Theory of Superconductivity

© Carsten Timm 2020

Version: January 6, 2025

Wintersemester 2017/2018 TU Dresden Institute of Theoretical Physics

Version: January 6, 2025
LATEX & Figures: S. Lange and C. Timm
Contents

1 Introduction 5
1.1 Scope and overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Books . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Basic experiments 7
2.1 Conventional superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Superfluid helium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Unconventional superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Bose-Einstein condensation in dilute gases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3 Bose-Einstein condensation 14

4 Normal metals 19
4.1 Electrons in metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.2 Semiclassical theory of transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

5 Electrodynamics of superconductors 24
5.1 London theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.2 Rigidity of the superfluid state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.3 Flux quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.4 Nonlocal response: Pippard theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

6 Ginzburg-Landau theory 31
6.1 Landau theory of phase transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.2 Ginzburg-Landau theory for neutral superfluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.3 Ginzburg-Landau theory for superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.4 Type-I superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.5 Type-II superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

7 Superfluid and superconducting films 54

7.1 Superfluid films . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
7.2 Superconducting films . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

8 Origin of the attractive interaction 73

8.1 Reminder on Green functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
8.2 Coulomb interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
8.3 Electron-phonon interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
8.4 Effective interaction between electrons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

9 Cooper instability and BCS ground state 85

9.1 Cooper instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
9.2 The BCS ground state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

3
10 BCS theory 93
10.1 BCS mean-field theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
10.2 Minimization of the free energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
10.3 Richardson’s solution of the BCS model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
10.4 Photoemission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
10.5 Isotope effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
10.6 Specific heat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
10.7 Density of states and single-particle tunneling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
10.8 Ultrasonic attenuation and nuclear relaxation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
10.9 Ginzburg-Landau-Gor’kov theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

11 Josephson effects 119

11.1 The Josephson effects in Ginzburg-Landau theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
11.2 Dynamics of Josephson junctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
11.3 The Bogoliubov-de Gennes Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
11.4 Andreev reflection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

12 Unconventional pairing 133

12.1 The gap equation for unconventional pairing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
12.2 Cuprates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
12.3 Pnictides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
12.4 Triplet superconductors and He-3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
12.5 Noncentrosymmetric superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

4
1

Introduction

1.1 Scope and overview

Superconductivity is characterized by a vanishing static electrical resistivity and an expulsion of the magnetic field
from the interior of a sample. We will discuss these basic experiments in the following chapter, but mainly this
course is dealing with the theory of superconductivity. We want to understand superconductivity using methods
of theoretical physics. Experiments will be mentioned if they motivate certain theoretical ideas or support or
contradict theoretical predictions, but a systematic discussion of experimental results will not be given.
Superconductivity is somewhat related to the phenomena of superfluidity (in He-3 and He-4) and Bose-Einstein
condensation (in weakly interacting boson systems). The similarities are found to lie more in the effective low-
energy description than in the microscopic details. Microscopically, superfluidity in He-3 is most closely related
to superconductivity since both phenomena involve the condensation of fermions, whereas in He-4 and, of course,
Bose-Einstein condensates it is bosons that condense. We will discuss these phenomena briefly.

fermions bosons
superconductivity superfluidity in He-4
superfluidity in He-3 Bose-Einstein condensation

After reviewing the basic experiments and Bose-Einstein condensation, we will discuss the electrodynamics of
superconductors. Then we will introduce the phenomenological Ginzburg-Landau theory for neutral superfluids
and for superconductors, which will naturally lead to the Anderson-Higgs mechanism. Ginzburg-Landau theory
already allows to understand many phenomena, as we will demonstrate by describing vortices. Vortices in thin
films can be created as thermal excitations and can form bound pairs, which leads to the Berezinskii-Kosterlitz-
Thouless theory of the phase transition in such films.
Turning to the microscopic description of superconductivity, we will examine the origin of the attractive
electron-electron interaction in conventional superconductors. This will form the basis for understanding the
Cooper instability of the normal Fermi sea and for the Bardeen-Cooper-Schrieffer (BCS) theory of superconductiv-
ity. With this theory in hand, we will discuss experimental consequences, e.g., in thermodynamics, single-particle
tunneling, and nuclear relaxation. Then we will turn to pair tunneling and Josephson effects. To improve our
theoretical toolbox, we will introduce the Bogoliubov-de Gennes equation for inhomogeneous superconductors.
The final chapters will be devoted to unconventional superconductors, for example cuprates and pnictides, to
Andreev scattering and Andreev bound states, and to topological superconductors.
The course assumes knowledge of the standard material from electrodynamics, quantum mechanics I, and
thermodynamics and statistics. We will also use the second-quantization formalism (creation and annihilation
operators), which are usually introduced in quantum mechanics II. A prior course on introductory solid state
physics would be useful but is not required. Formal training in many-particle theory is not required, necessary
concepts and methods will be introduced (or recapitulated) as needed.

5
1.2 Books
There are many textbooks on superconductivity and it is recommended to browse a few of them. None of them
covers all the material of this course. M. Tinkham’s Introduction to Superconductivity (McGraw-Hill, 2nd edition,
1996) is well written and probably has the largest overlap with this course.
A much broader, deeper, and more modern book is Superfluid States of Matter by B. Svistunov, E. Babaev,
and N. Prokof’ev (CRC Press, 2015). It certainly goes beyond this course in many places.
A classic book is J. R. Schrieffer’s Theory of Superconductivity (Addison-Wesley, 1983). It is deeper than
Tinkham’s book. Schrieffer uses methods from many-particle theory, which are, however, introduced in the book.
The presentation is not always pedagogical.
Many books on solid-state theory or on many-particle theory contain chapters on superconductivity at various
levels. One of them is Many-Body Quantum Theory in Condensed Matter Physics by H. Bruus and K. Flensberg
(Oxford University Press, 2004). The authors use a modern, pedagogical style. The book mainly uses Green-
function methods. Another excellent text is Condensed Matter Field Theory by A. Altland and B. D. Simons
(Cambridge University Press, 2010). It introduces the BCS theory of superconductivity both in the conventional
way, like in this lecture, and within the functional-integral formalism. The discussion is modern and lucid. There
is also a section on Berezinskii-Kosterlitz-Thouless theory.

6
2

Basic experiments

In this chapter we will review the essential experiments that have established the presence of superconductivity,
superfluidity, and Bose-Einstein condensation in various materials classes. Experimental observations that have
helped to elucidate the detailed properties of the superconductivity or superfluidity in specific systems are not
covered; some of them are discussed in later chapters.

2.1 Conventional superconductors

After H. Kamerlingh Onnes had managed to liquify Helium, it became possible to reach temperatures low enough
to achieve superconductivity in some chemical elements. In 1911, he found that the static (dc) resistivity of
mercury abruptly fell to zero at a critical temperature Tc of about 4.1 K.
ρ

normal metal

Hg

superconductor
0
0 Tc T

In a normal metal, the resistivity decreases with decreasing temperature but saturates at a finite value for T → 0.
The most stringent bounds on the resisitivity can be obtained not from direct measurement but from the decay
of persistent currents, or rather from the lack thereof. A current set up (by induction) in a superconducting ring
is found to persist without measurable decay after the electromotive force driving the current has been switched
off.
B

7
Assuming exponential decay, I(t) = I(0) e−t/τ , a lower bound on the decay time τ is found. From this, an
upper bound of ρ . 10−26 Ωm has been extracted for the resistivity (File and Mills, 1963). For comparison, the
resistivity of copper at room temperature is ρCu ≈ 1.7 × 10−8 Ωm.
The second essential observation was that superconductors not only prevent a magnetic field from entering—
this will be discussed in Sec. 5.1—but actively expel the magnetic field from their interior. This was observed by
W. Meißner and R. Ochsenfeld in 1933 and is now called the Meißner or Meißner-Ochsenfeld effect.

B
1111
0000
0000
1111
0000
1111
0000
1111
B=0

0000
1111
0000
1111
From the materials relation B = µH with the permeability µ = 1 + 4πχ and the magnetic suceptibility χ (note
that we are using Gaussian units) we thus find µ = 0 and χ = −1/4π. Superconductors are diamagnetic since
χ < 0. What is more, they realize the smallest (most diamagnetic) value of µ consistent with thermodynamic
stability.1 The field is not just diminished but completely expelled. They are thus perfect diamagnets.
It costs energy to make the magnetic field nonuniform although the externally applied field is uniform. It is
plausible that at some externally applied magnetic field Hc (T ) ≡ Bc (T ) this cost will be so high that there is no
advantage in forming a superconducting state. For typical conventional superconductors, the experimental phase
diagram in the temperature-magnetic field plane looks like this:
H
normal metal
Hc (T= 0)
1s
to
rd
er

superconductor 2nd order

Tc T

To specify which superconductors discovered after Hg are conventional, we need a definition of what we want
to call “conventional superconductors.” There are at least two inequivalent but often coinciding definitions:
Conventional superconductors
• show a superconducting state of trivial symmetry (essentially, the superconductor does not break any
symmetry beyond a global U(1) phase symmetry),
1 An argument for why µ cannot be negative goes as follows: with a suitably defined free energy F and the volume V one has
4π ∂F
H= .
V ∂B
Within the range of validity of linear-response theory, we have H = B/µ (if µ 6= 0) and thus
4πµ ∂F
B= .
V ∂B
This implies
V
F (B) = F (0) + B2
8πµ
so that µ < 0 would make the free energy unbounded from below.

8
 • result from an attractive interaction between electrons in which phonons play a dominant role.
Conventional superconductivity was observed in quite a lot of elements at low temperatures. The record critical
temperature for elements are Tc = 9.3 K for Nb under ambient pressure and Tc = 29 K for Ca under high pressure.
Superconductivity is in fact rather common in the periodic table, about 53 pure elements show it under some
conditions. Many alloys and intermetallic compounds were also found to show conventional superconductivity
according to the above criteria. Of these, for a long time Nb3 Ge had the highest known Tc of 23.2 K. But it is now
thought that MgB2 (Tc = 39 K, discovered to be superconducting in 2001) and a few related compounds are also
conventional superconductors in the above sense. They nevertheless show some interesting properties. The rather
high Tc = 39 K of MgB2 is interesting since it is on the order of the maximum Tcmax ≈ 30 K that was originally
expected for phonon-driven superconductivity. To increase Tc further, the interaction between electrons and
phonons would have to be stronger. However, it was thought that stronger interactions would make the material
unstable towards a charge density wave. MgB2 would thus be an “optimal” conventional superconductor.
Superconductivity with comparably high Tc has also been found in fullerites, i.e., compounds containing
fullerene anions. The record Tc in this class is at present Tc = 38 K for body-centered cubic Cs3 C60 under
pressure. Superconductivity in fullerites was originally thought to be driven by phonons with strong molecular
vibration character but there is recent evidence that it might be unconventional (not phonon driven).
The received wisdom was challenged by the discovery of superconductivity in hydrogen sulfide at up to 203 K
under a high pressure of above 150 GPa (corresponding to about 1.5 million standard atmospheres), see Drozdov
et al., Nature 525, 73 (2015). The responsible compound is thought to be H3 S forming a body-centered cubic
lattice. Amazingly, superconductivity in this system appears to be phonon driven and in this sense conventional.
The presence of the light hydrogen atoms is crucial for this. That H atoms could favor superconductivity at
high temperatures was already proposed by N. W. Ashcroft in 1968. There is an ongoing search for high-Tc
superconductivity in other hydrogen compounds. Recently, this effort has led to the discovery of superconductivity
with Tc ≈ 250 K in LaH10 at high pressure, see Drozkov et al., Nature 569, 528 (2019). The material is a clathrate,
in which the metal atom is surrounded by a cage of hydrogen atoms.

2.2 Superfluid helium

In 1937, P. Kapiza and independently Allen and Misener discovered that helium shows a transition at Tc = 2.17 K
under ambient pressure, below which it flows through narrow capilaries without resistance. The analogy to
superconductivity is obvious but here it was the viscosity instead of the resistivity that dropped to zero. The
phenomenon was called superfluidity. It was also observed that due to the vanishing viscosity an open container
of helium would empty itself through a flow in the microscopically thin wetting layer.
11111
00000
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
He
00000
11111
00000
11111

On the other hand, while part of the liquid flows with vanishing viscosity, another part does not. This was shown
using torsion pendulums of plates submerged in helium. A temperature-dependent normal component oscillates
with the plates.

9
 1111111
0000000
He

Natural atmospheric helium consists of 99.9999 % He-4 and only 0.0001 % He-3, the only other stable isotope.
The observed properties are thus essentially indistinguishable from those of pure He-4. He-4 atoms are bosons
since they consist of an even number (six) of fermions. For weakly interacting bosons, A. Einstein predicted in
1925 that a phase transition to a condensed phase should occur (Bose-Einstein condensation). The observation
of superfluidity in He-4 was thus not a suprise—in contrast to the discovery of superconductivity—but in many
details the properties of He-4 were found to differ from the predicted Bose-Einstein condensate. The reason for
this is that the interactions between helium atoms are actually quite strong. For completeness, we sketch the
temperature-pressure phase diagram of He-4:

p (10 6 Pa)
solid
3 phases

normal
superfluid liquid
1

ambient critical endpoint

gas
0
0 1 2 3 4 5 T (K)

The other helium isotope, He-3, consists of fermionic atoms so that Bose-Einstein condensation cannot take
place. Indeed no superfluid transition was observed in the temperature range of a few Kelvin. It then came
as a big suprise when superfluidity was finally observed at much lower temperatures below about 2.6 mK by D.
Lee, D. Osheroff, and R. Richardson in 1972. In fact they found two new phases at low temperatures. (They
originally misinterpreted them as possible magnetic solid phases.) Here is a sketch of the phase diagram, note
the temperature scale:

10
 p (10 6 Pa)
solid phases
superfluid
3 phase
A

2 superfluid
phase B normal
liquid

ambient gas
0
0 1 2 3 T (mK)

Superfluid He-3 shows the same basic properties as He-4. But unlike in He-4, the two superfluid states are sensitive
to an applied magnetic field, suggesting the presence of magnetic degrees of freedom at low energies.

2.3 Unconventional superconductors

By the late 1970’s, superconductivity seemed to be a more or less closed subject. It was well understood based on
the BCS theory and extensions thereof that dealt with strong interactions. It only occured at temperatures up to
23.2 K (Nb3 Ge) and thus did not promise widespread technological application. It was restricted to non-magnetic
metallic elements and simple compounds. This situation started to change dramatically in 1979. Since then,
superconductivity has been observed in various materials classes that are very different from each other and from
the typical low-Tc superconductors known previously. In many cases, the superconductivity is unconventional
and often Tc is rather high. We now give a brief and incomplete historical overview.
• In 1979, Frank Steglich et al. observed superconductivity below Tc ≈ 0.5 K in CeCu2 Si2 . This material is
not a normal metal in its normal state. Instead is is a heavy-fermion material. The electrons at the Fermi
energy have strong Ce f -orbital character. The very strong Coulomb repulsion between electrons in the
f -shell leads to a large effective mass m∗  me at the Fermi energy, hence the name. (The small overlap
between localized f orbitals also contributes to the large mass but interactions are crucial.) Since then,
superconductivity has been found in various other heavy-fermion compounds. BCS theory has difficulties to
explain superconductivity in these highly correlated metals. Nuclear magnetic resonance (discussed below)
and other experimental techniques have shown that many of these heavy-fermion superconductors show
unconventional symmetry of the superconducting state.
• Also in 1979, D. Jérome et al. (Klaus Bechgaard’s group) observed superconductivity in an organic salt
called (TMTSF)2 PF2 with Tc = 1.1 K. Superconductivity has since been found in various organic materials
with a maximum Tc of about 18 K. The symmetry of the superconducting state is often unconventional.
(We do not include fullerites under organic compounds since they lack hydrogen atoms.)
• While the previously mentioned discoveries showed that superconductivity can occur in unexpected materials
classes and probably due to unconventional mechanisms, the Tc values did not surpass the Tc ≈ 23 K of
Nb3 Ge. In 1986, J. G. Bednorz and K. A. Müller observed superconductivity in La2−x Bax CuO4 (the layered
perovskite cuprate La2 CuO4 with some Ba substituted for La) with Tc on the order of 35 K. In the following
years, many other superconductors based on the same type of nearly flat CuO2 planes sketched below were
discovered. The record transition temperatures for cuprates are Tc = 138 K for Hg0.8 Tl0.2 Ba2 Ca2 Cu3 O8+δ
at ambient pressure and Tc = 164 K for HgBa2 Ca2 Cu3 O8+δ under high pressure. Many experimental probes
show that the cuprates are unconventional superconductors. High-Tc superconductivity in the cuprates was
historically important since many advanced methods of many-body theory have been developed motivated
by the desire to understand this phenomenon. We will come back to this materials class below.

11
 Cu O
• In 1991, A. F. Hebard et al. found that the fullerite K3 C60 = (K− )3 C3− 60 became superconducting below
Tc = 18 K. Tc in this class has since been pushed to Tc = 33 K for Cs2 RbC60 at ambient pressure and
Tc = 38 K for (b.c.c., while all the other known superconducting fullerites are f.c.c.) Cs3 C60 under high
pressure. The symmetry of the superconducting state appears to be trivial but, as noted above, there is an
ongoing debate on whether the pairing is phonon-mediated.
• In 1994, Maeno et al. discovered superconductivity in Sr2 RuO4 with Tc ≈ 1.5 K. The (perovskite) lattice
structure is the same as for the earliest studied cuprates, such as doped La2 CuO4 , but with copper replaced
by ruthenium. The superconducting state appears to be magnetic and was long thought to be similar to
the superfluid states of He-3. Very recent experiments cast doubt upon this picture but superconductivity
in Sr2 RuO4 is certainly unconventional.
• In 2001, Nagamatsu et al. reported superconductivity in MgB2 with Tc = 39 K. The high Tc and the
layered crystal structure, reminiscant of cuprates, led to the expectation that superconductivity in MgB2 is
unconventional. However, most experts now think that it is actually conventional, as noted above.
• A new series of important discoveries started in 2006, when Kamihara et al. (H. Hosono’s group) observed
superconductivity with Tc ≈ 4 K in LaFePO, another layered compound. While this result added a new
materials class based on Fe2+ to the list of superconductors, it did not yet cause much excitement due to
the low Tc . However, in 2008, Kamihara et al. (the same group) found superconductivity with Tc ≈ 26 K
in LaFeAsO1−x Fx . Very soon thereafter, the maximum Tc in this iron-pnictide class was pushed to 55 K.
Superconductivity was also observed in several related materials classes, some of them not containing oxygen
(e.g., LiFeAs) and some with the pnictogen (As) replaced by a chalcogen (e.g., FeSe). The common structural
element is a flat, square Fe2+ layer with a pnictogen or chalcogen sitting alternatingly above and below
the centers of the Fe squares. Superconductivity is thought to be unconventional. The mechanism is not
necessarily the same for all members of this diverse family.

Fe As
• Drozdov et al. reported superconductivity in H3 S under high pressure with a maximum critical tempera-
ture of Tc = 203 K in 2015 and in LaH10 under high pressure with Tc ≈ 250 K in 2019, as noted above.
Although not completely understood yet, superconductivity is thought to be due to phonons and in this
sense conventional.

2.4 Bose-Einstein condensation in dilute gases

An important related breakthrough was the realization of a Bose-Einstein condensate (BEC) in a highly diluted
and very cold gas of atoms. In 1995, Anderson et al. (C. E. Wiman and E. A. Cornell’s group) reported conden-
sation in a dilute gas of Rb-87 below Tc = 170 nK (!). Only a few months later, Davis et al. (W. Ketterle’s group)
reported a BEC of Na-23 containing many more atoms. About a year later, the same group was able to create
two condensates and then merge them. The resulting interference effects showed that the atoms were really in a
macroscopic quantum state, i.e., a condensate. All observations are well understood from the picture of a weakly
interacting Bose gas. Bose-Einstein condensation will be reviewed in the following chapter.

12
13
3

Bose-Einstein condensation

In this short chapter we review the theory of Bose-Einstein condensation. While this is not the correct theory
for superconductivity, at least in most superconductors, it is the simplest description of a macroscopic quantum
condensate. This concept is central also for superconductivity and superfluidity.
We consider an ideal gas of indistinguishable bosons. “Ideal” means that we neglect any interaction between
the particles. There are two cases with completely different behavior depending on whether the particle number is
conserved or not. Rb-87 atoms are bosons (they consist of 87 nucleons and 37 electrons) with conserved particle
number, whereas photons are bosons with non-conserved particle number. Photons can be freely created and
destroyed as long as the usual conservation laws (energy, momentum, angular momentum, . . . ) are satisfied.
Bosons without particle-number conservation show a Planck distribution,
1
nP (E) = (3.1)
eβE − 1
with β := 1/kB T , for a grand-canonical ensemble in equilibrium. Note the absence of a chemical potential, which
is due to the non-conservation of the particle number. This distribution function is an analytical function of
temperature and thus does not show any phase transitions.
The situation is different for bosons with conserved particle number. We want to consider the case of a given
number N of particles in contact with a heat bath at temperature T . This calls for a canonical description (N, T
given). However, it is easier to use the grand-canonical ensemble with the chemical potential µ given. For large
systems, fluctuations of the particle number become small so that the descriptions are equivalent. However, µ
must be calculated from the given N .
The grand-canonical partition function is
Y  Y 1
Z= 1 + e−β(i −µ) + e−2β(i −µ) + · · · = , (3.2)
i i
1 − e−β(i −µ)

where i counts the single-particle states of energy i in a volume V . The form of Z expresses that every state
can be occupied not at all, once, twice, etc. For simplicity, we assume the volume to be a cube with periodic
boundary conditions. Then the states can be enumerated be wave vectors k compatible with these boundary
conditions. Introducing the fugacity
y := eβµ , (3.3)
we obtain
Y 1
Z= (3.4)
1 − ye−βk
k
X 1 X
ln 1 − ye−βk . (3.5)


⇒ ln Z = ln −β
=−
1 − ye k
k k

14
The fugacity has to be chosen to give the correct particle number
!
X X 1 X 1
N= hnk i = = . (3.6)
eβ(k −µ) −1 y −1 eβk −1
k k k

Since hnk i must be non-negative, µ must satisfy

µ ≤ k ∀ k. (3.7)

For a free particle, k = k 2 /2m and the lowest possible eigenenergy is k = 0 for k = 0 so that we obtain µ ≤ 0
and thus y ≤ 1.
For a large volume V , the allowed vectors k become dense and we can replace the sums over k by integrals
according to
Z∞
d3 k √
Z
X 2πV
... → V 3
. . . = 3 (2m) 3/2
d  . . . (3.8)
(2π) h
k 0

In the last equation we have used the density of states (DOS) of free particles in three dimensions. Note that the
DOS would be different for a different number of dimensions. For example, in two dimensions, the DOS scales
as 0 = 1. The DOS is also different if one assumes a harmonic instead of a square-well confining potential. The
replacement in Eq. (3.8) contains a fatal mistake, though. The DOS for  = 0 vanishes so that any particles in
the state with k = 0, k = 0 do not contribute to the results. But that is the ground state! For T = 0 all bosons
should be in this state. We thus expect incorrect results at low temperatures.
Our mistake was that Eq. (3.8) does not hold if the fraction of bosons in the k = 0 ground state is macroscopic,
i.e., if N0 /N := hn0 i /N remains finite for large V . To correct this, we treat the k = 0 state explicitly (the same
would be necessary for any state with macroscopic occupation). We write
Z∞
X 2πV √
... → (2m)3/2 d  . . . + (k = 0 term). (3.9)
h3
k 0

Note that splitting off the ground state would be not just unnecessary but incorrect for a two-dimensional gas.
Treating the ground state explicitly, we obtain
Z∞ Z∞
2πV √ −β by parts 2πV 2 3/2
ln Z = − 3 (2m)3/2 (2m)3/2 β


d  ln 1 − ye − ln(1 − y) = d − ln(1 − y)
h h3 3 y −1 eβ −1
0 0
(3.10)
with
Z∞ Z∞ √
2πV √ 1 1 2πV  y
N = 3 (2m)3/2 d  −1 −β + = 3 (2m)3/2 d + . (3.11)
h y e − 1 y −1 − 1 h y −1 e−β − 1 1 − y
0 0

Defining
Z∞
1 xn−1
gn (y) := dx (3.12)
Γ(n) y −1 ex−1
0

for 0 ≤ y ≤ 1 and n ∈ R, and the thermal wavelength

s
h2
λ := , (3.13)
2πmkB T

we obtain
V
ln Z = g5/2 (y) − ln(1 − y), (3.14)
λ3

15
 V y
N= g (y) +
3 3/2
. (3.15)
λ 1 −
| {z } | {z y}
=: N =: N0

We note the identity

∞
X yk
gn (y) = , (3.16)
kn
k=1

which implies

gn (0) = 0, (3.17)
∞
X 1
gn (1) = ≡ ζ(n) for n > 1 (3.18)
kn
k=1

with the Riemann zeta function ζ(x). Furthermore, gn (y) increases monotonically in y for y ∈ [0, 1[.
gn ( y)
g3/2

2
g5/2

1 goo

0
0 0.5 1 y
We now have to eliminate the fugacity y from Eqs. (3.14) and (3.15) to obtain Z as a function of the particle
number N . In Eq. (3.14), the first term is the number of particles in excited states (k > 0), whereas the second
term is the number of particles in the ground state. We consider two cases: 1. If y is not very close to unity
(specifically, if 1 − y  λ3 /V ), N0 = y/(1 − y) is on the order of unity, whereas N is an extensive quantity. Thus
N0 can be neglected and we get
V
N∼ = N = 3 g3/2 (y). (3.19)
λ
Since g3/2 ≤ ζ(3/2) ≈ 2.612, this equation can only be solved for the fugacity y if the concentration satisfies
N ζ(3/2)
≤ . (3.20)
V λ3
Note that the right-hand side decreases with decreasing temperature since λ3 ∝ T −3/2 increases. Hence, below a
critical temperature Tc , the inequality is no longer fulfilled. From
N ! ζ(3/2)
= 3/2 (3.21)
V h2
2πmkB Tc

we obtain 2/3
h2

1 N
kB Tc = . (3.22)
[ζ(3/2)]2/3 2πm V
2. If, on the other hand, y is very close to unity, i.e., 1 − y . λ3 /V , N0 cannot be neglected. Moreover, in this
case we find
V V
N = 3 g3/2 (y) = 3 g3/2 (1 − O(λ3 /V )), (3.23)
λ λ

16
where O(λ3 /V ) is a correction of order λ3 /V  1. Thus, by Taylor expansion,

V V
N = g3/2 (1) − O(1) = 3 ζ(3/2) − O(1). (3.24)
λ3 λ
The intensive term O(1) can be neglected compared to the extensive one so that

V
N ∼
= 3 ζ(3/2). (3.25)
λ
This is the maximum possible value at temperature T . Furthermore, N0 = y/(1 − y) is solved by

N0 1
y= = . (3.26)
N0 + 1 1 + 1/N0

For y to be very close to unity, N0 must satisfy N0  1. Since

V
N0 = N − N ∼
= N − 3 ζ(3/2) (3.27)
λ
must be positive, we require

N ζ(3/2)
> (3.28)
V λ3
⇒ T < Tc . (3.29)

We conclude that the fraction of particles in excited states is

3/2
λ3 (Tc )

N ∼ V T
= ζ(3/2) = = . (3.30)
N N λ3 λ3 (T ) Tc

The fraction of particles in the ground state is then

 3/2
N0 ∼ T
=1− . (3.31)
N Tc

In summary, we find in the thermodynamic limit

(a) for T > Tc :
N ∼ N0
= 1,  1, (3.32)
N N
(b) for T < Tc :
 3/2  3/2
N ∼ T N0 ∼ T
= , =1− . (3.33)
N Tc N Tc

1
N0 / N

Nε / N

0
0 0.5 1 T / Tc

17
We find a phase transition at Tc , below which a macroscopic fraction of the particles occupy the same single-
particle quantum state. This fraction of particles is said to form a condensate. While it is remarkable that
Bose-Einstein condensation happens in a non-interacting gas, the BEC is analogous to the condensate in strongly
interacting superfluid He-4 and, with some added twists, in superfluid He-3 and in superconductors.
We can now use the partition function to derive equations of state. As an example, we consider the pressure
∂Φ ∂
p=− =+ kB T ln Z
∂V  ∂V 
∂ V
= kB T g5/2 (y) − ln(1 − y) . (3.34)
∂V λ3

(Φ is the grand-canonical potential). In principle, the fugacity y depends on the volume V since we have to invert

V y
N= g3/2 (y) + , (3.35)
λ3 1−y

which gives us y(N, V ). Derivatives with respect to V are taken at fixed particle number N . Thus we get
 
kB T V 0 1 ∂y
p = 3 g5/2 (y) + kB T g (y) +
λ λ3 5/2 1 − y ∂V
 
kB T V g3/2 (y) 1 ∂y
= 3 g5/2 (y) + kB T + . (3.36)
λ λ3 y 1 − y ∂V

For T > Tc , we can find y and thus p numerically. For T < Tc we may set y = 1 so that ∂y/∂V = 0 and

kB T
p= ζ(5/2) ∝ T 5/2 . (3.37)
λ3
Note that only the excited states contribute to the pressure. This is plausible since particles in the condensate
have vanishing kinetic energy.
The result should be compared to the classical ideal gas. In that case the equation of state gives

p ∝ T. (3.38)

For the BEC, the pressure drops more rapidly since more and more particles condense and thus no longer
contribute to the pressure.

18
4

Normal metals

To be able to appreciate the remarkable poperties of superconductors, it is useful to review what we know about
normal conductors.

4.1 Electrons in metals

Let us ignore electron-electron Coulomb interaction and deviations from a perfectly periodic crystal structure
(due to defects or phonons) for now. Then the exact single-particle states are described by Bloch wavefunctions

ψαk (r) = uαk (r) eik·r , (4.1)

where uαk (r) is a lattice-periodic function, α is the band index including the spin, and ~k is the crystal momentum
in the first Brillouin zone. Since electrons are fermions, the average occupation number of the state |αki with
energy αk is given by the Fermi-Dirac distribution function
1
nF (αk ) = . (4.2)
eβ(αk −µ) + 1
If the electron number N , and not the chemical potential µ, is given, µ has to be determined from
X 1
N= , (4.3)
αk
eβ(αk −µ) +1

cf. our discussion for ideal bosons. In the thermodynamic limit we again replace

d3 k
X Z
... → V ... (4.4)
(2π)3
k

Unlike for bosons, this is harmless for fermions, since any state can at most be occupied once so that macroscopic
occupation of the single-particle ground state cannot occur. Thus we find

N X Z dk 3 1
= 3 eβ(αk −µ) + 1
. (4.5)
V α
(2π)

If we lower the temperature, the Fermi function nF becomes more and more step-like. For T → 0, all states with
energies αk ≤ EF := µ(T → 0) are occupied (EF is the Fermi energy), while all states with αk > EF are empty.
This Fermi sea becomes fuzzy for energies αk ≈ EF at finite temperatures but remains well defined as long as
kB T  EF . This is the case for most materials we will discuss.
The chemical potential, the occupations nF (αk ), and thus all thermodynamic variables are analytic functions
of T and N/V . Thus there is no phase transition, unlike for bosons. Free fermions represent the special case with
only a single band with dispersion k = ~2 k 2 /2m. If we replace m by a material-dependent effective mass, this

19
gives a reasonable approximation for simple metals such as alkali metals. Qualitatively, the conclusions are much
more general.
Lattice imperfections and interactions result in the Bloch waves ψαk (r) not being exact single-particle eigen-
states. (Electron-electron and electron-phonon interactions invalidate the whole idea of single-particle states.)
However, if these effects are in some sense small, they can be treated pertubatively in terms of scattering of
electrons between single-particle states |αki.

4.2 Semiclassical theory of transport

We now want to derive an expression for the current in the presence of an applied electric field. This is a question
about the response of the system to an external perturbation. There are many ways to approach this type of
question. If the perturbation is small, the response, in our case the current, is expected to be a linear function
of the perturbation. This is the basic assumption of linear-response theory. In the framework of many-particle
theory, linear-response theory results in the Kubo formula (see lecture notes on many-particle theory). We here
take a different route. If the external perturbation changes slowly in time and space on atomic scales, we can use
a semiclassical description. Note that the following can be derived cleanly as a limit of many-particle quantum
theory.
The idea is to consider the phase space distribution function ρ(r, k, t). This is a classical concept. From
quantum mechanics we know that r and p = ~k are subject to the uncertainty principle ∆r∆p ≥ ~/2. Thus
distribution functions ρ localized in a phase-space volume smaller than on the order of ~3 violate quantum
mechanics. On the other hand, if ρ is much broader, quantum effects should be negligible.
The Liouville theorem shows that ρ satisfies the continuity equation
∂ρ ∂ρ ∂ρ dρ
+ ṙ · + k̇ · ≡ =0 (4.6)
∂t ∂r ∂k dt
(phase-space volume is conserved under the classical time evolution). Assuming for simplicity a free-particle
dispersion, we have the canonical (Hamilton) equations
∂H p ~k
ṙ = = = , (4.7)
∂p m m
1 1 ∂H 1 1
k̇ = ṗ = − = − ∇V = F (4.8)
~ ~ ∂r ~ ~
with the Hamiltonian H and the force F. Thus we can write
 
∂ ~k ∂ F ∂
+ · + · ρ = 0. (4.9)
∂t m ∂r ~ ∂k
This equation is appropriate for particles in the absence of any scattering. For electrons in a uniform and time-
independent electric field we have
F = −eE. (4.10)
Note that we always use the convention that e > 0. It is easy to see that then
 
eE
ρ(r, k, t) = f k + t (4.11)
~
is a solution of Eq. (4.9) for any differentiable function f . This solution is uniform in real space (∂ρ/∂r ≡ 0)
and shifts to larger and larger momenta ~k for t → ∞. It thus describes the free acceleration of electrons in an
electric field. There is no finite conductivity since the current never reaches a stationary value. This is obviously
not a correct description of a normal metal. Our mistake has been to ignore scattering.
Scattering will change ρ as a function of time beyond what as already included in Eq. (4.9). We collect all
processes not included in Eq. (4.9) into a scattering term S[ρ]:
 
∂ ~k ∂ F ∂
+ · + · ρ = −S[ρ]. (4.12)
∂t m ∂r ~ ∂k

20
This is the famous Boltzmann equation. The notation S[ρ] signifies that the scattering term is a functional of ρ.
It is generally not simply a function of the local density ρ(r, k, t).
While expressions for S[ρ] can be derived for various types of scattering—interesting and unconventional
transport properties in the normal state can be studied in this way—for our purposes it is sufficient to employ the
simple but common relaxation-time approximation. It is based on the observation that ρ(r, k, t) should relax to
thermal equilibrium if no force is applied (F ≡ 0). For fermions, the equilibrium distribution is ρ0 (k) ∝ nF (k ).
This relaxation is ensured by the ansatz
ρ(r, k, t) − ρ0 (k)
S[ρ] = . (4.13)
τ
Here, τ is the relaxation time, which determines how fast ρ relaxes towards ρ0 .
If there are different scattering mechanisms that act independently, the scattering integral is just a sum of
contributions of these mechanisms,
S[ρ] = S1 [ρ] + S2 [ρ] + . . . (4.14)
Consequently, if the relaxation-time approximation is valid the relaxation rate 1/τ can be written as
1 1 1
= + + ... (4.15)
τ τ1 τ2
This is Matthiessen’s rule. There are three main scattering mechanisms:
• Scattering of electrons by disorder: This gives an essentially temperature-independent contribution, which
dominates at low temperatures.
• Electron-phonon interaction: This mechanism is strongly temperature-dependent because the available
phase space shrinks at low temperatures. One finds a scattering rate
1
∝ T 3. (4.16)
τe-ph
However, this is not the relevant rate for transport calculations. The conductivity is much more strongly
affected by scattering that changes the electron momentum ~k by a lot than by processes that change it
very little. Backscattering across the Fermi sea is most effective.

ky

∆k = 2 kF

kx

Since backscattering is additionally suppressed at low T , the transport scattering rate relevant here scales
as
1
trans ∝ T 5. (4.17)
τe-ph

• Electron-electron interaction: For a parabolic free-electron band its contribution to the resistivity is actually
zero since Coulomb scattering conserves the total momentum of the two scattering electrons and therefore
does not relax the current. However, in a real metal, umklapp scattering can take place that conserves the
electrons’ total momentum only modulo a reciprocal lattice vector. Thus the electron system can transfer
momentum to the crystal as a whole and thereby degrade the current. The temperature dependence is
typically
1
umklapp
∝ T 2. (4.18)
τe-e

21
 We now consider the force F = −eE and calculate the current density
d3 k ~k
Z
j(r, t) = −e ρ(r, k, t). (4.19)
(2π)3 |{z}
m
=v

To that end, we solve the Boltzmann equation in the relaxation-time approximation,

 
∂ ~k ∂ eE ∂ ρ − ρ0
+ · − · ρ=− . (4.20)
∂t m ∂r ~ ∂k τ
We are interested in the stationary solution (∂ρ/∂t = 0), which for a uniform field must be spatially uniform
(∂ρ/∂r = 0) if it is unique. This gives
eE ∂ ρ(k) − ρ0 (k)
− · ρ(k) = − (4.21)
~ ∂k τ
eEτ ∂
⇒ ρ(k) = ρ0 (k) + · ρ(k). (4.22)
~ ∂k
We iterate this equation by inserting it again into the last term:
 
eEτ ∂ eEτ ∂ eEτ ∂
ρ(k) = ρ0 (k) + · ρ0 (k) + · · ρ(k). (4.23)
~ ∂k ~ ∂k ~ ∂k
To make progress, we assume that the applied field E is small so that the response j is linear in E. Under this
assumption we can truncate the iteration after the linear term,
eEτ ∂
ρ(k) = ρ0 (k) + · ρ0 (k). (4.24)
~ ∂k
By comparing this to the Taylor expansion
 
eEτ eEτ ∂
ρ0 k + = ρ0 (k) + · ρ0 (k) + . . . (4.25)
~ ~ ∂k
we see that the solution is, to linear order in E,
 
eEτ
ρ(k) = ρ0 k + ∝ nF (k+eEτ /~ ). (4.26)
~
Thus the distribution function is simply the Fermi sea shifted in k-space by −eEτ /~. Since electrons carry
negative charge, the distribution is shifted in the direction opposite to the applied electric field.

E
ρ0

eE τ /h

The current density now reads

d3 k ~k d3 k ~k d3 k ~k eEτ ∂ρ0
Z   Z Z  
eEτ ∼
j = −e ρ 0 k + = −e ρ 0 (k) − e · . (4.27)
(2π)3 m ~ (2π)3 m (2π)3 m ~ ∂k
| {z }
=0

22
The first term is the current density in equilibrium, which vanishes. In components, we have

e2 τ X d3 k ∂ρ0 by parts e2 τ X d3 k ∂kα e2 τ d3 k

Z Z Z
jα = − Eβ k α = + E β ρ 0 = E α ρ0 . (4.28)
m (2π)3 ∂kβ m (2π)3 ∂kβ m (2π)3
β β |{z}
= δαβ

Here, the integral is the concentration of electrons, n := N/V . We thus obtain Ohm’s law

e2 nτ !
j= E = σE (4.29)
m
so that the conductivity is
e2 nτ
σ= . (4.30)
m
This is the famous Drude formula. For the resistivity ρ = 1/σ we get, based on our discussion of scattering
mechanisms, !
m m 1 1 1
ρ= 2 = 2 + transport + umklapp . (4.31)
e nτ e n τdis τe-ph τe-e

large T :
∝ T 5 mostly due
to electron−
phonon scattering
residual resistivity
due to disorder

23
5

Electrodynamics of superconductors

Superconductors are defined by electrodynamic properties—ideal conduction and magnetic-field expulsion. It is

thus appropriate to ask how these materials can be described within the formal framework of electrodynamics.

5.1 London theory

In 1935, F. and H. London proposed a phenomenological theory for the electrodynamic properties of superconduc-
tors. It is based on a two-fluid picture: For unspecified reasons, the electrons from a normal fluid of concentration
nn and a superfluid of concentration ns , where nn + ns = n = N/V . Such a picture seemed quite plausible
based on Einstein’s theory of Bose-Einstein condensation (1925), although nobody understood how the fermionic
electrons could form a superfluid. The normal fluid is postulated to behave normally, i.e., to carry an ohmic
current
jn = σ n E (5.1)
governed by the Drude conductivity
e2 nn τ
σn = . (5.2)
m
The superfluid is assumed to be insensitive to scattering. As noted in section 4.2, this leads to free acceleration
of the charges. The Londons have assumed that nn and ns are both uniform (constant in space) and stationary
(constant in time). These are serious restrictions of London theory, which will be overcome by Ginzburg-Landau
theory. With the supercurrent
js = −e ns vs (5.3)
and Newton’s equation of motion
d F eE
vs = =− , (5.4)
dt m m
we then obtain
∂js e2 ns
= E. (5.5)
∂t m
The same follows from the Boltzmann equation in the absence of scattering. This is the First London Equation.
Note that the curl of the First London Equation is

∂ e2 ns Faraday e2 ns ∂B
∇ × js = ∇×E = − . (5.6)
∂t m mc ∂t
This equation can be integrated in time to give

e2 ns
∇ × js = − B + C(r), (5.7)
mc

24
where the last term represents a constant of integration at each point r inside the superconductor. C(r) should
be determined from the initial conditions. If we start from a superconducting body in zero applied magnetic
field, we have js ≡ 0 and B ≡ 0 initially so that C(r) = 0. To describe the Meißner-Ochsenfeld effect, we have
to consider the case of a body becoming superconducting (by cooling) in a non-zero applied field. However, the
outcome cannot be derived within London theory since we have assumed the superfluid density ns to be constant
in time.
To account for the flux expulsion, the Londons postulated that C ≡ 0 regardless of the history of the system.
This leads to
e2 ns
∇ × js = − B, (5.8)
mc
which is the Second London Equation.
Taking the curl of Ampère’s law
4π 4π
∇×B= js + jn (5.9)
c c
(there is no displacement current in the stationary state) we get

4πe2 ns 4π 4πe2 ns 4π ∂B
∇×∇×B=− 2
B + σ n ∇ × E = − 2
B− σn . (5.10)
mc c mc c ∂t
We drop the last term since we are interested in the stationary state and use an identity from vector calculus
(BAC-CAB rule), which gives
4πe2 ns
− ∇(∇ · B) + ∇2 B = B. (5.11)
mc2
Introducing the London penetration depth s
mc2
λL := , (5.12)
4πe2 ns
this equation assumes the simple form
1
∇2 B = B. (5.13)
λ2L
Let us consider a semi-infinite superconductor filling the half space x > 0. A magnetic field Bapl = Hapl = Bapl ŷ
is applied parallel to the surface. One can immediately see that the equation is solved by

B(x) = Bapl ŷ e−x/λL for x ≥ 0. (5.14)

The magnetic field thus decresases exponentially with the distance from the surface of the superconductor. In
bulk we indeed find B → 0, which is the Meißner-Ochsenfeld effect.
B

Bapl

outside inside

0 λL x

The Second London Equation

c
∇ × js = − B (5.15)
4πλ2L
and the continuity equation
∇ · js = 0 (5.16)
can now be solved to give
c
js (x) = − Bapl ẑ e−x/λL for x ≥ 0. (5.17)
4πλL

25
Thus the supercurrent flows in the direction parallel to the surface and perpendicular to B and decreases into
the bulk on the same scale λL . js can be understood as the screening current required to keep the magnetic field
out of the bulk of the superconductor.
The two London equations (5.5) and (5.8) can be summarized using the vector potential since 1c ∂A ∂t = −E
and ∇ × A = B:
e2 ns
js = − A. (5.18)
mc
This equation is evidently not gauge invariant since a change of gauge

A → A + ∇χ (5.19)

changes the supercurrent. (The two London equations are gauge invariant since they are expressed in terms of E
and B.) Charge conservation requires ∇ · js = 0 and thus the vector potential must be transverse,

∇ · A = 0. (5.20)

This is called the London gauge in this context and the Coulomb gauge in general. Furthermore, the supercurrent
through the surface of the superconducting region is proportional to the normal component A⊥ , with constant
of proportionality determined by Eq. (5.18). For a simply connected region these conditions uniquely determine
A(r). For a multiply connected region this is not the case; we will return to this point below.

5.2 Rigidity of the superfluid state

F. London has given a quantum-mechanical justification for the London equations. If the many-body wave
function of the electrons forming the superfluid is Ψs (r1 , r2 , . . . ) then the supercurrent in the presence of a vector
potential A is
Z      
1 X ~ ∂ e ~ ∂ e
js (r) = −e d3 r1 d3 r2 · · · δ(r − rj ) Ψ∗s + A(rj ) Ψs + Ψs − + A(rj ) Ψ∗s . (5.21)
2m j i ∂rj c i ∂rj c

Here, j sums over all electrons in the superfluid, which have position rj and momentum operator

~ ∂
pj = . (5.22)
i ∂rj

Splitting the kinetic and electromagnetic terms, we obtain

Z  
e~ X ∂ ∂ ∗
js (r) = − d3 r1 d3 r2 · · · δ(r − rj ) Ψ∗s Ψs − Ψs Ψs
2mi j ∂rj ∂rj
e2 X Z
− A(r) d3 r1 d3 r2 · · · δ(r − rj ) Ψ∗s Ψs
mc j

e2 ns
Z  
e~ X
∗ ∂ ∂ ∗
=− 3 3
d r1 d r2 · · · δ(r − rj ) Ψs Ψs − Ψs Ψs − A(r). (5.23)
2mi j ∂rj ∂rj mc

In general, the wave function Ψs depends on A. Now London proposed that the wave function is rigid under
the application of a transverse vector potential. More specifically, he suggested that Ψs does not contain a term
of first order in A, in the gauge ∇ · A = 0.1 Then, to first order in A, the first term on the right-hand side
in Eq. (5.23) contains the unperturbed wave function one would obtain for A = 0. The first term is thus the
1 A possible argument for this goes as follows: a linear term would necessarily be proportional to c · A(r), where c is some vector.

However, the whole current density and also its first term only depend (implicitly) on a single vector, namely A(r). Since there is no
c there cannot appear a linear term.

26
supercurrent for A ≡ 0, which should vanish due to Ampère’s law. Consequently, to first order in A we obtain
the London equation
e2 ns
js = − A. (5.24)
mc
The rigidity of Ψs was later understood in the framework of BCS theory as resulting from the existence of a
non-zero energy gap for excitations out of the superfluid state.

5.3 Flux quantization

We now consider two concentric superconducting cylindrical shells that are thick compared to the London pene-
tration depth λL . A magnetic flux Z
Φ = d2 rB⊥ (5.25)

penetrates the inner hole and a thin surface layer with a thickness on the order of λL of the inner cylinder. The
only purpose of the inner cylinder is to prevent the magnetic field from touching the outer cylinder, which we are
really interested in. The outer cylinder is completely field free. We want to find the possible values of the flux Φ.
y

ϕ
Φ
x

thick compared to λL
Although the region outside of the inner cylinder has B = 0, the vector potential does not vanish. The relation
∇ × A = B implies I ZZ
ds · A = d2 r · B = Φ. (5.26)

By symmetry, the tangential part of A is

Φ
Aϕ = . (5.27)
2πr
The London gauge requires this to be the only non-zero component. Thus outside of the inner cylinder we have,
in cylindrical coordinates,
Φ
A= ϕ̂. (5.28)
2πr

27
This can be rewritten as
Φϕ
A=∇ . (5.29)
2π
Since this is a pure gradient, we can get from A = 0 to A = (Φ/2πr) ϕ̂ by a gauge transformation

A → A + ∇χ (5.30)

with
Φϕ
χ= . (5.31)
2π
χ is continuous but multivalued outside of the inner cylinder. Its gradient is single valued, though. We recall
that a gauge transformation of A must be accompanied by a transformation of the wave function,
 
ieX
Ψs → exp − χ(rj ) Ψs . (5.32)
~c j

This is most easily seen by noting that it guarantees the current js , Eq. (5.21) in Sec. 5.2, to remain invariant
under gauge transformations. Thus the wavefuncion at Φ = 0 (A = 0) and at non-zero flux Φ are related by
   
i e X Φϕj e X
Φ
Ψs = exp − 0
Ψs = exp − i Φ ϕj Ψ0s , (5.33)
~ c j 2π hc j

where ϕj is the polar angle of electron j. It may seem strange that a gauge transformation connects two states with
different observable properties, namely different fluxes. However, this is possible here since the transformation
is not done in all of space but only outside of the inner cylinder, which constitutes a region that is not simply
connected.
For ΨΦs as well as Ψs to be single valued and continuous, the exponential factor must not change for ϕj →
0

ϕj + 2π for any j. This is the case if

e hc
Φ∈Z ⇔ Φ=n with n ∈ Z. (5.34)
hc e
We find that the magnetic flux Φ is quantized in units of hc/e. Note that the inner cylinder can be omitted:
Assume we are heating it enough to become normal-conducting. Then the flux Φ will fill the whole interior of
the outer cylinder plus a thin (on the order of λL ) layer on its inside. But if the outer cylinder is much thicker
than λL , this should not affect Ψs appreciably, away from this thin layer.
The quantum hc/e is actually not correct. Based in the idea that two electrons could form a boson that
could Bose-Einstein condense, Onsager suggested that the relevant charge is 2e instead of e, leading to the
superconducting flux quantum
hc
Φ0 := , (5.35)
2e
which is indeed found in experiments.

5.4 Nonlocal response: Pippard theory

Experiments often find a magnetic penetration depth λ that is significantly larger than the Londons’ prediction
λL , in particular in dirty samples with large scattering rates 1/τ in the normal state. Pippard explained this
on the basis of a nonlocal electromagnetic response of the superconductor. The underlying idea is that the
quantum state of the electrons forming the superfluid cannot be arbitrarily localized. The typical energy scale of
superconductivity is expected to be kB Tc . Only electrons with energies  within ∼ kB Tc of the Fermi energy can
contribute appreciably. This corresponds to a momentum range ∆p determined by

kB Tc ! ∂
= = vF (5.36)
∆p ∂p =EF

28
 kB Tc
⇒ ∆p = , (5.37)
vF
with the Fermi velocity vF . From this and Heisenberg’s uncertainty principle, we estimate that the electrons
cannot be localized on scales smaller than
~ 1 ~vF
∆x ≈ = . (5.38)
2∆p 2 kB Tc
Therefore, Pippard introduced the coherence length
~vF
ξ0 = α (5.39)
kB Tc
as a measure of the minimum extent of electronic wavepackets. α is a numerical constant of order unity. BCS
theory predicts α ≈ 0.180. Pippard proposed to replace the local equation

e2 ns
js = − A (5.40)
mc
of London theory by the nonlocal expression

3 e2 ns ∆r(∆r · A(r0 )) −∆r/ξ0

Z
js (r) = − d3 r0 e (5.41)
4πξ0 mc (∆r)4

with
∆r = r − r0 . (5.42)
The special form of this equation was motivated by an earlier nonlocal generalization of Ohm’s law. The main
point is that electrons within a distance ξ0 of the point r0 where the field A acts have to respond to it because of
the minimum localization of the electronic state. If A does not change appreciably on the scale of ξ0 , we obtain

3 e2 ns ∆r(∆r · A(r)) −∆r/ξ0 3 e2 ns ∆r(∆r · A(r)) −∆r/ξ0

Z Z
js (r) ∼
=− d3 r 0 e = − d3 ∆r e . (5.43)
4πξ0 mc (∆r)4 4πξ0 mc (∆r)4

The result has to be parallel to A(r) by symmetry (since it is a vector depending on a single vector A(r)). Thus

3 e2 ns (∆r · Â(r))2 −∆r/ξ0

Z
js (r) = − A(r) d3 ∆r e
4πξ0 mc (∆r)4
Z1 Z∞
3 e2 ns 2(∆r)
 2
cos2 θ −∆r/ξ0
=− A(r) 2π d(cos θ) d∆r  (∆r)
 e
4πξ0 mc (∆r)
 4
−1 0 
3 e2 ns 2 e2 ns
=− A(r) 2π ξ0 = − A(r). (5.44)
4πξ0 mc 3 mc
We recover the local London equation. However, for many conventional superconductors, ξ0 is much larger than
λ. Then A(r) drops to zero on a length scale of λ  ξ0 . According to Pippard’s equation, the electrons respond
to the vector potential averaged over regions of size ξ0 . This averaged field is much smaller than A at the surface
so that the screening current js is strongly reduced and the magnetic field penetrates much deeper than predicted
by London theory, i.e., λ  λL .
To be a bit more quantitative, we can estimate the averaged vector potential and thus the screening current
to be reduced by a factor on the order of λ/ξ0 . From Ampère’s law ∇ × B = 4π c js , we infer that the length scale
of the exponential decay of B, i.e., the effective penetration depth, should then be enhanced by the inverse of this
factor. We thus get
ξ0
λ ≈ λL (5.45)
λ

29
 ⇒ λ2 ≈ λL ξ0 (5.46)
(5.47)
p
⇒ λ ≈ λL ξ0 ,

i.e., we estimate the effective penetration depth to be the geometric mean of the London penetration depth and
Pippard’s coherence length. Note √ that for λL  ξ0 we obtain λL  λ  ξ0 , consistent with our arguments.
Remember that the result λ ≈ λL ξ0 only holds in this limit.
The above motivation for the coherence length ξ0 relied on having a clean system. In the presence of strong
scattering the electrons can be localized on the scale of the mean free path l := vF τ . Pippard phenomenologically
generalized the equation for js by introducing a new length ξ where
1 1 1
= + (5.48)
ξ ξ0 βl

(β is a numerical constant of order unity) and writing

3 e2 ns ∆r(∆r · A(r0 )) −∆r/ξ

Z
js = − d3 r0 e . (5.49)
4πξ0 mc (∆r)4

Note that ξ0 appears in the prefactor but ξ in the exponential. This expression is in good agreement with
experiments for series of samples with varying disorder. It is essentially the same as the result of BCS theory.
Also note that in the dirty limit l  ξ0 , λL we again recover the local London result following the same argument
as above, but since the integral gives a factor of ξ, which is not canceled by the prefactor 1/ξ0 , the current is
reduced to
e2 ns ξ e2 ns βl
js = − A∼=− A. (5.50)
mc ξ0 mc ξ0
Note that the small mean free path in the dirty limit makes the response local again.
Taking the curl, we obtain

e2 ns βl
∇ × js = − B (5.51)
mc ξ0
4πe2 ns βl
⇒ ∇×∇×B=− B (5.52)
mc2 ξ0
4πe2 ns βl
⇒ ∇2 B = B, (5.53)
mc2 ξ0
in analogy to the derivation in Sec. 5.1. This equation is of the form
1
∇2 B = B (5.54)
(λdis
L )
2

with the London penetration depth

s s s
mc2 ξ0 ξ0
λdis
L = = λL . (5.55)
4πe2 ns βl βl

Thus the penetration depth is increased by a factor of order ξ0 /l in the dirty limit, l  ξ0 , λL . Since the
p

response is local in the dirty limit, the effective penetration depth equals the modified London result, i.e.,
s
ξ0
λdis = λL . (5.56)
βl

30
6

Ginzburg-Landau theory

Within London and Pippard theory, the superfluid density ns is treated as given. There is no way to understand
the dependence of ns on, for example, temperature or applied magnetic field within these theories. Moreover, ns
has been assumed to be constant in time and uniform in space—an assumption that is expected to fail close to
the surface of a superconductor.
These deficiencies are cured by the Ginzburg-Landau theory put forward in 1950. Like Ginzburg and Landau
we ignore complications due to the nonlocal electromagnetic response. Ginzburg-Landau theory is developed as
a generalization of London theory, not of Pippard theory. The starting point is the much more general and very
powerful Landau theory of phase transitions, which we will review first.

6.1 Landau theory of phase transitions

Landau introduced the concept of the order parameter to describe phase transitions. In this context, an order
parameter is a thermodynamic variable that is zero on one side of the transition and nonzero on the other. In
ferromagnets, the magnetization M is the order parameter. The theory neglects fluctuations, which means that
the order parameter is assumed to be constant in time and space (Ginzburg-Landau theory will go beyond this!).
Landau theory is thus a mean-field theory. Now the appropriate thermodynamic potential can be written as a
function of the order parameter, which we call ∆, and certain other thermodynamic quantities such as pressure
or volume, magnetic field, etc. We will always call the potential the free energy F , but whether it really is a
free energy, a free enthalpy, or something else depends on which quantities are given (pressure vs. volume etc.).
Hence, we write
F = F (∆, T ), (6.1)
where T is the temperature, and further variables have been suppressed. The equilibrium state at temperature
T is the one that minimizes the free energy. Generally, we do not know F (∆, T ) explicitly. Landau’s idea was to
expand F in terms of ∆, including only those terms that are allowed by the symmetry of the system and keeping
the minimum number of the simplest terms required to get nontrivial results.
For example, in an isotropic ferromagnet, the order parameter is the three-component vector M. The free
energy must be invariant under rotations of M because of isotropy. Furthermore, as usual in physics, we assume
F to be a smooth, i.e., differentiable, function of M. Then the leading terms, apart from a trivial constant, are
∼ α M · M + β (M · M)2 + O (M · M)3 . (6.2)


F =
2
Denoting the coefficients by α and β/2 is just convention. α and β are functions of temperature (and pressure
etc.).
What is the corresponding expansion for a superconductor or superfluid? Lacking a microscopic theory,
Ginzburg and Landau assumed based on the analogy with Bose-Einstein condensation that the superfluid part is
described by a single one-particle wave function Ψs (r). They imposed the plausible normalization
Z
2
d3 r |Ψs (r)| = Ns = ns V ; (6.3)

31
Ns is the total number of particles in the condensate. They then chose the complex amplitude ψ of Ψs (r) as the
order parameter, with the normalization
2
|ψ| ∝ ns . (6.4)
Thus the order parameter in this case is a complex number. They thereby neglect the spatial variation of Ψs (r)
on an atomic scale.
The free energy must not depend on the global phase of Ψs (r) because the global phase of quantum states is
not observable. Thus we obtain the expansion
2 β 4 6
F = α |ψ| + |ψ| + O(|ψ| ). (6.5)
2
Odd powers are excluded since they are not differentiable at ψ = 0. If β > 0, which is not guaranteed by symmetry
but is true for superconductors and superfluids, we can neglect higher order terms since β > 0 then makes sure
that F (ψ) is bounded from below. Now there are two cases:
• If α ≥ 0, F (ψ) has a single minimum at ψ = 0. Thus the equilibrium state has ns = 0. This is clearly a
normal metal (nn = n) or a normal fluid.
• If α < 0, F (ψ) has a ring of minima with equal amplitude (modulus) |ψ| but arbitrary phase. We easily see
r
∂F 3 α
= 2α |ψ| + 2β |ψ| = 0 ⇒ |ψ| = 0 (this is a maximum) or |ψ| = − (6.6)
∂ |ψ| β
Note that the radicand is positive.
F
α>0 α=0

α<0

0 Re ψ

α /β

F (ψ) for α < 0 is often called the Mexican-hat or wine-bottle potential (imagine this figure rotated around the
vertical axis to find F as a function of the complex ψ). In Landau theory, the phase transition clearly occurs
when α = 0. Since then T = Tc by definition, it is useful to expand α and β to leading order in T around Tc .
Hence,
α∼
= α0 (T − Tc ), α0 > 0, (6.7)
β∼
= const. (6.8)
Then the order parameter below Tc satisfies
s s
α0 (T − Tc ) α0 p
|ψ| ∼
= − = T − Tc . (6.9)
β β

Tc T

32
The scaling |ψ| ∝ (T − Tc )1/2 is characteristic for mean-field theories.
All solutions with this value of |ψ| minimize the free energy. In a given experiment, only one of them is
realized. This equilibrium state has an order parameter ψ = |ψ| eiφ with some fixed phase φ. This state is clearly
not invariant under rotations of the phase. We say that the global U(1) symmetry of the system is spontaneously
broken since the free energy F has it but the particular equilibrium state does not. It is called U(1) symmetry
since the group U(1) of unitary 1 × 1 matrices just contains phase factors eiφ .
Note that the expansion of F up to fourth order is only justified as long as ψ is small. Its quantitative validity
is thus limited to temperatures not too far below Tc . However, one can often gain insight into the qualitative
behavior of a system by applying Landau theory outside of this range.

Specific heat
Since we now know the mean-field free energy as a function of temperature, we can calculate further thermody-
namic variables. In particular, the free entropy is
∂F
S=− . (6.10)
∂T
Since the expression for F used above only includes the contributions of superconductivity or superfluidity, the
entropy calculated from it will also only contain these contributions. For T ≥ Tc , the mean-field free energy is
F (ψ = 0) = 0 and thus we find S = 0. For T < Tc we instead obtain
α2 1 α2 ∂ α2
r   
∂ α ∂
S=− F − =− − + =
∂T β ∂T β 2 β ∂T 2β
∼ ∂ (α0 )2 (α0 )2 (α0 )2
= (T − Tc )2 = (T − Tc ) = − (Tc − T ) < 0. (6.11)
∂T 2β β β
We find that the entropy is continuous at T = Tc . By definition, this means that the phase transition is continuous,
i.e., not of first order. The corresponding heat capacity of the superconductor or superfluid is
∂S
C=T , (6.12)
∂T
which equals zero for T ≥ Tc but is
(α0 )2
C= T (6.13)
β
for T < Tc . Thus the heat capacity has a jump discontinuity of
(α0 )2
∆C = − Tc (6.14)
β
at Tc . The phase transition is of second order since a second derivative of a thermodynamic potential (F ) with
respect to its natural variables (T ) is the lowest discontinuous one. Adding the other contributions, which are
analytic at Tc , the specific heat c := C/V is sketched here:
c

tion
co n tribu
al
norm
Tc T
The enhanced specific heat can be understood as being due to the “soft” phase fluctuations in the wine-bottle
potential. Recall that Landau theory only works close to Tc . In particular, for the specific heat it does not give
reasonable results for T  Tc .

33
6.2 Ginzburg-Landau theory for neutral superfluids
To be able to describe also spatially nonuniform situations, Ginzburg and Landau had to go beyond the Landau
description for a constant order parameter. To do so, they included gradients. We will first discuss the simpler
case of a superfluid of electrically neutral particles (think of He-4). We define a macroscopic condensate wave
function ψ(r), which is essentially given by Ψs (r) averaged over length scales large compared to atomic distances.
We expect any spatial changes of ψ(r) to cost energy—this is analogous to the energy of domain walls in magnetic
systems. In the spirit of Landau theory, Ginzburg and Landau included the simplest term containing gradients
of ψ and allowed by symmetry into the free energy
Z  
2 β 4
F [ψ] = d3 r α |ψ| + |ψ| + γ (∇ψ)∗ · ∇ψ , (6.15)
2

where we have changed the definitions of α and β slightly. We require γ > 0 so that the system does not
spontaneously become highly nonuniform. The above expression is a functional of ψ, also called the Landau or
Ginzburg-Landau functional. Calling it a “free energy” is really an abuse of language since the free energy proper
is only the value assumed by F [ψ] at its minimum.
If we interpret ψ(r) as the (coarse-grained) condensate wave function, it is natural to identify the gradient
term as a kinetic energy by writing
" #
Z 2
β 1 ~
F [ψ] ∼
2 4
3
= d r α |ψ| + |ψ| + ∇ψ , (6.16)
2 2m∗ i

where m∗ is an effective mass of the particles forming the condensate. Evidently, the two expressions for F [ψ]
are related by γ = ~2 /2m∗ .
From F [ψ] we can derive a differential equation for the function ψ(r) that minimizes F . The derivation is
very similar to the derivation of the Lagrange equation (of the second kind) from Hamilton’s principle δS = 0
known from classical mechanics. Here, we start from the extremum principle δF = 0. We write

ψ(r) = ψ0 (r) + η(r), (6.17)

where ψ0 (r) is the as yet unknown solution and η(r) is a small deviation from it. Then

F [ψ0 + η] = F [ψ0 ]
~2 ~2
Z  
+ d3 r αψ0∗ η + αη ∗ ψ0 + βψ0∗ ψ0∗ ψ0 η + βψ0∗ η ∗ ψ0 ψ0 + (∇ψ 0 )∗
· ∇η + (∇η)∗
· ∇ψ0
2m∗ 2m∗
+ O(η, η ∗ )2
by parts
= F [ψ0 ]
~2 ~2 ∗ 2
Z  
+ d3 r αψ0∗ η + αη ∗ ψ0 + βψ0∗ ψ0∗ ψ0 η + βψ0∗ η ∗ ψ0 ψ0 − (∇ 2
ψ 0 ) ∗
η − η ∇ ψ 0
2m∗ 2m∗
+ O(η, η ∗ )2 . (6.18)

If ψ0 (r) minimizes F , the terms linear in η and η ∗ must vanish for any η(r), η ∗ (r). Noting that η and η ∗ are
linearly independent, this requires the prefactors of η and of η ∗ to vanish for all r. Thus we conclude that

~2 2
αψ0 + β ψ0∗ ψ0 ψ0 − ∇ ψ0 = 0. (6.19)
| {z } 2m∗
= |ψ0 |2

Dropping the subscript “0” and rearranging terms we find

~2 2
− ∇2 ψ + αψ + β |ψ| ψ = 0. (6.20)
2m∗

34
This equation is very similar to the time-independent Schrödinger equation but, interestingly, it contains a
nonlinear term. It is also called the (time-independent) Gross-Pitaevskii equation, in particular when derived
from a microscopic description of an interacting superfluid. We now apply it to find the variation of ψ(r) close
to the surface of a superfluid filling the half space x > 0. We impose the boundary condition ψ(x = 0) = 0 and
assume that the solution only depends on x. Then

~2 00 2
− ψ (x) + αψ(x) + β |ψ(x)| ψ(x) = 0. (6.21)
2m∗
Since all coefficients are real, the solution can be chosen real. For x → ∞ we should obtain the uniform solution
r
α
lim ψ(x) = − . (6.22)
x→∞ β

Writing r
α
ψ(x) = − f (x) (6.23)
β
we obtain
~2
− ∗ f 00 (x) + f (x) − f (x)3 = 0. (6.24)
| 2m
{z α}
>0

This equation contains a characteristic length scale ξ > 0 with

~2 ∼ ~2
ξ 2 := − = > 0, (6.25)
2m∗ α 2m∗ α0 (Tc − T )

which is called the Ginzburg-Landau coherence length. It is not the same quantity as the Pippard coherence length
ξ0 , which is often also denoted by ξ. The Ginzburg-Landau ξ has a strong temperature dependence and actually
diverges at T = Tc , whereas the Pippard ξ has at most a weak temperature dependence. Microscopic BCS theory
reveals how the two quantities are related.
Equation (6.24) can be solved analytically. It is easy to check that
x
f (x) = tanh √ (6.26)
2ξ

is a solution satisfying the boundary conditions at x = 0 and x → ∞. (In the general, three-dimensional case, the
solution can only be given in terms of Jacobian elliptic functions.) The one-dimensional tanh solution is sketched
here:
ψ(x)

α
β

0 x
ξ

35
Fluctuations for T > Tc
So far, we have only considered the state ψ0 of the system that minimizes the Landau functional F [ψ]. This is
the mean-field state. At nonzero temperatures, the system will fluctuate about ψ0 . For a bulk system we have

ψ(r, t) = ψ0 + δψ(r, t), (6.27)

with uniform ψ0 . We consider the cases T > Tc and T < Tc separately.

For T > Tc , the mean-field solution is just ψ0 = 0 so that ψ(r, t) = δψ(r, t). We now write the partition
function Z as a sum of Boltzmann factors containing the Landau functional of all possible states,
Z
Z = D2 ψ e−F [ψ]/kB T . (6.28)

This equation requires some discussion. We are obviously using F [ψ] as if it were the Hamiltonian of the field
ψ(r, t), while we have introduced it as a generalization of a Landau “free energy.” This apparent inconsistency is
resolved by recalling that ψ(r, t) is the macroscopic condensate wave function. We have said above that ψ(r, t) is
averaged over length scales large compared to atomic distances. More precisely, fluctuations of the wave function
on shorter length scales have been integrated out (in a renormalization-group sense) so that F [ψ] can indeed be
understood as a free energy with respect to these small-scale fluctuations but it is the energy for the longer-scale
fluctuations we are considering here. This has been explained nicely by Ma in his book “Modern Theory of Critical
Phenomena” (Benjamin, Reading, 1976).
The notation D2 ψ in Eq. (6.28) expresses that the integral is over infinitely many complex variables, namely
the values of ψ(r) for all r. This means that Z is technically a functional integral. The mathematical details
go beyond the scope of this course but are not essential for understanding the physics. The integral is difficult
to evaluate. A common approximation is to restrict F [ψ] to second-order terms, which is reasonable for T > Tc
4
since the fourth-order term (β/2) |ψ| is not required to stabilize the theory (i.e., to make F → ∞ for |ψ| → ∞).
This is called the Gaussian approximation. It allows Z to be evaluated by Fourier transformation: we start from
Z   ∗ 
∼ 1 ~ ~
3 ∗
F [ψ] = d r αψ (r)ψ(r) + ∇ψ(r) · ∇ψ(r) (6.29)
2m∗ i i
and insert
1 X ik·r
ψ(r) = √ e ψk , (6.30)
V k
which gives
Z  
∼ 1 X 3 −ik·r+ik0 ·r ∗ 1 ∗ 0
F [ψ] = d re αψk ψk0 + (~kψk ) · ~k ψk0
V 2m∗
kk0 | {z }
V δkk0
X ~2 k 2 ∗

∗
= αψk ψk + ψ ψk
2m∗ k
k
X ~2 k 2

= α+ ψk∗ ψk . (6.31)
2m∗
k

Thus
!
~2 k 2
Z Y   
1 X
Z∼
= 2
d ψk exp − α+ ∗
ψk ψk
kB T 2m∗
k k
YZ ~2 k 2
   
2 1 ∗
= d ψk exp − α+ ψk ψk
C kB T 2m∗
k
YZ ~2 k 2
   
1
= 2
d ψ exp − α+ ∗
ψ ψ . (6.32)
C kB T 2m∗
k

36
The integral is now of Gaussian type (hence “Gaussian approximation”) and can be evaluated exactly:
Y πkB T
Z∼
= 2 2 . (6.33)
k
α + ~2mk∗

From this, we can obtain the thermodynamic quantities. For example, the heat capacity is

∂S ∂2F ∂2
C=T = −T 2
=T kB T ln Z
∂T ∂T ∂T 2
∼ ∂2 X πkB T
= kB T T ln 2 2 . (6.34)
∂T 2 α + ~2mk∗
k

We only consider the term that is singular at Tc . It stems from the temperature dependence of α ∼
= α0 (T − Tc ),
not from the explicit factors of T . This term is
2 X
~2 k 2
 
2 ∂
Ccrit ∼
= −k B T ln α +
∂T 2 2m∗
k
∂ X α0
= −kB T 2 2 2
∂T
k
α + ~2mk∗
X (α0 )2
= kB T 2 2 2
2
k α + ~2mk∗
 ∗ 2
2m X 1
= kB T 2
(α 0 2
) ∗ α
2
. (6.35)
~2 k + 2m
2
k ~2

Going over to an integral over k, corresponding to the thermodynamic limit V → ∞, we obtain

2
2m∗ d3 k
 Z
∼ 0 2 1
Ccrit = kB T 2 (α ) V
~2 (2π)3 k 2 + 2m2∗ α 2


~
kB T 2 (m∗ )2 0 2 1
= (α ) V q
2π ~4 2m∗ α
~2
2 ∗ 3/2
kB T (m ) 1
= √ (α0 )2 V √
2 2π ~3 α
1
∝√ . (6.36)
T − Tc

The exponent −1/2 of the specific heat is characteristic for Gaussian fluctuations.1 Recall that √ at the mean-field
level, C just showed a step at Tc . Including fluctuations, we obtain a divergence of the form 1/ T − Tc for T → Tc
from above. This is due to superfluid fluctuations in the normal state. The system “notices” that superfluid states
exist at relatively low energies, while the mean-field state is still normal.
We note for later that the derivation has shown that for any k we have two fluctuation modes with dispersion

~2 k 2
k = α + . (6.37)
2m∗
The two modes correspond for example to the real and the imaginary part of ψk . Since α = α0 (T − Tc ) > 0, the
dispersion has an energy gap α.
√
1 Note that the proportionality to (2m∗ α/~2 )−1/2 ∝ 1/ α already follows from dimensional analysis for the momentum integral.

37
 εk

0 kx

In the language of field theory, one also says that the superfluid above Tc has two degenerate massive modes,
with the mass proportional to the energy gap.

Fluctuations for T < Tc

Below pTc , the situation is a bit more complex due to the wine-bottle potential: All states with amplitude
|ψ0 | = −α/β are equally good mean-field solutions.
F
Im ψ

Re ψ

ring−shaped
minimum
It is plausible that fluctuations of the phase have low energies since global changes of the phase do not increase
F . To see this, we write
ψ = (ψ0 + δψ) eiφ , (6.38)
where ψ0 and δψ are now real. The Landau functional becomes
Z 
∼ β
F [ψ] = d r α(ψ0 + δψ)2 + (ψ0 + δψ)4
3
2
 ∗  
1 ~ ~
+ (∇δψ)e iφ
+ ~(ψ 0 + δψ)(∇φ)e iφ
· (∇δψ)e iφ
+ ~(ψ0 + δψ)(∇φ)eiφ
. (6.39)
2m∗ i i
As above, we keep only terms up to second order in the fluctuations (Gaussian approximation), i.e., terms
proportional to δψ 2 , δψ φ, or φ2 . We get
Z "  ∗
∼ 3
((
((3 δψ 2 3 2 1 ~ ~
F [δψ, φ] = d r ( 2αψ
(0 δψ
(( +( 2βψ0 + αδψ + 3βψ0 δψ + ∗
∇δψ · ∇δψ
2m i i
 ∗ ! #
1 ~ ∗ ~ 1 ∗
+ ∇δψ · ~ψ0 ∇φ + (~ψ0 ∇φ) · ∇δψ + (~ψ0 ∇φ) · ~ψ0 ∇φ (+ const). (6.40)
2m∗ i i 2m∗
| {z }
=0

Note that the first-order terms cancel since we are expanding about a minimum. Furthermore, up to second order
there are no terms mixing amplitude fluctuations δψ and phase fluctuations φ. We can simplify the expression:
∗
~2 α
Z   
∼ 3 2 1 ~ ~ ∗
F [δψ, φ] = d r −2αδψ + ∇δψ · ∇δψ − (∇φ) ∇φ
2m∗ i i 2m∗ β

38
 X  ~2 k 2 α ~2 k 2 ∗
 
= −2α + δψk∗ δψk − φ φ k , (6.41)
|{z} 2m∗ β 2m∗ k
k >0 | {z }
>0

in analogy to the case T > Tc . We see that amplitude fluctuations are gapped (massive) with an energy gap
−2α = 2α0 (Tc − T ). They are not degenerate. They are called amplitude modes.
On the other hand, phase fluctuations are ungapped (massless) with quadratic dispersion

α ~2 k 2
φk = − . (6.42)
β 2m∗
The appearance of ungapped so-called Goldstone modes is characteristic for systems with spontaneously
√ broken
continuous symmetries. We state without derivation that the heat capacity diverges like C ∼ 1/ Tc − T for
T → Tc− , analogous to the case T > Tc .

Time-dependent Ginzburg-Landau theory

The Ginzburg-Landau approach can be extended to describe dynamics. The field ψ(r, t) is then time dependent.
The relevant quantity is the Lagrange functional L[ψ, ∇ψ, ψ̇] of this field, from which one obtains the action
Z tf Z tf Z
S= dt L = dt d3 r L, (6.43)
ti ti

where ti and tf are initial and final times, respectively, and L is the Lagrange density. Equations of motion for
ψ(r, t) follow from the variational principle δS = 0, in analogy to the Lagrange equations in classical mechanics
and, even more appositely, the Maxwell equations in electrodynamics.
We only give the main expressions for the case of a neutral superfluid, without derivations. Here,

∂ψ β ~2
L = i~ ψ ∗ − α |ψ|2 − |ψ|4 − |∇ψ|2 (6.44)
∂t 2 2m∗
so that the Lagrange functional reads Z
∂ψ
L= d3 r i~ ψ ∗ − F [ψ], (6.45)
∂t
with the Landau functional F . From δS = 0, one then obtains (see the lecture notes on Quantum Theory 2)

∂ψ ~2
i~ = − ∗ ∇2 ψ + α ψ + β |ψ|2 ψ. (6.46)
∂t 2m
This equation is consistent with the time-independent limit, Eq. (6.20). It is clearly a nonlinear analogue of the
time-dependent Schrödinger equation. It is also called the time-dependent Gross-Pitaevskii equation. The result
makes sense since for a noninteracting field (β = 0) we obtain the single-particle Schrödinger equation.

6.3 Ginzburg-Landau theory for superconductors

To describe superconductors, we have to take the charge q of the particles forming the condensate into account.
We allow for the possibility that q is not the electron charge −e. Then there are two additional terms in the
Landau functional:
• The canonical momentum has to be replaced by the kinetic momentum (minimal coupling):

~ ~ q
∇ → ∇ − A, (6.47)
i i c
where A is the vector potential.

39
 • The energy density of the magnetic field, B 2 /8π, has to be included.2
If the charge carriers (later to be identified as Cooper pairs) had a nonvanishing spin, there would be an additional
term proportional to B · S. We consider the zero-spin case here, which applies to all conventional and most
unconventional superconductors. Thus we obtain the functional
"  2 #
B2
Z 
∼ 3 2 β 4 1 ~ q
F [ψ, A] = d r α |ψ| + |ψ| + ∇− A ψ +
2 2m∗ i c 8π
∗ 
B2
Z    ∗  ~  
3 2 β 4 1 ~ ~ q 1  q q
= d r α |ψ| + |ψ| + ∇ψ · ∇− A ψ+ − Aψ · ∇− A ψ+
2 2m∗ i i c 2m∗ c i c 8π
∗
B2
Z        
3 2 β 4 1 ∗ ~ ~ q 1 ∗
 q  ~ q
= d r α |ψ| + |ψ| − ψ ∇ · ∇− A ψ+ ψ − A · ∇− A ψ+
2 2m∗ i i c 2m∗ c i c 8π
" 2 #
2
Z 
2 β 4 1 ~ q B
= d3 r α |ψ| + |ψ| + ψ∗ ∇− A ψ+ . (6.48)
2 2m∗ i c 8π

Minimizing this free-energy functional with respect to ψ, we obtain, in analogy to the previous section,
 2
1 ~ q 2
∇ − A ψ + α ψ + β |ψ| ψ = 0. (6.49)
2m∗ i c
To minimize F with respect to A, we write A(r) = A0 (r) + a(r) and inset this into the terms containing A:
Z    ∗ 
3 1 ~ q q
F [ψ, A0 + a] = F [ψ, A0 ] + d r ∇ − A 0 ψ · − aψ
2m∗ i c c
  
1  q ∗ ~ q 1
+ − aψ · ∇ − A0 ψ + (∇ × A0 ) · (∇ × a) + O(a2 )
2m∗ c i c 4π | {z }
= a·(∇×∇×A0 ) − ∇·((∇×A0 )×a)
Z   ∗ 
q ~ ∗~
= F [ψ, A0 ] + d3 r − ∇ψ ψ + ψ ∇ψ ·a
2m∗ c i i
q2

2 1
+ ∗ 2 |ψ| A0 · a + (∇ × B0 ) · a + O(a2 ), (6.50)
m c 4π
where we have used ∇ × A0 = B0 and Gauss’ theorem. At the minimum, the coefficient of the linear term must
vanish. Dropping the subscript, we obtain
q~ q2 2 1
[∇ψ ∗ ]ψ − ψ ∗ ∇ψ + ∗ 2 |ψ| A + (6.51)


−i ∗
∇ × B = 0.
2m c m c 4π
With Ampère’s law we find
c q~ q2 2
∇ × B = i ∗ [∇ψ ∗ ]ψ − ψ ∗ ∇ψ − ∗ |ψ| A, (6.52)


j=
4π 2m m c
where we have dropped the subscript “s” of j since we assume that the normal current is negligible. Equations
(6.49) and (6.52) are called the Ginzburg-Landau equations.
In the limit of uniform ψ(r), Eq. (6.52) simplifies to
2
q 2 |ψ|
j=− A. (6.53)
m∗ c
This should reproduce the London equation
e2 ns
j=− A, (6.54)
mc
2 In principle, one also has to include the energy density of the electric field E but E vanishes in equilibrium in any conductor.

40
which obviously requires
2
q 2 |ψ| e2 ns
= . (6.55)
m∗ m
As noted in Sec. 5.3, based on flux-quantization experiments and on the analogy to Bose-Einstein condensation,
it is natural to set q = −2e. For m∗ one might then insert twice the effective electron (band) mass of the metal.
However, it turns out to be difficult to measure m∗ independently of the superfluid density ns and it is therefore
common to set m∗ = 2m ≡ 2me . All system-specific properties are thus absorbed into
2 2
m q 2 |ψ| m ψ e2 |ψ| 2
ns = = = 2 |ψ| . (6.56)
e2 m ∗ e2 2m
With this choice of normalization of ψ(r), we have shown that London theory follows as a special case from
Ginzburg-Landau theory.
We can now write the penetration depth as
s s s
mc2 mc2 ∼ mc2
λ= = 2 =
. (6.57)
2
4πe ns 8πe2 |ψ| 8πe2 − α
β

We have found two characteristic length scales:

• λ: penetration of the magnetic field,
• ξ: variation of the coarse-grained superconducting wave function (order parameter).
Within the mean-field theories we have so far employed, both quantities scale as
1
λ, ξ ∝ √ (6.58)
Tc − T
close to Tc . Thus their dimensionless ratio
λ(T )
κ := (6.59)
ξ(T )
is roughly temperature independent. It is called the Ginzburg-Landau parameter and turns out to be very
important for the behavior of superconductors in an applied magnetic field. For elemental superconductors, κ is
small compared to unity, i.e., the B field is less stiff than the wave function.

Fluctuations and the Anderson-Higgs mechanism

When discussing fluctuations about the mean-field state of a superconductor, we have to take the coupling to the
electromagnetic field into account. We proceed analogously to the case of a neutral superfluid and employ the
Gaussian approximation throughout. For T > Tc , we note that the kinetic term
  2
1 ~ q
∇− A ψ (6.60)
2m∗ i c

is explicitly of second order in the fluctuations δψ = ψ so that electromagnetic-field fluctuations appear only
in higher orders. Thus to second order, the order-parameter fluctuations decouple from the electromagnetic
fluctuations,
" #
2 X B∗ · Bk
B2 X ~2 k 2
Z 
∼ 1 ~
3 ∗
F [ψ, B] = d r αψ (r)ψ(r) + ∇ψ + = α + ψ ∗
k ψ k + k
. (6.61)
2m∗ i 8π 2m∗ 8π
k k

The superconducting fluctuations consist of two degenerate massive modes with dispersion k = α + ~2 k 2 /2m∗ ,
like for the neutral superfluid. The electromagnetic-field fluctuations decouple and are those of a free field.

41
 The case T < Tc is more interesting. Writing, as above, ψ = (ψ0 + δψ) eiφ , we obtain to second order
Z "  ∗
∼ d r 2αψ
3 ((3 δψ (( 2 2 2 1 ~ ~
F [δψ, φ, A] = ((0 δψ(+( 2βψ 0 + αδψ + 3βψ 0 δψ + ∇δψ · ∇δψ
2m∗ i
(
i
 ∗ ( ( (!
1 ~ ((∇φ)((∗(~ 1 ∗
+ ∗
∇δψ · ~ψ 0(∇φ( +((~ψ 0 · ∇δψ + (~ψ0 ∇φ) · ~ψ0 ∇φ
2m (i (( ( ( ( i 2m∗
(
 (
 ∗  (( (((
1  ~ q  ( q ( ∗ ~
+ ∇δψ · − ( (0(+((
Aψ − Aψ0 · ∇δψ 
2m∗ i (((((( c c i
( (
1   q   q  ∗ 
∗
+ (~ψ 0 ∇φ) · − Aψ 0 + − Aψ 0 · ~ψ 0 ∇φ
2m∗ c c #
1  q ∗  q  1
+ − Aψ0 · − Aψ0 + (∇ × A) · (∇ × A) (+ const)
∗
2m∗ c c 8π
Z   ∗
1 ~ ~
= d3 r − 2αδψ 2 + ∗
∇δψ · ∇δψ
2m i i
2

~ α  q  ∗  q  1
− ∇φ − A · ∇φ − A + (∇ × A) ∗
· (∇ × A) . (6.62)
2m∗ β ~c ~c 8π
Note that the phase φ of the macroscopic wave function and the vector potential appear in the combination
∇φ − (q/~c)A. Physical properties are invariant under the gauge transformation

A → A + ∇χ, (6.63)
1
Φ → Φ − χ̇, (6.64)
c
ψ → eiqχ/~c ψ, (6.65)

where Φ is the scalar electric potential and χ(r, t) is an arbitrary scalar field. We make use of this gauge invariance
by choosing
~c
χ = − φ. (6.66)
q
Under this tranformation, we get
~c
A → A− ∇φ =: A0 , (6.67)
q
ψ = (ψ0 + δψ) eiφ → (ψ0 + δψ) e−iφ+iφ = ψ0 + δψ. (6.68)

The macroscopic wave function becomes purely real (and positive). The Landau functional thus tranforms into
Z   ∗
1 ~ ~
F [δψ, φ, A] → F [δψ, A0 ] ∼
= d3 r − 2αδψ 2 + ∇δψ · ∇δψ
2m∗ i i
2

α q 0 ∗ 1
− (A ) · A 0
+ (∇ × A 0 ∗
) · (∇ × A0
) (6.69)
β 2m∗ c2 8π

(note that ∇×A0 = ∇×A). Thus the phase no longer appears in F ; it has been absorbed into the vector potential.
Furthermore, dropping the prime, we obtain
X  ~2 k 2 α q2
 
∼ ∗ ∗ 1 ∗
F [δψ, A] = −2α + δψk δψk − A · Ak + (k × Ak ) · (k × Ak )
2m∗ β 2m∗ c2 k 8π
k
X  ~2 k 2 α q2 k2 ∗
 
1
= −2α + ∗
δψk δψk − ∗
A · Ak + A · Ak − ∗
(k · Ak )(k · Ak ) . (6.70)
2m∗ β 2m∗ c2 k 8π k 8π
k

42
Obviously, amplitude fluctuations decouple from electromagnetic fluctuations and behave like for a neutral su-
perfluid. The amplitude modes are now called Higgs modes.
We discuss the electromagnetic fluctuations further. The term proportional to α/β is due to superconductivity.
Without it, we would have the free-field functional
X 1 
k 2 A∗k · Ak − (k · A∗k )(k · Ak ) . (6.71)

Ffree [A] =
8π
k

Decomposing A into longitudinal and transverse components,

Ak = k̂(k̂ · Ak ) + Ak − k̂(k̂ · Ak ) (6.72)

| {z } | {z }
=: Ak
k =: A⊥
k

with k̂ := k/k, we obtain

X 1 h k∗ k

k∗

k
i
Ffree [A] = k 2 Ak · Ak + k 2 A⊥∗ ⊥
k · Ak − k · Ak k · Ak
8π
k
X 1 h i X 1
2 k∗ k 2 k∗ k
2 ⊥∗ ⊥
k 2 A⊥∗ ⊥
(6.73)
 
= k A  ·
 A + k A k · A k − k A  ·
 A k = k · Ak .
8π  k k  k
 
8π
k k

Thus only the two transverse components appear—only they are degrees of freedom of the free electromagnetic
field.3 They do not have an energy gap.
Including the superconducting contribution, we get
X  α q2 k∗ k α q2 1 2 ⊥∗

∼
F [A] = − A · Ak − ⊥∗ ⊥
A · Ak + ⊥
k Ak · Ak . (6.74)
β 2m∗ c2 k β 2m∗ c2 k 8π
k

All three components of A appear now, the longitudinal one has been introduced by absorbing the phase φ(r).
Even more importantly, all components obtain a term with a constant coefficient, i.e., a mass term. Thus the
electromagnetic field inside a superconductor becomes massive. This is the famous Anderson-Higgs mechanism.
The same general idea is also thought to explain (most of) the masses of elementary particles, although in a more
complicated case (the broken symmetry is more complicated than the U(1) symmetry of superconductivity). The
“Higgs bosons” in our case are the amplitude-fluctuation modes described by δψ. Contrary to what is said in
popular discussions, they are not responsible for giving mass to the field A. Rather, they are left over when the
phase fluctuations are eaten by the field A.
The mass term in the superconducting case can be thought of as leading to the Meißner effect (finite penetration
depth λ). Indeed, we can write
X ~2 k 2
 X 1 1 
F [δψ, A] ∼
= −2α + δψk
∗
δψk + A ∗
· A k + k 2 ∗
A · Ak − (k · A ∗
)(k · A k ) . (6.75)
2m∗ 8π λ2 k k k
k k

The photon mass is proportional to 1/λ. (To see that the dispersion relation is ~2 ω 2 = m2 c4 + p2 c2 , we would
have to consider the full action including a temporal integral over F and terms containing time-derivatives.)

Elitzur’s theorem
One sometimes reads that in the superconducting state gauge symmetry is broken. This is not correct. Gauge
symmetry is the invariance under local gauge transformations. S. Elitzur showed in 1975 that a local gauge
symmetry cannot be spontaneously broken. Rather, superconductors and superfluids spontaneously break a
global U(1) symmetry in that the ordered state has a prefered macroscopic phase, as noted above.
3 This result persists in quantum electrodynamics: since the electromagnetic field is a spin-one field one would naively expect it

to have three independent degrees of freedom at each momentum, corresponding to the three magnetic quantum numbers −1, 0, +1.
However, for a massless field, the longitudinal component is not a dynamical field (it is fully determined by the charges) and the
electromagnetic field thus only has two degrees of freedom per momentum. These can be described as photons with positive and
negative helicity, i.e., with spin parallel or antiparallel to their momentum, respectively.

43
6.4 Type-I superconductors
Superconductors with small Ginzburg-Landau parameter κ = λ/ξ are said to be of type I. The exact condition is
1
κ< √ . (6.76)
2
It turns out that these superconductors have a uniform state in an applied magnetic field, at least for simple
geometries such as a thin cylinder parallel to the applied field.
The appropriate thermodyamic potential for describing a superconductor in an applied magnetic field is the
Gibbs free energy G (natural variable H) and not the Helmholtz free energy F (natural variable B), which we
have used so far. The reason is that the magnetic field H, not the magnetic induction B, is controlled by the
experimenter. The two potentials are related by the Legendre transformation
H·B
Z
G = F − d3 r . (6.77)
4π

Since the equilibrium state in the bulk is uniform the order parameter is |ψ| = −α/β and the magnetic induction
p

B as well as the vector potential (in the London gauge) vanish. Thus the Gibbs free-energy density is

2 β 4 α2 β α2 α2
gs = fs = α |ψ| + |ψ| = − + = − . (6.78)
2 β 2 β2 2β
On the other hand, in the normal state ψ vanishes, but the field penetrates the system and B ≡ H so that
HB B2 HB H2
gn = fn − = − =− . (6.79)
4π 8π 4π 8π
The system will only become superconducting if this reduces the Gibbs free energy, i.e., if
α2 H2
gs − gn = − + < 0. (6.80)
2β 8π
Thus in an applied magnetic field of H ≥ Hc , with the critical field
s
α2
Hc (T ) := 4π , (6.81)
β

superconductivity does not occur.

We can use the relation (6.81) and

mc2 mc2 β
λ2 (T ) = 2 =− (6.82)
8πe2 |ψ| 8πe2 α

to express the phenomenological parameters α and β in terms of the measurable quantities λ and Hc :
2e2 2 2
α=− λ Hc , (6.83)
mc2
 2 2
2e
β = 4π λ4 Hc2 . (6.84)
mc2

Domain-wall energy and intermediate states

We can now calculate the energy per area of a domain wall between superconducting and normal regions. We
assume that ψ(r) is only a function of x and impose the boundary conditions
(
ψ∞ := −α/β for x → ∞,
p
ψ(x) → (6.85)
0 for x → −∞.

44
What are reasonable boundary conditions for the local induction B(x)? In the superconductor we have

B(x) → 0 for x → ∞. (6.86)

At the other end, if we have limx→−∞ B(x) > Hc the bulk superconductor has a higher Gibbs free-energy density
than the normal conductor, as seen above. Thus superconductivity is not stable. For limx→−∞ B(x) < Hc , the
Gibbs free-energy density of the superconductor is lower and it eats up the normal phase. Thus a domain wall
can only be stable for s
α2
B(x) → Hc = 4π for x → −∞, (6.87)
β
which we will assume from now on. This obviously requires fine tuning that cannot be done exactly in the real
world. Nevertheless, we can get useful insight since domain walls can occur, without fine tuning, for less trivial
geometries, as we will see.
Technically, we have to solve the Ginzburg-Landau equations under these boundary conditions, which can
only be done numerically. The qualitative form of the solution is clear, though:

B(x) ~λ
ψ(x)

0 x
~ξ

The Gibbs free energy of the domain wall, per unit area, will be denoted by γ. To derive it, first note that for the
given boundary conditions, gs (x → ∞) = gn (x → −∞), i.e., the superconducting free-energy density deep inside
the superconductor equals the normal free-energy density deep inside the normal conductor. This bulk Gibbs
free-energy density is, as derived above,

Hc2 H2
gs (x → ∞) = gn (x → −∞) = fn (x → −∞) − =− c. (6.88)
4π 8π
The additional free-energy density due to the domain wall is

H2 H2
 
gs (x) − − c = gs (x) + c . (6.89)
8π 8π

The corresponding free energy per area is

Z∞
Hc2
 
γ= dx gs (x) +
8π
−∞
Z∞ z }| {
H2

B(x)Hc
= dx fs (x) − + c using that H(x) = const = Hc
4π 8π
−∞
Z∞ 2
B2 H2
   
2 β 4 1 ~ ∂ q BHc
= dx α |ψ| + |ψ| + x̂ − A(x) ŷ ψ + − + c , (6.90)
2 2m∗ i ∂x c |8π 4π 8π}
−∞
{z
= (B−Hc )2 /8π

where we have used B = ẑ B(x) and the gauge choice A = ŷ A(x), which requires B = ∂A/∂x.

45
 We can simplify this expression by multiplying the first Ginzburg-Landau equation (6.49) by ψ ∗ and integrating
over x:
Z∞ "  2 #
2 4 1 ~ ∂ q
0= dx α |ψ| + β |ψ| + ψ∗ x̂ − A(x) ŷ ψ
2m∗ i ∂x c
−∞
Z∞      
by parts 2 4 1 ~ ∂ q ~ ∂ q
= dx α |ψ| + β |ψ| + − x̂ ∗
− A(x) ŷ ψ · x̂ − A(x) ŷ ψ see Eq. (6.48)
2m∗ i ∂x c i ∂x c
−∞
Z∞ "   2#
2 1 4 ~ ∂ q
= dx α |ψ| + β |ψ| + x̂ − A(x) ŷ ψ . (6.91)
2m∗ i ∂x c
−∞

Thus
Z∞ Z∞ " 2 #
(B − Hc )2 Hc2
  
β 4 β 4 B
γ= dx − |ψ| + = dx −β 2 |ψ| + 1 −
2 8π 8π α Hc
−∞ −∞
Z∞ " 2 4
#
Hc2 B |ψ|
= dx 1− − 4
, (6.92)
8π Hc ψ∞
−∞

where we have drawn out the characteristic energy density Hc2 /8π. The domain wall energy is given by the differ-
ence of the energy cost of expelling the magnetic field and the energy gain due to superconducting condensation.
For strong type-I superconductors, ξ  λ, there is a region of thickness ξ − λ > 0 in which the first term is
already large, while the second only slowly approaches its bulk value (see√sketch above). Thus γ > 0 for strong
type-I superconductors. One can show that γ > 0 persists for all κ < 1/ 2. Type-I superconductors therefore
tend to minimize the total area of domain walls.
Note that even in samples of relatively simple shape, magnetic flux will penetrate for nonzero applied field. It
will do so in a way that minimizes the total Gibbs free energy, only one contribution to which is the domain-wall
energy. For example, in a large slab perpendicular to the applied field, the flux has to penetrate in some manner,
since going around the large sample would cost much energy. A careful analysis shows that this will usually
happen in the form of normal stripes separated by superconducting stripes, see Tinkham’s book. Such a state is
called an intermediate state. It should not be confused with the vortex state to be discussed later.

S N S N S N S

6.5 Type-II superconductors

Type-II superconductors are defined by a large Ginzburg-Landau parameter
1
κ> √ . (6.93)
2
The analysis in the previous section goes through. But now domain walls have a region where the condensate is
nearly fully developed but the flux is not completely expelled.

46
 B(x) ψ(x)

~λ

0 x
~ξ

Therefore, the domain-wall energy is negative, γ < 0. Hence, the system tends to maximize the total area of
domain walls. This tendency should be counterbalanced by some other effect, otherwise the system would become
inhomogeneous on microscopic (atomic) length scales and Ginzburg-Landau theory would break down. The effect
in question is flux quantization—the penetrating magnetic flux cannot be split into portions smaller than the flux
quantum Φ0 = hc/2e, see Sec. 5.3.

Fluxoid quantization
We now revisit the quantization of magnetic flux. Consider an arbitrary closed path ∂S forming the edge of a
surface S, where at least the edge ∂S must lie inside a superconductor,
B

0oS
S

The magnetic flux Φ through this loop is

Z Z I
Stokes
Φ = da · B = da · (∇ × A) = ds · A. (6.94)
S S ∂S

With the second Ginzburg-Landau equation

q~ q2 2
j=i ([∇ψ ∗
] ψ − ψ ∗
∇ψ) − |ψ| A (6.95)
2m∗ m∗ c
solved for A, we obtain
( )
m∗ c
I
~c
Φ= ds · − 2 j+i 2 ([∇ψ ∗ ] ψ − ψ ∗ ∇ψ)
q 2 |ψ| 2q |ψ|
∂S
I  
mc ~c
= ds · − 2 j + i (−2i)∇φ
e ns 2q
∂S
I I
mc ~c
= ds · vs − ds · ∇φ, (6.96)
e 2e
∂S ∂S

where vs := −j/ens is the superfluid velocity. The last term contains the change of the phase of the macroscopic
wave function along the whole path, which must be an integer multiple of 2π for ψ(r) to be continuous:
I
ds · ∇φ = −2πn, n ∈ Z. (6.97)
∂S

47
(The minus sign is conventional.) Thus we find for the so-called fluxoid
I
mc ~c hc
0
Φ := Φ − ds · vs = 2πn = n = nΦ0 , n ∈ Z. (6.98)
e 2e 2e
∂S

We see that it is not the flux but the fluxoid Φ0 that is quantized. However, deep inside a superconducting region,
the current vanishes and Φ0 equals Φ.

Vortices
The smallest amount of fluxoid that can penetrate a superconductor is one flux quantum Φ0 . It does so as a
vortex (or vortex line).
vortex core
B

Following the above arguments, the phase φ of ψ changes by −2π as one circles a vortex in the positive direction,
where, by convention, the direction of a vortex is the direction of the magnetic field B. Thus in the center of the
vortex line (the vortex core), the phase is undefined. This is only consistent with a continuous ψ if ψ = 0 in the
vortex core. Because of the phase winding, a vortex cannot end anywhere in the interior of a superconductor.
Vortices can thus only terminate at a surface or form closed loops. One can understand the winding of the phase
as a topological property of the vortex since it cannot be changed by any continuous deformation of ψ(r) unless
ψ is tuned to zero. In this sense, vortices are called topological defects of the superconducting condensate.
For a vortex along the z-axis we choose cylindrical coordinates %, ϕ, z. We then have to solve the Ginzburg-
Landau equations together with Ampère’s law with the boundary conditions

ψ(% = 0) = 0, (6.99)
|ψ(% → ∞)| = ψ0 , (6.100)
B(% → ∞) = 0. (6.101)

From symmetry, we have

ψ(r) = |ψ| (%) eiφ = |ψ| (%) e−iϕ , (6.102)

vs (r) = ϕ̂ vs (%), (6.103)
B(r) = ẑB(%) (6.104)

and we can choose

A(r) = ϕ̂A(%) (6.105)
so that
1 ∂
B(%) = %A(%). (6.106)
% ∂%

48
Choosing a circular integration path of radius r centered at the vortex core, the enclosed fluxoid is
I
mc mc !
Φ0 (r) = Φ(r) − ds · vs = 2πrA(r) − 2πr vs (r) = Φ0
e e
∂S
mc Φ0
⇒ A(r) − vs (r) = . (6.107)
e 2πr
This relation follows from fluxoid quantization and thus ultimately from the second Ginzburg-Landau equation.
To obtain j(r), ψ(r), and A(r), one has to use the first Ginzburg-Landau equation
 2
1 ~ 2e 2
∇+ A ψ + α ψ + β |ψ| ψ = 0 (6.108)
4m i c
and Ampère’s Law
c
j= ∇ × B. (6.109)
4π
2
This cannot be done analytically because of the nonlinear term β |ψ| ψ. For small distances % from the core,
one can drop this term, thereby linearizing the Ginzburg-Landau equation. The solution is still complicated, the
result is that |ψ| increases linearly in %. Numerical integration of the full equations gives the results sketched here
for a cut through the vortex core:

~λ ψ(r)

B(r)

0 x
~ξ

Another useful quantity is the free energy per unit length of a vortex line (its line tension). An analytical
expression can be obtained in the strong type-II case of κ  1. We only give the result here: The vortex line
tension is 2
H2

Φ0
v ≈ ln κ = c 4πξ 2 ln κ. (6.110)
4πλ 8π
We can now calculate the field for which the first vortex enters the superconductor, the so-called lower critical field
Hc1 . This happens when the Gibbs free energy for a superconductor without vortices (Meißner phase) equals the
Gibbs free energy in the presence of a single vortex. We assume the sample to have cross section A and thickness
L, parallel to the applied field.

Then we have the condition

0 = Gone vortex − Gno vortex

49
 Z
1
= s + Lv −
F d3 r H · B − 
F
s+0
4π
Z
Hc1
= Lv − d3 r B(r)
4π
Hc1 L
= Lv − Φ0 . (6.111)
4π
Thus
4πv
Hc1 = . (6.112)
Φ0
The line tension in Eq. (6.110) can also be written as
√ √ Φ0 Hc ξ Φ H ln κ
v = v v ≈ √ ln κ = 0 √ c (6.113)
4πλ 2 4π 2 κ
so that
Hc ln κ
Hc1 = √ . (6.114)
2 κ
Recall that this expression only holds for κ  1, where thus Hc1  Hc . Hc is the thermodynamic critical field
defined above. In a type-II superconductor, nothing interesting happens at H = Hc .

The Abrikosov vortex lattice

We have considered the structure of an isolated vortex line. How does a finite magnetic flux penetrate a type-II
superconductor? Based on Ginzburg-Landau theory, A. A. Abrikosov proposed in 1957 that flux should enter as
a periodic lattice of parallel vortex lines carrying a single flux quantum each. He proposed a square lattice, which
was due to a small mistake. The lowest-free-energy state is actually a triangular lattice.
y
B

x
As noted above, the magnetic flux starts to penetrate the superconductor at the lower critical field Hc1 . Further-
more, since flux expulsion in type-II superconductors is not required to be perfect, they can withstand stronger
magnetic fields than type-I superconductors, up to an upper critical field Hc2 , which is larger than Hc and Hc1 .
H
Hc2 (T )
normal metal
Shubnikov
(vortex)
phase
2n
d

Hc1 (T )
ord
e

2nd
r

ord
Meißner er
phase
Tc T

50
We will now review the basic ideas of Abrikosov’s approach. Abrikosov’s results are quantitatively valid only
close to Hc2 since he assumed the magnetic flux density B to be uniform, which is valid for

λ  l, (6.115)

where r
Φ0
l= (6.116)
B
is the typical distance between vortices (B/Φ0 is the two-dimensional concentration of vortex lines). The magnetic
flux associated with neighboring vortices then strongly overlaps. For B = B ẑ = const ≈ H ẑ (close to Hc2 !) we
can choose the gauge A = ŷ Hx. Then the first Ginzburg-Landau equation becomes (note m∗ = 2m)
 2
1 ~ 2eH 2
∇+ ŷ x ψ + α ψ + β |ψ| ψ = 0. (6.117)
2m∗ i c
Slightly below Hc2 , |ψ| is expected to be small (this should be checked!) so that we can neglect the nonlinear
term. Introducing the cyclotron frequency of a superconductor,
2eH
ωc := , (6.118)
m∗ c
we obtain
~2 2
 
∂ 1 ∗ 2 2
− ∗ ∇ − i~ωc x + m ωc x ψ(r) = −α ψ(r). (6.119)
2m ∂y 2 |{z}
>0

This equation has the same form as the Schrödinger equation for a particle of mass m∗ and charge q = −2e in a
uniform magnetic field H. This well-known problem is solved by the ansatz

ψ(x, y) = eiky y f (x). (6.120)

We obtain
~2
 
∂ 1
e−iky y − ∗ ∇2 − i~ωc x + m∗ ωc2 x2 eiky y f (x)
2m ∂y 2
2 2 2 2
~ d f ~ ky 1
=− ∗ + f − i~ωc x iky f + m∗ ωc2 x2 f
2m dx2 2m∗ 2 !
2 2
~2 d2 f 1 ∗ 2 ~k y ~ k y
=− ∗ + m ωc x 2 + 2 ∗ + f = −αf. (6.121)
2m dx2 2 m ωc (m∗ )2 ωc2

Defining x0 := −~ky /m∗ ωc , we get

~2 d2 f 1
− + m∗ ωc2 (x − x0 )2 f = −αf. (6.122)
2m∗ dx2 2
This is the Schrödinger equation for a one-dimensional harmonic oscillator with shifted minimum. Thus we obtain
solutions fn (x) as shifted harmonic-oscillator eigenfunctions for
 
1
− α = ~ωc n + , n = 0, 1, 2, . . . (6.123)
2
For at least one solution to exist, we need
~ωc ~eH
− α = −α0 (T − Tc ) = α0 (Tc − T ) ≥ = ∗ . (6.124)
2 m c
Keeping T ≤ Tc fixed, a solution can thus only exist for H ≤ Hc2 with the upper critical field
m∗ c
Hc2 = − α. (6.125)
~e

51
Using ξ 2 = −~2 /2m∗ α, we obtain
~c Φ0
Hc2 (T ) = 2
= . (6.126)
2e ξ (T ) 2π ξ 2 (T )
√
Note that since ξ ∼ 1/ Tc − T close to Tc , Hc2 (T ) sets in linearly, as shown in the above sketch.
Hc2 should be compared to the thermodynamic critical field Hc ,
2
mc2 1 ~2 c2 1

1 hc 1
Hc2 = −
2 2
α= =
2e λ 8e2 λ2 ξ 2 8π 2 2e λ2 ξ 2
Φ Hc2 ξ Hc2
⇒ Hc = √0 = √ =√
2π 2 λξ 2λ 2κ
√
⇒ Hc2 = 2 κ Hc . (6.127)
√
For κ = 1/ 2, Hc2 equals Hc . This is the transition between a type-II and a type-I superconductor.
So far, our considerations have not told us what the state for H . Hc2 actually looks like. To find out,
one in principle has to solve the full, not the linearized, Ginzburg-Landau equation. (We have seen that the
linearized equation is equivalent to the Schrödinger equation for a particle in a uniform magnetic field. The
solutions are known: The eigenfunctions have uniform amplitude and the eigenenergies form discrete Landau
levels, very different from what is observed in the Shubnikov phase.) Abrikosov did this approximately using a
variational approach. He used linear combinations of solutions of the linearized equation as the variational ansatz
and assumed that the solution is periodic in two dimensions, up to a plane wave.
We call the periods in the x- and y-directions ax and ay , respectively. The function

ψn (x, y) = eiky y fn (x) (6.128)

has the period ay in y if

2π
ky = q, q ∈ Z. (6.129)
ay
Then the harmonic oscillator is centered at
~ 2π Φ0
x0 = − q=− q. (6.130)
m∗ ωc ay Hay

Since the lowest-energy solution is obtained for n = 0 (the ground state of the harmonic oscillator), Abrikosov
only considered the n = 0 solutions
2 !
1 m∗ ωc
    
2πy 2πy Φ0
ψ0 (x, y) = exp iq f0 (x) = C exp iq exp − x+ q . (6.131)
ay |{z} ay 2 ~ Hay
normalization | {z }
−x0

In the Gauss function we find the quantity

~ ~c ~c ~2
= ≈ = − = ξ 2 (T ), (6.132)
m∗ ωc 2eH 2e Hc2 2m∗ α
as long as H ≈ Hc2 . Thus
   2 !
2πy 1 Φ0
ψ0 (x, y) = C exp iq exp − 2 x + q . (6.133)
ay 2ξ Hay

This is a set of functions enumerated by q ∈ Z. Abrikosov considered linear combinations

∞    2 !
X 2πy 1 Φ0
ψ(x, y) = Cq exp iq exp − 2 x + q . (6.134)
q=−∞
ay 2ξ Hay

52
To be periodic in x with period ax , up to a plane wave (the corresponding discussion in Tinkham’s book is not
fully correct), this ansatz has to satisfy
∞    2 !
X 2πy 1 Φ0
ψ(x + ax , y) = Cq exp iq exp − 2 x + ax + q
q=−∞
ay 2ξ Hay
∞     2 !
X 2πy 1 Φ0 Hax ay
= Cq exp iq exp − 2 x + q+
q=−∞
ay 2ξ Hay Φ0
!
∝ ψ(x, y) ∀ x, y. (6.135)

This requires
Hax ay
=: ν ∈ N. (6.136)
Φ0
(Note that this quantity is positive.) Then
∞    2 !
X 2πy 1 Φ0
ψ(x + ax , y) = Cq exp iq exp − 2 x + (q + ν)
q=−∞
ay 2ξ Hay
  X∞    2 !
2πy 2πy 1 Φ0
= exp iν Cq−ν exp iq exp − 2 x + q . (6.137)
ay q=−∞ ay 2ξ Hay

This equals ψ(x, y) up to a plane-wave factor if

Cq−ν = Cq ∀ q, (6.138)

i.e., if Cq is periodic. Abrikosov considered the case ν = 1, which leads to a square lattice. The lowest-free-energy
solution is obtained for ν = 2 and C1 = iC0 , which gives a triangular lattice, sketched above. Note that
Hax ay
ν= (6.139)
Φ0
has a simple interpretation: It is the number of flux quanta passing through a rectangular unit cell of the vortex
lattice.
The vortex lattice is a rather complex system: It is a lattice of interacting lines with a line tension v . At
nonzero temperatures, the vortices fluctuate, which can lead to the melting of the vortex lattice. The resulting
vortex liquid can be pictured as a pot of boiling spaghetti, with the constraint that the vortex lines must either
terminate at the surface or form closed loops (!). Moving vortices lead to ohmic resistance, even though most
of the sample is still superconducting. Moreover, the interaction of vortices with defects (“pinning”) plays an
important role. There is an extensive research literature on vortex matter, which we cannot review here.

53
7

Superfluid and superconducting films

Generally, fluctuations are stronger in systems of lower dimension. Indeed, they change the key properties of
two-dimensional superfluid and superconducting films qualitatively compared to three-dimensional samples. We
will consider such films within the Ginzburg-Landau theory.

7.1 Superfluid films

We start from the two-dimensional Landau functional
Z  
2 β 4
2 ∗
F [ψ] = d r α |ψ| + |ψ| + γ(∇ψ) · ∇ψ . (7.1)
2

We consider temperatures T < Tc . As shown in Sec. 6.2, fluctuations of the amplitude |ψ| then have an energy
gap, whereas fluctuations of the phase φ are ungapped. Not too close to Tc , phase fluctuations will thus dominate
and we neglect amplitude fluctuations, writing

ψ(r) = ψ0 eiφ(r) with ψ0 = −α/β. (7.2)

p

Thus, up to an irrelevant constant,

Z   X 
α α
2
F [φ] = d r −γ ∗
(∇φ) · ∇φ = −γ k 2 φ∗k φk . (7.3)
β β
| {z } k
>0

We can now calculate the correlation function

Z
D E 1
hψ(r)∗ ψ(0)i = ψ02 e−iφ(r) eiφ(0) ≡ ψ02 Dφ e−iφ(r)+iφ(0) e−F [φ]/kB T . (7.4)
Z
This is a Gaussian average since F [φ] is bilinear in φk . We now expand the exponential in the average into a
Taylor series,
∞
X 1
hψ(r)∗ ψ(0)i = ψ02 (−i)n h(φ(r) − φ(0))n i . (7.5)
n=0
n!
Making use of known properties of Gaussian averages, we write this as
∞
X 1 n/2
hψ(r)∗ ψ(0)i = ψ02 (−i)n 1 × 3 × 5 × · · · × (n − 1) (φ(r) − φ(0))2
n=0
n!
n even
∞
n=2m 2
X 1 × 3 × 5 × · · · × (2m − 1) m
= ψ0 (−1)m (φ(r) − φ(0))2
m=0
1 × 2 × 3 × · · · × (2m)

54
 ∞
X 1 m
= ψ02 (−1)m (φ(r) − φ(0))2
m=0
2 × 4 × · · · × (2m)
| {z }
1
= 2m m!
∞  m
X 1 1 m
= ψ02 − (φ(r) − φ(0))2
m=0
m! 2
 
1
= ψ02 exp − (φ(r) − φ(0))2 . (7.6)
2
Herein, we have, with the system area A,

(φ(r) − φ(0))2 = (φ(r) − φ(0))∗ (φ(r) − φ(0))

1 X −ik·r 0
= (e − 1)(eik ·r − 1) hφ∗k φk0 i
A 0
kk
1 1 X −ik·r 0 kB T β 1
= (e − 1)(eik ·r − 1) δkk0 from Eq. (7.3)
A 2 0 −γα k 2
kk
1 1X kB T β 1
= (2 − 2 cos k · r) . (7.7)
A 2 −γα k 2
k

The factor of 1/2 appearing in the third line is somewhat subtle. It stems from φ(r) being a real field so that the
integrals over φk and φ∗k double count the physical degrees of freedom. Going over to the thermodynamic limit,
i.e., the continuum description, we obtain
d2 k 1 − cos k · r
Z
kB T β
(φ(r) − φ(0))2 =
−γα (2π)2 k2
Z2π ZΛ
kB T β 1 1 − cos(kr cos ϕ)
= 2
dϕ dk
−γα (2π) k
0 0
ZΛ
kB T β 1 1
= dk (1 − J0 (kr) ). (7.8)
−γα 2π k | {z }
0 Bessel function

The k-integral is cut off at Λ, which is the inverse of some microscopic length scale, Λ = 1/r0 . The idea is that
contributions from shorter length scales r < r0 have been integrated out to obtain the Landau functional. The
substitution u = kr gives
r/r
Z 0
kB T β 1 (r/r0 )2 (r/r0 )2
 
kB T β 1 1 − J0 (u)
(φ(r) − φ(0))2
= du = F
p q 1, 1; 2, 2, 2; − . (7.9)
−γα 2π u −γα 2π 8 4
0

The resulting expression contains a generalized hypergeometric function. However, for large r/r0  1, there
exists a simple expansion:
"  −3/2 #
kB T β 1 r r ∼ kB T β 1 r
2
(φ(r) − φ(0)) = ln − ln 2 + γE + O = ln (7.10)
−γα 2π r0 r0 −γα 2π r0

and thus    −η

1 kB T β 1 r r
hψ(r)∗ ψ(0)i ∼
= ψ02 exp ln = ψ02 (7.11)
2 γα 2π r0 r0
with
1 kB T β
η=− > 0. (7.12)
4π γα

55
Thus the correlation function of the order parameters decays like a power law of distance in two dimensions.
We do not find long-range order, which would imply limr→∞ hψ(r)∗ ψ(0)i 6= 0. This result agrees with the
Mermin-Wagner theorem, which forbids long-range order for the two-dimensional superfluid. The power-law
decay characterizes so-called quasi-long-range order (short range order would have an even faster, e.g., exponential,
decay).

Isolated vortices
We have argued that fluctuations in the amplitude are less important because they have an energy gap proportional
to −2α > 0. This is indeed true for small amplitude fluctuations. However, there exist variations of the amplitude
that are, while energetically costly, very stable once they have been created. These are vortices. In two dimensions,
a vortex is a zero-dimensional object; the order parameter goes to zero at a single point at its center. The simplest
form of a vortex at the origin can be represented by

ψ(r) = |ψ(r)| eiφ(r) = |ψ(r)| ei(ϕ−ϕ0 ) , (7.13)

where r and ϕ are (planar) polar coordinates of r. An antivortex would be described by

ψ(r) = |ψ(r)| e−i(ϕ−ϕ0 ) . (7.14)

In both cases, limr→0 |ψ(r)| = 0. Note that we have changed the convention for the sign in the exponent compared
to superconducting vortices.
In the presence of vortices, the phase φ(r) of the order parameter is multivalued. Moreover, it is of course
undefined at the vortex centers. On the other hand, the phase gradient v = ∇φ is single-valued (but still undefined
at the vortex centers). For any closed loop C not touching any vortex cores, we have
I
ds · v = total change in phase along C = 2π NC , (7.15)
C

where NC ∈ Z is the enclosed winding number or vorticity. It is a topological invariant of the vortices since it
cannot be changed by a continuous deformation of ψ(r) unless ψ is made zero somewhere on the loop C. Like in
three dimensions, vortices are topological defects of the condensate.
The vorticity can be written as the sum of the vorticities Ni = ±1 of all vortices and antivortices inside the
loop, X
NC = Ni . (7.16)
i

N1 = +1
N2 = +1

N3 = −1

We now evaluate
∂ ∂ ∂ ∂φ ∂ ∂φ
∇×v = vy − vx = − . (7.17)
∂x ∂y ∂x ∂y ∂y ∂x
Note that in two dimensions the curl is a scalar. We assume v to be differentiable where it exists. Then ∇ × v
vanishes where v exists since the derivative commute. However, v does not exist at vortex cores. ∇ × v can be
extended to these points in the sense of distributions. For a single vortex at the origin, we make the ansatz

∇ × v = c δ(r) (7.18)

56
with c an unknown constant and integrate over an area S that contains the vortex core
ZZ ZZ I
Stokes
c= d2 r c δ(r) = d2 r ∇ × v = ds · v = 2π. (7.19)
S S ∂S

Hence for a single vortex we find

∇ × v = 2π δ(r). (7.20)
Generalization to multiple vortices at positions Ri gives
X
∇ × v = 2π Ni δ(r − Ri ) = 2π nv (r), (7.21)
i

where we have used the vortex concentration

X
nv (r) := Ni δ(r − Ri ). (7.22)
i

Now it is always possible to decompose a vector field into a curl-free (irrotational) and a divergence-free
component,
v = vph + vv , (7.23)
with

∇ × vph = 0, (7.24)
∇ · vv = 0. (7.25)

Since the curl of and thus the vortex concentation associated with vph vanishes, the component vph does not
contain any vortices. Alternatively, note that the first equation implies that there exists a single-valued scalar
field Ω(r) so that
vph = ∇Ω. (7.26)
Ω is a single-valued component of the phase, which cannot be due to vortices. This is the contribution from small
phase fluctuations, which we have already discussed. It leads to quasi-long-range order etc.
Conversely, vv is the vortex part, for which

∇ × vv = 2π nv , (7.27)
∇ · vv = 0. (7.28)

These relations suggest an electrostatic analogy. It is advantageous to rescale and rotate the field vv :
r
α
E(r) := − −8πγ ẑ × vv (r). (7.29)
β
| {z }
>0

Then the energy density far from vortex cores, where |ψ| ∼ −α/β, is
p
=
 −1
α α α α 1
w = −γ (∇φv ) · ∇φv = −γ vv · vv = −γ −8πγ E·E= E · E. (7.30)
β β β β 8π

Also, we find r r
α α
∇·E=− −8πγ (−∇ × vv ) = 2π −8πγ nv (7.31)
β β
and
∇ × E = 0. (7.32)

57
These equations reproduce the fundamental equations of electrostatics if we identify the charge density with
r
α
ρv = −8πγ nv . (7.33)
β
The factor in Gauss’ law ∇ · E = 2πρv is 2π instead of 4π since we are considering a two-dimensional system. We
can now derive the pseudo-electric field E(r) for a single vortex,
I r
α
da · E = 2πr E = 2π −8πγ (7.34)
β
q
−8πγ αβ
⇒ E(r) = (7.35)
r
and thus q
α
−8πγ β
E(r) = r̂. (7.36)
r
From this, we obtain the energy of a single vortex,
Z Z Z
1 1 α 2 1 α 1
E1 = Ecore + d r 2
E · E = Ecore − 8πγ d r 2 = Ecore − 2πγ dr . (7.37)
8π 8π β r β r
2
Now we note that the derivation does not hold for small distances from the vortex center since there |ψ| is not
close to −α/β. Thus we cut off the radial integral at the lower end at some vortex core radius r0 and put all
energy contributions from the core into Ecore . r0 is on the order of the coherence length ξ since this is the length
scale on which |ψ| changes. But the integral still diverges at the upper limit; if our system has a characteristic
size of L, we obtain
ZL
α dr α L
E1 = Ecore − 2πγ = Ecore − 2πγ ln . (7.38)
β r β r0
r0 | {z }
>0

Thus the energy of a single, isolated vortex diverges logarithmically with the system size. This suggests that
isolated vortices will never be present as thermal fluctuations as long as α < 0. This is not true, though. The
probability density or average concentration of such vortices should be
 
1 1 2π α L
p1 ∝ 2 e−E1 /kB T = 2 e−Ecore /kB T exp γ ln
r0 r0 kB T β r0
 k2πT γ αβ  −1/2η
1 L B 1 L
= 2 e−Ecore /kB T = 2 e−Ecore /kB T . (7.39)
r0 r0 r0 r0
The total number of vortices is, on average,
 2−1/2η
L
2
Nv = L p 1 ∝ e −Ecore /kB T
. (7.40)
r0
For η > 1/4, Nv diverges in the thermodynamic limit so that infinitely many vortices are present. For η < 1/4,
Nv → 0 for L → ∞ and according to our argument, which is essentially due to Kosterlitz and Thouless, there are
no vortices. It is plausible and indeed true that free vortices destroy quasi-long-range order and in this sense also
superfluidity. Note that
1 kB T β ! 1
η=− = (7.41)
4π γ α(T ) 4
is an equation for a critical temperature for the appearance of free vortices. We thus find that the critical
temperature in superfluid films should be reduced from the point where α = 0 (η → ∞) to the one where η = 1/4
due to vortices appearing as fluctuations of the order parameter. While qualitatively true, our description is still
incomplete, though, since we have so far neglected interactions between vortices.

58
 c
0 T single vortex T MF T
←
η=0 η = 1/4 η

8
(α = 0)

Vortex interaction
The energy of two vortices with vorticities ±1 can easily be obtained from the electrostatic analogy. We assume
that core regions do not overlap, i.e., the separation is R ≥ 2r0 . The pseudo-electric field of the two vortices,
assumed to be located at ±R/2 = ±R x̂/2, is

α r − R/2
r r
α r + R/2
E(r) = −8πγ − −8πγ
β |r − R/2|2 β |r + R/2|2
2 2
α |r + R/2| (r − R/2) − |r − R/2| (r + R/2)
r
= −8πγ 2 2 . (7.42)
β |r − R/2| |r + R/2|

Thus the energy is

Z
1
E2 = 2Ecore + d2 rE·E
8π
!2
2 2
|r + R/2| (r − R/2) − |r − R/2| (r + R/2)
Z
1 α
= 2Ecore − 8πγ d2 r 2 2
8π β |r − R/2| |r + R/2|
Z∞ Z∞
α
= 2Ecore − γ 2 dx dy
β
−∞ 0
4 2 4 2 2 2
|r + R/2| |r − R/2| + |r − R/2| |r + R/2| − 2 |r + R/2| |r − R/2| (r2 − R2 /4)
× 4 4 . (7.43)
|r − R/2| |r + R/2|

We introduce elliptic coordinates σ, τ according to

R
x= στ, (7.44)
2
Rp 2 p
y= σ − 1 1 − τ 2, (7.45)
2
where σ ∈ [1, ∞[, τ ∈ [−1, 1]. Then
 2 Z
α R σ2 − τ 2
E2 = 2Ecore − 2γ dσ dτ √ √
β 2 σ2 − 1 1 − τ 2
R 4

2
4
2
(σ + τ )4 R2 (σ − τ )2 + R2 (σ − τ )4 R2 (σ + τ )2


2
×
R 4

4
(σ − τ )4 R2 (σ + τ )4


2

2
2
2
− 2 R2 (σ + τ )2 R2 (σ − τ )2 R2 (σ 2 τ 2 + (σ 2 − 1)(1 − τ 2 ) − 1)

(σ + τ )2 + (σ − τ )2 − 2(σ 2 + τ 2 − 2)
Z
α 1
= 2Ecore − 2γ dσ dτ √ √
β σ2 − 1 1 − τ 2 σ2 − τ 2
Z
α 1 4
= 2Ecore − 2γ dσ dτ √ √ 2 − τ2
β 2
σ −1 1−τ 2 σ
Z
α 1
= 2Ecore − 8πγ dσ . (7.46)
β σ(σ 2 − 1)

59
We have to keep in mind that the integrals in real space have a lower cutoff r0 . The minimum separation from
the vortex at R x̂/2 is (σ − 1)R/2. For this separation to equal r0 , the lower cutoff for σ must be
2r0
σ0 = 1 + . (7.47)
R
With this cutoff, we obtain
α 1 (R + 2r0 )2
E2 = 2Ecore − 8πγ ln , (7.48)
β 2 4r0 (R + r0 )
which for R  r0 becomes
α R
E2 = 2Ecore − 4πγ ln . (7.49)
β 4r0
We absorb an R-independent constant into Ecore and finally obtain
α R
E2 ≡ 2Ecore + Vint (R) = 2Ecore − 4πγ ln . (7.50)
β r0
| {z }
>0

Note that the energy of a vortex-antivortex pair is independent of the system size L. Also, the vortex-antivortex
interaction Vint (R) increases with R, i.e., vortex and antivortex attract each other. Conversely, one can show that
vortices with the same vorticity repel each other.
Vint

0 r0 R

One can also show that for arbitrary vorticities N1 , N2 ∈ Z, the interaction reads
α R
Vint (R) = 4πγ N1 N2 ln . (7.51)
β r0
| {z }
<0

Since the equations of (pseudo-) electrostatics are linear, the superposition principle applies and we do not have
additional 3-, 4-, etc. body interactions. The energy of a system of vortices is thus
X 1 X
E= Ecore + Ni Nj vint (|ri − rj |) (7.52)
i
2
ij,i6=j

with
α r
vint (r) := 4πγ ln , (7.53)
β r0
provided that i Ni = 0. If the total vorticity i Ni does not vanish, the energy diverges logarithmically with
P P
the system size, as we have seen. As long as the total vorticity is zero, the energy per vortex is finite and thus
we expect a non-zero concentration of vortices for all temperatures T > 0. We now want to understand the
consequences of their presence.

60
Berezinskii-Kosterlitz-Thouless theory
The vortices in a two-dimensional superfluid behave like a Coulomb gas—a gas of charged particles with Coulomb
interaction, which is logarithmic in 2D. We consider the case that the (perhaps suitably modified) exponent η is
smaller than 1/4 so that the total vorticity (charge) has to vanish. It is then possible to group the vortices into
vortex-antivortex pairs. We do this using the following simple algorithm:
1. find the vortex and the antivortex with the smallest separation r,
2. mark this vortex and this antivortex as a pair,
3. repeat these steps for all remaining vortices until none are left.

+
+
+
− −

+
− −

The energy of an isolated vortex-antivortex pair (we will use the term “vortex pair ” from now on) of size r is
α r
E2 (r) = 2Ecore − 4πγ ln . (7.54)
β r0
Thus the probability density of such pairs is
1 −E2 (r)/kB T
p2 (r) ∝ e . (7.55)
r04

We can then write    r 2πK0

1 2 r 1 0
p2 (r) = 4 y0 exp −2πK0 ln = 4 y02 . (7.56)
r0 r0 r0 r
Here,
y0 := N0 e−Ecore /kB T (7.57)
is called the vortex fugacity. N0 is a constant of order unity; is the constant of proportionality implied in Eq.
N02
(7.55). More precisely, y02 is a vortex-pair fugacity. Furthermore,
1 α
K0 := − 2γ > 0 (7.58)
kB T β
is a dimensionless measure for the interaction strength in units of the thermal energy. K0 is the stiffness of the
condensate against phase fluctuations. Note that the previously defined exponent η is related to K0 through
η = 1/(2πK0 ).
The crucial idea is now that smaller pairs are polarized in the pseudo-electric field of the vortex and antivortex
forming a given pair. This leads to the screening of the vortex-antivortex interaction and thus reduces the energy
of large pairs. Formally, this can be described by renormalization-group (RG) theory (Kosterlitz 1974), which we
will summarize in the following. The grand-canonical partition function of the vortex-antivortex system is
 
Z 2 2
|r − |
X 1 Z
d r 1 d r2N 1 X i r j
Z0 = y 2N ··· exp  2π Ni Nj K0 ln , (7.59)
(N !)2 0 r02 r02 2 r0
N D1 D2N i6=j

61
where we have already implemented the constraint that the number of vortices, N , equals the number of antivor-
tices. The ranges of integration, Di , comprise the two-dimensional space R2 excluding disks of radius r0 centered
at all vortices (and antivortices) with numbers j < i. This means that the minimum separation is r0 . The idea of
RG theory is to perform the integrals for the smallest pairs of sizes between r0 and r0 + dr and rewrite the result
(approximately) in a form identical to Z0 but with changed (“renormalized”) parameters. Physically, we thereby
omit the smallest pairs and take their effects into account by renormalizing the parameters.

11111111111111111111
00000000000000000000
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
dr
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
r 0

00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
In this way, the partition function, the fugacity, and the stiffness become functions of the smallest length scale,
which we now denote by r. These “running” quantities are written as Z, y, and K, respectively. Thus
 
Z 2 2
|r − |
X 1 Z
d r1 d r2N 1 X i rj
Z= y 2N ··· exp  2π Ni Nj K ln . (7.60)
(N !)2 r2 r2 2 r
N D1 D2N i6=j

We now perform the integrals over the smallest separations between r and r0 = r + dr. A crucial assumption is
that this only involves vortex-antivortex pairs, not pairs of equal vorticity. This is plausible since a vortex and
an antivortex attract each other. The integration starts by splitting the integrals,
Z Z Z Z
d2 r1 · · · d2 r2N ∼
= d2 r1 · · · d2 r2N
D1 D2N D10 0
D2N
Z Z Z Z
1X
+ δNi ,−Nj d2 r1 · · · d2 r2N d2 rj d2 ri . (7.61)
2
i6=j D10 0
D2N D̃i,j r≤|ri −rj |<r+dr
| {z }
excluding i,j

The Di0 correspond to the Di but with the minimum separation increased to r + dr. D̃i,j is the full R2 excluding
disks of radius r around all vortices n 6= i, j. Inserting this into the expression for Z, we obtain the integral, from
the second term,

|rn − rj |
Z Z X
d2 rj d2 ri exp 2π Nn Nj K ln
r
n6=i,j
D̃i,j r≤|ri −rj |<r+dr

X |rn − ri | |rj − ri | 
− 2π Nn Nj K ln − 2π K ln
r r
n6=i,j
 
2
− · −
Z
∼
X (rm r j ) (r n r j ) r
= d2 rj 2πr dr 1 + π 2 K 2 Nm Nn 2 2

m,n6=i,j
|r m − r j | |rn − r j|
D̃i,j

62
 !
(rm − rj ) · (rn − rj )
X Z
= 2πr dr A + π 2 K 2 Nm Nn r2 d2 rj 2 2 , (7.62)
m,n6=i,j
|rm − rj | |rn − rj |
D̃i,j

where A is the area of the system. Herein, we have

(rm − rj ) · (rn − rj ) |rm − rn |

Z
L
d2 rj 2 2 = 2π ln r − 2π (1 − δmn ) ln . (7.63)
|rm − rj | |rn − rj | r
D̃i,j

The first term diverges for L → ∞ but drops out when the sum over m, n is performed, due to overall vanishing
vorticity. The result is  
X |rm − rn | 
3 2 2
(7.64)

A − 2π K r
· · · = 2πr dr  Nm Nn ln 
r 
m,n6=i,j
m6=n

and with the sum over i, j:

 
1X X |rm − rn | 
· · · = 2πr dr N 2 A − 2π 3 K 2 r2 (N − 1)2 Nm Nn ln , (7.65)
2 r
i6=j m6=n

neglecting some terms of order N 0 . Inserting everything into Z, we obtain two terms: The first corresponds to
Z with Di replaced by Di0 and the second reads
X 1  y 2N Z Z
2
d r1 · · · d2 r2N 2πr dr
(N !)2 r2
N
D10 0
D2N
| {z }
2N −2 integrals
! !
X |rm − rn | 1 X |rm − rn |
× 2
N A − 2π K r 3 2 2 2
(N − 1) Nm Nn ln exp 2πK Ni Nj ln . (7.66)
r 2 r
m6=n i6=j
| {z }
(2N −2)(2N −3)terms

We rename the summation index N as N + 1 in this second term. Then both terms contain 2N integrals under
the sum over N . We also put all terms containing dr into the exponent using 1 + a dr = ea dr . We obtain
   X Z Z
0 y 2 1  y 2N 2
Z = exp 2π 2 r dr A d r1 · · · d2 r2N
r (N !)2 r2
N
D10 0
D2N
 
1 X dr

|ri − rj |
× exp − −2πK + 8π 4 y 2 K 2 Ni Nj ln . (7.67)
2 r r
i6=j

Next, we have to express Z 0 in terms of the new length scale r0 := r + dr. This is only relevant in expressions not
already linear in dr. This applies to, on the one hand,

1 1 1 + 2 drr0
= = , (7.68)
r2 (r0 − dr)2 (r0 )2

and, on the other,

   
1 X 1 X
exp − 2πK Ni Nj ln r = exp − 2πK Ni Nj ln(r0 − dr)
2 2
i6=j i6=j

63
    
1X 1X dr 
= exp − 2πK Ni Nj ln r0  exp  2πK Ni Nj
2 2 r0
i6=j i6=j
 
 2N
1X dr
= exp − 2πK Ni Nj ln r0  1 − πK 0 , (7.69)
2 r
i6=j

neglecting terms of order N 0 compared to N . The renormalized partition function is finally

"  2 #
y X 1  y 2N  dr
2N
0 0
Z = exp 2π r dr A 1 + (2 − πK) 0
(r0 )2 (N !)2 (r0 )2 r
N
| {z }
=: Zpair
 
Z Z
1 X dr

|r i − r |
j 
× d2 r1 · · · d2 r2N exp  2πK − 8π 4 y 2 K 2 0 Ni Nj ln . (7.70)
2 r r0
D10 0
D2N i6=j

Here, Zpair is the partition function of the small pair we have integrated out. It is irrelevant for the renormalization
of y and K. Apart from this factor, Z 0 is identical to Z if we set
 
dr
0
y(r ) ≡ y(r + dr) = 1 + (2 − πK(r)) 0 y(r) (7.71)
r
dr
K(r0 ) ≡ K(r + dr) = K(r) − 4π 3 y 2 (r)K 2 (r) 0 . (7.72)
r
On the right-hand sides, we can replace dr/r0 by dr/r since the difference is of second order in dr:
 
dr
y(r0 ) ≡ y(r + dr) = 1 + (2 − πK(r)) y(r) (7.73)
r
dr
K(r0 ) ≡ K(r + dr) = K(r) − 4π 3 y 2 (r)K 2 (r) . (7.74)
r
Introducing the logarithmic length scale
r dr
l := ln ⇒ dl = , (7.75)
r0 r
we obtain the Kosterlitz RG flow equations
dy
= (2 − πK) y, (7.76)
dl
dK
= −4π 3 y 2 K 2 . (7.77)
dl
The initial conditions are

y(l = 0) = y0 = e−Ecore /kB T , (7.78)

1 α(T )
K(l = 0) = K0 = − 2γ , (7.79)
kB T β

i.e., the parameters assume their “bare” values at r = r0 (l = 0).

We will now discuss the physics encoded by the RG flow equations. First, note that the quantity
2
C := 2π 2 y 2 − − ln K (7.80)
πK

64
is invariant under the RG flow:
dC dy 2 dK 1 dK
= 4π 2 y − − = 4π 2 y 2 (2 − πK) − 8π 2 y 2 + 4π 3 y 2 K = 0. (7.81)
dl dl πK 2 dl K dl
Thus C is a “first integral” of the flow equations. We can calculate C from the initial values y0 , K0 and obtain
 
2 1 1 K
2π 2 y 2 = 2π 2 y02 + − + ln (7.82)
π K K0 K0
s  
1 1 1 1 K
⇒ y = y02 + 3 − + 2 ln . (7.83)
π K K0 2π K0

y
T

ix T = Tc
ratr
sepa

initial
conditions

0 2/π K

The RG flow is along curves decribed by this expression, where the curves are specified by y0 , K0 . The direction
of the flow is towards smaller K since dK/dl < 0. The parameters y0 , K0 change with temperature as given in
Eqs. (7.78) and (7.79). These initial conditions are sketched as a dashed line in the figure. We see that there are
two distinct cases:
• For T < Tc , K flows to some finite value K(l → ∞) > 2/π. This means that even infinitely large pairs feel
a logarithmic attraction, i.e., are bound. Moreover, the fugacity y flows to zero, y(l → ∞) = 0. Thus large
pairs are very rare, which is consistent with their (logarithmically) diverging energy.
• For T > Tc , K flows to K(l → ∞) = 0. Thus the interaction between a vortex and an antivortex that are far
apart is completely screened. Large pairs become unbound. Also, y diverges on large length scales, which
means that these unbound vortices proliferate. This divergence is an artifact of keeping only the leading
order in y in the derivation. It is cut off at finite y if we count (in particular, large) vortex-antivortex pairs
consistently. But the limit K → 0 remains valid.
At T = Tc we thus find a phase transition at which vortex-antivortex pairs unbind, forming free vortices. It is called
the Berezinskii-Kosterlitz-Thouless (BKT) transition. In two-dimensional films, vortex interactions thus suppress
the temperature where free vortices appear and quasi-long-range order is lost from the point T = Tsingle c
vortex
where
1 kB T β 1 1 ! 1 2
η=− = = ⇒ K0 = (7.84)
4π γα 2π K0 4 π
to the one where K(l → ∞) = 2/π and y0 , K0 lie on the “separatrix” between the two phases,
2 ! 2 2
C = 2π 2 y02 − − ln K0 = 2π 2 y 2 (l → ∞) − − ln K(l → ∞) = 0 − 1 − ln (7.85)
πK0 πK(l → ∞) π
2 2
⇒ + ln K0 = 1 + ln + 2π 2 y02 . (7.86)
πK0 π
Clearly the two criteria agree if y0 = 0. This makes sense since for y0 = 0 there are no vortex pairs to screen the
interaction. In addition, a third temperature scale is given by the mean-field transition temperature TMF , where
α = 0.

65
 quasi−long−range free vortices
order no condensate

000000000
111111111
0 Tc c
Tsingle vortex TMF T

01 K( l oo ) = 2/π K 0 = 2/π α= 0
K0 = 0

In the low-temperature phase, the largest pairs determine the decay of the correlation function hψ ∗ (r)ψ(0)i for
large r. Thus we find
 −η
∼ r
∗
hψ (r)ψ(0)i = ψ02
(7.87)
r0
with
1 1
η= . (7.88)
2π K(l → ∞)
We note that the exponent η changes with temperature (one could say that the whole low-temperature phase is
critical) but assumes a universal value at Tc : There,
2 1
lim K = ⇒ η(Tc ) = . (7.89)
l→∞ π 4
Due to the screening of the vortex interaction, the condensate is less stiff than it would be in the absence of
vortices. This is described by the renormalization of K from K0 to K(l → ∞). It is customary but somewhat
misleading to express this as a renormalization of the superfluid density ns , which is, after all, proportional to
K0 . Assuming that the temperature dependence of α and thus of the bare superfluid density
α
n0s ∝ ψ02 = − ∝ K0 (7.90)
β
is negligible close to Tc , the renormalized superfluid density
K 0
ns (T ) := n (7.91)
K0 s
obtains its temperature dependence exclusively from K/K0 . (The argument l → ∞ is implied here and in the
following.) For T < Tc but close to the BKT transition, we have
2
K= + ∆K. (7.92)
π
The invariant C = 2π 2 y02 − 2
πK0 − ln K0 is an analytic function of temperature and its value at Tc is
 
2 2
Cc = 2π 2 y 2 − − ln K = −1 − ln . (7.93)
πK T =Tc π

Thus we can write, close to Tc ,

2
C∼
= −1 − ln + b (T − Tc ) (7.94)
π
with some constant b := dC/dT |T =Tc . On the other hand,
 
2
C = 2π 2 y 2 − − ln K
πK
T 
2 2
=− − ln + ∆K
π π2 + ∆K


π
1 2  π 
=− π − ln − ln 1 + ∆K
1 + 2 ∆K π 2

66
 ∼ π  π2 2 π  1 π2
= −1 + ∆K  − ∆K 2 − ln − ∆K
 + ∆K 2
2 4 π 
2 2 4
2 π2
= −1 − ln − ∆K 2 . (7.95)
π 8
Thus
π2 2√ p
− ∆K 2 ∼
= b (T − Tc ) ⇒ ∆K ∼
= 2b Tc − T . (7.96)
8 |π {z }
const

Consequently, ns = (K/K0 ) n0s jumps to a finite value at Tc and then increases like a square root.
ns

Tc T

This behavior was indeed measured in torsion-pendulum experiments on He-4 films (Bishop and Reppy, 1980).

7.2 Superconducting films

In this section we concentrate on what is different for charged superconductors compared to neutral superfluids.
Recall that for a superfluid the gradient term in the Landau functional reads
Z
d2 r γ (∇ψ)∗ · ∇ψ. (7.97)

If we move around a vortex once, the phase has to change by 2π, regardless of the distance from the vortex core.
This phase change leads to an unavoidable contribution to the gradient term of
Z Z  ∗
α 1 ∂ iφ 1 ∂ iφ
d2 r γ (∇ψ)∗ · ∇ψ ∼
= −γ dr dϕ r e e
β r ∂ϕ r ∂ϕ
Z  2 Z Z
α 1 ∂φ α 1 α dr
= −γ dr dϕ = −γ dr dϕ = −2π γ , (7.98)
β r ∂ϕ β r β r
which diverges logarithmically with system size. On the other hand, for a superconducting film the gradient term
reads Z   2
1 ~ q
2
d r ∇− A ψ . (7.99)
2m∗ i c
Again, the phase of ψ winds by 2π around a vortex, but the associated gradient can, in principle, be compensated
by the vector potential. Pearl (1964) showed that this indeed leads to a finite energy of a single vortex in a
superconducting film. In addition, the free energy contains the magnetic-field term
B 2 (r)
Z
d3 r . (7.100)
8π
Note that the integral is three dimensional —the field is present in all of space, also outside of the film.
We first note that the film has a new length scale: In the London gauge, Ampère’s law reads
4π 4π
∇ × ∇ × A = ∇(∇ · A}) − ∇2 A = j ⇒ ∇2 A = − j. (7.101)
| {z c c
=0

67
On the other hand, from the London equation we have
c
j=− A, (7.102)
4πλ2
which is valid within the film and away from vortex cores so that |ψ| ∼
2
= −α/β. However, the current is confined
to the thin film. We can write
j(r) = K(x, y) δ(z) (7.103)
with a surface current density K. Thus
Z∞ Z  c 
K(x, y) = dz j(r) = dz − A(r). (7.104)
4πλ2
−∞ film

If the thickness is d and A is approximately constant across the thickness, we have

c d
K(x, y) = − A(x, y, 0) (7.105)
4π λ2
c d
⇒ j(r) = − A(r) δ(z) (7.106)
4π λ2
and finally
d
∇2 A =A δ(z). (7.107)
λ2
This result exhibits the new length scale that controls the spatial variation of A and thus of the current j,
λ2
λ⊥ := , (7.108)
d
which is large, λ⊥  λ, for a thin film. We see that in thin films, λ⊥ assumes the role of the penetration depth
λ. Since λ⊥ is large for thin films, one could say that thin films are always effectively of type II.
We here do not discuss the full derivation of Pearl but only consider the “far field” for large r  λ⊥ and show
that it does not lead to a diverging
p free energy for a single vortex. We assume that most of the magnetic flux of
Φ0 penetrates the film for % = x2 + y 2 . λ⊥ . Then the magnetic field above (and below) the film looks like a
monopole field for r  λ⊥ .

z B

00000000000000000000d
11111111111111111111

2
λ⊥ = λ / d
This field is easy to obtain from symmetry:
Z
da · B = Φ0 ⇒ 2πr2 B = Φ0 (7.109)
half space
for z>0

68
 Φ0 Φ0
⇒ B(r) = and B(r) = r̂ for z > 0. (7.110)
2πr2 2πr2
By symmetry,
Φ0
B(r) = sgn z r̂. (7.111)
2πr2
This gives a contribution to the field energy of
Z∞
B2 Φ20 Φ20 Φ2 1
Z Z
1 dr
3
d r = 3
d r 4 = 4π = 02 , (7.112)
8π 32π 3 r 32π 3 r 2 8π λ⊥
λ⊥

which is finite for an infinite film. The value of the lower cutoff does not matter for this. The cutoff has to be
present, since the monopole field is not a valid approximation for small r.
We next obtain the current from Ampère’s law in integral form:
∆r
B

111111111
000000000
d

I Z
4π
dr · B = da · j (7.113)
c
r+∆r
Z
Φ0 4π
⇒ 2 dr0 = ∆r d j(r) (7.114)
2π(r0 )2 c
r
 
4π Φ0 1 1 ∼ Φ0 ∆r
⇒ ∆r d j(r) = − = (7.115)
c π r r + ∆r π r2
Φ0 1
⇒ j(r) = (7.116)
4π 2 r2 d
and with the vector character restored
Φ0 ϕ̂
j(r) = . (7.117)
4π 2 r2 d
Note that the sheet current is thus
Φ0 ϕ̂
K(r) = . (7.118)
4π 2 r2
For large r we have |ψ| = ψ0 = −α/β (note that typically λ⊥  ξ). Then the second Ginzburg-Landau
p

equation gives
q~ q2 q  q 
j = i ∗ (−i∇φ − i∇φ)ψ02 − ∗ ψ02 A = ψ02 ∗ ~∇φ − A . (7.119)
2m m c m c
Thus we can rewrite the gradient term in the free energy as
∗ 2
2
m∗
Z Z Z
1  q 2 2 ψ0 (m )
d d r2
ψ 2
0 ~∇φ − A = d d r j · j = d d2
r j · j. (7.120)
2m∗ c 2m∗ ψ04 q 2 2q 2 ψ02

The contribution from large r is

Z∞
dm∗ Φ2 1 m∗ Φ20 m∗ Φ20
Z
dr 1
d r 04 4 2 =
2
2π = , (7.121)
2q 2 ψ02 16π r d 32π 4 q 2 ψ02 d r3 8π 3 q 2 ψ02 d λ2⊥
λ⊥

which is also finite for an infinite film.

69
 We conclude that the free energy of an isolated vortex is finite in a superconducting film. It is then plausible
and indeed true that the interaction energy of a vortex-antivortex pair does not diverge for large separations r but
saturates for r  λ⊥ . Since for the far field of a single vortex the magnetic-field energy, Eq. (7.112), dominates
over the energy due to the gradient term, we expect the large-r interaction to be dominated by the Coulomb-type
attraction of the magnetic monopoles in the upper and lower half spaces. This supposition is borne out by a
proper analysis. Consequently, for large r the interaction behaves like
 2
Φ0 1
Vint (r) ∼
= const − , (7.122)
2π r

where Φ0 /2π is the monopole strength, according to Eq. (7.111).

111111111
000000000
N

All this shows that, strictly speaking, there will be a non-zero concentration of free vortices at any temperature
T > 0. Thus there is no quasi-long-range order. However, the relevant length scale is λ⊥ = λ2 /d, which can
be very large for thin films, even compared to the lateral size L of the sample. In this case, the large-r limit is
experimentally irrelevant. But for vortex separations r  λ⊥ , the magnetic-field expulsion on the scale r is very
weak since λ⊥ is the effective penetration depth. Then the fact that the condensate is charged is irrelevant and
we obtain the same logarithmic interaction as for a neutral superfluid.
Thus for thin films of typical size we can use the previously discussed BKT theory. For superconducting
films we even have the advantage of an additional observable, namely the voltage for given current. We give a
hand-waving derivation of V (I). The idea is that a current exerts a Magnus force on a vortex, in the direction
perpendicular to the current. The force is opposite for vortices and antivortices and is thus able to break vortex-
antivortex pairs. As noted above, free vortices lead to dissipation. A vortex moving through the sample in the
orthogonal direction between source and drain contacts leads to a change of the phase difference ∆φ by ±2π. We
will see in the chapter on Josephson effects why this corresponds to a non-zero voltage. Since free vortices act
independently, it is plausible to assume that the resistance is

R ∝ nv , (7.123)

where nv now denotes the concentration of free vortices. To find it, note that the total potential energy due to
vortex-antivortex attraction and Magnus force can be written as

V = Vint − 2FMagnus r, (7.124)

with (we assume r  λ⊥ )

r
Vint = 2π kB T K ln . (7.125)
r0

70
 V

Vint

rbarrier x

There is a finite barrier for vortex-antivortex unbinding at a separation rbarrier determined from ∂V /∂r = 0. This
gives
2π kB T Kbarrier
rbarrier = , (7.126)
2FMagnus
where Kbarrier := K(rbarrier ). The barrier height is

∆E := V (rbarrier ) − V (r0 ) ∼
= V (rbarrier )
rbarrier
= 2π kB T Kbarrier ln − 2FMagnus rbarrier
r0
 
rbarrier
= 2π kB T Kbarrier ln −1 . (7.127)
r0
For small currents we have rbarrier  r0 and thus
rbarrier
∆E ∼
= 2π kB T K(l → ∞) ln . (7.128)
| {z } r0
≡K

The rate at which free vortices are generated is

 −2πK
rbarrier
Rgen ∝ e−β∆E ∼
= . (7.129)
r0
The recombination rate of two vortices to form a pair is

Rrec ∝ n2v , (7.130)

since two vortices must meet. In the stationary state we have

 −πK
rbarrier
(7.131)
p
Rgen = Rrec ⇒ nv ∝ Rgen ∝
r0
and thus a resistance of  −πK
rbarrier
R ∝ nv ∝ . (7.132)
r0
Since the Magnus force FMagnus is proportional to the current I we have
1 1
rbarrier ∝ ∝ (7.133)
FMagnus I

so that  −πK
1
R∝ = I πK . (7.134)
I

71
Finally, the voltage measured for a current I is

V = RI ∝ I πK I = I 1+πK , (7.135)

where K ≡ K(l → ∞) is the renormalized stiffness. Since

2 2 √ p 
K= + ∆K ∼
= 1 + 2b Tc − T (7.136)
π π
we find for the exponent √ p
1 + πK ∼
= 3 + |2 {z2b} Tc − T (7.137)
= const

for T . Tc .

1+ π K
3

1
ohmic

Tc T

Above Tc , we have K = 0 and thus ohmic resistance, V ∝ I, as expected. At Tc , the exponent jumps to the
universal value 3, i.e., V ∝ I 3 . For T ≤ Tc , the voltage is sub-ohmic, i.e., the voltage is finite for finite current
but rises more slowly than linearly for small currents. This behavior has been observed for thin superconducting
films. Thus they are not superconducting in the sense of ideal conduction.

72
8

Origin of the attractive interaction

In the following chapters we turn to the microscopic theory of superconductivity. While the BCS theory is
reasonably easy to understand if one assumes an attractive interaction between electrons in a superconductor,
it is far from obvious where such an interaction should come from. The only fundamental interaction that is
relevant for (non-radioactive) solids is the electromagnetic one, which naively gives a repulsive interaction of a
typical strength of several eV between nearby electrons. How can this lead to an attraction at the low energy
scale kB Tc ∼ 1 meV? We will see that the lattice of ion cores (nuclei with tightly bound inner electrons) plays
an important role, in particular its dynamics. We will use Feynman diagrams for Green functions to describe
the physics. Unfortunately, we do not have the time to introduce these concepts rigorously; this is done in many
good textbooks on many-particle physics as well as in the lecture notes on Many-Particle Theory (in German),
which are available online. For those familiar with Feynman diagrams, they rigorously represent mathematical
expressions, for the others they should at least be useful as cartoons of the relevant processes.

8.1 Reminder on Green functions

Nevertheless, we start by briefly summarizing relevant properties of Green functions. In many-particle physics,
we usefully express the Hamiltonian in terms of an electronic field operator Ψσ (r, t), where σ = ↑, ↓ is the
electron spin. The field operator can be expanded in any convenient basis of single-particle states characterized
by wavefunctions ϕνσ (r), where ν represents all relevant quantum numbers,
X
Ψσ (r, t) = ϕνσ (r) cνσ (t), (8.1)
ν
X
Ψ†σ (r, t) = ϕ∗νσ (r) c†νσ (t). (8.2)
ν

cνσ and c†νσ are annihilation and creation operators of electrons in the single-particle state, respectively. Note
that we are working in the Heisenberg picture, i.e., the wavefunctions ϕνσ (r) are independent of time, whereas
the time dependence is carried by the operators.
For example, the Hamiltonian for free electrons reads

−~2 ∇2
Z X
H = d3 r Ψ†σ (r) Ψσ (r). (8.3)
σ
2m

In this case it is useful to expand into plane waves,

1 X ik·r
Ψσ (r) = √ e ckσ (8.4)
V k
~2 (k 0 )2
Z
1 X 0
e−ik·r+ik ·r c†kσ
X
⇒ H= d3 r ck0 σ
V 0 σ
2m
kk

73
 X ~2 k 2
= c†kσ ckσ . (8.5)
2m
kσ

Two types of Green functions are defined as follows:

• greater Green function: D E
G> (rσt, r0 σ 0 t0 ) := −i Ψσ (r, t)Ψ†σ0 (r0 , t0 ) , (8.6)

where h. . .i is the equilibrium average

Tr Ae−βH 1
hAi = Tr Aρeq = ≡ Tr Ae−βH , (8.7)
Tr e−βH Z
where β = 1/kB T is the inverse temperature and H is the many-particle Hamiltonian. For non-interacting
electrons at temperature T  EF /kB , ρeq describes the Fermi sea. G> describes the conditional probability
amplitude for an electron created at point r0 with spin σ 0 at time t0 to be found at point r with spin σ at
time t.
• lesser Green function: D E
G< (rσt, r0 σ 0 t0 ) := +i Ψ†σ0 (r0 , t0 )Ψσ (r, t) (8.8)

describes the propagation of a hole from time t to time t0 .

These are not the most useful Green functions since in quantum theory propagation of an electron forward in
time cannot be separated from propagation of a hole backward in time. It is also useful to distiguish between the
cases t > t0 and t < t0 . This is accomplished by these definitions:
• retarded Green function:
Dn oE
GR (rσt, r0 σ 0 t0 ) := −i Θ(t − t0 ) Ψσ (r, t), Ψ†σ0 (r0 , t0 ) , (8.9)

where {A, B} := AB+BA is the anti-commutator (appropriate for fermions) and the Heaviside step function
(
1 for t > t0
Θ(t − t0 ) = (8.10)
0 for t < t0

selects only contributions with t > t0 (“forward in time”).

• advanced Green function:
Dn oE
GA (rσt, r0 σ 0 t0 ) := +i Θ(t0 − t) Ψσ (r, t), Ψ†σ0 (r0 , t0 ) , (8.11)

this Green function analogously contains only contributions “backward in time.”

The thermal averages in the Green functions introduce operators e−βH , whereas the time evolution of operators
introduces time-evolution operators according to

A(t) = eiHt/~ A e−iHt/~ . (8.12)

Specifically, we have
1  0 0
GR (rσt, r0 σ 0 t0 ) =−i Θ(t − t0 ) Tr e−βH eiHt/~ Ψσ (r) e−iH(t−t )/~ Ψ†σ0 (r0 ) e−iHt /~
Z 
0 0
+ eiHt /~ Ψ†σ0 (r0 ) e−iH(t −t)/~ Ψσ (r) e−iHt/~ . (8.13)

In practice, we will want to write the Hamiltonian in the form H = H0 + V and treat V in perturbation theory.
It is not surprising that this is complicated due to the presence of H in several exponential factors. However,

74
these factors are of a similar form, only in some the prefactor of H is imaginary and in one it is the real inverse
temperature. Can one simplify calculations by making all prefactors real? This is indeed possible by formally
replacing t → −iτ , t0 → −iτ 0 , which is the main idea behind the imaginary-time formalism. We cannot discuss
it here but only state a few relevant results.
It turns out to be useful to consider the Matsubara (or thermal) Green function
D E
G(rστ, r0 σ 0 τ 0 ) := − Tτ Ψσ (r, τ )Ψ†σ0 (r0 , τ 0 ) , (8.14)

where for any operator

A(τ ) := eHτ /~ A e−Hτ /~ (8.15)
and Tτ is the time-ordering directive
(
A(τ )B(τ 0 ) for τ > τ 0 ,
0
Tτ A(τ )B(τ ) = (8.16)
±B(τ 0 )A(τ ) for τ < τ 0 ,

where the upper (lower) sign holds for bosonic (fermionic) operators. For time-independent Hamiltonians, the
Green function only depends on the difference τ − τ 0 . One can then show that the resulting Green function
G(rσ, r0 σ 0 , τ ) is defined only for τ ∈ [−~β, ~β] and satisfies

G(rσ, r0 σ 0 , τ + ~β) = −G(rσ, r0 σ 0 , τ ) (8.17)

for fermions. This implies that the Fourier transform is a discrete sum over the fermionic Matsubara frequencies

(2n + 1)π
ωn := , n ∈ Z. (8.18)
~β
The imaginary-time formalism is useful mainly because G is easier to obtain or approximate than the other Green
functions and these can be calculated from G based on the following theorem: The retarded Green function GR (ω)
in Fourier space is obtained from G(iωn ) by means of replacing iωn by ω + i0+ , where i0+ is an infinitesimal
positive imaginary part,
GR (ω) = G(iωn → ω + i0+ ). (8.19)
This is called “analytic continuation.” Analogously,

GA (ω) = G(iωn → ω − i0+ ). (8.20)

G(i ω n)
iωn ω + i 0+

R
G (ω )
A
G (ω )

iωn ω − i 0+

It will be useful to write the electronic Green function in k space. In the cases we are interested in, momentum
and also spin are conserved so that the Green function can be written as
D E
Gkσ (τ ) = − Tτ ckσ (τ ) c†kσ (0) . (8.21)

75
Analogously, the bosonic Matsubara Green function of phonons can be written as
D E
Dqλ (τ ) = − Tτ bqλ (τ ) b†qλ (0) , (8.22)

where λ enumerates the three polarizations of acoustic phonons. We also note that for bosons we have

D(τ + ~β) = +D(τ ), (8.23)

i.e., the opposite sign compared to fermions. Therefore, in the Fourier transform only the bosonic Matsubara
frequencies
2πn
νn := , n∈Z (8.24)
~β
occur.

8.2 Coulomb interaction

We now discuss the effect of the electron-electron Coulomb interaction. We will mainly do that at the level of
Feynman diagrams but it should be kept in mind that these represent mathematical expressions that can be
evaluated as needed. The Coulomb interaction can be written in second-quantized form as
Z
1 X
Vint = d3 r1 d3 r2 Ψ†σ1 (r1 )Ψ†σ2 (r2 ) VC (|r2 − r1 |) Ψσ2 (r2 )Ψσ1 (r1 ), (8.25)
2σ σ
1 2

with
e2
, VC (r) = (8.26)
r
using Gaussian units. GlossingPover potential problems related to commuting operators, this is just the Coulomb
energy of a charge density −e σ Ψ†σ (r)Ψσ (r).
In momentum space, we have
1 X ik·r
Ψσ (r) = √ e ckσ (8.27)
V k
so that
1 XZ 0 00
·r2 +ik000 ·r1 †
ckσ1 c†k0 σ2
X
Vint = d3 r1 d3 r2 e−ik·r1 −ik ·r2 +ik VC (|r2 − r1 |) ck00 σ2 ck000 σ1
2V 2
kk0 k00 k000 σ1 σ2
1 XZ 0 00
+k000 )·R i(k−k0 +k00 −k000 )·ρ/2 †
ckσ1 c†k0 σ2
X
= d3 R d3 % ei(−k−k +k e VC (ρ) ck00 σ2 ck000 σ1 , (8.28)
2V 2
kk0 k00 k000 σ1 σ2

where
r1 + r2
R= , (8.29)
2
ρ = r2 − r1 . (8.30)

We can now perform the integral over R:

1 X XZ 0 00 000
Vint = d3 ρ δk+k0 −k00 −k000 ,0 ei(k−k +k −k )·ρ/2 c†kσ1 c†k0 σ2 VC (ρ) ck00 σ2 ck000 σ1
2V 0 00 000 σ σ
kk k k 1 2
Z
1 X X 0 00
= d3 ρ e−i(k −k )·ρ VC (ρ) c†kσ1 c†k0 σ2 ck00 σ2 ck+k0 −k00 ,σ1 . (8.31)
2V 0 00 σ σ
kk k 1 2

Substituting new momentum variables

k1 = k + k0 − k00 , (8.32)

76
 k2 = k00 , (8.33)
q=k −k , 0 00
(8.34)
we obtain
1 X X
Vint = VC (q) c†k1 −q,σ1 c†k2 +q,σ2 ck2 σ2 ck1 σ1 (8.35)
2V σ σ
k1 k2 q 1 2

with Z
VC (q) = d3 r e−iq·r VC (r). (8.36)

The interaction Vint is a sum over all processes in which two electrons come in with momenta k1 and k2 , a
momentum of q is transfered from one to the other through the Coulomb interaction, and the electrons fly out
with momenta k1 − q and k2 + q.

k1− q,σ1 k2 + q,σ2

Vc (q)

k1σ1 k 2 σ2

VC (q) is obviously the Fourier transform of the Coulomb interaction. It is most easily obtained by Fourier
transforming the Poisson equation for a point charge,
∇2 φ(r) = −4π ρ(r) = −4πQ δ(r) (8.37)
Z
⇒ d3 r e−iq·r ∇2 φ(r) = −4πQ (8.38)
Z
by parts
⇒ d3 r (−iq)2 e−iq·r φ(r) = −4πQ (8.39)

⇒ q 2 φ(q) = 4πQ (8.40)

Q
⇒ φ(q) = 4π 2 (8.41)
q
so that
e2
VC (q) = 4π . (8.42)
q2

Screening and RPA

The bare Coulomb interaction VC is strongly repulsive, as noted above. But if one (indirectly) measures the
interaction between charges in a metal, one does not find VC but a reduced interaction. First of all, there will be
a dielectric function from the polarizability of the ion cores,
4π e2
VC (q) → , (8.43)
 q2
but this only leads to a quantitative change, not a qualitative one. We absorb the factor 1/ into e2 from now
on. More importantly, a test charge in a metal is screened by a cloud of opposite total charge so that from far
away the effective charge is strongly reduced. Diagrammatically, the effective Coulomb interaction VCfull (q, iνn ) is
given by the sum of all connected diagrams with two external legs that represent the Coulomb interaction. With
the representations
−VC ≡ (8.44)

(the minus sign is conventional) and

−VCfull ≡ (8.45)

77
as well as

G0 ≡ (8.46)

for the bare electronic Green function corresponding to the non-interacting Hamiltonian H0 , we obtain1

+ + +

+ + + +

+ ··· (8.47)

This is an expansion in powers of e2 since VC contains a factor of e2 . We have exhibited all diagrams up to order
e6 . If we try to evaluate this sum term by term, we encounter a problem: At each vertex •, momentum and
energy (frequency) must be conserved. Thus in partial diagrams of the form

k k

q=0

the Coulomb interaction carries momentum q = 0. But

e2
VC (0) = 4π (8.48)
02
is ill defined. However, one can show that the closed G loop corresponds to the average electron density so
that the diagram signifies the Coulomb interaction with the average electronic charge density (i.e., the Hartree
energy). But this is compensated by the average charge density of the nuclei. Thus we can omit all contributions
containing the “tadpole” diagram shown above.
Since we still cannot evaluate the sum in closed form we need an approximation. We first consider the limiting
cases of small and large q in VCfull (q, iνn ):
• For large q, corresponding to small distances, the first diagram is proportional to 1/q 2 , whereas all the
others are at least of order 1/q 4 and are thus suppressed. We should recover the bare Coulomb interaction
for large q or small distances, which is plausible since the polarization of the electron gas cannot efficiently
screen the interaction between two test charges that are close together.
• For small q we find that higher-order terms contain higher and higher powers of 1/q 2 and thus become
larger and larger. This is alarming. The central idea of our approximation is to keep only the dominant
term (diagram) at each order in e2 . The dominant term is the one with the highest power in 1/q 2 . Only
the VC lines forming the backbone of the diagrams (drawn horizontally) carry the external momentum q
due to momentum conservation at the vertices. Thus the dominant terms are the ones with all VC lines in
the backbone:

q k q k’ q

k+q k’+ q
1 We do not write down the corresponding mathematical expressions here. Expressions for some of the diagrams will be derived

below.

78
Summing up these dominant terms, we obtain the approximate effective interaction

−VCRPA := +

+ +

+ ··· (8.49)

This is called the random phase approximation (RPA), for historical reasons that do not concern us here, or the
Lindhard approximation. The most important part of RPA diagrams is clearly the bubble diagram

Π0 ≡ , (8.50)

which stands for

1X 1 X 0 0
Π0 (q, iνn ) = − Gk+q,σ (iωn + iνn ) Gkσ (iωn )
β iω V
n kσ
2 XX 1 1
=− , (8.51)
β iω iωn + iνn − ξk+q iωn − ξk
n k

where the factor of 2 is due to the two-valued spin degree of freedom and ξk ≡ k − µ is the bare electronic
dispersion including the chemical potential µ. Π0 is essentially the negative of the electric susceptibility of the
free electron gas. We do not show this here. It is plausible, however, that the susceptibility, which controls the
electric polarization of the electron gas, should enter into a calculation of the screened Coulomb interaction.
We can write the diagrammatic series also as

−VCRPA (q, iνn ) = −VC (q) + VC (q) Π0 (q, iνn ) VC (q) − VC (q) Π0 (q, iνn ) VC (q) Π0 (q, iνn ) VC (q) + · · ·
(8.52)
 
= −VC (q) 1 − Π0 (q, iνn ) VC (q) + Π0 (q, iνn ) VC (q) Π0 (q, iνn ) VC (q) − · · · .

This is a geometric series, which we can sum up, with the result

VC (q)
VCRPA (q, iνn ) = . (8.53)
1 + VC (q) Π0 (q, iνn )

Note that VCRPA has a frequency dependence since Π0 (or the susceptibility) has one.
In the static limit iν → 0 + i0+ and at low temperatures T  EF /kB ∼ = µ/kB , one can show that

Π0 (q, 0) ∼
= const = N (EF ), (8.54)

where N (EF ) is the electronic density of states at the Fermi energy, including a factor of two for the spin. Thus
we obtain
2
4π qe2 e2
VCRPA (q) = 2 = 4π
1 + 4π qe2 Π0 q 2 + 4πe2 Π0

∼ e2
= 4π
q 2 + 4πe2 N (EF )
e2
= 4π 2 (8.55)
q + κ2s

with
(8.56)
p
κs := 4πe2 N (EF ).

79
Note that summing up the more and more strongly diverging terms has led to a regular result in the limit q → 0.
The result can be Fourier-transformed to give
e−κs r
VCRPA (r) = e2 (8.57)
r
(not too much confusion should result from using the same symbol “e” for the elementary charge and the base of
the exponential function). This is the Yukawa potential, which is exponentially suppressed beyond the screening
length 1/κs . This length is on the order of 10−8 m = 10 nm in typical metals. While we thus find a strong
suppression of the repulsive interaction at large distances, there is no sign of it becoming attractive.
On the other hand, for very high frequencies ν & EF /~, the electron gas cannot follow the perturbation, the
susceptibility and Π0 go to zero, and we obtain the bare interaction,

e2
VCRPA (q, ν → ∞) ∼
= VC (q) = 4π . (8.58)
q2

8.3 Electron-phonon interaction

The nuclei (or ion cores) in a crystal oscillate about their equilibrium positions. The quanta of these lattice
vibrations are the phonons. The many-particle Hamiltonian including the phonons has the form

H = Hel + Hph + Hel-ph , (8.59)

where we have discussed the electronic part Hel before,

 
1
Ωqλ b†qλ bqλ +
X
Hph = (8.60)
2
qλ

is the bare Hamiltonian of phonons with dispersion Ωqλ , and

1 XX  
Hel-ph = gqλ c†k+q,σ ckσ bqλ + b†−q,λ (8.61)
V
kσ qλ

describes the electron-phonon coupling. gqλ is the coupling strength. Physically, an electron can absorb (bqλ )
or emit (b†−q,λ ) a phonon under conservation of momentum. Hence, electrons can interact with one another by
exchanging phonons. Diagrammatically, we draw the simplest possible process as

k1− q,σ1 0 k2+ q,σ2

Dqλ

k1σ1 k 2σ2

In detail, we define, in analogy to the Coulomb interaction

−VC (q) ≡ , (8.62)

the interaction due to phonon exchange,

1 2 0
−Vph (q, λ, iνn ) ≡ − |gqλ | Dqλ (iνn ) ≡ . (8.63)
V
We quote the expression for the bare phonon Green function, i.e., the one obtained from Hph alone:
1 1 2 Ωqλ
0
Dqλ (iνn ) = + = . (8.64)
iνn − Ωqλ −iνn − Ωqλ (iνn )2 − Ω2qλ

80
The two terms in the intermediate form correspond to the two possible directions of propagation of the phonon.
The phonon-mediated interaction is thus frequency-dependent, whereas the Coulomb interaction is static. How-
ever, we could write the bare Coulomb interaction in a very similar form as the exchange of photons. Since the
speed of light is so much larger than the speed of sound, we can neglect the dynamics for photon exchange but
not for phonon exchange.

Jellium phonons
For our discussions we need a specific model for phonons. We use the simplest one, based on the jellium approxima-
tion for the ion cores (i.e., the nuclei with the tightly bound electrons). In this approximation we describe the ion
cores by a smooth positive charge density ρ+ (r, t). In equilibrium, this charge density is uniform, ρ+ (r, t) = ρ0+ .
We consider small deviations
ρ+ (r, t) = ρ0+ + δρ+ (r, t). (8.65)
Gauss’ law reads
∇ · E = 4π δρ+ (r, t) (8.66)
since ρ0+ is compensated by the average electronic charge density (the Hartree term). The density of force acting
on ρ+ is f = ρ+ E ∼
= ρ0+ E to leading order in δρ+ . Thus

∇·f ∼
= 4π ρ0+ δρ+ . (8.67)

The conservation of charge is expressed by the continuity equation

∂ ∂
ρ+ + ∇ · j+ ≡ ρ+ + ∇ · ρ+ v = 0. (8.68)
∂t ∂t
To leading order this reads (note that the velocity v vanishes in equilibrium)

∂
δρ+ + ρ0+ ∇ · v ∼
=0 (8.69)
∂t
∂2 ∂ Newton f
⇒ δρ+ ∼
= −ρ0+ ∇ · v = −ρ0+ ∇ · , (8.70)
∂t2 ∂t ρm
where ρm is the mass density of the ion cores, which is dominated by the nuclear mass. With the charge Zeff e
and mass M of the ion cores, we obtain

∂2 Zeff e Zeff e
2
δρ+ ∼
=− ∇·f =− 4π ρ0+ δρ+ , (8.71)
∂t M M
which is solved by
δρ+ (r, t) = δρ+ (r) e−iΩt , (8.72)
with r r
2 e2
Zeff
Zeff e 0
Ω= 4π ρ = 4π n0+ , (8.73)
M + M
where n0+ is the concentration of ions (or nuclei). We thus obtain optical phonons with completely flat dispersion,
i.e., we find the same frequency Ω for all vibrations.
One can also calculate the coupling strength gq . It is clear that it will be controlled by the Coulomb interaction
between electrons and fluctuations δρ+ in the jellium charge density. We refer to the lecture notes on many-particle
theory and only give the result:
1 2 Ω
|gq | = VC (q). (8.74)
V 2
Consequently, the electron-electron interaction due to phonon exchange becomes

1 2 Ω 2Ω Ω2
Vph (q, iνn ) = |gq | Dq0 (iνn ) = VC (q) = V C (q) . (8.75)
V 2 (iνn )2 − Ω2 (iνn )2 − Ω2

81
It is thus proportional to the bare Coulomb interaction, with an additional frequency-dependent factor. The
retarded form reads
Ω2 Ω2
R
Vph (q, ν) = Vph (q, iνn → ν + i0+ ) = VC (q) = V C (q) , (8.76)
(ν + i0+ )2 − Ω2 ν 2 − Ω2 + i0+ sgn ν
where we have used 2ν i0+ = i0+ sgn ν and have neglected the square of infinitesimal quantities.

8.4 Effective interaction between electrons

Combining the bare Coulomb interaction and the bare interaction due to phonon exchange, calculated for the
jellium model, we obtain the bare effective interaction between electrons,

Veff (q, iνn ) := VC (q) + Vph (q, iνn )

Ω2
= VC (q) + VC (q)
(iνn )2 − Ω2
(iνn )2
= VC (q) . (8.77)
(iνn )2 − Ω2
The retarded form is
ν2
R
Veff (q, ν) = Veff (q, iνn → ν + i0+ ) = VC (q) . (8.78)
ν 2 − Ω2 + i0+ sgn ν
This expression is real except at ν = Ω and has a pole there. Moreover, Veff
R
is proportional to VC with a negative
prefactor as long as ν < Ω.

VeffR

VC (q )
0
Ω ν

The effective interaction is thus attractive for 0 < ν < Ω. The exchange of phonons overcompensates the repulsive
Coulomb interaction. On the other hand, for ν → 0, the effective interaction vanishes. This means that in a
quasi-static situation the electrons do not see each other at all.
What happens physically is that the electrons polarize the (jellium) charge density of the ion cores. The
ions have a high inertial mass. Their reaction to a perturbation has a typical time scale of 1/Ω or a frequency
scale of Ω. For processes slow compared to Ω, the ion cores can completely screen the electron charge, forming a
polaron, which is charge neutral. For frequencies ν > 0, we have to think in terms of the response of the system
to a test electron oscillating with frequency ν. The jellium acts as an oscillator with eigenfrequency Ω. At the
present level of approximation it is an undamped oscillator. The jellium oscillator is excited at the frequency
ν. For 0 < ν < Ω, it is driven below its eigenfrequency and thus oscillates in phase with the test electron. The
amplitude, i.e., the jellium polarization, is enhanced compared to the ν = 0 limit simply because the system is
closer to the resonance at Ω. Therefore, the oscillating electron charge is overscreened. On the other hand, for
ν > Ω the jellium oscillator is driven above its eigenfrequency and thus follows the test electron with a phase
difference of π. Thus the electron charge is not screened at all but rather enhanced and the interaction is more
strongly repulsive than the pure Coulomb interaction.

82
Screening of the effective interaction
From our discussion of the Coulomb interaction we know that the real interaction between two electrons in a
metal is strongly screened at all except very short distances. This screening is well described within the RPA.
We now apply the RPA to the effective interaction derived above. We define

− Veff ≡ = + (8.79)

and
RPA
− Veff ≡ := + + + ··· (8.80)

or
RPA
−Veff (q, iνn ) = −Veff (q, iνn ) + Veff (q, iνn ) Π0 (q, iνn ) Veff (q, iνn )
− Veff (q, iνn ) Π0 (q, iνn ) Veff (q, iνn ) Π0 (q, iνn ) Veff (q, iνn ) + · · · (8.81)

As above, we can sum this up,

(iνn )2
RPA Veff (q, iνn ) (iνn )2 −Ω2
Veff (q, iνn ) = = VC (q) (iνn )2
1 + Veff (q, iνn )Π0 (q, iνn ) 1 + VC (q) Π0 (q, iνn )
(iνn )2 −Ω2
2
(iνn )
= VC (q)
(iνn )2 − Ω2 + (iνn )2 VC (q)Π0 (q, iνn )
VC (q) (iνn )2 + (iνn )2 VC (q)Π0 (q, iνn )
=
1 + VC (q)Π0 (q, iνn ) (iνn )2 − Ω2 + (iνn )2 VC (q)Π0 (q, iνn )
| {z }
= VCRPA (q,iνn )

(iνn )2
= VCRPA (q, iνn ) Ω2
(iνn )2 − 1+VC (q)Π0 (q,iνn )
(iνn )2
= VCRPA (q, iνn ) (8.82)
(iνn ) − ωq2 (iνn )
2

with the renormalized phonon frequency

Ω
ωq (iνn ) := p . (8.83)
1 + VC (q)Π0 (q, iνn )

To see that this is a reasonable terminology, compare Veff

RPA
to the bare effective interaction

(iνn )2
Veff (q, iνn ) = VC (q) . (8.84)
(iνn )2 − Ω2

Evidently, screening leads to the replacements VC → VCRPA and Ω → ωq .

For small momenta and frequencies, we have Π0 → N (EF ), the density of states at EF . In this limit we thus
obtain
Ω Ω Ω Ω
ωq ∼
=q =q ∼
=q 2 = q. (8.85)
e2
1 + 4π q2 N (EF ) κ2s
1 + q2
κs κs
q2

Due to screening we thus find an acoustic dispersion of jellium phonons. This is of course much more realistic
than an optical Einstein mode.
Beyond the low-frequency limit it is important that Π0 and thus ωq obtains a sizable imaginary part. It smears
out the pole in the retarded interaction Veff
RPA
(q, ν) or rather moves it away from the real-frequency axis—the
lattice vibrations are now damped. The real part of the retarded interaction is sketched here for fixed q:

83
 Re VeffRPA, R

VCRPA (q )
0
Re ωq ν

Note that
• the interaction still vanishes in the static limit ν → 0,
• the interaction is attractive for 0 < ν < Re ωq , where Re ωq ∼ q.
It is important that the static interaction is not attractive but zero. Hence, we do not expect static bound states
of two electrons.
To obtain analytical results, it is necessary to simplify the interaction. The main property required for
superconductivity is that the interaction is attractive for frequencies below some typical phonon frequency. The
typical phonon frequency is the material specific Debye frequency ωD . We write the effective RPA interaction in
terms of the incoming and transferred momenta and frequencies,
RPA
Veff RPA
= Veff (k, iωn ; k0 , iωn0 ; q, iνn ). (8.86)

k − q, i ω n − i νn q, i νn k’+ q , i ω’n + i νn

k, iω n k’, i ω’n

We then approximate the interaction (very crudely) by a constant −V0 < 0 if both incoming frequencies are
smaller than ωD and by zero otherwise,
(
−V0 for |iωn | , |iωn0 | < ωD ,
RPA
Veff ≈ (8.87)
0 otherwise.

This approximate interaction is evidently nonzero in the static limit and therefore could lead to bound states.
Such bound states would be an artifact of the approximation and should be disregarded.

84
9

Cooper instability and BCS ground state

In this chapter we will first show that the attractive effective interaction leads to an instability of the normal
state, i.e., of the Fermi sea. Then we will discuss the new state that takes its place.

9.1 Cooper instability

Let us consider the scattering of two electrons due to the effective interaction. A single scattering event is
represented by the diagram
k’, i ω’n k’+ q , i ω’n + i νn

k, iω n k − q, i ω n − i νn
Electrons can also scatter multiple times:

··· + + + + ··· (9.1)

An instability occurs if this series diverges since then the scattering becomes infinitely strong. The diagrams
represent a perturbative expansion in the interaction strength V0 about the noninteracting Fermi gas. A divergence
of the series signals a breakdown of perturbation theory. This means that the true equilibrium state cannot be
obtained from the equilibrium state for V0 = 0, i.e., the Fermi gas, by perturbation theory. A state that is
perturbatively connected to the free Fermi gas is called a Landau Fermi liquid. It is an appropriate description
for normal metals. Hence, a scattering instability implies that the equilibrium state is not a Fermi liquid.
Like in the RPA, it turns out to be sufficient to consider the dominant diagrams at each order. These are
the ladder diagrams, which do not contain crossing interaction lines. Moreover, the instability occurs first for
the scattering of two electrons with opposite momentum, frequency, and spin. We thus restrict ourselves to the
diagrams describing this situation. It is plausible that the scattering between states with opposite momenta
dominates: the total momentum of the two electrons is then zero, of course, which gives the largest phase space
of possible final states.

p
k
−p
−k

85
 We define the scattering vertex Λ by
− k − iωn − p − iΩ n

−Λ ≡
1111
0000
0000
1111 Λ

k iωn
0000
1111 p iΩ n
− k − iωn − p − iΩ n − k − iωn − k1 − i ω1n − p − iΩ n

k − k1 , k1 − p,
:= k − p, i ω n − iΩ n + + ··· (9.2)
i ω n − i ω1n i ω1n − iΩ n

k iωn p iΩ n k iωn k1 i ω1n p iΩ n

This is a geometric series, which we can sum up,

1111
0000
0000
1111
0000
1111
Λ = 1 + + + ··· = . (9.3)

1−

With our approximation (

−V0 for |iωn | , |iΩn | < ωD ,
RPA
Veff ≈ (9.4)
0 otherwise,
we obtain
+V0
− Λ(iωn , iΩn ) ≈ 1 1 (9.5)
Gk00 ↑ (iωn0 ) G−k
0 0
P P
1 − V0 β 0 , |iω 0 | < ω
iωn n D V k0 0 ↓ (−iωn )

for |iωn | , |iΩn | < ωD and zero otherwise. Thus


−V0
for |iωn | , |iΩn | < ωD ,


1 1 0 0 0 0
P P
Λ(iωn , iΩn ) ≈ 1 − V0 β 0 , |iω 0 | < ω
iωn n D V k0 Gk0 ↑ (iωn ) G−k0 ↓ (−iωn ) (9.6)
otherwise.

0

We see that the scattering vertex Λ diverges if

1 X 1 X 0
V0 Gk0 ↑ (iωn0 ) G−k
0 0
0 ↓ (−iωn ) = 1. (9.7)
β 0
V 0
iωn k
|iωn0 | < ωD
This expression depends on temperature. We now evaluate it explicitly:
Z∞
X 1 X 1 1 X 1
. . . = V0 kB T 0 0
= V0 kB T dξ D(µ + ξ) , (9.8)
0
V 0 iωn − ξk −iωn − ξk0
0
0
(ωn0 )2 + ξ2
iωn k iωn −∞
0 0
|iωn | < ωD |iωn | < ωD
where D() is the density of states per spin direction and per unit cell. Assuming the density of states to be
approximately constant close to the Fermi energy, we get (with ~ = 1)
Z∞
X dξ
· · · ≈ V 0 kB T D(EF )
0
(ωn0 )2 + ξ 2
iωn −∞
|iωn0 | < ωD | {z }
0 |
= π/|ωn

86
 βωD /2π βωD /2π
X 1 X 1
= V0 kB T D(EF ) π 2 (2n+1)π
= V0 
kB
T D(EF ) β

1
n=0


n=0
n+ 2
β
  
∼ βωD
= V0 D(EF ) γ + ln 4 . (9.9)
2π

Here, γ ≈ 0.577216 is the Euler constant. In the last step we have used an approximation for the sum over n that
is valid for βωD  1, i.e., if the sum has many terms. Since Tc for superconductors is typically small compared
to the Debye temperature ωD /kB (a few hundred Kelvin), this is justified. Altogether, we find

−V0
Λ≈   (9.10)
1 − V0 D(EF ) γ + ln 2βω
π
D

for |iωn | < ωD . Coming from high temperatures, but still satisfying kB T  ωD , multiple scattering enhances Λ.
Λ diverges at T = Tc , where
 
2ωD
V0 D(EF ) γ + ln =1 (9.11)
πkB Tc
2eγ ωD 1
⇒ ln = (9.12)
πkB Tc V0 D(EF )
2eγ
 
1
⇒ kB Tc = ωD exp − . (9.13)
π
|{z} V0 D(EF )
≈ 1.13387 ∼ 1

This is the Cooper instability. Its characteristic temperature scale appears to be the Debye temperature of a
few hundred Kelvin. This is disturbing since we do not observe an instability at such high temperatures, except
perhaps in H3 S and other hydrides. However, the exponential factor tends to be on the order of 1/100 so that we
obtain Tc of a few Kelvin.
It is important to realize that we find an instability of the normal metal towards a superconducting state for
any, arbitrarily weak attractive interaction. It is therefore called a weak-coupling instability. This behavior is
not shared by all mean-field theories: for example, the Stoner theory of ferromagnetism in metals predicts an
instability only if the (Hubbard) interaction U exceeds a critical value of Uc ∝ 1/D(EF ). The weak-coupling
Cooper instability helps to understand why superconductivity is so common among the elemental metals.

0.6
kBTc / ωD

0.4

0.2

0
0 0.5 1 1.5 2
V0 D(EF)

Also note that kB Tc is not analytic in V0 at V0 = 0 (the function has an essential singularity there). Thus
kB Tc cannot be expanded into a Taylor series around the non-interacting limit. This means that we cannot obtain
kB Tc in perturbation theory in V0 to any finite order. BCS theory is indeed non-perturbative.

87
9.2 The BCS ground state
We have seen that the Fermi sea becomes unstable due to the scattering of electrons in states |k, ↑i and |−k, ↓i.
Bardeen, Cooper, and Schrieffer (BCS) have proposed an ansatz for the new ground state. It is based on the
idea that electrons from the states |k, ↑i and |−k, ↓i form (so-called Cooper) pairs and that the ground state is
a superposition of states built up of such pairs. The ansatz reads
Y 
|ψBCS i = uk + vk c†k↑ c†−k,↓ |0i , (9.14)
k

where |0i is the vacuum state without any electrons and uk , vk are as yet unknown complex coefficients. Normal-
ization requires
1 = hψBCS |ψBCS i
exact reordering,
uk0 + vk0 c†k0 ↑ c†−k0 ,↓ |0i
Y
Y
u∗k + vk∗ c−k,↓ ck↑


= h0|
combining factors with same k
k k0

u∗k vk c†k↑ c†−k,↓

Y 2 2
+ uk vk∗ c−k,↓ ck↑ + |vk |


= h0| |uk | + |0i
k
Y 2 2
(9.15)


= |uk | + |vk | .
k

2 2
This is certainly satisfied if we demand |uk | + |vk | = 1 for all k, which we will do from now on.
Note that the occupations of |k, ↑i and |−k, ↓i are maximally correlated; either both are occupied or both are
empty. Also, |ψBCS i is peculiar in that it is a superposition of states with different total electron numbers. This
means that the electron number fluctuates; we are dealing with quantum fluctuations here since we are considering
a pure state. The fluctuations between a certain pair being present or absent agree with the earlier statement that
the superconducting state should not consist of statically bound pairs. (We could imagine the superconductor to be
entangled with a much larger electron reservoir so that the total electron number in superconductor and reservoir
is fixed.) As a consequence, expressions containing unequal numbers of electronic creation and annihilation
operators can have non-vanishing expectation values. For example,
c†k↑ c†−k,↓ BCS
≡ hψBCS | c†k↑ c†−k,↓ |ψBCS i
u∗k0 + vk∗ 0 c−k0 ,↓ ck0 ↑ c†k↑ c†−k,↓ uk00 + vk00 c†k00 ↑ c†−k00 ,↓ |0i
Y
Y

= h0|
k0 k00

h0| vk∗ c−k,↓ c†−k,↓ uk00 + vk00 c†k00 ↑ c†−k00 ,↓ |0i

Y
Y
u∗k0 + vk∗ 0 c−k0 ,↓ ck0 ↑


=
k0 6=k k00

since cc = 1 − c c and h0| c = 0

† † †

uk00 + vk00 c†k00 ↑ c†−k00 ,↓ |0i

Y
Y
= h0| vk∗ u∗k0 + vk∗ 0 c−k0 ,↓ ck0 ↑

k0 6=k k00

= h0| vk∗ uk + vk c†k↑ c†−k,↓

Y 2 2

|uk0 | + |vk0 | |0i = vk∗ uk . (9.16)
k0 6=k
| {z }
=1

The coefficients uk , vk are chosen so as to minimize the expectation value of the energy, hψBCS |H| ψBCS i, under
2 2
the constraint |uk | + |vk | = 1 for all k. |ψBCS i is thus a variational ansatz.
We write the Hamiltonian as
ξk c†kσ ckσ + Vint
X
H= (9.17)
kσ
and, in the spirit of the previous section, choose the simplest non-trivial approximation for Vint that takes the
following into account:
1. Only electrons with energies |ξk | . ωD relative to the Fermi energy are important. Note that this surrepti-
tiously replaces what was originally a cutoff in frequency by a cutoff in momentum.

88
 2. The instability is due to the scattering between electrons in the single-particle states |k, ↑i and |−k, ↓i.
These are related to each other by time reversal.
This leads to
1 X
Vint = Vkk0 c†k↑ c†−k,↓ c−k0 ,↓ ck0 ↑ (9.18)
N 0
kk

with (
−V0 for |ξk | < ωD and |ξk0 | < ωD ,
Vkk0 = (9.19)
0 otherwise.
This simplified model is in fact somewhat strange and should not be overinterpreted. We come back to this in
Sec. 10.3.
We assume that scattering without momentum transfer, k0 = k, contributes negligibly compared to k0 6= k
since there are many more scattering channels for k0 6= k. We also assume that u−k = uk , v−k = vk , which is, at
worst, a restriction of our variational ansatz. With this, we obtain
Y 
u∗q + vq∗ c−q,↓ cq↑ c†kσ ckσ uq0 + vq0 c†q0 ↑ c†−q0 ,↓ |0i
X Y

hψBCS |H| ψBCS i = ξk h0|
kσ q q0
1 X Y 
u∗q + vq∗ c−q,↓ cq↑ c†k↑ c†−k,↓ c−k0 ↓ ck0 ↑ uq0 + vq0 c†q0 ↑ c†−q0 ,↓ |0i . (9.20)
Y

+ Vkk0 h0|
N 0 q 0
kk q

Only those terms in the products over q, q0 contribute that make sure that the same fermions are first created
and then annihilated. For the interaction term, this is somewhat subtle. We discuss the first product, over q.
Special values of q that must be considered separately are q = k and q = k0 . For q = k, the two electrons created
by c†k↑ c†−k,↓ must be annihilated by c−q,↓ cq↑ so that only the term containing vq∗ = vk∗ survives. For q = k0 ,
the vq∗ term drops out since the same annihilation operators appear squared. Thus only the term containing
u∗q = u∗k0 remains. The arguments for the second product, over q0 , are analogous. For all q, q0 6= k, k0 , the
four operatorsQ in the center are irrelevant and we obtain the same expression as in the normalization integral,
containing q (|uq |2 + |vq |2 ) = 1. Altogether, we obtain

ξk h0| |vk | c−k,↓ ck↑ c†k↑ ck↑ c†k↑ c†−k,↓ |0i + ξk h0| |v−k | ck↓ c−k,↑ c†k↓ ck↓ c†−k,↑ c†k↓ |0i
X 2
X 2
hψBCS |H| ψBCS i =
k k
1 X
+ Vkk0 h0| vk∗ u∗k0 uk vk0 c−k,↓ ck↑ c†k↑ c†−k,↓ c−k0 ,↓ ck0 ↑ c†k0 ↑ c†−k0 ,↓ |0i
N 0
kk
X 2 1 X
= 2ξk |vk | + Vkk0 vk∗ uk u∗k0 vk0 =: EBCS . (9.21)
N 0
k kk

This energy should be minimized with respect to the uk , vk . For EBCS to be real, the phases of uk and vk must
be the same. But since EBCS is invariant under

uk → uk eiϕk , vk → vk eiϕk , (9.22)

we can choose all uk , vk real. The constraint from normalization then reads u2k + vk2 = 1 and we can parameterize
the coefficients by
uk = cos θk , vk = sin θk . (9.23)
Then
X 1 X
EBCS = 2ξk sin2 θk + Vkk0 sin θk cos θk sin θk0 cos θk0
N 0
k kk
X 1 X Vkk0
= ξk (1 − cos 2θk ) + sin 2θk sin 2θk0 . (9.24)
N 0 4
k kk

89
We obtain the minimum from
∂EBCS 1 X Vqk0 1 X Vkq
= 2ξq sin 2θq + cos 2θq sin 2θk0 + sin 2θk cos 2θq
∂θq N 0 2 N 2
k k
1 X !
= 2ξq sin 2θq + Vqk0 cos 2θq sin 2θk0 = 0. (9.25)
N 0
k

We replace q by k and parameterize θk by

∆k
sin 2θk =: p (9.26)
ξk2
+ ∆2k

and write
ξk
cos 2θk = p . (9.27)
ξk2+ ∆2k
The last equality is only determined by the previous one up to the sign. We could convince ourselves that the
other possible choice does not lead to a lower energy EBCS . Equation (9.25) now becomes

ξk ∆ k 1 X ξk ∆k0
2p 2 2
+ Vkk0 p 2 p =0 (9.28)
ξk + ∆ k N 0 ξk + ∆2k ξk2 0 + ∆2k0
k
1 X ∆k 0
⇒ ∆k = − Vkk0 p 2 . (9.29)
N 0 2 ξk0 + ∆2k0
k

This is called the BCS gap equation for reasons to be discussed below. When we have solved it, it is easy to
obtain the original variational parameters in terms of ∆k ,
!
1 ξk
u2k = 1+ p 2 , (9.30)
2 ξk + ∆2k
!
1 ξk
2
vk = 1− p 2 , (9.31)
2 ξk + ∆2k
∆k
u k vk = p 2 . (9.32)
2 ξk + ∆2k

The relative sign of uk and vk is thus the sign of ∆k . The absolute sign of, say, uk is irrelevant because of the
invariance of EBCS under simultaneous phase rotations of uk , vk (consider a phase factor of eiπ = −1).
For our special interaction (
−V0 for |ξk | , |ξk0 | < ωD ,
Vkk0 = (9.33)
0 otherwise,
the BCS gap equation (9.29) becomes

1 X ∆k0

− N

 (−V0 ) p 2 for |ξk | < ωD ,
2 ξk0 + ∆2k0
∆k = k0 (9.34)
 |ξ k0 | < ω D

otherwise.

0


Note that the right-hand side is independent of k as long as |ξk | < ωD holds. The gap equation can be solved by
the ansatz (
∆0 > 0 for |ξk | < ωD ,
∆k = (9.35)
0 otherwise.

90
We obtain
V0 ∆
p 0
X
∆0 = (9.36)
N 2 ξk2 0 + ∆20
k0
|ξk0 | < ωD
ZωD
V0 X 1 1
⇒ 1= p = V0 dξ D(µ + ξ) p , (9.37)
N 0 2 ξk2 0 + ∆20 2 ξ 2 + ∆20
k −ωD
|ξk0 | < ωD

where D() is again the density of states per spin direction and per unit cell. If the density of states is approxi-
mately constant within ±ωD of the Fermi energy, we obtain
ZωD
1 dξ ωD
1 = V0 D(EF ) p = V0 D(EF ) Arsinh (9.38)
2 ξ 2 + ∆20 ∆0
−ωD
1 ωD
⇒ sinh = (9.39)
V0 D(EF ) ∆0
1
⇒ ∆0 = ωD 1 . (9.40)
sinh V0 D(E F)

In the so-called weak-coupling limit of small V0 D(EF ) and thus large 1/V0 D(EF ), this result simplifies to
 
∼ 1
∆0 = 2ωD exp − . (9.41)
V0 D(EF )

Interestingly, apart from a numerical factor, the value of ∆0 agrees with kB Tc for the Cooper instability. We will
return to this observation below. The non-analyticity of the function ∆0 (V0 ) means that we cannot obtain ∆0
and thus |ψBCS i within perturbation theory for small V0 .
We can now find the energy gain due to the superconducting state, i.e., the condensation energy. For this, we
insert uk , vk into EBCS ,
!
X 1 ξk 1 X ∆k ∆k0
EBCS = 2

ξk 1− p 2 2
+ Vkk0 p 2 p . (9.42)
2 ξk + ∆ k N 0 4 ξk + ∆2k ξk2 0 + ∆2k0
k
kk

We use the simple form of Vkk0 and assume that D() is constant within ±ωD of the Fermi energy but not outside
of this interval. This gives
−ω
Z D
EBCS = N dξ D(µ + ξ) 2ξ
−∞
ZωD !
ξ
+N dξ D(µ + ξ) ξ 1− p
ξ 2 + ∆20
−ωD
Z∞ 

+N dξ D(µ +ξ)
 ξ (1 − 1)
 

ωD
ZωD ZωD
∆20
+N dξ dξ 0 D(µ + ξ)D(µ + ξ 0 )(−V0 ) p p
4 ξ 2 + ∆20 (ξ 0 )2 + ∆20
−ωD −ωD
−ω
Z D  
ωD
q
∼
= 2N 2 2 2
dξ D(µ + ξ) ξ + N D(EF ) −ωD ωD + ∆0 + ∆0 Arsinh
∆0
−∞

91
 ωD
− N V0 D2 (EF ) ∆20 Arsinh2 . (9.43)
∆0
With the gap equation
ωD
V0 D(EF ) Arsinh = 1, (9.44)
∆0
this simplifies to
−ω
Z D q
EBCS ∼
= 2N dξ D(µ + ξ) ξ − N D(EF ) ωD 2 + ∆2 .
ωD (9.45)
0
−∞

The normal-state energy should be recovered for ∆0 → 0. The energy difference is

q
∆EBCS := EBCS − EBCS |∆0 →0 ∼ = −N D(EF ) ωD ωD 2 + ∆2 + N D(E ) ω 2 .
0 F D (9.46)

Since for weak coupling we have ∆0  ωD , we can expand this in ∆0 /ωD ,

s  2
∼ ∆0 2 ∼ 1
2
∆EBCS = −N D(EF ) ωD 1 + + N D(EF ) ωD = − N D(EF ) ∆20 . (9.47)
ωD 2

The condensation-energy density is thus (counted positively)

1N
eBCS ∼
= D(EF ) ∆20 . (9.48)
2V
Type-I superconductivity is destroyed if eBCS equals the energy density required for magnetic-field expulsion. At
H = Hc , this energy is Hc2 /8π, as we have seen above. We thus conclude that
r
∼ N
Hc = 4π D(EF ) ∆0 . (9.49)
V
This prediction of BCS theory is in reasonably good agreement with experiments for simple superconductors.

92
10

BCS theory

The variational ansatz of Sec. 9.2 has given us an approximation for the many-particle ground state |ψBCS i. While
this is interesting, it does not yet allow predictions of thermodynamic properties, such as the critical temperature.
As a variational ansatz, it also does not provide much insight into the conceptual character of the approximation.
We will now consider superconductors at non-zero temperatures within mean-field theory, which will also provide
a new perspective on the BCS gap equation and on the meaning of ∆k .

10.1 BCS mean-field theory

We start again from the Hamiltonian
1 X
ξk c†kσ ckσ + Vkk0 c†k↑ c†−k,↓ c−k0 ,↓ ck0 ↑ .
X
H= (10.1)
N 0
kσ kk

A mean-field approximation consists of replacing products of operators A, B according to

AB ∼
= hAi B + A hBi − hAi hBi . (10.2)

Note that the error introduced by this replacement is

AB − hAi B − A hBi + hAi hBi = (A − hAi)(B − hBi), (10.3)

i.e., it is of second order in the deviations of A and B from their averages. A well-known mean-field approximation
is the Hartree or Stoner approximation, which for our Hamiltonian amounts to the choice A = c†k↑ ck0 ↑ , B =
c†−k,↓ c−k0 ,↓ . However, Bardeen, Cooper, and Schrieffer realized that superconductivity can be understood with
the help of a different choice, namely A = c†k↑ c†−k,↓ , B = c−k0 ,↓ ck0 ↑ . This leads to the mean-field BCS Hamiltonian

1 X  
ξk c†kσ ckσ + Vkk0 hc†k↑ c†−k,↓ i c−k0 ,↓ ck0 ↑ + c†k↑ c†−k,↓ hc−k0 ,↓ ck0 ↑ i − hc†k↑ c†−k,↓ ihc−k0 ,↓ ck0 ↑ i .
X
HBCS =
N 0
kσ kk
(10.4)
We define
1 X
∆k := − Vkk0 hc−k0 ,↓ ck0 ↑ i (10.5)
N 0
k

so that
1 X
∆∗k = − Vkk0 hc†k0 ↑ c†−k0 ,↓ i. (10.6)
N 0
k

At this point it is not obvious that the quantity ∆k is the same as the one introduced in Sec. 9.2 for the special
case of the ground state. Since this will turn out to be the case, we nevertheless use the same symbol from the

93
start. We can now write

ξk c†kσ ckσ − ∆k c†k↑ c†−k,↓ + const.

X X X
HBCS = ∆∗k c−k,↓ ck↑ − (10.7)
kσ k k

The constant is irrelevant for the following derivation and is omitted from now on. Since HBCS is bilinear in
c, c† it describes a non-interacting effective system. But what is unusual is that HBCS contains terms of the form
cc and c† c† , which do not conserve the electron number. We thus expect that the eigenstates of HBCS do not
have a sharp electron number. We had already seen that the BCS ground state has this property. This is a bit
strange since superpositions of states with different electron numbers are never observed (keyword: superselection
rules). One can formulate the theory of superconductivity in terms of states with fixed electron number, but this
formulation is cumbersome and we will not pursue it here.
To diagonalize HBCS , we introduce new fermionic operators, which are linear combinations of electron creation
and annihilation operators,    ∗  
γk↑ uk −vk ck↑
† = . (10.8)
γ−k,↓ vk∗ uk c†−k,↓
This mapping is called Bogoliubov (or Bogoliubov-Valatin) transformation. Again, it is not clear yet that uk , vk
are related to the previously introduced quantities denoted by the same symbols. For the γ to satisfy fermionic
anticommutation relations, we require
n o
† † †
γk↑ , γk↑ = γk↑ γk↑ + γk↑ γk↑

= u∗k uk ck↑ c†k↑ − u∗k vk∗ ck↑ c−k,↓ − vk uk c†−k,↓ c†k↑ + vk vk∗ c†−k,↓ c−k,↓
+ uk u∗k c†k↑ ck↑ − uk vk c†k↑ c†−k,↓ − vk∗ u∗k c−k,↓ ck↑ + vk∗ vk c−k,↓ c†−k,↓
n o n o n o
= |uk | ck↑ , c†k↑ −u∗k vk∗ {ck↑ , c−k,↓ } −vk uk c†−k,↓ , c†k↑ + |vk | c†−k,↓ , c−k,↓
2 2

| {z } | {z } | {z } | {z }
=0
=1 =0 =1
2 2 !
= |uk | + |vk | = 1. (10.9)

Using this constraint, we find the inverse transformation,

    
ck↑ uk vk γk↑
= . (10.10)
c†−k,↓ −vk∗ u∗k †
γ−k,↓

Insertion into HBCS yields

Xn  †

†
 
†

†

HBCS = ξk u∗k γk↑ + vk∗ γ−k,↓ uk γk↑ + vk γ−k,↓ + ξk −vk∗ γk↑ + u∗k γ−k,↓ −vk γk↑ + uk γ−k,↓
k
     o
† † † †
− ∆∗k −vk γk↑ + uk γ−k,↓ uk γk↑ + vk γ−k,↓ − ∆k u∗k γk↑ + vk∗ γ−k,↓ −vk∗ γk↑ + u∗k γ−k,↓
Xn 2 2

†
= ξk |uk | − ξk |vk | + ∆∗k vk uk + ∆k u∗k vk∗ γk↑ γk↑
k
 
2 2 †
+ −ξk |vk | + ξk |uk | + ∆∗k uk vk + ∆k vk∗ u∗k γ−k,↓ γ−k,↓
 
2 † †
+ ξk u∗k vk + ξk u∗k vk +∆∗k vk2 − ∆k (u∗k ) γk↑ γ−k,↓
| {z }
= 2 ξk u ∗
k vk
  o
2
+ ξk vk∗ uk + ξk vk∗ uk −∆∗k u2k + ∆k (vk∗ ) γ−k,↓ γk↑ + const. (10.11)
| {z }
∗u
= 2 ξk v k k

The coefficients uk , vk should now be chosen such that the γγ and γ † γ † terms vanish. This requires
2
2ξk u∗k vk + ∆∗k vk2 − ∆k (u∗k ) = 0 (10.12)

94
for the γ † γ † term and the condition for γγ is redundant. Writing

∆k = |∆k | eiφk , (10.13)

uk = |uk | e iαk
, (10.14)
vk = |vk | e iβk
, (10.15)

we obtain  
2 2
2ξk |uk | |vk | ei(βk −αk ) + |∆k | |vk | ei(2βk −φk ) − |uk | ei(φk −2αk ) = 0. (10.16)

A special solution of this equation (we do not require the general solution) is given by

αk = 0, (10.17)
βk = φk ,  (10.18)
2 2
2ξk |uk | |vk | + |∆k | |vk | − |uk | = 0. (10.19)
From the last equation we obtain
 
2 2 2 4 2 2 4
4ξk2 |uk | |vk | = |∆k | |vk | − 2 |vk | |uk | + |uk | (10.20)
     2
2 2 2 2 4 2 2 4 2 2 2 2
⇒ 4 ξk2 + |∆k | |uk | |vk | = |∆k | |vk | + 2 |vk | |uk | + |uk | = |∆k | |vk | + |uk | = |∆k | (10.21)
|∆k |
⇒ |uk | |vk | = q (10.22)
2
2 ξk2 + |∆k |

so that
2 2 2ξk |uk | |vk | ξk
|uk | − |vk | = =q . (10.23)
|∆k | 2
ξk2 + |∆k |
2 2
Together with |uk | + |vk | = 1 we thus find
 
2 1 ξk
|uk | = 1+ q , (10.24)
2 2
ξk2 + |∆k |
 
2 1 ξ k
|vk | = 1 − q . (10.25)
2 2 2
ξk + |∆k |

Restoring the phases in Eq. (10.22), we also conclude that

∆k
uk vk = q . (10.26)
2
2 ξk2 + |∆k |

∆k , uk , and vk need not be real1 but if they are, these relations are the same as in Sec. 9.2 for the BCS ground
state.
The BCS Hamiltonian now reads, ignoring a constant,
 
2 2
X ξ k |∆ k | 
 γ † γk↑ + γ † γ−k,↓

HBCS = q +q k↑ −k,↓
2 2
k ξk2 + |∆k | ξk2 + |∆k |
Xq 2

† †

= ξk2 + |∆k | γk↑ γk↑ + γ−k,↓ γ−k,↓ . (10.27)
k
1u was chosen real in the specific solution but this was not necessary.
k

95
Using ξ−k = ξk and the plausible assumption |∆−k | = |∆k |, both based on time-reversal symmetry, we obtain
the simple form
†
X
HBCS = Ek γkσ γkσ (10.28)
kσ

with the dispersion q

2
Ek := ξk2 + |∆k | . (10.29)
It is instructive to first consider the normal state, for which ∆k → 0. Then
 (
for ξk < 0,

2 1 ξk 0
|uk | = 1+ = (10.30)
2 |ξk | 1 for ξk > 0,
 (
for ξk < 0,

2 1 ξk 1
|vk | = 1− = (10.31)
2 |ξk | 0 for ξk > 0.

It follows that γkσ ∝ c†−k,−σ for ξk < 0, whereas γkσ ∝ ckσ or ξk > 0. Hence, the Bogoliubov quasiparticles
described by γ, γ † , are holes for energies below the chemical potential (ξk < 0) and electrons for energies above
(ξk > 0). Their dispersion is Ek = |ξk |. For a parabolic normal dispersion ξk :
Ek

hole electron
excitations excitations
0 kF k
√
Here, kF = 2mµ. The excitation energies Ek are always positive except at the Fermi surface—it costs energy
to create a hole in the Fermi sea and also to insert an q
electron into an empty state outside of the Fermi sea.
2
Superconductivity changes the dispersion to Ek = ξk2 + |∆k | :

Ek

∆ kF
0 kF k

Superconductivity evidently opens an energy gap of magnitude |∆kF | in the excitation spectrum.
We should recall that in deriving HBCS we have ignored a constant, which we now reinsert,
†
X
HBCS = EBCS + Ek γkσ γkσ . (10.32)
kσ

The energy of the system is EBCS if no quasiparticles are present and is increased (by at least |∆kF |) if quasi-
particles are excited. The state without any quasiparticles is the pure condensate. The fact that EBCS depends

96
on temperature through hc−k,↓ ck↑ i shows that the condensate is not generally the BCS ground state discussed
previously. However, one can show that it coincides with the ground state in the limit T → 0.
2 2
Furthermore, in the superconducting state we find 0 < |uk | < 1 and 0 < |vk | < 1, i.e., the Bogoliubov
quasiparticles are superpositions of particles and holes. Deep inside the Fermi sea, the quasiparticles are mostly
hole-like, while far above EF they are mostly electron-like. But right at the Fermi surface we find, for example
for spin σ = ↑,
1 1
γk↑ = √ ck↑ − √ eiφk c†−k,↓ . (10.33)
2 2
The quasiparticles here consist of electrons and holes with the same amplitude. This means that they are
electrically neutral on average.

1 2
uk

2
vk
0 kF k

So far, we have not determined the gap function ∆k . This can be done by inserting the Bogoliubov transfor-
mation into the definition
1 X
∆k = − Vkk0 hc−k0 ,↓ ck0 ↑ i, (10.34)
N 0
k

which yields
1 X D  E
∆k = − Vkk0 −vk0 γk† 0 ↑ + uk0 γ−k0 ,↓ uk0 γk0 ↑ + vk0 γ−k
†
0 ,↓
N 0
k
1 X n D E D E
=− Vkk0 − vk0 uk0 γk† 0 ↑ γk0 ↑ + uk0 vk0 γ−k0 ,↓ γ−k
†
0 ,↓
N 0
k
D E o
− vk0 γk† 0 ↑ γ−k
2 †
0 ,↓ + u2k0 hγ−k0 ,↓ γk0 ↑ i . (10.35)

For selfconsistency, the averages have to be evaluated with the BCS Hamiltonian HBCS . This gives
D E
γk† 0 ↑ γk0 ↑ = nF (Ek0 ), (10.36)
D E
†
γ−k0 ,↓ γ−k 0 ,↓ = 1 − nF (Ek0 ), (10.37)
D E
γk† 0 ↑ γ−k
†
0 ,↓ = 0, (10.38)
hγ−k0 ,↓ γk0 ↑ i = 0, (10.39)

and we obtain the BCS gap equation, now at arbitrary temperature,

1 X
∆k = − Vkk0 uk0 vk0 [1 − 2 nF (Ek0 )]
N 0
k
1 X ∆k0
=− Vkk0 q [1 − 2 nF (Ek0 )]
N 0 2
k 2 ξk2 0 + |∆k0 |
1 X ∆k0
≡− Vkk0 [1 − 2 nF (Ek0 )] . (10.40)
N 0 2Ek0
k

97
We see that nF (Ek0 ) → 0 for Ek0 > 0 and T → 0 so that the zero-temperature BCS gap equation (9.29) is
recovered as a limiting case.
For the simplified model interaction given by Eq. (9.19) and assuming a k-independent real gap, we obtain,
in analogy to the ground-state derivation,
p 
ZωD 1 − 2 nF ξ 2 + ∆20 ZωD
tanh β2 ξ 2 + ∆20
p
1 = V0 dξ D(µ + ξ) p = V0 dξ D(µ + ξ) p . (10.41)
2 ξ 2 + ∆20 2 ξ 2 + ∆20
−ωD −ωD

Note that this only works for ∆0 6= 0 since we have divided by ∆0 . If the density of states is approximately
constant close to EF , the equation simplifies to
ZωD β
p
∼ V0 D(EF ) tanh ξ 2 + ∆20
1= dξ p2
. (10.42)
2 ξ 2 + ∆20
−ωD

This equation can be solved numerically—the numerically evaluated integral needs to be plugged into a numerical
root finder—leading to the temperature dependence of ∆0 :
∆0

∆ 0 (0)

0 Tc T

For weak coupling we have already seen that

 
1
∆0 (0) ∼
= 2ωD exp − . (10.43)
V0 D(EF )
We can also obtain an analytical expression for Tc : If T approaches Tc from below, we can take the limit ∆0 → 0
in the gap equation. This leads to the linearized gap equation
ZωD
tanh β2 |ξ|
1∼
= V0 D(EF ) dξ
2 |ξ|
−ωD
ZωD
tanh β2 ξ
= V0 D(EF ) dξ
ξ
0
βω
ZD /2
tanh x
= V0 D(EF ) dx
x
0
( βω
ZD /2 )
by parts βωD βωD ln x
= V0 D(EF ) ln tanh − dx . (10.44)
2 2 cosh2 x
0

In the weak-coupling limit we have βωD = ωD /kB T  1 (this assertion should be checked a-posteriori ). Since
the integrand of the last integral decays exponentially for large x, we can send the upper limit to infinity,
( Z∞ )  
∼ βω D βω D ln x ∼ βωD π
1 = V0 D(EF ) ln tanh − dx = V0 D(EF ) ln + γ − ln , (10.45)
2 | {z 2 } cosh2 x 2 4
0
∼
=1

98
where γ is again the Euler gamma constant. This implies
 
1 ∼ 2βωD γ
exp = e (10.46)
V0 D(EF ) π
2eγ
 
∼ 1
⇒ kB Tc = ωD exp − . (10.47)
π V0 D(EF )

This is exactly the same expression we have found above for the critical temperature of the Cooper instability
in Sec. 9.1. Since the approximations used are quite different, this agreement is not trivial. The gap at zero
temperature and the critical temperature thus have a universal ratio in BCS theory,

2∆0 (0) ∼ 2π
= γ ≈ 3.528. (10.48)
kB Tc e
This ratio is close to the result measured for simple elemental superconductors. For example, for tin one finds
2∆0 (0)/kB Tc ≈ 3.46. For superconductors with stonger coupling, such as mercury, and for unconventional
superconductors the agreement is not good, though. In part, this is due to the crude model for the (temperature-
independent) interaction, not so much to a failure of the mean-field approximation.

10.2 Minimization of the free energy

In the previous section, we have obtained the BCS state by decoupling the interaction in the “Cooper channel,”
leading to Eq. (10.4), and choosing averages such as hc†k↑ c†−k,↓ i selfconsistently. The state so obtained is the one
that minimizes the free energy, which is surely the right thing to do. It also gives a hint that BCS theory is
consistent with Landau theory. We will return to this point below. In this section, we derive the BCS state by
minimizing the free energy explicitly.
We consider a constant attractive interaction V = −V0 with high-energy cutoff Λ, see Eq. (9.19). Above, this
was the Debye energy. We continue to assume that ∆k = ∆ is momentum independent and use this assumption
from the start be writing
V0 X D E
∆= c−k,↓ ck↑ , (10.49)
N
k
V0
XD † † E
∆∗ = ck↑ c−k,↓ . (10.50)
N
k

In the derivation of the free energy, it is crucial to keep all energy contributions. The BCS Hamiltonian of Eq.
(10.4) now reads
  
V0 N N
ξk c†kσ ckσ − ∆ c†k↑ c†−k,↓ +
X X X
HBCS = ∆∗ c−k,↓ ck↑ − − ∆∗ − ∆
N V0 V0
kσ k k
∗
∆ ∆
ξk c†kσ ckσ − ∆ c†k↑ c†−k,↓ + N
X X X
= ∆∗ c−k,↓ ck↑ − . (10.51)
V0
kσ k k

The terms containing operators can usefully be written in terms of two-component (“Nambu-spinor”) operators
 
ck,↑  
Ψk := † , Ψ†k = c†k↑ , c−k,↓ . (10.52)
c−k,↓

Note that the spinors each contain a creation and an annihilation operator. We rewrite the normal part of the
Hamiltonian as

ξk c†kσ ckσ = ξk c†k↑ ck↑ + ξk c†k↓ ck↓

X X X

kσ k k

99
 ξk c†k↑ ck↑ + ξ−k c† c−k,↓
X X
=
|{z} −k,↓
k k
= ξk

ξk c†k↑ ck↑ ξk c−k,↓ c†−k,↓ +

X X X
= − ξk . (10.53)
k k k

This allows us to write  

ξk −∆ N
Ψ†k
X X
HBCS = Ψk + ξk + |∆|2 . (10.54)
−∆∗ −ξk V0
k k

The term k ξk is a constant energy that does not depend on the gap parameter ∆ and can thus be dropped
P
when we determine ∆. The next step is to diagonalize the 2 × 2 matrices. Here, we only need the eigenvalues.
We do not require the eigenvectors, or, equivalently, the explicit unitary transformation. Of course, this is just
the Bogoliubov transformation discussed above. We introduce
 
γk,↑  
Γk := † , Γ†k = γk↑ †
, γ−k,↓ . (10.55)
γ−k,↓

and obtain
p 2 
ξk + |∆|2 p 0 N
Γ†k
X
HBCS = Γk + |∆|2
0 − ξk2 + |∆|2 V0
k
Xq †
Xq N
= ξk2 + |∆|2 γkσ γkσ − ξk2 + |∆|2 + |∆|2 . (10.56)
V0
kσ k

Since the quasiparticles do not interact in the BCS approximation the partition function can be written as a
product:
  !
N Xq Y  q 
2
Z = exp β |∆| exp β ξk2 + |∆|2 1 + exp −β ξk2 + |∆|2
V0
k kσ
 Y  q   q 2
N 2 2 2 2 2
= exp β |∆| exp β ξk + |∆| 1 + exp −β ξk + |∆|
V0
k
Y" p ! p !#2
β ξk2 + |∆|2 β ξk2 + |∆|2

N 2
= exp β |∆| exp + exp −
V0 2 2
k
p
β ξk2 + |∆|2
 Y
N
= exp β |∆|2 4 cosh2
V0 2
k
 Y
N βEk
= 4N exp β |∆|2 cosh2 , (10.57)
V0 2
k

with Ek = ξk2 + |∆|2 . Strictly speaking, this is the grand-canonical partition function, which is not obvious
p

since the chemical potential µ is hidden in ξk = k − µ. The free energy (or more correctly the grand potential)
is then
N X βEk
F = −kB T ln Z = −N kB T ln 4 + |∆|2 − kB T ln cosh2
V0 2
k
N X βEk
= −2N kB T ln 2 + |∆|2 − 2kB T ln cosh . (10.58)
V0 2
k

It is useful to also obtain the limit for T → 0:

N X eβEk /2
lim F = |∆|2 − lim 2kB T ln
T →0 V0 T →0 2
k

100
 N X Ek
= |∆|2 − lim 2k
BT

V0 T →0 2k
 BT

k
X N
=− Ek + |∆|2 . (10.59)
V0
k

This agrees with Eq. (10.56): At zero temperature, none of the quasiparticle states is occupied since they all have
positive energies Ek > 0.
The BCS gap equation is obtained by minimizing F with respect to |∆|:

! dF 2N X sinh βEk β dEk

2
0= = |∆| − 
2kBT

βEk 2
 d|∆|
d|∆| V0 cosh 2
 x
k
2N βEk d
X q
= |∆| − tanh ξk2 + |∆|2
V0 2 d|∆|
k
2N X βEk |∆|
= |∆| − tanh p , (10.60)
V0 2 2
ξk + |∆|2
k

which gives
V0 X |∆| βEk
|∆| = tanh . (10.61)
2N Ek 2
k

With
βEk eβEk /2 − e−βEk /2 eβEk − 1 eβEk + 1 − 2 2
tanh = βE /2 −βE /2
= βE
= βE
= 1 − βE = 1 − 2nF (Ek ) (10.62)
2 e k +e k e k +1 e k +1 e k +1
we obtain
V0 X |∆|
|∆| = [1 − 2nF (Ek )] , (10.63)
2N Ek
k

which is the BCS gap equation of the last section, for the special case of constant V and ∆. Hence, the two
approaches are equivalent, as they should be.
Now let us look at the free energy more closely. We consider a parabolic band structure ξk = k 2 /2m − µ but
the essential results do not depend on this. Our goal is to identify the leading dependence on |∆|. Dropping a
|∆|-independent term, we obtain from Eq. (10.58),
N X βEk
F = |∆|2 − 2kB T ln cosh
V0 2
k
Z 3
N d k βEk
= |∆|2 − 2kB T N vuc ln cosh vuc is the volume of the primitive unit cell
V0 (2π)3 2
Z
N 1 βEk
= |∆|2 − 2kB T N vuc 2 dk k 2 ln cosh . (10.64)
V0 2π 2
√
Substituting  = k 2 /2m, we get k = 2m and thus
Z Λ
N kB T N vuc 3/2 1
√ βp
F = |∆|2 − 2
(2m) d  ln cosh ( − µ)2 + |∆|2 , (10.65)
V0 π 2 0 2
where we have made the cutoff Λ explicit. Now comes a subtle point: while the free energy is dominated by
contributions from energy  far from the chemical potential (for such  the integrand is large), the strongest |∆|
dependence results from  close to µ. This is plausible: only if ( − µ)2 under the root is small can |∆|2 play a
significant role. Hence in evaluating the leading |∆| dependence, we can assume  ≈ µ outside of Ek :
Z Λ
∼ N 2 kB T N vuc (2m)3/2 √ βp
F = |∆| − µ d ln cosh ( − µ)2 + |∆|2
V0 2π 2 0 2

101
 Λ−µ
kB T N vuc (2m)3/2 √
Z
N βp 2
= |∆|2 − µ d ln cosh  + |∆|2 . (10.66)
V0 2π 2 −µ 2

The cutoff is large compared to all other energies so that we can replace Λ − µ by Λ as the upper limit. Now
as assume, in addition, that µ  |∆|, which is the case for typical superconductors. As noted, |∆| only plays a
significant role for || . |∆|  µ. For the leading |∆| dependence, we can thus extend the lower limit to some
value that is large in magnitude. For convenience, we use the same cutoff as for the upper limit (using different
cutoffs does not lead to any new insight) and write this as
Λ
kB T N vuc (2m)3/2 √
Z
N βp 2
F ∼
= |∆|2 − µ  + |∆|2
d ln cosh
V0 2π 2 −Λ 2
Z Λ
N kB T N vuc (2m)3/2 √ βp 2
= |∆|2 − µ d ln cosh  + |∆|2 . (10.67)
V0 π2 0 2

At low temperatures kB T  |∆|, the argument of the hyperbolic cosine is large compared to unity and we have

ex + e−x ∼ ex
ln cosh x = ln = ln = x − ln 2 ∼
=x (10.68)
2 2
and thus
Z Λ
N kB
T N vuc (2m)3/2 √ βp 
F ∼
= |∆|2 −  2
µ d
2 + |∆|2 u=
V0 π 0 2 |∆|
3/2 2 Z Λ/|∆|
N N vuc (2m) √ |∆| p
= |∆|2 − µ du u2 + 1. (10.69)
V0 π2 2 0

The integral is
Z Λ/|∆| iΛ/|∆|
p 1h p 2  p
du u2 + 1 = u u + 1 + ln u + u2 + 1
0 2 0
p p !
1 Λ Λ2 + |∆|2 Λ + Λ2 + |∆|2
= + ln (10.70)
2 |∆|2 |∆|

and with Λ  |∆|,

Λ2 Λ2
   
1 2Λ 1 |∆|
... ∼
= 2
+ ln ∼
= 2
− ln . (10.71)
2 |∆| |∆| 2 |∆| Λ
This gives
√ 
N vuc (2m)3/2 µ

N |∆|
F ∼
= |∆|2 − Λ2
− |∆| 2
ln
V0 4π 2 Λ
3/2 √
N N vuc (2m) µ |∆|
= |∆|2 + |∆|2 ln + const, (10.72)
V0 4π 2 Λ
as far as the |∆| dependence is concerned. The first term results from the operator-free term hAihBi in the general
mean-field decoupling, while the second is due to the effect of the gap opening on the quasiparticle density of
states. The term ln(|∆|/Λ) is called the “Cooper logarithm.” It is crucial for understanding the low-temperature
behavior, as we discuss next.
Note that the first term, (N/V0 ) |∆|2 , is positive. In the second term, the prefactor and |∆|2 are positive but
the Cooper logarithm is negative since |∆| < Λ. We write
√
N vuc (2m)3/2 µ
 
1 |∆|
F ∼
= const + N + 2
ln |∆|2 . (10.73)
V0 4π Λ
|{z} | {z } | {z }
>0 >0 <0

102
For |∆| → 0, the logarithm diverges to −∞, though very slowly. Hence, for sufficiently small |∆|, it will always
overpower the constant 1/V0 so that the free energy initially decreases when |∆| is increased starting from zero.
Now we consider the free energy at large |∆|. Here, we have
q
Ek = ξk2 + |∆|2 ∼ = |∆| (10.74)

and the free energy of Eq. (10.58) becomes

N β|∆|
F ∼
X
= −2N kB T ln 2 + |∆|2 − 2kB T ln cosh . (10.75)
V0 2
k

Using |∆|  kB T , we get

N X eβ|∆|/2
F ∼
= −2N kB T ln 2 + |∆|2 − 2kB T ln
V0 2
k
( N β|∆|
−2N |∆|2 − X
(
=( (( kB(T(
ln 2 + 2kXTN A
BX +(
2k(
BT(N(ln(2
V0 2A
N
= |∆|2 − N |∆|. (10.76)
V0
Thus the positive, quadratic term is large compared to the negative, linear one and the free energy increases
with increasing gap parameter |∆|. In summary, for small |∆|, the contribution from the quasiparticle energies
increases towards negative values slightly (logarithmically) faster than the term from the mean-field decoupling
increases towards positive values. On the other hand, at large |∆|, the increasing decoupling term dominates over
the quasiparticle contribution.

F(∆) −F(0)

small
|∆|

The range of F (∆) − F (0) < 0 is small and restricted to very small |∆| since it relies on the logarithm of |∆|
being large. Nevertheless, the minimum free energy always occurs at |∆| > 0, i.e., there is always superconductivity
at low temperatures when the interacting is attractive. Hence, we have recovered the weak-coupling nature of
the instability discussed in the previous chapter. Note that we have not considered competing instabilities—the
system might for example order magnetically, preventing the superconducting instability.

10.3 Richardson’s solution of the BCS model

We return to the simplified BCS Hamiltonian of Eqs. (9.18) and (9.19), repeated here:
1 X
ξk c†kσ ckσ + Vkk0 c†k↑ c†−k,↓ c−k0 ,↓ ck0 ↑ ,
X
H= (10.77)
N 0
kσ kk

where (
−V0 for |ξk | < ωD and |ξk0 | < ωD ,
Vkk0 = (10.78)
0 otherwise.

103
The coupling Vkk0 is thus constant within a shell surrounding the Fermi surface and zero outside. Since ωD is
typically small compared to the chemical potential this shell is thin and we can expand the normal-state dispersion
to linear order. This implies that in the direction normal to the Fermi surface, the shell extends by ωD /vF in
momentum, where vF is the Fermi velocity.
Above, we have solved this interacting model within the mean-field approximation. Would it not be nice to be
able to go beyond this or ideally to solve the model exactly? Amazingly, this is possible, at least in the number
of relevant momenta is finite. The solution goes back to Robert Richardson [R. W. Richardson, Phys. Lett. 3,
277 (1963); see also G. Gorohovsky and E. Bettelheim, Phys. Rev. B 84, 224503 (2011)] for an equivalent model
applied to atomic nuclei. It uses a variant of the so-called Bethe ansatz. We will not present the solution here
but make a few remarks.
The BCS Hamiltonian involves scattering of electron pairs with momenta k and −k. Evidently, there is no
scattering and hence no interaction between electrons with k outside of the thin shell. These electrons are thus
free and the solution of the many-particle problem for them is trivial—the single-particle states are occupied
according to the Fermi distribution. Furthermore, if a pair with momenta k and −k is singly occupied the
interaction gives zero and such electrons are also effectively free. Let us denote the number of empty and doubly
occupied pairs of momenta within the shell by nk . Richardson showed that all eigenvalues of the many-particle
Hamiltonian of these 2nk particles can be obtained by solving a set of nk coupled nonlinear algebraic equations.
While this is not trivial, it is much easier than diagonalizing the many-particle Hamiltonian, which has dimension
2nk (not 22nk since singly occupied pairs are excluded), corresponding to 2nk coupled linear algebraic equations.
So far, so good. But what does the solution tell us? Fourier transformation onto Wannier modes localized at
lattice sites reads as
1 X ik·r
crσ = √ e ckσ , (10.79)
N k
1 X −ik·r
ckσ = √ e crσ . (10.80)
N r

First ignoring the momentum cutoff, the interaction in momentum space is

V0 X † †
Vint = − ck↑ c−k,↓ c−k0 ,↓ ck0 ↑ . (10.81)
N 0kk

The interaction in real space is then

V0 0 0 0
eik ·(r2 −r1 ) c†r1 ↑ c†r2 ↓ cr02 ↓ cr01 ↑
X X X
Vint = − eik·(r1 −r2 )
N3
r1 r2 r01 r02 k k0
| {z }| {z }
= N δr1 r2 = N δr0 r0
1 2

V0 X † †
=− c c cr0 ↓ cr0 ↑ . (10.82)
N 0 r↑ r↓
rr

This is very strange. The interaction term scatters a localized pair from r0 to r and thus violates the local
conservation of charge. It not even decreases with separation; Vint describes arbitrary-range pair hopping. After
restoring the momentum cutoff, the k and k0 sum are performed over the momentum shell surrounding the Fermi
surface. This can be done explicitly but the result is clear: The Kronecker delta symbols are replaced by decaying
function of r1 − r2 and r01 − r02 . The characteristic decay length is the inverse of the smallest characteristic
momentum, which is the thickness 2ωD /vF of the momentum shell. Hence, Vint scatters a pair centered at r0 and
delocalized on a scale of vF /2ωD into a pair centered at r and similarly delocalized. Thus the cutoff does not
affect the main unphysical feature of Vint , i.e., the arbitrary-range hopping.
Exactly solving an unphysical model likely leads to unphysical results. The conclusion is that one should
not use the simplified BCS Hamiltonian together with the exact Richardson solution for extended systems. The
model does make a lot of sense within BCS mean-field theory. The restriction of the interaction to Vint was mainly
motivated by the well-founded assumption that only the zero-momentum averages hc†k↑ c†−k,↓ i become nonzero at

104
the mean-field transition. The choice of Vint is in this sense part of the mean-field approximation. It should be
added that the exact Richardson solution is useful when the nonlocality of the hopping is not crucial. This is the
case for small systems such as nuclei and superconducting quantum dots.

10.4 Photoemission
The most direct probe of the quasiparticle dispersion in superconductors is angle-resolved photoemission spec-
troscopy (ARPES). Dropping the constant term from mean-field decoupling, the BCS Hamiltonian reads
†
X X
HBCS = Ek γkσ γkσ − Ek
kσ k
1X † 1X 
†
 X
= Ek γkσ γkσ + Ek 1 − γkσ γkσ − Ek
2 2
kσ kσ k
1 X † †

= Ek γkσ γkσ − Ek γkσ γkσ , (10.83)
2
kσ

with Ek = ξk2 + |∆|2 or, generalized for a momentum-dependent gap parameter, Ek = ξk2 + |∆k |2 . The
p p
†
final form of HBCS exhibits the symmetry of the dispersion relative to zero energy (i.e., the Fermi energy). γkσ
can be understood as creating a quasiparticle with positive energy Ek and γkσ can be understood as creating a
quasiparticle with negative energy −Ek . This interpretation double counts the degrees of freedom—there now
appear to be four instead of two per momentum k—which is cured by the prefactor 1/2.
Ek

0 kF k

visible in ARPES

In this form, the quasiparticle states with negative energies are occupied in the ground state and it is thus
impossible to create an additional quasiparticle here because of the Pauli principle. This is consistent with our
new interpretation: these states would be created by γkσ , which also annihilates quasiparticles with positive
energies, which do not exist in the ground state, hence the process is impossible. At nonzero temperatures, the
occupation is described by the Fermi function so that some negative-energy states are empty and some positive-
energy states are occupied.
In photoemission experiments, light of known frequency is directed at the sample and excites the quasiparticles.
If the frequency and thus the photon energy is sufficiently high, the quasiparticle can be lifted up to the vacuum
energy. Now a Bogoliubov quasiparticle cannot leave the superconductor since such a electron-hole superposition
does not exist in vacuum. However, it does contain a nonzero electron component, which can leave the solid.

105
This process requires that the photon energy is large enough to lift a quasiparticle from an occupied state, i.e.,
essentially from below the Fermi energy, to the vacuum energy. Photoemission experiments can thus map out
the density of states of occupied states, up to matrix elements describing the coupling of the quasiparticles to the
light and the probability of high-energy electrons leaving the solid.
ARPES in addition records the momentum of the emitted electron. This allows to reconstruct the momentum
of the quasiparticle before photon absorption, which is very similar since photons of relevant energies have
relatively small momenta. The upshot is that the whole dispersion of occupied states can, in principle, be
mapped out. While ARPES is an extremely powerful method, it does not always work or does not work well, for
a number of reasons:
• ARPES is difficult at low temperatures T . 10 K since the sample is heated by the light.
• ARPES requires good surface quality.
• ARPES is highly surface sensitive since the penetration length of light into a superconductor is very short.
However, superconductors can have electronic states bound to the surface that might be difficult to disen-
tangle from bulk contributions. In unconventional superconductors one is often interested specifically in the
surface states, in which case ARPES is very useful if it works.

10.5 Isotope effect

How can one check that superconductivity is indeed governed by a phonon-mediated interaction? BCS theory
predicts  
1
kB Tc , ∆0 ∝ ωD exp − . (10.84)
V0 D(EF )
It would be ideal to compare kB Tc or ∆0 for superconductors that only differ in the Debye frequency ωD , not in
V0 of D(EF ). This is at least approximately possible by using samples containing different isotopes (or different
mixing ratios of isotopes) of the same elements.
The eigenfrequency of a harmonic oscillator scales with the mass like
1
ω∼√ . (10.85)
m
The entire phonon dispersion, and thus in particular the Debye frequency, also scales like
1 1
ωqλ ∼ √ , ωD ∼ √ (10.86)
M M
with the atomic mass M for an elemental superconductor. The same scaling has been found above for the jellium
model, see Eq. (8.73). Consequently, for elemental BCS weak-coupling superconductors,
1
kB Tc , ∆0 ∼ M −α with α = . (10.87)
2
This is indeed found for simple superconductors. The exponent is found to be smaller or even negative for materials
that are not in the weak-coupling regime V D(EF )  1 or that are not phonon-mediated superconductors. In
particular, if the relevant interaction has nothing to do with phonons, we expect α = 0. This is observed for
optimally doped (highest Tc ) cuprate high-temperature superconductors.

10.6 Specific heat

We now discuss further predictions following from BCS theory. We start by revisiting the heat capacity or specific
heat. The BCS Hamiltonian
†
X
HBCS = EBCS + Ek γkσ γkσ (10.88)
kσ

106
with q
2
Ek = ξk2 + |∆k | (10.89)
leads to the internal energy X
U = hHBCS i = EBCS + 2 Ek nF (Ek ). (10.90)
k

However, this is inconvenient for the calculation of the heat capacity C = dU/dT since the condensate energy
EBCS depends on temperature through hc−k,↓ ck↑ i. We better consider the entropy, which has no contribution
from the condensate. It reads
X
S = −kB [(1 − nF ) ln(1 − nF ) + nF ln nF ], (10.91)
kσ

where nF ≡ nF (Ek ). From the entropy, we obtain the heat capacity

dS dS
C=T = −β
dT dβ
X d X dnF
= 2kB β [(1 − nF ) ln(1 − nF ) + nF ln nF ] = 2kB β [− ln(1 − nF ) − 1 + ln nF + 1]
dβ | {z } dβ
k k nF
= ln 1−nF = ln e−βEk = −βEk
X d nF (Ek )
= −2kB β 2 Ek . (10.92)
dβ
k

Note that
q nF (Ek ) depends on β or temperature both explicitly and through the temperature dependence of
2
Ek = ξk2 + |∆k (T )| :
! !
2 2
X ∂nF ∂nF 1 d |∆k | X ∂nF 1 d |∆k |
C = −2kB β 2
Ek + = −2kB β Ek2 + β . (10.93)
∂β ∂Ek 2Ek dβ ∂Ek 2 dβ
k | {z } k
Ek ∂nF
= β ∂Ek

The first term is due to the explicit β dependence, i.e., to the change of occupation of quasiparticle states with
temperature. The second term results from the temperature dependence of the quasiparticle spectrum and is
2
absent for T ≥ Tc , where |∆k | = 0 = const. The sum over k contains the factor

∂nF ∂ 1 eβEk
=β = −β 2 = −β nF (Ek )[1 − nF (Ek )] (10.94)
∂Ek ∂ βEk eβEk + 1 (eβEk + 1)

so that !
2
X 1 d |∆k |
C = 2kB β 2
nF (1 − nF ) Ek2 + β . (10.95)
2 dβ
k

Here, nF (1 − nF ) is exponentially small for Ek  kB T . This means that for kB T  ∆min , where ∆min is the
minimum superconducting gap, all terms in the sum are exponentially suppressed since kB T  ∆min ≤ Ek . Thus
the heat capacity is exponentially small at low temperatures. This result is not specific to superconductors—all
systems with an energy gap for excitations show this behavior.
For the simple interaction used above, the heat capacity can be obtained in terms of an integral over energy.
The numerical evaluation gives the following result:

107
 C

∆C

l state
norma
0 Tc T

We find a downward jump at Tc , reproducing a result obtained from Landau theory in section 6.1. The jump
occurs for any mean-field theory describing a second-order phase transition for a complex order parameter. Since
BCS theory is such a theory, it recovers the result.
The height of the jump can be found as follows: The Ek2 term in Eq. (10.95) is continuous for T → Tc−
(∆0 → 0). Thus the jump height is given by

1 X d∆20
∆C = 2 T3 nF (ξk ) [1 − nF (ξk )] . (10.96)
kB c dβ T →Tc−
k

To obtain ∆0 close to Tc , we have to solve the gap equation

ZωD ZωD
tanh β2 ξ 2 + ∆20 tanh β2 ξ 2 + ∆20
p p
1 = V0 D(EF ) dξ p = V0 D(EF ) dξ p (10.97)
2 ξ 2 + ∆20 ξ 2 + ∆20
−ωD 0

for small ∆0 . Writing

1 1
β= = (10.98)
kB T kB (Tc − ∆T )
and expanding for small ∆T and small ∆0 , we obtain
ZωD ZωD
tanh 2kBξ Tc V0 D(EF ) dξ
1∼
= V0 D(EF ) dξ + ∆T 2 ξ
ξ 2kB Tc2 cosh 2kB Tc
0 0
| {z }
= 1 (gap equation)
ZωD
2kB Tc tanh 2kBξ Tc
!
V0 D(EF ) 2 dξ 1
+ ∆0 − . (10.99)
4kB Tc ξ2 cosh2 ξ
2kB Tc
ξ
0

In the weak-coupling limit, ωD  kB Tc , we can extend the integrals to infinity, which yields
7ζ(3)
V0 D(EF ) V0 D(EF ) 2 π2
0∼ = ∆T 2kB Tc − ∆0
2kB Tc2 4kB Tc 2kB Tc
2
∆T 7ζ(3) ∆0
= V0 D(EF ) − V0 D(EF ) (10.100)
Tc 8π 2 (kB Tc )2
2
8π
⇒ ∆20 ∼
= k 2 Tc ∆T, (10.101)
7ζ(3) B

where ζ(z) is the Riemann zeta function. Thus

d∆20 d∆T d∆20 d∆20 ∼ 8π 2 3 3

= = kB Tc2 = k T (10.102)
dβT →Tc− dβ d∆T ∆T →0 d∆T ∆T →0 7ζ(3) B c
1 X 8π 2 3 3
⇒ ∆C = 2 3 nF (ξk ) [1 − nF (ξk )] k T
kB Tc 7ζ(3) B c
k

108
 Z∞
8π 2
= kB N D(EF ) dξ nF (ξ) [1 − nF (ξ)]
7ζ(3)
−∞
| {z }
2
8π = k B Tc
= 2
N D(EF ) kB Tc , (10.103)
7ζ(3)
where N is the number of unit cells. The specific-heat jump is
∆C 8π 2
∆c = = 2
d(EF ) kB Tc , (10.104)
V 7ζ(3)

where d(EF ) = D(EF ) N/V is the density of states per volume. Note that the ratio ∆c/d(EF )kB
2
Tc is another
universal number within BCS theory.

10.7 Density of states and single-particle tunneling

Detailed experimental information on the excitation spectrum in a superconductor can be obtained from single-
particle tunneling between a superconductor and either a normal metal or another superconductor. We discuss
this in the following assuming, for simplicity, that the density of states D(E) in the normal state is approximately
constant close to the Fermi energy. We restrict ourselves to single-electron tunneling; pair tunneling, which leads
to the Josephson effects, will be discussed later.

Quasiparticle density of states

The density of states per spin of the Bogoliubov quasiparticles created by γ † is easily obtained from their disper-
sion:  
1 X 1 X
q
2
Ds (E) = δ(E − Ek ) = 2
δ E − ξk + |∆k | . (10.105)
N N
k k

The subscript “s” of Ds (E) stands for “superconducting.” For the case of approximately constant normal-state
density of states Dn and constant gap ∆0 , we find
Z∞  q 
Ds (E) = dξ Dn (µ − ξ) δ E − ξ 2 + ∆20
−∞
Z∞  q 
∼
= Dn (EF ) dξ δ E − ξ 2 + ∆20
−∞
Z∞  q 
= Dn (EF ) 2 dξ δ E − ξ 2 + ∆20 . (10.106)
0

For 0 ≤ E < ∆0 , the equation E − ξ 2 + ∆20 = 0 does not have a real solution for ξ and the integral vanishes.
p

On the other hand, in the case E > ∆0 we use a standard result for the δ distribution to write
 
Z∞
p
δ ξ − E 2 − ∆20
. . . = Dn (EF ) 2 dξ  p  note ξ ≥ 0
∂ 2 + ∆2
0 ∂ξ E − ξ 0
 
Z∞ δ ξ − E 2 − ∆2
p
0
= Dn (EF ) 2 dξ
√ 2ξ 2
0 ξ +∆0
E
= Dn (EF ) 2 p (10.107)
E 2 − ∆20

109
so that
Dn (EF ) p 2E

for E > ∆0 ,
∼
Ds (E) = E 2 − ∆20 (10.108)
0 for 0 ≤ E < ∆0 .


Ds (E )
Dn (EF)

0 ∆0 E

There are of course no states in the energy gap. Importantly, we find a divergence at the gap edge at E = ∆0 .
In the normal-state limit, ∆0 → 0, we obtain Ds (E)/Dn (EF ) → 2, deviating from the result of unity given in
Tinkham’s book. The origin is that E is an excitation energy relative to the Fermi sea, i.e., both electron and
hole excitations contribute to the density of states at positive E.

Single-particle tunneling
It is plausible that the density of states Ds (E) can be mapped out by tunneling experiments, for example using a
normal-metal/insulator/superconductor structure. However, we have to keep in mind that the particles tunneling
out of or into a normal metal are real electrons, whereas the quasiparticles in a superconductor are superpositions
of electrons and holes. To study tunneling effects theoretically, we employ a tunneling Hamiltonian of the form

H = HL + HR + HT , (10.109)

where HL and HR describe the materials to the “left” and “right” of the tunneling region. Either can be a normal
metal or a superconductor and is assumed to be unaffected by the presence of the tunneling region. For example,
while translational invariance is necessarily broken in a tunneling device, we nevertheless write HL,R in terms of
lattice-momentum states. HT describes the tunneling between the two materials,

tkq c†kσ dqσ + h.c.

X
HT = (10.110)
kqσ

Here, c and d are electronic operators referring to the two sides of the tunneling barrier and tkq is a tunneling
matrix element, which might depend on the momenta of the incoming and outgoing electron. We assume a
non-magnetic tunneling barrier so that the electron spin is conserved and tkq does not depend on it. We treat
the two bulk materials in the mean-field approximation so that HL,R are effectively non-interacting mean-field
Hamiltonians. We further assume constant normal-state densities of states and momentum-independent tunneling
matrix elements t = tkq .
For tunneling between two normal metals, we can calculate the current for an applied voltage V using the
Landauer formula:
Inn = IL→R − IR→L , (10.111)
| {z } | {z }
left to right right to left

with
Z∞
2e 2
IL→R ∝ n
dξ DL n
(µ + ξ) DR (µ + ξ + eV ) |t| nF (ξ) [1 − nF (ξ + eV )], (10.112)
h
−∞

110
 Z∞
2e 2
IR→L ∝ n
dξ DL n
(µ + ξ) DR (µ + ξ + eV ) |t| nF (ξ + eV ) [1 − nF (ξ)], (10.113)
h
−∞

where n
DL,R is the density of states per spin direction in the left/right normal metal. The factors of nF and 1 − nF
can be understood as the probabilities that the relevant initial states exist and are occupied and the final states
exist and are empty. We get
Z∞
2e n n 2
Inn ∝ dξ DL (µ + ξ) DR (µ + ξ + eV ) |t| [nF (ξ) − nF (ξ + eV )]
h
−∞
Z∞
∼ 2e n n 2
= D (EF ) DR (EF ) |t| dξ [nF (ξ) − nF (ξ + eV )]
h L
−∞
2e2 n 2
= n
DL (EF ) DR (EF ) |t| V. (10.114)
h
We thus find ohmic behaviour,
Inn = Gnn V. (10.115)
We next consider the case that one material is normal while the other (without loss of generality the left) is
superconducting. Now there are additional factors because the quasiparticles in the superconductor are not pure
electrons or holes. Let us say an electron is tunneling out of the superconductor with energy ω > 0. First of all,
this is only possible if ω ≥ ∆0 because of the energy gap. Now the electron can either come from an electron-like
quasiparticle (k > kF ), which contains an electron portion of
! p !
2 1 ξk 1 ω 2 − ∆20
|uk | = 1+ p 2 = 1+ , (10.116)
2 ξk + ∆20 √ 2 2 2 ω
ξk = ω +∆0

or from a hole-like quasiparticle (k 0 < kF ) with an electron portion of

! p !
2 1 ξk0 1 ω 2 − ∆20
|uk0 | = 1+ p 2 = 1− . (10.117)
2 2 √
ξk0 + ∆0 ξ = − ω2 +∆2 2 ω
k0 0

Ek

k’ k
ω

0 kF k
On the other hand, an electron tunneling out with energy ω < 0 is best described as a hole tunneling in with
energy −ω > 0. The relevant factors are
p !
2 1 ω 2 − ∆20
|vk | = 1− (10.118)
2 |ω|

and
p !
2 1 ω 2 − ∆20
|vk0 | = 1+ . (10.119)
2 |ω|

111
Thus the current flowing from left to right is
Z∞
2e s n 2
IL→R ∝ dω DL (ω) DR (µ + ω + eV ) |t| nF (ω) [1 − nF (ω + eV )]
h
0
( p ! p !)
1 ω 2 − ∆20 1 ω 2 − ∆20
× 1+ + 1−
2 ω 2 ω
| {z }
=1
Z0
2e s n 2
+ dω DL (|ω|) DR (µ + ω + eV ) |t| nF (ω) [1 − nF (ω + eV )]
h
−∞
( p ! p !)
1 ω 2 − ∆20 1 ω 2 − ∆20
× 1− + 1+
2 |ω| 2 |ω|
| {z }
=1
Z∞
2e 2
= s
dω DL n
(|ω|) DR (µ + ω + eV ) |t| nF (ω) [1 − nF (ω + eV )], (10.120)
h
−∞

where DL s
(ω) now is the superconducting density of states, neither containing a spin factor of 2 nor the factor
of 2 due describing both electrons and holes as excitations with positive energy—positive and negative energies
are here treated explicitly and separately. We see that the electron-hole mixing does not lead to any additional
factors beyond the changed density of states. With the analogous expression
Z∞
2e 2
IR→L ∝ s
dω DL n
(|ω|) DR (µ + ω + eV ) |t| nF (ω + eV ) [1 − nF (ω)] (10.121)
h
−∞

we obtain
Z∞
2e s n 2
Isn ∝ dω DL (|ω|) DR (µ + ω + eV ) |t| [nF (ω) − nF (ω + eV )]
h
−∞
Z∞ s
∼ 2e n 2 DL (|ω|)
= n
D (EF ) DR (EF ) |t| dω [nF (ω) − nF (ω + eV )], (10.122)
h L n
DL (EF )
−∞

where 
s
DL (|ω|)  p |ω|

for |ω| > ∆0 ,
n (E ) = 
ω 2 − ∆20 (10.123)
DL F 0 for |ω| < ∆0 .
Thus
Z∞ s
Gnn DL (|ω|)
Isn = dω n [nF (ω) − nF (ω + eV )]. (10.124)
e DL (EF )
−∞

It is useful to consider the differential conductance

Z∞
Ds (|ω|)
 
dIsn ∂ nF (ω + eV )
Gsn := = Gnn dω nL −
dV DL (EF ) ∂ω
−∞
Z∞ s
DL (|ω|)
= Gnn dω n β nF (ω + eV )[1 − nF (ω + eV )]. (10.125)
DL (EF )
−∞

112
In the limit kB T → 0 this becomes
Z∞ s
DL (|ω|) Ds (|eV |)
Gsn = Gnn dω n δ(ω + eV ) = Gnn Ln . (10.126)
DL (EF ) DL (EF )
−∞

Thus low-temperature tunneling directly measures the superconducting density of states. At non-zero tempera-
tures, the features are smeared out over an energy scale of kB T . In the band picture we can see that the Fermi

1
0
energy in the normal material is used to scan the density of states in the superconductor.

0
1
0
1
0
1
µ 1111
0000
0
1
1111
0000
0
1 1111
0000
0
1
2∆ 0

0000 1111
1111 0000
0
1
µ + eV

0
1
0
1S I N
For a superconductor-insulator-superconductor contact, we only present the result for the current without deriva-
tion. The result is very plausible in view of the previous cases:
Z∞ s s
Gnn DL (|ω|) DR (|ω + eV |)
Iss = dω n n (E ) [nF (ω) − nF (ω + eV )]. (10.127)
e DL (EF ) DR F
−∞

1
0
0
1
1111
0000
0
1
1111
0000 0
1
1111
0000
0
1
0
1
2∆ L

0
10
1
µ 2∆ R

0000 0000
1111 1111
0
10
1
0000
1111
µ + eV

0
1
0
1S I S
Now we find a large change in the current when the voltage V is chosen such that the lower gap edge at one
side is aligned with the upper gap edge at the other side. This is the case for |eV | = ∆L + ∆R . This feature
will remain sharp at non-zero temperatures since the densities of states retain their divergences at the gap edges
as long as superconductivity is not destroyed. Effectively, we are using the density-of-states singularity of one
superconductor to scan the density of states of the other. A numerical evaluation of Iss gives the following typical
behavior (here for ∆L = 2kB T , ∆R = 3kB T ):
3.5

3.0

2.5

2.0
Iss

1.5

1.0

|∆L − ∆ R|
0.5 ∆L + ∆R
0.0
0 1 2 3 4

eV

113
 The roughly ohmic transport at small voltages is due to thermal activation over the gaps; note that kB T
is not small compared to the gaps in this example. The peak at eV = |∆L − ∆R | is due to the coincidence
of the singularities in the densities of states at the upper gap edge on both sides. The nearly constant current
for |∆L − ∆R | < eV < ∆L + ∆R is, at first glance, surprising. More detailed analysis shows that both the
contributions from tunneling below the gap and from tunneling above the gap approach constants as eV increases.
What happens here is that the density-of-states peak at the upper gap edge on the left scans the states above
the gap on the right, where both the density of states (far above the gap) and the occupation (essentially zero
far above the chemical potential) are constant. Analogously, the density-of-states peak at the lower gap edge on
the right scans the occupied states below the gap on the left. At eV = ∆L + ∆R , these two contributions are
featureless. But now a new transport channel from below the gap on the left to above the gap on the right opens.
Since the product of the densities of states diverges at the gap edges, the current increases with a jump. For
higher applied voltage, additional states become available and the current increases continuously. In the limit
eV  ∆L , ∆R , the superconducting gaps should be irrelevant and we expect and observe the ohmic behavior of
a normal-normal contact.

10.8 Ultrasonic attenuation and nuclear relaxation

To conclude the brief survey of experimental consequences of BCS theory, we discuss the effects of time-dependent
perturbations. They will be exemplified by ultrasonic attenuation and nuclear relaxation, which represent two
distinct ways in which quasiparticle interference comes into play. Quite generally, we write the perturbation part
of the Hamiltonian as
Bk0 σ0 kσ c†k0 σ0 ckσ ,
XX
H1 = (10.128)
kk0 σσ 0

where Bk0 σ0 kσ are matrix elements of the perturbation between single-electron states of the non-interacting system.
In the superconducting state, we have to express c, c† in terms of γ, γ † ,
†
ck↑ = uk γk↑ + vk γ−k,↓ , (10.129)
†
ck↓ = uk γk↓ − vk γ−k,↑ , (10.130)

where we have assumed u−k = uk , v−k = vk . Thus, with σ, σ 0 = ±1,

  
Bk0 σ0 kσ u∗k0 γk† 0 σ0 + σ 0 vk∗ 0 γ−k0 ,−σ0 uk γkσ + σvk γ−k,−σ
†
XX
H1 =
kk0 σσ 0

Bk0 σ0 kσ u∗k0 uk γk† 0 σ0 γkσ + σu∗k0 vk γk† 0 σ0 γ−k,−σ
†
XX
=
kk0 σσ 0

†
+ σ 0 vk∗ 0 uk γ−k0 ,−σ0 γkσ + σσ 0 vk∗ 0 vk γ−k0 ,−σ0 γ−k,−σ . (10.131)

It is useful to combine the terms containing Bk0 σ0 kσ and B−k,−σ,−k0 ,−σ0 since both refer to processes that change
momentum by k0 − k and spin by σ 0 − σ. If the perturbation couples to the electron concentration, which is the
case for ultrasound, one finds simply
B−k,−σ,−k0 ,−σ0 = Bk0 σ0 kσ . (10.132)
This is often called case I. Furthermore, spin is conserved by the coupling to ultrasound, thus

Bk0 σ0 kσ = δσσ0 Bk0 σkσ . (10.133)

Adding the two terms, we obtain

1 XX 
Hultra = Bk0 σkσ u∗k0 uk γk† 0 σ γkσ + σu∗k0 vk γk† 0 σ γ−k,−σ
†
2 0 σ
kk
† †
+ σvk∗ 0 uk γ−k0 ,−σ γkσ + vk∗ 0 vk γ−k0 ,−σ γ−k,−σ + u∗k uk0 γ−k,−σ γ−k0 ,−σ

114
 
†
− σu∗k vk0 γ−k,−σ γk† 0 σ − σvk∗ uk0 γkσ γ−k0 ,−σ + vk∗ vk0 γkσ γk† 0 σ . (10.134)

Assuming uk , vk , ∆k ∈ R for simplicity, we get, up to a constant,

1 XX h  
Hultra = Bk0 σkσ (uk0 uk − vk0 vk ) γk† 0 σ γkσ + γ−k,−σ
†
γ−k0 ,−σ
2 0 σ
kk
 i
+ σ (uk0 vk + vk0 uk ) γk† 0 σ γ−k,−σ
†
+ γ−k0 ,−σ γkσ . (10.135)

We thus find effective matrix elements Bk0 σkσ (uk0 uk − vk0 vk ) for quasiparticle scattering and Bk0 σkσ σ(uk0 vk +
vk0 uk ) for creation and annihilation of two quasiparticles. Transition rates calculated from Fermi’s golden rule
contain the absolute values squared of matrix elements. Thus the following two coherence factors will be impor-
tant. The first one is
( ! !
2 1 ξk0 ξk
(uk0 uk − vk0 vk ) = 1+ p 2 1+ p 2
4 ξk0 + ∆2k0 ξk + ∆2k
! !)
ξk0 ξk ξk0 ξk
− 2p 2 p + 1− p 2 1− p 2
ξk0 + ∆2k0 ξk2 + ∆2k ξk0 + ∆2k0 ξk + ∆2k
 
1 ξk ξk0 ∆ k ∆k 0
= 1+ − (10.136)
2 Ek Ek0 Ek Ek0

with Ek = ξk2 + ∆2k . If the coefficient B and the normal-state density of states are even functions of the
p

normal-state energy
P relative to the Fermi energy, ξk , the second term, which is odd in ξk and ξk0 , will drop out
under the sum kk0 . Hence, this term is usually omitted, giving the coherence factor
 
1 ∆k ∆k0
F− (k, k0 ) := 1− (10.137)
2 Ek Ek0
relevant for quasiparticle scattering in ultrasound experiments. Analogously, we obtain the second coherence
factor  
∆k ∆k0
σ 2
(u
|{z} k k0 v + v 0
k ku )2
= 1 + =: F+ (k, k0 ) (10.138)
Ek Ek0
=1
for quasiparticle creation and annihilation.
We can gain insight into the temperature dependence of ultrasound attenuation by making the rather crude
approximation that the matrix element B is independent of k, k0 and thus of energy. Furthermore, typical
ultrasound frequencies Ω satisfy ~Ω  ∆0 and ~Ω  kB T . Then only scattering of quasiparticles by phonons but
not their creation is important since the phonon energy is not sufficient for quasiparticle creation. The annihilation
of thermally excited quasiparticles is also irrelevant since it only contributes to the equilibrium background of
phonons and directly only affects high-energy phonons in any case.
The rate of ultrasound absorption (attenuation) can be written in a plausible form analogous to the current
in the previous section:
Z∞
2
αs ∝ dω Ds (|ω|) Ds (|ω + Ω|) |B| F− (ω, ω + Ω) [nF (ω) − nF (ω + Ω)] , (10.139)
−∞

where now
∆20
 
1
0
F− (ω, ω ) = 1− . (10.140)
2 |ω| |ω 0 |
With the approximations introduced above we get
Z∞
∆20
 
2 Ds (|ω|) Ds (|ω + Ω|) 1
αs ∝ Dn2 (EF ) |B| dω 1− [nF (ω) − nF (ω + Ω)] . (10.141)
Dn (EF ) Dn (EF ) 2 |ω| |ω + Ω|
−∞

115
The normal-state attentuation rate is found by letting ∆0 → 0:
Z∞
2 nF (ω) − nF (ω + Ω) 1 2
αn ∝ Dn2 (EF ) |B| dω = Dn2 (EF ) |B| Ω (10.142)
2 2
−∞
Z∞
αs 1 Ds (|ω|) Ds (|ω + Ω|) |ω| |ω + Ω| − ∆20
⇒ = dω [nF (ω) − nF (ω + Ω)] . (10.143)
αn Ω Dn (EF ) Dn (EF ) |ω| |ω + Ω|
−∞

Since Ω is small, we can expand the integral up to linear order in Ω, noting that the term containing the Fermi
functions is of at least first order,
Z∞ 2 2
ω − ∆20

αs ∼ 1 Ds (|ω|) ∂nF
= dω 2
(−Ω)
αn Ω Dn (EF ) ω ∂ω
−∞
 −∆
Z 0 Z∞
 " #2
ω ω 2 − ∆20 ∂nF
= − +  dω p
ω 2 − ∆20 ω2 ∂ω
−∞ ∆0
 −∆
Z 0 Z∞

∂nF
= − +  dω
∂ω
−∞ ∆0

= −nF (∞) + nF (∆0 ) − nF (−∆0 ) + nF (−∞)

| {z } | {z }
=0 =1
= nF (∆0 ) + 1 − nF (−∆0 )
2
= 2 nF (∆0 ) = β∆0 . (10.144)
e +1
Inserting the BCS prediction for ∆0 (T ), we can plot αs /αn vs. temperature:

11
00
αs
αn
1
1111111
0000000

0 Tc T

The suppression of the attenuation below Tc is clearly due to the opening of the superconducting gap—at low
temperatures, there are exponentially few thermally excited quasiparticles that can scatter phonons. However,
the simplicity of the result is deceptive. The density of states at the gap edge, where most of the quasiparticles
reside, is singular so that one might expect a strong enhancement of αs /αn . The enhancement is canceled by the
coherence factor F− , which is coming from the matrix elements of the perturbation.
We now turn to the relaxation of nuclear spins due to their coupling to the electrons. We note without
derivation that the hyperfine interaction relevant for this process is odd in momentum if the electron spin is not
changed but is even if the electron spin is flipped, i.e.,
B−k,−σ,−k0 ,−σ0 = −σσ 0 Bk0 σ0 kσ (10.145)
(recall σσ 0 = ±1). This is called case II. The perturbation Hamiltonian now reads
1 XX 
HNMR = Bk0 σ0 kσ u∗k0 uk γk† 0 σ0 γkσ + σu∗k0 vk γk† 0 σ0 γ−k,−σ
†
2 0 0
kk σσ

116
 † †
+ σ 0 vk∗ 0 uk γ−k0 ,−σ0 γkσ + σσ 0 vk∗ 0 vk γ−k0 ,−σ0 γ−k,−σ − σσ 0 u∗k uk0 γ−k,−σ γ−k0 ,−σ0

†
+ σu∗k vk0 γ−k,−σ γk† 0 σ0 + σ 0 vk∗ uk0 γkσ γ−k0 ,−σ0 − vk∗ vk0 γkσ γk† 0 σ0 . (10.146)

Assuming uk , vk , ∆k ∈ R, we get, up to a constant,

1 XX h  
HNMR = Bk0 σ0 kσ (uk0 uk + vk0 vk ) γk† 0 σ0 γkσ − σσ 0 γ−k,−σ
†
γ−k0 ,−σ0
2 0 0
kk σσ
 i
+ σ (uk0 vk − vk0 uk ) γk† 0 σ0 γ−k,−σ
†
− σσ 0 γ−k0 ,−σ0 γkσ . (10.147)

Compared to ultrasonic attenuation (case I) there is thus a change of sign in both coherence factors. An analogous
derivation now gives the interchanged coherence factors
 
1 ∆k ∆k0
0
F+ (k, k ) = 1+ (10.148)
2 Ek Ek0

for quasiparticle scattering and  

1 ∆k ∆k0
0
F− (k, k ) = 1− (10.149)
2 Ek Ek0
for quasiparticle creation and annihilation. The relevant energy ~Ω is the Zeeman energy of a nuclear spin in the
applied uniform magnetic field and is small compared to the gap ∆0 . Thus we can again restrict ourselves to
the small-Ω limit. The derivation is initially analogous to case I, but with F− replaced by F+ . The nuclear-spin
relaxation rate is
Z∞
2
αs ∝ dω Ds (|ω|) Ds (|ω + Ω|) |B| F+ (ω, ω + Ω) [nF (ω) − nF (ω + Ω)] (10.150)
−∞

with
∆20
 
1
F+ (ω, ω ) = 0
1+ . (10.151)
2 |ω| |ω 0 |
Thus
Z∞
∆20
 
2 Ds (|ω|) Ds (|ω + Ω|) 1
αs ∝ Dn2 (EF ) |B| dω 1+ [nF (ω) − nF (ω + Ω)] (10.152)
Dn (EF ) Dn (EF ) 2 |ω| |ω + Ω|
−∞
Z∞
αs 1 Ds (|ω|) Ds (|ω + Ω|) |ω| |ω + Ω| + ∆20
⇒ = dω [nF (ω) − nF (ω + Ω)] . (10.153)
αn Ω Dn (EF ) Dn (EF ) |ω| |ω + Ω|
−∞

Note that the only difference compared to the corresponding Eq. (10.143) for ultrasound attenuation is the sign
of ∆20 in the numerator. If we now expand the integral for small Ω as above, we encounter a problem:
Z∞ " #2 Z∞
αs ∼ 1 ω ω 2 + ∆20 ∂nF ω 2 + ∆20 ∂nF
= 2 dω (−Ω) = −2 dω . (10.154)
2 ω 2 − ∆20 ∂ω
p
αn Ω ω 2 − ∆20 ω ∂ω
∆0 ∆0

This integral diverges logarithmically at the lower limit. Keeping a non-zero but realistically small Ω removes the
divergence. However, the calculated αs /αn is still too large compared to experiments. The origin of this problem
is the strong singularity in the superconducting density of states. A k-dependent gap ∆k removes the problem;
in a realistic theory ∆k always has a k dependence since it cannot have higher symmetry than the underlying
normal dispersion ξk . Introducing some broadening of the density of states by hand, we numerically find the
following temperature dependence:

117
 1
0
αs
αn
case II (NMR)

1
111111
000000
case I

0 Tc T

There is a large maximum below the transition temperature, called the Hebel-Slichter peak. It results from the
factor Ds2 (|ω|) (for Ω → 0) in the integrand,

ω2
Ds2 (|ω|) ∼
= Dn2 (EF ) , (10.155)
ω 2 − ∆20

which for nuclear relaxation is not canceled by the coherence factor F+ , whereas for ultrasonic attenuation it is
canceled by F− . Physically, the strong enhancement below Tc of the density of states of both initial and final
states at ω & ∆0 leads to increased nuclear relaxation.

10.9 Ginzburg-Landau-Gor’kov theory

We conclude this chapter by remarking that Lev Gor’kov managed, two years after the publication of BCS theory,
to derive Ginzburg-Landau theory from BCS theory. The correspondence is perfect if the gap is sufficiantly small,
i.e., T is close to Tc , and the electromagnetic field varies slowly on the scale of the Pippard coherence length ξ0
(see Sec. 5.4). These are indeed the conditions under which Ginzburg and Landau expected their theory to be
valid.
Gor’kov used equations of motion for electronic Green functions, which he decoupled with a mean-field-like
approximation that allowed for spatial variations of the mean-field decoupling term ∆(r). The derivation is given
in Schrieffer’s book and we omit it here. A more modern exposition based on the function-integral method is
given by Altland and Simons in Condensed Matter Field Theory. Gor’kov found that in order to obtain the
Ginzburg-Landau equations, he had to take

q = −2e, (10.156)
m∗ = 2m, (10.157)

as anticipated, and (using our conventions)

p
7ζ(3) p 0 ∆(r)
ψ(r) = ns , (10.158)
4π kB Tc
where
ns
n0s := . (10.159)
1 − TTc T →Tc−

Recall that ns ∼ 1 − T /Tc close to Tc in Ginzburg-Landau theory and thus n0s ≈ const. The spatially dependent
gap is thus locally proportional to the Ginzburg-Landau “condensate wavefunction” or order parameter ψ(r).
Since we have already found that London theory is a limiting case of Ginzburg-Landau theory, it is also a
limiting case of BCS theory. But London theory predicts the two central properties of superconductors: Ideal
conduction and flux expulsion. Thus Gor’kov’s derivation also shows that BCS theory indeed describes a super-
conducting state. (Historically, this has been shown by BCS before Gor’kov established the formal relationship
between the various theories.)

118
11

Josephson effects

Brian Josephson made two important predictions for the current flowing through a tunneling barrier between two
superconductors. The results have later been extended to various other systems involving two superconducting
electrodes, such as superconductor/normal-metal/superconductor heterostructures and superconducting weak
links. We will first formulate his predictions and then derive them for the case of a weak link.
Rather generally, for vanishing applied voltage, a supercurrent Is is flowing which is related to the phase
difference ∆φ of the two condensates by
 Z 
2π
Is = Ic sin ∆φ − ds · A . (11.1)
Φ0
We will discuss the critical current Ic presently. Note that the form of the argument of the sine function is
prescribed by gauge invariance. We consider the case without magnetic field so that we can choose the gauge
A ≡ 0. Then the Josephson relation simplifies to
Is = Ic sin ∆φ. (11.2)
It describes the DC Josephson effect. The current continues to flow as long as the phase difference ∆φ is
maintained, although no bias voltage is applied.
Secondly, Josephson predicted that in the presence of a bias voltage V , the phase difference would evolve
according to
d 2e
∆φ = − V (11.3)
dt ~
(recall that we use the convention e > 0) so that for a constant voltage an alternating current would flow,
2e
∆φ(t) = ∆φ0 − V t, (11.4)
~ 
2e
⇒ Is (t) = Ic sin ∆φ0 − Vt . (11.5)
~
This is called the AC Josephson effect. The frequency
2eV
ωJ := (11.6)
~
of the current is called the Josephson frequency. The AC Josephson effect relates frequencies (or times) to voltages
through the ratio 2e/~ of fundamental constants, which makes it important for metrology.

11.1 The Josephson effects in Ginzburg-Landau theory

We consider a weak link between two identical bulk superconductors. The weak link is realized by a short wire
of length L  ξ and cross section A made from the same material as the bulk superconductors. We choose this

119
setup since it is the easiest to treat in Ginzburg-Landau theory since the parameters α and β are uniform, but
the only property that really matters is that the phase φ of the order parameter ψ(r) only changes within the
weak link.
1111111
0000000
0000000
1111111
0000000
1111111
0000000
1111111
0000000
1111111
0000000
1111111
0000000
1111111
0000000
1111111
0 L x
We employ the first Ginzburg-Landau equation for A ≡ 0 assuming ψ(r) to depend only on the coordinate x
along the wire,
ξ 2 f 00 (x) + f (x) − |f (x)|2 f (x) = 0 (11.7)
with r
ψ(x) β
f (x) = = − ψ(x). (11.8)
|ψ(±∞)| α
We assume the two bulk superconductors to be uniform and to have a relative phase of ∆φ. This allows us to
write (
1 for x ≤ 0,
f (x) = (11.9)
ei∆φ for x ≥ L.
For the wire we have to solve Eq. (11.7) with the boundary conditions

f (0) = 1, f (L) = ei∆φ . (11.10)

Since L  ξ, the first term in Eq. (11.7) is larger than the other two by a factor of order ξ 2 /L2 , unless ∆φ = 0,
in which case the solution is trivially f ≡ 1. It is thus sufficient to solve f 00 (x) = 0, which has the solution1
L−x x
f (x) = + ei∆φ , (11.11)
L L
which linearly interpolates between f (0) and f (L). Inserting f (x) into the second Ginzburg-Landau equation
(with A ≡ 0), we obtain
 
q~ e~ α
js = i ∗ [(ψ 0 )∗ ψ − ψ ∗ ψ 0 ] = −i − [(f 0 )∗ f − f ∗ f 0 ]
2m 2m β
     
e~ 1 1 L−x x L−x x 1 1
= −i 2ns − + e−i∆φ + ei∆φ − + e−i∆φ − + ei∆φ
2m L L L L L L L L
 
e~ns x L − x i∆φ
− 2 ei∆φ − e−i∆φ − − e−i∆φ



= −i e
m L L2
 
e~ns x L−x
=2 − 2 sin ∆φ − sin ∆φ
m L L2
e~ns
= −2 sin ∆φ. (11.12)
mL
The current is obviously obtained by integrating over the cross-sectional area,
2e~ns A
Is = − sin ∆φ ≡ Ic sin ∆φ (11.13)
m L
1 Note that f 00 (x) = 0 implies a linear function f (x), which is distinct from a linear dependence of the phase, ei∆φ x/L .

120
so that we get
2e~ns A
Ic = − . (11.14)
m L
The negative sign is due to the negative charge −2e of the Cooper pairs. The amplitude of the current-phase
relation is clearly |Ic |. Note that the second Ginzburg-Landau equation is essentially the relation for the current
density associated with a wave function in quantum mechanics. Thus the DC Josephson effect only relies on the
presence of a macroscopic wave function of electrically charged entities, namely the Cooper pairs, or, in other
words, on the presence of a charge condensate.
Ginzburg-Landau theory also gives us the free energy of the wire. Since we have neglected the α and β terms
when solving the Ginzburg-Landau equation, we must for consistency do the same here,
ZL ZL 2 ZL
~2 0 ~2 ~2 2ns
 
2 α 1 1
F =A dx |ψ (x)| = A dx − − + ei∆φ =A dx 2 (1 − cos ∆φ)
4m 4m β L L 4m L2
0 0 0
A ~2 ns
= (1 − cos ∆φ) . (11.15)
L m
F

−π 0 π ∆φ

The free energy is minimal when the phases of the two superconductors coincide. Thus if there existed any
mechanism by which the phases could relax, they would approach a state with uniform phase along the wire, a
highly plausible result.
We can now also derive the AC Josephson effect. Assuming that the free energy of the junction is only changed
by the supercurrent, we have
d
F = Is V, (11.16)
dt
i.e., the electrical power. This relation implies that
∂F d
∆φ = Is V (11.17)
∂∆φ dt
A ~2 n s d 2e~ns A
⇒ sin ∆φ ∆φ = − sin ∆φ V (11.18)
L m dt m L
d 2e
⇒ ∆φ = − V, (11.19)
dt ~
as stated above. Physically, if a supercurrent is flowing in the presence of a bias voltage, it generates power. Since
energy is conserved, this power must equal the change of (free) energy per unit time of the junction.

11.2 Dynamics of Josephson junctions

For a discussion of the dynamical current-voltage characteristics of a Josephson junction, it is crucial to realize
that a real junction also
1. permits single-particle tunneling (see Sec. 10.7), which we model by an ohmic resistance R in parallel to the
junction,
2. has a non-zero capacitance C.
This leads to the resistively and capacitively shunted junction (RCSJ) model represented by the following circuit
diagram:

121
 C junction
R

The current through the device is the sum of currents through the three branches,
V dV
I= +C − Ic sin ∆φ, (11.20)
R dt
where we take Ic > 0 and have made the sign explicit. With
d 2e
∆φ = − V, (11.21)
dt ~
we obtain
~ d ~C d2
I=− ∆φ − ∆φ − Ic sin ∆φ. (11.22)
2eR dt 2e dt2
We introduce the plasma frequency r
2eIc
ωp := (11.23)
~C
and the quality factor
Q := ωp RC (11.24)
of the junction. This leads to
I 1 d2 1 d
= − 2 2 ∆φ − ∆φ − sin ∆φ (11.25)
Ic ωp dt Qωp dt
and with τ := ωp t finally to
d2 1 d I
2
∆φ + ∆φ + sin ∆φ = − . (11.26)
dτ Q dτ Ic
Compare this equation to the Newton equation for a particle moving in one dimension in a potential Vpot (x) with
friction (Stokes drag),

dVpot
mẍ + αẋ = − (11.27)
dx
α 1 dVpot
⇒ ẍ + ẋ = − . (11.28)
m m dx
This Newton equation has the same form as the equation of motion of ∆φ if we identify

t ↔ τ, (11.29)
x ↔ ∆φ, (11.30)
α 1
↔ , (11.31)
m Q
1 I
Vpot (x) ↔ ∆φ − cos ∆φ. (11.32)
m Ic
Thus the time dependence of ∆φ(τ ) corresponds to the damped motion of a particle in a “tilted-washboard”
potential

122
 I ∆φ
Ic
Vpot m

0 ∆φ

Equation (11.26) can be used to study a Josephson junction in various regimes. First, note that a stationary
solution exists as long as the effective potential Vpot has local minima or at least points with vanishing derivative,
which is the case for |I| ≤ Ic . Then the stationary solution satisfies

I
sin ∆φ = const = − and V ≡ 0. (11.33)
Ic
This solution does not exist for |I| > Ic . What happens if we impose a time-independent current that is larger
than the critical current? We first consider a strongly damped junction, Q  1. Then we can neglect the
acceleration (inertial) term and write

1 d I
∆φ + sin ∆φ = − (11.34)
Q dτ Ic
1 d I
⇒ ∆φ = − − sin ∆φ (11.35)
Q dτ Ic
d ∆φ
⇒ − I = Q dτ (11.36)
Ic + sin ∆φ
∆φ
Z∆φ 0
d ∆φ0 I > Ic 2 1+ I
Ic tan ∆φ
2
⇒ Q (τ − τ0 ) = − I
= −q
arctan q . (11.37)
Ic + sin ∆φ0 I 2
−1 I 2


−1
0 Ic Ic 0

We are interested in periodic solutions for ei∆φ or ∆φ mod 2π since we expect the motion to approach a periodic
function of time after a transient. One period T is the time it takes for ∆φ to change from 0 to −2π (note that
d∆φ/dτ < 0). For one period, we obtain
−2π
d ∆φ0
Z
I > Ic 2π
Qωp T = − I
= q (11.38)
Ic + sin ∆φ0 I 2


−1
0 Ic
2π 1 ~ 1 π~ 1
⇒ T = q = 2π q = p . (11.39)
Qωp I 2


−1 2eIc R I 2


−1 eR I − Ic2
2
Ic Ic

The voltage V ∝ −d∆φ/dt is of course time-dependent but the time-averaged voltage is simply

ZT ZT
1 ~ 1 d ~ 1 π~ 1
(11.40)
p
V̄ = dt V (t) = − dt ∆φ = − [∆φ(T ) − ∆φ(0)] = = R I 2 − Ic2
T 2e T dt 2e T | {z } e T
0 0 = −2π

for I > Ic . By symmetry, V̄ = −R I 2 − Ic2 for I < −Ic . The current-voltage characteristics for imposed direct
p

current (DC) I thus look like this:

123
 I

Ic
V/R

0 V
− Ic

For |I| ≤ Ic , the current flows without resistance. At Ic , non-zero DC and AC voltages set in gradually. For
|I|  Ic , the DC voltage approaches the ohmic result for a normal contact.
The solution for general Q requires numerical calculation but we can analyze the opposite case of weak
damping, Q  1. The stationary solution ∆φ = const, V = 0 still exists for I ≤ Ic . The mechanical analogy
suggests that the time-dependent solution with periodic ei∆φ will be a very rapid slide down the washboard,
overlaid by a small-amplitude oscillation,
∆φ ∼= −ωt + δφ, (11.41)
where ω  ωp and δφ is periodic in time and small. Inserting this ansatz into the equation of motion we find

I
ω=Q ωp  ωp , (11.42)
Ic
ωp2
δφ ∼
= − 2 sin ωt. (11.43)
ω
|{z}
1

Thus
ω ωp2 ω
∆φ ∼
=− τ − 2 sin τ. (11.44)
ωp ω ωp
We convince ourselves that this is a good solution for Q  1: Inserting it into Eq. (11.26), we obtain for the
left-hand side
!
ω 1 ω 1 ωp ω ω ωp2 ω
sin τ− − cos τ − sin τ + 2 sin τ
ωp Q ωp Q ω ωp ωp ω ωp
   2  
∼ ω I 1 Ic ω ω ω 1 Ic ω 1
= sin τ − − 2 cos τ − sin τ − cos τ 2
cos τ +O . (11.45)
ω Ic Q I ωp ω ωp Q I ωp Q4
 p  p


To leading order in 1/Q this is just −I/Ic , which agrees with the right-hand side. We thus find an averaged
voltage of
ZT  
1 ~ d ~ω ~ I ~ 2eIc I
V̄ = dt − ∆φ = = Q ωp = RC = RI, (11.46)
T 2e dt 2e 2e Ic 2e ~C Ic
0

i.e., the ohmic behavior of the normal junction. Note that the time-dependent solution exists for all currents, not
just for |I| > Ic . Thus for |I| ≤ Ic there are now two solutions, with V ≡ 0 and with V̄ = RI. If we change the
imposed current we expect hysteretic behavior. This is indeed observed.

124
 I

V/R
Ic

0 V
− Ic

If we instead impose a constant voltage we obtain, for any Q,

d 2e d2
∆φ = − V = const ⇒ ∆φ = 0 (11.47)
dt ~ dt2
and thus
 
1 2e 2e I(t)
− V + sin − V t + ∆φ0 = − (11.48)
Qωp ~ ~ Ic
 
Ic 2e 2e
⇒ I(t) = V + Ic sin V t − ∆φ0 . (11.49)
Qωp ~ ~
The averaged current is just
Ic 2e ~C Ic 2e V
I¯ = V = V = , (11.50)
Qωp ~ 2eIc RC ~ R
i.e., it is ohmic. We emphasize that this result holds for any damping. It is evidently important to carefully
specify whether a constant current or a constant voltage is imposed.

11.3 The Bogoliubov-de Gennes Hamiltonian

It is often necessary to describe inhomogeneous systems, Josephson junctions are typical examples. So far, the only
theory we know that is able to treat inhomogeneity is the Ginzburg-Landau theory, which has the disadvantage
that the quasiparticles are not explicitly included. It is in this sense not a microscopic theory. We will now discuss
a microscopic description that allows us to treat inhomogeneous systems. The essential idea is to make the BCS
mean-field Hamiltonian spatially dependent. This leads to the Bogoliubov-de Gennes Hamiltonian. It is useful to
revert to a first-quantized description. To this end, we introduce the condensate state |ψBCS i as the ground state
of the BCS Hamiltonian

ξk c†kσ ckσ − ∆k c†k↑ c†−k,↓ + const.

X X X
HBCS = ∆∗k c−k,↓ ck↑ − (11.51)
kσ k k

|ψBCS i agrees with the BCS ground state defined in Sec. 9.2 in the limit T → 0 (recall that ∆k and thus HBCS
is temperature-dependent). We have
HBCS |ψBCS i = EBCS |ψBCS i , (11.52)
where EBCS is the temperature-dependent energy of the condensate.
Since HBCS is bilinear, it is sufficient to consider single-particle excitations. Many-particle excitations are
simply product states, or more precisely Slater determinants, of single-particle excitations. We first define a
two-component spinor, also called Nambu spinor,
   †   † 
|Ψk1 i ck↑ |ψBCS i ck↑
|Ψk i ≡ := = |ψBCS i (11.53)
|Ψk2 i c−k,↓ |ψBCS i c−k,↓

125
describing the single-particle excitations. The |Ψk,1 i and |Ψk,2 i are not eigenstates of the BCS Hamiltonian,
though. It is easy to show that

[HBCS , c†k↑ ] = ξk c†k↑ − ∆∗k c−k,↓ , (11.54)

[HBCS , c−k,↓ ] = −ξk c−k,↓ − ∆k c†k↑ . (11.55)

With these relations we obtain

 
HBCS |Ψk1 i = HBCS c†k↑ |ψBCS i = ξk c†k↑ − ∆∗k c−k,↓ + c†k↑ HBCS |ψBCS i
= (EBCS + ξk ) |Ψk1 i − ∆∗k |Ψk2 i (11.56)

and
 
HBCS |Ψk2 i = HBCS c−k,↓ |ψBCS i = −ξk c−k,↓ − ∆k c†k↑ + c−k,↓ HBCS |ψBCS i
= (EBCS − ξk ) |Ψk2 i − ∆k |Ψk1 i . (11.57)

Thus for the basis {|Ψk1 i , |Ψk2 i}, the Hamiltonian has the matrix form
 
EBCS + ξk −∆k
. (11.58)
−∆∗k EBCS − ξk

This is the desired Hamiltonian in first-quantized form, except that we want to measure excitation energies relative
to the condensate energy. Thus we write as the first-quantized Hamiltonian in k space
 
ξk −∆k
HBdG (k) = . (11.59)
−∆∗k −ξk

This is the Bogoliubov-de Gennes Hamiltonian for non-magnetic superconductors. The upper left (lower right)
component describes spin-up electrons (spin-down holes). The electron-hole transformation performed for the
spin-down particles allows one to implement superpositions of spin-up electrons and spin-down holes by nonzero
off-diagonal components −∆k , −∆∗k .
The eigenvalues of the Bogoliubov-de Gennes Hamiltonian are
q
2
± ξk2 + |∆k | = ±Ek (11.60)

with corresponding eigenstates

 
uk |Ψk1 i − vk∗ |Ψk2 i = uk c†k↑ − vk∗ c−k,↓ |ψBCS i = γk↑
†
|ψBCS i (11.61)

and
 
vk |Ψk1 i + u∗k |Ψk2 i = vk c†k↑ + u∗k c−k,↓ |ψBCS i = γ−k,↓ |ψBCS i (11.62)

with uk , vk defined as above. (A lengthy but straightforward calculation has been omitted.) We can now un-
derstand why the second eigenvalue comes out negative: The corresponding eigenstate contains a quasiparticle
annihilation operator, not a creation operator. The excitation energies are the positive energies that we already
know.
The next step is to Fourier-transform the Hamiltonian to obtain its real-space representation, which we write
as  
1 X ik·r H0 (r) −∆(r)
HBdG (r) := e HBdG (k) = , (11.63)
N −∆∗ (r) −H0 (r)
k

where
~2 2
H0 (r) = − ∇ − µ + V (r) (11.64)
2m

126
is the noninteracting-electron Hamiltonian. One often uses effective lattice models of the form of tight-binding
models for both H0 (r) and ∆(r). It is now easy to include spatially inhomogeneous situations: Both V (r) and
∆(r) can be chosen spatially dependent. The relevant case is when they are not just lattice periodic. The
corresponding Schrödinger equation
HBdG (r) Ψ(r) = E Ψ(r) (11.65)
with  
Ψ1 (r)
Ψ(r) = (11.66)
Ψ2 (r)
is called the Bogoliubov-de Gennes equation. Note that in this context the gap ∆(r) is usually defined with the
opposite sign, which is just a phase change, so that the explicit minus signs in the off-diagonal components of
HBdG are removed. Furthermore, in Bogoliubov-de Gennes theory, the gap function is typically not evaluated
selfconsistantly from the averages hc−k,↓ ck↑ i. Rather, ∆(r) is treated as a given function characterizing the
strength of superconducting pairing. It is often called the “pairing potential.” But if required it is certainly
possible to evaluate ∆(r) selfconsistently by solving a BCS gap equation in real space.

11.4 Andreev reflection

As an application of the Bogoliubov-de Gennes approach, we study what happens to an electron that impinges on
a normal-superconducting interface from the normal side. We model this situation by the Bogoliubov-de Gennes
Hamiltonian !
~2
− 2m ∇2 − µ ∆0 Θ(x)
HBdG = ~2 2
(11.67)
∆0 Θ(x) 2m ∇ + µ

(note the changed sign of ∆(r)) appearing in

HBdG Ψ(r) = E Ψ(r). (11.68)
In the normal region, x < 0, the component Ψ1 (r) (Ψ2 (r)) is just a superposition of plane waves with wave vectors
k1 (k2 ) which must satisfy
k12 = 2m(µ + E) = kF2 + 2mE, (11.69)
k22 = 2m(µ − E) = kF2 − 2mE, (11.70)
where ~ = 1. In the superconductor, x > 0, we have
 
1 2
− ∇ − µ Ψ1 (r) + ∆0 Ψ2 (r) = E Ψ1 (r), (11.71)
2m
 
1 2
∇ + µ Ψ2 (r) + ∆0 Ψ1 (r) = E Ψ2 (r) (11.72)
2m
1
E + 2m ∇2 + µ
⇒ Ψ2 (r) = Ψ1 (r) (11.73)
∆0
 2
1 2
⇒ ∇ + µ Ψ1 (r) = (E 2 − ∆20 ) Ψ1 (r) (11.74)
2m

and analogously
 2
1 2
∇ + µ Ψ2 (r) = (E 2 − ∆20 ) Ψ2 (r). (11.75)
2m
If the energy is above the gap, |E| > ∆0 , the solutions are plane wave vectors q1 , q2 , where now
!2
2
q1,2
−µ = E 2 − ∆20 > 0 (11.76)
2m

127
  q 
⇒ 2
q1,2 2 2
= 2m µ + E − ∆0 , (11.77)

and amplitudes coupled by Eq. (11.73).

We are here interested in the more surprising case |E| < ∆0 . Since the solution must be continuous across
the interface and is plane-wave-like in the normal region, we make the ansatz

Ψ1 (r) = ei(k1y y+k1z z) Φ1 (x), (11.78)

from which, for x ≥ 0,

2
!2
k1k
1 d2
− + + µ Φ1 (x) = (E 2 − ∆20 ) Φ1 (x) (11.79)
2m 2m dx2
with k1k := (k1y , k1z ). Since this equation is linear with constant coefficients, we make an exponential ansatz

Φ1 (x) = eκx+iqx (11.80)

with κ, q ∈ R. This leads to

2
!2
k1k (κ + iq)2
− + +µ = E 2 − ∆20 (11.81)
2m 2m | {z }
<0
2 2 2
!2 !2
k1k +q iκq κ k1k2 + q2 κ2 iκq
⇒ − +µ+ + = −µ− −
2m m 2m 2m 2m m
2
! 2 !
k1k + q2 κ2 κ2 q 2
2
iκq k1k + q
2
κ2
= −µ− − −2 −µ− = E 2 − ∆20 . (11.82)
2m 2m m2 m 2m 2m

Since the right-hand side is real, we require

2
k1k + q2 κ2
−µ−
=0 (11.83)
2m 2m
2 2 2 2
⇒ κ = k1k + q − 2mµ = k1k − kF2 + q 2 . (11.84)

For the real part it follows that

κ2 q 2
− = E 2 − ∆20 (11.85)
m2  
2 2
2 2
κ q k 1k + q − 2mµ q2
⇒ = = ∆20 − E 2 > 0 (11.86)
m2  m

2

⇒ q 4 + k1k
2
− 2mµ q 2 − m2 (∆20 − E 2 ) = 0 (11.87)
s
2 2mµ − k1k2 2
2mµ − k

2 1k
⇒ q = ± + m2 (∆20 − E 2 )
2 2
" r #
1 2 2
= 2
k − k1k ± 2 2 2 2
kF − k1k + 4m (∆0 − E ) . 2 (11.88)
2 F

Both solutions are clearly real but the one with the minus sign is negative so that q would be imaginary, contrary
to our assumption. Thus the relevant solutions are
s r
1 2
q = ±q1 := ± √ 2 2
kF − k1k + 2
kF2 − k1k + 4m2 (∆20 − E 2 ). (11.89)
2

128
It also follows that
" r #
1 2 2
2 2
κ = k1k − kF2 + 2
k − k1k + 2 2 2 2 2
kF − k1k + 4m (∆0 − E )
2 F
" r #
1 2  2
= k − kF2 + kF2 − k1k
2 + 4m2 (∆20 − E 2 ) (11.90)
2 1k

and s r 2
1
κ = −κ1 := − √ 2
k1k − kF2 + kF2 − k1k
2 + 4m2 (∆20 − E 2 ). (11.91)
2
The positive root exists but would lead to a solution that grows exponentially for x → ∞. For Ψ2 (r) the derivation
is completely analogous. However, Ψ1 (r) and Ψ2 (r) are related by Eq. (11.73), which for exponential functions
becomes a simple proportionality. Therefore, we must have k2k = k1k , which already implies q2 = q1 and κ2 = κ1 .
Then we have
" 2
#
1
E + 2m ∇2 + µ 1 (−κ1 ± iq1 )2 k1k
Ψ2 (r) = Ψ1 (r) = E+ − + µ Ψ1 (r)
∆0 ∆0 2m 2m
" 2
#
1 k1k + q2 κ21 κ1 q1
= E− +µ+ ∓i Ψ1 (r). (11.92)
∆0 | 2m {z 2m} m
=0

Since we already know that

κ21 q12
= ∆20 − E 2 (11.93)
m2
and κ1 , q1 have been defined as positive, we get
p
E ∓ i ∆20 − E 2
Ψ2 (r) = Ψ1 (r), (11.94)
∆0
where the signs ∓ correspond to q = ±q1 . We now write down an ansatz as a linear combination of solutions
and show that it satisfies the Bogoliubov-de Gennes equation and the continuity conditions at the interface. The
ansatz reads  
eik1 ·r + r eik̃1 ·r
Ψ(r) = for x ≤ 0, (11.95)
a eik2 ·r
with k̃1 := (−k1x , k1y , k1z ) and k2 = (k2x , k1y , k1z ) with k2x > 0, where k1x2
+ k1k2
= kF2 + 2mE and k2x
2 2
+ k1k =
kF − 2mE, and
2

α+ ei(q1 x+k1y y+k1z z) + α− ei(−q1 x+k1y y+k1z z)

 
Ψ(r) = e−κ1 x for x ≥ 0. (11.96)
β+ ei(q1 x+k1y y+k1z z) + β− ei(−q1 x+k1y y+k1z z)

Note that k̃1 is the wave vector of a specularly reflected electron. From Eq. (11.94) we get
p
E ∓ i ∆20 − E 2
β± = α± . (11.97)
∆0
From the continuity of Ψ1 , Ψ2 , and their x-derivatives we obtain

1 + r = α+ + α− , (11.98)
a = β+ + β− , (11.99)
ik1x − r ik1x = α+ (−κ1 + iq1 ) + α− (−κ1 − iq1 ), (11.100)
a ik2x = β+ (−κ1 + iq1 ) + β− (−κ1 − iq1 ). (11.101)

129
We thus have six coupled linear equations for the six unknown coefficients r, a, α+ , α− , β+ , β− . The equations
are linearly independent so that they have a unique solution, which we can obtain by standard methods. The
six coefficients are generally non-zero and complex. We do not give the lengthy expressions here but discuss the
results physically.
• The solution in the superconductors decays exponentially, which is reasonable since the energy lies in the
superconducting gap.
• In the normal region there is a specularly reflected electron wave (coefficient r), which is also expected. So
far, the same results would be obtained for a simple potential step. However, explicit evaluation shows that
2
generally |r| < 1, i.e., not all electrons are reflected.
• There is also a term
Ψ2 (r) = a eik2 ·r for x ≤ 0. (11.102)
Recall that the second spinor component was defined by
|Ψk2 i = c−k,↓ |ψBCS i . (11.103)
Hence, the above term represents a spin-down hole with wave vector −k2 . Now we have k2k = k1k ,
2
k2x = kF2 − k1k
2
− 2mE, and k1x
2
= kF2 − k1k
2
+ 2mE. But the last terms ±2mE are small since

kF2
|E| < ∆0  µ = (11.104)
2m
in conventional superconductors. Thus |k2x − k1x | is small and the hole is traveling nearly in the opposite
direction compared to the incoming electron wave. This phenomenon is called Andreev reflection.
11
00
N
00
11
y,z S
electron
~
k1 00
11
00
11
electron−like
quasiparticle
00
11 (evanescent)

00
11
00
11
electron k1
00
11 hole−like

hole
− k2
00
11
00
11
quasiparticle
(evanescent)

00
11 x
Since not all electrons are reflected and in addition some holes are generated, where does the missing charge go?
The quasiparticle states in the superconductor are evanescent and thus cannot accommodate the missing charge.
The only possible explanation is that the charge is added to the superconducting condensate, i.e., that additional
Cooper pairs are formed. (The whole process can also run backwards, in which case Cooper pairs are removed.)
Recall that the condensate does not have a sharp electron number and can therefore absorb or emit electrons
without changing the state. But it can only absorb or emit electrons in pairs. The emerging picture is that if
an incoming electron is not specularly reflected, a Cooper pair is created, which requires a second electron. This
second electron is taken from the normal region, creating a hole, which, as we have seen, travels in the direction
the original electron was coming from.
N
11
00 S
00
11
00
11
e
00
11
00
11 Cooper
h
00
11 pair
00
11
130
Andreev bound states
An interesting situation arises if a normal region is delimited by superconductors on two sides. We here only
qualitatively consider a superconductor-normal-superconductor (SNS) heterostructure. Similar effects can also
occur for example in the normal core of a vortex.
If no voltage is applied between the two superconductors an electron in the normal region, with energy within
the gap, is Andreev reflected as a hole at one interface. It is then Andreev reflected as an electron at the other
interface. It is plausible that multiple reflections can lead to the formation of bound states. The real physics is
somewhat more complicated since the electron is also partially specularly reflected as an electron. It is conceptually
clear, though, how to describe Andreev bound states within the Bogoliubov-de Gennes formalism: We just have
to satisfy continuity conditions for both interfaces. We expect a discrete set of bound states to exist, in analogy

11
00
to the finite square well in elementary quantum mechanics.
0110
E S
00
11
00
11
N
1010
S

00
11 electron
2∆ 0
00
11 10 2∆ 0
00
11
00
11
hole
1010
x
00
11 10
If Andreev reflection dominates, as assumed for the sketch above, a Cooper pair is emitted into the right su-
perconductor for every reflection at the right interface. Conversely, a Cooper pair is absorbed from the left
superconductor for every reflection at the left interface. This corresponds to a supercurrent through the device.
Andreev bound states thus contribute to the Josephson effect in superconductor-normal-superconductor junctions.
If we apply a voltage V , the situation changes dramatically. The main insight was formulated by Klapwijk,
Blonder, and Tinkham (1982). If an electron moving, say, to the right, increases its kinetic energy by eV due
to the bias voltage, an Andreev reflected hole traveling to the left also increases its kinetic energy by eV since
it carries the opposite charge. An electron/hole Andreev-reflected multiple times can thus gain arbitrarily high

11
00 11
00
energies for any non-vanishing bias voltage.

E S
00
11
00
11
N
00
11
00
11
S

2∆ 0
00
11
00
11 00
11
00
11
2eV 00
11 00
11
00
11 00
11
eV

x
00
11 00
11
In particular, an electron-like quasiparticle from an occupied state below the gap in, say, the left superconductor
can after multiple reflections emerge in a previously unoccupied state above the gap in the right superconductor.
A new transport channel becomes available whenever the full gap 2∆0 is an odd integer multiple of eV :
2∆0 = (2n + 1) eV, n = 0, 1, . . . (11.105)
∆0
⇒ eV = , n = 0, 1, . . . (11.106)
n + 12
If 2∆0 is an even multiple of eV a hole emerges above the gap in the left superconductor. At the gap edge, the
Bogoliubov quasiparticles consist of equal parts electrons and holes. Hence, an incoming hole can just as easily
be absorbed as an incoming electron. Thus, new transport channels actually open at all multiples of eV , i.e., at
2∆0 = n eV, n = 1, 2, . . . (11.107)

131
 2∆0
⇒ eV = , n = 1, 2, . . . (11.108)
n
The case n = 0 corresponds to direct quasiparticle transfer from one superconductor to the other, similar to
quasiparticle tunneling in a superconductor-insulator-superconductor junction.
The opening of new transport channels leads to structures in the current-voltage characteristics below the gap,
specifically to peaks in the differential conductance dI/dV :

dI
dV

2
− 2∆ 0 0 3 ∆0 ∆0 2∆ 0 eV
1
2 ∆0

132
12

Unconventional pairing

In this chapter we first discuss why interactions different from the phonon-mediated one might lead to unconven-
tional pairing, that is to a gap function ∆k with non-trivial symmetry. Then we will briefly consider the origin
of such interactions.

12.1 The gap equation for unconventional pairing

We will still use the BCS gap equation even when discussing unconventional superconductors. While the BCS
mean-field theory might not be quantitatively correct in such cases, it will give a clear understanding of why
the gap ∆k can have non-trivial symmetry. To get started, we briefly review results from BCS theory. The
phonon-mediated interaction was derived in Sec. 8.4,

(iνn )2
RPA
Veff (q, iνn ) = VCRPA (q, iνn ) , (12.1)
(iνn )2 − ωq2 (iνn )

where VCRPA is the screened Coulomb interaction and ωq is the renormalized phonon dispersion. The retarded
interaction at small frequencies is

e2 ν2
R
Veff (q, iνn ) ∼
= 4π , (12.2)
q 2 + κ2s ν 2 − ωqR (ν)2 + i0+ sgn ν

where κs is the inverse screening length. The behavior at small distances r and small but non-zero frequency
ν is determined by VeffR
at large q, where ωq can by approximated by the Debye frequency. The interaction is
thus attractive at low but nonzero frequencies (because of the last factor) and decays like 1/r for small r. The
interaction is strongest at the same site in a tight-binding model (the pole of 1/r is cut off by the finite extent of
the atomic orbitals). In order to understand the physics, it makes sense to replace the interaction by a simplified
one that is completely local (attractive Hubbard model) or, equivalently, constant in k space, as we have done
above. However, the BCS gap equation (10.40),

1 X ∆k0
∆k = − Vkk0 [1 − 2 nF (Ek0 )] (12.3)
N 0 2Ek0
k

is in fact far more general. In the gap equation, Vkk0 describes the amplitude for scattering of two electrons with
momenta k0 and −k0 and opposite spins into states with momenta k and −k.
Let us first consider the case that the interaction is local in real space (flat in k space) but repulsive. This
would apply if the phonons were for some reason ineffective in overscreening the Coulomb interaction. Then we
obtain
1 X ∆k0
∆k = − V0 [1 − 2 nF (Ek0 )] (12.4)
N 0 2Ek0
k

133
with V0 > 0. The right-hand side is clearly independent of k so that we have ∆k = ∆0 and can cancel a factor
of ∆0 if it is non-zero:
V0 X 1 − 2 nF (Ek0 )
1=− . (12.5)
N 0 2Ek0
k

But now the right-hand side is always negative. Consequently, there is no solution with ∆0 6= 0 and thus no
superconductivity for a k-independent repulsion.
Now let us look at a strong interaction between nearest-neighbor sites. We consider a two-dimensional square
lattice for simplicity and since it is thought to be a good model for the cuprates. In momentum space, a nearest-
neighbor interaction is written as

Vkk0 = 2V1 [cos (kx − kx0 )a + cos (ky − ky0 )a], (12.6)

where V1 > 0 (V1 < 0) for a repulsive (attractive) interaction.

It is now important to realize which terms in the momentum sum in the gap equation (12.3) are dominant.
The factor 1 − nF (Ek0 ) is on the order of unity since Ek0 ≥ |∆k0 | and |∆k0 |/kB T = O(1). On the other hand, the
factor 1/Ek0 is largest on the normal-state Fermi surface, where Ek0 = |∆k0 |. We therefore concentrate on these
states. The coupling to states away from the Fermi surface is additionally suppressed by the factor ∆k0 /Ek0 . This
is a relatively weak, power-law suppression.
Let us first consider the repulsive case V1 > 0. Then Vkk0 is most strongly repulsive (positive) for momentum
transfer k − k0 → 0. But ∆k should be a smooth function of k, thus for k and k0 close together, ∆k and ∆k0 are
also similar. In particular, ∆k will rarely change its sign between k and k0 . Consequently, the right-hand side
of the gap equation always contains a large contribution with sign opposite to that of ∆k , coming from the sum
over k0 close to k. Hence, a repulsive nearest-neighbor interaction is unlikely to lead to superconductivity.
For the attractive case, V1 < 0, Vkk0 is most strongly attractive for k − k0 → 0 and most strongly repulsive
for k − k0 → (π/a, π/a) and other vectors related by lattice symmetries. The attraction at small q = k − k0 is
always favorable for superconductivity. However, we also have an equally strong repulsion around q ≈ (π/a, π/a).
A critical situation thus arises if both k and k0 lie close to the Fermi surface and their difference is close to
(π/a, π/a). The central insight is that this can still help superconductivity if the gaps ∆k and ∆k0 at k and
k0 ≈ k + (π/a, π/a), respectively, have opposite sign. In this case the contribution to the right-hand side of the
gap equation from such k0 has the same sign as ∆k since Vkk0 > 0 and there is an explicit minus sign. Note that
superconductivity can even be stabilized by a purely repulsive interaction around q ≈ (π/a, π/a) if the interaction
around q ≈ 0 is not too strongly repulsive. We will get back to this point shortly.
This effect is crucial for the cuprates, which do have an effective attractive nearest-neighbor interaction and
have a large normal-state Fermi surface shown here for a two-dimensional model:
ky

0 kx

The vector Q in the sketch is Q = (π/a, π/a). Following the previous discussion, ∆k close to the Fermi surface
should have different sign between points separated by Q. On the other hand, the small-q attraction favors gaps
∆k that change sign “as little as possible.” By inspection, these conditions are met by a gap changing sign on the
diagonals:

134
 ky

0 kx

This type of gap is called a dx2 −y2 -wave (or just d-wave) gap since it has the symmetry of a dx2 −y2 -orbital
(though in k-space, not in real space). The simplest gap function with this symmetry and consistent with the
lattice structure is
∆k = ∆0 (cos kx a − cos ky a). (12.7)
Recall that the gap function away from the Fermi surface is of limited importance because there Ek ∼ = |ξk |. The
d-wave gap ∆k also differs from the conventional, approximately constant s-wave gap in that it has zeros on the
normal-state Fermi surface. These zeros are called (gap) nodes. In the present case, they appear in the (11) and
equivalent directions. The quasiparticle dispersion in the vicinity of such a node is
q q
Ek = ξk2 + |∆k | ∼
2
= (k − µ)2 + ∆20 (cos kx a − cos ky a)2 . (12.8)

One node is at  
kF 1
k0 = √ . (12.9)
2 1
Note that kF depends on the direction in momentum space. Evidently, the dx2 −y2 -wave gap has point nodes for
our two-dimensional model. For a three-dimensional system of layers with negligible coupling, the normal-state
Fermi surface becomes a cylinder along the kz direction and the nodes become straight lines in the same direction.
If the dispersion along kz is not negligible but still small the Fermi surface is deformed but retains the topology of
a cylinder and the nodes remain lines where the Fermi surface intersect with the high-symmetry planes ky = ±kx .
Writing k = k0 + q and expanding for small q, we obtain
q
Ek0 +q ∼
= (vF · q)2 + ∆20 [−(sin kx0 a)qx a + (sin ky0 a)qy a]2 , (12.10)

where  
∂k vF 1
vF := =√ (12.11)
∂k k=k0 2 1
is the normal-state Fermi velocity at the node. Thus
s   2 q
∼ kF a kF a
2 2
Ek0 +q = (vF · q) + ∆0 −a sin √ , a sin √ · q = (vF · q)2 + (vqp · q)2 , (12.12)
2 2

where  
kF a −1
vqp := ∆0 a sin √ ⊥ vF . (12.13)
2 1
Thus the quasiparticle dispersion close to the node is a cone like for massless relativistic particles, but with
different velocities in the directions normal and tangential to the Fermi surface. Usually one finds vqp  vF since
vqp is controlled by ∆0  µ. The following sketch shows equipotential lines of Ek .

135
 ky

kx
The fact that the gap closes at some k points implies that the quasiparticle density of states does not have a gap.
At low energies we can estimate it from our expansion of the quasiparticle energy,
1 X
Ds (E) = δ(E − Ek )
N
k
 
4 X
q
∼
= 2
δ E − (vF · q) + (vqp · q) 2
N q
d2 q
Z  q 
= 4auc 2
δ E − (vF · q) + (vqp · q) 2
(2π)2
d2 u 
Z
4auc q
2 + u2

= δ E − u x y
vF vqp (2π)2
Z ∞
2auc 2auc
= du u δ(E − u) = E, (12.14)
πvF vqp 0 πvF vqp
where auc is the area of the two-dimensional unit cell. We see that the density of states starts linearly at small
energies. The full dependence is sketched in the following figure, where ∆max is the maximum of ∆k for k on the
Fermi surface.
Ds (E )
Dn (EF)

0 ∆ max E
The linear instead of exponential temperature dependence at low T obviously leads to distinct thermodynamic
properties. For example, the electronic contribution to the specific heat is not exponentially suppressed but a
power law c ∼ T 2 .
An additional nice feature of the d-wave gap is the following: The interaction considered above is presumably
not of BCS (Coulomb + phonons) type. However, there is also a strong short-range (local) Coulomb repulsion,
which we here assume not to be overscreened by phonon exchange. This repulsion can again be modeled by a
constant V0 > 0 in k-space. This additional interaction adds the term
V0 X 1 − 2 nF (Ek0 )
− ∆k0 (12.15)
N 0 2Ek0
k

to the gap equation. But since Ek0 does not change sign under rotation of k0 by π/2 (i.e., 90◦ ), while ∆k0 does
change sign under this rotation, the sum over k0 vanishes. d-wave pairing is thus robust against on-site Coulomb
repulsion.

136
12.2 Cuprates
Estimates of the critical temperature Tc of cuprates based on phonon-exchange and using experimentally known
values of the Debye frequency, the electron-phonon coupling, and the normal-state density of states are much
lower than the observed critical temperatures. Also, as we have seen, such an interaction tends to be flat in
k-space, which favors an s-wave gap. An s-wave gap is inconsistent with nearly all pertinent experiments on the
cuprates. The last section has shown that dx2 −y2 -wave pairing in the cuprates is plausible if there is a repulsive
interaction for momentum transfers q ≈ (π/a, π/a). We will now discuss the possible origin of this attraction.

AFM
Mott SC
0 hole doping x

A glance at typical phase diagrams shows that the undoped cuprates tend to be antiferromagnetic. Weak hole
doping or slightly stronger electron doping destroy the antiferromagnetic order, and at larger doping, supercon-
ductivity emerges. At even larger doping (the “overdoped” regime), superconductivity is again suppressed. Also
in many other unconventional superconductors superconductivity is found in the vicinity of, but rarely coexist-
ing with, magnetic order. This is true for most pnictide and heavy-fermion superconductors. The vicinity of a
magnetically ordered phase makes itself felt by strong magnetic fluctuations and strong, but short-range, spin
correlations. These are reflected by an enhanced spin susceptibility.
At a magnetic second-order phase transition, the static spin susceptibility χq (ν = 0) diverges at q = Q, where
Q is the ordering vector. It is Q = 0 for ferromagnetic order and Q = (π/a, π/a) for checkerboard (Néel) order
on a square lattice. Even some distance from the transition or at non-zero frequencies ν, the susceptibility χq (ν)
tends to have a maximum close to Q. Far away from the magnetic phase or at high frequencies this remnant of
magnetic order becomes small. This discussion suggests that the exchange of spin fluctuations, which are strong
close to Q, could provide the attractive interaction needed for Cooper pairing.

The Hubbard model

The two-dimensional, single-band, repulsive Hubbard model is thought (by many experts but not by everyone) to
be the simplest model that captures the principal physics of the cuprates. The Hamiltonian reads, in real space,

tij c†iσ cjσ + U

X †
ci↑ ci↑ c†i↓ ci↓
X
H=− (12.16)
ijσ i

with U > 0 and, in momentum space,

U X †
ξk c†kσ ckσ + c† 0
X
H= c ck0 ↓ ck↑ , (12.17)
N 0 k+q,↑ k −q,↓
kσ kk q

where the chemical potential was absorbed into the dispersion ξk = k − µ. The Hubbard (U ) term approximates
the screened Coulomb repulsion, assuming that screening makes the non-local repulsion negligibly small. The
underlying lattice in real space is a two-dimensional square lattice with each site i corresponding to a Cu+ ion.

137
 Cu+ O 2−
The undoped cuprate parent compounds have, from simple counting, an odd number of electrons per unit cell.
Thus for them the single band must be half-filled, while for doped cuprates it is still close to half filling. The anti-
ferromagnetic parent compounds are nevertheless insulators. This is attributed to the strong Hubbard interaction
U , i.e., they are Mott insulators.
The transverse spin susceptibility is defined by

χ+− (q, τ ) = − Tτ S + (q, τ ) S − (−q, 0) , (12.18)

where τ is the imaginary time, Tτ is the time-ordering directive, and

S ± (q, τ ) := S x (q, τ ) ± iS y (q, τ ) (12.19)

with
1 X † σα 0
S α (q, τ ) := √ ck+q,σ (τ ) σσ ckσ0 (τ ) (12.20)
N kσσ0 2
are electron-spin operators. σ = (σ x , σ y , σ z ) is the vector of Pauli matrices. The susceptibility can be rewritten
as
1 XD E
χ+− (q, τ ) = − Tτ (c†k+q,↑ ck↓ )(τ ) (c†k0 −q,↓ ck0 ↑ )(0) . (12.21)
N 0
kk
In the non-interacting limit of U → 0, the average of four fermionic operators can be written in terms of products
of averages of two operators (Wick’s theorem). The resulting bare susceptibility reads
1 XD †
E D
†
E
χ+−
0 (q, τ ) = − T c
τ k+q,↑ (τ ) c k+q,↑ (0) T c
τ k↓ (τ ) c k↓ (0)
N 0 0
k
1 D E D E
Tτ ck+q,↑ (0) c†k+q,↑ (τ ) Tτ ck↓ (τ ) c†k↓ (0)
X
=
N 0 0
k
1 X 0
= 0
Gk+q,↑ (−τ ) Gk↓ (τ ). (12.22)
N
k

The Fourier transform as a function of the bosonic Matsubara frequency iνn is

Zβ
χ+−
0 (q, iνn ) = dτ eiνn τ χ+−
0 (q, τ )
0
Zβ
1 X1X 0 1 X −iωn τ 0
= dτ eiνn τ e−iωn (−τ ) Gk+q,↑
0
(iωn0 ) e Gk↓ (iωn )
N β 0 β iω
0 k iωn n

1 X 1 X 1 X1X 0
= 0
β δνn +ωn0 ,ωn Gk+q,↑ (iωn0 ) Gk↓
0
(iωn ) = G 0
(iωn − iνn ) Gk↓ (iωn )
N β2 0
N β iω k+q,↑
k iωn ,iωn k n

1 X1X 1 1
= . (12.23)
N β iω iωn − iνn − ξk+q iωn − ξk
k n

138
This expression can be written in a more symmetric form concerning the appearing momenta and frequencies by
making use of the identity ξk = ξ−k and replacing the summation variables k by −k − q and iωn by iωn + iνn .
The result is
1 X1X 1 1
χ+−
0 (q, iνn ) = . (12.24)
N β iω iωn − ξk iωn + iνn − ξk+q
k n

The Matsubara frequency sum can be evaluated using methods from complex analysis (contour integration, residue
theorem). We here give the result without proof,
1 X nF (ξk ) − nF (ξk+q )
χ+−
0 (q, iνn ) = . (12.25)
N iνn + ξk − ξk+q
k

Note that the same result is found for the bare charge susceptibility except for a spin factor of 2. Diagrammatically,
the result can be represented by the bubble diagram, introduced in Sec. 8.2,
k , , iωn
1
χ+−
0 (q, iνn ) = − Π0 (q, iνn ) = − . (12.26)
2
k + q , , iωn + i νn
What changes when we switch on the Hubbard interaction U ? In analogy to the RPA theory for the screened
Coulomb interaction we might guess that the RPA spin susceptibility is given by

? U
− χ+−
RPA = + + ··· (12.27)

but the second and higher terms vanish since they contain vertices at which the Hubbard interaction supposedly
flips the spin,
U , (12.28)

which it cannot do. On the other hand, the following ladder diagrams do not vanish and represent the RPA
susceptibility:

− χ+−
RPA = + U + + ··· (12.29)

Since the Hubbard interaction does not depend on momentum, this series has a rather simple mathematical form,

χ+− +− +− +− +− +− +−
RPA (q, iνn ) = χ0 (q, iνn ) + χ0 (q, iνn ) U χ0 (q, iνn ) + χ0 (q, iνn ) U χ0 (q, iνn ) U χ0 (q, iνn ) + · · ·
= χ+− +− +− +−
0 (q, iνn )[1 + U χ0 (q, iνn ) + U χ0 (q, iνn ) U χ0 (q, iνn ) + · · · ]
χ+−
0 (q, iνn )
= (12.30)
1 − U χ+−
0 (q, iνn )

[the signs in the first line follow from the Feynman rules, in particular each term contains a single fermionic loop,
which gives a minus sign, which cancels the explicit one in Eq. (12.29)]. The RPA spin susceptibility can be
evaluated numerically for given dispersion ξk . It is clear that it predicts an instability of the Fermi liquid if the
static RPA spin susceptibility

χ0+−,R (q, 0)
χ+−,R +− +
RPA (q, 0) = χRPA (q, iνn → ν + i0 )|ν→0 ≡ (12.31)
1 − U χ+−,R
0 (q, 0)
diverges at some q = Q, i.e., if
U χ+−,R
0 (Q, 0) = 1. (12.32)

139
Since the spin susceptibility diverges, this would be a magnetic ordering transition with ordering vector Q. Note
that the RPA is not a good theory for the antiferromagnetic transition of the cuprates since it is a resummation
of a perturbative series in U/t, which is not small in cuprates. t is the typical hopping amplitude. Nevertheless
it gives qualitatively reasonable results in the paramagnetic phase, which is of interest for superconductivity.
The numerical evaluation in the paramagnetic phase at intermediate doping and not too low temperatures
gives a broad and high peak in χ+−,R
RPA (q, 0) centered at Q = (π/a, π/a). This is consistent with the ordering at
the same Q observed at weak doping.

χ +RPA
−, R
((q, q), 0)

q
0 π 2π
a a

At lower temperatures, details become resolved that are obscured by thermal broadening at high T . The RPA
and also more advanced approaches are very sensitive to the electronic bands close to the Fermi energy; states
with |ξk |  kB T have exponentially small effect on the susceptibility. Therefore, the detailed susceptibility at low
T strongly depends on details of the model Hamiltonian. Choosing nearest-neighbor and next-nearest-neighbor
hopping in such a way that a realistic Fermi surface emerges, one obtains a spin susceptiblity with incommensurate
peaks at πa (1, 1 ± δ) and πa (1 ± δ, 1).

qy

π
a

0 π qx
a

These peaks are due to nesting: Scattering is enhanced between parallel portions of the Fermi surface, which in
turn enhances the susceptibility [see M. Norman, Phys. Rev. B 75, 184514 (2007)].

140
 ky

π
a
π (1− δ, 1)
a
π (1, 1− δ)
a

π kx
a

The results for the spin susceptibility are in qualitative agreement with neutron-scattering experiments. However,
the RPA overestimates the tendency toward magnetic order, which is rectified by more advanced approaches.

Spin-fluctuation exchange
The next step is to construct an effective electron-electron interaction mediated by the exchange of spin fluctua-
tions. The following diagrammatic series represents the simplest way of doing this, though certainly not the only
one:

U
Veff := + + + ··· (12.33)

Veff flips the electron spins, indicating that it is indeed due to magnetic fluctuations. (As a technical remark,
note that the external legs do not represent electronic Green functions but only indicate the states of incoming
and outgoing electrons.) The series is very similar to the one for the RPA susceptibility. Indeed, the effective
interaction is

Veff (q, iνn ) = U + U χ+− +− +−

0 (q, iνn ) U + U χ0 (q, iνn ) U χ0 (q, iνn ) U + · · ·
= U + U 2 [χ+− +− +−
0 (q, iνn ) + χ0 (q, iνn ) U χ0 (q, iνn ) + · · · ]
= U + U 2 χ+−
RPA (q, iνn ). (12.34)

Typically, one goes beyond the RPA at this point by including additional diagrams. In particular, also charge
fluctuations are included through the charge susceptibility and the bare Green function G 0 is replaced by a
selfconsistent one incorporating the effect of spin and charge fluctuations on the electronic self-energy. This
leads to the fluctuation-exchange approximation (FLEX ). One could now obtain the Cooper instability due to
the RPA or FLEX effective interaction in analogy to Sec. 9.1 and use a BCS mean-field theory to describe the
superconducting state. However, since the effective interaction has a nontrivial frequency dependence, one usually
employs an extension of BCS theory that includes the full frequency dependence, known as Eliashberg theory. It
is typically more successful than BCS theory if the interaction is not weak (i.e., the typical interaction times the
electronic density of states is not small). Since the effective interaction is, like the spin susceptibility, strongly
peaked close to (π/a, π/a) and repulsive, it favors dx2 −y2 -wave pairing, as we have seen.
The numerical result of the FLEX for Tc and for the superfluid density ns are sketched here:

141
 Tc ns

0 doping x

The increase of Tc toward low doping is reasonable, in principle: Superconductivity is driven by spin fluctuations,
which are enhanced as one approaches the antiferromagnetic phase transition. However, the monotonous increase
of Tc for decreasing doping does not agree with the experimentally observed dome. There are several aspects that
make the region of low doping (underdoping) difficult to treat theoretically. One is indicated in the sketch: ns is
strongly reduced, which indicates that superconductivity may be in some sense fragile in this regime. Furthermore,
the cuprates are nearly two-dimensional solids. In fact we have used a two-dimensional model so far. If we take
this seriously, we know from chapter 7 that any mean-field theory, which Eliashberg theory with FLEX effective
interaction still is, fails even qualitatively. Instead, we expect a BKT transition at a reduced critical temperature.
We now interpret the FLEX critical termperature as the mean-field termperature TMF and the FLEX superfluid
density as the unrenormalized superfluid density n0s . We have seen in chapter 7 that the bare stiffness K0 is
proportional to n0s . The BKT transition temperature Tc is defined by K(l → ∞) = 2/π. It is thus reduced by
small n0s , corresponding to a small initial value K(0) ≡ K0 . This is physically clear: a small stiffness makes it
easy to create vortex-antivortex pairs. The real cuprates are three-dimensional materials, although the hopping
in the direction normal to the CuO2 planes is small. For this reason, and also because of the coupling to the
electromagnetic field discussed in Sec. 7.2, the behavior close to the transition is not described by BKT theory.
However, BKT theory does give a reasonable estimate for the doping dependence of the critical temperature Tc
in the underdoped regime.
Tc is of course also limited by TMF —it can never be larger than TMF —which is most important in the overdoped
regime. A BKT theory on top of the FLEX gives the following phase diagram, which is in qualitative agreement
with experiments:

TMF ns0

vortex
fluctuations

Tc superconductor
0 x
This scenario is consistent with the Nernst effect (an electric field measured normal to both an applied magnetic
field and a temperature gradient) in underdoped cuprates, which is interpreted in terms of free vortices in a broad
temperature range, which we would understand as the range from Tc to TMF .
Also note that spin fluctuations strongly affect the electronic properties in underdoped cuprates up to a
temperature T ∗ significantly higher than TMF . For example, below T ∗ the electronic density of states close to
the Fermi energy is suppressed compared to the result of band-structure calculations. The FLEX describes this
effect qualitatively correctly. This suppression is the well-known but not well-understood pseudogap.

Quantum critical point

The previously discussed approach relies on a resummation of a perturbative series in U/t. This is questionable
for cuprates, where U/t is on the order of 3. Many different approaches have been put forward that supposedly
work in this strong-coupling regime. They emphasize different aspects of the cuprates, showing that it is not even

142
clear which ingredients are the most important for understanding the phase diagram. Here we will review a line
of thought represented by Chandra Varma and Subir Sachdev, among others. Its starting point is an analysis of
the normal region of the phase diagram.

T
T*
’’strange metal’’
ρ~T
pseudo− Fermi
gap liquid

AFM superconductor
0 x
Roughly speaking, there are three regimes in the normal-conducting state:
• a pseudo-gap regime below T ∗ at underdoping, in which the electronic density of states at low energies is
suppressed,
• a “strange-metal” regime above the superconducting dome, without a clear suppression of the density of
states but with unusual temperature and energy dependencies of various observables, for example a resistivity
linear in temperature, ρ ∝ T ,
• an apparently ordinary normal-metal (Fermi-liquid) regime at overdoping, with standard ρ ∝ const + T 2
dependence.
The two crossover lines look very much like what one expects to find for a quantum critical point (QCP), i.e., a
phase transition at zero temperature.
T
quantum
critical
affected affected
by phase I by phase II

phase I phase II
QCP tuning parameter δ
δc

The regions to the left and right have a characteristic energy scale (δ) inherited from the respective ground
state (at T = 0), which dominates the thermal fluctuations. The energy scales go to zero at the QCP, i.e., for
δ → δc from both sides. Right at the QCP there then is no energy scale and energy or temperature-dependent
quantities have to be power laws. The emerging idea is that the superconducting dome hides a QCP between
antiferromagnetic (spin-density-wave) and paramagnetic order at T = 0. The spin fluctuations associated with
this QCP become stronger as it is approached. Consequently, the superconductivity caused by them is strongest
and has the highest Tc right above the QCP. However, it is not immediately clear why antiferromagnetic long-
range order appears to be suppressed already at very weak doping, while the QCP is located in the middle of
the superconducting dome. Compare the previous argument: There, the spin fluctuations are assumed to be
strongest close to the (finite-temperature) antiferromagnetic phase and one needs to invoke small ns and vortex
fluctuations to argue why the maximum Tc is not close to the antiferromagnetic phase.

143
12.3 Pnictides
The iron pnictide superconductors are a more heterogeneous group than the cuprates. However, for most of
them the phase diagram is roughly similar to the one of a typical cuprate in that superconductivity emerges at
finite doping in the vicinity of an antiferromagnetic phase. This antiferromagnetic phase is metallic, not a Mott
insulator, though, suggesting that interactions are generally weaker in the pnictides.
T
CeFeAsO1−x Fx

tet horo
140 K

ort
rag mb
on
al
ic
AFM superconductor
metal
0 0.05 doping x

The crystal structure is quasi-two-dimensional, though probably less so than in the cuprates. The common
structural motif is an iron-pnictogen, in particular Fe2+ As3− , layer with Fe2+ forming a square lattice and As3−
sitting alternatingly above and below the Fe2+ plaquettes.

As 3−
Fe 2+

While the correct unit cell contains two As and two Fe ions, the glide-mirror symmetry with respect to the Fe
plane allows to formulate two-dimensional (but not three-dimensional) models using a single-iron unit cell. The
fact that different unit cells are used leads to some confusion in the field. Note that since the single-iron unit
cell is half as large as the two-iron unit cell, the corresponding single-iron Brillouin zone is twice as large as the
two-iron Brillouin zone.
A look at band-structure calculations or ARPES data shows that the Fermi surface of pnictides is much more
complicated than the one of cuprates. The kz = 0 cut typically shows four or five Fermi pockets; the small pocket
at the M point may be absent.

144
 ky
M
X’ hole−like

Q2
hole−like
electron−like
Γ X kx
Q1

single−iron Brillouin zone

The (probably outer) hole pocket at Γ and the electron pockets at X and X0 are well nested with nesting vectors
Q1 = (π/a, 0) and Q2 = (0, π/a), respectively. Not surprisingly, the spin susceptibility is peaked at Q1 and Q2 in
the paramagnetic phase and in the antiferromagnetic phase the system orders antiferromagnetically at either of
three vectors, i.e., stripe like. Incidentally, the antiferromagnet emerges through the formation and condensation
of electron-hole pairs (excitons), described by a BCS-type theory. The same excitonic instability is for example
responsible for the magnetism of chromium.
Assuming that the exchange of spin fluctuations in the paramagnetic phase is the main pairing interaction,
the gap equation
1 X 1 − 2 nF (Ek0 )
∆k = − Vkk0 ∆k 0 , (12.35)
N 0 2Ek0
k

with Vkk0large and positive for k − k = ±Q1 or ±Q2 , favors a gap function ∆k that changes sign between k and
0

k + Q1 and between k and k + Q2 , following the general discussion in Sec. 12.1. This is most easily accomodated
by a nodeless gap changing sign between the electron and hole pockets:

ky

kx

This gap has s-wave symmetry since it does not show lower symmetry than the lattice, unlike d-wave. To

145
emphasize the sign change, it is often called an s± -wave (or s+− -wave) gap. The simplest realization reads

∆k = ∆0 cos kx a cos ky a. (12.36)

12.4 Triplet superconductors and He-3

So far, we have assumed that Cooper pairs are formed by two electrons with opposite spin so that the total
spin of the pair vanishes (spin-singlet pairing). This assumption becomes questionable in the presence of strong
ferromagnetic interactions, which favor parallel spin alignment. If superconductivity is possible at all in such a
situation, we could expect to find spin-1 Cooper pairs. Since they would be spin triplets, one is talking of triplet
superconductors. This scenario is likely realized for a few organic salts and certain heavy-fermion compounds
such as UPt3 . Furthermore, it has long been favored for Sr2 RuO4 , which is, interestingly, isostructural to
the prototypical cuprate La2 CuO4 . Recent experiments indicate, however, that the situation might be more
complicated. A triplet pairing state is more certain to be responsible for the superfluidity of He-3, where neutral
He-3 atoms instead of charged electrons form Cooper pairs, see Sec. 2.2.
Formally, we restrict ourselves to a BCS-type mean-field theory. We generalize the effective interaction to
allow for an arbitrary spin dependence,
1 X X
ξk c†kσ ckσ + Vστ σ0 τ 0 (k, k0 ) c†kσ c†−k,τ c−k0 ,τ 0 ck0 σ0 ,
X
H= (12.37)
N 0 0 0
kσ kk στ σ τ

where σ, τ, σ 0 , τ 0 =↑, ↓ are spin indices. In decomposing the interaction, we now allow the averages hc−k,τ ckσ i to
be non-zero for all τ , σ. Thus the mean-field Hamiltonian reads
1 X X D E 
ξk c†kσ ckσ + c†kσ c†−k,τ c−k0 ,τ 0 ck0 σ0 + c†kσ c†−k,τ hc−k0 ,τ 0 ck0 σ0 i + const.
X
HMF = Vστ σ0 τ 0 (k, k0 )
N 0
kσ kk στ σ 0 τ 0
(12.38)
We define
1 XX
∆στ (k) := − Vστ σ0 τ 0 (k, k0 ) hc−k0 ,τ 0 ck0 σ0 i (12.39)
N 0 0 0
k σ τ

so that
1 XX D E
∆∗στ (k) = − Vσ0 τ 0 στ (k0 , k) c†k0 σ0 c†−k0 ,τ 0 . (12.40)
N 0 0 0
k σ τ

Here, we have used that ∗

Vστ σ0 τ 0 (k, k0 ) = Vσ0 τ 0 στ (k0 , k), (12.41)


which follows from hermiticity of the Hamiltonian H. Then

ξk c†kσ ckσ − ∆στ (k) c†kσ c†−k,τ + const.

X X X
HMF = ∆∗στ (k) c−k,τ ckσ − (12.42)
kσ kστ kστ

The gap functions ∆στ (k) now form a matrix in spin space,
 
∆↑↑ (k) ∆↑↓ (k)
ˆ
∆(k) = . (12.43)
∆↓↑ (k) ∆↓↓ (k)

The function ∆(k)

ˆ has an important symmetry property that follows from the symmetry of averages hc−k,τ ckσ i:
It is clear that
hckσ c−k,τ i = −hc−k,τ ckσ i. (12.44)
Furthermore, by relabeling σ ↔ τ , σ 0 ↔ τ 0 , k → −k, k0 → −k0 in the interaction term of H, we see that the
interaction strength must satisfy the relation

Vστ σ0 τ 0 (k, k0 ) = Vτ στ 0 σ0 (−k, −k0 ). (12.45)

146
Thus
1 XX 1 XX
∆τ σ (−k) = − Vτ στ 0 σ0 (−k, −k0 ) hck0 σ0 c−k0 ,τ 0 i = + Vστ σ0 τ 0 (k, k0 ) hc−k0 ,τ 0 ck0 σ0 i = −∆στ (k)
N 0 0 0 N 0 0 0
k σ τ k σ τ
(12.46)
or, equivalently,
ˆ
∆(−k) ˆ T (k),
= −∆ (12.47)
where the superscript T denotes the matrix transposition. We now want to write ∆(k)
ˆ in terms of singlet and
triplet components. From elementary quantum theory, a spin-singlet pair is created by
c†k↑ c†−k,↓ − c†k↓ c†−k,↑
s†k := √ , (12.48)
2
while the m = 1, 0, −1 components of a spin-triplet pair are created by
t†k1 := c†k↑ c†−k,↑ , (12.49)
c†k↑ c†−k,↓ + c†k↓ c†−k,↑
t†k0 := √ , (12.50)
2
t†k,−1 := c†k↓ c†−k,↓ , (12.51)
respectively. Alternatively, we can transform onto states aligned with the x, y, and z axes. This is analogous to
the mapping from (l = 1, m) eigenstates onto px , py , and pz orbitals for the hydrogen atom. The new components
are created by
−c†k↑ c†−k,↑ + c†k↓ c†−k,↓
t†kx := √ , (12.52)
2
c†k↑ c†−k,↑ + c†k↓ c†−k,↓
t†ky := i √ , (12.53)
2
c†k↑ c†−k,↓ + c†k↓ c†−k,↑
t†kz := √ = t†k0 , (12.54)
2
which form a vector t†k . The term in HMF involving ∆στ (k) can now be expressed in terms of the new operators,

−t†kx − it†ky t†kx − it†ky

" #
X † †
X s†k + t†kz −s†k + t†kz
− ∆στ (k)ckσ c−k,τ = − ∆↑↑ (k) √ + ∆↑↓ (k) √ + ∆↓↑ (k) √ + ∆↓↓ (k) √
kστ k
2 2 2 2
! √ h i
2 ∆k s†k + d(k) · t†k
X
=− (12.55)
k
√
(the factor 2 is conventional), which requires
∆↑↓ (k) − ∆↓↑ (k)
∆k = , (12.56)
2
−∆↑↑ (k) + ∆↓↓ (k)
dx (k) = , (12.57)
2
∆↑↑ (k) + ∆↓↓ (k)
dy (k) = −i , (12.58)
2
∆↑↓ (k) + ∆↓↑ (k)
dz (k) = . (12.59)
2
The term in HMF involving ∆∗στ (k) is just the hermitian conjugate of the one considered. We have now identified
the singlet component of the gap, ∆k , and the triplet components, d(k). Since
 
−dx (k) + idy (k) ∆k + dz (k)
ˆ
∆(k) = , (12.60)
−∆k + dz (k) dx (k) + idy (k)

147
we can write the gap matrix in a compact form as
ˆ
∆(k) = (∆k 1 + d(k) · σ) iσ y , (12.61)
where σ is the vector of Pauli matrices. The appearance of the factor
 
0 1
iσ y = (12.62)
−1 0
should not be misinterpreted as showing that the y-component of the Cooper-pair spin is in some way singled out.
Rather, the factor stems from the anticommutativity of fermionic operators and is only expressed in terms of the
Pauli matrix σ y for convenience. It can also be understood as the unitary part of the antiunitary time-reversal
operator.
The mean-field Hamiltonian HMF is diagonalized by a Bogoliubov transformation and the gap function ∆(k) ˆ
is obtained selfconsistently from a gap equation in complete analogy to the singlet case discussed in Sec. 10.1,
except that ∆(k)
ˆ is now a matrix and that we require four coefficients uk↑ , uk↓ , vk↑ , vk↓ . We do not show this
here explicitly.
A few remarks on the physics are in order, though. The symmetry ∆(−k)
ˆ = −∆ ˆ T (k) implies

(∆−k 1 + d(−k) · σ) iσ y = −i (σ y )T ∆k 1 + d(k) · σ T (12.63)

| {z }
iσ y
∆−k 1 + d(−k) · σ = σ y
∆k 1 + d(k) · σ T σ y


⇒
 y x T y 
σ (σ ) σ
= ∆k 1 + d(k) ·  σ y (σ y )T σ y 
σ y (σ z )T σ y
 y x y 
σ σ σ
= ∆k 1 + d(k) ·  −σ y σ y σ y 
σy σz σy
= ∆k 1 − d(k) · σ. (12.64)
Thus we conclude that ∆k is even,
∆−k = ∆k , (12.65)
whereas d(k) is odd,
d(−k) = −d(k). (12.66)
Hence, the d -vector can never be constant, unlike the singlet gap in Sec. 10.1. Furthermore, we can expand ∆k
into even basis functions of irreducible representations of the point group of the system (which contains rotations,
inversion, and mirror reflections that leave the crystal structure invariant), e.g.,
∆k = ∆s ψs (k) + ∆dx2 −y2 ψdx2 −y2 (k) + ∆d3z2 −r2 ψd3z2 −r2 (k) + ∆dxy ψdxy (k) + ∆dyz ψdyz (k) + ∆dzx ψdzx (k) + . . . ,
(12.67)
and expand (the components of) d(k) into odd basis functions,
d(k) = dpx ψpx (k) + dpy ψpy (k) + dpz ψpz (k) + . . . (12.68)
For the singlet case, we have already considered the basis functions ψs (k) = 1 for conventional superconductors,
ψs (k) = cos kx a cos ky a for the pnictides, and ψdx2 −y2 (k) = cos kx a − cos ky a for the cuprates. Typical basis
functions for a triplet superconductor would be ψpi = sin ki a, i = x, y, z. He-3 in the so-called B phase realized
at not too high pressures and temperatures (see the phase diagram in Sec. 2.2) has the d -vector
(12.69)
 
d(k) = ∆1 x̂ ψpx (k) + ŷ ψpy (k) + ẑ ψpz (k) ,
where the basis functions are that standard expressions for px , py , and pz orbitals (note that there is no Brillouin
zone since He-3 is a liquid),
r
3
ψpx (k) = sin θk cos φk , (12.70)
4π

148
 r
3
ψpy (k) = sin θk sin φk , (12.71)
4π
r
3
ψpz (k) = cos θk (12.72)
4π
so that r
3
d(k) = ∆1 k̂. (12.73)
4π
Hence, He-3 shows p-wave spin-triplet pairing. This is the so-called Balian-Werthamer state. Note that it has
the property that r
3
|d(k)| = |∆1 | = const (12.74)
4π
so that ∆(k)
ˆ is nonzero everywhere. What is important here is that the superconducting gap is nonzero everywhere
on the normal-state Fermi surface. The Balian-Werthamer state does not have nodes.
The A phase covers a smaller part of the phase diagram. Here, the d-vector reads

(12.75)
 
d(k) = ∆1 d̂ ψpx (k) + iψpy (k) ,

where d̂ is a fixed unit vector. By symmetry, all states obtained by rotating d̂ or the “orbitals” px and py are
degenerate. This is called the Anderson-Brinkman-Morel state. Unlike the Balian-Werthamer state, it has point
nodes. For the d vector above, these appear at kx = ky = 0 since then ψpx (k) = ψpy (k) = 0.
There are not many pure spin-triplet superconductors as opposed to the superfluid He-3. Triplet pairing is
favored for the superconductor UPt3 . As discussed above, fermionic anticommutation implies that the order
parameter must have the form
ˆ
∆(k) = d(k) · σ iσ y (12.76)
with an odd function d(k). The correct way to classify pairing states uses irreducible representations of the point
group of the crystal. We will not explain this type of analysis but only note that the favored d-vectors belong
to the irreducible representation E2u of the point group D6h of UPt3 [see R. Joynt and L. Taillefer, Rev. Mod.
Phys. 74, 235 (2002)]. They are thought to be linear combinations of two basis vectors d1 (k) and d2 (k), which
under point-group operations transform like

d1 (k) = (kx2 − ky2 )kz ẑ, (12.77)

d2 (k) = kx ky kz ẑ. (12.78)

These basis functions are of third order in momentum, meaning that they describe a form of f-wave pairing.
It should be noted that d1 (k) and d2 (k) cannot be correct d -vectors for all momenta since they violate the
periodicity of momentum space. It is, however, common practice to write down basis functions obtained by
group-theoretical analysis in terms of lowest-order polynomials.

12.5 Noncentrosymmetric superconductors

In crystals lacking inversion symmetry, i.e., noncentrosymmetric crystals, a term of the form
X †
HASOC = λ ck g(k) · σ ck , (12.79)
k

with the spinor operator  

ck↑
ck = (12.80)
ck↓
and an odd function g(k) is allowed by symmetry. This terms represents antisymmetric spin-orbit coupling
(ASOC). It is forbidden in the presence of inversion symmetry since under inversion g(k) (being odd in k)

149
changes sign but the spin σ (being a pseudovector) does not. The ASOC splits the bands for the normal state,
except at high-symmetry points with g(k) = 0.
Due to the ASOC, spin is not conserved. This means that also the spin of a Cooper pair is not a good quantum
number. We can still write the pairing matrix as
ˆ
∆(k) = (∆k 1 + d(k) · σ) iσ y (12.81)

with even ∆k and odd d(k) since this is the most general 2 × 2 matrix satisfying the condition ∆(−k)ˆ = −∆ˆ T (k)
imposed by fermionic anticommutation. But since the spin is not conserved, generically both the singlet amplitude
∆k and the triplet amplitudes dx (k), dy (k), dz (k) are nonzero. In other words, noncentrosymmetric crystals show
a superposition of singlet and triplet pairs, called singlet-triplet mixing. It was shown by Frigeri et al. [Phys. Rev.
Lett. 92, 097001 (2004)] that pairing states with d(k) k g(k) are energetically favorable. This is plausible
since the Cooper pairs are then “compatible” with the ASOC in the normal-state Hamiltonian. Examples for
noncentrosymmetric superconductors are YPtBi, CePt3 Si, CeRhSi3 , and Y2 C3 .
As an example, let us consider a cubic crystal with point group Td , which does not include inversion symmetry.
Then, the lowest-order polynomial form for the g-vector reads

g(k) = kx (ky2 − kz2 ) x̂ + ky (kz2 − kx2 ) ŷ + kz (kx2 − ky2 ) ẑ. (12.82)

The d -vector should be parallel to g(k). On the other hand, the singlet amplitude ∆k is not constrained by
symmetry, beyond being even. This leads to the simplest ansatz
ˆ = ∆s 1 + ∆t kx (ky2 − kz2 ) σx + ky (kz2 − kx2 ) σy + kz (kx2 − ky2 ) σz iσ y , (12.83)
 

∆(k)

with a singlet amplitude ∆s and a triplet amplitude ∆t (with some dimensionfull factors absorbed). The eigen-
values of the factor in large parentheses are
q 2 2
2
(12.84)

∆s ± ∆t kx (ky2 − kz2 ) + [ky (kz2 − kx2 )] + kz (kx2 − ky2 ) .

Evidently, one of the eigenvalues crosses zero if |∆t /∆s | is sufficiently large. This happens on a surface in
momentum space since there is one constraint on (kx , ky , kz ). This surface generically intersects with the normal-
state Fermi surface along a line. Hence, we expect line nodes if the triplet component is sufficiently large. Recall
that for the cuprates the line nodes occur in the high-symmetry (mirror) planes ky = ±kx . Here, the situation is
very different: the vanishing eigenvalue and thus the line nodes are not tight to high-symmetry planes.

150