Superconductivity

12 Superconductivity in metals
In this chapter we focus on the phenomenon of superconductivity and the Bardeen–Cooper–

Schrieffer (BCS) (BCS1957) theory behind it. Superconductivity obtains when a finite frac-
tion of the conduction electrons in a metal condense into a quantum state characterized by
a unique quantum-mechanical phase. The specific value of the quantum-mechanical phase
varies from one superconductor to another. The locking in of the phase of a number of elec-
trons on the order of Avogadro’s number ensures the rigidity of the superconducting state.
For example, electrons in the condensate find it impossible to move individually. Rather, the
whole condensate moves from one end of the sample to the other as a single unit. Likewise,
electron scattering events that tend to destroy the condensate must disrupt the phase of
a macroscopic number of electrons for the superconducting state to be destroyed. Hence,
phase rigidity implies collective motion as well as collective destruction of a superconduct-
ing condensate. The only other physical phenomenon that arises from a similar condensation
of a macroscopic number of particles into a phase-locked state is that of Bose–Einstein con-
densation. There is a crucial difference between these effects, however. The particles that
constitute the condensate in superconductivity are Cooper pairs, which do not obey Bose
statistics. In fact, it is the Pauli principle acting on the electrons comprising a Cooper pair that
prevents the complete mapping of the superconducting problem onto a simple one of Bose
condensation. As we will see, it is the Pauli principle that makes BCS theory work so well.
What do we mean by this? In BCS theory, it is assumed that electrons form Cooper pairs, and
the pairs are strongly overlapping. Such a strong overlap would imply a strong correlation
between pairs. In fact, it is the correlations between pairs that accounts for most of the ob-
served properties of superconductors, for example the energy gap and the Meissner effect. In
BCS theory, however, there is no explicit dynamical interaction between Cooper pairs. The
only interaction, if it can be thought of in these terms, is that arising from the Pauli exclusion
principle which precludes two Cooper pairs from occupying the same momentum state. That
BCS theory works so well speaks volumes for the real nature of pair–pair correlations in met-
als. It would suggest that real pair–pair interactions in a metal arise primarily from the Pauli
exclusion principle, rather than from some additional dynamical interaction. It is primarily
for this reason that the simple pairing hypothesis of BCS has had such profound success.
12.1 Superconductivity: phenomenology
At the outset, we lay plain the experimental facts that any theory of superconductivity must
explain.
189
Downloaded from https:/www.cambridge.org/core. La Trobe University, on 06 Apr 2017 at 16:32:52, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139031066.013
B(z)
B0
vacuum superconductor
z
Fig. 12.1 Fall-off of the magnetic field in a Type I superconductor.
H c (T )
H c (0)
S N
Al Hg Pb
T(K)
1.2 4.15 7.19
Fig. 12.2 The dependence of the critical field as a function of temperature. The temperatures indicated on the horizontal axis
represent the superconducting transition temperatures for a series of metals.
(a) Zero resistance The typical signature of superconductivity is the vanishing of the
electrical resistance below some critical temperature Tc . The superconducting state is a
thermodynamically distinct state of matter. Below Tc , a current flows without any loss.
Until the high-Tc materials were made, Nb held the highest transition temperature at
9.26 K.
(b) Meissner effect Another feature is the exclusion of magnetic fields, the Meissner effect.
Materials in which the Meissner effect is complete are known as Type I superconductors.
Consequently, the interior of a Type I superconductor is a perfect diamagnet. A magnetic
field applied at the boundary of a Type I superconductor falls off exponentially with
distance in the interior of the material, as illustrated in Fig. 12.1. The penetration depth,
λL , is defined as the distance over which the magnetic field decreases by the factor 1/e.
Below Tc , the field needed to destroy superconductivity increases to some critical value
Hc (T ), as illustrated in Fig. 12.2. Because the magnetic field inside a superconductor
191 12.1 Superconductivity: phenomenology
I II
-4 M -4 M
N N
S S
H H
Hc H c1 Hc H c2
(a) (b)
Fig. 12.3 (a) Magnetization vs applied field for a Type I superconductor. (b) Magnetization vs applied field for a Type II
superconductor. Between Hc1 and Hc2 , magnetic field lines penetrate the superconductor but they do not destroy
superconductivity. The field lines form a vortex lattice.
is zero,
B = 0 = Happl + 4π M, (12.1)
where Happl is the applied field and M the magnetization. Solving this equation, we find
that the magnetization is
1
M =− Happl . (12.2)
4π
The negative value of M signals that the interior of a superconductor is diamagnetic. At

any temperature less than Tc , the magnetization should be a linear function of the applied
field. A material having a magnetization of this form is called a Type I superconductor.
The normal state is indicated with an N and the superconducting state with an S. Above
Hc , B = 0 and the magnetization no longer obeys Eq. (12.2).
In some materials, superconductivity is observed up to an upper critical field Hc2 , but
an incomplete Meissner effect is seen between a lower critical field Hc1 and Hc2 . The
resultant magnetization is shown in Fig. 12.3(b). Materials exhibiting a magnetization
of this kind are known as Type II superconductors. Between Hc1 and Hc2 , the magnetic
field penetrates the material but superconductivity is not destroyed. The field lines form
a regular array known as the Abrikosov (A1957) vortex lattice. All high-Tc cuprate
superconductors are Type II.
We are concerned primarily with Type I materials. To understand the penetration
depth in a superconductor, we resort to the London equations. First, we need the
Maxwell equation for the curl of an electric field: −∂B/∂t = c∇ ×E = cρ∇ ×J, where
ρ is the resistivity. In a perfect conductor ρ = 0 and, as a consequence, ∂B/∂t = 0.
In a superconductor, ρ = 0 as well. However, it is an experimental fact that B = 0
inside a superconductor. This result cannot be deduced from the Maxwell equations.
It is the Meissner effect that sets superconductivity apart from materials that just
display perfect conductivity. Inside a superconductor, expulsion of magnetic flux is
mediated by the current that flows. London proposed in 1935 that everywhere in a
superconductor
J = −const.A, (12.3)
where A is the vector potential and B = ∇ × A. From this ansatz, London was able to
show that a magnetic field decays exponentially inside a superconductor. Dimensionally,
the constant has units of 1/(L · time). Let us write the constant as
c
const. = , (12.4)
4π λ2L
where c is the speed of light. If we take the curl of both sides of Eq. (12.3), we find that
c
∇ ×J=− B. (12.5)
4π λ2L
From Ampère’s law,
4π
∇ ×B= j, (12.6)
c
we find that
4π
∇ × (∇ × B) = ∇ × J, (12.7)
c
which implies that
1
∇ 2B = B· (12.8)
λ2L
The solution to this equation,
B(r) = B(0)e−r/λL , (12.9)
is an exponentially decaying magnetic induction on a length scale λL . Exponential

decay of the magnetic field into a superconductor is the Meissner effect. Let vs , m∗ ,
and e∗ , respectively, be the velocity, mass, and charge of the current carriers in a
superconductor. Then
m∗ v̇s = −e∗ E. (12.10)
The current of these electrons is defined as J = −e∗ vs ns , which, combined with

Eq. (12.10), yields
∂J ns e∗2
= E (12.11)
∂t m∗
for the time evolution of the current. Taking the curl of both sides, we find that
∂ ns e2
0 = (∇ × J) − ∇ ×E
∂t m∗

∂ ns e∗2
= ∇ ×J+ B . (12.12)
∂t m∗ c
Comparing Eq. (12.12) with Eq. (12.5), we obtain that the penetration depth is
1/2
m∗ c2
λL = . (12.13)
4π ns e∗2
We see then that as the superconducting density increases, λL decreases.

We can justify the main assumption in the London approach by appealing to the theory
of Ginsburg and Landau (GL1950). The crucial ingredient in this phenomenological
theory is that the difference in the free energy density between the superconducting
and normal states can be written as a functional of an order parameter, ψ (r), for the
superconducting state. Physically, |ψ (r)|2 is proportional to the charge density in the
superconducting state, ns . Consequently, we can interpret ψ (r) as the wavefunction
of the superconducting state. In BCS theory, ψ (r) plays the role of the center-of-
mass wavefunction for a Cooper pair. Near Tc , the superfluid density is small; hence
|ψ|2 ne . Consequently, Ginsburg and Landau expanded the free energy density for
the superconducting state in the vicinity of Tc as a power series in |ψ|2 ,
2

F = FN + dr |∇ψ| 2
+ a(T )|ψ (r)| 2
+ b(T )|ψ (r)| 4
, (12.14)
2m∗
retaining the kinetic energy term, |∇ψ|2 , to account explicitly for spatial variations of
the field ψ (r). The free energy of the normal state is FN . The coefficients a(T ) and
b(T ) are real and temperature-dependent and, for stability, b(T ) > 0.
To find the ground state of the system, we minimize the free energy density with
respect to ψ ∗ (r):
2 2
− ∇ ψ (r) + a(T )ψ (r) + 2b(T )|ψ (r)|2 ψ (r) = 0. (12.15)
2m∗
Because the free energy density contains the term |∇ψ|2 , which is always positive, the
free energy is minimized by demanding that ∇ψ (r) = 0 or, equivalently, that ψ be uni-
form in space. Consequently, the solution to the saddle point equation is either ψ = 0 or
a(T )
|ψ0 |2 = − = ns , (12.16)
2b(T )
which implies that a(T ) < 0. Should a(T ) exceed zero, ψ = 0, and the system would
be in the normal state. Since superconductivity vanishes at Tc , we must have that
a(Tc ) = 0. A Taylor expansion of a(T ) around Tc to first order leads to the result that
a(T ) = a1 (T − Tc ) (12.17)
with a1 > 0.
We obtain the London conjecture by recalling that the current density in the presence
of a vector potential, A, is
e∗ ∗ ∗ e∗2
J= (ψ ∇ψ − ψ∇ψ ) − |ψ|2 A. (12.18)
2im∗ m∗ c
If we assume that the magnetic field is sufficiently small that the equilibrium value of
ψ0 is unchanged, then substitution of ψ0 into Eq. (12.18) yields the London result
e∗2 ns
J=− A. (12.19)
m∗ c
Using Eq. (12.13), we find that this result is consistent with Eq. (12.4). Hence, from
this simple phenomenological approach, we are able to justify the London ansatz for
the current density. Indeed, as we will see, the key intellectual content of the BCS
theory of superconductivity is the existence of an order parameter describing a charge
2e condensate with a well-defined phase, as in the Ginzburg–Landau theory. Such a
condensate breaks the U (1) gauge symmetry, as discussed in Chapter 1. In fact, the very
existence of the Meissner effect implies the breaking of a continuous symmetry. The
fact that a magnetic field cannot penetrate a superconductor tells us immediately that in
a superconductor, the photon is massive. Consequently, it can no longer be assumed that
the electrons are the propagating degrees of freedom. In fact, in a superconductor they
are not. The charge is quantized in units of 2e, implying that U (1) symmetry is broken.
(c) Heat capacity In the superconducting state, the entropy decreases continuously but
dramatically, signaling the formation of a highly ordered state. This is depicted in
Fig. 12.4(a). As a result, the temperature derivative of the entropy must be steeper on
the superconducting side than on the normal side of the transition. Consequently, the
heat capacity is discontinuous at Tc , and superconductivity is a second-order phase
transition. As shown in Fig. 12.4(b), in the superconducting state, the heat capacity,
cs , falls off as cs ∝ exp − /kB T , where is an energy scale. A heat capacity of this
form is indicative of an energy gap in the excitation spectrum, with p > μ. Let us
verify this with the simple calculation:

∂ p dp T →0 ∂
cV = −→ p e−β(p −μ) dp . (12.20)
∂T (eβ(p −μ) + 1) ∂T
If (p − μ) ≈ , then cV ∼ exp(− (T = 0)/kB T ). Consequently, in the super-
conducting state, we adopt the picture for the energy gap shown in Fig. 12.5.
The formation of a gap at the Fermi level in the superconducting state results in
a lowering of the ground state energy of the system. The gap is actually 2 , not .
Hence, p − μ accounts for only half the gap. Experimentally, the gap can be measured
by tunneling or ultrasound attenuation experiments. Thermodynamically, the gap gives
rise to a discontinuity in the heat capacity. That is, cs (Tc− ) − cN (Tc+ ) = 0. Across
S Cv
N N
S S
Tc Tc
T T
(a) (b)
Fig. 12.4 (a) Behavior of the entropy across the superconducting transition. (b) Behavior of the heat capacity in the normal and
superconducting states.
F F 2Δ
(a) (b)
Fig. 12.5 (a) Filled energy levels in the normal state. (b) Formation of an energy gap in the superconducting state. The full gap is
2 , not .
the superconducting transition, both first derivatives of the free energy vanish. Hence,
no latent heat is associated with the superconducting transition. Above Tc , = 0,
and at T = 0, has its largest value. A typical plot of (T )/ (T = 0) is shown
in Fig. 12.6. A weak-coupling superconductor has a ratio of 2 (T = 0)/kB Tc in the
range 1 to 3. Strong coupling corresponds to 2 (T = 0)/kB Tc > 4. The basic energy
scale for the creation of an electron–hole pair in a superconductor is 2 .
(d) Microwave and infrared properties As a result of the gap, photons possessing
energies less than 2 are not absorbed: all such photons are reflected. Perfect reflection
occurs for ω < 2 (T = 0)/. When this condition is true, photons see a completely
resistanceless surface. As ω > 2 (T = 0)/ at absolute zero, the resistance begins to
approach that of the normal state. We estimate this energy by assuming a weak-coupling
description is valid for the superconductor. Then ∼ 2kB Tc , and ω ∼ 4kB Tc /. For
a Tc of 5 K, ω ∼ 1012 s−1 . This frequency is in the infrared. Infrared radiation can then
penetrate a superconductor and scatter the electrons.
(T )
0 Tc
T
Fig. 12.6 The behavior of the superconducting gap, , as T → Tc .
–1
T1
T
Tc
Fig. 12.7 Behavior of the spin-lattice relaxation time at and below Tc . The enhancement in 1/T1 is a signature that the spins in a
superconductor are acting in consort.
(e) Ultrasonic attenuation No damping of an impinging beam of phonons is observed if

ωq < 2 /. As in the microwave absorption case, ωq must exceed the energy needed
to create an electron–hole pair.
(f) Nuclear-spin relaxation Consider a set of nuclei that have been forced to align with
a magnetic field. The rate at which the equilibrium magnetization is recovered is the
spin lattice relaxation rate, 1/T1 . In a superconductor, (1/T1 )S > (1/T1 )N just below
Tc . That is, there is an enhancement in the relaxation rate that is brought on by the
formation of the superconducting state. Hebel and Slichter (HS1959) were the first to
see this effect experimentally. The peak in the relaxation rate just below Tc is known
as the Hebel–Slichter peak.
(g) Isotope effect Experimentally, it is observed that if the mass of the ions is changed
isotopically, Tc changes accordingly:
1
Tc ∝ √ ∝ ωD . (12.21)
M
197 12.2 Electron–phonon effective interaction
This observation implies that electron–phonon coupling is at the heart of supercon-

ductivity (at least traditional superconductivity).
12.2 Electron–phonon effective interaction
The isotope dependence of Tc confirms that electron–phonon coupling is central to the

mechanism of superconductivity. We show now how electron–phonon coupling can produce
an attractive interaction between two electrons. The starting point for our analysis is an
ordered band of electrons with kinetic energies k coupled to the quantized vibrational
modes of the crystal. As a result, our Hamiltonian H = Ho + He-ph is a sum of an ordered
part,

Ho = ωq b†q bq + k a†k ak , (12.22)
q k
and the electron–lattice interaction

He-ph = Mq a†k+q ak bq + b†−q . (12.23)
k,q
The electron–phonon coupling constant is defined in Section 11.3. We will assume

that the electron–phonon interaction is weak and do perturbation theory to second or-
der in this interaction. As a second-order perturbation, we anticipate that the electron–
phonon interaction will provide a negative correction to the energy. While the final result
can be established quite straightforwardly (see Problem 12.3) using second-order per-
turbation theory, we will use a similarity transformation, as this method has numerous
applications in electron–phonon problems. Our analysis will mirror the Schrieffer–Wolff
transformation. Eliminating the linear electron–phonon interaction will produce a second-
order interaction, from which we will be able to deduce the conditions under which phonons
effectively bind electrons.
We seek a transformation S, such that
[Ho , S] = −He-ph . (12.24)
The transformed Hamiltonian will be
H̃ = e−S H eS
1
= Ho + [He-ph , S] + · · · . (12.25)
2
We posit that S is of the form

S= Ak,q b†−q + Bk,q bq Mq a†k+q ak , (12.26)
k,q
where A and B are to be determined by the constraint in Eq. (12.24). To proceed, we need
the commutators

nk , a†k1 ak2 = a†k1 ak2 (δk1 k − δk2 k ),

b†q , bq = −1,

b†q bq , bq = −bq δqq . (12.27)
We find then that

[Ho , S] = ωq b†q bq , Ak,q b†−q + Bk,q bq a†k+q ak Mq
q ,k,q

+ k Mq a†k ak , Ak,q b†−q + Bk,q bq a†k+q ak
k ,k,q

= Mq b†−q a†k+q ak ω−q + k+q − k Ak,q
k,q

+ Mq bq a†k+q ak k+q − k − ωq Bk,q . (12.28)
k,q
This quantity must equal −He-ph . The Ak,q and Bk,q coefficients that satisfy this constraint
are
−1
Ak,q = − ω−q + k+q − k (12.29)
and
−1
Bk,q = − k+q − k − ωq . (12.30)
Our unitary transformation is then

b†−q bq
S= + Mq a†k+q ak . (12.31)
k,q
k − k+q − ω−q k − k+q + ωq
Note the form of the energy denominators. The denominator of b†−q corresponds to the
energy difference for emission of a phonon of energy ω−q and the denominator of bq to
absorption. Let us define ± (k, q) = k − k+q ± ω±q . The effective Hamiltonian is
given by

1 b†−q bq
† † †
H̃ = Ho + Mq Mq b−q + bq ak+q ak , + a ak .
2 k,q,k ,q − (k , q ) + (k , q ) k +q
(12.32)
We need to evaluate a commutator of the form

b†−q + bq a†k+q ak , b†−q a†k +q ak . (12.33)
Specifically, we are interested in the two-electron terms that are produced from this commu-
tator. If we were to evaluate the electron commutator part, we would obtain only a one-body
potential. Such one-body potentials are not of interest here as they do not affect the inter-
action between two electrons. The commutator involving the phonon operators produces
199 12.3 Model interaction
k′−q k+q
k′ k
Fig. 12.8 Electron–phonon interaction giving rise to BCS pairing near the Fermi surface. The wavy line represents the
interaction in Eq. (12.35).
an effective two-body potential that, interestingly enough, is negative in a narrow energy

scale around the Fermi energy. To see this, we note that the phonon part of the commutator
in Eq. (12.33) yields the constraint δq+q . Consequently, our effective Hamiltonian is
†
1 ak+q ak a†k −q ak a†k+q ak a†k −q ak
H̃ = Ho + Mq M−q −
2 k,k ,q − (k , −q) + (k , −q)
2 † ωq
= Ho + Mq a †
k+q ak −q ak ak 2 (12.34)
k,k ,q k ,q − (ωq )2
with k ,q = k − k −q . In general, k and k −q vastly exceed ωq . However, if

k − k−q < ωq , then the two-electron potential in Eq. (12.34) is negative, and the
electrons experience a net attraction. Note that we have shown that a net attraction exists in
k-space, not real-space. The k-space attraction allows two electrons to bind together so that
they have a total energy that is lower than their non-interacting counterparts. This attraction
is the basis for pairing in BCS superconductivity. Diagrammatically, this interaction is
depicted in Fig. 12.8, where the wavy line represents the electron–phonon interaction,
2
Mq ωq

V (k , q) = . (12.35)
(k − k −q )2 − (ωq )2
12.3 Model interaction
In a metal, of course, we must consider the electron–electron repulsion. We can represent

this potential as
Ve (q)
Veeo = , (12.36)
ε(q, ω)
with ε(q, ω) the dynamic dielectric screening function. There is of course another source
of electron–electron interaction, the attraction induced by phonons. The full interaction
Vee (t)
0
ωq t
Fig. 12.9 Fourier transform of the full electron interaction including the contribution from phonons. The delta-function
contribution from the electron repulsions has been smeared slightly in the vicinity of t = 0.
between electrons is of the form

2
Ve (q) ωq Mq
Vee = + 2 , (12.37)
ε(q) k,q − (ωq )2
where we have taken the static limit of the screening function. Now ε(q) > 0, unless the
static compressibility is negative, as discussed in Chapter 9. We exclude this situation here
as a negative compressibility occurs in the dilute regime where non-perturbative treatments
of the Coulomb interaction are required. Typically, Ve (q) is much greater than the phonon
part of the potential. However, when k,q ωq , the phonon part dominates, and a net
attraction is introduced. We see then that on different energy scales, either the phonon or the
electron part can dominate. The largest phonon frequency that is relevant is ωD , the Debye
energy. Hence, for k,q ≤ ωD , a net attraction is produced that dominates the electron
repulsion. When both electrons have an energy close to the Fermi energy, then k,q ≈ 0,
and an attractive interaction obtains. This is an essential feature of the BCS theory.
Before we leave this section, it is instructive to consider the time Fourier transform of
the interaction potential, Vee . Let ω = k,q . The Fourier transform of Vee is
∞
dω −iωt ωq |Mq |2
Vee (t ) = e Veeo +
−∞ 2π (ω)2 − (ωq )2
∞
ωq e−iωt
= Veeo δ(t ) + |Mq |2 dω
2π −∞ (ω − ωq )(ω + ωq )
= Veeo δ(t ) − |Mq |2 sin ωqt. (12.38)
The repulsive part acts only at t = 0, while the attraction is an oscillatory function of time.
The total potential is shown in Fig. 12.9. We find that the electron potential is attractive
only over a time interval where the phonon attraction dominates. This is a key point; Vee
201 12.4 Cooper pairs
in the time domain is not always attractive. Electron repulsions and the phonon-mediated
attraction act on different time scales and hence pair-binding of electrons is possible.
Let us evaluate the two-body scattering amplitude,
As = p4σ1 p3σ2 |Vee |p1σ1 p2σ2 , (12.39)
where

Vee = V (k, k − q)a†k+q a†k −q ak ak . (12.40)
k,k ,q
As discussed in Chapter 5, any two-body amplitude of this form separates into a difference
of direct and exchange Coulomb integrals. For a general momentum-dependent potential,
the direct and exchange terms will enter with different combinations of the momenta and
spin. For example, the exchange term will be of the form V (p1 , p3 ) with σ1 = σ2 , whereas
the direct term will enter with no restriction on the spins and will depend on V (p1 , p4 ). Let
us assume that the net attractive potential is a constant of the form
$
−Vo , kF ,q < ωD ,
Vee = (12.41)
0, k ,q > ωD .
F
Because the potential is momentum-independent, the direct and exchange integrals are
equal, and the scattering amplitude vanishes when σ1 = σ2 . The only non-zero contribution
arises from σ1 = σ2 . As a consequence, for a constant interaction, the scattering amplitude
is non-zero only if the electrons are locked into a singlet state. In this case, phonons induce
a net attraction between electrons of opposite spin. The corresponding matrix element is of
the form
p4↑ p3↓ |Vee |p1↑ p2↓ = −V0 δp1 +p2 ,p3 +p4 (12.42)
for scattering in the vicinity of the Fermi surface. Two particles locked into a singlet state
will give rise to a gap at the Fermi level.
12.4 Cooper pairs
Consider now the somewhat artificial problem of a full Fermi sea containing N non-
interacting electrons with two additional interacting electrons outside the sea. As a result
of the Pauli exclusion principle, the momentum of the electrons outside the Fermi sea must
exceed pF . We take the potential of interaction to be the constant singlet pair potential
derived in the previous section, Eq. (12.42). The spin wavefunction is hence antisymmetric
with respect to spin. The corresponding spatial part must be symmetric to satisfy the overall
antisymmetry requirement. The eigenvalue equation for our subsystem of two particles is

2 2
− ∇ + ∇2 + V (r1 − r2 ) − E (r1 , r2 ; ↑1 , ↓2 ) = 0.
2
(12.43)
2m 1
In general, the two-body wavefunction is a composite state
(r1 , r2 ; ↑1 , ↓2 ) = ψ (r1 , r2 )χS (↑1 , ↓2 ), (12.44)
where χS (↑1 , ↓2 ) represents the singlet spin state and ψ contains the spatial dependence.
For two particles, we can express the spatial wavefunction in terms of the relative coordinate,
r = r1 − r2 , and a center of mass, R = (r1 + r2 )/2, as
ψ (r1 , r2 ) = ϕ(r)eiQ·R/ . (12.45)
Likewise, the momenta of interest are the center of mass, Q = p1 + p2 , and the relative
momentum, q = (p1 − p2 )/2. We expand ϕ(r) in a Fourier series as

eik·r/
ϕ(r) = √ αk , (12.46)
k
V
in which the prime indicates that the restriction k > kF is restricted to all states whose
energy exceeds F . Because k · r = k · r1 − k · r2 , we see that the pair state has momenta
(k, −k). Note, if k1 + k2 = 0, then the center-of-mass motion drops out of the problem.
We focus first on the Q = 0 solution. To this end, we define the Fourier components of
the interaction potential,

dr −i(k−k )·r/
Vkk = e V (r), (12.47)
V
and introduce the center-of-mass Schrödinger equation

2
− ∇ 2 + V (r) ϕ(r) = Eϕ(r), (12.48)
2μ
where μ is the reduced mass, μ = m/2. Substituting the Fourier representation of ϕ(r),
multiplying by exp(−ik · r/), and integrating, we obtain

(E − 2k ) αk = Vkk αk (12.49)
k
as our new eigenvalue equation. Here, k = k 2 /2m. The matrix element Vkk is equivalent
to
Vkk = k, −k |Vee | k , −k . (12.50)
A typical scattering process in Vkk is shown in Fig. 12.10.
If we now introduce the approximation that

−V0 , k, k > pF ,
Vkk = (12.51)
0, otherwise,
the eigenvalue equation can be recast as
1
1 = −V0
E − 2k
k>kF
= −V0 φ(E ). (12.52)
This equation is satisfied as long as φ(E ) = −1/V0 . The poles of φ(E ) occur at E = 2k ,
the total energy of the pair, which is bounded from below by 2F . In a finite system, k
takes on discrete values because k is quantized. As E approaches 2k from below, φ(E )
approaches −∞. Just above 2k , φ(E ) is ∼ +∞. For all E < 2F , φ(E ) is negative. Hence,
203 12.4 Cooper pairs
−k′ k
−k k′
Fig. 12.10 Scattering between a pair of electron states across the Fermi surface.
φ(E)
2 F 2k
−1
V0
Fig. 12.11 A plot of φ(E ) versus k in the Cooper-pair problem. The intersection of φ(E ) with the straight line −V0−1
determines the bound-state solutions for the pair.
a bound state forms when φ(E ) crosses −1/V0 for E < 2F . The intersection of −1/V0
with φ(E ) is illustrated graphically in Fig. 12.11. The existence of such a solution is the
Cooper (C1956) pair problem. To find the precise energy of the bound state, we convert the
sum in Eq. (12.52) to an integral,

F +ωD
1 N (x)dx
1 = V0 → V0 (12.53)
(k)
2k − E F 2x − E
by introducing the density of states, N (x). If we assume that N (x) does not change sig-
nificantly in the narrow range of integration, we can set N (x) ≈ N (F ). Under these
assumptions, the integral yields

2 F − E2
− = ln , (12.54)
V0 N (F ) F − E2 + ωD
which implies that

E 2 E
F − + ωD exp − = F − . (12.55)
2 V0 N (F ) 2
This linear equation can be solved immediately for the eigenenergy E:
2ωD exp (−2/V0 N (F ))
E = 2F − . (12.56)
1 − exp (−2/V0 N (F ))
In the limit that 2 V0 N (F ), the exponential in the denominator can be expanded. The
pair-binding energy

2
E 2F − 2ωD exp − (12.57)
V0 N (F )
in the weak-coupling limit results. Either of these expressions indicates that the Cooper pair
is bound with an energy < 2F whenever V0 is non-zero and positive. This is a profound
result. It implies that two electrons directly below the Fermi surface can lower their energy
by being excited into a Cooper pair with momentum (k, −k) just above the Fermi surface
provided that an attractive interaction of the form in Eq. (12.51) exists. This is known as
the Cooper instability.
Further, we can estimate how the pair-binding energy depends on the center of mass
of the Cooper pair. At the onset, we suspect that this quantity might scale as Q2 . We will
show that this is not the case. To proceed, we extend the pair-binding criterion to the case
in which Q = 0. For non-zero Q, the bare energy of the pair becomes q+Q/2 + −q+Q/2 .
Consequently, the pair-binding condition becomes
1
1 = −V0 . (12.58)
q
E − q+Q/2 − −q+Q/2
For small Q, we can rewrite Eq. (12.58) as
F +vF Q/2+ωD
N (q )dq
1 = −V0 , (12.59)
F +vF Q/2 E − 2q
dropping terms of O(Q2 ). The center-of-mass simply shifts the zero of the Fermi energy.
The new pair-binding energy,
2ωD
E = 2F + QvF − , (12.60)
exp (2/V0 N (F )) − 1
205 12.5 Fermi liquid theory
is a linear function of the center-of-mass momentum. Translation of the center of mass

strongly reduces the binding energy and could eventually break up the pair. To show this,
we set F = 0 and evaluate the value of Q at which the Cooper pair loses most of its binding
energy. We must then solve
2ωD
QvF =
exp (2/V0 N (F )) − 1
≈ kB Tc . (12.61)
Equivalently, Q/ ∼ kB Tc /vF ∼ 104 cm−1 , which is roughly the reciprocal of the Pippard
coherence length, ξ ∼ 10−4 cm. This is the effective radius of gyration of a Cooper pair, an
enormous distance when compared to interatomic spacings. Such a large coherence length
is a typical feature of phonon pairing mechanisms.
12.5 Fermi liquid theory
Of course, our problem is somewhat artificial in that we have ignored all interactions between
the electrons save for the pair just above the Fermi surface. It is certainly reasonable to
expect the simple picture of the Cooper instability to break down once we consider repulsive
interactions among all of the electrons. That is, when electrons are interacting, we cannot
a priori regard the non-interacting eigenstates as a valid description of our system (as we
have done in the Cooper problem). We then are led to the question, is the instability real?
We will answer this question by appealing to simple physical considerations arising from
the scattering of electrons near the Fermi surface and a more formal argument involving
the scaling of the full interacting Lagrangian. It is now well accepted that the normal state
of a metal is described by Landau–Fermi liquid theory. In this account, it is claimed that the
dominant effect of electron interactions in a metal is to renormalize the effective mass of the
electron. The observed shift is on the order of 10 to 50 percent. Another essential claim of
Fermi liquid theory is that there is a one-to-one correspondence between the excited states
of the normal state of a metal and those of a non-interacting electron gas. The elementary
excitations in Fermi liquid theory are called quasi-particles. A quasi-particle is a composite
particle with a lifetime. The lifetime stems from collisions with other quasi-particles. When
the lifetime (τ ) of a quasi-particle is infinite, the state with such a particle is an eigenstate
of the system. However, the minimum constraint that must hold for a quasi-particle state to
be an eigenstate of a system is that /τ ˜p , where ˜p is the energy of the quasi-particle.
We will see below that as the energy of a quasi-particle approaches the Fermi level, its
lifetime goes to infinity. The stability of quasi-particles at the Fermi level is a crucial tenent
of Fermi liquid theory.
The vanishing of the scattering rate of electrons in the vicinity of the Fermi level can be
shown as follows. Consider a near T = 0 distribution in which all but one of the electrons
is below the Fermi surface. Let 1 be the energy of the electron above the Fermi surface.
For an electron with this energy to scatter, it must interact with some electron with energy
2 < F . The Pauli exclusion principle requires that after the scattering event, the electrons
p+q k+q
k+q
(a) (b)
Fig. 12.12 (a) The scattering of an excited quasi-particle against particles in the ground state. (b) A diagrammatic representation
of the momentum space scattering.
must occupy two states above the Fermi surface. Let the energy of these states be 3 and
4 . Energy conservation requires that 1 + 2 = 3 + 4 . If 1 = F , then all the other states
must be at the Fermi level as well to satisfy energy conservation. As a consequence, for
electrons at the Fermi level, the number of states into which they can be scattered is zero.
Hence, the scattering rate, which is proportional to the density of scattering states, must
vanish. As a result, the lifetime of electrons or quasi-particles at the Fermi level is infinite.
In the event that 1 moves away from the Fermi level, there is a window of states of width
1 − F from which we can choose the other three electronic states. Once we have chosen
two of them from the narrow window 1 − F , the energy of the last state cannot be chosen
freely as a result of energy conservation. Consequently, there are 1 − F choices for 2
and 1 − F choices for 3 ; 4 is now fixed. As a result, the scattering rate should scale as
(1 − F )2 .
To see this more rigorously, we calculate the scattering rate for two particles in momentum
states −k and −k − q scattering to states p and p + q. A diagramatic depiction of this
interaction is shown in Fig. 12.12. The scattering rate for this process is
1 2π
= |V (q)|2 nkσ (1 − nk−qσ )(1 − np+q )δ(ωpq + ωkq ), (12.62)
τ q,k,σ
where V (q) is the Fourier transform of the Coulomb potential and ωpq = p+q − p . From
energy conservation, it follows that |p + q| < p and |p + q| > pF . These conditions are
summarized by the constraint p2F < |p + q|2 < p2 or, equivalently,
p2F − p2 < 2pq cos θ + q2 < 0. (12.63)
As in our analysis of screening, we can simplify the sum over k using the definition of the
screening function (see Eq. (9.103)) for free particles. At T = 0, we obtain
1 ne
= 2 |V (q)|2 S0 (q, ωpq )(1 − np+q ). (12.64)
τ 2 q
Because we are interested primarily in p − pF ≈ 0, we can take q to be small. We then

take the q → 0 limit of the screened Coulomb potential, V (q → 0) = 4π e2 /V κTF 2
, in
the Thomas–Fermi approximation. Also, we take the q → 0 form of the structure function
derived in Chapter 9:
ωpq
S(q, ωpq ) ∝ . (12.65)
q
Consequently, the scattering rate simplifies to

1 1
∝ sin θ dθ qωpq dq. (12.66)
τ F
We can replace the q-integral with one over ωpq by solving ωpq = (2pq cos θ + q2 )/2m.
The scattering rate is proportional to
⎛ ⎞
1
1 1 p −F xp
ω ⎝1 − ⎠ dx
F
∝ dω
τ F 0 −1 x p +ω
2 2
F
p −F
1 (p − F ) 2
∝ ωdω ∝ . (12.67)
F 0 F
We see clearly then that the scattering rate scales as (p − F )2 and hence vanishes for states
at the Fermi level. Setting p − F ∝ kB T , we find that the scattering rate is proportional
to T 2 . As mentioned in the previous chapter, the resistivity (GF1987) in the normal state
of the cuprate high-Tc materials is linear in temperature down to Tc . In light of the Fermi
liquid result, this behavior strongly suggests that the normal state of these materials is a
non-Fermi liquid.
The quadratic dependence of the quasi-particle scattering rate on temperature is a central
prediction of Fermi liquid theory. At room temperature, (kB T )2 /F is of the order of 10−4 eV.
This gives rise to a scattering lifetime that is of the order of 10−10 s at room temperature.
A typical relaxation time in a metal is four orders of magnitude shorter (see Chapter 11).
Hence, electron–electron interactions are not the dominant scattering mechanism for elec-
trons in the vicinity of the Fermi level. It is for this reason that the non-interacting picture
works so well and is a good approximation for the normal-state properties of a metal, at least
at room temperature. Of course, at sufficiently low temperatures, the importance of electron–
electron interactions increases tremendously, superconductivity being a case in point.
The ultimate statement (P1992; SM1991; BG1990) of Fermi liquid theory is that short-
range repulsive interactions are irrelevant at a Fermi surface. This statement can be proven
from a scaling analysis of the Lagrangian for the interacting system once we establish a few
preliminary results on fermionic coherent states. Fermionic coherent states and a fermionic
operator are conjoined by the eigenvalue equation
|ψ = ψ|ψ, (12.68)
where |ψ is the fermionic coherent state. The eigenvalue ψ is not an ordinary number,
however. Its square, ψ 2 , vanishes, thereby implying the series of equalities
2 |ψ = ψ 2 |ψ = 0|ψ (12.69)
holds. ψ is a Grassmann variable, that is an anticommuting number. While this may seem
weird, the vanishing of the square of a Grassmann is precisely what we need to engineer
the Pauli principle into a fermionic path integral formulation.
So what are the states |ψ that make Eq. (12.68) possible? We make an ansatz
|ψ = |0 − ψ|1 (12.70)
that fits the bill for a fermionic coherent state. Here |0 = 0 = † |1 = 0 and |1 = |0.
We perform the check on our ansatz by first noting that
|ψ = −ψ|1 = ψ|1 = ψ|0. (12.71)
However, |0 = |ψ + ψ|1. Substituting, we obtain
|ψ = [|ψ + ψ|1] = ψ|ψ, (12.72)
our desired result. Hence, |ψ constitutes a fermionic coherent state. Consider now the
adjoint ψ̄| † . The standard rules of quantum mechanics can be applied to the adjoint. For
example,
ψ̄| † ≡ ψ̄|ψ̄. (12.73)
Using the fact that ψ̄1| = −1|ψ̄, we obtain that
ψ̄|ψ = 0|0 + ψ̄ψ1|1

= 1 + ψ̄ψ = eψ̄ψ . (12.74)
This expression might seem a bit strange since in no sense does the truncation follow from
the smallness of ψ̄ψ. This quantity is neither big nor small. Rather, the truncation follows
because the square of a Grassmann variable is zero.
In proceeding, we will need a few integrals:

ψdψ = 1, (12.75)

dψ = dψ̄ = 0, (12.76)

ψ̄ψdψ̄dψ = −1, (12.77)

e−aψ̄ψ dψ̄dψ = (1 − aψ̄ψ )dψ̄dψ = a, (12.78)

e−ψ̄Mψ dψ̄dψ = DetM. (12.79)
These integrals allow us to establish several identities. First, the resolution of the identity,

|ψψ̄|e−ψ̄ψ dψ̄dψ = |00| + |11| = I, (12.80)
takes on a particlularly simple form in terms of fermionic coherent states. In writing the
path integral in terms of fermionic coherent states, we will need an expression for the trace
of a bosonic operator. By bosonic we simply mean an operator that is made up of an even
number of fermionic operators. Consider the quantity

−ψ̄|ψ = 0| − ψ̄1| |0 − ψ|1
= 0|0 − ψ̄ψ1|1. (12.81)
To obtain just the sum of the diagonal elements, we perform the integral

0|0 − ψ̄ψ1|1 e−ψ̄ψ dψ̄dψ = 0|0 + 1|1. (12.82)
Consequently, the trace formula is given by

Tr = −ψ̄||ψe−ψ̄ψ dψ̄dψ. (12.83)
With these relationships, we proceed to express the partition function,

Z = Tr e−βH , (12.84)
in terms of coherent states. Making contact with the standard Schrödinger propagator, we
identify β as an imaginary time. We subdivide this “time” interval by defining = β/N
and raise our Boltzmann weight,
N
Z = lim Tr e−H , (12.85)
N→∞
to the Nth power. In the large N-limit, we can truncate the exponential factor at the first
term. For compactness, we define D̂ ≡ D̂ (ψ † , ψ ) = 1 − H . We then use the spectral
resolution of the identity, Eq. (12.80), to obtain

Z = −ψ̄0 |D̂ |ψN−1 e−ψ̄N−1 ψN−1 ψ̄N−1 |D̂ |ψN−2 eψ̄N−2 ψN−2
N−1
× ψ̄N−2 | · · · |ψ1 e−ψ̄1 ψ1 ψ̄1 |D̂ |ψ0 e−ψ̄0 ψ0 dψ̄i dψi (12.86)
i=0
as our working expression for the partition function. Assume all operators in H are normal-
ordered so that H |0 = 0. As a consequence,
†
†
ψ̄i+1 |D̂ |ψi = ψ̄i+1 |D (ψi+1 , ψi )|ψi = eψ̄i+1 ψi e−H (ψi+1 ,ψi ) . (12.87)
We are almost done. We simply insert Eq. (12.87) into Eq. (12.86) N times,

N−1 (ψ̄i+1 −ψ̄i ) ψ −H
i
Z = lim e dψ̄i dψi , (12.88)
N→∞
i=0
where we have defined ψ̄N = −ψ̄0 and ψN = −ψ0 , and ψN and ψ̄N are explicitly not
integrated over. In the N → ∞ limit, we can convert the sum in the exponential to an
integral. Upon integrating the time-derivative term by parts with the boundary condition
that ψ (0) = ψ (β ) = 0, we reduce the partition function to the standard form

Z = Dψ̄Dψe−SE , (12.89)
where SE ,

SE = ψ̄∂τ ψ + H dd xdτ, (12.90)
is the Euclidean action with the τ integration performed on the interval [0, β]. This is the
starting point for all fermionic quantum field theories.
Consider the case of non-interacting electrons. The Hamiltonian,

H= (p − F )ψσ† (p)ψσ (p), (12.91)
p,σ
is conveniently written in terms of creation and annihilation operators in momentum space

with (p) the free particle dispersion and F the Fermi energy. The corresponding action is

S= dtd3 p iψσ† (p)∂t ψσ (p) − ψσ† (p)(p − F )ψσ (p) , (12.92)
where we have switched to real time. Technically we should have written the action in terms
of fermionic coherent states rather than the fermionic creation and annihilation operators.
While this distinction will not be crucial for the application we consider here, it is crucial
in the fermionic path integral that the fermionic coherent states be used. Nonetheless, it
is understood that path integrals written in terms of creation and annihilation operators
actually imply coherent states as the operative basis.
We want to consider scalings towards the Fermi surface. We first need to introduce the
concept of a scaling dimension. Consider the simple function f (x) = x2 . As a consequence
of the scaling x → sx, f (x) → s2 x2 , where s is assumed to be positive. If s → 0, the
function, f , vanishes. We define the scaling dimension of the parameter x as the power
to which the scale factor, s, must be raised to obtain the scaling transformation of x.
Operationally,
df s d ln f
Dim[x] = = = 2. (12.93)
ds f d ln s
In the context of a field theory, the scaling dimension plays a fundamental role. Let us
partition our action as
S = S0 + Sint , (12.94)
where all the interactions are contained in

Sint = dtdd kO(k, t ). (12.95)
Here O is some operator (of course a composite operator made out of the elemental fields)
that determines the interactions. We choose the scaling of time and momenta and the bare
fields such that S0 → s0 S0 = S0 . Under such a choice for the scaling, the interaction term
will transform as sb Sint . If b > 0, then the interactions generated by O are irrelevant as they
vanish in the scaling limit. In this case, we call O an irrelevant operator. If b < 0, then O is
relevant. The marginal case arises when b = 0. Since we are considering scalings towards
the Fermi surface, we want to scale all of the fields and spatial and time coordinates so that
the associated free particle action has a scaling dimension of zero. We will then show that
when the same scalings are applied to any repulsive short-range four-fermion interaction,
the scaling dimension will be positive provided that d ≥ 2. Following Polchinski (P1992),
we write the momentum of an electron,
p = k + , (12.96)
as a sum of two momenta, one on the Fermi surface, k, and one perpendicular to it, , which
will represent the deviations from the Fermi surface. As a result, we expand the dispersion
relationship of an electron around the Fermi surface,
∂
(p) = F + + O(2 ), (12.97)
∂p
retaining only the linear term. Since the zero of energy is set to be the Fermi surface, the
relevant scale changes which preserve the Fermi surface are k → k, → s and E → sE.
As a result of the latter scaling for the energy,
∂ ∂
→s , (12.98)
∂t ∂t
implying that t → s−1t. Likewise, dd p → dd−1 kd(s). Under these scale changes, the
action,

s dtdd−1 kd iψσ† (p)∂t ψσ (p) − vF ψσ† (p)ψσ (p) , (12.99)
scales linearly with s times the scaling dimension of the product ψσ† (p)ψσ (p). Our goal of
obtaining a scale dimension of zero for the action requires that we scale the fields in the
manner
ψσ (p) → s−1/2 ψσ (p), (12.100)
to absorb the overall factor of s arising from the scaling of and the energy. From our
definition of the scaling dimension, it follows that
1
Dim[ψσ (p)] = − . (12.101)
2
Our theory on the Fermi surface then has scale dimension zero. However, we have not yet
specified the chemical potential. Such a term,

dtdd−1 kdμ(k)ψσ† (p)ψσ (p), (12.102)
takes the form of a mass term in our effective low-energy theory. This term scales as
s−1+1−2/2 = s−1 . Hence, this term is always relevant. Nonetheless, this term can be absorbed
into the kinetic energy term, implying that the sense in which we expand around the Fermi
surface is with respect to the completely filled state, not one with a vacuum we can
(conveniently) consider to be empty.
A theory is natural if there are no relevant perturbations. Consider the short-range four-
fermion interaction term,
4
dt dd−1 ki diV (k1 , . . . , k4 )ψσ† (p1 )ψσ (p3 )ψσ† (p2 )ψσ (p4 )δ d (p1 + p2 − p3 − p4 ).
i=1
(12.103)
At this point, the sign of V is arbitrary. Our focus on short-ranged interactions is warranted
since we have shown that, at metallic densities, the interactions in metals are well screened.
Under the scalings used for the free part of the action, we find that the interaction term
scales as s4−1−4/2 = s times the scaling of the delta-function. The s4 term arises from the
four factors of d, s−1 from the time integration and s−2 from the four factors of the fields,
ψσ (p). What about the scaling of the δ-function? If we simply substitute Eq. (12.96) into
the argument of the δ-function, we find that
δ d (p1 + p2 − p3 − p4 ) = δ d (k1 + k2 − k3 − k4 + l1 + l2 − l3 − l4 )
≈ δ d (k1 + k2 − k3 − k4 ). (12.104)
Hence, the δ-function does not scale and we arrive at the conclusion that short-range
interactions are irrelevant on the Fermi surface as they scale linearly with s.
However, this argument concedes too much to the Fermi surface. Superconductivity can
be understood after all as an instability of the Fermi surface with respect to attractive
contact interactions. Hence, the argument above is too broad. We fine-tune the argument
by considering scattering processes that are connected by large momenta as a precursor
to the superconducting problem where the scattering involves electrons on opposite sides
of the Fermi surface depicted in Fig. 12.13. Specifically, consider a pair of electrons with
momenta p1 and p2 that scatter to a state with momentum p3 and p4 . At the outset, we
consider a generic case in which the momenta p1 and p2 share no special relationship,
as depicted in Fig. 12.13. We write the scattered momenta as p3 = p1 + δk3 + δ3 and
p4 = p2 + δk4 + δ4 . The momentum conservation condition now becomes
δ d (δk3 + δk4 + δ3 + δ4 ). (12.105)
(a) (b)
δk 2
δk2
1 2
2
δk1
1
δk 1
Fig. 12.13 (a) Momenta for two electrons generically located on the Fermi surface. (b) Here the momenta are opposite; that is,
p1 = −p2 . In this case, δk1 is parallel to δk2 and δ1 and δ2 point in exactly opposite directions.
When the δki s are linearly independent, the δ-function can be treated as before. Specifically,
it has no scaling and our previous conclusion holds that four-fermion interactions are
irrelevant. Consider the case shown in Fig. 12.13b, in which p1 = −p2 and hence δk3
and δk4 are parallel to one another. In this case, the δ-function that maintains momentum
conservation,
δ d (δk3 + δk4 )δ d (δ3 + δ4 ), (12.106)
factorizes because both δ3 and δ4 are perpendicular to the δk3 and δk4 . Under the scaling
→ s, the second of the two δ-functions scales as s−1 . This factor cancels the linear s
factor derived earlier. Consequently, we find quite generally that the four-fermion interaction
when the pair of electrons has zero net momentum is neither relevant nor irrelevant. Rather,
it scales as s0 and hence is marginal.
So where does this leave us? We need to work harder to determine the fate of our Fermi
surface when the electrons scatter to parity-symmetric momentum states. To this end, we
evaluate the first correction in the hope of constructing a recursion relationship similar
to those derived for the exchange coupling in the Kondo problem. Through order V 2 the
relevant diagrams are shown in Fig. 12.14. The first diagram is just the bare electron–
electron interaction vertex. The second diagram is the vertex through second-order in the
interaction. We associate with each internal fermion line (that is the lines of the bubble)
with a propagator of the form 1/(E − vF ), where E is the energy of the internal fermion
line. The second-order vertex takes the form

dE d2 kd 1
V2 .
2π 4 [(1 + i)(E + E ) − vF (k ) ] [(1 + i)(E − E ) − vF (k ) ]

(12.107)
(a) (b) p′, E −p′, E

−p′, E
p′, E
p″, E+E′ −p″, E−E′
p, E −p, E
p, E −p, E

Fig. 12.14 (a) Bare interaction vertex for electrons in state (p, E ) and (−p, E ) scattering to (−p , E ) and (−p , E ), and
(b) O(V 2 ) vertex. The integral corresponding to this vertex is evaluated in Eq. (12.107).
The (1 + i) factors serve to regularize the integral. In the end, we will take the limit
of → 0. In performing the integrals in the bubble vertex, it is helpful to rewrite the
denominator as
−1
, (12.108)
(1 + 2i) [E + E − (1 − i)vF (k ) ] [E − E + (1 − i)vF (k ) ]
where we have explicitly approximated (1 + i)2 ≈ (1 + 2i) since is strictly of O(0).

To evaluate this integral, we close in the upper half-plane. There are two poles of interest
which are located at
E = ±E ∓ vF (k ) ± ivF (k ) , (12.109)
where the upper sign corresponds to > 0 and the lower sign to < 0 so that both poles
reside in the upper half-plane. Using the fact that the residue at each of these poles is
2π i
, (12.110)
±2E ∓ 2vF (k ) ± 2ivF (k )
we reduce the second-order vertex to
E0 0
−iV 2 db db
N + , (12.111)
2 0 E − b + i −E0 b − E − i
where we have changed variables to b = vF (k ) , defined

d2 k 1
N= , (12.112)
(2π )3 vF (k )
and ignored any prefactor on the terms proportional to the infinitesimal, . Each integral
yields a logarithm:

E0
iV 2 N ln . (12.113)
|E|
Noting that the bare four-fermion vertex is proportional to −iV , the effective interaction to
second order becomes

E0
V (E ) = V − V 2 N ln + O(1) + O(V 3 ). (12.114)
|E|
We differentiate both sides of this equation,

dV (E )
= NV (E )2 /E, (12.115)
dE
to obtain the scaling equation for the interaction vertex. This equation is identical to the
scaling equation derived previously for the growth of the exchange interaction in the Kondo
problem in Chapter 8. The solution to this equation is simply
V
V (E ) = . (12.116)
1 + NV ln(E0 /E )
We see clearly that for a positive bare value of the interaction, V > 0, the effective interaction
decreases at low energy, that is, E < E0 . However, for V < 0, the interaction increases. This
is the BCS instability and the fundamental statement that, except for short-range attractions,
all renormalizations are towards the Fermi surface. Note that the bare interaction V is truly
marginal. Consequently, superconductivity arising within BCS theory is an example of
a marginal coupling becoming large and generating new degrees of freedom, namely a
condensate of electron pairs, at strong coupling.
It is a crucial assumption of the BCS pairing theory of superconductivity that the
normal-state properties are described by non-interacting quasi-particles with infinite life-
times (L1964; L1965) which stand in a one-to-one correspondence with the bare electrons.
This assumption is borne out by the scaling analysis just presented and experimentally in
the normal state of a wide range of Type I, as well as Type II, materials. Since the normal
state is essentially non-interacting, the energy of a particular quasi-particle is the energy
required to add or subtract that particle from a state in the system. To illustrate this concept,
we consider an interacting system described by some distribution function, n0pσ , generally
taken to be the Fermi–Dirac distribution. Suppose a particle is added or subtracted from
the system ever so slowly, so that the system remains in the ground state. This process rep-
resents an adiabatic change in the particle number. Nonetheless, this process will change
the particle distribution to some new quantity, npσ , and in turn the ground state energy will
change. Let δnpσ represent the resultant change in the particle distribution function. We
can calculate the change in the ground state energy,

1 1
δE = pσ δnpσ + fpσ,p σ δnpσ δnp σ + · · · , (12.117)
V pσ 2V 2 pσ,p σ
by expanding in powers of the fluctuation, δnpσ . The Landau (L1956) parameter fpσ,p σ
describes how quasi-particles interact. This quantity is invariant with respect to interchange
of p and p . We identify the energy of a quasi-particle,
1
˜pσ = pσ + fpσ,p σ δnp σ + · · · , (12.118)
V p σ
by taking the variation of the new ground state energy with respect to δnpσ . We see clearly
then that the quasi-particle energy is itself a function of the distribution function. More
explicitly, quasi-particles obey a Fermi–Dirac distribution function with p − μ replaced
with ˜pσ − μ. At the Fermi surface, quasi-particles move with a velocity

∂pσ
vF = (12.119)
∂p p=pF
that is equal to the Fermi velocity. Because vF = pF /m∗ , the formal definition of the
effective mass of a quasi-particle is

1 1 ∂pσ
= . (12.120)
m∗ pF ∂p p=pF
Consequently, in the vicinity of the Fermi surface, the quasi-particle energy
pσ = F + vF (p − pF ) (12.121)
is linear in the displacement momentum p − pF . As most of our focus will be on processes
in the vicinity of the Fermi surface, this expression should suffice to describe the energy of
a quasi-particle.
12.6 Pair amplitude
At this point it is customary to write down the BCS wavefunction and go on our merry
way to elucidate the superconducting transition. However, we want to take a step back and
explore the expectation value of the pairing amplitude for a Cooper pair. In so doing, we will
be able to swindle BCS theory. Thus far, we have established that two electrons above the
Fermi surface can lower their energy by forming a bound singlet Cooper pair just below F ,
if the temperature is sufficiently low. This tells us that we should be able to define an order
parameter whose expectation value should discern if the Cooper instability has occurred.
The appropriate pairing operator for a (p ↑, −p ↓) pair is
b†p = a†p↑ a†−p↓ , (12.122)
217 12.6 Pair amplitude
while
bp = a−p↓ ap↑ (12.123)
is the corresponding pair annihilation operator. Let |N represent the
ground state of
an N-particle system. The Cooper instability suggests that N − 2 bp N = 0 in the
superconducting state. Let us define

αp = N − 2 bp N, (12.124)
which is generally referred to as the pair amplitude. What we will show is that below a
certain temperature, αp grows exponentially. In the normal state, αp equals 1 or 0. In BCS
theory, αp is closely related to the order parameter.
We start by showing that αp is the general expansion coefficient for a pair state. To
proceed, we write the general singlet pair state as

|ψ = ηp b†p |0 = ηp |p ↑ −p ↓, (12.125)
p>pF p>pF
where ηp is an expansion coefficient. To determine ηp , we note that

†
0 bp ψ = ηp 0bp bp 0
p

= ηp δpp = ηp . (12.126)
p
Because |ψ simply differs from |0 by two particles, ηp = αp .

The pair amplitude is the expansion coefficient for the general pairing state. Let us
calculate the time evolution of αp . To do this, we write the BCS Hamiltonian as
H = H0 + Hint , (12.127)
where

H0 = p a†pσ apσ
p,σ

Hint = Vpp b†p bp . (12.128)
pp
In the Schrödinger picture, the time dependence of αp is carried in the wavefunctions. As

a consequence,
∂ ∂
i αp (t ) = i 0(t ) bp ψ (t )
∂t ∂t
= 0(t ) bp , H ψ (t ) . (12.129)
To evaluate [bp , H ], we will find the commutators useful:
[bp , np σ ] = bp (δpp δσ ↑ + δp,−p δσ ↓ ),

bp , b†p = δpp (1 − np↑ − n−p↓ ). (12.130)
We see explicitly that the occupancy factors, np↑ + n−p↓ , render the pair-creation operators
non-commutating. These terms preserve the Pauli principle between electrons forming the
pair. We can now evaluate [bp , H ]:

[bp , H ] = bp , p np σ + Vp p bp , b†p bp
p σ p p

= 2p bp + Vpp 1 − n−p↓ − np↑ bp . (12.131)
p
The prime on the sum indicates a restricted sum over the thin momentum shell of width 2ωD
around F . We can gain physical insight into this result by noting that (1 − n−p↓ − np↑ ) =
(1 − np↑ )(1 − n−p↓ ) − np↑ n−p↓ . If we use this result with Eq. (12.131), we obtain

∂
i − 2p αp (t ) = 1 − np↑ 1 − n−p↓ − np↑ n−p↓ Vpp αp (t ) (12.132)
∂t p
as the time evolution of the pair amplitude. The probability that the particles forming the
pair lie outside the Fermi surface is determined by the product (1 − np↑ )(1 − n−p↓ ), while
np↑ n−p↓ is the probability that they lie inside. This term can be ignored if we are interested
strictly in pair formation between two momentum states just outside the Fermi surface, as
in the Cooper problem. Our effective equation of motion is

∂
i − 2p αp (t ) = 1 − np↑ 1 − n−p↓ Vpp αp (t ) (12.133)
∂t p
in this limit. Because i∂/∂t → E, the above expression is precisely the Schrödinger
equation we solved previously to determine the Cooper bound state energy. However, the
general pair amplitude equation, Eq. (12.132), dictates that all particles be involved in the
pairing process whether they lie inside or outside the Fermi surface. This is the essence
of BCS theory. Cooper included only the particles outside F . BCS simply included the
np↑ n−p↓ term in addition. This simple change made all the difference.
12.6.1 Instability: superconducting state

We turn to the evaluation of the pair amplitude. We want to show that below some character-
istic temperature, Tc , the pair amplitude grows exponentially. We seek a solution of the form
αp (t ) = e−i αp (t = 0).
zt
(12.134)
Substituting this expression into the pairing equation, we obtain

z − 2p αp (0) = 1 − np↑ − n−p↓ Vpp αp (0) (12.135)
p
as the Fourier transform of the pair amplitude equation.
219 12.6 Pair amplitude
For the model attractive potential in Eq. (12.51), the pair amplitude evolution equation
becomes

z − 2p αp (0) = −V0 1 − np↑ − n−p↓ αp (0). (12.136)
p
As before, we sum both sides of this expression to obtain

1 − np↑ − n−p↓
αp (0) = −V0 αp (0), (12.137)
p p
z − 2p p
which implies that

1 − np↑ − n−p↓
1 = −V0 . (12.138)
p
z − 2p
This is almost the expression we had previously except we are now including pairing
between all particles. To obtain a solution, we plot the intersection of the left- and right-
hand sides of the above equation. As before, we plot the RHS in the vicinity of the discrete
energies p .
As is evident, a solution exists only if 2F − 2ωD ≤ Re z ≤ 2F + 2ωD . To solve our
pair condition, we convert the sum to an integral and note that np = n−p for free particles.
Also, because p↑ = p↓ , we have immediately that
−1
np↑ = n−p↓ = eβ(p −μ) + 1 (12.139)
and
ωD
tanh β/2
1 = −V0 N (F ) d , (12.140)
−ωD z − 2μ − 2
where N (F ) is the density of states at the Fermi level. Exponential growth of the pairing
amplitude will occur if z has a positive imaginary part. Let z = 2μ + x + iy, where x and y
are real. Consider first the case in which y = 0. We obtain in this limit that
ωD
β 1
1 = −V0 N (F ) d tanh P − iπ δ (x − 2) . (12.141)
−ωD 2 x − 2
To satisfy this equation, the imaginary part must vanish. This obtains only if |x|/2 > ωD
or, equivalently, Re z > 2F + 2ωD . As illustrated graphically in Fig. 12.15, the real part
of the integrand is positive in this energy range. Consequently, y = 0 is not permissible.
For a solution to exist on the energy shell, y = 0. We write our pair constraint as
ωD
(x − 2 − iy) β
1 = −V0 N (F ) tanh d, (12.142)
−ωD (x − 2) + 2
y2 2
z
2 F – 2 ωD 2 F + 2 ω D
Fig. 12.15 The right-hand side of the pair amplitude equation as a function of the complex energy, z.
which is only true if the imaginary part,

ωD
tanh β/2d
, (12.143)
−ωD (x − 2)2 + y2
vanishes. Because tanh(x) = − tanh(−x), we rewrite this integrand as

1 ωD β 1 1
0= d tanh −
2 −ωD 2 (x − 2)2 + y2 (x + 2)2 + y2
ωD
1 β 8x
= d tanh . (12.144)
2 −ωD 2 (x − 2)2 + y2 (x + 2)2 + y2
It is now clear that x = 0 because tanh β/2 is positive everywhere. Consequently, our
solution must be of the form z = 2μ + iy. We must solve then
ωD
(−iy − 2) tanh β/2
1 = −N (F ) V0 d . (12.145)
−ωD 4 2 + y2
The imaginary part of this integral is odd in and hence vanishes exactly. We are left with
the single equation
ωD
tanh β/2
1 = 4N (F ) V0 d . (12.146)
0 4 2 + y2
Here again the integrand is positive. As y increases, the integrand decreases. The same
thing is true as T increases. Hence, for T greater than some temperature, no solution exists.
This defines the critical temperature Tc . As T → 0, we find that

N (F ) V0 (2ωD ) da
2
1=
2 0 a + y2

N (F ) V0 (2ωD )2 + y2
= ln . (12.147)
2 y2
221 12.7 BCS ground state
Exponentiation of both sides of this equation,

−1
2
y = (2ωD ) exp
2 2
−1
N (F ) V0

2
≈ (2ωD )2 exp − , (12.148)
N (F ) V0
results in the familiar equation for y, namely,

1
y = ±2ωD exp − . (12.149)
N (F ) V0
Choosing the positive solution, we conclude that as T → 0, the pair amplitude grows as
αp (t ) e−izt
= e−2iμt e2ωD t exp(−1/N (F )V0 ) . (12.150)
The existence of an exponentially growing pair amplitude is the signature of an instability

in the N-particle ground state. We have essentially derived BCS theory by focusing solely
on the pair amplitude, αp . To obtain an exact expression for Tc , we consider our pair binding
equation in the limit that y = 0. In this limit, the temperature that satisfies Eq. (12.146) is
maximized. Hence, the y = 0 solution can be used to estimate Tc . Let us define βc = 1/kB Tc .
We find that
ωD βc /2
tanh x
1 = N (F ) V0 dx. (12.151)
0 x
This integral must be done numerically. The transition temperature consistent with
Eq. (12.151) is

1
kB Tc = 1.14ωD exp − . (12.152)
N (F ) V0
We will show that the parameter y plays the role of the energy gap in BCS theory.
12.7 BCS ground state
Thus far, we have shown how the pairing hypothesis leads to an instability in the ground
state for T < Tc . Our focus now is on the many-particle wavefunction describing this state.
The starting point for our analysis is the many-body state for N non-interacting electrons:
| 0 (N ) = b†p |0. (12.153)

p<pF
Because each momentum state is doubly occupied, we can construct such a state by suc-
cessively operating on the vacuum with the pairing operator b†p . Relaxing the restriction
p < pF , we recast | 0 (N ) in the form
0
(N ) = u0p + vp0 b†p |0, (12.154)
p
where

0, p < pF ,
u0p = (12.155)
1, p > pF ,

1, p < pF ,
vp0 = (12.156)
0, p > pF ,
†
0 of | is identical to Eq. (12.153). Note also that because the ap
0
This representation
anticommute, (N ) is properly antisymmetrized.
BCS postulated that the general electron superconducting ground state can be written as

|BCS = up + vp b†p |0 , (12.157)
p
where the coefficients up and vp are to be determined variationally and are referred to as
coherence factors. The normalization condition BCS |BCS = 1 imposes the constraint

1 = 0| u∗p + vp∗ bp up + vp b†p |0 . (12.158)
p,p
Products of the form appearing in Eq. (12.158) can be simplified using the commutation
relation in Eq. (12.130) and by noting that bp b†p |0 = 1. We find that only the diagonal
(p = p ) survives:

BCS |BCS = up 2 + vp 2 , (12.159)
p
implying that
2 2
up + vp = 1. (12.160)
If we expand the product in the BCS ground state, it becomes clear that |BCS is a sum of
all 2m particle states,

|BCS = A2m |2m , (12.161)
m
223 12.7 BCS ground state
where m is an integer, A2m is an expansion coefficient, and |2m is the ground state for 2m
particles. We can relate |2m to |BCS by writing the expansion coefficients (S1964) as
A2m = |A2m |e2imφ so that
φ 2imφ

BCS = e |A2m ||2m . (12.162)
m
That is, we associate with each 2m-particle state an overall phase, 2mφ. To this end, we
rewrite the BCS ground state
φ
= up + e2iφ
v b†
p p |0 (12.163)
BCS
p
by multiplying each pair-creation operator by exp(2iφ). Multiplying Eq. (12.162) by

exp(−i2mφ) and integrating over φ, we obtain a concise relation
2π
1
|2m = dφe−i2mφ up + e2iφ vp b†p |0 (12.164)
2π |A2m | 0 p
between each 2m-particle state and the BCS ground state. That only states with 2m particles
survive on the right-hand side of Eq. (12.164) follows from the fact that
2π
1
eiφ(N−N ) dφ = δNN . (12.165)
2π 0
Because the 2m-particle states are mutually orthogonal, when we multiply Eq. (12.164) by
2m |, the left-hand side becomes unity, allowing us to identify the square of the expansion
coefficient as
2π 2π
dφ dφ −i2m(φ−φ )
u∗p + e−2iφ vp∗ bp up + e2iφ vp b†p |0

|A2m | =
2
e 0|
0 0 2π 2π p,p
2π
1
= dφe−2imφ up 2 + e2iφ vp 2 . (12.166)
2π 0 p
Equation (12.166) should be construed as the probability distribution for 2m particles in the
BCS state. This quantity is strongly peaked at the average number of particles in the BCS
state and has a width proportional to the square root of the average number of particles in
the system (see Problem 12.8).
Our rewriting of the BCS state as a linear superposition of all possible 2m-particle states,
all with the same phase, φ, suggests that in a superconducting state, a conjugacy relation-
ship exists between the phase and the particle number. Indeed, such a relationship exists.
However, such a relationship cannot exist if one of the variables is bounded from below.
Without loss of generality, we can treat the particle number as being a continuous variable
from −∞ < N < ∞. We can establish the conjugacy relationship straightforwardly by
φ
acting −i∂/∂φ on |ψBCS in Eq. (12.164) and then integrating by parts. The result
2π
1 −i∂
dφe−i2mφ up + e2iφ vp b†p |0 = 2m|2m (12.167)
2π |A2m | 0 ∂φ p
implies that the particle number and the phase are related by
∂
N ↔ −i , (12.168)
∂φ
as long as we regard the particle number as a continuous variable. Similarly, by applying

i∂/∂ (2m) to Eq. (12.162), we establish analogously that
∂
i ↔ φ. (12.169)
∂N
Hence, the phase and the particle number are conjugate dynamical variables as long as we
consider the particle number to be a continuous variable. As such, the particle and phase
must satisfy the Heisenberg uncertainty relationship φ N ≥ 2π for two canonically con-
jugate variables. Consequently, complete certainty in the phase implies infinite uncertainty
in the particle number, as is the case in a superconductor. In addition, the phase and the
particle-number operators should also satisfy Hamilton’s equations
∂ Ĥ
iN̂˙ = [Ĥ , N̂] = i ,
∂ φ̂
˙ ∂ Ĥ
iφ̂ = [Ĥ , φ̂] = −i (12.170)
∂ N̂
for two canonically conjugate variables. We will use these equations in the last section of
this chapter where we formulate the Josephson effect.
12.8 Pair fluctuations
We focus now on computing bp for a BCS wavefunction. In the BCS ground state, the
average of the pair operator

bp BCS = 0| u∗p + vp∗ bp bp up + vp b†p |0
p ,p
= 0| u∗p vp bp b†p |0 = u∗p vp (12.171)

is non-zero. However, for a normal state, up and vp are never non-zero simultaneously (see
Eqs. (12.155) and (12.156)). Hence, bp = 0 in the normal state. Although bp is not
225 12.8 Pair fluctuations
identical in form to the pair amplitude studied in the previous sections, it contains the same
information. This state of affairs obtains because the BCS ground state is made up of states
with different numbers of electrons. A non-zero value of bp for a particular ground state
indicates that that state favors pair formation.
It is reasonable then to define
1
χ= bp (12.172)
V p
as the effective order parameter for superconductivity. The average value of χ ,
1 ∗
χ BCS = u vp , (12.173)
V p p
has a definite value in the superconducting state. To prove this, it is sufficient to show that
the second-order fluctuations of the order parameter vanish. We proceed by evaluating the
second moment
) *
1
χ BCS = 2
2
bp bp . (12.174)
V p,p
When we insert |BCS , the only non-zero terms will come from the states whose momenta
coincide with the indices in Eq. (12.174). Consequently, χ 2 BCS is a sum of
1
†

∗ ∗ ∗ ∗ †
0| up + vp bp up + vp bp bp bp up + vp bp up + vp b
p |0
V 2 p,p ,p=p
(12.175)
when p = p and
1
∗ ∗ †
+ v + v p |0
2
0| up p bp bp up p b (12.176)
V2 p
for p = p . The second expression vanishes because b2p = 0. Hence, χ 2 BCS is off-
diagonal in momentum space, and Eq. (12.175) reduces to
1 ∗
χ 2 BCS = 2 u vp u∗ vp
V p,p ,p=p p p
1
= 2
bp bp 1 − δpp
V p,p
1
= χ 2BCS − bp 2 . (12.177)
V2 p
The second-order fluctuation in the order parameter is

1 1
δχ 2 = χ 2 BCS − χ 2BCS =− bp 2 . (12.178)
V V p
Once we determine the coefficients up and vp , we will be able to show that δχ 2 ∝ V −1
and hence vanishes in the thermodynamic limit. On this basis, we will argue that χ takes
on a definite value in the superconducting state and, as advertised, is the order parameter
for the superconducting transition.
12.9 Ground state energy
We determine the coefficients up and vp by minimizing the ground state energy subject to
the normalization constraint (Eq. (12.160)) and a particle-number constraint on the ground-
state energy. We impose that the average number of particles in the BCS ground state is
the desired value (supposedly even) by introducing a Lagrange multiplier into the average
value of the Hamiltonian:
E = H − μN, (12.179)
where μ is the chemical potential. To simplify this calculation, we rescale the single-particle
energies such that p ≡ p2 /2m − μ. The energy E is now

E = p npσ + Vpp b†p bp . (12.180)
p,σ p,p
Let us first compute npσ :

npσ = 0| u∗p + vp∗ bp a†pσ apσ up + vp b†p |0
p p
2
= vp 0| bp npσ b† |0(δp,p δσ,↑ + δp,−p δσ,↓
p
p ,p
$ 2
v , σ =↑,
= p 2 (12.181)
v−p , σ =↓ .
Reflection symmetry requires that the expansion coefficients be invariant under p → −p.
The dominant term in b†p bp arises from p = p because b2p = 0. In this case the
average can be factorized: b†p bp = b†p bp = up u∗p vp∗ vp . Consequently, the average
value of the ground-state energy is
2
E = 2 vp p + Vpp up u∗p vp∗ vp . (12.182)
p p,p
We need to vary E ,

δE = 0 = 2 p δvp∗ vp + Vpp δvp∗ up u∗p vp + Vpp δu∗p up vp∗ vp , (12.183)
p p,p p,p
227 12.9 Ground state energy
with respect to vp∗ and u∗p to find the minimum energy. To eliminate δu∗p , we note that the
variation of the normalization constraint can be rewritten as
up δu∗p + vp δvp∗ = 0 (12.184)
or, equivalently,
vp ∗
δu∗p = − δv . (12.185)
up p
Substitution of this expression into Eq. (12.183) yields

⎡ ⎤
v 2
δvp∗ ⎣2p vp + Vpp up u∗p vp − Vp p up vp∗ ⎦,
p
0= (12.186)
p p
u p
which implies that

vp2
0 = 2p vp − p up + ∗p , (12.187)
up
where we have defined

p = − Vpp u∗p vp . (12.188)
p
As we will see, p will play the role of the energy gap. In fact, we can see this immediately
by dividing Eq. (12.187) by up ,
2
vp vp
0 = 2p + ∗p − p , (12.189)
up up
and solving this quadratic equation for vp /up ,
2
vp − p ± p2 + p
= . (12.190)
up ∗p
Let us define
2
εp ≡ p2 + p , (12.191)
which will turn out to be the quasi-particle energy in the superconducting state. At this
point we can see, at least heuristically, that p does introduce a gap into the single-particle
spectrum, p . We can simplify Eq. (12.190) further by noting that
vp p2 − εp2
=
up −p ∓ εp ∗p
p
= . (12.192)
p ± εp
We assess which sign is appropriate by noting that, in the free system, vp /up → 0 when
p > F . Consequently, we should choose the + sign.
Coupled with the normalization constraint,

2 2 p 2 2 εp p + εp
1 = up + up 2 = 2 up 2 , (12.193)
p + εp p + εp
we determine the magnitude of up to be
2
up = p + εp . (12.194)
2εp
There is clearly a phase degree of freedom associated with the definition of up . To make
life simple, we choose up to be positive and real. We find then that
6
p + εp
up = (12.195)
2εp
and
6 6
p p + εp p εp − p
vp = = . (12.196)
p + εp 2εp ∗p 2εp
For the simple attractive interaction chosen, p is real if motion of the center-of-mass is
ignored and vp simplifies to
6
εp − p
vp = . (12.197)
2εp
Returning to our original problem of determining the ground state energy, we reduce the
average energy
2
Es = 2 p vp + Vpp u∗p vp up vp∗
p p,p
2
= 2p vp − p up vp∗
p

εp − p 2
=− (12.198)
p
2εp
229 12.10 Critical magnetic field
to a single sum using the form of the coherence factors just derived. In the absence of the
attractive interaction, we had that

En = 2p θ (pF − p) (12.199)
p
with θ the Heaviside step function. Because we have redefined p ≡ p2 /2m − μ, p < 0
for p < pF . Consequently, the energy difference between the superconducting and normal
states is given by the integral
0
ωD ε − 2
p
δE = −N (F − μ) d + 2d
−ωD 2εp −ωD
ωD 2
εp −
= −N (F − μ) d. (12.200)
0 εp
To obtain the final result, we substitute the expression for εp and perform the integral over
. In obtaining the final result, it is expedient to introduce the new variable = sinh y.
Consequently,
2
2
εp − = sinh y + 1 − sinh y
2
= ( cosh y − sinh y)2 = 2 e−2y . (12.201)

As a result,
sinh−1 (ωD / )
δE = −N (F − μ) 2
e−2y dy. (12.202)
0
In general, ωD / 1, and the upper limit can then be extended to ∞. The y-integration
yields 1/2, and the energy difference is
−N (F − μ) 2
δE = . (12.203)
2
That is, the ground state energy of the superconducting state is lower than that of the
corresponding normal state. The condensation energy in a superconductor is δE.
12.10 Critical magnetic field
We showed in the previous section that the condensation energy in a superconductor is

2δE = −N (F − μ) 2 . The negative sign ensures that the superconducting state is lower
in energy than the normal state. The condensation energy is related to the critical field, Hc ,
as follows.
Consider placing a superconductor in a magnetic field with the cylindrical geometry
shown in Fig. 12.16. The energy per unit volume produced by the magnetic field is B2 /8π .
Fig. 12.16 Cylindrical piece of a superconductor surrounded by a magnetic coil.
The current in the magnetic coils is held constant by the field H . Hence, there must be a
contribution from the interaction of the vector field, A(r), with the current produced by the
magnetic field. Ampere’s law states that
4π
∇ ×H= J, (12.204)
c
where J is the current in the coils that produce the magnetic field. The resultant contribution
to the energy density is

∇ × H dr
W = − A·
4π V

∇ ×A dr
=− ·H
4π V

dr
= − B·H . (12.205)
4πV
In a uniform system, B = H, and the total internal energy due to the field is H 2 /8π −
H 2 /4π = −H 2 /8π . In the normal state, this energy must be included in the total ground
state energy. As a consequence, the energy of the normal state in the presence of a magnetic
field is given by E0n = En − H 2 /8π . Because the field does not penetrate a superconductor,
the internal energy of the superconductor remains unchanged from Es . Superconductivity
will persist until En − H 2 /8π > Es . Consequently,
H2
> En − Es = −δE (12.206)
8π
is the criterion for the critical value of the field. We find that the critical field,
√
Hc = −8π δE = 4π N (F − μ) , (12.207)
is directly proportional to the gap parameter, .
231 12.11 Energy gap
12.11 Energy gap
The physical motivation for introducing p ,

p = − Vpp u∗p vp
p
p
=− Vpp , (12.208)
p
2εp
as the gap is that this quantity breaks the continuous single-particle spectrum p into
two branches. Specializing to the model potential which is non-zero only over a narrow
momentum shell around the Fermi energy, we rewrite the gap equation as

p
p = V0 . (12.209)
p 2 p2 + 2p
The prime on the sum indicates the restriction |p − pF | < ωD . Within the momentum
shell, p is relatively independent of p only if the Fermi surface is isotropic. For s-wave
pairing, the Fermi surface is entirely isotropic, and the gap can be taken to be independent
of p. For higher angular momentum pairing states such as d-wave, the Fermi surface has
nodes, and the momentum dependence of the gap cannot be ignored. For our purposes,
we will assume here that p is independent of p, at least within the thin momentum shell
around the Fermi surface. An immediate solution to the gap equation is = 0. If = 0,
then bp = 0 and only the normal phase is possible.
To obtain a non-trivial solution to the gap equation, we convert the sum to an integral:
ωD
d
1 = N (F − μ) V0 √
−ωD 2 2 + 2
ωD /
d
= N (F − μ) V0 √
0 2 + 1
ωD
= N (F − μ) V0 sinh−1 . (12.210)

Consequently, the gap is given by
ωD
= . (12.211)
sinh (1/N (F − μ) V0 )
Typically, N (F − μ) V0 1. In this limit, the gap takes on the familiar form

1
(T = 0) 2ωD exp − . (12.212)
N (0)V0
In the Cooper problem, we showed that the binding energy of a pair is identical to (T =
0), except 1/ (N (0)V0 ) →2/ (N (0)V0 ). In the full theory, as remarked earlier, pairing of
Δ(T )
0 Tc
T
Fig. 12.17 Asymptotic behavior of the gap for
1/2the model BCS interaction. Asymptotically the gap scales as
(T ) = (T = 0) 1 − TTc .
particles below the Fermi energy is included as well. As a consequence, the exact exponent
is 2/(2N (0)V0 ) = 1/(N (0)V0 ). In this sense, pairing of particles below the Fermi surface
effectively doubles the density of states at the Fermi level. Comparing the gap with Tc
(Eq. (12.152)), we find that
2 4ωD
= = 3.52, (12.213)
kB Tc 1.14ωD
which is the crucial ratio in BCS theory. It determines whether a superconductor is in the
weak coupling (with respect to the phonon interaction) limit. Strong coupling corresponds
to 2 /kB Tc > 3.5.
To evaluate the gap at finite temperature, we include the Fermi distribution weight factor,
1 − np↑ − n−p↓ , which accounts for pairing among all the particles. Because np↑ = n−p↓ ,
we can write the finite temperature gap equation as

1 − 2np
p = − Vpp p . (12.214)
p
2εp
For an isotropic system and the BCS model potential, we obtain

ωD √
tanh β 2 + 2 /2
1 = V0 N (0) √ d (12.215)
−ωD 2 2 + 2
as the defining equation for the finite temperature gap in the excitation spectrum. This
equation must be solved numerically. However, the approximate solution is (T ) =
√
(T = 0) 1 − T /Tc . This behavior is illustrated in Fig. 12.17.
Recalling that the second-order fluctuation in the order parameter,
1 ∗2 2
δχ 2 = − 2 u v , (12.216)
V p p p
233 12.12 Quasi-particle excitations
is determined by the same combination of coherence factors as is the gap, we see imme-
diately that if the gap is well behaved then δχ 2 vanishes in the thermodynamic limit.
Consequently, bp takes on a definite value in the superconducting state.
12.12 Quasi-particle excitations
To be able to determine the energy required to add an extra particle to a superconductor,

we first determine the normalized creation operator for adding an extra electron to |BCS .
We note at the outset that

a†p ↑ |BCS = a†p ↑ up + vp b†p |0
p

= a†p ↑ up up + vp b†p |0
p=p

= up ψp ↑ (12.217)
and

a−p ↓ |BCS = a−p ↓ vp a†p ↑ a†−p ↓ up + vp b†p |0
p=p

= −vp ψp ↑ , (12.218)
where

ψp ↑ = a† up + vp b†p |0 (12.219)
p↑
p=p
is the normalized wavefunction when an extra electron with momentum p and spin ↑
is added to the BCS ground state. Consequently, a†p↑ and a−p↓ are not the operators that
should be used to construct the excited states because the resultant states are not normalized.
However, the linear combination of the electron operators,
†
γp↑ = u∗p a†p↑ − vp∗ a−p↓ , (12.220)
when acting on |BCS ,

†
γp↑ |BCS = u∗p a†p↑ − vp∗ a−p↓ up + vp b†p |0
p
2
= up + vp ψp↑
2

= ψp↑ (12.221)
does produce the normalized state |ψp↑ . The corresponding creation operator for the state
|ψ−p↓ is
†
γ−p↓ = u∗p a†−p↓ + vp∗ ap↑ . (12.222)
Because the γp s are linear combinations of the ap s, they must obey the standard Fermi–Dirac
anticommutation relationships, namely

γpσ , γp† σ = δpp δσ σ ,
+

†
γpσ , γp σ + = γpσ , γp† σ = 0. (12.223)
+
The anticommutation relationships guarantee that
γp↑ |BCS = γ−p↓ |BCS = 0. (12.224)
That is, no quasi-particles exist in the ground state of a superconductor.

Because the γp† s are linear combinations of an electron and a hole operator with antiparal-
lel spins, the excitations created by these operators do not carry a well-defined charge. That
is, creation of a quasi-particle in a superconductor does not conserve charge. Quasi-particles
have a well-defined spin, however, because adding an electron with spin /2 or a hole with
spin −/2 creates an eigenstate of Sz with spin /2. Cooper pairs, on the other hand, carry
charge 2e but are spinless in the singlet channel. Consequently, the quantities that carry
well-defined spin and charge in a standard singlet BCS superconductor are distinct entities.
It is in this sense that spin–charge separation obtains in singlet BCS superconductors. We
show now that the energy dispersion relation for the spin carrier in a singlet superconductor
is εp . To prove this assertion, we start with the equations of motion for ap↑ ,

iȧp↑ = p ap↑ − Vp1 p2 ap↑ , b†p1 bp2
p1 ,p2

= p ap↑ + Vpp1 bp1 a†−p↓ , (12.225)
p1
and for a†−p↓ ,

iȧ†−p↓ = −p a†−p↓ + Vpp2 b†p2 ap↑ . (12.226)
p2
#
To simplify these equations of motion, we replace p1 Vpp1 bp1 by its average value: − p =
#
p1 Vpp1 bp1 . As a consequence, the effective Hamiltonian

Heff = p npσ − p b†p + bp (12.227)
pσ p
235 12.12 Quasi-particle excitations
is linear in the pairing operator b†p , as opposed to quadratic as in the full Hamiltonian.
As a result, particle number is not conserved in the reduced Hamiltonian. Nonetheless,
the mean-field treatment of the BCS Hamiltonian provides an accurate description of the
superconducting state. This state of affairs occurs partly because the BCS ground state does
not have a fixed number of particles.
The mean-field closure leads to a coupled set of equations:
iȧp↑ = p ap↑ − p a†−p↓ ,
iȧ†−p↓ = −p a†−p↓ − p ap↑ , (12.228)
linear in the gap parameter, p . For economy of notation, we rewrite this set of equations
in matrix form:

∂ ap↑ p − p ap↑
i † =
∂t a−p↓ − p −p a†−p↓

ap↑
= Heff † . (12.229)
a−p↓
To aid in diagonalizing the reduced Hamiltonian matrix, we introduce the angle θp , such
that
p
cos θp = (12.230)
εp
and
p
sin θp = . (12.231)
εp
The utility of this transformation is immediate because the BCS coefficients

1 p 1/2 1 1/2
up = √ 1 + = √ 1 + cos θp (12.232)
2 εp 2
and

1 p 1/2 1 1/2
vp = √ 1 − = √ 1 − cos θp (12.233)
2 εp 2
form the eigenvector basis of the reduced Hamiltonian matrix,

cos θp −sin θp
Heff = εp . (12.234)
−sin θp −cos θp
The eigenvalues of this matrix are ±1. Consequently, the eigenvectors are of the form

sin θp
. (12.235)
cos θp ± 1
The ratio of the coefficients in the eigenvectors is

sin θp θp ±1
= ± tan . (12.236)
cos θp ± 1 2
Because
θp 1 − cos θp
tan =± , (12.237)
2 1 + cos θp
the eigenvectors of Heff can be written as

vp sin θp /2
e1 = = (12.238)
up cos θp /2
and

up cos θp /2
e2 = = . (12.239)
−vp − sin θp /2
It is now clear that Heff e1 = −εp e1 and Heff e2 = εp e2 . Simply put, ±εp are the two
eigenenergies of Heff . To lay plain the utility of e1 and e2 , we multiply the equations of
motion by e1 and e2 . The result for e2 ,

∂ †
ap↑
i ap↑ up − vp a−p↓ = e2 Heff †
T
∂t a−p↓

a p↑
= εp eT2 † , (12.240)
a−p↓
demonstrates that
iγ̇p↑ = εp γp↑ . (12.241)
Similarly, for e1 we have that

∂ ∗ ∗ †

† ap↑
i v ap↑ + up a−p↓ = −εp e1 † (12.242)
∂t p a−p↓
and
† †
iγ̇−p↓ = −εp γ−p↓ . (12.243)
These equations of motion imply that the diagonalized form for the reduced Hamiltonian is
† †
Heff = εp γp↑ γp↑ + γ−p↓ γ−p↓ . (12.244)
p
237 12.13 Thermodynamics
(a) (b)
p p
p
F
p p
F F
Fig. 12.18 Quasi-particle spectrum in the vicinity of the Fermi surface in the (a) superconducting and (b) normal states. The value
of p at the Fermi level is chosen to vanish. Consequently, the minimum excitation energy required at the Fermi
surface is the gap energy.
Consequently, creating any quasi-particle, regardless of its spin and momentum, costs an
energy εp . To reiterate, the quasi-particles of BCS theory are not electrons. Rather they are
spinful entities with no well-defined charge, in contrast to singlet Cooper pairs, which have
charge 2e but no spin. In this sense, spin–charge separation can be thought to obtain in a
BCS superconductor. A plot of the quasi-particle energies is shown in Fig. 12.18. As is
evident, adding or subtracting a particle costs at least the minimum energy p .
12.13 Thermodynamics
We are now in a position to calculate the thermodynamic properties of a superconductor.

Of special interest are the heat capacity, Cv , and the Helmholtz free energy, the former
of which has a discontinuity between the normal and superconducting states. To calculate
these quantities, we need to express our equations of motion in a manner that is amenable to
extracting finite temperature information. This is most easily done by rewriting the model
Hamiltonian in terms of the quasi-particle operators. To proceed, we invert the quasi-particle
operators and solve for the electron operators,
†
ap↑ = u∗p γp↑ + vp γ−p↓ (12.245)
and
†
a−p↓ = −vp γp↑ + u∗p γ−p↓ . (12.246)
We will show that the thermal average

†
γpσ γpσ = npσ , (12.247)
where npσ is the Fermi–Dirac distribution function evaluated at the quasi-particle energy,
εp . We will assume that the entropy associated with such quasi-particle excitations is given
by the non-interacting form

S = −kB npσ ln npσ + 1 − npσ ln 1 − npσ (12.248)
p,σ
presented in Chapter 2. Coupled with Eq. (12.244), the Fermi-Dirac distribution follows
necessarily. Nonetheless we will establish the result explicitly.
We evaluate first the Helmholtz free energy, F = E − T S, where E = H − μN. We
need to express the internal energy at the mean-field level,

E = p a†pσ apσ + Vpp b†p bp , (12.249)
p,σ p,p
in terms of the γp s. The thermal average in these expressions is to be interpreted as a trace

over all states of the system. An arbitrary excitation above the ground state is represented
by
|excited state = γp†i σi |BCS (12.250)

i
and hence contains a definite number of excited quasi-particles. As a consequence, any

average of the form γ γ identically vanishes. From Eqs. (12.245) and (12.246), it follows
that
† †
a†pσ apσ = up γp↑ + vp∗ γ−p↓ u∗p γp↑ + vp γ−p↓
p,σ p
† †
+ −vp∗ γp↑ + up γ−p↓ −vp γp↑ + u∗p γ−p↓
p
2 2 2
= up − vp np↑ + n−p↓ + 2 vp . (12.251)
p
In the ground state, we showed previously (see Eq. (12.171)) that bp = u∗p vp . At finite
temperature, the average of the pair operator
† †
bp = a−p↓ ap↑ = − vp γp↑ + u∗p γ−p↓ u∗p γp↑ + vp γ−p↓

= vp u∗p 1 − n−p↓ − np↑ (12.252)
contains the thermal factors that reflect the pairing of electrons above and below the Fermi
level. Consequently,

Vpp b†p bp = Vpp up u∗p vp∗ vp 1 − np↑ − n−p↓ 1 − n−p ↓ − np ↑ , (12.253)
p,p p,p
239 12.13 Thermodynamics
and the total Helmholtz energy is

2 2 2
F = p up − vp np↑ + n−p↓ + 2 vp
p

+ Vpp 1 − n−p↓ − np↑ 1 − n−p ↓ − np ↑ up vp∗ u∗p vp
p,p

+ kB T npσ ln npσ + 1 − npσ ln 1 − npσ . (12.254)
p,σ
Minimizing F with respect to up and vp , we obtain the standard result that up =

1/2 1/2
1 + cos θp /2 and vp = 1 − cos θp /2 with cos θp = p /εp . In this case,
however, the gap is temperature dependent through

p = − Vpp bp
p

=− Vpp 1 − np ↑ − n−p ↓ u∗p vp , (12.255)
p
where sin θp = p /εp . Varying F with respect to np↑ , we find that

∂F 2 np↑

= 0 = p up − vp
2 + kB T ln + 2 p up vp∗ , (12.256)
∂np↑ 1 − np↑
holding the coherence factors up and vp fixed. The condition that is now imposed on np↑ ,
namely

kB T ln n−1
p↑ − 1 = p cos θp + 2 p up vp∗
= p cos θp + p sin θp
= εp , (12.257)
is simply the requirement that np↑ is the Fermi–Dirac distribution evaluated at the quasi-
energy εp . The Fermi–Dirac distribution is the most likely distribution of the quasi-particle
excitations at finite temperature.
The free-energy calculation is now straightforward if we use the mean-field value of the
gap, Eq. (12.255):
2 2 2 sin θp
E = p up − vp np↑ + np↓ + 2 vp − p 1 − 2np
p p
2

θp sin θp
= 2p cos θp np + sin2 − p 1 − 2np . (12.258)
p
2 2

Coupled with the fact that ln 1/npσ − 1 = εp /kB T , we simplify the entropy term to

T S = −2kB T βεp np + ln 1 + e−βεp . (12.259)
p
The total free energy is then

θp

sin θp

F = 2p cos θp np + sin2 − p 1 − 2np
p
2 2

+ 2kB T βεp np + ln 1 + e−βεp . (12.260)
p
The evaluation of this expression is left as a homework problem (see Problem 12.10).
The heat capacity,

∂S
Cv = T
∂T V
∂npσ npσ
= −kB T ln
p,σ
∂T 1 − npσ
∂npσ
= εp , (12.261)
p,σ
∂T
contains the temperature derivative

∂np n2p εp ∂εp βεp
= − e , (12.262)
∂T kB T T ∂T
which is compounded by the implicit temperature dependence of the gap, p . From the
definition of εp , it follows that
∂εp ∂εp ∂ p p ∂ p 1 ∂ 2p
= = = . (12.263)
∂T ∂ p ∂T εp ∂T 2εp ∂T
Because ∂εp /∂T vanishes in the normal state, we write the difference between the heat
capacity in the superconducting and normal states at Tc as
S ∂ 2p
Cv − CvN T =Tc = −β n2p eβεp
p
∂T
∂np ∂ 2p
= . (12.264)
p
∂εp ∂T T =Tc
In the vicinity of T = Tc , the gap is well-approximated by 2 ∼ |1 − T /Tc |. We find then

that ∂ 2 /∂T ∼ 1/Tc , and the discontinuity in the heat capacity is identically
N (F )
CvS − CvN = (12.265)
Tc
at T = Tc . The positive sign associated with this difference indicates that the heat capacity
in the superconducting state just below Tc exceeds CvN .
The second result we seek is the exponential fall-off of CvS just below Tc . To obtain this
behavior, we analyze Eq. (12.261) in the limit that T → 0. In this limit, ∂ 2p /∂T → 0,
241 12.14 Experimental applications
np → e−βεp , and the heat capacity is dominated by

lim Cv = 2β 2 kB εp2 np (1 − np )
T →0
p
∞ √
de−β (T =0)+
2 2
2β 2 kB N (F ) 2 (T = 0)
∞0
2 −β 0
de−β /2 0
2
2β kB N (F ) 0 e
2
0

π −β 0
= N (F ) kB β 3/2 5/2
0 e , (12.266)
2
which decays exponentially as advertised with 0 = (T = 0). The exponential fall-off
is dominated by the gap at T = 0. In simplifying the integrals leading to this result, we
used the fact that the integrals are largest when .
12.14 Experimental applications
One of the key successes of the BCS theory is the ease with which physical observables,
such as the phonon attenuation rate, can be calculated. In fact, it was the calculation
of the spin-lattice relaxation time using the BCS theory and its subsequent experimental
confirmation that led to the wide acceptance of BCS theory. In this section, we perform three
of these calculations explicitly: (1) electromagnetic absorption, (2) ultrasonic attenuation,
and (3) the spin-lattice relaxation rate. Appearing in each of these calculations are particular
combinations of the coherence factors, up and vp . We will see that each combination of the
coherence factors can generate completely different physics. That the physical properties
are so intimately connected with the coherence factors is a hallmark of the BCS pairing
hypothesis. In each of these calculations, we will treat the coherence factors as being real.
As we have seen, for the simple model potential we have chosen, Eq. (12.41), the coherence
factors are in fact real.
12.14.1 Electromagnetic absorption

In a BCS superconductor, absorption of electromagnetic radiation is suppressed for fre-
quencies less than /. We have now the machinery to derive this result explicitly. In the
electromagnetic problem, the perturbation term is obtained by replacing the momentum
with p − eA(r, t )/c, where A(r, t ) is the vector potential. To lowest order in A(r, t ), the
perturbation is of the form, p · A + A · p. We will require that the vector potential satisfy
the transversality condition, ∇ · A = 0. As a consequence, the second-quantized form of
the perturbation for absorption of an electromagnetic wave with frequency ωq is
e
Habs = − A · pa†p+qσ apσ . (12.267)
mcV p,σ
q q
p p+q p+q p
(a) (b)
−p
p+q
(c)
†
Fig. 12.19 Quasi-particle excitations arising from the matrix element for electromagnetic absorption: (a) γp+q γp , (b) γp† γp+q ,
† †
and (c) γp+q γ−p . The wavy line represents the photon interaction.
We must compute a matrix element of the form

e
f |Habs | i = − A · pMifσ , (12.268)
V mc pσ
where

Mifσ = f a†p+qσ apσ i (12.269)
and |i and | f are the initial and final quasi-particle states, respectively. As in the calculation
of the heat capacity, we rewrite this matrix element,
†
Mifσ = f up γp† ↑ + vp γ−p ↓ up γp↑ + vp γ−p↓ i
σ

+ f − v−p γ−p ↑ + u−p γp† ↓ − v−p γ−p↑
†
+ u−p γp↓ i ,
in terms of the quasi-particle operators. There are a number of distinct kinds of

quasi-particle excitations that arise from the terms above. These are illustrated in
Fig. 12.19.
At T = 0, there are no quasi-particles in the ground state. Hence, only the last process
shown in Fig. 12.19, which creates two quasi-particles, survives at T = 0. We will specialize
to T = 0, as in this case the essential physics is easily unearthed. We find that
† †
† †

Mif↑ + Mif↓ = up+q vp f γp+q↑ γ−p↓ i − u−(p+q) v−p f γp+q↓ γ−p↑ i (12.270)
and
e † † † †
f |Habs | i = − (A · p) f |up+q vp γp+q↑ γ−p↓ − u−(p+q) vp γp+q↓ γ−p↑ |i .
mcV p
(12.271)
To simplify this expression, we note that if we shift the momentum in the second term by
p → − (p + q), then p · A → −A · (p + q) = −A · p. The last equality follows because
the vector potential is transverse to the electromagnetic field. We find that
−e † †

f |Habs | i = (A · p) up+q vp − up vp+q f γp+q↑ γ−p↓ i. (12.272)
mcV p
In the matrix element in Eq. (12.272), we write the final state as | f = |p ↑, p ↓.
The quasi-particle matrix element reduces to
† †

f γ p+q↑ γ−p↓ i = δp ,−p δp ,p+q . (12.273)
The total absorption rate per unit volume,

2π e 2 2
s = (A · p)2 δp+p ,q up vp − up vp δ εp + εp − ωq , (12.274)
mcV p,p
involves the coherence factors

6 6 6 6 2
2 εp + p εp − p εp − p εp + p
up vp − up vp = −
2εp 2εp 2εp 2εp
1
= εp εp − p p − 2 . (12.275)
2εp εp
We convert the sums to integrals,

1 dp
→ N (F ) dp , (12.276)
V p
4π
where dp represents an integral over the solid angle. The absorption rate is now

2π εε − − 2
s = N (F )2 dd δ ε + ε − ωq
2εε
e 2
d d
× δ p + p − q p·A . (12.277)
4π 4π mc
We note first that symmetry about the Fermi surface dictates that we must obtain the same
result under the transformation ↔ −. As a consequence, the terms vanish. If we
are interested primarily
in the contribution from the vicinity of the Fermi surface, then
we can set |p| = p = pF . The angular terms are trivial and yield (pF Ae)2 /3(mc)2 . The
remaining integration,
∞
8π pF Ae 2 ∞ εε − 2
s = N (F )2 d d δ ε + ε − ωq , (12.278)
3 mc 0 0 2εε
2 q
Fig. 12.20 The ratio of the electromagnetic absorption in the superconducting state to that in the normal state. The x-axis is the
frequency and is the gap energy.
is over the free particle energies. The factor of 4 arises from the conversion of the lower
limit of integration from −∞ to zero. Because s > 0, the integral must be positive for all
ε and ε . We see immediately that εε > 2 or, equivalently, that ωq > 2 . Thus, s must
vanish for ωq < 2 . This is the principal result of this section.
We can obtain a more quantitative prediction by changing variables and noting that
because ε = ( 2 + 2 )1/2 , εdε = d. This variable change introduces a lower limit cut-off
∞ ∞

εε − 2
s = const. dε dε δ ε + ε − ωq 1/2 1/2 (12.279)
ε 2 − 2 ε2 − 2
on the range of integration on ε and ε . The δ-function constraint, ε − ωq = −ε, allows
us to rewrite the ε-integral as
ωq −

ε ωq − ε − 2
s = const. dε 1/2 2 1/2 , (12.280)

ε2 − 2 ωq − ε − 2
where we approximated the upper limit by the lower bound ωq − . When = 0, the
integrand reduces to a constant and s → const.ωq . If we refer to the = 0 value of s
as N and plot s /N , we obtain the result shown in Fig. 12.20. The form we have derived
here for s agrees well with experiment.
12.14.2 Ultrasonic attenuation

Consider now the problem of a beam of phonons impinging on a superconductor. We have
advertised that a superconductor should be transparent to phonon absorption unless the
energy of the impinging phonon beam exceeds 2 . To show this, we rewrite the electron–
phonon Hamiltonian, Eq. (12.23), in terms of the quasi-particle operators,

He-ph = Mq bq + b†−q †
up+q γp+q↑ + vp+q γ−(p+q)↓ †
up γp↑ + vp γ−p↓
p,q

† †
+ u−(p+q) γp+q↓ − v−(p+q) γ−(p+q)↑ u−p γp↓ − v−p γ−p↑ , (12.281)
using Eqs. (12.245) and (12.246). There are two distinct kinds of term that arise from the
quasi-particle–phonon coupling: (1) single quasi-particle excitations and (2) quasi-particle
pair creation and annihilation. The single-quasi-particle creation and annihilation processes,

Heqp
-ph = Mq bq + b†−q up+q up γp+q↑
† †
γp↑ + vp+q vp γ−(p+q)↓ γ−p↓
p,q

† †
+ u−(p+q) u−p γp+q↓ γp↓ + v−(p+q) v−p γ−(p+q)↑ γ−p↑ , (12.282)
arise from a conjugate γp , γp† pair. The additional scattering channel,

Hepair
-ph = Mq bq + b†−q up+q vp γp+q↑ † †
γ−p↓ †
− u−(p+q) v−p γp+q↓ †
γ−p↑
p,q

+ vp+q up γ−(p+q)↓ γp↑ − v−(p+q) u−p γ−(p+q)↑ γp↓ , (12.283)
is mediated by quasi-particle pair creation and annihilation.

Both types of process are expected to occur when phonons are excited in a superconductor.
However, as in the electromagnetic absorption problem, pair creation and annihilation is
expected only when ωq > 2 . Because our goal is to show that absorption of phonons
is attenuated at frequencies less than 2 , we can ignore the pair scattering terms. We
specialize then to the ωq < case in which only single-particle scattering contributes to
the transition amplitude. The two representative classes of emission and absorption terms
are shown in Figs. 12.21(a) and 12.21(b). Each term contributes identically for ↓ and ↑
electron states. First, consider emission. The operator describing emission of a phonon in
a single quasi-particle exchange is

qp
Vemiss = Mq b†−q up+q up γp+qσ
† †
γpσ + v−(p+q) v−p γpσ γp+qσ , (12.284)
where we have set −p → p and −p − q → p in the second term in Eq. (12.282). From the
anticommutation relation for the γ s, we can rewrite the emission term as
†
qp
Vemiss = Mq b†−q up+q up − v−(p+q) v−p γp+qσ γpσ + v−(p+q) v−p δp,p+q . (12.285)
The last term requires that p = p + q or, equivalently, q = 0. However, Mq is linear in q and
hence vanishes when q = 0. In the emission process, the final composite electron–phonon
state must contain the phonon with momentum q and the quasi-particle with momentum
p + q and spin σ . The quasi-particle with momentum p is absent from this state. The initial
q −k
k+q k k+q
(a)
−k q
k+q k k+q
(b)
Fig. 12.21 (a) Emission and (b) absorption processes contributing to the ultrasonic attenuation. The wavy line represents the
quasi-particle–phonon interaction.
state is in general some determinantal state containing any number of quasi-particles, but
in particular the quasi-particle with momentum p and spin σ . Let |p + qσ, q and |pσ be
the final and initial states, respectively. The emission matrix element simplifies to

p + qσ, q|Vemiss
qp
|pσ = Mq up+q up − v−(p+q) v−p Nq + 1 np 1 − np+q , (12.286)
where np is the occupation of quasi-particles in momentum state p. Consequently, when
both ↑ and ↓ spins are included, the Fermi golden rule rate for emission is
2 2
emiss
qp
= Nq + 1 |Mq |2 up+q up − vp+q vp
p

× np 1 − np+q 2π δ ωq + εp − εp+q . (12.287)
Analogous arguments can be applied to absorption of a phonon. There are two principal
differences in the form of the transition rate. First, the factors of Nq + 1 are replaced by Nq .
This change arises because, in the case of absorption, the final state has one fewer phonon.
Second, the initial state and final states are of the form |p + qσ and |pσ, q , respectively.
The rate of absorption between all such initial and final states,
qp 2 2
abs = Nq |Mq | up+q up − vp+q vp np+q 1 − np 2π δ ωq + εp − εp+q ,
p
(12.288)
is a sum of the individual rates, which we treat at the level of the Fermi golden rule.
The difference between the emission and absorption rates,

qp
q = abs − emiss
qp

2 2 2
= Mq up+q up − vp+q vp 2π δ ωq + εp − εp+q
p

× Nq np+q − np − np 1 − np+q , (12.289)
contains a part which depends on the phonon occupation (through Nq ) and a term that does
not. In combining the absorption and emission terms, we used the time-reversal invariance
of the the coherence factors. The part independent of the phonon occupation defines the
effective spontaneous emission rate, whereas the first term defines the total rate at which
Nq phonons are absorbed. We define then
dNq 2 2 2
=− Mq up+q up − vp+q vp 2π δ ωq + εp − εp+q Nq np+q − np
dt p
= −αq Nq (12.290)
as the net acoustic attenuation rate.

To evaluate αq , we convert the sum to an integral

4π 2 2
αq = Mq g (0) dp dp up up − vp vp np − np

dp dp
× δ ωq + εp − εp δ p − p − q (12.291)
4π 4π

in which momentum conservation is ensured by the angular factor δ p − p − q . To
simplify Eq. (12.291), we note that the angular factors are identical in the normal and
superconducting states. Because the gap is assumed to be isotropic, we can separate out
the angular dependence. Let us call this factor A . We need now an expression for the
coherence factors:
2 1
up up − vp vp = εp εp + p p − 2 , (12.292)
2εp εp
where we have assumed that the gap is a constant, = p = p . We note that np and εp
are both even in p . Hence, integration over the linear p p factors vanishes because the
limits are even. As a further simplification, we change the variable of integration from p
to εp by noting that dεp /dp = εp /p . We are left with
∞ ∞
8π 2 2 εp εp 2
αq = Mq g (0)A dεp dεp 1−
p p εp εp

× np − np δ ωq + εp − εp . (12.293)
The quantity N (0)dp /dεp = N (0)εp /p is roughly the density of quasi-particle exci-
tations. At the Fermi surface, p → 0 and dp /dεp → ∞. However, over the range of
αqS
αqN
Tc
Fig. 12.22 Ratio of the ultrasound attenuation in the superconducting state to that in the normal state, as predicted by
Eq. (12.297). The fall-off of this ratio in the superconducting state is a signature of Cooper pair formation.
integration, εp /p is finite. In the limit that ωq , εp ≈ εp . Consequently,

εp εp 2 εp2 εp2 − 2
1− → 2 = 1. (12.294)
p p εp εp p εp2
It is certainly valid to work within this limit, because we are considering phonon absorption
as a result of single-quasi-particle excitations only. Evaluating the remaining integrand at
εp = εp + ωq , we find that
∞
8π 2 2
αq = Mq g (0)A dεp n εp + ωq − n εp

8π 2 2
− Mq g (0)A ωq n( ). (12.295)

In the normal state, = 0 and resultantly,
8π 2 2 4π 2 2
αqN = − Mq g (0)A ωq n(0) = − Mq g (0)A ωq . (12.296)

The ratio of αqs /αqN reduces to
αqs 2
= . (12.297)
αqN eβ (T ) +1
At T = Tc , = 0, and this ratio yields unity, as illustrated in Fig. 12.22. The calcula-
tion (S1964) of the ultrasonic attenuation rate is in excellent agreement with experimental
results. Here again, we see that it is the gap that ultimately determines the absorption of
sound waves in the superconducting states. This calculation is valid, of course, for longitu-
dinal phonons only. Unlike the longitudinal case in which Mq is the same in the normal as
well as in the superconducting state, Mq for transverse phonons changes drastically from
S
αq
αqN
Tc
Fig. 12.23 Ratio of the superconducting and normal state ultrasound absorption rates for transverse phonons as a function of
temperature T .
the normal to the superconducting state. In this case, the magnetic dependence of Mq leads
to a discontinuous drop in the attenuation rate at T = Tc , as depicted in Fig. 12.23 (S1964).
12.14.3 Spin-lattice relaxation

The final calculation we present using the BCS coherence factors is that of the spin-lattice
relaxation time, T1 . For historical reasons, this is a key calculation because the prediction
that a peak should occur in 1/T1 just below Tc predated the experiment by Hebel and
Slichter (HS1959). In fact, experimental confirmation of the enhancement of 1/T1 just
below Tc was the crowning evidence that led to the wide acceptance of the BCS theory. To
define T1 , we consider applying a magnetic field to a collection of nuclear spins. Once the
field is turned off, the spins will relax back to achieve the equilibrium magnetization. The
time for this process to occur is referred to as T1 . In insulators, it is the coupling between
the nuclei and electronic impurities that causes the nuclei to relax. Left alone, nuclei cannot
achieve an equilibrium spin temperature. The lattice temperature must be communicated
to the nuclei by some agent. In a metal, it is the conduction electrons that provide the
coupling. Because the electrons in a superconductor are locked into a coherent state, we
expect T1 to be drastically altered. Consider a nucleus with magnetic moment SN located at
the origin, r = 0. The net electron spin at r = 0 is given by the local electron spin density,
S = † (0)σ (0)/2, at the origin, where (0) is the electron wavefunction at r = 0 and
σ are the Pauli spin matrices. The Fermi contact,
8π
Hint = μB SN · S, (12.298)
3
defines the coupling between a nucleus and an electron at r = 0. In Eq. (12.298), μB is the
Bohr magneton.
To apply this coupling to a calculation of T1 in a superconductor, we simply need to
express S in terms of the quasi-particle operators. To do this, we introduce the spinor field
operators

1 ip·r ap↑
(r) = √ e . (12.299)
V p ap↓
In spinor notation, the local electron spin density at the origin becomes

† † a
S= ap ↑ ap ↓ σ p↑ . (12.300)
2V
ap↓
p,p
We now express the Fermi contact interaction in terms of (r). Because the product
S1 · S2 = S1z S2z + 2 S1+ S2− + S1− S2+ , we write the Fermi contact,
4π z †
Hint = SB SN ap ↑ ap↑ − a†p ↓ ap↓ + 2S+ † − †
N ap ↓ ap↑ + 2SN ap ↑ ap↓ , (12.301)
3V p,p
in the familiar Kondo form, where μB = μB /.

We are particularly interested in those terms in which the nuclear spin is flipped. Hence,
we can focus on either the S− + + +
N or the SN term. Consider SN . Let Hint be that part of the
interaction in which the nuclear spin is flipped up. Here again, we introduce the quasi-
particle representation of
a†p ↓ = u−p γp† ↓ − v−p γ−p ↑ (12.302)
and
†
ap↑ = up γp↑ + vp γ−p↓ , (12.303)
+
so that Hint is transformed to
4π +
+
Hint = μ S u−p up γp† ↓ γp↑ − v−p vp γ−p ↑ γ−p↓
†
3V B N p,p

− up v−p γ−p ↑ γp↑ + u−p vp γp† ↓ γ−p↓
†
. (12.304)
The first two terms describe processes in which quasi-particles of opposite spin are created
and annihilated, while the latter terms contain quasi-particle pair production or annihilation.
Typically, nuclear Zeeman energies are small relative to the quasi-particle gap energy.
Hence, if we focus on the frequency range ω ≈ μB H , with H the external magnetic field,
only the former terms are relevant. We have reduced our effective interaction to
4π +
+
Hint μB SN u−p up γp† ↓ γp↑ + v−p vp γ−p↓
†
γ−p ↑ . (12.305)
3V p,p
Matrix elements of this quantity will be non-zero if the initial state is of the form |p ↑, SzN
and the final state is |p ↓, SzN + 1. Let αN = SzN + 1|S+ N |SN . The resultant matrix element
z

pp 8π
SLR = μ αN u−p up 1 − np ↓ np↑ + v−p vp 1 − n−p↓ n−p ↑
3 B
4π
= μB αN u−p up + v−p vp 1 − np ↓ np↑ (12.306)
3
leads to the nuclear-spin lattice relaxation rate
32π 3 2 2
T1−1 = SB |αN |2 u−p up + v−p vp 1 − np ↓ np↑ δ εp − εp . (12.307)
9 p,p
At this level of theory, we are ignoring, of course, the Kondo effect. The Kondo regime
of a magnetic impurity in a superconductor is a subtle problem indeed, which we will not
consider here.
The procedure from here on is now standard: (1) convert the sum to an integral by
introducing the density of states, (2) evaluate the coherence factors, and (3) cancel all
integrals linear in the bare energy, . From the ultrasonic attenuation calculation, we have
that
2 1
up up + vp vp = εp εp + p p + p p . (12.308)
2εp εp
Here again, we assume that the gap is constant and the relaxation rate reduces to
∞ ∞
64π 3 2 εp εp + 2
T1−1 = μB |αN |2 g2 (0) dp dp np 1 − np δ εp − εp
9 0 0 εp εp
∞
3
64π 2 ε +
2 2
= μB |αN |2 g2 (0) dεn(ε) (1 − n (ε)) 2 . (12.309)
9 ε + 2
At T = 0, the product n(1 − n) = 0 because n = 1 at T = 0.
At T = 0, there is a divergence when ε = ± . In the vicinity of ± , we write the
integrand as
∞ 2
64π 3 2 ε + 2
T1−1 → μB |αN |2 g2 (0) n (ε) (1 − n (ε)) dε
ε −
9 2 2
∞
ε+
∝ dε, (12.310)
ε−
which is logarithmically divergent. The coefficient of the ln-divergence is proportional to

the gap, . This implies then that the ln-divergence starts at T = Tc where the gap begins
to arise. In experiments, the gap is anisotropic as a result of its momentum dependence.
Consequently, the divergence of T1 is smeared out slightly over a range comparable to the
average value of . Hebel and Slichter (HS1959) were the first to observe this behavior,
which is illustrated in Fig. 12.7.
The ln-divergence we have predicted at this low order in perturbation theory might be
removed, or at least broadened, by the summation of all higher-order terms. Hence, it is
worthwhile to investigate precisely how robust the ln-divergence is. The full series for the
electron–nucleus interaction is
+ +
+ f Hint p pH i

int = f Hint i +
full int
+ 0(H 3 ) + · · · . (12.311)
p
E − ε p + iη
In the context of the local-moment problem, we defined the Green function to be

|p p|
G(E + iη) = . (12.312)
p
E − εp + iη
Schematically, the perturbation series is of the form

+ + +
int
full
= f |Hint |i + f |Hint G (E + iη) Hint |i + · · ·

H +
= f int i . (12.313)
1 − H + G (E + iη)
int
To determine the divergent part of int

full
, we convert the sum in the Green function to an
integral,

1
G (E + iη) = N (0) dp
E − εp + iη
∞
dp dεp
= N (0) . (12.314)
dεp E − εp + iη
We have shown previously that N (0)dp /dεp defines the density of states,
N (0)εp εp
ρ εp = = N (0) , (12.315)
εp2 − 2p p
which is divergent at εp = p . Consequently,

∞
1
lim G (E + iη) = ρ εp P − iπ δ E − εp dεp
η→0 E − εp
= ReG − iπ ρ(E ), (12.316)
and the perturbation series becomes

) *
+
Hint
int = f
full
+
i . (12.317)
1 − Hint ReG + iπ Hint ρ(E )
In the vicinity of E = , the denominator of Eq. (12.317) is dominated by the ρ (E ) term.

In this limit, int
full
→ 1/ (iπ ρ (E )). The square of this quantity exactly cancels the divergent
term, 1/(εp − 2 ), in our previous expression for the relaxation rate.
2
253 12.15 Josephson tunneling
To see this more clearly, we rewrite the relaxation rate as

∞
−1 64π 3 2 ρ (ε) 2
T1 = μ |αN | g (0)
2 2
dε n(ε) (1 − n (ε))
9 B ε
2
ε + 2
× 2 , (12.318)
+ + 2 2
1 − Hint ReG + π 2 |Hint | ρ (ε)
which reduces to
∞
ε2 + 2
T1−1 ∝ dε (1 − n (ε)) n (ε) + 2 . (12.319)
1 + π 2 Hint g2 (0) ε2 − 2
In deriving Eq. (12.319), we ignored the real part of the Green function as it serves no
relevant purpose as far as the convergence is concerned. This expression is completely
convergent at ε2 = 2 . In fact, because 1 + π 2 |Hint |2 g2 (0) > 1, the integral does not
diverge over the complete integration range. Consequently, summing high-order terms in
the perturbation series results in a smearing of the peak in the relaxation rate immediately
below Tc .
12.15 Josephson tunneling
Consider two superconductors separated from one another by a thin insulating barrier.
Naively, we would expect no appreciable transport of charge between the two superconduc-
tors in the absence of an applied voltage, save possibly for single quasi-particle tunneling
through the insulating barrier. Josephson (J1962) showed that this naive picture is not cor-
rect. In particular, he proved that in the absence of an applied voltage for a sufficiently
thin barrier, Cooper pairs flow coherently between the two superconductors, thereby es-
tablishing a supercurrent through the barrier. Further, the transport of Cooper pairs across
the barrier does not result in the creation of quasi-particles in either superconductor. When
an applied voltage is present, the supercurrent oscillates with a well-defined period. The
essence of both of these effects, dc and ac Josephson tunneling, rests in the phase coherence
that obtains in the superconducting state.
We focus first on the dc Josephson effect. Consider two superconductors separated by a
thin insulating barrier. Let HT represent the Hamiltonian for single-particle tunneling across
the thin barrier. The specific form of this term is not essential here. The only important
feature is that HT transfers only one electron at a time. We assume at the outset that there
is no voltage difference between the two superconductors, and hence they are at the same
chemical potential. We can derive the Josephson effect by making an analogy with electron
transport in a 1d periodic chain in the tight-binding approximation. In this approximation,
a single orbital is placed on each lattice site and a hopping term mediates transport among
nearest-neighbor sites. In such a system, no energy is required to transport an electron
across m lattice sites. Likewise, it requires no energy to translate m Cooper pairs across the
barrier in the absence of an external voltage differential between the two superconductors.
Let |2m represent the many-body state when 2m Cooper pairs are transferred across the
barrier. The degeneracy of these states is split by the single-electron tunneling term, HT .
Consequently, we expand the total state of our system as a linear combination

|φ ≡ |φ = e2imφ |2m (12.320)
m
over all such pair states. The phase φ plays the role of the wavevector k in the 1d periodic
tight-binding model. As we have established earlier, the particle number, 2m, and the phase,
φ, are conjugate variables.
To compute the energy shift as a result of the tunneling processes, we employ perturbation
theory. The first-order term, φ|HT |φ, vanishes identically because φ is a sum of all pair
states and HT is a one-body operator. Consequently, the first non-zero term appears in
second order. Let
H
T |II| H

(2) = H
T (12.321)
T
E − EI
represent the tunneling operator at second order with EI the energy of the intermediate
state, |I. The second-order correction to the energy,

(2) |φ
Eφ = φ|HT

=

(2) |2m
e2iφ(m−m ) 2m |HT
m,m

=
(2) |2(m + 1) + e−2iφ 2m|H
e2iφ 2m|H
(2) |2(m − 1) , (12.322)
T T
m
is a sum of all matrix elements that differ by a single Cooper pair. We have assumed
that φ|φ = 1. To simplify this expression, we note that the energy of the intermediate
state involves a particle–hole excitation, and hence EI must exceed E by at least 2 .
Consequently, E − EI < 0. If we regard the tunneling term to be purely real, we simplify
the energy shift to
J0
Eφ = − cos 2φ, (12.323)
2
with

J0 = 4
(2) |2(m + 1) .
2m|H (12.324)
T
m
The minus sign in the energy shift arises from the sign of the excitation energy. From the
Hamilton equation, Eq. (12.170), it is clear that if the energy shift depends on the phase,
255 Problems
then the pair number fluctuates on either side of the barrier. This fluctuation is due entirely
to the tunneling processes. We calculate the pair current directly,

d2m dEφ
I = 2e = 2e = 2eJ0 sin 2φ, (12.325)
dt dφ
by differentiating the energy shift with respect to the phase. Consequently, in the absence
of an applied voltage, a dc supercurrent flows across the barrier. The value of the current
ranges from −2eJ0 to 2eJ0 . A supercurrent of this form was first observed by Anderson and
Rowell (AR1963). If a potential difference V exists across the barrier, then a term of the
form 2mV must be added to the Hamiltonian. Consequently, from Hamilton’s equations,
Eq. (12.170), the phase fluctuates in time according to
dφ
= 2eV. (12.326)
dt
Together, these two equations, Eq. (12.325) and Eq. (12.326), completely determine the
behavior of the supercurrent across the barrier. To illustrate, consider the simplest case in
which the voltage V is a constant in time. In this case, the phase φ varies linearly with time
and, as a consequence, the current oscillates as sin(2eV t/). Hence, an alternating current
flows with a frequency of 2eV/.
Summary
We have shown that the pairing hypothesis of BCS is sufficient to account for all relevant
experimental observables of low-temperature superconductors. In fact the BCS pairing
mechanism is the only account available currently to describe the transition to a supercon-
ducting state. In contrast to low-Tc materials, superconductivity in the cuprates originates
from doping an insulator. Further, the insulator possesses a partially-filled band and hence
falls into the class of Mott insulators in which an absence of transport originates from strong
electron repulsions. Consequently, we know a priori that we are not justified in starting
from Fermi liquid theory to describe even the normal state properties. Simply stated, the
deep phenomenology of the cuprates lies in the physics of doped Mott insulators. Whether a
theory as succinct and crystal clear as the BCS account can be formulated for such systems
remains to be seen.
Problems
12.1 Within the Ginsburg–Landau phenomenological approach, determine the form of

the free energy density when a magnetic field is present. Show that the free energy
difference between the superconducting and normal states is given by

Hc2 (T )
FS − FN = − . (12.327)
8π
√
12.2 Writing the Ginsburg–Landau wavefunction as ψ (r) = n(r)eiθ (r) , show that the
current density in terms of the variables θ and n(r) is given by

e eA
j= ∇θ − n(r). (12.328)
m c
Now assume that in the bulk of a material, the current density vanishes. As a conse-
quence, ∇θ = eA. Integrate both sides of this expression around a closed loop in a
superconducting ring and show that the resultant magnetic flux enclosed is quantized.
What is the correct value of e for a superconductor?
12.3 Use second-order perturbation theory directly to show that the electron–phonon
interaction is negative and given by the second term in Eq. (12.34).
12.4 Redo the Cooper pair instability calculation for triplet pairing between the electrons.
12.5 Evaluate r2 for a singlet Cooper pair.
12.6 In the problem of the instability of the superconducting state in the presence of the
BCS pairing interaction, determine the form of the growth rate of the pair amplitude
as T → Tc .
12.7 Evaluate the commutator [bk , b†k ], where the bk s are the Cooper pair annihilation
operators. What does the lack of commutativity of the Cooper pair creation and
annihilation operators mean?
12.8 Calculate the average number of particles in a superconductor. Let | represent the
BCS pair state. Show that the average value of the number operator, N, is given by
†
|N| = | akσ akσ | = 2 |vk |2 (12.329)
k,σ k
in the pair state. Also evaluate the fluctuation (N − N)2 . You should obtain a
simple result involving uk and vk only. For what special value of uk and vk is the
fluctuation maximized? Interpret your result.
12.9 So far we have ignored any spatial inhomogeneities in the gap. Consider a gap of the
form q = 0 e2iq·r , where q pF . Find the new self-consistent condition for 0 .
At T = 0, show that is independent of q for q < qc ≈ 0 /vF . Near Tc , expand
the gap equation to find that
2
(T ) 8π 2 2 2 pF
≈ (1 − T /Tc ) − q2 . (12.330)
kB Tc 7ζ (3) 3 mkB Tc
Then determine the critical value of q that makes the gap vanish.
12.10 Evaluate the sums explicitly in Eq. (12.260) and show that for T /Tc 1, FN − FS ∝
1 − (T /Tc )2 .
12.11 An Anderson-type impurity is placed in a superconductor. You are to formulate this
problem and develop a criterion for local moment formation. There are a number
of assumptions that can be applied. First, when you transform to the quasi-particle
basis, ignore all terms that do not conserve spin and particle number. The problem
257 References
should now be straightforward. You should be able to redo the Anderson problem
completely. Discuss clearly when the local moment exists and when it does not.
References
[A1957] A. A. Abrikosov, Sov. Phys. JETP 5, 1174 (1957).

[AR1963] P. W. Anderson and J. M. Rowell, Phys. Rev. Lett. 10, 230 (1963).
[BCS1957] J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev. 106, 162 (1957); 108,
1175 (1957).
[BG1990] G. Benefatto and G. Gallavotti, J. Stat. Phys. 59, 541 (1990).
[C1956] L. N. Cooper, Phys. Rev. 104, 1189 (1956).
[GL1950] G. V. L. Ginsburg and L. D. Landau, J. Exptl. Theor. Phys. (USSR) 20, 1064
(1950).
[GF1987] M. Gurvitch and A. T. Fiory, Phys. Rev. Lett. 59, 1337 (1987).
[HS1959] L. C. Hebel and C. P. Slichter, Phys. Rev. 113, 1504 (1959); L. C. Hebel, Phys.
Rev. 116, 79 (1959).
[J1962] B. D. Josephson, Phys. Lett. 1, 251 (1962).
[L1956] L. D. Landau, Sov. Phys. JETP 3, 920 (1956); 8, 70 (1959).
[L1964] A. I. Larkin, Sov. Phys. JETP 19, 1478 (1964).
[L1965] A. J. Leggett, Phys. Rev. Ser. A 140, 1869 (1965).
[P1992] J. Polchinski, arXiv:hep-th/9210046.
[SM1991] R. Shankar, Physica A177, 530 (1991).
[S1964] J. R. Schrieffer, Theory of Superconductivity (Benjamin, New York, 1964).

Superconductivity

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Superconductivity

Uploaded by

Copyright:

Available Formats

12 Superconductivity in metals

In this chapter we focus on the phenomenon of superconductivity and the Bardeen–Cooper–

12.1 Superconductivity: phenomenology

The negative value of M signals that the interior of a superconductor is diamagnetic. At

B(r) = B(0)e−r/λL , (12.9)

is an exponentially decaying magnetic induction on a length scale λL . Exponential

The current of these electrons is defined as J = −e∗ vs ns , which, combined with

We see then that as the superconducting density increases, λL decreases.

(e) Ultrasonic attenuation No damping of an impinging beam of phonons is observed if

This observation implies that electron–phonon coupling is at the heart of supercon-

12.2 Electron–phonon effective interaction

The isotope dependence of Tc confirms that electron–phonon coupling is central to the

and the electron–lattice interaction

The electron–phonon coupling constant is defined in Section 11.3. We will assume

[Ho , S] = −He-ph . (12.24)

The transformed Hamiltonian will be

We find then that

an effective two-body potential that, interestingly enough, is negative in a narrow energy

with k ,q = k − k −q . In general, k and k −q vastly exceed ωq . However, if

12.3 Model interaction

In a metal, of course, we must consider the electron–electron repulsion. We can represent

between electrons is of the form

12.4 Cooper pairs

assumptions, the integral yields

which implies that

For small Q, we can rewrite Eq. (12.58) as

is a linear function of the center-of-mass momentum. Translation of the center of mass

12.5 Fermi liquid theory

p2F − p2 < 2pq cos θ + q2 < 0. (12.63)

Because we are interested primarily in p − pF ≈ 0, we can take q to be small. We then

|ψ = |0 − ψ|1 (12.70)

|ψ = −ψ|1 = ψ|1 = ψ|0. (12.71)

However, |0 = |ψ + ψ|1. Substituting, we obtain

|ψ =  [|ψ + ψ|1] = ψ|ψ, (12.72)

ψ̄| † ≡ ψ̄|ψ̄. (12.73)

Using the fact that ψ̄1| = −1|ψ̄, we obtain that

ψ̄|ψ = 0|0 + ψ̄ψ1|1

Consequently, the trace formula is given by

With these relationships, we proceed to express the partition function,

is conveniently written in terms of creation and annihilation operators in momentum space

where all the interactions are contained in

ψσ (p) → s−1/2 ψσ (p), (12.100)

δ d (δk3 + δk4 + δ3 + δ4 ). (12.105)

δ d (δk3 + δk4 )δ d (δ3 + δ4 ), (12.106)

(a) (b) p′, E −p′, E

p″, E+E′ −p″, E−E′

where we have explicitly approximated (1 + i)2 ≈ (1 + 2i) since is strictly of O(0).

E = ±E ∓ vF (k ) ± ivF (k ) , (12.109)

where we have changed variables to b = vF (k ) , defined

We differentiate both sides of this equation,

can calculate the change in the ground state energy,

12.6 Pair amplitude

b†p = a†p↑ a†−p↓ , (12.122)

where ηp is an expansion coefficient. To determine ηp , we note that

Because |ψ  simply differs from |0  by two particles, ηp = αp .

In the Schrödinger picture, the time dependence of αp is carried in the wavefunctions. As

12.6.1 Instability: superconducting state

Substituting this expression into the pairing equation, we obtain

as the Fourier transform of the pair amplitude equation.

As before, we sum both sides of this expression to obtain

which implies that

which is only true if the imaginary part,

Exponentiation of both sides of this equation,

|ψ = |0 − ψ|1 (12.70)

|ψ = −ψ|1 = ψ|1 = ψ|0. (12.71)

However, |0 = |ψ + ψ|1. Substituting, we obtain

|ψ = [|ψ + ψ|1] = ψ|ψ, (12.72)

ψ̄| † ≡ ψ̄|ψ̄. (12.73)

Using the fact that ψ̄1| = −1|ψ̄, we obtain that

ψ̄|ψ = 0|0 + ψ̄ψ1|1

δ d (δk3 + δk4 + δ3 + δ4 ). (12.105)

δ d (δk3 + δk4 )δ d (δ3 + δ4 ), (12.106)

E = ±E ∓ vF (k ) ± ivF (k ) , (12.109)

where we have changed variables to b = vF (k ) , defined

Because |ψ simply differs from |0 by two particles, ηp = αp .

| 0 (N ) = b†p |0. (12.153)

= 0| u∗p vp bp b†p |0 = u∗p vp (12.171)

Let us first compute npσ :

when acting on |BCS ,

γp↑ |BCS = γ−p↓ |BCS = 0. (12.224)

|excited state = γp†i σi |BCS (12.250)

To determine the divergent part of int