Contents

5 Superconductivity 1
5.1 Phenomenology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
5.2 Electron-phonon interaction . . . . . . . . . . . . . . . . . . . . . . 3
5.3 Cooper problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
5.4 Pair condensate & BCS Wavefctn. . . . . . . . . . . . . . . . . 11
5.5 BCS Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
5.5.1 BCS wave function, gauge invariance, and number con-
servation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.5.2 Is the BCS order parameter general? . . . . . . . . . . . . . . . 16
5.6 Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.6.1 Bogoliubov transformation . . . . . . . . . . . . . . . . . . . . . . . 18
5.6.2 Density of states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.6.3 Critical temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.6.4 Specic heat. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.7 Electrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.7.1 Linear response to vector potential . . . . . . . . . . . 25
5.7.2 Meissner Eect. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.7.3 Dynamical conductivity. . . . . . . . . . . . . . . . . . . . . . . . . 33
6 Ginzburg-Landau Theory 34
6.1 GL Free Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.2 Type I and Type II superconductivity . . . . . . . . . . . . . 43
6.3 Vortex Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.4 Properties of Single Vortex. Lower critical eld H
c1
56
6.5 Josephson Eect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
0
5 Superconductivity
5.1 Phenomenology
Superconductivity was discovered in 1911 in the Leiden lab-
oratory of Kamerlingh Onnes when a so-called blue boy
(local high school student recruited for the tedious job of mon-
itoring experiments) noticed that the resistivity of Hg metal
vanished abruptly at about 4K. Although phenomenological
models with predictive power were developed in the 30s and
40s, the microscopic mechanism underlying superconductiv-
ity was not discovered until 1957 by Bardeen Cooper and
Schrieer. Superconductors have been studied intensively for
their fundamental interest and for the promise of technologi-
cal applications which would be possible if a material which
superconducts at room temperature were discovered. Until
1986, critical temperatures (T
c
s) at which resistance disap-
pears were always less than about 23K. In 1986, Bednorz and
Mueller published a paper, subsequently recognized with the
1987 Nobel prize, for the discovery of a new class of materials
which currently include members with T
c
s of about 135K.
1
Figure 1: Properties of superconductors.
Superconducting materials exhibit the following unusual be-
haviors:
1. Zero resistance. Below a materials T
c
, the DC elec-
trical resistivity is really zero, not just very small. This
leads to the possibility of a related eect,
2. Persistent currents. If a current is set up in a super-
conductor with multiply connected topology, e.g. a torus,
2
it will ow forever without any driving voltage. (In prac-
tice experiments have been performed in which persistent
currents ow for several years without signs of degrading).
3. Perfect diamagnetism. A superconductor expels a
weak magnetic eld nearly completely from its interior
(screening currents ow to compensate the eld within a
surface layer of a few 100 or 1000 A, and the eld at the
sample surface drops to zero over this layer).
4. Energy gap. Most thermodynamic properties of a su-
perconductor are found to vary as e
/(k
B
T)
, indicating
the existence of a gap, or energy interval with no allowed
eigenenergies, in the energy spectrum. Idea: when there
is a gap, only an exponentially small number of particles
have enough thermal energy to be promoted to the avail-
able unoccupied states above the gap. In addition, this gap
is visible in electromagnetic absorption: send in a photon
at low temperatures (strictly speaking, T = 0), and no
absorption is possible until the photon energy reaches 2,
i.e. until the energy required to break a pair is available.
5.2 Electron-phonon interaction
Superconductivity is due to an eective attraction between
conduction electrons. Since two electrons experience a re-
pulsive Coulomb force, there must be an additional attrac-
tive force between two electrons when they are placed in a
metallic environment. In classic superconductors, this force
3
is known to arise from the interaction with the ionic system.
In previous discussion of a normal metal, the ions were re-
placed by a homogeneous positive background which enforces
charge neutrality in the system. In reality, this medium is
polarizable the number of ions per unit volume can uctuate
in time. In particular, if we imagine a snapshot of a single
electron entering a region of the metal, it will create a net
positive charge density near itself by attracting the oppositely
charged ions. Crucial here is that a typical electron close to
the Fermi surface moves with velocity v
F
= k
F
/m which is
much larger than the velocity of the ions, v
I
= V
F
m/M. So
by the time ( 2/
D
10
13
sec) the ions have polarized
themselves, 1st electron is long gone (its moved a distance
v
F
10
8
cm/s 1000

A, and 2nd electron can happen by

to lower its energy with the concentration of positive charge
before the ionic uctuation relaxes away. This gives rise to an
eective attraction between the two electrons as shown, which
may be large enough to overcome the repulsive Coulomb in-
teraction. Historically, this electron-phonon pairing mech-
anism was suggested by Fr olich in 1950, and conrmed by
the discovery of the isotope eect, wherein T
c
was found to
vary as M
1/2
for materials which were identical chemically
but which were made with dierent isotopes.
The simplest model for the total interaction between two
electrons in momentum states k and k
/
, with q k k
/
, in-
teracting both by direct Coulomb and electron-phonon forces,
is given by
4
Figure 2: Eective attraction of two electrons due to phonon exchange
V (q, ) =
4e
2
q
2
+ k
2
s
+
4e
2
q
2
+ k
2
s

2
q

2
q
, (1)
in the jellium model. Here rst term is Coulomb interac-
tion in the presence of a medium with dielectric constant
= 1+k
2
s
/q
2
, and
q
are the phonon frequencies. The screen-
ing length k
1
s
is 1A or so in a good metal. Second term is
interaction due to exchange of phonons, i.e. the mechanism
pictured in the gure. Note it is frequency-dependent, re-
ecting the retarded nature of interaction (see gure), and in
particular that the 2nd term is attractive for <
q

D
.
Something is not quite right here, however; it looks indeed
as though the two terms are of the same order as 0;
indeed they cancel each other there, and V is seen to be al-
ways repulsive. This indicates that the jellium approximation
is too simple. We should probably think about a more care-
ful calculation in a real system as producing two equivalent
terms, which vary in approximately the same way with k
TF
and
q
, but with prefactors which are arbitrary. In some ma-
terials, then, the second term might win at low frequencies,
depending on details. The BCS interaction is sometimes re-
5
ferred to as a residual attractive interaction, i.e. what is
left when the long-range Coulomb
5.3 Cooper problem
A great deal was known about the phenomenology of super-
conductivity in the 1950s, and it was already suspected that
the electron phonon interaction was responsible, but the mi-
croscopic form of the wave function was unknown. A clue
was provided by Leon Cooper, who showed that the noninter-
acting Fermi sea is unstable towards the addition of a single
pair of electrons with attractive interactions. Cooper began
by examining the wave function of this pair (r
1
, r
2
), which
can always be written as a sum over plane waves
(r
1
, r
2
) =

kq
u
k
(q)e
ikr
1
e
i(k+q)r
2
(2)
where the u
k
(q) are expansion coecients and is the spin
part of the wave function, either the singlet [ > /

2
or one of the triplet, [ >, [ >, [ + > /

2. In
fact since we will demand that is the ground state of the
two-electron system, we will assume the wave function is re-
alized with zero center of mass momentum of the two elec-
trons, u
k
(q) = u
k

q,0
. Here is a quick argument related to
the electron-phonon origin of the attractive interaction.
1
The
electron-phonon interaction is strongest for those electrons
with single-particle energies
k
within
D
of the Fermi level.
1
Thanks to Kevin McCarthy, who forced me to think about this further
6
k
p
-q
k-q
p+q
k
p
g
g

D
p+q
k-q
Figure 3: Electrons scattered by phonon exchange are conned to shell of thickness
D
about Fermi
surface.
In the scattering process depicted in Fig. 3, momentum is
explicitly conserved, i.e. the total momentum
k + p = K (3)
is the same in the incoming and outgoing parts of the diagram.
Now look at Figure 4, and note that if Kis not 0, the phase
space for scattering (attraction) is dramatically reduced. So
the system can always lower its energy by creating K = 0
pairs. Henceforth we will make this assumption, as Cooper
did.
Then (r
1
, r
2
) becomes

k
u
k
e
ik(r
1
r
2
)
. Note that if u
k
is
even in k, the wave function has only terms cos k (r
1
r
2
),
whereas if it is odd, only the sin k (r
1
r
2
) will contribute.
This is an important distinction, because only in the former
case is there an amplitude for the two electrons to live on
7
-p
K
k

D
Figure 4: To get (attractive) scattering with nite cm momentum K, need both electron energies
to be within
D
of Fermi level very little phase space.
top of each other at the origin. Note further that in order to
preserve the proper overall antisymmetry of the wave function,
u
k
even (odd) in k implies the wave function must be spin
singlet (triplet). Let us assume further that there is a general
two-body interaction between the two electrons (the rest of
the Fermi sea is noninteracting in the model!) V (r
1
, r
2
), so
that the Hamiltonian for the system is
H =

2
1
2m

2
2
2m
+ V (r
1
, r
2
). (4)
Inserting the assumed form of into the Schrodinger equation
H = E, and Fourier transforming both sides with respect
to the relative coordinate, r = r
1
r
2
, we nd
(E 2
k
)u
k
=

k>k
F
V
kk
/ u
k
/ , (5)
8
where
k
= k
2
/2m and the V
kk
/ =
_
d
3
rV (r)e
i(k
/
k)r
are the
matrix elements of the two-body interaction.
Recall k, k
/
correspond to energies at the Fermi level
F
in
the absence of V . The question was posed by Cooper, is it
possible to nd an eigenvalue E < 2
F
, i.e. a bound state of
the two electrons? To simplify the problem, Cooper assumed
a model form for V
kk
/ in which
V
kk
/ =
_
V
k
,
k
/ <
c
0 otherwise
(6)
where as usual
k

k

F
. The BCS interaction V
kk
/ is
sometimes referred to as a residual attractive interaction,
i.e. the attractive, short-distance part left when the long-
range Coulomb interaction has been subtracted out, as in (1).
The bound state equation becomes
u
k
=
V

k
/
/
u
k
/
2
k
E
, (7)
where the prime on the summation in this context means sum
only over k such that
f
<
k
<
F
+
c
. Now u
k
may be
eliminated from the equation by summing both sides

k
,
yielding
1
V
=

k
/
1
2
k
E
(8)

1
2
N
0
_

F
+
c

F
d
1
2 E
=
1
2
N
0
log
2
F
+ 2
c
E
2
F
E
.(9)
9
For a weak interaction N
0
V 1, we expect a solution (if at
all) just below the Fermi level, so we treat 2
F
E as a small
positive quantity, e.g. negligible compared to 2
c
. We then
arrive at the pair binding energy

Cooper
2
F
E 2
c
e
2/N
0
V
. (10)
There are several remarks to be made about this result.
1. Note (for your own informationCooper didnt know this
at the time!) that the dependence of the bound state en-
ergy on both the interaction V and the cuto frequency
c
strongly resembles the famous BCS transition temperature
dependence, with
c
identied as the phonon frequency

D
, as given in equation (I.1).
2. the dependence on V is that of an essential singularity,
i.e. a nonanalytic function of the parameter. Thus we
may expect never to arrive at this result at any order in
perturbation theory, an unexpected problem which hin-
dered theoretical progress for a long time.
3. The solution found has isotropic or s-symmetry, since it
doesnt depend on the

k on the Fermi surface. (How would
an angular dependence arise? Look back over the calcula-
tion.)
4. Note the integrand (2
k
E)
1
= (2
k
+
Cooper
)
1
peaks
at the Fermi level with energy spread
Cooper
of states
involved in the pairing. The weak-coupling (N
0
V 1)
solution therefore provides a bit of a posteriori justica-
tion for its own existence, since the fact that
Cooper

c
10
implies that the dependence of V
kk
/ on energies out near
the cuto and beyond is in fact not terribly important, so
the cuto procedure used was ok.
5. The spread in momentum is therefore roughly
Cooper
/v
F
,
and the characteristic size of the pair (using Heisenbergs
uncertainty relation) about v
F
/T
c
. This is about 100-
1000A in metals, so since there is of order 1 electron/ unit
cell in a metal, and if this toy calculation has anything
to do with superconductivity, there are certainly many
electron pairs overlapping each other in real space in a
superconductor.
5.4 Pair condensate & BCS Wavefctn.
Obviously one thing is missing from Coopers picture: if it is
energetically favorable for two electrons in the presence of a
noninteracting Fermi sea to pair, i.e. to form a bound state,
why not have the other electrons pair too, and lower the en-
ergy of the system still further? This is an instability of the
normal state, just like magnetism or charge density wave for-
mation, where a ground state of completely dierent charac-
ter (and symmetry) than the Fermi liquid is stabilized. The
Cooper calculation is a T=0 problem, but we expect that as
one lowers the temperature, it will become at some critical
temperature T
c
energetically favorable for all the electrons to
pair. Although this picture is appealing, many things about
it are unclear: does the pairing of many other electrons alter
11
the attractive interaction which led to the pairing in the rst
place? Does the bound state energy per pair change? Do all
of the electrons in the Fermi sea participate? And most im-
portantly, how does the critical temperature actually depend
on the parameters and can we calculate it?
5.5 BCS Model.
A consistent theory of superconductivity may be constructed
either using the full eective interaction or our approxima-
tion V (q, ) to it. However almost all interesting questions
can be answered by the even simpler model used by BCS.
The essential point is to have an attractive interaction for
electrons in a shell near the Fermi surface; retardation is sec-
ondary. Therefore BCS proposed starting from a phenomeno-
logical Hamiltonian describing free electrons scattering via an
eective instantaneous interaction `a la Cooper:
H = H
0
V

kk
/
q

/
/
c

k
c

k+q
/
c
k
/
+q
/ c
k
/

, (11)
where the prime on the sum indicates that the energies of the
states k and k
/
must lie in the shell of thickness
D
. Note the
interaction term is just the Fourier transform of a completely
local 4-Fermi interaction

(r)

(r)(r)(r).
2
Recall that in our discussion of the instability of the nor-
mal state, we suggested that an innitesimal pair eld could
2
Note this is not the most general form leading to superconductivity. Pairing in higher angular momentum
channels requires a bilocal model Hamiltonian, as we shall see later.
12
produce a nite amplitude for pairing. That amplitude was
the expectation value c

k
c

k
. We ignore for the moment
the problems with number conservation, and ask if we can
simplify the Hamiltonian still further with a mean eld ap-
proximation, again to be justied a posteriori. We proceed
along the lines of generalized Hartree-Fock theory, and rewrite
the interaction as
c

k
c

k+q
/
c
k
/
+q
/ c
k
/

= [c

k
c

k+q
/
+ (c

)]
[c
k
/
+q
/c
k
/

+ (cc)], (12)
where, e.g. (cc) = c
k
/
+q
/c
k
/

c
k
/
+q
/ c
k
/

is the uctu-
ation of this operator about its expectation value. If a mean
eld description is to be valid, we should be able to neglect
terms quadratic in the uctuations when we expand Eq (20).
If we furthermore make the assumption that pairing will take
place in a uniform state (zero pair center of mass momentum),
then we put c
k
/
+q
/ c
k
/

= c
k
/

/c
k
/

q,0
. The eective
Hamiltonian then becomes (check!)
H H
0
(

k
c

k
c

k
+ h.c.) + c

k
c

, (13)
where
= V

k
/
c
k
c
k
. (14)
What BCS (actually Bogoliubov, after BCS) did was then to
treat the order parameter as a (complex) number, and
calculate expectation values in the approximate Hamiltonian
(13), insisting that be determined self-consistently via Eq.
13
(14) at the same time.
3
5.5.1 BCS wave function, gauge invariance, and number con-
servation.
What BCS actually did in their original paper is to treat the
Hamiltonian (11) variationally. Their ansatz for the ground
state of (11) is a trial state with the pairs k , k occupied
with amplitude v
k
and unoccupied with amplitude u
k
, such
that [u
k
[
2
+ [v
k
[
2
= 1:
[ >=
k
(u
k
+ v
k
c

k
c

k
)[0 > . (15)
This is a variational wave function, so the energy is to be
minimized over the space of u
k
, v
k
. Alternatively, one can
diagonalize the Hartree-Fock (BCS) Hamiltonian directly, to-
gether with the self-consistency equation for the order param-
eter; the two methods turn out to be equivalent. I will follow
the latter procedure, but rst make a few remarks on the form
of the wave function. First, note the explicit violation of par-
ticle number conservation: [ > is a superposition of states
describing 0, 2, 4 , N-particle systems.
4
In general a quan-
tum mechanical system with xed particle number N (like,
e.g. a real superconductor!) manifests a global U(1) gauge
symmetry, because H is invariant under c

k
e
i
c

k
. The
state [ > is characterized by a set of coecients u
k
, v
k
,
which becomes u
k
, e
2i
v
k
after the gauge transformation.
3
If the pairing interaction is momentum dependent the self-consistency or gap equation reads
k
=

k
V
kk
c
k

c
k

, which reduces to (14) if one sets V

kk
= V .
4
What happened to the odd numbers? In mesoscopic superconductors, there are actually dierences in the
properties of even and odd-number particle systems, but for bulk systems the distinction is irrelevant.
14
The two states [ > and (), where = 2, are inequiva-
lent, mutually orthogonal quantum states, since they are not
simply related by a multiplicative phase factor.
5
Since H
is independent of , however, all states [() > are contin-
uously degenerate, i.e. the ground state has a U(1) gauge
(phase) symmetry. Any state [() > is said to be a bro-
ken symmetry state, becaue it is not invariant under a U(1)
transformation, i.e. the system has chosen a particular
out of the degenerate range 0 < < 2. Nevertheless the ab-
solute value of the overall phase of the ground state is not an
observable, but its variations (r, t) in space and time are.
It is the rigidity of the phase, i.e. the energy cost of any of
these uctuations, which is responsible for superconductivity.
Earlier I mentioned that it was possible to construct a num-
ber conserving theory. It is now instructive to see how: states
of denite number are formed [Anderson 1958] by making co-
herent superpositions of states of denite phase
[(N) >=
_
2
0
de
iN/2
[() > . (16)
[The integration over gives zero unless there are in the ex-
pansion of the product contained in [ > precisely N/2 pair
creation terms, each with factor exp i.] Note while this state
has maximal uncertainty in the value of the phase, the rigidity
of the system to phase uctuations is retained.
6
5
In the normal state, [ > and () dier by a global multiplicative phase e
i
, which has no physical consequences,
and the ground state is nondegenerate.
6
The phase and number are in fact canonically conjugate variables, [N/2, ] = i, where N = 2i/ in the
representation.
15
It is now straightforward to see why BCS theory works.
The BCS wave function [ > may be expressed as a sum
[ >=

N
a
N
[(N) > [Convince yourself of this by calcu-
lating the a
N
explicitly!]. IF we can show that the distribu-
tion of coecients a
N
is sharply peaked about its mean value
< N >, then we will get essentially the same answers as work-
ing with a state of denite number N =< N >. Using the
explicit form (15), it is easy to show
N = [

k
n
k
[ = 2

k
[v
k
[
2
; (NN)
2
=

k
u
2
k
v
2
k
.
(17)
Now the u
k
and v
k
will typically be numbers of order 1, so
since the numbers of allowed k-states appearing in the k sums
scale with the volume of the system, we have N V , and
(NN)
2
V . Therefore the width of the distribution of
numbers in the BCS state is (N N)
2

1/2
/N N
1/2
.
As N 10
23
particles, this relative error implied by the
number nonconservation in the BCS state becomes negligible.
5.5.2 Is the BCS order parameter general?
Before leaving the subject of the phase in this section, it is
worthwhile asking again why we decided to pair states with
opposite momenta and spin, k and k . The BCS argu-
ment had to do 1) with minimizing the energy of the entire
system by giving the Cooper pairs zero center of mass mo-
mentum, and 2) insisting on a spin singlet state because the
phonon mechanism leads to electron attraction when the elec-
16
trons are at the same spatial position (because it is retarded
in time!), and a spatially symmetric wavefunction with large
amplitude at the origin demands an antisymmetric spin part.
Can we relax these assumptions at all? The rst require-
ment seems fairly general, but it should be recalled that one
can couple to the pair center of mass with an external mag-
netic eld, so that one will create spatially inhomogeneous
(nite-q) states with current ow in the presence of a mag-
netic eld. Even in zero external eld, it has been proposed
that systems with coexisting antiferromagnetic correlations
could have pairing with nite antiferromagnetic nesting vec-
tor

Q [Baltensberger and Strassler 1963]. The requirement
for singlet pairing can clearly be relaxed if there is a pair-
ing mechanism which disfavors close approach of the paired
particles. This is the case in superuid
3
He, where the hard
core repulsion of two
3
He atoms supresses T
c
for s-wave, sin-
glet pairing and enhances T
c
for p-wave, triplet pairing where
the amplitude for two particles to be together at the origin is
always zero.
In general, pairing is possible for some pair mechanism if
the single particle energies corresponding to the states k and
k
/

/
are degenerate, since in this case the pairing interaction
is most attractive. In the BCS case, a guarantee of this degen-
eracy for k and k in zero eld is provided by Kramers
theorem, which says these states must be degenerate because
they are connected by time reversal symmetry. However,
there are other symmetries: in a system with inversion sym-
17
metry, parity will provide another type of degeneracy, so k ,
k , k and k are all degenerate and may be paired
with one another if allowed by the pair interaction.
5.6 Thermodynamics
5.6.1 Bogoliubov transformation
We now return to (13) and discuss the solution by canonical
transformation given by Bogoliubov. After our drastic ap-
proximation, we are left with a quadratic Hamiltonian in the
cs, but with c

and cc terms in addition to c

cs. We can di-

agonalize it easily, however, by introducing the quasiparticle
operators
k0
and
k1
by
c
k
= u

k0
+ v
k

k1
c

k
= v

k0
+ u
k

k1
. (18)
You may check that this transformation is canonical (preserves
fermion comm. rels.) if [u
k
[
2
+ [v
k
[
2
= 1. Substituting into
(13) and using the commutation relations we get
H
BCS
=

k
[([u
k
[
2
[v
k
[
2
)(

k0

k0
+

k1

k1
) + 2[v
k
[
2
+2u

k
v

k1

k0
+ 2u
k
v
k

k1

k0
]
+

k
[(
k
u
k
v

k
+

k
u

k
v
k
)(

k0

k0
+

k1

k1
1)
+(
k
v
2
k

k
u
2
k
)
k1

k0
+ (

k
v
2
k

k
u
2
k
)

k0

k1
+
k
c

k
c

, (19)
18
which does not seem to be enormous progress, to say the least.
But the game is to eliminate the terms which are not of the
form

, so to be left with a sum of independent number-type

terms whose eigenvalues we can write down. The coecients
of the

and type terms are seen to vanish if we choose

2
k
u
k
v
k
+

k
v
2
k

k
u
2
k
= 0. (20)
This condition and the normalization condition [u
k
[
2
+[v
k
[
2
=
1 are both satised by the solutions
[u
k
[
2
[v
k
[
2
=
1
2
_
1

k
E
k
_
, (21)
where I dened the Bogolibov quasiparticle energy
E
k
=
_

2
k
+ [
k
[
2
. (22)
The BCS Hamiltonian has now been diagonalized:
H
BCS
=

k
E
k
_

k0

k0
+

k1

k1
_
+

k
_

k
E
k
+
k
c

k
c

_
.
(23)
Note the second term is just a constant, which will be impor-
tant for calculating the ground state energy accurately. The
rst term, however, just describes a set of free fermion exci-
tations above the ground state, with spectrum E
k
.
5.6.2 Density of states
The Bogoliubov quasiparticle spectrum E
k
is easily seen to
have a minimum
k
for a given direction k on the Fermi sur-
face dened by
k
= 0;
k
therefore, in addition to playing
19
the role of order parameter for the superconducting transi-
tion, is also the energy gap in the 1-particle spectrum. To
see this explicitly, we can simply do a change of variables in
all energy integrals from the normal metal eigenenergies
k
to
the quasiparticle energies E
k
:
N(E)dE = N
N
()d. (24)
If we are interested in the standard case where the gap is
much smaller than the energy over which the normal state dos
N
N
() varies near the Fermi level, we can make the replace-
ment
N
N
() N
N
(0) N
0
, (25)
so using the form of E
k
from (22) we nd
N(E)
N
0
=
_
E

E
2

2
E >
0 E <
. (26)
This function is sketched in Figure 5.
5.6.3 Critical temperature
The critical temperature is dened as the temperature at
which the order parameter
k
vanishes. We can now cal-
culate this with the aid of the diagonalized Hamiltonian. The
20
N(E)
N
0
E
Normal
SC
1
0
1

| |

| |
energy

Figure 5: a) Normalized density of states; b) Quasiparticle spectrum.

self-consistency condition is

k
= V

k
/
c

k
/

k
/

= V

k
/
u
k
/ v

k
/
1

k0

k0

k1

k1

= V

k
2E
k
(1 2f(E
k
)) . (27)
Since 1 2f(E) = tanh[E/(2T)], the BCS gap equation
reads

k
= V

k
/

k
/
2E
k
/
tanh
E
k
/
2T
(28)
This equation may now be solved, rst for the critical temper-
ature itself, i.e. the temperature at which 0, and then
for the normalized order parameter /T
c
for any temperature
T. It is the ability to eliminate all normal state parameters
21
from the equation in favor of T
c
itself which makes the BCS
theory powerful. For in practice the parameters
D
, N
0
, and
particularly V are known quite poorly, and the fact that two
of them occur in an exponential makes an accurate rst prin-
ciples calculation of T
c
nearly impossible. You should always
be suspicious of a theory which claims to be able to calculate
T
c
! On the other hand, T
c
is easy to measure, so if it is the
only energy scale in the theory, we have a tool with enormous
predictive power.
First note that at T
c
, the gap equation becomes
1
N
0
V
=
_

D
0
d
k
1

k
tanh

k
2T
c
. (29)
This integral can be approximated carefully, but it is useful
to get a sense of what is going on by doing a crude treatment.
Note that since T
c

D
generally, most of the integrand
weight occurs for > T, so we can estimate the tanh factor
by 1. The integral is log divergent, which is why the cuto

D
is so important. We nd
1
N
0
V
0
= log

T
c
T
c

D
e
1/N
0
V
(30)
The more accurate analysis of the integral gives the BCS result
T
c
= 1.14
D
e
1/N
0
V
(31)
We can do the same calculation near T
c
, expanding to lead-
ing order in the small quantity (T)/T, to nd (T)/T
c

22
3.06(1 T/T
c
)
1/2
. At T = 0 we have
1
N
0
V
=
_

D
0
d
k
1
E
k
=
_

D

dEN(E)/E (32)
=
_

D

dE
1

E
2

2
ln(2
d
/), (33)
so that (0) 2
D
exp 1/N
0
V , or (0)/T
c
1.76. The
full temperature dependence of (T) is sketched in Figure 6).
In the halcyon days of superconductivity theory, comparisons

T
T
c
(T)
1. 76 T
c
Figure 6: BCS order parameter as fctn. of T.
with the theory had to be compared with a careful table of
/T
c
painstakingly calculated and compiled by M uhlschlegl.
Nowadays the numerical solution requires a few seconds on
a PC. It is frequently even easier to use a phenomenological
approximate closed form of the gap, which is correct near
T = 0 and = T
c
:
(T) =
sc
T
c
tanh

sc
_
a
C
C
N
(
T
c
T
1), (34)
23
where
sc
= (0)/T
c
= 1.76, a = 2/3, and C/C
N
= 1.43
is the normalized specic heat jump.
7
This is another of the
universal ratios which the BCS theory predicted and which
helped conrm the theory in classic superconductors.
5.6.4 Specic heat.
The gap in the density of states is reected in all thermody-
namic quantities as an activated behavior e
/T
, at low T,
due to the exponentially small number of Bogoliubov quasi-
particles with thermal energy sucient to be excited over the
gap at low temperatures T . The electronic specic
heat is particularly easy to calculate, since the entropy of the
BCS superconductor is once again the entropy of a free gas
of noninteracting quasiparticles, with modied spectrum E
k
.
The expression (II.6) then gives the entropy directly, and may
be rewritten
S = k
B
_

0
dEN(E)f(E)lnf(E)+[1f(E)]ln[1f(E)],
(35)
where f(E) is the Fermi function. The constant volume spe-
cic heat is just C
el,V
= T[dS/dT]
V
, which after a little alge-
bra may be written
C
el,V
=
2
T
_
dEN(E)(
f
E
)[E
2

1
2
T
d
2
dT
]. (36)
A sketch of the result of a full numerical evaluation is shown
in Figure 1. Note the discontinuity at T
c
and the very rapid
7
Note to evaluate the last quantity, we need only use the calculated temperature dependence of near T
c
, and
insert into Eq. (47).
24
fallo at low temperatures.
It is instructive to actually calculate the entropy and specic
heat both at low temperatures and near T
c
. For T ,
f(E) e
E/T
and the density of states factor N(E) in the
integral cuts o the integration at the lower limit , giving
C (N
0

5/2
/T
3/2
)e
/T
.
8
Note the rst term in Eq. (47)is continuous through the
transition 0 (and reduces to the normal state specic
heat (2
2
/3)N
0
T above T
c
), but the second one gives a dis-
continuity at T
c
of (C
N
C
S
)/C
N
= 1.43, where C
S
= C(T

c
)
and C
N
= C(T
+
c
). To evaluate (36), we need the T depen-
dence of the order parameter from a general solution of (28).
5.7 Electrodynamics
5.7.1 Linear response to vector potential
The existence of an energy gap is not a sucient condition for
superconductivity (actually, it is not even a necessary one!).
Insulators, for example, do not possess the phase rigidity
which leads to perfect conductivity and perfect diamagnetism
which are the dening characteristics of superconductivity.
We can understand both properties by calculating the cur-
8
To obtain this, try the following:
replace the derivative of Fermi function by exp-E/T
do integral by parts to remove singularity at Delta
expand around Delta E = Delta + delta E
change integration variables from E to delta E
Somebody please check my answer!
25
rent response of the system to an applied magnetic or electric
eld.
9
The kinetic energy in the presence of an applied vector
potential A is just
H
0
=
1
2m

_
d
3
r

(r)[i(
e
c
)A]
2

(r), (37)
and the second quantized current density operator is given by
j(r) =
e
2m

(r)(i
e
c
A)(r) + [(i
e
c
A)

(r)](r) (38)
= j
para

e
2
mc

(r)(r)A, (39)
where
j
para
(r) =
ie
2m

(r)(r) (

(r))(r), (40)
or in Fourier space,
j
para
(q) =
e
m

k
kc

kq
c
k
(41)
We would like to do a calculation of the linear current re-
sponse j(q, ) to the application of an external eld A(q, )
to the system a long time after the perturbation is turned
on. Expanding the Hamiltonian to rst order in A gives the
interaction
H
/
=
_
d
3
rj
para
A =
e
mc

k
k A(q)c

kq
c
q
. (42)
9
To see this, note that we can choose a gauge where E = (1/c)A/t = iA/c for a periodic electric eld.
Then the Fourier component of the current is
j(q, ) = (q, )E(q, ) = K(q, )A(q, ),
so (q, ) = icK(q, )/.
26
The expectation value < j > may now be calculated to linear
order via the Kubo formula, yielding
j(q, ) = K(q, )A(q, ) (43)
with
K(q, ) =
ne
2
mc
+ [j
para
, j
para
](q, ). (44)
Note the rst term in the current
j
dia
(q, )
ne
2
mc
A(q, ) (45)
is purely diagmagnetic, i.e. these currents tend to screen the
external eld (note sign). The second, paramagnetic term is
formally the Fourier transform of the current-current corre-
lation function (correlation function used in the sense of our
discussion of the Kubo formula).
10
Here are a few remarks on
this expression:
Note the simple product structure of (43) in momentum
space implies a nonlocal relationship in general between j
and A., i.e. j(r) depends on the A(r
/
) at many points r
/
around r.
11
Note also that the electric eld in a gauge where the elec-
trostatic potential is set to zero may be written E(q, ) =
10
We will see that the rst term gives the diamagnetic response of the system, and the second the temperature-
dependent paramagnetic response.
11
If we transformed back, wed get the convolution
j(r) =
_
d
3
r

K(r, r

)A(r

) (46)
27
iA(q, ), so that the complex conductivity of the sys-
tem dened by j = E may be written
(q, ) =
i

K(q, ) (47)
What happens in a normal metal? The paramagnetic sec-
ond term cancels the diamagnetic response at = 0,
leaving no real part of K (Im part of ), i.e. the conduc-
tivity is purely dissipative and not inductive at , q = 0
in the normal metal.
5.7.2 Meissner Eect.
There is a theorem of classical physics proved by Bohr
12
which
states that the ground state of a system of charged particles
in an external magnetic eld carries zero current. The essen-
tial element in the proof of this theorem is the fact that the
magnetic forces on the particles are always perpendicular to
their velocities. In a quantum mechanical system, the three
components of the velocity do not commute in the presence
of the eld, allowing for a nite current to be created in the
ground state. Thus the existence of the Meissner eect in
superconductors, wherein magnetic ux is expelled from the
interior of a sample below its critical temperature, is a clear
proof that superconductivity is a manifestation of quantum
mechanics.
12
See The development of the quantum-mechanical electron theory of metals: 1928-33. L. Hoddeson and G.
Baym, Rev. Mod. Phys., 59, 287 (1987).
28
The typical theorists geometry for calculating the penetra-
tion of an electromagnetic eld into a superconductor is the
half-space shown in Figure 7, and compared to schematics of
practical experimental setups involving resonant coils and mi-
crowave cavities in Figs. 7 a)-c). In the gedanken experiment

H
0
a)
L (2 r)
b)
I
cavity
c)
Figure 7: a) Half-space geometry for penetration depth calculation; b) Resonant coil setup; c)
Microwave cavity
case, a DC eld is applied parallel to the sample surface, and
currents and elds are therefore functions only of the coordi-
nate perpendicular to the surface, A = A(z), etc. Since we
are interested in an external electromagnetic wave of very long
wavelength compared to the sample size, and zero frequency,
we need the limit = 0, q of the response. We will
assume that in this limit K(0, 0) const, which we will call
(c/4)
2
for reasons which will become clear! Equation
(43) then has the form
j =
c
4

2
A, (48)
29
This is sometimes called Londons equation, which must be
solved in conjunction with Maxwells equation
B =
2
A =
4
c
j =
2
A, (49)
which immediately gives A e
z/
, and B = B
0
e
z/
. The
currents evidently screen the elds for distances below the
surface greater than about . This is precisely the Meissner
eect, which therefore follows only from our assumption that
K(0, 0) = const. A BCS calculation will now tell us how the
penetration depth depends on temperature.
Evaluating the expressions in (44) in full generality is te-
dious and is usually done with standard many-body methods
beyond the scope of this course. However for q = 0 the cal-
culation is simple enough to do without resorting to Greens
functions. First note that the perturbing Hamiltonian H
/
may
be written in terms of the quasiparticle operators (18) as
H
/
= (50)

e
mc

k
k A(q)
_
(u
k
u
k+q
+ v
k
v
k+q
)(

k+q0

k0

k+q1

k1
)
+(v
k
u
k+q
u
k
v
k+q
)(

k+q0

k1

k+q1

k0
)
_

q0

e
mc

k
k A(0)(

k0

k0

k1

k1
) (51)
If you compare with the A = 0 Hamiltonian (23), we see that
30
the new excitations of the system are
E
k0
E
k

e
mc
k A(0)
E
k1
E
k
+
e
mc
k A(0) (52)
We may similarly insert the quasiparticle operators (18) into
the expression for the expectation value of the paramagnetic
current operator(41):
j
para
(q = 0) =
e
m

k
k(

k0

k0

k1

k1
)
=
e
m

k
k(f(E
k0
) f(E
k1
)) . (53)
We are interested in the linear response A 0, so that when
we expand wrt A, the paramagnetic contribution becomes
j
para
(q = 0) =
2e
2
m
2
c

k
[k A(0)] k
_

f
E
k
_
. (54)
Combining now with the denition of the response function
K and the diamagnetic current (45), and recalling

k

N
0
_
d
k
(d/4), with N
0
= 3n/(4
F
)
13
and
_
(d/4)kk =
1/3, we get for the static homogeneous response is therefore
K(0, 0) =
ne
2
mc
1
_
d
k
(
f
E
k
)1 (55)

n
s
(T)e
2
mc
1 (56)
13
Here N
0
is single-spin DOS!
31
where in the last step, I dened the superuid density to be
n
s
(T) n n
n
(T), with normal uid density
n
n
(T) n
_
d
k
_
f
E
k
_
. (57)
Note at T = 0, f/E
k
0, [Not a delta function, as in
the normal state casedo you see why?], while at T = T
c
the
integral n
n
1.
14
Correspondingly, the superuid density as
dened varies between n at T = 0 and 0 at T
c
. This is the BCS
microscopic justication for the rather successful phenomeno-
logical two-uid model of superconductivity: the normal uid
consists of the thermally excited Bogoliubov quasiparticle gas,
and the superuid is the condensate of Cooper pairs.
15
Y(T)
~ exp-/ T
1
1
0
n / n
n
T/ T
c
a)
T/ T
c
1
1
0
1
0

T/ T
c
b)
c)
n / n
s

(T)
Figure 8: a) Yoshida function; b) superuid density ; c) penetration depth
Now lets relate the BCS microscopic result for the static
homogeneous response to the penetration depth appearing in
the macroscopic electrodynamics calculation above. We nd
14
The dimensioness function n
n
(T/T
c
)/n is sometimes called the Yoshida function, Y (T), and is plotted in Fig.8.
15
The BCS theory and subsequent extensions also allow one to understand the limitations of the two-uid picture:
for example, when one probes the system at suciently high frequencies , the normal uid and superuid
fractions are no longer distinct.
32
immediately
(T) = (
mc
2
4n
s
(T)e
2
)
1/2
. (58)
At T = 0, the supercurrent screening excludes the eld from
all of the sample except a sheath of thickness (0). At small
but nite temperatures, an exponentially small number of
quasiparticles will be excited out of the condensate, depeleting
the supercurrent and allowing the eld to penetrate further.
Both n
n
(T) and (T) (0) may therefore be expected to
vary as e
/T
for T T
c
, as may be conrmed by explicit
expansion of Eq. (57). [See homework.] Close to T
c
, the pen-
etration depth diverges as it must, since in the normal state
the eld penetrates the sample completely.
5.7.3 Dynamical conductivity.
The calculation of the full, frequency dependent conductivity
is beyond the scope of this course. If you would like to read an
old-fashioned derivation, I refer you to Tinkhams book. The
main point to absorb here is that, as in a semiconductor with
a gap, at T = 0 there is no process by which a photon can be
absorbed in a superconductor until its energy exceeds 2, the
binding energy of the pair. This threshold for optical ab-
sorption is one of the most direct measurements of the gaps of
the old superconductors. If one is interested simply in the zero
DC resistance state of superconductors, it is frustrating to nd
this is not discussed often in textbooks. In fact, the argument
is somewhat oblique. One notes that the Ferrell-Tinkham-
33
Glover sum rule
_
d(0, ) = ne
2
/(2m) requires that the
integral under (q = 0, ) be conserved when one passes
through the superconducting transition. Thus the removal of
spectal weight below 2 (found in calculation of BCS con-
ductivity, rst by Mattis and Bardeen) implies that the lost
spectral weight must be compensated by a delta-function in
() at = 0, i.e. innite DC conductivity.
6 Ginzburg-Landau Theory
6.1 GL Free Energy
While the BCS weak-coupling theory we looked at the last
two weeks is very powerful, and provides at least a qualita-
tively correct description of most aspects of classic supercon-
ductors,
16
there is a complementary theory which a) is simpler
and more physically transparent, although valid only near the
transition; and b) provides exact results under certain circum-
stances. This is the Ginzburg-Landau theory [ V.L. Ginzburg
and L.D. Landau, Zh. Eksp. Teor. Fiz. 20, 1064 (1950)],
which received remarkably little attention in the west until
Gorkov showed it was derivable from the BCS theory. [L.P.
Gorkov, Zh. Eksp. Teor Fiz. 36, 1918 (1959)]. The the-
ory simply postulated the existence of a macrosopic quantum
wave function (r) which was equivalent to an order param-
eter, and proposed that on symmetry grounds alone, the free
16
In fact one could make a case that the BCS theory is the most quantitatively accurate theory in all of condensed
matter physics
34
energy density of a superconductor should be expressible in
terms of an expansion in this quantity:
f
s
f
n
V
= a[[
2
+ b[[
4
+
1
2m

[(+
ie

A)[
2
, (59)
where the subscripts n and s refer to the normal and super-
conducting states, respectively.
Lets see why GL might have been led to make such a
guess. The superconducting-normal transition was empiri-
cally known to be second order in zero eld, so it was natural
to write down a theory analogous to the Landau theory of
a ferromagnet, which is an expansion in powers of the mag-
netization, M. The choice of order parameter for the super-
conductor corresponding to M for the ferromagnet was not
obvious, but a complex scalar eld was a natural choice
because of the analogy with liquid He, where [[
2
is known
to represent the superuid density n
s
;
17
a quantum mechani-
cal density should be a complex wave function squared. The
biggest leap of GL was to specify correctly how electromag-
netic elds (which had no analog in superuid He) would
couple to the system. They exploited in this case the simi-
larity of the formalism to ordinary quantum mechanics, and
coupled the elds in the usual way to charges e

associated
with particles of mass m

. Recall for a real charge in a

17
in the He case has the microscopic interpretation as the Bose condensate amplitude.
35
magnetic eld, the kinetic energy is:
< [1
kin
[ > =
1
2m
_
d
3
r

(+
ie
c

A)
2
(60)
=
1
2m
_
d
3
r[(+
ie
c

A)[
2
, (61)
after an integration by parts in the second step. GL just re-
placed e, m with e

, m

to obtain the kinetic part of Eq. (59);

they expected that e

and m

were the elementary electron

charge and mass, respectively, but did not assume so.
T>T
T<T
c
c
f
||
Figure 9: Mexican hat potential for superconductor.
A system described by this free energy will undergo a second-
order phase transition in zero eld when a = 0: clearly when
a is positive, the system can minimize f by having = 0 (no
superconductivity), whereas if a is negative, f has a mini-
mum with ,= 0. The free energy (59) is a functional of
36
the order parameter , meaning the actual value of the order
parameter realized in equilibrium satises f/ = 0.
18
No-
tice f is independent of the phase of the order parameter,
[[e
i
, and so the ground state for a < 0 is equivalent
to any state related to it by multiplication by a pure phase.
This is the U(1) gauge invariance of which we spoke earlier.
This symmetry is broken when the system chooses one of the
ground states (phases) upon condensation (Fig 1.).
For a uniform system in zero eld, the total free energy
F =
_
d
3
rf is minimized when f is, so one nd for the order
parameter at the minimum,
[[
eq
= [
a
2b
]
1/2
, a < 0 (62)
[[
eq
= 0, a > 0. (63)
When a changes sign, a minimum with a nonzero value be-
comes possible. For a second order transition as one lowers the
temperature, we assume that a and b are smooth functions of
T near T
c
. Since we are only interested in the region near T
c
,
we take only the leading terms in the Taylor series expansions
in this region: a(T, H) = a
0
(T T
c
) and b = constant. Eqs.
(62) and (63) take the form:
[(T)[
eq
= [
a
0
(T
c
T)
2b
]
1/2
, T < T
c
(64)
[(T)[
eq
= 0, T > T
c
. (65)
18
Thus you should not be perturbed by the fact that f apparently depends on even for a > 0. The value of f
in equilibrium will be f
n
= f[ = 0].
37
Substituting back into Eqs.59, we nd:
f
s
(T) f
n
(T) =
a
2
0
4b
(T
c
T)
2
, T < T
c
(66)
f
s
(T) f
n
(T) = 0, T > T
c
. (67)
The idea now is to calculate various observables, and de-
termine the GL coecients for a given system. Once they
are determined, the theory acquires predictive power due to
its extreme simplicity. It should be noted that GL theory is
applied to many systems, but it is in classic superconductors
that it is most accurate, since the critical region, where de-
viations from mean eld theory are large, is of order 10
4
or
less. Near the transition it may be taken to be exact for all
practical purposes. This is not the case for the HTSC, where
the size of the critical region has been claimed to be as much
as 10-20K in some samples.
Supercurrents. Lets now focus our attention on the
term in the GL free energy which leads to supercurrents, the
kinetic energy part:
F
kin
=
_
d
3
r
1
2m

[(+
ie

A)[
2
(68)
=
_
d
3
r
1
2m

[([[)
2
+ ( e

/cA)
2
[[
2
]. (69)
These expressions deserve several remarks. First, note that
the free energy is gauge invariant, if we make the transforma-
tion

A

A+, where is any scalar function of position,
while at the same time changing exp(ie

/c) . Sec-
ond, note that in the last step above I have split the kinetic
38
part of f into a term dependent on the gradients of the order
parameter magnitude [[ and on the gradients of the phase
. Let us use a little intuition to guess what these terms
mean. The energy of the superconducting state below T
c
is
lower than that of the normal state by an amount called the
condensation energy.
19
From Eq. (59) in zero eld this is of
order [[
2
very close to the transition. To make spatial vari-
ations of the magnitude of must cost a signicant fraction
of the condensation energy in the region of space in which it
occurs.
20
On the other hand, the zero-eld free energy is ac-
tually invariant with respect to changes in , so uctuations
of alone actually cost no energy.
With this in mind, lets ask what will happen if we apply
a weak magnetic eld described by A to the system. Since
it is a small perturbation, we dont expect it to couple to [[
but rather to the phase . The kinetic energy density should
then reduce to the second term in Eq. (69), and furthermore
we expect that it should reduce to the intuitive two-uid ex-
pression for the kinetic energy due to supercurrents,
1
2
mn
s
v
2
s
.
Recall from the superuid He analogy, we expect [[
2
n

s
to be a kind of density of superconducting electrons, but that
we arent certain of the charge or mass of the particles. So
lets put
f
kin

1
2m

[(+
ie

A)[
2
. =
_
d
3
r
1
2m

(+e

/cA)
2
[[
2

1
2
m

s
v
2
s
.
(70)
Comparing the forms, we nd that the superuid velocity
19
We will see below from the Gorkov derivation of GL from BCS that it is of order N(0)
2
.
20
We can make an analogy with a ferromagnet, where if we have a domain wall the magnetization must go to zero
at the domain boundary, costing lots of surface energy.
39
must be
v
s
=
1
m

( +
e

A). (71)
Thus the gradient of the phase is related to the superuid
velocity, but the vector potential also appears to keep the
entire formalism gauge-invariant.
Meissner eect. The Meissner eect now follows imme-
diately from the two-uid identications we have made. The
supercurrent density will certainly be just

j
s
= e

s
v
s
=
e

s
m

( +
e

A). (72)
Taking the curl of this equation, the phase drops out, and we
nd the magnetic eld:

j
s
=
e

2
n

s
m

B. (73)
Now recall the Maxwell equation

j
s
=
c
4

B, (74)
which, when combined with (14), gives
c
4

B =
c
4

B =
e

2
n
s
m

B, (75)
or

B =

B, (76)
where
=
m

c
2
4e

2
n

s
1/2
. (77)
40
Notice now that if we use what we know about Cooper pairs,
this expression reduces to the BCS/London penetration depth.
We assume e

is the charge of the pair, namely e

= 2e, and
similarly m

= 2m, and [[
2
= n

s
= n
s
/2 since n

s
is the
density of pairs.
Flux quantization. If we look at the ux quantization
described in Part 1 of these notes, it is clear from our sub-
sequent discussion of the Meissner eect, that the currents
which lead to ux quantization will only ow in a small part
of the cross section, a layer of thickness . This layer encloses
the ux passing through the toroid. Draw a contour C in the
interior of the toroid, as shown in Figure 10. Then v
s
= 0
everywhere on C. It follows that
j
s
C

Figure 10: Quantization of ux in a toroid.

0 =
_
C
d

v
s
=
1
m

_
C
d

( +
e

A). (78)
The last integral may be evaluated using
_
C
d

= 2 integer, (79)
41
which follows from the requirement that be single-valued as
in quantum mechanics. Having n ,= 0 requires that one not
be able to shrink the contour to a point, i.e. that the sample
has a hole as in our superconducting ring. The line integral
of the vector potential is and
e

c
_
C
d

A =
e

c
_
S
d

S

A (80)
=
e

c
_
S
d

S

B (81)
=
e

c
. (82)
Here S is a surface spanning the hole and the ux through
the hole. Combining these results,
= 2
c
2e
n = n
hc
2e
= n
0
, (83)
where n is a integer,
0
is the ux quantum, and Ive rein-
serted the correct factor of in the rst step to make the units
right. Flux quantization indeed follows from the fact that the
current is the result of a phase gradient.
21
Derivation from Microscopic Theory. One of the
reasons the GL theory did not enjoy much success at rst
was the fact that it is purely phenomenological, in the sense
that the parameters a
0
, b, m

are not given within any micro-

scopic framework. The BCS theory is such a framework, and
gives values for these coecients which set the scale for all
21
It is important to note, however, that a phase gradient doesnt guarantee that a current is owing. For example,
in the interior of the system depicted in Fig. 2, both and

A are nonzero in the most convenient gauge, and cancel
each other!
42
quantities calculable from the GL free energy. The GL theory
is more general, however, since, e.g. for strong coupling su-
perconductors the weak coupling values of the coecients are
simply replaced by dierent ones of the same order of mag-
nitude, without changing the form of the GL free energy. In
consequence, the dependence of observables on temperature,
eld, etc, will obey the same universal forms.
The GL theory was derived from BCS by Gorkov. The
calculation is beyond the scope of this course, but can be
found in many texts.
6.2 Type I and Type II superconductivity
Now lets look at the problem of the instability of the normal
state to superconductivity in nite magnetic eld H. At what
magnetic eld to we expect superconductivity to be destroyed,
for a given T < T
c
?
22
Well, overall energy is conserved, so
the total condensation energy of the system in zero eld,f
s

f
n
(T) of the system must be equal to the magnetic eld energy
_
d
3
rH
2
/8 the system would have contained at the critical
eld H
c
in the absence of the Meissner eect. For a completely
homogeneous system I then have
f
s
(T) f
n
(T) = H
2
c
/8, (84)
and from Eq. (8) this means that, near T
c
,
H
c
=
_
2a
2
0
b
(T
c
T). (85)
22
Clearly it will destroy superconductivity since it breaks the degenerace of between the two componenets of a
Cooper pair.
43
Whether this thermodynamic critical eld H
c
actually rep-
resents the applied eld at which ux penetrates the sample
depends on geometry. We assumed in the simplied treat-
ment above that the eld at the sample surface was the same
as the applied eld. Clearly for any realistic sample placed
in a eld, the lines of eld will have to follow the contour
of the sample if it excludes the eld from its interior. This
means the value of H at dierent points on the surface will
be dierent: the homogeneity assumption we made will not
quite hold. If we imagine ramping up the applied eld from
zero, there will inevitably come a point H
appl
= H
appl,c
where
the eld at certain points on the sample surface exceeds the
critical eld, but at other points does not. For applied elds
H
appl,c
< H
appl
< H
c
, part of the sample will then be normal,
with local eld penetration, and other parts will still exclude
eld and be superconducting. This is the intermediate state
of a type I superconductor. The structure of this state for a
real sample is typically a complicated striped pattern of su-
perconducting and normal phases. Even for small elds, edges
and corners of samples typically go normal because the eld
lines bunch up there; these are called demagnetizing eects,
and must be accounted for in a quantitatively accurate mea-
surement of, say, the penetration depth. It is important to
note that these patterns are completely geometry dependent,
and have no intrinsic length scale associated with them.
In the 50s, there were several materials known, however,
in which the ux in suciently large elds penetrated the
44
sample in a manner which did not appear to be geometry de-
pendent. For example, samples of these so-called type II
superconductors with nearly zero demagnetizing factors (long
thin plates placed parallel to the eld) also showed ux pen-
etration in the superconducting state. The type-II materials
exhibit a second-order transition at nite eld and the ux
B through the sample varies continuously in the supercon-
ducting state. Therefore the mixed state must have currents
owing, and yet the Meissner eect is not realized, so that the
London equation somehow does not hold.
The answer was provided by Abrikosov in 1957 [A.A.A.,
Sov. Phys. JETP 5, 1174 (1957).] in a paper which Landau
apparently held up for several years because he did not be-
lieve it. Let us neglect the eects of geometry again, and go
back to our theorists sample with zero demagnetizing factor.
Can we relax any of the assumptions that led to the Lon-
don equation (72)? Only one is potentially problematic, that
n

s
(r) = [(r)[
2
= constant independent of position. Lets
examineas Abrikosov didthe energy cost of making spatial
variations of the order parameter. The free energy in zero
eld is
F =
_
d
3
r[a[[
2
+
1
2m

[[
2
+ b[[
4
], (86)
or
1
a
F =
_
d
3
r[[[
2
+
2
[[
2
+
b
a
[[
4
], (87)
45
where Ive put
= [
1
2m

a
]
1/2
= [
1
2m

a
0
(T
c
T)
]
1/2
. (88)
Clearly the length represents some kind of stiness of the
quantitiy [[
2
, the superuid density. [Check that it does in-
deed have dimensions of length!] If , the so-called coherence
length, is small, the energy cost of n
s
varying from place to
place will be small. If the order parameter is somehow changed
from its homogeneous equilibrium value at one point in space
by an external force, species the length scale over which
it heals. We can then investigate the possibility that, as
the kinetic energy of superuid ow increases with increasing
eld, if is small enough it might eventually become favorable
to bend [[
2
instead. In typical type I materials, (T = 0)
is of order several hundreds or even thousands of Angstrom,
but in heavy fermion superconductors, for example, coher-
ence lengths are of order 50-100A. The smallest coherence
lengths are attained in the HTSC, where
ab
is of order 12-
15A, whereas
c
is only 2-3A.
The general problem of minimizing F when depends on
position is extremely dicult. However, we are mainly in-
terested in the phase boundary where is small, so life is a
bit simpler. Lets recall our quantum mechanics analogy once
more so as to write F in the form:
F =
_
d
3
r[a[[
2
+ b[[
4
]+ < [

H
kin
[ >, (89)
46
where

H
kin
is the operator

1
2m

(+
ie

A)
2
. (90)
Now note 1) suciently close to the transition, we may
always neglect the 4th-order term, which is much smaller;
2) to minimize F, it suces to minimize <

H
kin
>, since
the [[
2
term will simply x the overall normalization. The
variational principle of quantum mechanics states that the
minimum value of < H > over all possible is achieved
when is the ground state (for a given normalization of ).
So we need only solve the eigenvalue problem

H
kin

j
= E
j

j
(91)
for the lowest eigenvalue, E
j
, and corresponding eigenfunction

j
. For the given form of

H
kin
, this reduces to the classic
quantum mechanics problem of a charged particle moving in
an applied magnetic eld. The applied eld H is essentially
the same as the microscopic eld B since is so small (at the
phase boundary only!). Ill remind you of the solution, due to
to Landau, in order to x notation. We choose a convenient
gauge,
A = Hy x, (92)
in which Eq. 44 becomes
1
2m

[(i

x
+
y

2
M
)
2

2
y
2


2
z
2
]
j
= E
j

j
, (93)
where
M
= (c/e

H)
1/2
is the magnetic length. Since the co-
ordinates x and z dont appear explicitly, we make the ansatz
47
of a plane wave along those directions:
= (y)e
ik
x
x+ik
z
z
, (94)
yielding
1
2m

[(k
x
+
y

2
M
)
2

2
y
2
+ k
2
z
](y) = E(y). (95)
But this you will recognize as just the equation for a one-
dimensional harmonic oscillator centered at the point y =
k
x

2
M
with an additional additive constant k
2
z
/2m

in the
energy. Recall the standard harmonic oscillator equation
(
1
2m

2
x
2
+
1
2
kx
2
) = E, (96)
with ground state energy
E
0
=

0
2
=
1
2
(k/m)
1/2
, (97)
where k is the spring constant, and ground state wavefunc-
tion corresponding to the lowest order Hermite polynomial,

0
exp[(mk/4)
1/2
x
2
]. (98)
Lets just take over these results, identifying

H
kin

k
x
,k
z
=
e

H
2m

k
x
,k
z
. (99)
The ground state eigenfunction may then be chosen as

k
x
,k
z
=
0
(

2
M
L
2
y
)
1/4
e
ik
x
x+ik
z
z
exp[(y + k
x

2
M
)
2
/2
2
M
)],
(100)
48
where L
y
is the size of the sample in the y direction (L
x
L
y
L
z
=
V = 1). The wave functions are normalized such that
_
d
3
r[
k
x
,k
z
[
2
=
2
0
(101)
(since I set the volume of the system to 1). The prefactors
are chosen such that
2
0
represents the average superuid den-
stity. One important aspect of the particle in a eld problem
seen from the above solution is the large degeneracy of the
ground state: the energy is independent of the variable k
x
,
for example, corresponding to many possible orbit centers.
We have now found the wavefunction which minimizes <

H
kin
>. Substituting back into (89), we nd using (99)
F = [a
0
(T T
c
) +
e

H
2m

c
]
_
d
3
r[[
2
+ b
_
d
3
r[[
4
. (102)
When the coecient of the quadratic term changes sign, we
have a transition. The eld at which this occurs is called the
upper critical eld H
c2
,
H
c2
(T) =
2m

ca
0
e

(T
c
T). (103)
What is the criterion which distinguishes type-I and type II
materials? Start in the normal state for T < T
c
as shown in
Figure 3, and reduce the eld. Eventually one crosses either
H
c
or H
c2
rst. Whichever is crossed rst determines the
nature of the instability in nite eld, i.e. whether the sample
expels all the eld, or allows ux (vortex) penetration (see
section C).
49
c1
H
H
H
Meissner phase
T
H
Normal state
c2
c
Figure 11: Phase boundaries for classic type II superconductor.
In the gure, I have displayed the situation where H
c2
is
higher, meaning it is encountered rst. The criterion for the
dividing line between type 1 and type II is simply
[
dH
c
dT
[ = [
dH
c2
dT
[ (104)
at T
c
, or, using the results (85) and (103),
(m

)
2
c
2
b
(e

)
2
=
1
2
. (105)
This criterion is a bit dicult to extract information from in
its current form. Lets dene the GL parameter to be the
ratio of the two fundamental lengths we have identied so far,
the penetration depth and the coherence length:
=

. (106)
Recalling that

2
=
m

c
2
4e

2
n

s
=
m

c
2
b
2e

2
a
(107)
50
and

2
=
1
2m

a
. (108)
The criterion (58) now becomes

2
=
m

c
2
b/2e

2
a
1/2m

a
=
(m

)
2
c
2
b
e

2
=
1
2
. (109)
Therefore a material is type I (II) if is less than (greater
than)
1

2
. In type-I superconductors, the coherence length is
large compared to the penetration depth, and the system is
sti with respect to changes in the superuid density. This
gives rise to the Meissner eect, where n
s
is nearly constant
over the screened part of the sample. Type-II systems cant
screen out the eld close to H
c2
since their stiness is too
small. The screening is incomplete, and the system must de-
cide what pattern of spatial variation and ux penetration is
most favorable from an energetic point of view. The result is
the vortex lattice rst described by Abrikosov.
6.3 Vortex Lattice
I commented above on the huge degeneracy of the wave func-
tions (100). In particular, for xed k
z
= 0, there are as many
ground states as allowed values of the parameter k
x
. At H
c2
it
didnt matter, since we could use the eigenvalue alone to de-
termine the phase boundary. Below H
c2
the fourth order term
becomes important, and the minimization of f is no longer an
eigenvalue problem. Lets make the plausible assumption that
if some spatially varying order parameter structure is going
51
to form below H
c2
, it will be periodic with period 2/q, i.e.
the system selects some wave vector q for energetic reasons.
The x-dependence of the wave functions

k
x
,k
z
=0
=
0
(

2
M
L
2
y
)
1/4
e
ik
x
x
exp[(y+k
x

2
M
)
2
/2
2
M
)]. (110)
is given through plane wave phases, e
ik
x
x
. If we choose k
x
=
qn
x
, with n
x
=integer, all such functions will be invariant
under x x + 2/q. Not all n
x
s are allowed, however: the
center of the orbit, k
x

2
M
should be inside the sample:
L
y
/2 < k
x

2
M
= q
2
M
n
x
< L
y
/2, (111)
Thus n
x
is restricted such that

L
y
2q
2
= n
max
/2 < n
x
< n
max
/2 =
L
y
2q
2
(112)
and the total number of degenerate functions is L
y
/(q
2
M
).
Clearly we need to build up a periodic spatially varying
structure out of wave functions of this type, with centers
distributed somehow. What is the criterion which determines
this structure? All the wave functions (110) minimize <

H
kin
>, and are normalized to
_
d
3
r[[
2
= [
0
[
2
. They are
all therefore degenerate at the level of the quadratic GL free
energy, F =
_
d
3
r[[
2
+ <

H
kin
>. The fourth order term
must stabilize some linear combination of them. We therefore
write
(r) =

n
x
C
n
x

n
x
(r), (113)
52
with the normalization

n
x
[C
n
x
[
2
= 1, which enforces
_
d
3
r[(r)[
2
=

2
0
. Note this must minimize <

H
kin
>. Lets therefore
choose the C
n
x
and q to minimize the remaining terms in F,
_
d
3
r[a[[
2
+b[[
4
]. Substituting and using the normalization
condition and orthogonality of the dierent
k
z
,k
x
, we nd
f = a
2
0
+

b
4
0
. (114)
with
a(H, T) = a
0
(T T
c
) +
e

H
2m

c
=
e

2m

c
(H H
c2
(T)), (115)

b(H) = sb, (116)

and
s = (

2
M
L
2
y
)
1
n
max

n
x1
,n
x2
,n
x3
,n
x4
C

n
x1
C

n
x2
C
n
x3
C
n
x4
_
dz
_
dx e
iq(n
x1
n
x2
+n
x3
+n
x4
)x

_
dy e

1
2
2
M
[(y+qn
x1

2
M
)
2
+(y+qn
x2

2
M
)
2
+(y+qn
x3

2
M
)
2
+(y+qn
x4

2
M
)
2
]
. (117)
The form of f[
0
] is now the same as in zero eld, so we
immediately nd that in equilibrium,

0
[
eq
= (
a
2

b
)
1/2
. (118)
and
f =
a
2
4

b
. (119)
This expression depends on the variational parameters C
n
x
, q
only through the quantity s appearing in

b. Thus if we mini-
mize s, we will minimize f (remember b > 0, so f < 0). The
53
minimization of the complicated expression (117) with con-
straint

n
x
[C
n
x
[
2
= 1 is dicult enough that A. Abrikosov
made a mistake the rst time he did it, and I wont inict the
full solution on you. To get an idea what it might look like,
however, lets look at a very symmetric linear combination,
one where all the C
n
x
s are equal:
C
n
= n
1/2
max
. (120)
Then
(r)

n
e
inqx
exp[(y + nq
2
M
)
2
/2
2
M
], (121)
which is periodic in x with period 2/q,
(x + 2/q, y) = (x, y), (122)
and periodic in y with period q
2
M
, up to a phase factor
(x, y + q
2
M
) = e
iqx
(x, y). (123)
in a suciently large system. Note if q =

2/
M
, [[
2
forms
a square lattice! The area of a unit cell is (2/q) (q
2
M
) =
2
2
M
, and the ux through each one is therefore

cell
= 2
2
M
H = 2
c
e

H
H =
hc
2e
=
0
(124)
where I inserted a factor of in the last step. We havent
performed the minimization explicitly, but this is a charac-
teristic of the solution, that each cell contains just one ux
quantum. The picture is crudely depicted in Fig. 12a). Note
by symmetry that the currents must cancel on the boundaries
54
s=1.18 s=1.16
Figure 12: a) Square vortex lattice; b) triangular vortex lattice.
of the cells. Since

j
s
= en
s
v
s
, integrating +
2e
c

A = 0
around each square must give, as in our previous discussion
of ux quantization in a toroid,

cell
= n
0
, n = integer. (125)
Somehow the vortex lattice consists of many such rings. The
problem with this idea is that the only way
_
d

=
around the boundary can be nonzero and the usual argument
about single-valuedness of the wave function carried through
is if there is a hole in the wave function. If there is no hole,
or region from which the wave function is excluded, the path
can be shrunk to a point, but the value of the integral must
remain the same since the integrand is the gradient of a scalar
eld. This is unphysical because it would imply a nite phase
change along an innitesimal path (and a divergence of the
kinetic energy!) The only way out of the paradox is to have
the system introduce its own hole in itself, i.e. have the am-
plitude of the order parameter density [[
2
go to zero at the
center of each cell. Intuitively, the magnetic eld will have an
accompanying maximum here, since the screening tendency
will be minimized. This reduction in order parameter ampli-
tude, magnetic ux bundle, and winding of the phase once by
55
2 constitute a magnetic vortex, which Ill discuss in more
detail next time.
Assuming C
n
= constant, which leads to the square lattice
does give a relatively good (small) value for the dimensionless
quantity s, which turns out to be 1.18. This was Abrikosovs
claim for the absolute minimum of f[[[
2
]. But his paper
contained a (now famous) numerical error, and it turns out
that the actual minimum s = 1.16 is attained for another set
of the C
n
s, to wit
C
n
= n
1/2
max
, n = even (126)
C
n
= in
1/2
max
, n = odd. (127)
This turns out to be a triangular lattice (Fig. 12b), for which
the optimal value of q is found to be
q =
3
1/4

1/2

M
, (128)
Again the area of the unit cell is 2
2
M
, and there is one ux
quantum per unit cell.
6.4 Properties of Single Vortex. Lower critical
eld H
c1
Given that the ux per unit cell is quantized, it is very easy to
see that the lattice spacing d is actually of order the coherence
length near H
c2
. Using (103) and (88) we have
H
c2
=
c
e

2
=

0
2
2
. (129)
56
On the other hand, as H is reduced, d must increase. To see
this, note that the area of the triangular lattice unit cell is
A =

3d
2
/2, and that there is one quantum of ux per cell,
A =
0
/H. Then the lattice constant may be expressed as
d =
4

3
(
H
c2
H
)
1/2
. (130)
Since is the length scale on which supercurrents and
magnetic elds vary, we expect the size of a magnetic vor-
tex to be about . This means at H
c2
vortices are strongly
overlapping, but as the eld is lowered, the vortices separate,
according to (124), and may eventually be shown to inuence
each other only weakly. To nd the structure of an isolated
vortex is a straightforward but tedious exercise in minimizing
the GL free energy, and in fact can only be done numerically
in full detail. But lets exploit the fact that we are inter-
ested primarily in strongly type-II systems, and therefore go
back to the London equation we solved to nd the penetration
depth in the half-space geometry for weak elds, allow n
s
to
vary spatially, and look for vortex-like solutions. For example,
equation (75) may be written

B =

B. (131)
Lets integrate this equation over a surface perpendicular to
the eld

B =

B(x, y) z spanning one cell of the vortex lattice:

2
_
(

B) d

S =
_

B d

S, (132)

2
4
c
_

j
s
d

=
0
. (133)
57
But we have already argued that

j
s
d

should be zero on
the boundary of a cell, so the left side is zero and there is
a contradiction. What is wrong? The equation came from
assuming a two-uid hydrodynamics for the superconductor,
with a nonzero n
s
everywhere. We derived it, in fact, from
BCS theory, but only for the case where n
s
was constant.
Clearly there must be another term in the equation when a
vortex-type solution is present, one which can only contribute
over the region where the superuid density is strongly varying
in space, i.e. the coherence length-sized region in the middle
of the vortex where the order parameter goes to zero (vortex
core). Lets simply add a term which enables us to get
the right amount of ux to Eq. (131). In general we should
probably assume something like

B +

B =
0
g(r) z (134)
where g(r) is some function which is only nonzero in the core.
The ux will then come out right if we demand
_
d
3
rg(r) =
1. But lets simplify things even further, by using the fact
that : lets treat the core as having negligible size,
which means it is just a line singularity. We therefore put
g(r) = (r). Then the modied London equation with line
singularity acting as an inhomogeneous source term reads

B +

B =
0

2
(r) z (135)

2
1

(
B
z

) + B
z
=
0

2
(r), (136)
58
where is the radial cylindrical coordinate. Equation (91)
has the form of a modied Bessels equation with solution:
B
z
=

0
2
2
K
0
(

). (137)
The other components of

B vanish. If you go to Abramowitz
& Stegun you can look up the asymptotic limits:
B
z
=

0
2
2
[log(

) + 0.116], < (138)

B
z
=

0
2
2

2
e
/
, . (139)
Note the form (93) is actually the correct asymptotic solution
to (91) all the way down to = 0, but the fact that the
solution diverges logarithmically merely reects the fact that
we didnt minimize the free energy properly, and allow the
order parameter to relax as it liked within the core. So the
domain of validity of the solution is only down to roughly the
core size, , as stated. In Figure 5 I show schematically
the structure of the magnetic and order parameter proles in
an isolated vortex. The solution may now be inserted into the
free energy and the spatial integrals performed, with some
interesting results:
F = L
z

2
0
16
2

2
log(). (140)
It is easy to get an intuitive feel for what this means, since
if we assume the eld is uniform and just ask what is the
59
core

0
supercurrents
||
B
z
Figure 13: Isolated vortex
magnetic energy, we get roughly
F
v
=
1
8
vortex volume B
2
(141)

1
8
(
2
L
z
) (
0
/
2
)
2
(142)
= L
z

2
0
8
2

2
, (143)
the same result up to a slowly varying factor. Now the lower
critical eld H
c1
is determined by requiring the Gibbs free
energies of the Meissner phase with no vortex be equal to the
Gibbs free energy of the phase with a vortex.
23
G diers from
F through a term
_
BH/4. In the Meissner state G = F,
so we may put
F = F + E
line
L
z

1
4
H
c1
_
Bd
3
r (144)
= F + E
line
L
z

1
4

0
L
z
, (145)
23
We havent talked about the Gibbs vs. Helmholtz free energy, but recall the Gibbs is the appropriate potential
to use when the external eld H is held xed, which is the situation we always have, with a generator supplying
work to maintain H.
60
where E
line
is the free energy per unit length of the vortex
itself. Therefore
H
c1
=
4E
line

0
(146)
is the upper critical eld. But the line energy is given precisely
by Eq. (95), E
line
=

2
0
16
2

2
log(), so
H
c1
(T) =

0
4
2
log(). (147)
6.5 Josephson Eect
In 1962 Brian Josephson
24
, then a 22year old graduate stu-
dent, made a remarkable prediction that two superconductors
separated by a thin insulating barrier should give rise to a
spontaneous (zero voltage) DC current, I
s
= I
c
sin , where
is the dierence in phase across the junction. And that
if a nite (DC) voltage were applied, an AC current with fre-
quency = 2eV/ would ow. I
c
is called the Josephson
critical current.
There is a myth that Brian Josephson did his calculation
(1962) and won the Nobel prize (1973) as part of the solu-
tion to a homework problem of Phil Andersons. The truth
is that Anderson was a lecturer on sabbatical at Cambridge
in 1961-62, and he gave a series of lectures in which he men-
tioned the problem of tunneling between two superconductors,
which Josephson then promptly solved. The idea was opposed
at rst by John Bardeen, who felt that pairing could not exist
24
Phys. Lett. 1, 251 (1962)
61
in the barrier region
25
. Thus much of the early debate cen-
tered on the nature of the tunneling process, whereas in fact
today we know that the Josephson eect occurs in a variety of
situations whenever two superconductors are separated by a
weak link, which can be an insulating region, normal metal,
or short, narrow constriction.
26
Lets rst consider the last example as the conceptually sim-
plest. The GinzburgLandau equation appropriate for this
situation may be written

2
d
2
f
dx
2
+ f f
3
= 0 (148)
where =
_

2m

a(T)
is the GL coherence length and f(x)
(x)/

. Take L , so the deviations of coming from

the bulk value
1
of the rst SC is small, and vice versa for
the second SC. Changes of in the constriction occur over
a length scale of L, so that the rst term is of O((/L)
2
)
f f
3
. So we must solve a Laplace equation for f, (
d
2
f
dx
2
= 0)
25
Physics Today, July 2001
26
In Je. 2009 I received an email from Brian Josephson correcting this version of the history:
Date: Wed, 10 Jun 2009 [Link] +0100
From: Brian Josephson <bdj10@[Link]>
To: pjh@[Link]
Subject: the Josephson myth
Dear Peter,
While browsing I came across your mention of the myth that I discovered the effect because
of a problem set by Anderson. Your correction is not completely correct either! It was
Pippard, my supervisor, who drew my attention to Giaevars tunnelling expts. and his theory,
which started me thinking (especially as to how one could get away without using coherence
factors). Anderson on the other hand told me of the Cohen/Falicov/Phillps calculation involving
a single superconductor when it came our in PRL, which gave me the idea of how to do the two-sc.
case. Previously I had got the broken symmetry idea which was crucial from a number of papers
including Andersons pseudospin model, and also expounded in his lecture course which I went to.
Best regards, Brian J.
62
with B.C. f(0) = 1, f(L) = e
i
. The solution will be
f =
_
1
x
L
_
+
x
L
e
i
. (149)
The solution can be thought of as two terms, the rst
1
,
beginning to leak into the constriction from the left, the
second
2
leaking into the constriction from the right. The
GL expression for the current will be
j =
e

2m

i
_

2
m

A
. .
zero
=
e

2m

__
1
x
L
+
x
L
e
i
__

1
L
+
1
L
e
i
_
c.c
_
=
e

L
sin , (150)
which means that the current will be
I = I
c
sin , (151)
I
c
=
e

L
A, (152)
where A is the crosssection.
Given that we have two weakly coupled QM systems, it is
not unreasonable (justied by microscopic theory) to write
down coupled Schrodinger equations
i

1
t
= E
1

1
+
2
(153)
i

2
t
= E
2

2
+
1
(154)
63
1
2

x
y
z
Figure 14:
where H
(i)
0

i
= E
i

i
and E
1
= E
2
= E
0
if the supercon-
ductors are identical. Take [
i
[
2
to be the density of pairs in
SC
i

i
=

n
i
e
i
i

i
=
1
2

n
i
n
i
e
i
i
+ i

n
i

i
e
i
i

n
1
2

n
1
+ i

n
1

1
= iE
1

n
1
i

n
2
e
i(
2

1
)
(155)
n
2
2

n
2
+ i

n
2

2
= iE
2

n
2
i

n
1
e
i(
1

2
)
. (156)
If we take the real parts and use n
1
= n
2
we get
n
1
2

n
1
=

n
2
sin(
2

1
)

n
1
2

n
2
=

n
1
sin(
1

2
)
n
1
= 2

n
1
n
2
sin(
2

1
), (157)
Note Ive put V=A=1. Then the current is just j = 2e n
1
. If
64
we take the imaginary parts we have

n
1

1
= E
1

n
1

n
2
cos(
2

1
) (158)

n
2

2
= E
2

n
2

n
1
cos(
1

2
), (159)
and by subtracting and assuming n
1
n
2
(lets couple 2
identical superconductors at rst) we get

2
= E
2
E
1
= 2e(V
1
V
2
) (160)
where for the second equality we used the fact that the po-
tential dierence between the superconductors shifts the pair
energies by 2eV . So we see that a nite voltage dierence
leads to a time changing phase dierence which means an
AC current via Eq. (157).
Magnetic elds. Now put a ux through the junction where
the B eld is along the y direction and A = Bx z. The
phase of the wave function must change by

2e
c
_
dS A (161)
for the theory to be gauge invariant. Notice that is now
spacedependent. So the Josephson equations will read
j = 4e

n
1
n
2
. .
j
c
sin
_

2e
c
_
2
1
dS A
_
, (162)
2e(V
1
V
2
) =

t
_

2e
c
_
2
1
dS A
_
, (163)
and since
_
2
1
S A =
_
d
0
dz(Bx) = Bxd, (164)
65
/
J
0 1 2
0
c
Figure 15:
we will have
J =
_
L
0
dxj(x) =
_
L
0
dxj
c
sin
_

2e
c
Bxd
_
=
Lj
c
2/
0
_
cos cos
_
+
2

0
__
, (165)
where
0
=
2c
2e
. What is the maximum current through the
junction for all possible ?
We have to calculate
dJ
d
= 0 which leads to the relation
tan = cot
_
/
0
_
(166)
and with a bit of tedious trigonometry to
J
c
= Lj
c

sin
_
/
0
_
/
0

. (167)
This formula produces the JosephsonFraunhofer interference
pattern. DC SQUID (Superconducting Quantum Interference
Device)
We ignore resistance and capacitance for now. The inside SC
thickness is assumed much greater than and since v
s
= 0
66
1 2
3 4
J
J

Figure 16:
we have = 2A/
0
. The ux will be
=
_
ds A =
_
2
1
ds A +

0
2
_
3
2
ds +
_
4
3
ds A+

0
2
_
1
4
ds
=

0
2
(
3

2
) +

0
2
(
1

4
) +
_
2
1
ds A+
_
4
3
ds A
=

0
2
(
1

2
) +
_
2
1
ds A
. .

12
+

0
2
(
3

4
) +
_
4
3
ds A
. .

34
=
43

12
. (168)
The Josephson current through the SQUID will be
J = J
c
_
sin
12
+ sin
43
_
= J
c
_
sin
12
+ sin(
12
+ )
_
. (169)
Which means that the current oscillates with the ux. And
as a result of that the SQUID can be a sensitive measure of
magnetic elds. In practice we include the capacitance and
resistance of the device.
67