Lecture 9

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Phys 769: Selected Topics in Condensed Matter Physics Summer 2010

Lecture 9: Basics of superconductivity


Lecturer: Anthony J. Leggett TA: Bill Coish
Introduction
Superconductivity refers to a complex of phenom-
ena which are usually found to occur together. It
occurs in a wide variety of metals and alloys, below
a transition temperature (T
c
) which currently ranges
up to 150 K (50% of room temperature). Of the
various constituent phenomena, the two which are
conceptually most important are (1) macroscopic dia-
magnetism and (2) persistent supercurrents. To il-
lustrate the dierence between these two phenomena
in the simplest possible context, consider a thin ring
(thickness the London penetration depth
L
, see
below) through which we can apply an external ux
(t) under Aharonov-Bohm conditions (no magnetic eld inside the metal itself). The
natural unit of ux is the (superconducting
1
) ux quantum
0
h/2e.
Imagine rst that we hold the ux steady at a value
0
(actually, any value <
0
/2 would
do). If the ring is normal (i.e. non-superconducting) such a ux has no eect; in particular,
no circulating current is induced in the ring. This will be the case for the superconductor
above its transition temperature T
c
. Suppose now that, holding the ux constant, we cool
through T
c
. We nd that a current starts to ow, with an amplitude that increases roughly
linearly with (T
c
T) and at zero temperature corresponds to circulation of all the electrons
in the ring with velocity (h/2m)(/
0
). It is clear that this phenomenon cannot simply be a
consequence of the innite conductivity which we will discuss below; innite conductivity
will prevent the system attaining equilibrium with the (stationary) crystal lattice, but once
such equilibrium has been attained (at T > T
c
), it cannot take the system out of it! Thus,
the macroscopic diamagnetism is a genuinely thermodynamic equilibrium phenomenon.
Of course, in the vast majority of experiments what is directly observed is not macroscopic
diamagnetism itself but the resulting Meissner eect the complete expulsion of a magnetic
1
In the theory of the quantum Hall eect, one sometimes denes 0 = h/e
1
eld from a simply connected (or in the simplest case also a multiply connected) supercon-
ductor. Imagine that we increase the thickness d of the ring. Then according to Maxwells
equations (applicable to a superconductor equally to a normal metal) the circulating current
will produce its own magnetic eld (even though in an Aharonov-Bohm geometry the exter-
nal magnetic eld was zero everywhere in the metal), and this in turn will aect the value
of the circulating current j(r). When we combine the London equation which quantitatively
describes the diamagnetism, namely (at T = 0)
j(r) =
ne
2
m
A(r) (1)
with the standard Maxwells equation (A(r)) =
0
j(r), we nd that both the eld
HHH(r) (and in the simple case considered here, the vector potential A(r)) and the current j(r)
fall o with distance z from the surface of the metal as e
z/
L
, where
L
(ne
2

0
/m)
1/2

c/
p
a few hundred

Ain most materials (
p
plasma frequency). Moreover, since the
only relevant magnetic variable is the vector potential A(r), we can now forget about the
Aharonov-Bohm condition and instead consider a simply connected sample the result
is the same. The magnetic eld (and current it produces) are screened out over a distance

L
. Hence, for any sample of dimension
L
, the magnetic eld is totally excluded
from the bulk superconductor.
Persistent currents
Now imagine going back to the original geometry, that we start in the superconducting
state (i.e. at T < T
c
) with an AB ux equal to n
0
(n =integer). It turns out that for
this special value of ux no diamagnetic current is induced. We now turn the ux down
to zero: according to Faradays law, this produces a transient e.m.f. V (t) = d(t)/dt
around the ring. In either a normal or a superconducting ring, this voltage will induce a
current. However, after some time is stationary at 0, so V = 0, and in a normal-metal
ring the current would decay over a timescale L/R (L = self-conductance of the ring, R =
resistance), typically s. However, if the metal is superconducting what is observed is
taht the current, once induced, circulates indenitely (experimental lower limit on lifetime
10
15
yrs.). The circulating-current states cannot be the thermodynamic equilibrium states
(reductio ad absurdum!) astronomically long-lived metastable state.
(Other phenomena: vortices (in type-II superconductors), Josephson eect, vanishing Peltier
coecient, London moment . . .).
2
Phenomenology (London, Ginzburg-Landau)
Can obtain a unied account if we assume (a) existence of an order parameter (macro-
scopic wavefunction, GL) (r) which has the general nature of a Schrodinger wave func-
tion in particular, is complex scalar and couples to an electromagnetic eld by minimal
coupling (i i e

AAA: following BCS, e

= 2e) (b) The order parameter is


thermodynamically stable, i.e. changing [[
2
from its equilibrium value costs energy.
Crucial point: if (r) [(r)[ exp i(r), then
_
dl = 2n (2)
(1) Macroscopic diamagnetism: basically the same argument as for the diamagnetism of a
rare-gas atom, i.e.:
in a magnetic ux A(r),
(r) 2eA(r)/, j =
e
m
[[
2
( 2eA/) (3)
but for
0
,
_
dl should remain 0, so
j(r) = const.
_
2e

A dl const.(/
0
) (const. [[
2
). (4)
(2) Persistent currents:
Initially, by hypothesis = p
0
(p integral) and circulating current is zero. By Eq. (3)
this
_
dl = 2p. Suppose that as (t) is decreased to zero p stays constant (cf.
below). Then after the end of the pulse, we have
_
dl = 2p, j(r) = pne
2
/m (p ,= 0). (5)
As we saw, this state cannot be the stable equilibrium state, which corresponds to
p = 0. But since is the phase of a complex scalar, p is a topological invariant (wind-
ing number cf. slinky): as long as [(r)[ / 0, cannot change.
Why does this argument not apply, e.g., to an electron in an
atomic p orbital (semiclassical approximation)? (see gure)
(t) = a(t)
p
() +b(t)
s
() ([a[
2
+[b[
2
= 1) (6)
E(t) = [a(t)[
2
E
p
+[b(t)[
2
E
s
(7)
hence if b(t) increases steadily from 0 to 1, E(t) monotonically
decreases (and ( : t) must 0 at some value of and t).
3
This would not work if the energy contained a term e.g.
E
nl
A
_
[()[
4
d (8)
In the Schrodinger case, such a term is forbidden by linearity. But in the Ginzburg-
Landau case, we can perfectly well have a term
_
A[(r)[
4
dr (A > 0) (9)
topological stability.
(Note (non)-generalization to SU(2) symmetry)
Microscopic interpreation of the order parameter
A clue as to the nature of the order parameter which occurs in the Ginzburg-Landau
phenomenology is given by the fact that the behavior of superuid liquid helium, which
is conceptually very similar to that of superconductors, can be understood microscopically
in terms of the phenomenon of Bose-Einstein condensation, which is reliably predicted to
occur, for a noninteracting Bose gas, below a critical temperature T
BEC
given by
T
BEC
= 3.31n
2/3

2
/mk
B
(10)
Although liquid He is of course by no means a noninteracting gas, it is believed that similarly
below a critical temperature T

a macroscopic number N
0
of atoms (a nite fraction of the
total) occupy a single one-particle state
0
(r). The order parameter for this system can
then be dened by the simple relation
(r)
_
N
0

0
(r) (11)
and has the properties required above (with e

0, of course). This suggests that the order


parameter in a superconductor may have a similar microscopic basis and we now explore
this possibility
2
.
A gas of noninteracting fermions is described in thermal equilibrium by the Fermi-Dirac
distribution, which allows a maximum of one particle per state; thus there is no question
of BEC or anything similar occurring. Consider, however, a very dilute system of fermions
(say for deniteness fermionic atoms, e.g. D) with an attractive interaction sucient to
2
a large part of the rest of this lecture is adapted slightly from sections 2, 5.2, and 5.3 of my book
Quantum Liquids (Oxford Graduate Texts, Oxford 2006)
4
bind two atoms into a molecule; for the moment let us assume for simplicity that there is
only one molecular bound state and that it is s-wave, and (perhaps less physically) that its
radius is large compared to the atomic size (but small compared to the average interatomic
distance). Such a molecular complex should behave as a boson, and if the gas is suciently
dilute we should be able to neglect intermolecular interactions to a rst approximation;
thus, applying the results of the previous section, we expect BEC of the molecules to set in
below a critical temerature given by Eq. (10), where n is now the density of the molecules
and m their mass; note that this temperature is, up to a factor of order unity, just the Fermi
temperature which would have characterized the atoms in the absence of interactions. Note
however that in the formation of the individual molecules themselves the Fermi statistics
played no role.
Now imagine that we gradually increase the density, while keeping the attractive interaction
for the moment constant. When the molecules start to overlap (i.e. when the intermolecular
distance becomes comparable to the molecular radius) we can of course no longer ignore
the molecule-molecule interactions. However, equally importantly, we can no longer ignore
the eects of the underlying Fermi statistics. A simple argument regarding this point goes
as follows: If a molecule has a radius r
0
(dened e.g. by the root-mean-square value of
the interatomic distance in the bound state), then by the uncertainty principle the kinetic
energy of the single-particle states involved in forming it is of order
2
/mr
2
0
E
0
. If we
work in volume , then the total number of single-particle states available in this energy
range is of order (2mE
0
/
2
)
3/2
r
3
0
. Thus for N molecules the average occupation
f of a given single-particle state is nr
3
0
, where n N/ is the density. When f is
small compared to unity (the dilute limit) the veto imposed by Fermi statistics on double
occupation is essentially irrelevant; however, when f nr
3
0
becomes comparable to one (i.e.
the molecules begin to overlap appreciably) then it is essential to take it into account.
Indeed, under those conditions the concept of an identiable molecule loses its meaning;
it is impossible to say which atom is paired with which.
In general we do not have much prior knowledge of how a given system of Fermi atoms will
behave under these dense conditions; we cannot for example necessarily exclude a priori
the possibility that it will form a solid. However, it is certainly not absurd to imagine
that a feature which is qualitatively similar to the formation of diatomic molecules and
their BEC which occurs in the dilute limit might persist as the density is increased. It
turns out, perhaps contrary to intuition, that the problem stated, which is dicult in the
intermediate regime, actually becomes easier to analyze again in the extremely dense limit
when the radius of the (putative) molecules is much larger than the interparticle spacing.
5
This is the so-called BCS limit, and was analyzed by Bardeen et al. in their classic 1957
paper; in this limit the molecules are called Cooper pairs. We will discuss the BCS
theory below. Whether one thinks of the process of Cooper pairing as a kind of BEC or
as something completely dierent is perhaps a matter of taste; however, it is important to
appreciate that it diers qualitatively from the BEC of dilute di-fermionic molecules in at
least two respects. In the rst place, it turns out (again perhaps somewhat contrary to
intuition) that the Fermi degeneracy which occurs in the high-density limit actually helps
the process of pair formation, so that even a two-particle attraction which is too weak to
bind a molecule in free space can nevertheless induce Cooper pairing. Secondly, while in
the dilute-gas limit the process of formation of the diatomic molecules can be thought of as
quite dierent from that of their BEC (so that the typical temperature at which dissociation
of the molecules into two free atoms occurs is orders of magnitude larger than T
c
for BEC),
in the BCS limit the processes of formation fo the pairs and of their condensation are
essentially identical it turns out to be thermodynamically unfavorable, in this limit, to
form pairs which are not condensed.
Despite these dierences, it is important to appreciate that there is no clear qualitative
distinction between the processes of BEC in the dilute-gas limit and of Cooper pairing in
the ultradense one, and indeed it is possible to construct an ansatz for the many-body ground
state wave function which interpolates smoothly between these limits: In recent years it
has become possible to probe this BEC-BCS crossover experimentally in ultracold Fermi
gases.
We are actually going to be interested not in fermionic atoms but in electrons, and this
poses a prima facie problem: The above argument rested on the assumption that the inter-
atomic interaction was attractive, but the bare interaction between electrons, namely
the Coulomb interaction, is uniformly repulsive! The resolution involves two dierent fea-
tures: (a) the interaction between any given pair of electrons is strongly screened by all the
remaining electrons, and hence the repulsion is much attenuated at distances beyond the
Debye (or Thomas-Fermi) screening length, typically 1

A. (b) there exists in addition an
indirect interaction mediated by the exchange of virtual phonons, which is attractive for
small enough exchanged energy and may, depending on the metal in question, outweigh the
screened Coulomb repulsion. So we are in business.
It is doubtful that the classic BCS theory, which has proved spectacularly successful in
accounting for the properties of essentially all superconductors discovered before 1975 (thats
a lot!), is equally applicable to all of the more recent classes (organics, heavy fermions,
cuprates, ferropnictides, ...). However, it covers the materials (Al and Nb) most commonly
6
employed in Josephson devices and with a slight generalization plausibly also superuid
3
He and Sr
2
RuO
4
; hence it is appropriate to review it in the present context.
Before plunging into the full details of BCS theory, it may be useful to reproduce a very
simple model calculation, originally due to Cooper, which historically played an important
role in the genesis of the full theory. It seems not always to be appreciated how useful this
toy model and simple generalizations of it can be, in particular in giving one a physical feel
for which kinds of ects are likely to inhibit (or not) the formation of the superconducting
state. I will present the model in its simplest (original) form.
Consider then two Fermi particles, constrained to be in a spin singlet state with center of
mass at rest, so that the two-particle orbital wave function has the generic form

orb
(r
1
, r
2
)
orb
(r
1
r
2
) =

k
c
k
e
ik(r
1
r
2
)
(12)
subject to the conditions (of which the rst is necessary to guarantee the overall Fermi
antisymmetry)
c
k
c
k
,

k
[c
k
[
2
= 1 (13)
With Cooper, let us attempt to take into account the eect of the other N 2 particles
by excluding possible values of [k[ less than the Fermi momentum k
F
. Then, if we dene
the single-particle kinetic energy
k

2
(k
2
k
2
F
)/2m relative to the Fermi energy
F
and moreover dene the energy eigenvalue E to be relative to 2
F
, the time-independent
Schrodinger equation for the c
k
takes the form
c
k
=
1
2
k
E

V
kk
c
k
(14)
where V
kk
is the matrix element of the interaction potential for scattering from states
(k, k) to (k

, k

). We take the latter to have the simple BCS form V


kk
= V
0
(
c

k
)
with
c

F
, and seek a solution where c
k
is independent of the direction of k (i.e. an
s-state) and thus a function only of
k
. Thus Eq. (14) takes the form
c() =
V
0
2 E
_
c
0
d

)c(

) (15)
where ()

k
(
k
) is the (single-spin) density of states. We are interested in the
possibility of a solution with E < 0. Now, were the absolute kinetic energy, the form of
() in three space dimensions would be const.
1/2
, and Eq. (15) would have no bound-state
(E < 0) solution for small enough V
0
. However, in the present (Cooper) problem () tends
7
to the constant N(0) as 0, and in fact may be reasonably approximated by this value
for all <
c
(recall that in the BCS model for V
kk
,
c

F
). Thus Eq. (15) becomes
c() =
N(0)V
0
2 E
_
c
0
d

c(

) (16)
This equation may be solved for the eigenvalue E by integration of both sides over , with
the result (assuming E < 0)
1 = N(0)V
0
_
c
0
d
2 E
=
N(0)
2
V
0
ln
_
2
c
[E[
+ 1
_
(17)
i.e.
E = 2
c
1
e
2/N(0)V
0
1
2
c
e
2/N(0)V
0
(18)
where the second approximate equality applies in the limit V
0
0.
We thus nd that however small the eective interelectron attraction V
0
, a bound state (i.e.
one with energy less than that of two free particles at the Fermi surface) always exists in
the Cooper problem, although its binding energy [E[ tends to zero exponentially as V
0
0.
Note that an essential ingredient in this result is the logarithmic divergence of the integral
on the RHS of Eq. (17) in the limit E 0, which in turn is a result of the fact that the
density of states available for scattering of the two particles remains constant in the limit
that the energy (relative to the Fermi energy cuto) tends to zero. It follows that any eect
(such as the Zeeman eect of a magnetic eld) which tends to destroy the availability of
these low-energy scattering states will tend to suppress the bound state.
Let us now examine the wave function corresponding to the bound state as a function of
the relative coordinate r [r
1
r
2
[. Taking into account that the angular dependence is
that of an s-state, we nd from Eq. (14) that up to a normalization the wave function is
given by the expression
(r) =
1
r

r
_
kc
k
F
cos kr
2
k
+[E[
dk (19)
where k
c
is the wave vector corresponding to
c
. The general structure of (r) is a sum of
terms of the form (cos k
F
r, sin k
F
r) multiplied by functions of r which fall o as (at most)
r
1
for small r but (at least) r
2
for large r, with the crossover between the two types of
behavior occurring at r v
F
/[E[ (v
F
/
c
) exp(2/N(0)V
0
)
c
. Thus the state is bound
and of radius
c
, in the sense that the total probability of nding the particles beyond a
distance r
c
apart tends to zero as r
1
; however, free particles with energy close to the
Fermi energy in a relative s-state (which would have (r) sin k
F
r/(k
F
r)).
It is tempting to try to generalize the Cooper calculation to nite temperature. The most
natural way to do so would seem to be to note that the pair can only occupy the pair of states
8
k , k if both of the latter are previously unoccupied by any of the other N2 electrons,
and that at nite temperature T 1/k
B
the probability of this is (1+exp
k
)
2
(where

k
can have either sign). Thus, it is natural to replace the density of states used in Eq. (15),
namely () = N(0)(), by the expression N(0)(1+e

)
2
, and correspondingly Eq. (17)
is replaced by
1 = N(0)V
0
_

d
(2 E)(1 + exp )
2
. (20)
The singularity at = 0 is now removed, and in fact it is clear (ignoring some messy details)
that the eect is qualitatively similar to replacing the lower limit in Eq. (17) by k
B
T. Thus
the equation should have no negative-energy solution when it is no longer possible to satisfy
a condition of the approximate form
_
c
k
B
T
d
2
> 1/N(0)V
0
(21)
i.e. above a critical temperature T
c
given by the order-of-magnitude estimate
T
c
(
c
/k
B
) exp 2/N(0)V
0
[E[/k
B
(22)
We will see in the next two sections that the three fundamental properties of the Cooper
pair state just described the exponentially small binding energy and critical temperature,
and the exponentially large radius are reected, albeit in a slightly dierent guise, in the
many-body bound state which occurs in the full BCS theory.
BCS theory at T=0
I shall outline here the essentials of the BCS theory of the ground state of a superconducting
metal. The discussion given here diers somewhat from that of the original BCS paper, but
the nal outcome is the same.
Let us consider a xed, even number N of fermions (in our case electrons) of spin 1/2
moving in free space and subject to a weak interaction whose precise form we do not for
the moment specify. It is convenient to subtract from the Hamiltonian the xed quantity
N, which is equivalent to measuring the single-particle energies from ; to make contact
with a realistic situation in which the metal in question can exchange electrons with leads,
etc., it will eventually be necessary to identify with the chemical potential E/N, but
we will verify below that in the context of classical superconductivity the latter may be
consistently approximated by its T = 0 normal-state value, namely the Fermi energy
F
.
9
Our Hamiltonian then consists of a kinetic energy

k
n
k
(23)
(where n
k
is the operator of the number of particles in plane-wave state k with spin )
and a potential energy

V
1
2

ij
V (r
i
r
j
) (24)
We start by dening a class of many-body states which I shall call generalized BCS states,
among which we hope to nd a good approximation to the true ground state of the system.
A generalized BCS state is by denition a state of the form

N
= ^ / (r
1
r
2

2
)(r
3
r
4

4
) . . . (r
N1
r
N

N1

N
) (25)
where ^ (like ^

, ^

below) is a normalization factor and / antisymmetrizes with respect


to the simultaneous exchange of coordinates and spins of any pair of electrons i, j; because
of the presence of this operator, we may as well choose (r
1
r
2

2
) from the outset to be
antisymmetric under the exchange r
1
r
2
,
1

2
. In second-quantized notation
N
takes the more compact form

N
= ^

_
_
_

_ _
drdr

(r, r

: , )

(r)

(r

)
_
_
_
N/2
[vac) (26)
where [vac) indicates the vacuum state. Our task now is to nd the form of the function
(r, r

, , ) which minimizes the sum of expressions Eq. (23) and Eq. (24). This formula-
tion of the problem is very general, and applies not only to classic superconductivity but also
to liquid
3
He and Sr
2
RuO
4
. As we will see below, however, the most physical quantity
is not the function (r, r

: , ) itself, but a related quantity F(r, r

: , ) which is es-
sentially the eigenfunction of the two-particle density matrix associated with a macroscopic
eigenvalue.
In the case of present interest, that of a superconducting metal, we will immediately restrict
our search for the ground state to a sub-class of the class of functions Eq. (25), namely
that dened by the conditions: (a) (r
1
r
2

2
) is a function only of the relative coordinate
r
1
r
2
(i.e. corresponds to center-of-mass momentum zero); (b) the spin structure is a
singlet. In other words we assume, in an obvious notation
(r
1
r
2

2
) = 2
1/2
(
1

2

1

2
) (r
1
r
2
). (27)
10
Then
N
can be written in the form

N
= ^

k
c
k
a

k
a

k
_
N/2
[vac) (28)
where the Fourier components c
k
of (r
1
r
2
) satisfy the condition c
k
c
k
but are
otherwise arbitrary. For convenience, however, we shall choose the normalization of the c
k
s
to satisfy the condition

k
[c
k
[
2
/(1 + [c
k
[
2
) = N/2. Note that the normal ground state
of the metal in the absence of interactions is a special case of the form Eq. (28), with
c
k
= (k
F
k).
The next step in the argument is to express the kinetic and potential energies in terms of
the coecients c
k
. Actually, it turns out to be more useful to work in terms of the quantity
F
k

c
k
1 +[c
k
[
2
(29)
F
k
actually turns out to be the value of the matrix element of the operator a
k
a
k
between
the ground states of the (N 2)-particle and N-particle systems:
F
k
= N 2[ a
k
a
k
[N) (30)
In terms of F
k
the deviation of the expectation value of the single-particle operator n
k
from its normal-state value (k
F
k) is given by
n
k
) =
1
2
_
1
_
1 4[F
k
[
2
_
sgn (
k
) (31)
and we can insert this expression in Eq. (23) to obtain the expectation value of the kinetic
energy (cf. below). With regard to the potential energy

V a delicate point arises. The
general expression for
_

V
_
in any many-body state is
_

V
_
=
1
2

ijmn
V
ijmn
_
a

i
a

j
a
m
a
n
_
(32)
where i, j, m, n denote possible values of the plane-wave vector k and spin characterizing
the single-particle states, and the matrix element V
ijmn
is given explicitly by
V
ijmn
=
2
_
dr
_
dr

V (r r

)e
i(k
i
kn)r
e
i(k
j
km)r

(33)
For a many-body wave function of the generic form Eq. (25) there are three types of terms
_
a

i
a

j
a
m
a
n
_
which are nonzero:
11
1. i = n, j = m
These are the Hartree terms: since the relevant matrix element V
ijmn
is simply a constant
V (0) which is independent of i and j, the corresponding contribution to V ) is
V )
Hartree
=
1
2
V (0)
_

N
2
_

1
2
V (0)N
2
(34)
where the second equality follows because Eq. (28) is by construction an eigenstate of

N. Thus this term is completely insensitive to the values of the individual F


k
and can
be ignored in the optimization process.
The second class of terms which are nonzero in Eq. (32) are
2. i = m, j = n
These are the Fock terms: the corresponding contribution to V ) has the form
V )
Fock
=
1
2

ij
V
ij,ji
n
i
n
j
) =
1
2

ij
V
ij,ji
n
i
) n
j
) (35)
V
ij,ji
=
1
_
dre
i(k
i
k
j
)r
V (r) (36)
where the second equality in Eq. (36) is valid, for a wave function of the generic form
Eq. (28), to order N
1
. Inserting the value of n
i
) from Eq. (31), we see that in general
V )
Fock
is sensitive to the detailed form of the function F
k
, and therefore in principle
needs to be taken into account in the optimization process. I will for the moment ignore,
apparently arbitrarily, this term.
3
The third class of terms which is nonzero in Eq. (32) are the so-called pairing terms:
3. i = j, m = n (i.e. k
i
= k
j
, etc.):
V )
pairing

kk

V
kk

_
a

k
a

k
a
k

a
k

_
(37)
where
V
kk

_
e
i(kk

)r
V (r)dr
_
V (k k

)
_
(38)
These have a very simple form when expressed in terms of the functions F
k
:
V )
pairing
=

kk

V
kk
F
k
F

k
(39)
3
In the original BCS procedure, one starts from a model Hamiltonian which omits the Hartree and Fock
terms.
12
Note that these are (the only) terms which would occur for scattering of two particles with
center-of-mass momentum zero.
Putting together the results Eq. (31) and Eq. (38), we nd that for any state of the form
Eq. (28) the expectation value of the Hamiltonian, up to a constant and the neglected Fock
term, is given by the expression
H) = T) +V )
pair
(40)
=

k
[
k
[(1
_
1 4[F
k
[
2
) +

kk

V
kk
F
k
F

k
(41)
Note that to the extent that we can expand the kinetic energy to lowest nontrivial order
in [F
k
[, Eq. (41) becomes H) =

k
2[
k
[[F
k
[
2
+

kk
V
kk
F
k
F

k
and is identical to the
expression for the total energy in the simple two-particle problem, with F
k
interpreted as
the Fourier component of the relative wave function and [
k
[
k
=
2
k
2
/2m. This is, of
course, not an accident.
Eq. (41) is valid for any generalized BCS wavefunction Eq. (25) subject to Eq. (27). We
now optimize the choice of the function F
k
by setting the derivative of Eq. (41) with respect
to F

k
equal to zero. This gives for F
k
the self-consistent equation

2[
k
[F
k
_
1 4[F
k
[
2
+

V
kk
F
k
= 0 (42)
Eq. (42) is actually nothing but the celebrated BCS gap equation in disguise. We can make
this explicit by introducing the notation
E
k

[
k
[
_
1 4[F
k
[
2
eq
(43)

k

_
F
k
[F
k
[
_
eq
(E
2
k

2
k
)
1/2
(44)
so that
(F
k
)
eq

k
/2E
k
(45)
where eq refers to the equillibrium value of F
k
, i.e. that value which solves Eq. (42).
Then Eq. (42) becomes

k
=

V
kk

2E
k

_
E
k

_

2
k
+[
k
[
2
_
(46)
which is the standard form of the BCS gap equation at T = 0. (Recall that V
kk
is dened as
the matrix element k , k [ V [k

, k

)). I reemphasize that, subject to the neglect of


13
the Fock term, the only assumption necessary to obtain Eq. (46) is that the (approximate)
ground state indeed lies within the subclass (Eq. (27)) of generalized BCS functions Eq.
(25).
Eq. (46) always possesses the trivial solution
k
0; according to Eq. (45) and Eq. (31)
this corresponds to F
k
= 0, n
k
) = (
k
), i.e. to the normal ground state (which as we
have noted is a special case of Eq. (28) and thus of Eq. (25)). The existence and nature
of one or more nontrivial solutions depends, of course, on the specic form of V
kk
. At this
point it is convenient to follow the original BCS paper and choose V
kk
to have the simple
form
V
kk

_
V
0
[
k
[, [
k
[ <
c
0 otherwise
(47)
That this apparently arbitrary choice gives such a good agreement with experiment for most
of the classic superconductors is at rst sight quite surprising, but can be understood from
a renormalization argument (no space for details here).
With the choice Eq. (47), the RHS of Eq. (46) is independent of k, so we can set
k
within
the shell [
k
[ <
c
equal to a constant :

k
=
_
[
k
[ <
c
0 [
k
[ >
c
(48)
where the magnitude of is given by the implicit equation
1 =
1
2
V
0

k
(
2
k
+[[
2
)
1/2
(49)
where the sum goes over the states within the shell [
k
[ <
c
. The phase of , which
according to Eq. (45) is the (uniform) phase of F
k
, is not a physically meaningful quantity,
any more than is the overall phase of the relative wave function in the two-particle problem.
For V
0
< 0 (repulsive interaction) Eq. (46) clearly has no solution, and the ground state
within the ansatz Eq. (28) is the normal one. The interesting case is V
0
> 0, N(0)V
0
1
(weak attractive interaction). In this case Eq. (49) yields the result
[[ 2
c
exp [1/N(0)V
0
] (
c
) (50)
Note that the exponent in Eq. (50) is half the value occurring in the expression for [E[ in
the Cooper problem.
4
With this solution both the value of F
k
, and the deviation n
k
) of the single-particle
occupation number n
k
) from its normal-state value (
k
), are substantial only for
4
This is because the sum over k now runs over negative as well as positive
k
14
[
k
[ [[; in fact, from Eq. (45) and Eq. (31) we have (cf. gure)
F
k
=

2(
2
k
+[[
2
)
1/2
[
k
[
1
(51)
n
k
) =
1
2
_
1

k
(
2
k
+[[
2
)
_
(
k
) [
k
[
2
(52)
where the asymptotic behavior for [[ [[ is indicated. The gure shows that the eect
of pairing on the single-particle distribution n
k
) is qualitatively similar to that of a nite
temperature of order /k
B
; the sharp discontinuity at k = k
F
is removed but the average
Fermi surface is unchanged.
Using Eq. (50) and Eq. (51) we can calculate
Figure 1: Dependence of the quantities (a) F
k
and (b) n
k
at T = 0 on the energy
k
relative to
the Fermi energy.
the value of the terms in Eq. (41) in the limit

c
[[:
5
T) = N(0)
2
_
ln
_
2
c

1
2
_
(53)
V )
pair
= V
0
N
2
(0)
2
ln
2
_
2
c

_
(54)
It should be noted that these results remain
valid even if [[ does not have the equilibrium
value specied by Eq. (50), provided F
k
still
has the form in Eq. (51). Dierentiation of
E() T) ()+V )
pair
() with respect to
gives back the result Eq. (50), and a condensa-
tion energy relative to the normal groundstate
energy of
E
cond
=
1
2
N(0)
2
(55)
Note that relative to the energy of the lled
Fermi sea E
cond
is only a fraction of order (/
F
)
2
,
which for classical superconductors is typically
10
7
10
8
. Consequently the relative shift
in chemical potential is also only of this order,
and can usually be safely neglected.
Let us study the structure of the Fourier transform F(r) of F
k
. This quantity is proportional
to the orbital part of the eigenfunction of the two-particle density matrix associated with
the single macroscopic eigenvalue, and its norm (i.e. the integral of [F(r)[
2
) just gives the
5
The exact result replaces the ln in Eq. (53) by sinh
1
(c/) and the condensation energy Eq. (55) by
N(0)
h

2
c
c(
2
c
+
2
)
1/2
i
=
1
2
N(0)
2
+O((/c)
2
).
15
value N
0
of this eigenvalue; thus, we may regard F(r) (or more precisely the normalized
quantity N
1/2
0
F(r)) as the wave function of the condensate and N
0
as the number
of Cooper pairs. The calculation of N
0
itself is straightforward: from Eq. (51), in the
approximation of a constant density of states N(0) and [[
c
N
0

_
dr[F(r)[
2
=

k
[F
k
[
2
=
2

k
(4E
2
k
)
1
=

4
N(0). (56)
Since in the free-electron (Sommerfeld) approximation N(0) is equal to (3
F
)
1
times the
total density N/ of electrons, the condensate fraction N
0
/N is of order /
F
10
3

10
4
.
We now consider the general form of the condensate wavefunction
F(r) =
1

k
(2E
k
)
1
e
ikr
(57)
where the sum goes over the shell [
k
[ <
c
. This expression is isotropic (F(r) F(r), r
[r[), and in the approximation of a constant density of states may be written
F(r) = N(0)k
2
F
r
1

r
_
cos k
F
rf
_
r

__
,


v
F
[[
(58)
f(z)
_
c/
c/
cos zxdx

1 +x
2
(59)
(where the coecient of a term proportional to sink
F
r vanishes because of the particle-
hole symmetry, i.e. the fact that the integrand is an even function of ).
As in the Cooper problem, the existence of a sharp cuto in F
k
(in this case at [
k
=
c
only)
leads to a power-law behavior of F(r) at long distances. However, unlike in that problem,
it is now possible to argue that this behavior is a pathological consequence of the articial
choice of the pairing potential and will be removed by a more physically reasonable choice.
This justies the extension of the limits of integration in Eq. (59) to , whereupon we
nd that f(z) is just the Bessel function K
0
(z); for z 1 it diverges as ln z
1
, but this
singularity is removed when we go back to a nite cuto, and is of no great interest to
us. More interesting is the behavior for z 1 where we nd f(z) exp(

2z). Thus,
anticipating the fact that the quantity

will turn out to be k


1
F
, we nd that the
approximate form of the pair (condensate) wave function is, for r k
1
F
, v
F
/
c
,
F(r) N(0)
sin k
F
r
k
F
r
e

2r/

(60)
We see that the radius of a pair is of order

. (For historical reasons, in the literature


it is more common to use the quantity
1

v
F
/, which is called the Pippard
16
coherence length. Using the fact that N(0) k
2
F
is of order (v
F
)
1
, we nd that the
integral of [F(r)[
2
is indeed of the order (Eq. (55)). For r

the pair wavefunction is


approximately proportional to that of two particles at the Fermi energy moving freely in a
relative s-wave state. In a typical classical superconductor

is of order 1 m, much larger


than the interelectron spacing ( 1

A).
17

You might also like