Bcs Theory
Bcs Theory
Lecture 7.3
BCS theory
electron attraction, and they adopted a particular cartoon of this attraction, none of that is implicit in
“BCS” as I use the term. However, weak coupling is implicit in this approximation.
2 The fermionic Bogoliubov approach for superconductors was inspired by the bosonic Bogoliubov
approach that models superfluid Bose gas. In that case, the elementary excitations were phonons.
c
Copyright
2009 Christopher L. Henley
756 LECTURE 7.3. BCS THEORY
Quantity Formula
(BCS) Coherence length ξ = ~vF /π∆
BCS gap ∆/Tc = 3.5
Critical temperature Tc = (...)~ω
p 0 exp(−1/N (0)V0 )
Quasiparticle dispersion Ek = |(k) − µ|2 + |∆(k)|2
(BCS coefficients) |uk |2 , |vk |2 = 12 (1 ± ((k) − µ)/Ek )
uk vk = ∆(k)/2Ek
they are adiabatically connected). The weak-coupling assumption hidden in the BCS
derivation(s) implies that the gap (= pair binding) is much smaller than EF , whereas it
is much larger than EF in the tight pair case.
Hamiltonian
Recall (from Lec. 7.0 ) our Hamiltonian is H = HKE + HV where the single-electron
part is
k c†kσ ckσ
X
HKE = (7.3.1)
kσ
with k measured as usual from the chemical potential µ, the spin index σ runs over ↑, ↓.
Sometimes we’ll assume spherical symmetry, with k ≈ ~vF (k − kF ), and unbounded
k, but only where specifically noted. Otherwise, the calculation is valid for generic
dispersions, and k may be a crystal wavevector bounded to a Brillouin zone.
The general form of interaction potential is HV = HV (0) + HV (1) , broken up ac-
cording to whether the electrons being scattered are in a singlet or triplet spin state,
respectively 3 .
X (0,p) †
HV (0) = Vkk0 (ck0 +p/2,↑ c†−k0 +p/2,↓ c−k+p/2,↓ ck+p/2,↑ ) (7.3.2)
pkk0
(0,p) (0,p)
where Vkk0 = −Vk,−k0 ensures the singlet symmetry. The triplet term is
X (1,p) †
HV (1) = Vkk0 ck0 +p/2,σ0 τ σ10 ,σ20 c†−k0 +p/2,σ0 · c−k+p/2,σ1 τ σ1 ,σ2 ck+p/2,σ2 . (7.3.3)
1 2
pkk0
In the case of the phonon-mediated coupling (Lec. 7.6 ), the potential is the same for
both channels. Which term enters the pairing Hamiltonian depends on the symmetry
of the superconducting order. This unit emphasizes singlet superconductivity so, where
it makes a difference, we’ll assume the singlet interaction (7.3.2).
It can be verified that the energy is lower when this is nonzero for only one choice of
p. The case p 6= 0 actually corresponds to two physical situations (i) a nonzero current
(see Lec. 7.8 [omitted]) (ii) an exotic ordering at nonzero wavevector, the “FFLO”
superconductor. But let’s specialize to P = 0 , since that normally has the lowest
energy. 4
In view of (7.3.5) and (7.3.4), the only terms which contribute to the anomalous
part of the Hartree-Fock effective Hamiltonian are the pairing terms 5 of the potential,
defined as those with p = 0:
X (00) † †
Hpair = Vkk0 ck0 ↑ c−k0 ↓ c−k↓ ck↑ (7.3.6)
kk0
In this story, we glossed over the large, non-anomalous (c† c type) expectations.
These same “Hartree” and “Fock” terms are encountered in the variational approach
(Lec. 7.3 Y, but canceled there by an appeal to Fermi liquid theory. The right viewpoint
is that we already dealt with this part of the interactions via the Landau quasiparticle
construction (Lec. 1.7 ). Then our c†k operators really make Fermi-liquid quasiparticles,
and the dispersion k already incorporates the effects of interactions, as long as k is
near the Fermi energy.
Now let’s substitute (7.3.4) into (7.3.6) using the Lec. 1.4Xdecoupling rule: this
produces an effective Hamiltonian, quadratic in electron operators:
X (00) ∗
eff
Hpair = Vkk0 (hc−k0 ↓ ck0 ↑ i c−k↓ ck↑ + hc−k↓ ck↑ ic†k0 ↑ c†−k0 ↓ ) − constV (7.3.7)
kk0
Having collected the terms from (7.3.7), we obtain the grand result of this section, the
quadratic effective Hamiltonian
This is weird, in that it does not conserve particle number: two fermions can be created
from nothing, or annihilated. Actually, this is just like Bogoliubov’s effective Hamilto-
nian for the (superfluid) dilute Bose gas from Lec. 7.2; we prefer to say the particles are
coming from, or going into, the condensate, which functions as a reservoir.
A close analog of the present calculation is the Hubbard model Hartree-Fock calcu-
lations of Lec. 5.3X . In that case, however, an electron creation operator at k merely
got mixed with one at k + Q, whereas now it gets mixed with an annihilation operator.
4 Details: If you have a p for which (7.3.5) is nonzero, that indicates you have ordering of pairs with
Any such quadratic form can be diagonalized, that is, put into the form
†
X
H̃Bogo. = Ek γkσ γkσ (7.3.11)
kσ
†
The fermion operators {γkσ } and {γkσ } create and annihilate the “Bogoliubov quasi-
particles”. They will be linear combinations of the creation and annihilation operators
of the real electrons:
γk↑ ≡ uk ck↑ − vk c†−k↓ , (7.3.12a)
uk = u−k (7.3.14)
As for vk : Though not obvious from what was written above, our assumption of a
superconductor with spatial s-wave pairing implies that vk in (7.3.12a) and (7.3.12b)
may be taken real without loss of generality. More generally, the phase of v k depends on
the orientation of k according to the angular momentum representation of the spatial
pairing. Both of these facts were suggested by the construction in Lec. 7.1 that produced
uk and vk in terms of a pair wave function χ(r1 − r2 ).
[Let’s start on the details of making the Bogoliubov transformation work.] First,
†
what are the necessary conditions for {γkσ } and {γkσ } to have canonical fermion anti-
commutation relations? The self-anticommutator gives a normalization condition:
†
1 = {γk↑ , γk↑ } = {u∗k c†k↑ − vk∗ c−k↓ , uk ck↑ − vk c†−k↓ } = u2k + |vk |2 . (7.3.15)
7 These are not the quasiparticles in the sense of Landau Fermi liquid theory (Lec. 1.7 ), since the
superconducting ground state does not connect adiabatically to the Fermi sea state.
7.3 B. BOGOLIUBOV QUASIPARTICLE OPERATORS 759
Otherwise, anticommutator of any pairs of Bogoliubov operators must give zero. The
only case where this cancellation is not obvious is the pair γk↑ and γ−k↓ , since the c
operators included in the former operator are the conjugates (with different coefficients)
of the c operators in the latter operator. We are demanding
0 = {γk↑ , γ−k↓ } = {uk ck↑ − vk c†−k↓ , u−k c−k↓ + v−k c†k↑ } = uk v−k − vk u−k . (7.3.16)
vk = v−k (7.3.17)
Note: the physical symmetry between k and −k merely implied |v−k | = |vk |. The phase
relationship in (7.3.17) is specific to pairing with an even angular momentum (which
implies the the pair wavefunction is even under spatial inversion.) Since fermions are
antisymmetric under exchange, that happens if and only if the spin symmetry of the
pairing is antisymmetric – as in the spin singlet we are assuming here. If it is strictly
s-wave symmetry, then vk /uk is independent of the direction of k, i.e. uk and |vk | are
functions only of |k|. [However, by a slight abuse of language, the term s-wave symmetry
is normally used whenever the actual symmetry coincides with s-wave symmetry under
rotations which belong to the point group describing the material’s symmetry.] For
antisymmetric (e.g. p-wave) spatial pairing, v−k = −vk instead; that normally entails
a multi-component order parameter, so is discussed among the exotic kinds of pairing
in Lec. 8.1 [2007] and Lec. 8.2 .
Note the arbitariness of the phase of vk in (7.3.17). (even after we fixed uk to be
real). If we wrote vk = |vk |eiθk , we’ll find that the only solution of the self-consistent
equations (below) has the same phase θk ≡ θ, for all k. This phase factor eiθ turns out
to be the same one appearing in the BCS wavefunction; and, as shown by Ex. [7.1.4(c)],
physical expectations don’t depend on θ. Hence θ is a symmetry as was already stated
in Lec. 7.1 . We will normally adopt the gauge θ = 0.
With these conditions, (7.3.12a) and (7.3.12b) define a canonical (unitary) transfor-
mation, which (lucky for us) breaks up into 2 × 2 blocks. The inverse transformation
reads:
†
ck↑ = u∗k γk↑ + vk∗ γ−k↓ (7.3.18a)
†
ck↓ = u∗k γk↓ − vk∗ γ−k↑ (7.3.18b)
where of course u∗k = uk by our convention.
Ek
∆k hole−like electron−like
= 2u kv k= sin 2 β k 1
Ek εk vk 1 uk 1
= ( uk2 v k 2 ) = cos 2 β k
Ek
γ c γ c
kσ −k−σ kσ kσ
k kF
∆ (kF )
εk
2∆
k
kF
Figure 7.3.1: (Left). Functions of wavevector k (or energy), related to angle parameter βk .
[It has been implicitly assumed that ∆ is nearly constant near the Fermi energy.]
p
the ground state energy is −Ek = − 2k + |∆k |2 . As with any other two-level sys-
tem, resonance is most important when the bare asymmetry is zero, in this case when
|k| = kF . From this viewpoint, the BCS gap is much like an avoided level crossing.
In the unpaired situation, when the occupation of this little subspace is just one,
(7.3.9) gives zero, so these states have energy zero. They are none other than the
Bogoliubov quasiparticles; in other words, when either of the terms in γk† actually hits
the BCS state, it makes the same state.
To solve efficiently for uk and vk we substitute Eqs. (7.3.12) into (7.3.11), collect
terms, and demand that the result reproduces the two terms of (7.3.10). The square
terms, compared with (7.3.1), give
∆k
2uk vk = . (7.3.19b)
Ek
of Bogoliubov quasiparticles:
q
Ek = 2k + |∆k |2 . (7.3.20)
Combining (7.3.19a) and (7.3.15) gives the second big result, the magnitudes of the
coefficients themselves:
2 1 k
uk = 1+ (7.3.21a)
2 Ek
1 k
|vk |2 = 1− (7.3.21b)
2 Ek
The uk and vk are the key parameters in this lecture; Sorry, they should have been
plotted in Lec. 7.1 ? see Fig. 7.3.1, in any case. The gap, of course, has a small effect
on the nature of quasiparticles, except right near kF . For k ∆(kF ), in (7.3.12a) and
(7.3.12b), uk ≈ 1, so the Bogoliubov quasiparticle is essentially an electron, as in the
normal state (Fig. 7.3.2), while for k −∆(kF ), vk ≈ 1 and it is essentially a hole.
Note: it’s sometimes handy to use trigonometric functions to parametrize the coef-
ficients: let
uk = sin βk (7.3.22a)
iθ
vk = e cos βk (7.3.22b)
The phase factor should really be eiθk , and we should really assume that, but it can
(will) be shown that all θk must be the same.
Let |Φ0 i be the “vacuum” of the quasiparticles, i.e., the state which is annihilated
by any of the operators {γk }. We can construct it as
Y
|Φ0 i ∝ γkσ |0i (7.3.23)
kσ
where |0i, as usual, denotes the bare vacuum – the state which is annihilated by any of
{ckσ }. Manifestly γkσ |Φ0 i = 0 is indeed satisfied for any (k, σ), since (γkσ )2 = 0.
This |Φ0 i must be the ground state of H̃Bogo. , since it has zero quasiparticles, and
each quasiparticle costs a nonvanishing energy Ek . This fact is the key step for future
manipulations: we now know how to figure out the ground state expectations of any
string of γ and γ † operators. It is the same as the expectation of the corresponding
string of c and c† operators in the vacuum of ordinary electrons (not the Fermi sea!).
Furthermore, if you substitute the definitions of γkσ into (7.3.23), it just reduces to
the BCS wavefunction (modulo a normalization factor):
where |Φ0 i was defined in (7.3.23). The (few) steps are left as an exercise, Ex. 7.3.3.
for k far away from the Fermi surface. It is a central object in BCS theory, as E k and
the coefficients uk and vk [Eqs. (7.3.21)] that enter all pysical expectations depend on
∆k . In this section, we’ll obtain the self-consistent equation for ∆k , and see how the
Bogoliubov approach extends to T > 0.
Gap equation
We still need to solve for ∆k . We just need to evaluate the factor hc−k↓ ck↑ i in (7.3.8),
is now easy to do using Eqs. (7.3.18) to convert electron into quasiparticle operators:
† †
hc−k↓ ck↑ i = uk vk (hγ−k↓ γ−k↓ i − hγk↑ γk↑ i). (7.3.25)
The terms of form γγ or γ † γ † had zero expectation [in |ΦBCS i, which we now know to
be the vacuum of the Bogoliubov operators.] Using (7.3.19b) we get
1 ∆k
hc−k↓ ck↑ i = (1 − hn̂qp qp
k↑ i − hn̂k↓ i) (7.3.26)
2 Ek
in terms of the number expectation defined in (7.3.13). At T = 0, we have hn̂qp
kσ i ≡ 0.
So substituting (7.3.26) back into (7.3.8), we obtain the gap equation
1 X (00) ∆k 0
∆k = − Vkk0 p (7.3.27)
2 0 |k0 |2 + |∆k0 |2
k
This is a nonlinear integral equation for a whole unknown function ∆k , which in general
could only be solved by numeric integration. [You would guess the function ∆k , insert
it in the right-hand side of (7.3.27) which provides a new set of values for ∆k , and keep
iterating until it converges.]
Cooper approximation
Since Cooper’s toy potential is just a constant, all that matters here is the energies k ,
not k itself.
7.3 D. THE GAP EQUATION 763
Let N () be the density of electron states for one spin state only, per site. [The
standard convention takes N () to be the DOS per unit volume, and takes V0 to have
units [energy][volume]. The conversion factor to that convention would be vcell , the
volume of a unit cell.]
We obtain Z +~ωc
1 1
1 = + V0 dN () p . (7.3.30)
2 −~ωc + |∆|2
2
Notice N () ≈ N (0) since ~ωc band , the bandwidth. Further assume ∆ ~ωc ;
otherwise, and the Cooper potential is no longer such a good approximation. The
integral (7.3.30) may be done exactly, but is better R understood by “a poor man’s”
approximation. The integrand is approximately d/ over most of the range, which
is log divergent both at → 0 and → ∞. The actual integrand stops diverging and
levels off once || < |∆. upper cutoff is ~ωc . Thus, the “poor man’s” approximation of
the integral is
Z ~ωc Z ∆ !
d
≈ + . (7.3.31)
∆ ~ωc
The gap equation says
~ωc
1 ≈ N (0)V0 ln + const (7.3.32)
∆
Inverting this function, you obtain the grand formula ∆ ≈ (const)~ωc exp[−1/N (0)V0 ].
From the exact integral of (7.3.30) we get the same grand result for the BCS gap formula,
but with the correct prefactor:
~ωc
∆= ≈ 2~ωc e−1/N (0)V0 (7.3.33)
sinh(1/N (0)V0 )
Rather like the charge-density-wave (CDW) gap of Lec. 3.4 , this formula combines
three parameters: one (N (0)) characterizing the free-electron energy scale, another
(~ωc ) characterizing the phonon energy scale, and a third (V0 ) related to the inter-
action term. In contrast to the CDW case, the third parameter is not directly the
electron-phonon coupling strength g, but is a function of it since V0 = O(g 2 /~ωc ). The
dependence of ∆ on V0 is non-analytic, indeed it is exactly the function exp(−1/z)
that is the cautionary example of analytic function theory. [Its Taylor series is zero
term-by-term.]
It is now evident, in (7.3.33), why the gap ∆ was small compared even to ~ωc . So is
Tc , which (we’ll shortly see) is essential the same quantity as ∆ within BCS theory. So
8 Still,it’s possible to have an instantaneous potential V (R) > 0 for all electron separations R in
(0)
real space, yet Vkk0 < 0 in Fourier space for the wavevectors that dominate scatterings near the Fermi
suface. Then the gap equation has a solution and superconductivity is expected.
764 LECTURE 7.3. BCS THEORY
(7.3.33) would would imply that (provided V0 > 0), every metal goes superconducting
at a sufficiently low temperature. In reality, if the predicted Tc is too small, the system
will be unstable first to some competing kind of correlated electron state, so as to gain
some energy from the interaction terms.
We are now equipped to compute, e.g., expectations of the order parameter operator
introduced in Lec. 7.1 ,
1 X
hΨ(r)i ≡ hc↓(r)c↑ (r)i = hc−k,↓ ck,↑ i (7.3.34)
N
k
The new gap equation looks like (7.3.27) except that a factor tanh(E k0 /2T ) is in-
serted in the right-hand side. If we adopt Cooper’s toy potential, we get
1 X 1 E
k
1= V0 tanh (7.3.36)
2N Ek 2T
k
9I have not shown that this is the correct generalization, and it would be difficult to do so here,
given my choice of an ad-hoc justification of the decoupling.
7.3 E. NONZERO TEMPERATURES 765
Is there a wrong factor, compare with (7.3.27)? As we raise T , the tanh factor gradually
takes over
p the job of suppressing the logarithmic divergence, which was handled by ∆
in Ek ≡ 2k + |∆|2 in the denominator. Correspondingly, the value of ∆ will decrease.
By definition, Tc is the temperature where ∆ = 0 and thus Ek = |k |. At the critical
βc ≡ 1/Tc , we get
Z +~ωc
1 X 1 1 1 1 1
1= V0 tanh( βc |k |) = N (0)V0 d tanh βc || (7.3.37)
2N |k | 2 2 ~ωc || 2
k
using the same approximations as in Sec. 7.3 D. Let’s estimate this integral by the
“poor man’s approximation” as in Sec. 7.3 D. The upper cutoff is ~ωc again, while
the lower cutoff is roughly k ≈ Tc , since that is where the tanh function in (7.3.37)
crosses over from unity to a linear behavior that cancels the 1/|| divergence. Thus T c
is playing the same role in (7.3.37) that ∆ played in the gap equation in Sec. 7.3 D; as
you might anticipate, we have 1 ≈ N (0)V0 [ln(~ωc /Tc ) + const. and Tc ∼ ∆(0). In fact
you can solve (7.3.37) exactly too, which gives the BCS result
Tc = 1.14 ~ωc e1/N (0)V0 ; (7.3.38)
comparing with (7.3.33), we get the ratio
2∆(0)
= 2πe−C ≈ 3.50, (7.3.39)
Tc
where C = 0.577 is Euler’s constant.
A different anomalous quantity is the BCS order parameter Ψ, as noted in Lec. 7.0 .
But its expectation (including phase factor) is proportional to ∆ within BCS theory.
Consequently, ∆ is often used as the order parameter – for both temperature and spatial
dependences. Commonly ∆ is called the “gap” even where it is being used as an order
parameter and the excitation gap is not in mind. See Sec. 7.4 E(?)about the BdG
equations.
Notice that even at T > 0, ∆(T ), is the excitation gap of a quasiparticle. It is
temperature dependent, somewhat like a semiconductor’s bandgap is, but stronger since
it vanishes at Tc .
Coherence length?
BCS theory is just a kind of mean-field theory, so it gives the same critical expo-
nents as the Ginzburg-Landau (GL) theory did. See Lec. 6.1 . I have mentioned the
GINZBURG CRITERION there, which explains the regime of validity for mean-field
theory in terms of T − Tc and the G-L coefficients.
In particular, if we solve (7.3.36) near Tc , we find
The specific heat has a jump at Tc , as in GL theory (or mean-field theory in general).
The exact ratio (for a weak-coupling, s-wave superconductor) of the superconducting
and normal state specific heats at Tc is 11
Cs (Tc ) 12
=1+ = 2.43. (7.3.44)
Cn (Tc ) 7ζ(3)
The specific heat ratio (7.3.44), and especially the gap/Tc ratio (7.3.39), are diagnostic
of weak coupling superconductors, well approximated by BCS theory. In strong-coupling
superconductors (see Lec. 7.7 [omitted 2007]) (7.3.39) is larger.
Also, the T = 0 condensation energy is −Econd = hHi|ΦBCS i − hHi|ΦF i . The result
is
1 1 ∆4
−Fcond (0) ≈ − N (0)∆2 + N (0) (7.3.45)
2 8 (~ωc )2
within BCS theory using the Cooper potential. [Compare this with the GL result!]
through the bulk. But as it travels, it carries no charge; the actual charge is carried
around the surface as a screening current (of the condensate of Cooper pairs.)
from Jörn 2005: surface current is carried by electrons in Cooper pairs.
In the BCS approximation, it sure looks like the quasiparticle is carrying a net
current. But this is, I think, an artifact of omitting the long-range Coulomb interaction
from our Hamiltonian. A superconductor excludes magnetic fields from its bulk, which
means it must exclude any current that could make a magnetic field, which it does by
making an equal and opposite screening current. [This “counterflow” of the condensate
adjacent to the quasiparticle somewhat resembles the counterflow in a classical fluid
with a hard sphere moving through it.] At the other side of the superconductor, the
charge can recombine with quasiparticles to reconstitute electrons.
In exotic, non-pairing theories of superconductivity, elementary excitations called
“spinons” appeared which are spin-1/2 chargeless fermions. It is amusing to find there
is really no line separating the (comparatively) prosaic Bogoliubov quasiparticles –
having spin (as a good quantum number) but zero charge – from the spinons.
I am slightly worried that one could assign physical meaning to the charge of the
quasiparticle within the bulk, by the Aharonov-Bohm interference effects it experiences
whenever it goes around some flux. It appears to me that that could be computed
using the Bogoliubov-de Gennes equations (Sec. 7.4 E(?)) in which each quasiparticle
is explicitly divided into particle and hole components. The flux acts differently on the
electron and the hole parts, so it tends to break up the quasiparticle. This was very
relevant, at least in the late 1990s, to understanding the thermal transport behavior in
the vortex lattice of a d-wave superconductor, which was measured by experiments on
the magnetic field effects on the thermal conductivity.
phonon-mediated coupling, think of the cutoff ωc as being the Debye frequency.15 With
this approximation, ∆k is independent of k so the gap equation becomes an equation
for a single unknown. In fact ∆ enters the equation as the cutoff of an otherwise
logarithmically divergent integral, so common when we have a Fermi surface. It looks
a lot like the CDW gap equation (Lec. 3.4 ). The exponential dependence of ∆ on the
parameters explains the sensitivity of Tc to details.
Actually, it isn’t so hard to solve the full gap equation numerically by iteration
(compare the Eliashberg and McMillan equations in Lec. 7.7 [omitted 2007]). They
still use one simplification: the gap function depends only on k , so the integral equations
involve one dimensional integrals over energies rather than three-dimensional integrals
over wavevector.
The variational approach incorporates the same physical ideas as the Bogoliubov
approach of Secs. 7.3 A ff and gives exactly the same results. 16
Hamiltonian
k c†kσ ckσ ≡
X X
HKE = k n̂kσ (7.3.47)
kσ kσ
where n̂kσ is the number of fermions in that state. To be most precise, we are trying to
minimize H − µN̂ where N̂ is the total number and µ is chemical potential. However,
our convention is that the energy scale has been shifted so that µ = 0; equiivalently we
are measuring k from the chemical potential (= Fermi energy, at T = 0).
The interaction energy can be written, quite generally, as
(p)
(c†k0 +p/2,σ1 c†−k0 +p/2,σ2 c−k+p/2,σ2 ck+p/2,σ1
X X
HV (0) = Vkk0 (7.3.48)
pkk0 σ1 ,σ2
I’ve assumed just two things: translational symmetry (so the total wavevector of each
term is zero, taking creation minus annihilation operators); and no spin dependence in
p
the interaction. I also assumed the Vkk 0 was real...
states within ~ωc interact equally. But electron states that are farther away from E F than
hbarωc are not very important and we can get away with omitting their interaction.
16 “Random-phase approximation in the Theory of Superconductivity” by P. W. Anderson, Phys.
Rev. 112, 1900 (1958), discusses the equivalence of the BCS variational approach and the Bogoliubov-
Valatin approach.
7.3 Z. ENERGY EXPECTATION 769
Let’s first consider the plain Fermi sea |ΨF i, which is actually just the special case of
|ΦBCS i with uk = 0, vk = 1 for k < EF , but uk = 1, vk = 0 for for k > EF . For a term
in HV to have a nonzero expectation in |ΨF i (or any other state with definite occupation
numbers), each creation operator must be paired with the annihilation operator of the
same state. As usual, there are two ways to do this.
First, the “Hartree” term has the final state of the two interacting electrons identical
to the initial state, |k1 σ1 , k2 σ2 i → |k1 σ1 , k2 σ2 i. This corresponds to k0 = k and the
result is simply X (p)
hHV (0) iHartree = Vkk nk+p/2 n−k+p/2 (7.3.50)
pk
(I NEED TO VERIFY THE - SIGN) These terms are large – of order eV per electron. 17
Those expectations are coming from the entire Fermi sea and not just the portion around
the Fermi energy.
Now, expectations (7.3.50) and (7.3.51) have just exactly the form assumed in Fermi
liquid theory, which they inspired. The spirit of Fermi liquid theory suggests we ought
to cancel these terms by making the unsual transformation to Fermi-liquid quasiparticles
(Lec. 1.7 ). After all, superconductivity will only involve the states close in energy to
EF , exactly those for which Fermi liquid theory is valid.
So, although we will continue to call {ck } “electron” operators, they are really
quasiparticle operators. The main consequences, for our purposes here, are (i) the value
of vF will be renormalized because of the interactions (ii) we can assume the Hartree
(p)
and Fock terms have already been accounted for in k (iii) the coefficients Vkk0 will be
changed in the transformation.
We’ve argued away the large energies that will (however) not differ much between
the normal and superconducting phases. To explain the latter’s stability, we must turn
to the small energies which depend on the superconducting correlations.
Pairing terms
Taking hHV (0) i in |ΦBCS i, one obtains the same terms as before, and a third type
of expectation, those of the pairing terms. 18 That means any term with p = 0.19
X (00) †
Hpair = Vkk0 bk0 bk (7.3.52)
kk0
Possibly should exclude k = k0 in this sum? In this case, we get a nonzero expectation by
(say) selecting the uk term in hΦBCS | and the vak term in |ΦBCS i in hΦBCS |HV (0) |ΦBCS i.
So let’s compute the expectation of a term in Hpair . We get
hb†k0 bk i|ΦBCS i = h0|(u∗k0 + vk∗ 0 b0k )(u∗k + vk∗ bk )b†k0 bk (uk + vk b†k )(uk0 + vk0 b†k0 )|0i (7.3.53)
In (7.3.53), the terms which give nonzero expectations are underlined. The factors
inside (. . .) all came from the hΦBCS | and |ΦBCS i factors. All the other factors in the
BCS wavefunction (with k00 different from k or k0 ) simply gave factors of |uk00 |2 +|vk00 |2 .
The result is X (00)
hHpair = Vkk0 (u∗k vk )(uk0 vk∗ 0 ) (7.3.54)
kk0
It is a bit tedious, but BCS also (i) could extract the excitation spectrum, which
has a gap equal to the minimum value of ∆(k) for k on the Fermi surface (ii) could
generalize the theory to T > 0 and thus extract Tc .
Sorry, I didn’t manage to convert the rest of these notes. As noted earlier, to a fair
extent it duplicates the Bogoliubov approach, so you can take your pick. However, there
are certain handy identities for our parametrization of the BCS theory (the angles β k )
which are summarized in the next section, and I think not included in the Bogoliubov
notes.
Exercises
General advice for these exercises
The questions here are mostly about the BCS wavefunction and could have been
asked in Lec. 7.1 , but they can be answered more efficiently using the machinery
introduced in this section. In many instances, one needs to evaluate the expectation of
a string of ckσ ’s and c†kσ ’s. One workable method is direct substitution into the BCS
formula. A second method is possible after Lec. 7.5 : rewrite ckσ and c†kσ in terms
†
of the Bogoliubov quasiparticle operators γkσ and γkσ . Then note that |ΦBCS i is the
“vacuum” for quasiparticles, i.e. γkσ |ΦBCS i ≡ 0. The second approach is a bit less
messy algebraically, at the price of demanding a bit more sophistication.
More hints – In general, for the exercises involving Bogoliubov manipulations, the
most convenient method seems to be writing {ckσ } and {c†kσ } in terms of {γkσ } and
†
{γkσ }. In every case, first give the result in terms of uk and vk . You can then re-
express it in terms of k , ∆ and (for convenience) Ek . As in lecture, I personally find it
convenient to use trigonometric functions: defining an angle βk by k = − ∆ cot 2βk .
Thus
uk = sin βk ; vk = cos βk (7.3.55)
You may leave results in terms of βk if you find it tedious to put them in terms of k
and ∆.
(a). Let’s start with a wavefunction containing just two factors from (7.3.23). Show
that
γk↑ γ−k↓ |0i = vk (uk + vk c†k↑ c†−k↓ )|0i (7.3.56)
by substituting from (7.3.12a) and (7.3.12b).
(b). Argue that
This verifies (7.3.24) (it should be obvious why the products of γ operators are the
same).
Hints: The key step is to show the only way to keep an annihilation operator factor
in (7.3.57) from making its term be zero, is to pair it with a creation operator (on its
right) from the factor γk↓ within the same parenthesis in the left-hand side of (7.3.57).
This problem continues the warm-up exercises of Lec. 7.1 by computing expectations
and correlation functions of the order parameter operator Ψ(r) defined there.
(a). Find hΨ(r)i, as an integral over k with integrand depending on ∆ and (k).
(b). You may need to adopt the “Cooper approximation” (that ∆ is a constant).
Then use the gap equation to simplify the above result and show hΨ(r)i = ∆/V0 .
Show the constant term in (7.3.7) is constV = |hΨi|2 , where Ψ is given by (7.3.34)
(see Ex. 7.3.4). )
WORRY: are factors of N right, here?
is very small (Tinkham, Fig. 2-1); in either case, the key point is that the sharp Fermi
surface has been smeared out.
There is a correlation length ξee such that this correlation function decays as e−r/ξee .
(c). As a fudge, replace the function fBCS (k) by the Fermi function fT (k) with
T = Tc . Then, assuming the result of Ex. 1.3.5x, find how ξee depends on ∆ and vF
(ignoring dimensionless prefactors!). How does this compare to the order parameter
correlation length ξ defined in Lec. 7.1 C?
In each part, sketch the behavior as a function of k /∆; be sure to indicate the
values at k /∆ = −∞, 0, +∞.
(a). What is the group velocity of the quasiparticles? (trivial).
(b). Check the value asserted in Sec. 7.3 X for the mean charge Qqp (k) carried by
a quasiparticle of wavevector k.
Hint: write the electron number N̂ = pσ c†pσ cpσ in terms of quasiparticle opera-
P
tors, and recall |ΦBCS i is the vacuum of the latter.