Chapter 2

ElectronPhonon Interaction

The electronphonon interaction is important not only in creating the phonon scattering of the electrons but also in the formation of Cooper pairs. This interaction is
indeed the cause of the superconductivity. The Frohlich Hamiltonian is derived. A
phonon exchange can generate an attraction between a pair of electrons.

2.1 Phonons and Lattice Dynamics

In this section, we will review a general theory of the heat capacity based on lattice
dynamics.
Let us take a crystal composed of N atoms. The potential energy V depends
on the configuration of N atoms, (r1 , r2 , , r N ). We regard this energy V as a
function of the displacements of the atoms,
u j r j r(0)
j ,

(2.1)

measured from the equilibrium positions r(0)

j . Let us expand the potential V =
V (u1 , u2 , , u N ) V (u 1x , u 1y , u 1z , u 2x , ) in terms of small displacements
{u j }:
V = V0 +

 
j

=x,y,z


u j

V
u j


+
0



2 V
1 
u j u k
+ ,
2 j k
u j u k 0
(2.2)

where all partial derivatives are evaluated at u1 = u2 = = 0, which is indicated

by subscripts 0. We may set the constant V0 equal to zero with no loss of rigor. By
assumption, the lattice is stable at the equilibrium configuration. Then the potential
energy V must have a minimum, requiring that all first-order derivatives vanish:


V
u j


= 0.

(2.3)

S. Fujita et al., Quantum Theory of Conducting Matter,

C Springer Science+Business Media, LLC 2009
DOI 10.1007/978-0-387-88211-6 2, 

15

16

ElectronPhonon Interaction

For small oscillations we may keep terms of the second order in u j only. We
then have
 1
A jk u j u k ,
(2.4)
V  V
2

j
k

A jk

2 V
u j u k


.

(2.5)

The prime ( ) on V indicating the harmonic approximation, will be dropped hereafter. The kinetic energy of the system is
T

m
j

r 2j =

m
j

u 2j

 m
j

u 2j .

(2.6)

A jk u j u k .

(2.7)

We can now write down the Lagrangian L T V as

L=

 m
j

u 2j

 1
j

This Lagrangian L in the harmonic approximation is quadratic in u j and u j .

According to theory of the principal-axis transformation [1], we can transform the
Hamiltonian (total energy) H = T + V for the system into the sum of the energies
of the normal modes of oscillations:
H=

3N

1
=1

(P2 + 2 Q 2 ),

(2.8)

where {Q , P } are the normal coordinates and momenta, and { } are normalmode frequencies. Note that there are exactly 3N normal modes.
Let us first calculate the heat capacity by means of classical statistical mechanics. This Hamiltonian H is quadratic in canonical variables (Q , P ). Hence the
equipartition theorem holds. We multiply the average thermal energy for each mode,
k B T , by the number of modes, 3N , and obtain 3N k B T for the average energy H .
Differentiating this 3N k B T with respect to T , we obtain 3N k B for the heat capacity,
which is in agreement with DulongPetits law: C = 3R. It is interesting to observe
that we obtained this result without knowing the actual distribution of normal-mode
frequencies. The fact that there are 3N normal modes played an important role.
Let us now use quantum theory and calculate the heat capacity based on Equation
(2.8). The energy eigenvalues of the Hamiltonian H are given by
E[{n k }] =

 1


+ n  ,

n = 0, 1, 2, .

(2.9)

2.1 Phonons and Lattice Dynamics

17

We can interpret Equation (2.9) in terms of phonons as follows: the energy of the
lattice vibrations is characterized by the set of the numbers of phonons {n } in the
normal modes {}. Taking the canonical-ensemble average of Equation (2.9), we
obtain
E[{n}] =

 1



 1
+ n   =
+ f 0 ( )  E(T ),
2
2

(2.10)

where
f 0 ()

1
exp(/k B T ) 1

(2.11)

is the Planck distribution function.

The normal-modes frequencies { } depend on the normalization volume V , and
they are densely populated for large V . In the bulk limit we may convert the sum
over the normal modes into a frequency integral, and obtain

E(T ) = E 0 +
E0

1
2

d  f 0 ()D(),

(2.12)

d  D(),

(2.13)

0
0

where D() is the density of states (modes) in angular frequency defined such that
Number of modes in the interval (, + d) D()d.

(2.14)

The constant E 0 represents a temperature-independent zero-point energy.

Differentiating E(T ) with respect to T , we obtain for the heat capacity at constant
volume:


E
CV =
T

=
V

d 

f 0 ()
D().
T

(2.15)

This expression was obtained under the harmonic approximation only, which is
expected to be valid at very low temperatures.
To proceed further, we have to know the density of normal modes, D(). To
find the set of characteristic frequencies { } requires solving an algebraic equation
of 3N -th order, and we need the frequency distribution for large N . This is not
a simple matter. In fact, a branch of mathematical physics whose principal aim is
to find the frequency distribution is called lattice dynamics. Figure 2.1 represents
a result obtained by Walker [2] after the analysis of the X-ray scattering data for
aluminum, based on lattice dynamics. Some remarkable features of the curve are

Fig. 2.1 The density of

normal modes in the angular
frequency for aluminum. The
solid curve represents the
data deduced from X-ray
scattering measurements due
to Walker [2]. The broken
lines indicate the Debye
distribution with D = 328 K

arbitrary units

18

ElectronPhonon Interaction

0.6
0.4
0.2
angular frequency

(2 x 1013 sec1)

0.8

1.0

max

(A) At low frequencies

D() 2 .

(2.16)

(B) There exists a maximum frequency max such that

D() = 0

for

max .

(2.17)

(C) A few sharp peaks, called van Hove singularities [3], exist below max .
The feature (A) is common to all crystals. The low frequency modes can be
described adequately in terms of longitudinal and transverse elastic waves. This
region can be represented very well by the Debyes continuum model [4], indicated
by the broken line and discussed in Book 1, Section 2.2. The feature (B) is connected
with the lattice structure. Briefly, no normal modes of extreme short wavelengths
(extreme high frequencies) exist. Hence there is a limit frequency max . The sharp
peaks, feature (C), were first predicted by van Hove [3] on topological grounds, and
these peaks are often refered to as van Hove singularities.
The cause of superconductivity lies in the electronphonon interaction [5]. The
microscopic theory can be formulated in terms of the generalized BCS Hamiltonian
[5], where all phonon variables are eliminated. In this sense the details of lattice dynamics are secondary to our main concern (superconductivity). The following point,
however, is noteworthy. All lattice dynamical calculations start with a real crystal
lattice. For example, to treat aluminum, we start with an fcc lattice having empirically
known lattice constants. The equations of motion for a set of ions are solved under
the assumption of a periodic lattice-box boundary condition. Thus, the k-vectors used
in lattice dynamics and Bloch electron dynamics are the same. The domain of the
k-vectors can be restricted to the same first Brillouin zone. Colloquially speaking,
phonons (bosons) and electrons (fermions) live together in the same Brillouin zone,
which is equivalent to say that electrons and phonons share the same house (crystal

2.2 ElectronPhonon Interaction

19

lattice). This affinity between electrons and phonons makes the conservation of momentum in the electronphonon interaction physically meaningful. Thus, the fact that
the electronphonon interaction is the cause of superconductivity is not accidental.

2.2 ElectronPhonon Interaction

A crystal lattice is composed of a regular arrays of ions. If the ions move, then the
electrons must move in a changing potential field. Frohlich proposed an interaction Hamiltonian, which is especially suitable for transport and superconductivity
problems. In the present section we derive the Frohlich Hamiltonian [6, 7].
Let us consider a simple cubic (sc) lattice. The normal modes of oscillations for
a solid are longitudinal and transverse running waves characterized by wave vector
q and frequency q . A longitudinal wave proceeding in the crystal axis x, which is
represented by
u q exp(iq t + iq r) = u q exp(iq t + iq x),

(2.18)

where u q is the displacement in the x-direction. The wavelength 2/q is greater

than twice the lattice constant a0 . The case: = 12 a0 is shown in Fig. 2.2.
If we imagine a set of parallel plates containing a great number of ions fixed
in each plate, then we have a realistic picture of the lattice vibration mode. From
Fig. 2.2 we see that the density of ions changes in the x-direction. Hence, the longitudinal modes are also called the density-wave modes. The transverse wave mode
can also be pictured from Fig. 2.2 by imagining a set of parallel plates containing a
great number of ions fixed in each plate and assuming the transverse displacements
of the plates. Notice that this mode generates no charge-density variation.
The Fermi velocity v F in a typical metal is of the order 106 ms1 while the
speed of sound is of the order 103 ms1 . The electrons are likely to move quickly
to negate any electric field generated by the density variations associated with the
lattice wave. In other words, the electrons may follow the lattice waves instantly.
Given a traveling normal wave mode in Equation (2.18), we may assume an electron
density deviation of the form:
Cq exp(iq t + iq r).

Fig. 2.2 A longitudinal wave

proceeding in the x-direction;
= 12a0

(2.19)

20

ElectronPhonon Interaction

Since electrons follow phonons immediately for all q , the coefficient Cq can be
regarded as independent of q . If we further assume that the deviation is linear in
the scalar product uq q = qu q and again in the electron density n(r), we obtain
Cq = Aq qu q n(r).

(2.20)

This is called the deformation potential approximation. The dynamic response

factor Aq is necessarily complex since the traveling wave is represented by the exponential form in Equation (2.19). There is a time delay between the field (cause)
and the density variation (result), and hence there is an exponential phase factor
change. Complex conjugation of Equation (2.19) yields Cq exp(iq t iq r). Using
this form we can reformulate the electrons response, but the physics must be the
same. From this consideration we obtain (Problem 2.2.1)
Aq = Aq .

(2.21)

The classical displacement u q changes, following the harmonic equation of motion:

u q + q2 u q = 0.

(2.22)

Let us write the corresponding Hamiltonian for each mode as

H=

1 2
( p + 2 q 2 ),
2

q u,

p q,

q ,

(2.23)

where we dropped the mode index q. If we assume the same quantum Hamiltonian
H and the quantum condition:
[q, p] = i,

[q, q] = [ p, p] = 0,

(2.24)

the quantum description of a harmonic oscillator is complete. The equations of motion are
q =

1
[q, H ] = p,
i

p =

1
[ p, H ] = 2 q,
i

(2.25)

(Problem 2.2.2).
We introduce the dimensionless complex dynamical variables:
a (2)1/2 ( p + iq),

a (2)1/2 ( p iq).

(2.26)

Using the last two equations, we obtain

a (2)1/2 (2 q + ip) = ia ,

a = ia.

(2.27)

2.2 ElectronPhonon Interaction

21

We can express (q, p) in terms of (a , a):

q = i(/2)1/2 (a a),

p = (/2)1/2 (a + a).

(2.28)

Thus, we may work entirely in terms of (a , a). After straightforward calculations, we obtain (Problem 2.2.3)
a a = (2)1 ( p + iq)( p iq)
1
= (2)1 [ p 2 + 2 q 2 + i(qp pq)] = H ,
2
1

 aa = H + ,
2

(2.29)

aa a a [a, a ] = 1,

(2.30)





1
1
1

H = (a a + aa ) =  a a +
 n +
.
2
2
2

(2.31)

The operators (a , a) satisfy the Bose commutation rules, Equation (2.30). We

can therefore use second quantization, which is summarized in Appendix A, Sections A.1 and A.2, and obtain

r
r
r
r

Eigenvalues of n a a: n  = 0, 1, 2, [see Equation (A.1)]

Vacuum ket |: a | = 0 [see Equation (A.14)]
Eigenkets of n: | , a | , (a )2 | having the eigenvalues 0, 1, 2,
[see Equation (A.16)]
Eigenvalues of H : 12 , 32 , 52 ,

In summary, the quantum Hamiltonian and the quantum states of a harmonic oscillator can be simply described in terms of the bosonic second quantized operators
(a, a ).
We now go back to the case of the lattice normal modes. Each normal mode
corresponds to a harmonic oscillator characterized by (q, q ). The displacements
u q can be expressed as

uq = i


2q

1/2

(aq aq ),

(2.32)

where (aq , aq ) are operators satisfying the Bose commutation rules:

[aq , ap ] aq ap ap aq = pq ,

[aq , ap ] = [aq , ap ] = 0.

(2.33)

22

ElectronPhonon Interaction

We can express the electron density (field) by

n(r) = (r)(r),

(2.34)

where (r) and (r) are annihilation and creation electron field operators, respectively, satisfying the following Fermi anticommutation rules:


(r), (r ) (r) (r ) + (r )(r) = (3) (r r ),


(r), (r ) = (r), (r ) = 0.

(2.35)

The field operators ( ) can be expanded in terms of the momentum-state

electron operators ck (ck ):

(r) =

1 
exp(ik r)ck ,
(V )1/2 k

(r) =

1 

exp(ik r)ck , (2.36)

(V )1/2 k

where operators c, c satisfy the Fermi anticommutation rules:

(3)
{ck , ck } ck ck + ck ck = k,k
,

{ck , ck } = {ck , ck } = 0.

(2.37)

Let us now construct an interaction Hamiltonian HF , which has the dimensions

of an energy and which is Hermitian. We propose

HF =

d 3r

1
q

Aq qu q exp(iq r) (r)(r) + h.c. ,

(2.38)

where h.c. denotes the Hermitian conjugate. Using Equations (2.20), (2.32) and
(2.36), we can re-express Equation (2.38) as (Problem 2.2.3):
HF =

 1
k

(Vq ck+q ck aq + h.c.),

Vq Aq (/2q )1/2 iq.

(2.39)

This is the Frohlich Hamiltonian. Electrons describable in terms of ck are now

coupled with phonons describable in terms of aq . The term

Vq ck+q ck aq

(Vq ck ck+q aq )

(2.40)

can be pictured as an interaction process in which a phonon is absorbed (emitted) by

an electron as represented by the Feynman diagram [8, 9] in Fig. 2.3 (a) [(b)]. Note: At
each vertex the momentum is conserved. The Frohlich Hamiltonian HF is applicable
for the longitudinal phonons only. As noted earlier, the transverse lattice normal modes
generate no charge density variations, making the interaction negligible.

2.3 PhononExchange Attraction

23
k

k q

q
k q

(a)

(b)

Fig. 2.3 Feynman diagrams representing (a) absorption and (b) emission of a phonon by an
electron

Problem 2.2.1. Prove Equation (2.21).

Problem 2.2.2. Verify Equation (2.27).
Problem 2.2.3. Verify Equation (2.31).
Problem 2.2.4. Verify Equation (2.37).

2.3 PhononExchange Attraction

By exchanging a phonon, a pair of electrons can gain attraction under a certain
condition. We treat this effect in this section by using the many-body perturbation
method [10, 11].
Let us consider an electronphonon system characterized by
H =


k

k cks cks

 1 
k


q

1
+ aq aq
2

Vq aq ck+q s cks + h.c.

Hel + H ph + HF H0 + V,

(V HF )

(2.41)

where the three sums represent: the total electron kinetic energy (Hel ), the total
phonon energy (H ph ), and the Frohlich interaction Hamiltonian HF , [see Equation
(2.39)].
For comparison we consider an electron gas system characterized by the
Hamiltonian
Hc =


k

k cks cks +


1
3, 4 | vc | 1, 2 c4 c3 c1 c2 Hel + Vc , (2.42)

2ks
k s
1 1

4 4

24

ElectronPhonon Interaction

3, 4|vc |1, 2 k3 s3 , k4 s4 | vc | k1 s1 , k2 s2 
=

4 e2 k0 1
k +k ,k +k k k ,q s3 s1 s4 s2 .
V q2 1 2 3 4 1 3

(2.43)

The elementary interaction process can be represented by the diagram in Fig. 2.4.
The wavy horizontal line represents the instantaneous Coulomb interaction vc . The
net momentum of a pair of electrons is conserved:
k1 + k2 = k3 + k4 ,

(2.44)

represented by the Kroneckers delta in Equation (2.43). Physically, the Coulomb

force between a pair of electrons is an internal force, and hence it cannot change the
net momentum.
We wish to find an effective Hamiltonian ve between a pair of electrons generated
by a phonon exchange. If we look for this ve in the second order in the coupling
constant , then the likely candidates are represented by two Feynman diagrams
in Fig. 2.5. Here, the time is measured upward. (Historically, Feynman represented
the elementary interaction processes by diagrams. Diagram representation is widely

k3

Fig. 2.4 The Coulomb

interaction represented by the
horizontal wavy line
generates the change in the
mometa of two electrons

k4

k1

k2

time

k q

k +q

k +q

k q

(a )

( b)

Fig. 2.5 A one-phononexchange process generates the change in the momenta of two electrons
similar to that caused by the Coulomb interaction

2.3 PhononExchange Attraction

25

used in quantum field theory [8, 9].) In the diagrams in Fig. 2.5, we follow the
motion of two electrons. We may therefore consider a system of two electrons and
obtain the effective Hamiltonian ve through a study of the evolution of two-body
density operator 2 . Hereafter, we shall drop the subscript 2 on indicating twobody system.
The system-density operator (t) changes in time, following the quantum Liouville equation:
i

(t)
= [H, ] H.
t

(2.45)

We assume the Hamiltonian H in Equation (2.41) and study the time evolution
of (t), using quantum many-body perturbation theory. Here, we sketch only the
important steps; more detailed calculations were given in FujitaGodoys book,
Quantum Statistical Theory of Superconductivity [10, 11].
Let us introduce a quantum Liouville operator
H H0 + V,

(2.46)

which generates a commutator upon acting on , see Equation (2.45). We assume

that the initial-density operator 0 for the electronphonon system can be factorized
as
0 = electron phonon ,

(2.47)

which is reasonable at 0 K, where there are no real phonons and only virtual phonons
are involved in the dynamical processes. We can then choose
phonon = |0 0| ,

(2.48)

where |0 is the vacuum-state ket for phonons:

aq |0 = 0

for any q.

(2.49)

The phonon vacuum average will be denoted by an upper bar or by angular

brackets:
(t) 0| (t) |0 (t)av .

(2.50)

Using a time-dependent perturbation theory and taking a phonon-average, we

obtain from Equation (2.45)
(t)
2
= 2
t


t
0



d V exp(i 1 H0 )V(t ) av .

(2.51)

26

ElectronPhonon Interaction

In the weak-coupling approximation, we may calculate the phononexchange

effect to the lowest (second) order in and obtain




2 V exp(i 1 H0 )V(t ) = 2 V exp(i 1 H0 )V av (t ).

(2.52)

Using the Markoffian approximation we may replace (t ) by (t) and take

the upper limit t of the -integration to . Using these two approximations, we
obtain from Equation (2.51)


(t)
= i2 1 lim V(H0 ia)1 V av (t),
a0
t

a > 0.

(2.53)

Let us now take momentum-state matrix elements of Equation (2.53). The lhs is

k1 s1 , k2 s2 | (t) | k3 s3 , k4 s4 
(1, 2 ; 3, 4, t),
t
t

(2.54)

where we dropped the upper bar indicating the phonon vacuum average. The rhs
requires more sophisticated computations due to the Liouville operators (V, H0 ).
After lengthy but straightforward calculations, we obtain from Equation (2.53)


(1, 2 ; 3, 4, t) =
i1 [1, 2 | ve | 5, 6 2 (5, 6 ; 3, 4, t)
t
k s k s
5 5

6 6

5, 6 | ve | 3, 4 2 (1, 2 ; 5, 6, t)] , (2.55)

3, 4 | ve | 1, 2 |Vq |2

q
k +k ,k +k k k ,q s3 s1 s4 s2 .
(3 1 )2 2 q2 1 2 3 4 3 1

(2.56)

Kroneckers delta k1 +k2 ,k3 +k4 in Equation (2.56) means that the net momentum
is conserved, since the phonon exchange is an internal interaction.
For comparison, consider the electron-gas system. The two-electron density matrix c for this system changes, following


c (1, 2 ; 3, 4, t) =
i1 [1, 2 | vc | 5, 6 c (5, 6 ; 3, 4, t)
t
k s k s
5 5

6 6

5, 6 | vc | 3, 4 c (1, 2 ; 5, 6, t)] ,

(2.57)

which is of the same form as Equation (2.55). The only differences are in the interaction matrix elements. Comparison between Equations (2.43) and (2.56) yields

References

27

4 e2 k0 1
V q2
q
|Vq |2
(k1 +q k1 )2 2 q2

(Coulomb interaction),
(phononexchange interaction).

(2.58)
(2.59)

In our derivation, the weak-coupling and the Markoffian approximations were

used. The Markoffian approximation is justified in the steady state condition in
which the effect of the duration of interaction can be neglected. The electron mass is
four orders of magnitude smaller than the lattice-ion mass, and hence the coupling
between the electron and ionic motion must be small by the mass mismatch. Thus,
expression (2.59) is highly accurate for the effective phononexchange interaction
at 0 K. This expression has remarkable features. First, the interaction depends on
the phonon energy q . Second, the interaction depends on the electron energy
difference k1 +q k1 before and after the transition. Third, if
|k1 +q k1 | < q ,

(2.60)

then the effective interaction is attractive. Fourth, the atraction is greatest when
k1 +q k1 = 0, that is, when the phonon momentum q is parallel to the constantenergy (Fermi) surface. A bound electron-pair, called a Cooper pair, may be formed
by the phononexchange attraction, which was shown in 1956 by Cooper [12].

References
1. T. W. B. Kibble, Classical Mechanics, (McGraw-Hill, Maidenhead, England, 1966), pp.
166171.
2. C. B. Walker, Phys. Rev. 103, 547 (1956).
3. L. Van Hove, Phys. Rev. 89, 1189 (1953).
4. P. Debye, Ann. Physik 39, 789 (1912).
5. J. Bardeen, L. N. Cooper and J. R. Schrieffer, Phys. Rev. 108, 1175 (1957).
6. H. Frohlich, Phys. Rev. 79, 845 (1950).
7. H. Frohlich, Proc. R. Soc. London A 215, 291 (1950).
8. R. P. Feynman, Statistical Mechanics (Addison-Wesley, Reading, MA, 1972).
9. R. P. Feynman, Quantum Electrodynamics (Addison-Wesley, Reading, MA, 1961).
10. S. Fujita and S. Godoy, Quantum Statistical Theory of Superconductivity, (Plenum, New York,
1996), pp. 150153.
11. S. Fujita and S. Godoy, Theory of High Temperature Superconductivity, (Kluwer, Dordrecht,
2001), pp. 5458.
12. L. N. Cooper, Phys. Rev. 104, 1189 (1956).

http://www.springer.com/978-0-387-88205-5