193 9.

3 The transverse dielectric function

For an EM field E(ω), the current density is given by


ji = σij Ej , (9.48)
j

where the conductivity tensor can be shown from perturbation theory:

e2 h̄2  f0 (Enk ) − f0 (En k )

σij = −i n k |∇i |nknk|∇j |n k , (9.49)
m2 ω En k − Enk − h̄ω − iη
nk
n  k

where η = 0+ . Using the relation (with η = 0+ )

 
1 1
=P − iπ δ(x), (9.50)
x + iη x
we obtain
πe2 h̄2 
  
 k ) n k ∇i |nk nk| ∇j n k δ(En k − Enk − h̄ω).
 
Reσij = f 0 (Enk ) − f 0 (En
m ω
2
nk
n  k
(9.51)

This is the Kubo–Greenwood expression3 for the real part of the conductivity, which
can be derived using different methods. Note that this formula is very general for quantum
systems. For example, nk need not be band states. Using Eq. (9.14), the diagonal part of
→
Eq. (9.51) yields (ê = unit vector  E )
4π ↔
2 (ω) = Re ê · σ · ê
ω
4π 2 e2 h̄2 

 
= f 0 (Enk ) − f 0 (En  k ) |n k |∇ · ê|nk| δ(En k − Enk − h̄ω).
2
m2 ω2
nk
n  k
(9.52)

Note that 2 (ω) now depends on ê. For a cubic system, we may write
 
| n k ê · ∇ |nk|2 = 13 | n k ∇ |nk|2 , (9.53)

which is independent of ê.

Fermi’s golden-rule derivation of 2 (ω). We may arrive at the expression in Eq. (9.52)
using perturbation theory. In a perturbing EM field,
 
1 1 |e|A(ri ) 2  
H= pi + + eφ(ri ) + V(ri ) = H0 + H1 , (9.54)
2 m c
i i i

3 R. Kubo, “Statistical-mechanical theory of irreversible processes I: General theory and simple applications
to magnetic and conduction problems,” J. Phys. Soc. Japan 12(1957); D. A. Greenwood, “The Boltzmann
equation in the theory of electrical conduction in metals,” Proc. Phys. Soc. 71(1958), 585.
194 9 Electronic transitions and optical properties of solids

with H0 the unperturbed Hamiltonian, and

1 |e|  1  e2 2 
H1 = (pi · A(ri ) + A(ri ) · pi ) + 2 A (ri ) + eφ(ri ). (9.55)
2c m 2c m
i i i

If we choose the Coulomb gauge

∇·A=0 (9.56)

and

φ = 0, (9.57)

then we can write for the ith electron (in the limit of small A)

|e|h̄ |e|h̄ e2 2
H̃i1 = −i A(ri ) · ∇ i − i ∇ i · A(ri ) + A (ri )
2mc 2mc 2mc2
|e|h̄ |e|
≈ −i A(ri ) · ∇ i = A(ri ) · pi . (9.58)
mc mc
For an EM wave of the form
A0 i(q·r−ωt)
A= [e − e−i(q·r−ωt) ], (9.59)
2
we have
1 ∂A
E=− (9.60)
c ∂t
or
c
A= E.
iω

Therefore, H̃i1 = Hi1 + Hi∗

1 , with

−|e|h̄ i(q·r−ωt) E 0
Hi1 = e · ∇. (9.61)
mω 2
We can now use Fermi’s golden rule to derive 2 (ω) for insulators. For absorption for
insulators at T = 0, the absorption coefficient η is defined through

energy absorbed/time–volume − dI
η≡ = 1 cdt2 , (9.62)
incident energy/area–time I n

where the incident energy is

I ∼ e−2x/δ , (9.63)
 195 9.3 The transverse dielectric function

(a) (b)
Conduction
band

t
nk Photon

ħω
nk Valence
band

Figure 9.4 Direct optical absorption process in an insulator. (a) Feynman diagram. (b) Band diagram.

where δ is the distance over which E is reduced by 1e . Using

 
dI −dI dx 2I c
− = = , (9.64)
dt dx dt δ n
we obtain, using Eqs. (9.63), (9.16), and (9.18),

− dI 2 2kω ω
η = 1 cdt2 = = = 2 . (9.65)
I n δ c nc

If we define Wtot (ω) ≡ number of photons absorbed/volume–time, then the energy

absorbed/time–volume = Wtot (ω)h̄ω. Hence, for an EM wave of the form in Eq. (9.59),
which has energy density u = 8π
1
(nE0 )2 ,

energy incident 1 c
= uv = (nE0 )2
area–time 8π n
ncE20
= , (9.66)
8π
and
8π h̄ω
η= Wtot (ω) (9.67)
ncE20

This gives
8π h̄
2 (ω) = Wtot (ω). (9.68)
E20

We may now obtain 2 (ω) from Eq. (9.68) by considering the transition rate from
(nk) → (n k ),
2π  
w(nk → n k ) = |n k |H1 |nk|2 δ(Enk − Enk − h̄ω)(fnk − fn k )
h̄
2
π h̄e2 E20
= ê · Mnn (k, k ) δ(En k − Enk − h̄ω)(fnk − fn k ),

(9.69)
2m ω2 2
196 9 Electronic transitions and optical properties of solids

with M an interband transition matrix element from H1 . (See Section 9.4.) Wtot is the sum
of all possible (nk → n k ), and we obtain Eq. (9.52) from Eq. (9.68).

9.4 Interband optical transitions in semiconductors and insulators

For insulating crystals, the interband matrix element M in Eq. (9.69) may be evaluated
using Eq. (9.61),

1 eh̄
H1 = E0 ei(q·r−ωt) ê · ∇, (9.70)
2 mω

and we have

 
Mnn (k, k ) ≡ n k eiq·r ∇ |nk = u∗n k e−ik ·r eiq·r ∇(unk eik·r )d3 r


= u∗n k (r)e−i(k −q−k)·r (∇ + ik)unk (r)d3 r. (9.71)

Since u is a periodic function, Eq. (9.71) gives the result that M is zero unless

k − q − k = G. (9.72)

This is the origin of the selection rule of conservation of crystal momentum. For q ∼ 0,
k ∼ k, we have

4π 2 e2 h̄2  
2 (ω) = |ê · Mvc (k)|2 δ(Ec − Ev − h̄ω). (9.73)
m2 ω2 vc
k

Here v denotes the occupied valence bands and c the empty conduction bands. The real
part of the dielectric function 1 (ω) can be obtained using the Kramers–Kronig relation.
We now introduce the joint density of states J(E) in analogy with the usual density of
states g(E) that is given for a paramagnetic system by ( is the sample volume)

1  2
g(E) = δ(E − Enk ) = d3 kδ(E − Enk ). (9.74)
 (2π )3
nk

The integral for 2 (ω) is over equal-energy-difference surfaces in k-space defined by Eck −
Evk − h̄ω = 0. In many cases, ê·Mvc can be assumed to be a slowly varying function of k
(except for the case M = 0 at some k-point, due to symmetry), and we can define the joint
density of states as
 
2 1
Jcv (ω) = δ[Ec (k) − Ev (k) − h̄ω] = ds , (9.75)
(2π )3 s ∇k (Eck − Evk )
v ck
 197 9.4 Interband optical transitions in semiconductors and insulators

Energy E (eV)
2
Approx. parallel
bands
0 Approx. parallel
bands
–2

–4
L Γ X Γ
Wavevector k

Figure 9.5 An example of parallel bands leading to critical points and van Hove singularities in the joint density of states.

where s = surface of constant h̄ω = Eck − Evk , and within this simplification

1 
2 (ω) ∼ |ê · Mvc |2 Jcv (ω), (9.76)
ω2 v c

where Mvc is a constant for a given (v, c) pair of bands. The joint density of states gives
a measure of the density of full and empty states of equal energy difference. This quantity
plays an important role in determining the interband contribution to 2 and the optical
properties of solids for certain classes of materials for which Mvc ≈ constant.
Similarly to the electron or phonon analysis of g(E), we define critical points in k-space
by the condition

∇k (Eck − Evk ) = 0, (9.77)

which associates a critical point with regions in k-space where the electronic energy bands
are parallel, as shown in Fig. 9.5.
Hence, near a critical point energy h̄ω(k0 ), we can expand the energy difference in
principal axes to obtain in three dimensions


3
h̄ω(k) = h̄ω0 (k0 ) + α (k − k0 )2 . (9.78)
=1

Depending on the signs of the α , we obtain van Hove singularities of different types (as
in the case of g(E)), M0 , M1 , M2 , and M3 in Jcv (E) for a three-dimensional crystal.
In many applications involving the optical properties of semiconductors, the behavior
of the absorption edge plays a very important role. The optical absorption edge of direct
gap materials, where the valence band maximum and conduction band minimum lie at
the same k, is quite different from those of the indirect gap materials. Direct transitions
constitute the lowest-energy transitions of the absorption spectrum of direct gap materials
and determine the shape of the absorption edge, as in the case of GaAs.
 198 9 Electronic transitions and optical properties of solids

Conduction
band

Energy E
Eg

Valence
band

Wavevector k

Figure 9.6 A direct bandgap material with parabolic bands.

For this case, near the band edges for parabolic bands (see Fig. 9.6),

h̄2 k2 h̄2 k2 h̄2 k2

Eck − Evk = Eg + − = Eg + , (9.79)
2mc 2mv 2μ
where μ is the reduced mass of the electron (me = mc ) and hole (mh = −mv ). This yields
 3/2
1 2μ
Jcv (ω) = (h̄ω − Eg )1/2 (9.80)
(2π ) h̄2
3

and, near the edge,

1 1
2 (ω) ∼ (h̄ω − Eg ) 2 (9.81)
ω 2

if one assumes that M is a constant.

Dipole-allowed vs. dipole-forbidden transitions. We now consider dipole-allowed vs.
dipole-forbidden transitions. In general, 2 (ω) (Eq. (9.73)) will depend on the matrix
element |ˆ · M|2 , with (where the u functions are normalized in a unit cell)

Mvc (k) = c, k + q| eiq·r ∇ |v, k


= u∗c,k+q (∇ + ik)uv,k (r)d3 r
  (9.82)
= u∗c,k+q ∇uv,k d3 r + ik u∗c,k+q uvk d3 r
f
≡ Mdvc (k) + ikMvc (k).

Usually, the second term of the sum can be neglected as compared with the first term
because of the orthogonality of Bloch functions with the same k, since q ≈ 0. Mdvc gives
 199 9.4 Interband optical transitions in semiconductors and insulators

h̄ω nk

mq
nk

Figure 9.7 Phonon-assisted transition caused by a photon of frequency ω through the absorption or emission of a phonon
having wavevector q.

the standard dipole selection rule: Mdvc = 0 between states of the same parity. For dipole-
f
forbidden transitions, Mvc in the second term will determine the transition intensity. Hence,
for dipole-allowed transitions (e.g. GaAs, in which the top of the valence band is p-like and
the bottom of the conduction band is s-like),

1 
2 ∼ |M |
d 2
h̄ω − Eg , (9.83)
ω2 v c
and, for dipole-forbidden transitions (e.g. NiO and the cuprates), owing to the extra k factor
in the second term in Eq. (9.82),

1 f
2 ∼ |Mvc |2 (h̄ω − Eg )3/2 . (9.84)
ω2

We note also that if Mdcv (k), although zero at k0 , has a linear k-dependent term, it will
contribute to give an effective total Mf in Eq. (9.84).
Indirect optical transitions. Up to this point, we have only considered direct transi-
tions. The wavevector or crystal momentum selection rule from the interaction H1 ∼ A · ∇
is

kf = ki + qphoton + G. (9.85)

However, if we have an indirect bandgap material, the absorption edge corresponding to

the indirect gap energy range arises from transitions involving impurities (which remove
the above selection rule) in the system or through the emission or absorption of phonons
(or other elementary excitations). (See Fig. 9.7.) Therefore, we cannot consider only the
term A · ∇.
For phonon-assisted transitions, the k-selection rule is

kf = k + qphoton ± qphonon + G. (9.86)

The perturbation Hamiltonian becomes

H1 = Helectron–photon + Helectron–phonon = Hrad + Hlat . (9.87)

Since H1 involves terms separately with a photon and a phonon, we need to evaluate it
to second order to obtain an indirect optical transition. The transition amplitude involves
200 9 Electronic transitions and optical properties of solids

summing up all virtual processes that lead to a final state of an electron in the conduction
band and an electron missing (hole created) in the valence band. For the system in its
ground state with Np photons and nq phonons, the initial state is

|i = electrons in their ground state, Np photons, and nq phonons , (9.88)

and the final state is


|f  = excited electron–hole pair, Np − 1 photons, and nq ± 1 phonons . (9.89)

Second-order perturbation theory gives the transition matrix element square for use in
Fermi’s golden rule as
2
 f | Hlat |m m| Hrad |i
+ (H ↔ H ) term , (9.90)
Em − Ei
rad lat
m

where |m is an intermediate state, and, as discussed in Chapter 10,

1
Hlat ∼ a†−q + aq )c†k+q ck , (9.91)

with
  √
nq − 1 Hlat nq ∼ nq (absorption) (9.92)

and
  
nq + 1 Hlat nq ∼ nq + 1 (emission), (9.93)

where the thermal averaged value nq  is

1
nq  = , (9.94)
eβ h̄ω(q) −1

and β = kB1T . Owing to the difference in the energy denominator, typically only one of the
two terms in Eq. (9.90) dominates. Since for absorption,

1
nq  = , (9.95)
eβ h̄ω(q) −1

and for emission,

1
nq  + 1 = , (9.96)
1 − e−β h̄ω(q)
 201 9.5 Electron–hole interaction and exciton effects

T>0 T=0

Square root of absorption

coefficient η1/2
Emission

Absorption Emission

ω
2ħωq Eg + ħωq

Figure 9.8 Phonon absorption and emission processes in optical absorption spectra. For T > 0, the emission onset is at lower
energy because, in general, Eg decreases with T, and both emission and absorption processes can be appreciable.

we have, for a simple three-band model with a single intermediate band β and all bands
parabolic,
 ck | H |βk  βk | H |vk  2
1 2 lat 1 1 rad 1
ε2 (ω) ∼ 2 d3 k1 d3 k2 δ(Ec (k2 )
ω Eβ (k1 ) − Ev (k1 ) − h̄ω
− Ev (k1 ) − h̄ω ± h̄ωq )
1 1 22 1
∼ h̄ω − Eg − h̄ωq nq  + 1) (for emission of phonon), (9.97)
ω2

or

1 1 22
∼ h̄ω − Eg + h̄ωq (nq ) (for absorption of phonon). (9.98)
ω2

As seen in the above expressions and in Fig. 9.8, there are characteristic frequency and tem-
perature dependences for phonon-assisted indirect transitions that are distinct from direct
transitions.

9.5 Electron–hole interaction and exciton effects

Up to this point, we have not considered the Coulomb interaction between the excited
electron and hole involved in an optical transition. If the electron and hole bind, this is
analogous to the case of positronium, and the resulting excitation is called an exciton. Even
without bound states, there are “exciton effects” owing to the correlations between the two
quasiparticles created. Often excitons are broadly classified as Wannier excitons, when the
 202 9 Electronic transitions and optical properties of solids

(a) (b)

Conduction
band 1s

Absorption coefficient
n=1

Energy E
l=0 2s

Eg Eex Free
3s

Valence
band
ω
Wavevector k Eg

Figure 9.9 Schematic of energy levels of excitons in a two-parabolic band model: (a) Quasiparticle band structure with bandgap
Eg and an exciton state at energy Eex indicated. (b) Sketch of absorption spectrum of a dipole-allowed direct bandgap
semiconductor with the lowest three excitonic transitions shown.

electron and hole are weakly bound, i.e. the electron and hole spatial correlation is large
compared to the lattice constant and Frenkel excitons, when they are strongly bound.
Wannier excitons can often be described approximately by an effective mass equation,
which at the simplest level is given by
 
p2 e2
− F(r) = EF(r), (9.99)
2μ r

where

E = Eex − Eg (9.100)

is the exciton binding energy and F(r) is its wavefunction with r the electron–hole rela-
tive coordinates. Here Eex is the energy of the excitonic state and Eg is the bandgap (see
Fig. 9.9), with

1 1 1
= +
μ mh me

and

r = re − rh . (9.101)

More rigorously, an exciton state is a coherent superposition of many free electron–hole

excitation configurations given in the form

|ψex  = αi |φi , (9.102)
i
 203 9.5 Electron–hole interaction and exciton effects

Energy E
Wavevector k

Figure 9.10 Schematic picture of an exciton state composed of many independent-particle transitions or electron–hole pairs
represented by the arrows.

where

|φi  = c†c,k+q cv,k |G, (9.103)

and c† and c are the one-particle (or, more rigorously, quasiparticle) creation and an-
nihilation operators, and |G represents the electronic ground state. This is depicted in
Fig. 9.10.
To determine the exciton energies and wavefunctions, we solve the many-body
Schrödinger equation, H |ψex  = Eex |ψex , where
 H is the many-electron Hamiltonian.
To solve this problem, we need to find φi | H φj . For example, the ground-state Slater
determinant is described by

G (1, . . . , N) = A{a1 (1)a2 (2) . . . aN (N)}, (9.104)

where A is the antisymmetry operator, ai are single-particle orbitals, and 1 denotes a com-
posite index x1 = (r1 , σ1 ). In the occupation number formalism, this can be written as
(where the orbitals are positioned in order of increasing energy)

|G = |1111 . . . 0000. (9.105)

It is convenient to separate the Hamiltonian into two parts (H = H1 +H2 ), the one-electron
Hamiltonian H1 and the electron–electron interaction term H2 , where


N N  2
 
p i
H1 = h1 (ri ) = + V(ri ) (9.106)
2m
i=1 i=1

and
1 1  e2
H2 = h2 (i, j) = . (9.107)
2 2 rij
i,j i=j
 204 9 Electronic transitions and optical properties of solids

Energy
Figure 9.11 Illustration of the different types of Slater determinants required to create all the matrix elements needed to solve the
exciton problem. The dots indicate occupied orbitals and arrows indicate a transition creating a free electron–hole
pair.

To solve this problem, which we will do in the next section, we need to calculate the
matrix elements between three types of Slater determinants, as shown in Fig. 9.11. These
matrix elements can be worked out if |A is a given single Slater determinant, as

A| H1 |A = ai | h1 |ai 
i

and
 1    

A| H2 |A = ai aj h2 ai aj − ai aj h2 aj ai , (9.108)

2
ij

where

 
ai aj h2 ai aj = a∗i (x1 )a∗j (x2 )h2 (r2 , r1 )ai (x1 )aj (x2 )dx1 dx2 . (9.109)

For |B = c†bk cak |A, as depicted in Fig. 9.11,

A| H1 |B = ak | h1 |bk  (9.110)

and
    

A| H2 |B = ak aj h2 bk aj − ak aj h2 aj bk . (9.111)

j

For |C = c†ck cak c†c ca |A, as depicted in Fig. 9.11,

A| H1 |C = 0 (9.112)

and

A| H2 |C = ak a | h2 |ck c  − ak a | h2 |c ck . (9.113)

 205 9.5 Electron–hole interaction and exciton effects

9.5.1 Weak binding limit in the two-band model

To illustrate the physical properties of excitons, we examine the weak binding limit in a
two-band model. We also note that the basis states or determinants defined above can have
finite center of mass momentum (see below). This is shown schematically in Fig. 9.12.
For an insulating material with weak spin–orbit coupling (i.e. the spin is not coupled to
the orbital motion), we may construct the ground state as
% &
|G = A φv,k1 (r1 )α(1)φv,k1 (r2 )β(2) . . . φv,k N (rN )β(N) , (9.114)
2

where v is the band index, and α = ↑, β = ↓ represent spin-up and -down spinor functions.
Hence, the free electron–hole excited configurations take the form

φck = c†cke se cvkh sh |G, (9.115)
e se ,v kh sh

where, at this point, we are not limiting ourselves to optical excitations, so ke and kh do
not need to be the same and the spin can flip. The wavevector kex = ke − kh characterizes
the center of mass momentum of the free electron–hole (e–h) pair excitation. The exciton
state can now be written as

|ψex  = Ai |φi , (9.116)
i

where |φi  is given by Eq. (9.115).

We now need to solve for the coefficients Ai in Eq. (9.116) by solving the many-electron
Schrödinger equation in order to obtain the exciton states by taking matrix elements of the
Hamiltonian, H = H1 + H2 , given in Eqs. (9.106) and (9.107). This Hamiltonian is spin
independent (since we neglect spin–orbit interaction, which can be important for heavier
atoms). Therefore, we should have a well-defined spin configuration for the excited states.
Combining two spin 12 particles (an electron and a hole) yields an S = 1 triplet state or

ke

kh

Wavevector k

Figure 9.12 Schematic picture showing a nonzero center of mass momentum for a free electron–hole pair configuration.
206 9 Electronic transitions and optical properties of solids

an S = 0 singlet state. We note here that the spin state of a hole is opposite to that of the
electron state being emptied.
For S = 1, we may have schematically (with the second spin being that of the hole)

Sz = 1: |↑ |↑,
 
Sz = 0: √1 |↑ |↓ − |↓ |↑ ,
2

Sz = −1: |↓ |↓,

and for S = 0,
 
Sz = 0: √1 |↑ |↓ + |↓ |↑ . (9.117)
2

Letting M denote a specific spin state, we have within a two-band model (i.e. neglecting
the sum over v and c)
M  
ψ = M
ex Ake ,kh φke ,kh , (9.118)
ke ,kh

where φ M is a composite Slater determinant of the form given by combining Eqs. (9.115)
and (9.117). The Hamiltonian also has crystalline translational symmetry, and, therefore,
the exciton has a well-defined center of mass wavevector kex · So kex is a good quantum
number and each of the terms in Eq. (9.118) must have kex = ke − kh . This implies that
ke − kh = ke − kh in Eq. (9.118). In other words, the exciton wavefunction is made up of
transitions with the same change in k. Thus, we can write the exciton wavefunction as
M  
ψ = M
ex A(k) φ 1 1
k+ k ,k− k
. (9.119)
2 ex 2 ex
k

We define the following “envelope function” in real space:


F(r) = A(k)eik·r . (9.120)
k

We may now derive a differential equation for F(r) using the many-body Schrödinger
equation

Hψex = Eex ψex , (9.121)

which becomes the matrix equation


Hk,k A(k ) = Eex A(k). (9.122)
k

At this point, we need to describe the matrix elements of H in Eq. (9.122). The diagonal
elements can be written as (using Koopmans’ theorem)
 
φkh ,ke H φke ,kh = Ec (ke ) − Ev (kh ) + G| H |G, (9.123)
 207 9.5 Electron–hole interaction and exciton effects

A (k)

k
0

Figure 9.13 The coefficient A(k) for the exciton state in Eq. (9.119) is schematically plotted in k-space. In the weak binding
approximation, the distribution is very narrow.

where Ec and Ev are the conduction and valence band state energies, respectively, and,
without any loss in generality, we set the last term (which is the ground-state energy) to
zero. Using the list of matrix elements above for the basis functions (free electron–hole
pair excitations) |A, |B, and |C, the off-diagonal matrix elements become

   −e2   2 

φkMe ,kh H φkM ,k = φcke φvk φck φvk − 2δM φck φvk −e φck φvk ,
e h e h e
e h h r12 h r12
(9.124)

where δM = 1 if S = 0, and δM = 0 if S = 1.
The first term is attractive and is called the direct term because it resembles a direct
interaction between two charges of opposite sign if ke = ke and kh = kh . Similarly, the
second term is called the exchange term because it looks like a Hartree–Fock exchange
interaction. Note that the exchange term is repulsive and is nonzero only for the spin singlet
case.
As we discussed above, a definite center of mass momentum implies that ke − kh =
ke − kh for all transitions that make up the exciton state. If we make the weak binding
approximation (that is, the wavefunction of the bound electron–hole pair in real space is
large), which amounts to a narrow distribution in k-space, then ke − ke ≈ 0. This is shown
in Figs. 9.13 and 9.14.
The direct term can be written as

 −e2   2 

φcke φvk φck φvk = φck Tφvk −e φck Tφvk , (9.125)
e h e h e
h r12 r12 h

where T is the time-reversal operator. This interaction is shown diagrammatically in

Fig. 9.15.
 208 9 Electronic transitions and optical properties of solids

Energy E

Wavevector k

Figure 9.14 All the transitions that make up the exciton state in the weak binding limit are close together in k-space.

vkh
cke
q
t

cke
vkh

Figure 9.15 Diagram of the direct interaction in Eq. (9.124).

If we explicitly write out the direct term, we have

   
−e2 −i(ke −ke )·(r1 −r2 ) ∗
= dr1 dr2 e ucke (r1 )ucke (r1 )u∗vk (r2 )uvkh (r2 ). (9.126)
dir r12 h

The term involving the periodic part of the Bloch function u’s (we shall call it g(r1 , r2 ))
can be seen in the weak binding approximation as g(r1 , r2 ) ≈ ρcke (r1 )ρvkh (r2 ) because all
the k-points are near each other in k-space. The function g is periodic and can be expanded
in planewaves of reciprocal lattice vectors

g(r1 , r2 ) = am,n eiGm ·r1 e−iGn ·r2 . (9.127)
m,n

If the density is approximately constant in a unit cell, then am,n = 1 for m = n = 0, and
am,n ≈ 0 otherwise. So, the direct term can now be written as

   
−e2 −i(ke −ke )·(r1 −r2 )
≈ dr1 dr2 e . (9.128)
dir r12
209 9.5 Electron–hole interaction and exciton effects

Similarly, for the exchange term in Eq. (9.124), we may argue that, to a good approximation
because uck is orthogonal to uvk ,

= Jex (kex )δM , (9.129)
ex

which is independent of ke − ke (which leads to a δ(r) in real space). Collecting all
the terms, we get the final equation for the exciton eigenvalues and eigenfunctions (i.e.
Eq. (9.122)) in the form
     
1 1
Ec k + kex − Ev k − kex − Eex A(k)
2 2
    (9.130)
e2  
+ − dr  e−i(k−k )·r + Jex (kex )δM A(k ) = 0.

r
k

We now introduce some new notation to simplify the evaluation and interpretation of
our results. We begin by defining
   
1 1
Ecv (k, kex ) = Ec k + kex − Ev k − kex . (9.131)
2 2

The envelope wavefunction F was introduced previously in Eq. (9.120). Since our goal is
to derive an equation for Eex and F, we use these definitions and Eq. (9.130) to obtain

  2 
e  −ik·r
Ecv (k, kex ) A(k) + − dr  F(r )e + Jex (kex )δM F(0) = Eex A(k). (9.132)
r

Since mathematically for any function with variable −i∇,


f(−i∇)F(r) = f(k)A(k)eik·r (9.133)
k

and
 e2  e2
dr 
F(r )e−ik·r eik·r = F(r), (9.134)
r r
k

if we multiply Eq. (9.132) by e−ik·r and then sum over k, we obtain

 
e2
Ecv (−i∇, kex ) − + Jex (kex )δM δ(r) F(r) = Eex F(r). (9.135)
r

This equation is the effective mass equation describing excitons. Two assumptions were
used to derive Eq. (9.135): a trial wavefunction represented by a linear combination of
Slater determinants describing free electron–hole pair excitations, and the weak binding
approximation.
It is useful at this point to make a few comments regarding the effective mass equation
(Eq. (9.135)). The excitation energy Eex is the energy needed to create an exciton. At this
 210 9 Electronic transitions and optical properties of solids

e– amplitude

Exciton wavefunction
Hole

Figure 9.16 Schematic drawing of the exciton wavefunction ψex (re ,rh ) in real space, with r the electron–hole relative coordinates
assuming a fixed position for the hole.

point, we can go beyond our rather simple treatment since the electron–hole interaction
2 2
will be screened by the other electrons by changing − er → − er , where  is the dielectric
function of the material. In most cases, the static  can be used since the relevant binding
energy Eb = Eg − Eex is small compared to the bandgap energy. Only the relative coor-
dinates between the electron and the hole remain within Eq. (9.135). We can expand the
interband transition energy term in the form

Ecv (−i∇, kex ) = αβ ∇α ∇β ,
Bnm n m
(9.136)
α,β
n,m

with the coefficients Bnm

αβ related to the band effective masses. Also, coming back to the
definition of A(k) in Eq. (9.119), one can show that within our approximations the full
exciton state is given by

|ψex  = F(R1 − R2 )w1† w2 |G , (9.137)
R1 R2

where now the wi ’s are the Wannier function annihilation operators. The electron–hole

amplitude or wavefunction in real space is ψex (re , rh ) = F(R1 − R2 )wc (re −
R1 ,R2
R1 )wv (rh − Rh ), with wc and wv the Wannier functions. The exciton wavefunction in
real space ψex (re , rh = 0) is illustrated schematically in Fig. 9.16.

9.5.2 Excitonic effects in the isotropic two-band model

As depicted in Fig. 9.17, we may assume for illustration purposes that both bands in an
h̄2 2
isotropic two-band model can be approximated as parabolas, Ecv (−i∇) = Eg − 2μ ∇ ,
where 1
μ = 1
me + 1
mh , and that kex = 0, which is the case for the optical transitions. Putting
 211 9.5 Electron–hole interaction and exciton effects

Conduction
band

Energy E
Eg

Valence
band

Wavevector k

Figure 9.17 The isotropic two-band model.

all of these into the effective mass equation gives

 
−h̄2 2 e2
∇ − + Jcv (0)δM δ(r) F(r) = (Eex − Eg )F(r), (9.138)
2μ r

where (Eex − Eg ) = E is the negative of the exciton binding energy. In the weak binding
approximation,  = constant and J is typically negligible because of the large spatial extent
of the exciton envelope function F. Equation (9.138) resembles the Schrödinger equation
for a hydrogen atom. Therefore, in three-dimensional systems, we expect bound states
(E < 0) of the form

R
E=− , (9.139)
n2

where R = μe2 2 . This energy is typically < 0.01 Ry for most bulk semiconductors. The
4

2h̄ 
wavefunction F becomes a hydrogen-like wavefunction Fn,,m (r). For the unbound states
E > 0, the wavefunctions are again hydrogen-like, Fs,l,m (r) = Rs,l (r)Yl,m , where s =

2μE
It should be noted that for E  R, F → eike ·r and A(k) → δ(k − ke ).
.
h̄2
These results for most bulk semiconductors are consistent with the weak binding
approximation. For example, the 1s exciton wavefunction has the form

1 r
−
F100 =  e aex , (9.140)
π a3ex

with a spread determined by aex = a0  mμe , where a0 is the Bohr radius. Now, in general,
me
μ > 1 and  ≈ 10 − 20 for semiconductors. So, aex ≈ 100a0 . Thus, for A(k), we have
212 9 Electronic transitions and optical properties of solids

(by Fourier-transforming Eq. (9.140))

1
A(k) = A0  2 , (9.141)
1 + (kaex )2


where A0 = 2πa2ex / πa3ex is a constant, and the ratio of A at the Brillouin zone edge
compared to the zone center is

A(kBZ ) 1 −4 −6
= 2 ≈ 10 − 10 , (9.142)
A(0)
1 + (kBZ aex )2

where we have used kBZ ≈ 1 in units of Bohr. Therefore, the weak binding approximation
is consistent. For this model, Jex affects only s-states since it is a delta function at r = 0,
and it is important only for more strongly bound excitons.
Returning to the optical response function, we have from Eq. (9.69) the expression
2 (ω) = 8π2h̄ Wtot for the imaginary part of , where Wtot is the number of photons ab-
E0
sorbed per volume per unit time. In the independent electron picture without considering
excitonic effects, this becomes (rewriting Eq. (9.73) and using the subscript F to denote
free or non-interacting particles)

4π 2 e2 h̄2 2d3 k 2

2F (ω) = ê · M c v (k) δ Ec (k) − Ev (k) − h̄ω , (9.143)

m2 ω2 BZ (2π )3

where ê is the polarization vector of the photon and M is the interband transition matrix
element. With excitonic effects included, the probability of a transition from the ground
state to an exciton state (from Fermi’s golden rule) is

2π 2

P|G→|ψex  = ψG | HI |ψex  δ(Eex − h̄ω), (9.144)
h̄
 |e|
where HI = mc A(ri ) · pi . Using the rules introduced previously for the matrix elements
i 
(in particular, A| i hi |B = ak | h1 |bk ), we find
  
π eA0 2
P|G→|ψex  = δkex δM h̄2 | A(k)ê · Mcv (k)|2 δ(Eex − h̄ω), (9.145)
2h̄ mc
k

where A0 is the amplitude of the vector potential of the EM wave, A(k) is the exciton
wavefunction in k-space, and

Mcv (k) = u∗c (k + q, r)(∇ + ik)uv (k, r)d3 r (9.146)

is the interband matrix element, as before.

 213 9.5 Electron–hole interaction and exciton effects

For dipole-allowed transitions, the first term on the right-hand side of Eq. (9.146) for the
matrix element (i.e. the ∇ term in the integrand) is assumed to be constant independent of

k, and one neglects the ik term for M. Since k A(k) = F(r = 0), Eq. (9.145) becomes

P|G|ψex  ∝ δkex δM |ê · Mcv (0)|2 |F(0)|2 δ(Eex − h̄ω). (9.147)

Since F(0)  = 0 only for s-states for the hydrogenic functions where Fn00 (0) = √ 1
,
πa3ex n3
this gives transition intensity to the excitons with principal quantum number n as In ∝ n3 .
1

Thus, we expect a series of lines at E = Eg − nR2 for n = 1, 2, . . . , with the intensity

decreasing as n13 (Fig. 9.9(b)). At high values of n, we can express the absorption intensity
in terms of the density of states
 
dn n3
D(E) = 2 = , (9.148)
dE R

yielding the result that the intensity/energy = In (E)D(E) goes to a constant for transitions
near Eg . This is shown as the shoulder in Fig. 9.18 near Eg .
For the unbound excitons (E > 0), we have for dipole-allowed bands

π χ eπ χ
2 (ω) = 2F (ω)|Fs00 (0)|2 = 2F (ω) , (9.149)
sinh(π χ )

where χ = There are two interesting limiting cases: (1) h̄ω − Eg  R, 2 (ω) →
R
h̄ω−Eg .
√
h̄ω−Eg
2F (ω), and (2) h̄ω − Eg  R, 2 (ω) → 2π χ 2F (ω) ≈ √ = constant, which is the
h̄ω−Eg
same constant that approaches from the E < 0 side. This behavior is sketched in Fig. 9.18.
Up to this point, we have considered the case of dipole-allowed transitions where optical
transitions lead to optically active (in linear response) excitons of hydrogen atom-like states
with  = 0 (s-states). But in the case of dipole-forbidden transitions, we shall see that the
optically active excitons must have  = 1, and they form hydrogen atom-like p-states. We
Absorption coefficient

ω
Eg

Figure 9.18 Plot of a typical optical spectrum with excitonic effects included.
214 9 Electronic transitions and optical properties of solids

begin with Eq. (9.146) and rewrite M in the form

M = Mallowed + ikMforbidden . (9.150)

For the dipole-forbidden case, Mallowed = 0. We use only the second term in Eq. (9.150)
for Eq. (9.145), and since
 ∂
A(k) k = F(r)|r=0 , (9.151)
∂r
k

the probability of creating an exciton state in the dipole-forbidden case is then

  2
π eA0 2 forbidden 2 ∂
P|G→|ψex  =
h̄ δkex δM M ê · F(r) δ(Eex − h̄ω). (9.152)
2h̄ mc ∂r r=0

Since there is only a nonzero transition probability for

∂
F(r)  = 0, (9.153)
∂r r=0

which is the case for hydrogen atom-like p-states with

 2 1
∂ n −1 2
Fn1m (r) = . (9.154)
∂r r=0 π a5ex n5

This results in having peaks in the absorption spectra at energies

R
E = Eg − (9.155)
n2
for n = 2, 3, 4, 5, . . . , ∞. Here, R is again the effective Rydberg for the exciton. The
intensity of each absorption line is proportional to the transition probability

n2 − 1
In ∼ (9.156)
n5
and
1
In ∼ as n → ∞. (9.157)
n3
The absorption spectra are shown schematically in Fig. 9.19. Note that as the energy ap-
proaches the gap energy Eg , the absorption per unit energy becomes constant. This occurs
because the intensity of each state times the density of states is a constant (as discussed
above for the dipole-allowed case):

lim In D(En ) = constant. (9.158)

n→∞
 215 9.5 Electron–hole interaction and exciton effects

n=2
(2p)

n=3

Absorption coefficient
(3p)

Free
n=4
(4p)

ω
Eg

Figure 9.19 Absorption spectra for dipole-forbidden transitions of exciton states.

For the case of unbound states, 2 is given by (again using hydrogenic wavefunctions)

π χ (1 + χ 2 )eπ χ
2 (ω) = 2F (ω) , (9.159)
sinh π χ

where 2F is the independent electron 2 and χ = (h̄ω−E R
g)
.
There are again two cases of interest. Case I is h̄ω − Eg  R (high energy), 2 (ω) ∼
2F (ω). Case II is h̄ω − Eg  R (near the continuum absorption edge), 2 (ω) ∼ 2F (ω)χ 3
= constant. The form of the absorption for energies approaching Eg from above is also a
constant. An example of a system with dipole-forbidden transitions is Cu2 O. In Cu2 O, the
absorption peaks fit very well into a hydrogen-like series4 vn = 17 250 − 786 n2
cm−1 with
n = 2, 3, . . .
So far, we have only been considering an onset of absorption with an M0 critical point
or singularity in the joint density of states. However, excitonic effects at van Hove singu-
larities of kinds M1 , M2 , and M3 can be significant. For the independent electron model,
a model dipole-allowed absorption spectrum in three dimensions is shown in Fig. 9.20.
The independent electron case is modified in several ways because of excitonic effects.
The effect of excitons near the van Hove singularity M3 is to reduce the oscillator strength
(Fig. 9.21)

π x eπ x
2 (ω) = 2F (ω) , (9.160)
sinh π x

where x = E3 −Rh̄ω , and there are no discrete lines above E3 like there are below E0 . For
the other two cases, near M1 excitonic effects enhance 2 (ω) and near M2 excitonic effects
weaken 2 (ω). In GaAs this leads to an apparent shift in the position of an absorption peak
by about 0.5 eV in the higher-frequency region, even though no bound exciton is formed.

4 P. W. Baumeister, “Optical absorption of cuprous oxide,” Phys. Rev. 121(1961), 359.

 216 9 Electronic transitions and optical properties of solids

Absorbtion coefficient
M1 M2

ω
M0 M3

Figure 9.20 Model of dipole-allowed absorption spectrum at different critical points for a three-dimensional independent
electron system.
Absorption coefficient

With exciton Free electron

effects case

ω
E3

Figure 9.21 Absorption at M3 van Hove singularity in the joint density of states is decreased due to excitonic effects.

Moving beyond the strict hydrogenic model, we note that the exchange interaction term,
which is given by

J = J(kex )δ(r)δM , (9.161)

is repulsive and short-ranged. Since photons cannot flip spin, we expect only singlet states
to be excited optically. However, if spin–orbit interactions are allowed and strong, the
singlet and triplet states will mix and extra weak transitions will be activated. In first-order
perturbation theory, the energy splitting between singlet and triplet states is given for s-like
states by (see Fig. 9.22)

1
E ∼ ψex | J |ψex  ∼ |F(0)|2 ∼ . (9.162)
n3
Hence, the exchange energy E for Wannier excitons is small except for n = 1 and for
strongly bound excitons.
 217 9.5 Electron–hole interaction and exciton effects

Singlet

Absorption coefficient
Triplet

ΔE

Figure 9.22 Single-triplet splitting of excitons.

For many materials, the electron bands are anisotropic. In these cases, the effective
masses and dielectric constants are tensors instead of scalars. For many materials, they
are described in terms of two constants representing an in-plane and an out-of-plane
component,

μ → (μ , μ⊥ ) (9.163)

and

 → ( , ⊥ ). (9.164)

For systems of this type, the anisotropic properties are characterized by the parameter γ ,
 μ
γ = , (9.165)
⊥ μ⊥
with

0 ≤ γ ≤ 1. (9.166)

This problem is difficult to solve analytically in general and is usually done numeri-
cally. For the specific case of γ = 0 (the case where μ⊥ → ∞, which corresponds to
the two-dimensional case, e.g. the two-dimensional electron gas, quantum wells, etc.), the
excitation energy for bound excitons (with dipole-allowed bands) is
R
Eex = Eg −  2 , (9.167)
n+ 1
2

for n = 0, 1, 2, . . . For n = 0, the excitation energy becomes

Eex = Eg − 4R. (9.168)

This is an example of how, in reduced dimensions, Coulomb interaction and correlation

effects become more important.
218 9 Electronic transitions and optical properties of solids

If we consider the effects of a magnetic field in the effective mass approximation, we

may use

e
pe → pe − A(re ), (9.169)
c
e
ph → ph + A(rh ), (9.170)
c

and the effective mass equation becomes

 
1  e 2 1  e 2 e2
pe − A(re ) + ph + A(rh ) − F(re , rh )
2me c 2mh c |re − rh | (9.171)
 
= Eex − Eg F(re , rh ).

For
1 ae nearly
2 uniform magnetic field, the canonical transformation F(re , rh ) =
i K+ c A(r) ·R
e φ(r) may be used, and Eq. (9.171) becomes
 
p2 e 1 1  e2 2 e2 2eh̄
+ − A(r) · p + A (r) − − K · A(r) φ(r)
2μ c mh mc 2μc2 r Mc
  (9.172)
h̄2 K2
= Eex − Eg − φ(r),
2M

 −1
where r = re − rh , R = memree +mh rh
+mh , M = me + mh , μ = 1
mc + 1
mh , and K = momentum
of the center of mass motion. Rearranging terms, Eq. (9.172) becomes

   
h̄2 K2
H0 + Hz + Hd + Hs φ(r) = Eex − Eg − φ(r), (9.173)
2M

p2 e2
where H0 = 2μ − r is the unperturbed Hamiltonian. For the unperturbed Hamiltonian,
we get

h̄2 K2 μe4
E0ex = Eg + − , (9.174)
2M 2n2 h̄2  2

and the other terms correspond to (using the gauge A = − 12 r × B)

   
e 1 1 eB 1 1
(1) Zeeman splitting: Hz = − A(r) · p = − Lz , (9.175)
c mh me 2c mh me
e2 2 e2 B2 2
(2) diamagnetic term: Hd = 2
A (r) = (x + y2 ), (9.176)
2μc 8μc2
219 9.5 Electron–hole interaction and exciton effects

and
2e 2e
(3) magneto-Stark effect: Hs = − K · A(r) = − B · (r × K)
Mc Mc
2e
=− r · (K × B) (9.177)
Mc
In the Hs term, K × B acts as an effective electric field and gives rise to a Stark effect.
The study of excitons and shallow impurity states in a magnetic field in semiconductors
is equivalent to studying atomic physics in extremely high magnetic fields. The relevant
parameter to measure the effect of an external B-field on a system with binding energy R∗
and an effective cyclotron frequency ωc∗ is

h̄ μeB∗ c   2
h̄ωc∗ h̄ωc 
δ= ∗
= ∗ e4
= μ∗ . (9.178)
2R μ 2Ry
2 2h̄ 2 m

For GaAs, δ = 1 at about a magnetic field of 7 Tesla. For an atomic system, for δ = 1, we
need
 μ∗ −2  
m 10 2
B = BGaAs ≈7 T = 105 T. (9.179)
 0.07

Magnetic field studies of exciton and impurity states in semiconductors may thus have
relevance in astrophysics because there are atoms and molecules in such large fields in
outer space.
10 Electron–phonon interactions

The interplay between electrons and lattice vibrations is one of the most interesting and
important physical processes in the study of solids. The static lattice with atoms assumed
frozen at their equilibrium positions yields the crystal potential, which governs many elec-
tronic properties such as band structure effects. With vibrations of the lattice comes a host
of new effects resulting from the interactions of electrons and phonons. These include
the scattering of electrons by phonons, which gives rise to a temperature-dependent re-
sistivity (Fig. 10.1). Other first-order processes include the non-radiative recombination
of electron–hole pairs and their generation, which is responsible for ultrasonic attenuation
(Fig. 10.2).
Second-order electron–phonon processes lead to extraordinary effects. The emission
and subsequent reabsorption of virtual phonons is the fundamental process for the cre-
ation of polarons in insulators and the mass enhancement of electrons in all types of solids
(Fig. 10.3(a)). This process is also responsible for part of the temperature dependence of
energy bandgaps. Another second-order process is the electron–electron interaction via the
exchange of a virtual phonon, which can lead to bipolarons, Cooper pairs, and supercon-
ductivity (Fig. 10.4). The emission of virtual electron–hole pairs (Fig. 10.3(b)) by phonons
and the subsequent destruction of the electron–hole pair renormalize the phonon energy.
This can lead to softening (lowering of energy) of a phonon mode, which in turn can cause
a structural phase transition.

10.1 The rigid-ion model

The rigid-ion model is one of the simplest ways of formulating and understanding electron–
phonon interactions. The idea is that each ion (or atom) is assumed to contribute rigidly
to the total potential seen by the electrons in the solid. As an ion is moved away from its
equilibrium position, the potential is no longer periodic, and this will scatter an electron
from one Bloch state to another. Let V(r) be the static crystal potential which changes
because of vibrations, i.e.
  
V(r) → V(r, t) = Va r − Ra (t) , (10.1)
,a

where
Ra (t) = R + τ a + δRa (t) = Ra + δRa (t), (10.2)
220
 221 10.1 The rigid-ion model

(a) Emission (b) Absorption

k–q k–q
k k

q q

Figure 10.1 Emission and absorption of a phonon by an electron.

k−q k
q q

k k+q

Figure 10.2 Emission of a phonon in non-radiative recombination of an electron–hole pair (left) and generation (right) of an
electron–hole pair.

(a) (b)
q
k

q q

k k–q k

k–q

Figure 10.3 Emission and absorption of a virtual phonon by an electron (a) and emission and reabsorption of a virtual
electron–hole pair by a phonon (b).

where  is the cell index, a is the atomic index, and τ a is the vector to atoms in the basis.
We assume the displacement is much less than the atomic spacing d, i.e.

ξ ≡ δRa  d. (10.3)

For typical vibrations, it can be shown that

ξ  m 1/4
∼ (10.4)
d M
and the excitation or phonon energy Eph scales as

Eph  m 1/2
∼  10−2 , (10.5)
Eel M
m
where Eel is a typical electron energy and M is the ratio of the electron and atomic masses.
Because of these scaling arguments, we can assume that the large-scale band structure is
 222 10 Electron–phonon interactions

k k+q

k k − q

Figure 10.4 Electron–electron interaction by the exchange of a virtual phonon.

not modified. However, the vibrations of the ions do affect the dynamics of electrons. For
example, in the case of a metal, electron–phonon interactions can cause a “wrinkle” in the
band structure within a few meV of the Fermi energy. This represents a modification of
the effective mass of the electrons by the phonons but only at energy scales comparable to
phonon energies.
The derivation of the electron–phonon interaction matrix element proceeds as follows.
Assuming a small displacement, at a specific time t, we can use the following Taylor
expansion approach:
   %    &
V(r, t) = Va r − Ra − δRa (t) ≈ Va r − Ra +δRa · ∇r Va r − Ra . (10.6)
,a ,a

Now the total Hamiltonian is Htot = Hel + Hph + Hel-ph , and we can write the electron–
phonon contribution as
  
Hel-ph = δRa · ∇ r Va r − Ra . (10.7)
,a

By expanding the displacement in phonon coordinates (see Chapter 4), we have

 a
 
δRa = Aaqα ˆ aqα eiq·R aqα + a†−qα , (10.8)
qα

where
(
h̄
Aaqα = , (10.9)
2Ma Nωqα

where q is the wavevector, a labels the atomic species, α is the phonon branch index, and
ˆ is the polarization vector. The phonon destruction and creation operators are represented
by aqα and a†qα . They are not related to the atomic species label using the same letter a.
223 10.1 The rigid-ion model

In the language of second quantization,


Hel = εnk c†nk cnk , (10.10)
nk
 1

Hph = a†qα aqα + h̄ωqα , (10.11)
qα
2

and

Hel-ph = n k |Hel-ph |nkc†n k cnk
n,k,n ,k
 αq
= Mnk→n k c†n k cnk (aqα + a†−qα ). (10.12)
n,k,n ,k ,q,α

Equation (10.12) defines the electron–phonon matrix element M.

Suppressing the band indices n and n by using an extended zone scheme so that k and

k are not limited to the first Brillouin zone, Eq. (10.7) becomes
 
   a
Hel-ph = δRa ·∇r Va (r−Ra ) = Aaqa ˆqα
a
· eiq·R ∇r Va (r − Ra ) (aqα +a†−qα ).
,a a qα 

  (10.13)

We now need k Hel-ph |k to evaluate the quantities in Eq. (10.12). Using the atomic form
factor of the potential,
 
1 −iq·r wN
Va (q) = e Va (r)dr = e−iq·r Va (r)dr, (10.14)
a 

where a is the average volume per atom, w is the number of atoms per cell,  is the
crystal volume, and N is the number of cells in the crystal. To evaluate the matrix element
we use the relations

i   iq ·r
∇r Va (r) = q e Va (q ) (10.15)
wN 
q

and
i   iq ·r  a
∇r Va (r − Ra ) = q e Va (q )e−iq ·R . (10.16)
wN 
q

So, Eq. (10.13) becomes

 i a a  
A ˆ · q eiq ·r ei(q−q )·R Va (q )(aqα + a†−qα ).
a
Hel-ph = (10.17)
wN qα qα
,a,α,q,q

Since  
ei(q−q )·R = Nδ(q − q + G),
a
(10.18)

224 10 Electron–phonon interactions

we obtain
 1 
k |Hel-ph |k = e−iG·τ α Va (q + G) ˆqα
a
· i(q + G)k |ei(q+G)·r |k. (10.19)
w
G

Hence, the electron–phonon part of the Hamiltonian in the second quantization form is
given by 
Hel-ph = Mqa (k − k )c†k ck (aqα + a†−qα ), (10.20)
k,k ,q,α

and the electron–phonon matrix element is given by

(
i  h̄
Mqα (k − k ) = ˆqα · (q + G)eiG·τ α Va (q + G)k |ei(q+G)·r |k. (10.21)
w 2Ma Nωqα
a,G

10.2 Electron–phonon matrix elements for metals,

insulators, and semiconductors


Let us consider an example appropriate for metals, which is to assume that k and |k are
planewaves, as one expects in a pseudopotential approach. Hence,

k |ei(q+G)·r |k = δk ,k+q+G (10.22)

and
(
i  h̄
Mqα (k → k ) = ˆqα · (q + G)eiG·τ α Va (q + G)δk ,k+q+G . (10.23)
w 2Ma Nωqα
a,G

Therefore, if we know the pseudopotential form factor, we can, in principle, get Mqα .
It is useful to distinguish the following two different processes. The first is the so-called
“normal” process or N-process, where k is not far from k. This implies that

k = k + q, (10.24)

with G = 0, and assuming w = 1:

Mqα (k → k + q) = iAqα ˆqα · qV(q). (10.25)

For these processes, there is no coupling if ˆ ⊥ q. Hence, transverse phonons are not
effective in scattering electrons.
The second process with G = 0 is called an Umklapp process or U-process (Fig. 10.5).
The relative values of the matrix elements for these two processes are

MN ˆα · qV(q)
= . (10.26)
MU ˆα · (q + G)V(q + G)
 225 10.2 Electron–phonon matrix elements for metals, insulators, and semiconductors

k = k + q + G
k

Figure 10.5 Scattering of an electron in an Umklapp process. If G = 0, this is a normal process.

ω
LO
ω LO
ω TO
TO

LA

TA

k
Γ X

Figure 10.6 Phonon dispersion relation for a polar diatomic crystal (schematic).

Usually V(q + G) < V(q) for a smooth pseudopotential, but Umklapp processes allow
coupling to transverse phonon modes, and there are many of these processes (i.e. many G
vectors) that contribute to the properties of metals.
Another example is an ionic insulator or polar semiconductor where the energies of the
longitudinal optical (LO) and transverse optical (TO) modes at long wavelengths are non-
degenerate. This splitting of the phonon energies (Fig. 10.6) may be shown to be given by
the ratio of the low- and high-frequency q = 0 dielectric constants

ωLO
2
0
= . (10.27)
ωTO
2 ∞

For charged ions and two atoms per unit cell, small q terms dominate because of the long-
range Coulomb interaction (∼ q12 in three dimensions). Here we use

4π Ze2
V1 (q) = −V2 (q) = (10.28)
∞ q2 a

at small q, where Z is the charge of one of the ions and ∞ is the electronic dielectric
constant of the polar material.
The polarization vectors of the two atoms for the longitudinal mode have the form

(1) (2)
ˆLO = q̂ = −ˆLO , (10.29)
 226 10 Electron–phonon interactions

–Z +Z –Z +Z

M M M M

Figure 10.7 Motion of ions in a diatomic ionic crystal (having charge Z and mass M) for long-wavelength LO phonon modes.

since the displacements are in opposite directions for nearest-neighbor atoms. (See
Fig. 10.7.) Using Eq. (10.21), with w = 2 and M1 = M2 = M,
( ( 
i h̄ h̄
MLO (k → k + q) = q̂ · qV1 (q) − q̂ · qV2 (q) . (10.30)
2 2MNωLO 2MNωLO

Using the relation

ωLO
2
= ωTO
2
+ 2p , (10.31)
where p is the ionic plasma frequency and Eq. (10.27), we obtain
  1/2
2π e2 h̄ωLO 1 1
MLO (k → k + q) = i − . (10.32)
q2  ∞ 0

Here, 0 is the dielectric constant including both the electronic and phonon contribu-
tions. This is the electron–phonon matrix element Fröhlich obtained by using an empirical
approach, and this coupling is often referred to as the Fröhlich interaction or Fröhlich
coupling.1
Our third example is the long-wavelength coupling of electrons to lattice vibrations
in covalent semiconductors. This is called deformation potential coupling. This electron–
phonon coupling is appropriate for excited electrons (or holes) in the conduction (valence)
bands of semiconductors and insulators in the q ≈ 0 limit (Fig. 10.8) to describe how
carriers are scattered to nearby states.
Consider the scattering of carriers near a band extremum in the q → 0 limit. Within
the framework of an effective Hamiltonian or effective mass theory (see Chapter 5), we
express the conduction electron energy as

E(k) = [E(k) − EC ] + EC (r), (10.33)

where the bracketed term on the right-hand side is the band energy relative to the energy of
the conduction band minimum EC , and the last term is the change in the conduction band
minimum energy for a deformed crystal at location r.
A periodic change of the lattice constant arising from a compressional wave will cause
a periodic change in EC (r). The perturbation (from the effective mass or effective Hamil-
tonian point of view) will scatter the Bloch states, and this is associated with the coupling

1 H. Fröhlich, “Electrons in lattice fields,” Adv. Phys. 3(1954), 325.

 227 10.2 Electron–phonon matrix elements for metals, insulators, and semiconductors

Ec

0
k

Figure 10.8 Energy band structure for carriers near band extrema.

of the electrons to the lattice or phonons

Hel-ph = EC (r) − EC = δEC (r). (10.34)

This is then the perturbation potential that scatters an electron from one state to another.
We may write δEC (r) in terms of the dilation (r) of the crystal volume :

δ(r)
(r) ≡ . (10.35)


Hence,
∂EC ∂EC
δEC (r) = δ =  (r) = C1 (r), (10.36)
∂ ∂
where C1 is defined as the deformation potential with units of energy.2 For example, con-
sider a unit volume around an electron at r. Suppose the unit volume is strained (Fig. 10.9),
then the original volume
 = x̂ · (ŷ × ẑ) (10.37)
would change to
 = x̂ · (ŷ × ẑ ) (10.38)
and
 − 
= . (10.39)

We now need to express the dilation of the crystal near the position of an electron r in
terms of the distortion of the crystal lattice sites. That is,

(r) = f({δR (r)}, r), (10.40)

2 J. Bardeen and W. Shockley, “Deformation potentials and mobilities in non-polar crystals,” Phys. Rev.
80(1950), 72.
 228 10 Electron–phonon interactions

(a) (b)

R艎 (r)
z z
r

x x
r艎

Figure 10.9 (a) Unstrained unit volume. (b) Unit volume under strain.

where the nearby lattice sites r (see Fig. 10.9(b)) when measured from the electron are
denoted as
R (r) = r − r. (10.41)
Upon a uniform strain, the position of R (r) will change:

R (r) → R (r) = R (r) + δR (r). (10.42)

Under uniform or slowly varying dilation, the atoms will move away from their unstrained
positions. The change in the lattice sites for the atoms near the electron will scale linearly
with the distance from the electron and

δR (r) → 0 as r (r) → r. (10.43)

We may write
 ∂δR(r)
R (r) = R (r) + Rμ . (10.44)
μ
∂Rμ r
Therefore, we can express the dilation as

 − 
(r) = = −∇r · δR(r). (10.45)

→
The electron–phonon coupling comes from the dependence of ∇ r · δR(r) on the position
of the electrons. Therefore, the perturbing Hamiltonian, using Eqs. (10.34) and (10.36),
becomes
Hel-ph = −C1 ∇r · δR(r). (10.46)
We now express δR(r) in phonon coordinates:
(
 h̄
δR(r) = q̂eiq·r (aq + a†−q ), (10.47)
2ρω (q)
q
229 10.3 Polarons

where the index  counts longitudinal modes only and ρ is the mass density. From
Eq. (10.45), we have
(
 h̄
(r) = −i |q|eiq·r (aq + a†−q ). (10.48)
2ρω (q)
q

So, we have

Hel-ph = −C1 ∇r · δR(r)

 
= C1 k  |k c†k ck
k k
(
 h̄  
= −iC1 |q| aq + a†−q c†k+q ck , (10.49)
2ρω (q)
kq

because k eiq·k |k = δk ,k+q for planewaves. Hence, we obtain
(
h̄
Mq (k → k + q) = −iC1 |q|. (10.50)
2ρω (q)

This is the deformation-potential coupling matrix element.

We can estimate C1 for a free electron metal, where
 2/3
h̄2 3π 2 N
EF = . (10.51)
2m 

Since
1 δEF 21
× =− , (10.52)
EF δ 3
we have
δEF 2 δ 2
=− = − . (10.53)
EF 3  3
Hence, for metals,
2 2
δE = − EF  ⇒ |C1 | = EF ∼ 2 − 8 eV. (10.54)
3 3
For a semiconductor like silicon, if all the valence electrons are considered to be free
electron-like, then EF is of order 16 eV, and C1 would be of order 10 eV.

10.3 Polarons

We next explore an example of the coupling of an electron to LO phonons in a polar crystal.

As shown in Fig. 10.10, when a dressed electron moves in an ionically bonded crystal, the
 230 10 Electron–phonon interactions

(a) (b)

e– e–

Figure 10.10 An electron moving in an ionic crystal (schematic): (a) rigid lattice, and (b) in the presence of lattice distortion.

Coulomb attraction and repulsion of the positive and negative ions with the electron will
cause distortions or strains of the lattice. The displacements of the lattice caused by the
distortion can be expressed in terms of phonon excitations, and the combination of the
electron and the “cloud of phonons” produced is viewed together as a new quasiparticle,
a polaron. This use of the term polaron, which is an electron in a polar crystal coupled to
LO phonons via the Fröhlich interaction, is often extended to electrons coupled via lattice
polarization in other situations and materials, such as polymers and metals.
The electron–lattice polarization lowers the electron energy. It also increases the effec-
tive mass of the electron since the electron composing the polaron is now traveling with
a lattice strain. In addition, there is a change in the electron mobility since the polaron
experiences scattering effects that are different from those of a free quasielectron.
Often polarons are sorted into two classes, large and small. If we define the size of
a polaron p to be an effective radius of the phonon cloud or polarization area, then
large polarons are those for which p > d (interatomic distance), and this indicates
weak electron–phonon coupling. For small polarons, p  d suggests strong coupling.
Sometimes the coupling can be so strong that an electron becomes “self-trapped” in its
polarization region. For large polarons and weak coupling, we are in a regime where
we may use perturbation theory for the electron–phonon interaction. More sophisticated
mathematical techniques are needed to solve for polaron properties for stronger coupling.
If we consider the weak coupling case, the Hamiltonian for this problem is (with the last
term on the right side considered to be small)

H = Hel + Hph + Hel-ph

   (10.55)
= εk c†k ck + (a†q aq + 12 )h̄ωq + Mq (k → k )c†k ck (aq + a†−q ).
k q k kq

For the present case, we consider that the electron only couples to LO phonons, hence M
can be written (Eq. (10.32)) as

 1/4  1/2
h̄ωLO h̄ 4π α
Mq = i , (10.56)
|q| 2mωLO 
 231 10.3 Polarons

E E

+ Phonon

k k

Figure 10.11 Self-energy shift for a polaron.

where the dimensionless Fröhlich coupling constant is defined as

 '
e2 1 1 m
α= − , (10.57)
h̄ ∞ 0 2h̄ωLO

where m is the electron band mass. Using the definitions given in Eqs. (10.56) and (10.57),
we arrive at our earlier expression for M given in Eq. (10.32).
Assuming weak coupling α < 1 allows the use of perturbation theory to compute the
self-energy shift of an electron, its effective mass m∗ to lowest order, and the number of
phonons associated with an electron in a polaron.
The self energy arising from electron–phonon interactions lowers the energy of an elec-
tron in a band, as shown in Fig. 10.11. The expression for the self energy at T = 0, using
second-order perturbation theory, is (for an electron in state a)

 |a|Hel-ph |b|2
a = Ea = , (10.58)
Ea − Eb
b

where b are all possible intermediate states and the scattering processes are illustrated in
Fig. 10.12.
For the initial state a, we assume an electron is in a free electron state |k with no phonons
present so that the total wavefunction can be represented by |k, 0 and the total energy of
2 2
state a is Ea = h̄2mk . When the electron is scattered to an intermediate state b, the electron
is in state |k − q and one phonon is created. Hence, the total wavefunction is a†q |k − q, 0
and the energy is now

h̄2 (k − q)2
Eb = + h̄ωq
2m
(10.59)
h̄2 k2 h̄2 k · q h̄2 q2
= − + + h̄ωLO
2m m 2m
 232 10 Electron–phonon interactions

a b c

k k–q k

Figure 10.12 Scattering diagrams for calculating the self energy of a polaron.

and
 |Mq |2  |Mq |2
Ea = =− . (10.60)
Ea − Eb q h̄ω + h̄2 q2
− h̄2 k·q
b LO 2m m
If we denote
A
|Mq |2 = , (10.61)
q2
where
 1/2
h̄ 4π α
A = (h̄ωLO )2 , (10.62)
2mωLO 
and we convert the sum to an integral,
 

→ 2π q2 dqdμ, (10.63)
q
(2π )3

k·q
where μ = |k||q| , then
⎛ ⎞
 
− 1 qBZ 1
Ek = A dμ dq ⎝ ⎠. (10.64)
(2π )3 −1 0 h̄ωLO + h̄2 q2
− h̄2 kqμ
2m m

Although this may be evaluated exactly, we can extract the main physical results by
expanding for small k, letting qBZ → ∞, and evaluating the definite integrals to obtain

h̄k2  α 
E = −α h̄ωLO − + O(k4 ). (10.65)
2m 6

The energy of the electron is

h̄2 k2
E(k) = + E(k). (10.66)
2m
233 10.3 Polarons

Table 10.1 Experimental values of α for a variety of polar

crystals.3
substance α substance α

KCl 4.0 ZnO 0.85

KBr 3.5 PbS 0.16
AgCl 2.0 GaAs 0.06
AgBr 1.7 InSb 0.014

Combining Eqs. (10.65) and (10.66), we see that the ground-state energy is decreased by
the factor α h̄ωLO , and the k2 -dependent part of the energy, i.e. Ekin (k), has the form

h̄2 1 h̄2 k2
Ekin (k) = (1 − α)k2 = . (10.67)
2m 6 2m∗

Hence, the effective mass m∗ of the polaron is

m  α
m∗ = α ≈m 1+ . (10.68)
(1 − 6 ) 6

As expected, the polaron mass is larger than the electron mass at the bottom of the
conduction band.
Another interesting calculation related to polarons is to estimate the average number of
phonons coupled to an electron in a polaron at T = 0. The phonon number operator is
given by

N̂ph = a†q aq , (10.69)
q

and its average value is


N̂ph  = ψ| a†q aq |ψ, (10.70)
q

where |ψ is the polaron wavefunction. For an electron wavevector k,

|ψk  = |k, 0 + |δψk , (10.71)

where |δψ is the change in the electron wavefunction |k, 0 due to the electron–phonon
interaction.
Using standard perturbation theory, to first order in M we have

 Mq a†q |k − q, 0
|δψk  = . (10.72)
q
Ek − (Ek−q + h̄ωLO )

3 C. Kittel, Introduction to Solid State Physics, 8th ed. (New York, NY: Wiley, 2005).
234 10 Electron–phonon interactions

Following the same procedure as was done in Eqs. (10.60)–(10.64), the expression for the
average number of phonons given by Eq. (10.70) in |ψk  becomes

 |Mq |2
N̂ph  =  2 . (10.73)
h̄2 q2 h̄2 kqμ
q h̄ωLO + 2m − m

Converting the above sum to an integral and expanding for small k, we obtain to lowest
order in k  2  ∞
 2m dq α
N̂ph  = 2A  2 = . (10.74)
(2π )2
h̄2
2mωLO 2
0
h̄ + q2

So α is, in some sense, a measure of the number of phonons traveling with an electron in a
polaron.
Although the above three results, the self-energy shift, the renormalization of the ef-
fective mass, and the relationship between α and the average number of phonons “in a
polaron” are calculated only for small α, they give considerable insight into the properties
of this quasiparticle. As α becomes larger, this approach breaks down, but there are tech-
niques for dealing with stronger coupling.4 Some representative values for α in different
materials are given in Table 10.1.

4 R. P. Feynman, R. W. Hellwarth, C. K. Iddings, and P. M. Platzman, “Mobility of slow electrons in a polar

crystal,” Phys. Rev. 127(1962), 1004.
11 Dynamics of crystal electrons in a magnetic field

In general, the motion of an electron is quantized in the presence of a magnetic field.

We can learn a great deal about the electronic properties of a crystalline solid by exam-
ining the behavior of the material in a magnetic field. Various magnetic field-induced
phenomena have been observed and have been used as detailed probes in characteriz-
ing materials. Moreover, they have led to the discovery of new phases of matter and
have been exploited in various applications. Examples include striking phenomena such
as the integer and fractional quantum Hall effects, Wigner electron crystallization, and
giant magnetoresistance.
Much of the behavior of electrons in a crystal with an applied magnetic field may be
understood by assuming that the electrons are non-interacting fermionic quasiparticles. In
this chapter, we shall consider the dynamics of crystal electrons in a magnetic field within
this picture. That is, the electrons are non-interacting fermions but with masses and energy-
wavevector dispersion relations that are modified by band structure, electron–electron, and
electron–phonon interaction effects. We defer the discussion of the integer quantum Hall
effect to Chapter 12 as part of the discussion on transport properties.

11.1 Free electrons in a uniform magnetic field and Landau levels

We first consider the simple case of free electrons in a uniform magnetic field B. Let
B = Bẑ and allow the electrons to be confined to a large box of dimension L on each side
(see Fig. 11.1). Classically, we know that the electron moves freely along the z-directon,
and its motion in the x–y plane is confined to cyclotron orbits. Quantum mechanically, the
energy eigenvalue ε and eigenfunction ψ of the electron are given by
 2 
1 q
Hψ(r) = p − A(r) ψ(r) = εψ(r), (11.1)
2m c

where q = −|e| is the electron charge and A(r) is the vector potential, which, in the
Landau gauge, is given by A(r) = (0, x, 0)B. In this gauge, A depends only on x. Since A
is static and translationally invariant along the y- and z-directions, the energy and y- and
z-components of the momentum of the electron are constants of motion. Equation (11.1)
may be transformed to that of a one-dimensional harmonic oscillator. Substituting

ei(kz z+ky y)
ψ(r) = φ(x) (11.2)
L
235
 236 11 Dynamics of crystal electrons in a magnetic field

Figure 11.1 Free electrons confined in a box of dimension L in the presence of a uniform magnetic field B.

in Eq. (11.1) yields

  2   
p2x q2 B2 h̄c h̄2 k2z
+ x − ky φ(x) = ε − φ(x). (11.3)
2m 2mc2 qB 2m
The above equation corresponds to the Schrödinger equation of a simple harmonic oscilla-
2 2
tor located at position x0 = qB
h̄c
ky with spring constant κ = qmcB2 . The electron energies are
given by
 1 h̄2 k2z
ελ (kz ) = λ + h̄ωc + , (11.4)
2 2m
with

λ = 0, 1, 2, 3, . . . , (11.5)

'
κ |e|B
ωc = = , (11.6)
m mc
and
 
2π
kz = nz , (11.7)
L
where nz are integers.
The expression for ωc has the same value as the classical cyclotron frequency. Thus,
the electron behaves like a free particle parallel to B and performs cyclotron motion in the
2 2
plane perpendicular to B. The free electron energy dispersion relation ε(k) = h̄2mk is now
changed into a series of parabolic curves as a function of kz labeled by the quantum number
λ, as illustrated in Fig. 11.2. Each of the curves is called a Landau level.1 The Landau levels
are highly degenerate because any value of ky yields the same set of solutions in Eq. (11.3).
For given λ and kz , it can be shown that the number of degenerate states D is given by
 
2|e| 2BL2
D= BL2 = . (11.8)
hc φ0

1 L. Landau, “Diamagnetism of metal,” Z. Phys. 64(1930), 629.