Vous êtes sur la page 1sur 56

OPTICAL TRANSITIONS, EXCITONS, AND POLARITONS IN

BULK AND LOW-DIMENSIONAL SEMICONDUCTOR STRUCTURES

Lucio Claudio Andreani


Dipartimento di Fisica "A. Volta"
Universita degli Studi di Pavia
via A. Bassi, 6
1-27100 Pavia, Italy

INTRODUCTION

Electrons in solids are subject to the crystal potential, as well as to the mutual electronelectron interaction. The resulting quantum-mechanical system represents a many-body
problem of great complexity. In weakly-correlated systems like the usual semiconductors, a good starting point is provided by the one-particle picture, in which the crystal
eigenstates are approximated by Slater determinants where the electrons occupy the
one-particle eigenstates called band levels. This is only an approximate picture, since
the electron-electron interaction yields corrections to the excited-state spectrum of the
crystal. In particular, two-particle excitations called excitons arise at energies below
the band gap, and excitonic corrections are found also at energies above the band gap.
The electronic states of a crystal can be probed using an external electromagnetic
field. The study of the optical properties gives very precise information on the electronic structure of semiconductors. In analyzing the radiation-matter interaction, it is
useful to distinguish between instantaneous and retarded parts of the electromagnetic
field. The unretarded (c -'; 00) part describes the instantaneous Coulomb interaction
and corresponds to the longitudinal electromagnetic field, while the retarded part is
identified with the transverse electromagnetic field, i.e., with the physical photons. For
one-particle states, interaction with the transverse electromagnetic field gives rise to
interband and intraband transitions. For two-particle (excitonic) states, interaction
with the longitudinal part of the electromagnetic field corresponds to the electron-hole
exchange interaction, while interaction with the transverse electromagnetic field gives
rise to polariton effects.
The above picture applies to bulk semiconductors as well as to mesoscopic structures like quantum wells, wires and dots. However both the electronic states and

Confined Electrons and Photons


Edited by E. Burstein and C. Weisbuch, Plenum Press, New York, 1995

57

the radiation-matter interaction are modified by the reduced dimensionality. One of


the most important modifications concerns the polaritonic effect. In infinite crystals,
the conservation of crystal momentum implies the formation of quasi-stationary states
called excitonic polaritons. In confined systems, the lack of crystal momentum conservation in one or more directions implies that an exciton interacts with a continuum
of photon states, thereby providing a mechanism for intrinsic radiative decay of free
excitons.
In these lecture notes we give an outline of the theory of linear optical properties in
semiconductors, considering both bulk and confined systems. We consider one-particle
properties (interband and intraband transitions) as well as exciton and polariton effects.
One main point is the modification of exciton states and polariton effects in going from
bulk to quantum-well systems, considering in particular the exciton radiative lifetime.
In Sec. 2 we treat optical properties in bulk semiconductors. After a review of
the classical theory of dielectric properties, including the Lorentz-oscillator model and
Kramers-Kronig relations, we describe the calculation of the dielectric constant by a
semiclassical theory of the radiation-matter interaction. We then give an outline of
the quantum theory of excitons and polaritons. In Sec. 2.5 we summarize the present
understanding of the radiative recombination of excitons in bulk semiconductors, which
is a basic problem but still not completely understood. In Sec. 3 we discuss optical
properties in confined systems, particularly in the quantum-well geometry. Following a
similar scheme, we first review the theory of one-particle transitions, and then consider
exciton states and polariton effects. In Sec. 3.4 we discuss the radiative lifetime of
excitons in going from bulk to confined systems. In Sec. 4 we summarize the main
points treated in these lectures. In the Appendix we give rules for converting from the
Gaussian (cgs) to the SI (MKSA) system of units.
2
2.1

OPTICAL PROPERTIES IN BULK SEMICONDUCTORS


Classical Theory of Dielectric Properties

Classical electromagnetic theory. Maxwell equations in matter must be supplemented by constitutive relations. We neglect magnetic effects, and consider only linear
response. In a homogeneous and isotropic medium, the Fourier components of the
polarization and displacement fields are related to the electric field by (Gaussian units)

XE,

+ 47rP = tE,

(1)
(2)

where t(w,k) = 1 + 47rX(w,k) is the frequency- and wavevector-dependent dielectric


function. The wavevector dependence of t is referred to as spatial dispersion [1]. It
can often be neglected since the wavelength of radiation is much larger than the lattice
spacing. In a crystal, which is invariant only under translations by lattice vectors, the
macroscopic dielectric function contains local-field effects [2]. In anisotropic media, the
dielectric function becomes a 3 x 3 tensor, which reduces to a scalar only for cubic
crystals in the long-wavelength limit.
The displacement current appearing in Maxwell equations is given by

J _~aE
d -

47r at

(3)

In the presence of free charges, an induced current appear according to Ohm's law,
J ind

58

= erE.

(4)

Both currents (3),(4) can be taken into account by introducing a complex dielectric
function,

(5)
At finite frequencies, the conductivity u(w) is also complex. Thus it is a matter of
convention to attribute the dispersive properties of a medium to a real dielectric function
plus a conductivity, or to a complex dielectric function E = E1 + iE2'
The index of refraction,
N = y

== n

+ iK"

(6)

is also complex. The absorption coefficient is obtained as

2w

a =

When

E2

(7)

-K,.

~ E1,

Eq. (7) can be approximated by

(8)
This is a good approximation for semiconductors, but it may fail for metals (see Sec.

2.2).
Oscillator model. A simple classical picture for the dielectric function is provided
by the Drude-Lorentz model [3, 4J. The electrons in a crystals are represented by a
collection of damped harmonic oscillators, which respond to an applied electric field
according to the equation
mo [x"

2 J =
+ ijX + WjX

-e E loc,

(9)

where Eloc is the local field acting on the oscillator and mo is the free-electron mass.
The induced dipole moment for each oscillator is
Pj

= -eXj = mo (2
Wj -

(10)

2 - tijW
. ).

The macroscopic polarization is found by multiplying Pj by the number of oscillators


per unit volume, which we denote by Ii IV, and summing over all resonances. If the
local field is identified with the applied field (i.e., if local-field corrections are neglected),
the resulting dielectric function is

E( w)

= 1

47re 2 1
mo V

Ii

L w~ _ w 2 _ ii"W
J

(11)

The oscillator strength Ii is a dimensionless quantity, which represents the number of


classical oscillators at frequency Wj. Since the total number of oscillators equals the
number N of electrons in the crystal, the oscillator strengths must satisfy the sum rule

LIi = N.

(12)

This sum rule can be derived by comparing the high-frequency limit of Eq. (11) with
the known limiting form of the dielectric function,

(13)
59

w;

where
= 47rNe 2/(moV) is the plasma frequency.
Equation (11) yields the real and imaginary parts of the dielectric function, the
index of refraction, absorption coefficient, etc. Close to a resonance frequency, the real
part l(W) shows anomalous dispersion, while 2(W) yields resonant absorption [3,4].
Free charges contribute a frequency-dependent conductivity u(w) = ne 2/(m*(( - iw)),
where m* is the effective mass. The relation between (11) and the electronic structure
is given by a microscopic calculation of the parameters Wj, Ii, and ij. In a solid, in
the energy region of interband transitions, the frequencies Wj form a continuum. A
quantum-mechanical calculation of the dielectric function, with a suitable definition of
the oscillator strength, will be described in Sec. 2.2.
Dispersion relations and sum rules. General properties of the optical constants
follow from the causality principle. The displacement field is related to the electric field
by

l)O x(r)E(t-r)dr,

D(t) = E(t)+47r

(14)

where the response function x(r) can be taken to vanish for r < O. As a consequence,
the Fourier transform X(w) is analytic in the upper half-plane. X(w) is also analytic
on the real axis (except for conductors, where (w) - t 47riu/w as w - t 0) and vanishes
rapidly for Iwl - t 00. Thus by the Cauchy theorem

1:

00

~'~~ 1 dw' -

i7r((w) - 1)

= 0,

(15)

where P denotes the principal value. Since the fields D( t) and E( t) are real, it follows
that (w)* = (-w). Separating real and imaginary parts of the above equation, we
obtain the Kramers-Kronig relations for the linear dielectric function [4, 5]
(16)
For conductors, the contribution from the pole at w = 0 has to be exhibited separately.
Dispersion relations analogous to (16) holds also for the real and imaginary parts of
the index of refraction, and of the logarithm of the reflectivity. This is of the utmost
importance for determining all optical constants with a single measurement, e.g. of the
reflectivity [5]. Kramers-Kronig relations are very useful in the design of devices like
optical modulators, in which the variation in phase of the electric field can be deduced
from absorption measurements.
Very useful sum rules can be obtained from (16) and the super convergence theorem
[6]. From the first dispersion relation in the limit w - t 00, and the asymptotic form
(13) for the dielectric function,
00

7r
W2(W) dw = -w;.

(17)
2
This is sometimes called the f-sum rule, since it implies the sum rule (12) when the
dielectric function has the Lorentz-oscillator form (11). Relation (17) is obviously more
general. From the second of (16) we obtain

Ino

10

00

(l(W) - 1) dw

Moreover, letting w

1(0) = 1 + ~

-t

(18)

0 in the first of (16),

roo 2(W) dw.

7r io

60

= O.

(19)

For semiconductors, where the second term dominates and interband absorption starts
at the band-gap energy, (19) shows that the electronic contribution to the dielectric
constant scales roughly with the inverse of the gap.
It has been found recently that Kramers-Kronig relations can be derived also for
the nonlinear optical functions [7, 8]. Nonlinear sum rules are also found to hold, and
have been verified experimentally.
2.2

Quantum Theory of Radiation-Matter Interaction

Time-dependent perturbation theory. The effect of a radiation field on the electronic states of a crystal can be studied by time-dependent perturbation theory. To
lowest order, the transition probability per unit time from an initial state Ii) to a final
state If) in the presence of a harmonic perturbation HI = e'fiwt is given by

(20)
where the upper (lower) sign corresponds to absorption (emission).
mechanical interaction between radiation and matter is
HI

=-

:L)A(ri,t). Pi

2moC i

e2

+ Pi' A(ri,t)) + - - 2 I:A(ri,t)2,


2moc

The quantum-

(21)

where the sum runs over all electrons in the system. The A2 term can be neglected
for linear processes. In the Coulomb gauge the terms A . P and P . A coincide. The
perturbation (21) induces transitions between crystal eigenstates. The first-order expression (20) describes one-photon absorption or emission, while higher-order terms in
the time-dependent perturbation series describe two-photon and higher-order processes.
We want to calculate the absorption coefficient at absolute zero, and therefore consider only absorption processes. Here we outline the semiclassical treatment, is which
the electron system is treated quantum-mechanically, but the radiation field is taken as
a classical field. We follow the treatment of Ref. [9], to which we refer for details. For
interband transitions, a second-quantized treatment of the electromagnetic field gives
the same result for the absorption coefficient.
In the one-electron approximation, the crystal ground state is given by a Slater
determinant in which the electrons occupy all Bloch states in the valence band:

(22)
where A is the antisymmetrizing operator. Spin coordinates are understood. Oneelectron excited states are given by

(23)
where the valence Bloch function 'ljJvkv has been replaced by the conduction Bloch
function 'ljJck c ' The matrix element of the interaction (21) between initial and final
states can be calculated using a well-known expression for the matrix element of a sum
of one-particle operators between determinantal states [9]. Keeping only the A P term,
the result is
(fIHIli)

= _e_('ljJ
k IA. pl'ljJ k ).
moc
c
v v
C

(24)

For a plane wave of frequency w, the vector potential is


A(r, t)

= Aoeei(q.r-wt) + h.c.,

(25)
61

where q is the wavevector of radiation and i is the polarization vector. The matrix
element (24) embodies the conservation of crystal momentum: kc = k y + q + G, where
G is a reciprocal lattice vector. Since the wavevector of light in the visible region is
much smaller than the dimension of the Brillouin zone, while the matrix element (24)
varies on a scale given by the Brillouin zone, we can neglect the light wavevector q.
Moreover, since the initial and final wavevectors are taken in the first Brillouin zone,
Umklapp processes (G =I 0) give no contribution. Thus we recover the well-known
result that only "vertical" transitions are allowed. The resulting expression for the
transition probability between a pair of initial and final states is

(26)
where

(27)
is the momentum matrix element between valence and conduction Bloch functions. The
total transition probability per unit time and volume is obtained by summing over all
pairs of conduction and valence bands and over all wavevectors and spin indexes, using
the relation

~L2L= L
V

Cy

Cy

JBZ

2dk
(211")3

(28)

where the integral extends over the first Brillouin zone. The absorption coefficient,
defined as
( ) _ energy absorbed/unit time & volume
a w - incident energy/unit time & surface'
is calculated to be
2 2

a(w)

= -411"2e- L
nmocw

Cy

1 - ()3I
BZ

2dk i
. Mcy(k)f5(Ec(k) - Ey(k) -fiw).
211"

(29)

(30)

Using the relation (8), the imaginary part of the dielectric constant is obtained as
2 2

2(W)

411" e
= -2-2
L
mow
Cy

1 -()3I i . Mcy(k)f5(Ec(k) - Ev(k) -fiw).


BZ

2dk
211"

(31)

This is the basic expression we set out to find. By using the relation

(32)
between momentum and dipole matrix elements, it can be checked that expression (31)
for 2(W) satisfies the sum rule (17). Using the Kramers-Kronig relation, the real part
of the dielectric function is obtained as

(33)
Oscillator strength. Quite generally, the oscillator strength for a transition from
a state Ii) to a state If) is a dimensionless quantity defined by

(34)

62

where liw = Er - E j , and depends on the polarization vector E. The oscillator strength
can also be expressed in terms of the dipole matrix element as
(35)
It obeys the sum rule Er f<. = N, where N is the total number of electrons. For the
case of direct interband transitions, (34) reduces to

(36)
Using this definition, it can be seen that close to a resonance frequency the quantummechanical expression (31)-(33) for the dielectric function coincides with the classical
expression (11), provided the a-function is broadened to a Lorentzian (to recover the
classical expression in the whole frequency range, emission terms must be included in
the quantum-mechanical calculation). The f-sum rule (12) can be seen to be fulfilled
by Eq. (36). The oscillator strength integrated over a frequency interval is related to
the absorption coefficient by

(37)
where

1
ftot = 2:::
V
cy

1
WI

2dk
- - 3 f<.
<W<W2 (27r)

(38)

Direct interband transitions: critical points. For allowed interband transitions, the momentum matrix element Mcv(k) is nonzero and slowly-varying. Thus the
structures in the dielectric function are dominated by the joint density of states,

(39)
This quantity has singularities when
( 40)

which usually occurs at high symmetry points (where VkEc(k) = VkEy(k) = 0) or


along symmetry lines. Close to a critical point ko, the energy difference can be expanded
up to quadratic terms in k - ko and the leading term in the energy-dependence of the
joint density of states can be determined. The calculations are described e.g. in Ref. [9).
In Fig. 1 we show the behavior of the joint density of states close to critical points in
three-dimensional, two-dimensional, and one-dimensional crystals. Note in particular
that the absorption coefficient just above the direct gap behaves as (fiw - E g )1/2 for 3D
systems, and as a constant for 2D systems. The optical spectra of many semiconductors
have been interpreted on the basis of the above critical-point analysis. We refer to
[3, 9, 10) for details.
Intraband transitions. When the top of the valence band and the bottom of the
conduction band occur at different points of the Brillouin zone, direct transitions are
not possible. This occurs e.g. in Si and Ce. In this case, optical absorption is still
observed as a tail of lower intensity, due to indirect transitions. Such transitions are
made possible by the electron-phonon interaction.

63

(a)

MO

(
E

(b)

(c)

i~ ~
I
I
I
I
I
I

l
E

Figure 1. Schematic behavior of the joint density of states for inter band transitions at critical points
for (a) three-dimensional, (b) two-dimensional, and (c) one-dimensional crystals.

64

In metals and in doped semiconductors, intraband transitions can occur at long


wavelengths. Intraband transitions cannot be obtained by the A . p interaction alone,
since the requirements of energy and crystal-momentum conservation cannot be satisfied
simultaneously; thus a dissipation mechanism like the electron-phonon interaction is
necessary. Intraband absorption is well described by free-carrier absorption [5], given
essentially by the Drude model of Sec. 2.1. Separating the conductivity contribution to
the dielectric function, we have
( w) = oo

47ri

+ -w

ne 2

( .)
m* 1 - 2W

(41)

The two frequency scales in this expression are the inverse I of the relaxation time,
and the screened plasma frequency Wp = (47rne 2/( oom*))1/2. In metals, usually I rv
1013 8- 1 , while wp is in the ultraviolet region. For W I, the absorption coefficient is
( 27rWlTO)1/2 ,
a (W ) _
- 2
c

(42)

and the electromagnetic field decays within a skin depth Ii = 2/a = c/(27rWlTO)1/2 (where
lTo = ne 2/( m*l) is the zero-frequency conductivity). From Fresnel formulas, the normal
incidence reflectivity is

R(w) = 1 - ( 2w )1/2,

(43)

7rlTo

which is known as the Hagen-Rubens formula. For


be approximated by
2

(w) ~ oo[l-

I, the dielectric constant can

w
w
-1
+ h-1].
w
w

(44)

The reflectivity has a plateau for I W wp and drops towards zero at Wp , where
the metal becomes transparent. For w > w p , Eq. (44) leads to an absorption a(w) =
y'E;;;lw;/( cw 2 ) and to a reflectivity which increases towards its asymptotic value [( y'E;;;1)/(y'E;;; + 1)]2. However this is rarely observed in metals, since interband transitions
modify the free-electron behavior, or in other words the contribution of bound electrons
to (w) becomes frequency-dependent.
2.3

Excitons in Bulk Semiconductors

Quantum theory of excitons. Excitons are excited states of the whole crystal,
which go beyond the one-electron approximation. They correspond to bound states
of an electron in the conduction band and a hole in the valence band. Excitons in
semiconductors are usually shallow, i.e., their radius is much larger than the interatomic
spacing: as such, they can be described by a two-particle effective mass equation. The
opposite extreme, i.e. that of tightly bound (or Frenkel) excitons, is appropriate to
molecular crystals. In this section, an outline of the theory of shallow (or WannierMott) excitons is given [11, 12,9].
We take as a starting point the N-electron Hamiltonian of the crystal, where, for
the moment, we do not consider spin-orbit terms. The system ground state is given by
the Slater determinant (22). In order to find exciton levels, we restrict ourselves to the
subspace of states of the form (23) (we consider only one pair of bands for simplicity).
The exciton wavefunction is represented as

(45)

65

where k is the relative electron-hole wavevector, and k ex is the exciton Bloch vector which describes the center-of-mass motion of the electron-hole pair. The exciton
wavevector k ex is a conserved quantum number, whose existence follows from translational symmetry of the crystal. The matrix elements of the N-electron Hamiltonian
between the basis states (23), referred to the ground state energy, are
(Wcke,vkv IHIWck~,vk~) - (WoIHlwo)

(Ec(kc) - Ev(kv))5kek~5kvk~

+ J(k,k') + Vc(k,k'),

(46)

where

(47)
(48)
and kc = k + k ex /2, etc. Note that the term J(k, k'), which corresponds to an electronhole exchange, comes from the direct matrix element between Slater determinants,
whereas the electron-hole Coulomb attraction Vc(k, k') comes from the exchange matrix
element between Slater determinants. As a result of many-body theory [13J, only this
latter term must be screened by the dynamical dielectric function. Using the above
matrix elements, the Schrodinger equation can be written in real space for the Fourier
transform of A(k),
F(r) = _1_2:A(k)eikor.

v'V

(49)

The Schrodinger equation for the envelope function takes the form

+ k;x) -

[Ec( -iV

Ev( -iV _ k;x) - EJF(r)

Jdr'

[Vc(r,r')

+ J(r,r')JF(r') = 0,

(50)

where

J(r, r')

= ~ 2: e ikor J(k, k')e-iklorl,


V

Vc(r, r')

(51)

kk'

= ~2: eikorVc(k, k')e-iklorl,

(52)

kk'

and V

= NO, is

the crystal volume.

Effective mass equation. Degenerate bands. In the effective mass approximation, the Coulomb and exchange potentials in real space can be considerably simplified
using the fact that the envelope function is strongly peaked in k-space. The Coulomb
potential can be shown to reduce to
,2

Vc(r, r') ~ -~5(r - r').

(53)

EbT

If the exciton radius is much larger than the polaronic radius, Eb must be identified with
the static dielectric constant EO, which includes the contribution of optical phonons to
screening (this is the case, e.g., of GaAs). If, on the other hand, the exciton and
66

the polaron radius are of the same order, the electron-phonon interaction must be
dynamically computed [14]: this is the case, e.g., of CuCl.
An exciton associated with a pair of parabolic bands has energy levels of the form
(neglecting the exchange interaction)

En{k ex ) = Eg -

R*
n2

fi 2 k ex 2

(54)

2M '

where
fLe 4
fi2
R* - - - = --:---:-::- 2~fi2
2fL{as)2

(55)

is the effective Rydberg, fL = (11m: + 1/mhtl is the reduced effective mass and M =
is the total effective mass. The exciton radius is given by
= fi2b/{fLe 2)
and is usually much larger than the hydrogenic Bohr radius, since semiconductors have
large dielectric constants and small effective masses. This result justifies a posteriori
the use of the effective mass approximation.
In semiconductors with zincblende structure, the degeneracy of the uppermost valence band (see Fig. 2) must be taken into account. The formalism described above
can be generalized to the case of degenerate bands [15], and the resulting effective mass
Hamiltonian is a matrix of the form

as

m: + mh

(56)
where TSSf{k) is the Luttinger Hamiltonian for holes [16] which describes the spin-orbit
interaction. The total electron-hole momentum k ex = ke + kh is a constant of the
motion, and is equal to zero for optically active excitons in direct gap materials. The
coupled set of effective mass equations cannot be solved exactly. However, a considerable simplification comes from the fact that the small electron effective mass appears
only in the diagonal terms, which are therefore much larger than the off-diagonal ones
[17]. The effective mass Hamiltonian in the relative coordinate can be separated into
an isotropic part, which depends on the parameter fLo = (11m: + Ilt l , and ad-like
anisotropic term, which can be treated in second-order perturbation theory. Hence the
spectrum of direct excitons in semiconductors with degenerate bands is found again to
be very nearly hydrogenic, the relevant effective mass being fLo.
Oscillator strength. Symmetry properties and selection rules. The excitonic oscillator strength is defined again by Eq. (34), where the initial and final states
now refer to the crystal ground state and the exciton state respectively. Evaluating the
momentum matrix element in the effective mass approximation yields [12]
(57)
where the additional factor of two comes from the choice of singlet exciton states. The
oscillator strength vanishes for triplet states. Note that the definition (34) assumes
that the excitonic state is normalized to unity: because of (49), the real-space envelope
function must also be normalized to unity. For hydrogenic s-states,

IFn{OW = 7rn 3 a sP

(58)

Thus Eq. (57) shows that the dimensionless oscillator strength is proportional to the
crystal volume. This is due to the fact that the exciton center-of-mass wavefunction

67

18 3/2

Figure 2. Schematic representation of the top of the valence band and the bottom of the conduction
band in direct-gap cubic semiconductors. Notations appropriate to the zincblende structure are used.

extends over the whole crystal. The meaningful quantity is the oscillator strength
per unit volume, which for an allowed transition decreases as n- 3 When polariton
effects (see Sec. 2.4) can be neglected, the oscillator strength per unit volume of an
isolated resonance can be related to the absorption coefficient a( w) integrated over the
absorption peak. The relation is

271' 2 e2

a(w) dw = - - - ,
lIine
nmoc V
J

(59)

and coincides with (37) found for interband transitions. It can be shown that the total
oscillator strength in the exciton line corresponds to that for interband transitions,
integrated over a frequency interval of the order of the exciton binding energy.
The excitonic effect introduces discrete absorption lines below the band gap. Above
the gap, excitonic effects give a correlation between the unbound electron-hole pair,
and the absorption coefficient must be computed from the solutions of the hydrogenic
Schrodinger equation in the continuous spectrum [18]. The result can be conveniently
expressed in terms of an enhancement factor (often called Sommerfeld factor), defined
as S(w) = aexc(w)/ainterband(W), where aexc(w) is the absorption coefficient including
the excitonic effect. In three dimensions, the Sommerfeld factor is given by

S( )
W

27l'X

= 1 _ exp( -271'x) ,

= (

R* ?/2.
fiw - Eg

(60)

A schematic plot of the absorption coefficient with and without excitonic effects is
shown in Fig. 3. The enhancement factor diverges as an inverse square root close
to the band gap, thereby making the absorption coefficient finite. By considering the
absorption due to closely spaced exciton lines just below the gap, it has been shown [18]
that the absorption coefficient including excitonic effects is continuous at the interband
absorption edge. This last result holds also for 2D and ID systems, although the form of

68

n=l

n=2 n=3

--,

,,

/
/

I
I

E
Figure 3. Schematic picture of the absorption coefficient close to the interband absorption edge for
3D crystals. Dashed line: without excitonic effect. Solid lines: including excitonic effects.

the joint density of states and of the Sommerfeld factor are modified in low-dimensiona:.
geometries (see Sec. 3.1).
Up to now we have not explicitly considered spin variables. If there is no spin-orbit
interaction, exciton states must be classified according to total spin, and are either
triplets or singlets. The oscillator strength is nonzero only for singlet states, since the
radiation-matter interaction (21) does not involve spin. When the full Hamiltonian (including spin-orbit interaction) is considered, excitons at kex=O are classified according
to the irreducible representations of the crystal point group, i.e., the symmetry group
at k = o. The exciton symmetry is contained in the product decomposition of [19, 9]
(61)
where rc (rv) are the irreducible representations of conduction (valence) bands at k =
0, and r env is the symmetry of the exciton envelope function. An important point
to note [19] is that Eq. (61) gives the exact exciton symmetry, although an effective
mass description of the exciton is only approximate. An exciton at k ex = 0 contains
conduction and valence states with finite k, which, however, enter in linear combinations
which have the same symmetry of the bands at k = O. In optical transitions between
the crystal ground state and exciton states, optically active excitons are those which
have the symmetry of the transition operator, i.e., of the dipole. The oscillator strength
in the presence of spin-orbit interaction is obtained from (57) by replacing the factor of
two by a spin-orbit factor g, which is two times the singlet component in each irreducible
representation.
To give an example, in semiconductors with zincblende structure and point group
Td the bottom of the conduction band is the twofold degenerate state r 6 , and the top
of the valence band has usually the fourfold degenerate rs symmetry [20] (see Fig. 2).
The lowest exciton state has multiplicity rs r6 = r3 EB r 4EB r s , and the threefold
degenerate dipole active state rs has a spin-orbit factor 9 = 4/3. For the case of CuCI,
where the spin-orbit splitting is negative and the top valence band has r7 symmetry,

69

the 8 excitons decompose into r7 r6 = r 2EB r s, and the optically active rs state has
a spin-orbit factor 9 = 2/3.
To obtain the absorption coefficient at the band edge [18], we use Eq. (30) (with
the factor of two for spin replaced by the spin-orbit factor g) and multiply by the
Sommerfeld factor (60). The crucial point is now to calculate the interband momentum
matrix element. This can be expressed in terms of the conduction band effective mass,
according to the Kane model [21]:
mo

rv

m; -

1 + --'-''----=-=---:.:'---'.---'-''--'-'-21(u~g)IPzlu~~)}12 _~----"-.,---_
Eg + ~Llso
moEg
Eg + Llso '

(62)

where u~g), u~~) are the zone-center Bloch function with Sand Z symmetry, respectively. This procedure is much more accurate than evaluating the dipole matrix element
(ucolrluvo) from the Bloch functions. The absorption coefficient at the band edge can
be put in the form

271" e 2 fl, 1
n ficm;

(63)

a~g----.

as

For zincblende semiconductors, the spin-orbit factor 9 = 4/3, and the reduced effective
+ 'Ylt 1 , as discussed above. Thus the absorption coefficient at the band
mass fl, = (~
me
edge scales essentially as the inverse exciton radius. For GaAs, using the value
= 144
Afor the exciton radius, we obtain a ~ 104 cm-I, in agreement with the experimental
value [22].

as

Exchange interaction. The electron-hole exchange interaction can usually be


treated as a first-order perturbation to the solutions of the effective-mass equation. It
is important, however, since it splits the exciton states belonging to different representations in Eq. (61) with respect to the crystal point group.
The evaluation of the matrix element (47) can be done in two different ways. In the
Wannier-function formulation, first introduced by Onodera and Toyozawa [23] and later
used by several authors [24, 9, 25, 26], the Bloch functions are expanded in Wannier
functions and the exchange matrix element is expressed as a sum over lattice sites r.
The exchange interaction is then separated into a short-range part, given by the r = 0
term, and a long-range part, given by the r ~ 0 terms in the sum. The short-range
part splits the excitons belonging to different irreducible representations. The longrange part is evaluated by expanding the Coulomb potential up to the dipole term.
It is nonanalytical as k ex --+ 0, and gives rise to a splitting between longitudinal and
transverse exciton states.
In the k-space formulation [27,28] the products of the periodic parts of the Bloch
functions are expanded in plane waves with reciprocal lattice vectors, and the exchange
matrix element is rewritten as

where
(65)
and Uck,Uvk are the periodic parts of the Bloch functions. In expression (64), the sum
of the G ~ 0 terms gives a contribution Jana which has a well-defined limit as k ex --+ 0,
and therefore is called the analytic part. The G = 0 term, instead, has different limits

70

as k ex ~ 0 depending on the direction of k ex with respect to the exciton dipole moment:


it is therefore called the nonanalytic part. The nonanalytic part can be evaluated by
expanding the periodic parts of the Bloch functions in k . p theory around the band
extremum (taken at k = 0), with the result

Jllon-ana(k ex ) =

~~:2 (uclkex ' rlu v) (uvlk ex . rluc).

(66)

ex

For k ex ~ 0, the nonanalytic part has a limiting value 47re 2 1(uclrlu v /n for an exciton
wavevector parallel to the dipole moment (longitudinal exciton), and a vanishin limiting value for an exciton wavevector perpendicular to the dipole moment (transverse
exciton). The division into analytic and nonanalytic parts is related, but not identical to the division into short-range and long-range parts. It is possible to show that
inclusion of the analytic part of the exchange interaction corresponds to taking into
account local-field effects, while inclusion of the nonanalytic part corresponds to taking
into account the depolarization field in the dielectric response [29].
In the effective mass approximation, both terms Jana, Jnon-ana are independent
of (k,k'). The corresponding term in the effective-mass equation (50) is gn(rna +
Jnon-ana)5(r)5(r'), where 9 is the same spin-orbit factor which appears in the oscillator
strength. When the Schrodinger equation for the exciton is projected onto a pair of
bands, the effect of exchange coupling to the other bands results in a screening of the
nonanalytic part of the exchange interaction with the background dielectric constant
[30, 28]. The splitting between longitudinal and transverse excitons is then calculated
to be

(67)
The LT splitting is proportional to the oscillator strength per unit volume according to
the relation

(68)
A schematic representation of the effect of the exchange interaction on the ground-state
exciton in zincblende semiconductors is shown in Fig. 4. The analytic part of the exchange interaction separates the optically active state r s , which is a linear combination
of singlet and triplet excitons, from the forbidden states which are pure triplets. In
addition, the nonanalytic part of the the exchange interaction splits the rs states into
a longitudinal and a transverse exciton, with degeneracies one and two respectively.

2.4

Polaritons in Bulk Semiconductors

The dispersion of optically active excitons taking into account the exchange interaction
presents a splitting between longitudinal and transverse modes. This result cannot hold
for the true eigenstates in the limit k ex ~ 0: in cubic crystals, states with the symmetry
of the dipole must be threefold degenerate at k cx = O. The solution to the paradox lies in
the observation that the interaction of the exciton with the transverse electromagnetic
field cannot be neglected for wavevectors kex < ko = nw / c. For wavevectors of the
order of the wavevector of light in the sample, excitons are strongly coupled to photons
and form the quasiparticles which are now called excitonic polaritons (for reviews, see
[31, 32, 33, 34, 35, 36]).

71

L
/

'"
I

IS (3)
/
T (2)
: L'lELT
...-------<.L---------it:;
I
I

I
I

__IB_X_r::_6_(8_)_-,-/ ___

13 +

r::

: L'l

(5)

t
I

Figure 4. Schematic representation of the effect of the exchange interaction on the ground-state
exciton in zincblende semiconductors. Numbers in parentheses indicate the degeneracy of the state.
11 is the analytic part of the exchange splitting.

Semiclassical theory. The classical theory of polaritons was first developed by


Huang [37, 38] to describe infrared absorption of light below the Reststrahl frequency
due to phonon resonances. Assuming a dipole-active resonance at frequency Wo, the
dielectric function is
(

w) =

47r{3w6

oo

+ Wo2 -w 2 '

(69)

where oo represents a frequency-independent contribution due to all other resonances


in the crystal, and the polarizability {3 is related to the oscillator strength per unit
volume by
e

{3

(70)

= mow5 V

Solving Maxwell equations with the constitutive relation D


dispersion relations:
c2 k 2
-2

(w)

= ( w)E yields the following


(71)

for transverse modes, and

(w) = 0

(72)

for longitudinal modes. The dispersion relations are schematically illustrated in Fig.
5. The transverse modes show a lower and an upper branch, which anticross close to
the wavevector of light in the sample. Both transverse modes are twofold degenerate.
In the instantaneous limit c --t 00, the transverse solution has a frequency woo The
longitudinal mode is a pure electrostatic solution. At kex = 0, the two upper polariton
modes are degenerate with the longitudinal exciton, so that a threefold degenerate state
is recovered as required by cubic symmetry. The electrostatic shift of the longitudinal
mode is
nWLT = nwo ( 1

47r{3 1/2
+) oo

27r{3
nwo c::: -nwo.

(73)

Eoo

This can be seen to coincide with expression (68) for the LT splitting. In other words,
the electrostatic shift of the longitudinal exciton calculated from the nonanalytic part

72

Figure 5. Schematic representation of the dispersion of the upper and lower polaritons and of the
longitudinal exciton (solid lines). The dispersion of the independent photon and transverse exciton

modes is also shown (dashed lines).

of the exchange interaction coincides with that calculated from the solution of Maxwell
equations in the limit of no retardation (this is the instantaneous, or electrostatic limit,
which is obtained by letting c -) 00).1
For excitons in semiconductors, the polariton dispersion turns upwards at large
wavevectors due to spatial dispersion, i.e., to the motion of the center of mass of the
exciton (see Fig. 5). For frequencies W > WL, this results in the presence of two propagating modes at the same energy [39]. If the optical response for light incident on a
semi-infinite crystal is calculated using the bulk dielectric function, Maxwell boundary
conditions are not sufficient in order to specify the relative amplitude of the two modes,
and additional boundary conditions (ABC) are needed. A vaste literature exists on the
ABC problem [39, 40, 33, 35, 36]. A commonly used ABC consists of assuming the
existence of a dead layer, where the exciton cannot penetrate, and taking the excitonic
polarization to vanish at the boundary of the dead-layer region [39, 40]. The ABC are
"mechanical" boundary conditions which describe the dynamics of the exciton close to
the surface. If the Schrodinger equation for the exciton is solved in the semi-infinite
crystal, and the nonlocal optical response is calculated using linear-response theory,
ABC need not be introduced. Such kinds of ABC-free theories have been recently
developed [41,42].
The meaning of Fig. 5 is that it is the dispersion of polaritons (not that of the excitons) which is measurable in good crystals at low temperature. In fact, the dispersion of
I We are using a dielectric function which describes the response to the total electromagnetic field,
in which case the exciton resonance frequency must not contain the exchange interaction. It is also
possible to work with a dielectric function which describes the response to the transverse electromagnetic field, and in this case the excitonic frequency must include the exchange interaction. This is
discussed in Ref. [1], 12, and in Ref. [29].

73

E(..v)

3..220

3..200

3.180

3.16O+--------L-:--------;1;l;-0----------:1iF5O::IP!77.(1-.rJrC~!"

Figure 6. Polariton dispersion for the Z3 exciton in CuCI, obtained by two-photon absorption (circles,
Ref. [43]) and hyper-Raman scattering (dots and crosses, Ref. [44]). From Ref. [34].

excitonic polaritons has been measured for many semiconductors in a variety of optical
experiments [35]. An example is shown in Fig. 6, which displays the dispersion of the
UP, LP and LE in CuCI obtained by nonlinear optical spectroscopy.
Quantum theory of polaritons. A quantum theory of polaritons consists of
setting up a second-quantized Hamiltonian which describes excitons and photons, with
their mutual interaction. The approximation of keeping only quadratic terms yields
a Hamiltonian which can be diagonalized exactly, and which (for an infinite crystal)
describes stationary states. Terms beyond the quadratic approximation, as well as
terms associated to the exciton-phonon interaction, describe the damping of polariton
states. The exciton-photon Hamiltonian can be obtained either from a microscopic
model for the exciton [45, 46] or from second quantization of Maxwell equations plus
the equation of motion for the excitonic polarization,

Ii
Mwo

1..

-P--\7P+P={3E

W5

(74)

(we consider only the case of cubic crystals). The resulting Hamiltonian is
H

1
2

1
2

= ~)livk(alak + -) + liwk(blbk + -)
k

+iCk(al

+ a_k)(bk -

b~k)

+ Dk(a! + a_k)(ak + a~k)],

(75)

where k = (k, A) is a combined index which includes wavevector and polarization vector
for the two transverse modes, at, ak (bt, bk ) are Bose operators for the photon (exciton),
v = c/...;e;;; is the speed of light in the crystal, liWk = liwo + r~!2 is the exciton energy
including spatial dispersion, and

~ (7r{3Wk )1/2
Ck -- /tWo
k
'
V

Eoo

D _ ~ 7r{3Wk
k - /tWo k .
V

(76)

Eoo

The normal modes of the quadratic Hamiltonian (75) can be found by an operator
transformation [47] which was extensively used by Hopfield [45]. New operators O'.kI, 1=

74

1,2 are introduced, which are related to the exciton and photon operators by the linear
relation

(77)

The new operators are determined by the condition that they satisfy Bose commutation
relations, and that the Hamiltonian becomes diagonal, i.e.,

(78)
This leads to the following secular equation for the eigenfrequencies Ok/:

(79)
This can be seen to be identical with the classical dispersion relation (71). The coefficients of the transformation are given e.g. in Refs. [45, 35J.
It should be remarked that the above diagonalization procedure gives the polariton
operators, but not the polariton states. In particular, the polariton vacuum defined by
the condition
(80)
does not coincide with the classical vacuum. It is shown in Ref. [48J that the conditions
(80) generate recursive equations, which allow an exact determination of the polariton
vacuum in terms of photon and exciton states. The polariton states can be obtained by
applying the polariton creation operators on the new vacuum. In general, the polariton
states are not simply given by a linear combination of one exciton and one photon states,
but contain instead additional components like three photons, two excitons plus one
photon, etc. The presence of these nonlinear components is associated to the intrinsic
squeezing of polariton states [49J. However the nonlinear components are appreciable
only for wavevectors k < f3k o, which on a frequency scale means w - WL < f32wo '"
wiT/wo. The frequency scale is very narrow, making such nonlinear and squeezing
effects difficult to observe.
Relevance of polariton concept. At this point I would like to address the following question: to which extent is the concept of polariton a relevant one, or, under
which conditions will polaritons be observed instead of the excitons? At present there
is no fully satisfactory answer to this question. However it seems clear that two criteria
can be given, namely temporal coherence and spatial coherence. These two criteria will
now be discussed, with particular reference to the case of GaAs, for which the validity
of the polariton description is a much debated question in the recent literature.
Temporal coherence in the exciton-photon coupling requires that dephasing processes are slower than the oscillation rate between exciton and photon in the polariton
state. The oscillation rate can be studied within a 2x2 model, in which only one exciton and one photon states are kept. The matrix element of the A . p interaction
(21) between exciton and photon states can be calculated by representing the vector
potential in second-quantized form,
ikr
A( r ) -- ,,(27rlic)1/2[A(A)
L. k
tk akA e
kA n V

+ h .c ..1

(81)

75

The result is
(excitonIHllphoton)

2 2
ko 1/2
= - (7r-ko-e-h-f)1/2 = -( -)
hwe,
1:00

k mo V

(S2)

where the coupling energy hWe is defined by


We =

r;l:iWo =

V;;:'

JWOWLT .
2

(S3)

Thus hWe gives the exciton-photon matrix element at the crossing point k = ko. The
polariton dispersion within this model is obtained by diagonalizing a 2 X 2 Hamiltonian.
At the crossing point, the polariton splitting is 2hwe (see Fig. 5) and (2we )-1 is the
oscillation time between exciton and photon. This simple model clearly fails for k - t 0,
where the matrix element diverges: this is a signature of the fact that the neglected
many-particle states become important. However this happens only very close to WL,
as discussed above.
This model suggests an analogy with a two-level atomic system interacting with
the radiation field of a single-mode cavity. The analog of the two atomic levels are the
crystal ground state, and the k ex = 0 exciton, which is the lowest electronic excited state
of the whole crystal. Interaction with a single mode of the radiation field is imposed by
the conservation of crystal momentum. The polariton splitting can thus be viewed as
vacuum-field Rabi splitting [50J. The Rabi frequency is given by the coupling frequency
We
The exciton-photon coupling energy hwe , which is much larger than the LT splitting,
is the relevant energy scale for the mixing between exciton and photon states. This can
also be recognized by examining the coefficients of the transformation (77). In fact, it is
shown in a paper by Hopfield [51J that the upper polariton is essentially excitonlike up
to energies Iw - Wo I :;; We (even if its dispersion relation is not). This has the important
consequence that inelastic light scattering can be observed also for frequencies such
that W - Wo ~ WLT, i.e., far from the exciton resonance [52, 32J. The LT splitting arises
from the electron-hole exchange interaction, i.e., from the instantaneous (longitudinal)
part of the electromagnetic field, and is not a polaritonic effect.
To give an example, let us take the case of GaAs, where hwo = 1.515 eV and the
LT splitting is as small as hWLT = O.OS meV [52J. The coupling energy is hWe = 7.S
meV, and the oscillation time (2we)-1 = 0.04 ps. The exciton-photon oscillation rate is
extremely fast compared to other characteristic times. For comparison, the dephasing
time in 190 nm thick GaAs layers has been measured to be T2 = 7.0 0.5 ps ~ (2wet1
[53J. The temporal behavior of polaritons is very stable. Comparing the thermal energy
to the LT splitting is not meaningful: in fact, polaritons can exist even for kBT ~ hWLT.
The correct criterion for temporal coherence is that the homogeneous broadening is
smaller than hwe.
In order for the exciton-photon coupling to be spatially coherent, the coherence
length le of the exciton wavefunction must by much larger than the wavelength of light,
otherwise the dipole matrix element between exciton and photon is reduced. In the
literature [54, 55], the requirement of spatial coherence is given in terms of an upper
bound for the homogeneous broadening of the exciton,

r < re =

)1/2
( SWLT
M c2
WOo

(S4)

This criterion can be understood as following from r- 1 vg > A, where Vg = dwk/dk is the
exciton velocity evaluated at the longitudinal frequency and r- 1 vg > VgT is the exciton
76

mean free path: in words, the exciton must not suffer scattering processes within a light
wavelength.
In general, spatial coherence provides a much more stringent criterion than temporal coherence. For example, for GaAs, re rv 0.1 meV. However many points are
still unanswered. It is not clear whether the coherence length le can always be identified with the mean free path. Moreover, the criterion (84) is probably not the whole
story. There is no reason why the exciton group velocity should be evaluated at the
longitudinal frequency. Intuitively, excitons at large wavevectors are less likely to be
scattered. It is conceivable that the coherence length of the exciton center-of-mass
motion is itself energy-dependent, and that spatial coherence is satisfied only in part
of the energy region of the polariton dispersion. Finally, the theory of Ref. [54] suggests that the requirements of temporal and spatial coherence do not have to be satisfied simultaneously: rather, temporal coherence would refer to an experiment with
fixed wavevector and imaginary frequency (which requires that the crystal is uniformly
excited), whereas spatial coherence would refer to an experiment performed at fixed
frequency with spatially-decaying fields. Thus the conditions for the validity of the
polariton description would depend on the experimental conditions.
From the experimental side, there is well-established evidence that excitonic polaritons are formed in good GaAs crystals up to 20 K. This conclusion follows from the
analysis of reflectivity [56], resonant light scattering [52, 32], time-of-flight [57], and
luminescence experiments [56,58,59]. In particular, the observation of Brillouin peaks
far from the exciton resonance and the spectacular decrease of the group velocity at
the transverse exciton frequency give direct evidence for the validity of the polariton
description. Still, the fact that the group velocity does not decrease down to the minimum theoretical value might indicate a partial, energy-dependent breakdown of the
polaritonic effect. Also, the existence of polaritons in GaAs is very sensitive to sample
quality.
2.5

Radiative Lifetime of Exciton-Polaritons

Optical absorption and radiative lifetime in the polariton framework. In the


preceding Section it has been shown that polaritons are the eigenstates of an infinite
crystal that result from the interaction of excitons with the radiation field. In the
absence of dissipative couplings, polaritons are stationary states. Thus the excitonradiation interaction alone does not give rise to optical absorption [45]. An incident
photon converts into a polariton at the crystal surface, which means that the energy
in the crystal oscillates back and forth between exciton and photon states. Absorption is determined by phonons or crystal defects, which scatter the polariton via its
exciton component [45]. Absorption is often treated by introducing a broadening r
in the dielectric function (69). The broadening parameter turns out to be strongly
energy-dependent [60]. There is also a more fundamental problem in this approach:
a broadening r appearing in (69) represents a decay rate of the exciton, whereas the
meaningful decay rate should be that of the polariton. As a further complication, absorption depends strongly on the ABC, reflecting the different exciton components of
the two polariton branches. To my knowledge, the problem of describing the absorption
lineshape within the polariton framework is still not completely solved.
The exciton-radiation interaction alone does not give rise to a radiative decay of
the exciton in an infinite crystal. Once again, the reason lies in the translational
invariance: because of conservation of crystal momentum, an exciton with a given
wavevector and polarization interacts with only one photon with the same wavevector
and polarization (neglecting Umklapp processes). Thus there is no density of states for

77

the decay, buth rather a coupling between two discrete states, which gives rise to the
stationary polariton states [45]. Within the polariton framework, radiative decay offree
excitons can only occur through conversion of polaritons into photons at the surface of
the sample. A polariton with a given wavevector propagates up to the crystal surface,
where it has a finite probability of being transmitted. Therefore, the "radiative decay
rate" of a polariton with well-defined wavevector must have a form like
(85)
where R is the reflection coefficient at the surface, and Vg is the polariton group velocity.
The radiative lifetime grows with the crystal size L, and represent the confinement time
of the polariton in the sample.
Polariton luminescence is actually a very complex phenomenon, since it results from
the interplay between polariton propagation, damping, and thermalization processes.
Thermalization within the LP branch is very efficient down to the knee of the polariton
disperion, just below the transverse frequency, due to the high density of states. Below
the transverse frequency, the decrease of the density of states and the large increase
of the group velocity cause the radiative lifetime to dominate over thermal relaxation.
This leads to the "polariton relaxation bottleneck" [61]: the polariton distribution
under steady-state excitation has a pronounced peak just below the exciton energy.
The observed luminescence for nonresonant excitation has two peaks, associated to
luminescence from LP and UP, as was observed in CdS [62] and GaAs [56]. The lifetime
of the luminescence is calculated to be in the ns range [58, 63], but it depends very
much on sample size and shape, as well as on excitation conditions. With resonant
excitation, a "cold", spatially nonuniform polariton distribution is created, leading to
a strongly enhanced radiative efficiency and to a single-peak luminescence with no
reabsorption dip at WL [59]. Although the luminescence lineshape and the radiative
lifetime depend on the detailed experimental conditions, the existing evidence strongly
suggests that the polariton picture is the relevant one for the interpretation of lowtemperature luminescence experiments in pure GaAs and CdS crystals.
Radiative lifetime of bound excitons. For excitons bound to impurities, the
situation is quite different since crystal momentum is not conserved. Thus there is a
density of states for radiative decay, given by the usual three-dimensional density of
states of photons. The radiative lifetime of bound excitons is described by the atomiclike formula [64]
3moc3 1
- 2ne 2 w 2 i'

(86)

r-----

and is inversely proportional to the oscillator strength. Excitons bound to impurities


are characterized by the so-called giant oscillator strength [65, 66]. Assuming that the
internal electron-hole wavefunction is undistorted, and that the wavefunction of the
center-of-mass motion is that of a bound state with a radius 1,

(87)
the oscillator strength is calculated to be
.f

78

2 1(U I
I ) 121
= --lic E' P Uy
mo w

JF ( )dR
R,O

12 = 871"1 3 -ifree ~ mo 871"(


1)3'
V
m* 71"

as

(88)

In CdS, the confinement length of excitons bound to impurities is 1 '" 15 A, which leads
to an oscillator strength of order unity. The corresponding lifetime is T '" 0.5 - 1 ns,
as observed experimentally [66]. The "giant oscillator strength" explains the fact that
bound excitons are easily observed in absorption, even for low impurity concentration.
Equation (88) is usually interpreted by saying that the oscillator strength of the
bound exciton is enhanced over the oscillator strength per molecule of the free exciton by
a factor 871'l3/n. This is the origin of the term "giant oscillator strength". This concept
must be used with caution: for Wannier excitons, since the electron-hole separation is
much larger than the lattice spacing, the "oscillator strength per molecule" cannot be
related to a property of the constituents of the crystal. The dimensionless oscillator
strength is proportional to the crystal volume, and is infinite in the thermodynamic
limit. Using an oscillator strength per molecule might suggest that a formula like
(86) could be used to calculate the radiative lifetime of a free exciton. This would
be incorrect. For a free exciton, when the Bloch vector is conserved, coupling to the
radiation field does not give rise to a radiative decay but rather to stationary polariton
states, as we have seen.

"Polariton" versus "coherence volume" lifetime. The exciton radiative lifetime is well understood, at least in principle, in the two cases of an exciton bound to
impurities, or when the exciton is free and exciton-polaritons are formed. No theory
exists which describes the radiative decay when the exciton is free, but the coherence
length is smaller than the wavelength of light and the polariton picture breaks down.
In a paper by 'tHooft et al. [67] the free-exciton lifetime is measured in ultrapure
MBE-grown GaAs crystals. The authors claim that the polariton concept is irrelevant
for their experiment, and interpret the results using Eq. (86) for the lifetime, which was
derived only for bound excitons. The temperature-dependence of the lifetime is interpreted as reflecting the density of states of the free exciton. In view of the experiments
discussed above [56, 59], the results of 'tHooft et al. might indicate that their sample
was of worse quality compared to samples used in previous works. Since the thickness
of their MBE-grown sample (1.5 f-Lm) was small compared to the thicknesses used in
previous works [56,59] (20 - 60 f-Lm), the breakdown of the polariton picture might be
due to the more important role played by interface scattering.
What is the mechanism for radiative decay of the free exciton when the coherence
length of the center-of-mass motion is smaller than the wavelength of light? It is
conceivable that in this case the lifetime is given by a formula similar to (86), with
the effective oscillator strength being proportional to the coherence volume Vc '" le 3
of the exciton center-of-mass motion. The lifetime for temperatures higher than the
homogeneous linewidth would increase like T 3 / 2 , reflecting the density of states of the
free exciton. A continuous transition might occur between the "polariton" regime le ~
A, when the lifetime is determined by polariton propagation and relaxation, and the
"coherence-volume" regime le ~ A, when the lifetime is given by a formula similar to
that for bound excitons. Which of the two pictures applies to a specific crystal depends
on the quality of the sample, on the nature of the interfaces, and on temperature.
However, in the lack of a complete theory, these considerations about a "coherencevolume" regime are purely speculative.

79

3
3.1

OPTICAL PROPERTIES IN CONFINED SYSTEMS


Interband and Intraband Transitions in Quantum Wells

In this Section we study optical absorption in single and multiple quantum-well systems.
This is an aspect of the more general problem, which is only partly solved, of calculating
the dielectric tensor of superlattices including spatial dispersion and local-field effects.
We refer to [68, 69] for recent work in this direction.
Electronic levels. The basic physical phenomenon taking place in semiconductor
heterostructures is the (partial) discretization of electronic levels. This is brought about
by the variation of the energy gap from one material to another. The band-edge profile
acts as an additional potential for the carriers, which can result in confinement of the
carriers along one or more directions. In these lecture notes I shall mainly be concerned
with optical properties in an isolated type-I Quantum Well (QW), where the electron
and the hole are confined in the same region of space.
Electronic levels in heterostructures are often described within the envelope-function
theory [70, 71]. In this approach the electronic wavefunction for a non degenerate band
(with extremum at k = 0) is expressed as
(89)
where ujo(r) is the Bloch function at the band edge and f(r) is an envelope function
which is assumed to be slowly varying on the scale of the lattice spacing. The envelope
function satisfies an effective-mass equation,

;,,2\1 2 ;,,2 8
1 8
[_II - _____
2m*
28zm*(z)8z

+ V(z)]f(r) =

(E - En(O))f(r),

(90)

where m*(z) is a position-dependent effective mass, and V(z) is the band-edge profile
plus any slowly-varying external potential. The envelope function satisfies continuity
and current-conserving boundary conditions at each interface.
Since the lowest conduction band is nondegenerate, quantum confinement of the
conduction levels is described by a simple particle-in-a-box problem. Complications
come from nonparabolicity of the bulk conduction band. The correct treatment of
nonparabolicity is a tricky problem. The effect of bulk nonparabolicity on the in-plane
effective mass is found to be about three times larger than on the longitudinal effective
mass. We refer to [72, 73] for details.
The dispersion of valence sub bands in quantum wells is more complicated due to the
degeneracy ofthe bulk valence-band edge (Fig. 2). Two series of levels arise, heavy holes
(HH) and light holes (LH), which have different confinement energies. In order to treat
the valence-band dispersion, the envelope-function approach must be generalized to the
degenerate-band case [70, 71]. HH and LH are coupled at finite in-plane wavevectors
due to off-diagonal terms of the Luttinger Hamiltonian [16]. The dispersion of valence
subbands in GaAs/ AIGaAs QWs is illustrated in Fig. 7 (see also [74]).
Absorption probability and absorption coefficient. Consider optical absorption from a single quantum well (SQW). For light propagating along the growth direction, and when the photon wavelength is much larger than the well width, the absorption
of a light beam cannot be described by the usual exponential decay: rather, each photon has a finite probability of being absorbed by the entire quantum well. Hence the

80

0
HH1

-30

LH1

Ql

Ql

3.0 kbor

LH1

s:3
:>,
bD

HH1

HH2

-90

HH3

I:i
Ql

LH2

(0)

-1500L---L-~1L-~--~2--~

(b)

k L

2
k L

Figure 7. Dispersion of the valence subbands in a 80 A wide GaAs-Gao.6Alo.4As quantum well (a)
with no applied stress (b) with an applied stress X = 3.0 kbar along the growth direction.

absorption probability for a single quantum well is a pure number, which can be defined
as
w

( ) _ energy absorbed/unit time & surface


w - incident energy/unit time & surface'

(91)

The absorption probability can be calculated by time-dependent perturbation theory,


as in Sec. 2.2. The result is
(92)
where kll is the in-plane wavevector. Note that w(w) is dimensionless, while the absorption coefficient (30) has the dimensions of em -1.
It is important to realize that the absorption coefficient (29) of a plane wave for
a sample containing a single quantum well is zero in the thermodynamic limit. This
implies that a plane wave propagating along the layer planes is not attenuated. There
is no contradiction with experiment, since the case of an infinite plane wave is never
realized in practice. In the real case of an incident light beam with diameter C, the
absorption per unit length is of the order of w( w) / C (apart from a factor of order
unity which depends on the electric field profile). Thus the attenuation of a light beam
propagating in the layer planes can be increased by decreasing the spot diameter. A
finite absorption of order w(w)/C is also obtained for light propagating in a waveguide
configuration, and in this case the relevant length C is the thickness of the waveguide
[75]. This is to be contrasted with the case of light propagating along the growth
direction, where the dimensionless absorption probability (91) is independent of the
spot size.

81

Multiple quantum wells (MQWs) are usually employed in order to increase the
optical efficiency. For a sample with N periods, the transmittance T is related to the
dimensionless absorption probability (91) by

= (1 -

wt

= exp(N .log(1 -

w)) ~ exp( -Nw),

(93)

and the usual formula for exponential decay is recovered. Macroscopically, the absorption coefficient per unit length is obtained as
a

= --log( 1 - w)
D

Nw
D

w
Lw

+ Lb

(94)

where Lw + Lb is the MQW period and D = N(Lw + L b) is the thickness of the sample.
This derivation shows that the proper length by which the dimensionless absorption
probability has to be divided is the period Lw + Lb (not the well width Lw).
A MQW sample can also be considered as a uniaxial crystal with a unit cell of size
Lw + Lb along the growth direction. The macroscopic absorption coefficient can again
be defined as in Eq. (29), and can be calculated by the procedure of Sec. 2.2. When
the electronic states are localized along the growth direction, the analog of Eq. (28) is

~ :E :E 2 :E =
V

cv

II

:E:E r

V.
J

cv

JBZ

2dkll

(27r)2

Lw

+ Lb

:E r
cv

JBZ

2dk ll

(27r)2'

(95)

where E j denotes the sum over the identical QWs and S is the area of the sample.
Thus we see that formula (94) gives the absorption coefficient of a MQW sample for
any direction of propagation. The physical meaning of Eq. (94) is that the attenuation
of a light beam in a MQW sample is proportional to the concentration of absorbing
regions, and therefore must depend on both well and barrier widths. A limiting case of
Eq. (94) is the case of a single quantum well, when the barrier width Lb --t 00 and the
macroscopic absorption coefficient vanishes.
A description of optical absorption in quantum well structures which is based on
the absorption probability has two advantages: first, the absorption probability for
interband transition is essentially independent of the well width (see below), without
the artificial 1/ Lw dependence which is sometimes reported in the literature as a result
of dividing the optical density by the well width [76J. Second, the absorption probability
integrated over the excitonic peak is directly related to the oscillator strength per unit
area, which is the basic quantity characterizing quasi-two-dimensional excitons (see Sec.
3.2).
Interband transitions. For transitions between valence and conduction subbands,
the momentum matrix element can be evaluated by keeping only the lowest-order term
in Eq. (89), and by neglecting the gradient of the slowly-varying envelope function. The
result is

(96)
Selection rules come from the envelope-function as well as from the Bloch-function part.
In the approximation of infinite barrier height, the envelope functions of valence and
conduction sub bands are identical, which leads to the selection rule D.n = 0 [77J. This is
only approximately true for a well of finite barrier height, since particles with different
effective masses have different penetration in the barriers. In any case, transitions with
D.n = 1,3, ... are forbidden by parity. Selection rules for the matrix element between
Bloch functions can be found using the Wigner-Eckart theorem [70J, or, in a more

82

elementary way, by observing that heavy holes at kll = 0 have an angular momentum
s = 3/2, while light holes and electrons have s = 1/2. For light polarized in the
layer planes, both heavy- and light-hole transitions are allowed. However, for light
polarized along the growth direction, the operator f. . p does not change the angular
momentum and only light-hole transitions are allowed. The ratios of oscillator strengths
for interband transitions at kll = 0 are f~~l : f~"'J) : frJ) = 4 : 3 : 1.
Equation (96) shows that the interband momentum matrix element between Bloch
functions is essentially the same for bulk and QW systems. Therefore, the oscillator
strength (34) is also the same. Differences in the optical absorption come from the joint
density of states. In an ideal two-dimensional system, the joint density of states has a
staircase form [77].
Absorption of a single quantum well is low, being of the order of a percent. This can
be easily estimated from Eqs. (92) and (96), which yield (for the heavy-hole transition in
the case of infinite barrier heights) w ~ (1Ie 2/( nne) )(fl,fm:). The absorption probability
is to a first approximation independent of the well thickness, since the two-dimensional
density of states P2D = p,/( 1In 2) is a constant. This has been verified experimentally [78].
Corrections to this result can come from leaking of the wavefunction in the barriers,
which reduces the overlap integral, and from the slight variation of the Sommerfeld
factor with the well width (see below).
The quantitative evaluation of interband absorption in QWs is made complicated
by valence band mixing. At kll -I- 0, heavy and light hole states become mixed and
transitions between all pairs of conduction and valence subbands are possible. The
ones which are forbidden at kll = 0 acquire a larger oscillator strength at those values
of kll for which the corresponding valence subband is strongly mixed with the subband of an allowed transition. The negative curvature of a subband also increases the
absorption, because it gives a large joint density of states. An example of calculated
interband absorption for transitions to the first conduction subband is shown in Fig. 8.
A phenomenological width r = 2 me V has been introduced. The peak at the energy
of the LHI-CBl transition comes from the negative curvature of the LHI subband (see
Fig. 7).
It must be stressed that the interband absorption shown in Fig. 8 is not a measurable
quantity, since the measured absorption contains excitonic effects. The Sommerfeld
factor in the strict two-dimensional limit has been calculated by Shinada and Sugano
[79], with the result

S(w) ___
2_

- 1 + e- rrx '

(97)

This factor is two at the subband edge, and decreases slowly with an energy scale given
by the effective Rydberg. For realistic quantum well excitons, the enhancement factor
is calculated to be between 1.3 and 1.4 at the absorption edge [80]. Including the
excitonic effect, the absorption probability in a GaAs-Gal_xAlxAs single quantum well
is w ~ 0.75% for the first HH transition, and w ~ 1.0% for the lowest HH and LH
transitions together (see Fig. 8 and Ref. [78]).
A schematic plot of the absorption in two-dimensional systems is shown in Fig.
9. As for three-dimensional excitons, the absorption probability corrected for excitonic
effects is continuous at the interband absorption edge [79]. Recently, analytical formulas
for the absorption lineshape in quantum wells including bound and continuum exciton
states have been developed [81, 82], which are based on the idea that the excitonic
spectrum has a dimensionality intermediate between 2D and 3D.

83

1.5
LH1
-I-

HH3

HH2
-I-

-I-

LH2
t

'"I 0

1.0

'-'

II

II

:.;:;
Q..
L-

a(f)

Ll
0

0.5

80

130

180

E - Egop (meV)
Figure 8. Interband absorption probability in a 80
excitonic effects.

n=l

n=2

n=3

A wide

GaAs-Gao.6Alo.4As quantum well, without

~
---------------

E
Figure 9. Schematic picture of the absorption probability in two-dimensional systems, without excitonic effects (dashed line) and with excitonic effects (solid line).

84

Intraband transitions. In bulk crystals, intraband transitions can only occur in


the presence of some scattering mechanism (see Sec. 2.2). In quantum-confined structures, intraband (or intersubband) transitions are made possible by the discretization
of the energy levels. Such transitions can only be observed in doped or photoexcited
samples [70, 83, 84].
The dipole matrix element for an intersubband transition between states of the
form (89) describes what is called an envelope-state transition. The term containing
the derivative of the Bloch function now gives no contribution, since the momentum
matrix element between Bloch functions of either the conduction or the valence band
vanishes. The term with the gradient of the envelope function is of the same order as
that coming from the correction to the lowest-order wavefunction on the r.h.s. of Eq.
(89). This second term in fact dominates, and has the effect of multiplying the matrix
element by maim:. The result is
(98)
For light polarized along the growth direction, transitions are allowed only between
conduction (or valence) states of opposite parities. For the transition between the
two lowest conduction levels, the matrix element in the case of infinite barriers is
(8/3)( -in/ Lw). The oscillator strength (34) for the lowest intersubband transition
is then evaluated as

fz

256 ma ~ 0.96 ma ,
271r2 m;
m;

(99)

i.e., it is nearly identical to that for interband transitions [85, 86]. However the joint
density of states is quite different: for parabolic subbands with the same effective mass,
the transition energy would be the same at all wavevectors, resulting in a 5-function
line where all the oscillator strength is concentrated. Line broadening comes from
conduction band nonparabolicity, as well as from many-body effects [87]. Assuming a
Lorentzian broadening, the intersubband absorption probability can be calculated from
(92) and (98), with the result
(100)
where ns is the areal density of electrons in the first conduction subband. Intersubband
absorption is smaller than interband absorption by a factor rv fi 2 n s /(m:,) [70,86].
When the transition matrix element is evaluated in the gauge -eE r, the so-called
"giant electric dipole" is found [85]. Of course the physical results are independent of
the choice of the gauge, and in fact the same expression (99) for the oscillator strength is
found in the two gauges. The physical meaning of the "giant electric dipole" is that an
oscillator strength comparable to that of inter band transitions is found in the infrared,
i.e., at a much lower energy as compared to interband transitions.
For light polarized in the planes, intersubband transitions become second-class (i.e.,
the matrix element is linear in the wavevector). In the valence band they become
allowed due to valence band mixing. A theoretical analysis of intersubband transitions
in the valence band has been given in Ref. [88].

3.2

Excitons in Confined Systems

Thin-film and QW regimes: overview. In this Section we discuss a few aspects


of the theory of excitons in thin layers [89, 90]. Two regimes can be distinguished,

85

according to the relative values of the well width L and the exciton radius aBo In the
limit L ~ ai3, which we shall call the thin-film regime, the excitonic Rydberg R* is
much larger than the quantization energy 1i 2/(m* L2), and the exciton is only weakly
perturbed by confinement. The internal electron-hole wavefunction is undistorted, but
the motion of the center of mass of the exciton is quantized [91, 92]. In the limit
L rv ai3 (which we call the quantum-well regime), the excitonic Rydberg is smaller than
the quantization energy of the subbands, and separate quantization of the electron
and hole subbands occurs. The distortion of the internal exciton wavefunction due to
a decrease of the average electron-hole separation leads to an increase of the binding
energy and of the oscillator strength per unit area as the well width is reduced [93,94].
Excitons in semiconductor heterostructures are usually described within the envelopefunction scheme, with the electron and hole confinement potentials being added to the
effective-mass Hamiltonian. Due to translational invariance in the layer planes, the exciton is characterized by an envelope function F(p, Ze, Zh), where p is the in-plane relative
coordinate. A suitable description of the exciton in the thin-film regime (neglecting the
valence-band degeneracy and in the assumption of infinite barriers) is provided by the
variational wavefunction of D'Andrea and Del Sole [92],

F(p,Ze,Zh)

= N[cos(KZ) - Fc(z)cosh(PZ) + Fo(z)sinh(PZ)]e- r / a ,

(101)

where r is the electron-hole relative coordinate, Z.is the center-of-mass coordinate along
the growth direction, N is a normalization factor, and P,a are variational parameters.
The functions Fe(z), Fo(z) are determined by the fulfillment of the no-escape boundary
conditions F(ze = ~) = F(Zh = ~) = o. The requirement that the wavefunction
has a continuous derivative leads to the following quantization condition for the centerof-mass wavevector K:

KL
KtanT

PL

+ Ptanh T

o.

(102)

The quantity 1/ P can be interpreted as a dead-layer thickness [40], and is of the order
of the exciton radius. For P L ~ 1, the quantization condition reduces to the center-ofmass quantization K = mr /(L - 2/ P). The exciton levels are those of a particle of mass
M in a box of width Lcff = L - 2/ P. For thin layers, the center-of-mass quantization is
only approximately true, and the full quantization condition (102) must be used. The
transition-layer thickness 1/ P decreases with the well width.
In the quantum-well regime, the exciton wavefunction is more suitably represented
in a basis consisting of products of conduction and valence subbands,

F(p, Ze, Zh) =

L Ajjle-UIPCj(Ze)Vj{Zh).

(103)

ijl

For very narrow wells, only one pair of valence/conduction subbands needs to kept. In
the limit L ~ 0, if the barriers are taken of infinite height, the binding energy tends
to the value appropriate to the two-dimensional Coulomb problem, R;D = 4R;D [79].
With finite barrier height, the binding energy reaches a maximum and tends to the
value appropriate to the barrier material as the well width tends to zero [95]. For wide
wells, several pairs of subbands must be kept in the expansion (103). The wavefunctions
(101) and (103) have been shown to match for L rv 3ai3 [96].
A quantitative determination of the exciton binding energy in quantum wells is
complicated by the interplay of several effects [73]. In real structures, the finite height
of the barriers and the variation of the band parameters from one material to the other
must be taken into account. Due to the degeneracy of the valence band, the kinetic

86

\t

100

20

HHI-CBI (Is)
18

:>

80

"e

-;;:: 14
\:!'

--x=O.2S
\

~
~60

16

/HHI-CBI (Is)

(b)

,,

- - - - x=OAO

co

,,

~
5

5 12

Iil

40

lS

10

c 20

8
6~~~~~-L~-L~-L~-L~~

50

100

150

O~~~~~~~~~~-L~~~

200

well widUl (A)

50

100

150

200

well width (A)

Figure 10. (a) Binding energy of the ground-state HHI-CBl exciton and (b) oscillator strengths
per unit area for in-plane polarization of the ground-state HHI-CBl and LHI-CBl excitons in
GaAs/Al"Gal_"As quantum wells. From Ref. [73].

energy of the holes must be described by the Luttinger Hamiltonian: thus the effectivemass Hamiltonian must be taken to be

H , = Ec(-iVe)5 , + T /(-iv\) -

Ie
Eb re -

rh

15 , + (Ve(Ze)

+ Vh(Zh))5 ,.

(104)

When the off-diagonal terms of the Luttinger Hamiltonian are neglected, heavy- and
light-hole excitons become uncoupled; however, coupling between the two series has an
important effect on the binding energies and on the oscillator strengths. Other effects of
a comparable size are nonparabolicity of the bulk conduction band and the difference in
dielectric constants between well and barrier materials. All the above effects go in the
direction of increasing the binding energy, and taken together result in very high binding
energies (see Fig. lOa), which in GaAs/ AlAs quantum wells can be even higher than the
two-dimensional limit [73]. This prediction has recently been confirmed experimentally

[81].
The combined effect of nonparabolicity and the dielectric mismatch results in a
decrease of the critical well width, below which the exciton binding energy tends to the
value appropriate to the barrier material. However a quantitative calculation of this
effect is still lacking. A similar problem arises in short-period superlattices: when the
superlattice period decreses below a critical value, the exciton wavefunction spreads out
and extends over several layers [97]. However the maximum value of the binding energy
is likely to be strongly increased by nonparabolicity and the dielectric mismatch. In
general, the increase of the binding energy due to the smaller dielectric constant of the
barrier (often called dielectric confinement [98]) is of great interest as a way to increase
the stability of the exciton against thermal dissociation.
Oscillator strength. The oscillator strength is defined, as usual, by Eq. (34).
For excitons in thin layers, the oscillator strength is proportional to the area S of the

87

10

<i'
.<t;

CuC!

";>
0

~
c
~
'"

cJ

'"0
2
0

20

40

60

80

100

well width (A)

Figure 11. Oscillator strength per unit area in CuCl quantum wells, with two types of variational
wavefunctions. From Ref. [96].

sample, and is given by

= 9

:fi) (Uv\E . p\ucW\

F(r = 0, Z) dZ\Z,

(l05)

where 9 is the spin-orbit factor.


In the thin-film limit, where the wavefunction (101) applies, the oscillator strength
per unit area is proportional to the effective film thickness Leff = L - 2/ P. This is due
to the fact that the center-of-mass wavefunction is coherent over the entire well. From
another point of view, the oscillator strength per unit volume must be constant and
close to the bulk value for thick films, so that the oscillator strength per unit area must
be proportional to the film thickness.
In the quantum-well limit, taking the separable wavefunction F = e-p/ac(Ze)V(Zh),
the oscillator strength per unit area is

(106)
and increases as the well width is reduced due to the decrease of the radius a.
Because of the above limiting forms, the oscillator strength per unit area must have
a minimum at some well width. Quantitative calculations have been carried out for
CuCI quantum wells, where the Bohr radius is ai3 = 6.6 A and the minimum of the
oscillator strength is found at L rv 3 ai3 [96], as shown in Fig. 11. The minimum of the
oscillator strength has been observed in CdTe/CdZnTe quantum wells [99]. For GaAs
quantum wells, the minimum is expected to occur around L rv 400 A.
The oscillator strength per unit area is related to the absorption probability (91) by

w(w) d(fiw)

27r2fie2 f
27r 2 f
-2
= ---s = --S(50 .10- 5 A )2.8.10- 2 meV,

nmoc

(107)

where the oscillator strength per unit area is measured in 50 . 10- 5 A-2. This is in fact
the order of magnitude of f / S for allowed excitonic transitions in quantum wells. An

88

accurate calculation of the oscillator strength requires theories which take into account
valence band mixing and coupling between excitons belonging to different subbands
[100, 73]. For quantum wells of medium width, the two-dimensional approximation
overestimates the oscillator strength by a factor between three and four. The oscillator
strengths of the ground-state excitons in GaAs-Gal_xAlxAs quantum wells are shown
in Fig. lOb, and are in good agreement with absorption and reflectivity measurements.
Excitons in QWs: symmetry properties and selection rules. Due to valence
band mixing, heavy- and light-hole subbands in quantum wells are coupled for finite
values of the in-plane wavevector. Since exciton states are built up from band states
with wavevectors k rv lias, valence band mixing can be expected to playa role on the
selection rules for excitonic transitions. Selection rules for excitons in quantum wells
have been derived in a model-independent way using symmetry arguments [101, 102].
Here we give just a few relevant examples.
The point group Td of the zincblende structure becomes D2d in quantum wells, due
to the reduction of symmetry. The lowest heavy-hole and conduction subbandshave r6
symmetry at kll=O, while the lowest light-hole subband has r7 symmetry. The symmetry of exciton states is given by the product decomposition of (61). For ground-state
excitons, the envelope function transforms according to the identity representation.
Thus the ground-state HH1-CB1 exciton transforms as r6 r6 = r 1E9 r 2E9 r s, while
the ground-state LH1-CB1 exciton has multiplicity r6 r7 = r3 E9 r 4 E9 rs. The
representations r 1 ,r 2 ,r3,r 4 are nondegenerate, while rs is twofold degenerate. Since
the z-component of the dipole operator has r 4 symmetry, while the x, y components
transform according to the rs representation, we see that for in-plane polarization both
heavy- and light-hole excitons are allowed, while for light polarized along the growth
direction only the light-hole exciton is allowed. It can also be shown that the oscillator
strength for the z-polarized LH exciton is four times that for in-plane polarization.
Although the zincblende lattice is not invariant under space inversion, effects related to the violation of inversion symmetry are small and difficult to observe in III-V
semiconductors. Therefore, parity selection rules hold for excitons in quantum wells to
a very good approximation. For example, excitons like LH1-CB2 or HH2-CB1 which
belong to conduction and valence sub bands of opposite parities are forbidden in the
s-like ground state. Such excitons can be observed only in p-like excited states. Parity
symmetry is broken in the presence of an electric field (applied or built-in). We refer
to [101, 102, 73] for details. Parity selection rules for two-photon transitions are opposite to those for one-photon transitions, thereby allowing very useful complementary
information to be obtained by two-photon spectroscopy [102].
The above selection rules maintain their validity also when valence band mixing
is taken into account. In fact, selection rules for ground-state excitonic transitions
are identical with those for the subbands at kll=O, since excitons at kcx=O transform
according to irreducible representations of the crystal point group. The role of valence
band mixing is to give a finite oscillator strength to some excitons not in s-states, like
LH1-CB2 (2p). Also, valence band mixing changes the ratio of the oscillator strengths
of the in-plane polarized ground-state HH1-CB1 and LH1-CB1 excitons from the value
3:1 (characteristic of the subbands at kll=O) to about two [73].
Excitons in QWs: exchange interaction. The electron-hole exchange interaction splits the exciton states corresponding to different irreducible representations of
the point group. It is proportional to the singlet component of a given exciton state,
via the spin-orbit factor g. The spin-orbit factors for QW excitons are: for the r5

89

14

(1)

;r----/
/
/
/

15

(2)

/,~-----

x.y

/ /
//
/1

LHI-CBl

""

r:;*

II

3 _(1_)_ _
_ _7_X_I6_(4_)--1.L ____r::_

15
HHI-CBl

It x I6

(2)

x.y

(4)

Figure 12. Schematic representation of the effect of the exchange interaction on the ground-state
excitons in quantum wells. Numbers in parentheses denote the degeneracy of the state.

representation, 9 = 1 and 9 = 1/3 for HH and LH excitons, respectively; for the r 4


representation, 9 = 4/3 for the light-hole exciton. All other states have 9 = 0 and
no exchange contribution to the energy. The exchange splittings of the ground-state
heavy- and light-hole excitons in QW s are illustrated in Fig. 12.
In bulk semiconductors, the dipolar part of the exchange interaction is nonanalytical
for k ex --) 0 and gives rise to a splitting between longitudinal and transverse states with
the symmetry of the dipole (see Sec. 2.3). For an isolated quantum well, we use the
following terminology: taking the exciton wavevector k ex = kxx along the x-axis, we
denote by T-mode the rs exciton with polarization vector E II ii, we call L-mode the
other rs state with polarization E II x, and we call Z-mode the r 4 exciton with Ell z. The
Z-mode exists only for the light-hole exciton. Thus the longitudinal-transverse splitting
is defined to be the exchange splitting between Land T modes with rs symmetry.
The exchange interaction for QW excitons has been calculated in Ref. [26] with
the Wannier-function formalism, and in Ref. [103] with the k-space methods. The
two procedures yield essentially the same results. In Fig. 13 we show the long-range
part of the exchange interaction for the light-hole exciton as a function of the in-plane
exciton wavevector k ex It can be seen that the longitudinal-transverse splitting vanishes
linearly in k ex for k ex --) O. This is an effect of reduced dimensionality, and it can be
understood from the fact that the LT splitting at k ex = 0 is proportional to the oscillator
strength per unit volume (see Eq. (68)): for a single QW, the dimensionless oscillator
strength is proportional to the area and the oscillator strength per unit volume vanishes
[26]. The situation should be different in the case of multiple quantum wells, since the
oscillator strength per unit volume is finite and a nonvanishing LT splitting between
rs excitons at kex=O is expected to occur [104].
It can be seen from Fig. 13 that the z-polarized (r4) light-hole exciton at k ex = 0 is
shifted upwards with respect to the rs exciton by an energy of the order of a meV. Such
a splitting, which is expected from symmetry arguments, arises from the depolarization
field for a polarization perpendicular to the interfaces. The ZT splitting of quantum well

90

x=O.4
LH exciton

1.0

r
>.

0.5
L

0.0

10

15

kL

Figure 13. Long-range exchange energies of the L, T and Z modes for the ground-state LHI-CBl
excitons in a 60 A wide GaAs/Alo.4Gao.6As quantum well. From Ref. [26].
excitons has been observed experimentally [105, 106]. Other features of the k-dependent
exchange interaction are much more difficult to observe, since they are overwhelmed by
spatial dispersion.
It has already been mentioned in Sec. 2.3 that taking into account the nonanalytic
part of the exchange interaction corresponds to including the depolarization field in the
dielectric response, i.e, to solve Maxwell equations in the instantaneous limit [1, 29].
Thus it is no surprise that the dispersion of the exciton taking into account the kdependent exchange interaction coincides with that found from the solution of the
electrostatic equations [107].
3.3

Polaritons in Confined Systems

Division into non radiative and radiative modes. Polariton effects are defined to
be the effects coming from the interaction between excitons and the retarded (transverse) part of the electromagnetic field. There is a fundamental difference between
polariton effects in bulk and in confined systems (see Fig. 14). In bulk crystals, due to
the conservation of crystal momentum, an exciton with a given wavevector k ex interacts
with only one photon with the same wavevector and polarization: there is no density
of states for radiative decay, and the mutual interaction gives rise to the stationary polariton states. In quasi-two-dimensional systems, due to the breaking of translational
invariance along the growth direction, an exciton with in-plane wavevector kll interacts
with photons with the same in-plane wavevector but with all possible values of k z Thus
there is a density of states for radiative decay, given by
~Vn2/iw

p(kll'w) == 'L,o(/iw - Vk~


kz

+ kn = sCi:)
7r

~B(ko - kll),

(108)

V ko - kll

where n is the index of refraction, ko = nw/ c is the wavevector of light in the sample,
and B(a::) is the Heavyside function (B(a::) = 1 for a:: > 0, B(a::) = 0 for a:: < 0). Only

91

bulk

exciton

confined systems

photon

exciton

photons

Figure 14. Schematic picture of the nature of the exciton-photon interaction. In the bulk, the interaction of the exciton with a single photon mode leads to stationary polariton states and to a splitting
of the eigenmodes. In low-dimensional systems, the interaction of the exciton with a continuum of
photon modes leads to a radiative decay of the exciton.

excitons with kll < ko decay radiatively. Thus the interaction of QW excitons with the
radiation field gives rise to two kinds of states [108, 109]:
1. States with kll > ko. They do not decay radiatively, and in this sense they
represent the analog of bulk polaritons. For these states, the electric field far from the
well is exponentially damped. These are thus nonradiative modes (analog to surface
modes) which lie on the right of the photon line in the kll-w plane, and do not couple
to incident light propagating along the growth direction.
2. States with kll < ko. These states have a finite radiative lifetime. They lie on the
left of the photon line in the kU-w plane, and have an oscillatory electric field far from
the well. These are the states which are observed in usual optical experiments.
Semiclassical theory with nonlocal response. Polariton effects can be treated
at two different levels. In a semiclassical theory, one looks for the solutions of Maxwell
equations together with a constitutive relation which accounts for the excitonic resonance. In a full quantum-mechanical theory, one has to diagonalize a Hamiltonian which
describes the exciton in interaction with the second-quantized electromagnetic field. As
shown in Sec. 2.4, both procedures yield the same dispersion relation for polaritons in
bulk semiconductors. The equivalence between semiclassical and quantum-mechanical
approaches is due to the fact that the hamiltonian contains only quadratic terms in the
exciton and photon operators, i.e., it describes a set of coupled harmonic oscillators, for
which the semiclassical description holds also for low quantum numbers. Thus it can
expected that semiclassical and quantum treatments of polaritons are equivalent also
in confined systems.
Here we shall describe the extension of the semiclassical theory to describe polaritons
in thin layers. Due to the lack of translational invariance along the growth direction,
the dielectric response of the system is intrinsically nonlocal. The constitutive relation
can be taken of the form

D(z)

= EooE(z) + 471" X(z,z')E(z')dz',

(109)

where the nonlocal susceptibility X(z, z') is given in linear response theory as [110]

x(z, z') =

L
>.

92

X>.(w, kcx)p>.(z)p>.(z'),

(110)

where A are quantum numbers of the excited states of the crystal. We consider only
the resonant terms, and incorporate the contribution of all nonresonant terms in the
background dielectric constant. For excitons in thin layers, we have

X.x (W, k)

_ g.xl(uc ler lu v }12


-

Ii ( Wk

W -

2,

(111)

. )'

= F.x(p = O,z,z),

p.x(z)

(112)

where F.x(p, Ze, Zh) is the exciton envelope function. Thus the semiclassical description
of polariton effects in thin layers consists of the following three steps [111, 92, 112]:
1. Determine the excitonic levels liw.x and wavefunctions F.x(p, Ze, Zh) by solving the
Schrodinger equation for the exciton;
2. Identify the resonant terms and calculate the nonlocal susceptibility (111);
3. Solve Maxwell equations with the constitutive relation (109).
Before presenting a few applications, let us discuss some features of the nonlocal theory.
III

The formalism can be applied to both the thin-film and quantum-well regimes for
the exciton. For QWs only one resonant level needs to be considered, while for
thin films one has to sum over the quantized levels of the center of mass.

III

The expression (110) has the important feature that each term in the sum is
separable in z, z'. This allows Maxwell equations to be solved analytically in
many cases [111].

III

In the thin-film regime, the formalism is intrinsically ABC-free. Additional boundary conditions are embodied in the no-escape boundary conditions for electrons
and holes separately. Also, the formalism can account for phenomena related to
the distortion of the excitonic wavefunction close to the surface, like a decrease
of the dead-layer depth for thin layers.

III

The formalism takes full account of polariton effects, including the radiative broadening (i.e. both real and imaginary parts of the self-energy are found). In expression (111), , is a nonradiative broadening term: the radiative width arises automatically from the solution of Maxwell equations with retardation, and manifests
itself as an additional broadening of absorption or reflectivity peaks.

At this point we should mention that an alternative procedure has been developed,
known as Stahl's coherent wave approach [41], in which the Schrodinger equation for the
excitons is formulated in a density-matrix scheme and is solved in the presence of the
electromagnetic field. Stahl's approach gives equivalent results for linear properties, but
can also be applied to nonlinear phenomena and to study polariton effects at frequencies
above the band edge.

Polariton effects in thin films: polariton interference vs. eM quantization.


Interference of polaritons was observed in CdS and CdSe slabs [55], in CuCI slabs [113],
and, most recently, in GaAs thin layers [114]. An example is given in Fig. 15.
The "traditional" explanation makes use of the dielectric function with spatial dispersion, which gives the two polariton branches j = 1,2 as the solutions of
c

kJ _

w2

Eoo

[1

+ Wk'2
J

2WLTWT]
-

w2

i,w

(113)

93

t;
I

I
D

11

1514

b'

ell

I'

,I

gl h'

1516
1518
photon energy (me-V)

1520

1522

Figure 15. Reflectance spectrum of a MBE grown GaAs thin layer, in the spectral region of the
exciton resonance. From Ref. [114].

The observed interference pattern can be explained as follows [55J: below the transverse
frequency, a single period is observed, which corresponds to Fabry-Perot interference
of the lower polariton according to the condition 2klL = 27rm (L is the width of the
slab). The period doubling above WL is interpreted as interference between UP and LP,
described by the condition (k1 - k2)L = 27rm. Moreover, a longer period appear, which
corresponds to interference of the UP with itself. The reflectivity can be calculated
imposing additional boundary conditions like Pekar's ABC I:i Pi = 0 at the surfaces.
For GaAs, the interference pattern is more complicated, due to the presence of two
exciton-polariton branches (heavy and light) [114J.
A related phenomenon which occurs in very thin layers is quantization of the center
of mass of the exciton, which can be observed in luminescence and reflectivity [115, 116,
117J. Assuming an excitonic wavefunction with CM confinement, selection rules for
optical transitions can be derived, which are in reasonable agreement with experiment
[115J.
The two phenomena can be treated in a unified way using the nonlocal scheme
described above. Numerical calculations of absorption and reflectivity profiles [112, 92J
show that the conditions of polariton interference are recovered for frequencies not
too close to the resonance region. When the polariton is excitonlike, the interference
condition 2kL = 27rm corresponds to the quantization condition k = (7r / L)m for the
center-of-mass wavevector: this is the case for the LP for k ~ ka, and for the UP close
to the longitudinal frequency. For very thin layers, the quantized wavevectors are too
large to involve the UP. Also, as we mentioned in Sec. 3.2, quantization of the center
of mass is only approximately described by the condition k = (7r / Lerr )m, and the more
precise quantization condition (102) must be used. In the resonance region between WT
and WL, the positions of absorption and reflectivity peaks cannot in general be simply
explained by any of the two pictures; instead, the full nonlocal theory must be used.
Polariton effects manifest themselves both as shifts with respect to the quantized CM
levels, and as additional broadenings of the peaks due to the radiative width.
The nonlocal approach has been developed so far only for the case of normal incidence. Polariton dispersion in thin films has been studied in Ref. [118J using a macroscopic theory with additional boundary conditions. The dispersion of stationary po-

94

lariton states has been calculated for both T-modes (s-polarization) and L, Z modes
(p-polarization). The case of p polarization is more complicated, since the longitudinal
mode is also excited in layers with L < A, and therefore the ABC have two components.
We refer to [118, 36] for details.
Polaritons in quantum wells: nonradiative modes. In the quantum-well limit,
the nonlocal formalism can be used to find the dispersion of nonradiative (waveguide)
modes by taking only one resonant term in the nonlocal susceptibility [26, 107, 119]
(the full quantum-mechanical theory is developed in Ref. [120]). The quantity p,\(z) =
FQw(O)c(z)v(z), where FQw(p) is the in-plane electron-hole wavefunction. We take a
propagation vector k = kxx along the x axis. We enclose the quantum well in a large
box of width d, such that the response function is essentially zero for Izl, Iz'l > d/2. We
assume an electric field which decays exponentially far from the well:
(114)
where the decay constant is given by
2
a -_ (k x

2
W )1/2
Eoo2
.

(115)

Here we give an example of the solution of Maxwell equations for the T-mode (T E, or
s-polarization). The electric and magnetic fields are given by
E

(116)

B
Maxwell equations imply

(117)
Since for TE modes V . E

= 0, the above equation gives, for Izl < d/2,


(118)

Maxwell boundary conditions require that Ell' Ell are continuous at z = d/2: from (116),

BEy/Bz is also continuous. Hence boundary conditions can be conveniently expressed


in terms of E y , yielding
(119)

It can be shown by direct substitution that a solution of Eq. (118) with boundary
conditions (119) is

Ey(z)

Jdz' p(z')e-alz-z'l.

(120)

Substituting in (118), we obtain the dispersion relation of T-mode polaritons in the


form
(121)

95

L=60 A. x=0.4

1623

~
51622

.,<1~
., 1621

1620

2
3
k / ko

Figure 16. Dispersion of the light-hole excitonic polariton in a 60


The dashed curve denotes the photon line. From Ref. [26].

A wide GaAs/ Alo.4Gao.6As

QW.

By similar manipulations, the dispersion relations of Land Z mode polaritons are


obtained in the form

(122)
27r
)1 2 X ( w, kx ) (-P(a)
k;
Z: -IFQw(O
- 21')
oo
a

= 1.

(123)

The quantities P(a) and I are defined by

P(a) =
I =

JJc*(z)v(z)v*(z')c(z')e-a1z-z'l dz dz',

Jic(zWlv(z)1 dz.
2

(124)
(125)

For all modes, the dispersion lies in the region of real a, which means that they lie
on the right of the photon line. For the L-mode, the condition X(w, k x ) < 0 implies
that the dispersion lies above that of the unretarded T-exciton, hence the dispersion
curve starts from the intersection of the photon line with the T -exciton. For the T - and
Z-modes, instead, the energies tend to zero as kx - t O. The dispersion of QW excitonic
polaritons is illustrated in Fig. 16. The polaritonic effect is seen to be important only
in a narrow region around ko. For k > ko, the effects of retardation are negligible,
and the dispersion of the exciton with the long-range exchange interaction (Fig. 13) is
recovered.
Luminescence from the L-polarized QW excitonic polariton has been reported in
Ref. [121), where coupling to the nonradiative modes is achieved by use of a grating.

96

Possible evidence for QW excitonic polaritons is given by the time-of-flight experiment


of Ref. [122J, in which an increase of the delay time of a light pulse propagating in the
layer planes is observed close to the exciton resonance. This could represent the analog
of the classical group-velocity experiments [57J.
It is of interest to investigate to which extent the quantum well can be described
within a local scheme with an effective dielectric function,

f(W) = foo[1

+ Wa - WLT
. J,
W - Z'Y

(126)

where WLT is a parameter which can be called "effective LT splitting". The relation
between the two approaches can be studied for the normal-incidence reflectivity. Within
the nonlocal scheme, the reflectivity of an isolated quantum well surrounded by infinite
barriers is found to be [123J

R(w)

(w - wa)2

r a2

+ (T + ra)2 ,

(127)

where

ra = ~~fxy
nmaC S

(128)

will be interpreted in the next section to be the decay rate of the electric field. Within
the local scheme, the reflectivity of a layer of width d is calculated to be

R(w)

2iqd
(f(W) -1)(1- e )
12,
(1 + Vf(W))2 - (1 - VE(w))2e 2iqd

=1

(129)

where q = E( W)w / c. The two approaches give identical results under the condition
qd ~ 1, which can be written as [124J

d)2
( "\
A

EooWLT
1
Wa -

W -

Z' Y 1 ~ 1.

(130)

If this condition is satisfied, the relation between the phenomenological parameter WLT
and the microscopic parameters is WLT = 2r a/(k ad). For a single quantum well, d
coincides with the well width L w , while for a multiple quantum well d must be identified
with the MQW period Lw+Lb [123, 124J. A sufficient condition for (130) to hold is that
the width 'Y be much larger than the effective LT splitting [123J. This is usually verified
experimentally. The parameter WLT can be expressed as the bulk LT splitting, multiplied
by an enhancement factor related to excitonic confinement. However we remark that
for a single quantum well WLT is not the exchange splitting between longitudinal and
transverse rs excitons, which, as we have seen, vanishes for kex ---t O. For MQWs,
on the other hand, WLT is expected to coincide with the finite exchange splitting at
k/l = O. A thorough comparison between local and nonlocal schemes for what concerns
the exciton-polariton dispersion is done in Ref. [124J, also for the case of MQWs, where
it is shown that the condition for the validity of the local description is given again by
Eq. (130).
The nonlocal scheme can be used to calculate the reflectivity and attenuated total
reflection of a quantum well. Such a calculation is done in Ref. [125J for both sand p
polarizations. Excitonic polaritons in the radiative region appear as peaks or dips in
the reflectivity, while nonradiative polaritons manifest themselves as dips in the ATR
spectrum [125J. In Ref. [126J the nonlocal formalism is applied to a calculation of the
reflectivity in quantum wells, quantum wire, and quantum dot structures.

97

0.2

0.1

L (Ie)
Figure 17. Radiative decay rate of the k =

a exciton

as a function of crystal thickness. From Ref.

[130].

3.4

Exciton Radiative Lifetime: from Bulk to Confined Systems

Crossover 2D--t3D: overview. Excitons in thin layers with thickness L < ,\ have
a finite radiative lifetime, as was first shown in Ref. [108]. For Frenkel excitons, the
decay rate is of the order of r ~ (,\2/a 2)rmob where a is the lattice spacing and rmol
is the decay rate of an isolated molecule. The decay is called "superradiant", because
the decay rate is proportional to the number of molecules within a wavelength, which
contribute in phase to the decay [127]. The superradiant decay of excitons in anthracene
films was observed in Ref. [128]. For Wannier-Mott excitons, the decay is still often
called superradiant, although it is not meaningful anymore to speak of an enhancement
with respect to the isolated molecule. I prefer the picture of Fig. 14, according to
which the intrinsic radiative decay of a free exciton in low-dimensional systems is due
to coupling with a continuum of photon states.
For excitons in thin layers, the "superradiant" decay must go over to the stable
polariton behavior as the film thickness becomes L >.. The behavior of the decay
rate as a function of thickness is calculated in Refs. [129, 130] (see Fig. 17, which is
calculated for Frenkel excitons). For L '\, the decay rate increases as r ()( L: for
Wannier-Mott excitons, this corresponds to the regime where the center of mass of the
exciton is quantized, and the oscillator strength per unit area increases linearly with the
well thickness (see Fig. 11). For L >., the decay rate has an oscillatory behavior and
decreases as 1/ L due to the reduced overlap between exciton and photon wavefunctions.
In fact, for L ,\ the physical picture is expected to coincide with the bulk polariton
behavior, for which the decay rate is an "escape rate" which goes as 1/ L (see Eq. (85)).
This crossover from superradiant decay to bulk polaritons is therefore well described
by the results of Refs. [129, 130]. In Ref. [130] it is also shown that the bulk polariton
behavior is obtained at all wavevectors, when proper approximations to the radiative
self-energy are made. The QW regime L ~
'\, where the oscillator strength (and
therefore the decay rate) increases as the well width is reduced, is not described in Refs.
[129, 130].

as

98

Radiative lifetime of QW excitons. Now we calculate the radiative decay rate


of a free exciton in an isolated QW, under the assumption that the in-plane wavevector
kll = kxx is conserved [131, 119, 123, 132]. We use the Golden Rule (20), with the
interaction Hamiltonian being given by (21). The initial state consists of the exciton
state with polarization vector e and no photons present, while the final state is the
crystal ground state plus a photon with wavevector k = (qll, kz) and polarization >..
Expressing the vector potential in second quantization according to (81), the matrix
element is calculated to be

(131)
The squared matrix element summed over the photon polarizations can be expressed
in terms of the exciton oscillator strength as 2
LI(il.C!fW
A

11" e 2 li,2

= 2-L!Ele.g(A)1
V

n ma

(132)

Now the Golden Rule gives

(133)
Evaluating the one-dimensional density of states as in (108) gives the decay rate in
terms of the oscillator strength per unit area as

r(kx)

= 211" ~ ka L !E Ie. g(A)1 2 0(ka n mackz

k x ),

(134)

where now kz = Jk6 - ki. We must now specify the exciton polarization vector e and
the photon polarization vectors g(A). For a given in-plane wavevector kll = kxx, the two
orthogonal photon polarization vectors can be chosen as

(135)
For the T-exciton, I:A f,le. g(A)1 2 = fxy, where fxy is the oscillator strength for in-plane
polarization. 2 For the L-exciton, we obtain I:A !Ele. g(A)1 2 = fxy(k z /k a)2, while for the
Z-exciton I:A !Ele . g(A)1 2 = fz(k x /k a)2. Thus we obtain the radiative widths of the T,
L, and Z excitons for kx < ka as follows:

211" e 2 fxy ko
-----n mac S k z '
211" e 2 fxu kz

---:;;: mac S k o '


211" e 2 fz k;
-----n mac S kokz

(136)

(the factor of two was missing in Ref. [123]). The decay rate vanishes for kx > ko. For
the light-hole exciton fxy = 4fz. Optically inactive states obviously have zero radiative
width.
2Note that for a given exciton polarization vector e, the oscillator strength can be thought to be
summed over all possible polarizations, since only the vector i = e contributes. Thus the oscillator
strengths fxy, fz are exactly those calculated e.g. in Ref. [73], and reported in Fig. 10.

99

At kll = 0, Land T modes have the same decay rate 2ro == (27re2/(nmoc))(fxlJ/S).
Taking an oscillator strength fx,,/ S = 50 .10- 5 A-2, which is appropriate for the HH1CB1 exciton in GaAs/ AlxGal_xAs quantum wells of about 100 A width [73], we obtain
firo = 0.026 meV. The decay time of an exciton state is TO = 1/(2ro) = 12 ps [132].
This coincides with the result found from the reflectivity calculation with the nonlocal
susceptibility (Sec. 3.3): the decay time of the electric field is l/r o, while the decay time
of the intensity is twice as short. Note that the Golden Rule gives directly the decay
rate of the exciton state. This lifetime is much longer than found in Ref. [131], where
the index of refraction is not considered and the two-dimensional limit for the exciton
is assumed. Observation of such short lifetimes has been reported in Refs. [133, 134].
The radiative width (136) represents the imaginary part of the exciton self-energy
due to interaction with photons. The real part of the self-energy, calculated e.g. in Refs.
[131, 119, 132], is a small effect. The largest polariton effect for quantum-well excitons
is the radiative lifetime. Note that the decay rate diverges as kx --t ko for the T and
Z modes. This divergence, which can be traced back to the density of states (108), is
integrable and disappears when the thermal average is taken. A slight broadening of
the wavevector due to inelastic scattering would wash out this divergence, which we
believe to have no physical consequences.
The intrinsic radiative decay of free excitons is calculated assuming conservation
of the in-plane wavevector, thereby disregarding the effects of interface roughness and
acoustic phonon scattering. Wavevector conservation is likely to be a good assumption
when the coherence length of the exciton is longer than the wavelength of light. Consequently, the short intrinsic lifetimes can only be observed in carefully selected samples
at low temperature [133, 134]. In general, several effects can change this simple picture.
At low temperature, excitons can be bound to impurities or interface defects. Also,
interface roughness acts as a disordered potential for the exciton motion, and produces
a mobility edge within the inhomogeneously broadened exciton line [135]: below the
mobility edge the exciton is localized by the disorder, while above the mobility edge
the exciton is mobile and interface roughness acts as a dephasing mechanism. Finally,
scattering with acoustic phonons has to be taken into account at finite temperature.
The interplay between all these effects constitutes a complicated problem, which is only
partly understood at time of writing. In the following I shall discuss a few models which
have been proposed [136].
Effect ofthermalization. Thermalization is due to inelastic scattering with acoustic phonons [53], which changes the exciton wavevector but does not change the total
exciton population (scattering with optical phonons is not expected to be relevant for
thermal energies smaller than 36 meV, which is the LO phonon energy in GaAs). The
key point is comparing the scattering rate with the radiative lifetime. The scattering
rate by acoustic phonons is measured to be linear in T with a coefficient I = 5 pe V / K,
for a QW of 135 A width [53]. Thus thermalization processes are faster than radiative
decay for T > 10 K. There is also other evidence for this conclusion, coming from
a high-energy Boltzmann tail of the exciton lines [137]. Thus it can be assumed as a
working assumption that thermalization processes are faster than radiative recombination, i.e., that excitons always have a thermal distribution while they decay radiatively.
However the possibility of a failure of this assumption must be kept in mind, particularly for thermalization between dipole-allowed and triplet exciton states (see Fig. 12),
since spin-flip scattering is likely to be slower. Also, having a thermalized distribution
depends on the conditions of excitation (resonant or nonresonant) and on the time of
observation [134].

100

In the assumption of a rapid thermalization, the decay rate of the luminescence is


given by the thermal average of the decay rate (136). The two characteristic energies
are the thermal energy kBT, and the kinetic energy of excitons which decay radiatively:
the latter is at most El = 1i 2 k5!(2M), where M is the exciton mass. Using M = 0.25
mo, we find 1i2k6/(2M) ~ 1.1 K. This means that, for T ~ 1 K, only a small fraction of
excitons occupy the states with kx < ko, which can decay radiatively. We introduce the
parameter a == Ed(kBT) = 1i 2 k5!(2Mk BT), which we assume to be ~ 1. Performing
the average over the Boltzmann distribution gives, for the T and L modes,

(137)

(138)
Assuming that the four spin states of the exciton (see Fig. 12) are equally populated, the
thermally-averaged decay rate of the HH exciton is given by r(T) = (r T (T)+r L (T))/4.
Thus the decay rate of the luminescence is given by
(139)
where TO = 1/(2ro ) is the radiative lifetime at kll = O. If spin-flip scattering is slow
and the exciton is thermalized only over the two optically-active states, the lifetime is
twice as short. In both cases, the radiative decay rate (136) is multiplied by the small
fraction of excitons with kx < ko, and is reduced accordingly. Equation (139) predicts
an effective radiative lifetime which rises linearly with temperature, and which is much
longer than the "bare" radiative lifetime at kll = o. A radiative lifetime which rises
linearly with temperature has been observed by several groups [138,139, 134], although
the numerical values for the slope are found to be strongly dependent on the sample.
"Coherence volume" lifetime. A different approach to the radiative lifetime
was proposed by Feldmann et al. [138]. It is based on the idea that excitons have a
"coherence area" A c , which is determined by the homogeneous linewidth rh via the
two-dimensional density of states:

1 _larh(T)
_ Mrh(T)
-A g(E) dE 2'
c

2~1i

(140)

where M is the exciton mass. It is assumed in Ref. [138] that the radiative decay is
described by Eq. (86), with the effective dimensionless oscillator strength given by
(141)
where f / S is the usual oscillator strength per unit area. The radiative decay rate is
then given by
4~n1i2e2w2

7(T)

= 3moMC3rh(T) S'

(142)

101

and depends on temperature via the homogeneous broadening. Assuming a thermal


distribution, and for kBT ~ E 1 , only states within r,,(T) contribute to the thermallyaveraged radiative decay rate, which is then obtained as
(143)
At temperatures kBT
perature,

r",

the radiative lifetime is found again to be linear in tem-

(144)
Equation (143) implies a relation between radiative lifetime and homogeneous broadening, which has been observed experimentally [140]. However the introduction of a
coherence area and the use of formula (86) (valid in principle only for bound excitons)
are not derived from a theory, and therefore remain an ad-hoc procedure (see discussion
at the end of Sec. 2.5).
Effect of dephasing. The "polariton" theory of the radiative lifetime is based on
the conservation of wavevector, i.e., on the assumption that the center-of-mass wavefunction of the exciton is coherent over distances ~ A. However, on increasing the
scattering rate, spatial coherence must eventually be lost. Qualitatively, dephasing
processes reduce the coherence length lc and make the radiative decay rate smaller
than that obtained from the polariton+thermalization treatment. Such a behavior has
been explicitly demonstrated with a microscopic model [141]. Thus scattering processes
act in two different ways: through reduction of the coherence length, they change the
way the system eigenstates are constructed; through thermalization, they change the
way the eigenstates are populated. If lc ~ A~/2 is the coherence length, there is a spread
in wavevector 5k ~ lc -1 which results in a homogeneous broadening r" ~ 1i2 j(Mlc 2). It
must be emphasized that these considerations are very qualitative: it is difficult to give
a rigorous definition of the coherence length, and also the relation between coherence
length and homogeneous broadening is not really understood.
A theory of the radiative lifetime which describes both "polariton" and "coherence
volume" pictures has been formulated by Citrin [132]. Starting from the polariton
picture, the effect of a homogeneous broadening can be described within the thermal
averaging procedure by introducing a spectral function for the exciton states,
_ fe-(3E fk<ko r(k)A(E, 1iwk) dkdE
'Y ( T ) f e-(3E dE
'

(145)

where the spectral function is taken to be a Lorentzian,


(146)
The results depend on the relative values of r", kBT, and E1 = 1i 2k5j(2M). Under
the conditions of fast thermalization and dephasing, two regimes can be distinguished:
for h ~ E 1 , homogeneous broadening has a negligible effect, and the results of Eq.
(139) are recovered. For
~ E 1 , instead, the decay width is given by the expression
(143). For kBT ~ r", E1 the two expressions coincide. Thus homogeneous broadening
is unimportant for kBT ~ rho

102

r"

It is interesting that the relation between radiative lifetime and homogeneous broadening can be derived without introducing the ad-hoc concept of a coherence area. However if r h rv li,2 j(Mle 2), the condition r" ~ El for the homogeneous broadening corresponds to weak spatial coherence le >.. Thus the concept of a coherence area with a
"giant oscillator strength" lifetime could be valid when le >.. The situation here is
totally analogous to exciton decay in bulk materials, as discussed in Sec. 2.5.
For GaAs-Gal_xAlxAs quantum wells, the intrinsic homogeneous broadening is r" rv
0.2 meV [140]. Since El rv 0.08 meV, the excitonic lifetime is probably described by the
coherence area rather than by the polariton picture. However it is very important that
the radiative lifetime is measured together with the homogeneous broadening, which
can vary from sample to sample.
In Ref. [132] the large sample-dependence of the lifetimes is interpreted in terms of
the effect of excitons bound to interface defects. Lifetimes of the order of 100-200 ps
are calculated for bound excitons: from this it can be shown that bound excitons have
an important effect on the thermally averaged lifetime, even when the temperature is
much larger than the binding energy. This fact makes it difficult to give a quantitative
theory for the measured lifetimes without introducing unknown parameters describing
the shape and concentration of defects.

Exciton lifetime: overview. To conclude this Section, we try to speculate about


a unified picture for the exciton radiative lifetime in bulk and in confined systems.
The polariton picture for the exciton radiative lifetime applies only when the coherence
length le ~ >., while in the opposite limit le >. the concept of a coherence volume
might apply, with the radiative lifetime being described by Eq. (86). In bulk semiconductors, the polariton picture leads to a lifetime which represents a confinement
time and which increases with the sample's size, while in all kinds of confined systems
(quantum wells, wires, and dots) the exciton has an intrinsic mechanism for radiative
recombination since it is coupled to a density of states of photons. For kBT ~ r h ,
the polariton and coherence-volume pictures coincide, and the behavior of the radiative lifetime as a function of temperature (under the conditions of rapid thermalization
and dephasing) reflects the dimensionality of the systems. In the bulk, the lifetime
for kBT ~ r h goes as r ()( T 3 / 2 , while the dependence is r ()( T in quantum wells:
the transition between the two regimes has been explicitly demonstrated in Ref. [96]
by averaging over the quantized center-of-mass levels. In one-dimensional systems like
quantum wires, the radiative lifetime for kBT ~ r" goes as r ()( T 1 / 2 [142]. In quantum
boxes, the complete quantization of electronic levels makes the problem of the coherence
length irrelevant: the excitonic decay is always described by the a formula like (86),
with the "giant oscillator strength" being proportional to the volume [131], and the
lifetime is expected to be independent of temperature as long as kBT is smaller than
the quantization energies.

CONCLUSIONS AND PROSPECTS

The present lectures have been focused on the following issues: oscillator strength
and optical absorption, excitons and polaritons in bulk semiconductors, excitons and
polaritons in confined systems, and the exciton radiative lifetime. I shall now give a
brief overview on these points.
Optical absorption above the gap is determined by three factors: the oscillator
strength, the joint density of states, and the excitonic enhancement. The oscillator
strength of a transition is a dimensionless quantity, which can be defined in both clas-

103

sical and quantum-mechanical terms, and whose main property is to satisfy the sum
rule (12). The oscillator strengths of allowed interband transitions in the bulk, interband transitions in quantum wells, and intraband transitions in quantum wells are
essentially equal (for type-I quantum wells with infinite barriers): differences in the
absorption come mainly from the joint density of states. The excitonic effect makes
the absorption coefficient continuous at the interband absorption edge. The absorption
coefficient can be accurately and quantitatively evaluated by expressing the interband
momentum matrix element in terms of the effective mass. A proper description of
absorption in quantum well structures is more suitably given in terms of a dimensionless absorption probability, and it is important to distinguish between the different
directions of propagation.
The exciton is the lowest electronic excitation of the whole crystal. The oscillator
strength of an excitonic transition is proportional to the crystal volume. Excitons in
semiconductors have a finite translational mass and therefore an energy dispersion. The
electron-hole exchange interaction, coming from the longitudinal (instantaneous) part
of the electromagnetic field, introduces a splitting between longitudinal and transverse
excitons with the symmetry of the dipole. For an exciton wavevector smaller than the
wavevector of light in the crystal, the exciton dispersion is modified by interaction with
the transverse (retarded) electromagnetic field, i.e., by polariton effects. The coupling
matrix element between exciton and photon at the crossing point is given by the energy
liwc (Eq. (83)), which is much larger than the LT splitting. The oscillation time between
exciton and photon within the polariton state (i.e., the period of Rabi oscillations)
is given by (2wctl. In an infinite crystal, polaritons are stationary states: what is
observed as a radiative lifetime is rather the confinement time of polaritons within a
sample of finite size. Breakdown of polariton effects is likely to occur because of the
loss of spatial coherence in the exciton-photon coupling, before temporal coherence is
lost. However, a microscopic description of the transition between polariton and exciton
pictures at finite temperature is still lacking, as well as a microscopic theory of optical
absorption in the polariton framework.

as,

where
Excitons in thin layers are characterized by the quantum-well regime L rv
electron and hole sub bands are separately quantized, and the thin-film regime L )i>
where the motion of the center of mass is quantized. Consequently, the oscillator
strength per unit area has a minimum at a critical well width. This qualitative picture
applies to quantum wires and dots as well. A realistic theory of excitons in quantum
wells must take into account valence band mixing and also the effects of conduction band
nonparabolicity and dielectric confinement. Taken together, these effects result in very
high binding energies. The electron-hole exchange interaction in single quantum wells
is qualitatively modified as compared to the bulk: in particular, there is no exchange
splitting between longitudinal and transverse excitons as kex - t O. This behavior is
modified in multiple quantum wells. The exchange splitting between longitudinal and
transverse excitons does not coincide with the effective parameter WLT appearing in an
effective, local dielectric function which approximately describes the optical response of
a single quantum well: this parameter is finite and increased by confinement.

as,

Exciton-polaritons in thin films are divided into radiative and nonradiative modes.
A semiclassical theory based on the nonlocal susceptibility gives a full description of
polariton effects, including the imaginary part of the self-energy, i.e., the radiative
lifetime. In the thin-film regime, polariton interference and center-of-mass quantization
occur. These phenomena have been studied so far mainly for normal incidence. In
the quantum-well regime, the stationary exciton-polariton states are the nonradiative
modes. In the radiative region, the real part of the polaritonic shift is very small: the
104

main polariton effect is the radiative lifetime. The intrinsic radiative lifetime of free
excitons in confined systems can be understood as arising from a coupling between the
exciton state and a continuum of photon states (Fig. 14). This intrinsic recombination
mechanism (as compared to polariton thermalization and escape in bulk crystals) is
probably one of the main reasons for the increased efficiency of free-exciton luminescence
in quantum-well structures as compared to the bulk. Polariton effects in multiple
quantum wells and superlattices have been only partly studied so far [124, 136J, and
only from the theoretical side. Polariton effects in micro cavities are likely to become a
major area of research in the next few years.
The radiative lifetime of free excitons in real crystals is only partly understood
at present. In bulk GaAs samples of good quality (as well as in other crystals with
smaller exciton radius) the polariton mechanism was shown in the past to determine
the luminescence lineshape of free excitons up to 20 K: this mechanism for light emission
is strongly dependent on crystal shape and excitation conditions. There is no theory
which describes samples of worse quality or at higher temperatures, when the coherence
length becomes smaller than the wavelength of light. In thin films, the role of interfaces
becomes progressively more important on reducing the thickness. In quantum wells,
localization and dephasing due to interface roughness are crucial phenomena. The
"polariton" theory, which neglects such effects, leads to short (rv 10 ps) lifetimes for
free excitons at low temperature, and to a linear increase with temperature for T ~ 1 K.
Both phenomena have been observed, but a quantitative agreement between theory and
experiment is not reached, indicating that the real situation is more complex. Major
progress has been made in clarifying the relation between polariton and coherencevolume pictures [132J. However a complete understanding of experimental results in
terms of a microscopic theory of exciton lifetime including localization, dephasing by
disorder, and phonon scattering processes is not yet reached.

ACKNOWLEDGEMENTS
I am indebted to F. Bassani for convincing me of the importance of the polariton
concept, and to him and F. Tassone for the collaboration during the research on polaritons in confined systems. I acknowledge useful discussions with M. Colocci, M. Gurioli,
and D. Wolford on the subject of exciton-polariton lifetimes. I greatly benefitted from
the stimulating athmosphere of the Erice School, and I would like to thank many of the
participants for criticism, suggestions, and encouragement: among them, V. M. Agranovich, G. Bastard, D. S. Citrin, R. G. Ulbrich, E. Yablonovitch, and, in particular, C.
Weisbuch.

APPENDIX: CONVERSION OF EQUATIONS BETWEEN GAUSSIAN


UNITS AND STANDARD INTERNATIONAL (SI) UNITS
In these lecture notes, the Gaussian system of units is employed. Gaussian units are
based on the cgs system of units for length, mass, and time. The electric charge is a
derived quantity, whose dimensions are determined by the Coulomb law written in the
form F = ql qz/ r2. No permittivity or permeability offree space is defined: instead, the
speed of light c explicitly appears in Maxwell equations. In the Gaussian system, the
electric and magnetic fields have the same dimensions, and have also the same magnitude for an electromagnetic wave in vacuum. The macroscopic fields D, P, H, and M
have also the same dimensions as the basic fields E and B. The relations between Gaus-

105

Table 1. Macroscopic Maxwell equations and a few other important relations in Gaussian and SI units.
Equation
Maxwell equations

Definitions of D and H

Gaussian units

SI (MKSA) units

V D = 471"p
471"
18D
VxH=-J+-c
c 8t
18B
V X E+-- = 0
c 8t
VB=O

VD=p

D = E + 471"P
H = B - 471"M

Constitutive relation
Relation between
fields and potentials
Minimal coupling
Coulomb force
Ohm's law

106

D = E = (1 + 47rX)E
18A
E=-V--c 8t
B=VxA
e
p -4 P --A
c
F = qlq2
r2

J = erE

8D
VxH=J+
8B t
VxE+
=O

8t

VB=O
D = oE+P
1
H=-B-M
/-Lo
D = E = (1 + X)oE
8A
E=-V--

8t

B=VxA
p

-4

P - eA

F = _1_qlq2
47r0 r2

J = erE

Table 2. Conversion table for formulas [4]. To convert any equation in Gaussian units
to the corresponding equation in SI units, on both sides of the equation replace the
relevant symbols listed below under "Gaussian" by the corresponding "SI" symbols
listed on the right. The reverse transformation is also allowed. The symbols for length,
mass, time and any other not electromagnetic quantities are unchanged. Note that the
physical quantities have different dimensions and numerical values in the two systems
of units. A conversion table for a given amount of a physical quantity is given in Ref.
[4], p. 820.
Quantity
Velocity of light
Charge (charge and current

Gaussian units

SI (MKSA) units

(/-lof-O)-1/2

q(p,J,I,P)

~q(p,J,I,P)

E (</>, V)

J47ft oE (</>, V)

47fto

density, current, polarization)


Electric field (potential, voltage)
Displacement field

Magnetic induction

B (A)

(vector potential)
Magnetic field
Magnetization

J47f/-loH

l!M
47f
(j

Conductivity

(j

Dielectric constant

Electric susceptibility

Permeability

/-l

47fto
t

to
X
47f
fi

tlO

107

sian and SI (or MKSA) units, as well as with other systems of electromagnetic units,
are discussed in detail in the Appendix of Jackson [4J. Here we collect a few important
formulas in both systems of units (Table 1), and describe how to convert any equation
from the Gaussian to the SI system and vice-versa (Table 2). To give an example of the
use of Table 2, relations like (30), (63), (68), (92), (107) are converted to the SI system
by the replacement e2 ~ e2 /( 47rEo) (provided all quantities are reinterpreted as MKSA
variables) .

READING GUIDE
At the request of the editor, we list some books and review papers that provide a
good introduction to the field, as well as some pioneering papers. Papers cited in the
text are listed at the end of this chapter.

1. F. Bassani and G. Pastori Parravicini, Electronic States and Optical Transitions in Solids (Perg-

2.

3.

4.

5.

6.

7.

8.

amon Press, Oxford, 1975). Monograph on band structure and optical properties of solids,
including excitons. The first two chapters contain a review of group theory, and symmetry
arguments are used throughout the book.
R. J. Elliott, Intensity of Optical Absorption by Excitons, Phys. Rev. 108, 1384 (1957). Classic
paper on the calculation of the absorption coefficient in semiconductors, including excitonic
effects.
J. O. Dimmock, Introduction to the Theory of Exciton States in Semiconductors, in Semiconductors and Semimetals, vo!' 3, edited by R. K. Willardson and A. C. Beer (Academic Press, New
York, 1967), p. 259. Introduction to the theory of Wannier excitons, including the effective-mass
approximation, optical absorption, and the effect of an external magnetic field.
J. J. Hopfield, Theory of the Contribution of Excitons to the Complex Dielectric Constant of
Crystals, Phys. Rev. 112, 1555 (1958). Key paper on excitonic polaritons. Gives a detailed
presentation of the quantum theory of polaritons, and a profound discussion of the problem of
exciton lifetime and absorption in the polariton framework.
D. D. Sell, S. E. Stokovski, R. Dingle, and J. V. DiLorenzo, Polariton Reflectance and Photoluminescence in High-Purity GaAs, Phys. Rev. B 7, 4568 (1973). First report of polariton effects
in GaAs, from a careful analysis of reflectivity and photoluminescence experiments.
C. Weisbuch, and R. G. Ulbrich, Resonant Light Scattering Mediated by Excitonic Polaritons
in Semiconductors, in Light Scattering in Solids III, edited by M. Cardona and G. Giinterodt,
Topics in Applied Physics, vo!' 51 (Springer, Berlin, 1982), p. 207. Review paper on excitonic
polaritons, with emphasis on light-scattering experiments.
V. M. Agranovich, and O. A. Dubovskii, Effect of Retarded Interaction on the Exciton Spectrum
in One-Dimensional and Two-Dimensional Crystals, JETP Lett. 3, 223 (1966). Treatment of
polariton effects in low-dimensional systems, with the distinction between radiative and nonradiative modes.
K. Cho, "AECn-Free Theol'y of Polaritons: from Semi-Infinite Medium to Quantum Well, J.
Phys. Soc. Jpn. 55, 4113 (1986). Formulation of linear response theory with the nonlocal susceptibility, with application to exciton-polaritons in thin films and quantum wells.

References
[1] V. M. Agranovich, and V. L. Ginzburg, Spatial Dispersion in Crystal Optics and the Theory of
Excitons (Interscience Pub!., London, 1966).
[2] See e.g. W. Hanke, Adv. Phys. 27, 287 (1978) and references therein.
[3] See e.g. F. Wooten, Optical Properties of Solids (Academic Press, London,1972).
[4] J. D. Jackson, Classical Electrodynamics (Wiley, New York, 1975).
[5] F. Stern, Solid State Physics, vo!' 15, F. Seitz, D. Turnbull, eds, p. 300 (1963).

108

[6] M. Altarelli, D. L. Dexter, H. M. Nussenzveig, and D. Y. Smith, Phys. Rev. B 6, 4502 (1972).
[7] F. Bassani, and S. Scandolo, Phys. Rev. B 44, 8446 (1991); S. Scandolo, and F. Bassani, Phys.
Rev. B 45, 13257 (1992); F. Bassani, Electronic Structure and Dielectric Properties of Superlattices, in Optical Properties of Semiconductors, edited by G. Martinez (Kluwer Academic Pub!.,
Amsterdam, 1993), p. 27.
[8] M. Sheik-Bahae et a!., IEEE J. Quantum Electron. QE-27, 1296 (1991).
[9] F. Bassani, and G. Pastori Parravicini, Electronic States and Optical Transitions in Solids (Pergamon Press, Oxford, 1975).
[10] D. Greenaway, and G. Harbeke, Optical Properties of Semiconductors (Pergamon Press, Oxford,
1966).
[11] R. S. Knox, Theory of Excitons, in Solid State Physics, supp!. 5, edited by F. Seitz and D.
Turnbull (Academic Press, New York, 1963).
[12] J. O. Dimmock, Introduction to the Theory of Exciton States in Semiconductors, in Semiconductol's and Semimetals, vo!' 3, edited by R. K. Willardson and A. C. Beer (Academic Press,
New York, 1967), p. 259.
[13] See e.g. G. D. Mahan, Many Particle Physics (Plenum Press, New York, N.Y., 1981).
[14] See e.g. G. Iadonisi, and F. Bassani, II Nuovo Cimento D 2, 1541 (1983) and references therein.
[15] J. Dresselhaus, J. Phys. Chem. Solids 1, 14 (1956).
[16] J. M. Luttinger, Phys. Rev. 102, 1030 (1956).
[17] A. Baldereschi, and Nunzio O. Lipari, Phys. Rev. B 3,439 (1971).
[18] R. J. Elliott, Phys. Rev. 108, 1384 (1957).
[19] J. J. Hopfield, J. Phys. Chem. Solids 15, 97 (1960).
[20] We use the group-theoretical notations of G. F. Koster, J. O. Dimmock, R. G. Wheeler, and H.
Statz, Properties of the Thirty-two Point Groups (MIT, Cambridge, 1963).
[21] E. O. Kane, Energy Band Theory, in Handbook of Semiconductors, edited by W. Paul (NorthHolland Amsterdam, 1982)' p. 193.
[22] M. Sturge, Phys. Rev. 127, 768 (1962); D. D. Sell, Phys. Rev. B 6, 3750 (1972); D. E. Hill, Solid
State Commun.11, 1187 (1972). See also: H. Mayer, U. Rossler, and M. Ruff, Phys. Rev. B 47,
12929 (1993).
[23] Y. Onodera and Y. Toyozawa, J. Phys. Soc. Jpn. 22,833 (1967).
[24] O. Schutz, and Ch. Uihlein, Phys. Status Solidi (b) 63, 545 (1974).
[25] K. Cho, Phys. Rev. B 14,4463 (1976).
[26] L. C. Andreani, and F. Bassani, Phys. Rev. B 41, 7536 (1990).
[27] M. M. Denisov, and V. P. Makarov, Phys. Status Solidi (b) 56, 9 (1973).
[28] U. Rossler, and H.-R. Trebin, Phys. Rev. B 23, 1961 (1981).
[29] K. Ehara, and K. Cho, Solid State Commun. 44,453 (1982).
[30] K. Cho, Solid State Commun. 33, 911 (1980).
[31] E. Burstein, F. De Martini (eds.): Polaritons, Proceedings of the First Taormina Conference on
the Structure of Matter (Pergamon, New York, 1974).
[32] C. Weisbuch, and R. G. Ulbrich, Resonant Light Scattering Mediated by Excitonic Polaritons
in Semiconductors, in Light Scattering in Solids III, edited by M. Cardona and G. Gunterodt,
Topics in Applied Physics, vo!' 51 (Springer, Berlin, 1982), p. 207.
[33] S. I. Pekar, Crystal Optics and Additional Light Waves (Benjamin-Cummings, Menlo Park,
California,l983) .
[34] B. Honerlage, R. Levy, J. B. Grun, C. Klingshirn, and K. Bohnert, Phys. Rep. 124, 163 (1985).
[35] F. Bassani, and L. C. Andreani, Ea:citon-Polariton States in Insulators and Semiconductors,
in Excited-State Spectroscopy in Solids, edited by U. Grassano and N. Terzi (Academic Press,
Amsterdam, 1987), p. l.
[36] P. Halevi, Exciton-Polaritons and Optical Properties of Direct-Gap Semiconductors, in Spatial
Dispersion in Solids and Plasmas, edited by P. Halevi (Elsevier, 1992), p. 340; R. Fuchs, and P.
Halevi, Basic Concepts and Formalism of Spatial Dispersion, ibid., p. 2.
[37] I<. Huang, Proc. Roy. Soc. A 208, 352 (1951).
[38] M. Born, and K. I-luang, Dynamical Theory of Crystal Lattices, Clarendon Press, Oxford, 1954.
[39] S. I. Pekar, J. Expt!. Theoret. Phys. 33, 1022 (1957) [Sov. Phys. JETP 6, 785 (1958)].
[40] J. J. Hopfield, and D. G. Thomas, Phys. Rev. 132,563 (1963).
[41] A. Stahl, and I. Balslev, Electrodynamics of the Semiconductor Band Edge (Springer, Berlin,
1987).
[42] A. D'Andrea, and R. Del Sole, Phys. Rev. B 32, 2337 (1985).
[43] D. Frohlich, E. Mohler, and P. Wiesner, Phys. Rev. Lett. 26, 554 (1971).
[44] B. Honerlage, A. Bivas, and Vu Duy Phach, Phys. Rev. Lett. 41, 49 (1978).
[45] J. J. Hopfield, Phys. Rev. 112, 1555 (1958).

109

[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]

[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]

[70]
[71]

[72]
[73]
[74]
[75]

[76]

[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]

110

V. M. Agranovich, J. Exptl. Theoret. Phys. 37,430 (1959) [Sov. Phys. JETP 37,307 (1960)].
U. Fano, Phys. Rev. 103, 1202 (1956).
A. Quattropani, 1. C. Andreani, and F. Bassani, II Nuovo Cimento D 7,55 (1986).
M. Artoni, and J. Birman, Phys. Rev. B 44, 3736 (1991); P. Schwendimann, and A. Quattropani,
Europhys. Lett. 17,355 (1992).
See lecture notes by J .-M. Raimond and S. Haroche, this volume.
J. J. Hopfield, Phys. Rev. 182,945 (1969).
R. Ulbrich, and C. Weisbuch, Phys. Rev. Lett. 38, 865 (1977).
L. Schultheis, J. Kuhl, A. Honold, and C. W. Tn, Phys. Rev. Lett. 57, 1797 (1986).
W. C. Tait, Phys. Rev. B 5, 648 (1972).
V. A. Kiselev, B. S. Razbirin, and 1. N. Uraltsev, Phys. Stat. Solidi (b) 72, 161 (1975).
D. D. Sell, S. E. Stokovski, R. Dingle, and J. V. DiLorenzo, Phys. Rev. B 7,4568 (1973).
R. G. Ulbrich, and G. W. Fehrenbach, Phys. Rev. Lett. 43, 963 (1979).
H. Sumi, J. Phys. Soc. Jpn. 41, 526 (1976).
C. Weisbuch, and R. G. Ulbrich, Phys. Rev. Lett. 39, 654 (1977); J. Lumin. 18/19, 27 (1979).
M. Dagenais, and W. F. Sharfin, Phys. Rev. Lett. 58, 1776 (1987); F. Vallee, F. Bogani, and C.
Flytzanis, Phys. Rev. Lett. 66, 1509 (1991). See also: W. C. Tait, and R. L. Weiher, Phys. Rev.
166, 769 (1968).
Y. Toyozawa, Prog. Theor. Phys. Suppl. 12, III (1959).
U. Heim, and P. Wiesner, Phys. Rev. Lett. 30, 1205 (1973).
W. J. Rappel, L. F. Feiner, and M. F. H. Schuurmans, Phys. Rev. B 38, 7874 (1988).
D. L. Dexter, Theory of the Optical PropeTties of Impe1jections in Nanometals, in Solid State
Physics, vol. 6, edited by F. Seitz and D. Turnbull (Academic, New York, 1958).
E. 1. Rashba, and G. E. Gurgenishvili, Fiz. Tverd. Tela 4, 1029 (1982) [Sov. Phys.-Solid State
4,759 (1962)].
C. H. Henry, and K. Nassau, Phys. Rev. B 1, 1628 (1970).
G. W. t' Hooft, W. A. J. A. van del' Poel, 1. W. Molenkamp, and C. T. Foxon, Phys. Rev. B
35, 8281 (1987).
V. M. Agranovich, and V. E. Kravtsov, Solid State Commun. 55, 85 (1985); V. M. Agranovich,
Solid State Commun. 78, 747 (1991).
The Dielectric Function of Condensed Systems, edited by L. V. Keldysh, D. A. Kirzhnitz, and
A. A. Maradudin (North-Holland, Amsterdam, 1989).
G. Bastard, Wave Mechanics Applied to Semiconductor Heterostructure (Les Editions de
Physique, Paris, 1989).
M. Altarelli, Envelope-Function Approach to Electronic States in HeterostructuTes, in Semiconductor Super/attices and Inte1jaces, Proceedings of the International School of Physics "E.
Fermi", edited by A. Stella (North-Holland, Amsterdam, 1993), p. 217.
U. Ekenberg, Phys. Rev. B 40, 7714 (1989).
L. C. Andreani, and A. Pasquarello, Phys. Rev. B 42, 8928 (1990).
Lecture notes by G. Bastard, this volume.
R. Grousson, V. Voliotis, P. Lavallard, M. L. Roblin, and R. Planel, Semicond. Sci. Techn. 8,
1217 (1993).
W. T. Masselink, P. J. Pearah, J. I<lem, C. K. Peng, H. Morko", G. D. Sanders, and Y.-C. Chang,
Phys. Rev. B 32, 8027 (1985); Y. Masumoto, M. Matsuura, S. Tarucha, and H. Okamoto, Phys.
Rev. B 32,4275 (1985).
R. Dingle, Confined Carrie1' Quantum States in Ultmthin Semiconductor Heterostructures, in
Festkorperprobleme XV, edited by H. J. Queisser (Vieweg, Braunschweig, 1975), p. 21.
S. Frisk, J.-L. Staehli, L. C. Andreani, A. Bosacchi, and S. Franchi, in Ref. [89], p. 183; B.
Rotelli, C. Arena, L. Tarricone, and C. Rigo, in Ref. [90], p. 315.
M. Shinada, and S. Sugano, J. Phys. Soc. Jpn. 21, 1936 (1966).
K. S. Chan, J. Phys. C, 20, 791 (1987).
M. Gurioli, J. Martinez-Pastor, M. Colocci, A. Bosacchi, S. Franchi, and 1. C. Andreani, Phys.
Rev. B 47, 15755 (1993).
P. Lefebvre, Ph. Christol, and H. Mathieu, in Ref. [90], p. 377; Phys. Rev. B 48, 17308 (1993).
C. Weisbuch, and B. Vinter, Quantum SemiconductoT StructuTes (Academic Press, 1991).
Lecture notes by J.-Y. Duboz, this volume.
1. C. West, and S. J. Eglash, Appl. Phys. Lett. 46, ll56 (1985).
J. Khurgin, Appl. Phys. Lett. 62, 1390 (1993).
See e.g. P. Von Allmen, PhD Thesis, Ecole Poly technique Federale, Lausanne, 1992 (unpublished).
Y.-C. Chang, and R. B. James, Phys. Rev. B 39, 12672 (1989).

[89] See e.g. Optics of Excitons in Confined Systems, edited by A. D'Andrea, R. Del Sole, R. Girlanda,
and A. Quattropani (Institute of Physics, 1992).
[90] See e.g. J. de Phys. IV, vo!' 3, Col!. C5, 1993 (Supp!. JP II, no. 10), Proceedings of the Third
International Conference on Optics of Excitons in Confined Systems, Montpellier, 1993.
[91] K. Cho, and M. Kawata, J. Phys. Soc. Jpn. 54,4431 (1985).
[92] A. D'Andrea, and R. Del Sole, Phys. Rev. B 41, 1413 (1990).
[93] R. C. Miller, D. A. Kleinman, W. T. Tsang, and A. C. Gossard, Phys. Rev. B 24, 1134 (1981).
[94] G. Bastard, E. E. Mendez, L. L. Chang, and E. Esaki, Phys. Rev. B 26, 1974 (1982).
[95] R. L. Greene, K. K. Bajaj, and D. E. Phelps, Phys. Rev. B 29, 1807 (1984).
[96] L. C. Andreani, A. D'Andrea, and R. Del Sole, Phys. Lett. A 168, 451 (1992).
[97] See e.g. P. Lefebvre, P. Christol, and H. Mathien, Phys. Rev. B 46, 13 603 (1992) and references
therein.
[98] M. Kumagai, and T. Takagahara, Phys. Rev. B 40, 12359 (1991).
[99] Y. Merle d'Aubigne, H. Mariette, N. Magnea, H. Tuffigo, R. T. Cox, G. Lentz, Le Si Dang, J.-L.
Pautrat, and A. Wasiela, J. Cryst. Growth 101, 650 (1990).
[100] G. E. W. Bauer, and T. Ando, Phys. Rev. B 38, 6015 (1988).
[101] L. C. Andreani, F. Bassani, and A. Pasquarello, in Symmetry in Nature, Quaderni della Scuola
Normale Superiore (Pisa, 1989), p. 19.
[102] R. Cingolani, and K. Ploog, Adv. Phys. 40, 535 (1991).
[103] S. Jorda, U. Rossler, and D. Broido, in Ref. [89], p. 49.
[104] B. S. Wang, and J. L. Birman, Phys. Rev. B 43, 12458 (1991).
[105] M. Berz, L. C. Andreani, E. F. Steigmeier, and F. I<. Reinhart, Solid State Commun. 80, 553
(1991).
[106] D. Frohlich, P. I(ohler, E. Meneses-Pacheco, G. Khitrova, and G. Weimann, in Ref. [89], p. 33.
[107] M. Nakayama, Solid State Commun. 55, 1053 (1985); M. Nakayama, and M. Matsuura, Surf.
Sci. 170, 641 (1986).
[108] V. M. Agranovich, and O. A. Dubovskii, JETP Lett. 3, 223 (1966).
[109] See also V. M. Agranovich, and S. Mukamel, Phys. Lett. A 147, 155 (1990) and references
therein.
[1l0] R. Kubo, J. Phys. Soc. Jpn. 12, 570 (1957).
[Ill] K. Cho, J. Phys. Soc. Jpn. 55,4113 (1986).
[112] I-I. Ishihara, and K. Cho, Phys. Rev. B 41, 1424 (1990).
[113] T. Mita, and N. Nagasawa, Solid State Commun. 44, 1003 (1982).
[114] Y. Chen, F. Bassani, J. Massies, C. Deparis, and G. Neu, Europhys. Lett. 14,483 (1991).
[115] H. Tuffigo, R. T. Cox, N. Magnea, Y. Merle d'Aubigne, and A. Million, Phys. Rev. 37, 4310
(1988); H. Tuffigo, PhD Thesis, Universite J. Fourier-Grenoble I, 1990 (unpublished).
[116] J. Kusano, Y. Segawa, M. Mihara, Y. Aoyagi, and S. Namba, Solid State Commun. 72, 215
(1989).
[117] A. Tredicucci, Y. Chen, F. Bassani, J. Massies, C. Deparis, and G. Neu, Phys. Rev. B 47,10348
(1993).
[118] R. N. Philp, and D. R. Tilley, Phys. Rev. B 44, 8170 (1991).
[119] F. Tassone, F. Bassani, and L. C. Andreani, Nuovo Cimento D 12, 1673 (1990).
[120] S. Jorda, U. Rossler, and D. Broido, Phys. Rev. B 48, 1669 (1993).
[121] M. Kohl, D. Heitmann, P. Grambow, and K. Ploog, Phys. Rev. B 37, 10927 (1988); 42, 2941
(1990).
[122] K. Ogawa, T. Katsuyama, and H. Nakamura, Phys. Rev. Lett. 64,796 (1990).
[123] L. C. Andreani, F. Tassone, and F. Bassani, Solid State Commun. 77, 641 (1991).
[124] E. L. Ivchenko, Fiz. Tverd. Tela 33, 2388 (1991) [Sov. Phys. Solid State 33, 1344 (1991)].
[125] F. Tassone, F. Bassani, and L. C. Andreani, Phys. Rev. B 45, 6023 (1992).
[126] E. L. Ivchenko, and A. V. I(avokin, Fiz. Tverd. Tela 34, 1815 (1992) [Sov. Phys. Solid State 34,
968 (1992)].
[127] R. H. Dicke, Phys. Rev. 93, 99 (1954).
[128] Ya. Aaviksoo, Ya. Lippmaa, and T. Reinot, Opt. Spectrosc. (USSR) 62,419 (1987).
[129] A. Sumi, J. Phys. Soc. Jpn. 50,2985 (1981).
[130] J. I(noester, Phys. Rev. Lett. 68,654 (1992).
[131] E. Hanamura, Phys. Rev. B 38, 1228 (1988).
[132] D. S. Citrin, Solid State Commun. 84, 281 (1992); Phys. Rev. B 47, 3832 (1993).
[133] B. Deveaud, F. Clerot, N. Roy, K. Satzke, B. Sermage, and D. S. Katzer, Phys. Rev. Lett. 67,
2355 (1991).
[134] B. Sermage, B. Deveaud, K. Satzke, F. Clerot, C. Dumas, N. Roy, D. S. Katzer, F. Mollot, R.
Planel, M. Berz, and J. L. Oudar, Super!. Microstr. 13,271 (1993).

111

[135]
[136]
[137]
[138]
[139]
[140]
[141]
[142]

J. Hegarty, L. Goldner, and M. D. Sturge, Phys. Rev. B 30,7346 (1984).


See also lecture notes by D. S. Citrin, this volume.
M. Colocci, M. Gurioli, and J. Martinez-Pastor, in Ref. [90], p. 3.
J. Feldmann, G. Peter, E. O. Gobel, P. Dawson, K. Moore, C. Foxon, and R. J. Elliott, Phys.
Rev. Lett. 59, 2337 (1987).
J. Martinez-Pastor, A. Vinattieri, L. Carraresi, M. Colocci, Ph. Roussignol, and G. Weimann,
Phys. Rev. B 47, 10456 (1993).
R. Eccleston, B. F. Feuerbacher, J. Kuhl, W. W. Ruhle, and K. Ploog, Phys. Rev. B 45,11403
(1992).
F. C. Spano, J. R. Kuklinski, and S. Mukamel, Phys. Rev. Lett. 65, 211 (1990).
D. S. Citrin, Phys. Rev. Lett. 69, 3393 (1992).

Lucio Claudio Andreani is research associate at the Department of Physics "A.


Volta" of the University of Pavia, where he conducts theoretical research on the optical
properties of mesoscopic semiconductor structures.

112

Vous aimerez peut-être aussi