PHYSICAL REVIEW E VOLUME 59, NUMBER 3 MARCH 1999

Density of states for a dielectric superlattice: TE polarization

I. Alvarado-Rodrı́guez,1,2,* P. Halevi,2 and J. J. Sánchez-Mondragón1,2
1
Centro de Investigaciones en Ingenierı́a y Ciencias Aplicadas, Universidad Autónoma del Estado de Morelos,
Avenida Universidad 1001, Colonia Chamilpa, Cuernavaca 62210, Mexico
2
Instituto Nacional de Astroı́sica, Optica y Electrónica, Apartado 51 Puebla, Puebla 72000, Mexico
~Received 12 August 1998!
We present a calculation of the band structure and the density of states ~DOS! for a dielectric one-
dimensional superlattice ~SL!. It is modeled by means of a periodic array of Dirac delta functions, character-
ized by the grating strength parameter g @Tocci et al., Phys. Rev. A 53, 2799 ~1996!#. The band structure for
TE or s-polarized modes is given by a simple, compact formula that reproduces well the qualitative features
exhibited by a real SL. We make use of equifrequency surfaces in wave-vector space—a concept similar to the
electron Fermi surfaces in solid state physics. This is helpful for deriving the DOS as a function of frequency
and the g parameter. The slope of the DOS exhibits discontinuities at all the edges of the band gaps. However,
the DOS is always finite, unlike the case that wave propagation is restricted to the SL axis ~where the DOS
vanishes in the band gap!. In fact, surprisingly, the DOS is actually enhanced relative to free space for all
values of the frequency and of g, and especially so at the lower edges of the band gaps. These results are
relevant to the spontaneous emission by an atom, or to dipole radiation in one-dimensional periodic structures.
@S1063-651X~99!11302-3#
PACS number~s!: 42.70.Qs 42.25Bs, 78.20.Bh

I. INTRODUCTION is not considered; thus our attention is limited to the TE case.

Our model can be viewed as a limiting case of very thin
Sine it was proposed that the emission of electromagnetic layers with a very large dielectric constant, positioned at
radiation can be modified by the environment @1,2#, there has equal distances. This is the Dirac d function model in which
been a lot of work in this respect. This work concerns the these thin layers are represented by a Dirac d function. It has
description of the modification of dipole radiation in several been used in Ref. @6# in the calculation of the power emis-
sion, considering only propagation of light parallel to the SL
environments like metallic cavities @3,4#, dielectric cavities
axis.
@5#, and superlattices ~SL’s! @6–8#. Such environments are
The model makes sense if the dielectric constant of the
inhomogeneous media, in which either the dielectric constant barriers is large enough and its thickness is small enough.
depends on position or else the medium is homogeneous but We derive a simple and compact dispersion relation that
bounded, giving rise to boundary conditions. In particular, gives rise to a band structure which is qualitatively the same
the periodic dielectric structures have earned a lot of atten- as that exhibited by a realistic SL. We have used the concept
tion because of the possibility of applications on low- of equifrequency surfaces to aid the calculation of the DOS.
threshold microlasers and, more recently, in relation to the Such surfaces arise from the dispersion relation and are the
advent of information technology @9#. optical equivalent of the Fermi surfaces of electrons in sol-
In the case of the SL structures, there are many studies ids. The results show an increment of the DOS relative to the
@10,11# on the determination and analysis of the band gaps, free space for all the frequencies and, especially, near the
the description of the bulk and surface modes, etc. Currently, lower band-gap edges. The DOS exhibits discontinuities in
there is much interest in the band structure of photonic crys- the slope at the band edges because of the change in the
tals @10#, that are two- and three-dimensional generalizations shape of the equifrequency surface. These results can be ex-
of the superlattice ~which may be considered as a special tended to a calculation of the power radiated by a dipole in
case!. this medium @13#.
The environmental effects of a periodic dielectric medium In Sec. II, we discuss the electromagnetic normal modes
have been described by the density of states ~DOS! for the for the Dirac d function model. We derive the corresponding
electromagnetic field in the medium @12,7#. The DOS is re- dispersion relation or the band structure and the TE field. We
lated to the transition rate by means of the Fermi golden rule. calculate the DOS in Sec. III using the equifrequency sur-
Here it is stated that the transition rate is proportional to the faces. These surfaces are the optical analogue of the Fermi
interaction dipole field times the DOS at the dipole transition surfaces for electrons in solids. The conclusions are pre-
frequency. This is valid to describe the power radiated in one sented in Sec. IV.
direction.
This paper deals with the calculation of the DOS for the II. NORMAL MODES
field in a one-dimensional SL. The case of TM polarization
A. Dirac d function model of a superlattice
A Dirac d function model of a SL is defined by the fol-
*Present address: Department of Electrical Engineering, Univer- lowing dependence of the dielectric constant on the position
sity of California at Los Angeles, Los Angeles, CA 90024-7594. @6#:

1063-651X/99/59~3!/3624~7!/$15.00 PRE 59 3624 ©1999 The American Physical Society

PRE 59 DENSITY OF STATES FOR A DIELECTRIC . . . 3625

where the dielectric constant e (r) is a function of the posi-

tion due to the inhomogeneity. In this equation, the Ekp (r)
functions represent the monochromatic solutions ~eigenvec-
tors! corresponding to the eigenfrequencies v kp . To ensure
that Ekp (r) are a complete set of orthonormal functions, they
have to fulfill the normalization and closure conditions as
stated in Refs. @6,14#.
As for any SL, the field can be expressed as a linear
combination of the two independent TE and TM polarization
modes. Each of these modes satisfies Eq. ~3!. Here we re-
strict our calculations of modes and DOS to the TE ~s!
modes, so that the mode index p, equal to TE, can be sup-
pressed. Because the dielectric constant is independent of y
and z, the fields must be proportional to exp @i(kyy1kzz)#. For
FIG. 1. A Dirac d superlattice. In this case, D→0 and our periodic structure the wave-vector component k x is given
e m → ` , while keeping the factor g5( e m 2 e 0 )D/d finite. The strati- by the Bloch vector k B which may be restricted to the first
fication direction is the x axis, and the period is denoted by d. Brillouin zone, namely, 2 p /d,k B < p /d. Thus the wave
vector is k5k B x̂1k y ŷ1k z ŷ. Moreover, by the Bloch-
` Floquet theorem @15,16# the amplitude of the wave is given
e ~ x ! 5 e 0 1gd (
n52`
d ~ x2nd ! . ~1! by a periodic function u k(x) which has the same period (d)
as the given system. Then the electric field is given by
This model can be obtained from Fig. 1 if the widths of
Ek~ r! 5u k B ~ x ! e ik B x e i ~ k y y1k z z ! êk ~4!
the ‘‘barriers’’ become vanishingly small (D→0), and, si-
multaneously, their dielectric constants increase beyond limit
( e m → ` ). To understand this approach, compare the inte- where êk is a unit vector with arbitrary direction in the yz
plane ~see Fig. 1!, and perpendicular to the wave vector k as
2d/2e (x)dx for the model given in Eq. ~1! and for the
grals * d/2
realistic SL consisting of alternating dielectric layers e o and well. Therefore, the polarization vector êk is
e m , whose widths are d2D and D, respectively. Then, if g
kz ky
in Eq. ~1! is given by êk52 ŷ1 ẑ52ŷ cos f 1ẑ sin f ~5!
ki ki
D
g5 ~ e m 2 e 0 ! , ~2! where we have defined k i as the projection of the wave vec-
d
tor on the yz plane, expressed as
then the integrals of e (x), taken for the two models just
described, both give ( e 0 1g)d. Thus taking e m @ e 0 and D k 2i 5k 2z 1k 2y , ~6!
!d while keeping g constant, we can give a physical mean-
ing to our model Eq. ~1!. It is expected to become realistic if and f is the angle that k forms with the z axis. Replacing x
the dielectric constant e m is very large and, at the same time, by x1d in Eq. ~4!, we obtain an alternative formulation of
the corresponding layers are very narrow. The g parameter is the Bloch theorem,
called the grating strength.
This consideration gives some justification to our model, Ek~ x1d,y,z ! 5u ~ x1d ! e ik B x e ik B d e i ~ k y y1k z z ! êk
so that its consequences, at least at the qualitative level, will
not significantly differ from those that may be obtained from 5e ik B d Ek~ x,y,z ! . ~7!
a realistic model. Moreover, the g parameter in Eq. ~1! can
be adjusted to yield realistic results within a limited fre- This means that the value of the function one period away
quency range. from any point is just a phase factor times the value of the
function at that point. This phase factor is given by the Bloch
B. TE modes wave vector. Substituting Eq. ~4! into Eq. ~3!, we obtain

F G
To start with, we have to solve the Maxwell wave equa-
tion for an inhomogeneous medium. For a given material d2 v 2k
E ~ x !1
2 k
e ~ x ! 2k 2y 2k 2z Ek~ x ! 50. ~8!
geometry, we can describe the field therein as a linear super- dx c2
position of the normal modes or eigenmodes. Each mode
may be labeled according to its wave vector k and polariza- The solution of this equation, within any two adjacent
tion index p. The modes for the electric field are then de- barriers x5nd , is the sum of two counterpropagating plane
fined by means of the Helmholtz equation waves in the x direction; this is reasonable as the medium is
stratified in this direction. Then
v kp
2
¹3¹3Ekp ~ r! 2 e ~ r! Ekp ~ r! 50 ~3!
c2 Ek~ x ! 5 @ Ae iKx 1Be 2iKx # êk, ~9!
3626 I. ALVARADO-RODRÍGUEZ et al. PRE 59

where K is given by the dispersion relation in the back- C. Boundary conditions

ground medium, which occupies the space between the bar- Because of the particular model in Eq. ~1! that we are
riers, considering, the TE boundary conditions have to be formu-
lated in a rather different way than for the usual problem of
v 2k propagation in a periodic medium @15#. Our derivation is
K 25 2
e 0 2k 2i . ~10!
c analogous to the scalar-wave calculation of Dowling and
Bowden @6#. Faraday’s law can be applied, as usual, to any
If Eq. ~9! represents the solution in the region n50, we loop centered at a barrier, x5nd. Because the magnetic field
can apply the Bloch theorem @Eq. ~7!# n times in order to must be finite, the electric field is continuous across the bar-
obtain the solution in the nth region, riers. Thus the first boundary condition is

E~kn ! ~ x ! 5e ink B d @ Ae iK ~ x2nd ! 1Be 2iK ~ x2nd ! # êk , ~11! E~kn ! ~ nd ! 5E~kn21 ! ~ nd ! ~13!

where
The first order derivative, however, is not continuous. To
nd,x, ~ n11 ! d. ~12! set the boundary condition for the first order derivative, we
integrate the Helmholtz equation ~8! over a small region near
Equation ~11! can be rearranged to satisfy Eq. ~4!. x5nd, using Eq. ~1!, that is

E nd1 e

nd2 e
dx
d2
dx
E ~ x ! 52K 2
2 k E nd1 e

nd2 e
dx Ek~ x ! 2
v 2k
c 2 E nd1 e

nd2 e
dx gd (n d ~ x2nd ! Ek~ x ! . ~14!

Here the integration interval is small enough so that the function can be taken as a constant which is the function evaluated at
x5nd. Then the integration results in

d ~n! d v 2k
Ek ~ nd ! 2 E~kn21 ! ~ nd ! 52 2 gdEk~ nd ! , ~15!
dx dx c

which is the second boundary condition.

D. Solution of the eigenvalue problem; the band structure

Substitution of Eq. ~11! into Eqs. ~13! and ~15! leads to the homogeneous system

M SD
A
B
50. ~16!

The characteristic matrix M of the system is defined as

M5 S 12exp@ i ~ K2k B ! d #
12exp@ i ~ K2k B ! d # 2i a ~ v k ,K !
12exp@ 2i ~ K1k B ! d #
211exp@ 2i ~ K1k B ! d # 2i a ~ v k ,K !
D , ~17!

where we have defined a ( v k ,K) as @6# for the case k i 50, namely, with K5( v /c) Ae 0 . K can be
either real or purely imaginary, since both types of solutions
gd v 2k fulfill the wave equation and the boundary conditions. Now
a ~ v k ,K ! 5 . ~18! we can see from Eq. ~19! that u f (K; v k) u can be greater than
2c 2 K
one for certain ranges of values of K, giving rise to complex
To avoid the trivial solution, we have that det M50. Then solutions for the Bloch wave vector k B . In these cases, there
we find the dispersion relation in terms of the Bloch wave are band gaps in which, for an ~infinite! periodic medium, no
vector k B and K( v k ,k i ), as defined in Eq. ~10!, solutions exist for the field. We plot this band structure in
Fig. 2 for a grating strength g50.1. In this plot, the shad-
cos~ k B d ! 5cos~ Kd ! 2 a ~ v k ,K ! sin~ Kd ! [ f ~ K; v k! . owed areas indicate the allowed regions. From this figure it
~19! can be seen that the Dirac d function model actually repro-
duces all the qualitative features of a realistic band structure
This equation gives, implicitly, the eigenfrequencies v k @15#. As the grating strength g increases, the band gaps be-
as a function of the wave-vector components k B and k i . This come wider. Nevertheless, this model has the interesting par-
dispersion relation was obtained by Dowling and Bowden ticularity that the upper edges of all the band gaps are inde-
PRE 59 DENSITY OF STATES FOR A DIELECTRIC . . . 3627

surface v (k B ,k y ,k z )5const simply by rotating the curves

around the k B axis.
To describe the surfaces depicted in Fig. 3, take a look at
the band structure in Fig. 2. In this figure, horizontal lines are
drawn at the frequency values selected in Figs. 3~a!, 3~b!,
3~c!, and 3~d!. Line ~a! lies entirely within the lowest al-
lowed band, so that the equifrequency surface is closed. Line
~b!, corresponding to a greater frequency, lies in part in the
first band gap and in part in the first band. This produces an
interruption in the corresponding equifrequency surface for
values k i d,1, as no solutions for the field exist therein. As
the frequency keeps increasing, the line v 5const cuts
through more and more passbands, so that the equifrequency
surfaces split into several sections. This can be seen from
Figs. 3~c! and 3~d! with the aid of the corresponding fre-
quency lines transversing the band structure. These surfaces
grow in size with the frequency, as we can see from the k i
FIG. 2. Band structure for the superlattice in Fig. 1, obtained range for each value of the frequency.
using the dispersion relation given in Eq. ~19! for e 0 51 and g Notice that k i can exceed the value of Ae 0 v /c. The
50.1. The width of the forbidden bands increases with g. This
modes located in the part of the surface where k i
means that, with increasing g, the upper band edges are lowered
~the lower band edges are ‘‘pinned’’ at integer multiples of p for
. Ae 0 v k /c correspond to the case of imaginary K, and
this model!. The free-space dispersion relation is recovered in the hence to evanescent solutions for E(n) k (x), Eq. ~21!, in the
limit g→0. This band structure is very similar to that obtained for a regions between the barriers.
realistic superlattice. To perform the DOS calculation, it is required to use its
formal definition which is the number of available photon
pendent of g. In fact, it is easy to show from Eqs. ~19! and modes per unit frequency range. We then construct two
~10! ~with k i 50) that these edges are given by v d/c5n p equifrequency surfaces, namely, v (k B ,k y ,k z )5 v and
(n51,2, . . . ). v (k B ,k y ,k z )5 v 1D v , where v is an arbitrary value of the
Using Eqs. ~16! and ~17!, we relate the coefficients A and frequency and D v is a small increment. We calculate the
B as volume therein, and divide it by the volume occupied by a
single mode. Then the differential volume element in k space
within the surfaces is given by ~see Fig. 4!
12e i ~ K2k B ! d
B52 A, ~20!
12e 2i ~ K1k B ! d DV k 52 p k i D k t D k n . ~22!

Here D k t 5 @ (Dk B ) 2 1(Dk i ) 2 # 1/2 is a differential segment

and the total field, from Eqs. ~11! and ~20!, may be expressed
parallel to the equifrequency surface, and D k n is the separa-
as
tion of the two surfaces that corresponds to the frequency
increment. In order to obtain Eq. ~22!, it is necessary to
2iAe ink B d rotate the area element D k t D k n around the k i axis. Now
E~kn ! ~ x ! 5 „sin K ~ x2nd ! D k n is perpendicular to the surface, and, hence, by defini-
12e 2i ~ K1k B ! d
tion,
2e 2ik B d sin K @ x2 ~ n11 ! d # …êk . ~21!
Dv
D k n5 . ~23!
The constant A is arbitrary. u ¹ kv u

Substituting Eq. ~23! into Eq. ~22!, and integrating over the
III. DENSITY OF STATES equifrequency surface, we have that the total phase-space
volume contributing to the frequency range ( v , v 1d v ) is
We calculate the photon DOS for our SL model for the

E E
TE polarization modes. In this case we have to consider that ki
we have continuous electromagnetic bands separated by the dV k 52 p d v dkt , ~24!
band gaps forbidden for propagation. vk vk u ¹ kv u

We study the surfaces of constant frequency ~equifre-

where we take the limit of infinitesimal increments. The
quency surfaces! in k space which are constructed with the
number of modes within the range ( v , v 1d v ) is obtained
aid of the dispersion relation, Eq. ~19!. Similar surfaces for
by dividing the volume calculated in Eq. ~24! by the volume
the electron energy eigenvalues are known in solid state
corresponding to one mode, (2 p ) 3 /V, where V is the volume
physics as Fermi surfaces @16#, a concept that is also useful
of the system. This yields
for electromagnetic propagation and optical properties. In

FE G
Fig. 3 several v (k B ,k y ,k z )5const curves are sketched in a
V ki
two-dimensional plot. Due to the azimuthal symmetry of the dN ~ v ! 5 d k t d v [D ~ v ! d v . ~25!
problem, we can obtain the three-dimensional equifrequency 4p 2 v ku ¹ kv u
3628 I. ALVARADO-RODRÍGUEZ et al. PRE 59

FIG. 3. Four cross sections (k z 50) of equifrequency surfaces v (k B ,k y ,k z )5 v 5const for g50.1. Due to the azimutal symmetry, the
surfaces are formed by rotating the curves around the k B axis. The dashed lines denote u k y u 5 v /c, so that the region between ~outside! these
lines is characterized by real ~imaginary! K values. ~a! Equifrequency surface for v d/c51. ~b! v d/c53; the region k i d,1 is excluded from
the surface because such values of k i d lie in the forbidden band ~see Fig. 2!. ~c! v d/c54. ~d! v d/c56; notice the further splitting of the
surfaces due to the apparition of more forbidden bands ~the regions k i d,2 and 5,k i d,6 are now excluded!.

Because v is a function of k, we can write

]v ]v ]v
¹kv 5 x̂1 ŷ1 ẑ. ~26!
]kB ]ky ]kz

By the use of Eqs. ~19!, ~10!, and ~6! we have that

u ¹kv u 5
c2
U K sin k B d
v k ~ g1 e 0 ! sin Kd1 ~ g v d/2c K ! e 0 @ cos Kd2 ~ 1/Kd ! sin Kd #
2 2
ky
x̂1 ŷ1
e0
kz
e 0-
ẑ . U ~27!

Then, the final expression for the DOS is

D~ v !5
V e 0v
4p c2 2 E
vk
dkt
k i F„v ,K ~ v ,k i ! …
$ K @ sin Kd1 a ~ v ,K ! sin 2Kd2 a 2 ~ v ,K ! sin2 Kd # 1k 2i F„v ,K ~ v ,k i ! …% 1/2
2 2
, ~28!

where we have defined the function F„v ,K( v ,k i )… as to note that, even though the group velocity u ¹ kv u vanishes
for on-axis propagation (k i 50), at the band edges the den-
F„v ,K ~ v ,k i ! …[ ~ g/ e 0 11 ! sin Kd1 a ~ v ,K ! sity of states remains finite. For other values of k i always

F G
u ¹ kv u Þ0. Nevertheless for frequencies at the band edges,
1 care should be taken when performing a calculation. Indeed,
3 cos Kd2 sin Kd . ~29!
Kd we took the limits of frequencies approaching these edges,
this process giving rise to convergence.
For frequencies above the first band, the gaps in the We plot the DOS @Eq. ~28!# for different values of g in
equifrequency surfaces must be carefully avoided @see Figs. Fig. 5. The inset in this figure shows a plot for a larger range
3~b!–3~d!# when performing the integration. It is important of frequencies. It is seen that there are sharp discontinuities
PRE 59 DENSITY OF STATES FOR A DIELECTRIC . . . 3629

in the first order derivative at certain values of the frequency.

These discontinuities arise from the band gaps of the band
structure. For g50.1 the band structure is given in Fig. 2.
This value of g(!1) corresponds to a weak modulation of
the dielectric constant, so that the D( v ) curve lies quite
close to the free-space DOS ~dashed!. Nevertheless, the slope
dD( v )/d v suffers an abrupt decrease for d v /c52.858, and
then an increase for d v /c5 p . These values of the frequency
are, precisely, the lower and upper edges of the first band gap
for k i 50; see Fig. 2. A similar decrease and subsequent
increase of the slope reoccurs for the normalized frequencies
5.75 and 2 p ~see the inset!. These values are just the bounds
of the second band gap. Of course, because we have allowed
for propagation in an arbitrary direction—not merely along FIG. 4. Partial cross sections in the k z 50 plane of two equifre-
the axis of the SL—there are allowed modes for every value quency surfaces denoted by v and v 1D v . It is seen that (D k t ) 2
of v , so that the DOS never vanishes. 5(Dk B ) 2 1(Dk i ) 2 . Using the azimutal symmetry, the total volume
For increasing values of the grating strength g the D( v ) element is DV k 52 p k i D k t D k n , and has the shape of a ring of a
curves move further away from the free-space curve. In ad- rectangular cross section.
dition, the discontinuities become longer, simply because the
gaps increase with g. The most eyecatching effect is that to ( e 0 1g) 3/2. For g50.9, this is an enhancement by a factor
sharp peaks appear at the lower edges of the gaps, and are, of about 2.6 with respect to vacuum. However, no need to
however, resolved only for substantial values of g. This is to say, the structure does matter greatly, as manifest especially
say that, not surprisingly, the DOS decreases every time that at the band edges. Thus, a reduction of the DOS with respect
v crosses the threshold between a band and a gap. Also to free space does not occur; this despite the fact that, for
notice that the increase in the slope occurs at d v /c5 p for one-dimensional propagation, the DOS actually vanishes for
all g. This is because the upper edges of the band gaps are frequencies within the one-dimensional gaps.
independent of g for our Dirac d function model, as noted
above. IV. CONCLUSION
In Fig. 5 an increment of the DOS relative to free space is
present for all frequencies, and is particularly pronounced for We have presented a calculation of the modes and of the
frequencies near the lower edges of the band gaps. The en- density of states for a periodic dielectric medium. The cal-
hancement increases rapidly with the grating strength param- culation is restricted to the TE polarization mode. We made
eter g. This can be understood qualitatively from the fact use of a Dirac d function model, characterized by the grating
that the DOS of an inlimited, homogeneous dielectric is strength parameter g, as an approximation to the real SL.
( v 2 /2p 2 c 3 ) e 3/2 per unit volume. For our model @Eq. ~1!#, the A solution of the inhomogeneous wave equation was pre-
average dielectric constant of the superlattice is ^ e (x) & 5 e 0 sented with the aid of the Bloch theorem. A very simple,
1g. Therefore, ignoring the structure of the superlattice, one compact formula was derived for the dispersion relation, and
could expect an enhancement of the DOS that is proportional the resulting band structure is found to be qualitatively simi-

FIG. 5. Density of states as a function of the reduced frequency v d/c for several values of the grating strength g. The inset extends the
frequency range to the interval (0,2p ). Here we can see the discontinuities in the slope for frequencies that define the band edges. Notice
that the DOS is enhanced for all the values of v and g, and especially at the lower edges of the band gaps.
3630 I. ALVARADO-RODRÍGUEZ et al. PRE 59

lar to that obtained for a realistic SL. Hence, the Dirac d propagation in all directions in space. As a consequence, the
function model is expected to give reasonable results for the DOS never vanishes, as occurs if the wave is restricted to
DOS. propagate along the axis of the SL @6#. Nevertheless, intu-
The DOS was calculated with the aid of the equifre- itively, one would have expected at least some reduction of
quency surfaces which are similar to the Fermi surfaces for the DOS associated with the partial gaps. The fact that this
electrons in solids. Within any one band the DOS increases can never happen is rather surprising. These conclusions
monotonously with the frequency, reaching a sharp and nar- should be helpful for understanding spontaneous emission
row maximum at the upper band edge ~for k i 50). Also, ~or dipole radiation! in laminated structures. A similar analy-
upon crossing from a stop band into a pass band ~again for sis of the DOS for the TM polarization mode is clearly de-
k i 50) there is an abrupt increase in the DOS. The results sirable, and will be undertaken in the future.
exhibit an enhancement of the DOS, in comparison to free
space, for all values of the frequency. This enhancement in-
creases with the grating strength g @as can be expected from ACKNOWLEDGMENTS
the fact that ^ e (x) & is proportional to ( e 0 1g) 3/2)#, and is
especially pronounced at the lower band edges of the band I.A.R. thanks the CONACyT and SNI for financial sup-
gaps. This behavior is the result of our taking into account port.

@1# E. M. Purcell, Phys. Rev. 69, 618 ~1946!. @8# A. Kamli, M. Babiker, A. Al-Hairy, and N. Enfati, Phys. Rev.
@2# D. Kleppner, Phys. Rev. Lett. 47, 233 ~1981!. A 55, 1454 ~1997!.
@3# P. W. Milloni and P. L. Knight, Opt. Commun. 9, 119 ~1973!; @9# J. D. Joannopoulos, Pierre R. Villenueve, and Shanhui Fan,
A. O. Barut and J. P. Dowling, Phys. Rev. A 36, 649 ~1987!; I. Nature ~London! 386, 143 ~1997!.
Abram and G. Bourdon, ibid. 54, 3476 ~1996!; F. de Martini, @10# J. D. Joannopoulos, R. D. Meade, and J. J. Winn, Photonic
M. Marrocco, P. Mataloni, L. Crescentini, and R. Loudon, Crystals: Molding the Flow of Light ~Princeton University
ibid. 43, 2480 ~1991!; M. A. Rippin and P. L. Knight, J. Mod. Press, Princeton, 1995!.
Opt. 43, 807 ~1996!. @11# F. Ramos-Mendieta and P. Halevi, Opt. Commun. 129, 1
@4# I. Alvardo-Rodrı́guez, P. Halevi, and J. J. Sańchez- ~1996!; J. Opt. Soc. Am. B 14, 370 ~1997!, and references
Mondragón, Rev. Mex. Fis. 44, 268 ~1998!. therein.
@5# Y. Yamamoto, S. Machida, Y. Horikoshi, and K. Igeta, Opt.
@12# Michael D. Tocci, Michael Scalora, Mark J. Bloemer, J. P.
Commun. 50, 337 ~1991!; R. J. Ram and D. I. Babic, IEEE J.
Dowling, and C. M. Bowden, Phys. Rev. A 53, 2799 ~1996!.
Quantum Electron. 31, 2 ~1995!; H. Rigneault and S. Mon-
@13# I. Alvarado-Rodrı́guez and P. Halevi ~unpublished!.
neret, Phys. Rev. A 54, 2356 ~1996!; S. Ciancaleoni, P. Mata-
@14# R. J. Glauber and M. Lewenstein, Phys. Rev. A 43, 467
loni, O. Jedrkiewicz, and F. De Martini, J. Opt. Soc. Am. B 14,
~1991!.
1556 ~1997!.
@6# J. P. Dowling and C. M. Bowden, Phys. Rev. A 46, 612 @15# P. Yeh, Optical Waves in Layered Media ~Wiley, New York,
~1992!. 1988!.
@7# Toshio Suzuki and Paul K. L. Yu, J. Opt. Soc. Am. B 12, 570 @16# C. Kittel, Introduction to Solid State Physics ~Wiley, New
~1995!. York, 1976!.