Shabanov2009 - BIC in Gratings

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 46

November 13, 2009 14:18 WSPC/140-IJMPB 05447

International Journal of Modern Physics B


Vol. 23, No. 27 (2009) 5191–5236
c World Scientific Publishing Company

RESONANT LIGHT SCATTERING AND HIGHER


HARMONIC GENERATION BY PERIODIC
SUBWAVELENGTH ARRAYS
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

SERGEI V. SHABANOV
Department of Mathematics, University of Florida,
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

Gainesville FL 32611, USA


shabanov@ufl.edu

Received 26 October 2009

Scattering of light on periodic subwavelength arrays is studied in the framework of the


resonant scattering theory. With various examples of periodic structures it is demon-
strated that: (i) an enhanced reflectance or transmittance is associated with the exis-
tence of trapped modes (quasi-stationary modes of light confined in the vicinity of the
scattering structure); (ii) scattering structures may have trapped modes due to peculiar-
ities their geometry (geometrical modes) and the dispersive properties of their material
(material modes); a practical criterion based on the scaling symmetry of Maxwell’s equa-
tions is proposed to distinguish them; (iii) the trapped mode field can be significantly
amplified, as compared to the incident wave amplitude, in some regions of the struc-
ture; (iv) the amplification increases with increasing the lifetime of the trapped mode;
(v) this effect can be used to enhance nonlinear optical effects (a resonant higher har-
monic generation is studied in detail as an example). The theory of coupled resonances
is developed and used to prove that there exist bound states of light in the radiation
continuum (resonances with the vanishing width) in periodic arrays. The bound states
are neither modes in metal cavities nor modes in photonic crystal defects. Structures
supporting the bound states of light can be used to enhance and control nonlinear optical
effects in subwavelength periodic arrays.

Keywords: Resonant scattering of light; enhanced transmittance; subwavelength periodic


arrays; gratings; bound states in the radiation continuum; field amplification; nonlinear
effects in photonics.

1. Introduction
The purpose of the present review is to analyze scattering of light on periodic sub-
wavelength arrays from the point of view of the resonant scattering theory, the
Breit–Wigner theory, in particular. In the author’s view, this approach, although
sometimes being too simplistic as compared to rigorous methods of the formal scat-
tering theory, provides a qualitatively correct picture of the phenomena like, e.g., the
enhanced light transmittance or reflectance and, in some case, gives even their accu-
rate quantitative explanations. One of the greatly desired goals of nanophotonics is
to create a technology for all-optical data processing. This quest ultimately includes

5191
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5192 S. V. Shabanov

the problem of controlling nonlinear optical effects in photonic devices. A possible


approach to this problem is proposed here. It is based on a controlled resonant am-
plification of the electromagnetic field within the scattering structure. The latter
turns out to be feasible due to a novel, recently predicted, phenomenon: the exis-
tence of bound states of light in the radiation continuum (the resonances with the
vanishing width). A proof of the concept is done by a study of a resonant generation
of higher harmonics by a periodic array of dielectric cylinders. Whenever possible,
mathematical technicalities are avoided — the reader is referred to the literature
— and results of numerical and experimental studies are used instead.
Diffraction properties of periodic gratings first attracted attention after obser-
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

vations of anomalous reflective properties.1–3 The Wood anomalies were interpreted


later by Lord Rayleigh and Fano.4–6 Even though the very existence of evanescent
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

electromagnetic modes was already known,7–10 it took some time to realize that
the anomalies observed were related to the excitation of evanescent electromagnetic
modes bound to the grating.6 A recent revitalization of interest to the scattering
properties of periodic two dimensional arrays is due to observations of the enhanced
(extraordinary) light transmittance of periodically perforated thin metal films at
the wavelength slightly larger than the period of the structure.11 It appears that the
film transmits more light (per hole) than Bethe’s theory 12 (for a single hole diffrac-
tion) predicts. Thanks to recent technological advances in manufacturing periodic
nanostructures,13 the enhanced transmittance has been observed and extensively
studied by many researchers. The phenomenon itself is not new to engineers and mi-
crowave antenna designers. A nearly 100% transmittance of periodically perforated
metal films in the microwave range has been used in various applications such as
filtering, etc.14–16 Due to the scaling invariance of Maxwell’s, equations it is natural
to expect the same scattering properties when the wavelength and structure dimen-
sions are scaled by the same factor. The scaling symmetry breaks when the structure
material becomes dispersive, which is the case for metals and dielectrics in the op-
tical frequency range. The incident light can excite polariton waves in dispersive
dielectrics and (surface) plasmon-polariton waves in metals.17,18 These excitations
affect the light scattering (in addition to the geometry). The latter has sparked an
intensive debate in the literature about the mechanism of the enhanced light trans-
mittance. Some researchers insist on the plasmon-polariton mechanism,19–24 some
emphasize the dynamical light diffraction where the structure geometry plays the
dominant role.25–28 Both the theories have been applied to interpret recent experi-
ments with a reasonable success,29–31 which suggests that they are, perhaps, more
complementary to one another rather than competitive.32,33 A further discussion
of this matter can be found in a recent review.34
The scattering theory of electromagnetic waves is well-developed and has been
applied to periodic subwavelength arrays. It is based on the Green function formal-
ism35–42 (see Refs. 43 and 44 for recent applications in photonics, and Ref. 45 for
a rigorous scattering theory formalism). Although resonant properties of periodic
subwavelength arrays can be deduced from the general scattering theory formalism,
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5193

a simpler (approximate) treatment based on the Breit–Wigner resonant scattering


theory is often applicable by making use of a close analogy between quantum me-
chanics and Maxwell’s theory. The simplicity of the theory allows one to make a
correct qualitative and, in many cases, quantitative analysis of scattering proper-
ties of photonic structures. One of its recent applications is a simple model for
omnidirectional absorption in nanostructured metals.46
The Breit–Wigner theory asserts that a sudden increase of the scattering cross-
section near a particular frequency is due to the existence of a quasi-stationary
(standing) wave localized in the vicinity of the scatterer. In electrodynamics, such
a wave is a trapped mode of light confined by the scattering structure; it is long-
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

lived and decays slowly by emitting a nearly monochromatic radiation (Secs. 2


by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

and 3). A scattering structure may have trapped modes due to both its geometry
and dispersive properties of its material. Numerical and experimental data show
that geometrical trapped modes are scale invariant (when all dimensions and the
wavelength are scaled by the same factor), while the material ones are not (Sec. 4).
In particular, the scaling symmetry of the transmittance spectra holds for thin
metal films with subwavelength periodic hole arrays.47 The implications of this
fact for the enhanced transmittance mechanism are discussed. In Sec. 5, the Breit–
Wigner theory is used to develop a simple theory of coupled trapped modes. One
of the consequences of this theory is a mechanism to control the lifetime of trapped
modes. In particular, it is proved that there is a (quantized) set of distances between
two periodic planar structures, each of which supports a trapped mode with a
finite lifetime, such that the combined structure supports trapped modes that never
decay; these are bound states in the radiation continuum in Maxwell’s theory.48
They are a new kind of standing waves different from those existing in metal cavities
or photonic crystal defects. Consequently, by varying the distance in the vicinity
of one of the critical values, the trapped mode can be stabilized as desired. The
distance between the planar structures to hold a bound state of light may even be
less than the wavelength.
In the formal scattering theory, such bound states correspond to resonances with
the vanishing width. A similar phenomenon was first predicted by von Neumann
and Wigner in quantum theory49 and discovered much later in atomic physics.50
The electromagnetic field of the trapped modes appears to be significantly en-
hanced in some regions of the scattering structure as compared with the amplitude
of the incident wave that excites the modes. When coupled to an optical nonlinear
material, this field may generate strong nonlinear optical effects even if the nonlin-
ear medium parameters are small. The field enhancement increases with increasing
lifetime of the mode. Therefore the above mechanism to stabilize trapped modes
may be used to control the field amplification (via controlling the trapped mode
lifetime) and, hence, nonlinear optical effects in photonic structures. A simple il-
lustration of this idea is given in Sec. 6. With an example of a periodic (double)
array of dielectric thin cylinders, it is shown that by tuning the resonance position
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5194 S. V. Shabanov

and width (in accord with the developed theory of coupled trapped modes), one can
achieve a far more efficient generation of higher harmonics than by the conventional
way, while the thickness of the structure may even be less than the wave length of
the fundamental (incident) wave.

2. Resonances in Electromagnetic Scattering


2.1. Array of dielectric cylinders
A simple system to illustrate the Breit–Wigner theory is an infinite periodic set of
parallel dielectric cylinders suspended in the vacuum,51–53 which is shown in the
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

left panel of Fig. 1. The dielectric material is not dispersive. The cylinders are thin,
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

meaning that both the incident light wavelength and period Dg are much larger
than the cylinder radius. The system is practically transparent for the incident
light. However, something interesting happens when the incident light wavelength
λ = 2π/k is close to k = |k| = |kx − nG| where |n| = 1, 2, . . . , and G = 2π/Dg is the
reciprocal lattice vector. Consider the simplest case of the normal incidence, k x = 0,
with only one scattering channel open, k < G (λ > Dg ; the zero-order diffraction).
The incident wave is linearly polarized along the cylinders. The reflectance is com-
puted numerically and shown in the right panel of Fig. 1 as a function of the
wavelength in units of the period Dg for two values of the dielectric constant, ε = 2
and ε = 4 (the red and blue curves, respectively).51 For a wavelength slightly larger
than Dg , the array becomes a perfect mirror. As the wavelength increases, the
reflectance quickly drops practically to zero, and the system becomes completely
transparent. Such a rapid change of the scattering cross-section is called a scattering

Dg
E
0 z
k
H

Fig. 1. (Color on line) Left: A periodic array of parallel dielectric cylinders. The cylinders are
parallel to the y-axis, their centers are in the plane z = 0, and Dg is the array period. The
incident radiation is linearly polarized with the electric field vector being parallel to the cylinders.
It propagates in the direction of the wave vector k. Right: Reflectance of the array as a function
of the wavelength in units of Dg . Normal incidence kx = 0. The dielectric constant ε = 2: the red
curve (a narrow resonance); ε = 4: the blue curve (a broad resonance).
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5195

resonance. In mechanical systems, resonances occur if there are internal degrees of


freedom of the scattering system that can be excited by incident particles. In quan-
tum mechanics, they are related to excitations of quasi-stationary states of the
scattering system. Such states, in contrast to stationary or bound states, decay
with time (e.g., excited states of atoms; they decay by emitting photons because
of the coupling of electrons to vacuum fluctuations of the quantum electomagnetic
field). Continuing this analogy, one might ask what causes such a sharp resonance
in the studied electromagnetic system.
To answer this question, the following numerical experiment is carried out. A
Gaussian pulse with the duration of 25 fs and the carrier frequency close to the
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

resonant frequency of the array impinges the structure. The pulse is propagated
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

through the array. The electric field as a function of time and z at x = 0 is shown
in the left panel of Fig. 2. The array is at z = 0. The light lines of the incoming,
transmitted, and reflected pulses are clearly visible. As expected, most of the pulse
energy is transmitted because the array is very transparent. An interesting feature
is a bright strip at z = 0. It indicates that a part of the pulse energy is trapped by
the array and remains in it long after the reflected and transmitted pulses are gone.
The pulse interacts with the array for about 25 fs (the pulse duration), while the
80 100 120 140 160
time (fs)

transmitted field (arb. units)

H = 2
60
40

H = 4
20

H=2
R / Dg = 0.1
0

-8 -6 -4 -2 0 2 4 6 8
-8 -6 -4 -2 0 2 4 6 8
z coordinatecol(units of Dg)
0 50 100 150 200 250
Min Max time (fs)
Fig. 2. (Color on line) Left: Scattering of an electromagnetic wave packet on the array of cylinders.
The magnitude of the electric field as a function of time and z at x = 0. At the initial time, the
pulse is localized at z = −4Dg and has the duration of 25 fs. The array is at z = 0. Right: The
electric field as a function of time at a fixed z = 4Dg . For t > 65 fs, only the radiation emitted
by the trapped mode is observed. For the narrow resonance shown in Fig. 1 (right) (ε = 2), the
trapped mode radiation is extremely long lasting and practically monochromatic. The frequency
coincides with the scattering resonance frequency. The broader resonance (ε = 4) corresponds to
a shorter-lived trapped mode.
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5196 S. V. Shabanov

trapped mode lives much longer than that. The brightness of the strip decreases with
time, which means that the trapped mode decays. The decay radiation is visible in
Fig. 2(left) on both sides of the array.
To find the spectral content of the decay radiation, the electric field at a fixed
positive z is measured as a function of time. As one might anticipate from Fig. 2(left)
looking, e.g., along the line z = 4Dg , there should be no signal until the transmit-
ted pulse arrives. After it passes the detector, one should see a long lasting decay
radiation of the trapped mode. The actual results are shown in the right panel of
Fig. 2. In the case of the narrow resonance (ε = 2), the trapped mode radiation
is long lasting and nearly monochromatic. Its frequency coincides with the reso-
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

nant frequency. For the broader resonance (ε = 4), the amplitude of the trapped
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

mode radiation decreases faster. Thus, the scattering resonance width is inversely
proportional to the lifetime of the corresponding trapped mode.
These results provide compelling evidence that the scattering resonance of the
array is associated with a quasi-stationary state or trapped mode of the electromag-
netic field confined by the scattering structure. This observation relates a particu-
lar property of the far field (scattering resonances) to a property of the near-field
(trapped modes).

2.2. The Breit–Wigner formula for the array of cylinders


The scattering problem for the array of cylinders can be solved in the small radius
approximation, kR  1 where R is the cylinder radius.52,53,74 There are many ways
to do so. In electrodynamics of structured interfaces, various perturbative methods
have been developed54–56 as well as the differential theory of gratings.57–59 There are
also extensive studies of various numerical methods based on the perturbation60–65
and differential66–69 theories applied to near-field optics. In particular, the theory
of scattering of electromagnetic waves on small particles is most suitable70 (in the
formal scattering theory, this is a solution of the Lippmann–Schwinger equation71,72
in a Born approximation73 for multiple scatterers). To illustrate the relation be-
tween the Breit–Wigner theory and the exact solution of the scattering problem and
thereby to get a better understanding of the range of validity of the former, the
Breit–Wigner formula for a resonant scattering amplitude is derived from the exact
solution. It approximates the resonant curves shown in Fig. 1(right) extremely well
near the resonance position, especially for narrow isolated resonances.
Let r be a position vector with components x and z and nr0 be positions of
cylinders, n = 0, ±1, ±2, . . . . Let E(r, t) = e−iωt E(r) be the electric field (ω is the
incident wave frequency). Then the scattering problem solution can be written as
a superposition of the incident wave and scattering fields of individual cylinders70 :
X
E(r) = eik·r + c0 eink·r0 E0+ (r − nr0 ) ,
n
X (2.1)
c0 = [1 − Σ̂0 ]−1 , Σ̂0 = e−ink·r0 E0+ (nr0 ) ,
n6=0
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5197

where E0+ (r) is a solution of the scattering problem for a single cylinder centered
at the origin; it can be found, e.g., in Ref. 75. The constant c0 determines the
field on the cylinders. By the Bloch theorem, E(jr0 ) = eijk·r0 E(0), and E(0) =
c0 [1 + E0+ (0)] = 2c0 because E0+ (0) = 1 (the incident wave has a unit amplitude).
Let all coordinates be measured in units of Dg and p = k/G be the wave vector in
units of G. The solution (2.1) can be written in the standard form52,53,74
X X
E = e2πip·r + El+ + Emev
; (2.2)
l m6=l

as the sum of three terms: the incident wave, scattered waves, and evanescent (near)
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

field, respectively. Depending on the wavelength of the incident radiation, there


are several diffraction channels open for the scattered wave. They are enumerated
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

by an integer l; its range is specified by the condition that the z-component of


the scattered wave vector pz,l = [p2 − (l − px )2 ]1/2 is real. Accordingly, closed
channels are labeled by an integer m whose range is determined by the condition
that αm = [(m − px )2 − p2 ]1/2 is real. The scattered field in each open diffraction
channel is
(2πpz,l )−1
El+ = Rl (p, px )e2πi[x(px −l)+|z|pz,l ] , Rl (p, px ) = , (2.3)
(S0 − 1)−1 − Σ̂0
where S0 = e2iδs , δs = (ε − 1)(kR/2)2 is the scattering phase of a single cylinder
in the s-wave approximation (this approximations is justified if kR  1), and
!
1 1 i 1 X 1 1
Σ̂0 = − + [γ + ln(p/2)] + p − , (2.4)
2πpz,0 2 π 2π p2 − (n − px )2 i|n|
n6=0

with γ being the Euler constant. The evanescent fields in each closed channel,
ev pz,0 −2π[αm |z|+i[x(px −m)]
Em = R0 e , (2.5)
αm
decay exponentially with the distance |z| from the array.
Even in the simple case of just one open diffraction channel (l = 0), the scat-
tering amplitude R0 seems to have a rather complicated structure from which the
resonant picture given in Fig. 1 is not obvious at all. The Breit–Wigner theory
asserts that the amplitude R0 should have a simple pole structure
iΓ1
R0 (p, px ) = − , (2.6)
p − p1 + iΓ1
near the resonance position p = p1 where Γ1 is called the resonance width (see
Ref. 83 for a review). Note that at p = p1 , R0 = −1, and in the asymptotic region
z → ∞, the scattered and incident fields in (2.2) (l = 0) would have the same
amplitudes and opposite phases so that the transmitted field vanishes and, hence,
by the flux conservation, the total reflection is achieved. In fact, R0 is the amplitude
of the reflected field. Since the amplitude of the incident field is normalized to
one, |R0 |2 = Γ21 /[(p − p1 )2 + Γ21 ] is the reflectance which gives a very accurate
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5198 S. V. Shabanov

approximation of the resonant curves in Fig. 1 near the resonance (p = p1 ). From


(2.3) it follows that Rl = (pz,0 /pz,l )R0 . Hence, if the Breit–Wigner formula (2.6)
is justified, the reflectance in higher diffraction orders must also have a similar
resonant structure. An important difference, though, is that |Rl | never reaches
the absolute maximum of unity for l 6= 0. The physical reason for this is the
flux conservation: the incident flux is distributed among several open diffraction
channels. The corresponding trapped modes have shorter lifetimes (the resonances
in |Rl |2 are broader) because they have more than one decay channel.
To prove the asserted statements, consider the electromagnetic flux along the
z-direction. Its conservation requires that
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

X X
pz,0 = pz,l |δ0l + Rl |2 + pz,l |Rl |2 .
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

l l

The first term represents the transmitted flux, and the reflected flux is given by the
second one. The flux conservation leads to the optical theorem:
X 1
Re[Σ̂0 − (S0 − 1)−1 ] = , (2.7)
2πpz,l
l

from which it follows that the amplitude |Rl | reaches its absolute maximal value if
there exists p = pr (px ) such that
h i
Im (S0 − 1)−1 − Σ̂0 = 0. (2.8)

This equation defines the position of resonances. To find them, analytical properties
of R0 (p, px ) as a function of p have to be investigated. Without loss of generality,
suppose that 0 ≤ px < 1/2. Consider first the case when there is only one open
scattering channel, p < 1 − px (zero-order diffraction). The scattering phase δs is
small and therefore the approximation (S0 − 1)−1 = −i/(2δs ) − 1/2 + O(δs ) is
justified to infer from (2.8) that
[2δs (p)]−1 = −Im Σ̂0 (p) . (2.9)
If this equation has a solution p = p1 (px ), then R0 (p1 , px ) = −1. When p < 1 − px ,
the sum in (2.4) contributes only to Im Σ̂0 and, in accord with the optical theorem,
Re Σ̂0 = 1/(2πpz )−1/2. Note that the scattering phase must be positive in order for
a solution of (2.9) to exist. It is indeed positive for dielectrics as ε > 1. By reducing
the cylinder radius, δs−1 can be made large. In order for −Im Σ̂0 to be large, p1 must
be close to the diffraction threshold p1 = 1 − px − q1 , where 0 < q1  1, because
the term n = 1 in (2.4) diverges as (q1 )−1/2 . Hence, Eq. (2.9) always has a solution.
In the leading order of perturbation theory for small δs ,
 2
δ1 1
p1 = 1 − p x − q 1 , q1 = , (2.10)
π 2(1 − px )
where δ1 = δs (1 − px ) is the scattering phase at the first diffraction threshold. The
Breit–Wigner formula (2.6) is obtained if the denominator of R0 is approximated
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5199

by a linear function of p − p1 in the vicinity of the resonant frequency;


 3
(2πpz,0 )−1 δ1 1
Γ1 = − ≈ √ , (2.11)
−1
∂p [(2δs ) + Im Σ̂0 ] π (1 − px ) 1 − 2px
p=p1

where ∂p denotes the derivative with respect to the frequency p. The leading con-
tribution comes from ∂p Im Σ̂0 because the derivative of the term n = 1 in Eq. (2.4)
−3/2
diverges as q1 ∼ δ1−3 .
The poles of R0 in other diffraction channels are found by the same technique.
For a small phase δs , the resonance position must be close to the nearest closed
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

scattering channel, p = |m − px |. If l = 0 (zero-order diffraction), then m = 1; when


two scattering channels are open, l = 0, 1, m = −1; for l = 0, ±1, m = 2 and so on.
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

In the sum (2.4) all open channels, n = l, contribute to Re Σ̂0 so that the optical
theorem (2.7) holds. A perturbation theory solution to (2.9) is given by
 2
δm 1
p = pm = |m − px | − qm , qm = , (2.12)
π 2|m − px |
where δm = δs (|m−px |). In a neighborhood of the resonant frequency, the scattering
amplitudes Rl can also be written in the Breit–Wigner form. Let alm be the value
of p−1
P
z,l at p = |m − px | and am = l alm . Then
 3
a1 iΓm a1 am δm
R0 = − = Rl ; Γm = , (2.13)
am p − pm + iΓm alm |m − px | π
similarly to (2.11). Note that all Rl have the same resonant factor so that the total
reflected flux has a resonance at p = pm of width Γm .
Thus, the reflection amplitude reaches its absolute maximum of unity only when
one scattering channel is open, p < 1 − px . As the frequency of the incident light
increases, the reflection amplitude attains consecutively smaller relative maxima
below p = |m − px | with |m| = 1, 2, . . . (the maximal value of |Rl | is alm /am < 1).
Larger values of |m| correspond to more open channels and the incoming energy flux
is distributed among all of them, px → px ±l. The reflected energy flux normal to the
array decreases and the total reflection never reaches unity at the resonance p = p m
(m 6= 1). The resonances also broaden as more open channels become available,
Γm > Γ1 . The conclusion is summarized in Fig. 3 where the exact (calculated)
total reflected flux l pz,l |Rl |2 as a function of the frequency k = ω/c = pG and
P

the x−component of the wave vector kx = px G.74 The maxima of the reflected flux
form a curve p = p(px ), which can be viewed as the dispersion curve of trapped
modes. The Breit–Winger theory of the resonant scattering in the system studied
is indeed in good agreement with the exact solution of Maxwell’s equations. Some
limitations of the Breit–Wigner theory for other periodic structures are presented
in Sec. 3.
In conclusion it is worth noting the following phenomenon. The field magnitude
on the cylinders |Ec | = 2|c0 |, or |Ec (p, px )| = 2πpz,0 (δs (p))−1 |R0 (p, px )|, reaches its
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5200 S. V. Shabanov
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

2
P
Fig. 3. Total reflection l pz,l |Rl | as a function of the frequency k = ω/c = pG and the
x−component of the wave vector kx = px G. The maxima are located right below the nearest
closed diffraction channel as given in (2.12). The curves p = pm (px ) where the reflection attains
its maximal values are the dispersion curves of the corresponding trapped modes.

maximum at p = p1 + O(Γ21 ). Therefore the field amplification of the trapped mode


can be related to the resonance width
 1/3
1 − 2px 2
|Ec (p1 , px )| = 1/3
. (2.14)
1 − px Γ1
−1/3
In particular, for the normal incidence, |Ec (p1 )| = 2Γ1 . Naturally, the field on
the cylinders decreases when more diffraction channels become open, |Ec (pm )| =
−1/3
2bm Γm where bm = (a2m /|m − px |)1/3 . Thus, the smaller the resonance width is
the higher is the field amplification which can be obtained. This effect can be used
to amplify nonlinear optical effects, e.g., a higher harmonic generation (Sec. 6).

2.3. Energy of trapped modes and Fano’s theory


When a structure is illuminated at the resonant frequency, the excitation of a
trapped mode (assuming that there is one) occurs at the very moment of the ra-
diation front arrival, which is not a stationary process. The energy is first pumped
into the evanescent fields before the stationary stage is reached, i.e., when the trans-
mitted and reflected energy flow becomes constant as described by the scattering
theory. How much energy can be trapped by the structure at a given frequency and
amplitude of the incident wave? The answer is quite remarkable.
Consider the array of cylinders and assume that only one scattering channel is
open, l = 0. Using (2.5) the electric energy of the evanescent modes per period of
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5201

the array can be computed,


p2z,0 X −3
Z ∞ X
ev 2
E(p, px ) = dz |Em | = |R0 |2 αm .
−∞ π
m6=0 m6=0

The energy reaches its maximum at the resonance frequency p = p1 . Near the
resonance frequency, by making use of (2.11) and (2.6) yet another classical result
of resonant scattering theory due to Fano76 is obtained,
A(px ) Γ21
E(p, px ) = , (2.15)
πΓ1 (p − p1 )2 + Γ21
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

where A(px ) = [(1−2px)(1−px )]1/2 . Thus, the maximal energy that can be trapped
by the system is inversely proportional to the resonance width, Emax = E(p1 , px ) ∼
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

1/Γ1 . When the incoming flux is terminated, the trapped mode looses its energy
exponentially through the radiation continuum (outgoing waves); its lifetime τ ∼
Γ−1
1 . From this point of view it is easy to understand the broadening of resonances
above the first diffraction threshold. The corresponding trapped modes have more
channels to decay and, hence, should have shorter lifetimes or larger widths. Note
that the factor am in the width Γm is the sum of positive terms over all open
channels. Each term represents a contribution to the width from the corresponding
open channel.

3. Trapped Modes in 2D Periodic Structures


3.1. Trapped modes in dielectric gratings
Suppose that instead of the array of cylinders there is a homogeneous thin dielectric
layer centered at z = 0. Then the solution (2.2) would have no evanescent fields, lo-
calized in the vicinity of the layer. Neither the scattering amplitude has resonances.
However Maxwell’s equations have solutions that are localized in the vicinity of the
layer, wave-guided modes with frequencies ω(kx ) < ck below the radiation contin-
uum. The wave-guided modes can propagate along the layer. They are decoupled
from the radiation modes and, hence, cannot be excited by the latter. A coupling
can be introduced by inhomogeneities in the layer surface as, e.g., in the grating
shown in Fig. 4. Now the incident radiation can excite wave-guided modes at the
slits, the modes propagate along the layer reaching neighboring slits where they
can excite the radiation modes. If the slits are narrow, the coupling is small and
long-lived trapped modes are expected to exist in the system and, consequently,
narrow resonances in the scattered far fields.
The grating is practically transparent if the dielectric constant is not large.
Their scattering properties have been well-studied both experimentally and theo-
retically.77–81 They are used here to further illustrate the relation between the res-
onances in scattering amplitudes and the (wave-guided) trapped modes82 as well
as to show some limitations of the Breit–Wigner theory. For simplicity, the normal
incidence in the zero-order diffraction is considered. The grating parameters are
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5202 S. V. Shabanov
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

Fig. 4. (Color on line) Left: Geometry of a dielectric grating suspended in the vacuum. Right: The
x-component of the electric induction of the trapped mode corresponding to the resonance shown
in the upper left panel of Fig. 7 as a function of x and z in units of D. Different colors correspond
to different signs, while the color is more intense for larger amplitudes. The mode can clearly be
associated with a standing wave-guide mode in the dielectric slab; it occupies the entire slab and
has an alternating sign along the slab.

Fig. 5. Reflectivity of the dielectric grating shown in Fig. 4 for ε = 2, D = 1.75 µm and
a = 0.3 µm as a function of the wave length (in units of D). Each panel shows the result for
a specific value of the thickness h indicated in the panel.

chosen so that the frequency range of interest is representative for infrared light,
D = 1.75 µm and a = 0.3 µm. The scattering of a broad Gaussian wave packet on
the structure is simulated. The results for ε = 2 are shown in Fig. 5. Each panel of
the figure corresponds to a specific value of the thickness h indicated in the panel.
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5203


Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

Fig. 6. Electric field as a function of time at z = 3.5D. The wave packet propagates from negative
to positive values of z. The grating is centered at z = 0. The grating parameters are the same as
in the upper left panel of Fig. 5.

Fig. 7. Reflectivity of the grating in the same case as in Fig. 6 but for ε = 4.

The number of resonant peaks grows with increasing h in accord with the number
of wave-guided modes in a dielectric slab.
The calculated reflectivity does not reach its absolute maximum of unity for
some resonances. This is, however, an artifact of numerical simulations associated
with extremely narrow resonances82 (cf. the results for broader resonances in Fig. 7
and the discussion in Sec. 3.3). Figure 6 shows the electric field as a function of time
at z = 3.5D (i.e., behind the grating) for the case given in the upper left panel of
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5204 S. V. Shabanov

Fig. 5. The grating is nearly transparent therefore most of the wave packet radiation
gets through it. The arrival of the wave packet at the observation point is clearly
visible. Its duration is about 25f s. After that a long lasting nearly monochromatic
radiation is seen coming from the structure, which indicates the existence of a long-
lived trapped mode. For a presentational reason the graph is truncated. In fact, the
mode lifetime is about 2ps. Other trapped modes have even longer lifetimes.
Similarly, to the array of cylinders, larger values of the dielectric constant cor-
respond to a stronger coupling of the trapped modes to the radiation continuum.
The modes have a shorter lifetime and the corresponding resonances broadens. The
reflectance for the same grating with ε = 4 is shown in Fig. 7. Now the calcu-
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

lated reflectance reaches its absolute maximum of unity. The broadening of the
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

resonances and/or a stronger coupling to the radiation continuum leads to a more


complex structure of the scattering amplitude as compared to the case ε = 2. It
can no longer be interpreted as that associated with the existence of trapped wave-
guided modes that do not interfere with each other as seen in the bottom panels
of Fig. 7. A simplified description of the scattering by means of the Breit–Wigner
theory becomes less accurate (see Sec. 3.3 for a further discussion).

3.2. Trapped modes in metal gratings. The effect of attenuation.


Nondispersive materials have been considered so far. Here the effects of the material
dispersion on trapped modes are investigated, in particular, the effect of attenu-
ation (causality requires the attenuation for dispersive materials according to the
Kronig–Kramer relation). From the point of view of the Breit–Wigner theory, the
material attenuation should shorten the lifetime of trapped modes; the correspond-
ing scattering resonances would be broadened. The physical reason is simple. In
addition to open radiation decay channels, trapped modes can also dump their en-
ergy into heat. The existence of additional decay channels leads to shortening the
lifetime as argued in the preceeding section.
Simplest periodic arrays made of a dispersive material are metal gratings. The
metal dielectric constant is described by the Drude model
ωp2
εM ≡ ε(ω) = 1 − , (3.1)
ω 2 + iηω
where η > 0 is the attenuation constant and ωp is the plasma frequency. The trans-
mittance of metal gratings have been intensively studied, especially, for the TM
polarization of the incident light, i.e., with the electric field perpendicular to the
slits.25,84–89 The main reason for using the TM polarization is that such incident ra-
diation can excite surface waves (plasmon polaritons) propagating normally to the
slits. The slits provide the needed coupling between the radiation continuum and
the surface waves, much like the coupling of the radiation and wave-guided modes
in dielectric gratings. The role of plasmon polariton modes in scattering properties
of periodically perforated metal films is discussed in the next section. Here the TE
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5205


Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

Fig. 8. (Color on line) Left: Metal grating with the slits filled by a nondispersive dielectric.
Right: Scattering of a Gaussian pulse on the grating. Electric field as a function of time and z at
x corresponding to the center of the grating slit. The red and blue colors correspond to positive
and negative values, respectively. The color intensity reflects the field amplitude. The grating
parameters are given in the text. For the presentation purpose the attenuation η in the metal is
set to zero to extend the lifetime of the trapped modes. The grating is positioned at 0 ≤ z ≤ h/D g .

polarization case (the electric field is parallel to the slits) is investigated. The inci-
dent radiation impinges normally at the grating. Due to the translational symmetry
along the slits, the plasmon polariton modes cannot be excited. Nevertheless the
system exhibits an enhanced transmittance at particular frequencies in the zero-
order diffraction if the slits are filled with a nondispersive dielectric. Let us find out
the origin of trapped modes responsible for these scattering resonances.90
The system is sketched in the left panel of Fig. 8. Each slit may be viewed as
a cavity which can support a standing wave. The standing wave mode would even-
tually decay because of the coupling to the radiation continuum, which seems as a
plausible scenario for resonant scattering. To find a rough estimate of wavelengths of
these modes, assume that the metal is perfect (ωp → ∞) so that the zero boundary
conditions hold at the metal-dielectric interface. This leads to quantization of the
x-component of the wave vector. The quantization of the z-component is obtained
by assuming that along the z-direction, the field variation can be approximated by
Fabri–Perot modes, that is, when a dielectric slab is completely transparent.91–94
This crude approximation yields the condition for the cavity mode wavelength λnm ,
ε(2/λnm )2 = (n/a)2 + (m/h)2 , (3.2)
where n, m = 1, 2, . . . . The zero-order diffraction is studied, λ ≥ Dg . The cavity
mode n = m = 1 has the largest wavelength. Hence, the cavity modes exist for
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5206 S. V. Shabanov

λ > Dg if
ε ≥ [(Dg /a)2 + (Dg /h)2 ]/4 .
To relate the analysis to the infrared region, the following geometrical parameters
are chosen: D = 1.75 µm; a = 0.35 µm; and h = 1.4 µm. With this choice, ε ≥ 9.7 in
order for the cavity modes to exist. A suitable dielectric that satisfies this condition
is Si whose dielectric constant ε = 11.9 which is used in the numerical study of the
scattering problem. Note that without any dielectric filling, the system reflects the
incident radiation in the zero-order diffraction as no trapped mode exists.
To verify that the above qualitative analysis correctly describes the physics
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

for realistic metal gratings, a silver grating is studied numerically 90 ; ωp = 9 eV


by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

and η = 0.1 eV. In contrast to nondispersive dielectric gratings, the metal has
attenuation and any trapped mode excited in the structure would dump its energy
into heat, which reduces its lifetime. To find the trapped modes, the calculations
are carried out with η = 0, first. In the right panel of Fig. 8, the electric field as
a function of time and z is shown at x corresponding to the center of the grating
slit. The initial wave packet is centered at z = −3Dg and begins to propagate in
the positive z-direction. Then most of it is reflected by the grating. The light line
trajectories of the incident and reflected packets are clearly seen. But some of the
wave packet energy is retained by the structure and slowly released, indicating the
excitation of trapped modes.
In Fig. 9 the transmittance of the grating is shown as a function of the wave-
length in units of Dg . The slits are filled with Si. The red and dashed black curves
correspond to η = 0.1 eV and η = 0, respectively. With no attenuation present, the
system becomes absolutely transparent for at least three wavelengths in zero-order
diffraction. The system has five trapped modes in the simulated frequency range.
The smallest resonant wavelength appears to be slightly less than Dg (see below).
The fact that the calculated peak of the dashed black curve at the longest reso-
nant wavelength does not reach unity at η = 0 is the same numerical artifact82 as
in the case of the dielectric grating presented in Fig. 5. For the shortest resonant
wavelength the dashed black curve does not reach unity because the correspond-
ing trapped mode can also decay into the first-order diffraction channel (cf. the
discussion of the array of cylinders).
The effect of the attenuation is quite clear. It destroys the perfect transparency
at the resonances because a substantial part of the trapped mode energy is dumped
into heat. The blue curve gives the sum of the incident and reflected flux. Its devia-
tion from unity (violation of the flux conservation) measures the loss of electromag-
netic energy into heat. The maximal losses occur at the resonant wavelength, thus
confirming that the enhanced transmission is indeed due to the trapped modes in
the system.
The above rough estimate of the resonant wavelengths can be improved by not-
ing that the field penetrates into the Drude metal (see Ref. 90 for more details).
The exact resonance positions generally appear to be red-shifted as compared to the
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5207


Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

Fig. 9. (Color on line) Left: Transmission of the grating in zero-order diffraction. The grating
parameters are given in the text. The red and dashed black curves correspond to η = 0.1 eV and
η = 0, respectively. The blue curve is the sum of the reflected and transmitted flux for η = 0.1 eV.
Its deviation from 1 defines the loss of electromagnetic energy into heat. The maximal losses are
clearly associated with resonances. Right: Electric field of the trapped modes inside the grating as
a function of the x and z-coordinates. Each colored strip corresponds to a trapped mode associated
with the related maximum of the transmission (left panel). The data (from top to bottom) starts
with the largest resonant wavelength λ ≈ 1.5Dg and ends with the shortest one λ ≈ 1Dg . The
red and blue colors correspond, respectively, to positive and negative values of the field. The
x-range for each strip (along the vertical axis) corresponds to −0.26 µm ≤ x ≤ 0.26 µm. The
range for z (horizontal axis) is shown in the figure in units of Dg . The grating is positioned in
0 ≤ z ≤ 0.8 = h/Dg . The spreading of the modes into the vacuum is clearly visible.

estimated ones. The latter observation has also been reported for the TM polariza-
tion. It can be explained by that the actual trapped modes spread into the vacuum
(in the z-direction), which can be seen in the right panel of Fig. 9. This effect is
not accounted for in the Fabri–Perot approximation. If a standing wave occupies a
larger volume, then its wavelength must be larger and, hence, its frequency is lower.
The localization of the field of the trapped modes in grating slits, as shown in
Fig. 9 (right), suggests that the slits can be viewed as an ensemble of independent
emitters. They are coherently excited by the incident wave, and their coherent emis-
sion builds up the radiation field associated with the resonantly enhanced transmis-
sion (reflection) properties of the grating. This observation is further validated by
calculating the transmission coefficient of the gratings with all the parameters being
the same except the period: Dg = D0 ≡ 1.75 µm; Dg = D0 /1.5; and Dg = D0 /4
(the results are not shown here; see Ref. 90). Variations of the grating period do
not affect positions of the peaks in the transmission coefficient, pointing at the
independence of the trapped field associated with different openings. This feature
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5208 S. V. Shabanov

is in contrast to the case of the TM polarization where the resonances tend to be


pinned to the grating period; the “pinning” is discussed in Sec. 4.2 with examples
of periodically perforated thin metal films. The transmission coefficient increases
when the grating period is reduced. This is naturally explained by the increase of
the density of emitters (openings). The decay time of the trapped mode remains
the same because it is set by the attenuation of the metal and by the coupling of
the trapped mode in each slit to the radiation modes in the vacuum. The results
are also in agreement with (3.2): the reduction of h, and primarily a, leads to a
blue shift of the whole resonance series. Thus, the trapped modes in the system are
indeed quasi-stationary cavity modes.
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

3.3. Breit–Wigner theory and extraordinary transmittance of


subwavelength arrays
The examples discussed show that the transmittance or reflectance always reaches
its absolute maximum provided there exists a trapped mode with a single radi-
ation decay channel (e.g., zero-order diffraction, no attenuation). Note also that
all systems studied are symmetric under the parity transformation z → −z. The
perfect transmittance or reflectance of such systems has an elegant explanation in
the Breit–Wigner theory. According to the latter, the existence of a trapped mode
implies that the scattering amplitude has a simple pole in the complex frequency
plane. It is not yet obvious that the very existence of the pole is sufficient for the
perfect reflectance or transmittance because the coefficient at the pole is unknown.
But under some relatively general assumptions, a proof of the perfect transmittance
or reflectance can be given.90
Suppose that the scattering structure is a layer suspended in the vacuum (or
in a homogeneous nondispersive medium). It is centered at z = 0 with the z-
axis being normal to the layer. A (periodic) perforation of the layer is such that
it is symmetric under the parity transformations: z → ±z. It is also assumed
that only one scattering channel is open, which implies that the polarization is
preserved in the scattering process. So the magnetic H and electric E fields have
one component in the asymptotic region |z| → ∞ for the TM and TE polarizations,
respectively. The latter is considered. The TM polarization case can readily be
studied by replacing E by H in the coming arguments. Suppose that without the
perforation the system is totally reflective (a perfect metal slab). Alternatively, one
could also assume the system is totally transparent with no “perforation”. The
perforation can then be viewed as the insertion of a periodic array of scatterers
(e.g., the array of dielectric cylinders). The former case is considered here; the
arguments are readily extended to the latter case.
The scattering problem solution for the nonperforated system (a mirror) is
E+
0 → ê0 [A0 e
ikz
+ c.c.] , z → −∞ ,
where ê0 is the unit polarization vector, and E+
0 → 0 as z → ∞. Since the trapped
mode in the perforated system has just one radiation decay channel, the scattered
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5209

field in the asymptotic region is

E+ → (±1)p ê0 A+ e±ikz , z → ±∞ .

where A+ is the scattering amplitude and p is the mode parity. Due to the par-
ity symmetry of the structure geometry, the trapped mode is either symmet-
ric (p = 0) or skew-symmetric (p = 1). The amplitude A+ can be found from
the energy flux conservation because no attenuation is present. The incident flux
is |A0 |2 . The outgoing flux reads |A0 ± A+ |2 + |A+ |2 . Let φ0 and φ+ be the
phases of A0 and A+ , respectively. Then from the flux conservation it follows that
|A+ |2 = |A0 |2 cos2 (φ0 −φ+ ). In accord with the Breit–Wigner theory, the scattering
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

amplitude can be approximated as A+ ∼ (ω − ω0 + iΓ)−1 near the resonance. In


by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

principle, A+ may contain a “background” scattering term described by an analytic


function in a neighborhood of the pole. To simplify the arguments, it is assumed
that the background scattering is small, and the scattering is dominated by the
contribution of the trapped mode; this is true for many of the above examples. It
follows then that φ+ , as a function of the frequency ω, rapidly changes by π over
a small interval containing ω0 (|ω − ω0 | ∼ Γ), while the phase φ0 describing the
nonresonant scattering is nearly constant, or changes slowly. Therefore by conti-
nuity of φ+ in the vicinity of ω = ω0 there exists a frequency ω = ωr at which
|A0 |2 = |A+ |2 , that is, the incoming flux coincides with the transmitted flux and
the structure becomes transparent.
The case when the background scattering cannot be neglected (or there are
several trapped modes with close frequencies) can be studied by means of the formal
scattering theory. A proper mathematical tool to describe the trapped modes is the
Ziegert states.95,96 The decomposition of the scattered field over simple poles is
known as the Mittag–Leffler theorem.97 The technicalities of the formalism can be
found, e.g., in Ref. 98. In particular, by including a background scattering, the Fano
profiles in the resonant scattering can be explained (cf. Fig. 7).

4. Material and Geometric Trapped Modes


Dispersive properties of the metal do not seem to play any significant role in the
trapped (cavity) mode formation in the metal grating studied in the previous sec-
tion. Indeed, the physics would be essentially the same if the perfect metal is used
(ωp → ∞). There would be the same number of the cavity modes which are respon-
sible for the enhanced transmittance of the grating. The metal dispersion merely
leads to a small shift of the resonant frequencies and attenuation. In other words,
the very mechanism of the enhanced transmittance remains in tact when the metal
dispersion is neglected. The structure geometry plays a dominant role in the trapped
mode formation and, hence, in the resonant scattering. The same applies to nondis-
persive dielectric structures. It is then natural to ask: when do the material disper-
sive properties lead to new physics (e.g., new trapped modes) in the wave scattering
on periodic arrays and how could it be detected in experiments?
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5210 S. V. Shabanov

For nondispersive media, Maxwell’s equations are invariant under the scaling
transformations: r → νr and λ → νλ, where λ = 2πc/ω is the wavelength. Con-
sequently, the scattering amplitude should exhibit the scaling symmetry. In par-
ticular, if the reflectance or transmittance has a resonance at λ = λr of width Γ,
then the structure whose dimensions are scaled by a factor ν must have a resonance
at λ = νλr of width Γ/ν. For example, the scattering amplitude for the array of
dielectric cylinders studied above depends only on scale invariant dimensionless
variables, λ/Dg and R/Dg . For dispersive media, the scaling symmetry is broken
because ε(ω/ν) 6= ε(ω). Thus, any feature of the scattering amplitude that does
not obey the scaling symmetry can only be attributed to the material dispersive
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

properties. The scaling symmetry breaking can be easily verified experimentally by


by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

manufacturing scaled structures and comparing the scattering data for them.

4.1. An array of ionic crystal cylinders


Consider a periodic array of cylinders suspended in the vacuum as shown in Fig. 1.51
Let the cylinders be made of a dispersive dielectric whose dielectric function is
described by the ionic crystal model
(ε0 − ε∞ )ωT2
ε(ω) = ε∞ + , (4.1)
ωT2 − ω 2 − iηω
where ωT is the resonant frequency, and η is the attenuation. The values ε∞,0
define the dielectric constant in the asymptotic regions ω → ∞ and ω → 0. The
system has three characteristic length scales, R, Dg and DT = 2πc/ωT (ignoring
the attenuation for the sake of discussion). The first two are associated with the
geometry of the system (geometric scales), while the last one defines the wavelength
of polariton excitations (material scale). The scattering amplitude depends on three
dimensionless parameters, λ/Dg , R/Dg , and DT /Dg , the latter of which changes
under the scaling transformations.
To investigate the effect of the material dispersion on resonances and associated
with them trapped modes, the transmission and reflection of the array are computed
as functions of λ/Dg in the zero-order diffraction (λ/Dg > 1) with a fixed filling
ratio R/Dg = 0.1 and DT /Dg = 0.5, 2.5, and 4. For DT /Dg = 0.5 the wavelength
of the polariton mode is out of the studied range, while in the other cases it is in
there. Experiments are often carried out in the infrared range. A possible material
might be the beryllium oxide99 for which ε∞ = 2.99, ε0 = 6.6, ωT = 87.0 meV, and
the damping η = 11.51 meV. With this choice DT = 2πc/ωT = 26.9 µm while the
geometric parameters, R and Dg , are varied. The results are shown in the left panel
of Fig. 10. Note that a logarithmic scale is used for the horizontal axis in order
to improve the resolution at small wavelengths. Consider first the following two
limiting cases. According to Eq. (4.1), for short wavelengths, λ  DT (Dg  DT ),
the medium behaves as a dielectric with ε ≈ ε∞ . In the long wavelength limit
λ  DT (Dg  DT ), the medium responds as a dielectric material characterized
by ε ≈ ε0 . In Fig. 10(left) the dashed and solid black curves represent the reflection
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5211

wavelength (units of the grating period)


5.0

wavelength (units of the grating period)


4.5
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

4.0
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

3.5

3.0

2.5

2.0

1.5

1.0
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
resonancewavelength
resonance wavelength //grating
gratingperiod
period

0.0 0.2 0.4 0.6 0.8 1.0


reflection coefficient

Fig. 10. (Color on line) Left: Calculated zero-order reflection (red curves) and transmission (blue
curves) coefficients for the periodic array of ionic crystal cylinders as functions of the incident
radiation wavelength measured in units of the period Dg . Different panels correspond to different
values of the period as compared to the polariton wavelength, DT = 2πc/ωT , while the ratio
R/Dg remains fixed (R being the cylinder radius). The dashed and solid black curves represent
the reflection coefficient calculated for the same array made of a lossless, nondispersive dielectric
characterized by ε = ε∞ , ε0 , respectively. Right: The reflection coefficient as a function of two
variables λ/Dg (vertical axis) and DT /Dg (horizontal axis) at R/Dg = 0.1.

coefficient of the array made of lossless, nondispersive dielectric cylinders with ε =


ε∞ and ε = ε0 , respectively. The reflection coefficient in these two cases reaches
unity near the first diffraction threshold λ ∼ Dg in accord with the early analysis
[cf. Fig. 1(right)]. This resonant pattern is explained by the existence of a quasi-
stationary (geometrical) trapped mode. The resonance width increases with ε, while
the resonant wavelength gets red-shifted, which explains the difference between the
dashed and solid black curves (ε0 > ε∞ ).
For DT /Dg = 0.5, the resonant excitation of polaritons is impossible within the
studied wavelength range, and the dielectric constant is close to ε0 . The result for
the reflection coefficient in this case is similar to the data shown by the black solid
curve. However, the reflection coefficient does not reach unity at the resonance
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5212 S. V. Shabanov

frequency because of the medium attenuation. The corresponding trapped mode


dumps some of its energy into heat.
For DT /Dg = 2.5, an enhanced reflectance occurs at two wavelengths. For the
resonance at λ/Dg ∼ 2.5, i.e., λ ∼ DT , the corresponding trapped mode is formed
by polaritonic excitations in the material. Hence, this is a material resonance. The
other resonance at λ ∼ Dg is a geometrical one. The dielectric constant in this
case approaches ε∞ for small wavelengths λ ∼ Dg . The width and position of
the geometric resonance are close to the data given by the dashed black curve.
The imaginary part of the dielectric function is large enough throughout the entire
wavelength range to produce substantial energy loss for both the resonances. When
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

the attenuation is reduced (η → η/20) in simulations, the reflection coefficient


by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

becomes close to unity at both the resonant frequencies.


Finally, for DT /Dg = 4 the polaritonic excitation appears at λ ∼ 4Dg and the
two resonances are well-separated. The geometrical resonance at λ ∼ D g closely
matches that for a lossless, nondispersive dielectric grating characterized by ε = ε ∞ .
Observe that the reflection coefficient is close to unity in this case because the
imaginary part of ε(ω) is small far from ω = ωT and, therefore, the energy loss is
small as well.
The main lesson is that, for arrays made of dispersive materials, two types of
resonances can occur, geometrical and material ones. The former are scale invariant,
while material resonances are not; the physics of the resonant scattering is different.
The scaling symmetry concept to distinguish the two types of trapped modes is
further illustrated in the right panel of Fig. 10 where the reflection coefficient is
shown as a function of two variables λ/Dg and DT /Dg , while R/Dg = 0.1 For
DT /Dg < 1 the maxima form a horizontal line reflecting the scale invariance of the
geometrical resonance at ε ≈ ε0 . For DT /Dg > 2 two sets of maxima emerge. The
horizontal line corresponds to the geometrical resonance for ε ≈ ε∞ . The material
resonances lie on the line λ/Dg ∼ DT /Dg . In the transition region, 1 < DT /Dg < 2,
the resonances overlap and cannot be resolved because the material resonance is
too broad due to a significant attenuation in the anomalous dispersion region.

4.2. Scaling in periodically perforated metal films


Consider a metal film deposited on a dielectric substrate with refraction index nd .
The film is perforated by square holes of size a so that the holes form a square lattice
with period Dg . The open area fraction of the array is f = (a/Dg )2 . This system
has been intensively studied because it exhibits an “extraordinary” or enhanced
light transmittance in the zero-order diffraction as noted in Introduction. An en-
hanced transmittance is said to occur if the normalized transmittance Tn = T /f
exceeds unity. It simply states that the array transmits more than the geometric
optics would allow, i.e., more than the open area fraction. It was observed that the
enhanced transmittance peaks near the diffraction thresholds.11 If the xy plane is
parallel to the film and kx,y,z are components of the wave vector of the transmitted
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5213

wave, then the periodic boundary conditions require that kx,y = Glx,y , where lx,y
are integers and G = 2π/Dg is the reciprocal lattice vector. For normal incidence,
the incoming wave vector has only a z component of the magnitude k = 2π/λ.
Since the incoming and transmitted waves have the same frequency, one obtains
q
2πnd /λ = G2 (lx2 + ly2 ) + kz2 . (4.2)
Hence, for λs = λ/(nd Dg ) > 1 only normal transmission is possible. The diffraction
thresholds are at λs = (lx2 + ly2 )−1/2 .
The mechanism of the enhanced transmittance is a subject of hot debates
in the literature. Some researchers favor the surface plasmon polariton mech-
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

anism,19–24 others the dynamical diffraction theory and coherent diffraction of


evanescent waves.25–28 In the former one assumes that surface plasmon polari-
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

ton waves coupled through the holes play the dominant role in the phenomenon.
The surface plasmon polariton is a wave that propagates along a planar metal-
dielectric interface and decays exponentially into the metal and dielectric with
increasing distance from the interface.18 The dispersion relation of the wave is
ksp = (ω/c)[εd εM (ω)/(εd + εM (ω))]1/2 where the subscripts d and M stand for “di-
electric” and “metal”. A film has two interfaces and two types of surface plasmon
polaritons exist. The two surface modes can interact through the holes as well as
become coupled to the radiation continuum. A periodic perforation naturally leads
to quantization of ksp , from which it readily follows that surface plasmon polaritons
can be excited near the diffraction thresholds as the real part of εM (ω) is large and
negative in the studied frequency range. Thus, the incident light excites a surface
plasmon polariton on one side, through the holes the latter excites a plasmon po-
lariton on the other side, which, in turn, decays by emitting the light on the other
side. This qualitative picture, though being quite appealing from the physical point
of view, still lacks a rigorous theoretical foundation. The diffraction mechanisms are
based on the diffraction theory of solving Maxwell’s equation in the far and near
field zones, much like in the above study of the array of cylinders and the metal
grating with dielectric fillings.
The enhanced light transmittance is a typical resonant phenomenon. Therefore
a mechanism of formation of the corresponding trapped mode should be studied.
The scaling symmetry can easily be verified to find out whether the dispersive
properties of the metal are relevant for the trapped mode formation or not.
The left panel of Fig. 11 shows experimental data47 for the transmittance of
five arrays, all with f = 0.25. Two (upper panel) were fabricated on fused silica
and three (lower panel) on ZnSe substrates. To obtain wide wavelength coverage,
fused silica substrates (nd = 1.4) were used for smaller period (Dg = 1 and 2 µm)
hole arrays and ZnSe substrates (nd = 2.4) for longer period (Dg = 4, 6, and
8 µm) arrays. The 70 nm thick Ag film on the fused silica substrate is thinner than
the 100 √ nm thick Ag film on the ZnSe substrate. The first, λs = 1, and second,
λs = 1/ 2, diffraction thresholds are shown as vertical dashed lines in the figure.
For each substrate, most of the spectral features scale well with λs , confirming,
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5214 S. V. Shabanov

0.4
0.2
0.0
-0.2
-0.4
-0.4 -0.2 0.0 0.2 0.4

0.4
0.2
0.0
-0.2
-0.4
-0.4 -0.2 0.0 0.2 0.4
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

0.4
0.2
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

0.0
-0.2
-0.4
-0.4 -0.2 0.0 0.2 0.4

Fig. 11. (Color on line) Left: Transmittance versus λs = λ/(nd Dg ), for five arrays with the same
open area fraction, f = 0.25. The upper panel shows the data for two arrays made on a fused silica
substrate (nd = 1.4) with Dg = 1 µm (blue curve) and Dg = 2 µm (red curve). The lower panel
shows the data for three arrays on ZnSe (nd = 2.4). Variations in Dg and nd yield a variation of λ
by a factor of 14. Right: The sum of the calculated transmittance and reflectance (left panels) near
λs = 1. The ohmic losses in the metal produced by the trapped modes correspond to dips in these
curves for λs > 1. The red dashed lines show the positions of the corresponding transmittance
peaks. The right panels show the energy density of the corresponding trapped mode in the plane
of the metal-dielectric interface as a function of x/Dg (horizontal) and y/Dg (vertical). The color
code varies from dark brown (zero) to bright yellow (maximum) on a linear scale.

within the accuracy of the experiment, that they are determined by the film ge-
ometry. Note that for the ZnSe substrate the scaling holds for a variation in Dg
(or λ) by a factor of 8. Similar results were also observed for f = 0.11 and 0.44
with the same variations of Dg . As shown in Ref. 47, the experimental data are in
good agreement with numerical solutions of Maxwell’s equations; the corresponding
simulated curves appear to have more narrow resonant peaks, which can naturally
be attributed to some imperfections of the actual film (the energy of the trapped
mode may leak into additional channels existing due to small geometrical defects
of the film). Calculated positions of the resonances exhibit a slight red shift which
cannot be resolved in the experiment due to the broadness of the peaks. The scaling
has been further verified by additional numerical simulations for Ag films on the
ZnSe substrate with f = 0.25 (Dg = 2a, a = 2 µm and 4 µm) and the film thick-
ness being 90 and 180 nm. For zero-order transmittance peak position, its width,
and its amplitude, the results show that the scaling symmetry holds extremely
well (it is exact within the accuracy of simulations, i.e., within a fraction of a per
cent).
Narrow resonances correspond to long-lived trapped modes. The longer the
mode lives, the more Ohmic losses occur at the resonance frequency. Hence, the
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5215

Ohmic losses should increase as the open area fraction f decreases. The sum of
the transmittance and reflectance as a function of λs is shown in the right part
of Fig. 11. For λs > 1, dips of these curves correspond to maximal Ohmic losses.
For small holes (the lowest panel), a dip is clearly visible, indicating substantial
Ohmic losses occurring at about the same wavelength as the transmittance peak
(red dashed line). For large holes, no structure at λs > 1 is present (upper panel),
and the first diffraction threshold appears as a sharp drop indicating the opening of
an off-axis diffraction channel. The very right panels of Fig. 11 show the computed
energy density of the trapped mode over the array unit cell in the metal-dielectric
interface plane. The hole is centered in the cell. The xy asymmetry of the energy
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

density is related to the y-polarization of the incident wave packet. Most of the en-
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

ergy is concentrated near the edges of the hole. Only for a small hole, the trapped
fields extend substantially over the metal surface as well as into the hole itself.
The Ohmic losses and the presence of a substrate reduce substantially the en-
hanced transmittance of the film. The Breit–Wigner theory predicts (see Sec. 3.3)
that the transmittance T can always reach unity if the attenuation is neglected
and the system has the reflection symmetry. Hence, the enhanced transmittance
always occurs because Tn = 1/f > 1. For a film on a substrate, the reflection
symmetry is broken and the statement no longer holds. More reflective substrates
(higher nd ) tend to suppress the measured transmittance T . The effect is quite
visible in the experimental data of Fig. 11(left). For the fused silica substrate, the
largest Tn = 2.8 (red peak), while for ZnSe, Tn = 1.1 (blue peak). Smaller f ’s
correspond to larger Ohmic losses as seen in Fig. 11(right) so that the trapped
mode dumps a substantial part of its energy into heat rather than into the radia-
tion field. However, the calculated transmittance for, e.g., f = 0.11 and the peak
at λs > 1, gives Tn = 0.5, while, without the substrate, Tn = 3.6; if, in addition,
the Ag attenuation is set to zero, then Tn = 1/f = 9.1 in the full agreement with
the Breit–Wigner theory prediction. The corresponding simulation results are not
shown here.
Thus, the experimental and numerical data show that the scaling symmetry
holds. Note that the film is thin enough so that wave-guided modes cannot be
excited (cf. Sec. 3.2) and, hence, the scaling symmetry should hold even without an
appropriate scaling of the film thickness. The enhanced transmittance is a resonant
phenomenon; the corresponding trapped modes are scale invariant and looks like
coherently excited emitters localized in the vicinity of each hole (Fig. 11 (very right
panels)). What are implications of this results for the plasmon mechanism of the
enhanced light transmittance?
First, the frequency range studied lies quite far from the plasmon frequency
of silver. This seems to be a plausible explanation of the data. The problem
can be solved analytically in the limit ωp → ∞ (perfect metal)34 by Babinet’s
principle with the conclusion that the hole array transmits 100% at some fre-
quency. The multiple scattering effect enhances the electric field in the film open-
ings thereby the trapped mode is formed (similarly to the case of the array of
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5216 S. V. Shabanov

cylinders). The entire effect of the metal dispersion is in a small shift of the reso-
nant frequency.34
Second, the periodicity of holes and dispersive properties of the metal film are
not necessary at all for the formation of trapped (evanescent) modes bound to
the film. This was already possible to deduce from the Bethe’s theory (see for a
review100 ) applied to the classical problem of diffraction by a hole in a perfectly con-
ducting screen, but the actual confirmation came later.101–107 Interestingly enough,
an enhanced transmittance has been observed for just a few holes.108,109 The corre-
sponding trapped modes are associated with Maxwell–Garret resonances. 110 When
the holes are arranged periodically, the interference of the decay radiation of the
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

coherently excited evanescent modes in each hole with the incident radiation and
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

with each other leads to the perfect transmittance of the array. This phenomenon
is well-known to exist for perforated metal films in the microwave range.111–114 So,
when the plasmon frequency is finite, but still large compared to the frequency in
question, the film geometry retains its dominant role in the trapped mode formation
and, therefore, in the resonant transmittance.
Does this mean that the surface plasmon polaritons are irrelevant for the en-
hanced transmittance mechanism? The question is, actually, ill-posed. Our under-
standing of a particular phenomenon in a complex dynamical system is often based
on identifying a dominant degree of freedom whose dynamics explains the phe-
nomenon in the leading order of perturbation theory. A conventional plasmon po-
lariton mode is defined as a solution of Maxwell’s equations localized near a flat
metal-dielectric interface. Similar solutions can be found for a thin metal film (two
interfaces). They are not scale invariant. A perforation of the film changes dras-
tically the surface plasmon polariton solution.17,18 It is also known that surface
modes in structured metal-dielectric composites may well become localized even for
ω/ωp  1 (in particular, in the holes of a metal film even if the holes are distributed
randomly),115 while the conventional plasmon polariton solution becomes delocal-
ized in this limit. So, the very notion of (localized) plasmon excitations strongly
depends on the structure geometry; such modes may well exhibit an approximate
scale invariance in some frequency range. In the above question, one should agree
upon what is exactly meant by the “plasmon polariton” modes for perforated films.
The debated “differences” between the two mechanisms might be more in semantics
than in physics.

5. Coupling of Trapped Modes


Resonant properties of composed layered photonic structures (when two or more
structures are adjacent) can be understood with the help of the concept of coupled
trapped modes. The coupling can change the corresponding resonance positions
and widths; it can even stabilize the trapped mode turning it into a bound state in
the radiation continuum (the trapped mode becomes decoupled from the radiation
modes and can no longer be excited by the incident radiation). Trapped modes
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5217

can interact through the radiative fields as well as through evanescent fields if the
distance between the structures is not too large.

5.1. Generalized scattering matrix


Consider a scattering problem for a plane electromagnetic wave of the frequency ω
impinging a periodic structure in a layer of thickness h. The layer is normal to the
z-axis and positioned so that z ∈ [0, h] in it. To reduce technicalities to a minimum,
the electric field of the incident wave is assumed to be parallel to the y-axis and
the structure geometry to be invariant under translations parallel to the y-axis (a
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

grating). The formalism can readily be extended to generic periodic layers. The two
polarization scattering channels, TM and TE, become coupled in general.116–120
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

The incident wave has the wave vector with two components kx (parallel to
the structure) and kz (normal to the structure). The electric field is represented
as E(t, x, z) = Eω (x, z)e−iωt+ikx x where the amplitude Eω satisfies Maxwell’s
equations
[∂z2 + V (x, z; ω)]Eω (x, z) = 0, (5.1)
where V (x, z; ω) = ε(ω, x, z)(ω 2 /c2 )+(∂x +ikx )2 . The dielectric function is piecewise
constant: ε(x, z; ω) = εm (ω) in regions occupied by the structure material and
ε(x, z; ω) = εa otherwise (e.g., εa = 1 if the structure is suspended in the vacuum).
A general solution of (5.1) is sought. Let D be the period of the structure. The
functions ε and Eω are periodic in x. The solution is expanded into a Fourier
−1/2 ijGx
P
series Eω (x, z) = j ϕj (x)Ej (z), where ϕj (x) = D e with G = 2π/D.
Define an ordered column E(z) whose entries are the amplitudes Ej (z). Using the
orthonormality of the Fourier harmonics, the equation for E(z) is derived from (5.1):
Z D
d2 E
+ W(z)E = 0 , (W)jj 0 (z) = ϕ̄j 0 (x)V (x, z; ω)ϕj (x) . (5.2)
d2 z 0

Outside the structure W is constant and diagonal: (W)jj 0 (z) ≡ (Wa )jj 0 =
1/2
δjj 0 [εa ω 2 /c2 − (Gj + kx )2 ]. Let ka = Wa , where the square root is defined so
that Im ka ≥ 0. The jth (diffraction) channel is said to be open if (Wa )jj > 0, and
closed otherwise. A general solution outside the structure is
E = eika z A+
L +e
−ika z −
AL , z < 0;
(5.3)
E=e ika (z−h)
A+
R +e ika (h−z)
A−
R , z > h.
The phase factors are chosen so that the elements of A± L,R define the maximal
amplitude of the corresponding wave propagating in the positive (+) or negative
(−) z-direction, referred to as the forward or backward propagating waves.
The generalized scattering matrix S is defined by 120
A+ A+ A+
! ! ! !
R L T + R− L
=S ≡ . (5.4)
A−
L A−
R R+ T − A−R
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5218 S. V. Shabanov

If a general solution of (5.2) is known, then S can be explicitly found. In particular,


for a a homogeneous layer (no structure present), S = SM whose blocks are R± = 0
and T ± = Φ = eihka .
To elucidate the physical significance of S, let us impose the scattering boundary
conditions. Suppose, first, that the incoming wave goes forward. Then the solution
should not have any backward propagating waves for z > h and, therefore, A − R = 0.
In this case A+R = T + +
A L defines the transmission amplitudes in each channel, while
− + +
AL = R AL are the reflection amplitudes in each channel. Similarly, if the incoming
wave is propagating backward, then A+ L = 0. The matrices T

and R− define
the backward transmission and reflection amplitudes, respectively. The difference
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

with the conventional scattering matrix is that S includes all closed channels in
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

addition to all open channels. In the asymptotic region, |z| → ∞, only open channels
contribute. Hence, the conventional scattering matrix is composed of the open-to-
open blocks of T ± and R± .

5.2. The convolution rule


Suppose that there are two layered structures for which the generalized scattering
matrices are known, say, S1,2 . Let the distance between the structures be L. What
is the scattering matrix S for the combined structure? There is a simple rule to
compute S and it has a transparent physical interpretation.
The problem has four interfaces which are set at z = 0, z1 , z2 , z3 where z2 −
z1 = L. A general solution of (5.2) for z1 < z < z2 is written in the form EM =
eikM (z−z1 ) A+
M +e
ikM (z2 −z) −
AM . For z < 0 and z > z3 , the solution is (5.3) where
h → z3 . The phase factors are chosen again so that A+ (A− ) define the maximal
amplitude of the corresponding wave propagating forward (backward). Let S 1 (S2 )
be associated with the structure in 0 < z < z1 (z2 < z < z3 ). By definition of S1,2
! ! ! !
A+M A+
L A+R ΦA+M
= S1 , = S2 ,
A− L ΦA−M A−
M A−R

where Φ = exp(ikM L). Combining these relations with the definition (5.4), the
total generalized scattering matrix can be written in the form:

S = S2 ? SM ? S1 , (5.5)
where the ?-product, S = S2 ? S1 , is defined via the block decomposition

T + = T2+ [1 − R− + −1 +
1 R2 ] T1 ; R+ = R + − + − −1 + +
1 + T1 [1 − R2 R1 ] R 2 T1 ,

T − = T1− [1 − R+ − −1 −
2 R1 ] T2 , R− = R − + − + −1 − −
2 + T2 [1 − R1 R2 ] R 1 T2 .
The ?-product is associative but not commutative: (S3 ? S2 ) ? S1 = S3 ? (S2 ? S1 ),
but S2 ?S1 6= S1 ?S2 , which is natural from the physical point of view. In particular,
for the unit matrix I we have, I ? S = S ? I = S. If the distance between the two
structues vanishes, L = 0, then SM = I in Eq. (5.5).
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5219

The ?-product has a transparent physical interpretation. Consider the forward


transmission T + of two structures next to each other and use the geometric series
to write it as T + = T2+ T1+ + T2+ R− + + + − + − + +
1 R2 T1 + T2 R1 R2 R1 R2 T1 + · · · . Take
the second term, for example, and read it from the right to left. First, the forward
incident wave is transmitted across the first structure (the factor T1+ ), then reflected
from the second one (the factor R+ 2 ), thus becoming backward propagating. Next
it is reflected by the first structure as a backward propagating wave (the factor
R−1 ), thus becoming forward propagating again, and, finally, it is transmitted by
the second structure (the factor T2+ ). The series describes nothing but the multiple
scattering effects where both open and closed channels are included, which makes
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

perfect sense if the structures are close. The other blocks of S can be interpreted
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

similarly. For the structures separated by a large distance, close channels do not
contribute, and the combined structure turns into a Fabri–Perrot interferometer.

5.3. Coupling of two trapped modes


Suppose that each of the two structures has scattering resonances. What are the
resonances of the combined structure? This question can be answered by studying
analytical properties of the generalized scattering matrix (5.5). To simplify the
discussion, the structures are assumed to be symmetric so that the forward and
backward reflection (transmission) matrices coincide. The superscript (±) may be
omitted. The forward reflection is
R+ = R1 + T1 ΦM−1 R2 ΦT1 , M = 1 − R2 ΦR1 Φ . (5.6)
The matrix M−1 is inversely proportional to det M. Therefore the poles of the
scattering amplitude must be zeros of det M. Finding the latter would be a tough
problem in general settings. Some simplifying assumptions based on the physics
of the problem should be made. First, in the frequency range of interest both the
structures have only one open channel (denoted by the subscript o) so that
! !
Rµoo Rµoc Rµoo Φ0 Rµoc Φc
Rµ = , Rµ Φ = , µ = 1, 2 ,
Rµco Rµcc Rµco Φ0 Rµcc Φc
where the open-to-open block, Rµoo , is just a complex number (a 1 × 1 matrix).
As the structures are symmetric, the column Rµco is the transposed raw Rµoc . The
determinant is decomposed over the first column or raw: det M = [1−R2oo R1oo Φ20 +
σc ] det Mcc , where Mcc is the closed-to-closed block of M and σc does not depend
on Rµoo ; its explicit form is not relevant for the discussion. Note that in the limit
L → ∞, the amplitudes Φc fall off exponentially and Mcc quickly approaches the
unit matrix, while σc tends (exponentially) to zero. In this limiting case, the modes
are coupled only through radiation (open channel) modes. Second, the resonances
are not far from each other and Rµoo are given by the Breit–Wigner formula.
Let ωµ and Γµ be the positions and widths of the resonances and Ωµ = ωµ −iΓµ .
The scattering amplitude in the vicinity of zeros of det M can be written in the
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5220 S. V. Shabanov

form:
√ !
A(ω) Ω1 i Γ 1 Γ2 Σ
Roo = , H= √ , (5.7)
det(H − ω) i Γ 1 Γ2 Σ Ω2
where Σ2 = Σ2 (ω) = Φ20 (1 + σc )−1 and A(ω) ≈ const. The amplitude (5.7) has
two simple poles that are determined by the generalized eigenvalue problem for the
matrix H: H(ω)ψ = ωψ. Remarkably, this is exactly the eigenvalue problem that
is used to describe the coupling of quasi-stationary states in quantum mechanics.

The coupling of the modes is proportional to Γ1 Γ2 . So, long-lived modes are
weakly coupled. In the weak coupling approximation (Γµ  ωµ ), and assuming
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

also |σc |  1 (which is always possible for L being large enough), the eigenvelue
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

problem can be solved and the resonance positions and widths of the combined
structures are
ω+ = Re Ω+ (ω1 ) , ω− = Re Ω− (ω2 ) , Γ± = −Im Ω± (ω± ) , (5.8)
where Ω± (ω) = (1/2)(Ω1 + Ω2 ) ± ((1/4)∆Ω2 − Γ1 Γ2 Σ2 )1/2 and ∆Ω = Ω1 − Ω2 . An
interesting question arises whether one can stabilize a trapped mode by increasing
its lifetime (or reducing the corresponding resonance width) through coupling it to
another trapped mode. Even more interestingly, is it possible to obtain a resonance
with the vanishing width, i.e., to turn the trapped mode into a bound state in
the radiation continuum? The answer is affirmative in quantum theory 49 ; the same
remarkable phenomenon takes place in electromagnetism.48

5.4. Bound states in the radiation continuum


Consider two identical structures separated by a distance L each of which supports a
trapped mode. Suppose that there is only one open channel (zero-order diffraction).
First, the limit of large L is studied so that the contribution of closed channels can be
neglected. Put R1 = R2 (the identical structures) and set Φc to zero and Φ0 = eik0 L
where k0 = k0 (ω, kx ) = [ω 2 /c2 − kx2 ]1/2 . Let R0 = Rµoo and T0 = Tµoo = 1 − R0
be the reflection and transmission coefficients of the structures, respectively. Then
the reflection amplitude is the same as that of the Fabri–Perot interferometer
R−1 −1 2
0 + [R0 − 2]Φ0
Roo = . (5.9)
(R−1 −1
0 − Φ0 )(R0 + Φ0 )

However, the key difference with unstructured interfaces of the Fabri–Perot inter-
ferometer is that R0 is given by the Breit–Wigner formula (2.6), thanks to the
presence of the trapped mode. This leads to quite an interesting conclusion. Since
Σ2 = Φ20 , Ω± (ω) = ω0 − iΓ[1 ± Φ0 (ω)]. The eigenvalues of H solve the equations
ω − Ω± (ω) = 0. Let Ω± (ω) be analytically continued into the complex frequency
plane (with an appropriate cut to define the square root in k0 (ω, kx )). Then either
of these equations has a real solution ω = ω0 at some specific (quantized) values of
the distance L. Indeed, for ω = ω0 , Im Ω± (ω0 ) = Γ[1 ± cos(Lk0 (ω0 , kx ))] vanishes
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5221

when L = 2πn/k0 (ω0 , kx ) or L = π(2n + 1)/k0 (ω0 , kx ) with n = 1, 2, . . . . In either


case, ω0 − Re Ω± (ω0 ) = 0. Remarkably, the conclusion is independent of the origi-
nal resonance width Γ. For Γ/ω0  1, the new resonances can easily be found by
perturbation theory (5.8),
ω± = ω0 ± Γ sin(Lk0 (ω0 , kx )) , Γ± = Γ[1 ± cos(Lk0 (ω0 , kx ))] . (5.10)
One of the coupled trapped modes turns into a bound state whenever the distance
between the layers is such that cos[k0 (ω0 , kx )L] = ±1, while the other has the
lifetime twice as short because the corresponding resonance width is doubled. The
bound state and double-width resonance have the same position ω = ω0 . The zero-
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

width trapped mode decouples from the radiation continuum and can live forever
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

confined within the structure. It cannot be excited by the incoming wave anymore.
The structure becomes a perfect waveguide for a wave of the frequency ω 0 (kx )
propagating between the layers. This waveguide is neither metal cavity nor defect
in photonic crystals. The mode supported by it lies in the radiation continuum. The
waveguide can have a very little of material as, e.g., a double array of thin dielectric
cylinders (see Sec. 5.5). The electromagnetic fields of this excitation exponentially
decay with the distance from the structure. So, the wave is not perfectly confined
within the structure, but still cannot escape to the asymptotic region |z| → ∞.
For L comparable with the wavelength, the contribution of closed channels be-
comes important. In the diagonal matrix Φ, the open channel corresponds to Φ0 as
before, and closed channels to Φc = e−kc L . Hence, the leading contribution of closed
channels comes from a closed channel with the smallest kc . In the perturbative ap-
proach, it is sufficient to retain only one closed channel. The generalized eigenvalue
problem can be solved in the leading order in Φc for weak coupling (Γ  ω0 ),
ω± = ω0 ± Γ[sin ϕ + ν sin(2ϕ + γ)] , Γ± = Γ{1 ± [cos ϕ + ν cos(2ϕ + γ)]} ,
where ϕ = k0 L, ν = qe−kc L , and Roc Rco = qeiγ , q ≥ 0. For any q, the parameter ν
can be made small enough (by increasing L) so that one of the two trigonometric
equations cos ϕ + ν cos(2ϕ + γ) = ±1 always has a solution for ϕ. The width Γ + or
Γ− vanishes (when cos γ < 0 or cos γ > 0) and the bound state emerges.
The contribution of closed channels channels gives rise to two effects. First, let
ϕ = ϕ0 = nπ be the solution when ν = 0. The bound state exists at L = Ln =
nπ/k0 . For small ν, thep solution is obtained by perturbation theory: ϕ = ϕ0 + θ
where θ = 2sγ ν sin γ ± 2ν| cos γ| + 4ν 2 +O(ν 3/2 and sγ = ±1 if Γ+ or Γ− vanishes.
In other words, either of the resonances can turn into the bound state at two values
of ϕ or, consequently, at two values of L, one is slightly larger than Ln and the
other is slightly less than Ln if cos γ 6= 0. When cos γ = 0, a bound state always
exists at L = Ln and also at L in a neighborhood of Ln . The second effect is that
the contribution of closed channels removes the degeneracy in position of the bound
state and the resonance with the doubled width: ω+ − ω− = 2Γ(θ + ν sin γ). This
difference can be used to judge whether near (evanescent) fields are relevant for the
bound state formation.
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5222 S. V. Shabanov

The formation of a bound state in the continuum through coupling of two res-
onances was first predicted by von Neumann and Wigner in quantum mechan-
ics.49 Later unusually stable resonances were observed in two-electron systems. 50
By means of the Feshbach projection formalism in the quantum scattering the-
ory,121–123 it was shown that a bound state in the radiation continuum is always
formed when the coupling between the quasi-stationary states is a continuous func-
tion of a parameter and the avoided crossing of the resonances takes place. 124–126
If more than one channels are open, then the analysis become more involved. The
appearance of the bound state depends strongly on the distribution of the flux be-
tween open channels. A complete nonperturbative analysis of this phenomenon is
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

too technical and will be presented elsewhere.


by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

5.5. Bound states in a double array of dielectric cylinders


To illustrate a nonperturbative effect of closed channels on the coupling of trapped
modes, consider two arrays of dielectric cylinders.48 The first array is located at
r1n = (n, 0) (the length is measured in units of the period Dg ) and the second one
at r2n = (a + n, h). Here h is the distance between two arrays, and 0 ≤ a < 1
is a relative shift of the arrays along the x-axis. Suppose that only one diffraction
channel is open for the incident wave, that is, p < 1 − px and px ≥ 0. The problem
can be solved analytically in the fashion similar to the single array case. However,
the technicalities are rather involved and, therefore, not given here. They can be
found in Ref. 127.
The reflection amplitude of the double array is given by
R−2 2
0 (1 + Φa ) − Φa (Σ̂+ + Σ̂− )
R(p, px ) = , (5.11)
R−2
0 − Σ̂+ Σ̂−

where R0 is the scattering amplitude of the single array (2.3) (l = 0), Φa =


e2πiapx Φ0 , Φ0 = e2πihpz,0 ,
X
Σ̂± = e±2πiapx [e2πihpz,0 − ipz,0 σ(±a)] , σ(a, p) = α−1
m e
−2πhαm 2πiam
e ,
m6=0

and αm = [(m − px )2 − p2 ]1/2 are wave vectors of closed channels. The poles of
the amplitude (5.11) are zeros of the denominator which is det M in (5.6) where
contributions of all closed channels are now included. Put σ = σc + iσs with σc,s
2
being real. The poles are obtained by solving R−2 0 = Σ̂+ Σ̂− = (Φ0 − ipz,0 σc ) −
2 2
pz,0 σs , in the complex p-plane. Note that the new resonance positions should not
exceed the first diffraction threshold, p± < 1 − px , because the expression (5.11)
is valid when only one diffraction channel is open. The single array resonance has
width Γ1 ∼ δ13  1 − px . Therefore the poles can be found by perturbation theory
in the scattering phase δ1 .
Although the analysis of bound states can be carried out for arbitrary values of
the incident angle (fixed by px ) and the relative shift a, the discussion is further
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5223

confined to a special case of σs = 0 to avoid excessive technicalities. This includes


the normal incidence (σs = 0 thanks to the symmetry αm = α−m when px = 0)
with arbitrary a, and the symmetric array (a = 0) with arbitrary px . In both
the cases, Σ+ = Σ− , σc = σ. For large values of h, σ is exponentially small and
therefore the resonance shift due to the coupling is of order Γ1 so that both the
new resonances remain below the diffraction threshold. In contrast, for h ∼ 1, the
coupling appears to be so strong that one of the resonances is shifted above the
diffraction threshold and disappears. Mathematically, the amplitude (5.11) has only
one pole for p < 1 − px , not two. The remaining resonance can have the vanishing
width. Approximating the scattering amplitude R0 by the Breit–Wigner formula
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

(2.6), and taking into account that σs = 0, new resonance positions and widths are
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

inferred in the leading order,


p± = p1 ± Γ1 [pz,0 σ − sin(2πhpz,0 )]|p=p1 , Γ± = Γ1 [1 ± cos(2πhpz,0 )]|p=p1 ,
where p1 = 1 − px − q1 and q1 is given in (2.10). Thus, either of the resonances
can have the vanishing width when the distance h between the arrays is such that

cos(2πhpz,0 ) = ±1 at p = p1 . Next, note that α1 ∼ q 1 ∼ δ1 and the leading
contribution comes from m = 1 in the sum σ, provided h is not large, hδ1  1. When
a = 0 and px 6= 0 the resonance positions become p± = p1 ± 2q1 = 1 − px − q1 ± 2q1 .
Clearly, one of the resonances crosses the diffraction threshold, p+ > 1 − px , which
is not a legitimate solution.
The left part of Fig. 12 shows the resonant properties of the symmetric double
array of cylinders (a = 0) calculated for the dielectric constant ε = 2. The total
reflected flux, l pz,l |Rl |2 , is given as a function of p and px in the left panel (a).
P

The maxima form a singe curve, meaning that there is only one trapped mode
for the value of h indicated in the upper left corner of the figure. The second
trapped mode is always shifted above the nearest closed diffraction threshold and
disappears as argued above. The left panel (b) presents the reflection (red) and
the field (dashed-blue) in the center of a cylinder for a single array as functions of
p at a fixed value of px = 0.2. The field on cylinders is amplified ten-fold for the
shown geometry. The left panel (c) shows the same quantities but for the double
array. The resonance becomes very narrow as compared to that in the left panel
(b). It is shifted to a lower frequency. As the distance between the arrays increases,
the second resonance should emerge. This is illustrated in the right part of Fig. 12
[panel (a)] where the reflectance as a function of the frequency p and distance h is
shown for a fixed value of px = 0.2. The horizontal white line indicates the position
of the first diffraction threshold. The width of the bright yellow curves at a fixed
value of h can be used to estimate the corresponding resonance width. The points
where the bound states are formed are clearly seen (the width vanishes). For small
h only one resonance exists. The second resonance emerges when h becomes greater
than 3. For large h the reflectance maxima converge to a single curve as positions of
the resonances become degenerate in accord with the Fabri-Perot argument when
the contribution of the closed channels is neglected. The geometry of the system
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5224 S. V. Shabanov

is symmetric under reflections about the midpoint of the interval separating the
two arrays. Therefore the trapped modes must have a specific parity relative the
reflection. The mode with a lower frequency is symmetric, while the mode with a
higher frequency is anti-symmetric. The right panel (b) of Fig. 12 shows the absolute
value of the electric field for the lower frequency trapped mode. The values of the
frequency and distance are indicated by the white circle in the right panel (a).
They correspond to a resonance with a nearly vanishing width. The mode is clearly
even. The electric field of the mode is amplified by several orders in magnitude at
cylinders and midpoints of the structure. The amplification can be controlled by
varying the distance h, hence, varying the width [cf. (2.14)].
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

When a 6= 0 and px = 0, the situation is more delicate. The term m = 1 may


no longer be dominant in σ. Even though α−1 ∼ δ1−1 is large, it comes with the
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

1
factor cos(2πa) which can be made arbitrarily small by taking a closer to 1/4. Put
a = 1/4. The singular term m = 1 in σ vanishes and one may set p = 1 in the
rest of the series. For all h > 0, the resonance position shift does not exceed 2Γ,
|q± | = Γ|σ(1/4, 1) − sin(2πh)| < 2Γ and, therefore, both the resonances are below
the diffraction threshold, p± < 1. The situation resembles the large h case. If a is
far enough from 1/4, σ(a, p1 ) may again be approximated by the term m = 1. In
this case, p± = 1 − q1 ± 2q1 cos(2πa) for hδ1  1. One of the resonances crosses the
diffraction threshold when the relative shift exceeds 1/6. The trapped mode that

0.805
1.0
(a)
0.8
k (units of G)

z D 0.795 0.5
(b)
reflection coefficient

h=0.635 D 0.2 D
x
0.79
(b)
field

0.0
1.4

(a) 0.785
1. 2. 3. 4. 5. 6. 7. 8.
k (units of G)
1.2

h (units of D)
k (units of G)
x (units of D)
1.0

1.
reflection coefficient

(c) (b)
0.8

field

0.
0.6

-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5


0.0 0.1 0.2 0.3 0.4 0.5 z (units of D)
kx (units of G)

k (units of G)
0.0 0.5 1.0 0 0.5x105 105 1.5x105

Fig. 12. (Color on line) Left: Double symmetric array of cylinders with ε = 2. The system
geometry is sketched in the upper left corner. Panel (a): The reflection flux as a function of p and
px . Panel (b): The reflection flux (red) and the field on cylinders (dashed-blue) as functions of p
at a fixed value of px = 0.2 for a single array. Panel (c): The same quantities but for the double
array. The range of p is roughly indicated by the while vertical bar in Panel (a). Right: Panel
(a): The reflectance of the double symmetric array of cylinders as a function of p and distance
h calculated for px = 0.2. The white horizontal line indicates the position of the first diffraction
threshold. Panel (b): The absolute value of the electric field of the lower frequency trapped mode
computed at p and h indicated by the white circle in Panel (a). The incident wave amplitude is
equal to unity.
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5225

remains below the diffraction threshold becomes a bound state at h = n/p− with
n = 1, 2, . . . .
A general analysis of the bound states in the double array of cylinder is given
in Ref. 127 where analytic solutions for the bound states are also found.

5.6. Bound states in double gratings


Consider two identical dielectric gratings, studied in Sec. 3, that are parallel and
separated by a distance h.48 The geometry is chosen so that there is only one
sharp resonance in the zero-order diffraction as indicated in the upper left panel of
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

Fig. 5. The calculations show48 that the field of the corresponding trapped mode
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

is amplified nearly ten-fold near the grating slits (the incident wave amplitude is
unity).
As the (non-coupled) trapped modes are farther from the diffraction threshold as
compared to those in the array of cylinders, both the symmetric and anti-symmetric
bound states are expected to exist in the zero-order diffraction. The geometry of

1.0
(a) 1.0
0.9
k (units of G)

0.8
0.5
0.7 (c)
0.6
(b) 0.0
0.5
0.5 1.0 1.5 2.0 2.5 3.0 3.5
h (units of D)

1.
x (units of D)

(b)

105
0.
-0.4 0.0 0.4 0.8 1.2 1.6
z (units of D)
x (units of D)

1.
(c) 0

0.
-0.4 0.0 0.4 0.8 1.2 1.6 2.0 2.4
z (units of D)

Fig. 13. (Color on line) Left: Bound state formation in the double grating. The geometry is
sketched above the panel (b). Panel (a): The reflection coefficient as a function of k and k x .
Panel (b): The reflectance (red) and electric field at the midpoint of the grating section (blue)
as functions of k at a fixed value of kx . The range of k and the value of kx are indicated by the
white vertical bar in the panel (a). The incident wave amplitude is one. Right: Panel (a) shows
the specular reflection coefficient as a function of the wave vector k of the incident radiation and
the distance h between the gratings for the fixed value of kx = 0.2G. The white horizontal line
indicates the opening of the first diffraction channel. Panels (b) and (c) show the absolute value
of the electric field for the symmetric (b) and antisymmetric (c) trapped modes as functions of
x and z. The amplitude of the incident field is unity. The fields are obtained for values of (k, h)
indicated by white circles in Panel (a) and corresponding to resonances with the nearly vanishing
width. The grating position is indicated with black lines.
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5226 S. V. Shabanov

the system is sketched above the left panel (b) of Fig. 13. The left panel (a) shows
the reflectance as a function of k and kx (in units of G). In contrast to the left
panel (a) of Fig. 12, there are two close curves along which the reflectance attains
its maxima. They are dispersion curves of the two close trapped modes emerged
due to the coupling. The corresponding pair of resonances is clearly seen in the
left panel (b) of Fig. 13 (red curves). The resonance with the vanishing width (the
bound state) lies at a lower frequency than the corresponding resonance in the
single grating. The resonance is so narrow that its profile cannot be resolved on the
scale of the panel. The profile is actually similar to that of a broader resonance at
a higher frequency, but inverted about its center. Note that the reflection attains
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

its absolute maximum (the system is a perfect mirror) at both the resonances.
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

The blue curve in the same panel gives the electric field at the midpoint of the
grating section as functions of k at a fixed value of kx for the incident wave of
the unit amplitude. The field amplification is higher for the narrow resonance. In
the right panel (a) of Fig. 13, the specular reflection coefficient is presented as a
function of the wave vector k of the incident radiation and the distance h between
the gratings for the fixed value of kx = 0.2G. The first-order diffraction threshold
is shown by a white horizontal line. In each order of diffraction, its maxima form
two curves converging to one another with increasing h. For h < 1 in the zero-order
diffraction, the resonances are well-separated, indicating a strong coupling effect
due to closed channels, while the Fabri–Perot regime is achieved when h > 1.5. Due
to the reflection symmetry of the system, the trapped modes corresponding to these
resonances have a specific parity, i.e., they are either symmetric or anti-symmetric.
Similar to quantum mechanics, the symmetric mode has a lower frequency (energy).
So, the lower curve corresponds to the symmetric mode, the upper one to the anti-
symmetric mode. The right panels (b) and (c) show the absolute value of the electric
field for the symmetric and anti-symmetric modes, respectively, computed at the
values of k and h indicated by the corresponding white circles in the right panel (a)
(the resonances with a nearly vanishing width). As in the case of cylinders, the
electric field is significantly enhanced at some “hot” spots of the structure as seen
in the right panels (b) and (c) of Fig. 13. The electric field is amplified by factor
105 (i.e., nearly 104 -fold of that for the single grating). The amplification factor
at a fixed frequency of the incident radiation can be regulated by changing the
distance h.

6. Resonant Generation of Higher Harmonics


Techniques to control nonlinear optical effects in nanostructures is essential for
achieving an ultimate goal of all-optical signal processing 128,129 (and, perhaps, com-
puting) whose key element should be an optical transistor, an essentially nonlinear
photonic device. The simplest nonlinear effect is a generation of higher harmonics.
The present studies, experimental and theoretical, of efficient methods of generat-
ing higher harmonics include: (i) The band gap effect in photonic crystals130–134 ;
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5227

(ii) enhancement of the fundamental field in defects created (in a controlled way) in
photonic crystals135,136 ; (iii) enhancement of the electromagnetic field due to con-
finement of surface plasmon polaritons for special interface surface geometries 137 ;
and (iv) enhancement of the fundamental field at “hot spots” of gratings.138–141
The field enhancement is determined by the lifetime of the corresponding trapped
mode and may well be described by the Breit–Wigner theory. Hence, one natu-
rally expects that the higher harmonic generation in grating structures is enhanced
when the fundamental field is in resonance with a trapped mode. This is indeed the
case.138–141 However, an even more significant effect can be achieved when both the
fundamental field and the generated higher harmonic are in resonance with suitable
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

trapped modes of the structure.74 The efficiency of this double-resonant higher har-
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

monic generation is determined by the lifetimes of the trapped modes which can be
controlled by using double arrays with variable distance between them. This opens
up a possibility to have controllable nonlinear effects in nanophotonic structures.
The concept is illustrated with the example of a periodic array of parallel dielectric
cylinders.
It has been found that [cf. (2.14)], when the incident radiation is in resonance
with a trapped mode, the electric field on the cylinders is enhanced by the factor
Γ−1/3 with Γ−1 being the trapped mode lifetime. Thus, for the cylinders made of
a nonlinear optical material, an enhancement of the higher harmonic generation is
expected. However, there is a catch in this seemingly natural guess which is related
to the resonant character of the phenomenon. To see this, the nonlinear scattering
problem has to be solved, c2 ∆E − ∂t2 D = 0, with D = εE + (χ/4π)E 2 being the
electric induction (the second harmonic generation is considered). The electric field
E satisfies the radiation boundary conditions at the spatial infinity. As before ε(r)
equals εc inside each cylinder and unity otherwise. Similarly, χ(r) equals a constant
χc inside the cylinders and vanishes otherwise. If χc E  4πεc , which is the case
for a typical nonlinear material, a perturbative approach is justified. The solution
is sought in the form E = Ef + Esh + O(χ2c ) where Ef,sh are the fundamental and
second harmonic fields. They satisfy the following linear system of equations,
χ 2 2
c2 ∆Ef − ε∂t2 Ef = 0 , c2 ∆Esh − ε∂t2 Esh = ∂ E . (6.1)
4π t f
If the fundamental field has the frequency ω, then it follows from (6.1) that Esh
must have the frequency Ω = 2ω. So, put Ef = e−iωt Eω and Esh = e−iΩt EΩ . Note
that the fields in the original system are real. For linear wave equations, one can
extend the solutions to complex valued ones. The physical fields are obtained by
taking the real part of the complex solution. If this rule is adopted, then the right
hand side of the second equation in (6.1) should contain (Re Ef )2 rather than the
complex field Ef2 . Making use of the identity, ∂t2 (Re Ef )2 = (1/2)Re(∂t2 Ef2 ), the
equations for amplitudes are obtained:
χ 2 2
c2 ∆Eω + εω 2 Eω = 0 , c2 ∆EΩ + εΩ2 EΩ = − Ω Eω . (6.2)

November 13, 2009 14:18 WSPC/140-IJMPB 05447

5228 S. V. Shabanov

The solution of the first equation in (6.2) is given by (2.2) and can be written in
the form Eω = E0ω +E + where E0ω = e2πip·r is the incident wave; the scattered field
E + satisfies the equation c2 ∆E + +εω 2 E + = (1−ε)ω 2 E0ω . This equation is identical
to the second equation in (6.2) (ω → Ω, and appropriate modifications of the right-
hand side). This observation immediately allows us to obtain the amplitude EΩ
from the solution E + given in (2.2). Thus, neglecting the variations of the field
within the cylinders,
χc
EΩ (r) = E 2 (p, px )E + (2p, 2px ; r) ,
8π(c − 1) c
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

where Ec (p, px ) is the magnitude of the field on the cylinders; the scattered field
E + (2p, 2px; r) includes all open and closed channels. As p and px are doubled in the
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

scattered field, the lth channel is open, provided (2p)2 − (l − 2px)2 ≥ 0. Suppose, for
example, the fundamental field has only one open channel, p < 1 − px . The second
harmonic field may have up to three open channels, l = 0, ±1.
Let Rshl be the amplitude of the second harmonic field in the lth open channel.
Its explicit form is obtained from (2.3) with the doubled p and px ,
χc
Rsh
l = E 2 (p, px )Rl (2p, 2px ) . (6.3)
8π(εc − 1) c
The functions Ec and Rl are proportional to R0 . The amplitude |R0 (p, px )| has
sharp maxima (resonances) at p = pm defined in (2.12), and quickly decreases as
p deviates from pm [cf. (2.13)]. Thus, the second harmonic amplitude significantly
differs from zero if the maxima of |R0 (p, px )| and |R0 (2p, 2px )| deviate from one
another no more than the corresponding resonance width. In other words, the inci-
dent and second harmonic waves must both be in resonance (or nearly in resonance)
with suitable trapped modes supported by the structure.
The calculated field on the cylinders |Ec | is plotted in the left part of Fig. 14 as a
function of p at a particular value of px = 0.125. The black curve graph represents
|Ec (p, px )|. The highest field is, of course, reached for the longest lived trapped
mode at p = p1 ∼ 1 − 0.125 = 0.875. The function |Ec (2p, 2px )| is graphed with the
dashed red curve. The double resonance can indeed be achieved as there are maxima
of the two functions that nearly coincide. The inset in Fig. 14(right) zooms in the
matching resonances at p ∼ 0.875 when the second harmonic generation is most
efficient. The resonances appears to be only slightly shifted relative one another
and the resonance matching is not far from optimal.
The conversion efficiency I is shown in the left part of Fig. 15 as a function of p
and px . The quantity I is defined as the total flux of the second harmonic generated
by the array (i.e., summed over all open diffraction channels) divided by the flux
of the second harmonic generated by the same incident wave in a homogeneous
dielectric layer that contains the same amount of the nonlinear optical material as
in the array of cylinders. The thickness of such a layer is L = πR 2 /Dg . The results
are presented in the logarithmic scale. The most intense lines correspond to the case
when both the fundamental and the second harmonic fields are in resonance with
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5229

a)

|E |
c
b)

|E |
c
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

|E |
c

0.86 0.87 0.88 c)

|E |
0.5 1 1.5 2 c
0.83 0.84 0.85 0.86 0.87 0.88
k ( units of G ) k ( units of G )

Fig. 14. (Color on line) Left: Field on the cylinders, |Ec (p, px )| (solid black) and |Ec (2p, 2px )|
(dashed red), as a function of the frequency k = pG at px = 1/8 and R/Dg = 0.1. The maxima
of the black curve correspond to the maxima on a section of the plot in Fig. 3 by the vertical line
px = 1/8. The double resonance occurs when the red and black maxima match. The inset is a
magnification of the double resonance at the lowest frequency (p < 1 − px ). Right: The inset at
three different values of R/Dg = 0.05, 0.1, 0.2 (Panels (a), (b), and (c), respectively). By varying
the cylinder radius the double resonance can be achieved (Panel (b)) as explained in the text.

trapped modes of the structure. Along these resonant lines the conversion efficiency
is up to 105 times larger than off-resonance. The less intense lines in the figure
correspond to the case when only the second harmonic field is in resonance with a
trapped mode, while the fundamental field is not. In the right part of Fig. 15, the
logarithm of the efficiency as a function of the frequency is shown at px = 1/8. This
is a cross-section of the color graph in Fig. 15(left) by the vertical line px = 1/8. The
efficiency is clearly maximal when both the fundamental and second harmonic fields
are in resonance with two trapped modes in the system. The absolute maximum
is achieved when the lowest frequency resonance for the fundamental field is used
that is slightly below p = 1 − px = 0.875. This corresponds to the maximal field
achievable on the cylinders in full accord with (6.3).
The double resonance condition needs to be further elucidated. If |R0 (p, px )|
has maxima at p = pm (px ), then the maxima of |R0 (2p, 2px )| occur at p =
pm (2px )/2. A perfect double resonance can be achieved if there exists m such that
qm (px ) = (1/2)q2m (2px ). This yields the condition on the scattering phase, δs (p) =
(1/2)δs (2p). For a small cylinder radius, the scattering phase is quadratic in p so
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5230 S. V. Shabanov

2.0
k total (units of G)

1.5
4

log [,(2Z)] ( arbitrary units )


1.0 3

2
0.5
1
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

0.0 0.1 0.2 0.3 0.4 0.5 0


by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

k parallel (units of G) -1

-2
0.5 1 1.5 2
-3 -2 -1 0 1 2 3 k ( units of G )

Fig. 15. The efficiency I of the resonant second harmonic generation by the array of cylinders as
a function of p and px (k = pG = ω/c). The efficiency I is the ratio of the total second harmonic
pz,l/2 |Rsh 2
P
flux generated by the array, l | , and the second harmonic flux generated by the same
incident wave in a homogeneous layer that contains the same amount of the nonlinear optical
material as in the array of cylinders (the thickness of the layer is L = πR2 /Dg ). The results are
presented in the log10 scale indicated at the bottom. Right: The efficiency of the second harmonic
generation as a function of p at px = 0.875; the cross-section of the left color graph by the vertical
line px = 0.125.

that this condition cannot be met. However, the match is not required to be perfect
for the double resonance to occur. It is sufficient that the distance between the max-
ima does not exceed the width Γ2m (2px ) of the second harmonic resonance. Recall
that Γm (px ) < Γ2m (2px ) according to (2.13). For a small cylinder radius, the max-
ima of |R0 (p, px )| lie closer to the diffraction thresholds than those of |R0 (2p, 2px )|.
The distance between the maxima is ∆qm = (1/2)q2m (2px ) − qm (px ) = 3qm (px ).
In the limiting case studied, Γm ∼ δs3 and ∆qm ∼ δs2 . Hence, for a sufficiently small
radius, the second harmonic generation is suppressed because the fundamental and
second harmonic fields are out of resonance. The distance between the resonances
increases quadratically with the cylinder radius, while the width of the resonances
grows as a cubic power of the radius. Thus, by varying the cylinder radius the res-
onances become broad enough in order for the matching to occur. The analysis is
summarized in the right part of Fig. 14: for a very small cylinder radius there is
a mismatch in the resonances [Panel (a)]; as the radius increases, the resonances
broaden and a good matching is indeed achieved [Panel (b)]; for a further increasing
cylinder radius, numerical simulations show that a mismatch happens again [Panel
(c)], which cannot be concluded from the above analysis because the resonances
become so broad and the perturbative approach no longer applies because R/D g is
not small enough.
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5231

If the dielectric constant of the cylinders depends on the frequency, εc = εc (p)


(a dispersive material), then the perfect matching can be achieved even for a small
cylinder radius, provided εc (p) ≈ 2εc (2p) − 1 in the vicinity p = p1 . Thus, the
resonance matching is facilitated if εc (p) decreases with p in the interval p1 < p <
2p1 . One should however bear in mind that the Kronig–Kramers dispersion relations
require εc to have an imaginary part. This introduces dumping which broadens the
resonance as some of the trapped mode energy goes into heat and, hence, |Rsh l |
decreases.
What happens for the double array of cylinders made of optical nonlinear ma-
terial (or similar double grating structures)? The width Γ1 of the resonance in the
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

fundamental field can be controllably reduced by changing h; the maximal value of


by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

|Ec (p, px )| increases according to (2.14). The effect is independent of the cylinder
radius as proved in Sec. 5.4. The flux carried by the second harmonics is estimated
by
−4/3
|Rsh 2 2 4 2 2
l | ∼ χc |Ec | |R0 (2p, 2px )| ∼ χc Γ1 ,

provided the second harmonic is in resonance ( |Rl (2p1 , 2px )| < 1 does not vanish).
It increases quite rapidly with decreasing Γ1 . The resonance matching can always
be achieved by adjusting the cylinder radius (by broadening the resonance in the
second harmonic field) because the formation of the bound state depends only on
h, not on the single array geometry (Sec. 5.4). Actually, the problem can be solved
analytically.142 The above conclusions indeed hold; the second harmonic generation
can be controlled by varying h and is even more efficient than in the case of the
single array. The analysis appears to be technically very involved and is not given
here; the details will be presented elsewhere.142

7. Concluding Remarks
The numerical and experimental data presented above show that the enhanced
(extraordinary) transmittance or reflectance of periodic subwavelength arrays can
naturally be understood as a resonant scattering phenomenon and therefore be ex-
plained by the resonant scattering theory which has been extensively developed in
quantum theory. In the author’s view, the close analogy between resonant phenom-
ena in quantum and Maxwell’s theory might be very fruitful and beneficial for the
latter. The above review is, in part, an attempt to demonstrate this.
The resonant scattering theory asserts that sharp resonances are associated with
quasi-stationary states (states with a finite life time). Such quasi-stationary states
in Maxwell’s theory are trapped modes of the electromagnetic fields confined in
the scattering structure vicinity. They decay by emitting a nearly monochromatic
radiation. The existence of such trapped modes may be due to both the structure
geometry and structure material (geometrical and material modes). The scaling
symmetry of Maxwell’s equation provides a natural criterion to find out whether
a particular experimentally observed resonance is associated with the structure
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5232 S. V. Shabanov

material dispersive properties or with the structure geometry. When the trapped
modes exhibits the scale invariance, the structure geometry is dominant. The latter
is quite easily verified in experiment by making several structures whose dimensions
are scaled by the same factor. In many cases, the Breit-Wigner theory for resonant
scattering can be applied to qualitatively, and often quantitatively, explain and
predict scattering properties of periodic subwavelength arrays. In particular, the so-
called extraordinary transmittance of periodically perforated thin metal films can
naturally be explained. The concept of coupled of quasi-stationary states developed
in quantum scattering theory is naturally extended to Maxwell’s theory. It predicts
a remarkable phenomenon: bound states of light in the radiation continuum (or
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

resonances with the vanishing width). The field of trapped modes corresponding
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

to narrow resonances is greatly amplified as compared to the field amplitude of


the incident radiation. This feature can be exploited to enhance non-linear optical
effects in subwavelength periodic arrays; in particular, the resonant generation of
higher harmonics provides a good example of this kind. The revealed mechanism
of the bound states formation opens up a possibility to control non-linear optical
effects by controlling the resonance width and, hence, the field amplification of the
corresponding trapped mode. It is believed that the theory of coupled resonances
in Maxwell’s theory may become useful in the future development of technologies
for all-optical data processing where the problem of controllable non-linear optical
effects has to be solved one way or the other in order to design an optical analog
of a transistor.

Acknowledgments
The author is deeply indebted to Andrei Borisov (LCAM, University of Paris-Sud,
Orsay) for his invaluable help with the figures and numerous discussions. The author
is also grateful for useful remarks and references as well as stimulating discussions to
F. J. Garca de Abajo, J.-P. Gauyacq, A. F. Hebard, J. R. Klauder, D. C. Marinica,
R. F. Ndangali, A. K. Sarychev, S. Selcuk, V. M. Shalaev, D. B. Tanner, T. Teperik
and A. P. Vinogradov and thanks the LCAM (University of Paris-Sud, Orsay) for
warm hospitality extended to him during his visit when a part of this work has been
done. The work was supported in part by the University of Florida opportunity
fund.

References
1. R. Wood, Philos. Mag. 4, 396 (1902).
2. R. Wood, Philos. Mag. 23, 310 (1912).
3. R. Wood, Phys. Rev. 48, 928 (1935).
4. L. Rayleigh, Philos. Mag. 14, 60 (1907).
5. U. Fano, Phys. Rev. 50, 573 (1936).
6. U. Fano, J. Opt. Soc. Am. 31, 213 (1941).
7. J. Zenneck, Ann. Phys. IV 23, 846 (1907).
8. A. Sommerfeld, Ann. Phys. IV 28, 665 (1909).
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5233

9. G. Mie, Ann. Phys. IV 25, 377 (1908).


10. P. Debye, Ann. Phys. IV 30, 57 (1909).
11. T. W. Ebbessen, H. J. Lezec, H. F. Ghaemi, T. Thio and P. A. Wolff, Nature (Lon-
don) 391, 667 (1998).
12. H. A. Bethe, Phys. Rev. 66, 163 (1944).
13. L. Venema, Nature Insight: Photonic Technology, Vol. 424 (Nature, London, 2003),
p. 809.
14. R. Ulrich, Infrared Phys. 7, 37 (1967).
15. C. C. Chen, IEEE Trans. Microwave Theory Tech. 19, 475 (1971).
16. R. C. McPhedran, G. H. Derrick and L. C. Botten, in Electromagnetic Theory of
Gratings, ed. R. Petit (Springer-Verlag, Berlin, 1980), p. 227.
17. H. Raether, Plasmons on Smooth and Rough Surfaces and on Gratings (Springer-
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

Verlag, Berlin, 1988).


by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

18. A. V. Zayats, I. I. Smolyaninov and A. A. Maradudin, Phys. Rept. 408, 131 (2005).
19. H. F. Ghaemi, T. Thio, D. E. Grupp, T. W. Ebbessen and H. J. Lezec, Phys. Rev.
B 58, 6779 (1998).
20. E. Popov, M. Neviére, E. Enoch and R. Reinisch, Phys. Rev. B 62, 16100 (2000).
21. L. Martín-Moreno, F. J. García-vidal, H. J. Lezec, K. M. Pellerin, T. Thio, J. B.
Pendry and T. W. Ebbessen, Phys. Rev. Lett. 86, 1114 (2001).
22. L. Salomon, F. Grillot, A. V. Zayats and F. de Fornel, Phys. Rev. Lett. 86, 1110
(2001).
23. R. Wannermacher, Opt. Commun. 195, 107 (2001).
24. W. L. Barnes, W. A. Murray, J. Dintinger, E. Devaux and T. W. Ebbessen, Phys.
Rev. Lett. 92, 107401 (2004).
25. M. M. J. Treacy, Phys. Rev. B 66, 195105 (2002).
26. M. M. J. Treacy, Appl. Phys. Lett. 75, 606 (1999).
27. M. Sarrazin, J.-P. Vigneron and J.-M. Vigoureux, Phys. Rev. B 67, 085415 (2003).
28. H. J. Lezec and T. Thio, Opt. Express 12, 3629 (2004).
29. H. F. Schouten, N. Kuzmin, G. Dubois, T. D. Visser, G. Gbur, P. F. A. Alkemade,
H. Blok, G. W. ’t Hooft, D. Lenstra and E. R. Eliel, Phys. Rev. Lett. 94, 053901
(2005).
30. G. Gay, O. Alloschery, B. Viaris de Lesegno, J. Weiner and H. J. Lezec, Phys. Rev.
Lett. 96, 213901 (2006).
31. G. Gay, O. Alloschery, B. Viaris de Lesegno, C. O’Dwyer, J. Weiner and H. J. Lezec,
Nature Phys. 2, 262 (2006).
32. P. Lalanne and J. P. Hugonin, Nature Phys. 2, 551 (2006).
33. L. Aigouy, P. Lalanne, J. P. Hugonin, G. Julié, V. Mathet and M. Mortier, Phys.
Rev. Lett. 98, 153902 (2007).
34. F. J. García de Abajo, Rev. Mod. Phys. 79, 1267 (2007).
35. H. Levin and J. Schwinger, Comm. Pure Appl. Math. 3, 355 (1950).
36. R. G. Newton, Scattering Theory of Waves and Particles (McGraw-Hill, New York,
1966).
37. O. Keller, Phys. Rev. B 34, 3883 (1986).
38. O. Keller, Phys. Rev. B 37, 10588 (1988).
39. O. Keller, Phys. Rev. B 38, 8041 (1988).
40. C. Girard, X. Bouju and A. Deraux, in Near-Field Optics (NATO ASI Series E242),
eds. D. Pohl and D. Courjon (Dordrecht, Kluwer, 1993), p. 199.
41. A. Dereux and D. Pohl, in Near-Field Optics (NATO ASI Series E242), eds. D. Pohl
and D. Courjon (Dordrecht, Kluwer, 1993), p. 189.
42. O. F. J. Martin, C. Girard and A. Dereux, Phys. Rev. Lett. 74, 526 (1995).
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5234 S. V. Shabanov

43. C. Girard and A. Dereux, Rep. Prog. Phys. 59, 657 (1996).
44. C. Girard, Rep. Prog. Phys. 68, 1883 (2005).
45. M. Reed and B. Simon, Methods of Modern Mathematical Physics. III: Scattering
Theory (Academic Press, New York, 1979).
46. T. V. Teperik, F. J. Garca de Abajo, A. G. Borisov, M. Abdelsalam, P. N. Bartlett,
Y. Sugawara and J. J. Baumberg, Nature Photonics 2, 299 (2008).
47. S. K. Selcuk, K. Woo, D. B. Tanner, A. F. Hebard, A. G. Borisov and S. V. Shabanov,
Phys. Rev. Lett. 97, 067403 (2006).
48. D. C. Marinica, A. G. Borisov and S. V. Shabanov, Phys. Rev. Lett. 100, 183902
(2008).
49. J. von Neumann and E. Wigner, Phys. Z 30, 465 (1929).
50. F. Capasso, C. Sirtori, J. Faist, D. L. Sivco, S.-N. G. Chu and A. Y. Cho, Nature
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

(London) 358, 565 (1992).


by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

51. A. G. Borisov and S. V. Shabanov, J. Comput. Phys. 209, 643 (2005).


52. R. Gómez-Medina, M. Laroche and J. J. Sáenz, Opt. Express 14, 3730 (2006).
53. M. Laroche, S. Albaladejo, R. Gómez-Medina and J. J. Sáenz, Phys. Rev. B 74,
245422 (2006).
54. J. M. Elson, Phys. Rev. B 12, 2541 (1975).
55. F. Toigo, A. Marvin, V. Celli and N. R. Hill, Phys. Rev. B 13, 5618 (1977).
56. G. S. Agarwal, Phys. Rev. B 15, 2371 (1977).
57. R. Petit, Electromagnetic Theory of Gratings. Topics in Current Physics. Vol. 22
(Springer-Verlag, Heidelberg, 1980).
58. D. Maystre, Progress in Optics. Rigorous Vector Theories of Diffraction Gratings,
Vol. XXI (Elsevier, Amsterdam, 1984).
59. F. Montiel and M. Naviere, J. Opt. Soc. Am. A 11, 3241 (1994).
60. D. Barchiesi and D. Van Labeke, J. Mod. Opt. 40, 1239 (1993).
61. D. Barchiesi and D. Van Labeke, Microsc. Microanal. Microstruct. 5, 1 (1994).
62. D. Van Labeke and D. Barchiesi, Opt. Commun. 114, 470 (1995).
63. A. Senetac and J.-J. Greffet, Ultramicroscopy 57, 246 (1995).
64. J.-J. Greffet, A. Sentenac and R. Carminati, Opt. Commun. 116, 20 (1995).
65. F. de Fornel, P. M. Adam, L. Salomon, J. P. Goudonner, A. Sentenac, R. Carminati
and J.-J. Greffet, J. Opt. Soc. Am. A 13, 35 (1999).
66. J. C. Weeber, C. Girard, J. R. Krenn, A. Dereux and J. P. Goudonnet, J. Appl.
Phys. 86, 2576 (1999).
67. L. Salomon, C. Charbonnier, F. de Fornel, P. M. Adam, P. Guérin and F. Carcenac,
Phys. Rev. B 62, 17072 (2000).
68. Y. Pagani, D. Van Labeke, B. Guizal, A. Vial and F. Baida, Opt. Commun. 209,
237 (2002).
69. C. Girard and R. Quidant, Opt. Express 12, 6141 (2004).
70. V. Twersky, J. Opt. Soc. Am. 52, 145 (1962).
71. B. A. Lippmann and J. Schwinger, Phys. Rev. 79, 469 (1950).
72. B. A. Lippmann, Phys. Rev. Lett. 16, 135 (1966).
73. L. D. Landau and E. M. Lifshitz, Field Theory, Theoretical Physics, Vol. II (Perga-
mon, London, 1960).
74. D. C. Marinica, A. G. Borisov and S. V. Shabanov, Phys. Rev. B 76, 085311 (2007).
75. P. M. Morse and H. Feshbach, Methods of Theoretical Physics (McGraw-Hill, New
York, 1953).
76. U. Fano, Phys. Rev. 124, 1866 (1961).
77. R. Magnusson and S. S. Wang, Appl. Phys. Lett. 61, 1022 (1992).
78. S. Peng and G. M. Morris, Opt. Lett. 21, 549 (1996).
November 13, 2009 14:18 WSPC/140-IJMPB 05447

RLS and HHG 5235

79. T. Peter, R. Bräuer and O. Bryngdahl, Opt. Commun. 139, 177 (1997).
80. L. Pilozzi, A. D’Andrea and R. Del Sole, Phys. Rev. B 54, 10763 (1996).
81. K. Koshino, Phys. Rev. B 67, 165213 (2003).
82. A. G. Borisov and S. V. Shabanov, J. Comput. Phys. 199, 742 (2004).
83. F.-M. Dittes, Phys. Rept. 339, 215 (2000).
84. F. J. Garcı́a-Vidal and L. Martn-Moreno, Phys. Rev. B 66, 155412 (2002).
85. Y. Takakura, Phys. Rev. Lett. 86, 5601 (2001).
86. Q. Cao and Ph. Lalanne, Phys. Rev. Lett. 88, 057403 (2002).
87. J. M. Steele, C. E. Moran, A. Lee, C. M. Aguirre and N. J. Halas, Phys. Rev. B 68,
205103 (2003).
88. A. Christ, T. Zentgraf, J. Kuhl, S. G. Tikhodeev, N. A. Gippius and H. Giessen,
Phys. Rev. B 70, 125113 (2004).
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

89. T. Zentgraf, A. Christ, J. Kuhl, N. A. Gippius, S. G. Tikhodeev, D. Nau and H.


by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

Giessen, Phys. Rev. B 73, 115103 (2006).


90. A. G. Borisov, F. J. Garcı́a de Abajo and S. V. Shabanov, Phys. Rev. B 71, 075408
(2005).
91. R. Harrington and D. Auckland, IEEE Trans. Antennas Propag. 28, 616 (1980).
92. Y. Takakura, Phys. Rev. Lett. 86, 5601 (2001).
93. J. R. Suckling, A. P. Hibbins, M. J. Lockyear, T. W. Preist, J. R. Sambles and C.
R. Lawrence, Phys. Rev. Lett. 92, 147401 (2004).
94. T. H. Isaac, J. Gómez Rivas, J. R. Sambles, W. L. Barnes and E. Hendry, Phys. Rev.
B 77, 113411 (2008).
95. A. J. F. Siegert, Phys. Rev. 56, 750 (1939).
96. R. G. Newton, Scattering Theory of Waves and Particles (Springer-Verlag, New
York, 1982).
97. R. M. More and E. Gerjuoy, Phys. Rev. A 7, 1288 (1973).
98. O. I. Tolstikhin, V. N. Ostrovsky and H. Nakamura, Phys. Rev. Lett. 79, 2026 (1997).
99. A. Rung and C. G. Ribbing, Phys. Rev. Lett. 92, 123901 (2004).
100. C. J. Bouwkamp, Rep. Prog. Phys. 17, 35 (1954).
101. Y. Leviatan, J. Appl. Phys. 60, 1577 (1986).
102. A. Roberts, J. Opt. Soc. Am. A 4, 1970 (1987).
103. A. Roberts, J. Appl. Phys. 65, 2896 (1989).
104. A. Roberts, J. Appl. Phys. 70, 4045 (1991).
105. U. Dürig, D. W. Pohl and F. Rohner, J. Appl. Phys. 59, 3318 (1986).
106. J. M. Vigoureux and D. Courjon, Appl. Opt. 31, 3036 (1992).
107. J. M. Vigoureux, F. Depasse and C. Girard, Appl. Opt. 31, 3036 (1992).
108. T. Thio, H. F. Ghaemi, H. J. Lezec, P. A. Wolff and T. W. Ebessen, J. Opt. Soc.
Am. 16, 1743 (1999).
109. C. Sonnichsen, A. C. Duch, G. Steininger, M. Koch, G. von Plessen and J. Feldmann,
Appl. Phys. Lett. 76, 140 (2000).
110. A. K. Sarychev, V. A. Podolskiy, A. M. Dykhne and V. M. Shalaev, IEEE J. Quant.
Elec. 38, 956 (2002).
111. A. G. Schuchlinsky, D. E. Zelenchuk and A. M. Lerer, J. Opt. A: Pure Appl. Opt.
7, 102 (2005).
112. N. Papasimakas, V. A. Fedotov, A. S. Schwanecke, N. I. Zheludev and F. J. Garcı́a
de Abajo, Appl. Phys. Lett. 91, 081503 (2007).
113. B. A. Munk, Frequency Selective Surfaces and Design (Wiley, New York, 2000).
114. G. Craven and R. Skedd, Evanescent Mode Microwave Components (Artech House
Publishers, New York, 1987).
115. A. K. Sarychev and V. M. Shalaev, Phys. Rept. 335, 275 (2000).
November 13, 2009 14:18 WSPC/140-IJMPB 05447

5236 S. V. Shabanov

116. J. B. Pendry and A. Mackinnon, Phys. Rev. Lett. 69, 2772 (1992).
117. J. B. Pendry, J. Mod. Opt. 41, 209 (1994).
118. S. G. Tikhodeev, A. L. Yablonskii, E. A. Muljarov, N. A. Gippius and T. Ishihara,
Phys. Rev. B 66, 045102 (2002).
119. C. D. Ager and H. P. Hughes, Phys. Rev. B 44, 13452 (1991).
120. D. M. Whittaker and I. S. Culshaw, Phys. Rev. B 60, 2610 (1999).
121. H. Feshbach, Ann. Phys. (N.Y.) 5, 357 (1958).
122. H. Feshbach, Ann. Phys. (N.Y.) 19, 287 (1962).
123. L. Fonda and R. G. Newton, Ann. Phys. (N.Y.) 10, 490 (1960).
124. H. Friedrich and D. Wintgen, Phys. Rev. A 32, 3231 (1985).
125. J. Okolowicz, M. Ploszajczak and I. Rotter, Phys. Rept. 374, 271 (2003).
126. A. Z. Devdariani, V. N. Ostrovsky and Yu. N. Sebyakin, Sov. Phys. JETP 44, 477
Int. J. Mod. Phys. B 2009.23:5191-5236. Downloaded from www.worldscientific.com

(1976).
by UNIVERSITY OF HONG KONG on 06/18/13. For personal use only.

127. R. F. Ndangali and S. V. Shabanov, Bound States in the Radiation Continuum for
Periodic Subwavelength Arrays of Dielectric Cylinders (2009) (submitted to J. Math.
Phys.).
128. M. Soljačić and J. D. Joannopoulos, Nature Materials 3, 211 (2004).
129. G. A. Wurtz, R. Pollard and A. V. Zayats, Phys. Rev. Lett. 97, 057402 (2006).
130. P. P. Markowicz, H. Tiryaki, H. Pudavar, P. N. Prasad, N. N. Lepeshkin and R. W.
Boyd, Phys. Rev. Lett. 92, 083903 (2004).
131. K. Sakoda and K. Ohtaka, Phys. Rev. B. 54, 5742 (1996).
132. M. Scalora, M. J. Bloemer, A. S. Manka, J. P. Dowling, C. M. Bowden, R.
Viswanathan and J. W. Haus, Phys. Rev. A 56, 3166 (1997).
133. Y. Dumeige, I. Sagnes, P. Monnier, P. Vidakovic, I. Abram, C. Mériadec and A.
Levenson, Phys. Rev. Lett. 89, 043901 (2002).
134. M. Centini, C. Sibilia, M. Scalora, G. D’Aguanno, M. Bertolotti, M. J. Bloemer, C.
M. Blowden and I. Nefedov, Phys. Rev. E 60, 4891 (1999).
135. T. V. Dolgova, A. I. Maidykovski, M. G. Martemyanov, A. A. Fedyanin, G.
Markowsky, V. A. Yakovlev and G. Mattei, Appl. Phys. Lett. 81, 2725 (2002).
136. M. W. McCutcheon, G. W. Rieger, J. F. Young, D. Dalacu, S. Frédérick, P. J. Poole
and R. L. Williams, Appl. Phys. Lett. 81, 2725 (2002).
137. E. Verhagen, L. Kuipers and A. Polman, Nano. Lett. 7, 334 (2007).
138. J. A. H. van Nieuwstadt, M. Sandtke, R. H. Harmsen, F. B. Segerink, S. Enoch and
L. Kuipers, Phys. Rev. Lett. 97, 146102 (2006).
139. M. Airola, Y. Liu and S. Blair, J. Opt. A: Pure Appl. Opt. 7, S118 (2005).
140. W. Fan, S. Zhang, N.-C. Panoiu, A. Abdenour, S. Krishna, R. M. Osgood, Jr., K.
J. Malloy and S. R. J. Brueck, Nano Lett. 6, 1027 (2006).
141. W. Fan, S. Zhang, K. J. Malloy, S. R. J. Brueck, N. C. Panoiu and R. M. Osgood,
Jr., Opt. Express 14, 9570 (2006).
142. R. F. Ndangali and S. V. Shabanov, Higher Harmonic Generation by Double Arrays
of Dielectric Cylinders (2009) (in preparation).

You might also like