structured - tspace
TRANSCRIPT
LINEAR AND NONLINEAR OPTICAL PROPERTIES OF ARTIFICIALLY STRUCTURED MATERIALS
Suresh Pereira
A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy
Graduate Depart ment of Physics University of Toronto
Copyright @ 2001 by Suresh Pereira
Natianat tibrafy I * D Bibliothèque nationaie du Canada
395 W ~ g O o n Street 395, nie weniigml Oitawa ON K 1 A W ûtütwaON K 1 A W Canada Canada
The author has granted a non- exclusive licence allowing the National Libmy of Canada to reproduce, loan, districbute or seli copies of t h thesis in rnicroform, paper or electronic formats.
The author retains ownership of the copyright in this thesis. Neither the thesis nor substantial extracts &om it may be printed or otherwise reproduced withouî the author's permission.
L'auteur a accordé une licence non exclusive pennem B la Bibiiothèque nationale du Canada de reproduire, prêter, distribuer ou vendre des copies de cette thèse sous la forme de microfiche/nIm, de reproduction çm papier ou sur fomat 6Iectronique.
L'auteur conserve la propriété du droit d'auteur qui protège cette thèse. Ni la thèse ni des extraits substantiels de celle-ci ne doivent être imprimés ou autrement reproduits sans son autorisation.
Abstract
Linear and Noalineax Opticd Properties of Arti6cially Structurecl Materiais
Suresh Pereira
Doctor of Philosophy
Graduate Department of Physics
University of Toronto
2001
We begin by deriving a set of equations that describe pulse propagation in one di-
mensional, periodic media, in the presence both of birefringence and a Kerr nonlinearity
'Ne use these equations to interpret the results ot a mies of experiments performed in
Bber gatings, which, with the appropriate approximations, c m be considered to have
one effective dimension.
WC ncxt turn to thc consideration of ho diamel a-a'~xguides coupleci bÿ a sequace of
periodicdy spaced microresonators, in the presence of a Kerr nonlinearity. We show that
h o distinct types of gaps open in the dispersion relation of the device. The frequency of
one type of gap is related to the spacing of the resonators. The fkequency of the other
type of gap is related to the radius of the resonators. We derive a set of coupled nonlinear
Schrodinger equations (NLSE) to describe the propagation of light in the system. We
show that the properties of the dispersion reiation in the vicinity of the two types of gaps
are markedly different, and that near the gap associateci wit h the radius of the resonators,
the group velocity dLspersion experienced by a pulse is very small. We then demonstrate
that a gap soliton should be observable at much lower intensities in this latter gap than
in a Bragg gap of the same frequency width.
We study the operation of a grating-~aveguide structure (GWS), where a grating
is used to coupled an incident plane wave into a guided mode of a layered medium.
We derive equations that determine the field everywhere in the presence of a grating of
arbitrary t hiclmes, and a Kerr nonlineari~. We demonstrate that the GWS can be used
as a low-loss, narrow-band ref'iector, or as an ail optical switch.
Finally, we construct a Hamiltonian formulation for pulse propagation equations in
a one dimensional, Kerr nonlinear, periodic medium. In doing so, we clear up some
confusion in the literature surrounding the nature of the conservai quantities associatecl
wit h Kerr nonlinear pulse propagation equations.
Acknowledgement s
This thesis is for Sun Young, Raoul, Phoebe, Kevin, Aubert and the Duck, each of
whom should know what they contnbuted.
I wodd like to thank the foilowing people for helping to make this thesis what it is.
Di. RE. Slusher and Dr. S. Spater for their perseverance on the fiber experiments and
Dr. G. Marowsky, Dr. M-A Bader, Dr. H-M Keller for their enthusiasm and guidance.
1 would also like to thank Dirk and Uwe for the^ generous hospitality in Gottingen!
Furthermore, for t heir help in maintainhg my sanity, 1 am indebted to Dr. J. Levhe and
Dr. S. Winters, the latter of whom, if justice d e s this world, will be Professor Wmters
by the t h e this thesis is bound.
Primarüy, though, 1 would like to thank Dr. J.E. Sipe for his concise explmations of
the ha points of phy~ics; and for providing iui deguit picture of cornpetence during his
imprisonrnent in administration. Hold tight, sir - I'm working on a plan to bust you out
by September 2002.
Contents
1 Introduction 1
2 Pulse propagation in birefrhgent. nonlinear media with deep gratings 11
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Linear Equations and Basis Functions . . . . . . . . . . . . . . . . . . . . 14
2.2.1 Periodic Structures . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Noniinearity and Multiple Scales Analysis . . . . . . . . . . . . . . . . . 17
2.3.1 Multiple Scales Analysis . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 One Principal Component; s=2: CNLSE . . . . . . . . . . . . . . . . . . 20
2.5 Two Principal Components; s=l: CME . . . . . . . . . . . . . . . . . . . 24
2.5.1 Weak Grating Limit of the NLCME . . . . . . . . . . . . . . . . . 29
2.6 Connecting the CNLSE and the NLCME . . . . . . . . . . . . . . . . . . 30
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7 Numerical Simulations 35
2.7.1 Cornparhg the CNLSE and CME . . . . . . . . . . . . . . . . . . 35
2.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3 Polarkation effects in birefringent. periodic. nonlinear media 41
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
. . . . . . . . . . . . . . . . . . . . . . . . 3.2 Theory for an hfbîte grating 44
. . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Modelling the Grating 44
3.2.2 Coupled Noniinear Schrddinger E<iuations . . . . . . . . . . . . . 46
3.3 Physicd Grating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.4 Approximate Solution for Polarization Evolution . . . . . . . . . . . . . . 52
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Experimental Data 55
3.5.1 Polarizat ion Evolut ion for high det unings . . . . . . . . . . . . . . 58
. . . . . . . . . 3.5.2 Polarization Instability as a function of Phase Lag 60
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4 Gap soliton propagation in a two-channe1 SCISSOR structure 66
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2 Linear theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.2.1 Dispersion relation for the bmChannel SCISSOR . . . . . . . . . 73
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Coupled NLSE 77
4.3.1 Multiple Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.3.2 Nonlinear Response . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.3.3 Coupled NLSEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5 Theory for a grating-waveguide structure with Kerr nonlinearity 94
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Introduction 94
5.2 The GWS and the Guided Modes . . . . . . . . . . . . . . . . . . . . . . 96
. . . . . . . . . . . . . . . . 5.3 Green Function Theory for Stratifiecl Media 98
5.3.1 Grating, No Nonlinearity . . . . . . . . . . . . . . . . . . . . . . . 104
5.3.2 No grating. Kerr Nonlineariw . . . . . . . . . . . . . . . . . . . . 105
5.3.3 nansfer matrices at the Interface . . . . . . . . . . . . . . . . . . 206
5.4 Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.4.1 Low loss rdector in the UV . . . . . . . . . . . . . . . . . . . . . 108
5.4.2 NonlineaxswitchinginaGWS . . . . . . . . . . . . . . . . . . . . 110
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6 Hamiltonian formulation for pulse propagation equations in a periodic.
nodnear medium 114
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.2 Canonical formulation of Maxwell's Equations . . . . . . . . . . . . . . . 116
6.2.1 Linear. Periodic Medium . . . . . . . . . . . . . . . . . . . . . . . 118
6.2.2 Periodic Medium with a Kerr noniinearity . . . . . . . . . . . . . 119
6.3 Muced Hwniltonian and the NLSE . . . . . . . . . . . . . . . . . . . . 122
6.3.1 Effective Fields and Envelope Functions . . . . . . . . . . . . . . 128
6.4 Conserved Quantities of the Hamiltonian . . . . . . . . . . . . . . . . . . 130
6.5 an the use of the Dual Field . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
7 Conclusion 137
Bibliography
Chapter 1
Introduction
In the past several decades, the wide research into artificidy structured materials (AS&)
has led to numerous technological applications, ranging kom biological sensors to waw
length division multiplexiug devices. Beyond their value to industry, ASiLIs offer a host
of challenges to researchem in basic physics. For cxamplc, the microresonator devices
that have been investigated in the past few years could potentiaily be wed to investigate
opt ical shock formation [Il, and have been used to study cavity quantum electrodynam-
ics [2]. In addition to these effects, ASMs present engaging geometries for the study of
nonlinear dynamics, and can be used as d-optical switches (31, and d-optical logic gates
141 - The material in this thesis is centred around the linear and Kerr nonlinear proper-
ties of optical pulse propagation in AS&. A Kerr nonlineazity is often described by
the introduction of a nonlinear index of refraction coefficient, n,, whereby a pulse with
intemie I will experknce an effective index of refraction nef/ = n + nzI, where n is the
background index of refraction in the lirnit I -r O (51. The pulses considered here are of
picosecond duration, with ca.rrier hequencies in the near IR to near W, so that the en-
velope function of the puise is slowly varying relative to the carrier kequency. There are
two m a h reasons for this. First, some of the most promishg applications of the systerns
that are studied here are in telecornmunicatiors, where the standard pulse duration is
in the picosecond range. Second, it is often better to M y understand nonlinear dects
for slowly varying pulses before moving to shorter pulses, where higher-order nonlinear
efTects can complicate the interpretation of experiments. The derivat ions presented in
this thesis c m alI be extendeci to describe the propagation of shorter pulses.
In chapters two and three of this thesis, the properties of birefnngent, periodic, Kerr
media with one effective dimension are studied. Here periodic means that both the index
of re£raction and the nonlinear index coefficient vary spatially with period d. Bir-gence
refers to the fact that the index of refkaction experienced by the two polarkations of
light are unequal, R, # Q, where x and y are the principal axes of polarization. We
introduce nb = - E, to quantify this birefringence. Examples of media that can
be considered to have one effective dimension include fiber Bragg gratings and etched
dielectric waveguides, when the etching is shallow. In these systems, the transverse
directions can be accotinted for by defining a mnde profile that remaine unchangecl during
propagation. For pulse propagation wit h intensities sufliciently low t hat nonlinear effects
are negligible, a medium is characterized by its dispersion relation, w (k) , which relates
the kequency of the Light, w , to the corresponding wave number, k. It is weil known
that an infinite periodic medium in one dimension always poseses a photonic band
gap - a range of frequencies in which light c m o t propagate [6]. This photonic band
gap is centred around the Bragg frequency, wo, of the medium, which is the bequency
at which the coherent rdections at the va.rious interfaces of the medium can build up.
Heuristically, this requires the reflected light kom one int d a c e t O be phase-matched
with that fiom another interface. The precise value of this Bragg bequency is diçcussed
below, after Equation (1.1).
In many practical periodic systems, such as fiber Bragg gratlligs, the strength of the
periodic variation in the index of rekaction (the index mntrast, bn) is very small relative
to the average background index of refraction (6nF N IO-^). This allowed previous
researchers to treat Light in the medium as a forward-propagating wave that experiencled a
weak coupling to a backward-propagating wave via the grating (and vice versa). A similar
analysis can be carried out with thinly etched waveguides, because there, aithough the
index contrast between the wawguide material and air is quite strong, the region of the
etching is s m d relative to the transverse dimensions of the mode profile. Furthemore,
it was assumed that the intensity of the pulses being described was such that the effect
of the Kerr nonlinearity codd be considered a small perturbation to the linear results.
Such analyses led to the derivation of the heuristic coupled-mode equations (CME), which
describe the evolution of slowly varying endope functions, A* (2, t ) , propagating in the
forward (+) and backward (-) directions , camed at the Bragg wavenumber (ko = r / d )
and the Bragg fiequency (wo = cko/x) of the periodic medium. The equations are [3]
5 a A - .dA, O = 2--
c at - 2- + KA+ +r { I A - ~ ~ + 21~+1' ) A-, dz
where n = (6n/2n) ko, and 7 = nzwo/c. In the derivation of these equations, it was
assumed that the pulse intensiw is SufEciently low that higher-order nonlinearities can
be ignored, and that the fkquency content of the pulse was close to the Bragg frequency.
In recent years fiber Bragg gratings have been grown with index contrasts as high as
4% of the background index, and Bragg stacks c m be constructed in which the index
constrast between adjacent layers is roughly equal to the average index. Both of these
situations cast the validity of the heuristic CME in doubt. Nevertheles, even in a medium
with a high index contrast, it is to be expected that the physics of pulse propagation, for
pulses whose &equency is close to the photonic band gap of the medium. is accurately
described by some sort of CME. This was confimeci by de Sterke et al. [7], who derived
the CNE using the underlying Bloch funaions of the periodic medium. The hear portion
of the CME derived by de Sterke et al. is equident to (1. l), except that the definitions
of wo and n are taken directly £rom the dispersion relation of the linear, periodic medium-
The nonlinear portion cont ains the terms in (1.1) , but adds a series of more complicated
nonlinear interactions.
In chapter two of this thesis, the e£Forts of de Sterke et cd. are extendeci to include
the e f k t s of birehgence. When attempting to derive equations that govem the prop
agation of slowly-varying envelope functions, it quickly becornes apparent that there are
a number of different length, t h e , and strength scales in the problem. The period, d,
is much srnatter than the spatial width of the pulse; the carrier frequency of the pulse is
much larger than the spread in frequencies containeci within the pulse; the value of the
background index of rekaction, n, is much Iarger than both the birehgence, nb, and
the nonlinear portion of the effective index, n21. We use the 'method of multiple scaies'
to carefdy account for these different quantities [8]. The value of this method is that it
effectively dram out the underlying physics of the systems being studied.
The equations that are deriveci for bire-ent media are similar in form to (1.1),
but include nonlinear couphg between orthogonal polarbations of light. The C m are
moût usehl when the frequency content of the puise is close to, or within, the photonic
band gap of the medium. When the frequency content of the pulse is away from a
photonic band gap, it is shown that the propagation of light is weli described by a set of
coupled nonlinear Schr&i.inger equat ions (NLSEs) . The orthogonal polarizat ions are st ill
coupled by the nonlinearityl but the effect of the grating is to modify the values of the
group velocity (aw (k) lak), and group velocity dispersion (a2w (k) / a k 2 ) relative to their
values in the absence of a grating. As the carrier hequency of the pulse is tuned closer
to the photonic band gap, the group velocity is reduced, tending towards zero, and the
group velocity dispersion is enhmced (by orders of magnitude relative to typical material
dispersions). The reduced group velociw and enhancd dispersion can be understood as
a consecpence of light being refiected inside the grating, but without the refîections
being entirely phase-matched, so that they do not build up. The reàuction in group
velociw leads to a kequency-dependent effenective nonlinear index co&cient, raYf (w),
that is enhanceci for frequencies cloaer to the band gap for two reasons: first, the puise
is travelling more slowly, and hence has more time to interact with the nonlinea.ri@; and
second, the intensity distribution of the underlying Bloch functions of the periodic system
affects the manner in which the pulse interacts with the nonlinearity in the system.
The CME and NLSEs are connected in the sense that both can be used to accurately
describe pulse propagation when the carrier frequency is near, but not inside the gap.
The CME are the appropriate equations when the c h e r frequency is inside or very close
to the gap; the NLSEs are appropriate when the carrier frequency is detuned from the
gap, or when the pulse width is sctremely broad. In fact, the NLSE can be used to
describe broad pulses whose frequency content is within the gap. It is well known that
the NLSE can support soliton solutions - solutions where the group velocity dispersion
is perfectly balanced by the nonlinearity so that the profile of the pulse does not change
[SI. If one excites a soliton at the band edge, then its group :*clccity is zcm, and thc
resulting energy distribution is c d e d a gap soliton.
The advantage of developing a set of coupled NLSEs is that it allows a convenient
point of connection with existing literature on nonlinear pulse propagation in birefringent
media. In particular, the literature contains a great deal of research on the existence
and observation of vector solitons in optical fibers without a Bragg grating, and on the
phenornenon of nonlinear energy exchange between the polarizations. The initial work on
birefringent d&ts in optical fibers was done under the assumption that the birefringence
(nb) of the system is large relative to the value of nonlinear index of refraction that is
induced by the intensity ( 7 4 . Under such an assumption, the effects of nonlinear energy
exchange between polarizations can be ignored, because it will not be phase-matched, and
hence not build up. This is analogous to the situation of light with fiequency content far
removed £iom a photonic band gap of the syjtem. It was later pointed out that inclusion
of nonlinear energy exchange can have a number of &riking effects, includùig radiation
Figure 1.1: A two-chnel sequence of spaced, side-coupled resonators is studied in chag
ter three. T m types of gaps open in the dispersion relation of the device. One is
associated with Bragg reflection, and related to the spacing, d. The other is associated
with the resonant fkequency of the resonators, and is related to the resonator radius.
of energy fiom a soliton-like state, and the formation of new vector soliton states 191.
Whether the effects of nonlinear energy exchange can be ignored in puise propagation
t hrough optical fibers wit h no grating can be detemiined by comparing the quant ity nz 1
to ng. In a grating the appropriate nodinear quanti@ is n 2 1 , which is a strong function
of fkequency dettrning. In s birefringent gmting, the bircfringcncc of thc background
medium is not as si&cmt as the effective birefringence, ntff, which is reiated to the
manner in which the orthogonal polarizations accumulate phase in the presence of the
grating. Both n;lf and nb are stmng functiom of the c-er fkequency of the pulse. This
means that for a pulse of given intensity, detuning the carrier hequency has the &ect of
moving the pulse fkom the regime in which energy evchange is disdowed ( n i f f » nyf 1)
to a regime where it is dowed (niff 5 n F f l ) . The effective birehgence, and the
fkequency dependence of the nonlinear regimes in a periodic system are studied in chapter
three of this thesis.
The basic theory presented in chapter two cm be extended to aid in the investigation
of periodic media where at any given point the light is conhed in a direction transverse to
its direction of propagation, but which require a more complicated analysis to determine
the Bloch fimctions. An example of such a medium is a series of periodically spaced
microresonators that sit between two Channel waveguide, shown in Figure 1.1. We study
this system in chapter four. Light travelling in either of the charineh can couple into,
and circulate around, the microresonators. A pulse sent through such a system will
propagate either forward ûr backward, but the one-dimensional expressions for the Bloch
hinctions that were used in chapter two are no longer appropriate. In moving to a higher
dimensional theory, it is necessary to generalize the mathematical tools of chapters two
and three.
In the 1 s t several years, microresonators have attracted a great deal of interest as
device elements that can couple light between waveguides, or allow light to turn a 90°
bend. They can be fabricated with low los, and high quality factors [IO]. If the medium
from which the microresonator is fabricated is linear, then a given microresonator has a
resonant Bequency w r = M ( C ~ E R ) , where M is an integer, n is the index of refkaction
of the microresonator, and R is its radius. If the resonators are spaced with period
d, then Bragg refktion leads to the opening of a photonic band gap centred around
f rq i i~nc ie s uf = Rr !m! ( ~ d ) ) (a Bragg gap), where N k an integer. Thio reflection
occurs because light travelling in the forward direction in the lower channel of the system
is weakly coupled, via the microresonator, to light travelling in the backward direction of
the upper Channel. However, in addition to the photonic band gaps associated with Bragg
rdection, t here also exist gaps associat ed wit h the resonances of the microresonat or (a
resonator gap). The Bragg gaps open because the relatively s m d coupling between
directions at hequency wb can build up; the resonator gaps open because at fiequency
w, the coupling between directions is immense.
The two types of gaps associated with the microresonator stmcture have a rather
different character. In both cases the group velocity tends t o m & zero, but while the
group velocity dispersion around a Bragg gap can becorne enormous, around a resonator
gap it becornes very s m d . In chapter four it is shown that in the presence of an opticd
Kerr nonhearity, the propagation of pulses with carrier fkquencies near either gap is
w d described by a NLSE, but since the group velocity dispersion near a resonator gap
Incident \
Figure 1.2: In a grating-waveguide structure (GWS) , st udied in chapt er four, an incident
plane wave is coupled to a guided mode of a l a y d medium via a gating. The guided
mode is coupled back out, making the device 100% reflecting when the appropriate
resonance condition is met.
is much smder, it should be possible to excite a gap soliton with much less energy.
In the lirnit where the upper channel is removeci, both types of photonic band gaps
of the system vanish (because it is assumed that coupling into the resonator incurs no
reflection, so the foxward and backward propagation directions are cornpletely uncou-
pled). In such a situation the spacing between resonators becomes irrelewt - only the
average densiw of resonators is important. An interesting feature of this system is that
as the caxrier kequency of a pulse nears w,, the group velocity of the pulse tends towards
zero, and the group velociw dispersion becomes zero (neglecting the dispersion of the
background medium). Roughly speaking, the group velocity becomes s m d because the
iight spends a long tirne circdating in each resonator. For a monochromatic excitation,
the intensity in each resonator can become enormous. This build-up in intensity c m be
used to obsenre low-threshold nonlinear switching, or to observe twephoton absorption
for biologicd sensing applications.
Another system in which the field intensitJt can become enonnous is a grating-
waveguide structure (GWS) (1 11, shown in Figure 1.2. This system consists of a thin
grating atop a waveguidmg layer. Underneath the guiding layer is a substrate. It is
well known that a grating can couple an incident plane wave into a mode of the guiding
structure - this is the principle behind a conventional grating coupler. Roughly speak-
hg, if the grating is t h , light couples slowly into the device, but bounces around for
a long time before escaping, which leads to a huge field intensity. Such an intensity can
be used either for nonlinear switching, or for sensing applications, but suffers bom the
fact that the fiequency range over which the reflection occurs is tiny. Conversely, if the
grating is thick, then light can couple in and out of the system much more quickly, so
the field intensity in the guiding layer is much smder? but the reflection occurs over a
large kequency range.
The theory of a GWS is examineci in chapter five, for the general case of a thick
m~ting, and a Kerr nonünear egcbg layer, ushg a Green Emtion theor). for surface O
optics. The theory is designed to integrate with the familiar Frgnel transmission and
rdection coeflicients of the layered medium. It is used to describe both the linear and
nonlinear properties of the system for plane-wave excitation. As an example of the
u s e f i e s of the GWS, we present a scheme for a narrow-band, low-loss reflector of W
light , for a GWS with no nonlinearity; we also present a scheme for d-optical switching
in a GWS with a Kerr nonlinear layer.
The material in this thesis has, to this point, followed a certain arc. The work in
chapters two and three is concerned with strictly onedimensional systems; in chapter
four, a quasi one-dimemional system is studied, in which the Bloch functions were in a
higher dimension, but where the pulse propagation was stiU in one dimension; in chapter
£ive, the GWS requires a M y dimensiona al th- The work in the last chapter
of this thesis returns to strictly one dimensional problems, but with a different goal in
mind: the construction of a canonical Hamiltonian formulation for pulse-propagation
equations, where 'canonical' means that the Hamiltonian is equal to the energy in the
electromagnetic field, and derives the correct equations of motion using the Heisenberg
equations of motion. The motivation for this work is three-fold. First, the dynamics of
Kerr nonlinear systems have often been studied using an effective IIainiltonian that was
not equd to the energy in the electrornagnetic field but that, newrtheless, generated the
appropriate equations of motion. This led to a certain confusion because it was uncertain
how to interpret the three conserved quantities: energy, momentum, and the effective
Hamiltonian. In this thesis, conhision about the conserved quantities is avoided by using
a Hamiltonian that is equal to the energy in the electromagnetic field. It is then shown
that the quantity previously Iabelled 'energy' is, in fact, the conserved charge associated
with phase-translation symmetry, and is equal to the tme energy oniy to zeroth order.
The second motivation for this work is to develop a more effective methodology for the
constmction of pulse propagation equations in media vvith periodicity in two and three
àimensions. The Harniltonian formulation presented here is used to derive a NLSE, and
it can easily be generalized to two or three dimensions. A third motivation for this work
is that the Hamiltonian formulation presented here can be generalized to give a quantum
mechanical description of the fields.
Chapter two of this thesis has previously been published in Physical Review E. Parts
of chapter three will be published in the near future as a chapter in a Springer-Verlag
adMncd topic book. The work in chapters fonr t h g h six is being prepared for p u b
lication in the near future.
Chapter 2
Pulse propagation in birefringent , nonlinear media wit h deep gratings
2.1 Introduction
In recent years, much effort has been devoted to the study of one dimensional photonic
bandgap materi& in the presence of a Kerr nonlinearity [3] [7] [12] [U] [14] (151. 4 great deal
of the experimental work in this field has concentrateci on fiber Bragg gratings, which
typicdy have refiactive index variations on the order of IO-' [l6] [17] [18]. With such
s m d index variations, it is reasonable to apply the heuristic coupled mode equations, or
the appropriate noalinear Schrodiuger equation, to analyze experimental results [3] [16].
However, index changes as high as 0.04 have been reported in fibers [19], and experiments
employing etched semiconductor waveguides have been proposed [20]. These systems
have SUfEiciently lwge index contrasts so as to cast doubt on the validiw of the heuristic
coupled mode equations. In a recent paper [7], a coupled mode theory was dewloped
that accounts for strong gratings, in which the index contrast varies over a significant
&action of the average background index, with a Kerr nonlinearity.
In this chapter we extend the strong grating, nonlinear coupled mode forrnalism to
Figure 2.1: Schematic of the system under study. The index of refkaction is periodic
in the z direction. The system is W t e and homogeneous in the x and y directions.
Birefringence is includcd by defimg E, # Q, where E, and E, are the average indicies
of refraction seen by the x and y polarizations respectively. The theory in this papa
accounts for propagation in only one dimension, so in the context of this figure, light can
propagate only in the z direction.
include birefringence. The system under study, shown in Figure 2.1, consists of an infinite
medium with a periodic variation in its index of &action in the z direction. The medium
is assumed to be both infinite and homogeneous in the x and y directions; yet the theory
presented here is wdl-suited to the description of a system such as a fiber grating, where a
mode profile can be dehed t hat accounts for the transverse directions, or a Bragg stack,
in which the transverse dimensions of the stack are much larger than those of the light
pulse. The birefringence in the system is accounted for by defining a Merent index of
refiaction for each polarization, that is, & # %, where E, and are the average indicies
of refiaction seen by the x and y polarizations respectively. The theory accounts ody
for propagation in one dimension. That is, in Figure 2.1, the iight can only propagate in
the z direction.
Birefiingence has the &ixt of separating the photonic band gaps of the two polariza-
tions so that, in certain frequency ranges, light of one polarization can propagate hely
while the other is blocked. This has immediate deleterious consequemes for proposed d e
vice based on circular polarization, where the hearly polarized signais are mixed. The
robustness of nonlinear effects, such as soliton formation and propagation t hrough grat ing
structures, has yet to be studied in the presence of birefringence. Although optical fibers
are nominally isotropie, the process of writing a pating introduces a birefnngence on the
order of loh6 [21]. The dynamics here can be expected to be more complicated than in
a bare optical fiber. In a bare optical fiber the two polarizations have diaerent group
velocities, but can be considered to &r equal dispersion [22]; this is not generally valid
in the presence of a grating. In addition to fiber experiments, semiconductor waveguides
with a X ( 3 ) nonlinearity, which possess TE and TM modes with dXerent group velocities
and dispersions, and which can have large index contrasts, have been stuciied experimen-
t d y [ZO]. Although our formalism is strictly one-dimensional, it provides a quslitative
insight into the properties of such structures. We note, too, that experiments in the
literature, such as the allsptical AND gate demonstrateci by Taverner et al. (181, require
a coupled mode formalism for their convenient a.ndysis, as will other experiments aimed
at exploithg polarization and nonünearity.
Weak-grating coupled mode equations for pulses in a nonlinear, birefringent, peri-
odic medium (231 have previously been reportecl. This chapter indudes derivations for
three sets of equations: weak- and strong- grating coupled mode equations, and coupled
nonlinear Schrodinger equations in the presence of birehgence. We use Bloch theory
to characterize the linear, birefringent problem, and the method of multiple scales to
include the nonlinearity and f i t e pulse width. Both the birefringence and nonlinearity
are assumeci to be weak, in a sense to be made precise below.
The o u t h e of this chapter is as follows. In Section 2.2 we dis= the linear properties
of a one dimensional, birefrkgent, periodic medium. In Section 2.3 we introduce the
method of multiple SC&, which we then use in Section 2.4 to derive a set of coupled
nonlinear Schrodinger equation, and in Section 2.5 to derive a set of of coupled mode
equations. In Section 2.6 we discuss the connection between the nonlinear Schrijdinger
equations and the coupled mode equations, and their respective regions of validity.
2.2 Linear Equations and Basis Functions
We begin with the linear Maxwell equations in the presence of a dielectnc tensor that is
a function of only one Cartesian component, E = ~ ( z ) . We assume that the (x, y) coorai-
nates can be chosen such that for a.il z the tensor is diagonal, E = dzag (E,, ( z ) , e, (2)).
Neglecting magnetic effects by setting the permeabiliw p equal to that of kee space,
p = po, we can then define indices of rehction associated with polarization dong the x
and y axe, n: ( x ) = Eii (2) where €0 is the permittivity of free space and where, for
the remainder of the text, the index i runs over x and y. We seek fields E(r, t ) , H(r, t )
that depend only on the coordinate z. To proceed, we introduce local mode amplitudes
where no is a reference refractive index and Zo = JE^)'/^ is the impedance of kee
space. Using (2.1) in Maxwell's equations we can derive the differential equations that
the A* fields satisfy,
with the cohimn vectors
the matrix differentid operators
where c is the speed of light in vacuum, and the index matrices
The similarity between our equations (2.2) and those of de Sterke et al. [7] dows us
to proceed in a marner analogous to theirs, except for the complication of h a d g both
x and y polarized fields. The idea is to assume an harmonic t h e dependence e-"rlt for
the A, fields, and then formally solve for the z-dependence in terms of the eigenvectors,
?Y,,, of the matrix qlm.
2.2.1 Periodic Structures
To find the @p* the particular E(L) must be specified. Since we assume e(z) is neri-
odic with p&od d, E ( Z + d ) = ~ ( z ) , we c m me Bloch's theorem to constmct the qPi
[24]. To connect with other literature it is convenient to write the in terms of the
correspondhg solutions q&i ( z ) for the electric fieid itself which satisfy (71
where w~ is considered positive, and are of the form
where u,,,*(k; z + d) = h= (k; 2); that is, the %*(k; z) have the periodicity of the lattice.
Note that the index p has been replaced by a discrete band index m and a reduced
wave number k, (-rrld < k 5 rr ld) . If we seek &,(k; z ) that satisfy periodic boundary
conditions over a nomalkation Iength L, then k must be of the form 2 q l L where p is
an integer. We denote the associated eigenfrequencies w,(k). For each po- . . the
Bloch functions are orthogonal through the metnc n!(z),
where the normalization constant N = Lld has been chosen to facilitate passage to the
L -. w Illnit. In terms of the (k; z) we h d [7]
with ic $2; (k; L) = - z) F - 1 a4mi(k; 2)
2 wmi(k) (2.10)
Properties of the dispersion relation such as group velociw and group velocity dispersion
at a given rn, k point, for a given polarization, can be determined via the 'k p' expansion
(7). The use sf the iY&) is prefmed over the use of the mual cl,,$; z ) bmust! the
former leads to a much simpler k p expansion and a much simpler implementation of
a multiple scdes analysis. We here simply give the key results. The velociw matrix
element vpqci)(k), at wavenumber k associated with bands p and q and associated with
polarization i is defined as
The group velocity and group velocity dispersion are giwn by
and
We note that the sum in (2.13) goes over positive and negative kequencies [25].
CHAPTER 2.
2.3 Nonlinearity and Multiple Scales Analysis
Having characterized the linear problem in the presence of birefringence, we now turn
to the inclusion of nonlinearity. In the presence of a nonlinear polarization, PNL(r, t) =
5PFL(z, t) + ijPYL(z, t ) , the Maxwell equations become
where
To describe the nonlinear polarization we adopt a nondispersive Kerr model,
with 2 , j , k, 2 = x, y. It is clear that this form of the nonünear polarization will couple
the A, vectors. Of course, the Kerr model (2.16) is only a reasonable assumption if the
intensities involveci are not large enough that higher order susceptibilities need also be
included. We refer to this as the weakly nonlinear regime.
2.3.1 Multiple Scales Analysis
We would Like to use the nonlinear equation (2.14) to treat pulsg descnbed by enw-
Lope functions that are slowly varying in time and space relative to a carrier frequency
and lattice period respectivdy. One method of carehlly accounting for the effects of a
'weak nonlinearity' and 'slowly varying' pulses is the method of multiple scales [3][7].
Thiç method requires the introduction of several t h e and Wace scales via a smallness
parmeter 7 < 1. One c m then M t e a typical function as
f(r,t) = F ( z , ~ ) z , ~ ~ z , ...; t , q t ~ ~ t ...), (2.17)
where F is assumed to vary equaiiy significantly as each of its spatial arguments varies
over a given range 1, and each of its temporal arguments varies over a given period T .
The multiple scales of the problem are defined by
and
For our purposes, the characteristic length scaie, 1, is the lattice period; and the character-
istic time scale is T = 27r/wo, where wo is on the order of a typicd carrier fkequency. These
quantities represent the shortest length and fastest time s c a k in the prohlem. One can
see fkom (2.18) t hat the rp , t , account for field variations over successively longer lengt h
and t h e scales.
For iliustrative purposes, consider a triai solution of our nonlinear equations of the
fom:
where the p subscript indexes a large principal c~mponent with band index p, and wave
vector k; the c subscript indexes srnaller cornpanion components with band index c and
wave-vector k. If ffi is not varying over too short a distance and the nonlinear dects are
not too strong, in a sense to be made more precise below, then we wouid expect that even
the nonhear Maxwell equations could be a p p r h a t e l y satisfied by having f,(k; z, t)
acquire a time dependence that involves variations on the order of time scdes long com-
pared to l / w * ( k ) . Of course, small corrections must be expected to this description,
which we see below can be described by adding srnail amplitudes of other Bloch functions
%(k)
To implement this strategy, we write:
The quantity 'a' has been introduced to characterize a typical amplitude of the fields; it
is set such that the F$) (k) are dirnensionless and of order uni@.
To set up the equation (2.14) for a multiple scales analysis, we cast it in terms of
these newly d&ed variables h, f. This can be done quite generdy, without specdjmg
whether there is one or more principal component in &. We find
where
and
We would like to solve (2.23) in successive powers of q, so we must chasacterize
the nonlinearity in terms of t). To do so, we set a typical component X$L(z) equal to
xNL?(2) where ~ ( z ) is of order and dimensionless. Then the quantity xN,a2 c m
be considered to characterize the 'strength' of the nonlinearity. If the value xNLa2 is of
order q" wit h s = 1,2, . . . t hen the intensity index of the nonlinearity is s, and the leading
term in Bi will be of order qt Although the solution to (2.23) can, in general, be pushed
to higher powers of i), it is not reasonable to take the analysis past the intensity index.
This is because the nonhearity has itself been approximated; to inciude higher orders
of i ) we would have to include higher susceptibilities in (2.16).
2.4 One Principal Component; s=2: CNLSE
For a pulse that is not too short, with carrier wavevector away hom the centre or edges
of the band structure, we seek a description in terms of one principal component for
each polarization; a sufnciently long pulse, with a corresponduigly narrow kequency
content, cm be detuned at the band edge, or even slightly within the gap and stili be
reasonably describeci by one principal component 126). The birefiingence introduces a
wavenumber (k) mismatch between the two polarizations, each of which is carried at the
same fkequency (w) , as shown in (a) of Figure 2.2. We write our fields as
where w, (k;) = w, (b). We stress that although the ca.rrier frequency w, (k,) is the
same for both polarizations, the derivatives will, in general, be unequal.
Equations describing light in periodic, Kerr-nonlinear media are oken presented in
terms of the electric 6eld or a similar quantity [3]. We here opt to rewrite our A field
in terms of quantities directly comparable to power, because this is the most readily
Figure 2.2: Dispersion relations in the vicinity of the Bragg wave-vector. There are two
situations: a) A carrier fkquency w l , for which there is one principal component for each
polarization. The hrequency gives a different wavevector for each polarization, which
accounts for birefigence; b) A carrier hequency w l in the bandgap; here one wodd
use two principal components. In this case the pulse is carrieci by the average of the
Bragg frequencies shown, which accounts for the birefikgence. The mismatch between
the incident hequency and the Bragg fiequencies can be included in the slowly vasring
amplitudes. The quantities Qz,Qu and O,+, are detuning parameters used in section
VI. The quantiw n, is the grating strength parameter, defineci in section V.
accessible experimental quantity. Using the form of the Ai fields in the definition of the
Poynting vector,
S = E(z, t ) x H(z, t ) ,
we find, using the velocity matriv elements, and group velocity expressions given by
(2.11) and (2.12), that we can express the t h e and space average of the Poynting vector
to O($) as
This equation (2.28) suggests a field-dation
where RE is an effective cross-sectional area in the (x, y) plane associateci with the
problem. The X and Y fields are defined such that !XI' is the power in the x-polatized
field and 1 Y l2 is the power in the y-polarized field.
To deal with the nonlinearity, we assume here an intensity index s = 2, which means
t hat our nodhearity enters the equations at the sarne scale as the grating group velocity
dispersion. Under this assumption our eigenvalue equation (2.23) becomes
where the nonlineax term Bi enters at order q2. Note that to order q0 (2.30) is satis
fied because at that order one simply recovers the linear eigenvalue equation (2.2). To
cornpiete the analysis we collect terms first in q1 and then q2, which givg us two sets
of equations for each polarization. By combining these equations we can extract a set
of CNLSE in a manner analogous to that presented by de Sterke et d [71. We hd , to
order ql,
and similar for Y. Rom the q2 order equations we find
The quantity Bi is defined in (2.15), but we oniy need to wite the electric-field
contributions to Bi to order qo to keep (2.32) self-consistent; recall that the nonlinear
susceptibility is of order T ~ , so the last term in (2.32) will be of order qO. The form of the
nonlinear susceptibih@ has been chosen as that of an isotropie medium, but in principle
any X ( 3 ) tensor codd be used. We note. though. that the birehgence is considered
s m d because of a Limitation imposed by our method, discussed after equation (2.33).
Thus, since the effect of the nonlinearity itself is already considered small, the deviations
in X ( 3 ) due to lack of isotropy will typically be of the next lowest order in 7, and hence
can be ignored. The overlap integral in (2.32) is evaluated as
where we note that the quantity e 2 * ( ~ - 4 ) ~ has not been integrated because we assume
that ( I E p - kz) = qK, where K is of the order of the average wavenumber (kz + kJ 12.
In this case 22 (k, - kz) zo = 2iqKzo = 2iKzl. Since zl and zo are considered to be
independent variables the quantity e2'(4-4)' remains. The value of the coeflicients a
are give in Table 2.1.
Table 2.1 : Coefficients fof the €NESE.
The y values are determineci by interchanging x y
a& d4%&)u&(h; Z O ) ~ ~ Z ; Z O ) We now relate our scaled derivatives to full time and space derivatives. Assembling
(2.20) (2.31), (2.32), (2.33), and noting that the equations for the Y fields can be derïved
by interchanging x - y in the preceding, we obtain the following coupled nonlinear
Schr~dmger equat ions,
The quantity
A = 2(k, - kz)
characterizes the birefiingence in the system. The coefncients cr are so subscripted be-
cause a,, accounts for self phase modulation; a,, accounts for cross phase modulation:
and a, accounts for p.hase conjugation. We note that equatims similac to (234) have
been studied extensively in the Literature 191 [22] [ Z j [28] (291.
2.5 Two Principal Components; s=l: CME
We now turn to describing pulses whose carrier bequencies are in or very close to a
photonic bandgap, either at the band center or the band edge. In (b) of Fig. 2.2, we
show the case where the fkequency of the pulse is within the photonic bandgap. The
pulse c m , however, be detuned outside the bandgap and still be well describecl by the
theory presented here (see section 2.6). As discussed above, this situation often requires
the use of two principal components in the description of our fields. We set the reference
wavenumber to be the same for the x and y polarizations; the fiequency mismatch be-
tween the Bragg frequencies of the two polarizations accounts for the birefkingence. We
find that a derivation of the coupled mode equations oniy requires us to carry our results
through to order vl, so we simply write our A fields as
where ko is the wavevector at the band edge (assurned in Fig. 2.2) or band centre.
The quantities fui and fli modulate Bloch functions amciated with the upper and lower
band of the given polarization, 2 , respectively; both are principal cornponents in the sense
defined above. The carrier frq~~ency, rommon to both polarizatinns, J = (wo, f wo9) i 2 ,
is the average of the Bragg fkequencies of the two polarizations, w ~ = (w, + w l i ) / 2 , where
w, is the frequency associateci with the lowest point of the upper band, and mli is the
kequency associateci with the highest point of the lower band on the dispersion relation
(Fig. 2.2) .
By using our expression for A, (2.36) in the matrix equation (2.23) we find, to order
TI 7
.a fui 2- = a, f - 2 - + - y.:, . B;,~W.~O,
l r a21 q 0
where we have iised the defhitions val,, = wl,, - 9 where Q and au, are of order i ~ % ;
this is quivalent to assuming that the bandgap is smaU relative to the ca.rrier frequency,
ie ( - W U < 1 Note that we c m satisfy this condition and still have a strong
grating in the sense we have discusseci here. Since we are only carsring the calculation
to order 7, equations (2.37) become:
afur 2- af ii at - v*- - (wu* -
dz W) fur + ki(z , 4 f,, fw, fl1 fip) = O, (2.38)
where
and the function B ( t , t ; f,, fup , f', , fi,) represents the complicated overlap integrah in
(2.37).
We now introduce [30]
Rom the definîtion of the Poynting vector (2.27) and using Maxwell's equations we h d ,
to 0(71°)
This expression suggests a definition
where has been defined following (2.29), and where the exponential factor eh6t'4 has
been included in anticipation of the form of the final equations, with
This is quivalent to using the Bragg kquencies w u to c a a y the XI, Y* fields. Thge
new fieids are travelling waws ~orm&ed such that the quantities (1 X+ (* - IX- 12) and
(IY+ 1 * - 1 ~ - 1 *) represent the power in each polarization. Using the definitions (2.40) and
(2.42) in (2.38), and evduating the overlap integrah, we can write our fidl coupled mode
equations as
The appropriate equations for the Y* can be found by interchanging X - Y in (2.44)
and chagging 6 + -6. In these equations the value
accounts for the grating strength, and 6 accounts for the strength of the intrinsic bire-
fkingence.
The coupling coefficients {a, P , y) have a rather involved definition. We st art by
defining
The indicies p,p, r, s can take on the values l,u, that is, they index the upper and lower
bands. Notice that in the definitions of Fm, and y,, the values of the Bloch functions
in the integral alternate bekveen x and y. The coefficients of the X* coupled mode
equations (2.44) are shown in Table 2.2. The y values of the coefficients can be found by
switching x - y in (2.46), and in Table 2.2.
Table 2.2: z coefficients for the NCME. The y values are
Value 1 Weak ~ra&J
2.5.1 Weak Grating Limit ofthe NLCME
Many fiber gratings have small index contrasts, which dows u s to simplify the coupled
mode equations (2.44) by considering a weak grating of the form
nt ( r ) = 4 + 6% c o s ( 2 ~ z ) . (2.47)
where & is the background index, 6n is the index modulation with 6% « n,, and is
the wavenumber that d&es the band edge. In the presence of a weak grating, the Bloch
functions at the band edge can be evaluated, and normalized via (2.8),
If we use these forms for the Bloch functions and assume a d o m nonlinearity, then
many of the coefficients in the coupled mode equations (2.44) are identically zero. We
confirm, using (2. Il), that in this limit the quantity v, is simply equal to the group
velocity in the absence of the grating, v* = c/&. With this in mind we rewrite (2.44)
as
with
The grating coefficient is
the Y* equations can be found by switching x * y and 6 -. -6 in (2.49) and
from which we note that Pz = f19 and 7, = y,.
a very weak birefnngence, where îï, E %, the coefficients in (2.50) are in the
ratio {a : p : +y} = (3 : 2 : l}. In the stationary k t these equations agree with those
given by Samir et al. [31].
2.6 Connecting the CNLSE and the NLCME
In the previous sections we derived two types of equations: a set of coupled nonlinear
Schrddinger equations, typicdy valid outside the bandgap, and a set of nonlinear coupled
mode equations, typicdy valid within or near the bandgap. As we will see in this section,
the coupIed mode equations make very definite predictions about the dispersion relation
and the Bloch functions of the periodic system. When these predict ed Bloch h c t ions and
dispersion relation deviate from the true values of the system, then the approximations
that have been used to derive the coupled mode equations have broken dom; this allows
us to determine the b i t s of validity of the equations. On the other hand, the nonlinear
SchMdinger equation relies on the local properties of the dispersion relation, so if the
nonlineaxit-y is sufficiently small it should always be valid as long as one is sufficiently
far away from a bandgap, or other portion of the dispersion relation with significant
higher-order curvature. If the kequency content of a pulse is very narrow. then higher
order dispersion will have Little &kt, so the Schrodinger equation should be valid at the
band edge and even slightly inside the bandgap.
A hirther point to be discussed is how the solutions to the nonlinear Schrodinger
equation relate to those of the coupled mode equations. Understanding this dows us
to identa the range where either approach could be used, an important goal because
dthough the coupled mode equations are ead,, solveable via niimpsical techniques, they
are diffcult to solve analyticdy. As mentioned, there is a great deal of work in the
literature on equations similar to our CNLSE [9] [22] [27/[28] [29], so if we understand how
solutions of the CNLSE are related to solutions of the NCME, then the CNLSE literature
becornes available to aid in the investigation of birefringence phenornena near the gap.
Specincally we want to know how to relate the two CNLSE fields, X and Y, to the four
NCME fields, X* and YI; and we want to get a sense of how close to the gap we must
be before the CNLSE cease to effectively describe the problern. Our method is to &art
with the weak grating nonlinex coupled mode equations and perform a M h e r multiple
scales analysis to derive the noniinear Schr~dinger equations. The use of the weak grating
equations simplifies the mathematics, and does not significantiy dec t the final results,
for reasons discussed bdow. The rnethod involveci foUows closely the analysis of de Sterke
and Sipe [3], except that in the present case the nonlinear te- are much more involved,
so we only sketch the results.
with which the h e m portion of the coupled mode equations can be written as
where we have used the Pauli spin matrices
and the unit mstrix ao. We seek solutions of (2.53) of the form Fi = te-'(%* '-Qiz) , where
the wavevector detuning is Qi = L - ko If the fidl frequency w, > w&, then we c d the
detuning parameter Q+ and otherwise we c d it Q-, with & = w, - w ~ . The R+ are
associated with the upper and lower band via the dispersion relation
which follows from substituting the Fi in (2.53)- Rom the dispersion r-12 the SQUP
velocity, and group velocity dispersion are
where pi (Q) = %RL(Q)/c is the ratio of the group velocity at a given wave-vector for a
point in the upper band, relative to the group velocity in the absence of the grating. The
eigenvectors have the form
fi(+' (Q
where the f,!') (Q j are associated with the 4+ respectiveiy.
Fkom these eigenvectors one can extract the Bloch functions of the periodic structure,
in the coupled mode equations E t ,
where the factor 1 / J i has been hchtded for propa n m h t i o n via (2.8). The
hinction multiplying eikZ can be identifiecl a s ~ ( * ) ~ ( k ; 2).
If we include the nonlineari~, then we can write the coupled mode equations as
L J
where Ni is the nonlinear term that follows immediately from (2.49). For simplicity we
concentrate on detuning into the upper band, &+(Qi) We represent our field vector Fi
as being mostiy in the upper band, but with a smail component in the lower band. We
&art by writing the field vectors as
where we have introduced the multiple scales variables h, t, as in (2.18). The upper-band
component aj dominates the expansion of Fi, and hence plays the role of a principal corn-
ponent; the bi terms are companion components. The numerical value of the detunings,
n+,(Qi) and Qi, WU be different for each polarization, but in each case we are detuning
to the sarne frequency w , as shown in (a) of Fig. 2.2. The normalization factor 116
has been introduced so that the envelope functions, ai, will be directly related to power.
Since the nonlineariw involves only cubic-type terms or higher, we can write:
Nz = r13Nz3 + ... (2.61)
To e d u a t e NZB rire combine (2.52), (ZS?), (Z.60), koxn w k c h it is apparent that to
Iowest order in 7)
where A is the bire£ringence parameter quoted earlier (235). Note that to order q1 the
forward and badrward going fields, XI, are associateci with the multiple scales envelope
fmction a,. This means that, were we to use the strong grating equations, the form
of wodd be the same, but the values of the coefEcients would change. However,
since the values of the weak grating Bloch functions are known, it is straightforward to
compare the nonhem Schrodinger equation derived fiom the weak grating NCME, to
the weak grating CNLSE.
Using this nonlinear operator in equation (6.17) contained in the article by de Sterke
and Sipe [3] dows us to write down the CNLSE:
where ai+, and O:+ are the group velocity and group velocity dispersion at the given
detuning (2.56), and the nonlinear coefEcients are:
The coefncients (2.65) lead to the concept of an effitzue nonl2nearity because their values
are dependent on Q, the detuning fiom the Bragg wavevector.
To connect (2.64) to the CNLSE giwn by (2.34), we recd that both the (X*, Y*)
fields used by the NCME, and the (X, Y) fields used by the CNLSE are normaüzed such
CHAPTER 2. 35
that their squared moduli represMt p0werOwer If we wish ta connect the cMSE and CME
fields we require that
and sirnilady for Y. We have used (2.62) for a,. Hence, our fields X, Y and a,, a, are
equivalent. Using the Bloch functions (2.58) we c m show that the coefncients given
above (2.65) agree with those in table 2.1.
2.7 Numerical Simulations
The simulations are intended to illustrate two points: Grst, we demonstrate the validity
of the CNLSE approximation with respect to the NCME approximation, as discussed
in Section 2.4; second, we investigate the e&ct of energy exchange between the two
poiarizations, which may be of importance in the development of new devices.
For the sarnple calculations, we used parameters of a typical optical fiber, given in
Table 2.3. Table 2.3: Parax.net ers used in numerical simulations
1 Index of Refraction (R,) 1 1.50 1 1 Index hlodulation ( 6 4 1 1.67 x 10-~ 1
1 Bragg Wavelength (nm) 1 1052.00 1
Bireikingence (% - &)
Nonlinem Index (n2; W / d )
2.7.1 Comparing the CNLSE and CME
2 x IO+
2.3 x 10-l6
To compare the CNLSE and the CME equations, we consider a pulse propagating through
a grating with parameters given in Table 2.3, using each set of equations. We solve the
CNLSE by a split-step Fourier technique: At each time step the hear portion of the
equations are solved in the Fourier domain, while the nonlinear portions are solved using
a 4t h order Runge-Kutta integration scherne [9]; we solve the equations (2.64) in a fiame
travelling with the average velocity of the two pulses. The CME are solved using a
collocation algorit hm [32].
Table 2.4: Cornparison of veiocities between
the NCME and CNLSE algorithms
A
where it has been assumeci that E, < g, so that (uch - uqr) > O. IR terms of these we
1.50 / 0.740
define the kequency detuning
NCME, x poln
0.740
where w is the carrier fiequency of the pulse.
CNLSE x poln
l To define a frrquency control parameter, we fi& define a total bandgap width
We start with simulations using the values of A given in Table 2.3. The initial intensiw
was 1.10GW/cm2 in each polarization, the initial pulse was a Gaussian with a full-width?
half-maximum ( F m ) of 200ps, and was chosen such that the initial frequency content
to the pulse did not extend appreciably into the gap. Table 2.4 compares the wlocities
observeci by the NCiyIE and the CNLSE for the x polazization after 3000ps of simulation
time, from which it can be seen that both algorithms predict the same vdocity even very
Figure 2.3: Cornparison of the CME and NLSE y polarbation pulse pro& after 2000ps
of simulation time for (a) A = 1.20 and (b) A = 1.10. It is evident in (b) that the two
algorithms are giving different resdts.
close to the band gap. Figure 2.3 compares the pulse shapes of the x polarization after
2000ps of simulation time for A = 1.20 and A = 1.10. It can be seen that although the
two dgonthms agree very closely for A = 1.20, at A = 1.10 the differences are more
maxked.
There are two reasons why the derivation in Section 2.6 wodd fail: firçt, the NCME
includes a l l orders of dispersion, while the CNLSE includes only 2d order dispersion;
second, the CNLSE derivation assumg that there is but slight build up of reflected
waves which, as one mars the bandgap, is decreasingly accurate. To quantify the effect
of the first objection, we calculate the quantity
in terms of which the expressions for 2d and rd order dispersion lengths, assumhg a
Gaussian pulse, [51 are
where TFWHM is the pulse width; if LD3 x LM, then 3rd order dispersion dects become
important. We thus have a criterion on TFwHM that
5 TFWHM >> -, (2.69)
Ri+
for 3rd order &&s to be ignored [33].
To quanti& the second limitation we note [5] that a Gaussian pulse with a given
TFWHM has a frequency width
Thus, for a given carrier fkequency, w, the fkequency spectrum of the pulse wili extend
into the band gap if
However, as the pulse frequency n e m the gap, it will, of course, scperience higher order
dispersion as well as, eventually, reflection, so that this criterion is not completely distinct
from the one presented in the preceding paragraph.
We present simulations to underscore the first objection. We use a gating with the
physical parameters in table 2.3, and a pulse with initial intensity 1 .50GW/cm2 and d e
tuning A = 2.00. We concentrate on a single polarization, since birehgence is incidental
to the higher order dispersion. Using the criterion (2.69) we fuid t hat TmHM > > 1 2 . 5 ~ ~ .
In Fig. 2.4 we plot the sirnulateci pulse profile after 5ûûps of simulation time using both
the NmIE and CNLSE for a TFWHM of 5ûps and 10ps. It can be seen that the lOps
pulse experiences a great deal of higher order dispersion. We note that only a small
amount of reflected mves build up in this simulation, so that the second objection is
irrelemt. We note, too, that since we have not attempted to sirnulate a soliton, the
Figure 2.4: Pulse profiles after 500ps simulation t h e for initial pulse widths of (a) 50ps
and (b) 10ps. It is clear that the lOps pulse is experiencing a great deal of higher order
dispersion under the CME.
self-phase modulation will tend to increase the fiequency spectrum of the puise, so that
eventually the results of the two integrations must diverge.
2.8 Conclusion
We have discussed the propagation of light through a strong grating structure in the
presence of birefrmgence and a Kerr nonlineariw. The &ect of the birehngence is to
separate the photonic band gaps associated with the two polarizations. Far fiom the
photonic band gaps, and even near the gaps if the pulses are not too short, the elec-
tromagnetic field can be w d describeci by two coupled nodinear Schrodinger equations,
one associated with each polarization. Here the situation is somewhat sirnilm to prop
agation in a 1D structure without a grating, with the dispersion due to the underlying
material medium. But the grating structure is richer in two respects. First, the two
polarization modes can have both different group velocities and different group wlocity
dispersions, whereas in iinifnrm Il3 structues difkences in the group uelocity dirper-
sions can typicdy be neglected. Second, the effective nonlinearity is a function of the
carrier frequency of the pulse, since it depends on how the appropriate Bloch function
samples the distribution of nonlineariw in the underlying medium.
At carrier frequencies close to the gap or within the gap, the electrornagnetic &Id is
desaibed by two sets of coupled mode equations. These two equations are the analog
of the familiar coupled mode equations in the absence of birefnngence, with one pair
of equations for each polarization. For a range of parameters either set of equations
can be useci, and we identifid the conditions required for this and confirmeci them with
numerical examples. Fkom the generd form of the equations we derive, it is clear that
whole new regimes of nonlinear phenornena can appear when biremgence exists in ID
photonic band gap structures, includuig d-optical switching geometries that have no
analog in isotropie structures. T'us the derivation of the sets of equations we presented
here is of interest not only in its own ri&, but as a starting point for addressing what
to date is the largely unexploreci territory of birehgent, nonlinear, photonic band gap
structures.
Chapter 3
Polarizat ion effect s in birefiingent , periodic, nonlinear media
Introduction
It is weil known that light propagating in an isotropic fiber grating, with frequency con-
tent slightly outside the photonic bandgap, can be described by a nonlinear SchrWger
equation (NLSE) [3j [7] [14] [16][17]. It is also known that the process of growing a UV-
induced fiber grating introduces a weak birefiingence into the nominally isotropic bare
fiber [21j. This birefringence leads to a separation of the photonic bandgaps of the two
orthogonal polaiizations. Propagation of light whose fiequency content is outside both
the photonic bandgaps is well described by a set of coupled NLSEs [23] (343, as was shown
in the previous chapter.
The coupled NLSEk r e l m t to a birefringent grating are similar in form to those used
for bare fibers [35], so rna.ny of the observations in that literature shodd be observable
in gratings as well. However, there are two major digerences between gratings and
bare fibers: h t , the grating dispersion is orders of magnitude higher than in a bare
Bber, so that, for a given pulse width, soliton formation intensities are much higher,
and interaction lengths are much shork~; seeond, the eoupled NLSE parameters, snch as
group velocity and phase velociw are fkequency dependent. Much of the literature on the
coupled NLSE has concentrated on energy exchange between the orthogonal polarizations
[9][28]. This energy exchange, if properly phase matched, can lead to a polarization
instability, whereby intense light polarized near the unstable axis (the axis with the lower
index of refiaction) shifts its energy to the stable axis (with higher index of refiaction)
[36][37]. In a ment paper it was demonstratecl that, unlike in a bare fiber, the threshold
intensity, I P I , for th& instability is a strong funetion of frequency detuning in a fiber
grating [38]. The t hreshold intensity, Igol, for isotropie soliton formation, which occurs
in bkfringent fibers if the light is exactly confineci to one of the principal axes of the
fiber, is also a function of fiquency detuniag [3]. In this chapter we investigate these
two phenornena and show that a grating dows us access to three distinct regimes: at
high detttnings we find IaOl c I p I ; at low detunings we h d Igd > IpI ; and for a middle
region we find I.& -- Ipl
Although the coupled NISEs provide an excellent heuristic guide to nonlinear phe-
nomena in fiber grating systems, in their usud form they are unable to describe accu-
rately the physical gratings used in experiments, because they account neither for the
finite grating length, nor for the apodization profile used to rninimize oscillations in the
linear transmission spectrum of the grating. Furthemore, they must be extendeci [39] if
they are to account for any slow spatial variation in the background index of refraction
of the grating. Therefore, a set of nonlinea,r coupled mode equations (CME) were used
to simulate light propagation [34]. For an infinite, unapodized grating with no slowly
varying spatial variation in the background index of refraction, it is known that the s e
lut ions to t hese nonhear coupled mode equations are àirectly related to the solutions
of the coupled NLSEs [34]. In the fkequency regimes of intergt here, there is a sort of
hierarchy, because the coupled NLSEk can be extracteci kom the nonlinear CME [34].
Thus, although the nonlinear CME provide the better description for pulse propagation,
a quantitative and qualitative understanduig of the iinderlying physics can he d e t d
using the coupleà NLSEs.
When Iad < IpI we can excite pulses that are soiiton-like, and hence maintain a
roughly constant amplitude profile without any nonlineu energy exchange. In this regime
we demonstrate numencaliy that we can use an a p p r h a t e theory (401 to predict the
nonlinear evolution of the Stokes parameters of the pulse. As we detune closer to the
photonic bandgap this approxhate theory breaks down due to reflect ion, more dispersive
&ts, and more nonlinear pulse shaping and energy exchange.
At very high intensities, for pulses with hquency content near but not inside the
photonic band gap, it may be possible to observe the vector solitary wave descnbed by
Akhmediev et aL (401. We present simulations that demonstrate that such a solitary
wave should exist in a fiber grating; its observation, though, would require an oblique
experimental procedure (411, for which the grating used in the experiments reported in
thk &aptw W tuo short. Nevwtheiess, the form of the vector solitary wave suggests that
the phenornenon of polarization instability, in addition to being hequency dependent,
is also strongly dependent on the initial phase lag between the field on the stable and
unstable axes. Eqeriments to v w this dependence on phase lag were performed by
Dr. Richart Slusher of Lucent Technologies, from which it was shown that with a 90'
degree phase lag between the components polarization instability is almost completely
suppressed.
This chapter is divided into six sections. In Section 3.2 a mode1 of a grating which ex-
tends infinitely in space is presented. Rom this model a dispersion relation is determineci,
from which the pulse propagation parameters necgsaxy to write down a set of coupled
NLSEs can be extracted. In Section 3.3 a more complete model of the physical grating
used in the experiment is presented, as are the nonlinear coupled mode equations that we
use to simulate light propagation. In Section 3.4 the approJamate nonlinear evolution of
the Stokes parameters is numerically simulated. In Section 3.5 sorne experimental r d t s
are presented, including the dependence of polarization instabiüty upon input phase 1%.
3.2 Theory for an -te grating
Here we present a mode1 of a birefkhgent grating £rom which we can derive an analytic
expression for the dispersion relation of the grating. Rom the dispersion relation we
extract pulse propagation parameters such as phase velocity mismatch, group velocity
and group velocity dispersion. We then use the pulse propagation parameters to e t e
d o m a set of coupled nonlinear Schrodinger equations that describe iight propagating in
the birefringent grating in the presence of a Kerr nonlinearitsf. Rom the coupled NLSEs
we can determine the threshold intensities required For soliton formation, and for the
onset of polarizaticn instabilky.
3.2.1 Modelling the Grating
We mode1 the index of refraction of a birefringent grating as
where i represents the x and y polarizations, & is the background index of the gating,
6% is the index contrast, and ko is the Bragg wavenumber. It is assumeci that 6ni « K. Both the background indices and the index constrasts are unequal: # q; 6% # dn,
[42]. In terms of these quantities the grating will have a strength = ko6q / (%) , and
a Bragg hequency wa = ck@+ We dehe the birefnngence (no) and gating strength
mismatch (bnb) - -
nb = % - r a z , anb =b%-anz; (3.2)
as well as the ratio, v , of the birefringence to the s grating strength, and the ratio, iM,
of the grating strength mismatch to the birefkingence,
" JV wi = fi k ni + (kt - 'O) , n,
where the k refer to detunings above and below the Bragg hequency. It is evident that
only fkequencies for which Iwi - w&( 2 CK,/% lie on the dispersion relation for the ith
polarization. If this condition is not met for a given polarization, then the kequency is
said to lie in the photonic bandgap of the grating; the width of the photonic band gap is
2 ~ ~ 4 % . We can invert (3.4) to tind the value of at a given
Using (3.4), the group velocity and group velocity dispersion
given by
fr€!qtleIlcy, W i ,
(3.5)
of the grating system are
In the simulations and experiments it is assumecl that a pulse is injected into the system
with a carrier fiequency common to both polarizations, J = w, = w,. Because the
system is birefiingent, this common carrier frequency will correspond to two dXerent
wave numbers, xz and x9, which can be found using (3.5). For the s m d birehgences
that we are considering, it is reasonable to assume that the group velocity and group
velocity dispersion of the two polarizations at t i j are roughiy equal:
The mismatch in phase accumulation, &, -k,, which is due to the linear birefhngence,
is non-negligible. We d&e a frequency detuning parameter,
where, for definiteness, are have measured the detiininp h m the x-pobnzation Bragg
kequency, and scaled to the half-width of the x-polarization photonic band gap. In terms
of this detuning parameter we can define an efféctive birefringence [42],
where the plus (+) sign refm to detiinings above the Bragg frequency, and the minus
(-) sign to those below the Bragg frequency.
3.2.2 Coupled Nonlinear Schrodinger Equations
A slowly varying pulse with hequmcy content sufnciently far fiom the photonic bandgaps
in a birefringent , Kerr nonlinear medium will satsfy a set of coupled nonlinear Schrodinger
equat ions (341,
where the fields a,, a,, are slowly-va,rying envelope functions, carried at the common fre
quency, g, but at unequal wavenumbers, k, # k,. The envelope functions are normalized
such that ~CQ l2 gives the intensity in the field. The quantity 17 is related to the phase ve-
locity mismatch, 7) = 2 (kg - &) = 2 n r f ~ / c . The quantities Z' and G" are definecl in
(3.7). In the weak grating limit (hi &) we assume here, the nonlinear coefficient a,,
is given by
where n2 is the nonlineax index of rehaction in the absence of a grating, and where f
is the nomalized detuning parameter defined in Equation (3.8). The value of a,, and
cr, can be determined by the ratio {a, : a,, : %} = (3 : 2 : 1). We also define an
effective nonlinear index of rehaction
rii the absence O£ equaths (%IO) d k b e pulses propagating wigh a
group velocity and group velocity dispersion defineci by (3.7). If we include nonlinear-
ity then three effects emerge: a,, governs self phase modulation; a,, governs cross
phase modulation; and a, governs phase conjugation, or energy exchange. The energy
exchange term is not phase matched, so if the efktive birefkgence is high, then we
would expect its effect to be small. If we ignore a,, we can define an eflective index of
refiaction for each polarization,
from which we can define a nonlinear birefringence,
When we include crk the concept of a noalinear birefruigence will still be valid, but the
dpamics of the pulse will be complicated, so it is best used as a heuristic guide.
In Figure 3.1 we plot the quantities @/q, (solid line), 3 1 (~/n,) (dash) and n;lf /n2
(linecircles) as a function of detuning, f, for M = 0.65 and v = 0.0375. The effective
birehgence is rougidy constant, due to the effect of the grating strength mismatch [42].
The group velocity is reduced as a consequence of the multiple rdections within the grat-
ing [3]. The nonlinearity is enhanced as a consequence both of the multiple refiections,
and the mamer in which the Bloch functions of the grating interact with the underlying
nonlinearitly (431.
The effective birekingence and enhanced nonheariîy help to determine two pulse
propagation efFects: isotropic soliton formation threshold and polarization instability. For
a given W-width, half-maximum (FWHM) pulse width, TFwHnr, an isotropic soliton,
Figure 3.1: Enective birefringence (solid h e ) , group velociS (dash-dot) and nonlinear
index of refraction (linecircles) in the fiber grating as a hindion of kequency detuning.
Near the edge of the photonic band gap (f = 1) the group velocity twds towards zero, and
the effective n q - W. For the parameters we have chosen, the effective birefkingence
rernains roughly constant .
polarized completely dong the x axis, d l form with a peak intensity [5]
The CW threshold for polarization instability is [5] defineci as the minimum pulse intensity
for which the nonlinear birehgence cancels the effective birehingence,
At this intensity, the phase term e*qz can be cancelled by self- and crossphase modu-
lation, so that energy exchange wiU build up. For dennitenes we set nb > O, so that to
cancei e*qz we need nNL (f ) = -niff (f ) which, giwn (3.14), can only occur if 1% 1 > 1% 1. Thus, if more energy lies on the x âxis (for nb > O), then energy exchange can occur,
while if more energy lies on the y axis, energy exchange will be suppressed. For this
reason we label the x axis unstable snd the y axis stable. We stress, though, that the
value IpI is baseci on a CW d y s i s , and that it refers to wmplete phase matchmg of
Figure 3.2: Intensity threshold for polarization instability ( IpI - line with circles) and
soliton formation (Isd - solid line) using material parameters given in the text. Polariza-
tion instsbiliv is a balamhg of birefnngence and nonlineari~, and since the nodineariQ
increases more steeply at low detuainp than does the birefiingence, the threshold for the
instability goes up. By contrast, a soliton formation is a balancing of nonlinearity and
group velocity disersion. .4lthough the nonlinearity incre-, the dispersion increass
at a lmger rate, so the threshold for soliton formation goes up at low detuning.
the energy exchange process. For intensities below IpI we would expect some energy
In Figure 3.2 we plot Iad (solid line) and IpI ( h e with circles) for M = 0.38, u = 0.07,
and TWHM = 50p, with ii. = 1.46, nb = 3.6 x 10-~, 6n. = 1.2 x 10-~. It is evident
that the birehgent grating gives us access to three distinct propagation regimes. For
large kequency detunings, we h d Isd < hl, so that we can excite pulses that roughly
retain th& shape upon propagation, but exchange no energy between polarizations; for
fi:equencies close to the gap Isd B IN; while for keqyencies between these extremes we
fkd Isd = I p I . Because the formula for Isd is a function of pulsewidth we could, for
larger pulsewidths, suppress the value of Isd so that Iad < IpI for all detunings where
we would expect (3.10) to hold; or, for shorter pulse widths, we could raise Isd > IPI at
3.3 Physical Grating
In this section we give a more adequate model for the experirnental grating. Since the
model accounts for the apodization of the grating, the concept of a dispersion relation
is no longer va.iid. Furthexmore, the coupled NLSEs will no longer provide an accurate
picture of light propagating in the grating. Instead we use a set of nonlinear coupled
mode equations to d d b e light propagation, but continue to use the coupled NLSEs as
a heuristic guide to nonlinear phenornena.
We model a physicai grating as
where we have allowed both the grating index contrast and the background index to
assume a siowly Msing z dependence. The z dependence accounts for three effects.
First, any experimental gating is of f i t e length, so we define both 6Ni (z) = O and
672, (2) = O for z < O and r > L, where the experimental grating extends only fkom O to
L. Second, it is cornmon to apodize the index contrast of the grating with a smoothly
varying index contrast profile 672 ( z ) , to eliminate sharp oscillations in the transmission
spectnun. Third, it is &O weiI known that the growth of an apodized UV fiber grating
creates both the desireci index contrast, hi (z) , and an overall rise in the background
index of refkaction, bNj (2). This rise in the background index leads to Fabry-Perot effkcts
in the transmission spectnim of the grating. We attempt to correct the background index,
so 6Ni (2) should be zero, but the correction technique is not perfect, and there is always
some z dependent background. The speciiîc model used for 6Ni (2) is given elsewhere
[4a. As a modei for 6% (z), we assume
where L is the length of the grating. This model assumes that the edges of the grating are
Gaussian-apodized with a half-width, half-maximum value zh,, while the centre portion -
bas a constant value of 6~ (2) = 6ni; since most of the grating has a constant index
contrast, the M and Y parameters can be used as defined above to characterize the
grating. In the simulations we use 6 = 2.4 x 10-~, = 1.46, nb = 4.2 x 10-~,
L = 7.7m and z h = 0.25.
As mentioned, we cannot use the coupled NLSEs (3.10) to give quantitative results for
the grating used in the experiments. Instead we rely on a set of nonlinear coupled mode
equations, which more easily amount for the apodization profile of the grating [34][44]:
where
and where J is the average of the two Bragg kequencies Z = (wo. + w ~ ) 12. The a p
propriate equations for the y polarization can be determined fÎom (3.19) by switching
x - y and 6 -+ -6. Here the are dowly varying endope functions modulating
forward and backward travelling waves, and are nomalized such that [AJ2 represents
the intensity in the field. The envelope functions are carried at their Bragg fiequency
( w ~ ) and at wavenumber ko.
3.4 Approximat e Solut ion for Polarizat ion Evolut ion
In a recent paper [40], Akhmediev et ai. presented a simple theory to describe nonlin-
ear pulse propagation governeci by the coupled NLSEs. In this section we summarize
th& theory, and demonstrate that for the correct parameters it provides an excellent de-
scription for pulse propagation through a grating. We 6rst define the nomalized Stokes
parameters
where So (z, t) = luJ2 + 1912. The quantity A$ is the phase lag between the two
polarizations, which we define by writing a= (z, t ) = laz (2, t ) ( e**=(2tt) and g (2, t ) =
1% (2, t)l e i 4 ~ ( z ~ t ) , so that
We now assume that the temporal profile of the two orthogonally polarized pulses is
cornmon and remains constant with propagation, so that
where h ( t ) is the dimensionles temporal profile. Using (3.23) the normalized Stokes
parameters become
where So = lx12 + IY l 2 h the absence of material absorption or other loss mechanisms,
the Stokes parameters can be cast in wctor notation with
and with So constant. Using the definitions (3.23) and (3.24) in the nonlinear Schrihhger
equation (3. IO), we find the vector S satisifes the differential quation
where ê1 and ê3 Xe unit vectors pointing in the SI and S3 directions respectively, and
where 0, half the phase-velocity mismatch, and g, the nonlinear parameter, are defined
1 / h4 ( t ) dt P = v / ~ , g = ~ a ~ l h 2 ( t ) & .
If we consîder puises that are detuned such that the group velociw dispersion is not
too high, and use a sufticiently l m intensity that the nonlinear reshaping of the pulse
is not too large, then the approximate theory just presented should be valid. In the
simulations we numerically integrate the nonlinear coupled mode equations (3.19), using
8Ops puises with f = 3.5 and Msying values of So (total intensity) . The control d a b l e
is Si, which we vary between f 1. The value of the Stokes parameters will Vary acrass the
pulse due to dispersive effects, so our quoted values of S2 and S3 are averages weighted
by intensity across the pulse. In Figure 3.3 we plot the output value of S2 as a function of
input SI for both the approxirnate theory and the numericd simulations, while in Figure
3.4 we plot the output value of S3 as a function of input Si. In both figures the qualitative
agreement is excellent. The exact d u e s for the theory and simulation disagree mostly
because in the integration of (3.26) it was sssumed that the grating M d its peak value - bn, throughout its eatire length, while in the simulation we used the apodized value
6% ( z ) (3.18).
Figure 3.3: Value of output S2 as a function of input Si for the approximate theory
(lower graph) and the numerical simulations (upper graph). We plot the values for iinear
intensities (square - line) ,2G W / n 2 (c ide - h e ) ,4G W / m 2 (cross - line) and 6G W/cm
(down triangle - iine). The simulations, which use the coupled mode equations and an
apodized grating, agree very well with the approximate theory based on the coupled
XLSES.
Figure 3.4: Value of output S3 as a function of input Si for the approximate theory
(lower graph) and the numerical simulations (upper graph). We plot the values for linear
intensit ies (square - Iine) ,2G W / m 2 (circle - line) ,4G W / m 2 (cross - line) and 6G W / m 2
(down triangle - Iine).
As we detune closer to the photonic band gap, the agreement between the apprmrimate
theory and the simulations worsen (for 80ps pulses the threshold is f < 2.5). This occurs
because the group velocity dispersion inmeases, which causes the pulse to broaden and
which makes the approximation of a constant temporal profile invalid. Authermore, at
lower values of f , parts of the incident pulse are rdected, which is not describeci by the
approximats t heory.
3.5 Experirnent al Data
The experimental arrangement. is shown in Figure 3.5. The pulsed light source is a Q
switched mode-locked YLF laser that produces 80ps wide pulses with center wavelengths
near 105311.. wavelengths. A single pulse from each Q-switched, mode-locked pulse train
is transmitted to the fiber using an electreoptic gate. The resulting 500 Hz pulse repeti-
tion rate at the grating is siow enough to eiiminate heating and to reduce the probabîiity
of damage to the input face of the fiber. The incident pulse is slightly chirped due to
n o n l i n e e in the laser gain crystal; this chirp is s m a l l compared to the nonlinear phase
shûts important for the phenornena studied in these experiments. The beam is coilimateci
to a diameter that matches the input microscope objective, and is attenuated using a
rotating neutral density filter. A . energy meter measures the pulse energies before the
input objective. Light exiting the fiber is collectecl by a microscope objective and focused
onto a fiber core at a distance of 2 m. This arrangement ensures that only light ftom
the fiber core is meastueci. Light from the collecting fiber is detected using a diode and
a sampling scope with a combined time response of 20 ps, sdiicient time resolution for
the experiments in this study.
The input polarization state of the optical pulse is set by the combined orientations
of a half-waw (HWPl) and quarter-wave (QWP1) plate immediately before the input
focusing lem as shown in Figure 3.5. The orientation of the polarization ellipse with
Figure 3.5: A schematic diagram of the experiniental apparatus used to measure the non-
iinear propagation of polarized light pulses in birefringent fiber gratings. A Q-switched,
mode-locked laser produces 80 ps pulses. An electreoptic puise picker (PP) transmits
one pulse fiom each Q-switched pulse train to the grating (G) through an attenuator
jATTj at a rate of 500 Hz. A combined setting of the half-wave plate (HWPI) and
quarter-wave plate (Q WP 1) determine the input orientation of the polarization st ate rel-
ative to the principal axes of the fiber grating as welI as the phase retardation between
the two field component along the principal axes. The collimateci light beam is focused
onto the fiber core ushg a microscope object lem (L). A similar objective recollimates
the beam d e r transmission through the grating and focuses it on a detector at a distance
of 2 meters. The Stokes components of the output field are selected for measurement by
the appropriate combination of setting for both the quarter-wave (QWPZ) and half-wave
plate (HWP2). The principal axes are shown as the stable and unstable axes. The output
polarization state to be measured is rotated to the vertical direction using (HWP2) so
that the light experiences a constant trammision coeflicient as it propagates through a
series of mirrors and lenses to the fast pho tdode detector and sampling scope.
Laser - PP t
ATT
-
respect to the pciacipal axes cd the gtating and the phase lag between 6he fa& and slow
axis polazïzation components are controlled in the following marner. If the polarization
of the light incident on the waveplates is along the fast principal axis of the grating, the
angles for the axes of QWP1, C, and Ml, Q, required to generate a polarization state
with a stabkto-unstable axis power ratio, r, and a phase lag between the polarization
components along the unstable and stable axes of A&n are found to be
The estimated experimentd error in setting At$ using these formulae is f 5O.
To measure the Stokes parameters (3.21) we use the output half-wave plate (HWP2)
and quarter-wave plate (QWP2) to rotate the polarization components required to mea-
sure the six intensiw components, I(0,O) dong the unstable axis, I(90,O) along the stable
&,I(45,0) along the axis at 45 degrees with respect to the principal axes, I(135,O) along
the axis at 135 degrees with respect to the principal axes, I(45, r / 2 ) for nght circularly
polarized light , and I(135,ir/2) for leR circularly polarized light , onto the transmission
axiç of a fixeci linear polarizer (LP2). This mangement maintains the polarization of the
output beam constant through the remainder of the collection opt ics, including polar-
ization dependent mirrors. The £kt d u e in the measured intensity parentheses is the
angle with respect to the principal axes and the second value is the phase retardation
between the unstable and stable components. The normalized Stokes wctor components
are calculateci fiom the set of six measured intensities, i.e. So = I(0,O) + I(gO,O), SI =
(I(0,O) - 1(90,0))/S0, S 2 = (I(45,O) - 1(135,0))/so, and S3 = ( I ( 4 5 , ~ / 2 ) - I(135, 7r/2))/So.
The sampling scope records the Stokes components throughout the pulse and we can
study the Stokes vector dynamics for each pulse, averaged on the sampling scope over a
series of individual p h .
The fiber grating has an overd Iength of 7 -7 cm with 0.75 cm apodized sections at
both ends. The average grating index is very d y constant tkoughout the entire fiber-
The fiber ends are cleaved so that there is less than 1 mm of normal fiber at each end
in order to simplify the analysis of the non1inea.r interactions. The center position of the
fiber grating bandgap is tuned with respect to the laser wavelength by strainùig the fiber.
The grating birefringence is much larger than the normal fiber birefringence due to the
UV grating writing process. The buefimgence in the grating used in these experiments
is a p p r k a t e l y 4.2 x IO-' [42], so that the biefringent phase shift after propagation
through the grating is near r / 4 , or a quarter of a beat period. Incident intensities at the
beginning of the grating in the fiber core are estimateci by using the messureci values of the
energy incident on the input microscope objective, the objective optical loss, the coupling
efnciency into the fiber and the effective fiber core area These intensity estimates give
the average intensity and are accurate to f 20%. Linear pulse propagation measurements
are made using peak pulse intensities l es than 0.5GW/cm2, where nonlinear effects
are negligible. For noniinear experiments, the peak intensity was about 10GW/m2.
Although higher peak intensities are available fiom the laser, they might damage the
grating and are avoided in the experiments.
3.5.1 Polarizat ion Evolut ion for high detunings
In Figure 3.6 we plot the normalized Stokes parameters (3.21), Si (dash), Sz (lin+squares)
and S3 (line-circles), as a function of time for experimental pulses with linear intensity
(3.6(0)), and nonlinear intensity (3.6(6)), detuned such that f = 4.3. For both intensities
the input pulse was polarized such that (y12 = 0.1 (x12, so that most of the intensity
lies dong the x axis, which has the lower index of rehction, and is thus unstable to
energy exchange. The input phase lag between the two polarizations is A4mpu, = 30°.
As a reference we also plot the total intensity in the pulse, normalized to its maximum
value (thick soiid line). At such a large detuning, the nonlinear dects are not evident
in the pulse intensiw - the width of the linear pulse (Ml-width, half-maximum) is 79ps,
Figure 3.6: Ekperirnentdy measured Stokes parameters for a pulse with Linear intensity
(a) and nonlinear int ensity (b) , where the hequency detuning is f = 4.3. The thick line is
the normaüzed pulse intensity. We show Si (dash), S2 (he-squares) and S3 (line-circles) . For the pulse with linear intensity, the Stokes parameters are roughly constant acmss the
entire pulse, except at the very edges, which can be attributed to noise. For the pulse
with nonlinear intmsity the Stokes parameters foUow the intensity of the pulse.
while for the nonlinear pulse it is 75ps. By contrast , the phase lag throughout the pulse
is a strong function of the intensity. In the h e a r regime (3.6(a)), the Stokes parameters
Si, S2 and S3 me esentidy constant throughout the pulse, except at the very edges,
which can be attributed to noise. In the nonlineax regime (3.6(b)), the value of the
Stokes parameters follow the intensity throughout the pulse. The phase lag between the
polarkations can be determineci from the definition of the Stokes parameters (3.24). We
h d that the phase lag of the linear pulse is A#, = 135O, while for the nonlinear pulse,
the phase lag at t = O is A#NL = 102'. We expect the phase lag for the nonlinear pulse
to be smder than for the linear pulse for the following reason. We have 5 > nz, so
that, for linear intensities, A#, = nif' (Tjlc) L > O, where L is the length of the grating.
For nonlinear intensities we expect the phase lag to be A4,, z A$, + nrL @/c) L. But
if 1912 < 1%12, then n r L < O, so A&L c &.
Figure 3.7: Output phase lag for pulses detuned at f = 4.3, with nonlinear intensity, a s
a fundion of input phase lag for experimental data (squares) and numerical simulations
with the coupled mode equations (line). The data and simuiations agree weil to within
experiment al accuracy.
In Figure 3.7 we plot the phase lag at the peak of the pulse (squares) as a function of
input phase Iag using nonlinear intensities and pulses detuned to f = 4.3. We compare
this to the soiid line, which is the phase lag found by using the coupled mode equations
(3.19) to simulate a Gaussian pulse with a peak Uitensity 10GW/n2, and full-width, half
maximum widt h 8Ops, propagating through our grating. We obt ain agreement between
theory and simulations that is well within the error of the phase lag measurements.
3.5.2 Polarization Instability as a function of Phase Lag
In a recent paper [38] it was demonstrated experimentdy that polarization instabiliw in
the grating system is a strong function of detuning, f . Howewr, the formula (3.16) for the
critical intensiw required for polarization instabiliw is only valid for a monochromatic
impulse, and says nothing about the phase relation of the two polarizations. Recent
work has shown that for certain pulse profiles the polarization instabilty can be forever
suppressed, and a vector solitary wave can form [a]. These solitary wavg, which have
is reached. The factor fi cornes from assuming a Gaussian pulse profile in the theory
presented by Ahkmediev et al. [40]. For an intensiw IT > Ikr the vector solitary wave
requins a phase lag 4 = */2, and an intensity distribution
where Ix is the intensity on the x (unst able) axis, and Iy is the intensity on the y (stable)
axis. Unfortunately, no maiytic expression for the pulse shape of vector solitary wave
is known. One interesthg feature about the solitary wave is that it requirg a larger
intensity of light on the unstable axis, which is couater-intuitive, since it suggests that
the energy should switch to the stable aJEis. The unique pulse shape, along with the 90'
phase Iag between the polarizations, d o w s the soiitary wave to retain its shape.
In Figure 3.8 we plot the results of a simulation of an 80ps pulse propagating through
a 100m grating, with a Gaussian pulse profile. We use grating psrameters such that
Iht = 8GW/cm2 (which bear no relation to the parameters in the physical grating).
We simulate a pulse with a total peak intensiw of 8.5GW/cm2, divided according to
Equation (3.30). The use of a Gaussian profile meam that the pulse is not a solitary
wave. Nevertheless in Figure 3.8, which shows the results of the simulation when we &art
with a 90° phase lag between the polarizations, the pulse maintains a roughly constant
intensity profile; the profile along the unstable avis is shown. In Figure 3.9, which shows
the results of simulations in which we let the initial phase lag be oO, the intensity in the
unstable axis diminishes upon propagation.
The simulations shown in Figures 3.8 and 3.9 suggest that the amount of polarization
instabi1iw experienced by a pulse is a strong hc t ion of its input phase lag. The reason
for this is fairly dear. The term accounts for accumulation of phase mismatch
due to the effective linear birefÎingence in the grating. Even if that term is cancded
Figure 3.8: Numerical simulation of a pulse with an initial Gaussian profile and with the
correct initial intensity, and initial phase lag, to form a vectory solitary wave. Because
the profile is Gaussian the pulse is not a solitary wave, but the initial phase lag of 90°
has the e f k t of supprgsing polarization instabiiity so that the intensiw and pulse width
dong the unstable axis remain roughly constant. The portion of the pulse dong the
unstable aKis is shown. The samples are taken every 1000ps, during which time they
advances 13.7cm, so that the group velocity of the pulse is O.Ol4cmlps = 0.68c/~,.
Figure 3.9: Numerical simulation of the pulse in Figure 8, but without the 90' phase
lag between the polarizations. The pulses are again taken every 1000ps, and stiU have a
group velocity of 0.68c/EZ. Without the 90' phase lag, though, they are susceptible to
nonlinear energy exchange. The intensit3f on the unstable axis (shown) is attenuated as
it switches to the stable axis.
by nonlinear birefringence, there will still be a mismatch due to the initial phase hg
between components. This was verified by the following experiment using a fiber grating
(conducted by Dr. Richart Slusher at Lucent technologies). We denote a (2) the ratio of
the energy of the pulse on the stable axis, Est, to the energy carried on the unst able axis,
L:
In the srperiments, pulses were injecteci with t~ (0) = t/9, so that m a t of the energy
was carrieci on the unstable axis. The input phase lag between the unstable and stable
axes was then varied, and the value of a (L), where L is the length of our grating, was
plotted versus the input phase lag, At&,. Two values of detuning were used: f = 4.3
and f = 2.0.
In Figure 10 we show the value of o (L) versus A#, for the experiments at f = 4.3
(squares) and f = 2 (circles). We &O present the r d t s of nunerical simulations
for f = 4.3 (dashed iinej and f = 2 (soiid linej. The qualitative agreement between
theory and experiment is excellent. The quantitative agreement is also good, but suffers
somewhat from the error involved in setting the input phase lag. For instance, the
experimental data at A#in k 90O should be equd, as it is in the simulations. However,
we note from the simulations that the point at Aqbin = &go0 lies on a region of the
m e with a high dope, and since we have about a &5O error in our input phase lag, the
deviat ion is underst andable.
The effect of the polarization instabiliw at f = 2 is much greater than at f =
4.3, despite the fact that the same total intensiw was used in both sets of data. This
observation agrees with the Ipy curve in Figure 3.2, and with the experiments in reference
[38]. The pulse at f = 2 experience a great deal of nonünear pulse comprgsion within
the grating, and this enhances its local intensity. This enhanced local intensity, when
taken in combination with the fact that the threshold for polarkation instabiliQ is lower,
means that the amount of energy exchange is higher for f = 2 than f = 4.3.
Figure 3.10: Polarization instabilim as a hinction of input phase lag for f = 2 (circles)
and f = 4.3 (squares). Also shown are the results of simulations with the nonlinear CME
for f = 2 (solid line) and f = 4.3 (dash) . The simulations agree wit h the experiments to
within experirnental error.
3.6 Conclusion
In this chapter we have presented a set of coupled NLSEs relevant to the propagation of
optical pulses through a Kerr nonlinear fiber Bragg grating. Using these equations we
have identifid three distinct regimes of nonlinear propagation. For p S e carnet frequen-
cies that are far from the stop band of the grating, we have shown that an approxhate
theos, can be used to predict the polarization evolution of the pulse, if the pulse has
reasonably soliton-like parameters. If the ca.rrier kquencies are closer t o the stop gap,
then it shouid be posible to observe a vector solitary wave. We have presented numericd
simulations, using the nonlinear CME, to support t his claim.
Experiments were conducted to observe a variety of polarization phenornena in a
physical grating. For carrier fkquencies far detuned hom the stop gap, but with intensi-
ties fiil. higher than those required for soliton formation, we have observeci an intensity-
dependence of the Stokes parameters of the pulse. This intensity dependence is wd-
predicted by our nonlinear C m , and accords well with our heuristic understanding of
Ken: naalinear &ects in gratings. F k t h ~ o r e , we have derno&rM the suppfesion
and enhancement of polarization instability as a function of the initial phase lag between
the orthogonal polarizations. For carnier frequencies closer to the stop band, the polar-
ization instability is much larger, due both to the increased value of the coefEicient that
govenis the instability, and to noniinear puise compression.
The work in this chapter can be seen as a springboard for Further investigation of
polarization instability effects either near or even within the stop gap of a birefiingent
grating. Conversely, it can been seen as a cautionary srample for those wishing to exploit
Bragg grating solitons in a nomindy isotropic Bragg grating. Aithough the birefiingence
in a Bragg grating is high, the intensities required to observe Bragg gating solitons near
the gap will almost assuredly put one in a situation where polarization instability becornes
important.
Chapter 4
Gap soliton propagation in a
two-channel SCISSOR structure
4.1 Introduction
A waveguide-coupled rnicrodisk (microresonator) is an d-pass or phase-ody filter - a
100% transrnitting device that imparts a frequency-dependent phase. These microres-
onators have been used in a variety of areas inciuding waveiength division multiplexing
[45], laser fkequency stabilization [46] and cavity quantum electrodynamics [2]. The prop
erties of a sidecoupleci, integrated spaced sequence of resonators (SCISSOR), in which
a number of raonators are coupled to a channel mveguide, have been examineci [1][47].
It has been shown that light whose fiequency corresponds to the resonant frequency of
the microresonators will propagate with a vastly reduced group velocity, and a Mnishing
group velociw dispersion [l]. The reduced group velocity corresponds to a huge build-
up of the electromagentic field intensity in the microresonators, which means that the
nodinear properties of the system are enhanceci.
In (a) of Figure 4.1 a single microresonator coupled to a chaMel waveguide is shown,
dong with the mode profiles of both. The dimensions and materials shown in the figure
are taken fkom a device cmrentiy being hbzicated [lq. The chaMel wavegu.de and
resonator are both 0.55pm thick, and the resonator radius is 4.5pm. Incident light cm
couple into the resonator, with a coupimg strength determineci by the overlap integral of
the two mode profiles shown in the figure. It is assumed that no rdection occurs at the
coupling point, and that the microremnator has no radiation losses. Because no reflection
occurs, and neither device element is lossy, the mode1 systern is 100% transmitting. A
physical microresonator will, of course, be lossy, but the losses can be contained to l e s
than 10dBlcm [IO] which is essentially lossless over the micrometer dimensions of the
device. Aithough light is not attenuated, it wperiences a phase delay due to the coupling
to the microresonator. In (b) of Figure 4.1 the cross section of the channel waveguide is
shown. The microresonator has the same cross section. The gui* in the y direction is
provided by the index contrast between the GaAs and AlGaAs layers. The guiding in the
x direction is provided by the index constrast between GaAs and air. This means that
the modes are very tightly confinecl in the x direct ion, and hence justifies the assumption.
given below, that the couphg between the channe1 guide and the microresonator occurs
at a single point.
The SCISSOR system in (a) of Figure 4.2 is a 100% transmitting device, because
each of the individual resonators is 100% transmitting. Recently, the properties of a
SCISSOR with two channel waveguides (see (b) of Figure 4.2), one above and one below
the microresonators, have been investigated [44. In such a system light propagating in
the forward (backward) direction in the bottom channe1 is coupled to Lght propagating in
the backward (forward) direction in the top channel. Because the microresonators in (b)
of Figure 4.2 are equdy spaced, with period d, there are two distinct &ects that should
be observable in such a device. First, when the fkequency of light is near the resonant
fkequency of an individual microresonator, a, which is related to the microresonator
radius, p (see (c) of Figure 4.2) then the large build-up of the field in the resonator leads
to a large coupling between the channel waveguides, and light cannot propagate; second,
Top Down View
I r
Mode b
T d t t e d Overlap
Cross Section
0)
Figure 4.1: (a) Top dom view of a singIe mïcroresonator coupled to a channe1 waveguide.
Shown, too, is a sketch of the mode profiles of the device elements, and the mode overlap
at the couphg point. (b) Cross section of the chamel guide. The gui- in the x
direction is provided by the index contrast between GaAs (n cz 3) and air (n zz 1), so
that the mode is very tightly confineci in that direction. The microresonator has the
same cross section.
when the frequency of light is near the Braggfkquency of the periodic system, which
is related to the pexiod d, then the small couphg between the channe1 waveguides can
build up vio a Bragg reflection-type effect, so that light cannot propagate. Both of these
effects lead to a gap in the dispersion relation of light in the system, but while the second
&ect corresponds to a wd-known photonic band gap, and can be described by a set of
coupled mode equations (CME), the former e f k t does not. We call the gap that is due
to the field build-up in the resonator a 'rcsonator gap' to distinguish it hrom a Bragg gap.
At a given frequency outside either type of gap, there are two propagation modes
asociated with each propagation direction. We label these modes the bottom and top
modes. For the bottom (top) mode associated with either propagation direction, there iç
a larger field intensity in the bottom (top) c h a e l than in the top ( bottom) channel. The
dispersion relations associated with the upper and lower modes are heavily dependent on
the values of the coupling coefficients a b , ~b and a,, rct where the subscripts b and t label
the bottom md top couphg points. Whcn a fidd in a chme1 guide impings upon a
couphg point, the rc coefficient determines how much of the field amplitude couples into
the resonator, and the D coefficient determines how much continues on in the Channel
guide. When ob = a,, which we c d a symmetric device, then the dispersion relation
associated with either the bottom or the top mode is the same. When a b # a,, which
we cail an asymmetric device, then the dispersion relations associated with the bottom
and top modes are distinct. In the absence of nonlinearity, Iight with kequency content
either outside a gap of the system, or only slightly inside a gap (in a sense to be made
more precise below), is well described by a set of uncoupleci h e a r Schrddinger equations
(LSEs). The two parameters that enter the LSE - group velocity and group velocity
dispersion - are taken hom the dispersion relation of the system. Thus, for a symmetric
device, the parameters of the LSEk that describe light in the bottom and top modes are
quivalent, while for an asymmetric device the parameters of the LSEs are distinct.
In the presence of a Kerr nonlineari@, bght in the upper mode is coupled nonlineazly to
Figure 4.2: (a) 100% transmitting SCISSOR structure. (b) Tmxhannel SCISSOR stmc-
ture. The resonators couple light between the bottom and top channels, leading to two
types of gaps in the dispersion relation of light. (c) One unit cell of the twwAannel
SCISSOR-
light in the lower mode. To describe pulse propagation for a pulse with fkquency content
either outside a gap, or ody slightly inside a gap, we derive a set of coupled nonlin-
Schrodinger equations (CNLSE). Near a Bragg gap, the group velocity dispersion is
much larger than it is near a microresonator gap. in fact, near a microresonator gap the
dispersion am drnost vanish. This means thst for the same pulse width, a gap sotiton can
be excited near the microresonator gap with a much smaller intensity [q. In addition to
simple gap soliton switching, the two chit~ll~lels can be used to form an all-optical AND
gate using a coupled gap soliton scheme [4]. Again, the coupled soliton will form at a
much lower energy in a resonator gap than in a Bragg gap.
In this section we review the Bloch theary for linear media with thzee effective spatial
dimensions, and a periodicity in the dielectric permittivity in up to three spatial dimen-
sions. We then present a mode1 with which we can determine the Bloch functions of
the tw~channel SCISSOR device, which has a periodicity in the dielectric permittivity
in only one dimension. We start with Maxwell's Equations in a linear. non-magnetic
( p = po) medium in three dimensions,
dD - - - V X H ; -- - -V x E, dB ât ât
with constitutive equations
where r = (2, y, T ) , and n (r), the index of refiaction of the medi*m, is pcriodic aith the
lattice, that is n (r) = n (r + R), where R is any lattice vector. Because of this periodicity,
Bloch's theorem guarantees that the stationary solutions to Maxwell's equat ions can be
chosen to be of the form
where the crystal wavevector k Les in the first Brillouin zone, m is a band index, and
We normalize these Bloch functions such that
/ Z n 2 (r) E h ( r ) (r) = 6,,diW,
where Q is a normalization volume. Using the r d t s of a k - p theory [49] (501 we define
in terms of which we define the group velocity, a w & ? k i , and the group velocity disper-
sion, a2wd/ (aki8kj), associateci with a given k-vector
For later use we define the cornplsr column [50]
and the matrix operator
so that
where we have defineci the index matrix
In terms of these coliimnc,
where
and where we have defined a notation for an overlap integral,
where, since the m e vectors of the two columns are the same (k)? the integration
proceeds over one unit cell of the system.
4.2.1 Dispersion relation for the twcShannel SCISSOR
In this chapter we assume that both the chwrnel waveguides and the microresonators
support a known mode profle with the electric field in the transverse direction (E = EY),
and the magnetic field, H, everywhere orthogonal to E. It is not necessary to assume that
the E-field is in the jj direction, but it makes the nonlinear interactions more tractable.
We &O assume that the coupling of light into and out of the microresonator occurs only
at the points b and t shown in (c) of Figure 4.2, and that at each coupling point there
is no refiection. As discussed, there are two types of propagation modes: a bottom (top)
mode, in which more intensity is containeci in the bottom (top) charnel than the top
(bottom) channel. Each of these two modes can be associated with forward or backward
propagation, so that there are a total of four modes in the system.
To determine the Bloch hinctions and dispersion relation of the coupled microres
onator system we use a transfer-matrix technique [51] to describe the light within one
unit cell of the qstem, and then impose the trançlaticin symmetry of the ptsiorlic
medium. We denote the electric field in the bottom channel L (r) = S (z, y) 1 (r) Y, and in the top channel U (r) = S ( x , y) u ( 2 ) y, where S (2, y) is the mode profile as
sociated with the channel mveguides. We denote the electric field in the microres-
onator Q (y, p, 0) = T (y, p) q ( O ) y , where T (y, p) is the mode profile associated with the
waveguide, p is the radius of the microresonator, and B is the angle within the resonator,
measured counter-clockwise from the bottom coupling point ( s e (c) of Figure 4.2). At
the coupling points we assume [l] [47]
[ d o + ) ] = ( - '%) [ " O - ) ] 1
1 (a+) 2 ~ ; ~ cq 1 (a-
with a = d / 2 , and where we have introduced the notation a* = a f 6a, n* = r & 675
Or = f 6?r, where 6a and 67r are infinitesimal quantities. In order to conserve energy, the
value of the couphg co&cients, obi uti S b r muSt be choscm such that (ail2 + 1&l2 = 1
and orni = o ~ K ~ ? where i = b, t . I f oi and 6 are r d , then the second condition is
Away from the coupling points, the ody effet of propagation is the accumulation of
phase. Since we are using channel guides, the phase accumulation will be gowrned by the
propagation constant associated with the mode profiles S (x, z ) , T (y, p) . We assume the
propagation constant is equal for the channel guides and the microresonator, and denote
it Y = ncffw/c , where w is the frequency of the light, and where nef/ is the effective
index of refraction associated with the waveguide. Strictly speaking, ne,l is a function
of hequency, but in the following we ignore its frequency dependence. We assume that
Light is travelling in the forward direction in the bottom chamel, and in the backward
direction in the top channel. This wili generate the wave numbers + 1 kb 1 and - 1 kt 1 for the
bottom and top modes propagating in the forward direction. The wave numbers for the
other two modes are - (kbl and + 1 hl. For the chamel guides we h d 1 (a- ) = 1 ( O ) eiva,
1 ( d ) = 1 (a,) eiua and u (a-) = u (O) e-iua, u (d) = u (a,) e-lm. In the microresonator we
&d q (n- ) = q (O+) eiY"P, q (O-) = q (A+) eimp. Cornbining these expressions for phase
accumulation with the expression for the coupling (4.14) we find
where
We now impose the translational symmetry associated with the periodic medium,
where k is the Bloch wwe niimber, W e use a wave n w n k ra.t.her than a wave v a b r
because the periodicity of the structure is ody in one dimension. Equating the right
hand sides of (4.15) and (4.17) we fmd that
where the matrix N is
Equation (4.18) has non-trivial solutions only when (NI = O, h m which we fllid an expres-
sion for the Bloch wave number, k (w ) , which we can invert to determine the dispersion
relation, w (k) . The Bloch functions can be found by determinhg the eigenfunctions of
expression (4.18), and using the phase accumulation and coupling matrices to detennine
the electric M d ewxyxhcrc in thc s)stem.
Two quantities of interest in the twcxhaanel SCISSOR structure are the resonance
frequency of a single resonator a, and the Bragg frequency Q,,
The nominal value of the Bragg hequency quoted here is determined by assuming that
the coupling between the ch=& is very weak. We can &O define the Bragg wave
number, ko = ald. The coupling coefncients of the two-charnel SCISSOR stmcture can
be chosen in three distinct manners. In what we caU a symmetric structure we choose
the couphg coefficients such that ab = oc. In what we c d an nsymmetric structure we
choose us # a,. For a onechannel SCISSOR structure we choose either ab = O or a, = O
(but not both) so that one chamel is effectively cut off. The one-channel stmcture has
been studied extensively [Il, so we do not consider it here.
Figure 4.3: Dispersion relation for a symmetric tw~chasu1el SCISSOR with q, = nt =
0.95. The resonator gaps open at w = 0.6nb and 1 .2Qb. The Bragg gap opens at w = Rb.
Note that the upper and lower edges of a Bragg gap are at the same k value, while in a
resonator gap they are at k = O, b.
In Fibwe 4.3 nTe plot thc dispersion relation for a qmmetr ic , b-cxhannel SCISSOR
structure with n,/l = 1.5, %/Rb = 0.6 and = at = 0.95. We note the opening of the
two types of gaps: at w 2 Clb, a Bragg gap opens; at w zz and w = 2% a resonator
gap opens. In the vicinity of a Bragg gap, the curvature of the dispersion relation is
very high, while near a resonator gap the bands are almost completely Bat. This means
that the group velocity dispersion of a pulse propagating with kequency content near a
Bragg gap is very high, whereas, when the hequency content is near a resonator gap,
the group velocity dispersion is v q low. At the band edges of both types of gaps the
group velocity Mnishes. The upper and lower edges of the photonic band gap occur at
k = b, while for the resonator gap they occur at k = O, b. Because of this, conventional
coupled mode theory (31, which assumes that the wavenumbers at the edges of the gap
are equal, cannot be used to describe pulse propagation throughout a resonator gap.
In Figure 4.4 we plot the dispersion relation for an asymmetric, two-charnel SCISSOR
stmcture with nef/ = 1.5. %/Ob = 0.6, and with 06 = 0.90, a, = 0.96. We concentrate
Figure 4.4: Dispersion relation in for the top mode (solid h e ) and the bottom mode
(dashed line) of an asymmetric SCISSOR with 0 6 = 0.9 and a, = 0.96, in the vicinity
of a resonator gap. The horizontal distance between the ling is related to the phase
velocity mismatch between the two moda, and is a strong function of fkequency.
on the hequencies in the vicinity of a resonator gap. The dashed line is the dispersion
relation of the bottom mode, while the solid line is the dispersion relation of the top
mode. For the s&e of clarity we have extended the Brillouin zone so that the dispersion
relations represent forward propagat h g light . The horizontal distance between the two
dispersion relations, which gives the phase velocity mismatch between the two modes, is
a strong function of frequency: the mismatch at the band edge just above the resonator
gap is twelve times larger than the mismatch w d away h m the gap (at w/Q, = 0.5 or
0.7). A similar effect has been observed in fiber Bragg gratings [38], but the magnitude
of the mismatch is much larger here.
4.3 Coupled NLSE
In this section we include the &ts of nonünearity in a twc+channel SCISSOR structure.
We specificaily discuss a I k r nonlinearity, but the theory is suffciently general that it
can be extended to invgtigate other typg of nonlinearity. In the presence of a Kerr
nonlineaxity we derive a set of coupled NLSEs that are Müd for pulses with fiequency
content anywhere on the dispersion relation that is not too deep withui either a resonator
or photonic band gap [7][34]. In such a situation we expect the pulse to be weU-described
by one Bloch function, centred at the carrier wave number of the pulse, and modulateci
by a slowly-varying envelope function.
When we introduce nonlinearity into the system, we must stiil satisfy the MaxweU
equations (4.1) , but the constitutive relation for D becornes
D (r, t ) = €on2 (r) E (r, t ) + P N L (r, t ) . (4.20)
Rat her than working directly wit h the electric and magnetic field, we first introduce the
potentials A and N, with
B = V x A ,
These potentials automaticdy satisfy the divergence equation of (4.1). It is easy to
verify hom the curl equations of (4.1) that the A and N fields ais0 satisfy
We write the dynamical equations (4.22) in matrix form as
a* in- = M x !P +i9, at
where the column Q of cornplex vector fields is given by
and the column
The index ma& n, and the matruc operator M, are dPfinPrl in Eqwtions (4.10) and
(4.8) respectively.
For an arbitrary complex column, 8, we can construct a column @ + $ representing
real potentials, where the physical conjugate, 9, of a column vector (4.24) is defined
- We write a real column as \y+ + \Y- where @- = a+. Defhing the operator
we see that the physical conjugate of L9+ is -Lq-. Hence, if we likewise have 8 =
0, + 8- and satisfy
LIS+ = ie+, (4.28)
then the dpamical equations will he .satifid. W' extract a set of coupled mEs h m
(4.28) using the method of multiple scales.
4.3.1 Multiple Scales
The vector notation introduced above is designed to deal with systems that have a one,
two or three dimensional perîodicity in their dielectric constant. In the ~ c h a . n n e l
SCISSOR structure, the periodicity is only in one dimension, so that the wave vectors
associated with the Bloch functions point in the z-direction. Because of the existence of
the two distinct modes, associated with light in the upper and lower channels respectively,
we label an arbitraxy wave vector k (w) = k, (w) 2, where for the remainder of the chapter
the subscript i = b, t indexes the bottom and top modes respectively.
In the absence of nonlinearity, the general solution of (4.28) is of the form
where the f& are r)imPnsidess expansion CQ* a ciirries the units of ?Zr+, and
the column 9<lki (r) is defined in Equation (4.7). In the presence of nonlinearity such a
solution cannot be used. Instead, we use the method of multiple scales [50] (71 to constmct
approximate solutions to (4.28). We replace the constant coefncients fqk in (4.29) with
hinctions fqk, (2, t) that vary slowly in space and time. To keep track of the slowness
we explicity separate different length and time scales in the problem by introducing a
smallness parameter, 7, and writing functions as
It is assumed that F va.ries equaily significsntly as each of its arguments varies over a
range d, or a period, r. These are chosen to be the shortest length and time scales in
the problem; d is taken as the size of a lattice constant, and T as l/w, where wili be
identifieci below. Then the ranges and periods d(") = d/qn and dn) = r/vn define the
multiple scales of the problem. We wiU quantify the value of below (after Equation
(4.55)). Derivatives are given by, for example,
We seek to describe the propagation of light pulses that v a q slowly relative to a given
carrier frequency? W. We define = kb (G) and & = kt (G). We seek an approha te
solution to (4.28) of the form
where we assume we have light in a bottom mode and a top mode, but we have not
specified the propagation direction of either mode. We have separated the cornponents
fmG (2, t) fkom the rest of the components because it is assumed that the fiequency and
wavenumber content of the pulse describeci by (4.32) is such that functions in band m
contribute most to the value of a+. These components will be called the principal comp*
nents. The 0 t h components, f&, (2, t), with q # rn, are called cornpanion components,
and are asstlmed to be smaller thaa the 1'. (z,t). To capture the relative stnmgths of
the principal and cornpanion components, we take
These slowly-varying quantities have no dependence on ro or to. Rom the dependence
of !P on N and A (4.24), we see that the ansatz for @+ (4.32) leads to
2 (2) ~ z ~ ~ ~ k ~ (r) f FS~H,,,~~ (r) + [r)f"i: + 11 Fqz, + ..] Ha& (i) i=u,l q#m
We construct an expansion of 138, in powers of r),
This construction requires us to find the expansions of the terms that appear in 13. These
tue s h p l y
where
and, for j > O,
where the matrix V is defineci in Equation (4.12).
We assume that the nonlinearity is weak and due to a third-order response which, at
kequencies of interest, is far off resonance. Then an appropriate mode1 for the nonlinear
polazization Par. (r , t ) is [52]
We ignore third harmonic generation on physical grouuds. We have assumed that the
underlying material is nondispersive, and while this may be valid for fiequencies near Y,
it will likely not be valid for frequency ranges extending to w 2 3W; furthmore, the
assumption of no absorption at w 2 3W will likely be in error. We expect, on physical
grounds, that in many cases the actual materia. dispersion and absorption will mske any
buildup of the third hannonic iinlikely, so that our mode1 will be adequate. Under these
assumpt ions the expression (4.38) reduces to
Then, since we have assumed that our electric field is polarized in the ij direction, we
find
PnL (r, t ) = PNL (r, t) 6 = ~ Q X ~ W Y Y (r) ( E (r, t)12 E (r, t ) e-"y + cc, (4.40)
where E (r, t) = E (r, t) y . To use this nonlinear polarization in our multiple scales
formalism, we make the following assumption for the strength of x" (r),
where €0- (r) has units of permittivity and is assumed to be on the order of the largest
linear media permittivity, or smaller.
Using the ansatz for N (r: t) (4.33), the definition of N (r, t) (4.21) and the Maxwell
equations (4.1) we find that , in light of the fact that the nonlinear terms are considered
to be O (q2) :
Using the expression for the ekctric field (442) we c m àetefmine an -ession for the
nonlinear polarization, which we write as
with
where
4.3.3 Coupled NLSEs
Using (4.34) and (4.45) we can solve the pulse propagation equation (4.28) in te= of
the multiple scales quantities. The mechanics of the derivation have been performed else-
where for a system with only one carrier wave vector (501. To include the nonlinear terrn
we follow previous work on birefkingence in periodic systems which expiicitly accounted
for the inclusion of two wave numbers (341. We find
The nonlinear co&cients are
where e,, (r) is the periodic portion of the Bloch function (4.3). The values of a z , a=
and fldt cm be determineci fkom those above by interchanging b t in aU expessions.
The noniinear coefficients are superscripted in such a m m e r because a x accounts for
self-phase modulation, ax accounts for cross-phase modulation, and #si accounts for
phase conjugat ion ra]. When the group velocity in (4.47) is non-zero, then the equations can be used to
describe two distinct situations: 1) the light in both modes is travelling in the same
direction; or 2) the Iight in one mode is t ravehg in the forward direction, while the
light in the other mode is travehg in the badcward direction. In the first situation,
the value of < is much smder than either a" and a", because in every portion
of the unit cell the light associated with the two Bloch functions is counterpropagating,
so the overlap integral is smd. In the second situation the light associated with the
Bloch functions is cepropagating, so the < codcient is simüar in strength to the other
nonlinear coefficients, but since the pulses would then have wave vectors of the opposite
sign, the quantity C would be enormous, and the phase conjugation process would still
not build up. A third situation occurs when the group velocity in (4.47) is zero. This only
occurs when the carrier fiequency of the light is considered to be at a band edge. h such
a situation the four modes of the system (top and bottom travelling in the forward and
backward direction) reduce to two modes, since the h a r d and backward propagation
directions no longer have meaning when the group velocity is zero. The two modes can
be considered as: light with phase velocity in the forward direction in the top (bottorn)
channel coupled to Iight with phase velocity in the backward direction in the bottom
(top). Again the < coefncient is very s d , because in every portion of the unit ceil
the light associated with the two modes is accumulating phase in the opposite direction,
so the overlap integral is s m d .
We can relate the f fields to physical quantities by using (4.42) to mite the full
eiectric field in terms of f , and using (4.21) and (4.24) to write the full magnetic field in
tems of f , and then writing the total energy at time t ,
We find that the energy in the ith fieid is
where we have introduced an effective area, Aeff, associated with the mode profiles in
the wave guides. If the quantity v, # O, then we can d&e a new field,
where lgd, ( z , t ) 1 represents the power in the ith field at a given z, t point. The total
energy in the ith field at spatial point z is
If we consider a carrier kequency and wave number that are not at the band edge, then
the g a (r, t) become the fields of interest. We can d&e an effective index of &action
associated with the g&, (r , t). To do so we consider a monochromatic excitation, and
consider its evolution as a b c t i o n of z. We h d that the nonlinear phase accumutation
CHAPTER 4. 86
- of the excitation is 4N, (2) = nef'
t 2 e kg, where the effective mnlincri.r index of &action
We have stresseci the hequency dependence of n;;{& because close to the gap vi 4 O and
nef' becomes enormous. This occurs for two reasons. First, when the group velocity (2)m;k,
tends to zero, Iight spends a much longer time interacting with the nonlinearity. Second,
the nonlinem coefkient n ; z i is referenced to the energy flw, which the nonlinearity
itself depends on the electric field strength. For a given energy flux, the electnc field
strength is greater for a smder group velocity.
This theory is only valid if the material parameters d o w a suitable choice of the
smallnes parameter, 7. The specific value of q for a given structure is
where A is the spatiai width of fmG, and g is a parameter
and
(4.55)
that is given by $01
Once (4.56) and (157) have been used to find the lotver bound on g, then the theory is
valid if q « 1.
4.4 Discussion
We first concentrate on gap soliton effects that should be obsewable in a symmetric hm
channel SCISSIOR structure. One of the most striking dec ts in Figure 4.3 is that the
bands in the dispersion relation just abow and below the resonator gap are extremely
%at. These Bat bands correspond to frequency ranges with a very low group velocity
dispersion (GVD). The d value of the GVD a& the edge of a resonabr gap is in
marked contrast to the G M at the edge of a photonic band gap. For a Bragg gap in a
fiber grating it is known that the GVD at the band edge is apprmimately (341
2 2 .b,, = - (C) , ( 6 4 n
where bw is the width of the gap. In Figure 4.5 we plot the G M at the upper band
edge for a microresonator gap in a synunetric ~ c h m e l SCISSOR structure (circle -
dashed iine), and for a Bragg gap in a fiber grating system (square - solid line) as a
function of gap width, 6w. We assume that 0, = 0.64 and that wb = (207r) ps - l , which
corresponds to d = 10pm, and p = 5.3pm. There we see that decreasing the gap width
i n m e s the GVD in the vicinity of a photonic band gap, but it derreases the GVD in
the vicinity of a resonator gap. For a gap width of 6w = Rb/lO, the GVD of the Bragg
gap is 80 times larger than that of the resonator gap; for 6w = 4 /105 , which is roughly
the gap width that was employed in fiber gratings, to observe Bragg grating solitons [l?] ,
the GVD of a photonic band gap is about 101° times Iarger than that of a resonator gap!
Although it is true that the two-chamel SCISSOR structure also supports Bragg gaps,
we compare the SCISSOR structure to a fiber grating because the form of the GVD near
a Bragg gap in a fiber grating (4.58) is particulady simple, and because fiber gatings
are often used to experimentdy observe nonlinear dects in the vicinity of a Bragg gap,
and hence ~epresent a competing teckoiogy for the gap &ton propagation we diseusa
below.
If light is propagating in only one of the channels of the two c h d SCISSOR stmc-
tue , Say the bottom channel, and with fiequency content near or within a raonator gap,
then the relevant equation of motion is the familiar NLSE,
where the group velocity term vanishes since vb = O at the band edge. We assume that
the fiequency content of the light is near the upper edge of the resonator gap, so that the
Figure 4.5: Group wlocity dispersion (GVD) at the upper band edge for a resonator
gap (circledashed line) aud a Bragg gap in a fiber grating with the same gap width
(squaresolid line). The GVD near a Bragg gap is always much larger, especiaily when
the gap width is smd.
GVD is of the correct sign to d o w soliton propagation. It is known that (4.59) supports
soliton solutions of the fom [53]
where
and the signs of the deturings 6 and A are chosen such that these coefficients corne out to
be real. The parameter 6 detennines the height and spatial width of the soliton, whereas
A determines the velocity. We label the frequency at the upper band edge at w z b . The
centre hequency of the soliton, then, is w, = w z b + 6 + A, and
the pulse is C. In order for most of the fkequencies of the pulse
the gap we require w, + 2C 5 w a b . This condition can be met f o ~
if we set A = C/ (2@) and 6 = -C (m/2), where M 2 16.
the frequency width of
to be cont ained wit hin
an arbitrary value of C
However, we note that
the pulse width is limited by the fact that the NLSE (4.59) is d y valid for f i ~ u m c k s
slightly inside the gap. If we fix M = 16 then we should have C 5 (6w) /20, where (6w)
is the width of the gap.
We now determine the energy required to excite the soiiton (4.60). Using the form of
f~ (4.60) in the expression for the energy (4.51) we find
Therefore, because the GVD near a resonator gap is so much smder than near a Bragg
gap, the energy required to excite a gap soliton with the same pulse width, and the
same depth within the gap, is much lower in a resonator gap. We define the ratio
Sad = &sGattm adittm ukb Jres/Ezb Jbom, where E,btmJ,(~qg) is the energy required to excite
a soliton in a resonator (bragg) gap. In Figure 4.6 we plot Ssd as a function of (6w).
When (6w) = 10-5nb26, the resonator gap has a GVD 10'' timg smaller than in the
correspondhg photonic band gap, so that the excitation energy required to form a gap
soliton in a resonator gap is los times smaller! We have not included the e t of
the underlying material dispersion in our calculations, so the enhancement factors in
a physical system would not be expected to be so large. We note that the reduced
formation energy corresponds to an inmeased soliton formation length, the length over
which a pulse with soliton-iike parameters will evolve into a true soliton. For instance,
in a bare optical fiber a typical formation length is 100m, but in a fiber grating the
formation length drops to about lem. For light in a resonator gap the formation length
will be quite large so that it would be difficdt to o b s m a true gap soliton. However,
part of the value of a gap soliton is its iw in switchmg schemes, where it is not essential
that the pulse be exactly a soliton. A Gaussian pulse with the same pulse width and
amplitude as in (4.60) will propagate over a fraction of a soliton width without a marked
change in its characteristics By constrast, a Gaussian pulse with the same pulse width,
but a much lower amplitude wilI decay very quiddy, since all of its hequencies w d d be
Figure 4.6: Plot of SSd, the ratio of the energy required to form a gap soliton in a
raonator gap, to that required in a Bragg gap of the same width. For s m d gap widths
Ssd is very tiny, indicating that a gap soliton will form in a resonator gap at much lower
energies .
within the gap.
If light is propagating in the fom-ard direction in both the bottom mcl top cliaf~lleis,
then the CNLSE (4.47) must be used to describe pulse propagation. We have mentioned
that the phase conjugation term can be ignoreci, but that the crossphase modulation
term m u t be included. The gap soliton solution (4.60) can be actended to consider a
generalized gap soliton [4]. When the SCISSOR stmcture is symmetric, so that the GVD
and nonlineax coefficients for the two c h m & are equal, then the generalized gap soliton
takes on the particularly simple form
fAb (L, t ) = \ I i ~ e x ~ (iBzz) exp [-i (6 + A) t] sech (Bir - Ct) ,
and f& (2, t ) = jmzb ( z , t). That is, the solution is just (4.60) with the amplitude
reduced by a factor If either fdb ( 2 , t) or fdt ( z , t ) is of the form (4.63) and is
propagating alone, then it d o s not have enough amplitude to form a soliton, and since
its frequency content is within the gap, it will be strongly attenuated. However, if both
pulses are propagating together, then they will form a generalized gap soliton and will
propagate without attenuation. La this mariner a simple AND gate c m he constmcted.
Note that the total energy requireà to form the generalized gap soliton with two puises
is twethiràs the energy requireà to form the slngie gap soliton with only one pulse.
This is because for pulses with the same hequency and polarization, the coefkient for
cross-phase modulation is twice as strong as for self-phase modulation. The situation
discussed by Lee and Ho (41 involveci orthogonally polarized pulses, for which the cross-
phase modulation codcient is only tw&hirds as strong as the self-phase modulation
coefncient, and so thek generalized gap soliton required more energy than a single gap
solit on.
We now briefly turn to the consideration of an asymmetric two-channe1 SCISSOR
structure. The above gap soliton theory assumeci that light propagating in either chamel
observed the same GVD. If, due to difEculty in the fabrication process, q, # a,, then
the light in the two cha.nneIs will see a different GVD. Nevertheless, the group velocity
at the band cdgc gcrp is still zero, and the generdized gap soliton describe3 above cm Le
further generalized to include the different GVD. In practice it is di€Ecult to obsewe gap
solitons. It is somewhat easier to observe what have been cded Bragg solitons, which
propagate in fiequency regions outside the gap, where the group velocity is much less
than the speed of light in the background medium. In optical fibers such solitons have
been observed with group velocities about 0.5c/n [16]. When the system has a b # a,,
then it should be possible to observe a vector soliton of the sort describeci by Menyuk
[54]. We do not discuss the observation of such solitons in this chapter. Mead we
investigate the mismatch in the Iinear properties and nonlinear coefEcients observed by
light in the two ch-& in an asymmetric device.
In Figure 4.7 we plot the value of the GVD at the upper band edge of the £kt
rgonator gap. We use the following parameters: Q = 0.97, w, = 0.6wb, SIb = (20n) ps- l ,
ne/f = 1.5. This corresponds to a resonator spacing of lOpm and a resonator radius of
5.31pm. We vary the value of a, hom 0.97 to 0.90, and plot the value ofwkb (circle-
Figure 4.7: Group velocity dispersion (GVD) a t the upper band edge for the bottom
mode (circledashed line) and the top mode (square-solid h e ) , in an asymmetric device
with ab = 0.97 and a, mried fiom 0.9 to 0.97. The higher the asymmetry, the higher the
mismatch ir: GVD between the modes.
dashed h e ) and wUEt (squaresolid line). We see that when ol < 0.96, the two values
diffa by about 30%. This, of course, changes the character of the type of coupled soliton
that c m be observed in the system. Nevertheless, a generalized gap soliton solution stiil
exists [4], as do vector soliton solutions with the fiequency content outside the gap (541.
4.5 Conclusion
We have demonstrated that a tw~chwinel SCISSOR structure posçesses two distinct
types of gaps: a Bragg gap, associateci with Bragg rdection, and a resonator gap, as
sociated with the resonant fkquencies of the microresonators. We have derived coupled
nonlinear Schr6dinge.r equations that govern the propagation of pulses with frequency
content anywhere except too deep iri within either type of gap, in the presence of a Kerr
nonIineariQ.
We have shown that the twcdmmel, SCISSOR structure supports a gap soliton, which
can be used for optical switching, in both types of gaps. However, the energy required
to observe such a gap soliton is much smaller if it is within a resonator gap. This is a
direct consequeme of the very flat dispersion relation in the vicinity of a resonator gap.
W e have shown, too, that the structure supports a coupled gap soiiton that can be used
to constmct a logical AND gate. Again, the energy requirements in the v i c i n i ~ of a
resonator gap are much smder. In a future work we will discuss the issue of coupling
iight into a gap soliton field pattern.
Chapter 5
Theory for a grat ing-waveguide
structure with Kerr nonlinearity
5.1 Introduction
The refiection properties of multilayer didectric based grating-waveguide structures (GWS) ,
comprised of a substrate, a waveguiàe layes, and a grating, have been investigated for
some t ime (1 11 [55] [56] [57] [Ml. It is known t hat such systems exhibit a resonant response
to an incident plane wave caused by a coupling of the plane wave to a waveguide mode
of the system via the grating (551. Such resonances, which Iead to a sharp spike in the
reflection spectrum, and a sharp dip in the t r d i o n spectrum of the GWS, have
been observed experiment aUy [55] [56] [SB], and described theoretically [56] (571. However,
both the theoretical and experimental discussions of the GWS have concentrated on thin
gratine in linear optical media, where by thin we mean that the thickness of the grating
is much less than the wavelength of light in the medium. In this chapter we consider the
general case of a GWS with a grating of arbitrary thickness in which one or more of the
layers is Kerr nonlinear.
In the absence of a Kerr nonlinearity, we deve1op a transfer-mat& formulation of the
GWS. The transfer matrix formulation is well h w n as a bol for waveguide design in
the absence of a grating, and for andyzing stacked dielectric layers [591. G' ~ven a mon*
chromatic impulse of known frequency and wave vector, each layer of the waveguide can
be characterized as supporthg either plane waves or ewescent waves, travelling either
upwards or downwards. The field in a given layer is related to the field in a neigh-
bouring Iayer vio Maxwell's saltw conditions, and the application of physical boundary
conditions determines the total field everywhere inside the structure. The addition of a
grating element complicates the transfer rnatrix approach, because the grating couples
light between a countably infinite set of diffrâction orders. We use a Green fùnction
approach [60] to derive the transfer matrix associateci with a grating for both TE and
TM polarized light. We also use the Green fùnction approach to describe the effects of a
Kerr nonlinearity in one or more of the layers. However, in the presence of nonlinearity a
simple tramfer-matrix approach cannot be used. Instead, we present a set of differential
equations that describe the build-up of the various fields in the system.
We examine the dec t of gating thickness on the GWS, in the absence of nonlinearity.
For a thin g r a t a , the bandwidth over which refiection occurs is nanow, and the intensity
of the field inside the waveguide mode can be very large. Conwsely, for a thicker grating,
the bandwidth of the reflection spectrum is larger, but the intensiw of the field in the
waveguide mode is much smder. We include the efFects of materiai absorption in the
system by defining a complex dielectric constant in one or more of the layers. We show
that since the intensity of the field in the waveguide mode is much smailer in the presence
of a thick grating, the thick grating GWS is much more tolerant to material absorption.
This allows us to design a low-loss rdector for W light.
In the presence of nonlinearity, it is known t hat the GWS exhibits optical bistabiliw
[61] [62]. Bistability can be used for ail-optical switching, but there is a tradeoff inwlwd,
because to obtain a low intensity threshold for the switching, a thin grating is required,
but a thin grating has a narrow bandwidth, and will thus lead to a slow device. In this
chapter we present a simple scheme based on crossphase modiilatinn that should d o w
for a much faster switching time-scale. We consider a CW beam being 100% rdected
by a very thin grating. We then inject a short pulse into the waveguide mode which will
change the effective index of refiaction of the guiding layer of the GWS. If the short puise
is sutiiciently intense, t hen it will shift the resonant wavelength of the system so t hat the
100% rdecting device becornes almost M y tr&tting for the duration of the pulse.
The theory in this chapter considers a grating that consists of a periodic variation in
the dielectric permittivity in one direction. The GWS, then, has two effective dimensions:
the dimension of the periodicity (x), and the dimension of the stacking of the layers (2).
However, the theory c m easily be extendecl to consider a medium in which the periodicity
in the dieletric permittivity extends in two dimensions (x, y) [63] (641. Such a system has
been investigated in the presence of a X ( 2 ) nonlinearity (641, but not X ( 3 ) , and mt in the
presence of material absorption.
This chapter is dividecl into fiw sections. Tn Section 5.2 we give a heuristic picture
of the operation of a GWS. In Section 5.3 we deveiop the transfer mat& theory for
a linear system, and derive the appropriate differential equations for a nonünear layer.
In Section 5.4 we present numerical simulations for the low-loss reflector, and for the
nonlinear switching scheme.
5.2 The GWS and the Guided Modes
We fbt give a generd description of the operation of the GWS in the absence of non-
lineari@. In the absence of both a grating and nonlinearily, it is well known that the
stratfied medium of Figure 5.1 can support guided modes [65] so that, for a giwn fie
quency, light cm propagate dong the x direction with an electric field profde
1 B a ~ h a d Mode: 4 - K \ F o w d Mode: k,+ r 1 I 1 - \ la, I
Figure 5.1: Schematic of the GWS system. The grating in layer one can couple an
incident plane wave into a guided mode of the stmcture if the appropriate resonance
condition is met. In principle there couid be any number of homogeneous layers beneath
the grating.
where ,û (w ) is the propagation constant associateci with the guided mode at frequency w,
the function Ê (r) is the normalized guided mode profile, and A is the amplitude of the
field in the mode. For a given fkequency, u, we cm determine P from a transcendentd
equation [65J.
A guided mode of the system must be evanescent in both the air and substrate regions,
so we cannot excite the waveguide mode by injecting a plane wave £rom the air region.
Instead, we would have to introduce an ev811escent wave, for which the z component of
the wave vector was equd to P. One common way to do so is to introduce a plane wave in
the vacuum region, and then use a grating to difbact sorne of the incident wave into the
guided mode, as in Figure 5.1. There, for a partiniltu combination of incident frequency
w and incident angle 4, a guided mode wiU be resonantly excited. Roughly speaking, the
where K = 27r/A is the momentum irnparted by the grating, and A is the period of
the grating. We can excite either a forward or backward propagating waveguide mode
(f respectively), and when # = O these two conditions become degenerate. For a fixeci
fiequency, w, the two angles associatecl with phase matchmg are
For a fixed angle, q50, we note that p - 2 m e J f /&, where Xo is the wavelength of light
in vacuum, and where nef , is only a weak function of wavelength, so that to good
approximation the forward and backward modes are atcited at vacuum wavelengths
In addition to coupling light into a waveguide mode, a grating will also couple light
out of a waveguide mode. Close to the resonance condition the reflection is enhanceci
because iight coupled into the waveguide mode by the grating is also coupled back out.
In fact, for a nmow range of wavelengths, almost 100% rdection can be achieved. The
presence of a Kerr nonlinearity in one or more of the waveguide layers will, of course,
affkct the position and the nature of the reaonmce, a d andits iheshape.
5.3 Green Function Theory for
We seek a physical description of the situation shown
S trat ified Media
in Figure 5.1, where a plane wave
with frequency w is incident from vacuum (layer O) ont0 a stratifieci, non-magnetic
(p = po) medium. The first layer (layer 1) of the stratifieci medium is a grating with
periodicity A, and with associated grating-momentum K = 27r/A. The grating materials
will have, in general, a cornplex dielectric constant. The subsequent layers are d o m ,
c c m & a n t a n d a K e r r m but have, in genera both a rnmpkx dielectnc ~esponSe,
To describe the system, we follow the strategy outiined by Sipe [60]. There it was noted
that a stratifieci medium can be describeci by using a set of fiesne1 co&cients to deter-
mine the &ect of the Maxweil salbtrs conditions at each interface, and by using a Green
hrnction technique to determine how the field evo1ves within each layer. We write the
relative dielectric permittivity and the Kerr co&cient of the ith layer as
where for the rernainder of the chapter the subscript i indexes the layer, and the super-
script rn indexes the Fourier order. We use the background permittivity, &, to determine
both the appropriate plane wave expansion for the electnc field in the ith layer, and the
Fresnel coefficients.
The &ect of the grating is to generate dSraction orders with wave vectors that
have the same magnitude, but different 4 and Z components. We label the set of wave
vectors appropriate to the system with two subscripts and one superscript: us, where
the f subscript denotes whether the wave is propagating (or decaying) in the upward or
downward direction. The wave vector associated with the mth diffraction order, in the
itqayer, propagating (or decaying) in the f direction is
with
where W r w / c and it is the x-component of the wave vector of the incident field, n =
W sin$, where t$ is the incident angle. The quaniw w?, the z-component of the wave
vector of the mth order ciifhacted field in the ih layer, is taken such that the ùn (tui) 2 O,
with Re (wi) 2 O if Im (wi) = O. This choice of up ensures that the v z wave vector gives
an upward-propagating (or ewescent) wave, while v: gives a domward-propagating
(or evangcent) wave. The full electric field in the ith layer is a superposition of the ail
the diffraction orders:
E, (r, t ) = e-"eG Er ( z ) eldx + c.c., m
where C.C. stands for complex conjugate, and r = (x, 2) . The x-dependence of the electric
field is contained entireiy in the exponential term. The field can be polarized in either
the s or p direction, with
In the absence of either a grating or nonlineanty, the weU-known transfer-matrix
technique can be used to solve the system of Figure 5.1. In such a situation there is
no couphg between the Fourier components of the electric field. The light within each
layer is described either by plane waves or evanescent waves; the &ect of the interfaces
is describeci by the Resnel coefficients. In the presence of either a grating and/or nonlin-
carie, Fresnel co&cients c m stil be used to describe the &kt of the interfaces [60], but
the evolution of the field amplitudes within each layer must be treated more carefully.
In order to determine the manner in which the Fourier co&cients of the expansion
(5.8) evolve within a layer, we turn to a Green function Eormitlatim of ~ldaxwell's qua-
tions [60]. We start by considering a homogeneous medium, with background dielectric
permittivity Fi, into which is embedded either a grating or a nonlinear slab. The grating
or slab is confinecl to the region 4 c z < 4-1 (see Figure 5.2). We define the total
poiarization, Fi (r), in the presence of an electric field, Ei (r),
- Pi (r) = E&& (P) + Pi (r) + C.C. (5.10)
where Pi (r) is the perturbation polarization - the portion not described by the back-
ground relative permittivityl 5, - and can include both linear and nonlinear terms. This
Homogeneous Background Penniftttvity z = d,,
i z z = di
Homogeneous Background Pennittivity
Figure 5.2: Grating or nonlinear slab, with average linear permittivity E , embedded in a
hornogeneous background medium with permittivity 2. The Green function theory c m
be used to soive for the fields within the perturbation.
perturbation polarization is given by
where PT (2) has been defined implicitly. In writing down the expression for the non-
linear susceptibility we have assumeci an instantanmus response, and we have ignored
both third-hannonic generation (THG) and the tensor nature of ~ ( ~ 1 . Tgn~ring THG is
a reasonable physical approximation because many media that are opticdy transparent
at w, are absorbing at 3w, and this absorption will make the build up of THG iinlikely.
lgnoring the tensor nature of reStrictS the validiw of the naLineor portion of this
theory to s-polarized light. H o ~ e ~ e r , in the application we discuss below, such a restric-
tion is not damaghg- In Figure 5.2, the perturbation polarization will be zero when
z > or r < 6 In the Greemfunction formulation we consider this perturbation
polarization as a source term for the electnc field. We write [601
EY (*) = p (4) p ? ' ( z - 4 > +p (di.,) e*F(-4-d + 1 Gr ( z - z') Pr (z') dzl, (5.12)
where G," is a Green function,
and where the coefficients Am (4) and Bm are related to the electric field d u e s
at z = di and z = &1 respectively. In writing the Green function we have used the step
function 8 (z),
and the Dirac delta function, 6(z). Of the three terms in (5.13), the first describes
the upward wave generated by that Fourier component of a sheet of polarization whose
variation in the x-direction is characterized by a wavenumber nm. The electric field
generated by that Fourier component also has a nm dependence in the x-direction, and
the s vector component of the electric field is proportional to s Pr, while the p z
components are proportional to - Pr. The factor l / w r [W&/wl = 1/ cos (4) if \lE;
is real and nm < W\IE;l can be simply understood in that case as being due to the fact
that the area in the plane, corresponding to an area, A, normal to the Poynting vector of
the field generated, is A/ cos (6). Of course, even if & is real, for nm > W& the field
dexribed by the first term of (5.13) is evanescent; analogous to the situation of total
interna1 refiection, the spatial variation in the x-direction is characterized by a length
shorter than the wavelength of light in the medium, and energy cannot be radiatecl. The
second term in (5. El), describing the downward wave generated, can be understood in
the same generd way. The third term d e s d e s a contribution to the electric field present
ody at the polarization perturbation; it is independent of n" and zur (except insofar
as e ( z ) depends on 2). In the n = O, W = O limit (in which case the first two terms of
(5.13) vaaish), i t can be sirnply understood as the electrostatic field that would result
if the polarization sheet were assembled, in a medium of dielectric constant E , fiom two
sheets of charge pardel to the x-direction.
The form of the Green function (5.13) motivata our choice of a form for the Fourier
components of the electric field,
where we have used UT (2, q) , (2, q) to denote the upward/downward components of
the q-polarized light, where q can take on the values s or p. We use y (2) to denote
the field generated by the t h d term in the Green function (5.13). We have Mplicitly
defineci (2) , Rm (2) , and ( 2 ) for notational convenience. Using (5.15) in (5.12)
we End, for the s-polarized fields,
The value of the F ( z ) will be zero outside the shaded region, but their presence affects
the mlution of the fields within the layer.
In the next two sections we consider specific forms of the perturbation potential,
Pm (z), and determine the field evolution as a consequence of those perturbations.
5.3.1 Grating, No Nonlinearity
For a grating with no Kerr coefEicient, X ( 3 ) = 0, we have a perturbation polarization
where we have substituted the dielectric permittivity (5.5) and the electric field form (5.8)
into the expression for the perturbation polarization (5. Il), and where, for notational
simplicity, we have definecl E: = O. Letting N = (m + n), we find
Using this expression for Pr (2) in (5.17) we find a matrix equation for the y,
so that the source polarization is equal to
in which the r" fields have been elhinateci.
Since we are assiiming that the grating has no nonlinem susceptibilitsf, the two polar-
izations, s and @, do not interad within the grating. Using the fom of the pertubation
polarization (5.22) in the integral equations (5.16) we can determine a dinerential equa-
tion for the ewlution of the ppolarized fields in the ith layer in the presence of a grating:
the matrices
and the quantity
We have suppressed the i subscript in the labelling of the matrices. If q = s then, since
2 - s = 0. the (tm and matricies both vanish. The equations for s-polarized light are
simpler than those for ppolarized light because the static fields, F, do not interact with
the s-polarizat ion.
For a thin pating, the equations (5.23) can be integrated immediately, Erom which
one can define a transfer matrix. If the grating is thick, then it can be divided up into a
number of regions, each of which is sufliciently thin that the thin grating mat& can be
used.
5.3.2 No grating, Kerr Nonlinearity
In the absence of a grat hg, the relative dielettric permit tivity is ~i (2) = Fi. Since we are
assuming that the nonlineaxity acists in a humogenous slab, we use only the Fourier
component of the susceptibility, Xy) (2) = Xi. We further we asnune that the light is
completely s-polarized, so the electric field hes the simple furm
Wedefme h . I = p + r n - n , so
Pi (2.3) = ex= C (s& (2 ) (Fm- (i)) 9 (2) eiMKx& (5.29) M f " P
Using t his form for the plarization in (5.16) we find
au( M M i x,z2 - = iwi ui ( z ) + - - (x st (z) $ ( z ) (Spirn- (2)) a) , (5 .XI) bz 2w,M m p
Were we to include ppolarized light in our description, the results would be similar,
but a larger range of nonlinear effects would ernerge, involving coupling between the two
polarizations. We lem such effects for hirther investigation.
5.3.3 Transfer Matrices at the Interface
The work of the preceding sections has shown how the Fourier wmponents of the electric
field evolve as the field propagates within a slab or grating layer. To fully solve for the
electric field everywhere in the system, we need to determine how the fields in a given
layer are related to those in the neighbouring lam. h fact, we can use the usual Fresnel
c d c i e n t s based upon the average relative dielectnc permittivity in each layer, Éi [60].
h m Snell's law, we know that the x-component of the wave-vector is conserved
across the interface, so that we can treat our difhction orders independently. To go
fmm the hyer j = i - 1 to l a . i we fin& for ppalarizedlight [59][601
[ u q - -- 1 ( 1 rq"] 1
Dr t g @ ] ( p ) 1 Dj"
and for or-polarized light (591 [6O]
[z ] = - ( 1 TC (s) r z ( 3 ) ) 1 [: 1 , where tg ( s / p ) , r c (slp) are the fiesne1 coefficients for s- and ppolarized light r g p e c -
t ively (593 (603 :
Use of the field evolution equations (5.16), along with the -ne1 coefficients (5.33)
d o w us to detennine the emlution of the electric field throughout the e n t i . structure of
Figure 5.1. To detemine the exact fields everywhere in the structure, we need to impose
boundary condit ions. The usual boundary conditions are that t here are no upward-going
waves in the Nth laYer (UN = O, Up = O), and that in the 0" layer, the only downward
going wave is the incident beam (dg* = O, D~~ = O with 4 and fixed).
In the absence of any nonlinear layas in the stmcture, the fiesne1 coefncients and the
field emlution equations can be used to generate a m a t e that relates the fields in the
O* layer to those in the N* layer. Application of the boundary conditions dows us to
determine dl the fields in the oth a d ndh layer; and the use of t hese fields coupled with
the appropriat e transfer and propagation equations will determine the fields anywhere in
the system. If a nonhem layer exîsts within the structure, then the situation is more
complicated, because the &ect of the nonlinear layer is determined by the strength of
the field, but the strengt h of the field is detemineci by the nonlinear layer. The system
response in the presence of nonlinearXty can be determined using a numerid shooting
5.4 Numerical Simulations
5.4.1 Low Ioss reflector in the UV
In the following numerical sirnulat ions we use the transfer matri. formalism. We include
several diffraction orders, and treat the grating element by discretizing it into a number of
t hin grat ing elements, each of which is well describeci by a tramfer matrk. The number
of diffraction orders, and the size of the discretization of the grating were deterrnined by
varying both until the results conver& to a solution that conserved energy. Typicdy
3-5 modes were sufEcient for convergence. The size of the discretization was a sensitive
hnction of the thickness of the grating, and the material absorption in the system.
We start by cornparhg the reflection spectra of the GWS for s- and ppolarized Light
in the absence of nolinearity and materid absorption. We do this by fMng the incident
angle of the light, and sweeping through the incident wavelength. &al1 that the resonant
refiection of a GWS occurs because the incident iight couples to a waveguide mode. We
know, though, that the two polarkations have waveguide modes with different values of
/3 (651, so the resonent couphg condition will be different . We perforrn the simulations
with the following materid parameters: n2 = 1.5, na = 1.4, A = IlOnm, a grating
thickness of 15nm, and a guiding layer thicknes of 100nm. We assume that there is no
material absorption or nonlinearity, end that the grating layer alternates evenly between
n = 1 and n = 1.5. In Figure 3 we plot the reflection spectra of the GWS for s-polarid
(solid line) and ppolarized (dashed line) iight at normal incidence. As predicted, the
spectra for the two polarizations are distinct even at normal incidence. Thus, the GWS
is a polarization-sensitive device, which, when used in transmission, completeiy blocks
one polarization in a narrow kquency range.
We now examine the toleninfe of the GWS to absorption. We use the same material
parameters as above, but we d&e an intensity-loas parameter, a, measured in units of
cm-'. This loss can be used to determine a cornplex dielectric coeEcient which we use in
1 55 1 56
Wavelength (nm)
Figure 5.3: Reflectivity of the GWS as a hinction of incident wavelength for s-polarized
(solid line) and ppolarized (dashed line) light. The device is dearly polarization sensitive.
the transfer matrix fornalism. SpecScally, we first define a cornplex index of rehact ion,
n, = n, + iw, in terms of which the electric field in a homogeneous medium with n,
can be written E (r) = ~ ~ e ' - ~ l ~ e - " ~ ~ l ~ ; the intensity in the field, 1 (;), will be
proportional to I E ~ I * e-4*mz/.b. We then define cr = $ml&, so that l/a is a measure of
how far into the medium the field can penetrate before having its intensity attenuated by
a factor of l/e. We thus find that n, = n,+i (c&/4x), kom which we can determine the
cornplex dielectric coefficient E, = n:. We also vary the grating height h m 5nm to 50nm.
In Figure 5.4(a) we plot the refiection spectra for a lOnm grating (dashed line), and a
50nm grating (solid line) , for normal incidence. The spectrum associated with the 50nm
grating is six t h e s wider (0.345nm as opposed to 0.06nm), and is shiRed in fkquency
by 0.3nm. The shift in kequency occurs because the propagation constant, 0, associated
with the mveguide mode is dependent on the grating thickness. The resonance is wider
for the thicker grating because the coupling to the guided mode is stronger. In Figure
5.4(b) we plot the intensity of the fkld in the waveguide mode as a hinction of kequency,
Figure 5.4: (a) Re£iection spectrum for a GWS with a 1On.m gr& ing (dashed line) and
a 50nm grating (solid line). (b) Field enhancement in the waveguide mode for a GWS
wit h a l h m grating (dashed line) and a 50nm grating (solid line) .
for a lOnm and a 50nm grating. When the grating is lOnm thick it is more difficult for
light to couple into the system, but also more difEcult for light to couple out! This means
t hat the field in the mveguide mode c m build up. In Figure 5.4(b) the intensity in the
mode is almost nine times larger for the lOnm grating than the 50nm grat ing. In Figure
5.5 we present simulations of the refkction spectrum of the GWS for normal incidence,
using losses that vary from cr = O to a = 100cm-'. We vary the grating height between
5nm and 25nm. We see t hat the t hinner grating is far more susceptible to material loss.
With a 25nm grating, the device is 85% dar ion even when tr = 100mt-l, wherects for
a 5 n m grating the device is only 23% rdecting at the same loss parameter.
5.4.2 NonIinear switching in a GWS
Our swit ching geometry is shown in Figure 5.6. It involves a normdy incident CW signal
beam that is 100% reflected by a GWS; and an intense pump beam, propagating in the
+x direction in a mveguide mode, that is injectecl from the left hand side of the figure.
We assume that the guiding layer (layer 2) is Kerr nonlinear, with d2) = 10- 14cm2/W,
O O 50 100
Loss Coefficient (cm-1 )
Figure 5.5: Peak re£lectivity of the GWS as a hinction of the los coefficient ai in units
cm-'. Shown are the rdectivities for a GWS with grating heights of 5- (circles), 15nm
(squares) and 25nm (triangles).
where d2) is the familiar nonlinear index of refkaction [5]. This value of is consistent
wit h chalcogenide glas (661. We use materid parameters nz = 2.4, n3 = 1.5. We assume
that the grating aiternates between n = 2.4 and n = 2.21. We set the grating thickness to
5nm, and the guiding layer thicknes to 350nm. In Figure 5.7 we show the transmission
spectrum of the grating. The separation between the zero transmission wavelength and
the peak transmission wavelength is 4.2 x 10-'nm.
In the presence of a pump beam with intensity I . , the nonlinear index of refraction in
the guiding layer will be d2) I,. To switch between zero transmission and peak transmis-
sion of the signal b m , we require d 2 ) I , = 8 x Using the quoted value of d2) this
requires I, = 8 x 108 W / m 2 . SU& intensities are mdiiy available using, for example, a
mode-locked Nd:YAG laser with 10 - 200ps pulses [67], with the beam confinecl in the ;
direction by the waveguide layer, and in the y direction by narrow focusing.
There are several engineering issues invohd in the design of this h t c h . Most impor-
Kerr nonlinear guiding layer
Figure 5.6: Schematic of the nonlinear switch proposed in the tact. In the absence of
a pump beam, the weak signal beam is 100% reflected by the GWS. In the presence of
the pump barn, the nonlinear indsc of rehction shifts the signal beam off-resonance,
so that the signal is 90% trarumitted
523.85 523.865
Wavelength (nm)
Figure 5.7: Trmmision spectnun for the GWS used in the switching scheme in the
absence of the pump beam. In the presenœ of the pump bearn, the resonanoe will shift
to a slightly higher wavelength so that the region of O transmission wiU be switched to a
region of about 90% trammMon.
tant among them is width of the signal beam. h the absence of nnnlinPnrity? we require
our signal beam to sit in the region of the trançmission spectruxn with less that 5% t rans
mission. This corresponds to a wavelengt h spread A A/& = 1.5 x IO-', which, for a k e d
frequency, using (5.2), corresponds to a spread in # of A4 = 1.5 x IO-'. This, in tum,
corresponds roughly to a 5cm signal beam. By contrast, the 200ps pulses t hat we requk
are only about 3 n long (in the waveguiding medium) .To make the required width of the
signal beam naxrower, we would have to increase the strength of the resonance, which
will increase the intensity required for switchuig.
5.5 Conclusion
We have developed a transfer mat& technique to describe light incident on a GWS
with a grating of arbitrary thickness. We have derived differentid equations that can be
used to describe the GWS operation when one or more of the layers is Kerr nonlinear.
We have shown that a GWS with a thick grating can be used as a highly loss-tolerant,
narrow-band reflector in the W. We have also described a simple switching technique in
which a highly reflected signal beam becornes highly transmitted in the presence of an
intense pump beam.
In a fûture work we wiU extend the theory in this chapter to consider both X(2) and ~ ( ~ 1
nnnlinearities in a twdimensional textured kyer. The investigation of second harmonie
generation in such a structure is an area of active research [Ml, but while the materials
being used possess both a hi& X(2) and ~ ( ~ 1 , the analysis of the systems involved have
only considered the effets of x(*).
Chapter 6
Hamiltonian formulation for pulse
propagation equat ions in a periodic,
nonlinear medium
6.1 Introduction
The investigation of optical pulse propagation in nonlinear Kerr media ofken proceeds
via the slowly-varying envelope hinction apprcxïmation [3] [5][22], wherein the kequency
content of an optical pulse is mnsidered to be narrowly centred around a giwn carrier
bequency, c. This apprcximation d o w s one to separate the pulse dyna~nics, containeci
in the slowly-varying envelope huictions, hom the phase accumulation due to the carrier
frequency. When applied to a homogenous, isotropic medium, the slowly-varying a p
praJcimation has been used to derive the familiar nonlinear Schrodinger equation (NLSE)
[5] as the dynamical equations for the envelope function; when applied t o a periodic,
isotropic medium the approximation has been used to derive bot h a NLSE [7] [Ml and a
set of nonlinear coupled mode equations (CME) [3] [?]. In the presence of birefringence,
a set of coupled NLSEs have been derived for a homogeneous medium (221; for a pe-
riodie me di^^^^, both a set of coupled NUES and a set of nnnlins CMEa have been
derived [34]. The dynamics of these envelope-hinction equations have often been studied
by const mct ing a Hamiltonian formulation of the dynamical equations [9] [29] (311 [40] (681.
Fkom such a Hamihonian two conserveci quantities can be identifiai, one energy-iike, and
one momentum-like. But the Hamiltonian itself is not equal to the energy-like quantity,
leading to a certain conhision in the literature [69]. One would naively scpect that a
nonlinear optical system would have two conserved quantities - energy and momentum.
But since the IIamiltonian itself is also conserved, the optical system has Mree c o n s e d
quantities, and the interpretation of this third conservecl quantity presented some dif-
ficulties [31] [68] [69]. A correct understanding of these three conserveci quantities is the
goal of this chapter.
In Section 6.2 of this chapter we construct a auionical Hamiltonian formulation of
Maxwell's equations in a one dimensional, periodic medium with a Kerr nonlinearity,
using a dual field h t p r o p e d by Hillery and Mlodinow [70][71][72]. By monical we
mean that our Hamiltonian can be used to derive the exact equations of motion using
the canonical commutation relations, und that it is numerically qua1 to the energy of
the (nonlinear) electrornagnetic field. Although we here only consider classical fields we
formally replace the canonical Poisson brackets with the associated cornmutators, with
a view towards eventudy quantizing the theory. In Section 6.3 we specialize our for-
mulation to consider an effective field that varies slowly relative to the underlying Bloch
h c t i o n s of the periodic medium. We then generate a reduced canonical Hamiltonian
in terms of this effective field. The dynsmical equation grnrerned by this reduced Hamil-
tonian is the farniliar NLSE. In Section 6.4 we use the reduced Hamiltonian, which,
within our apprmimations, is conserved and equd to the energy, to identify two more
conserved quant it ies: the moment um, asçociated with space-translation symmetry; and
a conserved charge, associated with phase-translation symmetry. We demonstrate that
the 'energy' of the electromagnetic field quoted in many papers is, in fact, the c o d
charge associateci with phase-translation symmetry- If ane is studying envelope hctian_c:
that vary slowly in both t h e and space, then this conserveci charge is equd to the energy
of the electromagnetic field only to zeroth order.
Although we have concentrated on deriving a NLSE, our rnethod c m be used to
construct reduced canonical Hamihonians associateci with the non1inea.r coupled mode
equat ions in bot h isotropie and birefringent periodic media. F'urthermore, the dud field is
generalizable to two and t hree dimensions [?'Il, so it can Uely be used t o derive equations
in higher dimensional photonic band gap materialS. ùi Section 6.5 we discuss the use of
the dual field and contraçt it to other fie1ds used in the literature to derive the NLSE in
periodic media.
6.2 Canonical formulation of Maxwell's Equat ions
We begin with Maxwell's equations in a one dimensional, non-magnetic ( p = po) medium:
where
P is the hll polarizat ion, €0 is the permit t ivity of kee space, and p, is the permeability of
free space. To wnstruct a canonid formulation of these dynamical equations we begh
by introducing the dual field, A [72], to satisfy
ThedualfkldwillsemeastheriuinnicalorwrUfiP.ld-
density [72], 1
where we have introduced the conjugate momentum fieid,
The canonical equations of motion that follow from this Hamiltonian density are
which, using (6.3,6.4) are found to be precisely (6.1). Alternately, for an infinite medium
one can use the equal-tirne cornmutators [72]
with equations of motion [72]
where the associated Hamiltonian,
which is the energy in the electromagnetic field.
6.2J Linear, Periodic Medium
For a linear, perioàic medium,
where E (z) = E (z + d) characterizes the linear dielectric response of the medium, and
d is the periodicity of the lattice. We have introduced the notation H L to stress that
t his Hamilt onian generates the linear dynamics of the electromagnet ic field. Using the
equations of motion (6.6) we find a l i n s wave equation that A must satisfy:
To determine the Bloch bc t i ons of (6.12) we use the usual ansatz [74]
where C.C. stands for 'complec conjugate'. The units of O, (t) are definecl by the orthog-
onality condit ion given below (6.16). Substitut ion of (6.13) in the wave equation (6.12)
gives an equation for the Bloch functions, O,,
Because the operator in the equation is self-adjoint, it admits r d eigenvalues and or-
thogonal eigenfunctions.
Rom Bloch's theorem [74], we c m write our Bloch functions in tenns of a discrete
band index m, and a reduced wave number, k (-sr/d < k 5 *Id), 0, + Bd, with
0, (2) = (i) e". (6.15)
We note that Wmk = Wm(-k) , so we can choose our Bloch hc t ions such that Bd (2) =
amc-, , . The h k have the periodicity of the lat t ice, (2) = ( z + d) . We normalize
the Bloch functions via
ly,: (2) 8,,,tkt ( r ) d* = N&nmtbkkt*
where L is a nomalization length, and where we have chosen the DSIIllâliZ&tian CQIlStazLt
N = L/d . This choice of L and N means that o u wave numbers take on only discrete
values, and that the difference between taro adjacent mve numbers is 2?r/L. The Bloch
huictions also sat isfy
where = dOmr/dz; this follows by using (6.16) and (6.14). An aample of a dispersion
relation in this reduced-wavenurnber scheme is shown in Figure 6.1.
6.2.2 Periodic Medium with a Kerr nonlinearity
We now turn to a periodic, Kerr nonlineaz medium. At hequencies far below any rem-
nances in the medium, for our one dimensional geometry the constitutive re la t i~a takes
the form [5]
D = E ( I ) E + ~ o ~ ( ~ ) (2) E ~ , (6.18)
where we assume that the nonlinearity ooefficient, X(3) ( z ) is periodic with period d,
~ ( ~ 1 (i + d) = x ( ~ ) ( z ) . To constmct the Hamiltonian we k t invert (6.18) t o get
where we have assumed that X(3) ( z ) E3 < E (2) E. This assumption of a weak nonhear-
ity is judfhd on phyucal groundsr we d y want ta discuss third order nonlinear eects,
but if the asnimption of a weak X(3) were not valid, then we w o d d have no justification
for not including fifth order or higher nonlinear &mts in (6.18). Using (6.19) in (6.5) we
h d
where HL is defineci above (6.1 1) and HNL is the portion of the full Hamiltonian respon-
able for the nonlinearity in the dynamics of the electromagnetic field. The expression
- 1 O 1 Wavenumber (k/(x/d))
Figure 6.1: Dispersion relation for a one dimensional, linear, periodic medium in the
reduced wave number scheme. The wave numbers are nomahed to = ~ / d . The
frequencies are nomalized to wb, which is the centre frequency of the fmt photonic band
gap. Note that the introduction of the nomaikation length, L, means that the wave
numbers are discretized with adjacent wavenumbers separated by 2n/ L. The solid band
in the diagram represents the hequency content of a forward-propagating pulse whoçe
dyna,rn.ics are well-described by the NLSE derived in this paper. The kequencies are
confineci to a narrow range so that third- and higher-order dispersion can be ignorecl.
If the Erequency content is brought closer to the photonic band gap, then the range of
bequencies must be made more narrow, since near the gap the curvature of the dispersion
relation is quite high.
equations of motion for A within the a p p r h a t i o n of a weak nonlinearity (6. El),
It will be usehl to express H in terms of the classical d o g of the raishg and lowering
operators associated with the Bloch modes. To do so, we k t q a n d A (2, t ) and p (z, t)
in terms of the Bloch modes of the periodic medium. We let
The reality of p and A requires that p h = and 4 = A,,+kl, so we can express
both the complex quantities p d and & in terms of one mode amplitude, a,,,k (t),
Using the (t ), the expansion (6.22) becomes
We note that since A (2, t ) and p (2, t) are written in terms of Bloch huictions that are
n o m M in the region -LI2 < z < L/2, they will bemme periodic with period L.
This has no &et on the underlying physics, because we can dways consider the lllnit
where L -r m. However, it does mean that when evaluating the Hamiltonian (6.20) in
te= of the A(z , t) and p ( r , t ) given by (6.24) we must restrict the integration t o the
region - L / 2 < 2 < L/2; and when evduating the equal-tirne commutation relations (6.7)
between A (2, t) and p (2, t) we mu& restrict both z and 2 to be within & L/2. Adhering
to these restrictbq we Itid that usin= (6.24) in (UQ), and applying the o r h g d t y
relations (6.16) to the portion of the Hamiltonian that generates the linear dynamics,
the full Hamiltonian
H =
Adopt ing cornmut at ion relations
guarantees the commutation relations between A (r , t ) and p ( 2' , t) (6.7) for z and z' wit hin
the nomalization length. In terms of the mode amplitudes, the canonical equations of
motion (6.8) become [73],
which, using (6.25) for H, give
This equation (6.28) is equivalent bo (6.1) and (6.2) wibh (6.19).
6.3 Reduced Hamiltonian and the NLSE
In this section we re-cast our H d t o n i a n in a form more suitable to the study of pulse
propagation. We build effective fields, g, (2, t), as a Fourier superposition of the 43 and
assume that the effective fields are centred at a given wave number, k, which corresponds
to a kquency w,. The g,,, (2, t ) can be used to rewrite the Hamiltonian (6.25) without
any l o s of generality.
This effective fields a p p r d is most valuable when the kequency amtent a£ the field
is narrow relative to a central kquency, w s , that lies in band m = 7E wit h wave number - k. We assume that is far fkorn a photonic band gap, and that the frequency content
of the pulse is entùely containeci within band B. Because the hquency content of our
effective fields is narrowly centered around w-,, we can ~tpand a fiequency ~ ~ p + ~ ) in
a Taylor series which wil l involve the local group velocity and group velocity dispersion.
We use a smallness parameter, , to characterize the strength of the tenns in our
Harniltonian. We examine the situation where the terms that are related to the group
velocity of the pulse are O (q) , and the t a m s that are relateà to the Kerr nonlinearity
and the group velocity dispersion of the pulse are both O (q2) relative to the kgest terms
in the Hamiltonian. Higher-order nonlinear &?ects, and higher order dispersion are not
considered, because both are assumeci to bc O ( $ 3 ) . We denote the resulting Hamiltonian
the 'reduced Hamiltonian' since it is equal to the energy of the electromagnetic field to
O (v2). Although our mode1 formdy includes third harmonic generation, we ignore its
effets in the following. We are justined in doing so by physical considerations. We
have assumed that the underlying matMd is nondispersive, and while this may be valid
for Erequencies nea. w*, it WU likely not be valid for fiequency ranges extending ta
w = 3wmE; fbrthermore, the assumption of no absorption at w E Ls will like1y be in
error. We expect, on physical grounds, that in many cases the actual material dispersion
and absorption will make any b d d u p of the third harmonic unIikely, so that our modd
will be adequate.
We start by using the Q (t) to d&e an effective field, g, (2, t), that is centred -
around the wave number k = k,
where we have intmduced the àetuniag
~ = k - k ,
and the mode amplitudes
Ultimately we seek to describe the evolution of our field A (z , t ) , which we asnune is a
smoothly varying b c t i o n of z as we rnove fiom a point in a unit cell to the corresponding
point in a neighbouring unit c d The function g, (z, t) will be such a smoothly varying
hnction of t only if the (t) are srnmthly varying hinctions of k. To ensure that the
amk ( t ) vary smoothly in k, one must choose the Bloch functions to vary smoothly in k,
which can be done using a k p expansion [7] [49] about k.
Using (6.26), we find that the equal-the commutation relations for the g, (2, t ) are
for r and z' both in our normslization length L, where the Dirac delta function, 6 (z - 3)
in (6.32) strictly appears only in the L -r oc Iimit. By inverting (6.29) we find
Using (6.33), the Harniltonian (6.25) can be wntten in tenns of the gm (r , t ) .
In the foilowing we restrict ourselves to consideration of effective fields for which at
t = O the hequency content of the pulse is contained entirely in band rn = m, and narrowly centred around so that, replacing the a++,) (t) with the g,, ( t ) , and
restricting the surnmation in the Harniltonian (6.25) to one band, we h d a reduced
where the Ki are waue number detltnings Sinœ are a ~ e &dering ooly one band,
m = m, we drop the rn subscript in the remainder of the chapter. We stress that
the Hamiltonian (6.34) is still exactly equal to the energy in the system at t = O. At
later tirnes the nonlinear interaction will generate new fiequencies, but in the foilowing
we ignore third harmonic generation, as discussed, so that for reasonable propagation
times and pulse intensities, the new frequencies that are generated will still lie in band
rn = m, and the reduced Hamiltonian will st il1 represent the exact energy in the system.
Mhermore , we assume that at t = O o d y fomd-travelling waves are present, so that
at Iater times there will be no interaction with any backward-travelling mves.
We fird consider the linear portion of the reduced Hamiltonian (6.34). We expaad
the frequency w P+K) as
where w = q, J' = l)wF+,)!8K! ~~0 and w" = 8 2 ~ C * + , ) / a K 2 ] K = o . Substituthg th&
expression for w F+K) and the expression for the effective fields (6.29) into the reduced
Hamiitonian (6.34) we find that the portion of the reduced Hamiltonian asociated with
the linear dynarnics of the field is
where 7 is a smahess parameter used to diaracterize the relative strength of terms in
the Hamiltonian. We can quantify i ) by letting I ) = g (,Id), where i, is an appropriate
measure of the width of the pulse, and the factor g must satisfy
where I<. is an appropriate m e m e of the width of the pulse- As discussed, the third-
and higher-order dispersion terms are mnsidered to be O (q3). The values of 3 and 8'
will depend on the dispenion relation itself; a w&ty QE techniques exist to àekrmhe
the dispersion relation of a one-dimensional, periodic system [74].
Turning to the portion of the reduced Hamiltonian (6.34) t hat generates the nonlinear
dynamics of the fields, we fi& recall that we are dealing with a weak nonlinearity (see
note foilowing (6.19)). We quantify the weahess of this nonlinearity by asserting that
the ratio of the largest nonlinear term to the large& linear term is O (q2). Because we are
only keeping te- in HR to O (v2), t hk means that we can replace 9k+,) with GeiKz,
and the s m d error that it introduces will enter at the nact level in the perturbation.
Similady, we replace w (*+KI with Z. The value of q set above determines the strength
of the nonlinear term that can be accomodated by this theory. For a stronger nonlinear
term (either through a lazger X(3), or through a higher intensity in the pulse), more
complicated nonlinear effects must be indudeci. We find
An integral that wiil be important in Hz, is
The portion in the square brackets contains only periodic quantities, with period d, and
can be expanded as a Fourier series
00 (n) ei2n.irz/d
7 G - - n-dl
with
where the integration proceeds over the length, d, of one unit ceil. Using the expansion
(6.40) in the integral (6.39) we b d
The integralwill be zero upless ( K r - K2+&- i C , + h / d ) =O, but we haveprevi-
ously st ipulated t hat aU our detunings are ail < ~ / d , so IKim only has a value for n = 0.
This means
[,, = ~ ( 0 ) lL" dle'(Ki-K2+K3-I(.)z - LI2
(6.43)
and
h writing down we are only considering the integrals that will mise in (6.38) that
contain terms with two complex conjugates. Terms with zero or four complex conjugates
lead to third harmonic generation which, as discussed, is ignored here. Terrns with one
or three complex conjugates vanish for the following reason. The expansion (6.40) could -
be made because the eh portions of the Bloch Function (see (6.15)) cancel out. If1 on the
other hand, we consider terms where e i t ha one or three of the Bloch funct ions are oonju-
gated, then the expansion (6.40) would be multiplieci by a prefactor e*". The integral
correspondhg to (6.423 muid be nonzero oniy if (k-1 - h; + K3 - K4 + 2 m j d f = O
which, since the detunings are all srnall, c m never occur. Thus all integrals with one or
t hree conjugates vanish.
R o m (6.38), there are six ways to generate the term l g14 so that, including counting
considerations, we find
where, t o sirnphfjr the expressions, we have definecl the nonlinear c d c i e n t
Collecting our results (6.36) and (6.45) we h d a reduced Hamiltonian,
with a reduced Hamiltonian density
where we have suppressed the z,t dependence of g(z, t). The Heisenberg equations of
motion foilow from (6.27)'
so that, using the commutation relations (6.32), the differential equation that governs
the dynamics of the g (z, t) field is
6.3.1 Effective Fields and Envelope hinctions
In Our treatment of the NLSE, we have constructed an effective field as a Fourier superpo-
sition of the mode amplitudes in the Harniltonian. This diners &om previous derivations
of the NLSE in a periodic medium, in which the field of interest was a slowly-varying
nnvelolopo fiincrion that modulatecl a Bloch function at a &en vave number, k, and byld
index, [?] [34]. In this section we relate our effective fields to the envelope functions
that would emage if we used the dual field in the approach of previous derivations of
the NLSE.
We start by noting that an arbitrsry (2) can be written as
where the detunine are defined in (6.30), and where the value of the connections, -
YI(E+K) ' can be determinecl using 'k + p' t heory [7] [49]. Using this expansion for the
?&& w e h d -
,gd = Crckr ,g -e'Kz r n k - t ~ ) (6.52)
C
and, using (6.52) in (6.24)' we 6nd
CWTER 6-
where
fc (LI t , = { 5 7 f F + K t a m P + K ) (t) PZ} .
The j, ( 2 , t), which are envelope functions that modulate Bloch functions at k, are related
to the a, Ir+K C ) eiKz via the connections. Previous derivations of t h e NLSE would require
the field A ( z , t) to be separated as foIlows,
where If, (2, t ) 1 < 1 fm ( z , t ) 1, since the kequency content of the field is assumed to be
narrowly centred around w - ~ . The f* ( 2 , t ) are typicdy calleci 'cornpanion' terms,
while fm (2, t ) is cded the 'principal' term. Using a method presented elsewhere [7]
it can be shown that the principal term f'(z,t) obeys a dynamical equation analgous
to (6.50). However, we have verified that the Hamiltonian h m which the dynamical
equation of the fm ( z , t) can be derived is not equal to the energy in the electromagnetic
field to the required order in perturbation theory.
We can use ( 6 . 5 4 ) to relate
We start by recognizing that,
series
the envelope h c t ion fni (2, t ) to the effective field g (2, t) . - ..
using k p theory, the y$ can be expandeci as a Taylor
b i n g this, and recalling that since the fkequency content of the pulse is confined to the
band E, so that u ~ F + ~ ) (t) O if p # TE, we find
where for envelope h c t i o n s that vary slowly in space, the first term on the right hand
side of t his equat ion will be much large. than the other terms.
6.4 Conserved Quant ities of the Hamiltonian
In t his section we discuss the conmeci quantities associated with the reduced Hamil-
tonian system described by (6.47). We h t use (6.29) to exhibit the reduced Hamiltonian
(6.47) in terms of the Fourier modes of the effective fields,
We re-mite this reduced Hamiltonian in terms of the field and mnjugate momentum
variables, which in t his problem are real and can be writ ten in terms of the g and gt as
Substitut ing t hese int O (6.47) the reduced Hamilt onian becornes
with equations of motion
R o m (6.58) it is clear that the reduced Hamiltonian is invariant under the two infki-
i tesimal transfomat ions
where it is assumed that o and uK are infinitesimal quantities. If we convert back to real
space, we can ident* the k t t rdormat ion as expressing the phase invariance of our
r e d u d Hamiltoniaq and the second exp~essing the translatid invariaxa We note
that the system itself does not possess fdl translational symmetry. Howwer, at the levd
of the effective fields, the periodicity of the underlying stmcture has been capt ured in the
dispersion relation, and the reduced H d t onian does possess transiat ional invariance.
The periodicity of the system is accounted for by the Bloch huictions. In terma of the
real coordinates, #K and XK, the two infinitesimal transformations correspond to
where p is either a or vK. We use the invaxiance of the reduced Hamiltonian to construct
the conserved quant it ies associated wit h t hese infinitesimai transformations. Under eit her
transformation
where we have used the equations of motion (6.61), and where we set 6HR = O since the
reduced Hamiltonian is invariant. We End two c o m e d quant ities. The first , associated
wit h phase invariance, we c d the charge, Q. The second, associated wit h translational
invariance, we c d the momentum, P. In Fourier space, the two conserved quantities
have the value
Q = &jL/* 1 9 1 2 d i , - L/2
The reduced
with
p = -Fi- a J ~ / ~ (g$gt - &3&) d i . 2 c - L i 2
Hamiltonian can be written in terms of the conservecl charge,
where Ht is obviously also conserved.
To understand the nature of these consenrd quantities, we wnsida the dikent ia1
equation satisfied by the g-field (6.50). In the absence of group velocity, group velocity
dispersion or nonlinearity (Gt = 3' = a = O), the solution to the differential equation
(6.50) is g ( z , t ) = g ( ~ , O ) e - = ~ , where & is a constant. With 3 = O, it is clear that
P = O. h thermore , the phase accumulation, e-zt, is directly related to the increase in
tirne, t, so that the accumulation of time and phase are proportional. This means that
the reduced Hamiltonian, HR, is identical to the charge, Q, and we effectively have only
one conserved quantity. If we adow group veIocity, 3 # O, but keep 3' = a = O, then the
equation of motion (6.50) describes a pulse that propagates a t a speed 3, and does not
distort its shape. We can solve the equations of motion as g (2, t ) = g (r - Jt , O) e-'zt,
h m which it is clear that an increase in the thne variable is equivalent to a displacement
in space plus an increas in the phase. That is, only two of the three displacements in
tirne, space and phase, are independent and henœ required to M y describe the effective
field dynamics. Associated with this, one of the conserved quantities can be expresed in
terms of the other two, HR = Q + cP, which means that only two of the three conserved
quantities are independent. Findy, if we place no restrictions on the coefficients in (6.50)
then there are no simple solutions to the equation of motion, and we find that time,
space and phase displacements mu& be braebed independently; and the tbee eoneerved
quant ities, H R l Q , P are independent. The independence of t hese quantities is forced upon
us by the introduction of eit her the group velocity dispersion, or the Kerr nonlinearity,
so the linear Schrodinger equation (a = O), will also have three independent conserved
quantit ies associated wit h t ime, spaœ and phase invanance.
To connect wit h the lit erat ure on nonlinear optical pulse propagation, we write our g
field as the product of an effective field and a carrier £kequency,
then
and
which le& to
ifiw4 HI = j-L)iz (T ( r&rt - cc) - i ~ f 1 lazr12 -
Although H' correctly determines the dynamics of the r fields, it is clearly not equal to the
energy in the elect romagnet ic field. In previous discussions, the nonlinear S chrodinger
equation (6.69) has been derived directly from Maxweil's equations, and the r fields
- effective fields that vary slowly in space and thne - have been the primary fields of
interest 131153 [7] [22]. It was o k e d that the quantity H' could be used in a Hamiltonian
formulation, such that the correct equations of motion (6.69) were derived 1681 (691; but
the quant ity Hl waa clearly not equal to the energy of the system. Thus, Hl was wnsidered
to be a conserved quantity that could be used to generate the correct equations of motion,
while the consenred quantity Q = $ /r l a d z was cded the energy. In iight of the preceding
we see why this appmch makes some sense. When the r fields are dective fields that
v a q slowly in time we find that
su that, ta zeroth ader, the energy of the syséem is
with Q conservecl since it is the conserved charge. Nevertheles, the quantity Q is not the
energy of the system, as sometimes clairned [9] [a]. This can be seen, too, by not ing t hat
the assertion that Q is the energy is quivalent to stating that each mode has the same
frequency, J, regardes of its wavenumba. This is a reasonable apprmimation only if
the spread in Eequencies is small relative to c.
Although we have discussed these consaved quantities in the context of the NLSE,
the concepts behind this extend to o t h a nonlinear systems of interest. The coupled
NLSEk relevant to birekngent systems are often daived h m a Hamiltonian that is
unequal to the energy (91 [4O], as are the nonlinear coupled mode equations that describe
periodic, Ken media (both isotropie and birefringent [31]). These equations can dI be
derived using the rnethociobgy in this chapter; a Hamiitonian can be identifieci that is
both equal to the energy in the systern, and derives the correct equations of motion.
6.5 On the useof the Dual Field
The reduced Hamihonian (6.47), used in conjuct ion wit h the commutation relations
(6.32) and the equations of motion (6.49), gives a NLSE t hat describes pulse propagation
in a periodic medium, under the restriction point4 out at the beginning of Section 6.3. A
similar equation was derived in by de Sterke et d [7]. The advantage to the formulation in
this chapter is that the reduced Hamiltonian is presented in a form ready for quantization.
However, since both approaches <ived at the NLSE, it might be asked whether one could
constmct a canonical Hamiltonian using the fomalism of de Sterke e t al., rather than
introducing the dual field, A. In this section we point out ciifference between these two
approaches, and show the advantages of using the dual field.
with
where n (t) is the index of rekaction and Zo = ,/z is the impedance of kee space.
In a Kerr nonlinear, periodic medium, the field A was shown to satisfy
('12 { [ A + + A-]3} 1 1 -- 2 nZ ( r ) ût
where n' (2) = y. One can readily construct a quantity, EA (A+, A-), which is equd
to the energy in the EM field. However, the construction of a canonical Hamiltonian
in te- of the mode amplitudes of the A field appears impossible. To show this, we
fùst imagine that one has constructeci such a Hamiltonian, HA (ak), where ak are the
appropriate mode amplitudes of the A field, with canonical commutation relations. One
muid then apply the Heisenberg equat ions of mot ion to h d
The portion on the right hand side will be some complicated combination of modes, a k .
Unfortunately, the 2& term on the right-hand side of equation (6.70) rnakes clear that
the time derivatives of the modes, ar, must be e x p r d in terms of combinations of
modes and their t irne derivat ives, so that the equations of motion (6.n) cannot be exact.
Nevert heless, if the nonlinearity itseif is weak, then the nonlinear contribut ion to the
time derivative, aA/ût wiU also be weak. In the spirit of perturbation, then, we can
replace the time derivatives of the nonlinear portion of (6.70) by their linear value. This
strategy a b w s the construction of e Ehdtonien Eowulekion of M d s etpebions
in the presence of a weak nonlinearity. We have verifid that such a Harniltonian can,
indeed, be constructed, but we do not present the results here.
The Hamiltonian generated by the use of the A field is of as much practical value as
that generated by the use of the dual field, A. The advantage to the dual field formulation
is that once a form of the hinction U (D) is chosen, no further apprcaimations need to be
made. Thus, for the investigation of the forma1 pmperties of the Hamiltonian system the
dual field approach is more usehil, while for the cdculation of Bcperimentd quantities
either approach will work.
6.6 Conclusion
We have constructed a canonical Hamiltonian formulation for light in a nonlinear, pe-
riodic Kerr medium, with the appropriate fiequency content such that the NLSE is the
relevant equation of motion. To do so we have introduced a reduced Hamiltonian that
is equal to the energy in the electromagnetic field to the required order in perturbation
theory. Using the reduced Hamiltonian we investigated the conserved quantities of the
system. We found that the quantity oRen labellecl energy in the literature is more prop
erly the conserved charge associated with phasetranslation symmetry. In the context of
the slowly-varying apprmùmabion the ~FKZ indueed by t h mistake is 3m&, st least as
concerns the cornparison of theoretical results to qeriments. However, for the purposes
of canonical quantization, it is essential that the Hamiltonian be equal to the energy of
the system. To underscore its later use in quantization, we have presented Hamilton's
equations of motion in terms of canonical commutation relations, although we stress
that the resdts in this chapter are purely classical. In a later work we will return to the
quantization of the EM field in a periodic, Kerr-nonlinear medium.
Chapter 7
Conclusion
We have examinecl the linear and Kerr nonlinear properties of a variety of artificidy
structured materials (ASMs). In chapter two we exLunineci a one dimensional, periodic
medium that was birefringent and Kerr nonlinear. W e derived a set of coupled nonlinear
Sduodingn cquations (NLSEs) and a çet of nonlinear coupled mode equations (CS=)
t hat describe pulse propagation in the system for pulses that are slowly-varying relative to
a central carrier hequency. We based o u . derivat ion on the Bloch funct ions of the periodic
medium in the absence of nonlinearity, so that the results are valid even for media with
very strong index contrasts. In chapter three we ex&ed some of the consequences of
the derivations presented in chapter two. Experiments on fiber gratine, which can be
considered to have only one effective dimension, were p e r f o d by scientists at Lucent
technology, and interpreted using the work in this thesis. We showed the hequency
dependenœ of the effective birefkingence in a periodic syçtem. We also showed that
hi&-intensity puises in the fiber were susceptible to polarizat ion instability (or energy
exchange between the polarizations), and that the initial phase lag between the two
orthogonal polarizations had a large &ect on the amount of this polarization instability.
In chapter four we examhed a system in which two channel waveguides were coupled
by a series of periodically spaced microresonators. W e showed that two distinct types
of gaps opened in the dispersion relation of the structure, OIE type, the &agg gap,
opened as a consequence of a weak coupling between the channels that built up due to
the spacing of the resonators. The other type, the resonator gap, opened because of the
enormous coupling between the channels when the resonant frequency of the resonators
was matched. We showed, too, that the gmup veiocity dispersion e x p d œ d by a
pulse whose Erequency content is near a resonator gap is much lower than a pulse show
hequency content is near a Bragg gap. This, then, led us to observe that it should be
possible to =cite a gap soliton in a resonator gap with much less energy relative to a gap
soliton with the same parameters within a Bragg gap of the same width.
In chapter five we exarnined a grating-waveguide structure in which an incident plane
wave is coupied, via a grating, to a guideci mode of a layered medium. We used a Green
function technique to describe the system with a grating of arbitrary thidmess, and a
Kerr nonlinearity in the layered medium. We showed that the device could be used as a
narrow-band, las-tolerant reflector in the W, or as an d-optical switch.
In chapter six we constmcted a Harniltonian formulation for pulse propagation q u a -
tions in a one dimensional, periodic, Kerr nonlinear medium. Because our Hamiltonian
is eqiial to the energy in the system, we were able to clear up some issues about the
nature of the conserved quant it ies in nonlinear pulse propagation equat ions. S pecificdy,
we showed that the quantity previously considered to be the energy of the system was,
in fact, the conswed charge associated with phase invariance.
Bibliography
(11 John E. Heebner, Robert W. Boyd and Q-Han Park, submitted to Phys. Rev. Lett.
[2] D. W. Vernooy, V. S. Ilchenko, H. ibbuchi, E. W. Streed and H. 3. Khble, Opt.
Lett. 23, 247-249 (1998).
[3] Cg M. de Sterke and J. E. Sipe, in Pmgress in Optics, edited by E. Wolf (North-
Hoiland, Amsterdam, 1994), Vol. 33, pp. 203-260.
141 Sangjae Lee and Seng-Tiong Ho, Opt. Lett. 18. 962-9ç;4 [ 1993).
[5] G . P. Agrawal, Non-Linenr Filer Optics (Academic, San Diego, 1989).
[6] John D. Joannopoulos, Robert D. Meade and Joshua N . W i n , Photonic Crystals
(Princeton University Press, Princeton, 1995).
[7] C. M. de Sterke, D. G. Salinas, and J. E. Sipe, Phys. Rev. E 54, 1969 (1996).
[8] Ali Nayfeh, Perturbation Methods (John W i e y and Sons, New York, 1973).
[9] N. Akhrnediev and J. M. SoteCrespo, Phys. Rev. E 49, 5742 (1994).
[IO] Brent E. Little and Sai T. Chu, Optics and Photonics News, Nov. 2000, 2429.
(111 A. Sharon, D. Rosenblatt, and A. A. Fkiesem, Appl. Phys. Lett. 69, 4154-4156
(1996).
[i2] H. G. Winful, J. H. Marburger, and E. Garmire, AppL Phys. Lett. 35, 379 (1979).
[131 W. Chen aad D. L Mills, Phys Rsu. Letk 58,160 (1987).
[14] C. M. de Sterke and J. E. Sipe, Phys. Rev. A 38, 5149 (1988).
[15] S. Radic, N. George, and G. P. Agrawal, J. Opt. Soc. Am. B 12, 671 (1995).
[16] B. J. Eggleton, C. M. de Sterke, and R. E. Slusher, J. Opt. Soc. Am. B 16, 587
(1999).
[17] B. J. Eggleton et al., Phys. Rev. Lett. 76, 1627 (1996).
1181 D. Taverne. et al., Opt. Lett. 20, 246 (1998).
[19] V. MiPahi et al., Appl. Phys. Lett. 63, 1727 (1993).
[20] N. G. R. Broderick et al., Quantum ElecEronics and Laer Sczmce Conference, OSA
Technicd Digest (Optical Society of America, Washington DC, lgW), p. 9.
[21] T. Erdogan and V. Mimahi, J. Opt. Soc. Am. B 11, 2100 (1994).
[22] C. R. Menyuk, Opt. Lett. 12, 614 (1987).
[23] Suresh Pereira and J. E. Sipe, Opt. Exp. 3, 418 (1998).
[24] P. S. J. Russell, J. Mod. Opt. 38, 1599 (1991).
[2q R o m Bloch's theorem we &te 8, (k) = h, (k) eh, where the periodic depen-
dence of (k) and hi (k) on z is suppressed. In the k p method h, (K + bk) is expanded in t erms of the h,,,, ( K) . ClearIy t hese include a,ll the hi ( K) . But note t hat
if e,, (K) is an eigenvector of M, with eigenvalue w, (K) , t hen @kï (K) is an
eigenvector of n; l IV& with eigendue -wmj(K). Hence, since wd(- K) = w d ( K ) ,
e k i ( - K ) is an eigenvector of %' M, with eigenvalue -wmi(K) = w ~ ( K ) ,
and is characterized by a crystal momentum K. We define h,.,,,(K) according to
82 (- K) = hmi (K)etKZ . When we wish to acpand h,,,(k) we must use the full set
k ( K ) , where the range of pussihie Z i i indudes the m and E This wap not dane
explicitly in de S terke et al. (71
[26] C. M. de Sterke and J. E. Sipe, Phys. Rev. A 42,550 (1990).
[27] Y. Silberberg and Y. Barad, Opt. Lett. 20, 246 (lCl95).
(281 J. M. Soto-Crespo, N. Akhmediev, and A. Ankiewicz, J. Opt. Soc. Am. B 12, 1100
(1995).
124 Y. Chen, Phys. Rev. E 57, 3542 (1998).
[30] C. M. de Sterke, Phys. Rev. E 48,4136 (1993).
[31] W. Samir, S. J. Garth, and C. Pask, J. Opt. Soc. Am. B 11, 64 (1994).
[32] C. M. de Sterke, K. R. Jackson, and B. D. Robert, J. Opt. Soc. Am. B 8,403 (1991).
[33] G. k n z , B. J. Eggleton, and N. Litchinit~er, J. Opt. Soc. Au. B 15, 715 (1998).
[34] Suresh Pereira and J. E. Sipe, Phys. Rw. E 62, 5745 (2000).
[35] Nail N. Akhmediev and Adrian Ankiewicz, Solitow: Nmlinem Pulses and Bearns
(Chapman and Hd, London, 19%').
[36] H. G. Winful, Appl. Phys. Lett. 47, 213 (1985).
[37] H. G. Wuifid, Opt. Lett. 11, 3 (1986).
(381 R. E. Slusher et al., Opt. Lett. 25, 749 (2000).
[39] C. M. de Sterke, Opt. Express 3, 418 (1998).
1401 Nail Akhmediev, Alexander Buryak and J. M. SotuCrespo, Opt. Comm. 112, 278
(1994).
[41] S. T. Cundin et d, Phys. Rev. Lett. 82, 3988 (1999).
[43] Hans-Martin Keller, Swesh Pereira and J. E. Sipe, Opt. Cornm. 170, 35 (1999).
[44] J. E. Sipe, L. Poladian, and C. M. de Sterke, J. Opt. Soc. Am. A 11, 1307 (1994).
[a] B. E. Little, S. T. Chu, H. A. Haus, J. Foresi and J.-P. Laine, J. Lightwave Technol.
15, 99S1005, 1997.
[46] V. V. Vassiliev et d, Opt. Comm. 158, 305-311 (1998).
[47j John E. Heebner, Robert W. Boyd and QHan Park, submitted to JOSA.
[a] Suresh Pereira et al, in preparation.
[49] M. Lax, Symmetry principles in solid stote and molecvlar physics, (Wiley, New York,
1974).
i50j N. A. R. Bhat and J. E. Sipe, accepteci for publication in Phys. h v . E.
[51] D. R. Smith et aL, J. Opt. Soc. Am. B 10, 314 (1993).
[52] R. W. Bayd, Nonlinear Optzcs, (Academic, San Diego, 1991).
[53] C. M. de Sterke and J. E. Sipe, Opt. Lett. 14, 871-873 (1989).
(541 Curtis R Menyuk, Opt. Lett. 12, 614616 (1987).
[55] A. Sharon, D. Rosenblatt, and A. A.. fiesem, J. Opt. Soc. Am. A 14, 2985-2993
(1997).
[56] S. Glasberg, A. Sharon, D. Rosenblatt, A. A. Ftiesem, Opt. Comm. 145, 291-299
(1998).
[57] S. M. Notron, T. Erdogan, and G. M. Morris, J. Opt. Soc. h. A 14, 629-639
(19%').
[581 Scott -W N a r t ~ f ~ , G. MichaeL Mnms, and ' h a n Erdogan, J. Opt. Soc Am. A 15,
464-472 (1998).
[59] Max Born and Emil WoLf, Principles of Dptics (Pergamon Press, New York, 1980).
[60] J. E. Sipe, J. Opt . Soc. Am. B 4, 481-489 (1987).
[61] R. Reinisch, M. Neviere, P. Vincent and G. Vitrant, Opt. Cornm. 91,51-56 (1992).
[62] U. Peschel et al., J. Opt. Soc. Am. B 12, 1249-1258 (1995).
[63] V. Pacradouni et al., Phys. Rev. B 62, 4204-4207 (2000).
[64] A. R Cowan and Jeff Young, in preparation.
[65] Dietrich Marcuse, Theory of Dielechic Optical Waveguides (Academic Press, New
York, 1974).
[66] G. Lenz et al., Opt. Lett. 25, 254256 (2000).
(671 Swesh Pereira, J. E. Sipe and R. E. Slusher, in preparation.
[681 C . M . de Sterke and J. E. Sipe, Phys. Rev. A 39, 5163 (1989).
[69] C. Pask, D. R. Rowland and W. Samir, J. Opt. Soc. Am. B 15, 1871 (1998).
[70] M. HiUery and L. D. Mlodinow, Phys. Rev. A 30, 1860 (1984).
[71] P. D. Drummond, Phys. Rw. A 42,6&15 (1990).
[72] P. D. Drummond, in Recent Developments zn Quantum @tics, edited by R. kiguva,
(Plenum Press, New York, 1993). pp. 65-76.
[73] L. Mandel and E. Wolf, Opticai Cohemce and Qaanturn Optics (Cambridge Uni-
versity Press, Cambridge, 1995). .