Introduction To Nonlinear Optics

This page intentionally left blank
Introduction to Nonlinear Optics
Since the early days of nonlinear optics in the 1960s, the field has expanded dramatically,
and is now a vast and vibrant field with countless technological applications. Providing
a gentle introduction to the principles of the subject, this textbook is ideal for graduate
students starting their research in this burgeoning area.
After basic ideas have been outlined, the book offers a thorough analysis of second
harmonic generation and related second-order processes, before moving on to thirdorder effects, the nonlinear optics of short optical pulses, and coherent effects such as
electromagnetically-induced transparency. A simplified treatment of high harmonic generation is presented at the end. More advanced topics, such as the linear and nonlinear optics of
crystals, the tensor nature of the nonlinear coefficients, and their quantum mechanical representation, are confined to specialist chapters so that readers can focus on basic principles
before tackling these more difficult aspects of the subject.
Geoffrey New is Professor of Nonlinear Optics with the Quantum Optics and Laser Science
Group in the Blackett Laboratory, Imperial College London. A highly regarded laser physicist and a Fellow of the Optical Society of America, he has spent nearly half a century
researching nonlinear optics and laser physics.
Introduction to Nonlinear Optics

GEOFFREY NEW
Quantum Optics & Laser Science Group
Blackett Laboratory
Imperial College London
cambridge university press

Cambridge, New York, Melbourne, Madrid, Cape Town,
Singapore, So Paulo, Delhi, Tokyo, Mexico City
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521877015
G. New 2011
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without
the written permission of Cambridge University Press.
First published 2011
Printed in the United Kingdom at the University Press, Cambridge
A catalogue record for this publication is available from the British Library
Library of Congress Cataloguing in Publication data
New, Geoffrey (Geoffrey H. C.)
Introduction to Nonlinear Optics / Geoffrey New.
p. cm.
Includes bibliographical references and index.
ISBN 978-0-521-87701-5 (Hardback)
1. Nonlinear optics. I. Title.
QC446.2.N475 2011
535 .2dc22
2010051428
ISBN 978-0-521-87701-5 Hardback
Cambridge University Press has no responsibility for the persistence or
accuracy of URLs for external or third-party internet websites referred to in
this publication, and does not guarantee that any content on such websites is,
or will remain, accurate or appropriate.
Contents
Preface
Acknowledgements
1 Introduction
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
1.10
1.11
1.12
Nonlinearity in physics
The early history of nonlinear optics
Optical second harmonic generation
Phase matching
Symmetry considerations
Optical rectification
The Pockels effect
Sum frequency generation
Difference frequency generation and optical parametric amplification
Third-order processes
Theoretical foundations
The growth of nonlinear optics
2 Frequency mixing
2.1
2.2
2.3
2.4
Introduction and preview

Electromagnetic theory
Second harmonic generation
2.3.1 Coupled-wave equations
2.3.2 The low-depletion approximation
2.3.3 ManleyRowe relations for second harmonic generation
2.3.4 Interlude: third harmonic generation
2.3.5 The antenna picture
2.3.6 Phase relationships and quasi-phase matching
2.3.7 Additional features
General three-wave processes
2.4.1 Coupled-wave equations and the ManleyRowe relations
2.4.2 Modified field variables
2.4.3 Sum and difference frequency generation
2.4.4 Optical parametric amplification
2.4.5 Tuning characteristics
2.4.6 Optical parametric oscillators
page xi
xv
1
1
2
3
5
6
7
9
9
11
12
13
15
19
19
19
21
21
22
23
24
25
26
29
29
29
31
32
32
35
36
Contents
vi
2.5
Focused beams
2.5.1 Introduction
2.5.2 Gaussian beams and the Gouy phase
2.5.3 Harmonic generation with Gaussian beams
2.5.4 Focused beams: summary
3 Crystal optics
3.1
3.2
3.3
3.4
3.5
Preview
Crystal symmetry
Light propagation in anisotropic media
3.3.1 Maxwells equations
3.3.2 The constitutive relation
3.3.3 The index ellipsoid: ordinary and extraordinary waves
3.3.4 Index surfaces
3.3.5 Walk-off
Wave plates
Biaxial media
3.5.1 General features
3.5.2 Propagation in the principal planes: the optic axes
3.5.3 Positive and negative crystals
4 Nonlinear optics in crystals

4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8

The linear susceptibility
Structure of the nonlinear coefficients
4.3.1 Formal definition
4.3.2 Intrinsic permutation symmetry
4.3.3 Full permutation symmetry and the ManleyRowe relations
4.3.4 Contracted suffix notation
4.3.5 The Kleinmann symmetry condition
Crystal symmetry
Second harmonic generation in KDP
Second harmonic generation in LBO
The Pockels effect and the Pockels cell
Optical rectification
5 Third-order nonlinear processes

5.1
5.2
Introduction
Basic third-order processes
5.2.1 Definitions
5.2.2 Symmetry considerations
5.2.3 Third harmonic generation
5.2.4 The DC Kerr effect and the Kerr cell
5.2.5 The optical Kerr effect
37
37
38
41
43
45
45
45
48
48
51
52
54
55
56
57
57
58
60
62
62
62
63
63
64
65
65
66
67
68
70
72
74
76
76
77
77
78
79
80
81
Contents
vii
5.3
5.4
5.5
5.6
5.7
5.2.6 Intensity-dependent refractive index

5.2.7 Spatial solitons, self-focusing, and self-phase modulation
Stimulated Raman scattering
5.3.1 Raman coupling of laser and Stokes waves
5.3.2 Anti-Stokes generation and higher-order effects
5.3.3 Raman waves
Interaction of optical and acoustic waves
5.4.1 Physical basis
5.4.2 Diffraction of light by ultrasound
5.4.3 The acousto-optic dispersive filter
Stimulated Brillouin scattering
5.5.1 Introduction
5.5.2 The coupled-wave equations of stimulated Brillouin scattering
5.5.3 Approximations and solutions
Optical phase conjugation
5.6.1 Phase conjugate beams
5.6.2 Phase conjugation by four-wave mixing
5.6.3 Phase conjugation by stimulated Brillouin scattering
Supercontinuum generation
6 Dispersion and optical pulses

6.1
6.2
6.3
6.4
6.5
6.6

Dispersion in the phase and group velocities
Dispersion and chirping of a Gaussian pulse
6.3.1 Propagation analysis
6.3.2 Characteristic dispersion distance
6.3.3 The bandwidth theorem and the timebandwidth product
6.3.4 Chirp
Hyperbolic secant profiles
Compression using gratings and prisms
Optical pulse propagation
6.6.1 Pulse propagation equation
6.6.2 Transformation to local time
7 Nonlinear optics with pulses

7.1
7.2
7.3
7.4
7.5
7.6
7.7

Wave equation for short pulses
Self-phase modulation
Self-phase modulation with dispersion: optical solitons
Self-steepening and shock formation
Compression of self-phase modulated pulses
Group velocity dispersion in second harmonic generation
82
83
87
87
90
91
95
95
97
99
101
101
103
104
107
107
110
112
114
116
116
116
120
120
121
122
122
124
125
128
128
128
131
131
131
132
135
138
140
145
Contents
viii
7.8
Optical parametric chirped pulse amplification

7.8.1 The basic principle
7.8.2 Bandwidth enhancement with a chirped pump
7.8.3 Bandwidth enhancement with non-collinear beams
7.8.4 Optical parametric generation in three dimensions
7.9 Nonlinear optical diagnosis of short pulses
7.10 Mode-locked pulse trains and carrier envelope phase control
8 Some quantum mechanics

8.1
8.2
8.3
8.4
8.5
8.6
8.7
8.8
Introduction
Basic ideas
The Schrdinger equation
The density matrix
8.4.1 Definition of the density matrix
8.4.2 Properties of the density matrix
Introduction to perturbation theory
First-order perturbation theory
Second-order perturbation theory
8.7.1 Second harmonic generation
8.7.2 Discussion and interpretation
8.7.3 The second harmonic coefficient
8.7.4 Sum frequency generation
8.7.5 Optical rectification
Third-order perturbation theory
9 Resonant effects
9.1
9.2
9.3
9.4
9.5
9.6
9.7
Introduction
Resonance in the linear susceptibility
Raman resonances
Parametric and non-parametric processes
Coherent effects
9.5.1 The optical Bloch equations
9.5.2 The Bloch vector
Self-induced transparency
9.6.1 Preview
9.6.2 Motion of the Bloch vector: the Rabi frequency
9.6.3 Pulse area and the area theorem
9.6.4 Outline proof of the area theorem
9.6.5 The steady-state solution
Electromagnetically-induced transparency and slow light
148
148
150
151
152
155
158
162
162
163
163
165
165
167
167
168
171
171
172
175
176
178
178
178
179
184
184
184
186
188
189
189
191
192
192
193
194
197
197
198
Contents
ix
10 High harmonic generation
204
Appendix A Conventions and units
218
Appendix B Linear and nonlinear susceptibilities in the time and frequency domains
221
Appendix C Definition of the nonlinear coefficients
225
10.1 Introduction to high harmonic generation

10.2 Classical model
10.2.1 Electron trajectories
10.2.2 Recollision energy
10.2.3 High harmonic spectra
10.2.4 Phase matching
10.3 Attosecond pulses
10.4 Discrete harmonics in high harmonic generation
A1
A2
A3
B1
B2
B3
C1
C2
C3
Introduction
The nonlinear polarisation
Coupled-wave equations
Non-instantaneous response: the time and frequency domain equations

The linear term
The second-order nonlinear term
Introduction
Second-order processes
C2.1 Sum frequency generation and intrinsic permutation symmetry
C2.2 Second harmonic generation
C2.3 Note on non-collinear beams
C2.4 The general second-order process
C2.5 Optical rectification and the Pockels effect
Third-order processes
C3.1 Summary of third-order results
C3.2 Example: intensity-dependent refractive index
204
206
206
209
211
213
214
216
218
219
220
221
222
223
225
225
225
227
228
228
229
229
229
230
Appendix D Non-zero d elements in non-centrosymmetric crystals
232
Appendix E Real fields, complex fields, and the analytic signal
234
Appendix F Geometry of the grating pair
235
Appendix G The paraxial wave equation
236
Appendix H Useful formulae for numerical simulations
237
Appendix I Useful constants
239
Contents
Answers to problems
Further Reading
References
Index
240
248
249
253
Preface
I set out to write this book in the firm belief that a truly introductory text on nonlinear
optics was not only needed, but would also be quite easy to write. Over the years, I have
frequently been asked by new graduate students to recommend an introductory book on
nonlinear optics, but have found myself at a loss. There are of course a number of truly
excellent books on the subject Robert Boyds Nonlinear Optics, now in its 3rd edition
[1], is particularly noteworthy but none of them seems to me to provide the gentle lead-in
that the absolute beginner would appreciate.
In the event, I found it a lot harder to maintain an introductory flavour than I had expected.
I quickly discovered that there are aspects of the subject that are hard to write about at all
without going into depth. One of my aims at the outset was to cover as much of the subject
as possible without getting bogged down in crystallography, the tensor structure of the
nonlinear coefficients, and the massive perturbation theory formulae that result when one
tries to calculate the coefficients quantum mechanically. This at least I largely managed
to achieve in the final outcome. As far as possible, I have fenced off the difficult bits of
the subject, so that six of the ten chapters are virtually tensor-free. Fortunately, many key
aspects of nonlinear optics (the problem of phase matching, for example) can be understood
without knowing how to determine the size of the associated coefficient from first principles.
In the process of writing, I also discovered several areas of the subject that I had not
properly understood myself. It is of course always said that the best way to learn a subject
is to teach it, so perhaps I should not have been too surprised. There were also things that I
had simply forgotten. There was one senior moment when, after struggling with a thorny
problem for several hours, I found the answer in a book which cited a paper by Ward
and New!
As to the subject itself, I have sometimes teased research students by telling them that
nonlinear optics was all done in the 1960s. This statement is of course manifestly ridiculous
as it stands what have all those thousands of papers on nonlinear optics published in the
last 40 years been about then? On the other hand, the statement that all the basic principles
of nonlinear optics were established in the 1960s, does embody an element of truth. There
really are very few fundamental concepts underlying the nonlinear optics of today that
were not known by 1970. What has happened is not so much that new principles have
been established, but rather that the old principles have been exploited in new ways, in
new materials, in new combinations, in new environments, and on new time and distance
scales. Technology has advanced so that many possibilities that could only be dreamt of in
the 1960s can now be realised experimentally and, more than that, may even be intrinsic to
commercial products. This point is elaborated in Section 1.12.
xi
Preface
xii
The book itself has a modular rather than a serial structure. Consequently, some chapters
stand on their own feet, and most readers will probably not want to read the ten chapters
in sequence. Beginners should certainly start with Chapter 1, but a reader wanting to get
up to speed on crystal optics could read Chapter 3 on its own, while someone interested
in dispersion could jump in at Chapter 6. The purpose of both these chapters is to support
those that follow. On the other hand, readers who are not interested in crystallography
should avoid Chapters 3 and 4, while those who are not concerned with the tensor nature of
the nonlinear coefficients should certainly skip Chapter 4 and maybe the start of Chapter 5
too. And anyone who is allergic to quantum mechanics should be able to survive the entire
book apart from Chapter 8 and parts of Chapter 9. Of course, some people may be looking
for more detailed information, which they may be able to find in one of the nine appendices.
The following table shows what the reader can expect in each chapter; ticks in brackets
imply a small amount of material.
Chapter
Tensors
Crystallography
Quantum Mechanics
1
2
3
4
5
6
7
8
9
10
( )
( )
( )
( )
( )
As the table suggests, the chapters have a varied flavour. Some are fairly specialist
in nature, while others contain mostly bookwork material. Chapter 1 serves as a gentle
introduction. The polarisation is expanded in the traditional way as a power series in the
electric field (treated as a scalar) with expansion coefficients that are regarded as constants.
A range of potential nonlinear interactions is explored in this way. The problem of phase
mismatch emerges naturally from the discussion. Although in one sense the procedure
amounts to little more than cranking a mathematical handle, a surprisingly large number
of important nonlinear processes can be discovered in this way. At the end of the chapter,
attention is drawn to the shortcomings of the simple approach adopted, and the necessary
remedies are outlined. The chapter ends with a brief overview of the entire field.
Chapter 2 starts with a detailed analysis of second harmonic generation, based on
Maxwells equations and proceeding via the coupled-wave equations. The fields and polarisations are now complex numbers, but their vectorial nature is still not taken into account.
The discussion broadens out into sum and difference frequency generation, optical parametric amplification and optical parametric oscillators. The chapter ends with a treatment
of harmonic generation in focused beams.
In Chapter 3, real nonlinear media are encountered for the first time. The chapter focuses
on the linear optics of crystals, and contains a fairly detailed treatment of birefringence in
xiii
Preface
uniaxial crystals. The principle behind quarter- and half-wave plates is outlined. A section
on biaxial media is included at the end.
Chapter 4 deals with second-order nonlinear effects in crystals, and there is now no way to
avoid defining the nonlinear coefficients with all their tensor trappings. Though I have tried
hard to keep it simple (as much material as possible has been relegated to Appendix C), the
treatment that remains is still rather indigestible. Permutation symmetries are outlined, the
contracted suffix notation for the nonlinear coefficients is introduced, and the Kleinmann
symmetry condition is explained. The structure of the nonlinear coefficients for the three
most important crystal classes is highlighted, and examples are given to show how the
results are applied in some typical nonlinear interactions.
Chapter 5 is devoted to a selection of third-order nonlinear processes. Although the coverage is restricted to amorphous media, there are still places (especially early in the chapter)
where the tensor structure of the coefficients in unavoidable. The later parts of the chapter
deal with stimulated Raman scattering, acousto-optic interactions, and stimulated Brillouin
scattering. A section on nonlinear optical phase conjugation is included at the end.
Chapter 6 focuses on dispersion. Although basically a linear effect, dispersion plays a
crucial role in many nonlinear processes, and the material covered here therefore provides
essential background for Chapter 7. After a general introduction, the evolution of a Gaussian
pulse in a dispersive medium is analysed in detail, and this leads on naturally to a discussion
of chirping, and pulse compression. A hand-waving argument then enables a rudimentary
pulse propagation equation to be deduced, which provides an opportunity to introduce the
local time transformation.
Nonlinear optical interactions involving short optical pulses are treated in Chapter 7.
The choice of examples is somewhat arbitrary, but the selection covers quite a wide range,
and includes self-phase modulation, nonlinear pulse compression, optical solitons, second
harmonic generation in dispersive media, optical parametric chirped pulse amplification
(OPCPA), pulse diagnostics, and the phase stabilisation of few-cycle optical pulses.
Up to this point in the book, the strength of the nonlinear interactions has been represented by the value of the appropriate nonlinear coefficient or, more exactly, the set of
numbers contained in the non-zero tensor elements for the process in question. Chapter
8 focuses on the quantum mechanical origin of the coefficients. The perturbation theory
formulae that emerge are large and cumbersome and, to this extent, they may leave the
reader not much the wiser! However, every effort has been made to simplify things by
taking special cases, and by highlighting dominant terms in the equations, the structure of
which is relatively simple. The discussion continues in Chapter 9, which deals with resonant effects including self-induced transparency (SIT) and electromagnetically-induced
transparency (EIT).
No book on nonlinear optics would be complete without a mention of high harmonic
generation (HHG), and the quest for attosecond pulses. The treatment of these topics in
Chapter 10 is essentially classical, and certainly simple. So if the promise in the title of an
introductory treatment to the subject has not always been fulfilled in some of the intermediate
chapters, at least it is kept in the last chapter.
Some more detailed material is included in the appendices, the last of which contains
values of useful constants in nonlinear optics.
xiv
Preface
The varied nature of the material covered in the book has affected how the reference
list has been compiled. A comprehensive list of sources would be almost as long as the
book itself, so I have had to adopt a selective policy on which papers to refer to. Broadly
speaking, references are of three types: historical papers describing things done for the first
time, books and recent journal articles where an aspect of the subject is reviewed, and (in a
few cases) papers on topics of particular current interest. To find further sources, it is always
worth using the Internet. The information available there may not always be of impeccable
quality, but one will almost always be led to an authoritative source within a few minutes.
There are bound to be errors of one sort or another in any book of this type. I have of
course made every effort to avoid them, but I doubt if anyone has ever written a physics
book without getting a few things wrong. Colleagues who read the book in draft picked up
quite a few glitches, but I am sure there are more. If you find something that you think is
wrong, please e-mail me on [email protected], so that I can build up a corrections file.
Geoff New
Le Buisson de Cadouin
France
July 2010
Acknowledgements
Many people contributed in many different ways to this book, some during the actual
writing, and others on a much longer time-scale.
I want first of all to express my thanks and appreciation to two people who had a strong
influence on my early research. Professor John Ward introduced me to nonlinear optics in
Oxford in 1964, and has been a good friend ever since, while the late Professor Dan Bradley
was a dynamic leader and a constant source of inspiration and support over the following
15 years or so, first at Queens University Belfast, and later at Imperial College London.
Over the years, I have benefited greatly from interactions with colleagues, collaborators,
research associates and graduate students, through which my understanding of nonlinear optics has been expanded. I mention particularly Professor John Elgin, Dr. Graham
McDonald, Dr. Paul Kinsler, Dr. Ian Ross, Dr. Mick Shaw, Professor Wilson Sibbett, Professor Derryck Reid, Professor Sabino Chavez-Cerda, Dr Jesus Rogel-Salazar, Dr. Harris
Tsangaris, Dr. Phil Bates, Dr. Jonathan Tyrrell, and Dr. Sam Radnor.
I am especially grateful for the help and advice on specific aspects of the book given by Dr.
Luke Chipperfield, Professor Ian Walmsley, Professor Majid Ebrahim-Zadeh, Dr. Louise
Hirst, Professor Jon Marangos, Dr. Stefan Scheel, Dr. Stefan Buhmann, and Professor Mike
Damzen.
I owe a special debt of gratitude to three colleagues in the Blackett Laboratory at Imperial
College. Neal Powell in the audio-visual unit drew almost all the non-graphical figures, and
improved several of the others as well. He also made a number of late changes at very
short notice, and altogether could not have been more helpful. Professor John Tisch read
Chapter 10 in detail, made numerous constructive comments that greatly improved the text,
and drew my attention to a number of key references. He also provided two figures for
that chapter, and one for Chapter 7 as well. Lastly, Professor Richard Thompson somehow
found the time to read most of the book in draft, and even said that he enjoyed it! I adopted
almost all of his invaluable suggestions and am very grateful for his help and support.
Finally, I would like to acknowledge the Leverhulme Trust, which provided financial
support in the form of an Emeritus Fellowship during the period in which the book was
being written, and to thank Cambridge University Press for putting up with countless missed
deadlines for delivery of the final text.
xv
Introduction
1.1 Nonlinearity in physics

The response of real physical systems is never exactly proportional to stimulus, which is a
way of saying that we live in an inherently nonlinear world. The deviation from linearity
may be very slight, especially if the stimulus is weak; the assumption of linearity will then
be an excellent approximation and probably the only route to an analytical solution. But
that does not change the fact that linearity is an idealisation.
A simple example of the linear approximation occurs in elementary mechanics where
one assumes that restoring force is proportional to displacement from equilibrium (Hookes
law). This leads to the equation of motion
mx = s1 x
(1.1)
where m is the mass and s1 is the Hookes law constant (the stiffness of the system). Equation
(1.1) has the simple harmonic motion solution
(1.2)
x = A cos{0 t + }
where 0 = s1 /m is the angular frequency of oscillation, and A and are fixed by the
initial conditions. Equation (1.1) is usually an excellent approximation when the amplitude
A of oscillations is small. But, since real systems are never perfectly linear, their oscillations
will never be precisely (co)sinusoidal, and will contain harmonics of 0 .
Nonlinearity can be incorporated into Eq. (1.1) by including additional terms on the righthand side to represent the fact that the restoring force is no longer linear in the displacement.
For example, it might be appropriate to write
mx = s1 x + s2 x 2 = (s1 s2 x)x
(1.3)
where the second form highlights the fact that the nonlinear term makes the stiffness
dependent on amplitude. Notice that the inclusion of a term in x 2 makes the stiffness
asymmetrical in the displacement; if s2 > 0, the net stiffness is lower for positive x and
higher for negative x.
Since Eq. (1.3) has no analytical solution, a numerical solution is generally called for.
However, the role of the extra term in introducing harmonics can be appreciated by regarding it as a perturbation, with Eq. (1.2) as the zeroth-order solution. To the first order of
approximation, the system will be subject to the additional force term
1
s2 x 2 = s2 A2 cos2 {0 t + } = 12 s2 A2 [1 + cos 2{0 t + }],
(1.4)
Introduction
and this will lead (through Eq. 1.3) to a second harmonic component in the motion. In
principle, the process can be repeated to higher orders of approximation, although the terms
quickly become negligible if the nonlinearity is weak.
In many situations in life, one seeks to minimise the effects of nonlinearity. Manufacturers of audio equipment are for example keen to advertise the lowest possible figures for
harmonic distortion. But in other circumstances, nonlinearity can be put to good use, and
this book is about how it can be exploited to spectacular effect in optical physics.
1.2 The early history of nonlinear optics

In optics, one is interested in the response of atoms and molecules to applied electromagnetic
(EM) fields. The interaction of light and matter is of course governed by the Schrdinger
equation, which is linear in the wave function but nonlinear in the response of the wave
function to perturbations. Despite this, optics proceeded quite successfully for many years
on the assumption that the response of optical materials was linear in the applied electric
field E. If P is the polarisation of the medium (i.e. the dipole moment per unit volume),1
one writes
P = 0 (1) E
(1.5)
where (1) is the linear susceptibility. So if E = A cos t, the consequence is that P =

0 (1) A cos t. It also follows from Eq. (1.5) that the electric displacement is
D = 0 E + P = 0 (1 + (1) )E = 0 E
(1.6)
dielectric constant. As we will see later, is the square

where = 1 + (1) is the relative

of the refractive index, n = 1 + (1) .
Equation (1.5) served as a good approximation for so long because the electric field
strengths that scientists were able to deploy in those early years were far weaker than the
fields inside atoms and molecules; the perturbations were therefore very small. It was not
until the 1870s that the Rev. John Kerr, a lecturer at the Free Church Training College in
Glasgow, UK, demonstrated that the refractive index of a number of solids and liquids is
slightly changed by the application of a strong DC field [2]. This phenomenon, now known
as the DC Kerr effect,2 was the first nonlinear optical effect to be observed.
Two decades later, in the 1890s, Friedrich Pockels at the University of Gttingen studied
a related process known today as the Pockels effect [3]. The Kerr effect and the Pockels
effect differ in two respects. In the Kerr effect, the refractive index change is proportional
to the square of the applied DC field, whereas in the Pockels effect, the change is directly
1 The word polarisation has two meanings in optics. It refers (as here) to the dipole moment per unit volume (in
coulombs m2 ), and would be represented by the vector P in a more rigorous treatment. But the word is also
used to describe the polarisation of light where it refers to the direction of the fields in the transverse EM wave.
The direction (or plane) of polarisation is normally taken to be that of the electric field in the wave.
2 Sometimes it is just called the Kerr effect, but it must be distinguished from the optical (or AC) Kerr effect.
1.3 Optical second harmonic generation
proportional to the field.3 Secondly, whereas the Kerr effect is observable in liquids and
amorphous solids, the Pockels effect occurs only in crystalline materials that lack a centre
of symmetry. This vital distinction and the reason behind it will be discussed in detail later.
1.3 Optical second harmonic generation

There now followed a long gap in the history of nonlinear optics. Further progress had to wait
for a source of strong optical frequency fields to become available, in other words for the
invention of the laser in 1960. With the arrival in the laboratory of the ruby laser, nonlinear
optics underwent a second birth, and it has been flourishing ever since. Indeed, since lasers
themselves are inherently nonlinear devices, one could even argue that laser physics is itself
a compartment within the wider field of nonlinear optics. Traditionally, however, the field
of nonlinear optics is taken to exclude lasers themselves, which is fortunate insofar as this
book would otherwise be much longer.
The first nonlinear optics experiment of the laser era was performed in 1961 by a team
led by the late Peter Franken at the University of Michigan in Ann Arbor [4]. As shown
in Fig. 1.1, a ruby laser was focused into a slab of crystalline quartz to discover if the
nonlinear response of the medium to the intense optical frequency radiation at 694.3 nm
was strong enough to create a detectable second harmonic component at a wavelength of
347.15 nm.4
The way to think about this experiment is diagrammed in Fig. 1.2. Consider the response
of the electrons in the quartz to the stimulus of the optical frequency electric field of the
laser beam written as E = A cos t. The displacement of the electrons creates a dipole
moment p per atom,5 or P = Np per unit volume where N is the atomic number density.
RUBY LASER
QUARTZ
CRYSTAL
L
L
PHOTOGRAPHIC
PLATE
Fig. 1.1
Schematic diagram of the rst second harmonic generation experiment by Peter Frankens group at the University of
Michigan in 1961 [4].
3 The Pockels effect is also known as the electro-optic effect, or sometimes (confusingly) as the linear electro-optic
effect because the index change is linearly proportional to the DC field.
4 The vacuum wavelength of the second harmonic component is of course half that of the fundamental.
5 One could also write p = E where is the atomic polarisibility, in which case (1) = N .
0
Introduction
c
b
a
b
a
c
Fig. 1.2
Linear and nonlinear response of polarisation P to applied electric eld E: (a) linear case (dotted), (b) quadratic
nonlinearity, (c) cubic nonlinearity.
We allow for the possibility of a nonlinear response by writing
P = 0 ( (1) E + (2) E 2 + ).
(1.7)
For (2) > 0, the dependence of P on E is represented by curve b in Fig. 1.2, while in
the absence of the nonlinear term, Eq. (1.7) reverts to Eq. (1.5), and the linear relationship
between E and P represented by the dotted straight line a in the figure is recovered.
If E = A cos t is substituted into Eq. (1.7), the polarisation now reads
P = 0 ( (1) A cos t + 12 (2) A2 (1 + cos 2t)).

linear term
nonlinear term
(1.8)
The analogy between Eqs. (1.7)(1.8) and Eqs. (1.3)(1.4) is obvious. Figure 1.3 shows
graphs of the total polarisation (in bold) and the linear term on its own (bold dotted); notice
how the waveform is stretched on positive half-cycles and flattened on negative half-cycles,
in accordance with curve b of Fig. 1.2. The solid grey line in Fig. 1.3 representing the
nonlinear term in Eq. (1.8) clearly contains a DC offset (dotted grey), the significance
of which is discussed in Section 1.6, and a second harmonic component, which is what
Franken and his team were looking for in 1961. A useful picture is to regard the second
harmonic polarisation of the nonlinear medium as an optical frequency antenna. Just as
radio and TV signals are broadcast by accelerating charges in the transmitter aerial, so the
oscillating nonlinear polarisation in Frankens quartz crystal should radiate energy at twice
the frequency of the incident laser field. The key question in 1961 was simply: will the
radiation from the second harmonic antenna be strong enough for photons at 347.15 nm to
be observed experimentally?
The University of Michigan experiment was a huge success to the extent that second
harmonic generation (SHG) was detected. The down side was that the harmonic intensity
was extremely weak, so weak in fact that the photographic plate reproduced in the 15
August 1961 issue of Physical Review Letters [4] appears to be totally blank! In fact, the
1.4 Phase matching
time
Fig. 1.3
Response (bold line) of a system exhibiting a quadratic nonlinearity to a (co)sinusoidal stimulus (dotted line). The
difference between the two (grey line) contains both second harmonic and DC frequency components.
energy conversion efficiency from fundamental to harmonic was about 1 part in 108 , a
clear demonstration that SHG was a real effect although, at this minuscule efficiency, it was
clearly a curiosity rather than a practical means of generating ultraviolet (UV) light.
1.4 Phase matching

Why was the efficiency so small? The reason is easy to understand as soon as one includes
the spatial dependence of the fields in the equations. Assuming that the ruby laser field is a
plane wave propagating in the z-direction, we write
E1 = A1 cos{t k1 z}
(1.9)
where the angular wave number of the fundamental beam is k1 = n1 /c, and n1 is the
refractive index.6 The second harmonic term in the polarisation is therefore
P2 = 12 0 (2) A21 cos{2t 2k1 z}
(1.10)
which seems fine at first sight. But now compare the space-time dependence in Eq. (1.10)
with that of a freely propagating field at 2 namely
E2 = A2 cos{2t k2 z}
(1.11)
where k2 = n2 2/c by analogy with k1 . Notice that the arguments of the cosines in
Eqs (1.10) and (1.11) are different unless 2k1 = k2 , which is only true if the refractive
indices n1 and n2 are the same. But dispersion ensures that the indices are usually not the
6
Here and throughout the book, the default space-time dependence of an optical field is (t kz), where
the parameter k is called the angular wave number, by analogy with angular frequency = 2 . However,
many authors use (kz t) instead of (t kz). The choice is immaterial under a cosine but, with complex
exponentials, it leads to a difference in sign that permeates into many subsequent equations. See Appendix A
for more discussion.
Introduction
same. The severity of the mismatch is easily quantified by asking over what distance the
cosine terms in Eqs (1.10) and (1.11) get radians out of step. This distance, known as the
coherence length for the second harmonic process, is easily shown to be7
Lcoh =
=
|k|
4 |n2 n1 |
(1.12)
where k = k2 2k1 is called the mismatch parameter. For typical optical materials,
Lcoh is a few tens of microns8 (m) after which, as we will show in the next chapter, the
SHG process goes into reverse, and energy is converted from the harmonic back into the
fundamental wave. Hence, only a few microns of the 1 mm quartz sample contributed to
the signal in the first SHG experiment, and the rest was redundant.
The phase-matching problem was soon solved. Within a few months of the University
of Michigan experiment, researchers at the Ford Motor Companys laboratories, around
30 miles from Ann Arbor in Dearborn, Michigan, had exploited birefringence in a KDP
crystal9 to keep the and 2 waves in step with each other [5]. They did this by finding
a particular direction of propagation in KDP for which the refractive index of the ordinary
wave at was the same as that of the extraordinary wave at 2. The process is called phase
matching or sometimes birefringent phase matching (BPM) to distinguish it from other
methods of achieving the same end; see Chapter 2. Under phase-matched conditions, the
SHG conversion efficiency jumped to tens of percent with careful experimental adjustment,
and from that moment on, nonlinear optics stopped being a curiosity and became a practical
proposition.
Phase matching is of vital importance in many nonlinear processes. The key principle
to grasp is that the direction of energy flow between fundamental and harmonic waves is
determined by the relative phase between the nonlinear polarisation and the harmonic field.
If the two can be held in step by making the fundamental and harmonic refractive indices
the same, the energy will keep flowing in the same direction over a long distance. Otherwise
it will cycle backwards and forwards between the fundamental and the harmonic, reversing
direction more frequently the more severe the phase mismatch.
1.5 Symmetry considerations

Before moving on, it is worth considering some other key features of the SHG process.
First of all, notice that the quadratic term in Eq. (1.6), represented by curve b in Fig. 1.2,
implies that the medium responds differently according to the direction of the electric field.
A positive field creates a slightly greater (positive) response than that of curve a, whereas a
negative field produces a slightly smaller response than in the linear case. This implies that
7 Sadly, there is no generally agreed definition of coherence length in nonlinear optics, and definitions with
1, 2, , and 2 in the numerator of Eq. (1.12) can be found in different textbooks. The definition in Eq. (1.12)
is the one used almost universally by researchers in the 1960s, and I see no reason to change it.
8 Micron is still widely used for micrometre in colloquial usage.
9 KDP stands for potassium dihydrogen phosphate.
1.6 Optical rectication
(a)
Fig. 1.4
(b)
(a) Centrosymmetric pattern and (b) non-centrosymmetric pattern.

SHG (and all other nonlinear processes that depend on (2) ) can occur only in media that
are structurally different in one direction from the opposite direction. Media of this kind
are variously said to lack inversion symmetry, to lack a centre of symmetry, or to be noncentrosymmetric. In simple terms, one could say that they have an inherent one-wayness, a
property that only crystalline materials can possess. Fortunately, many non-centrosymmetric
media also exhibit double refraction (birefringence), and so are candidates for birefringent
phase matching.
A more detailed discussion of inversion symmetry is given in Chapter 3, but the simple
schematic pictures shown in Fig. 1.4 may be helpful at this stage. Imagine you are looking
into a crystal along the line of the laser beam. The structure shown on the left possesses
inversion symmetry. It can be flipped horizontally (x x) or vertically (y y) and
the pattern is unchanged; the centre of symmetry at the mid-point is marked with a dot. It is
therefore impossible in principle for this medium to exhibit the asymmetrical characteristics
shown in Fig. 1.3. The structure on the right is different because, in this case, a flip in either
direction results in a different pattern. So left is different from right, up is different from
down, and the response in Fig. 1.3 is no longer forbidden.
A simple mathematical proof of the principle runs as follows [6]. In one dimension,
consider the term Px = 0 (n) Exn . Under inversion symmetry, Px must change sign if Ex
changes sign. But, when n is even, Exn is unchanged if Ex Ex , and it follows that
(2) , (4) , etc. must be zero in this case. No such restriction applies for odd n, so processes
involving (3) for example (see Section 1.10) can occur in centrosymmetric media.

After the successful demonstration of second harmonic generation, the nonlinear optics
bandwagon began to roll, and a multitude of other nonlinear processes were discovered in a
gold rush period in the mid-1960s. One such process was optical rectification, which has
already appeared in the polarisation in Eq. (1.8) above [7]. This is probably the simplest
nonlinear optical process both to visualise and to observe. To detect optical rectification,
all one has to do is to pass a laser beam through a non-centrosymmetric crystal located
Introduction
CRYSTAL
LASER
Fig. 1.5
Schematic diagram of optical rectication. Positive and negative charges in the nonlinear medium are displaced,
causing a potential difference between the plates and current to ow in an external circuit.
Fig. 1.6
Optical rectication signal (top trace) and laser monitor (lower trace).
between a pair of capacitor plates, as shown in Fig. 1.5. As suggested in the figure, the
asymmetric response of the material to the field of the laser displaces the centre of gravity
of the positive and negative charges in the medium, creating a DC polarisation in the
medium. Dipoles oscillating at the optical frequency are of course created in all materials,
and these are represented by the linear polarisation of Eq. (1.5). What is shown in Fig. 1.5
is rather the DC offset of Fig. 1.3, and its effect is to induce a potential difference between
the plates, which allows a rectified component of the optical frequency field to be detected;
see Problem 1.2.
An example of an optical rectification (OR) signal is shown in the dual-beam oscilloscope
record of Fig. 1.6, where the upper and lower traces are the OR and the laser monitor
signals, respectively. In fact, so accurately does the OR signal track the laser intensity that
it is impossible to tell them apart from the record itself.10 Since the OR process also leaves
the laser field essentially unchanged, it was considered as a possible basis for laser power
measurement in the early days of the subject. But other technologies prevailed in the end.
Notice that the phase-matching issue does not arise in the case of optical rectification,
because the DC term in the polarisation (analogous to the second harmonic contribution in
10 The fact that one trace is the negative of the other is a trivial result of the polarity of the electrical connections.
1.8 Sum frequency generation
Eq. 1.10) is
Pdc = 12 0 (2) A21
(1.13)
which has no space-time dependence.
1.7 The Pockels effect

The Pockels effect [3] relates to the change in the refractive index of a non-centrosymmetric
medium when a DC electric field is applied. The effect can be discovered within the
framework of Eq. (1.7) if a DC term is included in Eq. (1.9) so that
E = A0 + A1 cos{t k1 z}.
(1.14)
Substituting this form into Eq. (1.6) produces a new nonlinear term in the polarisation with
the same space-time dependence as the linear polarisation. When the two are combined, the
net contribution at frequency is
P = 0 ( (1) + 2 (2) A0 )A1 cos{t k1 z} +
(1.15)
The new term is proportional to the DC field A0 , and is in phase with the optical frequency
field provided (2) is real. No phase-matching considerations arise since, as for optical rectification, phase matching is automatically ensured. It is clear from Eq. (1.15) that the linear
(2)
susceptibility has been
modified by 2 A0 , and the refractive index has correspondingly
(1)
changed from n = 1 + to n = 1 + (1) + 2 (2) A0 .
The Pockels effect allows the polarisation properties of light to be controlled electrically
and, in combination with a polarising beam splitter, is routinely used to create an optical
switch. A more detailed discussion of a Pockels cell can be found in Chapter 4.
1.8 Sum frequency generation

The catalogue of second-order nonlinear processes can be extended further by considering
the process of sum frequency generation (SFG), a possibility that emerges if the incident
electric field contains two different frequency components namely
E = A1 cos{1 t k 1 .r} + A2 cos{2 t k2 .r}.
(1.16)
The angular wave numbers have now been written as vectors to allow for the possibility that
the waves at 1 and 2 are non-collinear (i.e. travelling in different directions).11 Inserting
Eq. (1.16) into the polarisation expansion of Eq. (1.7) yields a term of the form
P = 0 ( + (2) A1 A2 cos{3 t (k1 + k2 ).r} + )
(1.17)
11 When written as a vector, the angular wave number is variously called the wave vector, the propagation vector
or simply the k-vector.
Introduction
10
k1
k2
k3
Fig. 1.7
Wave vector triangle for phase-matching non-collinear beams.

where 3 = 1 + 2 . It must be stressed that the subscripts 13 are now merely labels,
their only numerical significance being that 3 is the highest frequency of the set. In this
notation, second harmonic generation is the special case where 1 = 2 = 3 /2.
If the polarisation wave of Eq. (1.17) is to match the space-time dependence of a freely
propagating field at the sum frequency 3 and space-time dependence cos{3 t k3 .r}, the
phase-matching condition to be satisfied is
k1 + k2 = k3
(1.18)
which is represented by the vector triangle of Fig. 1.7. For collinear beams, this equation
becomes
n1 1 + n2 2 = n3 3
(1.19)
which is more complicated than for SHG where phase matching simply requires that the
fundamental and harmonic refractive indices are the same. The term phase matchingis to be
preferred to index matching for this reason. In media that exhibit normal dispersion,12 the
right-hand side of Eq. (1.19) will always be larger than the left-hand side (see Problem 1.2).
This means that adjusting the angle between k1 and k2 is not sufficient on its own to fix the
phase-mismatch problem, although it may still be useful for adjusting the phase-matching
condition. An example of the use of non-collinear beams for increasing the bandwidth of a
nonlinear interaction can be found in Chapter 7.
At the photon level, one can regard the SFG process as the annihilation of photons 1
and 2 , and the creation of a photon 3 , so that
1 + 2 3
(1.20)
which is a statement of conservation of energy. This equation has been written to suggest
that the process may go into reverse, which will happen for certain phase relationships
between the three fields.
Notice that multiplying all terms in Eq. (1.18) by yields
k1 + k2 = k3 .
(1.21)
Since momentum is p = k, this equation demonstrates that the phase-matching condition for the interacting waves is equivalent to the momentum-matching condition for the
interacting photons.
12 Normal dispersion means that the refractive index rises with frequency (or falls with wavelength).
11
1.9 Difference frequency generation and optical parametric amplication
1.9 Difference frequency generation and optical parametric

amplication
We now consider a further possibility. Suppose that the relationship 1 + 2 = 3 still
applies, but that the two waves entering the nonlinear medium are 2 and 3 (> 2 ) rather
than 1 and 2 . In this case, the relevant term in the polarisation is
P = 0 ( + (2) A3 A2 cos{1 t (k3 k2 ).r} + ).
(1.22)
This represents the process known as difference frequency generation (DFG) in which 1 is
generated through the interaction of 3 and 2 .13 Equations (1.18)(1.21) are all unchanged
in the DFG case, although it perhaps makes more sense to write Eq. (1.20) the other way
round, namely
3 1 + 2 .
(1.23)
This equation highlights an important feature of DFG. Since the difference frequency
component grows as photons at 3 divide into photons at 1 and 2 , it follows that the
other input beam at 2 grows at the same time, indeed it gains a photon for every photon
created at 1 .
A further scenario now presents itself. What would happen if the sole input wave were
the one at 3 ? Since there is always some background noise covering the optical spectrum,
might this be able to provide seed signals at 1 and 2 to get the DFG process off the
ground? If so, how would an 3 photon know how to divide its energy between the two
smaller photons? After all, there is an infinite number of ways to break a pencil into two
parts, so which of the infinite set of possibilities would the photon choose?
The quick answer to these questions is that the process does indeed work as suggested,
and that the split that most closely satisfies the phase-matching condition is the one that
prevails. If you like, you can think of the photons trying out all options at once, with
the most efficient being the one that wins. The process under discussion is called optical
parametric amplification (OPA), and it can be performed in an optical cavity to create an
optical parametric oscillator (OPO) [8]. OPOs offer a route to generating tunable coherent
radiation because the output frequency can be varied by controlling the phase matching. As
an alternative source of coherent radiation, the OPO should be regarded as a close relative
of the laser in which the fundamental physical process is optical parametric amplification
rather than stimulated emission.
Frequency-mixing processes are conveniently represented diagrammatically as shown
in Fig. 1.8, where the first three frames depict SHG, SFG, and DFG/OPA, respectively.
Up-arrows indicate waves that are (ideally) being depleted, and down-arrows those that are
being enhanced. But unless the phase-matching conditions are satisfied, these directions
apply only to the first (and subsequent odd-numbered) coherence lengths, and are reversed
13 Of course, the difference frequency ( ) would also have appeared in Eq. (1.17) had that equation been
2
1
written out in full. The focus on (3 2 ) is to emphasise the fact that difference frequency generation is sum
frequency generation in reverse.
Introduction
12
1
SHG
Fig. 1.8
SFG
DFG/OPA
THG
Diagrammatic representation of various nonlinear processes including second harmonic generation (SHG), sum
frequency generation (SFG), difference frequency generation (DFG), optical parametric amplication (OPA), and third
harmonic generation (THG).
in the others. More sophisticated versions of these diagrams featuring virtual energy levels
will be introduced in Chapter 8.
1.10 Third-order processes

A number of important third-order processes can be identified within the framework used
in this chapter if we include a cubic term in Eq. (1.7), in which case the equation becomes
P = 0 ( (1) E + (2) E 2 + (3) E 3 + ).
(1.24)
The kind of nonlinear response represented by the new term is typified by curve c in Fig.1.2.
Notice that, unlike the quadratic term, the way the response depends on the magnitude of
E is independent of the sign of E, so the behaviour is symmetrical between positive and
negative fields. The consequence is that (3) is non-zero in centrosymmetric materials in
which (2) = 0 by symmetry, so an ordinary block of glass (and liquids and gases too)
exhibit processes arising from the third-order nonlinearity.
What new processes does the cubic term give rise to? If one inserts Eq. (1.14) into
Eq. (1.24), one obtains a number of new polarisation components including
P = 0 ( + 3 (3) (A20 + 14 A21 )A1 cos{t k1 z}
+ 32 (3) A0 A21 cos{2t 2k1 z} + 14 (3) A31 cos{3t 3k1 z} + ).
(1.25)
As we discovered in Section 1.6, terms in cos t represent refractive index changes, and
indeed the DC Kerr effect is represented by the term in A20 in Eq. (1.25). However, there
is clearly now a further index change based on the following term, which involves A21 =
2I /c0 n where I is the intensity (i.e. the power per unit area) of the field at . This term
13
1.11 Theoretical foundations
arises from the optical-frequency Kerr effect (sometimes called the AC Kerr effect), and it
is the origin of the intensity-dependent refractive index, a process in which a beam of light
changes its own refractive index. It is quite easy to show from Eq. (1.25) that the index
change is14

3 (3)
n
I.
(1.26)
4c0 n2
This effect underlies the potentially damaging effect of self-focusing, as well as the temporal
effects of self-phase modulation (SPM) and self-steepening. Self-phase modulation plays
a crucial role in a variety of pulse compression processes, as well as in optical soliton
formation.
A generalised version of the intensity-dependent refractive index can be discovered if one
inserts the two-frequency field of Eq. (1.16) into Eq. (1.24). In this case, the polarisation
includes the terms

P = 0 ( + 32 (3) A21 A2 cos{2 t k2 .r} + 32 (3) A22 A1 cos{1 t k1 .r}). (1.27)
The first term represents a change in the refractive index at 2 associated with the intensity
of the wave at 1 , while in the second term the roles are reversed. This process in which one
beam modifies the refractive index of another is known as cross-phase modulation (XPM).
The case of the intensity-dependent refractive index mentioned earlier is the special case
where a single beam modifies its own propagation environment.
The two final terms in Eq. (1.25) represent DC-induced second harmonic generation
and third harmonic generation, respectively. SHG is of course normally forbidden in centrosymmetric materials, but an applied DC field lifts the symmetry and enables the process
to occur. Third harmonic generation (THG; see the right-hand frame of Fig. 1.8) is the
third-order analogue of second harmonic generation. Unlike SHG, it can occur in a centrosymmetric medium, although the larger frequency difference between the fundamental
and the harmonic means that birefringent phase matching is harder to achieve.
1.11 Theoretical foundations

We now consider the strengths and weaknesses of the simplified treatment given in this
introductory chapter. By expanding the polarisation in powers of the electric field (see
Eq. 1.24), we have certainly succeeded in discovering quite a number of nonlinear optical phenomena, all of which exist, and most of which are in the mainstream of current
technology. The vital importance of achieving phase matching (in processes where it is
not automatic) has also emerged, and the entire treatment has been simple and straightforward. It is in fact remarkable how much activity in nonlinear optics today has its roots in
phenomena that have featured in this chapter.
14 It is common to write n = n I where n is known as the nonlinear refractive index; see Section 5.2.6. We
2
2
have not done that here to avoid notational confusion with the refractive index at frequency 2 .
Introduction
14
The approach we have adopted is, however, deficient in a number of respects, and one or
two false impressions have been given. Firstly, the fundamental origin of optical nonlinearity
has never even been mentioned nor, in consequence, has any proper justification been given
for writing the polarisation as a power series in the electric field to begin with. Indeed,
plugging different electric field combinations into the expansion, and finding the resulting
contributions to the polarisation feels more like a mathematical exercise than a serious
exercise in nonlinear physics.
The key point is that, so far, all the physics at the atomic and molecular level has been
hidden inside the nonlinear coefficients (n). To do things properly, one needs to study the
origin of optical nonlinearities quantum mechanically. Perturbation theory can normally be
used for this purpose provided the applied electric fields are weak compared to those within
the nonlinear material. The expansion of the polarisation in powers of the electric field then
arises naturally as the theory is taken to successive orders. Elaborate quantum mechanical
expressions for the nonlinear coefficients emerge from the analysis (take a quick glance at
Chapter 8!), and numerical values can be calculated, at least in principle. One discovers that
the coefficient for a given process depends on the frequencies of all the participating fields,15
so the implication in this chapter that all nonlinear processes of a given order share a single
nonlinear coefficient (i.e. that all second-order interactions depend on the same (2) ), turns
out to be incorrect. For example, the nonlinear coefficient governing optical rectification is
not the same as the one controlling second harmonic generation, nor (at third order) is the
coefficient of the DC Kerr effect the same as that for the optical Kerr effect.
Things become more complicated under resonant conditions, in other words when one
or more of the field frequencies approaches transition frequencies in the nonlinear medium.
Resonance does not of itself necessarily invalidate the perturbation theory approach. In
some circumstances, the nonlinear coefficients simply become complex, and a range of
resonant nonlinear interactions (such as two-photon absorption and stimulated scattering)
then appear. However, perturbation theory certainly fails in the case of coherent interactions
(such as self-induced transparency) where relatively strong fields act on time-scales that are
shorter than the damping times. These issues are discussed in Chapters 5 and 9. Resonance
or no resonance, perturbation theory invariably fails if the field strength is sufficiently
high, an obvious example being high harmonic generation, which is based on the brute
force ionisation of an atom and the subsequent recollision of the extracted electron with the
parent ion. The topic is treated in Chapter 10.
Other important issues that have so far been ignored include the vector nature of the
field and polarisation, along with all structural aspects of the nonlinear media, many of
which are crystalline and probably birefringent too. The parameters in Eq. (1.24) were all
written as scalars, whereas the polarisation and the field are in fact vectors, and the nonlinear
coefficients are actually tensors. So the equation should really be written
P = 0 ( (1) E + (2) EE + (3) EEE + )
15
(1.28)
Some deeper questions about the nature of optical susceptibilities, and whether the equations for P as a
function of E should be interpreted in the time domain or the frequency domain, are considered in Appendix
B.
1.12 The growth of nonlinear optics
15
where (1) , (2) and (3) are second-, third-, and fourth-rank tensors, respectively. Readers
who are daunted by tensors will be reassured to know that, just as a simple vector (a firstrank tensor) can be written as a column matrix containing its Cartesian components, so a
second-rank tensor can be written as a two-dimensional matrix and, in general, an nth-rank
tensor written as an n-dimensional matrix. In matrix form, Eq. (1.28) duly becomes
(1)
(2)
(3)
Pi = 0
(i, j , k, l = x, y, z).
ij Ej +
ij k Ej Ek +
ij kl Ej Ek El
j
jk
j kl
(1.29)
(1)
Notice that Eq. (1.29) appears to suggest that the linear susceptibility ij has 32 = 9
(2)
components, the second-order nonlinear coefficient ij k has 33 = 27 components, and the

(3)
third-order coefficient ij kl has 34 = 81. Fortunately, the situation is not as bad as it seems.
(1)
At worst, it turns out that there are three independent components of ij , while in many
(2)
common nonlinear crystals, most elements of ij k are zero. Again, third-order processes
are most often studied in isotropic media,16 where most of the 81 potential coefficients are
zero by symmetry. These issues are treated in more detail in Chapters 3, 4 and 5.

It was suggested in the Preface that most of the foundations of nonlinear optics were
laid in the 1960s. It is certainly a salutary experience to re-read the seminal 1962 paper
of Armstrong, Bloembergen, Ducuing and Pershan (ABDP) [9] and see the depth and
breadth of its coverage of the subject at such an early date. By the end of that decade,
most of the basic principles of the field as we know it today were established. However,
knowing the principles is one thing; exploiting them is quite another. Lasers themselves were still in their infancy in the 1960s, and the story of the last half-century
has been of laser physics and nonlinear optics growing in parallel, each fertilising the
other, and with technological advances always making yesterdays dreams todays reality. The basic themes have constantly been re-orchestrated using superior materials, in
new combinations and environments, and especially on increasingly challenging time and
distance scales.
Nowhere has progress been more crucial than in materials technology where developments in ultrahigh purity crystal growth and the fabrication of complex semiconductor
heterostructures, designer nonlinear media, photonic crystals, and quantum-confined
materials have been astonishing. Consider, for example, the early history of the optical
parametric oscillator (or OPO). The possibility of generating tunable coherent light using
OPOs was understood as early as 1962, and the first OPO was demonstrated in 1965 [10].
16 The distinction between structural and optical isotropy is addressed in Chapter 5.
Introduction
16
But for the next 20 years, OPO research proceeded only slowly, primarily because the nonlinear materials required to create viable devices were not available. But by the 1980s, the
necessary fabrication techniques had been developed, and sophisticated OPO sources based
on high-grade crystals are now in widespread use.
A similar story can be told about quasi-phase matching (QPM). Suggested as a phasematching technique in ABDP (see Fig. 10 in [9]), QPM could not be implemented reliably
for many years, because it was not possible at that time to grow thin-layered stacks of
nonlinear crystals with alternating orientation. But once periodic poling techniques were
perfected [11], QPM quickly became standard, to the point where, today, some researchers
regard birefringent phase matching (BPM) as outdated.
Of the nonlinear optical phenomena discovered in this chapter, almost all the frequencymixing processes are now commonplace. It turns out that a remarkably large number of
nonlinear optical techniques are based in one way or another on the optical Kerr effect, which
appeared in Eq. (1.25). As mentioned in Section 1.10, this leads to self-phase modulation
in which a beam or pulse of light changes its own local refractive index. This effect is
the physical basis of self-focusing, spatial and temporal soliton formation, optical bistable
devices, as well as standard techniques for optical pulse compression and laser modelocking.
As noted already, cross-fertilisation between nonlinear optics and laser physics has been
an important feature of developments in both areas; indeed the two fields are so closely
related, that it is sometimes hard to draw a clear line between them. The technique known
as Kerr-lens mode-locking (KLM) is an example of this synergy [12]. The spatial properties of a laser beam are altered by the optical Kerr effect because the refractive index
near the beam centre, where the intensity is greatest, is increased relative to that at the
periphery. If the effect becomes strong enough to overcome diffractive beam spreading,
catastrophic self-focusing can occur. But at a weaker level, the effect on an optical pulse
can be exploited to remarkable effect. If it can be arranged that, at an aperture placed somewhere in the beam path, the beam size decreases as the intensity increases, the transmission
through the aperture will be greatest at the peak of the pulse. But this is precisely the
recipe for a passive mode-locking device, and this is basically how KLM mode-locking
works. The technique is widely used today and, in Ti:sapphire lasers (among others), it
enables pulses shorter than 10 fs to be generated. Of course, a modest amount of energy
compressed into 10 fs corresponds to high peak power, and this is just what is needed to
excite optical nonlinearities. In one scenario, after further compression into the sub-10 fs
regime (using other nonlinear optical techniques), the conditions are met for high harmonic
generation (HHG) in which a wide range of harmonics is generated in a gas jet serving
as the nonlinear medium. Detailed analysis (see Chapter 10) indicates that a train of ultrabroadband pulses is generated with a periodicity that is twice the optical carrier period.
Under optimum conditions, the signal emerges as a train of attosecond pulses,17 or possibly as a single pulse. At the time of writing, this particular problem lies at the research
frontier.
17 1 attosecond (1 as) = 1018 s.
17
The principle behind high harmonic generation is radically different from anything envisaged in the 1960s, and this process certainly involves new principles. They may have been
called the swinging sixties, but it is very doubtful if anyone in that decade had the idea of
tearing an electron out of an atom in an intense optical field, and then using the field to drive
the electron back to collide with the parent ion [13]. Of course, Eq. (1.24) is a series expansion, and series expansions always break down when signal strengths are high. It would
therefore not be surprising to find that other strong-field effects involve new principles too.
The foundations for many strong-field coherent effects in optics were in fact laid in the
classic papers of McCall and Hahn on self-induced transparency (SIT) [14], and the effects
studied there can certainly not be accommodated within Eq. (1.24). However, some of the
principles and much of the terminology and mathematical notation in McCall and Hahns
work were inherited from nuclear magnetic resonance, which was discovered by Isidor
Rabi in 1938.18 The terms Rabi frequency and Rabi flopping are now commonplace in
optics (see Chapter 9). Coherent nonlinear optical processes studied more recently include
electromagnetically-induced transparency (EIT) [15,16], lasing without inversion (LWI)
[17] and slow light [18,19]. Both EIT and LWI arise when there are three atomic levels
coupled by two EM fields, one strong and the other weak, and the strong field transition
modifies the properties of the weak field transition. Rabi flopping on the strong transition
is crucial in both cases, as it is in SIT, but the way in which the process is exploited is novel.
Most of the topics mentioned in the foregoing discussion will be treated in more detail
later: OPOs in Chapter 2, self-focusing in Chapter 5, pulse compression and solitons in
Chapter 7, SIT and EIT in Chapter 9, and high harmonic generation and attosecond pulses
in Chapter 10. But nonlinear optics is a huge subject, and no book, let alone an introductory
text like this one, can include everything of importance in the field.
Problems
1.1 Consider second harmonic generation in lithium niobate for a fundamental field whose
(vacuum) wavelength is 1.064 m. If the effective refractive indices are 2.2339 and
2.2294 for the fundamental and second harmonic fields, respectively, find the coherence
length.
1.2 Assuming that the DC polarisation in Fig. 1.5 is uniform throughout the crystal, and
that edge effects are neglected, show that the open circuit voltage between the plates is
Vopen = Pdc d/0 , where d is the plate separation and is the DC dielectric constant
of the crystal. What is the polarity of the induced voltage? Determine the magnitudes
and directions of the vectors P, E and D in the crystal.
1.3 Prove Eq. (1.17).
1.4 Show that Eq. (1.19) can never be satisfied in a normally dispersive medium (one where
the refractive index rises monotonically with frequency).
1.5 A phase-matching configuration is possible in beta-barium borate (BBO) in which
two separate non-collinear beams at 1.064 m generate a second harmonic beam at
18 Rabi received a Nobel Prize for this achievement in 1944.
18
Introduction
0.532 m. If the effective refractive indices at the two wavelengths are 1.65500 and
1.55490, respectively, find the angle between the two fundamental beams.
1.6 What are the dimensions of the three coefficients in Eq. (1.25)?
1.7 (a) Show that the refractive index change implied at the end of Section 1.7 is n
=
(2) A0 /n.
(b) Prove Eq. (1.26).
Frequency mixing
2.1 Introduction and preview

In this chapter, we will consider several of the basic frequency-mixing processes of nonlinear
optics. The simplest is second harmonic generation (SHG), and we will take this as our basic
example. In SHG, a second harmonic wave at 2 grows at the expense of the fundamental
wave at . As we will discover, whether energy flows from to 2 or vice versa depends on
the phase relationship between the second harmonic field and the nonlinear polarisation at
2. Maintaining the optimal phase relationship is therefore of crucial importance if efficient
frequency conversion is to be achieved.
The SHG process is governed by a pair of coupled differential equations, and their
derivation will be our first goal. The analysis in Section 2.2 is somewhat laborious, although
the material is standard, and can be found in many other books, as well as in innumerable
PhD theses. The field definitions of Eqs (2.4)(2.5) are used repeatedly throughout the
book, and are worth studying carefully.
In Section 2.3, the coupled-wave equations are solved for SHG in the simplest approximation. The results are readily extended to the slightly more complicated cases of sum and
difference frequency generation, and optical parametric amplification, which we move on
to in Section 2.4. The important case of Gaussian beams is treated in Section 2.5, where the
effect of the Gouy phase shift in the waist region of a focused beam is highlighted.
2.2 Electromagnetic theory

We start from Maxwells equations for a non-magnetic medium with no free charges namely
div D = 0;
div B = 0;
curl E = B/t;
curl B = 0 D/t.
(2.1ad)
By taking the curl of Eq. (2.1c) and the time derivative of Eq. (2.1d), the B vector can be
eliminated. This leads to1
2 E +
1 2E
2P
=
0
c2 t 2
t 2
(2.2)
1 In writing this equation, a term 1 grad div P has been dropped from the right-hand side. While div D is
0
always zero in a charge-free medium, it cannot necessarily be assumed that div E and div P are zero for waves
propagating in a medium that is not optically isotropic. However, in practice one is usually justified in assuming
that div E and div P are small.
19
Frequency mixing
20
where c = (0 0 )1/2 is the velocity of light in vacuum. Ignoring the vectorial nature of
the fields for the moment, and specialising Eq. (2.2) to the case of plane waves travelling
in the +z-direction, the equation reduces to
2E
1 2E
2 (P L + P NL )
+
=
0
z2
c2 t 2
t 2
(2.3)
where P L and P NL are the respective linear and nonlinear contributions to the polarisation,
and the electric field and the polarisation are so far both real quantities.
We now express the field and polarisation waves in complex form, writing them (as is
conventional in nonlinear optics) as sums of discrete frequency components with angular
frequencies n namely2

E(z, t) = 12
E n exp{in t} + c.c. = 12
E n exp{i(n t kn z)} + c.c.
(2.4)
n
P (z, t) =
1
2
Pn exp{in t} + c.c. =
1
2
Pn exp{i(n t kn z)} + c.c.
(2.5)
where c.c. is short for complex conjugate. As we will discover shortly, the natural choice
of angular wave number3 is kn = nn n /c, where nn is the refractive index at n .
Although the time dependence of the optical carrier waves has been separated out, the
complex amplitudes E n , Pn , E n and Pn in Eqs (2.4) and (2.5) may themselves vary in
time in the case of pulsed excitation. However, we will assume for the moment that the
waves are strictly monochromatic, so the complex amplitudes are time independent. The
two parameters carrying tildes are distinguished from those carrying carets by the fact that
the rapidly varying spatial factors exp{ikn z} have been removed.4
In terms of E n and Pn , the linear polarisation is PnL = 0 (1) E n where (1) is the linear
susceptibility. Substituting Eqs (2.4)(2.5) into Eq. (2.3), and taking the linear part of the
polarisation over to the left-hand side, we obtain

2 (1 + (1) )
2 NL
E n
n
n = n Pn
2ikn
E
(2.6)
+ kn2
z
c2
c 2 0
where the second z-derivative of E n has been neglected. The natural choice of angular
wave number is clearly the one that makes E n /z = 0 in the linear approximation (where
PnNL = 0), and we duly set5

n
nn n
kn =
1 + (1) =
(2.7)
c
c
2 Some authors omit the factors of 1 in Eqs (2.4) and (2.5), and some use exp{i(k z t)} instead of exp{i( t
n
n
n
2
kn z)}. These choices affect many subsequent equations, so you should always check what conventions are being
used when looking for definitive results in books or journal articles. See Appendix A for a review of different
conventions.
3 The term angular wave number is used throughout this book for k = 2/.
4 Note that the complex amplitudes are not uniquely defined in terms of the real field and polarisation. See
Appendix E for further discussion of this issue.
5 Of course, we could have made a different choice for k at the price of making E a function of z.
n
n
21
2.3 Second harmonic generation

where nn = 1 + (1) is the refractive index at frequency n . The susceptibility (1) will
turn out to be frequency dependent, which is a manifestation of dispersion.
Equation (2.6) reduces, with the help of Eq. (2.7), to
E n
n NL
P .
= i
z
2cnn 0 n
(2.8)
This simple differential equation is the basis for all the coupled-wave equations of nonlinear optics. It contains critical information about the effect of different phase relationships
between the nonlinear polarisation and the field. If the polarisation wave has a phase
lead of /2 with respect to the field (i.e. the arguments of PnNL and i E n are the same),
Eq. (2.8) indicates that E n will grow, whereas, if the sign of the phase difference is reversed
(a phase lag of /2), the field will decay. If, on the other hand, the polarisation and field
are in phase, the amplitude of the field will be unchanged, but the phase rather than the
amplitude of the field will change under propagation. In this case, it is the angular wave
number and hence the refractive index that are modified.
It must be stressed that any time dependence of the complex amplitudes has been ignored
in the derivation of Eq. (2.8), so a more sophisticated equation will have to be developed
before we can study the nonlinear optics of short pulses. This is done in Section 7.2.

2.3.1 Coupled-wave equations
While discussing second harmonic generation, we will use the labels n = 1, 2 for the fundamental and harmonic waves, so 1 = and 2 = 2. The first task is to find P1NL and
P2NL for use in Eq. (2.8), which can be done by evaluating 0 (2) E 2 for a field of the form
E(z, t) = 12 [E 1 exp{i1 t} + E 2 exp{i2 t}] + c.c.
(2.9)
The relevant contributions to the second harmonic polarisation (see Eq. 2.5) are
P2NL = 12 0 (2) E 12 and P1NL = 0 (2) E 2 E 1 or, equivalently P2NL = 12 0 (2) E 12 exp{ikz}
and P1NL = 0 (2) E 2 E 1 exp{ikz} in terms of parameters that are slowly varying in
space. Notice in the second two equations the appearance of the phase-mismatch parameter
k defined by
k = k2 2k1 = 2(n2 n1 )/c.
(2.10)
This originates from the slightly different spatial dependencies of the nonlinear polarisations
and the fields they are driving.
Substituting the expressions for the nonlinear polarisation into Eq. (2.8), and relabelling
(2) as SHG yields
Frequency mixing
22
E 2
2 SHG 2 ikz
E1 e
= i
z
4cn2
E 1
1 SHG ikz
E2 E1 e
= i
.
z
2cn1
(2.11)
(2.12)
These are the coupled-wave equations governing second harmonic generation. Notice the
slight asymmetry in the factors on the right-hand side.6 This can be traced back to the factor
of 2 in the cross-term of (a + b)2 = a 2 + 2ab + b2 ; Eq. (2.12) contains a cross-term in the
electric fields on its right-hand side, whereas Eq. (2.11) does not.
2.3.2 The low-depletion approximation

The simplest solution to the coupled-wave equations is obtained when the energy
conversion to the second harmonic is so small that the fundamental field remains essentially
undepleted. In this low-depletion approximation, Eq. (2.12) can be ignored, and E 1 treated
as a constant in Eq. (2.11), which is now readily integrated. Setting E 2 = 0 at z = 0, one
obtains7

2 SHG E 12
2 SHG E 12 eikz 1
kz
E 2 (z) = i
= ieikz/2
z sinc
(2.13)
4cn2
ik
4cn2
2
where sinc x = (sin x)/x. Equation (2.13) is easily written in terms of intensities (in W m2 )
rather than fields (in V m1 ) with the help of the formula

2

In (z) = 12 c0 nn E n (z)
(n = 1, 2).
(2.14)
The second harmonic intensity is readily shown to be
I2 (z) =

(2 SHG I1 )2
2 kz
sinc
.
2
80 c3 n2 n21
(2.15)
This is a crucial result. Notice that I2 varies as the square of both the nonlinear coefficient,
and the fundamental intensity I1 . Moreover, if k = 0, the sinc function goes to unity, and
I2 grows as the square of the distance too. On the other hand, when k = 0, Eq. (2.15)
becomes
I2 (z) =
(2 SHG I1 )2 sin2 {kz/2}

.
k 2
20 c3 n2 n21
(2.16)
In this case, the intensity oscillates between zero and a maximum of

I2max =
1
20 c3 n2 n21
2 SHG I1
k
2
.
(2.17)
6 Had we written 2 instead of in Eq. (2.11), it would have been even easier to overlook the asymmetry.
1
2
7 It is a common notational convention to replace SHG with 2d ; see Chapter 4.
eff
23
SH intensity
PM
non-PM
Fig. 2.1
2
3
coherence lengths
Second harmonic intensity as a function of propagation distance under phase-matched (PM) and two non-PM
conditions. The distance scale applies to the non-PM solid line. The dotted line is for a coherence length that is 1.5
times greater.
The intensity reaches this value for the first time at z = Lcoh where
Lcoh =
|k|
(2.18)
which is known as the coherence length.8

Within the present plane-wave analysis, the conversion efficiency is given by I2 (z)/I1 (0).
However, in the more realistic case where the fundamental and harmonic are in the form of
beams (see Section 2.5), a more sophisticated definition that takes the beam profiles into
account is needed.
Figure 2.1 shows the behaviour of I2 (z) under phase-matched (PM), and two non-phasematched (non-PM) conditions. The distance axis is in units of coherence length for the
non-phase-matched case plotted as a solid line; the dotted line is for a coherence length 1.5
times larger. Notice that I2max increases as k gets smaller, in line with Eq. (2.17).
2.3.3 ManleyRowe relations for second harmonic generation

Naturally, when significant energy is transferred to the second harmonic, the fundamental
field will be depleted, and the approximation on which Eqs (2.15)(2.17) are based will break
down. It is then necessary to solve Eqs (2.11) and (2.12) as a coupled pair. In the present
context, we simply note that these equations ensure that the total energy (and therefore
the intensity) of the field is conserved in the energy exchange between fundamental and
8 Sadly, there is no generally agreed definition of coherence length in nonlinear optics. Definitions with 1, 2, ,
and 2 in the numerator of Eq (2.18) can be found in different textbooks. The definition in Eq (2.18) is the one
that was used almost universally in the 1960s, and I see no reason to change it.
Frequency mixing
24
harmonic. For it is easy to show (see Problem 2.1) that Eqs (2.11) and (2.12) ensure that
dI1
dI2
=
.
dz
dz
(2.19)
In terms of the number of photons )n in each beam, it is also easy to show that
d)1
d)2
= 2
dz
dz
(2.20)
which is a statement that for every photon gained (or lost) at 2, two photons are lost (or
gained) at . Equation (2.20), and its counterpart in more general situations (see Section
2.4.1), are known as ManleyRowe relations.
2.3.4 Interlude: third harmonic generation

We interrupt the discussion of second harmonic generation in order to highlight the
close analogy with third harmonic generation (THG), which is its third-order counterpart. The equations governing the two processes are in fact so similar that one can almost
write them down by guesswork. The key difference is of course that SHG requires a
non-centrosymmetric medium, whereas THG can be observed in centrosymmetric media
including amorphous solids, liquids and gases.
Third harmonic generation arises from the frequency combination + + = 3 and,
for the purposes of this section, we write 1 = and 3 = 3. The analogue of Eq. (2.13) is
E 3 (z) = ieikz/2

3 14 THG E 13
kz
z sinc
2cn3
2
(2.21)
where the mismatch parameter is now defined as k = k3 3k1 . The corresponding

expression for the third harmonic intensity is
I3 (z) =
(3 THG )2 3
I1
402 c4 n3 n31
sin {kz/2}
k
2
(2.22)
which is analogous to Eq. (2.16) for SHG. Notice the cubic relationship between the fundamental and the third harmonic intensities, a feature that is clearly evident in the experimental
results of Fig. 2.2.
When k = 0, Eq. (2.22) indicates that I3 (z) (like I2 (z)) increases as the square of the
distance of propagation. However, it is much more difficult to achieve phase matching for
THG than for SHG. Under non-phase-matched conditions, it is clear from Eq. (2.22) that
I3 (z) varies as ( THG /k)2 . Since the nonlinear coefficient and the mismatch parameter
are both proportional to the number density of the nonlinear medium, this implies that the
harmonic intensity is independent of the number density. It follows that the harmonic signals
generated by gases and solids may be of comparable magnitude.
25
I3
Gradient
3 line
I1
(Units arbitrary)
Fig. 2.2
Third harmonic intensity as a function of fundamental intensity plotted on a log-log scale and showing wide
experimental scatter.
2.3.5 The antenna picture

After this brief excursion into the third harmonic case, we return to second harmonic generation. A useful way of understanding the SHG process is to write the second harmonic field
at the exit surface of the nonlinear medium as the sum of contributions from all upstream
points in the nonlinear medium, shown as slices of width dz in Fig. 2.3. Each slice acts
as an optical frequency antenna, radiating its contribution to the total harmonic field in the
forward direction.9 In this spirit, we write
i
E 2 (z) E 2 (z) exp{ik2 z} =
0 cn2
z
0
dz P2 (z ) exp{ik2 (z z )}

harmonic propagation
(2.23)
where P2 (z )dz represents the strength of the antenna, and the final term is the phase
change of the harmonic field generated at z in propagating from z to z. Substituting
9 The antennae actually radiate in all directions, but the phasing is correct for constructive interference only in
the forward direction.
Frequency mixing
26
dz'
Fig. 2.3
z'
The antenna model. The harmonic signal at the exit of the crystal is viewed as the sum of contributions from upstream
sources.
P2 (z ) = 12 0 SHG E 12 exp{i2k1 z } into the equation yields
E 2 (z) =
z
i SHG E 12
dz exp{ikz }
2cn2
(2.24)
which leads immediately to Eq. (2.13).

Within the present plane-wave approach, the antenna picture of Eq. (2.23) merely offers
an alternative interpretation of Eq. (2.13). It will, however, be used in the next section to
provide important insight into the SHG process, and will prove to be an indispensable tool
for the treatment of focused beams in Section 2.5.
2.3.6 Phase relationships and quasi-phase matching

We will now focus our attention on the way in which the phase of the harmonic field develops
in second harmonic generation, particularly when the process is not phase matched. To
display the results, we will use a graphical representation of the complex line integral of
Eq. (2.24), known as a phasor diagram. This is constructed by treating the integral as a sum
of infinitesimal vectors in the complex plane, as shown in Fig. 2.4.
If the fundamental field E 1 is real and positive, Eq. (2.24) indicates that E 2 is negative
imaginary under perfect phase-matching conditions (k = 0). In this case, the phasor diagram consists of a straight line running in the negative imaginary direction from O towards
A in Fig. 2.4. The length of the line (representing the second harmonic field amplitude)
increases linearly with propagation distance, so the intensity goes as the square of the
distance, as in Fig. 2.1.
In the presence of dispersion, the phasor plot becomes a curve. For normal dispersion
(n2 > n1 ), Eq. (2.10) indicates that k > 0, and the effect (through Eq. 2.24) is to cause
the phasor line to veer away from OA, and to trace out the circular arc OB.10 The change
in the phase-matching conditions does not affect the length of the phasor path, only its
shape. At point B in the figure, the net field E 2 (zB ) is given by the straight line OB. It
is easy to see from Eq. (2.24) that the (obtuse) angle between OA and the tangent to the
10 The distinction between normal dispersion (n > n ) and anomalous dispersion (n < n ) is discussed in
2
1
2
1
Section 6.2 of Chapter 6.
27
imag
F
real
Fig. 2.4
Phasor representation of second harmonic generation in the complex plane. OA represents the phase-matched
harmonic eld, OBC the phasor path for the rst coherence length under non-phase-matched conditions, and CD part
of the second coherence length. For quasi-phase-matching, the path is OBCEF.
circle at B (dotted in Fig. 2.4) is kz, and it can also be shown (see Problem 2.3) that
OB = OC sin{kz/2}.
After one coherence length, the phasor plot reaches C, where z = Lcoh = / |k|.
The net harmonic field is real at this point, so it is tempting to conclude that the fundamental and harmonic fields are in some sense in phase. However, the slowly varying
envelopes only provide phase information in combination with the appropriate exponential factors from Eq. (2.4) and, in any case, one should always treat statements like in
phase or in quadrature with caution for two waves of different frequency. (After all,
if the fundamental and second harmonic fields interfere constructively at one point, they
will interfere destructively half a cycle later.) What counts in SHG is the relative phases
of the harmonic field and the associated nonlinear polarisation, which is basically the 2
component of the fundamental field squared. After one coherence length, this has changed
by , which is why the phasor path at C in Fig. 2.4 is running in the positive imaginary
direction, in exact opposition to its initial direction at O. The amplitude of the second
harmonic field E 2 (z = Lcoh ) has clearly reached a maximum at C and, thereafter, the
phasor path will continue via D, returning to O (zero harmonic field) after two coherence
lengths.
But Fig. 2.4 invites us to consider another possibility. Suppose that at C, the sign of
the nonlinear coefficient could be changed. The phasor path would then reverse direction,
proceeding via E to F, where the amplitude of the harmonic field would be double what it
was at C. And if, at F, the sign could be changed again, and the reversal repeated after every
coherence length, the harmonic field could be increased without limit.
Fortunately, changing the sign of the nonlinear coefficient at regular intervals can be
realised experimentally by growing a non-centrosymmetric crystal as a series of slabs,
each slab one coherence length thick, and with the orientation of alternate slabs reversed.
Imagine a loaf of sliced bread with every other slice turned upside down. This method of
Frequency mixing
28
overcoming refractive index mismatch was in fact proposed by Armstrong, Bloembergen,

Ducuing and Pershan (see Fig. 10 of [9]). However, it was many years before it was
reliably achieved in the laboratory by using a process known as periodic poling [11].
This approach to overcoming the phase-mismatch problem is called quasi-phase matching
(QPM) to distinguish it from birefringent phase matching (BPM) [5]. Notice from Fig.
2.4 that the second harmonic field generated in QPM is phased in a direction close to
the real axis, roughly in quadrature to the harmonic field in BPM under phase-matched
conditions.
The poling period of a periodically poled material is defined as the thickness of one pair
of inverted slabs or, in other words, two coherence lengths. Hence
Lpole = 2Lcoh =
2
.
|k|
(2.25)
In practice, this is typically a few tens of microns, so the slices of bread are very thin!
The intensity growth of a second harmonic under QPM conditions is shown in Fig. 2.5;
notice that the curve follows the non-phase-matched curve for the first coherence length,
but the paths separate as soon as the polarity is reversed. The step-like nature of the QPM
line follows directly from the geometry of Fig. 2.4, although that figure represented the field
whereas Fig. 2.5 shows the intensity. Had the field rather than the intensity been plotted,
the steps would have been of equal height.
It is clear from Figs 2.4 and 2.5 that the second harmonic field under QPM is smaller
than for perfect phase matching. After integral numbers of coherence lengths, the relevant
factor is 2/ = 0.637, which is the ratio of the diameter of a circle to half its circumference
and a result that follows immediately from Fig. 2.4. Since intensity goes as the square of
the field, the QPM intensity after two coherence lengths (at F in Fig. 2.4) is four times that
at C, and the intensity ratio is (2/ )2 = 0.405.
QPM
SH intensity
PM
non-PM
0
Fig. 2.5
2
3
coherence lengths
Second harmonic intensity as a function of propagation distance for quasi-phase-matching (QPM). The dotted lines
are reproduced from Fig. 2.1 for comparison.
29
2.4 General three-wave processes
2.3.7 Additional features

A number of important features have been left out of the simple treatment given in this
section. In particular:
(A) The plane wave is an idealisation that is never realised in practice. SHG experiments
are normally conducted using beams of finite lateral extent that may pass through a
focus in or near the nonlinear medium.
(B) Most nonlinear materials used for SHG are birefringent, and at least one of the coupled
fields is usually an extraordinary wave. As we shall see in the next chapter, this means
that the energy will angle off to the side, reducing the efficiency of the nonlinear
interaction.
(C) The nonlinear coefficient that couples the fields is a tensor, and the fundamental and
harmonic fields are often orthogonally polarised.
Item (A) is discussed in Section 2.5 below, and items (B) and (C) in Chapters 3 and 4.
Notwithstanding its limitations, the simplified treatment we have presented has revealed
the crucial importance of phase matching for SHG, and indeed for many other second-order
frequency-mixing processes, some of which we consider next.

2.4.1 Coupled-wave equations and the ManleyRowe relations
We now move on to the general case where three waves with frequencies linked by the
equation 1 + 2 = 3 interact through the second-order nonlinearity.11 As for SHG, we
need to find a set of coupled-wave equations based on Eq. (2.8), but this time we must use
a field of the form
E(z, t) = 12 [E 1 exp{i1 t} + E 2 exp{i2 t} + E 3 exp{i3 t}] + c.c.
(2.26)
We work out 0 (2) E 2 as before and, of the terms that then appear, those of special interest are P3NL = 0 (2) E 1 E 2 exp{ikz}, P2NL = 0 (2) E 3 E 1 exp{ikz} and
P1NL = 0 (2) E 3 E 2 exp{ikz}. Substitution of these expressions into Eq. (2.8) yields
the three coupled differential equations.
3 TWM ikz
E 3
E 1 E2 e
= i
z
2cn3
(2.27)
11 In this section, the only numerical significance of the subscripts is that is the highest frequency of the three.
3
Frequencies 2 and 3 are no longer the second and third harmonics as in earlier sections.
Frequency mixing
30
E 2
2 TWM ikz
E 3 E1 e
= i
z
2cn2
E 1
1 TWM ikz
E 3 E2 e
= i
z
2cn1
(2.28)
(2.29)
where the label TWM stands for three-wave mixing, and k = k3 k2 k1 .12 These
equations are evidently close relatives of Eqs (2.11)(2.12). The equation paralleling
Eq. (2.19) is
dI1
dI2
dI3
+
=
dz
dz
dz
(2.30)
which is a statement of conservation of energy among the three interacting fields. For the
photon numbers, Eq. (2.20) generalises to
d)1
d)2
d)3
=
=
dz
dz
dz
(2.31)
where the set of identities within Eq. (2.31) are the ManleyRowe relations for the general
three-wave mixing process. In physical terms, they state that, for every photon gained (or
lost) at 3 , one is lost (or gained) at both 1 and 2 .
Notice that the derivation of Eqs (2.27)(2.29) was based on selecting terms of special
interest, and discarding all the others. But what allowed us to throw other terms away?
What about terms at 21 for example? Or at 1 + 3 ? The answer is that we are assuming that the 1 + 2 = 3 process is the only one close to phase matching, while other
frequency combinations are not phase matched and can therefore be ignored. The assumption is normally justified in practice, which highlights the importance of phase matching in
nonlinear frequency mixing.
While Eqs (2.27)(2.29) govern all nonlinear processes of this type, there are a number
of different cases depending on which waves are present initially and their relative strengths
and phases. In brief these are:
Sum frequency generation (SFG), where two waves at 1 and 2 add to give 3 =
1 + 2 . As the wave at 3 is enhanced, the waves at 1 and 2 are depleted (or vice
versa if and when the process goes into reverse). When one of the incident frequencies
is much stronger than the other, SFG may be termed parametric up-conversion, the
thought being that the strong wave converts the weak wave to the sum frequency.
Second harmonic generation is a special case of SFG. In quantitative work, it makes a
difference whether there is a single fundamental beam, or two distinct beams at the same
frequency that might for instance be non-collinear. The analysis of Section 2.3 applies to
the former case. In the latter case, the formulae for SFG should be used with 1 = 2 ;
see Section C2.3 of Appendix C for further discussion.
12 This is our standard definition of the mismatch parameter in which the high-frequency value comes first; see
Appendix A.
31
Difference frequency generation (DFG), where 3 and one of the two lower frequencies
(say 2 ) subtract to give 3 2 = 1 . Once 1 is present, it also subtracts from 3

to interact with 2 . DFG is just SFG in reverse, so 3 is depleted, and both 1 and 2
grow. It may perhaps come as a surprise that the wave at 2 increases in strength, but
this becomes obvious as soon as one writes the DFG process in terms of photons, i.e.
3 1 + 2 ; see also Eq. (1.23). Every time a photon at 1 is created, so is a
photon at 2 . Again, when one of the incident frequencies is much stronger than the
other, DFG may be termed parametric down-conversion.
Optical parametric amplification, which is based on exactly the same equations as DFG.
The only difference is that the two lower frequencies 1 and 2 are normally very weak
initially, and one or both may even be generated from noise. As shown below, this affects
the nature of the solution to Eqs (2.27)(2.29).
Although general analytical solutions to Eqs (2.27)(2.29) can be obtained, approximate
solutions are usually more than adequate, and more manageable too. As for SHG, approximations are possible if at least one of the incident fields is sufficiently strong that the change
in its magnitude resulting from the nonlinear interaction can be neglected. However, the
results exhibit some quite subtle features, and care is necessary.
2.4.2 Modified field variables

For the discussion of specific three-wave processes such as sum and difference frequency
generation, we will write the coupled-wave equations in terms of the modified field
variables [20]

ni
Ai =
Ei .
(2.32)
i
Equations (2.27)(2.29) then simplify to
A 3
= iA 1 A 2 eikz ;
z
A 2
= iA 3 A 1 eikz ;
z
A 1
= iA 3 A 2 eikz (2.33)
z
where the coupling coefficient common to all three equations is

3 2 1 TWM
=
.
(2.34)
n3 n2 n1 2c
2

The simplification occurs because A i is directly proportional to the photon flux Fi
(photons per area per time) through the relation
2

1
c
n
Ei
0
i
2
Ii
c0 2
Fi =
=
=
(2.35)
Ai .
i
2
i
Since the coupled-wave equations govern processes where gain (or loss) of a photon at 3
is accompanied by the corresponding loss (or gain) of photons at 1 and 2 , the nearer one
can get to photon language, the simpler the equations are bound to become.
32
Frequency mixing
2.4.3 Sum and difference frequency generation

In the case of SFG, if 1 and 2 are roughly equal in strength and the energy conversion to
3 is small, depletion of both incident waves can be ignored, which leads to

kz
ikz/2
A3 (z) = ie
A1 A2 z sinc
.
(2.36)
2
This equation is clearly a close analogue of Eq. (2.13), and it is worth comparing the
formulae carefully to appreciate the close relationships between then.
A parallel result can be obtained for DFG (3 2 = 1 ) if the energy conversion to
1 is sufficiently small that both A 3 and A 2 can be treated as constants. In this case, one
obtains

kz
ikz/2
A1 (z) = ie
A3 A2 z sinc
.
(2.37)
2
However, care must be exercised with this equation, as it is only valid under these very
restricted conditions. The more general case where A 2 is allowed to vary is treated in the
following section.
2.4.4 Optical parametric amplification

In an optical parametric amplifier (OPA), the high frequency 3 is called the pump, the
lower frequency of primary interest is called the signal, and the remaining frequency is
called the idler. In practice, the signal usually has a higher frequency than the idler, and we
will identify the signal with the wave at 2 for the purpose of the present discussion.
We now examine the behaviour of Eqs (2.33) in the case where depletion of the pump
beam at 3 can be neglected, so that A 3 is a constant, and the first of the three coupled
equations can be ignored. Both A 1 and A 2 are allowed to vary, and combining the remaining
two differential equations yields
2 A 2
= g 2 A 2 gk A 1 eikz
z2
2 A 1
= g 2 A 1 gk A 2 eikz
z2
(2.38)
(2.39)
where g is the parametric gain coefficient given by

TWM A 3
g = A 3 =
2c
3 2 1
.
n3 n2 n 1
(2.40)
For convenience, we will assume that TWM and A 3 (and hence also g) are real quantities.13
13 The nonlinear coefficient will be real provided the fields are well away from material resonances. The pump
field can be taken to be real without any loss of generality.
33
Remarkably, Eqs (2.38)(2.39) possess the analytic solution14

sinh g z
A 2 (z)eikz/2 = A 2 (0) cosh g z + i 12 k A 2 (0) g A 1 (0)
(2.41)
g

sinh g z
ikz/2
A1 (z)e
= A1 (0) cosh g z + i 2 k A1 (0) g A2 (0)
(2.42)
g

where g = g 2 (k/2)2 . The important general conclusion to draw from Eqs (2.41)
(2.42) is that, provided g > k/2, the equations support the exponential growth of both
signal and idler with an effective gain coefficient of g . On the other hand, when g < k/2,
g becomes imaginary, the hyperbolic functions become trigonometric functions, and the
solutions are oscillatory.
One can gain valuable insight (and have some fun too) by plotting graphs of Eqs (2.41)
(2.42). A good starting point is the case where A 1 (0) = 0 (zero initial idler), for which the
signal and idler fields are

k sinh g z
ikz/2
cosh g z + i
(2.43)
A2 (z) = A2 (0)e
2g

g sinh g z
A 1 (z) = A 2 (0)eikz/2 i
.
(2.44)
g

2

2

Figure 2.6 traces the growth of A 1 (z) and A 2 (z) in the case where A 2 (0) = 1, g = 3
and k = 4. The distance scale is in units of the nominal coherence length (= /k),
15
10
|Ai |2
0
0
0.2
0.4
0.6
0.8
Fig. 2.6
Solutions of Eqs (2.41)(2.42) for g = 3, k = 4, and A 2 (0) = 1. The distance scale is in units of the nominal
coherence length (= /k).
14 This result is quoted by Yariv in Chapter 17 of the 3rd edition of Quantum Electronics [20], albeit in a slightly
different notation, and with a different definition of k.
Frequency mixing
34
2
2
|Ai |2
0
0
Fig. 2.7
4
z
Solutions of Eqs (2.41)(2.42) for g = 1.5, and other parameters as Fig. 2.6.
although that is of little significance for these parameter values. The two curves are identical
in shape, and are separated by 1 unit in the vertical direction (see Problem 2.5). After
5 units of distance, they rise to roughly 2 108 , and are virtually indistinguishable. At
this huge gain factor, the assumption that the pump is undepleted will probably have been
infringed.
The dramatic effect of reducing g from 3 to 1.5 is shown in Fig. 2.7. Since
g is now less
than k/2, g is imaginary, and it makes sense to define k /2 = ig = (k/2)2 g 2
and to write Eqs (2.43) and (2.44) in terms of trigonometric functions (see Problem 2.6).
2

2

As in Fig. 2.6, the curve for A 1 (z) is identical to the one for A 2 (z) apart from a vertical
displacement of 1 unit, but this is the only significant feature the two figures have in
common. Both curves are evidently now oscillatory and restricted
to low values. Notice the
increase in the coherence length by the factor of k/k = 4/ 7 1.5. The dotted line
traces the solution predicted by Eq. (2.37), which is based on conditions that are clearly not
fulfilled in this case (see Problem 2.7).
The case of perfect phase matching (k = 0) is also interesting. Equations (2.41)(2.42)
then become
A 2 (z) = A 2 (0) cosh gz i A 1 (0) sinh gz
(2.45)
A 1 (z) = A 1 (0) cosh gz
(2.46)
i A 2 (0) sinh gz
which highlight the fact that, when A 1 (0) = 0, the initial phase relationship of the three
waves has a crucial effect on the evolution, particularly when A 2 and A 1 are of comparable
magnitude. This is a standard feature of multi-wave nonlinear interactions whenever all the
interacting fields are present at the outset.
35
2
|Ai|2
2
1
0
0
Fig. 2.8
2
z

Solutions of Eqs (2.41)(2.42) for g = 1, k = 0, A 2 (0) = 1 and A1 (0) = 0.9. It is left to the reader to
discover the argument of A 1 (0); see Problem 2.8.

An example is shown in Fig. 2.8 in which g = 1, A 2 (0) = 1 (and real), and A 1 (0) = 0.9.
It is left to the reader to deduce the phase of A1 (0), and to work out what would happen if
A 1 (0) were increased to 1 (see Problem 2.8).
2.4.5 Tuning characteristics

In a typical practical application, the idler alone might be absent initially (as in Eqs (2.43)
(2.44) above), or the signal and idler might both grow from noise. In the former case, the
idler phase adjusts automatically to optimise the interaction. In the latter case, the division
of the pump photon energy between signal and idler is determined by the phase-matching
conditions, as explained in Chapter 1. It follows that varying the phase-matching conditions
enables the optical parametric amplification process to be tuned.
A thorough discussion of angle tuning using birefringent phase matching (BPM) must
wait until we have studied the properties of birefringent media in Chapter 3, and the tensor
nature of the nonlinear coefficients in Chapter 4. However, the principle of tunability can
be illustrated in the case of quasi-phase matching (QPM), which was discussed in Section
2.3.6. As we saw there, if the mismatch parameter of a nonlinear interaction is k, effective phase matching can be achieved by using a periodically poled medium with a poling
period of Lpole = 2/ |k|. Figure 2.9 illustrates how the signal and idler wavelengths
in an optical parametric amplifier depend on the poling period in lithium niobate. For a
given period (on the abscissa), the two wavelengths can be read off the ordinate. The upper
and lower branches of the curve meet at 1.568 m, which is twice the pump wavelength
of 784 nm.
Frequency mixing
36
2.8
wavelength (microns)
2.4
2.0
1.6
1.2
0.8
18.8
19.2
19.6
20.0
20.4
poling period (microns)
Fig. 2.9
Tuning curve for optical parametric amplication as a function of poling period. The lower and upper branches of the
curve trace the signal and idler wavelengths.
Unfortunately, once a poled sample has been fabricated, the period is fixed, so continuous
tuning is not possible using this technique. However, samples are frequently made with
several different periodicities on the same wafer, so a limited degree of tunability can be
achieved by switching from one to the next.
2.4.6 Optical parametric oscillators

Figure 2.6 showed that an optical parametric amplifier delivers gain at the signal and idler
frequencies in the presence of a suitably strong pump, while it was shown in the previous
section how the process can be tuned. Taken together, these ingredients provide a solution
to one of the key problems of the laser era: how to create tunable coherent radiation.
Of course, broadband lasers can sometimes be tuned to a limited degree, but the tuning
range is invariably restricted, and the freedom to generate any desired wavelength at will
is lacking. But now, all that is necessary to achieve this long-sought-after goal is to put an
optical parametric amplifier within an optical cavity, where the nonlinear crystal replaces
the amplifying medium of a traditional laser. Such a device is called an optical parametric
oscillator (or OPO).
As noted already, the first OPO was demonstrated in 1965 [10], although 20 years or so
were to elapse before efficient and reliable OPOs could be realised. In designing an OPO,
one has to decide which of the three participating frequencies (pump, signal and idler) the
cavity mirrors should be designed to reflect and which they should transmit. Obviously the
pump beam has to reach the nonlinear medium somehow or other. Various different OPO
configurations can be employed but, for longitudinal pumping, it is clearly necessary for
at least one of the mirrors to transmit the pump beam. As far as the signal and idler are
2.5 Focused beams
37
SIGNAL
PUMP
Fig. 2.10
NONLINEAR
CRYSTAL
SIGNAL
+ IDLER
Schematic diagram of an optical parametric oscillator.

concerned, there are two options: in a singly-resonant OPO, the mirrors reflect the signal
and transmit the idler, whereas in a doubly-resonant device, both beams are strongly
reflected.
Broadly speaking, singly-resonant OPOs tend to be better behaved, because problems
with conflicting mode conditions that afflict doubly-resonant systems are avoided. The point
at issue here is that an optical cavity of length L exhibits FabryPerot resonances at angular
frequencies given by q = q c/L where q is a (normally very large) integer. Even if the
signal is the only frequency of the three constrained by this feature, the smooth tunability of
an OPO may be somewhat impaired. But if both the signal and idler must simultaneously
satisfy a mode condition of this kind, the problems become much more severe.
A schematic diagram of a longitudinally pumped singly-resonant OPO is shown in
Fig. 2.10. If birefringent phase matching is employed, tuning can be achieved by rotating the crystal in conjunction with other tuning elements. Short-pulse OPOs can be realised
by synchronous pumping with a mode-locked laser, which involves accurate matching of
the length of the OPO cavity to the periodicity of the mode-locked pulse train of the pump.
In this case, fine tuning can be achieved by harnessing group velocity dispersion (GVD).
Because of GVD (see Chapter 6), the cavity round-trip period of the signal pulse will be
wavelength dependent. But the period is tied to that of the pump so, if the physical length
of the cavity is changed, the signal wavelength changes too to maintain synchronism. In
other words, the signal shifts to a wavelength where the group time delay in the nonlinear
medium cancels the change in the mirror spacing.
Further material on OPOs can be found in [8] and [21].
2.5 Focused beams

2.5.1 Introduction
Everything said so far has been based on a plane-wave analysis in which the transverse
variation of the interacting fields has been ignored. But in the real world, one is invariably
working with beams, and this affects nonlinear interactions in several different ways.
Frequency mixing
38
The most obvious property of a beam is that it is more intense in the centre line than on
the periphery. So, since the efficiency of nonlinear frequency mixing depends on intensity
(see, e.g. Eqs 2.16 and 2.22), the conversion efficiency will be higher in the centre than
at the edges, and we should expect output beams in nonlinear frequency conversion to
be narrower than input beams. In a rough-and-ready approach to SHG with finite beams,
one could simply apply the formulae of Sections 2.3 and 2.4 at arbitrary points across the
fundamental wavefront, and work out the associated profile of the harmonic. Beams expand
and contract as they propagate, of course, and that would need to be taken into account.
However, if a beam passes through a focus in the course of a nonlinear interaction, the
crucial phase relationships between the various frequency components will be upset by a
feature known as the Gouy phase shift, and a more sophisticated analysis is needed. The
origin and effects of the Gouy phase on nonlinear frequency mixing will be studied in the
following sections.
2.5.2 Gaussian beams and the Gouy phase

The simplest beam-like profile is the Gaussian beam. This is an exact solution of the paraxial
wave equation (PWE), which is derived in Appendix G, and reads
E
i
= T2 E
z
2k
(2.47)
where T2 = 2 /x 2 + 2 /y 2 is the transverse Laplacian operator. In most practical

situations, the PWE is an excellent approximation to Maxwells equations, which means in
turn that a Gaussian beam is very close to an exact solution of Maxwells equations.
The spatial structure of a Gaussian beam is given by
z) =
E(r,
E0
1 i
r 2
exp
w02 (1 i )

(2.48)

where r = x 2 + y 2 is the transverse displacement from the axis, w0 is a measure of the
beam width in the focal plane (z = 0), and = z/zR is the axial distance15 normalised to
the Rayleigh length defined by16
zR =
w02
= 12 kw02 .
(2.49)
Note that and k in this equation are the wavelength and angular wave number in the
medium of refractive index n.
The basic geometry of a Gaussian beam is shown in Fig. 2.11. The beam is symmetrical
about the origin (or focal point) at x = y = z = 0 where the transverse profile is narrowest.
15 The symbol is lower-case zeta, the sixth letter of the Greek alphabet.
16 The confocal parameter b = 2z is sometimes defined instead of the Rayleigh length.
R
2.5 Focused beams
39
Fig. 2.11
Geometry of a Gaussian beam. The distance scale is in units of the Rayleigh length. The solid lines track the
half-maximum intensity contour.
The beam converges towards the focal point from the upstream side, and expands away
from it in an entirely symmetrical manner on the downstream side. In the vicinity of the
focus, the beam exhibits a waist-like structure that extends between roughly z = zR (or
= 1). The field strength on-axis (r = 0) is given by the first term in Eq. (2.46), which
can be rewritten

E
0
z) =
E(0,
exp{iGouy }.
(2.50)
1 + 2
z) yields the on-axis intensity, namely
Squaring the modulus of E(0,

2

E(0, z) =
E02
.
1 + 2
(2.51)
The phase term in Eq. (2.50) is the crucial Gouy phase mentioned at the start of this
section and given by
Gouy = tan1 .
(2.52)
As shown in Fig. 2.12, Gouy rises from at = to + at = +. In combination

with the fast-varying spatial term exp{ikz}, its effect is to reduce the effective value of k in
the waist region, and correspondingly to increase the local wavelength. As we will discover
shortly, the Gouy phase has a profound effect on nonlinear frequency-mixing processes in
focused beams.
The beam profile transverse to the axis is governed by the second term in Eq. (2.48)
which can be recast in the form

r 2
r2
r2
exp
.
(2.53)
i
=
exp
w2
w2
w02 (1 i )
Frequency mixing
40
Gouy
0
4
Fig. 2.12
The Gouy phase as a function of normalised distance.

The parameter w( ), which governs the variation of the beam width along the axis, is
given by

w( ) = w0 1 + 2 .
(2.54)
It is easy to show that the beam diameter measured at half-maximum intensity is given by

d( ) = w( ) 2 log 2 = 1.18w0 1 + 2 .
(2.55)
Clearly w0 is the value of w at the focal point where the diameter is d(0) = 1.18w0 . The
solid lines in Fig. 2.11 trace the half-maximum intensity contours.
The complex term in Eq. (2.53) shows how the phase of the beam varies as one moves
away from the axis, and this enables the radius of curvature R of the beam wavefront to be
found. By using the well-known intersecting chords theorem, it can be shown (see Problem
2.9) that the link between the phase (r, ) = r 2 /w2 and R is

|(r, ))|
r = (2R)
.
2
(2.56)

z2
1
R
z
+
= z+ R.
= R
(2.57)
and it follows that
This result indicates firstly that R as z 0 (the wavefront is plane at z = 0), and
secondly that R z as z , so when z zR the centre of curvature tends to the
focal point. The curved dotted lines in Fig. 2.11 indicate the curvature of the wavefront at
= 1 and = 4.
2.5 Focused beams
41
2.5.3 Harmonic generation with Gaussian beams

The theory of harmonic generation with Gaussian beams is quite involved [22, 23], so we
will merely present the most important results for second and third harmonic generation
(SHG and THG), and argue that they are reasonable.
The first important result to emerge from a detailed study of the problem is that the
Rayleigh length of the harmonic beam (at q) is the same as that of the fundamental beam
that created it. This means that the dimensionless distance measure = z/zR is the same
for both beams. Furthermore, since zR = 12 kw02 with k = n(q)/c (see Eq. 2.49), it follows
that the harmonic beam is narrower than the fundamental. For SHG, the factor is roughly17
2, a result that is entirely reasonable given that I2 I12 (see Eq. 2.15). In the general case
q
of qth harmonic generation, Iq I1 and the narrowing factor is roughly q.
The most insightful way of presenting the expression for the harmonic field is through the
antenna picture of Eq. (2.23). Assuming the conversion efficiency is small so that depletion
of the fundamental field can be ignored, the formulae for the harmonic field in the SHG
and THG cases are18

SHG
2
E02
2
2k
r
d exp{i2 }
1
E 2 (z) = i
exp
zR
4cn2
(1 i )
2zR (1 i )
(1 i )
E 3 (z) = i
3 THG
8cn3

E03
3k1 r 2
exp
(1 i )
2zR (1 i )

zR
(2.58)
d exp{i3 }
(1 i )2
(2.59)
where q = kq zR = (kq qk1 )zR and = z /zR .

These equations contain critical information about the effect of the Gouy phase on optical
harmonic generation. An important point to appreciate is that, in each case, the overall
Gaussian beam structure of the harmonic beam resides in the central bracket, while the final
integral governs the phase-matching characteristics including the role played by the Gouy
phase. Think carefully about the (1 i )q1 factor in the denominators of these integrals.
Why is this factor raised to the (q 1)th power? The reason is that it represents the difference
between the strength and phase of the polarisation antenna driving the harmonic field, and
the harmonic field itself. Notice how this factor introduces a phase difference between
the extremes of the integral for the SHG case, and a 2 phase difference for THG.
The key properties of the phase-matching integrals are summarised in Table 2.1, which
also includes the general case of qth harmonic generation.19 The most important feature to
emerge is that all the integrals are zero when kq > 0. This means that when a fundamental
beam is focused through a medium that exhibits normal dispersion and is of infinite extent,
17 The factor will only be exact under phase-matched conditions when the refractive indices at the fundamental
and harmonic frequencies are the same.
18 Throughout this section, the indices 2 and 3 refer once more to the second and third harmonic waves.
19 The leading factor of i from Eqs (2.58)(2.59) has been included with the integrals in Table 2.1.
Frequency mixing
42
Table 2.1 Properties of the integral of Eqs (2.53)(2.54).

kq > 0
kq = 0
kq < 0
SHG
d exp{ik2 z }
(1 i )
2i exp{k2 zR }
THG
d exp{ik3 z }
(1 i )2
2i |k3 | zR exp{k3 zR }
qth HG
d exp{ikq z }
q1
(1 i )
q2

2i kq zR
exp{k3 zR }
no second or third harmonic will be generated, or in fact, any higher harmonic either. In
practice, infinite means that the focus lies many Rayleigh lengths from either the front or
back surface.
Further insight into these properties can be gained by displaying the integrals as phasor
diagrams, as explained in Section 2.3.5 above. Some examples are shown in Fig. 2.13. Finite
limits were used to create some of the plots in this diagram to prevent them from running off

the page. For example, curve a shows the phasor plot of i d (1 i )1 with = 16,
which applies to SHG under perfect phase matching. For infinite limits, the length of the
curve is itself infinite, so the two arms of the curve would disappear off the right-hand side
of the page. The grey arrow connecting the end points at = 16 represents the value of
the integral. This is clearly negative imaginary, and inspection shows that its magnitude is
roughly 3, which is consistent with its value of i for infinite limits (see Table 2.1). The
locations of = 1 on the phasor plot are also indicated. The phase difference between
the extremes of the integral (associated with the factor (1 i )1 ) is clearly evident in the
180 degree swing in the direction of the curve in the figure.
Curve b in Fig. 2.13 represents SHG for k2 zR = +0.02 between integration limits of
= 1000. The two end-points converge to the same point on the real axis in this case, so
the value of the integral is tending to zero, as indeed it does for any positive value of k2 .
Curves c and d are for THG with k3 zR = 0 and 0.5, respectively, and corresponding
limits of = 10 and 1000. In the former case, the end-points converge towards +1 on
the real axis; the curve is in fact a perfect circle. Because the factor in the denominator of the
integral is (1i )2 , the phase difference between the ends of the curve is now 2. For curve
d, Table 2.1 gives the value of the integral for infinite limits as 2 i |k3 | zR exp{k3 zR },
which is 1.905iin this case.
Finally, it is reassuring to see that Eqs (2.58) and (2.59) are consistent with earlier results
in the plane-wave limit. For instance, if we consider Eq. (2.58) on-axis (r = 0), close to
the beam waist ( 0), and for a thin nonlinear medium extending from z = 0 to
z = z ( zR ), the second harmonic field reduces to
E 2 (z) =
i(2) SHG E02

4cn2
z
0
dz exp{ik2 z }
(2.60)
2.5 Focused beams
43
2
=16
a
=1
Imag
Real
2
=+1
=+16
Fig. 2.13
Phasor diagrams for harmonic generation in a Gaussian beam: (a) SHG at k2 = 0 for a symmetric focus in a
medium extending from = 16 to +16, (b) k2 zR = +0.02, (c) THG for k3 = 0 in the range = 16 to
+16, (d) k3 zR = 0.5. See text for details.
which is equivalent to Eq. (2.23). If k2 = 0, the field is clearly negative imaginary, just
as in the plane-wave case; see the line OA in Fig. 2.4.
2.5.4 Focused beams: summary

The key conclusion of Section 2.5 is that if a beam is focused into an infinite nonlinear medium exhibiting normal dispersion (k > 0), no net harmonic radiation will be
generated. In practical terms, this mathematical statement means that negligible harmonic
generation will result if the focal point lies within the medium and many Rayleigh lengths
from either the front or back surface. The physical interpretation of this remarkable result
is that the Gouy phase conspires with normal dispersion to ensure that the net contributions to the harmonic from sources upstream and downstream of the focus interfere
destructively.
If there is no way of making k negative (and there rarely is), the best strategy for
circumventing the problem is to restrict the extent of the nonlinear medium within the
focal region. This ensures that at least one of the integration limits in Table 2.1 is no
longer , which means that the cancellation of the contributions to the harmonic signal from the two sides of the focus is no longer complete. For a relatively thick nonlinear
medium, the focus can be moved towards the front or the back surface, or even located
outside the medium altogether, which may be a sensible measure anyway if one wants
to avoid destroying expensive nonlinear crystals. The narrow gas jets used in experiments on high harmonic generation are good examples of thin nonlinear media; see
Chapter 10.
44
Frequency mixing
Problems
2.1 Use Eq. (2.14) to show that Eqs (2.11) and (2.12) ensure that the total energy in the
two waves is conserved in the SHG interaction.
2.2 A square law relationship between the harmonic and fundamental intensities applies
for the second harmonic (Eq. 2.16), and a cubic law for the third harmonic (Eq. 2.22).
Convince yourself that, in the case of pulsed excitation, the same relationships apply
to the energy.
2.3 Assuming perfect phase matching in a second harmonic generation experiment, find
the second harmonic intensity when SHG = 0.5 pm V1 , the sample thickness is
3 mm, the fundamental vacuum wavelength is 600 nm, and the fundamental intensity
is 1 GW cm2 . Assume that the fundamental and harmonic indices are 1.52. Assume
the fundamental beam is effectively undepleted, and comment on the validity of this
approximation under these circumstances. (Hint: Watch out for mixed units in this
question.)
in Fig. 2.4 is equal to kz/2, and hence that
2.4 Prove that the angle AOB
OB = OC sin{kz/2}.
2.5 Show that, under quasi-phase-matchedconditions, the field amplitude during the sec-
ond coherence length is E(z) = ECL 12 (5 3 cos{k(z Lcoh )/2}) where ECL is
the field after one coherence length.
2
2

2.6 Convince yourself that A 2 (z) A 1 (z) in Eqs (2.41)(2.42) is a constant. What
fundamental physical principle does this fact reflect?
2.7 Rewrite Eqs (2.41)(2.42) in terms of trigonometric functions in the case where
g < k/2.
2.8 Under what conditions is Eq. (2.37) a good approximation to Eq. (2.42)?
2.9 Deduce the modulus and argument of the initial idler field in Fig. 2.8, and discover
how the figure would change if the modulus were increased from 0.9 to 1. Could this
outcome be achieved experimentally?
2.10 Verify that the Gaussian beam solution of Eq. (2.48) is a solution of the paraxial wave
equation of Eq. (2.47).
2.11 Prove Eq. (2.56) and sketch a graph of R( ) in Eq. (2.57).
2.12 A beam at 1m and its second harmonic (at 500 nm) are phased in such a way that zeros
of the fundamental are aligned with positive peaks of the harmonic.After what distance
of propagation in air do zeros of the fundamental harmonic coincide with zeros of the
harmonic? (nair 1 = 2.7412 104 at 1 m; nair 1 = 2.7897 104 at 500 nm.)
(As noted in Section 2.3.6, phase relationships in SHG are best handled by comparing
the harmonic field with the square of the fundamental field.)
2.13 Since the cosh and sinh terms in Eqs (2.41)(2.42) separately satisfy the original differential equations, why are both necessary in the solution? Equally, why does the initial
idler field occur in the solution for the signal, and vice versa? After all, Eqs (2.38)
(2.39) seems to suggest that signal and idler are completely decoupled when k = 0.
2.14 Show that curve c in Fig. 2.13 traces a perfect circle.
Crystal optics
3.1 Preview
The central feature of this chapter is the phenomenon of birefringence, also known as double
refraction, which occurs in crystals that are optically anisotropic. Given that birefringence
is a linear optical effect, why is the whole of Chapter 3 being devoted to it? Firstly, most
nonlinear crystals are birefringent, and so one naturally needs to know how light propagates
in these media. Secondly, several important nonlinear optical techniques (the most obvious
being phase matching) exploit birefringence to achieve their goal. Lastly, the material in
this chapter provides essential background for the following chapter on the nonlinear optics
of crystals.
Section 3.2 is a brief tutorial on crystal symmetry. Crystallography is something of a
world on it own, and many people find it a complete mystery. Although the summary
offered here is very basic, it should provide everything needed for what comes later.
Section 3.3 discusses the propagation of EM waves in optically anisotropic media, and
contains a fairly detailed analysis of birefringence (double refraction), ordinary and extraordinary waves, and associated topics. The treatment is mainly centred on uniaxial media
because of their relative simplicity and the fact that most nonlinear crystals are of this type.
Section 3.4 describes how birefringence can be exploited in the construction of wave
plates, while Section 3.5 is reserved for a brief mention of biaxial media in which the
propagation characteristics are considerably more complicated.
3.2 Crystal symmetry

Two questions about a crystalline medium are of critical importance in nonlinear optics:
(1) Is the crystal centrosymmetric or non-centrosymmetric? In other words, does it or does
it not possess inversion symmetry?
(2) Is the crystal optically isotropic or anisotropic? Does it, in other words, exhibit
birefringence (double refraction)?
45
If the crystal is non-centrosymmetric, second-order nonlinear optical effects can of course

be observed. Because the second-order tensor governing the nonlinear electromechanical
(2)
phenomenon of piezoelectricity has identical symmetry properties to ij k in Eq. (3.1), all
Crystal optics
46
second-order nonlinear crystals are also piezoelectric.1 If, on the other hand, the medium is
centrosymmetric (i.e. it possesses inversion symmetry), all even-order nonlinear processes
are forbidden, and so one is restricted to observing odd-order effects. A simple explanation
of this principle was given in Chapter 1. A slightly more general version is based on the
tensor/matrix form of Eq. (1.29), in which the second-order polarisation reads
(2)
Pi = 0
ij k Ej Ek (i, j , k = x, y, z).
(3.1)
jk
If the medium possesses inversion symmetry, the polarisation must change sign if all components of the field change sign. But if both fields change sign in Eq. (3.1), the right-hand
side is unchanged. So the polarisation must be zero, which can only be true if all 27 elements
(2)
of ij k are zero. The way this conclusion arises in a quantum mechanical calculation of
the coefficients is discussed in Chapter 8. The argument obviously applies equally to any
even-order term in the polarisation of order higher than 2.
We turn now to the second question at the start of this section. As we shall discover
shortly, most common nonlinear optical crystals are both non-centrosymmetric and optically
anisotropic, but it is important to understand that there is no fundamental connection between
the two properties. Some crystals have one property without the other, some have both, and
some neither. An optically isotropic crystal will not of course exhibit birefringence and so,
even if it is non-centrosymmetric and second-order nonlinear processes are allowed, it will
not be possible to employ birefringent phase matching (BPM). But there are of course other
methods of phase matching (e.g. quasi-phase matching).
Every crystalline medium belongs to one of 32 point symmetry classes, which are grouped
into seven symmetry systems as shown in Table 3.1 [24,25]. Materials in six of the seven
systems (containing 27 of the 32 classes) are optically anisotropic, while those in the seventh
(the cubic system, containing five classes) are optically isotropic.2 The seven crystal systems
are referred to by name, and the 32 classes by point group.3 The classes are sometimes
referred to by class number, but there is more than one numbering scheme, so caution is
necessary. The numbers in Table 3.1 are those used by Cady [25].
The data in column 3 of the table indicate whether or not a crystal possesses inversion
symmetry. It can be seen immediately that the presence of birefringence says nothing about
whether or not a crystal is non-centrosymmetric. For example, copper sulphate (class 2) is
biaxial (the more complicated kind of anisotropy), yet it possesses a centre of symmetry.
On the other hand, gallium arsenide (class 31) is optically isotropic, but lacks inversion
symmetry.
1 Indeed, the word piezoelectric is sometimes used as a synonym for non-centrosymmetric.
2 The word isotropic meaning the same in all directions, carries two distinct meanings in optics and crystal-
lography depending on whether the reference is to optical isotropy or to structural isotropy. Gases, liquids and
amorphous solids are structurally isotropic, but crystals are by definition structurally anisotropic. On the other
hand, media that do not exhibit birefringence are said to be optically isotropic, and these include all crystals in
classes 2832 in Table 3.1. But since they are crystals, none is structurally isotropic. In the absence of external
perturbations, all structurally isotropic media are optically isotropic too.
3 Each point group possesses particular symmetry properties with respect to a point. Fortunately, it is not necessary
to have a detailed understanding of the point group codes listed in HermannMauguin (international) notation in
the second column of Table 3.1. Readers are referred to [6] or [7], or to other specialist texts on crystallography.
47
Table 3.1 The 32 crystal classes.

System/Class No.
Symmetry code
Biaxial crystals
Triclinic system
1
1
2
1
Monoclinic system
3
2
4
m
5
2/m
Orthorhombic system
6
222
7
mm2
8
2/m 2/m 2/m
Uniaxial crystals
Tetragonal system
9
4
10
4
11
4 2 m
12
422
13
4/m
14
4mm
15
4/m 2/m 2/m
Trigonal system
16
3
17
3
18
32
19
3m
20
3 2/m
Hexagonal system
21
6
22
6 2 m
23
6
24
622
25
6/m
26
6mm
27
6/m 2/m 2/m
Optically isotropic crystals
Cubic system
28
23
29
432
30
3m = 2/m 3
31
4 3 m
32
4/m 3 2/m = m3m
Inversion sym.
Examples
no
yes
Copper sulphate
no
no
yes
no
no
yes
no
no
no
no
yes
no
yes
no
yes
no
no
yes
no
no
no
no
yes
no
yes
no
no
yes
no
yes
LBO, KTP, KTA
KDP, ADP, CDA

Nickel sulphate
Sodium periodate
-quartz
BBO, Lithium niobate
Calcite
Gallium selenide
Lithium iodate
-quartz
Cadmium selenide
Sodium chlorate
Pyrite
Gallium arsenide, zinc blende
Sodium chloride, diamond
Crystal optics
48
Table 3.2 Crystal classes with key properties summarised.

Non-centrosymmetric
Centrosymmetric
All
Anisotropic
Isotropic4
All
18
9
27
3
2
5
21
11
32
The statistics summarised in Table 3.2 indicate that 18 of the 32 crystal classes are both
non-centrosymmetric and optically anisotropic. Second-order nonlinear processes (such as
second harmonic generation) are allowed, and birefringent phase matching can be used to
enhance them. As noted earlier, most common nonlinear materials fall into this category,
and a number of these are listed against their class in the right-hand column of Table 3.1.
Most common nonlinear crystals happen to belong to classes 7, 11 or 19.
The shape and size of a crystals unit cell is defined by three vectors a, b and c, and the
associated abc-axes are known as the crystallographic axes. The abc-axes are orthogonal
in the orthorhombic, tetragonal, and cubic systems, but are non-orthogonal in the others. A
right-handed set of orthogonal xyz-axes, known as the principal dielectric axes (or sometimes the physical axes), is invariably used in addition, but unfortunately the conventions
for relating these to the abc-axes are not as consistent as one might wish. Even when the
crystallographic axes are themselves orthogonal, it cannot necessarily be assumed that abc
maps to xyz. In the orthorhombic system, for instance, the translation might be bac xyz.
Reference works such as Dmitriev et al. [26] should be consulted for detailed information
of this kind. We shall return to this confusing feature in Chapter 4.
Interestingly, the geometric structure of a crystal (i.e. the location of the atoms within
the lattice) does not in itself determine its symmetry properties. For example, the geometry
of gallium arsenide (class 31; see Fig. 3.1) is identical to that of diamond (class 32), but
diamond possesses inversion symmetry while gallium arsenide does not. The reason is that
the centre of symmetry in diamond lies mid-way between two nearest neighbours, both
of which are of course carbon atoms in that case. But nearest neighbours in GaAs are
always gallium and arsenic, and since these are different atoms, the inversion symmetry of
diamond is lost [27]. Conversely, crystals of the same point symmetry group may have a
wide variety of different internal geometries. For example, diamond and sodium chloride
are both members of class 32, but the arrangement of atoms in the unit cell is different.
3.3 Light propagation in anisotropic media

3.3.1 Maxwells equations
We now discuss how birefringence originates, and how light propagates in birefringent
media (see also [28,29]). As the first step, we undertake a simple exercise in electromagnetic
4 In this context isotropic refers to optical isotropy. See footnote 2 on page 46.
49
Fig. 3.1
The face-centred cubic structure of GaAs. Ga and As have equivalent status in the lattice so, if each black dot
represents Ga, each grey dot represents As (or vice versa). Every atom lies at the centre of a tetrahedron with atoms of
the other species at the vertices. If the cube in the gure is divided into eight cubes of half the side length, then four of
the eight contain tetrahedra. The stubs indicate bonds to atoms in adjacent cubes. Diamond has the same geometrical
structure, but all atoms are carbon.
(EM) wave propagation to determine the constraints imposed by Maxwells equations on
the directions of the four vectors E, D, B and H. In most of what follows, we consider
plane EM waves of angular frequency propagating in the direction of the unit vector s.
The space-time dependence is then of the form exp i(t k.r) where the wave vector
k (= ks) is perpendicular to the wavefront, which is the plane of constant phase. To make
the argument simpler (but with no loss of generality), we will temporarily assume that k
points in the +z-direction. In this case, the four vectors depend only on z through their
space-time dependence exp i(t kz). In a non-magnetic medium (in which B = 0 H),
containing no free charges, Maxwells equations read
div D = 0;
div B = 0;
curl E = B/t;
curl B = D/t.
(3.2ad)
Three simple deductions can now be made:

Since B (like the other three field vectors) is independent of x and y, Eq. (3.2b) reduces
to Bz /z = 0, which can only be true if Bz = 0; hence B and H lie in the x-y plane,
which is the plane of the wavefront.
Applying the same argument to Eq. (3.2a) leads to a similar conclusion for D, which
therefore lies in the plane of the wavefront too.
If for convenience we choose the y-axis to lie along B and H, the only non-zero term on
the left-hand side of Eq. (3.2d) is xBy /z, and it follows that D lies along the x-axis;
hence B, H and D all lie in the plane of the wavefront, with D perpendicular to B and H.
50
Crystal optics
B,H
s,k
N = ExH
Fig. 3.2
Field vector directions in a non-magnetic medium as prescribed by Maxwells equations. D, B and H lie in the plane of
the wavefront, while s and k are normal to the wavefront. E is perpendicular to B and H, and so is coplanar with D, s
and k. But E is not normally parallel to D in an anisotropic medium, so the Poynting vector N = E H is not usually
perpendicular to the wavefront. The angle between N and k is the walk-off angle.
It remains to consider the direction of E, which is restricted by Eq. (3.2c). Since the only
non-zero terms in curl E are xEy /z+ y Ex /z, and since B only has a y-component,
it follows that Ey = 0. Hence E is perpendicular to B, but Maxwells equations impose
no other constraint.
The directions of the four field vectors are therefore as shown in Fig. 3.2. The vectors
D, B (and H), and k are mutually perpendicular, while E is coplanar with D and k, but
its direction is otherwise not determined. These conclusions are in no way affected by
the assumption that k points along the z-axis, which was merely a convenient choice that
led quickly to the general conclusion. From here on, the assumption is withdrawn and, in
subsequent figures, the direction of k is arbitrary once again.
The direction in which light is polarised is related to the directions of the vectors in
Fig. 3.2. As mentioned in Chapter 1, the word polarisation is used in optics in two senses:
in relation to the polarisation of light, on the one hand, and to the polarisation of matter,
on the other. The vector P representing the macroscopic electric dipole moment per unit
volume of a medium is of course the polarisation in its second sense. Since D = 0 E + P,
the three vectors are coplanar.
Most modern authors associate the polarisation of light (i.e. the direction in which it is
polarised) with the direction of the electric field vector E. However, Born and Wolf [29]
comment that it has been more common in the history of optics to associate the polarisation of
light with the direction of the magnetic field! To avoid ambiguity, they speak of the direction
of vibration or the plane of vibration, taking care to specify the vector concerned in all
cases.
In this book, we define the plane of polarisation of light as the plane perpendicular to B
and H, and containing the four vectors D, E, P and k, which is drawn as a horizontal plane
in Fig. 3.2.
51
3.3.2 The constitutive relation

If Maxwells equations do not define a unique direction for E, what does? The answer
is the so-called constitutive relation of the medium, which determines the magnitude and
direction of E for a given D, and vice versa. In its simplest form, this relation is simply
D = 0 E
(3.3)
where the scalar = 1 + (1) is called the dielectric constant, and (1) is the linear
susceptibility. However, this only applies in an optically isotropic medium. In an anisotropic
(non-isotropic) material, the constitutive relation reads
D = 0 E
(3.4)
where is now the dielectric constant tensor. Written in matrix form, Eq. (3.4) reads
DX
XX XY XZ
EX
DY = 0 Y X Y Y Y Z EY
(3.5)
DZ
ZX
ZY
ZZ
EZ
in which the vectors have become column matrices and the (second-rank) tensor has become
the square dielectric constant matrix. Capital letters have been used to define an arbitrary set
of orthogonal axes; lower case letters will be reserved for the principal axes to be introduced
shortly.
The key message of Eqs. (3.4) and (3.5) is that D and E are not necessarily parallel in
an anisotropic material, and the tensor/matrix machinery is needed to specify the precise
relation between them. The matrix in Eq. (3.5) makes it look as if nine numbers are needed
for this purpose but, fortunately, only three are sufficient even for a medium of the most
general symmetry. For a start, it is easy to show that the dielectric constant matrix is
symmetric, and consequently that a special set of orthogonal axes known as the principal
dielectric axes always exists with respect to which the matrix is diagonal. We shall write
these axes xyz. A mathematician would say that the dielectric constant matrix can be
diagonalised in a similarity transformation. With respect to the principal axes, Eq. (3.5)
becomes
Dx
xx
0
0
Ex
Dy = 0 0 yy 0 Ey
(3.6)
Dz
0
0 zz
Ez
where the three diagonal elements are the three principal dielectric constants of the medium.
Note that the principal axes are not the same as the crystallographic axes (abc) which, as
noted earlier, are not necessarily even orthogonal. More information about the relationship
between the two sets of axes in a given crystal can be found in the next chapter.
From Eq. (3.6), there are clearly three different types of material. Media in which all
three dielectric constants are different are called biaxial, those in which two of the three
are the same (by convention xx = yy ) are called uniaxial, and when all three are the
same, Eq. (3.6) reverts to Eq. (3.3), which is the isotropic case. Optically anisotropic media
therefore divide into biaxial and uniaxial materials, represented respectively, by classes 18
and 927 in Table 3.1 Members of classes 2832 are optically isotropic.
Crystal optics
52
We can now begin to understand some of the characteristic features of EM wave propagation in anisotropic media. The conditions imposed by Maxwells equations are shown
in Fig. 3.2, while the missing condition determining the direction of E for a given D is
provided by Eq. (3.6). A little thought suggests however that these conditions will normally
conflict. What happens, for example, if the constitutive relation specifies a direction of E
that is not perpendicular to B and H, as it almost invariably does? The answer is that such a
wave will not propagate. Indeed, it turns out that, for an arbitrary direction of k, Maxwells
equations and the constitutive relation are simultaneously satisfied for only two directions
of D. These directions are mutually orthogonal, and define the planes of polarisation of
the two allowed EM wave solutions. A wave entering the medium polarised in any other
direction will divide into two orthogonally polarised components, which will propagate
according to different rules.
3.3.3 The index ellipsoid: ordinary and extraordinary waves

The characteristics of anisotropic media can be displayed pictorially in several different
ways. The first of the two representations used in this chapter is called the index ellipsoid
(or the wave normal ellipsoid), and is defined by the formula
x2
y2
z2
+
+
=1
xx
yy
zz
(3.7)
y2
z2
x2
+ 2 + 2 = 1.
2
nx
ny
nz
(3.8)
or equivalently
In Eq. (3.8), the three principal dielectric constants have been replaced by the three
associated refractive indices, the significance of which will emerge shortly.
In the general (biaxial) case, the three semi-axes of the ellipsoid nx , ny and nz are all
different. However, to keep things simple, we will confine ourselves mostly to uniaxial
media in which nx = ny = no (the ordinary refractive index) and nz = ne = no (the
extraordinary refractive index). In this case, Eq. (3.8) reduces to
z2
x2 + y2
+
= 1.
n2o
n2e
(3.9)
The x and y semi-axes are now of equal length, the x-y section is circular (so the ellipsoid
technically becomes a spheroid), and the z-axis is known as the optic axis. There are now
two distinct cases depending on which index is the larger. If ne > no , the spheroid is
prolate (like a rugby ball) and the medium is termed positive uniaxial; on the other hand,
if ne < no , the spheroid is oblate (more like a pumpkin) and the medium is negative
uniaxial. Most crystals used in nonlinear optics happen to be negative uniaxial, and the
form of the spheroid in this case is shown in Fig. 3.3.
Detailed analysis [28, 29] reveals that sections of the spheroid through the origin O have
a particular significance in determining the properties of EM wave propagation. Consider
a wave for which the vector k is directed at an arbitrary angle to the z- (or optic) axis.
53
z (OPTIC AXIS)
n~e ()
ne
no
Fig. 3.3
The index ellipsoid for a negative uniaxial medium. The section in the x-y plane is a circle, so the object is technically a
spheroid. The shaded ellipse is perpendicular to k.
The plane perpendicular to k and parallel to the wavefront (shaded in the figure), generally
cuts the spheroid in an ellipse, except in the special case when = 0 when the section is
circular. This ellipse has several important properties:
(a) The axes of the ellipse define the two orthogonal directions of D that simultaneously
satisfy Maxwells equations and the constitutive relation. One of the two axes (the major
axis in the negative uniaxial case, shown dash-dotted in the figure) always lies in the
x-y plane. It corresponds to the direction of polarisation of the ordinary wave, and has
the same length irrespective of the direction of k. The other axis (shown dotted) relates
to the extraordinary wave, and its length depends on the angle between k and the
z-axis.
(b) The lengths of the semi-axes of the ellipse are the refractive indices of the associated
waves, namely no for the ordinary wave, and n e ( ) for the extraordinary wave. The
value of n e ( ) is easily deduced from Fig. 3.4. This shows the projection of the index
ellipsoid onto the kz plane or, equivalently, the view along the dash-dotted line in
Fig. 3.3. The length of the bold line perpendicular to k in Fig. 3.4 is the value of n e ( )
and, since the section of a spheroid is necessarily an ellipse, it is easy to show that

n e ( ) =
cos2
sin2
+
n2o
n2e
1/2
.
(3.10)
Notice that n e (90 ) = ne and, for propagation along the optic axis, n e (0 ) = no , in
which case the refractive index for all polarisations is the same.
Crystal optics
54
z (OPTIC AXIS)
k
N
~
ne ()
Fig. 3.4
Projection of the index ellipsoid onto the k-z plane.

(c) Where the axes of the ellipse (representing the directions of D) intersect the spheroid,
the normals to the surface define the corresponding directions of E. For ordinary waves,
D and E are parallel and lie in the x-y plane, and this wave propagates in the usual way.
However, the two vectors are generally not parallel for the extraordinary wave (see Fig.
3.4), which leads to peculiar propagation properties discussed in Section 3.3.5 below.
The fact that the extraordinary refractive index ne is not the same as the refractive index
of the extraordinary wave given by n e ( ) in Eq. (3.10) is a perennial source of confusion. It
is vital to distinguish between the two. The parameter n e ( ) is angle-dependent and varies
between no at = 0 and ne at = 90 ; on the other hand, ne is the maximum value of
n e ( ) for positive uniaxial media (ne > no ), and the minimum value of n e ( ) for negative
uniaxial media (ne < no ). Values of no and ne as a function of wavelength are quoted for
different materials in the literature.
3.3.4 Index surfaces

An alternative method of displaying the properties of an anisotropic medium now presents
itself. For every direction of the propagation vector k, we measure out the values of no
and n e ( ), creating a double surface of rotation comprising a sphere of radius no for the
ordinary wave, and a spheroid formed by rotating the ellipse of Eq. (3.10) about the optic
axis for the extraordinary wave.5 For a positive uniaxial medium, the spheroid lies outside
5 Instead of creating the double surface from the two effective refractive indices for each direction of propagation,
Yariv and Yeh [28] plot the components of k; their diagram, which they call the normal surface, is therefore a
scaled version of ours. On the other hand, Born and Wolf [29] construct what they also call the normal surface
by plotting the two phase velocities in each direction. Their diagram is therefore based on the inverse refractive
indices.
55
z
D
k
N
~
ne ()
no
Fig. 3.5
Index surfaces for a negative uniaxial medium.

the sphere; for a negative medium, it lies inside. The two surfaces touch along the optic axis.
A section through the index surfaces for a negative uniaxial medium is shown in Fig. 3.5.
Careful thought shows that the inner extraordinary surface is simply the ellipse of Fig. 3.4
rotated by 90 . Notice that the bold line representing n e ( ), which was perpendicular to k
in Fig. 3.4, lies along k in Fig. 3.5. The radius of the circle outside the ellipse corresponds
to the ordinary refractive index no .
The fact that Figs 3.3 and 3.5 both involve spheroids, circles and ellipses is another
potential source of confusion. It is worth repeating that the index ellipsoid of Fig. 3.3 is a
single surface object from which the refractive indices of the two orthogonally polarised
waves in a given direction of propagation can be obtained by a geometrical construction.
The double-layered surface of Fig. 3.5, on the other hand, traces those two refractive indices
for any given direction of propagation. The plane of polarisation of the waves can also be
deduced from the figures. You should convince yourself that the ordinary wave is polarised
perpendicular to the plane of the paper in Figs 3.4 and 3.5, while the extraordinary wave is
polarised in the plane of the paper.
3.3.5 Walk-off
The fact that E and D are generally not parallel for an extraordinary wave means that the
wave exhibits peculiar propagation properties. According to EM theory, the direction of
energy flow in an EM wave is given by the Poynting vector N = E H. Hence, if E is not
parallel to D, the energy does not flow in the same direction as the normal to the wavefront
defined by k, but walks off to the side as indicated in both Figs 3.4 and 3.5.
As mentioned in Section 3.3.3, E lies along the normal to the ellipse of Fig. 3.4. The walkoff angle, (the angle between D and E and equivalently between N and k), is therefore
Crystal optics
56
ord
ext
Fig. 3.6
Walk-off of an extraordinary wave, showing the walk-off angle .

the angle between the radial line from the origin to the ellipse, and the normal to the ellipse
at that point. Some tricky trigonometry leads to the formula

() = tan1 {(no /ne )2 tan }
(3.11)
where the upper and lower signs apply to negative and positive crystals, respectively. Positive means that the Poynting vector points away from the optic axis (as in Fig. 3.5). The
walk-off angle is typically a few degrees; see Problem 3.5.
Figure 3.6 contrasts the propagation of an ordinary and an extraordinary wave with
parallel wavefronts. It highlights the way in which the energy in an extraordinary wave slips
sideways, so that the extraordinary beam ultimately separates completely from the ordinary
beam. The effect is similar to the behaviour of an aircraft attempting to fly (say) north in
a strong easterly cross-wind. Though the plane is flying on the correct northerly bearing,
the sideways air flow pushes it to the right of its intended flight path. Some supermarket
trolleys exhibit a similar property, although the problem there is wear and tear!
3.4 Wave plates

The principle behind quarter- and half-wave plates can now be understood. Consider the
propagation of ordinary and extraordinary waves in the plane normal to the optic (z-) axis,
where = 90 . To simplify the discussion, we assume that the direction of propagation is
along the y-axis which points into the paper in Fig. 3.7. Suppose that a wave that is linearly
polarised at 45 to the x- and z-axes (along the line PP) is incident on the entry surface of the
sample at y = 0. The wave must be resolved into two equal components, an ordinary component (refractive index no ) polarised in the x-direction, and an extraordinary component
(refractive index ne ) polarised in the z-direction. The respective electric fields are
3.5 Biaxial media
57
z
H
Fig. 3.7
Change in polarisation direction from PP to HH in a half-wave plate.

A
Ex = cos{(t no y/c)}
2
A
Ez = cos{(t ne y/c)}
2
(3.12)
where A is the amplitude of the original field. Although the two components are in phase at
y = 0, a phase difference develops during propagation given by (y) = 2 yn/0 , where
0 is the vacuum wavelength and n = no ne . After propagating a distance y = 0 /4|n|,
the components are /2 out of phase (technically they are said to be in quadrature), and
careful inspection will convince you that the total field is now circularly polarised (see
Problem 3.3). An element that creates circularly polarised light in this way is called a
quarter-wave plate.
The phase difference continues to increase under further propagation, and reaches at
y = 0 /2 |n|. The components are now in anti-phase, so that Ex = Ez . This implies
that the total field is once again linearly polarised, but at 90 to the original orientation,
as indicated by the dotted line HH in Fig. 3.7. A device with these properties is called a
half-wave plate.
Quarter- and half-wave plates are used routinely for manipulating the polarisation of
optical beams. Bear in mind that, because the refractive indices are wavelength dependent,
wave plates have to be specified for a particular wavelength.
3.5 Biaxial media

3.5.1 General features
Professor Arthur Schawlow, one of the inventors of the laser and a great humourist too,
liked to define a diatomic molecule as a molecule with one atom too many. In the same
Crystal optics
58
spirit, one might define a biaxial crystal as a crystal with one axis too many, because the
optical properties of biaxial media are much more complicated than those of their uniaxial
counterparts. Happily, most common nonlinear materials are uniaxial, but several members
of biaxial class 7 (e.g. LBO and KTP) are in widespread use, so one cannot escape the topic
of biaxial media altogether.
In a biaxial medium, the semi-axes of the index ellipsoid, which represent the values of
nx , ny and nz in Eq. (3.8), are all different, and the principal axes are always chosen so that ny
is intermediate between nx and nz .6 Most of the ideas set out in Section 3.3.3 for finding the
polarisation of waves propagating in a particular direction are unchanged in biaxial crystals,
but the details are more complicated than before. Two waves with orthogonal directions
of D still propagate for every value of k but, except in special cases, both solutions have
extraordinary wave characteristics, with the Poynting vector angled to k and associated
walk-off effects. Moreover, these media are called bi-axial rather than uni-axial because
they have two optic axes rather than one.
3.5.2 Propagation in the principal planes: the optic axes

Fortunately, biaxial crystals are almost invariably used in nonlinear optics with all the
interacting waves in one of the principal planes. The propagation characteristics are then
relatively straightforward, and similar to those in uniaxial media.
The index surfaces of a biaxial crystal with nz > ny > nx are represented in Fig. 3.8,
where sections of the double surface in the principal planes are shown. The three solid
lines are quadrants of circles; the three dotted lines are quadrants of ellipses. Diagrams of
nx
nz
ny
z
k
Fig. 3.8
Index surfaces for a biaxial medium showing sections in the principal planes. Solid lines indicate circular quadrants
and dotted lines elliptical quadrants.
6 Some highly respected sources (e.g. Born and Wolf [29] and Yariv and Yeh [28]) choose principal axes to ensure
that nz > ny > nx . While this is certainly a possible convention in principle, leading reference works such as
Dmitriev et al. [26] quote the indices of some biaxial crystals (e.g. KNbO3 and KB5) with nx > ny > nz .
3.5 Biaxial media
59
this kind can be quite confusing, so it is worth spending a little time understanding how
to interpret this one. Propagation along one of the principal axes is the simplest case. The
figure indicates that for propagation along (for example) the z-axis, the two indices are nx
and ny (> nx ); these are the radii of the two solid circles touching the z-axis. The waves are,
of course, polarised transversely to the z-direction, in the x and y directions respectively.
The D and E vectors are parallel in this special case, so walk-off effects do not occur, and
both solutions can be regarded as quasi-ordinary waves.
When k is not aligned with one of the principal axes, but still lies in one of the three
principal planes, the situation is straightforward too, because there is a close analogy with
uniaxial media. Under these circumstances, one of the two solutions is always a quasiordinary wave; for any direction in the x-z plane, for instance, one wave is always polarised
in the y-direction and has index ny . A similar situation applies in the x-y and y-z planes.
However, the choice of nx and nz (in either order) as the highest and lowest of the three
refractive indices means that the x-z plane is different from the other two. This is evident
from Fig 3.8, which shows that the two index surfaces intersect in the x-z plane. The line
through the intersection (drawn dashed in Fig. 3.8) defines one of the two optic axes of a
biaxial medium, and the other one lies symmetrically on the other side of the z-axis in the
x-z plane, as shown in Fig. 3.9.
The directions of the optic axes are easily found. Consider a wave propagating in the x-z
plane with wave vector k at an angle to the z-axis. We already know that the y-polarised
wave has index ny for all , and it is easy to show that the index of the other wave rises
from nx at = 0 to nz at = 90 , varying as it does so according to the formula

n(
)=
cos2
sin2
+
n2x
n2z
1/2
(3.13)
z
k
nx
x
ny
nz
2
Fig. 3.9
The x-z sections of the index surfaces of a biaxial medium, showing the two optic axes angled at ; to the z-axis.
60
Crystal optics
which is analogous to Eq. (3.10). The directions of the optic axes can now be found from
Eq. (3.13) by setting n(
) = ny . Both waves then have the same refractive index and the
same phase velocity, which is the defining characteristic of an optic axis. The angles of the
optic axes to the z-axis, conventionally written ; (or Vz ), are readily found from Eq. (3.13)
to be
!
n
2 n2
n
z
y
x
; = cos1
.
(3.14)
ny n2z n2x
Notice from this equation that when nx = ny (the case of a uniaxial medium), the two axes
merge into the single axis of a uniaxial medium along the z-axis, i.e. ; = 0.
It is a surprising fact that the two optic axes of a biaxial crystal are the only directions for
which the two effective refractive indices are the same [29]. This is strange because, given
that the optic axis in Fig. 3.8 lies at the intersection of the two index surfaces, one would
expect there to be other directions close to the optic axis (but out of the x-z plane, indicated
by the grey double arrow) where the surfaces would also intersect and the indices were also
identical. After all, two planes normally intersect in a line, not at a point. The answer to this
conundrum is that the index surfaces do not behave as simple surfaces, and it turns out that
the surface intersection is avoided (in the mathematical sense of an avoided crossing) for
all directions other than the two optic axes.7
3.5.3 Positive and negative crystals

As in the uniaxial case, biaxial crystals are classified into positive and negative types,
although the criterion is unfortunately not immediately obvious. The rule is that, if 2; is
acute, a biaxial medium is positive if nz > nx and negative if nx > nz . The designations
are reversed if 2; is obtuse. There is a logic to this apparently arbitrary system insofar as
positive biaxial media are those in which the optic axes are closer (an acute angle between
them) to the axis of the index ellipsoid exhibiting the highest refractive index. This matches
the designation in uniaxial media (with a single optic axis along the z-axis), where the
crystal is positive uniaxial if nz ne > no .
Problems
3.1 Express the elliptical x-z section of Eq. (3.9) in polar form. How does it differ from
Eq. (3.10)?
3.2 Consider a wave propagating along the x-axis of a uniaxial medium. In what directions are the two allowed wave solutions polarised? What are the respective refractive
indices? Do any walk-off effects occur in this case?
7 The fact that there really are just two directions for which the indices are the same follows also from a geo-
metrical property of an ellipsoid. It can be shown that an ellipsoid with unequal axes has only two circular
sections through its centre, which are perpendicular to the plane containing the largest and smallest semi-axes.
The two optic axes are normals to these circular sections.
61
3.5 Biaxial media
3.3 A wave propagates for a distance /4 |n| in the +y-direction of a positive uniaxial
crystal (ne > no ); see Fig. 3.7. Is the resulting circularly polarised wave polarised in a
clockwise or an anticlockwise direction (looking in the direction of wave propagation)?
3.4 Prove Eq. (3.14).
3.5 Beta-barium borate (BBO) is used in a second harmonic generation experiment. The
fundamental wavelength is = 1.064 m, and the angle between the direction of
propagation and the optic axis is 22.8 . Find the walk-off angle.
(BBO refractive indices at 0.532 m: no = 1.67421, ne = 1.55490.)
Nonlinear optics in crystals

As noted in Section 1.8, the treatment presented in Chapter 1 was greatly over-simplified.
The fact that the frequency dependence of the coefficients in the polarisation expansion
was neglected gave the false impression that the coefficients governing all processes of a
given order are the same, apart from simple factors. The tensor nature of the coefficients
was completely ignored too.
Unfortunately, if one wants to understand nonlinear optical interactions in crystalline
media, one cannot avoid getting to grips with the tensor nature of the nonlinear coefficients,
a topic that is intricate and hard to simplify. At the very least, one needs to be able to interpret
the numerical notation used to label the coefficients, and to know how to apply the data
supplied in standard reference works in a given crystal geometry.
There is no disguising the fact that Section 4.3 in the present chapter is rather complicated.
Readers who want to avoid the worst of the difficulties should skim it, always bearing in
mind that the situation turns out to be far less alarming at the end of the journey than it
seemed it might be at the beginning. Early on in that section, it looks as if there could
be literally hundreds of separate nonlinear coefficients to deal with. But it soon emerges
that the number is almost certainly no larger than 18, and perhaps only 10. And even this
worst-case scenario applies only in crystals of the triclinic class, of which there are no
members in regular use in nonlinear optics. Indeed, in most common crystals, the number
of independent coefficients is typically only two or three and, for a given geometry, there
may be only one.
4.2 The linear susceptibility

We start by writing the relationship between the polarisation and the field in linear optics
in its general form. The defining equation is1
(1)
Pi () = 0
ij ()E j () (i, j = x, y, z)
(4.1)
j
1 The summation sign may be omitted because, under the repeated suffix convention, summation is implied over
any repeated suffix on the right-hand side.
62
4.3 Structure of the nonlinear coefcients
63
(1)
where ij () is the frequency-dependent matrix of the linear susceptibility (the component

form of the associated tensor), and Pi () and E i () are the (complex) amplitudes of quasimonochromatic terms in the summations of Eqs (2.4) and (2.5). Since the indices i and j
can each be x, y or z, Eq. (4.1) is shorthand for three equations.
When combined with D = 0 E + P, Eq. (4.1) becomes
(1)

D i () = 0 E i () + 0
ij ()E j () = 0
ij ()E j ()
(4.2)
j
where ij () is the dielectric constant matrix from Section 3.3.2. From Eq. (3.6), we know
that, with respect to the principal dielectric axes, only the three diagonal elements of ij ()
are non-zero and, from the subsequent discussion, that these elements are the squares of the
associated refractive indices. The frequency dependence of Eq. (4.2) is therefore simply the
well-known phenomenon of dispersion.
A more detailed discussion of Eq. (4.1) can be found in Appendix B. The connection
between the frequency dependence of the susceptibility and the non-instantaneous response
of the polarisation to the field is established, and it is shown that the equation is in fact a
hybrid form, with some time-domain and some frequency-domain characteristics.

4.3.1 Formal definition
We now turn our attention to the structure of the second-order nonlinear coefficients, which
is the central topic of this chapter. For the second-order nonlinear polarisation governing
the three-wave interaction 1 + 2 = 3 of sum frequency generation (SFG), we write by
analogy with Eq. (4.1) the expression2

Pi (3 ) = 12 0
ijSFG
k (3 ; 1 , 2 )Ej (1 )Ek (2 )
p
= 12 0
jk

jk
ijSFG
k (3 ; 1 , 2 ) Ej (1 )Ek (2 )

SFG
+ikj
(3 ; 2 , 1 )E j (2 )E k (1 )
where 1 = 2 , and the suffices i, j and k can each be x, y, or z. The symbol
(4.3)
%
p
in the
first line of the equation implies that the terms with 1 and 2 permuted must be included,
as shown explicitly in the second line. The pre-factor of 12 is a consequence of the field
definitions of Eqs (2.4) and (2.5).
Equation (4.3) merits detailed discussion. Notice first of all that the nonlinear coefficients
have three frequency arguments representing the frequency of the polarisation to the left of
2 A detailed discussion of Eq. (4.3) can be found in Appendix C.
64
the semicolon, and those of the two fields in the order that they appear to the right. Since 3 =
1 +2 , only two frequency arguments are strictly necessary, but it is conventional to quote
all three. As regards the coordinate suffices ijk, each of which can be x, y or z, the component
of the polarisation (i) comes first, followed by those of the fields ( jk) in the appropriate
%
order. Under the repeated suffix convention, the summation j k {rhs} is redundant, and
could have been omitted. Negative frequencies are associated with complex conjugate fields
and polarisations, and must be treated as distinct from their positive counterparts.
4.3.2 Intrinsic permutation symmetry

Equation (4.3) can be simplified once it is recognised that some pairs of terms on
SFG ( ; , )
the right-hand side are indistinguishable in their effect, for example xyz
3
1 2
SFG
Ey (1 )Ez (2 ) and xzy (3 ; 2 , 1 )Ez (2 )Ey (1 ). As explained in Appendix C, it is common practice in this situation to equate the two contributions, in other words to assume that
a coefficient is unchanged if the two field frequencies are reversed, together with the asso(SFG)
(SFG)
ciated coordinate indices, i.e. ij k (3 ; 1 , 2 ) = ikj (3 ; 2 , 1 ). This assumption
of intrinsic permutation symmetry (IPS) enables Eq. (4.3) to be abbreviated to

Pi (3 ) = 0
ijSFG
(4.4)
k (3 ; 1 , 2 )Ej (1 )Ek (2 )
jk
(2)
(2)
where ij k (3 ; 1 , 2 ) and ikj (3 ; 2 , 1 ) are both included in this expression. It must

be stressed that IPS is a convenience, not a fact that can be proved. The point is that only
the sum of the two coefficients can be measured, and so it is sensible to assume that the
individual components are the same.
Equation (4.4) needs to be adjusted when 1 = 2 , or when any of the frequencies is
zero. Further discussion of the issues involved, and a general formula covering all situations,
can be found in Appendix C. Here we simply quote results for the most important cases:
Second harmonic generation (3 = 2; 1 = 2 = )

Pi (2) = 12 0
ijSHG
k (2; , )Ej ()Ek ().
(4.5)
The Pockels effect (3 = ; 1 = , 2 = 0)

Pi () = 20
ijPEk (; , 0)E j ()Ek (0).
(4.6)
Optical rectification (3 = 0; 1 = , 2 = )

Pi (0) = 12 0
ijOR
k (0; , )Ej ()Ek ()
(4.7)
jk
jk
jk
where E j () = E j (). As explained in Appendix C, in all these equations the order of the
two frequencies after the semicolon is fixed by definition, and terms with the frequencies
65
in reverse order do not appear. Notice that the pre-factors are identical to those that arose
naturally in the simple treatment of Chapter 1.
4.3.3 Full permutation symmetry and the ManleyRowe relations

At this point, one is bound to feel daunted by the accelerating complexity of the analysis. Consider, for example, the case of sum frequency generation. It is already clear from
Eq. (4.4) that there is the potential for 27 separate nonlinear coefficients (since ijk can each
be x, y or z), and that is bad enough. But what about negative frequencies? Do we have to
(2)
deal with coefficients like xyz (3 ; 1 , 2 ) as well, which would double the number of coefficients to 54? And there is another problem too. In Chapter 2, we discovered
that a proper treatment of SFG involves three coupled-wave equations, but so far we have
only considered the coefficient driving the equation for the field at 3 . If the coefficients
involved in the other two equations have to be included as well, that could multiply the
number of coefficients by a further factor of 3, raising the total to 162.
Happily, the worry about the large number of coefficients turns out to be far less serious
that it appears at first sight. Firstly, reversing the signs of all participating frequencies turns
a coefficient into its complex conjugate. But in a lossless medium where the frequencies
of the fields are far from atomic and molecular resonances, the coefficients are real to an
excellent approximation, so reversing all the signs makes no difference.
Secondly, under loss-free conditions, a principle known as full permutation symmetry
can be invoked under which the nonlinear coefficients are unchanged if the three Cartesian
indices ijk are permuted together with their associated frequency arguments. This enables
one to write, for example,
(2)
(2)
(2)
xyz
(3 ; 1 , 2 ) = yzx
(1 ; 2 , 3 ) = zxy
(2 ; 3 , 1 )
(4.8)
where the signs have been adjusted to maintain the convention that the first frequency is the
sum of the other two. The three coefficients in Eq. (4.8) are those needed in the coupled-wave
equations for SFG. We recall that the ManleyRowe relations of Eqs (2.20) and (2.31) rely
on the three coefficients being the same, and it is entirely consistent that this is ensured by
the principle of full permutation symmetry, which only applies under loss-free conditions.
That brings us back to the 27 independent coefficients with which we started this section.
But we will shortly discover that, for second harmonic generation at least, the number is
actually only 18. And, better still, even that worst-case scenario never arises in practice
because all common nonlinear materials belong to symmetry classes in which most of the
coefficients are zero in any case.
4.3.4 Contracted suffix notation

We now introduce the commonly used contracted suffix notation for the nonlinear
coefficients. We start from Eq. (4.5) for SHG which reads (for j = k)

SHG
Pi (2) = 12 0 ijSHG
(4.9)
k (2; , ) + ikj (2; , ) Ej ()Ek () (j = k).
66
Table 4.1 Contracted sufx notation for uniaxial media.

i
n
jk
p
x
1
xx
1
y
2
yy
2
z
3
zz
3
yz, zy
4
xz, zx
5
xy, yx
6
The two coefficients in this equation are clearly indistinguishable, so we take them as equal
SHG = 2d
(under IPS) and define ijSHG
i(j k) , where the brackets around jk serve as
k = ikj
a reminder that the order is irrelevant.3 The first suffix i and the suffix pair jk are now
translated into numbers n and p (di(j k) dnp ) according to the prescription shown in
SHG = SHG = 2d and for these coefficients Eq. (4.9)
Table 4.1.4 Thus, for example xyz
14
xzy
yields the contribution
Px (2) = 20 d14 E y ()E z ().
(4.10)
(2)
Similarly, when the second two suffices are the same, we have for example yzz 2d23 ,
which leads to
Py (2) = 0 d23 E y2 ().
(4.11)
Combining everything into a single equation we have
Px (2)
d11

Py (2) = 0 d21
d31
Pz (2)
d12
d22
d32
d13
d23
d33
d14
d24
d34
d15
d25
d35
d16
d26
d36
E x2 ()
E y2 ()
E z2 ()
2E y ()E z ()
2E z ()E x ()
2E x ()E y ()
. (4.12)
From the form of the d-matrix, it is clear that the number of independent SHG coefficients
is now no more than 18, even for a crystal of the most general symmetry. In practice, the
number is normally far less, as we will shortly discover.
4.3.5 The Kleinmann symmetry condition

The notational contraction of Table 4.1 can be applied more generally if all the frequencies
involved in the interaction are small compared to the resonance frequencies of the medium.
3 The origin of the factor 2 in the definition of d is lost in the history of the subject. Note that when d appears in
the literature with three suffices, the contracted form is equal to the original separate coefficients, not to their
sum. So, for instance, d14 = dxyz (= d123 ) = dxzy (= d132 ). We could incidentally have defined 2di(j k) =

1 (2) + (2) in which case we would not have needed to equate (2) and (2) ; see Appendix C.
2
ij k
ikj
ij k
ikj
4 The mapping in Table 4.2 is not necessarily the same in biaxial media; see Table 4.4.
67
Dispersion can then be neglected, making the nonlinear susceptibilities independent of

frequency, in which case all coefficients with the same set of ijk indices become equal,
e.g. xyy = yxy = yyx , etc. In fact, this identity, which is called the Kleinmann symmetry condition (KSC), is often approximately true even when significant dispersion is
present.
Under the KSC, ij k is the same if j and k are interchanged, so the contracted notation
of Eq. (4.10) can now be used in the general three-wave case, not just for SHG. But the
KSC has the further consequence that the number of independent coefficients drops from
18 to 10, one for each of the sets xxx, xxy, xxz, xyy, xyz, xzz, yyy, yyz, yzz and zzz. Hence,
some of the elements in the 36 d matrix of Eq. (4.12) are now interrelated, namely
d11 d12 d13 d14 d15 d16

d11 d12 d13 d14 d15 d16
d21 d22 d23 d24 d25 d26 d16 d22 d23 d24 d14 d12
d31
d32
d33
d34
d35
d36
d15
d24
d33
d23
d13
d14
(4.13)
where the eight elements now linked to others are printed in bold. The equation for sum
frequency generation analogous to Eq. (4.12) now reads
Px (3 )
d11 d12 d13 d14 d15 d16

Py (3 ) = 20 d16 d22 d23 d24 d14 d12

d15 d24 d33 d23 d13 d14
Pz (3 )
E x (1 )E x (2 )
E y (1 )E y (2 )
Ez (1 )Ez (2 )

(4.14)
.
Ey (1 )E z (2 ) + E z (1 )E y (2 )
E z (1 )E x (2 ) + E x (1 )E z (2 )
E x (1 )E y (2 ) + E y (1 )E x (2 )

Invariably, crystal symmetry drastically reduces the number of non-zero nonlinear coefficients still further. The non-zero SHG coefficients for each of the 21 non-centrosymmetric
crystal classes, and the relationships between them, are detailed in Appendix D. The most
important conclusion is that the worst-case scenario of 18 independent coefficients only
occurs in triclinic class 1 crystals, no member of which is in regular use in nonlinear optics.
In fact, most popular nonlinear materials are in uniaxial classes 11 and 19, and biaxial
class 7; common members of these three categories are listed in Tables 4.2 and 4.3, and
the non-zero coefficients of the two uniaxial classes are shown in Table 4.4. The situation
is more complicated in biaxial class 7 because of the awkward relationship between the
physical and the crystallographic axes in these materials shown in column 6 of Table 4.3.
The implications of this latter point will be discussed later.
68
Table 4.2 Common uniaxial crystals.

Class 11 (4 2 m)
Potassium dihydrogen phosphate
Ammonium dihydrogen phosphate
Cesium dihydrogen arsenate
Class 19 (3 m)
Lithium niobate
-barium borate
Proustite
Abbreviation
Formula
Sign
KDP
ADP
CDA
KH2 PO4
NH3 H2 PO4
CsH2AsO4
negative
negative
negative
BBO
LiNbO3
-BaB2 O4
Ag3AsS3
negative
negative
negative
Table 4.3 Common biaxial crystals.

Class 7 (m m 2)
Lithium triborate
Potassium titanyl phosphate
Potassium niobate
Potassium pentaborate tetrahydrate
Potassium titanyl arsenate
Abbrev. Formula
Sign
Indices
LBO
KTP
negative
negative
negative
positive
positive
nx
nx
nx
nx
nx
LiB3 O5
KTiOPO4
KNbO3
KB5 O8 .4H2 O
KTiOAsO4
KB5
KTA
< ny
< ny
> ny
> ny
< ny
xyz =
< nz
< nz
> nz
> nz
< nz
acb
abc
bac
abc
abc
Table 4.4 Non-zero d-elements in class 11 and 19 crystals.

1
Class 11 (3 m): d14 = d25 ; d36

1
2
3
4
d14
Class 19 (4 2 m): d15 = d24 ; d16 = d21 = d22 ; d31 = d32 ; d33
1
2
= d16
= d16
= d15
3
d31
= d31
d33
= d14
d15
d36
d16
4.5 Second harmonic generation in KDP

Let us consider second harmonic generation by birefringent phase matching in a uniaxial
class 11 crystal such as KDP, which is a fairly straightforward case. The simplest scenario,
envisaged in Chapter 2, and shown in Fig. 4.1, is where two ordinary waves at combine to
create an extraordinary wave at 2, a process that is described by the shorthand o + o e
4.5 Second harmonic generation in KDP
69
z (optic axis)
k1
e
o
e
x
Fig. 4.1
Type 1 phase-matching geometry in a class 11 crystal. k1 is the wave vector of the fundamental, which is polarised in
the x-y plane (indicated by o-o) and is therefore an ordinary wave. The polarisation of the extraordinary second
harmonic wave (shown as e-e) is perpendicular to o-o and to k1 .
and called Type-I phase-matching. Phase matching requires that

no
n 2
e ( )
cos2
sin2
+
2
2
(n2
(n2
o )
e )
1/2
(4.15)
where the second step follows from Eq. (3.10), and is the angle between the direction of
propagation and the z-axis, as shown in the figure. Rearranging Eq. (4.15) yields
!
2
2
n2
(n2
e ) (no )
o
cos =
.
(4.16)
2
2
2
no (ne ) (no )2
2
For a fundamental wavelength of 1.064 m, no = 1.4938, n2
o = 1.5124 and ne =
1.4705, and if these values are inserted, Eq. (4.16) gives = 41.21 .
But this is not the end of the story, because it is still necessary to find the optimum value
of the azimuthal angle . From Table 4.4, the three elements d14 , d25 and d36 are non-zero
in KDP and, since the ordinary fundamental wave will be polarised in the x-y plane (see
the double arrow o-o in the figure), it is clear from Eq. (4.12) that the operative term in the
polarisation is
Pz (2) = 20 d36 E x ()E y ().
(4.17)
From Fig. 4.1, it can be seen that E x () = E o () sin and E y () = E o () cos , where
E o () is the field amplitude of the incident ordinary wave, while the second harmonic
polarisation transverse to the direction of propagation (represented by the double arrow e-e
70
in the figure) is Pz (2) sin . The conclusion is that

Pe (2) = 20 d36 E o ()2 sin sin cos
= 0 [d36 sin sin 2] E o ()2 = 0 deff E o ()2
(4.18)
where the final step defines the effective nonlinear coefficient deff for the ooe interaction
as deff = d36 sin sin 2. To ensure phase matching, must be close to the value set by
Eq. (4.16); the optimum value of is clearly 45 .
Notice that the role of the effective nonlinear coefficient is to link the polarisation of
the extraordinary wave to the field of the ordinary wave, in contrast to the individual d
coefficients which relate the Cartesian components of P and E.
Phase-matched SHG in KDP is also possible by combining an ordinary and an extraordinary fundamental wave to create an extraordinary second harmonic. The condition for this
oee interaction (Type II phase matching) is that no () + ne () = 2ne (2). This is satisfied
at = 59.13 , although it is not possible to write down an exact analytical formula in this
case. An intricate piece of three-dimensional geometry (see Problem 4.3) shows that [30]
(
)
Pe (2) = 0 12 (d36 + d14 ) sin 2 sin 2 E o ()E e () = 0 deff E o ()E e () (4.19)
where the effective nonlinear coefficient in this case is5 deff = 12 (d36 + d14 ) sin 2 cos 2.
4.6 Second harmonic generation in LBO

It is quite easy to treat the case of second harmonic generation in one of the principal planes
of a biaxial crystal such as lithium triborate (LBO), once a potentially confusing feature is
understood. For class 7 crystals, the crystallographic axes abc on which the fundamental
symmetry properties are based do not always translate in the obvious way to the xyz axes
used to define the index ellipsoid and the direction of the optic axes. Of course, problems
of this kind are to be expected given that the crystallographic axes are not even orthogonal
in four of the seven crystal symmetry systems. But even in the orthorhombic system (that
includes class 7) where the crystallographic axes are orthogonal, one cannot necessarily
assume that a x, b y, c z. The different mappings for the five class 7 crystals
listed in Table 4.3 are shown in column 6.
Why is this a problem? Because it is the abc axes, not the xyz axes, that determine the
structure of the d-matrices, and the numerical suffix notation is based on the former, not the
latter. For LBO, the axes mapping is a x, b z, c y, and the non-zero d-elements
are d31 = dcaa = dyxx , d32 = dcbb = dyzz , d33 = dccc = dyyy as shown in Table 4.5.
Notice that 3 now means y, and 2 means z!
Apart from this, it is straightforward to analyse SHG in LBO. The refractive indices at
1.06 m and 0.53 m are listed in Table 4.6, and the x-y section of the index surfaces is
5 The formula for the effective coefficient quoted by Dmitriev et al. [26] assumes that d
14 = d36 which is a
reasonable approximation in KDP.
4.6 Second harmonic generation in LBO
71
Table 4.5 Non-zero d-elements in LBO.

1 = aa
= xx
1=a=x
2=b=z
3=c=y
2 = bb
= zz
d31
d32
3 = cc
= yy
d33
4 = bc, cb
= zy, yz
5 = ac, ca
= xy, yx
6 = ab, ba
= xz, zx
d15
d24
Table 4.6 Refractive indices of LBO.

LBO
1.06 m
0.53 m
nx
ny
nz
1.5648
1.5785
1.5904
1.6065
1.6053
1.6216
nz ()
nx (2)
x
n z () n y (2)
Fig. 4.2
The x-y section of the LBO index surfaces showing the phase-matching direction for SHG. The index variations are
exaggerated.
shown in Fig. 4.2. Two waves can propagate at an angle to the x-axis: the one polarised
in the z-direction is a quasi-ordinary wave6 with refractive index nz , while the refractive
index of the other wave falls from ny at = 0 to nx at = 90 . Since ny (2) > nz () >
nx (2), phase matching is possible at a value of that can be found from
!
2
2
(n2
n2
y ) (nz )
x
cos =
(4.20)
2
2 2
nz (n2
y ) (nx )
6 A quasi-ordinary wave has the propagation characteristics of an ordinary wave in a uniaxial medium, although
there are strictly speaking no ordinary waves in biaxial media; see Section 3.5.
72
which has the same structure as Eq. (4.16). Using the data of Table 4.6 yields a value of
around 11.6 .
The d coefficient mediating the generation of the second harmonic wave is dyzz = dcbb =
d32 ; another possibility would be dxzz = dabb = d12 , but this coefficient is zero in class 7
crystals. Since the angle between the y-axis and the plane of the wavefront is , the effective
nonlinear coefficient is deff = d32 cos pm which is close to d32 because pm is small.
4.7 The Pockels effect and the Pockels cell

As described in Chapter 1, the Pockels effect relates to the change in refractive index of a
non-centrosymmetric medium induced by an applied DC field. From Eq. (4.6) we have

ijPEk (; , 0)E j ()EkDC
(4.21)
Pi () = 20
jk
where, as explained in Section 4.3.3, the order of the fields on the right-hand side is fixed
by definition.
The Pockels effect is in fact more usually described in terms of the effect of the DC field
on the inverse dielectric constant matrix, which reads
2
no
0
0
(4.22)
1 = 0
n2
0
o
0
0
n2
e
when referred to the principal axes of a uniaxial medium. Use of 1 turns out to be
convenient insofar as the equation of the index ellipsoid (Eq. 3.6) can be written
r 1 r = I
(4.23)
where r is the row (xyz), r is the associated column, and I is the identity matrix. In terms
of its effect on 1 , the Pockels effect can be represented by

rij k EkDC .
(4.24)
(1 )ij =
k
Since the symmetry of 1 means that the order of the indices ij is irrelevant, a similar
contracted notation to that of Table 4.1 can be introduced. Equation (4.24) then reads

1
=
rpn EnDC (n = 1 3, p = 1 6)
(4.25)
(1 )p =
2
n p
n
where the index p represents a pair of subscripts as before. The relationship between the r
and coefficients can be shown to be
rij k = rj ik = rpn =
ijPEk (; , 0) + jPE
ik (; , 0)
ii jj
2ijPEk (; , 0)
ii jj
(4.26)
73
4.7 The Pockels effect and the Pockels cell
Consider the simple case of a class 11 crystal such as KDP. In close analogy with the
d coefficients in the upper section of Table 4.4, the only non-zero r coefficients are r41 =
r52 = r63 . For a wave polarised in the x-y plane propagating in the +z-direction through a
sample of thickness l with a DC voltage V between the exit and entrance surfaces, Eq. (4.25)
reads

1
1
r63 V

=
=
.
(4.27)
n2 xy
n2 yx
l
Equations (4.22) and (4.23) now yield for the equation of the index ellipsoid
x2 + y2
z2
2xy r63 V
+ 2+
= 1.
2
no
ne
l
(4.28)
It is readily shown that the z = 0 section of the ellipsoid is an ellipse whose axes are
inclined at 45 to the x- and y-axes, as shown in Fig. 4.3. The lengths of the semi-axes,
which represent the refractive indices of the two orthogonally polarised fields that can
propagate in the z-direction are
n
= n0 (1 12 n20 r63 V /l).
(4.29)
The nonlinear medium therefore behaves as a wave plate whose strength can be controlled
by varying the applied voltage; this is the principle on which the Pockels cell is based.
Section 3.4 gives the quarter-wave distance as /4 |n| and this, with n = n30 r63 V /l
from Eq. (4.29), yields for the quarter-wave voltage the formula
V/4 =
4n30 r63
(4.30)
y
y'
x'
Fig. 4.3
The Pockels effect causes the x-y section of the index ellipsoid in a uniaxial medium to become elliptical. The circular
(zero DC eld) prole is shown in grey.
74
Notice that the thickness of the crystal has cancelled out. This is because doubling the
crystal length doubles the interaction length but halves the longitudinal electric field for a
given applied voltage.
Unfortunately, the voltages required to operate Pockels cells on this basis are very high
(see Problem 4.5), but it is possible in principle to construct wave plates, modulators,
and beam deflectors using these principles. An alternative configuration in which the DC
potential is transverse to the beam has the advantage that the interaction length can be
increased without lowering the DC field.

In Chapter 1, we saw how a beam propagating in a non-centrosymmetric medium can create
a DC polarisation proportional to the intensity. This process, known as optical rectification,
was studied extensively in the early days of nonlinear optics, and was then largely forgotten
for 20 years until it was rediscovered as a means to create terahertz radiation. The process
is described by the formula

PiDC = 12 0
ijOR
(4.31)
k (0; , )Ej ()Ek ()
jk
from Eq. (4.7). At this point, the principle of full permutation symmetry enables us to write
OR
PE
kj
i (0; , ) = ij k (; , 0).
(4.32)
This exact relationship between the Pockels effect and optical rectification coefficients can
be incorporated into Eq. (4.26) to yield
rij k = rj ik = rpn =
2ijPEk (; , 0)
ii jj
OR (0; , )
2kj
i
ii jj
(4.33)
We have therefore discovered two nonlinear optical effects with a direct numerical
relationship between their coefficients.
Problems
4.1 What are the units of the second-order nonlinear coefficients defined in this chapter?
4.2 Consider a possible set-up for implementing the situation envisaged in Problem 1.5.
What beam directions could be used and what nonlinear coefficient would mediate the
interaction?
4.3 Try to prove Eq. (4.19). A good start is to write down the components of unit vectors
in the o and e directions. It does not matter which direction of the double arrows you
count as positive, provided you are consistent throughout the calculation. This is a
tricky question.
75
4.4 Prove Eq. (4.29). This formula, which involves an approximation, was used to plot
Fig. 4.3.
4.5 The electro-optic coefficient r63 for ammonium dihydrogen phosphate (ADP) is
8.5 pm V1 . Find the quarter-wave voltage at the wavelength of a HeNe laser.
(n0 (632.8 nm) in ADP = 1.522.)
Third-order nonlinear processes
5.1 Introduction
Third-order nonlinear processes are based on the term 0 (3) E 3 in the polarisation
expansion of Eq. (1.24). Just as the second-order processes of the previous chapter coupled
three waves together, so third-order processes couple four waves together, and hence are
sometimes called four-wave processes.
Third-order processes are in some ways simpler and in other ways more complicated than
their second-order counterparts. Perhaps the most important difference is that third-order
interactions can occur in centrosymmetric media. This means that, while crystals can be used
if desired, one is often dealing with optically isotropic materials in which the complexity
of crystal optics is absent. However, the tensor nature of the third-order coefficients is still
not entirely straightforward, even in isotropic media.
A second issue is that phase matching is automatic for many third-order processes and,
even when it is not, the existence of four waves, potentially travelling in four different
directions, makes phase matching easier to achieve. Of course, phase matching was also
automatic at second order for the Pockels effect and optical rectification, but those were
somewhat special cases involving DC fields.
At first sight, automatic phase matching sounds like a good thing, but the benefit (if it
is such) comes at a serious price. When phase matching was critical, and one had to work
to achieve it for a particular combination of frequencies, it did at least provide a way of
promoting one particular nonlinear process, while discriminating against all the others. The
problem with the third-order processes tends to be that, when one of them occurs, so do
several others, and disentangling them can then be very difficult.
In principle, we could follow the same routine at third order as we did for the secondorder processes in Chapter 4. Although all the ideas are the same, the higher order makes
things more complicated than before, and then it was bad enough! Here, we shall make
things simpler by confining attention almost exclusively to structurally isotropic media, i.e.
non-crystalline media such as gases, liquids and amorphous solids.
In Section 5.2, we will discuss several basic third-order interactions including third harmonic generation (THG), the DC and optical Kerr effects, and intensity-dependent refractive
index (IDRI). IDRI in particular has widespread applications in optical pulse technology,
some of which are covered in Chapter 7. Stimulated scattering processes are introduced in
Section 5.3 by extending the Kerr effect analysis to include a material resonance. The stimulated Raman effect is discussed from two different perspectives, one of which highlights
76
5.2 Basic third-order processes
77
the travelling wave nature of the Raman excitation. The insight gained from this exercise
carries over into Section 5.4, which is concerned with the interaction of optical and acoustic
waves. We consider two cases involving the diffraction of light by ultrasound, before moving on to stimulated Brillouin scattering in Section 5.5. Section 5.6 deals with optical phase
conjugation, and the chapter ends with a brief comment on supercontinuum generation in
Section 5.7.

5.2.1 Definitions
The most general third-order nonlinear process involves the interaction of waves at four
different frequencies linked by 1 + 2 + 3 = 4 . Fortunately, in all common cases, some
of the frequencies are the same, and some may also be zero, or the negatives of others.
Important examples including third harmonic generation and three different variants of the
Kerr effect were mentioned in Chapter 1 in connection with Eqs (1.25) and (1.27). Those
equations were of course based on a simple formalism where the nonlinear coefficients were
assumed to be scalar quantities. However, to appreciate all the features of these effects, and
especially their response to different polarisation states of light, one cannot avoid treating the
coefficients as tensors, even in the case of structurally isotropic media. Since the associated
mathematics is intricate, most of the detail has been relegated to Appendix C.
According to the most general formula (see Eq. C3.1), and in close analogy with Eq. (4.3),
the polarisation at 4 is given by
(3)
ij kl (4 ; 1 , 2 , 3 )E j (1 )E k (2 )E l (3 )
(5.1)
Pi (4 ) = 14 0
p
j kl
%
where ijkl can each be x, y or z, and p indicates that the right-hand side is to be summed
over all distinct permutations of 1 , 2 and 3 . This means that the form of the polarisation
depends on the frequency set, and so is process-specific. Some particular examples are:
Third harmonic generation ( + + = 3)

Pi (3) = 14 0
ijTHG
kl (3; , , )Ej ()Ek ()El ().
(5.2)
j kl
Third harmonic generation is closely analogous to second harmonic generation at second

order, and is discussed in Section 5.2.3 below.
The DC Kerr effect ( = 0 + 0 + )

Pi () = 30
ijKkl (; 0, 0, )Ej (0)Ek (0)E l ().
(5.3)
j kl
The first nonlinear effect to be observed (see Chapter 1), this concerns refractive index
changes caused by an applied DC field; see Section 5.2.4.
78
The optical Kerr effect (1 = 2 2 + 1 )

Pi (1 ) = 32 0
ijOK
kl (1 ; 2 , 2 , 1 )Ej (2 )Ek (2 )El (1 ).
(5.4)
j kl
In this case, the refractive index of an optical wave at 1 is modified in the presence of a
wave at 2 ; more details are given in Section 5.2.5. If the difference between the frequencies
of the two waves coincides with a resonance in the nonlinear medium, the same formula
governs stimulated Raman scattering, which is discussed in Section 5.3.
Intensity-dependent refractive index ( = + )

Pi () = 3 0
ijIDRI
kl (; , , )Ej ()Ek ()El ().
4
(5.5)
j kl
This is a special case of the optical Kerr effect in which the two waves of Eq. (5.4) are one
and the same, and a single wave modifies its own refractive index; see Section 5.2.6.
A full justification of Eqs (5.2)(5.5) is given in Section C3 of Appendix C. It is worth
pointing out that the different numerical pre-factors in Eqs (5.2)(5.5) result from the permutation operation in Eq. (5.1). Once this has been performed, there is no more permutation
to do and, thereafter, the order of the field frequencies in the coefficient arguments is fixed
by definition. The pre-factors are incidentally identical to those in Eqs (1.25) and (1.27) of
Chapter 1.
It should also be noted that in cases where two (or more) optical waves are involved,
these need not necessarily travel in the same direction. There is, for example, no reason in
principle why the two waves in Eq. (5.4) need be collinear, and in this case their frequencies
could even be the same. An important special case known as degenerate four-wave mixing
(DFWM) involves four waves with the same frequency, but travelling in different directions;
see Eq. (5.104).
5.2.2 Symmetry considerations

Now that the tensor character of the nonlinear coefficients is explicit, we need to consider
the effect of symmetry. All the principles developed at second order in Chapter 4 extend to
the third-order case. The key conclusion is that, in a lossless crystal of the most general triclinic symmetry, there are 34 = 81 independent nonlinear coefficients. For other symmetry
classes, the number is lower, and tables listing the independent coefficients in each class can
be found in Boyd [1]. We are restricting ourselves to structurally isotropic media, in which
there are no intrinsic axes, all directions are equivalent, and the orientation of the xyz-axes
can be chosen to make calculations as simple as possible.1 In this case, only 21 of the 81
coefficients are non-zero, and these are of four types: type 1 (three members) in which all
indices are identical (1 iiii ), and types 2, 3 and 4 (six members each) in which two
pairs of indices are the same, namely 2 = jj kk , 3 = j kj k and 4 = j kkj (j = k). The
1 Once again, it is important to distinguish between the structural isotropy of gases, liquids and amorphous solids,
and optical isotropy, which is exhibited by cubic crystals too (see Table 3.1). By its very nature, a crystal is not
structurally isotropic although, in cubic crystals, all directions are optically equivalent.
79
numbering scheme ensures that indices 1 and 2 are the same in 2 , indices 1 and 3 in 3 , and
1 and 4 in 4 . Within each type, all members are equal and, as we shall show in a moment,
the symmetry of a structurally isotropic medium imposes the further constraint that
1 = 2 + 3 + 4 .
(5.6)
In terms of indices, the 21 non-zero coefficients can be listed as follows:

type 1 : xxxx = yyyy = zzzz (= xxyy + xyxy + xyyx)
type 2 : xxyy = yyzz = zzxx = yyxx = zzyy = xxzz
type 3 : xyxy = yzyz = zxzx = yxyx = zyzy = xzxz
(5.7)
type 4 : xyyx = yzzy = zxxz = yxxy = zyyz = xzzx

where the final step of the first line follows from Eq. (5.6). The key conclusion of Eqs (5.6)
and (5.7) is that a structurally isotropic medium has just three independent third-order coefficients. As we shall discover in the following sections, further links between the different
types of coefficient apply in certain cases. For collinear beams, it makes sense to set the zaxis along the direction of propagation, in which case all coefficients involving z in Eq. (5.7)
can be ignored; this reduces the number of relevant non-zero coefficients from 21 to 8. Moreover, if all beams are plane polarised in the same direction, the x-axis can be chosen as the
direction of polarisation, in which case the only relevant coefficient is xxxx = 1 .
In cubic crystals, which are optically (but not structurally) isotropic, the same 21 elements
as in Eq. (5.7) are non-zero, but there are fewer links between them. For crystal classes 29,
31, and 32, the number of independent elements is four, the only change in this case being
that Eq. (5.6) no longer applies. For classes 28 and 30, on the other hand, there are seven
independent elements.2

This is the simplest third-order effect to analyse. The generation process is a direct extension
of the second harmonic case at second order, and has already been treated in Section 2.3.4
of Chapter 2. Since all field frequencies on the right-hand side of Eq. (5.2) are the same,
permuting the last three coordinate indices (j , k and l) of the THG coefficients makes no
difference, and so 2 = 3 = 4 . In the case of plane-polarised incident light, everything
is especially straightforward since the only element of interest is 1 . Of course, there is
nothing to stop one choosing coordinate axes that are rotated (say by 45 ) with respect to
the incident field, as shown in Fig. 5.1. A simple exercise (see Problem 5.1) demonstrates
that the effective coefficient with respect to the new axes is
eff = 12 (1 + 2 + 3 + 4 ).
(5.8)
Clearly eff must equal 1 , since the coefficient cannot depend on the choice of axes, and
this constitutes a proof of Eq. (5.6). It is also quite easy to show (see Problem 5.2) that the
THG polarisation goes to zero in the case of circularly polarised light.
2 The type 2, 3 and 4 coefficients divide into two sets of three at the mid-point in Eq. (5.7).
80
E1
Fig. 5.1
Slanting axes used to prove Eq. (5.8); see Problem 5.1.
5.2.4 The DC Kerr effect and the Kerr cell

In the DC Kerr effect,3 a strong DC field changes the refractive index of a medium. If the
DC field is y-polarised, Eq. (5.3) indicates that the respective polarisations in the x- and
y-directions are
K (; 0, 0, )E 2 (0)E
x () = 30 K Ey2 (0)E x ()
Px () = 30 xyyx
y
4
K (; 0, 0, )E 2 (0)E
y () = 30 K Ey2 (0)E y ().
Py () = 30 yyyy
y
1
(5.9)
It follows that the DC field creates a refractive index difference between the two polarisations
given by
n n
=
3(1K 4K )Ey2 (0)

2n
32K Ey2 (0)

n
(5.10)
where n and n are the respective indices for light polarised parallel and perpendicular
to the DC field, and n is the zero field index. The second step follows from Eq. (5.6), and
because 2K and 3K are indistinguishable from Eq. (5.3). The Kerr constant K of a medium
is defined by the equation
n n n = 0 KE 2 (0)
(5.11)
where 0 is the vacuum wavelength. This implies that K = 32K /(0 n) for consistency
with Eq. (5.10). The sign of K is normally positive, although exceptions occur.
An optical switch based on the DC Kerr effect can be based on the same ideas as those used
in a Pockels cell (see Section 4.7). A Kerr cell is placed between a pair of crossed polarisers
set at 45 to the x- and y-axes. In the end-on view shown in Fig. 5.2, the electrodes are set to
apply a DC field in the y-direction, while PP and AA indicate the respective alignments of
the polariser and analyser. From Section 3.4, we know that to rotate the plane of polarisation
of light by 90 , a phase change of needs to be introduced between the x- and y-components
of the field. From Eq. (5.10), this is clearly realised when the light propagates through the
half-wave distance is given by
L =
0
1
.
=
2 |n|
2 |K| E(0)2
3 Sometimes called simply the Kerr effect, this was the first nonlinear optical effect to be observed [2].
(5.12)
81
Fig. 5.2
Schematic diagram of a Kerr cell looking along the line of the beam. The lines PP and AA show the alignment of
polariser and analyser.
For liquids (such a nitrobenzene) commonly used in Kerr cells values of K are typically on
the order of 1012 m/V. This implies (see Problem 5.3) that Kerr cell switching voltages
tend to be in the tens of kV range. Kerr liquids exhibit a multiplicity of relaxation times,
mostly in the subpicosecond regime, but the response times of real Kerr cells are determined
by other factors, and 100 ps is certainly achievable.
5.2.5 The optical Kerr effect

In the optical Kerr effect,4 a strong wave at frequency 2 and intensity I (2 ) changes the
refractive index of a weak probe wave at 1 , a process known as cross-phase modulation; the
acronym XPM is often used. If the strong and the weak waves have the same polarisation,
the operative term in the polarisation is

2

OK
Px (1 ) = 32 0 xxxx
(1 ; 2 , 2 , 1 ) E x (2 ) E x (1 )
(5.13)
which implies that the refractive index of the weak wave is changed by
nx
=
OK I ( )
3xxxx
2
.
2n(1 )n(2 )co
(5.14)
If, on the other hand, the weak and strong waves are, respectively, x- and y-polarised,
Eq. (5.13) becomes

2

OK
Px (1 ) = 32 0 xyyx
(1 ; 2 , 2 , 1 ) E y (2 ) E x (1 ).
(5.15)
4 This is sometimes known as the AC Kerr effect.
82
This is the same as Eq. (5.13) apart from the fact that it contains a type 4 coefficient, and
so the index change is weaker. One cannot say that it is three times as weak, because the
type 2, 3 and 4 coefficients are not necessarily equal in this case.

An important special case of the optical Kerr effect occurs when a single beam at = 1 =
2 modifies its own refractive index. In the next section, we will discuss some physical
effects that arise from this important self-action effect, which is known as intensitydependent refractive index (IDRI).
For the case of plane-polarised light, Eq. (5.5) can be written in the simple form

2

Px () = 34 0 1IDRI E x () E x ().
(5.16)
This implies that the refractive index is changed to5

31IDRI
n = n0 +
I = n0 + n2 I
4n20 c0
(5.17)
where I is the intensity, n0 is the low-intensity index, and the equation defines n2 as
the nonlinear refractive index. It is no surprise that the refractive index change implied by
Eq. (5.17) is essentially the same as that of Eq. (5.14); the extra factor of 2 in the denominator
of n2 arises from the different pre-factors in Eqs (5.4) and (5.5).
Equations (5.16)(5.17) include only the type 1 coefficient. However, IDRI exhibits some
particularly interesting features in the case of circularly and elliptically polarised light, and
so we should also consider these more complicated situations in which other coefficients are
involved too. Because the second and fourth frequency indices of the IDRI coefficient are
the same (see Eq. 5.5), it follows that2IDRI = 4IDRI ( = 3IDRI ). Under these circumstances,
it can be shown (see Eq. C3.8 of Appendix C, and Problem 5.4) that

E
)E x + 1 B(E.
E)
E x
Px = 14 0 A(E.
2
(5.18)

E
)E y + 1 B(E.
E)
E y
Py = 1 0 A(E.
4
where we have adopted the traditional convention [31] that A = 62IDRI and B = 63IDRI .
Since we continue to focus on waves propagating in the z-direction, we are only considering
E = E x E x + E y E y etc.
the x- and y-components of the field and polarisation, and so E.
A feature of special interest in Eq. (5.18) is the presence of Ex and Ey in the B terms.
The complex conjugation indicates that, for circularly polarised light, the handedness of
this contribution to the polarisation is opposite to that of the incident field. This becomes
clearer if we write down the clockwise component of the polarisation namely6

E
)(E x i E y ) + 1 B(E.
E)(
E x + i E y ) .
Px i Py = 14 0 A(E.
(5.19)
2
5 See also Eq. (1.26).
6 Clockwise looking in the +z-direction.
83
Incident light with counter-clockwise circular polarisation in E will therefore induce a

clockwise contribution to the polarisation P of the medium.7
A detailed analysis of this feature and its ramifications can be found in Chapter 4 of Boyd
[1]. The most important conclusion is that, for elliptically polarised light, the axes of the
ellipse rotate under propagation. In contrast, for plane polarised and circularly polarised
beams, the state of polarisation is unaffected by the intensity-dependent index change.
In the former case, the change is given by Eq. (5.17). For circularly polarised light, the
corresponding equation is

A
n = n0 +
I
(5.20)
4n20 c0
so the B coefficient plays no part in this case.
5.2.7 Spatial solitons, self-focusing, and self-phase modulation

Intensity-dependent refractive index (IDRI) is probably the most important of the third-order
processes in terms of its consequences. The spatial modulation of refractive index that it
creates has a direct influence on optical beam propagation, while the corresponding temporal
modulation affects the amplitude and phase structure of optical pulses, and the spectral
structure of optical signals in general. In this section, we will concentrate on the spatial
aspects of IDRI. Its contribution to supercontinuum generation is discussed in Section 5.7,
and a detailed analysis of its effects on optical pulse propagation (and the many ways in
which these can be exploited) is presented in Sections 7.37.6.
Consider first an intense optical beam with a bell-shaped transverse intensity profile
propagating through a block of glass. The index change of Eq. (5.17) introduces an intensitydependent spatial phase modulation, altering the shape of the wavefront, and hence its
focusing and defocusing properties. Since the nonlinear index n2 is almost invariably positive,8 the refractive index will be raised at the centre of the beam where the intensity is
greatest. This leads directly to a reduction in the wavelength, with the further consequence
that the wavefronts bend in on themselves as shown in Fig. 5.3. If this effect is strong enough
Fig. 5.3
Beam collapse due to self-focusing.

7 Notice both meanings of the word polarisation in this sentence. See footnote 1 in Chapter 1.
8 In fused silica, n is on the order of 3 1016 cm2 W1 .
2
84
to overcome diffraction, which works in the opposite direction, the self-focusing tendency
will predominate, and the beam will get narrower, causing the intensity to rise further and the
process to accelerate. In the worst-case scenario, the end result is the catastrophic collapse
of the beam, leading to severe and irreversible damage to the optical material.
Fortunately, the competition with diffraction means that whole beam self-focusing,
which is what has been described, has a definite threshold. Unfortunately, a closely related
phenomenon called small-scale self-focusing also occurs, and leads to local hot-spots and
filamentation. These effects were first considered by Kelley [32], and were observed by
Alfano and Shapiro [33]. Precautions against all forms of self-focusing are routine in large
laser systems. These are usually based on spatial filters in which the beam is focused through
a small aperture to smooth out distortions in the wavefront. To prevent breakdown near the
focus, the process takes place in a vacuum.
The theory of all types of self-focusing has had a long and chequered history. We will
concentrate on the simplest case of whole-beam self-focusing in the case of one transverse
dimension, i.e. a slit geometry. As we will discover, the competition in this case between
diffraction and self-focusing can result in a stable balance, and the formation of spatial
solitons. In two transverse dimensions, the analogous solution is normally unstable, which is
unfortunate insofar as it would otherwise be possible to create optical beams that propagated
without diffractive spreading.
We start by inserting the nonlinear polarisation term from the standard coupled-wave
equation (2.8) into the paraxial wave equation (2.47). The resulting equation is
i
E
1 2 n2 I E
=
E +
z
2k T
c
(5.21)
where the transverse Laplacian T2 = 2 /x 2 + 2 /y 2 takes care of diffraction, and a

simplified version of Eq. (5.16) has been used to form the nonlinear term.
It is important to remember that, without the second term on the right-hand side, Eq. (5.21)
is simply the paraxial wave equation, which has the Gaussian beam of Eq. (2.47) as an exact
solution. However, when the second term is present, the potential exists for the equation
to have a stationary solution, where the transverse profile does not vary with z. This is
because, provided n2 is positive (which it normally is), the phase of the second term is that
of +E , whereas, for a beam with a bell-shaped transverse profile and a plane wavefront,
the phase of the first term is that of E . As we now show, in one transverse dimension,
an analytical solution to Eq. (5.21) can be obtained in which the effects of diffraction and
nonlinear refraction are in perfect balance.
To adapt Eq. (5.21) to a single transverse dimension, we assume that the electric field
is constant in (say) the y-dimension, so the second derivative in y can be dropped. We
introduce three characteristic length parameters, a beam width W , the diffraction length Ld
linked to W through Ld = kW 2 (where k is the angular wave number), and a nonlinear
propagation distance defined by
Lnl =
c
|n2 | Ipk
(5.22)
85
where Ipk is the peak intensity. Defined in this way, Lnl is the distance over which the
nonlinear index change introduces a phase change of 1 radian, while Ld = kW 2 is related
(though not identical) to the Rayleigh length in Section 2.5.2.9
We now use W and Lnl to define the dimensionless variables = z/Ld and = x/W .10
E pk and, for reasons that
We also introduce a dimensionless variable for the field U = E/
will emerge very shortly, we define

(5.23)
N = Ld /Lnl = k02 n0 |n2 | Ipk W 2 .
In terms of these new variables, and assuming that n2 > 0, Eq. (5.21) reduces to the simple
form

2
2
U
2
1
i = 2 2 + N U U .
(5.24)
Equation (5.24) is in the form of the celebrated nonlinear Schrdinger equation, which possesses a hierarchy of remarkable analytical solutions known as solitons [34]. The parameter
N defined in Eq. (5.23) is the soliton order.
Solitons arise in a wide variety of physical systems when nonlinear and dispersive effects
(diffraction in the present case) are in opposition. In the lowest order (N = 1) case, the two
terms on the right-hand side of Eq. (5.24) are in perfect balance, and the soliton propagates
as a beam of constant width. The N = 1 solution of Eq. (5.24) is
U ( , ) = sech{ } exp{ 12 i }.
(5.25)
When this is substituted into the right-hand side of Eq. (5.24), the result is zero, which is
of course the mark of a stationary solution. Moreover, since N = 1 in this case, it follows
from Eq. (5.23) that
c
(= Lnl ) = kW 2 (= Ldisp )
(5.26)
|n2 | Ipk
so the peak intensity of the soliton is related to the width parameter W through the equation
Ipk W 2 =
c2
.
2 n0 |n2 |
(5.27)
The physical meaning of this equation is that the smaller the beam width W , the stronger the
diffraction, and the higher the peak intensity needs to be to maintain the balance between
the focusing effect of the nonlinear term, and the defocusing effect of diffraction. Rewriting
Eq. (5.25) in terms of the original parameters yields
z) = E pk sech{x/W } exp{ 1 iz/Lnl }.
E(x,
2
(5.28)
9 If W is identified with w in Gaussian beam theory, the diffraction length L = kW 2 corresponds to twice
0
d
the Rayleigh length, otherwise known as the confocal parameter b = 2zR .
10 The prime on reflects the 2 difference between L and the Rayleigh length noted in the previous footnote.
d
86
We could of course have written Ld instead of Lnl because the two are the same when
N = 1. The final term on the right-hand side implies that the phase velocity of the soliton
solution is slightly less than that of a free-propagating wave, which is expected given that
the index at the centre of the beam is raised by the nonlinear refractive index term.
Spatial solitons in a slit geometry have been observed experimentally [35,36], and have
potential applications in all-optical devices. However, the problem we have been discussing
has an exact parallel in the context of optical pulse propagation, where becomes a time
variable, and group velocity dispersion takes the role that diffraction plays here. This important topic is treated in Section 7.4, where higher-order solitons are discussed, and pictures
of the soliton solutions presented.
To what extent can the results we have obtained be extended to the case of two transverse
dimensions? Even without doing any more mathematics, an interesting feature of this new
case can be deduced from Eq. (5.27). In circular geometry where the beam widths in the
x and y dimensions are the same, W 2 will be a measure of the beam area, and Eq. (5.27)
now indicates that the balance between diffraction and self-focusing depends on power!11
Moreover, according to the equation, the critical power for self-focusing depends only on
the wavelength of the light, and the nonlinear refractive index of the medium. The standard
formula for critical power, which is the same as the right-hand side of Eq. (5.27) apart from
a numerical factor, is [37]
Pcrit = (2 /4 n0 n2 ).
(5.29)
The parameter depends on the precise beam profile, and is usually close to 2.
Typical values of Pcrit are remarkably modest. For fused silica, which has a notably
low value of n2 , Pcrit is still only a few megawatts, a very small laser power by any standard. This raises an immediate question about what can be done to prevent self-focusing
occurring. Fortunately, the distance scale over which self-focusing occurs is linked to
Lnl , which depends on intensity not on power. The power may exceed Pcrit but, provided the beam is wide and the intensity low, nothing dramatic will happen for a long
distance.
There are several other issues too. As we have seen, stable soliton solutions exist in the
case of one transverse dimension. However, the corresponding solutions in two transverse
dimensions are unstable if the Kerr effect is the only nonlinear process involved. Stable
solutions can, however, exist, at least in principle, if saturable absorption occurs at the
same time. As noted at the beginning of this section, rather than whole-beam self-focusing,
which is what we have been discussing, what tends to happen in practice is that beam
profiles become subject to local filamentation, a process that is driven by a complicated
four-wave mixing interaction. The detailed theory of self-trapping is complex, and beyond
the scope of the present book.
Whereas intensity-dependent refractive index causes self-focusing when the intensity
varies in space, it leads to temporal self-phase modulation (SPM) when intensity varies in
11 Previously, W 2 was the square of the beam width in the x-direction, and the extent of the beam in the y-direction
was arbitrary.
5.3 Stimulated Raman scattering
87
time.12 SPM plays a crucial role in many techniques involving short optical pulses, and the
topic is treated in detail in Sections 7.37.6 of Chapter 7.

5.3.1 Raman coupling of laser and Stokes waves
We now move on to consider the process of stimulated Raman scattering (SRS). SRS was
discovered accidentally by Woodbury and Ng in 1962 [3839], while they were working
with a ruby laser that was Q-switched using a nitrobenzene Kerr cell. As well as 694.3 nm
(431.8 THz) radiation, they found that laser action was also occurring at 765.8 nm (391.5
THz), and noticed that the 40.3 THz frequency difference corresponded exactly to a vibrational resonance frequency of the nitrobenzene molecule. They called the device a Raman
laser, and the principle is now in widespread use, especially in fibre Raman systems.
SRS is the stimulated version of spontaneous Raman scattering, a process that was first
observed in the 1920s using conventional light sources. In those experiments, the scattered
frequency component was called the Stokes wave, and this terminology has been retained
for the stimulated process.13 In practice, SRS is frequently based on vibrational resonances,
although rotational or electronic states may also be involved. The terms stimulated rotational Raman scattering and stimulated electronic Raman scattering are sometimes used.
A further possibility is that the laser and Stokes waves interact via an acoustic wave; this
case is called stimulated Brillouin scattering, and is discussed in Section 5.5 below.
For all the third-order processes discussed earlier in the chapter, the optical field frequencies and their simple combinations (i.e. sums and differences) have been assumed to
be far from any material resonances, whatever their origin. As a consequence, the nonlinear
coefficients were real, and the energy in the optical fields conserved. To discover how SRS
originates, we start by writing Eq. (5.13) for the optical Kerr effect in the modified form

2

P (S ) = 32 0 (3) (S ; L , L , S ) E(
)
L E(
S ).
(5.30)
To make things simple, the spatial aspects of the original equation have been dropped, and
the polarisation and the fields have been replaced by their slowly varying envelopes.14 In
anticipation of their role in stimulated Raman scattering, the frequencies of the strong and
weak waves have been written L and S (< L ) where L and S stand respectively for
laser and Stokes.
12 The word temporal in this sentence is normally omitted, so self-phase modulation (SPM) invariably refers
to the time-dependent case.
13 The reference is to Sir George Stokes, who first discussed the change of wavelength of light in the context of
fluorescence in a famous paper of 1852. The terminology is increasingly used to describe any down-shifted
frequency.
14 The relationship between P and P , and E and E is specified in Eqs (2.4)(2.5). As explained later, the process
is automatically phase matched, so direct replacements can be made.
88
As long as the third-order coefficient in Eq. (5.30) remains real, the Stokes wave polarisation will be in phase with the Stokes field, and the equation will therefore mediate the
intensity-dependent refractive index change characteristic of the optical Kerr effect. But
suppose the coefficient becomes complex. What will be the effect of the imaginary part?
If we substitute Eq. (5.30) with (3) = + i into the general coupled-wave equation
(2.8), we obtain
2
E S
3S

=
( i ) E L E S
z
4cnS
(5.31)
where the field frequencies are now indicated by suffices. Clearly the sign of is critical,
since it determines whether the Stokes field grows or decays. To resolve the issue, we
anticipate a result from Chapter 9 where the quantum mechanical basis of SRS is discussed.
Equation (9.10) indicates that is positive, in the absence of a population inversion
in the nonlinear material. If the population is concentrated in the ground state, we duly
conclude that the Stokes wave in Eq. (5.31) will grow, and this is the basis of the SRS
process. The corresponding equation governing the growth of the Stokes wave intensity
2

IS = 1 nS c0 E S is readily shown to be
2
IS
=
z
3S R IL
c 2 0 nP nS

IS = gS IS
(5.32)
where the second step defines the factor in brackets as the Stokes gain coefficient gS . The
subscript R (for Raman) has been added to the nonlinear coefficient at the same time. If
the energy conversion from L to S is small, we can assume that the laser intensity is
undepleted, in which case Eq. (5.32) has the straightforward exponential solution
IS (z) = IS (0) exp{gS z}.
(5.33)
According to this equation, the Raman gain coefficient is proportional to the pump laser
intensity, so enormous gain factors are potentially available by increasing IL .15 Indeed, the
Stokes wave will grow from noise in a cell of length L, once exp{gS L} gets sufficiently
large. When gS L 25, for example, the gain factor is approaching 1011 , which is likely
to be more than sufficient. Once the stimulated Raman process takes off, the assumption that
the pump laser field is undepleted is of course likely to break down, and we will consider
this aspect of the problem shortly.
A simple energy level diagram of SRS is shown in Fig. 5.4(a). Peak gain occurs when
the difference frequency (L S ) between the laser and Stokes waves coincides with
a resonance (at angular frequency ;) in the medium. As mentioned already, the level
associated with the resonance (labelled 2 in Fig. 5.4) is often a vibrational state, but it could
also be rotational or electronic in origin.16 As already noted (and suggested by the figure),
the growth of the Stokes field will be accompanied by depletion of the laser field, which is
15 We refer to g as the gain coefficient and exp{g z} as the gain factor. It is important to distinguish the two,
S
S
and good practice to avoid using the word gain on its own.
16 Strictly speaking, all frequency labels in Fig. 5.4 should be multiplied by as befits an energy level diagram.
89
L
S
AS
AS
S 2
(a)
Fig. 5.4
(b)
(c)
(d)
Schematic energy level diagrams for stimulated Raman scattering (a) rst Stokes, (b) second Stokes, (c) rst
anti-Stokes, (d) parametric anti-Stokes.
governed by equations analogous to Eqs (5.31) and (5.32), namely
2
E L
3L
R iR E S E L
=
z
4cnL
3L R
IL
= 2
IS IL .
z
c 0 nL nS
(5.34)
(5.35)
For every photon gained by the Stokes wave, one is lost by the laser, but optical energy is
not conserved because the population of the Raman level grows at the same time. The sign
changes in Eqs (5.34)(5.35) compared to Eqs (5.31)(5.32) arise because the nonlinear
coefficient governing the polarisation at L is the conjugate of the coefficient in Eq. (5.30)
since
(3) (L ; S , S , L ) = (3) (S ; L , L , S ).
(5.36)
There has been no mention of phase matching in the discussion of SRS simply because it is
automatic, as indeed it is for the optical Kerr effect. Had the spatial aspects of the travelling
waves been shown explicitly in Eq. (5.30), that equation would have read

2

P (S ) exp{i(S t k S .r)} = 32 0 (3) E(
(5.37)
L ) E(
S ) exp{i(S t kS .r)}
where
2
2

E(L ) = E(
L ) exp{i(L t kL .r)} .
(5.38)
All the space-time terms therefore cancel out, which is another way of saying that the wave
vector mismatch is zero, namely
k = kS (kL kL + kS ) = 0.
(5.39)
90
This equation also highlights the fact that, since the SRS process is automatically phase
matched, the Stokes wave is free to grow in any direction it chooses. Of course, the laser
field is normally in the form of a beam, and so Stokes waves propagating in roughly the
same direction will receive preferential growth simply because that gives them the best
opportunity to feed off the laser energy. But this criterion could equally be met by a Stokes
wave travelling in the opposite direction to the pump and, indeed, backward stimulated
Raman scattering is a viable process.
5.3.2 Anti-Stokes generation and higher-order effects

It would be more accurate to refer to the frequency component at S = L ; as the first
Stokes wave because, once generated, it can become the pump for a further Raman process
in which a second Stokes wave at S2 = L 2; is created; see Fig. 5.4(b). A cascade of
further down-shifted frequencies may occur.
Up-shifted (anti-Stokes) frequency components can also be generated through the Raman
resonance. There are at least two possible ways in which the first anti-Stokes (AS) component at AS = L + ; can be produced. First of all, by analogy with Eq. (5.30), it could
arise from the anti-Stokes polarisation term

2

P (AS ) = 32 0 (3) (AS ; L , L , AS ) E(
(5.40)
L ) E(
AS ).
As with Eqs (5.30) and (5.31), the key issue is the sign of the imaginary part of the coefficient.
Detailed examination of the quantum mechanics (see Chapters 8 and 9) reveals that if most
of the population is in the ground state, the wave at AS will decay, but if a population
inversion exists between states 0 and 2 in Fig. 5.4, the anti-Stokes wave will grow and the
Stokes wave will decay. The anti-Stokes process is represented by Fig. 5.4(c); a photon is
lost at L , one is gained at AS , and the difference in energy is provided by de-excitation
of the medium. These conclusions are entirely reasonable. If the population is mostly in the
ground state, the process that moves population from state 0 to state 2 dominates, while if
the population of state 2 is larger than that of state 0, the reverse process takes precedence.
Up-shifts are also possible through the sequence shown in Fig. 5.4(d), where the antiStokes wave is produced by the four-wave mixing combination L +L = AS +S , where
two laser waves combine to create a Stokes wave and an anti-Stokes wave. This is sometimes
referred to as parametric anti-Stokes Raman generation, even though a Stokes wave is driven
as well. Whereas the Raman processes discussed so far have all been automatically phase
matched, this parametric process17 involves the phase-matching condition
kAS = kAS + kS 2kL .
(5.41)
The corresponding wave vector diagram for phase-matched anti-Stokes generation is shown
in Fig. 5.5. If the optical dispersion curve in the region between S and AS were a straight
line, it is easy to show that the anti-Stokes generation process would be phase matched
when all three waves propagated collinearly. Normally, the curvature of the line ensures
17 See Section 9.4 for a discussion of the word parametric.
91
kL
kL
kAS
Fig. 5.5
kS
Wave vector diagram for parametric anti-Stokes scattering.

that phase matching occurs when the Stokes and anti-Stokes waves are at a small angle
to the laser beam. This gives rise to the ring structure which is characteristic of Raman
generation via this (and other) four-wave mixing process.
There are many other possibilities too. For example, the four-wave process L + S =
AS + S2 , where collinear laser and Stokes waves create angled anti-Stokes and second
Stokes waves, can be phase matched in a similar scheme to Fig. 5.5. And a new and exciting
scenario is created if two strong waves with frequency separation ; are inserted into the
Raman medium. In this case, a wide frequency comb of waves extending in both the Stokes
and anti-Stokes directions is created in a process called ultra-broadband multi-frequency
Raman generation [4042].
5.3.3 Raman waves

In the treatment of Section 5.3.1, the physical origin of the Raman resonance was hidden
inside the nonlinear coefficient. We noted that the resonance could be electronic, vibrational,
or rotational in origin, but the characteristics of the associated oscillations did not feature
anywhere in the analysis. We will now discuss an alternative approach to SRS in which the
Raman oscillations appear explicitly in the calculation.
For the purposes of discussion, we will assume that the Raman resonance has its origin
in the molecular vibrations in a liquid. An important concept to grasp at the outset is that
the vibrational oscillators at different positions in the nonlinear medium are driven by the
laser and Stokes waves in such a way that specific phase relationships are established
between them. Together, these oscillators create a travelling vibrational wave, with its
own wavelength, wave vector, and dispersion characteristics.18 The idea that disconnected
oscillators can be phased to create a travelling wave may be unfamiliar to some readers, but
sports fans will immediately recognise the analogy with the audience wave phenomenon,
where spectators stand and sit in a controlled time sequence, creating a wave that travels
around a large stadium.19
We start by going back to first principles and writing the equation for the linear polarisation
in the simple form
P = 0 N E
(5.42)
18 These have been called Raman waves in the section heading to cover the more general possibility that they
might alternatively be rotational or electronic in origin.
19 This is also known as a stadium wave or (in the UK) as a Mexican wave because it first attracted widespread
attention during the 1986 football World Cup in Monterrey, Mexico.
92
where N (= (1) = n2 1) is the linear susceptibility, and n is the refractive index.

Let us suppose that the microscopic susceptibility of a molecule depends weakly on a
coordinate (call it q) that measures the vibrational amplitude such that
(q) = 0 + 1 q +
(5.43)
where 1 = (/q)|q=0 . Suppose also that, within the liquid, the molecular displacements
are so phased that as a function of time and position, q forms a travelling wave of angular
frequency ; and wave vector K given by
q = 12 q exp{i(;t K.r)} + c.c.
(5.44)
with an associated wave of refractive index

(5.45)
n(q)
= n0 + (1 /2n0 ) [q exp{i(;t K.r)} + c.c.] .

Thus, if E = 12 E L exp{i(t kL .r)} + c.c. , it follows from Eqs (5.43)(5.44) that the
polarisation in Eq. (5.42) will include terms such as

P = + 12 0 N 1 E L q exp{i(S t (kL K).r}

+ E L q exp{i(AS t (kL + K).r)} +
(5.46)
where S = L ; and AS = L + ;. These terms serve as source antennae for the first
Stokes and first anti-Stokes waves. Inserting the Stokes term into the general coupled-wave
equation (2.8) yields
S
E S
N 1 E L q
= i
z
2cnS
(5.47)
while the analogous equation for the laser field is

E L
L
N 1 E S q.
= i
z
2cnL
(5.48)
The corresponding differential equation for q is

2q
2 q
F
+
+ ;20 q =
2
t
t
m
(5.49)
where F is the effective force driving the vibrations, m is the mass, ;0 the natural resonance
frequency, and the damping time. An expression for the force can be found from F =
du/dq, where u = 12 0 (q)E 2 is the oscillator energy. With the help of Eq. (5.43),
we obtain F = 12 0 1 E 2 , where the time-averaging overbar removes the optical frequency
terms. We now substitute F into Eq. (5.49) along with Eq. (5.44) for q, and we assume that
both the laser and Stokes fields are already present so that

(5.50)
E = 12 E L exp{i(L t kL .r)} + E S exp{i(S t kL .r)} + c.c. .
93
we obtain
Neglecting terms in q and D q,

q
q
0 1
+ = i
E L E S
t
4m;
(5.51)
where we have set ; = ;0 and kL kS = K.

Equations (5.47), (5.48) and (5.51) are the differential equations governing the three
waves involved in the stimulated Raman interaction. In the strongly damped case, it is easy
to get from here to the results of the previous subsection. When in Eq. (5.51) is small, the
time derivative can be dropped, and the vibrational amplitude, which is now slaved to the
laser and Stokes fields, is

0 1
q = i
E L E S .
(5.52)
4m;
When this result is substituted into Eqs (5.47) and (5.48), these equations become

0 N 12 2
E S
S
(5.53)
=+
E L E S
z
2cnS
4m;

0 N 12 2
E L
L
(5.54)
=
E S E L
z
2cnL
4m;
which are consistent with Eqs (5.32) and (5.35) if
R =
N 0 12
.
6m;
(5.55)
The terms in curly brackets in Eqs (5.53) and (5.54) are the coefficients governing gain at
the Stokes frequency, and loss at L . At the beginning of the process, the Stokes field is
small, and the laser field E L is approximately constant. At all stages, the local value of q
follows from Eq. (5.52).
Notice how the final few steps have served to hide the vibrational wave from view once
again. As soon as Eq. (5.52) is substituted into Eqs (5.47) and (5.48), and the brackets in
Eqs (5.53) and (5.54) have been replaced by R from Eq. (5.55), all explicit reference to
the molecular vibrations disappears.
To gain further physical insight, we now consider the properties of the vibrational wave
of Eq. (5.44). We have seen that the wave consists of a spatial distribution of vibrating molecules whose phases are determined by the interference of the laser and Stokes
waves according to Eq. (5.52). In fact, the wave vectors of all three waves are linked by
kL = kS + K, which can be represented by the vector triangle of Fig. 5.6. The figure
includes a hint of the wavefront structure of the Raman wave, and this raises the question
of its dispersion characteristics. In most circumstances, the oscillators participating in SRS
interact with each other extremely weakly, and this means that there is virtually no constraint on the magnitude of K, or on the wavelength of the wave. The dispersion curve is
therefore essentially a flat horizontal line positioned at frequency ; for all |K|, as shown
schematically in Fig. 5.7. The velocity of the wave is of course ;/K and, since ; is essentially fixed while K is free to vary, the velocity varies widely too. Similar attributes apply
94
kL
K
kS
Fig. 5.6
Wavevector triangle for stimulated Raman scattering showing the wave vector K of the Raman wave.
Raman waves
optical phonons
acoustic
waves
K
Fig. 5.7
Schematic dispersion diagram (not to scale) for Raman waves (and optical phonons) (top), and acoustic waves
(bottom).
incidentally to optical phonons, which mediate the stimulated Raman process in crystalline
materials.20
Since the magnitude of K is unconstrained, the direction of ks in Fig. 5.6 is not fixed, and
this confirms the fact that the Stokes wave is free to grow in any direction it chooses. This
is another way of saying that the stimulated Raman process is automatically phase matched
in all directions.
20 Acoustic phonons in a solid are basically sound waves, in which neighbouring atoms are generally displaced
in the same direction. For optical phonons, on the other hand, neighbouring atoms are displaced in opposite
directions; the oscillations are therefore of a very different kind, and have very different dispersive properties.
95
5.4 Interaction of optical and acoustic waves
Figure 5.6 highlights another interesting aspect of the Raman wave approach, namely that
it makes SRS look like a three-wave process, whereas the viewpoint of Section 5.3.1 made it
seem like a four-wave process. We have seen the transition from one viewpoint to the other
in Eqs (5.52)(5.55). The distinction between three- and four-wave processes is evidently
less clear-cut than we might have thought. Earlier in the book, three-wave processes have
normally involved three electromagnetic waves coupled by a (2) nonlinearity, and we have
become accustomed to the idea that these can only be supported in media lacking a centre of
symmetry. As we have seen, however, the Raman waves we are dealing with here are waves
of vibration or rotation or (in the case of electronic states) waves of excitation, created in
the interference between the laser and Stokes waves. There is no one-wayness in them,
and so no violation of symmetry principles.

5.4.1 Physical basis
In their simplest form, acoustic waves are waves of compression and rarefaction, waves
therefore in which the density varies in space and time about its mean value. Since refractive
index depends on density, acoustic waves are associated with waves of refractive index,
and we have already seen in Section 5.3.3 how a refractive index wave (see Eq. 5.45)
couples the laser and the Stokes waves in stimulated Raman scattering. This suggests that
light and acoustic waves can interact too, and acousto-optics is the name given to this field
of study.
The dictionary definition of acoustic is relating to sound, but it must be stressed at
the outset that the acoustic frequencies we will be dealing with in this section are far higher
than 20 kHz, which is normally regarded as the upper limit of hearing for a young person.
We shall rather be considering the frequency range from a few MHz up to a few GHz, which
is at the high end of the ultrasound spectrum.
A number of different physical processes are potentially involved in the interaction
of light and ultrasound, and it is worth listing some of them and summarising their key
features.
The photoelastic effect: Also known as the piezo-optic effect, this concerns the change
in the dielectric properties of a material in the presence of strain. Since photoelasticity is

often measured by recording changes in birefringence, it is sometimes thought to occur
only in anisotropic media. But in fact the symmetry properties of the photoelastic tensor
are identical to those of the third-order coefficients described in Section 5.2.1 above, and
so photoelasticity can certainly occur in amorphous media.
Electrostriction: This is a universal property of dielectrics in which stress and strain are
created in the presence of an electric field. The effect varies as the square of the applied
96
field (so it is not reversed when the sign of the field is reversed), and it is sometimes
known as quadratic electrostriction for this reason. Thermodynamic arguments indicate
that electrostriction and photoelasticity are closely interrelated.
Piezoelectricity: The piezoelectric effect describes the appearance of an electrical polari-
sation in a material when a strain is applied. Although this sounds a little like photoelasticity,
it differs insofar as piezoelectricity occurs only in non-centrosymmetric crystals. Indeed the
symmetry properties of the piezoelectric tensor are identical to those governing second-order
nonlinear effects such as second harmonic generation.
The inverse piezoelectric effect:21 As the name implies, this is piezoelectricity in
reverse. Instead of stress/strain causing the medium to become polarised, the inverse piezoelectric effect relates to the creation of stress/strain in a non-centrosymmetric medium
by the application of an electric field. Unlike electrostriction, the effect is directly proportional to the applied field, and the sign changes when the field is reversed. In broad
terms, inverse piezoelectricity is to electrostriction as the Pockels effect is to the DC Kerr
effect.
Optical absorption: If two optical beams interfere in an absorbing medium, the material
will tend to heat up in the region of the anti-nodes, and the associated thermal expansion
can drive an acoustic wave.
The two most important processes involved in the coupling of optical and acoustic waves
are usually photoelasticity and electrostriction. The first describes how ultrasound affects
light through the effect of stress and strain on refractive index, while the second describes
how light creates stress and strain, enabling two optical waves to drive an acoustic wave. It
is worth noting that processes of these two kinds were implicated in stimulated Raman
scattering; see Eqs (5.42)(5.45) for the first kind and the discussion after Eq. (5.49)
for the second. As we shall discover, a Raman-type process known as stimulated Brillouin scattering (SBS) exists in which an acoustic wave plays the role of the vibrational
wave of Section 5.3.3. However, there are important differences between SBS and SRS,
which arise because acoustic frequencies are much lower than vibrational frequencies,
and because the dispersion characteristics of ultrasound are very different from those of
Raman waves. For the vibrational waves of Section 5.3.3, we have seen that frequency
is virtually independent of wavelength, so the schematic dispersion relation in Fig. 5.7 is
essentially horizontal. In contrast, the dispersion relation for an acoustic wave is much
closer to a straight line through the origin, and this is represented by the lower plot in
Fig. 5.7; the gradient of the line is basically the velocity of an acoustic wave. It must be
stressed that the figure is purely schematic, and completely out of scale. The ultrasound
frequencies we will be considering are four to five orders of magnitude smaller than typical
vibrational frequencies so, if Fig. 5.7 were drawn to scale and the Raman wave line were
21 Also known as the converse, reciprocal, or reverse piezoelectric effect.
97
visible at the top, the acoustic dispersion line would be indistinguishable from the K-axis
at the bottom.
In Sections 5.4.2 and 5.4.3, we will consider two situations where a laser signal is
diffracted from an acoustic wave generated externally, that is itself essentially unaffected
by the diffraction process. Then, in Section 5.5, we will move on to stimulated Brillouin
scattering, in which the acoustic wave plays the role of the Raman wave in stimulated
Raman scattering.
5.4.2 Diffraction of light by ultrasound

Consider an acoustic wave of angular frequency ;A travelling in the y-direction. The
longitudinal acoustic displacement Q can be written

exp{i(;A t KA y)} + c.c. ,
Q = 12 Q
(5.56)
is the slowly varying complex amplitude, KA = ;A V is the angular wave number,
where Q
and V is the wave velocity. Since in one dimension, strain S = Q/y, the associated
acoustic strain wave is given by

exp{i(;A t KA y)} + c.c. .
S = 12 iKA Q
(5.57)
The presence of strain influences the propagation of optical fields through changes in the
dielectric constant given by
= e S
where e is the electrostrictive constant.22 It follows that

exp{i(;A t KA .r)} iKA Q
exp{i(;A t + KA .r)}
= 12 e iKA Q
(5.58)
(5.59)
where the equation has been generalised to allow for propagation in an arbitrary
direction. The change in the dielectric constant leads to a nonlinear polarisation
P NL
= 0 E. If the field in this
equation is an incident laser beam given by E =
1
L exp{i(L t kL .r)} + c.c. and we use Eq. (5.58) for , we obtain
E
2

exp{i(L + ;A )t (kL + KA ).r)}
P NL = 14 0 e KA i E L Q

exp{i(L ;A )t (kL KA ).r)} + c.c. .
i E L Q
(5.60)
The terms in this equation clearly represent optical antennae driving up-shifted and downshifted fields in new directions, in other words the diffraction of light by ultrasound. A
schematic diagram of the interaction is shown in Fig. 5.8, where the acoustic wave is
22 The electrostrictive constant is defined by = (/)
e
= where is the mean value of the density .
98
k+
kL
k
ACOUSTIC WAVE
Fig. 5.8
Diffraction of an optical wave by an acoustic wave.

shown propagating at right angles to the incident laser wave, creating diffracted waves in
the directions of kL KA . The process is used in acousto-optic modulators and deflectors.
The diffraction angles drawn in the figure are roughly 15 , but a simple calculation
shows that deflections of this magnitude are hard to achieve. The angular wave number of
an optical wave at 1 m is 2 106 rad s1 and, since tan 15
= 0.25, the magnitude of
1
6
1
KA needs to be about 2 10 rad s . Since the velocity of ultrasound in solids is around
4000m s1 , the necessary acoustic frequency is
;A
106 4000
=
= 1 GHz.
(5.61)
2
4
While acousto-optical devices at a few GHz are now available, frequencies in the 20200
MHz range are more common.
Although 1 GHz is a high frequency for ultrasound, it is very small compared to typical
optical frequencies of around 300 THz. It follows that the relative optical frequency shifts
in acousto-optical diffraction are minute, and so the angular wave numbers of the incident
and scattered beams are virtually the same (i.e. k+
= k
= kL ). This is relevant to the phase
matching conditions. Figure 5.9(a) shows that, if the acoustic wave is perpendicular to the
incident light wave, the wave vectors cannot form a closed triangle. A simple calculation
(see Problem 5.7) shows that the wave vector mismatch in this case is k
= 12 kL 2 , and
so the coherence length for the process is
L I2A
Lcoh
= 2 =
(5.62)
where L and IA are the respective optical and acoustic wavelengths. This works out as
around 16 m for L = 1 m and = 15 , although the figure rises to around 150 m
at 5 .
Acousto-optic devices in which the interaction length is substantially smaller than Lcoh
are said to operate in the RamanNath regime. Longer interaction lengths can be realised
under phase-matched conditions. Phase matching is easily achieved by angling the light
and acoustic waves so that one of the triangles is closed, as shown in Fig. 5.9(b). The figure
99
k+
KA
k+
KA
kL
kL
KA
k
(a)
Fig. 5.9
KA
k
(b)
Wavevector triangles for (a) the acoustic wave perpendicular to the optical wave (as in Fig. 5.8), (b) a slanted acoustic
wave to ensure the phase matching of the k wave.
shows the case where
kL = k + KA .
(5.63)
Notice that, since kL

= k , the closed triangle in Fig. 5.9(b) is effectively isosceles. Equation
(5.63) is similar to the Bragg condition in X-ray diffraction, and indeed the case where phase
matching is achieved in acousto-optics is known as the Bragg regime. Whereas in X-ray
diffraction, the grating arises from a stationary spatial pattern of atoms in a crystal, here
it characterises a travelling acoustic wave. In fact, it is precisely because the acoustic wave
presents a refractive index grating moving slightly away from the incident laser beam in
Fig. 5.9(b), that the k beam is slightly down-shifted (by ;A ) in angular frequency. The
frequency change is consistent with the Doppler shift from a moving mirror. An even closer
analogy with X-ray diffraction occurs in a process called stimulated Rayleigh scattering
where the laser beam scatters from an essentially stationary refractive index pattern in the
nonlinear medium.
5.4.3 The acousto-optic dispersive filter

The acousto-optic programmable dispersive filter (AOPDF) devised by Tournois [43], commonly known by its trade name (the Dazzler), is designed to control and customise the
structure of femtosecond optical pulses. Consider, in the simplest case, a pulse of acoustic
waves given by

t) exp{i(;A t KA z)} + c.c. .
Q(z, t) = 12 Q(z,
(5.64)
The pulse travels through a birefringent crystal in the z-direction, collinearly with two
optical pulses, the input (1) and the output (2). One of the optical pulses (say the input)
travels as an ordinary wave, and the other (the output) as an extraordinary wave. A schematic
diagram is shown in Fig. 5.10.
Phase matching is crucial to the operation of the device. The key idea is to tailor the
acoustic pulse so that the phase-matching condition determines the point in the crystal
at which the input channel couples to the output channel. Because all three waves are
collinear, the condition can be written in scalar form as k2 = k1 + KA . The frequency
condition 2 = 1 + ;A naturally applies too and, since acoustic frequencies (typically
100
ORDINARY
EXTRAORDINARY
k1
KA
k2
Fig. 5.10
Collinear conguration for an acousto-optic dispersive lter.

50 MHz in an AOPDF) are tiny compared to optical frequencies, 1 and 2 will be the
same to about 1 part in 107 . However, k1 and k2 will be significantly different because
they are associated with ordinary and extraordinary waves, which have different refractive
indices. The relationship between the three angular wave numbers under phase-matched
conditions is shown in the lower part of Fig. 5.10.
The velocity of acoustic waves is typically more than five orders of magnitude smaller
than the velocity of light, so the acoustic pulse is effectively frozen in the crystal as far as
the optical pulses are concerned, as indicated in the figure. Hence we can remove the time
dependence from Eq. (5.64) and write

Q(z) = 12 Q(z)
(5.65)
exp{iKA (z)z} + c.c.
where the phase characteristics of the acoustic pulse have been absorbed into KA , which is
now a function of z. Each spectral component of the input pulse will therefore couple into
the output channel at the point in the acoustic wave where the phase-matching condition is
satisfied, in other words when23
|n e () no ()|
.
(5.66)
c
Once the dispersion characteristics of the crystal are known, this equation can be solved
numerically to give z as a function of . For a crystal of length L, the frequency-dependent
phase accumulated by the optical pulses in the crystal is easily shown to be

() = no z() + n e ()(L z())
(5.67)
c
which controls the phase of the transfer function T () from input to output. However, the
process also allows
the

amplitude of the transfer function to be controlled, because this is

determined by Q(). Overall we can write
KA (z) =
E 2 () = T ()E 1 ()
23 n () is the refractive index of the extraordinary wave; see Section 3.3.3.
e
(5.68)
5.5 Stimulated Brillouin scattering
101
which implies that amplitude and phase modulation of the optical pulse can be effected
simultaneously provided an acousto-optic pulse with the appropriate amplitude and phase
structure can be constructed. The problem in practice is of course to find the acoustic profile
needed to deliver a desired T (), while compensating for the group velocity dispersion
introduced by the device itself at the same time.24 This non-trivial task has to be done using
a specially designed software package.
In summary, the AOPDF enables the structure of a femtosecond pulse to be managed in
a programmable way through the mediation of a precisely tailored acoustic signal. However, the details are usually rather more complicated than the simple treatment given here
suggests. For example, the acoustic pulse might be a shear wave rather than a compression
wave and, for various reasons, a non-collinear beam configuration is necessary in the popular
acousto-optic crystal TeO2 . Moreover, since acousto-optic crystals tend to be mechanically
as well as optically anisotropic, the acoustic Poynting vector will not necessarily be parallel
to KA , and so acoustic walk-off will need to be taken into account as well.

5.5.1 Introduction
The previous two sections dealt with the diffraction of light by ultrasound generated externally by an acoustic transducer. We now consider a process, similar to stimulated Raman
scattering, in which an acoustic wave plays the role of the Raman wave, and the Stokes and
acoustic waves are driven by the laser. First observed in 1964 [44], the process might logically be called stimulated acoustic Raman scattering, but it is in fact known as stimulated
Brillouin scattering (SBS).
There are close parallels between SBS and SRS, but significant differences too. The
most important is that, for reasons connected with the dispersion characteristics of acoustic
waves, the Stokes wave in SBS is normally observed in the backward direction. In the typical
configuration shown in Fig. 5.11, the laser beam is focused into the nonlinear medium where
A
L
Fig. 5.11
Schematic diagram of backward stimulated Brillouin scattering.

24 The group velocity dispersion of optical pulses is treated in Chapters 6 and 7.
102
the SBS interaction takes place in the region of the focus. The laser drives a forward-going
acoustic wave and a backward-directed Stokes wave, which is picked off by the partially
reflecting mirror beyond the lens.
The most common process by which two optical waves drive an acoustic wave is based
on the longitudinal electrostrictive effect. The existence of an electrostrictive force can be
deduced from Eq. (5.58) by inferring that the stored electrostatic energy density is changed
in the presence of strain by ue = 12 0 e SE 2 . This in turn implies the existence of a
pressure p = ue /S = 12 0 e E 2 , from which it follows that the force on unit volume of
material resulting from optical fields travelling along the z-axis is
Fv =
p
(E)2
= 12 0 e
.
z
z
(5.69)
The equation of motion for the acoustic displacement Q in response to the force is
(E)
1
2 0 e z
1 2Q
Q
2Q
D
=
C z2
t
t 2
(5.70)
where the force terms on the left-hand side respectively represent electrostriction, elasticity,
and loss. The parameter C is the compressibility of the medium, is the mass density, and
D is the acoustic damping coefficient.
It has been mentioned already that the Stokes wave in SBS travels in the backward
direction, and we now need to consider why that is. It is in fact quite easy to see that
collinear propagation in which all waves travel in the same direction is not a viable option.
As for the programmable dispersive filter (PDF) of Section 5.4.3, the conditions to be
satisfied are L = S + ;A , and kL = kS +K A and, as in that case, the laser and Stokes
frequencies (L and S ) will be virtually identical. For the PDF, the magnitudes of kL
and kS (the angular wave numbers kL and kS ) were significantly different because one
wave was ordinary and the other extraordinary. But SBS is normally observed in isotropic
materials, so kL and kS will now be virtually equal, while the magnitude of KA (=KA ) will
be determined by the angle in the vector triangle of Fig. 5.12(a). Clearly, if the laser and
Stokes waves co-propagate, kL and kS will be parallel, KA will be extremely small, the
acoustic wavelength will be extremely large, and ;A (= KA V ), which was already small,
will be effectively zero.
It will also emerge later that the Brillouin gain is proportional to KA , so the gain would
in any case be negligible under collinear conditions. Anyway, with this point in mind, it
is desirable to make KA as large as possible and, from Fig. 5.12(a), it is clear that this
is achieved when the triangle is opened right out, and the laser and Stokes wave counterpropagate, as shown in Fig. 5.12(b). The acoustic wave travels in the +z-direction and the
laser wave scatters off it, as from a moving mirror, leading to a Stokes wave travelling
in the z-direction with a marginal frequency down-shift. It is clear from the figure that
KA = kL + kS
= 2kL
= 2kS , and it follows that the frequency shift is given by
;A 2nL V
= vac
2
L
(5.71)
103
kL
(a)
kS
kS
KA
kL
KA
z=0
z=L
(b)
Fig. 5.12
Wave vector orientations in stimulated Brillouin scattering: (a) general angle; (b) the backward case.
where V (= ;A /KA ) is the speed of ultrasound in the nonlinear medium, and vac
L is the
vacuum wavelength of the laser. Values of ;A /2 around 5 GHz are typical for liquids,
with rather lower frequencies in gases, and rather higher frequencies (1525 GHz) in solids.
5.5.2 The coupled-wave equations of stimulated Brillouin scattering

To analysis the SBS process, we write the optical and acoustic waves in the form

E = 12 E L exp{i(L t kL z)} + E S exp{i(S t + kS z)} + c.c. ,

exp{i(;A t KA z)} + Q
exp{i(;A t KA z)}
Q = 12 Q
(5.72)
(5.73)
where the first equation includes the incident laser beam and the counter-propagating Stokes
wave; notice the plus sign in the Stokes wave argument. We substitute Eqs (5.72) and (5.73)
into Eq. (5.70) and, after some algebra, we obtain
Q
DQ
0 e
EL ES
+
=
z
2V
8V 2
(5.74)
where we have used L S = ;A and kL + kS = KA . The proof uses the relation V =

have been dropped.
(C)1/2 , and several first and second space and time derivatives of Q
Equation (5.74) governs the process in which the optical fields drive the acoustic wave;
notice particularly the presence of the electrostrictive constant with the laser and Stokes
fields on the right-hand side. The equation for the Stokes field can be found from the second
term on the right-hand side of Eq. (5.60). The nonlinear polarisation that drives the wave
at S is

E L exp{i(S t + kS z)} + c.c.
PSNL = 12 12 i0 e KA Q
(5.75)
1
= 2 PS exp{i(S t + kS z)} + c.c.
104
where the second step defines PS . We again customise the standard coupled-wave equation
(2.8) to the wave at S and, when PS is inserted into the right-hand side, the result is

E S
S e KA
EL Q .
(5.76)
=
z
4cnS
is
The analogous equation for the laser field (based on PL = 12 i0 e K E S Q)

E L
L e KA
=
ES Q.
z
4cnL
(5.77)
The three coupled differential equations (5.74) and (5.76)(5.77) govern the evolution of
the three fields in the Brillouin interaction. If we now introduce the new parameters
A = Q,
a=
D
,
2V
b=
0 e
,
4V D
cS =
S e KA
L e KA
and cL =
,
4cnS
4cnL
the equations take on the more manageable appearance

A
+ a A = abE L E S
z
E S
= cS E L A
(z)
E L
= cL E S A.
z
(5.78)
(5.79)
(5.80)
We have defined A as the negative of the acoustic displacement since this makes the signs
more convenient, and we have also included a negative sign in the derivative of Eq. (5.79)
to highlight the fact that the Stokes wave propagates in the backward direction.
5.5.3 Approximations and solutions

Solutions to Eqs (5.78)(5.80) can be obtained within several different approximations.
First of all, the equation for A can be simplified if one or other of the terms on the left-hand
side can be ignored. Discarding the second term amounts to neglecting acoustic damping
which, though rarely appropriate, leads to an interesting solution, which we discuss later. In
the case of strong acoustic damping, which is normally more realistic, the spatial derivative
is dropped, and this leads immediately to
A = bE L E S .
(5.81)
This means that the acoustic wave is slaved to the laser and Stokes waves, just like the
vibrational wave in the SRS case in Eq. (5.52). When Eq. (5.81) is substituted into Eqs (5.79)
and (5.80), we obtain

2
E S

(5.82)
= + bcS E L E S
(z)

2
E L

(5.83)
= bcL E S E L
z
105
from which it is easy to show that

1 IS
1 IL
=
.
S d(z)
L dz
(5.84)
This means that for each Stokes photon gained in the z-direction, a laser photon is lost in
the +z-direction.
The equations can be simplified further if depletion of the laser field is negligible.
2

Equation (5.83) can then be discarded, and E L treated as a constant in Eq. (5.82), which
now has a solution in which the Stokes wave grows exponentially in the z-direction, with
a gain coefficient given by
2

2
S e2 KA 0 E L

gS = bcS EL =
.
16cnS V D
(5.85)
In a cell of length L, as suggested in Fig. 5.12(b), the solution is

E S (z) = E S (L) exp{gS (L z)}
(5.86)
where E S (L) is the Stokes field at the rear surface, which would be the input field in the case
of a Brillouin amplifier. The form of the solution is plotted in Fig. 5.13, where E S (L) = 1
(arbitrary units), and gS L = 5. The Stokes field grows from right to left, and because the
acoustic wave is tied to the Stokes wave through Eq. (5.81), A is strongest at z = 0 and falls
away exponentially thereafter. This is despite the fact that the acoustic wave propagates in
the +z-direction. As for SRS, the Stokes wave will grow from noise if the gain coefficient,
which depends on the laser intensity through Eq. (5.83), is sufficiently high.
amplitudes (arb.units)
150
STOKES
100
50
ACOUSTIC
0
0
0.2
0.4
0.6
0.8
z/L
Fig. 5.13
Stokes and acoustic wave amplitudes as a function of distance for stimulated Brillouin scattering in the case of strong
acoustic damping. Laser depletion is ignored.
106
An analytical solution is also possible when laser depletion and acoustic damping are both
ignored. As noted earlier, the latter assumption is usually unrealistic, but it is instructive to
pursue this case because it has some interesting features. As an added bonus, the system
comes with the impressive title of contraflow Hermitian parametric coupling [45], so it is
clearly worthy of attention! The equations to be solved are now
A
= r E S ;
z
E S
= s A
(z)
(5.87)
where r = abE L and s = cS E L . It is easy to show that a solution of the type A(z)
=
A(0) exp{gz} and ES (z) = ES (0) exp{gz} exists. However, g is now imaginary, and given
by
g = i rs = i
(5.88)
where

= abcS |EL |2 =
S e2 KA 0 |EL |2
.
32cnS V 2
(5.89)
This implies that the solutions are oscillatory, but there turns out to be a singularity after a
quarter of a cycle, so complete oscillations are never seen.
Once again, we consider the evolution of A and E S in the configuration of Fig. 5.12(b).
As before, the pump wave enters at z = 0, while the Stokes wave develops in the backward
direction. If we set the boundary condition that A(0) = 0, the solution is
A(z)
=
where
r E S (L) sin z
;
cos L
r
= eiL
cos z
E S (z) = E S (L)
cos L
(5.90)
c0 nS
.
2S KA V 2
The key feature of this result is the dependence of both amplitudes on (cos L)1 . This
implies that the gain factors of the Stokes and acoustic waves become infinite when L
/2, which occurs when the laser intensity reaches
IL =
4 2 c2 nS nL V 2
.
S e2 KA L2
(5.91)
This is the threshold for stimulated Brillouin scattering in this albeit unrealistic case.
A plot of the solutions given in Eq. (5.90) is shown in Fig. 5.14 for the case where
L = 0.99 (/2) for which (cos L)1 = 63.7. The field units are arbitrary and we have
set E S (L) = r/ = 1 for the purposes of illustration. The acoustic wave grows to the right
and the Stokes wave to the left, each field being linked to the gradient of the other through
Eq. (5.87).
Notice especially the presence of KA in the numerator of Eq. (5.89) and the denominator
of Eq. (5.91), confirming that the Stokes wave in stimulated Brillouin scattering grows most
strongly when KA is maximised, which is in the backward direction according to Fig. 5.12.
Recent review material on SBS can be found in [46, 47].
5.6 Optical phase conjugation
107
70
STOKES
ACOUSTIC
amplitudes (arb. units)
60
50
40
30
20
10
0
0
0.2
0.4
0.6
0.8
z/L
Fig. 5.14
As for Fig. 5.13, but with acoustic damping neglected.

5.6.1 Phase conjugate beams
When an optical beam passes through an aberrating medium, it emerges with a distorted
wavefront. If it is then reflected back through the medium by a normal mirror, the damage is
compounded. By contrast, phase conjugate reflection creates a beam that retraces its steps
through the medium as in a time-reversed movie so that the beam emerges with its aberrations removed. Phase conjugate beams have important applications in laser engineering
and, in this section, we will consider how they can be created.
Before discussing how a phase conjugate wave can be generated, we need to understand
what it is, and how it differs from a normal wave. As usual, we start by considering a beam
(call it wave 1) travelling broadly in the +z-direction, namely

E1 (r, t) = 12 E 1 (r) exp{it} + E 1 (r) exp{it} .
(5.92)
Here E 1 (r) = E 1 (r) exp{ik1 z}, and E 1 (r) is slowly varying with respect to r =
(x, y, z).25 As the name suggests, the corresponding phase conjugate wave EC (r, t) is
obtained by replacing E 1 (r) in Eq. (5.92) by its complex conjugate E 1 (r) so that

EC (r, t) = 12 E 1 (r) exp{it} + E 1 (r) exp{it} .
(5.93)
But the terms could equally be written in reverse order, in which case we would have

EC (r, t) = 12 E 1 (r) exp{it} + E 1 (r) exp{it} = E1 (r, t)
(5.94)
25 We are assuming a continuous (CW) beam. More generally, the fields would of course be time dependent.
108
where the final step shows that the phase conjugate wave is identical to the time reversed
version of the original wave, which is obtained by changing t to t in Eq. (5.92). The
conclusion is that replacing E 1 (r) in Eq. (5.92) by its conjugate is equivalent to reversing
the sign of t, so it is immaterial whether one regards the phase conjugate field as being E 1 (r)
running in forward time, or E 1 (r) running in reverse time. A movie tracing the evolution
of EC (r, t) would be identical in every respect to a movie of E1 (r, t) run backwards.
A trivial first example is the plane wave. While E 1 (z) = A exp{ik1 z} travels in the +zdirection, its conjugate E 1 (z) = A exp{+ik1 z} travels in the z-direction, and is equivalent
to E 1 (z) under time reversal. As a second example, consider the Gaussian beam, which is
given (see Section 2.5) by

2
A
r
exp 2
E 1 (r, z) =
(5.95)
ik1 z
1 iz/zR
w0 (1 iz/zR )
where r 2 = x 2 + y 2 , zR is the Rayleigh length, and w2 (z) = w02 (1 + (z/zR )2 ). Once again,
taking the complex conjugate is equivalent to replacing z by z, so the direction of travel
(or the direction of time if you prefer) is reversed.
It is now easy to see that phase conjugate reflection is fundamentally different from what
happens when a beam encounters a conventional mirror. The two cases are contrasted in
Fig. 5.15. The focus of the Gaussian beam is at z = 0, and the solid black line shows the
spherical wavefront (labelled 1) as it approaches the normal mirror M at z = zm . After
reflection, the beam continues to expand in the z-direction, as shown by the dotted line
RR, which is now curved in the opposite sense. The reflected wave is in fact equivalent
to a leftward travelling Gaussian beam centred at z = 2zm ; see Problem 5.9. By contrast,
the phase conjugate wave simply retraces the path of the original wave, a possibility that
is represented in the figure by the dotted line CC, which is curved in the same sense as
the original wave, and is contracting back towards z = 0. The difference in the width and
curvature of RR and CC highlights the difference between normal reflection and phase
conjugated reflection.
As noted at the start of this section, interest in phase conjugation is centred on the potential
cancellation of optical aberrations, so we now need to consider how aberrated beams behave
after reflection at a phase conjugate mirror. The key point always to keep in mind is that
M
R
C
1
0
zm
2zm
Fig. 5.15
Conventional reection (RR) and phase conjugate reection (CC) for a simple spherical wavefront.
109
2R
1
2C
Fig. 5.16
Spherical wavefront (1) is distorted (2) after propagating through an aberrating plate. Conventional reection creates
wavefront 2R, which will be further aberrated by its backward passage through the plate. The phase conjugated
wavefront 2C will have its aberrations removed as it returns through the plate creating wavefront C.
the phase conjugate wave is the same as the time-reversed wave so, if a beam acquires
aberrations in (say) an optically imperfect window, its phase conjugate reflection will pass
back through the window and emerge with the aberrations removed. A double pass through
a laser amplifier using a phase conjugate mirror to return the beam is clearly an attractive
prospect in laser engineering. However, it is important to appreciate that the procedure only
cancels linear optical aberrations, and that distortions stemming from nonlinear optical
phenomena are added rather than subtracted in the second pass.
To illustrate these points, consider the spherical wavefront labelled 1 in Fig. 5.16, where
it is shown approaching an aberrating plate from the left. As it passes through the plate, it
picks up phase distortions (x, y), and emerges as the distorted wavefront 2 given by
E 2 (r) = E 1 (r) exp{i(x, y)}.
(5.96)
The beam is shown expanding away to the right with ripples in its wavefront. If it were
reflected back by an ordinary mirror, it would return as wavefront 2R, which is curved in
the opposite sense. After passing through the plate a second time, the beam would emerge
doubly distorted, and would continue to expand away to the left. On the other hand, a phase
conjugate reflector would return wavefront 2 to the plate as wavefront 2C given by
E 2 (r) = E 1 (r) exp{i(x, y)}.
(5.97)
This has exactly the same shape as wavefront 2, but is now travelling in the opposite direction. Moreover, since it is equivalent to a time-reversed version of wavefront 2, it retraces
the steps of that wavefront, and emerges on the left as wavefront C with its aberrations
removed. The operation can be represented by the equation
E C (r) = E 1 (r) exp{i(x, y)} exp{i(x, y)} = E 1 (r).
(5.98)
110
The phase distortions cancel out, and a phase conjugate replica of the original clean beam
is recovered.
5.6.2 Phase conjugation by four-wave mixing

Various ways of generating phase conjugate beams have been studied. As we shall discover, the process is closely analogous to holography, but holography performed in real
time through a nonlinear interaction, rather than via the creation and development of a
photographic plate. Any medium in which a refractive index grating can be imprinted
and instantaneously read out is a potential candidate for mediating the phase conjugation
process.
Let us assume (see Fig. 5.17) that the wave to be phase conjugated (again denoted wave 1)
travels in the +z-direction, while a pair of waves (2 and 3) travel along a different line but
with oppositely directed wave vectors k2 and k2 . The three waves can be written

E1 = 12 E 1 (r) exp{i(t k1 z)} + E 1 (r) exp{i(t k1 z)}
(5.99)

E2 = 12 E 2 (r) exp{i(t k2 .r)} + c.c.
(5.100)

E3 = 12 E 3 (r) exp{i(t (k2 .r))} + c.c. .
(5.101)
Even without going into the fundamental mechanism of the process, the possibility of
phase conjugate wave generation can be spotted within the basic formalism of third-order
nonlinear processes. For if one works out the cube of the total field (E1 + E2 + E3 ), a term
appears in the nonlinear polarisation of the form
P4NL (r) E 2 (r)E 3 (r) exp{i2t} E 1 (r) exp{i(t k1 z)}
(5.102)
which comes from the product of the second term in Eq. (5.99) and the first terms in
Eqs (5.100) and (5.101). If waves 2 and 3 are plane waves, then E 2 (r)E 3 (r) = A2 A3 , and
3
4
1
Fig. 5.17
Phase conjugation by four-wave mixing. Wave 1 and plane wavefront 2 write a grating in the nonlinear medium. The
reverse plane wavefront 3 interacts with the grating to create the phase conjugated beam 4.
111
we have
P4NL (r) A2 A3 E 1 (r) exp{i(t + k1 z)}.
(5.103)
This polarisation antenna will therefore radiate a field (wave 4) in the backward direction,
with a wavefront that is the phase conjugate of the original wave. The interaction could be
mediated by the non-resonant four-wave mixing process defined by26
P4NL (r) = 32 0 (3) (; , , )E 2 (r)E 1 (r)E 3 (r)
(5.104)
where E 2 (r) = E 2 (r) exp{ik2 .r}, etc. in accordance with the definitions of
Eqs (2.4)(2.5).
It is instructive to divide the process into two distinct stages. In the first stage, the
interference of wave 1 and wave 2 (assumed plane) writes a stationary pattern in the medium
with the structure
12 (r) A2 E 1 (r) exp{i(k1 z k2 ).r}.
(5.105)
This comes from the product of the second term in Eq. (5.99) with the first term in
Eq. (5.100). In the second stage, wave 3 interacts with this pattern, creating a term in
the polarisation
P4NL (r) A3 12 (r) exp{i(t (k2 .r))} = A3 A2 E 1 (r) exp{i(t + k1 z)}
(5.106)
which is the form given by Eq. (5.103). In Fig. 5.17, the first stage is displayed in black and
the second stage in grey. The fine lines indicate the orientation of the stationary interference
pattern.Although these lines are drawn straight in the figure, in practice they will be distorted
by aberrations in wave 1, and it is of course precisely these distortions that wave 3 reads to
create the phase conjugate wave 4.
The roles of waves 2 and 3 could equally be reversed. In that case, instead of Eqs (5.105)
and (5.106), we would have
13 (r) A3 E 1 (r) exp{i(kz + k2 ).r}
(5.107)
and
P4NL (r) A2 13 (r) exp{i(t (k2 .r))} = A2 A3 E 1 (r) exp{i(t + k1 z)}
(5.108)
which is the same as Eq. (5.106). The corresponding diagram is shown in Fig. 5.18. Notice
that the interference pattern is at right angles to the one in Fig. 5.17, and the lines forming
the pattern are also drawn closer together because the wavelength will be smaller at these
angles.
The patterns of Eqs (5.105) and (5.107) take different physical forms depending on the
type of interaction involved. In the case of four-wave mixing envisaged in Eq. (5.104),
26 It is in fact necessary only for waves 2 and 3 to be conjugates of each other for the phase structure of the
2
waves to cancel out, and for Eq. (5.103) to apply. The pre-factor comes from 6 12 because six cross-terms
appear when the cube is taken.
112
3
1
4
2
Fig. 5.18
As Fig. 5.17, but with the roles of wavefronts 2 and 3 reversed.

12 (r) and 13 (r) would be second-order terms in the density matrix (see Chapter 8), and
the symbol has been chosen with that in mind. The interference patterns would certainly
register on a photographic plate, which points to the parallel with conventional holography.
In that case, the intention is normally the re-creation of the original wavefront, rather than the
production of a phase conjugated wave. The initial stage of the process would then involve
the combination of the first term in Eq. (5.99) with the second term in the reference wave
of Eq. (5.100) to produce the pattern
12 (r) E 1 (r)A2 exp{i(k2 k1 z ).r}.
(5.109)
The second stage would involve the illumination of this pattern with the reference wave to
re-create the original wavefront through the combination
P NL (r) 12 (r)A2 exp{i(t k2 .r)} = A22 E 1 (r) exp{i(t k1 z)}.
(5.110)
There would of course be an interval between the two stages, during which the photographic development of the three-dimensional hologram would take place. By contrast, the
two stages of phase conjugation by four-wave mixing occur simultaneously, and the term
real-time holography has been coined for that reason.
5.6.3 Phase conjugation by stimulated Brillouin scattering

Phase conjugation can also be effected by stimulated Brillouin scattering. In early experiments on SBS, it was noticed that the backward Stokes wave was so similar in structure to a
time-reversed version of the pumping wave that it worked its way back into the laser source
and interfered with its performance. It was soon recognised that the phase conjugating
property was an inherent feature of SBS.
At first sight, this phenomenon is unexpected, because there is nothing in Eq. (5.82) to
suggest that the Stokes wavefront E S should have any specific relationship to the conjugate
of the laser wavefront E L . On the other hand, Fig. 5.11 (shown enlarged in Fig. 5.19),
113
S L
LS
seed
Fig. 5.19
Schematic diagram of phase conjugation by stimulated Brillouin scattering.

indicates that the Stokes beam, propagating in the backward direction from the focal region
where the interaction is strongest, is likely to be far closer to E L than to E L for purely
geometrical reasons.After all, we have seen in Fig. 5.15 the difference between the reflection
from an ordinary mirror and from a phase conjugate mirror. A mirror placed at the focal point
in Fig. 5.19 would certainly cause the incoming beam broadly to retrace its steps, but this
would not be the case for any other mirror position. In particular, if we wanted to seed the
Stokes beam by launching a weak signal into the cell from the right, the wavefront (shown
as the dotted line S on the right of Fig. 5.19) should clearly be curved to match the rightward
propagating wavefront E L (labelled L in the figure). It should therefore correspond to the
time reversal of E L , in other words to E L . After passing through the SBS interaction region,
the Stokes beam wavefront will emerge on the left, curved in the same sense as the laser
wave as it approaches the focal region from the left, as shown in the figure.
The issue is therefore not so much whether the structure of the Stokes beam is closer to
the conjugate of E L than to E L itself it clearly will be. Rather it is whether the wavefront
of E S will match E L sufficiently accurately to display all the attributes of a phase conjugate
beam. The evidence of experiment is clearly that it does, so the burden is on the theorist to
explain why that is the case.
A qualitative argument runs broadly as follows.27 If the laser wavefront is aberrated, the
spatial structure of the beam throughout the focal region will exhibit corresponding spatial
complexity. If one then imagines a range of possible backward-propagating Stokes beams,
all competing for the energy of the incident laser through the SBS process, the winner of
the competition will be the beam that can march most accurately backwards through the
pattern laid down by the laser. This will be the beam that is closest to the phase conjugate
of E L .
A question remains. Does the fact that the wavelength of the Stokes beam is slightly
higher than that of the laser affect the argument? Does it compromise the accuracy needed
if the Stokes beam is to negotiate its way back along the path laid down by the laser? The
wavelengths (and frequencies) of the two differ in SBS typically by 1 part in 2 104 so,
for a laser at 1 m, serious mismatch between the two will occur over a distance of around
20 mm. But this is far longer than the effective length of the interaction region in a typical
SBS experiment, and suggests that the difference is not an issue.
Reviews of phase conjugate optics can be found in [4951].
27 A semi-quantitative argument along these lines is offered by Boyd [1]. A more complicated picture of phase
conjugation in SBS, based on Brillouin-assisted four-wave mixing, is presented by He and Liu [48].
114
5.7 Supercontinuum generation

At the beginning of the chapter, it was mentioned that when one third-order process occurs,
several others often do so at the same time. Nowhere is the tendency more evident than in
supercontinuum generation.28
In supercontinuum generation (SCG), an intense narrow-band incident pulse undergoes
dramatic spectral broadening through the joint action of several third-order processes, resulting in an output pulse of essentially white light. The effect was first observed by Alfano
and Shapiro [33] in various glasses, and output spectra spanning the visible spectrum were
recorded. The spectral broadening was attributed to self-phase modulation (SPM) but, since
it was ten times stronger than anything previously observed, it became known as a supercontinuum [5253]. In bulk media, the threshold for SCG coincides with the threshold
for self-focusing, and the increase in intensity that accompanies beam collapse promotes a
family of third-order processes including self-phase modulation, cross-phase modulation,
self-steepening, and plasma formation. The mechanism of self-focusing is complicated
enough in itself but, despite this, a detailed model of SCG in bulk materials was presented
by Gaeta in 2000 [54].
Understanding SCG in optical fibres is significantly simpler because the complexities
of self-focusing are absent. Even there, however, a whole family of third-order processes
conspires together to deliver the strong spectral broadening, including stimulated Raman
scattering and associated four-wave mixing processes, as well as self- and cross-phase
modulation [55]. For a recent book on supercontinuum generation in optical fibres, see [56].
When driven by a mode-locked pulse train, SCG is the basis for the generation of wideband frequency combs that have applications in optical frequency metrology. If combs
at the fundamental and second harmonic frequencies are present at the same time, and
both span a substantial fraction of an octave, the two will overlap, and this is the basis
of an important technique in laser technology for the phase stabilisation of femtosecond
pulses; this is explained in more detail in Section 7.10. More generally, SCG can be used
as a broadband light source in a wide variety of contexts including spectroscopy, optical
coherence tomography, and telecommunications.
Problems
5.1 Show that, with respect to the axes defined in Fig. 5.1, the effective third harmonic
generation coefficient for a plane-polarised fundamental field is eff = 12 (1 + 2 +
3 + 4 ) (Eq. 5.8).
5.2 Show that third harmonic radiation cannot be generated using circularly polarised light.
5.3 Find the half-wave voltage for a Kerr cell based on nitrobenzene (K = 4.4
1012 m V2 ). The cell is 10 cm long, and the plate separation is 1 cm.
28 I am indebted to Dr Louise Hirst for providing me with some useful background material on supercontinuum
generation
115
5.7 Supercontinuum generation
5.4 Verify Eq. (5.18) starting from the relevant equations in Appendix C.
5.5 Show that the phase velocity of an N = 1 soliton is approximately (cn2 Ipk /2n20 ) lower
than its low-intensity value in the same medium.
5.6 Show that Eq. (5.35) follows from Eq. (5.34), and that Eqs (5.32) and (5.35) together
ensure that the photon flux (photons/area/time) is conserved.
5.7 Prove Eq. (5.62) for the coherence length of the acousto-optic process diagrammed in
Fig. 5.8.
= 0.
5.8 Equations (5.90) give the solutions to Eqs (5.87) for the boundary condition A(0)
Show that for the alternative boundary condition E S (L) = 0, the analogous solutions are
cos (L z) ; E S (z) = A (0) sin (L z) .

A(z)
= A(0)
cos L
r
cos L
5.9 Show that when the Gaussian beam of Eq. (5.95) encounters the ordinary mirror at
z = zm in Fig. 5.15, the reflected wave corresponds to a backward-travelling Gaussian
beam centred at z = 2zm , as indicated in the figure.
Dispersion and optical pulses

Just as Chapter 3 on crystal optics served as a prelude to Chapter 4, so the present chapter
lays the foundations for Chapter 7 on nonlinear optics in ultrashort pulse technology. The
central topic is the phenomenon of dispersion. Although itself a linear effect, dispersion
plays a vital role in many nonlinear interactions, and in many techniques used for short pulse
generation too. We have already seen how dispersion in the phase velocity leads to phase
mismatching in frequency-mixing processes and how, in combination with birefringence,
it can also become part of the solution. However, whenever optical pulses are involved,
dispersion in the group velocities invariably becomes important, a feature that will crop up
repeatedly in the following chapter.
In Section 6.2, we consider the relationship between phase velocity dispersion (PVD) and
group velocity dispersion (GVD). The ideas are then applied and elaborated in Section 6.3
where we demonstrate how, in the presence of GVD, a Gaussian pulse broadens and
develops a frequency sweep. Hyperbolic secant pulses are contrasted with Gaussians in
Section 6.4, and a discussion of chirped pulse compression using gratings and prisms follows in Section 6.5. A partial differential equation governing pulse propagation in the
presence of GVD is developed in Section 6.6 for applications in Chapter 7.
6.2 Dispersion in the phase and group velocities

In Section 1.2, we showed
that the linear susceptibility is related to refractive index through
the equation n() = 1 + (1) (). But the susceptibility, and hence also the refractive
index, are frequency dependent (as indicated), and this underlies the phenomenon of dispersion, a word that means distribution in different directions or spreading out. Dispersion
describes the separation of white light into colours by a prism, and equally the temporal
spreading of an optical pulse as it propagates through a dispersive material. These and
similar phenomena arise from the effect of the frequency dependence of n on two key
characteristics of light propagation: the phase velocity and the group velocity
For a simple travelling wave of the form cos{t kz}, the phase velocity is
vphase =
116
=
k
n()
(6.1)
117
where k = n()/c. On the other hand, the group velocity is given by

vgroup =
d
.
dk
(6.2)
For an optical pulse, one can say in general terms that vphase is the speed of the highfrequency optical oscillations (the optical carrier wave), while vgroup is the speed of the
pulse envelope. So one might expect a propagating pulse to have a profile of the form
E(z, t) A(t z/vgroup ) cos{(t z/vphase )}. The analytic result derived in Eq. (6.14)
below broadly confirms this conclusion, but also shows that the envelope broadens, and the
pulse develops a frequency sweep, known as a chirp.
The phase and group velocities can be conveniently represented on the plot of against
k, shown schematically in Fig. 6.1. This is called a dispersion diagram because it embodies
the frequency dependence of the refractive index, which is the origin of dispersion in both
vphase and vgroup . It is clear from Eqs (6.1) and (6.2) that the gradient of OA represents the
phase velocity, while the gradient of the tangent TT to the dispersion curve at A represents
the group velocity. In the case shown, the group velocity is lower than the phase velocity.
If one could watch the propagation of an optical pulse under these conditions, the optical
carrier wave oscillations would be seen to be travelling faster than the envelope, growing
as they approached the peak from the rear, and dying away as they departed towards the
front. There are points on the dispersion curve where the situation is reversed, and the group
velocity is higher than the phase velocity.
The excursions of the dispersion curve in Fig. 6.1 are in fact grossly exaggerated. Graphs
of the phase and group velocities in fused silica, shown dotted in Fig. 6.2, indicate that
the difference between the two (and their individual variation over the wavelength range
shown) is barely more than 1%. The eye would therefore find it hard to distinguish a scale
drawing of a typical dispersion curve from a straight line through the origin.
T
A
Fig. 6.1
Schematic dispersion diagram. The excursions of the dispersion curve are exaggerated.
118
0.8
wavelength (microns)
1.2
1.6
1.96
GTDD (ps/nm/km)
200
phase
400
1.94
1.92
group
600
1.90
phase/group velocity (108 m/s)
800
Fig. 6.2
Group time delay dispersion characteristic of fused silica (solid line). The phase and group velocities (plotted against
the right-hand axis) are shown dotted.
It is evident from Eq. (6.1) that the frequency dependence of the refractive index leads
directly to variation in the phase velocity. This is known as phase velocity dispersion
(PVD) or sometimes chromatic dispersion. In regions of the spectrum when n() rises
with frequency (dn()/d > 0), the dispersion is said to be normal, because this is the
case for most transparent optical materials in the visible part of the spectrum. The opposite
case (dn()/d < 0) is referred to as anomalous dispersion.1
The characteristics of group velocity dispersion (GVD) are conveniently expressed in
terms of the group time delay (GTD), defined as the time it takes a wave group to propagate
a distance z, namely
group =
z
vgroup
=z

dk
z
dn
z
dn
=
n+
=
n
.
d
c
d
c
d
(6.3)
The final step in the equation expresses the result in terms of the vacuum wavelength. In
the context of optical pulse propagation, one is interested in the variation of the group
time delay with frequency or wavelength, a property that is known as the group time delay
dispersion (GTDD), and is equivalent to GVD in the information it provides. The key
formulae, obtained by differentiating Eq. (6.3), are

dgroup
z
d 2k
dn
d 2n
=z 2 =
2
+ 2
d
d
c
d
d
(6.4a)
dgroup
z d 2 n
.
=
d
c d2
(6.4b)
and
1 These definitions are frequently quoted in terms of the (vacuum) wavelength rather than the frequency. The
index falls with increasing wavelength in the case of normal dispersion.
119
Table 6.1 Signs of parameters controlling PVD and GVD.

dn
d
dvphase
d
d 2n
d2
dvgroup
, 2
d
D ,
dgroup
d
Chirp
Up/+
Down/
Phase velocity (chromatic) dispersion

Normal
+
Anomalous
+
Group velocity dispersion

Positive
Negative
Notice the simplicity of the group time delay dispersion (GTDD) when expressed in terms
of the vacuum wavelength. The GTDD is commonly characterised (especially in optical
fibres) by the dispersion parameter
D =
d 2n
.
c d2
(6.5)
Convenient units for D are picoseconds (of pulse spread) per nanometre (of bandwidth)
per kilometre (of fibre).
The wavelength dependence of D in fused silica is plotted as a solid line in Fig. 6.2.
Below 1.3 m, d 2 n/d2 is positive, so D and dgroup /d are both negative. The
group time delay falls with increasing wavelength, and the group velocity correspondingly rises, as shown in the figure. Close to 1.3 m, d 2 n/d2 passes through zero
and, at higher wavelengths, d 2 n/d2 goes negative, while D and dgroup /d become
positive.
The GVD is said to be positive or negative according to the sign of d 2 n/d2 . Occasionally people speak of normal and anomalous GVD, but this usage is best avoided
because it encourages the erroneous belief that dispersion in the group velocity neatly parallels dispersion in the phase velocity. But Fig. 6.2 makes it quite clear that there is no
necessary correspondence between the two kinds of dispersion. The phase velocity of fused
silica rises with wavelength (and the refractive index therefore falls) throughout the range
displayed, and so the PVD is normal everywhere. However, the group velocity rises on
the left-hand side of the figure (positive GVD), reaches a maximum near 1.3 m, and falls
away thereafter (negative GVD).
The signs of the various dispersion parameters are summarised in Table 6.1, which
highlights the point that the signs of PVD and GVD are determined by different
characteristics.
The fact that the GVD goes to zero near 1.3 m makes this an attractive wavelength
for high bit-rate optical communications in optical fibres. The zero dispersion wavelength
can also be shifted to some extent by adjusting the fibre geometry and, in practice, it is
normally moved to around 1.55 m where InGaAsP lasers operate and the transmission of
fused silica is maximised.
120
6.3 Dispersion and chirping of a Gaussian pulse

6.3.1 Propagation analysis
Group velocity dispersion (the fact that different frequency groups travel at different speeds)
has two important consequences: the pulse envelope broadens, and the carrier wave within
the envelope acquires a frequency sweep. For positive GVD, low frequencies travel faster
than high frequencies, and so arrive earlier. The carrier frequency duly rises through the
pulse, which is said to have acquired a positive chirp (or an up-chirp). In the opposite
case, the frequency sweep is negative, and the pulse exhibits a down-chirp.
We will now perform a simple exercise on a Gaussian pulse to see how it changes during
propagation in a dispersive medium.2 We use the general definition of the real field
t) exp{i0 t} + c.c.)
E(z, t) = 12 (E(z,
(6.6)
where 0 is the angular frequency of the optical carrier wave before dispersion is introduced.
t) is generally complex, but we will assume that initially (at z = 0)
The field envelope E(z,
it has the real Gaussian form
t) = A exp{t 2 /2t02 }.
E(0,
(6.7)
The full width

at half maximum intensity (FWHM) of the intensity profile is easily shown
to be t0 = t0 4 ln 2 1.665t0 .
To discover how the pulse changes as it propagates, we take the Fourier transform of the
t) exp(i0 t), namely3
analytic signal V (0, t) = E(0,
1
V (0, ) =
2
+
*
At0
t) exp{i( 0 )t}dt =
E(0,
exp 12 ( 0 )2 t02
2
(6.8)
which is a Gaussian centred on the carrier frequency 0 .

The evolution of the pulse as it propagates through the dispersive medium is governed
t); see Eq. (2.4). The effect of dispersion can
by a factor of the form exp{ikz} within E(z,
therefore be found by applying a phase change of () = k()z to V (0, ) namely
V (z, ) = V (0, ) exp{ik()z}.
We expand the phase () as a Taylor series about 0 namely

2
d
1 d
() k()z = 0 +
( 0 ) + 2
( 0 )2 +
d 0
d2 0
(6.9)
(6.10)
where 0 = k(0 )z, and it follows from Eqs (6.3) and (6.4) that the derivatives are

dgroup
d
d 2
=
;
=
.
(6.11)
group

2
d 0
d 0
d
2 Readers who are not interested in the details of the following proof can skip to Eq. (6.14).
3 See Appendix E for a discussion of the analytic signal.
121
It will also be convenient for later use to write the analogous expansion for k() in the form
k() = k(0 ) + 1 ( 0 ) + 12 2 ( 0 )2 +
(6.12)
where the expansion coefficients, which will feature in several later formulae, are

group
dk
d 2 k
1 dgroup
D 2
1
1
=
v
=
=
;
=
.
(6.13)
2
group

2
d 0
z
d 0
z d
2 c
Substituting Eq. (6.10) into Eq.(6.9), and transforming back to the time domain yields
t) exp(i0 t} =
E(z,
V (z, ) exp{it}d

2
2
tgroup
i tgroup
A
=
exp{i0 tphase } exp 2
exp

1 + i
2t0 (1 + 2 )
2t02 (1 + 2 )

term 2
term 1
term3
term 4
(6.14)
where tphase = t z/vphase , tgroup = t group , and is a dimensionless distance given by

2 z
D 2
= 2 =
z.
(6.15)
t0
2 ct02
Of the four terms on the right-hand side of Eq. (6.14), two (terms 1 and 3) affect the pulse
envelope, and two (terms 2 and 4) affect the phase of the carrier. The result has five key
features:
the optical carrier wave (term 2) moves at the phase velocity since tphase = t z/vphase ;
the envelope (term 3) moves at the group velocity since tgroup = t z/vgroup ;
2;
the pulse width (term 3) is increased by the factor 1 +
the pulse peak intensity (term 1) is reduced by the factor 1 + 2 , which follows when
one works out the modulus squared;

the pulse acquires a frequency chirp (term 4), which means that the optical carrier
frequency changes through the pulse.
These issues are reviewed in the following subsections.
6.3.2 Characteristic dispersion distance

The first question to ask is: how far must the pulse travel through the dispersive medium
before | | = 1? This characteristic
dispersion distance Ldisp marks the pointat which
the peak intensity has fallen by 2 and the pulse duration has increased by 2. From
= (2 /t02 )z, it follows that
Ldisp =
t02
t02
=
.
|2 |
4 log 2 |2 |
(6.16)
122
We conclude that the larger |2 | the smaller Ldisp (as expected), and that Ldisp increases
as the square of the initial pulse duration. The square
arises because a pulse of twice the
duration needs to travel twice as far to be broadened by 2, but also has half the bandwidth
to be dispersed.
6.3.3 The bandwidth theorem and the timebandwidth product

Next, we consider the relationship between the pulse duration and the spectral bandwidth.
From Eq. (6.8), the FWHM of the spectral intensity is easily shown to be
=
0.441
1 2 log 2
.
=
t0
t0
(6.17)
Moreover, we have seen from Eq. (6.14) that dispersion causes the pulse duration to increase
according to

(6.18)
t = t0 1 + 2 .
Hence the timebandwidth product is

t = 0.441 1 + 2 .
(6.19)
This is a special case of the bandwidth theorem, which states that

t K
(6.20)
where K, a number of order unity, is 0.441 for Gaussian profiles. When the minimum
timebandwidth product is realised, which occurs at = 0 in the present example, a pulse
is said to be bandwidth-limited or transform-limited.
6.3.4 Chirp
The magnitude of the frequency chirp can be determined by writing term 4 of Eq. (6.14)
as exp{i}, and remembering that d/dt represents the shift in the carrier frequency. The
chirp (in rad s2 ) is therefore
d
d 2
= 2
=
.
2
dt
dt
t0 (1 + 2 )
(6.21)
For long propagation distances (| | 1), this simplifies to

2 c
1
d 1
= 2
;
= 2 =
dt
2 z
D z
t0
c
d 1
= 2
=
dt
22 z
D z
(| | 1)
(6.22)
where the result has been quoted in both rad s2 and in Hz s1 .

It is worth trying to understand the physics behind these formulae. First of all,
Eqs (6.21) and (6.22) indicate that, if 2 > 0, the chirp is positive. Positive 2 means
that the GTD increases with frequency, so higher frequencies suffer a greater delay than
lower frequencies, which is consistent with an increase in carrier frequency through
the pulse.
123
0.5
relative frequency shift
field (arb. units)
5.0
10
10
local time
Fig. 6.3
Prole of a chirped Gaussian pulse with = +3. The parent pulse is shown dotted. The slanting dotted line is the
relative frequency shift (plotted against the right-hand axis), and conrms that the chirp is linear.
Secondly, Eq. (6.21) indicates that, as the pulse propagates through the dispersive
medium, the chirp increases until | | = 1, and decreases thereafter. The initial increase
occurs as the pulse spectrum begins to be spread out, but the chirp ultimately falls as the
fixed bandwidth is distributed over an increasingly long time.
Let us follow this line of thought a little further. For | | 1, the pulse duration is
t = | | t0 from Eq. (6.18), so dividing the bandwidth from Eq. (6.17) by the pulse
duration yields

1 2 log 2
c
=
= 2
t
|D | z
| | t02
(6.23)
with the help of Eq. (6.15). This is very reassuring, because it is in perfect agreement with
Eq. (6.22), apart from the sign which has not been considered.
An example of a pulse chirped by dispersion is shown in Fig. 6.3. In order to make the
chirping clearly visible, a pulse only two optical cycles in duration has been chosen as the
parent (shown dotted). The value of is +3, and careful inspection of the figure shows that
the pulse is indeed up-chirped; the carrier frequency is clearly higher for tg > 0 than for
tg < 0.
The slanting dotted line (plotted against the right-hand axis in the figure) records the
frequency shift relative to that of the optical carrier. This ranges from around 0.3 on
the leading edge to around +0.3 on the trailing edge which, in real world terms, are high
numbers. However, had more realistic values been used, the chirp would have been hard to
see by eye.
Finally, notice that dispersion has reduced the amplitude of the pulse because of term 1 in
Eq. (6.14). If a quick glance at the figure suggests that energy is not conserved, remember
that the field rather than the intensity is plotted in the figure.
124
6.4 Hyperbolic secant proles

Gaussian profiles possess some convenient mathematical properties; for example, no other
bell-shaped function would have allowed the calculation leading to Eq. (6.14) to have been
performed so simply. However, hyperbolic secant profiles arise in the analysis of optical
solitons, and indeed most real optical signals are probably closer to sech than to Gaussian
pulses. It is therefore worth considering the basic properties of sech profiles.
The analogue of Eq. (6.7) for a hyperbolic secant profile is
E0 (0, t) = A sech{t/t0 }.
(6.24)
2
Remember that if the field profile is a sech, the intensity varies
as sech , the FWHM of the
intensity profile is then easily shown to be t = t0 ln{3 + 8} 1.763 t0 .
Just as the Fourier transform of a Gaussian is another Gaussian, so the Fourier transform
of a sech is another sech, namely

,
V (0, 0 ) = At0 /2 sech 12 ( 0 )t0 .
(6.25)

2
The timebandwidth product comes out to be t = 2 ln{3 + 8} 0.315,
slightly less than the figure of 0.441 for Gaussians (see Eq. 6.20).
The key difference between Gaussian and sech (or sech2 ) profiles lies in the behaviour
in the wings. This is best appreciated by viewing the profiles on a log scale as in Fig. 6.4.
Whereas the sech2 function (curve S) has exponential wings, which appear as straight
lines on the log scale, the wings of the Gaussian (curve G) plunge increasingly steeply
1x100
log10(intensity)
1x102
S
G
1x104
1x106
1x108
1x1010
4
Fig. 6.4
0
time
Comparison of Gaussian and sech2 proles of the same intensity FWHM plotted on a logarithmic intensity scale. In
reality, the wings are always embedded in a noise background (shown in grey).
6.5 Compression using gratings and prisms
125
downwards. Naturally, in any experimental situation, the wings ultimately end in the noise
background, as indicated in grey.4

The pulse broadening and chirping that result from GVD could easily be reversed if a material with the opposite GVD characteristics could be found. However, materials exhibiting
negative GVD are needed to reverse the effects of positive GVD, and these are virtually
unobtainable. Fortunately, the same effect can be realised by using arrangements of diffraction gratings or prisms; indeed the grating arrangement shown in Fig. 6.5, first proposed by
Treacy in 1969 [57], has been widely used ever since for compressing up-chirped pulses.
A simple geometrical argument shows that the optical path of a plane wave negotiating
the Z-shaped path between the gratings is shorter for short wavelengths and longer for long.
As the wavelength increases, the diffraction angle in Fig. 6.5 increases too; points B and
C will lie higher in the diagram (see arrows), and the path length ABC will increase. Hence,
the lower wavelengths (higher frequencies) at the rear of an up-chirped pulse will take a
shorter path, and hence will catch up the lower frequencies at the front. In terms of the
frequency-dependent phase, we can write
() = 12 (2 z)( 0 )2 + 12 (2 p)( 0 )2

GVD
grating pair
G2
+
C
G1
A
Fig. 6.5
Geometry of a grating pair.

4 The fluctuating noise background shown in Fig. 6.4 is numerically generated random noise.
(6.26)
126
Fig. 6.6
Four-prism compressor.
where the first term comes from Eqs (6.10) and (6.12), and p is the perpendicular distance
between the gratings. If 2 p is positive, the second term counters the quadratic term in 2 z
in the case of normal GVD (see Section 6.3).
To find 2 in Eq. (6.26), it is necessary to calculate the group time delay group of different
frequency groups travelling from the point of incidence A on grating G1 in Fig. 6.5 to B on
the second grating G2, and thence to C. The result (see Appendix F for details) is
dgroup
4 2 cp
= 2 p = 2 3
d
I cos3
(6.27)
where I is the ruling spacing of the gratings, and is the mean value of angle in Fig. 6.5.
Notice that 2 p is necessarily positive, so it is correctly signed to offset positive 2 z in
Eq. (6.26). On the other hand, dgroup /d is negative, which means that higher frequencies
overtake lower frequencies. These are the characteristic features of negative GVD.
Grating pairs are often used in a double-pass configuration. The output is either reflected
back through the system by a plane mirror, or two grating pairs are used in series. Apart
from doubling the dispersion, both schemes have the advantage that different frequency
components are not spatially separated in the output beam.
Gratings can deliver substantial amounts of negative GVD, but they introduce significant
loss at the same time. An alternative technique based on chromatic dispersion in sets of
prisms offers much higher transmission, and is therefore suitable for dispersion control
inside laser cavities.5 Prisms are also less expensive that gratings, and prism systems can
deliver dispersion of either sign too. On the other hand, the amount of GVD they can
introduce is smaller than for gratings, basically because chromatic dispersion in glass is
much weaker than diffractive dispersion from gratings. The four-prism configuration shown
in Fig. 6.6 is a popular arrangement; as for gratings, placing two prism pairs in series ensures
that different frequency components are not spatially separated in the output beam.
The properties of a single prism pair can be understood from the geometry of Fig. 6.7
in which a typical ray follows the path shown from A to B. As pointed out by Martinez
et al. [58], the frequency-dependent phase introduced by refraction through the prisms can
be measured between any pair of points on the input and output wavefronts, so the line
between the prism vertices at V and W can be used as a convenient reference in this case.6
5 Another option for intra-cavity dispersion control is to use chirped dielectric mirrors.
6 This technique could equally have been used to analyse the grating pair (see Problem 6.6).
127
Fig. 6.7
Geometry of a prism pair.

M
G1
Fig. 6.8
G2
Schematic grating stretcher.

If the distance VW is given by s, and the angle between VW and the ray path is , we have
() =
and hence
dgroup
s cos
d 2
= 2 =
d
d
c
s cos ()
c
d
d
2
s sin
(6.28)

d
d 2
2
+ 2 .
d
d
(6.29)
The key conclusion is that dgroup /d will be negative if the first term is dominant,
which will be the case for small . Under these conditions, the prism pair, like the grating
pair, exhibits the characteristic of negative dispersion. However, an arbitrary amount of
positive dispersion can easily be included by increasing the distance a pulse travels in the
body of one or both of the prisms. Moreover, if one of the prisms is translated along a line
perpendicular to its base, as suggested by the arrows on the third prism in Fig. 6.6, the
amount of positive GVD can be controlled.
Finally, we note that placing a telescope between a pair of gratings, as shown in Fig. 6.8,
reverses the sign of the dispersion, and this kind of arrangement is routinely used to introduce
positive GVD for stretching pulses in chirped-pulse amplification systems. Mirrors are often
used instead of the lenses shown in the figure; indeed placing a single spherical mirror of the
correct strength between a pair of gratings in a reflective double-pass arrangement offers
an elegant solution.
128
6.6 Optical pulse propagation

6.6.1 Pulse propagation equation
The simple differential equations developed in Chapter 2 (see e.g. Eq. 2.8) were suitable
for tracing the growth and decay of signals that are essentially monochromatic. To study
the propagation and nonlinear interaction of optical pulses, more sophisticated differential
equations are required. For a detailed derivation, one needs to go back to Maxwells wave
equation, but it is possible to advance a reasonable argument in favour of the correct result
by developing some of the formulae obtained in this chapter in a simple hand-waving way.
For an incremental propagation distance z, Eq. (6.9) can be written in the differential
form V () = ik()V ()z. With the help of Eq. (6.12), this result can be used to
construct the differential equation
d V (z, )
= i(0 + 1 ( 0 ) + 12 2 ( 0 )2 + )V (z, ).
dz
(6.30)
This equation can now be Fourier transformed, term by term, into the time-domain equation

t)
E(z,
2
t)
(6.31)
= i0 1 + 21 i2 2 + E(z,
z
t
t
where standard formulae for transforming differential operators have been used. Since
1 , Eq. (6.31) can be rewritten in the form
1 = vgroup

2
1
1
t)
(6.32)
+
E(z, t) = i 0 + 2 2 2 + E(z,
z vgroup t
t
which contains all the ingredients needed to produce the features seen in Eq. (6.14). The
directional derivative on the left-hand side represents the rate of change of the field with
distance as seen by an observer travelling with the pulse, and ensures that the pulse envelope
travels at the group velocity. The first term on the right-hand side ensures that the carrier
wave travels at the phase velocity, while the term in 2 underlies the development of chirp
and the change in the pulse shape.
6.6.2 Transformation to local time

A number of refinements to Eq. (6.32) can now be introduced. First of all, the phase velocity
t) = E(z,
t) exp{ik0 z} where k0 = 0 . The reduces
term can be removed by writing E(z,
the equation to

2
1
t) = 1 i2 E(z, t) .
+
E(z,
(6.33)
2
z vgroup t
t 2
A further very common procedure is to transform to the so-called local time frame, where
local time is measured relative to the time of arrival of a particular point on the pulse. This
removes the motion of the pulse from the equations, which has the advantage that changes
6.6 Optical pulse propagation
129
to the pulse profile are highlighted, and that a steady-state pulse is represented by a function
with zero derivatives in the direction of propagation.
The local time coordinates (Z, T ) are duly defined by Z = z, T tgroup = t z/vgroup .
By using the chain rule, it is easy to show that
z vgroup t
Z
and
,
t
T
and Eq. (6.33) therefore becomes

E (Z, T )
2 E (Z, T )
= 12 i2
Z
T 2
(6.34)
t) by
where the field E (Z, T ) in transformed coordinates is related to E(z,
t)=E (Z, T ) E (z, t z/vgroup ).
E(z,
(6.35)
Strictly speaking, E and E are different mathematical functions, a point that is often glossed
over. The distinction can be understood by considering the trivial case of a pulse travelling
at the group velocity without change of shape, i.e. with 2 = 0. A schematic representation
t) in this case is shown in the lower part of Fig. 6.9, where the slanting lines are
of E(z,
t), and thicker lines represent higher field values. Notice how the peak of
contours of E(z,
Z
T
0
z/vgroup
Fig. 6.9
Pulse propagation in (top) the local time frame (Z-T), and (bottom) in the normal z-t frame.
130
the pulse moves from t = 0 to t = z/vgroup in propagating a distance z. This should be

compared with the upper part of the figure, which shows the corresponding representation
of E (Z, T ). The contours now run parallel to the Z-axis, at fixed values of local time T . The
difference between the two parts of Fig. 6.9 is precisely the difference between E and E .
For future purposes, we will set these niceties aside, and write Eq. (6.34) in the form
E
2 E
= 12 i2 2
z
T
(6.36)
where the distinction between E and E has been ignored, and Z has been replaced by z on
the left-hand side on the understanding that the derivative is to be performed at constant T ,
not at constant t. The parameter T has been retained on the right-hand side as a reminder that
this is a local time-frame equation. Equation (6.36) will be used to construct the differential
equation governing short pulse propagation at the beginning of the next chapter.
Problems
6.1 Prove the final step of Eq. (6.3).
6.2 It appears from Fig. 6.2 that the zero-dispersion wavelength (where the group time
delay dispersion changes sign) might perhaps correspond to a point of inflexion in the
phase velocity as a function of wavelength.
Is this suggestion correct or incorrect?
6.3 (a) Prove the formula t0 = t0 4 ln 2 1.665t0 for the intensity FWHM of the
Gaussian pulse defined in Eq. (6.7).
(b) A transform-limited Gaussian pulse at 1 m has an intensity FWHM of 1 ps. Find
its bandwidth in both THz and nm.
(c) What is the maximum chirp (in GHz/ps) that can be realised by dispersing the pulse
of part (b)?
6.4 Prove the result for the intensity FWHM of a sech2 pulse quoted after Eq. (6.24).
6.5 Find the characteristic dispersion distance Ldisp for a 100 fs Gaussian pulse at 1 m in
a medium with a dispersion parameter of 20 ps nm1 km1 .
6.6 Apply the technique used to find the dispersion characteristics of the prism pair in
Eqs (6.28) and (6.29) to derive Eq. (6.27) for the grating pair.
Nonlinear optics with pulses

While everything in Chapter 6 came under the umbrella of linear optics, in this chapter
we will be looking at a range of nonlinear phenomena associated with optical pulses. We
will consider the process of self-phase modulation in considerable detail, and see how it
can be used, in combination with dispersion, to stretch and compress optical pulses, and to
generate optical solitons. We will examine the adverse effects of group velocity dispersion
on second harmonic generation, and examine various ways of optimising the bandwidth
in optical parametric chirped pulse amplification. Finally, we will discover how nonlinear
optical techniques can be used in the diagnosis of ultrashort pulses and for the stabilisation
of the carrier-envelope phase.
7.2 Wave equation for short pulses

A detailed derivation of the differential equation governing the propagation of short optical
pulses under nonlinear conditions is long and intricate; see for example [59]. To avoid this,
we will take a series of reasonable steps that lead to the correct conclusion.
We start, as in Section 2.2, by substituting Eqs (2.4) and (2.5) into Eq. (2.3), but we now
retain three time-dependent terms that were previously discarded. The result is

L + P NL )
E
(
P
2i E
2
2
2 L
NL
2ik
+k E+ 2
2 E = 0 (P + P ) 2i
. (7.1)
dz
c dt
c
dt
We will mainly be dealing with signals close to a single frequency , so only one frequency
component has been included, and the subscripts on E andP NL have therefore been dropped
We now set P L = 0 (1) E and k = n0 /c, where n0 = 1 + (1) is the refractive index.1
Equation (7.1) then becomes

E
2i
1 E
1
+
= i
P NL
dz
vphase dt
2c0 n0
dt
(7.2)
1 The symbol n is used for the low-intensity value of the refractive index in this chapter. The symbol n will be
0
reserved for the value when intensity dependence effects are taken into account; see Section 7.3.
131
132
where vphase = c/n0 . When the complex envelopes have no time variation, this equation
reduces to Eq. (2.8).
The most obvious deficiency of Eq. (7.2) is that it takes no account of dispersion. This
shortcoming can be remedied by switching to the frequency domain and expanding the
angular wave number as a Taylor series in , as was done in Sections 6.3 and 6.6. Equation
(7.2) then becomes

E
1 E
2 E
2i
+
= 12 i2 2 i
1
(7.3)
P NL
dz
vgroup dt
t
2c0 n0
dt
where the dispersion parameter is given by
2 =
1 dgroup
z d
as explained in Chapter 6.
When Eq. (7.3) is transformed to the local time frame Z = z, T = t z/vgroup (see
Section 6.6.2), it reduces to2

t)
E
2 E(z,
i
1
1
= 2 i2
i
P NL .
(7.4)
dz
T 2
2c0 n 0
dT
In writing this equation, some of the finer details discussed at the end of the last chapter
have been glossed over. Although Z has been replaced by z on the left-hand side, it must be
remembered that the derivative is to be performed at constant T , not at constant t. Notice that
the factor 2 in the final operator of Eq. (7.3) has disappeared in the local time conversion. In
fact, it is frequently a good approximation to ignore this operator altogether and, for much
of this chapter, we will abbreviate Eq. (7.4) to
2 E
E
P NL .
= 12 i2 2 i
z
T
2c0 n0
(7.5)
However, the term that has been omitted plays an essential role in a number of important
nonlinear processes, especially when pulses containing very few optical cycles are involved.
We shall return to this issue at several points later in this chapter.
7.3 Self-phase modulation

We first encountered the process of intensity-dependent refractive index (IDRI) in Chapter 1
and then (in more detail) in Chapter 5. We now consider the effect of IDRI on the propagation
of an optical pulse. Written in terms of slowly varying envelopes, and with the coordinate
subscripts dropped, Eq. (5.16) reads
2

P NL = 34 0 1IDRI E E.
(7.6)
2 The step from Eq. (7.3) to Eq. (7.4) is not straightforward; see [59] for a detailed treatment.
133
7.3 Self-phase modulation
As shown in Chapter 5, this implies that the refractive index acquires an intensity-dependent
term given by

31IDRI
n = n0 +
I = n0 + n2 I
(7.7)
4n20 c0
where n0 is the refractive index at low intensities, and the second step defines the nonlinear
index n2 .3 In fused silica n2 31016 cm2 W1 ; in most other materials, the value is one
or two orders of magnitude larger. The sign of n2 is almost always positive, which means
that refractive index normally increases with intensity.
The most obvious consequence of Eq. (7.7) for an optical pulse is that the phase velocity
(= c/n) varies with intensity, and this causes the phase of the optical carrier wave to become
modulated in the process known as self-phase modulation (SPM). Specifically, since the
angular wave number is k = n/c, and an optical signal acquires a phase of ikz as it
propagates, a pulse will develop a phase modulation given by = n2 I (t)z/c. The same
conclusion follows from Eq. (7.5). If we ignore the GVD term for the moment, and insert
Eq. (7.6) for the nonlinear polarisation, the equation reads

E
n2 I
= i
E.
(7.8)
z
c
The solution is clearly E(z)

= E(0)
exp {i(z)}where
(z, t) = n2 I (t)z/c.
(7.9)
Notice that these equations imply that the pulse envelope will not be affected by the nonlinearity, and this is certainly borne out in practice when the IDRI is weak and/or the distance
of propagation is small. Under other conditions, the more sophisticated Eq. (7.4) has to be
used as the governing equation, and one such case is dealt with in Section 7.5 below.
Equation (7.9) shows that the phase change at the peak of the pulse rises to 1 radian
after a propagation distance of c/n2 Ipk , and this provides a convenient definition of the
characteristic distance for SPM, namely
c
Lspm =
(7.10)
|n2 | Ipk
which is identical to Eq. (5.22). After travelling a distance of (say) 4 Lspm , the phase
change at the peak will therefore be 4, and Fig. 7.1 shows what the field profile looks like
at this point. Although superficially similar to Fig. 6.3 (the case of GVD), the structure of the
frequency modulation is more complicated in the present figure. For SPM, the frequency
shift (plotted as a dotted line in the figure) is
n2 z I
=
t
c t
from Eq. (7.9), while the rate of change of frequency (the chirp) is
d
d 2
n2 z 2 I
= 2 =
.
dt
dt
c t 2
3 The definition of the nonlinear refractive index is sometimes based on the equation n = n + n |E|2 .
0
2
134
0.5
field (arb. units)
0.5
10
0
time (carrier cycles)
10
Prole of a pulse chirped by self-phase modulation. The dotted line traces the relative frequency shift. Compare
Fig. 6.3.
0.5
frequency shift & chirp
phase (units of )
Fig. 7.1
relative frequency shift
0.5
10
10
time (carrier cycles)
Fig. 7.2
Characteristics of Fig. 7.1: (solid line) time-dependent phase; (dotted line) relative frequency shift (= rst phase
derivative); (grey line) chirp (= second phase derivative).
Whereas in the dispersive case (shown in Fig. 6.3), the chirp is constant and positive across
the entire profile, now it is negative on both wings, but positive in around the peak of the
profile.
More detailed information is contained in Fig. 7.2, which shows the phase (t), and
curves based on its first and second time derivatives. From Eq. (7.9), (t) I (t),
and this is the solid line in Fig. 7.2. The dotted line is the relative frequency shift from
Fig. 7.1, which varies as d/dt, while the solid grey line is the chirp. Notice that the
7.4 Self-phase modulation with dispersion: optical solitons
Intensity
135
10
0
frequency
10
Intensity
10
10
Fig. 7.3
Spectra of pulses chirped by SPM: peak phase shift (top), 4.5(bottom).

region of positive chirp lies between the two points of inflexion of I (t), which are close to
the half-maximum intensity points.
Figure 7.3 shows spectra of self-phase modulated Gaussian pulses for peak phase shifts
of and 4.5 (top and bottom, respectively). The frequency scale is normalised to the
intensity FWHM of the parent pulse spectrum. In general, the spectra consist of a series
of lobes, the number corresponding roughly to the peak phase shift in units of [60, 61].
We will discuss how the increased bandwidth can be exploited to create shorter pulses in
Section 7.6 below.

Dispersion has been ignored in Section 7.3, so we must now consider what happens to a
pulse when both SPM and GVD occur together. We duly restore the dispersion term to its
rightful place in Eq. (7.8), which now reads
E
2 E
= 12 i2 2 i
z
T
n2 I
c
E.
(7.11)
136
We now use the dispersion distance Ldisp from Eq. (6.16), and the pulse width measure
t0 from Eq. (6.24) to define dimensionless distance and time coordinates = z/Ldisp =
E pk , Eq. (7.11) then becomes
|2 | z/t02 and = T /t0 . In terms of the relative field U = E/
U
=i

Ldisp 2
sgn{2 } 2
U .
U

2
2
Lspm
(7.12)
This equation has very different properties depending on the sign of the GVD parameter
2 . As we have seen earlier, both GVD and SPM cause pulse chirping. Since the nonlinear
index n2 is almost invariably positive, SPM generally creates an up-chirp. Positive GVD
(2 > 0; see Table 6.1) creates an up-chirp too so, in this case, GVD and SPM reinforce
each other; see Section 7.6 below. On the other hand, when 2 < 0, the two processes work
in opposition, and Eq. (7.12) then reads
U
i
=
1
2
2
2
2
+ N U U
2
where
!
N=
Ldisp
=
Lspm
(2 < 0)
n2 Ipk t02
.
c |2 |
(7.13)
(7.14)
Equation (7.13) is in the form of the nonlinear Schrdinger equation (NLSE), and is essentially identical to Eq. (5.24). In Chapter 5, we were discussing spatial solitons; now we are
dealing with temporal solitons time-dependent pulse solutions of Eq. (7.13). Whereas,
before, the dispersive term involved the transverse spatial coordinate , and was associated
physically with diffraction, now the corresponding parameter is , and the term represents
GVD. The scale length Ldiff (diffraction) is replaced by Ldisp (GVD), while the definitions
of Lspm and Lnl are identical. Once again, N is the soliton order.
For the lowest-order (N = 1) solution of Eq. (7.13), the GVD and SPM terms are in
perfect balance, and the soliton propagates as an unchanging solitary pulse. The solution is
U ( , ) = sech{ } exp{ 12 i }.
(7.15)
It is easy to verify that this result is a stationary solution of Eq. (7.13); i.e. U / = 0.
Since N = 1, it follows from Eq. (7.14) that
t2
c
(= Lspm ) = 0 (= Ldisp )
|2 |
|n2 | Ipk
(7.16)
so the peak intensity of the soliton is related to the duration through the equation
Ipk t02 =
c |2 |
.
|n2 |
(7.17)
The equations imply that if the soliton duration is halved, the strength of the dispersion is
quadrupled, and the peak intensity must be quadrupled to compensate if a stationary N = 1
137
ntens ty
60
30
200
100
tan
ce
dis
Fig. 7.4
100
0
0
10
e
l tim
loca
Steady-state propagation of the N = 1 soliton.
ntens ty
200
100
200
50
pro 1
pag
00
atio 1
50
nd
ista
nce
Fig. 7.5
100
0
0
10
0
e
l tim
loca
Propagation of the N = 2 soliton over the rst period.

solution is to be maintained. Rewriting Eq. (7.15) in terms of the original parameters yields
T ) = E pk sech{T /t0 } exp{ 1 iz/L}
E(z,
2
(7.18)
where L = Ldisp = Lspm . While the soliton envelope travels at the group velocity, the
phase term implies that the phase velocity is reduced from its original value of c/n0 ; see
Problem 7.3.
A numerical solution to Eq. (7.13) for N = 1 is shown in Fig. 7.4. The figure is entirely
unexciting visually, but its truly remarkable feature lies precisely in the fact that the profile does not evolve as it propagates. In fact, since we already know that Eq. (7.15) is a
stationary solution to the NLSE, the figure is largely a test of the numerics. However, the
simulation does provide the additional information that the first-order soliton is stable under
propagation.
More complex analytical solutions of Eq. (7.13) exist for higher integer values of N ,
i.e. when SPM is N 2 times stronger relative to GVD than for N = 1. For all higher-order
solitons, the solutions are periodic rather than stationary. The profiles evolve cyclically
under propagation with a universal period of z0 = 12 Ldisp that does not depend on N ; in
all cases, the profile returns to its initial form at integral multiples of z0 . Figure 7.5 shows
the structure of the N = 2 soliton over the first complete period. The contortions within the
period become increasingly complicated as N increases.
138
An interesting practical issue arises if the pulse launched into a soliton-supporting system
does not exactly match a soliton solution. Broadly speaking, the initial pulse evolves into the
soliton whose order Nf is the nearest integer to the initial (non-integer) value Ni given by
Eq. (7.14). The peak intensity and pulse width adjust themselves until Eq. (7.14) is satisfied
for N = Nf , and some fraction of the initial energy is dissipated in the process. If Ni < 12 ,
no soliton is formed, and the pulse simply disperses.4
Optical soliton propagation was first reported in 1980 by Mollenauer, Stolen and Gordon, who observed picosecond pulse narrowing and subsequent steady-state propagation in
an optical fibre [62]. Because the spreading associated with GVD was perfectly balanced
by SPM, solitons immediately became candidates for high bit-rate optical communications. However, the solitons still tended to lengthen under propagation because of energy
losses (see Eq. 7.17), and so frequent replenishment of the energy was necessary. In 1988,
Mollenauer and Smith demonstrated transmission of 55 ps pulses over 4000 km by supplying Raman gain at 42 km intervals [63]. Despite this, very few soliton-based fibre links are
currently in commercial use, largely because of advances in conventional fibre transmission.
Further information can be found in [64, 65].
7.5 Self-steepening and shock formation

When the effects of intensity-dependent refractive index are weak, its principal effect is on
the phase and spectral structure of an optical pulse. However, when it is strong, the pulse
envelope is distorted as well as the optical carrier wave [60,66]. To analyse this case, it is
necessary to go back to Eq. (7.4), which includes the differential operator on the right-hand
side that has been ignored in the previous two sections. When Eq. (7.6) is inserted for the
nonlinear polarisation and the terms rearranged, the equation becomes

(ngroup + n2 I )
2 E
n2
+
E = 12 i2 2 i
i
I E.
(7.19)
z
c
t
t
c
t
The most significant feature of this result is that the group index is now intensity dependent,
so the group velocity falls with intensity in the same way as the phase velocity. This means
that the pulse peak travels more slowly than the wings, so the trailing edge of the pulse
envelope will begin to steepen, leading potentially to the formation of a shock.
The simple geometry of Fig. 7.6 enables the propagation distance before a shock forms
on the pulse envelope to be calculated. Time runs horizontally in the figure, and distance
vertically. The slanting lines trace the paths of two neighbouring points on the pulse profile,5
which travel at different speeds because of the different local intensities. The figure shows
the lines meeting at S after propagating a distance L, at which point the intensity gradient
becomes infinite. The trajectory starting at A travels at speed v and reaches S in time t,
while the neighbouring trajectory from point B travels at v dv and takes t + dt to reach
4 See Agrawal [61] for more details.
5 These lines are known mathematically as the characteristics of the governing differential equation.
7.5 Self-steepening and shock formation
139
v dv
L
v
A
dt
Fig. 7.6
Space-time geometry of self-steepening and shocking.

the same point. Clearly L = vt = (v dv)(t + dt) and it follows that
dv
v
v2
c2
= =
=
dt
t
L
Ln2group
where ngroup is the group index. On the other hand, we also know that
c
v=
,
ngroup + n2 I
so
dv
cn2
dI
cn2 dI
=
2
.
2
dt
(ngroup + n2 I ) dt
ngroup dt
(7.20)
(7.21)
Combining Eqs (7.20) and (7.21) yields

L=
c
.
n2 dI /dt
The shock first occurs at the minimum value of this function, in other words at the point
where the intensity is falling most rapidly. Hence
Lshock = Min L =
c
ct
.
=
n2 Min {dI /dt}
1.83n2 Imax
(7.22)
Note that Min {dI /dt} in the denominator refers to the steepest negative gradient; the final
step applies to a Gaussian pulse with a full width at half maximum intensity of t. Figure 7.7
shows the intensity profile as the point of shocking is approached.
While this analysis is entertaining, one should probably not get too excited by the possibility of optical shock formation. The steepening of the trailing edge of an intense pulse (known
as self-steepening) is certainly a real phenomenon, but the analysis of shocking given here
is very simplistic. We have for instance ignored the other terms in Eq. (7.19), particularly
the one representing group velocity dispersion, which is the principal counter-balance to
shock formation. After all, as the profile steepens and a shock is imminent, the bandwidth
will be widening rapidly, and dispersive effects will be correspondingly strengthened.
An even more esoteric possibility is that steepening and shock formation might occur
not on the envelope of a pulse but within its carrier-wave structure. Surprisingly, the
140
intensity
0
4
Fig. 7.7
0
time
A self-steepened pulse envelope.

theory of this process was studied by Rosen as early as 1965 [67], before work on
envelope shocking had been developed. The issue was revisited in the late 1990s [68],
and carrier wave steepening has been considered as a tool in high harmonic generation
experiments [69].
7.6 Compression of self-phase modulated pulses

We saw in Chapter 6 that when a pulse is subject to positive group velocity dispersion
(GVD), the spectral width is not changed, but the temporal profile broadens, and the pulse
acquires a positive frequency sweep. In Section 6.5, we studied ways in which combinations
of gratings and prisms can be used to reverse this chirp, and potentially return the profile to
its original shape. We discovered in Section 7.4 that solitons occur when SPM and negative
GVD are judiciously balanced. On the other hand, we saw in Section 7.3 that, when a pulse
is up-chirped by SPM acting on its own, the temporal profile is largely unaffected, but the
spectrum is broadened, and the broadening can be dramatic (see Fig. 7.3). But once the
spectral width is increased, the bandwidth theorem (Section 6.3.3) tells us that the potential
pulse duration after compression is decreased, and this suggests that it should be possible
to use gratings and prisms to achieve shorter pulse durations than at the outset. Our aim in
the present section is therefore to discover how to manipulate the spectral phase profile to
compress pulses that have been chirped by SPM.
Although gratings and prisms work well for countering GVD, this is because GVD
creates an essentially linear chirp that is readily reversed. Unfortunately, things are not so
straightforward for SPM. There is no problem with the signs; SPM produces an up-chirp,
and requires negative GVD to reverse it. The difficulty is that, as indicated in Figs 7.1 and
141
phase
4
20
Fig. 7.8
10
0
relative frequency
10
20
Spectral proles of a pulse chirped by SPM with a peak phase shift of 6.5: spectral phase (bold), spectral intensity
(grey), spectral phase after optimal quadratic correction (dotted).
7.2, the chirp produced by SPM is not linear. The point is further evidenced in Fig. 7.8,
where the phase variation across the spectrum of a self-phase modulated pulse with a peak
phase shift of 6.5 is displayed (solid black line); the spectral intensity is shown as a grey
line for reference. The complexity of Fig. 7.8 is the manifestation in the frequency domain
of the non-uniform frequency sweep within a pulse chirped by SPM; see Fig. 7.1.
So what happens when one tries to compress the pulse of Fig. 7.8? First of all, let us try
the effect of an ideal compressor, one that simply eliminates the spectral phase variations
in the figure. Such a device is virtually impossible to realise in the laboratory, but the
operation is easily performed numerically by a single line of code that sets all the spectral
phase to zero. The result is shown as the bold line in Fig. 7.9, where the intensity scale is
normalised to the peak intensity of the original pulse (shown dotted grey), and the time-scale
is normalised to its FWHM. The peak intensity has increased by a factor of over 11, and
the width reduced by roughly the same amount. However, the quality is poor as evidenced
by the lobes in the wings.
A more realistic possibility is to try compressing the parent profile using a quadratic
phase correction. Of course, the phase variation in Fig. 7.8 is not quadratic so, to optimise
the process, one has to scan a range of values and choose the one that yields the highest peak
power. The corrected phase profile for this optimum condition is shown dotted in Fig. 7.8,
and the dotted black line in Fig. 7.9 shows the corresponding temporal profile. The peak
intensity is slightly lower than for ideal compression, and the side lobes have been replaced
by a broad pedestal supporting the central peak.
The key to improving the quality of the compressed pulse is to combine SPM with an
appropriate amount of positive GVD. This recipe turns out to produce a broad temporal
profile with a much smoother spectrum than when SPM acts alone, and it can therefore be
more efficiently compressed. However, the exact way in which SPM and GVD conspire to
142
relative Intensity
10
0
1
0.5
0.5
relative time
Fig. 7.9
Temporal intensity proles of a pulse chirped by SPM: under ideal compression (bold), under optimal quadratic
compression (dotted), parent (pre-compression) prole (grey).
2
phase
4
20
10
10
20
relative frequency
Fig. 7.10
Spectral proles of a pulse chirped by SPM combined with GVD: spectral phase (bold), spectral intensity (grey). The
corresponding proles without GVD from Fig. 7.8 are shown dotted.
achieve this desirable end is not entirely clear, and whole research papers have been written
about how to optimise the process. In practical compressors, one typically introduces SPM
and GVD in an optical fibre. Both processes act throughout the length of the fibre, but
whereas the effect of GVD increases with distance as the spectral bandwidth gets larger,
the SPM process weakens as the pulse spreads out and the peak power drops.
For simulation purposes, it is convenient to use the same scale lengths as in the soliton
studies. In the examples that follow, the strength of the initial pulse is N = 12, and the distance of propagation is 2.05Ldisp (= 1.3z0 where z0 is the soliton distance measure defined
at the end of Section 7.4).
The solid lines in Fig. 7.10 show the spectral characteristics of the broadened pulse
(before compression) for comparison with those of Fig. 7.8 (shown dotted). In the lower
143
intensity
0.2
0
30
20
10
10
20
30
intensity
10
0
1
Fig. 7.11
0
relative time
Temporal intensity proles corresponding to Fig. 7.10: broadened prole before compression (top), compressed
prole (bottom). In both frames the original parent prole is shown in grey.
part of the figure (in grey), the multiple peaks in the original spectral intensity profile have
been smoothed out, while the ripples in the original phase profile (shown higher up) have
largely been eliminated.
The corresponding temporal profiles are shown in Fig. 7.11. The broadened pulse before
compression is shown in the upper frame, and the corresponding profile after compression
in the lower frame. In both cases, the original parent pulse is shown in dotted grey, and
time and intensity are normalised to the characteristics of the parent, which has unit peak
intensity and unit FWHM on the scales used. The compression factor (relative to the parent)
is about 12, but the pulse still exhibits small side lobes, so it would be wise to display the
results on a logarithmic intensity scale to see the wing structure more clearly. After all,
experimentalists are interested in pulse contrast (the dynamic intensity range between the
pulse peak and its wings), and this cannot be assessed properly in the figure as it stands.
The profiles from Fig. 7.11 are displayed on a log scale in Fig. 7.12. The compressed
profile is shown in bold, while the dotted line shows the profile after quadratic compression
144
1x101
relative intensity
1x100
1x101
1x102
1x103
1x104
1x105
1x106
1
Fig. 7.12
0
relative time
Proles from the lower frame of Fig. 7.11 plotted on a logarithmic intensity scale. The dotted line corresponds to the
prole under optical quadratic compression from Fig. 7.9 (without GVD).
1.0
intensity
0.8
0.6
0.4
0.2
0.0
40
Fig. 7.13
20
0
times (fs)
20
40
Experimental pulse compression result. A 30 fs parent pulse (grey) is narrowed to under 7 fs (black) using a hollow
bre compressor. (Courtesy of J.W.G. Tisch, Imperial College Attosecond Laboratory.)
from Fig. 7.9. The grey line shows the profile of the original pulse. It can now be seen that
the broad pedestal of Fig. 7.9 has been eliminated, and the contrast ratio in the wings has
been improved by nearly four orders of magnitude.
The idea of first broadening the bandwidth of an optical pulse, and subsequently compressing it using diffraction gratings or their equivalent, goes back as far as 1969 [70].
However, it was not until the 1980s that the development of optical fibre technology allowed
the principles to be put effectively into practice. Some remarkable experimental results with
fibre-grating compressors were soon achieved. In 1984, for example, Johnson et al. [71]
compressed a 33 ps pulse to 410 fs, which corresponds to a compression ratio of 80, an
extremely impressive figure for a single-stage compressor. The recent result [72] shown in
Fig. 7.13 is on a much more challenging time-scale. The combination of a hollow fibre and
145
7.7 Group velocity dispersion in second harmonic generation
a grating was used to compress a 30 fs parent pulse (shown in grey) to under 7 fs (shown
in black). Since the pulse energy was around 0.5 mJ, the peak power after compression is
approaching 100 GW.

We now turn from self-action effects, in which a pulse modifies its own propagation characteristics, to inter-pulse effects in which pulses (often at different frequencies) interact.
We start by returning to second harmonic generation (SHG), and considering what happens when the process is driven by pulses rather than by quasi-monochromatic beams.
Pulses are attractive because the high peak power they offer makes for high conversion efficiency. But there is also a down side if dispersion causes the fundamental and
harmonic pulses to travel at different speeds. To understand this problem, which was
recognised in the early days of nonlinear optics [73, 74], we must consider once again
the relationship between phase velocity and group velocity that we first encountered in
Section 6.2.
Phase matching is, of course, as crucial to achieving high-efficiency nonlinear interactions with pulses as it is with monochromatic beams. In the case of SHG, the term implies
phase velocity matching between the fundamental and second harmonic waves.6 The question is: what about group velocity matching? In an SHG experiment with optical pulses,
what happens if the group velocities of the fundamental and second harmonic pulses are
different?
The key point to understand here is that, as explained in Chapter 6, phase and group
velocity effects are manifestations of the same phenomenon dispersion and hence they
are closely interrelated. In particular, when significant group velocity mismatch occurs in
SHG, it means that part of the bandwidth covered by the fundamental and harmonic pulses
is not phase matched.
To appreciate this important idea, consider the schematic dispersion diagram shown in
Fig. 7.14. The centre of the fundamental spectrum lies at A, and the centre of the corresponding second harmonic spectrum at B. If OAB is a straight line (as drawn), it means that the
centre frequencies are phase matched, while OB is double the length of OA because SHG
doubles the frequency. The dotted line CAD represents a section of the dispersion curve
around A, and its gradient corresponds to the group velocity of the fundamental pulse, just
as the gradient of OA represents the phase velocity. Now consider second harmonic generation for the frequency component represented by point C. The second harmonic component
that is phase matched lies at E, where OCE is a straight line, and OE is twice the length of
OC. A similar argument shows that the phase matched second harmonic of component D
lies at F. Since it follows from simple geometry that OCD and OEF are similar triangles,
the gradient of EBF, which is the section of the dispersion curve near B, is equal to that of
6 As we saw in Chapter 1, a more complicated condition applies in a general three-wave interaction.
146
C
A
Fig. 7.14
Schematic dispersion diagram showing the phase-matching conditions for second harmonic generation using a short
optical pulse.
CAD. So the group velocities of the fundamental and second harmonic pulses must be the
same for the pulse to be fully phase matched.
In practice, CAD and EBF are of course dispersion curves, and reflect properties of the
nonlinear material that cannot be adjusted, except sometimes to a limited extent in optical
fibres. Their gradients at A and B will not generally be the same, so the group velocities
of the fundamental and harmonic pulses will be different. Hence OCD and OEF will no
longer be similar triangles and, if the A and B components are themselves phase matched,
the same will not be the case for adjacent components.
It is worth mentioning that, under birefringent phase matching, CAD and EBF would be
sections of separate ordinary and extraordinary dispersion curves. For quasi-phase matching, one could perhaps think of EBF as a section of the curve that has been shifted to a
lower value of k by the periodic poling.
We now return to the original question: what is the effect of a group velocity mismatch
on SHG? As the fundamental pulse propagates through the medium, it starts to generate a
pulse of second harmonic in the normal way. But the harmonic pulse travels at a different
speed from the fundamental usually slower and so it begins to fall behind. Meanwhile,
the fundamental pulse continues to generate second harmonic waves as before, and these
continue to exude from the back end of the fundamental pulse. The final result is a flat second
harmonic signal, stretching out behind the parent pulse over a time interval determined by
the group time delay dispersion (GTDD) between the two frequencies accumulated through
the nonlinear medium.
A numerical simulation of the behaviour is shown in Fig. 7.15. A 1 ps fundamental pulse
(1 nJ at 1560 nm) generates a second harmonic signal at 780 nm in a 75 mm slab of
periodically poled lithium niobate (PPLN). The group time delay dispersion between the
two signals is 304 fs mm1 , which amounts to 22.8 ps in the entire slab. Centred on
zero local time in the figure are the original fundamental pulse (dotted grey), and its
147
500
power (W)
400
300
200
100
0
0
Fig. 7.15
10
local time (ps)
20
Second harmonic generation by a short optical pulse in the presence of group velocity dispersion.
depleted profile (solid grey) after propagation through the slab, by which time 44% of
the original energy has been lost to the harmonic. The harmonic signal at four stages
during the interaction (12.5 mm, 25 mm 37.5 mm and 75 mm) is spread out behind
the fundamental, as expected, and the extent of the final harmonic profile corresponds
to the 22.8 ps group time delay noted earlier. The slanting tops of the harmonic signals are caused by depletion of the fundamental; without depletion the tops would be
essentially flat.
Group velocity mismatch not only distorts the shape of the SHG pulse, but it also lowers the overall conversion efficiency too. This is because the second harmonic intensity
increases as the square of the distance under phase-matched conditions but, once a section
of the second harmonic becomes detached from the parent, its growth is ended and so its
effective interaction distance has been reduced. For example, consider a simplistic picture
in which the GTDD through the sample was ten times the fundamental pulse duration. The
first tranche of second harmonic would detach itself from the parent one-tenth of the way
through the nonlinear medium, so its peak intensity would reach only one-hundredth of
the level it would have achieved had detachment not occurred. This is partially compensated by the fact that the final second harmonic pulse is ten times as long as the parent,
but the overall account shows an energy deficit of a factor of ten. The bottom line is that
group velocity mismatch is bad for efficiency, and smears out the second harmonic pulse
as well.
Another example of the same phenomenon is shown in Fig. 7.16, this time for a fundamental pulse (5 fs duration) containing only a few optical cycles [75]. This result was
obtained by the direct solution of Maxwells equations using a variant of the well-known
finite difference time domain (FDTD) method. As in Fig. 7.15, the group velocity of the
second harmonic is lower than that of the fundamental, so the high-frequency harmonics
lie to the rear (the right) of the fundamental pulse in the figure.
148
1
0.8
0.6
Electric field [arb]
0.4
0.2
0
0.2
0.4
0.6
0.8
1
840
860
880
900
920
940
960
Time [fs]
Fig. 7.16
Second harmonic generation by a few-cycle pulse computed by the direct solution of Maxwells equations; from [75]
by permission.
7.8 Optical parametric chirped pulse amplication

7.8.1 The basic principle
Optical parametric chirped pulse amplification (OPCPA) [76, 77] combines the technique
of chirped pulse amplification (CPA) with optical parametric amplification (OPA).7 In
chirped pulse amplification [78], a principle originally developed in radar engineering
(and known as chirp radar) is applied to short-pulse laser physics. The problem to be
solved is how to amplify pulses of EM radiation to extreme peak power without damaging
the equipment that supplies the energy. The recipe is simple. Start with a short pulse of
the desired duration but only modest energy: stretch it in a systematic way, normally by
chirping; amplify it to form a broad high-energy pulse; and finally compress the pulse
back to its original duration. This creates high-energy pulses that are both extremely short
and extremely intense. The peak power needs to be kept within acceptable limits in all
system components. In the optical case, both stretcher and compressor are normally based
on diffraction gratings; the arrangement of Fig. 6.8 is used for the stretcher, and the standard
zigzag configuration of Fig. 6.5 for compression. Damage to the compressor is avoided by
the use of large gratings, and by putting the entire compression system under vacuum if
necessary.
7 The process is occasionally called chirped pulse optical parametric amplification (CPOPA).
149
Local time (compressed pulse) (ps)
2.0
1.5
1.0
0.5
5x107
6x106
4x106
Power (W)
3x107
2x107
2x106
1x107
0x100
0x100
5
Fig. 7.17
Power (compressed pulse) (W)
4x107
0
Local Time (ps)
10
Pulse proles in OPCPA showing inefcient use of the available bandwidth: signal pulse (bold line), idler pulse
(dotted), pump pulse (grey with parent dotted). The 65 fs compressed signal pulse is shown on the right.
[Pump: 50 J, 10 ps, 400 nm (centre). Signal pulse: 2 nJ, 7 ps, 714 nm (centre) chirped at 10 THz/ps (= 17 nm ps1 ).
Crystal: 8 mm LBO, all beams in the x-y plane, beam angle 31.14 .]
The attraction of combining chirped pulse amplification with optical parametric amplification is the wide bandwidth offered by parametric amplifiers. The standard procedure is to
use a signal pulse with a relatively broad temporal profile and a strong chirp. These characteristics necessarily imply wide spectral bandwidth, and hence the potential for exceptionally
short and intense pulses after compression. The potential of pulses with such extreme characteristics in novel scientific applications is immense, and this is why OPCPA is the focus of
so much current attention [79]. The pump pulse must naturally be wide enough to amplify
the full temporal (and spectral) extent of the signal. The pump itself need not be chirped,
although a weak chirp can sometimes be applied to enhance the amplification bandwidth
as we shall see shortly.
As a practical example, we demonstrate the OPCPA process for a 10 ps pump pulse
driving a 7 ps signal pulse chirped at 10 THz ps1 (= 17 nm ps1 ) in an 8 mm sample of
LBO. Figure 7.17 shows the various profiles after the interaction, the signal in black, the
idler in dotted black, the pump in grey, and the initial pump in dotted grey. The separate
profile on the right (plotted against the top time axis and the right-hand power axis) shows
the 65 fs signal pulse after optimal quadratic compression. A full list of parameter values is
given in the figure caption.8
8 Further background to the numerical techniques used to create to Fig. 7.17 and similar figures can be found
in [80].
150
signal/idler wavelength (nm)
1000
900
800
700
600
396
Fig. 7.18
398
400
402
pump wavelength (nm)
404
Phase-matched signal/idler wavelengths in LBO as a function of the pump wavelength.
Why are the signal and idler profiles in Fig. 7.17 so narrow? Why has only a thin slice
in the centre of the pump pulse been used, and the rest of its energy wasted? The reason is that the signal pulse is strongly chirped, and only the central part of the signal
spectrum is phase matched. Time corresponds to frequency in a chirped pulse, and so
only a small part of the temporal profile can participate in an efficient nonlinear interaction. Different methods for overcoming this problem are described in the following
sections.
7.8.2 Bandwidth enhancement with a chirped pump

Insight into the problem is provided by Fig. 7.18, where the phase matched signal wavelength for LBO is plotted as a function of the pump wavelength. The upper branch of the
graph (dotted) traces the idler wavelength. The positive slope of the signal branch near 400
nm indicates that slightly lower (or higher) pump wavelengths are needed to phase-match
lower (or higher) signal wavelengths. So the problem could be at least partially solved if
the pump pulse carried a chirp that was accurately matched to that of the signal. Careful
measurement of Fig. 7.18 reveals that the gradient at 400 nm is around 17, and this sug1
gests that if the pump chirp were 17
th that of the signal, or roughly 1 nm ps1 (which
1
converts to 1.875 THz ps ), phase matching could be maintained over a much wider
bandwidth.
Figure 7.19 shows how Fig. 7.17 is transformed when a chirped pump pulse with the
specified characteristics is used. A much wider time (and frequency) slice is now involved
in the interaction and, with the wider bandwidth, the compressed profile narrows from 65 fs
to 18 fs. Further refinements are possible too. One could, for instance, introduce a nonlinear
151
local time (compressed pulse) (ps)

1.5
1.0
2.0
0.5
8x108
6x106
power (W)
4x106
4x108
2x106
2x108
power (compressed idler) (W)
6x108
0x100
0x100
5
10
local time (ps)
Fig. 7.19
Improved OPCPA bandwidth usage (compared to Fig. 7.17) achieved by chirping the pump pulse at 1.875 THz/ps. See
caption to Fig. 7.17 for other details.
ks
ki
kp
Fig. 7.20
Non-collinear phase matching in OPCPA for bandwidth optimisation.
chirp on the pump or signal, a possibility that has been demonstrated successfully in the
laboratory [81].
7.8.3 Bandwidth enhancement with non-collinear beams

Another approach to bandwidth enhancement in OPCPA is to use non-collinear beams. To
understand how this might help, consider the case of a narrow-band pump at 3 driving
a broadband signal at 2 [79]. The k-vector triangle that ensures phase matching for the
signal and idler centre frequencies is shown in Fig. 7.20, where the signal propagates at
an angle to the pump. The question is now whether there is a particular value of that
maximises the phase-matched bandwidth.
We will suppose that the signal and idler are quasi-ordinary waves, while the pump is
an extraordinary wave. In this case, the magnitudes of k2 and k1 are fixed, but the length
152
of k3 depends on its angle to the x-axis.9 For particular values of and that close the
triangle, the cosine rule gives
2k3 k2 cos = k32 + k22 k12 .
(7.23)
Keeping the angle fixed, we now seek the condition for which the phase matching is
unaffected (i.e. the k-vector diagram remains closed) when small changes are made to 2 .
As this is done, there will be changes of opposite sign to 1 , as well as to k1 , k2 , and the
angles and in the figure. Differentiating Eq. (7.23) with respect to frequency, and using
the fact that d2 = d1 , yields
k1
dk1
dk2
= (k2 k3 cos )
d1
d2
(7.24)
which reduces to
k3 cos = k2 + k1 (vg2 /vg1 )
(7.25)
where vg2 and vg1 are the signal and idler group velocities. By combining Eqs (7.23) and
(7.25) and with further use of the sine and cosine rules, one obtains for the angle the
expression

sin
1
= tan
(7.26)
k2 /k1 + cos
where the angle (defined in the figure) is given by = cos1 {vg2 /vg1 }.
The application of this technique in LBO is illustrated in Fig. 7.21 where the results of
simulations based on collinear and non-collinear beams are compared directly. The dotted
and solid black lines in the figure show the signal profiles at the end of the interaction
in the respective cases, while the corresponding pump profiles are shown in grey. For the
parameter values used, which are similar to those in Figs 7.17 and 7.19, the angles are
= 1.55 , 4.03 and 2.48 . As before, since the signal is strongly chirped,
a wider pulse translates immediately to a wider spectrum. The compressed signal pulse
widths in the collinear and non-collinear cases are respectively 31.0 and 12.5 fs, which
underscores the benefits of non-collinear geometry.
7.8.4 Optical parametric generation in three dimensions

Virtually every example in this book has been based on the plane-wave approximation in
which transverse variations in intensity and phase across the wavefronts of the interacting
beams are ignored. By contrast, the contour plots displayed in Fig. 7.22 show the transverse
intensity profiles of the pump (top), signal (centre), and idler (bottom) after interacting in
an optical parametric amplifier. The three beams have been separated purely for display
purposes; in reality they lie on top of each other, and the cross-hairs define the common
centre line. Once again, the nonlinear medium is LBO, with all beams in the x-y plane.
9 LBO is a biaxial crystal (see Sections 3.5 and 4.6). If, as we are assuming, all beams lie in the x-y plane, those
polarised in the z-direction propagate as quasi-ordinary waves.
153
1x107
power (W)
8x106
4x106
0x100
5
local time (ps)
Fig. 7.21
Enhanced OPCPA bandwidth utilisation by non-collinear phase matching: signal pulse (bold line), nal pump pulse
(grey). Corresponding collinear proles are shown dotted.
[Pump: 100 J, 10 ps, 523.5 nm. Signal: 2 nJ, 7 ps, 850 nm (centre) chirped at 10 THz ps1 (= 24 nm ps1 ). Crystal:
8 mm LBO, all beams in the x-y plane, pump propagation direction 15.58 , non-collinear angle = 1.545 .]
pump
400
300
200
100
signal
0
100
200
idler
300
400
200 100
Fig. 7.22
0
100
microns
200
Two-dimensional pump, signal, and idler proles in OPCPA, with the three proles separated for display purposes.
White crosses mark the common centre line. The seed signal pulse is angled to the right, and the idler runs to the left
to ensure phase matching; see Fig. 7.20. Pump walk-off (to the left) is evident.
[Pump 523.5 nm, 1 mJ, 40 ps, 240 m super-Gaussian. Signal 850 nm, 25 pJ, 20 ps, 240 m super-Gaussian, 1.3 deg.
Crystal: LBO xy, 7 mm, 14.49 deg, 0.85 pm V1 .]
154
250
200
150
idler
microns
100
50
0
pump
50
100
signal
150
200
651
0
mm
Fig. 7.23
Top view of an OPCPA interaction similar to Fig. 7.22, but with narrower beams to illustrate the separation of signal
and idler more clearly.
[Pump: 0.5 mJ, 125 m super-Gaussian. Signal 0.1 mJ, 125 m super-Gaussian, 1.3 deg. Other data as Fig. 7.22.]
The signal and idler are quasi-ordinary waves, but the pump is polarised in the x-y plane
and so is subject to walk-off, which is clearly visible in its leftward movement of about
60 m in the 7 mm crystal. The walk-off angle is about 0.5 in this case. A non-collinear
beam configuration is used, in which the signal beam is angled rightward at about 1.3 , and
this leads to an idler angled leftward at 2.1 (see Fig. 7.20). Over 7 mm, these angles would
suggest a 160 m rightward shift of the signal, and a 250 m leftward shift of the idler,
neither of which is evident in Fig. 7.22.
A vivid demonstration of what is happening is provided by Fig. 7.23, which shows an
aerial view of the process, with the three beam paths superimposed. To make the point more
dramatically, this figure was generated using narrower beams than in Fig. 7.22, and the
relative intensities of the three signals were also adjusted for the same reason.10 For the
first 4 mm in the crystal, there is no sign of signal or idler; the pump appears undepleted
and, apart from pump walk-off, nothing appears to be happening at all. The parametric
generation process is however slow-cooking. Detailed examination of the data shows
that the signal and idler are in fact growing steadily in this region, but this is invisible on
the linear scale used in the figure. After all, according to the coupled-wave equations, the
signal and idler both need the presence of the other (and of course the pump) in order to
grow. Since the interaction would cease if the two parted company, they are forced to cling
together for survival. However, about two-thirds of the way through the sample, the pot
comes to the boil, and the signal and idler burst out on opposite sides of the pump beam,
travelling at the expected angles.11
Further results on transverse effects in OPCPA are presented in [82].
10 The angling of the signal and idler beams is reversed in Fig. 7.22, but this does not affect the argument in any
way.
11 Because the axial and transverse scales are different, the beam angles in Fig. 7.22 are magnified by about a
factor 10.
155

In the late 1960s, the development of laser mode-locking techniques enabled picosecond
optical pulses to be generated for the first time and, since these were beyond the time
resolution of oscilloscopes, new measurement techniques had to be devised. Two approaches
were explored: streak cameras, and autocorrelation techniques based on nonlinear optics.
Versions of the latter approach, highly developed to various levels of sophistication, continue
to be used for optical pulse diagnosis today, when pulses in the femtosecond (fs) regime are
commonplace, and pulses deep in the attosecond regime are hot topics of current research.12
To measure something, one always needs a benchmark. But optical pulses are shorter
than all the normal benchmarks, so one resorts to using an optical pulse as its own measure,
and this is what is done in autocorrelation methods. The principle is to overlap two replicas
of an optical signal with a controllable time displacement between them, and to record some
parameter that is sensitive to the degree of overlap as a function of the time displacement.
One needs to find a property of the system that behaves differently when the two replicas
overlap than when they travel separately (i.e. without overlapping).
A device that immediately springs to mind is the Michelson interferometer, in which an
optical signal is divided into two beams that follow paths of different length before being
recombined. It is easy to show that the fraction of the pulse energy that reaches the detector
under these circumstances is13
f ( ) =
where GE =
1
2
GE ( )
1+
GE (0)

(7.27)
E(t)E(t + )dt is the autocorrelation function of the (real) electric field
E(t). However, it is easy to show that GE is the Fourier transform of the intensity spectrum,
so a Michelson interferometer records the same information as a spectrometer, albeit in a
different form.
Additional information can be obtained from a Michelson interferometer if nonlinear
optical detection is used, the most common practice being to record the second harmonic of
the interferometer output. To analyse this case, we introduce the complex analytical signal,
which is related to the real field by
E(t) = ReV (t) = 12 (V (t) + V (t)).
(7.28)
This equation is actually just a version of Eq. (2.4) in which the electric field is evaluated at
a fixed point in space (say z = 0), and V (t) includes the complete carrier wave structure.14
12 1 fs = 1015 s and 1 as = 1018 s.
13 The rest of the energy is reflected back to the source. Note that, in what follows, E(t) is the field amplitude
after two encounters with the 5050 beam splitting, and therefore one-half that entering the interferometer.
14 Equation (7.28) does not define V (t) uniquely; a rigorous procedure for doing this is given in Appendix E.
156
The second harmonic signal depends on the time-averaged fourth power of the total
optical field at the output of the interferometer, namely
S( ) = (E(t) + E(t + ))4 .
(7.29)
A laborious calculation now reveals that the relative harmonic signal registered at the output
is given by

S( )
2GI ( ) + R( )
= 1+
(7.30)
S()
GI (0)
where GI ( ) is the autocorrelation function of the intensity (defined here as I (t) = |V (t)|2 )
given by

GI ( ) =
I (t)I (t + )dt.
(7.31)
The background harmonic signal represented by S() in Eq. (7.30) is the net harmonic
signal created by the two replicas travelling separately (i.e. with such a large time separation
that there is effectively no overlap at all). Rapidly varying terms contained in R( ) in
Eq. (7.29) are given by

R( ) = Re ei20
V 2 V2 dt + 2ei0
(V 2 V V + V V V2 )dt
(7.32)
where V (t) = V (t)ei0 t is the slowly varying part of V (t), and V and V are abbreviations
for V (t) and V (t + ), respectively
If the rapidly varying terms in Eq. (7.30) are ignored, it is easy to see that the equation
predicts that the signal at the centre of the overlap region ( = 0) is three times what it is
in the wings ( = ). The contrast ratio of the intensity autocorrelation function (IACF)
is therefore said to be 3. Unfortunately, however, the mathematics is tending to obscure the
physics of what is going on, so we shall adopt a simple-minded approach to try to interpret
the result. Consider two square replica pulses of unit width and intensity I . Outside the
overlap region, where the pulses travel separately, 2I 2 units of harmonic intensity will be
registered, where the 2 arises because there are two pulses in succession, and the square
because the second harmonic varies as the square of the fundamental intensity. By contrast,
at the centre of the overlap region ( = 0), the intensity will presumably be doubled, which
suggests that the harmonic signal should be (2I )2 = 4I 2 , twice what it is in the wings.
Although this argument indicates correctly that a higher signal is registered within the
overlap region, it is wrong in detail because the fine structure in the interference between the
two replicas (and contained in R( )) has been ignored. Constructive interference doubles
the field at the anti-nodes, so the anti-node intensity rises by 4 (not 2), and the corresponding harmonic signal rises to 16I 2 , or eight times what it is in the wings. Away from
the anti-nodes, the interference pattern will be characterised by cos2 fringes in the intensity,
which will lead to cos4 fringes in the harmonic signal. Since the cycle average of cos4 is
3
2
8 , the cycle-averaged harmonic signal is 6I , which is 50% higher than the crude result
157
relative intensity
0
8
Fig. 7.24
0
relative time
SHG autocorrelation trace showing the ne structure resolved and (grey) unresolved.
obtained in the previous paragraph, and a factor 3 above the 2I 2 background value. This
confirms the 3 peak-to-background ratio implied by Eq. (7.30) when the rapidly varying
terms are ignored.
Figure 7.24 shows the IACF of the linearly chirped pulse of Fig. 6.3. The full record
(including rapidly varying terms) is shown in black, and the cycle-averaged record in grey.
Notice that the fine structure does not extend over the entire overlap region, let alone beyond
it. This is a consequence of the chirp, which means that strong constructive and destructive
interference between the two replicas of the pulse only occurs near the centre of the pattern,
where the temporal displacement is small. If one moves sufficiently far enough from the
centre, the fine structure is completely smeared out.
The IACF, however accurately measured, does not provide sufficient information to
reconstruct even the complete intensity profile of the original pulse, let alone its phase
structure [83]. A technique (crude by modern standards) for observing the frequency sweep
of a chirped pulse was demonstrated by Treacy as early as 1970 [84]. In Treacys experiment,
pulses were spectrally dispersed, and IACFs of the separate components were recorded
simultaneously. The tilted autocorrelation record created by a chirped pulse provided a
direct visual demonstration of the frequency sweep.15
A number of highly sophisticated techniques of this general type are now available which,
in conjunction with suitable numerical tools, enable the complete field structure of the original waveform to be reconstructed. Most of the popular techniques are known by memorable
acronyms such as FROG (Frequency-Resolved Optical Gating) [85] and SPIDER (Spectral
Phase Interferometry for Direct Electric field Reconstruction) [86]. The field of optical pulse
measurement has become increasingly complex, and a full-length book has been written on
one single technique [87].
15 This kind of measurement is known technically as an optical sonogram.
158
Exhaustive studies of all the available options for optical pulse characterisation can be
found in [88, 89]. Most of these methods involve nonlinear optics, although there is no
reason in principle why linear techniques cannot be used, provided a non-time stationary
filter of some kind is involved.16 Linear methods would certainly offer increased sensitivity
and they continue to be studied for this reason [90].

In a typical mode-locked laser, a parent pulse circulating in the laser cavity creates a train of
replicas that emerge through the output mirror with a temporal spacing close to the roundtrip period of the cavity. Even in a CW laser, there are always sources of jitter, so the parent
pulse inevitably changes slightly from one round trip to the next, and one replica pulse
is never precisely identical to the one that came before. However, even in the ideal case
where the pulse envelope itself is perfectly stable, differences between the phase and group
velocities will cause the optical carrier wave to shift under the pulse envelope from transit
to transit. This phase slip is highly undesirable in experiments where precise waveform
control is needed, and methods are being developed to control it.
The spectrum of a mode-locked pulse train can be found by convolving the spectrum of
an individual pulse (e.g. Eq. 6.8) with the longitudinal mode structure of the cavity. In the
idealised case, the modes are represented by a simple comb function of the form

C() =
( j )
(7.33)
j
where the angular frequencies of the modes j follow from the standing-wave condition
L = j j /2 in a cavity of length L. If we assume the cavity to be filled with a material of refractive index nj = n(j ), it is easy to show that the mode frequencies are given
by j = j (c/nj L). The presence of nj in this formula means that, if the material is
dispersive, the modes will not be uniformly spaced, so a pulse circulating freely in the
cavity will spread out, just as it would if it propagated through the same length of material under any other circumstances. However, the mode-locking mechanism, be it active
or passive, will counter this tendency, enforcing equal mode spacing , and enabling us
to write
j = j + 0
(7.34)
where 0 (< ) is a small frequency remainder that takes account of the fact that j will
generally not be an integral multiple of . Under mode-locked conditions,
=
c
Ln
(7.35)
16 We saw earlier that a basic Michelson interferometer yields nothing more than an intensity spectrum, but no
non-time stationary filter is involved in this case.
159
Fig. 7.25
50
100
radians
105
200
Few-cycle pulses with a carrier-envelope phase slip between them.

where n is an effective refractive index that depends on the values of nj within the
bandwidth, and also on the mode-locking mechanism.
The time-dependent real field comes from the sum of all participating modes, namely

E(t) = 12
Aj exp{i(j t + )} + c.c.
j
1
2
Aj exp{i(j t + 0 t + )} + c.c.
(7.36)
where the amplitudes Aj are real, and is the common phase of all modes at t = 0. Careful
thought indicates that the pulse envelope repeats with a periodicity of T = 2/, while
the phase of the optical carrier changes from one pulse to the next by cep = T 0 =
2 (0 /), which is known as the carrier-envelope phase slip (CEP slip).
Figure 7.25 shows an example of the phenomenon described. In the pulse on the left, the
peak of the envelope coincides with a peak of the carrier wave, and the pulse structure is
cosine like, i.e. symmetrical about the centre; in this case the carrier-envelope phase offset
is zero. On the other hand, the carrier wave of the pulse on the right has slipped leftward
relative to the envelope by 0.4, and the structure is therefore skewed. If the inter-pulse
phase slip were raised to 0.5, the pulse would be anti-symmetrical (and sine-like). In
pulses containing many optical cycles, CEP effects are hard to spot by eye, but in few-cycle
pulses of the kind shown in Fig. 7.24, they are much more obvious.
Since in Eq. (7.35) depends on L, fine adjustment of the cavity length should enable
the frequency comb j to be adjusted, and 0 (and hence cep ) to be brought to zero. In
this case, CEP will not change from pulse to pulse. Various techniques for CEP stabilisation
have been devised. Consider, for instance, what happens if a mode-locked signal with
a spectrum given by Eq. (7.34) is subject to second harmonic generation. The frequency
comb governing the harmonic can be found by summing the frequencies of two fundamental
160
combs so that17
l = j + k = (j + k) + 20 .
(7.37)
If the mode-locked pulse bandwidth is sufficiently wide, the high-frequency wing of the
fundamental comb will overlap the low-frequency wing of the second harmonic comb.
Now, comparison of Eqs (7.34) and (7.37) indicates that the two combs will be aligned in
the region of overlap only if 0 = 0. But in that case, the CEP slip will be eliminated,
so it follows that CEP stabilisation can be achieved if the two combs can be forced into
synchronism.
This desirable outcome can be realised experimentally by heterodyning the frequency
components of the fundamental and harmonic waves in the region of spectral overlap, and
using a servo loop controlling the cavity mirror spacing L to bring the beat note to zero.
This creates a train of mode-locked pulses in which the location of the carrier waves under
the pulse envelope is stabilised, and no inter-pulse phase slip occurs [9193].
It must be emphasised that stabilising the CEP simply stops it varying, and does not of
itself determine the absolute carrier phase at the peak of the envelope, which is given by
in Eq. (7.36). However, it is important to recognise that few-cycle pulses are extremely
fragile objects and, even if pulses with a specific CEP could be generated on demand, the
CEP would be significantly altered if the pulses passed through a few microns of glass;
see Problem 7.6. This property can in fact be turned into an opportunity, because it means
that the CEP can be tuned by fine adjustment of the path length through a rotatable plate.
As explained in Chapter 10, the absolute CEP is of crucial importance in applications such
as high harmonic generation and attosecond pulse generation, and CEP tuning is used to
optimise the process.
Problems
7.1 Calculate Lspm (the characteristic length for self-phase modulation) in fused silica at
an intensity of 1 TW cm2 . Express the answer in units of the vacuum wavelength.
[n2 (fused silica) = 3 1016 cm2 W1 .]
7.2 Verify that Eq. (7.15) is a solution to Eq. (7.13).
7.3 Show that the phase velocity of an N = 1 soliton is approximately (cn2 Ipk /2n20 ) lower
than its low-intensity value in the same medium. (This question is essentially the same
as Problem 5.5. It is equally applicable to the spatial solitons of Section 5.7 and to the
temporal solitons of Section 7.4.)
7.4 Prove the final step of Eq. (7.22) for Lshock . Show that the ratio Lshock /Lspm is roughly
four times the number of cycles contained within the intensity FWHM of an optical
pulse.
7.5 Try the effect of adding a 3 component to cos t and notice how the distortions of the
carrier wave depend on the relative phase. This is most conveniently done by plotting
the waveforms using a PC.
17 Notice that the second harmonic process includes sum frequency generation among the components of the
fundamental.
161
7.6 CEP phase slip arises in the propagation of an optical pulse through fused silica as
a manifestation of the difference between the phase and group velocities. Using the
information given in Fig. 6.2 as a guide, estimate the distance the pulse needs to travel
before the phase and group time delays differ by a quarter optical cycle (i.e. before a
phase slip of /2 has occurred). Assume that the vacuum wavelength is 800 nm. Think
about the implication of the answer.
Some quantum mechanics
8.1 Introduction
A key weakness of the simple approach to nonlinear optics adopted in Chapter 1 was
that the physical origin of nonlinearity in the interaction of light and matter was hidden
inside the (n) coefficients of the polarisation expansion.1 This is such a fundamental issue
that it is difficult to avoid some mention of how nonlinearity arises within a quantum
mechanical framework, even in an introductory text. Unfortunately, the standard technique
for calculating the nonlinear coefficients is based on time-dependent perturbation theory,
and the expressions that emerge begin to get large and unwieldy even at second order.
While every effort has been made in this chapter to provide a gentle lead-in to this aspect
of the subject, this is almost impossible to achieve given the inherent complexity of the
mathematical machinery.
From a mathematical point of view, Schrdingers equation is linear in the wave function,
but nonlinear in its response to perturbations. At a fundamental level, this is where nonlinear
optics comes from. The perturbations of atoms and molecules referred to here arise from
external electromagnetic fields. When the fields are relatively weak, the perturbations are
relatively small, and the theoretical machinery of time-dependent perturbation theory can
be deployed to quantify the effects. This is the regime where the traditional polarisation
expansion of Eq. (1.24) applies, indeed the terms in the expansion correspond to successive
orders of perturbation theory.
The early part of this chapter is devoted to a brief review of quantum mechanical principles. Perturbation theory techniques are then developed, and expressions for the coefficients
governing many of the basic nonlinear processes described elsewhere in this book are
derived. The focus is mainly on simple cases, and no attempt is made to obtain expressions
for the coefficients in their most general form.
When the perturbations become sufficiently strong and/or the time-scales sufficiently fast,
the polarisation expansion breaks down, and other analytical techniques have to be developed.
This situation frequently arises close to a resonance, although resonances can sometimes be
treated within perturbation theory. Examples of resonant effects, including the stimulated
Raman effect, self-induced transparency and electromagnetically-induced transparency are
discussed in Chapter 9. High harmonic generation also occurs within the strong-field regime,
and this topic is treated in Chapter 10, albeit from an essentially classical standpoint.
1 The other defect was that the tensor nature of the coefficients was ignored.
162
163
8.3 The Schrdinger equation
8.2 Basic ideas

The state of a quantum mechanical system can be specified in several different ways: the
wave function (x), or the state vector |$ of Dirac are two obvious examples. The most
fundamental approach to quantum mechanics is based on |$, which is a vector in an abstract
Hilbert space, whereas (x) is one of many possible representations of |$ in particular
Hilbert spaces [94].2
The idea that the state of a system can be represented in different ways should not come
as a surprise. After all, the position-dependent wave function (x) is related by Fourier
- k, and x and k
transformation to the momentum space wave function O(p), where p = h
are transform variables. Since (x) and O(p) both contain complete information about the
same state, they are equivalent representations of the system.
Although we will use Dirac formalism in the early part of this chapter, readers who are not
familiar with the notation can treat it simply as shorthand for the traditional wave function
expressions. As noted by Rae [95], in an equation such as

2 dx = %1| A |2$
1 A
(8.1)
the object on the right-hand side can be regarded as a code for the expression on the left-hand
side. The bra vector %1| translates into the conjugated wave function 1 , the ket vector |2$
into 2 , and integration is implied.
The words represented and representation have appeared several times in the last few
paragraphs. In the early history of quantum mechanics, the words were also used to describe
different ways of apportioning the time dependence of the system between the state vectors
and the operators. Most modern texts use the word picture in this latter connotation.3 In
the Schrdinger picture, the time dependence resides entirely in the state vectors, while
in the Heisenberg picture, it resides entirely in the operators. On the other hand, in the
interaction picture, the time dependence is split, with the intrinsic element in the operators
and the remainder (which is usually the interesting part) in the state vectors.
8.3 The Schrdinger equation

In state vector notation, the Schrdinger equation reads
|$
i
= - H |$
t
h
(8.2)
where H = H 0 + V includes the unperturbed Hamiltonian H 0 and a perturbation Hamiltonian V . In the context of nonlinear optics, V is often written in the dipole approximation as
= er.E, where e is the elementary charge and
= er is the dipole operator.
V = .E
2 For simplicity, a single space coordinate is applied to wave functions in this section
3 Throughout Messiah [94], the word representation is put in quotation marks when it is used in this sense.
164
It is convenient to expand |$ in terms of the orthonormal energy eigenstates of the system

using one of the two alternative forms4

|$ =
bn |nt $ =
an |n$.
(8.3)
n
Here, the state vector |nt $ contains the intrinsic time dependence of the system, and is
related to the time-independent |n$ by
|nt $ = |n$ exp{in t}
(8.4)
where n = En /h- and En is the energy eigenvalue. It follows that an = bn exp(in t} so,
in the second form of Eq. (8.3), the intrinsic time dependence is contained in the coefficients
an , rather
than in |nt $. The occupation probabilities of the different energy eigenstates are

|an |2 = |bn |2 , and these sum to unity so that5

|an |2 =
|bn |2 = 1.
(8.5)
Written in terms of wave functions, Eq. (8.3) reads

=
bn n =
an n
(8.6)
- } and the n are time independent.

where the energy eigenfunctions n = n exp{iEn t/h

. /
The orthonormality of the energy eigenstates implies that n n dr = nn = n n
where dr stands for dxdydz.
Substituting Eq. (8.3) into Eq. (8.2) allows the time evolution of the quantum system to
be expressed is terms of either an or bn via the equations
i
a n = in an Vnl al ;
h
i
bn = Vnlbl
h
(8.7)
where Vnl = %n|V |l$ is the perturbation matrix element, Vnl = Vnl exp{inl t}, and nl =
n l . Since V = .E,
we can write the matrix element out in full as
y
Vln = (xln Ex + ln Ey + zln Ez ) = (%x$ln Ex + %y$ln Ey + zln Ez )
(8.8)
where the j th Cartesian component of ln is written ln , or equivalently as %j $ln which is

slightly easier on the eyes.
The expectation value of an arbitrary operator A is given by

%| A |$ =
al an Anl
bl bn Anl exp{inl t}
(8.9)
nl
nl
wave vector has unit norm.
%nt | bn =
%n|an .
n
5 This is easily proved from Eq. (8.3) using the orthonormality of the energy eigenstates and the fact that the
4 In terms of bras rather than kets, the equation reads %| =
8.4 The density matrix
165
where Aln = %l| A |n$ . In nonlinear optics, the operator of primary interest is the dipole
= er, because it leads directly to the polarisation via the equation
operator

P = N %$ = N
nl al an .
(8.10)
nl
8.4 The density matrix

8.4.1 Definition of the density matrix
Later in the chapter, we will treat a range of nonlinear optical problems using state vectors
and wave functions as in the previous section. However, an alternative formalism based
on the density operator (or density matrix) offers significant advantages in some contexts,
particularly when resonant effects are involved and/or damping becomes important. We
will review the density matrix approach in this section.
The Schrdinger equation (8.2) is entirely equivalent to the equation

i
= - H ,
(8.11)
t
h

where = |$%| is known as the density operator, and the commutator H , =
H H . Using Eq. (8.3), we can write the density operator in the form

= |$%| =
am al |m$%l|
(8.12)
ml
and define the density matrix as

nn = %n||n
$ = an an .
(8.13)
Readers who are not familiar with Dirac notation should feel reassured by the second form
of Eq. (8.13), which defines the density matrix in terms of the probability amplitudes from
Eq. (8.3). Indeed, for the simplest two-level system (shown in Fig. 8.1), the density matrix
becomes simply

|a0 |2 a0 a1
00 01
=
=
(8.14)
10 11
a1 a0 |a1 |2
1
10
Fig. 8.1
Two-level system.
166
in which the diagonal elements nn = |an |2 are just the population probabilities of the energy
eigenstates 0 and 1. The sum of the diagonal elements is therefore unity. The population
density Nn of level n is given by N nn where N is the total atomic/molecular number
density. As we shall discover shortly, the off-diagonal elements govern the dipole moment
of the atoms, and hence the polarisation of the medium. Notice that when (for example)
a0 = 1, it follows that a1 = 0, from Eq. (8.5), and that the off-diagonal elements of are
zero.
The time evolution of can be found by differentiating Eq. (8.13) and using Eqs (8.7).
The result is

i
nn = inn nn Vnq qn nq Vqn
h q
(8.15)
where the first term on the right-hand side indicates that the intrinsic time dependence of
the off-diagonal element nn is exp{inn t}. However, it is easy to remove this intrinsic
motion by defining nn = nn exp{inn t} = bn bn . The time dependence of nn is then
given by

i

nn = Vnq qn nq Vqn

h q
(8.16)
= V exp{i t}, etc. The switch from to represents a transformation

where Vnq
nq
nq
to the interaction picture where the intrinsic time dependence has been transferred to the
operators, and the remaining time dependence in nn is the result of external perturbations.
At this point, an important issue must be raised: if the density matrix method is entirely
equivalent to the state vector method of Section 8.2, what is its advantage? Why not stay
with the more familiar state vectors and wave functions? The answer is that the density
matrix approach offers clear benefits when one is dealing with mixed (as opposed to pure)
states. If all members of an ensemble of atoms/molecules are in the same quantum state, the
state is said to be pure. However, it is far more common for different atoms/molecules to
be subject to different stochastic influences, and it is much easier to incorporate these effects
within a density matrix treatment than in the state vector approach. Using the two-level case
as an example, we can adopt the generalised definition of the density matrix given by

|a0 |2 a0 a1
=
(8.17)
a1 a0 |a1 |2
where the overbars imply averages over all members of the representative ensemble.
Stochastic processes affecting the diagonal elements of the density matrix (e.g. spontaneous emission) redistribute the population between the different states, of which there are
just two in this simple case. But effects also exist that randomise the phase relationships
between the states but have no effect on the populations, and these cause the off-diagonal
elements of to decay on a faster time-scale.
We can take these stochastic effects into account by adding phenomenological damping
terms to Eq. (8.15). For off-diagonal elements of the generalised density matrix, the equation
8.5 Introduction to perturbation theory
167
becomes

i
nn = i nn nn Vnq qn nq Vqn
h q
(n = n )
(8.18)
where nn = nn inn and nn is the rate of loss of phase coherence between states
n and n . For diagonal elements, the general form is

i
nn =
Vnq qn nq Vqn
(8.19)
m mm nn h
m
where the first term represents a population arriving from other levels, and the second
population departing to lower levels. These results could equally have been written in terms
of nn .
8.4.2 Properties of the density matrix

It is clear from its definition that the density matrix is Hermitian, which means that nn =
n n , and that its trace is unity (see Eqs 8.5 and 8.14). But it also possesses the key property
0 1
In the context of nonlinear
that the expectation value of any operator A is A = Tr{ A}.
. /
%
= Tr{ }
=
optics, this enables one to find the dipole moment from
nn n n , and
nn
once the dipole moment is known, the polarisation follows immediately from P = N %$,
which is consistent with Eq. (8.10). Typically, only the off-diagonal elements of the dipole
moment matrix are non-zero so, for the simplest two-level model, we have

0
01
=
.
(8.20)
10
0
It follows that
P = N %$ = N (01 10 + 10 01 )
(8.21)
in this special case. The result highlights the important point mentioned earlier that the
polarisation is associated with the off-diagonal elements of the density matrix. If P is nonzero, it necessarily implies that the system is in a superposition state; otherwise, all but one
of the an would be zero, and hence all the off-diagonal elements of the density matrix would
be zero too.
Notice that %$ is real because 10 = 01 . We shall in fact assume that the matrix elements
themselves are real, in which case 10 = 01 ; this can usually be arranged without loss
of generality. As already noted, we use the notation %j $10 to represent the j th Cartesian
component of 10 .
8.5 Introduction to perturbation theory

In the context of nonlinear optics, perturbation theory involves the successive approximation
of the solution for the quantum mechanical state of the system. The procedure can be
168
implemented using either the probability amplitudes an (or bn ), or the density matrix nn .
It is good to be familiar with both approaches, because each has its particular merits and
demerits. The density matrix method includes stochastic effects in a natural way, which is
an obvious advantage. But the associated algebra is often more awkward, and one needs
to keep a careful watch out for spurious cancellations situations where terms appear to
cancel, but only if causality is ignored. We shall encounter some of the problems that may
arise in Section 8.8.2 below.
The application of perturbation theory proceeds in the same way whichever approach
is used. Starting from a zeroth-order solution, the relevant evolution equation is used to
calculate a first-order correction. The modified solution is then reinserted into the equation,
and a second-order correction duly emerges. The operation can in principle be repeated
ad infinitum, but the algebraic labour increases at an alarming rate. In practice, perturbation
theory is rarely extended to fourth order (at least in nonlinear optics), but the complexity at
third order is already severe.
In standard perturbation theory notation, one writes H = H 0 + V , and the successive
corrections to the probability amplitudes in the form
an = an(0) + an(1) + 2 an(2) + 3 an(3) +
(8.22)
The parameter (which is nothing to do with wavelength) is used to keep track of terms
of different order, and is set to unity at the end of the calculation. In nonlinear optics,
however, there are numerous other indicators of term order (the number of matrix elements,
for example), and so we simply write
an = an(0) + an(1) + an(2) + an(3) +
(8.23)
or the analogous density matrix equation

(0)
(1)
(2)
(3)
nn = nn + nn + nn + nn +

From Eq. (8.13), we have nn =
(2)
an an ,
and so it is tempting to assume (for example) that
(0) (2)
21 = a2 a1
(8.24)
(1) (1)
+ a 2 a1
(2) (0)
+ a 2 a1
(8.25)
in which orders of the amplitudes in the right-hand side terms always sum to 2. One must
be a little cautious here, because the density matrix potentially includes damping, while
the probability amplitudes do not. However, in the absence of damping, Eq. (8.25) can
be verified by taking the time derivatives of both sides and using the relevant evolution
equations; see Problem 8.1.
8.6 First-order perturbation theory

In deriving perturbation theory formulae for the linear and nonlinear susceptibilities, we
will focus mostly on simple cases where only two or three levels are involved. Readers
wishing to see the results in their most generalised form should consult books such as Boyd
[1] or Butcher and Cotter [96].
8.6 First-order perturbation theory
169
First-order perturbation theory in a two-level model system is very simple, and leads to a
formula for the linear susceptibility (1) , a parameter that we first encountered in Chapter 1.
(0)
(0)
In the state vector approach, we take the zeroth-order solution as a0 = 1, a1 = 0, which
means that all the atoms are in the ground state. Equations (8.7) then read
i
(1)
(1)
a 1 + i10 a1 = - V10
h
(8.26)
(1)
where a1 is the first-order contribution to a1 . If the atoms are subject to a (real) perturbing
field whose components are

Ej = 12 E j exp{it} + c.c.
(j = x, y, z)
(8.27)
then
V10 = 12

%j $10 E j exp{it} + c.c.
(8.28)
where %j $ln ln andj = x, y, z. Equation (8.26) can now be integrated to give

(1)
a1
1
%j $10
= 2h
E j exp{it} E j exp{it}
+
.
10 +
10
The analogous result within the density matrix approach is

1
(1)
(1)
%j $10
10 = 01 = +
2h
10 +
10
(8.29)
(8.30)
which is identical to Eq. (8.29) apart from the presence of damping via the complex
frequency 10 = 10 i10 .
The polarisation of the medium is
Pi = 12 Pi exp{it} + c.c.

= N a0 a1 %i$10 + a1 a0 %i$01

(1)
(1)
= N 01 %i$10 + 10 %i$01

%i$01 %j $10
N %j $01 %i$10
= +
E j exp{it} + c.c.

2h
10
10 +
(8.31)
where the more general form of the result with damping included has been quoted. Notice
(1)
(1)
that terms in exp{it} from both 10 and its conjugate 01 have contributed to the final
% (1)
result. From Pi = 0 j ij E j , it follows that the linear susceptibility is
(1)
ij
N
= 0 h
%j $01 %i$10
%i$ %j $
+ 01 10
10
10 +

.
(8.32)
170
For a many-level system, Eq. (8.30) generalises to

(1)
n0
1
%j $n0
= 2h
+
n0 +
n0

(8.33)
and Eq. (8.32) then becomes

(1)
ij =

%i$0n %j $n0
N %j $0n %i$n0
+
.

0 h- n
n0
n0 +
(8.34)
In summary, the applied electric field of Eq. (8.27), acting via the perturbation matrix
(1)
elements in Eq. (8.28), creates first-order components a1 in Eq. (8.29), or equivalently
(1)
(1)
the off-diagonal density matrix elements 01 and 10 of Eq. (8.30). These lead directly to
corresponding terms in the polarisation of the medium (Eq. 8.31), and hence in the linear
susceptibility (Eqs 8.32 and 8.34). The resonance structure of Eq. (8.34) will be explored
in Section 9.2.
As in any forced oscillator system, the correction terms oscillate at the applied frequency
rather than the natural resonance frequency, and this fact can be represented by adding
virtual levels at h- into the energy level scheme of Fig. 8.1, as shown in Fig. 8.2. We
will offer an interpretation of Eqs (8.32) and (8.34) in Section 8.7.2, where the significance
of the virtual levels is considered in greater detail.
Finally, it is worth remembering that the linear susceptibility is directly related to the
(1)
dielectric constant matrix through the equation ij = 1 + ij , which we encountered in
Chapter 3; see the discussion surrounding Eqs (3.5) and (3.6). In general, ij is Hermitian (ij = ji ) but, away from resonance, the elements are real and the matrix becomes
symmetric. It is this property that allows it to be diagonalised (Eq. 3.6), and the principal
dielectric axes defined. However, Eq. (8.34) indicates that, if ij is to be real, %j $0n %i$n0
must be real too. We have, of course, assumed the individual matrix elements themselves
to be real, so the issue does not arise. But had we not done so, the need for the product to be
real when i = j raises some interesting issues. When i = j , the product is real in any case.
1
+1
10
Fig. 8.2
Two-level system showing virtual levels accessed in rst-order perturbation theory.
8.7 Second-order perturbation theory
171

8.7.1 Second harmonic generation
We have just played round 1 of perturbation theory, and we now move on to round 2. First
of all, we add a third level (2) to the two-level scheme of Fig. 8.2, to create the three-level
system shown in Fig. 8.3; the stack of virtual levels to be discussed later is shown on the
right. We will now work out the second harmonic generation coefficient for this system on
the assumption that the dipole matrix elements between all three pairs of levels are nonzero, as indicated by the double arrows in the top-left-hand section of the figure. We ignore
damping (setting nn = nn ), and we use the probability amplitude approach because it
turns out to be algebraically simpler.
(1)
At first order, we already have a1 from Eq. (8.29) and, since levels 1 and 2 are
functionally equivalent, we have by analogy
(1)
a2
1
%j $20
= 2h
j
+
.
20
20 +
(8.35)
The second-order corrections can now be found from

i
i
(2)
(2)
(1)
%k$12 (E k eit + c.c.)a2(1) .
a 1 + i10 a1 = - V12 a2 = h
2h
k
2
+2
1
20
10
+1
Fig. 8.3
Three-level system showing virtual levels accessed in second-order perturbation theory
(8.36)
172
Substituting Eq. (8.35) into Eq. (8.36) yields

E j E k exp{i2t}
E j E k exp{i2t}
1
(2)
%k$12 %j $20
a1 = - 2
+
+
4h
(10 + 2)(20 + ) (10 2)(20 )
j, k
(8.37)
(2)
where only terms at 2 are shown explicitly. The analogous result for a2 is obtained by
the exchange 1 2. The second harmonic polarisation can now be found from6
PiSHG = 12 Pi exp{i2t} + c.c.

(2) (0)
(2) (0)
(1) (1)
= N a2 a0 %i$02 + a1 a0 %i$01 + a1 a2 %i$21 + c.c.
(8.38)
and the final result is7

%j $01 %k$12 %i$20
%j $02 %k$21 %i$10
N
Pi2 = - 2
+
2h
(10 )(20 2)
(20 )(10 2)
j, k
%k$01 %i$12 %j $20

%k$02 %i$21 %j $10
+
(10 )(20 + ) (20 )(10 + )

%i$01 %j $12 %k$20
%i$02 %j $21 k10
+
+
E j E k .
(10 + 2)(20 + ) (20 + 2)(10 + )
(8.39)
The same result can of course be obtained using the density matrix approach, but the algebra
is less straightforward. The problem is that when Pi2 is evaluated in that case, the result
has eight terms rather than six, and four contain factors of the type (21 ), etc. in the
denominator. These combine to form terms 3 and 4 in Eq. (8.39), but the step is best avoided.
Equation (8.39) is readily generalised to a many-level system and made more compact at
the same time. By introducing summations over multiple intermediate levels, the formula
becomes

%j $0n %k$nn %i$n 0
%k$0n %i$nn %j $n 0
N
2
Pi = - 2
+
2h
(n0 )(n 0 2) (n0 )(n 0 + )
j k nn

%i$0n %j $nn %k$n 0
E j E k .
(n0 + 2)(n 0 + )
(8.40)
Equation (8.39) is just the special case of Eq. (8.40) where there are just two intermediate
levels, and the terms with (n, n ) = (1,2) and (2,1) are written separately.
8.7.2 Discussion and interpretation

Equations (8.39)(8.40) merit detailed discussion. In fact, it will be helpful to start by
revisiting the first-order case, and relating the terms in the linear susceptibility (Eq. 8.32)
6 Note that all six terms in the second line contain terms at +2.
7 Since j and k are equivalent dummy indices in Eqs (8.39) and (8.40), they can be exchanged at will within the
matrix element products, to make the expressions look neat.
173
term1
term2
1
+1
Fig. 8.4
Diagrammatic representation of the structure of Eq. (8.32).
to Fig. 8.2. As noted in Section 8.5, this diagram represents the two real levels (0 and 1)
that feature in the equation by solid lines, and the two virtual levels at h- by dotted
lines. Both terms in the equation use the same real level sequence 0 1 0, and the
) and %i$ %j $ /(
matrix elements in %j $01 %i$10 /( 10
01
10 10 + ) have been ordered so
that the subscript sequence 0110 reflects this. On the other hand, the first term involves
- . This can be seen by
the virtual level at +h- and the second term the virtual level at h
)
inspecting the respective denominators. If the complex nature of 10 is ignored, ( 10
becomes (10 ), which is the angular frequency separation between the real level 1 and
the virtual level +1; equally, (10 + ) is the separation between the same real level and
the virtual level 1.
Figure 8.4 uses these ideas to construct a more detailed diagrammatic representation
of the mathematics. The two solid arrows in the centre represent the real level sequence
0 1 0, which is common to both terms. The two grey arrows on the left indicate
the virtual level sequence 0 +1 0 used by the first term, while the grey arrows
on the right indicate the sequence 0 1 0 used by the second term. The centreleft portion of Fig. 8.3 therefore represents the first term (term 1), while the centre-right
portion represents the second term (term 2) as indicated. The broad double arrows indicate
the energy (or frequency) differences between the participating real and virtual levels in
each case, a short double arrow implying a strong term. Term 1 (with the negative sign in
the denominator and the smaller detuning) will naturally dominate term 2 in this case. In a
multi-level model, the situation will naturally be more complicated, but it often remains true
that a good approximation to the susceptibility can be obtained from a single dominant term.
It is also worth writing the dominant term in Eq. (8.31) in a slightly different way in order
to highlight the structure of the result. In terms of Pi , the equation reads

%j $01 E j exp{it}
Pi = N %i$10
)
h- ( 10
j
(8.41)
174
where the term in brackets is the (dimensionless) ratio of an interaction energy (on the top)
and the detuning energy (on the bottom).
We now turn to Eq. (8.39), which is the product of second-order perturbation theory and
so naturally more complicated. There are now two intermediate states in the real and virtual
level sequences and, accordingly, three matrix elements in the numerator of each term and
two brackets in the denominator representing successive angular frequency differences.
The equation has been organised so that reading the brackets in the denominator from left
to right reveals the sequence of participating real and virtual levels, by direct extension of
the procedure used at first order. Thus, in the first term of Eq. (8.39), the appearance of
10 and 20 in the successive brackets indicates that the sequence of real levels is 0
1 2 0, while the respective appearance of and 2 implies that the removal of a
fundamental photon from the field first takes the system to virtual state +h- and the removal
of a second then raises the system further to +2h- . Finally, a second harmonic photon is
emitted, which returns the system to its ground state. Notice that the matrix elements in
the numerator have been arranged so that the same real level sequence appears when the
indices are read from right to left, namely 011220.
The six terms in Eq. (8.39) represent two different real level sequences each with three
different virtual level sequences. Terms 1, 3 and 5 (the three on the left) are for the real level
sequence 0 1 2 0, while the other three are for the sequence 0 2 1 0.
A diagrammatic representation of the first three is presented in Fig. 8.5, which is based on
the same ideas as Fig. 8.4. The black arrows (on the far left and right) indicate the real level
sequence, and the three possible virtual level sequences are shown in the central columns.
The sequences for each of the six terms are summarised in Table 8.1.
As in Fig. 8.4, the double arrows in Fig. 8.5 indicate the values of the frequency denominators, and there are, of course, now two per term whose product affects the term strength. The
equation, the table, and the figure have all been arranged so that term 1 is the strongest and
term 6 the weakest. The matrix elements have been ordered to reflect the level sequences.
term 1
term 3
term 5
+2
2
1
+1
Fig. 8.5
Diagrammatic representation of the structure of Eq. (8.39).
175
Table 8.1 Real and virtual level sequences in Eq. (8.39).

Term
Real sequence
Virtual sequence
1
2
3
4
5
6
0120
0210
0120
0210
0120
0210
- +2h
- 0
0 +h
- 0
0 +h +2h
0 +h h 0
- h
- 0
0 +h
- 0
0 h 2h
- 0
0 h 2h
Consider, for example, term 4, which involves %k$02 %i$21 %j $10 /((20 )(10 + )).
Reading both top and bottom from left to right, the interpretation is that in the first step, a
fundamental photon is absorbed and the system goes to real state 2, in the second a harmonic
photon is emitted and the system goes to state 1, and in the final step, a further fundamental
photon is absorbed and the system returns to the ground state.
The most important point to come out of this discussion is probably the existence of a
dominant path determined by real levels that lie close to the virtual levels at +h- and
- . Taking the dominant term on its own, Eq. (8.39) reduces to
+2h

%j $01 %k$12 %i$20
N
Pi2 = - 2
E j E k
2h
(10 )(20 2)
(8.42)
jk
which in many cases will give a good approximation for the second harmonic polarisation.
For comparison with Eq. (8.41), we can reorganise the equation into the form
Pi2
1
2N
%i$20

jk

%k$12 E k
.
h- (10 ) h- (20 2)
%j $01 E j
(8.43)
This shows that the most basic effect of taking perturbation to the next order has been to
add a second energy ratio into the formula.
8.7.3 The second harmonic coefficient

We now need to reconcile Eq. (8.40) with Eq. (4.5) where the SHG coefficient was defined
via the equation
Pi2 = 12 0

jk

ijSHG
k Ej Ek .
(8.44)
176
The notation differs slightly from that in Chapter 4, but in an obvious way. The conclusion
is that

N
%j $0n %k$nn %i$n 0
%k$0n %i$nn %j $n 0
SHG
ij k = - 2
+

0 h
(
)(
2)
(
n0
n0 )(n 0 + )
n0
nn

%i$0n %j $nn %k$n n
+
.
(8.45)
(n0 + 2)(n 0 + )
We consider the special case where i = x and the field contains y and z components. Then
Eq. (8.44) becomes

SHG
SHG
Px2 = 12 0 xyz
+ xzy
Ey Ez
(8.46)
while Eq. (4.10) reads
Px2 = 20 d14 E y E z .
If we focus just on the dominant path, Eq. (8.45) gives

N %z$01 %y$12 %x$20 + %y$01 %z$12 %x$20
Px2 = - 2
E y Ez
2h
(10 )(20 2)
and the conclusion is that

N
%z$01 %y$12 %x$20 + %y$01 %z$12 %x$20
SHG
SHG
1
=
2d
=
.
14
xyz
xzy
2
2h- 2 0
(10 )(20 2)
(8.47)
(8.48)
(8.49)
SHG and SHG with the individual terms in the numerator

One should not seek to associate xyz
xzy
of Eq. (8.49). Not only is there no way of distinguishing the two contributions by experiment,
but the order of the fields has not been preserved in the derivation in any case.
Finally, we must address the symmetry issue. Back in Chapter 1, we used simple symmetry
principles to show that SHG (and other second-order processes) are forbidden in media
possessing inversion symmetry. So what basic principle ensures that the SHG coefficient
of Eq. (8.44) is zero under these circumstances?
The answer is that quantum states possess definite parity in a medium with inversion
symmetry. But transitions between states of the same parity are forbidden and the associated
dipole matrix elements are zero. In Eq. (8.40), the sequence of states in the matrix element
triple products is 0 n n 0, and there is no way of choosing even and odd parity for
the three states alternately to close the loop. In general, it follows that the product of any odd
number of dipole matrix elements involving a closed sequence of states is necessarily zero.
8.7.4 Sum frequency generation

The case of sum frequency generation 1 + 2 = 3 can be treated by direct extension
of the second-order perturbation theory techniques deployed above in the special case of
SHG. The most important message is that there will be twice as many terms as for SHG;
twice as many diagrams of the type shown in Fig. 8.5 can be drawn because there are now
177
two distinct orders in which the upward arrows can appear: 1 first and then 2 or vice
versa. Equation (8.42) for the dominant term now generalises to

%k$01 %j $12 %i$20 E j (1 )E k (2 ) %k$01 %j $12 %i$20 E j (2 )E k (1 )
N
Pi (3 ) = - 2
+
2h
(10 2 )(20 3 )
(10 1 )(20 3 )
jk
(8.50)
which, by exchanging j and k in the second term (which is OK because they are dummy
indices under a summation) can be compacted to

N
%k$01 %j $12 %i$20
%j $01 %k$12 %i$20
Pi (3 ) = - 2
+
E j (1 )E k (2 ).
2h
(10 2 )(20 3 ) (10 1 )(20 3 )
jk
(8.51)
As expected, this equation contains two terms, whereas there was only one in Eq. (8.42).
For the same special case that we used in Section 8.7.3, the x-component of polarisation
is now

N
%z$01 %y$12 %x$20
%y$01 %z$12 %x$20
Px (3 ) = - 2
+
E y (1 )E z (2 )
2h
(10 2 )(20 3 ) (10 1 )(20 3 )

%y$01 %z$12 %x$20
%z$01 %y$12 %x$20
+
+
Ey (2 )Ez (1 ) (8.52)
(10 2 )(20 3 ) (10 1 )(20 3 )
which parallels Eq. (8.48) for SHG. From the definition of the sum frequency generation
(SFG) coefficient in Eq. (4.4), we also have

SFG
SFG
Px (3 ) = 0 xyz
Ey (1 )E z (2 ) + xzy
Ez (1 )E y (2 ) .
(8.53)
SFG and SFG are distinct, and can be associated with the
Notice that, in contrast to SHG, xyz
xzy
respective terms in Eq. (8.52). The first coefficient relates to the situation where the fields
at 1 and 2 are, respectively, in the y- and z-directions, and the second to the case where
the frequencies are reversed.
A related issue is intrinsic permutation symmetry (IPS), which was first discussed in
Section 4.3.2. When Eq. (4.3) is written out in full for SFG, the x-component reads

SFG
SFG
Px (3 ) = 12 0 xyz
(3 ; 1 , 2 ) + xzy
(3 ; 2 , 1 ) E y (1 )E z (2 )

SFG
SFG
+ 12 0 xzy
(3 ; 1 , 2 ) + xyz
(3 ; 2 , 1 ) E y (2 )E z (1 ).
(8.54)
Given the structural similarity between Eqs (8.54) and (8.52), it is tempting to try to match
the coefficients in the latter equation with the individual terms in the former. However,
SFG ( ; , ) and SFG ( ; , ) cannot be
as noted in Chapter 4, the effect of xyz
3
1 2
3
2 1
xzy
distinguished, since both involve the same product E y (1 )E z (2 ), and nor can that of
SFG ( ; , ) and SFG ( ; , ) in the second row of Eq. (8.54). There is in fact
xzy
3
1 2
3
2 1
xyz
no justification for making the suggested identifications.
Under IPS, it is conventional (but not obligatory) to assume that the paired coefficients in Eq. (8.54) make equal contributions, and hence to write ijSFG
k (3 ; 1 , 2 ) =
178
SFG ( ; , ). This enables the equation to be contracted to

ikj
3
2 1

SFG
SFG
Px (3 ) = 0 xyz
(3 ; 1 , 2 )E y (1 )E z (2 ) + xzy
(3 ; 1 , 2 )E y (2 )E z (1 ) .
(8.55)
Matching Eqs (8.55) and (8.52) now yields

%z$01 %y$12 %x$20
N
%y$01 %z$12 %x$20
SFG
xyz (3 ; 1 , 2 ) = - 2
+
(8.56)
2h 0 (10 2 )(20 3 ) (10 1 )(20 3 )
SFG ( ; , ). Notice that the order of the frewith a closely analogous formula for xzy
3
1 2
SFG
SFG ( ; , ). In
quencies is now fixed because xzy (3 ; 2 , 1 ) is included within xyz
3
1 2
conclusion, remember that IPS is merely a sensible convention, not a fundamental principle.
8.7.5 Optical rectification

The case of optical rectification (OR) can be treated if one goes back to Section 8.7.1 and
extracts the zero frequency terms instead of those at 2. Like most perturbation theory
derivations, the calculation is arduous. The result for the DC polarisation is

N
%i$0n %j $nn %k$n 0
%i$0n %k$nn %j $n 0
PiDC = - 2
+
4h
n0 (n 0 )
n0 (n 0 + )

j k nn

%j $0n %i$nn %k$n 0
%k$0n %i$nn %j $n 0
+
(n0 )(n 0 ) (n0 + )(n 0 + )

%j $0n %k$nn %i$n 0
%k$0n %j $nn %i$n 0
+
+
E j E k
(n0 )n 0
(n0 + )n0
and an expression for the optical rectification coefficient can be deduced from

Pi (0) = 12 0
ijOR
k (0; , )Ej ()Ek ().
(8.57)
(8.58)
jk
The matrix element involving the coordinate i is associated with the polarisation and those
carrying j and k with the incident field. As before, the terms have been organised so that,
by reading from left to right, the sequence of real and virtual states can be worked out.
An intriguing aspect of Eq. (8.57) is the need for the right-hand side to be real, since the
DC polarisation is real by definition. But even if we take the matrix elements to be real,
the presence of Ej Ek makes it less than obvious that the equation satisfies the condition.
In fact, detailed examination shows that the expression really is real; see Problem 8.2.
8.8 Third-order perturbation theory

There is no problem in principle with taking perturbation theory to third order, but the
mathematics becomes hugely tedious. Fortunately, there is no need for this labour if one
179
Fig. 8.6
Four-level system with virtual levels for the dominant term in third harmonic generation.
is prepared to trust the diagrams of Fig. 8.5, and make the necessary extensions. Indeed
for a system of four levels 0123 of alternate parity, the contribution to the third harmonic
polarisation from the dominant path shown in Fig. 8.6 can be written down by inspection,
namely

N
%j $01 %k$12 %l$23 %i$30
Pi3 = - 3
E j E k E l .
4h
(10 )(20 2)(30 3)
(8.59)
j kl
Notice the extra matrix element, the extra bracket, the extra field component, and the extra
2h- in the denominator. In reality, the single path 0 1 2 3 0 would naturally
be summed over multiple intermediate paths.

To end this chapter, we tackle the more challenging case of intensity-dependent refractive
index (IDRI). Some intriguing problems arise in the calculation of the IDRI coefficient,
which provide valuable insight into the inner workings of perturbation theory, and highlight
the relative advantages and disadvantages of the probability amplitude (PA) and density
matrix (DM) approaches.
We simplify the discussion by restricting attention to a three-level system in which
states 0 and 2 have the same parity (and hence 02 = 20 = 0), and we also assume
that all fields are polarised in the x-direction. The terms we need to calculate in order
to obtain the third-order polarisation using the two different approaches are listed in
Table 8.2.
180
Table 8.2 Terms needed to calculate the IDRI coefcient

0
1
(0)
Probability
amplitude
Density matrix
a0
(0)
=1
00 = 1
(0)
a0 , a2
(1)
00 , 11 , 20
a1
10
(2)
(2)
(2)
(2)
Third- order polarisation

(3)
(3) (0)
a1
(2)
(3)
a1 a0
(3)
10 , 21
(3)
(2) (1)
, a2 a1
, +c.c.
(3)
10 , 21 +c.c.
2
+2
1
+1
Fig. 8.7
Real level sequence (bold) for Eq. (8.60), with virtual level sequences (grey) for each of the six terms in the equation.
It can be shown quite straightforwardly (using either the PA or the DM method) that
the contribution to the IDRI polarisation from the real level sequence 0 1 2
1 0 is
IDRI
P(via
state2)

2 2
N x01 x12 E x2 E x
1
1
=
2
= +
4h- 3
10
20 10
10 20 10
1
1
1
1
+
+ + +
+ +
+ + + ++ +
10 20 10
10 20 10
10 20 10
10 20 10

(8.60)
= = 2 and
where we have adopted a shorthand notation where n0
= n0 , n0
n0
++
n0 = n0 + 2. Diagrams representing the terms in this equation are shown in Fig. 8.7.
The real state sequence (0 1 2 1 0) is indicated by the black arrows at the far
left, while the six sets of grey arrows show the virtual state sequences for the six terms in
Eq. (8.60) in the same order. With the pre-factor of 2, these are in fact the 12 terms in Table
VII of Ward [97].8
8 Each path in Fig. 8.7 represents two terms because one downward arrow corresponds to the E in Eq. (8.60)
x
and the other to the polarisation; these can come in either order.
181
We now need to incorporate the state sequence 0 1 0 1 0 into this result,

and this is where the problems begin. It turns out that the DM approach provides the easiest
route to the right answer, although it is decidedly trickier than the PA method at the first step.
(2)
Consider, for example, what happens when one tries to find 11 . If we assume (correctly)
(2)
(1) (1)
that 11 = a1 a1 in the absence of damping, the answer in that case is easily found from
Eq. (8.29) to be

10 01
(E x2 ei2t + c.c.)
E x E x
E x E x
(1) (1)
a 1 a1 =
+
.
(8.61)
+
4h- 2
(10 + )(10 ) (10 + )2
(10 )2
On the other hand, if we try to derive the more general form of this result starting from
(2)
Eq. (8.15), the time derivative of 11 comes out as

i
(2)
(1)
(1)
V10 01 10 V01
11 = h q

1
1
1
1
2
i2t
2
i2t
+
Ex e
Ex e

+
10 01 10
10 +
10
10
.

= i -2
1
1
1
1
4h
Ex Ex
10 + 10 + 10 10
(8.62)
We now face a puzzling situation. For if damping is ignored and 10 replaced by 10 ,
integration yields the terms in ei2t in Eq. (8.61), but the terms in E x E x are not recovered
because the curly bracket in Eq. (8.62) goes to zero.
What has gone wrong? The answer is not connected with damping. It is certainly true that,
(2)
when damping is included, the cross-term in Eq. (8.62) no longer vanishes, and 11 then
diverges on integration. This feature is associated physically with the population growth of
level 1 caused by absorption from level 0 in the wings of the resonance. But the PA approach
on which Eq. (8.61) is based does not include damping in any case, so this cannot be the
origin of the extra terms in the equation.
We can recover the missing terms if we use a device suggested by Takatsuji [98], and
include a causality factor et in the electric fields, where is a small positive constant that is
taken to be zero at the end of the calculation. As well as attaching this factor to the fields in
Eq. (8.62), it is also necessary to take account of its effect at first-order, which can be done
simply by reinterpreting 10 as (10 i). With thus included, the curly bracket in Eq.
(8.62) is no longer zero, and when the equation is integrated, the missing terms in Eq. (8.61)
reappear. The moral of this story is that one should exercise caution when cancelling terms
under an integral that would diverge if integrated separately.
We have in fact now surmounted the main obstacle within the DM approach to finding
the contribution to the IDRI polarisation from the level sequence 0 1 0 1 0,
i.e. where the central level is the ground state. This is the same problem that arises from the
appearance in the general formula for the IDRI coefficient quoted in Ward [97] of terms
such as ((n0 )m0 (n 0 ))1 . The four central terms in Eq. (8.60) are of this type,
and of course presented no difficulty when m = 2. But the term becomes singular if we try
to set m = 0, and that is precisely what we are now seeking to do.
182
Within the PA approach, this problem manifests itself in a different but revealing way
(2)
when one tries to calculate a0 from the equation
2
3

|01 |2
i
1
1
(2)
(1)
(0)
(8.63)
+
E x E x a0
a 0 = - V01 a1 = + i
h
4h- 2
10 + 10
where only time-independent terms on the right-hand side are shown. The expression in
the square brackets clearly represents a frequency change in a0 , which is associated with
a shift in the energy of the ground state (0) in the presence of the excited state (1) and the
coupling field. This is a real effect, known as the AC Stark shift. Since the intrinsic time
- ), it is clear from Eq. (8.63) that the ground state energy
dependence of a0 is exp{iE0 t/h
is lowered if < 10 . An upward shift of the same amount will occur for level 1, which
means that the two levels move apart. One would therefore expect that, for these two levels
alone, IDRI causes a reduction in the polarisibility.
There are several ways of managing this problem:
Orr and Ward [99] used a technique based on Bogoluibov and Mitropolskys method
of averages to remove the singularities and to derive a modified formula for the IDRI
coefficient;
Takatsuji [98] obtained the identical result by renormalising the zeroth-order Hamiltonian
to include the AC Stark shifts;
Boyd [1] cites Hanna, Yuratich and Cotter [100] for an algebraic device that leads to the
same conclusion.
But, as noted already, we have no need to adopt any special measures because the difficulties
we encountered with Eqs (8.61) and (8.62) were the manifestation within the DM approach
of the same basic difficulty. Now that we have solved them, we can simply press on.
The contribution to the IDRI polarisation from the sequence 0 1 0 1 0 is
included in

(3)
(3)
P = N 01 10 + 10 01
(8.64)
where
(2)

i 10
i
(3)
(3)
(2)
(2)
10 + i10 10 = - V10 00 11 V10 = 11E x eit + c.c. .
h
h
(8.65)
(2)
A term in V12 20 has been omitted from this equation because its contribution has already
(2)
(2)
been included in Eq. (8.60). The final step in Eq. (8.65) follows because 00 = 11 .
(2)
Of course, we already have 11 from Eq. (8.61), so the only hurdle left to surmount is the
algebra. The final result is

4
|
|
2
2
10
E x2 E x
(8.66)
Pvia
state0 = N
2G + +
+
4h- 3
10
(2)10
10 (2)10
where
G=
+ 3
(10
)
+ 2
(10
) 10
+
2
10
(10
)
3.
(10
)
(8.67)
183
Notice that the first and last terms in Eq. (8.66) mirror the first and last terms in Eq. (8.60)
and the first and last diagrams in Fig. 8.7. On the other hand, the terms in G replace the
central four terms and diagrams. The total IDRI polarisation is therefore
P IDRI = Pstate2
+ Pstate0
(8.68)
where the two contributions come from Eqs (8.60) and (8.66), respectively. Notice that the
contribution from level 0 is invariably negative. The contribution from level 2 could be of
either sign although, for a multi-level system, the overall sign is almost invariably positive.
Refractive index almost always increases with intensity.
The conclusion to draw from this section depends on ones point of view. On the one
hand, perturbation theory clearly has some nasty surprises in store for the unwary; on the
other, it holds some fascinating secrets for the enthusiast.
Problems
(0)
8.1 Verify Eq. (8.25) for the three-level system of Fig. 8.3 in the case where a0 = 1, and
(0)
(0)
hence a1 = a2 = 0.
8.2 Show that the right-hand side of Eq. (8.57) is real. (This is quite easy once you know
what to look for but, if you do not spot the trick quickly, you could waste a lot of time.)
8.3 For a class 11 crystal such as KDP, Eq. (4.14) reads
Px (3 ) = 20 d14 (E y (1 )E z (2 ) + E z (1 )E y (2 )).
Why is there only one d coefficient in this equation when there are two distinct
coefficients in the analogous Eq. (8.39)?
Resonant effects
9.1 Introduction
Although resonant effects were largely ignored in Chapter 8, some of the machinery for
handling them was included at the start of the chapter. Damping terms were, for example,
introduced in Section 8.4.1, and complex frequencies were retained through much of the
discussion of Section 8.6.
Moving close to a resonance does not necessarily invalidate the perturbation theory
approach, as indeed the inclusion of damping in Eqs (8.30)(8.34) indicated. We will
examine the properties of those equations in the next section, and extend the discussion
to include Raman resonances in Section 9.3. However, the application of sufficiently strong
fields on fast time-scales certainly violates the weak perturbation assumption on which
the approach of Chapter 8 was based. In Section 9.5, we will explore the properties of
the density matrix equations in this coherent limit and, in Sections 9.6 and 9.7, we will
apply the results to two well-known coherent effects: self-induced transparency (SIT) and
electromagnetically-induced transparency (EIT).
9.2 Resonance in the linear susceptibility

The linear susceptibility of a multi-level system was shown in Chapter 8 (see Eq. 8.34) to be
(1)

20n
20n
N
N
() =
=
0 n n0
n0 + in0
0
n
(9.1)
where the definition n0 = n0 in0 has been used in the second step, and only the
dominant terms have been included. The tensor nature of the susceptibility has been ignored,
and the matrix elements 0n have been assumed to be real as before.
In the single-level case, Eq. (9.1) becomes
(1) () =
184
N 2
0
i
2 + 2
= () i ()
(9.2)
9.2 Resonance in the linear susceptibility
185
where = 10 , and other subscripts have been dropped. The real and imaginary parts
of () are

N 2
N 2

()
=
.
(9.3)
() =
0 2 + 2
0 2 + 2
Equation (2.6) can now be used to link the susceptibility to the angular wave number of the
field via the sequence

n
n 1
i

k=
1 + i =
(9.4)
1 2 =
2 i
c
c
2n
c
where the refractive index is n() = 1 + () and the approximation is valid if is
small. The negative imaginary part of k leads to absorption (via eikz ), and is the intensity
absorption coefficient given by
() =
N 2 g()
()
=
cn
c0 n
(9.5)
where g() = 1 /(2 + 2 ) is the normalised Lorentzian lineshape function.

The linear dispersion characteristics of the medium are evidently governed by (), and
the absorption profile by (). Graphs of these two parameters (normalised to (0)) are
presented in Fig. 9.1. Since n()
= 1 + 12 (), the graph of ()/ (0) (shown in solid
black) shows how the refractive index varies in the vicinity of the resonance; notice that the
abscissa scale runs from positive to negative (= 0 ) to ensure that the field frequency
increases to the right. The refractive index falls sharply in the immediate vicinity of the resonance (anomalous dispersion), but rises with frequency (normal dispersion) further out on
both wings. The Lorentzian frequency profile of the absorption coefficient is shown in grey.
Real systems are of course normally characterised by a large number of resonances,
not just one. The summation in Eq. (9.1) takes care of this, and the way in which two
neighbouring resonances run together is shown in Fig. 9.2.
(relative)
/
10
10
Fig. 9.1
Real part (bold) and imaginary part (grey) of the linear susceptibility near a resonance; see Eq. (9.3). Both graphs are
normalised to the peak value of the imaginary part.
Resonant effects
186
(relative)
/
20
20
Fig. 9.2
As Fig. 9.1, but for two adjacent resonances.
9.3 Raman resonances

Several two-photon resonant effects could be examined in detail, but stimulated Raman
scattering is particularly interesting and was of course discussed in Chapter 5. We therefore
focus on that case. We will not engage in the same level of detail as in Chapter 8, but
concentrate rather on the main features by including only the dominant path in the analysis.
A simplified energy level scheme for stimulated Raman scattering (SRS) is shown in
Fig. 9.3. As explained in Chapter 5, SRS involves beams at frequencies L and S (< L ),
where subscripts L and S designate the Laser and Stokes waves, respectively. When the
frequency difference (L S = D ) is close to the frequency separation 20 of level 2
and the ground state 0 (the Raman resonance frequency), gain is recorded at S , loss at L ,
and population is transferred from level 0 to level 2 at the same time. Level 2 could be a
vibrational state, a rotational state, or even an electronic state.
Under these circumstances, the methods developed in Chapter 8 lead to the hierarchy of
first-, second- and third-order density matrix elements
(1)
01 =
(2)
02 =
(3)
01 =
01 E L exp{iL t}
22 (10 L )
(9.6)
01 12 E L E S exp{i(L S )t}
)
42 (10 L )( 20
D
(9.7)
01 12 21 E L E S E S exp{iL t}
)( )
82 (10 L )( 20
D
10
L
(9.8)
where 20 = 20 i2 . These lead to a term in the polarisation at L of the form

2
01 12 21 10

PL = 2
)( ) ES EL .
4 (10 L )( 20
D
10
L
(9.9)
9.3 Raman resonances
187
2
20
Fig. 9.3
Real and (dominant) virtual levels for stimulated Raman scattering in a three-level system.
The focus of attention here is on the middle term in the denominator. Away from resonance
(|20 D | 2 ), the polarisation in Eq. (9.9) is a manifestation of the optical Kerr effect,
in which the refractive index of the wave at L is modified by the presence of the wave at
= i , and it follows that P i E
L, a
S . On resonance, on the other hand, 20
D
2
L
1
condition that we know from Chapter 2 leads to attenuation of the wave at L .
Clearly there must be an equation analogous to Eq. (9.9) for the Stokes polarisation PS . It
(3)
turns out that this is driven by the third-order density matrix element 21 and the associated
polarisation is
PS =
42 (
2
01 12 21 10

E L E S .
)(
)(
)
10
L
20
D
10
L
(9.10)
The interpretation of this equation is closely analogous to that of Eq. (9.9). Away from the
Raman resonance, this equation governs the change in the refractive index of the Stokes
wave in the presence of the laser field. On resonance, however, it implies that the Stokes
wave will grow; notice that the middle term in the denominator of Eq. (9.10) is the conjugate
of the corresponding term in Eq. (9.9), so PS +i E S which implies gain. Indeed, for
every photon lost at L , a photon is gained at S . But of course the energies do not balance,
because the photon energies are different. The energy difference is delivered to the medium,
as population is transferred from level 0 to level 2.
The analysis of this section has been based on the implicit assumption that the initial
(unperturbed) population of the medium is in the ground state 0. However, if the calculation
were reworked with the initial population in level 2, it would emerge that an anti-Stokes
wave at frequency AS = L + 20 would receive gain, accompanied by population
transfer from level 2 to level 0. But, as explained in Chapter 5, an anti-Stokes wave can
also be generated without initial population in level 2 through the frequency combination
1 See the discussion after Eq. (2.8).
Resonant effects
188
L S + L , a process that is mediated by the polarisation

PAS =
01 12 21 10
E L E S E L .
42 (10 L )( 20 D )(10 AS )
(9.11)
Notice that when the basic spatial dependence of the fields is made explicit, we have
E L E S E L = E L E S E L exp{i(2kL kS ).r}
(9.12)
and exp{i(2kL kS ).r} needs to match exp{i(kAS ).r} if the interaction is to proceed
efficiently. This anti-Stokes generation mechanism is therefore subject to a phase-matching
condition, in contrast to the Raman generation process represented by Eqs (9.9) and (9.10).
We remark that density matrix formulae have been quoted in this section because they
enable stochastic damping terms to be included. However, the same results (albeit without
the damping terms) can be obtained more easily using the state vector approach, where the
algebra is more straightforward. The advantages and disadvantages of the two approaches
were discussed in detail in Chapter 8.
9.4 Parametric and non-parametric processes

The process described in Eqs (9.11) and (9.12) is sometimes called parametric anti-Stokes
Raman generation to distinguish it from the anti-Stokes mechanism mentioned earlier which
was based on the initial population of level 2 and which (like Stokes wave generation) counts
as non-parametric. Although widely used, the meaning of the word parametricis obscure,
and it would be as well to avoid the word altogether were it not already entrenched in the
terminology of nonlinear optics.
Three questions are relevant in attempting to make the distinction between parametric
and non-parametric processes:
1. Is the energy in the participating optical fields conserved?
2. Does the medium return to its initial energy level at the end of the interaction?
3. Is the process subject to a phase-matching condition?
If the answer to all three questions is Yes, the process can confidently be classed as parametric. An obvious example is second harmonic generation; two fundamental photons combine
to form one harmonic photon, the state of the medium is not changed, and there is certainly
a phase-matching condition to satisfy. On the other hand, if the answer to all the questions
is No, the process is almost certainly non-parametric. The generation of a Stokes wave in
stimulated Raman scattering seems to be a good example; a laser photon is lost, a Stokes
photon of lower energy is gained, and the nonlinear medium is lifted to an excited state.
Phase matching is automatically satisfied.
Unfortunately, things are not in fact so straightforward, and there are a number of complications and pitfalls in this area. How, for example, should one classify the optical Kerr
effect? Far from the Raman resonance, Eqs (9.9) and (9.10) merely govern refractive index
changes. On resonance, however, the same equations govern stimulated Raman scattering
189
9.5 Coherent effects
(SRS), which we have just argued is non-parametric. Admittedly, no phase-matching condition is involved, but energy in the EM fields is certainly conserved. However, as Butcher
and Cotter [96] point out, when SRS occurs in a solid, the phonon can be treated as a wave
that is subject to a phase-matching condition, as indeed has been pointed out in Chapter 5
with regard to Raman waves in general. And one could of course argue that there is actually
a phase-matching condition in SRS, albeit one that is automatically satisfied!
Again, how should one regard a process that is subject to a phase-matching condition,
but has an intermediate resonance? Parametric anti-Stokes Raman scattering governed
by Eq. (9.11) above is a case in point. As to question 2, which is sometimes taken to
be definitive, the problem here is that different viewpoints of the same process can give
different answers. After all, the matrix element sequences in Eqs (9.9) and (9.10) start and
end at level 0, but that does not make SRS a parametric process.
The word parametric is clearly of doubtful value and should be used with caution.

9.5.1 The optical Bloch equations
We now move on to study coherent effects in nonlinear optics where the interactions take
place on time-scales that are short compared to the relevant damping times. Coherent processes cannot normally be handled using perturbation theory, so a new set of mathematical
tools must be developed to describe them.
We start by treating a two-level atomic system excited by a near-resonant optical pulse
given by E = A(t) cos t. To simplify the analysis, we will take the pulse envelope A(t) to
be real, and we will also assume that it is substantially shorter than the relaxation times of
the medium so that damping effects can be ignored. Under these conditions, the off-diagonal
and diagonal terms in Eq. (8.15) read
i
V (11 00 )

iV
=
(10 01 )
10 = i10 10 +
(9.13)
00 = 11
(9.14)
where V = A cos t, and we have assumed = 10 = 01 .

These two equations have several important features. Firstly, the upper state population
has been included; notice that the final term on the right-hand side involves the population
difference between the two levels. Secondly, the damping terms of Eqs (8.18) and (8.19)
have not been included, so Eqs (9.13) and (9.14) will provide an accurate representation of
a real system only on very fast time-scales where damping plays an insignificant role.
The intrinsic time dependence of 10 governed by the first term on the right-hand side
of Eq. (9.13) could, of course, be removed by using the representation based on 10 =
Resonant effects
190
10 exp{i10 t} from Eq. (8.16). In this case, Eqs (9.13) and (9.14) become
i
V (11 00 )
10

i

= 11 = V01
10 V10
01
10 =
(9.15)
00
(9.16)
= V exp{i t} and
where V10
10
10
jj = jj . If we now substitute V10 = 10 A cos t,
etc., assume that = 10 = 01 , and neglect rapidly varying terms (in what is known as
the rotating wave approximation (RWA)), it follows that
A
(11 00 ) exp{i(10 )t}
2
iA
= 11 =
(10 exp{i(10 )t} 01 exp{i(10 )t}) .
2
10 = i
(9.17)
00
(9.18)
Actually, it is more common to make a subtly different transformation of Eqs (9.13) and
(9.14), one that is based not on the transition frequency between levels 0 and 1, but rather
= exp{it}, the equations
on the frequency of the applied field. If we duly define 10
10
read
)
iA(t)(11
00
2

iA(t) it

= 11
=
e + eit 01 eit 10 eit
2

iA(t)
10 01
2

= i10
10

00
(9.19)
(9.20)
where = 10 , and the RWA has been invoked to drop rapidly varying terms in the
final form of Eq. (9.20).
It is now a simple matter to obtain the celebrated optical Bloch equations. We define
} +
u = 2Re{10
10
01
} i( )
v = 2Im{10
10
01
(9.21)

w = 11
11
00
00
and Eqs (9.19) and (9.20) can now be written

u = v
wA(t)
v = u +

vA(t)
w =
.
(9.22)
The sign conventions have been chosen to correspond to those of Allen and Eberly [101].
191
9.5.2 The Bloch vector

A powerful geometrical interpretation of the atomic dynamics now presents itself. For if we
define a vector r = (u, v, w), Eq. (9.22) can be written [14, 101, 102]
r = r
(9.23)
where = (A(t)/, 0, ). Equations (9.22)(9.23) are known as the optical Bloch
equations, and r as the optical Bloch vector. Since r is perpendicular to r according to
Eq. (9.23), the magnitude of the Bloch vector is clearly constant, a property that can also
be seen from Eqs (9.22) where uu + v v + ww is evidently zero. From Eqs (9.21), it is
also clear that the vector has unit magnitude, so u2 + v 2 + w2 = 1, and the vector moves
on the surface of a sphere of unit radius, which is known as the Bloch sphere. However,
it must be remembered that the present analysis neglects damping and, when stochastic effects are taken into account, the components of the Bloch vector are subject to
relaxation.
A diagram of the Bloch sphere is shown in Fig. 9.4, where the 1, 2 and 3 axes correspond to
the respective components of r (i.e. u = r1 , v = r2 , w = r3 ). The first point to appreciate
is that, when the Bloch vector points to the south pole of the sphere (w = 1), the system is
in the ground state while, at the north pole (w = +1), it is in the excited state; this should be
clear from the third of Eqs (9.21). A useful connection can also be established between the
transverse components (u and v) of r, and the polarisation of the medium. If we combine
3
w = +1
w
v
u
1
w = 1
Fig. 9.4
The Bloch vector r = (u, v, w) showing the Bloch sphere and its north and south poles.
Resonant effects
192
= exp{it}, we obtain
the definition of the polarisation in Eq. (8.21) with 10
10
P = 12 N (u + iv)eit + c.c.
(9.24)
Hence, when w = 1 and u = v = 0, the polarisation is zero. However, at all other points
on the sphere, the system is in a superposition state, and the polarisation is non-zero, while
maximum polarisation occurs on the equator, where the magnitude of P is N .
Stochastic damping of the off-diagonal elements of the density matrix will ultimately
cause u and v, and hence the polarisation P , to relax to zero. But we are, of course,
assuming the pulse envelope A(t) to be much shorter than the relevant relaxation times
precisely so that these effects can be ignored. However, another source of dephasing arises
in the presence of inhomogeneous line broadening, when there is a distribution of detuning
among the different two-level atoms/molecules. Under these circumstances, Eq. (9.24)
generalises to
P = 12 N (u + i v)e
it + c.c.
(9.25)
where

u(z,
t) =
u(, z, t)gi ()d

v(z,
t) =
v(, z, t)gi ()d.
(9.26)
The parameters u and v are averages over the inhomogeneous lineshape function gi (),
which is normalised so that gi ()d = 1. Inhomogeneous broadening plays an important role in several coherent dynamic effects including self-induced transparency, which we
turn to next.
9.6 Self-induced transparency

9.6.1 Preview
The process of self-induced transparency (SIT) was first identified by McCall and Hahn in
1967 [14]. They showed that the equations derived in the previous section, in combination
with Maxwells equations, allow the propagation of a resonant pulse through a two-level
medium without loss, provided the pulse is short compared to the damping times. In the
simplest terms, the front of the 2-pulse swings the Bloch vector from the south to the
north pole of the Bloch sphere. Energy is removed from the pulse in this process, and
transferred to the medium, which becomes (temporarily) fully excited. But the pulse drives
the vector on over the north pole, returning it to the south pole down the opposite side of
the sphere. The population falls back to the ground state as the energy is returned from the
medium to the field.
193
The detailed theory of SIT is complicated. Whole books have been written on coherenttransient phenomena, processes that include SIT itself, and related effects such as photon
echoes; see [101]. We shall restrict ourselves to a highly simplified version of the theory,
in which the tendency of the mathematics to obscure the physics will hopefully be avoided
as far as possible.
9.6.2 Motion of the Bloch vector: the Rabi frequency

Consider first the motion of the Bloch vector in the presence of a time-independent field
that is tuned to exact resonance. In this case, = (A/, 0, 0), so r rotates about the
1-axis at an angular frequency A/, which is known as the Rabi frequency ;R . If the
medium is initially in the ground state, the vector starts at the south pole of the sphere and
rotates in the 23 plane, cycling backwards and forwards between the south and north poles
at the Rabi frequency, with the polarisation oscillating correspondingly. The motion of the
Bloch vector in this simple case is easily shown to be described by the equations
u=0
(9.27)
v = sin ;R t
w = cos ;R t.
Figure 9.5 illustrates the motion, if the angle is interpreted as ;R t; motion of this type
is sometimes called Rabi flopping. From Eq. (9.24), it is easy to show that the associated
polarisation is
P = N sin ;R t sin t.
(9.28)
3
w = +1
w = 1
Fig. 9.5
Motion of the Bloch vector in the v-w plane for a pulse of area = 5/4. The diagram also illustrates the motion
dened in Eq. (9.27) if is interpreted as ;R t.
Resonant effects
194
A much more elaborate calculation is needed to solve the problem when the system is
detuned from resonance. If, as before, w = 1 initially, the solution can be shown to
be [101]
u = ab(1 cos ;R t)
v = b sin ;R t

w = a 2 + b2 cos ;R t
(9.29)
where a = / ;R , b = ;R / ;R and ;R is the Rabi frequency, modified to take detuning
into account, and given by

(9.30)
;R = ;2R + 2 .
It is easy to show (see Problem 9.4) that Eqs (9.29) describe motion in a plane that passes
through the south pole of the Bloch sphere and is slanted from the 3-axis at an angle whose
tangent is / ;R .
9.6.3 Pulse area and the area theorem

We now consider the more general case of pulsed excitation, where the field is given by
E(t) = A(t)cos t and the pulse envelope A(t) (assumed real) is short compared to the
damping times. As before, we assume that the Bloch vector is initially at w = 1, and
the detuning is zero ( = 0); this, of course, ensures that u = 0 at all times, and restricts
the motion to the v-w plane. It is easy to show that the solution in this case is
u=0
(9.31)
v = sin
w = cos
where
(t) =
t
A(t )dt .
(9.32)
The final position of the Bloch vector is determined by the angle
= () =
A(t )dt
(9.33)
which is commonly referred to as the pulse area. Notice that is related to the time integral
of the field profile A(t), not to the that of A2 (t), which is related to the pulse energy. It is
therefore possible for the pulse energy to rise while the pulse area remains unchanged, as
we shall shortly discover.
The swing of the Bloch vector in the v-w plane for pulse area of = 5/4 is shown in
Fig. 9.5. As noted after Eq. (2.8), when the polarisation has a /2 phase lag with respect to
energy flows from the field to the medium. This is the
the electric field (so that P i E),
195
case for v < 0 (see Eq. 9.24), which applies in the initial part of the process in Fig. 9.5 where
w is rising. But as the Bloch vector passes the north pole (r = (0, 0, +1)) of the sphere,
v goes positive, the phase of the polarisation changes from a lag to a lead, the direction of
energy flow reverses, and w begins to fall.
As in the case of constant field treated earlier, the path of the Bloch vector is confined to
the v-w plane only for perfect tuning. When the detuning = 0, the vector traces a looped
path on the surface of the sphere, starting at the south pole and returning to it, provided
is sufficiently large. The motion is no longer in a plane, as it was for a time-independent
field, except in the case of perfect tuning when the vector passes through the north pole
(w = +1). Otherwise, the path veers away from the pole, reaching a maximum value of
w (< +1), and then retreating south again. For very small detuning, the furthest north
point will be close to the north pole; for large detuning, the vector will never leave the
southern hemisphere, and for very large detuning, it will never get far from the south pole.
The question that must now be addressed is: how does evolve as the resonant pulse propagates? The answer is contained in the celebrated area theoremof McCall and Hahn, which
applies in the presence of inhomogeneous line broadening, The area theorem states that,
irrespective of the pulse shape, the pulse area evolves with propagation distance according to

d
N 0 2 gi (0)

=
sin = sin (absorbing case)
(9.34)
dz
2cn0
2
where the final step defines
=
N 0 2 gi (0)
.
c0 n
(9.35)
Here, gi (0) is the inhomogeneous lineshape function at line centre, and n is the refractive
index. The signs in Eqs (9.34) and (9.35) are appropriate for an absorbing medium, which
is the case we have been treating throughout this section.
We now consider the implications of the area theorem. It is evident from Eq. (9.34) that
d /dz = 0 when = m (integral m), but it is easy to see that the solution is stable for even
m, and unstable for odd m. For example, if is close to (m = 1), the sign of d /dz ensures
that runs away from , whereas, near m = 2, converges on 2 from either direction.
Pulses with = 2 (known as 2 pulses) are not only stable in pulse area but, because the
medium is returned to the lower state at the end of the process, the pulse energy is constant
too. This suggests that 2 pulses with stationary pulse profiles might exist, a conjecture
that turns out to be correct, and will be examined in a little more depth in a moment.
The graphical solutions to Eq. (9.34) shown in Fig. 9.6 provide full details of the evolution of the pulse area. The abscissa is in units of z, so the characteristic length-scale is
determined by ( )1 . For very weak fields, sin , and the solution to Eq. (9.34) is

(z) = (0)e z/2 . Now, the expressions for in Eq. (9.35) and in Eq. (9.5) are identical
apart from the nature of the respective lineshape functions, so it is tempting to assume that

the pulse energy decays as (e z/2 )2 = e z in this weak-field limit. However, extreme
caution is necessary in this context because, if the pulse envelope develops a negative field
lobe, the area can fall to zero while energy still remains. A detailed analysis of small-area
pulses has been given by Crisp [103].
Resonant effects
196
area
5
4
3
2
1
<<< amplifying
absorbing >>>
0
Distance (units of z)
Fig. 9.6
Evolution of pulse area with propagation distance according to Eq. (9.34). In the case of an amplier (Eq. 9.36), the
evolution is from right to left.
An intriguing feature of Eq. (9.35) is the presence of gi (0) in the numerator. This implies
that in the absence of inhomogeneous broadening (IHB), because gi (0) then
becomes a delta function. This in turn means that the distance scale of Fig. 9.6 shrinks to zero,
and the pulse area reaches its stable limit immediately. What is the physical interpretation of
this surprising feature? According to Eqs (9.24) and (9.31), a pulse of initial area i leaves
the medium with a polarisation P = 12 iN sin i eit + c.c., and this optical antenna will
continue to radiate until it is either washed out by IHB, or all the stored energy is used
up. So, in the absence of IHB, the radiation will continue until the final area f becomes
m (m even) and w = 1. If i < , the radiation adds a negative lobe to the rear of the
pulse, making f = 0 while, if < i < 2, a positive lobe is added and f = 2. The
extension to higher values of i is straightforward.
So far, the discussion has been based on the assumption that the medium is in the lower
state (wi = 1) before the arrival of the pulse. In the case of an amplifying medium,
wi = +1, and the area theorem changes to
d

= + sin
dz
2
(amplifying case).
(9.36)
The sign reversal means that the stable values of are now m with m odd; Fig. 9.6 can still
be used, provided it is read from right to left since, in effect, the sign change means z z.
In the amplifying case, an initial pulse with 0 < i < 2 will gravitate towards a so-called
-pulse with = . However, a 180 swing takes the Bloch vector for perfectly tuned
atoms with = 0 from wi = +1 to wf = 1 so, even though the area has stabilised,
the energy will continually increase as the pulse picks up energy from the medium. And the
only way this combination of circumstances can occur is if the pulse continually narrows.
197
9.6.4 Outline proof of the area theorem

The proof of the area theorem is very involved, and only a brief sketch will be attempted
here. All parameters must now be treated as functions of space and time, and A(t) is replaced
t) which is assumed real. The field equation is given by
by E(z,
E
n E
NL
+
= i
P
dz
c dt
2c0 n
(9.37)
which is a simplified version of Eq. (7.2). Equation (9.37) leads, with the help of Eqs (9.25)
and (9.26), to
E
N 0
=
dz
2c0 n

v(, z, t)g()d
1 E
.
c dt
(9.38)
From the definition of pulse area, and with the help of the previous equation, we have
d
= Lim
t
dz
= Lim
t
dt
t
t )
E(z,
z
dt
N 0
2c0 n
1 E(z, t )
v(, z, t )g()d
.
c dt
(9.39)
) E(z,
)),
The final term in this equation, when integrated, involves the factor (E(z,
which is zero for a pulse. A comparison of the remaining term with Eq. (9.34) shows that
the area theorem is proved if it can be demonstrated that
t
Lim
dt
d v(, z, t )g() = g(0) sin .
(9.40)
It is in this final step that most of the mathematical difficulty lies, and a lengthy analysis
is necessary to verify the identity [101]. The appearance of g(0) on the right-hand side
underlines the fact that the double integral is dominated by the part of the lineshape in the
immediate vicinity of = 0. In view of the second of Eqs (9.31), the result of Eq. (9.40)
seems quite reasonable.
9.6.5 The steady-state solution

The fact that a 2 pulse propagates without change in either area or energy suggests that a
steady-state solution might exist. This is indeed the case, although the proof leading to the
solution is even more unfriendly than the one for the area theorem. The result is

t) = 2 sech t z/V
E(z,
(9.41)
Resonant effects
198
where is a measure of the pulse width, and can take any value that keeps the pulse within
the coherent regime. The pulse envelope travels at speed V given by
1
n
=
V
c
2
2 gi (0)

gi ( )
d .
1 + ( )2
(9.42)
This formula can be approximated when the pulse duration is either very short or very
long compared to the inverse linewidth; see Problem 9.3.
9.7 Electromagnetically-induced transparency and slow light

To end this chapter, we apply the density matrix machinery to the study of another strongfield effect known as electromagnetically-induced transparency (EIT), which was first
studied by Boller et al. in 1991 [104,15]. The treatment of EIT presented here is similar
to that in Section III of Fleischhauer et al. [16], where research on EIT and related topics
such as slow light is reviewed. The energy level scheme shown in Fig. 9.7 is the same as in
[16]; much of the notation is the same too, although the numbering of the energy levels is
different.
Under optimal EIT conditions, absorption at line centre of a transition from the ground
state (0) to an excited state (1) is eliminated in the presence of a strong resonant coupling
field between state 1 and a third state 2. At the same time, a narrow region of strong
normal dispersion appears at the centre of the resonance in which the group velocity is
very low; this leads to the effect known as slow light. Moreover, while absorption from the
ground state is artificially removed by EIT, stimulated emission to the ground state turns
1
p
coupling field
probe field
2
2
Fig. 9.7
Three-level system used for analysing electromagnetically-induced transparency, showing the three detunings
dened at the beginning of Section 9.7.
199
out to be unaffected. This opens the way to a further phenomenon known as lasing without
inversion [17].
The excited state 2 has the same parity as the ground state, and both are coupled to state 1,
which has the opposite parity. Of the two EM waves involved in the process, the strong
coupling field at c is tuned close to the spacing 12 between state 1 and state 2, while
a weak probe field at p scans the properties of the 01 transition in the presence of c .
We introduce detuning parameters p = 10 p and c = 12 c for the respective
probe and coupling fields. The two-photon detuning 2 = p c = 20 (p c )
also turns out to be useful. The coupling field is shown twice (by double arrows) in Fig. 9.7,
in order to highlight all three detuning parameters.
Nine density matrix equations can be written down for this three-level system, three for
the diagonal elements and six for the off-diagonal elements (of which three are conjugates
of the other three). It turns out, however, that the key features of EIT can be understood by
considering the equations for just two off-diagonal elements, namely2

01 E p
1
01 = 10 01 2 i ;Rc 02 exp{ic t} +
(9.43)
(00 11 ) exp{ip t}

02 = 20 02 12 i;Rc 01 exp{ic t}.
(9.44)
In these equations, ;Rc = 12 Ac / is the Rabi frequency associated with the coupling
field, and E p is the probe field; damping terms have been included. An equation for 12
can easily be written down too, but this element plays no direct part in the action. As for
the diagonal elements, it turns out to be sufficient simply to assume that 00
= 1 and
11 = 22 = 0.
The steady-state solutions of the equations are readily obtained once one spots that
01 varies as exp{ip t} and 02 as exp{i2 t}. This enables the replacements 01
ip 01 and 02 i2 02 to be made, and one then quickly obtains the solutions
02 =
;Rc exp{ic t}
01
2(2 + i20 )
(9.45)
01 =
01 E p exp{ip t}
;Rc exp{ic t}
02 +
.
2(p + i10 )
2(p + i10 )
(9.46)
Combining these equations yields

01 = exp{ip t}
01 E p (2 + i20 )

2D
(9.47)
where
D=
1
2 ;Rc
2
(p + i10 )(2 + i20 ).
(9.48)
The structure of the 01 transition can now be explored by studying the susceptibility at
the probe frequency. From Eq. (8.21), it can be shown that the dominant term in the
2 The damping constants defined in Eqs (9.43) and (9.44) are half the corresponding factors in [16].
Resonant effects
200
polarisation is
N
Pp = N 01 10 exp{i(10 p )t} =
|01 |2 (p + i20 )

E p .

D
(9.49)
Writing the susceptibility in the form = i as in Section 9.2, we obtain

=
N |01 |2 2 (2 p ( 12 ;Rc )2 ) + (20 )2 p

0
|D|2
(9.50)
=
N |01 |2 22 10 + 20 (( 12 ;Rc )2 + 10 20 )

.
0
|D|2
(9.51)
Despite the relative simplicity of Eqs (9.43) and (9.44) with which we started, the structure
of the susceptibility is evidently quite complicated. However, careful examination of the
formulae yields a rich reward, because EIT, slow light, and indeed the AutlerTownes
splitting too, are all hidden inside Eqs (9.50) and (9.51).
We will look at the special case where 20 is negligible, and the coupling field is perfectly
tuned to the 12 transition, so that c = 0 and 2 = p . Equations (9.50) and (9.51) then
reduce to
2
3
1
2
2
N |01 |2 10 p (p ( 2 ;Rc ) )

(9.52)
=
0 10
|D|2
2 2 23
N |01 |2 p 10

=
(9.53)
0 10
|D|2
2.
where |D|2 = (( 12 ;Rc )2 2p )2 + 2p 10
In Figs 9.8 and 9.9, the square-bracketed terms in Eqs (9.52) and (9.53) are plotted as a
function of the normalised detuning p /10 for two different values of ;Rc . Notice that the
abscissa scale decreases to the right, which ensures that the probe frequency p = 10 p
increases in that direction. Positive values of represent absorption.
Figure 9.8 shows the behaviour when ;Rc = 4.010 , with graphs for ;Rc = 0 (i.e.
zero coupling field) shown dotted for comparison. Several important features are apparent
from the figure. Firstly, the absorption profile, which consists of a single maximum when
;Rc = 0 (dotted line), is split into two in the presence of the coupling field. This effect is
known at the AutlerTownes splitting. Inspection of Eqs (9.52) and (9.53) reveals that the
peak separation corresponds to the Rabi frequency, and this is confirmed by the figure.
Secondly, the absorption near exact resonance is very small, and careful inspection reveals that it is actually zero at p = 0; see Eq. (9.53). This is the basis
of electromagnetically-induced transparency (EIT). Exactly on resonance, the medium
becomes perfectly transparent!
Thirdly, there are now two regions of anomalous dispersion, one associated with each
component of the double peak. But between them lies a region of normal dispersion so, at
the point of exact resonance (p = 0), the type of dispersion has changed.
What happens when the strength of the coupling field is reduced is no less remarkable. In
Fig. 9.9, where ;Rc = 0.610 , the AutlerTownes splitting has narrowed, but the feature at
201
0.5
0.5
4
detuning
Fig. 9.8
Real part (top) and imaginary part (bottom) of the susceptibility of Eqs (9.52) and (9.53) as a function of normalised
probe detuning for c = 0 and ;Rc = 410 , showing the AutlerTownes splitting, strong normal dispersion near
the centre of the resonance, and transparency at exact resonance. The results for zero coupling eld are shown dotted.
0.5
0.5
2
Fig. 9.9
0
detuning
As Fig. 9.8 but for ;Rc = 0.610 .
exact resonance still exists, and has become notably sharper. And it continues to sharpen as
the coupling field is further reduced. Indeed, under these idealised conditions, the frequency
dependence of the real part at p = 0 increases without limit, tending to infinity as ;Rc 0
at which point the EIT features naturally vanish. From Section 6.2,
vgroup =
dn
n+
d
(9.54)
202
Resonant effects
and, since dn/d

= 12 d /d, EIT is associated with very low values of the group velocity.
This is the phenomenon of slow light.
While the existence of transparency at exact resonance is fascinating from a theoretical
perspective, it is not in itself of particular importance experimentally. After all, if one wants
to avoid absorption, one can easily do so by moving away from line centre. The crucial
point here is that the third-order nonlinear coefficient (3) can be shown to be subject
to constructive rather than destructive interference at the resonance, and so resonantly
enhanced nonlinear optical interactions can take place at the centre of the transparency
window. A number of four-wave mixing schemes involving the coherent linkage of states
1 and 2 have been considered.
Numerous experiments demonstrating the dramatic reduction in the optical group velocity
at the centre of the EIT resonance have been conducted [18,19]. In one example, based on
ultracold sodium atoms in a Bose condensate, light was slowed to 17 ms1 , or roughly 60
kmhr1 [105], which means that c/vgroup
= 2 107 . It is easy to see that this massive
speed reduction will be accompanied by a corresponding spatial compression of the pulse
envelope by the same factor. After all, a 1 s pulse is spread over 300 m at the speed of light
in vacuum, but at 17 ms1 its extent will be only 17 m. However, it is only the envelope
that is slowed down and compressed; the optical carrier wave continues to travel at the
normal speed, and its wavelength is unaltered. The strength of the electric field and hence
the photon flux are also not significantly changed by the slowing down, and the consequence
is that the number of photons in the pulse will also be reduced by the compression factor.
This raises a serious question about where the energy has gone, and it turns out that almost
all the incident energy is temporarily stored in the medium.
It is also possible to bring a light pulse to a complete halt by switching off the coupling
field while the pulse is confined within the interaction region [106]. And, when the coupling
field is switched back on, the pulse starts moving again. This has potential applications in
optical data processing and data storage.
Problems
9.1 Convince yourself that the approximation in Eq. (9.4) is correct.
9.2 Use Eq. (2.8) in conjunction with Eqs (9.9)(9.10) to verify that the energies gained
at S and lost at L in stimulated Raman scattering are in the ratio of the respective
photon energies.
9.3 When the steady-state pulse duration in self-induced transparency is very small
compared to the inverse inhomogeneous linewidth gi (0)1 , show that Eq. (9.42) can
1
1
n 1
n 2
be approximated to
. Show similarly that
=
=
2 in the
c
2 gi (0)
V
c
V
opposite limit ( gi (0)1 ).
9.4 Show that Eq. (9.29) represents motion in a plane that passes through w = 1 and is
slanted away from the 3-axis at an angle of tan1 {/ ;R }.
203
9.5 Show that the refractive index gradient at the centre of the EIT resonance in Fig. 9.9 is
given by
2N |01 |2
dn
.
=
d
0 (;Rc )2
9.6 Use a PC to explore the properties of Eqs (9.52) and (9.53) in a more general case
than that considered in this chapter. For example, how are Figs 9.8 and 9.9 modified
when 20 = 0? See if you can discover the Raman resonance, when c = 0 and 2 is
scanned through zero.
10
High harmonic generation

The treatment of frequency mixing in the early chapters of this book was based on the
assumption that the applied EM fields were weak compared to the internal fields within the
nonlinear media. This enabled the polarisation to be expanded as a power series in the field,
and perturbation theory to be used to calculate the nonlinear coefficients.
In this chapter, we consider the generation of optical harmonics in the strong-field regime,
where the fields are sufficiently intense to ionise the atoms in a gaseous medium, and the
electrons released then move freely in the field, at least to a good approximation. As we
shall discover, an enormous number of harmonics can be generated in these circumstances,
extending across a broad spectral plateau that extends deep into the soft X-ray region.
High harmonic generation (HHG) therefore creates a table-top source of coherent X-rays,
which has already been applied in lensless diffraction imaging [107]. Moreover, if the huge
spectral bandwidth is suitably organised, HHG enables pulses as short as 100 attoseconds1
or even less to be generated [108]. These can then serve as diagnostic tools on unprecedented
time-scales. They have already been used to probe proton dynamics in molecules with a
100-as time resolution [109], and to measure the time delay of electron emission in the
photoelectric effect for the first time [110]. Other applications are described in [111].
Experiments on third harmonic generation in the noble gases in the late 1960s involved
laser intensities of around 1 GW cm2 , far below those employed in current HHG experiments, which are typically five or six orders of magnitude higher. The early experiments
[22] were definitely in the perturbative regime, and they were performed under nonphase-matched conditions in an extended nonlinear medium. Conversion efficiencies were
minuscule, at around 1 in 1014 .
Experimental results on harmonic generation in gases and metal vapours in the 1970s
and early 1980s continued to exhibit the typical characteristics of the perturbative regime,
namely that the conversion efficiency for successive harmonics dropped off rapidly.2 However, by the late 1980s, much higher laser intensities were being deployed, and some
remarkable experimental results on the noble gases [113,114] suggested that a new strongfield regime was being entered. For lower harmonics, the conversion efficiency still fell
away as before but, beyond a certain point (typically around the seventh to the ninth
1 1 attosecond (1 as) = 1018 s.

2 See [112] for citations to this work.
204
205
harmonic), the development of the spectral plateau mentioned earlier was clearly seen.
Once on the plateau, the efficiency remained essentially constant up to a fairly well-defined
high-frequency cut-off. The plateau could be extended and the cut-off pushed further into
the UV by raising the laser intensity, although a saturation intensity existed beyond which
further extension was not possible.
By the early 1990s, harmonic orders well into the hundreds had been recorded in Ne
[115,116] and, by the end of the decade, orders around 300 were reported [117]. At the time
of writing, the figure is over 1000 [118] which, with a fundamental wavelength of 800 nm,
corresponds to a harmonic wavelength of around 0.8 nm at the spectral extremity, which is
deep in the soft X-ray part of the spectrum beyond the water window.
Thin gas jets are normally used in HHG work, as this gets round at least some of the
problems associated with focused beams in normally dispersive media (see Section 2.5).
Conversion efficiencies are many orders of magnitude higher than in the very early work,
but are still small at around 1 in 106 . Excitation intensities are typically around 1014 1015
W cm2 for Nd:YAG and Ti:sapphire lasers or perhaps a little higher for KrF excitation at
249 nm.3 These are the levels at which ionisation of the atoms becomes significant, and
this means that an entirely new set of concepts is needed to model the HHG process.
Highly sophisticated theoretical techniques are needed to do full justice to the HHG
problem, and some aspects lie at the current frontier of knowledge. However, a general
appreciation of how high harmonics are produced in the strong-field regime can be gained
by using the three-step model, the foundations of which were laid in [119,120]. The process
is divided into the following stages:
(1) Tunnel-ionisation in which an electron is removed from an atom by the field.
(2) The subsequent motion of the ionised electron in the oscillating EM field, in the course
of which the electron acquires significant kinetic energy.
(3) The recollision of the energised electron with the parent atom, in which the accumulated energy is emitted in the form of radiation. The frequency of the emission
corresponds to the kinetic energy of the electron, plus the energy of ionisation.
The schematic diagram of Fig. 10.1 illustrates the concept behind stage 1. It shows a
one-dimensional Coulomb potential (frame b) tilted by a strong electric field, which is
positive on the left (frame a) and negative on the right (frame c). In an oscillating field,
the slope naturally changes with time, so a movie would show a see-saw motion with the
potential sloping to the left and to the right on successive half-cycles. The key point is that
the potential barrier is lowered on each side in turn, increasing the probability of ionisation
in the favoured direction, the more so as the amplitude of the field is increased.
Many of the main features of the second stage (the motion of the ionised electron in the
field) can be understood within the simple classical model outlined in the next section. It is
fairly easy to understand that linearly polarised light is a prerequisite if the electron is to be
returned accurately to its starting point in the parent atom. We already know from Chapter 5
that the effective third harmonic generation coefficient is zero for circularly polarised light,
so this feature of HHG should not come as a complete surprise. However, the criterion on
3 Intensities are traditionally quoted in W cm2 in the research literature.
206
(b)
(c)
potential energy
(a)
Fig. 10.1
Schematic diagram showing a Coulomb potential (frame b) tilted by a positive electric eld (frame a) and a negative
electric eld (frame c). Notice the lowering of the potential barrier on the left in frame a, and the right in frame c.
the polarisation is more stringent in the HHG case; indeed it turns out that any significant
ellipticity will cause the electron to move away from the atom along a curved path which is
highly unlikely ever to return to the starting point.4 The need to employ linearly polarised
light has the benefit that a meaningful quantum mechanical treatment based on the timedependent Schrdinger equation in one dimension can be used to trace the evolution of the
electron wave packet in the interval between ionisation and recollision. We shall, however,
confine ourselves to a classical approach in this chapter.
The simplest statement that can be made about the recollision event is that the electron
returns to its bound state, and the energy of the emitted photon corresponds to the kinetic
energy acquired by the electron during its journey in the continuum, plus the energy of
ionisation. Quantum mechanically, the optical antenna driving the emission is formed in
the interference between the electron wave packet and the bound-state wave function of
the atom. The cut-off in the spectrum is determined by the maximum kinetic energy that
the electron can acquire, and a simple formula for this is easily obtained from the classical
model that we now describe.

10.2.1 Electron trajectories
The treatment of HHG that follows is essentially classical, although Plancks constant is
used to convert recollision energy to emission frequency, and a formula derived from a more
sophisticated model is used for a weighting factor (see Eq. 10.11 below). It is remarkable
how far one can progress with the HHG problem on this very simplistic basis. The reason is
4 If the electron and the ion are treated classically as point particles, even the slightest ellipticity in the field
polarisation precludes a recollision; see Problem 10.2. Quantum mechanically, however, the electron wave
packet and the ion have finite size, which reduces the strength of the recollision while making it more likely to
occur.
207
that classical physics does a relatively good job at handling the motion of the ionised electron
in the continuum, which turns out to be the critical stage of the process. The conclusion
is confirmed by quantum mechanical results in the strong-field approximation [121]. The
excursions of the electron are also quite large (see Problem 10.4), so it is an acceptable
approximation to neglect the influence of the parent ion.
We assume that the HHG process is excited by an optical field, linearly polarised in the
x-direction, and given byE(t) = A(t) cos t, where A(t) is in general the time-dependent
envelope of an optical pulse. The equation of motion of the electron is
mx = eA(t) cos t
(10.1)
where e is the elementary charge and m the electron mass. If we ignore the time variation
of A for the moment and assume the field is monochromatic, simple integration yields the
electron velocity as
x = xdrift
eA sin t
m
(10.2)
where the constant of integration is the cycle-averaged velocity. Setting xdrift = 0 yields
a formula for the quiver energy (or ponderomotive energy) of the electron in the field,
namely
UP = 12 m(x 2 ) =
(eA)2 2
(eA)2
sin
t
=
.
2m2
4m2
(10.3)
Consider now the situation of an electron immediately after ionisation. If the electron is
released from the atom at time t0 , and its initial velocity is assumed to be zero, the solution
to Eq. (10.1) is

4UP
x = V
(10.4)
m
where
V (t) = (sin t0 sin t) .
(10.5)
The nature of this solution depends on the point in the cycle where ionisation occurred (i.e.
on t0 ). The electron is initially accelerated away from the atom by the field but, sooner
or later, the field changes sign and the electron is slowed to a standstill (V (t) = 0). The
velocity then changes sign, and the electron starts moving back toward the parent ion, but
whether it actually reaches it before the velocity changes sign again depends on t0 .
If ionisation occurs while the field magnitude is rising (i.e. in a quarter-cycle preceding
an extremum), it can be shown (see Problem 10.3) that the electron will never return to the
parent atom. On the other hand, in quarter-cycles where the field magnitude is falling, the
time of recollision can be found by integrating Eq. (10.4) and setting the net excursion to
zero, i.e.
tR
x=
t0
xdt
=
eAX
=0
m2
(10.6)
208
field
excursion X, velocity V
0
0
6
phase (rads)
Fig. 10.2
Set of one-dimensional electron trajectories showing the displacements X (in grey) from Eq. (10.7), and the
velocities V (in black) from Eq. (10.5), both as a function of t (in radians). The (co)sinusoidal eld prole is shown at
the top.
where
X(t) = (cos t cos t0 + (t t0 ) sin t0 ) .
(10.7)
The transcendental equation X(tR ) = 0 can now be solved for the recollision time tR , and
the result substituted into Eq. (10.5) to obtain the velocity and kinetic energy at the point of
recollision. The equation is in fact represented by an interesting geometrical construction
that links the times of ionisation and recollision (see Problem 10.5). If the tangent to E(t) =
A cos t at t = t0 intersects E(t) at some subsequent time, this is the point of recollision.
As noted earlier, the light must be plane-polarised if the electron is to remain on the correct
line; otherwise the electron trajectory will be curved and, classically at least, the electron
will never return to its starting point (see Problem 10.2).
Some results are displayed in Fig. 10.2 in which the solid grey lines are a set of onedimensional electron trajectories X(t) from Eq. (10.7) for departure times from the parent
atom at 12 equally-spaced values of t0 between 0 and /2 radians. The electric field is
shown at the top for reference. The black lines trace the corresponding variations of V (t) =
(sin t0 sin t) from Eq. (10.5), and represent the electron velocities as a function of
time. Both sets of lines terminate at the point of recollision where X(tR ) = 0. On the graphs
of V (t), the length of the verticals to the axis where the lines end represents the recollision
velocity VR . The kinetic energy at recollision is therefore given by UKE = 12 mx 2 = 2UP VR2
from Eqs (10.3) and (10.5).
All trajectories shown in Fig. 10.2 involve a negative X excursion from the phase (or
time) axis, and end at recollision.5 Electrons leaving the atom at exactly the peak of the field
(t0 = 0) have the longest trajectories in terms of both distance and duration, but these
5 The excursion is negative because negatively charged electrons move in a field that is initially positive.
209
electrons only just manage to complete the return journey, arriving back at the starting
point at t = 2 with zero velocity and zero kinetic energy. For electrons ionised after
the peak, the trajectories do not extend quite so far from the parent ion, although they are
still counted as long trajectories at this stage. The recollision velocities rise quite rapidly
as the ionisation time increases, reaching a maximum of |V |
= 1.26 when t0 0.314,
which corresponds to a recollision kinetic energy of 2 1.262 UP . This is the origin of
the equation, frequently repeated in the literature, that the maximum energy available at
recollision is [120]
URmax = 3.17UP + UI .
(10.8)
The ionisation energy UI is included here because the electron returns to a bound state in
the parent atom.
As t0 increases further, the trajectories continue to shorten, and the recollision velocities
for these short trajectories fall away too, until t0 = /2 where the field is zero. Throughout the next quarter-cycle of the field when the field magnitude is rising, no trajectories
return to the starting point, and all electrons escape. The quarter-cycle ends at t0 = ,
after which the entire sequence repeats with the opposite polarity.
It is worth emphasising that trajectories are classified as long or short according to the
time interval between ionisation and recollision. From Fig.10.2, it is clear that the earlier
the time of ionisation, the later the time of recollision, so the long trajectories correspond to
early departures and late arrivals, and short trajectories to late departures and early arrivals.
The boundary between the long and short trajectories is taken to be the point where the
recollision energy is maximised according to Eq. (10.8).
10.2.2 Recollision energy

The kinetic energy component UKE of the recollision energy is plotted (in units of UP ) in
Fig. 10.3. The solid black line (on the left) shows UKE plotted as a function of the phase
at ionisation. The graph shows a sharp initial rise in the energy, the peak at 3.17, and the
subsequent fall to zero. The solid grey curve to the right is exactly the same data, but now
plotted as a function of the phase at recollision, rather than at ionisation. Since electrons
ionised early follow long trajectories and recollide late, the left-hand side of the black curve
corresponds to the right-hand side of the grey curve, which is labelled long for that reason.
Similarly, the right-hand wing of the black curve and the left-hand wing of the grey curve
are associated with the short trajectories. These two wings meet where the phase is /2
(arrowed), which corresponds to the first zero of the field. Notice that ionisation times
are spread over a quarter-period of the field, whereas the recollisions are spread over the
remaining three-quarters of the cycle.
As explained in Section 10.1, when the electron returns to the parent atom at recollision,
the available energy, made up of the acquired kinetic energy plus the ionisation energy, is
emitted in the form of radiation at frequency = (UKE + UI )/h. Written in terms of laser
intensity, the kinetic energy at recollision is given by

e2
I 2
(10.9)
UKE = 3.17
8 2 0 c3 m
3
short
long
0
0
departure phase
Fig. 10.3
w (t,)
Kinetic Energy (units of Up)
210
4
arrival phase
The kinetic energy at recollision (in units of UP ) plotted as a function of the phase at ionisation t0 (on the left) and
the phase at recollision tR (on the right). The dotted lines show the corresponding variation of the weighting
function of Eq. (10.11).
where Eqs (10.3) and (10.8) have been used. In terms of intensity and wavelength, the
corresponding number of harmonics is
HKE =
UKE
= 2.39 1013 ((m))3 I (W cm2 ).
h
(10.10)
An ionisation component (see Eq. 10.8) must of course be added to both these equations to
obtain the corresponding values at recollision.
Notice especially that the kinetic energy element of the recollision energy is proportional
to the laser intensity, and so varying the intensity has a direct effect on the cut-off point.
However, the range of viable intensities is quite limited. At the lower end, Eq. (10.10)
indicates that, for = 1 m, little of interest is likely to happen at intensities under about
51013 W cm2 . On the other hand, an upper limit on the intensity is set by the point at which
the atoms become fully ionised, and the efficiency of the recollision process is consequently
seriously impaired. In practice, this limit is maybe 3 1015 W cm2 . Incidentally, at even
higher intensities, the Lorentz force associated with the magnetic component of the EM
field would become significant. This would drive the electron along a curved path and,
as for elliptically-polarised light, recollisions would be prevented. But this regime is well
above the ionisation limit.
Equations (10.9) and (10.10) also suggest that higher recollision energies can be realised
by moving to higher wavelengths. This is a controversial issue at the time of writing because
it seems that the advantage may be offset by a severe reduction in conversion efficiency as
the wavelength is increased. At a lower fundamental frequency, the time interval between
ionisation and recollision is lengthened, and this gives the electron wave packet more time
211
spectral intensity
1x108
1x104
1x100
1x104
0
Fig. 10.4
20
40
harmonic order
60
80
Simulated emission spectra for a neon atom excited by 750 nm radiation at 5 1014 W cm2 , under three different
conditions: no weighting (black line), full weighting factor of Eq. (10.11) (upper grey line), ionisation weighting term
only (lower grey line).
to disperse, reducing the strength of the recollision. Some degree of trade-off between the
two effects may be possible but, as noted earlier, conversion efficiencies in HHG are already
small, with best results in the region of 1 in 105 .
10.2.3 High harmonic spectra

A rough idea of the overall emission spectrum can be gained by computing the emission
frequency for a large number of electron trajectories spread over a quarter-period, and
plotting the distribution of values. The result of this exercise for a neon atom excited by a
5 1014 W cm2 beam at 750 nm is plotted as the black line in Fig. 10.4. At this frequency,
the 21.56 eV ionisation potential of neon lies close to the thirteenth harmonic, while the
kinetic energy gained in the continuum (UKE = 1.33 1017 J) corresponds to roughly
50 harmonic orders from Eq. (10.10). The figure duly shows that the spectrum exhibits a
broad plateau, starting at the thirteenth harmonic and running up to a sharp cut-off close
to the sixty-third harmonic, which is at a wavelength of approximately 12 nm, in the XUV
region of the EM spectrum.
There are several weaknesses in the argument as it stands, apart from the obvious one
that it is essentially classical. First of all, it is based on the false assumption that the atom
is equally likely to be ionised at all points within the quarter-cycle. In fact, the ionisation
rate will inevitably drop as the magnitude of the field falls, so the contribution of the late
departing electrons to the low-frequency side of the spectrum will certainly have been exaggerated in our analysis. However, this effect is offset to some extent in more sophisticated
treatments by the fact that the quantum mechanical wave packet representing the electron
in the continuum has more time to spread out on the long trajectories, and this weakens the
212
gas in
vacuum chamber
dispersed harmonics
laser
pulse
pulsed
valve
gas-jet
grating
vacuum
window
Soft X-ray detector
(microchannel plate)
vacuum
pump
Fig. 10.5
Schematic diagram of the experimental set-up used to generate the data of Fig. 10.6, showing the grazing incidence
at-eld diffraction grating and the soft X-ray sensitised microchannel plate detector. (Courtesy of J.W.G. Tisch,
Imperial College Attosecond Laboratory.)
emission at recollision. The precise weighting functions that should be used to account for
these effects is hotly debated, but one possible expression that takes both into account is
[122,123]

3 !
2(2UI )3/2
T
E(0)
w(t, ) =
.
(10.11)
exp
E(t)
3E (t)

spreading
ionisation
In the first factor, which governs wave packet spreading, is the time the electron spends
in the continuum and T is the optical period. Primes on the ionisation potential and the
optical field in the final factor indicate that values in atomic units are to be used.6
The dotted lines in Fig. 10.3 show plots of Eq. (10.11) for the parameter values used for
Fig. 10.4, while the grey lines in Fig. 10.4 show the effect of the weighting factor on the
spectrum. The lower of the two shows the result when the spreading factor in the equation
is omitted; as expected, weighting the distribution in this way has a stronger effect at lower
frequencies.
A schematic diagram of a typical HHG experimental set-up is shown in Fig. 10.5 [72].
The fundamental laser beam is focused into a pulsed gas jet, and the resulting harmonics
are dispersed by a grating onto a soft X-ray detector. All the equipment must be enclosed
in a vacuum chamber in order for the gas jet to function properly, and of course to ensure
that the harmonics are transmitted to the detector.
An experimental spectrum obtained under broadly similar conditions to those of
Fig. 10.4 is shown in Fig. 10.6 [72]. This confirms the presence of a broad plateau of
harmonics, extending to around the fifty-ninth harmonic in this case, after which there is
a sharp cut-off. The main qualitative difference between the two figures is the appearance
of discrete odd harmonics in Fig. 10.6, which arise from the periodic nature of the driving
6 1 a.u. of potential = 27.212 V, and 1 a.u. of field = 5.142 1011 V m1 .
213
Spectral intensity
59
25
55
35
40
50
45
60
70
80
90
100
eV
Fig. 10.6
Experimental HHG spectrum in neon excited by a 7 fs 250 J CEP stabilised Ti:sapphire laser pulse centred at 780 nm
(average over 25 shots). See the caption to Fig. 10.5 for more details. (Courtesy of J.W.G. Tisch, Imperial College
Attosecond Laboratory.)
field. By contrast, Fig. 10.4 was based on ionisations within a single quarter-cycle of the
field. This point is discussed further in Section 10.4 below.
10.2.4 Phase matching

A further weakness in the model outlined so far is that it is based on the behaviour of a single
atom, rather than on a spatial distribution of atoms. The propagation of harmonic radiation
from individual source atoms to the detector, through the gas, and quite possibly through
the focal region of the driving field too, needs to be taken into account in a fully-fledged
analysis.
The phase-matching problem for second and third harmonic generation in a focused
beam was treated in Section 2.5. There are of course now large numbers of harmonics in the
analysis, rather than just one, although, as indicated in Table 2.1, the diffraction integrals
are zero under normal dispersion for all harmonics higher than the second. Thin gas jets are
used in HHG work for this reason. But there are two major complications in the HHG case.
For one thing, the presence of ionised free electrons in the interaction region changes the
local refractive index. But a more serious problem is that whereas, for harmonic generation
in crystals, the phase of the polarisation antenna is linked directly to the phase of the relevant
power of E, now the antenna phase is dependent on intensity, and it therefore varies both
spatially and temporally within the interaction region. It may turn out to be possible to
balance this effect against the variation of the Gouy phase, but this is certainly a difficulty
that one could well do without.
A full discussion of these issues is beyond the scope of the present chapter, although
an overall feature that emerges is that phase matching is harder to achieve for radiation
from the long trajectories than from the short trajectories. There is therefore this further
aspect to include in the balance between long- and short-trajectory emissions, in addition
to ionisation weighting and wave-packet spreading.
For a wider discussion of HHG and related issues, readers are referred to [112,124].
214

The sequence shown in Fig. 10.2 repeats every half-cycle of the field with alternating
polarity, but the same emission spectrum. Two bursts of harmonic radiation are therefore
emitted in each cycle of the field, one on the positive field excursion, and one on the
negative. These are commonly known as half-cycle bursts. The overall behaviour is shown
in Fig. 10.7 where the emission frequency (in units of harmonic order) is plotted as a function
of time (in cycles). The driving waveform is shown at the bottom for reference. The section
of each burst where w(t, ) > 0.1 (see Eq. 10.11) is drawn as a solid line. The tails where the
weighting factor is smaller are drawn dotted. The leftmost burst in the figure derives from
electrons ionised in the first quarter-cycle of the field. A burst further to the left, generated
by electrons ionised in the quarter-cycle between 12 and 14 is not shown.
The key question that now arises is: what kind of intensity profile is associated with
the time-dependent spectra of Fig. 10.7? Clearly there will be some kind of high harmonic
pulse train, but what will be the form of its individual components? The overall bandwidth,
covering a range up to and beyond the sixtieth harmonic, is certainly sufficient to support
pulses as short as 30 as (1 as = 1018 s), provided the phases can be properly organised, but
how can that be done? While more sophisticated theoretical models are needed to answer
this question definitively, some aspects of the classical picture are preserved in the various
quantum mechanical treatments. For example, Fig. 10.7 implies (correctly) that, if attention
is focused on a band of harmonics, between (say) order 40 and 50, this will receive two
distinct contributions in each half-cycle, an early burst from the short trajectories to the
left of each peak, and a later burst from the long trajectories to the right. The figure also
harmonic order
60
40
20
Fig. 10.7
2
cycles
Emission frequency as a function of time for neon excited by 750 nm CW laser radiation at an intensity of
5 1014 W cm2 showing the half-cycle bursts. Frequency is plotted in units of harmonic order. Sections of the
spectral prole where the weighting factor of Eq. (10.11) is less than 0.1 are shown dotted.
215
harmonic order
60
40
20
Fig. 10.8
2
cycles
As Fig. 10.7 but for a pulse containing only two optical cycles (full width at half maximum intensity). The peak
intensity is 5 1014 W cm2 .
suggests that the pulses are likely to be chirped, since the frequency obviously rises with
time before each peak, and falls with time subsequently.
Once a train of attosecond pulses (i.e. pulses significantly shorter than 1 fs) has been
created, the next question is how to isolate a single pulse from the train. A number of
strategies have been devised for this purpose, the obvious first step being to use a pulsed
rather than a monochromatic fundamental field to drive the HHG process. Figure 10.8
shows a typical prediction from the classical model for a fundamental pulse with an intensity
FWHM of two optical cycles. The first thing to notice about Fig. 10.8 is its asymmetry. As
in Fig. 10.7, only those parts of the profiles where w(t, ) > 0.1 are drawn as solid lines
and, whereas the first three bursts can be ignored on this criterion, the corresponding bursts
on the right contain solid sections. The main reason for the skewing is that the weighting
factor depends on the field at the time of ionisation, whereas the spectra are based on the
electron kinetic energy at the time of recollision which may be up to one period later.7
The second noteworthy aspect of Fig. 10.8 is that significant harmonic content is confined
to just three half-cycles; the spectrum extends beyond the fortieth harmonic on only two
of these, and beyond the sixtieth on only one. This feature is quite easy to understand
from Eq. (10.8) once UP (defined in Eq. 10.3) is recognised as time dependent. It certainly
suggests that using a few-cycle fundamental pulse is a valuable first step.
Further refinement is possible by tuning the carrier-envelope phase (CEP), although a
potential complication is that the CEP changes as a few-cycle pulse passes through a focus
as a result of the Gouy phase; see Section 2.5. Various other tactics for creating single
attosecond pulses have been explored. For example, a high-pass spectral filter with a cutoff around the fiftieth harmonic would deliver the highest harmonics in the central burst
7 Another consequence of pulsed excitation is that on the leading edge where the pulse amplitude is rising, the
bursts are incomplete, even when the weighting factor is ignored.
216
of Fig. 10.8, while strongly attenuating the other bursts in the sequence. Another effective
strategy is to modulate the polarisation of the driving pulse in a technique called polarisation gating. A simple argument has already been offered to suggest that HHG depends
critically on the use of linearly polarised light. So if the polarisation can be changed rapidly
from elliptical to plane and back to elliptical, a powerful switch can be created. Pulses of
130-as duration containing 1.2 optical cycles, and with spectra extending to 36 eV (corresponding to the twenty-third harmonic of 750 nm radiation) have been achieved using this
technique [125].
Reviews of recent work on attosecond pulse generation can be found in [13, 111, 124].

We end this chapter, and the book, with two fundamental questions about HHG. The first is
this. Why does Fig. 10.4 show a broad continuous spectrum with no evidence of individual
harmonics, while Fig. 10.6 clearly indicates the presence of discrete odd harmonics? Given
the way in which Fig. 10.4 was created, there is of course no way it could have been
been any different. A continuous range of electron trajectories was used to generate a
continuous spread of recollision energies, which were divided by Plancks constant to
create a continuous spectrum.
Where then does the discreteness in Fig. 10.6 come from? The answer lies in Fourier
analysis. A spectrum containing regularly spaced discrete components is necessarily associated with some degree of temporal periodicity, so a single half-cycle burst can never
deliver a spectrum in which individual harmonic components are identifiable. Conversely,
the appearance of discrete spectral components in Fig. 10.6 indicates that two or more
half-cycle bursts are involved [111].
A second issue now arises. The fact that the discrete components in Fig. 10.6 correspond
to odd harmonics is of course associated with the centrosymmetric nature of the gaseous
medium. However, the figure is not particularly convincing on this score. The harmonics
are reasonably well separated up to about the forty-fifth harmonic but, from there on, the
spectrum becomes more continuous in nature.
One possible reason for this is simply that the spectral resolution is limited. However,
only a single half-cycle burst in Fig. 10.8 extends to the sixtieth harmonic, so one should
not expect to see significant discreteness at the high end of the spectrum anyway.8 But that
argument also suggests that even harmonics really can be generated in HHG experiment,
which raises the further question of how this can be so without violating basic symmetry
principles.
The answer is simply that few-cycle driving pulses possess their own inherent onewayness and hence themselves lack inversion symmetry. For example, the two-cycle pulse
used to create Fig. 10.8, and shown at the base of that figure, is certainly different in the
8 The fact that there is some discreteness in Fig. 10.6 up to the highest harmonic is probably because the driving
pulse was slightly longer than two optical cycles and/or the CEP was different from that in Fig. 10.8.
217
positive and negative directions; the central maximum is at +10 (with respect to the lefthand scale), while the two neighbouring negative peaks are at 9.1. In fact, it is this very
feature that has created only one burst in Fig. 10.8 with spectral content beyond the sixtieth
harmonic.
With regard to this aspect of nonlinear optics, we have therefore come full circle in this
book. At the beginning of Chapter 1, it was stressed that a non-centrosymmetric nonlinear
medium was essential for generating even harmonics, and that amorphous solids, liquids
and gases were ruled out for symmetry reasons. Now, at the end of Chapter 10, we have
discovered an exception to the rule. But there is, of course, no violation of fundamental
symmetry principles, merely a new source of asymmetry. It turns out that even harmonics
can be generated in a centrosymmetric medium provided a non-centrosymmetric optical
pulse is used to excite the nonlinearity.
Problems
10.1 Show that when a one-dimensional Coulomb potential is modified in the presence of
an optical field (see Fig. 10.1), the potential barrier is lowered by an amount that is
proportional to the inverse square root of the field. (But note that a proper treatment
of ionisation suggests a far stronger dependence on field; see Eq. 10.11.)
10.2 Show that if the electron and the ion are treated as point particles, even the slightest
degree of ellipticity in the polarisation will prevent a recollision
10.3 If ionisation occurs while the field magnitude is rising (i.e. in a quarter-cycle preceding
an extremum), show that the electron will never return to the parent atom.
10.4 On the basis of the classical model, estimate the maximum excursion of the electron
from the parent ion if the fundamental wavelength is 800 nm and the laser intensity
is 1015 W cm2 .
10.5 In Fig. 10.2, show that the tangent to the electric field curve E(t) = A cos t at the
time of ionisation intersects E(t) at the time of recollision.
10.6 In an HHG experiment, He atoms are excited by 800 nm Ti:sapphire laser pulses with
a peak intensity of 2 1015 W cm2 . On the basis of the classical model developed in
this chapter, what values would you predict for (a) the highest harmonic and (b) the
minimum wavelength. [Ionisation potential of helium = 24.58 eV.]
Appendix A Conventions and units
A1 Introduction
The formulae governing nonlinear frequency mixing appear in a bewildering variety of
forms in the literature. This is caused by the use of different unit systems, different conventions for defining complex field amplitudes, and different sign conventions in wave
propagation. To make matters worse, apparent discrepancies (often by factors of 2) have
sometimes turned into real discrepancies (errors) as formulae have been copied without
allowing for the conventions in use when they were originally written. The purpose of this
appendix is therefore to restate the conventions used in this book, and to attempt to reconcile
standard results quoted in different sources.
Many of the issues can be traced back to the different ways of defining the complex
envelopes of real variables. We start by defining electric fields in the general form1
E(z, t) = p
E n (z, t) exp {ie n t} + c.c.
=p
E n (z, t) exp {ie (n t kn z)} + c.c.
(A1.1)
where E n = E n exp{ikn z}, and the envelope function E n is slowly varying in both time
and space. Analogous formulae apply for the polarisation (see Eq. 2.5) and for the associated
magnetic field variables.
The parameter e (= 1) in Eq. (A1.1) allows for a choice of sign in the space-time
coefficient of the exponential, while the pre-factor p can be either 12 or 1. The rationale
for p = 12 is that the moduli of the complex fields are then the same as the amplitudes
of the real signal. On the other hand, setting p = 1 makes Eq. (A1.1) marginally neater,
and simplifies the pre-factors in several common formulae in nonlinear optics at the same
time.
The values of p and e work their way through into all subsequent equations. The choices
used by different authors are listed in columns 2 and 5 of Table A1. The table also includes
further parameters defined in the next section.
1 Equation (2.4) is the special case where p = 1 and = +1.
e
2
218
A2 The nonlinear polarisation
219
Table A1 Conventions used in different sources.

p
Boyd [1] (2nd edn)

Boyd [1] (3rd edn)
Sutherland [126]
Yariv [20] (3rd edn)
1
1
1
This book
e m
f = 2p qe
1
2
1
4
0
01
K = q
1
01
1
1
1
1
1
1
1
1
1
1
1
1
8
2
2K
01
1
2
01
note 1
note 2
Note 1: Sutherland, who uses the symbol K for our q , quotes results in both SI (K = 1) and
Gaussian units (K = 4 ).
Note 2: Yariv uses the symbol deff for our 0 deff . This means that an extra 0 appears in the
denominator of his coupled-wave equations, although the equations are actually identical to ours.
A2 The nonlinear polarisation

We take the case of sum and difference frequency generation as our basic example. As
explained in Chapter 2, the frequencies of the three interacting fields are linked by the
equation
1 + 2 3 .
(A2.1)
In the case of parametric amplification, which is governed by the same set of differential
equations, the wave at 3 is the pump, and those at 2 and 1 are the signal and the idler.
The signal frequency is normally the higher of the two, although this is not prescriptive.
In terms of the effective nonlinear coefficient deff (see Section 4.5), the equations for the
polarisations are
2 )E(
1 );
P (3 ) = 4pqdeff E(
3 )E (1 );
P (2 ) = 4pqdeff E(
3 )E (2 ).
P (1 ) = 4pqdeff E(
(A2.2)
In Gaussian units, q = 1, P and E are both in statvolts per cm, and hence deff is in cm
per statvolt.2 In SI units, it is common to write q = 0 , in which case deff is in metres per
volt (or more often pm V1 ). However, some authors (e.g. Yariv [20]) use SI units, but set
q = 1 and absorb 0 into deff . Equation (A2.2) then becomes
2 )E(
1 ), etc.
P (3 ) = 4pdeff E(
(A2.3)
where the modified coefficient deff = 0 deff is in coulombs per volt (C V1 ).3 Conversion
factors between the various unit systems are listed in Table A2.
For a type I interaction in KDP, for example, deff is given by
deff = d36 sin sin 2
2 1 statvolt = 300 V.
3 Yariv writes d
eff as deff , so his equations appear superficially to be different from ours.
(A2.4)
Conventions and units
220
Table A2 Conversion factors linking Gaussian and SI Units.

Multiply value in
by
to get value in
cm/statvolt (esu)
cm/statvolt (esu)
C/V
C/V
pm/V
pm/V
(4 0 /3) 104 = 3.709 1015

(4/3) 108 = 4.189 108
(3/4 0 ) 104 = 2.696 1014
1012 /0 = 1.129 1023
(3/4 ) 108 = 2.387 109
0 1012 = 8.854 1024
C/V
pm/V
cm/statvol (esu)
pm/V
cm/statvolt (esu)
C/V
where is the phase-matching angle, is the azimuthal angle, and d36

= 0.43 pm V1 in
the SI system; see Eq. (4.18).
A3 Coupled-wave equations
The general form of the coupled-wave equation of Eq. (2.8) is
E n
n NL
P
= i( e )
z
2cnn n
(A3.1)
where = 4 in Gaussian units or 01 in SI. Using Eq. (A2.2) for the nonlinear polarisation,
and recalling that E n = E n exp{ikn z} and Pn = Pn exp{ikn z}, we obtain the coupledwave equations in the general form

E 3
3
= if
deff E 2 E 1 exp {i kz}
(A3.2a)
z
n3 c

E 2
2
(A3.2b)
= if
z
n2 c

E 1
1
= if
(A3.2c)
z
n1 c
where f = 2p qe , and n3 , n2 and n1 are the respective refractive indices. The parameter
in the exponential coefficients is the product e m , where m (= 1) determines the
sign convention used in defining the mismatch parameter
k = m (k3 k2 k1 )
(A3.3)
in which k3 = n3 3 /c, etc.

The equation for intensity (power per unit area) is
I=
2p2 nc |E|2
(A3.4)
where n is the refractive index. Notice the presence of p in this equation.

The conventions used by different authors for the further parameters introduced in this
section are included in Table A1.
Appendix B Linear and nonlinear susceptibilities

in the time and frequency domains
B1 Non-instantaneous response: the time and frequency
domain equations
In this appendix, the structure of the polarisation expansion that has appeared at various levels of sophistication in this book will be examined in more detail. The introductory treatment of Chapter 1 was based on the simple statement of Eq. (1.24),
namely
P = 0 ( (1) E + (2) E 2 + (3) E 3 + ).
(B1.1)
The tensor nature of the nonlinear coefficients has been ignored. The polarisation, P ,
and electric field, E, are real variables in C m2 and V m1 , respectively, which implies
that the nth-order susceptibilities (n) are in (m V1 )n1 , or more conveniently in
(pm V1 )n1 .
In Chapter 1, the frequency dependence of the nonlinear coefficients was ignored. The
coefficients in Eq. (B1.1) are therefore simple constants, which means that the polarisation
responds instantaneously to changes in the field. This property is never realised in actual
physical systems, and more sophisticated mathematical machinery is needed to describe
what happens in practice.
The following sections provide a brief review of what this involves. The discussion will
also address an issue surrounding equations such as Eq. (4.5), namely
Pi (2) = 12 0

jk
ijSHG
k (2; , )Ej ()Ek ().
(B1.2)
In this, the defining equation for second harmonic generation, the polarisation, the fields
and the coefficient are all written as functions of frequency, which immediately raises the
question of whether this is a time-domain or a frequency-domain equation. As we will
show below, the equation is in reality a hybrid form. On the one hand, E j () retains
its normal units of V m1 , and represents the electric field in a narrow frequency band
around . (If E j () were the Fourier transform of the time-dependent field, it would of
course be in V m1 Hz1 .) But if Eq. (B1.2) is not a frequency-domain equation, it is not
a true time-domain equation either. Equation (B3.1) below shows what that would look
like.
221
Linear and nonlinear susceptibilities in the time and frequency domains
222
B2 The linear term

We focus on the linear term, because it is the easiest to handle mathematically and the
principles extend to the higher-order terms. To take account of non-instantaneous response,
the first term on the right-hand side of Eq. (B1.1) should be replaced by
P
(1)

(t) = 0
1 ( )E(t )d .
(B2.1)
The equation specifies the polarisation of the medium at time t in terms of the electric field
at all previous times t , etc. through the response function 1 , which contains all the
information about the material properties, and must be zero for < 0 to preserve causality.
The response function is properly a tensor quantity, but we ignore this feature here for the
sake of simplicity.1
The connection between non-instantaneous response and the frequency dependence of the
linear susceptibility can be established by introducing E(),

which is the Fourier transform

it d. Equation (B2.1) can now be written
of E(t), given by E(t) =

E()e
(1)

(t) = 0

d1 ( )e
it
E()e
= 0
it
d (1) ()E()e
(B2.2)
where the final step defines the frequency-dependent coefficient
(1)

() =
d1 ( )ei.
(B2.3)
If we also introduce the Fourier transform of P (1) (t) and match terms, we obtain
P () = 0 (1) ()E()
(B2.4)
which is the frequency-domain analogue of Eq. (B2.1). The form of this equation is similar
to that of Eq. (4.1), which reads
P () = 0 (1) ()E().
(B2.5)
However, the key difference between Eqs (B2.4) and (B2.5) is that P () and E()
are the
Fourier transforms of P (t) and E(t), whereas P () and E() are the amplitudes of quasimonochromatic components of the polarisation and electric field; see Eqs (2.4) and (2.5)
in Chapter 2. As we shall now show, these are different things; indeed the corresponding
parameters do not even have the same dimensions!
1 Refer to Chapter 4 and Appendix C for detailed discussion on the tensor nature of the coefficients.
B3 The second-order nonlinear term
223
In Eq. (2.4), the electric field was written

n )ein t + c.c.
E(t) = 21
E(
(B2.6)
(n ) since E(t) is real.2 The Fourier transform of E(t) is

where E(
n) = E
1
E()
=
2
dt 12

n )ein t + c.c eit
E(
n

n )( n ) + E(
= 12
E(
n )( + n )
(B2.7)
which is a set of delta functions, as expected. Substituting Eq. (B2.7) into Eq. (B2.2) leads to
P
(1)
it
d (1) ()E()e
= 12 0
(t) = 0
n )ein t + c.c.
(1) (n )E(
(B2.8)
If we now write the polarisation as in Eq. (2.5), we have

P (1) (t) = 12
P (n )ein t + c.c.
(B2.9)
Matching terms with Eq. (B2.8), we recover Eq. (B2.5).

n ) is in V m1 , like E (1) (t), whereas E()
Notice that E(
is in V m1 Hz1 because it
is linked to E (1) (t) via a Fourier integral. Similarly, P (n ) is in C m2 , whereas P () is
in C m2 Hz1 . The parameters E(t), P (1) (t) and (1) ( ) in Eqs (B2.1) (B2.3) are real
quantities, while E(),

P (1) () and (potentially) (1) () are complex.
In a medium with no memory (the case of instantaneous response), 1 ( ) = (1) ( ).
The linear term then becomes P (1) (t) = 0 (1) E(t) in the time domain, or P (1) () =
0 (1) E()
in the frequency domain, where (1) is now a real constant that is independent
of both time and frequency.
B3 The second-order nonlinear term

A similar procedure can be applied to the second-order nonlinear term. Written in an
analogous form to Eqs (B2.1) and (B2.2), the results are
P
(2)

(t) = 0

d

= 0
d 2 ( , )E(t )E(t )

d
2 E(
n ) was written E n in Eq. (2.4).
(B3.1)
) exp{i( + )t}
d (2) (, )E()
E(
Linear and nonlinear susceptibilities in the time and frequency domains
224
where 2 ( , ) = 0 for , < 0 (to ensure causality) and
(2)
(, ) =

d

d 2 ( , )ei( + ) .
(B3.2)
Once again, if the field spectrum is made up of delta functions (as in Eq. B3.2), the frequency
integrals in Eq. (B3.1) reduce to summations, and we obtain

n )E(
n ) exp{i(n + n )t}.
P (2) (t) = 14 0
(2) (n , n )E(
(B3.3)
n, n
If P (2) (t) is now written in the form of Eq. (B2.5), we obtain terms of the type
n )E(
n ).
P (2) (n + n ) = 12 0 (2) (n , n )E(
(B3.4)
This result is the basis of the sum frequency generation formula of Eq. (4.3), which reads

(2)
(2)
Pi (3 ) = 12 0
ij k (3 ; 1 , 2 )E j (1 )E k (2 ) + ij k (3 ; 2 , 1 )E j (2 )E k (1 ) .
jk
(B3.5)
The notation in these last two formulae is slightly different, mainly because the second includes the tensor nature of the coefficients and the first does not. Also, in line
with convention, the susceptibilities in Eq. (B3.5) carry three frequency arguments rather
than two.
Appendix C Denition of the nonlinear coefcients
C1 Introduction
The defining equations for nonlinear optical coefficients are inherently intricate and the
complexity is worsened by the use of different conventions by different authors. A perennial
problem is that the equations are huge when written out in full, but opaque when shortened
by the use of summations and other notational devices.
This appendix contains a blow-by-blow discussion of the most important definitions, with
a selection of examples and special cases that will hopefully serve to clarify the general
equations.
C2 Second-order processes
C2.1 Sum frequency generation and intrinsic permutation symmetry
The nonlinear coefficient for the basic second-order process of sum frequency generation
(SFG) is defined by1
(2)
Pi (3 ) = 12 0
ij k (3; 1 , 2 )E j (1 )E k (2 )
(C2.1)
p
jk
where 1 + 2 = 3 , and the indices ij k can each be x, y or z. The symbol
%
p
indicates
that the right-hand side should be summed over all distinct permutations of 1 and 2 . So,
provided 1 = 2 , terms with the two frequencies in reverse order should be included.
This feature has its origin in the trivial fact that when one works out (E(1 ) + E(2 ))2 ,
the cross-term is E(1 )E(2 ) + E(2 )E(1 ) = 2E(1 )E(2 ).
When the sum over permutations is carried out, Eq. (C2.1) becomes

(2)
(2)
Pi (3 ) = 12 0
ij k (3 ; 1 , 2 )E j (1 )E k (2 ) + ij k (3 ; 2 , 1 )E j (2 )E k (1 )
jk

(2)
(2)
= 12 0
ij k (3 ; 1 , 2 ) + ikj (3 ; 2 , 1 ) E j (1 )E k (2 )
jk
225
1 The pre-factor of 1 arises from the field definitions of Eq. (2.4).

2
(C2.2)
Denition of the nonlinear coefcients
226
where the second line follows by exchanging the dummy indices j and k in the second
term of the first line. Since ij k (3 ; 1 , 2 ) and ikj (3 ; 2 , 1 ) are associated with the
same pair of fields, there is no way of distinguishing their effect, and any measurement will
record the sum of the two. It therefore makes sense to contract Eq. (C2.2), which can be
done in two slightly different ways. One possibility is to write
Pi (3 ) = 0

jk
(2)
ij k (3 ; 1 , 2 )E j (1 )E k (2 )
(C2.3)
where
(2)
ij k (3 ; 1 , 2 ) =
1
2

(2)
(2)
ij k (3 ; 1 , 2 ) + ikj (3 ; 2 , 1 )
(C2.4)
and the overbar indicates that an average has been taken. The other possibility is to invoke
the convention known as intrinsic permutation symmetry (IPS) and assume that
(2)
(2)
ij k (3 ; 1 , 2 ) = ikj (3 ; 2 , 1 ).
(C2.5)
In this case Eq. (C2.2) becomes

Pi (3 ) = 0

jk
(2)
ij k (3 ; 1 , 2 )E j (1 )E k (2 )
(C2.6)
which is identical to Eq. (C2.3) apart from the overbar. The key point about both Eq. (C2.3)
and Eq. (C2.6) is that, one way or another, coefficients with 1 and 2 in reverse order have
been eliminated, and the ordering is now fixed by definition.
In some ways, the averaging approach is preferable to the imposition of IPS. An arbitrary
assumption is avoided, and the overbar on the coefficient in Eqs (C2.3) and (C2.4) serves
as a reminder that the frequency ordering is fixed. However, it is common practice to use
IPS, and so we will do so here. Reminders that the frequency ordering in Eqs (C2.3), (C2.6)
and similar equations is fixed by definition are included where appropriate.
As an example, consider the case where i = x and j k = yz, zy. Equation (C2.2) then
reads

(2)
(2)
Px (3 ) = 12 xyz
(3 ; 1, 2 )E y (1 )E z (2 ) + xyz
(3 ; 2 , 1 )E y (2 )E z (1 )

(2)
(2)
+ 12 xzy
(3 ; 1, 2 )E z (1 )E y (2 ) + xzy
(3 ; 2 , 1 )E z (2 )E y (1 )
(C2.7)
which can be rearranged as
Px (3 ) =

(2)
(2)
xyz
(3 ; 1 , 2 ) + xzy
(3 ; 2 , 1 ) E y (1 )E z (2 )

(2)
(2)
+ 12 xzy
(3 ; 1 , 2 ) + xyz
(3 ; 2 , 1 ) E z (1 )E y (2 ).
1
2
(C2.8)
C2 Second-order processes
227
(2)
(2)
(2)
(2)
By assuming that xyz (3; 1 , 2 ) = xzy (3 ; 2 , 1 ) and xzy (3 ; 1 , 2 ) = xyz (3 ;

2 , 1 ) under IPS, Eq. (C2.8) reduces to

(2)
(2)
Px (3 ) = 0 xyz
(3 ; 1 , 2 )E y (1 )E z (2 ) + xzy
(3 ; 1 , 2 )E z (1 )E y (2 ) .
(C2.9)
The corresponding result for i = x and j k = yy is
(2)
Px (3 ) = 0 xyy
(3 ; 1 , 2 )E y (1 )E y (2 )
(2)
(C2.10)
(2)
where xyy (3 ; 2 , 1 ) has been identified with xyy (3 ; 1 , 2 ).
C2.2 Second harmonic generation

The results of Section C2.1 have to be adjusted in the case of second harmonic generation.
Provided 1 = 2 = , and only a single beam at is involved (see Section C2.3 below),
%
the summation
is irrelevant and can be removed. Equation (C2.1) now reads
p
Pi (2) = 12 0

jk
ijSHG
k (2; , )Ej ()Ek ().
(C2.11)
Once again, we treat the special case where i = x and j k = yz, zy, in which case (C2.11)
becomes

SHG
SHG
Px (2) = 12 0 xyz
(2; , ) + xzy
(2; , ) E y ()E z ().
(C2.12)
At this point it is conventional to make the replacement

2dx(yz) =
1
2

SHG
SHG
xyz
(2; , ) + xzy
(2; , )
(C2.13)
which leads to
Px (2) = 20 dx(yz) E y ()E z () = 20 d14 E y ()E z ()
(C2.14)
where (yz) means that both yz and zy are included. The second step incorporates the
numerical suffix notation that is normally introduced at this point. More details are given
in Table 4.1; see also Eq. (4.10).
The factor of 2 in Eqs (C2.13) and (C2.14) is a historical artefact. It was introduced at
some point in the history of the subject, probably as a way of wiping out the pre-factor of 12
that has featured in most of the previous equations. As an example the corresponding result
228
for i = x and j k = yy is
SHG
(2; , )E y2 () = 0 d12 (2; , )E y2 ().
Px (2) = 12 0 xyy
(C2.15)
C2.3 Note on non-collinear beams

A caveat needs to be entered at this point. It is important to understand that the distinction
between 1 and 2 disappears in the case of second harmonic generation not merely because
the two waves have the same frequency but because they are derived from a single beam.
It follows that in the case of non-collinear SHG, when two separate beams are involved,
albeit of the same frequency, the formulae of sum frequency generation should be used with
1 = 2 = . Two distinct beams interact in this case, even if they were originally derived
from the same laser source.
C2.4 The general second-order process

Further adjustments are needed if any of the participating frequencies is zero. The most
inclusive form of the expression for the nonlinear polarisation is
(2)
Pi (3 ) = 0 K2 (3 ; 1 , 2 )
ij k (3 ; 1 , 2 )E j (1 )E k (2 )
(C2.16)
jk
where K2 (3 ; 1 , 2 ) = 12 p12 2rl . The three components in K2 arise as follows:

The initial factor of 12 has its origin in the definition of the complex fields, and is the
same 12 that appeared in Eq. (C2.1).

The factor p12 is the number of distinct permutations of 1 and 2 ; so p12 = 2 if the
frequencies are different and 1 if they are the same. This factor is the direct result of the
%
operation in Eq. (C2.1) and, once it is implemented, the ordering of the frequencies
p
within the nonlinear coefficient is fixed.

The parameters l and r are the numbers of zero frequencies to the left and right of the
semicolon in the argument of the nonlinear coefficient.
Why does the number of zeros matter? The answer is that, whereas for a DC field, E(0) = A,
1
it
for
an AC field of the same amplitude, A cos{t + } = 2 E()e + c.c. to ensure that

E() = A. As a result, the factor 2rl will appear in K2 . On the other hand, if 1 , 2
and 3 are all non-zero, r = l = 0, and K2 = 12 p12 , which links Eq. (C2.16) to Eqs (C2.6)
and (C2.11).
Values of K2 , and the various components within it, are listed in Table C1 for four
particular second-order processes. The key point to notice is that the values are exactly
the same as those that appeared naturally in the simple treatment of Chapter 1. This is
either reassuring, or a frustratingly poor return on the investment made in this appendix,
depending on your point of view! The relevant equation numbers are listed in the right-hand
column.
C3 Third-order processes
229
Table C1 Pre-factors for second-order nonlinear processes.

SFG
SHG
Pockels
OR
1 , 2
p12
K2 = 12 p12 2rl
Equations
3
2
1 , 2
,
, 0
, -
2
1
2
2
0
0
0
1
0
0
1
0
(1.17) & (C2.6)

(1.8) & (C2.11)
(1.14) & (C2.19)
(1.8) & (C2.17)
1
2
1
2
C2.5 Optical rectification and the Pockels effect

We now consider two further second-order examples involving DC fields.
For optical rectification (OR), 3 = 0 and a suitable choice for the other two frequencies
is 1 = and 2 = . Negative frequencies must be treated as distinct from positive
frequencies, so p12 = 2, and since the polarisation is DC, l = 1 and r = 0. Hence K2 = 12 ,
and Eq. (C2.16) becomes

ijOR
(C2.17)
Pi (0) = 12 0
k (0; , )Ej ()Ek ()
jk
where E k () = E k (). We stress once again that the ordering of frequencies to the right
of the semicolon in Eq. (C2.17) is a matter of definition, but the choice is arbitrary and we
could equally have written

Pi (0) = 12 0
ijOR
(C2.18)
k (0; , )Ej ()Ek ()
jk
where the frequencies + and have been reversed both in the coefficient and in the
fields.
For the Pockels effect, K2 = 2, and of the two options for the defining equation we
choose

Pi () = 20
ijPEk (; , 0)E j ()Ek (0).
(C2.19)
jk
C3.1 Summary of third-order results
Section C2.1 began with the equation for sum frequency generation. The parallel effect at
third order is the process 4 = 1 + 2 + 3 where all four frequencies are non-zero. The
defining equation, analogous to Eq. (C2.1), is
(3)
ij kl (4 ; 1 , 2 , 3 )E j (1 )E k (2 )E l (3 )
(C3.1)
Pi (4 ) = 14 0
p
j kl
230
Table C2 Pre-factors for third-order nonlinear processes.

3
1 , 2 , 3
p123
K3 = 14 p123 2rl
Equations
SFG
1 , 2 , 3
(C3.1) with 6 permutations
THG
,,
DC SHG
,,0
3
2
1
4
3
2
DC KERR
, 0, 0
(1,25)
AC KERR
, ,
(1.27)
IDRI
, ,
3
2
3
4
where, as before,
%
p
(1.25)
(1.25)
(1.25) & (C3.4)
indicates that all distinct permutations of 1 , 2 and 3 must be
included. This multiplies the number of terms by 6 if all three frequencies are different, and
by 3 when two are the same.
The most general third-order result, analogous to Eq. (C2.16) at second order, is
Pi (4 ) = 0 K3 (4 ; 1 , 2 , 3 )

j kl
(3)
ij kl (4 ; 1 , 2 , 3 )E j (1 )E k (2 )E l (3 ) (C3.2)
where K3 (4 ; 1 , 2 , 3 ) = 14 p123 2rl and p123 is the number of distinct permutations

of 1 , 2 and 3 . The structure of K3 is entirely analogous to that of K2 , which has been
discussed in Section C2.4. The values of K3 in a number of special cases are listed in Table C2
and, as for K2 , are precisely those that arose in the simple treatment of Chapter 1. We stress
%
again that the summation
in Eq. (C3.1) is an instruction to permute the frequencies
p
to the right of the semicolon, and that the p123 factor in K3 is the consequence of the
permutation. As soon as p123 is present in the pre-factor, the ordering of the frequencies in
the nonlinear coefficient is fixed, because the possible permutations have been taken into
account.
C3.2 Example: intensity-dependent refractive index

As an example, we will consider intensity-dependent refractive index in detail. In this special
case, Eq. (C3.1) reads
Pi () = 14 0
(3)
ij kl (; , , ) E j ()E k ()E l ()
j kl
(3)
+ ij kl (;
, , )E j ()E k ()E l ()

(3)
+ ij kl (; , , )E j () E k ()E l () .
(C3.3)
231
By invoking IPS, and juggling the indices, Eq. (C3.3) can be reduced to any of the following
equivalent forms
(3)
(a) Pi () = 34 0
ij kl (; , , )E j ()E k ()E l ()
j kl
(b) Pi () = 34 0
(c) Pi () = 34 0

j kl

j kl
(3)
ij kl (; , , )E j ()E k ()E l ()
(C3.4)
(3)
ij kl (; , , )E j ()E k ()E l ().
These differ only in the position of the negative frequency in the sequence, and which one
to use is entirely a matter of choice. We will adopt form (b) in which 2 = .
In a structurally isotropic medium, it can be shown that coefficients with 60 of the 81
possible sets of ijkl are zero, leaving 21 non-zero combinations. These 21 sets include three
of type jjjj, and 18 (six each) of types jjkk, jkjk, and jkkj. Invoking IPS and general symmetry
principles shows that, within each type, all six members are equal, and that the types are
also related by
jjjj = jj kk + j kj k + j kkj
(C3.5)
where j = k. There are therefore three independent elements.2

Further constraints apply for particular processes. Under form (b) in Eq. (C3.4), for
example, the frequencies associated with the second and fourth indices are the same, and it
follows that
j kkj = jj kk = j kj k .
(C3.6)
Notice that the unequal member of the set has been determined by the choice of form (b)
in Eq. (C3.4). Had we chosen forms (a) or (c), Eq. (C3.6) would be changed.
With i = x, and for the case where Ez = 0, Eq. (C3.4b) reads

2

(3)
(3)
(3)
3
Px () = 4 0 xxxx
Ex () E x () + xyxy
Ex ()E y2 () +2xxyy
Ex ()E y ()E y ()
(C3.7)
where Eq. (C3.6) has been used. With further help from Eq. (C3.5), the equation can be
written

4

2
2 5

(3)
2
2
(3)
3
Px () = 4 0 xyxy Ex () + Ey () Ex () + 2xxyy Ex () + Ey () Ex () .

(C3.8)
In the case where Ez = 0, Eq. (C3.8) generalises to

(3)
(3)
Px () = 34 0 xyxy
(E.E)E x + 2xxyy
(E.E )E x .
2 See Section 5.2.2 for details about the situation in cubic crystals.
(C3.9)
Appendix D Non-zero d elements in

non-centrosymmetric crystals
Table D1 contains information about which d elements are non-zero for each of the 21
non-centrosymmetric crystal classes. Column 3 lists the number of independent elements
for each class and (in brackets) the lower number of independents when the Kleinmann
symmetry condition (KSC) applies.
Detailed information is given in column 4 where semicolons separate independent elements. Elements that are equal to others under all circumstances (in value if not in sign)
are so indicated, while elements grouped within brackets are identical if the KSC applies.
In a few cases (e.g. class 24) d14 = d25 , but both are zero under Kleinmann symmetry as
indicated by the = 0. The rare class 29 is non-centrosymmetric, but other aspects of the
symmetry force all elements of the d matrix to be zero.
Note that, in biaxial media, the mapping from the crystallographic (abc) to the physical
(xyz) axes is not necessarily straightforward. It cannot be assumed that xyz 123, which
many people would take for granted. For a detailed discussion, see Sections 4.4 and 4.6, as
well as Chapter 2 of Dmitriev et al. [26].
Table D1 Non-zero d elements in non-centrosymmetric crystals.

System & class
Symmetry
Independents
Links and examples
Biaxial
Triclinic
1
Monoclinic
3
4
Orthorhombic
6
7
18
all elements are independent
2
m
8 (4)
10 (6)
(14, 25, 36); (16, 21); 22; (23, 34): e.g. NPLO
11; (12, 26); (13, 35); (15, 31); (24, 32): 33
222
mm2
3 (1)
5 (3)
(14, 25, 36): e.g. barium formate

(15, 31); (24, 32); 33
e.g.LBO, KTP, KTA, KNbO3 , KB5
Uniaxial
Tetragonal
9
10
11
12
232
4
4
4 2 m
422
4 (2)
4 (2)
2 (1)
1 (0)
(14 = 25, 36); (15 = 24, 31 = 32)

(14 = 25, = 0); (15 = 24, 31 = 32); 33
(14 = 25, 36) e.g. KDP, ADP, CDA
(14 = 25, = 0) e.g. nickel sulphate
Non-zero d elements in non-centrosymmetric crystals
233
Table D1 (Continued).
14
4mm
3 (2)
(15 = 24, 31 = 32); 33

e.g. potassium lithium niobate
Trigonal
16
6 (4)
18
32
2 (1)
19
3m
4 (3)
Hexagonal
21
22
23
(11 = 12 = 26); (14 = 25, = 0); (15 = 24,

31 = 32)
(16 = 21 = 22); 33 e.g. sodium periodate
(11 = 12 = 26); (14 = 25, = 0) e.g.
-quartz
(15 = 24, 31 = 32); (16 = 21 = 22); 33
e.g. BBO, LiNbO3 , proustite
6
6 2 m
6
2
1
4 (2)
24
26
622
6mm
1 (0)
3 (2)
(12 = 26 = 11); (16 = 21 = 22)

(16 = 21 = 22) e.g. gallium selenide
(14 = 25, = 0); (15 = 24, 31 = 32); 33
e.g. Lithium iodate
(14 = 25, = 0) e.g. -quartz
(15 = 24, 31 = 32); 33 e.g. CdSe
Optically isotropic
Cubic
28
29
31
23
432
4 3 m
1
0
1
(14 = 25 = 36) e.g. sodium chlorate

Non-centrosymmetric, but all element zero.
(14 = 25 = 36)
Appendix E Real elds, complex elds, and the

analytic signal
In Chapter 2, the real electric field was defined as the sum of a set of discrete frequency
components namely

E(z, t) = 12
E n exp{in t} + c.c. = 12
E n exp{i(n t kn z)} + c.c.
(E1.1)
n
Each term in the summation represents the component of the field in the vicinity of n . The
complex amplitudes E n contain fast-varying spatial terms exp{ikn z}, which are removed
to form the slowly varying complex envelope function E n = E n exp{ikn z}. Both E n and
E n vary slowly in time, or not at all.
In the case of a narrow-band signal in the vicinity of 0 and at a fixed point in space (say
z = 0), we have
exp{i0 t} + c.c. = 1 V (t) + c.c.
E(t) = 12 E(t)
2
(E1.2)
where V (t) is called the analytic signal. For a given real field, V (t) is not uniquely defined,
because its imaginary part can be changed at will without affecting E(t) and, even if V (t)
were known, E(t)would

not be uniquely defined either, because the choice of 0 is strictly
arbitrary. The standard prescription for defining V (t) is

V (t) =
2E()
exp{it}d
(E1.3)
where E()
is the Fourier transform of E(t) given by
1
E()
=
2

E(t) exp{it}dt.
(E1.4)
Notice that the definition of V (t) in Eq. (E1.3) is based solely on the positive frequency
components of E(),
and the factor of 2 has been included to ensure that E(t) = Re{V (t)}.
This process involves no loss of information because the reality of E(t) means that E()
=
E () in Eq. (E1.4).
It follows that the Fourier transform of V (t) is
1 V (t) exp{it}dt = 2E()
( > 0)
V () = 2
(E1.5)
0
( < 0)
which in fact comes immediately from Eq. (E1.3).
234
Appendix F Geometry of the grating pair
This appendix contains the detailed working leading to Eq. (6.27), which summarises the
dispersive properties of the grating pair arrangement of Fig. 6.5.
The coefficient of the quadratic term in Eq. (6.26) is
dgroup
d 2
=
(F1.1)
2
d
d
where p is the perpendicular distance between the gratings, and the final step follows from
Eq. (6.11). To find 2 , we first need to work out the group time delay group of different
frequency groups travelling from the point of incidence A on the first grating G1 in Fig. 6.5
to B on the second grating G2, and thence to C. The precise positions of B and C will of
remains a right angle, and all points C
course depend on the frequency, but the angle ACB
lie on the finishing line AF. By dividing the distance ABC by c, the group time delay is
easily shown to be
2 p =
p(1 + cos( ))
(F1.2)
c cos
where and are the angles of incidence and diffraction. Group velocity dispersion in the
medium (usually air) between the gratings has been ignored. Differentiating Eq. (F1.2), we
have
dgroup
p(sin + sin ) d
=
.
(F1.3)
d
c cos2
d
For first-order diffraction, the standard grating law reads
group =
2 c
=
(F1.4)
I
I
where is the wavelength and I is the ruling spacing, and differentiating this equation
gives
(sin + sin ) =
d
2 c
= 2
.
d
I cos
Substituting Eqs (F1.4) and (F1.5) into Eq. (F1.3) yields
dgroup
4 2 cp
= 2 p = 2 3
d
I cos3
where is the mean value of over the input spectrum. This is Eq. (6.27).
235
(F1.5)
(F1.6)
Appendix G The paraxial wave equation
Maxwells wave equation in a linear isotropic medium reads

2 E +
n2 2 E
=0
c2 t 2
(G1.1)

which follows from Eq. (2.2) if P = 0 (1) E, and the refractive index n = 1 + (1) . We
now look for beam-type solutions by writing a component of E in the form
E = 12 U (x, y, z) exp{i(t kz)} + c.c.
(G1.2)
where k = n/c as usual. The rapid spatial dependence in exp{ikz} implies a beam
travelling basically in the z-direction, while the spatial dependence of U (x, y, z) is assumed
to be relatively slow in comparison.
Substituting Eq. (G1.2) into Eq. (G1.1) yields

U
i
2U
2
(G1.3)
=
T U +
z
2k
z2
where the transverse Laplacian operator governing diffraction is defined by T2 = 2 /x 2 +
2 /y 2 .
In the so-called paraxial approximation, the second (non-paraxial) term on the right-hand
side of Eq. (G1.3) is neglected, leading to the paraxial wave equation
U
i
= T2 U .
z
2k
The Gaussian beam of Eq. (2.47) is an exact solution of this equation.
236
(G1.4)
Appendix H Useful formulae for numerical

simulations
This appendix contains some useful formulae for anyone performing numerical simulations
that involve Fourier transforms between the time and frequency domains.
Discrete Fourier transform (DFT) routines typically perform the operation
Ak
N1

Aj e2 isj k/N
(H1.1)
j =0
where the input complex array Aj (j = 0, . . . , N 1) is replaced on output, and s = 1

offers a choice of sign. This operation can be used to perform either part of the discrete
Fourier transform pair defined by
Ek =
N1

E j e2 ij k/N
j =0
E j =
N1

(H1.2)
Ek e2 ij k/N
k=0
where = 1/N . Note that two successive calls to the routine (with opposite signs s in
the exponents) return the original array multiplied by the array dimension N . In the early
CooleyTukey fast Fourier transform (FFT) algorithm [127], N was restricted to a power
of 2, although a wider range of possible dimensions is available in modern FFT packages.
To use the DFT effectively, the following simple relationships are helpful. The full widths
of the time and frequency windows are T = N t = 2/ and ; = N = 2/t, where
the link between the two domains provided by the second step in these formulae is
t = 2/N.
(H1.3)
Consider a transform from a Gaussian profile in the time domain to the corresponding
Gaussian profile in the frequency domain. We define the full widths at half maximum
intensity in the time domain as t = gt t; the corresponding width in the frequency
domain is = g . The parameters gt and g are the intensity full widths in grid
units. It follows that
t =
237
gt tg
gt g
t
=
=
.
2
2
N
(H1.4)
238
Useful formulae for numerical simulations
The timebandwidth product for a transform-limited pulse is t = K where K depends

on the pulse shape; see Chapter 6. It follows that
gt g = N K.
(H1.5)
This statement of the bandwidth theorem expressed in grid units comes in handy when
testing computer codes.
Appendix I Useful constants

c = 2.997 925 108 ms1
0 = 8.854 188 1012 Fm1 (= C V1 m1 )
0 (= 4 107 ) = 1.256 637 106 H m1
m = 9.109 382 1031 kg
e = 1.602 176 1019 C
h = 6.626 068 1034 J s
= h/2 = 1.054 572 1034 J s rad1
See Appendix A for unit conversion factors.
Velocity of light
Vacuum permittivity
Vacuum permeability
Electron mass
Elementary charge
Plancks constant
239
Answers to problems
Chapter 1
6
1.1 Lcoh = 1.06410

40.0045 = 59.11m.
1.2 Under open circuit conditions, there is no net free charge on either of the plates, so
D (= 0 E + Pdc ) = 0. So, since P is directed upwards in the figure, E = Pdc /(0 )
is directed downwards, and since |E| = Vopen /d, the answer follows immediately.
1.3 Use 2 cos A cos B = cos(A + B) + cos(A B).
1.4 For Eq. (1.19) to be satisfied, we require that n3 3 n2 2 +n1 1 where 1 +2 = 3 .
Write 1,2 = 12 (3 ) where = 2 1 .
The condition then becomes n3 12 (n2 + n1 ) + 12 (n2 n1 ) where = /3 1.
But n3 > n2 n1 for normal dispersion, and it is easy to show that the two conditions
are inconsistent.
1.5 Figure 1.7 (for sum frequency generation) applies to SHG if the k1 and k2 arms are of
equal length. The acute angle of the triangle is cos1 {1.5549/1.6550} = 20.03 and
the angle between the two fundamental beams is twice as large.
1.6 1 is dimensionless; 2 is in m V1 ; and3 is in (m V1 )2 .

(2) A
(2) A
0
1.7 (a) n + n = 1 + (1) + 2 (2) A0 = (1 + (1) ) 1 + 2

= n + n 0.
1+ (1)
(b) The effective change in (1) is 34 (3) A21 and the intensity is 21 nc0 A21 . A similar
procedure to that in part (a) leads to the answer.
Chapter 2

2

d E 1
dI1 1
E 1
2.1
= 2 n1 c0 dz = 12 n1 c0 E 1 ddz
+ c.c.
dz

= 14 1 0 i E 2 E 12 eikz i E 2 E 12 eikz ,
2

d E 2
d E 2
dI2
= 12 n2 c0
= 12 n2 c0 E 2
+ c.c.
dz
dz
dz

1
= 2 0 i E 2 E 12 eikz i E 2 E 12 eikz .
8
Since 2 = 21 , it follows that
240
dI1
dz
2
= dI
dz .
Chapter 2
241
2.2 This is perhaps obvious. If the fundamental field is A exp{t 2 }, the second harmonic field varies as A2 exp{2t 2 }, while the
intensities vary as A2 exp{2t 2 }

and A4 exp{4t 2 }. The energies come from A2 exp{2t 2 }dt = A2 /2
and A4 exp{4t 2 }dt = A4 /4, so the power-law relationship is the same

for energy as for intensity. A similar argument applies for the third harmonic.
2.3 Putting the numbers into Eq. (2.15) gives I2 = 132.3 MW/cm2 . The low-depletion
approximation will be slightly infringed.
2.4 In Eq. (2.13), the phase of the harmonic field is governed by the term ieikz/2 .
is clearly kz/2. Since O

Since i points along OA, AOB
BC = /2, it follows that

O
CB = kz/2 too.

coh )/2}
2.5 The real and imaginary parts of E(z) are, respectively, EC + EC 1cos{k(zL
2

sin{k(zLcoh )/2}
and EC
. Take the roots of the sum of the squares to obtain the result.
2
2.6 You can of course slog out the answer from Eqs (2.41) and (2.42), but it is more
2
2

elegant to use Eq. (2.33) to show that the derivative of A 2 (z) A 1 (z) is zero.
The result reflects the fact that the changes in the number of photons in the signal and
idler beams is the same; see Section 2.4.2.

sin k z/2
1
ikz/2
2.7 A2 (z)e
= A2 (0) cos k z/2 + i 2 k A2 (0) g A1 (0)
,
k /2
A 1 (z)eikz/2 = A 1 (0) cos k z/2 + i
2.8
2.9
2.10
2.11
sin k z/2
2 k A1 (0) g A2 (0)
k /2

where k /2 = (k/2)2 g 2 .
When A 1 (0) = 0 and g k/2.
A 1 (0) = 0.9i.
When A 1 (0) = i A 2 (0), both fields tend to zero at z = according to the equations.
In practice, the fields would drop to a low value, and then grow again from noise.
This is solved by straightforward substitution.
The intersecting chords theorem gives r 2 = (2R )
= 2R where R is the radius
of curvature of the wavefront, r is distance from the axis, and is the bulge in the
wavefront. The relationship between and the phase (r, ) (= r 2 /w2 ) is =
2 /. Combining these relationships, and using zR = w02 / and w2 = w02 (1 + 2 )
R
from Eqs (2.49) and (2.54) yields zR = 1+
2 . The answer follows immediately.
1
2.12 The condition is (2k1 k2 )L = /2. This leads to L = 8|n2n
= 25.77 mm.
1|
2.13 The solution of any second-order ODE requires two boundary conditions, the initial
value and its gradient. And it is the gradient that introduces the other field component
through Eqs (2.28) and (2.29).
2.14 Starting
the origin, the arc length swept out by the right-hand side of the curve
dfrom

is 0 1+
=
tan1 = Gouy . As , the arc length goes to /2, which is the
2
circumference of a semicircle of radius 12 . At the same time, the phasor direction at

the end of the arc is given by the phase of (1 + i )2 = 2Gouy which is also consistent
with a circular path of the same radius.
Answers to problems
242
Chapter 3
2
3.1 For y = 0 and in polar form, Eq. (3.9) reads r cos

+ r sin
= 1, where is the
n2o
n2e
angle relative to the z-axis. But in the geometrical construction for n e ( ), the angle
refers to the direction of the wave vector k, which is perpendicular to the line defining
n e ( ); hence the complementary angle is needed. Equation (3.10) is therefore obtained
by replacing r by n e ( ), and exchanging cos and sin .
3.2 The two allowed waves are polarised in the z-direction (index ne ) and in the y-direction
(index no ). No walk-off occurs.
3.3 Anticlockwise in the first case, and clockwise in the second.
3.4 Set n(
) = ny and sin2 = 1 cos2 . The result follows by straightforward algebra.
3.5 The walk-off angle is 3.18 from Eq. (3.11).
Chapter 4
4.1 m V1 .
4.2 The waves could lie in the x-y plane (perpendicular to the optic axis) with the interaction
based on the d31 coefficient of BBO; see Table 4.4.
4.3 Unit vectors in the o-o and e-e directions are o = (sin , cos , 0) and e =
(cos cos , cos sin , sin ). These enable one to write
Ex = Eo (sin ) + Ee ( cos cos ) and Ey = Eo ( cos ) + Ee ( cos sin ).
Hence the term involving both Eo and Ee in the product Ex Ey (mediated by d36 to
create Pz ) is Eo Ee (cos cos 2). But Pe = Pz sin , so the overall geometrical factor
is 12 sin 2 cos 2. A similar calculation to find the contribution mediated by d14 yields
the same result.
axes rotated 45 clockwise w.r.t. xy, we have x = (x + y )/ 2 and

4.4 In terms of x y
y = (x y )/ 2. Substituting these results into Eq. (4.28) yields
2 2
x 2 (n2
o + r63 V /T) + y (no r63 V /T) = 1
and it follows that the modified semi-axes representing the refractive indices are
no (1 n2o r63 V /T)1/2
= no (1 12 n2o r63 V /T).
4.5 V/4 =
4n3o r63
= 5279 V for the parameters given.
Chapter 5
5.1 If real arithmetic is used, the polarisation along the direction of the field is simply P =
(3)
0 1 E 3 . Along the slanting axes, the field components are Ex = rE and Ey = rE,
Chapter 5
243
where r = 1/ 2. It is now easy to show that Px = 0 s r 3 E 3 and Py = 0 s r 3 E 3

where s = 1 + 2 + 3 + 4 . Since P = r(Px Py ), it follows that P = 0 ( 12 s )E 3 ,
which proves the result. (If complex numbers are used, the replacements P 12 P and
E 12 E should be made throughout.)
5.2 If real arithmetic is used, light is circularly polarised when Ex = E cos t and Ey =
E sin t. In this case,

Px = 0 1 Ex3 + (2 + 3 + 4 )Ey2 Ex = 0 1 E 3 cos3 t + sin2 t cos t
where 1 = 2 +3 +4 (Eq. 5.6) has been used. But cos3 t +sin2 t cos t = cos t,
so there is no third harmonic component in the nonlinear polarisation. Naturally, a
similar result holds for Py .
5.3 E(0) = (2KL )1/2 = 1.066 MV m1 ; so V = 10.66 kV.
5.4 Equation (C3.4b) with i = x reads

1 E x E x E x + 2 E x E y E y + 2 E x E z E z
Px () = 34 0
3 E y E x E y + 3 E z E x E z + 4 E y E y E x + 4 E z E z E x
where all relevant non-zero coefficients from Eq. (5.7) have been included. Since
1 = 2 +3 +4 from Eq. (5.6) (or Eq. C3.5), and 2 = 4 for IDRI from Eq. (C3.6),
the equation can be rewritten

22 E x E x E x + 22 E x E y E y + 22 E x E z E z
3
Px () = 4 0
+3 E x E x E x 3 E y E x E y + 3 E z E x E z
5.5
5.6
5.7
5.8
and the first of Eqs (5.18) follows immediately. The proof of the second equation is
similar.
n I
The final term in Eq. (5.28) changes k = n0c to k = n0c + 2 pk , where Eq. (5.26)
has been used. Hence

n2 Ipk 1 c
n2 Ipk
cn2 Ipk
c
c
vphase = =
1+
1
=
.
=
k
n0
2n0
n0
2n0
n0
2n20
2
2
E

E L
+ E L zL and use IL = 12 nL c0 E L and IS = 12 nS c0 E S . The
Work out E L z
result follows immediately.
The intersecting chords theorem gives (2kL k)k
= 2kL k
= KA2 . It follows
1
1
2
2
that k
= 2 kL and the answer follows from Lcoh = / |k| and
= 2 kL (KA /kL )
kL = 2/L .
This involves some intricate algebra. Start by writing
A(z)
= A + eiz + A eiz
B(z)
= B + eiz + B eiz .
= A + + A , while B = B + e2iL ensures that B(L)

= 0. It also follows
Now A(0)
from from Eq. (5.78) that i A + = r B + , i A = r B , etc. Combining the equations
leads to the answer.
Answers to problems
244
5.9 The profile of the forward beam at the mirror is found by setting z = zm in Eq. (5.95). The
profile of the backward wave at the mirror must be the negative of this (to ensure zero
field at the mirror surface). A backward-travelling Gaussian beam centred at z = zm can
be found by replacing z by (z 2zm ) in Eq. (5.95), and taking the complex conjugate
(to reverse the direction). Setting z = zm in this equation gives the required profile at
the mirror apart from the minus sign, which arose from the phase change at the mirror.
But this corresponds simply to a phase change of the backward beam, and so it does
not affect the conclusion.
Chapter 6
6.1
6.2
6.3
z
dn
z
2 c dn d
z
2 c 2 dn
z
dn
c n + d = c n + d d = c n 2 c d = c n d .
dn 2
d 2 vphase
d 2 vphase
d2n
d2n
= n2c3 d
nc2 d
and d
2 , so the zeros of
2 are only coincident when
d2
d2
dn
d = 0. The suggestion is therefore generally incorrect.
2
0
= ln 2. The
(a) Since the intensity varies as exp{(t/t0 )2 }, it follows that t
2t0
answer to part (a) follows immediately.

(b) = 0.441/t = 0.441 THz; = 2 /c = 1.47 nm.
(c) Equation (6.21) gives maximum chirp at = 1, so d
dt =
221 GHz ps1 .
6.4 If = t/t0 , then sech2 =
4
e2 +2+e2
1
4 t02
ln 2
t02
1
2
2 = 6. This is equivalent
2 when e + e
0, which has the roots e2 = 3 8. This leads to = t/t0 =
to e4 6e2 + 1 =
ln{3 8}.
6.5 34 cm.
6.6 Choose an arbitrary reference line of length s between the gratings, defined by = 0 ,
0)
so that s = p/ cos 0 and = s cos(
. Then
c

d 2
s
d
d 2
group = 2
=
2
+ 2 sin( 0 )
d =0
c
d
d

2
2

d
d
p

+
cos( 0 )
=
.
d
c cos 0 d =0
=0
But
d
d =0
p
c
= 2 I2cos
, so group = c cos
0
2 c
2 I cos 0
cp
= 3 I42 cos
3 .
0
Chapter 7
L
1
7.1 Lspm = 2 n2 Ipk so spm
= 2 31016 1012 = 531.
Care always needs to be taken over the use of centimetres in questions of this kind. All
parameters must of course be in the same units.
Chapter 8
245
7.2 This is straightforward algebra.

n I
7.3 The final term in Eq. (7.18) with L = Lspm changes k = n0c to k = n0c + 2 pk .
Hence

n2 Ipk 1 c
n2 Ipk
cn2 Ipk
c
c
vphase = =
1+
1
=
.
=
k
n0
2n0
n0
2n0
n0
2n20
7.4 For a Gaussian intensity profile I = Ipk exp{t 2 }, it iseasy to show that the steepest
1/2
e
8 ln 2
gradient occurs at t = (2)1/2 where dI
= 1.43
dt =
t
t and t is the

intensity FWHM given by
t
=
(4
ln
2)/
.

ct
cT
4.4 t where T = 2/ is the
It follows that Lshock =

=
Lspm
1.43n2 Ipk
2 n2 Ipk
optical period.
7.5 This question is relevant to the phenomenon of carrier-wave shocking, and to the use
of two-colour pumping in high harmonic generation experiments.
L|vphase vgroup |
7.6 The time difference is t
where L is the distance of travel and v is
=
v 2
1
the mean velocity. Setting t = 4 T = /4c yields
2
800109 (1.92108 )2
L = v
= 12.3 m, where values of v and v have been
=
8
6
4c v
4 310
210
estimated from Fig. 6.2.

This is a very short distance, and it shows that the carrier-envelope phase is scrambled
extremely easily. The situation inside a mode-locked laser is more complicated than
this, but the result shows that the slip from one transit to the next will be very large
and therefore essentially random. However, this in no way affects the efficacy of the
technique for CEP stabilisation described in Section 7.10.
Chapter 8
(2)
(1) (1)
8.1 The formula 21 = a2 a1 is verified if the derivative of both sides can be shown
to be the same. (The initial conditions match, because all three parameters are initially
(2)
(1) (1)
(1) (1)
zero.) It is therefore necessary to show that 21 = a 2 a1 + a2 a 1 .
From Eqs (8.7) and (8.15), we know that

i
(2)
(2)
(1)
(1)
21 = i21 21
V20 01 20 V01

i
(1)
(1)
(0)
a 2 = i2 a2 V20 a0

i
(1)
(1)
(0)
a 1 = +i1 a1 + V01 a0

(1)
(1) (0)
and the answer follows immediately once it is recognised that 20 = a2 a0 , etc.

8.2 For j = k, take the complex conjugate, and satisfy yourself that the two expressions
are the same. Remember that j , k, and n, n are all dummy indices, and so the pairs can
be exchanged at will. The conclusion applies equally when j = k, although the issue
would only arise in that case if the matrix elements were complex.
Answers to problems
246
8.3 The contraction to d14 for SFG is valid only when the Kleinmann symmetry condition
applies, and in this limit the two coefficients become identical; see Section 4.3.
Chapter 9
9.1 This is similar to Problem 1.7. The sequence is
!

i
n
i

1 + i =
(1 + ) 1
1
k=
=
c
c
1 +
c
2n2
where n = 1 + . The approximation is good provided 2n2 .

= i which leads to
9.2 On resonance, 20
D
2
2
2

AL E S E L
AS E L E S
E L
E S
and
=
=
z
nL
z
nS
where A is positive and contains constants. It follows that
2 2
nL E L E L
nS E S E S

= A E S E L =
.
S z
L z
IL
1
1
S
ES
EL
But I
z = 2 c0 nS ES z + c.c. and z = 2 c0 nL EL z + c.c., where IS and IL are
1 IL
S
the Stokes and laser energies, respectively. So 1S I
z = L z , which completes the
proof.
9.3 In the first case, the lineshape function is very narrow, and the Lorentzian factor in the
denominator of the integral can be neglected. The integral is then unity (because the
lineshape function is normalised), and the answer follows trivially.
In the opposite limit, the lineshape profile is much broader than the Lorentzian factor,
and so gi ( ) can be replaced bygi (0). The integral is now gi (0)/ and the answer
follows directly.

9.4 We have u = ab(1 cos ;R t) and 1 + w = 1 a 2 b2 cos ;R t = b2 1 cos ;R t .
Hence 1+uw = ab = ;R is independent of t, and the motion projected onto the u-w
plane is a line slanted with respect to the 3-axis at an angle of tan1 {/;R }.

dn
dn
9.5 d
= d
. The algebra is made much easier by the fact that the

=
p
10
p =0
derivative is to be evaluated at p = 0.
Chapter 10
10.1 If U = Ex
dU
|x| , dx
= E +
|x|2
= 0 when |x| =
|E|
|E|1/2 I 1/4 ,
because intensity is proportional to the square of the field. While this may be an
interesting exercise, the result has little bearing on what actually happens under strongfield ionisation.
Chapter 10
247
10.2 Elliptically polarised light can always be reduced to two orthogonal electric field
components in quadrature. However, the quasi-classical model shows that recollisions
only occur in alternate quarter-cycles, so the recollision can never occur in both
quadratures, let alone in both quadratures at the same time.
10.3 Consider the quarter-cycle /2 t0 < 0 in which the field is approaching a
maximum. From Eq. (10.5), the electron velocity is proportional to V (t) = sin t0
sin t (t > t0 ). And since sin t0 < 0, V (t) is a sine function with a negative shift.
t
This in turn means that V (t )dt , which controls the net displacement (see Eq. 10.6),
t0
can occur.
is negative for all t > t0 , so no recollision
2
eA
e
2I
10.4 The factor in Eq. (10.6) is m
2 = 4 2 c2 m
c0 = 2.75 nm. The peak value of X in
Fig. 10.2 is about 2, so the maximum excursion is maybe 5.5 nm.
10.5 If the statement is true, it follows from trigonometry that the derivative of E(t) at
cos t0
the time of ionisation (= A sin t0 ) must equal A cos t(tRR A
. This equality is
t0 )
confirmed by Eq. (10.7)
10.6 The fundamental photon energy in eV is hc/e = 1.55 eV. The ionisation energy
therefore corresponds to 15.85 harmonic units. Equation (10.10) gives 244.4 for
the recollision kinetic energy in harmonic units, so the maximum harmonic is
int{260.3} = 260, and the minimum wavelength is 800 260.3 = 3.07 nm. This
is close to the value achieved by Chang [43], although the nominal intensity used in
the classical model can hardly be expected to yield an accurate answer.
Further Reading
R.W. Boyd, Nonlinear Optics (3rd edn), Elsevier, Amsterdam, 2007.

A comprehensive treatment of the subject at a more advanced level, presented in a tutorial
style. The third edition is in SI units; earlier editions used Gaussian units.
G.S. He and S.H. Liu, Physics of Nonlinear Optics, World Scientific, Singapore, 1999.
A wide-ranging treatment of nonlinear optics at an advanced level, with an especially
generous coverage of third-order processes and stimulated scattering. The style is more that
of a review than a tutorial text; the book contains a large number of research references.
G. Agrawal, Nonlinear Fibre Optics, Academic Press, New York, 1989.
A clear and accessible treatment of nonlinear optics in optical fibres.
N. Bloembergen, Nonlinear Optics, Benjamin, New York, 1965.
A classic by one of the leading pioneers of nonlinear optics, albeit written at a very early
date.
A.C. Newell and J.V. Moloney, Nonlinear Optics, Addison Wesley, New York, 1992.
A full-length text on nonlinear optics with a slight bias towards aspects of the subject
relevant to optical communications.
A. Yariv, Quantum Electronics (3rd edn), John Wiley, New York, 1988.
Chapters 1419 contain a very useful summary of nonlinear optics.
P.N. Butcher and D. Cotter, The Elements of Nonlinear Optics, Cambridge University Press,
Cambridge, 1990.
An authoritative account of the fundamentals of the subject.
H. Zernike and J.E. Midwinter, Applied Nonlinear Optics, Dover, New York, 2007 (reprint).
A valuable account of nonlinear optics as it was in the early 1970s, with a tutorial flavour
and at an introductory level.
R.L. Sutherland, Handbook of Nonlinear Optics (2nd edn), Marcel Dekker, New York,
2003.
An indispensable handbook of nonlinear optics for the specialist.
V.G. Dmitriev, G.G. Gurzadyan and D.N. Nikogosyan, Handbook of Nonlinear Optical
Crystals (3rd edn), Springer, Berlin, 1999.
An indispensable repository of data on nonlinear optical crystals.
248
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
249
R.W. Boyd, Nonlinear Optics (3rd edn), Elsevier, Amsterdam, 2007.

J. Kerr, Phil. Mag. (4th Series), 50 (1875) 337.
F. Pockels, Abr. Gott., 39 (1894) 1.
P.A. Franken, A.E. Hill, C.W. Peters, and G. Weinreich, Phys. Rev. Lett., 7 (1961)
118.
P.D. Maker, R.W. Terhune, M. Nisenoff, and C.M. Savage, Phys. Rev. Lett., 8 (1962)
21.
P.A. Franken and J.F. Ward, Rev. Mod. Phys., 35 (1963) 35.
M. Bass, P.A. Franken, J.F. Ward, and G. Weinreich, Phys. Rev. Lett., 9 (1962) 446.
M. Ebrahim-Zadeh, Phil. Trans. Roy. Soc. Lond. A, 263 (2003) 2731.
J.A. Armstrong, N. Bloembergen, J. Ducuing, and P.S. Pershan, Phys. Rev., 127
(1962) 1918.
J.A. Giordmaine and R.C. Miller, Phys. Rev. Lett., 14 (1965) 973.
E.J. Lim, M.M. Feyer, R.L. Byer, and W.J. Kozlovsky, Electron. Lett., 25 (1989)
731.
D.E. Spence, P.N. Kean, and W. Sibbett, Opt. Lett., 16 (1991) 42.
P.B. Corkum and M. Krausz, Nature Physics, 3 (2007) 381.
S.L. McCall and E.L. Hahn, Phys. Rev. Lett., 18 (1967) 908; S.L. McCall and
E.L. Hahn, Phys. Rev. 183 (1969) 457.
S.E. Harris, Physics Today, 507, (July, 1997) 36.
M. Fleischhauer, A. Imamoglu, and J.P. Marangos, Rev. Mod. Phys., 77 (2005) 633.
S.E. Harris, Phys. Rev. Lett., 62 (1989) 1033; A. Imamoglu and S.E. Harris, Opt.
Lett., 14 (1989) 1344.
Z. Dutton, N.S. Ginsberg, C. Slowe, and L. Vestergaard Hau, Europhys. News
35 (2) (2004) 33.
Nature Photonics 2 (8) (August 2008), special issue on slow light.
A. Yariv, Quantum Electronics (3rd edn), John Wiley, New York, 1988.
M. Ebrahim-Zadeh, Optical parametric devices, in Handbook of Laser Technology
and Applications, IOP Publishing, Bristol, 2003, 1347; Continuous-wave optical
parametric oscillators, in Handbook of Optics VI, Optical Society of America,
McGraw-Hill, New York, 2010.
G.H.C. New and J.F. Ward, Phys. Rev. Lett., 19 (1967) 556; J.F. Ward and G.H.C.
New, Phys. Rev., 185 (1969) 57.
G.C. Bjorklund, IEEE J. Quantum Electron., QE-11 (1975) 267.
J.F. Nye, Physical Properties of Crystals, Oxford University Press, Oxford, 1985.
W.G. Cady, Piezoelectricity (2 vols.), Dover, New York, 1964.
V.G. Dmitriev, G.G. Gurzadyan, and D.N. Nikogosyan, Handbook of Nonlinear
Optical Crystals (3rd edn), Springer, Berlin, 1999.
R.W. Boyd, Nonlinear Optics (3rd edn), Elsevier, Amsterdam, 2007, p. 51.
A. Yariv and P. Yeh, Optical Waves in Crystals, Wiley Interscience, New York, 2003.
M. Born and E. Wolf, Principles of Optics (7th edn), Cambridge University Press,
Cambridge, 1999.
250
References
[30] H. Zernike and J.E. Midwinter, Applied Nonlinear Optics, Dover, New York, 2007,
p. 64.
[31] P.D. Maker and R.W. Terhune, Phys. Rev., 137 (1965) A801.
[32] P.L. Kelley, Phys. Rev. Lett., 15 (1965) 1005.
[33] R.R. Alfano and S.L. Shapiro, Phys. Rev. Lett., 24 (1970) 592.
[34] P.G. Drazin and R.S. Johnson, Solitons: An Introduction (2nd edn) Cambridge
University Press, Cambridge, 1989.
[35] A. Barthelemy, S. Maneuf, and C. Froehly, Opt. Commun., 55 (1985) 201.
[36] J.S. Aitchison et al., Electron. Lett., 28 (1992) 1879.
[37] G. Fibich and A.L. Gaeta, Opt. Lett., 25 (2000) 335.
[38] E. J. Woodbury and W. K. Ng, Proc. IRE, 50 (1962) 2367.
[39] G.E. Eckhardt et al., Phys. Rev. Lett., 9 (1962) 455.
[40] A.P. Hickman et al., Phys. Rev. A, 33 (1986) 1788.
[41] G.S. McDonald et al., Opt. Lett., 19 (1994) 1400.
[42] E. Takahashi et al., Opt. Express, 15 (2007) 2535.
[43] P. Tournois, Opt. Commun., 140 (1997) 245.
[44] R.Y. Chiao, C.H. Townes, and B.P. Stoicheff, Phys. Rev. Lett., 12 (1964) 592.
[45] C.F. Quate, C.D.W. Wilkinson, and D.K. Winslow, Proc. IEEE, 53 (1965)
1604.
[46] A. Kobyakov, M. Sauer, and D. Chowdhury, Adv. Opt. and Phot., 2 (2010) 1.
[47] M.J. Damzen et al., Stimulated Brillouin Scattering: Fundamentals and Applications,
Taylor & Francis, London, 2003.
[48] G.S. He and S.H. Liu, Physics of Nonlinear Optics, World Scientific, Singapore,
1999.
[49] D.M. Pepper, Opt. Eng., 21 (1982) 156.
[50] M. Gower, Optical Phase Conjugation, Springer-Verlag, Berlin, 1994.
[51] A. Brignon and J-P. Huignard (Eds.), Phase-Conjugate Laser Optics, Wiley Interscience, New York, 2003.
[52] J.I. Gersten, Phys. Rev. A, 21 (1980) 1222.
[53] J.T. Manassah, R.R. Alfano, and M. Mustafa, Phys. Lett., A 107 (1985) 305.
[54] A.L. Gaeta, Phys. Rev. Lett., 84 (2000) 3582.
[55] I. Ilev et al. Appl. Opt., 35 (1996) 2548.
[56] J.M. Dudley and J.R. Taylor (Eds), Supercontinuum Generation in Optical Fibres,
Cambridge University Press, Cambridge, 2010.
[57] E.B. Treacy, IEEE J. Quantum Electron., QE-5 (1969) 454.
[58] O.E. Martinez, J.P. Gordon, and R.L. Fork. J. Opt. Soc. Am. A, 1 (1984) 1003.
[59] R.W. Boyd, Nonlinear Optics (3rd edn), Elsevier, Amsterdam, 2007, Chapter 13.
[60] T.K. Gustafson et al. Phys. Rev., 177 (1969) 306.
[61] G.P. Agrawal, Nonlinear Fibre Optics (2nd edn), Academic Press, New York, 1989,
Chapter 4.
[62] L.F. Mollenauer, R.H. Stolen, and J.P. Gordon, Phys Rev. Lett., 45 (1980) 1095.
[63] L. F. Mollenauer and K. Smith, Opt. Lett., 13 (1988) 675.
[64] L.F. Mollenauer and J.P. Gordon, Solitons in Optical Fibres: Fundamentals and
Applications, Academic Press, New York, 2006.
[65] J.R. Taylor (Ed.), Optical Solitons Theory and Experiment, Cambridge University
Press, Cambridge, 1992.
[66] F. DeMartini, C.H. Townes, T.K. Gustafson, and P.L. Kelley, Phys. Rev., 164 (1967)
312.
[67] G. Rosen, Phys. Rev., 139 (1965) A539.
[68] R.G. Flesch, A. Pushkarev, and J.V. Moloney, Phys. Rev. Lett., 76 (1996) 2488;
L. Gilles, J.V. Moloney, and L. Vasquez, Phys. Rev. E, 60 (1999) 1051.
References
251
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]
[91]
[92]
[93]
[94]
[95]
[96]
[97]
[98]
[99]
[100]
[101]
[102]
[103]
[104]
[105]
[106]
[107]
[108]
[109]
[110]
[111]
[112]
[113]
S.B.P. Radnor et al., Phys. Rev. A, 77 (2008) 033806.

R.A. Fisher, P.L. Kelley, and T.K. Gustafson, Appl. Phys. Lett., 14 (1969) 140.
A.M. Johnson, R.H. Stolen, and W.H. Simpson, Appl. Phys. Lett., 44 (1984) 729.
J.W.G. Tisch (Imperial College Attosecond Laboratory), private communication.
J. Comly and E. Garmire, Appl. Phys. Lett., 12 (1968) 7.
W.H. Glenn, IEEE J. Quantum Electron., QE- 5 (1969) 284.
J.C.A. Tyrrell, P. Kinsler, and G.H.C. New, J. Mod. Opt., 52 (2005) 973.
A. Dubietis et al., Opt. Commun., 88 (1992) 433.
I. N. Ross et al., Opt. Commun., 144, (1997) 125.
D. Strickland and G. Mourou, Opt. Commun., 56 (1985) 219.
I.N. Ross, P. Matousek, K. Osvay, and G.H.C. New, J. Opt. Soc. Am. B, 19 (2002)
2945.
E.J. Grace, C. Tsangaris, and G.H.C. New, Opt. Commun., 261 (2006) 225.
Y. Tang et al., Opt. Lett., 33 (2008) 2386.
C.L. Tsangaris, PhD thesis, Imperial College London, 2005.
D.J. Bradley and G.H.C. New, Proc IEEE, 62 (1974) 313.
E.B. Treacy, J. Appl. Phys., 42 (1971) 3848.
D.J. Kane and R. Trebino, IEEE J. Quantum Electron., QE-29 (1993) 571.
C. Iaconis and I.A. Walmsley, Opt. Lett., 23 (1998) 792.
Rick Trebino, Frequency-Resolved Optical Gating: The Measurement of Ultrashort
Laser Pulses, Kluwer Academic, Dordrecht, 2000.
I.A. Walmsley and V. Wong, J. Opt. Soc. Am. B, 13 (1996) 2453.
I.A. Walmsley and C. Dorrer, Adv. Opt. and Photonics, 1 (2009) 308.
C. Dorrer and I. Kang, J. Opt. Soc. Am. B, 25 (2008) A1.
H. R. Telle et al., Appl. Phys. B 69 (1999) 327.
D.J. Jones et al., Science, 288 (2000) 635.
T. Brabec and F. Krausz, Rev. Mod. Phys., 72 (2000) 545.
A. Messiah, Quantum Mechanics, Vol.1, Dover, New York, 2003.
A. Rae, Quantum Mechanics (5th edn), Taylor & Francis, London, 2008, p. 131.
P.N. Butcher and D. Cotter, The Elements of Nonlinear Optics, Cambridge University
Press, Cambridge, 1990.
J.F. Ward, Rev. Mod. Phys., 37 (1965) 1.
M. Takatsuji, Phys. Rev. B, 2 (1970) 340.
B.J. Orr and J.F. Ward, Mol. Phys., 20 (1971) 513.
D.C. Hanna, M.A. Yuratich, and D. Cotter, Nonlinear Optics of Free Atoms and
Molecules, Springer-Verlag, Berlin, 1979.
L. Allen and J.H. Eberly, Optical Resonance and Two-Level Atoms, Dover,
New York, 1988.
R.P. Feynman, F.L. Vernon Jr. and R.W. Hellwarth., J. Appl. Phys., 28 (1957) 49.
M.D. Crisp, Phys. Rev. A, 1 (1970) 1604.
K.J. Boller, A. Imamoglu, and S.E. Harris, Phys. Rev. Lett., 66 (1991) 2593.
L. Vestergaard Hau, S.E. Harris, Z. Dutton, and C. H. Behroozi, Nature, 397 (1999)
594.
C. Liu, Z. Dutton, C.H. Behroozi, and L. Verstergaard Hau, Nature, 409 (2001) 490.
R.L. Sandberg et al., Phys. Rev. Lett., 99 (2007) 098103.
E. Goulielmakis et al., Science, 320 (2008) 1614
S. Baker et al., Science, 312 (2006) 424.
A.L. Cavalieri et al., Nature, 448 (2007) 1029.
M. Krausz and M. Ivanov, Rev. Mod. Phys., 81 (2009) 163.
J.G. Eden, Prog. Quantum Electron., 28 (2004) 197.
A. McPherson et al. J. Opt. Soc. Am B, 4 (1987) 595.
References
252
[114]
[115]
[116]
[117]
[118]
[119]
[120]
[121]
[122]
[123]
[124]
[125]
[126]
M. Ferray et al., J. Phys. B, 21 (1988) L31.

J.J. Macklin, J.D. Kmetec, and C.L. Gordon III, Phys. Rev. Lett., 70 (1993) 766.
A. LHuillier et al., Phys. Rev. Lett., 70 (1993) 7674.
Z. Chang et al., Phys. Rev. Lett., 79 (1997) 2967.
E. Seres, J. Seres, and C. Spielmann, Appl. Phys. Lett., 89 (2006) 18919.
P.B. Corkum, Phys. Rev. Lett., 71 (1993) 1994.
K.J. Schafer et al., Phys. Rev. Lett., 70 (1993) 1599.
M. Lewenstein et al., Phys. Rev. A, 49 (1994) 2117.
G.L. Yudin and M. Yu. Ivanov., Phys. Rev. A, 64 (2001) 013409.
L.E. Chipperfield, private communication.
L.E. Chipperfield et al., Laser and Photonics Reviews, in press.
G. Sansone et al., Science, 314 (2006) 443.
R.L. Sutherland, Handbook of Nonlinear Optics (2nd edn), Marcel Dekker, New York
2003.
[127] J.W. Cooley and J.W. Tukey, Math. Comput., 19 (1965) 297.
Index
AC Kerr effect. See optical Kerr effect

acoustic waves. See acousto-optics
diffraction of light by, 98100
acousto-optics, 95102
programmable dispersive filter, 95102
ammonium dihydrogen phosphate (ADP), 47, 68,
76, 232
analytic signal, 120, 234
angular wave number, 5, 9, 20, 185
anisotropic media, 4561
light propagation in, 4860
tables, 4748
anomalous dispersion, See dispersion, anomalous
antenna picture, See optical antenna
anti-Stokes gain, 9091, 187188
anti-Stokes generation, 9091
area theorem, 194197
asymmetry
in few-cycle pulses, 217218
in nonlinear response, 1,4, 67, 217218
attosecond pulse generation, 16, 215217
AutlerTownes splitting, 200
bandwidth theorem, 122
beta-barium borate (BBO), 17, 47, 61, 68, 233
biaxial media, 46, 47, 5152, 5760 , 7072
positive and negative, 60
birefringence, 67, 14, 28, 4561, 95
birefringent phase matching, 67, 13, 16, 28, 37, 46,
48, 6870, 100102
Bloch equations. See optical Bloch equations
Bloch sphere, 191195
Bloch vector, 191194
Bragg regime, 100
carrier-envelope phase, 160162, 216
centre of symmetry, 3, 7, 48, 95;
See also inversion symmetry, centrosymmetric
medium
centrosymmetric medium, 67, 12, 4548
chirping, 122123, 130, 121
in OPCPA, 151153
of a Gaussian pulse, 120121
circularly-polarised light
in intensity-dependent refractive index, 8283
in third harmonic generation, 79, 117, 205206
253
in wave plates, 45
coherence length, 17, 6, 11, 23 , 2628, 118
coherent effects, 14, 17, 189203
constitutive relation, 5152
contracted suffix notation, 6566
contraflow Hermitian parametric coupling, 107
conversion efficiency, 56, 23, 148, 204205,
210211
coupled-wave equations, 2122, 2931
general form, 220
for stimulated Brillouin scattering, 104108
for stimulated Raman scattering, 8889
critical power (for self-focusing), 86
cross-phase modulation, 13, 81
crystal symmetry, 4548 , 67
crystal classes, 47
crystallographic axes, 48, 67, 70
cubic crystals, 7879
DC Kerr effect, 23, 12, 77, 8081, 230

DC-induced second harmonic generation, 13, 230
density matrix
advantages, 166167
definition, 165167
interaction picture, 163, 166
properties, 167
density matrix approach
compared to probability amplitude approach, 168,
180184
density operator, 165
dielectric constant, 2, 5152
inverse, 72
matrix, 51, 63, 170
tensor, 51,
difference-frequency generation (DFG), 1112, 3135
diffraction
of light by acoustic waves, 98100
diffraction grating pair, 125127, 235
dipole approximation, 163
dipole moment, 2, 3, 50, 167
dipole operator, 163167
Dirac notation, 163165
Index
254
dispersion, 6, 21, 26, 63, 67, 91, 101, 184185

See also group velocity dispersion
of acoustic waves (ultrasound), 94, 97, 103
anomalous, 118, 185, 200
normal, 10, 26, 41, 43, 118, 185, 200, 213
in phase and group velocity, 116119
normal vs anomalous, 118119
dispersion curve, 9395, 9798, 90, 117
dispersion diagram, 94, 117, 147
dispersion distance, 121122, 130, 136139
dispersion parameter, 119
dominant path, 174180
double refraction. See birefringence
electric field definitions. See field definitions
electromagnetically induced transparency (EIT), 17,
198202
electrostriction, 9697, 103
electrostrictive constant, 98, 104
elliptically-polarised light
in intensity-dependent refractive index, 8283
energy eigenstate, 164, 166
expectation value, 164, 167
extraordinary refractive index. See refractive index,
extraordinary
extraordinary wave, 5254, 6870
field definitions, 20
filamentation, 84, 86
focused beams, 3743
harmonic generation in, 4143
four-wave processes, 76
Fourier transform
discrete, 237
Franken, P, 34
frequency comb, 117, 160162
full permutation symmetry, See permutation symmetry
Gaussian beam, 3843, 44, 109
harmonic generation in, 4143
Gaussian pulse, 120121, 130
Gouy phase, 3843 , 213
grating pair. See diffraction grating pair
group time delay, 118
group time delay dispersion, 118119, 148149
group velocity, 116119
group velocity dispersion (GVD), 37, 86, 102,
116122, 125132, 135149
in second harmonic generation, 146149
positive vs. negative, 119
half-wave plate, See wave plates
Heisenberg picture, 163
high-harmonic generation, 14, 1617, 204218
attosecond pulses from, 215217
classical model, 206214

discrete harmonics in, 217218
spectrum, 211213
half-cycle bursts, 215218
long and short trajectories, 208212, 213
phase matching, 213
recollision energy, 209211
spectral plateau, 204205, 211213
three-step model, 205
Hookes Law, 1
hyperbolic secant pulse, 124125 130
idler wave, 32, 3537
index ellipsoid, 5254
equation of, 52, 7273
index matching. See phase matching
index surfaces,
biaxial media, 5860
uniaxial media, 5455
inhomogeneous broadening, 192, 195198
intensity-dependent refractive index (IDRI), 8287,
132146
nonlinear coefficient definition, 78, 230231
perturbation theory formula, 180184
interaction picture, 163, 166
intrinsic permutation symmetry, See permutation
symmetry, intrinsic
inverse dielectric constant, 72
inversion symmetry, 3, 67, 4548, 95, 177
isotropic media,
(optical isotropy) 47, 45
(structural isotropy) 7879, 231
KDP. See potassium dihydrogen phosphate
Kerr cell, 8081, 117
Kerr constant, 80
Kerr effect. See DC Kerr effect; optical Kerr
effect; etc.
Kerr, J, 2
Kerr-lens mode-locking, 16
Kleinmann symmetry, 6667, 232233,
KTP. See potassium titanyl phosphate
laser, invention of, 3
lasing without inversion, 17, 199
LBO. See lithium triborate
linear response, 14
linear susceptibility, 2, 21, 51, 6263, 222223
frequency dependence, 184185
matrix, 15
resonance, 184185
tensor, 15
lineshape function, 185
inhomogeneous, 192, 195
lithium niobate (LiNbO3 ), 17, 35, 47, 68, 148, 233
Index
255
lithium triborate (LBO), 47, 68, 70, 150155, 232

second harmonic generation in, 7072,
local time transformation, 129130
low-depletion approximation, 2224
ManleyRowe relations, 2324, 2931, 6565
Maxwells equations, 1921, 4850
mismatch parameter, 6, 21, 24, 30, 35, 223224, 99,
220
modified field variables, 31
momentum matching, 10
multi-frequency Raman generation, 91
non-centrosymmetric medium, 67, 4548
non-centrosymmetric crystals,
tables, 4748, 232233
non-collinear beams, 910, 78, 9091, 94, 103108,
111115, 153156, 228
non-instantaneous response, 221224
nonlinear coefficients,
DC-induced second harmonic generation, 230
DC Kerr effect, 77
definition, 225231
general second-order process, 228229
general third-order process, 229230
in non-centrosymmetric crystals, 232233
intensity dependent refractive index, 78, 180184,
230231
matrix form, 15
optical Kerr effect, 78, 230
optical rectification, 64, 179, 229
Pockels effect, 64, 229
second harmonic generation, 64, 172177, 221,
227229
second-order, 6368, 172179, 225229
sum-frequency generation, 6364, 177179,
224230
tensor form, 15
third harmonic generation, 77, 117, 179180, 230
third-order, 7778, 179184, 229230
time and frequency dependence, 221224
non-linear propagation distance, 8486
non-linear response, 4
origin of, 14
nonlinear Schrdinger equation, 85, 136
normal dispersion, See dispersion, normal
normal surface, See index surfaces
numerical simulation
useful formulae for, 237
OPCPA. See optical parametric chirped pulse
amplification
optic axes (biaxial medium), 5960
optic axis (uniaxial medium), 5256
optical antenna, 4, 2526, 41, 92, 98, 112, 196, 206,
213
optical bistability, 16
optical Bloch equations, 189190

optical Kerr effect, 16, 78, 8182, 230
optical nonlinearity
origin of, 14, 162
optical parametric amplification, 1112, 3236
tuning characteristics, 3536
optical parametric chirped pulse amplification,
150156
bandwidth enhancement in, 151154
in three dimensions, 154156
optical parametric oscillator, 11, 15, 3637
singly- vs doubly resonant, 37
synchronously pumped, 37
tuning characteristics, 37
optical phase conjugation, 109115
by four-wave mixing, 111114
by stimulated Brillouin scattering, 114115
optical pulse compression. See pulse compression
optical rectification, 79, 17, 74
and the Pockels effect, 74
non-linear coefficient definition, 64, 229
optical shock formation, 139141, 162
optical solitons, 13, 16
order, 85, 136
spatial, 8387, 118
temporal, 135139
optics of anisotropic media, 4860
ordinary refractive index. See refractive index,
ordinary
ordinary wave, 5254
orthonormality, 164
parametric anti-Stokes generation, 90, 188189
parametric gain coefficient, 32
parametric (vs. non-parametric) processes, 9091,
188189
parametric up-conversion, 30
parametric down-conversion, 31
paraxial wave equation, 38, 44, 84, 236
periodic poling, 16, 28, 35
permutation symmetry
full, 65, 74
intrinsic, 6465, 178, 225227
perturbation Hamiltonian, 163
perturbation matrix element, 164
perturbation theory. See time-dependent perturbation
theory
phase conjugation. See optical phase conjugation
phase matching, 56, 913, 1618, 2628, 3237,
4144, 6870, 8991, 99102, 146147, 151,
188189
See also birefringent phase matching, quasi-phase
matching
in anti-Stokes generation, 9091, 188189
in high-harmonic generation, 213
Index
256
phase matching (cont.)

type I , 6870
type II, 70, 219
phase mismatch, 6
phase velocity, 116119
phasor diagram, 2627 , 4243
photoelastic effect, 95
photon flux, 31
physical axes (of a crystal). See principal dielectric
axes
physical constants, 239
pi pulse, 196
piezoelectric effect, 4546, 97,
inverse, 97
piezo-optic effect, 95
Pockels cell, 73
Pockels effect, 9, 7274
and optical rectification, 74, 229
Pockels, Friedrich, 2
point symmetry group, 46
polarisation
non-instantaneous response, 221224
of light, 50
of matter, 2, 50
polarisation expansion, 4, 1215, 17, 221224
polarisation gating, 217
poling period, 28, 35
ponderomotive energy, 207
potassium dihydrogen phosphate (KDP), 47, 68,
6870, 73, 184, 232
potassium titanyl phosphate (KTP), 47, 68, 232
Poynting vector, 55, 58, 102
principal dielectric axes, 48, 51
biaxial medium, 59, 68
principal dielectric constants, 51
prism pair/quadruplet, 125127
probability amplitude approach
compared to density matrix approach, 168, 180184
pulse area, 194196
pulse compression, 125127, 141146
pulse propagation equation, 128132
pump wave, 32, 3537
quarter-wave plate, See wave plates
quarter-wave voltage, 74, 76
quasi-ordinary wave, 59
quasi-phase matching, 16, 2628, 3536
Rabi flopping, 193
Rabi frequency, 193194, 199
Rabi, Isidor, 17
Raman gain, 8889, 186188
Raman gain coefficient, 88
Raman laser, 87
Raman-Nath regine, 99
Raman resonance, 186188
Raman waves, 9195

dispersion, 91
Rayleigh length, 38, 41, 43
refractive index, 2, 21
extraordinary, 5255
of the extraordinary wave, 54
ordinary, 5255
repeated suffix convention, 62, 64
representation (quantum mechanical), 163
resonance
in the linear susceptibility, 184185
Raman, 186188
resonant effects, 14, See Chapter 9
rotating wave approximation, 190
Schawlow, AL, 57
Schrdinger equation, 2, 163167, 162
Schrdinger picture, 163
second harmonic generation, 37, 2130
group velocity dispersion in, 146149
in a focused beam, 4143
in KDP, 6870
in LBO, 7072
second-order processes. See Chapter 2
228229
self-focusing, 16, 8387, 117
critical power for, 86
self-induced transparency, 14, 17, 192198, 202
inhomogeneous broadening in, 195198
self-phase modulation, 16, 13, 8387 , 117, 132146
characteristic distance, 133, 162
self-steepening, 13, 117, 139141
signal wave, 32, 3537
slow light, 17, 198202
soliton order, 85, 136
state vector, 163
state vector notation. See Dirac notation
stimulated Brillouin scattering, 102109
stimulated Raman scattering, 8795, 186188
stochastic effects, 166167
Stokes wave, 87
strong-field regime, 204
sum-frequency generation, 910, 3035
non-linear coefficient definition, 6364, 225230
supercontinuum generation, 117
symmetry
of nonlinear medium, 67, 4548
third harmonic generation, 13, 24
nonlinear coefficient definition, 77, 117, 179180,
230
in a focused beam, 4143
Index
257
third-order processes. See Chapter 3

1213, 229230
three-wave processes, 2937
time-bandwidth product, 122
time-dependent perturbation theory, 168184
first-order, 173177
introduction to, 168168
second-order, 172179
third-order, 179184
Treacy, EB, 125
two-level atom, 165, 167, 189
two-pi pulse, 195
type I phase matching, 6870
type II phase matching, 70, 219
ultrashort pulse diagnostics, 157160
ultrasound, 95
diffraction of light by, 98100

uniaxial media, 47, 5157
positive and negative, 52, 54
unit cell, 48
unit conversion factors, 219220
units
SI vs. Gaussian, 218220
virtual levels, 170176, 179184
walk-off, 5556, 58, 61
angle, 56
wave-normal ellipsoid. See index ellipsoid
wave plates, 5657
zero-dispersion wavelength, 119, 130

Introduction To Nonlinear Optics

Uploaded by

Copyright:

Available Formats

Introduction To Nonlinear Optics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction To Nonlinear Optics

Uploaded by

Copyright:

Available Formats

What are some of the main topics covered in the textbook?

What are some of the main topics covered in the textbook?

What are some of the advanced topics discussed in specialist chapters?

What are some of the advanced topics discussed in specialist chapters?

This page intentionally left blank

Introduction to Nonlinear Optics

Introduction to Nonlinear Optics

cambridge university press

Introduction and preview

4 Nonlinear optics in crystals

Introduction and preview

5 Third-order nonlinear processes

5.2.6 Intensity-dependent refractive index

6 Dispersion and optical pulses

Introduction and preview

7 Nonlinear optics with pulses

Introduction and preview

Optical parametric chirped pulse amplification

8 Some quantum mechanics

10 High harmonic generation

Appendix A Conventions and units

Appendix C Definition of the nonlinear coefficients

10.1 Introduction to high harmonic generation

Non-instantaneous response: the time and frequency domain equations

Appendix D Non-zero d elements in non-centrosymmetric crystals

Appendix E Real fields, complex fields, and the analytic signal

Appendix F Geometry of the grating pair

Appendix G The paraxial wave equation

Appendix H Useful formulae for numerical simulations

Appendix I Useful constants

1.1 Nonlinearity in physics

s2 x 2 = s2 A2 cos2 {0 t + } = 12 s2 A2 [1 + cos 2{0 t + }],

1.2 The early history of nonlinear optics

where (1) is the linear susceptibility. So if E = A cos t, the consequence is that P =

dielectric constant. As we will see later, is the square

1.3 Optical second harmonic generation

1.3 Optical second harmonic generation

effect because the index change is linearly proportional to the DC field.

1.4 Phase matching

1.4 Phase matching

1.5 Symmetry considerations

1.6 Optical rectication

(a) Centrosymmetric pattern and (b) non-centrosymmetric pattern.

1.6 Optical rectication

1.8 Sum frequency generation

which has no space-time dependence.

1.7 The Pockels effect

1.8 Sum frequency generation

or simply the k-vector.

Wave vector triangle for phase-matching non-collinear beams.

1.9 Difference frequency generation and optical parametric amplication

1.9 Difference frequency generation and optical parametric

1.10 Third-order processes

1.11 Theoretical foundations

1.11 Theoretical foundations

1.12 The growth of nonlinear optics

components, the second-order nonlinear coefficient ij k has 33 = 27 components, and the

1.12 The growth of nonlinear optics

17 1 attosecond (1 as) = 1018 s.

(2 SHG I1 )2 sin2 {kz/2}

where the mismatch parameter is now defined as k = k3 3k1 . The corresponding

different notation, and with a different definition of k.

where q = kq zR = (kq qk1 )zR and = z /zR .