Theory of Photon Acceleration

Series in Plasma Physics
Series Editors:
Professor Peter Stott, CEA Caderache, France
Professor Hans Wilhelmsson, Chalmers University of Technology, Sweden
Other books in the series
An Introduction to Alfv´ en Waves
R Cross
Transport and Structural Formation in Plasmas
K Itoh, S-I Itoh and A Fukuyama
Tokamak Plasma: a Complex Physical System
B B Kadomtsev
Electromagnetic Instabilities in Inhomogeneous Plasma
A B Mikhailovskii
Instabilities in a Confined Plasma
A B Mikhailovskii
Physics of Intense Beams in Plasma
M V Nezlin
The Plasma Boundary of Magnetic Fusion Devices
P C Stangeby
Collective Modes in Inhomogeneous Plasma
J Weiland
Forthcoming titles in the series
Plasma Physics via Computer Simulation, 2nd Edition
C K Birdsall and A B Langdon
Nonliner Instabilities in Plasmas and Hydrodynamics
S S Moiseev, V G Pungin, and V N Oraevsky
Laser-Aided Diagnostics of Plasmas and Gases
K Muraoka and M Maeda
Inertial Confinement Fusion
S Pfalzner
Introduction to Dusty Plasma Physics
P K Shukla and N Rao
Series in Plasma Physics
Theory of Photon Acceleration
J T Mendonc¸a
Instituto Superior T´ ecnico, Lisbon
Institute of Physics Publishing
Bristol and Philadelphia
c IOP Publishing Ltd 2001
All rights reserved. No part of this publication may be reproduced, stored in a
retrieval system or transmitted in any form or by any means, electronic, mechan-
ical, photocopying, recording or otherwise, without the prior permission of the
publisher. Multiple copying is permitted in accordance with the terms of licences
issued by the Copyright Licensing Agency under the terms of its agreement with
the Committee of Vice-Chancellors and Principals.
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library.
ISBN 0 7503 0711 0
Library of Congress Cataloging-in-Publication Data are available
Commissioning Editor: John Navas
Production Editor: Simon Laurenson
Production Control: Sarah Plenty
Cover Design: Victoria Le Billon
Marketing Executive: Colin Fenton
Published by Institute of Physics Publishing, wholly owned by The Institute of
Physics, London
Institute of Physics Publishing, Dirac House, Temple Back, Bristol BS1 6BE, UK
US Office: Institute of Physics Publishing, The Public Ledger Building, Suite
1035, 150 South Independence Mall West, Philadelphia, PA 19106, USA
Typeset in T
E
X using the IOP Bookmaker Macros
Printed in the UK by Bookcraft, Midsomer Norton, Somerset
Onda que, enrolada, tornas,
Pequena, ao mar que te trouxe
E ao recuar te transtornas
Como se o mar nada fosse
Fernando Pessoa
To my children:
Dina, Joana and Pedro
Contents
Acknowledgments xi
1 Introduction 1
1.1 Definition of the concept 1
1.2 Historical background 3
1.3 Description of the contents 5
2 Photon ray theory 8
2.1 Geometric optics 8
2.2 Space and time refraction 11
2.2.1 Refraction 12
2.2.2 Time refraction 13
2.2.3 Space–time refraction 15
2.3 Generalized Snell’s law 17
2.4 Photon effective mass 22
2.5 Covariant formulation 27
3 Photon dynamics 31
3.1 Ionization fronts 31
3.2 Accelerated fronts 39
3.3 Photon trapping 43
3.3.1 Generation of laser wakefields 43
3.3.2 Nonlinear photon resonance 44
3.3.3 Covariant formulation 49
3.4 Stochastic photon acceleration 51
3.4.1 Motion in two wakefields 52
3.4.2 Photon discrete mapping 54
3.5 Photon Fermi acceleration 57
3.6 Magnetoplasmas and other optical media 63
4 Photon kinetic theory 67
4.1 Klimontovich equation for photons 68
4.2 Wigner–Moyal equation for electromagnetic radiation 70
4.2.1 Non-dispersive medium 70
4.2.2 Dispersive medium 74
viii Contents
4.3 Photon distributions 75
4.3.1 Uniform and non-dispersive medium 76
4.3.2 Uniform and dispersive medium 77
4.3.3 Pulse chirp 78
4.3.4 Non-stationary medium 81
4.3.5 Self-blueshift 82
4.4 Photon fluid equations 83
4.5 Self-phase modulation 87
4.5.1 Optical theory 88
4.5.2 Kinetic theory 90
5 Photon equivalent charge 97
5.1 Derivation of the equivalent charge 97
5.2 Photon ondulator 103
5.3 Photon transition radiation 105
5.4 Photon Landau damping 108
5.5 Photon beam plasma instabilities 113
5.6 Equivalent dipole in an optical fibre 115
6 Full wave theory 123
6.1 Space and time reflection 123
6.1.1 Reflection and refraction 123
6.1.2 Time reflection 125
6.2 Generalized Fresnel formulae 126
6.3 Magnetic mode 128
6.4 Dark source 133
7 Non-stationary processes in a cavity 140
7.1 Linear mode coupling theory 140
7.2 Flash ionization in a cavity 143
7.3 Ionization front in a cavity 146
7.4 Electron beam in a cavity 149
7.5 Fermi acceleration in a cavity 152
8 Quantum theory of photon acceleration 157
8.1 Quantization of the electromagnetic field 157
8.1.1 Quantization in a dielectric medium 157
8.1.2 Quantization in a plasma 161
8.2 Time refraction 165
8.2.1 Operator transformations 165
8.2.2 Symmetric Fock states 167
8.2.3 Probability for time reflection 170
8.2.4 Conservation relations 172
8.3 Quantum theory of diffraction 173
Contents ix
9 New developments 177
9.1 Neutrino–plasma physics 177
9.2 Photons in a gravitational field 183
9.2.1 Gravitational redshift 183
9.2.2 Gravitational lens 186
9.2.3 Interaction of photons with gravitational waves 187
9.2.4 Other metric solutions 191
9.3 Mean field acceleration processes 193
Appendix Derivation of the Wigner–Moyal equation 195
A.1 Non-dispersive media 195
A.2 Dispersive media 198
References 203
Glossary 209
Index 215
Acknowledgments
This book results from a very fruitful collaboration, at both local and international
levels. First of all, I would like to thank Robert Bingham and Padma K Shukla for
their contagious enthusiasm about science, and for their collaboration in several
different subjects. I also would like to thank Nodar L Tsintsadze, for his active
and creative collaboration during his long stay in Lisboa. I extend my thanks to
my many other collaborators in this field.
At the local level, I would like to thank my students, with particular emphasis
to Luis O Silva, who efficiently transformed some of my vague suggestions into
coherent scientific papers. I am also gratefull to Joao M Dias, Nelson Lopes,
Gonc¸alo Figueira, Carla C Rosa, Helder Crespo, Madalena Eloi and Ricardo
Fonseca, who by their successful experiments and numerical simulations helped
me to get a better understanding of this field. I would also like to refer to the more
recent collaboration with Ariel Guerreiro and my colleague Ana M Martins on
the quantum theory. J M Dias, G Figueira and J A Rodrigues prepared some of
the figures in this book.
My acknowledgments extend to my colleagues Armando Brinca and Filipe
Romeiras, who looked carefully at a first version of this book and provided very
useful suggestions and comments.
J T Mendonc¸a
July 1999
xi
Chapter 1
Introduction
We propose in this book to give a simple and accurate theoretical description of
photon acceleration, and of related new concepts such as the effective photon
mass, the equivalent photon charge or the photon Landau damping.
We also introduce, for the first time, the concepts of time reflection and
time refraction, which arise very naturally from the theory of wave propagation
in non-stationary media. Even if some of these concepts seem quite exotic,
they nevertheless result from a natural extension of the classical (and quantum)
electrodynamics to the cases of very fast processes, such as those associated with
the physics of ultra-short and intense laser pulses.
This book may be of relevance to research in the fields of intense laser–
matter interactions, nonlinear optics and plasma physics. Its content may also help
to develop novel accelerators based on laser–plasma interactions, new radiation
sources, or even to establish new models for astrophysical objects.
1.1 Definition of the concept
The concept of photon acceleration appeared quite recently in plasma physics. It
is a simple and general concept associated with electromagnetic wave propaga-
tion, and can be used to describe a large number of effects occurring not only in
plasmas but also in other optical media. Photon acceleration is so simple that it
could be considered a trivial concept, if it were not a subtle one.
Let us first try to define the concept. The best way to do it is to establish a
comparison between this and a few other well-known concepts, such as with re-
fraction. For instance, photon acceleration can be seen as a space–time refraction.
Everybody knows that refraction is the change of direction suffered by a light
beam when it crosses the boundary between two optical media. In more technical
terms we can say that the wavevector associated with this light beam changes,
because the properties of the optical medium vary in space.
We can imagine a symmetric situation where the properties of the optical
medium are constant in space but vary in time. Now the light wavevector remains
1
2 Introduction
constant (the usual refraction does not occur here) but the light frequency changes.
This effect, which is as universal as the usual refraction, can be called time
refraction. A more general situation can also occur, where the optical medium
changes in both space and time and the resulting space–time refraction effect
coincides with what is now commonly called photon acceleration.
Another natural comparison can be established with the nonlinear wave pro-
cesses, because photon acceleration is likewise responsible for the transfer of
energy from one region of the electromagnetic wave spectrum to another. The
main differences are that photon acceleration is a non-resonant wave process,
because it can allow for the transfer of electromagnetic energy from one region of
the spectrum to an arbitrarily different one, with no selection rules.
In this sense it contrasts with the well-known resonant wave coupling pro-
cesses, like Raman and Brillouin scattering, harmonic generation or other three-
or four-wave mixing processes, where spectral energy transfer is dictated by well-
defined conservation laws. We can still say that photon acceleration is a wave
coupling process, but this process is mainly associated with the linear properties
of the space and time varying optical medium where the wavepackets propagate,
and the resulting frequency shift can vary in a continuous way.
Because this is essentially a linear effect it will affect every photon in the
medium. This contrasts, not only with the nonlinear wave mixing processes,
but also with the wave–particle interaction processes, such as the well-known
Compton and Rayleigh scattering, which only affect a small fraction of the in-
cident photons. We can say that the total cross section of photon acceleration is
equal to 1, in contrast with the extremely low values of the Compton or Rayleigh
scattering cross sections. Such a sharp difference is due to the fact that Compton
and Rayleigh scattering are single particle effects, where the photons interact with
only one (free or bounded) electron, while photon acceleration is essentially a
collective process, where the photons interact with all the charged particles of the
background medium.
This means that photon acceleration can only be observed in a dense
medium, with a large number of particles at the incident photon wavelength
scale. For instance, in a very low density plasma, an abrupt transition from
photon acceleration to Compton scattering will eventually occur for decreasing
electron densities. This may have important implications, for instance, in
astrophysical problems.
The concept of photon acceleration is also a useful instrument to explore
the analogy of photons with other more conventional particles such as electrons,
protons or neutrons. Using physical intuition, we can say that the photons can be
accelerated because they have an effective mass (except in a vacuum).
In a plasma, the photon mass is simply related with the electron plasma
frequency, and, in a general optical medium, this effective mass is a consequence
of the linear polarizability of the medium.
This photon effective mass is, in essence, a linear property, but the particle-
like aspects of the electromagnetic radiation, associated with the concept of the
Historical background 3
photon, can also be extended in order to include the medium nonlinearities.
The main nonlinear property of photons is their equivalent electric charge,
which results from radiation pressure, or ponderomotive force effects. In a
plasma, this ponderomotive force tends to push the electrons out of the regions
with a larger content of electromagnetic wave energy. Instead, in an optical fibre
or any similar optical medium, the nonlinear second-order susceptibility leads to
the appearance of an equivalent electric dipole. Because these equivalent charge
distributions, monopole or dipole charges according to the medium, move at very
fast speeds, they can act as relativistic charged particles (electrons for instance)
and can eventually radiate Cherenkov, transition or bremsstrahlung radiation.
These and similar effects have recently been explored in plasmas and in
nonlinear optics, especially in problems related to ultra-short laser pulse prop-
agation [3, 23], or to new particle accelerator concepts and new sources of radia-
tion [42]. The concept of photon acceleration can then be seen as a kind of new
theoretical paradigm, in the sense of Kuhn [51], capable of integrating in a unified
new perspective, a large variety of new or already known effects associated with
electromagnetic radiation.
Furthermore, we can easily extend this concept to other fields, for instance
to acoustics, where phonon acceleration can also be considered [1]. This photon
phenomenology can also be extended to the physics of neutrinos in a plasma
(or in neutral dense matter), if we replace the electromagnetic coupling between
the bound or free electrons with the photons by the weak coupling between the
electrons and the neutrino field [106].
Taking an even more general perspective, we can also say that photon ac-
celeration is a particular example of a mean field acceleration process, which can
act through any of the physical interaction forces (electromagnetic, weak, gravi-
tational or even the strong interactions). This means that, for instance, particles
usually considered as having no electric charge can efficiently be accelerated by
an appropriate background field.
This has been known for many centuries (since the invention of slingshots),
but was nearly ignored by the builders of particle accelerators of our days. It
also means that particles with no bare electric charge can polarize the background
medium, and become ‘dressed’ particles with induced electric charge, therefore
behaving as if they were charged particles.
1.2 Historical background
Let us now give a short historical account of this concept. The basic equations
necessary for the description of photon acceleration have been known for many
years, even if their explicit meaning has only recently been understood. This is
due to the existence of a kind of conceptual barrier, which prevented formally
simple jumps in the theory to take place, which could provide a good example of
what Bachelard would call an obstacle epistemologique [6].
4 Introduction
The history of the photon concept is actually rich in these kinds of conceptual
barrier, and the better known example is the 30 year gap between the definition
by Einstein of the photon as a quantum of light, with energy proportional to
the frequency ω, and the acceptance that such a particle would also have a mo-
mentum, proportional to the wavevector
¯
k, introduced in the theory of Compton
scattering [86].
It is therefore very difficult to have a clear historical view and to find out
when the concept of photon acceleration clearly emerged from the already ex-
isting equations. The assumed subjective account of the author of the photon
acceleration story will be proposed, accepting that other and eventually better and
less biased views are also possible.
One of the first papers which we can directly relate to photon acceleration
in plasma physics was published by Semenova in 1967 [97] and concerns the fre-
quency up-shift of an electromagnetic wave interacting with a moving ionization
front. At this stage, the problem could be seen as an extension of the old problem
of wave reflection by a relativistic mirror, if the mirror is replaced by a surface of
discontinuity between the neutral gas and the ionized gas (or plasma).
The same problem was later considered in greater detail by Lampe et al in
1978 [53]. In these two papers, the mechanism responsible for the ionization
front was not explicitly discussed. But more recently it became aparent that
relativistic ionization fronts could be produced in a laboratory by photoionization
of the atoms of a neutral gas by an intense laser pulse.
A closely related, but qualitatively different, mechanism for photon acceler-
ation was considered by Mendonc¸a in 1979 [63] where the ionization front was
replaced by a moving nonlinear perturbation of the refractive index, caused by
a strong electromagnetic pulse. In this work it was shown that the frequency
up-shift is an adiabatic process occurring not only at reflection as previously
considered, by also at transmission.
At that time, experiments like those of Granatstein et al [36, 87], on
microwave frequency up-shift from the centimetric to the milimetric wave range,
when reflected inside a waveguide by a relativistic electron beam, were exploring
the relativistic mirror concept. The idea behind that theoretical work was to
replace the moving particles by moving field perturbations, easier, in principle, to
be excited in the laboratory.
In a more recent work produced in the context of laser fusion research,
Wilks et al [118] considered the interaction of photons with plasma wakefield
perturbations generated by an intense laser pulse. Using numerical simulations,
they were able to observe the same kind of adiabatic frequency up-shift along
the plasma. This work also introduced for the first time the name of photon
acceleration, which was rapidly adopted by the plasma physics community and
stimulated an intense theoretical activity on this subject.
Arelated microwave experiment by Joshi et al, in 1992 [95] was able to show
that the frequency of microwave radiation contained in a cavity can be up-shifted
to give a broadband spectrum, in the presence of an ionization front produced by
Description of the contents 5
an ultraviolet laser pulse. These results provided the first clear indication that the
photon acceleration mechanism was possibly taking place.
In the optical domain, the observation of a self-produced frequency up-shift
of intense laser pulses, creating a dense plasma when they are focused in a neutral
gas region, and the measured up-shifts [120], were also pointing to the physical
reality of the new concept and giving credit to the emerging theory. Very recently,
the first two-dimensional optical experiments carried out by our group [21], where
a probe laser beam was going through a relativistic ionization front in both co- and
counter-propagation, were able to demonstrate, beyond any reasonable doubt, the
existence of photon acceleration and to provide an accurate quantitative test of the
theory.
Actually, the spectral changes of laser beams by ionization of a neutral gas
were reported as early as 1974 by Yablonovich [122]. In these pioneering experi-
ments, the spectrum of a CO
2
laser pulse was strongly broadened and slightly up-
shifted, when the laser beam was focused inside an optical cavity and ionization
of the neutral gas inside the cavity was produced. This effect is now called
flash ionization for reasons that will become apparent later, and it can also be
considered as a particular and limiting case of the photon acceleration processes.
In parallel with this work in plasma physics, and with almost completely
mutual ignorance, following both theoretical and experimental approaches clearly
independently, research on a very similar class of effects had been taking place in
nonlinear optics since the early seventies. This work mainly concentrated on the
concept of phase modulation (including self-, induced and cross phase modula-
tion), and was able to prove both by theory and by experiments, that laser pulses
with a very large spectrum (called the supercontinuum radiation source) can be
produced. This is well documented in the book recently edited by Alfano [3].
As we will see in the present work, the theory of photon acceleration as
developed in plasma physics is also able to explain the phase modulation effects,
when we adapt it to the optical domain. This provides another proof of the interest
and generality of the concept of photon acceleration. We will attempt in this work
to bridge the gap between the two scientific communities and between the two
distinct theoretical views.
1.3 Description of the contents
Four different theoretical approaches to photon acceleration will be considered in
this work: (1) single photon trajectories, (2) photon kinetic theory, (3) classical
full wave models and (4) quantum theory.
The first two chapters will be devoted to the study of single photon equations
(also called ray equations), derived in the frame of geometric optics. This is the
simplest possible theoretical approach, which has several advantages over more
accurate methods. Due to its formal simplicity, we can apply it to describe,
with great detail and very good accuracy, various physical configurations where
6 Introduction
photon acceleration occurs. These are ionization fronts (with arbitrary shapes and
velocities), relativistic plasma waves or wakefields, moving nonlinearities and
flash ionization processes.
It can also be shown that stochastic photon acceleration is possible in several
physical situations, leading to the transformation of monocromatic radiation into
white light. An interesting example of stochastic photon behaviour is provided by
the well-known Fermi acceleration process [29], applied here to photons, which
can be easily described with the aid of single photon equations.
Apart from its simplicity and generality, photon ray equations are formally
very similar to the equations of motion of a material particle. This means that
photon acceleration happens to be quite similar to electron or proton acceleration
by electromagnetic fields, even if the nature of the forces acting on the photons
is not the same. For instance, acceleration and trapping of electrons and photons
can equally occur in the field of an electron plasma wave.
Chapter 2 deals with the basic concepts of this single photon or ray theory,
as applied to a generic space- and time-varying optical medium. The concept
of space–time refraction is introduced, the generalized Snell’s laws are derived
and the ray-tracing equations are stated in their Hamiltonian, Lagrangian and
covariant forms.
Chapter 3 deals with the basic properties of photon dynamics, illustrated with
examples taken from plasma physics, with revelance to laser–plasma interaction
problems. Extension to other optical media, and to nonlinear optical configura-
tions will also be discussed.
This single photon theory is simple and powerful, but it can only provide a
rough description of the laser or other electromagnetic wavepackets evolving in
non-stationary media. However, an extension of this single particle description to
the kinetic theory of a photon gas is relatively straightforward and can lead to new
and surprising effects such as photon Landau damping [14]. This is similar to the
well-known electron Landau damping [54].
Such a kinetic theory is developed in chapter 4 and gives a much better
description of the space–time evolution of a broadband electromagnetic wave
spectrum. This is particularly important for ultra-short laser pulse propagation.
In particular, self-phase modulation of a laser pulse, propagating in a nonlinear
optical medium, and the role played by the phase of the laser field in this process,
will be discussed.
Chapter 5 is devoted to the discussion of the equivalent electric charge of
photons in a plasma, and of the equivalent electric dipole of photons in an optical
fibre. We will also discuss the new radiation processes associated with these
charge distributions, such as photon ondulator radiation, photon transition radi-
ation or photon bremstrahlung.
The geometric optics approximation, in its single photon and kinetic ver-
sions, provides a very accurate theoretical description for a wide range of different
physical configurations. Even very fast time events, occurring on a timescale
of a few tens of femtoseconds, can still be considered as slow processes in the
Description of the contents 7
optical domain and stay within the range of validity of this theory. But, in several
situations, a more accurate theoretical approach is needed, in order to account for
partial reflection, for specific phase effects, or for arbitrarily fast time processes.
We are then led to the full wave treatment of photon acceleration, which
is presented in chapters 6 and 7. Most of the problems discussed in previous
chapters are reviewed with this more exact approach and a comparison is made
with the single photon theory when possible. New aspects of photon acceleration
can now be studied, such as the generation of a magnetic mode, the multiple
mode coupling or the theory of the dark source which describes the possibility of
accelerating photons initially having zero energy.
In chapter 8 we show that a quantum description of photon acceleration
is also possible. We will try to establish in solid grounds the theory of time
refraction, which is the basic mechanism of photon acceleration.
The quantum Fresnel formulae for the field operators will be derived. We
will also show that time refraction always leads to the creation of photon pairs,
coming out of the vacuum. More work is still in progress in this area.
Finally, chapter 9 is devoted to new theoretical developments. Here, the
photon acceleration theory is extended in a quite natural way to cover new physi-
cal problems, which correspond to other examples of the mean field acceleration
process. These new problems are clearly more controversial than those covered
in the first eight chapters, but they are also very important and intellectually very
stimulating.
Two of these examples are briefly discussed.The first one concerns collective
neutrino plasma interaction processes, which were first explicitly formulated by
Bingham et al in 1994 [13] and have recently received considerable attention
in the literature. The analogies between the photon and the neutrino interaction
with a background plasma will be established. The second example will be the
interaction of photons with a gravitational field and the possibility of coupling
between electromagnetic and gravitational waves. One of the consequences of
such an interaction is the occurrence of photon acceleration in a vacuum by
gravitational waves.
Chapter 2
Photon ray theory
It is well known that the wave–particle dualism for the electromagnetic radiation
can be described in purely classical terms. The wave behaviour is described by
Maxwell’s equations and the particle behaviour is described by geometric optics.
This contrasts with other particles and fields where the particle behaviour is de-
scribed by classical mechanics and the wave behaviour by quantum mechanics.
Geometric optics is a well-known and widely used approximation of the
exact electromagnetic theory and it is presented in several textbooks [15, 16].
We will first use the geometric optics description of electromagnetic wavepackets
propagating in a medium. These wavepackets can be viewed as classical particles
and can be assimilated to photons. We will then apply the word ‘photon’ in
the classical sense, as the analogue of an electromagnetic wavepacket. A single
photon can be used to represent the mean properties of a given wavepacket. The
photon velocity will then be equal to the group velocity of the wavepacket.
This single photon approach will be used in the present and the next chapters.
A more accurate description of a wavepacket can still be given in the geometric
optics approximation, by using a bunch of photons, instead of a single one. The
study of such a bunch will give us information on the internal spectral content
of the wavepacket. This will be discussed in chapter 4. The use of the photon
concept in a quantum context will be postponed until chapter 8.
2.1 Geometric optics
It is well known that a wave is a space–time periodic event, where the time
periodicity is characterized by the angular frequency ω and the space periodicity
as well as the direction of propagation are characterized by the wavevector
¯
k. In
particular, an electromagnetic wave can be described by an electric field of the
form
¯
E(¯ r, t ) =
¯
E
0
exp i{
¯
k · ¯ r −ωt } (2.1)
where
¯
E
0
is the wave field amplitude, ¯ r is the position and t the time.
8
Geometric optics 9
In a stationary and uniform medium the frequency and the wavevector are
constants, but they are not independent from each other. Instead, they are related
by a well-defined expression (at least for low amplitude waves) known as the
dispersion relation. In a vacuum, the dispersion relation is simply given by ω =
kc, where the constant c is the speed of light in a vacuum and k is the absolute
value of the wavevector (also known as the wavenumber).
In a medium, the dispersion relation becomes
ω =
kc
n
=
kc

=
kc

1 ÷χ
(2.2)
where n is the refractive index of the medium, its dielectric constant and χ its
susceptibility. In a lossless medium these quantities are real, and in dispersive
media they are functions of the frequency ω and the wavevector
¯
k.
In particular, for high frequency transverse electromagnetic waves propa-
gating in an isotropic plasma [82, 108], we have = 1 − (ω
p
/ω)
2
, where ω
p
is the electron plasma frequency. It is related to the electron density n
e
by the
expression: ω
2
p
= e
2
n
e
/
0
m, where e and m are the electron charge and mass,
and
0
is the vacuum permittivity.
From equation (2.2) we can then have a dispersion relation of the form
ω =

k
2
c
2
÷ω
2
p
. (2.3)
A similar equation is also valid for electromagnetic waves propagating in a
waveguide [25]. The plasma frequency is now replaced by a cut-off frequency ω
0
depending on the field configuration and on the waveguide geometry.
In the most general case, however, the wave dispersion relation is a com-
plicated expression of ω and
¯
k, and such simple and explicit expressions for the
frequency cannot be established. It is then preferable to state it implicitly as
R(ω,
¯
k) = 0. (2.4)
The above description of wave propagation is only valid for uniform and
stationary media. Let us now assume that propagation is taking place in a non-
uniform and non-stationary medium. If the space and time variations in the
medium are slow enough (in such a way that, locally both in space and in time,
the medium can still be considered as approximately uniform and constant), we
can replace the wave electric field (2.1) by a similar expression:
¯
E(¯ r, t ) =
¯
E
0
(¯ r, t ) exp iψ(¯ r, t ) (2.5)
where ψ(¯ r, t ) is the wave phase and
¯
E
0
(¯ r, t ) a slowly varying wave amplitude.
We can now define a local value for the wave frequency and for the wavevec-
tor, by taking the space and time derivatives of the phase function ψ:
¯
k =

∂¯ r
ψ, ω = −

∂t
ψ. (2.6)
10 Photon ray theory
This local frequency ω and wavevector
¯
k are still related by a dispersion
relation, which is now only locally valid:
R(ω,
¯
k: ¯ r, t ) = 0. (2.7)
The parameters of the medium, for instance its refractive index or, in a
plasma, its electron plasma frequency ω
p
, depend on the position ¯ r and on time
t . Such a dispersion relation is satisfied at every position and at each time. This
means that its solution can be written as ω = ω(¯ r,
¯
k, t ).
Starting from Maxwell’s equations, it can be shown that this expression
stays valid as long as the space and timescales for the variations of the medium
are much slower than k
−1
and ω
−1
, or, in more precise terms, if the following
inequality is satisfied:

ω


∂t
ln ξ

÷

k
[∇ ln ξ[ <1 (2.8)
where ξ is any scalar characterizing the background medium, for instance the
refractive index or, for a plasma, the electron density.
It can easily be seen from the above definitions of the local frequency and
wavevector that

¯
k
∂t
= −∇ω = −
¸
∂ω
∂¯ r
÷

¯
k
∂¯ r
·
∂ω

¯
k

. (2.9)
The last term in this equation was established by noting that, because ω
is assumed to be a function of ¯ r and
¯
k, it varies in space not only because of
its explicit dependence on ¯ r but because the value of
¯
k is also varying. Let us
introduce the definition of group velocity
¯ v
g
=
∂ω

¯
k
. (2.10)
This is the velocity of the centroid of an electromagnetic wavepacket moving
in the medium, ¯ v
g
= d¯ r/dt . The above equation can be written as


∂t
÷ ¯ v
g
·

∂¯ r

¯
k = −
∂ω
∂¯ r
. (2.11)
The differential operator on the left-hand side is nothing but the total time
derivative d/dt . It means that we can rewrite the last two equations as
d¯ r
dt
=
∂ω

¯
k
,
d
¯
k
dt
= −
∂ω
∂¯ r
. (2.12)
These equations of motion can be seen as describing the evolution of point
particles: the photons. We notice that they are written in Hamiltonian form. The
canonical variables are here the photon position ¯ r and the wavevector
¯
k, while the
Space and time refraction 11
frequency plays the role of the Hamiltonian function, ω ≡ h(¯ r,
¯
k, t ). In general,
it will be time dependent according to

dt
=
∂ω
∂t
. (2.13)
These equations are well known in the literature and their derivation is given
in textbooks [55] and in papers [8, 9, 117]. We see that they explicitly predict
a frequency shift as well as a wavevector change. Equations (2.12) and (2.13)
have, in fact, an obvious symmetrical structure. But, for some historical reason,
the physical implications of such a structure for media varying both in space and
time have only recently been fully understood [66].
These ray equations can also be written with implicit differentiation as
d
¯
k
dt
=
∂ R/∂¯ r
∂ R/∂ω
,
d¯ r
dt
= −
∂ R/∂
¯
k
∂ R/∂ω
(2.14)
and

dt
= −
∂ R/∂t
∂ R/∂ω
. (2.15)
This implicit version, even if it is appropriate for numerical computations
(this is currently used for magnetized plasmas with complicated spatial config-
urations), loses the clarity and elegance of the Hamiltonian approach. For this
reason we will retain our attention on the explicit Hamiltonian version of the ray
or photon equations.
2.2 Space and time refraction
In order to understand the physical meaning of the photon equations stated above,
let us use a simple and illustrative example of a photon propagating in a non-
dispersive (but space- and/or time-dependent) dielectric medium, described by
the dispersion relation
ω ≡ ω(¯ r,
¯
k, t ) =
kc
n(¯ r, t )
. (2.16)
From equations (2.12), we have
d¯ r
dt
=
ω
k
2
¯
k,
d
¯
k
dt
= ω

∂¯ r
ln n. (2.17)
From equation (2.13), we can also obtain

dt
= −ω

∂t
ln n. (2.18)
Let us successively apply these equations to three distinct situations: (i) an
inhomogeneous but stationary medium; (ii) an homogeneous but time-dependent
medium; (iii) an inhomogeneous and non-stationary medium.
12 Photon ray theory
2.2.1 Refraction
In the first situation, we assume two uniform and stationary media, with refractive
indices n
1
and n
2
, separated by a boundary layer of width l
b
located around the
plane y = 0. In order to describe such a configuration we can write
n(¯ r, t ) ≡ n(y) = n
1
÷
n
2
[1 ÷tanh(k
b
y)] (2.19)
where n = n
2
−n
1
, and k
b
= 2π/l
b
.
The hyperbolic tangent is chosen as a simple and plausible model for a
smooth transition between the two media. Clearly, the ray equations (2.17, 2.18)
are only valid when l
b
is much larger than the local wavelength 2π/k. On the
other hand, if we are interested in the study of the photon or ray propagation across
a large region with dimensions much larger than l
b
, such a smooth transition can
be viewed from far away as a sharp boundary, and the above model given by
equation (2.19) corresponds to the usual optical configuration for wave refraction.
In this large scale view of refraction, the above law can be approximated by
n(y) = n
1
÷nH(y) (2.20)
where H(y) = 0 for y < 0 and H(y) = 1 for y > 0 is the well-known step
function or Heaviside function.
First of all, it should be noticed that from equations (2.18) and (2.19) we
have

dt
= 0. (2.21)
This means that the wave frequency is a constant of motion ω(¯ r,
¯
k, t ) =
ω
0
= const. If the plane of incidence coincides with z = 0, the time variation of
the wavevector components is determined by
dk
x
dt
= 0 (2.22)
dk
y
dt
= ω
0

∂y
ln n(y) =
ω
0
n
2n(y)
k
b
sech
2
(k
b
y). (2.23)
We see that the wavevector component parallel to the gradient of the re-
fractive index is changing across the boundary layer, and that the perpendicular
component remains constant. Defining θ(y) as the angle between the wavevector
¯
k and the normal to the boundary layer ˆ e
y
, we can obtain, from the first of these
equations,
k
x
=
ω
0
c
n(y) sin θ(y) = const. (2.24)
Considering the asymptotic values of θ(y) as the usual angles of incidence
and of transmission θ
1
= θ(y → −∞) and θ
2
= θ(y → ÷∞), we reduce this
result to the well-known Snell’s law of refraction
n
1
sin θ
1
= n
2
sin θ
2
. (2.25)
Space and time refraction 13
n
x
θ
1
1
ω
=
constant
n
2
θ
2
y
Figure 2.1. Photon refraction at a boundary between two stationary media.
Returning to equations (2.17) we can also see that refraction leads to a
change in the group velocity, or photon velocity, with asymptotic values v
g
=
c/n
1
, for y → −∞, and v
g
= c/n
2
, for y → ÷∞. In a broad sense, we
could be led to talk about photon acceleration during refraction. But, as we will
see later, this is not appropriate because this change in group velocity is exactly
compensated by a change in the photon effective mass, in such a way that the
total photon energy remains constant. This will not be the case for the next two
examples.
2.2.2 Time refraction
Let us turn to the opposite case of a mediumwhich is uniformin space but changes
its refractive index with time. We can describe this change by a law similar to
equation (2.19):
n(¯ r, t ) ≡ n(t ) = n
1
÷
n
2
[1 ÷tanh(
b
t )] . (2.26)
Here, the timescale for refractive index variation 2π/
b
is much larger than
the wave period 2π/ω. From equation (2.19), we now have

dt
= −ω

∂t
ln n(t ) = −
ωn
2n(t )

b
sech
2
(
b
t ). (2.27)
We see that the photon frequency is shifted as time evolves, following a law
similar to that of the wavevector change during refraction. On the other hand, if
14 Photon ray theory
n x
α
1
1
k = constant
ct
n
2
α
2
Figure 2.2. Time refraction of a photon at the time boundary between two uniform media.
propagation is taken along the x-direction, we can see from equation (2.18) that
dx
dt
=
c
n(t )
= v
g
(t ),
dk
dt
= 0. (2.28)
The photon momentum is now conserved and the change in the group ve-
locity is only related to the frequency (or energy) shift. Snell’s law (2.24) is now
replaced by
k =
ω(t )
c
n(t ) = const. (2.29)
Defining the initial and the final values for the frequency as the asymptotic
values ω
1
= ω(t →−∞) and ω
2
= ω(t →÷∞), we can then write
n
1
ω
1
= n
2
ω
2
. (2.30)
This can be called the Snell’s law for time refraction. Actually, in the plane
(x, ct ) we can define an angle α, similar to the usual angle of incidence θ defined
in the plane (x, y), such that tan α = v
g
/c = 1/n. This could be called the angle
of temporal incidence. Replacing it in equation (2.30), we get
ω
1
tan α
2
= ω
2
tan α
1
. (2.31)
This equation resembles the well-known Snell’s law of refraction, equa-
tion (2.25). However, there is an important qualitative difference, related to
Space and time refraction 15
the mechanism of total reflection. We know that this can occur when a photon
propagates in a medium of decreasing refractive index. From equation (2.25)
we see that, for n
1
> n
2
, it is possible to define a critical angle θ
c
such that
sin θ
c
= n
2
/n
1
. For θ
1
≥ θ
c
we would have sin θ
2
≥ 1 which means that
propagation is not allowed in medium 2 and that total reflection at the boundary
layer will occur.
In contrast, equations (2.30) or (2.31) are always satisfied for arbitrary (pos-
itive) values of the asymptotic refractive indices n
1
and n
2
. Not surprisingly, no
such thing as a reflection back in time (or a return to the past) can ever occur. This
is the physical meaning of the replacement of the sine law for reflection by the
tangent law for time refraction.
2.2.3 Space–time refraction
Let us now assume a more general situation where the optical medium is both
inhomogeneous in space and non-stationary in time. As a simple generalization
of the previous two models, we will assume that the boundary layer between
media 1 and 2 is moving with a constant velocity u along the x-direction.
Equations (2.19) and (2.26) are now replaced by a similar expression, of the
form
n(¯ r, t ) ≡ n(x −ut ) = n
1
÷
n
2
[1 ÷tanh(k
b
x −
b
t )] (2.32)
with
b
= k
b
u. From the ray equations (2.12) we can now derive a simple
relation between the time variation of frequency and wavenumber:
dk
dt
=
1
u

dt
. (2.33)
This relation means that, when a photon interacts with a moving boundary
layer, the wave frequency and wavevector are both shifted, in a kind of space–
time refraction. We notice that in order to establish the Snell’s laws (2.24, 2.25),
or their temporal counterparts (2.29, 2.30), we had to define some constant of
motion.
In the first case, the constants of motion were the photon frequency ω and
the wavevector component parallel to the boundary layer k
x
. In the second case,
the constant of motion was the photon wavevector. Now, for the case of refraction
in a moving boundary, or space–time refraction, we need to find a new constant
of motion because neither the frequency nor the wavevector are conserved.
This new constant of motion can be directly obtained from equation (2.33),
but it is useful to explore the Hamiltonian properties of the ray equations (2.12,
2.13). For such a purpose, let us then define a canonical transformation from the
variables(x, k) to the new pair of variables (x
/
, k
/
), such that
x
/
= x −ut, k
/
= k. (2.34)
16 Photon ray theory
The resulting canonical ray equations can be written as
dx
/
dt
=
∂ω
/
∂k
/
,
dk
/
dt
= −
∂ω
/
∂x
/
. (2.35)
It can easily be seen that the new Hamiltonian ω
/
appearing in these equa-
tions is a time-independent function defined by
ω
/
≡ ω
/
(x
/
, k
/
) = ω(x
/
, k
/
) −uk
/
=
¸
c
n(x
/
)
−u
¸
k
/
. (2.36)
In a more formal way, we can define a generating function for this canonical
transformation, as
F(x, k
/
, t ) = (x −ut )k
/
(2.37)
such that
k =
∂ F
∂x
= k
/
(2.38)
x
/
=
∂ F
∂k
/
= x −ut. (2.39)
The Hamiltonian (2.36) is determined by
ω
/
= ω ÷
∂ F
∂t
= ω −uk
/
. (2.40)
Because, based on this canonical transformation, we were able to define a
new constant of motion ω
/
= const, we can take the asymptotic values of n(x
/
)
and k
/
for regions far away from the boundary layer (x
/
→±∞), and write
ω
/
=

c
n
1
−u

k
1
=

c
n
2
−u

k
2
(2.41)
or, equivalently, in terms of the frequency
ω
/
= ω
1

1 −
n
1
c
u

= ω
2

1 −
n
2
c
u

. (2.42)
We are then led to the following formula for the total frequency shift associ-
ated with space–time refraction:
ω
2
= ω
1
1 −β
1
1 −β
2
(2.43)
where β
i
= un
i
/c, for i = 1, 2. We see that the resulting frequency up-shift can
be extremely large if the boundary layer and the photons are moving in the same
direction at nearly the same velocity: β
2
. 1.
Generalized Snell’s law 17
When compared with the frequency shifts associated with the case of a
purely time refraction, this shows important qualitative and quantitative differ-
ences. This new effect of almost unlimited frequency shift is a clear consequence
of the combined influence of the space and time variations of the medium, or
in other words, of the synergy between the usual refraction and the new time
refraction considered above.
A comment has to be made on the exact resonance condition, defined by
β
2
= 1. At a first impression, it could be concluded from equation (2.43) that
the frequency shift would become infinitely large if this resonance condition were
satisfied. However, because in this case the photons are moving with the same
velocity as the boundary layer, they would need an infinite time to travel across
the layer and to be infinitely frequency shifted.
A closer look at the resonant photon motion shows that this trajectory would
be x
/
= const and, according to equation (2.36), no frequency shift would occur.
On the other hand, the resonant photon trajectory is physically irrelevant because
it represents an ensemble of zero measure in the photon phase space (x, k). What
is relevant is that, in the close vicinity of this particular trajectory, a large number
of possible photon trajectories will lead to extremely high-frequency transforma-
tions taking place in a finite but large amount of time.
Let us now apply the above analysis to the well-known and famous problem
of photon reflection by a relativistic mirror. This can be done by assuming that
n
1
= 1 and n is such that n(x
/
) = 0 for x
/
≥ 0.
The mirror will move with constant velocity −u and its surface will coincide
with the plane x
/
= 0. A photon propagating initially in a vacuum in the opposite
direction, with a wavevector
¯
k = k
1
ˆ e
x
, will be reflected by the mirror and come
back to medium 1 with a wavevector
¯
k = −k
1
ˆ e
x
.
This means that in the above equation (2.43) we can make β
2
= β = u/c
and β
1
= −β. It will then become
ω
2
= ω
1
1 ÷β
1 −β
. (2.44)
This is nothing but the well-known formula for the relativistic mirror, which
was derived here in a very simple way, without invoking relativity or using any
Lorentz transformation. Such a result is possible because the photon equations are
exactly relativistic by nature. Apart from providing a very simple and alternative
way to derive the relativistic mirror effect, this result is also interesting because
it shows that such an effect can be seen as a particular case of the more general
space–time refraction or photon acceleration processes.
2.3 Generalized Snell’s law
We generalize here the above discussion of space–time refraction, by considering
oblique photon propagation with respect to the moving boundary between two
18 Photon ray theory
different dielectric media. The moving boundary will now be described by the
following refractive index:
n(¯ r, t ) = n
1
÷
n
2
¸
1 ÷tanh(
¯
k
b
· ¯ r −
b
t )
¸
(2.45)
where
b
=
¯
k
b
· ¯ u and ¯ u is the velocity of the moving boundary. As before,
we assume that the velocity and the slope of the boundary layer, defined by
¯
k
b
,
remain constant.
It is useful to introduce parallel and perpendicular propagation, by defining
¯ r = r
|
ˆ
b ÷¯ r

and
¯
k = k
|
ˆ
b ÷
¯
k

, where
ˆ
b =
¯
k
b
/k
b
. We conclude from the photon
equations of motion that
dk
|
dt
=
kc
n
2
∂n
∂r
|
(2.46)
d
¯
k

dt
= 0. (2.47)
The conservation of the perpendicular wavevector is due to the fact that the
refractive index is only a function of r
|
: n(¯ r, t ) = n(r
|
, t ). We can now use the
canonical transformation from (¯ r,
¯
k) to a new pair of variables (¯ r
/
,
¯
k
/
), generated
by the transformation function
F(¯ r,
¯
k
/
, t ) = (¯ r − ¯ ut ) ·
¯
k
/
. (2.48)
This is a straightforward generalization of the canonical transformation used
for the one-dimensional problem. This leads to
¯
k =
∂ F
∂¯ r
=
¯
k
/
(2.49)
¯ r
/
=
∂ F

¯
k
/
= ¯ r − ¯ ut. (2.50)
The new Hamiltonian ω
/
(¯ r
/
,
¯
k
/
, t ) appearing in these equations is determined
by
ω
/
= ω ÷
∂ F
∂t
= ω − ¯ u ·
¯
k
/
. (2.51)
This Hamiltonian is a constant of motion because in the new coordinates
neither n nor ω depend explicitly on time. It is clear that we can now define two
constants of motion for photons crossing the moving boundary. Let us call them
I
1
and I
2
. The first one is simply the new Hamiltonian ω
/
, and the other one is the
perpendicular wavenumber k

:
I
1
= ω(¯ r
/
,
¯
k
/
) − ¯ u ·
¯
k
/
(2.52)
I
2
= [
¯
k
/

ˆ
b[. (2.53)
Generalized Snell’s law 19
Notice that our analysis is valid even if the velocity of the moving boundary
¯ u is not parallel to the gradient of the refractive index
ˆ
b. However, in order to
place our discussion on simple grounds, we assume here that ¯ u is parallel to
ˆ
b. If
θ(¯ r
/
) is the angle between the photon wavevector and the front velocity, defined
for each photon position ¯ r
/
along the trajectory, we can rewrite the two invariants
as
I
1
= ω(¯ r
/
,
¯
k
/
)
¸
1 −
u
c
n(¯ r
/
) cos θ(¯ r
/
)
¸
(2.54)
I
2
= k
/
sin θ(¯ r
/
). (2.55)
In the first of these equations, we made use of the local dispersion relation
n(¯ r
/
)ω(¯ r
/
,
¯
k
/
) = k
/
c. Let us retain our attention on photon trajectories which start
in medium 1 far away from the boundary, at ¯ r
/
→ −∞, cross the boundary layer
(at ¯ r
/
= 0) and then penetrate in medium 2 moving towards ¯ r
/
→÷∞.
Denoting by the subscripts 1 and 2 the frequencies and angles at the extreme
ends of such trajectories, we can conclude, from the above two invariants, that
ω
2
= ω
1
1 −β
1
cos θ
1
1 −β
2
cos θ
2
(2.56)
where β
i
= (u/c)n
i
, for i = 1, 2. This expression establishes the total frequency
shift. We can also conclude that
n
1
sin θ
1
=
ω
2
ω
1
n
2
sin θ
2
. (2.57)
This last equation can be seen as the generalized Snell’s law, valid for refrac-
tion at a moving boundary betwen two different dielectric media. It shows that
the relation between the angles of incidence and transmission depends not only
on the refractive indices of the two media but also on the photon frequency shift.
Of course, when the velocity of the boundary layer tends to zero, u → 0,
the frequency shift also tends to zero, ω
2
→ ω
1
, as shown by equation (2.56).
In this case equation (2.57) reduces to the usual form of Snell’s law (2.25), as it
should. On the other hand, for a finite velocity u ,= 0, but for normal incidence
θ
1
= θ
2
= 0, equation (2.56) reduces to ω
2
= ω
1
(1−β
1
)/(1−β
2
), in accordance
with the discusssion of the previous section.
We should notice that equation (2.56) is valid for an arbitrary value of the
velocity of the moving boundary. It stays valid, in particular, for supraluminous
moving boundaries, such that u > c. It is well known that such boundaries can
exist without violating the principles of Einstein’s theory of relativity [31]. They
can even be built in a medium where the atoms are completely at rest, as will be
discussed in the next chapter.
Let us consider the limit of an infinite velocity u → ∞. Equation (2.56) is
then reduced to equation (2.30): n
2
ω
2
= n
1
ω
1
. This means that it reproduces the
above result for a pure time refraction.
20 Photon ray theory
α
1
c t
x
n
2
n
1
α
2
(a)
n
1
y
u
ut
0
θ
1
n
2
θ
2
x u t
(b)
Figure 2.3. Illustration of the generalized Snell’s law: photon refraction at the moving
boundary between two uniform media. (a) Diagram (x, ct ) for one-dimensional propaga-
tion; (b) oblique propagation.
Now, for an arbitrary oblique incidence, but for highly supraluminous fronts,
u ·c, we get from equations (2.56, 2.57)
sin(θ
1
−θ
2
) . 0. (2.58)
We see that, in this limit of a highly supraluminous front, there is no momen-
tum transfer from the medium to the photon, θ
1
. θ
2
. The resulting frequency
shift is determined in this case by the law of pure time refraction, showing that
Generalized Snell’s law 21
sin

θ
c
β
1
Figure 2.4. Dependence of the critical angle θ
c
on the front velocity β
1
, for n
2
/n
1
= 0.8.
space changes of the refractive index of the medium do not contribute to the final
frequency shift. The dominant processes are associated with the time changes.
Let us now examine the conditions for the occurrence of total reflection.
From the above expression of the generalized Snell’s law (2.57) we can define a
critical angle θ
1
= θ
c
such that θ
2
= π/2, or
sin θ
c
=
n
2
n
1
ω
2
ω
1
. (2.59)
Using equation (2.56), we also see that
sin θ
c
=
n
2
n
1
(1 −β
1
cos θ
c
). (2.60)
This can be explicitly solved and gives
sin θ
c
=
n
1
n
2
n
2
1
÷β
2
1
n
2
2

¸
¸
1 ÷[β
1
[

1 −

n
2
n
1

2
(1 −β
2
1
)
¸
¸
¸
. (2.61)
For a boundary at rest, β
1
= 0, this expression reduces to the well-known
formula for the critical angle sin θ
c
= n
2
/n
1
. We also notice that this result
guarantees that in the absence of any boundary (which means n
2
= n
1
) we always
get sin θ
c
= 1, or θ
c
= π/2 (absence of total reflection) for arbitrary values of
β
1
, as expected. The same occurs for purely time-varying media ([β
i
[ → ∞),
confirming that no time reflection is possible.
For angles of incidence larger than this critical angle θ
1
> θ
c
, the photon
will be reflected. This means that the direction of the reflected wavevector will be
reversed with respect to that assumed in equation (2.56).
If the boundary layer is moving towards the incident photon (which is the
usual configuration of the relativistic mirror effect), we have to replace β
1
by −β
1
22 Photon ray theory
in this equation. Moreover, we will replace the subscript 2 by the subscript 3,
in order to stress that the final photon state is given by reflection and not by
transmission:
ω
3
= ω
1
1 ÷β
1
cos θ
1
1 −β
1
cos θ
3
. (2.62)
Here the angle of reflection θ
3
is not equal to the angle of incidence (as
occurs for a stationary boundary) but, according to equation (2.57), it is given by
sin θ
3
=
ω
1
ω
3
sin θ
1
. (2.63)
These two formulae state the relativistic mirror effect for oblique incidence,
as a particular case of photon acceleration at the moving boundary. Actually,
these formulae stay valid even for partial reflection at the boundary, for angles
of incidence smaller than the critical angle. But in order to account for partial
reflection we will have to use a full wave description and not use the basis of the
geometric optics approximation.
Let us return to the general expression for the two photon invariants I
1
and
I
2
, and make a short comment on the case where the vector velocity, ¯ u, and
¯
k
b
,
characterizing the gradient of the refractive index across the boundary layer, are
not parallel to each other. We can then write ¯ u ·
¯
k
/
= uk
/
cos α(¯ r
/
) and [
¯
k
/

ˆ
b[ =
k
/
sin θ(¯ r
/
), where these two angles α(¯ r
/
) and θ(¯ r
/
) are not necessarily equal to
each other. If ¯ u is contained in the plane of incidence, we have θ(¯ r
/
) = α(¯ r
/
)÷θ
0
,
where θ
0
is a constant. In the expressions of the two invariants I
1
and I
2
, as given
by equation (2.55), the angle θ(¯ r
/
) has to be replaced by α(¯ r
/
) = θ(¯ r
/
) −θ
0
.
In our discussion we have kept our attention on very simple but paradigmatic
situations of boundary layers moving with constant velocity. This is because our
main objective here was to extend the existing geometric optics theory to the
space–time domain and to obtain simple but also surprising generalizations of
well-known formulae and well-known concepts.
Our Hamiltonian approach to photon dynamics can however be applied to
more complicated situations of space–time varying media, for instance to non-
stationary and accelerated boundary layers.
2.4 Photon effective mass
Let us now go deeper in the analysis of the photon dynamics and explore the
analogies between the photon ray equations and the equations of motion of an
arbitrary point particle with finite rest mass. From the above definitions of the
local wavevector
¯
k and local frequency ω, it is obvious that the total variation of
the phase ψ, as a function of the space and time coordinates, is
dψ =
¯
k · d¯ r −ωdt. (2.64)
Photon effective mass 23
If we identify the photon frequency ω with the particle Hamiltonian, and
the wavevector
¯
k with its momentum, as we did before, we can easily see from
classical mechanics [33, 56] that this phase is nothing but the photon action
ψ =

¯
k · d¯ r −ω(¯ r,
¯
k, t ) dt. (2.65)
In order to establish the photon trajectories we can apply the principle of
minimum action, and state that the phase has to obey the extremum condition
δψ = 0. (2.66)
It is well known that, from this variational principle, it is possible to rederive
the photon canonical equations already established in section 2.1. On the other
hand, for a dielectric medium with refractive index n = kc/ω, and from the
definitions of
¯
k and ω, we can get the following equation for the phase:

∂ψ
∂¯ r

2
=
n
2
c
2

∂ψ
∂t

2
. (2.67)
In the particular case of a stationary medium, we have ω = ω
0
= const, and
equation (2.65) reduces to
ψ = ψ
0
−ω
0
t =

¯
k · d¯ r −ω
0
t (2.68)
where ψ
0
is the reduced photon action. Equation (2.67) is then reduced to the
usual eikonal equation of geometric optics [15]

∂ψ
0
∂¯ r

2
=
n
2
c
2
ω
2
0
. (2.69)
We can then refer to the more general equation for the phase, equation (2.67),
as for the generalized eikonal equation, valid for space–time varying media.
We also know from classical mechanics that the action can be defined as the
time integral of a Lagrangian function. The photon Lagrangian is then determined
by
ψ =

t
2
t
1
L(¯ r, ¯ v, t ) dt (2.70)
where ¯ v = d¯ r/dt is the photon velocity, or group velocity. Comparing this with
equation (2.65), we obtain
L(¯ r, ¯ v, t ) =
¯
k · ¯ v −ω(¯ r,
¯
k, t ). (2.71)
It should be noticed that, in this expression, the photon wavevector is not an
independent variable because it is determined by
¯
k =
∂L
∂ ¯ v
. (2.72)
24 Photon ray theory
We can also determine the force acting on the photon as the derivative of the
Lagrangian with respect to the coordinates (or the gradient of L):
¯
f =
∂L
∂¯ r
. (2.73)
With these definitions we can reduce the Lagrangian equation
d
dt
∂L
∂ ¯ v

∂L
∂¯ r
= 0 (2.74)
to Newton’s equation of motion
d
¯
k
dt
=
¯
f = −
∂ω
∂¯ r
. (2.75)
Obviously, this also agrees with the second of the Hamiltonian ray equa-
tions (the first one can be seen as the definition of ¯ v = d¯ r/dt ). We have given
here the three equivalent versions of the photon equations of motion in a space–
time varying medium: (1) the Newtonian equation (2.75); (2) the Lagrangian
equation (2.74) and (3) the Hamiltonian equations already stated in section 2.1.
This means that the photon trajectories can be described by the formalism of
classical mechanics, in a way very similar to that of point particles with finite
mass.
Let us explore further this analogy between the photon and a classical parti-
cle. The example of high-frequency transverse photons in an isotropic plasma is
particularly interesting because of the simple form of the associated dispersion
relation: ω =

k
2
c
2
÷ω
2
p
(¯ r, t ). If we multiply this expression by Planck’s
constant (divided by 2π), redefine the photon energy as =
¨
hω and the photon
momentum as ¯ p =
¨
h
¯
k, we get
=

p
2
c
2
÷
¨
h
2
ω
2
p
(¯ r, t ) =

p
2
c
2
÷m
2
eff
c
4
. (2.76)
This means that the photon in a plasma is a relativistic particle with an
effective mass defined by
m
eff
= ω
p ¨
h/c
2
. (2.77)
We see that the photon mass in a plasma is proportional to the plasma fre-
quency (or to the square root of the electron plasma density). Therefore, in a
non-stationary and non-uniform medium, this mass is not a constant because it
depends on the local plasma properties.
In the following, we will use
¨
h = 1. We can also see that the photon velocity
in this medium is determined by v = pc
2
/, or v = kc
2
/ω. The corresponding
relativistic gamma factor is then
γ =

1 −
1
c
2

d¯ r
dt

2

−1/2
=

1 −
k
2
c
2
ω
2

−1/2
=
ω
ω
p
. (2.78)
Photon effective mass 25
We can also adapt Einstein’s famous formula for the energy of a relativistic
particle to the case of a photon moving in a plasma, as
ω = γ ω
p
= m
eff
γ c
2
. (2.79)
Let us now write the photon Lagrangian in an explicit form. Using equa-
tion (2.71), we have
L =
v
c

ω
2
−ω
2
p
−ω = −ω
p

γ −
v
c

γ
2
−1

. (2.80)
This can also be written as
L(¯ r, ¯ v, t ) = −
1
γ
m
eff
c
2
= −ω
p
(¯ r, t )

1 −
v
2
c
2

1/2
. (2.81)
The photon momentum
¯
k, and the force acting on the photon
¯
f , are deter-
mined by
¯
k =
∂L
∂ ¯ v
= −ω
p

∂ ¯ v

1 −
v
2
c
2

1/2
=
ω
p
c
2
γ ¯ v = m
eff
γ ¯ v (2.82)
and
¯
f =
∂L
∂¯ r
= −

1 −
v
2
c
2

1/2
∂ω
p
∂¯ r
= −
ω
p
ω
∂ω
p
∂¯ r
. (2.83)
The Newtonian equation of motion (2.75) can then be written as
d
¯
k
dt
= −
ω
p
ω
∂ω
p
∂¯ r
. (2.84)
The same kind of description can be used for electromagnetic wave propa-
gation in a waveguide. The effective mass of the confined photons will then be
equal to m
eff
= ω
0
/c
2
, where ω
0
is the cut-off frequency for the specific mode of
propagation.
The same equations of motion apply here, but only for one-dimensional
propagation (along the waveguide structure) and, for non-stationary and non-
uniform waveguides, the same effective mass changes and frequency shifts are
expected. The implications of the effective mass of a photon in a waveguide were
considered, in a somewhat speculative maner, by Rivlin [90].
In general terms, a photon always propagates in a medium with velocity less
than c. In the language of modern quantum field theory it can be considered as
a ‘dressed’ photon, because of the polarization cloud. For a generic dispersive
dielectric medium, the photon velocity is determined by
¯ v ≡
∂ω

¯
k
=


¯
k
kc
n
=


¯
k
kc

1 ÷χ(ω)
(2.85)
26 Photon ray theory
where χ(ω) is the susceptibility of the medium. We then have
v =
c

1 ÷χ
(1 ÷χ) ÷(ω/2)(∂χ/∂ω)
. (2.86)
The relativistic γ factor associated with the photon in the medium is then,
by definition,
γ =

1 −
v
2
c
2

−1/2
=
(1 ÷χ) ÷(ω/2)(∂χ/∂ω)
{[(1 ÷χ) ÷(ω/2)(∂χ/∂ω)]
2
−(1 ÷χ)}
1/2
. (2.87)
In the case of a plasma, we have χ = −ω
2
p

2
= −1/γ
2
. This is in
agreement with equation (2.78), as expected. For a non-dispersive medium we
have ∂χ/∂ω = 0, and we get the following simple result:
γ =

1 ÷
1
χ

1/2
=
n

n
2
−1
. (2.88)
The gamma factor tends to infinity when the refractive index tends to one
(the case of a vacuum) because the photon velocity then becomes equal to c. Let
us now establish a definition of the photon effective mass, valid for a dispersive
optical medium. Using Einstein’s energy relation for a relativistic particle, we get
m
eff
=
ω
γ c
2
. (2.89)
This means that, in general, m
eff
is a function of the frequency ω. Exceptions
are the isotropic plasma and the waveguide cases. As a consequence, we are not
allowed to write the dispersion relation as ω = (k
2
c
2
÷ m
2
eff
c
4
)
1/2
, except for
these important but still exceptional cases, because for a refractive index larger
than one this would simply imply imaginary effective masses, in contrast with the
well-behaved definition stated above. For instance, for a non-dispersive dielectric
medium with n > 1 we have, from equation (2.89),
m
eff
=
ω
nc
2

n
2
−1. (2.90)
Let us now write the Lagrangian for a photon in a generic dielectric medium.
Using equation (2.71), we have
L = −ω

1 −
v
c

1 ÷χ

. (2.91)
This reduces to equation (2.80) for a plasma, where χ = −ω
2
p

2
. In con-
trast, for a non-dispersive medium, we have v = c/

1 ÷χ and the Lagrangian is
equal to zero. Notice that, in general, we are not allowed to write L = −m
eff
c
2
/γ ,
except for plasmas and waveguides.
Covariant formulation 27
From equation (2.73) we can calculate the force acting on the photon
¯
f =
∂L
∂¯ r
= ω

∂¯ r
ln n =
1
2
ω
(1 ÷χ)
∂χ
∂¯ r
. (2.92)
This reduces to equation (2.83) for χ = −ω
2
p

2
.
We will now make a final comment on the nature of this force. For a non-
uniform but stationary medium, the force acting on the photon is responsible for
the change in the photon velocity according to the usual laws of refraction: a
gradient of the refractive index always leads to a change in the photon velocity, as
stated by this expression for the force. We have then a variation in the relativistic
γ factor: ∂γ/∂t ,= 0.
However this is not equivalent to photon acceleration, because the total en-
ergy of the photon remains unchanged: ∂ω/∂t = 0. The reason is that the
variation in velocity (or in kinetic energy) is exactly compensated by an equal
and opposite variation in the effective mass (or in the rest energy)

∂t
ln γ = −

∂t
ln m
eff
. (2.93)
If, in contrast, the properties of the medium are also varying with time, this
equality is broken and we can say that photon acceleration takes place. The
medium exchanges energy with the photon, and the photon frequency (or its total
energy) is not conserved: ∂ω/∂t ,= 0.
2.5 Covariant formulation
The similarities between space and time boundaries noted in the preceding sec-
tions suggest and almost compel the use of four-vectors, defined in relativistic
space–time. Let us then define the four-vector position and momentum as
x
i
= {ct, ¯ r}, k
i
=
¸
ω
c
,
¯
k
¸
. (2.94)
The flat space–time of special relativity [43, 55] is described by the
Minkowski metric tensor g
i j
, such that g
00
= −1, g
i i
= 1 (for i = 1, 2, 3) and
g
i j
= 0 (for i ,= j ). From the relation k
i
= g
i j
k
j
, we get k
i
= k
i
(for i = 1, 2, 3)
and k
0
= −k
0
= ω/c.
The total variation of the photon phase (or action) will then be given by
dψ =
¯
k · d¯ r −ωdt =
3
¸
i =1
k
i
dx
i
−k
0
dx
0
= k
i
dx
i
(2.95)
where, in the last expression, we have used Einstein’s summation rule for repeated
indices.
28 Photon ray theory
We can then establish an expression for the variational principle (2.66), as
δψ = δ

k
i
dx
i
= 0. (2.96)
The photon canonical equations can now be written as
dx
i
dx
0
= −
∂k
0
∂k
i
,
dk
i
dx
0
=
∂k
0
∂x
i
. (2.97)
Notice that, for i = 0, the second of these equations simply states that
dω/dt = ∂ω/∂t , and the first one is an identity. For i = 1, 2, 3, we obtain
the above three-dimensional canonical equations.
It should also be noticed that, in these equations, the momentum component
k
0
is not an independent variable, but a function of the other variables k
0

k
0
(k
1
, k
2
, k
3
, x
i
). The explicit form of this function depends on the properties
of the medium. For a generic dielectric medium, we have
k
0
= −
1
n(k
0
, x
i
)

3
¸
j =1
k
j
k
j

1/2
. (2.98)
We can certainly recognize here the dispersion relation ω = kc/n, written
in a more sophisticated notation. The square of the four-vector momentum is
determined by
k
i
k
i
= −(k
0
)
2
÷
3
¸
j =1
(k
i
)
2
= k
2

ω
c

2
. (2.99)
Noting that k = ωn/c, and using n =

1 ÷χ, we obtain
k
i
k
i
= (k
0
)
2
χ(k
0
, x
i
). (2.100)
This is the dispersion relation in four-vector notation. In the case of a plasma,
we simply have
k
i
k
i
= −
ω
2
p
c
2
= −m
2
eff
c
2
. (2.101)
Now, the moving boundary between two dielectric media can be described
by the law
n(x
i
) = n
1
÷
n
2
¸
1 ÷tanh(K
j
x
j
)
¸
(2.102)
where K
j
= {
b
/c =
¯
k
b
· ¯ u/c,
¯
k
b
} is the four-vector defining the space–time
properties of that boundary.
The invariant I
1
can also be written in the same notation as
I
1
= −u
i
k
i
(2.103)
Covariant formulation 29
where the four-vector velocity u
i
= {c, ¯ u} was used.
What we have written until now is nothing but the equations already es-
tablished for the photon equations of motion in the new language of the four-
dimensional space–time of special relativity. This new way of writing can be
useful in, at least, two different ways.
First, it stresses the fact that the apparently distinct phenomena of refraction
and photon acceleration are nothing but two different aspects of a more general
physical feature: the photon refraction due to a variation of the refractive index in
the four-dimensional relativistic space–time. Second, it can be useful for future
generalizations of photon equations to the case of a curved space–time, where
gravitational effects can also be included.
We can now make a further qualitative jump. Given the formal analogy
between equation (2.95) and the phase defined in the usual three-dimensional
space coordinates, we can generalize it and write
dψ = k
i
dx
i
−h(x
i
, k
i
) dτ. (2.104)
Here we have used a timelike variable τ, to be identified later with the photon
proper time, and a new Hamiltonian function such that h(x
i
, k
i
) = 0. This
function is sometimes called a super-Hamiltonian [33, 76].
The covariant form of the variational principle (2.96) can now be written as
δφ = δ

k
i
dx
i
−h dτ
=

δk
i

dx
i

∂h
∂k
i



dk
i
÷
∂h
∂x
i

δx
i
= 0 (2.105)
where the integration is performed between two well-defined events in space–
time. For arbitrary variations δk
i
and δx
i
, we can derive from here the photon
canonical equations in covariant form:
dx
i

=
∂h
∂k
i
,
dk
i

= −
∂h
∂x
i
. (2.106)
In contrast with the above equations of motion (2.97), these new equations
are formally analogous to the three-dimensional canonical ray equations (2.12).
The question here is how to define the appropriate function h(x
i
, k
i
) such that
these new canonical equations are really equivalent to equations (2.97). This is
simple for the plasma case, but not obvious for an arbitrary dielectric medium.
We notice that, for a plasma, the photon effective mass is not a function of
the frequency. In this case, we can use the analogue of the covariant Hamiltonian
for a particle with a finite rest mass:
h(x
i
, k
i
) =
k
j
k
j
2m
eff
÷
1
2
m
eff
c
2
. (2.107)
30 Photon ray theory
This is equivalent to
h(x
i
, k
i
) =
(k
j
k
j
)c
2

p
(x
i
)
÷
1
2
ω
p
(x
i
)
=
1

p
(x
i
)
¸
k
2
c
2
÷ω
2
p
(x
i
) −ω
2
¸
. (2.108)
It can easily be seen that, for i = 0, the use of this Hamiltonian function
in equations (2.106) leads to the appropriate definition of the relativistic γ factor.
For i = 1, 2, 3, it leads to equations (2.12), just showing that equations (2.106)
are indeed an equivalent form of the photon canonical equations.
In the general case of a dielectric medium with refractive index n (where
the photon effective mass is, in general, a function of the photon frequency), the
covariant Hamiltonian h(x
i
, k
i
) = 0 has to satisfy the following conditions:
∂h
∂ω
= −γ,
∂h
∂t
= −γ ω

∂t
ln n (2.109)
and
∂h

¯
k
= γ ¯ v,
∂h
∂¯ r
= −γ ω

∂¯ r
ln n. (2.110)
For a non-dispersive medium, where v = c/n, we see that these four condi-
tions are satisfied by the covariant Hamiltonian
h(x
i
, k
i
) = γ

kc
n
−ω

=
γ c
n(x
i
)
¸
k
0
÷

3
¸
i =1
k
i
k
i

1/2
¸
. (2.111)
The photon ray theory for non-stationary plasmas was first proposed by
us [66]. This work has been considerably extended in the present chapter.
The covariant formulation was considered in reference [65], where the
Hamiltonian (2.107) was first stated.
Chapter 3
Photon dynamics
Two different kinds of moving perturbation can be imagined in a non-stationary
medium. The first one is a shock front or moving discontinuity, similar to the
moving boundary between two media considered in the previous chapter. The
second is a wave-type perturbation of the refractive index.
Both kinds of perturbation can be excited in a plasma by an intense laser
pulse. They can also be analytically described in a simple way. The moving
boundary will be the ionization front created by the laser pulse, and the wave
perturbation will be the wakefield left behind the pulse.
For the sake of simplicity we will focus here on the dynamical properties
of the photon motion in a non-magnetized plasma, in the presence of ionization
fronts and of wakefields. At the end of this chapter, we will discuss the ways in
which the same kind of perturbation can also be excited in magnetized plasmas
and in other optical media.
As a particular example of photon acceleration in a dielectric medium, we
will consider the well-known induced phase modulation processes.
3.1 Ionization fronts
One simple way of producing a moving discontinuity of the refractive index is to
create an ionization front, which is the boundary between the neutral state and the
ionized state of a given background gas. In other words, the ionization front is the
boundary between a neutral and a plasma medium. The motion of this boundary
is independent of the motion of the atoms, ions or electrons of the medium, which
means that we can eventually produce a relativistic moving boundary in a medium
where the particles stay nearly at rest.
The ionization front is one of the main problems in the theory of photon
acceleration in plasmas, as referred to in chapter 1. A relativistic front can be
efficiently produced by photoionization of a gas by an intense laser pulse [4, 47].
Because the photoionization is a fast process, occurring within a timescale of a
31
32 Photon dynamics
laser
pulse
electron
density
n / n
q = x - u t
e 0
Figure 3.1. Ionization front scheme: an intense laser pulse photoionizes a background gas.
The resulting plasma boundary follows the pulse, moving at nearly the same speed.
few femtoseconds, the front will tend to replicate the form of the ionizing laser
pulse and will move with nearly the same velocity.
The formation and propagation of such a front is a very complicated process.
First of all, the laser pulse is moving at the boundary between two different media
and, for that reason, its group velocity cannot be determined in a simple way.
Furthermore, its spectral content is also changing, due to the self-frequency
shift, or acceleration of the laser photons by the laser itself, which will be dis-
cussed later. On the other hand, the photoionization process cannot be isolated
from other ionization mechanisms because the primary electrons (created and
subsequently accelerated by the laser field) will also contribute to the ionization
process, by electron-impact ionization [60].
Here we are not interested in the details of the ionization front structure. It is
only interesting to note that, behind the ionizing laser pulse, the plasma state will
have a much longer lifetime than the duration of the pulse, because of the long
timescales of the diffusion and recombination processes, which will eventually
destroy it. We will therefore restrict our attention to the dynamics of individual
probe or test photons, interacting with a given ionization front.
We start with a discussion of the different types of photon trajectory, as given
by the one-dimensional ray equations
dq
dt
=

∂p
,
dp
dt
= −

∂q
(3.1)
where q = x − ut , p = k and u is the front velocity. We restrict our analysis to
high-frequency photons, which is pertinent for laser propagation.
Ionization fronts 33
We can then neglect the contribution of the ion response and, for an isotropic
or non-magnetized plasma, the Hamiltonian function is simply determined by
≡ (q, p, t ) =

p
2
c
2
÷ω
2
p0
f (q) −up. (3.2)
Here, ω
p0
is the maximum value of the electron plasma frequency attained
by the ionization front and f (q) is a function increasing with q from 0 to 1
and describing the form of the front. One possible choice, already mentioned
in chapter 2, is f (q) = [1 ÷tanh(k
f
q)]/2, where k
f
determines the front width.
Another possible choice, for fronts created by a single step ionization of
the background neutral atoms, by a Gaussian pulse, is f (q) = exp(−k
2
f
q
2
), for
q < 0, and f (q) = 1, for q ≥ 0. Other, more complicated but more realistic
forms of front profiles can also be proposed, but will not significantly modify our
view of the various possible classes of photon trajectories.
From equation (3.2) we can determine the photon trajectory p(q) in the two-
dimensional phase space (q, p)
p(q) = γ
2

c
¸
¸
sβ ±

1 −
1
γ
2
ω
2
p0

2
f (q)

(3.3)
where β = [u[/c, s is the sign of the velocity u and γ
−2
= (1 −β
2
).
From this result we can establish two different conditions for photon reflec-
tion by the front. First, we can say that a reversal in direction of the photon
trajectory in phase space (q, p) will occur if
∂p
∂q
→−∞. (3.4)
Using equation (3.3), we see that such a condition is satisfied if, for a given
point q = q
r
, we have
=
ω
p0
γ

f (q
r
). (3.5)
Noting that the maximum value for f (q) is equal to 1, we can say that the
photon trajectory will be sooner or later reflected in phase space if < ω
p0
/γ .
For a photon with initial frequency ω
1
, which moves with a positive velocity
along the x-axis (or, equivalently, along the q-axis) and then interacts with a front
moving in the oposite direction (such that s = −1), we can write = ω
1
(1 −
sβ) = ω(1 ÷β).
This allows us to establish an upper limit ω
q
for the frequencies of those
photons with trajectories which reverse sense in phase space (q, p)
ω
1
<
ω
p0
1 ÷β

1 −β
2
≡ ω
q
. (3.6)
34 Photon dynamics
On the other hand, we have reflection in real space (a reflection observed in
the laboratory frame of reference) if there is a turning point x where the photon
momentum reverse changes its sign: p = k = 0. According to equation (3.3) this
implies that
= ω
p0

f (q). (3.7)
More generally, we can say that reflection in real space occurs if < ω
p0
.
This allows us to establish a new upper limit ω
x
for the frequency of photons with
this kind of behaviour
ω
1
<
ω
p0
1 ÷β
≡ ω
x
. (3.8)
From equations (3.6, 3.8) we can get a simple relation between the two upper
limits
ω
x
=
ω
q

1 −β
2
. (3.9)
This shows that we will always have ω
x
> ω
q
. We are now ready to establish
the distinct regimes of photon propagation across the ionization front or, in other
words, the different types of trajectory in phase space. If we assume that the
parameter ω
p0
is fixed and if we vary the value of the incident photon frequency
ω
1
, we successively obtain:
(a) type I trajectories, for low-frequency photons, such that ω
1
< ω
q
—there is
reflection in both the real space and the phase space;
(b) type II trajectories, for moderate values of the frequency, such that ω
q
<
ω < ω
x
—there is reflection in real space but transmission (no reversal) in
phase space. This means that the reflected photon and the front move both
in the backward direction, but the front moves faster, which means that the
photon can still cross the entire front region;
(c) type III trajectories, for high-frequency photons with frequencies such that
ω
1
> ω
x
—there is transmission in both the real space and the phase space.
If instead, we had assumed a fixed incident frequency ω
1
and had increased the
value of the maximum plasma frequency of the front, we would have type III
trajectories for 0 < ω
p0
< ω
a
, where ω
a
= ω
1
(1 ÷ β), type II trajectories for
moderate densities ω
a
< ω
p0
< ω
b
, where ω
b
= γ ω
a
, and type I trajectories for
high-density trajectories, such that ω
p0
> ω
b
. These three types of trajectory are
illustrated in figure 3.2.
The maximum values for the photon frequency shifts can easily be obtained
from the invariance of , as already shown in chapter 2, independently of the
choice of the form function f (q). For type I trajectories, we have
ω
2
= ω
1
1 ÷β
1 −β
(3.10)
where ω
2
is the final value of the frequency. This result is identical to that of
reflection of an incident photon in a relativistic mirror with normalized velocity
β.
Ionization fronts 35
-10 -5 0 5 10
-1.2x10
7
-1.0x10
7
-8.0x10
6
-6.0x10
6
-4.0x10
6
-2.0x10
6
0.0
2.0x10
6
4.0x10
6
6.0x10
6
Reflection, type I
Reflection, type II
Transmission, type III
Co-propagation, type IV (*)
W
a
v
e
n
u
m
b
e
r

[
m
-
1
]
q
Figure 3.2. Trajectories of photons interacting with an ionization front. The three
counter-propagating types of trajectory are shown for the following parameters: front
velocity β = 0.9, electron density n
e
= 5 10
20
cm
−3
. A co-propagating trajectory
is also shown, but for a different electron density n
e
= 10
20
cm
−3
, for numerical reasons.
This trajectory would not intersect the others if the plasma parameters were the same.
-10 0 10
0
2
4
6
8
10
12
Reflection, type I
Reflection, type II
Transmission, type III
Co-propagation, type IV (*)
V
a
c
u
u
m

w
a
v
e
l
e
n
g
t
h
[
µ
m
]
q
Figure 3.3. The same four trajectories as in figure 3.2, but represented in terms of the
vaccuum wavelength λ = 2πc/ω, which is the prefered unit for the experimentalists to
denote the measured photon frequency ω.
This is a very interesting result because, as we explained above, the particles
of the media can stay at rest, and we still have a moving mirror effect.
The boundary between two different states of the medium (the neutral and
the plasma state) act then as a moving material obstacle. Such an equivalence be-
36 Photon dynamics
tween the ionization front and a material moving mirror is however not complete,
as discussed in chapter 4.
Using the same kind of argument, we can also easily obtain for type II and
III trajectories the following frequency transformation:
ω
2
=
ω
1
1 −β
¸
1 −β

1 −(ω
p0

1
)
2
(1 −β)
¸
. (3.11)
For high-frequency photons, such that ω
1
·ω
p0
, this leads to the following
frequency shift:
ω = ω
2
−ω
1
.
1
2
ω
2
p0
ω
1
β
1 ÷β
. (3.12)
As expected, the frequency shift for these highly energetic photons interact-
ing with low-density fronts is very small. The interesting thing however is that
even in this case a positive shift is expected. Actually, for β . 1, the frequency
shift is half of the value expected for the case of flash ionization (β → ∞). This
limiting case, and its physical meaning, will be discussed later.
Until now we have only considered photons counter-propagating with re-
spect to the moving front. However, in order to get the full picture of the pho-
ton phase space, we have to consider type IV trajectories, corresponding to co-
propagation. In this case, the frequency shift is
ω
2
=
ω
1
1 −β
¸
1 −β

1 −(ω
p0

1
)
2
¸
. (3.13)
For high frequencies ω
1
· ω
p0
, this reduces to an expression similar to
equation (3.12), but where (1 ÷ β) is replaced by (1 − β), in the denominator.
This means that, in co-propagation, much larger frequency shifts are expected.
Furthermore, two independent measurements of the frequency shifts, for
co- and counter-propagation, will allow us to determine the two independent
parameters of the ionization front: the normalized velocity β and the maximum
plasma density or electron plasma frequency ω
p0
.
Such a approach was followed in the experiments by Dias et al [21], who es-
tablished a proof of principle of the photon accelerated process and also made the
first observation of frequency shift in counter-propagation. Previous experiments
of frequency up-shift by an ionization front propagating inside a microwave cav-
ity, where essentially the co-propagating signal was observed, were reported by
Savage et al [95]. Several other observations of frequency up-shift in microwave
experiments were reported [52, 121].
These four types of trajectory correspond to four distinct regions in phase
space, which are delimited by three separatrix curves. If we use = ω
p0
/γ in
equation (3.3), we obtain an expression for two of the separatrix curves
p(q) = γ
ω
p0
c
¸
−β ±

1 − f (q)
¸
. (3.14)
Ionization fronts 37
p
q
type I
type II
type III
type IV
Figure 3.4. Photon interaction with ionization fronts: the three separatrix curves and the
four types of trajectory.
If, instead, we use = ω
p0
, we get the third separatrix
p(q) = γ
2
ω
p0
c
¸
−β ÷

1 − f (q)/γ
2
¸
. (3.15)
These three curves, and the corresponding four regions of phase space, are
shown in figure 3.4.
In the above discussion we have used an arbitrary form function f (q) but
where f (q) = 0 for q → −∞. Another possible version of the front assumes
that the ionizing laser pulse is propagating not in a neutral gas, but in a partially
ionized gas, with a residual plasma frequency equal to δω
p0
, with δ < 1. This
model of the front was considered by Kaw et al [46]. It can be described by the
following form function:
f (q) = δ ÷
1
2
(1 −δ)[1 ÷tanh(k
f
q)]. (3.16)
This improvement in the model slightly changes the photon trajectories and
the values of the resulting frequency shifts, but the basic picture of the photon
phase space remains valid. For type I trajectories, the invariance of now leads
to the expression
ω
2
=

1
)
1 −β
2
¸
1 ÷β

1 −δ(1 −β
2
)[ω
p0
/(ω
1
)]
2
¸
. (3.17)
Here, we have used (ω
1
) = ω
1
[1 ÷ β

1 −δ(ω
p0

1
)
2
]. For δ = 0 this
reduces to equation (3.10), and, for δ ,= 0 but for ω
1
. ω
p0
, this simplifies to
ω
2
=
ω
1
1 −β
2
¸
1 ÷β
2
÷2β

1 −δ
¸
. (3.18)
38 Photon dynamics
Similarly, for type II and III trajectories, equation (3.11) is generalized to
ω
2
=

1
)
1 −β
2
¸
1 −β

1 −(1 −β
2

p0
/(ω
1
)
2
¸
(3.19)
where (ω
1
) is the same as in equation (3.17).
Finally, for the co-propagating type IV trajectories, we have, instead of equa-
tion (3.13),
ω
2
=

/

1
)
1 −β
2
¸
1 ÷β

1 −δ(1 −β
2
)[ω
p0
/
/

1
)]
2
¸
. (3.20)
Here we have to use
/

1
) = ω
1
[1 −β

1 −(ω
p0

1
)
2
]. Let us now turn
to another (and eventually more realistic) model of the ionization front where, in
the absence of the front, we have a neutral gas, but where the influence of the
neutral gas on the refractive index n is not neglected as it was before.
If we consider a value n ,= 1 (but still neglect dispersion in the neutral
region), we can use the following model:
(q, p) =
¸
p
2
c
2
1 ÷χ[1 − f (q)]
÷ω
2
p0
f (q)
¸
1/2
−up. (3.21)
Here χ = n
2
− 1 is the susceptibility of the background neutral gas, and a
singly ionized plasma is assumed. With this model we still get the same picture
of the photon phase space and similar expressions for the frequency shift. For
instance, for type I trajectories, we have now
ω
2
= ω
1
1 ÷βn
1 −βn
. (3.22)
For types II and III, we get
ω
2
= ω
1
1 ÷βn
1 −β
2

¸
¸
1 −β
¸
1 −
ω
2
p0
ω
2
1
(1 −β)
2
(1 ÷βn)
2

1/2
¸
¸
¸
(3.23)
and, for type IV trajectories, we obtain
ω
2
=
ω
1
1 −βn
¸
1 −β

1 −(ω
p0

1
)
2
¸
. (3.24)
Let us briefly consider oblique interaction of photons with an infinite ioniza-
tion front. In this case, the generalization of the above one-dimensional analysis
to the two-dimensional case allows us to determine the cut-off frequencies.
Accelerated fronts 39
In particular, the condition for the existence of a type I trajectory (corre-
sponding to reflection both in real space and in phase space) corresponds to
incident photon frequencies ω
1
smaller than some limit ω
q
, defined by
ω
q
=
ω
p0

1 −β
2

[1 ÷β cos(θ
1
)]
2
−(1 −β
2
) sin
2

1
)
(3.25)
where β = [u[/c.
For normal incidence, such that θ
1
= 0, this expression reduces to equa-
tion (3.6). But a new and very interesting physical situation occurs when the
incident photon is co-propagating in the vacuum region (or, more precisely, in the
neutral gas region) and is overtaken by the ionization front.
Even if the photon travels with a velocity nearly equal to c, its velocity
perpendicular to the front can be much smaller due to oblique propagation. In this
case, we have π/2 ≤ θ
1
≤ π and photon reflection can still take place, as shown
by the above expression. This new effect can be called co-propagating relativistic
mirror and it is due to oblique interaction with the front. The corresponding
frequency shift can be obtained from the expressions of the two invariants I
1
and
I
2
, defined in chapter 2, and the result is
ω
2
= ω
1
¸
1 ÷β cos(θ
1
) ÷β[β ÷cos(θ
1
)[
1 −β
2
¸
. (3.26)
This effect was observed in the experiments by Dias et al [21] mentioned
already.
3.2 Accelerated fronts
A front moving with constant velocity can only be considered as an ideal and
limiting case. It is therefore quite natural to extend the above analysis to the
more general situation of accelerated ionization fronts. This can easily be done
by considering a plasma frequency space–time dependence of the form
ω
2
p
(¯ r, t ) = ω
2
p
(¯ r −
¯
R(t )). (3.27)
For a density perturbation moving with constant velocity, we would simply
have
¯
R(t ) = ¯ vt , which is the case discussed in the previous section. But now we
can have d
2
¯
R/dt
2
,= 0, which corresponds to accelerated density perturbations.
In order to treat our problem we can easily generalize the above canonical
transformation (¯ r,
¯
k) →(¯ q, ¯ p), using the following generating function:
F(¯ r,
¯
k, t ) =
¯
k · (¯ r −
¯
R(t )). (3.28)
This leads to
¯ q =
∂ F

¯
k
= ¯ r −
¯
R(t ), ¯ p =
∂ F
∂¯ r
=
¯
k. (3.29)
40 Photon dynamics
The transformed Hamiltonian is
= ω ÷
∂ F
∂t
=

p
2
c
2
÷ω
2
p
(¯ q) − ¯ p ·

¯
R
∂t
. (3.30)
Now, the velocity of the front ∂
¯
R/∂t is not constant, which means that is
no longer a constant of motion. New qualitative aspects of photon acceleration
can be associated with accelerated fronts.
For co-propagation (photons and front propagating in the same direction)
we can improve the frequency up-shifting by increasing the interaction time. For
counter-propagation (photons and front propagating in the opposite sense) two
sucessive reflections can take place.
We will illustrate these new aspects, by focusing our discussion on the one-
dimensional problem. As before, we can write ω
2
p
(q) = ω
2
p0
f (q). But, instead
of using our tanh model, we will assume an even simpler form for the ionization
front, described by
ω
2
p
(q) =

¸
¸
0 (q < −1/k
f
)
ω
2
p0
(1 ÷k
f
q) (−1/k
f
< q < 0)
ω
2
p0
(q > 0)
. (3.31)
The width of the front is obviously equal to 1/k
f
. It is also clear that the
photon frequency will only change as long as the photon travels inside the gradient
region, or when the photon coordinates are such that
−1/k
f
< q(t ) < 0. (3.32)
Inside that region, the photon frequency is determined by the expression
ω(t ) =

p
2
(t )c
2
−ω
2
p0
k
f
q(t ). (3.33)
In order to obtain an explicit expression for the frequency shift, we have to
consider the photon equations of motion, allowing us to determine q(t ) and p(t ).
They can be written as
dq
dt
=
pc
2
ω(t )

∂ R
∂t
(3.34)
dp
dt
=
1
2
ω
2
p0
k
f
ω(t )
. (3.35)
These equations can easily be integrated by noticing that equation (3.33)
allows us to write
q(t ) = p
2
(t )
c
2
ω
2
p0
k
f
÷ Q(t ) (3.36)
Accelerated fronts 41
with
Q(t ) = −
ω
2
(t )
ω
2
p0
k
f
. (3.37)
Using equations (3.34, 3.35) we realize that the total time derivative of equa-
tion (3.36) leads to
dQ
dt
= −
∂ R
∂t
. (3.38)
This can be integrated to give
Q(t ) = Q
0
− R(t ). (3.39)
Going back to equations (3.33, 3.37), we obtain
ω(t ) =

ω
2
0
÷ω
2
p0
k
f
R(t ) (3.40)
where ω
0
≡ ω(t = 0) is the initial value of the photon frequency (just before
entering the acceleration region).
The constant of integration in equation (3.39) is Q
0
= −ω
2
0
/(ω
2
p0
k
f
), as can
be seen from equation (3.37).
In the particular case of a front moving with constant velocity, we have
R(t ) = v
f
t . This expression shows that the photon frequency grows with

t ,
as first noticed by Esarey et al [27]. This expression is valid as long as the photon
stays inside the gradient region (3.32). It means that the closer the photon velocity
is to the front velocity v
f
, the longer t will be and the larger the final value of the
frequency shift will be, as already shown in the previous section.
However, because of the photon acceleration process itself, the photon will
have a tendency to escape from the gradient region, and a phase slippage between
the photons and the front will always exist. Such a slippage can however be
avoided (or at least significantly reduced) if the ionization front is accelerated as
well.
Then, in principle, we can imagine an ideal situation where the photon co-
propagating with the front is indefinitely accelerated up to arbitrary high frequen-
cies, just by using a very small (but accelerated) plasma density perturbation. The
external source responsible for the creation of such an accelerated plasma density
perturbation would provide the energy necessary to accelerate the co-propagating
photons.
Let us illustrate this idea with the simple case of a front moving with constant
acceleration a:
R(t ) = v
f
t ÷
1
2
at
2
. (3.41)
In this case, we can estimate the maximum frequency up-shift for very un-
derdense fronts (or, equivalently, for very high-frequency photons), such that
ω
2
0
· ω
2
p0
. The photon velocity can be assumed to be nearly equal to c and
42 Photon dynamics
the total interaction time can be determined by the equality ct
in
− R(t
in
) . 1/k
f
.
The solution is
t
in
=
c
a
¸
(1 −β
f
) −

(1 −β
f
)
2
÷2a/(k
f
c
2
)
¸
(3.42)
where we have used β
f
= [v
f
[/c. For small accelerations, such that at
in
<
2c(1 − β
f
), we can simplify this expression and, using equation (3.40), we can
write
ω ≡ ω(t
in
) −ω
0
.
ω
2
p0

2
0
β
f
1 −β
f
¸
1 ÷
a
2(1 −β
f
)
2
c
2
¸
. (3.43)
The first term in this expression coincides with that derived in the previ-
ous section for fronts moving with constant velocity. The second one gives the
contribution of the front acceleration. We can see that a positively accelerated
front (a > 0) introduces a significant increase in the total frequency up-shift,
especially due to the factor (1−β
f
)
3
. This illustrates our previous statement that,
by increasing the time of interaction between the photons and the front, we can
optimize the photon acceleration process.
If we are not entirely satisfied with these qualitative arguments and want to
replace them by a more detailed description of the photon dynamics, we can go
back to equations (3.34, 3.35, 3.40) and derive explicit expressions for the photon
trajectories. In particular, from equation (3.35), we can get
dp
dt
=
1
2
ω
p0

k
f

−Q(t )
. (3.44)
The photon trajectory will then be determined by
p(t ) = p
0
÷
ω
p0
2

k
f

t
0
dt
/

R(t
/
) − Q
0
. (3.45)
For positively accelerated fronts (a > 0), this leads to
p(t ) = p
0
÷ω
p0

k
f
/2a ln

ω(t )ω
p0

k
f
/2a
ω
0
÷ω
p0
v
k

k
f
/2a
∂ R
∂t

. (3.46)
With equation (3.36), this completes the explicit integration of the photon
trajectories. Until now we have focused our qualitative discussions on photons co-
propagating with the front. But it can easily be seen that the above equations stay
valid for the counter-propagating photons, except that in equations (3.42, 3.43)
the factor (1 −β
f
) is replaced by (1 ÷β
f
). The frequency shift is still increased
by a positive front acceleration.
However, a qualitatively new effect occurs for co-propagating photons: more
than one turning point, defined by
dq
dt
= 0 (3.47)
Photon trapping 43
can be observed along the same trajectory.
This means that a counter-propagating photon can be first reflected by the
front, suffering the corresponding double Doppler shift, and then it is caught again
by the front, and reflected a second time into the co-propagating direction. These
doubly reflected trajectories were numerically identified by Silva [99] and can
attain considerably large frequency shifts.
3.3 Photon trapping
3.3.1 Generation of laser wakefields
It is well known that a plasma can support several kinds of electrostatic wave and
oscillations. One of the basic types of such waves is called an electron plasma
wave, because its frequency is nearly equal to the electron plasma frequency [82,
108]. To be more precise, the electron plasma waves are characterized by the
following dispersion relation:
ω
2
= ω
2
p
÷3k
2
v
2
the
(3.48)
where ω and k are here the frequency and wavenumber of the electrostatic oscil-
lations, and v
the
=

T
e
/m is the electron thermal velocity and T
e
the electron
temperature.
The electron plasma waves can only persist in a plasma if their wavenumber
satisfies the inequality k < ω
p
/v
the
. Otherwise, they will be strongly attenuated
by electron Landau damping.
Such a damping is due to the energy exchange between the waves and the
electrons travelling with a velocity nearly equal to the wave phase velocity v
φ
=
ω/k (the so-called resonant electrons). This is equivalent to saying that the
electron plasma waves can only exist if their wavelength is much larger than a
characteristic scale length of the plasma λ < λ
De
, where λ
De
= v
the

p
is called
the electron Debye length.
The existence of the electron Landau damping implies that the frequency of
the allowed electron plasma waves is always very close to the electron plasma
frequency, ω ∼ ω
p
. As can be seen from the above dispersion relation, the phase
velocity of the electron plasma waves can be arbitrarily high: v
φ
v
g
= 3v
2
the
.
It is then possible to excite electrostatic waves with relativistic phase veloc-
ities v
φ
. c, and nearly zero group velocities. These waves will not be Landau
damped, because resonant electrons will be nearly absent (except for extremely
hot plasmas).
Relativistic electron plasma waves can be excited by intense laser beams in
a plasma. Using a simplified but nevertheless accurate view of this problem, we
can say that there are two different excitation mechanisms: the beat-wave and the
wakefield.
The first mechanism[64, 77, 92] is valid for long laser pulses, with a duration
t much larger than the period of the electron plasma wave: t · ω
−1
p
. In this
44 Photon dynamics
laser
pulse
wakefield
Figure 3.5. Schematic of the laser wakefield mechanism. A short laser pulse propagating
in a plasma excites a relativistic electron plasma wave.
case, the superposition of two parallel laser pulses with different but very close
frequencies, ω
1
and ω
2
, such that their difference matches the electron plasma
frequency, ω
1
− ω
2
. ω
p
can resonantly excite an electron plasma wave. The
phase velocity of this wave is determined by the laser beating v
φ
= (ω
1

ω
2
)/(k
1
−k
2
).
The second mechanism [12, 34, 89] is valid for short laser pulses, with a
duration of the order of the electron plasma period t . ω
−1
p
. Asingle laser pulse
can then excite a tail of electron plasma oscillations, which is usually called the
laser wakefield. It is important to notice that the phase velocity of this wakefield
is nearly equal to the group velocity of the laser pulse.
With both the beat-wave and the wakefield mechanisms, relativistic electron
plasma waves can be excited. Interaction of these waves with fast electrons can
accelerate them further and provide the basis for a new generation of particle
accelerators [109].
The problem of electron acceleration is not our concern here. Our interest
is mainly focused on the analogies between charged particles and photons and,
especially, on the possibility of accelerating and trapping photons in the field
of an electron plasma wave. But, it should also be mentioned that the study of
photon acceleration and trapping by a relativistic plasma wave can be an important
diagnostic tool for particle accelerator research [22, 107].
3.3.2 Nonlinear photon resonance
From now on, we will simply refer to a relativistic electron plasma wave as
a wakefield. In order to describe such a wakefield, we will assume a plasma
Photon trapping 45
frequency perturbation of the form
ω
2
p
(¯ q) = ω
2
p0
¸
1 ÷ cos(
¯
k
p
· ¯ q)
¸
(3.49)
where < 1 is the relative amplitude of the wakefield and
¯
k
p
is the wakefield
wavevector, and the subscript p is added in order to avoid confusion with the
photon wavevector.
The photon equations of motion can be explicitly written as
d¯ q
dt
=
c
2
¯ p
ω(¯ q, ¯ p)
− ¯ u (3.50)
d ¯ p
dt
=

2
ω
2
p0
¯
k
p
ω(¯ q, ¯ p)
sin(
¯
k
p
· ¯ q) (3.51)
where ¯ u . (ω
p0
/[k
p
[
2
)
¯
k
p
is the wakefield phase velocity and
ω(¯ q, ¯ p) =
¸
p
2
c
2
÷ω
2
p0
[1 ÷ cos(
¯
k
p
· ¯ q)]
¸
1/2
. (3.52)
Let us determine the fixed points, defined by
d¯ q
dt
= 0,
d ¯ p
dt
= 0. (3.53)
This is equivalent to writing
c
2
¯ p = ω(¯ q, ¯ p)¯ u, sin(
¯
k
p
· ¯ q) = 0. (3.54)
At this point we could split the variables ¯ q and ¯ p into their parallel and
perpendicular components with respect to the direction of the wakefield propa-
gation ¯ u/u, or
¯
k
p
/k
p
. However, it can easily be realized that the perpendicular
components play only a secondary role in the photon dynamics.
We will now proceed by just retaining the simple case of one-dimensional
motion, where q
|
= q, p
|
= p, and q

= p

= 0. From equations (3.53) we
can then explicitly establish the elliptic fixed points as
q
e
=
π
k
p
, p
e
= sγβ
ω
p0
c

1 − (3.55)
and the hyperbolic fixed points as
q
h
= 0,

k
p
, p
h
= sγβ
ω
p0
c

1 ÷. (3.56)
Here we have used β = [u[/c, s = sign of u and γ = (1 − β
2
)
−1/2
. We
should notice that, in contrast with the simple pendulum, the momenta p
e
and p
h
46 Photon dynamics
p
k

q
p
Figure 3.6. The separatrix curves of the photon nonlinear resonance in phase space (q, p).
corresponding to the elliptic and hyperbolic fixed points are not identical. The
photon nonlinear resonance described by equations (3.50, 3.51) appears then as a
kind of asymmetric pendulum [98].
Furthermore, the resonance asymmetry increases with the wakefield rela-
tivistic γ -factor. For small wakefield amplitudes <1, the distance between the
two fixed points can be written as
[ p
e
− p
h
[ . γβ
ω
p0
c
. (3.57)
Aqualitative representation of the photon nonlinear resonance in phase space
(q, p) is shown in figure 3.6.
It is also interesting to notice that, for the one-dimensional case, the invariant
can be written as
= ω(q, p) −up = γ
−2
ω(q, p). (3.58)
This means that the elliptic fixed point given by equation (3.55) can be
defined by
p
e
= sγ
2
β

e
c
2
(3.59)
where

e
=
ω
p0
γ

1 −. (3.60)
The value of p
h
, for the hyperbolic fixed point, is also determined by an
expression similar to equation (3.59), where
e
is replaced by
h
such that

h
=
ω
p0
γ

1 ÷. (3.61)
Photon trapping 47
Replacing this expression in the definition of the invariant Hamiltonian ,
we obtain the equation for the separatrix
(
h
−up)
2
= p
2
c
2
÷ω
2
p
(q), (3.62)
or, more explicitly,
p = γ
ω
p0
c

1 ÷
¸
¸
sβ ±

1 −
1 ÷ cos(k
p
q)
1 ÷

. (3.63)
For q = 0, (2π/k
p
), this reduces to p = p
h
. On the other hand, for q =
(π/k
p
), this expression determines the maximum and the minimum values of the
photon momentum on the separatrix
p
±
= γ
ω
p0
c

1 ÷
¸
sβ ±

2
1 ÷

. (3.64)
The width of the nonlinear resonance will then be given by the difference
between these two extreme values of the separatrix curves:
p
max
= p
÷
− p

= 2γ
ω
p0
c

2. (3.65)
From this analysis we recognize that the photons are trapped by the elec-
tron plasma wave if their trajectories are associated with values of the invariant
Hamiltonian such that

e
≤ <
h
. (3.66)
Inside the separatrix, the photon frequency will oscillate according to the
value of characterizing its trajectory. We will have no frequency variation
only for exactly resonant photon trajectories such that p = p
e
, or =
e
,
corresponding to the elliptic fixed point.
The frequency variation will grow when we approach the separatrix curve,
but such a change will require longer and longer times to take place. We will also
attain a maximum for the frequency variation at the separatrix, between the points
p
÷
and p

defined by equation (3.64), if we wait an infinite time.
For very high-frequency photons, such that ω · ω
p0
, we have ω . pc
and the maximum frequency shift will be determined by ω
max
. cp
max
=
2γ ω
p0

2. For untrapped photons, the photon frequency will also oscillate, but
with smaller and smaller amplitudes when we displace the photon trajectories
away from the separatrix. This is illustrated in figure 3.7.
It is now interesting to look at the deeply trapped trajectories, oscillating
around p = p
e
. If we introduce q = (π/k
p
) ÷ ˜ q and linearize the photon
equations of motion around the elliptic fixed point, we can easily obtain
d
2
˜ q
dt
2
÷ω
2
b
˜ q = 0 (3.67)
48 Photon dynamics
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
0
1
2
3
4
5
V
a
c
u
u
m

w
a
v
e
l
e
n
g
t
h

[
µ
m
]
Position on the plasma wave [Plasma wavelength]
Figure 3.7. Examples of photon trajectories around a nonlinear resonance, showing
trapped and untrapped motion, corresponding to the following parameters: β = 0.9995,
n
e
= 10
18
cm
−3
, = 0.5. The photon frequency ω is represented in terms of the vacuum
wavelength λ = 2πc/ω.
which corresponds to a linear oscillator with a frequency
ω
b
=
ck
p
γ

2(1 −)
. (3.68)
This can be called the photon bounce frequency. It is instructive to compare
its value with the bounce frequency of an electron trapped in the field of the same
electron plasma wave [82]
ω
be
=

(e/m)E
0
k
p
(3.69)
where E
0
is the amplitude of the electric field associated with this wave.
Using Poisson’s equation we can relate it to the electron density amplitude
perturbation
k
p
E
0
=
e ˜ n

0
=
en
0

0
. (3.70)
This allows us to write the electron bounce frequency as
ω
be
= ω
p0

. (3.71)
This shows the striking similarities of the photons and electrons trapped in
the field of an electron plasma wave. Other and eventually even more surprising
similarities will be discussed later.
Photon trapping 49
3.3.3 Covariant formulation
As an example of an application of the covariant equations established in sec-
tion 2.5, we can describe the photon trapping by using equations (2.106, 2.108)
with
ω
2
p
(x
i
) = ω
2
p0
¸
1 ÷ cos(K
i
x
i
)
¸
(3.72)
where K
i
is the four-momentum associated with the plasma wakefield and is,
as before, its amplitude. Let us write
K
i
x
i
= K
1
x
1
− K
0
x
0
= k
/
x
1
−ω
/
t (3.73)
where ω
/
. ω
p0
is the wakefield frequency and is related to k
/
by the dispersion
relation (3.48).
In this particular case, the covariant Hamiltonian (2.108) is independent of x
2
and x
3
, which means that the components k
2
and k
3
of the photon momentum are
two constants of motion. We will assume a one-dimensional photon propagation
by making these two constants equal to zero: k
2
= k
3
= 0. We can then write the
reduced one-dimensional form of the covariant Hamiltonian as
h(x
0
, x
1
, k
0
, k
1
) =
k
2
1
−k
2
0

p
(x
0
, x
1
)
c
2
÷
1
2
ω
p
(x
0
, x
1
). (3.74)
The corresponding equations of motion in the relativistic space–time are
dx
0

=
∂h
∂k
0
= −
k
0
c
2
ω
p
(x
0
, x
1
)
(3.75)
dx
1

=
∂h
∂k
1
=
k
1
c
2
ω
p
(x
0
, x
1
)
(3.76)
and
dk
0

= −
∂h
∂x
0
= −
1
2
¸
1 −
k
2
1
−k
2
0
ω
2
p
(x
0
, x
1
)
c
2

∂ω
p
∂x
0
(3.77)
dk
1

= −
∂h
∂x
1
= −
1
2
¸
1 −
k
2
1
−k
2
0
ω
2
p
(x
0
, x
1
)
c
2

∂ω
p
∂x
1
(3.78)
where τ is the photon proper time.
In order to write these equations in a more convenient form, it is useful to
introduce new adimensional variables, such that
x = K
1
x
1
= k
/
x
1
, y = k
/
x
0
(3.79)
and
v =
k
1
c
ω
p0
, u =
k
0
c
ω
p0
. (3.80)
50 Photon dynamics
We can see from the definition of the variable y that K
0
x
0
= β
/
y, where
β
/
= ω
/
/k
/
c is the normalized phase velocity of the wakefield. The normalized
proper time z, and Hamiltonian h
0
, are defined by
dz = k
/
c dτ, h
0
=
h
ω
p0
. (3.81)
We can state the new equations of motion as
dx
dz
=
∂h
0
∂v
,
dy
dz
=
∂h
0
∂u
(3.82)
and
dv
dz
= −
∂h
0
∂x
,
du
dz
= −
∂h
0
∂y
. (3.83)
The explicit form of the normalized Hamiltonian is
h
0
(x, v: y, u) =
v
2
−u
2
÷1 ÷ cos(x −β
/
y)
2

1 ÷ cos(x −β
/
y)
. (3.84)
By definition, we always have h
0
= 0. Noting that the denominator in this
equation cannot become imaginary because the wakefield modulation is always
less than one, < 1, we can write
u
2
= v
2
÷1 ÷ cos(x −β
/
y). (3.85)
This is nothing but the photon dispersion equation, written in adimensional
form. Noticing that h
0
only depends on x and y through (x −β
/
y), we conclude
that
∂h
0
∂x
= −
1
β
/
∂h
0
∂y
. (3.86)
Replacing this in the canonical equations (3.83), we obtain
dv
dz
= −
1
β
/
du
dz
. (3.87)
This expression means that, apart from the Hamiltonian H
0
, it is possible to
define another constant of motion I , such that
I = u ÷β
/
v = const. (3.88)
The existence of these two independent constants of motion, h
0
and I , proves
that the motion of a photon in a sinusoidal wakefield is integrable, as expected.
Notice that this equation is equivalent to stating that ω −kv
/
is a constant, where
v
/
= ω
/
/k
/
is the phase velocity of the wakefield. This invariant was already iden-
tified in the non-covariant formulation of photon acceleration. It will appear again
in the full wave description, where it corresponds to the wave phase invariance.
Stochastic photon acceleration 51
The position of the nonlinear resonance contained in the Hamiltonian (3.84)
is determined by the stationary condition (in the proper time z), or
d
dz
(x −β
/
y) = 0. (3.89)
Noting that, from the canonical equations (3.82), we have
dx
dz
=
v

p
,
dy
dz
= −
u

p
(3.90)
where
p
=

1 ÷ cos(x −β
/
y) is a positive quantity, we can write this sta-
tionary condition as
v ÷β
/
u = 0. (3.91)
In non-normalized variables this is equivalent to kc
2
/ω = ω
/
/k
/
. This means
that the nonlinear resonance corresponds to the equality between the photon group
velocity v
g
≡ kc
2
/ω and the phase velocity of the wakefield perturbation ω
/
/k
/
.
Replacing this resonance condition in equation (3.85), we obtain
u
2
(1 −β
/
) = 1 ÷ cos(x −β
/
y). (3.92)
The centre of this resonance corresponds to (x − β
/
y) = π/2, which leads
to the following coordinates:
u
0
= −
1

1 −β
/2
≡ −γ
/
, v
0
=
β
/

1 −β
/2
= −β
/
u
0
. (3.93)
The value of the invariant I , defined by equation (3.88), corresponding to
this particular photon trajectory is equal to I
0
= −

1 −β
/
. We can also see,
from equation (3.92), that the maximum excursion of u
2
associated with trapped
trajectories inside the nonlinear resonance is determined by δu
2
(1 −β
/2
) = 2.
The resonance half-width is then given by
δu
0
=

2
1

1 −β
/
. (3.94)
This completes the characterization of the trapped photon trajectories in the
four-dimensional relativistic space–time.
3.4 Stochastic photon acceleration
In his famous experiment, Isaac Newton inaugurated spectroscopy by decompos-
ing white light into its spectral components with the help of a prism. In very recent
days it was discovered that the opposite process can also occur and, starting from
a nearly monochromatic spectral light source, we can regenerate white light. For
that purpose, stochastic acceleration of photons can be used.
52 Photon dynamics
3.4.1 Motion in two wakefields
Let us first discuss the simple case of photon motion in the presence of two
different wakefields (or, in other words, two relativistic electron plasma waves)
with amplitudes
1
and
2
, distinct wavevectors
¯
k
1
and
¯
k
2
and distinct phase
velocities ¯ u
1
= (ω
/
/[k
1
[
2
)
¯
k
1
and ¯ u
2
= (ω
/
/[k
2
[
2
)
¯
k
2
.
The photon dispersion relation can now be written as
ω =
¸
k
2
c
2
÷ω
2
p0
¸
1 ÷
¸
i =1,2

i
cos(
¯
k
i
· ¯ r −ω
i
t )
¸
1/2
. (3.95)
Using the canonical variables ¯ q = ¯ r − ¯ u
1
t and ¯ p =
¯
k, we can write
the photon equations of motion in the form of equations (3.1), with the new
Hamiltonian (which is now time dependent)
= ω − ¯ u
1
· ¯ p =
¸
p
2
c
2
÷ω
2
p0
[1 ÷
1
cos(
¯
k
1
· ¯ q)
÷
2
cos(
¯
k
2
· (¯ q − ¯ vt ))]
¸
1/2
− ¯ u
1
· ¯ p (3.96)
where
¯ v = ¯ u
1
− ¯ u
2
. ω
p0
[(
¯
k
1
/k
2
1
) −(
¯
k
2
/k
2
2
)]. (3.97)
In order to simplify the discussion, we will assume that the wakefields prop-
agate in the same direction, and we will concentrate on the one-dimensional
case. The extension to three dimensions is straightforward and does not lead
to qualitatively new effects.
We can see from the above Hamiltonian (or from its one-dimensional ver-
sion) that two nonlinear resonances exist in the (q, p) phase plane. They are
defined by the elliptic fixed points
q
i
=
π
k
p
, p
i
= s
i
γ
i
β
i
ω
p0
c

1 −
i
(3.98)
for i = 1, 2. The following definitions were used here: s
i
= sign of u
i
, β
i
=
[u
i
[/c and γ
i
= (1 −β
2
i
)
−1/2
.
It is also quite well known that the second of these two resonances (the
one corresponding to i = 2) can only be seen in the phase plane if we use a
stroboscopic plot of the photon motion, at the instants t
n
= (2nπ/k
2
v), for n
integer. Such a discrete representation of the motion is usually called a Poincar´ e
map.
In accordance with equation (3.65) we can also say that the the width of these
two resonances is determined by
(p)
i
= 2γ
i
ω
p0
c

2
i
. (3.99)
Stochastic photon acceleration 53
It is well known from the theory of dynamical systems [59, 124] that the
interaction between the two resonances leads to the destruction of the separatrix
curves. These curves will both break up into a thin region of stochastic motion
and the width of such a region will exponentially grow with the amplitude of the
resonances
i
. With increasing values of
i
, the two stochastic regions will even-
tually merge, and a large fraction of the photon phase space (q, p) will be filled
with stochastic trajectories, leading to what is called large-scale stochasticity.
Occurrence of large-scale stochastic acceleration is dictated by a qualitative
criterion, called the Chirikov criterion or the resonance overlapping criterion. This
is confirmed by numerical calculations within an error of a fewper cent, for a large
variety of similar nonlinear Hamiltonian motions. The criterion states that large-
scale stochasticity occurs when the sum of the resonance half-widths becomes
larger than the distance between these resonances.
In our case it can be written as
(p)
1
÷(p)
2
≥ 2[ p
2
− p
1
[. (3.100)
Using equations (3.98, 3.99) we can write the overlapping criterion as
γ
1

2
1
÷γ
2

2
2
[s
2
β
2
γ
2

1 −
2
−s
1
β
1
γ
1

1 −
1
[
≥ 1. (3.101)
It is important to notice that this criterion is independent of the mean electron
plasma frequency ω
p0
, and that it only depends on the velocities and amplitudes of
the two wakefields. If we assume that the two amplitudes are equal,
1
=
2
= ,
we can simplify the overlapping criterion and write

2
1 −
1 ÷ν
[s
2
β
2
−s
1
β
1
ν[
≥ 1 (3.102)
where ν = γ
1

2
.
In the weakly relativistic limit, where we have γ
i
. 1 and ν . 1, we
conclude from this expression that large-scale stochasticity occurs for

1
2
3
(s
2
β
2
−s
1
β
1
)
2
. (3.103)
Noting that β
i
are always smaller than one, we conclude that this criterion
is compatible with low values of the wakefield amplitude < 1. In the opposite
limit of strongly relativistic phase velocities of the plasma wakefields, where γ
i
·
1 and β
i
. 1, we obtain from equation (3.102)

1
2
3
(1 −ν)
2
(3.104)
where we have assumed that the wakefields propagate in the same direction, s
1
=
s
2
, and that their relativistic γ factors are similar, ν . 1.
54 Photon dynamics
We see that, again, the criterion is compatible with low wakefield amplitudes
< 1. Even in the most unfavourable situation of very different relativistic
factors, ν · 1, we get from equation (3.103) ≥ 1/3. We can then say
that, in very broad terms, both in the weakly and the strongly relativistic limits,
transition from regular to stochastic trajectories can be expected with moderately
low wakefield amplitudes.
However, the strongly relativistic case is physically more interesting, be-
cause the width of the region between the two resonances where stochastic ac-
celeration can occur is proportional to the relativistic gamma factors, and a larger
frequency spread of a bunch of photons with nearly identical initial frequencies
can be obtained. From initial monochromatic radiation we can then obtain broad-
band radiation (white light) [66].
A similar situation occurs if, instead of having two different wakefields, we
can excite a single wakefield in a plasma, but with a modulated amplitude. Such
a modulation can be due, for instance, to the existence of ion acoustic waves
propagating in the background plasma. Here, instead of two nonlinear resonances,
we have three nearby resonances to which we can apply the same overlapping
criterion. This was studied, using the covariant formulation, in reference [65].
3.4.2 Photon discrete mapping
It is also interesting to consider the interaction of a photon moving in a plasma
with an electrostatic wavepacket containing a large spectrum of electron plasma
waves. Here we will follow a procedure similar to that used by Zaslavski et
al [125] for studying charged particles.
The photon dispersion relation (3.95) is now replaced by
ω =
¸
k
2
c
2
÷ω
2
p0
¸
1 ÷

¸
n=−∞

n
cos(
¯
k
n
· ¯ r −ω
n
t )
¸
1/2
. (3.105)
The corresponding one-dimensional photon equations of motion are
dx
dt
=
kc
2
ω
,
dk
dt
=
ω
2
p0


¸
n=−∞

n
k
n
sin(k
n
x −ω
n
t ). (3.106)
As a simple and reasonable model for the wavepacket, we can use

n
k
n
= a, ω
n
= ω
0
. ω
p0
, k
n
= k
0
÷nk (3.107)
where a is a constant and k < k
0
is the characteristic distance between two
consecutive spectral components.
This model is convenient from the physical point of view because, in con-
trast with similar models used in nonlinear dynamics where the amplitudes are
assumed constant, it corresponds to a spectrum with variable amplitudes, where
Stochastic photon acceleration 55
the lower amplitudes correspond to the higher wavenumbers. This is compatible
with the idea that electron Landau damping of the wakefield spectrum prevents
the higher wavenumbers.
The sum in equation (3.106) can then be replaced by
a sin θ

¸
n=−∞
cos(nkx) (3.108)
where θ = k
0
x −ω
0
t is the phase of the central spectral component. We can also
use the identity

¸
n=−∞
cos(nkx) =

k

¸
n=−∞
δ(x − x
n
) (3.109)
with x
n
= (2nπ/k) ≡ nL. This means that, from equation (3.106), we can
write
dk
dt
=
ω
2
p0

aL sin θ

¸
n=−∞
δ(x − x
n
). (3.110)
This shows that the photon moves with a constant wavenumber (or a constant
velocity), except when it crosses the point x = x
n
, where it suffers a sudden kick
and abruptly changes its wavenumber. This new wavenumber remains constant
until it crosses the next point x = x
n÷1
, and so on.
This picture of the photon motion means that we can transform the variable
position x inside the delta function argument into a time variable, just by using
the relation x = [v
g
[t = ([k[c
2
/ω)t . The above equation is then replaced by
dk
dt
=
ω
2
p0
2[k[c
2
aL sin θ

¸
n=−∞
δ(t −t
n
). (3.111)
The instants t
n
are defined by x(t
n
) = x
n
, apart from a minor detail resulting
from the fact that the photon can eventually be reflected at some point x
n
, and in
that case the instant t
n÷1
will be defined by x(t
n÷1
) = x
n−1
. Of course, such a
reversal in the direction of photon propagation is not likely to occur for energetic
photons, with initial frequencies much larger than the electron plasma frequency.
Let us now introduce an adimensional variable w, such that
w =
kc
2
ω
2
p0
[k[ =
k
2
c
2
ω
2
p0
s
k
(3.112)
where s
k
is the sign of k. This is equivalent to writing
k =
ω
p0
c
[w[
1/2
s
w
(3.113)
56 Photon dynamics
where s
w
is the sign of w. According to equation (3.111) the new variable w will
evolve in time as
dw
dt
= aL sin θ

¸
n=−∞
δ(t −t
n
). (3.114)
This has to be completed with an equation describing the time evolution for
the phase variable θ. By definition, we have dθ/dt = (k
0
c
2
/ω)k −ω
0
. But, at this
point, we should notice that it is physically quite natural to assume that
n
<1.
According to equation (3.105), this allows us to replace ω by ¨ ω and to write,
using the new variable w

dt
= k
0
c
ω
p0
¨ ω
s
w
−ω
0
. (3.115)
This equation, with ¨ ω in the denominator, is also physically very convenient
because it guarantees that θ varies continuously across the points t = t
n
. We
now have two coupled and closed equations for the variables w and θ, equa-
tions (3.114, 3.115), and we can build up a map on the new phase plane (w, θ).
For that purpose, we define
w
n
= w(t

n
), θ
n
= θ(t

n
) (3.116)
where t
±
n
= t
n
±δ, with δ → 0, represent the instants imediately before (−) and
after (÷) the critical instants t
n
.
Equations (3.114, 3.115) show that, at t = t
n
, the variable w suffers a sudden
jump, and that the phase θ remains constant. This can be stated as
w(t
÷
n
) −w(t

n
) = aL sin θ, θ(t
÷
n
) −θ(t

n
) = 0. (3.117)
During the interval (t
÷
n
, t

n÷1
) the variable w, and the corresponding photon
momentum p, remain unchanged. This is equivalent to stating that w(t

n÷1
) =
w(t
0
n
), or
w
n÷1
= w
n
÷aL sin θ
n
. (3.118)
We can also see from equation (3.115) that the time derivative of the phase
θ also stays constant over the same time interval. We can then write
θ(t

n÷1
) = θ(t
÷
n
) ÷
d
dt
θ(t
÷
n
)t
n
. (3.119)
Here, t
n
is the interval between two consecutive kicks
t
n
= L
¨ ω
n÷1
k
n÷1
c
2
. (3.120)
Using equation (3.115, 3.119), we can then obtain the following result:
θ
n÷1
= θ
n

ω
0
L
c

1 ÷[w
n÷1
[

[w
n÷1
[
÷k
0
Ls
n÷1
(3.121)
Photon Fermi acceleration 57
where s
n÷1
is the sign of w
n÷1
.
We see that equations (3.118, 3.121) define a discrete mapping on the phase
plane (w, θ), which can be written as
w
n÷1
= w
n
÷ A sin θ
n
(3.122)
θ
n÷1
= θ
n
− B

1 ÷[w
n÷1
[
−1
÷
˜
θs
n÷1
. (3.123)
This mapping depends on two parameters, A and B, determined by
A = aL, B =
ω
0
c
L. (3.124)
The map (3.123) also shows a constant phase shift determined by
˜
θ = k
0
L.
For small values of the adimensional variable w, which corresponds to small val-
ues of the photon wavenumber, this mapping reduces to the
ˆ
L-mapping introduced
in [125], which shows an intermittency behaviour for A ·1.
This means that the photons can suffer intermittency acceleration when they
interact with an electrostatic wavepacket in the limit of small photon wavenum-
bers (when they are close to the cut-off conditions).
3.5 Photon Fermi acceleration
We shall now discuss a different mechanism for photon acceleration which can
occur inside an electromagnetic cavity with moving boundaries. This can be seen
as the photon version of the well-known mechanism for cosmic ray acceleration
first proposed by Fermi [29], where charged particles can gain energy by bouncing
back and forth between two magnetic clouds.
This mechanism became extremely successful and it is dominantly used in
the current models for cosmic ray acceleration in shocks. In its simplest and
most popular versions, the charged particle can be described by two-dimensional
discrete maps [59].
We will show in this section that a similar mechanism can be applied to pho-
tons, and that the photon motion can also be described by a discrete mapping [30].
In the case of photons, the magnetic clouds of the original model are replaced
by mirrors or by plasma walls with a sharp density gradient. We can assume that
one of the walls is fixed at x = 0 and the other oscillates around x = L
0
with a
frequency ω
/
.
This can be described by a plasma frequency space–time variation of the
form
ω
2
p
(x, t ) = ω
2
p0
¸
f (x − L(t )) x > L(t )
0 L(t ) ≥ x > 0
f (−x) x < 0
. (3.125)
The function f (x) describes the plasma density profile of the two plasma
walls (assuming that they have identical profiles), to be specified later. The
58 Photon dynamics
x
0
L (t)
photon
fixed
mirror
moving
mirror
Figure 3.8. Fermi acceleration of photons inside an oscillating cavity.
moving wall is supposed to oscillate according to
L(t ) = L
0
(1 ÷ cos ω
/
t ) (3.126)
where the amplitude of this oscillation is assumed to be very small: <1.
The plasma walls are also supposed to be sufficiently dense in order to act
as mirrors and to reflect all the incoming photons, which then remain trapped
in a kind of one-dimensional cavity. Each time the photons are reflected by the
moving plasma wall, they suffer a double Doppler shift, and their initial frequency
ω
i
is transformed into a final frequency ω
f
, defined by the well-known law
ω
f
= ω
i
1 ÷β
1 −β
. (3.127)
The velocity of this moving wall is just the time derivative of L(t ) and we
can write
β = −
L
0
ω
/
c
sin ω
/
t. (3.128)
If we define two adimensional quantities
b =
L
0
ω
/
c
, θ = ω
/
t (3.129)
we obtain the following law for the frequency shift after successive reflections at
the moving plasma wall:
ω
n÷1
= ω
n
1 −b sin θ
n
1 ÷b sin θ
n
. (3.130)
Photon Fermi acceleration 59
Now we have to find a corresponding transformation law for the phase θ
between two successive reflections. Clearly, we can write θ
n÷1
= θ
n
÷ ω
/
t ,
where t is the time spent by the photon between two successive reflections at
the moving wall. This time interval can be divided into two distinct parts: t =
t
0
÷ t
p
. The first part is the time spent by the photon in the vaccum region
between the two plasma walls.
For <1, we can use, as a good approximation
t
0
= 2
L
0
c
. (3.131)
In order to calculate the time spent by the photon inside the two plasma
regions, t
p
, we notice that the photon velocity is determined by:
dx
dt
= c

1 −(ω
p0
/ω)
2
f (x, t ). (3.132)
If the time spent inside the moving plasma region is much shorter than
the period of the wall oscillations, t
p
< (4π/ω
/
), we can neglect the plasma
motion during the process of photon reflection, and replace f (x, t ) by f (x) in
this equation. This leads to
t
p
=
4
c

x
c
0
dx

1 −(ω
p0
/ω)
2
f (x)
. (3.133)
Here, the factor of 4 was introduced in order to account for the four distinct
paths inside the two plasma walls. This integral extends from the plasma bound-
ary to the cut-off position x
c
, where the photon frequency equals the electron
plasma frequency ω = ω
p
(x
c
). Let us also introduce a normalized frequency
u =
ω
ω
p0
. (3.134)
Using equations (3.130, 3.133), we can establish the following discrete map
on the phase plane (u, θ):
u
n÷1
= u
n
F(θ
n
) (3.135)
θ
n÷1
= θ
n
÷ G(u
n÷1
). (3.136)
The function F(θ
n
) is solely dependent on the double Doppler shift at the
moving wall, and the function G(u
n÷1
) is determined by the plasma density
profile
F(θ) =
1 −b sin θ
1 ÷b sin θ
(3.137)
G(u) = 2b

1 ÷
2
L
0

x
c
0
dx

1 − f (x)/u
2

. (3.138)
60 Photon dynamics
Let us take the particularly interesting example of a parabolic density profile
f (x) =
α
2
π
2
L
2
0
x
2
(3.139)
where the parameter α defines the plasma density slope.
Then equation (3.138) reduces to
G(u) = 2b

1 ÷
u
α

. (3.140)
The map (3.135, 3.136) also takes a simpler form for very small plasma wall
oscillations, such that b <1. In this case, we have
F(θ) = 1 ÷2b sin θ. (3.141)
We should note that all these maps are not area preserving. This can easily be
seen by considering the Jacobian of the transformation (u
n
, θ
n
) →(u
n÷1
, θ
n÷1
):
[ J[ =

∂u
n÷1
/∂u
n
∂u
n÷1
/∂θ
n
∂θ
n÷1
/∂u
n
∂θ
n÷1
/∂θ
n

= [F(θ)[ ,= 1. (3.142)
This is precisely the factor by which the photon frequency is double Doppler
shifted by the moving plasma wall. However, because F(θ) is a periodic function
of θ, we see that sets consisting of thin layers in phase space, of infinitesimal
width δu and extending in phase from θ = 0 to θ = 2π, are area preserving.
It is also quite useful to determine the fixed points of the map, and their sta-
bility. This will give us important information concerning the qualitative aspects
of photon motion. The first-order fixed points are determined by
u
n÷1
= u
n
F(θ
n
) = u
n
(3.143)
θ
n÷1
= θ
n
÷ G(u
n÷1
) = θ
n
÷2mπ (3.144)
for m integer. This is equivalent to
F(θ) = 1, G(u) = 2mπ. (3.145)
For the parabolic density profile (3.139), this leads to
θ
m
= 0, u
m
= α


b
−1

. (3.146)
The stability of these fixed points can be determined by locally linearizing
the map around each of them. Let us first linearize on the variable u, by defining
u = u
m
÷ ¨ u. (3.147)
Photon Fermi acceleration 61
Replacing it in equations (3.135, 3.136, 3.141), we obtain
¨ u
n÷1
= ¨ u
n
÷2u
m
sin θ
n
(3.148)
θ
n÷1
= θ
n
÷
2b
α
¨ u
n÷1
÷2mπ. (3.149)
Introducing a new variable I and a new parameter K, such that
I =
2b
α
¨ u, K =
4b
2
α
u
m
, (3.150)
this reduces to the well-known standard map, first studied by Chirikov [17, 59]:
I
n÷1
= I
n
÷ K sin θ
n
(3.151)
θ
n÷1
= θ
n
÷ I
n÷1
. (3.152)
This shows that the photon dynamics around a given fixed point (θ
m
, u
m
)
is approximately described by the standard map. It is well known that such a
map suffers a topological transition into large-scale stochasticity if K > 1. This
means that, near the fixed points, the photon motion will become stochastic in a
significant fraction of the available phase space, if
u
m
>
α
4b
2

. (3.153)
We now examine the stability of the fixed points (3.145) of the Fermi map-
ping for the parabolic density profile. Following the usual procedure [59], we
linearize the map on both variables, and obtain
¯ x
n÷1
= A · ¯ x
n
(3.154)
where ¯ x = ( ¨ u,
¨
θ) = (u − u
m
, θ − θ
m
), and the matrix transformation A is the
linearized Jacobian matrix
A =
¸
1 −2bu
m
2b
α
1 −4
b
2
α
u
m
¸
. (3.155)
It can easily be seen that this linear map is area preserving:
[A[ = 1. (3.156)
Stability of the fixed points (3.145) requires that [Tr A[ < 2, or, more explic-
itly, that
u
m
<
α
b
2
. (3.157)
We can then say that, for high enough integers m such that this inequality is
not verified, the first-order fixed points are all unstable. This means that, for such
values of the photon normalized frequency u, the photon motion is essentially
62 Photon dynamics
Figure 3.9. Phase space of the Fermi photon map. Regular and stochastic trajectories are
shown.
stochastic. However, for much lower frequencies, significant stochastic motion
had already taken place, according to the much less stringent threshold criterion
(3.153). This is well illustrated by numerical calculations.
From here we conclude that a broad spectrum of radiation (which can be
called white light) can be generated from nearly monochromatic light trapped
Magnetoplasmas and other optical media 63
inside an oscillating plasma cavity.
It is interesting to compare this qualitative aspect of the photon Fermi map
with the usual Fermi mappings for charged particles [59] where, in contrast with
the threshold criterion given by equation (3.157), an upper limit exists for particle
energies, above which they cannot be accelerated. In some sense the photon Fermi
map is similar to the inverted Fermi map for charged particles, where the energy
axis is turned upside down.
This model of Fermi acceleration of photons could, in principle, be experi-
mentally tested using an oscillating optical cavity, if the cavity parameters were
conveniently chosen and if the photons could bounce back and forth several times
inside the cavity, before escaping out, or before being absorbed by the walls.
It can also be useful in the context of astrophysics. For instance, the spectrum
emitted by some very high-redshift radio galaxies reveals a strong asymmetry in
the Lyman α line profile, indicating a clear blueshift [115]. This can be interpreted
as a result of Fermi acceleration of the Lyman α photons by a moving shock [11].
Finally, we should notice that the concept of photon acceleration in a plasma
can be extended to the acceleration of plasmons, or quanta of electron plasma
waves, when they are trapped inside an unstable cavity. In this case [24], the plas-
mon Fermi acceleration will mainly lead to a change in the plasmon wavenumber
because the plasmon frequency is always nearly equal to the electron plasma
frequency.
3.6 Magnetoplasmas and other optical media
In this chapter we have only focused on processes occurring in isotropic plasmas
because most of the published work on photon acceleration concerns this medium.
However, our theoretical approach remains valid for other optical media.
Let us then conclude the chapter with some brief comments on the photon
processes associated with moving perturbations of the refractive index in magne-
tized plasmas and in non-ionized optical media, such as a neutral gas, a glass or
an optical fibre.
The photon dispersion relation in a magnetized plasma depends, not only on
the electron plasma frequency ω
p
, but also on the value and direction of the static
magnetic field
¯
B
0
. We can represent it generally as
R(ω,
¯
k: ω
p
(¯ r, t ), ω
c
(¯ r, t )) = 0 (3.158)
where ω
c
= e[B
0
[/m is the electron cyclotron frequency.
This means that we can expect to obtain photon acceleration, not only by
using moving electron density perturbations, such as ionization fronts and wake-
fields, but also if we excite similar forms of moving magnetic field perturbations.
They can be produced by non-stationary currents applied to external coils, or by
propagating low-frequency electromagnetic waves, such as Alfven waves.
64 Photon dynamics
The explicit expression of this dispersion relation is quite complicated, even
in the limit of high-frequency waves for which the ion dispersion effects can be
neglected, but we can still consider some simple examples.
First, if we assume photon propagation in a direction perpendicular to the
static magnetic field
¯
B
0
, there are two distinct polarization states. One corre-
sponds to the ordinary mode, with photons linearly polarized along
¯
B
0
, for which
the dispersion relation is identical to that considered before for a non-magnetized
plasma. The other corresponds to the extraordinary mode, with photons ellipti-
cally polarized in the plane perpendicular to
¯
B
0
. The dispersion relation for this
new mode is
ω
2
= k
2
c
2
÷ω
2
p
ω
2
−ω
2
p
ω
2
−ω
2
uh
(3.159)
where ω
uh
=

ω
2
p
÷ω
2
c
is called the upper-hybrid frequency (the lower-hybrid
one would only appear for low-frequency waves where the influence of the ion
motion has to be retained).
We see that, for a photon frequency close to this resonance frequency, a small
time change in the static magnetic field, or equivalently in the electron cyclotron
frequency ω
c
, will significantly alter the value of the refractive index and will lead
to a frequency shift. In order to calculate this effect with the aid of the canonical
equations for the photons we have to make use of the Hamiltonian function, valid
for the extraordinary mode.
From the above dispersion relation, we can easily get
ω(¯ r,
¯
k, t ) =
¸
1
2
(k
2
c
2
÷ω
2
p
÷ω
2
uh
)
÷
1
2

(k
2
c
2
÷ω
2
p
÷ω
2
uh
)
2
−4(k
2
c
2
ω
2
uh
÷ω
4
p
)
¸
1/2
(3.160)
where both the electron plasma frequency and the upper-hybrid frequency are
functions of ¯ r and t .
Similarly, for propagation parallel to the static magnetic field, we have two
photon polarization states, with the corresponding dispersion relation
ω
2
= k
2
c
2
÷ω
2
p
ω
ω ±ω
c
. (3.161)
The plus sign corresponds to the L-mode, which is left circularly polarized,
and the minus sign corresponds to the R-mode, which is right circularly polarized.
Because, by definition, we have ω
c
> 0, we can see that the L-mode has no
resonances, while the R-mode is resonant for ω = ω
c
. This is the so-called
cyclotron resonance. This means that a small space–time change in the cyclotron
frequency will lead to a significant change in the refractive index for the R-mode.
However, in dealing with photon motion very close to resonances, we have to
take into account the wave absorption mechanisms, which will eventually reduce
Magnetoplasmas and other optical media 65
the efficiency of the expected frequency shift or photon acceleration processes.
These absorption mechanisms are well known in plasma physics and will not be
discussed here.
Let us now turn to non-ionized optical media. The typical form for the linear
dispersion relation in these media is
ω
2
= k
2
c
2
÷
ω
2
ω
2
f
ω
2
−ω
2
0
(3.162)
where ω
0
is the frequency of the nearest resonant transition between two quantum
levels and ω
f
plays the role of the plasma frequency. It depends on the density of
atoms or molecules in the medium and on the value of the transition probabilities
between these two quantum levels.
For ω · ω
0
this dispersion relation reduces to that of the non-magnetized
plasma. However, for visible light propagating in the usual optical media, we
have ω <ω
0
, which explains why the refractive index is usually greater than one
(in contrast with the plasma case where it is less than one).
Two ways can be foreseen to produce a space–time change in the refractive
index and a subsequent photon acceleration in such media. The first one is to
change the transition frequency ω
0
. This can, for instance, be done with the aid
of an externally applied electric field. The resulting Stark effect will lead to a
detuning (quite often a splitting) of the atomic transition energy levels. We can
then imagine a physical configuration where the optical medium is located inside
an elongated capacitor.
By applying a sudden voltage signal at one extremity of this capacitor, the
voltage signal will propagate along the capacitor arms and produce a moving
transition between two different values of the refractive index of the optical
medium [28]. The closer we are to the resonance condition, the stronger will be
the change in the refractive index.
A more drastic change in the refractive index can be obtained by using
electromagnetic induced transparency (EIT) [38]. This corresponds to exactly
equating to zero the value of ω
f
, by inhibiting transitions between the two quan-
tum states with the help of an auxiliary light source. The photons of this auxiliary
source couple one of these two energy states with a third one. In this way,
the refractive index of a nearly resonant medium (which is opaque, because it
completely absorbs the resonant photons) can be reduced exactly to the vacuum
value, corresponding to a complete transparency.
It should be noticed that, in recent experiments with EIT produced by auxil-
iary short laser pulses, a large spectral broadening is observed [39], which can be
interpreted as the result of photon acceleration.
These ways of changing the refractive index of a medium are essentially lin-
ear, but we can also imagine similar effects resulting from nonlinear mechanisms.
For instance, it is known that a strong laser beam can produce a change in the
66 Photon dynamics
refractive index, according to
n = n
0
÷n
2
I (¯ r, t ) (3.163)
where n
0
is the linear refractive index, n
2
is the nonlinear refractive index (pro-
portional to the nonlinear susceptibility of the medium) and I (¯ r, t ) is the intensity
profile of the strong laser beam.
If photons having a different frequency and belonging to a probe beam co-
propagate with the first one and cross the strong-beam boundary (because their
group velocities are different), they will suffer a frequency shift described by
the above canonical equations and similar to the one observed for the ionization
front. The difference here is that the frequency shift will be negative for a photon
crossing the front of the strong beam, and positive for a photon crossing the rear of
the beam, because n
2
is positive, in contrast with the ionization front case which
led to a decrease in the refractive index.
This effect is well understood in the frame of our single photon dynamical
approach, but it is currently known as induced phase modulation. It is quite
clear that the field phase is not an essential aspect of the problem, because the
phase is completely absent from our photon canonical equations, and we still get
a frequency shift.
The problem of the influence of the field phase in the so-called phase modu-
lation effects is very interesting and will be discussed in detail later. Here we only
would like to stress that the photon dynamical theory discussed in this chapter can
equally well treat the cases of co-propagating and counter-propagating photons,
in contrast with the usual theory of induced phase modulation which is only used
for the co-propagation case [3].
Chapter 4
Photon kinetic theory
Until now, we have considered single particle trajectories, which (as we have
shown) can account for several new and interesting features of electromagnetic
wavepackets travelling in a non-stationary medium. However, this approach is
not capable of describing the change in the internal structure of the wavepackets
themselves, because such a structure is simply forgotten.
The easiest way to obtain a more detailed description of the wavepackets is
to describe them, not as a single photon, but as an ensemble of photons with a
given spectral and space–time distribution. The electromagnetic wave spectrum
will then be seen as a gas of photons evolving in an optical medium. The kinetic
equation describing the space–time evolution of such a gas is derived here.
First, we will use simple and intuitive arguments which are similar to those
leading to the Klimontovich equation for charged particles in a gas. A second and
more elaborated method will also be described, where the analogue of the Wigner
function for the electromagnetic field is introduced.
These kinetic equations sometimes contain too much information for several
problems, for which a simpler description of the photon distributions is required.
For that reason it is also interesting to derive photon conservation equations which
allow us to describe the electromagnetic field as a fluid of identical particles, and
to consider only averaged photon properties.
This chapter will be completed by a detailed discussion of the phase-space
representation of a short electromagnetic pulse. Examples of time evolution of
short pulses with chirp and frequency shift are given. In particular, we present a
model for the self-induced blueshift, which is produced when a strong laser pulse
propagates along a neutral gas and produces an ionization front.
We conclude the chapter by considering the well-known effect of self-phase
modulation and by showing that it can be described as a particular example of
photon acceleration.
67
68 Photon kinetic theory
4.1 Klimontovich equation for photons
Let us first consider a single photon trajectory, as defined by the ray equations
in their canonical form. The dynamical state of the photon (position and local
frequency) can be determined by its location in the six-dimensional phase space
(¯ r,
¯
k).
We can also associate with this single photon a microscopic density distribu-
tion defined by the product of two Dirac δ functions
N
1
(¯ r,
¯
k, t ) = δ[¯ r − ¯ r(t )]δ[
¯
k −
¯
k(t )] (4.1)
where ¯ r(t ) and
¯
k(t ) represent the actual photon trajectory, as determined by the
solutions of the ray equations.
Of course, in order to accurately describe the evolution of a given elec-
tromagnetic pulse propagating in a medium, it is more convenient to consider,
instead of a single photon trajectory (which gives no information on the detailed
structure of the pulse), a large number of nearby photon trajectories. Let us then
assume n different photon trajectories, and the associated density distribution
N(¯ r,
¯
k, t ) =
n
¸
j =1
δ[¯ r − ¯ r
j
(t )]δ[
¯
k −
¯
k
j
(t )]. (4.2)
Integration in phase space clearly gives the total number of photons (or ray
trajectories)
n =

d¯ r

d
¯
kN(¯ r,
¯
k, t ). (4.3)
If we want to establish an equation for the time evolution of the photon
density distribution, we can take the partial time derivative of equation (4.2). We
have

∂t
N(¯ r,
¯
k, t ) =
n
¸
j =1

d¯ r
j
dt
·

∂¯ r
j
÷
d
¯
k
j
dt
·


¯
k
j

δ[¯ r − ¯ r
j
(t )]δ[
¯
k −
¯
k
j
(t )]. (4.4)
Now, we can use the obvious relation ∂/∂x f (x − y) = −∂/∂y f (x − y), and
transform this expression into

∂t
N(¯ r,
¯
k, t ) =
n
¸
j =1

d¯ r
j
dt
·

∂¯ r
÷
d
¯
k
j
dt
·


¯
k

δ[¯ r − ¯ r
j
(t )]δ[
¯
k −
¯
k
j
(t )]. (4.5)
Another simplification can be introduced by using the following property of
the Dirac δ function: xδ(x − y) = yδ(x − y). This allows us to replace the
remaining ¯ r
j
and
¯
k
j
by ¯ r and
¯
k inside the differential operator acting on the δ
functions.
The result is
d
dt
N(¯ r,
¯
k, t ) = 0 (4.6)
Klimontovich equation for photons 69
with
d
dt
=

∂t
÷
d¯ r
dt
·

∂¯ r
÷
d
¯
k
dt
·


¯
k
. (4.7)
Equation (4.6) can be called a kinetic equation for photons in its Klimon-
tovich form [49]. It simply states that the photon number density is conserved.
The problem with this simple result is that it was obtained by assuming that the
photons are point particles, which is obviously not the case, and its validity has
to be confirmed a posteriori by using Maxwell’s equations. We will see below
that this equation is only approximately valid, as should be expected, and that its
range of validity is nearly (but not exactly) coincident with the range of validity
of geometric optics.
From the canonical ray equations, we see that this kinetic equation can also
be written as
∂ N
k
∂t
÷[N
k
, ω] = 0 (4.8)
where N
k
≡ N
k
(¯ r, t ) ≡ N(¯ r,
¯
k, t ), and the Poisson bracket is
[N
k
, ω] =
∂ N
k
∂¯ r
·
∂ω

¯
k

∂ N
k

¯
k
·
∂ω
∂¯ r
. (4.9)
At this point we could follow the usual statistical procedure and introduce
some coarse-graining in the photon phase space, which would replace the quite
spiky quantity N
k
(¯ r, t ) by a smooth and well-behaved function like its ensemble
average !N
k
(¯ r, t )). This would lead us too far fromour present purpose. It is more
interesting here to establish a link between this quantity and the electromagnetic
energy density.
By definition, we can write the total energy as
W(t ) =

w(¯ r, t ) d¯ r (4.10)
where the energy density is
w(¯ r, t ) = 2

¨

k
N
k
(¯ r, t )
d
¯
k
(2π)
3
. (4.11)
This equation states that the energy of each photon is equal to
¨

k
, as we
know from quantum theory. The factor of 2 is introduced because of the existence
of two possible states of polarization. On the other hand, we can exactly establish,
from the classical theory of radiation [57], that
w(¯ r, t ) =

0
4

∂ωR
∂ω

k
[E
k
[
2
d
¯
k
(2π)
3
(4.12)
where [E
k
[
2
is the module square of the electric field amplitude of the Fourier
component
¯
k, and R ≡ R(ω,
¯
k) = 0 is the dispersion relation of the medium:
R(ω,
¯
k) = (ω,
¯
k) −
k
2
c
2
ω
2
÷[
¯
k · ˆ e[
2
c
2
ω
2
= 0. (4.13)
70 Photon kinetic theory
Here ˆ e is the unit polarization vector. By comparing these two expressions
for the electromagnetic energy density, we obtain
N
k
(¯ r, t ) =

0
8
¨
h

∂ R
∂ω

k
[E
k
[
2
. (4.14)
Obviously, this definition of the photon number density can only make sense
if we assume that the electric field amplitude of each Fourier component is a
slowly varying function of space and time. A more refined way of establishing
the definition of N
k
(¯ r, t ) is based on the concept of the Wigner functions for the
electromagnetic field, which will be considered next.
4.2 Wigner–Moyal equation for electromagnetic radiation
4.2.1 Non-dispersive medium
We will consider first a non-dispersive medium, in order to clearly state our
procedure. We will also assume that the medium is isotropic and with no losses.
In the absence of charge and current distributions, we have, from Maxwell’s
equations,

2
¯
E −∇(∇ ·
¯
E) −µ
0

2
∂t
2
¯
D = 0 (4.15)
where
¯
D =
0

¯
E is the displacement vector. We also have = 1 ÷ χ, where χ
is the susceptibility of the medium.
Assuming, for simplicity, that the fields are transverse (∇ ·
¯
E = 0), we can
write

2
¯
E −
1
c
2

2
¯
E
∂t
2
=
1
c
2

2
∂t
2

¯
E). (4.16)
For a wave with a given frequency ω and wavenumber
¯
k, we can define the
Wigner function for the electric field as
F(¯ r, t : ω,
¯
k) =

d¯ s


¯
E

¯ r ÷
¯ s
2
, t ÷
τ
2

·
¯
E

¯ r −
¯ s
2
, t −
τ
2

e
−i
¯
k·¯ s÷iωτ
.
(4.17)
This quantity is formally quite similar to the Wigner function for a quantum
system[40]. In contrast with our classical approach, the quantumWigner function
is well understood and of current use in quantum optics [58, 116].
Following a procedure explained in detail in appendix A, we can derive from
the above wave equation an equation describing the space–time evolution of the
Wigner function. The same procedure was used in reference [111] to study the
case of relativistic plasmas.
In our case of a non-dispersive medium, the evolution equation for
F(¯ r, t : ω,
¯
k) takes the form


∂t
÷
c
2
¯
k
ω
· ∇

F ÷


∂t

F = −ω( sin F) (4.18)
Wigner–Moyal equation for electromagnetic radiation 71
where is a differential operator, which acts both backwards on and forwards
on F. It can be defined by
=
1
2

¸

∂¯ r
·


¯
k


∂t

∂ω
¸

. (4.19)
The right and left arrows are here to remind us that, in each of the two terms,
the first differential operator acts backwards on and the second one acts forwards
on F. The sine differential operator in equation (4.18) is, in fact, an infinite series
of differential operators, according to
sin =

¸
l=0
(−1)
l
(2l ÷1)'

2l÷1
. (4.20)
At the cost of such an unusual operator, we were able to derive from
Maxwell’s equations a closed evolution equation for the Wigner function F of
the electric field. This is valid in quite general conditions, apart from our basic
assumptions that the medium should be non-dispersive and that the dielectric
constant should only evolve on a slow timescale. Its relation to the geometric
optics approximation will become apparent below.
Equation (4.18) is formally quite similar to the Wigner–Moyal equation for
quantum systems [40, 80], except for the term on the time derivative of the refrac-
tive index, which has no equivalent in the quantum mechanical problem. For that
reason it can be called the Wigner–Moyal equation for the electromagnetic field.
Clearly, it is significantly more complex than the kinetic equation established at
the begining of this chapter.
In order to compare these two approaches, it is useful to introduce a few
simplifying assumptions. The first one is associated with the character of the
electromagnetic spectrum. We can assume that this spectrum is just a superposi-
tion of linear waves. For each spectral component, the value of the frequency ω
has to satisfy the linear dispersion relation of the medium
ω = ω
k
= kc/

. (4.21)
The corresponding group velocity is
¯ v
k
=
∂ω
k

¯
k
=
c

¯
k
k
=
c
2
ω
k

¯
k. (4.22)
In this case of a linear wave spectrum, the Wigner function F takes the form
F ≡ F(¯ r, t : ω,
¯
k) = F
k
(¯ r, t )δ(ω −ω
k
). (4.23)
Replacing it in equation (4.26), and noting that the reduced Wigner function
F
k
is independent of ω and consequently that

m
F
∂ω
m
= F
k

m
∂ω
m
δ(ω −ω
k
) = (−1)
m
δ(ω −ω
k
)

m
F
k
∂ω
m
= 0, (4.24)
72 Photon kinetic theory
we can write the Wigner–Moyal equation in a simplified form


∂t
÷ ¯ v
k
· ∇

F
k
÷
∂ ln
∂t
F
k
= −
ω
k

[ sin
k
F
k
] . (4.25)
Here,
k
is a reduced differential operator defined by

k
=
1
2


∂¯ r
·


¯
k

. (4.26)
Because, in the Wigner–Moyal equation, the sine operators are too compli-
cated to be calculated in specific problems, it is useful to simply retain the first
term in the development (4.20):
sin
k
.
k
. (4.27)
This is valid for a slowly varying medium, where the gradients contained in
the operator
k
are very small. In such a case, we are close to the conditions
where the geometric optics approximation is valid, and the Wigner–Moyal equa-
tion reduces to


∂t
÷ ¯ v
k
· ∇

F
k
÷

∂ ln
∂t

F
k
. −
ω
k
2


∂¯ r
·
∂ F
k

¯
k

. (4.28)
On the other hand, if we neglect the logarithmic derivative in this equation,
we notice that this equation implies that a triple equality exists, namely
dt =
d¯ r
¯ v
k
=
d
¯
k

k
/2)(∂/∂¯ r)
. (4.29)
This is equivalent to stating that
d¯ r
dt
= ¯ v
k
=
∂ω
k

¯
k
(4.30)
d
¯
k
dt
=
ω
k
2

∂¯ r
=
kc
2
3/2

∂¯ r
= −
∂ω
k
∂¯ r
. (4.31)
We recover here the ray equations of the geometric optics approximation,
identical to those used before. They are nothing but the characteristic equations of
the simplified version of the Wigner–Moyal equation, which can then be written
as
d
dt
F
k


∂t
÷ ¯ v
k
· ∇ ÷
d
¯
k
dt
·


¯
k

F
k
. 0. (4.32)
This equation, which states the conservation of the Wigner function F
k
, is
valid when the logarithmic time derivative, as well as the higher order deriva-
tives associated with the diffraction terms l > 0 in the development of the sine
Wigner–Moyal equation for electromagnetic radiation 73
operator sin
k
, can be neglected. Furthermore, from equation (4.17) we can
define F
k
(¯ r, t ) as the space Wigner function for the electric field, as shown in
appendix A:
F
k
(¯ r, t ) =

¯
E(¯ r ÷ ¯ s/2, t ) ·
¯
E

(¯ r − ¯ s/2, t )e
−i
¯
k·¯ s
d¯ s. (4.33)
It is now useful to define the number of photons N
k
(¯ r, t ) in terms of this
reduced Wigner function, as
N
k
(¯ r, t ) =

0
8
¨
h

∂ R
∂ω

ω
k
F
k
(¯ r, t ) (4.34)
where R = 0 is the dispersion relation of the medium. For the case considered
here of a non-dispersive medium it reduces to
R ≡ R(ω,
¯
k) = −c
2
k
2

2
= 0. (4.35)
The expression for the number of photons (4.34) is then reduced to
N
k
(¯ r, t ) =

0
4
¨
h

ω
k
F
k
(¯ r, t ). (4.36)
We can now return to the somewhat more exact expression for the Wigner–
Moyal equation (4.28) and rewrite it as
d
dt
F
k
= −

∂ ln
∂t

F
k
(4.37)
where the total derivative is determined by equation (4.32).
On the other hand, if we take the total time derivative of the number of
photons (4.36), and if we notice that

k
dt
=
∂ω
k
∂t
= −
ω
k
2

∂ ln
∂t

, (4.38)
we can then obtain
d
dt
N
k
=
¸
1
2

∂t
÷ ¯ v
k
· ∇

ln
¸
N
k
. (4.39)
Neglecting the slow variations of the refractive index appearing on the right-
hand side, we can finally state an equation of conservation for the number of
photons, in the form
dN
k
dt


∂t
÷ ¯ v
k
· ∇ ÷
d
¯
k
dt
·


¯
k

N
k
= 0. (4.40)
74 Photon kinetic theory
This is identical to the Klimontovich equation derived at the begining of this
chapter. This new derivation, which is much more complicated, has however the
advantage of using a more general definition for N
k
.
On the other hand, we understand from this that the conservation equation
for the number of photons is only valid when the higher order terms contained
in the sine operator of the Wigner–Moyal equation can be neglected. This means
that these terms represent the diffraction corrections to the geometric optics ap-
proximation.
Finally, we note that the classical Wigner function for the electromagnetic
field introduced in this section is sometimes used to characterize ultra-short laser
pulses with a time-dependent spectrum, as measured in optical experiments [45].
In contrast, an evolution equation of this quantity seems to have been ignored.
The Wigner–Moyal equation described here can eventually be used to understand
the space–time evolution of such short pulses along a given optical circuit.
4.2.2 Dispersive medium
The above derivation is conceptually quite interesting because it establishes a
clear link between the exact Maxwell’s equations and the heuristically derived
Klimontovich equation. However, its range of validity is not very wide because
we have neglected dispersion effects.
The generalization to the case of a dispersive medium is considered in this
section. For simplicity, we still neglect the losses in the medium, which can easily
be included afterwards.
First of all, if the electromagnetic radiation propagates in a dispersive
medium, our starting equation (4.16) has to be replaced by


2

1
c
2

2
∂t
2

¯
E = µ
0

2
∂t
2
¯
P (4.41)
where
¯
P =
0
¯
E −
¯
D is the vector polarization of the medium. In general terms it
can be related to the electric field
¯
E by the integral
¯
P(¯ r, t ) =
0

d¯ r
/

dt
/
χ(¯ r, t, ¯ r
/
, t
/
)
¯
E(¯ r − ¯ r
/
, t −t
/
). (4.42)
Again, we can derive from here an evolution equation for the double Wigner
function for the electric field. The derivation is detailed in appendix B and, to the
lowest order of the space and time variations of the medium, the result is


∂t
÷ ¯ v
g
· ∇

F = −
2
2ω ÷∂η
0
/∂ω
(ηF) (4.43)
where is the differential operator defined by equation (4.19) and ¯ v
g
is the group
velocity defined by
¯ v
g
=
2c
2
¯
k −ω
2
∂/∂
¯
k
2ω ÷ω
2
∂/∂ω
(4.44)
Photon distributions 75
with = 1 ÷χ = 1 ÷η/ω
2
.
This means that, if we had used the full developement of η
±
around (¯ r, t ),
instead of the first terms, we would have obtained the operator sin instead of just
, characteristic of the Wigner–Moyal equation. This equation therefore gener-
alizes the above derivation of this equation to the case of a dispersive medium.
Obviously, for a non-dispersive medium, such that ∂η
0
/∂ω = 0, this would
reduce to the result of the previous section.
Let us assume that the electromagnetic wave spectrum is made of a superpo-
sition of linear waves, such that we can use equation (4.23), F = F
k
δ(ω − ω
k
).
Then, equation (4.43) becomes


∂t
÷ ¯ v
k
·


¯
k
÷
1
(∂ω
2
/∂ω)
ω
k
∂η
k
∂¯ r
·


¯
k

F
k
= 0 (4.45)
where ¯ v
k
= (∂ω/∂
¯
k)
ω
k
and η
k
= ω
2
k
χ(¯ r,
¯
k, t ).
As an example of a dispersive medium, we can consider an isotropic plasma,
where we have η
k
= −ω
2
p
. In this case, the gradient of η
k
appearing in the last
term of this equation reduces to the gradient of the electron plasma density, or
equivalently, to the gradient of the square of the plasma frequency.
We have then


∂t
÷ ¯ v
k
·


¯
k

1

k
∂ω
2
p
∂¯ r
·


¯
k

F
k
= 0 (4.46)
where ω
k
=

k
2
c
2
÷ω
2
p
(¯ r, t ).
This is equivalent to stating that the reduced Wigner function F
k
is con-
served:
d
dt
F
k


∂t
÷ ¯ v
k
·


¯
k
÷
d
¯
k
dt
·


¯
k

F
k
= 0 (4.47)
because we know, from the photon ray equations, that
d
¯
k
dt
= −
∂ω
k
∂¯ r
= −
1

k
∂ω
2
p
∂¯ r
. (4.48)
From these reduced forms of the Wigner–Moyal equation for a dispersive
medium, we can then justify the use of the equation of conservation for the
number of photons N
k
(¯ r, t ), equation (4.40), which can also be called the kinetic
equation for photons propagating in slowly varying dispersive media. The prac-
tical interest of this kinetic approach will now be illustrated with a few specific
examples.
4.3 Photon distributions
In order to illustrate the interest of this kinetic approach, let us give some exam-
ples and introduce some definitions. First of all, it should be noticed that we have
76 Photon kinetic theory
introduced in equation (4.34) a more general definition for the number of photons
than those usually found in the literature. As it states, it can be applied to arbitrary
forms of wavefields (plane, spherical or cylindrical waves).
In particular, if we take the simple case of plane waves, such that
¯
E(¯ r, t ) =
¯
E
0
exp(i
¯
k
0
· ¯ r −iω
0
t ), our definition reduces to
N
k
(¯ r, t ) =

0
8
¨
h
∂ R
∂ω
[E
0
[
2
δ(
¯
k −
¯
k
0
). (4.49)
This is just the definition commonly found in the literature [93, 114] which
is not very useful to describe, for instance, short laser pulses.
4.3.1 Uniform and non-dispersive medium
Let us now consider some examples of solutions of the photon kinetic equation.
For simplicity, we will discuss one-dimensional propagation. The pertinent ki-
netic equation will be


∂t
÷v
k

∂x
÷ f
k

∂k

N
k
(x, t ) = 0. (4.50)
The simplest possible case corresponds to a photon beam propagation in a
uniform and non-dispersive medium. The third term in this equation will then be
equal to zero, due to uniformity:
f
k

dk
dt
= −
∂ω
∂x
=
kc
n
2
∂n
∂x
= 0 (4.51)
where the refractive index n is time independent.
Furthermore, in a non-dispersive medium, we also have v
k
= v
0
= c/n =
const. This means that the one-dimensional kinetic equation (4.50) is reduced to


∂t
÷v
0

∂x

N
k
(x, t ) = 0. (4.52)
A possible solution of this equation is a Gaussian pulse, with a duration τ,
which propagates along the medium without changing its shape. This can be
represented by
N
k
(x, t ) = N(k) exp
¸
−(x −v
0
t )
2

2
x
¸
(4.53)
where σ
x
= v
0
τ is the spatial pulse width, and N(k) describes the spectral
content.
Taking its space and time derivatives, we can easily see that this solution
satisfies equation (4.52) for an arbitrary function N(k). A useful choice is that of
a spectral Gaussian distribution, centred around some wavenumber value k
0
, with
a spectral width σ
k
N(k) = N
0
exp
¸
−(k −k
0
)
2

2
k
¸
. (4.54)
Photon distributions 77
Compatibility between the spectral and the spatial distributions implies that
we always have σ
k
≥ 1/σ
x
. The equality σ
k
σ
x
= 1 corresponds to the case of
a transform limited pulse, which minimizes the position–momentum uncertainty
relations.
4.3.2 Uniform and dispersive medium
As a first step in complexity, we consider a uniform but dispersive medium, where
the refractive index is frequency dependent. The third term in the photon kinetic
equation (4.50) will still be equal to zero, but the group velocity v
k
will not be a
constant. In order to calculate its value, let us linearize the refractive index around
the central pulse frequency
n(ω) . n
0
÷(ω −ω
0
)

dn

ω
0
(4.55)
where ω
0
= ω(k
0
) and n
0
= n(ω
0
). The dispersion relation can then be written
as
kc = n
0
ω ÷(ω −ω
0
)ωn
/
0
(4.56)
with n
/
0
= (dn/dω)
ω
0
. This can be solved for ω to give
ω =
1
2

ω
0

n
0
n
/
0

÷
1
2


0
−n
0
/n
/
0
)
2
÷4kc/n
/
0
. (4.57)
The group velocity becomes
v
k
=
∂ω
∂k
=
c

b
2
÷4kcn
/
0
(4.58)
with b = (ω
0
n
/
0
−n
0
).
We see that, for n
/
0
> 0, the higher frequencies inside the pulse (with
higher values of k) will travel with lower velocities and will be retarded along
propagation v
k
< v
0
, for k > k
0
. In such a medium, the pertinent equation to be
solved is


∂t
÷v
k

∂x

N
k
(x, t ) = 0. (4.59)
Instead of the Gaussian distribution (4.53), we can try here a solution of the
form
N
k
(x, t ) = N(k) exp
¸
−(x −v
k
t )
2

2
x
¸
(4.60)
wher v
k
is determined by equation (4.58) and is independent of x and t .
This is clearly a solution of equation (4.59). Let us see the meaning of this
new solution and compare it with (4.53), by linearizing v
k
around k
0
, v
k
. v
0
÷
(k −k
0
)v
/
0
, with
v
/
0
=

∂v
k
∂k

k
0
= −
2c
2
n
/
0
(b
2
÷4k
0
cn
/
0
)
3/2
. (4.61)
78 Photon kinetic theory
Replacing this development of v
k
in equation (4.60), and using N(k) as given
by equation (4.54), we obtain a photon distribution formally analogous to (4.53),
but with N(k) replaced by a space- and time-dependent function:
N
k
(x, t ) = N(k, x, t ) exp
¸
−(x −v
0
t )
2

2
x
¸
(4.62)
with
N(k, x, t ) = N
0
exp
¸
−(k −k
0
)
2

k
(t )
2
¸
exp
¸
2t (k −k
0
)v
/
0
(x −v
0
t )/σ
2
x
¸
. (4.63)
Here, we have used
1
σ
k
(t )
2
=
1
σ
2
k
÷
v
/
0
t
2
σ
2
x
. (4.64)
4.3.3 Pulse chirp
The above result shows that, when an initially Gaussian beam propagates in a
uniform but dispersive medium it maintains its spatial Gaussian shape but its
spectral distribution is distorted in time. At this point, the concept of pulse chirp
has to be introduced, because it is associated with such a pulse distortion.
In the frame of our kinetic description of a photon beam, the chirp is deter-
mined by the space–time distribution of the averaged wavenumber:
!k) =
1
n
γ
(x, t )

kN
k
(x, t )
dk

(4.65)
where the normalization factor
n
γ
=

N
k
(x, t )
dk

(4.66)
is the photon mean density.
The concept of pulse chirp is very important for the optics of short laser
pulses. We can say that a given pulse (or photon beam) is chirped if its averaged
wavenumber is not constant across the pulse. As an example, let us consider the
distribution (4.62–4.64). Replacing it in equation (4.65), we obtain
!k) =


−∞
k dk exp(−ak
2
÷2bk)


−∞
dk exp(−ak
2
÷2bk)
(4.67)
where we have used
a =
1
σ
k
(t )
2
, b =
k
0
σ
k
(t )
2
÷
v
/
0
σ
2
x
(x −v
0
t )t. (4.68)
Photon distributions 79
Using the following solutions for these two integrals [35]:


−∞
exp(−ak
2
÷2bk) dk =

π
a
exp

b
2
a

(4.69)
and


−∞
k exp(−ak
2
÷2bk) dk =
b
a

π
a
exp

b
2
a

, (4.70)
we obtain
!k) =
b
a
= k
0
÷v
/
0
σ
k
(t )
2
σ
2
x
(x −v
0
t )t. (4.71)
This shows that the mean value of the wavenumber inside the photon beam is
time dependent, which means that propagation of a Gaussian pulse in a dispersive
medium leads to pulse chirping, simply because some of the photons will travel
with larger velocities than the others. In a similar way, we can calculate the
spectral width of the chirp, by defining a new mean value
!k
2
) =
1
n
γ
(x, t )

k
2
n
k
(x, t )
dk

. (4.72)
In our case, it can be written as
!k
2
) =


−∞
k
2
dk exp(−ak
2
÷2bk)


−∞
dk exp(−ak
2
÷2bk)
. (4.73)
Using the following solution for the integral in the numerator [35]:


−∞
k
2
exp(−ak
2
÷2bk) dk =
1
2a

π
a

1 ÷
2b
2
a

exp

b
2
a

, (4.74)
we obtain
!k
2
) =
1
2a

1 ÷
2b
2
a

=
σ
k
(t )
2
2
÷!k)
2
. (4.75)
This is equivalent to writing
!(k)
2
) ≡ !(k −!k))
2
) =
σ
k
(t )
2
2
. (4.76)
According to the definition of σ
k
(t ), as given by equation (4.64), this means
that the square mean deviation decreases with time, while the pulse chirp associ-
ated with the mean value !k) increases in time. Starting from a Gaussian spectral
distribution, at t = 0, such that !k) is constant across the pulse spatial width, we
obtain a thin and elongated spectral distribution as time evolves (or as the pulse
propagates).
80 Photon kinetic theory
x
k
N
k
( a)
x
k
N
k
( b)
Figure 4.1. Representation of the photon distribution N
k
(x, t ) in phase space (x, k), at a
given time t , for (a) a transform limited Gaussian pulse; (b) a chirped pulse.
A more general class of solutions for the photon kinetic equation for uniform
dispersive media can be stated as
N
k
(x, t ) = N(k, x, t ) exp
¸
−(x −v
k
t )
2

2
x
¸
(4.77)
with
N(k, x, t ) = N
0
e
−[k−g(x,t )]
2

2
k
. (4.78)
Using this in equation (4.59) we can easily realize that this is a solution of the
photon kinetic equation, for a uniform but dispersive medium, for every function
Photon distributions 81
satisfying the equation


∂t
÷v
k

∂x

g = 0. (4.79)
As an example of a solution, we can consider the family of functions
g(x, t ) = k
0
÷ g
0
(x −v
k
t )
n
(4.80)
where k
0
and g
0
are constants and n is an integer.
4.3.4 Non-stationary medium
Let us now assume a non-dispersive but non-stationary and non-uniform medium,
such that the refractive index is determined by
n ≡ n(x, t ) = n
0
[1 ÷δ f (x, t )] (4.81)
where f (x, t ) describes a small space–time perturbation and δ < 1 is the scale
of the perturbation.
The dispersion relation can then be written as
ω =
kc
n
0
[1 ÷δ f (x, t )]
.
kc
n
0
[1 −δ f (x, t )] (4.82)
and the photon, or group velocity, as
v
k
≡ v
g
(x, t ) . v
0
[1 −δ f (x, t )] (4.83)
where we have used v
0
= c/n
0
.
The photon kinetic equation becomes


∂t
÷v
g
(x, t )

∂x
÷kv
0
δ
∂ f
∂x

∂k

N
k
(x, t ) = 0 (4.84)
where we have used v
0
= c/n
0
. Let us try a solution of the form (4.77, 4.78), but
with v
k
replaced by the constant v
0
. Using this in equation (4.84) we notice that
the function g(x, t ) has to satisfy the following equation:


∂t
÷v
g

∂x

g = δv
0
k
∂ f
∂x
÷δ f v
2
0
σ
2
k
σ
2
x
(x −v
0
t )
(k − g)
. (4.85)
For a given perturbation of the refractive index of the medium f (x, t ), this
equation allows us to obtain the solution g(x, t ). However, such solutions, even
if they exist, are in general very difficult to find.
In order to obtain an approximate solution for g(x, t ), we can assume that
we are interested in the main region of the pulse, such that x . v
0
t , and the last
82 Photon kinetic theory
term in this equation can be neglected. If we also take v
g
. v
0
, the equation for
g(x, t ) is reduced to


∂t
÷v
0

∂x

g . δv
0
k
∂ f
∂x
. (4.86)
An important example is that of a wakefield perturbation of the refractive
index, such that f (x, t ) = cos k
p
(x − ut ). In this case, the approximate solution
for g(x, t ) is
g(x, t ) = k
0
÷ g
0
cos k
p
(x −ut ) (4.87)
with
g
0
= −δ
kv
0
u −v
0
. (4.88)
This shows that the spectral shifts are more important when the phase ve-
locity of the perturbation is nearly equal to the group velocity of the photons,
u ∼ v
0
, as already found in the analysis of single photon trajectories. But, of
course, these analytical results are quite rough and, for this and other situations,
numerical solutions of the photon kinetic equation are required.
4.3.5 Self-blueshift
An interesting result [91] was obtained from the numerical integration of the pho-
ton kinetic equation, for the self-blueshift of a laser pulse penetrating in a neutral
gas. This blueshift is due to the sudden ionization of the gas and subsequent
creation of an ionization front.
Self-blueshift is a particular aspect of photon acceleration where the space–
time changes in the refractive index are due to the incident laser pulse itself.
Experimentally, this has been well known since the early sixties [120, 122], and
it was studied theoretically by several authors [41, 48].
Let us describe the ionization model. For very intense incident laser fields,
the photoionization processes are described by the tunnelling ionization theory.
If the initial density of the neutral atoms of the gas is n
0
(¯ r, t ), this number will
decrease due to field ionization as
dn
0
(¯ r, t )
dt
= −w
1
(¯ r, t )n
0
(¯ r, t ) (4.89)
where w
1
is the ionization rate for the atom.
The number of singly ionized atoms n
1
(¯ r, t ) will then be determined by the
balance equation
dn
1
(¯ r, t )
dt
= w
1
(¯ r, t )n
0
(¯ r, t ) −w
2
(¯ r, t )n
1
(¯ r, t ) (4.90)
where w
2
is the probability of double ionization occurring.
Photon fluid equations 83
In more general terms, the number density of j -charged ions is determined
by
dn
j
(¯ r, t )
dt
= w
j
(¯ r, t )n
j −1
(¯ r, t ) −w
j ÷1
(¯ r, t )n
j
(¯ r, t ). (4.91)
Finally, the density of completely ionized atoms is given by
dn
Z
(¯ r, t )
dt
= w
z
(¯ r, t )n
Z
(¯ r, t ) (4.92)
where Z is the atomic number of the atoms in the gas.
The various ionization probabilities w
j
depend on the local amplitude of the
laser electric field and are well known from the standard theory of field ionization
[4, 47]. The values of these ionization probabilities can establish the coupling
between these balance equations for the atomic states and the above kinetic equa-
tion for the photons. Furthermore, the local plasma dispersion relation can be
consistently determined by considering charge neutrality.
The electron plasma density will then be determined by
n
e
(¯ r, t ) =
j =Z
¸
j =0
j n
j
(¯ r, t ). (4.93)
A numerical integration of the kinetic equation for photons, coupled with
these simple balance equations, clearly shows that the front of a Gaussian beam
penetrating in a neutral gas region will be significantly blueshifted (see figure
4.2). Furthermore, the blueshift is not uniform and several wavelike perturbations
can be observed in the up-shifted spectrum, one for each ionization state. These
spectral undulations will be eventually attenuated by electron impact ionization
which was not contained in this model.
4.4 Photon fluid equations
In this chapter we have been able to show that the photon kinetic equation can be
derived from Maxwell’s equations, in the limit of slowly space- and time-varying
media. Even if the range of validity of the photon kinetic theory is limited, it
nevertheless contains detailed information about the photon spectrum. It is then
useful to derive fromthis equation a set of conservation laws describing the space–
time behaviour of photon averaged quantities, in the same way as fluid equations
can be derived from kinetic equations for the particles of an ordinary gas.
Let us first define the mean density of the photon gas as the integral of the
photon distribution over the entire spectrum:
n
γ
(¯ r, t ) = 2

N
k
(¯ r, t )
d
¯
k
(2π)
3
. (4.94)
84 Photon kinetic theory
Figure 4.2. Numerical simulation of the self-blueshift of a laser beam with an intensity of
10
15
W cm
−2
, propagating in neutral argon, with an initial Gaussian spectrum. The initial
density of the neutral atoms is 10
18
cm
−3
for x > 0.
Similarly, we can define the mean photon velocity, or mean group velocity,
as
¯ u(¯ r, t ) =
2
n
γ

¯ v
k
N
k
(¯ r, t )
d
¯
k
(2π)
3
. (4.95)
The factor of two appearing in these two definitions is due to the existence of
the two independent polarization states. The continuity equation for the photon
gas, or photon density conservation equation, can be derived by integrating the
kinetic equation in
¯
k. Integration of the first term gives

∂t

N
k
d
¯
k
(2π)
3
=
1
2
∂n
γ
∂t
. (4.96)
Integration of the second term leads to

¯ v
k
· ∇N
k
d
¯
k
(2π)
3
= ∇ · (¯ un
γ
) −

N
k
(∇ · ¯ v
k
)
d
¯
k
(2π)
3
. (4.97)
Finally, integration of the third term gives

¯
f ·
∂ N
k

¯
k
d
¯
k
(2π)
3
=

N
k
∇ · ¯ v
k
d
¯
k
(2π)
3
. (4.98)
The sum of these three contributions leads to the continuity equation for the
photon gas
∂n
γ
∂t
÷∇ · (¯ un
γ
) = 0. (4.99)
Photon fluid equations 85
The second conservation equation to be derived is the photon momentum
conservation law. Multiplying the photon kinetic equation by ¯ v
k
and integrating
it over the entire spectrum, we get the following contribution from the first term:

¯ v
k
∂ N
k
∂t
d
¯
k
(2π)
3
=
1
2

∂t
(¯ un
γ
) −

N
k
∂ ¯ v
k
∂t
d
¯
k
(2π)
3
. (4.100)
The contribution of the second term can be written as

¯ v
k
¯ v
k
· ∇N
k
d
¯
k
(2π)
3
=
1
2
∇ ·

n
γ
!¯ v
k
¯ v
k
)

N
k
∇ · (¯ v
k
¯ v
k
)
d
¯
k
(2π)
3
(4.101)
where we have used the following quantity:
!¯ v
k
¯ v
k
) =
2
n
γ

N
k
¯ v
k
¯ v
k
d
¯
k
(2π)
3
. (4.102)
Here, we can introduce the concept of photon pressure, P
γ
, such that
!¯ v
k
¯ v
k
) = ¯ u¯ u ÷
P
γ
n
γ
I (4.103)
where I is the 3 3 identity matrix.
Using equation (4.102) and adding the diagonal terms of this tensorial rela-
tion, it can easily be shown that the photon pressure is determined by
P
γ
=
2
3

N
k
(v
k
−u)
2
d
¯
k
(2π)
3
. (4.104)
We can also define an effective temperature for the photon gas, assuming
that it can be considered an ideal gas, such that
P
γ
= n
γ
T
eff
. (4.105)
Finally, the contribution from the third term of the photon kinetic equation
to the momentum conservation law is determined by
¯ v
k

¯
f ·
∂ N
k

¯
k

d
¯
k
(2π)
3
. (4.106)
Adding the contributions from the three terms and using the continuity equa-
tion, we can finally establish the momentum conservation law of the photon gas,
in the form
∂ ¯ u
∂t
÷ ¯ u · ∇¯ u = −
∇P
γ
n
γ
÷
¯
F
γ
(4.107)
where the mean force acting on the photons is determined by
¯
F
γ
=
2
n
γ

N
k
¸
∂ ¯ v
k
∂t
÷∇ · (¯ v
k
¯ v
k
) ÷


¯
k
· (
¯
f ¯ v
k
)
¸
d
¯
k
(2π)
3
. (4.108)
86 Photon kinetic theory
As a first example, let us calculate this force for the case of a non-dispersive
dielectric medium, with refractive index n ≡ n(¯ r, t ). In this case, we have
ω =
kc
n
, ¯ v
k
=
c
n
¯
k
k
. (4.109)
Using these in the above expression for
¯
F
γ
, we obtain
¯
F
γ
=
2
n
γ
c
2
n
2

N
k
¸

∂n
∂t
¯
k
k
÷
1
n
∂n
∂¯ r
·

I −2
¯
k
¯
k
k
2

d
¯
k
(2π)
3
. (4.110)
We see that this mean force acting on the photon gas is due to the space and
time variations of the refractive index. As a second example, let us consider an
isotropic plasma, such that
ω
2
= k
2
c
2
÷ω
2
p
, ¯ v
k
=
¯
kc
2
ω
. (4.111)
The mean force now becomes
¯
F
γ
= −
1
n
γ

N
k
ω
2

∇ω
2
p
÷ ¯ v
k
∂ω
2
p
∂t

d
¯
k
(2π)
3
. (4.112)
If the distribution is even with respect to the quantity (¯ v
k
− ¯ u), this can also
be written as
¯
F
γ
= −
U
n
γ

∇ω
2
p
÷ ¯ u
∂ω
2
p
∂t

(4.113)
with
U =

N
k
ω
2
d
¯
k
(2π)
3
. (4.114)
In a similar way, we can also derive the energy conservation equation, by
multiplying the kinetic equation by
¨
hω and integrating it over
¯
k. The contribution
of the first term is

¨

∂ N
k
∂t
d
¯
k
(2π)
3
=
1
2

∂t
W
γ

¨
h

N
k
∂ω
∂t
d
¯
k
(2π)
3
(4.115)
where the mean photon energy is determined by
W
γ
=
2
n
γ

¨
hωN
k
d
¯
k
(2π)
3
. (4.116)
The contribution of the second term is

¨
hω¯ v
k
· ∇N
k
d
¯
k
(2π)
3
=
1
2
∇ · (¯ uW
γ
) −
¨
h

N
k
∇ · (ω¯ v
k
)
d
¯
k
(2π)
3
. (4.117)
Self-phase modulation 87
Finally, the contribution of the third term is

¨

¯
f ·
∂ N
k

¯
k
d
¯
k
(2π)
3
= −
¨
h

N
k


¯
k
· (ω
¯
f )
d
¯
k
(2π)
3
(4.118)
which exactly cancels the last term in equation (4.117).
Adding these three contributions, we obtain

∂t
W
γ
÷∇ · (¯ uW
γ
) = 2
¨
h

N
k
∂ω
∂t
d
¯
k
(2π)
3
. (4.119)
In the case of a non-dispersive dielectric medium, we have
∂ω
∂t
= −
ω
n
∂n
∂t
(4.120)
where n is the refractive index, and, in the case of a plasma,
∂ω
∂t
=
1

∂ω
2
p
∂t
(4.121)
where ω
p
is the electron plasma frequency.
We see that, in contrast with the momentum conservation equation (4.107,
4.108), where both the time and the space derivatives of the refractive index
(or of the plasma frequency) contribute to the change in the mean value of the
momentum of the photon gas, only the time derivatives can be a source of energy.
4.5 Self-phase modulation
An important property of the photon kinetic theory is that it can describe the
nonlinear changes of a photon bunch or a short laser pulse due to its own spectrum.
In other words, it can explain the following self-consistent process: the photon
bunch changes the optical properties of the medium which, in turn, modify the
spectral composition of the photon bunch. This was already shown in the example
of the self-blueshift, given at the end of section 4.4.
The best example of such a process is however the well-known self-phase
modulation of a laser pulse in an optical fibre or in other optical media. By this
process, an initially nearly monochromatic laser pulse can be transformed into a
broadband radiation pulse.
In more exact terms, if the initial laser pulse is a nearly transform limited
short pulse, where the spectral width is close to its minimum value allowed by the
uncertainty principle, the nonlinear process, usually called self-phase modulation,
can transform it into nearly white light.
This astonishing effect has been known experimentally since the early 1970s
and the current theory is based on the calculation of the nonlinear contributions to
the total time phase dependence. However, the photon kinetic theory described in
88 Photon kinetic theory
this chapter suggests the possible use of a totally independent theoretical expla-
nation, where the phase is completely ignored and where only the photon number
distribution is considered.
It will be shown here that this new approach is equally capable of describing
the spectral changes characterizing the so-called self-phase modulation processes,
which leads us to two important conclusions. The first one is that the phase is not
an essential ingredient of self-phase modulation. This is also valid for the induced
phase modulation, briefly discussed at the end of chapter 3. The second is that
self-phase modulation (as well as the other phase modulation processes, such as
induced and crossed phase modulation) is nothing but a particular aspect of the
phenomenology of photon acceleration described in this work.
Physically this means that, starting from a nearly monochromatic laser pulse,
the nonlinear properties of the optical media can accelerate and decelerate the
photons contained inside the pulse, spreading them over a considerable range
of the optical spectrum, thus leading to nearly white light. In order to clarify
these few very important physical statements, and to make them accessible to the
nonspecialist, let us first remind ourselves of the usual optical theory of self-phase
modulation and then compare it to the photon kinetic approach.
4.5.1 Optical theory
If we start from Maxwell’s equations in a dielectric medium, in the absence of
charge and current distributions, we can easily derive the following equation of
propagation for the electric field
¯
E associated with the laser pulse:

2
¯
E −∇(∇ ·
¯
E) −
1
c
2

2
¯
E
∂t
2
= µ
0

2
¯
P
∂t
2
. (4.122)
The vector ploarization appearing in this equation can be divided into two
distinct parts:
¯
P =
¯
P
L
÷
¯
P
NL
. The linear part is determined by
¯
P
L
(t ) =
0


0
χ
(1)
(τ)
¯
E(t −τ) dτ. (4.123)
Here we assume that the medium is isotropic, otherwise the linear (or first-
order) susceptibility χ
(1)
would be replaced by a tensor. For the nonlinear part of
the polarization vector, we can neglect dispersion and simply write
¯
P
NL
(t ) =
0
χ
(3)
[E(t )[
2
¯
E(t ). (4.124)
Assuming that the laser field is transverse (∇ ·
¯
E = 0), we can then write the
propagation equation as

2
¯
E −
1
c
2

2
∂t
2
¸
¯
E ÷


0
χ
(1)
(τ)
¯
E(t −τ) dτ
¸
=
1
c
2

2
∂t
2
χ
(3)
[E[
2
¯
E. (4.125)
Self-phase modulation 89
Let us assume a solution of the form
¯
E(¯ r, t ) =
1
2
¸
¯ aE
0
(z, t ) exp(ik
0
z −iω
0
t ) ÷c.c.
¸
. (4.126)
This represents a laser pulse propagating along the z-axis, with a unit po-
larization vector ¯ a and a slowly varying amplitude E
0
(z, t ). This means that the
scales of variation of this amplitude are such that

1
E
0
∂ E
0
∂t


0
,

1
E
0
∂ E
0
∂z

<k
0
. (4.127)
We can then write

2
¯
E
∂t
2
. −ω
2
0
¯
E ÷
1
2
¸
−2iω
0
¯ a exp(ik
0
z −iω
0
t )
∂ E
0
∂t
÷c.c.
¸
(4.128)

2
¯
E . −k
2
0
÷
1
2
¸
2ik
0
¯ a exp(ik
0
z −iω
0
t )
∂ E
0
∂z
÷c.c.
¸
.
We will also assume weak linear dispersion, which means that the function
χ
(1)
(τ) is sharply peaked around τ = 0. The absence of dispersion would
correspond to identifying this function with a Dirac delta function δ(τ).
In the weak dispersion medium, we can expand the electric field
¯
E(t − τ)
appearing in equation (4.123) as
¯
E(t −τ) .
¯
E(t ) −τ

∂t
¯
E(t ) ÷· · · . (4.129)
Using this expansion in equation (4.123), and noting that
χ(ω) =


0
χ
(1)
(t )e
iωt
dt (4.130)

∂ω
χ(ω) = i


0
t χ
(1)
(t )e
iωt
dt
we obtain
¯
P
NL
(t ) =

0
2
¸
¯ a exp(ik
0
z −iω
0
t )
¸
χ(ω
0
) ÷i
∂χ(ω
0
)
∂ω

∂t
¸
E
0
(z, t ) ÷c.c.
¸
.
(4.131)
We can now use equations (4.129, 4.131) in the propagation equa-
tion (4.125). Noting that the frequency ω
0
and the wavenumber k
0
are related
by the linear dispersion relation k
2
c
2
= ω
2
(ω), where (ω) = 1 ÷ χ(ω) is the
dielectric function of the medium, we obtain an equation describing the slow
space–time evolution of the envelope field amplitude:


∂z
÷
1
v
0

∂t

E
0
= iω
0
α[E
0
[
2
E
0
. (4.132)
90 Photon kinetic theory
Here v
0
is the group velocity, determined by
1
v
0
=
k
0
ω
0
÷
ω
2
0
2c
2
∂(ω
0
)
∂ω
(4.133)
and α is the nonlinear coefficient, resulting from the existence of a third-order
susceptibility
α =
ω
0
k
0
1
c
2
χ
(3)
. (4.134)
The solution of the envelope equation (4.132) can be written in the form
E
0
(z, t ) = A(η) e
iφ(η,t )
(4.135)
where we have η = z −v
0
t and the nonlinear phase
φ(η, t ) = φ
0
÷ω
0
α[ A(η)[
2
t. (4.136)
Such a solution describes a laser pulse propagating with an invariant pulse
shape, but with a phase which depends on time (or on the travelled distance),
as well as on the form of that shape. The result is that the pulse frequency will
not remain constant along the propagation and will depend on the actual position
inside the pulse envelope:
ω = ω
0

∂φ
∂t
= ω
0
¸
1 ÷α
t
v
0

∂η
[ A(η)[
2
¸
. (4.137)
As shown before, this space–time dependence of the frequency (or of the
wavevector) is called the pulse chirp. In order to have a more precise idea of
the chirp, and of the amount of frequency shift ω = ω − ω
0
introduced by the
nonlinear susceptibility of the medium, we can take the simple and illustrative
example of a Gaussian laser pulse, such that
[ A(η)[
2
= A
2
0
e
−η
2

2
. (4.138)
From equation (4.137), we get
ω = −αt
ω
0
v
0

σ
2
[ A(η)[
2
. (4.139)
We see that the frequency shift grows linearly with time, and that it is nega-
tive at the pulse front, for η > 0, and positive at the rear, for η < 0.
4.5.2 Kinetic theory
An alternative description of the same effect is now given in terms of photon
acceleration [103]. Instead of using the envelope field equation, we use, as our
Self-phase modulation 91
starting point, the equation describing the space–time evolution of the photon
number distribution N
k
(¯ r, t ), which can be written as
d
dt
N
k
(¯ r, t ) = 0 (4.140)
where the total time derivative is
d
dt
=

∂t
÷
d¯ r
dt
·

∂¯ r
÷
d
¯
k
dt
·


¯
k
. (4.141)
In order to determine this total time derivative operator we have to make use
of the ray equations
d¯ r
dt
=
∂ω

¯
k
,
d
¯
k
dt
= −
∂ω

¯
k
(4.142)
where, in the expression of the frequency ω ≡ ω
k
(¯ r, t ), we retain the nonlinear
corrections to the refractive index
ω =
kc
n
=
kc
n
0
÷n
2
I (¯ r, t )
. (4.143)
Here, n
0
is the linear refractive index, n
2
is the nonlinear one and I (¯ r, t ) is
the laser pulse intensity, where
I (¯ r, t ) =
¨
h

ω
k
N
k
(¯ r, t )
d
¯
k
(2π)
3
. (4.144)
This is a slowly varying function as compared with the frequency timescale:
ω ·


∂t
ln I (¯ r, t )

. (4.145)
Using equation (4.143) in (4.142), we get
d¯ r
dt
=
c
n
0
÷n
2
I (¯ r, t )
(4.146)
d
¯
k
dt
=
kc
[n
0
÷n
2
I (¯ r, t )]
2
n
2

∂¯ r
I (¯ r, t ). (4.147)
We know that the formal solution of the photon kinetic equation (4.140) can
be written as
N(¯ r,
¯
k, t ) = N(
¯
k
0
(¯ r,
¯
k, t ), ¯ r
0
(¯ r,
¯
k, t ), t
0
) (4.148)
where (
¯
k
0
, ¯ r
0
, t
0
) are the initial conditions.
This means that, by solving equations (4.146, 4.147) we are able to deter-
mine the evolution of N
k
(¯ r, t ). In order to illustrate the procedure, let us solve
these equations in a one-dimensional situation, where propagation is taken along
the z-axis.
92 Photon kinetic theory
We will also assume that the nonlinear corrections to the refractive index
are small, n
0
· n
2
I (¯ r, t ), which is usually the case, even for very intense laser
pulses. We can then reduce equations (4.146, 4.147) to
dz
dt
=
c
n
0
¸
1 −
n
2
n
0
I (z, t )
¸
(4.149)
dk
dt
=
kc
n
0
n
2
n
0

∂z
I (z, t ). (4.150)
Let us also assume that the pulse propagates without significant profile de-
formation, with a group velocity c/n
0
. This means that
I (z, t ) . I

z −
c
n
0
t

. (4.151)
This suggests the use of a canonical transformation from (z, k) to a new pair
of variables (η, p), such that
η = z −
c
n
0
t, p = k. (4.152)
The resulting Hamiltonian function is
(η, p, t ) = ω(η, p, t ) − p
c
n
0
. (4.153)
Using the approximate expression
ω .
kc
n
0
¸
1 −
n
2
n
0
I (z, t )
¸
=
pc
n
0
¸
1 −
n
2
n
0
I (η)
¸
(4.154)
we get
(η, p) = −pwI (η) (4.155)
where we have used
w =
n
2
n
0
c
n
0
. (4.156)
The ray equations can now be written in the form

dt
=

∂p
= −wI (η),
dp
dt
= −

∂η
= pw
∂ I (η)
∂η
. (4.157)
The integration of these photon equations of motion is straighforward, and
the result can be written as

η
η
0

/
I (η
/
)
= −w(t −t
0
) (4.158)
Self-phase modulation 93
and
p
0
= p exp
¸
−w

t
t
0
∂ I (η)
∂η
dt
¸
. (4.159)
The first of these expressions shows that the initial value of the position
coordinate η is independent of the momentum variable p:
η
0
≡ η
0
(η, t ). (4.160)
With regard to the second expression, it corresponds to an implicit integra-
tion, but it will be very useful for the calculation of the spectral changes of the
laser pulse along its propagation. Let us show this by considering the chirp of the
pulse !ω).
By definition, the chirp is the averaged frequency of the pulse, at a given
position η and at a given instant t :
!ω)
η,t
=

N
p
(η, t )ω(η, p, t ) dp. (4.161)
Using equations (4.148, 4.154), we get
!ω)
η,t
=

pc
n
0
¸
1 −
n
2
n
0
I (η)
¸
N(η
0
, p
0
, t
0
) dp. (4.162)
Using equation (4.159) we can write p dp in terms of p
0
dp
0
, which leads to
!ω)
η,t
=

p
0
c
n
0
¸
1 −
n
2
n
0
I (η)
¸
N(η
0
, p
0
, t
0
) exp
¸
2w

∂ I (η)
∂η
dt
¸
. (4.163)
But for a non-dispersive pulse, we have I (η) = I (η
0
), as already stated in
equation (4.151). This means that we can write the pulse chirp in the form
!ω)
η,t
= !ω)
0
exp
¸
2w
∂ I (η)
∂η
(t −t
0
)
¸
(4.164)
where !ω)
0
≡ !ω)
η
0
,t
0
is the initial pulse chirp.
This is a simple but important result, which will allow us to rediscover the
main features of the so-called self-phase modulation, this time without consider-
ing the field phase. In particular, we can determine the condition for an extremum

∂η
!ω)
η,t
= 0. (4.165)
This will correspond to the stationary points !ω)
max
of the chirp curve. From
equation (4.164) it is obvious that

∂η
!ω)
η,t
= 2w(t −t
0
)!ω)
η,t
¸

2
∂η
2
I (η)

= 0. (4.166)
94 Photon kinetic theory
From this we conclude that the point η = η
max
, for which the maximum
value of the chirp !ω)
max
is observed, is determined by the condition

2
∂η
2
I (η) = 0. (4.167)
Let us illustrate these results by assuming a Gaussian pulse
I (η) = I
0
e
−η
2

2
. (4.168)
For this particular pulse shape, we have

2
∂η
2
I (η) =


2
σ
2
−1

2
σ
2
I (η). (4.169)
Comparing this with condition (4.167), we conclude that the chirp maxima
are located at the two points determined by 2η
2
= σ
2
, or equivalently, by
η
max
= ±
1

2
σ. (4.170)
From equation (4.168) we then get

∂η
I (η)

η
max
= ∓

2
σ
I
0
e
−1/2
. (4.171)
Using equation (4.164) we finally obtain the condition for the maximum
chirp:
!ω)
max
= !ω)
0
exp
¸

2

2
σ
wI
0
e
−1/2
(t −t
0
)

. (4.172)
It should be noticed that this concept of maximum chirp can also be physi-
cally identified with the condition for the maximum frequency shift with respect
to the initial pulse frequency. Introducing the frequency shift as ω = !ω)
max

!ω)
0
, we get, for small arguments of the above exponential,
ω . ±
2

2
σ
n
2
n
0
c
n
0
I
0
e
−1/2
t (4.173)
where we have used t = t −t
0
.
Notice that such an expansion is valid for small time intervals, or for small
nonlinearities. This result for the pulse frequency shift is in agreement (not only
qualitatively, but also quantitatively) with that obtained above for the self-phase
modulation of the laser pulse in a nonlinear medium: the spectral width grows
linearly with time and the maxima down-shifts and the up-shifts are equal to each
other.
Self-phase modulation 95
<ω>
max
<ω>
0
τ
Figure 4.3. Time evolution of the maximum frequency shift associated with self-phase
modulation, as determined by the photon kinetic theory.
We should however notice that, for long time intervals t → ∞, the above
expansion in not valid and the Stokes and anti-Stokes sidebands become quite
asymmetric. This is clearly seen in the more exact equation (4.172) where, for
long time intervals, one of the branches of the exponential tends to zero, while the
other tends to infinity. Such a spectral feature is well confirmed by experiments
and is illustrated in figure 4.3.
This asymmetry is a natural consequence of the present theoretical formula-
tion, even for an initially symmetric laser pulse. In contrast, it cannot be derived
from the optical theory of self-phase modulation discussed above.
Another important feature of this effect, which is the pulse steepening, can
also be explained by our kinetic model. In order to study the changes in the pulse
shape, we can no longer use the non-dispersive solution stated in equation (4.151),
which was one of the basic assumptions of our analytical calculation of the pulse
chirp. Instead, we have to solve numerically the photon kinetic equation. The
result is illustrated in figure 4.4, where the pulse steepening, resulting from the
group (or photon) velocity dispersion, is clearly shown.
We conclude this section by stating that the effect usually called self-phase
modulation can be accurately described with the photon kinetic equation, where
the wavefield phase is completely ignored. The three main experimental features
of this effect are well described by this kinetic model: the pulse chirp, the spectral
asymmetry between the Stokes and the anti-Stokes sidebands, and the steepening
of the pulse shape. In many respects, the analytical solutions of the kinetic theory
are more accurate than the analytical solutions of the optical theory, in particular
for the prediction of spectral asymmetry.
We are then led to two different kinds of conclusion. The first one is that
the so-called self-phase modulation can be seen as a particular example of the
more general concept of photon acceleration in a non-stationary medium. In this
96 Photon kinetic theory
Figure 4.4. Illustration of pulse steepening associated with self-phase modulation, as
described by the kinetic theory. A Gaussian pulse is also shown, for comparison.
particular case, the non-stationarity of the medium is due to the laser pulse itself
(or to the photon number distribution associated with the pulse) which locally
changes the refractive index, due to the nonlinear response of the medium.
The second, and more surprising, conclusion is that the wavefield phase is
not the essential ingredient of this effect: two laser pulses, with different phase
contents, would lead to nearly the same spectral broadening in the same medium.
This means that the explanation of such a frequency shift or spectral change is
conceptually more accurate in terms of photon kinetics than in terms of phase
modulation.
Chapter 5
Photon equivalent charge
Until now, we have only considered the linear properties of photons in a medium,
which essentially derive from their effective mass. In some sense these are the
most relevant ones.
In this chapter we turn our attention to the influence of the nonlinear prop-
erties of an optical medium, and we show that the photons possess an equivalent
electric charge. In a plasma, this equivalent charge can be directly associated with
the ponderomotive force (or the radiation pressure) of the photon gas acting on
the plasma electrons. We will later see that in a non-ionized medium, for instance
an optical fibre, this electric charge is replaced by an electric dipole.
This new, and somewhat unexpected property of a photon is due to the space–
time varying polarization effects that it induces in the medium. We can then say
that a photon in a vacuum is a ‘bare’ particle with no electric charge, but a photon
with the same frequency (or energy) travelling in a plasma has to be considered
a ‘dressed’ particle with an electric charge which is not quantized and depends
on the plasma density [69, 104, 112]. Furthermore, the existence of this charge
allows us to formulate a new series of problems, related to possible radiation
mechanisms due to accelerated ‘dressed’ photons, or to their motion across the
boundary between two different media.
Finally, a newproperty associated with this electric charge (or in other words,
with the radiation pressure effects) is the possibility of resonant Cherenkov inter-
actions between photons and electron plasma waves, leading to the concept of
photon Landau damping [14]. Plasma instabilities, induced by particular photon
spectral distributions (such as photon beams), can then be envisaged.
5.1 Derivation of the equivalent charge
One possible way to derive the photon equivalent charge is to consider an elec-
tromagnetic wavepacket propagating in a plasma. Once again, we will assume
for simplicity that the plasma is unmagnetized. The electric field
¯
E ≡
¯
E(¯ r, t )
97
98 Photon equivalent charge
associated with this wavepacket is described by the wave equation

2
¯
E −∇(∇ ·
¯
E) −
1
c
2

2
¯
E
∂t
2
= µ
0

¯
J
∂t
. (5.1)
For high-frequency waves, we can neglect the ion motion, and the current in
this equation is reduced to the electron current
¯
J = −en¯ v. (5.2)
Phase velocities of transverse electromagnetic waves in an unmagnetized
plasma are always larger than the speed of light c, which means that kinetic effects
are usually negligible and that we can use the electron fluid equations in order to
determine the electron density n and the electron mean velocity.
In their simple non-relativistic version (which is valid only for moderate
wave intensities) the electron fluid equations are
∂n
∂t
÷∇ · n¯ v = 0 (5.3)
∂ ¯ v
∂t
÷ ¯ v · ∇¯ v = −
e
m

¯
E ÷ ¯ v
¯
B

. (5.4)
In the first of these equations we recognize the continuity equation of the
electron fluid, and in the second one, the momentum conservation equation. The
magnetic field
¯
B appearing in the Lorentz force term is the magnetic field of the
electromagnetic wave itself and can be expressed in terms of the electric field
¯
E
by using the Faraday equation

¯
E = −

¯
B
∂t
. (5.5)
Equations (5.4) are nonlinear and their solution is in general quite compli-
cated. However we can, as a first approximation, linearize them with respect to
the wave electric field. Let us call
¯
E
1
this linear solution for the electric field.
If we restrict our discussion to purely transverse waves, we have ∇ ·
¯
E
1
= 0
and the wave equation (5.1) reduces to

2
¯
E
1

1
c
2

2
¯
E
1
∂t
2
= µ
0

¯
J
1
∂t
(5.6)
where the linear current is
¯
J
1
= −en
0
¯ v
1
. (5.7)
Here n
0
is the equilibrium (or unperturbed) electron plasma density. The
linear response of the plasma is determined by the linearized version of equa-
tions (5.3, 5.4)
∂n
1
∂t
÷n
0
∇ · ¯ v = 0 (5.8)
∂ ¯ v
1
∂t
= −
e
m
¯
E
1
. (5.9)
Derivation of the equivalent charge 99
The solution of this equation can take the general form
¯
E
1
(¯ r, t ) =
1
2
¯
E
0
(¯ r, t ) exp(i
¯
k
0
· ¯ r −iω
0
t ) ÷c.c. (5.10)
where ω
0
and
¯
k
0
are the mean frequency and wavevector of the wavepacket and
¯
E
0
(¯ r, t ) is a slowly varying amplitude. We should notice that such a general
expression could also be a solution for nonlinear wavepackets.
According to equations (5.8, 5.9), the electron density and velocity pertur-
bations associated with this solution have similar expressions, with amplitudes
determined by
n
1
= n
0
¯
k
0
ω
0
· ¯ v
1
, ¯ v
1
=
e

0
m
¯
E
0
. (5.11)
We notice that, for purely tranverse wave solutions, such that
¯
k
0
·
¯
E
0
= 0, the
electron density perturbations are equal to zero: n
1
= 0. Finally, the amplitude of
the wave magnetic field is, according to equation (5.5),
¯
B
1
=
¯
k
0
ω
0

¯
E
0
. (5.12)
Once this simple linear solution is understood, we can return to the wave
equation (5.1) and try to obtain a more adequate solution which takes the nonlin-
ear terms into account. This can be done by assuming that the total electric field
in this equation is divided in two parts: the main field (which corresponds to the
linear solution) and a small correction due to radiation or polarization effects, i.e.
¯
E =
¯
E
1
÷
¯
E
2
(5.13)
with
[
¯
E
1
[ ·[
¯
E
2
[. (5.14)
To the first order in the small nonlinear field
¯
E
2
, we can obtain from equa-
tion (5.1)

2
¯
E
2
−∇(∇ ·
¯
E
2
) −
1
c
2

2
∂t
2
¯
E
2
= µ
0

∂t
¯
J
2
(5.15)
where the nonlinear current is determined by
¯
J
2
= −en
0
¯ v
2
. (5.16)
To the same order of approximation, and assuming that equations (5.8, 5.9),
stay valid, we can also get from equations (5.3, 5.4),
∂n
2
∂t
÷n
0
∇ · ¯ v
2
= 0 (5.17)
∂ ¯ v
2
∂t
÷ ¯ v
1
· ∇¯ v
1
= −
e
m
(
¯
E
2
÷ ¯ v
1

¯
B
1
). (5.18)
100 Photon equivalent charge
Now we should make use of the vector identity
∇(¯ a ·
¯
b) = (¯ a · ∇)
¯
b ÷(
¯
b · ∇)¯ a ÷ ¯ a (∇
¯
b) ÷
¯
b (∇ ¯ a). (5.19)
If we identify the vector ¯ a with the first-order velocity perturbation ¯ v
1
and
the vector
¯
b with its complex conjugate, we can easily realize that equation (5.18)
can be rewritten as
∂ ¯ v
2
∂t
÷
1
2
∇[v
1
[
2
= −
e
m
¯
E
2
. (5.20)
In doing so we are neglecting the high-frequency part of the nonlinear pertur-
bation ¯ v
2
, which would not be valid if we wanted to study harmonic generation.
After elimination of ¯ v
2
from equations (5.17, 5.20) we get the following evolu-
tion equation for the low-frequency nonlinear perturbation of the electron plasma
density:

2
n
2
∂t
2

en
0
m
∇ ·
¯
E
2
=
n
0
2

2
[v
1
[
2
. (5.21)
Using equation (5.11) this can also be written as the equation of a forced
linear oscillator

2
n
2
∂t
2
÷ω
2
p
n
2
=
1
2
e
2
n
0
m
2
ω
2
0

2
[E
0
(¯ r, t )[
2
. (5.22)
The eigenfrequency of this linear oscillator is nothing but the unperturbed
electron plasma frequency ω
p
= (e
2
n
0
/
0
m)
1/2
. In this equation [E
0
(¯ r, t )[
2
describes the slowly varying envelope of the electromagnetic wavepacket.
If this wavepacket propagates along the x-axis, with a group velocity ¯ v
g
=
v
g
¯ e
x
, with negligible shape deformation, we can write
¯
E
0
(¯ r, t ) ≡
¯
E
0
(ξ, ¯ r

) (5.23)
with
ξ = x −v
g
t. (5.24)
We can now perform a coordinate transformation from (x, t ) to (ξ, τ), with
τ = t . This means that

∂x
=

∂ξ
,

∂t
=

∂τ
−v
g

∂ξ
. (5.25)
Using this in the equation for the forced oscillator (5.22), we obtain


2
∂τ
2
÷ω
2
p
÷v
2
g

2
∂ξ
2
−2v
g

2
∂τ∂ξ

n
2
=

0
2m

ω
p
ω
0

2


2
∂ξ
2
÷∇
2

[E
0
(ξ, ¯ r

)[
2
. (5.26)
Derivation of the equivalent charge 101
Before solving this equation we can simplify it in two ways. First we notice
that the gradients in the perpendicular direction can be considered negligible in
several relevant physical situations. In particular, for a ‘pancake’-like wavepacket
(a very short laser pulse), or for wavepackets not very much different from plane
waves, we can always make ∇
2

<∂
2
/∂ξ
2
.
Second, if we assume that, in the frame of the wavepacket the pulse shape
and the electron perturbations induced by it are not significantly changing, we
can also use the so-called quasi-static approximation: ∂/∂τ . 0. With these two
drastic but physically acceptable approximations the above equation is reduced to
a much simpler form:


2
∂ξ
2
÷k
2
p

n
2
=

0
2m
k
2
p
ω
2
0

2
∂ξ
2
[E
0
(ξ)[
2
(5.27)
with k
p
= ω
p
/v
g
.
It is useful, at this point, to introduce a dimensionless space variable η = k
p
ξ
and to replace the square module of the electric field by the number of photons
N(ξ) =

4
¨

0
[E
0
(ξ)[
2
. (5.28)
Equation (5.27) now becomes


2
∂η
2
÷1

n
2
= f (η) (5.29)
where the force term is
f (η) =
2
¨
hk
2
p

0

2
∂η
2
N(η). (5.30)
An appropriate solution of equation (5.29) is
n
2
(η) =

η

f (η
/
) sin(η −η
/
) dη
/
. (5.31)
Integrating by parts, we have
n
2
(η) =
¸
g(η
/
) sin(η −η
/
)
¸
η

÷

η

g(η
/
) cos(η −η
/
) dη
/
. (5.32)
Here we have used f (η) = dg/ dη. Noting that g(η
/
) = 0 for η
/
→∞, and
that sin(η −η
/
) = 0 for η
/
= η, we obtain
n
2
(η) =

η

g(η
/
) cos(η −η
/
) dη
/
(5.33)
or, more explicitly,
n
2
(η) =
2
¨
hk
2
p

0

η

cos(η −η
/
)

∂η
/
N(η
/
) dη
/
. (5.34)
102 Photon equivalent charge
This equation is appropriate for a second integration by parts. But for our
purpose this will not be necessary because we notice that, in the limit of very short
pulses, such that the pulse duration t is shorter than the period of the electron
plasma oscillations t <ω
−1
p
, we have η = (η −η
/
) <1. This means that we
can take cos(η −η
/
) . 1, and obtain
n
2
(η) =
2
¨
hk
2
p

0
N(η). (5.35)
We can now define a photon equivalent charge Q
ph
, such that
−en
2
(η) = Q
ph
N(η). (5.36)
This means that
Q
ph
= −2
¨
h
ek
2
p

0
. (5.37)
The current associated with the N(η) particles of charge Q
ph
moving with
velocity ¯ v
g
= v
g
¯ e
x
is obviously given by
¯
J
2NL
= Q
ph
N(η)v
g
¯ e
x
. (5.38)
The same expression for the second-order nonlinear current can also be
derived if we start from the time derivative of the total second-order current
¯
J
2
=
¯
J
2L
÷
¯
J
2NL
:

¯
J
2
∂t
= −en
0
∂ ¯ v
2
∂t
=
e
2
n
0
m
¯
E
2
÷

0
2m
ω
2
p
ω
2
0
∇[E
0
[
2
. (5.39)
Writing this equation in terms of the variables (ξ, τ) and making the above
two assumptions of quasi-static (∂/∂τ = 0) and of a ‘pancake’ pulse (∇

<
∂/∂ξ) approximations, we get
−v
g

¯
J
2
∂ξ
=
e
2
n
0
m
¯
E
2
−v
2
g

∂ξ
Q
ph
N(ξ)¯ e
x
. (5.40)
We see that the second term leads to an expression for
¯
J
2NL
which coincides
with equation (5.38). This justifies the use of the concept of photon equivalent
charge introduced above.
Our derivation of Q
ph
was simplified by considering the integration of equa-
tion (5.34) in the limit of a very short pulse or wavepacket. But it can also be
obtained if we consider a pulse with an arbitrary duration. In this case, the electron
second-order perturbation will consist of two terms: the above term (5.35) which
is directly related to the photon effective charge Q
ph
, and a second term which
represents the wakefield or plasma oscillation left behind the pulse [69].
Photon ondulator 103
To illustrate the existence of these two components, let us return to equa-
tion (5.34) and integrate it by parts. We are led to
n
e
(η) =
2
¨
hk
2
p

0
¸
N(η) −

η

N(η
/
) sin(η −η
/
) dη
/
¸
(5.41)
or, using equation (5.37),
n
2
(η) = −
1
e
Q
ph
N(η) ÷
1
e
Q
ph

η

N(η
/
) sin(η −η
/
) dη
/
. (5.42)
The first term was already given by equation (5.36) and represents the equiv-
alent charge of the total photon distribution. The second term represents the
wakefield left behind it. It is nothing but an electron plasma oscillation moving
with a phase velocity equal to the photon velocity v
g
, and which follows the pulse
as a wake. This justifies the statement made in chapter 3 about the possibility
of generating relativistic electron plasma waves by means of a short laser pulse
propagating in a plasma.
Another interesting aspect of the above calculations is the negative value of
the photon effective charge. The sign of Q
ph
can be physically well understood
if we notice that the ponderomotive force appearing on the right-hand side of
equation (5.21) tends to push the plasma electrons out of the region occupied by
the electromagnetic wavepacket. This means that the photons tend to repel the
electrons. Their equivalent charge is therefore negative.
5.2 Photon ondulator
We have just seen how the nonlinear current associated with an electromagnetic
wavepacket propagating in a plasma can be derived. Our result (5.37, 5.38) shows
that such a current is due to a flow of a number N(ξ) of photons with charge Q
ph
moving with velocity ¯ v
g
. Replacing this in the wave equation (5.15) we obtain

2
¯
E
2
−∇(∇ ·
¯
E
2
) −
1
c
2

2
∂t
2
¯
E
2
=
ω
2
p
c
2
÷µ
0

∂t
Q
ph
N(ξ)¯ v
g
. (5.43)
This equation determines the secondary field radiated by the localized
wavepacket (or the primary photon bunch) moving across the medium. An
example of such a radiation effect is discussed here. Let us restrict our discussion
to purely transverse fields, such that ∇ ·
¯
E
2
= 0.
Using a time Fourier transformation
¯
E
2
(¯ r, t ) =


−∞
¯
E
ω
(¯ r) e
−iωt


(5.44)
we obtain

2
¯
E
ω
÷
ω
2
c
2
(ω)
¯
E
ω
= −iωµ
0


−∞
¯ v
g
Q
ph
N(ξ) e
iωt
dt. (5.45)
104 Photon equivalent charge
Here, we have used the plasma dielectric constant (ω) = 1 − ω
2
p

2
.
Neglecting the transverse structure of N(ξ), or the dependence of the photon
distribution on the transverse variable ¯ r

, we can use
¯
E
ω
(¯ r) = A
ω
(x) e
i
¯
k

·¯ r

¯ e
ω
(5.46)
where ¯ e
ω
is the unit polarization vector and A
ω
is the spectral field amplitude.
Equation (5.45) now reduces to

2
A
ω
∂x
2
÷k
2
A
ω
= −iωµ
0
(¯ e
ω
· ¯ e
x
)


−∞
v
g
Q
ph
N(ξ) e
iωt
dt (5.47)
where
k
2
=
ω
2
c
2
(ω) −k
2

. (5.48)
The solution of equation (5.47) for radiation emitted in the backward direc-
tion (or in the direction of negative values of x) is given by
A
ω
(x) = −µ
0
ω
2kc
(¯ e
ω
· ¯ e
x
) e
−ikx


−∞
dt

x
x
0
dxv
g
Q
ph
N(ξ) e
ikx÷iωt
. (5.49)
Notice that, due to the factor (¯ e
ω
· ¯ e
x
) we can only observe a radiated trans-
verse field if k

is not equal to zero. Let us assume the case of radiation due to a
modulation in the electron mean density
n
0
(x) = n
0
÷ ˜ n cos(
˜
kx). (5.50)
This leads to
k
2
p
=
ω
2
p
v
2
g
.
ω
2
p0
v
2
g0
¸
1 ÷
˜ n
n
0
cos(
˜
kx)
¸
. (5.51)
The modulations in the photon group velocity also exist, but they were ne-
glected (v
g
. v
g0
) because they would only give second-order contributions.
Using this expression in equation (5.37) we can get from equation (5.49)
A
ω
(x) = A e
−ikx


−∞
dt

x
x
0
dx N(ξ) cos(
˜
kx) e
ikx÷iωt
(5.52)
with
A = µ
0
(¯ e
ω
· ¯ e
x
)
ω
ω
0
e
m
ω
2
p0
kcv
g0
˜ n
n
0
. (5.53)
The asymptotic field radiated very far away from the emitting region can
then be written as
A
ω
(x →−∞) = A e
−ikx


−∞
dt


−∞
dξ N(ξ)
cos[
˜
k(ξ ÷v
g0
t )] e
ik(ξ÷v
g0
t )÷iωt
. (5.54)
Photon transition radiation 105
Integration in time leads to
A
ω
(x →−∞) = π A

dξ N(ξ)
¸
e
i(
˜
k÷k)ξ
δ(ω −ω
1
)
÷ e
−i(
˜
k−k)ξ
δ(ω −ω
2
)
¸
(5.55)
with
ω
1
= v
g0
(
˜
k −k), ω
2
= v
g0
(
˜
k ÷k). (5.56)
This result shows that an electromagnetic wavepacket (or equivalently a
photon bunch described by N(ξ)) travelling in a plasma with a modulated electron
density, radiates in two characteristic frequencies which depend on the periodicity
scale of the modulation. This is similar to the radiation of electrons travelling in
a vacuum in the presence of a modulated magnetic field (an electron ondulator),
showing once more the analogies between a photon in a plasma and a charged
particle. Such a radiation process [69] can then be called a photon ondulator.
5.3 Photon transition radiation
Let us examine another example of a radiation process directly due to the exis-
tence of the equivalent charge: the possibility of transition radiation emitted by
a photon bunch at a plasma boundary. This process is asociated with the sudden
disappearence of the equivalent charge [71].
In order to simplify the problem, we will assume a nearly monochromatic
burst of photons moving along the x-axis. We have then
N
k
(¯ r, t ) = (2π)
3
N(ξ, ¯ r

)δ(
¯
k −
¯
k
0
) (5.57)
where ξ = x −v
0
t , and v
0
= v
k
is the group velocity for
¯
k = k
0
¯ e
x
.
Let us also assume a plasma–vacuum transition layer described by
ω
2
p
(x) = ω
2
p0
f (x). (5.58)
A plausible choice for the shape function is f (x) = [1 ÷ tanh(x/)]/2,
where is the width of the boundary layer. The nonlinear current
¯
J
2NL
reduces
to
¯
J
2NL
= Q
0
N(ξ, ¯ r

)v
0
(x) f (x)¯ e
x
(5.59)
with Q
0
= Q
ph
(k) for k = k
0
and for ω
p
= ω
p0
. We know that the group velocity
v
0
is determined by
v
0
(x) = c

1 −(ω
p0

0
)
2
f (x). (5.60)
In order to give a simple physical explanation of the photon transition radi-
ation, we consider the particular form of a sharp plasma boundary. We consider
106 Photon equivalent charge
the simplest case of a boundary with infinitesimal width → 0, such that
f (x) = H(−x), where H(x) is the Heaviside function. We also assume that
the duration of the bunch of photons is negligible, so that we can write
N(ξ, ¯ r

) = N(¯ r

)δ(ξ). (5.61)
In this case, the nonlinear current becomes
¯
J
2NL
= −Q
0
N(¯ r

)v
0
H(−x)δ(x −v
0
t )¯ e
x
. (5.62)
This means that the source term in the wave equation can be written as

∂t
¯
J
2NL
= Q
0
N(¯ r

)v
2
0
H(−x)δ
/
(x −v
0
t )¯ e
x
= − Q
0
N(¯ r

)v
0
δ(t )δ(x)¯ e
x
. (5.63)
We see that the source term is located both in space and time. In space, it is
located precisely at the plasma boundary. In time it is located at the instant when
the photon bunch crosses the boundary. This means that, in the vacuum region
(x > 0), the transverse field radiated by the boundary is determined by


2

1
c
2

2
∂t
2

¯
E = −µ
0
Q
0
v
0
N(¯ r

)δ(t )δ(x)¯ e
x
. (5.64)
This equation shows that an infinitely sharp boundary will radiate an
infinitely large spectrum. This is very similar to the well-known case of
bremsstrahlung radiation, where the particle velocity changes and its charge
remains constant (because it is an invariant). In contrast here, the velocity of the
radiating particle remains nearly constant and its equivalent charge goes to zero.
In order to obtain a finite spectral width, we can replace equation (5.61) by
a more realistic description of a bunch of photons coming out of the plasma. For
instance, we can assume a Gaussian bunch described by
N(ξ, ¯ r

) =
N(¯ r

)
σ

π
exp
¸

ξ
2
σ
2

(5.65)
where σ is the bunch width. The nonlinear current becomes
¯
J
2NL
= −Q
0
v
0
N(¯ r

)H(−x)
¯ e
x
σ

π
exp
¸

(x −v
0
t )
2
σ
2

. (5.66)
The wave equation for the secondary radiated field can now be written as


2

1
c
2

2
∂t
2

¯
E
2
= −µ
0
Q
0
v
0
N(¯ r

)
¯ e
x
τ

π
exp
¸

t
2
τ
2

δ(x) (5.67)
Photon transition radiation 107
where τ is here σ/v
0
.
Following a procedure similar to that of the previous section, we arrive at a
general solution that can be written in the following simple way:
A
ω
(z) = A
0
(ω) e
ikx
(5.68)
with
A
0
(ω) = i
µ
0
2k
Q
0

τ
v
0
(ˆ e

ω
· ¯ e
x
)N(
¯
k

) exp
¸

ω
2
τ
2
4

(5.69)
where k
2
= (ω/c)
2
−k
2

, and
N(
¯
k

) =

N(¯ r

) e
−i
¯
k

·¯ r

d¯ r

. (5.70)
This result shows that the energy radiated by the disappearing equivalent
charge is proportional to the square of this charge, Q
2
0
, and to the square of the
photon velocity, v
2
0
, as in the usual linear [57] and nonlinear [62] transition radi-
ation processes. Furthermore, radiation is not emitted along the particle direction
of motion because of the geometric factor (ˆ e

ω
· ¯ e
x
). As already noticed, this factor
is equal to zero for parallel radiation
¯
k

= 0, because of the transverse nature of
the radiated field.
Finally, let us discuss the importance of the transverse dimensions of the
burst of particles crossing the boundary. If these transverse dimensions are neg-
ligible, we can write N(¯ r

) = N
0
δ(¯ r

). In this case, we can use N(
¯
k

) = 1 in
equation (5.70), and the radiated field will have a maximum for nearly perpen-
dicular direction of propagation, where (¯ e

ω
· ¯ e
x
) . 1. In the opposite limit of an
infinitely large bunch, such that N(¯ r

) = N
0
, and N(
¯
k

) = (2π)
2
N
0
δ(
¯
k

), we
will have no transition radiation at all.
These simple calculations can obviously be refined, and the general case
of an arbitrary photon distribution crossing an arbitrary plasma boundary can be
studied with the aid of equations (5.37, 5.43). This shows that, in quite general
conditions, a primary distribution of photons can radiate a large spectrum of elec-
tromagnetic waves (or secondary photons) when they cross a plasma boundary
with nearly constant velocity. This is a direct consequence of their equivalent
electric charge in a plasma.
This new kind of transition radiation is qualitatively similar to the usual
transition radiation of charged particles (for instance electrons) moving across
a dielectric discontinuity with constant velocity. In particular, the radiated power
is proportional to the square of the particle velocity and of the particle equivalent
charge. However, in some respects, the new type of transition radiation discussed
here is also quite similar to the usual bremsstrahlung, with the difference that,
instead of a change in velocity, we have a change in the value of the particle
charge.
108 Photon equivalent charge
5.4 Photon Landau damping
In section 4.3 we considered the interaction of a photon distribution with an
electron plasma wave moving with a relativistic phase velocity. But it can be
easily recognized that our approach was not self-consistent in the sense that the
electron plasma wave was defined a priori, and its evolution was assumed to
be independent of the radiation spectrum. Here we return to the same problem,
but with a more consistent view of the physics of the photon–plasma interaction,
in the sense that the electron plasma oscillations will be coupled to the photon
field [14].
In equilibrium, we can characterize the plasma by its electron mean density
n
0
, and the radiation spectrum by some distribution of photons, N
k0
, for instance
the Planck distribution for a given plasma temperature T. If the plasma is per-
turbed, the photon field will also be perturbed because they are coupled to each
other.
This means that we can write, for the total electron plasma density n and for
the total photon distribution N
k
, the following expressions:
n = n
0
÷ ˜ n, N
k
= N
k0
÷
˜
N
k
. (5.71)
The objective of the present section is to show how the coupling between
these two perturbed quantities can be described, and what are the new physical
processes associated with it. If we use the electron fluid equations and Poisson’s
equation for the electrostatic field, we can derive the following equation for the
electron density perturbation:

2
˜ n
∂t
÷ω
2
p0
˜ n −3v
2
e

2
˜ n = −n
0
∇ ·
¸
∂ ¯ v
∂t

(5.72)
where ω
p0
is the unperturbed electron plasma frequency and v
e
=

T/m is the
electron thermal velocity.
In this equation, the right-hand side describes the ponderomotive force due
to the photon field. We can write it more explicitly as
¸
∂ ¯ v
∂t

= −
1
2
∇[v[
2
= −
1
2

e
m

2

[E
k
[
2
ω
2
k
d
¯
k
(2π)
3
. (5.73)
Here we have assumed the linear solution for the electron motion in the
radiation field ¯ v = −i(e
¯
E
k
/mω
k
). This can also be written in terms of the photon
number distribution, if we use the plane wave definition
N
k
=

8
¨
h

∂ R
∂ω

k
[E
k
[
2
(5.74)
where R ≡ R(ω, k) = 0 is the photon dispersion relation.
Photon Landau damping 109
Replacing this in equation (5.72), we obtain

2
˜ n
∂t
÷ω
2
p0
˜ n −3v
2
e

2
˜ n = 4
¨
h
ω
2
p0
m

2

˜
N
k
ω
2
k
(∂ R/∂ω)
k
d
¯
k
(2π)
3
. (5.75)
In the absence of any radiation field,
˜
N
k
= 0, the right-hand side of this
equation is equal to zero and, for perturbations of the form ˜ n ∼ exp i(
¯
k · ¯ r −ωt ),
this equation will lead to the well-known linear dispersion relation for electron
plasma waves
ω
2
= ω
2
p0
÷3v
2
e
k
2
. (5.76)
It should be kept in mind that, in this equation, the frequency ω and the
wavenumber k are related to the electron plasma wave spectrum, and not to the
photon spectrum. Furthermore, it is known from the plasma kinetic theory that
this dispersion relation is only valid if the electron Landau damping is negligible,
which implies that the phase velocity of the electrostatic wave has to be much
larger than the electron thermal velocity: (ω/k) ·v
e
. We will come back to that
point later.
In the general case where the radiation field is present and we have
˜
N
k
,= 0,
equation (5.75) is coupled with the kinetic equation for the photon field which,
after an appropriate linearization, determines
˜
N
k
as a function of N
k0
and ˜ n:

˜
N
k
∂t
÷ ¯ v
k
·

˜
N
k
∂¯ r
= −
¯
f ·
∂ N
k0

¯
k
. (5.77)
According to our theory of photon acceleration, the force acting on the
photons is determined by
¯
f = −∇ω
k
, where ω
k
=

ω
2
p
÷k
2
c
2
. This leads
to the following explicit expression for the force acting on the photons:
¯
f = −
1

k
e
2

0
m
∇˜ n. (5.78)
The pair of equations (5.75, 5.77) will then describe the coupling between the
electron plasma waves associated with the density perturbation ˜ n and the photon
number perturbations
˜
N
k
. Let us assume sinusoidal perturbations of the form
˜ n(¯ r, t ) = ˜ n(t ) e
i
¯
k·¯ r
,
˜
N
k
/ (¯ r, t ) =
˜
N
k
/ (t ) e
i
¯
k·¯ r
. (5.79)
Here, the value of the wavevector
¯
k, characterizing the scale of the perturba-
tion in both the electron density and the photon spectrum, should not be confused
with the wavevector
¯
k
/
characterizing a specific Fourier component of the photon
spectrum. In physical terms we are dealing with a situation which only makes
sense when we have
¯
k
/
·
¯
k. This means that the electron plasma wave has a much
larger scale length than the photons described by
˜
N
k
/ , which makes plausible the
description of the transverse radiation field as a fluid of particles (the photons)
110 Photon equivalent charge
evolving in a slowly varying background medium (the plasma and its electrostatic
wave).
Using equation (5.79) in equations (5.75, 5.77), we obtain

2
˜ n
∂t
2
÷(ω
2
p0
÷3k
2
v
2
e
) ˜ n = −4
¨
hk
2
ω
2
p0
m

˜
N
k
/
ω
/2
(∂ R/∂ω)
ω
/
d
¯
k
/
(2π)
3
(5.80)
and

˜
N
k
/
∂t
÷i
¯
k · ¯ v
k
/
˜
N
k
/ =
i

/
e
2

0
m
˜ n
¯
k ·
∂ N
k
/
0

¯
k
/
(5.81)
where ω
/
≡ ω(
¯
k
/
).
We can nowfollowthe usual Landau approach [82], but apply it to the photon
distribution and not to the electron distribution. This implies the use of a time
Laplace transformation
˜ n(ω) =


0
˜ n e
iωt
dt (5.82)
and a similar transformation for
˜
N
k
/ (ω), where ω is here a complex quantity.
From equation (5.81) we then get a relation between ˜ n(ω) and
˜
N
k
/ (ω):
˜
N
k
/ = −
˜ n
2n
0
ω
2
p0
ω
/
¯
k · (∂ N
k
/
0
/∂
¯
k
/
)
ω −
¯
k · ¯ v(
¯
k
/
)
. (5.83)
Similarly, from equation (5.80) we get

ω
2
p0
÷3k
2
v
2
e
−ω
2

˜ n
= 2ω
4
p0
¨
hk
2
m
˜ n
n
0

1
ω
/3
(∂ R/∂ω)
ω
/
¯
k · (∂ N
k
/
0
/∂
¯
k
/
)
ω −
¯
k · ¯ v(
¯
k
/
)
d
¯
k
/
(2π)
3
. (5.84)
We can use (∂ R/∂ω)
ω
/ . 2/ω
/
. From these two equations we can then
derive the following expression:
ω
2
= 3k
2
v
2
e
÷ω
2
p0
¸
1 −
¨
hk
2
mn
0
ω
2
p0

1
ω
/2
¯
k · (∂ N
k
/
0
/∂
¯
k
/
)
ω −
¯
k · ¯ v(
¯
k
/
)
d
¯
k
/
(2π)
3

. (5.85)
In order to develop this integral we have to consider separately the paral-
lel and the perpendicular photon motion with respect to the propagation of the
electron plasma wave. This can be done by defining
¯
k
/
= p
¯
k
k
÷
¯
k
/

, ¯ v(
¯
k
/
) = u( p,
¯
k
/

)
¯
k
k
÷ ¯ v

. (5.86)
The integral in equation (5.85) becomes

1
ω
/2
¯
k · (∂ N
k
/
0
/∂
¯
k
/
)
ω −
¯
k · ¯ v(
¯
k
/
)
d
¯
k
/
(2π)
3
= −

d
¯
k
/

(2π)
3

dp
ω
/2
∂ N
k
/
0
/∂p
u −(ω/k)
. (5.87)
Photon Landau damping 111
Developing the parallel photon velocity u around the resonant value defined
by u( p
0
) = (ω/k)
u( p,
¯
k
/

) . u( p
0
,
¯
k
/

) ÷( p − p
0
)

∂u
∂p

p
0
(5.88)
we get

dp
ω
/2
∂ N
k
/
0
/∂p
u −(ω/k)
.
1
(∂u/∂p)
p
0

dp
ω
/2
∂ N
k
/
0
/∂p
p − p
0
. (5.89)
This last integral can be written in the standard form
I (z) ≡

h(z)
z − z
0
dz = P

h(z)
z − z
0
dz ÷iπh(z
0
) (5.90)
where P

means the Cauchi principal part of the integral.
Replacing this result in equation (5.85) and using ω = ω
r
÷iγ , we get from
the real part of the resulting equation
ω
2
r
= 3k
2
v
2
e
ω
2
p0
¸
1 ÷
¨
hk
2
mn
0
ω
2
p0
P

1
p − p
0
∂G
p
∂p
dp

, (5.91)
and from the imaginary part
γ = π
¨
hk
2
ω
3
p0
2mn
0

∂G
p
∂p

p
0
. (5.92)
In these two expressions, G
p
is a kind of reduced distribution function for
the photon gas, which resulted from the integration of the equilibrium photon
distribution N
k
/
0
over the perpendicular directions:
G
p
=

1
ω
/2
( p
0
)
N
k
/
0
(∂u/∂p)
p
0
d
¯
k
/

(2π)
3
. (5.93)
Equations (5.91, 5.92) are the main results of this section. The first one
represents the dispersion relation of electron plasma waves (in the high phase
velocity regime) in the presence of a photon distribution. The second one deter-
mines the damping of these plasma waves due to the resonant interaction with the
photon spectrum. Due to the similarities of this expression with the well-known
electron Landau damping, this new damping effect can be called the photon
Landau damping.
The similarities between the electron and the photon contributions can be
stated in a more appropriate way if we notice that the dispersion relation (5.91)
can be rewritten as
1 ÷χ
e
÷χ
ph
= 0 (5.94)
where χ
e
is the electron susceptibility and χ
ph
is the photon susceptibility.
112 Photon equivalent charge
According to equation (5.91) the electron susceptibility is
χ
e
= −
ω
2
p0
ω
2
r
−3
k
2
v
2
e
ω
2
r
(5.95)
and the photon susceptibility is
χ
ph
= −
¨
hk
2
mn
0
ω
2
p0
P

1
p − p
0
∂G
p
∂p
dp. (5.96)
As we have noticed already, the above expression for the electron suscepti-
bility is only valid in the limit of very high phase velocities. In the general case,
χ
e
would depend on the actual electron distribution function and moreover, an
electron Landau damping would have to be added to the above photon Landau
damping. These electron kinetic effects are well documented in plasma physics
textbooks and will not be described here.
The important point to notice here is that these electron kinetic effects are
negligible for electron plasma waves with high phase velocities, like those pro-
duced by intense laser pulses propagating in a plasma. In this case the resulting
electron plasma waves (or wakefields) have relativistic phase velocities nearly
equal to the laser pulse group velocity from which they originate.
The important conclusion of the above calculation is that, even in the ab-
sence of resonant electron populations which could exchange efficiently their
energy with the electron plasma wave, the relativistic plasma perturbations can
nevertheless exchange energy with the photon gas. The resulting wave damping
is described by equation (5.92). This means that photon Landau damping can
replace the electron Landau damping, showing that in many respects the photon
dynamics in the field of an electrostatic wave in a plasma is similar to the electron
dynamics, as already documented in chapter 3 for single photon trajectories.
For a plasma in thermal equilibrium at a temperature T, the Planck distribu-
tion
N
k
/
0
=
ω
/2
π
2
c
3
1
exp(
¨

/
/T) −1
(5.97)
where ω
/
=

ω
2
p0
÷k
/2
c
2
should be used in the above equations.
Because of the derivative contained in the expression for the photon Landau
damping (5.92) this equilibrium distribution will always lead to a negative value
of γ , or to a real wave damping. However, the situation can change if a photon
beam is superposed on the Planck distribution. Then, in some regions of phase
velocities, the derivative can be positive, indicating the possibility of a wave
instability or wave growth. This will be illustrated below.
It should also be noticed that a quasi-linear diffusion coefficient for the pho-
tons can be derived from this theory, showing how the photon spectrum changes
in the presence not of a single plasma wave as before, but in the presence of a
spectrum of electron plasma waves [14].
Photon beam plasma instabilities 113
5.5 Photon beam plasma instabilities
In order to illustrate the new physical aspects contained in the dispersion relation
(5.94, 5.96) let us consider a one-dimensional problem where
N
k
/
0
= (2π)
2
N
k
/ δ(
¯
k
/

). (5.98)
The photon susceptibility will then reduce to
χ
ph
=
¨
hk
2
mn
0
ω
2
p0
ω
2

1
ω
2
k
/
∂ N
k
/ /∂k
/
(ω/k) −v(k
/
)
dk
/

(5.99)
where k
/
≡ k
/
|
. Using the photon (group) velocity
v(k
/
) =
c
2
k
/
ω
k
/
(5.100)
we obtain, after integration by parts,

1
ω
k
/
∂ N
k
/ /∂k
/
(ω/k) −(c
2
k
/

k
/ )
dk
/

= −
1
c
2

1
ω
k
/
N
k
/
[(ω/k)(ω
k
/ /c
2
) −k
/
]
2
dk
/

. (5.101)
Let us assume a mono-energetic photon beampropagating across the plasma:
N
k
/ = 2πN
0
δ(k
/
−k
/
0
). (5.102)
Using ω
/
0
= ω
k
/
0
we can write, after performing the integration,
χ
ph
= −
¨
hk
4
mn
0
ω
4
p0
ω
2
ω
/
0
k
2
c
2
ω
/
0
N
0
(ω −ku
0
)
2
(5.103)
where u
0
= v(k
/
0
).
Furthermore, if we neglect the electron thermal effects, the electron suscep-
tibility reduces to
χ
e
= −
ω
2
p0
ω
2
. (5.104)
The dispersion relation for the electron plasma waves (5.94) can then be
written as
1 −
ω
2
p0
ω
2
¸
1 ÷

2
(ω −ku
0
)
2

= 0 (5.105)
where we have used

2
=
¨
hk
4
c
2
mn
0
ω
2
p0
ω
/3
0
N
0
. (5.106)
114 Photon equivalent charge
We can see from equation (5.105) that this quantity plays nearly the same
role in the photon susceptilibity that is played by the electron plasma frequency
ω
p0
in the electron susceptibility. Due to that, we can call it the effective photon
plasma frequency.
But the analogy with a plasma frequency can be explored even further, and
we can define a photon plasma frequency
ph
, formally identical to the electron
plasma frequency, such that

2
ph
=
Q
2
ph
N
0

0
m
eff
(5.107)
where Q
ph
is the photon equivalent charge as determined by equation (5.37), and
m
eff
=
¨

p0
/c
2
is the photon effective mass introduced in chapter 2. This means
that we can write

2
ph
=
4
¨
hN
0
mn
0
ω
5
p0

/
0
c)
2
. (5.108)
Comparing this with equation (5.106), we obtain

2
=
1
4

kc
ω
p0

4
ω
p0
ω
/
0

2
ph
. (5.109)
Noting that, for the electron plasma waves resonant with the photon beam,
we always have k
2
c
2
.
2
p0
. ω
2
, we can also write the approximate expression

2
.
ω
p0

/
0

2
ph
. (5.110)
We see that the effective photon plasma frequency is always significantly
larger than the photon plasma frequency
ph
as defined by equation (5.107). But,
the fact that this quantity
ph
appears in the photon susceptibility shows that the
concept of the photon equivalent charge Q
ph
is already implicit in the photon
susceptibility term and in the photon Landau damping. If the photon effective
charge did not exist, the photon Landau damping would not be possible. We can
then conclude that the high-frequency photons behave in a plasma like any other
charged particle.
Coming back to the dispersion relation (5.105), we can assume the resonant
condition ω
p0
= ku
0
, and we can write ω = ω
p0
÷ iµ. We can then obtain the
growth rate for the photon beam plasma instability as
µ =

3
2
ω
p0

2

2
p0

1/3
. (5.111)
We notice that this growth rate is proportional to the power 1/3 of the density
of photons in the beam:
µ ∝ N
1/3
0
. (5.112)
Equivalent dipole in an optical fibre 115
This instability is comparable to the usual beam plasma instability produced
by the interaction of a monoenergetic electron beam (with mean velocity u
0
) with
a plasma, for which the growth rate is also proportional to the density of particles
in the beam to the power 1/3. This shows again that the photons behave, in this
respect, just like electrons.
The photon beam plasma instability should also be compared with the for-
ward Raman scattering [50], which occurs when a photon beam (with a well-
defined field phase) interacts with a plasma. Electron plasma waves are also
produced here, but the physical mechanism is completely different because it
implies that the usual resonant conditions for wave decay are satisfied: ω
/
1
=
ω
/
0
±ω and k
/
1
= k
/
0
±k, where ω
/
1
and k
/
1
are the frequency and wavenumber of
the scattered photons. This corresponds to the well-known energy and momentum
conservation rules, which are clearly distinct from the resonance condition ω =
ku
0
, appling to our photon beam plasma instability.
Furthermore, the maximum growth rate for the forward Raman scattering is
determined by
µ .
ω
2
p0
2


/
0
v
os
c
(5.113)
where v
os
= eE
0
/mω
/
0
, and E
0
is the amplitude of the electric field associated
with the incident photon beam.
We see that this new growth rate is proportional to N
1/2
0
, in contrast with the
result of equation (5.112), thus showing that the photon beam plasma instability
is a new kind of instability, which is clearly distinct from the usual Raman decay
instability.
5.6 Equivalent dipole in an optical fibre
We have shown that it is possible to define an equivalent photon charge in a
plasma, which is a different way of describing the ponderomotive force or ra-
diation pressure effects. Here we examine a similar concept for non-ionized
dielectric media.
In particular, we examine a laser pulse propagation along an optical fibre
having a second-order nonlinearity and show that an equivalent electric dipole
can be derived [70]. The difference with respect to the plasma case is related to
the absence of free electrons in this medium.
From Maxwell’s equations, in the absence of charge and current distribu-
tions, we can establish the following equation for the propagating electric field:

2
¯
E −∇

∇ ·
¯
E


1
c
2

2
¯
E
∂t
2
= µ
0

2
¯
P
∂t
2
. (5.114)
116 Photon equivalent charge
The polarization vector is
¯
P =
¯
P
L
÷
¯
P
NL
, with the linear part defined by
¯
P
L
(t ) =
0


0
χ
(1)
(τ) ·
¯
E(t −τ) dτ. (5.115)
The nonlinear part of the polarization vector, if we neglect dispersion effects,
can be written as
¯
P
NL
(t ) =
0
χ
(2)
·
¯
E(t )
¯
E(t ) ÷
0
χ
(3)
·
¯
E(t )
¯
E(t )
¯
E(t ). (5.116)
We now assume, as we already did for the plasma case, that the electric field
can be divided into two terms: the dominant field associated with the propagating
pulse
¯
E
p
, and a secondary perturbation or radiation field,
¯
E
r
, such that [E
r
[ <
[E
p
[:
¯
E =
¯
E
p
÷
¯
E
r
. (5.117)
To the lowest order in the perturbation field
¯
E
r
, we can write from equa-
tions (5.114–5.117)

2
¯
E
p
−∇

∇ ·
¯
E
p


1
c
2

2
∂t
2
¸
¯
E
p
÷


0
χ
(1)
(τ) ·
¯
E(t −τ) dτ
¸
=
1
c
2

2
∂t
2
χ
(3)
· [E
p
[
2
¯
E
p
. (5.118)
From this equation we can obtain linear (if χ
(3)
= 0) or nonlinear (if χ
(3)
,=
0) wave pulse solutions. In general, for propagation along an optical fibre, we can
write both types of solution in the general form
¯
E
p
(¯ r, t ) =
ˆ a
p
2
¸
R(¯ r

)A(z, t ) e
i(kz−ωt )
÷c.c.
¸
. (5.119)
This field is assumed to propagate along the fibre axial directon z with neg-
ligible pulse shape deformation, and with a group velocity v
g
.
To the first order in the radiation field
¯
E
r
, and assuming for convenience that
χ
(2)
is of the same order of
¯
E
r
, we get from equations (5.114–5.117)

2
¯
E
r
−∇

∇ ·
¯
E
r


1
c
2

2
∂t
2
¸
¯
E
r
÷


0
χ
(1)
·
¯
E
r
(t −τ) dτ
¸
= µ
0

∂t
¯
J
r
(5.120)
where
¯
J
r
is the nonlinear polarization current, defined by
¯
J
r
=
0

∂t
¸
χ
(2)
·
¯
E
p
¯
E
p
÷χ
(3)
· [E
p
[
2
ˆ a
p
ˆ a
p
¯
E
r
¸
. (5.121)
Equivalent dipole in an optical fibre 117
Using equation (5.119), we can see that
¯
E
p
(¯ r, t )
¯
E
p
(¯ r, t ) =
1
4
ˆ a
p
ˆ a
p
¸
R
2
(¯ r

)A
2
(z, t ) e
2i(kz−ωt )
÷[R(¯ r

)[
2
[a(z, t )[
2
÷c.c.
¸
. (5.122)
The first term oscillates in time as exp(2iωt ) and is associated with harmonic
generation. Its physical content is well understood and, for this reason, we can
ignore it here.
The term in χ
(3)
describes the scattering of a probe radiation field by the
main pulse and can also be ignored because it does not play any role in the effect
to be discussed here. For simplicity, we will use χ
(2)
· ˆ a
p
ˆ a
p
= χ
(2)
ˆ e in the
remaining term.
The nonlinear polarization current
¯
J
r
is then reduced to
¯
J
r
=

0
2
χ
(2)

∂t
W(¯ r, t )ˆ e (5.123)
where we have used
W(¯ r, t ) = [R(¯ r

)[
2
[ A(z, t )[
2
. (5.124)
It is obvious that, if the main pulse moves along the fibre with no significant
change in its envelope shape, we can write
W(¯ r, t ) ≡ W(¯ r

, z −v
g
t ). (5.125)
This means that the time derivative appearing in equation (5.123) can be
replaced by a space derivative

∂t
W = −v
g

∂z
W. (5.126)
As a result of this replacement, we can rewrite the nonlinear polarization
current as the product of a charge with a velocity
¯
J
r
= v
g
Q(¯ r, t )ˆ e. (5.127)
This quantity Q(¯ r, t ) can be called the equivalent charge distribution of the
main pulse, and it is determined by
Q(¯ r, t ) = −

0
2
χ
(2)

∂z
W(¯ r

, z −v
g
t ). (5.128)
This is clearly a dipole charge distribution because the space derivative re-
verses its sign when we move from the front to the rear of the pulse. This clearly
contrasts with the plasma case where, due to the existence of free electrons which
are pushed away by the electromagnetic pulse, the equivalent charge distribution
reduces to a negative monopole charge.
118 Photon equivalent charge
An interesting feature is that the total equivalent charge here is equal to zero,
as can be seen by integrating the distribution (5.128) along the pulse distribution:


−∞
Q(z, t ) dz = −

0
2
χ
(2)


∂z
W dz = −

0
2
χ
(2)
[W]

−∞
= 0. (5.129)
This means that we have here total charge conservation, which is physically
quite reassuring. Furthermore, this property is independent of the actual shape of
the pulse.
Let us consider the interesting limiting case of a rectangular pulse: the
equivalent charge distribution reduces to two point charges, one negative and the
other positive, corresponding to
Q(¯ r, t ) = −Q
eq
(¯ r

)
¸
δ

z −
L
2
−v
g
t

−δ

z ÷
L
2
−v
g
t
¸
(5.130)
where
Q
eq
=

2
χ
(2)
W
0
(¯ r

) (5.131)
for a pulse of length L and maximum intensity W
0
.
Replacing equation (5.127) in equation (5.120) we can determine the sec-
ondary field
¯
E
r
created by this moving dipole:

2
¯
E
r
−∇

∇ ·
¯
E
r


1
c
2

2
∂t
2
¸
¯
E
r
÷


0
χ
(1)
·
¯
E
r
(t −τ) dτ
¸
=
1
c
2

∂t
v
g
Q(¯ r, t )ˆ e. (5.132)
It should be noticed that this secondary field is not always a radiation field.
For instance, in a homogeneous medium, where both the main pulse velocity v
g
and the charge amplitude Q
eq
are constant, it will reduce to a near field. But even
in this case we can imagine an external loop where the existence of such a field
can eventually be detected.
Radiation will typically occur in two distinct situations: first, when the group
velocity v
g
changes due to a space dependence of the linear susceptibility of
the medium χ
(1)
; second, when the charge amplitude changes due to the space
variation of the nonlinear susceptibility χ
(2)
.
These two radiation mechanisms are clearly distinct from those usually con-
sidered in the literature [2], which are mainly concerned with soliton propagation
in inhomogeneous but centro-symmetric media, where the second-order nonlin-
earities are forbidden (χ
(2)
= 0). In this case, the source of radiation is a
linear term of the form δχ
(1)
E
p
, where δχ
(1)
is the perturbation of the linear
susceptibility due to the inhomogeneities of the medium. In contrast with our
present radiation mechanism, which is clearly associated with an acceleration of
the moving dipole, these more common aspects of secondary radiation can be
Equivalent dipole in an optical fibre 119
Q

(-)
(+)
(a)
eq
(z - v

t)
g
(b)
Q
eq
(z - v

t)
g
(-)
(+)
Figure 5.1. Equivalent dipole distribution for (a) a sech laser pulse, and (b) a nearly
rectangular pulse.
described as scattering of the lowest order pulse field by the local inhomogeneities
of the medium.
Let us study equation (5.132) in more detail, and introduce a time Fourier
transformation of the field as
¯
E
r
(¯ r, t ) =


−∞
¯
E
ω
(¯ r) e
−iωt


. (5.133)
120 Photon equivalent charge
The Fourier components
¯
E
ω
will be determined by the equation
¸

2
÷
ω
2
c
2
(ω)

¯
E
ω
−∇

∇ ·
¯
E
ω

= i
ω
c
ˆ e


−∞
v
g
(z)Q(¯ r

, z, t ) e
iωt
dt
(5.134)
where (ω) is the linear dielectric function of the medium.
Let us retain our attention on radiated transverse fields, such that ∇·
¯
E
ω
= 0.
Furthermore, in a cylindric geometry, we can use
¯
E
ω
(¯ r) = R
ω
(¯ r

)A
ω
(z)ˆ e
ω
. (5.135)
Replacing it in the above equation, we get
¸
A
ω

2

R
ω
÷ R
ω

2
∂z
2
A
ω

÷
ω
2
c
2
(ω)R
ω
A
ω
= i
ω
c
(ˆ e · ˆ e
ω
)v
g
(z)


−∞
Q(¯ r, t ) e
iωt
dt. (5.136)
This equation can be solved by considering first the corresponding homoge-
neous equation

2

R
ω
÷

2
∂z
2
A
ω
÷
ω
2
c
2
(ω) = 0. (5.137)
If we assume that the radial field amplitude R
ω
is determined by the equation

2

R
ω
= −k
2

R
ω
(5.138)
where k

is the perpendicular wavenumber, the axial field amplitude A
ω
will have
the solution
A
ω
= A
±
e
±ikz
. (5.139)
The axial wavenumber k is determined by the relation
k
2
=
ω
2
c
2
(ω) −k
2

. (5.140)
Returning to equation (5.136), we can write the inhomogeneous equation for
the axial field amplitude A
ω
in the form

2
A
ω
∂z
2
÷k
2
A
ω
= i
ω
c
(ˆ e · ˆ e
ω
)v
g
(z)


−∞
q(z, t ) e
iωt
dt (5.141)
where q(z, t ) is the axial equivalent charge distribution, determined by
q(z, t ) =

R

ω
(¯ r

)Q(¯ r

, z, t ) d¯ r

[R
ω
(¯ r

)[
2
d¯ r

. (5.142)
Equivalent dipole in an optical fibre 121
We now use the well-known method of the variation of parameters to solve
equation (5.141). For forward propagation, we obtain
A
ω
(z)
÷
=
i
w
ω
c
(ˆ e · ˆ e
ω
) e
ikz

dt e
iωt

z
z
0
dz
/
v
g
(z
/
)g(z, t ) e
−ikz
(5.143)
where the Wronskian is w = 2ik.
To be more specific, let us assume that we have an axial modulation of the
nonlinear properties of the optical fibre, which can be described by
v
g
(z) = v
0
= const, χ
(2)
= χ
20
[1 ÷δ cos(k
0
z)] . (5.144)
This means that we can replace, in the above solution, the axial equivalent
charge by
q(z, t ) = q
0
(z, t )δ cos k
0
z. (5.145)
The forward field solution becomes
A
ω
(z)
÷
= A
0
e
ikz

dt e
iωt

z
z
0
dz
/
q
0
(z
/
, t ) cos(k
0
z) e
−ikz
(5.146)
with an amplitude
A
0
=
ω
2kc
(ˆ e · ˆ e
ω
)v
0
δ. (5.147)
In the particular case of a rectangular pulse shape, as described by equa-
tion (5.130), we can write
q
0
(z, t ) = q
0
¸
δ

z −
L
2
−v
0
t

−δ

z ÷
L
2
−v
0
t
¸
. (5.148)
We can now easily integrate equation (5.146) in z, for a very long fibre, and
we obtain
A
ω
(z) = A
0
q
0
e
ikz

e
iωt
¸
cos

k
0
L
2
÷k
0
v
0
t

e
−ik(L/2)−ikv
0
t
−cos

k
0
L
2
−k
0
v
0
t

e
ik(L/2)−ikv
0
t
¸
dt. (5.149)
From this equation it can then be concluded that two distinct frequencies are
radiated by the main pulse when it propagates in a modulated optical fibre:
A
ω
(z) = 2πi A
0
q
0
¸
sin
¸
(k
0
−k)
L
2
¸
δ(ω −ω
1
)
÷sin
¸
(k
0
÷k)
L
2
¸
δ(ω −ω
2
)
¸
e
ikz
. (5.150)
The values of the two frequencies are determined by
ω
1
= v
0
(k −k
0
), ω
2
= v
0
(k ÷k
0
). (5.151)
122 Photon equivalent charge
When the wavelength of the radiated field is nearly equal to the characteristic
wavelength of the fibre inhomogeneity, k . k
0
, we have ω
1
. 0 and ω
2
. 2v
0
k
0
.
This means that, in this case, ω
2
is a kind of second harmonic.
Finally, we note that, for the more general situation of a pulse with an
arbitrary shape, the solution (5.150) is replaced by
A
ω
(z)
÷
= π A
0
e
ikz

q
0
(ξ)
¸
e
i(k
0
−k)ξ
δ(ω −ω
1
)
÷e
−i(k
0
−k)ξ
δ(ω −ω
2
)
¸
dξ (5.152)
with ξ = x −v
0
t .
This radiation process is a direct consequence of the existence of an equiv-
alent dipole charge of the optical pulse propagating in an optical medium with
second-order nonlinearities. As already noticed for the plasma case treated in
section 5.2, it can be considered as the analogue of the electron ondulator, or
another version of the photon ondulator.
Chapter 6
Full wave theory
In previous chapters we were able to describe a large number of different physical
phenomena using the Hamiltonian approach to photon dynamics, either in its
simple form of single photon trajectories or in its more elaborated version of the
photon kinetic theory. However, the geometric optics approximation associated
with this elegant and powerful approach is not capable of explaining such basic
and sometimes important phenomena as partial reflection and mode coupling.
A full wave description of the electromagnetic radiation is therefore neces-
sary, and it will be the subject of the present and the next chapters. Several papers
have been devoted to this problem [26, 27, 78, 118, 123].
In chapter 3 we showed that an effect similar to the usual refraction, which
we called time refraction, could take place. Here we will see that a new effect,
which, in the same spirit, can be called time reflection, can also take place.
6.1 Space and time reflection
For comparison with the cases of time-varying media, we start first by reminding
ourselves of the well-known derivation of the Fresnel formulae, by considering
reflection and transmission of an incident electromagnetic wave at the boundary
between two stationary dielectric media.
6.1.1 Reflection and refraction
Let us assume two different media, with dielectric constants
1
and
2
, with a
sharp boundary at the plane x = 0. If a plane wave with frequency ω and
wavevector
¯
k
i
propagates along the x-axis and interacts with this boundary, the
total electric field will be determined by
¯
E(x, t ) =
¸
¯
E
i
exp i(k
i
x −ωt ) ÷
¯
E
r
exp i(k
r
x −ωt ), (x < 0)
¯
E
t
exp i(k
t
x −ωt ), (x > 0).
(6.1)
123
124 Full wave theory
This field has three different terms, corresponding to the incident, the re-
flected and the transmitted waves. These waves have to satisfy the dispersion
relations of both media, which implies that
k
i
=
ω
c
n
1
, k
r
= −k
i
, k
t
= k
i
n
2
n
1
(6.2)
with the refractive indices n
1,2
=

1,2
.
In order to establish a relation between the field amplitudes
¯
E
i
,
¯
E
r
and
¯
E
t
,
we have to state the associated boundary conditions. These are determined by
Maxwell’s equations. In the absence of charge and current distributions in the
media (ρ = 0,
¯
J = 0), these equations are

¯
E = −

¯
B
∂t
, ∇
¯
H =

¯
D
∂t
(6.3)
and
∇ ·
¯
D = 0, ∇ ·
¯
B = 0. (6.4)
It is well known that, in order to satisfy the first pair of equations, the
components of
¯
E and
¯
H tangent to the boundary have to be continuous across
this boundary:
(
¯
E
1

¯
E
2
) ¯ e
x
= 0, (
¯
H
1

¯
H
2
) ¯ e
x
= 0. (6.5)
From equations (6.3) we realize that, for plane waves propagating in non-
magnetic media, we can write
¯
B ≡ µ
0
¯
H =
¯
k
¯
E
ω
. (6.6)
Assuming perpendicular polarization (
¯
E·¯ e
x
= 0), and using the wavenumber
relations (6.2), we can write the boundary conditions (6.5) in the form
E
i
÷ E
r
= E
t
, E
i
− E
r
= E
t
n
2
n
1
. (6.7)
From this we derive the well-known Fresnel formulae for normal wave inci-
dence
T ≡
E
t
E
i
=
2n
1
n
1
÷n
2
, R ≡
E
r
E
i
=
n
1
−n
2
n
1
÷n
2
. (6.8)
These expressions verify the simple relation
1 ÷ R = T. (6.9)
These results are extremely well known and we should not insist on their
physical significance. We should instead use them as a reference guide for the
less understood time-varying processes.
Space and time reflection 125
6.1.2 Time reflection
Let us turn to the opposite situation of a time discontinuity (or a time boundary)
occurring in a homogeneous and infinite medium. For times t < 0, the medium
will have a refractive index n
1
everywhere and, for t = 0 the refractive index will
be suddenly shifted to a new value n
2
.
The electric field associated with a plane wave propagating in the medium,
with the initial frequency ω
i
, will be described by an expression similar to equa-
tion (6.1)
¯
E(x, t ) =
¸
¯
E
i
exp i(kx −ω
i
t ), (t < 0)
¯
E
r
exp i(kx −ω
r
t ) ÷
¯
E
t
exp i(kx −ω
t
t ), (t > 0).
(6.10)
As before, the three distinct plane waves have to satisfy the linear dispersion
relations of the medium, which implies that
ω
i
=
kc
n
1
, ω
r
= −ω
t
, ω
t
= ω
i
n
1
n
2
. (6.11)
We should take notice of the symmetry between these relations and those
of equation (6.2). It can easily be understood that the negative frequency mode
corresponds indeed to a reflected wave because, in order to describe a real wave,
we have to add to fields (6.1, 6.10) their complex conjugates. This operation
will then lead to the appearance of a plane wave with positive frequency and
propagating in the opposite direction with respect to that of the initial wave:
¯
E

r
exp[−i(kx ÷[ω
r
[t )].
In order to determine the relative amplitudes of the transmitted and the re-
flected waves, we have to establish the time continuity conditions for the electro-
magnetic field. We can see from Maxwell’s equations (6.3, 6.4) that they only
contain the time derivatives of the fields
¯
D and
¯
B. These equations can only stay
valid for all times if these two fields are continuously varying in time.
The time continuity conditions are then given by
¯
D(t = 0

) =
¯
D(t = 0
÷
),
¯
B(t = 0

) =
¯
B(t = 0
÷
). (6.12)
Expressing the fields
¯
D and
¯
B in terms of the electric field
¯
E for the three
distinct waves present in equation (6.10), we can rewrite these continuity condi-
tions as
n
2
1
E
i
= n
2
2
(E
r
÷ E
t
), E
i
= −E
r
÷ E
t
n
2
n
1
. (6.13)
This means that the well-known continuity conditions (6.7), valid for a sharp
spatial boundary in the dielectric properties of the propagating medium, are re-
placed by these new conditions in the case of a sharp time boundary. The explicit
result for the reflected and the transmitted wave amplitudes is
R ≡
E
r
E
i
=
1
2

n
2
1
n
2
2
−1

, T ≡
E
t
E
i
=
1
2

n
2
1
n
2
2
÷1

. (6.14)
126 Full wave theory
A simple relation between R and T can also be established here:
T ÷ R =
n
2
1
n
2
2
. (6.15)
These equations can be called the Fresnel formulae for time reflection. They
are very similar to the usual Fresnel formulae (6.8), but they also show quite
surprising differences.
The first surprise is that a time discontinuity can lead to a reflected wave,
that is, a wave propagating in the opposite direction with respect to the initial
wave. The second is that the amplitudes of both the reflected and the transmitted
waves can be very large if the value of the new refractive index n
2
is very low.
This can be obtained, for instance, by plasma creation (or flash ionization) with a
nearly resonant plasma frequency ω
p
∼ ω
t
. This means that we could expect to
produce electromagnetic energy out of nearly nothing, in exactly the same way as
photons can be created from a vacuum in quantum models. In this case, of course,
the amount of energy of the waves created by time reflection would have to be
furnished by the external agent responsible for the sudden ionization process.
We should however notice that the limit of n
2
= 0 cannot be properly
described by the above model because it does not take into account the existence
of the electrostatic mode, which can be excited in the medium when its dielectric
constant is equal to zero. As we know, for a plasma, this is the electron plasma
wave. The field of this electrostatic wave would have to be added to the above
continuity conditions. Furthermore, for the plasma case, there are important
qualitative differences related, not only to dispersion, but also to the appearance
of a magnetic mode, as will be seen later.
A large amount of work has been devoted to the problem of wave trans-
formation due to a sudden creation of a plasma medium, in several different
configurations including the presence of a static magnetic field [44]. However,
the concept of time refraction was never explored.
6.2 Generalized Fresnel formulae
We can generalize the above treatement of independent space and time disconti-
nuities by considering the case of a sharp boundary of the refractive index moving
in space with a constant velocity ¯ u = u¯ e
x
. An earlier discussion of this problem
can be found in reference [110].
Let us assume an incident wave of the form
¯
E
i
exp i(k
i
x −ω
i
t ) (6.16)
with k
i
= (ω
i
/c)n
1
. In the reference frame moving with the boundary, the same
wave will be represented by
¯
E
/
i
exp i(k
/
i
x
/
−ω
/
i
t
/
). (6.17)
Generalized Fresnel formulae 127
Let us invoke the Lorentz transformations from the moving frame to the rest
frame:
x = γ (x
/
÷ut
/
), t = γ

t
/
÷
β
c
x
/

(6.18)
with β = u/c and γ
−2
= (1 −β
2
).
We know that the field phase in equations (6.16, 6.17) is a relativistic invari-
ant, which means that k
i
x − ω
i
t = k
/
i
x
/
− ω
/
i
t
/
. Replacing equations (6.18) in
this equality, and separately equating the terms containing x
/
and t
/
, we obtain the
appropriate Lorentz transformations for the frequency and the wavenumber:
ω
/
= γ ω
i
(1 −βn
1
), k
/
i
= γ k
i

1 −
β
n
1

. (6.19)
From this we get the dispersion relation in the moving frame
k
/
i
=
ω
/
c
n
/
1
(6.20)
with
n
/
1
=
n
1
−β
1 −βn
1
. (6.21)
This equation relates the values of the refractive index as seen in the moving
and in the rest frames. It is a very simple and useful result which, surprisingly,
cannot easily be found in the literature.
Of course, this transformation is only valid for a non-dispersive medium.
The case of a plasma will be considered below. We are now able to repeat, in
the moving frame, the above derivation of the Fresnel formulae (6.8). To do this
we assume that the total electric field in this frame is formally identical to that of
equation (6.1), i.e.
¯
E
/
(x
/
, t
/
) =
¸
¯
E
/
i
exp i(k
/
i
x
/
−ω
/
t
/
) ÷
¯
E
/
r
exp i(k
/
r
x
/
−ω
/
t
/
), (x
/
< 0)
¯
E
/
t
exp i(k
/
t
x
/
−ω
/
t
/
), (x
/
> 0).
(6.22)
In analogy with equation (6.2), the three different wavenumbers are related
by k
/
r
= −k
/
i
and k
/
t
= k
/
i
(n
/
2
/n
/
1
). This leads to the result
E
/
r
E
/
i
=
1 −w
1 ÷w
,
E
/
t
E
/
i
=
2
1 ÷w
(6.23)
with w = (k
/
t
/k
/
i
) = (n
/
2
/n
/
1
).
Let us see how such a result can be extended to the plasma case. To do that
we assume that medium 1 is just the vacuum region and medium 2 is a plasma
moving in a vacuum with velocity u. This means that the particles contained in
the plasma medium (electrons and ions) all move with an average velocity equal
to u.
128 Full wave theory
It can easily be realized that the plasma frequency is a relativistic invariant
ω
2
p
= k
2
c
2
−ω
2
= k
/2
c
2
−ω
/2
. (6.24)
This means that, for the plasma case, equation (6.21) is not valid and we
have
w =

k
/
t
c
ω
/

=

1 −
ω
2
p
ω
/2
. (6.25)
We clearly see that w = 0 defines a cut-off condition and that total reflection
will occur for
ω
2
p
= ω
2
1 ÷β
1 −β
. (6.26)
This is in agreement with our single photon model of chapter 3. The trans-
formations (6.23) stay valid for the moving plasma case if we assume that w is
determined by equation (6.25).
We can now make a Lorentz transformation of the electric fields back to the
rest frame. For the field of the incident wave, we have
E
/
i
= γ (E
i
−uB
i
) = γ E
i
(1 −βn
1
). (6.27)
For the reflected wave β is replaced by −β and for the transmitted wave n
1
is replaced by n
2
. From equations (6.23) we then get
R ≡
E
r
E
i
=
1 −βn
1
1 ÷βn
1
1 −w
1 ÷w
, T ≡
E
t
E
i
=
1 −βn
1
1 −βn
1
2
1 ÷w
. (6.28)
We can easily find that
1 ÷ R = T
1 −βn
2
1 −βn
1
. (6.29)
These equations can be called the generalized Fresnel formulae for moving
dielectric perturbations. We notice that for a stationary boundary, β = 0, these
equations reduce to (6.8, 6.9), as expected.
In the limit of an infinite velocity, β → ∞, which would be the case for a
time discontinuity, equation (6.29) also reduces to (6.15). However, in this limit,
equations (6.28) do not reduce to the Fresnel formulae for time reflection (6.14).
This apparent incongruence is due to the fact that the calculations presented in
this section are based on Lorentz transformations and therefore they are not valid
for β > 1.
6.3 Magnetic mode
An interesting new effect occurs when the moving discontinuity of the refractive
index is associated with an ionization front. We are refering to the excitation of
Magnetic mode 129
+
-
+
-
moving
plasma
vacuum
( a)
+
-
+
-
plasma
neutral

gas
moving
boundary
plasma
( b)
Figure 6.1. Comparison between the two cases: (a) a moving plasma and (b) a moving
ionization front.
a purely magnetic mode [53, 97]. This will be examined here in some detail and
will be compared with the (at first sight) generic results of the previous section.
Let us study the problem in the frame of the moving ionization front. In this
frame, the atoms of the neutral gas region are seen to flow with a velocity −u and
to disappear at the front discontinuity x
/
= 0.
On the other hand, the plasma electrons (assuming that they are created with
zero kinetic velocity) will appear to flow away from that front with exactly the
same speed. The electron equation of motion can be written, in the same reference
frame, as
d¯ v
dt
/
= −
e

(
¯
E
/
−u¯ e
x

¯
B
/
) (6.30)
with
d
dt
/
=

∂t
/
−u

∂x
/
. (6.31)
Notice that, if instead of an ionization front we were considering a plasma
moving with the same velocity, we would have to use u = 0 in these equations,
because the plasma electrons would be moving with the front. The electric and
magnetic fields
¯
E
/
and
¯
B
/
are determined by Maxwell’s equations (6.3), which
can now be written as

/

¯
E
/
= −µ
0

¯
H
/
∂t
/
, ∇
/

¯
H
/
=
¯
J
/
÷
0

¯
E
/
∂t
/
(6.32)
where the electron current is simply
¯
J
/
= −en
0
γ ¯ v
/
(6.33)
130 Full wave theory
and n
0
is the electron mean density in the rest frame.
We can now solve these equations by assuming that the fields evolve in space
and time according to exp i(k
/
x
/
−ω
/
t
/
). We obtain
i(ω
/
÷k
/
u)¯ v
/
=
e

¸
¯
E
/

u
ω
/
¯ e
x
(
¯
k
/

¯
E
/
)
¸
(6.34)
i
¸
¯
k
/
(
¯
k
/

¯
E
/
) ÷
ω
/2
c
2

¯
E
/
= −en
0
γ µ
0
ω
/
¯ v
/
. (6.35)
Notice that we cannot divide the first of these equations by (ω
/
÷k
/
u) because
such a factor can eventually be equal to zero, as will be shown below. From these
two equations we can easily derive the following dispersion relation, valid for
transverse modes (
¯
k
/
·
¯
E
/
= 0) in the plasma region:
(k
/2
c
2
÷ω
2
p
−ω
/2
)(ω
/
÷k
/
u) = 0. (6.36)
This shows that, for a given frequency, two distinct modes are possible:
k
/
=
ω
/
c

1 −(ω
p

/
)
2
(6.37)
and
k
/
= k
/
m
= −
ω
/
u
. (6.38)
The first mode is the usual transverse electromagnetic mode in a plasma. The
second mode can be called the magnetic mode because its electric field is equal
to zero in the rest frame. This can be seen by noting that, according to Maxwell’s
equations (6.32), we have
E
/
m
=
ω
/
k
/
m
B
/
m
= −uB
/
m
. (6.39)
Using the Lorentz transformation for the electric field we get, in the rest
frame,
E
m
= γ (E
/
m
÷uB
/
m
) = 0. (6.40)
Similarly, it can be shown that its frequency in the rest frame is also equal to
zero:
ω
m
= γ (ω
/
÷uk
/
m
) = 0, k
m
= γ

k
/
m
÷β
ω
/
c

=
ω
c
1 −β
1 ÷β
. (6.41)
This shows that the magnetic mode is a purely magnetostatic perturbation in
the rest frame, with zero frequency but a finite wavelength 2π/k
m
,= 0. From the
Magnetic mode 131
above discussion we can conclude that the total electric field associated with an
ionization front will have four (and not three) distinct waves:
¯
E
/
(x
/
, t
/
) =
¸
¯
E
/
i
exp i(k
/
i
x
/
−ω
/
t
/
) ÷
¯
E
/
r
exp i(k
/
r
x
/
−ω
/
t
/
), (x
/
< 0)
¯
E
/
t
exp i(k
/
t
x
/
−ω
/
t
/
) ÷
¯
E
/
m
exp i(k
/
m
x
/
−ω
/
t
/
), (x
/
> 0).
(6.42)
The continuity conditions for the transverse fields
¯
E
/
and
¯
B
/
can then be
stated as
E
/
i
÷ E
/
r
= E
/
t
÷ E
/
m
, E
/
i
− E
/
r
= E
/
t
k
/
t
k
/
i
− E
/
m
k
/
m
k
/
i
(6.43)
where
k
/
t
k
/
i
= w =

1 −(ω
p

/
)
2
,
k
/
m
k
/
i
= −
1
β
. (6.44)
We need to complement the two continuity conditions (6.43) by another con-
dition because we have here three unknown field amplitudes. This extra condition
is provided by the second of Maxwell’s equations (6.32).
We notice that the current
¯
J
/
is continuous across the front boundary if we
assume that the electrons are created with zero net velocity:
¯
J
/
= ¯ v
/
= 0 for
x
/
= 0. Because the electric field is also continuous, this implies that ∂ B
/
/∂x
/
will also be continuous across the boundary:

∂ B
/
∂x
/

x
/
=0

=

∂ B
/
∂x
/

x
/
=0
÷
. (6.45)
This new continuity condition can be rewritten as
k
/
i
B
/
i
÷k
/
r
B
/
r
= k
/
t
B
/
t
÷k
/
m
B
/
m
. (6.46)
Noting that k
/
r
= k
/
i
, and that B
/
= (k
/
c/ω
/
)E
/
, we can derive from this and
from equations (6.43) the three conditions for the electric field amplitudes:
E
/
i
÷ E
/
r
= E
/
t
÷ E
/
m
E
/
i
− E
/
r
= wE
/
t
÷
E
/
m
β
E
/
i
÷ E
/
r
= w
2
E
/
t
÷
E
/
m
β
2
. (6.47)
This can be solved as
E
/
r
E
/
i
=
1 ÷β
1 −β
1 −w
1 ÷w
E
/
t
E
/
i
=
2
1 ÷βw
1 ÷β
1 ÷w
E
/
m
E
/
i
=

2
(1 −w)
(1 −β)(1 ÷βw)
. (6.48)
132 Full wave theory
This result has to be compared with equations (6.23), which are valid in the
absence of the magnetic mode. The difference between these two results can
better be understood if we calculate the energy of the reflected wave.
In order to perform this calculation we first notice that the energy of an
electromagnetic wave is (apart from the constant
¨
h, taken here to be equal to 1) the
product of the wave frequency with the number of photons: W = ωN. Because
the number N is a relativistic invariant, the energy will be Lorentz transformed in
a way that is the reverse of the Lorentz transformation for the frequency.
We can then write, for the number of photons associated with both the inci-
dent and the reflected wave
N
i
=
W
i
ω
=
W
/
i
ω
/
, N
r
=
W
r
ω
r
=
W
/
r
ω
/
. (6.49)
This means that, for both cases (absence and presence of the magnetic mode),
we can write the fraction of reflected energy as
W
r
W
i
=
ω
r
ω
W
/
r
W
/
i
=
ω
r
ω

E
/
r
E
/
i

2
. (6.50)
Using the Lorentz transformation for the the frequencies, we get
ω
/
=
ω
γ (1 ÷β)
=
ω
r
γ (1 −β)
(6.51)
which leads to
W
r
W
i
=
1 −β
1 ÷β

E
/
r
E
/
i

2
. (6.52)
In the absence of the magnetic mode (in the case of a moving plasma, con-
sidered in the previous section), and assuming total reflection (w = 0), we can
obtain from equations (6.23, 6.52)
W
r
W
i
=
1 −β
1 ÷β
=
ω
r
ω
. (6.53)
This means that, for a counter-propagating ionization front (β < 0), we have
a significant gain in the energy of the reflected wave. From equation (6.49) we
can also see that the number of photons is conserved, N
r
= N
i
, which means that
all the incident photons are reflected, but with a different frequency.
In contrast, when the magnetic mode is present (in the case of an ionization
front, considered in this section), and assuming once again that total reflection is
taking place (w = 0), we get from equation (6.48)
W
r
W
i
=
1 ÷β
1 −β
=
ω
ω
r
. (6.54)
Dark source 133
This means that, for β < 0, we now have a significant decrease in the energy
of the reflected wave. From equation (6.49) we can also see that the number of
photons is not conserved upon reflection, even if it is a total reflection:
N
r
N
i
=

ω
ω
r

2
. (6.55)
The missing photons are, in this case, converted into a static magnetic field
(zero energy photons), and we could be talking about a photon freezing or photon
condensation effect. In the next section we will consider a situation which, in
some respects, can be considered as the reverse of this photon freezing.
6.4 Dark source
Let us assume a static electric field, applied to a region of space containing a
neutral gas. No net current is produced by this field, as long as the atoms or
molecules of the gas stay in the neutral state. But, if an ionization front propagates
across that region, the static electric field will accelerate the free electrons created
by the front, and the resulting space- and time-varying currents will eventually
become a source of electromagnetic radiation.
To be specific, we first assume a sinusoidal electrostatic field, described by
¯
E
0
(¯ r) =
¯
E
0
cos(k
0
x) =
¯
E
0
2

e
ik
0
x
÷c.c.

,
¯
B
0
(¯ r) = 0. (6.56)
Let us also assume that the ionization front moves with velocity ¯ u along the
x-axis, but in the negative direction, and that the electrostatic field is perpendicular
to this velocity:
¯
E
0
= E
0
¯ e
y
, ¯ u = −u¯ e
x
. (6.57)
Assuming that the ion motion is irrelevant to the fast processes associated
with high-frequency wave radiation, we can write the electron current in the
plasma region as
¯
J = −en
0
H(x ÷ut )¯ v (6.58)
where the Heaviside function H(x ÷ ut ) represents the sharp boundary between
the neutral and the plasma regions, and the electron velocity ¯ v is determined by
the linearized equation of motion
∂ ¯ v
∂t
= −
e
m
¸
¯
E
0
(¯ r) ÷
¯
E
¸
. (6.59)
Here,
¯
E represents the radiation field. This field is determined by the wave
equation
134 Full wave theory

2
¯
E −
1
c
2

2
¯
E
∂t
2
= µ
0

¯
J
∂t
. (6.60)
Taking the time derivative of equation (6.58) and assuming that the electrons
are created at the plasma boundary x = ut with no kinetic energy, we obtain

¯
J
∂t
= −en
0
uδ(x ÷ut )¯ v −en
0
H(x ÷ut )
∂ ¯ v
∂t
=
e
2
n
0
m
H(x ÷ut )
¸
¯
E
0
(¯ r) ÷
¯
E
¸
. (6.61)
If we want to calculate the field radiated in the plasma region x > ut and
propagating in the forward direction, along the x-axis and in the positive direction,
we can write from these two equations


2
∂x
2

1
c
2

2
∂t
2

ω
2
p
c
2

¯
E =
ω
2
p
c
2
H(x ÷ut )
¯
E
0
(¯ r

, x). (6.62)
In order to solve this equation, let us introduce a time Fourier transformation
defined by
¯
E(x, t ) =

¯
E
ω
(x) e
−iωt


. (6.63)
The one-dimensional wave equation becomes

2
∂x
2
¯
E
ω
÷
1
c
2

2
−ω
2
)
¯
E
ω
=
ω
2
p
c
2
¯
E
0
(x)

H(x ÷ut ) e
iωt
dt. (6.64)
The time integral in the source term can easily be solved and we are led to
the following equation:

2
∂x
2
E
ω
÷k
2
E
ω
= i A
¸
exp
¸
i

k
0

ω
u

x
¸
÷exp
¸
−i

k
0
÷
ω
u

x
¸¸
(6.65)
with a wavenumber k determined by
k
2
c
2
= ω
2
−ω
2
p
(6.66)
and a source term amplitude A where
A =
1
2ωu
ω
2
p
c
2
(
¯
E
0
· ˆ a). (6.67)
Here we have used the unit polarization vector ˆ a =
¯
E
ω
/[E
ω
[. The solution
of equation (6.65), for forward propagation, is
E
ω
(x) =
A
2k
e
ikx

x
−∞
e
−ikx
/
¸
exp
¸
i

k
0

ω
u

x
/
¸
÷ exp
¸
−i

k
0
÷
ω
u

x
/
¸¸
dx
/
. (6.68)
Dark source 135
The asymptotic value of the radiated field, observed in the plasma side but
very far away from the ionization front (x → ∞), will contain at most two
spectral components:
E
ω
(x) =
π
k
A e
ikx
¸
δ

k −k
0
÷
ω
u

÷δ

k ÷k
0
÷
ω
u
¸
(6.69)
with the allowed wavenumbers determined by
k = −
ω
u
±k
0
. (6.70)
We can see that only the positive sign is allowed here because we are con-
sidering forward propagation (k > 0). This expression also implies that we have
an upper limit for the frequency of the radiation produced in the plasma side:
ω < k
0
u.
The value of ω can be obtained by using this equation in the dispersion
relation (6.66):

k
0

ω
u

2
c
2
= ω
2
−ω
2
p
. (6.71)
The explicit result is
ω = k
0

2
¸
¸
1 −β

1 −
ω
2
p
γ
2
k
2
0
u
2

(6.72)
with β = u/c and γ
2
= (1 −β
2
)
−1
.
This result was first obtained by Mori et al [79], who used a different ap-
proach. The interest of our present approach is that we did not use any Lorentz
transformation on the front frame. Let us consider the case where ω
2
p
< γ
2
k
2
0
u
2
,
which corresponds to a large gamma factor or a small plasma density.
Equation (6.72) then reduces to
ω .
k
0
u
2
÷
ω
2
p
2k
0
u
. (6.73)
This result is very interesting because it shows that a large value of the
radiated field frequency ω can be obtained even with very small values of k
0
,
which means, for a long length scale of the static electric field
¯
E
0
(x), if we use
k
0
u <ω
p
, we obtain a radiation frequency ω . ω
2
p
/2k
0
u ·ω
2
p
.
The result described by the present model can be seen as a special case of
photon acceleration: (virtual) photons initially with zero frequency, and associ-
ated with the static electric field (6.56), are accelerated into high frequencies by
the relativistic ionization front and propagate along the plasma medium in the
forward direction.
136 Full wave theory
The conversion efficiency of the process is stated by the radiation field solu-
tion, equation (6.69):
[E
ω
[
2

π
k
A

2
=
π
2
4k
2

ω
p
c

4
[E
0
[
2
ω
2
u
2
. (6.74)
Let us now consider the field radiated into the neutral gas side, for x < ut ,
and propagating in the backward direction. If we restrict again our attention to
the one-dimensional problem, this field is determined by the wave equation


2
∂x
2

1
c
2

2
∂t
2

¯
E =
ω
2
p
c
2
H(x ÷ut )
¯
E
0
(x). (6.75)
Following the above procedure, we can write the asymptotic solution, valid
far away from the ionization front, for x →−∞, as
E
ω
(x) .
π
k
A e
−ikx
¸
δ

k ÷k
0

ω
u

÷δ

k −k
0

ω
u
¸
. (6.76)
The two possible spectral components are
k =
ω
u
±k
0
. (6.77)
In contrast with the previous case where the resulting radiation was prop-
agating in the plasma side, here the waves propagate in a vacuum and have to
satisfy the vacuum dispersion relation ω = kc. Using this in the above condition,
this leads to
ω =
k
0
u
1 −β
. (6.78)
This means that, for relativistic ionization fronts such that (1−β) ∼ γ
2
·1,
very high frequency radiation fields can eventually be emitted. However, the total
energy associated with this new wave is much smaller than the energy of the
lower frequency emitted into the plasma side, because the value of k appearing in
equation (6.76) is now much larger, typically by a factor of γ
2
.
The apparatus necessary to produce this kind of radiation process can be
based on an array of capacitors, periodically spaced along the x-axis with pe-
riodicity 2π/k
0
, in order to create the static periodic electric field defined in
equation (6.56). Such a configuration can be named a dark source [79] because it
transforms a static field into very high frequency photons propagating in both the
plasma and the vacuum region.
However, other configurations can also be imagined, based on the same
principle, as will be illustrated in the following pages. Let us then generalize
the above procedure, by replacing the static periodic electric field by an arbitrary
electrostatic field:
¯
E
0
(¯ r) =

¯
E
q
(¯ r

) e
iqx
dq

. (6.79)
Dark source 137
The previous case obviously corresponds to
¯
E
q
(¯ r

) = π
¯
E
0
[δ(q −k
0
) ÷δ(q ÷k
0
)] . (6.80)
The one-dimensional wave equation for the time Fourier component of the
radiated field, valid in the plasma region for this case of an arbitrary electrostatic
field, is

2
∂x
2
E
ω
÷
1
c
2

2
−ω
2
p
)E
ω
= i

A
q
e
i(q−ω/u)x
dq

(6.81)
with
A
q
=
ω
u
ω
2
p
c
2

¯
E
q
· ˆ a

. (6.82)
The general solution for radiation in the forward direction is then given by
E
ω
(x) =
1
2k
e
ikix


−∞
dx
/

dq

A
q
e
−i(k−q÷ω/u)x
/
. (6.83)
Asymptotically, for x →∞, we get
E
ω
(x) =
π
k
A
0
e
ikx
(6.84)
where A
0
= A
q
for q = q
0
, and q
0
is determined by
q
0
= k ÷
ω
u
. (6.85)
A similar result could equally well be obtained for radiation emitted in the
backward direction and propagating in the vacuum region.
As an example of an application of this more general formulation of the
problem, let us consider the case of an electric field acting on a very small region
of space, around some point x = x
0
. This could, for instance, be obtained by a
capacitor of infinitesimal width (two pointlike electrodes) located at that position:
¯
E
0
(¯ r) =
¯
E
0
δ(x − x
0
). (6.86)
This can be described by equation (6.79) by using an infinite Fourier spec-
trum with amplitudes
¯
E
q
(¯ r

) = 2π
¯
E
0
e
−iqx
0
. (6.87)
In this case, we could generate an infinitely large spectrum of radiation, in
both the forward and the backward direction. Such a configuration, different
(but very similar in its principle) from the original idea of a dark source using
a capacitor array as described above, could be interesting for the production of
ultra-short (or even sub-cycle) radiation pulses, if the resulting broad spectrum of
radiation was subsequently compressed by some appropriate optical system.
138 Full wave theory
moving
front
plasma
Capacitor array
+ + + + + + +
- - - - - -

-
( a)
+
-
moving
front
plasma
Point-like E field
( b)
Figure 6.2. Examples of dark sources: (a) a periodic E-field produced by a capacitor
array; (b) a pointlike E-field.
Exploring further this idea of producing high-frequency radiation out of a
sharp ionization front moving with relativistic velocity, we can even imagine a
situation where the static electric field is absent. This means that we do not need
to consider any capacitor system acting on the gas medium:
¯
E
0
(¯ r) = 0. But now,
some other ingredient has to be introduced in order to replace this field.
We can, for instance, assume that the density of the background neutral gas
medium is modulated in space. Such a modulation can be obtained, for instance,
by producing a sound wave, or by forcing the gas to flow through a parallel grid
before entering the interaction zone.
When the ionization front moves across the gas, a space modulation of the
electron plasma density will take place. The resulting electron current in the
plasma region left behind the front is now described by
¯
J = −en
0
[1 ÷ cos(k
0
x)]H(x ÷ut )¯ v (6.88)
Dark source 139
where is the amplitude of the density modulation and k
0
determines the period
of the space modulation.
A radiation field
¯
E can eventually result from this configuration, as deter-
mined by the wave equation (6.60). But now the electron velocity ¯ v is determined
by equation (6.59) with
¯
E
0
(¯ r) = 0.
Instead of equation (6.61), we will be led to a Mathieu-type wave equation,
which shows unstable solutions inside some region of the space of parameters. In
this case, a dark source radiation instability will occur. A brief discussion of the
Mathieu equation is postponed to the end of the next chapter.
Chapter 7
Non-stationary processes in a cavity
In this new chapter devoted to the full wave theory of photon acceleration, we will
explore a simple theoretical model for mode coupling inside an electromagnetic
cavity. This is formally analogous to the elementary quantum theory of collisions
and it has the merit of reducing the space–time-varying evolution problem to a
purely time-varying problem.
Moreover, this cavity model can also be adequately applied to several exper-
imental configurations. It was initially developed in references [67, 100].
7.1 Linear mode coupling theory
Let us consider an electromagnetic cavity with metalic walls containing a neutral
gas, which can be ionized by some external agent (for instance, by an intense laser
pulse or by a high voltage applied to adequate electric probes). In quite general
conditions, we can describe the electric field associated with the electromagnetic
modes contained in the cavity by the following equation:


2

1
c
2

2
∂t
2

¯
E = µ
0

¯
J
∂t
. (7.1)
The electric field has to satisfy certain boundary conditions and its value will
be determined by the source term which is described by the electric current
¯
J.
The boundary conditions will imply that the general solution of this equation can
take the form of a superposition of eigenmodes, such that
¯
E(¯ r, t ) =
¸
l
e
l
(t )E
l
(¯ r) (7.2)
where e
l
(t ) are time-dependent amplitudes, and E
l
(¯ r) are the cavity eigenmodes
depending on three quantum numbers labelled by l.
In order to be more specific, let us assume the simplest possible example of a
rectangular cavity with the length of the three sides given by L
x
, L
y
and L
z
. If we
140
Linear mode coupling theory 141
restrict our discussion to transverse electric modes, the fields E
l
(¯ r) are determined
by
E
l x
(¯ r) = E
0
cos(mπx/L
x
) sin(nπy/L
y
) sin( pπz/L
z
),
E
l y
(¯ r) = E
0
m
n
L
y
L
x
sin(mπx/L
x
) cos(nπy/L
y
) sin( pπz/L
z
), (7.3)
E
l z
(¯ r) = 0.
The quantities m, n and p are integers and, in this case, we have l ≡
(m, n, p). The generalization of the present discussion to another type of cavity,
and the inclusion of transverse magnetic modes, is straightforward.
In order to guarantee that the eigenmodes are normed and orthogonal we
define the constant E
0
as
E
0
=

8
V

1 ÷
m
2
n
2
L
2
y
L
2
x

−1/2
(7.4)
where we have introduced the cavity volume V = L
x
L
y
L
z
. This choice for E
0
allows us to write the orthonormality condition

V
E
l
(¯ r) · E
l
/ (¯ r) d¯ r = δ
ll
/ . (7.5)
Let us now turn to the electric current appearing in equation (7.1). If we
make some simple and plausible assumptions for the gas ionization process, this
source term can be written as a function of the cavity electric field
¯
E(¯ r, t ) [7].
We first notice that the plasma created out of the neutral gas can be seen as
containing an infinity of electronic species, corresponding to the electrons created
at different times:
¯
J = −e
¸
i
n
i
¯ v
i
. (7.6)
The density n
i
of each of these electron populations is determined by
n
i
=

∂n
∂t

t
i
t (7.7)
where t is the elementary time interval around t
i
during which these populations
are created.
The velocity ¯ v
i
of the electrons created at the successive instants t
i
is deter-
mined by
¯ v
i
(t ) = −
e
m

t
t
i
¯
E(t
/
) dt
/
÷ ¯ v
i 0
. (7.8)
We can replace this integral inside equation (7.7), assume that the electrons
are created initially with zero kinetic energy, ¯ v
i 0
= 0, and take the limit t →0.
The result is
¯
J =
e
2
m

t
−∞
dt
/

∂n
∂t

t
/

t
t
/
dt
//
¯
E(t
//
). (7.9)
142 Non-stationary processes in a cavity
If we take the time derivative of the current, we get

¯
J
∂t
=
e
2
m
¸
t
−∞

∂n
∂t

t
/
¯
E(t
/
) dt
/
÷

∂n
∂t

t

t
t
¯
E(t
//
) dt
//
¸
. (7.10)
The second term in this expression is obviously equal to zero, and therefore
the equation reduces to

¯
J
∂t
=
e
2
m

t
−∞

∂n
∂t

t
/
¯
E(t
/
) dt
/
=
e
2
m
¸
n(t
/
)
¯
E(t
/
)
¸
t
−∞

t
−∞
n(t
/
)

¯
E
∂t
/
dt
/
. (7.11)
If the electric field varies much faster than the electron density (which im-
plies once again the existence of two timescales), this expression reduces to

¯
J
∂t
=
0
ω
2
p
(t )
¯
E(t ). (7.12)
This allows us to write the field equation (7.1) in the following closed form:


2

1
c
2

2
∂t
2

¯
E =
ω
2
p
c
2
¯
E. (7.13)
We can now replace the mode decomposition (7.2) in this equation and, after
using the orthonormality condition (7.5), we obtain


2
∂t
2
÷k
2
l
c
2

e
l
(t ) = −
¸
l
/
C
ll
/ (t )e
l
/ (t ) (7.14)
where we have used
k
2
l
≡ k
2
mnp
= (mπ/L
x
)
2
÷(nπ/L
y
)
2
÷( pπ/L
z
)
2
. (7.15)
This equation describes the time evolution of the mode amplitudes due to the
linear mode coupling induced by the ionization of the gas inside the cavity. The
mode coupling coefficients C
ll
/ are determined by
C
ll
/ (t ) =

V
ω
2
p
(¯ r, t )E
l
(¯ r) · E
l
/ (¯ r) d¯ r. (7.16)
The mode coupling equation (7.14) can also be written in the following form:
¸

2
∂t
2
÷ω
2
l
(t )

e
l
(t ) = −
¸
l
/
,=l
C
ll
/ (t )e
l
/ (t ). (7.17)
Flash ionization in a cavity 143
Here we have used the time-dependent mode frequency
ω
2
l
(t ) = k
2
l
c
2
÷C
ll
(t ). (7.18)
Here, the self-coupling coefficient (l
/
= l) corresponds to a kind of averaged
plasma frequency, taken over the mode space configuration
C
ll
=

v
ω
2
p
(¯ r, t )[E
l
(¯ r)[
2
d¯ r. (7.19)
The mode coupling equation (7.17) is the basic equation of the present
model, and can be used for a detailed description of the spectral changes inside
the electromagnetic cavity. Its physical content and its relevance to our problem
will be discussed in the next few sections.
7.2 Flash ionization in a cavity
The simplest possible model of plasma formation inside a cavity is to admit that
ionization is uniform over the entire volume of the cavity: ω
2
p
(¯ r, t ) ≡ ω
2
p
(t ). In
this case, from equation (7.16), we have
C
ll
/ (t ) = ω
2
p
(t )δ
ll
/ . (7.20)
This means that the cavity eigenmodes are decoupled from each other and
that they evolve according to the equation

2
e
l
∂t
2
÷ω
2
l
(t )e
l
= 0 (7.21)
where ω
2
l
(t ) = k
2
l
c
2
÷ω
2
p
(t ).
Assuming that the plasma frequency ω
p
changes over a slow timescale, we
can integrate this equation and obtain the following WKB solution:
e
l
(t ) = E
l

k
l
c/ω
l
(t ) exp
¸
−i

t
ω
l
(t
/
) dt
/
)
¸
. (7.22)
Here E
l
is a constant amplitude. In order to be more specific, let us use the
following explicit law for the plasma creation inside the cavity:
ω
2
p
(t ) = ω
2
p0

1 −e
−γ t

. (7.23)
This means that the plasma is formed on a timescale of 1/γ . The integral in
equation (7.22) can now be written as

t
ω
l
(t
/
) dt
/
= k
l
c
¸¸
1 ÷
1
2

ω
p0
k
l
c

2

t ÷
1

ω
p0
k
l
c

2
e
−γ t
¸
. (7.24)
144 Non-stationary processes in a cavity
Let us see what happens during the first stage of the plasma formation when
we have γ t <1. We simply get
e
l
(t ) . E
l
exp

−ik
l
ct −i
ω
2
p0
2γ k
l
c

. (7.25)
This means that, in this short time limit, the cavity eigenmode keeps oscil-
lating at the same frequency, but suffers a small phase shift. In the opposite limit
of very long times, such that γ t ·1, we have
e
l
(t ) . E
l

k
l
c/ω
l
(∞) exp(−iω
l
(∞)t ) (7.26)
where ω
l
(∞) is the asymptotic value of the mode eigenfrequency for very large
times: ω
2
l
(∞) = k
2
l
c
2
÷ω
2
p0
.
We can see that this WKB solution gives, for long times, a result which
coincides with that of the simple theory of flash ionization outlined in previous
chapters: the frequency shift of a given eigenmode of an electromagnetic cavity,
where a plasma is uniformly created by some external agent, is simply determined
by the asymptotic value of the plasma frequency.
On the other hand, in order to compare this new approach with the pre-
vious one, we should not forget that an eigenmode in a cavity is equivalent to
two travelling waves propagating in opposite directions. This can be seen from
equations (7.3), where the factor sin( pπz/L
z
) can be decomposed into the two
factors exp(±ik
z
z), with k
z
= pπ/L
z
, representing propagation along the z-axis
in opposite directions.
However, as we will see next, this view of flash ionization is still too simple
and we have to refine it. A more realistic model for flash ionization inside a
cavity will have to contain the description of some spatial structure. In general,
the plasma creation will not be completely uniform over the entire cavity volume,
and we have to retain the space dependence of ω
2
p
(¯ r, t ). This is illustrated in
figure 7.1.
Let us then return to the mode coupling equation (7.17) and try a WKB
solution of the form (7.22), but where E
l
represents a time-dependent amplitude.
Taking the second time derivative of this new solution we get

2
e
l
∂t
2
. −ω
2
l
e
l
−2i

ω
l
k
l
c
∂ E
l
∂t
exp
¸
−i

t
ω
l
(t
/
) dt
/
¸
. (7.27)
Using this in the mode coupling equation, we obtain an equation for the
slowly varying amplitude, in the form
∂ E
l
∂t
= −i
¸
l,=l
/
B
ll
/ (t )E
l
/ . (7.28)
The new mode coupling coefficients are determined by
B
ll
/ (t ) =
C
ll
/ (t )
2

ω
l
(t )ω
l
/ (t )
exp

−i

t

l
/ (t
/
) −ω
l
(t
/
)] dt
/

. (7.29)
Flash ionization in a cavity 145
plasma
Figure 7.1. Non-uniform plasma formation inside a cavity.
Let us examine the simple but physically meaningful case where a given
mode l = l
0
pre-exists in the cavity for t < 0, and all the other modes are absent
before the plasma formation. If the coupling is not too strong, the amplitude of
this mode can still be considered as approximately constant for times t ≥ 0.
Using this parametric approximation, we can write from equation (7.27) that,
for any other mode l ,= l
0
, the amplitudes are
E
l
(t ) = −iE
l
0

t
B
ll
/ (t
/
) dt
/
. (7.30)
This equation allows us to calculate explicitly, and in a simple way, the
mode coupling efficiency. The explicit calculation can be performed by using
the following model for non-uniform ionization inside the cavity:
ω
2
p
(¯ r, t ) = ω
2
p0
(1 −e
−γ t
) exp[−a(¯ r − ¯ r
0
)
2
]. (7.31)
Here 1/γ is the timescale for plasma formation and a
−1/2
is the dimension
of the ionized region around some point ¯ r
0
. According to their definition, the
mode coupling coefficients C
ll
/ (t ) can be split into two factors, one containing
the time dependence and the other containing the spatial mode coupling.
Using equations (7.31) and (7.16), we obtain
C
ll
/ (t ) = ω
2
p0
(1 −e
−γ t
)I
ll
/ (7.32)
where
I
ll
/ =

t
e
−a(¯ r−¯ r
0
)
2
E
l
(¯ r) · E
l
/ (¯ r) d¯ r. (7.33)
Using the eigenmodes (7.3), or any others, this quantity can then be explicitly
evaluated. From our analysis, two new aspects relevant to flash ionization can be
clearly identified [67].
146 Non-stationary processes in a cavity
plasma
Figure 7.2. Moving ionization front inside a cavity.
First, the ionization processes inside a cavity not only produce a shift in
the frequency of a pre-existing mode but also give rise to a large spectrum of
radiation associated with the amplitudes e
l
(t ), for l ,= l
0
, due to linear mode
coupling. Second, the frequency shift associated with each mode is not simply
determined by the asymptotic value of the plasma frequency ω
p0
but by the self-
coupling coefficient C
ll
< ω
p0
defined by equation (7.19), which is a kind of
plasma frequency averaged over the eigenmode spatial field distribution.
This means that the spectrum generated by flash ionization inside a cavity is
broader, and the frequency shift of the pre-excited mode is smaller, than predicted
by single photon dynamics. These two qualitative aspects are well confirmed by
experimental results in microwave cavities, where the observed frequency shifts
were always smaller than the expected ω
l
=

k
2
l
c
2
÷ω
2
p0
, and a broad spectrum
of radiation was observed.
7.3 Ionization front in a cavity
Let us now turn to the problem of an ionization front travelling across the cavity,
as illustrated in figure 7.2. Then ω
2
p
(¯ r, t ) will be of the general formω
2
p
(¯ r −¯ vt ). In
the simplest but physically realistic case of a front which is uniform in the planes
z = const, and starts its motion at the position z = L
z
, moving along the z-axis
with a velocity v, we can write
ω
2
p
(¯ r, t ) = ω
2
p0
f (z ÷vt ) (7.34)
where f (z ÷vt ) describes the form of the front.
In this case, the coupling coefficients (7.16) can be written as
C
ll
/ (t ) =
2
L
z
ω
2
p0
δ
mm
/ δ
nn
/ I
pp
/ (t ) (7.35)
where the integral I
pp
/ (t ) is determined by
I
pp
/ (t ) =

L
z
0
sin( pπz/L
z
) sin( p
/
πz/L
z
) f (z ÷vt ) dz. (7.36)
Ionization front in a cavity 147
In order to have an order of magnitude estimate of the amplitude of the cavity
modes excited by the ionization front, we can use the simple model for the front
form
f (z ÷vt ) = H(z ÷vt − L
z
) (7.37)
where H(z −a) is the Heaviside function and the particular value a = L
z
−vt is
chosen in order to describe a sharp front starting its motion at (z = L
z
, t = 0) and
moving across the cavity along the z-axis towards the point z = 0 with a negative
velocity.
For this particular case, the integral (7.36) can be easily calculated and gives
I
pp
/ (t ) = −
L
z

1
p − p
/
sin
¸
( p − p
/
)π(1 −vt /L
z
)
¸
÷
L
z

1
p ÷ p
/
sin
¸
( p ÷ p
/
)π(1 −vt /L
z
)
¸
. (7.38)
It can be noticed from equation (7.35) that the mode coupling is only as-
sociated with the quantum numbers p and p
/
, leaving the other two quantum
numbers m and n invariant. This is due to the fact that the plasma frequency is
constant along the x- and the y-axis with which these two quantum numbers are
associated. This means that the mode coupling will only occur along the direction
of the electron density gradient.
Using this expression in equation (7.35) and noting that k
z
= pπ/L
z
and
k
/
z
= p
/
π/L
z
, we obtain
C
ll
/ (t ) = −
ω
2
p0
π
δ
mm
/ δ
nn
/

¸
(−1)
p−p
/
p − p
/
sin[(k
z
−k
/
z
)vt ] −
(−1)
p÷p
/
p ÷ p
/
sin[(k
z
÷k
/
z
)vt ]
¸
. (7.39)
This expression is valid for p ,= p
/
. In the case of p = p
/
, it is replaced by
C
ll
(t ) = −
ω
2
p0
2pπ
{2k
z
÷sin[2k
z
L
z
(1 −vt /L
z
)]} . (7.40)
In order to discuss the physical meaning of this result, let us go back to the
evolution equation (7.17). Let us also assume that the mode l
/
, pre-existing in the
cavity for times t < 0, is still the dominant mode for subsequent times (which
implies that the mode coupling is not very efficient).
For our qualitative discussion it is appropriate to neglect the slow variation
of the frequency eigenmode, such that we can use the approximate expression
e
l
/ (t ) . E
/
e
−iω
l
/ t
. (7.41)
148 Non-stationary processes in a cavity
We can then see that only four distinct forced terms exist, for which the
mode amplitudes e
l
are resonantly excited, and these terms verify the resonance
condition
ω
l
= ω
l
/ ±[(k
z
±k
/
z
)v]t. (7.42)
If we were in a free space configuration, and not inside a cavity, we would
simply have k
z
= ω
l
/c and k
/
z
= ω
l
/ /c, which means that this resonance condition
would reduce to the well-known expression for the relativistic mirror:
ω
l
= ω
l
/
1 ±β
1 ±β
(7.43)
where β = v/c.
The only and significant difference with respect to the relativistic mirror
effect is that here we generally have partial and not total reflection. The four
resonance conditions therefore have a very simple explanation: the pre-existing
cavity mode would correspond in free space to two travelling waves with the same
frequency, but propagating in opposite directions. Each of these travelling waves
interacts with the front and, from its partial reflection, two new waves result,
propagating in opposite directions and with distinct frequencies.
Let us come back to the mode coupling equation and retain one of the four
resonant terms. We can then write

d
2
dt
2
÷ω
2
l

e
l
= E
/
e
−iω
l
t
(7.44)
where E
/
depends on the coupling with the pre-existing mode l
/
, and ω
l
is given
by one of the possible four choices of equation (7.42).
Using
e
l
(t ) = E
l
(t ) e
−iω
l
t
(7.45)
we obtain the following saturation amplitude for the resonant mode p · p
/
, after
the completed journey of the ionization front across the cavity:
E
l
(sat) .
ω
2
p0

E
/

l
L
z
v
. (7.46)
We can see that a larger value for the front velocity v corresponds to a larger
value for the quantum number p compatible with the resonance condition, and
to a smaller saturation amplitude. This qualitative discussion suggests that the
resonant mode with a lower frequency shift will have a larger saturation amplitude
than the resonant mode with a larger frequency shift. This eventually explains the
absence of this mode in the first experiments of ionization fronts in a microwave
cavity [95].
But, apart from the resonant modes which are more strongly coupled with
the pre-existing mode in the cavity, many other modes can also be excited by
Electron beam in a cavity 149
C
ln
|E |
|E |
|E

|
2
3
4
t
Figure 7.3. Time variation of the mode coupling coefficient and of amplitudes of the
lowest order excited modes.
linear mode coupling, leading to an electromagnetic energy cascade over a large
spectral width, as shown by numerical integration of the above mode coupling
equations [100], and qualitatively confirmed by experiments. See figure 7.3 for a
numerical example.
7.4 Electron beam in a cavity
Instead of considering ionization processes inside a cavity, we can also obtain a
similar moving boundary if an electron beam is sent across the cavity. The mode
coupling due to an electron beamis formally analogous to that due to an ionization
front. However, the two processes also present very distinctive features.
First of all, it should be noticed that the field equation (7.1) can also be
written as

c
2

2


2
∂t
2

¯
A = −
¯
J (7.47)
where
¯
A is the vector potential.
In our discussion of the flash ionization and the ionization front we used
equation (7.12), which can be rewritten as
¯
J = −
e
2
m

t
−∞

∂n
∂t

t
/
[
¯
A(t ) −
¯
A(t
/
)] dt
/
. (7.48)
Using this expression in equation (7.47) leads to our previous field equa-
tion (7.13). However, the current associated with an electron beam cannot be
150 Non-stationary processes in a cavity
described by the same source current because equations (7.12, 7.48) were derived
for electrons created at specific times and positions, with a velocity equal to zero.
In contrast, the electrons of an electron beam will exist for all times, and will
always move with a non-negligible velocity. Using the equation of motion for the
beam electrons we can easily obtain the following new expression for the electric
current:
¯
J = −en(¯ r, t )¯ v = −
e
2
m
n(¯ r, t )
¯
A. (7.49)
Using it in equation (7.47), we obtain

c
2

2


2
∂t
2

¯
A = ω
2
p
(¯ r, t )
¯
A. (7.50)
This is formally identical to equation (7.13), but with the electric field re-
placed by the vector potential. Because the spatial eigenfunctions for the electric
field and for the vector potential are identical, we can again use a linear expansion
for
¯
A of the form
¯
A(¯ r, t ) =
¸
l
a
l
(t )E
l
(¯ r). (7.51)
The evolution equation for the mode amplitudes can now readily be found:


2
∂t
2
÷ω
2
l
(t )

a
l
(t ) = −
¸
l
/
,=l
C
ll
/ (t )a
l
/ (t ). (7.52)
We see that it is formally identical to equation (7.17), but with a different
physical meaning, the mode amplitudes a
l
(t ) being related to the vector potential
eigenmodes. We can then repeat the calculations of the previous section. The
result is that the frequency spectum generated by the electron beam will be iden-
tical to that generated by the ionization front, because the modes with the same
frequency are coupled in the same way.
However, the energy content of each mode is different because the vector
potential amplitudes are related to the electric field amplitude by e
l
(t ) = a
l
(t )ω
l
.
This means that, for instance, the saturation amplitude obtained after the com-
pleted journey of the electron beam front across the cavity is now determined
by
A
l
(sat) .
ω
2
p0

A
/

l
L
z
v
(7.53)
where A
/
is the amplitude of the source term and where we have assumed that
a
l
(t ) . A
l
exp(−iω
l
t ).
Again, this is formally identical to equation (7.46), with the electric field
amplitudes replaced by vector potential amplitudes. But, if we rewrite this ex-
pression in terms of the electric field amplitudes, we realize that the saturation
amplitude is now larger, by a factor of ω
l
/ /ω
l
, than that for the ionization front
Electron beam in a cavity 151
case. Also, as we have seen from the resonance condition for the frequency shift
(7.42, 7.43), this factor can be much larger than one.
This means that the electron beam is a much more efficient way to up-shift
the electromagnetic energy in the frequency domain, not because the frequency
up-shift itself is different, but because the amplitudes of the shifted modes are
much larger. This can be illustrated by a numerical integration of the coupled
mode equations for both cases [100]. Using the same argument we can also con-
clude that the down-shifted modes will have, in contrast, a much lower amplitude
in the case of the electron beam.
The physical differences between these two processes of frequency up-
shifting can be better understood if we consider the total magnetic field generated
inside the cavity. We know that the magnetic field is determined by
¯
B = ∇
¯
A
and, using equation (7.51), we can write, for the case of the electron beam
¯
B(¯ r, t > L
z
/v) =
¸
l
[∇ E
l
(¯ r)]A
l
exp(−iω
l
t ). (7.54)
This means that the total magnetic field is just the sum of the magnetic fields
associated with each of the modes excited in the cavity.
A different situation occurs for the case of an ionization front. If we write
the total magnetic field in terms of the total electric field, we get
¯
B(¯ r, t ) = −∇

t
0
¯
E(¯ r, t ) = −
¸
l
[∇ E
l
(¯ r)]

t
0
e
l
(t
/
) dt
/
. (7.55)
The asymptotic value for this magnetic field will then be
¯
B(¯ r, t > L
z
/v) =
¸
l
1

l
[∇ E
l
(¯ r)]

δ
ll
/ ÷iω
l

L
z
/v
0
e
l
(t
/
) dt
/
− E
l
e
−iφ
l
÷ E
l
e
−iω
l
t

(7.56)
where we have φ
l
= ω
l
L
z
/v.
It can be seen from this expression that the last term, which oscillates at
the frequencies ω
l
, corresponds to the sum of the magnetic fields associated with
each of the modes excited in the cavity. But, the remaining terms describe a static
magnetic field which was excited by the ionization process and which can be
identified with the magnetic mode discussed in the previous chapter.
This means that, in the case of the ionization front, a large amount of energy
is transferred to the static magnetic field, which explains why this process of
frequency up-shifting is much less efficient than that of an electron beam where
such a static field is absent.
Finally, we should keep in mind that other possible non-stationary processes
occurring inside the cavity, which do not create new electron populations, are
described by the same mode coupling equations as the electron beamcase. This is,
152 Non-stationary processes in a cavity
for instance, the situation where, in a pre-existing plasma medium, a space–time
variation of the plasma density is produced, associated with an electron plasma
wave with relativistic velocity (wakefield). As shown in the above discussion, the
up-shifting frequency effect will be as efficient as in the case of the electron beam
and no static magnetic field will be excited.
7.5 Fermi acceleration in a cavity
The linear mode coupling formalism explored in this chapter allows us to give a
new description of the Fermi acceleration of photons, using a full wave descrip-
tion [101]. Let us then assume that, due to some externally applied perturbation,
the neutral gas inside the cavity is ionized in a limited domain near the boundary
z = L
z
, in such a way that the width of the plasma region oscillates in time with
a frequency , much smaller than the cavity mode eigenfrequencies ω
l
.
This can be described by the following space–time distribution for the elec-
tron plasma frequency:
ω
2
p
(¯ r, t ) = ω
2
p0
H[z − L
z
÷ A(1 −cos t )]. (7.57)
Here we have used the Heaviside function H(z − a). This simple law
describes a plasma with constant plasma frequency ω
p0
but oscillating in a region
of the cavity between L
z
and L
z
−2A.
The coupling coefficients (7.39) now take the form, for p ,= p
/
,
C
ll
/ = −
ω
2
p0
π
δ
mm
/ δ
nn
/
¸
1
p − p
/
sin[π( p − p
/
)(1 −(1 −cos t ))]

1
p ÷ p
/
sin[π( p ÷ p
/
)(1 −(1 −cos t ))]
¸
(7.58)
where = A/L
z
. For p = p
/
, equation (7.41) becomes
C
ll
= ω
2
p0
(1 −cos t ) −
ω
2
p0
2pπ
sin[2pπ(1 −cos t )]. (7.59)
Using this explicit expression in the mode coupling equations (7.17) we can
calculate the time evolution of the eigenmode field amplitudes. Let us start by
neglecting the mode coupling and let us study the evolution of a single mode
e
l
(t ), when according to equation (7.59), the effective value of plasma frequency
is modulated by a much lower frequency . Notice that mode coupling would be
absent if the plasma frequency were oscillating uniformly over the entire cavity
volume.
In this case, we can write the single mode evolution equation as
d
2
dt
2
e
l
(t ) ÷
¸
k
2
l
c
2
÷ω
2
p0
(1 −cos t )
Fermi acceleration in a cavity 153
÷

2pπ
sin[2pπ(1 −cos t )]
¸
e
l
(t ) = 0. (7.60)
Here it is appropriate to use a dimensionless time variable τ = t /2, and to
introduce two parameters:
δ =
4

2
(k
2
l
÷ω
2
p0
), γ
0
= 2
ω
2
p0

2
. (7.61)
Equation (7.60) can then be reduced to a more familiar form:
d
2

2
e
l
(τ) ÷
¸
δ −2γ
0
cos(2τ) −2
γ
0
2pπ
sin[2pπ(1 −cos(2τ))]
¸
e
l
(τ) = 0.
(7.62)
For high-frequency cavity modes such that 2pπ · 1 the last term can be
neglected and this reduces to the well-known Mathieu equation. In order to illus-
trate the main physical processes in the cavity let us first restrict our discussion to
this equation.
It is well known [81] that the Mathieu equation has bounded solutions, as
well as unbounded ones, depending on the values of the two parameters δ and γ
0
.
The existence of unbounded (or unstable) solutions is physically very interesting
because this means that it is possible to excite cavity modes, starting from the
noise level up to arbitrary amplitudes. The energy source responsible for the
creation of this electromagnetic field, nearly out of nothing, can only be identified
with the external source producing the plasma oscillations.
The stability diagramof the Mathieu equation shows that the most favourable
modes to be excited verify the resonance condition k
l
c . m/2, where m is some
positive integer. However, because we are assuming that is much smaller than
the eigenmode frequencies ω
l
, this resonance condition implies that m has to be
quite large. Also, in this case, we known from the properties of the Mathieu
equation that the unstable domain in the parameter space (δ, γ
0
) is extremely
narrow and very difficult to experimentally satisfy. This means that, in principle,
generation of electromagnetic energy inside an oscillating and empty cavity is
possible, but it is very unlikely to be observed, due to the fact that it will only
occur inside very narrow regions of the parameter space.
We now turn to the stable regions of the parameter space, where it is pos-
sible to derive approximate analytical expressions for the time evolution of the
amplitude of a given uncoupled mode e
l
(t ), and to describe the entire frequency
spectrum associated with this single mode. Due to the frequency modulations of
the self-coupling coefficient (7.59), the frequency spectrum of a single mode no
longer corresponds to a well-defined frequency ω
l
, but to an infinite number of
frequencies.
This can be clearly seen if we assume that the mode amplitude e
l
(t ) is
described by
e
l
(t ) =
¸
k
e
lk
exp[−i( ¨ ω
l
÷k)t ] (7.63)
154 Non-stationary processes in a cavity
where we have used ¨ ω
2
l
= k
2
l
c
2
÷ ω
2
p0
, and have assumed that the amplitudes
e
lk
(t ) vary over a timescale much larger than 1/ ¨ ω
l
.
Using this solution in the single mode evolution equation (7.62), and retain-
ing only the terms with the same fast timescale, we can easily derive an equation
for the new amplitudes e
lk
(t ):
−2i( ¨ ω
l
÷k)
d
dt
e
lk
÷[ ¨ ω
2
l
−( ¨ ω
l
÷k)
2
]e
lk


2i
ω
2
p0
(e
l,k−1
−e
l,k÷1
) ÷i
ω
2
p0
4pπ
¸
m
J
m
(2pπ)

¸
i
m
e
2i pπ
e
l,k−m
−i
−m
e
−2i pπ
e
l,k÷m
¸
= 0. (7.64)
In order to obtain this expression we have used the expansion of the sine
term of equation (7.62) in Bessel functions of integer order J
m
. We can simplify
it considerably if we notice that, for < ¨ ω
l
, the second term (proportional
to e
lk
) can be neglected. On the other hand, if the oscillation parameter is
small and if the quantum number p is large, in such a way that the inequality
2pπ < 1 is satisfied, we can also neglect the term containing the sum over the
Bessel functions.
With these two simplifying assumptions, we can reduce equation (7.64) to
d
dt
e
lk
=
ω
2
p0

k
(e
l,k−1
−e
l,k÷1
) (7.65)
where we have used ω
k
= ¨ ω
l
÷k.
This system of coupled equations for the amplitudes e
lk
can be easily inte-
grated by noting that it is formally identical to the well-known recurrence relation
for Bessel functions J
k
[84]. This means that it satisfies the following solution:
e
lk
(t ) = J
k

ω
2
p0

k
t

. (7.66)
Let us now consider the opposite limit of small and large p, such that
2pπ · 1. In this case, the sum term dominates over the term proportional to
in equation (7.64). Moreover, we also have J
1
(2pπ) · J
m
(2pπ), for m ,= 1,
which means that we can also neglect all the terms in the series expansion except
those for m = ±1.
The resulting equation is
d
dt
e
lk
= i
ω
2
p0
4pπω
k
J
1
(2pπ) cos(2pπ)(e
l,k−1
÷e
l,k÷1
). (7.67)
Here again, the integration becomes possible because of the formal analogy
with the recurrence relation between the Bessel functions with purely imaginary
Fermi acceleration in a cavity 155
argument I
k
. The solution can now be written as
e
lk
(t ) = (−1)
k
i
−k
J
k

ω
2
p0
2pπω
k
cos(2pπ)J
1
(2pπ)t

. (7.68)
We clearly see from these solutions that, in the two opposite limits of the
parameter (2pπ), an infinitely large spectrum of frequencies with decreasing
amplitudes can be generated out of a single mode contained in a variable cavity.
This means that Fermi acceleration results in a considerable line broadening.
This effect of energy spreading over the frequency domain is even more dra-
matic when we include the linear mode coupling associated with the coefficients
C
ll
/ established above. Once again, it is not possible to integrate the resulting
mode equations, unless we introduce some simplifying assumptions.
First of all, let us assume that the cavity contains a dominant mode character-
ized by the index l
/
, such that coupling occurs mainly between this and the other
cavity modes.We can write the amplitude e
l
/ of the dominant mode in the form
e
l
/ (t ) = E
l
/ e
−i ¨ ω
l
/ t
(7.69)
where E
l
/ can be assumed nearly constant.
The amplitudes of all the other modes, l ,= l
/
, can then be described by
¸
d
2
dt
2
÷ω
2
l
(t )

e
l
(t ) = −C
ll
/ E
l
/ e
−i ¨ ω
l
/ t
(7.70)
where ω
2
l
(t ) and C
ll
/ (t ) are determined by equations (7.18, 7.58).
We can see from here that the modes which are excited with a larger ampli-
tude satisfy the nearly resonant condition
ω
l
(t ) . ¨ ω
l
= ¨ ω
l
/ ±m (7.71)
where m is an integer.
For these nearly resonant modes, we can write the following evolution equa-
tion:
¸
d
2
dt
2
÷ω
2
l
(t )

e
l
(t ) = ±iE
l
/
ω
2
p0

¸
J
m
(( p ÷ p
/
)π)
p ÷ p
/

J
m
(( p − p
/
)π)
p − p
/
¸
e
−i ¨ ω
l
/ t
. (7.72)
This equation can easily be integrated if we neglect the slow time evolution
of the eigenfrequency ω
l
. ¨ ω
l
, and assume that
e
l
(t ) = E
l
e
−i ¨ ω
l
t
(7.73)
156 Non-stationary processes in a cavity
where E
l
(t ) is the slow amplitude of the nearly resonant modes.
In this approximation, the solution for short times is given by
E
l
(t ) = ∓E
l
/
ω
2
p0
4π ¨ ω
l
¸
J
m
(( p ÷ p
/
)π)
p ÷ p
/

J
m
(( p − p
/
)π)
p − p
/
¸
t. (7.74)
This result shows that the modes that are more easily excited are those for
which, in addition to the resonance condition (7.71), we have p . p
/
. Of course,
for low modulation frequencies < ω
l
, these resonance conditions are satisfied
by several modes. In the limit < (ω
l÷1
− ω
l
) all the modes in the cavity will
be nearly resonant, with an amplitude approximately given by equation (7.74).
If we complement this with the tendency of each mode considered indi-
vidually to suffer a considerable resonant broadening, as shown by the above
discussion of the single mode evolution, we conclude that Fermi acceleration
inside an oscillating cavity provides an efficient mechanismfor generating a broad
spectrum of electromagnetic radiation.
This is in qualitative agreement with our previous discussion of the Fermi
photon acceleration using the Hamiltonian approach, where it was shown that
the photon trajectories would become stochastic under certain conditions. Here
the particle stochastic behaviour is replaced by the mode coupling, in the same
way as stochasticity in classical particle motion is replaced by a non-stochastic
description based on the wave equation for a quantum particle.
Chapter 8
Quantum theory of photon acceleration
In order to complete our theoretical framework, it is important to showthat photon
acceleration is not a result of the classical or semi-classical approximations, but
that it can easily be identified and described in purely quantum grounds.
As an introduction to this chapter, we will describe first the well-known
quantization procedure of the electromagnetic field [37, 61], with small changes,
in order to adapt the notation to our specific needs. We will also describe the less
well-known procedure for the field quantization in a plasma.
We will then examine the quantumtheory of space and time refraction, which
not only confirms the main features of the classical theory, but also identifies
intrinsic quantum processes such as the possibility of photon creation from the
vacuum.
8.1 Quantization of the electromagnetic field
8.1.1 Quantization in a dielectric medium
We consider an infinite, non-dispersive and non-dissipative dielectric medium,
characterized by a real dielectric constant, . We give a phenomenological de-
scription of the problem of the field quantization in this medium, by assuming
that is a well-known constant, given a priori and not explicitly calculated by a
microscopic theory. This approach will be useful for our problem.
We start with Maxwell’s equations which can be written, in the absence of
charge and current distributions, as

¯
E = −

¯
B
∂t
, ∇ ·
¯
D = 0

¯
H =

¯
D
∂t
, ∇ ·
¯
B = 0
(8.1)
with
¯
D =
¯
E and
¯
B = µ
0
¯
H.
157
158 Quantum theory of photon acceleration
We define the scalar and the vector potentials, φ and
¯
A, such that
¯
E = −

¯
A
∂t
−∇φ,
¯
B = ∇
¯
A. (8.2)
It is convenient to choose the Coulomb gauge, defined by the condition
∇ ·
¯
A = 0. (8.3)
In this case, we can derive from equations (8.1) the following equations for
the scalar and vector potentials:

2
φ = 0 (8.4)

2
¯
A −
n
2
c
2

2
¯
A
∂t
2
=
n
2
c
2

∂φ
∂t
(8.5)
with n =

/
0
.
This obviously implies that φ = 0, and

2
¯
A −
n
2
c
2

2
¯
A
∂t
2
= 0. (8.6)
We know that, in classical theory, the general solution of this equation can
be written as
¯
A(¯ r, t ) = 2

¯
A
k
e
i(
¯
k·¯ r−ω
k
t )
d
¯
k
(2π)
3
(8.7)
where ω
k
= (kc/n).
In this expansion, the integration extends over both the negative and the
positive values of the three components of the wavevector
¯
k. But we know that,
in order to guarantee that the electromagnetic fields are real quantities, we have
to assume that
¯
A
−k
=
¯
A

k
.
This allows us to rewrite the above general solution as
¯
A(¯ r, t ) =

¯
A
k
(¯ r, t )
d
¯
k
(2π)
3
(8.8)
with
¯
A
k
(¯ r, t ) =
¸
¯
A
k
e
i(
¯
k·¯ r−ω
k
t )
÷
¯
A

k
e
−i(
¯
k·¯ r−ω
k
t )
¸
. (8.9)
The corresponding electric and magnetic fields are, according to equa-
tions (8.2),
¯
E(¯ r, t ) =

¯
E
k
(¯ r, t )
d
¯
k
(2π)
3
,
¯
B(¯ r, t ) =

¯
B
k
(¯ r, t )
d
¯
k
(2π)
3
(8.10)
Quantization of the electromagnetic field 159
with
¯
E
k
(¯ r, t ) = −iω
k
¸
¯
A
k
e
i(
¯
k·¯ r−ω
k
t )

¯
A

k
e
−i(
¯
k·¯ r−ω
k
t )
¸
(8.11)
¯
B
k
(¯ r, t ) = i
¯
k
¸
¯
A
k
e
i(
¯
k·¯ r−ω
k
t )

¯
A

k
e
−i(
¯
k·¯ r−ω
k
t )
¸
. (8.12)
Let us now consider the total electromagnetic energy density
W =

W
k
d
¯
k
(2π)
3
(8.13)
with
W
k
=
1
2
¸
[
¯
E
k
(¯ r, t )[
2
÷
1
µ
0
[
¯
B
k
(¯ r, t )[
2
¸
= 2ω
2
k
(
¯
A
k
·
¯
A

k
). (8.14)
We know that the amplitudes of the vector potential are, in general, complex
quantities and it is useful to consider their real and imaginary parts separately, by
defining
q
k
=

(A
k
÷ A

k
), p
k
= iω
k

(A
k
− A

k
). (8.15)
Here we have considered scalar quantities, by using
¯
A
k
= A
k
¯ e
k
, where ¯ e
k
is
the unit polarization vector. From these two quantities, we can write
A
k
=
1

k

k
q
k
÷i p
k
). (8.16)
Using this in equation (8.14), we obtain for the energy density of the mode
¯
k
W
k
=
1
2
( p
2
k
÷ω
2
k
q
2
k
). (8.17)
This is formally identical to the energy of a one-dimensional oscillator with
unit mass, frequency ω, position q and momentum p:
W ≡ H(q, p) =
p
2
2
÷ω
2
q
2
2
. (8.18)
The first term represents the kinetic energy and the second term the parabolic
potential. This classical oscillator can be quantized by defining a Hamiltonian
operator
ˆ
H, such that
ˆ
H =
1
2
( ˆ p
2
÷ω
2
ˆ q
2
) (8.19)
where the position and the momentum operators satisfy the commutation relations
[ ˆ q, ˆ p] ≡ ˆ q ˆ p − ˆ p ˆ q = i
¨
h. (8.20)
160 Quantum theory of photon acceleration
The Hamiltonian operator can also be written in terms of the destruction and
creation operators, ˆ a and ˆ a
÷
, such that
ˆ q =

¨
h

( ˆ a ÷ ˆ a
÷
), ˆ p = −i

¨

2
( ˆ a − ˆ a
÷
). (8.21)
This is equivalent to
ˆ a =
1

2
¨

(ωˆ q ÷i ˆ p), ˆ a
÷
=
1

2
¨

(ωˆ q −i ˆ p). (8.22)
From this and from equations (8.17, 8.20), we have
ˆ a
÷
ˆ a =
1
2
¨


2
ˆ q
2
÷ ˆ p
2
÷iω[ ˆ q, ˆ p]) =
1
¨

ˆ
H −
1
2
¨

. (8.23)
We also have
ˆ a ˆ a
÷
=
1
¨

ˆ
H ÷
1
2
¨

. (8.24)
This leads to the following commutation relation:
[ ˆ a, ˆ a
÷
] = ˆ a ˆ a
÷
− ˆ a
÷
ˆ a = 1. (8.25)
From equations (8.19, 8.24) we can also write the Hamiltonian operator in
the form
ˆ
H =
¨

1
2
÷
ˆ
N

(8.26)
where
ˆ
N is the quantum number operator
ˆ
N = ˆ a
÷
ˆ a. (8.27)
Returning to the electromagnetic field, we see that we can associate with
each mode
¯
k a one-dimensional quantum oscillator with destruction and creation
operators ˆ a
k
and ˆ a
÷
k
, such that the field amplitude operators for each mode are
established by the equivalence
¯
A
k

¨
h

k
ˆ a
k
¯ e
k
,
¯
A

k

¨
h

k
ˆ a
÷
k
¯ e

k
. (8.28)
At this point we should notice that the electromagnetic radiation has two
independent polarization vectors. In order to take them both into account we
change the operators in the following way:
ˆ a
k
¯ e
k

¸
λ=1,2
a(
¯
k, λ)¯ e(
¯
k, λ). (8.29)
Quantization of the electromagnetic field 161
From now on, we will remove the hat in the operator notation. The two unit
polarization vectors are orthogonal:
¯ e(
¯
k, λ) · ¯ e(
¯
k, λ
/
) = δ
λλ
/ . (8.30)
An obvious choice of these two orthogonal polarization vectors is two linear
polarizations in the plane perpendicular to the direction of wave propagation:
¯
k · ¯ e(
¯
k, λ) = 0 (λ = 1, 2) (8.31)
¯ e(
¯
k, 1) ¯ e(
¯
k, 2) =
¯
k/[k[. (8.32)
The total vector potential operator will then be given by
¯
A(¯ r, t ) =
¸
λ=1,2

d
¯
k
(2π)
3

¨
h

k
¸
a(
¯
k, λ)¯ e(
¯
k, λ) e
i(
¯
k·¯ r−ω
k
t )
÷a
÷
(
¯
k, λ)¯ e

(
¯
k, λ) e
−i(
¯
k·¯ r−ω
k
t )
¸
. (8.33)
From equations (8.2) we can also establish the electric and the magnetic field
operators:
¯
E(¯ r, t ) = i
¸
λ=1,2

d
¯
k
(2π)
3

¨

k
2
¸
a(
¯
k, λ)¯ e(
¯
k, λ) e
i(
¯
k·¯ r−ω
k
t )
−a
÷
(
¯
k, λ)¯ e

(
¯
k, λ) e
−i(
¯
k·¯ r−ω
k
t )
¸
(8.34)
and
¯
B(¯ r, t ) = i
¸
λ=1,2

d
¯
k
(2π)
3

¨
h

k
¸
a(
¯
k, λ)
¯
k ¯ e(
¯
k, λ) e
i(
¯
k·¯ r−ω
k
t )
−a
÷
(
¯
k, λ)
¯
k ¯ e

(
¯
k, λ) e
−i(
¯
k·¯ r−ω
k
t )
¸
. (8.35)
These field operators are strictly equivalent to those obtained for propagation
in a vacuum, the only difference being the replacement of the vacuum permitivity

0
by the appropriate dielectric constant of the medium, .
8.1.2 Quantization in a plasma
We consider an infinite, homogeneous and unmagnetized plasma. In contrast with
the previous case, we are now dealing with a dispersive medium. Instead of using
the plasma dielectric function, it is more appropriate to describe this medium as a
vacuum plus charge and current distributions. Maxwell’s equations, which are the
162 Quantum theory of photon acceleration
starting point of our quantization procedure, can then be written in the following
form:

¯
E = −

¯
B
∂t
, ∇ ·
¯
E =
ρ

0
(8.36)

¯
H =
¯
J ÷
0

¯
E
∂t
, ∇ ·
¯
B = 0 (8.37)
with
¯
B = µ
0
¯
H.
Assuming that the plasma ions are at rest, with a mean density n
0
, we can
write the charge and current densities as
¯
J = −en¯ v, ρ = −e(n −n
0
). (8.38)
The electron density and mean velocity are described by the continuity and
the momentum conservation equations:
∂n
∂t
÷∇ · n¯ v = 0 (8.39)
∂ ¯ v
∂t
÷ ¯ v · ∇¯ v = −
e
m
(
¯
E ÷ ¯ v
¯
B) − S
2
e
∇ ln n. (8.40)
We define here the electron thermal velocity by the quantity v
e
=

T/m,
where T is the plasma temperature, and use S
2
e
= 3v
2
e
. By treating the plasma
electrons as a fluid we are neglecting the resonant particle effects which can, for
instance, lead to purely kinetic effects such as the electron Landau damping. The
inclusion of these effects would require a more refined and purely microscopic
quantization procedure.
Apart from the fluid approach, we will also restrict our discussion to low
field intensities. This allows us to linearize the fluid equations (8.39, 8.40) around
the equilibrium state.
Using n = n
0
÷ ˜ n, where [ ˜ n[ <n
0
is the density perturbation, we get
∂ ˜ n
∂t
÷n
0
∇ · ¯ v = 0
∂ ¯ v
∂t
= −
e
m
¯
E −
S
2
e
n
0
∇n.
(8.41)
The linearized electron current and charge densities are
¯
J = −en
0
¯ v, ρ = −e ˜ n. (8.42)
Let us introduce the scalar and vector potentials φ and
¯
A, by using equa-
tions (8.2), and retain the Coulomb gauge, defined by the condition (8.3). The
Quantization of the electromagnetic field 163
corresponding potential equations can be derived from equations (8.37) and take
the form

2
φ =
e

0
˜ n (8.43)


2

1
c
2

2
∂t
2

¯
A =
en
0
c
2

0
¯ v

(8.44)
where ¯ v

is the transverse part of the electron velocity, determined by the condi-
tion
∇ · ¯ v

= 0. (8.45)
On the other hand, from the electron fluid equations we can easily derive


2
∂t
2
− S
2
e

2

˜ n = −
en
0
m

2
φ (8.46)

2
¯ v

∂t
2
=
e
m

2
¯
A
∂t
2
. (8.47)
This allows us to write


2
∂t
2
− S
2
e

2

˜ n = −ω
2
p
˜ n (8.48)
¯ v

=
e
m
¯
A (8.49)
which, according to equations (8.43, 8.44), implies that


2

1
S
2
e

2
∂t
2

φ =
ω
2
p
S
2
e
φ (8.50)


2

1
c
2

2
∂t
2

¯
A =
ω
2
p
c
2
¯
A. (8.51)
We notice here that, if the electron thermal velocity could be made equal to
c/

3, such that S
2
e
= c
2
, these two equations would reduce to a single Klein–
Gordon equation for the four-vector potential A
ν
≡ (φ/c,
¯
A), in the form


2

1
c
2

2
∂t
2

A
ν
=
m
eff
c
¨
h
A
ν
(8.52)
where m
eff
= ω
p ¨
h/c
2
would be the mass of the vector field.
However, it is obvious that we have S
2
e
,= c
2
for general plasma conditions,
which means that the electromagnetic field in a plasma will not be exactly equiva-
lent (but it will be similar) to a massive vector field. In order to prepare for the field
164 Quantum theory of photon acceleration
quantization procedure we can write the general solution of equations (8.50, 8.51)
in the form
¯
A(¯ r, t ) =

¯
A
k
(¯ r, t )
d
¯
k
(2π)
3
, φ(¯ r, t ) =

φ(¯ r, t )
d
¯
k
(2π)
3
(8.53)
with
¯
A
k
(¯ r, t ) =
¸
¯
A
k
e
i(
¯
k·¯ r−ω
k
t )
÷
¯
A

k
e
−i(
¯
k·¯ r−ω
k
t )
¸
(8.54)
and
φ
k
(¯ r, t ) =
¸
φ
k
e
i(
¯
k·¯ r−ω
k
t )
÷φ

k
e
−i(
¯
k·¯ r−ω
k
t )
¸
. (8.55)
These two equations for the vector and the scalar potentials are formally
identical to each other. But it should be noticed that, according to equa-
tions (8.50, 8.51), the frequencies for the vector potential (the first equation) are
determined by the dispersion relation
ω
k
=

k
2
c
2
÷ω
2
p
(8.56)
and for the scalar potential (the second equation) by
ω
k
=

k
2
S
2
e
÷ω
2
p
. (8.57)
The corresponding electric and magnetic fields are still represented by equa-
tions (8.2, 8.10), but now equation (8.12) is replaced by
¯
E
k
(¯ r, t ) = iω
k
¯
A
k
(¯ r, t ) −i
¯

k
(¯ r, t ),
¯
B
k
(¯ r, t ) = i
¯
k
¯
A
k
(¯ r, t ). (8.58)
Let us now consider the total energy density
w
k
=
1
2
¸

0
[
¯
E
k
(¯ r, t )[
2
÷
1
µ
0
[
¯
B
k
(¯ r, t )[
2
¸
÷w
part
(
¯
k)
= 2k
2
φ
k
φ

k
÷2ω
2
k
¯
A
k
·
¯
A

k
. (8.59)
Here, the first term corresponds to longitudinal (or electrostatic) oscillations,
and the second termto the transverse electromagnetic waves. For convenience, we
have added to the purely electromagetic energy the kinetic energy of the particles
associated with the oscillations of the scalar potential.
It is well known from plasma theory that, in the electron plasma oscillations,
the averaged energy is equally divided between the electrostatic field energy and
the kinetic energy of the plasma electrons. The result is the appearance of a factor
of 2 in the first term of this equation.
Field quantization is obtained by introducing destruction and creation opera-
tors a(
¯
k, λ) and a
÷
(
¯
k, λ), for each of the three distinct modes, such that we have,
for the transverse modes (λ = 1, 2),
¯
A
k

¨
h
2
0
ω
k
a(
¯
k, λ)¯ e(
¯
k, λ),
¯
A

k

¨
h
2
0
ω
k
a
÷
(
¯
k, λ)¯ e

(
¯
k, λ) (8.60)
Time refraction 165
and, for the longitudinal mode (λ = 3),
φ
k

¨
h
2
0
k
a(
¯
k, λ), φ

k

¨
h
2
0
k
a
÷
(
¯
k, λ). (8.61)
The total vector potential operator will then be determined by
¯
A(¯ r, t ) =
¸
λ=1,2

d
¯
k
(2π)
3

¨
h
2
0
ω
k
¸
a(
¯
k, λ)¯ e(
¯
k, λ) e
i(
¯
k·¯ r−ω
k
t )
÷a
÷
(
¯
k, λ)¯ e

(
¯
k, λ) e
−i(
¯
k·¯ r−ω
k
t )
¸
. (8.62)
The total scalar potential operator is
φ(¯ r, t ) =

d
¯
k
(2π)
3

¨
h
2
0
k
¸
a(
¯
k, 3) e
i(
¯
k·¯ r−ω
k
t )
÷a
÷
(
¯
k, 3) e
−i(
¯
k·¯ r−ω
k
t )
¸
. (8.63)
The resulting total energy operator will be given by
W =
3
¸
λ=1

d
¯
k
(2π)
3
¨

k
(λ)
¸
a
÷
(
¯
k, λ)a(
¯
k, λ) ÷
1
2
¸
(8.64)
such that
ω
k
(λ) =

k
2
c
2
(λ) ÷ω
2
p
(8.65)
with c
2
(λ = 1, 2) = c
2
, and c
2
(λ = 3) = S
2
e
.
The first two modes correspond to the two transverse photons and the third
mode corresponds to plasmons (or longitudinal photons). Here we find the three
independent polarization states of a massive vector field (which has spin one) [88],
but with two distinct characteristic velocities: for the transverse particles it is the
speed of light c, and for the plasmons it is the thermal velocity S
e
.
The similarities of the electromagnetic field quantization in a plasma with
the quantization of a massive vector field was noticed long ago by Anderson [5].
It has served as a phenomenological model for the theory of the Higgs boson [88].
8.2 Time refraction
8.2.1 Operator transformations
Let us consider a time discontinuity in an infinite dielectric medium such that, at
time t = 0, the dielectric constant suddenly changes from a value
1
to a new
value
2
. This transformation law can be described by the expression
(t ) =
1
H(−t ) ÷
2
H(t ) (8.66)
166 Quantum theory of photon acceleration
where H(t ) is the Heaviside function.
We know from equation (8.34) that, for a given polarization state of the
photons in the dielectric medium (λ = 1, or λ = 2), the electric field operator
can be written as
¯
E(¯ r, t ) = i

¨

k
2
¸
a(
¯
k, t ) e
i
¯
k·¯ r
−a
÷
(
¯
k, t ) e
−i
¯
k·¯ r
¸
¯ e
k
. (8.67)
Here we have used a real polarization vector ¯ e
k
= ¯ e

k
, and introduced time-
dependent destruction and creation operators
a(
¯
k, t ) = a(
¯
k) e
−iω
k
t
, a
÷
(
¯
k, t ) = a
÷
(
¯
k) e

k
t
. (8.68)
We also know that the displacement vector and the magnetic field operators
can be determined by
¯
D(
¯
k, t ) =
¯
E(
¯
k, t ),
¯
B(
¯
k, t ) =
¯
k
ω
k

¯
E(
¯
k, t ). (8.69)
This means that we can write
¯
D(
¯
k, t ) = i¯ e
k

¨
h
2
ω
j

j
¸
a
j
(
¯
k, t ) e
i
¯
k·¯ r
−a
÷
j
(
¯
k, t ) e
−i
¯
k·¯ r
¸
(8.70)
¯
B(
¯
k, t ) = i(
¯
k ¯ e
k
)

¨
h

j

j
¸
a
j
(
¯
k, t ) e
i
¯
k·¯ r
−a
÷
j
(
¯
k, t ) e
−i
¯
k·¯ r
¸
. (8.71)
For t < 0 we use the index j = 1, and for t > 0, we use j = 2, in these
expressions. Our main problem is to relate the new operators a
2
and a
÷
2
, to the
old ones, a
1
and a
÷
1
. These operators are different from each other because the
meaning of a photon (or of an elementary excitation of the field) changes with the
refractive index at t = 0.
In order to obtain such a relation we use the continuity conditions for the
fields
¯
D(¯ r, t = 0

) =
¯
D(¯ r, t = 0
÷
),
¯
B(¯ r, t = 0

) =
¯
B(¯ r, t = 0
÷
). (8.72)
Noting that these equalities are independent of ¯ r, we can easily reduce them
to the following relations between the new and the old operators:

ω
1

1
[a
1
(
¯
k) −a
÷
1
(−
¯
k)] =

ω
2

2
[a
2
(
¯
k) −a
÷
2
(−
¯
k)] (8.73)
1

ω
1

1
[a
1
(
¯
k) ÷a
÷
1
(−
¯
k)] =
1

ω
2

2
[a
2
(
¯
k) ÷a
÷
2
(−
¯
k)]. (8.74)
Let us define the parameter
α =

ω
1

1
ω
2

2
=

n
2

1
n
1

2
=

n
1
n
2
. (8.75)
Time refraction 167
Equations (8.74) can be rewritten as
α[a
1
(
¯
k) −a
÷
1
(−
¯
k)] = a
2
(
¯
k) −a
÷
2
(−
¯
k)
a
1
(
¯
k) ÷a
÷
1
(−
¯
k) = α[a
2
(
¯
k) ÷a
÷
2
(−
¯
k)]. (8.76)
By adding and subtracting these two equations, we obtain
a
1
(
¯
k) = Aa
2
(
¯
k) − Ba
÷
2
(−
¯
k)
a
÷
1
(−
¯
k) = Aa
÷
2
(−
¯
k) − Ba
2
(
¯
k). (8.77)
Here we have used the real coefficients A and B, defined by
A =
1 ÷α
2

, B =
1 −α
2

. (8.78)
The reciprocal relations can also be obtained:
a
2
(
¯
k) = Aa
1
(
¯
k) ÷ Ba
÷
1
(−
¯
k)
a
÷
2
(−
¯
k) = Aa
÷
1
(−
¯
k) ÷ Ba
1
(
¯
k). (8.79)
This shows that each field mode existing for t < 0, with a given wavevector
¯
k, will be coupled to two modes existing for t > 0, with wavevectors
¯
k and −
¯
k.
Such a coupling provides an explanation, at the quantum level, of the effect of
time reflection obtained with the classical theory of chapter 6.
8.2.2 Symmetric Fock states
It is particularly important to look at the behaviour of the number states, or
Fock states, when we go through a time discontinuity. They are defined as the
eigenstates of the number operator N
k
≡ a
÷
(
¯
k)a(
¯
k):
a
÷
(
¯
k)a(
¯
k)[n
k
) = n
k
[n
k
). (8.80)
The eigenvalues n
k
of the number operator are the occupation numbers, or
the number of photons in the mode
¯
k. The energy eigenvalues are
!n
k
[H
k
[n
k
) =
¨

k
¸
n
k
[

a
÷
(
¯
k)a(
¯
k) ÷
1
2

[n
k

=
¨

k

n
k
÷
1
2

. (8.81)
The ground state, corresponding to an occupation number equal to zero,
n
k
= 0, is called the vacuum state. It is well known (and can easily be derived
from the above equations) that the eigenvectors [n
k
), corresponding to an arbitrary
excited state, can be derived by applying n
k
times the creation operator a
÷
(
¯
k) to
the vacuum eigenstate [n
k
= 0) ≡ [0
k
).
For normalized eigenvectors we can write
[n
k
) =
1

n'
[a
÷
(
¯
k)]
n
k
[0
k
). (8.82)
168 Quantum theory of photon acceleration
We have shown that, in the process of time discontinuity, the modes
¯
k and −
¯
k
are coupled to each other. It is then useful to introduce the following symmetric
state vectors:
[n, n
/
)
j
≡ [n
k
, n
/
−k
)
j
= [n
k
)
j
[n
/
−k
)
j
. (8.83)
The index j = 1 pertains to the Fock states valid for t < 0, and j = 2 to
the Fock states valid for t > 0. Using equation (8.71), we can then define these
symmetric Fock states in terms of the symmetric vacuum:
[n, n
/
)
j
=
1

n'n
/
'
[a
÷
j
(
¯
k)]
n
[a
÷
j
(−
¯
k)]
n
/
[0, 0)
j
. (8.84)
This expression means that, if we want to establish a relation between some
initial symmetric state [n, n
/
)
1
and some final symmetric state [m, m
/
)
2
, we have
to establish a relation between the symmetric vacuum states, before and after the
time discontinuity, [0, 0)
1
and [0, 0)
2
. This can be achieved by representing the
initial vacuum states [0, 0)
1
in terms of the final state vectors
[0, 0)
1
=
¸
m,m
/
C
m,m
/ [m, m
/
)
2
. (8.85)
It is obvious that such a development has to be perfectly symmetric with
respect to the modes
¯
k and −
¯
k, simply because the vacuum is a state of zero total
momentum. This implies that
C
m,m
/ = C
m
δ
mm
/ . (8.86)
If we apply the destruction operator a
1
(
¯
k) to equation (8.85), and use equa-
tion (8.77), we obtain
a
1
(
¯
k)[0, 0)
1
≡ 0 =
¸
m
C
m
¸
Aa
2
(
¯
k) − Ba
÷
2
(−
¯
k)
¸
[m, m)
2
= C
m
¸
A

m[m −1, m)
2
− B

m ÷1
¸
[m, m ÷1)
2
= (AC
m÷1
− BC
m
)

m ÷1[m, m ÷1)
2
. (8.87)
This leads to the following recurrence relation for the coefficients C
m
:
C
m
=
B
A
C
m−1
=

B
A

m
C
0
. (8.88)
The initial symmetric vacuum states (8.85) can then be represented as
[0, 0)
1
= C
0

¸
m=0

B
A

m
[m, m)
2
. (8.89)
Time refraction 169
Now, we can use the normalization condition in order to determine the re-
maining coefficient:
1
!0, 0[0, 0)
1
= [C
0
[
2

¸
m=0

B
A

2m
= [C
0
[
2
1
1 −(B/A)
2
= 1. (8.90)
This means that
C
0
= e

0

1 −(B/A)
2
(8.91)
where θ
0
is an arbitrary phase.
Using equation (8.89), and assuming that θ
0
= 0, we can write the final
expression for the symmetric vacuum decomposition (8.85) as
[0, 0)
1
=

1 −(B/A)
2

¸
m=0

B
A

m
[m, m)
2
. (8.92)
This result shows that a time boundary at t = 0 will be able to generate,
from an initial vacuum state described by the vector states [0, 0)
1
, a number of
2m photons which will appear in symmetric pairs of modes,
¯
k and −
¯
k. According
to the above equation, the probability of finding such a symmetric Fock state with
2m photons at times t > 0, will be given by
p(m) =
2
!m, m[0, 0)
11
!0, 0[m, m)
2
=
¸
1 −

B
A

2

B
A

2m
. (8.93)
p
(m)
m
Figure 8.1. Probability of photon creation from the vacuum, for B/A = 0.5.
170 Quantum theory of photon acceleration
In order to have a more precise idea of the physical consequences of this
result, let us assume the following plausible situation of a very small time per-
turbation of the refractive index of the medium, such that n
2
= n
1
(1 ÷ δ), with
δ <1:
α =

n
1
n
2
. 1 −
δ
2
. (8.94)
It means that we have (B/A) . (δ/2), which leads to
p(m) .

δ
2

2m
. (8.95)
This gives an order of magnitude estimate for the probability of creation of
photons out of the vacuum due to a sudden change of the refractive index. This
probability decreases with the number of photons m according to a power law.
The possible generation of photons out of the vacuum inside an optical cavity
with a time-dependent dielectric was recently considered in reference [19].
8.2.3 Probability for time reflection
The above results on the transformation of an initial vacuum by a time discontinu-
ity can be generalized to an arbitrary initial state [φ
1
) ≡ [n, n
/
)
1
, not necessarily
perfectly symmetric (n ,= n
/
). The probability of observing a given final state

2
) ≡ [m, m
/
)
2
is determined by
p(m, m
/
) =
2
!m, m
/

1
)!φ
1
[m, m
/
)
2
. (8.96)
In order to remain as close as possible to our previous discussion of the
classical model for time reflection, we will concentrate on initial completely
asymmetric states, where n ,= 0 photons pre-exist in the mode
¯
k and no photons at
all propagate in the opposite direction, which means that n
/
= 0. This corresponds
to an initial photon beam propagating along
¯
k for t < 0. The initial photon state
is [φ
1
) = [n, 0)
1
.
Using equations (8.84, 8.92), we obtain
[n, 0)
1
=
1

n'
[a
÷
1
(
¯
k)]
n
[0, 0)
1
=
1

n'

1 −(B/A)
2

¸
m=0

B
A

m
[a
÷
1
(
¯
k)]
n
[m, m)
2
. (8.97)
We can now use equation (8.77) and, noting that the operators a
÷
2
(
¯
k) and
a
2
(−
¯
k) commute, we obtain, after a binomial expansion
[a
÷
1
(
¯
k)]
n
= A
n
n
¸
r=0
n'
(n −r)'r'
(−1)
n−r

B
A

r
[a
÷
2
(
¯
k)]
n−r
[a
2
(−
¯
k)]
r
. (8.98)
Time refraction 171
Using this in equation (8.97), we arrive at
[n, 0)
1
=
n
¸
r=0

¸
s=0
b
sr
(n)[n ÷s, s)
2
(8.99)
where we have used the new index s = m −r, and the coefficients
b
sr
(n) =
1

n'
A
n

1 −(B/A)
2
n'
(n −r)'r'
(−1)
n−r

B
A

s÷2r
. (8.100)
From this we can calculate the probability for observing at a time t > 0 a
given photon state [n ÷s, s)
2
:
p(n, s) =
2
!n ÷s, s[n, 0)
11
!n, 0[n ÷s, s)
2
=

n
¸
r=0
b
sr
(n)

2
. (8.101)
This result shows that, after the occurrence of a sudden time change at t = 0
of the refractive index of an infinite dielectric medium, there is a probability
p(n, 0) ,= 1 of observing a state [n, 0)
2
with the same number of photons propa-
gating with the same wavevector (but with a shifted frequency).
But there is also a finite probability p(n, s) ,= 0 of observing a number s > 0
of photons propagating in the opposite direction −
¯
k. This gives us the quantum
explanation for the effect of time reflection already described with the classical
theory of chapter 6.
The above discussion can easily be generalized to include the case of coher-
ent states. It is well known that these particular quantum states have the advantage
of tending to a classical field in the limit of large occupation numbers.
By definition, a coherent state for a given field mode
¯
k is
[α) = e
−[α[
2
/2

¸
n=0
α
n

n'
[n
k
). (8.102)
This means that we can, in our symmetric representation, define an initial
coherent state as
[α, 0)
1
= e
−[α[
2
/2

¸
n=0
α
n

n'
[n, 0)
1
. (8.103)
Using equation (8.99) we can write this state vector in terms of the final
states
[α, 0)
1
= e
−[α[
2
/2

¸
n,m=0
n
¸
r=0
α
n
b
sr
(n)

n'
[n ÷s, s)
2
. (8.104)
This expression shows that, in general, an initial coherent state will not
lead to time-transmitted and time-reflected coherent states. This means that the
time discontinuity does not preserve the classical-like properties of the initial
172 Quantum theory of photon acceleration
photon state. However, it is also obvious that, for very large values of the initial
occupation number n ≡ n
k
the transmitted state will be very similar to the initial
one and close to a similar coherent state, propagating in a similar direction but
with a shifted frequency. This problems would deserve a detailed numerical
study [74].
Furthermore, the non-conservation of the classical-like properties of the ini-
tial states has to be related to the creation of squeezed states (states with no
classical counterpart), which is also associated with the time discontinuity. This
problem is not directly related to photon acceleration and will not be discussed
here.
8.2.4 Conservation relations
We conclude the quantum theory of time refraction by briefly discussing the
energy and momentum conservation relations. Let us first consider the total
energy operator
W =

w
k
d
¯
k
(2π)
3
=

¨

k
¸
a
÷
(
¯
k)a(
¯
k) ÷
1
2
¸
d
¯
k
(2π)
3
. (8.105)
For an initial state [n, 0)
1
, the expectation value of this operator is
!W)
1
=
1
!n, 0[W[n, 0)
1
=
¨

1

!n
k
[a
÷
(
¯
k)a(
¯
k)[n
k
) ÷!0[a
÷
(−
¯
k)a(−
¯
k)[0) ÷1

=
¨

1
(n
k
÷1) (8.106)
with ω
1
= [k[c/n
1
.
We have seen that the time discontinuity generates final states [n ÷ s, s)
2
,
with a finite probability p(n, s). The expectation value for the energy operator for
t > 0 will then be
!W)
2
=
2
!n ÷s, s[W[n ÷s, s)
2
=
¨

2

!n ÷s[a
÷
(
¯
k)a(
¯
k)[n ÷s) ÷!s[a
÷
(−
¯
k)a(−
¯
k)[s) ÷1

=
¨

2
(n
k
÷2s ÷1) (8.107)
with ω
2
= [k[c/n
2
.
This means that the energy variation introduced by the time discontinuity of
the medium on the electromagnetic spectrum is
W = !W)
2
−!W)
1
=
¨
hωn
k
÷
¨

2
2s. (8.108)
We see that the energy is not conserved for two distinct reasons. The first one
was already observed in the classical description and corresponds to the frequency
shift ω = ω
2
−ω
1
of the n
k
photons initially existing in the medium. The second
Quantum theory of diffraction 173
one is specific to the quantum nature of light and is due to the generation at t = 0
of oppositely propagating s pairs of photons.
Let us consider next the total momentum operator
P =

p
k
d
¯
k
(2π)
3
=

¨
h
¯
k
¸
a
÷
(
¯
k)a(
¯
k) ÷
1
2
¸
d
¯
k
(2π)
3
. (8.109)
Using the same kind of approach we can calculate the initial and the final
expectation values of this operator. The result is
!P)
1
=
¨
h
¯
kn
k
, !P)
2
=
¨
h
¯
k(n ÷s −s). (8.110)
This means that the total momentum is conserved, as expected: P =
!P)
2
−!P)
1
= 0.
8.3 Quantum theory of diffraction
Let us consider a sharp boundary between two stationary dielectric media with
no dispersion. For simplicity, we assume that the media, with dielectric constants
equal to
1
and
2
, have a boundary at x = 0, and that the propagation is along
the x-axis.
If photons with frequency ω and initial wavevector k
i
are propagating in
medium 1 and interact with this boundary, we can write for the associated electric
field operator, valid in the semi-infinite region x < 0, and for a given polarization
(λ = 1, or 2)
¯
E(x, t ) =
¯
E
i
(x, t ) ÷
¯
E
r
(x, t ) (8.111)
where the incident and the reflected field operators are
¯
E
i
(x, t ) = i

¨

2
1
¸
a
1
(k
i
, t ) e
ik
i
x
−a
÷
1
(k
i
, t ) e
−ik
i
x
¸
¯ e(k
i
) (8.112)
¯
E
r
(x, t ) = i

¨

2
1
¸
a
1
(k
r
, t ) e
ik
r
x
−a
÷
1
(k
r
, t ) e
−ik
r
x
¸
¯ e(k
r
) (8.113)
with k
i
= ωn
1
/c and k
r
= k
i
.
In the second medium (x > 0), we can define the electric field operator
associated with the transmitted wave as
¯
E
t
(x, t ) = i

¨

2
2
¸
a
2
(k
t
, t ) e
ik
t
x
−a
÷
2
(k
t
, t ) e
−ik
t
x
¸
¯ e(k
t
) (8.114)
with k
t
= ωn
2
/c = k
i
(n
2
/n
1
).
174 Quantum theory of photon acceleration
The corresponding magnetic field operators are determined by similar ex-
pressions:
¯
B
i
(x, t ) =
k
i
ω
E
i
(x, t )[¯ e
x
¯ e(k
i
)] (8.115)
¯
B
r
(x, t ) =
k
r
ω
E
r
(x, t )[¯ e
x
¯ e(k
r
)] (8.116)
and
¯
B
t
(x, t ) =
k
t
ω
E
t
(x, t )[¯ e
x
¯ e(k
t
)]. (8.117)
The quantization of the electromagnetic field is based on the assumption that
the field operators satisfy Maxwell’s equations. This means that the boundary
conditions for these operators have to be formally identical to those for the classi-
cal fields. We know, from classical theory, that the components of the electric and
magnetic fields tangent to the boundary between the two media are continuous.
We can then write, for the operators
(
¯
E
i
÷
¯
E
r
) ¯ e
x
=
¯
E
t
¯ e
x
, (
¯
B
i
÷
¯
B
r
) ¯ e
x
=
¯
B
t
¯ e
x
. (8.118)
Using equation (8.117) and noting that the field is polarized in the perpen-
dicular direction ¯ e(k
i
) = ¯ e(k
r
) = ¯ e(k
t
), we obtain
E
i
(0, t ) ÷ E
r
(0, t ) = E
t
(0, t ), B
i
(0, t ) ÷ B
r
(0, t ) = B
t
(0, t ). (8.119)
At this point we notice that the time-dependent destruction and creation
operators in equations (8.113, 8.114) are of the form
a(k, t ) = a(k) e
−iωt
, a
÷
(k, t ) = a
÷
(k) e
iωt
. (8.120)
Equating separately the terms with the same time dependence, we obtain
a
1
(k
i
) ÷a
1
(−k
i
) = α
2
a
2
(k
t
)
a
1
(k
i
) −a
1
(−k
i
) = a
2
(k
t
) (8.121)
with
α =

n
1
n
2
=

1

2

1/4
. (8.122)
From this we obtain
a
2
(k
t
) =
2
1 ÷α
2
a
1
(k
i
), a
1
(−k
i
) = −
1 −α
2
1 ÷α
2
a
1
(k
i
). (8.123)
These expressions can be seen as the Fresnel formulae relating the incidence,
the transmission and the reflection destruction operators. We can easily realize
that the same expressions are valid for the creation operators. We can then derive
Quantum theory of diffraction 175
from (8.123) a more familiar version of the Fresnel formulae, by multiplying them
by exp(−iωt ), and by adding their Hermitian conjugates. We therefore get
E
i
(0, t ) ÷ E
r
(0, t ) = E
t
(0, t ), E
i
(0, t ) − E
r
(0, t ) =
1
α
2
E
t
(0, t ). (8.124)
This is nothing but a different version of equation (8.119). From this, we get
R ≡
E
i
(0, t )
E
r
(0, t )
= −
1 −α
2
1 ÷α
2
, T ≡
E
t
(0, t )
E
i
(0, t )
=

2
1 ÷α
2
. (8.125)
These operator relations are equivalent to equation (8.123), and they are
formally identical to the Fresnel formulae for the classical fields. The same result
could be obtained by using a different method [94].
For comparison with the time reflection case, it is interesting to consider
the energy and momentum conservation relations. Let us start with the number
operator associated with the incident photon states: N
i
≡ a
÷
(k
i
)a(k
i
). By using
equations (8.123) and noting that a
2
(k
t
) and a
1
(−k
i
) commute, we obtain
N
i
= [a
÷
1
(−k
i
) ÷a
2
(k
t
)][a
1
(−k
i
) ÷a
2
(k
t
)]
= a
÷
1
(−k
i
)a
1
(−k
i
) ÷a
÷
2
(k
t
)a
2
(k
t
) = N
r
÷ N
t
. (8.126)
We see from here that, for a generic quantum state of the radiation field,
the expectation value of the incident number operator N
i
is equal to the sum of
the expectation values of the reflected and transmitted photons: n
i
= n
r
÷ n
t
.
In addition, because the photons maintain their frequency upon reflection and
refraction, the total energy is also conserved:
W = (W
r
÷ W
t
) − W
i
=
¨
hω(n
r
÷n
t
) −
¨
hωn
i
= 0. (8.127)
We see that photon creation from a vacuum around a space boundary cannot
exist, in contrast with the case of a time boundary. On the other hand, the total
momentum is not conserved, in agreement with the results of the classical theory:
P = (P
r
÷ P
t
) − P
i
=
¨
h(k
r
n
r
÷k
t
n
t
) −
¨
hk
i
n
i
=
¨
h(k
i
÷k
t
)n
t
=
¨
hk
i

1 ÷
1
α
2

n
t
. (8.128)
Let us conclude with a comment on the more general case of a moving space
boundary. It is clear from the above results that we can follow the approach used
in the classical theory: first, we make a Lorentz transformation to the reference
frame moving with the dielectric boundary. Second, we repeat the above quan-
tization procedure in the moving frame and, finally, we make an inverse Lorentz
transformation back to the rest frame. The result cannot be other than a relation
between the incidence, reflection and transmission field operators identical to that
obtained for the classical fields.
176 Quantum theory of photon acceleration
This leads us to the conclusion that the quantum theory of photon refraction
at a moving boundary will not predict the creation of photons from a vacuum,
because such an effect is already absent in the moving frame (or for a boundary at
rest). This is in contrast with the quantum theory of section 8.2 which shows that
such an effect is present for a time discontinuity, or equivalently, for a boundary
moving with an infinite velocity.
Chapter 9
New developments
The theory of photon acceleration, as presented in its various versions in the
present book, has recently been extended to new and exciting areas of physics.
Two examples are given here.
One example is related to the new field of neutrino interactions with dense
plasmas [13]. Here, the striking similarities between the photon and the neutrino
dispersion relations in a plasma can be explored [68]. A considerable amount
of theoretical work has already been performed by several authors, leading to an
already quite coherent view of the collective neutrino–plasma interactions [106].
The other example concerns the coupling between photons and the gravita-
tional field. Of particular interest is the attempt to describe photon acceleration
processes associated with gravitational waves [72]. Given the extreme difficulty
of this subject, these results have to be received with some caution, and can be
considered as merely tentative. But, we firmly believe that these attempts can be
very positive for the progress of knowledge in this yet quite unexplored area.
9.1 Neutrino–plasma physics
The neutrino interaction with very dense plasmas is of considerable importance in
the early universe and during supernova explosions. The mechanisms for photon
acceleration in non-stationary media can easily be transposed to neutrino–plasma
physics.
It is well known that, in the presence of matter, the neutrino effective mass
is changed due to the weak-current interaction. In particular, the charged current
couples the electron neutrinos with the electrons existing in a dense plasma.
In order to describe this weak coupling we can use the dispersion relation of
a neutrino in a plasma, relating its momentum ¯ p and its energy E [10, 119]:
(E − V)
2
− p
2
c
2
−m
2
ν
c
4
= 0 (9.1)
177
178 New developments
where m
ν
is the (eventually existing) neutrino rest mass and V is an equivalent
potential energy such that
V =

2G
F

0
m
e
2
ω
2
p
≡ gω
2
p
. (9.2)
Here G
F
is the Fermi constant for weak interactions, e and m are the electron
charge and mass and
0
is the vacuum permittivity. Because the coupling constant
g is very small, we can assume, even for extremely dense plasmas, that the
equivalent potential energy V is always much smaller than the total neutrino
energy E.
We also know that the standard theory for electro-weak interactions assumes
a zero value for the neutrino rest mass m
ν
[88], but some recent observations of
solar neutrinos point to the existence of a small rest mass [32]. For that reason,
we opt here for retaining the rest mass in the neutrino dispersion relation. This
relation can be rewritten in terms of the neutrino frequency ω and of the neutrino
wavevector
¯
k (using
¨
h = 1), for a space- and time-varying plasma, as
ω(¯ r,
¯
k, t ) =

k
2
c
2
÷m
2
ν
c
4
÷ gω
2
p
(¯ r, t ). (9.3)
This is indeed very similar to the photon dispersion relation. Here we can
also use the ray equations for the neutrino field, which are formally identical to
the equations for single photon trajectories. In the presence of electron plasma
perturbations moving with a velocity ¯ u, the electron plasma frequency is such
that ω
2
p
(¯ r, t ) ≡ ω
2
p
(¯ η), with ¯ η = ¯ r − ¯ ut .
Following the procedure described in chapter 3, we can establish an invariant
for the neutrino trajectories in the form I = ω −
¯
k · ¯ u. From this we conclude
that the total energy exchange between the neutrino and the background plasma
medium, when it crosses the moving plasma boundary (for instance, during the
collapse of an exploding neutron star) is
E = gω
2
p0
β
1 −β
(9.4)
where β = [u[/c and where we have supposed counter-propagation between the
neutrino and the plasma front.
It should be noticed that, as before for the photon case, this process of energy
exchange between the particle and its background plasma medium is linear and
not resonant, which means that all the particles with the same initial energy will
exchange the same amount of energy with the medium.
This simple approach is based on the classical description of the neutrino,
equivalent to the geometric optics approximation for the photon field. This is
clearly valid as long as the plasma space and timescales are much larger than
those characterizing the neutrino state.
But we can do better and use a quantum mechanical description of this non-
resonant interaction between the neutrinos and the plasma medium. For that
Neutrino–plasma physics 179
purpose, we derive from the above dispersion relation (9.1) a Klein–Gordon wave
equation by making the usual replacement:
¯ p →−i
¨
h

∂¯ r
, E →i
¨
h

∂t
. (9.5)
From this we get

¨
h
2
c
2

2
−m
2
ν
c
4

¨
h
2

2
∂t
2

ψ = 2i
¨
hV
∂ψ
∂t
(9.6)
where ψ is the normalized wavefunction associated with the neutrino field.
We could improve our quantum description of the neutrino field by replacing
this Klein–Gordon equation by a more adequate Dirac equation. But, the present
formulation stays valid as long as we neglect the coupling between different
helicity states.
We are interested in the case where the effective potential V(¯ r, t ) is associ-
ated with a plasma perturbation moving with a velocity ¯ u. If the perturbation is
nearly uniform in the plane perpendicular to its velocity, we can choose the x-axis
as parallel to ¯ u and reduce the wave equation with its one-dimensional version:


2
∂x
2

1
c
2

2
∂t
2

m
2
ν
c
2
¨
h
2

ψ(x, t ) =
2i
¨
hc
2
V(x −ut )

∂t
ψ(x, t ). (9.7)
Let us now use the D’Alembert transformation of variables:
ξ = x −ut, η = x ÷ut. (9.8)
We can easily see that


2
∂x
2

1
c
2

2
∂t
2

= (1 −β
2
)


2
∂ξ
2
÷

2
∂η
2

÷2(1 ÷β)
2

2
∂η∂ξ
. (9.9)
When u = c, or β = 1, this reduces to the expression


2
∂x
2

1
c
2

2
∂t
2

= 4

2
∂η∂ξ
. (9.10)
When written in terms of the new variables, the wave equation (9.7) becomes
¸
(1 −β
2
)


2
∂ξ
2
÷

2
∂η
2

÷2(1 ÷β)
2

2
∂η∂ξ

m
2
ν
c
2
¨
h
2

ψ(ξ, η)
=
2iβ
¨
hc
V(ξ)


∂η


∂ξ

ψ(ξ, η). (9.11)
180 New developments
Because the potential V is independent of the variable η, we can introduce a
Fourier transformation in that variable, such that
ψ(ξ, η) =

ψ
q
(ξ) e
iqη
dq

. (9.12)
This means that we can reduce the wave equation to
¸
(1 −β
2
)


2
∂ξ
2
−q
2

÷2iq(1 ÷β)
2

∂ξ

m
2
ν
c
2
¨
h
2

ψ
q
(ξ)
=
2iβ
¨
hc
V(ξ)

iq −

∂ξ

ψ
q
(ξ). (9.13)
In order to understand the meaning and to test the validity of this form of
wave equation, let us first briefly consider the particular case of neutrino motion in
a vacuum: V(ξ) = 0. In this trivial case, we can perform a Fourier transformation
in the remaining variable, ξ, of the form
ψ
q
(ξ) =

ψ
qp
e
i pξ
dp

. (9.14)
The wave equation then leads to a neutrino dispersion relation of the form
(1 −β
2
)( p
2
÷q
2
) ÷2(1 ÷β
2
) pq ÷
m
2
ν
c
2
¨
h
2
= 0. (9.15)
This strange expression has to be equivalent to the dispersion relation (9.3),
with ω
p
= 0, which means it is equivalent to
ω
2
−k
2
c
2

m
2
ν
c
4
¨
h
= 0. (9.16)
In order to be convinced of such an equivalence, we notice that

∂x
=

∂ξ
÷

∂η
,

∂t
= u

∂η


∂ξ
. (9.17)
This leads to the two relations
k = p ÷q, ω = u( p −q). (9.18)
Using this in equation (9.15), we obtain equation (9.16). It means that this is
indeed the neutrino dispersion relation in a vacuum, written in terms of the more
complicated quantities p and q.
The physical meaning of these new variables is given by (9.18): their sum is
the neutrino momentum and their difference is its energy divided by its velocity.
Neutrino–plasma physics 181
If the neutrino rest mass was made equal to zero, these two quantities, k and ω/u,
would be identical.
We can now return to the non-trivial case of a slowly varying potential,
V(ξ) ,= 0. Here, we can try a WKB solution for the wave equation (9.13) of
the form
ψ
q
(ξ) = ψ
q0
exp

i

ξ
p(ξ
/
) dξ
/

. (9.19)
With this, we can obtain a dispersion relation, which is locally valid in ξ:
(1−β
2
)
¸
p(ξ)
2
÷q
2
¸
÷2(1÷β
2
) p(ξ)q÷
m
2
ν
c
2
¨
h
2
=

¨
hc
V(ξ) [q − p(ξ)] . (9.20)
For this type of solution to be valid, equations (9.18) have to be replaced by
similar ones, valid only locally, and which can be written as
2q = k(ξ) −
ω(ξ)
u
, 2p(ξ) = k(ξ) ÷
ω(ξ)
u
. (9.21)
Here, the quantities k(ξ) and ω(ξ) represent the local values of the
wavenumber and frequency for a given neutrino state characterized by the
quantity q. This quantity remains constant over the entire trajectory. Due to the
existence of such an invariant (which also appeared in the classical theory), we
can easily calculate the change of the neutrino energy when it interacts with a
non-stationary plasma background, in the same way as we have established the
photon frequency shift in a non-stationary optical medium.
In particular, if we assume that the plasma potential tends to zero at infinity,
V(ξ → ∞) = 0, and if in the other side of the discontinuity at ξ = 0 its value is
V
0
, then we can derive from the invariance of q an energy shift equal to the one
given by equation (9.4). But, apart from confirming the classical results for the
energy variation, the quantum description can also reveal a qualitative new effect,
namely the possibility of quantum reflection at the moving plasma boundary.
Using equations (9.21), we can easily obtain a relation between the energy
of the incident and the reflected energy states, E
i
=
¨

i
and E
r
=
¨

r
:
−2qu = k
i
u ÷ω
i
= −k
r
u ÷ω
r
. (9.22)
Assuming that we nearly have kc = ω, we get fromthis the relativistic mirror
relation ω
r
= ω
i
(1 ÷ β)/(1 − β). According to the same equations (9.21), the
values of the quantum number p associated with the incident and the reflected
wave solution, for the same value of q, are different:
2p
i
u = k
i
u −ω
i
. −ω
i
(1 −β), 2p
r
u = −k
r
u −ω
r
. −ω
r
(1 ÷β). (9.23)
Using the above relativistic mirror relation, we obtain
p
r
. p
i
(1 ÷β)
2
(1 −β)
2
. (9.24)
182 New developments
The probability amplitudes for the reflected neutrino state can also be derived
from the wave equation, by using a mode coupling approach. If we assume wave
function solutions of the form
ψ
q
(ξ) =
¸
j =1,2
ψ
qj
(ξ) e
i

ξ
p
j

/
) dξ
/
(9.25)
where ψ
qj
(ξ) are slowly varying amplitudes such that
∂ψ
qj
∂ξ
< p
j
ψ
qj
, (9.26)
this means that we can write

2
∂ξ
2
ψ
qj
. −p
2
j
ψ
qj
÷i p
j
∂ψ
qj
∂ξ
. (9.27)
Using the solution (9.25) in the wave equation (9.13), and assuming that the
local dispersion relation (9.20) stays valid for both the incident and the reflected
particle state, we obtain
¸
j
¸
(1 −β
2
) p
j
(ξ)
¸
∂ψ
qj
∂ξ
e
i

ξ
p
j

/
) dξ
/
.

¨
hc
V(ξ)
¸
l,=j
[q − p
l
(ξ)] ψ
ql
e
i

ξ
p
l

/
) dξ
/
. (9.28)
We know that the neutrino–plasma interaction is very weak, even for very
dense plasmas, due to the extremely small value of the Fermi constant G
F
. This
means that we can assume, in quite general conditions, that the incident mode ψ
q1
is dominant, and that its amplitude is only slightly perturbed by the interaction
with the moving plasma discontinuity.
We can then write the evolution equation for the reflected mode in the para-
metric form
∂ψ
q2
∂ξ
= w(ξ)ψ
q1
(9.29)
where the coupling coefficient is determined by
w(ξ) =
2
¨
hc
β
(1 −β)
2
V(ξ)
p
1
(ξ)
e
i

ξ
( p
1
−p
2
) dξ
/
. (9.30)
This expression shows that the coupling between the initial and the reflected
state is proportional to the velocity of the moving plasma perturbation, or equiv-
alently to the value of β. It is also proportional to the Fourier component of the
potential V(ξ) for p = p
1
− p
2
.
Photons in a gravitational field 183
This can be seen by integrating equation (9.29) and assuming that the initial
value of the reflected state amplitude is zero (or ψ
q2

0
) = 0):
ψ
q2
(ξ) = ψ
q1

ξ
ξ
0
w(ξ
/
) dξ
/
. (9.31)
We can further explore the analogy between the neutrino and the photon
coupling with a non-stationary background plasma by developing a kinetic theory
for the neutrino gas present in the medium. The result is a neutrino kinetic
equation, or neutrino fluid equations, similar to those derived in chapter 4 for
the photon gas [112].
We should also keep in mind that the neutrino gas will react back on the
plasma electrons. This can be described by a neutrino pressure term, similar to
the radiation pressure considered before, which has to be added to the electron
equations of motion. Associated with this pressure term we can likewise define
an equivalent electric charge for the neutrino gas [69, 83, 85, 113].
The result of this mutual coupling between the neutrino and the electron gas
in a plasma leads to the possibility of neutrino Landau damping of relativistic elec-
tron plasma oscillations [73], and to the possibility of transferring a considerable
fraction of the energy of a neutrino beam into the plasma, by exciting electron
plasma waves in the medium [102]. Other surprising results arising from the
study of this collective neutrino plasma are the generation of magnetic fields and
inhomogeneities in the early stages of the universe [105].
9.2 Photons in a gravitational field
The large variety of effects leading to a frequency shift (or to a spectral change) of
the electromagnetic radiation, which we have classified under the name of photon
acceleration, present some analogies with the well-known gravitational frequency
shift occurring when photons escape from regions of strong gravitational fields.
Even if the gravitational effects are described by a completely different theoretical
approach, one can question the possible connections between these gravitational
frequency shifts and our concept of photon acceleration.
Such a question is relevant, not only because it can lead to a deeper and more
global view of photon dynamics, but also because it is known that, at least in some
limit, a gravitational field can be adequately described as a dielectric medium in a
flat space [55, 126]. We can then make the bridge between the photon propagation
in a dielectric medium (in the absence of a gravitational field) as described in this
book, and the photon propagation in the presence of a gravitational field (and in
the absence of a medium).
9.2.1 Gravitational redshift
We showfirst that the Hamiltonian ray theory can be easily extended to include the
influence of the gravitational fields. This can be done by choosing a convenient
184 New developments
definition for the photon frequency, as explained below. This new theoretical
approach will then provide an alternative derivation of the gravitational redshift.
Once this question is clarified, we can eventually use a kinetic photon theory in a
gravitational field, in the same way as we did for plasmas and for other dielectric
media.
Let us consider an electromagnetic wavepacket characterized by the four-
vector k
i
≡ (ω/c,
¯
k), where ω is the mean frequency and
¯
k the mean wavevector.
We know that, in a vacuum and in the absence of a gravitational field, the absolute
value of this four-vector vanishes:
k
i
k
i
= 0. (9.32)
If a wave is propagating in a slowly varying gravitational field we can intro-
duce an eikonal ψ, such that:
k
i
= ∂ψ/∂x
i
. (9.33)
If we replace this in the above condition for the absolute value of k
i
we
obtain the eikonal equation [55]
g
i k
∂ψ
∂x
i
∂ψ
∂x
k
= 0 (9.34)
where g
i k
are the components of the metric tensor.
We will illustrate our ideas by using a very simple metric tensor such that,
for an arbitrary but constant gravitational field, the spatial coordinate system is
Cartesian at a given point and at a given time: g
i k
= −δ
i k
, where the indices
i and k are supposed to take the values 1, 2 and 3, corresponding to the space
coordinates. As concerns the time coordinates, we use g
00
,= 1. More realistic
(but also more complicated) metric tensors will be discussed at the end of this
section.
If we define the frequency of the electromagnetric wave ω as the derivative
of the eikonal function ψ with respect to the time variable
ω = −c
∂ψ
∂x
0
, (9.35)
we obtain from equation (9.34) the following dispersion relation:
ω
2
= k
2
c
2
g
00
(9.36)
where we noticed that g
00
= 1/g
00
.
The same result could be obtained by noting [55, 126] that the influence
of a (quasi-static) gravitational field on the electromagnetic wave propagation
is equivalent to the change of the dielectric constant of a vacuum from
0
to

0
/

g
00
, and a similar change of the magnetic permeability of a vacuum from
µ
0
to µ
0
/

g
00
.
Photons in a gravitational field 185
Introducing a scalar potential V, such that g
00
= 1 ÷2V/c
2
, we can rewrite
this photon dispersion relation as
ω = kc

1 ÷2
V
c
2
. kc ÷
k
c
V. (9.37)
We know that the frequency ω = ω(¯ r,
¯
k, t ) can be used as the photon
Hamiltonian for the ray equations written in the canonical form
d¯ r
dt
=
∂ω

¯
k
.

c ÷
V
c

¯
k
k
d
¯
k
dt
= −
∂ω
∂¯ r
. −
k
c
∂V
∂¯ r
.
(9.38)
For the total time derivative of the Hamiltonian function, we can also write

dt
=
∂ω
∂t
.
k
c
∂V
∂t
. (9.39)
This equation shows that, for a static gravitational field, the frequency ω, as
defined by equation (9.35), is a constant of motion. This will be discussed in more
detail below.
If the potential V is due to a single massive object (for instance a star),
of mass M and radius R, we can write V(¯ r) = −GM/r, for r ≥ R. One
photon emitted at the surface of that star will have the initial wavenumber k
1
such that ω = k
1
c −(k
1
/c)GM/R. This photon will be observed on earth with a
wavenumber k
2
, such that (r →∞) ω = k
2
c.
We obtain from these two expressions of ω the following change in the
wavenumber:
k = k
2
−k
1
= −(k
1
/c)GM/R < 0. (9.40)
This means that the observed wavelength observed on earth will be larger
than the one emitted at the star surface (λ > 0). This is the well-known
gravitational redshift.
In the present formulation, we could be led to the conclusion that the grav-
itational redshift appears only as a shift in wavelength, and not a shift in fre-
quency, because the photon frequency ω, as defined by equation (9.35), is an
invariant. However, this apparent contradiction with the conventional description
of the gravitational redshift disappears if we notice that we are using a different
definition for the photon frequency. It is well known that the conventional view
is based on the photon ‘proper frequency’ ω
τ
, which is defined as the derivative
of the eikonal with respect to the proper time τ (notice that this is the observer
proper time and not the photon proper time):
ω
τ
= −
∂ψ
∂τ
= −
∂ψ
∂x
0
∂x
0
∂τ
=
ω

g
00
. (9.41)
186 New developments
With this local definition of frequency we get ω
τ
= kc everywhere, meaning
that ω
τ
= kc. In this way, the gravitational redshift becomes a frequency shift
as well.
We propose to use the universal time ω, instead of the proper time ω
τ
, be-
cause it coincides with the Hamiltonian function appearing in the photon canoni-
cal equations (9.38). In addition, as discussed below, these equations can easily be
extended to other situations and used to describe several other important effects,
such as the photon bending in the vicinity of a star or the photon acceleration by
a gravitational wavepacket.
9.2.2 Gravitational lens
Let us first discuss a gravitational lens, or photon bending near a massive star.
The massive astrophysical objects that can produce important light bending are
also very often surrounded by dense and warm plasmas. It is then useful to study
the photon equations of motion in a plasma and in a gravitational field. In order
to keep the discussion formally simple, we include the plasma dispersive effects
but neglect the influence of static magnetic fields.
For V = 0, the dispersion relation for photons in a non-magnetized plasma
is simply: ω
2
τ
≡ ω
2
= k
2
c
2
÷ ω
2
p
, where ω
p
is the electron plasma frequency. If
a static gravitational field is also present, we have V ,= 0, and, using the above
metric transformation, we obtain ω
2
= [k
2
c
2
÷ω
2
p
]g
00
, or more explicitly
ω =

(k
2
c
2
÷ω
2
p
)

1 ÷2
V
c
2

. (9.42)
Let us again assume that the static gravitational field is related to a star of
mass M and radius R. We can then replace V(r) = −GM/r, for r > R in
this dispersion relation and use it as the appropriate Hamiltonian function for the
photon canonical equations
ω ≡ ω(¯ r,
¯
k) =

(k
2
c
2
÷ω
2
p
)

1 −
α
r

(9.43)
with α = 2GM/c
2
.
In general, the plasma density will decay exponentially away from the sur-
face of the star (r > R), according to a Boltzmann distribution. This means that
we can rewrite it as
ω
2
p
= ω
2
p0
e
−mV(r)/T
. ω
2
p0

1 −
β
r

(9.44)
where T is the plasma temperature, β = mGM/T and the approximate expres-
sion is valid for a warm plasma such that T ·mV(r).
Photons in a gravitational field 187
The photon equations of motions (9.38) can now be written as
d¯ r
dt
=
c
2
ω

1 −
α
r

¯
k
d
¯
k
dt
= −
∂ω
∂r
¯ e
r
.
(9.45)
In the second of these equations we have used spherical coordinates ¯ r =
(r, θ, φ). For motion in the plane φ = const, we can write
¯
k = k
r
¯ e
r
÷(L/r)¯ e
θ
(9.46)
where L = const is the angular momentum of the photon trajectory. Assuming
that along the photon trajectory we always have (α/r) < 1 and neglecting the
term in 1/r
2
, we obtain an approximate expression for the Hamiltonian:
ω . c

k
2
r
÷
L
2
r
2
÷m
2
eff
c
2

a
r
. (9.47)
Here we have used the photon equivalent mass in a plasma: m
eff
=
ω
p0
/c
2
.We have also introduced the new parameter
a = αω
¸
1 −
ω
2
p0
ω
2

1 −
β
α

. (9.48)
This approximate photon Hamiltonian is formally identical to that of a rel-
ativistic particle with mass m
eff
, moving in a Coulomb field. The corresponding
trajectories are well known from the textbooks [55] and it is not necessary to state
them here.
What is important to notice is that this description of photon motion around
a massive object generalizes the usual description of gravitational bending for the
case of a plasma in a gravitational field. Notice also that diffraction due to plasma
inhomogeneities around a star can dominate over purely gravitational bending if
β <α. In this case we have a . βω
2
p0
/ω. In the absence of a plasma, we recover
the well-known results concerning light bending by a star.
9.2.3 Interaction of photons with gravitational waves
It is already quite surprising and rewarding that we could rederive, from our
simple photon equations of motion, the gravitational redshift and the gravitational
lens effects, and moreover, that we could easily include the plasma refraction
effects. The inclusion of plasma effects is quite interesting because stars or other
massive objects are usually surrounded by warm and dense plasmas which can
eventually influence the total lensing effect.
188 New developments
Let us now consider a much less obvious case, where the static gravitational
field of the previous two cases is replaced by a time-dependent gravitational field.
This is, in general, a difficult problem because in order to solve it, we have to
make explicit use of the Einstein equations of general relativity.
However, if this time-varying field is due to an infinitesimal gravitational
wavepacket, propagating in an otherwise flat space–time, we can still define a
universal frequency and use the dielectric description of gravitation as an accept-
ably good approximation [126]. This means that we can use, as a first-order
approximation, a flat space where the spatial part of the metric tensor is still
Euclidian and the temporal metric component g
00
becomes a function of time.
Let us neglect plasma effects and consider the propagation of a photon in a
vacuum. Equation (9.36) shows that the universal photon frequency ω is no longer
a constant, and the same will happen to the proper photon frequency ω
τ
. We can
then say that, for a time variation g
00
, photon acceleration (or energization) by the
gravitational field can eventually occur.
In order to derive explicit results, let us use the following scalar potential:
V(¯ r, t ) = V
0
÷
˜
V(¯ r − ¯ v
f
t ). (9.49)
Here ¯ v
f
is the velocity of the infinitesimal gravitational field perturbation
and
˜
V its amplitude. In analogy with several other similar occasions, we can
introduce a canonical transformation from (¯ r,
¯
k) to a new pair of variables (¯ η, ¯ p),
such that ¯ η = ¯ r − ¯ v
f
t and ¯ p =
¯
k.
The photon equations of motion become
d¯ η
dt
=
∂ω
/
∂ ¯ p
d ¯ p
dt
= −
∂ω
/
∂ ¯ η
.
(9.50)
The new Hamiltonian associated with these canonical equations is
ω
/
(¯ η, ¯ p) = ω(¯ η) − ¯ v
f
· ¯ p
=
¸
( p
2
c
2
÷ω
2
p
)
¸
1 ÷2
V
0
c
2
÷2
˜
V(¯ η)
c
2
¸
1/2
− ¯ v
f
· ¯ p. (9.51)
Several physical configurations can be studied with these Hamiltonian for-
mulations and all lead, even for very small moving gravitational perturbations,
to strong photon acceleration. This is due to the nearly resonant character of the
photon–gravitational-wave interaction because of their close group and phase ve-
locities. Moreover, such an interaction takes place over extremely large distances,
which enhances the process of photon acceleration.
In order to illustrate this general feature, let us consider two different situ-
ations. In the first one we assume that, near some radiating source (for instance
Photons in a gravitational field 189
a collapsing star), a gravitational shock wave starts to form before being radiated
away, and interacts with the surrounding photons. Strictly speaking, this first
example is not quite compatible with our assumption of a small gravitational
perturbation travelling in a flat space. However, we will keep it here as a first
and illustrative attempt to formulate a new problem that will be interesting for a
future and more accurate solution.
We will retain the flat space approximation and describe the shock front in
the following simple form of a potential perturbation:
˜
V(¯ η) =
˜
V
2
[1 ÷tanh(
¯
k
f
· ¯ η)]. (9.52)
Here
¯
k
f
determines the scale of the moving shock front. The photon equa-
tions of motion (9.50) could be solved numerically. However, because we are
concerned with order of magnitude estimates, based on a very rough description,
it is more interesting to extract some results from the invariance of ω
/
. This will
allow us to discuss the relevant properties of the photon motion.
If the photons are emitted at some point η < 0, their initial frequency ω
1
is
such that
ω
/
= ω
1
− ¯ v
f
·
¯
k
1
(9.53)
where we assume that the shock front velocity ¯ v
f
and the initial photon wavevec-
tor
¯
k
1
are nearly parallel.
After crossing the entire shock front, the photons will acquire a final fre-
quency ω
2
and a final wavevector
¯
k
2
, such that
ω
/
= ω
2
− ¯ v
f
·
¯
k
2
. (9.54)
Equating these two different expressions for the invariant ω
/
, we obtain
ω
2
= ω
1
1 −β(1 −δ)
1 −β(cos θ
2
/ cos θ
1
)(1 −δ ÷
˜
V/c
2
)
(9.55)
where θ
1
and θ
2
are the initial and final angles between the photon wavevector
and the front velocity, and where we have introduced the following auxiliary
parameters:
β =
v
f
c
cos θ
1
, δ =
V
0
c
2
÷
ω
2
p
2k
2
1
c
2
. (9.56)
In order to have an estimate of the total frequency up-shift, let us assume that
θ
1
= 0 and that β ∼ 1. We obtain
ω = ω
2
−ω
1
= ω
1
˜
V
c
2
β
1 −β
. (9.57)
For a perturbation
˜
V/c . 10
−1
and = 1 − β . 10
−8
, this leads to
ω
2
. ω
1
10
7
, which means that the background infrared or visible photons
190 New developments
existing near the exploding star can be accelerated up to the gamma ray frequency
range. Notice that the time necessary for the photon to cross the front will be, in
this case, equal to τ = (2/k
f
) seconds, if k
f
is given in cm
−1
.
As a second example, let us consider photons interacting in a vacuum with
an infinitesimal gravitational wavepacket. In contrast with the previous example
this is now well inside the domain of validity of the dielectric model for the
gravitational field. Because we are considering propagation in a vacuum, we use
ω
p
= 0 and write
˜
V =
˜
V cos(
¯
k
f
· ¯ η). (9.58)
The photon frequency shift will occur when it travels from a region of grav-
itational wave maximum to a wave minimum. In this case, the same kind of
analysis leads to a maximum frequency shift which is twice the value given by
equation (9.57).
But now we have a much smaller field perturbation (
˜
V/c
2
<1) and, because
the photons and the gravitational wave travel almost exactly at the same speed, we
have to assume that the angles θ
1
and θ
2
are small but not exactly zero. Otherwise,
the total frequency shift would be zero.
We can then write
ω . ω
1
4
θ
2
2
˜
V
c
2
. (9.59)
For an extremely small perturbation
˜
V/c
2
. 10
−9
we still obtain photon
accelerations up to the gamma ray energies ω
2
. 4ω
1
10
7
for a very small
angle between the photon wavevectors and the gravitational wavevechtor: θ
2
.
10
−8
radians. The distance necessary for the photons to travel from a maximum
to a minimum of the gravitational wavepacket will be d . k
−1
f
10
−2
parsecs,
which is quite small on a cosmological scale.
We should point out that in these estimates we have neglected the focusing
effects associated with the difference between the initial and the final angles, θ
1
and θ
2
. However these two angles can easily be related, due to the existence of
a second invariant for the photon trajectories, as discussed in chapter 3. This
invariant states the conservation of the transverse photon wavevector, and can be
written as k
1
sin θ
1
= k
2
sin θ
2
.
The result of this two-dimensional effect is to focus the photon trajectories
in the direction of the gravitational wave propagation, leading to a decrease in the
local angles θ and an increase in the effective β. This is due to the fact that the
photon acceleration process increases the parallel photon wavevector, while the
transverse one is kept constant. It then enhances the acceleration process (at the
expense of a larger interaction distance) and broadens the region of photon phase
space with a significant frequency up-shift.
Notice that the value of β = 1 is never exactly attained by these trajectories.
This means that the exactly resonant condition between the photons and the grav-
itational waves, for which no acceleration exists, is an ensemble of zero measure
Photons in a gravitational field 191
in the photon phase space and, quite fortunately, is physically irrelevant to this
problem.
It should be noted that, for arbitrary field amplitudes, the wave solutions of
the Einstein equations generally imply that, apart from g
00
, other time-dependent
components of the metric tensor should also be included. This is equivalent to
saying that, in general, a universal frequency cannot be defined. This will lead to
the necessity of using a different dispersion relation.
9.2.4 Other metric solutions
Let us briefly show how the above description of photon motion in a gravita-
tional field can be improved, by using more accurate metric tensors. First of all,
we should notice that the general dispersion relation in a vacuum (9.34) can be
explicitly written as
g
00
ω
2
−2g

ωk
α
c ÷ g
αβ
k
α
k
β
c
2
= 0 (9.60)
where α = 1, 2, 3, ω = −ck
0
, and we have used the symmetry g

= g
α0
.
If the vacuum was replaced by a plasma medium, we would have to replace
the zero, in the right-hand side of this equation, by ω
2
p
, due to the existence of a
cut-off frequency. We can then use the explicit form of the Hamiltonian function
ω = ω(k
α
, x
α
, t ), where t = x
0
/c in the photon canonical equations (2.97).
These equations (9.60, 2.97) are valid in quite general conditions. Let us
give some examples of metric solutions, which can be physically more accurate
than the simple example used above.
For instance, if we have a weak gravitational field, the interval is determined
by
ds
2
= g
i j
dx
i
dx
j
=

1 ÷
V
c
2

c
2
dt
2

1 −
V
c
2

dl
2
. (9.61)
If the field is created by a star with mass M, we have V = −2GM/r, as
stated above. The corresponding metric tensor will have components g
00
= (1 ÷
V/c
2
) and g
αα
= −(1 − V/c
2
), for α = 1, 2, 3. The dispersion relation in a
plasma will be given by
ω
2
=

ω
2
p

k
2
c
2
g
αα

g
00
. (9.62)
This is not very much different from equation (9.42). Let us now consider a
Schwarzschild type of metric, which is valid for a non-rotating spherical distribu-
tion of mass. The interval can be written in spherical coordinates as
ds
2
= A(r)c
2
dt
2
− B(r) dr
2
−r
2

2
−r
2
sin
2
θ dξ
2
(9.63)
where A(r) and B(r) are functions of the radial coordinate r.
192 New developments
The photon dispersion equation can now be written as
ω
2

ω
2
p
B(r)
=
A(r)
B(r)
k
2
r
÷
A(r)
r
2
k
2

(9.64)
with
k
2

= k
2
θ
÷
k
2
ξ
sin
2
θ
. (9.65)
This expression shows a clear asymmetry between the radial and the perpen-
dicular propagation.
As a final example of a photon dispersion relation in a gravitational field, let
us consider the metric solution of a gravitational wavepacket of the form
ds
2
= c
2
dt
2
−dl
2
÷ f (x, t )(c dt −dx)
2
. (9.66)
We can use
f (x, t ) = a cos(k
0
x
0
÷k
1
x
1
) = a cos(qx −t ) (9.67)
where a and are the gravitational wave amplitude and frequency.
It can easily be shown that the photon dispersion relation in a vacuum takes
the form
(1 − f )

ω
c

2
÷2 f

ω
c

k
|
−(1 ÷ f )k
2
|
−k
2

= 0. (9.68)
Here, we have used the photon wavenumbers parallel and perpendicular to
the direction of the gravitational wave propagation: k
|
= k
1
and k
2

= k
2
2
÷ k
2
3
.
Solving for ω, we obtain
ω = −
f
1 − f
k
|
c ±c

k
2
|
(1 − f )
2
÷
k
2

1 − f
. (9.69)
From here we can see that, for photon propagation perpendicular to the
direction of the gravitational wave, the dispersion relation is simply given by
ω = k

c/(1 − f ), and for parallel photon propagation, we are reduced to the
case of pure vacuum ω = k
|
c. This last expression shows that photons with an
exactly parallel motion cannot be accelerated by the gravitational wave, as already
stated before.
These various examples of metric solutions, and of the corresponding pho-
ton dispersion relations, show that the exchange of energy between the electro-
magnetic (or photon) field and the gravitational field is possible in a variety of
situations. For a stationary gravitational field, our approach leads to a new and
simple derivation of the well-known gravitational redshift and gravitational lens
effects.
For moving gravitational field perturbations, we have shown that the photons
can resonantly interact (in a plasma or in a vacuum) with gravitational shocks and
Mean field acceleration processes 193
with gravitational wavepackets. According to our rough estimates, this interaction
will eventually lead to very high frequency up-shifts, over distances well inside
the cosmological constraints.
This could eventually provide a natural explanation for the recently observed
gamma ray bursts [75]. However, our estimates need to be validated by more
credible metric solutions and more accurate numerical calculations.
Nevertheless, it results from our equations that such an interaction is univer-
sal, in the sense that it affects every photon moving in a gravitational field. We
also believe that, apart from its specific astrophysical implications, our simple
approach points to a new and fundamental coupling mechanism between the
electromagnetic and the gravitational waves which should be investigated in the
future.
Actually, it has been known for quite some time that a gravitational wave
can be Landau damped by a background photon gas [18], but the physical conse-
quences of such damping, which is the statistical counterpart of the acceleration
of photons by a gravitational wave, has not yet been explored.
9.3 Mean field acceleration processes
This section contains a few concluding remarks. We have started with very simple
ideas concerning the processes of time refraction and time reflection, which cor-
respond to a natural extension of the familiar concepts of refraction and reflection
into the space–time domain.
The frequency shift resulting from these basic processes is what we call
photon acceleration. We have shown that it can occur in a large variety of physical
situations, in plasmas, in optical fibres or other optical media, and that it is more
effective when the space–time disturbance of the medium travels with a velocity
nearly equal to that of the photons.
Presently, a significant fraction of the physics community is still reluctant to
talk about photon acceleration and prefers to use other terms such as frequency
shift or phase modulation. This is not, in principle, a big problem, because what
is important in physics is to have an accurate view of the physical processes,
independent of the way they are known. However, the choice of the words is
never completely innocent or arbitrary, and reflects the ideas that we have about
the physical reality.
We have shown in this work that the photons in a medium are subjected to
a force, proportional to the time derivative of the refractive index. Furthermore,
their dynamical interaction with electrostatic waves in a plasma is very similar to
that of a charged particle (an electron or an ion) interacting with the same wave.
In particular, photons can oscillate and can be trapped in the wavefield. On
the other hand, the plasma waves can be damped by photon Landau damping, in
the same way as they are Landau damped by the electrons.
We have also found that an effective photon mass, and an equivalent electric
194 New developments
charge (or a dipole) for the photons in a medium, could be defined. This shows
that our familiar view of photons as particles with no rest mass, and with no
electric charge, can only apply to ‘bare’ photons moving in a vacuum and not to
‘dressed’ photons moving in a background medium. This means that we should
not deny for photons what we accept as true for other particles: that, by receiving
energy from the fields with which they interact, they are energized or, in other
words, they are accelerated.
We can still argue that photons correspond to a particle description of the
electromagnetic field, and that more generally, a wave description is necessary.
But the same is also true for the other fields and for the other particles.
This means that the use of photon acceleration as a genuine physical concept
can lead to a more global view of the physical processes and of the elementary
interactions between the various particles and the various fields. What, at first
sight, is seen as a more fashionable choice of terminology can lead us to ask new
questions and eventually to get a deeper understanding about physics.
As an example, we could say that the equivalent charge of a photon in a
plasma is nothing but a different way of describing the well-known ponderomo-
tive force, or radiation pressure effects. However, the fact that we were able to
isolate the new concept of an equivalent charge led us immediately to the prob-
lem of secondary radiation emitted by accelerated photons, such as the photon
ondulator effects of the photon transition radiation. Other questions related to this
concept, but not considered here, are the possibility of photon bending in a static
magnetic field or the attraction between two parallel photon beams.
We can also explore the idea of photon acceleration as a particular example
of particle acceleration by a time-varying mean field. Such a mean field process
could operate not only with photons and the electromagnetic field, but also with
other particles and with other fields. We saw in this chapter that the ideas of
photon acceleration in an optical medium can be extrapolated to the case of
photons interacting in a vacuum with a gravitational field, or to neutrinos moving
in a dense plasma.
These different mean field processes involve the electromagnetic, the weak
and the gravitational interactions. Similar processes can also be found for the
strong interaction. In particular, the possibility of particle acceleration by the
non-stationary nuclear matter produced by relativistic heavy ion collisions [20] is
presently being explored [96].
Appendix
Derivation of the Wigner–Moyal equation
A.1 Non-dispersive media
We consider transverse electromagnetic fields (∇ ·
¯
E = 0), in a non-dispersive
medium, as described by

2
¯
E −
1
c
2

2
¯
E
∂t
2
=
1
c
2

2
∂t
2

¯
E). (A.1)
Now, we use the notation
¯
E
i

¯
E(¯ r
i
, t
i
) and χ
i
≡ χ(¯ r
i
, t
i
), for i = 1, 2, and
we write


2
1

1
c
2

2
∂t
2
1

¯
E
1
=
1
c
2

2
∂t
2
1
χ
1
¯
E
1
(A.2)


2
2

1
c
2

2
∂t
2
2

¯
E
2
=
1
c
2

2
∂t
2
2
χ
2
¯
E
2
. (A.3)
Let us multiply the first of these equations by
¯
E

2
and the complex conjugate
of the second one by
¯
E
1
. Noticing that, in the absence of losses, the refractive
index is always real and for this reason we can write χ
i
= χ

i
, we obtain, after
subtracting the resulting two equations,
¸
(∇
2
1
−∇
2
2
) −
1
c
2


2
∂t
2
1


2
∂t
2
2

C
12
=
1
c
2


2
∂t
2
1
χ
1


2
∂t
2
2
χ
2

C
12
(A.4)
with
C
12
=
¯
E
1
·
¯
E

2
. (A.5)
For convenience, let us introduce new space and time variables, such that
¯ r =
1
2
(¯ r
1
÷ ¯ r
2
), ¯ s = ¯ r
1
− ¯ r
2
(A.6)
195
196 Derivation of the Wigner–Moyal equation
and
t =
1
2
(t
1
÷t
2
), τ = t
1
−t
2
. (A.7)
In an equivalent manner, we could have stated that
¯ r
1
= ¯ r ÷
¯ s
2
, ¯ r
2
= ¯ r −
¯ s
2
(A.8)
and
t
1
= t ÷
τ
2
, t
2
= t −
τ
2
. (A.9)
Using these variable transformations, we can easily realize that


2
∂t
2
1
χ
1


2
∂t
2
2
χ
2

=

1
4

2
∂t
2
÷

2
∂τ
2


1
−χ
2
) ÷

2
∂t ∂τ

1
÷χ
2
). (A.10)
This expression can be simplified by noting that τ is a fast timescale and t
is a slow timescale, as will become more obvious in the following. Furthermore,
we can assume that the susceptibility χ is a slowly varying function and that its
dependence on the fast time variable τ is negligible. Using (χ
1
÷ χ
2
) . 2χ, we
can then write this equation as
2

∇ · ∇
s


c
2

2
∂t ∂τ

C
12
.
1
c
2

1
−χ
2
)

2
∂τ
2
C
12
÷2
∂χ
∂t

∂τ
C
12
. (A.11)
We know that, by making a Taylor expansion of a function of time f (t ÷τ)
around f (t ), we can obtain
f (t ÷τ) = f (t ) ÷

¸
m=1
1
m'
τ
m

m
∂t
m
f (t ) . f (t ) ÷τ
∂ f (t )
∂t
÷· · · . (A.12)
Introducing an exponential operator, this can be written in a more elegant
and more compact form as
f (t ÷τ) = exp

τ

∂t

f (t ). (A.13)
A power series development of this exponential operator clearly shows that
this is equivalent to equation (A.12). Similarly, a function of the coordinates
f (¯ r ÷ ¯ s) can be expanded around f (¯ r) as
f (¯ r ÷ ¯ s) = exp(¯ s · ∇) f (¯ r). (A.14)
This means that, by performing a double (space and time) Taylor expansion
of the susceptibilites χ
1
and χ
2
around ¯ r and t , we obtain
χ
1
= χ(¯ r ÷ ¯ s/2, t ÷τ/2) = exp

¯ s
2
· ∇ ÷
τ
2

∂t

χ(¯ r, t ) (A.15)
Non-dispersive media 197
and
χ
2
= χ(¯ r − ¯ s/2, t −τ/2) = exp


¯ s
2
· ∇ −
τ
2

∂t

χ(¯ r, t ). (A.16)
This means that the difference between the two values of the susceptibility
of the medium can be written as

1
−χ
2
) = 2 sinh

¯ s
2
· ∇ ÷
τ
2

∂t

χ
= 2

¸
l=0
1
(2l ÷1)'
¸
¯ s
2
· ∇

÷

τ
2

∂t
¸
2l÷1
χ. (A.17)
At this point it is useful to introduce the double Fourier transformation of
C
12
:
C
12
≡ C(¯ r, ¯ s, t, τ) =

d
¯
k
(2π)
3



F(¯ r, t : ω,
¯
k) e
i
¯
k·¯ s−iωτ
. (A.18)
The corresponding inverse transformation is related to the electric field as
follows:
F(¯ r, t : ω,
¯
k) =

d¯ s

dτC(¯ r, ¯ s, t, τ) e
−i
¯
k·¯ s÷iωτ
=

d¯ s


¯
E

¯ r ÷
¯ s
2
, t ÷
τ
2

·
¯
E

¯ r −
¯ s
2
, t −
τ
2

e
−i
¯
k·¯ s÷iωτ
. (A.19)
This quantity is the Wigner function for the electric field. Using this defini-
tion in equation (A.11), we obtain


∂t
÷
c
2
¯
k
ω
· ∇

F ÷
∂ ln
∂t
F = i
ω
2

1
−χ
2
)F. (A.20)
Using equation (A.17), we can write on the right-hand side of this equation

1
−χ
2
)F = 2

¸
l=0
F
(2l ÷1)'
¸
¯ s
2
· ∇

÷

τ
2

∂t
¸
2l÷1
χ. (A.21)
But, from the definition of F, we can also write

m

¯
k
m
= (−i¯ s)
m
F,

m
∂ω
m
= (iτ)
m
F. (A.22)
This means that we can rewrite equation (A.21) as

1
−χ
2
)F = 2

¸
l=0
(−1)
l
(2l ÷1)'
¸
1
2


¯
k
· ∇ = −
1
2

∂ω

∂t
¸
2l÷1
χF. (A.23)
198 Derivation of the Wigner–Moyal equation
Using this result in equation (A.20) we finally obtain


∂t
÷
c
2
¯
k
ω
· ∇

F ÷


∂t

F = −ω( sin F) (A.24)
with
=
1
2

¸

∂¯ r
·


¯
k


∂t

∂ω
¸

. (A.25)
For a linear wave spectrum, we have
F ≡ F(¯ r, t : ω,
¯
k) = F
k
(¯ r, t )δ(ω −ω
k
). (A.26)
Using this in the definition of C
12
, we get
C
12
≡ C(¯ r, ¯ s, t, τ) = e
−iω
k
τ

F
k
(¯ r, t ) e
i
¯
k·¯ s
d
¯
k
(2π)
3
= e
−iω
k
τ
C(¯ r, ¯ s, t, τ = 0). (A.27)
According to equation (A.19), this means that we can define F
k
(¯ r, t ) as the
space Wigner function for the electric field:
F
k
(¯ r, t ) =

C(¯ r, ¯ s, t, τ = 0) e
−i
¯
k·¯ s
d¯ s
=

¯
E(¯ r ÷ ¯ s/2, t ) ·
¯
E

(¯ r − ¯ s/2, t ) e
−i
¯
k·¯ s
d¯ s. (A.28)
A.2 Dispersive media
In a dispersive medium, we have


2

1
c
2

2
∂t
2

¯
E = µ
0

2
∂t
2
¯
P (A.29)
with
¯
P =
0
¯
E −
¯
D.
Returning to the procedure followed in appendix A.1, we can see that equa-
tions (A.2, A.3) have to be replaced by


2
i

1
c
2

2
∂t
2
i

¯
E
i
= µ
0

2
∂t
2
i
¯
P
i
(A.30)
for i = 1, 2.
Dispersive media 199
Again, we can derive from here an evolution equation for the quantity C
12
=
¯
E
1
·
¯
E

2
. The result is
¸
(∇
2
1
−∇
2
2
) −
1
c
2


2
∂t
2
1


2
∂t
2
2

C
12
= µ
0
¸

2
∂t
2
1
(
¯
P
1
·
¯
E

2
) −

2
∂t
2
2
(
¯
P

2
·
¯
E
1
)

. (A.31)
Let us now introduce the space and time variables defined by equa-
tions (A.6, A.7). This equation becomes
2

∇ · ∇
s

1
c
2

2
∂t ∂τ

C
12
= µ
0

1
4

2
∂t
2
÷

2
∂τ
2

¯
P
1
·
¯
E

2

¯
P

2
·
¯
E
1

÷µ
0

2
∂t ∂τ

¯
P
1
·
¯
E

2
÷
¯
P

2
·
¯
E
1

. (A.32)
Here we can introduce the Fourier transformation
¯
E
i

¯
E(¯ r
i
, t
i
) =


i

d
¯
k
i
(2π)
3
¯
E(ω
i
,
¯
k
i
) e
i
¯
k
i
·¯ r
i
−iω
i
t
i
. (A.33)
A similar transformation for the polarization vector is defined by
¯
P
i

¯
P(¯ r
i
, t
i
) =


i

d
¯
k
i
(2π)
3
¯
P(¯ r
i
, t
i
: ω
i
,
¯
k
i
) e
i
¯
k
i
·¯ r
i
−iω
i
t
i
(A.34)
such that
¯
P(¯ r
i
, t
i
: ω
i
,
¯
k
i
) =
0
χ(¯ r
i
, t
i
: ω
i
,
¯
k
i
)
¯
E(ω
i
,
¯
k
i
). (A.35)
The susceptibility of the medium χ(¯ r, t : ω,
¯
k), appearing in this expression,
is assumed to be a slowly varying function of space and time. We can rewrite
the quantity C
12
in terms of the Fourier components of the electric field. But
because this would lead to quite cumbersome expressions, we prefer to introduce
new frequency and wavevector variables, such that
¯ q =
¯
k
1
÷
¯
k
2
,
¯
k =
1
2
(
¯
k
1

¯
k
2
) (A.36)
and
= ω
1
÷ω
2
, ω =
1
2

1
−ω
2
). (A.37)
Equivalently, we could have stated that
¯
k
1
=
¯ q
2
÷
¯
k,
¯
k
2
=
¯ q
2

¯
k (A.38)
200 Derivation of the Wigner–Moyal equation
and
ω
1
=

2
÷ω, ω
2
=

2
−ω. (A.39)
As in the case of the space and time variable transformations (A.6–A.9),
the Jacobian of the new transformations is equal to one: dω
1

2
= ddω and
d
¯
k
1
d
¯
k
2
= d¯ q d
¯
k. In terms of these new variables, the quantity C
12
becomes for-
mally identical to equation (A.18), as it should be, with the quantity F(¯ r, t : ω,
¯
k)
defined now as
F(¯ r, t : ω,
¯
k) =

d

d¯ q
(2π)
3
J(¯ q,
¯
k, , ω) e
i ¯ q·¯ r−it
(A.40)
with
J(¯ q,
¯
k, , ω) =
¯
E(ω ÷/2,
¯
k ÷ ¯ q/2) ·
¯
E(−ω ÷/2, −
¯
k ÷ ¯ q/2). (A.41)
Returning to equation (A.32) and retaining on its right-hand side only the
dominant term, the one proportional to ∂
2
/∂τ
2
, we can write
2
¸
∇ · ∇
s

1
c
2

2
∂t ∂τ

C
12
= −
1
c
2

d

d¯ q
(2π)
3

÷
−η

)
J(¯ q,
¯
k, , ω) e
i ¯ q·¯ r−it
e
i
¯
k·¯ s−iωτ
(A.42)
where, in order to simplify the expression, we have introduced the quantities
η
±
=

ω ±

2

2
χ

¯ r ±
¯ s
2
, t ±
τ
2
: ω ±

2
,
¯
k ±
¯ q
2

. (A.43)
Here, we should notice that [ω[ · [[ because ω is associated with the fast
timescale τ, whereas the frequency is associated with the slow timescale t . In
the same way, we can assume that [
¯
k[ · [ ¯ q[. Developing these quantities around
the values (ω,
¯
k) and (¯ r, t ), we obtain
η
±
. η
0
±

2
∂η
0
∂ω
±
¯ q
2
·
∂η
0

¯
k
±
τ
2
∂η
0
∂t
±
¯ s
2
·
∂η
0
∂¯ r
÷· · · (A.44)
where we have considered that
η
0
= ω
2
χ(¯ r, t : ω,
¯
k). (A.45)
This means that, in equation (A.42), we can use
η
÷
−η

=

∂η
0
∂ω
÷ ¯ q ·
∂η
0

¯
k

÷

τ
∂η
0
∂t
÷ ¯ s ·
∂η
0
∂¯ r

. (A.46)
But we also notice that the quantity η
0
and its derivatives are independent
of and ¯ q. It means that, in equation (A.42), they can be taken out of the
Dispersive media 201
integrations in these variables. This allows us to make the following replacements,
in the same equation:

d

d¯ q
(2π)
3
J(¯ q,
¯
k, , ω) e
i ¯ q·¯ r−it
= i

∂t
F(¯ r, t : ω,
¯
k) (A.47)
and

d

d¯ q
(2π)
3
¯ q J(¯ q,
¯
k, , ω) e
i ¯ q·¯ r−it
= −i∇F(¯ r, t : ω,
¯
k). (A.48)
The result is
2
¸
∇ · ∇
s

1
c
2

2
∂t ∂τ

C
12
= −
1
c
2


d
¯
k
(2π)
3
¸
i

∂η
0
∂ω

∂t

∂η
0

¯
k
· ∇

F(¯ r, t : ω,
¯
k)
÷

τ
∂η
0
∂t
÷ ¯ s · ∇η
0

F(¯ r, t : ω,
¯
k)
¸
e
i
¯
k·¯ s−iωτ
. (A.49)
We can now replace C
12
≡ C(¯ r, t, ¯ s, τ) by its Fourier integral, as defined by
equation (A.18), and we obtain
2i

¯
k · ∇ ÷
ω
c
2

∂t

F(¯ r, t : ω,
¯
k) = −
i
c
2

∂η
0
∂ω

∂t

∂η
0

¯
k
· ∇

F(¯ r, t : ω,
¯
k)

1
c
2

d¯ s


/

d
¯
k
/
(2π)
3

τ
∂η
0
∂t
÷ ¯ s · ∇η
0

C(¯ r, t, ¯ s, τ) e
i(
¯
k
/

¯
k)·¯ s
e
−i(ω
/
−ω)τ
. (A.50)
But, it is also clear that we can write

τ e
−i(ω
/
−ω)τ
dτ = i

∂ω
/

e
−i(ω
/
−ω)τ
dτ = 2πiδ(ω
/
−ω)

∂ω
/
(A.51)
and

¯ s e
i(
¯
k
/

¯
k·¯ s)
d¯ s = −(2π)
3
iδ(
¯
k
/

¯
k)


¯
k
/
. (A.52)
This means that we can finally transform equation (A.51) into a closed dif-
ferential equation for the Wigner function F ≡ F(¯ r, t, ω,
¯
k), which takes the
form


∂t
÷
c
2
¯
k
ω
· ∇

F = −

∂η
0
∂ω
∂ F
∂t

∂η
0

¯
k
· ∇F

÷

∂ F
∂ω
∂η
0
∂t

∂ F

¯
k
· ∇η
0

. (A.53)
202 Derivation of the Wigner–Moyal equation
After rearranging the terms in this equation, we can rewrite it in a more
suitable form:


∂t
÷ ¯ v
g
· ∇

F = −
2
2ω ÷∂η
0
/∂ω

0
F) (A.54)
where is defined by equation (A.25), and
¯ v
g
=
2c
2
¯
k −ω
2
∂/∂
¯
k
2ω ÷ω
2
∂/∂ω
(A.55)
with = 1 ÷χ = 1 ÷η
0

2
.
References
[1] Abdullaev S S 1993 Chaos and Dynamics of Rays in Waveguide Media (London:
Gordon and Breach)
[2] Abdullaev F 1994 Theory of Solitons in Inhomogeneous Media (New York: Wiley)
[3] Alfano R R (ed) 1989 The Supercontinuum Laser Source (New York: Springer)
[4] Ammosov M V, Delone N B and Krainov V P 1985 Tunnel ionization of complex
atoms and of atomic ions in an alternating electromagnetic field Sov. Phys.– JETP
64 1191
[5] Anderson P W 1963 Plasmons, gauge invariance, and mass Phys. Rev. 130 439
[6] Bachelard G 1975 La Formation de l’Esprit Scientifique (Paris: Librairie
Philosophique J Vrin)
[7] Banos A Jr, Mori W B and Dawson J M 1993 Computation of the electric and
magnetic fields induced in a plasma created by ionization lasting a finite interval
of time IEEE Trans. Plasma Sci. 21 57
[8] Bernstein I B 1975 Geometric optics in space- and time-varying plasmas Phys. Fluids
18 320
[9] Bernstein I B and Friedland L 1983 Geometric optics in space and time varying
plasmas Handbook of Plasma Physics vol 1, ed M N Rosenbluth and R Z Sagdeev
(Amsterdam: North-Holland)
[10] Bethe H 1986 Possible explanation of the solar-neutrino puzzle Phys. Rev. Lett. 56
1305
[11] Binette L, Joguet B and Wang J C L 1998 Evidence of Fermi acceleration of Lyman
α in the radio galaxy 1243+036 Astrophys. J. 505 634
[12] BinghamR, DeAngelis U, Amin MR, Cairns RAand MaNamara B1992 Relativistic
Langmuir waves generated by ultrashort laser pulses Plasma Phys. Control.
Fusion 34 557
[13] Bingham R, Dawson J M, Su J J and Bethe H A 1994 Collective interactions between
neutrinos and dense plasmas Phys. Lett. A 193 279
[14] Bingham R, Mendonc¸a J T and Dawson J M 1997 Photon Landau damping Phys.
Rev. Lett. 78 247
[15] Born M and Wolf E 1970 Principles of Optics (Elmsford, NY: Pergamon)
[16] Budden K G 1961 Radio Waves in the Ionosphere (Cambridge: Cambridge Univer-
sity Press)
[17] Chirikov B V 1979 Universal instability of many-dimensional oscillator sytems Phys.
Rep. 52 263
[18] Chesters D 1973 Dispersion of gravitational waves by a collisionless gas Phys. Rev.
D 7 2863
203
204 References
[19] Cirone M, Rzazewski K and Mostowski J 1997 Photon generation by time-dependent
dielectric: a soluble model Phys. Rev. A 55 62
[20] Csernai L P 1994 Introduction to Relativistic Heavy Ion Collisions (Chichester:
Wiley)
[21] Dias J M, Stenz C, Lopes N, Badiche X, Blasco F, Dos Santos A, Silva L O,
Mysyrowicz A, Antonetti A and Mendonc¸a J T 1997 Experimental evidence of
photon acceleration of ultrashort laser pulses in relativistic ionization fronts Phys.
Rev. Lett. 78 4773
[22] Dias J M, Silva L O and Mendonc¸a J T 1998 Photon acceleration versus frequency-
domain interferometry for laser wakefield diagnostics Phys. Rev. ST—Accelerators
Beams 1 031301
[23] Diels J C and Rudolph W 1996 Ultrashort Laser Pulse Phenomena (New York:
Academic)
[24] Dysthe F B, Pecseli H L and Trulsen J 1983 Stochastic generation of continuous wave
spectra Phys. Rev. Lett. 50 353
[25] Elliott R S An Introduction to Guided Waves and Microwave Circuits (Englewood
Cliffs, NJ: Prentice-Hall)
[26] Esarey E, Ting A and Sprangle P 1990 Frequency shifts induced in laser pulses by
plasma waves Phys. Rev. A 42 3526
[27] Esarey E, Joyce G and Sprangle P 1991 Frequency up-shifting of laser pulses by
co-propagating ionization fronts Phys. Rev. A 44 3908
[28] Faro A and Mendonc¸a J T 1982 Interaction of electromagnetic waves with a moving
perturbation in a stationary gas J. Phys. D: Appl. Phys. 16 287
[29] Fermi E 1949 On the origin of the cosmic radiation Phys. Rev. 75 1169
[30] Figueira G, Mendonc¸a J T and Silva L O 1996 Discrete Fermi mapping for photon
acceleration in a time varying plasma cavity Transport, Chaos and Plasma
Physics 2 (Singapore: World Scientific) p 237
[31] Fisher D L and Tajima T 1993 Superluminous laser pulse in an active medium Phys.
Rev. Lett. 71 4338
[32] Fukuda Y et al 1999 Measurement of the flux and zenith-angle distribution of upward
through going muons by Super Kamiokande Phys. Rev. Lett. 82 2644
[33] Goldstein H 1980 Classical Mechanics 2nd edn (Reading, MA: Addison-Wesley)
[34] Gorbunov L M and Kirsanov V I 1987 Excitation of plasma waves by an electromag-
netic wave packet Sov. Phys.– JETP 66 290
[35] Gradshteyn I S and Ryzhik I M 1980 Table of Integrals, Series and Products (New
York: Academic) pp 307, 338
[36] Granatstein V L, Sprangle P, Parker R K, Pasour J, Herndon M, Schlesinger S P and
Seftor J L 1976 Realization of a relativistic mirror: electromagnetic backscattering
from the front of a magnetized relativistic electron beam Phys. Rev. A 14 1194
[37] Haken H 1981 Light, Vol. 1: Waves, Photons, Atoms (Amsterdam: North-Holland)
[38] Harris S E 1997 Electromagnetically induced transparency Phys. Today 50 36
[39] Hau L V, Harris S E, Dutton Z and Behroozi C H 1999 Light speed reduction to 17
metres per second in an ultracold atomic gas Nature 397 594
[40] Hillary M, O’Connel R F, Scully M O and Wigner E P 1984 Distribution functions
in physics: fundamentals Phys. Rep. 106 121
[41] Hizanidis K, Vomvoridis J L, Mendonc¸a J T and Franszeskakis D J 1996 Relativistic
theory of frequency blue-shift of an intense ionizing laser beam in a plasma IEEE
Trans. Plasma Sci. 24 323
References 205
[42] Special issue on Second Generation Plasma Accelerators 1996 IEEE Trans. Plasma
Sci. 24 (2)
[43] Jackson J D 1999 Classical Electrodynamics 3rd edn (New York: Wiley)
[44] Kalluri D K 1999 Electromagnetics of Complex Media: Frequency Shifting by a
Transient Magnetoplasma Medium (Boca Raton, FL: CRC)
[45] Kane D J and Trebino R 1993 Single-shot measurements of the intensity and phase
of an arbitrary ultrashort pulse by using frequency resolved optical grating Opt.
Lett.18 823
[46] Kaw P K, Sen A and Katsouleas T 1992 Nonlinear 1D laser pulse solitons in a plasma
Phys. Rev. Lett. 68 3172
[47] Keldysh L V1965 Ionization in the field of a strong electromagnetic wave Sov. Phys.–
JETP 20 1307
[48] Kim A V, Lirin S F, Sergeev A M, Vanin E V and Stenflo L 1990 Compression and
frequency up-conversion of an ultrashort ionizing pulse in a plasma Phys. Rev. A
42 2493
[49] Klimontovich Yu L 1967 The Statistical Theory of Non-equilibrium Processes in a
Plasma (Cambridge, MA: MIT Press)
[50] Kruer W L 1988 The Physics of Laser Plasma Interactions (Redwood City, CA:
Addison-Wesley)
[51] Kuhn T S 1970 The Structure of Scientific Revolutions 2nd edn (Chicago, IL:
University of Chicago Press)
[52] Kuo S P and Ren A 1993 Experimental study of wave propagation through a rapidly
created plasma IEEE Trans. Plasma Sci. 21 53
[53] Lampe M, Ott E and Walker J H 1978 Interaction of electromagnetic waves with a
moving ionization front Phys. Fluids 21 42
[54] Landau L D 1946 On the vibration of the electric plasma J. Phys. (USSR) 10 25
[55] Landau L D and Lifshitz E M 1996 The Classical Theory of Fields (Oxford:
Butterworth-Heinemann)
[56] Landau L D and Lifshitz E M 1994 Mechanics 3rd edn (Oxford: Butterworth-
Heinemann)
[57] Landau L D and Lifsjitz E M 1984 Electrodynamics of Continuous Media 2nd edn
(Oxford: Butterworth-Heinemann)
[58] Leonhardt U 1997 Measuring the Quantum State of Light (Cambridge: Cambridge
University Press)
[59] Lichtenberg A J and Lieberman M A 1983 Regular and Stochastic Motion (New
York: Springer)
[60] Lotz W 1967 An empirical formula for the electron impact ionization cross-section
Z. Phys. 206 205
[61] Loudon R 1983 The Quantum Theory of Light (Oxford: Clarendon)
[62] Mendonc¸a J T 1979 Nonlinear transition radiation in a plasma Phys. Rev. Lett. 43 354
[63] Mendonc¸a J T 1979 Nonlinear interaction of wavepackets J. Plasma Phys. 22 15
[64] Mendonc¸a J T 1985 Beat-wave excitation of plasma waves J. Plasma Phys. 34 115
[65] Mendonc¸a J T, Hisanidis K, Frantzeskakis D J, Silva L O and Vomvoridis Y L 1997
Covariant formulation of photon acceleration J. Plasma Phys. 58 647
[66] Mendonc¸a J T and Silva L O 1994 Regular and stochastic acceleration of photons
Phys. Rev. E 49 3520
[67] Mendonc¸a J T and Silva L O 1995 Mode coupling theory of flash ionization IEEE
Trans. Plasma Sci.
206 References
[68] Mendonc¸a J T, Bingham R, Shukla P K, Dawson J M and Tsytovich V N 1995
Interaction between neutrinos and nonstationary plasmas Phys. Lett. A 209 78
[69] Mendonc¸a J T, Silva L O, Bingham R, Tsintsazde N L, Shukla P K and Dawson J M
1998 Effective charge of photons and neutrinos in a plasma Phys. Lett. A 239 373
[70] Mendonc¸a J T, Hizanidis K and Franszeskakis D J 1998 Method for generating
tunable high-frequency harmonics in periodically modulated ξ
(2)
materials Opt.
Commun. 146 245
[71] Mendonc¸a J T, Shukla P K, Bingham R and Tsintsadze N L 1999 Transition radiation
of photons and neutrinos at a plasma boundary Phys. Scr. at press
[72] Mendonc¸a J T, Shukla P K and Bingham R 1998 Photon acceleration by gravitational
waves Phys. Lett. A 250 144
[73] Mendonc¸a J T, Shukla P K, Bingham R, Dawson J M and Silva L O 1999 Neutrino
Landau damping and collective neutrino–plasma processes J. Plasma Phys. at
press
[74] Mendonc¸a J T, Guerreiro A and Martins A M 1999 Quantum theory of time
refraction, in preparation
[75] Meszaros P, Laguna P and Rees M J 1993 Gas dynamics of relativistically expanding
gamma-ray burst sources: kinematics, energetics, magnetic fields, and efficiency
Astrophys. J. 415 181
[76] Misner C W, Thorne K S and Wheeler J A 1970 Gravitation (San Francisco:
Freeman)
[77] Mori W B 1987 On beat-wave excitation of relativistic plasma waves IEEE Trans.
Plasma Sci. 15 88
[78] Mori W B 1991 Generation of tunable radiation using an underdense ionization front
Phys. Rev. Lett. 44 5118
[79] Mori W B, Katsouleas T, Dawson J M and Lai C H 1995 Conversion of dc fields in
a capacitor array to radiation by a relativistic ionization front Phys. Rev. Lett. 74
542
[80] Moyal J E 1949 Quantum mechanics as a statistical theory Proc. Camb. Phil. Soc. 45
99
[81] Nayfeh A 1993 Introduction to Perturbation Theory (New York: Wiley) p 253
[82] Nicholson D R 1983 Introduction to Plasma Physics (New York: Wiley)
[83] Nieves J F and Pal P B 1994 Induced charge of neutrinos in a medium Phys. Rev. D
49 1398
[84] Olver F W J 1965 Handbook of Mathematical Functions ed M Abramowitz and
I A Stegun (New York: Dover) p 355
[85] Oraevskii V N, Semikov V B and Smorodinskii Ya A 1994 Electrodynamics of
neutrinos in a medium Phys. Part. Nucl. 25 1063
[86] Pais A 1982 ‘Subtle is the Lord. . .’ the Science and Life of Albert Einstein (Oxford:
Oxford University Press)
[87] Pasour J A, Granatstein V L and Parker R K 1997 ‘Relativistic mirror’ experiment
with frequency tuning and energy gain Phys. Rev. A 16 2441
[88] Quang Ho-Kim and Pham Xuan Yem 1998 Elementary Particles and Their Interac-
tions (Berlin: Springer)
[89] Rax J M and Fisch N J 1993 Ultrahigh intensity laser–plasma interaction: a
Lagrangian approach Phys. Fluids B 5 2578
[90] Rivlin L A 1997 Photons in a waveguide (some thought experiments) Phys. Usp. 40
291
References 207
[91] Rosa C C, Oliveira e Silva L, Lopes N and Mendonc¸a J T 1998 Formation of ioniza-
tion fronts by intense short laser pulses propagating in a neutral gas Superstrong
Fields in Plasmas ed M Lontano, G Mourou, F Pegoraro and E Sindoni (Verena:
American Institute of Physics) pp 135–40
[92] Rosenbluth M N and Liu C S 1972 Excitation of plasma waves by two laser beams
Phys. Rev. Lett. 29 701
[93] Sagdeev R Z and Galeev A A 1969 Nonlinear Plasma Theory (New York: Benjamin)
[94] Santos D J and Loudon R 1995 Electromagnetic field quantization in inhomogeneous
and dispersive one-dimensional systems Phys. Rev. A 52 1538
[95] Savage R L Jr, Joshi C and Mori W B 1992 Frequency upconversion of electromag-
netic waves upon transmission into an ionization front Phys. Rev. Lett. 68 946
[96] Seixas J and Mendonc¸a J T 1999 Moving boundary effects in QCD plasma formation,
in preparation
[97] Semenova V I 1967 Reflection of electromagnetic waves from an ionization front
Sov. Radiophys. Quantum Electron. 10 599
[98] Silva L O and Mendonc¸a J T 1992 Asymmetric pendulum Phys. Rev. A 46 6700
[99] Silva L O and Mendonc¸a J T 1996 Photon acceleration in superluminous and
accelerated ionization fronts IEEE Trans. Plasma Sci. 24 316
[100] Silva L O and Mendonc¸a J J 1996 Full wave theory of photon acceleration in a
cavity IEEE Trans. Plasma Sci. 24 503
[101] Silva L O, Mendonc¸a J T and Figueira G 1996 Full wave theory of Fermi photon
acceleration Phys. Scr. T 63 288
[102] Silva L O, Bingham R, Dawson J M, Mendonc¸a J T and Shukla P K 1999 Neutrino
driven streaming instabilities in a dense plasma Phys. Rev. Lett. 83 2703
[103] Silva L O and Mendonc¸a J T 1999 Where is phase in self-phase modulation? to be
submitted
[104] Shukla P K, Tsintsadze N L, Mendonc¸a J T and Stenflo L 1999 Equivalent charge
of photons in magnetized plasmas Phys. Plasmas 6 627
[105] Shukla P K, Bingham R, Mendonc¸a J T and Stenflo L 1998 Neutrinos generating
inhomogeneities and magnetic fields in the early universe Phys. Plasmas 5 2815
[106] Shukla P K, Silva L O, Bethe H, Bingham R, Dawson J M, Stenflo L, Mendonc¸a J T
and Dalhed S 1999 The physics of collective neutrino–plasma interactions Plasma
Phys. Control. Fusion 41 A699
[107] Solokov A A, Mora P and Chessa P 1999 Simulation of photon acceleration in a
plasma wakefield Phys. Plasmas 6 503
[108] Stix T H 1962 The Theory of Plasma Waves (New York: McGraw-Hill)
[109] Tajima T and Dawson J M 1979 Laser electron accelerator Phys. Rev. Lett. 43 267
[110] Tsai C S and Ault B A 1967 Wave interactions with moving boundaries J. Appl.
Phys. 38 2106
[111] Tsintadze N L and Mendonc¸a J T 1998 Kinetic theory of photons in a plasma Phys.
Plasmas 5 3609
[112] Tsintsadze N L, Mendonc¸a J T and Tsintsadze L N 1998 Kinetic formulation of
neutrino–plasma interactions Phys. Plasmas 5 3512
[113] Tsintsadze N L, Mendonc¸a J T and Shukla P K 1998 Kinetic theory of induced
electric charges of intense photons and neutrinos in an unmagnetized plasma Phys.
Lett. A 249 110
[114] Tsytovich V N 1970 Nonlinear Effects in Plasmas (New York: Plenum)
208 References
[115] van Ojik R, Roettgering H J A, Miley G K and Hunstead R W 1997 The gaseous
environments of radio galaxies in the early universe: kinematics of the Lyman α
emission and spatially resolved HI absorption Astron. Astrophys. 317 358
[116] Walls D F and Milburn G J 1995 Quantum Optics (Berlin: Springer)
[117] Weinberg S 1962 Eikonal method in magnetohydrodynamics Phys. Rev. 126 1899
[118] Wilks S C, Dawson J M, Mori W B, Katsouleas T and Jones M E 1989 Photon
accelerator Phys. Rev. Lett. 62 2600
[119] Wolfenstein L 1978 Neutrino oscillations in matter Phys. Rev. D 17 1369
[120] Wood W M, Siders C W and Downer M C 1993 Femtosecond growth dynamics
of an underdense ionization front measured by spectral blueshifting IEEE Trans.
Plasma Sci. 21 20
[121] Xu X, Yugami N and Nishida Y 1997 Frequency upshift in the interaction of a high
power microwave with an inhomogeneous plasma Phys. Rev. A 55 3328
[122] Yablonovitch E 1974 Self-phase modulation and short-pulse generation from laser-
breakdown plasma Phys. Rev. A 10 1888
[123] Xiong C, Yang Z and Liu S 1995 Frequency shifts by beam-driven plasma wake
waves Phys. Lett. A 43 203
[124] Zaslavsky G M 1988 Chaos in Dynamical Systems (New York: Harwood)
[125] Zaslavsky G M, Sagdeev R Z, Usikov D A and Chernikov A A 1991 Weak Chaos
and Quasi-Regular Patterns (Cambridge: Cambridge University Press)
[126] Zeldovich Ya B and Novikov I D 1971 Relativistic Astrophisics, Vol 1: Stars and
Relativity (Chicago, IL: University of Chicago Press)
Glossary
Dark source
This is the name of an experimental configuration where an ionization front radi-
ates high-frequency electromagnetic waves by moving across a region filled with
a static and spatially periodic electric field. This field can be produced by an array
of capacitors with alternate polarities. In the language of photon acceleration, this
radiation process can be seen as an acceleration of photons initially having zero
frequency.
Flash ionization
When an intense laser pulse is focused into a region filled with a neutral gas it
photoionizes the gas around the focal region and creates a plasma. This process
of nearly instantaneous plasma creation, with the corresponding frequency up-
shift of the laser frequency, is usually referred to as flash ionization. It should
be contrasted with the ionization front case, where the boundaries of the plasma
region can move with relativistic velocities. In some sense, flash ionization is a
limiting case of an ionization front where the velocity of the front tends to infinity,
because the whole plasma is created in the focal region simultaneously.
Frequency up-shift
Frequency up-shift (or the equally possible, but usually less exciting, frequency
down-shift) is another name for photon acceleration. It comes naturally from
the full-wave description of wave propagation in non-stationary media, where the
concept of the photon is not explicitly considered. However, photon acceleration
is a more suggestive concept, if we use geometric optics or quantum theory, or if
we compare it with similar effects in other physical domains.
209
210 Glossary
Induced phase modulation
This is the frequency shift of a probe laser pulse which propagates in a nonlinear
optical medium in the presence of a second laser pulse. If the intensity of this
second pulse is non-negligible, it produces a time-dependent phase modulation
of the first pulse. If the two pulses have similar intensities they can mutually
modulate their phases, and both shift their frequencies, thus leading to cross-
phase modulation. These two processes of induced and cross-phase modulation
can be seen as particular examples of photon acceleration.
Ionization front
This is the boundary between two distinct regions of a gaseous medium. On one
side of the boundary the gas is neutral, and on the other side, the gas is ionized
(a plasma). If the plasma state is created by an intense laser pulse propagating
across a neutral gas, such a boundary can move with relativistic velocities, even if
the particles in these two media stay nearly at rest. The interest of the ionization
front is that it creates a moving discontinuity of the refractive index, leading to
the acceleration (or frequency shift) of photons which interact with this moving
boundary.
Landau damping
In a famous paper, Landau was able to prove that an electron plasma wave can
be damped in a collisionless plasma. This damping process is due to the resonant
interaction of the wave with the plasma electrons travelling with a velocity nearly
equal to the wave phase velocity. Of course, for electron plasma waves with
relativistic phase velocities (such as those associated with laser wakefields) the
electron Landau damping is negligible because the number of resonant electrons
tends to zero. However, in this case, the electron plasma waves can still be Landau
damped, not by the electrons, but by the resonant photons. This surprising result
is a consequence of the acceleration of those many photons resonantly interacting
with a relativistic plasma wave. Its existence reveals the possibility of energy
exchange between the non-stationary medium (in this case a plasma perturbed by
the electron plasma wave) and the photon gas.
Magnetic mode
When an incident electromagnetic wave interacts with a moving ionization front
it excites, not only the expected reflected and transmitted waves, but also a new
kind of wave which is (in the laboratory frame of reference) a purely magnetic
perturbation with zero frequency. This is called the magnetic mode. This mode
has not yet been observed in experiments, but the theoretical arguments clearly
Glossary 211
point to its existence. In the language of photon acceleration we could say that
the creation of a magnetic mode corresponds to a deceleration of the incident
high-frequency photons down to zero-frequency photons.
Photon
The word photon has two different meanings, depending on whether we are re-
ferring to classical or quantum theory. In classical theory, it refers to an elec-
tromagnetic wavepacket with a length and time duration much shorter than the
characteristic space and timescales of the medium. In the frame of geometric
optics, the photon trajectories are determined by the ray equations. In quantum
theory, it refers to the elementary excitation of the electromagnetic field, and can
be identified with a quantum of electromagnetic energy and momentum.
Photon acceleration
This is the total energy variation of a photon when it moves in a non-stationary
medium. The change in energy is equivalent to a shift in the photon frequency,
with a change in the (group) velocity, which justifies the name. It should be
noticed that the total energy of a photon can be divided in two parts: a kinetic
part, associated with its wavevector, and a rest-mass part, associated with the
polarization of the medium. In inhomogeneous but stationary media, the kinetic
energy of the photon is not conserved, even if its total energy stays constant. In
this case, we cannot talk about photon acceleration. Such an acceleration can only
occur in non-stationary media.
Photon effective mass
Aphoton always moves in a material mediumwith a velocity less than the velocity
of light in vacuum c. For that reason, we can say that the photon behaves like a
particle with an effective mass. If the medium is an isotropic plasma, the photon
mass is simply proportional to the electron plasma frequency. In general, it will be
determined by the susceptibility of the medium. We should notice that, when we
say a photon in a medium we are refering to a complex entity, which we can call
a dressed photon, in contrast with the bare photon which only exists in a vacuum.
Photon equivalent charge
The electrons in a plasma are pushed away fromthe regions of intense electromag-
netic energy (or regions containing a large photon density) by the ponderomotive
force or radiation pressure. Alternatively, we can say that the photons behave as if
they had a negative electric charge, thus pushing away the other negative charges
212 Glossary
(the electrons). This property can only be described with nonlinear equations, in
contrast with the photon effective mass, which is intrinsically linear. We use here
a different adjective (equivalent, instead of effective) in order to keep in mind the
different nature of the photon charge and mass, one being linear and the other
nonlinear. We can also derive a similar concept for a non-ionized medium. For
instance, we can associate an electric dipole with the photons moving along an
optical fibre. The main difference with respect to the plasma case is the non-
existence of free electrons in the fibre.
Self-blueshift
The process of ionization of a neutral gas by an incident laser pulse produces a
temporal change in the refractive index leading to an upshift, or a blueshift, of
the laser frequency. This is a particular case of photon acceleration where the
acceleration process is due to the incident photon beam itself.
Self-phase modulation
When an intense and short laser pulse propagates in a nonlinear optical medium
(for instance along an optical fibre) it produces a nonlinear modulation of its
own phase, thus leading to a significant frequency shift. The resulting spectral
width can be significantly larger than that of the initial laser pulse and the final
light pulses are sometimes referred to as supercontinuum radiation. Self-phase
modulation can also be seen as a photon acceleration process.
Time reflection
A sudden change in the refractive index of a dielectric medium leads to the
appearance of a wave with the same wavenumber as the pre-existing wave but
propagating in the opposite direction. Fresnel formulae for time reflection, similar
to the well-known Fresnel formulae for the usual (space) reflection, can be found.
Time refraction
A sudden change in the refractive index of a dielectric medium leads to a change
in the frequency of the photons moving in this medium, but maintaining their
wavevector. This is similar, but in some sense symmetric, to the case of the
usual (space) refraction, where the refractive index changes in space and not in
time, leading to the conservation of the photon frequency but a variation of the
wavevector. A very simple and general Snell’s law for time refraction can be
found.
Glossary 213
Wakefield
We call the laser wakefield, or more simply the wakefield, the fast electron plasma
wave created by a short and intense laser pulse propagating in a plasma. The phase
velocity of this electron plasma wave is nearly equal to the group velocity of the
laser pulse.
Index
accelerated front, 39–42
accelerators, 1, 44
adiabatic process, 4
Alfven waves, 63
angle of,
incidence, 19
temporal incidence, 14
transmission, 19
anti-Stokes sidebands, 95
area preserving map, 61
asymmetric,
pendulum, 46
states, 170
astrophysical,
objects, 1, 186
problems, 2, 63
atomic number, 83
atomic transitions,
detuning, 65
averaged frequency, 93
‘bare’ particle, 97, 194, 211
beat-wave, 43
Bessel functions, 154
binomial expansion, 170
blueshift, 63, 67
Boltzmann distribution, 186
bounce frequency, 48
boundary,
conditions, 174
layer, 12
moving, 175
bremsstrahlung, 3, 106, 107
broadband radiation, 4, 87
canonical,
equations, 188, 192
transformations, 15, 18, 39, 92
canonical variables, 10, 52
covariant form, 29
capacitors, 136, 137, 209
elongated, 65
Cartesian coordinates, 184
Cauchi principal part, 111
cavity, 57, 140, 152
centro-symmetric media, 118
Cherenkov radiation, 3
Chirikov criterion, 53
chirp, 67, 78, 90, 93, 94, 95
classical theory of radiation, 69
coarse-graining, 69
coherent states, 171, 172
commutation relations, 159, 160
completely ionized atoms, 83
conservation,
equation, 67
rules, 115
constants of motion, 15, 16, 18
continuity,
conditions, 166
equation, 84
co-propagating photons, 36, 40
cosmical scale, 190, 193
cosmic ray acceleration, 57
Coulomb,
field, 187
gauge, 158
coupling coefficients, 152, 182
counter-propagating photons, 36, 40
215
216 Index
covariant formulation, 6, 49, 54
critical angle, 21
curved space–time, 29
cut-off,
condition, 57, 128
frequency, 38, 191
cyclotron resonance, 64
D’Alembert transformation, 179
dark source, 7, 132, 136, 209
density profile, 57
parabolic, 60
deeply trapped trajectories, 47
Debye length, 43
dielectric,
constant, 9, 157, 161, 165, 173,
184
function, 89, 120
medium, 31, 115, 123, 157, 183
diffraction, 72, 74
diffusion, 32
Dirac equation, 179
discrete map, 57, 59
dispersion relation, 9, 43, 50, 52, 54,
81, 130, 135, 164, 184, 186,
191, 192
general form, 69
linear, 71
of electron plasma waves, 109,
113
dispersive medium, 9
Doppler shift, 43, 58
‘dressed’ particle, 97, 194, 211
dynamical systems, 53
early universe, 177, 183
effective potential, 179
eigen
frequencies, 152
functions, 150
modes, 140, 150, 152, 167
eikonal equation, 23, 184
Einstein’s
energy formula, 25
equations, 188, 191
summation rule, 27
theory of relativity, 19
electromagnetic,
cavity, 57
wavepacket, 97
induced transparency, 65
electron,
acceleration, 44
beam, 149
charge, 9
cyclotron frequency, 63
density, 9
fluid equations, 98, 108, 162
impact ionization, 83
mass, 9
plasma frequency, 9, 24, 33, 39,
128, 178
plasma wave, 6, 43, 97, 108, 112,
126, 152, 164, 183, 193,
211, 210, 213
temperature, 43, 162, 186
thermal velocity, 43, 108, 162,
165
electron-impact ionization, 32
electrostatic waves, 43
energy,
cascade, 149
conservation law, 86–87
density, 69
eigenvalues, 167
envelope equation, 89
extraordinary mode, 64
Faraday equation, 98
Fermi,
acceleration, 6, 57, 152–156
constant, 178, 182
mapping, 61
field,
configuration, 9
operators, 7
phase, 127
quantization, 157–165
Index 217
fixed points, 45–47, 52
stability, 60, 61
flash ionization, 126, 143, 149, 209
flat space–time, 27
Fock states, 167–169
force acting on the photons, 109,
193
forced linear oscillator, 100
form function, 37
Fourier transformation, 119, 134,
180, 197, 199
four-vectors, 27, 28, 29, 49, 163,
184
frequency,
cut-off, 9, 25
local, 8
frequency shift, up-shift, 4, 39, 40,
58, 64, 146, 171, 172, 181,
190, 193, 209, 209
maximum value, 34, 47, 94
Fresnel formulae, 7, 123, 124, 212
generalized, 126–128
for time reflection, 126
quantum, 174, 175
front,
acceleration, 41
profiles, 33
velocity, 40
width, 33
full wave models, 5, 209
gamma rays, 189, 192
Gaussian,
distribution, 77, 78
pulse, 33, 83, 90, 94, 106
geometric optics, 5, 69, 71, 123,
178, 209
generating function, 16, 18, 39
glass, 63
gravitational,
field, 7, 29, 177, 183, 194
lens, 186, 192
redshift, 183–186, 192
shock, 189, 192
wavepackets, 186, 187, 193
group velocity, 8, 10, 13, 23, 25,
61, 74, 77, 81, 90, 105, 112,
113, 213
growth rates, 114
Hamiltonian, 6, 10, 11, 23, 33, 40,
47, 50, 52, 64, 92, 156, 185,
186, 187, 188, 191
covariant, 49
operator, 159, 160
theory, 183
harmonic generation, 100, 117, 122
Heaviside function, 12, 133, 147,
152, 166
heavy ion collisions, 194
helicity states, 179
Higgs boson, 165
hyperbolic tangent model, 12
ideal gas, 85
inhomogeneous medium, 11
instabilities,
decay, 115
dark source radiation, 139
photon beam-plasma, 113–115
wave, 112
intensity profile, 66
interactions,
elementary, 194
gravitational, 3
laser–plasma, 6
laser–matter, 1
neutrino–plasma, 7, 177
strong, 3, 194
weak, 3, 177, 178
interaction time, 42
intermittency, 57
invariants, 28, 46, 47, 39, 50, 178,
181, 185, 189, 190
ionization,
flash, 5, 36
front, 4, 31-39, 67, 82–84, 128,
133, 138, 146, 148, 141,
209, 210
218 Index
probabilities, 82, 83
single step, 33, 38
Jacobian, 60, 200
matrix, 61
kinetic equation, 67, 69, 74, 75, 83,
91
Klein–Gordon equation, 163, 179
Klimontovich equation, 67–69
Lagrangian, 6, 23, 25, 26
equation, 24
Landau damping, 210
electron, 43, 55, 109, 111, 112,
162, 193
of gravitational waves, 193
neutrino, 183
photon, 1, 6, 97, 108, 111, 112,
114, 193
Laplace transformation, 110
large-scale stochasticity, 53, 61
laser beating, 44
laser pulse,
CO
2
, 5
intense, 1, 31, 43, 65, 67, 112,
140, 210, 212, 213
short, ultra-short, 1, 65, 74, 76,
78, 87, 101, 103, 137
ultraviolet, 5
linear map, 61
line broadening, 155
L-map, 57
L-mode, 64
local wavevector, 8
Lorentz,
force, 98
transformation, 17, 127, 128, 130,
132, 175
lossless medium, 9
lower hybrid frequency, 64
Lymann α line, 63
magnetic,
clouds, 57
mode, 7, 126, 129, 130, 141, 151,
210
magnetoplasmas, 11, 63
massive vector field, 163, 165
Maxwell’s equations, 69, 70, 71, 74,
83, 88, 115, 124, 125, 129
matrix transformation, 61
Mathieu equation, 139, 153
mean field acceleration, 3, 193
mean force, 85
method of the variation of
parameters, 121
metric tensor, 184
microscopic density distribution, 68
microwave,
cavity, 4, 148
experiments, 36
Minkowski’s metric tensor, 27
mirrors, 57
mode coupling, 7, 123, 142, 144,
148, 152
momentum conservation law, 85
mono-energetic photon beam, 113
moving,
boundaries, 15, 31, 57
magnetic field perturbations, 63
neutrino, 7, 194
dispersion relation, 177, 178, 180
effective mass, 177
fluid equations, 183
kinetic equation, 183
reflection, 181
neutron star, 178
Newtonian equation, 23
nonlinear,
coefficient, 90
current, 99, 106, 107
dynamics, 54
optical medium, 88, 210
optics, 1
phase, 90
polarization, 116
wavepacket, 99
Index 219
nonlinear resonance, 52
width, 47, 51, 53
nonstationary medium, 11
normal incidence, 124
nuclear matter, 194
number of photons, 73, 132, 167
oblique propagation, 39
obstacle epistomologique, 3
occupation numbers, 167
operator,
creation and destruction, 160,
164, 166, 174
displacement vector, 166
electric field, 161, 166, 173
energy, 165, 172
number, 160, 167, 175
optical,
cavity, 5, 170
circuits, 74
experiments, 74
fibre, 6, 63, 87, 97, 115, 121, 193,
212
medium, 31, 67, 193, 194
ordinary mode, 64
overlapping criterion, 53
pancake-like wavepackets, 101, 102
paradigm, 3
partially ionized gas, 37
particle accelerators, 44
phase,
function, 9, 23
slippage, 41
velocity, 98
phase modulation, 193
crossed, 5, 88, 210
induced, 5, 31, 66, 88, 210
self, 5, 67, 87, 212
phase space,
two-dimensional, 33
six-dimensional, 68
photon,
acceleration, 211
action, 23
averaged quantities, 83
bending, 186
classical concept, 8
distribution, 108
effective mass, 1, 13, 97, 114,
187, 193, 211
effective temperature, 85
equivalent charge, 1, 2, 3, 97,
102, 103, 107, 114, 193, 211
equivalent dipole, 6, 97, 115–122,
194, 212
gas, 6, 67
mean density, 78, 83
mean energy, 86
mean velocity, 84
number density, 70
ondulator, 6, 105, 122, 194
pair creation, 7
plasma frequency, 114
position, 10
pressure, 85
proper frequency, 185, 188
proper time, 29, 49
quantum concept, 8, 166
trajectories, 5, 42
universal frequency, 188
velocity, 8, 23, 25
photon acceleration,
stochastic, 6
photoionization, 4, 32, 82
Planck’s constant, 24
Planck distribution, 108, 112
plasma physics, 1
plasma,
boundary, 105
instabilities, 97
isotropic, un-magnetized, 9, 24,
33, 75, 86, 97, 161, 211
kinetic theory, 109
magnetized, 11, 31, 63
temperature, 112
wave, 6
plasmons, 63, 165
220 Index
Poincar´ e map, 52
Poisson bracket, 69
Poisson’s equation, 48, 108
polarization states, 64, 69, 84, 160–
161, 166
ponderomotive force, 2, 97, 103,
108, 115, 194, 211
pulse,
distortion, 78
invariant shape, 90
steepening, 95
quantum,
numbers, 140
optics, 70
oscillator, 160
theory of collisions, 140
quasi-linear diffusion coefficient,
112
quasi-static approximation, 101
R-mode, 64
radiation,
mechanisms, 97
microwave, 4
monochromatic, 6, 51, 54, 62, 87,
88
pressure, 2, 97, 115, 183, 194,
211
supercontinuum, 5, 212
radio galaxies, 63
ray equations, 5, 72, 82, 178, 185,
211
recombination, 32
rectangular pulse, 118, 121
reduced,
distribution function, 111
photon action, 23
reflection, 4, 33, 175, 193
partial, 7, 123, 148
total, 15, 21, 128, 132, 148
refraction, 1, 3, 12-13, 175, 193
refractive index, 9, 31, 64, 125, 193,
195
nonlinear, 4, 66, 91
relativistic,
electron plasma waves, 103
gamma factor, 24, 26, 27
invariants, 127, 128, 132
phase velocities, 43, 108, 112
space–time, 49
relativistic mirror, 4, 17, 21, 22, 34,
148, 181
co-propagating, 39
residual plasma frequency, 37
resonance
asymmetry, 46
condition, 148, 153, 155, 190
resonant,
electrons, 44, 112
transition, 65
saturation amplitude, 148, 150
scattering,
Brillouin, 2
Compton, 3, 4
Raman, 2, 115
Rayleigh, 3
Schwarzschild metric, 191
secondary,
field, 116, 118
photons, 107
self blueshift, 82, 87, 212
self-coupling coefficient, 143, 146,
153
self-frequency shift, 32, 67
separatrix, 36–37, 46, 47
destruction, 53
shocks, 31, 57
simple pendulum, 45
sine differential operator, 71, 74
slingshots, 3
Snell’s law, 12, 14, 15, 212
generalized, 6, 17–19, 21
soliton propagation, 118
sound wave, 137
space–time refraction, 15–17
special relativity, 27
Index 221
spectroscopy, 51
spectral,
asymmetry, 95
broadening, 65
undulations, 83
width, 79, 87
spin one, 165
squeezed states, 172
standard map, 61
stars, 185, 186, 187, 191
static,
electric field, 133, 135, 136
magnetic field, 63
Stark effect, 65
stationary medium, 11
stochastic,
acceleration, 6, 51
trajectories, 156
Stokes sideband, 95
stroboscopic plot, 52
sub-cycle pulses, 137
successive reflections, 40, 43
super-Hamiltomian, 29
supernova explosion, 177
supraluminous boundaries, 19, 20
susceptibility, 9, 26, 70, 118, 196,
199, 211
electron, 111, 112, 113, 114
linear, 88
nonlinear, 66
photon, 111, 112, 113, 114
second-order, 116
third-order, 90, 117
theory of field ionization, 83
threshold criterion, 62
time,
boundary, 169, 175
continuity conditions, 125
discontinuity, 125, 165–172
reflection, 1, 13–15, 19, 123,
125–126, 167, 171, 175,
193, 212
refraction, 1, 13–15, 123, 193,
212
time-dependent medium, 11
timescales, 10
topological transition, 61
transition probabilities, 65
transition radiation, 3, 105, 194
nonlinear, 107
transform limited pulse, 77, 87
transverse electromagnetic waves,
9, 24
trapping, 6
tunnelling ionization, 82
turning point, 34, 42
two wakefields, 52
types of trajectories, 34–38
upper-hybrid frequency, 64
uncertainty,
principle, 87
relations, 77
underdense fronts, 41
unlimited frequency shift, 17
unstable solutions, 153
vacuum, 9, 180, 184, 188, 194, 211
magnetic permeability, 184
permittivity, 9, 161, 178
photon creation, 126, 157, 170,
175, 176
symmetric, 168, 169
states, 167, 168
variational principle, 23, 28
covariant formulation, 29
virtual photons, 135
visible light, 65
wakefield, 31, 43, 102, 109, 112,
152, 210, 213
wavevector, 45
wave absorption, 64
wave equation, 70, 74, 88, 98, 103,
106, 115, 134, 137, 179
wavefunction, 179
222 Index
waveguide, 4, 9, 25
wavenumber, 9
averaged, 78
wavepacket, 8, 54, 184
centroid, 10
envelope, 100
internal structure, 67
wave-particle dualism, 8
white light, 6, 51, 62, 87, 88
Wigner function, 67, 70–73, 197,
198, 201
quantum, 70
reduced, 71
Wigner-Moyal equation, 71, 72, 74,
75, 195–202
WKB solutions, 143, 144, 181
wronskian, 121
zero,
energy, 7
frequency photons, 209, 211

Series in Plasma Physics
Series Editors: Professor Peter Stott, CEA Caderache, France Professor Hans Wilhelmsson, Chalmers University of Technology, Sweden Other books in the series An Introduction to Alfv´ n Waves e R Cross Transport and Structural Formation in Plasmas K Itoh, S-I Itoh and A Fukuyama Tokamak Plasma: a Complex Physical System B B Kadomtsev Electromagnetic Instabilities in Inhomogeneous Plasma A B Mikhailovskii Instabilities in a Confined Plasma A B Mikhailovskii Physics of Intense Beams in Plasma M V Nezlin The Plasma Boundary of Magnetic Fusion Devices P C Stangeby Collective Modes in Inhomogeneous Plasma J Weiland Forthcoming titles in the series Plasma Physics via Computer Simulation, 2nd Edition C K Birdsall and A B Langdon Nonliner Instabilities in Plasmas and Hydrodynamics S S Moiseev, V G Pungin, and V N Oraevsky Laser-Aided Diagnostics of Plasmas and Gases K Muraoka and M Maeda Inertial Confinement Fusion S Pfalzner Introduction to Dusty Plasma Physics P K Shukla and N Rao

Series in Plasma Physics

Theory of Photon Acceleration

J T Mendonca ¸ Instituto Superior T´ cnico, Lisbon e

Institute of Physics Publishing Bristol and Philadelphia

c IOP Publishing Ltd 2001 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the publisher. Multiple copying is permitted in accordance with the terms of licences issued by the Copyright Licensing Agency under the terms of its agreement with the Committee of Vice-Chancellors and Principals. British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. ISBN 0 7503 0711 0 Library of Congress Cataloging-in-Publication Data are available

Commissioning Editor: John Navas Production Editor: Simon Laurenson Production Control: Sarah Plenty Cover Design: Victoria Le Billon Marketing Executive: Colin Fenton Published by Institute of Physics Publishing, wholly owned by The Institute of Physics, London Institute of Physics Publishing, Dirac House, Temple Back, Bristol BS1 6BE, UK US Office: Institute of Physics Publishing, The Public Ledger Building, Suite 1035, 150 South Independence Mall West, Philadelphia, PA 19106, USA Typeset in TEX using the IOP Bookmaker Macros Printed in the UK by Bookcraft, Midsomer Norton, Somerset

tornas. ao mar que te trouxe E ao recuar te transtornas Como se o mar nada fosse Fernando Pessoa To my children: Dina. Joana and Pedro .Onda que. enrolada. Pequena.

.

3 Photon trapping 3.3 Space–time refraction 2.2 Dispersive medium xi 1 1 3 5 8 8 11 12 13 15 17 22 27 31 31 39 43 43 44 49 51 52 54 57 63 67 68 70 70 74 2 3 4 .3 Description of the contents Photon ray theory 2.1 Geometric optics 2.3 Covariant formulation 3.5 Covariant formulation Photon dynamics 3.2.2 Nonlinear photon resonance 3.3 Generalized Snell’s law 2.4.1 Klimontovich equation for photons 4.1 Generation of laser wakefields 3.4 Stochastic photon acceleration 3.4.1 Non-dispersive medium 4.2.6 Magnetoplasmas and other optical media Photon kinetic theory 4.3.2 Time refraction 2.3.2 Space and time refraction 2.1 Refraction 2.2 Accelerated fronts 3.3.1 Ionization fronts 3.1 Motion in two wakefields 3.2 Photon discrete mapping 3.5 Photon Fermi acceleration 3.2.2.2 Historical background 1.2.2 Wigner–Moyal equation for electromagnetic radiation 4.Contents Acknowledgments 1 Introduction 1.1 Definition of the concept 1.4 Photon effective mass 2.

1 Quantization in a dielectric medium 8.2 Photon ondulator 5.2 Flash ionization in a cavity 7.1 Uniform and non-dispersive medium 4.1.3 Contents Photon distributions 4.3.3.1 Derivation of the equivalent charge 5.5 Self-blueshift Photon fluid equations Self-phase modulation 4.1.2 Symmetric Fock states 8.1 Optical theory 4.2 Quantization in a plasma 8.3 Quantum theory of diffraction .4 4.2 Generalized Fresnel formulae 6.1.6 Equivalent dipole in an optical fibre 6 Full wave theory 6.2.4 Dark source 7 Non-stationary processes in a cavity 7.5 Photon beam plasma instabilities 5.2.1 Space and time reflection 6.3 Probability for time reflection 8.5 5 Photon equivalent charge 5.3 Pulse chirp 4.5 Fermi acceleration in a cavity 8 Quantum theory of photon acceleration 8.2 Kinetic theory 75 76 77 78 81 82 83 87 88 90 97 97 103 105 108 113 115 123 123 123 125 126 128 133 140 140 143 146 149 152 157 157 157 161 165 165 167 170 172 173 4.4 Non-stationary medium 4.1 Quantization of the electromagnetic field 8.5.2 Uniform and dispersive medium 4.4 Conservation relations 8.4 Electron beam in a cavity 7.3 Photon transition radiation 5.2 Time refraction 8.1 Operator transformations 8.1 Reflection and refraction 6.2.3.2.3.viii 4.3 Magnetic mode 6.1 Linear mode coupling theory 7.3.2 Time reflection 6.1.3 Ionization front in a cavity 7.4 Photon Landau damping 5.5.

3 Interaction of photons with gravitational waves 9.1 Non-dispersive media A.2 Photons in a gravitational field 9.4 Other metric solutions 9.2.2.2 Gravitational lens 9.2.1 Gravitational redshift 9.1 Neutrino–plasma physics 9.2 Dispersive media References Glossary Index .3 Mean field acceleration processes ix 177 177 183 183 186 187 191 193 195 195 198 203 209 215 Appendix Derivation of the Wigner–Moyal equation A.2.Contents 9 New developments 9.

Acknowledgments

This book results from a very fruitful collaboration, at both local and international levels. First of all, I would like to thank Robert Bingham and Padma K Shukla for their contagious enthusiasm about science, and for their collaboration in several different subjects. I also would like to thank Nodar L Tsintsadze, for his active and creative collaboration during his long stay in Lisboa. I extend my thanks to my many other collaborators in this field. At the local level, I would like to thank my students, with particular emphasis to Luis O Silva, who efficiently transformed some of my vague suggestions into coherent scientific papers. I am also gratefull to Joao M Dias, Nelson Lopes, Goncalo Figueira, Carla C Rosa, Helder Crespo, Madalena Eloi and Ricardo ¸ Fonseca, who by their successful experiments and numerical simulations helped me to get a better understanding of this field. I would also like to refer to the more recent collaboration with Ariel Guerreiro and my colleague Ana M Martins on the quantum theory. J M Dias, G Figueira and J A Rodrigues prepared some of the figures in this book. My acknowledgments extend to my colleagues Armando Brinca and Filipe Romeiras, who looked carefully at a first version of this book and provided very useful suggestions and comments.

J T Mendonca ¸ July 1999

xi

Chapter 1 Introduction

We propose in this book to give a simple and accurate theoretical description of photon acceleration, and of related new concepts such as the effective photon mass, the equivalent photon charge or the photon Landau damping. We also introduce, for the first time, the concepts of time reflection and time refraction, which arise very naturally from the theory of wave propagation in non-stationary media. Even if some of these concepts seem quite exotic, they nevertheless result from a natural extension of the classical (and quantum) electrodynamics to the cases of very fast processes, such as those associated with the physics of ultra-short and intense laser pulses. This book may be of relevance to research in the fields of intense laser– matter interactions, nonlinear optics and plasma physics. Its content may also help to develop novel accelerators based on laser–plasma interactions, new radiation sources, or even to establish new models for astrophysical objects.

1.1 Definition of the concept
The concept of photon acceleration appeared quite recently in plasma physics. It is a simple and general concept associated with electromagnetic wave propagation, and can be used to describe a large number of effects occurring not only in plasmas but also in other optical media. Photon acceleration is so simple that it could be considered a trivial concept, if it were not a subtle one. Let us first try to define the concept. The best way to do it is to establish a comparison between this and a few other well-known concepts, such as with refraction. For instance, photon acceleration can be seen as a space–time refraction. Everybody knows that refraction is the change of direction suffered by a light beam when it crosses the boundary between two optical media. In more technical terms we can say that the wavevector associated with this light beam changes, because the properties of the optical medium vary in space. We can imagine a symmetric situation where the properties of the optical medium are constant in space but vary in time. Now the light wavevector remains 1

2

Introduction

constant (the usual refraction does not occur here) but the light frequency changes. This effect, which is as universal as the usual refraction, can be called time refraction. A more general situation can also occur, where the optical medium changes in both space and time and the resulting space–time refraction effect coincides with what is now commonly called photon acceleration. Another natural comparison can be established with the nonlinear wave processes, because photon acceleration is likewise responsible for the transfer of energy from one region of the electromagnetic wave spectrum to another. The main differences are that photon acceleration is a non-resonant wave process, because it can allow for the transfer of electromagnetic energy from one region of the spectrum to an arbitrarily different one, with no selection rules. In this sense it contrasts with the well-known resonant wave coupling processes, like Raman and Brillouin scattering, harmonic generation or other threeor four-wave mixing processes, where spectral energy transfer is dictated by welldefined conservation laws. We can still say that photon acceleration is a wave coupling process, but this process is mainly associated with the linear properties of the space and time varying optical medium where the wavepackets propagate, and the resulting frequency shift can vary in a continuous way. Because this is essentially a linear effect it will affect every photon in the medium. This contrasts, not only with the nonlinear wave mixing processes, but also with the wave–particle interaction processes, such as the well-known Compton and Rayleigh scattering, which only affect a small fraction of the incident photons. We can say that the total cross section of photon acceleration is equal to 1, in contrast with the extremely low values of the Compton or Rayleigh scattering cross sections. Such a sharp difference is due to the fact that Compton and Rayleigh scattering are single particle effects, where the photons interact with only one (free or bounded) electron, while photon acceleration is essentially a collective process, where the photons interact with all the charged particles of the background medium. This means that photon acceleration can only be observed in a dense medium, with a large number of particles at the incident photon wavelength scale. For instance, in a very low density plasma, an abrupt transition from photon acceleration to Compton scattering will eventually occur for decreasing electron densities. This may have important implications, for instance, in astrophysical problems. The concept of photon acceleration is also a useful instrument to explore the analogy of photons with other more conventional particles such as electrons, protons or neutrons. Using physical intuition, we can say that the photons can be accelerated because they have an effective mass (except in a vacuum). In a plasma, the photon mass is simply related with the electron plasma frequency, and, in a general optical medium, this effective mass is a consequence of the linear polarizability of the medium. This photon effective mass is, in essence, a linear property, but the particlelike aspects of the electromagnetic radiation, associated with the concept of the

This photon phenomenology can also be extended to the physics of neutrinos in a plasma (or in neutral dense matter). in an optical fibre or any similar optical medium. This means that. but was nearly ignored by the builders of particle accelerators of our days. move at very fast speeds. The basic equations necessary for the description of photon acceleration have been known for many years. especially in problems related to ultra-short laser pulse propagation [3. which could provide a good example of what Bachelard would call an obstacle epistemologique [6]. if we replace the electromagnetic coupling between the bound or free electrons with the photons by the weak coupling between the electrons and the neutrino field [106]. This has been known for many centuries (since the invention of slingshots). they can act as relativistic charged particles (electrons for instance) and can eventually radiate Cherenkov. we can also say that photon acceleration is a particular example of a mean field acceleration process. and become ‘dressed’ particles with induced electric charge. gravitational or even the strong interactions). which prevented formally simple jumps in the theory to take place. The concept of photon acceleration can then be seen as a kind of new theoretical paradigm. which results from radiation pressure. weak. Because these equivalent charge distributions. In a plasma. Instead. this ponderomotive force tends to push the electrons out of the regions with a larger content of electromagnetic wave energy. where phonon acceleration can also be considered [1]. therefore behaving as if they were charged particles. Taking an even more general perspective. a large variety of new or already known effects associated with electromagnetic radiation. The main nonlinear property of photons is their equivalent electric charge. capable of integrating in a unified new perspective.Historical background 3 photon. particles usually considered as having no electric charge can efficiently be accelerated by an appropriate background field. transition or bremsstrahlung radiation. . These and similar effects have recently been explored in plasmas and in nonlinear optics. 23]. which can act through any of the physical interaction forces (electromagnetic. in the sense of Kuhn [51]. 1. monopole or dipole charges according to the medium. for instance to acoustics. even if their explicit meaning has only recently been understood. This is due to the existence of a kind of conceptual barrier. or to new particle accelerator concepts and new sources of radiation [42]. the nonlinear second-order susceptibility leads to the appearance of an equivalent electric dipole. for instance. Furthermore.2 Historical background Let us now give a short historical account of this concept. It also means that particles with no bare electric charge can polarize the background medium. we can easily extend this concept to other fields. or ponderomotive force effects. can also be extended in order to include the medium nonlinearities.

The assumed subjective account of the author of the photon acceleration story will be proposed. on microwave frequency up-shift from the centimetric to the milimetric wave range. 87]. they were able to observe the same kind of adiabatic frequency up-shift along the plasma. in 1992 [95] was able to show that the frequency of microwave radiation contained in a cavity can be up-shifted to give a broadband spectrum. A related microwave experiment by Joshi et al.4 Introduction The history of the photon concept is actually rich in these kinds of conceptual barrier. the problem could be seen as an extension of the old problem of wave reflection by a relativistic mirror. and the acceptance that such a particle would also have a momentum. and the better known example is the 30 year gap between the definition by Einstein of the photon as a quantum of light. In a more recent work produced in the context of laser fusion research. The idea behind that theoretical work was to replace the moving particles by moving field perturbations. which was rapidly adopted by the plasma physics community and stimulated an intense theoretical activity on this subject. but qualitatively different. were exploring the relativistic mirror concept. This work also introduced for the first time the name of photon acceleration. the mechanism responsible for the ionization front was not explicitly discussed. In these two papers. easier. mechanism for photon acceler¸ ation was considered by Mendonca in 1979 [63] where the ionization front was replaced by a moving nonlinear perturbation of the refractive index. A closely related. At this stage. experiments like those of Granatstein et al [36. Using numerical simulations. when reflected inside a waveguide by a relativistic electron beam. Wilks et al [118] considered the interaction of photons with plasma wakefield perturbations generated by an intense laser pulse. caused by a strong electromagnetic pulse. to be excited in the laboratory. In this work it was shown that the frequency up-shift is an adiabatic process occurring not only at reflection as previously considered. if the mirror is replaced by a surface of discontinuity between the neutral gas and the ionized gas (or plasma). accepting that other and eventually better and less biased views are also possible. introduced in the theory of Compton scattering [86]. in principle. It is therefore very difficult to have a clear historical view and to find out when the concept of photon acceleration clearly emerged from the already existing equations. in the presence of an ionization front produced by . One of the first papers which we can directly relate to photon acceleration in plasma physics was published by Semenova in 1967 [97] and concerns the frequency up-shift of an electromagnetic wave interacting with a moving ionization front. But more recently it became aparent that relativistic ionization fronts could be produced in a laboratory by photoionization of the atoms of a neutral gas by an intense laser pulse. by also at transmission. with energy proportional to the frequency ω. At that time. proportional to the wavevector k. The same problem was later considered in greater detail by Lampe et al in 1978 [53].

As we will see in the present work. We will attempt in this work to bridge the gap between the two scientific communities and between the two distinct theoretical views. In the optical domain. the spectrum of a CO2 laser pulse was strongly broadened and slightly upshifted. The first two chapters will be devoted to the study of single photon equations (also called ray equations). were also pointing to the physical reality of the new concept and giving credit to the emerging theory. This effect is now called flash ionization for reasons that will become apparent later. This provides another proof of the interest and generality of the concept of photon acceleration. (2) photon kinetic theory. and it can also be considered as a particular and limiting case of the photon acceleration processes. Very recently. derived in the frame of geometric optics. were able to demonstrate. These results provided the first clear indication that the photon acceleration mechanism was possibly taking place.and counter-propagation. beyond any reasonable doubt. when the laser beam was focused inside an optical cavity and ionization of the neutral gas inside the cavity was produced. 1. creating a dense plasma when they are focused in a neutral gas region. Due to its formal simplicity. the theory of photon acceleration as developed in plasma physics is also able to explain the phase modulation effects. Actually. and the measured up-shifts [120].3 Description of the contents Four different theoretical approaches to photon acceleration will be considered in this work: (1) single photon trajectories. which has several advantages over more accurate methods. induced and cross phase modulation). In parallel with this work in plasma physics. the first two-dimensional optical experiments carried out by our group [21]. and was able to prove both by theory and by experiments. following both theoretical and experimental approaches clearly independently. where a probe laser beam was going through a relativistic ionization front in both co. This is well documented in the book recently edited by Alfano [3]. various physical configurations where . the existence of photon acceleration and to provide an accurate quantitative test of the theory. research on a very similar class of effects had been taking place in nonlinear optics since the early seventies. the observation of a self-produced frequency up-shift of intense laser pulses.Description of the contents 5 an ultraviolet laser pulse. In these pioneering experiments. the spectral changes of laser beams by ionization of a neutral gas were reported as early as 1974 by Yablonovich [122]. This is the simplest possible theoretical approach. we can apply it to describe. with great detail and very good accuracy. when we adapt it to the optical domain. (3) classical full wave models and (4) quantum theory. and with almost completely mutual ignorance. This work mainly concentrated on the concept of phase modulation (including self-. that laser pulses with a very large spectrum (called the supercontinuum radiation source) can be produced.

The geometric optics approximation. This single photon theory is simple and powerful. For instance. in its single photon and kinetic versions. Chapter 2 deals with the basic concepts of this single photon or ray theory. applied here to photons. and the role played by the phase of the laser field in this process. even if the nature of the forces acting on the photons is not the same. leading to the transformation of monocromatic radiation into white light. We will also discuss the new radiation processes associated with these charge distributions. as applied to a generic space. moving nonlinearities and flash ionization processes. illustrated with examples taken from plasma physics. Extension to other optical media. In particular. However. The concept of space–time refraction is introduced. the generalized Snell’s laws are derived and the ray-tracing equations are stated in their Hamiltonian. self-phase modulation of a laser pulse. This is similar to the well-known electron Landau damping [54]. which can be easily described with the aid of single photon equations. can still be considered as slow processes in the . will be discussed. provides a very accurate theoretical description for a wide range of different physical configurations. occurring on a timescale of a few tens of femtoseconds. an extension of this single particle description to the kinetic theory of a photon gas is relatively straightforward and can lead to new and surprising effects such as photon Landau damping [14]. but it can only provide a rough description of the laser or other electromagnetic wavepackets evolving in non-stationary media. with revelance to laser–plasma interaction problems. such as photon ondulator radiation. This is particularly important for ultra-short laser pulse propagation. This means that photon acceleration happens to be quite similar to electron or proton acceleration by electromagnetic fields.6 Introduction photon acceleration occurs. Even very fast time events. Such a kinetic theory is developed in chapter 4 and gives a much better description of the space–time evolution of a broadband electromagnetic wave spectrum. An interesting example of stochastic photon behaviour is provided by the well-known Fermi acceleration process [29]. and to nonlinear optical configurations will also be discussed. It can also be shown that stochastic photon acceleration is possible in several physical situations. relativistic plasma waves or wakefields. propagating in a nonlinear optical medium. and of the equivalent electric dipole of photons in an optical fibre. Apart from its simplicity and generality. Lagrangian and covariant forms. photon transition radiation or photon bremstrahlung. photon ray equations are formally very similar to the equations of motion of a material particle. Chapter 5 is devoted to the discussion of the equivalent electric charge of photons in a plasma. These are ionization fronts (with arbitrary shapes and velocities). acceleration and trapping of electrons and photons can equally occur in the field of an electron plasma wave.and time-varying optical medium. Chapter 3 deals with the basic properties of photon dynamics.

We are then led to the full wave treatment of photon acceleration.Description of the contents 7 optical domain and stay within the range of validity of this theory. which is presented in chapters 6 and 7.The first one concerns collective neutrino plasma interaction processes. for specific phase effects. which were first explicitly formulated by Bingham et al in 1994 [13] and have recently received considerable attention in the literature. the multiple mode coupling or the theory of the dark source which describes the possibility of accelerating photons initially having zero energy. which correspond to other examples of the mean field acceleration process. in several situations. One of the consequences of such an interaction is the occurrence of photon acceleration in a vacuum by gravitational waves. More work is still in progress in this area. coming out of the vacuum. We will also show that time refraction always leads to the creation of photon pairs. These new problems are clearly more controversial than those covered in the first eight chapters. the photon acceleration theory is extended in a quite natural way to cover new physical problems. which is the basic mechanism of photon acceleration. such as the generation of a magnetic mode. The quantum Fresnel formulae for the field operators will be derived. chapter 9 is devoted to new theoretical developments. in order to account for partial reflection. The analogies between the photon and the neutrino interaction with a background plasma will be established. In chapter 8 we show that a quantum description of photon acceleration is also possible. Most of the problems discussed in previous chapters are reviewed with this more exact approach and a comparison is made with the single photon theory when possible. Finally. . but they are also very important and intellectually very stimulating. a more accurate theoretical approach is needed. But. Here. We will try to establish in solid grounds the theory of time refraction. New aspects of photon acceleration can now be studied. or for arbitrarily fast time processes. Two of these examples are briefly discussed. The second example will be the interaction of photons with a gravitational field and the possibility of coupling between electromagnetic and gravitational waves.

These wavepackets can be viewed as classical particles and can be assimilated to photons. In particular. A more accurate description of a wavepacket can still be given in the geometric optics approximation. instead of a single one. This will be discussed in chapter 4. 8 . t) = E 0 exp i{k · r − ωt} (2. This single photon approach will be used in the present and the next chapters. The study of such a bunch will give us information on the internal spectral content of the wavepacket. an electromagnetic wave can be described by an electric field of the form E(r . The use of the photon concept in a quantum context will be postponed until chapter 8. r is the position and t the time. A single photon can be used to represent the mean properties of a given wavepacket. The photon velocity will then be equal to the group velocity of the wavepacket. as the analogue of an electromagnetic wavepacket. 16]. 2.Chapter 2 Photon ray theory It is well known that the wave–particle dualism for the electromagnetic radiation can be described in purely classical terms.1 Geometric optics It is well known that a wave is a space–time periodic event. by using a bunch of photons. Geometric optics is a well-known and widely used approximation of the exact electromagnetic theory and it is presented in several textbooks [15. We will then apply the word ‘photon’ in the classical sense.1) where E 0 is the wave field amplitude. We will first use the geometric optics description of electromagnetic wavepackets propagating in a medium. where the time periodicity is characterized by the angular frequency ω and the space periodicity as well as the direction of propagation are characterized by the wavevector k. The wave behaviour is described by Maxwell’s equations and the particle behaviour is described by geometric optics. This contrasts with other particles and fields where the particle behaviour is described by classical mechanics and the wave behaviour by quantum mechanics.

the dispersion relation becomes ω= kc kc kc =√ =√ n 1+χ (2. the wave dispersion relation is a complicated expression of ω and k. In a vacuum. where ωp is the electron plasma frequency. 108]. the medium can still be considered as approximately uniform and constant). by taking the space and time derivatives of the phase function ψ: k= ∂ ψ.2) we can then have a dispersion relation of the form ω= 2 k 2 c 2 + ωp . they are related by a well-defined expression (at least for low amplitude waves) known as the dispersion relation. k) = 0. In a lossless medium these quantities are real. where e and m are the electron charge and mass. for high frequency transverse electromagnetic waves propagating in an isotropic plasma [82. and 0 is the vacuum permittivity. It is then preferable to state it implicitly as R(ω. In a medium.1) by a similar expression: E(r . and such simple and explicit expressions for the frequency cannot be established. Let us now assume that propagation is taking place in a nonuniform and non-stationary medium. t) = E 0 (r . We can now define a local value for the wave frequency and for the wavevector. locally both in space and in time. where the constant c is the speed of light in a vacuum and k is the absolute value of the wavevector (also known as the wavenumber).3) A similar equation is also valid for electromagnetic waves propagating in a waveguide [25]. however.6) . the dispersion relation is simply given by ω = kc. From equation (2. In the most general case. The plasma frequency is now replaced by a cut-off frequency ω0 depending on the field configuration and on the waveguide geometry. we have = 1 − (ωp /ω)2 . but they are not independent from each other.2) where n is the refractive index of the medium. ∂r ω=− ∂ ψ. If the space and time variations in the medium are slow enough (in such a way that. t) is the wave phase and E 0 (r .5) where ψ(r . t) (2. t) a slowly varying wave amplitude. its dielectric constant and χ its susceptibility. In particular.4) The above description of wave propagation is only valid for uniform and stationary media. Instead. and in dispersive media they are functions of the frequency ω and the wavevector k. ∂t (2. (2. t) exp iψ(r . (2. we can replace the wave electric field (2. It is related to the electron density n e by the 2 expression: ωp = e2 n e / 0 m.Geometric optics 9 In a stationary and uniform medium the frequency and the wavevector are constants.

7) The parameters of the medium. (2. Starting from Maxwell’s equations. vg = dr /dt. ∂r (2. dt ∂k ∂ω dk =− . which is now only locally valid: R(ω. if the following inequality is satisfied: 2π ∂ 2π ln ξ + |∇ ln ξ | ω ∂t k 1 (2. The canonical variables are here the photon position r and the wavevector k. for instance its refractive index or. while the .10 Photon ray theory This local frequency ω and wavevector k are still related by a dispersion relation. or. it can be shown that this expression stays valid as long as the space and timescales for the variations of the medium are much slower than k −1 and ω−1 .8) where ξ is any scalar characterizing the background medium. because ω is assumed to be a function of r and k.9) = −∇ω = − + · ∂t ∂r ∂r ∂ k The last term in this equation was established by noting that. This means that its solution can be written as ω = ω(r .10) This is the velocity of the centroid of an electromagnetic wavepacket moving in the medium. (2. dt ∂r (2. (2. It can easily be seen from the above definitions of the local frequency and wavevector that ∂k ∂ω ∂ k ∂ω . in more precise terms. k.11) The differential operator on the left-hand side is nothing but the total time derivative d/dt. t) = 0. Such a dispersion relation is satisfied at every position and at each time. Let us introduce the definition of group velocity vg = ∂ω ∂k . it varies in space not only because of its explicit dependence on r but because the value of k is also varying. The above equation can be written as ∂ ∂ + vg · ∂t ∂r k=− ∂ω . It means that we can rewrite the last two equations as dr ∂ω = . for a plasma. depend on the position r and on time t.12) These equations of motion can be seen as describing the evolution of point particles: the photons. in a plasma. r . its electron plasma frequency ωp . for instance the refractive index or. t). We notice that they are written in Hamiltonian form. k. the electron density.

Equations (2. (2. t) From equations (2. (ii) an homogeneous but time-dependent medium.and/or time-dependent) dielectric medium. t).13).12) and (2. t) = . the physical implications of such a structure for media varying both in space and time have only recently been fully understood [66]. But. (iii) an inhomogeneous and non-stationary medium. 117]. We see that they explicitly predict a frequency shift as well as a wavevector change. in fact. k. it will be time dependent according to dω ∂ω = . 9. These ray equations can also be written with implicit differentiation as dk ∂ R/∂ r = . dt ∂t (2.Space and time refraction 11 frequency plays the role of the Hamiltonian function. dt ∂t (2.13) These equations are well known in the literature and their derivation is given in textbooks [55] and in papers [8. we can also obtain dω ∂ = −ω ln n. for some historical reason. (2. dt ∂ R/∂ω and dr ∂ R/∂ k =− dt ∂ R/∂ω (2. let us use a simple and illustrative example of a photon propagating in a nondispersive (but space.16) n(r .12). For this reason we will retain our attention on the explicit Hamiltonian version of the ray or photon equations. ω ≡ h(r . described by the dispersion relation kc ω ≡ ω(r . 2. we have dr ω = 2 k. . loses the clarity and elegance of the Hamiltonian approach. even if it is appropriate for numerical computations (this is currently used for magnetized plasmas with complicated spatial configurations). dt ∂r (2.13) have.14) dω ∂ R/∂t =− .2 Space and time refraction In order to understand the physical meaning of the photon equations stated above.17) From equation (2. In general. k.18) Let us successively apply these equations to three distinct situations: (i) an inhomogeneous but stationary medium. an obvious symmetrical structure.15) dt ∂ R/∂ω This implicit version. dt k ∂ dk = ω ln n.

(2. if we are interested in the study of the photon or ray propagation across a large region with dimensions much larger than lb . k. and that the perpendicular component remains constant.20) where H (y) = 0 for y < 0 and H (y) = 1 for y > 0 is the well-known step function or Heaviside function. If the plane of incidence coincides with z = 0.19) we have dω = 0. The hyperbolic tangent is chosen as a simple and plausible model for a smooth transition between the two media.17. the ray equations (2. the above law can be approximated by n(y) = n 1 + n H (y) (2. dt ∂y 2n(y) (2. separated by a boundary layer of width lb located around the plane y = 0.19) corresponds to the usual optical configuration for wave refraction.2.1 Photon ray theory Refraction In the first situation. t) = ω0 = const. t) ≡ n(y) = n 1 + n [1 + tanh(kb y)] 2 (2.18) and (2. (2.12 2. ω0 kx = n(y) sin θ(y) = const.25) .21) dt This means that the wave frequency is a constant of motion ω(r . 2. and kb = 2π/lb . from the first of these ˆ equations. such a smooth transition can be viewed from far away as a sharp boundary. we can obtain. with refractive indices n 1 and n 2 . In this large scale view of refraction. In order to describe such a configuration we can write n(r .24) c Considering the asymptotic values of θ(y) as the usual angles of incidence and of transmission θ1 = θ (y → −∞) and θ2 = θ(y → +∞).18) are only valid when lb is much larger than the local wavelength 2π/k. Clearly. and the above model given by equation (2. Defining θ(y) as the angle between the wavevector k and the normal to the boundary layer e y .19) where n = n 2 − n 1 .22) (2. First of all. (2. we reduce this result to the well-known Snell’s law of refraction n 1 sin θ1 = n 2 sin θ2 . the time variation of the wavevector components is determined by dk x =0 dt dk y ∂ ω0 n = ω0 ln n(y) = kb sech2 (kb y). it should be noticed that from equations (2. we assume two uniform and stationary media. On the other hand.23) We see that the wavevector component parallel to the gradient of the refractive index is changing across the boundary layer.

for y → +∞. we could be led to talk about photon acceleration during refraction.2 Time refraction Let us turn to the opposite case of a medium which is uniform in space but changes its refractive index with time.26) [1 + tanh( b t)] . with asymptotic values vg = c/n 1 .19). we now have n(r . But. if .19): n (2. following a law similar to that of the wavevector change during refraction. and vg = c/n 2 . This will not be the case for the next two examples. in such a way that the total photon energy remains constant. the timescale for refractive index variation 2π/ b is much larger than the wave period 2π/ω. Returning to equations (2. or photon velocity. From equation (2.Space and time refraction 13 ω constant y θ2 n2 n1 θ1 = x Figure 2.27) We see that the photon frequency is shifted as time evolves. Photon refraction at a boundary between two stationary media. for y → −∞. as we will see later. this is not appropriate because this change in group velocity is exactly compensated by a change in the photon effective mass.1. 2 Here. t) ≡ n(t) = n 1 + dω ∂ ω n = −ω ln n(t) = − dt ∂t 2n(t) b sech2 ( b t). We can describe this change by a law similar to equation (2. 2.17) we can also see that refraction leads to a change in the group velocity. In a broad sense.2. On the other hand. (2.

28) The photon momentum is now conserved and the change in the group velocity is only related to the frequency (or energy) shift. (2. such that tan α = vg /c = 1/n.30) This can be called the Snell’s law for time refraction. propagation is taken along the x-direction. we get ω1 tan α2 = ω2 tan α1 . similar to the usual angle of incidence θ defined in the plane (x. Actually. dt n(t) dk = 0.18) that dx c = = vg (t).24) is now replaced by ω(t) k= n(t) = const. y).31) This equation resembles the well-known Snell’s law of refraction.14 Photon ray theory k = constant ct α2 n 2 x α1 n1 Figure 2. related to .25). This could be called the angle of temporal incidence. Replacing it in equation (2. we can then write n 1 ω1 = n 2 ω2 . However. (2. Time refraction of a photon at the time boundary between two uniform media. dt (2. ct) we can define an angle α. there is an important qualitative difference.29) c Defining the initial and the final values for the frequency as the asymptotic values ω1 = ω(t → −∞) and ω2 = ω(t → +∞). we can see from equation (2. Snell’s law (2. equation (2. in the plane (x. (2.2.30).

32) [1 + tanh(kb x − b t)] 2 with b = kb u. In contrast.3 Space–time refraction Let us now assume a more general situation where the optical medium is both inhomogeneous in space and non-stationary in time.34) .24.25) we see that.12) we can now derive a simple relation between the time variation of frequency and wavenumber: dk 1 dω = . We know that this can occur when a photon propagates in a medium of decreasing refractive index. Now. This new constant of motion can be directly obtained from equation (2. the constant of motion was the photon wavevector. or their temporal counterparts (2. no such thing as a reflection back in time (or a return to the past) can ever occur.25).30) or (2. We notice that in order to establish the Snell’s laws (2. 2.31) are always satisfied for arbitrary (positive) values of the asymptotic refractive indices n 1 and n 2 .13).29.19) and (2. dt u dt (2. k = k. t) ≡ n(x − ut) = n 1 + (2. let us then define a canonical transformation from the variables(x. 2. we will assume that the boundary layer between media 1 and 2 is moving with a constant velocity u along the x-direction. For such a purpose. such that x = x − ut. In the second case. Not surprisingly. when a photon interacts with a moving boundary layer. for n 1 > n 2 . for the case of refraction in a moving boundary. (2. or space–time refraction.30). In the first case. k ). of the form n n(r . k) to the new pair of variables (x .12.Space and time refraction 15 the mechanism of total reflection.33) This relation means that. it is possible to define a critical angle θc such that sin θc = n 2 /n 1 . the constants of motion were the photon frequency ω and the wavevector component parallel to the boundary layer k x . From equation (2. Equations (2. For θ1 ≥ θc we would have sin θ2 ≥ 1 which means that propagation is not allowed in medium 2 and that total reflection at the boundary layer will occur. 2. we had to define some constant of motion.26) are now replaced by a similar expression. we need to find a new constant of motion because neither the frequency nor the wavevector are conserved. equations (2. the wave frequency and wavevector are both shifted. in a kind of space– time refraction. 2. but it is useful to explore the Hamiltonian properties of the ray equations (2.33). This is the physical meaning of the replacement of the sine law for reflection by the tangent law for time refraction.2. From the ray equations (2. As a simple generalization of the previous two models.

39) Because. we can take the asymptotic values of n(x ) and k for regions far away from the boundary layer (x → ±∞). We see that the resulting frequency up-shift can be extremely large if the boundary layer and the photons are moving in the same direction at nearly the same velocity: β2 1. we were able to define a new constant of motion ω = const. k ) − uk = c −u k . k ) = ω(x . 2. as F(x.42) We are then led to the following formula for the total frequency shift associated with space–time refraction: ω2 = ω1 1 − β1 1 − β2 (2. ∂k k= The Hamiltonian (2. for i = 1. = dt ∂k dk ∂ω .43) where βi = un i /c. in terms of the frequency ω = ω1 1 − n1 n2 u = ω2 1 − u . . k .36) is determined by ω =ω+ ∂F = ω − uk .41) or. n(x ) (2. c c (2. equivalently.36) In a more formal way.40) (2. we can define a generating function for this canonical transformation. ∂t (2. based on this canonical transformation.38) (2. and write ω = c − u k1 = n1 c − u k2 n2 (2.35) It can easily be seen that the new Hamiltonian ω appearing in these equations is a time-independent function defined by ω ≡ ω (x .37) such that ∂F =k ∂x ∂F x = = x − ut. =− dt ∂x (2. t) = (x − ut)k (2.16 Photon ray theory The resulting canonical ray equations can be written as dx ∂ω .

because in this case the photons are moving with the same velocity as the boundary layer. without invoking relativity or using any Lorentz transformation. A photon propagating initially in a vacuum in the opposite direction. the resonant photon trajectory is physically irrelevant because it represents an ensemble of zero measure in the photon phase space (x. 1−β (2. 2. This means that in the above equation (2. with a wavevector k = k1 ex . k). it could be concluded from equation (2. which was derived here in a very simple way. This can be done by assuming that n 1 = 1 and n is such that n(x ) = 0 for x ≥ 0.36). this result is also interesting because it shows that such an effect can be seen as a particular case of the more general space–time refraction or photon acceleration processes. However. this shows important qualitative and quantitative differences. It will then become ω2 = ω1 1+β . This new effect of almost unlimited frequency shift is a clear consequence of the combined influence of the space and time variations of the medium. The mirror will move with constant velocity −u and its surface will coincide with the plane x = 0. no frequency shift would occur. they would need an infinite time to travel across the layer and to be infinitely frequency shifted. a large number of possible photon trajectories will lead to extremely high-frequency transformations taking place in a finite but large amount of time. by considering oblique photon propagation with respect to the moving boundary between two . A comment has to be made on the exact resonance condition. Let us now apply the above analysis to the well-known and famous problem of photon reflection by a relativistic mirror. A closer look at the resonant photon motion shows that this trajectory would be x = const and. Apart from providing a very simple and alternative way to derive the relativistic mirror effect. or in other words. Such a result is possible because the photon equations are exactly relativistic by nature.44) This is nothing but the well-known formula for the relativistic mirror. At a first impression. in the close vicinity of this particular trajectory. What is relevant is that.Generalized Snell’s law 17 When compared with the frequency shifts associated with the case of a purely time refraction. On the other hand.3 Generalized Snell’s law We generalize here the above discussion of space–time refraction. of the synergy between the usual refraction and the new time refraction considered above.43) we can make β2 = β = u/c and β1 = −β. according to equation (2.43) that the frequency shift would become infinitely large if this resonance condition were satisfied. defined by β2 = 1. will be reflected by the mirror and come ˆ ˆ back to medium 1 with a wavevector k = −k1 ex .

It is useful to introduce parallel and perpendicular propagation. As before.48) This is a straightforward generalization of the canonical transformation used for the one-dimensional problem. (2. remain constant.46) (2.52) (2.51) ∂t This Hamiltonian is a constant of motion because in the new coordinates neither n nor ω depend explicitly on time. ∂k k= by (2. The moving boundary will now be described by the following refractive index: n(r . where b = kb /kb .49) (2. t) appearing in these equations is determined ∂F =ω−u·k .50) The new Hamiltonian ω (r . k . We conclude from the photon equations of motion that dk kc ∂n = 2 dt n ∂r dk⊥ = 0. t). k ). we assume that the velocity and the slope of the boundary layer. The first one is simply the new Hamiltonian ω .53) . generated by the transformation function F(r .18 Photon ray theory different dielectric media. It is clear that we can now define two constants of motion for photons crossing the moving boundary. k ) − u · k ˆ I2 = |k × b|. We can now use the canonical transformation from (r . (2. This leads to ∂F =k ∂r ∂F r = = r − ut. k) to a new pair of variables (r . t) = (r − ut) · k . k . dt (2. t) = n(r . Let us call them I1 and I2 . defined by kb .45) where b = kb · u and u is the velocity of the moving boundary. (2. by defining ˆ ˆ ˆ r = r b + r⊥ and k = k b + k⊥ .47) The conservation of the perpendicular wavevector is due to the fact that the refractive index is only a function of r : n(r . and the other one is the perpendicular wavenumber k⊥ : ω =ω+ I1 = ω(r . t) = n 1 + n 1 + tanh(kb · r − 2 b t) (2.

that ω2 = ω1 1 − β1 cos θ1 1 − β2 cos θ2 (2. It is well known that such boundaries can exist without violating the principles of Einstein’s theory of relativity [31]. the frequency shift also tends to zero. Denoting by the subscripts 1 and 2 the frequencies and angles at the extreme ends of such trajectories.25). k ) 1 − I2 = k sin θ (r ).54) (2. If θ (r ) is the angle between the photon wavevector and the front velocity.57) This last equation can be seen as the generalized Snell’s law.56) is valid for an arbitrary value of the velocity of the moving boundary. valid for refraction at a moving boundary betwen two different dielectric media.30): n 2 ω2 = n 1 ω1 . ω2 → ω1 . This means that it reproduces the above result for a pure time refraction. It stays valid. . ω1 (2. for supraluminous moving boundaries. for a finite velocity u = 0. Let us retain our attention on photon trajectories which start in medium 1 far away from the boundary.55) In the first of these equations. we can rewrite the two invariants as I1 = ω(r .56). In this case equation (2. in particular. but for normal incidence θ1 = θ2 = 0. as it should. This expression establishes the total frequency shift. in accordance with the discusssion of the previous section. equation (2. u n(r ) cos θ(r ) c (2. On the other hand. k ) = k c. for i = 1.56) reduces to ω2 = ω1 (1 − β1 )/(1 − β2 ). in order to ˆ place our discussion on simple grounds. Of course. as shown by equation (2. It shows that the relation between the angles of incidence and transmission depends not only on the refractive indices of the two media but also on the photon frequency shift. defined for each photon position r along the trajectory. at r → −∞. such that u > c. we made use of the local dispersion relation n(r )ω(r . We should notice that equation (2. u → 0. They can even be built in a medium where the atoms are completely at rest. when the velocity of the boundary layer tends to zero. we assume here that u is parallel to b. 2. Let us consider the limit of an infinite velocity u → ∞. from the above two invariants. we can conclude. We can also conclude that n 1 sin θ1 = ω2 n 2 sin θ2 .Generalized Snell’s law 19 Notice that our analysis is valid even if the velocity of the moving boundary ˆ u is not parallel to the gradient of the refractive index b. as will be discussed in the next chapter.56) is then reduced to equation (2.56) where βi = (u/c)n i . However. Equation (2. cross the boundary layer (at r = 0) and then penetrate in medium 2 moving towards r → +∞.57) reduces to the usual form of Snell’s law (2.

we get from equations (2.56. there is no momentum transfer from the medium to the photon. u Now. θ1 θ2 .58) We see that.20 Photon ray theory (a) ct α2 n α1 (b) 2 x n 1 y u n 1 ut t u θ2 x n2 θ1 0 Figure 2.57) sin(θ1 − θ2 ) 0. (2. in this limit of a highly supraluminous front. The resulting frequency shift is determined in this case by the law of pure time refraction.3. (a) Diagram (x. (b) oblique propagation. ct) for one-dimensional propagation. for an arbitrary oblique incidence. Illustration of the generalized Snell’s law: photon refraction at the moving boundary between two uniform media. but for highly supraluminous fronts. c. showing that . 2.

 (2. The same occurs for purely time-varying media (|βi | → ∞).60) This can be explicitly solved and gives  n1n2  sin θc = 2 1 + |β1 | 1 − 2 n 1 + β1 n 2  2 n2 n1 2 2 (1 − β1 ) . as expected. or θc = π/2 (absence of total reflection) for arbitrary values of β1 .57) we can define a critical angle θ1 = θc such that θ2 = π/2. The dominant processes are associated with the time changes.56). We also notice that this result guarantees that in the absence of any boundary (which means n 2 = n 1 ) we always get sin θc = 1. Dependence of the critical angle θc on the front velocity β1 . confirming that no time reflection is possible.Generalized Snell’s law 21 sin θ c β 1 Figure 2. we have to replace β1 by −β1 .8. Let us now examine the conditions for the occurrence of total reflection.59) Using equation (2. From the above expression of the generalized Snell’s law (2. n1   (2. n 1 ω1 (2.61) For a boundary at rest.4.56). For angles of incidence larger than this critical angle θ1 > θc . or sin θc = n 2 ω2 . this expression reduces to the well-known formula for the critical angle sin θc = n 2 /n 1 . If the boundary layer is moving towards the incident photon (which is the usual configuration of the relativistic mirror effect). for n 2 /n 1 = 0. space changes of the refractive index of the medium do not contribute to the final frequency shift. we also see that sin θc = n2 (1 − β1 cos θc ). This means that the direction of the reflected wavevector will be reversed with respect to that assumed in equation (2. the photon will be reflected. β1 = 0.

If u is contained in the plane of incidence.55).57). Let us return to the general expression for the two photon invariants I1 and I2 . are ˆ not parallel to each other. and kb . as given by equation (2. Our Hamiltonian approach to photon dynamics can however be applied to more complicated situations of space–time varying media. This is because our main objective here was to extend the existing geometric optics theory to the space–time domain and to obtain simple but also surprising generalizations of well-known formulae and well-known concepts. From the above definitions of the local wavevector k and local frequency ω. But in order to account for partial reflection we will have to use a full wave description and not use the basis of the geometric optics approximation. and make a short comment on the case where the vector velocity. ω3 (2. (2. 2. u. We can then write u · k = uk cos α(r ) and |k × b| = k sin θ(r ). is dψ = k · dr − ω dt. where these two angles α(r ) and θ(r ) are not necessarily equal to each other. for instance to nonstationary and accelerated boundary layers. Actually.22 Photon ray theory in this equation. in order to stress that the final photon state is given by reflection and not by transmission: 1 + β1 cos θ1 ω 3 = ω1 . according to equation (2. as a particular case of photon acceleration at the moving boundary. the angle θ (r ) has to be replaced by α(r ) = θ(r ) − θ0 . we will replace the subscript 2 by the subscript 3. (2. it is obvious that the total variation of the phase ψ. Moreover. where θ0 is a constant. as a function of the space and time coordinates.4 Photon effective mass Let us now go deeper in the analysis of the photon dynamics and explore the analogies between the photon ray equations and the equations of motion of an arbitrary point particle with finite rest mass. In the expressions of the two invariants I1 and I2 . characterizing the gradient of the refractive index across the boundary layer. we have θ(r ) = α(r ) + θ0 .64) . it is given by sin θ3 = ω1 sin θ1 . for angles of incidence smaller than the critical angle.62) 1 − β1 cos θ3 Here the angle of reflection θ3 is not equal to the angle of incidence (as occurs for a stationary boundary) but. In our discussion we have kept our attention on very simple but paradigmatic situations of boundary layers moving with constant velocity. these formulae stay valid even for partial reflection at the boundary.63) These two formulae state the relativistic mirror effect for oblique incidence.

we obtain L(r . we have ω = ω0 = const.1.71) It should be noticed that.67) In the particular case of a stationary medium. (2.66) It is well known that. and equation (2. t) dt (2. we can get the following equation for the phase: ∂ψ ∂r 2 = n2 c2 ∂ψ ∂t 2 .72) . (2. c2 0 (2. and state that the phase has to obey the extremum condition δψ = 0. and from the definitions of k and ω. v. t) dt. or group velocity. valid for space–time varying media. as for the generalized eikonal equation. from this variational principle.65) In order to establish the photon trajectories we can apply the principle of minimum action. k. 56] that this phase is nothing but the photon action ψ= k · dr − ω(r . k.Photon effective mass 23 If we identify the photon frequency ω with the particle Hamiltonian. the photon wavevector is not an independent variable because it is determined by k= ∂L . v.67) is then reduced to the usual eikonal equation of geometric optics [15] ∂ψ0 ∂r 2 = n2 2 ω .68) where ψ0 is the reduced photon action.65) reduces to ψ = ψ0 − ω0 t = k · dr − ω0 t (2. ∂v (2. it is possible to rederive the photon canonical equations already established in section 2. On the other hand.70) where v = dr /dt is the photon velocity. for a dielectric medium with refractive index n = kc/ω. we can easily see from classical mechanics [33.65). and the wavevector k with its momentum. The photon Lagrangian is then determined by ψ= t2 t1 L(r .67). in this expression. as we did before. Comparing this with equation (2. equation (2.69) We can then refer to the more general equation for the phase. We also know from classical mechanics that the action can be defined as the time integral of a Lagrangian function. t) = k · v − ω(r . (2. Equation (2. t). (2.

24

Photon ray theory

We can also determine the force acting on the photon as the derivative of the Lagrangian with respect to the coordinates (or the gradient of L): f = ∂L . ∂r (2.73)

With these definitions we can reduce the Lagrangian equation d ∂L ∂L − =0 dt ∂ v ∂r to Newton’s equation of motion dk ∂ω = f =− . dt ∂r (2.75) (2.74)

Obviously, this also agrees with the second of the Hamiltonian ray equations (the first one can be seen as the definition of v = dr /dt). We have given here the three equivalent versions of the photon equations of motion in a space– time varying medium: (1) the Newtonian equation (2.75); (2) the Lagrangian equation (2.74) and (3) the Hamiltonian equations already stated in section 2.1. This means that the photon trajectories can be described by the formalism of classical mechanics, in a way very similar to that of point particles with finite mass. Let us explore further this analogy between the photon and a classical particle. The example of high-frequency transverse photons in an isotropic plasma is particularly interesting because of the simple form of the associated dispersion 2 relation: ω = k 2 c2 + ωp (r , t). If we multiply this expression by Planck’s constant (divided by 2π), redefine the photon energy as = h ω and the photon ¯ momentum as p = h k, we get ¯ = p 2 c2 + h 2 ωp (r , t) = ¯ 2 p 2 c2 + m 2 c4 . eff (2.76)

This means that the photon in a plasma is a relativistic particle with an effective mass defined by m eff = ωp h /c2 . (2.77) ¯ We see that the photon mass in a plasma is proportional to the plasma frequency (or to the square root of the electron plasma density). Therefore, in a non-stationary and non-uniform medium, this mass is not a constant because it depends on the local plasma properties. In the following, we will use h = 1. We can also see that the photon velocity ¯ in this medium is determined by v = pc2 / , or v = kc2 /ω. The corresponding relativistic gamma factor is then 1 dr γ = 1− 2 c dt
2 −1/2

k 2 c2 = 1− 2 ω

−1/2

=

ω . ωp

(2.78)

Photon effective mass

25

We can also adapt Einstein’s famous formula for the energy of a relativistic particle to the case of a photon moving in a plasma, as ω = γ ωp = m eff γ c2 . (2.79)

Let us now write the photon Lagrangian in an explicit form. Using equation (2.71), we have L= v v 2 ω2 − ωp − ω = −ωp γ − γ2 − 1 . c c (2.80)

This can also be written as 1 v2 L(r , v, t) = − m eff c2 = −ωp (r , t) 1 − 2 γ c
1/2

.

(2.81)

The photon momentum k, and the force acting on the photon f , are determined by k= and f = ∂L ∂ v2 = −ωp 1− 2 ∂v ∂v c ∂L v2 =− 1− 2 ∂r c
1/2

=
1/2

ωp γ v = m eff γ v c2

(2.82)

∂ωp ωp ∂ωp =− . ∂r ω ∂r

(2.83)

The Newtonian equation of motion (2.75) can then be written as ωp ∂ωp dk =− . dt ω ∂r (2.84)

The same kind of description can be used for electromagnetic wave propagation in a waveguide. The effective mass of the confined photons will then be equal to m eff = ω0 /c2 , where ω0 is the cut-off frequency for the specific mode of propagation. The same equations of motion apply here, but only for one-dimensional propagation (along the waveguide structure) and, for non-stationary and nonuniform waveguides, the same effective mass changes and frequency shifts are expected. The implications of the effective mass of a photon in a waveguide were considered, in a somewhat speculative maner, by Rivlin [90]. In general terms, a photon always propagates in a medium with velocity less than c. In the language of modern quantum field theory it can be considered as a ‘dressed’ photon, because of the polarization cloud. For a generic dispersive dielectric medium, the photon velocity is determined by v≡ ∂ω ∂k = ∂ kc kc ∂ = √ ∂k n ∂ k 1 + χ (ω) (2.85)

26

Photon ray theory

where χ (ω) is the susceptibility of the medium. We then have √ c 1+χ v= . (1 + χ ) + (ω/2)(∂χ /∂ω)

(2.86)

The relativistic γ factor associated with the photon in the medium is then, by definition, γ = 1− v2 c2
−1/2

=

(1 + χ ) + (ω/2)(∂χ /∂ω) . {[(1 + χ ) + (ω/2)(∂χ /∂ω)]2 − (1 + χ )}1/2

(2.87)

2 In the case of a plasma, we have χ = −ωp /ω2 = −1/γ 2 . This is in agreement with equation (2.78), as expected. For a non-dispersive medium we have ∂χ /∂ω = 0, and we get the following simple result:

γ = 1+

1 χ

1/2

=√

n n2 −1

.

(2.88)

The gamma factor tends to infinity when the refractive index tends to one (the case of a vacuum) because the photon velocity then becomes equal to c. Let us now establish a definition of the photon effective mass, valid for a dispersive optical medium. Using Einstein’s energy relation for a relativistic particle, we get m eff = ω . γ c2 (2.89)

This means that, in general, m eff is a function of the frequency ω. Exceptions are the isotropic plasma and the waveguide cases. As a consequence, we are not allowed to write the dispersion relation as ω = (k 2 c2 + m 2 c4 )1/2 , except for eff these important but still exceptional cases, because for a refractive index larger than one this would simply imply imaginary effective masses, in contrast with the well-behaved definition stated above. For instance, for a non-dispersive dielectric medium with n > 1 we have, from equation (2.89), m eff = ω n 2 − 1. nc2 (2.90)

Let us now write the Lagrangian for a photon in a generic dielectric medium. Using equation (2.71), we have L = −ω 1 − v 1+χ . c (2.91)

2 This reduces to equation (2.80) for a plasma, where χ = −ωp /ω2 . In con√ trast, for a non-dispersive medium, we have v = c/ 1 + χ and the Lagrangian is equal to zero. Notice that, in general, we are not allowed to write L = −m eff c2 /γ , except for plasmas and waveguides.

Covariant formulation
From equation (2.73) we can calculate the force acting on the photon f = ∂L ∂ 1 ω ∂χ = ω ln n = . ∂r ∂r 2 (1 + χ ) ∂ r

27

(2.92)

2 This reduces to equation (2.83) for χ = −ωp /ω2 . We will now make a final comment on the nature of this force. For a nonuniform but stationary medium, the force acting on the photon is responsible for the change in the photon velocity according to the usual laws of refraction: a gradient of the refractive index always leads to a change in the photon velocity, as stated by this expression for the force. We have then a variation in the relativistic γ factor: ∂γ /∂t = 0. However this is not equivalent to photon acceleration, because the total energy of the photon remains unchanged: ∂ω/∂t = 0. The reason is that the variation in velocity (or in kinetic energy) is exactly compensated by an equal and opposite variation in the effective mass (or in the rest energy)

∂ ∂ ln γ = − ln m eff . ∂t ∂t

(2.93)

If, in contrast, the properties of the medium are also varying with time, this equality is broken and we can say that photon acceleration takes place. The medium exchanges energy with the photon, and the photon frequency (or its total energy) is not conserved: ∂ω/∂t = 0.

2.5 Covariant formulation
The similarities between space and time boundaries noted in the preceding sections suggest and almost compel the use of four-vectors, defined in relativistic space–time. Let us then define the four-vector position and momentum as x i = {ct, r }, ki = ω ,k . c (2.94)

The flat space–time of special relativity [43, 55] is described by the Minkowski metric tensor g i j , such that g 00 = −1, g ii = 1 (for i = 1, 2, 3) and g i j = 0 (for i = j). From the relation k i = g i j k j , we get k i = ki (for i = 1, 2, 3) and k 0 = −k0 = ω/c. The total variation of the photon phase (or action) will then be given by
3

dψ = k · dr − ω dt =
i=1

ki dx i − k 0 dx 0 = ki dx i

(2.95)

where, in the last expression, we have used Einstein’s summation rule for repeated indices.

3. The square of the four-vector momentum is determined by 3 ω 2 ki k i = −(k0 )2 + (ki )2 = k 2 − . we simply have 2 ωp ki k i = − 2 = −m 2 c2 . but a function of the other variables k0 ≡ k0 (k1 . k2 . we obtain the above three-dimensional canonical equations. x i ). (2. as δψ = δ ki dx i = 0. The invariant I1 can also be written in the same notation as I1 = −u i k i (2. 2.96) The photon canonical equations can now be written as dx i ∂k0 =− .99) c j=1 Noting that k = ωn/c.100) This is the dispersion relation in four-vector notation. x i ). (2. ∂ki dx 0 dk i ∂k0 = .28 Photon ray theory We can then establish an expression for the variational principle (2. x i ) 3 1/2 kjk j j=1 . in these equations. k3 . the momentum component k0 is not an independent variable.102) 2 where K j = { b /c = kb · u/c. For a generic dielectric medium. the moving boundary between two dielectric media can be described by the law n n(x i ) = n 1 + 1 + tanh(K j x j ) (2. written in a more sophisticated notation.66).97) Notice that.103) . (2. kb } is the four-vector defining the space–time properties of that boundary. It should also be noticed that. the second of these equations simply states that dω/dt = ∂ω/∂t. (2. we have k0 = − 1 n(k0 . For i = 1. ∂ xi dx 0 (2.101) eff c Now. The explicit form of this function depends on the properties of the medium. and using n = √ 1 + χ .98) We can certainly recognize here the dispersion relation ω = kc/n. (2. we obtain ki k i = (k0 )2 χ (k0 . and the first one is an identity. In the case of a plasma. for i = 0.

ki ) such that these new canonical equations are really equivalent to equations (2. First. We notice that. it stresses the fact that the apparently distinct phenomena of refraction and photon acceleration are nothing but two different aspects of a more general physical feature: the photon refraction due to a variation of the refractive index in the four-dimensional relativistic space–time. In this case.106) In contrast with the above equations of motion (2. for a plasma. it can be useful for future generalizations of photon equations to the case of a curved space–time. 76]. Second. and a new Hamiltonian function such that h(x i .107) . Given the formal analogy between equation (2. This function is sometimes called a super-Hamiltonian [33. to be identified later with the photon proper time. What we have written until now is nothing but the equations already established for the photon equations of motion in the new language of the fourdimensional space–time of special relativity. For arbitrary variations δki and δx i .104) Here we have used a timelike variable τ . but not obvious for an arbitrary dielectric medium.12). The question here is how to define the appropriate function h(x i . ki ) = 0. u} was used. ki ) = kjk j 1 + m eff c2 . We can now make a further qualitative jump. we can use the analogue of the covariant Hamiltonian for a particle with a finite rest mass: h(x i .97). (2.105) where the integration is performed between two well-defined events in space– time. The covariant form of the variational principle (2.95) and the phase defined in the usual three-dimensional space coordinates. ki ) dτ. these new equations are formally analogous to the three-dimensional canonical ray equations (2. at least. we can derive from here the photon canonical equations in covariant form: dx i ∂h . This new way of writing can be useful in. 2m eff 2 (2. two different ways.96) can now be written as δφ = δ = =0 ki dx i − h dτ δki dx i − ∂h dτ ∂ki − dki + ∂h dτ δx i ∂ xi (2.97). dτ ∂x (2. the photon effective mass is not a function of the frequency. where gravitational effects can also be included. we can generalize it and write dψ = ki dx i − h(x i . = dτ ∂ki dki ∂h =− i.Covariant formulation 29 where the four-vector velocity u i = {c. This is simple for the plasma case.

. ki ) = γ kc γc −ω = k0 + n n(x i ) 3 1/2 k i ki i=1 . 2ωp (x i ) (2. ki ) = 0 has to satisfy the following conditions: ∂h = −γ . for i = 0.106) leads to the appropriate definition of the relativistic γ factor.109) (2. In the general case of a dielectric medium with refractive index n (where the photon effective mass is. ∂h ∂ = −γ ω ln n ∂t ∂t ∂h ∂ = −γ ω ln n. 2. we see that these four conditions are satisfied by the covariant Hamiltonian h(x i .107) was first stated.108) It can easily be seen that. The covariant formulation was considered in reference [65]. ∂ω and ∂h = γ v.110) ∂k For a non-dispersive medium. it leads to equations (2. in general. a function of the photon frequency).111) The photon ray theory for non-stationary plasmas was first proposed by us [66]. For i = 1. ∂r ∂r (2. This work has been considerably extended in the present chapter. (2. the covariant Hamiltonian h(x i .12).30 Photon ray theory This is equivalent to h(x i .106) are indeed an equivalent form of the photon canonical equations. 3. where the Hamiltonian (2. the use of this Hamiltonian function in equations (2. just showing that equations (2. where v = c/n. ki ) = (k j k j )c2 1 + ωp (x i ) i) 2ωp (x 2 1 2 = k 2 c2 + ωp (x i ) − ω2 .

Both kinds of perturbation can be excited in a plasma by an intense laser pulse. They can also be analytically described in a simple way. similar to the moving boundary between two media considered in the previous chapter. The motion of this boundary is independent of the motion of the atoms. Because the photoionization is a fast process. 47]. The moving boundary will be the ionization front created by the laser pulse. ions or electrons of the medium. The second is a wave-type perturbation of the refractive index.1 Ionization fronts One simple way of producing a moving discontinuity of the refractive index is to create an ionization front. occurring within a timescale of a 31 . 3. which means that we can eventually produce a relativistic moving boundary in a medium where the particles stay nearly at rest.Chapter 3 Photon dynamics Two different kinds of moving perturbation can be imagined in a non-stationary medium. The first one is a shock front or moving discontinuity. As a particular example of photon acceleration in a dielectric medium. we will discuss the ways in which the same kind of perturbation can also be excited in magnetized plasmas and in other optical media. we will consider the well-known induced phase modulation processes. At the end of this chapter. which is the boundary between the neutral state and the ionized state of a given background gas. in the presence of ionization fronts and of wakefields. and the wave perturbation will be the wakefield left behind the pulse. The ionization front is one of the main problems in the theory of photon acceleration in plasmas. the ionization front is the boundary between a neutral and a plasma medium. In other words. A relativistic front can be efficiently produced by photoionization of a gas by an intense laser pulse [4. as referred to in chapter 1. For the sake of simplicity we will focus here on the dynamical properties of the photon motion in a non-magnetized plasma.

the photoionization process cannot be isolated from other ionization mechanisms because the primary electrons (created and subsequently accelerated by the laser field) will also contribute to the ionization process. behind the ionizing laser pulse. as given by the one-dimensional ray equations dq ∂ = . The formation and propagation of such a front is a very complicated process. Furthermore. the plasma state will have a much longer lifetime than the duration of the pulse. We will therefore restrict our attention to the dynamics of individual probe or test photons. We restrict our analysis to high-frequency photons. because of the long timescales of the diffusion and recombination processes. p = k and u is the front velocity. which will be discussed later. . First of all.32 Photon dynamics laser pulse electron density n e / n0 q=x-ut Figure 3.1) where q = x − ut. which will eventually destroy it. Ionization front scheme: an intense laser pulse photoionizes a background gas. which is pertinent for laser propagation. dt ∂p dp ∂ =− dt ∂q (3. its spectral content is also changing. On the other hand. The resulting plasma boundary follows the pulse. It is only interesting to note that. due to the self-frequency shift. for that reason. its group velocity cannot be determined in a simple way. the front will tend to replicate the form of the ionizing laser pulse and will move with nearly the same velocity. or acceleration of the laser photons by the laser itself.1. few femtoseconds. the laser pulse is moving at the boundary between two different media and. interacting with a given ionization front. moving at nearly the same speed. Here we are not interested in the details of the ionization front structure. by electron-impact ionization [60]. We start with a discussion of the different types of photon trajectory.

Other. is f (q) = exp(−k 2 q 2 ). One possible choice. but will not significantly modify our view of the various possible classes of photon trajectories. From equation (3. we can say that a reversal in direction of the photon trajectory in phase space (q.Ionization fronts 33 We can then neglect the contribution of the ion response and. for q ≥ 0. by a Gaussian pulse. more complicated but more realistic forms of front profiles can also be proposed. p)   2 1 ωp0 p(q) = γ 2 sβ ± 1 − 2 2 f (q)  (3. we have ωp0 = f (qr ).2) we can determine the photon trajectory p(q) in the twodimensional phase space (q. the Hamiltonian function is simply determined by ≡ (q.6) . for a given point q = qr . t) = 2 p 2 c2 + ωp0 f (q) − up. This allows us to establish an upper limit ωq for the frequencies of those photons with trajectories which reverse sense in phase space (q.3) c γ where β = |u|/c. for an isotropic or non-magnetized plasma. p.3). ∂q (3. (3. we can say that the photon trajectory will be sooner or later reflected in phase space if < ωp0 /γ . First.2) Here. For a photon with initial frequency ω1 . p) will occur if ∂p → −∞.5) γ Noting that the maximum value for f (q) is equal to 1. which moves with a positive velocity along the x-axis (or. already mentioned in chapter 2. where k f determines the front width. for fronts created by a single step ionization of the background neutral atoms. p) ω1 < ωp0 1+β 1 − β 2 ≡ ωq . along the q-axis) and then interacts with a front moving in the oposite direction (such that s = −1). we see that such a condition is satisfied if. s is the sign of the velocity u and γ −2 = (1 − β 2 ). we can write = ω1 (1 − sβ) = ω(1 + β). is f (q) = [1 + tanh(k f q)]/2. (3. for f q < 0. Another possible choice. and f (q) = 1. equivalently. From this result we can establish two different conditions for photon reflection by the front. (3.4) Using equation (3. ωp0 is the maximum value of the electron plasma frequency attained by the ionization front and f (q) is a function increasing with q from 0 to 1 and describing the form of the front.

7) More generally. If we assume that the parameter ωp0 is fixed and if we vary the value of the incident photon frequency ω1 . These three types of trajectory are illustrated in figure 3. (3. we had assumed a fixed incident frequency ω1 and had increased the value of the maximum plasma frequency of the front. for low-frequency photons. This result is identical to that of reflection of an incident photon in a relativistic mirror with normalized velocity β. 3.6. independently of the choice of the form function f (q). for high-frequency photons with frequencies such that ω1 > ωx —there is transmission in both the real space and the phase space. in other words.8) ≡ ωx .10) where ω2 is the final value of the frequency. type II trajectories for moderate densities ωa < ωp0 < ωb . as already shown in chapter 2.34 Photon dynamics On the other hand. for moderate values of the frequency. we have reflection in real space (a reflection observed in the laboratory frame of reference) if there is a turning point x where the photon momentum reverse changes its sign: p = k = 0. We are now ready to establish the distinct regimes of photon propagation across the ionization front or. where ωb = γ ωa . which means that the photon can still cross the entire front region.2. (b) type II trajectories. we successively obtain: (a) type I trajectories. 1+β From equations (3.9) 1 − β2 This shows that we will always have ωx > ωq . such that ωq < ω < ωx —there is reflection in real space but transmission (no reversal) in phase space. The maximum values for the photon frequency shifts can easily be obtained from the invariance of . but the front moves faster. where ωa = ω1 (1 + β). such that ω1 < ωq —there is reflection in both the real space and the phase space. This means that the reflected photon and the front move both in the backward direction. (3. and type I trajectories for high-density trajectories. we can say that reflection in real space occurs if < ωp0 . . we have ω2 = ω1 1+β 1−β (3. (c) type III trajectories. the different types of trajectory in phase space. we would have type III trajectories for 0 < ωp0 < ωa . For type I trajectories. If instead.8) we can get a simple relation between the two upper limits ωq ωx = .3) this implies that = ωp0 f (q). According to equation (3. This allows us to establish a new upper limit ωx for the frequency of photons with this kind of behaviour ωp0 ω1 < (3. such that ωp0 > ωb .

2x10 6 6 6 6 R eflection.0x10 2. but for a different electron density n e = 1020 cm−3 . and we still have a moving mirror effect. for numerical reasons.0x10 -6.2.0x10 4.3. type I R eflection. type I R eflection. type II 7 Transm ission. as we explained above.Ionization fronts 6. type III C o-propagation.0x10 -4. the particles of the media can stay at rest. type IV (*) 7 -10 -5 0 5 10 q Figure 3. This trajectory would not intersect the others if the plasma parameters were the same. 12 R eflection. type II 10 V acuum w avelength [ µ m ] Transm ission.2.0x10 -1. which is the prefered unit for the experimentalists to denote the measured photon frequency ω.0 -2.9. The same four trajectories as in figure 3. A co-propagating trajectory is also shown. Trajectories of photons interacting with an ionization front. This is a very interesting result because. electron density n e = 5 × 1020 cm−3 . The three counter-propagating types of trajectory are shown for the following parameters: front velocity β = 0. Such an equivalence be- . The boundary between two different states of the medium (the neutral and the plasma state) act then as a moving material obstacle.0x10 -8. but represented in terms of the vaccuum wavelength λ = 2πc/ω. type IV (*) 8 6 4 2 0 -10 0 10 q Figure 3. type III C o-propagation.0x10 6 6 35 6 W avenu m be r [m ] -1 0.0x10 -1.

Such a approach was followed in the experiments by Dias et al [21]. 121]. and its physical meaning. in order to get the full picture of the photon phase space. This limiting case. as discussed in chapter 4. the frequency shift is ω2 = ω1 1 − β 1 − (ωp0 /ω1 )2 . will allow us to determine the two independent parameters of the ionization front: the normalized velocity β and the maximum plasma density or electron plasma frequency ωp0 . we obtain an expression for two of the separatrix curves p(q) = γ ωp0 −β ± c 1 − f (q) . much larger frequency shifts are expected. The interesting thing however is that even in this case a positive shift is expected. These four types of trajectory correspond to four distinct regions in phase space. If we use = ωp0 /γ in equation (3. we can also easily obtain for type II and III trajectories the following frequency transformation: ω2 = ω1 1 − β 1 − (ωp0 /ω1 )2 (1 − β) . such that ω1 ωp0 .12) 2 ω1 1 + β As expected. However. where essentially the co-propagating signal was observed.12). this reduces to an expression similar to equation (3. 1−β (3. in co-propagation. in the denominator. for co. for β 1.3). corresponding to copropagation. the frequency shift for these highly energetic photons interacting with low-density fronts is very small. Using the same kind of argument. will be discussed later. (3. but where (1 + β) is replaced by (1 − β). the frequency shift is half of the value expected for the case of flash ionization (β → ∞). 1−β (3. Actually. who established a proof of principle of the photon accelerated process and also made the first observation of frequency shift in counter-propagation. Previous experiments of frequency up-shift by an ionization front propagating inside a microwave cavity. Furthermore. which are delimited by three separatrix curves.11) For high-frequency photons. This means that. (3.and counter-propagation. were reported by Savage et al [95]. In this case. Several other observations of frequency up-shift in microwave experiments were reported [52. Until now we have only considered photons counter-propagating with respect to the moving front.14) . we have to consider type IV trajectories. this leads to the following frequency shift: 2 1 ωp0 β ω = ω2 − ω1 .13) For high frequencies ω1 ωp0 . two independent measurements of the frequency shifts.36 Photon dynamics tween the ionization front and a material moving mirror is however not complete.

we get the third separatrix ωp0 −β + c 1 − f (q)/γ 2 . with δ 1. with a residual plasma frequency equal to δωp0 .Ionization fronts 37 type III p type I type IV type II q Figure 3. Another possible version of the front assumes that the ionizing laser pulse is propagating not in a neutral gas. Photon interaction with ionization fronts: the three separatrix curves and the four types of trajectory. This model of the front was considered by Kaw et al [46].10). but in a partially ionized gas.17) Here.18) 2 1−β . and the corresponding four regions of phase space. For δ = 0 this reduces to equation (3. we use = ωp0 . (3. For type I trajectories. we have used (ω1 ) = ω1 [1 + β 1 − δ(ωp0 /ω1 )2 ].15) p(q) = γ 2 These three curves. In the above discussion we have used an arbitrary form function f (q) but where f (q) = 0 for q → −∞. the invariance of now leads to the expression ω2 = (ω1 ) 1 + β 1 − δ(1 − β 2 )[ωp0 / (ω1 )]2 . and.4. are shown in figure 3. It can be described by the following form function: 1 f (q) = δ + (1 − δ)[1 + tanh(k f q)]. 1 − β2 (3. this simplifies to √ ω1 ω2 = 1 + β 2 + 2β 1 − δ .4. for δ = 0 but for ω1 ωp0 .16) 2 This improvement in the model slightly changes the photon trajectories and the values of the resulting frequency shifts. instead. (3. but the basic picture of the photon phase space remains valid. (3. If.

instead of equation (3.23) (3. p) = p 2 c2 2 + ωp0 f (q) 1 + χ [1 − f (q)] 1/2 − up.38 Photon dynamics Similarly. for type II and III trajectories. For instance. With this model we still get the same picture of the photon phase space and similar expressions for the frequency shift. If we consider a value n = 1 (but still neglect dispersion in the neutral region).17). (3. for the co-propagating type IV trajectories. 1 − βn   (3. (3.21) Here χ = n 2 − 1 is the susceptibility of the background neutral gas. . for type IV trajectories. we obtain ω2 = ω1 1 − β 1 − (ωp0 /ω1 )2 . we get  2 ωp0 (1 − β)2 1 + βn  ω2 = ω1 1−β 1− 2 1 − β2  ω1 (1 + βn)2 and. for type I trajectories. and a singly ionized plasma is assumed.20) Here we have to use (ω1 ) = ω1 [1 − β 1 − (ωp0 /ω1 )2 ].19) where (ω1 ) is the same as in equation (3. equation (3. we have a neutral gas. we have. Let us now turn to another (and eventually more realistic) model of the ionization front where. ω2 = (ω1 ) 1 + β 1 − δ(1 − β 2 )[ωp0 / 1 − β2 (ω1 )]2 . we have now ω 2 = ω1 1 + βn . in the absence of the front. Finally.13).11) is generalized to ω2 = (ω1 ) 1 − β 1 − (1 − β 2 )ωp0 / (ω1 )2 1 − β2 (3.22) For types II and III. the generalization of the above one-dimensional analysis to the two-dimensional case allows us to determine the cut-off frequencies. In this case. 1 − βn 1/2  (3.24) Let us briefly consider oblique interaction of photons with an infinite ionization front. but where the influence of the neutral gas on the refractive index n is not neglected as it was before. we can use the following model: (q.

which corresponds to accelerated density perturbations. t) = k · (r − R(t)). in the neutral gas region) and is overtaken by the ionization front.25) where β = |u|/c. This leads to q= ∂F ∂k = r − R(t). t) = ωp (r − R(t)). The corresponding frequency shift can be obtained from the expressions of the two invariants I1 and I2 . 3. For normal incidence. which is the case discussed in the previous section. defined in chapter 2. and the result is ω 2 = ω1 1 + β cos(θ1 ) + β|β + cos(θ1 )| .27) For a density perturbation moving with constant velocity.2 Accelerated fronts A front moving with constant velocity can only be considered as an ideal and limiting case. as shown by the above expression.26) This effect was observed in the experiments by Dias et al [21] mentioned already. such that θ1 = 0. defined by ωq = ωp0 1 − β 2 [1 + β cos(θ1 )]2 − (1 − β 2 ) sin2 (θ1 ) (3. This new effect can be called co-propagating relativistic mirror and it is due to oblique interaction with the front. the condition for the existence of a type I trajectory (corresponding to reflection both in real space and in phase space) corresponds to incident photon frequencies ω1 smaller than some limit ωq .28) . k) → (q. But a new and very interesting physical situation occurs when the incident photon is co-propagating in the vacuum region (or. In this case. we would simply have R(t) = vt. k. In order to treat our problem we can easily generalize the above canonical transformation (r . more precisely. This can easily be done by considering a plasma frequency space–time dependence of the form 2 2 ωp (r . Even if the photon travels with a velocity nearly equal to c.29) (3. ∂r (3. It is therefore quite natural to extend the above analysis to the more general situation of accelerated ionization fronts. we have π/2 ≤ θ1 ≤ π and photon reflection can still take place. p).Accelerated fronts 39 In particular. this expression reduces to equation (3. But now we can have d2 R/dt 2 = 0. (3. using the following generating function: F(r .6). p= ∂F = k. 1 − β2 (3. its velocity perpendicular to the front can be much smaller due to oblique propagation.

We will illustrate these new aspects. or when the photon coordinates are such that −1/k f < q(t) < 0. ωp (q) =  2 (q > 0) ωp0 (3. we can write ωp (q) = ωp0 f (q). ∂t (3. the velocity of the front ∂ R/∂t is not constant. we will assume an even simpler form for the ionization front. the photon frequency is determined by the expression ω(t) = 2 p 2 (t)c2 − ωp0 k f q(t). For co-propagation (photons and front propagating in the same direction) we can improve the frequency up-shifting by increasing the interaction time. allowing us to determine q(t) and p(t). For counter-propagation (photons and front propagating in the opposite sense) two sucessive reflections can take place. (3. They can be written as dq pc2 ∂R = − dt ω(t) ∂t 2 ωp0 k f dp 1 = . described by  (q < −1/k f ) 02 2 ωp0 (1 + k f q) (−1/k f < q < 0) . which means that is no longer a constant of motion. But.33) In order to obtain an explicit expression for the frequency shift. by focusing our discussion on the one2 2 dimensional problem.40 Photon dynamics The transformed Hamiltonian is =ω+ ∂F = ∂t 2 p 2 c2 + ωp (q) − p · ∂R .33) allows us to write c2 q(t) = p 2 (t) 2 + Q(t) (3. It is also clear that the photon frequency will only change as long as the photon travels inside the gradient region.32) Inside that region. dt 2 ω(t) (3.30) Now. we have to consider the photon equations of motion.31) The width of the front is obviously equal to 1/k f . As before.36) ωp0 k f . (3. instead of using our tanh model.35) These equations can easily be integrated by noticing that equation (3. New qualitative aspects of photon acceleration can be associated with accelerated fronts.34) (3.

just by using a very small (but accelerated) plasma density perturbation. because of the photon acceleration process itself. The photon velocity can be assumed to be nearly equal to c and . (3. Let us illustrate this idea with the simple case of a front moving with constant acceleration a: 1 R(t) = v f t + at 2 . Going back to equations (3.41) 2 In this case. It means that the closer the photon velocity is to the front velocity v f .40) where ω0 ≡ ω(t = 0) is the initial value of the photon frequency (just before entering the acceleration region). 3.39) is Q 0 = −ω0 /(ωp0 k f ). we have √ R(t) = v f t. This expression is valid as long as the photon stays inside the gradient region (3. This expression shows that the photon frequency grows with t. as first noticed by Esarey et al [27].35) we realize that the total time derivative of equation (3. equivalently. we can imagine an ideal situation where the photon copropagating with the front is indefinitely accelerated up to arbitrary high frequencies.39) (3. for very high-frequency photons).33.37) Using equations (3. In the particular case of a front moving with constant velocity. 3. 2 2 The constant of integration in equation (3. as can be seen from equation (3. we obtain ω(t) = 2 2 ω0 + ωp0 k f R(t) (3.32). (3. such that 2 2 ω0 ωp0 .34. 2 ωp0 k f 41 (3.38) dt ∂t This can be integrated to give Q(t) = Q 0 − R(t). we can estimate the maximum frequency up-shift for very underdense fronts (or. the longer t will be and the larger the final value of the frequency shift will be.Accelerated fronts with Q(t) = − ω2 (t) . the photon will have a tendency to escape from the gradient region. Such a slippage can however be avoided (or at least significantly reduced) if the ionization front is accelerated as well. Then.36) leads to dQ ∂R =− . The external source responsible for the creation of such an accelerated plasma density perturbation would provide the energy necessary to accelerate the co-propagating photons.37). as already shown in the previous section. and a phase slippage between the photons and the front will always exist. However.37). in principle.

3. we can simplify this expression and.45) For positively accelerated fronts (a > 0).42 Photon dynamics 1/k f . dt 2 −Q(t) The photon trajectory will then be determined by p(t) = p0 + ωp0 kf 2 t 0 (3.40). we can go back to equations (3. we can get dp 1 ωp0 k f = √ . from equation (3.44) √ dt . 3. The frequency shift is still increased by a positive front acceleration. 2(1 − β f )2 c2 (3. we can write ω ≡ ω(tin ) − ω0 2 ωp0 2 2ω0 βf 1 − βf 1+ a .46) With equation (3.43) The first term in this expression coincides with that derived in the previous section for fronts moving with constant velocity.47) . especially due to the factor (1 − β f )3 .34. using equation (3. This illustrates our previous statement that. Until now we have focused our qualitative discussions on photons copropagating with the front. The second one gives the contribution of the front acceleration. a qualitatively new effect occurs for co-propagating photons: more than one turning point.40) and derive explicit expressions for the photon trajectories.42) the total interaction time can be determined by the equality ctin − R(tin ) The solution is tin = c (1 − β f ) − a (1 − β f )2 + 2a/(k f c2 ) where we have used β f = |v f |/c. except that in equations (3. this completes the explicit integration of the photon trajectories. In particular.42. For small accelerations. this leads to p(t) = p0 + ωp0 k f /2a ln ω(t)ωp0 k f /2a ∂ R ω0 + ωp0 vk k f /2a ∂t . by increasing the time of interaction between the photons and the front. We can see that a positively accelerated front (a > 0) introduces a significant increase in the total frequency up-shift.35). defined by dq =0 dt (3.35. such that atin 2c(1 − β f ). (3. If we are not entirely satisfied with these qualitative arguments and want to replace them by a more detailed description of the photon dynamics. R(t ) − Q 0 (3.36). But it can easily be seen that the above equations stay valid for the counter-propagating photons. 3. (3.43) the factor (1 − β f ) is replaced by (1 + β f ). we can optimize the photon acceleration process. However.

Otherwise. 92] is valid for long laser pulses. This means that a counter-propagating photon can be first reflected by the front.1 Generation of laser wakefields It is well known that a plasma can support several kinds of electrostatic wave and oscillations.48) where ω and k are here the frequency and wavenumber of the electrostatic oscil√ lations. and reflected a second time into the co-propagating direction.3 Photon trapping 3.3. These waves will not be Landau damped. where λDe = vthe /ωp is called the electron Debye length. Such a damping is due to the energy exchange between the waves and the electrons travelling with a velocity nearly equal to the wave phase velocity vφ = ω/k (the so-called resonant electrons). because resonant electrons will be nearly absent (except for extremely hot plasmas). 108]. the phase 2 velocity of the electron plasma waves can be arbitrarily high: vφ vg = 3vthe . One of the basic types of such waves is called an electron plasma wave. we can say that there are two different excitation mechanisms: the beat-wave and the wakefield. 3. Using a simplified but nevertheless accurate view of this problem.Photon trapping 43 can be observed along the same trajectory. the electron plasma waves are characterized by the following dispersion relation: 2 2 ω2 = ωp + 3k 2 vthe (3. and vthe = Te /m is the electron thermal velocity and Te the electron temperature. As can be seen from the above dispersion relation. Relativistic electron plasma waves can be excited by intense laser beams in a plasma. 77. It is then possible to excite electrostatic waves with relativistic phase velocities vφ c. with a duration −1 t much larger than the period of the electron plasma wave: t ωp . suffering the corresponding double Doppler shift. The existence of the electron Landau damping implies that the frequency of the allowed electron plasma waves is always very close to the electron plasma frequency. To be more precise. they will be strongly attenuated by electron Landau damping. This is equivalent to saying that the electron plasma waves can only exist if their wavelength is much larger than a characteristic scale length of the plasma λ λDe . In this . and then it is caught again by the front. The first mechanism [64. because its frequency is nearly equal to the electron plasma frequency [82. ω ∼ ωp . The electron plasma waves can only persist in a plasma if their wavenumber satisfies the inequality k ωp /vthe . and nearly zero group velocities. These doubly reflected trajectories were numerically identified by Silva [99] and can attain considerably large frequency shifts.

on the possibility of accelerating and trapping photons in the field of an electron plasma wave.44 Photon dynamics laser pulse wakefield Figure 3. ω1 and ω2 . The phase velocity of this wave is determined by the laser beating vφ = (ω1 − ω2 )/(k1 − k2 ). we will simply refer to a relativistic electron plasma wave as a wakefield. the superposition of two parallel laser pulses with different but very close frequencies. It is important to notice that the phase velocity of this wakefield is nearly equal to the group velocity of the laser pulse. 89] is valid for short laser pulses. In order to describe such a wakefield. Interaction of these waves with fast electrons can accelerate them further and provide the basis for a new generation of particle accelerators [109]. But. case. we will assume a plasma .2 Nonlinear photon resonance From now on. such that their difference matches the electron plasma frequency.3. which is usually called the laser wakefield. especially. The problem of electron acceleration is not our concern here. 107]. Schematic of the laser wakefield mechanism. With both the beat-wave and the wakefield mechanisms. The second mechanism [12. it should also be mentioned that the study of photon acceleration and trapping by a relativistic plasma wave can be an important diagnostic tool for particle accelerator research [22. A single laser pulse can then excite a tail of electron plasma oscillations. relativistic electron plasma waves can be excited.5. A short laser pulse propagating in a plasma excites a relativistic electron plasma wave. 34. Our interest is mainly focused on the analogies between charged particles and photons and. ω1 − ω2 ωp can resonantly excite an electron plasma wave. 3. with a −1 duration of the order of the electron plasma period t ωp .

kp ph = sγβ ωp0 √ 1+ . c (3. 2π . kp pe = sγβ ωp0 √ 1− c (3. From equations (3. (3. (3. and q⊥ = p⊥ = 0. or kp /kp .51) where u (ωp0 /|kp |2 )kp is the wakefield phase velocity and 2 ω(q.Photon trapping frequency perturbation of the form 2 2 ωp (q) = ωp0 1 + cos(kp · q) 45 (3. We should notice that. dt (3. p)u. s = sign of u and γ = (1 − β 2 )−1/2 . and the subscript p is added in order to avoid confusion with the photon wavevector.54) dp = 0. where q = q.53) we can then explicitly establish the elliptic fixed points as qe = π . the momenta pe and ph .53) At this point we could split the variables q and p into their parallel and perpendicular components with respect to the direction of the wakefield propagation u/u. it can easily be realized that the perpendicular components play only a secondary role in the photon dynamics. However. in contrast with the simple pendulum. p) 2 ωp0 kp dp = sin(kp · q) dt 2 ω(q. p) = p 2 c2 + ωp0 [1 + cos(kp · q)] 1/2 . sin(kp · q) = 0. p) (3.50) (3. p = p.49) where < 1 is the relative amplitude of the wakefield and kp is the wakefield wavevector. defined by dq = 0.52) Let us determine the fixed points. We will now proceed by just retaining the simple case of one-dimensional motion. dt This is equivalent to writing c2 p = ω(q. The photon equations of motion can be explicitly written as dq c2 p = −u dt ω(q.55) and the hyperbolic fixed points as qh = 0.56) Here we have used β = |u|/c.

60) = ωp0 √ 1− . The photon nonlinear resonance described by equations (3. p) is shown in figure 3. the resonance asymmetry increases with the wakefield relativistic γ -factor.55) can be defined by pe = sγ 2 β where e e c2 (3.6. 3. the distance between the two fixed points can be written as | pe − ph | γβ ωp0 . for the one-dimensional case.59) (3.59). (3. for the hyperbolic fixed point.51) appears then as a kind of asymmetric pendulum [98].46 Photon dynamics p kpq Figure 3. The separatrix curves of the photon nonlinear resonance in phase space (q. Furthermore.61) . γ The value of ph . p).57) A qualitative representation of the photon nonlinear resonance in phase space (q. corresponding to the elliptic and hyperbolic fixed points are not identical. γ (3. p) − up = γ −2 ω(q. where e is replaced by h such that h = ωp0 √ 1+ .58) This means that the elliptic fixed point given by equation (3. For small wakefield amplitudes 1.50. is also determined by an expression similar to equation (3.6. the invariant can be written as = ω(q. It is also interesting to notice that. c (3. p).

the photon frequency will also oscillate.65) c From this analysis we recognize that the photons are trapped by the electron plasma wave if their trajectories are associated with values of the invariant Hamiltonian such that < h. This is illustrated in figure 3. or = e. (3. but such a change will require longer and longer times to take place. (3. for q = (π/kp ). For untrapped photons. between the points p+ and p− defined by equation (3. 1− 1+ h 2 − up)2 = p 2 c2 + ωp (q). oscillating around p = pe . If we introduce q = (π/kp ) + q and linearize the photon ˜ equations of motion around the elliptic fixed point. It is now interesting to look at the deeply trapped trajectories. We will also attain a maximum for the frequency variation at the separatrix. We will have no frequency variation only for exactly resonant photon trajectories such that p = pe . On the other hand. if we wait an infinite time. this reduces to p = ph . (3. we have ω pc c pmax = and the maximum frequency shift will be determined by ωmax √ 2γ ωp0 2 . the photon frequency will oscillate according to the value of characterizing its trajectory. (3. this expression determines the maximum and the minimum values of the photon momentum on the separatrix p± = γ ωp0 √ 1+ c sβ ± 2 1+ .62) (3.63) For q = 0.7.  ωp0 √ p=γ 1 + sβ ± c  1 + cos(kp q) .64). more explicitly. For very high-frequency photons.64) The width of the nonlinear resonance will then be given by the difference between these two extreme values of the separatrix curves: ωp0 √ pmax = p+ − p− = 2γ 2 . (2π/kp ). corresponding to the elliptic fixed point. such that ω ωp0 .67) . but with smaller and smaller amplitudes when we displace the photon trajectories away from the separatrix.66) e ≤ Inside the separatrix. 47 . The frequency variation will grow when we approach the separatrix curve. we can easily obtain d2 q ˜ 2 + ωb q = 0 ˜ dt 2 (3.Photon trapping Replacing this expression in the definition of the invariant Hamiltonian we obtain the equation for the separatrix ( or.

48 Photon dynamics 5 V acuum w avelength [ µ m ] 4 3 2 1 0 -0.9995.7.70) This allows us to write the electron bounce frequency as √ ωbe = ωp0 . = 0. corresponding to the following parameters: β = 0. It is instructive to compare its value with the bounce frequency of an electron trapped in the field of the same electron plasma wave [82] ωbe = (e/m)E 0 kp (3. n e = 1018 cm−3 .2 0. . (3. showing trapped and untrapped motion.68) This can be called the photon bounce frequency. (3.0 0.71) This shows the striking similarities of the photons and electrons trapped in the field of an electron plasma wave. Using Poisson’s equation we can relate it to the electron density amplitude perturbation kp E 0 = en ˜ 0 = en 0 0 .69) where E 0 is the amplitude of the electric field associated with this wave. Other and eventually even more surprising similarities will be discussed later. Examples of photon trajectories around a nonlinear resonance.2 0. which corresponds to a linear oscillator with a frequency ωb = ckp γ 2(1 − ) .4 -0. The photon frequency ω is represented in terms of the vacuum wavelength λ = 2π c/ω.6 -0.6 P osition on the plasm a w ave [P lasm a w avelength] Figure 3. (3.5.4 0.

x 1 ) and k2 − k2 ∂ωp dk0 ∂h 1 = − 0 =− 1 − 21 0 01 c2 dτ 2 ∂x ωp (x . x 1 .108) with 2 2 ωp (x i ) = ωp0 1 + cos(K i x i ) (3.108) is independent of x 2 and x 3 . it is useful to introduce new adimensional variables.80) . In order to write these equations in a more convenient form.76) (3. We will assume a one-dimensional photon propagation by making these two constants equal to zero: k 2 = k 3 = 0. (3. 2. and v= k1 c .Photon trapping 3. In this particular case.78) where τ is the photon proper time.75) (3.77) (3.5. Let us write Ki x i = K 1 x 1 − K 0 x 0 = k x 1 − ω t is. such that x = K 1 x 1 = k x 1. x ) ∂x0 k2 − k2 ∂ωp dk1 ∂h 1 = − 1 =− 1 − 21 0 01 c2 dτ 2 ∂x ωp (x .48). ωp0 (3.106.79) (3. ωp0 u= y = k x0 k0 c . k0 . We can then write the reduced one-dimensional form of the covariant Hamiltonian as h(x 0 . which means that the components k 2 and k 3 of the photon momentum are two constants of motion. x ) ∂x1 (3. its amplitude.3. k1 ) = 2 2 k1 − k0 1 c2 + ωp (x 0 .72) where K i is the four-momentum associated with the plasma wakefield and as before. x 1) 2 2ωp (x (3. the covariant Hamiltonian (2. we can describe the photon trapping by using equations (2.74) The corresponding equations of motion in the relativistic space–time are dx 0 ∂h k0 c2 = =− dτ ∂k0 ωp (x 0 .3 Covariant formulation 49 As an example of an application of the covariant equations established in section 2. x 1 ) dx 1 k1 c2 ∂h = = dτ ∂k1 ωp (x 0 . 0. x 1 ).73) where ω ωp0 is the wakefield frequency and is related to k by the dispersion relation (3.

88) The existence of these two independent constants of motion. such that I = u + β v = const. proves that the motion of a photon in a sinusoidal wakefield is integrable.83). (3.86) ∂x β ∂y Replacing this in the canonical equations (3. dz ∂y (3. (3. h0 = h . where β = ω /k c is the normalized phase velocity of the wakefield.85) This is nothing but the photon dispersion equation. dz ∂v and dv ∂h 0 =− . u) = v 2 − u 2 + 1 + cos(x − β y) 2 1 + cos(x − β y) . we always have h 0 = 0. we obtain dv 1 du =− . as expected. The normalized proper time z. v.82) (3. written in adimensional form.81) We can state the new equations of motion as dx ∂h 0 = . This invariant was already identified in the non-covariant formulation of photon acceleration.87) This expression means that.84) By definition. we can write u 2 = v 2 + 1 + cos(x − β y). it is possible to define another constant of motion I . (3. Noting that the denominator in this equation cannot become imaginary because the wakefield modulation is always less than one. y. dz ∂x dy ∂h 0 = dz ∂u du ∂h 0 =− . are defined by dz = k c dτ. Noticing that h 0 only depends on x and y through (x − β y). It will appear again in the full wave description. . where it corresponds to the wave phase invariance. apart from the Hamiltonian H0 . we conclude that ∂h 0 1 ∂h 0 =− . < 1. and Hamiltonian h 0 . Notice that this equation is equivalent to stating that ω − kv is a constant. dz β dz (3. where v = ω /k is the phase velocity of the wakefield. h 0 and I .50 Photon dynamics We can see from the definition of the variable y that K 0 x 0 = β y.83) The explicit form of the normalized Hamiltonian is h 0 (x. ωp0 (3. (3.

3. stochastic acceleration of photons can be used.91) In non-normalized variables this is equivalent to kc2 /ω = ω /k . In very recent days it was discovered that the opposite process can also occur and. (3.93) The value of the invariant I .84) is determined by the stationary condition (in the proper time z). Replacing this resonance condition in equation (3.82). we can write this stationary condition as v + β u = 0. dz p u dy =− dz p (3.92) The centre of this resonance corresponds to (x − β y) = π/2.92).85). from the canonical equations (3.90) (3.94) 2 This completes the characterization of the trapped photon trajectories in the four-dimensional relativistic space–time.Stochastic photon acceleration 51 The position of the nonlinear resonance contained in the Hamiltonian (3. that the maximum excursion of u 2 associated with trapped trajectories inside the nonlinear resonance is determined by δu 2 (1 − β 2 ) = 2 . dz Noting that. For that purpose. defined by equation (3. (3. (3. we obtain u 2 (1 − β ) = 1 + cos(x − β y).4 Stochastic photon acceleration In his famous experiment. or d (x − β y) = 0. The resonance half-width is then given by δu 0 = √ 1 . starting from a nearly monochromatic spectral light source. Isaac Newton inaugurated spectroscopy by decomposing white light into its spectral components with the help of a prism. we can regenerate white light.88). v0 = β 1−β2 = −β u 0 . from equation (3. 1−β (3. . which leads to the following coordinates: u0 = − 1 1−β 2 ≡ −γ . This means that the nonlinear resonance corresponds to the equality between the photon group velocity vg ≡ kc2 /ω and the phase velocity of the wakefield perturbation ω /k . We can also see. we have dx v = . corresponding to √ this particular photon trajectory is equal to I0 = − 1 − β .89) where p = 1 + cos(x − β y) is a positive quantity.

95) Using the canonical variables q = r − u 1 t and p = k. we will assume that the wakefields propagate in the same direction. p) phase plane. The following definitions were used here: si = sign of u i .96) + where 2 cos(k2 · (q − vt))] 1/2 − u1 · p v = u1 − u2 2 2 ωp0 [(k1 /k1 ) − (k2 /k2 )]. in other words.1). (3.98) for i = 1. c (3. We can see from the above Hamiltonian (or from its one-dimensional version) that two nonlinear resonances exist in the (q.4. They are defined by the elliptic fixed points qi = π . 2.65) we can also say that the the width of these two resonances is determined by ( p)i = 2γi ωp0 2 i. distinct wavevectors k1 and k2 and distinct phase velocities u 1 = (ω /|k1 |2 )k1 and u 2 = (ω /|k2 |2 )k2 . for n integer. In accordance with equation (3. we can write the photon equations of motion in the form of equations (3. It is also quite well known that the second of these two resonances (the one corresponding to i = 2) can only be seen in the phase plane if we use a stroboscopic plot of the photon motion. The photon dispersion relation can now be written as 1/2 2 ω = k 2 c2 + ωp0 1 + i=1.99) . βi = |u i |/c and γi = (1 − βi2 )−1/2 . with the new Hamiltonian (which is now time dependent) 2 = ω − u 1 · p = p 2 c2 + ωp0 [1 + 1 cos(k1 · q) (3.1 Photon dynamics Motion in two wakefields Let us first discuss the simple case of photon motion in the presence of two different wakefields (or.2 i cos(ki · r − ωi t) . The extension to three dimensions is straightforward and does not lead to qualitatively new effects. at the instants tn = (2nπ/k2 v).97) In order to simplify the discussion. kp pi = si γi βi ωp0 1− c i (3. Such a discrete representation of the motion is usually called a Poincar´ e map. (3. and we will concentrate on the one-dimensional case. two relativistic electron plasma waves) with amplitudes 1 and 2 .52 3.

99) we can write the overlapping criterion as √ √ γ1 2 1 + γ2 2 2 ≥ 1. (3. the two stochastic regions will eventually merge.101) √ √ |s2 β2 γ2 1 − 2 − s1 β1 γ1 1 − 1 | It is important to notice that this criterion is independent of the mean electron plasma frequency ωp0 .102) ≥ 1 (1 − ν)2 23 (3. and a large fraction of the photon phase space (q. 3. we obtain from equation (3.Stochastic photon acceleration 53 It is well known from the theory of dynamical systems [59.103) Noting that βi are always smaller than one. These curves will both break up into a thin region of stochastic motion and the width of such a region will exponentially grow with the amplitude of the resonances i . where we have γi 1 and ν conclude from this expression that large-scale stochasticity occurs for ≥ 1 (s2 β2 − s1 β1 )2 . and that their relativistic γ factors are similar. In our case it can be written as ( p)1 + ( p)2 ≥ 2| p2 − p1 |.104) where we have assumed that the wakefields propagate in the same direction. With increasing values of i . ν 1. for a large variety of similar nonlinear Hamiltonian motions. If we assume that the two amplitudes are equal. In the weakly relativistic limit. . we can simplify the overlapping criterion and write 1+ν 2 ≥1 1 − |s2 β2 − s1 β1 ν| where ν = γ1 /γ2 .98. we conclude that this criterion is compatible with low values of the wakefield amplitude 1. p) will be filled with stochastic trajectories.100) Using equations (3. and that it only depends on the velocities and amplitudes of the two wakefields. where γi 1 and βi 1. 23 (3. 1 = 2 = . called the Chirikov criterion or the resonance overlapping criterion.102) 1. The criterion states that largescale stochasticity occurs when the sum of the resonance half-widths becomes larger than the distance between these resonances. 124] that the interaction between the two resonances leads to the destruction of the separatrix curves. (3. This is confirmed by numerical calculations within an error of a few per cent. s1 = s2 . leading to what is called large-scale stochasticity. Occurrence of large-scale stochastic acceleration is dictated by a qualitative criterion. we (3. In the opposite limit of strongly relativistic phase velocities of the plasma wakefields.

95) is now replaced by ω= k c 2 2 2 + ωp0 1+ ∞ n n=−∞ 1/2 cos(kn · r − ωn t) .4. but with a modulated amplitude. This model is convenient from the physical point of view because. transition from regular to stochastic trajectories can be expected with moderately low wakefield amplitudes. in reference [65]. A similar situation occurs if. we have three nearby resonances to which we can apply the same overlapping criterion.2 Photon discrete mapping It is also interesting to consider the interaction of a photon moving in a plasma with an electrostatic wavepacket containing a large spectrum of electron plasma waves.54 Photon dynamics We see that. We can then say that. 3. instead of having two different wakefields. it corresponds to a spectrum with variable amplitudes. the criterion is compatible with low wakefield amplitudes 1.107) where a is a constant and k k0 is the characteristic distance between two consecutive spectral components. the strongly relativistic case is physically more interesting. From initial monochromatic radiation we can then obtain broadband radiation (white light) [66]. dt ω 2 ωp0 dk = dt 2ω ∞ n kn n=−∞ sin(kn x − ωn t). (3. However. in contrast with similar models used in nonlinear dynamics where the amplitudes are assumed constant.106) As a simple and reasonable model for the wavepacket. ν 1. to the existence of ion acoustic waves propagating in the background plasma. again. The photon dispersion relation (3. we can use n kn = a. (3. because the width of the region between the two resonances where stochastic acceleration can occur is proportional to the relativistic gamma factors. where . both in the weakly and the strongly relativistic limits. we can excite a single wakefield in a plasma. ω n = ω0 ωp0 . and a larger frequency spread of a bunch of photons with nearly identical initial frequencies can be obtained.105) The corresponding one-dimensional photon equations of motion are dx kc2 = . instead of two nonlinear resonances. kn = k0 + n k (3. Such a modulation can be due. Here. for instance. Even in the most unfavourable situation of very different relativistic factors. Here we will follow a procedure similar to that used by Zaslavski et al [125] for studying charged particles. using the covariant formulation.103) ≥ 1/3. we get from equation (3. in very broad terms. This was studied.

just by using the relation x = |vg |t = (|k|c2 /ω)t.106) can then be replaced by a sin θ ∞ cos(n kx) n=−∞ (3. dt 2|k|c2 n=−∞ (3. such that w= kc2 k 2 c2 |k| = 2 sk 2 ωp0 ωp0 (3. we can write 2 ∞ ωp0 dk δ(x − xn ). The sum in equation (3. and in that case the instant tn+1 will be defined by x(tn+1 ) = xn−1 . This is equivalent to writing k= ωp0 |w|1/2 sw c (3.106). This means that. We can also use the identity ∞ n=−∞ cos(n kx) = 2π k ∞ n=−∞ δ(x − xn ) (3. This is compatible with the idea that electron Landau damping of the wakefield spectrum prevents the higher wavenumbers.Stochastic photon acceleration 55 the lower amplitudes correspond to the higher wavenumbers. with initial frequencies much larger than the electron plasma frequency. Let us now introduce an adimensional variable w. apart from a minor detail resulting from the fact that the photon can eventually be reflected at some point xn . from equation (3. (3.113) . Of course.110) = a L sin θ dt 2ω n=−∞ This shows that the photon moves with a constant wavenumber (or a constant velocity). The above equation is then replaced by 2 ∞ ωp0 dk = a L sin θ δ(t − tn ). This picture of the photon motion means that we can transform the variable position x inside the delta function argument into a time variable.111) The instants tn are defined by x(tn ) = xn .108) where θ = k0 x − ω0 t is the phase of the central spectral component. and so on. This new wavenumber remains constant until it crosses the next point x = xn+1 . except when it crosses the point x = xn . where it suffers a sudden kick and abruptly changes its wavenumber.109) with xn = (2nπ/ k) ≡ n L.112) where sk is the sign of k. such a reversal in the direction of photon propagation is not likely to occur for energetic photons.

dt n (3.114) = a L sin θ dt n=−∞ This has to be completed with an equation describing the time evolution for the phase variable θ . 3. at this point. the variable w suffers a sudden jump.119). (3. with ω in the denominator. − θn = θ(tn ) (3. tn is the interval between two consecutive kicks tn = L ωn+1 ¯ . remain unchanged.115) show that.111) the new variable w will evolve in time as ∞ dw δ(t − tn ). According to equation (3.120) Using equation (3.115) This equation. we should notice that it is physically quite natural to assume that n 1. this allows us to replace ω by ω and to write. or w(tn (3.118) wn+1 = wn + a L sin θn . This is equivalent to stating that w(tn+1 ) = 0 ).115).114. We now have two coupled and closed equations for the variables w and θ . 3.115.115) that the time derivative of the phase θ also stays constant over the same time interval. and the corresponding photon − momentum p. kn+1 c2 (3. we have dθ/dt = (k0 c2 /ω)k − ω0 . Equations (3. But. By definition. tn+1 ) the variable w. at t = tn .116) ± where tn = tn ± δ. We can then write − + θ (tn+1 ) = θ (tn ) + d θ(t + ) tn . According to equation (3. This can be stated as + − w(tn ) − w(tn ) = a L sin θ. dt ω ¯ (3. is also physically very convenient ¯ because it guarantees that θ varies continuously across the points t = tn . and that the phase θ remains constant. and we can build up a map on the new phase plane (w.119) Here.117) + − During the interval (tn . with δ → 0.114. represent the instants imediately before (−) and after (+) the critical instants tn . For that purpose.121) √ c |wn+1 | . (3. equations (3. + − θ(tn ) − θ(tn ) = 0. we define − wn = w(tn ). ¯ using the new variable w ωp0 dθ = k0 c sw − ω0 .105). We can also see from equation (3. 3. θ ). we can then obtain the following result: √ ω0 L 1 + |wn+1 | θn+1 = θn − + k0 Lsn+1 (3.56 Photon dynamics where sw is the sign of w.

123) also shows a constant phase shift determined by θ = k0 L. The . to be specified later. θ ). We will show in this section that a similar mechanism can be applied to photons. For small values of the adimensional variable w. In its simplest and most popular versions. the magnetic clouds of the original model are replaced by mirrors or by plasma walls with a sharp density gradient. In the case of photons. A and B. We can assume that one of the walls is fixed at x = 0 and the other oscillates around x = L 0 with a frequency ω . where charged particles can gain energy by bouncing back and forth between two magnetic clouds.Photon Fermi acceleration 57 where sn+1 is the sign of wn+1 . This means that the photons can suffer intermittency acceleration when they interact with an electrostatic wavepacket in the limit of small photon wavenumbers (when they are close to the cut-off conditions). which can be written as wn+1 = wn + A sin θn ˜ θn+1 = θn − B 1 + |wn+1 |−1 + θ sn+1 . which shows an intermittency behaviour for A 1.125) f (−x) x <0 The function f (x) describes the plasma density profile of the two plasma walls (assuming that they have identical profiles). and that the photon motion can also be described by a discrete mapping [30]. We see that equations (3.124) (3. this mapping reduces to the L-mapping introduced in [125]. determined by A = aL. ωp (x. t) = ωp0 0 (3. the charged particle can be described by two-dimensional discrete maps [59]. This mechanism became extremely successful and it is dominantly used in the current models for cosmic ray acceleration in shocks. c (3.5 Photon Fermi acceleration We shall now discuss a different mechanism for photon acceleration which can occur inside an electromagnetic cavity with moving boundaries. This can be described by a plasma frequency space–time variation of the form f (x − L(t)) x > L(t) 2 2 L(t) ≥ x > 0 .121) define a discrete mapping on the phase plane (w. B= ω0 L. 3. This mapping depends on two parameters.123) ˜ The map (3. 3.122) (3. This can be seen as the photon version of the well-known mechanism for cosmic ray acceleration first proposed by Fermi [29].118. which corresponds to small valˆ ues of the photon wavenumber.

and their initial frequency ωi is transformed into a final frequency ωf . moving wall is supposed to oscillate according to L(t) = L 0 (1 + cos ω t) (3.8. which then remain trapped in a kind of one-dimensional cavity.126) where the amplitude of this oscillation is assumed to be very small: 1.130) . defined by the well-known law ωf = ωi 1+β . The plasma walls are also supposed to be sufficiently dense in order to act as mirrors and to reflect all the incoming photons. c θ =ωt (3.128) sin ω t. Each time the photons are reflected by the moving plasma wall.129) we obtain the following law for the frequency shift after successive reflections at the moving plasma wall: ωn+1 = ωn 1 − b sin θn .127) The velocity of this moving wall is just the time derivative of L(t) and we can write L 0ω β=− (3. 1 + b sin θn (3. they suffer a double Doppler shift. c If we define two adimensional quantities b= L 0ω . Fermi acceleration of photons inside an oscillating cavity.58 Photon dynamics fixed mirror photon moving mirror x 0 L (t) Figure 3. 1−β (3.

(3. c (3. (3.Photon Fermi acceleration 59 Now we have to find a corresponding transformation law for the phase θ between two successive reflections. (3. θ): u n+1 = u n F(θn ) θn+1 = θn + G(u n+1 ). tp . the factor of 4 was introduced in order to account for the four distinct paths inside the two plasma walls. dt (3. Clearly. 3.132) If the time spent inside the moving plasma region is much shorter than the period of the wall oscillations. This leads to tp = 4 c xc 0 dx 1 − (ωp0 /ω)2 f (x) . t) by f (x) in this equation. and replace f (x.133).137) dx 1 − f (x)/u 2 .131) In order to calculate the time spent by the photon inside the two plasma regions. we notice that the photon velocity is determined by: dx = c 1 − (ωp0 /ω)2 f (x.135) (3. The first part is the time spent by the photon in the vaccum region between the two plasma walls. we can use. For 1. This time interval can be divided into two distinct parts: t = t0 + tp . where t is the time spent by the photon between two successive reflections at the moving wall. we can neglect the plasma motion during the process of photon reflection. we can write θn+1 = θn + ω t. as a good approximation t0 = 2 L0 . where the photon frequency equals the electron plasma frequency ω = ωp (xc ). tp (4π/ω ). Let us also introduce a normalized frequency u= ω . and the function G(u n+1 ) is determined by the plasma density profile F(θ ) = 1 − b sin θ 1 + b sin θ 2 L0 xc 0 (3. we can establish the following discrete map on the phase plane (u.138) G(u) = 2b 1 + . This integral extends from the plasma boundary to the cut-off position xc . t).133) Here.130.134) Using equations (3. ωp0 (3.136) The function F(θn ) is solely dependent on the double Doppler shift at the moving wall.

It is also quite useful to determine the fixed points of the map. θn+1 ): |J | = ∂u n+1 /∂u n ∂θn+1 /∂u n ∂u n+1 /∂θn = |F(θ )| = 1. this leads to θm = 0. and their stability. um = α mπ −1 . G(u) = 2mπ. (3. by defining u = u m + u.144) For the parabolic density profile (3. such that b 1. However. Let us first linearize on the variable u. This will give us important information concerning the qualitative aspects of photon motion. ∂θn+1 /∂θn (3. are area preserving. 3. b (3.140) The map (3.143) (3.141) We should note that all these maps are not area preserving. we have F(θ ) = 1 + 2 b sin θ.145) (3.60 Photon dynamics Let us take the particularly interesting example of a parabolic density profile f (x) = α2π 2 2 x L2 0 (3.135. we see that sets consisting of thin layers in phase space.147) .146) The stability of these fixed points can be determined by locally linearizing the map around each of them.138) reduces to G(u) = 2b 1 + u . Then equation (3. because F(θ ) is a periodic function of θ . This can easily be seen by considering the Jacobian of the transformation (u n .139) where the parameter α defines the plasma density slope. α (3.142) This is precisely the factor by which the photon frequency is double Doppler shifted by the moving plasma wall. The first-order fixed points are determined by u n+1 = u n F(θn ) = u n θn+1 = θn + G(u n+1 ) = θn + 2mπ for m integer. In this case. (3.136) also takes a simpler form for very small plasma wall oscillations. ¯ (3. θn ) → (u n+1 . of infinitesimal width δu and extending in phase from θ = 0 to θ = 2π .139). This is equivalent to F(θ ) = 1.

This means that. ¯ α Introducing a new variable I and a new parameter K .141). This means that. we obtain u n+1 = u n + 2 u m sin θn ¯ ¯ 2b θn+1 = θn + u n+1 + 2mπ. the first-order fixed points are all unstable.148) (3.145) requires that |Tr A| < 2.156) Stability of the fixed points (3. we linearize the map on both variables.151) (3. more explicitly.145) of the Fermi mapping for the parabolic density profile. ¯ α K = 4b2 um . if um > α . that α um < 2 .150) this reduces to the well-known standard map. 3. (3. α 61 (3.153) We now examine the stability of the fixed points (3. 4b2 (3. θ ) = (u − u m . first studied by Chirikov [17.157) b We can then say that. u m ) is approximately described by the standard map. the photon motion is essentially . It is well known that such a map suffers a topological transition into large-scale stochasticity if K > 1. 3.155) It can easily be seen that this linear map is area preserving: |A| = 1. and obtain xn+1 = A · xn (3.152) This shows that the photon dynamics around a given fixed point (θm .135. and the matrix transformation A is the ¯ ¯ linearized Jacobian matrix A= 1 2b α −2 bu m . for high enough integers m such that this inequality is not verified. for such values of the photon normalized frequency u.136. near the fixed points. (3. 59]: In+1 = In + K sin θn θn+1 = θn + In+1 . (3. b2 1 − 4 α um (3. or. Following the usual procedure [59].149) (3. the photon motion will become stochastic in a significant fraction of the available phase space. θ − θm ).154) where x = (u.Photon Fermi acceleration Replacing it in equations (3. such that I = 2b u.

From here we conclude that a broad spectrum of radiation (which can be called white light) can be generated from nearly monochromatic light trapped . This is well illustrated by numerical calculations. stochastic. for much lower frequencies. However.9. Regular and stochastic trajectories are shown.62 Photon dynamics Figure 3. Phase space of the Fermi photon map. according to the much less stringent threshold criterion (3. significant stochastic motion had already taken place.153).

158) where ωc = e|B0 |/m is the electron cyclotron frequency. However. or by propagating low-frequency electromagnetic waves. t). They can be produced by non-stationary currents applied to external coils. This can be interpreted as a result of Fermi acceleration of the Lyman α photons by a moving shock [11]. t)) = 0 (3. The photon dispersion relation in a magnetized plasma depends. In some sense the photon Fermi map is similar to the inverted Fermi map for charged particles. In this case [24]. such as Alfven waves. This model of Fermi acceleration of photons could. above which they cannot be accelerated. a glass or an optical fibre. but also if we excite similar forms of moving magnetic field perturbations. in principle. such as ionization fronts and wakefields.157). For instance. Finally. . where the energy axis is turned upside down. we should notice that the concept of photon acceleration in a plasma can be extended to the acceleration of plasmons. but also on the value and direction of the static magnetic field B0 . not only on the electron plasma frequency ωp . or before being absorbed by the walls. indicating a clear blueshift [115]. or quanta of electron plasma waves. Let us then conclude the chapter with some brief comments on the photon processes associated with moving perturbations of the refractive index in magnetized plasmas and in non-ionized optical media. be experimentally tested using an oscillating optical cavity. when they are trapped inside an unstable cavity. such as a neutral gas. It is interesting to compare this qualitative aspect of the photon Fermi map with the usual Fermi mappings for charged particles [59] where. We can represent it generally as R(ω. ωc (r . an upper limit exists for particle energies.6 Magnetoplasmas and other optical media In this chapter we have only focused on processes occurring in isotropic plasmas because most of the published work on photon acceleration concerns this medium.Magnetoplasmas and other optical media 63 inside an oscillating plasma cavity. the spectrum emitted by some very high-redshift radio galaxies reveals a strong asymmetry in the Lyman α line profile. our theoretical approach remains valid for other optical media. 3. if the cavity parameters were conveniently chosen and if the photons could bounce back and forth several times inside the cavity. before escaping out. in contrast with the threshold criterion given by equation (3. not only by using moving electron density perturbations. This means that we can expect to obtain photon acceleration. the plasmon Fermi acceleration will mainly lead to a change in the plasmon wavenumber because the plasmon frequency is always nearly equal to the electron plasma frequency. It can also be useful in the context of astrophysics. k. ωp (r .

will significantly alter the value of the refractive index and will lead to a frequency shift. we have ωc > 0. This means that a small space–time change in the cyclotron frequency will lead to a significant change in the refractive index for the R-mode. which is left circularly polarized. if we assume photon propagation in a direction perpendicular to the static magnetic field B0 . In order to calculate this effect with the aid of the canonical equations for the photons we have to make use of the Hamiltonian function.160) where both the electron plasma frequency and the upper-hybrid frequency are functions of r and t. for propagation parallel to the static magnetic field. valid for the extraordinary mode. which will eventually reduce . we can easily get ω(r .64 Photon dynamics The explicit expression of this dispersion relation is quite complicated. Because. This is the so-called cyclotron resonance. for a photon frequency close to this resonance frequency. and the minus sign corresponds to the R-mode. However.159) 2 ω2 − ωuh 2 2 where ωuh = ωp + ωc is called the upper-hybrid frequency (the lower-hybrid one would only appear for low-frequency waves where the influence of the ion motion has to be retained). by definition. with photons elliptically polarized in the plane perpendicular to B0 . Similarly. which is right circularly polarized. while the R-mode is resonant for ω = ωc . for which the dispersion relation is identical to that considered before for a non-magnetized plasma. with photons linearly polarized along B0 . a small time change in the static magnetic field. there are two distinct polarization states. The other corresponds to the extraordinary mode. From the above dispersion relation. ω ± ωc (3. or equivalently in the electron cyclotron frequency ωc . we can see that the L-mode has no resonances.161) The plus sign corresponds to the L-mode. t) = + 1 2 2 2 2 (k c + ωp + ωuh ) 2 1/2 1 2 2 2 4 (k 2 c2 + ωp + ωuh )2 − 4(k 2 c2 ωuh + ωp ) 2 (3. we have to take into account the wave absorption mechanisms. even in the limit of high-frequency waves for which the ion dispersion effects can be neglected. we have two photon polarization states. but we can still consider some simple examples. The dispersion relation for this new mode is 2 ω 2 − ωp 2 ω 2 = k 2 c 2 + ωp (3. One corresponds to the ordinary mode. k. in dealing with photon motion very close to resonances. First. with the corresponding dispersion relation 2 ω 2 = k 2 c 2 + ωp ω . We see that.

It depends on the density of atoms or molecules in the medium and on the value of the transition probabilities between these two quantum levels. which explains why the refractive index is usually greater than one (in contrast with the plasma case where it is less than one). be done with the aid of an externally applied electric field. It should be noticed that. the voltage signal will propagate along the capacitor arms and produce a moving transition between two different values of the refractive index of the optical medium [28].Magnetoplasmas and other optical media 65 the efficiency of the expected frequency shift or photon acceleration processes. Let us now turn to non-ionized optical media. However. because it completely absorbs the resonant photons) can be reduced exactly to the vacuum value. We can then imagine a physical configuration where the optical medium is located inside an elongated capacitor. The first one is to change the transition frequency ω0 . In this way. by inhibiting transitions between the two quantum states with the help of an auxiliary light source. The closer we are to the resonance condition. corresponding to a complete transparency. Two ways can be foreseen to produce a space–time change in the refractive index and a subsequent photon acceleration in such media. for visible light propagating in the usual optical media. but we can also imagine similar effects resulting from nonlinear mechanisms. For ω ω0 this dispersion relation reduces to that of the non-magnetized plasma. for instance. This can. The photons of this auxiliary source couple one of these two energy states with a third one.162) where ω0 is the frequency of the nearest resonant transition between two quantum levels and ω f plays the role of the plasma frequency. A more drastic change in the refractive index can be obtained by using electromagnetic induced transparency (EIT) [38]. For instance. we have ω ω0 . which can be interpreted as the result of photon acceleration. the refractive index of a nearly resonant medium (which is opaque. The resulting Stark effect will lead to a detuning (quite often a splitting) of the atomic transition energy levels. By applying a sudden voltage signal at one extremity of this capacitor. the stronger will be the change in the refractive index. in recent experiments with EIT produced by auxiliary short laser pulses. These ways of changing the refractive index of a medium are essentially linear. This corresponds to exactly equating to zero the value of ω f . The typical form for the linear dispersion relation in these media is ω =k c + 2 2 2 ω2 ω2 f 2 ω 2 − ω0 (3. These absorption mechanisms are well known in plasma physics and will not be discussed here. it is known that a strong laser beam can produce a change in the . a large spectral broadening is observed [39].

because n 2 is positive. because the phase is completely absent from our photon canonical equations. Here we only would like to stress that the photon dynamical theory discussed in this chapter can equally well treat the cases of co-propagating and counter-propagating photons. The problem of the influence of the field phase in the so-called phase modulation effects is very interesting and will be discussed in detail later. If photons having a different frequency and belonging to a probe beam copropagate with the first one and cross the strong-beam boundary (because their group velocities are different). and positive for a photon crossing the rear of the beam. t) is the intensity profile of the strong laser beam.66 Photon dynamics refractive index. in contrast with the ionization front case which led to a decrease in the refractive index. they will suffer a frequency shift described by the above canonical equations and similar to the one observed for the ionization front. t) (3. but it is currently known as induced phase modulation. according to n = n 0 + n 2 I (r . The difference here is that the frequency shift will be negative for a photon crossing the front of the strong beam. It is quite clear that the field phase is not an essential aspect of the problem. in contrast with the usual theory of induced phase modulation which is only used for the co-propagation case [3]. and we still get a frequency shift.163) where n 0 is the linear refractive index. . n 2 is the nonlinear refractive index (proportional to the nonlinear susceptibility of the medium) and I (r . This effect is well understood in the frame of our single photon dynamical approach.

not as a single photon. For that reason it is also interesting to derive photon conservation equations which allow us to describe the electromagnetic field as a fluid of identical particles. 67 . which is produced when a strong laser pulse propagates along a neutral gas and produces an ionization front. we present a model for the self-induced blueshift.Chapter 4 Photon kinetic theory Until now. The electromagnetic wave spectrum will then be seen as a gas of photons evolving in an optical medium. we have considered single particle trajectories. because such a structure is simply forgotten. A second and more elaborated method will also be described. In particular. The easiest way to obtain a more detailed description of the wavepackets is to describe them. we will use simple and intuitive arguments which are similar to those leading to the Klimontovich equation for charged particles in a gas. These kinetic equations sometimes contain too much information for several problems. but as an ensemble of photons with a given spectral and space–time distribution. First. Examples of time evolution of short pulses with chirp and frequency shift are given. where the analogue of the Wigner function for the electromagnetic field is introduced. However. which (as we have shown) can account for several new and interesting features of electromagnetic wavepackets travelling in a non-stationary medium. for which a simpler description of the photon distributions is required. We conclude the chapter by considering the well-known effect of self-phase modulation and by showing that it can be described as a particular example of photon acceleration. and to consider only averaged photon properties. The kinetic equation describing the space–time evolution of such a gas is derived here. this approach is not capable of describing the change in the internal structure of the wavepackets themselves. This chapter will be completed by a detailed discussion of the phase-space representation of a short electromagnetic pulse.

k). k.68 Photon kinetic theory 4. we can take the partial time derivative of equation (4.2) Integration in phase space clearly gives the total number of photons (or ray trajectories) n= dr dk N (r .5) Another simplification can be introduced by using the following property of the Dirac δ function: xδ(x − y) = yδ(x − y). k.4) Now.1 Klimontovich equation for photons Let us first consider a single photon trajectory. t). t) = 0 (4. as defined by the ray equations in their canonical form. t) = ∂t n j=1 dr j ∂ dk j ∂ · + · dt ∂ r dt ∂ k δ[r − r j (t)]δ[k − k j (t)]. k. Of course. (4. a large number of nearby photon trajectories.2). as determined by the solutions of the ray equations. We have ∂ N (r .3) If we want to establish an equation for the time evolution of the photon density distribution. instead of a single photon trajectory (which gives no information on the detailed structure of the pulse). (4. and the associated density distribution n N (r . t) = δ[r − r (t)]δ[k − k(t)] (4. we can use the obvious relation ∂/∂ x f (x − y) = −∂/∂ y f (x − y). (4. k. The dynamical state of the photon (position and local frequency) can be determined by its location in the six-dimensional phase space (r . (4. k.1) where r (t) and k(t) represent the actual photon trajectory. in order to accurately describe the evolution of a given electromagnetic pulse propagating in a medium. The result is d N (r . Let us then assume n different photon trajectories.6) dt . t) = j=1 δ[r − r j (t)]δ[k − k j (t)]. This allows us to replace the remaining r j and k j by r and k inside the differential operator acting on the δ functions. We can also associate with this single photon a microscopic density distribution defined by the product of two Dirac δ functions N1 (r . it is more convenient to consider. and transform this expression into ∂ N (r . k. t) = ∂t n j=1 dr j dk j ∂ ∂ · · + dt ∂ r j dt ∂ k j δ[r − r j (t)]δ[k − k j (t)].

as we ¯ know from quantum theory. On the other hand. ˆ 2 ω ω (4. The problem with this simple result is that it was obtained by assuming that the photons are point particles.12) where |E k |2 is the module square of the electric field amplitude of the Fourier component k. (2π )3 (4. (4. k) = (ω. From the canonical ray equations. This would lead us too far from our present purpose. we see that this kinetic equation can also be written as ∂ Nk (4. The factor of 2 is introduced because of the existence of two possible states of polarization. t) ¯ dk .Klimontovich equation for photons with 69 d ∂ dr ∂ dk ∂ = + · + · . we can exactly establish.13) .7) dt ∂t dt ∂ r dt ∂ k Equation (4. k. It simply states that the photon number density is conserved. t) . which would replace the quite spiky quantity Nk (r .10) This equation states that the energy of each photon is equal to h ωk . (4. k) = 0 is the dispersion relation of the medium: R(ω. It is more interesting here to establish a link between this quantity and the electromagnetic energy density.8) + [Nk . By definition. t) by a smooth and well-behaved function like its ensemble average Nk (r . and R ≡ R(ω. that w(r . ω] = W (t) = where the energy density is w(r . which is obviously not the case.11) w(r . We will see below that this equation is only approximately valid. t) = 0 4 ∂ω R ∂ω |E k |2 k dk (2π )3 (4. ω] = 0 ∂t where Nk ≡ Nk (r . t) dr (4.6) can be called a kinetic equation for photons in its Klimontovich form [49].9) − · ∂r ∂k ∂ k ∂r At this point we could follow the usual statistical procedure and introduce some coarse-graining in the photon phase space. t). t) ≡ N (r . k) − k 2 c2 c2 + |k · e|2 2 = 0. we can write the total energy as [Nk . and the Poisson bracket is ∂ Nk ∂ω ∂ Nk ∂ω · . and its validity has to be confirmed a posteriori by using Maxwell’s equations. from the classical theory of radiation [57]. t) = 2 h ωk Nk (r . as should be expected. and that its range of validity is nearly (but not exactly) coincident with the range of validity of geometric optics.

for simplicity. that the fields are transverse (∇ · E = 0). we can derive from the above wave equation an equation describing the space–time evolution of the Wigner function. 116]. In the absence of charge and current distributions.2 Wigner–Moyal equation for electromagnetic radiation 4. A more refined way of establishing the definition of Nk (r . we can write 1 ∂2 E 1 ∂2 ∇ 2 E − 2 2 = 2 2 (χ E).17) This quantity is formally quite similar to the Wigner function for a quantum system [40]. In contrast with our classical approach. Following a procedure explained in detail in appendix A. In our case of a non-dispersive medium. this definition of the photon number density can only make sense if we assume that the electric field amplitude of each Fourier component is a slowly varying function of space and time. we obtain Nk (r . in order to clearly state our procedure. where χ is the susceptibility of the medium. By comparing these two expressions ˆ for the electromagnetic energy density. t − e−ik·s+iωτ . which will be considered next. we can define the Wigner function for the electric field as s τ · E∗ r − . ω.15) ∂t where D = 0 E is the displacement vector.18) s τ dτ E r + . The same procedure was used in reference [111] to study the case of relativistic plasmas. t + 2 2 . t) = 0 8h ¯ ∂R ∂ω |E k |2 .2. k) takes the form F(r . t) is based on the concept of the Wigner functions for the electromagnetic field.70 Photon kinetic theory Here e is the unit polarization vector. k) = ds ∂ c2 k + ·∇ F + ∂t ω ∂ ∂t F = −ω( sin F) (4. the evolution equation for F(r . the quantum Wigner function is well understood and of current use in quantum optics [58.1 Non-dispersive medium We will consider first a non-dispersive medium. 4. Assuming.14) Obviously. ω. ∂2 ∇ 2 E − ∇(∇ · E) − µ0 2 D = 0 (4.16) c ∂t c ∂t For a wave with a given frequency ω and wavenumber k. we have. We also have = 1 + χ . We will also assume that the medium is isotropic and with no losses. k (4. t. from Maxwell’s equations. 2 2 (4. (4. t.

22) In this case of a linear wave spectrum. (4.21) The corresponding group velocity is vk = ∂ωk c k c2 =√ = k.18) is. k) = Fk (r . The sine differential operator in equation (4. it is useful to introduce a few simplifying assumptions.20) At the cost of such an unusual operator. Equation (4. k ωk ∂k (4.18) is formally quite similar to the Wigner–Moyal equation for quantum systems [40. We can assume that this spectrum is just a superposition of linear waves. in each of the two terms. the value of the frequency ω has to satisfy the linear dispersion relation of the medium √ ω = ωk = kc/ . The first one is associated with the character of the electromagnetic spectrum. 80]. (4. we were able to derive from Maxwell’s equations a closed evolution equation for the Wigner function F of the electric field. the first differential operator acts backwards on and the second one acts forwards on F. in fact. In order to compare these two approaches.24) . the Wigner function F takes the form F ≡ F(r . according to sin = ∞ l=0 (−1)l (2l + 1)! 2l+1 . an infinite series of differential operators. t)δ(ω − ωk ). Its relation to the geometric optics approximation will become apparent below. which has no equivalent in the quantum mechanical problem. m ∂ω ∂ω ∂ωm (4.Wigner–Moyal equation for electromagnetic radiation where is a differential operator. ω. It can be defined by = 1← ∂ ∂ ∂ ∂ · − 2 ∂r ∂ k ∂t ∂ω → 71 and forwards . (4. it is significantly more complex than the kinetic equation established at the begining of this chapter. For that reason it can be called the Wigner–Moyal equation for the electromagnetic field. except for the term on the time derivative of the refractive index. Clearly. (4. t.26). and noting that the reduced Wigner function Fk is independent of ω and consequently that ∂m F ∂m ∂ m Fk = Fk m δ(ω − ωk ) = (−1)m δ(ω − ωk ) = 0.23) Replacing it in equation (4.19) The right and left arrows are here to remind us that. This is valid in quite general conditions. which acts both backwards on on F. For each spectral component. apart from our basic assumptions that the medium should be non-dispersive and that the dielectric constant should only evolve on a slow timescale.

namely dt = dr dk = . vk (ωk /2 )(∂ /∂ r ) (4.27) This is valid for a slowly varying medium. Fk ≡ + vk · ∇ + · dt ∂t dt ∂ k This equation. In such a case. which can then be written as d ∂ dk ∂ (4. it is useful to simply retain the first term in the development (4. as well as the higher order derivatives associated with the diffraction terms l > 0 in the development of the sine .28) On the other hand. where the gradients contained in the operator k are very small. (4. and the Wigner–Moyal equation reduces to ∂ + vk · ∇ Fk + ∂t ∂ ln ∂t Fk − ωk 2 ∂ Fk ∂ · ∂r ∂ k . we are close to the conditions where the geometric optics approximation is valid. is valid when the logarithmic time derivative. · 2 ∂r ∂ k (4.25) is a reduced differential operator defined by k = 1← ∂ ∂ → . They are nothing but the characteristic equations of the simplified version of the Wigner–Moyal equation.26) Because. if we neglect the logarithmic derivative in this equation. in the Wigner–Moyal equation.31) We recover here the ray equations of the geometric optics approximation. dt 2 ∂r ∂r ∂r 2 (4.20): sin k k. the sine operators are too complicated to be calculated in specific problems. we notice that this equation implies that a triple equality exists. (4. which states the conservation of the Wigner function Fk . (4. identical to those used before.32) Fk 0.30) (4.72 Photon kinetic theory we can write the Wigner–Moyal equation in a simplified form ∂ ∂ ln ωk + vk · ∇ Fk + Fk = − [ sin ∂t ∂t Here. k k Fk ] .29) This is equivalent to stating that dr ∂ωk = vk = dt ∂k dk ωk ∂ kc ∂ ∂ωk = = 3/2 =− .

The expression for the number of photons (4.28) and rewrite it as d ∂ ln Fk = − dt ∂t Fk (4. (4. Furthermore. and if we notice that dωk ∂ωk ωk = =− dt ∂t 2 we can then obtain d Nk = dt 1 ∂ + vk · ∇ ln 2 ∂t Nk . in the form dNk ≡ dt dk ∂ ∂ + vk · ∇ + · ∂t dt ∂ k Nk = 0. k) = − c2 k 2 /ω2 = 0.32).37) where the total derivative is determined by equation (4. t) · E ∗ (r − s/2. t) = 0 8h ¯ ∂R ∂ω ωk Fk (r .34) is then reduced to Nk (r .40) .36).33) It is now useful to define the number of photons Nk (r . t). t) = 0 (4. can be neglected. (4. as shown in appendix A: Fk (r . (4.36) We can now return to the somewhat more exact expression for the Wigner– Moyal equation (4. t) = E(r + s/2. we can finally state an equation of conservation for the number of photons. For the case considered here of a non-dispersive medium it reduces to R ≡ R(ω. t) (4. t) as the space Wigner function for the electric field. as Nk (r . (4.38) Neglecting the slow variations of the refractive index appearing on the righthand side. t)e−ik·s ds. if we take the total time derivative of the number of photons (4.35) 4h ωk ¯ Fk (r .34) where R = 0 is the dispersion relation of the medium. On the other hand. from equation (4. (4.39) ∂ ln ∂t . t) in terms of this reduced Wigner function.Wigner–Moyal equation for electromagnetic radiation 73 operator sin k .17) we can define Fk (r .

16) has to be replaced by ∇2 − 1 ∂2 c2 ∂t 2 E = µ0 ∂2 P ∂t 2 (4. This new derivation. This means that these terms represent the diffraction corrections to the geometric optics approximation. t ) E(r − r . r . an evolution equation of this quantity seems to have been ignored. t − t ). which can easily be included afterwards. which is much more complicated. On the other hand. t. we still neglect the losses in the medium. to the lowest order of the space and time variations of the medium. its range of validity is not very wide because we have neglected dispersion effects. 4. However. we understand from this that the conservation equation for the number of photons is only valid when the higher order terms contained in the sine operator of the Wigner–Moyal equation can be neglected. In contrast. The derivation is detailed in appendix B and.41) where P = 0 E − D is the vector polarization of the medium. as measured in optical experiments [45].2. t) = 0 dr dt χ (r . First of all. we note that the classical Wigner function for the electromagnetic field introduced in this section is sometimes used to characterize ultra-short laser pulses with a time-dependent spectrum. the result is ∂ 2 + vg · ∇ F = − (η F) ∂t 2ω + ∂η0 /∂ω (4. our starting equation (4. For simplicity. In general terms it can be related to the electric field E by the integral P(r .2 Dispersive medium The above derivation is conceptually quite interesting because it establishes a clear link between the exact Maxwell’s equations and the heuristically derived Klimontovich equation. The Wigner–Moyal equation described here can eventually be used to understand the space–time evolution of such short pulses along a given optical circuit. (4.42) Again. has however the advantage of using a more general definition for Nk .19) and vg is the group velocity defined by 2c2 k − ω2 ∂ /∂ k vg = (4.43) where is the differential operator defined by equation (4. The generalization to the case of a dispersive medium is considered in this section.44) 2ω + ω2 ∂ /∂ω . Finally.74 Photon kinetic theory This is identical to the Klimontovich equation derived at the begining of this chapter. we can derive from here an evolution equation for the double Wigner function for the electric field. if the electromagnetic radiation propagates in a dispersive medium.

We have then 2 ∂ ∂ 1 ∂ωp ∂ − + vk · · ∂t 2ωk ∂ r ∂ k ∂k Fk = 0 (4. equation (4. we can then justify the use of the equation of conservation for the number of photons Nk (r .3 Photon distributions In order to illustrate the interest of this kinetic approach. to the gradient of the square of the plasma frequency. t). Let us assume that the electromagnetic wave spectrum is made of a superposition of linear waves. This is equivalent to stating that the reduced Wigner function Fk is conserved: d ∂ ∂ dk ∂ (4.48) From these reduced forms of the Wigner–Moyal equation for a dispersive medium. for a non-dispersive medium. this would reduce to the result of the previous section. t). Obviously. which can also be called the kinetic equation for photons propagating in slowly varying dispersive media. let us give some examples and introduce some definitions.47) + Fk = 0 Fk ≡ + vk · · dt ∂t dt ∂ k ∂k because we know.43) becomes ∂ ∂ηk ∂ 1 ∂ + + vk · · 2 /∂ω) ∂t (∂ω ∂k ∂k ωk ∂ r Fk = 0 (4. The practical interest of this kinetic approach will now be illustrated with a few specific examples. Then. the gradient of ηk appearing in the last term of this equation reduces to the gradient of the electron plasma density. such that ∂η0 /∂ω = 0. that 2 dk ∂ωk 1 ∂ωp =− =− . characteristic of the Wigner–Moyal equation. As an example of a dispersive medium. equation (4. In this case. if we had used the full developement of η± around (r . k. F = Fk δ(ω − ωk ).45) 2 where vk = (∂ω/∂ k)ωk and ηk = ωk χ (r . dt ∂r 2ωk ∂ r (4. or equivalently. we would have obtained the operator sin instead of just . such that we can use equation (4. t). we can consider an isotropic plasma. it should be noticed that we have .40). instead of the first terms. 2 where we have ηk = −ωp . t).46) 2 where ωk = k 2 c2 + ωp (r . First of all. from the photon ray equations. This means that. This equation therefore generalizes the above derivation of this equation to the case of a dispersive medium.Photon distributions with 75 = 1 + χ = 1 + η/ω2 . 4.23).

52) for an arbitrary function N (k).50) The simplest possible case corresponds to a photon beam propagation in a uniform and non-dispersive medium. short laser pulses. t) = 0. we also have vk = v0 = c/n = const. we will discuss one-dimensional propagation. if we take the simple case of plane waves. For simplicity. it can be applied to arbitrary forms of wavefields (plane.54) . Taking its space and time derivatives.3. such that E(r .49) This is just the definition commonly found in the literature [93.34) a more general definition for the number of photons than those usually found in the literature.76 Photon kinetic theory introduced in equation (4. we can easily see that this solution satisfies equation (4. t) = ∂R |E 0 |2 δ(k − k0 ).50) is reduced to ∂ ∂ + v0 ∂t ∂x Nk (x. and N (k) describes the spectral content. spherical or cylindrical waves). Furthermore. with a spectral width σk 2 N (k) = N0 exp −(k − k0 )2 /σk .51) where the refractive index n is time independent. (4. In particular. 8h ∂ω ¯ 0 (4. (4. t) = 0. due to uniformity: fk ≡ dk ∂ω kc ∂n =− = 2 =0 dt ∂x n ∂x (4. with a duration τ . (4. which propagates along the medium without changing its shape. 114] which is not very useful to describe. t) = E 0 exp(ik0 · r − iω0 t). in a non-dispersive medium. This can be represented by 2 Nk (x. A useful choice is that of a spectral Gaussian distribution.52) A possible solution of this equation is a Gaussian pulse. for instance. As it states. The pertinent kinetic equation will be ∂ ∂ ∂ + vk + fk ∂t ∂x ∂k Nk (x. our definition reduces to Nk (r . This means that the one-dimensional kinetic equation (4. t) = N (k) exp −(x − v0 t)2 /σx (4.53) where σx = v0 τ is the spatial pulse width.1 Uniform and non-dispersive medium Let us now consider some examples of solutions of the photon kinetic equation. 4. The third term in this equation will then be equal to zero. centred around some wavenumber value k0 .

We see that. t) = N (k) exp −(x − vk t)2 /σx (4.59) + vk Nk (x.58) and is independent of x and t. 2 c b2 + 4kcn 0 (4. In order to calculate its value.2 Uniform and dispersive medium As a first step in complexity. ∂t ∂x Instead of the Gaussian distribution (4.61) .3. we can try here a solution of the form 2 Nk (x. 4. t) = 0.50) will still be equal to zero. The dispersion relation can then be written as kc = n 0 ω + (ω − ω0 )ωn 0 (4. by linearizing vk around k0 . where the refractive index is frequency dependent. The third term in the photon kinetic equation (4.53).59). we consider a uniform but dispersive medium.56) with n 0 = (dn/dω)ω0 . vk v0 + (k − k0 )v0 . the higher frequencies inside the pulse (with higher values of k) will travel with lower velocities and will be retarded along propagation vk < v0 . for n 0 > 0. In such a medium. This can be solved for ω to give ω= 1 n0 ω0 − 2 n0 + 1 (ω0 − n 0 /n 0 )2 + 4kc/n 0 . with v0 = ∂vk ∂k =− k0 (b2 2c2 n 0 . let us linearize the refractive index around the central pulse frequency n(ω) n 0 + (ω − ω0 ) dn dω (4.55) ω0 where ω0 = ω(k0 ) and n 0 = n(ω0 ). + 4k0 cn 0 )3/2 (4. This is clearly a solution of equation (4.Photon distributions 77 Compatibility between the spectral and the spatial distributions implies that we always have σk ≥ 1/σx . Let us see the meaning of this new solution and compare it with (4. the pertinent equation to be solved is ∂ ∂ (4.57) The group velocity becomes vk = ∂ω = ∂k (4. which minimizes the position–momentum uncertainty relations. The equality σk σx = 1 corresponds to the case of a transform limited pulse.53).60) wher vk is determined by equation (4. but the group velocity vk will not be a constant. for k > k0 .58) with b = (ω0 n 0 − n 0 ).

and time-dependent function: 2 Nk (x. In the frame of our kinetic description of a photon beam. As an example.53). we obtain a photon distribution formally analogous to (4. and using N (k) as given by equation (4.64) is the photon mean density. let us consider the distribution (4. t) = N (k. σk (t)2 b= v k0 + 0 (x − v0 t)t. t) k Nk (x. when an initially Gaussian beam propagates in a uniform but dispersive medium it maintains its spatial Gaussian shape but its spectral distribution is distorted in time. σk (t)2 σx σk 4. x.54). t) dk 2π (4.3. but with N (k) replaced by a space.67) .66) 1 n γ (x.63) Here.65) (4.60).62) with N (k. because it is associated with such a pulse distortion. 2 2 σk (t) σx (4. Replacing it in equation (4. At this point. the chirp is determined by the space–time distribution of the averaged wavenumber: k = where the normalization factor nγ = Nk (x.68) ∞ 2 −∞ k dk exp(−ak + 2bk) ∞ 2 −∞ dk exp(−ak + 2bk) (4. we obtain k = where we have used a= 1 .78 Photon kinetic theory Replacing this development of vk in equation (4. we have used v t2 1 1 = 2 + 02 . x. the concept of pulse chirp has to be introduced.65). t) exp −(x − v0 t)2 /σx (4.62–4. We can say that a given pulse (or photon beam) is chirped if its averaged wavenumber is not constant across the pulse. t) = N0 exp −(k − k0 )2 /σk (t)2 2 × exp 2t (k − k0 )v0 (x − v0 t)/σx . The concept of pulse chirp is very important for the optics of short laser pulses.3 Pulse chirp The above result shows that. t) dk 2π (4.64). (4.

74) we obtain k2 = 1 2a 1+ = σk (t)2 + k 2.75) This is equivalent to writing ( k)2 ≡ (k − k )2 = σk (t)2 . 2 (4. we can calculate the spectral width of the chirp.72) In our case. simply because some of the photons will travel with larger velocities than the others. 2 (4.Photon distributions Using the following solutions for these two integrals [35]: ∞ −∞ 79 exp(−ak 2 + 2bk) dk = b2 π exp a a b a b2 π exp a a (4. as given by equation (4.71) This shows that the mean value of the wavenumber inside the photon beam is time dependent. (4.70) we obtain k = b σk (t)2 (x − v0 t)t. it can be written as k 2 = ∞ 2 2 −∞ k dk exp(−ak + 2bk) . t) k 2 n k (x. while the pulse chirp associated with the mean value k increases in time. at t = 0. which means that propagation of a Gaussian pulse in a dispersive medium leads to pulse chirping. t) dk .64). ∞ 2 −∞ dk exp(−ak + 2bk) (4.73) Using the following solution for the integral in the numerator [35]: ∞ −∞ k 2 exp(−ak 2 + 2bk) dk = 1 2a 2b2 a π a 1+ 2b2 a exp b2 a . In a similar way. this means that the square mean deviation decreases with time. we obtain a thin and elongated spectral distribution as time evolves (or as the pulse propagates). 2π (4.76) According to the definition of σk (t). Starting from a Gaussian spectral distribution. such that k is constant across the pulse spatial width. . by defining a new mean value k2 = 1 n γ (x. = k 0 + v0 2 a σx (4.69) and ∞ −∞ k exp(−ak 2 + 2bk) dk = . (4.

at a given time t.80 Photon kinetic theory (a) Nk x k (b) Nk x k Figure 4.t)] 2 /σ 2 k .78) Using this in equation (4. (4.77) with N (k.1. t) exp −(x − vk t)2 /σx (4.59) we can easily realize that this is a solution of the photon kinetic equation. for a uniform but dispersive medium. x. (b) a chirped pulse. x. t) in phase space (x. for (a) a transform limited Gaussian pulse. Representation of the photon distribution Nk (x. t) = N0 e−[k−g(x. A more general class of solutions for the photon kinetic equation for uniform dispersive media can be stated as 2 Nk (x. k). t) = N (k. for every function .

2 ∂x σx (k − g) (4.3. The photon kinetic equation becomes ∂ ∂ ∂f ∂ + vg (x.79) As an example of a solution. t)] (4. such that the refractive index is determined by n ≡ n(x.85) For a given perturbation of the refractive index of the medium f (x. Let us try a solution of the form (4. but with vk replaced by the constant v0 . t)] where f (x. t)] kc [1 − δ f (x. t) = k0 + g0 (x − vk t)n where k0 and g0 are constants and n is an integer. such solutions. as vk ≡ vg (x. t) has to satisfy the following equation: ∂ ∂ + vg ∂t ∂x g = δv0 k 2 ∂f 2 σ (x − v0 t) + δ f v0 k . such that x v0 t.77. t) describes a small space–time perturbation and δ of the perturbation. t) = 0 (4. t) + kv0 δ ∂t ∂x ∂ x ∂k Nk (x. and the last .80) (4. even if they exist. are in general very difficult to find.78). In order to obtain an approximate solution for g(x. (4.4 Non-stationary medium Let us now assume a non-dispersive but non-stationary and non-uniform medium. Using this in equation (4.81) 1 is the scale (4. t). t)] n0 (4.83) where we have used v0 = c/n 0 .84) we notice that the function g(x. The dispersion relation can then be written as ω= kc n 0 [1 + δ f (x. t). However.82) and the photon. we can consider the family of functions g(x. this equation allows us to obtain the solution g(x.Photon distributions satisfying the equation ∂ ∂ + vk ∂t ∂x 81 g = 0. t) = n 0 [1 + δ f (x. or group velocity. t) v0 [1 − δ f (x. 4.84) where we have used v0 = c/n 0 . we can assume that we are interested in the main region of the pulse. 4. t).

for this and other situations.90) . t). t) is reduced to ∂ ∂ ∂f + v0 g δv0 k . t) = cos kp (x − ut). (4. t) will then be determined by the balance equation dn 1 (r . If we also take vg g(x. u − v0 (4. Experimentally. the photoionization processes are described by the tunnelling ionization theory. this has been well known since the early sixties [120. t)n 0 (r .5 Self-blueshift An interesting result [91] was obtained from the numerical integration of the photon kinetic equation. such that f (x.82 Photon kinetic theory v0 . Let us describe the ionization model. t)n 1 (r . as already found in the analysis of single photon trajectories. 122]. numerical solutions of the photon kinetic equation are required. t) = k0 + g0 cos kp (x − ut) (4. these analytical results are quite rough and. ∂t ∂x ∂x An important example is that of a wakefield perturbation of the refractive index. this number will decrease due to field ionization as dn 0 (r . 48]. For very intense incident laser fields. But. Self-blueshift is a particular aspect of photon acceleration where the space– time changes in the refractive index are due to the incident laser pulse itself. The number of singly ionized atoms n 1 (r . for the self-blueshift of a laser pulse penetrating in a neutral gas. and it was studied theoretically by several authors [41. If the initial density of the neutral atoms of the gas is n 0 (r .3. of course. t)n 0 (r .87) with g0 = −δ kv0 .89) where w1 is the ionization rate for the atom. t) = −w1 (r . t) is g(x. u ∼ v0 . In this case.86) term in this equation can be neglected. the equation for (4. t) dt (4.88) This shows that the spectral shifts are more important when the phase velocity of the perturbation is nearly equal to the group velocity of the photons. 4. This blueshift is due to the sudden ionization of the gas and subsequent creation of an ionization front. the approximate solution for g(x. t) − w2 (r . t) = w1 (r . t) dt where w2 is the probability of double ionization occurring.

t) = 2 Nk (r .2). Furthermore. Furthermore.and time-varying media. it nevertheless contains detailed information about the photon spectrum. The electron plasma density will then be determined by j=Z n e (r . in the same way as fluid equations can be derived from kinetic equations for the particles of an ordinary gas. t) = j=0 jn j (r .92) where Z is the atomic number of the atoms in the gas. the blueshift is not uniform and several wavelike perturbations can be observed in the up-shifted spectrum. t) = w j (r . The various ionization probabilities w j depend on the local amplitude of the laser electric field and are well known from the standard theory of field ionization [4. coupled with these simple balance equations. t) − w j+1 (r .Photon fluid equations by dn j (r . dt Finally. t). the density of completely ionized atoms is given by dn Z (r . t) dk . 47]. Let us first define the mean density of the photon gas as the integral of the photon distribution over the entire spectrum: n γ (r . 4. The values of these ionization probabilities can establish the coupling between these balance equations for the atomic states and the above kinetic equation for the photons. the local plasma dispersion relation can be consistently determined by considering charge neutrality. t)n j−1 (r . clearly shows that the front of a Gaussian beam penetrating in a neutral gas region will be significantly blueshifted (see figure 4. in the limit of slowly space. (2π )3 (4. t) = wz (r . t) dt 83 In more general terms. t)n j (r . (4.94) .93) A numerical integration of the kinetic equation for photons. one for each ionization state. t). Even if the range of validity of the photon kinetic theory is limited.4 Photon fluid equations In this chapter we have been able to show that the photon kinetic equation can be derived from Maxwell’s equations.91) (4. These spectral undulations will be eventually attenuated by electron impact ionization which was not contained in this model. It is then useful to derive from this equation a set of conservation laws describing the space– time behaviour of photon averaged quantities. the number density of j-charged ions is determined (4. t)n Z (r .

propagating in neutral argon. Integration of the first term gives ∂ ∂t Nk dk 1 ∂n γ = .95) The factor of two appearing in these two definitions is due to the existence of the two independent polarization states. (2π )3 (4. (2π )3 (4.98) The sum of these three contributions leads to the continuity equation for the photon gas ∂n γ (4. or mean group velocity.99) + ∇ · (un γ ) = 0. integration of the third term gives f · ∂ Nk dk = ∂ k (2π)3 N k ∇ · vk dk . t) dk . we can define the mean photon velocity. (2π )3 (4. 3 2 ∂t (2π ) (4.84 Photon kinetic theory Figure 4. The continuity equation for the photon gas. with an initial Gaussian spectrum. t) = 2 nγ vk Nk (r . ∂t . Similarly. Numerical simulation of the self-blueshift of a laser beam with an intensity of 1015 W cm−2 .97) Finally. or photon density conservation equation.96) Integration of the second term leads to vk · ∇ Nk dk = ∇ · (un γ ) − (2π)3 Nk (∇ · vk ) dk . can be derived by integrating the kinetic equation in k.2. as u(r . The initial density of the neutral atoms is 1018 cm−3 for x > 0.

we can finally establish the momentum conservation law of the photon gas. (4. Multiplying the photon kinetic equation by vk and integrating it over the entire spectrum.102) Here. in the form ∇ Pγ ∂u + u · ∇u = − + Fγ (4.101) where we have used the following quantity: vk vk = 2 nγ Nk vk vk dk . such that Pγ = n γ Teff . we can introduce the concept of photon pressure. we get the following contribution from the first term: vk ∂ Nk dk 1 ∂ (un γ ) − = 3 ∂t (2π) 2 ∂t Nk ∂ v k dk .100) The contribution of the second term can be written as vk vk · ∇ Nk dk 1 = ∇ · n γ vk v k 3 2 (2π) − Nk ∇ · (vk vk ) dk (2π )3 (4.Photon fluid equations 85 The second conservation equation to be derived is the photon momentum conservation law.102) and adding the diagonal terms of this tensorial relation.105) Finally. it can easily be shown that the photon pressure is determined by Pγ = 2 3 Nk (vk − u)2 dk .103) where I is the 3 × 3 identity matrix. ∂t (2π )3 (4. Pγ . (2π )3 (4. ∂t (2π )3 ∂k (4.104) We can also define an effective temperature for the photon gas.108) .106) Adding the contributions from the three terms and using the continuity equation. assuming that it can be considered an ideal gas. the contribution from the third term of the photon kinetic equation to the momentum conservation law is determined by vk f · ∂ Nk ∂k dk . (2π )3 (4. such that vk vk = u u + Pγ I nγ (4. Using equation (4. (2π )3 (4.107) ∂t nγ where the mean force acting on the photons is determined by Fγ = 2 nγ Nk ∂ vk ∂ dk · ( f vk ) + ∇ · (vk vk ) + .

ω (4.112) If the distribution is even with respect to the quantity (vk − u).114) ω2 (2π )3 In a similar way.110) We see that this mean force acting on the photon gas is due to the space and time variations of the refractive index. n vk = ck . nk (4. In this case. let us calculate this force for the case of a non-dispersive dielectric medium. with refractive index n ≡ n(r . we have ω= kc .117) . (4. (2π )3 (4. As a second example. (2π )3 (4. (2π )3 (4. The contribution ¯ of the first term is U= hω ¯ ∂ Nk dk 1 ∂ = Wγ − h ¯ 3 ∂t (2π) 2 ∂t Nk ∂ω dk ∂t (2π )3 (4.109) Using these in the above expression for Fγ .111) The mean force now becomes Fγ = − 1 nγ Nk ω2 2 ∇ωp + vk 2 ∂ωp ∂t dk . by multiplying the kinetic equation by h ω and integrating it over k.86 Photon kinetic theory As a first example. (2π )3 (4.115) where the mean photon energy is determined by Wγ = 2 nγ h ωNk ¯ dk . we can also derive the energy conservation equation. vk = kc2 . we obtain Fγ = 2 c2 nγ n2 Nk − ∂n k 1 ∂n kk + · I−2 2 ∂t k n ∂r k dk .116) The contribution of the second term is h ωvk · ∇ Nk ¯ 1 dk = ∇ · (uWγ ) − h ¯ 3 2 (2π) Nk ∇ · (ωvk ) dk . such that 2 ω 2 = k 2 c 2 + ωp . let us consider an isotropic plasma.113) nγ ∂t with Nk dk . t). this can also be written as 2 ∂ωp U 2 ∇ωp + u Fγ = − (4.

an initially nearly monochromatic laser pulse can be transformed into a broadband radiation pulse.4. the nonlinear process. in turn. where the spectral width is close to its minimum value allowed by the uncertainty principle. where both the time and the space derivatives of the refractive index (or of the plasma frequency) contribute to the change in the mean value of the momentum of the photon gas. the photon kinetic theory described in . By this process. However. In more exact terms. ∂t (2π )3 (4. it can explain the following self-consistent process: the photon bunch changes the optical properties of the medium which. This was already shown in the example of the self-blueshift. The best example of such a process is however the well-known self-phase modulation of a laser pulse in an optical fibre or in other optical media. In other words. if the initial laser pulse is a nearly transform limited short pulse.121) where ωp is the electron plasma frequency. Adding these three contributions. This astonishing effect has been known experimentally since the early 1970s and the current theory is based on the calculation of the nonlinear contributions to the total time phase dependence. and.108). ∂ω 1 ∂ωp = ∂t 2ω ∂t 2 (4. the contribution of the third term is hω f · ¯ ∂ Nk dk = −h ¯ ∂ k (2π)3 Nk ∂ ∂k · (ω f ) dk (2π )3 87 (4. usually called self-phase modulation. can transform it into nearly white light. 4. 4. only the time derivatives can be a source of energy. given at the end of section 4.117). in the case of a plasma.119) In the case of a non-dispersive dielectric medium. in contrast with the momentum conservation equation (4.118) which exactly cancels the last term in equation (4. we obtain ∂ Wγ + ∇ · (uWγ ) = 2h ¯ ∂t Nk ∂ω dk .Self-phase modulation Finally. modify the spectral composition of the photon bunch. We see that.5 Self-phase modulation An important property of the photon kinetic theory is that it can describe the nonlinear changes of a photon bunch or a short laser pulse due to its own spectrum.107.120) (4. we have ∂ω ω ∂n =− ∂t n ∂t where n is the refractive index.

briefly discussed at the end of chapter 3. (4.123) Here we assume that the medium is isotropic.5. where the phase is completely ignored and where only the photon number distribution is considered. we can easily derive the following equation of propagation for the electric field E associated with the laser pulse: ∇ 2 E − ∇(∇ · E) − 1 ∂2 E ∂2 P = µ0 2 . spreading them over a considerable range of the optical spectrum. (4. It will be shown here that this new approach is equally capable of describing the spectral changes characterizing the so-called self-phase modulation processes. (4. which leads us to two important conclusions. the nonlinear properties of the optical media can accelerate and decelerate the photons contained inside the pulse. The second is that self-phase modulation (as well as the other phase modulation processes. Physically this means that. we can neglect dispersion and simply write PN L (t) = 0χ (3) |E(t)|2 E(t). otherwise the linear (or firstorder) susceptibility χ (1) would be replaced by a tensor.88 Photon kinetic theory this chapter suggests the possible use of a totally independent theoretical explanation. In order to clarify these few very important physical statements. The linear part is determined by PL (t) = ∞ 0 0 χ (1) (τ ) E(t − τ ) dτ.124) Assuming that the laser field is transverse (∇ · E = 0). thus leading to nearly white light. such as induced and crossed phase modulation) is nothing but a particular aspect of the phenomenology of photon acceleration described in this work. For the nonlinear part of the polarization vector.125) c2 ∂t 2 . starting from a nearly monochromatic laser pulse.122) The vector ploarization appearing in this equation can be divided into two distinct parts: P = PL + PN L .1 Optical theory If we start from Maxwell’s equations in a dielectric medium. 4. This is also valid for the induced phase modulation. let us first remind ourselves of the usual optical theory of self-phase modulation and then compare it to the photon kinetic approach. we can then write the propagation equation as ∇2 E − 1 ∂2 c2 ∂t 2 E+ 0 ∞ χ (1) (τ ) E(t − τ ) dτ = 1 ∂ 2 (3) 2 χ |E| E. and to make them accessible to the nonspecialist. The first one is that the phase is not an essential ingredient of self-phase modulation. in the absence of charge and current distributions. c2 ∂t 2 ∂t (4.

128) We will also assume weak linear dispersion.131) We can now use equations (4. 2 ∂z (4. . with a unit polarization vector a and a slowly varying amplitude E 0 (z. .c.132) . (4. The absence of dispersion would correspond to identifying this function with a Dirac delta function δ(τ ).c. we obtain an equation describing the slow space–time evolution of the envelope field amplitude: PN L (t) = 0 a exp(ik0 z − iω0 t) χ (ω0 ) + i ∂ 1 ∂ + ∂z v0 ∂t E 0 = iω0 α|E 0 |2 E 0 . ∂t (4. where (ω) = 1 + χ (ω) is the dielectric function of the medium.129.c.129) Using this expansion in equation (4. (4. and noting that χ (ω) = 0 ∞ χ (1) (t)eiωt dt tχ (1) (t)eiωt dt (4.123) as E(t − τ ) E(t) − τ ∂ E(t) + · · · . 2 ∂t 1 ∂ E0 2 −k0 + 2ik0 a exp(ik0 z − iω0 t) + c.Self-phase modulation Let us assume a solution of the form E(r . which means that the function χ (1) (τ ) is sharply peaked around τ = 0.c. . 2 89 (4.130) ∂ χ (ω) = i ∂ω we obtain ∞ 0 ∂χ (ω0 ) ∂ E 0 (z. Noting that the frequency ω0 and the wavenumber k0 are related by the linear dispersion relation k 2 c2 = ω2 (ω). This means that the scales of variation of this amplitude are such that 1 ∂ E0 E 0 ∂t We can then write ∂2 E ∂t 2 2 −ω0 E + ω0 . t) = 1 a E 0 (z.125). we can expand the electric field E(t − τ ) appearing in equation (4.126) This represents a laser pulse propagating along the z-axis. t) + c.127) ∇2 E 1 ∂ E0 −2iω0 a exp(ik0 z − iω0 t) + c. t). In the weak dispersion medium. 1 ∂ E0 E 0 ∂z k0 .131) in the propagation equation (4. 4.123). t) exp(ik0 z − iω0 t) + c. 2 ∂ω ∂t (4.

∂t v0 ∂η (4. t) = φ0 + ω0 α|A(η)|2 t. and positive at the rear.135) Such a solution describes a laser pulse propagating with an invariant pulse shape. for η > 0. In order to have a more precise idea of the chirp.133) and α is the nonlinear coefficient. The result is that the pulse frequency will not remain constant along the propagation and will depend on the actual position inside the pulse envelope: ω = ω0 − t ∂ ∂φ = ω0 1 + α |A(η)|2 . determined by ω2 ∂ (ω0 ) 1 k0 = + 0 v0 ω0 2c2 ∂ω (4.139) 2 /σ 2 .137).134) k0 c2 The solution of the envelope equation (4. (4. such that |A(η)|2 = A2 e−η 0 From equation (4.138) We see that the frequency shift grows linearly with time.2 Kinetic theory An alternative description of the same effect is now given in terms of photon acceleration [103]. resulting from the existence of a third-order susceptibility ω0 1 (3) α= χ .132) can be written in the form E 0 (z. this space–time dependence of the frequency (or of the wavevector) is called the pulse chirp. as well as on the form of that shape. (4. we can take the simple and illustrative example of a Gaussian laser pulse.137) As shown before. t) = A(η) eiφ(η. and of the amount of frequency shift ω = ω − ω0 introduced by the nonlinear susceptibility of the medium. as our .5.t) where we have η = z − v0 t and the nonlinear phase φ(η. (4.136) (4. we get ω = −αt ω0 2η |A(η)|2 . for η < 0. Instead of using the envelope field equation. v0 σ 2 (4.90 Photon kinetic theory Here v0 is the group velocity. 4. and that it is negative at the pulse front. we use. but with a phase which depends on time (or on the travelled distance).

t) . (4.147) we are able to determine the evolution of Nk (r . (4. 2 2 ∂r dt [n 0 + n 2 I (r .143) in (4.147) We know that the formal solution of the photon kinetic equation (4. let us solve these equations in a one-dimensional situation.146) (4. . where I (r .148) where (k0 . t). n 2 is the nonlinear one and I (r .142) = =− dt dt ∂k ∂k where. (2π )3 (4. t) = h ¯ ωk Nk (r . t). r0 (r . t). t) = N (k0 (r . t0 ) (4. t) = 0 dt where the total time derivative is d ∂ dr ∂ dk ∂ = + · + · . This means that. k. t) is the laser pulse intensity. t) dk ∂ kc n = I (r . in the expression of the frequency ω ≡ ωk (r . t). where propagation is taken along the z-axis. r0 .142).145) Using equation (4. ∂t (4.141) dt ∂t dt ∂ r dt ∂ k In order to determine this total time derivative operator we have to make use of the ray equations dr dk ∂ω ∂ω . k.140) Here.144) This is a slowly varying function as compared with the frequency timescale: ω ∂ ln I (r .Self-phase modulation 91 starting point.146. t).140) can be written as N (r . the equation describing the space–time evolution of the photon number distribution Nk (r . t). n 0 is the linear refractive index. t) (4. we get dr c = dt n 0 + n 2 I (r . In order to illustrate the procedure. we retain the nonlinear corrections to the refractive index ω= kc kc = . by solving equations (4. t0 ) are the initial conditions. k.143) (4. t)] (4. t) dk . n n 0 + n 2 I (r . 4. which can be written as d Nk (r .

k) to a new pair of variables (η. This means that I (z.158) . t) = I (η) n0 n0 n0 n0 (4. dt ∂η ∂η (4. p) = − pw I (η) where we have used w= n2 c .150) Let us also assume that the pulse propagates without significant profile deformation.151) This suggests the use of a canonical transformation from (z. t) = dt n0 n0 dk kc n 2 ∂ = I (z. p). t). n0 p = k. n 0 pulses. t) = ω(η.157) The integration of these photon equations of motion is straighforward.156) kc n2 n2 pc 1− 1− I (z. p. 4. even for very intense laser are small. n0 (4.154) c . t) − p Using the approximate expression ω we get (η. We can then reduce equations (4.153) The ray equations can now be written in the form dη ∂ = = −w I (η).92 Photon kinetic theory We will also assume that the nonlinear corrections to the refractive index n 2 I (r . which is usually the case. (4. t) I z− c t . such that η=z− c t. and the result can be written as η η0 dη = −w(t − t0 ) I (η ) (4.149) (4. t). n0 n0 (4.146. n0 (4. dt n 0 n 0 ∂z (4. dt ∂p dp ∂ ∂ I (η) =− = pw .155) (4. with a group velocity c/n 0 . p.147) to dz n2 c 1− I (z.152) The resulting Hamiltonian function is (η.

we have I (η) = I (η0 ).t = n2 p0 c 1− I (η) N (η0 . (4. 4. This means that we can write the pulse chirp in the form ω η.Self-phase modulation and p0 = p exp −w t t0 93 ∂ I (η) dt .t ∂2 I (η) = 0. (4. t) d p.163) ∂η But for a non-dispersive pulse. t0 ) exp 2w n0 n0 ∂ I (η) dt . which leads to ω η.148.151).159) we can write p d p in terms of p0 d p0 .160) With regard to the second expression.154). max (4. but it will be very useful for the calculation of the spectral changes of the laser pulse along its propagation. ∂η (4. t)ω(η. as already stated in equation (4.164) where ω 0 ≡ ω η0 . p. This is a simple but important result. Let us show this by considering the chirp of the pulse ω . In particular.166) . which will allow us to rediscover the main features of the so-called self-phase modulation. (4. From This will correspond to the stationary points ω equation (4. t).t = Np (η.161) Using equations (4. n0 n0 (4. t0 ) d p. we can determine the condition for an extremum ∂ ω ∂η η. ∂η2 (4.t = ω 0 exp 2w ∂ I (η) (t − t0 ) ∂η (4. p0 . By definition. it corresponds to an implicit integration.159) The first of these expressions shows that the initial value of the position coordinate η is independent of the momentum variable p: η0 ≡ η0 (η. we get ω η.162) Using equation (4.t = 0. p0 .164) it is obvious that ∂ ω ∂η η. this time without considering the field phase.t0 is the initial pulse chirp.t = n2 pc 1− I (η) N (η0 . the chirp is the averaged frequency of the pulse. at a given position η and at a given instant t: ω η.165) of the chirp curve.t = 2w(t − t0 ) ω η.

172) It should be noticed that this concept of maximum chirp can also be physically identified with the condition for the maximum frequency shift with respect to the initial pulse frequency. ∂η2 Let us illustrate these results by assuming a Gaussian pulse I (η) = I0 e−η For this particular pulse shape. but also quantitatively) with that obtained above for the self-phase modulation of the laser pulse in a nonlinear medium: the spectral width grows linearly with time and the maxima down-shifts and the up-shifts are equal to each other. σ (4. by 1 ηmax = ± √ σ. or equivalently. (4.164) we finally obtain the condition for the maximum chirp: ω max √ 2 2 w I0 e−1/2 (t − t0 ) . is determined by the condition ∂2 I (η) = 0.168) we then get ∂ I (η) ∂η √ ηmax (4. √ 2 2 n2 c ω ± I0 e−1/2 t (4. for small arguments of the above exponential. we have ∂2 I (η) = ∂η2 2η2 2 −1 I (η).167).173) σ n0 n0 where we have used t = t − t0 . 2 From equation (4. we conclude that the chirp maxima are located at the two points determined by 2η2 = σ 2 . σ2 σ2 (4. or for small nonlinearities. Notice that such an expansion is valid for small time intervals. = ω 0 exp ∓ σ (4.94 Photon kinetic theory From this we conclude that the point η = ηmax .170) =∓ 2 I0 e−1/2 . for which the maximum value of the chirp ω max is observed.168) Comparing this with condition (4.167) .169) 2 /σ 2 (4.171) Using equation (4. Introducing the frequency shift as ω = ω max − ω 0 . we get. This result for the pulse frequency shift is in agreement (not only qualitatively. .

in particular for the prediction of spectral asymmetry. resulting from the group (or photon) velocity dispersion. the spectral asymmetry between the Stokes and the anti-Stokes sidebands. We are then led to two different kinds of conclusion.4. can also be explained by our kinetic model. as determined by the photon kinetic theory. This is clearly seen in the more exact equation (4. In this . one of the branches of the exponential tends to zero.172) where. even for an initially symmetric laser pulse. This asymmetry is a natural consequence of the present theoretical formulation. Such a spectral feature is well confirmed by experiments and is illustrated in figure 4. which is the pulse steepening. Instead. the analytical solutions of the kinetic theory are more accurate than the analytical solutions of the optical theory. is clearly shown. Time evolution of the maximum frequency shift associated with self-phase modulation. The result is illustrated in figure 4. where the wavefield phase is completely ignored. The three main experimental features of this effect are well described by this kinetic model: the pulse chirp. In many respects.Self-phase modulation 95 <ω> 0 <ω> max τ Figure 4. the above expansion in not valid and the Stokes and anti-Stokes sidebands become quite asymmetric. Another important feature of this effect. and the steepening of the pulse shape. it cannot be derived from the optical theory of self-phase modulation discussed above. we have to solve numerically the photon kinetic equation. We conclude this section by stating that the effect usually called self-phase modulation can be accurately described with the photon kinetic equation.3. We should however notice that. The first one is that the so-called self-phase modulation can be seen as a particular example of the more general concept of photon acceleration in a non-stationary medium. where the pulse steepening. while the other tends to infinity. for long time intervals t → ∞. for long time intervals.3.151). we can no longer use the non-dispersive solution stated in equation (4. In contrast. which was one of the basic assumptions of our analytical calculation of the pulse chirp. In order to study the changes in the pulse shape.

for comparison. as described by the kinetic theory. the non-stationarity of the medium is due to the laser pulse itself (or to the photon number distribution associated with the pulse) which locally changes the refractive index. due to the nonlinear response of the medium. and more surprising. Illustration of pulse steepening associated with self-phase modulation. with different phase contents. would lead to nearly the same spectral broadening in the same medium.4. The second.96 Photon kinetic theory Figure 4. This means that the explanation of such a frequency shift or spectral change is conceptually more accurate in terms of photon kinetics than in terms of phase modulation. conclusion is that the wavefield phase is not the essential ingredient of this effect: two laser pulses. A Gaussian pulse is also shown. particular case. .

We can then say that a photon in a vacuum is a ‘bare’ particle with no electric charge. 104. induced by particular photon spectral distributions (such as photon beams). We will later see that in a non-ionized medium. a new property associated with this electric charge (or in other words. In this chapter we turn our attention to the influence of the nonlinear properties of an optical medium. The electric field E ≡ E(r . this electric charge is replaced by an electric dipole. Finally. In some sense these are the most relevant ones. Furthermore. This new. but a photon with the same frequency (or energy) travelling in a plasma has to be considered a ‘dressed’ particle with an electric charge which is not quantized and depends on the plasma density [69.1 Derivation of the equivalent charge One possible way to derive the photon equivalent charge is to consider an electromagnetic wavepacket propagating in a plasma. Once again. 5. we have only considered the linear properties of photons in a medium. for instance an optical fibre. the existence of this charge allows us to formulate a new series of problems. can then be envisaged. and we show that the photons possess an equivalent electric charge. 112].Chapter 5 Photon equivalent charge Until now. which essentially derive from their effective mass. with the radiation pressure effects) is the possibility of resonant Cherenkov interactions between photons and electron plasma waves. In a plasma. we will assume for simplicity that the plasma is unmagnetized. related to possible radiation mechanisms due to accelerated ‘dressed’ photons. leading to the concept of photon Landau damping [14]. t) 97 . this equivalent charge can be directly associated with the ponderomotive force (or the radiation pressure) of the photon gas acting on the plasma electrons. and somewhat unexpected property of a photon is due to the space– time varying polarization effects that it induces in the medium. or to their motion across the boundary between two different media. Plasma instabilities.

However we can. 5. If we restrict our discussion to purely transverse waves. Let us call E 1 this linear solution for the electric field. we have ∇ · E 1 = 0 and the wave equation (5. which means that kinetic effects are usually negligible and that we can use the electron fluid equations in order to determine the electron density n and the electron mean velocity. The linear response of the plasma is determined by the linearized version of equations (5.3) ∂t ∂v e + v · ∇v = − E +v× B .3. and in the second one. we can neglect the ion motion.1) ∂t c2 ∂t 2 For high-frequency waves.98 Photon equivalent charge associated with this wavepacket is described by the wave equation 1 ∂2 E ∂J = µ0 . the momentum conservation equation. (5.1) reduces to ∇×E =− ∇ 2 E1 − where the linear current is J1 = −en 0 v1 .4) ∂n 1 + n0∇ · v = 0 ∂t ∂ v1 e = − E1. The magnetic field B appearing in the Lorentz force term is the magnetic field of the electromagnetic wave itself and can be expressed in terms of the electric field E by using the Faraday equation ∂B .2) Phase velocities of transverse electromagnetic waves in an unmagnetized plasma are always larger than the speed of light c. (5. (5. In their simple non-relativistic version (which is valid only for moderate wave intensities) the electron fluid equations are ∂n + ∇ · nv = 0 (5. (5.4) are nonlinear and their solution is in general quite complicated.4) ∂t m In the first of these equations we recognize the continuity equation of the electron fluid. (5. linearize them with respect to the wave electric field.8) (5.9) ∂ J1 1 ∂ 2 E1 = µ0 2 ∂t 2 ∂t c (5.5) ∂t Equations (5.6) . as a first approximation.7) Here n 0 is the equilibrium (or unperturbed) electron plasma density. and the current in this equation is reduced to the electron current ∇ 2 E − ∇(∇ · E) − J = −en v. ∂t m (5.

18) .12) Once this simple linear solution is understood. we can obtain from equation (5.13) To the same order of approximation. (5. 2 99 (5. we can return to the wave equation (5.e.8. 5.1) and try to obtain a more adequate solution which takes the nonlinear terms into account. we can also get from equations (5. for purely tranverse wave solutions. We should notice that such a general expression could also be a solution for nonlinear wavepackets. ω0 (5.3. i. According to equations (5. E = E1 + E2 with | E1| | E 2 |. and assuming that equations (5.8.9). Finally.11) ω0 iω0 m We notice that. (5.5). t) = 1 E 0 (r .4). t) exp(ik0 · r − iω0 t) + c. ∂n 2 + n 0 ∇ · v2 = 0 ∂t ∂ v2 e + v1 · ∇ v1 = − ( E 2 + v1 × B1 ). 5. v 1 = (5. the electron density perturbations are equal to zero: n 1 = 0. · v1 .16) (5.1) ∂ 1 ∂2 ∇ 2 E 2 − ∇(∇ · E 2 ) − 2 2 E 2 = µ0 J2 (5. ∂t m (5.14) To the first order in the small nonlinear field E 2 .15) ∂t c ∂t where the nonlinear current is determined by J2 = −en 0 v2 . This can be done by assuming that the total electric field in this equation is divided in two parts: the main field (which corresponds to the linear solution) and a small correction due to radiation or polarization effects. B1 = k0 × E0.17) (5. such that k0 · E 0 = 0.c.9).10) where ω0 and k0 are the mean frequency and wavevector of the wavepacket and E 0 (r . with amplitudes determined by k0 e n1 = n0 E0. the amplitude of the wave magnetic field is. stay valid. the electron density and velocity perturbations associated with this solution have similar expressions. 5. according to equation (5.Derivation of the equivalent charge The solution of this equation can take the general form E 1 (r . t) is a slowly varying amplitude.

21) ∇ · E 2 = ∇ 2 |v1 |2 .20) + ∇|v1 |2 = − E 2 . τ ). m 2 ∂t 2 Using equation (5. (5. with negligible shape deformation. ∂t 2 m In doing so we are neglecting the high-frequency part of the nonlinear perturbation v2 .17.100 Photon equivalent charge Now we should make use of the vector identity ∇(a · b) = (a · ∇)b + (b · ∇)a + a × (∇ × b) + b × (∇ × a).18) can be rewritten as ∂ v2 e 1 (5. ∂x ∂ξ ∂ ∂ ∂ = − vg . t)|2 . we can easily realize that equation (5. If this wavepacket propagates along the x-axis. ∂t ∂τ ∂ξ (5.24) We can now perform a coordinate transformation from (x. This means that ∂ ∂ = . r⊥ )|2 .26) = 0 2m ωp ω0 2 ∂2 2 + ∇⊥ |E 0 (ξ. which would not be valid if we wanted to study harmonic generation.20) we get the following evolution equation for the low-frequency nonlinear perturbation of the electron plasma density: ∂ 2n2 en 0 n0 − (5. After elimination of v2 from equations (5.23) Using this in the equation for the forced oscillator (5. ∂ξ 2 . t) ≡ E 0 (ξ.19) If we identify the vector a with the first-order velocity perturbation v1 and the vector b with its complex conjugate.22) 2 2 m 2 ω0 ∂t 2 The eigenfrequency of this linear oscillator is nothing but the unperturbed electron plasma frequency ωp = (e2 n 0 / 0 m)1/2 . 5. with τ = t.25) (5. (5.22). we obtain 2 ∂2 ∂2 2 2 ∂ + ωp + vg 2 − 2vg ∂τ ∂ξ ∂τ 2 ∂ξ n2 (5. (5. t) to (ξ. r⊥ ) with ξ = x − vg t. with a group velocity vg = vg ex .11) this can also be written as the equation of a forced linear oscillator ∂ 2n2 1 e2 n 0 2 2 + ωp n 2 = ∇ |E 0 (r . we can write E 0 (r . In this equation |E 0 (r . t)|2 describes the slowly varying envelope of the electromagnetic wavepacket.

we have n 2 (η) = g(η ) sin(η − η ) η ∞ + η ∞ g(η ) cos(η − η ) dη . we can also use the so-called quasi-static approximation: ∂/∂τ 0.31) Integrating by parts. Noting that g(η ) = 0 for η → ∞. Second. (5.30) n 2 (η) = f (η ) sin(η − η ) dη . we obtain n 2 (η) = or.27) with kp = ωp /vg . more explicitly.32) Here we have used f (η) = dg/ dη.33) cos(η − η ) ∂ N (η ) dη . n 2 (η) = 2h k p ¯ 2 mω0 η ∞ η ∞ g(η ) cos(η − η ) dη (5. if we assume that. (5. (5. It is useful.34) . First we notice that the gradients in the perpendicular direction can be considered negligible in several relevant physical situations. for a ‘pancake’-like wavepacket (a very short laser pulse). and that sin(η − η ) = 0 for η = η.29) (5. we can always make ∇⊥ ∂ 2 /∂ξ 2 . ∂η (5. or for wavepackets not very much different from plane 2 waves. In particular. mω0 ∂η2 An appropriate solution of equation (5. in the frame of the wavepacket the pulse shape and the electron perturbations induced by it are not significantly changing. With these two drastic but physically acceptable approximations the above equation is reduced to a much simpler form: 2 2 ∂2 0 kp ∂ 2 + kp n 2 = |E 0 (ξ )|2 2 2m ω0 ∂ξ 2 ∂ξ 2 (5.Derivation of the equivalent charge 101 Before solving this equation we can simplify it in two ways.28) (5. to introduce a dimensionless space variable η = kp ξ and to replace the square module of the electric field by the number of photons N (ξ ) = Equation (5.27) now becomes ∂2 + 1 n 2 = f (η) ∂η2 where the force term is 2h k p ∂ 2 ¯ 2 N (η).29) is f (η) = η ∞ 4 h ω0 ¯ |E 0 (ξ )|2 . at this point.

38). Our derivation of Q ph was simplified by considering the integration of equation (5.38) The same expression for the second-order nonlinear current can also be derived if we start from the time derivative of the total second-order current J2 = J2L + J2N L : 2 ∂ J2 ∂ v2 e2 n 0 0 ωp ∇|E 0 |2 .35) which is directly related to the photon effective charge Q ph . such that the pulse duration t is shorter than the period of the electron −1 plasma oscillations t ωp . But it can also be obtained if we consider a pulse with an arbitrary duration. such that −en 2 (η) = Q ph N (η).34) in the limit of a very short pulse or wavepacket.40) We see that the second term leads to an expression for J2N L which coincides with equation (5. This means that we can take cos(η − η ) 1. we have η = (η − η ) 1.36) mω0 . and a second term which represents the wakefield or plasma oscillation left behind the pulse [69]. = −en 0 = E2 + 2 ∂t ∂t m 2m ω0 (5. This justifies the use of the concept of photon equivalent charge introduced above. This means that Q ph = −2h ¯ 2 ekp (5.37) The current associated with the N (η) particles of charge Q ph moving with velocity vg = vg ex is obviously given by J2N L = Q ph N (η)vg ex . .102 Photon equivalent charge This equation is appropriate for a second integration by parts.35) We can now define a photon equivalent charge Q ph . ∂ξ m ∂ξ (5. But for our purpose this will not be necessary because we notice that. in the limit of very short pulses. we get −vg ∂ J2 e2 n 0 2 ∂ = E 2 − vg Q ph N (ξ )ex . (5. the electron second-order perturbation will consist of two terms: the above term (5. τ ) and making the above two assumptions of quasi-static (∂/∂τ = 0) and of a ‘pancake’ pulse (∇⊥ ∂/∂ξ ) approximations. (5.39) Writing this equation in terms of the variables (ξ. In this case. and obtain n 2 (η) = 2h k p ¯ 2 mω0 N (η). (5.

37). Our result (5. Another interesting aspect of the above calculations is the negative value of the photon effective charge.37.45) . Replacing this in the wave equation (5. (5. 1 1 n 2 (η) = − Q ph N (η) + Q ph e e η ∞ N (η ) sin(η − η ) dη . 5. The second term represents the wakefield left behind it. Their equivalent charge is therefore negative. This means that the photons tend to repel the electrons. This justifies the statement made in chapter 3 about the possibility of generating relativistic electron plasma waves by means of a short laser pulse propagating in a plasma. Using a time Fourier transformation ∇ 2 E 2 − ∇(∇ · E 2 ) − E 2 (r .44) ω2 (ω) E ω = −iωµ0 c2 ∞ −∞ vg Q ph N (ξ ) eiωt dt.36) and represents the equivalent charge of the total photon distribution. t) = we obtain ∇ 2 Eω + ∞ −∞ E ω (r ) e−iωt dω 2π (5.21) tends to push the plasma electrons out of the region occupied by the electromagnetic wavepacket. (5. let us return to equation (5. and which follows the pulse as a wake. ∂t c2 ∂t 2 c This equation determines the secondary field radiated by the localized wavepacket (or the primary photon bunch) moving across the medium. such that ∇ · E 2 = 0. We are led to n e (η) = 2h k p ¯ 2 mω0 N (η) − η ∞ N (η ) sin(η − η ) dη (5. 5.34) and integrate it by parts. using equation (5. It is nothing but an electron plasma oscillation moving with a phase velocity equal to the photon velocity vg .42) The first term was already given by equation (5.38) shows that such a current is due to a flow of a number N (ξ ) of photons with charge Q ph moving with velocity vg .2 Photon ondulator We have just seen how the nonlinear current associated with an electromagnetic wavepacket propagating in a plasma can be derived.41) or.Photon ondulator 103 To illustrate the existence of these two components.15) we obtain 2 ωp ∂ 1 ∂2 (5. The sign of Q ph can be physically well understood if we notice that the ponderomotive force appearing on the right-hand side of equation (5. An example of such a radiation effect is discussed here.43) E 2 = 2 + µ0 Q ph N (ξ )vg . Let us restrict our discussion to purely transverse fields.

48) c2 The solution of equation (5. ω0 m kcvg0 n 0 (5. n0 (5. (5. due to the factor (eω · ex ) we can only observe a radiated transverse field if k⊥ is not equal to zero. Equation (5.54) .47) ω2 2 (ω) − k⊥ .52) ˜ ω e ωp0 n . Using this expression in equation (5.51) The modulations in the photon group velocity also exist.47) for radiation emitted in the backward direction (or in the direction of negative values of x) is given by Aω (x) = −µ0 ω (eω · ex ) e−ikx 2kc ∞ −∞ x dt x0 dxvg Q ph N (ξ ) eikx+iωt . we can use E ω (r ) = Aω (x) eik⊥ ·r⊥ eω (5. Neglecting the transverse structure of N (ξ ).37) we can get from equation (5.45) now reduces to ∂ 2 Aω + k 2 Aω = −iωµ0 (eω · ex ) ∂x2 where k2 = ∞ −∞ vg Q ph N (ξ ) eiωt dt (5.104 Photon equivalent charge 2 Here. (5.53) The asymptotic field radiated very far away from the emitting region can then be written as Aω (x → −∞) = A e−ikx ˜ × cos[k(ξ + vg0 t)] e ∞ ∞ −∞ −∞ ik(ξ +vg0 t)+iωt dt dξ N (ξ ) (5.46) where eω is the unit polarization vector and Aω is the spectral field amplitude. ˜ This leads to 2 kp = 2 ωp 2 vg 2 ωp0 2 vg0 (5. Let us assume the case of radiation due to a modulation in the electron mean density ˜ n 0 (x) = n 0 + n cos(kx).49) Aω (x) = A e−ikx with A = µ0 (eω · ex ) ∞ −∞ x dt x0 ˜ dx N (ξ ) cos(kx) eikx+iωt 2 (5. but they were neglected (vg vg0 ) because they would only give second-order contributions. we have used the plasma dielectric constant (ω) = 1 − ωp /ω2 . .50) 1+ n ˜ ˜ cos(kx) . or the dependence of the photon distribution on the transverse variable r⊥ .49) Notice that.

3 Photon transition radiation Let us examine another example of a radiation process directly due to the existence of the equivalent charge: the possibility of transition radiation emitted by a photon bunch at a plasma boundary. r⊥ )v0 (x) f (x)ex (5.Photon transition radiation Integration in time leads to Aω (x → −∞) = π A dξ N (ξ ) ei(k+k)ξ δ(ω − ω1 ) ˜ ˜ 105 + e−i(k−k)ξ δ(ω − ω2 ) with ˜ ω1 = vg0 (k − k). ˜ ω2 = vg0 (k + k). showing once more the analogies between a photon in a plasma and a charged particle. (5. Let us also assume a plasma–vacuum transition layer described by 2 2 ωp (x) = ωp0 f (x).59) with Q 0 = Q ph (k) for k = k0 and for ωp = ωp0 . radiates in two characteristic frequencies which depend on the periodicity scale of the modulation. This is similar to the radiation of electrons travelling in a vacuum in the presence of a modulated magnetic field (an electron ondulator). We know that the group velocity v0 is determined by v0 (x) = c 1 − (ωp0 /ω0 )2 f (x). and v0 = vk is the group velocity for k = k0 ex . (5. we consider the particular form of a sharp plasma boundary. Such a radiation process [69] can then be called a photon ondulator.56) This result shows that an electromagnetic wavepacket (or equivalently a photon bunch described by N (ξ )) travelling in a plasma with a modulated electron density. t) = (2π)3 N (ξ.58) A plausible choice for the shape function is f (x) = [1 + tanh(x/ )]/2. (5.55) (5.57) (5. we will assume a nearly monochromatic burst of photons moving along the x-axis. We consider . where is the width of the boundary layer. We have then Nk (r . 5. In order to simplify the problem. The nonlinear current J2N L reduces to J2N L = Q 0 N (ξ.60) In order to give a simple physical explanation of the photon transition radiation. This process is asociated with the sudden disappearence of the equivalent charge [71]. r⊥ )δ(k − k0 ) where ξ = x − v0 t.

106 Photon equivalent charge the simplest case of a boundary with infinitesimal width → 0. This is very similar to the well-known case of bremsstrahlung radiation.67) . we can assume a Gaussian bunch described by N (ξ. In space. r⊥ ) = N (r⊥ )δ(ξ ). (5. it is located precisely at the plasma boundary. the transverse field radiated by the boundary is determined by ∇2 − 1 ∂2 c2 ∂t 2 E = −µ0 Q 0 v0 N (r⊥ )δ(t)δ(x)ex . (5. σ2 σ π (5. r⊥ ) = N (r⊥ ) ξ2 √ exp − 2 σ σ π (5. We also assume that the duration of the bunch of photons is negligible. The nonlinear current becomes ex (x − v0 t)2 J2N L = −Q 0 v0 N (r⊥ )H (−x) √ exp − . This means that. where the particle velocity changes and its charge remains constant (because it is an invariant). we can replace equation (5. (5. For instance. in the vacuum region (x > 0). In this case.66) The wave equation for the secondary radiated field can now be written as ∇2 − 1 ∂2 c2 ∂t 2 t2 ex E 2 = −µ0 Q 0 v0 N (r⊥ ) √ exp − 2 δ(x) τ τ π (5. the nonlinear current becomes J2N L = −Q 0 N (r⊥ )v0 H (−x)δ(x − v0 t)ex .61) This means that the source term in the wave equation can be written as ∂ 2 J2N L = Q 0 N (r⊥ )v0 H (−x)δ (x − v0 t)ex ∂t = − Q 0 N (r⊥ )v0 δ(t)δ(x)ex .62) (5.61) by a more realistic description of a bunch of photons coming out of the plasma. In time it is located at the instant when the photon bunch crosses the boundary. In order to obtain a finite spectral width. such that f (x) = H (−x).64) This equation shows that an infinitely sharp boundary will radiate an infinitely large spectrum.63) We see that the source term is located both in space and time. the velocity of the radiating particle remains nearly constant and its equivalent charge goes to zero. where H (x) is the Heaviside function. so that we can write N (ξ.65) where σ is the bunch width. In contrast here.

In the opposite limit of an infinitely large bunch.70). . such that N (r⊥ ) = N0 . In this case. Finally. and the general case of an arbitrary photon distribution crossing an arbitrary plasma boundary can be studied with the aid of equations (5. a primary distribution of photons can radiate a large spectrum of electromagnetic waves (or secondary photons) when they cross a plasma boundary with nearly constant velocity. This is a direct consequence of their equivalent electric charge in a plasma. let us discuss the importance of the transverse dimensions of the burst of particles crossing the boundary. the new type of transition radiation discussed here is also quite similar to the usual bremsstrahlung.43). Furthermore. If these transverse dimensions are negligible. and to the square of the 0 2 photon velocity. These simple calculations can obviously be refined. instead of a change in velocity. this factor ˆ∗ is equal to zero for parallel radiation k⊥ = 0. in quite general conditions. as in the usual linear [57] and nonlinear [62] transition radiation processes. the radiated power is proportional to the square of the particle velocity and of the particle equivalent charge. As already noticed. we arrive at a general solution that can be written in the following simple way: Aω (z) = A0 (ω) eikx with A0 (ω) = i µ0 Q 0 ω2 τ 2 ˆ∗ √ v0 (eω · ex )N (k⊥ ) exp − 2k τ 4 (5. However. in some respects. This shows that. we have a change in the value of the particle charge. Following a procedure similar to that of the previous section. we can write N (r⊥ ) = N0 δ(r⊥ ). This new kind of transition radiation is qualitatively similar to the usual transition radiation of charged particles (for instance electrons) moving across a dielectric discontinuity with constant velocity. because of the transverse nature of the radiated field.Photon transition radiation 107 where τ is here σ/v0 . v0 . and the radiated field will have a maximum for nearly perpen∗ dicular direction of propagation.70) This result shows that the energy radiated by the disappearing equivalent charge is proportional to the square of this charge.68) 2 where k 2 = (ω/c)2 − k⊥ . where (eω · ex ) 1.37. and N (k⊥ ) = N (r⊥ ) e−ik⊥ ·r⊥ dr⊥ .69) (5. and N (k⊥ ) = (2π )2 N0 δ(k⊥ ). (5. with the difference that. 5. Q 2 . we will have no transition radiation at all. In particular. we can use N (k⊥ ) = 1 in equation (5. radiation is not emitted along the particle direction of motion because of the geometric factor (eω · ex ).

This can also be written in terms of the photon number distribution. and what are the new physical processes associated with it.72) √ where ωp0 is the unperturbed electron plasma frequency and ve = T /m is the electron thermal velocity. the right-hand side describes the ponderomotive force due to the photon field. but with a more consistent view of the physics of the photon–plasma interaction. and its evolution was assumed to be independent of the radiation spectrum. for instance the Planck distribution for a given plasma temperature T . In this equation. In equilibrium. 2 ωk (2π )3 (5.74) where R ≡ R(ω. ˜ ˜ Nk = Nk0 + Nk . and the radiation spectrum by some distribution of photons.71) The objective of the present section is to show how the coupling between these two perturbed quantities can be described. . We can write it more explicitly as ∂v 1 1 e = − ∇|v|2 = − ∂t 2 2 m 2 ∇ |E k |2 dk .4 Photon Landau damping In section 4.73) Here we have assumed the linear solution for the electron motion in the radiation field v = −i(e E k /mωk ). But it can be easily recognized that our approach was not self-consistent in the sense that the electron plasma wave was defined a priori. Here we return to the same problem. k) = 0 is the photon dispersion relation. This means that we can write.3 we considered the interaction of a photon distribution with an electron plasma wave moving with a relativistic phase velocity. in the sense that the electron plasma oscillations will be coupled to the photon field [14].108 Photon equivalent charge 5. Nk0 . if we use the plane wave definition Nk = 8h ¯ ∂R ∂ω |E k |2 k (5. the photon field will also be perturbed because they are coupled to each other. we can characterize the plasma by its electron mean density n 0 . for the total electron plasma density n and for the total photon distribution Nk . we can derive the following equation for the electron density perturbation: ∂ 2n ˜ ∂v 2 2 ˜ ˜ + ωp0 n − 3ve ∇ 2 n = −n 0 ∇ · ∂t ∂t (5. If we use the electron fluid equations and Poisson’s equation for the electrostatic field. the following expressions: n = n 0 + n. (5. If the plasma is perturbed.

75) is coupled with the kinetic equation for the photon field which. ˜ In the general case where the radiation field is present and we have Nk = 0. we obtain ωp0 2 ∂ 2n ˜ 2 2 ˜ ˜ + ωp0 n − 3ve ∇ 2 n = 4h ∇ ¯ ∂t m 2 109 ˜ Nk dk . (5. t) = n(t) eik·r . for perturbations of the form n ∼ exp i(k · r − ωt). determines Nk as a function of Nk0 and n: ˜ ˜ ˜ ∂ Nk ∂ Nk0 ∂ Nk + vk · =−f · . ˜ 2ωk 0 m (5. characterizing the scale of the perturbation in both the electron density and the photon spectrum. 2 (∂ R/∂ω) (2π )3 ωk k (5. (5.Photon Landau damping Replacing this in equation (5. We will come back to that point later. Let us assume sinusoidal perturbations of the form n(r . the force acting on the 2 photons is determined by f = −∇ωk . Nk = 0. and not to the photon spectrum. which makes plausible the description of the transverse radiation field as a fluid of particles (the photons) . t) = Nk (t) eik·r . the right-hand side of this equation is equal to zero and. Furthermore.77) According to our theory of photon acceleration. the value of the wavevector k.78) The pair of equations (5. ∂t ∂r ∂k (5. ˜ after an appropriate linearization. where ωk = ωp + k 2 c2 . ˜ ˜ ˜ ˜ Nk (r .76) It should be kept in mind that. it is known from the plasma kinetic theory that this dispersion relation is only valid if the electron Landau damping is negligible. This means that the electron plasma wave has a much ˜ larger scale length than the photons described by Nk . ˜ this equation will lead to the well-known linear dispersion relation for electron plasma waves 2 2 ω2 = ωp0 + 3ve k 2 . which implies that the phase velocity of the electrostatic wave has to be much larger than the electron thermal velocity: (ω/k) ve . in this equation.79) Here.75. should not be confused with the wavevector k characterizing a specific Fourier component of the photon spectrum. 5. In physical terms we are dealing with a situation which only makes sense when we have k k.72). the frequency ω and the wavenumber k are related to the electron plasma wave spectrum.77) will then describe the coupling between the electron plasma waves associated with the density perturbation n and the photon ˜ ˜ number perturbations Nk .75) ˜ In the absence of any radiation field. This leads to the following explicit expression for the force acting on the photons: f =− 1 e2 ∇ n. equation (5.

79) in equations (5. Using equation (5. 5. ω 2 u − (ω/k) (5.82) ˜ and a similar transformation for Nk (ω).77). ω 2 ω − k · v(k ) (2π )3 (5. from equation (5. This can be done by defining k k = p + k⊥ . where ω is here a complex quantity.81) where ω ≡ ω(k ). k k v(k ) = u( p.110 Photon equivalent charge evolving in a slowly varying background medium (the plasma and its electrostatic wave).85) becomes 1 k · (∂ Nk 0 /∂ k ) dk =− ω 2 ω − k · v(k ) (2π)3 dk⊥ (2π )3 d p ∂ Nk 0 /∂ p . k (5. We can now follow the usual Landau approach [82].84) We can use (∂ R/∂ω)ω 2/ω . 2n 0 ω ω − k · v(k ) 2 (5.83) Similarly. This implies the use of a time Laplace transformation n(ω) = ˜ 0 ∞ n eiωt dt ˜ (5. ω − k · v(k ) (2π )3 (5. From these two equations we can then derive the following expression: 2 2 ω2 = 3k 2 ve + ωp0 1 − hk2 2 ¯ ω mn 0 p0 1 k · (∂ Nk 0 /∂ k ) dk .80) we get 2 2 ωp0 + 3k 2 ve − ω2 n ˜ 4 = 2ωp0 hk2 n ˜ ¯ m n0 1 ω 3 (∂ R/∂ω) ω k · (∂ Nk 0 /∂ k ) dk . k⊥ ) + v⊥ . ˜ From equation (5.87) .80) and ˜ ∂ Nk ∂ Nk 0 i e2 ˜ + ik · vk Nk = nk · ˜ ∂t 2ω 0 m ∂k (5.75.81) we then get a relation between n(ω) and Nk (ω): ˜ n ωp0 k · (∂ Nk 0 /∂ k ) ˜ ˜ Nk = − .85) In order to develop this integral we have to consider separately the parallel and the perpendicular photon motion with respect to the propagation of the electron plasma wave. but apply it to the photon distribution and not to the electron distribution. we obtain ωp0 ∂ 2n ˜ 2 2 + (ωp0 + 3k 2 ve )n = −4h k 2 ˜ ¯ 2 m ∂t 2 Nk 2 (∂ R/∂ω) ω ˜ ω dk (2π )3 (5.86) The integral in equation (5.

Replacing this result in equation (5. ω 2 p − p0 (5. p0 (5.Photon Landau damping 111 Developing the parallel photon velocity u around the resonant value defined by u( p0 ) = (ω/k) u( p.90) where P means the Cauchi principal part of the integral. 5.89) This last integral can be written in the standard form I (z) ≡ h(z) dz = P z − z0 h(z) dz + iπ h(z 0 ) z − z0 (5. p − p0 ∂ p (5.93) Equations (5.91) can be rewritten as 1 + χe + χph = 0 (5. k⊥ ) + ( p − p0 ) ∂u ∂p (5.91. this new damping effect can be called the photon Landau damping.85) and using ω = ωr + iγ . The similarities between the electron and the photon contributions can be stated in a more appropriate way if we notice that the dispersion relation (5.92) are the main results of this section.92) In these two expressions. which resulted from the integration of the equilibrium photon distribution Nk 0 over the perpendicular directions: Gp = 1 ω 2( p dk⊥ Nk 0 .94) where χe is the electron susceptibility and χph is the photon susceptibility.91) and from the imaginary part γ =π 3 h k 2 ωp0 ¯ 2mn 0 ∂G p ∂p . Due to the similarities of this expression with the well-known electron Landau damping. The second one determines the damping of these plasma waves due to the resonant interaction with the photon spectrum. G p is a kind of reduced distribution function for the photon gas. k⊥ ) we get u( p0 . 3 0 ) (∂u/∂ p) p0 (2π ) (5. The first one represents the dispersion relation of electron plasma waves (in the high phase velocity regime) in the presence of a photon distribution.88) p0 d p ∂ Nk 0 /∂ p ω 2 u − (ω/k) 1 (∂u/∂ p)p0 d p ∂ Nk 0 /∂ p . we get from the real part of the resulting equation 2 2 2 ωr = 3k 2 ve ωp0 1 + hk2 2 ¯ ω P mn 0 p0 1 ∂G p dp . .

92) this equilibrium distribution will always lead to a negative value of γ .92). However.91) the electron susceptibility is χe = − 2 ωp0 2 ωr −3 2 k 2 ve 2 ωr (5.96) As we have noticed already. For a plasma in thermal equilibrium at a temperature T . The important point to notice here is that these electron kinetic effects are negligible for electron plasma waves with high phase velocities. but in the presence of a spectrum of electron plasma waves [14]. even in the absence of resonant electron populations which could exchange efficiently their energy with the electron plasma wave. Because of the derivative contained in the expression for the photon Landau damping (5. These electron kinetic effects are well documented in plasma physics textbooks and will not be described here. This means that photon Landau damping can replace the electron Landau damping. the situation can change if a photon beam is superposed on the Planck distribution.112 Photon equivalent charge According to equation (5. p − p0 ∂ p (5. This will be illustrated below. like those produced by intense laser pulses propagating in a plasma. In this case the resulting electron plasma waves (or wakefields) have relativistic phase velocities nearly equal to the laser pulse group velocity from which they originate. indicating the possibility of a wave instability or wave growth. Then.97) π c exp(h ω /T ) − 1 ¯ 2 where ω = ωp0 + k 2 c2 should be used in the above equations. The resulting wave damping is described by equation (5. as already documented in chapter 3 for single photon trajectories. It should also be noticed that a quasi-linear diffusion coefficient for the photons can be derived from this theory. the Planck distribution 1 ω2 Nk 0 = 2 3 (5. or to a real wave damping. The important conclusion of the above calculation is that. an electron Landau damping would have to be added to the above photon Landau damping. χe would depend on the actual electron distribution function and moreover.95) and the photon susceptibility is χph = − hk2 2 ¯ ω P mn 0 p0 1 ∂G p d p. . in some regions of phase velocities. showing that in many respects the photon dynamics in the field of an electrostatic wave in a plasma is similar to the electron dynamics. the above expression for the electron susceptibility is only valid in the limit of very high phase velocities. the relativistic plasma perturbations can nevertheless exchange energy with the photon gas. In the general case. the derivative can be positive. showing how the photon spectrum changes in the presence not of a single plasma wave as before.

106) . after integration by parts.98) ∂ Nk /∂k 1 dk 2 (ω/k) − v(k ) 2π ωk (5.101) Let us assume a mono-energetic photon beam propagating across the plasma: Nk = 2π N0 δ(k − k0 ). mn 0 ω03 2 (5. Using the photon (group) velocity v(k ) = we obtain.102) χph = − 4 h k 4 ωp0 k 2 c2 N0 ¯ mn 0 ω2 ω0 ω0 (ω − ku 0 )2 (5.105) ω (ω − ku 0 )2 where we have used 2 = h k 4 c2 ωp0 ¯ N0 .100) (5.99) where k ≡ k .5 Photon beam plasma instabilities In order to illustrate the new physical aspects contained in the dispersion relation (5. The photon susceptibility will then reduce to χph = h k 2 ωp0 ¯ mn 0 ω2 2 (5. after performing the integration.Photon beam plasma instabilities 113 5. ωk [(ω/k)(ωk /c2 ) − k ]2 2π c c2 k ωk (5. 1 ∂ Nk /∂k dk ωk (ω/k) − (c2 k /ωk ) 2π 1 1 Nk dk =− 2 .104) ω The dispersion relation for the electron plasma waves (5.96) let us consider a one-dimensional problem where Nk 0 = (2π )2 Nk δ(k⊥ ).103) where u 0 = v(k0 ).94.94) can then be written as 2 2 ωp0 1− 2 1+ =0 (5. 5. if we neglect the electron thermal effects. (5. Furthermore. Using ω0 = ωk we can write. 0 (5. the electron susceptibility reduces to 2 ωp0 χe = − 2 .

109) Noting that. the photon Landau damping would not be possible. (5.107). 2 we always have k 2 c2 ω2 .105). and we can define a photon plasma frequency ph . we can also write the approximate expression p0 2 ωp0 4ω0 2 ph . We can then obtain the growth rate for the photon beam plasma instability as √ 3 µ= ωp0 2 2 2 2ωp0 1/3 .114 Photon equivalent charge We can see from equation (5. This means ¯ that we can write 5 4h N0 ωp0 ¯ 2 . (5. such that Q 2 N0 ph 2 = (5.107) ph 0 m eff where Q ph is the photon equivalent charge as determined by equation (5.105) that this quantity plays nearly the same role in the photon susceptilibity that is played by the electron plasma frequency ωp0 in the electron susceptibility. the fact that this quantity ph appears in the photon susceptibility shows that the concept of the photon equivalent charge Q ph is already implicit in the photon susceptibility term and in the photon Landau damping. But. If the photon effective charge did not exist. (5. we can assume the resonant condition ωp0 = ku 0 .37).112) . Due to that.106). We can then conclude that the high-frequency photons behave in a plasma like any other charged particle. we can call it the effective photon plasma frequency. (5. (5. we obtain 2 = 1 4 kc ωp0 4 ωp0 ω0 2 ph .108) ph = mn 0 (ω0 c)2 Comparing this with equation (5.110) We see that the effective photon plasma frequency is always significantly larger than the photon plasma frequency ph as defined by equation (5.111) We notice that this growth rate is proportional to the power 1/3 of the density of photons in the beam: 1/3 µ ∝ N0 . and m eff = h ωp0 /c2 is the photon effective mass introduced in chapter 2. for the electron plasma waves resonant with the photon beam. Coming back to the dispersion relation (5. and we can write ω = ωp0 + iµ. formally identical to the electron plasma frequency. But the analogy with a plasma frequency can be explored even further.

appling to our photon beam plasma instability. we can establish the following equation for the propagating electric field: ∇2 E − ∇ ∇ · E − ∂2 P 1 ∂2 E = µ0 2 .112). which are clearly distinct from the resonance condition ω = ku 0 . In particular. in contrast with the result of equation (5. The photon beam plasma instability should also be compared with the forward Raman scattering [50]. where ω1 and k1 are the frequency and wavenumber of the scattered photons. The difference with respect to the plasma case is related to the absence of free electrons in this medium. which is clearly distinct from the usual Raman decay instability. in this respect. Electron plasma waves are also produced here. We see that this new growth rate is proportional to N0 . we examine a laser pulse propagation along an optical fibre having a second-order nonlinearity and show that an equivalent electric dipole can be derived [70]. Here we examine a similar concept for non-ionized dielectric media. This corresponds to the well-known energy and momentum conservation rules. 2 ∂t 2 c ∂t (5. thus showing that the photon beam plasma instability is a new kind of instability. 1/2 5.114) . but the physical mechanism is completely different because it implies that the usual resonant conditions for wave decay are satisfied: ω1 = ω0 ± ω and k1 = k0 ± k. Furthermore. the maximum growth rate for the forward Raman scattering is determined by µ 2 ωp0 vos √ 2 2ω0 c (5.Equivalent dipole in an optical fibre 115 This instability is comparable to the usual beam plasma instability produced by the interaction of a monoenergetic electron beam (with mean velocity u 0 ) with a plasma. just like electrons. and E 0 is the amplitude of the electric field associated with the incident photon beam. This shows again that the photons behave. which is a different way of describing the ponderomotive force or radiation pressure effects. From Maxwell’s equations. for which the growth rate is also proportional to the density of particles in the beam to the power 1/3. which occurs when a photon beam (with a welldefined field phase) interacts with a plasma.6 Equivalent dipole in an optical fibre We have shown that it is possible to define an equivalent photon charge in a plasma. in the absence of charge and current distributions.113) where vos = eE 0 /mω0 .

for propagation along an optical fibre. can be written as PNL (t) = 0χ (2) · E(t) E(t) + 0χ (3) · E(t) E(t) E(t).117) ∇ 2 Er − ∇ ∇ · Er − = µ0 ∂ Jr ∂t 1 ∂2 Er + c2 ∂t 2 ∞ 0 χ (1) · E r (t − τ ) dτ (5.116 Photon equivalent charge The polarization vector is P = PL + PNL . t) = ap ˆ R(r⊥ )A(z.116) We now assume.117) To the lowest order in the perturbation field E r .115) The nonlinear part of the polarization vector. In general. if we neglect dispersion effects. defined by Jr = 0 ∂ ˆ ˆ χ (2) · E p E p + χ (3) · |E p |2 ap ap E r . ∂t (5.120) where Jr is the nonlinear polarization current. that the electric field can be divided into two terms: the dominant field associated with the propagating pulse E p . and with a group velocity vg . (5. To the first order in the radiation field E r .121) .114–5.119) This field is assumed to propagate along the fibre axial directon z with negligible pulse shape deformation.c. we can write from equations (5. (5. 2 (5. as we already did for the plasma case. with the linear part defined by PL (t) = ∞ 0 0 χ (1) (τ ) · E(t − τ ) dτ. and a secondary perturbation or radiation field. c2 ∂t 2 From this equation we can obtain linear (if χ (3) = 0) or nonlinear (if χ (3) = 0) wave pulse solutions. (5. such that |E r | |E p |: E = Ep + Er. we get from equations (5. we can write both types of solution in the general form E p (r . E r . t) ei(kz−ωt) + c. .117) ∇ 2 Ep − ∇ ∇ · Ep − = 1 ∂2 Ep + c2 ∂t 2 ∞ 0 χ (1) (τ ) · E(t − τ ) dτ (5.118) 1 ∂ 2 (3) χ · |E p |2 E p . and assuming for convenience that χ (2) is of the same order of E r .114–5.

z − vg t). we will use χ (2) · ap ap = χ (2) e in the ˆ ˆ ˆ remaining term. (5. The nonlinear polarization current Jr is then reduced to Jr = where we have used W (r .123) can be replaced by a space derivative ∂ ∂ W = −vg W.126) As a result of this replacement. t) = |R(r⊥ )|2 |A(z.123) This means that the time derivative appearing in equation (5. due to the existence of free electrons which are pushed away by the electromagnetic pulse. we can ignore it here.125) 0 2 χ (2) ∂ W (r . (5. 117 (5. t) = 1 ap ap R 2 (r⊥ )A2 (z.c. t) ≡ W (r⊥ . we can see that E p (r .127) This quantity Q(r . ∂z (5. ∂t ∂z (5. This clearly contrasts with the plasma case where. the equivalent charge distribution reduces to a negative monopole charge. z − vg t). and it is determined by Q(r .Equivalent dipole in an optical fibre Using equation (5. t) e2i(kz−ωt) ˆ ˆ 4 + |R(r⊥ )|2 |a(z. For simplicity. t)e. t) = − 0 2 χ (2) ∂ W (r⊥ . for this reason.122) The first term oscillates in time as exp(2iωt) and is associated with harmonic generation. The term in χ (3) describes the scattering of a probe radiation field by the main pulse and can also be ignored because it does not play any role in the effect to be discussed here. t) can be called the equivalent charge distribution of the main pulse. .119). t)e ˆ ∂t (5. we can write W (r . t)|2 + c. . if the main pulse moves along the fibre with no significant change in its envelope shape. we can rewrite the nonlinear polarization current as the product of a charge with a velocity Jr = vg Q(r . t) E p (r .128) This is clearly a dipole charge distribution because the space derivative reverses its sign when we move from the front to the rear of the pulse.124) It is obvious that. t)|2 . Its physical content is well understood and. ˆ (5.

ˆ c2 ∂t It should be noticed that this secondary field is not always a radiation field. t)e.128) along the pulse distribution: ∞ −∞ Q(z. where the second-order nonlinearities are forbidden (χ (2) = 0). second. But even in this case we can imagine an external loop where the existence of such a field can eventually be detected.127) in equation (5. one negative and the other positive. Furthermore. t) = −Q eq (r⊥ ) δ z − where L L − vg t − δ z + − vg t 2 2 (5. t) dz = − 0 2 χ (2) ∂ 0 W dz = − χ (2) [W ]∞ = 0. For instance. −∞ ∂z 2 (5. Radiation will typically occur in two distinct situations: first. when the charge amplitude changes due to the space variation of the nonlinear susceptibility χ (2) . In this case. it will reduce to a near field. which is clearly associated with an acceleration of the moving dipole. as can be seen by integrating the distribution (5. corresponding to Q(r . the source of radiation is a linear term of the form δχ (1) E p .131) χ (2) W0 (r⊥ ) 2 for a pulse of length L and maximum intensity W0 . In contrast with our present radiation mechanism. this property is independent of the actual shape of the pulse. where both the main pulse velocity vg and the charge amplitude Q eq are constant. These two radiation mechanisms are clearly distinct from those usually considered in the literature [2]. which are mainly concerned with soliton propagation in inhomogeneous but centro-symmetric media. where δχ (1) is the perturbation of the linear susceptibility due to the inhomogeneities of the medium.129) This means that we have here total charge conservation. Replacing equation (5. in a homogeneous medium. Let us consider the interesting limiting case of a rectangular pulse: the equivalent charge distribution reduces to two point charges.132) 1 ∂ vg Q(r . these more common aspects of secondary radiation can be . when the group velocity vg changes due to a space dependence of the linear susceptibility of the medium χ (1) . which is physically quite reassuring.120) we can determine the secondary field E r created by this moving dipole: Q eq = ∇ 2 Er − ∇ ∇ · Er − = 1 ∂2 Er + c2 ∂t 2 ∞ 0 χ (1) · E r (t − τ ) dτ (5.130) (5.118 Photon equivalent charge An interesting feature is that the total equivalent charge here is equal to zero.

2π (5. Equivalent dipole distribution for (a) a sech laser pulse.Equivalent dipole in an optical fibre 119 (a) Q eq (+) (-) (z .133) . described as scattering of the lowest order pulse field by the local inhomogeneities of the medium.vg t) (b) Q eq (+) (-) (z . Let us study equation (5. and (b) a nearly rectangular pulse.v t) g Figure 5. t) = ∞ −∞ E ω (r ) e−iωt dω .132) in more detail.1. and introduce a time Fourier transformation of the field as E r (r .

ˆ Replacing it in the above equation. t) eiωt dt (5. t) is the axial equivalent charge distribution. Furthermore. z. (5.120 Photon equivalent charge The Fourier components E ω will be determined by the equation ∇2 + ω ω2 ˆ (ω) E ω − ∇ ∇ · E ω = i e c c2 ∞ −∞ vg (z)Q(r⊥ . (5. |Rω (r⊥ )|2 dr⊥ (5.136).136) This equation can be solved by considering first the corresponding homogeneous equation ∂2 ω2 2 ∇⊥ Rω + 2 Aω + 2 (ω) = 0.137) ∂z c If we assume that the radial field amplitude Rω is determined by the equation 2 2 ∇⊥ Rω = −k⊥ Rω (5. determined by q(z. t) dr⊥ .141) where q(z. t) eiωt dt. Let us retain our attention on radiated transverse fields.134) where (ω) is the linear dielectric function of the medium. we get 2 Aω ∇⊥ Rω + Rω (5. in a cylindric geometry.139) The axial wavenumber k is determined by the relation k2 = ω2 2 (ω) − k⊥ .142) . t) = ∗ Rω (r⊥ )Q(r⊥ . z. the axial field amplitude Aω will have the solution Aω = A± e±ikz . c2 (5. such that ∇ · E ω = 0.135) ω2 ∂2 Aω + 2 (ω)Rω Aω ∂z 2 c ∞ −∞ ω = i (e · eω )vg (z) ˆ ˆ c Q(r . we can write the inhomogeneous equation for the axial field amplitude Aω in the form ∂ 2 Aω ω ˆ ˆ + k 2 Aω = i (e · eω )vg (z) c ∂z 2 ∞ −∞ q(z.138) where k⊥ is the perpendicular wavenumber.140) Returning to equation (5. we can use E ω (r ) = Rω (r⊥ )Aω (z)eω . (5. t) eiωt dt (5.

t) = q0 δ z − L L − v0 t − δ z + − v0 t 2 2 . for a very long fibre. (5. and we obtain Aω (z) = A0 q0 eikz − cos k0 eiωt cos k0 L − k 0 v0 t 2 L + k 0 v0 t 2 e−ik(L/2)−ikv0 t (5. t) = q0 (z. in the above solution.144) This means that we can replace. To be more specific. ω2 = v0 (k + k0 ). which can be described by vg (z) = v0 = const. let us assume that we have an axial modulation of the nonlinear properties of the optical fibre. ˆ ˆ 2kc In the particular case of a rectangular pulse shape.141).145) The forward field solution becomes Aω (z)+ = A0 eikz with an amplitude dt eiωt z z0 dz q0 (z . we obtain Aω (z)+ = i ω (e · eω ) eikz ˆ ˆ wc dt eiωt z z0 dz vg (z )g(z.147) (e · eω )v0 δ. we can write A0 = q0 (z.143) where the Wronskian is w = 2ik.146) ω (5. t) cos(k0 z) e−ikz (5. the axial equivalent charge by q(z. (5.150) L δ(ω − ω2 ) eikz . as described by equation (5.151) . t) e−ikz (5.Equivalent dipole in an optical fibre 121 We now use the well-known method of the variation of parameters to solve equation (5.149) eik(L/2)−ikv0 t dt. (5. For forward propagation. t)δ cos k0 z.146) in z. χ (2) = χ20 [1 + δ cos(k0 z)] . 2 The values of the two frequencies are determined by ω1 = v0 (k − k0 ). From this equation it can then be concluded that two distinct frequencies are radiated by the main pulse when it propagates in a modulated optical fibre: Aω (z) = 2πiA0 q0 sin (k0 − k) + sin (k0 + k) L δ(ω − ω1 ) 2 (5.130).148) We can now easily integrate equation (5. (5.

Finally. As already noticed for the plasma case treated in section 5.152) +e−i(k0 −k)ξ δ(ω − ω2 ) dξ with ξ = x − v0 t.150) is replaced by Aω (z)+ = π A0 eikz q0 (ξ ) ei(k0 −k)ξ δ(ω − ω1 ) (5. This radiation process is a direct consequence of the existence of an equivalent dipole charge of the optical pulse propagating in an optical medium with second-order nonlinearities. in this case. k k0 . the solution (5. it can be considered as the analogue of the electron ondulator.2. ω2 is a kind of second harmonic. . we note that. or another version of the photon ondulator. for the more general situation of a pulse with an arbitrary shape. This means that.122 Photon equivalent charge When the wavelength of the radiated field is nearly equal to the characteristic wavelength of the fibre inhomogeneity. we have ω1 0 and ω2 2v0 k0 .

6. by considering reflection and transmission of an incident electromagnetic wave at the boundary between two stationary dielectric media.Chapter 6 Full wave theory In previous chapters we were able to describe a large number of different physical phenomena using the Hamiltonian approach to photon dynamics. 78.1 Reflection and refraction Let us assume two different media. t) = E i exp i(ki x − ωt) + E r exp i(kr x − ωt). However. and it will be the subject of the present and the next chapters.1) 123 .1. which we called time refraction. Several papers have been devoted to this problem [26. A full wave description of the electromagnetic radiation is therefore necessary.1 Space and time reflection For comparison with the cases of time-varying media. can also take place. 118. (6. can be called time reflection. could take place. which. with a sharp boundary at the plane x = 0. In chapter 3 we showed that an effect similar to the usual refraction. the total electric field will be determined by E(x. either in its simple form of single photon trajectories or in its more elaborated version of the photon kinetic theory. Here we will see that a new effect. 27. we start first by reminding ourselves of the well-known derivation of the Fresnel formulae. the geometric optics approximation associated with this elegant and powerful approach is not capable of explaining such basic and sometimes important phenomena as partial reflection and mode coupling. with dielectric constants 1 and 2 . E t exp i(kt x − ωt). If a plane wave with frequency ω and wavevector ki propagates along the x-axis and interacts with this boundary. 6. 123]. (x < 0) (x > 0). in the same spirit.

6) Assuming perpendicular polarization ( E ·ex = 0). n1 (6. (6.2) √ with the refractive indices n 1. In the absence of charge and current distributions in the media (ρ = 0.124 Full wave theory This field has three different terms. for plane waves propagating in nonmagnetic media.3) we realize that. these equations are ∇×E =− and ∇ · D = 0.8) Ei n1 + n2 Ei n1 + n2 These expressions verify the simple relation 1 + R = T. E r and E t . ∇ · B = 0. which implies that ki = ω n1. c kr = −ki . These are determined by Maxwell’s equations.5) ∂B . (6.2 . the components of E and H tangent to the boundary have to be continuous across this boundary: ( E 1 − E 2 ) × ex = 0. (6.2 = 1.9) These results are extremely well known and we should not insist on their physical significance.2). ∂t ∇×H = ∂D ∂t (6. ω (6. . corresponding to the incident. the reflected and the transmitted waves. J = 0).5) in the form Ei + Er = Et. We should instead use them as a reference guide for the less understood time-varying processes. R≡ = .4) It is well known that. we have to state the associated boundary conditions. in order to satisfy the first pair of equations. ( H1 − H2 ) × ex = 0. and using the wavenumber relations (6. Ei − Er = Et n2 .7) From this we derive the well-known Fresnel formulae for normal wave incidence Et 2n 1 Er n1 − n2 T ≡ = . we can write the boundary conditions (6. we can write B ≡ µ0 H = k×E .3) From equations (6. (6. In order to establish a relation between the field amplitudes E i . kt = ki n2 n1 (6. These waves have to satisfy the dispersion relations of both media.

11) We should take notice of the symmetry between these relations and those of equation (6.12) Expressing the fields D and B in terms of the electric field E for the three distinct waves present in equation (6. E r exp i(kx − ωr t) + E t exp i(kx − ωt t). n2 (6. are replaced by these new conditions in the case of a sharp time boundary. The time continuity conditions are then given by D(t = 0− ) = D(t = 0+ ). B(t = 0− ) = B(t = 0+ ).10). the three distinct plane waves have to satisfy the linear dispersion relations of the medium. (t < 0) (t > 0). 6. for t = 0 the refractive index will be suddenly shifted to a new value n 2 .7).1) E(x. we have to add to fields (6. will be described by an expression similar to equation (6. (6. which implies that ωi = kc . For times t < 0. ω t = ωi n1 .3.14) . We can see from Maxwell’s equations (6.4) that they only contain the time derivatives of the fields D and B. 6.Space and time reflection 6.2 Time reflection 125 Let us turn to the opposite situation of a time discontinuity (or a time boundary) occurring in a homogeneous and infinite medium.2).13) 1 2 n1 This means that the well-known continuity conditions (6. in order to describe a real wave. These equations can only stay valid for all times if these two fields are continuously varying in time. (6. we have to establish the time continuity conditions for the electromagnetic field. It can easily be understood that the negative frequency mode corresponds indeed to a reflected wave because. with the initial frequency ωi .1. This operation will then lead to the appearance of a plane wave with positive frequency and propagating in the opposite direction with respect to that of the initial wave: ∗ E r exp[−i(kx + |ωr |t)]. E i = −E r + E t . we can rewrite these continuity conditions as n2 n 2 E i = n 2 (E r + E t ). (6.1. The explicit result for the reflected and the transmitted wave amplitudes is R≡ Er 1 = Ei 2 n2 1 n2 2 −1 .10) As before. the medium will have a refractive index n 1 everywhere and. The electric field associated with a plane wave propagating in the medium. In order to determine the relative amplitudes of the transmitted and the reflected waves. (6. T ≡ Et 1 = Ei 2 n2 1 n2 2 +1 . valid for a sharp spatial boundary in the dielectric properties of the propagating medium.10) their complex conjugates. n1 ωr = −ωt . t) = E i exp i(kx − ωi t).

there are important qualitative differences related. (6. This means that we could expect to produce electromagnetic energy out of nearly nothing. A large amount of work has been devoted to the problem of wave transformation due to a sudden creation of a plasma medium. but also to the appearance of a magnetic mode. this is the electron plasma wave. for a plasma.16) with ki = (ωi /c)n 1 . Furthermore. (6. the amount of energy of the waves created by time reflection would have to be furnished by the external agent responsible for the sudden ionization process. Let us assume an incident wave of the form E i exp i(ki x − ωi t) (6. which can be excited in the medium when its dielectric constant is equal to zero. This can be obtained. However. a wave propagating in the opposite direction with respect to the initial wave. by plasma creation (or flash ionization) with a nearly resonant plasma frequency ωp ∼ ωt . The field of this electrostatic wave would have to be added to the above continuity conditions. the concept of time refraction was never explored. in exactly the same way as photons can be created from a vacuum in quantum models. The first surprise is that a time discontinuity can lead to a reflected wave. As we know. for instance. The second is that the amplitudes of both the reflected and the transmitted waves can be very large if the value of the new refractive index n 2 is very low. We should however notice that the limit of n 2 = 0 cannot be properly described by the above model because it does not take into account the existence of the electrostatic mode. not only to dispersion.15) These equations can be called the Fresnel formulae for time reflection. the same wave will be represented by E i exp i(ki x − ωi t ). An earlier discussion of this problem can be found in reference [110].8). as will be seen later. but they also show quite surprising differences.17) . 6. for the plasma case. They are very similar to the usual Fresnel formulae (6. In the reference frame moving with the boundary.2 Generalized Fresnel formulae We can generalize the above treatement of independent space and time discontinuities by considering the case of a sharp boundary of the refractive index moving in space with a constant velocity u = uex . in several different configurations including the presence of a static magnetic field [44]. In this case. that is.126 Full wave theory A simple relation between R and T can also be established here: T+R= n2 1 n2 2 . of course.

It is a very simple and useful result which.23) E i exp i(ki x − ω t ) + E r exp i(kr x − ω t ).18) c with β = u/c and γ −2 = (1 − β 2 ). We are now able to repeat.19) From this we get the dispersion relation in the moving frame ki = with n1 = ω n c 1 (6. This means that the particles contained in the plasma medium (electrons and ions) all move with an average velocity equal to u.20) n1 − β . Replacing equations (6. The case of a plasma will be considered below.21) This equation relates the values of the refractive index as seen in the moving and in the rest frames. this transformation is only valid for a non-dispersive medium. we obtain the appropriate Lorentz transformations for the frequency and the wavenumber: ω = γ ωi (1 − βn 1 ). Of course.16. i. (x < 0) (x > 0).8). with w = (kt /ki ) = (n 2 /n 1 ). cannot easily be found in the literature. (6. and separately equating the terms containing x and t . in the moving frame. t ) = Er 1−w .Generalized Fresnel formulae 127 Let us invoke the Lorentz transformations from the moving frame to the rest frame: β x = γ (x + ut ). ki = γ ki 1 − β n1 .e. We know that the field phase in equations (6. surprisingly. . the above derivation of the Fresnel formulae (6.1).18) in this equality. To do this we assume that the total electric field in this frame is formally identical to that of equation (6. 6. Let us see how such a result can be extended to the plasma case.22) In analogy with equation (6. To do that we assume that medium 1 is just the vacuum region and medium 2 is a plasma moving in a vacuum with velocity u. E t exp i(kt x − ω t ). This leads to the result E (x .17) is a relativistic invariant. 1 − βn 1 (6.2). t = γ t + x (6. which means that ki x − ωi t = ki x − ωi t . (6. the three different wavenumbers are related by kr = −ki and kt = ki (n 2 /n 1 ). = Ei 1+w Et 2 = Ei 1+w (6.

9). β = 0.28) do not reduce to the Fresnel formulae for time reflection (6.23) we then get R≡ Er 1 − βn 1 1 − w . = Ei 1 − βn 1 1 + w (6. For the field of the incident wave. we have E i = γ (E i − u Bi ) = γ E i (1 − βn 1 ). 1 − βn 1 (6. We can now make a Lorentz transformation of the electric fields back to the rest frame. these equations reduce to (6. In the limit of an infinite velocity. β → ∞.8. However.29) also reduces to (6. as expected.128 Full wave theory It can easily be realized that the plasma frequency is a relativistic invariant 2 ωp = k 2 c 2 − ω 2 = k 2 c 2 − ω 2 . which would be the case for a time discontinuity.25). for the plasma case. This apparent incongruence is due to the fact that the calculations presented in this section are based on Lorentz transformations and therefore they are not valid for β > 1. in this limit. (6. 6. 6.29) These equations can be called the generalized Fresnel formulae for moving dielectric perturbations. (6. equation (6.28) We can easily find that 1+ R = T 1 − βn 2 . (6. equations (6.23) stay valid for the moving plasma case if we assume that w is determined by equation (6. The transformations (6.3 Magnetic mode An interesting new effect occurs when the moving discontinuity of the refractive index is associated with an ionization front.27) For the reflected wave β is replaced by −β and for the transmitted wave n 1 is replaced by n 2 . (6. From equations (6.24) This means that.25) We clearly see that w = 0 defines a cut-off condition and that total reflection will occur for 1+β 2 ωp = ω2 .26) 1−β This is in agreement with our single photon model of chapter 3.21) is not valid and we have w= kt c = ω 1− 2 ωp ω2 .15). equation (6. We notice that for a stationary boundary. We are refering to the excitation of .14). = Ei 1 + βn 1 1 + w T ≡ Et 1 − βn 1 2 .

On the other hand.Magnetic mode (a) 129 vacuum (b) + - - moving + plasma moving boundary + plasma .31) dt ∂t ∂x Notice that. the plasma electrons (assuming that they are created with zero kinetic velocity) will appear to flow away from that front with exactly the same speed.3). 97]. in the same reference frame. because the plasma electrons would be moving with the front. we would have to use u = 0 in these equations. In this frame.32) J = −en 0 γ v (6. The electron equation of motion can be written. (6. which can now be written as ∂H . ∂t where the electron current is simply ∇ × E = −µ0 ∇ ×H =J + 0 with ∂E ∂t (6.1. a purely magnetic mode [53. if instead of an ionization front we were considering a plasma moving with the same velocity. Comparison between the two cases: (a) a moving plasma and (b) a moving ionization front. The electric and magnetic fields E and B are determined by Maxwell’s equations (6.30) ( E − uex × B ) dt mγ d ∂ ∂ = −u . Let us study the problem in the frame of the moving ionization front. the atoms of the neutral gas region are seen to flow with a velocity −u and to disappear at the front discontinuity x = 0. This will be examined here in some detail and will be compared with the (at first sight) generic results of the previous section. as dv e =− (6.plasma + neutral gas Figure 6.33) .

(6. km = γ km + β ω c = ω1−β . We obtain i(ω + k u)v = e u E − ex × (k × E ) mγ ω ω2 c2 E = −en 0 γ µ0 ω v .37) (6.39) Using the Lorentz transformation for the electric field we get. From the . as will be shown below.35) i k × (k × E ) + Notice that we cannot divide the first of these equations by (ω +k u) because such a factor can eventually be equal to zero. (6. for a given frequency. two distinct modes are possible: k = and k = km = − ω c 1 − (ωp /ω )2 ω . km m (6. c 1+β (6. valid for transverse modes (k · E = 0) in the plasma region: 2 (k 2 c2 + ωp − ω 2 )(ω + k u) = 0. u (6.34) (6.130 Full wave theory and n 0 is the electron mean density in the rest frame.36) This shows that. according to Maxwell’s equations (6.41) This shows that the magnetic mode is a purely magnetostatic perturbation in the rest frame. with zero frequency but a finite wavelength 2π/km = 0. The second mode can be called the magnetic mode because its electric field is equal to zero in the rest frame. We can now solve these equations by assuming that the fields evolve in space and time according to exp i(k x − ω t ). we have Em = ω B = −u Bm . E m = γ (E m + u Bm ) = 0.32). in the rest frame.38) The first mode is the usual transverse electromagnetic mode in a plasma. This can be seen by noting that.40) Similarly. From these two equations we can easily derive the following dispersion relation. it can be shown that its frequency in the rest frame is also equal to zero: ωm = γ (ω + ukm ) = 0. (6.

E t exp i(kt x − ω t ) + E m exp i(km x − ω t ). = Ei (1 − β)(1 + βw) (6.43) ki ki E (x .43) the three conditions for the electric field amplitudes: Ei + Er = Et + Em E E i − E r = wE t + m β E E i + E r = w2 E t + m .47) .44) E i exp i(ki x − ω t ) + E r exp i(kr x − ω t ). This extra condition is provided by the second of Maxwell’s equations (6. we can derive from this and from equations (6. t ) = where kt =w= ki 1 − (ωp /ω )2 .32). Because the electric field is also continuous.Magnetic mode 131 above discussion we can conclude that the total electric field associated with an ionization front will have four (and not three) distinct waves: (x < 0) (x > 0). Ei − Er = Et t − Em m (6. We need to complement the two continuity conditions (6. (6. this implies that ∂ B /∂ x will also be continuous across the boundary: ∂B ∂x = ∂B ∂x .43) by another condition because we have here three unknown field amplitudes.42) The continuity conditions for the transverse fields E and B can then be stated as k k Ei + Er = Et + Em. We notice that the current J is continuous across the front boundary if we assume that the electrons are created with zero net velocity: J = v = 0 for x = 0.46) Noting that kr = ki . (6. ki β (6. km 1 =− . and that B = (k c/ω )E . (6. β2 This can be solved as Er 1+β 1−w = Ei 1−β 1+w Et 2 1+β = Ei 1 + βw 1 + w Em 2β 2 (1 − w) .48) (6.45) x =0− x =0+ This new continuity condition can be rewritten as ki Bi + kr Br = kt Bt + km Bm .

The difference between these two results can better be understood if we calculate the energy of the reflected wave. = Wi 1−β ωr (6. and assuming total reflection (w = 0).23. we can obtain from equations (6. we can write the fraction of reflected energy as Wr ωr W r ωr E r = = Wi ω Wi ω Ei 2 .53) This means that.52) Wr 1−β ωr = = . we have a significant gain in the energy of the reflected wave. which means that all the incident photons are reflected. 6.23).51) . Wi 1+β ω (6. considered in the previous section). (6. but with a different frequency. In contrast. Nr = Ni .132 Full wave theory This result has to be compared with equations (6. and assuming once again that total reflection is taking place (w = 0). (6. when the magnetic mode is present (in the case of an ionization front. taken here to be equal to 1) the ¯ product of the wave frequency with the number of photons: W = ωN .49) This means that.50) Using the Lorentz transformation for the the frequencies. for the number of photons associated with both the incident and the reflected wave Ni = W Wi = i.49) we can also see that the number of photons is conserved. We can then write. ωr ω (6. Because the number N is a relativistic invariant. we get from equation (6. From equation (6. considered in this section). In order to perform this calculation we first notice that the energy of an electromagnetic wave is (apart from the constant h . we get ω = which leads to Wr 1 − β Er = Wi 1 + β Ei 2 ω ωr = γ (1 + β) γ (1 − β) (6.54) . the energy will be Lorentz transformed in a way that is the reverse of the Lorentz transformation for the frequency. ω ω Nr = Wr W = r.48) Wr ω 1+β = . for a counter-propagating ionization front (β < 0). for both cases (absence and presence of the magnetic mode).52) In the absence of the magnetic mode (in the case of a moving plasma. which are valid in the absence of the magnetic mode.

and time-varying currents will eventually become a source of electromagnetic radiation. To be specific. and the resulting space. in some respects. (6. B0 (r ) = 0. and we could be talking about a photon freezing or photon condensation effect. in this case. In the next section we will consider a situation which. (6.59) Here. we now have a significant decrease in the energy of the reflected wave. and that the electrostatic field is perpendicular to this velocity: E 0 = E 0 e y . as long as the atoms or molecules of the gas stay in the neutral state. . described by E 0 (r ) = E 0 cos(k0 x) = E0 2 eik0 x + c.4 Dark source Let us assume a static electric field. the static electric field will accelerate the free electrons created by the front. and the electron velocity v is determined by the linearized equation of motion ∂v e =− E 0 (r ) + E . for β < 0. From equation (6. ∂t m (6. applied to a region of space containing a neutral gas.49) we can also see that the number of photons is not conserved upon reflection.c. This field is determined by the wave equation .55) The missing photons are. E represents the radiation field. u = −uex . No net current is produced by this field. but in the negative direction. (6. if an ionization front propagates across that region. 6. can be considered as the reverse of this photon freezing. even if it is a total reflection: Nr = Ni ω ωr 2 . But.58) where the Heaviside function H (x + ut) represents the sharp boundary between the neutral and the plasma regions.57) Assuming that the ion motion is irrelevant to the fast processes associated with high-frequency wave radiation. converted into a static magnetic field (zero energy photons).Dark source 133 This means that. we can write the electron current in the plasma region as J = −en 0 H (x + ut)v (6.56) Let us also assume that the ionization front moves with velocity u along the x-axis. we first assume a sinusoidal electrostatic field.

ˆ 2ωu c2 2 (6.58) and assuming that the electrons are created at the plasma boundary x = ut with no kinetic energy.62) In order to solve this equation. t) = E ω (x) e−iωt .67) Here we have used the unit polarization vector a = E ω /|E ω |. ω x u (6. m If we want to calculate the field radiated in the plasma region x > ut and propagating in the forward direction. x). we can write from these two equations 2 ωp ∂2 1 ∂2 − 2 2− 2 ∂x2 c ∂t c E= 2 ωp c2 H (x + ut) E 0 (r⊥ .66) and a source term amplitude A where A= 1 ωp ( E 0 · a).64) The time integral in the source term can easily be solved and we are led to the following equation: ∂2 ω ω E + k 2 E ω = iA exp i k0 − x + exp −i k0 + x 2 ω u u ∂x with a wavenumber k determined by 2 k 2 c 2 = ω 2 − ωp (6.63) 2π The one-dimensional wave equation becomes 2 ωp ∂2 1 E ω + 2 (ω2 − ω2 ) E ω = 2 E 0 (x) ∂x2 c c H (x + ut) eiωt dt. (6.134 Full wave theory 1 ∂2 E ∂J = µ0 . we obtain ∇2 E − ∂J ∂v = − en 0 uδ(x + ut)v − en 0 H (x + ut) ∂t ∂t 2n e 0 = (6.68) + exp −i k0 + . along the x-axis and in the positive direction.60) ∂t c2 ∂t 2 Taking the time derivative of equation (6. (6. is E ω (x) = A ikx e 2k x −∞ e−ikx exp i k0 − ω x u dx .61) H (x + ut) E 0 (r ) + E . for forward propagation.65). let us introduce a time Fourier transformation defined by dω E(x. (6. The solution ˆ of equation (6. (6.65) (6.

72) with β = u/c and γ 2 = (1 − β 2 )−1 .71) u The explicit result is  ω = k0 uγ 2 1 − β 1− 2 ωp 2 γ 2 k0 u 2   (6. which corresponds to a large gamma factor or a small plasma density.69) with the allowed wavenumbers determined by k=− ω ± k0 .66): ω 2 2 2 k0 − c = ω 2 − ωp . will contain at most two spectral components: E ω (x) = π ω ω A eikx δ k − k0 + + δ k + k0 + k u u (6. Equation (6. observed in the plasma side but very far away from the ionization front (x → ∞).70) We can see that only the positive sign is allowed here because we are considering forward propagation (k > 0).72) then reduces to ω 2 ωp k0 u + . are accelerated into high frequencies by the relativistic ionization front and propagate along the plasma medium in the forward direction. for a long length scale of the static electric field E 0 (x). which means. k0 u The result described by the present model can be seen as a special case of photon acceleration: (virtual) photons initially with zero frequency. This expression also implies that we have an upper limit for the frequency of the radiation produced in the plasma side: ω < k0 u.73) This result is very interesting because it shows that a large value of the radiated field frequency ω can be obtained even with very small values of k0 . (6. The value of ω can be obtained by using this equation in the dispersion relation (6. This result was first obtained by Mori et al [79]. if we use 2 2 ωp . u (6.56). we obtain a radiation frequency ω ωp /2k0 u ωp . The interest of our present approach is that we did not use any Lorentz 2 2 transformation on the front frame. and associated with the static electric field (6. 2 2k0 u (6.Dark source 135 The asymptotic value of the radiated field. who used a different approach. . Let us consider the case where ωp γ 2 k0 u 2 .

(6.76) is now much larger. by replacing the static periodic electric field by an arbitrary electrostatic field: dq E 0 (r ) = E q (r⊥ ) eiq x . However. typically by a factor of γ 2 . we can write the asymptotic solution. If we restrict again our attention to the one-dimensional problem. u . Using this in the above condition. (6. ω2 u 2 (6. this leads to k0 u ω= . Such a configuration can be named a dark source [79] because it transforms a static field into very high frequency photons propagating in both the plasma and the vacuum region. (6. However. periodically spaced along the x-axis with periodicity 2π/k0 . in order to create the static periodic electric field defined in equation (6. here the waves propagate in a vacuum and have to satisfy the vacuum dispersion relation ω = kc.56). very high frequency radiation fields can eventually be emitted. and propagating in the backward direction.79) 2π .76) The two possible spectral components are k= (6. valid far away from the ionization front.69): |E ω |2 ∼ π A k 2 = π2 4k 2 ωp c 4 |E 0 |2 .74) Let us now consider the field radiated into the neutral gas side.78) 1−β This means that. as will be illustrated in the following pages. Let us then generalize the above procedure. this field is determined by the wave equation ∂2 1 ∂2 − 2 2 ∂x2 c ∂t E= 2 ωp c2 H (x + ut) E 0 (x). for x → −∞. other configurations can also be imagined.77) In contrast with the previous case where the resulting radiation was propagating in the plasma side. for relativistic ionization fronts such that (1−β) ∼ γ 2 1. equation (6. for x < ut.136 Full wave theory The conversion efficiency of the process is stated by the radiation field solution. (6.75) Following the above procedure. the total energy associated with this new wave is much smaller than the energy of the lower frequency emitted into the plasma side. as E ω (x) π ω ω A e−ikx δ k + k0 − + δ k − k0 − k u u ω ± k0 . The apparatus necessary to produce this kind of radiation process can be based on an array of capacitors. based on the same principle. because the value of k appearing in equation (6.

u (6. in both the forward and the backward direction. u c2 The general solution for radiation in the forward direction is then given by 2 Aq = E ω (x) = 1 ikix e 2k ∞ −∞ dx dq Aq e−i(k−q+ω/u)x . and q0 is determined by q0 = k + ω .79) by using an infinite Fourier spectrum with amplitudes E q (r⊥ ) = 2π E 0 e−iq x0 . if the resulting broad spectrum of radiation was subsequently compressed by some appropriate optical system. This could. (6. 137 (6.80) The one-dimensional wave equation for the time Fourier component of the radiated field. could be interesting for the production of ultra-short (or even sub-cycle) radiation pulses.81) 2 ω 2π ∂x c with ω ωp ˆ (6. around some point x = x0 .Dark source The previous case obviously corresponds to E q (r⊥ ) = π E 0 [δ(q − k0 ) + δ(q + k0 )] . for instance.86) This can be described by equation (6.84) where A0 = Aq for q = q0 . for x → ∞. .82) Eq · a . (6. Such a configuration. different (but very similar in its principle) from the original idea of a dark source using a capacitor array as described above.83) Asymptotically. is ∂2 dq 1 2 E + 2 (ω2 − ωp )E ω = i Aq ei(q−ω/u)x (6.85) A similar result could equally well be obtained for radiation emitted in the backward direction and propagating in the vacuum region. valid in the plasma region for this case of an arbitrary electrostatic field. we could generate an infinitely large spectrum of radiation. we get E ω (x) = π A0 eikx k (6. be obtained by a capacitor of infinitesimal width (two pointlike electrodes) located at that position: E 0 (r ) = E 0 δ(x − x0 ). 2π (6. let us consider the case of an electric field acting on a very small region of space. As an example of an application of this more general formulation of the problem.87) In this case.

We can. assume that the density of the background neutral gas medium is modulated in space.88) . Exploring further this idea of producing high-frequency radiation out of a sharp ionization front moving with relativistic velocity. (b) a pointlike E-field. The resulting electron current in the plasma region left behind the front is now described by J = −en 0 [1 + cos(k0 x)]H (x + ut)v (6. for instance. by producing a sound wave. But now. When the ionization front moves across the gas. Such a modulation can be obtained. This means that we do not need to consider any capacitor system acting on the gas medium: E 0 (r ) = 0. we can even imagine a situation where the static electric field is absent. some other ingredient has to be introduced in order to replace this field. a space modulation of the electron plasma density will take place.2. or by forcing the gas to flow through a parallel grid before entering the interaction zone.138 Full wave theory (a) + + + + + + + moving front plasma - - - - - - - Capacitor array (b) + moving front plasma Point-like E field Figure 6. for instance. Examples of dark sources: (a) a periodic E-field produced by a capacitor array.

In this case. But now the electron velocity v is determined by equation (6.60). A radiation field E can eventually result from this configuration.Dark source 139 where is the amplitude of the density modulation and k0 determines the period of the space modulation. as determined by the wave equation (6.61). Instead of equation (6. which shows unstable solutions inside some region of the space of parameters.59) with E 0 (r ) = 0. . a dark source radiation instability will occur. A brief discussion of the Mathieu equation is postponed to the end of the next chapter. we will be led to a Mathieu-type wave equation.

and El (r ) are the cavity eigenmodes depending on three quantum numbers labelled by l. In order to be more specific. which can be ionized by some external agent (for instance. It was initially developed in references [67. Moreover.1 Linear mode coupling theory Let us consider an electromagnetic cavity with metalic walls containing a neutral gas.Chapter 7 Non-stationary processes in a cavity In this new chapter devoted to the full wave theory of photon acceleration. 100]. 7. This is formally analogous to the elementary quantum theory of collisions and it has the merit of reducing the space–time-varying evolution problem to a purely time-varying problem.2) where el (t) are time-dependent amplitudes. such that E(r . If we 140 . t) = l el (t)E l (r ) (7. In quite general conditions. we can describe the electric field associated with the electromagnetic modes contained in the cavity by the following equation: ∇2 − 1 ∂2 c2 ∂t 2 E = µ0 ∂J .1) The electric field has to satisfy certain boundary conditions and its value will be determined by the source term which is described by the electric current J . by an intense laser pulse or by a high voltage applied to adequate electric probes). we will explore a simple theoretical model for mode coupling inside an electromagnetic cavity. The boundary conditions will imply that the general solution of this equation can take the form of a superposition of eigenmodes. let us assume the simplest possible example of a rectangular cavity with the length of the three sides given by L x . L y and L z . ∂t (7. this cavity model can also be adequately applied to several experimental configurations.

5) Let us now turn to the electric current appearing in equation (7. n Lx Elz (r ) = 0. this source term can be written as a function of the cavity electric field E(r . n and p are integers and.8) E(t ) dt + vi0 . is straightforward. p). The generalization of the present discussion to another type of cavity. and take the limit t → 0. (7.4) where we have introduced the cavity volume V = L x L y L z . vi0 = 0. This choice for E0 allows us to write the orthonormality condition El (r ) · El (r ) dr = δll . In order to guarantee that the eigenmodes are normed and orthogonal we define the constant E0 as E0 = 8 V 1+ m2 L 2 y n2 L 2 x −1/2 (7.6) i The density n i of each of these electron populations is determined by ni = ∂n ∂t t ti (7. If we make some simple and plausible assumptions for the gas ionization process. and the inclusion of transverse magnetic modes. we have l ≡ (m.7).7) where t is the elementary time interval around ti during which these populations are created. The velocity vi of the electrons created at the successive instants ti is determined by e t vi (t) = − (7.1). corresponding to the electrons created at different times: J = −e n i vi . in this case. t) [7].Linear mode coupling theory 141 restrict our discussion to transverse electric modes. m ti We can replace this integral inside equation (7. V (7. n. The result is t e2 t ∂n J= dt dt E(t ).9) m −∞ ∂t t t . (7.3) The quantities m. the fields El (r ) are determined by Elx (r ) = E0 cos(mπ x/L x ) sin(nπ y/L y ) sin( pπ z/L z ). m Ly Ely (r ) = E0 sin(mπ x/L x ) cos(nπ y/L y ) sin( pπ z/L z ). assume that the electrons are created initially with zero kinetic energy. We first notice that the plasma created out of the neutral gas can be seen as containing an infinity of electronic species. (7.

after using the orthonormality condition (7. t)El (r ) · El (r ) dr . Cll (t)el (t) l (7. (7. (7.12) This allows us to write the field equation (7. we obtain ∂2 + kl2 c2 el (t) = − ∂t 2 where we have used 2 kl2 ≡ kmnp = (mπ/L x )2 + (nπ/L y )2 + ( pπ/L z )2 .13) We can now replace the mode decomposition (7. and therefore the equation reduces to ∂J e2 = ∂t m = t −∞ ∂n ∂t E(t ) dt t t −∞ e2 n(t ) E(t ) m − t −∞ n(t ) ∂E dt .10) The second term in this expression is obviously equal to zero. The mode coupling coefficients Cll are determined by Cll (t) = V 2 ωp (r . this expression reduces to ∂J = ∂t 2 0 ωp (t) E(t). we get ∂J e2 = ∂t m t −∞ ∂n ∂t E(t ) dt + t ∂n ∂t t t t E(t ) dt .2) in this equation and. (7.14) (7.16) The mode coupling equation (7.5).15) This equation describes the time evolution of the mode amplitudes due to the linear mode coupling induced by the ionization of the gas inside the cavity. ∂t (7.1) in the following closed form: ∇2 − 1 ∂2 c2 ∂t 2 E= 2 ωp c2 E. ∂t 2 l =l (7.142 Non-stationary processes in a cavity If we take the time derivative of the current.14) can also be written in the following form: ∂2 + ωl2 (t) el (t) = − Cll (t)el (t).17) .11) If the electric field varies much faster than the electron density (which implies once again the existence of two timescales). (7.

t)|El (r )|2 dr .23) This means that the plasma is formed on a timescale of 1/γ .19) The mode coupling equation (7.18) Here.21) 2 where ωl2 (t) = kl2 c2 + ωp (t). the self-coupling coefficient (l = l) corresponds to a kind of averaged plasma frequency. Assuming that the plasma frequency ωp changes over a slow timescale. we can integrate this equation and obtain the following WKB solution: el (t) = El kl c/ωl (t) exp −i t ωl (t ) dt ) . taken over the mode space configuration Cll = 2 ωp (r .22) can now be written as t ωl (t ) dt = kl c 1+ 1 2 ωp0 kl c 2 t+ 1 2γ ωp0 kl c 2 e−γ t . and can be used for a detailed description of the spectral changes inside the electromagnetic cavity. 143 (7. let us use the following explicit law for the plasma creation inside the cavity: 2 2 ωp (t) = ωp0 1 − e−γ t . from equation (7.17) is the basic equation of the present model. t) ≡ ωp (t). (7. In order to be more specific. we have 2 Cll (t) = ωp (t)δll . 7. (7. The integral in equation (7. (7.20) This means that the cavity eigenmodes are decoupled from each other and that they evolve according to the equation ∂ 2 el + ωl2 (t)el = 0 ∂t 2 (7. (7.22) Here El is a constant amplitude.2 Flash ionization in a cavity The simplest possible model of plasma formation inside a cavity is to admit that 2 2 ionization is uniform over the entire volume of the cavity: ωp (r . Its physical content and its relevance to our problem will be discussed in the next few sections. v (7.16). In this case.24) .Flash ionization in a cavity Here we have used the time-dependent mode frequency ωl2 (t) = kl2 c2 + Cll (t).

26) where ωl (∞) is the asymptotic value of the mode eigenfrequency for very large 2 times: ωl2 (∞) = kl2 c2 + ωp0 . This is illustrated in figure 7. we should not forget that an eigenmode in a cavity is equivalent to two travelling waves propagating in opposite directions.17) and try a WKB solution of the form (7. in this short time limit. In general. (7. the cavity eigenmode keeps oscillating at the same frequency. we have el (t) El kl c/ωl (∞) exp(−iωl (∞)t) (7. a result which coincides with that of the simple theory of flash ionization outlined in previous chapters: the frequency shift of a given eigenmode of an electromagnetic cavity. Let us then return to the mode coupling equation (7. but suffers a small phase shift. t).3). where the factor sin( pπ z/L z ) can be decomposed into the two factors exp(±ik z z). in order to compare this new approach with the previous one. Taking the second time derivative of this new solution we get ∂ 2 el ∂t 2 −ωl2 el − 2i ωl kl c ∂ El exp −i ∂t t ωl (t ) dt . = −i ∂t l =l The new mode coupling coefficients are determined by Cll (t) Bll (t) = √ exp −i 2 ωl (t)ωl (t) t (7. 2 and we have to retain the space dependence of ωp (r . This can be seen from equations (7. we obtain an equation for the slowly varying amplitude. the plasma creation will not be completely uniform over the entire cavity volume. On the other hand. is simply determined by the asymptotic value of the plasma frequency.29) . as we will see next. but where El represents a time-dependent amplitude. (7.22). this view of flash ionization is still too simple and we have to refine it. in the form ∂ El Bll (t)El . for long times.25) This means that. We can see that this WKB solution gives. However. representing propagation along the z-axis in opposite directions. A more realistic model for flash ionization inside a cavity will have to contain the description of some spatial structure. We simply get el (t) El exp −ikl ct − i 2 ωp0 2γ kl c .1.28) [ωl (t ) − ωl (t )] dt .27) Using this in the mode coupling equation. where a plasma is uniformly created by some external agent. with k z = pπ/L z .144 Non-stationary processes in a cavity Let us see what happens during the first stage of the plasma formation when we have γ t 1. such that γ t 1. (7. In the opposite limit of very long times.

33) Using the eigenmodes (7. or any others. for any other mode l = l0 .Flash ionization in a cavity 145 plasma Figure 7.30) This equation allows us to calculate explicitly. the mode coupling efficiency.31) and (7.3). one containing the time dependence and the other containing the spatial mode coupling. Using this parametric approximation. . According to their definition. The explicit calculation can be performed by using the following model for non-uniform ionization inside the cavity: 2 2 ωp (r . this quantity can then be explicitly evaluated. the amplitude of this mode can still be considered as approximately constant for times t ≥ 0. we obtain 2 Cll (t) = ωp0 (1 − e−γ t )Ill (7. If the coupling is not too strong.32) where Ill = t e−a(r −r0 ) El (r ) · El (r ) dr . two new aspects relevant to flash ionization can be clearly identified [67]. Using equations (7. and in a simple way. 2 (7. From our analysis. the amplitudes are El (t) = −iEl0 t Bll (t ) dt . t) = ωp0 (1 − e−γ t ) exp[−a(r − r0 )2 ]. Non-uniform plasma formation inside a cavity.16). the mode coupling coefficients Cll (t) can be split into two factors.27) that. (7.1. we can write from equation (7. Let us examine the simple but physically meaningful case where a given mode l = l0 pre-exists in the cavity for t < 0.31) Here 1/γ is the timescale for plasma formation and a −1/2 is the dimension of the ionized region around some point r0 . (7. and all the other modes are absent before the plasma formation.

(7.35) where the integral I pp (t) is determined by I pp (t) = 0 Lz sin( pπ z/L z ) sin( p π z/L z ) f (z + vt) dz. First. the ionization processes inside a cavity not only produce a shift in the frequency of a pre-existing mode but also give rise to a large spectrum of radiation associated with the amplitudes el (t).3 Ionization front in a cavity Let us now turn to the problem of an ionization front travelling across the cavity. These two qualitative aspects are well confirmed by experimental results in microwave cavities. and starts its motion at the position z = L z .146 Non-stationary processes in a cavity plasma Figure 7. Moving ionization front inside a cavity. Second. the frequency shift associated with each mode is not simply determined by the asymptotic value of the plasma frequency ωp0 but by the selfcoupling coefficient Cll < ωp0 defined by equation (7. the coupling coefficients (7.34) where f (z + vt) describes the form of the front. we can write 2 2 ωp (r . 2 2 as illustrated in figure 7. In the simplest but physically realistic case of a front which is uniform in the planes z = const. t) will be of the general form ωp (r − vt).2. This means that the spectrum generated by flash ionization inside a cavity is broader. which is a kind of plasma frequency averaged over the eigenmode spatial field distribution.36) . In this case. where the observed frequency shifts 2 were always smaller than the expected ωl = kl2 c2 + ωp0 . due to linear mode coupling. Then ωp (r . and the frequency shift of the pre-excited mode is smaller.16) can be written as Cll (t) = 2 2 ω δmm δnn I pp (t) L z p0 (7. moving along the z-axis with a velocity v.19).2. t) = ωp0 f (z + vt) (7. 7. and a broad spectrum of radiation was observed. for l = l0 . than predicted by single photon dynamics.

Let us also assume that the mode l .and the y-axis with which these two quantum numbers are associated. we obtain Cll (t) = − × 2 ωp0 π δmm δnn (−1) p− p (−1) p+ p sin[(k z − k z )vt] − sin[(k z + k z )vt] .35) that the mode coupling is only associated with the quantum numbers p and p .37) where H (z − a) is the Heaviside function and the particular value a = L z − vt is chosen in order to describe a sharp front starting its motion at (z = L z .38) It can be noticed from equation (7. it is replaced by Cll (t) = − 2 ωp0 2 pπ {2k z + sin[2k z L z (1 − vt/L z )]} . (7. (7. This is due to the fact that the plasma frequency is constant along the x. is still the dominant mode for subsequent times (which implies that the mode coupling is not very efficient). t = 0) and moving across the cavity along the z-axis towards the point z = 0 with a negative velocity. such that we can use the approximate expression el (t) E e−iωl t . the integral (7. For our qualitative discussion it is appropriate to neglect the slow variation of the frequency eigenmode.40) In order to discuss the physical meaning of this result. Using this expression in equation (7.39) p− p p+ p This expression is valid for p = p . This means that the mode coupling will only occur along the direction of the electron density gradient. For this particular case.36) can be easily calculated and gives I pp (t) = − 1 Lz sin ( p − p )π(1 − vt/L z ) 2π p − p Lz 1 + sin ( p + p )π(1 − vt/L z ) . pre-existing in the cavity for times t < 0.Ionization front in a cavity 147 In order to have an order of magnitude estimate of the amplitude of the cavity modes excited by the ionization front. leaving the other two quantum numbers m and n invariant. (7. let us go back to the evolution equation (7.41) .35) and noting that k z = pπ/L z and k z = p π/L z . we can use the simple model for the front form f (z + vt) = H (z + vt − L z ) (7. 2π p + p (7.17). In the case of p = p .

Let us come back to the mode coupling equation and retain one of the four resonant terms. The four resonance conditions therefore have a very simple explanation: the pre-existing cavity mode would correspond in free space to two travelling waves with the same frequency. we would simply have k z = ωl /c and k z = ωl /c.44) where E depends on the coupling with the pre-existing mode l . Each of these travelling waves interacts with the front and.45) we obtain the following saturation amplitude for the resonant mode p the completed journey of the ionization front across the cavity: El (sat) 2 ωp0 E L z . but propagating in opposite directions.46) We can see that a larger value for the front velocity v corresponds to a larger value for the quantum number p compatible with the resonance condition. after (7. (7. and not inside a cavity. Using el (t) = El (t) e−iωl t (7. from its partial reflection. This qualitative discussion suggests that the resonant mode with a lower frequency shift will have a larger saturation amplitude than the resonant mode with a larger frequency shift. 4π pωl v p . We can then write d2 + ωl2 el = E e−iωl t dt 2 (7. for which the mode amplitudes el are resonantly excited.42) If we were in a free space configuration. But. and to a smaller saturation amplitude. apart from the resonant modes which are more strongly coupled with the pre-existing mode in the cavity.148 Non-stationary processes in a cavity We can then see that only four distinct forced terms exist. which means that this resonance condition would reduce to the well-known expression for the relativistic mirror: ωl = ωl 1±β 1±β (7.42). and ωl is given by one of the possible four choices of equation (7. The only and significant difference with respect to the relativistic mirror effect is that here we generally have partial and not total reflection. and these terms verify the resonance condition ωl = ωl ± [(k z ± k z )v]t. many other modes can also be excited by . two new waves result. This eventually explains the absence of this mode in the first experiments of ionization fronts in a microwave cavity [95]. propagating in opposite directions and with distinct frequencies.43) where β = v/c.

Time variation of the mode coupling coefficient and of amplitudes of the lowest order excited modes.4 Electron beam in a cavity Instead of considering ionization processes inside a cavity. t (7. First of all.47) ∂t where A is the vector potential.1) can also be written as ∂2 c2 ∇ 2 − 2 A = − J (7.13).3 for a numerical example. leading to an electromagnetic energy cascade over a large spectral width. as shown by numerical integration of the above mode coupling equations [100].47) leads to our previous field equation (7. the current associated with an electron beam cannot be .12). The mode coupling due to an electron beam is formally analogous to that due to an ionization front. 7. which can be rewritten as J =− e2 m t −∞ ∂n ∂t [ A(t) − A(t )] dt . the two processes also present very distinctive features.Electron beam in a cavity 149 |E2 | C ln |E3 | |E4 | t Figure 7. it should be noticed that the field equation (7. However.48) Using this expression in equation (7. See figure 7. and qualitatively confirmed by experiments. we can also obtain a similar moving boundary if an electron beam is sent across the cavity.3. linear mode coupling. However. In our discussion of the flash ionization and the ionization front we used equation (7.

48) were derived for electrons created at specific times and positions.13).47). by a factor of ωl /ωl . Because the spatial eigenfunctions for the electric field and for the vector potential are identical. (7. than that for the ionization front . Again. and will always move with a non-negligible velocity. for instance.150 Non-stationary processes in a cavity described by the same source current because equations (7. we realize that the saturation amplitude is now larger. we obtain c2 ∇ 2 − ∂2 ∂t 2 2 A = ωp (r . the energy content of each mode is different because the vector potential amplitudes are related to the electric field amplitude by el (t) = al (t)ωl . but with a different physical meaning. (7. But.53) 4π pωl v where A is the amplitude of the source term and where we have assumed that al (t) Al exp(−iωl t). with a velocity equal to zero. 7.50) This is formally identical to equation (7. This means that. (7. t) = al (t)El (r ). with the electric field amplitudes replaced by vector potential amplitudes. the saturation amplitude obtained after the completed journey of the electron beam front across the cavity is now determined by 2 ωp0 A L z Al (sat) (7. this is formally identical to equation (7.12. Using the equation of motion for the beam electrons we can easily obtain the following new expression for the electric current: e2 J = −en(r . The result is that the frequency spectum generated by the electron beam will be identical to that generated by the ionization front. if we rewrite this expression in terms of the electric field amplitudes. the electrons of an electron beam will exist for all times. ∂t 2 l =l (7.46). the mode amplitudes al (t) being related to the vector potential eigenmodes.51) l The evolution equation for the mode amplitudes can now readily be found: ∂2 + ωl2 (t) al (t) = − Cll (t)al (t). because the modes with the same frequency are coupled in the same way.49) m Using it in equation (7. However.17). t)v = − n(r . we can again use a linear expansion for A of the form A(r . t) A. In contrast.52) We see that it is formally identical to equation (7. t) A. but with the electric field replaced by the vector potential. We can then repeat the calculations of the previous section.

Electron beam in a cavity

151

case. Also, as we have seen from the resonance condition for the frequency shift (7.42, 7.43), this factor can be much larger than one. This means that the electron beam is a much more efficient way to up-shift the electromagnetic energy in the frequency domain, not because the frequency up-shift itself is different, but because the amplitudes of the shifted modes are much larger. This can be illustrated by a numerical integration of the coupled mode equations for both cases [100]. Using the same argument we can also conclude that the down-shifted modes will have, in contrast, a much lower amplitude in the case of the electron beam. The physical differences between these two processes of frequency upshifting can be better understood if we consider the total magnetic field generated inside the cavity. We know that the magnetic field is determined by B = ∇ × A and, using equation (7.51), we can write, for the case of the electron beam B(r , t > L z /v) =
l

[∇ × El (r )]Al exp(−iωl t).

(7.54)

This means that the total magnetic field is just the sum of the magnetic fields associated with each of the modes excited in the cavity. A different situation occurs for the case of an ionization front. If we write the total magnetic field in terms of the total electric field, we get B(r , t) = −∇ ×
0 t

E(r , t) = −
l

[∇ × El (r )]
0

t

el (t ) dt .

(7.55)

The asymptotic value for this magnetic field will then be B(r , t > L z /v) = × δll + iωl
0 l L z /v

1 [∇ × El (r )] iωl el (t ) dt − El e−iφl + El e−iωl t (7.56)

where we have φl = ωl L z /v. It can be seen from this expression that the last term, which oscillates at the frequencies ωl , corresponds to the sum of the magnetic fields associated with each of the modes excited in the cavity. But, the remaining terms describe a static magnetic field which was excited by the ionization process and which can be identified with the magnetic mode discussed in the previous chapter. This means that, in the case of the ionization front, a large amount of energy is transferred to the static magnetic field, which explains why this process of frequency up-shifting is much less efficient than that of an electron beam where such a static field is absent. Finally, we should keep in mind that other possible non-stationary processes occurring inside the cavity, which do not create new electron populations, are described by the same mode coupling equations as the electron beam case. This is,

152

Non-stationary processes in a cavity

for instance, the situation where, in a pre-existing plasma medium, a space–time variation of the plasma density is produced, associated with an electron plasma wave with relativistic velocity (wakefield). As shown in the above discussion, the up-shifting frequency effect will be as efficient as in the case of the electron beam and no static magnetic field will be excited.

7.5 Fermi acceleration in a cavity
The linear mode coupling formalism explored in this chapter allows us to give a new description of the Fermi acceleration of photons, using a full wave description [101]. Let us then assume that, due to some externally applied perturbation, the neutral gas inside the cavity is ionized in a limited domain near the boundary z = L z , in such a way that the width of the plasma region oscillates in time with a frequency , much smaller than the cavity mode eigenfrequencies ωl . This can be described by the following space–time distribution for the electron plasma frequency:
2 2 ωp (r , t) = ωp0 H [z − L z + A(1 − cos t)].

(7.57)

Here we have used the Heaviside function H (z − a). This simple law describes a plasma with constant plasma frequency ωp0 but oscillating in a region of the cavity between L z and L z − 2A. The coupling coefficients (7.39) now take the form, for p = p , Cll = − − where
2 ωp0

π

δmm δnn

1 sin[π( p − p )(1 − (1 − cos t))] p− p (7.58)

1 sin[π( p + p )(1 − (1 − cos t))] p+ p

= A/L z . For p = p , equation (7.41) becomes Cll =
2 ωp0 (1 − cos

t) −

2 ωp0

2 pπ

sin[2 pπ (1 − cos t)].

(7.59)

Using this explicit expression in the mode coupling equations (7.17) we can calculate the time evolution of the eigenmode field amplitudes. Let us start by neglecting the mode coupling and let us study the evolution of a single mode el (t), when according to equation (7.59), the effective value of plasma frequency is modulated by a much lower frequency . Notice that mode coupling would be absent if the plasma frequency were oscillating uniformly over the entire cavity volume. In this case, we can write the single mode evolution equation as d2 2 el (t) + kl2 c2 + ωp0 (1 − cos t) dt 2

Fermi acceleration in a cavity
+ 2 pπ sin[2 pπ (1 − cos t)] el (t) = 0.

153 (7.60) t/2, and to

Here it is appropriate to use a dimensionless time variable τ = introduce two parameters: δ= 4 (k 2 2 l +
2 ωp0 ),

γ0 = 2

2 ωp0 2

.

(7.61)

Equation (7.60) can then be reduced to a more familiar form: d2 γ0 e (τ ) + δ − 2γ0 cos(2τ ) − 2 sin[2 pπ (1 − cos(2τ ))] el (τ ) = 0. 2 l 2 pπ dτ (7.62) For high-frequency cavity modes such that 2 pπ 1 the last term can be neglected and this reduces to the well-known Mathieu equation. In order to illustrate the main physical processes in the cavity let us first restrict our discussion to this equation. It is well known [81] that the Mathieu equation has bounded solutions, as well as unbounded ones, depending on the values of the two parameters δ and γ0 . The existence of unbounded (or unstable) solutions is physically very interesting because this means that it is possible to excite cavity modes, starting from the noise level up to arbitrary amplitudes. The energy source responsible for the creation of this electromagnetic field, nearly out of nothing, can only be identified with the external source producing the plasma oscillations. The stability diagram of the Mathieu equation shows that the most favourable modes to be excited verify the resonance condition kl c m /2, where m is some positive integer. However, because we are assuming that is much smaller than the eigenmode frequencies ωl , this resonance condition implies that m has to be quite large. Also, in this case, we known from the properties of the Mathieu equation that the unstable domain in the parameter space (δ, γ0 ) is extremely narrow and very difficult to experimentally satisfy. This means that, in principle, generation of electromagnetic energy inside an oscillating and empty cavity is possible, but it is very unlikely to be observed, due to the fact that it will only occur inside very narrow regions of the parameter space. We now turn to the stable regions of the parameter space, where it is possible to derive approximate analytical expressions for the time evolution of the amplitude of a given uncoupled mode el (t), and to describe the entire frequency spectrum associated with this single mode. Due to the frequency modulations of the self-coupling coefficient (7.59), the frequency spectrum of a single mode no longer corresponds to a well-defined frequency ωl , but to an infinite number of frequencies. This can be clearly seen if we assume that the mode amplitude el (t) is described by el (t) = elk exp[−i(ωl + k )t] ¯ (7.63)
k

154

Non-stationary processes in a cavity

2 where we have used ωl2 = kl2 c2 + ωp0 , and have assumed that the amplitudes ¯ ¯ elk (t) vary over a timescale much larger than 1/ωl . Using this solution in the single mode evolution equation (7.62), and retaining only the terms with the same fast timescale, we can easily derive an equation for the new amplitudes elk (t):

d ¯ ¯ elk + [ωl2 − (ωl + k )2 ]elk dt 2 ωp0 2 − ωp0 (el,k−1 − el,k+1 ) + i Jm (2 pπ ) 2i 4 pπ m − 2i(ωl + k ) ¯ × im e2i pπ el,k−m − i−m e−2i pπ el,k+m = 0. (7.64)

In order to obtain this expression we have used the expansion of the sine term of equation (7.62) in Bessel functions of integer order Jm . We can simplify it considerably if we notice that, for ωl , the second term (proportional ¯ to elk ) can be neglected. On the other hand, if the oscillation parameter is small and if the quantum number p is large, in such a way that the inequality 2 pπ 1 is satisfied, we can also neglect the term containing the sum over the Bessel functions. With these two simplifying assumptions, we can reduce equation (7.64) to
2 ωp0 d (el,k−1 − el,k+1 ) elk = dt 4ωk

(7.65)

where we have used ωk = ωl + k . ¯ This system of coupled equations for the amplitudes elk can be easily integrated by noting that it is formally identical to the well-known recurrence relation for Bessel functions Jk [84]. This means that it satisfies the following solution: elk (t) = Jk
2 ωp0

2ωk

t .

(7.66)

Let us now consider the opposite limit of small and large p, such that 2 pπ 1. In this case, the sum term dominates over the term proportional to in equation (7.64). Moreover, we also have J1 (2 pπ ) Jm (2 pπ ), for m = 1, which means that we can also neglect all the terms in the series expansion except those for m = ±1. The resulting equation is ωp0 d J1 (2 pπ ) cos(2 pπ )(el,k−1 + el,k+1 ). elk = i dt 4 pπωk
2

(7.67)

Here again, the integration becomes possible because of the formal analogy with the recurrence relation between the Bessel functions with purely imaginary

The solution can now be written as elk (t) = (−1)k i−k Jk 2 ωp0 155 2 pπ ωk cos(2 pπ )J1 (2 pπ )t .69) where El can be assumed nearly constant. We can see from here that the modes which are excited with a larger amplitude satisfy the nearly resonant condition ωl (t) ωl = ωl ± m ¯ ¯ (7. in the two opposite limits of the parameter (2 pπ ). (7. 7.73) . it is not possible to integrate the resulting mode equations. For these nearly resonant modes. Once again.71) where m is an integer.72) − Jm (( p − p )π ) −iω t e ¯l .We can write the amplitude el of the dominant mode in the form ¯ el (t) = El e−iωl t (7. This effect of energy spreading over the frequency domain is even more dramatic when we include the linear mode coupling associated with the coefficients Cll established above. First of all. p− p This equation can easily be integrated if we neglect the slow time evolution of the eigenfrequency ωl ωl .70) where ωl2 (t) and Cll (t) are determined by equations (7. we can write the following evolution equation: 2 ωp0 d2 + ωl2 (t) el (t) = ± iEl 2π dt 2 Jm (( p + p )π ) p+ p (7. The amplitudes of all the other modes.58). an infinitely large spectrum of frequencies with decreasing amplitudes can be generated out of a single mode contained in a variable cavity. unless we introduce some simplifying assumptions. let us assume that the cavity contains a dominant mode characterized by the index l . can then be described by d2 ¯ + ωl2 (t) el (t) = −Cll El e−iωl dt 2 t (7.68) We clearly see from these solutions that. such that coupling occurs mainly between this and the other cavity modes.18.Fermi acceleration in a cavity argument Ik . This means that Fermi acceleration results in a considerable line broadening. l = l . and assume that ¯ ¯ el (t) = El e−iωl t (7.

This is in qualitative agreement with our previous discussion of the Fermi photon acceleration using the Hamiltonian approach. we conclude that Fermi acceleration inside an oscillating cavity provides an efficient mechanism for generating a broad spectrum of electromagnetic radiation. we have p p . the solution for short times is given by El (t) = ∓El 2 ωp0 4π ωl ¯ Jm (( p + p )π ) Jm (( p − p )π ) t. If we complement this with the tendency of each mode considered individually to suffer a considerable resonant broadening. Of course. In this approximation.74). in addition to the resonance condition (7. as shown by the above discussion of the single mode evolution.71). . for low modulation frequencies ωl . in the same way as stochasticity in classical particle motion is replaced by a non-stochastic description based on the wave equation for a quantum particle. these resonance conditions are satisfied by several modes. with an amplitude approximately given by equation (7. − p+ p p− p (7. Here the particle stochastic behaviour is replaced by the mode coupling. In the limit (ωl+1 − ωl ) all the modes in the cavity will be nearly resonant.74) This result shows that the modes that are more easily excited are those for which.156 Non-stationary processes in a cavity where El (t) is the slow amplitude of the nearly resonant modes. where it was shown that the photon trajectories would become stochastic under certain conditions.

As an introduction to this chapter. we will describe first the well-known quantization procedure of the electromagnetic field [37.1) with D = E and B = µ0 H . 61]. We will then examine the quantum theory of space and time refraction. . ∇·D=0 ∂t ∂D ∇×H = .1. ∇·B=0 ∂t (8.Chapter 8 Quantum theory of photon acceleration In order to complete our theoretical framework. in order to adapt the notation to our specific needs. We give a phenomenological description of the problem of the field quantization in this medium. as ∇×E =− ∂B . non-dispersive and non-dissipative dielectric medium. 157 . it is important to show that photon acceleration is not a result of the classical or semi-classical approximations. with small changes. characterized by a real dielectric constant.1 Quantization of the electromagnetic field 8. This approach will be useful for our problem. We start with Maxwell’s equations which can be written. We will also describe the less well-known procedure for the field quantization in a plasma. by assuming that is a well-known constant. 8. but that it can easily be identified and described in purely quantum grounds. which not only confirms the main features of the classical theory.1 Quantization in a dielectric medium We consider an infinite. given a priori and not explicitly calculated by a microscopic theory. in the absence of charge and current distributions. but also identifies intrinsic quantum processes such as the possibility of photon creation from the vacuum.

the integration extends over both the negative and the positive values of the three components of the wavevector k. defined by the condition ∇ · A = 0. t) = 2 Ak ei(k·r −ωk t) (8. (2π)3 B(r . t) dk (2π )3 (8.7) (2π )3 where ωk = (kc/n). c2 ∂t 2 (8. t) = with Ak (r .2). such that E =− ∂A − ∇φ.8) The corresponding electric and magnetic fields are. But we know that. k This allows us to rewrite the above general solution as A(r .3) In this case. E(r . φ and A.158 Quantum theory of photon acceleration We define the scalar and the vector potentials. the general solution of this equation can be written as dk A(r . (8. according to equations (8. in classical theory.9) Ak (r . t) = E k (r . t) dk (2π )3 (8.6) We know that.2) It is convenient to choose the Coulomb gauge. (8. we can derive from equations (8. t) = Bk (r . and ∇2 A − n2 ∂ 2 A = 0. we have to assume that A−k = A∗ .10) .4) (8. k (8. t) dk .1) the following equations for the scalar and vector potentials: ∇ 2φ = 0 ∇2 A − n2 ∂2 A c2 ∂t 2 = n2 c2 ∇ ∂φ ∂t (8. In this expansion. This obviously implies that φ = 0. t) = Ak ei(k·r −ωk t) + A∗ e−i(k·r −ωk t) . ∂t B = ∇ × A.5) √ with n = / 0. in order to guarantee that the electromagnetic fields are real quantities.

14) We know that the amplitudes of the vector potential are. (8. t)|2 + 1 2 | Bk (r .11) (8.19) ˆ 2 where the position and the momentum operators satisfy the commutation relations [q. ˆ ˆ ˆˆ ˆˆ ¯ (8. pk = iωk (Ak − A∗ ). we can write Ak = 1 (ωk qk + i pk ). 2 2 (8.Quantization of the electromagnetic field with E k (r . by using Ak = Ak ek . 4 ωk (8. This classical oscillator can be quantized by defining a Hamiltonian ˆ operator H .13) (8. where ek is the unit polarization vector.15) k k Here we have considered scalar quantities.20) . t) = ik × Ak ei(k·r −ωk t) − A∗ e−i(k·r −ωk t) .17) This is formally identical to the energy of a one-dimensional oscillator with unit mass. 2 k (8. From these two quantities. k Let us now consider the total electromagnetic energy density W = with Wk = 1 2 | E k (r .14). p] ≡ q p − pq = ih . by defining √ √ qk = (Ak + A∗ ). in general. p) = p2 q2 + ω2 . such that 1 ˆ H = ( p 2 + ω2 q 2 ) ˆ (8. position q and momentum p: W ≡ H (q.12) (8. t)|2 = 2 ωk ( Ak · A∗ ). k µ0 Wk dk (2π )3 159 (8. t) = − iωk Ak ei(k·r −ωk t) − A∗ e−i(k·r −ωk t) k Bk (r .18) The first term represents the kinetic energy and the second term the parabolic potential. we obtain for the energy density of the mode k Wk = 1 2 2 2 ( p + ωk qk ). frequency ω.16) Using this in equation (8. complex quantities and it is useful to consider their real and imaginary parts separately.

17. ˆ ˆ 2ω p = −i ˆ hω ¯ (a − a + ).22) From this and from equations (8.24) we can also write the Hamiltonian operator in the form 1 ˆ ˆ H = hω +N (8.27) Returning to the electromagnetic field.2 a(k. ˆ ˆ (8. p]) = hω 2h ω ¯ ¯ aa+ = ˆˆ 1 hω ¯ 1 ˆ H + hω .28) At this point we should notice that the electromagnetic radiation has two independent polarization vectors. a + ] = a a + − a + a = 1. 8.29) .24) This leads to the following commutation relation: [a.23) We also have (8.25) From equations (8. 8.26) ¯ 2 ˆ where N is the quantum number operator ˆ N = a + a. ˆ ˆ 2 (8. (8. ˆ 2 ωk k k (8. ˆ ˆ 2h ω ¯ (8. λ)e(k.21) This is equivalent to a=√ ˆ 1 (ωq + i p). such that the field amplitude operators for each mode are ˆ ˆ+ established by the equivalence Ak → h ¯ ak ek . ¯ 2 1 ˆ H − hω . we see that we can associate with each mode k a one-dimensional quantum oscillator with destruction and creation operators ak and ak .160 Quantum theory of photon acceleration The Hamiltonian operator can also be written in terms of the destruction and creation operators. such that ˆ ˆ q= ˆ h ¯ (a + a + ). In order to take them both into account we change the operators in the following way: ak ek → ˆ λ=1. ˆ 2 ωk A∗ → k h + ∗ ¯ a e . a and a + . λ). we have a+a = ˆ ˆ 1 1 ˆ ˆ ˆ ˆ (ω2 q 2 + p 2 + iω[q.20). ¯ 2 (8. ˆ ˆ 2h ω ¯ a+ = √ ˆ 1 (ωq − i p).19. ˆ ˆ ˆˆ ˆ ˆ (8.

t) = i λ=1. λ) e−i(k·r −ωk t) . λ)e∗ (k. λ)e(k. λ) ei(k·r −ωk t) 2 ωk (8. λ)k × e(k.2 dk (2π)3 h ωk ¯ a(k. t) = i λ=1. 2) e(k. λ)e(k. we are now dealing with a dispersive medium. In contrast with the previous case. λ) e−i(k·r −ωk t) .Quantization of the electromagnetic field 161 From now on.1.34) −a + (k. we will remove the hat in the operator notation. . λ)e∗ (k. 8. These field operators are strictly equivalent to those obtained for propagation in a vacuum. homogeneous and unmagnetized plasma. it is more appropriate to describe this medium as a vacuum plus charge and current distributions. (8.35) −a + (k. 1) × e(k. From equations (8. Instead of using the plasma dielectric function. The two unit polarization vectors are orthogonal: e(k. λ)k × e∗ (k. λ ) = δλλ . 2) = k/|k|. λ) ei(k·r −ωk t) 2 ωk (8.32) dk (2π)3 h ¯ a(k.30) An obvious choice of these two orthogonal polarization vectors is two linear polarizations in the plane perpendicular to the direction of wave propagation: k · e(k.2) we can also establish the electric and the magnetic field operators: E(r . t) = λ=1. which are the . λ) · e(k.2 dk (2π)3 h ¯ a(k. λ) ei(k·r −ωk t) 2 (8.2 (8. The total vector potential operator will then be given by A(r .33) +a + (k. λ) e−i(k·r −ωk t) and B(r . Maxwell’s equations. the only difference being the replacement of the vacuum permitivity 0 by the appropriate dielectric constant of the medium.2 Quantization in a plasma We consider an infinite. λ) = 0 (λ = 1.31) (8.

37) with B = µ0 H . ∇·B=0 ∂t ∇×E= − (8. defined by the condition (8. can then be written in the following form: ∂B ρ . ρ = −e(n − n 0 ).162 Quantum theory of photon acceleration starting point of our quantization procedure. The .39. The inclusion of these effects would require a more refined and purely microscopic quantization procedure.40) √ We define here the electron thermal velocity by the quantity ve = T /m.36) (8. for instance.42) (8. with a mean density n 0 . 8.39) (8. ∇·E= ∂t 0 ∂E ∇×H =J+ 0 .41) Let us introduce the scalar and vector potentials φ and A. we get ∂n ˜ + n0∇ · v = 0 ∂t ∂v e S2 = − E − e ∇n. Apart from the fluid approach.38) The electron density and mean velocity are described by the continuity and the momentum conservation equations: ∂n + ∇ · nv = 0 ∂t ∂v e 2 + v · ∇ v = − ( E + v × B) − Se ∇ ln n. Using n = n 0 + n. ∂t m n0 The linearized electron current and charge densities are J = −en 0 v. 2 2 where T is the plasma temperature. By treating the plasma electrons as a fluid we are neglecting the resonant particle effects which can. we will also restrict our discussion to low field intensities. where |n| ˜ ˜ n 0 is the density perturbation. Assuming that the plasma ions are at rest. we can write the charge and current densities as J = −en v. ˜ (8. and retain the Coulomb gauge. (8.2).3). and use Se = 3ve .40) around the equilibrium state. by using equations (8. lead to purely kinetic effects such as the electron Landau damping. This allows us to linearize the fluid equations (8. ρ = −en. ∂t m (8.

which means that the electromagnetic field in a plasma will not be exactly equivalent (but it will be similar) to a massive vector field. 8.Quantization of the electromagnetic field 163 corresponding potential equations can be derived from equations (8.44). A). ¯ 2 However. 2 m ∂t 2 ∂t This allows us to write ∂2 2 2 ˜ − Se ∇ 2 n = − ωp n ˜ ∂t 2 e v⊥ = A m which.51) √ We notice here that. from the electron fluid equations we can easily derive ∂2 en 0 2 2 ˜ − Se ∇ 2 n = − ∇ φ 2 m ∂t ∂ 2 v⊥ e ∂2 A = .50) c2 (8. according to equations (8.43.46) (8. implies that ∇2 − ∇2 − 1 ∂2 2 Se ∂t 2 1 ∂2 c2 ∂t 2 φ= A= 2 ωp 2 Se 2 ωp (8. determined by the condition ∇ · v⊥ = 0.47) (8. (8. (8.37) and take the form e ∇ 2φ = n ˜ (8. in the form ∇2 − 1 ∂2 c2 ∂t 2 Aν = m eff c ν A h ¯ (8.43) 0 ∇2 − 1 c2 ∂t 2 ∂2 A= en 0 v⊥ c2 0 (8. it is obvious that we have Se = c2 for general plasma conditions.44) where v⊥ is the transverse part of the electron velocity.52) where m eff = ωp h /c2 would be the mass of the vector field. such that Se = c2 . these two equations would reduce to a single Klein– Gordon equation for the four-vector potential Aν ≡ (φ/c. In order to prepare for the field . if the electron thermal velocity could be made equal to 2 c/ 3.45) On the other hand.49) φ A.48) (8.

50. according to equations (8. λ). (8. 8. (8. the first term corresponds to longitudinal (or electrostatic) oscillations. t) = iωk Ak (r . λ)e∗ (k. and the second term to the transverse electromagnetic waves.60) . such that we have. The result is the appearance of a factor of 2 in the first term of this equation. the frequencies for the vector potential (the first equation) are determined by the dispersion relation ωk = 2 k 2 c 2 + ωp (8. For convenience. t) dk . But it should be noticed that. t) − ikφk (r . λ). Ak → h ¯ a(k. 2).164 Quantum theory of photon acceleration quantization procedure we can write the general solution of equations (8.58) Let us now consider the total energy density wk = 1 2 0 | E k (r .57) The corresponding electric and magnetic fields are still represented by equations (8. for the transverse modes (λ = 1. 8. t) = φ(r . in the electron plasma oscillations.50.10).12) is replaced by E k (r . t) = ik × Ak (r .55) These two equations for the vector and the scalar potentials are formally identical to each other. t) = with and Ak (r . λ) and a + (k. t) = φk ei(k·r −ωk t) + φk e−i(k·r −ωk t) . Field quantization is obtained by introducing destruction and creation operators a(k.56) and for the scalar potential (the second equation) by ωk = 2 2 k 2 Se + ωp .54) (8. 8. t)|2 + wpart (k) µ0 (8. t) dk (2π )3 (8. t)| 2 + 1 | Bk (r . Bk (r . (2π)3 φ(r .53) Ak (r .51). λ)e(k.2. but now equation (8. λ) 2 0 ωk (8. for each of the three distinct modes. (8. t). 2 0 ωk A∗ → k h ¯ a + (k.59) ∗ 2 = 2 k 2 φk φk + 2ωk Ak · A∗ . we have added to the purely electromagetic energy the kinetic energy of the particles associated with the oscillations of the scalar potential. t). It is well known from plasma theory that. the averaged energy is equally divided between the electrostatic field energy and the kinetic energy of the plasma electrons. t) = Ak ei(k·r −ωk t) + A∗ e−i(k·r −ωk t) k ∗ φk (r . k Here.51) in the form A(r .

Time refraction and.2 dk (2π)3 h ¯ a(k.61) The total vector potential operator will then be determined by A(r . λ). t) = λ=1. t) = dk (2π)3 h ¯ a(k. The first two modes correspond to the two transverse photons and the third mode corresponds to plasmons (or longitudinal photons). The total scalar potential operator is φ(r . λ) + ¯ 2 (2π)3 ωk (λ) = 2 k 2 c2 (λ) + ωp (8. φk → h ¯ 2 0k a(k. and c2 (λ = 3) = Se . λ)e(k.1 Operator transformations Let us consider a time discontinuity in an infinite dielectric medium such that. at time t = 0. 2) = c2 .62) +a + (k. the dielectric constant suddenly changes from a value 1 to a new value 2 . λ)e∗ (k. but with two distinct characteristic velocities: for the transverse particles it is the speed of light c. λ).66) . Here we find the three independent polarization states of a massive vector field (which has spin one) [88].2 Time refraction 8.63) 2 0k The resulting total energy operator will be given by 3 W = λ=1 dk 1 h ωk (λ) a + (k.2. λ)a(k. ∗ φk → 165 h ¯ 2 0k a + (k. and for the plasmons it is the thermal velocity Se . for the longitudinal mode (λ = 3). 3) ei(k·r −ωk t) + a + (k.65) 2 with c2 (λ = 1. The similarities of the electromagnetic field quantization in a plasma with the quantization of a massive vector field was noticed long ago by Anderson [5]. (8. (8. 8. 3) e−i(k·r −ωk t) . It has served as a phenomenological model for the theory of the Higgs boson [88]. This transformation law can be described by the expression (t) = 1 H (−t) + 2 H (t) (8.64) such that (8. λ) ei(k·r −ωk t) 2 0 ωk (8. λ) e−i(k·r −ωk t) .

75) . t) e−ik·r . j (8. (8.166 Quantum theory of photon acceleration where H (t) is the Heaviside function. t) eik·r − a + (k. t) eik·r − a + (k. (8. n2 (8. and for t > 0. t) = k × E(k. t = 0+ ). in these + expressions. t) = a(k) e−iωk t . These operators are different from each other because the meaning of a photon (or of an elementary excitation of the field) changes with the refractive index at t = 0.71) B(k. ωk (8.68) We also know that the displacement vector and the magnetic field operators can be determined by D(k.34) that. or λ = 2). t) = E(k. t) eik·r − a + (k. t). t). t) = i(k × ek ) h ¯ 2ω j j For t < 0 we use the index j = 1. to the + old ones. This means that we can write D(k. t = 0+ ). B(r .72) Noting that these equalities are independent of r . and introduced timedependent destruction and creation operators a(k. a1 and a1 . a + (k.74) √ ω1 1 ω2 2 Let us define the parameter α= ω1 ω2 1 2 = n2 n1 1 2 = n1 . we can easily reduce them to the following relations between the new and the old operators: √ √ + + ω1 1 [a1 (k) − a1 (−k)] = ω2 2 [a2 (k) − a2 (−k)] (8. In order to obtain such a relation we use the continuity conditions for the fields D(r . t) e−ik·r ek .69) a j (k. for a given polarization state of the photons in the dielectric medium (λ = 1.73) 1 1 + + [a1 (k) + a1 (−k)] = √ [a2 (k) + a2 (−k)]. 2 (8. t) = a + (k) eiωk t . t = 0− ) = D(r .70) (8. t) = i h ωk ¯ a(k. Our main problem is to relate the new operators a2 and a2 . (8. t) e−ik·r j a j (k. t = 0− ) = B(r . We know from equation (8.67) ∗ Here we have used a real polarization vector ek = ek . we use j = 2. t) = iek h ¯ ωj 2 j B(k. the electric field operator can be written as E(r .

74) can be rewritten as + + α[a1 (k) − a1 (−k)] = a2 (k) − a2 (−k) + + a1 (k) + a1 (−k) = α[a2 (k) + a2 (−k)]. or Fock states. with a given wavevector k. with wavevectors k and −k. when we go through a time discontinuity. n k = 0.78) The reciprocal relations can also be obtained: + a2 (k) = Aa1 (k) + Ba1 (−k) + + a2 (−k) = Aa1 (−k) + Ba1 (k).76) By adding and subtracting these two equations. (8.79) This shows that each field mode existing for t < 0. (8. 167 (8.2 Symmetric Fock states It is particularly important to look at the behaviour of the number states. 2α B= 1 − α2 . can be derived by applying n k times the creation operator a + (k) to the vacuum eigenstate |n k = 0 ≡ |0k .81) The ground state. It is well known (and can easily be derived from the above equations) that the eigenvectors |n k .80) The eigenvalues n k of the number operator are the occupation numbers. or the number of photons in the mode k. we obtain + a1 (k) = Aa2 (k) − Ba2 (−k) + + a1 (−k) = Aa2 (−k) − Ba2 (k). of the effect of time reflection obtained with the classical theory of chapter 6. defined by A= 1 + α2 . 2α (8. is called the vacuum state. For normalized eigenvectors we can write 1 |n k = √ [a + (k)]n k |0k . will be coupled to two modes existing for t > 0.Time refraction Equations (8. 8. at the quantum level. ¯ 2 2 (8. They are defined as the eigenstates of the number operator Nk ≡ a + (k)a(k): a + (k)a(k)|n k = n k |n k . (8.2. corresponding to an occupation number equal to zero. Such a coupling provides an explanation. The energy eigenvalues are n k |Hk |n k = h ωk n k | a + (k)a(k) + ¯ 1 1 |n k = h ωk n k + .77) Here we have used the real coefficients A and B. n! (8. corresponding to an arbitrary excited state.82) .

and j = 2 to the Fock states valid for t > 0. (8. (8. m 2 .85) can then be represented as |0. n −k j = |n k j |n −k j . before and after the time discontinuity.168 Quantum theory of photon acceleration We have shown that. (8.m Cm. 0 1 = C0 ∞ m=0 B A m |m. 0 1 and |0. we can then define these symmetric Fock states in terms of the symmetric vacuum: |n. the modes k and −k are coupled to each other.84) This expression means that.89) . we obtain a1 (k)|0. n j ≡ |n k . 0 1 = m. j j (8. 0 1 in terms of the final state vectors |0. n j =√ 1 n!n ! [a + (k)]n [a + (−k)]n |0.85).87) This leads to the following recurrence relation for the coefficients Cm : Cm = B Cm−1 = A B A m C0 . m + 1 2 √ = (ACm+1 − BCm ) m + 1|m. (8. m + 1 2 . It is then useful to introduce the following symmetric state vectors: |n.m = Cm δmm . simply because the vacuum is a state of zero total momentum.77). m 2 . Using equation (8. if we want to establish a relation between some initial symmetric state |n. m 2 √ √ = Cm A m|m − 1. This implies that Cm.83) The index j = 1 pertains to the Fock states valid for t < 0. 0 j . m 2 − B m + 1 |m.71). m 2. we have to establish a relation between the symmetric vacuum states. (8. n 1 and some final symmetric state |m. 0 2 .86) If we apply the destruction operator a1 (k) to equation (8. and use equation (8.88) The initial symmetric vacuum states (8.85) It is obvious that such a development has to be perfectly symmetric with respect to the modes k and −k. 0 1 ≡0= m + Cm Aa2 (k) − Ba2 (−k) |m. in the process of time discontinuity.m |m. This can be achieved by representing the initial vacuum states |0. |0. (8.

.5. Probability of photon creation from the vacuum. and assuming that θ0 = 0. from an initial vacuum state described by the vector states |0. a number of 2m photons which will appear in symmetric pairs of modes. the probability of finding such a symmetric Fock state with 2m photons at times t > 0. 0 1 . According to the above equation. m 2 . 1 − (B/A)2 (8.90) This means that C0 = eiθ0 1 − (B/A)2 (8. we can use the normalization condition in order to determine the remaining coefficient: 0. 0|0. for B/A = 0. m|0. Using equation (8. 0|m.1.93) (m) p m Figure 8.92) This result shows that a time boundary at t = 0 will be able to generate.85) as |0. m 2 = 1− B A 2 B A 2m . will be given by p(m) =2 m. 0 11 0.89). k and −k. we can write the final expression for the symmetric vacuum decomposition (8.Time refraction 169 Now. 0 1 = 1 − (B/A)2 ∞ m=0 B A m |m. (8. (8. 0 = |C0 |2 ∞ m=0 1 1 B A 2m = |C0 |2 1 = 1.91) where θ0 is an arbitrary phase.

The possible generation of photons out of the vacuum inside an optical cavity with a time-dependent dielectric was recently considered in reference [19].3 Probability for time reflection The above results on the transformation of an initial vacuum by a time discontinuity can be generalized to an arbitrary initial state |φ1 ≡ |n.170 Quantum theory of photon acceleration In order to have a more precise idea of the physical consequences of this result.77) and. noting that the operators a2 (k) and a2 (−k) commute.84. m 2. such that n 2 = n 1 (1 + δ). m ) =2 m. which means that n = 0. 0 1 . let us assume the following plausible situation of a very small time perturbation of the refractive index of the medium. we obtain. 0 1 n! ∞ 1 =√ 1 − (B/A)2 n! m=0 B A m + [a1 (k)]n |m. (8. we obtain |n.95) This gives an order of magnitude estimate for the probability of creation of photons out of the vacuum due to a sudden change of the refractive index. which leads to δ 2 2m . 8. (8. n 1 . m |φ1 φ1 |m.96) In order to remain as close as possible to our previous discussion of the classical model for time reflection. (8. 8. (8. The initial photon state is |φ1 = |n. after a binomial expansion + [a1 (k)]n = An n r =0 n! (−1)n−r (n − r )!r ! B A r + [a2 (k)]n−r [a2 (−k)]r . 0 1 1 + = √ [a1 (k)]n |0.92).98) .2. (8.97) + We can now use equation (8. This corresponds to an initial photon beam propagating along k for t < 0. m 2 . with δ 1: δ n1 α= 1− . where n = 0 photons pre-exist in the mode k and no photons at all propagate in the opposite direction.94) n2 2 It means that we have (B/A) p(m) (δ/2). not necessarily perfectly symmetric (n = n ). This probability decreases with the number of photons m according to a power law. The probability of observing a given final state |φ2 ≡ |m. we will concentrate on initial completely asymmetric states. Using equations (8. m 2 is determined by p(m.

101) This result shows that. s) = 0 of observing a number s > 0 of photons propagating in the opposite direction −k. It is well known that these particular quantum states have the advantage of tending to a classical field in the limit of large occupation numbers.104) |n + s. s 2 . s 2 : n 2 p(n. The above discussion can easily be generalized to include the case of coherent states. we arrive at n 171 |n. This gives us the quantum explanation for the effect of time reflection already described with the classical theory of chapter 6.103) √ |n. in general. define an initial coherent state as ∞ αn 2 |α.99) r =0 s=0 where we have used the new index s = m − r . s) =2 n + s. n! (8.99) we can write this state vector in terms of the final states ∞ n α n bsr (n) 2 |α. s 2 = r =0 bsr (n) .Time refraction Using this in equation (8. 0 1 . √ n! n. there is a probability p(n. 0|n + s. But there is also a finite probability p(n. and the coefficients 1 n! bsr (n) = √ An 1 − (B/A)2 (−1)n−r (n − r )!r ! n! B A s+2r . 0 1 = ∞ bsr (n)|n + s.100) From this we can calculate the probability for observing at a time t > 0 a given photon state |n + s. a coherent state for a given field mode k is |α = e−|α| 2 /2 ∞ n=0 αn √ |n k .97). 0 1 = e−|α| /2 (8. 0 2 with the same number of photons propagating with the same wavevector (but with a shifted frequency).102) This means that we can. an initial coherent state will not lead to time-transmitted and time-reflected coherent states. 0) = 1 of observing a state |n. This means that the time discontinuity does not preserve the classical-like properties of the initial . 0 11 n. (8. n! n=0 Using equation (8. By definition. in our symmetric representation. s 2 (8.m=0 r =0 This expression shows that. s|n. 0 1 = e−|α| /2 (8. after the occurrence of a sudden time change at t = 0 of the refractive index of an infinite dielectric medium. (8.

The first one was already observed in the classical description and corresponds to the frequency shift ω = ω2 −ω1 of the n k photons initially existing in the medium. for very large values of the initial occupation number n ≡ n k the transmitted state will be very similar to the initial one and close to a similar coherent state. the expectation value of this operator is W 1 = 1 n. Let us first consider the total energy operator W = wk dk = (2π)3 h ωk a + (k)a(k) + ¯ 1 dk . which is also associated with the time discontinuity. Furthermore. However. it is also obvious that.108) We see that the energy is not conserved for two distinct reasons. s 2 .4 Conservation relations We conclude the quantum theory of time refraction by briefly discussing the energy and momentum conservation relations. with a finite probability p(n. s).2. propagating in a similar direction but with a shifted frequency. The second . 0 = h ω1 ¯ + 1 n k |a (k)a(k)|n k + 0|a + (−k)a(−k)|0 + 1 (8. This means that the energy variation introduced by the time discontinuity of the medium on the electromagnetic spectrum is W = W 2 − W 1 = h ωn k + h ω2 2s. the non-conservation of the classical-like properties of the initial states has to be related to the creation of squeezed states (states with no classical counterpart). s = h ω2 ¯ + 2 n + s|a (k)a(k)|n + s + s|a + (−k)a(−k)|s + 1 (8. 2 (2π )3 (8.172 Quantum theory of photon acceleration photon state. 8. This problem is not directly related to photon acceleration and will not be discussed here. This problems would deserve a detailed numerical study [74].107) = h ω2 (n k + 2s + 1) ¯ with ω2 = |k|c/n 2 .105) For an initial state |n. ¯ ¯ (8. The expectation value for the energy operator for t > 0 will then be W 2 = 2 n + s.106) = h ω1 (n k + 1) ¯ with ω1 = |k|c/n 1 . s|W |n + s. 0|W |n. 0 1 . We have seen that the time discontinuity generates final states |n + s.

we assume that the media. 2 (2π )3 (8.110) P = P 2 This means that the total momentum is conserved. we can write for the associated electric field operator. 8. t) + E r (x. as expected: − P 1 = 0.112) (8. ¯ (8. t) = i E r (x. or 2) E(x.114) with kt = ωn 2 /c = ki (n 2 /n 1 ).111) where the incident and the reflected field operators are E i (x. t) = i hω ¯ + a1 (ki . t) eikr x − a1 (kr .3 Quantum theory of diffraction Let us consider a sharp boundary between two stationary dielectric media with no dispersion. The result is P 1 = h kn k . t) e−iki x e(ki ) 2 1 hω ¯ + a1 (kr . t) = i hω ¯ + a2 (kt . t) eikt x − a2 (kt . For simplicity. valid in the semi-infinite region x < 0. t) (8.Quantum theory of diffraction 173 one is specific to the quantum nature of light and is due to the generation at t = 0 of oppositely propagating s pairs of photons. and for a given polarization (λ = 1.113) with ki = ωn 1 /c and kr = ki . . we can define the electric field operator associated with the transmitted wave as E t (x. Let us consider next the total momentum operator P= pk dk = (2π)3 h k a + (k)a(k) + ¯ 1 dk . t) e−ikr x e(kr ) 2 1 (8. If photons with frequency ω and initial wavevector ki are propagating in medium 1 and interact with this boundary. and that the propagation is along the x-axis. t) eiki x − a1 (ki . In the second medium (x > 0).109) Using the same kind of approach we can calculate the initial and the final expectation values of this operator. have a boundary at x = 0. ¯ P 2 = h k(n + s − s). with dielectric constants equal to 1 and 2 . t) = E i (x. t) e−ikt x e(kt ) 2 2 (8.

119) At this point we notice that the time-dependent destruction and creation operators in equations (8. t) = E r (x. ( Bi + Br ) × ex = Bt × ex . 1 + α2 a1 (−ki ) = − 1 − α2 a1 (ki ). 8. t) + Br (0. t) = E t (0. we obtain a1 (ki ) + a1 (−ki ) = α 2 a2 (kt ) a1 (ki ) − a1 (−ki ) = a2 (kt ) with α= From this we obtain a2 (kt ) = 2 a1 (ki ). (8. t)[ex × e(kt )]. t)[ex × e(kr )] ω Bi (x. we obtain E i (0.123) n1 = n2 1 2 1/4 (8. t)[ex × e(ki )] ω kr Br (x. t) = Bt (0. (8. for the operators ( E i + E r ) × ex = E t × ex . t) + E r (0. t) = (8. (8. 1 + α2 (8. a + (k. This means that the boundary conditions for these operators have to be formally identical to those for the classical fields. that the components of the electric and magnetic fields tangent to the boundary between the two media are continuous. We can then derive . t) = and Bt (x. t) = a + (k) eiωt .116) kt (8.114) are of the form a(k. Bi (0. (8.120) Equating separately the terms with the same time dependence. We can then write. t). ω The quantization of the electromagnetic field is based on the assumption that the field operators satisfy Maxwell’s equations.115) (8.122) These expressions can be seen as the Fresnel formulae relating the incidence.121) . t).174 Quantum theory of photon acceleration The corresponding magnetic field operators are determined by similar expressions: ki E i (x. t) = a(k) e−iωt .113.118) Using equation (8. We know. the transmission and the reflection destruction operators. from classical theory.117) E t (x. We can easily realize that the same expressions are valid for the creation operators.117) and noting that the field is polarized in the perpendicular direction e(ki ) = e(kr ) = e(kt ).

The same result could be obtained by using a different method [94]. we make a Lorentz transformation to the reference frame moving with the dielectric boundary.125) These operator relations are equivalent to equation (8.123). t) + E r (0. reflection and transmission field operators identical to that obtained for the classical fields. On the other hand. It is clear from the above results that we can follow the approach used in the classical theory: first. (8. in contrast with the case of a time boundary. we repeat the above quantization procedure in the moving frame and. =− E r (0. t) 1 + α2 T ≡ E t (0. the expectation value of the incident number operator Ni is equal to the sum of the expectation values of the reflected and transmitted photons: n i = n r + n t . because the photons maintain their frequency upon reflection and refraction. and they are formally identical to the Fresnel formulae for the classical fields. the total energy is also conserved: W = (Wr + Wt ) − Wi = h ω(n r + n t ) − h ωn i = 0. t). we obtain + Ni = [a1 (−ki ) + a2 (kt )][a1 (−ki ) + a2 (kt )] + + = a1 (−ki )a1 (−ki ) + a2 (kt )a2 (kt ) = Nr + Nt . Second. We therefore get E i (0. we make an inverse Lorentz transformation back to the rest frame. Let us start with the number operator associated with the incident photon states: Ni ≡ a + (ki )a(ki ).124) This is nothing but a different version of equation (8. The result cannot be other than a relation between the incidence. From this.Quantum theory of diffraction 175 from (8. t) = 1 E t (0. ¯ ¯ α (8. by multiplying them by exp(−iωt). t) 1 − α2 . for a generic quantum state of the radiation field. t) − E r (0. E i (0. ¯ ¯ (8.119). the total momentum is not conserved.123) a more familiar version of the Fresnel formulae. t) 2α 2 . α2 (8. in agreement with the results of the classical theory: P = (Pr + Pt ) − Pi = h (kr n r + kt n t ) − h ki n i ¯ ¯ 1 = h (ki + kt )n t = h ki 1 + 2 n t . t). . it is interesting to consider the energy and momentum conservation relations. = E i (0. finally.126) We see from here that. For comparison with the time reflection case. and by adding their Hermitian conjugates. t) = E t (0.127) We see that photon creation from a vacuum around a space boundary cannot exist. In addition. By using equations (8. we get R≡ E i (0.123) and noting that a2 (kt ) and a1 (−ki ) commute. t) 1 + α2 (8.128) Let us conclude with a comment on the more general case of a moving space boundary.

2 which shows that such an effect is present for a time discontinuity. for a boundary moving with an infinite velocity. because such an effect is already absent in the moving frame (or for a boundary at rest).176 Quantum theory of photon acceleration This leads us to the conclusion that the quantum theory of photon refraction at a moving boundary will not predict the creation of photons from a vacuum. This is in contrast with the quantum theory of section 8. . or equivalently.

Chapter 9 New developments

The theory of photon acceleration, as presented in its various versions in the present book, has recently been extended to new and exciting areas of physics. Two examples are given here. One example is related to the new field of neutrino interactions with dense plasmas [13]. Here, the striking similarities between the photon and the neutrino dispersion relations in a plasma can be explored [68]. A considerable amount of theoretical work has already been performed by several authors, leading to an already quite coherent view of the collective neutrino–plasma interactions [106]. The other example concerns the coupling between photons and the gravitational field. Of particular interest is the attempt to describe photon acceleration processes associated with gravitational waves [72]. Given the extreme difficulty of this subject, these results have to be received with some caution, and can be considered as merely tentative. But, we firmly believe that these attempts can be very positive for the progress of knowledge in this yet quite unexplored area.

9.1 Neutrino–plasma physics
The neutrino interaction with very dense plasmas is of considerable importance in the early universe and during supernova explosions. The mechanisms for photon acceleration in non-stationary media can easily be transposed to neutrino–plasma physics. It is well known that, in the presence of matter, the neutrino effective mass is changed due to the weak-current interaction. In particular, the charged current couples the electron neutrinos with the electrons existing in a dense plasma. In order to describe this weak coupling we can use the dispersion relation of a neutrino in a plasma, relating its momentum p and its energy E [10, 119]: (E − V )2 − p 2 c2 − m 2 c4 = 0 ν (9.1) 177

178

New developments

where m ν is the (eventually existing) neutrino rest mass and V is an equivalent potential energy such that √ 0m 2 2 V = 2G F 2 ωp ≡ gωp . (9.2) e Here G F is the Fermi constant for weak interactions, e and m are the electron charge and mass and 0 is the vacuum permittivity. Because the coupling constant g is very small, we can assume, even for extremely dense plasmas, that the equivalent potential energy V is always much smaller than the total neutrino energy E. We also know that the standard theory for electro-weak interactions assumes a zero value for the neutrino rest mass m ν [88], but some recent observations of solar neutrinos point to the existence of a small rest mass [32]. For that reason, we opt here for retaining the rest mass in the neutrino dispersion relation. This relation can be rewritten in terms of the neutrino frequency ω and of the neutrino wavevector k (using h = 1), for a space- and time-varying plasma, as ¯ ω(r , k, t) =
2 k 2 c2 + m 2 c4 + gωp (r , t). ν

(9.3)

This is indeed very similar to the photon dispersion relation. Here we can also use the ray equations for the neutrino field, which are formally identical to the equations for single photon trajectories. In the presence of electron plasma perturbations moving with a velocity u, the electron plasma frequency is such 2 2 that ωp (r , t) ≡ ωp (η), with η = r − ut. Following the procedure described in chapter 3, we can establish an invariant for the neutrino trajectories in the form I = ω − k · u. From this we conclude that the total energy exchange between the neutrino and the background plasma medium, when it crosses the moving plasma boundary (for instance, during the collapse of an exploding neutron star) is
2 E = gωp0

β 1−β

(9.4)

where β = |u|/c and where we have supposed counter-propagation between the neutrino and the plasma front. It should be noticed that, as before for the photon case, this process of energy exchange between the particle and its background plasma medium is linear and not resonant, which means that all the particles with the same initial energy will exchange the same amount of energy with the medium. This simple approach is based on the classical description of the neutrino, equivalent to the geometric optics approximation for the photon field. This is clearly valid as long as the plasma space and timescales are much larger than those characterizing the neutrino state. But we can do better and use a quantum mechanical description of this nonresonant interaction between the neutrinos and the plasma medium. For that

Neutrino–plasma physics

179

purpose, we derive from the above dispersion relation (9.1) a Klein–Gordon wave equation by making the usual replacement: p → −ih ¯ From this we get h 2 c2 ∇ 2 − m 2 c4 − h 2 ¯ ¯ ν ∂2 ∂t 2 ψ = 2ih V ¯ ∂ψ ∂t (9.6) ∂ , ∂r E → ih ¯ ∂ . ∂t (9.5)

where ψ is the normalized wavefunction associated with the neutrino field. We could improve our quantum description of the neutrino field by replacing this Klein–Gordon equation by a more adequate Dirac equation. But, the present formulation stays valid as long as we neglect the coupling between different helicity states. We are interested in the case where the effective potential V (r , t) is associated with a plasma perturbation moving with a velocity u. If the perturbation is nearly uniform in the plane perpendicular to its velocity, we can choose the x-axis as parallel to u and reduce the wave equation with its one-dimensional version: ∂2 1 ∂2 m 2 c2 − 2 2 − ν2 ∂x2 c ∂t h ¯ ψ(x, t) = 2i ∂ V (x − ut) ψ(x, t). 2 ∂t hc ¯ (9.7)

Let us now use the D’Alembert transformation of variables: ξ = x − ut, We can easily see that ∂2 1 ∂2 − 2 2 ∂x2 c ∂t = (1 − β 2 ) ∂2 ∂2 + 2 ∂ξ 2 ∂η + 2(1 + β)2 ∂2 . ∂η∂ξ (9.9) η = x + ut. (9.8)

When u = c, or β = 1, this reduces to the expression ∂2 1 ∂2 − 2 2 ∂x2 c ∂t =4 ∂2 . ∂η∂ξ (9.10)

When written in terms of the new variables, the wave equation (9.7) becomes (1 − β 2 ) = ∂2 ∂2 + 2 ∂ξ 2 ∂η + 2(1 + β)2 ψ(ξ, η). m 2 c2 ∂2 − ν2 ψ(ξ, η) ∂η∂ξ h ¯ (9.11)

2iβ ∂ ∂ − V (ξ ) ∂η ∂ξ hc ¯

180

New developments

Because the potential V is independent of the variable η, we can introduce a Fourier transformation in that variable, such that ψ(ξ, η) = ψq (ξ ) eiqη dq . 2π (9.12)

This means that we can reduce the wave equation to (1 − β 2 ) ∂ ∂2 m 2 c2 ψq (ξ ) − q 2 + 2iq(1 + β)2 − ν2 ∂ξ ∂ξ 2 h ¯ 2iβ ∂ = V (ξ ) iq − ψq (ξ ). hc ∂ξ ¯

(9.13)

In order to understand the meaning and to test the validity of this form of wave equation, let us first briefly consider the particular case of neutrino motion in a vacuum: V (ξ ) = 0. In this trivial case, we can perform a Fourier transformation in the remaining variable, ξ , of the form ψq (ξ ) = ψqp ei pξ dp . 2π (9.14)

The wave equation then leads to a neutrino dispersion relation of the form (1 − β 2 )( p 2 + q 2 ) + 2(1 + β 2 ) pq + m 2 c2 ν h2 ¯ = 0. (9.15)

This strange expression has to be equivalent to the dispersion relation (9.3), with ωp = 0, which means it is equivalent to ω2 − k 2 c2 − m 2 c4 ν = 0. h ¯ (9.16)

In order to be convinced of such an equivalence, we notice that ∂ ∂ ∂ = + , ∂x ∂ξ ∂η This leads to the two relations k = p + q, ω = u( p − q). (9.18) ∂ ∂ ∂ =u − . ∂t ∂η ∂ξ (9.17)

Using this in equation (9.15), we obtain equation (9.16). It means that this is indeed the neutrino dispersion relation in a vacuum, written in terms of the more complicated quantities p and q. The physical meaning of these new variables is given by (9.18): their sum is the neutrino momentum and their difference is its energy divided by its velocity.

20) hc ¯ For this type of solution to be valid. u (9. apart from confirming the classical results for the energy variation.21). V (ξ ) = 0. the quantities k(ξ ) and ω(ξ ) represent the local values of the wavenumber and frequency for a given neutrino state characterized by the quantity q. would be identical. then we can derive from the invariance of q an energy shift equal to the one given by equation (9. We can now return to the non-trivial case of a slowly varying potential. This quantity remains constant over the entire trajectory. valid only locally.21). According to the same equations (9. in the same way as we have established the photon frequency shift in a non-stationary optical medium. (1 − β)2 (9. Due to the existence of such an invariant (which also appeared in the classical theory). Using equations (9.24) .23) Using the above relativistic mirror relation. 2 pr u = −kr u − ωr −ωr (1 + β).22) Assuming that we nearly have kc = ω.13) of the form ψq (ξ ) = ψq0 exp i ξ p(ξ ) dξ .Neutrino–plasma physics 181 If the neutrino rest mass was made equal to zero. these two quantities. we can try a WKB solution for the wave equation (9. (9. In particular. which is locally valid in ξ : (1−β 2 ) p(ξ )2 + q 2 +2(1+β 2 ) p(ξ )q + m 2 c2 ν h2 ¯ = 2β V (ξ ) [q − p(ξ )] . (9. and if in the other side of the discontinuity at ξ = 0 its value is V0 . But. namely the possibility of quantum reflection at the moving plasma boundary. and which can be written as 2q = k(ξ ) − ω(ξ ) .21) Here. we can easily calculate the change of the neutrino energy when it interacts with a non-stationary plasma background. the quantum description can also reveal a qualitative new effect. Here. we obtain pr pi (1 + β)2 . we get from this the relativistic mirror relation ωr = ωi (1 + β)/(1 − β). we can obtain a dispersion relation. k and ω/u. we can easily obtain a relation between the energy of the incident and the reflected energy states.19) With this. are different: 2 p i u = k i u − ωi −ωi (1 − β). u 2 p(ξ ) = k(ξ ) + ω(ξ ) .18) have to be replaced by similar ones. (9. equations (9. if we assume that the plasma potential tends to zero at infinity. V (ξ → ∞) = 0. E i = h ωi and E r = h ωr : ¯ ¯ −2qu = ki u + ωi = −kr u + ωr . the values of the quantum number p associated with the incident and the reflected wave solution.4). (9. for the same value of q.

This means that we can assume.30) This expression shows that the coupling between the initial and the reflected state is proportional to the velocity of the moving plasma perturbation. If we assume wave function solutions of the form ψq (ξ ) = j=1. (9.29) = w(ξ )ψq1 ∂ξ where the coupling coefficient is determined by w(ξ ) = 2 V (ξ ) i β e 2 p (ξ ) h c (1 − β) 1 ¯ ξ ( p1 − p2 ) dξ .25) where ψq j (ξ ) are slowly varying amplitudes such that ∂ψq j ∂ξ this means that we can write ∂2 ψq j ∂ξ 2 − p 2 ψq j + i p j j ∂ψq j . .20) stays valid for both the incident and the reflected particle state.25) in the wave equation (9. (9.182 New developments The probability amplitudes for the reflected neutrino state can also be derived from the wave equation. that the incident mode ψq1 is dominant. by using a mode coupling approach. and that its amplitude is only slightly perturbed by the interaction with the moving plasma discontinuity. It is also proportional to the Fourier component of the potential V (ξ ) for p = p1 − p2 . and assuming that the local dispersion relation (9.13). we obtain (1 − β 2 ) p j (ξ ) j ∂ψq j i e ∂ξ ξ p j (ξ ) dξ 2β V (ξ ) [q − pl (ξ )] ψql ei hc ¯ l= j ξ pl (ξ ) dξ .2 ψq j (ξ ) ei ξ p j (ξ ) dξ (9. We can then write the evolution equation for the reflected mode in the parametric form ∂ψq2 (9. due to the extremely small value of the Fermi constant G F . even for very dense plasmas.28) We know that the neutrino–plasma interaction is very weak.26) Using the solution (9. in quite general conditions. (9. or equivalently to the value of β. ∂ξ (9.27) p j ψq j .

9. 9.31) We can further explore the analogy between the neutrino and the photon coupling with a non-stationary background plasma by developing a kinetic theory for the neutrino gas present in the medium. (9. Other surprising results arising from the study of this collective neutrino plasma are the generation of magnetic fields and inhomogeneities in the early stages of the universe [105]. We should also keep in mind that the neutrino gas will react back on the plasma electrons. but also because it is known that. and to the possibility of transferring a considerable fraction of the energy of a neutrino beam into the plasma. which has to be added to the electron equations of motion. 85.29) and assuming that the initial value of the reflected state amplitude is zero (or ψq2 (ξ0 ) = 0): ψq2 (ξ ) = ψq1 ξ ξ0 w(ξ ) dξ . at least in some limit. present some analogies with the well-known gravitational frequency shift occurring when photons escape from regions of strong gravitational fields. The result is a neutrino kinetic equation.2 Photons in a gravitational field The large variety of effects leading to a frequency shift (or to a spectral change) of the electromagnetic radiation. This can be described by a neutrino pressure term. not only because it can lead to a deeper and more global view of photon dynamics. The result of this mutual coupling between the neutrino and the electron gas in a plasma leads to the possibility of neutrino Landau damping of relativistic electron plasma oscillations [73]. 126]. which we have classified under the name of photon acceleration. Associated with this pressure term we can likewise define an equivalent electric charge for the neutrino gas [69. by exciting electron plasma waves in the medium [102]. similar to those derived in chapter 4 for the photon gas [112]. Even if the gravitational effects are described by a completely different theoretical approach.Photons in a gravitational field 183 This can be seen by integrating equation (9. similar to the radiation pressure considered before. or neutrino fluid equations. a gravitational field can be adequately described as a dielectric medium in a flat space [55. 83. 113].1 Gravitational redshift We show first that the Hamiltonian ray theory can be easily extended to include the influence of the gravitational fields. and the photon propagation in the presence of a gravitational field (and in the absence of a medium). We can then make the bridge between the photon propagation in a dielectric medium (in the absence of a gravitational field) as described in this book. one can question the possible connections between these gravitational frequency shifts and our concept of photon acceleration.2. This can be done by choosing a convenient . Such a question is relevant.

for an arbitrary but constant gravitational field. We will illustrate our ideas by using a very simple metric tensor such that. (9. (9. we can eventually use a kinetic photon theory in a gravitational field. ∂x0 (9. 2 and 3. where ω is the mean frequency and k the mean wavevector. such that: ki = ∂ψ/∂ x i . corresponding to the space coordinates.36) where we noticed that g 00 = 1/g00 .32) If a wave is propagating in a slowly varying gravitational field we can introduce an eikonal ψ. and a similar change of the magnetic permeability of a vacuum from √ µ0 to µ0 / g00 . If we define the frequency of the electromagnetric wave ω as the derivative of the eikonal function ψ with respect to the time variable ω = −c ∂ψ . This new theoretical approach will then provide an alternative derivation of the gravitational redshift.184 New developments definition for the photon frequency.34) where g ik are the components of the metric tensor. More realistic (but also more complicated) metric tensors will be discussed at the end of this section. We know that. Let us consider an electromagnetic wavepacket characterized by the fourvector k i ≡ (ω/c. the spatial coordinate system is Cartesian at a given point and at a given time: g ik = −δik . .33) If we replace this in the above condition for the absolute value of k i we obtain the eikonal equation [55] g ik ∂ψ ∂ψ =0 ∂xi ∂xk (9. Once this question is clarified. k). the absolute value of this four-vector vanishes: k i ki = 0.34) the following dispersion relation: ω2 = k 2 c2 g00 (9. As concerns the time coordinates. The same result could be obtained by noting [55. in the same way as we did for plasmas and for other dielectric media. we use g 00 = 1. where the indices i and k are supposed to take the values 1. 126] that the influence of a (quasi-static) gravitational field on the electromagnetic wave propagation is equivalent to the change of the dielectric constant of a vacuum from 0 to √ 0 / g00 .35) we obtain from equation (9. as explained below. in a vacuum and in the absence of a gravitational field.

and not a shift in frequency. we can write V (r ) = −G M/r . because the photon frequency ω. If the potential V is due to a single massive object (for instance a star). One photon emitted at the surface of that star will have the initial wavenumber k1 such that ω = k1 c − (k1 /c)G M/R.37) We know that the frequency ω = ω(r . we can also write (9. t) can be used as the photon Hamiltonian for the ray equations written in the canonical form dr ∂ω = dt ∂k dk ∂ω =− dt ∂r dω ∂ω = dt ∂t c+ V k c k k ∂V − . c ∂r k ∂V . This will be discussed in more detail below. This photon will be observed on earth with a wavenumber k2 . which is defined as the derivative of the eikonal with respect to the proper time τ (notice that this is the observer proper time and not the photon proper time): ωτ = − ∂ψ ∂ψ ∂ x 0 ω =− 0 =√ . such that (r → ∞) ω = k2 c.38) For the total time derivative of the Hamiltonian function.41) . as defined by equation (9. we can rewrite this photon dispersion relation as ω = kc 1 + 2 V c2 k kc + V. as defined by equation (9. for a static gravitational field. is a constant of motion. this apparent contradiction with the conventional description of the gravitational redshift disappears if we notice that we are using a different definition for the photon frequency. c ∂t (9. c (9. k.40) This means that the observed wavelength observed on earth will be larger than the one emitted at the star surface ( λ > 0).35). of mass M and radius R. This is the well-known gravitational redshift. We obtain from these two expressions of ω the following change in the wavenumber: k = k2 − k1 = −(k1 /c)G M/R < 0. In the present formulation.Photons in a gravitational field 185 Introducing a scalar potential V . is an invariant. the frequency ω. such that g00 = 1 + 2V /c2 . for r ≥ R.35). we could be led to the conclusion that the gravitational redshift appears only as a shift in wavelength.39) This equation shows that. However. ∂τ g00 ∂ x ∂τ (9. It is well known that the conventional view is based on the photon ‘proper frequency’ ωτ . (9.

We can then replace V (r ) = −G M/r . If a static gravitational field is also present. We propose to use the universal time ω. In order to keep the discussion formally simple. using the above 2 metric transformation.43) with α = 2G M/c2 . as discussed below. In addition.42) Let us again assume that the static gravitational field is related to a star of mass M and radius R. or photon bending near a massive star. c2 (9. It is then useful to study the photon equations of motion in a plasma and in a gravitational field. because it coincides with the Hamiltonian function appearing in the photon canonical equations (9. the gravitational redshift becomes a frequency shift as well. and. The massive astrophysical objects that can produce important light bending are also very often surrounded by dense and warm plasmas.38). For V = 0. we have V = 0. for r > R in this dispersion relation and use it as the appropriate Hamiltonian function for the photon canonical equations ω ≡ ω(r . In this way. .44) where T is the plasma temperature. meaning that ωτ = kc. such as the photon bending in the vicinity of a star or the photon acceleration by a gravitational wavepacket. these equations can easily be extended to other situations and used to describe several other important effects. the plasma density will decay exponentially away from the surface of the star (r > R). according to a Boltzmann distribution. This means that we can rewrite it as 2 2 ωp = ωp0 e−mV (r )/T 2 ωp0 1 − β r (9.2. we obtain ω2 = [k 2 c2 + ωp ]g00 .186 New developments With this local definition of frequency we get ωτ = kc everywhere. we include the plasma dispersive effects but neglect the influence of static magnetic fields. k) = 2 (k 2 c2 + ωp ) 1 − α r (9. or more explicitly ω= 2 (k 2 c2 + ωp ) 1 + 2 V . In general. the dispersion relation for photons in a non-magnetized plasma 2 2 is simply: ωτ ≡ ω2 = k 2 c2 + ωp . β = mG M/T and the approximate expression is valid for a warm plasma such that T mV (r ). where ωp is the electron plasma frequency.2 Gravitational lens Let us first discuss a gravitational lens. 9. instead of the proper time ωτ .

What is important to notice is that this description of photon motion around a massive object generalizes the usual description of gravitational bending for the case of a plasma in a gravitational field. we obtain an approximate expression for the Hamiltonian: ω 2 c kr + L2 a + m 2 c2 − . dt ∂r 187 (9.Photons in a gravitational field The photon equations of motions (9. In the absence of a plasma. The inclusion of plasma effects is quite interesting because stars or other massive objects are usually surrounded by warm and dense plasmas which can eventually influence the total lensing effect. .48) This approximate photon Hamiltonian is formally identical to that of a relativistic particle with mass m eff . moving in a Coulomb field.2. The corresponding trajectories are well known from the textbooks [55] and it is not necessary to state them here. the gravitational redshift and the gravitational lens effects.45) In the second of these equations we have used spherical coordinates r = (r. eff r r2 (9. and moreover. For motion in the plane φ = const. we recover the well-known results concerning light bending by a star.47) Here we have used the photon equivalent mass in a plasma: m eff = ωp0 /c2 .46) where L = const is the angular momentum of the photon trajectory.38) can now be written as dr c2 α = 1− k dt ω r dk ∂ω = − er . from our simple photon equations of motion. that we could easily include the plasma refraction effects. Assuming that along the photon trajectory we always have (α/r ) 1 and neglecting the term in 1/r 2 .We have also introduced the new parameter a = αω 1 − 2 ωp0 ω2 1− β α . 9. we can write k = kr er + (L/r )eθ (9. (9. θ. Notice also that diffraction due to plasma inhomogeneities around a star can dominate over purely gravitational bending if 2 β α. In this case we have a βωp0 /ω. φ).3 Interaction of photons with gravitational waves It is already quite surprising and rewarding that we could rederive.

we can still define a universal frequency and use the dielectric description of gravitation as an acceptably good approximation [126]. we can introduce a canonical transformation from (r . The photon equations of motion become dη ∂ω = dt ∂p dp ∂ω =− . we have to make explicit use of the Einstein equations of general relativity. t) = V0 + V (r − v f t). such an interaction takes place over extremely large distances. Moreover. k) to a new pair of variables (η.188 New developments Let us now consider a much less obvious case. even for very small moving gravitational perturbations. a difficult problem because in order to solve it. as a first-order approximation. We can then say that. p) = ω(η) − v f · p = 2 ( p 2 c 2 + ωp ) 1 + 2 (9. In order to derive explicit results. Equation (9. for a time variation g00 . photon acceleration (or energization) by the gravitational field can eventually occur. let us use the following scalar potential: ˜ V (r . let us consider two different situations. This is due to the nearly resonant character of the photon–gravitational-wave interaction because of their close group and phase velocities. and the same will happen to the proper photon frequency ωτ . (9. propagating in an otherwise flat space–time. which enhances the process of photon acceleration. This means that we can use. such that η = r − v f t and p = k. a flat space where the spatial part of the metric tensor is still Euclidian and the temporal metric component g00 becomes a function of time. p). In the first one we assume that. This is.51) Several physical configurations can be studied with these Hamiltonian formulations and all lead. However. Let us neglect plasma effects and consider the propagation of a photon in a vacuum. dt ∂η The new Hamiltonian associated with these canonical equations is ω (η. In order to illustrate this general feature.49) Here v f is the velocity of the infinitesimal gravitational field perturbation ˜ and V its amplitude. where the static gravitational field of the previous two cases is replaced by a time-dependent gravitational field. to strong photon acceleration.50) ˜ V0 V (η) +2 2 c2 c 1/2 − v f · p. (9. in general. In analogy with several other similar occasions. near some radiating source (for instance . if this time-varying field is due to an infinitesimal gravitational wavepacket.36) shows that the universal photon frequency ω is no longer a constant.

let us assume that θ1 = 0 and that β ∼ 1. which means that the background infrared or visible photons ω1 × 10 . it is more interesting to extract some results from the invariance of ω .50) could be solved numerically.52) Here k f determines the scale of the moving shock front.56) c c 2k1 c2 In order to have an estimate of the total frequency up-shift. (9. c2 1 − β (9.55) where θ1 and θ2 are the initial and final angles between the photon wavevector and the front velocity. the photons will acquire a final frequency ω2 and a final wavevector k2 . However. After crossing the entire shock front. and where we have introduced the following auxiliary parameters: 2 ωp vf V0 β= cos θ1 . their initial frequency ω1 is such that ω = ω1 − v f · k 1 (9.57) ω2 ˜ For a perturbation V /c 10−1 and = 1 − β 10−8 . 2 (9.53) where we assume that the shock front velocity v f and the initial photon wavevector k1 are nearly parallel. (9. We will retain the flat space approximation and describe the shock front in the following simple form of a potential perturbation: ˜ V ˜ V (η) = [1 + tanh(k f · η)]. If the photons are emitted at some point η 0. because we are concerned with order of magnitude estimates. However. this leads to 7 . we obtain ω2 = ω1 1 − β(1 − δ) ˜ 1 − β(cos θ2 / cos θ1 )(1 − δ + V /c2 ) (9. and interacts with the surrounding photons. δ = 2 + 2 . We obtain ω = ω 2 − ω 1 = ω1 ˜ V β . this first example is not quite compatible with our assumption of a small gravitational perturbation travelling in a flat space. The photon equations of motion (9. based on a very rough description.Photons in a gravitational field 189 a collapsing star).54) Equating these two different expressions for the invariant ω . we will keep it here as a first and illustrative attempt to formulate a new problem that will be interesting for a future and more accurate solution. such that ω = ω2 − v f · k 2 . Strictly speaking. This will allow us to discuss the relevant properties of the photon motion. a gravitational shock wave starts to form before being radiated away.

58) The photon frequency shift will occur when it travels from a region of gravitational wave maximum to a wave minimum. As a second example. equal to τ = (2/k f ) seconds. the total frequency shift would be zero. In this case. as discussed in chapter 3.190 New developments existing near the exploding star can be accelerated up to the gamma ray frequency range. leading to a decrease in the local angles θ and an increase in the effective β. the same kind of analysis leads to a maximum frequency shift which is twice the value given by equation (9. This invariant states the conservation of the transverse photon wavevector. ˜ But now we have a much smaller field perturbation (V /c2 1) and. The distance necessary for the photons to travel from a maximum to a minimum of the gravitational wavepacket will be d k −1 × 10−2 parsecs. We should point out that in these estimates we have neglected the focusing effects associated with the difference between the initial and the final angles. This means that the exactly resonant condition between the photons and the gravitational waves.59) θ2 c ˜ For an extremely small perturbation V /c2 10−9 we still obtain photon accelerations up to the gamma ray energies ω2 4ω1 × 107 for a very small angle between the photon wavevectors and the gravitational wavevechtor: θ2 10−8 radians. The result of this two-dimensional effect is to focus the photon trajectories in the direction of the gravitational wave propagation. Notice that the time necessary for the photon to cross the front will be.57). let us consider photons interacting in a vacuum with an infinitesimal gravitational wavepacket. We can then write ˜ 4 V ω ω1 2 2 . θ1 and θ2 . It then enhances the acceleration process (at the expense of a larger interaction distance) and broadens the region of photon phase space with a significant frequency up-shift. This is due to the fact that the photon acceleration process increases the parallel photon wavevector. In contrast with the previous example this is now well inside the domain of validity of the dielectric model for the gravitational field. in this case. (9. while the transverse one is kept constant. we use ωp = 0 and write ˜ ˜ V = V cos(k f · η). Notice that the value of β = 1 is never exactly attained by these trajectories. for which no acceleration exists. However these two angles can easily be related. Otherwise. if k f is given in cm−1 . is an ensemble of zero measure . (9. due to the existence of a second invariant for the photon trajectories. because the photons and the gravitational wave travel almost exactly at the same speed. Because we are considering propagation in a vacuum. we have to assume that the angles θ1 and θ2 are small but not exactly zero. f which is quite small on a cosmological scale. and can be written as k1 sin θ1 = k2 sin θ2 .

a universal frequency cannot be defined. If the vacuum was replaced by a plasma medium.60. This is equivalent to saying that.2. The corresponding metric tensor will have components g00 = (1 + V /c2 ) and gαα = −(1 − V /c2 ). for arbitrary field amplitudes. For instance. by ωp . 2.4 Other metric solutions Let us briefly show how the above description of photon motion in a gravitational field can be improved. we should notice that the general dispersion relation in a vacuum (9. These equations (9. t). (9. 2.34) can be explicitly written as g 00 ω2 − 2g 0α ωkα c + g αβ kα kβ c2 = 0 (9.60) where α = 1. if we have a weak gravitational field. We can then use the explicit form of the Hamiltonian function ω = ω(kα .62) This is not very much different from equation (9. we have V = −2G M/r .61) c c If the field is created by a star with mass M. 3. other time-dependent components of the metric tensor should also be included. and we have used the symmetry g 0α = g α0 . 9. apart from g00 .97) are valid in quite general conditions. 2. which can be physically more accurate than the simple example used above. This will lead to the necessity of using a different dispersion relation. by using more accurate metric tensors. as stated above. in general. the interval is determined by V V ds 2 = gi j dx i dx j = 1 + 2 c2 dt 2 − 1 − 2 dl 2 .97). ω = −ck0 . quite fortunately. The interval can be written in spherical coordinates as ds 2 = A(r )c2 dt 2 − B(r ) dr 2 − r 2 dθ 2 − r 2 sin2 θ dξ 2 where A(r ) and B(r ) are functions of the radial coordinate r . due to the existence of a cut-off frequency.Photons in a gravitational field 191 in the photon phase space and.63) . in the right-hand side of this equation. x α . Let us now consider a Schwarzschild type of metric.42). (9. the wave solutions of the Einstein equations generally imply that. we would have to replace 2 the zero. It should be noted that. is physically irrelevant to this problem. where t = x 0 /c in the photon canonical equations (2. for α = 1. 3. The dispersion relation in a plasma will be given by 2 ω 2 = ωp − k 2 c2 gαα g00 . (9. First of all. which is valid for a non-rotating spherical distribution of mass. Let us give some examples of metric solutions.

We can use f (x. t)(c dt − dx)2 . let us consider the metric solution of a gravitational wavepacket of the form ds 2 = c2 dt 2 − dl 2 + f (x.192 New developments The photon dispersion equation can now be written as ω2 − 2 ωp B(r ) 2 k⊥ = A(r ) 2 A(r ) 2 k + 2 k⊥ B(r ) r r + 2 kξ (9. It can easily be shown that the photon dispersion relation in a vacuum takes the form ω 2 ω 2 (1 − f ) +2f (9. (9. For moving gravitational field perturbations. show that the exchange of energy between the electromagnetic (or photon) field and the gravitational field is possible in a variety of situations. and of the corresponding photon dispersion relations. we obtain ω=− k2 k2 f + ⊥ . we are reduced to the case of pure vacuum ω = k c. we have used the photon wavenumbers parallel and perpendicular to 2 2 2 the direction of the gravitational wave propagation: k = k1 and k⊥ = k2 + k3 . for photon propagation perpendicular to the direction of the gravitational wave. our approach leads to a new and simple derivation of the well-known gravitational redshift and gravitational lens effects.68) k − (1 + f )k 2 − k⊥ = 0.67) (9. as already stated before.66) where a and are the gravitational wave amplitude and frequency. and for parallel photon propagation. As a final example of a photon dispersion relation in a gravitational field. c c Here.69) From here we can see that. k c±c 2 1− f 1− f (1 − f ) (9. the dispersion relation is simply given by ω = k⊥ c/(1 − f ). Solving for ω.65) sin2 θ This expression shows a clear asymmetry between the radial and the perpendicular propagation. t) = a cos(k0 x 0 + k1 x 1 ) = a cos(q x − t) (9. we have shown that the photons can resonantly interact (in a plasma or in a vacuum) with gravitational shocks and . This last expression shows that photons with an exactly parallel motion cannot be accelerated by the gravitational wave.64) with = 2 kθ . These various examples of metric solutions. For a stationary gravitational field.

However. Furthermore. This is not. our estimates need to be validated by more credible metric solutions and more accurate numerical calculations. This could eventually provide a natural explanation for the recently observed gamma ray bursts [75]. Nevertheless. but the physical consequences of such damping. over distances well inside the cosmological constraints. apart from its specific astrophysical implications. it results from our equations that such an interaction is universal. We have also found that an effective photon mass. the plasma waves can be damped by photon Landau damping. The frequency shift resulting from these basic processes is what we call photon acceleration. in optical fibres or other optical media. Actually. has not yet been explored. photons can oscillate and can be trapped in the wavefield. in the same way as they are Landau damped by the electrons.3 Mean field acceleration processes This section contains a few concluding remarks. independent of the way they are known. their dynamical interaction with electrostatic waves in a plasma is very similar to that of a charged particle (an electron or an ion) interacting with the same wave. the choice of the words is never completely innocent or arbitrary. In particular. Presently. in the sense that it affects every photon moving in a gravitational field. this interaction will eventually lead to very high frequency up-shifts. and reflects the ideas that we have about the physical reality. and that it is more effective when the space–time disturbance of the medium travels with a velocity nearly equal to that of the photons. a significant fraction of the physics community is still reluctant to talk about photon acceleration and prefers to use other terms such as frequency shift or phase modulation. We have started with very simple ideas concerning the processes of time refraction and time reflection. However. because what is important in physics is to have an accurate view of the physical processes. We have shown that it can occur in a large variety of physical situations. a big problem. which is the statistical counterpart of the acceleration of photons by a gravitational wave. According to our rough estimates. it has been known for quite some time that a gravitational wave can be Landau damped by a background photon gas [18]. and an equivalent electric . in principle.Mean field acceleration processes 193 with gravitational wavepackets. which correspond to a natural extension of the familiar concepts of refraction and reflection into the space–time domain. in plasmas. We have shown in this work that the photons in a medium are subjected to a force. On the other hand. 9. We also believe that. our simple approach points to a new and fundamental coupling mechanism between the electromagnetic and the gravitational waves which should be investigated in the future. proportional to the time derivative of the refractive index.

However. the fact that we were able to isolate the new concept of an equivalent charge led us immediately to the problem of secondary radiation emitted by accelerated photons. What. they are accelerated. We can also explore the idea of photon acceleration as a particular example of particle acceleration by a time-varying mean field. but also with other particles and with other fields. or to neutrinos moving in a dense plasma. in other words. Other questions related to this concept. and that more generally. at first sight. is seen as a more fashionable choice of terminology can lead us to ask new questions and eventually to get a deeper understanding about physics. We saw in this chapter that the ideas of photon acceleration in an optical medium can be extrapolated to the case of photons interacting in a vacuum with a gravitational field. This means that the use of photon acceleration as a genuine physical concept can lead to a more global view of the physical processes and of the elementary interactions between the various particles and the various fields. by receiving energy from the fields with which they interact. but not considered here. they are energized or. or radiation pressure effects. such as the photon ondulator effects of the photon transition radiation. . We can still argue that photons correspond to a particle description of the electromagnetic field. and with no electric charge. This means that we should not deny for photons what we accept as true for other particles: that. could be defined. are the possibility of photon bending in a static magnetic field or the attraction between two parallel photon beams. Similar processes can also be found for the strong interaction. As an example. a wave description is necessary. the possibility of particle acceleration by the non-stationary nuclear matter produced by relativistic heavy ion collisions [20] is presently being explored [96]. This shows that our familiar view of photons as particles with no rest mass. we could say that the equivalent charge of a photon in a plasma is nothing but a different way of describing the well-known ponderomotive force. can only apply to ‘bare’ photons moving in a vacuum and not to ‘dressed’ photons moving in a background medium. These different mean field processes involve the electromagnetic. But the same is also true for the other fields and for the other particles. the weak and the gravitational interactions. Such a mean field process could operate not only with photons and the electromagnetic field. In particular.194 New developments charge (or a dipole) for the photons in a medium.

1) Now.2) (A.Appendix Derivation of the Wigner–Moyal equation A. (A.6) 195 . 2 2 (∇1 − ∇2 ) − 1 c2 ∂2 ∂2 − 2 2 ∂t1 ∂t2 C12 = 1 c2 ∂2 ∂2 χ1 − 2 χ2 C12 2 ∂t1 ∂t2 (A. 2. after subtracting the resulting two equations. 2 2 2 c2 ∂t2 (A. the refractive index is always real and for this reason we can write χi = χi∗ . c2 ∂t 2 c ∂t (A. ti ).1 Non-dispersive media We consider transverse electromagnetic fields (∇ · E = 0).5) For convenience. for i = 1. Noticing that.4) with ∗ C12 = E 1 · E 2 . ti ) and χi ≡ χ (ri . in the absence of losses. and we write 2 ∇1 − 1 ∂2 2 c2 ∂t1 1 ∂2 2 c2 ∂t2 E1 = E2 = 1 ∂2 χ E 2 1 1 c2 ∂t1 1 ∂2 χ E . we use the notation E i ≡ E(ri . let us introduce new space and time variables. as described by ∇2 E − 1 ∂2 E 1 ∂2 = 2 2 (χ E). 2 s = r1 − r2 (A.3) 2 ∇2 − ∗ Let us multiply the first of these equations by E 2 and the complex conjugate of the second one by E 1 . such that r= 1 (r1 + r2 ). we obtain. in a non-dispersive medium.

we can assume that the susceptibility χ is a slowly varying function and that its dependence on the fast time variable τ is negligible.7) (A. by performing a double (space and time) Taylor expansion of the susceptibilites χ1 and χ2 around r and t.12). τ = t1 − t2 . 2 In an equivalent manner.9) (A.12) Introducing an exponential operator. (A. 2 1 ∂t ∂τ c ∂τ (A. we obtain χ1 = χ (r + s/2. we could have stated that t= s r1 = r + .15) . as will become more obvious in the following.14) This means that. ∂t (A. by making a Taylor expansion of a function of time f (t + τ ) around f (t). t2 = t − . t + τ/2) = exp τ ∂ s ·∇ + 2 2 ∂t χ (r . we can easily realize that t1 = t + ∂2 ∂2 χ1 − 2 χ2 2 ∂t1 ∂t2 = 1 ∂2 ∂2 + 2 4 ∂t 2 ∂τ (χ1 − χ2 ) + ∂2 (χ1 + χ2 ). 2 2 Using these variable transformations.11) c2 We know that. (A. this can be written in a more elegant and more compact form as f (t + τ ) = exp τ ∂ ∂t f (t). 2 r2 = r − s 2 (A. Using (χ1 + χ2 ) 2χ . we can then write this equation as 2 ∇ · ∇s − ∂2 ∂t∂τ C12 1 ∂2 ∂χ ∂ (χ − χ2 ) 2 C12 + 2 C12 . Similarly.196 and Derivation of the Wigner–Moyal equation 1 (t1 + t2 ).8) and τ τ . a function of the coordinates f (r + s) can be expanded around f (r ) as f (r + s) = exp(s · ∇) f (r ). we can obtain f (t + τ ) = f (t) + 1 m ∂m f (t) τ m! ∂t m m=1 ∞ f (t) + τ ∂ f (t) + ···.13) A power series development of this exponential operator clearly shows that this is equivalent to equation (A. ∂t∂τ (A. Furthermore. t) (A.10) This expression can be simplified by noting that τ is a fast timescale and t is a slow timescale.

t. t − 2 2 e−ik·s+iωτ . Using this definition in equation (A. ∂ωm (A. (A. (A. (A. ω. 197 (A.21) But.17).21) as (χ1 − χ2 )F = 2 1 ∂ ∂ 1 ∂ (−1)l ·∇ =− (2l + 1)! 2 ∂ k 2 ∂ω ∂t 2l+1 χ F. ω. t).23) . we obtain ∂ c2 k ∂ ln ω + ·∇ F + F = i (χ1 − χ2 )F.11). we can write on the right-hand side of this equation (χ1 − χ2 )F = 2 ∞ l=0 F (2l + 1)! s ·∇ + 2 τ ∂ 2 ∂t 2l+1 χ. ∂m = (iτ )m F. t − τ/2) = exp − · ∇ − 2 2 ∂t χ (r .20) Using equation (A. t. 2π (A. s. k) = = ds ds dτ C(r . ∂t ω ∂t 2 (A.Non-dispersive media and s τ ∂ χ2 = χ (r − s/2.17) At this point it is useful to introduce the double Fourier transformation of C12 : C12 ≡ C(r . τ ) e−ik·s+iωτ s τ · E∗ r − .22) This means that we can rewrite equation (A. s. k) eik·s−iωτ .16) This means that the difference between the two values of the susceptibility of the medium can be written as (χ1 − χ2 ) = 2 sinh =2 ∞ l=0 s τ ∂ ·∇ + 2 2 ∂t 1 (2l + 1)! χ τ ∂ 2 ∂t 2l+1 s ·∇ + 2 χ.19) s τ dτ E r + . τ ) = dk (2π)3 dω F(r .18) The corresponding inverse transformation is related to the electric field as follows: F(r . we can also write ∂m ∂ km ∞ l=0 = (−is)m F. t. (A. t. t + 2 2 This quantity is the Wigner function for the electric field. from the definition of F.

2 Dispersive media In a dispersive medium. k) = Fk (r .198 Derivation of the Wigner–Moyal equation Using this result in equation (A. we have F ≡ F(r . A. t.26) = e−iωk τ C(r . Using this in the definition of C12 . we get C12 ≡ C(r .20) we finally obtain ∂ c2 k + ·∇ F + ∂t ω ∂ ∂t F = −ω( sin F) (A. we can see that equations (A. t. this means that we can define Fk (r . τ = 0). t.30) .29) with P = 0 E − D. According to equation (A. s.2. we have ∇2 − 1 ∂2 c2 ∂t 2 E = µ0 ∂2 P ∂t 2 (A.24) with = 1← ∂ ∂ ∂ ∂ − · 2 ∂r ∂ k ∂t ∂ω → . 1 ∂2 c2 ∂ti2 E i = µ0 ∂2 Pi ∂ti2 (A.19). t) as the space Wigner function for the electric field: Fk (r . τ ) = e−iωk τ Fk (r . τ = 0) e−ik·s ds E(r + s/2. 2. t) = = C(r .28) A. s. ω.27) (A. (A. t)δ(ω − ωk ).25) For a linear wave spectrum. t) e−ik·s ds. t) · E ∗ (r − s/2. t) eik·s dk (2π )3 (A.3) have to be replaced by ∇i2 − for i = 1.1. t. Returning to the procedure followed in appendix A. s. (A.

we can derive from here an evolution equation for the quantity C12 = ∗ E 1 · E 2 . ω. ki ) eiki ·ri −iωi ti (2π )3 (A. and = ω1 + ω2 . appearing in this expression. ωi . ti . ti ) = dki E(ωi . k). 2 k2 = (A. ki ) E(ωi . we could have stated that k1 = q + k. such that q = k1 + k2 . 2 q −k 2 (A.37) Equivalently. ω= k= 1 (k1 − k2 ) 2 1 (ω1 − ω2 ). t.32) Here we can introduce the Fourier transformation E i ≡ E(ri . ti . ti ) = such that P(ri . We can rewrite the quantity C12 in terms of the Fourier components of the electric field. (A.31) = µ0 ∂2 ∂2 ∗ ∗ ( P1 · E 2 ) − 2 ( P2 · E 1 ) .35) The susceptibility of the medium χ (r . is assumed to be a slowly varying function of space and time.34) (A. ti . A. ki ) eiki ·ri −iωi ti .Dispersive media 199 Again. dωi 2π dki P(ri . The result is 2 2 (∇1 − ∇2 ) − 1 c2 ∂2 ∂2 − 2 2 ∂t1 ∂t2 C12 (A. ωi . ki ). This equation becomes 2 ∇ · ∇s − = µ0 + µ0 1 ∂2 c2 ∂t∂τ C12 ∗ ∗ P1 · E 2 − P2 · E 1 1 ∂2 ∂2 + 2 2 4 ∂t ∂τ ∂2 ∂t∂τ dωi 2π ∗ ∗ P1 · E 2 + P2 · E 1 .6.36) (A.7). ki ) = 0 χ (ri . (2π )3 (A.33) A similar transformation for the polarization vector is defined by Pi ≡ P(ri .38) . But because this would lead to quite cumbersome expressions. we prefer to introduce new frequency and wavevector variables. ωi . 2 ∂t1 ∂t2 Let us now introduce the space and time variables defined by equations (A.

(A.200 and Derivation of the Wigner–Moyal equation ω1 = + ω. −k + q/2). we can write 2 ∇ · ∇s − 1 ∂2 c2 ∂t∂τ C12 = − 1 c2 d 2π dq (η+ − η− ) (2π )3 t ik·s−iωτ × J (q.42). t). k. 2 2 2 2 (A. ω. t. k) = with J (q. k + q/2) · E(−ω + /2. whereas the frequency is associated with the slow timescale t. ω) = E(ω + /2. k.42).39) 2 2 As in the case of the space and time variable transformations (A.t ± . . we can use η+ − η− = ∂η0 ∂η0 +q · ∂ω ∂k + τ ∂η0 ∂η0 +s· ∂t ∂r .44) where we have considered that η0 = ω2 χ (r . (A.43) Here. in equation (A.46) (A. In terms of these new variables. . the one proportional to ∂ 2 /∂τ 2 . ω) eiq·r −i e (A. t. .18). In the same way. we should notice that |ω| | | because ω is associated with the fast timescale τ . we can assume that |k| |q|. ω. k). ω2 = − ω.32) and retaining on its right-hand side only the dominant term.9).42) where. Developing these quantities around the values (ω.45) of But we also notice that the quantity η0 and its derivatives are independent and q. k. we obtain η± η0 ± τ ∂η0 q ∂η0 s ∂η0 ∂η0 ± ± · ± · + ··· 2 ∂ω 2 ∂k 2 ∂t 2 ∂r (A. t.41) d 2π dq J (q. k) and (r . the Jacobian of the new transformations is equal to one: dω1 dω2 = d dω and dk1 dk2 = dq dk. (A.6–A. ω) eiq·r −i (2π )3 t (A. It means that. in equation (A. as it should be.40) Returning to equation (A. we have introduced the quantities 2 η± = ω ± 2 s τ q χ r ± . they can be taken out of the . in order to simplify the expression. with the quantity F(r . ω. k) defined now as F(r . This means that.k ± . the quantity C12 becomes formally identical to equation (A.ω ± .

t. k) ∂t (A. ∂ω ∂t ∂k (A. τ ) ei(k −k)·s e−i(ω −ω)τ . t.18). t. ω. . k.49) + s · ∇η0 F(r . ∂t We can now replace C12 ≡ C(r . But. τ ) by its Fourier integral. ω.50) dk (2π )3 × C(r . as defined by equation (A. it is also clear that we can write τ e−i(ω −ω)τ dτ = i and s ei(k −k·s) ds = −(2π )3 iδ(k − k) ∂ ∂ω e−i(ω −ω)τ dτ = 2π iδ(ω − ω) ∂ ∂ ∂ω (A. ω.51) . k) − 2π ∂ω ∂t (2π)3 ∂k ∂η0 τ (A. t. k. t. . k) eik·s−iωτ . ω. ω. k) = − dτ dω 2π i c2 ∂η0 ∂ ∂η0 · ∇ F(r . t. k). ω.51) into a closed differential equation for the Wigner function F ≡ F(r . in the same equation: d 2π and d 2π dq q J (q. and we obtain 2i k · ∇ + − 1 c2 ω ∂ c2 ∂t ds F(r . ω) eiq·r −i (2π)3 t dq (2π)3 J (q.53) . which takes the form 2ω ∂ c2 k + ·∇ F = − ∂t ω + ∂η0 ∂ F ∂η0 − · ∇F ∂ω ∂t ∂k ∂ F ∂η0 ∂F − · ∇η0 . t. This allows us to make the following replacements. ω. s.Dispersive media 201 integrations in these variables.52) ∂k This means that we can finally transform equation (A. s.47) = −i∇ F(r . (A. k) − ∂ω ∂t ∂k ∂η0 τ + s · ∇η0 ∂t (A. k). (A. t. ω) eiq·r −i t =i ∂ F(r .48) The result is 2 ∇ · ∇s − =− + 1 c2 1 ∂2 c2 ∂t∂τ C12 ∂η0 ∂ dω dk ∂η0 i · ∇ F(r . t.

25). we can rewrite it in a more suitable form: ∂ 2 + vg · ∇ F = − (η0 F) ∂t 2ω + ∂η0 /∂ω where is defined by equation (A. 2c2 k − ω2 ∂ /∂ k 2ω + ω2 ∂ /∂ω (A.55) (A.54) . and vg = with = 1 + χ = 1 + η0 /ω2 .202 Derivation of the Wigner–Moyal equation After rearranging the terms in this equation.

Fusion 34 557 [13] Bingham R. 21 57 [8] Bernstein I B 1975 Geometric optics in space. gauge invariance.References [1] Abdullaev S S 1993 Chaos and Dynamics of Rays in Waveguide Media (London: Gordon and Breach) [2] Abdullaev F 1994 Theory of Solitons in Inhomogeneous Media (New York: Wiley) [3] Alfano R R (ed) 1989 The Supercontinuum Laser Source (New York: Springer) [4] Ammosov M V. Lett. DeAngelis U. 52 263 [18] Chesters D 1973 Dispersion of gravitational waves by a collisionless gas Phys. ed M N Rosenbluth and R Z Sagdeev (Amsterdam: North-Holland) [10] Bethe H 1986 Possible explanation of the solar-neutrino puzzle Phys. A 193 279 [14] Bingham R.– JETP 64 1191 [5] Anderson P W 1963 Plasmons. 78 247 [15] Born M and Wolf E 1970 Principles of Optics (Elmsford. Lett. Rev.and time-varying plasmas Phys. NY: Pergamon) [16] Budden K G 1961 Radio Waves in the Ionosphere (Cambridge: Cambridge University Press) [17] Chirikov B V 1979 Universal instability of many-dimensional oscillator sytems Phys. Lett. 56 1305 [11] Binette L. D 7 2863 203 . ¸ Rev. Delone N B and Krainov V P 1985 Tunnel ionization of complex atoms and of atomic ions in an alternating electromagnetic field Sov. Phys. Cairns R A and MaNamara B 1992 Relativistic Langmuir waves generated by ultrashort laser pulses Plasma Phys. Su J J and Bethe H A 1994 Collective interactions between neutrinos and dense plasmas Phys. Rep. Dawson J M. Rev. Rev. Amin M R. and mass Phys. Mori W B and Dawson J M 1993 Computation of the electric and magnetic fields induced in a plasma created by ionization lasting a finite interval of time IEEE Trans. J. Control. Mendonca J T and Dawson J M 1997 Photon Landau damping Phys. Fluids 18 320 [9] Bernstein I B and Friedland L 1983 Geometric optics in space and time varying plasmas Handbook of Plasma Physics vol 1. 130 439 [6] Bachelard G 1975 La Formation de l’Esprit Scientifique (Paris: Librairie Philosophique J Vrin) [7] Banos A Jr. Joguet B and Wang J C L 1998 Evidence of Fermi acceleration of Lyman α in the radio galaxy 1243+036 Astrophys. Plasma Sci. 505 634 [12] Bingham R.

Rep. Rev. ST—Accelerators Beams 1 031301 [23] Diels J C and Rudolph W 1996 Ultrashort Laser Pulse Phenomena (New York: Academic) [24] Dysthe F B. Parker R K.– JETP 66 290 [35] Gradshteyn I S and Ryzhik I M 1980 Table of Integrals. Phys. Badiche X. Phys. A 44 3908 [28] Faro A and Mendonca J T 1982 Interaction of electromagnetic waves with a moving ¸ perturbation in a stationary gas J. Silva L O. Lett. Series and Products (New York: Academic) pp 307. Mendonca J T and Franszeskakis D J 1996 Relativistic ¸ theory of frequency blue-shift of an intense ionizing laser beam in a plasma IEEE Trans. Pasour J. Atoms (Amsterdam: North-Holland) [38] Harris S E 1997 Electromagnetically induced transparency Phys. Photons. Blasco F. Dutton Z and Behroozi C H 1999 Light speed reduction to 17 metres per second in an ultracold atomic gas Nature 397 594 [40] Hillary M. Dos Santos A. Rev. Rev. 16 287 [29] Fermi E 1949 On the origin of the cosmic radiation Phys. D: Appl. O’Connel R F. Stenz C.204 References [19] Cirone M. Chaos and Plasma Physics 2 (Singapore: World Scientific) p 237 [31] Fisher D L and Tajima T 1993 Superluminous laser pulse in an active medium Phys. NJ: Prentice-Hall) [26] Esarey E. 24 323 . Vol. Plasma Sci. 71 4338 [32] Fukuda Y et al 1999 Measurement of the flux and zenith-angle distribution of upward through going muons by Super Kamiokande Phys. A 42 3526 [27] Esarey E. Today 50 36 [39] Hau L V. Mysyrowicz A. Rev. Rev. Harris S E. Rev. Phys. Rzazewski K and Mostowski J 1997 Photon generation by time-dependent dielectric: a soluble model Phys. Antonetti A and Mendonca J T 1997 Experimental evidence of ¸ photon acceleration of ultrashort laser pulses in relativistic ionization fronts Phys. Lopes N. 106 121 [41] Hizanidis K. Silva L O and Mendonca J T 1998 Photon acceleration versus frequency¸ domain interferometry for laser wakefield diagnostics Phys. Lett. Rev. Mendonca J T and Silva L O 1996 Discrete Fermi mapping for photon ¸ acceleration in a time varying plasma cavity Transport. Rev. Rev. Ting A and Sprangle P 1990 Frequency shifts induced in laser pulses by plasma waves Phys. Rev. Scully M O and Wigner E P 1984 Distribution functions in physics: fundamentals Phys. 75 1169 [30] Figueira G. Schlesinger S P and Seftor J L 1976 Realization of a relativistic mirror: electromagnetic backscattering from the front of a magnetized relativistic electron beam Phys. Sprangle P. Vomvoridis J L. MA: Addison-Wesley) [34] Gorbunov L M and Kirsanov V I 1987 Excitation of plasma waves by an electromagnetic wave packet Sov. Pecseli H L and Trulsen J 1983 Stochastic generation of continuous wave spectra Phys. 338 [36] Granatstein V L. 78 4773 [22] Dias J M. A 14 1194 [37] Haken H 1981 Light. 50 353 [25] Elliott R S An Introduction to Guided Waves and Microwave Circuits (Englewood Cliffs. Lett. Joyce G and Sprangle P 1991 Frequency up-shifting of laser pulses by co-propagating ionization fronts Phys. 1: Waves. A 55 62 [20] Csernai L P 1994 Introduction to Relativistic Heavy Ion Collisions (Chichester: Wiley) [21] Dias J M. Lett. 82 2644 [33] Goldstein H 1980 Classical Mechanics 2nd edn (Reading. Herndon M.

References 205 [42] Special issue on Second Generation Plasma Accelerators 1996 IEEE Trans. Plasma Sci. . 43 354 ¸ [63] Mendonca J T 1979 Nonlinear interaction of wavepackets J. Plasma Sci. Plasma Phys. 58 647 [66] Mendonca J T and Silva L O 1994 Regular and stochastic acceleration of photons ¸ Phys. Sergeev A M. 24 (2) [43] Jackson J D 1999 Classical Electrodynamics 3rd edn (New York: Wiley) [44] Kalluri D K 1999 Electromagnetics of Complex Media: Frequency Shifting by a Transient Magnetoplasma Medium (Boca Raton. Lirin S F. MA: MIT Press) [50] Kruer W L 1988 The Physics of Laser Plasma Interactions (Redwood City. 22 15 ¸ [64] Mendonca J T 1985 Beat-wave excitation of plasma waves J.18 823 [46] Kaw P K. Phys. Rev. Rev. Frantzeskakis D J. Sen A and Katsouleas T 1992 Nonlinear 1D laser pulse solitons in a plasma Phys. Phys. Lett. 34 115 ¸ [65] Mendonca J T. 206 205 [61] Loudon R 1983 The Quantum Theory of Light (Oxford: Clarendon) [62] Mendonca J T 1979 Nonlinear transition radiation in a plasma Phys.– JETP 20 1307 [48] Kim A V. Fluids 21 42 [54] Landau L D 1946 On the vibration of the electric plasma J. Plasma Sci. Silva L O and Vomvoridis Y L 1997 ¸ Covariant formulation of photon acceleration J. Rev. E 49 3520 [67] Mendonca J T and Silva L O 1995 Mode coupling theory of flash ionization IEEE ¸ Trans. Ott E and Walker J H 1978 Interaction of electromagnetic waves with a moving ionization front Phys. A 42 2493 [49] Klimontovich Yu L 1967 The Statistical Theory of Non-equilibrium Processes in a Plasma (Cambridge. 68 3172 [47] Keldysh L V 1965 Ionization in the field of a strong electromagnetic wave Sov. CA: Addison-Wesley) [51] Kuhn T S 1970 The Structure of Scientific Revolutions 2nd edn (Chicago. Phys. Lett. Plasma Phys. Vanin E V and Stenflo L 1990 Compression and frequency up-conversion of an ultrashort ionizing pulse in a plasma Phys. IL: University of Chicago Press) [52] Kuo S P and Ren A 1993 Experimental study of wave propagation through a rapidly created plasma IEEE Trans. Hisanidis K. 21 53 [53] Lampe M. FL: CRC) [45] Kane D J and Trebino R 1993 Single-shot measurements of the intensity and phase of an arbitrary ultrashort pulse by using frequency resolved optical grating Opt. Plasma Phys. Lett. (USSR) 10 25 [55] Landau L D and Lifshitz E M 1996 The Classical Theory of Fields (Oxford: Butterworth-Heinemann) [56] Landau L D and Lifshitz E M 1994 Mechanics 3rd edn (Oxford: ButterworthHeinemann) [57] Landau L D and Lifsjitz E M 1984 Electrodynamics of Continuous Media 2nd edn (Oxford: Butterworth-Heinemann) [58] Leonhardt U 1997 Measuring the Quantum State of Light (Cambridge: Cambridge University Press) [59] Lichtenberg A J and Lieberman M A 1983 Regular and Stochastic Motion (New York: Springer) [60] Lotz W 1967 An empirical formula for the electron impact ionization cross-section Z. Rev.

J. Plasma Phys. Dawson J M and Silva L O 1999 Neutrino ¸ Landau damping and collective neutrino–plasma processes J. Dawson J M and Lai C H 1995 Conversion of dc fields in a capacitor array to radiation by a relativistic ionization front Phys. Bingham R and Tsintsadze N L 1999 Transition radiation ¸ of photons and neutrinos at a plasma boundary Phys. Usp. A 16 2441 [88] Quang Ho-Kim and Pham Xuan Yem 1998 Elementary Particles and Their Interactions (Berlin: Springer) [89] Rax J M and Fisch N J 1993 Ultrahigh intensity laser–plasma interaction: a Lagrangian approach Phys. Lett. Bingham R. Rev. A 209 78 [69] Mendonca J T. 146 245 [71] Mendonca J T. Commun. A 250 144 [73] Mendonca J T. Phil. Shukla P K. Lett. Nucl. Camb. 15 88 [78] Mori W B 1991 Generation of tunable radiation using an underdense ionization front Phys. Hizanidis K and Franszeskakis D J 1998 Method for generating ¸ tunable high-frequency harmonics in periodically modulated ξ (2) materials Opt. Shukla P K. energetics. 415 181 [76] Misner C W. Shukla P K and Bingham R 1998 Photon acceleration by gravitational ¸ waves Phys. Granatstein V L and Parker R K 1997 ‘Relativistic mirror’ experiment with frequency tuning and energy gain Phys. Katsouleas T. . 44 5118 [79] Mori W B. Part. Bingham R. Semikov V B and Smorodinskii Ya A 1994 Electrodynamics of neutrinos in a medium Phys. at press [74]