Optics and Quantum Electronics - Research Laboratory of Electronics PDF

Physical Foundations of by David Klyshko
Quantum Electronics
7930.9789814324502-tp.indd 2 3/23/11 3:11 PM

This page intentionally left blank
David Klyshko
Quantum Electronics
by
Physical Foundations of
Editors
Maria Chekhova
Lomonosov Moscow State University, Russia
Sergey Kulik
Lomonosov Moscow State University, Russia
World Scientific
NEW JERSEY LONDON SINGAPORE BEIJING SHANGHAI HONG KONG TA I P E I CHENNAI
7930.9789814324502-tp.indd 1 3/23/11 3:11 PM

Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
PHYSICAL FOUNDATIONS OF QUANTUM ELECTRONICS BY DAVID KLYSHKO

Copyright 2011 by World Scientific Publishing Co. Pte. Ltd.
All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to
photocopy is not required from the publisher.
ISBN-13 978-981-4324-50-2
ISBN-10 981-4324-50-7
Printed in Singapore.
Alvin - Physical Foundations of Quantum.pmd 1 3/1/2011, 9:45 AM

March 23, 2011 16:14 World Scientific Book - 9in x 6in ws-book9x6
Preface
This book belongs to the series of textbooks in electronics and radiophysics writ-
ten at the Physics Department of Lomonosov Moscow State University. Similarly
to the other books of this series [Migulin (1978); Vinogradova (1979)], it is writ-
ten for undergraduate Physics students and aims at introducing the readers to the
most general concepts, rules, and theoretical methods. The main focus is on the
three directions in physical optics that appeared after the advent of lasers: nonsta-
tionary interactions between light and matter (Chapter 5), optical anharmonicity
of matter (Chapter 6) and quantum properties of light (Chapter 7). The first four
chapters describe the theoretical base of more traditional parts of quantum elec-
tronics. The book starts with a short review of the history of quantum electronics
with its main concepts, ideas, and terms. Further, basic methods for describing
the interaction of optical radiation with matter are considered, based on quantum
transition probabilities (Chapter 2), the density matrix formalism (Chapter 3), and
the linear dielectric susceptibility of matter (Chapter 4).
The author tried to combine a systematic approach with a more detailed in-
sight into several interesting ideas and effects, such as, for instance, superradiance
(Sec. 5.3), phase conjugation (Sec. 6.5), and photon antibunching (Sec. 7.6).
The reader is expected to know the basics of quantum mechanics and statistical
physics; however, much attention is paid to explaining the notations used in the
book. The author tried to gradually increase the presentation complexity within
each section as well as within the whole book. Each section or chapter starts with
a simplified qualitative picture of the phenomenon considered. More complicated
sections providing additional information are marked by circles.
The book uses the Gaussian system of units, which is most common in quan-
tum electronics; however, in the numerical estimates, energy and power are given
in Joules and Watts.
v
vi Physical Foundations of Quantum Electronics
A large number of general guides in quantum electronics have already

been published [Klimontovich (1966); Zhabotinsky (1969); Bertin (1971); Fain
(1972); Pantell (1969); Yariv (1989); Piekara (1973); Khanin (1975); Tarasov
(1976); Loudon (2000); Apanasevich (1977); Maitland (1969); Svelto (2010);
Strakhovskii (1979); Kaczmarek (1981); Tarasov (1981); Elyutin (1982)] at all
levels of presentation, from popular books [Klimontovich (1966); Zhabotinsky
(1969); Piekara (1973)] to fundamental monographs [Fain (1972); Khanin (1975);
Apanasevich (1977)], and in many cases the reader will be referred to them. For
instance, the present book does not consider the design and parameters of lasers
and masers as well as their various applications. The theory of optical resonators
and waveguides is presented, in particular, in the University course of wave the-
ory [Vinogradova (1979)] (see also [Maitland (1969); Yariv (1976)]), while the
self-oscillation theory, dynamics, and classical statistics of laser systems can be
found in the textbooks on the oscillation theory [Migulin (1978)] and statistical
radiophysics [Akhmanov (1981)] (see also [Khanin (1975); Rabinovich (1989)]).
The book is based on the lecture course in quantum electronics taught by the
author to undergraduate students for 20 years. This course was started in 1960,
after a suggestion by S. D. Gvozdover, even before the appearance of lasers. At
first, the course was completely devoted to masers (paramagnetic amplifiers and
molecular generators) and radio-spectroscopy. The advent of lasers and the laser
revolution in optics, spectroscopy and other fields of science made the author
move the center of gravity of the course from the microwave range to the op-
tical one and supply the course with new sections. However, one should keep in
mind that lasers and masers are based on common principles and that quantum
electronics originated from radio spectroscopy and radiophysics. The latter pro-
vided quantum electronics with one of its basic notions, the feedback, and it is not
by chance that the founders of quantum electronics and nonlinear optics, such as
Basov, Bloembergen, Khokhlov, Prokhorov, Townes, and many others, worked in
radiophysics. Sometimes quantum electronics is called quantum radiophysics.
Both the Quantum Electronics lecture course and this book were hugely
influenced by Rem Viktorovich Khokhlov whose advice and friendship are un-
forgettable. The author is indebted to P. V. Elyutin, A. M. Fedorchenko and
A. S. Chirkin, who have read the manuscript and helped to eliminate many flaws.
The author is also grateful to V. B. Braginsky who stimulated the writing of this
book.
D. N. Klyshko
Foreword
Below, we present the translation of a book by David Klyshko (19292000), which

was originally published in 1986. This is a remarkable book by a remarkable per-
son whose insight into physics in general and quantum electronics in particular
was so deep that even now, after nearly 25 years, a lot of new ideas can still be
found in this book. The main advantage of the book is that it generalizes seem-
ingly unique effects and joins together seemingly different approaches. Because
it is mainly at the boundaries of the explored that one should look for new ideas
and discoveries, this book will be helpful for both a researcher and an ambitious
student aiming at research in nonlinear optics, laser physics, quantum or atom
optics.
Although some parts of the book look very new even now, others are definitely
outdated. This statement relates not to the sections or even subsections of the
book; rather, it is about numerous references to the technology or parameters of
the equipment that were available when the book was written. This requires addi-
tional comments and explanations, which we have endeavored to make throughout
the whole text, mostly as footnotes but sometimes as additional sections (Secs. 1.3,
7.2.10 and 7.5.7).
At the same time, we by no means think that the additional parts provide a
complete view at the modern state of quantum electronics. For this reason, we
have also included an additional list of references, containing books or review
articles that appeared after the original book had been published.
Maria Chekhova
Sergey Kulik
The Editors
vii

List of Notation and Acronyms
a, transverse size, cm; photon annihilation operator

A, area, cm2 ; probability of spontaneous transition, s1 ; vector potential,
(erg/cm)1/2
B, scaling coefficient between the stimulated transition probability and
the energy spectral density, cm3 /(ergs2 )
c, state amplitude
d, dipole moment, (ergcm3)1/2
D, electric induction, (erg/cm3)1/2
e, unit polarization vector
E, electric field, (erg/cm3)1/2
E, energy, erg
f , frequency, s1 , oscillator strength
F, photon flux density, cm2 s1 ; free energy, erg
g, degeneracy; form factor, s
G, transfer coefficient, Greens function; field correlation function,
erg/cm3
H, magnetic field, (erg/cm3)1/2
H, Hamiltonian, erg
I, intensity of radiation, erg/(cm2s); identity operator
j, current density, erg/(cm3s2 )1/2
k, wave vector, cm1
l, length, cm
n, refractive index
N, density of molecules or photons, cm3 ; number of photons per mode
Ni , population of a level, cm3
N, mean number of photons per mode in equilibrium radiation
p, momentum, gcm/s; pressure, erg/cm3
ix
x Physical Foundations of Quantum Electronics
P, polarization, (erg/cm3)1/2 ; probability

P, power, erg/s
q, generalized coordinate
Q, quality factor; generating function
r, radius vector, cm
R, Bloch vector; reflectivity coefficient
s, angular momentum, ergs
S, Poynting vector, erg/(cm2s)
T, time interval, s; temperature, K
u, group velocity, cm/s
U, internal energy, erg; evolution operator
v, phase velocity, cm/s
V, volume, cm3
V, interaction energy, erg
w, relaxation transition probability per unit time, s1
W, transition probability per unit time, s1
Z, statistical sum
, linear polarisability, cm3 ; absorption or amplification coefficient,
cm1
, quadratic polarisability, (cm9 /erg)1/2
, cubic polarisability, cm6 /erg; dissipation constant, s1
, relative population difference
, dielectric permittivity
, quantum efficiency
, angle or angle of precession, rad
, Heaviside step function
, Boltzmanns constant, erg/K
, wavelength, cm; o = /2
, magnetic dipole moment, (ergcm3)1/2 ; Fermi level, erg
, polarization index; wavenumber, cm1
, operator of projection or summation over permutations
, density operator or matrix; mass density, g/cm3 ; charge density,
(erg/cm5)1/2
, interaction cross-section, cm2 ; Pauli matrix
, relaxation or correlation time, s
, phase or azimuthal angle, rad; eigenfunctions of the energy operator
(n) , n-th order susceptibility of the medium, (erg/cm3)(1n)/2 =(Hs)1n
, , wave function
List of Notation and Acronyms xi
, circular frequency, rad/s

, Rabi frequency, rad/s; solid angle, sr
CARS, coherent anti-Stokes Raman scattering
CF, correlation function
EPR, electronic paramagnetic resonance
FDT, fluctuation-dissipation theorem
IR, infrared
MBS, Mandelshtam-Brillouin scattering
MW, microwave
NMR, nuclear magnetic resonance
OPO, optical parametric oscillator
PC, phase conjugation
PDC, parametric down-conversion
PMT, photomultiplier tube
SHG, second harmonic generation
SIT, self-induced transparency
SPDC, spontaneous parametric down-conversion
SRS, spontaneous Raman scattering
StRS, stimulated Raman scattering
StPDC, stimulated parametric down-conversion
StTS, stimulated temperature scattering
SVA, slowly varying amplitude
UV, ultraviolet

Contents
Preface v
Foreword vii
List of Notation and Acronyms ix
1. Introduction 1
1.1 Basic notions of quantum electronics . . . . . . . . . . . . . . . 2
1.1.1 Stimulated emission . . . . . . . . . . . . . . . . . . . 2
1.1.2 Population inversion . . . . . . . . . . . . . . . . . . . 2
1.1.3 Feedback and the lasing condition . . . . . . . . . . . 3
1.1.4 Saturation and relaxation . . . . . . . . . . . . . . . . 4
1.2 History of quantum electronics . . . . . . . . . . . . . . . . . . 5
1.2.1 First steps . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.2 Radio spectroscopy . . . . . . . . . . . . . . . . . . . 6
1.2.3 Masers . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.4 Lasers . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Recent progress in quantum electronics (added by the Editors) . 9
1.3.1 Physics of lasers . . . . . . . . . . . . . . . . . . . . . 9
1.3.2 Laser physics . . . . . . . . . . . . . . . . . . . . . . 10
1.3.3 New trends in nonlinear optics . . . . . . . . . . . . . 10
1.3.4 Atom optics . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.5 Optics of nonclassical light . . . . . . . . . . . . . . . 11
2. Stimulated Quantum Transitions 15

2.1 Amplitude and probability of a transition . . . . . . . . . . . . 15
2.1.1 Unperturbed atom . . . . . . . . . . . . . . . . . . . . 16
xiii
xiv Physical Foundations of Quantum Electronics
2.1.2 Atom in an alternating field . . . . . . . . . . . . . . . 18

2.1.3 Perturbation theory . . . . . . . . . . . . . . . . . . . 19
2.1.4 Linear approximation . . . . . . . . . . . . . . . . . . 20
2.1.5 Probability of a single-quantum transition . . . . . . . 21
2.2 Transitions in monochromatic field . . . . . . . . . . . . . . . . 21
2.2.1 Dipole approximation . . . . . . . . . . . . . . . . . . 21
2.2.2 Transition probability . . . . . . . . . . . . . . . . . . 22
2.2.3 Finite level widths . . . . . . . . . . . . . . . . . . . . 24
2.3 Absorption cross-section and coefficient . . . . . . . . . . . . . 26
2.3.1 Relation between intensity and field amplitude . . . . . 26
2.3.2 Cross-section of resonance interaction . . . . . . . . . 27
2.3.3 Population kinetics . . . . . . . . . . . . . . . . . . . 28
2.3.4 Photon kinetics . . . . . . . . . . . . . . . . . . . . . 28
2.3.5 Coefficient of resonance absorption . . . . . . . . . . . 29
2.3.6 Amplification bandwidth . . . . . . . . . . . . . . . . 30
2.3.7 Degeneracy of the levels . . . . . . . . . . . . . . . . 31
2.4 Stimulated transitions in a random field . . . . . . . . . . . . . 33
2.4.1 Correlation functions . . . . . . . . . . . . . . . . . . 33
2.4.2 Transition rate . . . . . . . . . . . . . . . . . . . . . . 34
2.4.3 Einsteins B coefficient . . . . . . . . . . . . . . . . . 35
2.4.4 Spectral field density . . . . . . . . . . . . . . . . . . 35
2.5 Field as a thermostat . . . . . . . . . . . . . . . . . . . . . . . 36
2.5.1 Spontaneous transitions . . . . . . . . . . . . . . . . . 37
2.5.2 Natural bandwidth . . . . . . . . . . . . . . . . . . . . 38
2.5.3 Number of photons, spectral brightness, and brightness
temperature . . . . . . . . . . . . . . . . . . . . . . . 39
2.5.4 Relaxation time . . . . . . . . . . . . . . . . . . . . . 41
3. Density Matrix, Populations, and Relaxation 43

3.1 Definition and properties of the density matrix . . . . . . . . . . 43
3.1.1 Observables . . . . . . . . . . . . . . . . . . . . . . . 43
3.1.2 Density matrix of a pure state . . . . . . . . . . . . . . 44
3.1.3 Mixed states . . . . . . . . . . . . . . . . . . . . . . . 45
3.1.4 More general definition of the density matrix . . . . . 47
3.1.5 Properties of the density matrix . . . . . . . . . . . . . 48
3.1.6 Density matrix and entropy . . . . . . . . . . . . . . . 49
3.1.7 Density matrix of an atom . . . . . . . . . . . . . . . 50
3.2 Populations of levels . . . . . . . . . . . . . . . . . . . . . . . 51
3.2.1 Equilibrium populations . . . . . . . . . . . . . . . . . 51
Contents xv
3.2.2 Two-level system and the negative temperature . . . . . 52

3.2.3 Populations in semiconductors . . . . . . . . . . . . . 53
3.2.4 Inversion in semiconductors . . . . . . . . . . . . . . 55
3.3 Evolution of the density matrix . . . . . . . . . . . . . . . . . . 56
3.3.1 Non-equilibrium systems . . . . . . . . . . . . . . . . 56
3.3.2 Von Neumann equation . . . . . . . . . . . . . . . . . 57
3.3.3 Interaction with the thermostat . . . . . . . . . . . . . 58
3.3.4 Evolution of a closed system . . . . . . . . . . . . . . 58
3.3.5 Transverse and longitudinal relaxation . . . . . . . . . 59
3.3.6 Interaction picture . . . . . . . . . . . . . . . . . . . . 62
3.3.7 Perturbation theory . . . . . . . . . . . . . . . . . . . 64
4. The Susceptibility of Matter 67

4.1 Definition and general properties of susceptibility . . . . . . . . 67
4.1.1 Symmetry . . . . . . . . . . . . . . . . . . . . . . . . 68
4.1.2 The role of causality . . . . . . . . . . . . . . . . . . . 69
4.1.3 Absorption of a given field . . . . . . . . . . . . . . . 70
4.1.4 Susceptibility of the vacuum . . . . . . . . . . . . . . 71
4.1.5 Thermodynamic approach . . . . . . . . . . . . . . . 72
4.2 Dispersion theory . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2.1 Dispersion law . . . . . . . . . . . . . . . . . . . . . . 75
4.2.2 The effect of absorption . . . . . . . . . . . . . . . . . 76
4.2.3 Classical theory of dispersion . . . . . . . . . . . . . . 77
4.2.4 Quantum theory of dispersion . . . . . . . . . . . . . . 79
4.2.5 Oscillator strength . . . . . . . . . . . . . . . . . . . 81
4.2.6 Isolated resonance . . . . . . . . . . . . . . . . . . . . 82
4.2.7 Polaritons . . . . . . . . . . . . . . . . . . . . . . . . 85
4.3 Two-level model and saturation . . . . . . . . . . . . . . . . . . 89
4.3.1 Applicability of the model . . . . . . . . . . . . . . . . 89
4.3.2 Kinetic equations . . . . . . . . . . . . . . . . . . . . 90
4.3.3 Saturation . . . . . . . . . . . . . . . . . . . . . . . . 91
4.3.4 Lineshape in the presence of saturation . . . . . . . . 92

4.4 Bloch equations . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.4.1 Kinetic equations for the mean values . . . . . . . . . . 95
4.4.2 Pauli matrices and expansion of operators . . . . . . . 96
4.4.3 The Bloch vector and the Bloch sphere . . . . . . . . . 99
4.4.4 Higher moments and distributions . . . . . . . . . . . . 100
4.4.5 Bloch equations . . . . . . . . . . . . . . . . . . . . . 101
4.4.6 Equation for polarization . . . . . . . . . . . . . . . . 103
xvi Physical Foundations of Quantum Electronics
4.4.7 Magnetic resonance . . . . . . . . . . . . . . . . . . . 104
5. Non-Stationary Optics 107

5.1 Stimulated non-stationary effects . . . . . . . . . . . . . . . . . 108
5.1.1 Atom as a gyroscope . . . . . . . . . . . . . . . . . . 108
5.1.2 Analytical solution . . . . . . . . . . . . . . . . . . . . 110
5.1.3 Nutation . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.1.4 Self-induced transparency . . . . . . . . . . . . . . . . 114
5.2 Emission of an atom . . . . . . . . . . . . . . . . . . . . . . . 115
5.2.1 Emission of a dipole . . . . . . . . . . . . . . . . . . . 116
5.2.2 Probability of a spontaneous transition . . . . . . . . . 117
5.2.3 Normally ordered emission . . . . . . . . . . . . . . . 118
5.2.4 Relation between spontaneous and thermal emission . . 120
5.2.5 On the emission of fractions of a photon . . . . . . . . 121
5.2.6 Quantum beats . . . . . . . . . . . . . . . . . . . . . 121
5.2.7 Resonance fluorescence . . . . . . . . . . . . . . . . 124
5.3 Collective emission . . . . . . . . . . . . . . . . . . . . . . . . 127
5.3.1 Superradiance . . . . . . . . . . . . . . . . . . . . . . 127
5.3.2 Analogy with phase transitions . . . . . . . . . . . . . 130
5.3.3 Photon echo . . . . . . . . . . . . . . . . . . . . . . . 131
6. Nonlinear Optics 135

6.1 Nonlinear susceptibilities: definitions and general properties . . 137
6.1.1 Nonlinear susceptibilities . . . . . . . . . . . . . . . . 138
6.1.2 Various definitions . . . . . . . . . . . . . . . . . . . 139
6.1.3 Permutative symmetry . . . . . . . . . . . . . . . . . 141
6.1.4 Transparent matter . . . . . . . . . . . . . . . . . . . 141
6.1.5 The role of the material symmetry . . . . . . . . . . . 144
6.2 Models of optical anharmonicity . . . . . . . . . . . . . . . . . 145
6.2.1 Anharmonicity of a free electron . . . . . . . . . . . . 146
6.2.2 Light pressure . . . . . . . . . . . . . . . . . . . . . . 149
6.2.3 Striction anharmonicity . . . . . . . . . . . . . . . . . 152
6.2.4 Anharmonic oscillator . . . . . . . . . . . . . . . . . . 154
6.2.5 Raman anharmonicity . . . . . . . . . . . . . . . . . . 157
6.2.6 Temperature anharmonicity . . . . . . . . . . . . . . . 162
6.2.7 Electrocaloric anharmonicity . . . . . . . . . . . . . . 164
6.2.8 Orientation anharmonicity . . . . . . . . . . . . . . . . 166
6.2.9 Quantum theory of nonlinear polarization . . . . . . . 169
Contents xvii
6.2.10 Probability of multi-photon transitions . . . . . . . . . 173

6.2.11 Conclusions . . . . . . . . . . . . . . . . . . . . . . . 177
6.3 Macroscopic nonlinear optics . . . . . . . . . . . . . . . . . . . 177
6.3.1 Initial relations . . . . . . . . . . . . . . . . . . . . . . 177
6.3.2 Classification of nonlinear effects . . . . . . . . . . . . 178
6.3.3 The role of linear and nonlinear dispersion . . . . . . . 181
6.3.4 One-dimensional approximation . . . . . . . . . . . . 182
6.3.5 The Manley-Rowe relation and the permutation
symmetry . . . . . . . . . . . . . . . . . . . . . . . . 187
6.3.6 Derivation of one-dimensional equations . . . . . . . 189
6.4 Non-parametric interactions . . . . . . . . . . . . . . . . . . . 191
6.4.1 Nonlinear absorption . . . . . . . . . . . . . . . . . . 191
6.4.2 Doppler-free spectroscopy . . . . . . . . . . . . . . . . 195
6.4.3 Raman amplification . . . . . . . . . . . . . . . . . . . 197
6.4.4 Spontaneous and stimulated scattering . . . . . . . . . 199
6.4.5 Self-focusing . . . . . . . . . . . . . . . . . . . . . . . 201
6.4.6 Self-focusing length . . . . . . . . . . . . . . . . . . 203
6.5 Parametric interactions . . . . . . . . . . . . . . . . . . . . . . 207
6.5.1 Undepleted-pump approximation the near field . . . 208
6.5.2 The far field . . . . . . . . . . . . . . . . . . . . . . . 210
6.5.3 Three-wave interaction . . . . . . . . . . . . . . . . . 212
6.5.4 Frequency up-conversion . . . . . . . . . . . . . . . . 213
6.5.5 Parametric amplification and oscillation . . . . . . . . 214
6.5.6 Backward interaction . . . . . . . . . . . . . . . . . . 216
6.5.7 Second harmonic generation . . . . . . . . . . . . . . 217
6.5.8 The scattering matrix . . . . . . . . . . . . . . . . . . 219
6.5.9 Parametric down-conversion . . . . . . . . . . . . . . 220
6.5.10 Light scattering by polaritons . . . . . . . . . . . . . 225
6.5.11 Four-wave interactions . . . . . . . . . . . . . . . . . 226
6.5.12 Nonlinear spectroscopy . . . . . . . . . . . . . . . . . 228
6.5.13 Dynamical holography and phase conjugation . . . . . 229
7. Statistical Optics 237

7.1 The Kirchhoff law for quantum amplifiers . . . . . . . . . . . . 239
7.1.1 The Kirchhoff law for a single mode . . . . . . . . . . 239
7.1.2 The Kirchhoff law for a negative temperature . . . . . . 241
7.1.3 Noise of a multimode amplifier . . . . . . . . . . . . . 245
7.1.4 Equilibrium and spontaneous radiation;
superfluorescence . . . . . . . . . . . . . . . . . . . . 246
xviii Physical Foundations of Quantum Electronics
7.1.5 Gain and bandwidth of a cavity amplifier . . . . . . . . 248

7.1.6 The Kirchhoff law for a cavity amplifier.
The Townes equation . . . . . . . . . . . . . . . . . . 251
7.2 Basic concepts of the statistical optics . . . . . . . . . . . . . . 252
7.2.1 Analytical signal . . . . . . . . . . . . . . . . . . . . . 253
7.2.2 Random intensity . . . . . . . . . . . . . . . . . . . . 254
7.2.3 Correlation functions . . . . . . . . . . . . . . . . . . 256
7.2.4 Temporal coherence . . . . . . . . . . . . . . . . . . . 257
7.2.5 Spatial coherence . . . . . . . . . . . . . . . . . . . . 259
7.2.6 Coherence volume and the degeneracy factor . . . . . . 260
7.2.7 Statistics of photocounts and the Mandel formula . . . 262
7.2.8 Photon bunching . . . . . . . . . . . . . . . . . . . . . 265
7.2.9 Intensity correlation . . . . . . . . . . . . . . . . . . . 266
7.2.10 Second-order coherence (added by the Editors) . . . . . 270
7.3 Hamiltonian form of Maxwells equations . . . . . . . . . . . . 273
7.3.1 Maxwells equations in the k, t representation . . . . . . 273
7.3.2 Canonical field variables . . . . . . . . . . . . . . . . 278
7.3.3 Hamiltonian of the field and the matter . . . . . . . . 280
7.3.4 Dipole approximation . . . . . . . . . . . . . . . . . 283
7.4 Quantization of the field . . . . . . . . . . . . . . . . . . . . . 285
7.4.1 Commutation relations . . . . . . . . . . . . . . . . . 285
7.4.2 Quantization of macroscopic field in matter . . . . . . 287
7.4.3 Quantization of the field in a cavity . . . . . . . . . . . 288

7.5 States of the field and their properties . . . . . . . . . . . . . . 288
7.5.1 Diracs notation . . . . . . . . . . . . . . . . . . . . . 289
7.5.2 Energy states . . . . . . . . . . . . . . . . . . . . . . . 291
7.5.3 Coherent states . . . . . . . . . . . . . . . . . . . . . . 294
7.5.4 Coordinate and momentum states . . . . . . . . . . . . 298
7.5.5 Squeezed states . . . . . . . . . . . . . . . . . . . . . 302
7.5.6 Mixed states . . . . . . . . . . . . . . . . . . . . . . . 305
7.5.7 Entangled states (added by the Editors) . . . . . . . . . 310

7.6 Statistics of photons and photoelectrons . . . . . . . . . . . . . 314
7.6.1 Photon statistics . . . . . . . . . . . . . . . . . . . . . 314
7.6.2 Photon bunching and anti-bunching . . . . . . . . . . . 318
7.6.3 Statistics of photoelectrons . . . . . . . . . . . . . . . 323

7.7 Interaction of an atom with quantized field . . . . . . . . . . . 327
7.7.1 Absorption and emission probabilities . . . . . . . . . 328
7.7.2 Spontaneous emission . . . . . . . . . . . . . . . . . . 329
7.7.3 Interaction of stationary systems . . . . . . . . . . . . 331
Contents xix
7.7.4 Spectral representation . . . . . . . . . . . . . . . . . 333

7.7.5 Equilibrium systems. FDT . . . . . . . . . . . . . . . 335
Bibliography 337
Index 343
Chapter 1
Introduction
Quantum electronics studies the interaction of electromagnetic field with matter

in various wavelength ranges, from radio to X-rays and gamma rays. Investigation
of the basic laws of this interaction led to the creation of lasers, sources of coher-
ent (i.e., monochromatic and directed) intense light. Optimization of the existing
lasers and the development of new laser types, as well as advances in experimental
technology, in their turn, stimulated further development of quantum electronics.
This avalanche process, typical for modern science, led to new directions in op-
tics (nonlinear and quantum optics, holography, optoelectronics) and spectroscopy
(nonlinear and coherent spectroscopy), to numerous applications of lasers in tech-
nology, communications, medicine. We are probably close to solving the problem
of laser thermonuclear fusion and laser isotope separation on an industrial scale.a
Not so diverse but also important applications were found by the elder broth-
ers of lasers, masers, which operate in the radio range, at wavelengths on the order
of 0.1 10 cm, and are used as super-stable frequency etalons and super-sensitive
paramagnetic amplifiers.
The term quantum electronics appeared as a counterpart of classical elec-
tronics, mainly dealing with free electrons, which have continuous energy spec-
trum and, as a rule, are well described by classical mechanics. However, some
essentially quantum devices, such as, for instance, the ones based on the Joseph-
son junction, are traditionally not considered as part of quantum electronics. The
other name, quantum radiophysics, is not quite appropriate either, since it does
not relate to the optical frequency range.
a Editorsnote: This opinion was quite common in the laser physics community at the time when the
book was written. However, further investigations reduced the optimism in this field, and we are now
still witnessing new attempts towards laser thermonuclear fusion (inertial confinement fusion).
1
2 Physical Foundations of Quantum Electronics
1.1 Basic notions of quantum electronics
The operation of lasers and masers rests on the three whales, basic notions of
quantum electronics namely, stimulated emission, population inversion, and
feedback.
1.1.1 Stimulated emission

Stimulated emission leads to the multiplication of photons: a photon hitting an
excited atom or molecule causes, with a probability W12 , the transition of the atom
to one of its lower levels (Fig. 1.1). The released energy, E2 E1 , is transferred
to the electromagnetic field in the form of the second photon. This other pho-
ton has the same parameters as the incident photon, i.e., energy ~ = E2 E1 ,
momentum p = ~k and the same polarization type. Then, there are two indistin-
guishable photons, which can turn into four photons through the interaction with
other excited atoms. In the classical language, this picture corresponds to the ex-
ponential amplification of the amplitude of a classical plane electromagnetic wave
with frequency and wavevector k.b
Fig. 1.1 Amplification of light under stimulated transitions. A resonant photon hits an excited atom,
which then gives its stored energy to the field. As a result, the field contains two indistinguishable
photons.
1.1.2 Population inversion

Interaction with atoms that are at the lower level, with the energy E1 , occurs
through the absorption of photons, i.e., attenuation of the electromagnetic wave.
It is important that the probability W21 of this process (per one atom) is exactly
b See Editors note in Sec. 2.5.3.
Introduction 3
(a) (b)
Fig. 1.2 Obtaining population inversion through optical pumping: (a) initial Boltzmanns population
distribution; (b) strong resonant radiation balances the populations of levels 1 and 3, so that N2 > N1 .
equal to the probability of stimulated emission, W21 = W12 , and therefore the
overall effect depends on the difference of numbers of atoms at the levels 1 and 2,
N N1 N2 . Usually, populations Nm of the levels are defined per unit volume.
If the matter is at thermodynamic equilibrium with a temperature T , then, ac-
cording to Boltzmanns distribution, Nm exp(Em /T ), with being the Boltz-
mann constant. Therefore, if E2 > E1 , then N2 < N1 (Fig. 1.2(a)). As a result,
stimulated up transitions occur more frequently than stimulated down transi-
tions, and external electromagnetic radiation in equilibrium medium is attenuated.
Thus, in order to amplify field, the medium should be in a non-equilibrium state,
with N2 > N1 . One says that such a state has population inversion, or negative
temperature.
A lot of methods have been developed for achieving population inversion.
The most important ones are pumping the medium (Fig. 1.2(b)) with auxiliary
radiation (used for solid and liquid doped dielectrics), electric discharge in gases
and injection in semiconductors.
1.1.3 Feedback and the lasing condition

In order to turn an amplifier into an oscillator, one should provide positive feed-
back, which can be realized using a pair of plane or concave spherical mirrors. (In
masers, the active medium is placed into a microwave cavity.)
Amplification (or attenuation) can be quantitatively described as follows. Let
F [s1 cm2 ] be the flux density of photons propagating along the z axis. The
increment of F scales as the product of the stimulated transition probability per
unit time, W, and the number of active particles, N:

dF/dz = WN. (1.1)
In its turn, the stimulated transition probability scales as F,
W = F, (1.2)
where [cm2 ] is the probability of a transition per unit time for a photon flux with
unit density. It is called the interaction cross-section. As a result,
dF/dz = NF F, (1.3)
which leads to exponential intensity variation for a plane wave in matter (for
> 0, it is called the Bouguer law):
F(z) = F(0)ez, N. (1.4)
The parameter is called absorption (at > 0) or amplification (at < 0) coef-
ficient. Its inverse, 1 , has the meaning of the mean free walk of a photon. The
interaction cross section , in principle, can be as large as 32 /2 ( = 2c/ is
the wavelength), so that in the optical range, where 104 cm, it is sometimes
sufficient to have N 109 cm3 for noticeable amplification at a length of 1 cm.
Let the active medium of length l be placed between two mirrors (a Fabry
Perot interferometer) with reflection coefficients R1 , R2 . Then, from (1.4), the
threshold condition of lasing is
R1 R2 e2l = 1. (1.5)
For mirrors with dielectric coating, one can easily have R & 0.99, and for lasing
with l = 10 cm it is sufficient to have = (ln R)/l = 0.001 cm1 . Usually,
the radiation is fed out from the laser by making one of the mirrors have lower
reflection coefficient.
1.1.4 Saturation and relaxation

Let us consider some other important notions of quantum electronics. Saturation
occurs when the populations of some pair of levels become equal (N1 = N2 ) due
to stimulated transitions in a sufficiently intense external radiation. This effect re-
stricts and stabilizes the intensity of quantum oscillators and the gain coefficient
of quantum amplifiers. Relaxation processes counteract saturation and tend to re-
store the equilibrium Boltzmann distribution of populations, which is determined
by the temperature of the thermostat. Relaxation processes determine the lifetimes
of particles at different levels and the spectral linewidths.
Introduction 5
Even in the absence of incident radiation or other external influence, an excited

molecule can make a transition into one of the lower-energy states by emitting
a photon. This kind of emission is called spontaneous. Spontaneous emission
plays the role of a seed for self-oscillations in quantum oscillators, restricts their
stability, and creates noise in quantum amplifiers. Spontaneous and stimulated
transitions in equilibrium matter lead to thermal radiation, which is described by
Plancks formula and Kirchhoffs law.
It is important that while stimulated effects can be rather well calculated in
the framework of classical electrodynamics with deterministic field amplitudes
E, H, spontaneous effects are consistently described only by the laws of quantum
statistical optics, where E and H are random values or operators.
The above-mentioned terms and notions relate to different fields of theoretical
physics: quantum mechanics (energy levels, transition probabilities), statistical
physics (relaxation, populations, fluctuations), oscillation theory (feedback, self-
oscillations). Quantum electronics, as a field of physics, is remarkable and attrac-
tive because it uses theoretical and experimental tools from a diversity of fields,
and also because it poses new problems for these fields and provides them with
new experimental methods.
1.2 History of quantum electronics
Quantum electronics can be considered as a new chapter in the theory of light

and, more generally, in the theory of the interaction between electromagnetic field
and matter. The earliest chapters of this theory were devoted to the empirical
description of normal dispersion of light in the transparency ranges of the matter,
which was studied by Newton and his contemporaries more than 300 years ago.
The next steps, made in the 19th century, were the study of anomalous dispersion
within the absorption bands and the classical dispersion theory by Lorentz. The
quantum era in optics and generally in physics started at the beginning of the
20th century from Plancks theory of equilibrium radiation, which led Einstein to
the notion of photon, and from Bohrs postulates. Quantum theory of dispersion
was formulated in the 1920s by Kramers and Heisenberg. Meanwhile, Dirac,
Heisenberg, and Pauli developed quantum electrodynamics.
The history of quantum electronics, in its turn, is quite interesting and instruc-
tive [Dunskaya (1974)]. In principle, at the beginning of the 20th century the level
of laboratory technique was high enough for building, for instance, a gas laser.
However, this could not happen before the discovery of certain concepts and laws,
which form the base of a quantum generator.
1.2.1 First steps

The first step along this way, which took several decades, was made in 1916
by A. Einstein who introduced the notions of stimulated emission and absorp-
tion. A quantitative theory of these effects was developed about ten years later by
P. Dirac. From the theory, it followed that the photons generated via stimulated
emission have all their parameters (energy, propagation direction, and polariza-
tion) the same as the ones of the incident photons. This property is called the
coherence of stimulated emission.
The first experiments demonstrating stimulated emission were reported in
1928 by Ladenburg and Kopfermann. These experiments studied the refractive
index dispersion for neon excited by electric discharge. (Note that in the first gas
laser, which was built only 33 years later, neon was used as well.) In their paper,
Ladenburg and Kopfermann have accurately formulated the condition of popula-
tion inversion and the resulting necessity to selectively excite the atomic levels. In
1940, V. A. Fabrikant has pointed out, for the first time, that the intensity of light
in a medium with population inversion should increase. (He considered this effect
only as a proof for the existence of stimulated emission but not as a phenomenon
that can have useful applications.) Unfortunately, this paper, as well as an applica-
tion for an invention filed by V. A. Fabrikant and his colleagues in 1951, was not
properly published in time and therefore did not influence further development of
quantum electronics.
1.2.2 Radio spectroscopy

The first devices of quantum electronics, masers, which were later used in ap-
plications such as generation and amplification of waves in the centimeter range,
were developed only in the middle of the 1950s. Remarkably, quantum electronics
has first conquered the radio range; lasers appeared at the beginning of the 1960s.
This is partly because in usual optics experiments, N1 N2 , and therefore stim-
ulated emission, as a rule, plays no role. At the same time, in radio spectroscopy,
N1 N2 |N1 N2 |, and the observed absorption of radio waves is caused by
the stimulated absorption slightly exceeding the stimulated emission.
An important role was also played by the advanced development of radio spec-
troscopy in the 1940s, in both theory and experiment. (Experimental base for
microwave radio spectroscopy was provided by the development of radar tech-
nique.) By that time, the theory of radio waves interaction with gas molecules
was developed, the structure of rotational spectra was calculated in detail, the role
of relaxation and saturation was understood. Of considerable importance were
Introduction 7
investigations with beam radio spectroscopes, which had been started as early as
in the 1930s. Probably, it was also important that radio spectroscopists, in contrast
to opticians, understood very well the operating principles of MW generators and
amplifiers based on free-electron beams (klystrons, magnetrons, traveling-wave
and backward-wave tubes), they were familiar with the notions of negative resis-
tance and positive feedback, and had practical experience with high-quality MW
cavities.
Among the works directly preceding the advent of masers, one should mention
the ones by Kastler (France), who developed in 1950 the optical pumping method
for increasing the population inversion of close levels in gases. Besides gas and
beam radio spectroscopy, an important role was also played by magnetic radio
spectroscopy, a direction that was started in the 1940s and studied the interac-
tion of radio waves with ferromagnetics and nuclear or electronic paramagnetics
(E. K. Zavoisky, 1944). These are namely the achievements in the theory and
technique of magnetic resonance that led to the development of paramagnetic am-
plifiers, which have an extremely small level of noise. Population inversion has
been first obtained in a system of nuclear spins placed into magnetic field (Parcell
and Pound, 1951).
1.2.3 Masers
The idea of using stimulated emission in a medium with population inversion for
the amplification and generation of MW electromagnetic waves was suggested
at several different conferences at the beginning of the 1950s by N. G. Basov
and A. M. Prokhorov (Lebedev Physics Institute, Academy of Sciences, USSR),
C. H. Townes (Columbia University, USA), and J. Weber (University of Mary-
land, USA). The first quantitative theory of a quantum generator was published
by Basov and Prokhorov in 1954. They have found the threshold population dif-
ference necessary for self-excitation and suggested a method for obtaining popula-
tion inversion in a molecular beam using inhomogeneous electrostatic field. Later,
Basov, Prokhorov, and Townes were awarded a Nobel Prize for their contributions
to the development of quantum electronics.
In 1954, description of the first operating maser was published by Gordon,
Zeiger, and Townes. The active medium was ammonium molecular beam, focused
with the help of electric field. Nowadays, beam masers are used in the national
standards of frequency and time.
The second basic maser type, paramagnetic amplifier, was created in 1957
by Scovill, Feher, and Seidel who followed a suggestion by Bloembergen. In
paramagnetic amplifiers, population inversion is created with the help of auxiliary
radiation, the pump, which saturates the populations of levels 1 and 3 (Fig. 1.2).
As a result, levels 1 and 2 (or 2 and 3) get population inversion. The idea of pump-
ing a three-level system, which was later widely used in solid-state and liquid
lasers, belongs to Basov and Prokhorov (1955). The active medium of paramag-
netic amplifiers, which is a diamagnetic crystal doped with a small amount (on the
order of 103 ) of paramagnetic atoms, i.e., atoms with odd electron numbers, is
cooled down to helium temperatures. Cooling is necessary for reducing the noise
and slowing down the relaxation processes, which counteract the population in-
version. (In paramagnetics, relaxation of populations is caused by the interaction
between crystal lattice vibrations and the magnetic moments of non-compensated
electrons.)
1.2.4 Lasers
Transition from radio to the optical frequencies took about five years: the first
operating laser emitting coherent red light was described by Maiman in 1960. As
the active medium, the laser used a pink ruby crystal (aluminium oxide doped
with chrome) and population inversion was achieved using blue and green light
from a pulsed flash lamp. An important step was realizing that a Fabry-Perot
interferometer, i.e., two parallel plane mirrors, is a high-quality resonator, i.e., an
oscillation system for light waves (Prokhorov, Dicke, 1958).
The laser era of physics started. Soon after the appearance of solid-state lasers
with optical pumping, a number of other laser types was developed: gas discharge
lasers (1961), semiconductor lasers based on p n transitions (1962), liquid lasers
based on the solutions of organic dyes (1966). Rather quickly, the wavelength
range from far infrared (IR) to far ultraviolet (UV) was covered. The parameters
of the lasers (power, monochromaticity, directivity, stability, tunability) were con-
tinuously improving; their field of application rapidly broadened. An important
role was played by the invention of methods to shorten the duration of laser light
pulses (q-switching and mode locking).
First experiments on light frequency doubling (Franken et al., 1961) started the
explosive development of nonlinear optics, which studies and uses the nonlinear-
ity of the matter at optical frequencies. Holography and optical spectroscopy had
their second birth; new fields appeared, such as optoelectronics, coherent spec-
troscopy, and quantum optics. X-ray and gamma-ray lasers are to arrive soon.c
It should be stressed once again that the rapid development of quantum elec-
tronics was provided by a large amount of ideas and information stored by the
c Editors note: While X-ray lasers have been indeed constructed in the end of the 20th century [Svelto
(2010)], making a gamma-ray laser is still a challenge.

Introduction 9
beginning of the 1950s in the fields of radio and optical spectroscopy. Such di-
rections of physics as magnetic resonance or molecular-beam spectroscopy, seem-
ingly far from practical applications, led to a laser revolution in many fields of
science and technology.
1.3 Recent progress in quantum electronics (added by the Editors)
This textbook was published in 1987, almost a quarter century ago. At that time, it
was a very modern book; it reflected the latest events in quantum electronics and
provided a complete picture of its directions and tendencies. Since then, many
changes took place in this field. New technologies appeared, new laser sources
were developed, and new effects were discovered. In this section, we will try to
briefly review the advances in quantum electronics that happened after the book
had been published.
1.3.1 Physics of lasers

During the last two decades, important progress has been achieved in laser tech-
nology, and all parameters of lasers have been considerably improved. Mean pow-
ers of laser radiation achieved at present amount to hundreds of kW, while peak
powers reach the petawatt (1015 W) range. Such radiation provides the values
of electric and magnetic fields comparable to atomic ones and threfore opens a
perspective for observing principally new effects in optics and particle physics.
The ultra-fast laser technology is now capable of producing pulses as short as tens
of attoseconds, containing only few optical cycles. The spectral range covered
by modern commercial laser systems, in particular, achieved by continuous fre-
quency tuning, is from vacuum UV (about 100 nm) to mid-IR (tens of microns).
These achievements became possible due to both the development of existing
methods, such as frequency conversion, generation of higher optical harmonics,
mode locking etc., and the discovery of new technologies. In particular, dye-laser
systems were gradually replaced by solid-state ones. The most famous among
them are titanium-sapphire lasers and similar systems, providing ultra-short pulse
generation, as well as optical combs, via mode locking. Huge progress has been
achieved in the development of semiconductor lasers. A totally novel step in laser
technology, with respect to the 1980s, was the invention of fibre laser systems,
which can have extremely high efficiency and therefore provide record output
powers.
Apparently, lasers became widely used devices which penetrate into all fields
of human activity starting from toys up to the high technologies and medicine.
1.3.2 Laser physics

Laser physics, or research in physics essentially based on the use of lasers, under-
went considerable progress as well. Modern laser physics covers several branches
of science and various applications like nonlinear and quantum optics, fiber op-
tics, optical pulse shaping, optoelectronics (including integrated optics), optical
communications, different aspects of general optics etc. New directions appeared,
such as, for instance, high resolution spectroscopy or atom optics. Some of the
new directions will be discussed in more detail below; the rest will be briefly men-
tioned here. Application of laser methods to metrology resulted in the develop-
ment of caesium atomic clock to a high-technology level; recently, this device has
been made on a chip and is now available as a consumer product. Laser methods
became extremely helpful in the manipulation with microscopic and nanoscopic
objects; in particular, the technique of laser tweezers enables trapping and dis-
placing small particles, including biological objects. Laser cooling of atoms and
molecules is another example of progress in laser physics. Finally, lasers are now
widely used in the technique of scanning near-field optical microscopy (SNOM),
which successfully complemented the existing methods of scanning tunnel mi-
croscopy and atomic-force microscopy.
1.3.3 New trends in nonlinear optics

Huge progress in nonlinear optics is due to the development of the material sci-
ence, which led to the production of new nonlinear optical materials. Among
them, there were newly synthesized crystals with high nonlinear susceptibilities
and broad transparency range, such as BBO, LBO, KTP, and many others. Further
opportunities in realizing various types of phase matching were provided by the
use of spatially inhomogeneous structures such as periodically and aperiodically
poled crystals, photonic crystals and microstructured fibres (photonic-crystal
fibres). The opportunities offered by such structures are: making use of new com-
ponents of nonlinear susceptibility tensors, non-critical phase matching and si-
multaneous phase matching for different nonlinear processes, as well as processes
in different frequency ranges.
One of the novel trends is development of integrated nonlinear optics. Due
to the miniaturization of optical elements, involving fibre optics and waveguide
structures, it became possible to realize most of nonlinear optical processes on
Introduction 11
a chip. Optical fibres are now used not only for light transmission, but also for
beam splitting, polarization transformations, as nonlinear elements and as active
elements [Agrawal (2007)]. Nonlinear waveguides, based on KTP and lithium
niobate crystals, and sometimes on semiconductor layers, are used as extremely
efficient and compact elements for frequency conversion, requiring very low pump
powers and allowing for relatively easy control. Integrated optics also uses plas-
monic structures, which form convenient interfaces between free space or di-
electrics and metal surfaces.
We now witness a certain shift of interest to novel frequency ranges. Among
them, attention is drawn to the terahertz (1012 Hz) range of frequencies, which is
important for spectroscopic studies in biology, for astronomy, and for the security
applications (detection of explosive materials and weapons). For more details
on the recent developments in nonlinear optics, one can see, for instance, [Boyd
(2008)].
1.3.4 Atom optics

A completely novel direction that appeared in the end of the 20th century is atom
optics, i.e., manipulation of individual atoms by means of laser beams. It is worth
noticing that manipulating single quantum objects characterizes the modern devel-
opment of quantum electronics and, probably, physics in general compared with
the last century when the ensemble approach dominated.
Forces acting on atoms due to the gradients of light intensity turn a standing
wave into a scatterer for atomic beams, causing diffraction, interference, and trap-
ping. Trapping of ions and atoms enables one to address these quantum objects,
single ones or in an array, and control their quantum state. In particular, it is possi-
ble to organize the interaction between single material quantum objects and single
photons. This is extremely important both in fundamental research and for various
applications like quantum information.
Furthermore, the effect of Bose-Einstein condensation, predicted as early as
in 1925, has been observed in 1995. A Bose-Einstein condensate (BEC), a large
group of atoms described by a single wave function, is one of the few examples
of a macroscopic object manifesting quantum behavior. Similarly to single atoms
and ions, a BEC can be manipulated by means of laser beams.
1.3.5 Optics of nonclassical light

Quantum optics, started by the famous Hanbury BrownTwiss experiment
(Sec. 7.2) in 1956, had explosive development in the end of the last century. New
types of nonclassical light have been generated. In addition to single-photon and

two-photon Fock states in superposition with the vacuum (Sec. 7.5), higher-order
Fock states can be conditionally prepared now by using spontaneous parametric
down-conversion [Bouwmeester (2000); Mandel (2004)]. The spectral and spa-
tial structure of such states has been studied in detail, as well as their polarization
properties. The concept of squeezed states (Sec. 7.5), which were first observed
about the same time as the book was published, and the idea of shot-noise sup-
pression [Yamamoto (1999)], were since then considerably developed. Squeezed
states became one of the main instruments of experimental quantum optics [Ba-
chor (2004); Walls (1994)], together with the two-photon states (photon pairs).
The phenomenon of polarization squeezing was observed and studied. Finally,
various types of entangled states [Scully (1997); Mandel (2004); Bouwmeester
(2000)], both faint (few-photon) and bright ones, based on quadrature squeezing,
were generated, and numerous experiments on testing Bells inequalities [Gryn-
berg (2010); Scully (1997); Mandel (2004); Klyshko (1998)] were carried out.
New sources of nonclassical light were discovered. Since the beginning of
the 21st century, optical fibres have been used as a very reliable and efficient
source of both squeezed states and photon pairs. This source is based on the cubic
susceptibility (Kerr nonlinearity), and the corresponding nonlinear optical effect
is spontaneous four-wave mixing (originally called hyper-parametric scattering,
Sec. 6.5). By applying fibres with specially tailored dispersion dependence, which
can be achieved by modifying the structure, by doping, or by tapering, one can
fully control the phase matching and provide its new types [Agrawal (2007)].
Photon pairs and squeezed light are also generated in waveguide structures having
high efficiency, compact sizes, and controllable properties. In addition, modern
sources of nonclassical light include nano- and micro-emitters such as quantum
dots, vacancies and color centers in diamond, and others. These sources are in a
sense similar to single atoms, which were used for generating nonclassical light in
the 1960s and the 1970s; however, an important advantage of solid-state emitters
is much easier handling, including preparation and control.
Huge progress has been made in the development of the detection tech-
niques [Leonhardt (1997)]. The only type of photon-counting detector mentioned
in the book is a photomultiplier tube (PMT); nowadays, much more common for
single-photon counting are avalanche photodiodes (APDs) operating in the Geiger
mode. Such detectors provide quantum efficiencies of up to 60% and time reso-
lution of about 50 ps in the visible (Si-based APDs) and near-IR (InGaAs- or
Ge-based APDs) ranges while having relatively low dark noise (up to tens of pA).
Other types of single-photon detectors appeared quite recently, namely, super-
conducting photodetectors, which can operate in the IR and even terahertz range,
Introduction 13
and transition-edge sensors (TES), capable of photon-number resolution. The lat-

ter possibility, nearly impossible at the time when this book was written, is also
achieved by combining single-photon counting with time or space multiplexing.
Finally, the technique of homodyne detection, which is hardly mentioned in the
book, has been hugely developed during the last two decades. Using this tech-
nique, it is possible not only to measure the distributions of coordinate and mo-
mentum for various quantum states (Sec. 7.5), but also to reconstruct the quasi-
probability distributions, such as Wigner or Husimi functions [Schleich (2001);
Bachor (2004)].
Probably the most important event in the development of quantum optics is
its application to quantum information, a field that emerged in the end of the 20th
century at the boundary of quantum mechanics, mathematics, and information sci-
ence [Nielsen (2000)]. Along with the quantum metrology, which is briefly men-
tioned in the book, quantum information and quantum communication technolo-
gies became a real practical output of quantum optics, which at first looked like
nothing but a collection of beautiful fundamental experiments. In quantum metrol-
ogy, in addition to the absolute calibration methods (Sec. 7.6), which were devel-
oped in the 1980s, there appeared the techniques of super-resolution and precise
positioning [Bachor (2004)] based on squeezed light or high-order Fock states. A
lot of experimental techniques, developed earlier in quantum optics for nonclas-
sical state generation, transformation and measurement, were simply transferred
to quantum communication. In quantum communication, various states of light
are used as information carriers, from qubits (quantum information bits), qutrits,
ququarts, and high-dimensional qudits to entangled states formed by these ele-
mentary carriers [Bouwmeester (2000); Nielsen (2000)]. Transformations of these
states by linear optical elements, as well as interactions between these states, can
form the basis for quantum gates, which, in their turn, may in the nearest fu-
ture become the key elements of a quantum computer [Nielsen (2000)]. Different
approaches to the measurement of quantum states serve as a powerful tool for
quantum state tomography and quantum process tomography. Finally, the most
advanced branch of quantum information is quantum key distribution, in which
a secret encryption key is distributed between several communicating parties in
such a way that eavesdropping is not possible due to the fundamental laws of
quantum physics.d
d This is true provided that the unavoidably introduced error rate exceeds some critical level, depending
on the specific type of protocol used.


Chapter 2
Stimulated Quantum Transitions
The most important notion in quantum electronics is the probability for an electron
in an atom or a molecule to make a quantum transition from one level to another.
In this chapter, we will first give the general expression for the probability of a
quantum transition in the first order of the perturbation theory (Sec. 2.1), then
calculate the probability of a transition due to monochromatic radiation (Sec. 2.2)
and find the interaction cross-section and the absorption coefficient (Sec. 2.3).
Further, we will consider stimulated transitions under fluctuating (noise) radiation
with a broad spectrum (Sec. 2.4). Noise radiation surrounding an atom can play
the role of a thermostat and cause relaxation (Sec. 2.5).
A consistent theory of electromagnetic processes should describe both the
matter and the field based on the principles of quantum mechanics. However,
most part of quantum electronics effects are sufficiently well described by the so-
called semiclassical theory of radiation, in which only the motion of particles
is quantized while the electromagnetic field is considered in terms of classical
Maxwells equations. By avoiding quantum electrodynamics, one gets the the-
ory considerably simplified but, at the same time, loses the chance to consistently
describe fluctuations of the electromagnetic field and, in particular, spontaneous
emission and the noise of quantum amplifiers. The present book mainly considers
stimulated effects in a classical deterministic field and therefore uses the semi-
classical theory of radiation. Quantization of the field and spontaneous effects are
considered in Chapter 7.
2.1 Amplitude and probability of a transition
In the simplest model of quantum electronics, matter is assumed to consist of sep-

arate non-interacting motionless atoms or molecules in external electromagnetic
field. Our first task is to find out what happens with a given atom in a given al-
15
ternating field E(t). (Usually the effect of the magnetic field is much weaker than
the one of the electric field.) At the second stage, we will find the back action of
the atoms on the field. The self-consistent solution to the two systems of equa-
tions describing the response of the matter to the field and the response of the field
to a given motion of charges, under certain simplifying conditions, is the main
problem in the theory of interaction between radiation and matter.
The behavior of material particles in given external fields is described by the
Schrodinger equation,
(i~/t H)(r, t) = 0. (2.1)
Here, is the wave function, whose arguments are the set of coordinates r =
{r1 , r2 , ...} and the time; H is the energy operator consisting of the non-perturbed
part, H0 , and the alternating energy of the particles in the external field, V(t),
H(t) = H0 + V(t). (2.2)
The non-perturbed energy, in its turn, includes the kinetic energy of the particles
and the energy of their interaction V0 . (The latter also includes the energy of the
particles in external static fields).
2.1.1 Unperturbed atom

In the absence of the alternating field, the wave function can be represented as
X
(0) (r, t) = c(0)
n n (r, t), (2.3)
n
n (r, t) = n (r)exp(iEn t/~), (2.4)

where En and n (r) are the eigenvalues and the eigenfunctions of H0 , satisfying
the stationary Schrodinger equation,
(H0 En )n = 0. (2.5)
The index n numerates the energy levels. (We assume that the particles move
within a bounded space domain and therefore the levels are discrete; we also as-
sume the levels to be non-degenerate.) The set of functions {n } is assumed to be
orthogonal and normalized,
Z
drn m = nm , (2.6)
so that
Z X
dr||2 = |cn |2 = 1. (2.7)
n
Stimulated Quantum Transitions 17
The cn coefficient in the expansion (2.3) gives the relative population |cn |2 of
the level n, i.e., the probability to measure the energy En or, as one says, the
probability to find the system at the level n. Indeed, according to the rule of
calculating mean values in quantum mechanics, the mean energy of the system,
with an account for Eqs. (2.3)(2.6), is
Z X
E hH0 i dr H0 = |cn |2 En . (2.8)
n
Note that, according to (2.3), in the general case the atom is not necessarily in
a stationary state with a definite energy En (even in the absence of the alternating
force, V(t) = 0). For instance,let only two coefficients cn of the superposition
(2.3) be nonzero: c1 = c2 = 1/ 2; then the mean ensemble energy of the atom is
(E1 + E2 )/2 but single energy measurements will give either E1 or E2 . Then the
electron cloud, i.e., the probability density to find the electron at point (r, t), will
oscillate with the Bohr frequency, 21 2 1 (E2 E1 )/~ (Fig. 2.1):
P(r, t) = |(r, t)|2 = |1 (r) + 2 (r)exp(i21 t)|2 /2
= 21 /2 + 22 /2 + 1 2 cos(21 t). (2.9)
(We assume that n = n .) Such nonstationary states are called coherent ones.
This term is often used in the case where many identical atoms are in a non-
(a) (b)
Fig. 2.1 Electron cloud of an atom that is in a coherent (non-stationary) state given by a superposition
of two stationary states 1 and 2 with different symmetries oscillates with the transition frequency
21 : (a) dependencies of the wave functions on one of the space coordinates; (b) corresponding con-
figurations of the electron cloud.
stationary state with the same phase. Then, electrons oscillate synchronously
and the system of atoms has a macroscopic dipole moment emitting intense light
with the frequency 21 . This effect, called superradiance, will be considered in
Sec. 5.3.
In the presence of external alternating field E(t), eigenoscillations of the elec-
tron cloud with the frequencies mn will be accompanied by stimulated oscilla-
tions with the frequency of the field .
2.1.2 Atom in an alternating field

Consider now the effect of an alternating field on the wave function (r, t) of
an atom or a molecule. At V(t) , 0, the function (2.3) does not satisfy the
Schrodinger equation (2.1) any more, but the expansion can be kept in the form
(2.3) if the coefficients cn are considered as time-dependent,
X
(r, t) = cn (t)n (r, t). (2.10)
n
(This possibility follows from the completeness of the eigenfunctions set n (r).)
Thus, due to the effect of the incident light, the relative populations |cn (t)|2 of
the levels are redistributed (with the normalization condition (2.7) maintained). In
other words, the atom makes stimulated transitions between the levels. Let us find
the probability of such transitions.
From the Schrrodinger equation (2.1) for the wave function, we will pass to
equations for cn (t). For this purpose, let us substitute expansion (2.10) in (2.1) and
take into account that, according to (2.5), i~n = H0 n :
X
(i~cn cn V)n = 0. (2.11)
n
Left-multiplying this equality by one of the functions m and integrating w.r.t. r,
we obtain
X Z Z
(i~cn drm n cn drm Vn ) = 0.

(2.12)
n
Let us take into account the orthogonality (2.6) of the eigenfunctions and introduce
the following notation for the matrix elements of the perturbation operator:
Z Z
0
Vmn drm Vn = Vmn exp(imn t), Vmn drm Vn . (2.13)
As a result, we find the system of equations for the coefficients cn (t), which is
equivalent to the initial Schrodinger equation:
X
0
i~dcm /dt = Vmn cn . (2.14)
n
Note that the coefficients cn form a function of a discrete argument (energy),

cn (En ), which can be considered as the wavefunction of the system in the
energy representation (while (r) is the wave function in the coordinate repre-
sentation). Correspondingly, (2.14) is the Schrodinger equation in the energy rep-
resentation. The functions and are related through a linear transformation
and provide the same information. The inverse of transformation
R (2.10) can be
obtained by left-multiplying it by the integral operator drm :
Z
cm = drm . (2.15)
The change of representation, , is similar to the change of the basis in

vector algebra, where the components of a vector are also linearly transformed.
The relation between different representations is most clearly manifested in
Diracs notation (Sec. 7.5). In this notation, the Schrodinger equation (2.1) is
written in the invariant form (without specifying the representation) as
i~d|ti/dt = H|ti. (2.16)
In order to pass to the energy representation, let us left-multiply (2.16) by the m-th
eigenvector of H0 ,
d X
i~ hm|ti = hm|H|ti = hm|H|nihn|ti. (2.17)
dt n
P
The last equation was obtained using the expansion of the unity, I = |nihn|. Let
us denote hm|V|ni Vmn and use (2.2) and (2.5), then
d X
i~ hm|ti = Em hm|ti + Vmn hn|ti.
dt n
Finally, if we separate the slowly varying part of the hm|ti factor,
hm|ti cm (t)exp(iEm t/~),
we once again obtain (2.14).
2.1.3 Perturbation theory

In the general case, the solution to the system (2.14) can be found using the pertur-
bation theory, as a series expansion in the external force. Alternatively, the system
can be solved without the perturbation theory, using the so-called two-level ap-
proximation, which will be considered below, in Sec. 4.3.
Thus, we look for the solution to (2.14) in the form of the sum
X
cn = c(0)
n + c (1)
n + = c(s)
n , (2.18)
s
in which the s-th term is proportional to the s-th power of the external force,
c(s) s
n V . Substituting (2.18) in (2.14) and setting equality between the terms of
the same order in V, we find
c(0)
m = 0, (2.19)
X
0
i~c(1)
m = Vmn c(0)
n , (2.20)
n
X
0
i~c(s)
m = Vmn c(s1)
n . (2.21)
n
The set of zero-order coefficients c(0)

m provides the initial conditions for equations
(2.14).
2.1.4 Linear approximation

Usually, it is assumed that only one of the coefficients, for instance, c(0)
1 , is
nonzero, so that at time t0 the system is in a state with a given energy,
cn (t0 ) = c(0)
n = n1 . (2.22)
In this case, the system of equations (2.20) for the first-order coefficients gets
decoupled,
0
c(1)
m = Vm1 /i~. (2.23)
Hence, it follows that in the linear approximation, the response of a quantum

system that is initially at level 1, to an external perturbation is given by the formula
1 t 0
Z
c(1)
m (t) = dt Vm1 (t0 )exp(im1 t0 ). (2.24)
i~ t0
Thus, at large time delays after the perturbation has been switched on (t t0
), the c2 coefficient, which determines the perturbed population of the level with
the energy E2 , scales as the Fourier transform of the external force at the Bohr
frequency 21 = (E2 E1 )/~. In other words, in the first-order perturbation
theory, a quantum system behaves as a set of linear oscillators and responds only
to the resonant harmonics of the external force. If there is no 21 harmonic in the
perturbation spectrum, or its amplitude is small, then level 2 will not be populated.
2.1.5 Probability of a single-quantum transition

Given the initial condition c(0) m = mn , the dimensionless complex number cm is
called the amplitude of the transition from level n to level m. (Often, the second
subscript is added to specify the initial condition, cm cmn .a ) The squared mod-
ulus of the transition amplitude, |cmn (t)|2 , is equal to the conditional probability
of finding the system at the level m at time t provided that at time t0 it was at the
level n. In the linear (single-photon) approximation, this probability, according to
(2.24), is
Z t
2
P(1)
mn P (1)
(m, t|n, t0 ) = ~ | dt0 Vmn (t0 )exp(imn t0 )|2 . (2.25)
t0
From (2.25), an important property of quantum systems follows: the proba-
bility of forward and backward transitions are equal:
P(1) (1)
mn = Pnm . (2.26)
Indeed, the V operator corresponds to an observable quantity, the energy, and
therefore its mean values are real, hVi = hVi . Hence, this operator is Hermi-
tian (self-conjugate), V = V . Matrix elements of Hermitian operators have the

property Vmn = Vnm ; therefore, (2.25) leads to (2.26).
Note that for the transition n m to be enabled, it is necessary, in addition to
the resonance condition, that its matrix element is nonzero,
Z
Vmn = drm Vn , 0.
This requirement, providing the selection rules, forbids some transitions for
high-symmetry systems. For instance, if n (r) = n (r) (central symmetry)
and V r (dipole approximation), then transitions between states with the same
parity are forbidden (since the integrand is odd in this case). If Vmn , 0, one says
that the perturbation couples or mixes the states m and n.
2.2 Transitions in monochromatic field
2.2.1 Dipole approximation

Let us apply the general formula (2.25) to the case of a harmonic perturbation.
Most problems of quantum electronics allow the dipole approximation for the
energy of the interaction between charges and field (Sec. 7.3),
V = d E = d (E0 eit + E0 eit )/2. (2.27)
a Recall that in quantum mechanics, the subscripts are often read from right to left; therefore, c
mn is
the amplitude of the transition n m.
Here, the amplitude E0 of the electromagnetic wave is assumed constant within

the system considered, an atom or a molecule, since in the optical frequency range
104 cm a0 108 cm, where a0 is the Bohr radius.
In the case of a single-electron atom, the dipole moment operator d is equal
to the product of the electron charge and its radius vector, d = er, so that the
matrix elements are
Vmn = dmn (E0 eit + c.c.)/2, (2.28)
Z
dmn = e d3 rrm n , (2.29)
where c.c. means the complex conjugated expression. For allowed transitions, the
integral in (2.29) is on the order of the atom size a0 , so that
dmn ea0 1018 CGS 1D. (2.30)
In the case of magneto-dipole transitions, which are used, in particular, in param-
agnetic amplifiers, the electric field E0 in (2.28) should be replaced by the mag-
netic field H0 and d should be replaced by the magnetic dipole moment whose
absolute value is on the order of Bohrs magneton,
mn 0 e~/2mc 0.9 1020 CGS. (2.31)
2.2.2 Transition probability

Substituting (2.28) in (2.25), for t0 = 0 we find
exp[i(mn )t] 1 exp[i(mn + )t] 1 2

1
Pmn = 2 dmn E0 + dmn E0 .
4~ mn mn +
(2.32)
Further, let us consider stimulated transitions up (mn > 0) or down (mn < 0)
under the condition that the frequency of the field ( > 0) is resonant to the
transition:
|mn | | |mn ||. (2.33)
Then, one of the terms in (2.32) is much larger than the other one, so that the latter
can be neglected (the so-called rotating-wave approximationb). For instance, stim-
ulated absorption is mainly determined by the first term, which is proportional to
the positive-frequency part of the field E0 eit /2, while for stimulated emission,
it is the second term that matters, proportional to the negative-frequency part,
E0 eit /2. Note that in quantum electrodynamics, the amplitudes E0 , E0 are oper-
ators scaling as the photon annihilation and creation operators, a, a , see Sec. 7.4.
b The title comes from the fact that in the complex plane, the eit vector is rotating, in contrast to the
vector cos t, which oscillates.

Fig. 2.2 Probability Pmn of a stimulated transition as a function of the field frequency and the
interaction time t.
Thus, in the first-order perturbation theory, the probability to find the atom at
level m at time t, under the initial condition cn (0) = 1, is
#2
dmn E0 2 sin(t/2)
"
Pmn = , (2.34)
2~ /2
where |mn |. According to this formula, the dependence of the transition
probability on the external field frequency is resonant, and the sharpness of the
resonance increases with time (Fig. 2.2). As a result, in the limit t , the
transition probability is given by Diracs delta function,
" #2
sin(t/2)
lim = 2t(). (2.35)
t /2
The factor by the delta function is verified by integrating both parts of (2.35) in .
Because the transition probability scales as the time duration t of the pertur-
bation, one can introduce the transition probability per unit time, also called the
transition rate,
Wmn Pmn /t = 2|dmn E0 /2~|2(). (2.36)
Thus, the transition rate scales as the square of the field, i.e., the intensity of the
wave. The presence in (2.36) of the delta function, which differs from zero only at
exact resonance, can be understood from the photon viewpoint: according to the
energy conservation law, the change in the atom energy by the value Em En
~mn should be accompanied by the absorption or emission of a photon with the
energy ~.
2.2.3 Finite level widths

In reality, however, there always exist additional perturbations, for instance, col-
lisions in the case of a gas. These perturbations broaden the energy levels and, as
a result, even at t the resonance has a finite width . (In the case of colli-
sions, = 2/, where is the mean time interval between the collisions.) The
finite width of the resonance is taken into account by changing the delta function
in (2.36) to a form factor g(), which describes the true shape of the spectral line
and is also normalized to unity:
Wmn = 2|dmn E0 /2~|2g(), (2.37)
Z
dg() = 1. (2.38)

If the broadening is due to collisions, spectral lines have Lorentzian shapes

(Fig. 2.3, curve 1):
2/()
gL () = . (2.39)
1 + (2/)2
The transition probability is maximal at = |mn |:
W0 = 2 /, (2.40)
where |dmn E0 |/~ is the Rabi frequency. It has the dimensionality of fre-
quency and characterizes the perturbation of an atom by resonant monochromatic
field.
Thus, the probability of a stimulated transition scales as the intensity of light,
squared matrix element of the dipole moment, and the inverse width of the spectral
line.
This dependence on the spectral line width is typical not only for collision
broadening: from the normalization condition (2.38) it follows, in the general
case, that g(0) 1/. For instance, in gases at low pressure is often deter-
mined by the Doppler effect, which results in a Gaussian line shape. Figure 2.3
shows the comparison of spectral line shapes due to collision broadening (2.39),
the Doppler effect,
!1/2 " 2 #
2 ln 2
gD () = exp 4 ln 2 , (2.41)

and finite interaction time t (c.w. Eq. (2.34)),
!
0.886 2 2.78 5.56
gt () = sinc , = (2.42)
t
Fig. 2.3 Spectral line shapes. Broadening due to collisions or spontaneous radiation leads to a
Lorentzian line shape (1); the Doppler effect results in a Gaussian line shape (2), while limited in-
teraction time t causes a sinc2 (t/2) shape (3).
where sinc x (sin x/x) and is the full line width at level 1/2. It is clear from
Fig. 2.3 that the particular mechanism of broadening has a noticeable effect only
on the tails of the lines. Note that the amplitude ratio of the area-normalized form
factors (2.39), (2.41), (2.42) at exact resonance and with the same is
!1/2
2 4 ln 2 5.56
: : 1 : 1.47 : 1.39. (2.43)
2
When deriving (2.37), we have replaced the function (2/)2 sin2 (t/2), which
was obtained from the perturbation theory, by 2t g(). (Here, is the frequency
mismatch between the field and the transition.) Let us explain this procedure using
the strong collision model. According to this model, every collision instantly
brings the atom back to the initial level, after which the interaction between the
field and the atom starts anew. Then, t in (2.34) should be replaced by t t0 t,
where t0 is the time instance of the last collision. In gas, the time interval t
between the last collision of an atom and some fixed time t is a random variable
with exponential distribution (see [Rytov (1976)]):
P(t) = exp(t/)/, (2.44)
where is the mean time interval between the collisions.

The power absorbed (or emitted) by the atom scales as the instant transition
rate at a given time t,
W(t) dP/dt = 2 sin(t)/2. (2.45)

Here, we have used expression (2.34) for the transition probability P. The mean
transition rate is given by averaging (2.45) using (2.44),
Z
2 /2
W d(t)P(t)W(t) = . (2.46)
0 1 + 2 2
The last expression is in agreement with (2.37) and (2.39) if is replaced by the
inverse half-width of the line, 2/. The integral in (2.46) can be easily done by
replacing sin x with Im[exp(ix)].
2.3 Absorption cross-section and coefficient
2.3.1 Relation between intensity and field amplitude

In order to pass from the transition rate W to the transition cross-section = W/F
and the absorption (or amplification) coefficient = N, we have to write the
squared field |E0 |2 in terms of either the photon flux density F [s1 cm2 ] or
intensity I = ~F [W/cm2 ].
Let us first find the energy E of the wave. From Maxwells equations, it fol-
lows that the instant energy of the field contained in a volume V of transparent
isotropic non-magnetic material with the dielectric constant is
Z
E(t) = d3 r(E 2 + H 2 )/8. (2.47)
V
In the case of a plane monochromatic wave,
E = (1/2)eE0ei(krt) + c.c., H = nk E. (2.48)
Here, e is the unity polarization vector, k = kn/c is the wave vector, k is the

unity vector in the propagation direction, n = is the refractive index. After
substituting (2.48) in (2.47) and time averaging, we get the following relationship
between the time-averaged energy and the amplitude of the plane wave:
E E(t) = n2 V|E0 |2 /8. (2.49)
Intensity of the wave is obviously given by the product of the energy density,
E/V, and the velocity of the wave, c/n,
I = cn|E0 |2 /8. (2.50)
Note that this formula was derived without considering the frequency dispersion
of the medium, (). The dispersion can be taken into account by replacing in
(2.47) by d()/d (see, for instance, Refs. [Landau (1982); Silin (1961)]). As a
result, (2.49), instead of n2 , will contain the expression
(1/2)[d()/d + ] = c2 /uv, (2.51)
where v = c/n is the phase velocity and u = d/dk, the group velocity. Now, the
energy density should be multiplied by u, which again leads to relationship (2.50).
2.3.2 Cross-section of resonance interaction

After replacing the squared field |E0 |2 in the expression (2.37) for the transition
probability W with the photon flux density F, we find the transition cross section,
which is, by definition, equal to W/F (we assume n = 1):
mn = (42 /~c)g()|dmn
(e) 2
|, (2.52)
(e)
where dmn dmn e.
The transition cross-section is maximal at exact resonance ( = mn > 0); in
the case of a Lorentzian line shape (2.39),
(e) 2
mn0 = 8mn |dmn | /~c . (2.53)
One can imagine as the area of the shadow made by the atom. Let us estimate
this area. The line width cannot be less than the so-called natural width,
determined by spontaneous transitions. In what follows (Sec. 2.5), it will be shown
that
nat = 43mn |dmn |2 /3~ c3 = 1/T 1 nat (2.54)
where T 1 nat is the lifetime of an atom in the excited state, which is finite due to
spontaneous transitions into the ground state. Let dmn k e, then, after substituting
(2.54) in (2.53), we get
0 = (3/2)(nat /)2 . (2.55)
(e) 2
If dmn has random orientation, we have |dmn |2 = 3|dmn | , hence a factor of 1/3
appears in (2.55).
Thus, if both collision and Doppler widths are much less than the natural one,
the shadow of the atom with respect to resonance optical transitions has a size
on the order of the wavelength, 104 cm, and not the atom size, a0 108 cm.
In rare gases, the main role is played by Doppler broadening, which is on the order
of f /2 1GHz. The natural width for allowed optical transitions is two
orders of magnitude as small; therefore, 2 /100. (In the case of magnetic
dipole transitions, fnat 103 Hz and 106 2 .)
2.3.3 Population kinetics

Let us consider the evolution of mean populations Nm of the levels, which are
defined as
Nm |cm |2 N0 , (2.56)
where N0 is the total number of atoms. Thus, from considering a single particle
we pass to studying a system of N0 identical non-interacting particles. The n m
transition rate is defined as (see Eq. (2.36))
Wmn d|cm |2 /dt. (2.57)

P
Hence, the population increase rate Nm of the final state m is N0 n Wmn . However,
we have assumed that the initial state n is occupied with a unity probability; here,
P
this probability is equal to Nn /N0 ; therefore, Nm = n Wmn Nn . If we also take into
account that particles leave the levels, we get the following system of equations
for populations:
X
dNm /dt = (Wmn Nn Wnm Nm ). (2.58)
n
Here, so far, the relaxation is ignored; in the general case, the rates Wmn should
also include contributions from chaotic fields created by the surrounding particles.
Such equations are studied in non-equilibrium thermodynamics; they are called
kinetic equations or population balance equations. If only two levels take part in
the exchange, then
N1 + N2 = N0 , (2.59)
and only one of equations (2.58) is sufficient. Because the perturbation operator
is Hermitian, it follows that W12 = W21 W (see (2.26)), hence
N1 = N2 = W(N2 N1 ), (2.60)
2.3.4 Photon kinetics

Each transition down is accompanied by the emission of a single photon while
each transition up, by the absorption of a single photon; as a result, the rate of
photon emissionc is N1 , and the transfer equation for photons takes the form
N/t + (uN) = dN1 /dt, (2.61)

c In the semiclassical theory of radiation, there is no concept of a photon, and it would be more con-
sistent to speak here of the field energy variation by ~. But photon language is more convenient as
it is more visual.
where N(r, t) is the photon concentration and uN = F is the photon flux density
vector. Hence, in the stationary one-dimensional case, where N/t = 0 and
F = Fz (z),
dF/dz = W(N2 N1 ). (2.62)
Further, from the definition of the transition cross-section, it follows that
dF/dz = (N1 N2 )F F. (2.63)
Let populations be independent of z, then (2.63) leads to the exponential variation

of the intensityd of light,
F(z) = F(0)eI , (2.64)
where the absorption (or amplification, at N1 < N2 ) coefficient, according to

(2.52), is
(e) 2
= (42 /~c) g()|d12 | (N1 N2 ). (2.65)
2.3.5 Coefficient of resonance absorption

Maximal (resonance) value of the absorption coefficient in the case of a Lorentzian
line shape (2.39) is
(e) 2
0 = (8/~c)(21/)|d12 | (N1 N2 ). (2.66)
Note that in the stationary case, Nm = 0, and at W , 0 it follows from (2.60)

that N1 = N2 , so that, seemingly, always = 0. (This population balancing due
to the effect of the radiation is called saturation.) However, relaxation processes
that are ignored in (2.60), such as spontaneous transitions, inelastic collisions of
atoms with each other and with electrons in gases, interaction with lattice oscilla-
tions in solids, radiation-free transitions etc., tend to restore the initial population
difference N1 N2 . Therefore, in the case of sufficiently weak fields, saturation
can be neglected.
In the optical range, for allowed transitions with natural broadening, can be
as high as 1 cm1 at relatively small numbers of active particles |N| = N2 N1 .
At = 0.5, according to (2.55), |N| = 20 /32 = 109 cm3 . In the X-ray
range, is 4 orders of magnitude as small, and |N| 1017 cm3 .
In microwave paramagnetic amplifiers, the line width is determined by the
dipole-dipole interaction of paramagnetic ions. In a ruby crystal (Al2 O3 +103 Cr),
d Recall that intensity scales as F, namely, I = ~F.
with the concentration of chrome ions 1019 cm3 , the line width is on the or-
der of 50 MHz. Substituting for d in (2.53), for = 1 cm we get =
82 /~ f = 5 1020 cm2 . A realistic number of active particles available in
paramagnetics is approximately equal to the equilibrium population difference,
N (0) = ~N0 /T g N0 /10 = 1018 cm3 , where g = 2S + 1 = 4 is the degener-
acy of the chrome ion ground level, which is lifted by a constant magnetic field,
and S is the spin number. Hence, 0 = 0.05 cm1 , and for obtaining amplification
G = 100 one needs the length of the crystal l = ln G/0 1 m. In order to reduce
l, the crystal is placed into a bulk resonator, where radiation can many times pass
through the matter, or into a slowing-down system. In the latter case, the above-
given formulas for and will be still valid, with the speed of light in the vacuum
replaced by the group velocity of waves in the slowing down medium, u = dK/dz,
where K is the propagation constant.
2.3.6 Amplification bandwidth

Due to the exponential relation (2.64) between the transfer coefficient of a layer
of thickness l, G F(l)/F(0), and the absorption coefficient , the shape of the
observed frequency dependence G() at || l 1 (large optical density) will differ
from the function (). It is easy to see that this effect will lead to the sharpening
of the observed resonance at < 0 and to its broadening at > 0 (Fig. 2.4).
Let < 0 and the () dependence be Lorentzian. Defining the amplification
bandwidth 0 by the condition of G() 1 two-fold reduction with respect to its
Fig. 2.4 Observed shape of the resonance in the case of a Lorentzian line with width at different
optical densities y = 0 l at the center of the line (x 2( 0 )/).
maximal value, we find from (2.64) that

)1/2
0
(
ln G0
= 1 , (2.67)
ln[(G0 + 1)/2]
with G0 exp(0 l). Hence, at G0 1 1 it follows that 0 = , while at
G0 1 1,
!1/2
0 ln 2 1
. (2.68)
ln G0 ln 2 |0 |l
Thus, the narrowing of the amplification band with the increase of the amplifier
length is rather slow. For instance, at G0 = 100 (0 l = 4.6) the ratio (2.67) takes
the value 0.417 (approximate expressions (2.68) yield 0.42 and 0.46).

2.3.7 Degeneracy of the levels
Expression for the amplification coefficient and the inversion condition N2 > N1
have been obtained above under the assumption that the atom energy levels were
not degenerate. Let now g1 different (with respect to some parameters) states have
the same energy E1 and g2 states have energy E2 ,
(H0 E1 )1i = 0 (i = 1, . . . , g1 ),
(2.69)
(H0 E2 )2 j = 0 ( j = 1, . . . , g2 ).
Note that the conclusions given below will be also valid in the case where the
degeneracy is lifted due to sufficiently small perturbations (Fig. 2.5). Now, the
Fig. 2.5 Degeneracy of the levels: g1 different states have the same energy E1 while g2 other states
have the energy E2 . In the right-hand side of the figure the degeneracy is lifted due to the external con-
stant field H0 , which breaks the symmetry of the system. Alternating field causes transitions between
a certain pair of states i, j.
subscripts 1 or 2 in the perturbation theory should be replaced by double indices,

1i or 2 j. The probability of a stimulated transition between the states 1i and 2 j,
according to (2.37), scales as the corresponding matrix element:
(e) 2
W1i,2 j = W2 j,1i |d1i,2 j| .
The number of transitions up or down is proportional to the population of the

initial state N1i or N2 j , therefore
g2
X
N1i = (W1i,2 j N2 j W2 j,1i N1i ). (2.70)
j=1
P
The variation rate of the total population of the level, N1 N1i , will be equal to
the double sum over the degenerate states,
X
N1 = W1i,2 j (N2 j N1i ). (2.71)
ij
Now, assume that the saturation effect is absent and relaxation or inversion
lead to a uniform distribution of sublevel populations,
N1i = N1 /g1 , N2 j = N2 /g2 . (2.72)
As a result, (2.71) takes the form (see (2.60))
N1 = W 0 N 0 , (2.73)
where
X
W0 W1i,2 j , N 0 N1 /g1 N2 /g2 . (2.74)
ij
Thus, degeneracy of the levels can be taken into account if W in Eq. (2.66)
is understood as W 0 and N, as the difference of state populations Nm /gm . The
inversion condition then takes the form
N1 /g1 < N2 /g2 . (2.75)
Let, for instance, g1 = 1 and g2 = 3, then, one needs N2 > 3N1 for amplification.
Recall that, according to the Boltzmann distribution,
N2(0) /N1(0) = (g2 /g1 ) exp(~21 /T ) (2.76)
and N2(0) < 3N1(0) .

2.4 Stimulated transitions in a random field
Up to now, the field stimulating a transition was considered as monochromatic.

Let now E(t) have an arbitrary time dependence. According to (2.24), the first-
order perturbation theory gives the following expression for the transition ampli-
tude in the dipole approximation, provided that c1 (t0 ) = 1:
Z t
1 X
c2 (t) = dt0 exp(i21 t0 ) d21 E (t0 ), (2.77)
i~ t0
where = x, y, z are indices in the Cartesian frame. Squared module of this

expression gives the transition probability,
X " t
P21 (t) = ~2
d21 d21 dt0 dt00 exp[i21 (t0 t00 )]E (t0 )E (t00 ). (2.78)
t0
2.4.1 Correlation functions

Next, consider the case of a chaotic, random field. Then, P should be averaged
over a corresponding probability distribution, so that instead of the pair product
E E , (2.78) will contain the matrix of second-order field momentse
hE (t0 )E (t00 )i G (t0 , t00 ) = G (t00 , t0 ). (2.79)
This equality defines a certain tensor, each of its nine components being a function
of two arguments. The matrix of second-order moments (2.79) is also called the
field correlation tensor. Another equivalent term is the field coherence function
(of the first order). Statistical properties of a random field are fully described by
a set of moments (coherence functions) of all orders and for all possible pairs of
points x r, t. Statistical optics will be described in more detail in Chapter
7; here, we only note that odd field moments, as a rule, are equal to zero, while
moments of order 2n determine the probabilities of n-photon transitions. Let us
also mention that the sum of the diagonal elements of the second-order moment
P
matrix, G (the trace of the matrix), with coinciding arguments defines the
mean energy density hE 2 i/8 of the electric field at the point under consideration.
The most important class of random fields are stationary fields, whose statis-
tical characteristics (intensity, spectrum, polarization) do not change with time.
The correlation function of a stationary process can only depend on the difference
of its two arguments,
G (t0 , t00 ) = G (t0 + t0 , t00 + t0 ) G (t0 t00 ). (2.80)
e Angular brackets denote averaging over a statistical ensemble of fields (Sec. 7.2).
From (2.80) and the definition (2.79), the symmetry property follows,
G (t) = G (t); (2.81)
in particular, G (t) should be an even function of time.
Thus, according to (2.78) and (2.79), the probability of a transition due to
stationary random radiation is determined by the field correlation tensor,
X " t
P21 = ~2
d21 d21 dt0 dt00 exp[i21 (t0 t00 )]G (t0 t00 ). (2.82)
t0
2.4.2 Transition rate

Consider the action of the perturbation at time intervals that are much larger than
the field correlation time E (Sec. 7.2). Then the integration limits in (2.82) can
be replaced by . Let us make a change of variables, t1 t0 t00 , t2 t0 + t00 .
Integration in t1 yields the Fourier transform of G (21 ), which is called the field
spectral density tensor,
Z
G () (2)1 dteit G (t) = G () = G () G+ (). (2.83)
The inverse transformation has the form

Z
G (t) = deit G (). (2.84)
Here, as usual, we omit infinite integration limits and denote the function and its
Fourier transform by the same letter. The second integration (in t2 ) simply yields
the observation time t t0 , so that one can define a time-independent transition
rate W P/(t t0 ). Let the dipole moment of the transition be parallel to the x
axis, then we finally find a simple expression for the transition rate,
W21 = 2~2 |d21 |2G xx (21 ). (2.85)
Thus, the rate of a stimulated transition due to a random (noise or incoher-
ent) perturbation scales as the spectral density G() of the perturbation at the
transition frequency. It is useful to compare (2.85) with formula (2.37), which
defines the transition rate in the case of a monochromatic field. The two formulas
coincide after the substitution |E0 |2 g() G().
In this consideration we did not take into account the broadening of the
levels due to relaxation processes. However, intuitively it is clear that the conclu-
sion should be still valid in the case where is much less than the width of the
perturbation spectrum, E 1/E . In this case, the field is called incoherent. In
the opposite case, the field can be obviously considered as monochromatic, i.e.,
coherent.
2.4.3 Einsteins B coefficient

Consider now isotropic non-polarized radiation with G = G . From G() we
will pass to the energy spectral density (). The latter is defined through the
energy density as follows (we assume n = 1):
Z
2 2
E/V = hE + H i/8 d(). (2.86)
0
In the radiation, E = H, therefore,

X Z
E/V = G (t = 0)/4 = 3 dG()/2, (2.87)
0
where the last equality was obtained using relation (2.84) with t = 0 and taking
into account that, according to (2.83), G() is an even function. From the com-
parison of (2.86) and (2.87) we get the relation between the spectral densities of
field amplitude and energy,
G() = 2()/3. (2.88)
By substituting (2.88) in (2.85), we finally find the transition rate in an isotropic

non-polarized noise field with a broad spectrum:
W21 = B21 (21 ), (2.89)
B21 = B12 (2|d21|/~)2 /3. (2.90)
The proportionality coefficient B between the transition rate and the energy
density is called the Einstein coefficient for a stimulated transition. In the next
section, using the Planck function (0) () for equilibrium radiation, we will find
the second Einstein coefficient, A, giving the rate of spontaneous transitions.

2.4.4 Spectral field density
Concluding this section, let us clarify the physical meaning of the field spectral
density G(). In order to do this, we formally represent E(t) as a Fourier integral
(the subscript is omitted),
Z
E(t) = deit E(), (2.91)
Z
E() = dteit E(t)/2. (2.92)
A rigorous definition for the Fourier representation of a random function can be

found in Ref. [Rytov (1976)]. With the help of (2.92) and definitions (2.79),
(2.80), we find the correlator of the field Fourier components,
"
0 0
hE()E(0 )i = dtdt0 eit+i t G(t t0 )/42. (2.93)
In the time integral, we make a change of variables, t1 t t0 , then the

integral in t1 yields, according to (2.83), 2G(), while the second integral yields
2( + 0 ), according to one of the delta-function representations,
Z T
lim dteit = 2(). (2.94)
T T
As a result,
hE()E(0 )i = G()( + 0 ). (2.95)
Because E(t) is real, it follows from (2.92) that
E() = E (); (2.96)
therefore, (2.95) can be also represented in the form
hE()E (0 )i = G()( 0 ). (2.97)
Thus, in a stationary field, only harmonics of opposite frequencies correlate,

and their correlation is determined by the spectral density G(). It means that the
reading of a photodetector measuring the field energy within a frequency band
centered at will scale as G().
2.5 Field as a thermostat
Consider population kinetics for atoms in an equilibrium field with the energy
spectral density (0) (). From the kinetic equation (2.60) for two nondegenerate
levels, it follows that
N2 = N1 = B(N1 N2 ), (2.98)
where B B12 = B21 and (21 ). Thus, a noise broadband field, similarly
to a monochromatic one, tends to equalize the populations of the levels, so that
in the stationary regime N1 = N2 . However, equilibrium radiation of temperature
T should heat or cool the matter to the same temperature, hence the population
distribution is given by the Boltzmann formula, according to which N1 > N2 .
2.5.1 Spontaneous transitions

This contradiction can be solved by adding to the kinetic equation (2.98) a term
describing spontaneous (i.e., field-independent) transitions from the excited level
2 to the ground level 1. Such transitions are accompanied by the emission of light
from the heated body, which prevents the equalization of the populations due to
the external field. According to Einstein, let us denote the rate of spontaneous
transitions as A12 A, then (2.98) takes the form
N2 = B(N1 N2 ) AN2 . (2.99)

The A coefficient can be calculated from Boltzmanns and Plancks distribu-
tions and the B value found above. If the atoms are in equilibrium with the field,
N2 = 0, and (2.99) leads to the relation
A/B = (N1 /N2 1). (2.100)
Substituting here the Planck distribution,

(0) () = ~k3 N()/2 , (2.101)
with
N() = [exp(~/T ) 1]1 , (2.102)
and the Boltzmann distribution,
N1(0) /N2(0) = exp(~21 /T ), (2.103)

we find the ratio of the coefficients for spontaneous and stimulated transitions:
A/B = ~k3 /2 , (2.104)

where k = /c = 1/o. Hence, taking into account expression (2.90) for the B
coefficient, we find that
A = 4k3 |d21 |2 /3~. (2.105)

For an allowed transition in the visible range (d21 = 1D, = 0.5), estimation
yields A = 2 106 s1 .
The notion of a spontaneous transition plays an important role in the theory
of interaction between field and matter and in quantum electronics. Spontaneous
transitions determine the minimal linewidths of emission and absorption. They
lead to the thermal radiation of heated matter. Similarly to relaxation processes
in general, they hinder obtaining population inversion. Also, since spontaneous
transitions occur independently of the external field, they are a source of noise and
therefore limit the sensitivity of quantum amplifiers and the monochromaticity of

quantum generators (Sec. 7.1).
Let us note that the probability of a spontaneous transition has a strong (cu-
bic) dependence on the frequency, which explains why creating UV and X-ray
lasers faces a certain difficulty. Although in the opposite spectral range, the mi-
crowave one, the probability of spontaneous transitions is very small, the minimal
noise temperature of paramagnetic amplifiers is namely determined by sponta-
neous transitions (see (7.10), (7.11)).
We have found A in an indirect way. Spontaneous transitions can be consis-
tently explained in the framework of quantum electrodynamics, by the interaction
between atoms and the vacuum (Sec. 7.7). However, they have simple classi-
cal and semiclassical analogues, the radiation of accelerated electrons in an atom
(Sec. 5.2). One can also provide a semi-quantum model: a classical current, ac-
cording to Glauber [Glauber (1965)], excites quantized field into a coherent state.
2.5.2 Natural bandwidth

Spontaneous transitions limit the lifetime T 1 of an isolated atom in an excited
state. One can expect that T 1 = 1/A. This simple dependence will be confirmed in
what follows, see (5.34). Further, according to the uncertainty relation Et = ~,
where E is the accuracy of the energy measurement and t is the measurement
time, a finite lifetime of an atom leads to a finite width of the energy level. As-
suming t = T 1 , we obtain E2 = ~A. This broadening of the level should man-
ifest itself, in stationary experiments, in the variance of the transition frequency,
21 = E2 /~, i.e., in the broadening of spectral lines,
nat = A. (2.106)
The width of spectral lines caused by spontaneous transitions is called the nat-
ural bandwidth. This term stresses that nat is the minimal possible linewidth,
which takes place even in the case of a single isolated atom. Note, however, that
natural broadening can be, in principle, eliminated by placing an atom into a bulk
resonator that has no oscillations with frequencies in the vicinity of 21 . In prac-
tice, observed lines have natural bandwidth only in the rare cases where other
perturbations, such as collisions and the Doppler effect in gases, interaction with
phonons in crystals, etc., have much smaller contributions and, in addition, the op-
tical thickness of the sample is small (Sec. 7.1). Note that in this case, the width
of absorption or amplification lines related to stimulated transitions is also equal
to A.
Natural broadening of isolated lines leads to Lorentzian lineshapes. Rig-

orously, this follows from the Wigner-Weisskopf theory (see, for instance,
Ref. [Louisell (1964)]). There is also a simple classical model, according to which
an excited atom emits an exponentially decaying quasi-monochromatic oscillation
(Sec. 5.2). Fourier transformation of this oscillation provides the Lorentzian (dis-
persion) lineshape of the emission spectrum.
Let us estimate, with the help of (2.106), the relative value of the natural broad-
ening. For allowed transitions, d ea0 (see (2.30)), so that
nat / (4/3)(1/137)(a0/o)2 , (2.107)
where we assumed the value 1/137 for the fine structure constant e2 /~c (recall
that this number also defines the ratio of the velocity of an electron in a hydrogen
atom to the speed of light). Assuming = 1/R 4 137a0 0.1, with R being
the Rydberg constant, we obtain
nat / 0.3/(137)2 107 . (2.108)
Displacement of atomic levels due to the interaction with the electromagnetic vac-
uum (the Lamb shift) is on the same order of magnitude or smaller. Thus, relative
perturbation of an atom by the vacuum is extremely small.
2.5.3 Number of photons, spectral brightness, and brightness

temperature
Let us find the ratio of stimulated and spontaneous transition probabilities in the
case of incoherent (noise) field. According to (2.104) and (2.89),
W st /W sp = B/A /~g N, (2.109)
where
g 2 /2 c3 , (2.110)
has the meaning of the spectral density of field modes in a unit volume (recall that
is the energy spectral density per unit volume). A mode, or an oscillation type, is,
roughly speaking, an oscillation degree of freedom (or a spatial harmonic) of the
field (Sec. 7.3). The inverse value, 1/g , is equal to the frequency interval between
the neighboring modes. According to definition (2.109), N has the meaning of the
field energy per one mode, in ~ units. In other words, N is the number of photons
per mode. This value is also called the degeneracy factor of photon gas. Note that
both energy and the number of photons fluctuate; here, and N are mean values.f
The value N is the most important parameter of incoherent radiation. Let us
show that it scales as the basic photometry characteristic, the spectral brightness
I . The latter is defined as the radiation intensity within a unit spectral interval
and unit solid angle and has dimensionality [W/(cm2 Hzsr)]. Radiation intensity
per unit frequency is equal to half the spectral energy density, /2, multiplied by
the speed of light. By adding the factor 1/4, we pass to the spectral brightness
I . Hence, with the help of (2.109), we find
I = c/8 = ~c3 N. (2.111)
Thus, according to (2.109), stimulated transitions due to incoherent field occur

N times as frequently as spontaneous ones. The total number of transitions up and
down can be represented as
w21 = AN, w12 = A(N + 1). (2.112)
Sometimes spontaneous transitions, which correspond to the second term in

the last expression, are interpreted as stimulated transitions due to zero-point (vac-
uum) fluctuations of the field. However, this interpretation leads to underestimat-
ing twice the probability of spontaneous transitions down and does not explain the
absence of spontaneous transitions up [Ginzburg (1983)]. The correct result is ob-
tained by distinguishing between normally and antinormally ordered fluctuations.
In Sec. 7.7, it is shown that spontaneous transitions are determined by normally
ordered fluctuations of the atom dipole moment and by antinormally ordered fluc-
tuations of the vacuum.
In equilibrium radiation, N only depends on the frequency and temperature
and is given by the Planck formula (2.102), N = N(). In the general case, N
depends, in addition to frequency, on the observation direction, polarization type,
f Editors note: Eq. (2.109) has an important consequence in the context of quantum information. As
it was first mentioned by Wooters and Zurek and independently by Milonni and Hardies in 1982, if the
ratio between the induced and spontaneous transitions tends to unity, then it leads to the impossibility to
clone the polarization state of a single photon. Indeed, if Win = Wsp , then the mean number of photons
per field mode equals unity. Then, amplification of a single photon through stimulated transitions in an
atom (which was initially prepared in excited state) will be accompanied by the spontaneous emission
of a photon that has random polarization with respect to the initial one. If the mean number of photons
per mode grows, then the contribution of spontaneous transitions goes down. According to (2.109),
cloning becomes possible in the limit of high N, which is often associated with classical field, therefore
this fact does not contradict to the non-cloning theorem. However, we would like to stress that this
conclusion has to be applied with caution because there is no criteria of non-classicality of light based
on the mean photon number! For instance, squeezed states of light or bright squeezed vacuum states
(Sec.7.5) are nonclassical despite having large photon numbers [Bachor (2004)].
and observation point: N = N(k, r, ). Here, k is the wavevector, which also

defines the frequency and the direction, and is polarization index taking two
values.
In non-equilibrium field, equality (2.102) is used as the definition of the bright-
ness temperature T e f (N) ~/ ln(1 + 1/N) for radiation with given frequency,
direction, and polarization. For instance, solar radiation, taken in the optical range
and within an appropriate angular interval, has T e f 6000 K, so that, according
to (2.102), for green light ( 0.5) N 102 . We see that stimulated transitions
in solar light are much less frequent than spontaneous ones. Thus, in the visible
range the probabilities of stimulated and spontaneous transitions become equal
only for radiation that is hundreds of times as bright as the Sun radiation, with
T e f 4 104 K. Such brightness can be only achieved with multimode lasers (the
notion of brightness cannot be applied to single-mode lasers).

2.5.4 Relaxation time
With the help of (2.99), let us now define the rate of population variation with an
account for both spontaneous and stimulated transitions. For this, we replace N1
by N0 N2 and use relation (2.112):
N2 = A[NN0 (2N + 1)N2 ]. (2.113)
Here N is the number of photons per mode and N0 = N1 + N2 is the total number
of atoms at two levels. Hence, in a stabilized regime,
N1(0) /N0 = (N + 1)/(2N + 1) = /(1 + ),
(2.114)
N2(0) /N0 = N/(2N + 1) = 1/(1 + ),
where exp(~21 /T ). The solution to (2.113) has the form
N2 (t) = N2(0) + [N2 (t0 ) N2(0) ]et/T 1 ,
(2.115)
1/T 1 A(2N + 1) = 2B + A = w12 + w21 .
Thus, the time T 1 of heating (or cooling) of the atomic internal degrees of freedom
due to the interaction with incoherent radiation at small N equals 1/A, the life time
of an atom due to spontaneous transitions, while at large N it reduces 2N + 1 =
coth(~21 /2T ) times.
In fact, here we have considered a simple model of relaxation where the ther-
mostat is formed by incoherent electromagnetic radiation surrounding the atom
(Sec. 7.7).

Chapter 3
Density Matrix, Populations, and

Relaxation
The probability method used above allowed us to describe the energy exchange
between radiation and atoms. At the same time, another known manifestation of
the interaction between field and matter, namely, the slowing down of the waves
propagation, was not considered. Another, and more important, drawback of the
probability approach is that it does not provide a sufficiently rigorous account
for relaxation processes, whose consistent consideration should be performed in
the framework of statistical physics and kinetics. A more complete theory of
the interaction of atoms with the external field and the thermostat is based on the
density matrix formalism, which combines quantum and statistical considerations.
Below, in Sec. 3.1, we discuss the definition and the general properties of the
density matrix. Itn Sec. 3.2, its diagonal elements are considered, which give the
populations of the levels, and the notion of negative temperature is introduced.
Section 3.3 describes the time evolution of the density matrix and the relaxation
processes.
3.1 Definition and properties of the density matrix
3.1.1 Observables
In Chapter 2, we have defined the transition probability in terms of the amplitudes
of energy states cn . Let us now write an arbitrary observable f of a quantum
system (further, f will be understood as the dipole moment of an atom, f d ) in
terms of similar coefficients. We will start from the basic measurement postulate
of the quantum mechanics: multiple measurements of a value f performed on an
ensemble of identical systems, i.e., systems prepared in the same state (r, t),
will yield, on the average, the value
Z
h f (t)i = dr (r, t) f(r, t) ht| f |ti, (3.1)
43
where f is the operator corresponding to f and r is the set of the system coordi-
nates. (In what follows, we will often omit the hats of the operators.)
It is important that f in (3.1) can be also understood as a product of operators,
f g2 or f gh. This enables one to determine not only the mean values, h f i,
called first-order moments, but also higher-order moments, hg2 i, hgn i, hghi, . . . ,
which characterize the quantum fluctuations of g and the quantum correlation of
g and h. Of course, Eq. (3.1) is written in the Schrodinger picture and is therefore
applicable only to the case where f is a single-time operator, for instance, f (t) =
g(t) h(t). In order to define the correlation functions hg(t1 ) h(t2)i, one has to switch
to the Heisenberg picture where the time dependence is attributed to the operators
and not to the wave functions.
Knowing the wave function, one can find not only the moments h f n (t)i of an
observable f but also its distribution at time t, P( f, t). This function is given by
Eq. (3.1) with the operator f replaced by the diade operator | f ih f | (Sec. 7.5).
3.1.2 Density matrix of a pure state

Let us expand the wave function over the set of the eigenstates of some operator
(not necessarily the energy operator),
X
(r, t) = bn (t)n (r). (3.2)
n
P
In the Dirac notation, simply |ti = |nihn|ti. Note that if n are energy func-
tions, then the coefficients bn and cn differ in only exponential factors (see (2.13)).
Substituting (3.2) in (3.1), we obtain
X
hfi = bn bm fnm . (3.3)
nm
Here, the matrix
Z
fnm drn f m hn| f |mi
is assumed to be known, and the problem is reduced to the one of finding pairwise
products bn bm , which also form a matrix, called the density matrix or the statistical
matrix,
mn bm bn . (3.4)
Thus, the state vector of the system is put into correspondence with a matrix.
One can also define an operator corresponding to :
Z
drm n bm bn , (3.5)
which, in Diracs notation, reads = |tiht| (Sec. 7.5).

Density Matrix, Populations, and Relaxation 45
In terms of the density matrix and operator, the mean (3.3) can be written in a
more compact manner,
X
hfi = mn fnm = Tr( f ), (3.6)
mn
P
where Tr f means the sum of the diagonal elements, fnn , called the trace, or
Spur, of the matrix. The trace of a matrix is one of its invariants, since it does not
change under the matrix transformations of the form f 0 = U f U 1 . Such operator
transformations describe changes of the representation in quantum mechanics, and
the invariance of the trace provides that the observable quantities are independent
of the choice of representation. The property Tr f 0 = Tr f immediately follows
from another property, Tr(gh) = Tr(hg), which can be easily verified from the
definitions of the Tr and multiplication operations.
3.1.3 Mixed states

In the transition from classical mechanics to statistical physics, the main notion is
the Gibbs ensemble, which is composed of identical systems distributed with the
probabilities P(q, p) over the possible states of the system. A quantum statistical
ensemble is constructed in a similar way: we assume that its P1 N systems are in
the state 1 , P2 N in the state 2 , Pi N in the state i , and so on. Here, N is the
P
total number of systems in the ensemble and Pi = 1.
A mixed state is a state of the system for which the exact wave function is
not defined but only the set of numbers Pi is known, each of them giving the
probability that the system is in the i state. The system is then characterized as a
weighed mixture of states, in contrast to a pure state, for which the wave function
of the system is known.
We stress that a linear combination 1 1 + 2 2 is still a pure state with a
well-defined wave function. The mean value of an operator contains in this case
an interference term, which depends on the relative phase of the states,
h f i = P1 f11 + P2 f22 + 2Re(1 2 f12 ), (3.7)
where
Z
Pi |i |2 , fi j dri f j .
In a similar mixed state, the last term in (3.7) is absent. There is an analogy
with the superposition of two light fields: coherent fields interfere, while a non-
coherent mixture simply yields double intensity.
Additional uncertainty of mixed states leads to additional thermal fluctua-

tions of observables over an ensemble and, according to the ergodicity hypothesis,
also over time. However, these fluctuations are not as principal and inevitable as
quantum ones.
In real experiments, the purity of the prepared states depends on the skill
of the experimentalist. Near the absolute zero, a crystal is in a pure (ground)
state with a definite energy, but the coordinates of its atoms still have quantum
fluctuations. In a good maser or laser with complete population inversion, the
atoms are in the excited state with the energy E2 . In this case, there are no energy
fluctuations, hH 2 i = E22 , but the coordinate of the electron and the dipole moment
still fluctuate.
Thus, depending on its prehistory, a quantum object can be found in one of the
three possible state types:
1) in an eigenstate n of a given operator f, where one knows a priori that
h f i = fnn and f does not fluctuate,
h f k i = h f ik ;
P
2) in a pure state formed by a superposition bn n , with quantum fluctuations
observed and only the probability |bn |2 of measuring a certain value fnn is known,
X
h f ki = |bn |2 ( fnn )k ;
3) in a mixed state, where quantum uncertainty is combined with the lack of in-
formation about the wave function.
In the case of mixed states, mean values should be calculated via double av-
eraging: quantum averaging over the wave function i , according to Eq. (1), and
classical averaging, with the help of the Pi distribution and the usual rules of the
probability theory,
X X Z
hfi = Pi fii = Pi dri f i . (3.8)
i i
Now, the amplitudes in expansion (3.2) and the density matrix (3.4) depend
on the index i,
X
i = b(i)
n n ,
so that (3.8) takes the form

X
(i)
hfi = Pi (b(i)
n ) bm fnm = Tr( f ), (3.9)
mni
where we have defined the density operator of a mixed state,
X
Pi i , mn bm bn . (3.10)
i
Due to its linearity, the averaging operation could be included into the definition
of the density matrix. Equation (3.6) maintains its form in this case. In future, the
bar denoting additional averaging will be omitted.

3.1.4 More general definition of the density matrix
Often, one defines a mixed state and the density matrix in a different, more general
way. In this case, the term mixed state is applied not to the whole system but to
a part of it.
Let a system consist of two parts, A and B. Its state is given by a wave function
(rA , rB), which, in the general case, is not factorable, (rA , rB) , A (rA )B (rB ).
Therefore, the wave function A of a subsystem does not exist. Indeed, factora-
bility means independence of the subsystems, hence it is impossible if A and B
interact or have interacted in the past. There is an analogy with classical statis-
tical physics: for interacting particles, the joint probability distribution P(rA , rB)
cannot be represented as P(rA )P(rB).
However, in classical statistical physics we can separately define the probabil-
ity distribution function for subsystem A by summing P over the variables that are
of no interest for us,
Z
PA (rA ) = drB P(rA , rB). (3.11)
The question is whether a similar procedure is possible in quantum mechanics.

In order to define the density matrix of a subsystem, let us expand over some
complete set of functions in (rA , rB ),
X
= bin in . (3.12)
in
Such a set in = i n is generated by two operators, each of them acting on the

variables of only one subsystem. For instance,
(HA EAn )n = 0, (HB EBi )i = 0.
Let fA be the observable of interest,
Z
h fA i = drA drB fA
X X
= bin bi0 n0 fAnn0 ii0 An0 n fAnn0 = Tr(A fA ). (3.13)
ii0 nn0 nn0
We have once again obtained Eq. (3.6) by introducing the notation

X
An0 n bin bin0 , (3.14)
i
which is equivalent to the definition (compare with (3.11))
A TrB (AB ). (3.15)
In quantum electronics, the A system usually corresponds to a particular atom

(or a molecule or an electron in a crystal) while the B system includes all other
matter particles and quantized electromagnetic field. (Apparently, a classical field
acting on a system does not destroy the purity of the state.) Sometimes, on
the contrary, A is understood as a particular field mode while B corresponds to the
matter. If system B has a sufficient number of degrees of freedom and a continuous
energy spectrum, i.e., has a large heat capacity, then its state can be considered as
independent of A and it plays the role of a thermostat. The influence of B on A
causes the relaxation of A.
If the back-action of A on B can be neglected, then A can be described by
means of a wave function or a density matrix of a pure state (3.4), and we come
back to the problem of a quantum system in a given noise field, which was con-
sidered in Sec. 2.4. Solving this problem in the framework of some model enables
one to calculate the relaxation properties and the shapes of spectral lines. In the
simplest model, the thermostat for an atom is formed by the equilibrium Planck
field, and the probabilities of relaxation transitions are given by the Einstein coef-
ficients A, B (Sec. 2.5).
3.1.5 Properties of the density matrix

Using definition (3.10), one can easily show that the density matrix has the fol-
lowing properties:
Tr = 1, 0 nn 1, + = . (3.16)
In most cases, one uses the energy representation, in which the diagonal elements
of the density matrix nn n have the meaning of relative occupation numbers
of the levels. The first property in (3.16) means that the probability to find the
system on some level is equal to unity, the second one provides the non-negativity
of the probability, and the third one (Hermiticity), that the observable quantities
are real,
X X
h f i = mn fnm

= nm fmn = h f i.
The non-diagonal element of the density matrix bm bn characterizes the de-

gree of correlation between the m and n states in a statistical ensemble. If the
state amplitudes of various systems of the ensemble contain random phase fac-
tors, b(i) (i)
n exp(in ), then, for m , n,
mn exp i(m n ) = 0, (3.17)

and the state of the ensemble is fully characterized by the state populations n .
For a pure state, definition (3.4) leads to the property |mn |2 = mm nn . In a
mixed state, the elements of the density matrix satisfy the Cauchy-Bunyakovsky
inequality,
|mn |2 < mm nn . (3.18)

3.1.6 Density matrix and entropy
Let a closed system be in a pure energy state, = 1 exp(iE1 t/~). Then, accord-
ing to definition (3.4), there is only one nonzero element of the density matrix,
mn = mn n1 . Such a trivial matrix satisfies the matrix equation
2 = . (3.19)
This is a property of all pure states. It follows from Eq. (3.4), the matrix multipli-
cation rule, and the normalization condition Tr = 1.
Violation of equality (3.19), or its corollary Tr2 = 1, can be a sign of a
mixed state. However, there exists a more convenient quantitative measure of
statistical indeterminacy of quantum systems, the entropy (see, for instance, [Fain
(1972); Landau (1964)]). Let us define the entropy operator in terms of the density
operator in the following way: S ln . Then the entropy S is equal to hS i, i.e.,
S = hln i = Tr( ln ). (3.20)
In the representation where is diagonal, (3.20) takes the form
X
S = n ln n . (3.21)
n
(This follows from the fact that in the diagonal representation, [F( f)]nn = F( fnn ).)
In a pure state, n equals 0 or 1, therefore S = 0, the indeterminacy (chaoticity)
is minimal. The opposite limiting case of a maximal indeterminacy is realized for
a uniform mixture of states, n = const = 1/g, where g is the number of states with
a given energy (the Gibbs microcanonical ensemble). Then, = I/g, 2 = I/g 2
and, according to (3.21)
g
X
S = (1/g) ln(1/g) = ln g. (3.22)
n=1
Thus, 0 S ln g.

3.1.7 Density matrix of an atom
In statistical physics, one usually considers macroscopic objects consisting of
N 1022 identical particles. The terms state, energy level, density matrix
relate in this case to the matter as a whole. In principle, one can speak about the
wave function of a 1 cm monocrystal, which depends on about 1022 space argu-
ments ri and time. Correspondingly, the number of possible states and, hence, the
dimensionality of the matrices fmn , mn are also extremely high. Furthermore, in
order to realize an ensemble one has to have, say, 103 similar crystals.
On the other hand, active media in quantum electronics, such as gases,
doped crystals or dye solutions, as a rule, consist of weakly interacting atoms
or molecules. Then, it is sufficient to consider the state of a single atom, or, to be
precise, of a single external electron. The rest of the particles are then considered
as a thermostat, which has a weak influence on the wave function of the atom.
This transition from about 1022 degrees of freedom to a few ones leads to a
crucial simplification of the theory, i.e., to the ideal gas model. The theory is fur-
ther simplified by excluding from considerations all states that are not populated
and not resonant with respect to the external field. This transition to the two-level
system is valid in the case of a quasi-monochromatic external field and the ab-
sence of degeneracy. Note that the density matrix of an n-level non-degenerate
system consists of n2 elements, n(n 1) of them being complex. However, the
normalization and Hermiticity conditions (3.16) reduce the number of indepen-
dent elements, so that the state of the system is described by n2 1 m real
numbers. For a two-level system, m = 3, and its state can be represented as a
point in a three-dimensional phase space, with the coordinates 2021 , 20021 , and
1 2 (Sec. 4.4). In the case of a pure state, condition 2 = reduces the
number of independent parameters to two, and the state can be shown by a point
on a unit sphere.
Since the atoms are identical and independent, the additive macroscopic pa-
rameters of the matter, such as polarization P, are calculated by simply multiply-
ing single-atom mean values by N, P = Nhdi. Note that if all gas atoms are under
the same conditions, the gas as a whole can be considered as an ensemble (quan-
tum or quantum-statistical) containing approximately 1022 systems. Summation
over atoms is then equivalent to ensemble averaging, and a diagonal element of
the density matrix, n , defines the average relative population Nn /N of a level En
in a real gas rather than in a hypothetical ensemble of 103 similar gas volumes.
3.2 Populations of levels
3.2.1 Equilibrium populations

In thermodynamic equilibrium, all statistical properties of a system are determined
by the Gibbs canonical distribution. This distribution is applicable to both isolated
macroscopic systems and systems of any size interacting with a thermostat. If
single atoms or molecules of an ideal gas are considered, the Gibbs distribution
corresponds to the density matrix of the form
(0) (0)
mn = mn m = mn exp(Em /T )/Z, (3.23)
where the normalization factor Z is called the statistical sum and can be found
from the normalization condition,
X
Z= exp(Em /T ). (3.24)
m
Here, m numerates various states of an atom, therefore the population of a gn -fold

degenerate level is
Nn(0) = gn N exp(En /T )/Z. (3.25)
This equation is called the Boltzmann distribution. Note that the equilibrium den-
sity operator (3.23) can be represented in the form
(0) = exp(H0 /T )/Tr{exp(H0 /T )}. (3.26)
As it was shown in Sec. 2.3, interaction of the external field with the matter
is determined by the populations of the resonant levels N1 , N2 . In the first order
of the perturbation theory, alternate field only creates the non-diagonal elements
of the density matrix, (1)
12 E, while the diagonal elements remain unchanged,
(1)
n 0. Therefore, at sufficiently weak fields one can calculate populations using
the Boltzmann distribution (3.25).
According to the Boltzmann distribution, the only populated states are the ones
that are apart from the ground state by an energy not much exceeding T . Hence,
field at frequencies much larger than T/~ T can only cause transitions up. At
room temperature, this boundary frequency is in the far IR range (t T /2c
200cm1, T = 1/T 50), while at helium temperatures, in the microwave
range (t 1cm1 ).
In the case of atomic gases and dopant ions in crystals, the lowest excited
levels, as a rule, are well above this boundary, and almost all particles are in the
ground state, so that they all participate in the absorption of light, N N1 N.
Often, the ground level has a degeneracy g1 , which can be lifted (completely or
partly) due to the spin-orbit interaction (the fine structure) or due to static fields
(the Stark and Zeeman effects). In this case, particles are distributed over sub-
levels, and if the splitting is much less than T , then the populations of the
sublevels are approximately equal to N/g1 , and the population differences are,
according to (3.25), on the order of
N (~/g1 T )N N. (3.27)
Transitions between such sublevels in doped crystals are used in paramagnetic
amplifiers, and relation (3.27) explains why it is necessary to cool the active media
of amplifiers down to helium temperatures.
In the case of molecular gases or solutions of organic dyes, the ground elec-
tronic level has a rich rotational-vibrational structure, which covers the microwave
and the middle-IR spectral ranges. Therefore, the molecules are distributed over
many levels, and the population differences are small as well.
3.2.2 Two-level system and the negative temperature

Consider populations of two non-degenerate levels as functions of the temper-
ature. Let the zero energy be placed exactly between the two levels, so that
E1,2 = ~/2, then it follows from (3.23) that 1,2 = ex /Z, with x ~/2T .
From the condition 1 + 2 = 1 we find that Z = e x + ex and, as a result,
1 = N1 /N = (e2x + 1)1 , 2 = N2 /N = (e2x + 1)1 , (3.28)
N1 N2
= = tanh x. (3.29)
N
The active medium of a laser, in principle, is in a strongly non-equilibrium
state, and the Boltzmann distribution (3.25) is not applicable to it, as is, strictly
speaking, any notion related to temperature. However, in the case of non-
equilibrium systems it is convenient to keep the equations in the form (3.28),
(3.29) but to understand T as some effective parameter. Effective, or spin, temper-
ature for a given pair of non-degenerate levels is defined through the population
ratio as follows:
Nm /Nn exp(~nm /T e f ), (3.30)
i.e., the effective temperature is simply a logarithmic measure of the population
ratio. It follows from (3.30) that in the case of population inversion, T e f < 0.
It is easy to see that Eqs. (3.28), (3.29) maintain their form even for non-
equilibrium systems, provided that T is understood as the effective temperature.
Figure 3.1 shows the relative population difference as a function of the effective
Fig. 3.1 Relative population difference and the entropy S of a two-level system as functions of the
parameter x = ~0 /2T .
temperature. The dependence is plotted according to Eq. (3.29) for all tempera-
tures, both positive and negative. Full inversion (1 2 = 1) corresponds
to T e f = 0, full saturation (1 = 2 = 1/2, = 0) corresponds to T e f = ,
at = 1 T e f = +0. Note that a two-level system with a negative temperature has
more energy stored than a system with a positive temperature:
E = 1 E1 + 2 E2 = (~/2) = (~/2) tanh x, (3.31)
where x ~/2T e f .
The entropy of a non-equilibrium two-level system can be also defined in
terms of T e f . According to definition (3.21) and to (3.28),
S = 1 ln 1 2 ln 2 = ln(2 cosh x) x tanh x. (3.32)
Thus, entropy is an even function of the temperature, with the maximum S 0 = ln 2
at T e f = (Fig. 3.1).
Further, we will show that the intensity of thermal radiation from a two-level
system can be also written in terms of the effective temperature (Sec. 7.1). At
T e f < 0, it is this radiation that causes the noise of quantum amplifiers (the Kirch-
hoff law for negative temperatures). In particular, at ~ |T e f | the noise temper-
ature of an amplifier has the same absolute value as the effective one, T n = |T e f |.

3.2.3 Populations in semiconductors
Boltzmanns distribution (3.25) is not valid for calculating the number of active
particles in the case of inter-band transitions in semiconductors. (Such transi-
tions are used in semiconductor lasers.) In contrast to bound electrons in gases
or in doped dielectric crystals, electrons in the valence and conduction bands of

a semiconductor are not localized and can exchange locations. This possibility
allows one to consider the multi-electron problem and to take into account the
anti-symmetry of the total wave function with respect to the permutation of two
electrons, which leads one to the Pauli principle.
In the first approximation, electrons behave like particles of an ideal quantum
gas with high density. By applying to an ideal gas the general Gibbs distribution,
which has the form (3.23) provided that m numbers all possible states of a multi-
particle system, and taking into account the Pauli principle, we come to the Fermi-
Dirac distribution f (0) (E). For comparing it with (3.25), we will represent this
distribution as (see Fig. 3.2(b))
Nm(0) = 2 f (0) (Em ) = 2{exp[(Em )/T ] + 1}1 , (3.33)
where the factor 2 takes into account spin saturation, is the Fermi level, defined
by the normalization condition Nm(0) = N, with N being the total number of elec-
P
trons, and Em are allowed energy values for one electron. The spectrum is discrete
due to periodic boundary conditions for the electron wave function. According to
(3.33), mean population of any level cannot exceed two electrons, in agreement
with the Pauli principle.
The energy levels Em of electrons in semiconductors have almost continuous
distribution within the allowed bands. As a result, population Nm can be consid-
ered as a function of a continuous argument E, and the normalization condition
P
Nm = N, which indirectly defines the Fermi level, takes the form
Z
dEg(E)N(E) = N, (3.34)
with the integration running over the valence and conduction bands and g(E) being
the energy density of states.
For pure semiconductors, the Fermi level is approximately at the centre of
the energy gap. If there were dopant levels, each of them would contain one
electron; the Fermi level can be formally defined as the one that is half-occupied.
At low temperatures, the boundary between full and empty levels is very sharp
(Fig. 3.2(b)).
In the case of sufficiently high levels, for which E T , one can neglect
the unity in the denominator of Eq. (3.33), and the equation takes the form of the
Boltzmann distribution (3.25),
Nm = 2Z 1 exp(Em /T ) 2, (3.35)
where Z = exp(/T ).
(a) (b) (c)
Fig. 3.2 Population inversion in a semiconductor: (a) relation between the momentum p and the
energy E, i.e., the dispersion relation, for electrons and holes, Eg being the energy gap; light with the
frequency and wavevector k transfers electrons from the E1 level to the E2 level (or vice versa); (b)
populations of energy levels in an equilibrium semiconductor (the Fermi-Dirac distribution); (c) due
to the injection of carriers, the Fermi level splits in quasi-levels v , c , and for some pairs of levels,
inversion takes place: f1 < f2

3.2.4 Inversion in semiconductors
Consider the condition for quantum amplification through inter-band transitions in
semiconductors. The incident field, with the frequency exceeding the gap width
Eg /~, leads to almost vertical transitions of electrons from level 1 of the valence
band to level 2 of the conduction band (Fig. 3.2(a)). The levels 1,2 within the
bands are unambigously defined by the conservation laws of energy, ~ = E2 E1 ,
and momentum (or, to be precise, quasi-momentum), ~k = p2 p1 .
The number of stimulated transitions up scales as the probability of filling
the ground level, N1 /2 = f (E1 ) f1 , multiplied, in accordance with the Pauli
principle, by the probability 1 f2 of a hole being on the excited level. Similarly,
the number of transitions down scales as f2 (1 f1 ), with the same proportionality
factor (see (2.26)). The overall effect of field energy amplification or absorption
scales as the difference,
f1 (1 f2 ) f2 (1 f1 ) = f1 f2 = (N1 N2 )/2. (3.36)
Thus, the contribution of a single pair of resonant levels into absorption scales
as the difference of their populations, similarly to the case of localized electrons,
and the inversion condition has the form
f (E2 ) > f (E1 ). (3.37)

In an equilibrium semiconductor, f = f (0) , and this condition is not satisfied.

However, if, for instance, a sufficient number of carriers (electrons and holes)
is injected into the bands with the help of an external DC current source, then
condition (3.37) can be satisfied, see Fig. 3.2(c). One can easily show that this
requires degeneracy of the carriers in the bands,
c v > ~ > Eg . (3.38)
Here, c , v are the Fermi quasi-levels in the conduction and valence bands. In
addition to the injection method, semiconductor lasers use optical pumping, either
single- or two-photon one, and pumping with an electron beam.
Let us mention here that amplifiers and oscillators using free electrons, such
as gyrotrons, free-electron lasers etc., can be also described in terms of population
inversion (occupation numbers). For instance, in a quasi-monochromatic beam
with the mean energy E0 , only a small group of levels in the vicinity of E0 is
occupied; therefore, inversion takes place with respect to all lower levels, f (E0 ) >
f (E).
3.3 Evolution of the density matrix
3.3.1 Non-equilibrium systems

The density matrix mn of a system, similarly to the distribution function P(q, p)
in classical physics, contains complete statistical information about the proper-
ties of the system, i.e., allows one to calculate ensemble means h f i = Tr( f ),
higher-order moments, correlation coefficients h f g . . . i = Tr( f g . . . ) etc. Ther-
modynamics mainly deals with equilibrium systems where the density matrix and
the ensemble means are time-independent, (0) = h fi(0) = 0. Note, however, that
the correlation functions h f (t)g(t0 )i(0) may depend on the time difference t t0 .
In quantum electronics, on the contrary, of most interest are systems where,
due to the effect of external fields, essentially non-equilibrium state is formed, ,
(0) . If the external perturbation is alternating, V = V(t), then the density matrix
and the ensemble means are naturally time-dependent, = (t), h f i = h f (t)i. On
the other hand, after the external field is switched off, the initially non-equilibrium
((t0 ) , (0) ) system will relax and tend to equilibrium, and its density matrix and
means will be again functions of time. However, the relaxation processes can be
also described by alternating perturbation, V(t), acting on the system from the
thermostat.
Non-equilibrium and non-stationary systems are studied by non-equilibrium
statistical thermodynamics, also called the kinetic theory. In contrast to dynamics,
kinetics studies not the time dependencies of coordinates and momenta of separate
particles, q(t), p(t), or of the wave function, (q, t), but the behavior of the means,
h f (p, q, t)i, distribution functions, P(q, p, t), or the density matrix, mn (t), for sys-
tems interacting with the thermostat and (or) with external alternating fields.
3.3.2 Von Neumann equation

Let us first consider, in terms of the Schrodinger equation, the dynamic problem
about the behavior of the density matrix for a system with a known energy operator
H. For this, we substitute expansion (3.2) in theR Schrodinger equation and left-
multiply the resulting equation by the operator drm . Because the functions
m are orthogonal and normalized, we obtain the following system of equations
determining the dynamics of the bm coefficients:
X
i~bm = Hmn bn . (3.39)
n
Recall that, in contrast to Eq. (2.14), this equation includes the matrix elements
of the total Hamiltonian H rather than the interaction operator V. This is due to
a different definition of state amplitudes. In addition, the basic functions m used
here are not necessarily eigenfunctions of the energy operator.
We multiply (3.39) by bk and write its complex conjugate,
X
i~bk bm = Hmn bk bn ,
n
X (3.40)
i~bk bm = Hnm bk bn .
n
Here, we used the Hermiticity of the energy operator, H + = H. Let us interchange
the m, k indices in the second equation and take the sum of the two equations. As
a result, taking into account the definition of the density matrix for a pure state,
(3.4), we find the following equation of motion:
X
i~mk = (Hmn nk mn Hnk ). (3.41)
n
According to definition (3.10), the equation for a mixed-state density matrix
has the same form. Using the matrix multiplication rule and the commutator no-
tation, [ f, g] f g g f , one can write Eq. (3.41) in a compact invariant form,
i~ = [H, ]. (3.42)
This equation, describing the evolution of the density matrix, is called the von
Neumann equation. It is the starting point for non-equilibrium thermodynam-
ics. Its classical analogue is the Liouville equation for the distribution function
P(q, p, t).
3.3.3 Interaction with the thermostat

In the most general approach, in Eq. (3.42) is understood as the density matrix
of a closed system in a pure state, with the energy containing the following terms:
H = H0 + V, H0 = HA + HB , V = V1 + V2 , (3.43)
where HA and HB are unperturbed Hamiltonians of the system and the thermo-
stat while V1 and V2 describe, respectively, the interaction of the system with
the thermostat, i.e., relaxation, and with the external field. The von Neumann
equation is solved in terms of the perturbation theory, and then averaging over the
variables of the thermostat is performed, see the second definition of the density
matrix (3.15).
In a less rigid approach, only relates to the system under consideration, i.e.,
an atom, a molecule, etc., H0 HA , and V1 is assumed to be a stochastic function
of the time with given statistical parameters. Let the indices k, m, n numerate non-
perturbed energy functions (H0 k = Ek k ), then Eq. (3.41) takes the form
!
d 1 X
+ imk mk = (Vmn nk mn Vnk ), (3.44)
dt i~ n
where the V operator includes the action of the thermostat and the external field.
Note, however, that this approach does not explain the non-equality of the up
and down relaxation transition probabilities, w12 > w21 , see the next section.
Finally, in quantum electronics, as a rule, relaxation is taken into account phe-
nomenologically, using a small number of constants, which are assumed to be
known from a more detailed theory or from experiment.
3.3.4 Evolution of a closed system

Before introducing relaxation parameters into the density-matrix equation, con-
sider the case of a closed system. Let n be eigenfunctions of the energy operator,
then Hmn = En mn , and Eq. (3.41) takes the form
mk = imk mk . (3.45)
Thus, the density matrix of a closed system has a trivial dependence on time,
mk (t) = mk (0) exp(imk t), (3.46)
i.e., non-diagonal elements of the density matric oscillate with the corresponding
Bohr frequencies while the diagonal elements (relative populations) are constant.
Note that this result also follows directly from the exponential time dependence
of the state amplitudes, bn = cn exp(iEn t/~), and the definition of (3.4).
The dipole moment of an isolated atom can be calculated from Eq. (3.46),
X
hd(t)i = Tr{d(t)} = dnm mn (0) exp(imn t). (3.47)
mn
At the same time, it follows from Maxwells equations that an oscillating dipole,
similarly to an aerial, emits electromagnetic waves into free space; therefore,
within a certain time an atom should lose all its energy and get into the ground
state, i.e., mn () = mn n0 . Thus, an atom cannot be isolated from electromag-
netic vacuum, which plays the role of the thermostat with T = 0. This example
reminds us that isolated systems do not exist, and hence (3.45) should be com-
pleted by relaxation terms describing the evolution into equilibrium, () (0) .
3.3.5 Transverse and longitudinal relaxation

The most simple models of relaxation, based, in particular, on the Markovian
approximation, lead to the following form of kinetic equations for the density
matrix (see, for instance, [Fain (1972); Apanasevich (1977)]):
!
d
+ imk mk = mk mk , m , k (3.48)
dt
dm X
= (wmn n wnm m ), m mm . (3.49)
dt n
According to (3.48), non-diagonal elements of the density matric behave like

amplitudes of exponentially decaying oscillators,
mk (t) = mk (0) exp[(imk mk )t]. (3.50)
The damping constant for a given pair of levels, 12 21 , is often denoted as

1/T 2 . The relaxation time T 2 of the non-diagonal component 12 is called the
time of spin-spin, or transverse, relaxation. (The meaning of the second term will
be clarified in Sec. 4.4.)
From the experimental viewpoint, transverse relaxation is usually revealed in
the broadening of spectral lines. (So far, we ignore nonstationary experiments,
which will be considered in Chapter 5.) It will be shown in Sec. 4.2 that (3.48)
leads to a Lorentzian line shape with the FWHM
= 212 = 2/T 2 . (3.51)
In rarefied gases, relaxation is only caused by the interaction of atoms with elec-
tromagnetic vacuum. This interaction leads to the spontaneous emission, with
the probability A12 = 212 , and the corresponding broadening of the upper level
E2 = ~A12 , as well as the spectral broadening, called the natural broadening,
nat = 212 = A12 . (3.52)
If the lower level of the transition under consideration is not the ground one, its
P
broadening has to be taken into account as well. Let 2n m<n Amn be the total
probability of a spontaneous transition from level n to all lower levels, then
mn = m + n . (3.53)
In reality, fnat is on the order of MHz for the case of visible-range allowed tran-
sitions, and T 2 106 s.
For sufficiently dense gases, natural broadening is masked by the collision
one, and T 2 coincides, within an order of magnitude, with the mean time be-
tween the collisions of atoms with each other. As a result, 2/, and the
line width scales as the pressure p, provided, of course, that the Doppler broad-
ening is smaller than the collision one. For rough estimates, one can assume that
at p = 1 mmHg, f 10...100 MHz. Note, however, that under certain con-
ditions, the increase of the pressure leads to line narrowing, 1/p, called
collision, or dynamical, narrowing. One of the models of this effect is considered
in Ref. [Akhmanov (1981)].
Interaction of atoms with the thermostat leads not only to the damping of the
states, but also to a certain shift of the transition frequency. In the case where
the thermostat is a vacuum, this shift is called the Lamb shift. Both effects can
be formally taken into account by substituting a complex value for the transition
frequency mn ,
mn = mn + mn imn . (3.54)
It is important that transverse relaxation is not always related to the energy
transfer to the thermostat. For instance, elastic collisions in a gas lead to random
changes in the phases of complex state amplitudes for separate atoms, b(i) m , and
(i)
their pair products, b(i)
m n b . If these phases are initially equal, mn , 0, then, after
a certain time interval T 2 = 1/mn , which is on the same order of magnitude as
the mean time interval between collisions, the phase will be uniformly distributed
within the interval 0 2, so that mn 0. A similar effect is caused by the
dipole-dipole interaction of neighboring dopant atoms in crystals. Perturbations
of this kind, which do not change populations, are called adiabatic. Certainly,
non-adiabatic perturbations, such as non-elastic collisions, also contribute to the
relaxation of non-diagonal elements, as they change both the amplitudes and the
phases of the coefficients b(i) m.
Let us now consider the relaxation of diagonal density matrix elements, i.e.,
populations. Kinetic equations (3.49) contain a set of phenomenological coeffi-
cients wmn with the dimensionality 1/s. The w21 coefficient defines the rate of
transition from state 1 into state 2 due to the effect of the thermostat. (Recall
that in quantum mechanics, transition indices are read from right to left.) The
role of the thermostat can be played, for instance, by lattice vibrations in crystals,
translational degrees of freedom of atoms in gases, and electromagnetic radiation.
In the case of two-level systems, one denotes
T 1 = (w12 + w21 )1 . (3.55)
T 1 determines the relaxation time of populations, i.e., of the mean energy, and is
called the time of spin-lattice, or longitudinal, relaxation. The time of longitudinal
relaxation depends on the thermostat temperature and varies within broad limits,
from 1012 s in the case of nonradiative optical transitions in condensed matter, to
hours and days in the case of nuclear magnetic resonance. (In this case, interac-
tion with lattice is weak due to the small value of the nuclei magnetic moment,
1023 CGS.) Note that adiabatic perturbations, like dipole-dipole interaction,
do not change the populations; therefore, usually T 1 > T 2 . In experiment, longi-
tudinal relaxation manifests itself in the saturation effect (Sec. 4.3).
Equations (3.48), (3.49) should also involve the case of thermodynamic equi-
librium, where = (0) and (0) = 0; hence, the following relation should hold
true:
X
(wmn n wnm m ) = 0. (3.56)
n
This equality is satisfied, in particular, if one assumes the principle of detailed

equilibrium,
wmn n = wnm m . (3.57)
Hence, taking into account the Boltzmann distribution, we find the relation be-
tween the probabilities of relaxation transitions and thus reduce the number of
independent parameters in Eq. (3.49) by a factor of two,
wmn /wnm = exp(~nm /T ), (3.58)
with T being the temperature of the thermostat. This condition provides dy-
namical equilibrium for the populations. Thus, w12 > w21 , in contrast to the
case of stimulated transitions in a classical field, where, according to (2.26),
W12 = W21 . When the thermostat is at low temperature, it has no excitations
(photons, phonons, etc.) with high energy, ~ > T . Therefore, it can only absorb
energy from the system under consideration, and transitions up are practically
absent. An extreme example is realized for transitions between nuclei levels in
the range. Even in condensed matter, such transitions usually occur only due to
spontaneous emission, with the probability w12 = A = 1/T 1. In the case of nuclear
isomers, A is extremely small because the transitions are forbidden, and T 1 can be
as large as days, similarly to the NMR case.
The parameters wmn can be calculated, in principle, using some model of the
thermostat. An example where the role of the thermostat is played by the field has
been considered in Sec. 2.5. Then,
w21 = B, w12 = B + A, (3.59)
1/T 1 = A coth(~21 /2T ) = A/(0) , (3.60)

where A, B are the Einstein coefficients for spontaneous and stimulated transitions,
(0) (21 ) is the spectral density of the equilibrium field, given by the Planck
formula, and (0) is the equilibrium relative population difference (see Eq. (3.29)).
3.3.6 Interaction picture

Usually, one has to solve the von Neumann equation for the density matrix by
means of the perturbation theory, i.e., a sequence of iterations. The only exception
is the case of a two-level system, which will be considered in Sec. 4.3. As in the
case of solving the Schrodinger equation in the energy representation (Sec. 2.1),
we will assume that the influence of the external alternating field on the electrons
in an atom is much weaker than the effect of the nucleus constant field, which
determines the unperturbed stationary states of a bound electron. As we will show
in what follows, a more precise formulation of the condition for the perturbation
theory to be valid has the form , where = |dmn E0 |/~ is the Rabi
frequency, i.e., the matrix element of the perturbation energy in frequency units,
and is the mismatch between the field frequency and the closest atom frequency,
i.e., the energy deficit | mn | in the virtual state (Sec. 6.2) or the transition
bandwidth mn given by the relaxation.
Before solving the von Neumann equation, it is convenient to transfer the triv-
ial time dependence of the unperturbed density matrix to the operators. To do
this, let us introduce the following notation for the matrix elements of an arbitrary
operator in the energy basis:
Z
0
fmn = fmn exp(imn t) = drm (r, t) f n (r, t), (3.61)
where the functions n = n exp(iEn t/~) satisfy the equation i~n = H0 n .

Transformation of the matrix elements (3.61) corresponds to the following unitary

transformation of the operators:
f 0 (t) = U0+ f U0 , (3.62)
U0 (t) exp(iH0 t/~), U0+ U0 = I. (3.63)

The unitary operator U0 is called the unperturbed evolution operator; it is diago-
nal in the energy representation and has the eigenvalues exp(iEn t/~), so that
n (t) = U0 (t)n , (3.64)
n = U0+ (t)n (t). (3.65)

In the Dirac notation, the time evolution of a state vector, for the case V = 0,
is described as
|ti = U0 (t t0 )|t0 i. (3.66)
The inverse transformation has the form
| i0 |t0 i = U0+ |ti. (3.67)
Let us now substitute into the von Neumann equation (3.42), for mn and Vmn ,
the primed values, according to (3.62), and take into account that mn +nk = mk .
As a result, we get the equation for the density matrix in the interaction picture,
also called the Dirac picture,
X
i~0mk = (Vmn0
0nk 0mn Vnk
0
). (3.68)
n
In the invariant notation, it is written as

i~0 = [V0 , 0 ]. (3.69)
Note that the time dependence of an arbitrary operator f is given by the
Heisenberg equation,
i~ f = [ f, H]. (3.70)
Here, it is assumed that f does not depend on the time directly, i.e., f /t = 0.
Transformations of operators of the form (3.63) accompanied by transforma-
tions of state vectors of the form (3.67) means passing to the interaction picture,
and in the case V = 0, to the Heisenberg picture. These transformations are
similar to passing to a rotating frame of reference.
In the initial Schrodinger picture, state vectors, and also, according to def-
inition (3.4), the density matrix elements are functions of time. Operators can
depend on time only due to a varying external force (as, for instance, the energy
operator V(t) = d E(t) in dipole interaction). On the contrary, in the Heisen-
berg picture all time dependence is transferred to the operators and their matrix
elements, except the density matrix operator, and the state vectors are constant.
The interaction picture is intermediate, and it has all values time-dependent.
However, it is important that the observables do not depend on the choice of
the picture,
h f i = Tr( f ) = Tr( f 0 0 ). (3.71)
This can be proven using definition (3.63), the unitarity, U0 U0+ = I, and the in-
variance of the trace to cyclic permutations, Tr(abc) = Tr(bca).

3.3.7 Perturbation theory
It is not difficult to find the formal solution to the von Neumann equation (3.69)
using the iteration method. For this, let us represent the density operator as a series
expansion (the primes will be temporarily omitted),
(t) = (0) + (1) (t) + (2) (t) + . . . (3.72)
and substitute it in (3.69). Here, (0) = (t0 ) is the initial condition. By setting
equalities between the terms of the same order in the perturbation V, we find the
relation
i~(k) = [V, (k1) ]. (3.73)
Integration yields
Z t Z t2
(k) k
(t) = (i~) dtk . . . dt1 [V(tk ), . . . [V(t1 ), (0) ] . . . ]. (3.74)
t0 t0
From this, we find the mean value of an arbitrary operator,

X Z t Z tk1
h f (t)i = (i~)k dt1 . . . dtk
k=0 t0 t0
h[. . . [ f (t), V (t1 )], . . . , V0 (tk )]i(0) .

0 0
(3.75)
In the last expression, averaging is over the initial (unperturbed) density matrix
(0) ; the initial time moment t0 is usually assumed to be . Eq. (3.75) was
derived using the property Tr(ab) = Tr(ba), which leads to the following equalities
under the Tr operation:
a[b, c] = [a, b]c, a[b, [c, d]] = [[a, b], c]d. (3.76)
Relation (3.75) determines the response (reaction) of a quantum system to an

external perturbation. For instance, assuming f = d , V(t) = d E(t) one can
find the mean dipole moment of an atom, i.e., the charge displacement due to a
given electric field, in the form
hd(t)i = E + E2 + E3 + . . . , (3.77)
where , , are some integral operators whose structure is clear from (3.75). Ex-
pansion of d(t) and E(t) in Fourier integrals or series determines the polarisability
tensors (), (, 0 ), . . . of an atom. Further, by multiplying atom polarisabil-
ities by the density N of the atoms, one can find the macroscopic susceptibility
tensors of the matter, (1) (), (2) (, 0 ), . . .
As a result of such calculations, some of which will be demonstrated in
Secs. 4.2 and 6.2 below, polarization P = Nhdi of the matter can be expressed
in terms of the external field and the parameters of the atoms, dipole matrix ele-
ments dmn and the transition frequencies mn .

Chapter 4
The Susceptibility of Matter
In classical electrodynamics, the interaction between matter and a field is con-

ventionally studied in two steps, the microscopic one and the macroscopic one.
The microscopic part is focused on the behaviour of charged material particles
in a given external field. As a result, one finds the averaged, macroscopic, pa-
rameters of the matter, such as the susceptibility tensor , which determines the
polarization P = E of homogeneous matter caused by the field, or the dielectric
function = 1+4. As a result of frequency Fourier expansion of the field, these
quantities become complex functions of the frequency, () = 0 () + i 00 ().
In the macroscopic approach, the susceptibility of the matter is assumed to be
known, and the emission and propagation of the field are studied using Maxwells
macroscopic equations.
The present chapter considers first the definition and the general properties
of the susceptibility tensor (Sec. 4.1). Further, in Sec. 4.2 the susceptibility is
calculated in the framework of the simplest model of identical, motionless and
non-interacting molecules. Both classical and quantum theory is used in this case.
Section 4.3 considers saturation, the most important effect in nonlinear optics,
in which populations of two levels get balanced due to a strong resonant field.
Finally, in Sec. 4.4 the Bloch equations, which are widely used in quantum elec-
tronics, are derived.
4.1 Definition and general properties of susceptibility
By definition, linear dielectric susceptibility () is the proportionality factor be-

tween a monochromatic macroscopic field E() at frequency and the polariza-
tion P() emerging in a homogeneous medium due to its effect. In an anisotropic
medium, polarization can be non-parallel to the field, so that in the general case,
67
susceptibility is a tensor,
X
P () = ()E (), (4.1)

with , = x, y, z. Usually, the summation sign is omitted. In more compact

notation,
P() = ()E(). (4.2)
4.1.1 Symmetry
Here, E() denotes the Fourier component of the field E(t). As it is common in
physics, a function of time f (t) and its Fourier transform f () are denoted by the
same symbol and only their arguments differ,
Z Z
f (t) = deit f (), f () = dteit f (t)/2. (4.3)
The absence of the integration limits means that they are . Note that if f (t) is
real, then (4.3) leads to the following symmetry property for f ():
f () = f (). (4.4)
Thus, the real part, f 0 (), is an even function while the imaginary part, f 00 (),
is an odd function, so it is sufficient to know f () for positive frequencies. From
the definition (4.2), it follows that all components of the susceptibility tensor also
satisfy (4.4),
() = (). (4.5)
The () tensor and its inverse Fourier transform, (t), called the Green func-
tion or the response function, have one more general property, typical for arbitrary
physical systems: they are symmetric, = , or
= . (4.6)
Here, is the transposed tensor, . This equality is an example of the

general Onsager symmetry principle for kinetic indices. It is also confirmed by the
microscopic theory (see (4.59)). The symmetry of is only violated in the case of
optical activity, either natural or caused by a constant magnetic field. In the latter
case, instead of (4.6) we have (H0 ) = (H0 ).
Additional relations between different components of are imposed by the
symmetry of the medium. For instance, in crystals with cubic symmetry, =
, as in isotropic media.
The Susceptibility of Matter 69
4.1.2 The role of causality

Dependence of susceptibility on the frequency, (), cannot be arbitrary. As we
will see, its real and imaginary parts, 0 () and 00 (), are related via the Hilbert
transformation.
Consider polarization induced in a dielectric by a very short field pulse,
E(t) (t), so that the pulse duration is much less than the period of the most
high-frequency eigenmode of the matter. In the spectrum of such a pulse, all
frequencies are distributed uniformly, E() = const, and the spectrum of the
polarization, according to (4.1), repeats the shape of (): P() (). (For
simplicity, the medium is considered as isotropic.) Hence, the polarization pulse
P(t) repeats the shape of the Fourier transform of the complex susceptibility,
Z
P(t) deit () 2(t).
Apparently, the system cannot respond before the external force is turned on;
hence, (t) should turn to zero at t < 0,
Z
deit () (t), (4.7)
where (t) is the Heaviside step function, which is unity at t > 0 and zero at t < 0.
The causality principle, according to (4.7), restricts considerably the allowed
class of () functions. Indeed, it follows that (), considered as a function of a
complex frequency = 0 + 00 , should be analytical in the upper semiplane. Let
us calculate the integral in (4.7) using the residue theory. The integrand contains
00
the factor e t ; therefore, at t > 0, the integral should run along a contour in
the lower semiplane (see Fig. 4.1), while at t < 0, the contour should be in the
upper semiplane. However, due to the causality principle, at t < 0 the integral
should turn into zero. Therefore, the () function cannot have poles in the upper
semiplane (see, for instance, [Vinogradova (1979); Landau (1982, 1964)]).
Further, according to the integral Cauchy formula, the real and imaginary parts
of an analytical function are related via the Hilbert transformations,
00 (1 ) 0 (1 )
Z Z
0 () = PV d1 , 00 () = PV d1 , (4.8)
1 1
where PV denotes the principal value of an integral. These integral equations
are called the Kramers-Kronig relations. They allow, for instance, the real part of
susceptibility to be calculated from the measured imaginary part. The above-given
derivation can be extended to the case of an anisotropic medium. Then, equations
(4.8) will be valid for all components of the tensor.
Fig. 4.1 Proof that the susceptibility of a dielectric () is an analytical function of complex fre-
quency in the upper semiplane: at t < 0, the response function (t) turns to zero due to the causality
principle. At the same time, it is equal to the integral of ()eit along the C contour. Hence,
according to the Cauchy theorem, C should not contain poles of ().
4.1.3 Absorption of a given field

In the linear optics approximation, susceptibility completely determines emis-
sion, propagation, and absorption of a macroscopic field in a homogeneous
medium, as well as the properties of surface waves, refraction and diffraction at the
boundaries. Moreover, according to the fluctuation-dissipation theorem (FDT),
also determines the equilibrium thermal field in matter (Sec. 7.7).
Let us show that the imaginary part of the susceptibility, 00 (), determines
the power of radiation absorbed or, at 00 < 0, emitted by the matter. We start
from macroscopic Maxwells equations for a linear non-magnetic medium with
D = E + 4P = (I + 4)E and B = H,
c rotH D = 4j, (4.9)
c rotE + H = 0, (4.10)
divD = 4, (4.11)
divH = 0, (4.12)
where j and are external (given) densities of current and charge.
At j = 0, the power P absorbed by a unit volume of the matter, due to the
energy conservation law, should be opposite to the divergence of the energy flux
density S,
P(t) = c div(E H)/4 = c(E rotH H rotE)/4.
From (4.9, 4.10), it follows that
P(t) = (E D + H H)/4.
In the case of a monochromatic field, this expression contains terms oscillating
at a double frequency. After time averaging, it turns into zero, so that the mean
power per period is
P P(t) = E P = Im(E0 P0 )/2 = i(E0 E0

E0 E0 )/4 (4.13)
Let us interchange the , indices in the first term and take into account that in a
non-gyrotropic material, susceptibility (4.6) is a symmetric tensor, then
P = 00 E0

E0 /2 E0 00 E0 /2. (4.14)
In the case of a gyrotropic material, the imaginary part of in (4.14) should be
replaced by the anti-Hermitian part, ( + )/2i. For an isotropic medium or a
cubic crystal, (4.14) takes the form
P = 00 |E0 |2 /2. (4.15)

4.1.4 Susceptibility of the vacuum
Further, let us find the field generated in a homogeneous medium by external
sources, i.e., a given polarization with a harmonic variation in time and space,
P = (1/2)P0eikrit + c.c. (4.16)
Here, k and are independent variables. In a homogeneous medium, P induces
a plane monochromatic wave with amplitudes E0 , H0 . Let us substitute (4.16) in
(4.9, 4.10) and take into account that j = P. We get a system of algebraic equations
for E0 , H0 (n ck/),
n H0 + E0 = 4P0 , (4.17)
n E0 H0 = 0. (4.18)
Excluding H0 , we get
n (n E0 ) + E0 = 4P0 . (4.19)
Double vector product in (4.19) projects the E0 vector onto the plane orthog-
onal to the propagation direction k. Let us denote this projection operation by .
Apparently, the tensor has the components
= k k /k2 . (4.20)
As a result, Eq. (4.19) takes the form
(n2 ) E0 = 4P0 . (4.21)
Thus, the problem is reduced to solving a system of two linear non-
homogeneous algebraic equations. The solution can be expressed, in a standard
way, in terms of the minors and determinant of the matrix (n2 ) . Instead
of solving the system directly, we express E0 in terms of P0 using the formalism
of the inverse matrix or tensor. By definition, A A1 = A1 A = I; therefore, it
follows from (4.18) and (4.21) that
E0 = G P0 , H0 = n (G P0 ), G 4(n2 )1 . (4.22)
The G (k, ) tensor is called the spectral Green function for Maxwells equa-
tions. It determines a macroscopic field induced by polarization (4.16), i.e., the
response of the electromagnetic vacuum in matter to an external excitation. The
tensor function G(k, ), similarly to or , satisfies the Kramers-Kronig relations.
Its Fourier transform, G(r, t), determines the fields emerging in a homogeneous
medium due to an arbitrary distribution of the polarization P(r, t) or a current.
Consider the case of a homogeneous medium. Let the z axis be along k, then
from (4.20) and (4.22) we find
E0x,y = 4P0x,y/(n2 ), (4.23)

E0z = 4P0z/; (4.24)
recall that here n ck/. The last equation for the field E0z , which is longitudinal
with respect to the propagation direction, shows that it is independent of k: we
usually exclude the effects of spatial dispersion where = (k, ). According
to (4.24), longitudinal field created by this polarization is maximal at frequencies
where |()| is minimal; these frequencies are given by the condition 0 () 0.
Note that the Green function for a longitudinal field can be also obtained from
Eq. (4.11) by assuming that = divP.
Transverse components of the field, E0x,y , considered
as functions of k, ac-
cording to (4.23), have a wave resonance at k = 0 /c, i.e., at n2 = 0 . Then,
G xx = Gyy = i/00 , and the radiation power is (c.w. (4.15))
P = (|P0x |2 + |P0y |2 )/200 . (4.25)

4.1.5 Thermodynamic approach
Between the ranges of strong absorption in matter, there are transparency win-
dows where one can neglect the energy dissipation, i.e., assume |00 /0 | 1.
(We consider the matter to be non-gyrotropic.) In the absence of dissipation, the
vibrational energy of the particles caused by the external field is conserved; hence,
the work of polarization can be defined as a function of the field amplitude, A(E0 ).
This work consists of displacing the charges and is performed by the sources of
alternating fields. Note that for A to be defined, a finite time is necessary for a
stationary amplitude P0 of polarization oscillations to be formed, which is only

possible in the presence of some finite absorption. After introducing the notion of
work A(E0 ) we can consider E0 as one of the thermodynamic parameters defining
the state of the matter, in addition to entropy S , density , etc. In the framework of
this approach, one can formulate thermodynamical definitions for the polarization
P0 (S , , E0 ) and susceptibility (S , , E0 ) as functions of the state of the matter.
In transparency windows, dispersion is small as well; therefore, polarization
follows the field almost instantaneously,
P(t) ()E(t), (4.26)
where is the central frequency of a quasi-monochromatic field. In the case of an

optical field, () is certainly defined without accounting for the inertial mecha-
nisms of polarization, for instance, orientation of the molecules by an alternating
field (Sec. 6.2). Such mechanisms only contribute to the static and radio-frequency
susceptibilities.
Let us first ignore the dispersion completely. Then the state of the matter has a
time dependence only via the field E(t). Then, in (4.13) one can assume P = E,
so that the variation rate of the macroscopic field (per unit volume) takes the forma
d E2 + H2 1
!
P(t) = (EE + HH)/4 + E E = + EE . (4.27)
dt 8 2
In the last equation, we have used the symmetry of the tensor. The expres-
sion in brackets is obviously the energy density of the macroscopic field, the first
term being energy of the field in the vacuum, at the same E, H, and the second
one having the meaning of additional work A performed by a field source in the
presence of the matter. Additional energy of the matter in a given field has the
opposite sign,
v = E E/2. (4.28)
Strictly speaking, the macroscopic field E inside the matter should be replaced
here by the external field E0 in the absence of the matter (see Ref. [Landau (1982)],
Sec. 11), but for the sake of simplicity we ignore the difference between E and E0 .
Equations (4.26)(4.28) assume a linear relation between P and E, which is
valid only for a sufficiently weak field. An evident generalization of (4.28) is
dv = P(S , , E) dE, (4.29)

a Taking dispersion into account leads, in the linear approximation, to replacing in (4.27) by
d()/d [Landau (1982)].

or
Z E
v= P(S , , E) dE. (4.30)
0
Polarization P and, hence, the elementary work of polarization, dv, certainly
depend not only on E but on the other parameters defining the state of the matter.
Therefore, the integral in (4.30) is along a curve, and in order to find v, this curve
should be specified. The work of polarization can be defined at constant entropy S
and density , i.e., for a thermally isolated material with a given concentration of
molecules N = /m. In this case, polarization will not change the internal energy
of the matter per unit volume in the absence of the field, U0 (S , ) (by definition,
dU0 = T dS + d, being the chemical potential). Therefore, the internal energy
of the matter in the presence of the field is
U(S , , E) = U0 (S , ) + v(S , , E), (4.31)
where E plays the role of an external thermodynamical parameter.
Now, one can define polarization and susceptibility thermodynamically as
functions of the state of the matter,
P (S , , E) U/E , (4.32)
(S , ) (2 U/E E ) E=0 . (4.33)

Thus, by defining the tensor in terms of the thermodynamical potential one
can provide its symmetry. In (4.27)(4.32), one can assume E = E(r, t) if the dis-
persion is neglected; hence, the state of the matter depends on time and coordinate
as parameters.
Further, doing the Taylor expansion of the internal energy U(E) or the energy
of adiabatic polarization v(E), near the E = 0 point one can define the nonlinear
polarization and the nonlinear susceptibility tensors (n) (Sec. 6.1).
It is often convenient to use, instead of U, other thermodynamic potentials
such as, for instance, the free energy F(T, , E). The field part of F, which has
the meaning of the work of polarization, vF , should be calculated at constant tem-
perature, so that in the general case vF , vU . However, in weak fields, the field
parts of all potentials are the same (see Ref. [Landau (1964)], Sec. 15) and equal
to v(E). As a result, various macroscopic effects in electromagnetic field, such as
electrostriction, electrocaloric effect etc., are determined by partial derivatives of
in density, temperature, and so on (Sec. 6.2).
Let now the field be quasi-monochromatic, then should be replaced by ().
Thus, transmission of light by transparent matter leads to an increase of thermo-
dynamic potentials by a value of
v(t) = [E0 () E0 + E0 () E0 ei2t + c.c.]/8. (4.34)
Only the constant, or slowly varying, part of the potential is of practical interest,
v = E0 () E0 /4. (4.35)
This expression for the effective potential of the matter in a monochromatic
field describes, according to the known thermodynamic equations, the effect of
light on the state of the matter. Variation of the state (temperature, density, etc.) of
the matter, in its turn, influences and the transmitted light, i.e., causes a nonlinear
optical effect (Sec. 6.2).
Note that, according to (4.35), one can define the polarization amplitude and
the susceptibility in terms of the effective potential,

P0 = 4v/E0 , = 42 v/E0 E0 . (4.36)
Let the density of the molecules be N, then in the approximation of non-
interacting molecules, polarizability of a single molecule is = /N, and from
(4.35) one can find the effective potential of a molecule in an alternating field,
1
V = E0 () E0 . (4.37)
4
This potential defines the mean force of light pressure acting on a molecule in a
monochromatic field in terms of the molecule polarizability,
F = V = (E0 () E0 )/4. (4.38)
This expression can be transformed as
1 1 E0
F = (E0 E0 )= d + c.c.
4 x 4 x 0
1
= d E0 (r) + c.c. = d(t) E(r, t). (4.39)
4 x 0 x
The factor 1/2 is absent here since we assume that the operator does not act on
the dipole moment d = E of the molecule. The force (4.39) corresponds to the
potential V(r) = d E(r). Light pressure will be considered in more detail in
Sec. 6.2.
4.2 Dispersion theory
4.2.1 Dispersion law

In the transparency windows, 00 = 0 and, according to (4.23), the Green function

turns to infinity at n = . Usually, it is this resonant value of the ratio ck/
that is denoted by n and called the refractive index.
The same condition provides the existence of a nontrivial solution (E0 , 0)

to homogeneous (P0 = 0) Maxwells equations. Therefore, a medium without
sources can only support propagation of waves with a certain relation between
the wavelength, = 2/k, and the frequency. This relation,
k() = [()]1/2 /c, (4.40)
or the inverse of it, (k), is called the dispersion law, and the waves satisfying
it are called free, or normal, ones. The condition for normal longitudinal field to
exist is () = 0.b It follows from (4.40) that the phase velocity of transverse

normal waves is times as small as the speed of light. The group velocity, as
we know, is given by the derivative d/dk u. It follows that longitudinal waves
do not propagate, since, according to (4.24), (k) = const and u = 0. (Here, we
again neglect the effect of spatial dispersion.)
In an anisotropic medium, the condition for normal waves to exist, or for the
Green function to turn to infinity, according to (4.22), has the form
det(n2 ) = 0. (4.41)
This condition is called the Fresnel equation. With the frequency and the
wavevector direction k/k fixed, Eq. (4.41) only has solutions for twoc particular
directions of the polarization vector e ( = 1, 2). In the general case, the polar-
ization vector is not orthogonal to k and may be complex, which corresponds to
the elliptical polarization of the normal wave (for more details, see Ref. [Landau
(1973)]). The two normal waves have different dispersion laws (k), which leads
to birefringence. In an anisotropic medium, the group velocity vector u is equal
to (k) and, in the general case, is not parallel to the phase velocity vector.
4.2.2 The effect of absorption

With an account for absorption, the Fresnel equation has solutions only for com-
plex and/or k. The choice depends on the particular problem. A stationary
experiment corresponds to a real frequency and a complex propagation constant.
If the wave vector is complex, a free monochromatic wave gets either damped
or amplified in the course of propagation. Let us make a replacement in (4.40),
k k k + i/2, then the dispersion law of a normal transverse wave in an
isotropic medium takes the form (k + i/2)2 = ( 0 + i 00 )2 /c2 . Hence,
k2 2 /4 = 2 0 /c2 , k = 2 00 /c2 ,
b We neglect the effects of spatial dispersion, which can be described by a (k) dependence (see, for
instance, Refs. [Vinogradova (1979); Landau (1982)]).

c Effects of spatial dispersion may double the number of normal waves at a given frequency. The
corresponding waves are called new ones (Fig. 4.5).

or
!1/2
|| + 0
k= Re = , (4.42)
c c 2
!1/2
|| 0
/2 = Im = . (4.43)
c c 2
The sign by the square root is chosen from physical considerations. These equa-
tions define the positions of the two poles of the Green functions G(k, ) in the k
plane.
In the case of weak absorbtion, 2 << k2 , Eqs. (4.42), (4.43) take the form

k = 0 /c, (4.44)
= k 00 / 0 . (4.45)
It should be stressed again that the dispersion law (k) and a fixed polariza-
tion e only take place for free waves, i.e., waves generated by distant sources. In
the presence of given sources, the spatial and temporal dependencies of stimu-
lated field are determined by the distribution of the currents and can be arbitrary.
In particular, thermal fluctuation field inside the matter is created by the chaotic
motion of charged particles and the field at a given frequency is a superposition of
plane waves of various lengths. Note that waves with maximal amplitudes do not
always satisfy (4.44) (see Eq. (4.79)).
Thus, the macroscopic theory enables all basic observable rules of waves emis-
sion, propagation, and absorption through the phenomenological function ().
The next step is calculating () in the framework of the microscopic theory.
This is a traditional problem of non-equilibrium thermodynamics, and its com-
plete solution is still absent.
4.2.3 Classical theory of dispersion

In order to find the order of magnitude and the dispersion dependence for linear
dielectric permittivity, let us use the simplest model of the matter as a set of inde-
pendent, motionless, and identical atoms or molecules. Due to alternating electro-
magnetic field, the electron cloud of a molecule oscillates (the nuclei are assumed
P
to be motionless), and the molecule gains the dipole moment d(t) = e ri (t),
which, in the first approximation, scales as the field. Here, e > 0 is the electron
charge and ri is the radius vector of the ith electron. As a rule, magnetic dipole
moment, quadruple moment and higher-order moments can be ignored since the
scale of the spatial field variation, > 105 cm, in the optical range exceeds much
the typical size of a molecule, a0 108 cm. The product of d and the con-
centration of molecules N is equal to the dipole moment per unit volume, i.e.,
polarization, P = N d = E.
Thus, the problem of calculating the susceptibility is reduced to calculating
the dipole moment of a molecule induced by an external field.
Thermal motion of charges can be taken into account in the framework of the
kinetic theory. In the quantum theory, ri and, hence, d, are operators; therefore,
one should do both quantum and statistical averaging, i.e., use the density matrix
formalism.
Consider first the classical Lorentz model, which represents a molecule as an
oscillator. The equation of motion of a linear isotropic oscillator has the form
r + 2r + 20 r = eEloc /m, (4.46)
where m, 0 and e are the effective mass, frequency, and charge of the oscillator,
respectively, is the phenomenological damping constant, and Eloc is the field at
the centre of the molecule, known as the local field. After multiplying (4.46) by
eN, we find the equation of motion for the polarization,
P + 2 P + 20 P = 2p Eloc /4, (4.47)
where p = (4e2 N/m)1/2 is the so-called plasma frequency.
The field Eloc at the centre of a motionless molecule differs from the space-
averaged macroscopic field E. According to Lorentz,
4 +2
Eloc = E + P= E, (4.48)
3 3
so that (4.47) takes the form
P + 2 P + 20 P = 2p E/4, (4.49)
20 20 2p /3. (4.50)
Hence, assuming the field to be monochromatic, we find
2p /4
= . (4.51)
20 2 2i
In what follows, we will assume that the eigenfrequency shifts (4.50) due to the
Lorentz correction are included into the definition of 0 .
Suppose now that there are several types of independent oscillators with eigen-
P
frequencies j and concentrations f j N, f j = 1, then
2p X fj
= . (4.52)
4 j
2j 2 2i
The parameter f j is called the oscillator strength. A similar expression, which

in many cases describes well the observed dispersion of susceptibility, will be
obtained below using the quantum theory.
Note that for very high frequencies or in the case of free electrons in a plasma
or metal, one can assume in (4.52) j , so that
2p
1 .
( + 2i)
4.2.4 Quantum theory of dispersion

Let us now start from the kinetic equations for the density matrix (3.48), (3.49)
with the phenomenological relaxation parameters mn , wmn . In the dipole approx-
imation, the perturbation energy V = d E, and its matrix elements in the case
of a monochromatic field have the form
Vmn (t) = ~mn eit /2 + h.c., (4.53)
where
mn dmn E0 /~. (4.54)
Notation h.c. stands for the Hermitian conjugate matrix,
dnm E0 eit /2 = dmn E0 eit /2.
Monochromatic perturbation will cause, in the linear approximation, the same

response; therefore, let us seek the density matrix in the form
it
(1) (1)
mn (t) = mn ()e + h.c. (4.55)
In the zeroth order of the perturbation theory, the density matrix is diagonal, (0)
mn =
(0)
m mn , so that, after substituting (4.53) and (4.55) in (3.48), (3.49), we find for
m,n
mn (0)
nm /2
(1)
mn () = , (4.56)
mn imn
where nm n m is the relative population difference for levels n and m.
The diagonal elements (1)
nn , according to (3.48), (3.49), will scale as the inverse
frequency of the perturbation; if one is only interested in resonance effects,
under the condition / 1 one can assume (1) nn = 0. Thus, the amplitude of
the response to a harmonic perturbation scales as the population difference and
reaches its maximum at resonance, = mn .
After substituting (4.56) in (3.6), we find the dipole moment of the molecule
and the polarization,
1
P = Nhd(t)i = P0 eit + c.c.
2
N X (0) nm d nm (dmn E0 )
P0 = .
~ mn mn imn
Hence, according to definition (4.1),
() ()
N X (0)nm dnm dmn
= , (4.57)
~ mn mn imn
where d() d is the projection of the molecule dipole moment onto the axis
= x, y, z.
One can easily verify that the obtained expression has the necessary symme-
try (4.5) and satisfies the causality principle (Fig. 4.1). Note that (4.57) can be
represented in a somewhat different form,
X
= Nn (n)
(),
n
() () () () (4.58)
1 X dnm dmn dmn dnm
(n)

() + .
~ m mn imn mn + + imn

Here, (n) has the meaning of the polarizability tensor of a molecule in state n.
In the absence of a static magnetic field, unperturbed wave functions and, hence,
the matrix elements dmn = dnm , can be considered to be real (see Ref. [Landau
(1964)]). Then, according to (4.6), (4.58) is invariant to the permutation of ,
indices,
() ()
2X mn dmn dmn
(n)
= (n)
= . (4.59)
~ m mn ( + imn )2
2
Using (4.57) and (4.14), one can easily show that the contribution of each pair
of levels (m,n) into the field energy is positive or negative depending on the sign
of mn nm , i.e., amplification of the field requires population inversion, see also
(4.60).
In the case of a gas, (4.57) should be averaged over random orientations and
velocities of the molecules. Due to orientation averaging, non-diagonal elements
() 2
of the d d turn into zero, and the diagonal ones become |dmn | = |dmn |2 /3. As a
result, the susceptibility tensor (4.57) becomes a scalar,
2N X mn (0) nm |dmn |
2
= , (4.60)
3~ m>n 2mn ( + imn )2
In the last equation, we have taken into account that the double sum contains twice
each term with m , n,
X X X
amn = ann + (amn + anm ), (4.61)
mn n m>n
and the diagonal terms in (4.60) are zero since nn = 0.

4.2.5 Oscillator strength
In order to compare (4.60) with the classical expression (4.52), let us define di-
mensionless oscillator strengths,
fmn 2mnm |dmn |2 /3~e2 . (4.62)
Note that the oscillator strength can be also defined phenomenologically, in terms
of 00 , see Ref. [Landau (1982)].
Let us number possible pairs of states (m, n) by a single index j {m, n}, as-
suming m > n. If we neglect the 2 terms in the denominator of (4.60) and put
f j fnm (0)
nm , Eq. (4.60) takes the form of (4.52). Therefore, quantum calcula-
tion confirms the Lorentz model: in the first approximation in the amplitude of
the external field, matter behaves like a set of linear oscillators with damping.
However, f j may now take negative values, which manifests itself in the effects of
quantum amplification (00 < 0) and negative dispersion (0 / < 0 outside of
the resonance).
Recall that, according to (4.7), a field -pulse causes a pulse of polarization
shaped as the Fourier transform of (). According to (4.60), the poles of (),
understood as the function of a complex frequency, are at points j = j i j
in the lower semi-plane; therefore, the polarization pulse is a sum of damped
harmonic oscillations,
2p X f j
(t) = (t) exp ( j t) sin ( j t). (4.63)
4 j j
This expression defines the Green function for the polarization of the matter in
terms of the eigenfrequencies and the oscillator strengths of the transitions.
The oscillator strengths satisfy the sum rules. For instance, for single-electron
transitions,
X
fnm = 1. (4.64)
m
This equation can be obtained from the commutation rule [x, p] = i~, p p x .
Let H0 = p2 /2m + V(r), then
[x, H0 ] = i~p/m, pmn = immn xmn ; (4.65)
hence,
mn |xmn |2 = ixmn pnm /m = ipmn xnm /m,

X
mn |xmn |2 = i[p, x]nn /2m = ~/2m.
m
From the last equation, we obtain (4.64).

For most strong optical transitions in atoms, | fmn | 1. For instance, for the
resonance line of atomic hydrogen, f = 0.416 (the 1s2p transition, = 0.12).
Hence, according to (4.62),
|xmn | = (oc omn fmn /2)1/2 = 108 cm, (4.66)
which corresponds to |dmn | = 4.8 1018 CGSE = 4.8 D. Here, oc ~/mc

41011 cm is the Compton wavelength. For allowed transitions between rotational
levels in the millimeter range, dmn is also on the order of 1D, but in this case,
according to (4.62), fmn 105 . Note that the sum in (4.64) should also include an
integral over the continuous spectrum of ionized states; for instance, in hydrogen
the ionized state for n = 1s has a contribution of f = 0.43.
4.2.6 Isolated resonance

In the vicinity of a narrow isolated resonance, only a single term in the double
sum of (4.60) has to be taken into account,
20
= + . (4.67)
20 2 2i
Here, Nd2 /3~ = f 2p /80 and is the contribution of other reso-

nances, which is real. In the denominator of (4.67), the term 2 has been omitted;
similarly to the Lorentz correction (4.50), it can be incorporated into the defini-
tion of 0 . The parameter , scaling as the product of the active particle density,
N, and the squared dipole moment of the transition, d2 , determines the maximal
value of 00 and the amplitude of 0 variation (Fig. 4.2).
In the optical range, the Q-factor of a resonance is usually high, 0 /2 1;
therefore, in the close vicinity of a resonance one can use a simple approximate
formula,

= , (4.68)
x+i
with
x ( 0 )/, 0 > 0.
Fig. 4.2 Dispersion dependence of susceptibility () in the vicinity of an isolated resonance, for
the Q-factor 0 /2 = 5 and = 0: solid lines correspond to (4.67), dashed lines, to approximate
formula (4.68), and dash-dotted lines, to approximate formula (4.69).
This equation yields an even dependence for 00 (x) and an odd dependence for
0 (x) (see Fig. 4.2). It is clear from Fig. 4.2 that even for a resonance with
a low Q-factor, this formula provides a good approximation for 00 and somewhat
worse one for 0 .
Note that far from the resonance, |0 | decays much slower than |00 |;
therefore, in the transparency windows, where 00 1, the refractive index n
can still noticeably differ from unity. At a sufficient distance from the resonance,
absorption can be neglected, and (4.67) takes another asymptotic form (Fig. 4.2),
f 2p /4 0
= , (4.69)
20 2 1 2 /20
where 0 (0).
In the description of optical experiments, instead of or = 1 + 4, one uses
parameters that are more close to experiment, the refractive index and the index
of absorption,

n kc/ = Re , c/2 = Im . (4.70)
The value 1 has the meaning of the length of the wave penetration into the
matter, in o/2 c/2 units. Figure 4.3 shows the dispersion dependence of these
parameters, according to (4.68), (4.69) and (4.42)(4.45), in the vicinity of an
isolated resonance.
Fig. 4.3 Dispersion dependence of the refractive index n and the absorption index in the vicinity of
an isolated resonance, for the resonance amplitude being 1 (a) and 5 (b): solid lines correspond
to (4.42), (4.43), and (4.68); dashed lines, to (4.44), (4.45) (i.e., weak absorption approximation),
and

dash-dotted lines, to (4.69) (i.e., approximation of 00 = 0, when n = at > 0 and = at
< 0); shading denotes the energy gap, where 0 < 0.
In laser media, as a rule, || 1 cm1 , so that | 00 / 0 | ||/k < 104 , and

approximations (4.44), (4.45) are certainly valid. Substituting into them (4.68),
we find
!
x
n n 1 , (4.71)
2 1 + x2
/cn
, (4.72)
1 + x2
1/2
where n = (1 + 4 )1/2 is the refractive index at x 1 and
f 2p 4Nd2
4 = = (4.73)
20 3~
is the resonance amplitude for . Note that (4.72) coincides with the result of
a probabilistic calculation, (2.65), and that population inversion leads to a sign
change in 0 , 00 , and . In this case, the dispersion dependence of the refractive

index looks opposite to the usual one: n decreases with the growth of the fre-
quency outside of the resonance and increases in the absorption range. This effect
is called negative dispersion.
In condensed matter, narrow resonances often have large amplitudes, > 1.
This is especially typical for the dispersion in the infrared range, near lattice
eigenvibrations of ion crystals. The corresponding elementary excitations (quasi-
particles) are called optical phonons. Let, for instance, d = 1D, N = 1020 cm3 ,
and 2 = 1 cm1 ; then, according to (4.73), = 4.d If, in addition, f = 1
and 0 = 1 cm1 , then the plasma frequency is much larger than but still much
smaller than 0 : p = (0 )1/2 = 200 cm1 .
From homogeneous Maxwells equations, it follows that longitudinal oscilla-
tions are possible, their dispersion dependence being (, k) = 0. In the neglection
of dissipation and spatial dispersion, such an oscillation has a fixed frequency l
and an arbitrary wave vector, i.e., zero group velocity u = d/dk. According to
(4.69), at f = = 1,
q
l = 20 + 2p 0 + 2p /20 = 0 + . (4.74)
Hence, at 1, the splitting between the longitudinal and transverse frequen-

cies, l and 0 , is much larger than the damping constant . The same condition
defines whether the eigenfrequency shifts of molecules due to their Coulomb in-
teraction, (4.50), are high, so 0 in (4.74) should be understood as 0 /3.
In the interval 0 l , according to (4.69), < 0, and the wave number

k = /c is purely imaginary, so that the field is not a wave any more. Thus,
this interval is a forbidden zone where the module of Fresnels reflectivity R =

( 1)/( + 1) becomes a unity, and the dielectric behaves as a metal. Note
that a metal, in its turn, is at p similar to a dielectric.

4.2.7 Polaritons
In the case of low absorption, macroscopic field in the matter can be treated quan-
tum mechanically (Sec. 7.4). In this case, the notion of a photon in matter, or
a polariton, emerges. (It should not be confused with a polaron, an electron in a
dielectric considered together with the polarization it induces.) A polariton (some-
times also called a light exciton) is an elementary excitation of a macroscopic field
and the molecules interacting with it, having an energy ~ and propagating with a
velocity u = d/dk. As approaches 0 , more and more of the polariton energy
is contained in the internal energy of the molecules.
d Here, the scaling factor 2c between and frequency in cm1 has been omitted.
When a photon enters a medium from a vacuum, it becomes a polariton, with

a probability of 1 |R|2 ; after passing an average distance of 1 , the polariton
gets absorbed. The momentum ~k of a photon in matter differs n times from the
momentum ~/c of a vacuum photon with the same energy.
Polaritons can be also excited through thermal energy, in which case their
mean number per mode is given by the Planck function N(). According to
FDT (Sec. 7.7), the k spectrum of equilibrium field fluctuations in matter scales
as N()G00 (, k), where G is the Greens function for macroscopic Maxwells
equations (Sec. 4.1).
Above, we discussed the dispersion law, i.e., the relation between the fre-
quency and the wavelength, for free waves created by a distant source. One can
suggest other definitions for the functions n() or (k), for instance, given by
the maximum of the Greens function imaginary part G00 (, k). The correspond-
ing dispersion dependence manifests itself in experiments on light scattering by
polaritons (Sec. 6.5).
Let us substitute () in the single-pole approximation (4.68) into Eqs. (4.23),
(4.24). Then, for > 0,
4
Gx = , (4.75)
y + /(x + i)
4
Gz = , (4.76)
+ /(x + i)
where
x ( 0 )/, y (ck/)2 .
Hence, the spectra of transverse and longitudinal field fluctuations are described
by the functions (Fig. 4.4)
4
G00x = , (4.77)
( + xy)2 + y2
4
G00z = 2
. (4.78)
( x )2 +
If the y dependence on is neglected, then the spectra of fluctuations at fixed k
have Lorentzian shapes with the central frequencies given by the equations
() = (ck/)2 , (l ) = 0, (4.79)
where () coincides with (4.68) under the condition = 0,
/x + f 2p /(20 2 ). (4.80)
Fig. 4.4 Spectral density of equilibrium field, hE2 ik , normalized to ~N/23 , as a function of the
frequency and wave vector in the vicinity of a dielectric function resonance for 0 / = 20 and = 5.
Solid lines refer to transverse (with respect to k) oscillations , dashed lines, to longitudinal ones. It
is clear from the figure that the frequency spectrum of thermal field fluctuations is described by the
dispersion dependence (dash-dot) without the anomalous part.
Thus, the dispersion law for field equilibrium fluctuations (4.79) differs from

the dispersion law for free waves (4.42) ck/ = Re () by the absence of the
damping parameter .
The dispersion law (4.79) corresponds to the condition G00 /x = 0. At the
same time, the condition for the G00x maximum at a fixed frequency, G00x /y = 0,
leads, according to (4.77), to the known dispersion law (4.44), Re() = (ck/)2 .
Hence, it follows that the dispersion law observed near a resonance depends on
the experimental conditions.
Dispersion properties of a medium can be qualitatively represented as a graph
showing the relation between and k, instead of the () or n() dependencies.
Figure 4.5 shows such a relation in different approximations and provides the
commonly used names of the corresponding quasi-particles.
If the coupling between the transverse field and the oscillations of charges is
negligible, which is possible at k 0 /c, the elementary excitation, i.e., the en-
ergy quantum of the matter, in the case of polar oscillations of ions in a crystal
lattice, is called an exciton, or an optical phonon. Excitons, similarly to photons,
depending on the wave packet describing them, can be either localized within a
certain area in the crystal, or spread over the whole space. In the = 0 ap-
proximation, dispersion dependencies of photons and excitons overlap without
interaction, which means that the incident field does not excite the excitons.
(a) (b) (c)
(d) (e) (f)

Fig. 4.5 Dispersion in various approximations. The numbers correspond to the following quasipar-
ticles: 1, photon; 2, mechanical exciton, or optical phonon; 3, Coulomb exciton (longitudinal and
transverse); 4, polariton (light exciton); 5, new waves. The following cases are shown: (a) oscillator
strength or the particle density is small, and dispersion dependencies anti-intersect with a small gap;
(b) Coulomb interaction between the molecules lifts the degeneracy between the frequencies of lon-
gitudinal and transverse excitons; (c) interaction between the molecules and the transverse field leads
to polariton effects, namely, to the energy gap (shaded area), to the dispersion of the polariton phase
velocity /k outside the gap and to its group velocity u turning into zero at both boundaries of the gap;
(d) dissipation leads to the smearing of the dispersion dependence, to finite life time and finite free
path length 1/ = u for polaritons; in this case, waves excited externally have anomalous dispersion
(dashed line); (e,f) anisotropy of the matter leads to the dependence of exciton and polariton frequen-
cies on the wavevector direction k/k. The figure shows frequency and angular dispersion, k(, ), for
the extraordinary wave in a uniaxial crystal in the cases of weak (e) and strong (f) anisotropy; is the
angle between the wave vector and the z axis of the crystal, z0 and zl are frequencies corresponding
to zz being and 0, respectively; 0x and lx are the same frequencies with respect to xx = yy .
At small and negligible dissipation, there is anti-crossing, or repulsion,

of the dispersion dependencies, which look, in the interaction area, like two hy-
perbolas separated by a small gap (Fig. 4.5(a)).
At 1, oscillations of the charges and the field strongly influence each
other, and the dispersion dependence changes considerably. There appears the
longitudinal branch and the energy gap, near which u 0 (Figs. 4.4, 4.5(c)).
The electromagnetic wave is followed by an in-phased ( < 0 ) or anti-phased

( > 0 ) polarization wave, whose contribution into the total energy density is
considerable. Note once again that longitudinal oscillations, in the u = 0 ap-
proximation, do not propagate, i.e., these are waves with a fixed frequency l and
arbitrary wavelengths.
Certainly, the simplest models considered here provide only a qualitative de-
scription of the field dispersion in realistic media. For the description of spatial
dispersion effects, one has to take into account the dependence of ad 0 on
k [Vinogradova (1979); Landau (1982)]. Doppler broadening can be included into
the model by integrating (4.67), (4.68), (4.69) w.r.t. 0 over the Maxwell dis-
tribution in the case of gases and over the Fermi-Dirac distribution in the case
of inter-band transitions in condensed matter. Transitions between narrow exci-
ton bands in semiconductors and molecular crystals are described by dispersion
dependencies (4.67), (4.68), (4.69) with , 0 , and k depending on , k. Calcu-
lation of these parameters is an interesting problem in solid state theory.
Note, in conclusion, that sometimes, instead of calculating , it is simpler to
calculate directly the Green function G or the spectral density of equilibrium fluc-
tuations [Zubarev (1971)], which is related to it through the Green-Kubo formula
(Sec. 7.7).
4.3 Two-level model and saturation
Susceptibility calculated above determines the response of the matter to an alter-

nating field only within the first order of the perturbation theory. This calculation
does not take into account saturation of populations and other effects nonlinear
in the field. Nonlinear effects are most pronounced under resonance conditions,
when the field frequency is close to the eigenfrequencies of the matter.
4.3.1 Applicability of the model

This section considers two-level model, which is widely used in quantum elec-
tronics and spectroscopy. The model is based on the assumption that the field is
quasi-monochromatic and a resonance 21 0 takes place only for a single
pair of non-degenerate levels of a molecule. Such a situation is typical for mag-
netic resonance effects, nuclear (NMR) or electronic (EPR) ones. If a molecule
has a single uncoupled electron, or the nucleus has a spin I = 1/2, then an exter-
nal magnetic field H0 splits each level in two Zeemans sublevels, with the tran-
sition frequency 0 = H0 , being the gyromagnetic ratio. Usually, 0 is in the
microwave range and differs much from all other frequencies of the molecule.
However, in the absence of a constant magnetic field the Hamiltonian, as a

rule, is invariant to certain symmetry operations, such as rotations etc., and there-
fore all energy levels are degenerate. Nevertheless, even in this case the two-level
model provides a qualitatively correct description.
Note that in some cases, the two-level approximation is not applicable at all.
For instance, it is not valid for NMR at I > 1/2 or for transitions between vi-
brational levels of molecules with weak anharmonicity. Such a molecule behaves
as an oscillator with nearly equidistant levels, so that a resonance takes place for
many pairs of levels simultaneously.
4.3.2 Kinetic equations

A two-level system is described by density-matrix equations (3.44) with m, n =
1, 2 and phenomenological relaxation times wmn and 21 1/T 2 (see (3.48),
(3.49)). Let Vnn = 0, then
21 = (i0 + 1/T 2 )21 i(11 22 )V21 /~, (4.81)

11 = w12 22 w21 11 + i(V21 12 21 V12 )/~. (4.82)
In the case of two levels, 11 + 22 = 1, hence 11 = 22 . Denote
11 22 , w12 + w21 1/T 1 . (4.83)
With the perturbation switched off, V = 0, populations should take their equilib-
rium values (0)
nn ; therefore, the relaxation rates w12 and w21 are related by (3.58),
which leads to
(w12 w21 )/(w12 + w21 ) = (0) . (4.84)
With an account for (4.83) and (4.84), Eq. (4.82) takes the form
= ((0) )/T 1 + 4Im(21 V12 )/~. (4.85)
Let the field be quasi-monochromatic and have the mean frequency > 0
close to 0 ,
E(t) = (1/2)E0 (t) exp(it) + c.c., (4.86)
where E0 (t) is the slowly varying amplitude of the field. Then, in the dipole and
resonance approximations, one can assume in (4.81) that
V21 (1/2)d21 E0 (t) exp(it) (1/2)~ exp(it + i). (4.87)

Here, we have omitted the non-resonant negative-frequency terme proportional to

eit and introduced the Rabi frequency |d21 E0 |/~ and the interaction phase
(t). For a linearly polarized field and real wave functions, is simply the phase
of the wave. Let us also define the slowly varying envelope of the non-diagonal
density matrix element,
21 (t) = 0 (t) exp(it). (4.88)
As a result, kinetic equations for a two-level system take the form
0 = [(i( 0 ) 1/T 2 ]0 + iei /2, (4.89)

= ((0) )/T 1 2Im(0 ei ). (4.90)
This system of three equations for three real functions 0 (t), 00 (t), and (t) deter-
mines the evolution of a two-level system due to the thermostat and the external
field. According to (4.89), the effect of the field on 0 scales as the population
difference , which, in its turn, is related to the field through 0 , according to
(4.90). It is this relation that leads to the nonlinearity of the two-level system
response. Under stationary conditions, this nonlinearity manifests itself in the sat-
uration effect, i.e., tending to zero provided that 2 1/T 1 T 2 . Non-stationary
effects caused by the anharmonicity of a two-level system will be considered in
Chapter 5.
4.3.3 Saturation
Consider now a stationary response of a two-level system to a monochromatic
field, with , 0 , , and being constant. Then it follows from (4.89) (compare
with (4.56)) that
/2
0 = ei , (4.91)
0 i/T 2
Substituting this expression in (4.90) allows one to find the stationary population
difference,
= (0) /(1 + 2T 1 W), (4.92)
where
2 T 2 /2
W . (4.93)
1 + (0 )2 T 22
e We also neglect higher-order harmonics of the density matrix, which oscillate with frequencies n
and are on the order of magnitude of (/)n .

These equations describe saturation, the decrease of population difference due

to a strong resonance field. Note that (4.93) coincides with the transition rate
calculated in Sec. 2.2 for the case of a Lorentzian shape with the unsaturated
width 0 = 2/T 2 . The second term in the denominator of (4.92) is called the
saturation factor,
s 2WT 1 = 2W/(w12 + w21 ). (4.94)
It is clear from (4.94) that saturation results from the competition between the
transition rates due to the noise field of the thermostat, (w12 + w21 )/2, and the
external monochromatic field, W.
According to (4.93), saturation is most pronounced at exact resonance, where
the saturation factor takes its maximal value,
s0 = 2 T 1 T 2 = 20 FT 1 F/F s . (4.95)
Here, we have introduced the transition cross section W/F, photon flux den-
sity F, and the photon flux density corresponding to a two-fold decrease in the
population difference,
F s 1/(20 T 1 ) = ~c/(80|d(e) 2
12 | T 1 T 2 ). (4.96)
Thus, in a resonant field the population difference decreases as

(0)
= . (4.97)
1 + F/F s

4.3.4 Lineshape in the presence of saturation
By substituting (4.92) and (4.93) into (4.91), one finds the value of 0 with an
account for population saturation. The dipole moment of a two-level system (we
consider a non-polar molecule, dnn 0) is
hd(t)i = d12 21 + c.c., (4.98)
so that the susceptibility of a medium consisting of N two-level molecules takes

the form (see (4.57))
() ()
(E0 ) = Nd12 d21 /~(0 i/T 2 )
() () 0 + i/T 2
= ~1 (0) Nd12 d21 . (4.99)
(0 )2 + (1 + s0 )/T 22
Recall that here, frequencies and 0 are positive and the values of at
< 0 are determined by (4.99) with , i replaced by , i. Note that at s0 , 0 it
follows from (4.99) that the susceptibility (, E0 ), considered as a function of the

complex variable = 0 +i00 , has poles both in the upper and lower semi-planes,
= 0 i(1 + s0 )1/2 /T 2 .
As a result, the function (, E0 ) does not satisfy the Kramers-Kronig relations

(4.8) at E0 , 0.
Let us find, using (4.99) and (4.14), the power of absorbed, or, at (0) < 0,
emitted radiation:
P = ~(0) NW0 /[1 + s0 +(0 )2 T 22 ] Pmax s0 /[1 + s0 +(0 )2 T 22 ], (4.100)
where W0 |d21 E0 /~|2 T 2 /2 is the stimulated transition rate at exact resonance.

It follows from this equation that at saturation, the spectral line maintains its

Lorentzian shape of the form 1/(1 + x2 ), but its width increases 1 + s0 times
(Fig. 4.6),
p
= 2 1 + s0 /T 2 . (4.101)
This effect is called radiation-induced broadening or field-induced broadening.

Let s0 1 + (0 )2 T 22 (strong saturation), then it follows from (4.100) that
P = ~(0) N/2T 1 Pmax . (4.102)
Thus, at strong saturation the power absorbed by the matter is no more de-
pendent on the intensity and frequency of the field and is only determined by the
Fig. 4.6 The absorbed power P as a function of the frequency at different saturation factors s0 ,
according to (4.100).
rate of energy transfer from the molecules to the thermostat. In the case of strong
saturation, the susceptibility real part, according to (4.99), scales linearly with the
frequency,
0 (0 )/s0 . (4.103)
Let us make a numerical estimate. For a wave with the intensity I = 1 W/cm2
and wavelength = 1, ~ = 2 1012 erg, F = 5 1018 photon/(s cm2 ), and

E0 = 8I/c = 0.1G= 30 V/cm. If d12 = 1D, then = 108 s1 , so that s0 = 1 at
T 1 = T 2 = 108 s. The transition rate is W = 5 107 s1 , and if (0) N = 1019 cm3 ,
as it is the case for a doped crystal or a gas at atmospheric pressure, then it follows
from (4.100) that P = 50 MW/cm3 . This estimate shows that in the optical range,
saturation is accompanied by strong heating of the matter. On the other hand, for
electronic paramagnetic resonance in the = 1 cm range and at T 1 = 103 s it
follows from (4.102) that P = 0.1 W/cm3 , so that stationary saturation is possible.
This is used in paramagnetic amplifiers.
Saturation is very important in quantum electronics. It is used for creating
population inversion by means of auxiliary radiation (pump) in lasers based on
doped solids and in paramagnetic amplifiers. It is also applied for producing short
strong light pulses via Q-switching and mode locking. Saturation stabilizes the
amplitude of quantum oscillators and limits the dynamical range of quantum am-
plifiers.
In the case of inhomogeneous broadening, for instance, due to the Doppler ef-
fect, saturation affects not all the line but only its part, with the width on the order
of collision or natural bandwidth. This effect, leading to the Bennett hole burn-
ing in the velocity distribution of the molecules (Fig. 4.7), is used for frequency
stabilization of lasers and in saturation spectroscopy (Sec. 6.4).
Fig. 4.7 Bennett hole burning. For a line inhomogeneously broadened due to the Doppler effect,
saturation only affects those molecules whose velocities have given projections vz () = (0 )/k on
the direction k of the wave propagation. N(vz ) is the velocity distribution of active particles.

4.4 Bloch equations
4.4.1 Kinetic equations for the mean values

In the previous section, we first solved kinetic equations for the density matrix
and then, with the help of the obtained solution (t), we found the necessary mean
(observed) value according to the formula h f (t)i = Tr{ f (t)}. It seems natural to
try to exclude and to find kinetic equations for the observables directly. Such
equations can be obtained from the equations for , but here we will find them in
a different way.
In the case of a closed system, equations for the observables can be found
by averaging the Heisenberg equations, which determine the time dependence of
operators in the Heisenberg picture,
i~d f /dt = [ f (t), H(t)]. (4.104)
(We assume that f has no direct time dependence.) The averaging is over the
initial density matrix, which is usually considered as equilibrium,
i~dh f i/dt = Tr{[ f (t), H(t)](t0)}. (4.105)
In the case of multi-particle systems, the derivative dh f i/dt depends, as a rule,
not only on the mean value h f (t)i but also on the second moments or correlation
functions, h f (t)g(t0 )i. One can write the Heisenberg equations for the second mo-
ments, but after averaging, the result will contain the third moments, and so on.
In order to unlink this infinite chain of equations for the moments, one has to
neglect the correlation of some values at a certain point, h f gi h f ihgi.
As a result, after excluding excessive variables, one can obtain relatively
simple kinetic equations for the observables of a single particle. In these equa-
tions, interaction with other particles and with the thermostat is taken into account
with the help of a few phenomenological parameters, like relaxation times T 1 , T 2
for a two-level system or kinetic transfer coefficients. It is noteworthy that, ac-
cording to FDT (Sec. 7.7), kinetic coefficients for the first moments provide infor-
mation about the second moments, i.e., fluctuations.
Approaches based on the density-matrix equations are certainly equivalent to
the ones based on the equations for the mean values, and they should yield simi-
lar results. Note that in classical statistical physics, similarly, there are two basic
methods of describing the kinetics: the first one is based on the distribution func-
tion (Liouvilles equation, Boltzmanns distribution, Fokker-Planck equation), and
the other one, on the moments (diffusion and transfer equations).
Macroscopic Maxwells equations are, in fact, kinetic equations for the first
field moments hEi, hHi, with the phenomenological function (). The relaxation
time of a monochromatic field is apparently equal to the ratio of the energy

density, E = 0 |E0 |2 /8, and the loss power, E = 00 |E0 |2 /8, i.e., = 0 / 00 .
The same result is obtained if we set equality between and the absorption
length 1/ divided by the velocity of the wave, c/n.
Below, we will consider equations for the observables of a two-level system
in two typical cases, namely, for electric dipole and magnetic dipole interactions,
the energy of interaction with the field being d E and H, respectively.
In the latter case, the observable is the magnetic moment hi of the particle or
magnetization M = Nhi; kinetic equations for these observables are called the
Bloch equations. In the case of electric dipole interaction, kinetic equations for
hdi and the population difference , i.e., energy in ~0 units, are similar to the
Bloch equations and are called the optical Bloch equations.
4.4.2 Pauli matrices and expansion of operators

Description of two-level systems is most convenient in terms of two-dimensional
Pauli matrices, which are defined as
! ! !
0 1 0 i 1 0
x , y , z . (4.106)
1 0 i 0 0 1
These matrices mn hm| |ni represent certain operators , whose eigenval-
ues are = 1. (Recall that the eigenvalues of a matrix fmn are defined as the
roots of the characteristic equation, det{ fmn mn } = 0.) The matrix represen-
tation (4.106) is the eigenrepresentation for the z operator. From (4.106) and
the rules of matrix multiplication, we find the multiplication table for the Pauli
operators,
x y = y x = iz ,
2 = I, y z = z y = i x , (4.107)
z x = x z = iy .
Thus, the Pauli matrices anti-commute with each other ( + = 2 ),
and their commutation relations coincide with the ones for the Cartesian compo-
nents of the angular momentum s.
It is also convenient to introduce elementary matrices called diadic tensors, or
outer products of vectors,
! !
(1) 1 0 (+) 0 1
|1ih1| = , |1ih2| =
0 0 0 0
! !
0 0 0 0
() |2ih1| = , (2) |2ih2| = .
1 0 0 1
In the general case, two arbitrary vectors |ai and |bi can always compose a di-
adic tensor, with the matrix elements given by the products of the corresponding
components of the vectors,
(ab) |aihb|, (ab)
mn = hm|aihb|ni.
A symmetric diadic operator (n) |nihn| is called a projector, since its action on
a vector consists of projecting it onto the |ni direction,
(n) |ai = |nihn|ai = const |ni.
Usually, |ni is a unit vector, hn|ni = 1. Note that the mean value of a diadic
operator (mn) coincides with the corresponding element of the transposed density
matrix,
X
h(mn) i = Tr{|mihn|} = kl hl|mihn|ki = nm .
kl
The Pauli operators are related to the diadic operators as

2(1) = I + z , 2(2) = I z , 2() = x iy ,
x = (+) + () , y = i(() (+) ), z = (1) (2) ,
I = (1) + (2) .
One can easily find the multiplication table and the commutation relations for the
diadic operators,
2 2
(1) = (1) , () = 0, (+) () = (1) ,
() (+) = (2) , [(+) , () ] = z , [() , z ] = 2() .
Note that () are non-Hermitian operators: (() )+ = () . They can be called
creation and annihilation operators for an energy quantum. Indeed, let |2i be
the ground-state wave function of the system. The (+) operator turns it into the
excited-state wave function, (+) |2i = |1i. Similarly, () |1i = |2i.
One can easily verify that any Hermitian operator acting in the Hilbert space
of a two-level system can be represented as a sum (sometimes we will keep the
hats over the operators),
f = aI + bx + cy + dz , (4.108)
where a, b, c, d are real numbers. Indeed, combining (4.106) and (4.108), we find
the relations that define the coefficients in the expansion (4.108) in terms of the
matrix elements fmn ,
!
a + d b ic
f = . (4.109)
b + ic a d
Recall that quantum mechanics, similarly to vector calculus, operates with

three types of variables: usual complex numbers (c-numbers, scalars), complex
vectors (wave functions of discrete or continuous variables), given by n numbers,
and operators (matrices, tensors), given by n2 numbers. (Here, n is the dimension-
ality of the vector space, equal to the number of states of the system.) For a given
operator, one can find, according to certain known rules, the corresponding scalars
(eigenvalues and trace), which are invariant with respect to a change of represen-
tation, i.e., to a rotation of the basic vectors. The Pauli vector, { x , y , z },
and the orbital momentum s, which is proportional to it, are vectors in a real three-
dimensional space and, at the same time, operators in the abstract space of states,
having two complex basic vectors |1i, |2i.
Note that the coefficients in the Pauli-matrix expansion (4.108) of an arbitrary
operator f have a clear physical meaning. They determine two values that the
observable f can take at single measurements. By writing the equation for the
eigenvalues of matrix (4.109), one can see that the spectrum fnn fn consists of
two numbers,
f1,2 = a (b2 + c2 + d2 )1/2 . (4.110)
Let the basic vectors for the representation (4.106) be the energy states of the
system. Then the operator H0 is diagonal and, for f H0 , it follows from (4.109)
that a = b = c = 0, d = ~0 /2. Therefore, the Hamiltonian of the system scales
as the z operator,
H0 = ~0 z /2, (4.111)
where ~0 H022 H011 . The relative population difference is in this case equal
to the mean value of z ,
hz i = 11 22 . (4.112)
Let the eigenfunctions |mi of the H0 operator and, hence, the matrix elements
of the electric dipole moment dmn be real, d12 = d21 = d0 . Let us also assume
that the diagonal elements are absent (the molecule is non-polar), dnn = 0, then
the operator d = er scales as x ,
d = d0 x , (4.113)
and the perturbation operator takes the form
V = (d0 E) x . (4.114)
4.4.3 The Bloch vector and the Bloch sphere

Let us define the Bloch vector R (also called pseudo-spin) as the mean value of
the Pauli operator . Using (4.106), R can be also written in terms of the density
matrix,
R hi = {2021 , 20021 , }. (4.115)
Thus, the R vector, similarly to the density matrix mn , fully determines the
state of the system. In other words, an arbitrary state of a two-level system is
given by three real numbers, which can be explicitly shown by a point or a radius
vector in some three-dimensional space. In the case of a spin-1/2 particle, this
vector is parallel to the mean orbital momentum. In the case of an electric dipole
two-level system, R does not correspond to any observable vector, but its x and z
components, according to (4.111), (4.113), have a clear physical meaning.
Let us find the length of the R vector. According to (4.115) and (3.18),
R2 = (11 22 )2 + 4|21 |2 1. (4.116)
In the case of a pure state, by definition (3.4), |21 |2 = 11 22 and R is a unit vector.
Thus, an arbitrary pure state of a system can be shown by a point on a sphere
called the Bloch sphere. If the state is mixed, then (see (3.18)), |21 |2 < 11 22 ,
and R < 1.
During time evolution, the depicting point (the R(t) vector) moves along some
trajectory on the unit sphere. This trajectory, for an arbitrary perturbation V(t)
(4.114), can be found using the Heisenberg equation (4.104) for and the com-
mutation rules (4.107). At V = 0, (3.46) immediately yields
R x = R x0 cos 0 t + Ry0 sin 0 t,
Ry = R x0 sin 0 t + Ry0 cos 0 t, (4.117)
Rz = Rz0 .
Thus, in the case of a closed system, the end of the R vector circles around
the z axis, similarly to the precession of a spinning top around the gravity force
direction (Fig. 4.8). According to (4.111) and (4.113), the energy is constant in
this case, while the dipole moment oscillates with the transition frequency. In
particular, for a pure coherent state, with cn = exp(in )/ 2, it follows from (3.4)
that
R x = cos(0 t + 1 2 ),
Ry = sin(0 t + 1 2 ), (4.118)
Rz = 0,
Fig. 4.8 The state of a two-level system represented geometrically using the Bloch vector R whose
components determine the dipole moment hdi = d0 R x and the population difference = Rz . The
z axis is directed downwards, so that points denoting the excited states of the system are above the
point denoting the ground state. Under free evolution of the system, R undergoes precession with the
frequency 0 , the angle of precession being determined by the initial conditions.
i.e., the depicting point moves along the equator of a unity sphere, and the pre-
cession angle is arctan(R /Rz ) = /2. For an energy state, the point rests at
one of the poles (cn = 0 or 1, = 0 or ). A weak resonant perturbation causes
a slow variation of the precession angle with the Rabi frequency, the so-called
nutation (Sec. 5.1).
4.4.4 Higher moments and distributions

Recall that the density matrix (or, according to (4.115)), the vector R = hi)
provides full statistical information about the system, i.e., allows one to find the
higher moments h f k i and the probability distributions P( f ) of an arbitrary observ-
able f . Moments can be easily expressed in terms of or h i using (4.108) and
the multiplication formulas (4.107). In particular, it follows from (4.107) that
2k+1
= , 2k
= I.
Hence, the main measure of fluctuations, the variance, is
2 h2 i h i2 = 1 R2 . (4.119)
For an energy state, where Rz = 1 and R x,y = 0, the energy variance is equal
to zero, while the variance of the dipole moment is a unity (in d0 units). In the
case of a coherent state, according to (4.118), the energy fluctuates with the unity
variance (in ~0 /2 units), while the variance of the transverse components, x,y ,
depends on the time instance of the measurement, as it oscillates with 20 fre-

quency between 0 and 1.
Consider the uncertainty relations containing the variances of the Pauli vector
components and limiting the accuracy of their simultaneous measurement. Ac-
cording to (4.119),
2 2 = 1 + R2 R2 R2 R2 ,
2 + 2 = 2 R2 R2 .
For an arbitrary state, the length of the R vector does not exceed unity (see
(4.116)); therefore, the following inequalities hold true:
2 2 R2 + R2 R2 , (4.120)
2 + 2 1 + R2 , (4.121)
where , , . In the case R = 0, inequality (4.120) takes the form of a standard
uncertainty relation,
f g |h[ f, g]i|/2. (4.122)
Let us now find the probability distributions. Let P (1) be the probability of
taking values 1, then
h i = P (1) P (1) = 2P (1) 1.
Hence,
P (1) = (1 R )/2.
For instance, for a coherent state (4.118),
P x (1) = cos2 [(0 t + 2 1 )/2].
Similarly, for an arbitrary observable of a two-level system,
P( f1 ) = 1 P( f2 ) = (h f i f2 )/( f1 f2 ). (4.123)
4.4.5 Bloch equations

Let us find the equations of motion for the Pauli vector. From (4.104), (4.107),
(4.111), and (4.114), it follows that
x = 0 y ,
y = 0 x + (t)z , (4.124)
z = (t)y ,
where (t) 2d0 E(t)/~ is the instantaneous Rabi frequency. Let us introduce
the vector A(t) {(t), 0, 0 }, then (4.124) can be represented as a vector product,
= A. (4.125)
From (4.125), we immediately find a similar equation for the Bloch vector,
R = R A. (4.126)
According to (4.126), the R vector, which represents all properties of a two-

level system, is precessing around the instantaneous direction of the effective field
vector A(t).
Further, let us take into account, in the simplest approximation, the interac-
tion between the particle and its environment. Assume that there is an exponential
relaxation with two positive parameters T 1 , T 2 , which characterize the rate of ap-
proaching thermodynamic equilibrium after the perturbation is off. As a result,
the Heisenberg equations (4.125) turn into the so-called optical Bloch equations,
R = R A R /T 2 z(Rz (0) )/T 1 , (4.127)
where z is a unit vector along the z axis.

One can easily verify that these equations, with an account for (4.115), co-
incide with the density-matrix equations (4.81), (4.85), and hence all results of
Sec. 4.3 are still valid. However, now the system behavior has an obvious geo-
metric interpretation.
It should be stressed that the Bloch equations (4.127) are kinetic equations,
describing only the first moments of the observables R = hi; they provide no
information about fluctuations and higher moments. The latter can be only found
by choosing some particular stochastic model of relaxation.
In the case of a monochromatic field, a settled stimulated motion of R is pre-
cession with the field frequency around the z axis (Fig. 4.8). The precession
angle , according to (4.91), is given by ( |d0 E0 |/~)
R 2|21 | sign()T 2
tan = = , (4.128)
Rz [1 + (0 )2 T 22 ]1/2
and the length of the R vector, according to (4.92), is
R = / cos = (0) /[cos (1 + tan2 )], (4.129)
where T 1 /T 2 . Let T 1 = T 2 and = 0 , then the R vector, due to

the resonance field, shrinks (1 + 2 2 )1/2 times and precesses at an angle =
arctan(sign()).
Fig. 4.9 Relaxation of the Bloch vector R. After the perturbation is switched off, the transverse
component goes down to zero at a rate of T 21 , while the longitudinal component takes its equilibrium
value (0) at a rate of T 11 .
If the perturbation is suddenly switched off, the R vector, according to (4.127),

will simultaneously precess with the Bohr frequency 0 and undergo relaxation to
the equilibrium value {0, 0, (0) }, i.e., move along a spiral (Fig. 4.9). If T 1 T 2 ,
then the first to disappear is the transverse component, R , i.e., the non-diagonal
element of the density matrix, 21 , and after that Rz takes its equilibrium value.
4.4.6 Equation for polarization

In the framework of the two-level approximation, the obtained equations (4.127)
fully determine the optical properties of the medium, both stationary (Sec. 4.3)
and non-stationary (Chapter 5) ones. To make it more obvious, let us pass from
the variables R x , Ry to polarization P = N d0 R x [Pantell (1969)], which enters
Maxwells equations. From (4.127), it follows that
P + P/T 2 = 0 Nd0 Ry . (4.130)
Taking the second derivative, we get
P + P/T 2 = 0 Nd0 (Ry /T 2 0 R x + (t)Rz ). (4.131)

In practice, always 0 T 2 1, so that at 0 , according to (4.130), one

can assume in the right-hand side of (4.131) that
0 Nd0 Ry P. (4.132)
As a result, we find that polarization satisfies the second-order linear differential
equation (we assume that d0 kE)
2 20 d02
P +P + 20 P = EN, (4.133)
T2 ~
where N NRz is the population difference per unit volume. Substituting
(4.132) into the equation for Rz , we find the equation for N,
1 2
N + (N (0) N) = E P. (4.134)
T1 ~0
This equation has a simple meaning: according to (4.13), E P is the power ab-
sorbed in the matter.
Thus, a two-level system behaves as a harmonic oscillator with the damping
1/T 2 whose coupling with the external force E depends on the force itself, with the
inertia time T 1 . In the case of a weak field, with the saturation factor 2 T 1 T 2 1,
the system is equivalent to a linear oscillator.
4.4.7 Magnetic resonance

As it was already mentioned, two-level approximation is most applicable to a
spin-1/2 particle. In a magnetic field, an electron acquires additional energy H =
H, with being the magnetic moment of the electron, anti-parallel to its
mechanical moment (spin) s, = (g0 /~)s. Here, g = 2.002 is the so-called g
factor of a free electron, 0 = e~/2mc = 0.927 1020 erg/G is the atomic unit of
magnetic dipole moment (Bohrs magneton). The spin is usually given in ~ units:
s0 s/~. The operators of spin projections, s , scale as the Pauli matrices, i.e.,
s0 = /2. Thus, if we ignore the difference between g and 2, then = 0 , and
H = 0 H . (4.135)
Now, using commutation relations for (4.107), we can easily find equations
of motion for any operator. For instance,
i~ x = 0 ([ x , y ]Hy + [ x , z ]Hz ) = 2i0 (Hy z Hz y ). (4.136)
Derivatives for other components can be written in a similar way. As a re-
sult, equation of motion for the Pauli vector takes a simple form (compare with
(4.125)),
= H, (4.137)
where 20 /~ = 2 2.8 MHz/G is the gyromagnetic ratio. Since the

vector scales as the magnetic and mechanical moment of an electron, equations
for and s have a similar form, for instance, s = s H.
The last equation has the same form as the classical equation for rotational
motion, according to which the rate of angular momentum variation is equal to the
torque of the forces acting on a dipole in a magnetic field, H = 20 s H/~.
Thus, an electron in a magnetic field behaves similarly to a spin top under the
action of two forces. In the case of a constant magnetic field, (4.137) describes
precession, the motion of the momentum vector along a cone around H0 (Fig. 4.8).
The precession frequency, 0 = ||H0 , coincides with Bohrs transition frequency,
(H22 H11 )/~.
However, in contrast to a classical spin top, the observable angular momen-
can have only a single absolute value, s ( s2 )1/2 , equal to
P
tum
of an electron
~ 3/2, since 2 = 1. Also, a single measurement of its projection onto an axis
can only yield one of the two values, ~/2. Note that s2 , hsi2 = hs i2 ; for in-
P
stance, in a mixed state with equal populations ( = 21 = 0), all three projections
are equal to zero, so that hsi2 = 0.
Let the magnetic field have, in addition to the constant component H0 , a time-
varying component H (t), orthogonal to H0 . Let the x axis be parallel to H and
the z axis be anti-parallel to H0 . Then the electron energy (4.135) can be written
in terms of the Pauli operators,
H = H0 + V(t) = ~(0 z + (t) x )/2, (4.138)
where, this time, (t) = H . We have chosen H0z to be negative for the subscript
1 to denote the lower level. Once again, the Heisenberg equations take the form
(4.124).
Let us average these equations over the initial density matrix, pass to magne-
tization M = 0 Nhi, and add relaxation. We get
M x = M x /T 2 + 0 My ,
My = My /T 2 0 M x + Mz H , (4.139)
Mz = (Mz(0) Mz )/T 1 My H ,
where Mz(0) 0 N(0) is the static equilibrium magnetization.

This system of equations, determining the magnetization kinetics of a param-
agnetic material (electronic or nuclear) in a constant or variable field, is called the
Bloch equations. The Bloch equations describe magnetic resonance, i.e., resonant
absorption of radio waves. This effect forms the basis of the most important di-
rections in radio spectroscopy, electronic paramagnetic resonance (EPR), nuclear
magnetic resonance (NMR), and ferromagnetic resonance. One can easily verify
that the Bloch equations are equivalent to the density-matrix equations in the case
of a two-level system; therefore, all above-made conclusions are also valid for the
case of magnetic resonance, with d E(t) replaced by 0 H (t).
As we have already mentioned, often T 1 > T 2 , since the relaxation of the
population energy hz i is only caused by non-adiabatic interactions of the particle
with the environment, for instance, non-elastic collisions in a gas or spin-lattice
(spin-phonon) interaction in a crystal. At the same time, variation of the transverse
components, h i (or 21 ), does not require energy transfer, and hence the life
time of h i reduces as a result of both adiabatic (spin-spin) and non-adiabatic
perturbations. One can say that perturbations disturb the precession phase 1 2
in (4.118), which is the argument of 21 , and hence the time and ensemble mean
values tend to zero, h i 0.
Chapter 5
Non-Stationary Optics
In previous chapters, we mostly considered settled, stationary cases of the inter-

action between a field and matter. With a harmonic field on, transient processes
decay due to the relaxation, and the oscillation of charges goes on with a station-
ary amplitude, given by the susceptibility . Susceptibility with an account for
saturation, (E0 ) (see (4.99)), determines the response of the matter, namely, of
the polarization amplitude P0 , to a quasi-monochromatic perturbation, provided
that either the amplitude E0 is constant or the time of its variation is much greater
than the relaxation time, E T 1,2 .
If the field is weak and there is no saturation, the population relaxation time
T 1 does not play any role, and the stationarity condition has the form E T 2 .
It is important that in the absence of saturation, the equations of motion of matter
are linear, and hence the susceptibility () also determines non-stationary, tran-
sient processes. For instance, the response of matter to a short (E T 2 ) weak
pulse scales as the Fourier transform of (), i.e., Greens function (t), and has
the form of a set of oscillations decaying with the times T 2mn (see (4.63)). The
response to a weak pulse of an arbitrary shape, E(t), is given by the convolution
of (t) and E(t).
A question arises: what will be the response of matter to short and strong
pulses? In this chapter, it is shown that under such conditions, several unusual
optical effects can be observed. Among these effects, called coherent a ones,
there are, for instance, self-induced transparency, optical echo, and superradi-
ance. Some of these essentially non-stationary effects have been observed in ra-
dio spectroscopy relatively long ago; however, in the optical range they have only
been observed after the invention of the lasers. Apart of their general theoret-
ical importance, optical non-stationary effects are of practical interest from the
a In this book, instead of the ambiguous term coherent we will use the term non-stationary, which
is more accurate for the description of the effects considered.
107
viewpoint of spectroscopic applications and for the optimization of lasers. They

also form the basis of coherent (non-stationary) spectroscopy [Steinfeld (1978);
Manykin (1984)]. More detailed description of non-stationary effects can be found
in Refs [Allen (1975); Macomber (1976)]. In Sec. 5.1, using the Bloch equations,
we find the variation of the state of a two-level system under a short resonant
light pulse. Further, we consider emission from a single atom (Sec. 5.2) and an
ensemble of N atoms (Sec. 5.3) under different initial conditions.
5.1 Stimulated non-stationary effects
Consider the reaction of a two-level system to a quasi-harmonic perturbation of

finite duration. For simplicity, assume that the pulse envelope is rectangular, with
the duration E T 2 , so that one can neglect relaxation and use the relations for
the density matrix mn or the equivalent Bloch equations for R = {2021 , 20021 , 11
22 } at T 1,2 = . Then, the Bloch equations (4.126) have the same form as the
Heisenberg equations for the operator (4.125). In other words, at time intervals
much shorter than the typical times of the interaction between the system and its
environment, it can be considered as isolated and described, instead of kinetic
equations, in terms of the Schrodinger equation for the wave function, the von
Neumann equation for (t), or the Heisenberg equations for the operators.
We have already solved this problem using the Schrodinger equation in Chap-
ter 2, where we have found the amplitudes of the perturbed states cmn (t) and the
transition probability Pmn = |cmn |2 for a multi-level system in the first order of the
perturbation theory. However, in the case of a two-level system, there is no need
to represent the solution as a perturbative series.
5.1.1 Atom as a gyroscope

Let us turn to the geometric representation of the instant state of the system us-
ing the Bloch vector R. According to (4.126), it behaves similarly to the orbital
momentum of a body with a single fixed point,
R = R A. (5.1)
Here, A {2d0 E(t)/~, 0, 0} is the effective field vector, d0 d21 = d12 is the
dipole moment of the transition, which is assumed to be real and parallel to the
field E(t), and 0 is the transition frequency. Equation (5.1) preserves the length
of the R vector, which is unity in the case of a pure state. The end of the R vector
moves then along the surface of a unit sphere.
Non-Stationary Optics 109
Consider the trajectory of this motion in a quasi-monochromatic field,
E(t) = (1/2)E0(t)eit+i0 + c.c.
Let us pass to a reference frame rotating with the field frequency around the
z axis,
21 = 0 eit ,
R x = R0x cos t + R0y sin t, (5.2)
Ry = R0x sin t + R0y cos t.
Let the conditions | 0 | 0 , d0 E0 ~0 hold, then the rotating-wave

approximation is applicable and the functions 0 (t), R0 (t) are slow compared to
eit . These functions have a simple physical meaning: they determine the ampli-
tude and the phase of the mean dipole moment, hdi = 2d0 021 = d0 R x . The power
absorbed from the external field is proportional to R0y (t), see (5.12).
It is not difficult to verify that in the rotating-wave approximation, equation of
motion (5.1) takes the form
R0 = R0 , (5.3)
where
R0 {200 , 2000 , },
{ cos 0 , sin 0 , 0 }, (5.4)
11 22 , d0 E0 /~,
0 is the initial phase of the field at the center of the atom; in the case of a plane
wave, 0 = kz + 1 . Recall that for simplicity, we assume the matrix element of
the transition to be real. In the general case, 0 = arg(d21 E0 ).
Thus, in a harmonic field the R0 vector, which represents the state of the sys-
tem, rotates with the rate = [2 + (0 )2 ]1/2 around the effective field
direction (Fig. 5.1). This slow rotation is called nutation; it adds to the fast
stimulated precession with the rate equal to the field frequency . In other words,
the angle of precession (t) (see (4.128)) slowly varies within the pulse duration,
so that the depicting point moves along a spiral on a sphere with the fixed radius
R = (2 + 4|21 |2 )1/2 .
In the resonant case, where |0 | , the nutation axis is in the equa-
torial plane. If, in addition, the initial state is an energy one, i.e., R(0) is at one of
the poles and (0) = 0 or , then, due to the effect of the field, R moves along a
meridian with the longitude 0 /2.
(a) (b)
Fig. 5.1 Nutation of the Bloch vector R. Due to a monochromatic field, R, in addition to precession,
rotates around the direction of the effective field with the nutation rate : (a) the frame of reference
rotates with the field frequency ; (b) a fixed frame of reference; solid line corresponds to the exact
resonance, dashed line, to , 0 .
As to the non-resonant case, |0 | , the nutation axis is almost parallel

to the z axis, so that the R vector moves approximately along latitudes, with the
angular rate 0 in the rotating reference frame, i.e., with the rate 0 in a
fixed frame.
In practice, of course, the envelope E0 (t) of the field increases and decreases
gradually, so that the rates and are slow functions of time. At exact reso-
nance, = 0 , variation of the precession angle is determined by the area of the
pulse [Allen (1975)],
Z t Z t
(t) (0) = dt = d0 E0 dt/~. (5.5)
0 0
The phase 0 of the field is assumed to be constant, i.e., only amplitude modula-
tion is considered.
Note that equation (5.3) provides a recipe for preparing any arbitrary pure state
of the system (with given latitude and longitude) with the help of a coherent
field: one should first cool the system, i.e., move the depicting point to the South
pole, and then apply a resonant pulse with a given area and phase.
5.1.2 Analytical solution

In the case E0 = const, it is not difficult to describe nutation algebraically. Let, for
instance, the initial state be stationary, i.e., the depicting point is at t = 0 at one
of the poles, and 0 = 0. Then (5.3) is satisfied by the following functions of the
time [Allen (1975)]:

2(0 ) 2
R0x = (0) sin (t/2),
2

R0y = (0) sin(t), (5.6)

22 2
" #
Rz = (0) 1 sin (t/2) .
2
The last equation can be understood in terms of the transition probability. Let
the system initially be at the ground level, (0) = +1, then the probability P of a
transition up is 22 = (1 Rz )/2, and (5.6) leads to the Rabi formula,
" #2

P= sin(t/2) . (5.7)

Note that at |0 |, (5.7) coincides with the transition probability (2.34),
which was found using the perturbation theory,
"
# 2
0
P= sin t . (5.8)
(0 ) 2
Hence, if one finds the transition rate W dP dt from (5.7) and averages it over
the exponential distribution of interaction times (see the end of Sec. 2.2), the re-
sulting expression will be equivalent to formula (4.100) describing the stationary
absorption with an account for saturation (at T 1 = T 2 ).
Thus, in the presence of a resonant perturbation, a quantum system passes
periodically from the ground level to the excited one and back (Fig. 5.2). The
transition time, according to (5.7), is
t = / = ~/d0 E0 . (5.9)
Fig. 5.2 Transition probability as a function of time, according to the Rabi formula, in the case of
exact resonance (solid line) and with 0 = 3 (dashed line).
This picture should be compared with the idea of quantum jumps, commonly
used at the early stage of quantum mechanics.
Let a transition with d0 = 1D be caused by a strong plane wave with the peak
intensity 1 GW/cm2 ; then, E0 = 1 MW/cm, and (5.9) leads to t = 1ps. This time
is in many cases much smaller than the relaxation time, hence our approximation
T 1,2 is valid.
But for usual optical experiments, t T 1,2 , and the R vector, depicting the
state of a two-level particle on the sphere, does not have enough time to move far
from the South pole before the next collision of the particle with the neighboring
ones, which returns R into the equilibrium position with R = 0. As a result, on
the average, dynamical equilibrium takes place, with a certain stationary angle of
precession (4.128).
Note that the amplitude of the field required for stationary saturation, E0

~/d0 T 1 T 2 (see (4.95)), does not exceed, on the order of magnitude, the pulse
amplitude required for observing coherent effects,
E0 ~/d0 T 2 . (5.10)
In other words, a field that can cause a transition within a time much shorter than
the relaxation time, will cause, on a long-time scale, strong saturation.
Below, we will consider several experimental methods of observing non-
stationary effects.

5.1.3 Nutation
Let an equilibrium two-level system interact with an electromagnetic wave with
the amplitude E0 and the resonant frequency = 0 . At short time intervals,
t T 2 , the response of the system is described by equation (5.3). According to
this equation, the Bloch vector R, which denotes the instant state of the system,
performs, like a spin top, nutational motion, i.e., periodic variation of the preces-
sion angle with the Rabi frequency = d0 E0 /~. At exact resonance, R moves
along a meridian from the South pole (the equilibrium state) through the equa-
tor (a coherent state) to the North pole (population inversion) and back, with the
angular rate .
Evidently, when the R vector moves upwards, towards the North, the
ensemble-averaged energy of the system increases due to the energy of the wave,
and at the time instance when = t = , the wave, on the average, loses exactly
one energy quantum ~. Further, when R goes downwards, towards the South,
the system gives the stored energy back. As a result, the wave becomes amplitude
modulated with the nutation frequency (Fig. 5.3). Provided that there are suf-
Fig. 5.3 Effects of transient optical nutation and free polarization decay. The front slope of a resonant
pulse (dashed line) creates in the matter a transient process, nutation, which leads to the intensity
modulation I(t) of the transmitted light (solid line) with the Rabi frequency. Here, is the absorption
coefficient with an account for saturation, l is the layer thickness, T 2 is the transverse relaxation time.
The back slope of the observed pulse gets a tail caused by the free polarization precession.
ficiently many atoms, the wave transmitted through the matter can have a 100%
modulation. This periodic variation of instantaneous optical density of the matter
is called optical nutation.
Let us find the power absorbed by the matter. According to (4.98), polarization
of a medium with the density N of two-level atoms is
P = 2d0 N021 = d0 N(R0x cos t + R0y sin t) Pc cos t + P s sin t, (5.11)
where the in-phase, Pc , and the quadrature, P s , polarization components have

been introduced.b (We assume that 0 = 0 and E = E0 cos t.) From (4.13), it
follows that the loss power averaged over the period 2/ per unit volume of the
matter is determined by the quadrature component of the polarization,
1
P(t) = E0 P s (t) = d0 E0 NR0y (t)/2. (5.12)
2
Hence, with the help of (5.6) we find
P(t) = ~(0)N sin(t)/2. (5.13)
This result agrees with the above-given qualitative consideration assuming that
the field and the atoms periodically exchange energy quanta. Note that, because at
t T 1,2 relaxation is too slow to manifest itself, real absorption, i.e., irreversible
energy dissipation into the thermostat, is absent here.
Further, at t > T 1,2 , the amplitude of atoms nutation reduces due to relaxation,
and a stationary angle of precession is established (4.128). Absorption of the field
b Editors note: the terms come from radio spectroscopy; in nonlinear and quantum optics they would
be called amplitude and phase quadratures.

occurs then in a usual way and is described, according to Eq. (4.15), by the imag-
inary part of the susceptibility. Modulation of the transmitted wave disappears,
and its amplitude is given by the Bouguer law with an account for saturation (see
Fig. 5.3 and Sec. 6.4).
If then the field is suddenly switched off, the matter will continue to emit light
for some time (Fig. 5.3) due to the free precession of the R vector; the matter
is then in a coherent superradiance state (Sec. 5.3). This effect is called free
polarization decay.
In the presence of inhomogeneous, for instance, Dopplers, broadening, these
formulas refer to only a small group of atoms with a certain projection of the
velocity. In order to find the total polarization, they should be integrated over
Maxwells distribution.
Optical nutation can be used for finding the transition dipole moment by mea-
suring the modulation frequency and for the study of relaxation processes by
observing the modulation decay rate.
5.1.4 Self-induced transparency

Further, let us consider interaction of two-level atoms with short resonant pulses
with duration E T 1,2 . According to (5.13), such pulses increase the energy per
unit volume of the matter by a value
Z E
E= Pdt = ~(0)N sin2 (E /2). (5.14)
0
Let, for instance, E = 2/, so that E0 E = 2~/d0 forms the so-called 2

pulse, then E = 0, and the pulse should pass through the matter without absorp-
tion! This theoretical prediction seems paradoxical: a material that is completely
opaque in the usual sense has full transmission for sufficiently short and strong
pulses.
This effect, named self-induced transparency (SIT), or self-bleaching, is in-
deed observed in experiment. (It should be distinguished from the saturation ef-
fect, which also leads to deviations from the Bouguer law.) The effect of the
matter on the radiation is manifested here only in the reduction of the propagation
velocity and in the pulse distortion. However, it is important that the area of the
pulse, = E (see (5.5)), remains equal to 2.
SIT can be qualitatively explained as follows. The first half of the pulse in-
cident on the matter ( = ) is absorbed by the atoms, which pass into the ex-
cited state with the population inversion. However, the second half of the pulse
eliminates this inversion, so that the absorbed energy is coherently returned to
the field. Note that irradiating the matter with short pulses is an example of a
non-stationary way to create population inversion in two-level atoms.
For the quantitative description of SIT, it is necessary to consider the inhomo-
geneous wave equation in combination with the equation of matter, i.e., Maxwells
and Blochs equations (see, for instance, Ref. [Allen (1975)]). Such a consider-
ation shows that the propagation velocity u of a 2 pulse is determined by its
length, cE l, and the usual absorption coefficient 0 ,
c/u = 1 + 0 l/2. (5.15)
Thus, the longer the pulse, the more delayed it gets. For instance, at E = 1ns
(l = 30 cm) and 0 = 102 cm1 the pulse is slowed down by three orders of
magnitude.
An interesting result of the wave theory in nonlinear materials is the prediction
of solitons, stable pulses whose shape and amplitude do not change in the course
of propagation [Rabinovich (1989)]. In the case of two-level anharmonicity,
solitons have a hyperbolic secant shape,
E0 () = (2~/d0E )sech(/E ), t z/u. (5.16)

R
One can easily verify that (5.16) is a 2 pulse, i.e., dtE0 = 2~/d0.
It follows from (5.14) that 4 and, generally, 2n pulses do not get absorbed
either. According to the wave theory, in the course of propagation such pulses
split in separate stable 2 solitons.
Note that SIT also takes place for inhomogeneously broadened transitions (for
instance, in the presence of the Doppler effect in gases or ingomogeneous static
fields in solids), where 2/ T 2 T 2 . It is important that the condition
E T 2 is not necessary. SIT is also observed under inter-band transitions in
semiconductors.
5.2 Emission of an atom
Consider an equilibrium two-level system after the incidence of a resonant /2

pulse with the duration E = /2. According to (5.3), in the end of the pulse
duration the system is in a coherent state with the coordinates = = /2
on the Bloch sphere (the initial field phase is assumed to be 0 = 0). Recall
that in a coherent state, the energy of a quantum ensemble has no definite value.
When measured, it randomly takes values 0 or ~0 , the mean value being half of
a quantum, E hH0 i = ~0 /2, if the energy is measured from the ground level.
According to (5.6), after the /2 pulse is over, the density matrix of the atom
has the following elements:
21 (t) = i(0)ei0 t /2, (t) 11 22 = 0. (5.17)
Hence, the mean dipole moment of the atom oscillates (see Fig. 2.1) with the
transition frequency 0 = 21 , its amplitude being equal to the transition matrix
element d0 d21 (we assume that d21 = d12 and (0) = 1),
hd(t)i = 2d0 021 (t) = d0 sin 0 t (5.18)
5.2.1 Emission of a dipole

Let us now take into account the spontaneous emission of the atom, which we
so far neglected. As it was already mentioned in Sec. 2.5, spontaneous transfer
of energy from the atom to the vacuum can be considered as relaxation with a
typical time T 1 = T 2 /2 = 1/A, where A is the spontaneous transition rate and
1/A is the radiative life time of the excited state. The vacuum plays then the role
of a thermostat with a zero temperature and infinite heat capacity. Spontaneous
emission is the simplest quantum model of an irreversible process.
Consider the spontaneous emission of an oscillating dipole (5.18) from the
classical viewpoint. It follows from Maxwells equations that the time-averaged
power emitted by moving charges is equal, in the dipole approximation, to [Lan-
dau (1973)]
P = 2dcl2 /3c3 . (5.19)

If we identify dcl with hdi, we will underestimate the power twice comparedto the
prediction of the quantum mechanics (see below). Assume then that dcl = 2hdi,
then
Pcoh = 4hdi2 /3c3 = 240 d02 /3c3 . (5.20)
In addition to estimating the total power of spontaneous emission, the semi-
classical model predicts its polarization and directivity diagram. For instance, in
the case of an electric dipole transition, the field in the far-field zone is
E(r, t) = hd (t0 )i/c2 r, (5.21)
where d is the projection of the dipole moment on a plane orthogonal to r and
t0 t r/c. Hence, the directivity diagram has a usual dipole shape P sin2 ,
where is the angle between r and hdi.
Note that an atom that is initially in a coherent state emits field with a definite
phase. This means that another atom, being in a ground state at a distance r from
the first one, with a certain probability, scaling as r2 , can get into a coherent state,
with the precession phase shifted by 0 r/c.
5.2.2 Probability of a spontaneous transition

Emission (5.20) will lead to a constant decrease in the stored energy and the os-
cillation amplitude of the dipole moment, so that the system will eventually come
into the ground state. The Bloch vector R will then move along a spiral con-
verging to the South pole (in a fixed reference frame), and the amplitude of the
dipole moment hdi oscillations in (5.18) and (5.20) will slowly (as A 0 ) de-
cay, see Fig. 4.9. Similar behavior will take place after exciting the system with
any pulse, except for n or 2n pulses, which yield hdi = 0. Thus, an atom in
a non-stationary state has an oscillating dipole moment and, according to the
classical electrodynamics, it spontaneously emits a quasi-monochromatic wave
with the frequency equal to the transition frequency 0 . The angle of precession
gradually tends to zero in this case, due to the radiative energy loss.
The semi-classical expression (5.4) for the power Pcoh emitted in the case of
a coherent state allows one to estimate the probability of a spontaneous transition
per unit time, A. To do this, let us postulate that the power of the atom emission
decays exponentially,
P(t) = P(0)eAt , (5.22)
then the total energy is
Z
E= P(t)dt = P(0)/A. (5.23)
0
Assuming that it is equal to the initial energy of the atom, ~0 /2, we find, with
the help of (5.20), that
A = 430 d02 /3~c3 , (5.24)
which coincides with the result of a more consistent calculation (Sec. 7.7).
In addition to using an arbitrary numerical coefficient,c this reasoning has
other, more serious, flaws. Namely, it does not result in an exponential decay and
predicts the stability of an excited energy state ( = 1, 12 = 0) with hd(t)i = 0.
One can try to improve the situation by substituting into (5.19) hd2 i instead of
hdi2 . However, the d2 operator is proportional to the unity operator,
d02 0
! ! !
2 0 d0 0 d0
d = = = d02 I, (5.25)
d0 0 d0 0 0 d02
c Recall

that we voluntarily assumed dcl /hdi = 2.
therefore hd2 i = d02 regardless of the state of the atom; this also follows from
(4.107). For instance, in this model the atom should emit even in the ground
state, which contradicts the energy conservation law. Instead of a stable excited
state we have obtained an unstable state. The same problem arises in the well-
spread interpretation of spontaneous transitions as caused by zero-point vacuum
fluctuations.
A consistent theory of the spontaneous emission, as well as statistical optics
in general, should be based on the quantum description of the field (Chapter 7).

5.2.3 Normally ordered emission
However, spontaneous emission can still be described correctly in the frame-
work of semiclassical theory, provided that the squared dipole moment in (5.19)
is replaced by the mean value of the normally ordered squared dipole moment
(Sec. 7.7),
P = (2/3c3)h: (d) :2 i = (4/3c3)hd() d(+) i. (5.26)
The colons denote normal ordering, i.e., placing positive-frequency operators f (+)
on the right of negative-frequency operators g() . By definition, an f (+) operator
in the Heisenberg picture contains only positive-frequency harmonics,
X
f (+) (t) = fn exp(in t), n > 0.
n
Similarly, the Fourier transform of a g() (t) is only nonzero for negative
frequencies.
Any operator can be represented as a sum of positive- and negative-frequency
parts (disregarding the constant component), d = d(+) + d() ; for Hermitian opera-
tors, d(+) = (d() )+ . Hence,
: d2 := d(+)2 + 2d() d(+) + d()2 = 2d() d(+) , (5.27)
while according to (5.25),
d2 d(+)2 + d() d(+) + d(+) d() + d()2 = d02 I (5.28)
(in what follows, we will show that d()2 = 0).

During a spontaneous transition, the d(t) dependence in the Heisenberg repre-
sentation can be approximately considered as unperturbed, since A 0 ,
!
0 exp(i0 t)
d(t) = d0 , (5.29)
exp(i0 t) 0
where d0 d21 = d12 , 0 = 21 > 0. Hence, d = 20 d and, by definition,

!
(+) 0 1
d (t) = d0 exp(i0 t) d0 (+) (t),
0 0
! (5.30)
() 0 0 ()
d (t) = d0 exp(i0 t) d0 (t),
1 0
where () x iy (Sec. 4.4). By multiplying these matrices, we find
!
0 0
: d2 := 2d02 = d02 (I z ); (5.31)
0 1
recall that d = d0 x . To justify this procedure, let us point out that it is namely
the normally ordered square of the field, E () E (+) d() d(+) that determines the
useful energy, which can be absorbed by the other atom (Sec. 7.7). Also, note
that this procedure does not require time averaging over high-frequency oscilla-
tions of the power (see (5.19)).
From (5.26), with an account for (5.31), it follows that P = 240 d02 hI z i/3c3 ,
or
P = ~0 A22 . (5.32)
Let us stress that this result is valid for any state of the atom, including a purely
energy or coherent one. In the latter case, 22 = 1/2, and (5.32) coincides with
(5.20). Thus, power of the spontaneous emission of a two-level atom at a given
time moment scales as the upper level population.d
Due to the energy conservation law, evidently, the following equality should
hold true:
~0 = 2P. (5.33)
The factor 2 takes into account that each transition changes the population differ-
ence by a value of 2. Hence, replacing 22 in (5.32) by (1 )/2, we find the
kinetic equation,
= (1 )T/T 1nat , T 1nat 1/A. (5.34)
Thus, we have confirmed the exponential law (5.22) of the excited state decay,
(t) = 1 + [(0) 1]eAt . (5.35)
A more rigorous description of the interaction between a two-level system and
the vacuum can be obtained using the Heisenberg equations (4.125) for the Pauli
d This conclusion becomes evident in Diracs notation, (+) = |1ih2|, () = |2ih1|, () (+) =
|2ih2|, h() (+) i = 22 .

vector, by substituting, for the external field, the emission field in the zeroth-order
approximation, i.e., the reaction field (for more detail, see Ref. [Allen (1975)]).
Such equations describe the effect of radiation-induced damping, well known
from classical electrodynamics [Landau (1973)]. For z , they lead to an equa-
tion of the form (5.34), while for () , to harmonic-oscillator equations with the
eigenfrequency 0 + iA/2, where is the correction to the transition fre-
quency, called the Lamb shift, A. For instance, for the resonant line L of
a hydrogen atom, /2 109 Hz. The imaginary correction to the transition
frequency has the meaning of the inverse time 1/T 2 of transverse relaxation due
to the radiation reaction. Thus, for an isolated atom, T 2nat = 2T 1nat = 2/A.
5.2.4 Relation between spontaneous and thermal emission

According to the semiclassical theory, spontaneous transitions in atoms lead to the
emission of exponential wave trains. Then, the field depends on the time as
E(t) = (t)E0 eAt/2 cos(0 t + 0 ), (5.36)
where (t) is the Heaviside step function. The Fourier transform of (5.36), giving
the spectrum of the radiation, has a Lorentzian (dispersion) shape,
!
iE0 exp(i0 ) exp(i0 )
E() = + . (5.37)
4 0 + iA/2 + 0 + iA/2
A similar result follows from the quantum theory. The spectral line width nat =
A due to the spontaneous emission is called the natural one.
Emission spectra observed in practice are usually created by a large number
of atoms excited at random time moments into states with random phases. (The
opposite case, leading to superradiance, will be considered later.) For instance,
in a heated gas at low pressure the atoms are excited through collisions. During
a collision, an atom is in the pulsed field created by the neighboring atom, and
this field changes its state according to the general formulas of the perturbation
theory (Sec. 2.1). After the collision, atoms, according to the semiclassical theory,
emit exponential wave trains of the form (5.36) with random initial phases. The
average length of a train is determined by the spontaneous life time 1/A or the
time between collisions.
Usually, radiation cooling of a gas is compensated for by the external heating,
so that the superposition of all trains creates the stationary field of thermal radia-
tion. In the case of a small optical density of the gas, l 1, the intensity will
scale as the upper level population. If l 1, one has to take into account not only
spontaneous transitions but also stimulated ones, leading to the absorption and fur-
ther re-emission of photons (the radiation trapping effect). In the limit l 1,
the gas emits equilibrium Plancks radiation. In Sec. 7.1 we will consider a simple
qualitative model of this process, also covering the case of amplified spontaneous
emission at < 0.
5.2.5 On the emission of fractions of a photon

Let us mention here an interesting paradox, namely, the apparent contradiction
between our theory and the traditional picture of photons. For a coherent state
with the precession angle , according to (5.14), the energy stored by the atom
is E = ~0 sin2 (/2). As a result of spontaneous emission, this energy after a
certain time will be transferred to the field. At the same time, in the photon
language this means that the atom emits a fraction of the photon energy, equal to
E/~0 = sin2 (/2). For instance, after an atom is excited by a /2 pulse into a
coherent state on the equator, it emits an exponential train with the energy equal
to half a photon, in contradiction with the initial postulates by Planck and Bohr.
However, this conclusion by no means contradicts to quantum electrodynam-
ics, which states that a field contains an integer number of photons N only in pure
energy states. This class of states with a definite photon number is a very special
one, and even exotic from the viewpoint of its experimental preparation, with the
exception for the vacuum state with N = 0. As to coherent, as well as mixed states
of the field, hNi = hHi/~0 can be any non-negative number.
The paradox, as usual, originates from a terminology confusion: E = ~0 /2
is the energy at t = 0 averaged over the atomic ensemble, hH(0)i = E, it has
no relation to the result of a single energy measurement, Ei , which yields either
E1 = 0 or E2 = ~0 with equal probabilities. At t 1/A, the energy of the
field is ~0 /2 only on the average; a single measurement of the photon number,
i.e., of the field energy, yields, according to the basic postulate of the quantum
mechanics, only an average number of photons 0, 1, 2, . . . .
Thus, the energy exchange between the atoms and the field is still possible via
fractions of quanta, but only if energy is understood as the ensemble-averaged
one, E = hHi. As to the energy that is transferred in a single interaction event, it
is not definite, as the initial state of the atom is almost always not an energy one.

5.2.6 Quantum beats
In the case of a multi-level system, the natural bandwidth can be explained by the
finite lifetimes of all excited levels,
X
tm = 1/Am 1/ Anm , (5.38)
n<m
due to spontaneous transitions to all lower levels. The probabilities Anm are de-
termined by (5.24) with 30 d02 3mn |dnm |2 . Due to the uncertainty relation
Et = ~, the lifetime tm of the system on a certain level corresponds to the
broadening of this level, which is equal, in circular frequency units, to the inverse
lifetime,
X
m Em /~ = 1/tm = Anm . (5.39)
n<m
Finally, the linewidth of emission or absorption corresponding to a certain pair
of levels is obviously given by the sum of the widths of the two levels,
mk = m + k . (5.40)
Hence, it follows that a weak line with d23 0 can have a large natural width due
to large d12 and d13 .
Spontaneous emission of a multilevel system excited by coherent radiation
has an interesting feature: its power oscillates in time. Let us consider the time
dependence of the dipole emission power.
According to the upgraded semiclassical formula (5.26), the power of radi-
ation emitted by an atom at a given time moment scales as the mean value of the
normally ordered square of the dipole moment second derivative,
4 X () (+)
P(t) = 3 kl dlm (t)dmk (t). (5.41)
3c klm
(For simplifying the notation, we assume that the dipole moments of all transitions
are parallel.) Here, kl is the density matrix of the atom at the initial time moment
t0 0. It is determined by the excitation, which should be pulsed for observing
quantum beats, i.e., the duration of the excitation pulse should be much smaller
than the beat period, 2/32 . In the case of a gas, this is achieved using a pulsed
laser or discharge while in the case of an atomic beam the latter should be passed
through a thin foil. As a result, a considerable number of atoms are in the same
pure non-energy state (0), so that the beat signal has a definite phase and a large
power.
After the excitation pulse is over, the dynamics of a given atom is determined
by its unperturbed Hamiltonian H0 . (Relaxation can be neglected provided that
the beat frequency is sufficiently large, 32 T 2 1.) The matrix elements in the
energy representation will then depend on time harmonically,
dmn (t) = dmn exp(imn t), dmn (t) = 2mn dmn (t). (5.42)
Let us number the levels according to increasing energy. Then the matrices of
positive- and negative-frequency operators have triangular shapes,
(+) ()
dmn (t) = dmn (t)nm , dmn (t) = dmn (t)mn , (5.43)
(a) (b)
Fig. 5.4 Quantum beats: (a) dipole moment of a three-level system in a coherent state with indefinite
energy (dashed line) oscillates with the frequencies 21 and 31 ; (b) as a result, the power P of the
spontaneous emission is modulated with the beat frequency 32 = 31 21 .
where mn 1 at m > n and 0 at m < n (see (5.30) for the case of a two-level
system). As a result, the only nonzero terms in the sum (5.41) are the ones with
m < k, m < l,
4 X 2 2
P(t) = 3 kl dlm dmk exp(ilk t). (5.44)
3c m<k,m<l lm mk
In particular, the power of radiation emitted by a three-level system contains
four terms,
P(t) = ~(31 A13 + 32 A23 )33 + ~21 A12 22
+ 2~(31 21 A13 A12 )1/2 |32 | cos(32 t + ), (5.45)
where
Anm 43mn |dmn |2 /3~c3, arg(23 d31 d12 ).
The first three terms in Eq. (5.45) scale as the populations of the excited levels and
correspond to usual spontaneous emission with the power constant in the frame-
work of our approximation. To take into account radiative energy losses, one
should add factors exp(Amn t) to these terms. The last term in (5.45) describes,
at Amn 32 21 31 , quantum beats, the periodic modulation of the total
power of spontaneous emission from two close levels 2 and 3, with the difference
frequency 32 = 31 21 (Fig. 5.4).e
Let us represent (5.45) in the form
P(t) = P0 [1 + m cos(32 t + )],
P0 ~21 A12 22 + ~31 A13 33 , (5.46)
1/2
m 2~(31 21 A13 A12 ) |32 |/P0 ,
e Note that if superscripts are omitted in (5.41), then (5.45) will contain beats with the frequency
21 , not observed in experiment.
where we have neglected the weak radiation at the difference frequency 32 , which
is usually in the microwave range. If the atom is in a pure state, then mn = cm cn ,
and
m = 2/( + 1/), |231 c3 d31 /221 c2 d21 |. (5.47)
Thus, the modulation coefficient m reaches a unity at A31 A21 , provided that
the atom is excited into a coherent state, |t0 i = (|2i + |3i)/ 2. From the classical
viewpoint, the atom emits two sine waves, which interfere, so that at some time
moments the radiation is completely canceled while at other ones it is doubled.
However, one should keep in mind that it is impossible to observe quantum beats
by measuring the field energy in the case of a single atom: multiple measurements
are required, with different t. In practice, one uses many atoms in the same state.
Usually, levels 2, 3 correspond to the fine or superfine structure, or Zeemans
sublevels, so that the beat frequency is in the microwave range and the modula-
tion can be discovered by radio electronic methods. (For instance, PMTs enable
observing modulation up to frequencies on the order of 100 MHz.) It is important
that quantum beats are a single-atom phenomenon; therefore, the Doppler effect
has nearly no influence on the beat frequency, which allows one to measure small
splitting of levels [Aleksandrov (1972)].
Note that there is another method, developed in the 1960-s, which also uses
the spectral analysis of the photocurrent rather than the direct analysis of light,
but in a stationary regime. This method, called spectroscopy of optical mixing
or spectroscopy of intensity fluctuations, allows one to observe extremely small
frequency splittings on the order of 1 Hz [Cummins (1974)].

5.2.7 Resonance fluorescence
Consider now the case of a cold gas, where T ~0 , and the atoms are excited
by external directed radiation. The total secondary field of a macroscopic sample
can be divided in two parts, the one that is coherent with the incident field, and
the scattered one.
The coherent part is determined by the space-averaged atom density, it inter-
feres with the incident primary field, and a joint resulting field propagates through
the medium in the same direction. Homogeneous matter, i.e., the constant compo-
nent in the space Fourier distribution of the matter mass, only slows the wave down
and, if the energy dissipation is taken into account, reduces its amplitude. These
effects are described by the macroscopic susceptibility of matter (Chapter 4).
The secondary radiation that is scattered sideways is caused by the atomic in-
homogeneity of the matter. The fields of different atoms are not coherent with
each other; therefore, the total power of the scattered radiation is additive, i.e.,
equal to the number of atoms times the emission power P of a single atom (pro-
vided that multiple scattering is negligible the Born approximation). Due to the
optical anharmonicity of the matter (Sec. 6.2), the spectra of the secondary radi-
ation contain, in addition to the elastic (unshifted) component, new components,
providing valuable information about the structure of the matter (fluorescence,
Raman scattering, etc.).
Here, using the gyroscopic model, we will consider resonance fluorescence
(resonance scattering), which is emission of radiation with a frequency close to
the transition frequency 0 , provided that the excitation frequency is close to
0 as well. This phenomenon, discovered by R. Wood in Na vapor as early as at
the beginning of the 20th century, attracts much attention nowadays in connection
with its two interesting features.
First, resonance fluorescence of single atoms brings the field into states with
unusual statistics, which cannot be described in terms of classical statistical optics.
These are states with photon anti-bunching and squeezed states (for more details,
see Chapter 7).
Second, in the case of a large intensity of the incident light (the pump), the
spectrum of the scattered light contains, near the elastic (Rayleigh) component,
two satellites with the frequencies and (Fig. 5.5). Here, [2 + (0
)2 ]1/2 is the generalized Rabi frequency and d0 E0 /~. Besides, amplification
of weak probe light at these frequencies due to the pump energy is observed.
Fig. 5.5 Resonance fluorescence. A monochromatic external field modulates the wave function of
a two-level system with the frequencies and . Therefore, the spectrum of the scattered field
consists of three components. The following cases are shown: (a) small mismatch and strong field; (b)
large mismatch; (c) in the case of a large mismatch, satellites appear due to the four-photon process in
which two pump photons (thick lines) become two photons with frequencies 0 and 2 0 .
The spectrum of the scattered light is observed under stationary conditions.

In this case, a quantitative theory requires an account for the relaxation and
quantization of the field (see, for instance, Ref. [Grishanin (1981)]); however, the
emerging of the satellites can be easily explained qualitatively in the framework
of the semiclassical theory and neglecting the relaxation. According to (5.3), a
monochromatic field, in addition to the precession of the R(t) vector with the field
frequency , also causes its nutation with the frequency (Fig. 5.1). As a result,
the mean dipole moment of the atom hd(t)i R x (t), and the emitted field should
contain frequencies and . Therefore, the nonlinearity of a two-level sys-
tem leads to the modulation of the field scattered by it, with the generalized Rabi
frequency . (This should be compared with the Raman scattering, where the
modulation frequency coincides with one of the frequencies of the system, so that
the satellites have frequencies 0 , and with the optical nutation, see Sec. 5.1.)
Consider two limiting cases. Let |0 | , then , and the spec-
trum of resonance fluorescence is a triplet symmetric with respect to the transition
frequency, 0 , 0 d0 E0 /~ (Fig. 5.5(a)). In the opposite case, |0 |,
and the spectrum consists of the central elastic component , a line at the transi-
tion frequency 0 , and the mirror component at 2 0 (Fig. 5.5(b)). From
the photon viewpoint, radiation at frequencies 0 and 2 0 is a result of
a single four-photon elementary process, i.e., absorption of two photons of the
incident light and emission of two secondary photons, according to the scheme
2~ ~0 + ~(2 0 ) (Fig. 5.5(c)).f
Note that in a non-resonance case, emission in the vicinity of a transition is
not exactly at frequency 0 but shifted. Let 0 > 0, then the frequency
of the anti-Stokes satellite (the right-hand one, on the frequency scale) is
00 = + [2 + (0 )2 ]1/2 0 + (d0 E0 /~)2 /2(0 ). (5.48)
This shift of the observed transition frequency, dependent on the intensity of
the exciting radiation, is called the high-frequency shift, or the optical Stark effect.
It follows from (5.48) that on the order of magnitude the shift is equal to the
squared excitation energy in frequency units, , divided by the mismatch.
Let us estimate the total power of emission. According to (5.7) and (5.32),
P(t) ~0 A22 (t) = ~0 A(/)2 sin2 (t/2). (5.49)
Thus, if the relaxation is negligible, P oscillates with the frequency , i.e., the
process is non-stationary and the notion of the spectral density is non-applicable.
It is noteworthy that in the non-resonant case, P scales as d04 /(0 )2 .
f Editors note: in modern terms, this process is called spontaneous four-wave mixing; for more details,
see Chapter 7.
Above, we did not take into account relaxation. It is intuitively clear that it
should lead to the broadening of discrete spectral components up to a width of
about 2/T 2 (Fig. 5.5). Hence, the condition for observing the satellites in the case
= 0 is > 2/T 2 , i.e., strong saturation should take place, which is only
possible with the help of lasers. (The effect was first observed in 1974 [Delone
(1975)].) Note that according to the Bloch equations with an account for relax-
ation (Sec. 4.4), nutation of the R vector disappears under stationary conditions,
so that the above-given explanation of the effect seems to be not valid. However,
the kinetic Bloch equations do not describe the fluctuations of R caused by the
thermostat and containing the frequency .
If we put aside the rotating-wave approximation, the two-level model will de-
scribe other nonlinear effects, for instance, emission at the frequency 3. In the
case of a polarized atom (or molecule) with nonzero diagonal matrix elements of
the dipole moment, dnn , 0, or in the case of the magnetic resonance, the model
will describe two-photon absorption and the Raman effect, i.e., emission at fre-
quencies 0 . Thus, according to the simplest two-level model of an atom, the
matter can scatter the incident light with changing its spectrum, which is a mani-
festation of the nonlinear properties of the matter, i.e., the optical anharmonicity.
Another simple nonlinear effect, saturation, has been considered in Sec. 4.3.
5.3 Collective emission
As we have already mentioned, emission from macroscopic matter depends

strongly on the conditions of its excitation. Under usual, chaotic (incoherent)
excitation, the states of different atoms are statistically independent, the matter
is described by a diagonal density matrix, and as a result, the emission power P
depends on the total number of atoms N linearly. (We assume that the linear size
of the sample is much smaller than the free path of a photon, l 1.) In the case
of a coherent excitation, all N atoms are in the same state, and the sample can
be described by a joint wave function. Then, as one can easily verify using the
semiclassical theory, P can depend on N quadratically, which should drastically
change the situation. In what follows, we will consider two unusual optical effects
originating from the quantum coherence formed in a macroscopic sample.
5.3.1 Superradiance
Let at the initial time moment there be N identical atoms in independent coherent
states,

j = {|1i j + |2i j exp[i(0 t + j )]}/ 2. (5.50)
This is a particular case of a possible state of a system containing N particles

distinguishable through their coordinates r j , with the wave function of the system
being factorable,
N
Y
= 1 2 . . . N = j. (5.51)
j=1
The mean dipole moment of the system is

X X X
hdi = h| d j |i = h j |d j | j i = d0 cos(0 t + j ), (5.52)
j j j
where the transition dipole moments d0 d12 are considered as real and oriented
along a certain axis. Here, the phase j of the j-th atom coherent state has a clear
meaning: this is the phase of the Bloch vector precession. If the state is prepared
by means of a /2 pulse, this phase is determined by the phase of the field at the
point r j of the atom. If N = 2, then, according to (5.52), the oscillation amplitude
of the total dipole moment is doubled at 1 = 2 and turns into zero at 1 2 = .
In the case of N atoms with the same j ,
hdi = Nd0 cos(0 t + ).
Suppose then that the linear size of the system is much smaller than the wave-
length, l 0 = 2c/0 , so that formula (5.19) for the dipole emission power is
valid. Replacing there dcl by hdi, in the case of equal phases we get Pcoh N 2 , i.e.,
the power emitted by dipoles oscillating with equal phases scales as the square
of their number. This effect of coherent (collective, cooperative) spontaneous
emission of a multi-atomic system is called superradiance. Although its classical
interpretation is trivial, the quantum theory has been first considered only in 1954
by Dicke [Fain (1972); Allen (1975)].
A close phenomenon had been observed earlier in the microwave range, in
nuclear magnetic resonance (NMR) experiments. In such experiments, a para-
magnetic sample is placed between two crossed inductance coils (the magnetic
induction method). A resonance current in one of the coils excites the stimulated
precession of the macroscopic magnetic moment M of the sample around the di-
rection of a constant magnetic field (see (4.139)). The rotating moment M induces
the induction electromotive force in the other coil, the receiver one. As a result,
the power of the observed signal scales as the squared number of nuclei in the
sample.
Coherent emission of phased dipoles is observed in many nonlinear optical
effects such as generation of optical harmonics. However, here, similarly to the
effect of nuclear induction, emission is at the frequency of the external force,
which in optics can be very different from the eigenfrequencies of the transitions.
It is important that the condition l is not necessary: at an appropriate phase
of stimulated or free precession along some direction, so that j = kz j , where
k is the wave vector and z j is the j-th atom coordinate (spatial phase matching
condition), the emission along this direction will be coherent. Thus, superradiance
of extended systems is not isotropic; it has a maximum along the phase matching
direction z.g
The semiclassical approach, based on the formula (5.19) for the power of co-
herent radiation with dcl replaced by hdi, yields an exact result only at N 1.
Recall that at N = 1, we had to introduce an additional factor of 2 into formula
(5.20). As we have mentioned above, in the general case we have to use, instead
of dcl2 , the normally-ordered square 2hd() d(+) i, which leads to the replacement of
N 2 by N(N + 1),
X
hd() d(+) i = hd() (+)
j dk i
jk
X X
= h j |d() (+)
j d j | j i + h j |d() (+)
j | j ihk |dk |k i
j j,k

X
= d02 N + cos( j k ) /2 = d02 N(N + 1)/4. (5.53)
j<k
Substituting (5.53) into (5.26), we obtain the superradiance power of N atoms in

coherent states,
Pcoh = ~0 AN(N + 1)/4. (5.54)
The last equality in (5.53) implies that all phases j are equal, i.e., the states are
mutually coherent.
Otherwise, if the precession phases are independent random variables, dis-
tributed uniformly within the 0..2 interval, then all cosines will turn into zero
after averaging, and (5.53) will only contain N diagonal terms,
Pincoh = ~0 AN/2. (5.55)
Here, the subscript incoh indicates that separate atoms, each of them being in a
coherent state on the equator of the Bloch sphere, emit incoherently.
Apparently, in addition to the limiting cases (5.53) and (5.56) considered here,
there are many other states of an N-atom system. In particular, as one can see from
the above-given example for two atoms, emission-free states are possible, in which
collective effects suppress dipole spontaneous emission.
g In Sec. 5.1, we have already come across such a phenomenon, free polarization decay.
Due to the energy losses accompanying the emission, spontaneous superradi-

ance has fast decay. The pulse duration of superradiance can be estimated as the
ratio of the energy ~0 N/2 stored by the system to the initial power of coherent
emission (5.54),
coh 2/A(N + 1). (5.56)
Thus, at N 1 and l the radiative lifetime of the coherent state of an N-

atom system is reduced N/2 times due to the collective effect. However, as it has
been already mentioned, superradiance in extended systems is directive, i.e., only
part of the field modes work efficiently and take the energy, and this leads to an
increase of coh . The increase is approximately 4/ times, with being the
effective solid angle into which the emission is directed. For a stretched sample
with the Fresnel number 2 /l 1, this slowing down of the superradiance is on
the order of (a/)2 l/ 1, where a2 is the sample cross section and l is the
length.
Consider now the case where the initial state of an N-atom system is an energy
Q
one, (0) = |2i j . The Bloch vector of each atom is directed up, towards the
North pole. Initially, the atoms emit independently,
P 2h|d () d(+) |i = Nd02 , (5.57)
and gradually pass from the energy state into coherent ones, with some precession
angles j and phases j .
However, one can expect that the atoms influence each other through the trans-
verse field, and therefore the precession phases should mutually synchronize, pro-
vided that the atom density is sufficiently high [Andreev (1980)]. As a result, after
a certain delay time t0 the system will pass by itself into a superradiance coherent
state with phased dipoles. At this moment, slow spontaneous emission will turn
into a short strong superradiance pulse with a definite phase of the field (Fig. 5.6).
5.3.2 Analogy with phase transitions

A consistent theory, as well as the experiment (see Ref. [Andreev (1980)]), per-
formed for the first time in 1973, confirm this qualitative picture. Often, it is
this surprising effect that is called superradiance. Spontaneous emerging of an or-
dered coherent state from an initially chaotic state of a macrosystem is generally
of huge interest in physics. Let us mention the analogy between the spontaneous
formation of a superradiance state in a system of excited atoms and phase tran-
sitions in equilibrium matter. Another example of such a phase transition in a
non-equilibrium system is emerging of auto-oscillations with a certain phase and
Fig. 5.6 Time dependence of the emission power for atoms excited at time t = 0: 1, usual (incoherent)
spontaneous emission, observed at low atom density; 2, superradiance of atoms that are initially in a
coherent state; 3, superradiance in the case of an energy initial state occurs with a typical delay t0 .
a macroscopic amplitude in a laser operating above the threshold [Haken (1977);

Klimontovich (1980); Arecchi (1974)]. In this case, again, the interaction of atoms
through the field of the cavity makes the chaotic noise field become a regular one
and leads to the formation of regular macroscopic alternating polarization. Note
that in equilibrium systems, phase transitions are also caused by the interaction
of atoms or molecules through the field, but a static one. A rough description of
the ferroelectric transition is provided by the Lorentz model: according to (4.50),
the frequency of the polarization resonance turns into zero provided that the atom
density is sufficiently high. In this case, susceptibility at zero frequency turns into
infinity and a macroscopic static polarization is created in the system.
Superradiance in a system with a population inversion should be distinguished
from superfluorescence (stationary or pulsed), amplified spontaneous emission
(Sec. 7.1). During superfluorescence, emission-free relaxation processes (such as
collisions in gases or interaction with phonons in solids) prevent a macroscopic
volume from emitting coherently, and therefore the emission depends on the num-
ber of particles linearly, and not quadratically.
5.3.3 Photon echo

In the case of inhomogeneous broadening, different atoms have slightly differ-
ent eigenfrequencies due to the Doppler effect, inhomogeneous static fields, etc.
Then the observed linewidth 2/T 2 of a transition is much greater than the
collision-induced or radiation-induced width 2/T 2 .
Let a short (E T 2 ) resonant /2 pulse drive the atoms of the sample from
the ground state into the same coherent state on the equator of the Bloch sphere
Fig. 5.7 Photon echo. After being excited by two short light pulses (dashed lines, top), the matter
emits a flash of light (solid line) delayed by a time t0 equal to the interval between the pulses. The
effect is explained by the fact that at time 2t0 , different molecules have the same phase of charge
oscillations (bottom).
(Fig. 5.7). It creates a polarization

N
X X X
P(t) = hd j (t)i = d0 R(xj) (t) = d0 sin((21j) t). (5.58)
j=1 j j
Here, N is the particle number density, d0 d(12j) the transition dipole moment,
R( j) the Bloch vector of the jth atom, with the components 2(21j)0 , 2(21j)00 , (11j) (22j) .
At the initial stage, when t T 2 , the difference between the eigenfrequencies
( j)
21 can be neglected, so that
P(t) = Nd0 sin(0 t), (5.59)
( j)
with 0 21 being the mean transition frequency. The macroscopic polarization
(5.59) is accompanied by strong superradiance. However, in a short time T 2 , the
dipoles in (5.58) get out of phase, P(t) becomes close to zero, and only slow
spontaneous emission remains, with a typical decay time T 1 = 1/A.
The dephasing process can be illustrated with the help of the vector model
(Sec. 5.1). In the frame of reference rotating with the frequency 0 , the R(0j) vec-
tors will undergo precession clockwise or anti-clockwise, with the angular rates
j (21j) 0 , which are within the interval +. Prior to the /2 pulse,
( j)
all R0 are directed down along the z axis, while immediately after the pulse, they
are directed along the y axis, R(0yj) (0) = 1. (We assume the pulse to be very short.)
Let T 2 be much smaller than the superradiance time coh and t T 1 , then the
radiative energy losses can be neglected and each Bloch vector is precessing with
its own velocity in the equatorial plane. The Bloch vectors will be unfolding like
a fan (Fig. 5.7), and at times t > T 2 they will uniformly fill the equatorial plane.
The precession of the jth atom can be described analytically in the rotating
frame of reference by the formula
(0j) = (i/2) exp(i j t), (5.60)
where (0j) (21j) exp(i0 t) is the slow amplitude of the density-matrix non-
diagonal element.
At first sight, this decay of the macroscopic order is irreversible. However,
there exists a smart way to restore the coherent state of the system provided that
t < T 2 , where T 2 is the typical time of irreversible precession dephasing caused by
collisions. The idea is that a second resonant pulse, with the area , is incident
on the system at a time t0 .
According to the gyroscopic Bloch equation (5.3), the pulse turns an R(0j)
vector by 180 around the x axis (5.7). Since R(0j) is in the equatorial plane, the
rotation is equivalent to the mirror reflection with respect to the xz plane or to the
change of the R(0yj) sign, i.e., complex conjugation of (0j) (t0 ).
One can easily see that after this operation, the fan or R(0j) vectors starts
folding, and at the time t = 2t0 all vectors will be parallel once again (5.7), this
( j) ( j)
time along y : R0y (2t0 ) = 1, 0 (2t0 ) = i/2. Again, we consider the pulse to
be very short.
At this time moment, a superradiance pulse is emitted again, its duration about
T 2 (Fig. 5.7). It is this pulse of emission from the sample, appearing in a time t0
after the second external pulse, that is called the photon, or spin, echo. As the
interval t0 increases, the amplitude of the echo signal reduces as exp(2t0 /T 2 ).
The appearance of echo at t = 2t0 can be illustrated by an analogy with runners
at a stadium, who start running at t = 0 with different speeds. At t = t0 they
simultaneously turn round and run back with the same speeds. Clearly, at t = 2t0
they will cross the start line simultaneously.
The coherence recovery under a pulse can be also described analytically.
Before the pulse,
(0j) (t0 0) = (i/2) exp(i j t0 ), (5.61)
while immediately after the pulse,h
(0j) (t0 + 0) = (0j) (t0 0) = (i/2) exp(i j t0 ). (5.62)

h Note that a similar operation of amplitude complex conjugation, or time reversal, forms the basis of
the phase conjugation effect (Sec. 6.5).

Further, at t > t0 , free precession takes place again, with the initial amplitude
given by (5.62),
(0j) (t) = (0j) (t0 + 0) exp[i j (t t0 )] = (i/2) exp[i j (t 2t0 )]. (5.63)
Hence, at t = 2t0 all (0j) become equal.

This effect, one of the most beautiful phenomena in quantum electronics, has
been discovered by Hahn in 1950 in NMR experiments, with inhomogeneous
broadening caused by the magnetic field inhomogeneity, and named as the spin
echo. In optics, it was first observed in ruby by Kurnit et al. in 1964 using a ruby
laser. Photon and spin echo are used for the measurement of relaxation parameters
and fine structure of transitions (echo spectroscopy, see Ref. [Manykin (1984)]).
Chapter 6
Nonlinear Optics
Some nonlinear optical phenomena have been already considered above (satura-
tion, resonance fluorescence). This chapter will present a more systematic de-
scription of these effects.
Although most nonlinear optical phenomena are well described by the semi-
classical radiation theory and do not require the quantization of the field, many
effects have a clear interpretation and classification in terms of photons. For in-
stance, frequency doubling can be represented as resulting from elementary three-
photon processes where, due to the interaction between the field and the matter,
two photons of the incident light (pump) are destroyed and a photon with the
double energy 2~ is created (Fig. 6.1(a)). It is clear from the figure that, since
the process does not change the energy of the matter (such processes are called
parametric ones), the energy of the created photon is exactly twice as large as the
energy of the pump photon.
In the case of a biharmonic pump with the frequencies 1 , 2 , the matter
emits photons with the combination frequencies, 0 = 1 2 . These are sum-
and difference-frequency generation effects, see Figs. 6.1(b) and 6.1(c). A simi-
lar description can be suggested for four-photon and higher-order processes. The
processes that take place in a parametric oscillator and in the corresponding spon-
taneous effect, spontaneous parametric down-conversion, are inverse to the sum
frequency generation. Namely, a pump photon is split in two photons with frac-
tional frequencies, 0 1 + 2 (Fig. 6.1(d)).
The efficiency of parametric processes in a macroscopic material is dramati-
cally increased under the condition of photon momentum conservation, k1 + k2 =
k0 . This equality is called the spatial phase-matching condition.
In non-parametric processes, the matter changes its energy and passes to other
energy levels. For instance, in two-photon absorption, two pump photons are
annihilated and an atom goes to an excited level (6.1(e)). Such effects lead to
135
nonlinear absorption scaling as the square of the light intensity. In a Raman two-
photon transition, a photon is annihilated and another one, with a different energy,
is created (6.1(g) and 6.1(h)).
All these effects, both parametric and non-parametric, are widely used in non-
linear spectroscopy, as well as for the variation of laser radiation frequency. An
important role is played by nonlinear optical effects in laser thermonuclear synthe-
sis, laser separation of isotopes, laser chemistry, and many other fields of quantum
electronics. Also, note that the possibility of quantum amplification is princi-
pally related to the nonlinearity of the material, since in a linear system the lev-
els are equidistant and the stimulated emission is always compensated for by the
absorption.
(a) (b) (c) (d)
(e) (f) (g) (h)

Fig. 6.1 Elementary multi-photon processes. Solid horizontal lines show real energy levels of the
matter; dashed horizontal lines denote virtual levels. Arrows pointing upwards are absorbed photons;
arrows pointing downwards, emitted ones. Bold arrows correspond to the photons of the primary
radiation (the pump), thin ones, to the secondary or spontaneous radiation. Top row: three-photon
parametric (coherent) processes; bottom row: two-photon non-parametric processes; (a) second har-
monic generation; (b, c) sum- and difference-frequency generation; (d) parametric down-conversion;
(e) non-resonance and resonance (cascaded) two-photon absorption; (f) two-photon emission (stimu-
lated, spontaneous-stimulated, and spontaneous); (g, h) Stokes and anti-Stokes Raman processes.
Nonlinear Optics 137
In the quantitative description of nonlinear optical effects, similar to the case

of linear optics, the theory is divided in two parts, the microscopic and the macro-
scopic ones. The microscopic theory is aimed at calculating the polarization P(E)
induced by a given field E. (In the general case, the magnetic field H should be
taken into account as well.) The equations of motion of charged particles are non-
linear; therefore, the function, or, rather, the functional P(E) can be written as a se-
ries expansion containing the quadratic term (2) E2 , the cubic one, (3) E3 , etc. At
the macroscopic stage of calculations, one substitutes the function P(E) into the
Maxwell equations, and a self-consistent solution E, H is searched at given field
sources and boundary conditions. According to this scheme, Sec. 6.1 discusses
the general properties of (n) , Sec. 6.2 considers various models of the optical
anharmonicity of the matter, which allow one to estimate (n) . Sections 6.36.5
describe the basic problems of macroscopic nonlinear optics, as well as the ways
to solve them, and some observable effects.
It should be stressed that nonstationary problems of nonlinear optics are solved
by writing joint equations for the field and the matter, without using the sus-
ceptibilities. For instance, a quantitative analysis of self-induced transparency
(Sec. 5.1) requires a joint solution to Maxwells and Blochs equations.
Nonlinear optics can be studied in more detail by reading Refs. [Vino-
gradova (1979); Akhmanov (1981); Fain (1972); Apanasevich (1977); Landau
(1982); Akhmanov (1964); Bloembergen (1965); Kielich (1980); Fabelinsky
(1965); Klyshko (1980); Letokhov (1975); Akhmanov2 (1981); Delone (1978);
Dmitriev (1982); Butylkin (1977); Schubert (1973); Zernike (1973); Walther
(1976); Letokhov (1983)].a Currently, there is also a new rapidly developing field,
nonlinear optics of the surface [Chang (1981)]. One of the most interesting phe-
nomena in this field is giant Raman scattering of light by molecules adsorbed on
a rough metal surface. Cross section of this scattering is by orders of magnitude
greater than the cross section of usual bulk scattering per one molecule.
6.1 Nonlinear susceptibilities: definitions and general properties
Before analyzing various models of optical anharmonicity, it is reasonable to find

out the general properties, such as the symmetry of the nonlinear response of the
matter, which do not depend on the choice of the model. For this, let us generalize
the notion of the phenomenological susceptibility (Sec. 4.1) to the nonlinear case.
a Editors note: See also [Agrawal (2007); Boyd (2008)].

6.1.1 Nonlinear susceptibilities

Let a field with a discrete spectrum,
X
E(t) = (1/2) En exp(in t) + c.c., n = 1, 2, . . . , (6.1)
n
be incident on the matter. Equations of motion for the charged particles of the
matter are nonlinear. As a result, charge displacements induced by the field (6.1)
and, hence, the polarization P(t) will contain not only the Fourier components
with the frequencies of the external force n but also the ones with the combina-
tions of these frequencies, n m , including multiple frequencies 2n and the
zero frequency, n n = 0.
Let us first reduce the consideration to the case of the nonlinearity quadratic
in the external field. Then, the phenomenological relation between the spectral
components of the polarization and the electric field has the form
P(2) (2)
0 = (1 , 2 )E1 E2 , (6.2)
where P0 is the complex amplitude of polarization oscillations with the frequency
0 = 1 + 2 , and we are so far assuming 1 , 2 . The quadratic susceptibility
(or quadratic polarizability) (2) defined this way sets a relation between three
vectors and is therefore a third-rank tensor. Notation (6.2) is not based on a certain
frame of reference (often, the relation is written with a colon, P0 = (2) : E1 E2 ).
If some Cartesian frame of reference is chosen, (6.2) takes the form
X
P(2)
0 = (2)
(1 , 2 )E 1 E 2 . (6.3)

In what follows, as usual, we will omit the summation over dumb indices , .
Each of the 27 components (2) 0
(, ) of the
(2)
tensor is a function of two
0
independent arguments , taking values from to +. Since the Fourier
components of the field and the polarization are complex, (2) is also complex,
and in total, there are 54 real functions of these variables. However, as we will
show below, there are many relations between these functions, and the number of
independent variables is reduced.
By analogy with (6.3), the nonlinear susceptibility of an arbitrary order is
defined as
P(m) (m)
0 = 1 ...m (1 , . . . , m )E 11 . . . E mm ,
(6.4)
0 = 1 + 2 + + m .
For instance, the cubic susceptibility (3)

(1 , 2 , 3 ) is a fourth-rank tensor, and
each of its 81 complex components depends on three continuous variables. Note
that the Cartesian indices , , . . . can be also considered as arguments of (m) ,

each of them taking three discrete values.
The relation between the polarization and the field can be written symbolically
as a power series,
X
P(E) = (m) Em . (6.5)
m=0
In the spectral representation, this relation is an algebraic one, while in the time
representation, (m) should be understood as integral operators. Their kernels,
(m) (t1 , . . . , tm ), called multi-time Greens functions or response functions of the
matter, are defined in terms of the spectral susceptibility (m) (1 , . . . , m ) using
the m-fold Fourier transformation (for m = 1, see (4.7)). By analogy with the
linear case (Sec. 4.1), the causality principle leads to integral relations between
the real and imaginary parts of (m) , similar to the Kramers-Kronig relations.
The effect of the magnetic field can be taken into account by doing a double
power-series expansion in (6.5), (mn) Em Hn . Some nonlinear optical effects
P
reveal spatial dispersion, which can be described as the dependence of (m) not
only on 1 , . . . , m but also on the wavevectors k1 , . . . , km .

6.1.2 Various definitions
Often, one uses the definition of spectral amplitudes En that differs from (6.4) by
the absence of the 1/2 factor,
X
E(t) = En exp(in t) + c.c., En En /2. (6.6)
n
Similarly, at 0 , 0, P0 = P0 /2, so (6.2) leads to the following relation for the

quadratic nonlinearity:
(2)
P0 = (1/2)(2) E1 E2 = 2(2) E1 E2 (2) E1 E2 . (6.7)
Thus, two different definitions for the field and polarization spectral ampli-
tudes correspond to two differently defined m-th-order susceptibilities, related as
(m) = 2m1 (m) , 0 , 0. (6.8)
An exception from this rule are even-order susceptibilities at 0 = 0, describing
the optical rectification effect. For this case, P0 = P0 , and
(m) = 2m (m) , 0 = 0.
Assume now that the inertia of the material response can be neglected. This
is the so-called inertia-free, or Kleinmans, approximation, which is valid in the
case where the field frequencies and their combinations are in the transparency
windows of the matter. In this case, the polarization follows the field instanta-
neously, and for m = 2,
P(2)
(t) = E (t)E (t), (6.9)
where is some real constant tensor. In particular,
P(2) 2
x (t) = xxx E x (t) + xxy E x (t)E y (t)
+ xyx Ey (t)E x (t) + xyy Ey2 (t) + . . . (6.10)

From this example, one can see that only the sum + has a physical
meaning; these components cannot be measured separately, and therefore the
tensor is symmetric in the last two indices while the (m) tensor is symmetric in
all indices except the first one.b Assuming the field in (6.9) to be biharmonic, we
obtain
X

P(2)
(t) = (1/2) Re{ [En En + En En exp(2in t)]
n=1,2

+ 2E1 E2 exp[i(1 + 2 )t] + 2E1 E2 exp[i(1 2 )t]}. (6.11)
The first term here describes optical rectification, the second one, harmonic gen-
eration, the third and the fourth ones correspond to the generation of sum and
difference frequencies.
On the other hand, from the definition of the susceptibility (6.3) it follows that
X

P(2)
(t) = (n , n )En En
n=1,2
X
+ Re{ (n , n )En En exp(2in t)
n=1,2
+ (1 , 2 )E1 E2 exp[i(1 + 2 )t]

+ (1 , 2 )E1 E2 exp[i(1 2 )t]}. (6.12)
(Here we have omitted the order index of the susceptibility, which we will some-
times do in what follows.)
Comparing (6.11) and (6.12), we see that in the dispersion-free approximation,
(, 0 ) = 2(, ) = 2(, ) = , (6.13)
where , 0 , 0. Similarly, one can show that (, 0) = 2. The first equality
in (6.13) is obviously still valid in the presence of dispersion, provided that and
0 are sufficiently close. Thus, every component of the (, 0 ) tensor, considered
b Below, we show from the energy considerations that the (m) tensors are fully symmetric.
as a function of two variables, and 0 , has a peculiarity on the lines 0 =

where its values are twice as small as at the neighbouring points.
Higher-order susceptibilities have similar peculiarities at coincident frequency
arguments. The corresponding coefficients can be found by repeating the deriva-
tion of (6.13).

6.1.3 Permutative symmetry
Definitions (6.3), (6.4) lead to the invariance of the (m) tensors to permutations
of their frequency arguments together with the corresponding Cartesian indices,
(1 , 2 ) (2 , 1 ), (6.14)
(1 , 2 , 3 ) (1 , 3 , 2 ) . . . (6.15)
Indeed, (6.3) can be rewritten in other equivalent forms,
P(2)
0 = (2 , 1 )E 2 E 1 = (2 , 1 )E 2 E 1 .
Comparison of the latter with (6.3) yields (6.14).

It follows from (6.14), (6.15) that tensors describing the generation of har-
monics are symmetric with respect to all indices except the first one.
Another general property of susceptibilities follows from the fact that both
the polarization and the field are real values, which requires that the amplitudes
should turn into their complex conjugates as the frequencies change their signs.
Changing all frequency signs in (6.4) and doing complex conjugation, we obtain
P0 = (1 , . . . , m )E1 . . . Em .
Comparing this with (6.4), we find (see (4.5))
(1 , . . . , m ) = (1 , . . . , m ). (6.16)

6.1.4 Transparent matter
Note that spatio-frequency permutative relations (6.14), (6.15) do not concern the
first index. Using (2) as an example, let us show that its permutation is possible
in the non-resonant case where all frequencies are away from the resonances of
the matter. For the sake of symmetry, we introduce the third argument into the
notation of the susceptibility,
(1 , 2 ) (0 ; 1 , 2 ) 012 = 012 , (6.17)
where the minus sign by the combination frequency 0 = 1 + 2 provides that
the sum of all three arguments of the susceptibility is zero. In the last equality, we
took into account relation (6.16). Note that sometimes, other notation can be used
as well,
(0 ; 1 , 2 ) (0 = 1 + 2 ).
Let us find the power absorbed in a unit volume of the matter due to the
quadratic nonlinearity in the case of a three-frequency field. According to (4.13),
2
(2)
X
P(2) = E P = (1/2) n ImEn P(2)
n P0 + P1 + P2 . (6.18)
n=0
Hence, according to definition (6.3), the partial powers are

P0 = (1/2)0 Im012
E 0 E 1 E 2 ,
2
P1 = (1/2)1 Im10
E 1 E 0 E 2 , (6.19)
10
P2 = (1/2)2 Im2 E2 E1 E0 .
Note that the signs and the values of the powers Pn depend on the field phases.
With the help of (6.16), we find from (6.19) that

P(2) = (1/2)Im[1 (012 102 012 210
) + 2 ( )]E 0 E 1 E 2 . (6.20)
If all three frequencies are far from resonances, then the absorption is absent and
the matter only redistributes the energy between the three frequency components
of the field. According to (6.19), the share of the nth component scales as n .
In a transparent matter, P = 0, and since the complex amplitudes Ena are arbi-
trary, the expression in the square brackets of (6.20) should turn to zero. Provided
that the dispersion is weak, the frequencies can be also considered as arbitrary,
so that each of the coefficients by 1 and 2 is zero as well. Hence, in addition
to the automatic symmetry relations (6.14), (6.15), which do not concern the first
index, a medium that is transparent at all three frequencies imposes the following
bounds:
012 102 210
= ) = . (6.21)
In the general case of an arbitrary-order non-resonant susceptibility (m) , there
is complete permutative symmetry with respect to all indices. A possibility to have
the permutation of the first indices leads to the Manley-Rowe relations (Sec. 6.3):
P0 /0 = P1 /1 = P2 /2 . Resonance nonlinear susceptibilities have symmetries
more limited than given by (6.21) (Sec. 6.3). For instance, the Raman susceptibil-
ity satisfies the relation
1xxxx
212
= 2xxxx
112
.
In the dispersion-free approximation, (m) have no frequency dependence at

all, and the susceptibility tensors are therefore symmetric in all indices.
As an example of using Eq. (6.21), consider the case 2 = 1 . Then, with
an account for (6.13),
4 (0 = 1 1 ) = (1 = 1 + 0). (6.22)
The presence of the factor 4 here can be verified by repeating the derivation of
(6.13) in the presence of a constant field. The susceptibility on the left-hand side
describes optical rectification, while the one on the right-hand side is responsible
for the linear electrooptic effect (the Pockels effect), i.e., variation of the refractive
index n(1 ) at the frequency 1 scaling as the static field E0 . Indeed, polarization
at the frequency 1 , with an account for linear and quadratic susceptibilities, is
P1 = [ (1 ) + (1 = 1 + 0)E0 ]E1 ( + )E1 . (6.23)
The susceptibility increase will manifest itself in an anisotropic variation of
n(1 ). Thus, (6.22) sets a quantitative relation between two different phenom-
ena, the Pockels effect, which has been known for a long time, and the optical
rectification, which has only been discovered after the invent of lasers. Another
example of such a pair of related phenomena is given by Faradays direct and
inverse effects. (The inverse Faraday effect is the emerging of static magnetization
scaling as the intensity of a circularly polarized light wave.)
Thus, according to (6.14), (6.15), and (6.21), nonlinear susceptibility tensors
(m) of a transparent material are invariant to all (m + 1)! permutations of their
space-frequency arguments.
The absence of dissipation in the transparency windows enables one to define
the energy v of the polarization of the matter (Sec. 4.1). If dispersion, i.e., delay
of the response, is completely neglected, then, by analogy with (4.28),
v(2) (t) = (1/3) E E E , (6.24)
where E E(t). In the thermodynamic approach, this energy should be added
to the free energy density of the matter F and to other thermodynamic potentials
(Sec. 4.1). Then, P, (1) , and (2) are determined, respectively, by the first, second,
and third derivatives of F(E) at the point E = 0, see (4.32), (4.33).
Let the field contain three harmonics,
X
E(t) = (1/2) En exp(in t), (6.25)
n
where
n = 0, 1, 2, 0 = 1 + 2 , 1 , 2 , n , 0,
n n , En En ,
then the time-averaged polarization energy is

v(2) v(2) (t) = (3!/3 23 ) E0 E1 E2 + c.c. (6.26)
Dispersion should be taken into account by replacing with (, 0 ), which
is possible due to the symmetry (6.16), (6.21). As a result,

v(2) ({En }) = (1/4)(012 012
E 0 E 1 E 2 + E 0 E 1 E 2 ). (6.27)
This relation allows one to calculate the nonlinear polarization, as well as (2) , in
terms of v(2) (see (4.36)),

P(2) (2)
n = 4v /E n . (6.28)
6.1.5 The role of the material symmetry

In different reference frames, the components of the field E and polarization P
vectors, as well as the tensors (m) relating them, are of course different. Under
the rotation of a Cartesian frame of reference, the susceptibility transforms as
X
(m)
0 0 ... = a0 a0 . . . (m)
... , (6.29)
,,...
with the tilde denoting the components in the new frame of reference. Here, a is
the matrix describing the transformation from the old components to the new ones,
and we assume that rotation also includes the sign changes in some or all coordi-
nates, i.e., inversion and mirror reflection. For instance, in the case of inversion,
a0 = 0 , and
(m)... = (1)m+1 (m)
... , (6.30)
where m + 1 is the rank of the tensor. Thus, as a result of the reference frame in-
version, the components of odd-rank tensors (in particular, vectors and quadratic
susceptibility tensors) change their signs, E = E , (2) (2)
= , . . . .
For tensors describing the physical properties of a material, there is a special
frame of reference where the tensor has its simplest form, with the maximal num-
ber of zero or equal components. In crystals, this natural frame of reference
coincides with the crystallographic one. For instance, real tensors of rank two are
diagonal in the natural frame of reference, (1)
= .
Any unbounded medium, either amorphous or crystalline, has a certain sym-
metry of particles positions averaged over their thermal motion. Formally, this
symmetry of the medium is determined by a set (group) of a certain number of
symmetry elements. In particular, the elements of a pointc group of symmetry are
c The term point is due to the fact that rotations leave one point (the origin) fixed, unlike translational
transformations of the coordinates.

all rotations a of the reference frame, including mirror reflections and inversion,
which leave the structure of the medium unchanged. For instance, many crys-
tals, as well as optically non-active liquids and gases, are invariant with respect to
inversion. Such media are called centrally symmetric.
Similarly, any macroscopic property of a medium, which is characterized by a
certain tensor, has its own group of symmetry elements. The symmetry elements
of a tensor are all rotations a(i) that act according to rule (6.29) and do not change
the components of the tensor. For instance, according to (6.30), all even-rank
tensors are invariant with respect to inversion.
It seems obvious that the symmetry of a macroscopic property of a medium
cannot be lower than the symmetry of its structure (the von Neumann principle,
Ref. [Nye (1957)]). In other words, the symmetry group of a property should in-
clude all symmetry groups of the structure, i.e., the latter is a subgroup of the sym-
metry group of the property. Hence, if a is an element of a point symmetry group
of a medium, then the tilde in (6.29) can be omitted. Then, (6.29) becomes a rela-
tion between different components of the (m) tensor. By substituting into (6.30)
all symmetry elements a(i) of the medium one by one, we obtain a homogeneous
system of equations for (m) . In isotropic media and in crystals, such equations
greatly reduce the number of susceptibility nonzero components, as well as make
many components equal, sometimes up to the sign.
The most bright example follows from (6.30) in the case of centrally sym-
metric media. According to the von Neumann principle, all tensors describing
the physical properties of such media should be also centrally symmetric, i.e., the
tilde in (6.30) can be omitted. Hence, for even m, it should be (m) (m)
... = ... ,
which leads to (m) = 0. Thus, in centrally symmetric media, all even-order sus-
ceptibilities are equal to zero.
Note that this conclusion is not valid in the case of susceptibilities describing
magnetic effects. This is because the magnetic field and the magnetization are
pseudovectors (axial vectors, as they do not change their signs under the inversion
of the coordinates). As a result, the corresponding susceptibilities are pseudo-
tensors and do not transform according to (6.30) under inversion. In particular,
the Faraday effect, described by the relation P1 = (1 = 1 + 0)E1 H0 , is also
possible in centrally symmetric media.
6.2 Models of optical anharmonicity
Depending on the features of the matter, its state, the frequencies of the incident
fields, and other experimental conditions, various mechanisms can contribute in
the observable effects. Below we will consider several particular classical models
of optical anharmonicity and then present the general quantum scheme of calcu-
lating nonlinear susceptibilities.
6.2.1 Anharmonicity of a free electron

Let an electron (or any charged particle) be in the field of a plane monochromatic
wave E polarized linearly along the x axis and propagating along the z axis. With
an account for the magnetic part of the Lorentz force, non-relativistic equations of
motion for the electron have the form (e < 0)
e 1 e 1
X + 2 X = E x ZHy = 1 Z E,
m c m c
e
Z + 2Z = XE, (6.31)
mc
where is a phenomenological damping constant providing that after the periodic
field is switched off, the motion proceeds with a stationary amplitude, and
E E x (Z, t) = ReE1 exp[ikZ(t) it] = Hy (Z, t) (6.32)
is the field at the particle location. Damping can be caused by collisions, as well as
by the reaction of the radiation, i.e., radiation losses (radiation friction). Suppose
that the transverse displacement Z of the electron is small, then in (6.32) one can
assume Z = 0 (dipole approximation, i.e., zero-order approximation in kZ).
Let us search the stationary solution to (6.31) in the form of a Fourier expan-
sion using the method of successive approximations in the field amplitude,
R = R(1) + R(2) + = Re(R1 eit + R2 e2t + . . . ), (6.33)
where R {X, Y, Z}. In the first approximation, one can neglect the effect of the
magnetic field, so that the response of the electron is linear,
e/m
X1(1) = E1 . (6.34)
2 + 2i
Hence, the linear polarisability tensor of an electron, which defines the relation
between the amplitudes of the dipole moment d1 = eR1 and the field E1 , is
e2
() = . (6.35)
m( + 2i)
In our model, there is no neutralizing positive charge, and the dipole moment is
defined with respect to the origin, R = 0. After multiplying by the density N
of electrons, we find the linear susceptibility (1) of cold (free of Doppler effect)
plasma, see (4.52).
Thus, at , the linear polarisability of an electron is
e2
= = re o2 , (6.36)
m2
where re e2 /mc2 3 1013 cm is the classical radius of the electron. Note that
re is related to other typical length parameters through the fine structure constant
e2 /~c2 1/137,
o0 /2 = 137a0 = 1372oc = 1373 re , (6.37)
where 0 2o 1/R 105 cm is the wavelength corresponding to the ion-

ization potential of the hydrogen atom 13.6 eV, R me4 /4~3c is the Rydberg
constant, a0 ~2 /me2 5 109 is Bohrs radius and oc ~/mc 4 1011 cm is
the Compton wavelength. Let = 0 , then
= 4a30 6 1025 cm3 , (6.38)
i.e., polarisability of a free electron in the UV range is, on the order of magnitude,
equal to the hydrogen atom volume, i.e., to the polarisability of a bound electron
in the absence of a resonance.
In order to find the second approximation, one should replace R by R(2) in
the left-hand sides of (6.31) and by R(1) in the right-hand sides. Then the Lorentz
force will have components with both zero and double frequency,
Fz(2) = (e/c)X (1) E = (1/2)kIm (|E1 |2 + E12 e2it ). (6.39)
The double-frequency force causes longitudinal oscillations of the electrons with

the frequency 2. Their amplitude, according to (6.31) and (6.39), is
e2 E12 1
Z2(2) = zxx E12 . (6.40)
8im2 c( + i)( + 2i) e
In the last equation, we have introduced the quadratic polarisability tensor of a
free particle, which provides the relation between the amplitudes of the field and
the dipole moment eZ2(2) at the double frequency. Thus, the quadratic polarisability
of a free electron at is
f ree
|| e3 /m2 c3 = (eo/mc2 ) /E NL . (6.41)
Here, E NL is a typical parameter equal to the field amplitude at which the linear
and quadratic responses are equal, Z (2) = X (1) . At = 0 ,
f ree
E NL = e/re o0 109 G, (6.42)
Fig. 6.2 Due to a plane monochromatic wave, an electron moves along a Lissajous pattern.
which corresponds to the intensity 1020 W/cm2 .d Below, we will show that the
anharmonicity of a bound electron is two orders of magnitude as large as that of a
free electron even in the absence of a resonance,
bound f ree
E NL /E NL 0 /137,
where
bound
E NL e/2a20 107 G. (6.43)
Multiplying by the density of the electrons, we get the quadratic susceptibility
of the plasma,
f ree
|(2) | = e3 N/m2 c3 = |(1) |/E NL . (6.44)
Thus, one of the fundamental sources of the matter anharmonicity is the
Lorentz force. Note that the denominator of (6.44) contains the speed of light,
which is typical for magnetic effects.
According to (6.41), the ratio Z (2) /X (1) has an order of E1 /E NL , while kZ (2)
(E1 /E NL )2 . This justifies for using the dipole approximation in the calculation of
at E1 E NL .
If the static force of light pressure is not taken into account, then, accord-
ing to (6.34) and (6.40), an electron in a magnetic field moves in the xz plane
along a figure of eight trajectory (Fig. 6.2). Stimulated oscillations of the elec-
tron along the field, X (1) (t), are accompanied by dipole emission in all directions,
except the exact x one. This is what is called Thomson scattering or, taking into
account recoil, Compton scattering. At the same time, longitudinal oscillations of
d In the Gaussian system of units, E and H have the same dimensionality; therefore, E can be measured
in Gausses (1G = 300 V/cm).

the electron, Z (2) (t), lead to dipole emission at the frequency of the incident field
second harmonic, i.e., the effect of frequency doubling. The second-harmonic
emission is maximal in the transverse plane, xy, and absent along the primary
field propagation direction. This structure of the free-electron quadratic polaris-
ability tensor prevents the second-harmonic amplitudes in a macroscopic sample,
such as plasma or a semiconductor, from adding up coherently.
Thus, electrons in a plasma, metal, or semiconductor provide, in addition to
Thomson scattering, also non-coherent scattering with the double frequency and
the intensity scaling as 2 N. Bound electrons in atoms or molecules also have
quadratic polarizability, which leads to a non-coherent scattering at the double fre-
quency, termed hyper-Rayleigh scattering. From the quantum-mechanical view-
point, it is interpreted as the absorption of two incident photons and the emission
of a double-energy photon (Fig. 6.1(a)). If the phase velocities of the incident
wave and its second harmonic are equal, n() = n(2) (the so-called phase match-
ing condition), then, provided that the structure of the (2) tensor is appropriate,
weak nearly isotropic hyper-Rayleigh scattering is accompanied by a much more
intense longitudinal emission, scaling as 2 N 2 = (2)2 (Sec. 6.5).

6.2.2 Light pressure
The constant component of the Lorentz force (6.39) determines the static light
pressure force F0 acting on the electron due to the traveling wave. According to
(6.39), F0 scales as the imaginary part of the electron linear polarisability, 00 , i.e.,
the power of the scattered light, P = 00 |E1 |2 /2 (see (4.15)), or the interaction
cross-section,
P/I = 4k00 , (6.45)
where I = c|E1 |2 /8 is the intensity of the plane wave. Therefore, the force of
light pressure can be represented in the form
F0 = k00 |E1 |2 /2 = P/c = I/c. (6.46)
This force accelerates the electron; however, collisions in plasma will lead
to a constant speed of electron drift, Z0 = F0 /m, where = 1/ and is the
time between collisions. (Recall that is the oscillation damping rate, which can
exceed ). The constant current emerging along the beam, scaling as |E1 |2 , can be
interpreted in terms of optical rectification. (In dielectrics, the term rectification,
or dc-effect, means the appearance of a static field E0 |E1 |2 .)
Let , then, according to (6.35), 00 = 2e2 /m3 . Let us estimate F0 in
the case where the damping of electron oscillations is only due to radiation losses
(radiation friction). According to (6.31), the friction force, in the first order, is
2mX (1) . Multiplying it by the velocity X (1) and averaging over a single period,
we find the power of losses, P = m2 |X1(1) |2 . Setting equality between this
expression and the power of dipole emission (5.19), we obtain
2/ = 00 /0 = 2re /3o. (6.47)
The same result follows from (5.24) with a unity oscillator strength (4.62). Now,
(6.46) takes the form
F0 = re2 |E1 |2 /3. (6.48)
Comparing (6.46) and (6.48), we find the Thomson cross-section of scattering by

a free electron,
T = 8re2 /3. (6.49)
From the photon viewpoint, light pressure is due to the fact that an electron ac-
quires the momenta of absorbed photons, which are then symmetrically re-emitted
in all directions via Thomsons (or Comptons) scattering. Let us stress that we
have only considered the average value of the force, which has quantum fluctua-
tions [Minogin (1986)].
We have found the force of light pressure acting on a free electron in the case
of a traveling wave. Similar analysis can be carried out for a more complicated
spatial configuration of light. For instance, an electron will be displaced in the
field of a standing wave, which is the Kapitsa-Dirac effect. It is important that
in a non-homogeneous field, the Lorentz force has a nonzero value averaged over
the period even at 00 = 0. In this case, F0 is determined by 0 , which is much
greater than 00 , so that the latter can be assumed to be zero. This force is caused
by the exchange between different plane waves and is called the stimulated one
(in contrast to the spontaneous force (6.46)).
Displacement of an electron R(t) due to a monochromatic field is
E(R0 , t)/e, where = me2 /2 is the polarisability and R0 is the non-perturbed
coordinate of the electron. Hence, we find the averaged Lorentz force,
F0 = E H/c = kIm E H /2, (6.50)
where the last equality is also valid for the case of a complex polarisability. Note
that in a plane traveling wave, EH sin t cos t 0, and that in the general
case the force F0 does not scale as the averaged Poynting vector, S0 = cRe E
H /8.
Consider first the case of two plane waves,

2
X
E = En exp(ikn R0 + in ),
n=1
Hn = kn En , kn = kn /c.
At 00 = 0, (6.50) contains only cross components,
F0 = (1/2)kIm (E1 H2 E2 H1 )ei

= (1/2)Im[k(E1 E2 ) + E1 (E2 k1 ) E2 (E1 k2 )]ei , (6.51)
where k R0 + , k k1 k2 , 1 2 .
It is easy to see that the part of F0 scaling as k can be represented in the
gradient form with the effective potential E 2 /2 (from (4.28) it is clear why
the factor 1/2 appears),
F M = |E |2 /4. (6.52)
Note that this part of F0 , called the Miller force, disappears, according to (6.51),
in the case where the waves are orthogonally polarized. For counter-propagating
waves (k2 = k1 ) with the same linear polarization, F0 = F M :
F0 = k1 E1 E2 sin(2k1 R0 + ). (6.53)
Due to this force, charged particles tend to bunch in the nodes of a standing wave.
In the general case, a field consists of a continuum of plane waves, and the light
pressure force can be found by integrating (6.51) in k1 and k2 . Note that in this
case, the gradient part (6.52) maintains its form. It should be stressed that in the
presence of the electron recoil, the interacting waves have different frequencies.
The corresponding phenomenon is called the stimulated Compton effect.
Consider further the pressure of light acting on bound electrons in an atom or
a molecule, i.e., on neutral polarisable particles. As the starting point we take the
effective potential V = d E(r), where d is the induced dipole moment and r is
the coordinate of the particle center of mass. Hence (see(4.39)),
F = d(t) E(r, t), (6.54)
or F = d E /x . Assuming
E x = (1/2)E1ei(kzt) + c.c. E (+) + E () ,

d x = ()E (+) + c.c.
and selecting the constant component, we find (6.46) once again.

Consider first resonance pressure. Let, as before, the damping be only due to
the emission by the particle, i.e., the resonance fluorescence (Sec. 5.2), then the
width of the resonance is minimal (the natural width), 2rad = A. Then, according
to (2.55), = 2o2 , and it follows from (6.45), (6.46) that 00 = o3 /2 and
F0 = o2 |E1 |2 /4. (6.55)
2
Thus, the resonance light pressure on a bound electron is o /re2
= 137 times6
as great as on a free one, provided that the damping is only caused by radiation.
This huge difference is due to the high ratio of the resonance and Thomsons cross
sections.
Estimate (6.55) relates to the case where only the lower level is populated. In
the general case,
00 = (1/2)o3 = o3 (0) /2(1 + 2W0 T 1 ), (6.56)
(0)
where and are relative population differences with an account for saturation
and without it, respectively (Sec. 4.3), W0 is the transition probability and T 1 is
the time of longitudinal relaxation. In the case of radiation relaxation, T 1 = 1/A.
According to (6.56), in the presence of population inversion the light pressure
force is directed oppositely to the light beam, which in the photon language can
be explained in terms of the recoil of photons emitted via stimulated transitions
forward.
At strong saturation, the power P absorbed by the atom, according to (4.102),
is ~(0) /2T 1 , so that (6.46) leads to
F0 = ~k(0) /2T 1 , (6.57)
the force scales as the momentum of the incident photon times the number of
photons scattered per unit time. Let = 1 and A = 106 s1 , then F0 = 3 1017
dyn, and at m = 3 1023 g the acceleration is as high as 106 cm/s2 . The intensity
of saturating light is in this case much greater than 104 W/cm2 .
Resonance pressure of laser light provides quite unusual applications. With its
help, one can accelerate, displace, and focus beams of neutral molecules, separate
isotopes, trap molecules within a small space domain, and reduce their thermal
velocities [Minogin (1986)].
6.2.3 Striction anharmonicity

Now, let all frequencies of the field be in the transparency range of the matter.
Then the dispersion can be neglected and the polarisability of a particle can be
considered as a real constant. In this case, the force takes the form
F = E 2 /2 = E(+) E() . (6.58)
Here, the bar denotes averaging over high-frequency components. Indeed, we are
only interested in the static part of the force acting on a molecule as a whole, or at
least in the part that is varying slowly compared with the frequencies of the field
and the molecule. For notation E () , see Sec. 7.2.
The gradient force (6.58) corresponds to the effective potential of the molecule
V = E 2 /2.
The additional energy density of the matter and the force density in an optical
field will be N times as high (see (4.35)),
v = E 2 /2 = (n2 1)E 2 /8, (6.59)
f = E 2 /2, (6.60)
where (1) N, N is the density of molecules, which is assumed to be

uniform, and n = is the refractive index.
Here, we have neglected the interaction between molecules, which is only
possible for a sufficiently small N. One can show that in the general case (see
Ref. [Landau (1982)]), one should make a substitution into (6.60),
! !

= , (6.61)
T 4 T
where is the density of the matter. For instance, from the Clausius-Mossotti
relation, which can be easily obtained from (4.48),
1 4
= , (6.62)
+2 3m
it follows that
! !
1 ln 3
= = ,
T T ( 1)( + 2)
so that in the case of a dense medium, (6.60) should be multiplied by the Lorentz
correction ( + 2)/3.
In a traveling plane wave, E 2 has no constant component, hence F = 0.
(Note that we are considering a range where 00 = 0 and the spontaneous force
(6.46) related to the dissipation is absent.) However, in a standing plane wave,
E = 2E1 cos(kz) cos(t), and from (6.58), it follows that (compare with (6.53))
Fz = kE12 sin(2kz). (6.63)
Thus, at > 0 particles gather in the anti-nodes of a wave. The force (6.63),
scaling as the polarisability, is called the stimulated force.
In limited beams of light, there is a static transverse gradient of the field
square, and at > 0 the particles tend to move towards the beam axis. Under
stationary conditions, the force density (6.60) should be compensated for due to
the increase in the pressure, p, and the density of particles, N, in the central
part of the beam,
p = v = E 2 /2, (6.64)
N/N = / = T p, (6.65)
where T is the isothermic compressability of the medium. These equations de-
scribe electrostriction in a light field.
The increased density of particles in the light beam will cause a change in the
susceptibility of matter,
= N = T 2 |E1 |2 /4. (6.66)
At the same time, by definition,
P(3) (3) 2
1 = ( = + )|E 1 | E 1 = E 1 ,
so that the electrostriction contribution to the cubic susceptibility is

2
(3) = T (1) /4. (6.67)
Hence, we find the typical nonlinear parameter
2
E NL (1) /(3) = 4/T (1) . (6.68)
In liquids, n 1.5 ( 0.1), and T 1010 dyn. (Recall that 1/v2 , where v
is the speed of sound.) Hence, (3) 1013 cm2 /erg, E NL 106 G.
The gradient force (6.58) and the corresponding pressure 6.64 are very im-
portant for applications: they enable one to generate strong ultra-sound waves
using bi-harmonic laser fields. The same force causes stimulated Mandelshtam-
Brillouin scattering (see below). Striction nonlinearity (6.67) is one of the reasons
for the self-focusing of light. Another important mechanism, the optical Kerr ef-
fect, i.e., orientation of anisotropic molecules of a liquid in a linearly polarized
light beam, will be considered further.e
6.2.4 Anharmonic oscillator

In classical Lorentzs dispersion theory (Sec. 4.2), electrons in atoms are assumed
to be harmonic oscillators. It is natural to consider the optical nonlinearity of mat-
ter in terms of the anharmonic oscillator model. Let a particle be in the potential
V(x) = m20 x2 /2 mx3 /3 eEx, (6.69)
e Editors note: The same mechanism is used in the atom optics, namely, in making atomic beams
scattered by an optical lattice.

where e is the particle charge and is a small parameter defining how much the
shape of the potential well differs from a parabolic one. (For simplicity, we con-
sider a one-dimensional model.) From (6.69), we find the equation of motion
Dx x + 2 x + 20 x = eE/m + x2 . (6.70)
Let the external field E be a bi-harmonic one. We search the stationary solution
as a perturbative series,
X
x(t) = x(m) (t), x(m) m1 E m , (6.71)
m
where m = 1, 2, 4, 8, . . . . Substituting (6.71) into (6.70) and setting equalities

between terms of the same order of magnitude, we find the solution in the form of
iterations,
Dx(1) = eE/m, Dx(2m) = x(m)2 . (6.72)
In the first order,
x(1) = Re(x1 ei1 t + x2 ei2 t ),
(6.73)
xn = (n )En /e, () = e2 /mD(),
where
D() 20 2 2i = D ().
In the second order, according to (6.72), there are components of x(t) with
frequencies 0, 21 , 22 , 1 2 . Consider the response x3 at the frequency 3
1 + 2 . It follows from (6.72) at m = 1 that
x3 = x1 x2 /D(3 ) (3 = 2 + 1 )E2 E1 /e,
where
(3 = 2 + 1 ) = e3 /m2 D(3 )D(2 )D(1 ). (6.74)
At 1 = 2 , contains an additional factor of 1/2 (see (6.13)). According to
(6.14), (6.15), (6.16), polarisability is invariant to the permutation of the last two
arguments, as well as to a simultaneous change in the signs of all three frequencies
and the imaginary unity. However, according to (6.74), the first argument can be
interchanged with the second or the third ones only in the absence of a resonance
at one of the frequencies, with |n 0 | (compare with (6.21)). From the
polarisability of a single oscillator one can pass to the susceptibility (2) of the
medium by multiplying the polarisability by the density N of the particles.
Thus, the model of an anharmonic oscillator predicts a dramatic increase
in the quadratic polarisability (by a factor of Q 0 /2) in the vicinity of
resonances, where one of its three arguments is close to the eigenfrequency 0 .

In this case, becomes a complex value. From the comparison of (6.72) and
(6.73), it follows that scales as the product of the linear susceptibilities at the
corresponding frequencies,
(3 = 2 + 1 ) (3 )(2 )(1 ). (6.75)
Note that the quadratic susceptibilities of many dielectric crystals in the trans-
parency range (between the lattice and electron eigenfrequencies) satisfy the rela-
tion
(2) n(3 )n(2 )n(1 ), (6.76)
with the same scaling factor for different crystals (Millers empirical rule).
Our model does not take into account the difference between the actual (lo-
cal) field Eloc and the macroscopic field E, which is averaged over the atomic
inhomogeneities. According to Lorentz, in a cubic crystal Eloc /E = ( + 2)/3.
(This correction is only valid for dielectrics, while in metals and semiconductors,
Eloc = E.) Polarization P of the medium, caused by a given external polarization
Pext , is also ( +2)/3 times as large as Pext . As a result, the quadratic susceptibility
(2) , defined with respect to macroscopic parameters, is
(3 ) + 2 (2 ) + 2 (1 ) + 2 (2)
(2) = loc , (6.77)
3 3 3
where (2) loc is defined for local parameters. In non-cubic crystals, this correction is
a tensor. Equations (6.75)(6.77) indicate that there is a close connection between
the linear and nonlinear properties of a medium.
Let a displacement x = a0 correspond to the case where the linear part of the
force in (6.70), eE0 = m20 a0 , is equal to the nonlinear one, ma20 . (Here, E0 is a
typical field keeping the charge near the equilibrium position.) Then, = 20 /a0 ,
and for , n 0 we have an estimate
E NL / m40 /e m20 a0 /e = E0 . (6.78)
Thus, the ratio of the quadratic polarization and the linear one is on the order of
E1 /E0 (compare with (6.42)). For a hydrogen atom, a0 should be understood as
Bohrs radius, ~2 /me2 , and 0 , as the ionization edge frequency, e2 /2~a0. Then,
E0 = e/2a20 107 G.
The cubic potential (6.69), according to (6.72), creates only even harmonics,
2, 4, 8, . . . . For the formation of odd harmonics, it is necessary that the poten-
tial has a term x4 . Note that even at E = 0, the potential (6.69) is not centrally
symmetric: it changes its sign under the coordinate inversion, x x. It is useful
to consider a three-dimensional model using the potential [Akhmanov (1964)]
V(r) = m20 x x /2 mx x x . (6.79)
6.2.5 Raman anharmonicity

Spontaneous Raman effect, or spontaneous Raman scattering (SpRS)f was dis-
covered by Raman and, independently, by Mandelshtam and Landsberg, in 1928,
much ahead of the advent of lasers. The corresponding stimulated effect (StRS)
was first observed only in 1962.
In SpRS, monochromatic light (the pump) incident on the matter polarizes
molecules with an optical frequency 1 . As a result, the molecules acquire the
dipole moment
d(t) = 1 E1 cos 1 t, (6.80)
where, for simplicity, we assume the linear polarisability of a molecule 1

(1 ) to be a real scalar. Emission of molecular dipoles (6.80) leads to the
Rayleigh scattering.
Let us now take into account the intra-molecular oscillations of the nuclei with
the eigenfrequency 0 1 , which are excited due to collisions. Oscillations
of the nuclei near the equilibrium positions, Q(t), slowly modulate the electron
cloud surrounding them. In this case, all electronic parameters of the molecule
are modulated as well, including its optical polarisability, (t) = (/Q)Q(t).
This picture is based on the so-called adiabatic approximation, which implies that
the electron eigenfrequencies 0 much exceed 0 (usually, 0 /2c 105 cm1 ,
0 /2c . 103 cm1 ). With the oscillations of the nuclei taken into account, (6.80)
takes the form of an amplitude-modulated oscillation,
!

d(t) = 1 + Q0 cos 0 t E1 cos 1 t, (6.81)
Q
where Q0 is the amplitude of the nuclei oscillations.
As a result, the radiation scattered by the dipoles contains, in addition to the
carrier (Rayleighs) frequency 1 , two sidebands: the Stokes one, 1 0 , and
the anti-Stokes one, 1 + 0 . In the case of a multi-atom molecule, the Raman
spectrum contains its normal oscillations (some oscillations do not influence
due to symmetry). With an account for the anisotropy of and the rotations of
molecules, the induced dipole moment d(t) will be also modulated by the typical
rotational frequencies.
This modulation approach is based on a single nonlinear parameter, /Q,
introduced by Placzek, and provides an explicit description of the sponta-
neous effect. (Here, the term spontaneous relates to the field components with
f Editors note: in the original text, the Russian term is used, which is spontaneous combination
scattering.
the frequencies 1 0 , which are absent in the incident field and appear
spontaneously.)
In order to describe stimulated nonlinear effects caused by the parametric cou-
pling between nuclei and electrons, it is useful to consider the model of two cou-
pled oscillators, with the potential
V(x, Q) = m20 x2 /2 + M20 Q2 /2 x2 Q eEx. (6.82)
Here, x and 0 are the coordinate and the eigenfrequency of the electron, Q and
0 are the corresponding values for the nuclei, is the coupling parameter scaling
as /Q. It follows from (6.82) that
x + 2 x + 20 x = eE/m + 2Qx/m, (6.83)
Q + 2Q + 20 Q = x2 /M. (6.84)
This model was proposed for the description of StRS by Platonenko and Khokhlov
in 1964. According to (6.84), the force acting on the the nuclei scales as the square
of the electron displacement; therefore, the nuclei oscillations will be enhanced in
the case where the difference between the two field frequencies is close to 0 . Let
the incident field be a biharmonic one, and 1 2 0 .
In the linear approximation in the field, Q(1) = 0, and
x(1) 2 2
n = eE n /mDn , Dn 0 n 2in , n = 1, 2. (6.85)
In the expression for the force, x2 , we only leave the resonance terms with
the frequency scaling as x(1) (1)
1 x2 , then the amplitude of stimulated molecular
oscillations with the frequency is
x(1)
1 x2
(1)
(/M)(e/m)2
Q = = E1 E2 . (6.86)
2M(20 2 2i) 2D0 D1 D2
Thus, an optical biharmonic field with a proper frequency difference excites

the intra-molecular nuclei oscillations through the electron shell. These oscil-
lations, coherent with the incident light, add to the equilibrium thermal ones and
cause additional incoherent scattering at the anti-Stokes frequency, 3 = 1 + =
21 2 , and at the second Stokes frequency, 4 2 = 22 1 . In ad-
dition, cubic polarization is induced at the initial field frequencies, leading to the
amplification of the field E2 with the lower frequency and attenuation of the field
E1 with the higher frequency. It is namely the effect of Raman amplification that
causes stimulated Raman scattering.
By substituting (6.85) and (6.86) into (6.83), (6.84), we find

x(1) Q 2 e3 /2Mm4
x(3)
1 = 2 = |E2 |2 E1 ,
mD1 D0 D21 D2 D2
x(1)
1 Q

2 e3 /2Mm4
x(3)
2 = = |E1 |2 E2 , (6.87)
mD2 D0 D1 D1 D22
x(1)
1 Q 2 e3 /2Mm4 2
x(3)
3 =
= E E .
mD3 D0 D21 D2 D3 1 2
After multiplying x(3)

n by eN, we find the cubic polarisability. As a rule, in
experiment the frequencies of the field n (n = 1, 2, 3) are much less than the elec-
tron transition frequencies 0 ; therefore, Dn 20 (non-resonance RS). Within
this approximation, (6.87) leads to
(3) (1 = 1 2 + 2 ) = (3) (2 = 2 1 + 1 )
= (3) (3 = 21 2 )
= C/[20 (1 2 )2 2i(1 2 )], (6.88)
where C 2 e4 N/Mm4 80
These nonlinear susceptibilities describe, respectively, Raman absorption
(since (3) (1 )00 > 0 at 1 > 2 ), Raman amplification ((3) (2 )00 < 0)
and coherent anti-stokes Raman scattering (CARS) with the intensity scaling as
|(3) (3 )|2 I1 I2 . Besides, it follows from (6.88) that in the presence of a monochro-
matic wave with a fixed frequency L within the transparency range, another
monochromatic wave, with a variable frequency , will have resonance disper-
sion in two regions, L 0 (Fig. 6.3). The width 2 of these resonances is
determined by the decay rate of molecular oscillations, and this Raman (induced)
dispersion is anomalous in the Stokes range.
Let us find the relation between the nonlinear parameters and /Q. We
substitute Q = Q0 cos 0 t into Eq. (6.83). In the first order in , it leads to the
relation x2 = eE1 Q0 /m2 D1 D2 . Comparing it with (6.81), we get
2e2 2e2
= 2 2 4 . (6.89)
Q m D1 D2 m 0
By means of the two-oscillator model, one can also describe the spontaneous
Raman scattering. To this end, the right-hand side of (6.84) should be replaced by
the stochastic Langevin force f (t) causing thermal (and quantum) fluctuations of
Q(t). This force is delta-correlated, and one can find its spectral density by setting
Fig. 6.3 Raman susceptibility. Due to the effect of the pump with the frequency , the susceptibility
of matter acquires additional resonances at frequencies L 0 , where 0 are the eigenfrequencies of
the molecules. It is important that the Stokes resonance has negative losses (bottom left) and anoma-
lous dispersion of the refractive index (top left).
equality between the fluctuation energy Q and the equilibrium energy of the os-
cillator. Another method of describing SpRS is based on the Raman analogue of
the FDT (Sec. 7.7), stating that the polarization fluctuations of matter at frequency
are determined by the imaginary part of the cubic susceptibility [Fain (1972);
Klyshko (1980)],
hP ()P(0 )i = (~/)( 0 )N()(3) ( = L + L )00 |E L |2 , (6.90)
where
N() [exp(~/T ) 1]1 = N() 1, L . (6.91)
Here, at < 0 (the anti-Stokes range), N has the meaning of the equilibrium
number of phonons N0 , while at > 0 (the Stokes range), N = (N0 + 1). In
the last expression, the unity describes the quantum fluctuations of the nucleus
coordinate Q, which cause the Stokes scattering even at T = 0, when the anti-
Stokes scattering is absent. The elementary process corresponding to the Stokes
scattering is a two-photon one. It includes the annihilation of a pump photon and
the creation of a Stokes photon and a phonon (Fig. 6.1(g)).
Certainly, the two-oscillator model, similarly to the anharmonic-oscillator one,
is only qualitative. A quantitative calculation of the susceptibility, even the linear
one, is very complicated and requires the knowledge of the wave functions and
eigenfrequencies of the system (see below).
Inelastic scattering can be also due to the excitation of other degrees of free-
dom of the medium, for instance, electronic ones. In this case, the radiation fre-
quency changes by a quantity equal to the frequency of some electron transition
in an atom or a molecule, 1 2 = mn (Em En )/~. If the incident field
contains two frequencies such that 1 + 2 = mn > 0 and the molecule is at the
ground level n, then two photons can be absorbed simultaneously. In the inverse
process, the excited molecule emits two photons, spontaneously or via stimulated
emission.
In macroscopic matter, light interacts not only with local internal oscillations
of the particles, but also with the collective excitations, for instance, with acous-
tic, temperature, spin, plasma waves, and with the oscillations of the molecules
orientations.
The equilibrium chaotic part of these waves modulate the refractive index (see
(6.66)), and the corresponding sidebands 2 = 1 appear in the spectrum of
the scattered light. The scattering can be viewed as resulting from the diffraction
of the incident light by a traveling grating formed by waves of pressure, tempera-
ture, and so on. From the quantum viewpoint, a photon ~1 of the incident light
(the pump) gets scattered with a simultaneous birth or annihilation of a matter
excitation quantum ~ (a phonon, a magnon, a plasmon, an exciton, a polariton
etc.). For the scattering by propagating excitations, it is typical that the modulation
frequency depends on the observation direction, i.e., on the angle of scattering,
= (). This dependence follows from the diffraction Braggs condition or,
in other words, the phase-matching condition (the momentum conservation law),
k1 k2 q = 0, and the dispersion dependence for the scattering wave, q = q(),
where q is the wavevector of the matter excitation. The effect of phase matching
on the Raman anharmonicity can be formally taken into account by assuming the
cubic susceptibility to depend not only on the frequencies but also on the wave
vectors (spatial dispersion).
For describing stimulated scattering by acoustic waves (MBS, Mandelshtam-
Brillouin scattering) and other collective excitations, one should take into account
the non-equilibrium coherent part of these excitations caused by the biharmonic
pump. It is clear from (6.64) how sound can be excited by light: due to elec-
trostriction, there appears a source of alternating pressure in the matter, with the
difference frequency ,
2
X
(1)

p(r, t) = (1/2) (En /2) exp(in t) + c.c.

n
= (1/8)(1) E1 E2 exp(it) + . . . (6.92)

This source creates density waves , propagating with the speed of sound
v. If the pump waves are plane ones, then exp[i(k1 k2 ) r], and the
stimulated sound wave will have the maximal intensity at |k1 k2 | = q = /v.
This phase-matching condition chooses from the continuous spectrum of acous-
tic excitations, spanning a range from zero to approximately 1011 Hz, one (or
two, considering the difference between v for transverse and longitudinal waves
in amorphous solids) discrete component, with the frequency
= v|k1 k2 | 2vk sin(/2). (6.93)
The width of this spectral component is determined by the sound absorption coef-
ficient.
6.2.6 Temperature anharmonicity

It follows from (6.93) that at , 0, light scattering by acoustic waves, i.e., waves
of pressure, p, and density, , is inelastic, v , 0. (To be precise, the maxi-
mum of the scattering corresponds to , 0.) According to (6.93), the scattering
can be centered around = 0 only if it is due to non-propagating excitations, for
which v = 0 or which decay sufficiently fast. Such scattering can be caused by
temperature fluctuations, T , or entropy fluctuations, S T , as well as by con-
centration fluctuations, C, in mixtures and solutions. The values x p, T, C, . . .
(or , S , . . . ) are thermodynamic parameters describing the macroscopic state of a
medium, and their oscillations, both thermal (spontaneous) and stimulated (co-
herent) ones, break the optical homogeneity of the medium (n = (n/x)x)
and lead to the scattering of light, spontaneous or stimulated. All these types
of scattering occur with a frequency shift that is small compared with the case
of scattering by molecular oscillations and are called the Rayleigh, or molecular,
scattering [Fabelinsky (1965)]. The last term emphasizes the difference from scat-
tering by macroscopic inhomogeneities, such as dust particles and other objects.
The integral intensity of spontaneous scattering by the parameter x scales as
the mean square x2 , and it can be calculated thermodynamically. At the same
time, the spectral distribution od the intensity is determined by the kinetic equa-
tions describing the evolution of the field x(r, t). For instance, for x = T the
kinetics is given by the diffusion equation,
T a2 T = P/c p , (6.94)
where a is the temperature conductance, P is the power of external heat sources

per unit volume, c p is the specific heat capacity at constant pressure. (Strictly
speaking, one should take into account not only T but also the simultaneous
variation of pressure, p, due to heating; here, for simplicity, we ignore the rela-
tion between temperature and pressure waves.)
At P = 0, (6.94) describes spontaneous temperature scattering, and its solution
can be represented as a sum of plane waves exponentially decaying in time,
X
T (r, t) = T q eiqrt , (6.95)
q
with aq2 . Similarly to the case of scattering by acoustic waves, the scattering
of light with the wavevector k1 in the direction k2 is caused by the temperature
grating with q = (k1 k2 ). However, according to (6.95), this grating is not
moving, and diffraction by it results in an elastic (unshifted) line in the spectrum
of the scattered light, with the bandwidth
= 2 = 8ak2 sin2 (/2). (6.96)
In liquids, 108 s1 at = 90 . The inverse linewidth of the temperature

(entropy) scattering, T 1/ = o2 /a, has the meaning of the typical time of
temperature diffusion by a distance of one wavelength o |k1 k2 |1 , i.e., the
relaxation time of the temperature grating.
The mechanism of the stimulated temperature scattering (StTR) and the corre-
sponding anharmonicity is evident in the presence of some absorption (StTR-2).
Indeed, in the case of a biharmonic field, the external force in (6.94) has a variable
component,
P(r, t) = 00 E 2 (r, t) = 00 ReE1 E2 ei(qrt) , (6.97)
which creates a temperature wave,
T = ReT ei(qrt) . (6.98)
Its amplitude can be found by substituting (6.97) and (6.98) into (6.94),
00 /c p
T = E1 E2 . (6.99)
i
Here, 1 2 || and 00 00 ().
The traveling coherent wave of temperature (6.98) modulates the susceptibility
(mainly due to the density decrease caused by temperature expansion),g
!
T
= T , (6.100)
T p T
g In an ideal gas, N = p/T , therefore, = p/T and, if the dependence of on T is neglected,
(/T ) p = /T .
therefore, the wave of the susceptibility has the amplitude

00 /c p
!
= E1 E2 . (6.101)
i T p
As a result, cubic polarization emerges, with the frequencies 1 , 2 , 3
21 2 , 4 22 2 ,
P(3) (t) = (t)E(t)
= Re (E2 ei1 t + E1 ei2 t + E1 ei3 t + E2 ei4 t )/2. (6.102)
Let E1 , E2 be parallel to the x axis, then, from the definition of the cubic suscep-
tibility we find that
00
!

121
xxxx
2
= 2121
xxxx = 3112
xxxx = 4221
xxxx = . (6.103)
2c p [aq2 + i(2 1 )] T p
Note that StRS-2 has an interesting feature: here, the anti-Stokes components
are amplified (compare with (6.88)), i.e., the energy of the field is transferred,
in the course of propagation, from the low-frequency components into the high-

frequency ones, since ( T ) p < 0.
In order to estimate the temperature anharmonicity, assume that
n2
!
5 1
= , c p = N, 00 , aq2 , (6.104)
T p T 2 4E T
where E = n/c is the relaxation time of the field. Then, at |1 2 |T 1,
2 (1) 40UE
E NL , (6.105)
(3) 3n2 T
where U = 3T N/2 is the internal energy density. Thus, if T = E , which
at E = 108 s corresponds to = 0.003 cm1 , then the nonlinear polarization
becomes equal to the linear one when the field energy n2 E12 /8 equals the thermal
energy of the matter. Note that if one passes from the temperature conductance a
2
to the heat conductance = ac p , then E NL can be represented as 8q2 T E .
6.2.7 Electrocaloric anharmonicity

Temperature anharmonicity can take place even in a completely transparent ma-
terial, due to the optical electrocaloric effect. The corresponding stimulated scat-
tering is called StTS-1.
Consider the simplest model describing the effect of the electric field on the
temperature of a non-absorbing material. After the field (continuous or alternat-
ing) is switched on, the energy levels of the molecules get shifted due to the Stark
effect, and their populations do not correspond any more to the temperature of
the thermostat. (The role of the thermostat is usually played by the translational
and rotational degrees of freedom of the molecules, or vibrations of the crystal
lattice.) During the relaxation time T 1 , populations are re-distributed, which is
accompanied by the change of the thermostat energy.h As a result, the thermostat
temperature is changed. Note that an analogous magnetocalorical effect is used
for obtaining ultra-low temperatures (adiabatic demagnetization). A more rigid
explanation follows from the definition of the temperature for a closed system
(microcanonical ensemble),
1/T S /U (ln g)/U, (6.106)
where S is the entropy, U the internal energy and g(U) the density of energy states.
The latter depends on the configuration of the levels and hence changes after the
field is turned on.
Let us estimate the contribution of the electrocaloric effect in the temperature
anharmonicity. According to (4.35), when a dielectric is polarized, its thermody-
namic potentials get an increase v = |E1 |2 /4 per unit volume. Let us choose T
and as independent parameters, then the entropy is determined in terms of the
temperature derivative of the Gibbs potential (T, p); therefore, the variation of
S due to the polarization is
! !
1
S = = |E1 |2 . (6.107)
T p 4 T p
After multiplying S by T , we obtain the increase of the heat Q, and after mul-
tiplying it by T/c p , we find the temperature increase,
!
T
T = |E1 |2 . (6.108)
4c p T p
Comparing this with (6.94), we see that the role of the absorbed power in a
transparent material is played by
T d|E1 |2
!
dQ
Pequiv = = . (6.109)
dt 4 T p dt
In the case of a biharmonic field, we obtain

!
T
Pequiv (r, t) = Im E1 E2 ei(qrt) . (6.110)
2 T p
h Here, the finite heat capacity of the external degrees of freedom is taken into account.
Comparing (6.110) and (6.97), we find the ratio of the electrocaloric anharmonic-
ity and the anharmonicity (6.103) due to the dissipation,
(3)
! !
ec T 2T
= . (6.111)
(3) 200 T p
ab
cn T p
Hence, it follows that the absorption coefficient equivalent to the electrocaloric

effect is not high,

2T
equiv 104 cm1 , (6.112)
cn T p
where we assumed = 108 s1 , T = 300 K, (/T ) p = 104 K1 .
Finally, note that due to the relation between the density and tempera-
ture waves, electrostriction also contributes into the temperature anharmonicity
(see [Apanasevich (1977)]).
6.2.8 Orientation anharmonicity

As we have already mentioned, rotation of anisotropic molecules in a gas also
modulates the scattered light, which leads, due to the quantization of the rota-
tional motion, to the appearance of discrete sidebands near the Rayleigh (non-
shifted) and Raman lines in the spectrum of spontaneous scattering. However, if
the density of the particles is high, a molecule cannot make a full rotation during
the orientation relaxation time ; therefore, in liquids the rotational lines overlap
and the Rayleigh line acquires a broad pedestal spanning tens of inverse centime-
ters, the so-called Rayleigh wing. Light scattering by orientation fluctuations of
molecules is also called anisotropic (depolarized) scattering. In the framework of
the macroscopic description, one can assume that anisotropic scattering is caused
by the symmetry breaking of the medium (which is otherwise isotropic), i.e., the
scattering is due to the symmetry fluctuations. A schematic shape of the scattered
spectrum, with an account for the matter excitations considered above, is shown
in Fig. 6.4.
Interaction of light with the orientation motion of molecules is another source
of optical anharmonicity. This type of anharmonicity manifests itself in the Kerr
effect, discovered as early as in the 19th century, in which n E02 , the optical
Kerr effect and self-focusing, with n |E1 |2 , and in stimulated Rayleigh-wing
scattering.
Let us estimate the contribution of the orientation anharmonicity in the cubic
susceptibility. Consider first a non-polar molecule. In a field E(t) it acquires
induced dipole moment d(t) () E(t) (we neglect the absorption) and the
Fig. 6.4 Main types of scattering, the corresponding nonlinearities and typical frequencies: 1, tem-
perature (entropy) scattering by temperature fluctuations is related to the electrocaloric anharmonicity
( 108 s1 ); 2, Mandelshtam-Brillouin scattering ( 1010 s1 ) by pressure fluctuations is
related to striction anharmonicity; 3, the Rayleigh wing is caused by anisotropy fluctuations and is re-
lated to the orientation anharmonicity ( 1011 s1 ); 4, Raman scattering by the internal vibrations
of molecules ( 1014 s1 ). The first three types are called molecular, or Rayleigh, scattering.
time-averaged energy (see (4.35))
V = ReE() E(+) . (6.113)
Hence, an anisotropic molecule tends to turn with respect to the field to max-
imize its polarisability. However, in equilibrium matter, rotation of molecules by
the field is hindered by their interaction with the neighbors: relaxation processes
restore the equilibrium state with chaotic orientations of the molecules, which the
field tends to order. Competition between the field and the thermal motion results
in a dynamic equilibrium, with the degree of orientation being on the order of
V/T . In this case, the liquid becomes birefringent, similarly to a uniaxial crystal
with the axis parallel to E (if the field is linearly polarized). This phenomenon is
called the optical Kerr effect.
Let the polarisability anisotropy be (). For instance, for a linear molecule,
= || . Then the susceptibility variation (2 ) at a frequency 2 induced
by the field at a frequency 1 will be equal, on the order of magnitude, to (2 )N
times the degree of orientation,
(1 )(2 )N|E1 |2 /T. (6.114)
Thus, the cubic susceptibility for non-polar molecules can be estimated as
(3) (1 )(2 )N/T. (6.115)
Assuming that the anisotropy is high, = (1) /N, we get

2
E NL (1) /(3) T/. (6.116)
Let a30 1024 cm3 and T = 300 K, then E NL = 2 105 G, and if (1) = 0.1,
then (3) = 1012 cm3 /erg (compare with (6.42), (6.43), (6.68)).
If a molecule has a constant dipole moment d0 and the orienting field is a static
one or it varies slowly compared with the orientation relaxation time ( 1012 s),
then the effective energy is V = d0 E. (The induced moment can be neglected
in this case.) The degree of orientation will then scale as (V/T )2 since linear
electrooptic effect is forbidden in liquids (Sec. 6.1). As a result,
(3) (, 0, 0) ()N(d0 /T )2 1010 cm3 /erg, (6.117)
E NL T/d0 104 G. (6.118)
(Here, we assumed d0 = 1 D.) Thus, the Kerr effect in polar liquids is much
stronger than in non-polar ones.
Suppose now that the orienting field is an optical biharmonic one, with
1 2 < 1/, then V and, correspondingly, the degree of orientation
V/T will contain an alternating component with the frequency . As a result,
the polarization at the probe field frequency 3 (which, in principle, can coin-
cide with 1 or 2 ) will be modulated, i.e., the matter will emit coherent field
with the frequencies 3 . Thus, orientation anharmonicity provides a reso-
nance contribution with the width 2/ to the cubic susceptibility (3) (3 , 2 , 1 )
at (1 2 ) 1/. If the orientation relaxation is taken into account, this con-
tribution becomes complex. Its imaginary part corresponds to the amplification
of the 2 field, with the maximum at 2 = 1 1/, and leads to the stimulated
Rayleigh-wing scattering.
Orientation anharmonicity can be quantitatively described in terms of the
molecular orientation distribution function, which is stationary at 1:
!2
V/T
V 1 V
P() = Ce = C 1
+ . . . , (6.119)
T 2 T
where C is a normalization factor (depending, of course, on the temperature and

the field), the energy V() is defined in (6.113) and is a set of three angles
determining the orientation of the molecule in a laboratory frame of reference (the
Euler angles).
In the case of non-polar molecules, it is sufficient to do the expansion up to
the linear in 1/T term,
h i
C = P(0) (1 + hVi(0) /T ), P() = P(0) 1 (V hVi(0) )/T , (6.120)
where
Z Z
hVi(0) d3 P(0) V, P(0) 1/ d3 = 1/82. (6.121)
The angle-averaged induced dipole moment has the form

hdi = h()i E + h()i : EE + . . . (6.122)
The angular brackets denote averaging with the perturbed distribution function
P(). If we neglect the intra-molecular anharmonicity ( = 0), then hdi = E,
where
Z
hi d3 P()()
= (0) (h()V()i(0) (0) hVi(0) )/T, (6.123)

Z
(0)
d3 P(0) () = ( xx + yy + zz )/3. (6.124)
Here, (0) is the linear polarisability averaged over the equilibrium distribution
function. Hence, we find the correction to polarisability,
Z
(0) = d3 [() (0) ] Re E() () E(+) /82 T. (6.125)
Here, () is the linear polarisability tensor in the laboratory frame of reference

for a molecule with a given orientation .
At & 1 it is necessary to take into account the variation of the distribution
function in time, P = P(, t), which is described by the kinetic equation [Apanase-
vich (1977)] or a Focker-Planck-type equation. Within the exponential-relaxation
approximation, these equations yield a usual dispersion dependence with a pole at
1 = 2 ,
(3) (2 = 2 1 + 1 ) 1/[1 + i(1 2 )]. (6.126)
(3)00
Hence, the low-frequency (Stokes) component of the field is amplified, < 0,
and this amplification is maximal at || = 1/.

6.2.9 Quantum theory of nonlinear polarization
Nonlinear polarisabilities , , . . . of molecules and the susceptibility of matter
(2) N, (3) N, . . . can be calculated similarly to the linear polarisability
(Sec. 4.2), using the density-matrix equation with the phenomenological damping
constants. However, it is easier to use the general formula (3.75) for the response
h f (t)i of a quantum system to an external perturbation with the energy V(t); the
damping should be added from general considerations at the final stage of the
calculation.
We are interested in the stationary response of the system to a periodic pertur-
bation; therefore, the lower integration limits in (3.75) should be chosen as .
The upper limits can be chosen as + provided that the causality is taken into
account by introducing the step functions (t t1 ), . . . , (tk1 tk ) into the inte-
grand. Let us introduce the damping, which is necessary for achieving a stationary
regime and practically inevitable in any system (see Fig. 6.5),
(t) = et (t > 0), (t) = 0 (t < 0), (6.127)
where is some positive constant, which will be further replaced by mn .

Assuming in (3.75) f = d and V = d E, for the induced dipole moment of
order k we obtain (see also Ref. [Fain (1972)])
Z Z
hd (t)i(k) = (i/~)k dt1 . . . dtk (t t1 ) . . . (tk1 tk )
Tr {[. . . [d0 (t), d0 1 (t1 )], . . . , d0 k (tk )]}E1 (t1 ) . . . Ek (tk ). (6.128)
Here, is the equilibrium density operator and the operators d0 (t) are considered
in the interaction picture. The integrand in (6.128) depends on k + 1 time argu-
ments, only k of them being independent, as one can easily verify. This tensor
function is called the response function of the system or its Greens function. One
also says that it is a causal Greens function and, at k > 1, a multi-time one.
Let the field have a discrete spectrum,
X
E(t) = E p exp(i p t)/2, p = 1, 2, . . . , (6.129)
p
then the integration in (6.128) is elementary. For instance, for k = 1 we find

Z t
i
hd (t)i(1) = dt1 E p exp[i p t1 + (t1 t)]
2~
h i
() () () ()
nn dnm dmn exp(inm t + imn t1 ) dnm dmn exp(inm t1 + imn t)
() () () ()
nn dnm dmn dnm dmn
= E p exp(i p t). (6.130)
2~ mn p i nm p i
Here, summation over the state indices m, n and over the Cartesian indices , =
x, y, z (which are sometimes written as superscripts, to make the notation more
compact), as well as over the frequency index p, is implied. If the damping is
taken into account by changing to the damping rate mn of the density matrix
non-diagonal element, then the linear susceptibility (1)
defined by (6.130) will
coincide with (4.57).
Similarly, at k = 2, (6.128) yields

Z t Z t1
1
hd i(2) = 2 dt1 dt2 E p Eq exp[i( p t1 + q t2 ) + (t2 t)]
4~
h
() () ()
i
nn dnm dml dln exp[(inm t + iml t1 + iln t2 )] + . . .
() () ()
nn dnm dml dln
= 2
+ . . .
4~ (mn p q i)(ln q i)
E p Eq exp[i( p + q )t]. (6.131)

This expression contains only the contribution of the first term of the double com-
mutator [[d(t), d(t1 )], d(t2 )], since the other three differ only in signs and in the
permutations of the state indices l, m, n.
Consider sum-frequency generation, 1 + 2 = 0 . At 1 , 2 , the double
sum over frequencies contains two terms oscillating with the frequency 0 : the
one with q = 1, p = 2 and the one with q = 2, p = 1. Therefore, the 0 component
of the dipole moment can be represented as a sum of two terms differing by a
permutation of indices 1, and 2, ,
21
() () ()
(2)
Y nn dnm dml dln
d0 = E2 E1 2
(0) (1) + . . . , (6.132)

2~ Dmn Dln
where
D(p)
mn mn p imn ,
Q
is the operator of summing over various permutations, and has been replaced
by mn . Note that the dispersion function 1/D is the Fourier transform of the step
function (t) (Fig. 6.5).
Fig. 6.5 The Heaviside step function multiplied by et and its Fourier transform are used for taking
into account the effects of causality and damping on the response of a system.
Thus, the quadratic susceptibility of matter consisting of N similarly oriented

non-interacting molecules can be written in terms of the unperturbed level popu-
lations Nn = nn N, frequencies mn , and the transition moments dmn as
21 () () () () () ()
012 1 Y X dnm dml dln dnm dml dln
= 2 Nn (0) (1) +
2~ lmn Dmn Dln Dnl D(1)
(0)
nm
() () () () () ()
dnm dml dln dnm dml dln
, (6.133)
D(0) (1)
lm Dln
(0) (1)
Dlm Dnm
where the superscripts 0 and 0 relate to the frequencies 0 .
It is easy to see that expression (6.133) has all symmetry features described in
Q
Sec. 6.1. The operator provides the spatio-frequency symmetry with respect to
the last two indices (6.14), (6.15). The property (6.16) follows from the relations
(p) ( p)
Dmn = Dnm and dmn = dnm , so that changing the signs of all frequencies and the
imaginary units only interchanges the terms in (6.133): the first one is swapped
with the second one and the third one, with the fourth one.
The fact that in centrally symmetric media (2) = 0 also follows from (6.133).
The eigenstates of a system with a center of symmetry possess a certain parity:
n (r) = n (r); therefore, dmn = 0 if m and n have the same parity. Hence, at
least one of the three moments relating the states l, m, n is equal to zero.
The symmetry property in the case of transparent matter, (6.21), also follows
from (6.133) at mn = 0. In this case, the last terms in (6.133) can be combined in
pairs,
1 1 1
+ = .
D(0) (1)
lm Dln D(0) (2)
lm Dnm D(1) (2)
ln Dnm
As a result, out of 8 terms only 6 = 3! are left, which differ by permutations

of index pairs (, 0), (, 2), (, 1),
021 () () ()
1 X Y dnm dml dln
021
= 2 Nn . (6.134)
2~ lmn
(mn 0 )(ln 1 )
Susceptibility of the next order is calculated similarly. In each new order of

the perturbation theory, factors of the form dln /~D(p)
ln are added. Hence, we get an
estimate for the optical anharmonicity in the transparency range,
(k) ~0
E NL (k+1)
E0 , (6.135)
ea0
where a0 , 0 , E0 are the typical size, frequency, and internal field of the molecule.
For the hydrogen atom, E0 = 13.6V/a0 107 G. Assuming (1) = 0.1, we obtain
(2) = 108 G1 , (3) = 1015 G2 . Even this crude estimate gives a correct idea
about the orders of magnitude of the susceptibilities. At resonance, (k) increases.

6.2.10 Probability of multi-photon transitions
If we are only interested in energies, then the nonlinear interaction between ra-
diation and matter can be described in terms of probabilities or cross sections of
multi-photon transitions, as it was done in Chapter 2 for the case of linear single-
photon interaction. Then, as a rule, the field does not have to be quantized, i.e., one
can use the semi-classical theory, but the results of calculations can be interpreted
in the photon language.
As an example, let us find the probability of a two-photon elementary process
describing Raman scattering and two-photon (induced) absorption or emission.
Substituting the transition amplitude in the first-order approximation (2.24) into
(2.21), for the amplitude of the two-photon transition from level a to level b we
obtain
Z t Z t2
c(2)
ba (t) = (i~) 2
dt V 0
(t
2 bn 2 ) 0
dt1 Vna (t1 ). (6.136)
t0 0
Here, the n index, in which summation is implied, numerates all intermediate (vir-
tual) non-perturbed states through which the transition can occur. This expression
reflects an important feature of quantum dynamics: all possible virtual states par-
ticipate in a transition, even those seemingly violating the conservation laws, with
the only restriction, following from the causality principle: t > t2 > t1 > t0 .
Let us substitute into (6.136) the dipole perturbation V0 = d0 E and the
biharmonic field (6.129), assuming that t0 , t = ,
X
c(2)
ba = (2~)
2
dbn Eq dna E p
nqp
Z Z t2
dt2 dt1 exp[i(bn q )t2 + i(na p )t1 ]

X dbn Eq dna E p Z
= dt2 exp[i(ba q p )t2 ]. (6.137)
nqp
4~2 i(na p )
The lower integration limit in the integral over t1 makes no contribution due to the
adiabatic start of the perturbation or due to the damping (see (6.127) atR +0).
The integral left in (6.137) is one of the delta-function representations: dt2 =
2(ba q p ). Thus, in the second order of the perturbation theory, the
field excites a molecule only provided that the algebraic sum of the two field
frequencies coincides with the frequency of the transition a b. This condition

generalizes Bohrs postulate for a single-photon resonance.
Let two field frequencies 1 and 2 satisfy the condition of the combination
resonance, 1 + 2 ba , then (6.137) yields
c(2) 12
ba = i(2/~)(E1 M E2 )(ba 1 2 ),
() () () ()
12 21 1 X dbn dna dbn dna
M = M + . (6.138)
4~ n na 2 na 1

If ba > 0, i.e., the initial state of the molecule is the lower one, then
(6.138) gives the probability amplitude for the molecule to absorb two photons
(at 1 , 2 > 0) or to absorb one photon (1 > 0) and emit another photon, the
Stokes one (2 < 0). In the last case, (6.138) is the amplitude of a Stokes Raman
process. Similarly, at ba < 0, (6.138) describes two-photon emission or anti-
Stokes scattering. One can say that the two terms in (6.138) differ in the sequence
the 1 and 2 photons are absorbed (or emitted).
Note that there are four types of two-photon emission: stimu-
lated, spontaneous-stimulated, stimulated-spontaneous and purely spontaneous
(Fig. 6.1(f)). They correspond to the four terms in the expression (N1 +1)(N2 +1) =
N1 N2 + N1 + N2 + 1, which follows from (6.138) after replacing E p with operators.
(Here, N p are the initial photon numbers.)
It follows from (6.138) that the maximal contribution to the transition ampli-
tude is provided by virtual states with the minimal energy deficit ~(na p ).
Note that the various transition paths (contributions of various virtual states)
may differ in signs and cancel each other (quantum interference of states).
In order to find the transition probability from (6.138), it is necessary to define
the square of the delta function:
Z T/2
|()|2 = () lim dteit /2 = ()T/2. (6.139)
T T/2
From (6.138), (6.139), we obtain the probability of a two-photon transition per

unit time, i.e., the transition rate (compare with the derivation of (2.36)),
(2) (2) 2
Wba = 2|Kba | (ba 2 1 )/~2 , (6.140)
(2)
Kba E1 M12 E2 . (6.141)
According to (6.140), the two-photon transition a b is possible if the virtual

transitions a n and n b are allowed. In a centrally symmetric medium, the a
and b levels should have the same parity; in this case, the single-photon transition
between them will be forbidden (the alternative prohibition). Thus, two-photon

spectroscopy and, in particular, Raman spectroscopy enable the study of levels
that are not accessible for linear spectroscopy.
For taking into account the finite width of the resonance, one should re-
place () by the normalized form factor g() (see (2.37)). The resulting ex-
pression coincides with the probability of a single-photon transition (2.37), with
the only exception that the Rabi frequency |dba E1 |/~ is replaced by
(2)
2Kba . Probabilities of k-photon transitions have a similar structure: each addi-
(k)
tional photon p participating in the transition adds to Kba a factor of the form
dmn E p /~(na p 1 ) /0 E p /E NL (compare with (6.135)).
As in the case of single-photon transitions (Sec. 2.3), the transition rate W (2)
determines the cross-section (2) and the absorption (amplification) coefficient
(2) , as well as the imaginary part of the cubic susceptibility (3) (2 , 1 , 1 ).
The cross section of a stimulated two-photon transition will be defined as the
transition rate in the case of unity densities of the photon fluxes,
W (2) Ip c
(2)
stim , Fp = |E p |2 . (6.142)
F1 F2 ~ p 8~ p
For taking into account the finite transition linewidth, we replace the delta
function in (6.140) by the normalized form factor g (Sec. 2.2),
(2) 3 12 2
stim = 128 k1 k2 |e1 M e2 | g(ba 2 1 ), (6.143)
where kn n /c.
The two-photon absorption coefficient for the field with the frequency 2 in
the presence of the second field, with the frequency 1 , is (compare with (1.4))
(2) (2 ) = (2)
stim NF 1 , (6.144)
where N Na Nb is the population difference per unit volume.
The power absorbed from the field 2 by a unit volume of matter is
P2 = 2 ImP(3) 2121
2 E2 /2 = 2 Im E 2 E 1 E 2 E 1 /2. (6.145)
On the other hand, P2 = (2) I2 . Comparing (6.145) and (6.144), for the case of a
real transition dipole moment dmn we find that
Im2
121
= 4~1N M
21 21
M g(ba 1 2 ). (6.146)
Finally, the complete expression for the cubic nonlinearity follows from (6.146)
after replacing g by 1/(ba 1 2 i). It is not difficult to verify that a
straightforward calculation of (3) from (6.128) in a single-resonance approxima-
tion yields the same result.
Equations (6.140)(6.143) describe purely stimulated two-photon transitions.

In the case of spontaneous-stimulated transitions (leading to the spontaneous Ra-
man scattering), the role of |E2 |2 is played by the quantum fluctuations of the
field, which provide an equivalent photon flux density of F2vac = c/L3 in each
field mode in the vicinity of the frequency 2 , with L3 being the quantization vol-
ume (Sec. 7.3). After summing over all modes in the vicinity of the frequency
ba 1 , we find
X Z Z X
(2)
sp.st W (2)
sp.st /F 1 = (c/L3
) (2)
stim = d2 d2 k22 (2)
stim /(2)
3
k2 4 2
Z X
d2 (d(2)
2 /d) sp.st , (6.147)
4 2
where 2 is the polarization index. After substituting here (6.143), we obtain the
differential cross-section per unit solid angle and a single polarization type of the
field 2 ,
(d(2) 3 12 2
2 /d) sp.st = 16k1 k2 |e1 M e2 | . (6.148)
Further, substituting F1 by c/L3 and summing over all modes 1 of the field,
from 0 up to the transition frequency ab , we find the rate of a purely spontaneous
transition,
X Z ba Z X
(2) 3 (2) 3
W sp = (c/L ) sp.st = (2/ ) d1 d1 d2 k13 k23 |e1 M12 e2 |2 .
k1 0 4 1 2
(6.149)
For a rough estimate of the two-photon decay probability in the optical range,
assume that all frequencies in (6.149) are equal to 0 = e2 /2~a0 = c/2a0 , and
the polarisability M of the transition is equal to a30 , with = 1/137. Neglecting
all other numerical factors, we obtain
(2)
W sp 0 (a/2)6 50 s1 .
An accurate calculation for the 2s 1s in a hydrogen atom yields 8 s1 .
(2) (1)
Thus, W sp is 1373 106 times as small as W sp . Despite the small probability,
two-photon decay can be easily observed using two PMTs and a coincidence cir-
cuit. Note that spontaneous (as well as thermal) two-photon emission, in contrast
to the single-photon one, has a continuous spectrum, which has no relation to the
discrete spectrum of an unperturbed atom. For the emission from a heated body,
the statistics of two-photon radiation also differs from the one of single-photon
radiation, which is caused by the fact that photons are emitted in pairs. Thus,
the anharmonicity of matter leads, in principle, to the deviation of the thermal
radiation statistics from a Gaussian one [Klyshko (1980)].
6.2.11 Conclusions
Thus, a large number of mechanisms contributes to the optical anharmonicity of
macroscopic matter. Quadratic susceptibility (2) , as a rule, is related to the non-
linearity of bonded electrons. It differs from zero only in piezoelectric crystals
and is on the order of 107 109 G1 provided that all frequencies belong to the
optical transparency window. The cubic susceptibility (3) in condensed transpar-
ent matter is also caused by the electronic nonlinearity ((3) 1015 G2 ) if all
frequencies belong to the optical range. However, if the difference of two fre-
quencies coincides with the frequency of a molecular vibration, then (3) becomes
as high as 1012 1013 due to the combined electron-nuclear (Placzek or Raman)
nonlinearity. At 0, the main contribution in the case of solids is that of elec-
trostriction ((3) 1013 ); in liquids, the orientation (Kerrs) nonlinearity is added
((3) 1012 ). Electrocaloric anharmonicity, usually, results in (3) . 1013 . Ex-
tremely strong nonlinear optical effects can be observed in liquid crystals and in
plasma. Note that the nonlinear electrodynamics of plasma is well described by
the Landau-Vlasov kinetic equations (see [Silin (1961)]).
6.3 Macroscopic nonlinear optics
Thus, using classical or quantum macroscopic models, we have found polarization

P of matter in a given field E, i.e., we have excluded the variables of the medium.
By substituting D = E + 4P into macroscopic Maxwells equations, we obtain
a closed system of equations for E, H describing the emission and propagation of
electromagnetic field in matter with an account for the matter nonlinearity. Mani-
festations of nonlinearity in the optical range are extremely diverse and depend on
the properties of both the medium and the initial field, such as the amplitude, the
spatial and temporal spectra. The most important parameter is of course the ratio
E/E NL , which, as a rule, is much less than the unity.
6.3.1 Initial relations

Below, we consider the main types of stationary effects. The field can be repre-
sented as a sum of independent spectral components,
X
E(r, t) = (1/2) En (r) exp(in t),
n (6.150)
En = En , n = n , n = 1, 2, . . . .
The amplitudes of monochromatic waves En (r) satisfy the system of

Helmholtz wave equations that are related to each other due to the anharmonicity
of the medium, which is considered to be non-magnetic,
c2 En 2n n En = 42n PnNL (r). (6.151)
These relations can be easily obtained from Maxwells equations (4.9)(4.12).
Here, the linear part of the polarization is included into n . The amplitude of the
nonlinear polarization at the frequency n , which is the source of the macro-
scopic field En1 , is defined through the nonlinear polarisability (Sec. 6.1),

.
(m1) (n1 ; n2 , . . . , nm )..En2 (r) . . . Enm (r),
X
NL
Pn1
(r) = (6.152)
m=3
where
m
X
ni = 0. (6.153)
i=1
For describing non-coherent (noise) fields, such as thermal or Raman radiation,

one should add Langevin random sources in the right-hand side of (6.151). Ac-
cording to the fluctuation-dissipation theorem (FDT), their spectral density scales
as (1)00 or (3)00 , respectively (see (6.90)). Another method of calculating the noise
radiation from a nonlinear medium is based on Kirchhoff-type equations, which
directly express the noise intensity in terms of the medium temperature and the so-
lutions to the dynamic equations (6.150) (consider, for instance, (6.225), (6.307),
(7.6)).
6.3.2 Classification of nonlinear effects

It follows from Maxwells equations (see (4.13)) that the specific power of radia-
tion absorbed from the field by the matter at point r due to anharmonicity within
a single period is
X
.
n1 (m1) ..En1 . . . Enm . (6.154)
X X
PNL (r) = (i/4) n En Pn = (i/4)
n m=3 n1
(m)
Each term P (r) in the last sum describes m-frequency interaction. If the field
amplitudes are represented in the form
X
En (r) x |En (r)| exp[in (r)], (6.155)

where = x, y, z, then both the sign and the absolute value of P(m) will depend
on (m) (r) n1 1 + + nm m , i.e, on the relation between the field spectral
components. Such interactions are called parametric. It is clear from (6.154) that
the parametric interaction of m harmonics provides an effect that is accumulated
in space only under the condition (m) (r) = const, which is possible only for
plane waves, such that n (r) = kn r, kn kn . Then, the wave vectors of the
interacting waves should form a closed polygon,
m
X
k(m) kni = 0. (6.156)
i=1
This condition, with an account for (6.153), is called the spatial phase-
matching condition.
Often, it is only a single low-order term in (6.154) that is important (m = 3
or 4, i.e., three- or four-frequency interaction), with a certain combination of fre-
quency signs sn . This combination can be put into correspondence to an elemen-
tary multi-photon process involving m photons; then (6.153) and (6.156) can be
P
interpreted as the conservation laws of the field energy, ~n , and momentum,
P
~kn . In a parametric process, the energy of matter does not change, i.e., the
initial level Ea and the final level Eb coincide, ba = 0 (Fig. 6.1(a)).
Further, one can consider resonance parametric processes, for which one or
more virtual levels coincides with the real ones. Then, non-parametric effects
become also important (see below), i.e., there appears linear (or multi-photon) ab-
sorption (or emission), , 0, and the energy of the field is not conserved. How-
ever, provided that the absorption is low, the phase-matching condition (6.156)
is still valid. Note that, for instance, at m = 4, single, double, and triple reso-
nances are possible. The efficiency of resonance interactions strongly depends on
the frequencies n (even without taking phase matching into account), and the
corresponding susceptibilities take complex values.
Outside of resonances, 0, susceptibilities are real, and their dispersion
is weak; therefore, the efficiency of non-resonance parametric processes has only
indirect frequency dependence, through the phase-matching condition, which at
fixed directions of kn is satisfied only for a particular set of frequencies and at
fixed frequencies, only for a certain set of directions. Therefore, the efficiency of
parametric interactions manifests joint frequency-angular dispersion.
Let us return to the sum (6.154). Among the terms of even orders m, there
are degenerate ones, containing subscript pairs ni = n j , for which (m) 0. The
corresponding interactions, depending only on the intensities of the field harmon-
ics, |En |2 , are called non-parametric. The phase-matching condition is satisfied
for them automatically. The power absorbed at the frequency 1 due to the non-
parametric interaction with an l-frequency field, according to (6.154), is

.
P1 (r) = (1/2)1Im(2l1) (1 ; 1 l , l )..E1 E1 . . . El El . (6.157)
It has a constant sign for a given set of frequencies and slowly (compared to eikr )
varies in space for any configuration of the field, since the phases of the harmonics
do not influence Eq. (6.157).
In non-parametric elementary processes, such as, for instance, two-photon ab-
sorption or a Raman transition, the final energy of the matter differs from the
initial one (Fig. 6.1), and therefore, the field energy is not preserved.i For a non-
parametric process, in addition to the trivial (in this case) condition (6.153) there
is a resonance at
l
X
n = ba . (6.158)
n=1
In the presence of additional intermediate resonances, a non-parametric pro-

cess is called a cascaded or resonance one. Such processes have complicated
dispersion dependence on each field frequency n separately. Examples are reso-
nance Raman scattering and cascaded two-photon absorption in a three-level sys-
tem (Fig. 6.1(e)).
In the case of non-resonance (non-cascaded) processes, (1) = 0, and the ef-
ficiency of non-parametric interaction depends, according to (6.158), only on the
P
sum of all field frequencies, n . This case can be called a single-resonance one.
Note that in non-parametric processes, new spectral components appear only
due to spontaneous or spontaneous-stimulated transitions, and in the absence of
a feedback the radiation is noisy, non-coherent, even if the pump is coherent and
has a fixed phase. An example is inelastic scattering of light. At the same time,
parametric processes can lead to the generation of coherent fields with new fre-
quencies (generation of harmonics).
The effects of nonlinear optics can be additionally classified according to the
number of essential spectral components or according to the number of plane
waves (modes). For instance, single-frequency non-parametric (degenerate) ef-
fects include nonlinear absorption and dispersion (l1 = ba ) and saturation
(1 1 + + 1 = ba ). Examples of two-frequency non-parametric effects
are induced absorption and dispersion (1 + 2 = ba ) and Raman interaction
(1 2 = ba ), leading to the stimulated Raman scattering (StRs) and the inverse
Raman effect, i.e., induced absorption at the anti-Stokes frequency.
i Note, however, that induced dispersion, accompanying multi-photon absorption, similarly to linear
dispersion, corresponds to a virtual process with ba = 0.

The second harmonic generation is a degenerate case of three-frequency

parametric interaction (1 2 3 = 0). Four-frequency parametric effects
include third-harmonic generation and coherent anti-Stokes Raman scattering
(CARS), which forms the basis of active spectroscopy. CARS is determined
by the resonance part of the cubic susceptibility (3) (4 = 3 2 + 1 ) at
1 2 = 4 3 ba . An example of a parametric single-frequency four-
wave interaction is optical phase conjugation due to (3) ( = + ) at
k2 = k1 , k4 = k3 .
All these effects will be considered in more detail below. Sometimes, several
effects can be manifested simultaneously; for instance, StRS can be accompanied
by self-focusing and CARS. However, often, by choosing the experimental condi-
tions one can select a single effect. In our analysis, for simplicity we will assume
that this is the case.
It is also convenient to distinguish between the effects according to other pair-
wise features: spontaneous stimulated, stationary non-stationary (Chapter 5).
It should be stressed that the terms spontaneous and stimulated have no
rigid definitions in quantum electronics. In linear optics, spontaneous and stim-
ulated emission is considered to be a non-stationary process involving a single
molecule, while the joint stationary radiation from heated matter is called thermal
radiation. At the same time, the term spontaneous scattering is understood as
scattering by thermal (or, at ~/T 1, quantum) fluctuations of various param-
eters of the matter (Sec. 6.2) at small pump intensity I1 , although this process is
explained in terms of spontaneous-stimulated two-photon transitions (Sec. 6.2). If
I1 is increased, the intensity of the scattered light grows first linearly and then, at
|(2) |l > 1, exponentially (for the Stokes component). As a result, the efficiency
of frequency conversion can be as high as tens of per cent, and it is to this sponta-
neously emitted and amplified radiation that the term StRS is usually applied. The
same term is sometimes used for the effect of external Stokes field amplification.
The term coherent is even more ambiguous, even if we restrict ourselves to
nonlinear optics. It is used, on the one hand, for the effects of non-stationary
resonance interaction, like SIP (Sec. 5.1), and on the other hand, for parametric
stationary effects like second harmonic generation. In the first case, it means that
the field and the matter have the same phase and in the second case, that different
components of the field have the same phase.
6.3.3 The role of linear and nonlinear dispersion

The nonlinearity of the wave equations of hydrodynamics and gas dynamics is
most apparent in the appearance of shock waves, i.e., in the transformation of sine
acoustic waves into sawtooth ones. (A well-known example is wave breaking in

shallow water.) In spectral language, this effect is explained by higher-harmonic
generation, which enriches the spectrum of an excitation in the course of its prop-
agation.
However, due to the refractive-index dispersion n(), light shock waves do
not emerge: harmonics with the frequencies 1 , 21 , . . . propagate with differ-
ent phase velocities c/n(p1 ), so that the sign of the interaction energy and,
correspondingly, the amplitudes of the harmonics have fast spatial oscillations
(see (6.154)). As a result, the amplitudes of higher harmonics do not in-
crease in the course of propagation (non-accumulating interaction), and an-
other small parameter appears in the theory, the ratio of the coherence length
lcoh = /k to the length of the medium l. Here, k is the wave mismatch
(6.156), which is equal, in the case of collinear pth-harmonic generation, to
k p pk1 = [n(p1 ) n(1 )]p1 /c.
It is only under special conditions that the phase velocities of two or three har-
monics can be matched, using birefringence or anomalous dispersion. Thus, the
phase matching condition (6.156) restricts the number of efficiently interacting
spectral components of the field in parametric interactions. For non-parametric
processes, the restriction is due to the resonance condition (6.158), i.e., the dis-
persion of (3) , (5) , . . . .
The efficiency of nonlinear processes is usually increased by using pulsed
pumping with Q-switched ( 10 ns) or mode locked ( 10 ps)j lasers. Clearly,
for the interaction of several light pulses with different central frequencies to ac-
cumulate in space, it is also necessary that their group velocities coincide,
u p = d/dk p = c(n p + p dn p /d)1
(the group matching condition); otherwise, the pulses will separate in the course
of propagation.

6.3.4 One-dimensional approximation
Further, as a rule, we will apply the one-dimensional approximation, which is
most useful in nonlinear (as well as linear) optics. Also known as plane-waves
approximation, it allows one to pass from partial-derivatives equations (6.151) to
usual equations. It reflects correctly the main features of many effects involving
pump beams not too much convergent or divergent and samples with the length l
not too large. The divergence can be often taken into account in the final formulas,
j Editors note: At present, mode locking provides pulse durations as small as tens of femtoseconds.
at least qualitatively, by summing up the contributions from all essential plane

waves. We will go beyond this approximation only in the description of self-
focusing in the end of Sec. 6.4.
An arbitrary field can be represented as a four-dimensional Fourier integral,
Z
E(r, t) = (L/2)3 d3 kdE(k, ) exp[i(k r t)], (6.159)
and similarly for H. The factor (L/2)3 is added from dimensionality considera-
tions, see (7.100). This is the so-called k, -representation. Here, k and are in-
dependent real variables. In the case of a homogeneous transparent linear medium
without internal sources, k and are not independent any more (Sec. 4.2). Then,
XX
E(k, ) = (1/2) e (k)E(s) (sk)( s (k)), (6.160)
=1,2 s=
E(+) (k) = E() (k) . (6.161)

We assume the medium to be non-gyrotropic; therefore, and e are real. Substi-
tuting (6.160) into (6.159), we find
1 L 3
Z X
E(r, t) = d3 k e (k)E(s) (k) exp[i s (k)], (6.162)
2 2 s
where
s (k) s[k r (k)t], s = , = 1, 2.
For instance, in the case of a plane monochromatic polarized wave,
E(s) (k) = (E1 s+ + E1 s ) (3) (k k1 ).
Unit polarization vectors e (k) and the dispersion law (k) are determined by the
() tensor,
c2 k [k e ] + 2 ( ) e = 0. (6.163)
Hence, if is real,
(k) = (k), e (k) = e (k) = e (k).
The wave vector k and the type of polarization define a plane monochromatic
wave, or a mode. It is convenient to make the set of modes countable using a cube
with periodicity L3 (Sec. 7.3); then the Fourier integral turns into a series, in which
a mode is labeled by a single subscript k {k, , s} (which also includes the sign
s of the frequency),
X
E(r, t) = (1/2) ek Ek exp(ik ), Ek E(s) (k). (6.164)
k
In a transparent linear medium without sources, the mode amplitudes Ek are inde-
pendent complex numbers defined by the boundary conditions.
Consider now a nonlinear absorbing plane layer (0 < z < l) in the medium,
which has the same 0 , and a stationary field incident on it from both sides. Then,
since the model has no dependence on x, y, t, the amplitudes Ek are only func-
tions of z. They are called the slowly varying amplitudes (SVA). Actually, the
transition from the real field E(r, t) to the z-dependent mode amplitudes Ek (z) is a
three-dimensional Fourier transform in the variables x, y, t (a non-complete trans-
formation, unlike (6.159)) or the transition to the , k , z-representation.
Note that a mode can be defined by fixing, instead of k, the frequency,
= (k), k , and the sign k of the longitudinal component kz . Then, kz is deter-
mined by the dispersion law (for an isotropic medium, kz = k ( 0 2 /c2 k2 )1/2 ).
In experiment, mode expansion is achieved by placing a spectral device into the
far-field zone of the emitter; in this case, the spherical angles k , k of k are also
fixed (taking into account refraction). Thus, the subscript k stands for one of the
following sets of values:
k {k, , s} = {, k , , , s} = {, , , , s}. (6.165)
As we will show in what follows, in the case of sufficiently small nonlinearity and
absorption, Fourier transformation of the Helmholtz equations (6.151) w.r.t. x, y
leads to a system of ordinary equations for SVA (for k > 0),
!
d k 2ik NL
+ Ek (z) = P (z) exp(ikz z). (6.166)
dz 2 cnk k
Here, the following notation was introduced:
k k ek 00 ek /cnk , nk nk cos k cos k ckz /k , (6.167)
where k is the angle between the ray (Poynting) vector sk of the mode and the z
axis, k is the anisotropy angle, i.e., the angle between k and sk , and PkNL is the
Fourier component of the nonlinear polarization with the frequency k > 0 and
the transverse wave vector k {k x , ky }, which is parallel to the ek vector. It can
be written as
PkNL (4|ukz|/L)PNL (k , q, z)
Z
= (2|ukz|/L3 ) dx dy dt ek PNL (r, t) exp[i(k t k x x ky y)], (6.168)
q {k x , ky }, r {x, y, z}, ukz (k)/kz. (6.169)

According to (6.166), linear absorbtion results in an exponential dependence
of the mode amplitude on z, while nonlinearity leads to a coupling (mixing) of
modes. Let us stress that in a one-dimensional model, a mode k can be only ex-
cited by a component of the source field having the same frequency and transverse
wavevector. The source field can have any dependence on z but the most efficient
component is the one that has the propagation constant Kz close to kz , due to the
phase matching condition (compare with (4.23)).
The boundary conditions for system (6.166) are given by the amplitudes Ek (z0 )
of the plane waves incident from the left and from the right: z0 = 0 for = + and
z0 = l for = . The solution to (6.166) yields the rule of mode transformation
by the nonlinear layer, i.e., the scattering matrix of the layer. Certainly, for de-
scribing real experiments the model should include reflection and refraction at the
boundaries (with an account for the nonlinearity, see Ref. [Bloembergen (1965)]),
as well as consider the limited interaction cross-section in x and y. In this case,
even in the linear approximation one gets the diffraction coupling between the
modes having close directions. This effect is more conveniently described in the
, r representation (6.151) (see the self-focusing section in 6.3).
Let us find, with the help of (6.166), the variation rate of the energy flux lon-
gitudinal component carried by the mode k, in ~k units:
Fkz Ik cos k /~k = c[Ek Hk ] z/16~k + c.c.

= cnk |Ek (z)|2 /8~k , (6.170)
where z is a unit vector along the z axis. For calculating Fkz , we multiply (6.166)
by Ek and sum with the complex conjugated expression,
(d/dz + k )Fkz = ImPkNL Ek exp(ikz z)/2~. (6.171)
The right-hand side has, according to (6.154), a simple meaning: this is the energy
(in ~k units) absorbed by a unit volume of the matter per unit time due to the
nonlinear polarization, i.e., this is the rate of the photon density decrease, Nk , in
mode k. In other words, (6.171) is the transfer equation (or continuity equation)
for the photons of the macroscopic field,
Fk + Nk = 0, Fk = uk Nk . (6.172)
The nonlinear polarization PkNL in the right-hand side of (6.166) or (6.171) is

given by equations (6.152) and (6.158) through the hierarchy of susceptibilities
(m) , m 1 2. In the case of an m-frequency interaction, equation (6.166) takes the
form
.
(d/dz + k /2)E1 = (2s1 1 /icn1 )(m1) ..e1 E2 . . . Em exp(ikz ), (6.173)
where
m
X
k si kiz , Ei ei E(si i ) (ki ), (6.174)
i=1
and integration w.r.t. x, y, t in (6.168) yields the following restrictions for the
interacting modes and the sign indices:
m
X
k si ki = 0, (6.175)
i=1
m
X
si i = 0. (6.176)
i=1
In addition, accumulation of the interaction along z requires that the longitudinal

wave mismatches are sufficiently small,
m
X
k si kiz . 1/l. (6.177)
i=1
Let us re-normalize the mode amplitudes so that their squares are equal
to the longitudinal photon flux density. (To simplify the notation, we assume
Fi Fki z > 0.)
ai (cni /8~i )1/2 Ei , (6.178)
|ai |2 = Fi . (6.179)
As a result, (6.171) and (6.173) take the form

!
d 1 s1
+ a = 1...m a2 . . . am exp(ikz), (6.180)
dz 2 1 2i
!
d
+ 1 F1 = s1 Im1...m a2 . . . am exp(ikz), (6.181)
dz
where
!m/2 !1/2
1 8~ 1 . . . m .
1...m (m1) ..e1 . . . em . (6.182)
2~ c n1 . . . nm
These equations will be analyzed for several typical cases in Secs. 6.4, 6.5.
6.3.5 The Manley-Rowe relation and the permutation symmetry

Suppose that only a single interaction between m-frequency field components
n1 , . . . , nm (see (6.154)) is essential, corresponding to an l-photon process. (For a
parametric interaction, l = m, while for a non-parametric one, l = m/2.) Let this
interaction be non-resonance for some pair of frequencies 1 , 2 . To be more ex-
act, let the photons with these frequencies be created or annihilated only together,
simultaneously, which is the case if the corresponding virtual level is far from any
real one. Then,
s1 N1 = s2 N2 , (6.183)
where
sn sign(n) sign(n ). (6.184)
These relations are similar to the Manley-Rowe equations used in the oscil-
lation theory for describing nonlinear circuits [Migulin (1978)]. The signs of the
frequencies are determined by the energy conservation law (6.153) or (6.158), Nn
is the rate of the flux variation for photons with the frequency |n |. Apparently,
~|n |Nn + Pn = 0, where Pn = n ImEn Pn /2 is the power absorbed by matter at
frequency n .
From (6.154) and (6.183), it follows for the case of a parametric interaction
that
. .
Im(12... ..E1 E2 21... ..E2 E1 . . . ) = 0, (6.185)
where
12... (1 ; 2 , . . . ) = (12... ) . (6.186)
Let us substitute (6.155) into (6.185),
X
|E1 E2 . . . |Im{(12... 21...
... ... ) exp[i(1 + 2 + . . . )]} = 0. (6.187)
...
Here, all amplitudes and phases of the field harmonics can be varied indepen-
dently; therefore, the expression in round brackets is identically equal to zero,
12... 21...
... = ... . (6.188)
This permutation relation is more general than (6.21) where the medium was as-
sumed to be transparent at all frequencies.
In the case of a non-parametric interaction, the phases are irrelevant, and we
obtain from (6.157) and (6.183) that
. .
Im(1122.....E1 E1 E2 E2 2211... ..E2 E2 E1 E1 . . . ) = 0, (6.189)
X
|E1 E1 E2 E2 . . . | Im(1122... 2211...
... ... ) = 0. (6.190)
...
Hence, we obtain the invariance to block permutations,
1122... 2211...
... = ... . (6.191)
We have already observed such symmetry for Raman transitions and, in gen-
eral, two-photon transitions (Sec. 6.2). Assuming in (6.191) 1 L > 0, 2
S < 0, we obtain, with the help of (6.16),
LLS S
S S LL
(L = L + S S ) = (S = S + L L ) . (6.192)
Strictly speaking, (6.192) leads to the symmetry of only the imaginary parts
of the susceptibilities; however, using relations of Kramers-Kronig type (4.8), one
can show that the real and imaginary parts of (m) have the same symmetry. Re-
lation (6.191) also follows from (6.188), if one considers a resonance parametric
process with the frequencies 3 = 1 , 4 = 2 :
1212... 2112...
... = ... . (6.193)
By performing complex conjugation and using partial symmetry (6.14), (6.15),

and (6.16), we obtain (6.191).
Permutation symmetry of susceptibilities, or the Manley-Rowe relations, lead
to several simple integrals of motion for SVA. For instance, in the case of non-
resonance interaction, the coupling coefficients (6.182) are invariant to the per-
mutations of all indices; therefore, the right-hand sides of Eqs. (6.181) for the
intensities differ only in signs. Hence, it follows for k = 0 that
s1 dF1 /dz = s2 dF2 /dz = = sm dFm /dz = Im(1...m a1 . . . am eikz ), (6.194)
where the sign factors si = 1 in the case kiz > 0 are determined by the en-
ergy conservation in the corresponding elementary process. Hence, for systems
(6.180), (6.181) we find m 1 integrals of motion, which allow the intensities in
m 1 modes to be expressed in terms of the intensity in a single mode,
Fk (z) = s1 sk F1 (z) + Ck , C1 = 0, (6.195)
and the constants Ck are determined by the boundary values Fk (0).

If all essential frequencies, including the combination ones, are in the trans-
parency windows (i.e., parametric processes are non-resonance), then there is no
dissipation of the field energy and there must exist one more integral of motion
corresponding to the conservation of the total light energy flux. However, one can
easily see that the total energy flux is expressed in terms of Ck constants. To this
end, let us multiply relations (6.195) by ~k and add them,
m
X m
X
Iz (z) = ~k Fk (z) = ~k Ck . (6.196)
k=1 k=1
In parametric interactions, there is yet another integral, which defines the

phase difference
m
X
(z) sk k (z) (6.197)
k=1

of the complex amplitudes ak Fk exp(isk k ) in terms of Fk (z) and (0),
sn Fn k + 2(F1 Fm )1/2 cos[(z) + kz] = C. (6.198)
Here, n is any of the subscripts 1, . . . , m. By differentiating the left-hand part of
(6.198), one can see, due to (6.180), (6.194), and the relation = , that it is
independent of z.

6.3.6 Derivation of one-dimensional equations
In order to derive equations (6.166) for SVA in the , q, z-representation, let us
write the frequency components of the field En (r),
E(, r) = (1/2)[En (r)( n ) + En (r)( + n )],
which enter the Helmholtz equations (6.151), in terms of the mode amplitudes
Ek (z) = 4|ukz|E(, q, z)/L defined in (6.164). Let > 0 and the polarization type
be fixed, then
Z
dt it
Z L 3
2E(, r) = e d3 k Ek (z) exp(ik r i(k)t)
2 2
L 3 Z
= d3 kEk (z) exp(ik r)( (k))
2
L 3 Z X
= d2 q |ukz |1 Ek (z) exp[iq ikz (, q)z], (6.199)
2
where q {k x , ky }, {x, y}, ukz (k)/kz . The function kz (, q) kz > 0

is defined here indirectly, through the dispersion law, (q, kz ) = . In deriving
(6.199), we have used the property of the delta function
X
[ f (x)] = ai (x xi ), f (xi ) 0, ai |d f /dx|1
x=xi . (6.200)
i
Further, assume that only a single mode with a certain sign of kz is excited effi-
ciently, then
L3
Z
E(, r) = d2 q|ukz|1 Ek (z) exp(ik r),
163 (6.201)
k {q x , qy , kz (, q)}.
While substituting (6.201) into (6.151), let us take into account that usually
both absorption and nonlinearity are low. Then, Ek (z) vary very little at a distance
of a wavelength, and the second derivatives can be neglected (for more detail on
the SVA method, see [Akhmanov (1964)]),
|d2 Ek /dz2 | |kz dEk /dz |. (6.202)
As a result,
eE(z)eikr eikr {k (k e) + ik ( e) + i (k e)}E(z),

(6.203)
where the k index is so far omitted. The first term here will be further canceled
due to (6.163). Let us take the inner products of (6.151) and (6.203) with e and
use the vector identity
e {k ( e) + (k e)} = 2{e (e k)} , (6.204)
then (6.151) takes the form
iL3
!Z
d2 q|ukz|1 eikr {2c2 [e (e k)] 2k e 00 e}Ek (z)
163
= 42k e PNL (k , r). (6.205)
Here, the vector e (e k) Ek Hk is parallel to the ray vector sk , i.e., to

the direction of the energy flux at 00 = PNL = 0, and = zd/dz. Let denote
the angle between sk and z and be the angle between sk and k, which is usually
small; then,
[e (e k)] z = k cos cos nk /c kz . (6.206)

R
Equation (6.166) is obtained by acting on (6.205) by the operator d2 exp(iq0
).
6.4 Non-parametric interactions
Non-parametric (non-coherent) effects of nonlinear optics, i.e., effects like multi-

photon absorption, are described by odd-order susceptibilities of the form
(2m1) (1 = 1 m + m ). The rate of the mode k1 amplitude varia-
tion scales then as the local amplitude E1 (z) of the same mode, there is no phase
matching factor exp(ikz), and one-dimensional SVA equations (6.166) can be
represented in the form
!
d 1 2i1 1 2 2
+ |E2 | |Em | E1 = 0, (6.207)
dz 2 cn1
.
1 (2m1) (1 = 1 m + m )..e1 e1 . . . em em . (6.208)
Thus, the real part of 1 determines the local variation of the propagation con-
stant (i.e., of the wavelength or the phase velocity) of mode k1 , which scales as the
local intensities of other modes (induced dispersion effect), while the imaginary
part of 1 determines additional absorption or amplification due to the energy of
other modes (nonlinear or induced absorption effect).
Multiplying (6.207) by E1 or directly using (6.181), we find the system of
equations for the intensities (Fi Fiz ),
[d/dz + 1 + (m)
1 (z)]F 1 (z) = 0, (6.209)
where
!m
1 8~ 1 m 00
(m) (m) (m)
1 1 F 2 F m , 1 .
2~ c n1 nm 1
Recall that the signs of Fi and ni coincide and are determined by the sign of the ki
projection onto the z axis; therefore, F1 can increase in the course of propagation
only if 001 < 0.
Below, we will consider the basic types of non-parametric effects: nonlinear
absorption (including the saturation effect), Raman amplification and absorption,
spontaneous and stimulated scattering, self-focusing and self-modulation, as well
as their role in optics applications and in spectroscopy.
6.4.1 Nonlinear absorption

Consider the single-mode case. The imaginary part of the cubic susceptibility,
.
(3)00 Im(3) ( = + )..e ee e,
Fig. 6.6 Variation of the intensity of light in the course of propagation and the limitation effect under
two-photon absorption: F is the photon flux density times l, where is the two-photon absorption
coefficient, and l is the thickness of the layer.
leads to the violation of the exponential Bougers law for the intensity variation in
matter. This effect can be related to two-photon absorption (see (6.146)), in which
(3)00 > 0, and to the saturation of a single-photon resonance, where, according to
(4.99),
(3)00 = (1)00 dab
2
T 1 T 2 /~2 [1 + (0 )2 T 22 ].
Let a plane monochromatic wave be incident on a transparent (in the linear ap-
proximation) medium orthogonally to the boundary. Then, according to (6.209),
dF/dz + F 2 = 0, 322~2 (3)00 /c2 n2 > 0. (6.210)
The solution to this equation can be easily found,
F(z) = F(0)/[1 + zF(0)] 1/z. (6.211)
The last equation takes place at F(0) 1/z and describes the limitation effect
(Fig. 6.6), in which the intensity F(z) of the transmitted light does not depend on
the incident light intensity F(0).
If F(0) has fluctuations (in particular, due to the photon structure of the field),
the two-photon absorption will smear them, i.e., at the output of the layer photons
will be distributed in time and space more uniformly than at the input (the anti-
bunching effect, see Sec. 7.6).
For two-photon inter-band transitions in semiconductors like CdS, (3)00
1013 cm3 /erg at 0 0.7 and n 2. Then, /~ 1 cm/GW, i.e., the limi-
tation level at l = 1 cm for I ~F is on the order of 1 GW/cm2 . Note that at
such intensities, the structure of the matter can be changed, there can be a phase
transition and plasma can be created. Such optical breakdown in a transparent gas
or condensed matter can start from multi-photon absorption or ionization.
Ionization of an inert-gas atom requires the absorption of about 10 photons
from the neodimium laser (~ 1 eV). According to (6.209), the number of such
transitions per unit volume and time is
N = dF/dz = (8/c)10(19)00 I 10 /2~ (E1 /E NL )20 , (6.212)
with E NL ~0 /d0 . Although this value is extremely small, the effect can be ob-
served by focusing a Q-switched laser [Delone (1978)]. Note that the high power
of the field in (6.212) increases the effect of its fluctuations. Indeed, according to
(7.41), radiation with Gaussian statistics is hI 10 i/hIi10 = 10! 3 106 times as
efficient as a non-fluctuating radiation with the same hIi.
As it was shown by Keldysh (see Ref. [Elyutin (1982)], Sec. 29), at very large
field values E1 or small frequencies , so that the condition d0 E1 < ~ is violated,
the dramatic power dependence of (6.212) on E1 gets slower due to the factor
exp(~0 /d0 E1 ), which is typical for the effect of tunnel ionization in a constant
field.
Power dependence of the form (6.212) can be also violated due to the dis-
sociation of multi-atom molecules in the field of IR-lasers under a single-photon
resonance with one of the vibrational frequencies of the molecule, v . In the
case of a CO2 laser ( 10 , ~ 0.1 eV), dissociation requires several tens
of photons. However, experiments with short pulses, where collision relaxation is
too slow to be revealed, manifest a weak dependence on E1 . This effect is proba-
bly caused by the fast transfer of the absorbed energy to a large number of other
vibrations (intra-molecular relaxation). Such effects are studied by a new field
in quantum electronics, IR photochemistry, and used for the laser separation of
isotopes.
In photochrome materials [Barachevsky (1977)], the limitation (darkening) ef-
fect is observed at much lower intensities, even in sunlight, where I 0.1 W/cm2 ,
due to the photochemical anharmonicity, which is very strong but inertial.
The lowest nonlinearity threshold is observed for the photographic process,
where the photochemical reaction (creation of blackening centers, silver particles
in AgBr microcrystallites) sometimes requires an energy density (exposure) as
low as It 108 J/cm2 . It is noteworthy that for the creation of a single center,
several photons are required (m > 1), i.e., this is a highly nonlinear process. At
small exposures, the degree of blackening D of a developed negative scales as the
number of centers; therefore, D = I m , where is the film sensitivity, and we
ignore the statistical variance of m. The relative intensity decrease for the probe
light after passing through a negative with a small thickness l is
I/I = D = (m+1) l, (6.213)
where (m+1) I m /l.

Various types of photochemical nonlinearity are used for holography and, in
general, for information recording. More fast optical nonlinearities form the
basis of dynamical holography and the optical phase conjugation effect (Sec. 6.5)
where recording and reading are simultaneous.
Thus, even a material that is transparent in the usual sense becomes absorbing
or reflecting at sufficiently high intensities. The study of such effects can have
very important applications in connection with the problem of laser thermonuclear
synthesis.
Saturation of a single-photon resonance (Sec. 4.3), on the contrary, leads to the
bleaching of matter (in the absence of the population inversion). In this case, the
intensity fluctuations become more pronounced, which is used for mode locking
in picosecond lasers. According to the two-level model, the absorption coefficient
(F) (or, at < 0, the amplification coefficient) at large intensity F has the form
0 /(1 + F/F s ), with the parameter F s determined by the relaxation times T 1 , T 2
and the transition moment dab (see (4.96)). Hence, (6.210) is replaced by the
nonlinear transfer equation for a plane wave in the presence of saturation (i.e., at
a single-photon resonance),
dF 0 F
+ = 0. (6.214)
dz 1 + F/F s
Note that here, in contrast to (6.210), contributions of an infinite number of odd-

order susceptibilities are taken into account. The solution to (6.214) can be easily
obtained in an indirect form,
0 z = ln(F/F0 ) + (F F0 )/F s , (6.215)
where F F(z), F0 F(0). The two terms in the right-hand side of (6.215) cor-
respond to the exponential and linear F(z) dependencies, respectively (Fig. 6.7).
At strong saturation, the exponential Bougers law becomes linear,
!
0 z
F 1 F0 , (6.216)
1 + F0 /F s
and in the limit F0 F s the matter is completely bleached (compare with the
nonstationary SIT effect described in Sec. 5.1).
Fig. 6.7 Intensity F of a wave versus the distance z in an amplifying or absorbing medium in the
presence of saturation. At strong saturation, the exponential variation of F is replaced by a linear one
(the dashed line), 0 is the amplification coefficient for a weak signal, F s is the signal intensity at
which 0 is reduced twice.
6.4.2 Doppler-free spectroscopy

Induced absorption due to a two-photon transition and induced bleaching due to
the saturation of a single-photon transition form the basis for two interesting spec-
troscopic methods that allow one to overcome Dopplers broadening of resonances
masking the fine details of the spectra.
The scheme of a two-photon Doppler-free spectroscope is shown in Fig. 6.8.
The gas under study is placed into a standing wave with the frequency , which
is scanned in the vicinity of the transition half-frequency 0 /2. Due to the linear
Doppler effect, a traveling wave gets a frequency shift 0 = k v, where
is the field frequency in a laboratory frame of reference, 0 is the same in the
reference frame of a molecule moving with the velocity v, k is the field wave vec-
tor. In a standing wave, half of the photons have the wavevector k and half have
the wavevector k; therefore, their Doppler shifts differ in signs. As a result, if
a two-photon transition occurs through one photon absorbed from the forward
wave and the other photon, from the backward one, the shifts are fully compen-
sated, so that the observed resonance will look like a sharp peak with the natural
or collision width, on a broad Dopplers pedestal.
(a) (b)
(c) (d)
Fig. 6.8 Doppler-free two-photon spectroscopy: (a) the levels ( is the scanned laser frequency and
0 is the frequency of the observed fluorescence); (b) the scheme of the spectroscope; (c) the idea of
Doppler-shift compensation; (d) the signal as a function of the laser frequency.
(a) (b)
Fig. 6.9 Doppler-free saturation spectroscopy: (a) the scheme of the saturation spectroscope; (b) the
observed signal as a function of the laser frequency.
Note that it is convenient to register the resonance indirectly, by observing

single-photon fluorescence accompanying the transition of a molecule from the
excited level to a third level. A disadvantage of the two-photon spectroscopy
is the necessity to use a strong field, because of the small transition probability,
which leads to a considerable shift of the resonance due to the optical Stark effect
(see (5.48)).
The Doppler-free saturation method [Letokhov (1975)] also uses counter-
propagating waves with the same frequency (Fig. 6.9). The forward wave has
a larger intensity and causes strong saturation of molecules whose velocities have
appropriate projections vz = ( 0 )/k on the wave propagation direction. The
backward probing wave at , 0 interacts with another group of molecules,

which have the opposite sign of vz , and therefore is strongly absorbed. However,
at = 0 , we have vz = 0, and the resonance is bleached for the probe wave due
to the saturation caused by the forward wave. Hence, the readings of the detector
registering the power of the probe wave after the sample will grow at 0 .
The width of the resonance will be determined by the natural, collision, or field-
induced (4.101) broadening.
For more detailed description of nonlinear spectroscopy, see Refs. [Letokhov
(1975); Akhmanov2 (1981); Bloembergen (1977); Steinfeld (1978); Walther
(1976); Letokhov (1983)]. Let us mention other types of Doppler-free spec-
troscopy: the methods of quantum beats (Sec. 5.2), molecular beams, buffer gas.
In the last method, an inert gas is added to reduce the free path of the molecules
under study, so that the free path becomes much less than the wavelength. This
leads to a narrow peak with the homogeneous broadening appearing in the center
of a broad Doppler line.
6.4.3 Raman amplification

Consider now non-parametric interaction of two modes with different frequencies.
Here, the most interesting effect is Raman amplification, which is closely related
to the inelastic scattering of light. In the macroscopic description, the scattering
is explained by a resonance of the cubic susceptibility (3) (2 , 1 , 1 ) at 1
2 0 > 0. Here, the combination frequency is close to the frequency
0 ba of some matter excitation. Usually, 0 is one of the eigenfrequencies
of molecular vibrations (0 102 103 cm1 ) or the frequency of an acoustic
wave (0 0 0.1 cm1 ). In the first case, the scattering is called the Raman one,
in the second one, the Mandelshtam-Brillouin one. In piezoelectric crystals, one
also observes light scattering by polaritons, 0 . 103 cm1 (Secs. 4.2, 6.5). In
this case, similar to the case of acoustic-phonon scattering, (3) and 0 depend on
the angle between k1 and k2 . Therefore these effects are actually resonance ones,
and they will be considered in more detail in Sec. 6.5.
It is important that the imaginary part of the cubic susceptibility is negative in
the vicinity of the Stokes resonance. As a result, the energy is transferred from
the high-frequency field components to the low-frequency ones. The spectrum
becomes redder and the matter is heated; as a result, the population difference
Na Nb for the transition with the frequency ba reduces.
Let two polarized plane waves with the frequencies 1,2 > 0 be incident on
a transparent medium, orthogonally to the boundary. From (6.209), we find the
equations for the energy fluxes,

dF1 /dz + F1 F2 = 0, (6.217)
dF2 /dz F1 F2 = 0, (6.218)

where
322~k1 k2 (3)00 /n21 n22 = (2)
st (Na Nb ) > 0, (6.219)
.
(3)00 Im(3) (2 = 2 1 + 1 )..e2 e2 e1 e1
.
= Im(3) (1 = 1 2 + 2 )..e1 e1 e2 e2 . (6.220)
The last equation follows from (6.88) (see also (6.146), (6.192)); it provides
conservation of the total number of photons in Raman two-photon transitions,
F1 (z) + F2 (z) = C. (6.221)
By substituting (6.221) into (6.217), (6.218), we find
C
F1 = F10 , (6.222)
F10 + F20 eCz
C
F2 = F20 , (6.223)
F20 + F10 eCz
where Fi0 Fi (0) and C F10 + F20 . At a sufficiently large distance,
F1 0, F2 C, i.e., all photons become Stokes ones (Fig. 6.10). The re-
leased energy is then spent on the excitation of matter; this phenomenon is used
for the generation of ultrasound via stimulated Mandelshtam-Brillouin scattering.
In practice, the energy conversion is not complete because of several effects we
did not take into account here. In particular, these are spontaneous-stimulated
transitions and the generation of new components with the frequencies 21 2
and 22 1 , as well as waves with the same frequencies 1 , 2 but with different
polarization etc.
Let now F10 F20 , then at sufficiently small z the variation of the high-
frequency field can be neglected (the undepleted-pump approximation). In this
case, (6.223) yields exponential Raman amplification,
F2 = F20 exp(2 z),
(6.224)
2 F10 = 3222 (3)00 I10 /c2 n1 n2 > 0.
In practice, for strong Raman resonances with molecular vibrations in liquids, 2
is as high as 1 cm1 at I & 0.1 GW/cm2 . Thus, Raman anharmonicity enables light
(a) (b)
(c) (d)
Fig. 6.10 Raman interaction of two waves at various initial intensities. The horizontal axis is the
distance z in 1/C units and the vertical one shows the intensities Fi in units C = F10 + F20 , i.e.,
the functions F1 = (1 + ez )1 and F2 = 1 F1 , where = F20 /F10 is the relative initial intensity
of the Stokes wave. (a) = 1/25, the dashed line shows the asymptotics describing the exponential
amplification of the Stokes wave in the undepleted-pump approximation; (b) = 1/5; (c) = 1; (d)
= 5. The anti-Stokes wave, roughly, decays exponentially.
amplification without the population inversion. Raman amplification (or, for anti-
Stokes frequencies, attenuation) is used in spectroscopy, as well as for shifting
the frequency of lasers. If the feedback is added, with the help of a mirror, an
amplifier becomes an oscillator (a Raman laser).
6.4.4 Spontaneous and stimulated scattering

Even without a feedback or an external signal at frequency 2 , with F10 , 0 the
matter will emit incoherent radiation at frequencies 1 0 . At 2 l 1, the
radiation at Stokes frequencies 2 , i.e., the intrinsic noise of the Raman amplifier,
can be comparable with the pump, F2 F1 . This effect is called stimulated
Raman scattering (StRS). The emerging noise field with the central frequency
1 0 plays the role of the pump for the second Stokes component, with the
frequency 1 20 , which excites the third one, and so on. In addition, due to
four-frequency parametric interactions like 21 2 3 1 + 0 , intense
anti-Stokes components appear, and a quantitative analysis of the phenomenon
becomes difficult.
Let us estimate the power of StRS in the undepleted-pump approximation,
without taking into account higher-order Stokes and anti-Stokes components.
At F20 = 0, the Stokes field is generated due to spontaneous (to be precise,

spontaneous-stimulated) transitions. According to the Kirchhoff law (Sec. 7.1),
which is valid for two-photon transitions as well [Klyshko (1980)], the intensity
of the noise at the frequency 2 at the output of the amplifier is (in photons per
mode)
N = (N + 1)(G 1), (6.225)
where
( " # )1
Nb ~(1 2 )
G exp(2 l), N = exp 1 , (6.226)
Na Nb T
T is the temperature of the matter, and we assume that the transition a b is not
saturated.
For passing from N to F, one has to multiply the photon flux density in one
mode, Nc/L3 (L3 is the quantization volume), by the effective number of modes
for the amplifier or the detector,
Nc Nc3 k
F = g = =N . (6.227)
L3 (2)3 22
Here, we have considered a single polarization type, assumed the refractive index
to be unity, and denoted the effective frequency and solid-angle bandwidths of the
amplifier or the detector as and , respectively.
If the scattering volume has the shape of a thread, with a 1 and G 1,
only longitudinal scattering ( 0) or backscattering ( 180) into the solid
angle A/l2 is considerable.k (Here, A a2 is the cross-section area of the
scattering volume and l is its length.) Then, from (6.227) we find the relation be-
tween the noise power per unit frequency band, P = ~FA/, and the number
of photons per mode,
P = ~n2F N/2, (6.228)
where the dimensionless number nF A/l is called the wave parameter or the
Fresnel number. Substitution of (6.225) into (6.228) yields ( 2 , 2 )
P = ~n2F (N + 1)(el 1)/2. (6.229)
The total output power is found by integrating this expression in , which

reduces to the multiplication by the effective noise band and the substitution
of 1 0 for . In the case of weak pumping, is determined by the width
k In stimulated Mandelshtam-Brillouin scattering, it is only backscattering that is efficient, since
(0) = 0 (see (6.93)).

of the Raman resonance, i.e., by the decay rate of molecular vibrations (usually,
/2c 1 10 cm1 ). At high amplification, becomes (l)1/2 times as
narrow (Sec. 2.3)
Thus, according to (6.229), the scattered power depends on the pump inten-
sity as exp(lF10 ) 1. At lF10 1, the exponential dependence turns into a
linear one, and then the scattering is called spontaneous (SpRS). Replacing A/l2
in (6.229) by , we find the power of spontaneous Stokes scattering per unit
frequency and solid angle,
P = (~c/3 )(N + 1)VF10 , (6.230)
where V = Al is the volume of the scattering area. It is not difficult to verify
that this expression agrees with the above-calculated cross section (6.148) of a
spontaneous-stimulated transition. In contrast to StRS, SpRS has a broad direction
diagram, determined by the convolution of (3) with the polarization unit vectors
(6.220).
At F2 F1 , the role of the pump is played by the lower-frequency field,
and the external (anti-Stokes) field F1 (z), according to (6.217), (6.218), decays
exponentially. This effect is used in the spectroscopic method of inverse Raman
scattering. The noise field is then described by (6.225), (6.229) or (6.230) with
the permutation of indices 1, 2 and the replacements N + 1 N, .
Therefore, spontaneous anti-Stokes scattering with a given frequency is
N/(N + 1) = Nb /Na = exp(~0 /T )
times as weak as the Stokes scattering, provided that all other conditions are the
same. With a sufficiently strong pumping, the power of the anti-Stokes component
is saturated. The number of photons per mode tends then, according to (6.225), to
N, i.e., the anti-Stokes field acquires the brightness temperature
T e f = T 1 /0 T. (6.231)
6.4.5 Self-focusing
This single-frequency effect is connected with the self-action of a quasi-plane
quasi-monochromatic wave due to the real part of the cubic susceptibility (3)
Re(3) (, , ). The main contribution in (3) is from the electro-striction and
orientational anharmonicity (then, (3) > 0). In an absorptive medium, tempera-
ture anharmonicity ((3) < 0) and saturation are added.
With an account for linear susceptibility, the polarization amplitude is (we
assume that (1)00 = (3)00 = 0)
P = ((1) + (3) |E|2 )E (|E|2 )E,
where E E1 (r, t) is the slowly varying amplitude, i.e., the envelope. As a result,
the refractive index depends on the local intensity of the field,
n(r, t) [1 + 4(|E|2)]1/2 n0 + n2 |E|2 , (6.232)
where n2 2(3) /n0 .

Let (3) > 0, then a light beam entering a medium forms a dielectric waveguide
with the profile n(r ) = n2 |E(r )|2 repeating the intensity distribution in the
beam. The propagation velocity c/n at the beam axis will be less than at its edges,
so that the side beams will bend towards the axis (nonlinear refraction), and the
beam will shrink. This will lead to an additional increase in n near the axis, as
well as in the rate of the shrinking, so that, as a result, the beam will collapse to
a certain minimal size, determined by competing processes (Fig. 6.11).
Fig. 6.11 Self-focusing and defocusing (dashed line): 1, a beam with low power, P P0 , slowly
diverges due to diffraction; 2, a beam with a critical power, P0 , becomes channeled; 3, a beam with
P > P0 collapses.
One of the competing effects is diffraction. It is described by the divergence

angle d 0 /2n0 a, where a is the initial beam radius and 0 2c/, while the
nonlinear refraction is characterized by the angle of total internal reflection due to
the refractive index variation n,
s
n0 2n
NL arccos , (6.233)
n0 + n n0
where the beam profile is assumed to be rectangular. If refraction and diffrac-

tion compensate each other exactly, then waveguide propagation occurs, with a
constant beam radius (self-channeling). Assuming NL = d /2, one can obtain a

rough estimate for the self-channeling condition:
nmin n2 |E|2min n0 2d /8 20 /32n0 a2 . (6.234)
It follows that the waveguide cross-section a2 scales as the inverse intensity of
light. In this case, the power of a self-channeling beam should be
P0 = cn0 |E|2min a2 /8 c20 /256n2. (6.235)
2
This estimate differs /4 times from a more accurate result, which will be
obtained below from the nonlinear quasi-optical wave equation (6.239) in the
aberration-free approximation, i.e., under the assumption that the wave front is
spherical,
P0 = c20 /642n2 . (6.236)
Let 0 = 1 , (3) = 1012 cm2 /erg (carbon sulfide), and n0 = 1.5, then,
according to (6.236), P0 104 W. This estimate shows that self-focusing can
considerably change the features of other nonlinear effects, such as harmonics
generation, StRS, and others, which are usually observed at strong pumping. Self-
focusing is also related to the important problem of the optical strength of matter,
which determines the maximal intensity of laser beams.

6.4.6 Self-focusing length
Clearly, at P > P0 the beam will compress (self-focus), and under the opposite
condition it will diverge due to diffraction. The length of self-focusing will appar-
ently depend on the value (P P0 )/P0 .
In order to find this dependence, consider the solution to the nonlinear wave
equation (6.151) for a monochromatic field Re E(r)eit in quasi-optical and
aberration-free approximations (see Ref. [Vinogradova (1979)]). In a homoge-
neous isotropic medium without external sources, the field is transverse, so that
(6.151) leads to the nonlinear Helmholtz equation,
2 E + k2 (1 + (2 /0 )|E|2 ) E = 0, (6.237)
(3)
where k = n0 /c, n0 = 0 , 2 4 = 2n0 n2 .
Let a polarized light beam with a limited cross-section but with a narrow di-
rection diagram (narrow angular spectrum) be incident on a nonlinear transparent
medium occupying a semispace z > 0. Then the field in the medium can be
searched in the form
E(r) = eE(r) exp(ikz), (6.238)
where |e|2 = 1, ez = 0, and E(r) is a slowly (compared to eikz ) varying amplitude

(SVA) describing the variation of the local amplitude of field oscillations along
the beam and over its cross section.
In practice, the typical distances zd , zNL at which E(z) changes noticeably due
to diffraction and nonlinearity are much greater than ; therefore, 2 E/z2 is neg-
ligible compared to 2k E/z (the SVA approximation). In this case, from (6.237)
and (6.238) we obtain a nonlinear parabolic equation for E(r),
2ik E/z + 2 E + (k2 2 /0 )|E|2 E = 0, (6.239)
where 2 2 /x2 + 2 /y2 . At 2 = 0, (6.239) is the basic equation of quasi-

optics, which studies the propagation of directed wave beams.
If we neglect the variation of E over the cross-section, (6.239) leads to
E(z) = E(0)eikz , k/k = n/n0 = 162(3) I/cn30 , (6.240)
i.e, the nonlinearity only changes the phase velocity /(k + k) = c/(n0 + n) of
the plane wave. For CS2 , at I = 1 GW/cm2 , n = 2 105 .
Now, let the incident wave have a Gaussian profile and a plane wave front
at the boundary: E(x, y, 0) = E0 exp(2 /a20 ), where 2 x2 + y2 and a0 is the
initial radius of the beam. Let us try to find the solution to (6.239) in the form of
a Gaussian beam with a variable radius, a(z) a0 f (z), and a parabolic wave front
with a variable curvature (z) on the axis,
E(r) = E0 exp[F(z)2 + i(z)]/ f (z), (6.241)
F(z) ik(z)/2 1/a20 f 2 (z). (6.242)
Here, (z) is the additional to kz phase delay on the axis, and the factor 1/ f pro-
vides the dependence of the wave power on z,
Z
P = (cn0 /8) dx dy |E|2 = (cn0 /16)E02a20 . (6.243)
At small , the wave (6.241) has approximately spherical wavefront with the ra-
dius 1/(z). Below, it will be shown that the curvature of the front and the width
of the beam at a given z are related through Eq. (6.247).
From (6.241), we find
E/z = (2 F 0 + i0 f 0 / f )E, 2 E = 4F(1 + 2 F)E, (6.244)
E02 22 E02

22
|E|2 = exp 1 .
f2
2 2
a0 f f2 a20 f 2

In the last equation, we have passed from a Gaussian profile to a parabolic one,
which is possible for the paraxial part of the beam (the aberration-free approxi-
mation). By substituting (6.244) into (6.239), one can find the functions f, , .
Indeed, setting to zero the sum of the coefficients by 2 , we find
2ikF 0 + 4F 2 k2 /z2NL f 4 = 0, (6.245)
where
z2NL n0 a20 /4n0 (6.246)
and n0 is the refractive index variation at point r = 0. The imaginary part of
(6.245) yields
= d(ln f )/dz, (6.247)
and the real part,
f 00 = 1/z20 f 3 . (6.248)
Here, the following notation has been introduced:
z20 (z2 2 1
d zNL ) = z2d /(1 P/P0 ) = z2NL /(P0 /P 1),
(6.249)
zd ka20 /2, z2d /z2NL = P/P0 = (n0 /n0 )(ka0 )2 .
The z0 parameter determines the length at which the beam collapses, and the zd
parameter has the meaning of the diffraction length; here, P0 and P are given
by Eqs. (6.236) and (6.243).
Hence, we finally obtain the z dependencies for the width and the curvature
radius R(z) of a parabolic beam due to its self-action,
a(z)/a0 = f = (1 + z2 )1/2 , (6.250)
R(z) = 1/ = z(1 + z2 ), (6.251)

with z z/z0 . The function (z) can be found from the coefficients of the ze-
roth order in appearing after the substitution of (6.244) into (6.239). The field
amplitude in the beam has the form

E0 2 ik2
E(r) = exp 2
+ 2
+ i(z) . (6.252)
(1 + z2 )1/2 2
a0 (1 + z ) 2z(1 + z )
This solution at P P0 (i.e., at z0 zd ) describes the so-called TEM00 wave,
which is the simplest solution to the nonlinear parabolic equation,

E0 2
ETEM = exp 2 . (6.253)
1 + iz/zd a0 (1 + iz/zd )
Fig. 6.12 Dependence of z0 (in ka20 /2 units) on the beam power P (in the units of critical power P0 ),
given by the expression |1 P|1/2 . At P < 1, z0 has the meaning of the beam diffraction length, and
at P > 1, it is the collapse length. The latter, at P 1, is approximately equal to the nonlinear
length zNL (dashed line).
(Note that 1 + iz = (1 + z2 )1/2 exp(i arctan z), i.e., here, (z) = arctan(z/zd ).)
The diffraction length zd determines the distance at which the cross-section of the
beam is doubled and the front radius achieves its minimal value 2zd = ka20 b (the
so-called confocal parameter of the beam). At P P0 , the role of zd is played by
z0 , and diffraction slows down. Finally, at P > P0 , the parameter z20 and, hence, z2 ,
becomes negative, the beam is narrowed, and its width and the wavefront radius
turn into zero at z = |z0 |; at P P0 , |z0 | zNL (Fig. 6.12). Certainly, quasi-
optical aberration-free approximation describes correctly only the initial stage of
self-focusing; however, the typical self-focusing length should be on the same
order of magnitude as |z0 |.
In the vicinity of the focus, an important role can be played by higher-order
nonlinearities (5) , . . . and multi-frequency nonlinear effects, for instance, stimu-
lated scattering. Note that in the case of pulsed fields, the focus position |z0 (t)| is
time-dependent: as the intensity increases, the focus moves from infinity to some
minimal distance, and then it goes to infinity again (the moving focus).
Next, let a wave with a plane wavefront and a complicated profile
|E(x, y, 0)|2 I(x, y) (a wave with transverse amplitude modulation) be inci-
dent on a nonlinear medium. In the course of propagation, diffraction tends to
smoothen the profile inhomogeneities, while the nonlinear refraction, in contrast,
makes them more pronounced (self-modulation). Qualitatively, these effects are
described by (6.250) if a(z) is understood as the width of an inhomogeneity. As a
result, a wave can split in numerous waveguided filaments, each of them carrying
a power of P0 . This splitting is an example of dynamic transverse instability for
waves in a nonlinear medium and is similar to the formation of solitons in the case
of longitudinal instability (see below).
Clearly, at (3) < 0, self-defocusing should take place (Fig. 6.11). It is de-
scribed by the above-given theory with negative P0 and z2NL . This effect can
be easily observed in continuous-wave laser beams by adding absorbing dyes to
transparent solutions.
In two-frequency (and/or two-wave) experiments, mutual focusing or defocus-
ing is observed due to Re(3) (2 = 2 1 + 1 ): an intense light beam with a
frequency 1 turns a plane-parallel plate into a collecting or dispersing lens for
the wave 2 . The same frequency component of the cubic susceptibility describes
induced dispersion and the optical Kerr effect: a plane pump with the frequency
1 turns an isotropic medium into a birefringent one for the second wave, with
the frequency 2 . The optical Kerr effect and mode-locked lasers enable the man-
ufacturing of optical shutters with switching times on the order of 1 ps.
Consider now a plane quasi-monochromatic wave with the time-modulated
envelope E(t). In a linear medium, within the first approximation of the disper-
sion theory, E(t) propagates without changing its shape, with the group velocity
u = (k). In the framework of the second approximation, d2 /dk2 , 0, and
the intensity spikes get spread; this effect is similar to the diffraction smooth-
ing of transverse inhomogeneities. In a nonlinear medium, self-action can lead
to an inverse effect: at (3) d2 /dk2 < 0, pulses with sufficient energy are com-
pressed (self-compression, see Ref. [Vinogradova (1979)], Sec. 9.5). A quasi-
monochromatic wave with a small initial modulation splits, as a result of self-
modulation, in separate pulses with fixed energy and shape, the solitons,l , which
are analogues of waveguided filaments in the case of self-focusing. In practice,
such effects are observed for picosecond pulses, so that inertial nonlinearity mech-
anisms (for instance, the temperature one) are not manifested.
6.5 Parametric interactions
In the case of a parametric interaction of m modes with positive frequencies i , the

stationarity condition has the form of the energy conservation law for an m-photon
process,
m
X
si i = 0, (6.254)
i=1
where si = 1, i = (ki ) > 0. According to the one-dimensional model, the

waves interacting in an infinite layer should satisfy the transverse momentum con-
l Recall that light solitons (with the area 2) are also formed in the range of resonance absorption due
to the two-level anharmonicity (Sec. 5.1)

servation law,
k x = 0, ky = 0, (6.255)
P
where k si ki . Formally, one obtains (6.254) and (6.255) by doing the three-
dimensional Fourier transformation of Maxwells equations (Sec. 6.3). In experi-
ment, the interaction time and cross-section A a2 are always limited; therefore,
(6.254) and (6.255) are valid only up to an accuracy of 1/, 1/a, respectively,
|| . 1/, |k | . 1/a. (6.256)
In this section, as a rule, for simplicity we will neglect linear absorption

(i = 0). Then, interaction of m modes can be described within the linear ap-
proximation, according to (6.173), by a system of linear equations,
dE1 2s1 1
= P (z)eiz , (6.257)
dz icn1 1
where
P1 (z) = (m1) (s1 1 ; s2 2 ; . . . ; sm m )E2 (z) Em (z), (6.258)
Ei = Eki at si = 1 and Ei = Eki at si = 1, (m1) is the convolution of the

nonlinear susceptibility tensor with the polarization unit vectors, n n cos cos ,
n is the refractive index, and are the angles formed by the Poynting vector with
the z axis and the k vector, respectively, kz . Equations for E2 , . . . , Em have
a similar form and can be obtained by subscript permutations.
It follows from (6.257) that parametric interaction is efficient only at suffi-
ciently small ||, i.e., under the condition that the longitudinal momentum of the
field is preserved,
X
|| | si kiz | < 1/l. (6.259)
The phase matching condition k 0 reduces dramatically the number of

essential interactions, i.e., combinations {ki , i , si }, especially in an anisotropic
medium where the refractive index n (, , ) = ck/ (k) depends not only on
the frequency, but also on the polarization index = 1, 2 and the direction k
k/k {, }.
6.5.1 Undepleted-pump approximation the near field

Let k1z > 0 and E1 (0) = 0, i.e., the mode labeled by 1 is not excited by an ex-
ternal source. If the nonlinearity, or the layer thickness, or the incident pump
fields Ei (0), i = 2, . . . , m, are sufficiently small, one can neglect the effect of the
nonlinearity on the pump, i.e, assume that the nonlinear polarization P1 (z) at fre-
quency 1 in the right-hand side of (6.257) is a given function of the coordinates,
determined by the spatial distribution of free incident fields.
For a given polarization, taking into account definition (6.168), from (6.257)
we find the expression for the amplitude of the mode (k, ) excited in a homoge-
neous transparent linear medium by a Fourier component of the polarization,
82 ikuk
Z
E (k) = 3 2 d3 r eikr e P(k , r), (6.260)
L nk cos k V
where uk is the group velocity. Note that (6.260) determines the radiation field
outside of the radiating area V regardless of its shape.
In the case of a plane-parallel layer with the thickness z, it follows from (6.257)
that
2s1 1 (m1) eiz 1

E1 (z) = E2 (0) . . . Em (0) , (6.261)
cn1
!m
21 |(m1) |2 z2
!
8 2 z
I1z (z) = sinc I2z (0) . . . Imz (0), (6.262)
c 16n1 . . . nm 2
where Ikz = cnk |Ek |2 /8 = ~k Fkz is the longitudinal component of the energy
flux density in mode k and sinc(x) (sin x)/x. Eqs. (6.261), (6.262) give the
efficiency of parametric frequency conversion.
Note that the new mode (k1 , 1 ) k1 is determined through the pump modes
k2 , . . . , km by conditions (6.254), (6.255) only up to the polarization type 1 and
the sign 1 of the longitudinal component k1z (see (6.165)). In other words, each
P
combination of polarized pump waves si ki , i = 2, . . . excites, in principle, four
waves with the frequency 1 , which differ by indices 1 = 1, 2 and 1 = 1.
However, the factor sinc2 (z/2) in (6.262) usually strongly suppresses the wave
with k1z < 0 (about (k1z z)2 times, see Fig. 6.13).
Thus, in the undepleted-pump approximation, excitation of new waves scales
as the product of the incident intensities, the squared nonlinear susceptibility,
and, provided that phase matching is satisfied (||z 1), as the square of
the interaction length. In the case of the opposite inequality, z2 is replaced by
(/2)2 sin2 (z/2) 2/2 , i.e., if the phase velocities are not matched, the ampli-
tude of the new wave periodically turns to zero. The distance /||, at which the
intensity of the wave varies monotonically, is called the coherence length.
Fig. 6.13 In the case of parametric frequency conversion, 2 + 3 4 1 , there are two waves,
k1() , satisfying the energy and transverse momentum conservation laws. However, the coherence
length /() for the k1() wave is extremely small, on the order of the wavelength.
6.5.2 The far field

In practice, Eqs. (6.261), (6.262), as well as, generally, the one-dimensional ap-
proximation with a finite number of transverse modes, determine only the near
field of the nonlinear sample, where the diffraction effects are not important due
to the finite cross-section A of the interaction area. However, in the undepleted-
pump approximation it is not necessary to use the mode expansion and the one-
dimensional model. Indeed, the problem is reduced to solving a single wave
equation (6.151) at a given distribution of monochromatic sources P(, r), i.e.,
to finding the Greens function G(, r) of the Helmholtz equation. This function
has especially simple form in the case of a homogeneous isotropic medium and a
large distance r from the sources to the observer.
We have already defined the Greens function of the Maxwell equations in the
, k representation (see (4.23)). One can show that its three-dimensional Fourier
transformation yields, in the first order in /r, i.e., for the wave field, the expres-
sion
2 ikr
G(, r) = e (r), (6.263)
c2 r
where = r r is the projector onto the plane orthogonal to the direction
r r/r and k n()/c. Let r kA (the far field), then (6.263) leads to the
following relation between the field and the external polarization (compare with
(6.260)):
2 ikr
Z
0
E(, r) = 2 e d3 r0 eikr (r) P(, r0 ), (6.264)
cr v
where k kr and r connects some point inside the V area containing the sources
and the observation point. Thus, at large V the far field scales as the Fourier
transform of the external polarization.
Consider generation of a field with the frequency 1 by an undepleted pump

consisting of m 1 plane waves. At V , the nonlinear polarization is also a
plane wave, its frequency and wave vector being
m
X m
X
1 = si i , K si ki . (6.265)
i=2 i=2
Let the tensor (m1) be nonzero within a rectangular box with dimensions a, b, c,
then, at s1 = 1, (6.264) yields
21 ik1 r .
E (1 , r) = 2
e ( (m1) ..E2 . . . Em )V f (r), (6.266)
cr
Z
0 0
f (r) d3 r0 eik r /V = sinc(k x a/2)sinc(ky b/2)sinc(kz c/2), (6.267)
k n1 1 r/c K. (6.268)
The direction diagram of the field 1 at fixed K is mainly determined by the

function f (r). For an efficient frequency conversion, the length of K should be
close to n1 1 /c = k1 ; in this case, the radiation intensity is maximal in the di-
rection K (for which f 1) and has a considerable value within a solid angle of
about 21 /A, where A is the area of the V projection onto a plane orthogonal to K
(Fig. 6.14).
Fig. 6.14 To the definition of the direction diagram I() for the field emitted by a polarization with
the frequency and the wave vector K from a domain with the dimensions a, b, c: the radiation has a
noticeable intensity only if the observed wave vector k belongs to a domain of dimensions 1/a, 1/b, 1/c
near the point K.
Fig. 6.15 Angular dispersion of the refractive index ne () for the extraordinary wave enables the
frequency dispersion to be compensated and the phase matching condition no (1 ) = ne (21 , 1 ) to be
satisfied for second-harmonic generation, as well as for other three-wave interactions.
6.5.3 Three-wave interaction

Let us consider, in the framework of the one-dimensional model, the interaction
of three waves or modes with the frequencies 1 + 2 = 3 , due to the quadratic
susceptibility, which has a noticeable value only in piezoelectric crystals. For the
phase matching condition, k k3 k2 k1 = 0, to be satisfied in the transparency
range, the crystal should be birefringent and have a certain orientation of optical
axes with respect to the incident beams (Fig. 6.15).
Let us re-normalize the wave amplitudes so that their squares are equal to Fi ,
the longitudinal photon flux densities in ~i units (see (6.178)). From (6.257) (or
from (6.180)) it follows, at s1 = s2 = s3 = 1 and in the case of all waves
co-propagating, ni > 0, that
da1 /dz = ieiz a2 a3 , (6.269)

da2 /dz = ieiz a1 a3 , (6.270)
da3 /dz = ieiz a1 a2 , (6.271)
where
(323~1 2 3 /c3 n1 n2 n3 )1/2 , (6.272)

.
kz , (2) (3 = 2 + 1 )..e3 e2 e1 . (6.273)
Here, the convolution can be considered as real and invariant to a simultaneous

permutation of the frequencies and the polarization unit vectors ei (Sec. 6.1).
After multiplying these equations by ai , we see that the rates of flux variations,
dFi /dz, are equal for modes 1 and 2, while for modes 1 and 3 they have opposite
P
signs. Thus, the energy flux ~i Fi is only redistributed, without absorption,
between the three modes, and the share of each mode scales as its frequency (the
Manley-Rowe relations, see Sec. 6.3),

F1 = F2 = F3 , (6.274)
where
Fi Fi (z) Fi (0). (6.275)
There is also the third independent integral of motion (6.198) for Eqs. (6.269)
(6.271), determining the phase difference. These integrals allow one to solve
Eqs. (6.269)(6.271) in terms of a single quadrature. As a result, the ai (z) depen-
dence is given by elliptic functions describing a periodic energy exchange between
the three modes [Bloembergen (1965)]. Here, we will consider the limiting cases
where one of the waves is much stronger than the other two (the undepleted-pump
approximation, or the parametric approximation).
6.5.4 Frequency up-conversion

Let F10 F2 , F3 (undepleted low-frequency pump), a10 = a10 , and = 0, then
(6.270), (6.271) yield
da2 /dz = ia3 , (6.276)
da3 /dz = ia2 , (6.277)
a10 = (3232 3 /c3 n1 n2 n3 I10 )1/2 . (6.278)

These equations are easily solved with the help of the substitution ai = ci eiz :
a2 = a20 cos z + ia30 sin z,
(6.279)
a3 = a30 cos z + ia20 sin z.
Thus, modes 2 and 3, similarly to coupled pendulums, periodically (at a dis-
tance /2) exchange energy (in ~i units) according to (6.274) (Fig. 6.16).
Let us estimate a typical length of the parametric interaction, zNL 1/
0 n/42 E10 . Let 2 = 3 = 1 , ni = 1, = 108 G1 , I10 = 100 MW/cm2 , then
zNL = 0.3 cm.
At a30 = 0, it follows from (6.279) that
F3 = F20 sin2 (z/zNL ), F2 = F20 cos2 (z/zNL ). (6.280)
This equation describes parametric sum-frequency generation, or frequency up-
conversion, which is used, in particular, for visualizing IR radiation. At the inter-
action length z = zNL /2, the output intensity of the converter reaches its maxi-
mum, equal (in photon units) to the input intensity. The conversion efficiency is
then equal, in ordinary units, to 3 /2 > 1.
(a) (b)
Fig. 6.16 Parametric interaction in the case of a low-frequency undepleted pump. (a) The intensity
versus the distance z; (b) two versions of the radiation spectrum (arrows show the direction of the
energy transfer).
Equation (6.280) can be easily generalized to the case , 0. The sum-

frequency generation is then described by the equation
F3 = F20 [ sin(0 z)/0 ]2 , 0 (2 + 2 /4)1/2. (6.281)
Hence, at 2 42 we obtain (6.262) once again. Comparison of (6.280),
(6.281) with (6.261), (6.262) shows that the approximation of two undepleted
pumps, a1,2 = const, is applicable only at (2/)2 1. Then, the period of
spatial modulation is determined by the wave mismatch and not by the pump
intensity .
6.5.5 Parametric amplification and oscillation

Let F30 F1 , F2 (high-frequency pump), then, instead of (6.276)(6.278), we
have
da1 /dz = ia2 , da2 /dz = ia1 ,
!1/2 (6.282)
3231 2 .
I30 (3 ; 1 , 2 )..e3 e1 e2 .
c3 n1 n2 n3
The solution to this system is (compare with (6.279))
a1 = a10 cosh z + ia20 sinh z,
(6.283)
a2 = a20 cosh z + ia10 sinh z.
If a10 = 0, then (compare with (6.280))
F1 = F20 sinh2 z (G 1)F20 ,
(6.284)
F2 = F20 cosh2 z GF20 ,
Fig. 6.17 Parametric amplification and difference-frequency generation (high-frequency pumping,

forward interaction). (a) Dependence of the intensity on the distance; (b) typical frequency spectrum
and the directions of the energy transfer; (c) relation between the wave vectors at phase matching.
Fig. 6.18 Optical parametric oscillator.
where G is the parametric amplification coefficient. These equations describe the

effects of difference-frequency generation, 3 2 1 , and parametric ampli-
fication. In contrast to the case of a low-frequency pump, here the intensities grow
exponentially (Fig. 6.17). Note that (6.284) satisfies the Manley-Rowe equations
(6.274), F1 = F2 = (G 1)F20 .
It is not difficult to show that in the case of , 0, amplification and difference-
frequency generation with high-frequency pumping are described by (compare
with (6.281))
G = 1 + [ sinh(z)/]2 , (2 2 /4)1/2. (6.285)
Note that at 2 < 2 /4, the exponential growth turns into beats.
If there is a positive feedback at the frequency 1 or/and 2 (Fig. 6.18), the
amplifier turns into an optical parametric oscillator (OPO). One of the frequen-
cies (for instance, 1 ) is then called the signal one and the other one (2 ), the
idler one. Certainly, the oscillation requires that the absorption and other losses
(which we ignore here) are compensated for by the parametric amplification. The
OPO oscillation frequencies are mainly determined by the phase-matching con-
dition, i.e., by the refractive indices ni ; therefore, at a fixed pump frequency 3
(a) (b)
Fig. 6.19 Parametric amplification and difference-frequency generation in the case of backward in-
teraction. (a) Intensity (in F10 units) as a function of the coordinate z (in 1/ units) at F2l = 0 and
l = 1.5; the parametric amplification coefficient is 1/ cos2 (1.5) 200; (b) the phase-matching trian-
gle for backward interaction.
the OPO can be smoothly tuned by varying the orientation or temperature of the
crystal. The existing pulse-pumped OPOs cover the range 0.420 , which
conveniently complements the range of tunable dye lasers. By optimizing the
parameters of the crystal, the focused pump beam, and the cavity, one can obtain
oscillation even in the cw regime.
6.5.6 Backward interaction

Let the pump and signal waves propagate in the layer from left to right, k1z , k3z >
0, while the idler wave, from right to left, k2z < 0. The phase matching condition
can then be satisfied due to strong birefringence or anomalous dispersion. For the
backward wave, 2 > /2 and n2 < 0,m so that instead of (6.282) we have
da1 /dz = ia2 , da2 /dz = ia1 . (6.286)
Let the layer have a thickness l, then the boundary conditions have the form
a1 (0) = a10 , a2 (l) = a2l . It is easy to check that the solutions to (6.286) are
(Fig. 6.19)
a1 = {a10 cos[(l z)] + ia2l sin z}/ cos l,

(6.287)
a2 = {a2l cos z + ia10 sin[(l z)]}/ cos l,
where is assumed to be real. At l /2, these solutions tend to infinity, i.e.,
oscillations are generated, despite the absence of the mirrors. One can say that
backward interaction provides a distributed feedback (a similar effect takes place
in a backward-wave tube).
m To be precise, the sign of n n cos cos is determined by the angle = between the Poynting
vector and the z axis; however, in the transparency range, the anisotropy angle does not exceed several
degrees, and we neglect it here.
6.5.7 Second harmonic generation

Second harmonic generation (SHG) in the one-dimensional approximation and in
the absence of linear absorption is described by equations (6.270), (6.270), with
the subscript replacement 3 2 1,
da1 /dz = ieiz a1 a2 , (6.288)
da2 /dz = ieiz a21 /2, (6.289)
where
(643~31 /c3 n21 n2 )1/2 , k2z 2k1z , (6.290)

. .
= (2) (1 ; 21 , 1 )..e1 e2 e1 = 2(2) (21 ; 1 , 1 )..e2 e1 e1 . (6.291)
The factor 2 in the last expression was added according to relation (6.13).
One can easily see that these equations satisfy the condition of the field energy
conservation, which in this case coincides with the Manley-Rowe equation,
|a1 |2 + 2|a2 |2 = C1 , (6.292)
and the factor 2 is due to the fact that each second-harmonic photon has an en-
ergy twice as large as the pump photon energy. The second integral of equations
(6.288), (6.289), according to (6.198), has the form
|a1 |2 [ 2|a2 | cos(2 21 + z)] = C2 , (6.293)
where i are the phases of complex amplitudes ai .

In the general case, the solutions to equations (6.288), (6.289) describe spa-
tially periodic energy exchange between the modes of the pump and the second
harmonic. (These solutions can be expressed in terms of the elliptic sine func-
tion [Bloembergen (1965); Dmitriev (1982)]).
Here, we will only consider the most important and simple case where =
0, a2 (0) = 0. Then, assuming in (6.293) z = 0, we find that C2 = 0 and, hence,
at any z we have cos = 0, where = 2 21 . Thus, the phase shift between
the modes is constant. One can easily show that the phases are constant as well.
Indeed, let us substitute into (6.289) ai bi exp (ii ), bi > 0:
db2 /dz = (1/2) b21 sin = (1/2)||b21, (6.294)
d2 /dz = ( b21 /2b2 ) cos = 0. (6.295)
From the second equation, it follows that both 1 and 2 are independent of z, and
from the first one, that the energy transfer direction (from the first harmonic into
Fig. 6.20 Second harmonic generation. The horizontal axis shows the distance (in 1/ units), the
vertical axis corresponds to photon fluxes for the pump and the harmonic (in F10 units). At z 1,
every two pump photons turn into a single double-frequency photon.
the second one or vice versa) depends on the signs of and . In the case consid-
ered, b2 is increasing (from zero); therefore, sin > 0, i.e., = (/2) sign.
Substituting (6.292) with C1 = b210 into (6.294), we easily find the solution
db2 /dz = ||(b210 2b22 )/2, (6.296)
Z
1 db2 2 2b2
z= 2 2
= arctanh . (6.297)
|| b10 /2 b2 ||b 10 b 10
Hence,

b2 = (1/ 2)b10 tanh z, b1 = b10 / cosh z, (6.298)
where
1/2
||b10 / 2 = 2(2/cn1)3/2 1 ||I1z . (6.299)
Thus, the intensity of phase-matched second harmonic I2z (z) increases mono-
tonically as tanh2 z, and at z 1 it achieves the initial intensity of the pump
I1z (0) (Fig. 6.20). The pump intensity goes in this case down to zero as 1/ cosh2 z,
with its phase being constant. At z = 1/, the SHG efficiency is 58% in the en-
ergy and 29% in the photon number. At , 0 or a2 (0) , 0, monotonic solutions
are replaced by periodic ones, and reduces.
In practice, powerful pulsed lasers provide as high as several tens per cent.
Note that phase-matched SHG in piezoelectric crystals has important applications,
such as shifting laser frequencies up by as much as an octave.
Let us also mention much weaker effects that do not require phase match-
ing: SHG under the reflection from a medium without the center of symme-
try [Bloembergen (1965)] and incoherent scattering of light at frequency 2 by
non-centrosymmetric molecules [Kielich (1980)], as well as by any molecules,
atoms, and free electrons, due to the magnetic anharmonicity (Sec. 6.2).
Note that the condition E2 (+0) = 0 accepted above does not fully correspond
to an ordinary experiment where it is the incident second-harmonic field that is
zero, while the transmitted and reflected fields E2 (+0), E2 (0) are non-zero due
to the requirement that the tangent components of the fields E2 , H2 should be
continuous at the boundary [Bloembergen (1965)]. However, in practice E2 (0)
are indeed very small.
6.5.8 The scattering matrix

Parametric interaction between two modes in a nonlinear layer, considered in the
undepleted-pump approximation, leads to a linear relation between the input and
output amplitudes, which in the general case can be written as
a1 = g11 a10 + g12 a20 , a2 = g21 a10 + g22 a20 . (6.300)
The coefficients gi j form the two-dimensional scattering matrix of the sample,
which depends on the layer length l, absorption i , nonlinearity , the pump am-
plitude a30 , the wave mismatch and so on. If reflections from the layer bound-
aries are taken into account, the matrix becomes four-dimensional, while in the
presence of diffraction (finite transverse dimensions of A) it becomes infinitely
dimensional.
The scattering matrix should possess a certain symmetry, which follows from
the general principles, in particular, from the Manley-Rowe relations (6.274). In
the case where the input fields a1 and a2 are statistically independent, photon
exchange between the modes is described by the energy scattering matrix Gi j
|gi j |2 ,
F1 = G11 F10 + G12 F20 , F2 = G21 F10 + G22 F20 . (6.301)
Substitution of (6.301) into (6.274) yields
(G11 1)F10 + G12 F20 = (G22 1)F20 + G21 F10 .
Because F10 and F20 can be varied independently, the scattering matrix of a trans-
parent layer should satisfy the relations
G11 1 = G21 , G22 1 = G12 . (6.302)
The system (6.282) is symmetric with respect to indices 1, 2; therefore, G11 =
G22 G and there is only a single independent element of the energy scattering
matrix, the transfer coefficient G. As a result, (6.301) takes the form
F1 = GF10 + (G 1)F20 ,
(6.303)
F2 = GF20 + (G 1)F10 .
Fig. 6.21 Spontaneous multi-photon processes can be taken into account in the semiclassical the-
ory, by adding a single extra photon into the output modes (the corresponding arrows are directed
downwards). The figure shows the direct and inverse three-photon parametric processes.

6.5.9 Parametric down-conversion
The quantum theory, in principle, allows a three-photon transition in which trans-
parent matter (an atom or crystal) absorbs a photon from mode k3 and emits a
pair of photons into modes k1 and k2 , returning into its initial state (Fig. 6.21).
The probability of such a transition scales as (N1 + 1)(N2 + 1)N3 , where Ni are
the initial photon numbers per mode and the unities appear due to the quantum
fluctuations of the mode amplitudes in the ground state (Chapter 7). The prob-
ability of the inverse process, where photons in modes k1 , k2 are absorbed and a
photon is emitted into mode k3 , scales as N1 N2 (N3 + 1). As a result, the rate of
pair generation scales as
N1 = N2 = N3 (N1 + N2 + 1)N3 N1 N2 . (6.304)
Hence, in the first order in N3 , we have
N1 N10 = N2 N20 N10 + N20 + 1. (6.305)
Thus, due to the effect of the radiation at frequency 3 , the matter generates pairs
of photons with frequencies 1 and 2 = 3 1 distributed within a broad
range, from zero to the incident radiation frequency. In the case of macroscopic
matter, the frequencies and directions of emission are related through the phase
matching condition k3 = k1 + k2 . This effect, observed in birefringent crystals, is
called parametric down-conversion (PDC)n or parametric fluorescence [Zhabotin-
sky (1969); Klyshko (1980); Akhmanov (1971)]. PDC can be interpreted as the
quantum noise of an optical parametric amplifier.
n Editorsnote: in the original text, D. N. Klyshko used the term parametric scattering, suggested
in his pioneering work (1967) on the theory of PDC [Klyshko (1980)]. Later, the term parametric
down-conversion became widely accepted.
The scaling factor in (6.305) should be equal to the conversion coefficient

G12 = G 1, which was found in (6.285) in the framework of classical nonlinear
optics. Hence (compare with (6.303)),
N1 = N10 + (G 1)(N10 + N20 + 1) = GN10 + (G 1)(N20 + 1). (6.306)
A similar expression can be obtained for N2 through the permutation of indices

1, 2. Here, G 1 does not have to be linear in the pump intensity any more.
The obtained expression allows the following rule to be formulated. In or-
der to take into account spontaneous emission in the classical description of a
parametric amplifier, one should add an extra single photon into each input idler
mode.o The same result follows from (6.306) in the case of a more general trans-
formation,
N10 N10 + p, N20 N20 + q, N1 N1 + p,
where p+q = 1, 0 6 p 6 1. In particular, one can add half a photon to all numbers
of photons per mode.
According to (6.306), quantum noise yields G1 photons in each output mode
of a parametric amplifier, with G being the amplification coefficient for this mode.
A similar result (Kirchhoffs law) is valid for quantum and Raman amplifiers (see
(7.7) and (6.225)).
Note that at N2 = N3 = 0, it follows from (6.304) that three-frequency up-
conversion with low-frequency pumping has no contribution from quantum fluc-
tuations. In the case of four-frequency interaction, it is possible that two pump
photons decay in a pair of photons with the frequencies and 23 , occupying
the range 0 2 (hyper-parametric scattering or light scattering by light).p
Let us pass from photon numbers per mode to the spectral brightness. Using
(6.306) and (6.227), we find that at Ni0 = 0 (we assume that 1 = 0)
I (k1 ) ~1 F (k1 ) = (~1 /221 )[G(k1 ) 1], (6.307)
where F dF/dd is the flux of photons in the k1 direction with the fre-
quency (k1 ) per unit solid angle, unit area (orthogonal to k1 ) and unit circular
frequency. The value I (k, r) is called the spectral brightness (sometimes, sim-
ply the intensity) of incoherent radiation. Recall that the brightness, in the case of
a transparent medium, does not vary in the direction of light propagation.
o An idler mode, with respect to the considered signal mode with the frequency , is the mode
1
with the frequency 2 = 3 1 .
p Editors note: now this process is commonly called spontaneous four-wave mixing. Starting from the
beginning of the 2000 s, it is applied for the generation of nonclassical light, both faint one (photon
pairs) and bright one (squeezed light).
Thus, the noise spectral brightness of an ideal amplifier, normalized to G 1

(i.e., relative to the input), is
vac
I = ~/22 = ~3 n2 /83 c2 . (6.308)
This value can be naturally called the spectral brightness of zero-point fluctua-
tions of macroscopic field (for a single polarization type) or the vacuum spectral
brightness. It corresponds to the presence of a single photon in each mode. At
= 1 and n = 1, the brightness of fluctuations per unit wavelength interval is
vac vac
I = |d/d|I = ~c2 /5 0.6W/( cm2 sr). (6.309)
Thus, the relation between the spectral brightness and the photon number per
mode is
vac
I (k) = I N(k). (6.310)
Let us also estimate the brightness (effective) temperature of superfluores-
cence at the output of an ideal amplifier. From the Planck formula (2.102) and
the Kirchhoff law N = G 1, it follows that
T e f f (k) ~/ ln(1 + Nk1 ) = ~/ ln(1 G1
k )
(~/)G 106 K. (6.311)

The last estimate was made for = 1 and G = 100.
Substituting (6.285) and (6.308) into (6.307), we obtain the intensity of the
parametric amplifier quantum noise,
vac
I (k) = I sinh2 [(2 2 (k/4))1/2 l]/(1 2 (k)/42 ), (6.312)
where l is the thickness of the nonlinear layer, (k) is the deviation from phase
matching for mode k, and (2) E3 (see (6.282)). According to (6.312), the PDC
intensity has a sharp maximum at frequencies and directions {, , } satisfying
the phase matching condition k = 0.
This condition determines the frequency-angular spectrum of PDC (),
where is the angle of scattering, i.e., the angle between the observed wave vec-
tor k and the pump wave vector k3 . The dependence of the frequency on the angle
, as a rule, can be neglected, i.e, the PDC spectrum is axially symmetric w.r.t. the
direction k3 . Field with a given frequency is emitted along a cone with a certain
angle (). As one can see from Fig. 6.22, the frequency spectrum of PDC from
a blue pump covers a broad range of IR and visible wavelengths. The visible ra-
diation is directed mostly forward, at angles not exceeding several degrees, which
is due to the fact that crystals have small birefringence in the transparency range.
When the idler frequency approaches the eigenfrequencies of the crystal lat-
tice, which are usually in the region of hundreds of inverse centimeters, PDC
Fig. 6.22 Frequency-angular spectrum of parametric down-conversion in lithium niobate for different
pump wavelengths 3 : 1 is the angle of the cone along which the wave 1 is emitted. The angle 3
between the pump beam and the crystal axis is 90 . When 3 is reduced, the gap in the spectrum (for
3 < 0.53 ) disappears.
continuously turns into the Raman scattering by polaritons and optical phonons.
Then, has a resonance growth due to the contribution from electron-nuclear an-
harmonicity (Sec. 6.2), but this growth is accompanied by an increase in the linear
absorption 2 at the idler frequency, so that the brightness integrated over the fre-
quency, I , does not change much. Certainly, when the idler frequency is small
and 2 l 1, (6.312) has to be multiplied by N(2 /T ) + 1, where T is the lattice
temperature and N is the Planck function.
By analogy with the scattering by polaritons, PDC can be defined as the scat-
tering of the pump radiation by the field fluctuations in the idler modes, i.e., as
light scattering by light, similarly to the way Mandelshtam-Brillouin scattering
can be called light scattering by sound.
The important features of PDC, distinguishing it from other types of light scat-
tering in matter, are, first, a broad continuous spectrum, not related directly to the
eigenfrequencies of the matter, and, second, the two-photon structure of the emit-
ted radiation: at small pumping (l 1), signal and idler photons are emitted
only in pairs, practically simultaneously.
Note that, in addition to coherent (forward) emission, there is incoherent PDC
by separate non-centrosymmetric molecules (to be precise, by the density and
orientation fluctuations of such molecules). Then, the phase matching condition
plays no important role, since the momentum deficit of the field is provided by the
molecule.
Let l 1 and n 1, then from (6.312) we find the intensity of spontaneous
PDC (SPDC) in the direction k with the frequency (k),

vac 2 2
I (k) = I l sinc2 [(k)l/2]
= 4~c5 4 2 l2 I30 sinc2 [(k)l/2], (6.313)
where 3 and the polarization indices are assumed to be chosen so that
is minimal. Note that the PDC intensity in the directions of exact phase matching
depends on the nonlinear layer thickness quadratically, which is typical for SPDC.
Let I30 = 1 W/cm2 and l = 1 cm, then the amplification in the phase-matched
direction is G = 1 + 2 l2 1 + 107 (see the estimate after (6.279)), and it follows
from (6.311) that at = 0.5 , T e f f 1800 K. Such radiation can be easily seen
by eye and looks like colored rings. Note that the actual transfer coefficient of
the sample G0 with such pumping is always less than the unity because of the
reflection, absorption, and scattering losses.
The effective frequency band of the SPDC spectrum at a fixed observation
direction is determined, similarly to the band of parametric amplification, by the
phase matching width, i.e., by the condition l = . Considering only the linear
expansion of (), we can write
= (2/l)|/|1 2/|1 2 |, (6.314)
where i l/ui . According to Eq. (6.314), which is valid in the case of collinear
phase matching with 1 , 2 ,q the width of the SPDC spectrum (in Hz) is equal
to the inverse time delay (in seconds) between the signal photon and the idler one
during their passage through the interaction region. Usually, /2c 10 cm1
at l = 1 cm.
At l 1, (6.312) describes parametric super-fluorescence, or stimulated
parametric down-conversion. According to the estimates made above, stimulated
PDC can be observed from pulsed pump with the intensity about 100 MW/cm2 or
higher. Certainly, stimulated PDC, similarly to parametric oscillation, is observ-
able in practice only in the direction 0, at the frequencies of collinear phase
matching, i (0). This is due to the fact that, as it follows from the geometry of
the experiment, the effective interaction length le f in the case of a narrow pump
beam (a/l 1) is dramatically reduced (as a/ sin ) at a/l. The total power
of stimulated PDC can be estimated as
vac 2l
P = I A = I e A/4, (6.315)
where we have introduced the effective frequency and solid-angle bands and the
effective cross-section (see the estimate of P for the StRS (6.229)). Stimulated
q Editors note: It is also valid in the case of type-II phase matching, in which the signal and idler
photons are orthogonally polarized and therefore have different group velocities.
PDC is used, similarly to parametric oscillation, for a controllable shift of laser

radiation frequencies.

6.5.10 Light scattering by polaritons
This process is described by the system (6.269)(6.271) with linear absorption
introduced into one of the equations. In the approximation of undepleted high-
frequency pump, the interaction between the Stokes (signal) and the polariton
(idler) waves is determined by the equations
da1 /dz = ia2 eiz , (6.316)
da2 /dz a2 = ia1 eiz , (6.317)
where 2 /2, = a30 (see (6.282)), and the lower signs correspond to the
case of the backward idler wave. In the polariton range of the PDC spectrum, i.e.,
in the vicinity of lattice eigenfrequencies, the idler wave is strongly absorbed, so
that usually ||, l 1, and the amplitude of the idler wave is determined by
the local amplitude of the signal wave.
For this reason, let us search the solution to (6.317) in the form a2 (z) =
b2 exp(iz),
db2 /dz b2 + ib2 = ia1 . (6.318)
Here, the derivative db2 /dz does not much exceed |b2 |; therefore, if || | + i|,
it can be neglected. Then,
b2 = ia1 /( i). (6.319)
In this approximation, the incident idler radiation does not influence the output
field, due to the high absorption, which is much stronger than the parametric am-
plification. Substituting (6.319) into (6.316), we obtain
da1 /dz = (g/2)a1, (6.320)
where
g 2||2 /( i) = 643 1 2 |(2) |2 I30 c3 n1 n2 n3 ( i).
Thus, strong absorption of one of the two interacting waves leads to the ex-
ponential increase, similarly to the case of usual Raman interaction due to (3)
(Sec. 6.4), with the only difference that here, phase matching condition || is
important.
Comparing g at = 0 with the Raman amplification coefficient (6.224), with

the definition of 2 (6.167) taken into account, we find the equivalent cubic sus-
ceptibility (assuming ni = 1),
(3)00 (2) 2 (1)00

equiv = | | / . (6.321)
The same relation between the resonance susceptibilities of the first, second, and
third order follows from the microscopic theory [Klyshko (1980)].
The spectral brightness of the Stokes field I (k1 ) in spontaneous or stimu-
lated light scattering by polaritons can be found through the Kirchhoff law,
vac
I (k1 ) = I [N(2 /T ) + 1]{exp[g0 (k1 )z] 1}, (6.322)
or by using the results of Sec. 6.4. Note that if a2 is understood as the amplitude
of the sound wave, the above analysis will describe the Mandelshtam-Brillouin
scattering (spontaneous and stimulated).
6.5.11 Four-wave interactions

In centrally symmetric media, macroscopic quadratic susceptibility (2) is equal
to zero, and therefore the simplest parametric process involves four field modes
interacting due to the cubic susceptibility (3) (4 = 3 + 2 + 1 ). In the corre-
sponding quantum transition, four photons are absorbed or emitted and the state
of the matter remains the same.
The most important four-wave parametric effects are (Fig. 6.23) generation of
combination frequencies (for instance, the third harmonic) for converting laser ra-
diation to the UV and IR ranges, coherent anti-Stokes Raman scattering (CARS),
and optical phase conjugation (OPC). Let us also mention three-wave effects due
to (3) (4 = 3 +2 +0) in a constant field E0 , which breaks the central symmetry
of the medium.
(a) (b) (c) (d)

Fig. 6.23 Basic types of four-wave single-resonance parametric processes and their applications:
(a,b) generation of coherent UV and IR radiation; (c) coherent anti-Stokes Raman scattering (active,
or CARS, spectroscopy); (d) phase conjugation.
(a) (b)
(c) (d)
Fig. 6.24 Phase matching conditions in four-wave interactions: (a) generation of UV radiation is
usually performed under collinear phase matching; (b) in CARS, all frequencies are close and phase
matching is satisfied at small angles of scattering ; (c) in the general case of frequency conversion of
the form 1 + 2 3 4 , the phase matching has the form k1 + k2 = k3 + k4 . The wave vectors
in this case can be non-planar: the phase matching quadrangle can be folded along the dashed lines;
(d) in OPC, all four frequencies are equal and the phase matching condition is satisfied for any two
standing waves.
In the frequency conversion, it is usually the electron anharmonicity that

works, CARS is due to the mixed electron-nuclear (Placzeks) anharmonicity,
and OPC, similarly to self-focusing, is due to the inertial types of anharmonicity:
the orientation and sriction ones (Sec. 6.2).
Note that in OPC, which will be described in more detail below, all frequen-
cies are degenerate, as in non-parametric interactions. Nevertheless, we classify
it as a parametric process, since it preserves the field energy and leads to the co-
herent excitation of new modes with the phase matching condition k1 + k2 =
k3 + k4 = 0.
In the case of frequency conversion in gases, collinear phase matching is sat-
isfied due to anomalous dispersion (it is convenient to use mixtures of gases for
this purpose). CARS in many condensed materials requires non-collinear phase
matching with the angles of scattering about 1 (in the case of a two-beam pump,
see Fig. 6.24).
In all effects mentioned so far, the incident (input) field has, in the general
case, three modes excited (Ni0 , 0, i = 1, 2, 3), and the output radiation has
new photons in the fourth mode, k4 (1 + 2 + 3 4 ). Such effects can
be called stimulated, they are well described by classical electrodynamics. In ad-
dition, there are spontaneous effects, in which the pump contains, in the general
case, two modes: 1 + 2 3 + 4 . Then, photons in the output modes 3, 4 ap-
pear simultaneously, in pairs, due to spontaneous-stimulated transitions (compare
with the parametric down-conversion). Such processes, which include hyperpara-
Fig. 6.25 Active spectroscopy. The matter is excited by two lasers with frequencies 1 and 2 such
that 1 2 equals the frequency 0 of molecular vibrations. The resonance can be detected by an
increase in the intensity Ii , polarization, or phase of any one of the four frequencies observed while
scanning 1 or 2 .
metric scattering,r or light scattering by light, and spontaneous CARS, are only
described in terms of nonlinear quantum optics and, in some cases, equations like
the Kirchhoff law [Klyshko (1980)].
6.5.12 Nonlinear spectroscopy

The absolute value of (3) and, correspondingly, the efficiency of four-wave para-
metric interactions increase dramatically near intermediate resonances, where one
of the virtual levels coincides with a real one (Fig. 6.23). Most interesting are
two-photon resonances like 1 2 = 3 4 0 , in which neither the pump
not the output field undergo single-photon resonance absorption.
Observation of (3) resonances is the main instrument for several methods
of nonlinear spectroscopy, the most important of which is, probably, active
spectroscopy, based on CARS [Akhmanov2 (1981); Bloembergen (1977)]. The
scheme of a nonlinear spectroscope for the study of (3) dispersion is shown in
Fig. 6.25. It uses two lasers with the frequencies 1 , 2 (let 1 > 2 ), one of
them (say, 2 ) tunable within the range 1 0 , where 0 is the frequency of the
molecular vibration under study. At 2 = 1 0 , the frequency components of
the cubic susceptibilities (1 = 1 2 + 2 ), (2 = 2 1 + 1 ), (A =
21 2 ), (S = 22 1 ) have resonances (see (6.88)) caused by the exci-
tation of atom oscillations in a molecule by bi-harmonic light due to the mutual
influence of electronic and nuclear degrees of freedom (Sec. 6.2).
r Editors note: called now spontaneous four-wave mixing.
In the case of active spectroscopy, the resonance manifests itself in the in-
tensity variation at the new frequencies A or S . (It is convenient to use the
anti-Stokes range, A , where there is less stray light caused by the fluorescence
of the sample and the optical elements.) According to (6.261), (6.262), in the
undepleted-pump and plane-wave approximations, the intensity of CARS is
!4
4 2A l2 I12 I2 (3)
IA (2 ) = 2
|NR + (3) 2
R (A = 21 2 )| , (6.323)
c n1 n2 n A
where the non-resonance real part of (3) , (3)

NR , has a weak dependence on the fre-
quency and causes the asymmetry of the observed spectrum. Active spectroscopy
has advantages over usual Raman spectroscopy in sensitivity, resolution, and the
amount of available information. Note that using tunable lasers allows one to do
without dispersive elements.
In the case of Raman amplification spectroscopy, one registers the intensity
increase, I2 , at the output, equal, according to (6.224), to
3222 lI1
I2 (2 ) = Im(3) (2 = 2 1 + 1 ). (6.324)
c2 n 1 n 2
One can also observe the decrease of the intensity I1 of the field with the higher
frequency 1 near the resonance. This phenomenon is called the inverse Raman
effect.
The resonance can be also registered in other ways: by measuring the phase
delay of the incident field 1 or 2 , which scales as Re(3) (the Raman Kerr-effect
method), or by measuring the polarization parameters of the fields i , i = 1, 2, 3, 4
(nonlinear ellipsometry) [Akhmanov2 (1981)].
All these nonlinear spectroscopy methods, together with the two-photon spec-
troscopy and saturation spectroscopy, considered in Sec. 6.4, became a consider-
able extension of the traditional pre-laser spectroscopy, which mainly used linear
effects and spontaneous Raman scattering. A significant broadening of the scope
of spectroscopy due to the use of lasers together with the abilities of nonlinear
spectroscopy allowed one to speak about the laser revolution in spectroscopy.
6.5.13 Dynamical holography and phase conjugation

The idea of the optical phase conjugation (OPC) method based on the four-wave
interaction is clear from Figs. 6.24(d) and 6.26. Let a standing monochromatic
pump wave (a reference wave, according to holography termonology), i.e., k2 =
k1 , be excited in a medium with the cubic nonlinearity (3) ( = +). Then,
if a third plane wave k3 with the same frequency and an arbitrary direction is
Fig. 6.26 Phase conjugation using degenerate four-wave interaction. A nonlinear medium excited by
a standing pump wave reflects back all waves of the same frequency along the way of their incidence.
As a result, a divergent spherical wave becomes a wave convergent to the source (dashed line).
incident on the medium, the fourth wave will appear, with the frequency and
the wavevector k4 = k1 + k2 k3 = k3 . Thus, the medium excited by the pump
acts as a mirror reflecting all plane waves back along the path of their arrival, in
contrast to a usual mirror, which performs the transformation kz kz .
In the case of an arbitrary spatial distribution of the signal (object) field E3 (r)
it will contain many Fourier components {k3 }, each of them creating its own con-
jugate component. As a result, the initial object wave will be restored around the
medium, RE3 (r), with the same wave front shape, but propagating in the opposite
direction (from the medium) and, of course, having a different energy (R , 1). It
is important that due to the backward parametric amplification (see (6.287)), |R|
can considerably exceed the unity. (Usually, this is only achieved in the pulsed
regime.)
The possibility of the OPC effect for an arbitrary optical field, which, in a
sense, provides the time inversion,s looks striking from the viewpoint of nonlinear
optics. In fact, this effect has been discovered in the optical range as early as in
1949, long before the birth of lasers and nonlinear optics, by Gabor, the author
of holography. In holography, OPC manifests itself in the appearance of twin
images, which were considered by Gabor only as sources of noise. The perspec-
tives of OPC applications were understood only much later, mainly in the sev-
enties, when the practical methods of dynamical holography have been invented,
which enabled OPC to work in real time, without the delay for photographic
film developing. These methods, in addition to four-wave parametric interaction,
s OPC is equivalent to the t t transformation only for a strictly monochromatic field. In the case
of a quasi-monochromatic field, OPC does not change the shape of the envelope E0 (t).
(a)
(b)
Fig. 6.27 Applications of the OPC effect: (a) correcting the wave front distortions. The plane front
of the wave incident from the left becomes distorted, but after being reflected by a phase conjugating
mirror and passing the same medium again it restores its initial shape. (Reflection from a usual mirror
would double the distortions.) (b) Focusing of strong laser radiation on small targets: light from a
weak laser (top) is scattered by a target, part of the scattered field is amplified by a strong amplifier,
gets reflected by the mirror, is amplified once again and returns to the target.
use three-wave degenerate interaction effects, 180 light scattering, and superflu-
orescence [Zeldovich (1985); Bespalov (1979)]. OPC methods have been also
developed for acoustic waves.
OPC is an example of adaptive optics, aiming for the automatic correction
of optical systems. OPC allows one to correct the distortions of the wave front
(phase distortions) appearing due to the signal wave passing through an optically
inhomogeneous medium, like an opaque glass or a quantum amplifier. To correct
for this, it is sufficient to reflect the wave back at the output of the inhomogeneous
medium by means of a phase-conjugating mirror and this way to make it travel
along the same path in the opposite direction. Then, all distortions of the wave
front formed on the way will be straightened and the front will restore its origi-
nal shape (Fig. 6.27(a)). (Of course, amplitude distortions caused by irreversible
absorption or amplification will not be compensated but accumulated.) This ef-
fect allows powerful but inhomogeneous amplifiers to be used for increasing the
energy of weak lasers providing single-mode beams with the minimal (diffrac-
tion) divergence and natural bandwidth. This technique provides record values
of spectral brightness.
Another application of OPC, important for solving the problem of laser ther-
monuclear synthesis, is automatic focusing of strong laser radiation onto small
targets (Fig. 6.27(b)).
Using a resonance susceptibility (3) (in particular, enhanced by two-photon
resonances in the vapors of alkali metals) allows one to reduce the pump power in
four-wave interactions (down to 1 W/cm2 , with the reflection coefficient |R| 1).
In OPC via stimulated scattering, a sufficiently strong monochromatic wave
ReE(r) exp(it) serves as a pump causing 180 stimulated Raman scattering
or stimulated Mandelshtam-Brillouin scattering in a nonlinear medium. If the
wave front of the pump is rather non-uniform, the back-reflected Stokes radiation
ReES (r) exp(iS t) (with S = 0 ) has approximately the same wave front
shape, ES (r) RE (r), where |R| is close to unity.
At first stages of backscattering (z . l) the Stokes field is chaotic, it has all
modes with different wavevectors k excited independently and homogeneously.
(Since the wavevector length k = n(S )S /c is fixed, the modes are determined
by the transverse component k q). In the case of a multi-mode pump, var-
ious Stokes modes (q, S ) have different Raman amplification coefficients (q).
Moreover, one can show [Zeldovich (1985)] that if some mode (q, ) is present in
the pump spectrum, then the Stokes mode (q, S ) has, on the average, an ampli-
fication coefficient twice as large: (q) 2. Because at l 1 this difference
is amplified exponentially, it is sufficient that the part of the Stokes field repeating
the pump spectrum is considerably more intense than the noise part.
The advantages of Brillouin or Raman mirrors are the absence of the pump
(they are analogues of reference-free holography) and almost 100% efficiency.
The disadvantages are the existence of a threshold, the necessity to have multiple
modes in the signal field, and the frequency shift of the reflected radiation. The
latter restricts the accuracy of the reconstruction. Of much interest are lasers in
which one of the mirrors is phase conjugating (a Brillouin one) and the other
one, a usual mirror, plane or concave. This scheme simultaneously provides Q-
switching (due to the threshold behavior of the stimulated scattering) and correc-
tion for the optical inhomogeneity of the active medium.
Let us consider OPC via four-wave interaction in a little more detail. An
arbitrary signal field (for simplicity, considered as scalar) can be represented as
E(r, t) = ReE0 (r)eit = |E0 (r)| cos[t + (r)]. (6.325)
The phase-conjugated field describing monochromatic waves with the inverse di-
rections of wavefront propagation, by definition, differs by the sign of the time,
E(r, t) E(r, t). (6.326)

The amplitude and phase of the phase-conjugated field will be defined by analogy
with (6.325),
E(r, t) = ReE0 (r)eit = |E0 (r)| cos[t + (r)]. (6.327)
Certainly, the transformation E E is possible in practice only in the absence

of irreversible processes. From (6.325)(6.327), we find the relations between the
spectral amplitudes of the incident field and the phase-conjugate one,
E0 (r, t) = E0 (r), (r) = (r). (6.328)
Hence, the wave surfaces of the monochromatic fields determined by the equa-
tions (r) = const and (r) = const coincide. Thus, the OPC effect reverses not
the wave fronts but the propagation directions.t Note that reflection by a usual
mirror, plane, spherical, or of a more complicated shape, also leads to the trans-
formation E E, but only in trivial cases where the mirror surface coincides
with the wave surface.
The reversed field E copies the signal one in all space outside of the nonlinear
medium, including any transparent or scattering bodies of arbitrary shapes (pro-
vided of course that the scattered field fits the mirror aperture, see Fig. 6.26).
However, as we have already mentioned, if real, irreversible absorption (or ampli-
fication) is present, these are only phase surfaces that are restored; the amplitude
of the reversed field will be further reduced (or increased) on the way back through
the absorber (amplifier).
Let a signal field, ReE3 exp(i3 t), and a pump field, ReE1 exp(i1 t), be
excited in a nonlinear medium, which for the sake of simplicity will be considered
isotropic. An electromagnetic field present in a medium is accompanied by other
fields, for instance, pressure p(r, t), temperature, vibrations of molecules, excited
electrons, and so on (Sec. 6.2). In photographic materials, there are fields of
metallic silver or other products of photochemical reactions. In the simplest cases,
the amplitudes of these fields scale as the constant or slowly varying part of the
squared local field E 2 (r, t). For instance, due to the optical electrostriction,
p(r, t) ReE1 (r)E3 (r) exp[i(3 1 )t]. (6.329)
The pressure field (6.329) is a bulk hologram, it contains full information

about the signal provided that the pump field is known. Certainly, a record made
by pressure will be erased soon (within a relaxation time, /v, where v is the sound
velocity) after the signal is turned off, in contrast to the fields of photochemical re-
actions. However, if the signal is changed, the record will stabilize fast as well. In
t Editors note: the author used the term wave front reversal, which was more popular earlier.
the case of monochromatic fields with the same frequencies, p(r) is a static field
materializing the spatial distribution of the interference field E1 (r)E3 (r). This
simple model illustrates the idea of dynamical holography, which is used for the
study of fast processes.
The record (6.329) can be read out with the help of a second reference wave
ReE2 exp(i2 t), which will be scattered by the field n(r, t) of the refractive
index induced by the pressure field. In other words, the readout field E2 induces
polarization scaling as pE2 , and the field it emits is
E4 (r) P4 (r) = (3) E1 (r)E2 (r)E3 (r), (6.330)
where 4 = 1 + 2 3 . The relation between E4 and P4 in the Born approxima-
tion is given by Eqs. (6.260), (6.264). One can see from (6.330) that if the product
E1 E2 has a weak dependence on r, the OPC effect takes place, E4 E3 = E3 .
In particular, this is the case for a plane monochromatic standing-wave pump, for
which 2 = 1 and k2 = k1 . Note that in usual, static holography, since the
nonlinearity is inertial, it is necessary that 1 = 3 and 2 = 4 .u
Note that OPC can be also obtained via three-wave interaction,
P4 (r) = (2) E1 (r)E3 (r). (6.331)
However, in this case, only a part of the signal field E3 is reversed, whose fre-
quency and angular spectrum is within the phase matching band |k|l < 1,
k k1 k3 k4 , which leads to a loss of fine spatial and temporal details
of the signal. In addition, the phase-conjugated field E4 propagates from left to
right, similarly to the signal field, and an additional mirror is required for reflect-
ing it back. A huge advantage of four-wave interaction is the possibility to obtain
phase matching automatically using standing plane waves.
Let us estimate the efficiency |R| of OPC via degenerate (i ) interaction
using one-dimensional (diffraction-free) approximation in the undepleted-pump
regime. To this end, let us use Eqs. (6.286) for backward interaction, with an
obvious replacement of frequency subscripts. From (6.287) at z = 0, we find the
ratio of the incident and reflected amplitudes in modes k and k,
|R| |E4 (k)/E3 (k)| = tan l, (6.332)
where, according to (6.278), with the replacement (2) (3) E2 and E1 = E2 ,
2 (3) 323
= | E1 E2 | = 2 |(3) |I1 . (6.333)
cn cn 0
Let (3) = 1012 cm3 /erg, l = 1 cm, 0 = 1 , n = 1.5, then l = 1 at
I1 1 GW/cm2 .
u In the case of a bulk hologram (kl 1), both a static or a dynamic one, the phase matching condition
(Fig. 6.24(c)) leads to the equality of all four frequencies.

It is important that at l = /2, the conversion coefficient turns to infinity,

and then the fields E3 , E4 are emitted spontaneously. Thus, a standing wave in a
cubic medium has instability with respect to the parametric generation of back-
ward waves, in addition to the instability of traveling waves with respect to
self-focusing, self-modulation, and stimulated scattering. Note that at l = 1 self-
focusing, according to (6.236), will be not significant at
a2 P0 /I = l/4, (6.334)
i.e., in the case where the pump has considerable diffraction divergence in the
layer. (In this case, the validity of the one-dimensional approximation is also
violated.)

Chapter 7
Statistical Optics
In classical electrodynamics, the electric field E (r, t) E(x), x {r, t, } is

assumed to be deterministic and, in principle, measurable with an accuracy as
good as required. (We speak only about the electric field as it is the field that
determines the observable effects.)
In classical statistical optics, which includes the coherence theory as its impor-
tant part, E(x) for every x is considered as a random variable, with x being a pa-
rameter. It is convenient to split space and time in numbered cells and to consider
x as a discrete parameter spanning a countable number of values xi . Thus, a fluc-
tuating light field is described by a set of random variables Ei E(xi ). (Another
way of discretizing the field, the mode decomposition, is described in Sec. 7.3.)
All properties of the random set ensemble {Ei } are given by a multi-dimensional
distribution function, or a set of moments (correlation matrices) hE1 E2 Em i of
various orders m (the angular brackets denote averaging over the distribution func-
tion.) In experiment, the averaging is of course not over an ensemble of fields but
over a certain spatial and temporal interval Vdet . In addition, the field is filtered in
frequency and in the propagation direction.
From the viewpoint of statistical optics, the macroscopic Maxwells equations
are kinetic equations for the first moments hEi i, hHi i. The intensity and spec-
trum of light are determined by the second moments hEi E j i, while nquantum
processes are given by moments of order 2n.
However, classical statistics is, strictly speaking, not applicable to the optical
range, since the degeneracy factor hNi = [exp(~/T e f ) 1]1 , which has the
meaning of the average number of photons in one mode, or the spectral brightness
vac
I in ~c/3 I units, is usually much less than unity. For instance, for the
green part of the sunlight spectrum (T e f 6000 K, 0.5 ), hNi 0.01, and
it reaches the unity only in the IR range, at = 3.5 . Among the few excep-
tions there are laser fields, for which effective (brightness) temperature exceeds
237
the one of the sunlight by many orders of magnitude. In connection with this, let
us mention one of the paradoxes in the history of physics: quantum optics started
its rapid development only during the laser age, when light fields with hNi 1
appeared (although the general principles of quantum electrodynamics were de-
veloped much earlier).
In quantum optics and electrodynamics, an ensemble of fields is given by a
wavefunction or a density operator . The angular brackets in the definition
of the correlation function denote now quantum averaging over or ; then, Ei
are operators acting on according to certain rules. It is important that, in the
general case, the fields at the neighbouring points of space-time do not commute,
which leads to the quantum fluctuations of the field, to the moments hEi E j i being
nonzero even for the vacuum, to the spontaneous emission of excited atoms, and
to the noise of quantum amplifiers and generators.
The statistical theory of laser radiation studies the most important parameters
of lasers, such as the maximal possible monochromaticity of oscillators and the
sensitivity of amplifiers. Similarly to a consistent theory of thermal light emission
by heated bodies, this theory should be based on quantum electrodynamics and
non-equilibrium thermodynamics [Zubarev (1971); Klimontovich (1982)]. A spe-
cial difficulty in the quantum-statistical analysis of a laser is that the nonlinearity
plays a principle role, as it determines, due to the saturation effect, the stationary
amplitude of oscillations (the limiting cycle of a classical auto-oscillator).
The most important achievement in the nonlinear quantum theory of
lasers [Loudon (2000); Klimontovich (1980); Lax (1968); Arecchi (1974);
Klauder (1968)] is the conclusion that in the cavity of a laser operating well above
threshold, the field is in a coherent state, a concept introduced into quantum optics
by Glauber [Glauber (1965)]. There is a close analogy between a field in a coher-
ent state with a large amplitude and a classical harmonic oscillation and, since the
saturation effect manifests itself only in the case of high amplitudes, the nonlinear
regime of a laser is rather accurately described by the semi-classical theory. Non-
linear theories predict all statistical characteristics of laser radiation: the intensity,
the spectral width, the coherence radius, and the higher moments.
Even more crude, but still useful is the approximation considered in the linear
theory of noise in quantum amplifiers and oscillators (Sec. 7.1), which ignores
the saturation effect and is therefore valid only below the oscillation threshold.
The most important results of this theory are the Kirchhoff law, giving the noise
intensity of an amplifier in terms of its amplification coefficient, and the Townes
formula, relating the natural bandwidth of an oscillator to its power. According
to the linear theory, the radiation has Gaussian statistics, and therefore these two
Statistical Optics 239
parameters, the intensity and the spectral width, fully describe the statistics of the
field.
This chapter is devoted to the foundations of quantum optics. The consid-
eration starts with the linear theory of noise in quantum amplifiers, which does
not require the quantization of the field (Sec. 7.1). In Sec. 7.2, basic notions of
classical statistical optics are considered. The next section (Sec. 7.3) describes
the initial stage of the field quantization, i.e., writing the Maxwell equations in
the canonical form, which allows the quantization to be performed in an easy way
(Sec. 7.4). Section 7.5 considers the basic classes of quantum states of the field,
and Sec. 7.6 describes the statistics of photons and photoelectrons for these states.
Finally, in Sec. 7.7 we once again return to the question about the probability
of a transition due to a noise field (Sec. 2.4), but this time we consider it in the
framework of the quantum theory.
7.1 The Kirchhoff law for quantum amplifiers
The intensity of the noise radiation of a laser or a maser operating in a linear sta-
tionary regime can be calculated without using quantum electrodynamics, from
the general considerations based on certain rules of non-equilibrium thermody-
namics like the Nyquist formula (or the fluctuation-dissipation theorem, FDT)
and the Kirchhoff law for the thermal radiation.
7.1.1 The Kirchhoff law for a single mode

Let an ideal waveguide be filled by homogeneous matter of temperature T , ther-
modynamical or effective. Consider the electromagnetic energy carried by a single
waveguided field type (for instance, H01 in a rectangular waveguide) in the station-
ary case. Let the waveguide be connected with a matched source and a matched
load, so that the backward wave is independent of the forward one (Fig. 7.1).
We will be calculating the power spectral density P f (, z) P/ f , i.e.,
the energy of incoherent radiation with the frequency = 2 f > 0 transferred
through a cross-section z in 1 s within the frequency bandwidth 1 Hz. It is easy to
show that in ~ = h f units, P f (, z) corresponds to the mean number of photons
N(, z) per one longitudinal mode of the waveguide,
u hfu hfu
Pf = Ef = Nf = g f N = h f N. (7.1)
L L L
Here, u = d/dk is the group velocity; k is the propagation constant; L is the
length of some part of the waveguide, which is much greater than the wavelength
Fig. 7.1 Waveguide Kirchhoff law. The noise of the waveguide is determined by the competition
between the absorption and the spontaneous emission j = 0 N, where N = N2 /(N1 N2 ) is the
Planck function and Ni are the level populations. As a result, the noise can be expressed in terms of
the transfer coefficient G = ez and the temperature T .
= 2/k but much less than the distance at which P f (z) changes noticeably; E f
is the spectral energy density of the macroscopic field within L; N f E f /h f is
the number of photons within L per 1 Hz, which is equal to N times the spectral
density of the longitudinal modes g f = Ld1 /d = L/u. The inverse value, u/L,
is equal to the interval between the eigenfrequencies of the neighboring modes
(for the notion of a mode in more detail, see Sec. 7.3).
Let us start from the linear kinetic equation for N(, z) of the form
dN/dz = N + j. (7.2)
Here, () is the absorption or amplification coefficient due to stimulated transi-

tions and j() 1 0 is a distributed source of noise due to spontaneous emission.
This equation provides a phenomenological description of the stationary inter-
action of matter with a single transverse mode of the field, not necessarily in a
waveguide.a It is useful to compare this equation with one-dimensional equations
(6.171) for the slow amplitudes and with Einsteins relations (2.99) for non-
stationary interactions. Apparently, B(N1 N2 ) and j AN2 , where A, B are
the Einstein coefficients and Ni are the populations of a pair of levels separated by
an interval ~.
The solution to (7.2) has the form
N = N0 G + ( j/)(1 G), G(, z) exp[()z], (7.3)
where N0 is the signal at the input of the amplifier and G is the transfer coefficient
of the waveguide. Here, the second term, which is independent of the input signal,
is the noise of the waveguide.
a Suchequations are called transfer equations (for photons, neutrons, etc.), they imply that a space
coordinate z, defined up to an accuracy of L/2, can be attributed to a photon. In the general case, a
function N(k, r, t) is used, which has the meaning of local spectral brightness in ~c/3 units.
Let us express the ratio j/ in terms of the temperature of the matter. Let
z 1, then, according to (7.3), N = j. On the other hand, in a sufficiently long
waveguide with damping, equilibrium radiation should be formed, with the mean
number of photons per mode given by the Planck function N. Hence,
j/ = N [exp(~/T ) 1]1
= (1/2)[coth(~/2T ) 1] = N2(0) /(N1(0) N2(0) ). (7.4)
This conclusion is rigorous only for a completely equilibrium system where the
populations Ni and the photon number N obey the Boltzmann and Planck distri-
butions, respectively.
However, it is reasonable to assume that (7.4) is approximately valid even
in the absence of equilibrium radiation, for instance, at z . 1 and N0 = 0.
Then, the parameter T relates only to the matter, whose temperature is maintained
constant and uniform, despite the radiation cooling (quasi-equilibrium). For
this, it is necessary that the degrees of freedom that emit and absorb radiation at
frequency interact with the thermostat much stronger than with the field.b In this
approximation, Eq. (7.3) at N0 = 0 describes the spatial spreading of equilibrium
Plancks field (with a single propagation direction) in a waveguide of a layer with
the length z, temperature T and absorption coefficient (),
N = AN, A 1 G. (7.5)
Here, A is the absorptivity of the waveguide or the layer (in the general case, with
an account for reflections at the input and output). It can be measured, according
to (7.3), using an external signal: A = 1 dN/dN0 .
Equation (7.5) is the Kirchhoff law relating the thermal radiation of a heated
body to its thermodynamic parameter T and kinetic parameter A. It can be easily
derived in the form (7.5) without using the kinetic equation [Landau (1964)]. Note
that this law is used in the construction of photometry reference sources for light
intensity, although, in principle, it is only applicable within the approximations of
strong coupling with the thermostat, linear optics, and geometric optics. For its
generalizations, see Ref. [Klyshko (1980)].
7.1.2 The Kirchhoff law for a negative temperature

The phenomenological relation of spontaneous and stimulated effects with the
level populations, j/ = N2 /(N1 N2 ), is confirmed by calculations in the frame-
b Let thetime T 1 characterize the interaction of the particles with the thermostat; then the condition for
the Kirchhoff law (and, generally, FDT-type equations) to be applicable to non-equilibrium problems
is, apparently of the form 1/T 1 A(2N + 1), compare with (2.115).
Fig. 7.2 The Planck function determining the mean number of photons with the frequency > 0
in a single mode of an equilibrium field, as well as the number of excited atoms N2 divided by the
population difference N1 N2 . Here, T is the effective temperature; in the case of a field, T > 0.
work of the two-level model. It is then generalized to the case of a non-equilibrium

matter with non-Boltzmann Ni . In this approximation, it is also convenient to keep
relation (7.4) and the Kirchhoff law in the form (7.5), but to interpret T as the ef-
fective (spin) temperature T e f , which is determined by the actual population ratio,
exp(~/T e f ) N1 /N2 (Sec. 3.2).
From the definition of the function N(T ), it follows that in the case of pop-
ulation inversion, where and T are negative, N is also negative (Fig. 7.2),
N(T ) = [N(T ) + 1]; therefore, the Kirchhoff law can be written as
N = [N(T ) + 1](G 1),
or (see Fig. 7.3)
N = N{1 exp[0 ()z/(2N + 1)]}, (7.6)
since = (N1 N2 ) = 0 tanh(~/2T ), where 0 is the absorption coefficient
at T = +0.
In particular, in the case of full inversion, N1 = 0, T = 0, N = 1, so that
the spectral noise density of a single-mode ideal quantum amplifier in h f units is
simply equal to the transfer coefficient with the unity subtracted,
N = G 1. (7.7)
The 1 appears here because the noise source is a distributed one. Hence, the
noise relative to the input, N/G, is 1 G1 , which at G 1 yields one photon per
mode, i.e., one photon in a unity frequency band per unit time.
Fig. 7.3 The spectral density of the thermal radiation emitted by a single-mode attenuator (T > 0) or
amplifier (T < 0) as a function of its length z and temperature T , according to the Kirchhoff law. 0 is
the absorption coefficient at T = +0.
If an incoherent signal with the power spectral density h f N0 is present at the

input, one should add GN0 to (7.7). The result can be represented in the form
N + 1 = G(N0 + 1). (7.8)
This equation provides the following algorithm for taking into account the spon-
taneous emission (or, in other words, the quantum fluctuations) of an ideal ampli-
fier: a unity is added to the number of photons at the input, the result is multiplied
by the classical amplification coefficient, and then the unity is subtracted, which
yields the output number of photons. Thus, the signal-to-noise ratio at the output
of an ideal quantum amplifier with large amplification is equal to N0 , the number
of signal photons per mode.
The unities in (7.8) can be interpreted as a result of zero-point field fluc-
tuations; however, this interpretation should be treated with caution [Ginzburg
(1983)]. Quantum fluctuations determine the limiting sensitivity and accuracy of
quantum amplifiers, as well as any measurement devices. At N1 , 0, thermal
fluctuations, scaling as N(T ), are added to the quantum ones (see (7.6)).
In the microwave range, one often uses the concept of the noise temperature
of an amplifier, T n . By definition, T n is the brightness temperature of the noise
radiation at the output relative to the input,
N(T n ) N/G = N(T e f )(ez 1), (7.9)
where N is the Planck function for the chosen frequency . This equation relates
T n with the effective (spin) temperature of matter, T e f . The notion of T n is conve-

nient only in the classical temperature range, where N T/~ and (7.9) takes
the form
T n = (ez 1)T e f ,
which in the two limiting cases of small absorption and large amplification yields
T n = zT e f , |z| 1, (7.10)
T n = T e f , z 1. (7.11)
Thus, the spin temperature determines not only the ratio of the populations but
also the limiting noise temperature of a quantum amplifier.
In paramagnetic amplifiers, |T e f | is on the order of the lattice temperature of
the active crystal, which is cooled down to helium temperatures for the sake of
lowering T n , reducing spin-lattice relaxation, and increasing the equilibrium pop-
ulation difference. As a result, the noise temperature of such amplifiers is as low
as several Kelvins. At the same time, in ordinary electronic amplifiers, T n 103
K, due to the high temperature of the cathode; in parametric microwave amplifiers
based on semiconductor diodes, T n 100 K.
In practice, T n in paramagnetic amplifiers is determined by the thermal emis-
sion not from the active medium, according to (7.11), but from the elements of the
input channel (the aerial, waveguides, and ferrite devices), according to (7.10).
Indeed, an input waveguide with the walls at room temperature and the losses of
only 1% makes a noticeable contribution, T n 3 K.
It is not difficult to generalize the Kirchhoff law (7.5) to the case where the
waveguide has several sources of homogeneous linear loss, i , and noise, ji , with
different effective temperatures T i (i = 1, 2, . . . ). Assuming again ji = i Ni , we
obtain the kinetic equation
X
dN/dz = i (N N), (7.12)
i
P P
which only differs from (7.2) by the replacements i and N i Ni .
As a result, (7.5) takes the form
X
N = (1 ez ) (i /)Ni , (7.13)
i
P
with i . Thus, the contribution of each element to the total thermal emis-
sion scales as its contribution i / to the total absorption coefficient.
Consider now the shape of the thermal emission spectrum in the vicinity of
a single narrow resonance at various optical densities z (Fig. 2.4). When ||z is
small, the emission spectrum repeats the absorption spectrum, N() = ()zN
(the dependence N() is too slow to be taken into account here.) At large positive
z, the line gets flattened and broadened, tending eventually to the equilibrium
spectrum N(), and at large negative z the linewidth
is reduced (similarly to the
amplification band, see (2.68)) and tends to zero as ln G.
7.1.3 Noise of a multimode amplifier

We have so far discussed a single transverse mode of a waveguide. In the general
case, where the radiation is delivered to the detector as several independent modes,
(7.5) or (7.12) should be summed over all essential modes. If the waveguide cross-
section A is much greater than 2 , summation can be replaced by integration.
In free space, the number of transverse modes (i.e., modes with the same fre-
quency k = ck) observed by a pointlike detector in the near-field zone scales as
the spatial aperture of the emitter, A ab, and the angular aperture of the detector,
= x y ,
g = (a1 1 2
x )(by ) = A/ , (7.14)
where = 2k, we consider only a single polarization type and assume that the
observation direction is orthogonal to A. According to (7.14), the angular interval
between the neighboring transverse modes is on the order of the diffraction angle
/a.
Multiplying (7.5) by the photon energy ~, the detector band f = /2,
and the number of transverse modes g , we find the power of multimode thermal
radiation with a single polarization type,
P = (~A/22 )N(1 G), (7.15)
where we assume that the transfer coefficient G = ez and the effective tempera-
ture T are the same for the modes having close directions.
The ratio P/A is called the spectral brightness; this is the main
energy characteristic of incoherent multi-mode radiation. In the general case,
I I(k, r) depends on the frequency, direction, and the observation point; how-
ever, in the absence of scattering, emission, and absorption between the points
r and r + a, the spectral brightness at these points is the same provided that the
argument k is parallel to a,
I(k, r + a) = I(k, r), k k a. (7.16)
Certainly, it makes sense to speak about the displacement a of the observation

point only at a .
In photometry, there are special terms for the integrals of I w.r.t. various vari-
ables: the brightness, the radiance, and the luminous intensity,c
Z Z
B( k, r) dk I(k, r), E(r) dB( k, r) cos k ,
0
Z
P(z) dxdyE(r). (7.17)
Here, k k/k, k is the angle between k and the z axis. The spectral and volume
energy density is
Z
1
(, r) = dI(k, r), (7.18)
u() 4
where u is the group velocity, k . In the case of the thermal radiation from
a heated body, I(k, r) cos k is called the emittance of a body (at point r of its
surface).
According to (7.15), the Kirchhoff law for the multimode emission from a
matter layer with the thickness z and the effective temperature T has the form
(compare with (6.225), (6.307))
vac
I (k) = I N(k), (7.19)
where
vac
I ~/22 , N(k) = N(1 e(k)z ), (7.20)
vac
where we take into account only a single polarization type. The value I corre-
sponds to the spectral brightness of radiation having a single photon in each mode
vac
(i.e., I is the doubled brightness of zero-point vacuum fluctuations), it is a
natural unit for the measurement of I : the mean number of photons per mode
vac
N is equal to the spectral brightness in I units.
7.1.4 Equilibrium and spontaneous radiation; superfluorescence

Consider three typical cases:
vac
1. At z 1, (7.19) leads to I = I N, which, in the case of isotropic
radiation, corresponds to the blackbody radiation and, according to (7.18), at u =
c/n, to the Planck formula,
2 = (8n/c)I = (8~/3)N; (7.21)
the factor 2 accounts here for two polarization types.
c Sometimes the attribute energy is added, indicating that they are measured in physical units and not
in light-engineering ones. The latter take into account the spectral characteristic of the human eye and
are based on the candela unit.
2. If ||z 1, then (7.19) leads to
vac vac
I = I Nz = I N2 l/ cos k , (7.22)
where the optical density of the layer z for the mode k was assumed to be (N1
N2 )l/ cos k . Let us write the transition cross-section in terms of the transition
dipole element d0 according to (2.52) and find, using (7.17), the total power of
spontaneous emission. Let n = 1, then
Z Z Z Z
P= d dxdy cos k dI = (42/~c)I
vac
N2 V0 d(d0 e k )2 ,
4 A 0 4
(7.23)
where V = Al is the matter volume. The integral here is so that P/VN2 8d02 /3,
coincides with the expression found above for the power of the spontaneous emis-
sion per one molecule, 440 d02 /3c3 (see (5.32)).
3. Now, let z 1, then N = GN, and
P = N2 (N2 N1 )1 I
vac
G0 Ae f e f e f , (7.24)
where we have introduced the effective apertures Ae f , e f and the frequency

bandwidth e f , which can be rigorously defined in terms of the correspond-
ing integrals, and the amplification coefficient at the center of the line, G0 . This
equation determines the power of superfluorescence, the amplified spontaneous
emission. Assume, for a crude estimate, that Ae f coincides with the cross-section
A of the amplifier, e f coincides with the amplification bandwidth , and
e f A/l2 , where l is the length of the amplifier, then at N1 = 0
vac
P = I G0 A2 /l2 . (7.25)
At = 0.5 , = 10 nm, A = 1 cm2 , l = 10 cm, and G = 103 , we obtain

P = 2 104 W.
For many transitions in gases, inversion can be obtained only for a short time,
. 109 s, even with the help of very powerful lasers. This is due to the extremely
long lifetime of the lower level. Then, if the active area length is larger than
c 30 cm, a mirror-based feedback is too slow to have an effect. The radiation
of lasers using such self-restricted transitions is, in fact, superfluorescence. An
important example is the nitrogen laser with the wavelength 330 nm.
Certainly, at sufficiently high G0 the superfluorescence power (7.25) can be
large enough for saturation; then the problem becomes nonlinear and the Kirchhoff
law is not valid any more.
Fig. 7.4 Cavity quantum amplifier: (a) in the optical range; (b) in the microwave range; (c) an
equivalent circuit.
7.1.5 Gain and bandwidth of a cavity amplifier

For increasing the gain at a given length of the active medium, one can use a
positive feedback provided by a bulk cavity. With the help of an equivalent circuit,
let us find the gain coefficient G() and the transmission band of a reflective
cavity amplifier, i.e., an amplifier with a single coupling waveguide or a single
semi-transparent mirror (Fig. 7.4). In the microwave range, the amplified reflected
wave is separated from the incident weak signal by means of non-reciprocal ferrite
devices (circulators).
Consider the spectral range in the vicinity of some eigenfrequency 0 of the
cavity. If the corresponding oscillation is non-degenerate and 0 is well separated
from other frequencies, then the field at an arbitrary point of the cavity depends
on the time similarly to the parameters of any other oscillator system with a single
degree of freedom, for instance, like the current in an LC contour. Then the elec-
tric field has the form E = u(r)q(t), where u(r) is a known function determined by
the shape of the cavity and q(t) obeys the equation of motion for a harmonic oscil-
lator. This allows the cavity response to be calculated using the equivalent circuit
shown in Fig. 7.4(c). Damping (and, according to the Nyquist theorem, noise
as well) are introduced into the circuit by three resistances: the external load Rl
(which, by definition, coincides with the impedance of a transmission line corre-
sponding to the coupling waveguide), the resistance R0 scaling as the losses in the
cavity walls, and the negative resistance Ra scaling as the emission of the active
medium. The width of the chosen spectral line is assumed to be much greater than
, and therefore Ra is a real parameter.
In the case of a transmission line, the reflection coefficient, i.e., the ratio of
the complex amplitudes for the backward and forward waves, is determined by
the ratio of the load (cavity) resistance Zc to the impedance of the line Rl . For
instance, the reflection coefficient for the current is
Rl Zc Rl Rc iX
K() = = , (7.26)
Rl + Zc Rl + Rc + iX
where
2
1
L = L 20 1 2L(0 ).

Rc R0 + Ra , X (7.27)
C
The relation between Rl and the other parameters depends on the coupling between
the cavity and the waveguide.
Let us pass from the equivalent parameters Ri to dimensionless variables that
have a direct physical meaning: the Q-factors Qi 0 L/Ri or the damping con-
stants di 1/Qi . The value Qa is called the (magnetic) quality factor of the active
medium. After dividing both the numerator and the denominator of (7.26) by 0 L,
we find
dl dc 2ix
K() = , (7.28)
dl + dc + 2ix
where x (0 )/0 . The gain (or attenuation) of the reflected wave power is
(dl dc )2 + 4x2
G() = |K()|2 = . (7.29)
(dl + dc )2 + 4x2
It is important that the coupling between the waveguide and the cavity can
be varied, for instance, by changing the orientation of the feedback loop or the
transmission of the mirror. Then, the normalized losses dl = Rl /0 L due to the
external circuit will also vary from zero (no feedback, a totally reflecting mirror)
to infinity (maximal feedback, no mirror). In the case of a resonance ( = 0 ) and
positive losses in the cavity (dc > 0), variation of the feedback strength from 1
to +1 leads to the variation of the amplitude reflection coefficient for the current
K0 . The dependence passes through zero at the point of load matching, dl = dc
(Fig. 7.5).
If dc = d0 + da < 0, then by decreasing the feedback strengths one makes K0
and G0 vary from 1 to . Thus, a cavity quantum amplifier with a sufficiently
high-quality cavity (Q0 > Qa ) enables a weak signal to be amplified as much as
possible due to the regeneration.
However, an increase in K0 is accompanied by a decrease in another parameter,
important for many applications, namely, the amplification bandwidth . As we
will show below, at K0 1 the product of the amplitude gain and the bandwidth
does not depend on the feedback,
K0 = 20 /|Qa |. (7.30)
The product K0 is called the regeneration parameter. It characterizes the qual-

ity of the active medium for paramagnetic amplifiers.
Fig. 7.5 The gain coefficient K0 of a cavity amplifier as a function of the cavity quality factor 1/dc .
By reducing the feedback dl , at a fixed dc < 0 one can make K0 as high as desired. Here, dc is the
cavity damping constant with an account for the negative contribution from the active medium; dl is
the damping of the load, i.e., the relative losses due to the external circuit.
According to (7.29),
dc dl
G1= , (7.31)
d2 /4 + x2
where d dl + dc 1/Q is the total damping constant and Q is the quality factor
of the loaded cavity. It follows that at G 1 the frequency characteristic of the
amplifier has a Lorentzian shape and the amplification bandwidth is determined
by the quality factor of the loaded cavity with an account for all losses,
1 1 1
=d= + + . (7.32)
0 Ql Q0 Qa
For achieving G 1, a nearly full compensation of all losses is necessary,
da d0 + dl , in which case K0 2|d p |/d, so that the product of the gain and
the bandwidth is a constant, K0 = 20 |dc |. Usually, the quality of the cavity,
Q0 & 104 , is much greater than the quality of the active medium, |Qa | . 103 ,
hence, in practice, Qc = Qa , which yields (7.30).
Let us write Qa in terms of the parameters of the matter by using the general
definition of the quality factor as the ratio of the energy stored in an oscillation
contour,
Z Z
0 2
E= 3 2
d r( E + H )/8 = d3 r 0 u2 (r)/8, (7.33)
Vc Vc
to the energy of losses during the time 1

0 ,
Z
P/0 = d3 r 00 (0 )u2 (r)/8. (7.34)
Vc
Here, we have used the equalities Eel = Emag , q2 (t) = 1/2, and Eq. (4.15) for the
losses per unit volume. Hence,
Q1 00 0
a = / = /k, (7.35)
where V/Vc 0 1 is the factor showing the proportion of the cavity filled
by the active medium and determined by the ratio of the integrals in (7.33) and
(7.34); = 1 + 4, is the linear amplification factor for plane waves with the
wavevector k; in the case of a magnetic transition, 00 should be understood as the
magnetic permeability. At electron paramagnetic resonance, 5 102 cm1
(Sec. 2.3), which yields, with = 3 cm and 0 = = 1, Qa = 40. Let K0 = 100,
then it follows from (7.30) that f = 5 MHz. This bandwidth is not sufficient for
many applications (communications, radar, radio astronomy); for this reason, one
uses several coupled cavities and slowing-down systems.
7.1.6 The Kirchhoff law for a cavity amplifier. The Townes equation
In order to take into account the noise of the active medium, let us add to the
equivalent circuit (Fig. 7.4) a source of current with the spectral density given by
the Nyquist equation (see Ref. [Akhmanov (1981)]),
i2f = 4T Ra /|Z|2 , (7.36)
where Z = Rl + Rc + iX and T is the effective (spin) temperature of the active

medium. (We neglect the noise related to Q0 .) For taking into account quan-
tum noise, T should be replaced by ~N. The load resistance will produce the
spectral power density P f = i2f Rl , which is, in ~ units,
N = 4Nda dl /(d2 + 4x2 ) = N(1 G). (7.37)
Here, we used (7.31) in the approximation Q0 |Qa |. Thus, we have again,

according to the Kirchhoff law, expressed the noise in terms of the effective tem-
perature and the gain. The width of the noise spectrum n at half maximum,
according to (7.37), coincides with the amplification bandwidth (see (7.32)),
and hence, it tends to zero at dc dl . Similarly to the case of a traveling-wave
amplifier, at G 1 and ~ T the noise temperature has the same absolute
value as the spin one.
Let us find the full noise power at dc da dl and 0 ,

~30 N ~30 |N|
Z Z
d
P= d f ~N = = . (7.38)
0 2Q2a 0 2 /4 + (0 )2 Q2a
This leads to the Townes equation giving the spectral width of the emission of a
cavity amplifier in terms of the total radiation power,
~30 N2
= . (7.39)
PQ2a (N2 N1 )
Let = 1 , P = 1 mW= 104 erg/s, N1 = 0, 1/oQa = 0.01 cm1 , then
f = ~c3 2 /P 1Hz. (7.40)
Certainly, this calculation of is valid only for the linear regime of the
amplifier below its oscillation threshold (Ql < Qa ). However, one can expect
that even above the threshold, (7.39) provides a correct order of magnitude for
the bandwidth of a quantum oscillator (in the limiting case where the technical
noise, like mirror vibration, is absent). More accurate calculations, taking into ac-
count the nonlinearity due to the saturation [Loudon (2000); Klimontovich (1980);
Lax (1968); Arecchi (1974); Klauder (1968)], which restricts the noise amplitude,
lead to a factor 1/2 appearing in (7.39) for the regime high above the threshold.
This can be explained by the suppression of amplitude fluctuations, which give
the same contribution to the spectral width as the phase fluctuations. One should
keep in mind that the random slow phase variation (phase drift) of a harmonic sig-
nal leads to the fluctuations of its frequency and to a finite width of its spectrum.
The difference between the oscillation shapes of a quantum amplifier (oscillator)
below threshold and above it is explained in what follows (Fig. 7.7).
It is only the absence of amplitude fluctuations that qualitatively distinguishes
the radiation of a single-mode laser from a narrow-band noise. In the case of a
multi-mode laser with independent modes, this difference vanishes as well.
7.2 Basic concepts of the statistical optics
Above, we have found, in the form of the Kirchhoff law, the spectral brightness
of the radiation I(k) as a function of the frequency and the observation direction
in the far-field zone. Although the intensity is an important characteristic of the
radiation, clearly, it does not provide the complete statistical information about
the electromagnetic field.
In the present section, we will consider, using several typical experimental

schemes, the observed quantities and the corresponding convenient theoretical
tools, the correlation functions G(n)
1...2n .
Because the most part of observable optical effects do not require the quanti-
zation of the field for their interpretation, we will restrict our consideration to the
more simple classical theory, and only in some cases the results of the quantum
theory (Secs. 7.47.7) will be given without the derivation.
7.2.1 Analytical signal

In statistical optics, it is convenient to use, instead of the real field E (r, t), a
complex function E(+) (r, t), which is called the analytical signal or the positive-
frequency field. It is unambiguously defined as
Z
E (+) (t) deit E(), (7.41)
0
where E() is the Fourier transform of the real field,

Z
E() dteit E(t)/2 = E (). (7.42)

By definition, the spectrum of the analytical signal E (+) (t) contains only pos-
itive frequencies ( > 0) while the spectrum of the complex conjugate (negative-
frequency) field E () = E (+) , only negative frequencies. From the definition, the
relation follows (Fig. 7.6)
E(t) = E (+) (t) + E () (t) = 2ReE (+) (t). (7.43)
(a) (b)
Fig. 7.6 The analytical signal E (+) (t) and the real field E(t) in the case of a monochromatic spectrum
(a) and in the general case (b).
In the case of a quasi-monochromatic field with a narrow spectrum, it is con-

venient to introduce a slow complex amplitude E0 (t), defined as
E(t) ReE0 (t)eit , (7.44)
where is some mean frequency. The absolute value |E0 (t)| is called the envelope
and the argument is called the slowly varying phase.The spectrum of E0 (t) is lo-
cated within the interval /2 around the zero frequency. Apparently, one can
write that
E (+) (t) = (1/2)E0(t)eit . (7.45)
The readings of optical detectors, such as PMTs, bolometers, photographic
films, etc. depend on the square (or higher powers) of the field averaged over a
certain time T due to the non-instantaneous response of the detector. (So far, we
do not take into account the finite spatial dimensions of the detector.) Assume that
1/T , i.e., the envelope is much broader than the detector response
time, which, in its turn, includes a large number of light periods. Then the reading
of the detector will scale as the instantaneous intensityd
I(t) E 2 (t)/2 |E (+) (t)|2 = |E0 (t)|2 /4. (7.46)
In Sec. 7.7, this statement will be justified more rigorously. By using here the
analytical signal, we automatically eliminate the terms oscillating with the double
optical frequency.
7.2.2 Random intensity

In a random field, the instantaneous intensity I(t) at point r varies: it fluctuates
both in time and in space.e Often, one measures only the simplest characteristics
of the field: the mean (ensemble-averaged or time-averaged) intensity at a single
point (E (+) E(+) (r, t)),
hIi = hE () E (+) i, (7.47)
and the mean square of the intensity,
hI 2 i = hE ()2 E (+)2 i, (7.48)
or the variance,
hI 2 i h(I hIi)2 i = hI 2 i hIi2 . (7.49)
d Sometimes we will omit the scaling factor, which in the case of a plane wave is c/2.
speaking, one should consider the random tensor I E() E(+) , where , = x, y, z or, in
e Strictly
the case of directed radiation, , = x, y.

In principle, one can measure the parameters (7.47)(7.49) by observing the mean
value and the variance of the readings of a broadband single-quantum detector (see
below). The readings of an n-quantum detector directly yield the n-th moment,
hI n i G(n) .
We will assume the field to be stationary, so that the mean values in (7.47)
(7.49) do not depend on the time, and ergodic. The angular brackets may then
denote the averaging either over time or over an ensemble (with some distribution
function P(I)).
After passing to the quantum theory (Sec. 7.4), E (+) is replaced by the operator
E , which is written in terms of the photon annihilation operators ak , while E ()
(+)
is replaced by the operator E () , which is written in terms of the photon creation

operators ak . Complex conjugation is then replaced by Hermitian conjugation, and
the angular brackets denote quantum averaging with the help of a wave function
or a density matrix. Most quantum states of the field also allow averaging using
the quasiprobability distributions P(z) (Sec. 7.5), which is similar to the classical
averaging. In quantum averaging, the order of the operators matters; the ordering
presented in (7.47), (7.48) is called the normal one. Of course, in the classical
theory the normal moment G(n) hE ()n E (+)n i can be replaced by h|E (+) |2n i.
In the vast majority of cases, the optical field is emitted by many independent
sources with random amplitudes and phases, such as, for instance, in the case
of the thermal radiation from heated matter, a quantum amplifier (Sec. 7.1), or a
multimode laser with independent modes. Then, the distribution of the complex
amplitude E0 E00 + iE000 is a normal (Gaussian) one, with independent E00 and
E000 , while the intensity distribution is exponential (Fig. 7.7),
PT (I) = hIi1 exp(I/hIi). (7.50)
Thus, the mean intensity hIi fully determines the statistics of a stationary
chaotic field (at a single point and for a single polarization type). Note that the
most probable intensity value is zero, P(0) 1 P(I). Using (7.50), one can easily
find the moments and the variance of the intensity,
G(n) n 2 2
T = n!hIi , hI iT = hIi , (7.51)
where the subscript T indicates that the field is chaotic (thermal).
Another typical case is the radiation of a single-mode laser with a stabilized
amplitude (Fig. 7.7). Then,
P(I) = (I I0 ), G(n) = I0n , I = 0. (7.52)
Equations (7.50)(7.52) do not take into account the discreteness of possible
energy values, i.e., they ignore the photon structure of the field; therefore, they
(a) (b) (c)

Fig. 7.7 Two basic types of the field states. The figure shows the approximate shape of the field vari-
ation in time from the classical viewpoint (a) and the corresponding distribution functions P(|E0 |): the
classical ones (b) and the quantum ones (c). Top: the quasi-monochromatic field of thermal radiation,
of a quantum amplifier or oscillator below the threshold (both the amplitude and the phase fluctuate);
bottom: the field of an above-threshold quantum oscillator (the amplitude fluctuations are suppressed
due to the saturation effect).
are only valid for classical fields for which the degeneracy factor hNi (see below)
is much greater than the unity. The general case will be considered in Sec. 7.6.
So far, let us only note that in the quantum theory, the continuous distribution
(7.50) is replaced by the discrete exponential one, while (7.52) is replaced by
the Poissonian distribution (Fig. 7.7(c)).
7.2.3 Correlation functions

The distributions or intensity moments considered above provide no information
about the correlations between the fields at neighboring points in space and time,
or between the different Cartesian components of the field. Complete informa-
tion is given by the set of multi-dimensional distributions or tensor correlation
functions (CF). The latter, according to Glauber [Glauber (1965)], are defined as
G(n) () () (+) (+)

1...2n hE 1 . . . E n E n+1 . . . E 2n i. (7.53)
(Here we use the notation that is also valid in the quantum description.) Each
subscript denotes a set of arguments, for instance, E1 E1 (r1 , t1 ). For simplicity,
we only consider CFs with even numbers of fields, since in optics the moments
of the form hE1 E2 E3 i are usually equal to zerof and, in addition, are difficult to
measure.
f Exceptions are fields at the output of nonlinear media excited by external radiation [Perina (1972);
Akhmanov (1971)] or heating [Klyshko (1980)].

The CFs of a stationary field are invariant to the choice of the initial time
moment, i.e., to the replacement of time arguments
t1 , . . . , t2n t1 + t, . . . , t2n + t, (7.54)
where t is arbitrary. It is convenient to choose t t1 .
As a result, G(n) depends on 2n1 time arguments, while its Fourier transform,
the spectral CF G(n) , depends on 2n1 frequencies. For instance, the first-order CF
has the form G(r1 , r2 , ); it is called the mutual coherence function of the field at
points r1 , r2 , while its Fourier transform, G(r1 , r2 , ), is called the mutual spectral
density. At r1 = r2 , the first-order CF, G(r, t), is called the autocorrelation func-
tion, while G(r, ) is the spectral density. Traditional polarization characteristics
of directed radiation, such as degree of polarization or the Stokes parameters, are
also determined by the first-order CF with an account for tensor indices [Perina
(1972)]. With all arguments coinciding, CFs become intensity moments (we omit
the tensor indices),
G(n) (+) 2n n
1...1 = h|E (r1 , t1 )| i = hI (r1 )i. (7.55)
Among the various statistical models of the field, a special role belongs to
the Gaussian model, in which all CFs are expressed in terms of the first-order
CF [Glauber (1965)],
0
X
G(n)
1...n10 ...n0 = G(1) (1)
110 G nn0 , (7.56)
P0
where denotes the sum of all n! permutations of the primed indices. For in-
stance,
G(2) (1) (1) (1) (1)
1234 = G 13 G 24 + G 14 G 23 . (7.57)
All information about a Gaussian (chaotic, thermal) field is contained in the
first-order correlation function G (r1 , r2 , ) or the mutual spectral density
G (r1 , r2 , ).
For a rough characterization of a quasi-monochromatic directed non-polarized
Gaussian field, it is sufficient to fix at each point the intensity G (r) and the co-
herence parameters coh 1/, coh for both polarization types = 1, 2. The
meaning of these parameters will be explained in the next sections.
7.2.4 Temporal coherence

Consider the field E 0 (t) at the output of a Michelson interferometer (Fig. 7.8). It
consists of two terms differing by a certain time delay, = t0 t,
E 0 (t) = [E(t) + E(t0 )]/2,
Fig. 7.8 The Michelson interferometer and the measurement of the longitudinal coherence length.
Fig. 7.9 Typical dependence of the mean intensity at the output of a Michelson interferometer, hI 0 i,
on the position of the moving mirror: g() is the normalized autocorrelation function of the field at the
input.
where E(t) is the field of the plane wave at the input of the interferometer.
According to (7.46), the intensity at the output is
I 0 (t) = [I(t) + I(t0 ) + 2ReE () (t)E (+) (t0 )]/4.
Hence, in the case of stationary radiation,
hI 0 i = [hIi + ReG()]/2 = hIi[1 + Reg()]/2, (7.58)
where
G() hE () (0)E (+) ()i = hE0 (0)E0 ()ieit /4 (7.59)
is the auto-correlation function of the field at the output, which is related, ac-
cording to the Wiener-Khinchine theorem, to the spectral density G() via the
Fourier-transformation, g() G()/hIi is the normalized CF, and hIi = G(0).
Note that in optics, CFs are usually normalized to the mean values and not to the
standard deviations I.
According to (7.58), the dependence of the intensity at the output of a Michel-
son interferometer on the delay time determines the real part of the first-order CF.
One can show that the function G(), similarly to E (+) (), is analytic in the lower
semi-plane. Therefore, its real and imaginary parts are related via the Hilbert
transformation, so that, in principle, from the interference pattern one can restore
the radiation spectrum. This method forms the basis for the Fourier spectroscopy.
Besides, the interference of the field under study with a reference coherent wave
is the source of information in holographic interferometry.
A typical interference pattern in the case of a single spectral line is shown in
Fig. 7.9. The relative amplitude of the oscillations in the output intensity is called
the interference visibility or the degree of coherence. According to (7.58), (7.59),
the visibility coincides with |g()|.
If we define the coherence time coh by the condition |g(coh )| = 1/2, then
it follows from the properties of the Fourier transformation that coh 2/.
Hence, the longitudinal coherence length ccoh is on the order of the inverse spec-
tral width in inverse centimeters,
lcoh 2c/ 1/. (7.60)
For instance, for lines with natural broadening in the visible range,
2 106 s1 , lcoh 105 cm. This can be viewed as the length of the wave train
emitted by an atom during spontaneous emission. In the case of a single-mode
laser, estimation with the Townes equation (7.40) yields a coherence length on the
order of a light second.
7.2.5 Spatial coherence

The correlation of fields at two points r1 and r2 can be measured with the help of
the Young interferometer, which is a screen with two pinholes placed orthogonally
to the direction towards the source (Fig. 7.10).
The interference pattern at an arbitrary point behind the screen, similarly to
(7.58), is determined by the first-order CF of the general form:
G12 () G(r1 , r2 , ) hE () (r1 , t)E (+) (r2 , t + )i, (7.61)

Fig. 7.10 Youngs interferometer and the measurement of the transverse coherence radius.
while the pattern on the symmetry axis is determined by G12 (0). The coherence
radius coh is determined by the distance between the pinholes at which the visi-
bility is 50% reduced.
An ideal laser with a single transverse mode emits plane or spherical waves,
for which coh = . In the case of a multi-mode laser, or a chaotic (thermal) light
source, coh in the far-field zone (z a2 /) is determined by the transverse size a
of the source and the distance z from the source (Fig. 7.10),
coh z/a zd /a , (7.62)
where d is the diffraction angle and a is the angular size of the source. Due
to diffraction, the transverse coherence radius increases in the course of light
propagation. The shape of the phase front in this case increases and tends to
the spherical one. The van Cittert-Zernicke theorem [Akhmanov (1981); Klauder
(1968); Perina (1972)], which describes this effect quantitatively, states that the
dependence of the first-order CF on (r1 r2 ) is given by the Fourier transform
of the brightness distribution over the source cross-section.
Relation (7.62) enabled Michelson to measure, with the help of his stellar
interferometer, the angular diameters of several stars, for which a & 107 rad and
coh . 10 m. At smaller a , the coherence radius is determined by the atmospheric
distortions of the wave front, which hinder the operation of the interferometer. The
Hanbury BrownTwiss interferometer (see below) is free of this drawback.
7.2.6 Coherence volume and the degeneracy factor

The coherence volume is defined as the product of the coherence area, 2coh , and
the coherence length, lcoh . For the far-field zone of a chaotic source, it follows
from (7.60) and (7.62) that

Vcoh = 4 /a , (7.63)
2 2
where = /lcoh and a (a/z) is the solid angle at which the observer sees
the source.
An important dimensionless statistical parameter of the radiation is the de-
generacy factor, which is the mean energy of the field (for a single polarization
mode), in ~ units, contained in the coherence volume. In other words, the de-
generacy factor is the number of photons crossing the coherence area during the
coherence time,
hEicoh /~. (7.64)
Let us write hEicoh in terms of the radiation spectral brightness,
hEicoh = I a coh 2coh = 22 I . (7.65)
Hence,
= 3 I /~c hNi. (7.66)
3
Thus, the degeneracy factor is equal to the spectral brightness in ~c/ units.
In other words, the mean photon number per coherence volume coincides with
the mean number of photons per single mode, hNi. In an equilibrium field, =
N() [exp(~/T ) 1]1 . In a non-equilibrium field, the temperature should
be interpreted as the brightness temperature, T e f .
The degeneracy factor = hNi is not only a convenient measure of the spectral
brightness; it also defines the applicability of classical statistics: at . 1 (i.e.,
~ . T ) the photon structure of the field becomes important.g
Let us divide the far-field space of a quasi-monochromatic source into cells
with volumes Vcoh . By definition, the radiation at any two points of the same
cell is mutually coherent. Hence, the field within one cell can be approximately
considered as single-mode, i.e., assumed to be a spherical monochromatic wave
with a definite amplitude |E0 | and phase . Passing from one cell to another, we
will observe random fluctuations of |E0 | and . Thus, the spatial distribution of a
stationary field forms the ensemble of harmonic-oscillator states.
Fields belonging to different coherence volumes are, by definition, uncorre-
lated and, generally, independent. Hence, by virtue of the central limit theorem of
g Editors note: It may seem that the degeneracy factor can be considered as a measure of nonclas-
sicality. However, this is not true, and this is not stated in the book. At 1, the shot noise, which
is related to the photon structure of light, prevails over excess intensity fluctuations and therefore be-
comes important. At the same time, there are nonclassical states of light with 1 (squeezed states,
squeezed vacuum).
Table 7.1 Basic distribution types for the photon number and the energy.
Quantum theory Classical theory

State Number of modes
1 1 1 1
Coherent (laser) Poissonian Poissonian (E E0 ) Gaussian
Chaotic (thermal) Geometric Poissonian Boltzmanns Gaussian
K-photon (mixed) PK = hNi/K, Poissonian
P0 = 1 P K
the probability theory, the energy distribution in a volume V that is much greater
than Vcoh will be Gaussian, with the variance inversely proportional to the number
of cells V/Vcoh . In the quantum case, where . 1, the Gaussian distribution for a
multi-mode field is replaced by the Poissonian one (Sec. 7.6)
Table 7.1 shows some types of distributions for the energy or the number of
photons. The states that are called K-photon ones have no classical analogues
and manifest photon anti-bunching and bunching (Sec 7.6).
7.2.7 Statistics of photocounts and the Mandel formula

The intensity distribution for the field at one point can be measured by means
of a PMT operating in the photon-counting mode.h Then the mean intensity of
light hIi should be sufficiently low, so that the photocurrent pulses at the PMT
output do not overlap (Fig. 7.11). By repeatedly counting the number of pulses
m coming within a certain fixed time interval T , one can find the distribution
P(m) of the number of primary photoelectrons released from the photocathode
by the incident light. (Of course, the total duration of the measurement should
considerably exceed the coherence time coh .)
Let us find the relation between the statistics of photocounts and the field
statistics. Suppose that T coh and A Acoh , then one can neglect the intensity
variation during the sampling time T and over the photocathode area A. Under
this condition, the P(m) distribution will be determined by the P(I) distribution
regardless of the detection volume Vdet cT A. Here, coh and Acoh 2coh are the
typical scales of the field fluctuations, and the above-given inequalities enable us
to consider the detector as pointlike and single-mode, i.e., measuring a single
degree of freedom of the field. (This also implies that the detector measures a
single polarization type.)
h Editors note: nowadays, much more convenient for single-photon counting are avalanche photodi-
odes (APDs) (see also Sec. 1.3).

Assume first that the intensity I is constant, i.e., it does not vary from sample
to sample. It is important that even in this case, the number of photoelectrons in
a single sample is, according to the quantum mechanics, random, unpredictable.
The very process of energy measurement necessarily introduces additional Poisso-
nian stochasticity into the detector readings. (We do not consider the non-realistic
case of a detector with a 100% quantum efficiency and a pure energy state of the
field with a fixed number of photons.) Using the semi-classical (Sec. 2.1) or purely
quantum (Sec. 7.7) perturbation theory, we can only find the ionization probabil-
ity W1 t I for a single atom of the photocathode during a small time interval
t.
In the case of a sufficiently small t, the probability for any on the N indepen-
dent atoms of the photocathode to be ionized is N times as large, and also scales
as I,
W = NW1 (/T )I. (7.67)
Here is the scaling factor, which can be represented as
= Vdet /2~, (7.68)
where Vdet cT A is the effective detection volume, lN0 = N/A is the

quantum yield of a thin photocathode with the thickness l, is the ionization
cross-section, and N0 N/Al is the concentration of atoms. We assume that is
constant within the spectral width of the field.
By definition, all time moments within T are equivalent, since the wave inci-
dent on the photocathode is a pure sine one, and an electron can appear within
any time interval t with the same probability It/T . This statistical model, as
one can easily show (see, for instance, Ref. [Rytov (1976)]), leads to the Poisso-
nian distribution with the parameter I,
P(m|I) = C(I)m /m!, C eI . (7.69)
Intensity fluctuations from sample to sample can be taken into account by

averaging (7.69) with the P(I) distribution,
Z
P(m) = dIP(m|I)P(I) hP(m|I)i. (7.70)
0
As a result, we obtain the semi-classical Mandel formula for the photocount dis-
tribution, i.e., for the probability of discovering m pulses at the PMT output,
P(m) = h(I)m eI i/m!. (7.71)

Fig. 7.11 Relation between the statistics of the field and the photocount statistics from the semi-
classical viewpoint: E is the field, i is the PMT output current; (a) in the case of a field with a constant
amplitude E0 , the number of photoelectrons mi emerging during a certain time interval T has a Poisso-
nian distribution; (b) the amplitude fluctuations of the field at T coh cause additional fluctuations in
the number of the photoelectrons (the bunching effect); (c) a multi-mode (T coh ) detector averages
over the field fluctuations, and the bunching effect is not observed.
The quantum-theoretical approach, mainly developed by Glauber (Sec. 7.6),

yields an expression of the similar form, with the only difference that the probabil-
ity P(I) is replaced by the quasiprobability, a function that takes negative values
or has singularities (like delta-function derivatives) for some states of the field.
By expanding (7.71) in powers of the detector quantum efficiency , one can
express P(m) in terms of higher-order intensity moments, G(k) hI k i, with k 1 m,

X
P(m) = (1)km k G(k) /m!(k m)! (7.72)
k=m
For a sufficiently small detection volume ( T A 0), as a rule, one can

take into account only the first term in this expansion, i.e., neglect the exponent in
(7.71). Then P(m) can be represented as
T m (m) (NT )m m
P(m) = W = hW1 i, (7.73)
m! m!
where W (m) is the mth derivative of P(m) in T , W1 is the ionization probability per
unit time for a single atom, N is the total number of atoms in the detector.
The Mandel formula (7.71) describes the ensemble of random samples differ-
ing by a shift in time (due to the assumed stationarity and ergodicity of the field).
One can easily see that in the case of radiation that is homogeneous and spatially
ergodic along x, y and propagates approximately along z, the same formula de-
scribes an ensemble of samples differing by a spatial shift in the (x, y) plane. This,
in principle, allows one to study the statistics of non-stationary fields in real time,
with the help of a large number of photon counters placed at different points of
the beam cross-section over an area much greater than Acoh .
Further, it is clear from the derivation of formula (7.71) that it is also valid in
the case of an arbitrary coherence volume Vdet = cT A provided that I is under-
stood as the intensity averaged over Vcoh ,
Z
t) 1
I(r, dx0 dy0 dt0 I(x0 , y0 , z0 ), (7.74)
AT
where the integration limits, x a/2, y b/2, t T/2, are determined by the sizes
and the time constant of the detector. In this case, the distribution P(I) should
be replaced by P(I). In the limiting case of A Acoh or/and T T coh (multi-
mode detection), the fluctuations of I are completely eliminated due to averaging:
P(I) = (I hIi), so that we once again obtain the Poissonian distribution P(m|hIi)
from (7.71), but now it is independent of the field statistics. By observing the
dependence of P(m) on A, T in the intermediate case, one can, in principle, obtain
information about the coherence time and area of the field.
7.2.8 Photon bunching

Relation (7.71) between the distributions also determines the relations between
the moments of the photocount numbers,
X
k
hm i mk P(m), (7.75)
m=0
and the intensity moments,
Z
k
hI i dII k P(I). (7.76)
0
Using the generating functions method (Sec. 7.6), one can easily find the general
rule,
hm(m 1) . . . (m k + 1)i = k hI k i. (7.77)
The linear combination of moments in the left-hand side is called the kth-order
factorial moment. From this, in particular, it follows that hmi = hIi and
hm2 i = hmi + 2 hI 2 i, (7.78)
hI 2 i hI 2 i hIi2 , (7.79)
and similarly for hm2 i. Thus, fluctuations of the photoelectron number contain,
in addition to the usual Poissonian (shot-noise) part, a contribution from the fluc-
tuations of the light intensity. Only in the case of a single-mode stabilized laser
(I = 0) this contribution is absent. In other cases, fluctuations of the photo-
electron number, according to (7.78), should at first sight exceed the shot noise,
since from the definition (7.79) and from the condition P(I) > 0 it follows that
hI 2 i > 0.
The existence of these excess fluctuations has been called the photocount
bunching effect, since in a Poissonian sequence of pulses, by definition, the ap-
pearance of a single pulse has no effect on the appearance of the next one, and the
inequality hm2 i > hmi means that the pulses have a tendency to bunch. A similar
effect for the photon numbers, hN 2 i > hNi, is called photon bunching or photon
correlation. A close effect has been discovered by Hanbury Brown and Twiss in
1956 in the chaotic light of a mercury lamp.
In chaotic light, it follows from (7.51) and (7.78) that
hm2 iT = hmi(1 + hmi), (7.80)
i.e., the excess part of the variance is hmi times as large as the Poissonian one, so
that the photon bunching effect is more pronounced in classical fields. From the
classical viewpoint, strong fluctuations of the amplitude |E0 | of a wave formed by
many independent sources with random phases are quite obvious. More surpris-
ing for the classical theory is the anti-bunching of photons and, correspondingly,
photocounts, so that m2 < hmi, in contradiction with (7.78) and the initial Man-
del formula (7.71) if P(I 1 0) is assumed. (As we have already mentioned, the
last condition is violated in quantum theory.)
7.2.9 Intensity correlation

The photocount number distribution (7.71) does not provide direct information
about the temporal or spatial radiation spectrum, as it only contains single-point
CFs with all arguments coinciding (see (7.72)). A complete characterization of an

nth-order CF requires the measurement of the field at 2n points of space-time. In
the case of the first-order CF, this can be done with the help of interferometers, as
it was schematically described above.
Consider now the measurement of the intensity correlation function, namely,
the second-order CF of the following particular form:
G(2) (x1 , x2 , x1 , x2 ) G(2)

12 () = h: I(x1 )I(x2 ) :i, (7.81)
where x1 {ri , ti }, t2 t1 , and the r2 r1 vector is orthogonal to the

propagation direction. The colons remind that in the quantum calculation, all E (+)
operators should be put on the right of E () operators before the averaging.
In the case of a Gaussian field, with the help of (7.57) we find
G(2) (1) 2
12 () = hI1 ihI2 i[1 + |g12 ()| ], (7.82)
where g(1)12 () is the normalized first-order CF. Thus, in a chaotic field, by mea-
suring the intensity correlation one also gains information about the amplitude
correlations. This relation between G(2) and G(1) forms the basis for the opti-
cal mixing spectroscopy [Cummins (1974)], also called spectroscopy of intensity
fluctuations or the method of photon correlations.
In the general case, G(2) is not related to G(1) , and the second-order coherence
parameters may differ from (1) (1) (2)
coh , coh . For instance, in two-photon light, coh
(1) i
coh .
The time dependence of G(2) () at = 0 is measured by means of a single
detector with a delay line and an electronic correlation circuit.j One can also use
the spectral analysis of photocurrent fluctuations. The first experiment of this type
has been performed by Forrester, Gudmundsen and Johnson as early as in 1955,
before the advent of lasers.k Modern technique enables one to achieve the spectral
resolution much less than 1 Hz.
In order to study the spatial second-order coherence G(2) (), one has to use
two detectors with a variable distance between them.l According to (7.82), when
is varied from to 0, G(2) is increased, in an ideal case, by a factor of two
(Fig. 7.12). This effect has been discovered by Hanbury Brown and Twiss in 1956
and used for the measurement of the angular diameters of stars [Hanbury Brown
i Editors note: This is the case in one of the possible definitions of second-order coherence parame-
ters, see the last subsection.

j Editors note: see Fig. 7.14(a).
k The possibility of similar experiments on the heterodyning of light had been discussed even earlier
by Gorelik [Gorelik (1948)].

l Editors note: see Fig. 7.14(b).
Fig. 7.12 Correlation and anti-correlation of light intensities: i1 , i2 are the photocurrents of the de-
tectors; is the displacement of one of the detectors from the symmetric position; 1, thermal radiation;
2, laser radiation; 3, two-photon radiation.
(1971)] whose intensities are correlated within distances on the order of hundreds
of meters.
The Hanbury BrownTwiss experiment on the measurement of G(2) () for the
light from a mercury lamp is shown in Fig. 7.13. Splitting of the beam with the
help of a semi-transparent mirror allows one to measure the correlation at points
that are arbitrarily close to each other. Let the PMTs operate in the photon count-
ing regime, then the correlation between the photocounts in the two channels,
hm1 m2 i, is 1 2 hI1 I2 i. Hence, with the help of (7.82), we find
hm1 m2 i = hm1 ihm2 i(1 + |g(1) 2
12 (0)| ). (7.83)
This result is only valid in the case of single-mode detectors, where the time
constant of the detector T is much less than coh and the detector aperture A is
much less than Acoh . If, for instance, T coh , then a small factor on the order of
coh /T appears by the second term of (7.83), which reduces the observed effect.
Fig. 7.13 Hanbury BrownTwiss experiment on the observation of light intensity correlation.
The intensity correlation effect is closely connected with the intensity fluctua-
tions of light at the input of Hanbury BrownTwiss interferometer. Indeed, let I1
and I2 be random intensities in the arms of the interferometer and I1 and I2 be
their fluctuations (In In hIn i). The output signal hi1 i2 i scales as
hI1 I2 i = hI1 ihI2 i + hI1 I2 i. (7.84)
The second term here characterizes the mutual correlation of intensities. From the
condition I1 + I2 = I, we find the relation
hI 2 i = h(I1 + I2 )2 i = hI12 i + hI22 i + 2hI1 I2 i. (7.85)
Thus, the correlation is determined by the variances,
hI1 I2 i = (1/2)(hI 2 i hI12 i hI22 i). (7.86)
Assume first that the incident light has a constant intensity (radiation of a
single-mode laser), then In = 0 and, according to (7.86), the correlation is equal
to zero. Then, hI1 I2 i = hI1 ihI2 i. Now, suppose that usual light from a thermal or
luminescent source be incident on the interferometer; then, according to (7.51),
hI 2 i = hIi2 . Suppose that similar relations hold for the secondary beams as well,
hIn2 i = hIn i2 . Hence, with the help of (7.86) we find (Fig. 7.12) that
hI1 I2 i = (1/2)(hIi2 hI1 i2 hI2 i2 ) = hI1 ihI2 i,

(7.87)
g(2)
12 hI1 I2 i/hI1 ihI2 i = 2.
This reasoning can be easily reproduced in the photon language by changing

In to photon numbers Nn . In this case, the shot noise (Poissonian noise) is added
to the variance, so that
hN 2 ilas = hNi, hN 2 iT = hNi + hNi2 . (7.88)
In fact, the result is the same: in thermal light, there is photon correlation caused
by photon bunching or, in other words, by the presence of excess noise in addition
to shot noise.
Further, consider field with a fixed number N of photons. These N photons
will be randomly split by the semi-transparent mirror between the two channels
with the probabilities p and q = 1 p. This picture corresponds to the well-
known Bernoullis probability model [Rytov (1976)], which gives the binomial
distribution for the probability of N1 photons going to channel 1,
P(N1 ) = C NN1 pN1 qNN1 . (7.89)

The moments of this distribution have the form

hN1 i = pN, hN12 i = p2 N 2 + pqN,
(7.90)
hN2 i = qN, hN22 i = q2 N 2 + pqN.
Hence, we find
hN1 N2 i = (1/2)(N 2 hN12 i hN22 i) = pqN(N 1). (7.91)
It is noteworthy that now the correlation is negative,
hN1 N2 i = hN1 N2 i hN1 ihN2 i = pqN.
Then (see Fig. 7.12),
g(2)
12 = 1 1/N. (7.92)
Thus, photon anti-bunching in the initial beam leads to photon anti-correlation

in the two output beams.
Consider, finally, the general case of a field with an arbitrary statistics. Then,
(7.91) should be additionally averaged over the photon-number distribution P(N)
in the incident field. As a result,
hN1 N2 i = pqh: N 2 :i, g(2) 2 2 (2)

12 = h: N :i/hNi g , (7.93)
where h: N 2 :i hN(N 1)i is the normalized (factorial) moment and the an-
gular brackets denote averaging over the distribution P(N). Thus, the relative
correlation of photon numbers at two field points, g(2)
12 1, is determined by the
normalized factorial moment of the field g(2) . The same result follows from the
rigorous quantum-theory approach (see (7.331)).
7.2.10 Second-order coherence (added by the Editors)

Unlike the first-order coherence parameters, which are defined unambiguously as
the widths of the first-order correlation function (CF) in time and space, and can
be measured as shown in Figs. 7.87.10, second-order coherence time and radius
allow different definitions. This is because the second-order CF G(2) (t1 , r1 , t2 , r2 )
has two time arguments and two space arguments, and there are several ways to
define its width. In the stationary case, the second-order correlation function de-
pends only on the difference of its time arguments; the same relates to the space
arguments in the spatially homogeneous case: G(2) (t1 , r1 , t2 , r2 ) G(2) (, ). Then,
the second-order coherence parameters can be defined as the widths of G(2) (, )
in and . In the non-stationary (spatially inhomogeneous) case, they can be
corr
t corr
tcorr
rcorr
(a) (b)
Fig. 7.14 Measurement of the temporal (a) and spatial (b) widths of the second-order correlation
function.
introduced by analogy, as the widths of G(2) (t1 , r1 , t2 , r2 ) in t1 t2 or r1 r2 (con-

ditional width of the CF). These parameters can be measured using two simple
setups shown in Fig. 7.14. Intensities at two space-time points are measured by
two detectors, usually photon-counting ones, and the coincidences of their counts
are registered with the help of a coincidence (correlation) circuit. Time delay
is usually introduced electronically, by delaying the output pulses of one of the
detectors (Fig. 7.14(a)).m Space delay is introduced by displacing one of the de-
tectors (Fig. 7.14(b)). Such setups (see, for instance, [Mandel (2004)]) allow one
to measure the second-order correlation function and, in particular, its widths in
space and time.
However, there is another way to introduce second-order coherence param-
eters. By analogy with the experimental schemes of measuring the first-order
coherence time and radius (Figs. 7.87.10), which are based on the first-order in-
terference, one can define the second-order coherence parameters using various
experimental schemes for observing second-order interference (intensity interfer-
ence). In particular, intensity interference can be observed using Michelsons or
Youngs interferometers (Fig. 7.15). One measures the coincidence counting rate
between two detectors, which register the intensities of the field at two points (us-
ing a beamsplitter if necessary, as in Fig. 7.15(a)). The interference phase can be
varied by moving the mirror in the Michelson interferometer (a) or by displacing
one of the slits in the Young interferometer (b). The interference pattern will be
formed by the dependence of the coincidence counting rate on the phase, provided
that the first-order interference is absent. The second-order coherence time can be
introduced then as the delay in the Michelson interferometer at which the visibility
of the intensity interference decays by a factor of two. Similarly, the second-order
coherence radius can be defined as the distance between the slits in the Youngs
m Inthe case of SPDC, the correlation time is usually less than the resolution of the detectors; in such
cases, other techniques should be used.
(2)
t
coh
corr
(2)
t corr
t r r
corr
coh
corr
(a) (b)
Fig. 7.15 Intensity interference observed using (a) Michelsons interferometer and (b) Youngs in-
terferometer. Coincidences of photocounts are observed for two detectors registering photon pairs
emitted by sources separated in time (a) or space (b).
experiment at which the visibility of intensity interference decays by a factor of

two.
The principal difference between this definition and the previous one is that
they relate the second-order coherence parameters to the radiation emitters (or
different contributions into radiation) and detectors, respectively. Note that it is
the second definition that is meant at the beginning of this section (paragraph after
Eq. (7.82)), leading to the conclusion that for two-photon light, (1) (2)
coh coh . In
order to distinguish between these two different definitions, it is convenient to use
the terms correlation time/correlation radius for the widths of G(2) (t1 , r1 , t2 , r2 )
in t1 t2 and |r1 r2 |, respectively (Figs. 7.14(a) and 7.14(b)), and the
terms coherence time/coherence radius for the widths of the intensity interfer-
ence patterns (Figs. 7.15(a) and 7.15(b)).
As an example, consider the two alternative definitions in the case of two-
photon light generated via SPDC. Let us discuss only temporal coherence, and
suppose that the pump is a sequence of relatively long Fourier-limited pulses
of duration T p . The correlation time will then be determined by the length of
the nonlinear crystaln and be close to the first-order coherence time, which is
given by the inverse spectral width of SPDC radiation [Goodman (1985)]:
corr (1)coh = 2/. At the same time, according to the second definition,
(2)
coh can be measured experimentally by observing two-photon interference with
SPDC radiation fed into a Michelson interferometer (Fig. 7.15(a)). Numerous
experiments on two-photon interference [Mandel (2004)] show that the visibil-
ity will be high as long as the time delay does not exceed the pump coherence
time, which in the case considered here coincides with T p . Thus, for this example
(2) (1) (2)
coh T p and coh corr coh .
n Strictly
speaking, it is the minimal value of the correlation time that is determined by the crystal
length; the correlation time can then be increased due to the propagation of light through a dispersive
medium.
The same reasoning will be valid for spatial coherence parameters, (1,2)
coh .
7.3 Hamiltonian form of Maxwells equations
In this section, we will show that Maxwells equations for the transverse part of the
field, i.e., for the radiation field, can be reduced to a system of independent equa-
tions for harmonic oscillators. These equations can be easily represented in the
form of classical Hamiltons equations. This enables one to use the quantization
algorithm which defines the commutator of two operators in terms of Poissons
brackets for the corresponding classical values.
7.3.1 Maxwells equations in the k, t representation

Suppose that we are interested in the evolution of a radiation field within a certain
bounded space domain during a limited time interval T . Imagine that all space is
divided in identical cubic cells with the linear size L > cT ,o , so that one of the
cells contains all the field that is measured during the time T .
Fig. 7.16 To the definition of the quantization length L: E(z) is the real field and E(z) is the fictitious
field that is periodic in space.
Consider the dependence of some field component on one of the coordinates,

for instance, E x (z), at a fixed time moment (Fig. 7.16). Let us define a spatially
periodic function E x (z) by the condition E x (z + nL) E x (z), where L/2 < z <
L/2, n = 0, 1, . . . . Within the observation interval, (L/2, L/2), the fictitious
periodic field E coincides with the real one, E; therefore, E and E are physically
equivalent, and in future we will omit the tilde.
o This condition is assumed to be well satisfied, so that the field at the boundaries stays equal to zero
during all observation time.

Fig. 7.17 The sinc(x) function in the cases of a discrete (points) and continuous (dashed line)
arguments.
A field that is periodic in z can be represented as a sum of spatial harmonics

(we omit the x index),
X
E(z) = Em exp(ikm z),
m= (7.94)
km 2m/L, m = 0, 1, 2, . . .
For finding the amplitude of the mth harmonic, we act on (7.94) from the left by
the operator
Z L/2
dz exp(ikn z),
L/2
then
Z L/2 X
dz exp(ikn z)E(z) = Em Lsinc[(m n)] = En L. (7.95)
L/2 m
The last inequality follows from the fact that the function of an integer value,
Z L/2
dz exp[i(km kn )z] = Lsinc[(m n)] = Lmn , (7.96)
L/2
differs from zero only at a single point m = n, where it is equal to unity (Fig. 7.17).
Since E(z) is real, Em = Em .
At the final stages of the calculation, the quantization length L can be usually
assumed to be infinite; in this case, the Fourier series (7.94) becomes an integral,
Z
E(z) = (L/2) dkE(k)eikz , (7.97)

where
Z
1
E(k) = L dzE(z)eikz . (7.98)

In the derivation of (7.98), the following representation of the delta function was
used:
Z
dzeikz = lim Lsinc(kL/2) = 2(k). (7.99)
L
The factor L/2 is added to (7.97) for making the discrete Fourier components
coincide at L with the continuous ones, Em E(km ). Comparing (7.94) and
(7.97), we find the rule for passing from summation to integration,
X Z
(L/2) dk . . . (7.100)
m
The value L/2 is called the density of modes (in the one-dimensional case
with a single polarization state). The inverse value, 2/L, is equal to the distance
between the neighboring modes m, m + 1 on the k axis.
By repeating this procedure for other components of the field Ey , Ez and for the
dependencies on x, y, we obtain a three-dimensional Fourier series for the field,
X X
E(r, t) = Elmn (t)e2i(lx+my+nz)/L E k (t)eikr , (7.101)
lmn k
Z
E k (t) = L3 d3 rE(r, t)eikr = Ek (t), (7.102)
L3
k (2/L){l, m, n}, l, m, n = 0, 1, 2, . . . (7.103)

A similar series expansion can be written for the magnetic field. Due to rep-
resenting an arbitrary field (7.101) as a sum of plane waves, a continuous spatial
distribution E (r) is determined by a countable set of complex numbers Ek . The
allowed vectors k form a grating in the k space, which is divided in cells of
volume (2/L)3 due to the periodicity condition. Note that because of the relation
Ek = Ek , not all numbers Ek are independent. The sum (7.101) can be written
in the following equivalent forms:
X X
E = Re E k eikr = Re(E + iF) = Re (E k + iF k )eikr , (7.104)
k k
where F(r, t) is an arbitrary real field and F k = Fk are its harmonics.

Expansion in spatial harmonics (7.101) enables the field E(r) to be unam-
biguously separated in two components: the transverse E (r) and the longitudinal
Ek (r) (the argument t is omitted),
XX
E (r) = e k E k eikr ,
k =1,2
X (7.105)
Ek (r) = e k3 E k3 eikr ,
k
where the unity orthogonal vectors e k form a right-hand triple, and e k3 k

k/k. It follows from (7.105) that
X
divE = ik E k eikr = divEk ,
k
X (7.106)
rotE = ik E k eikr = rotE ,
k
and similar relations are valid for the magnetic field. By substituting (7.106) into
Maxwells equations (4.9)(4.12) and assuming = 1, we obtain
crotH E = 4 j , crotE + H = 0, (7.107)
Ek = 4 jk , Hk = 0, (7.108)
divEk = 4, div Hk = 0. (7.109)
Hence, the longitudinal part of an alternating magnetic field is equal to zero,

while the longitudinal part of the electric field is determined by the positions of the
charges at the same time moment, without retardation. Therefore, the radiation
field in the vacuum, which is of interest for optics, is transverse, and it is deter-
mined by the dynamical equations (7.107) through the transverse part of given
(external) currents j j. (Hereafter, we omit the index.)
By substituting (7.105) into (7.107), we find the equations of motion for the
spatial harmonics,
E k ick H k = 4 j k , (7.110)
H k + ick E k = 0, (7.111)
which, after excluding H k , yields
E k + 2k E k = 4 j k . (7.112)
Here, k ck and
Z
3
j k (t) L d3 r k j(r, t)eikr = jk (t), (7.113)
L3
with k being the projection tensor (see (4.20)). Thus, Maxwells equations for
the transverse field in the k, t representation are reduced to a system on inhomo-
geneous equations for independent harmonic oscillators. Note that the harmonics
E k and Ek are always excited simultaneously since E k = Ek .
In the case of a free-space field, i.e., in the absence of currents in L3 , the

spatial harmonics, according to (7.110)(7.112), oscillate without damping with
the eigenfrequencies of the modes k ,
E k (t) f ree = (E k0 ei k t + E0k0 ei k t )/2,

(7.114)
H k (t) f ree = k (E k0 ei k t E0k0 ei k t )/2.
Here, E k0 is the initial amplitude of the plane wave propagating in the +k direc-
tion, while E0k0 is the independent amplitude of the backward wave, propagating
in the k direction. The condition E k = Ek yields E0k0 = Ek0 . Hence, by
summing over all k, we obtain
X
E(r, t) f ree = Re E k0 ei(kr k t) ,
k
X (7.115)
H(r, t) f ree = Re k E k0 ei(kr k t) .
k
Thus, the state of a free-space field at an arbitrary point r, t is given by a set of

complex vectors E k0 .
In the presence of external currents in L3 , induced field is added to the free
one. The induced field is determined by the j k (t) functions according to the in-
homogeneous equation (7.112). For instance, a monochromatic plane wave of
the current will excite the induced field with its frequency , which may differ
from k (compare (4.23)). In the general case, E k (t) is certainly not a harmonic
function. The induced field can be also searched in the form (7.115), by assuming
E k0 to be slow functions of the coordinates in the case of stationary currents (see
Chapter 6 where we used the notation E k0 = E(+) k (z)) or functions of the time in
the case of non-stationary problems, typical for the quantum mechanics.
Sometimes, it is convenient to describe the field using the vector potential
A(r, t). In the case of the Coulomb gauge, the field A is assumed to be transverse,
and it is unambiguously defined by the relations
rotA H, divA = 0. (7.116)
By substituting rot A for H into (7.107), we obtain rot(cE + A) = 0, i.e.,
E = A/c. (7.117)
Hence, we find the relations between the spatial harmonics of a real field and its
potential,
A k = cE k , A k = ik H k /k2 , H k = ik A k . (7.118)
7.3.2 Canonical field variables

Equations (7.110)(7.112) for H k and E k resemble the Hamilton equations for
the canonical coordinates and momenta of a system of particles, qi , pi . The cur-
rents j k play the role of generalized forces. However, in experiment, one usu-
ally observes traveling waves in the far-field zone of the source, having a certain
propagation direction, for instance, along +k; therefore, it is preferable that the
canonical variables with the index k relate only to the forward wave.
The set of the four numbers {l, m, n, } {k, } k defines a plane wave or
a mode (oscillation type) in free space. (In what follows, we will numerate the
modes by a single index k.) In the presence of currents, the instantaneous state of
the field in two modes k and k {k, } with the same linear polarizations e k is
given by two complex scalars or four real ones,
Ek Ek ek Ek0 + iEk00 , Hk ( k ek ) Hk Hk0 + iHk00 ,

Ek Ek ek = Ek0 iEk00 , Hk ( k ek ) Hk Hk0 + iHk00 . (7.119)
Instead of the magnetic field, one can use the vector potential. According to
(7.118) and (7.119),
Ak Ak ek = iHk /k = (Hk00 iHk0 )/k,
Ak = (Hk00 + iHk0 )/k,
where k denotes simultaneously the absolute value of the k vector and the mode
index.
Let us form linear combinations,
qk (L3 /42k )1/2 (Ek00 + Hk00 ),
(7.120)
pk (L3 /4)1/2(Ek0 + Hk0 ).
(the choice of the coefficients will be explained below, from (7.134).) With the
help of (7.119), we see that the variables qk , pk for the backward mode are inde-
pendent of qk , pk ,
qk Ek00 + Hk00 = Ek00 + Hk00 ,

pk Ek0 Hk0 = Ek0 + Hk0 .
It is convenient to join the real coordinate qk and momentum pk of a mode

in a single complex dimensionless variable,
ak (2~k )1/2 (k qk + ipk ) = (Ek + Hk )/2ick ,
(7.121)
ck (2~k /L3 )1/2 .
The inverse transformations can be easily found as well,
Ek = ick (ak ak ) = (/L3 )1/2 [pk pk + ik (qk qk )],

(7.122)
Hk = ick (ak + ak ) = (/L3 )1/2 [pk + pk + ik (qk + qk )].
Note that in the case of a standing plane wave, ak = ak , and therefore the variables
qk , pk scale as the magnetic and electric fields, respectively,
E(r, t) = 4(/L3 )1/2 pk (t) cos(k r),

(7.123)
H(r, t) = 4(/L3 )1/2 k qk (t) sin(k r).
In the new variables, the plane-wave expansion takes the form
X
E(r, t) = i ck ak (t)eikr + c.c.,
k
X
H(r, t) = i ck k ak (t)eikr + c.c., (7.124)
k
X
A(r, t) = i (ck /k)ak (t)eikr + c.c.,
k
P
where complex vectors a k e k a k have been introduced.
After substituting (7.122) into (7.110)(7.112), we find the equations for the
new variables,
qk = pk (4L3 /k )1/2 j00k , (7.125)

pk = 2k qk + (4L3 )1/2 j0k , (7.126)
ak = ik ak + (2i/ck ) jk . (7.127)
The general solution to the last equation has the form

Z t
2i 0
ak (t) = ak (0)eik t + dt0 eik (tt ) jk (t0 ). (7.128)
ck 0
Expanding ak (t) and jk (t) in Fourier frequency integrals, from (7.127) we im-
mediately find the induced part of the field in the k, representation (compare
with (4.23)),
2c1
k
ak ()ind = jk (),
k ik
where we have added the damping, k > 0. Hence, it is clear that at k k , the
spectrum ak () of the amplitude ak (t) mainly contains only positive frequencies
close to the eigenfrequency k . If we neglect the negative-frequency part of ak (t),

i.e., assume that
Z
ak (t) a(+)
k (t) deit ak (), (7.129)
0
then every term in the sum (7.124) describes a plane wave propagating in the +k
direction (in contrast to the sum (7.101)).
In a free-space field, this approximation, according to (7.128), is at jk = 0
valid rigorously,
ak (t) f ree = ak (0)eik t . (7.130)
By comparing (7.114) and (7.122), we find the relation Ek0 = 2ick ak (0).
Hence, the positive-frequency part of the field is determined by ak (t) functions
while the negative-frequency part, by ak (t) functions,
X
E(+) (r, t) = i ck ak (t)eikr ,
k
X (7.131)
()
E (r, t) = i ck ak (t)eikr .
k
In the quantum theory, these functions become photon creation and annihilation
operators, ak ak , ak ak .

7.3.3 Hamiltonian of the field and the matter
From Maxwells equations, it follows (see, for instance, (4.27)) that the instanta-
neous energy of the field is
Z
E(t) = (1/8) d3 r(E 2 + H 2 ) H0 . (7.132)
L3
Let us accept this expression as the Hamilton function of the free-field transverse
part. By substituting here the plane-wave expansion (7.101) and taking into ac-
count the orthogonality condition (7.96), we obtain a diagonal quadratic form,
X
H0 = (L3 /8) (|Ek |2 + |Hk |2 ). (7.133)
k
According to (7.114), at j = 0, H0 does not depend on t. With the help of (7.124),

we find that
X X
H0 = (p2k + 2k q2k )/2 = ~ k |ak |2 . (7.134)
k k
One can easily verify that the Hamilton equations,

qk = H/pk , pk = H/qk, (7.135)
at H = H0 lead to the oscillator equations for qk , pk , ak following from (7.125
7.127) at j = 0. This confirms that (7.132) was chosen correctly.
The joint Hamiltonian of the field and the system of charged particles inside
the volume L3 in the non-relativistic case (see Ref. [Landau (1973)]) is
X
H = H0 + (Pi ei Ai /c)2 /2mi + Hc ,
i
X (7.136)
Ai A(Ri (t), t), Hc ei e j /|Ri R j |,
i< j
where Hc is the energy of the Coulomb (longitudinal-field) interaction between

the particles and Ri , Pi are the canonical variables of the ith particle whose charge
and mass are ei , mi .
Eqs. (7.135) and (7.136) lead to the relation between the kinetic and canoni-
cal momenta of the particles,
mi V i = Pi ei Ai /c, (7.137)
where V i Ri is the i-th particle velocity. Hence, the Hamiltonian (7.136) can be
represented in a simple form,
X
H = H0 + mi Vi2 /2 + Hc . (7.138)
i
According to (7.136), the Hamiltonian of the interaction between the particles and
the transverse field is
X ei e2i

2
V=
Pi A i + 2
Ai . (7.139)
i
mi c 2mi c
In the case of particles with internal magnetic moment i , one should add the
energy of the spin interaction, Hi .
Let us show that (7.135), (7.136) lead to the usual Newtons equations with
the Lorentz force for the particles and to Maxwells equations with the external
currents (7.107) for the field. In order to obtain Newtons equations, let us differ-
entiate (7.137) in time. Taking into account that, according to (7.135), (7.136),
Pi = (ei /c)Vi Ai /Ri , (7.140)
we find that
!
ei Ai Ai ei
mi Ri = Pi + Vi = ei Ei + [V i Hi ] . (7.141)
c t Ri c
Recall that the fields are taken at the point of the particle location, therefore,
d Ai /dt , Ai /t = cEi .
The field equations can be found by differentiating the total Hamiltonian
(7.136) w.r.t. the canonical variables pk , qk of the field. Then the forces act-
ing on the field from the particles are determined by the second term in (7.138).
Let us first differentiate it in Ai with the help of (7.137):
mi Vi2 Vi ei ei
= mi Vi = Vi = V i Ai . (7.142)
Ai 2 Ai c Ai c
It follows that the Hamiltonian of the interaction between the transverse field and
non-relativistic spin-free particles can be represented, instead of (7.139), as
Z
0
X ei 1
V V i Ai = d3 r j A, (7.143)
i
c c
where j(r, t) is the external current density determined by the coordinates and
velocities of the particles,
X
j(r, t) = ei V i (t)(3) (r Ri (t)), (7.144)
i
and the prime reminds that, according to (7.142), the Hamiltonian (7.143) pro-
vides an exact description only for the perturbation of the field by the particles
and not vice versa.
Hamiltons equations (7.135) immediately lead to the equations of motion for
an arbitrary function of canonical coordinates, f (qk , pk , t),
d f /dt = f /t + { f, H}, (7.145)
X f g !
g f
{ f, g} . (7.146)
k
qk pk qk pk
One can easily verify that after the linear transformation (7.121) from qk , pk to
the new independent variables ak , ak , the Poisson bracket (7.146) takes the form
!
1 X f g g f
{ f, g} = . (7.147)
i~ k ak ak ak ak
Assuming f ak , with the help of (7.134) and (7.143) we obtain
Z
1 H i Ai
ak = = ik ak + d3 r j . (7.148)
i~ ak ~c ak
With an account for (7.124) and (7.144), this equation coincides with Eq. (7.127),
which was obtained from Maxwells equations with external currents.
Thus, we have written the equations for the field and the matter in the canon-
ical form (7.135) with the Hamiltonian (7.136). Before using this result for the
quantization of the field equations, let us consider the interaction Hamiltonian in
the dipole approximation.

7.3.4 Dipole approximation
In quantum electronics, it is often possible to use approximations instead of the
exact expressions for the perturbation energy (7.139), (7.143). In the case of a
plane free monochromatic wave, H = E. Therefore, in the first order in Vi /c, one
can neglect the magnetic part of the Lorentz force in the Newton equation,
m Ri ei E(Ri , t). (7.149)
Further, let the particles occupy a restricted space area with the linear size a
much less than the scale of the field variation, o = c/. Then the field can be writ-
ten as a series expansion in Ri and only the first few terms should be taken into
account. In this case, (7.139) leads to the multi-field expansion of the perturbation
Hamiltonian for the particles in powers of Ri /o. In the zeroth (dipole) approxi-
mation, A(Ri ) A(r0 ) A0 , where r0 is some fixed point inside the system of
particles (for instance, the center of mass). Then, according to (7.140), Pi = 0,
and (7.141) takes the form (compare with (7.149))
mi Ri = ei E0 , (7.150)
where E0 E(r0 , t). This equation, according to (7.135), follows from the inter-
action Hamiltonian of the form
Vdip d(t) E0 , (7.151)
where
X
d(t) ei (Ri (t) r0 ), |Ri r0 | o. (7.152)
i
Here, in contrast to (7.139), the field is a given external parameter. Note also that
the dipole moment of a neutral system does not depend on the choice of r0 .
Let the matter consist of N separate motionless molecules with the dipole
moments d j and centers at r j . Then the energy of the matter in the external field,
according to (7.151) is (compare with (4.28))
N
X Z
Vdip = d j (t) E(r j , t) = d3 rP(r, t) E(r, t), (7.153)
j=1
N
X
P(r, t) = d j (t)(3) (r r j ). (7.154)
j=1
The interaction Hamiltonian for the field in the dipole approximation follows
from (7.143) after replacing Ai by A0 ,
0 1 1d
Vdip = d A0 = d E0 (d A0 ). (7.155)
c c dt
If we restrict the consideration to the case of quasi-monochromatic currents and
fields, then d A0 contains two components: a constant one and one oscillating
with the double frequency. As a result, only the first term in (7.155), coinciding
with (7.151), provides an accumulating interaction,
0
Vdip Vdip = d E0 . (7.156)
Thus, the dipole Hamiltonian Vdip can be also used for calculating the emitted
field in single-frequency problems. It follows from (7.156) that
1 Vdip ck
ak + ik ak = = dk exp(ik r0 ), (7.157)
i~ ak ~
where dk d ek . The same result can be obtained from the exact equation
(7.127) in the case of a neutral system, after taking into account (7.144), replacing
exp(ik Ri ) by exp(ik r0 ), and replacing V i by ik Ri .
Often, instead of (7.139), one uses the approximation
X
V ei Pi Ai /mi c, (7.158)
i
i.e., neglects the term that is quadratic in eA. (Note that in the case of a single
electron in a harmonic field, this term is on the order of Ei2 where = o2 re is the
polarisability of a free electron (6.36).) This approximation is only valid in the first
order of the perturbation theory, i.e., in the calculation of single-quantum effects.
Eq. (7.158) also follows from (7.143) if the canonical and kinetic momentums are
assumed to be the same. Eq. (7.158) leads to the following equation of motion for
the particle:
ei dAi
mi Ri = , (7.159)
c dt
which at a o coincides with (7.150).
For bonded electrons in atoms and small molecules, a 108 cm, and the con-
dition for the validity of the dipole approximation (7.151), (7.155) is satisfied up
to the X-ray range. Recall that magnetic moments related to spin and orbital mo-
tion are on the order of the Bohr magneton 0 , which is two orders of magnitude
as small as one Debye,
20 e~/mc = eoc ea0 /137. (7.160)
However, despite the relatively small value of multi-field effects, their mani-
festation in the optical range is important and can be easily observed: for instance,
in the effect of the optical activity (polarization rotation) and in the appearance of
forbidden lines in spectra.
A free electron in a harmonic field, according to (7.149), oscillates with the
amplitude a1 = eE1 /m2 and the velocity a1 ; therefore, the conditions a o
and V c have the same form,
E1 mc2 /eo 108 G. (7.161)
This estimate was made for = 1 and corresponds to a practically impossible
intensity 1018 W/cm2 . Nevertheless, by taking into account the magnetic-field
effect in a light wave (Sec. 6.2) one can calculate the quadratic polarisability of a
free electron and observable nonlinear effects.
7.4 Quantization of the field
Thus, we have represented the field equations in the form of Hamiltons equations
for the spatial harmonicsq Ek (t), Hk (t) (or their linear combinations qk , pk , ak ).
Now, we can pass to the main stage of the quantum description, which is finding
the commutation rules for the dynamical field variables.
7.4.1 Commutation relations

After passing to the quantum description, all canonical variables qk , pk and their
functions f (qk , pk ) become linear operators qk , pk , fk acting according to certain
rules on the state vector of the system. The difference between the actions of
the operator products f g and g f p can be defined in terms of the Poisson brackets
(7.146),
f g g f [ f, g] = i~{ f, g}. (7.162)
Here, all variables are considered at the same time moment. In particular, we find
that
[qk , pk0 ] = i~k,k0 , [qk , qk0 ] = [pk , pk0 ] = 0, (7.163)
p The hat sign over the operators will be only used where necessary.
[ak , ak0 ] = k,k0 , [ak , ak0 ] = [ak , ak0 ] = 0, (7.164)

[ f, ak ] = f /ak , [ak , f ] = f /ak , (7.165)
[Ek , Hk0 ] = 2c2k k,k0 , [Ek , Ek0 ] = [Hk , Hk0 ] = 0, (7.166)
where
c2k = 2~k /L3 , k {k, }, k {k, }. (7.167)
From (7.166), using the linear relations (7.124), one can also find the commutators
for the fields E(r, t), H(r, t).
The classical complex variable ak is put into correspondence with the pho-
ton creation operator ak , which is Hermite conjugated to the photon annihilation
operator ak . The Hermitian operator ak ak = (ak ak ) Nk is called the photon-
number operator, and most often, it is this operator that corresponds to optical
observables. For instance, the spectral brightness can be expressed in terms of Nk
as
I (k, ) = ~c3 hNk i. (7.168)
This expression, in contrast to (7.19), contains averaging over the wave function
or the density operator .
Often, in addition to Nk , other operators can be of interest. For instance, the
mean value of ak ak0 characterizes the statistical coupling between modes k and k0 .
One can show that the rate of an m-quantum stimulated transition scales as the
mean value of the operator am m m
k ak : Nk : . Here, colons denote normal ordering,
i.e., putting all ak operators on the right of ak . The mean values of these operators,
m m
G(m) m
k h: Nk :i = hak ak i, (7.169)
are called normal (normally ordered, factorial) moments of order m for mode k.
The relation between the factorial moments G(m) and the usual ones hN m i can be
easily found from the operator identities following from (7.162) or (7.164),
[am , N] = mam , [N, am ] = mam .
Hence,
: N m := N(N 1) . . . (N m + 1). (7.170)
(Hereafter, we omit the subscript k whenever a single mode is considered.)
In the Heisenberg picture, the wave function and the density matrix are con-
stant, while the operators depend on time according to the Heisenberg equations,
which can be obtained from (7.145) and (7.162),
d f /dt = f /t + [ f, H]/i~. (7.171)
For instance, assuming f ak and using (7.124), (7.134), (7.143) and (7.165), we
obtain the Heisenberg equation for the annihilation operator in the form (7.127).
Similarly, all other relations from Sec. 7.3 remain valid after changing the classical
dynamical variables to operators in the Heisenberg representation. It is important
that the operator products are written in the symmetrized form, for instance,
|a|2 (a a + aa )/2 = a a + 1/2 = aa 1/2, (7.172)
where we used (7.165) for obtaining the last equalities. Note that in the operator

identities similar to (7.172), 1/2 is understood as I/2, where I is the unity, or
= .
identity, operator, I
Spatial harmonics of a free field depend on time harmonically with the fre-
quency k = ck. Hence, we find the two-time commutators,
[ak (t), ak0 (t0 )] f ree = kk0 exp[ik (t t0 )], (7.173)
and similar relations for other field variables. In the presence of external currents,
(7.173) can be replaced by a more complicated dependence on t and t0 , but it
should turn into (7.164) at t = t0 . This conservation of commutation relations
follows from the unitarity of the operators time evolution.
7.4.2 Quantization of macroscopic field in matter

Macroscopic field in a non-magnetic material is described by Maxwells equations
with the phenomenological dielectric function (in the linear approximation). In
the transparency windows, 00 () 0, and the energy of the free field is preserved,
so that we can again use the Hamiltonian formalism and quantize the field vari-
ables. If we also neglect the dispersion 0 (), then the procedure will be similar
to the one of Sec. 7.3, with the only changes in the speed of light (c c/n) and
in the orientation of the polarization unit vectors ek (in the case of an anisotropic
medium).
One can show [Klyshko (1980)] that with an account for linear dispersion, the
relation between the macroscopic field E(r, t) and the photon creation and annihi-
lation operators ak , ak in a transparent medium has the form (7.124) provided that
the ck coefficients are multiplied by a factor of
!1 1/2 !1/2
2
uv 1
k 2 e e = 2 , (7.174)
k
c cos k nk
where vk k /k = c/nk and uk k /k cos k are the phase and group veloci-
ties, respectively and k is the angle between the ray and wave vectors. The spatial
harmonics of the magnetic field in (7.124) should be then multiplied by nk .
7.4.3 Quantization of the field in a cavity

The field in a closed cavity with ideally reflecting walls can be represented as a
sum of real orthogonal eigenfunctions uk (r), vk (r) of the corresponding boundary
problem,
X X
E= pk (t)uk (r), H = k qk (t)vk (r), (7.175)
k k
where k are the cavity eigenfrequencies.

For instance, free field in a rectangular cavity is a superposition of stand-
ing plane waves. (The field is not transverse in this case, see [Landau (1982)])
The allowed values of the wavevector are determined by the cavity dimensions
L , = x, y, z (compare with (7.103)),
( )
l m n
k= , , , l, m, n = 0, 1, 2, . . . (7.176)
L x Ly Lz
A standing plane wave is a superposition of two counter-propagating waves
with the equal amplitudes ak = ak ; in this case, according to (7.123), E
pk cos kz and H k qk sin kz. (The scaling coefficient can be found from (7.134)
at L3 L x Ly Lz and is on the order of (16/L3 )1/2 .) Hence, taking into account the
uncertainty relation qp 1 ~/2, one comes to the conclusion that the accuracy
of a simultaneous measurement of E(r, t) and H(r, t) inside the cavity is limited.
Sometimes, the free-space field is also expanded in standing waves of the form
cos k r, sin k r, but then the amplitudes of the standing waves do not have a direct
relation to the values observed in experiment. Indeed, for selecting the plane wave
+k, the detector should be placed in the far-field zone of the emitter, where the
k wave is absent. The relation between the far field and the ak operators is
considered in [Klyshko (1980)].

7.5 States of the field and their properties
Next, we consider the various states of the field, both pure and mixed ones, and
their properties, as well as the mean values and distributions of the observables in
these states. It is convenient to use some basis set of wave functions, so that an
arbitrary state can be represented as an expansion over this basis. This procedure
is similar to expanding an arbitrary vector over the set of unit vectors of some
frame of reference in real space. We will consider the basis sets generated by
various operators: energy H, coordinate q, momentum p, photon annihilation
a q + ip, as well as relations between these bases. In this consideration,
we will use compact Diracs notation, which will be briefly described in the next
subsection.
7.5.1 Diracs notation

An arbitrary instantaneous state of a quantum system is defined by the wave func-
tion (x) hx|i, where x is some set of variables, discrete or continuous, that are
sufficient for the complete description of the system. A complete description of a
one-dimensional spinless particle, in particular, of an oscillator or a field mode,
is given by a single variable: the coordinate (x q), the momentum (x p), or
the energy (x E).q
The function hx|i is called the x-representation of the system. The state itself,
without specifying the representation, is denoted by |i, or | i, or |ti. The complex
conjugated function, (x), is denoted by h|xi = hx|i , i.e., one can write | i =
h |, h | = | i.
In the x-representation, the state | i is given by a set (discrete or continuous)
of numbers hx1 | i c1 , hx2 | i c2 , . . . , which can be naturally considered as
the components of some vector in a multi-dimensional space. Then, hxn | i is an
analogue of the inner product of the unit vector hxn | and the state vector | i, i.e.,
the projection of | i onto the n-th axis. Any vector can be represented as a sum of
the unit vectors multiplied by the cn coefficients,
X X
|i= cn |xn i |xn ihxn | i, (7.177)
n n
or, in more compact notation,

X
|i= |nihn| i. (7.178)
n
Similarly,
X
h|= h |nihn|. (7.179)
n
In the case of a continuous variable, summation in (7.177) and other similar equa-
tions is replaced by integration,
Z
|i= dx|xihx| i. (7.180)
The vectors h | and | i are called, respectively, the bra- and ket-vectors (being parts
of a bracket).
q Recall that in classical mechanics, the state is given by the numbers q, p, while in quantum mechanics,
by the function (q) (or (p), (E), . . . ).

Projections of different unity vectors onto one another are equal to zero (in the
case of an orthogonal system),
hn|n0 i = nn0 , hx|x0 i = (x x0 ). (7.181)
One says that a frame of reference (a basis) is complete if any vector can be
represented in the form (7.177). The completeness can be expressed in the form
of a tensor equation
X
I = |nihn|. (7.182)
n
Here, I is a unit tensor (I| i = | i), and |aihb| denotes a dyadic tensor, or the outer
product of the vectors |ai and hb|. The action of a dyadic on the vectors is obvious
from its notation,
{|aihb|}| i |aihb| i = hb| i|ai,
(7.183)
h |{|aihb|} h |aihb| = hb|h |ai.
The tensor |aiha| Pa with ha|ai = 1 is called a projector since its action on a
vector | i selects the component of this vector along |ai: Pa | i = |aiha| i ca |ai.
The expansion of the unity (7.182) provides an easy way for forming various
representations of scalars ha|bi, vectors | i, tensors (operators) f :
X
ha|bi = ha|I|bi = ha|nihn|bi, (7.184)
n
X
| i = I| i = |nihn| i, (7.185)
n
X
f = IfI = fnn0 |nihn0 |, (7.186)
nn0
with fnn0 hn| f |n0 i in the last equation.

An operator f can act on the right on a ket vector and on the left on a bra
vector, creating new vectors, f |ai |bi and ha| f hc|, with other directions and
lengths. (The length or, more precisely, the norm of a vector |ai is defined as the
number ha|ai1/2 .) If it is only the length of a vector that is changed, the vector is
called the eigenvector (right or left) for this operator. It is convenient to denote an
operator and its eigenvectors and eigenvalues by the same character,
f| fn i = fn | fn i. (7.187)
The set of eigenvectors | fn i usually forms a basis, not necessarily orthogonal. The
operator f , Hermite conjugated to f , is defined by the equation
f |ai = {ha| f } , (7.188)
or ( f )ab = ( fba ) . If f = f , f is an Hermitian operator, and fab = fba

, fn =

fn , h fn | fn0 i = 0 (at fn , fn0 ).
In quantum mechanics, it is postulated that the probability distribution P( f |t)
for the observable f at a time t for an ensemble of systems being in the same
quantum state |ti is determined by the projections of |ti onto the eigenvectors | f i
of the operator f,
P( f |t) = C|h f |ti|2 = Cht|P f |ti, (7.189)
1
where C is the normalization sum or integral (C = 1 for normalized vectors).
If the observable f has a continuous spectrum, then P( f ) has dimensionality 1/ f
and is the probability distribution density. The distribution (7.189) is defined in
terms of the Schrodinger variables. In the Heisenberg representation, it has the
form
P( f |t) = Cht0 |P f (t)|t0 i, (7.190)
P f (t) | f (t)ih f (t)|. (7.191)
In the case of a mixed state, the operator P f should be averaged with the
density operator. For instance, in the Schrodinger representation,
P( f |t) = Tr{| f ih f |(t)} = h f |(t)| f i, (7.192)
i.e., the distribution of an observable f is defined by the diagonal elements of the
density matrix in the f -representation (the populations).
According the postulate of the wave function reduction, when some value (for
instance, energy) is measured by means of a classical device, the measurement
brings the system from the initial state | i into the state |E1 i, where E1 is deter-
mined by the reading of the detector. Thus, a measurement is simultaneously the
preparation of a system with a known wave function. In order to prepare a system
in a given state |E1 i, one has to measure the energy of a sufficiently large number
of systems in different initial states until a necessary reading E1 is achieved. When
a classical device shows a reading f this means it has brought the system into the
state P f | i = | f i; the back-action of the device on the state is described by the
projector P f | f ih f |.
7.5.2 Energy states

Usually, it is the energy that is measured in quantum optics; correspondingly, as
the basis one usually chooses the set of the eigenstates of the energy operator for
separate modes of the free field, i.e., the harmonic-oscillator Hamiltonian,
H0 = (p2 + 2 q2 )/2 = ~(N + I/2), (7.193)
where N a a is the photon-number operator for the chosen mode. The k index
will be omitted, as a rule, whenever we consider only a single mode. By definition,
the eigenvalues and eigenfunctions satisfy the equality
H0 |Ni = EN |Ni. (7.194)
According to (7.193), the energy states |Ni or, in short, the N-states, also
known as Fock states, are also eigenstates for the a a operator, (N N)|Ni = 0,
where N E/~ 1/2.
The H0 operator is Hermitian; therefore, the vectors |Ni form a complete or-
thogonal normalized frame of reference,
X
hN|N 0 i = NN 0 , |NihN| = I, (7.195)
N
X X
|i= |NihN| i, f = fNN 0 |NihN 0 |. (7.196)
N NN 0
Using the commutation rule [a, a ] = I, it is not difficult to show (see [Klyshko
(1980)]) that N are integers,
EN = (N + 1/2)~, N = 0, 1, 2, . . . (7.197)
Thus, the energy of a single mode can only take a set of discrete equidistant
values differing by ~, the energy of a photon. A mode has the smallest possible
energy if N = 0, which corresponds to the vacuum state |0i, while the states with
N > 0 are called N-photon states.
It is also easy to find out how the a, a operators act on N states,
a|Ni N 1/2 |N 1i, (7.198)
a |Ni (N + 1)1/2 |N + 1i. (7.199)

These relations explain why a, a are called photon annihilation and creation oper-
ators; they also show that N-states are not eigenstates for the operators q, p, a, a.
Hence, if the field is in some N-state (including the vacuum state), the measure-
ment of the electric field will reveal quantum fluctuations. This conclusion imme-
diately follows from the fact that H0 does not commute with q, p.
According to (7.199), the N-state can be obtained by acting N times on the
vacuum state by the operator a ,
|Ni = (N!)1/2 (a )N |0i. (7.200)
However, in practice it is very difficult to bring some free-field or cavity mode into
a pure N state (except the vacuum one), especially for N 1 2. Usually, the actual
state of a mode is an incoherent mixture of a few first N states, while the state of
an ideal laser is a coherent mixture of many N-states.
All considerations given above related to a fixed time moment. The time
dependence of the state vector of a single free-field mode is described in the
Schrodinger representation by the equation
i~d|ti/dt = H0 |ti, (7.201)
and if at time t = 0 the mode was in an N-photon state, then, according to (7.194),
|tiN = |NieiNt |N, ti, (7.202)
where = ck. Hence, in an N-state, the mean values, moments, and distributions
of all observables, including q, p, are stationary. An arbitrary pure state of a mode
depends on the time as
X
|ti = cN |N, ti = c0 |0i + c1 eit |1i + . . . , (7.203)
N
where cN hN|t0 i. In the presence of currents, the coefficients in this expansion

are time-dependent, although this dependence is usually slow compared to the one
of exp(it).
So far, we have been discussing the state of a single mode. In the case of
independent modes, the energy wave function of the total field is obtained by
simply multiplying the energy wavefunctions for the modes with definite photon
numbers,
|{Nk }i k |Nk ik |N1 , N2 , . . . i, (7.204)

X
|{Nk }, ti = |{Nk }i exp(iEt/~), E ~ Nk k . (7.205)
k
Thus, an energy state is fixed by giving the photon numbers {Nk } in all modes
(occupation numbers), while an arbitrary state can be represented as a superposi-
tion of states with all possible combinations {N},
X
|ti = c({Nk })|{Nk }, T i. (7.206)
{Nk }
In the presence of external currents, the state amplitudes c({Nk }) become time-
dependent, and can be only determined by applying the perturbation theory (com-
pare with Sec. 2.1).
7.5.3 Coherent states

As it was shown by Glauber, another convenient basis is formed by the eigenstates
of the non-Hermitian photon annihilation operator,
a|zi = z|zi. (7.207)
The states |zi are called coherent states. Since a q + ip, one can expect that the
spectrum of a is continuous and complex, i.e., z = z0 + iz00 is an arbitrary complex
number.
From the definition (7.207), it follows that the action of an arbitrary operator
function f (a) on the vector |zi is reduced to a simple multiplication of this vector
by a usual (c-number) function f (z),
f (a)|zi = f (z)|zi. (7.208)
The equation conjugated to (7.207) has the form
hz|a = z hz|, (7.209)
i.e., ket-vectors hz| are left-hand eigenvectors for the a operator.

Using (7.207), (7.208), (7.209), we immediately find the mean number of pho-
tons in a coherent state |zi,
hNiz hz|N|zi = |z|2 . (7.210)
In the case of a coherent state, all factorial moments of the photon number are
also calculated in a simple way (see (7.170)),
G(m) m
z hz| : N : |zi = |z|
2m
= hNim
z . (7.211)
This equality shows that the moments are factorable.

In (7.211), we have used the notation : N m :: a a . . . a a : am am . Gen-
erally, the colons denote the operation of normal ordering, which means placing
all a operators on the right of all a operators. This operation ignores the non-
commutativity of the operators, i.e., between the colons the operators can be writ-
ten in any order. Note that : : is a nonlinear operation; for instance, the operator
: aa :=: a a + I := a a is not equal to the operator : a a : +I = a a + I.
From (7.207), (7.209), (7.208), it follows that
h: f (a , a) :iz = f (z , z). (7.212)
For finding the mean value of an arbitrary operator f (a , a) in a coherent state, it

is sufficient to represent this operator, using the equality aa = a a + I, as a sum
of normally ordered operators, and then replace a by z and a by z. Often, normal

ordering can be performed through the series expansion,
X
f (a , a) = cmn am an ,
mn
X (7.213)
h f (a , a)iz = cmn zm zn .
mn
For instance,
haa iz = ha a + Iiz = |z|2 + 1,
N 2 a aa a = a (a a + I)a =: N 2 : +N, (7.214)
2 4 2
hN iz = |z| + |z| .
Using the last equality, one can write the second moment and the variance hN 2 i
hN 2 i hNi2 in terms of the first moment,
hN 2 iz = hNiz (hNiz + 1), hN 2 iz = hNiz . (7.215)
This relation is typical for a Poissonian random variable (see below).
In the general case, normal ordering of an arbitrary operator is not a simple
task (see examples in Ref. [Louisell (1964)]). Sometimes, the following operator
identity is helpful [Klauder (1968)]:
exp(a + a) = C exp(a ) exp(a) = C 1 exp(a) exp(a ), (7.216)
where C exp(/2). At = , the operator in (7.216) is called the displace-
ment operator and is denoted as D(),
D() exp(a a). (7.217)
From (7.200), (7.217), and the equality obtained below, (7.222), it follows that
D(z)|0i = |zi. (7.218)
One can show [Bloembergen (1965)] that a classical external current jk converts
a mode from a vacuum state, |0i, into a coherent one, |zi, i.e., its action can be
described by the displacement operator. The amplitude z coincides in this case
with the classical amplitude found from (7.128).
Let us show that an oscillator in a coherent state will indeed have a Poissonian
distribution of the energy, i.e., that in a coherent state photons behave, in a sense,
like a chaotic flow of sand grains. For this, we will find the transformation matrix
hN|zi relating the bases |Ni and |zi. Let us left-multiply (7.207) by hN|, which
yields
hN|a|zi = zhN|zi. (7.219)
Hence, taking into account (7.188) and (7.199), we obtain

(N + 1)1/2 hN + 1|zi = zhN|zi.
From this recurrent relation, we find
hN|zi = (N!)1/2 zN h0|zi. (7.220)
The remaining unknown factor h0|zi can be assumed to be real; then it can be
found from the normalization condition,
X
hz|zi = |hN|zi|2 = h0|zi2 exp |z|2 = 1. (7.221)
N
Hence, the expansion of a coherent state over Fock states follows,

X
|zi = |NihN|zi, (7.222)
N
hN|zi = (N!)1/2 zN exp(|z|2 /2). (7.223)

Recall that hN|zi is the N-representation of a coherent wave function, while
hz|Ni = hN|zi is the z-representation of a number-state wave function. Therefore,
the probability of measuring N photons per mode in a coherent state is determined
by a Poissonian distribution with the parameter hNi = |z|2 ,
P(N|z) = hNiN exp(hNi)/N! (7.224)
With the help of (7.223), one can easily verify that different z-vectors are not
orthogonal to each other (in contrast to N-vectors),
X
hz1 |z2 i = hz1 |NihN|z2 i = exp[|z1 z2 |2 /2 + iIm(z1 z2 )], (7.225)
N
|hz1 |z2 i|2 = exp(|z1 z2 |2 ). (7.226)

This fact does not cancel the completeness property (as in the case of a usual
oblique-coordinate reference system): any arbitrary vector or diadic tensor can
be still expanded over the set of |zi vectors or |z1 ihz2 | diadic tensors, respectively.
Indeed, it follows from (7.222) that
2
X zM zN
|zihz| = e|z| |MihN|. (7.227)
MN
(M!N!)1/2
Let us sum these diadic tensors over all z ei ,

Z X Z Z 2
2
d2 z|zihz| = |MihN|(M!N!)1/2 de M+N+1 dei(NM) ,
MN 0 0
(7.228)
where d2 z = dz0 dz00 = d d. The integral in yields 2N M , while the integral

in is equal to N!/2. Hence, taking into account (7.195), we obtain the z-diadic
expansion of the operator unity, i.e., the completeness condition, in the form
Z
I = 1 d2 z|zihz|. (7.229)
Hence, an arbitrary vector can be represented as

Z
| i = 1 d2 z|zihz| i. (7.230)
In particular,
Z
1
|z1 i = d2 z|zihz|z1 i. (7.231)
This relation, together with (7.225), shows that the basis vectors |zi can be ex-
pressed in terms of each other, i.e., the z basis is over-complete. Roughly speak-
ing, it means that the number of coordinates exceeds the dimensionality of the
space.
The time dependence of a coherent state can be easily found by substituting
(7.202) into (7.222),
X
|z, ti = |NihN|zieiNt = |zeit i |z(t)i. (7.232)
Thus, free evolution does not turn a coherent state into some other type of
state, similarly to the case of an energy state (see (7.202)) and in contrast to the
cases of q- and p-states (see below).
If all modes are in coherent states, then the state vector of the field will be
|{zk }i = |z1 i1 |z2 i2 |z1 , z2 , . . . i. (7.233)
According to (7.131), this vector is the eigenvector for the positive-frequency
(+)
field operator E (r, t) with the eigenvalue
X
E(+) (r, t) i ck ek zk exp(ik r ik t). (7.234)
k
The mean value of the field in a coherent state is equal to the real part of this
expression,
h{zk }| E(r, t)|{zk }i = 2ReE(+) (r, t). (7.235)
Further, according to (7.212), all normally ordered field moments (correlation
functions),
G(n) () () (+) (+)
1...2n h E 1 . . . E n E n+1 . . . E 2n i, (7.236)
which usually determine the readings of optical detectors, are factorable in the
case of a coherent state, i.e., can be expressed in terms of the products of the first
moments,
G(n) () (+)
1...2n ({zk }) = E 1 . . . E 2n . (7.237)
Here, the eigenvalues of the fields Ei() Ei (ri , ti ) are determined by the set {zk }
according to Eq. (7.234).
7.5.4 Coordinate and momentum states

By definition, the eigenvectors |qi of the coordinate operator q satisfy the condi-
tion
q|qi = q|qi. (7.238)
From q = q, it follows that q = q and hq|q = qhq|. Similarly, one can define
the eigenvectors |pi of the momentum operator, p|pi = p|pi. The spectrum of
q is continuous; therefore, the q-representation of an arbitrary state vector, the
orthonormality condition and the completeness condition have the form
Z
|i= dq|qihq| i, (7.239)
hq|q0 i = (q q0 ), (7.240)
Z
I = dq|qihq|. (7.241)
According to (7.240), q-vectors have infinite norm, hq|qi = , which leads to

certain difficulties. As an example, let us find, using the general rule (7.189), the
probability density for the coordinate of a system in a state | i = |q1 i:
P(q|q1 ) = C|hq|q1 i|2 = C[(q q1 )]2 , (7.242)
Z Z
C 1 = dq|hq|q1i|2 = dq[(q q1 )]2 . (7.243)
The squared delta-function has a meaning provided that one of its represen-
tations with a finite width q is used. In this case, (0) = 1/q (see (6.139)),
and one can replace (q)2 by (q)/q. The width q is chosen from the physical
considerations: it should be much less than the interval on which the functions
that are multiplied by the delta function before the integration vary considerably.
In this example, however, q is canceled since C = q,
P(q|q1) = (q q1 ). (7.244)
Sometimes, it is more convenient to use discrete q and p representations with

the usual normalization, hqm |qn i = hpm |pn i = mn . For passing to discrete spectra
{qn } and {pn }, the wave functions (q) hq| i or their Fourier transforms (p)
hp| i should be considered either periodic or differing from zero only within finite
intervals, L and ~K, respectively (compare with the discretization procedure for
wavenumbers in Sec. 7.3). This restriction is equivalent to the assumption that
(q), (p) vary very little on the intervals q 1/K, p ~/L, and is always
valid for sufficiently large L and K.
Further, let us find the transformation function hq|pi, i.e., the q-representation
of a p-state, by assuming
d
hq| p = i~ hq|. (7.245)
dq
In this relation, q is a continuous parameter of the vector hq|, and the differentiation
operator acts on this parameter. Multiplying (7.245) by |pi, we obtain the equation

i~ hq|pi = phq|pi, (7.246)
q
whose solution, evidently, is
hq|pi = (2~)1/2eipq/~ . (7.247)
The normalization constant here was found by substituting into (7.240) the diadic
expansion of a unity,
Z
d phq|pihp|q0i = (q q0 ).
According to (7.247), in a q-state, all momentums are equally probable, while in

a p-state, all coordinates are equally probable,
P(p|q) |hp|qi|2 = 1/2~,
(7.248)
P(q|p) |hq|pi|2 = 1/2~.
Similarly, one can find a function of two variables hq|zi z (q), whose square
determines the probability density distribution P(q|z) for the coordinate in a z state.
(Note that the symbol P(z|q) is meaningless since the non-Hermitian operator a
does not correspond to any physical observable.) From (7.121) and (7.245), we
obtain that

hq|a = (q + d/dq)hq|/ 2, (7.249)
where q (m/~)1/2 q and we have introduced the mass m of an equivalent oscil-
lator. Multiplication by |zi with an account for (7.207), (7.209) yields

(/q + q 2z)hq|zi = 0. (7.250)
This equation is satisfied by the function

2
hq|zi = C1 (z) exp[(q 2z) /2]. (7.251)
Similarly,

hp|zi = C2 (z) exp[( p + i 2z)2 /2], (7.252)
where p p/(~m)1/2. The normalization constants are defined here only up to
the phase factors,
C1 = (m/~)1/4 exp[z002 + i1 (z)],
(7.253)
C2 = (~m)1/4 exp[z02 + i2 (z)].
According to the definition (7.121),

q = (a + a )/ 2, p = (a a )/i 2, (7.254)
hence it follows from (7.207), (7.209) that

hqiz hz|q|zi = 2z0 , h piz = 2z00 . (7.255)
As a result, the distributions following from (7.251), (7.252) can be represented
in the form
P(q|z) = 1/2 exp[(q hqi)2 ],
(7.256)
P( p|z) = 1/2 exp[( p h pi)2 ].
Thus, for an oscillator in a coherent state, the coordinate and amplitude have
Gaussian distributions with the variances
hq2 i = ~/2m, hp2 i = ~m/2, hq2 i = h p2 i = 1/2 (7.257)
r
and the minimal possible product of uncertainties,
qp = ~/2. (7.258)
The distributions of the coordinate and momentum at z , 0 differ from the
0
vacuum
00 one only by the displacement of the center of reference by 2z and
2z ; their variances are not increased, in contrast to the energy variance (see
(7.215)). The relative widths of the distributions, q/hqi and p/hpi, are inversely
proportional to z0 and z00 .
It follows from (7.232) that the evolution of an oscillator
in a coherent state
is described by changing z to z0 eit . Let z0 = z0 q0 / 2, then, according to
(7.255), one should set in (7.256)
hqi = q0 cos t, h pi = q0 sin t. (7.259)
r Here, as usual, x denotes two different values, the operator x hxi and the number h(x hxi)2 i1/2 .
Fig. 7.18 Distribution of the coordinate for an oscillator in a Fock state (top) and
a coherent state
(bottom) with the same mean photon numbers, equal to 5. The coordinate q is in ~/m units.
As a result, the distributions shift without changing their shapes (Fig. 7.18),
P(q|z, t) = 1/2 exp[(q q0 cos t)2 ],
(7.260)
P( p|z, t) = 1/2 exp[( p + q0 sin t)2 ].
Thus, the mean coordinate and momentum in the case of a coherent state de-
pend on time the same way as the corresponding values of a classical oscillator.
As |z0 | increases, the relative fluctuations are reduced and a quantum oscillator
becomes more and more similar to a classical one.
Note that the wave function (7.251) with the substitution zt z(t) = z0 eit
should satisfy the Schrodinger equation in the q representation,
2
!
2i 2
+ q z (q, t) = 0,
t q2 (7.261)
z (q, t) hq|zt i.
(Here and below, we use dimensionless variables q q, p p.) This condition
allows one to find the phase in (7.253). As a result, the coherent wave function
can be represented as

z (q, t) = 1/4 exp[(q 2zt )2 /2 z002 0 00
t + i(zt zt t/2)]
= 1/4 exp{(q hqt i)2 /2 + i[(q hqt i/2)hpt i t/2]}. (7.262)

Above, we have found the matrices of transitions from the q-representation
to p- and z-representations. Similarly, one can find the functions hN|qi N (q)
determining the coordinate distribution in Fock states and the photon-number dis-
tribution in q-states. These functions satisfy Eq. 7.261 with /t replaced by
iN; they are equal to the Hermite polynomials multiplied by the vacuum func-
tion h0|qi = exp(q2 /2). They are obtained if one multiplies (7.200) by hq| and
replaces hq|a with 21/2 (q d/dq)hq|,
hq|Ni = (2N N!1/2 )1/2 (q d/dq)N exp(q2 /2). (7.263)
7.5.5 Squeezed states

One should keep in mind that in the case of all states except the energy (Fock)
ones, the distributions and moments are time-dependent (see (7.259), (7.260) for
a coherent state). For instance, one can show that q-states periodically become
p-states and vice versa (Fig. 7.20).
Consider the evolution of variances for the coordinate and momentum in the
case of an arbitrary initial state. Solutions to the Heisenberg equations for the
coordinate and momentum operators have the classical form,
q(t) = q cos + p sin , p(t) = p cos q sin , (7.264)
where t, q q(0), p p(0).

In the general case, from (7.264) we can find the time dependence of the co-
ordinate and momentum variances,
Dq () = D p ( /2) = Dq cos2 + D p sin2 + Dqp sin 2. (7.265)
Here, the following notation was introduced:
D x (t) h[x(t)]2 i, x(t) x(t) hx(t)i, D x D x (0),

Dqp (t) hq(t)p(t) + p(t)q(t)i/2
= ha(t)2 a (t)2 i/2i hq(t)ihp(t)i;
the averaging runs over the initial state of the oscillator |t0 i. Thus, the variances
of the coordinate and momentum oscillate anti-phased with the frequency 2, and
their sum is an integral of motion,
Dq (t) + D p () = Dq + D p = 2(hNi ha ihai) + 1 = 2Daa . (7.266)
According to (7.265), the variances are constant only under the condition that
Dq = D p and Dqp = 0. Using (7.254), one can verify that this is the case for Fock
states (Dq = D p = N + 1/2) and coherent states (Dq = D p = 1/2).
Fig. 7.19 Coherent (a) and squeezed (b,c) states of an oscillator. The figure shows the time depen-
dencies of the mean coordinate and the coordinate uncertainty.
The dependencies are calculated using
Eq. (7.265) with q0 = 5, Dqp = 0, qp = 1/2 and q = 1/ 2 (a), 0.2 (b), 2.5 (c).
Recently, it was proposed to generate and measure the so-called squeezed

states [Walls (1983)], for which Dq 1/2 (or D p 1/2) and Dq D p = 1/4
(Fig. 7.19).s For a mechanical oscillator or electromagnetic field in such a state,
repeated stroboscopic measurements with an appropriate phase will reveal fluc-

tuations reduced with respect to the zero-point vacuum ones, ~/2m. Thus, in
principle, zero-point vacuum fluctuations do not restrict the limiting accuracy of
the coordinate or momentum measurement. Squeezed states can be of interest for
the information transmission and for the measurement of tiny forces caused, for in-
stance, by gravitational waves [Braginsky (1980)]. Note that up to now, attention
s Editors note: About the same time as the book was published, squeezed states were produced in
experiment, first through four-wave mixing and then via parametric amplification, see [Bachor (2004);
Walls (1994)]. Amplitude squeezing of photocurrent was also observed using a negative feedback
loop [Yamamoto (1999)].
Fig. 7.20 Various states of a quantum oscillator shown on the phase plane. The horizontal and verti-
cal sizes of the figures correspond to the uncertainties in the coordinate and momentum, respectively;
the sizes along the radius and the azimuth correspond, respectively, to the amplitude and phase un-
certainties. 1, the vacuum state; 2, a coherent state; 3, a coordinate state; 4, a squeezed state; 5, a
momentum state; 6, a photon-number state; 7, a phase state.
is attracted to certain questions in the quantum measurement theory [Braginsky

(1975, 1980)].t
We have considered four types of states generated by the operators p2 +q2 , p+
iq, p, q, as well as the relations between these states. Similarly, one can construct
a lot of other states. Worth mentioning are the eigenstates of the phase opera-
tor [Fain (1972); Loudon (2000)], which are most close to a classical oscillation
with a fixed phase.
It is convenient to show various states on a phase diagram (q, p) as figures of
different shapes whose linear dimensions are equal to the uncertainties q, p in
these states (Fig. 7.20). The area of each figure cannot be less than about a unity.
A classical state of an oscillator is shown by a point (q1 , p1 ), a coherent state, by a
circle of a unity diameter and the center at point (q1 , p1 ); an N-photon state, by a
thin circle with the diameter (N + 1/2)1/2 centered at the origin; a q-state, by a thin
t Editors note: nowadays, squeezed states still attract a great attention in connection with different
applications. We refer to the following ones:

optical communication and optical measurements, due to the fact that weaker signals can be trans-
mitted with the same signal-to-noise ratio and the same light power;
gravitational-wave detection;
precise measurement of spatial displacements with multimode squeezed states; super-resolution;
quantum imaging with surpassing the quantum noise limit;
one-way quantum computing with cluster states based on squeezed states;
quantum memory based on squeezed-light interaction with atomic ensembles and single atoms;
research in fundamental quantum physics.
vertical straight line; a p-state, by a thin horizontal straight line.u However, one
should keep in mind that these figures have only qualitative meaning; strictly, they
do not correspond to any joint distributions P(q, p), which do not exist in quantum
mechanics.
Similar to the states themselves, the figures depicting them on the phase plane
change in time due to the natural evolution (described by the Schrodinger equa-
tion) or due to the reduction as a result of the measurement back-action. For
instance, after an accurate measurement of q, a coherent circle in Fig. 7.20 will
turn into a vertical line. The evolution due to free oscillations is described by the
counter-clockwise rotation of the figure around the origin, with the angular rate
, or, alternatively, the clockwise rotation of the frame of reference q, p.
7.5.6 Mixed states

If the field is interacting (or has interacted) with another quantum object, for in-
stance, with an atom, then, by definition, there are no separate wave functions
for the field, (xE , t), and the atom, (xA , t); one can only speak of the joint wave
function, (xE , xA , t). Similarly, one cannot speak of the state vector | ik of a given
mode k if it is coupled with another mode k0 , or several modes. For instance, a
classical point-like source with the frequency excites a spherical wave, in which
case all plane waves with |k| = /c are coupled. Similar coupling of modes with
the same frequency (transverse modes) occurs due to the diffraction. Modes
with different frequencies (longitudinal modes) can be coupled due to the matter
anharmonicity [Klyshko (1980)].v
In all cases where the system is described by a non-complete set of variables,
one says that it is in a mixed state (Sec. 3.1). Then, instead of a state vector | i,
the system is characterized by a certain operator , called the density operator. In
the special case of a pure state, the density operator is a projector, pure = | ih |,
while for a mixed state, is given by a sum of projectors (see (7.186)).
The density operator, like any other field operator, can be written in vari-
ous bases (representations), in terms of the density matrices of the form NN 0
u Editors note: Probably one should add squeezed vacuum, a state generated at the output of an
unseeded optical parametric amplifier. In the diagram, it will be shown by an ellipse at the origin,
with the area being the same as for a coherent state [Bachor (2004); Walls (1994)].
v Editors note: Note that the description of a composite system consisting of two or several interacting
sub-systems (or sub-systems having interacted in the past) directly relates to the concept of entangled
states ([Peres (1993); Nielsen (2000); Bouwmeester (2000)], Sec. 7.5.7). This family of states was not
considered in the original book at all, partly due to the fact that in 1986, entangled states were not as
popular as nowadays. At the same time the author put a lot of efforts to avoid using vague terms while
describing physical phenomena ([Klyshko (1994)]).
hN||N 0 i, qq0 , zz0 and so on. In most cases, the density operator of the field is
written in the N-basis, but in some cases the z-basis is more convenient. For a
single mode, can be written in terms of N- and z-projectors as (the mode index
k is omitted)
X
= |NihN 0 | (7.267)
NN 0
Z
= d2 zP(z)|zihz|. (7.268)
Note that here, the diagonal z-representation is used, which is possible in many
cases and is provided by the over-completeness of the z-basis. Equation (7.268)
is called the Glauber-Sudarshan representation or the P-representation. The nor-
malization and Hermiticity conditions of are
X
NN = 1, NN 0 = N 0 N ,
N
Z
d2 z P(z) = 1, P(z) = P (z).
The mean value of any field operator f can be expressed in terms of accord-
ing to the relation h f i = Tr( f ) (Sec. 3.2), or according to (7.267), (7.268),
X Z
hfi = NN 0 fN 0 N = d2 zP(z)hz| f |zi. (7.269)
NN 0
Hence, with the help of (7.212), we find

Z
h: f (a , a) :i = d2 zP(z) f (z , z). (7.270)
Thus, the P(z) function provides an easy way to calculate the mean values of
normally ordered operators. In particular, the normally ordered moments can be
found as
Z
(n)
G = d2 zP(z)|z|2n . (7.271)
Equations (7.269), (7.270) show that the weighting function P(z) plays the
role
0of the probability
for the oscillator to have a complex amplitude z, i.e., q =
2z , p = 2z00 . However, P(z) can take negative values; besides, even with
P(z) = (2) (z z1 ), i.e., in the case of a pure coherent state, q and p have zero-point
fluctuations.w Therefore, P(z) is called a quasi-probability.
w Editors note: also called shot-noise fluctuations.
Moreover, the quasi-probability of a coherent state P(z) allows one to find the
probability distribution P( f ) of an arbitrary observable f . For this, one should
replace f in (7.269) by the projector P( f ) = | f ih f |,
Z Z
P( f ) = d2 zP(z)|hz| f i|2 = d2 zP( f |z)P(z). (7.272)
The two-dimensional Fourier-transform of the quasi-probability P(z) is called

the normally ordered characteristic function,
Z
a a
(, ) he e i = d2 zP(z)ez z . (7.273)
This definition yields a usual (not generalized) function for all states, in contrast
to P(z).x From the definition of , it follows that G(n) can be calculated by means
of differentiation, instead of more complicated integration,
!n

(n)
G = (, ) |=0 . (7.274)

Thus, a mixed state of a mode can be described by the NN 0 matrix, or one of
the functions P(z), (, ).
A mixed state of a multi-mode field is given by the density matrix
hN1 , N2 , . . . ||N10 , N20 , . . . i = h{Nk }||{Nk0 }i,
or by the quasi-probability P({zk }), or by its Fourier transform, ({k , k }). In
the case of independent modes, these values are factorable, reducible. It should
be stressed that with the help of the quasi-probability function, the operation of
quantum averaging of normally ordered operators (which are usually of interest)
takes the classical form (7.270), which is also maintained in the case of a multi-
mode field. For instance, the correlation functions (7.236) are found by averaging
their mean values in a coherent state (7.237) with the quasi-probability,
Z Z Y
G(n)
1...2n = . . . P({zk })G (n)
1...2n ({zk }) d2 zk . (7.275)
k
Recall that here, zk z(k, ) has the meaning of the amplitude (in 2ck units) of
the plane wave Ek0 propagating in the k direction and having polarization e .
Further, let us consider some examples of mixed states of the field.
In the case of a stationary field, the density operator does not depend on time,
i.e., [, H] = 0, which in the N-representation yields
X
h{Nk }||{Nk0 }i (Nk0 Nk )k = 0.
k
x For instance, P(z) for N states contains the 2Nth-order derivative of the delta function [Perina (1972)].
Hence, a stationary density matrix is diagonal in the occupation numbers of modes

with different frequencies. The function P({zk }) depends then only on the absolute
values |zk |, since zk (t) eit .
In a stationary state, all simultaneous moments, h f m (t)i, and correlation func-
tions, h f (t)g(t + ) . . . i, are independent of t. Usually, one assumes that the system
is ergodic, i.e., that the ensemble means h f i coincide with the values measured
in experiment, fexp (t), and averaged over time, and that the observed variations
(fluctuations) of fexp (t) in time are caused by the uncertainty of f in a pure quan-
tum or mixed ensemble. An important role is also played by the periodically
non-stationary states, which include, in particular, coherent states.
Often, one can use the approximation of statistically independent modes, =
Q
k k . Then, the diagonal element hNk |k |Nk i has the meaning of the population

(occupation number) of the kth mode N-photon state. In the case of a stationary
field with independent modes, populations fully characterize the properties of the
field. In particular, the mean photon number per mode, which determines the main
photometry parameter, the spectral brightness, is

X
hNk i = Tr(ak ak k ) = NhN|k |Ni. (7.276)
N=0
In an equilibrium state, the density operator is determined by the Gibbs dis-

tribution, k exp(Hk /T ), where T is the thermostat temperature (Sec. 3.2).
In an equilibrium state, the population of an N-photon level, i.e., the probability
to find simultaneously N photons in one mode, or the probability for the mode
energy to take the value ~(N + 1/2), depends on N exponentially,
PT (N) = CeN x , (7.277)
where
C = P(0) = 1 ex , x ~/T.
Equation (7.277) is called the Plancks, or geometric, distribution, as it forms

a geometric sequence, or the Bose-Einstein distribution. By substituting (7.277)
into (7.276), we find the mean number of photons per mode, also called the de-
generacy factor of the photon gas,
hNiT = (e x 1)1 N . (7.278)
This equality enables to be used instead of x as the parameter of the distribution;

then (7.277) takes the form
PT (N) = P(0)/(1 + 1/)N , P(0) = 1/(1 + ). (7.279)

In a non-equilibrium field, the modes can also have exponential energy distri-
bution, provided that they are chaotically excited by many independent sources.
This is the case for thermal radiation, fluorescence, or superfluorescence (in the
linear regime, see Sec. 7.1). Then, the dependence of hNk i on |k| = /c de-
termines the frequency spectrum of the radiation, while the dependence on the
direction, k/k, determines the angular spectrum and the direction of incoherent
radiation. Recall that in non-laser light, usually all hNk i 1. For instance, for the
green part of the sunlight spectrum, hNk i 102 , hence the probabilities to find
0, 1, and 2 photons in one mode are approximately equal to 0.99, 102 , and 104 .
One can show [Glauber (1965)] that the quasi-probability of a chaotically ex-
cited mode is a two-dimensional Gaussian function with the variance hNi/2,
PT (z) = exp(|z|2 /hNi)/hNi. (7.280)
Hence, taking into account (7.273),
T (, ) = exp( hNi) (7.281)
and, according to (7.274), only even symmetric moments are nonzero,
G(m) m m
T h: N :iT = m!hNi . (7.282)
It follows from (7.282) and (7.211) that an m-quantum transition is m! times as
probable in a thermal field than in a coherent field with the same hNi (see (6.212)),
which is due to the long tail of the thermal distribution. At m = 2, (7.282)
describes photon bunching (Secs. 7.2, 7.6).
By substituting (7.280) into (7.272), one can see that the distribution of the
coordinate and momentum for a thermal state are Gaussian as well, with zero
mean values and the variances determined by the relation hNi + 1/2 = h p2 + q2 i/2,
i.e.,
hq2 iT = h p2 iT = hNiT + 1/2 = (1/2) coth(x/2). (7.283)
Note that additive multi-mode parameters of the field, such as, for instance, the
electric field amplitude E at point (r, t), will have Gaussian distribution regardless
of the states of separate modes (provided that they are independent), by virtue of
the central limit theorem.
As we have already mentioned, a pure coherent state does not belong to the
class of stationary states, since z(t) = exp(it+i). However, one can construct
a stationary coherent state by forming a mixture of coherent states with the same
amplitudes and random phases . Such a state, apparently, is described by the
quasi-probability of the form [Glauber (1965)]
P(z) = (|z| )/2 = (|z|2 2 )/, (7.284)
which describes an ensemble of ideal lasers with uncertain phases. It is easy to

check, with the help of (7.272), (7.224), that the energy distribution for such an
ensemble will be still Poissonian, with hNi = 2 .
7.5.7 Entangled states (added by the Editors)

There is an important class of quantum states, named entangled states, which at
the time when the book was published were only starting to be discussed, mainly
in connection with the experiments on Bell tests. Now, these states form the base
for such important branches of modern science as quantum information and quan-
tum communications. The details can be found in many beautiful books and re-
view articles [Peres (1993); Scully (1997); Nielsen (2000); Bouwmeester (2000);
Bachor (2004); Bruss (2002)]. Below, we present just a brief review of their prop-
erties.
For the first time the entangled state was introduced by Schrodinger in 1935,
who mentioned that the best possible knowledge of the whole does not include the
best possible knowledge of its parts [Schrodinger (1935)]. One can interpret this
statement as follows: it is impossible to describe the subsystems A, B, . . . , forming
a composite system, in terms of wave functions while the whole state does possess
a wave function. A pure state of a bipartite system is called separable if and only
if it can be written as
| i = |A i |B i; (7.285)
otherwise it is entangled.
In the simplest case of two qubits, |1 i = 1 |01 i + 1 |11 i and |2 i = 2 |02 i +
2 |12 i, their joint state belongs to the 2 2 = 4-dimensional Hilbert space and, in
the general case, is not separable,
4
X
|12 i = c1 |01 i|02 i + c2 |01 i|12 i + c3 |11 i|02 i + c4 |11 i|12 i, |c j |2 = 1; (7.286)
j=1
apparently, |12 i , |1 i |2 i.
However, definition (7.285) does not tell us whether a given state is more
entangled or less entangled. For quantifying entanglement, there is a simple pa-
rameter introduced by Wootters, the concurrence C, which indicates how much
entanglement is stored in a composite state of two qubits:
0 C 2|c1 c4 c2 c3 | 1. (7.287)
If the state is separable, then C = 0. An example of maximally entangled two-
qubit states (C = 1) are the Bell states, which form a complete basis for two-qubit
systems:
1
()
12
[|01 i|12 i |11 i|02 i], (7.288)
2
1
()
12 [|01 i|02 i |11 i|12 i]. (7.289)
2
The state ()
12 is called the singlet state and plays a special role due to its remark-
able symmetry properties. The other three states form the triplet. For instance
the singlet state is invariant with respect to choice of basis. The Bell states are an
important tool in modern quantum optics but probably they are most famous for
violating the Bell inequality [Klyshko (1998); Grynberg (2010)]
Another way to quantify entanglement is through the so-called Schmidt de-
composition, which exists for any bipartite composite system:
D
X D
X
p
AB = j |u j i|v j i, = 1, (7.290)
j=1 j=1
where |u j i and |v j i are bases for subsystems A and B, both having dimensionality
D. The degree of entanglement can be estimated with the Schmidt number,
D 1
X
1 K 2j D, (7.291)
j=1
which can be interpreted as the number of nonzero Schmidt modes in the expan-
sion (7.290). It is easy to calculate the Schmidt number for the Bell states (7.288),
(7.289): K = 2. A state is separable if and only if K = 1. The Schmidt decom-
position is a very useful approach from the physical viewpoint as it allows one to
interpret the natural eigenmodes |u j i and |v j i of the system under study (called the
Schmidt modes) in terms of entanglement [Mandel (2004)].
Another closely related measure of entanglement is the von Neumann entropy,
D
X
S j log2 j = S (A ) = S (B ), (7.292)
j=1
where S (A,B) TrB,AAB is the reduced density matrix for subsystem A (B).
This definition returns us to the initial meaning of entanglement introduced by
Schrodinger. Indeed, the notion of the entropy relates to the uncertainty in the
(sub)system. For a pure state the von Neumann entropy equals zero, while for
completely mixed states it takes the maximal value.
The definition (7.285) can be also formulated for mixed states. Namely, a
mixed state is called separable if and only if it can be written as
X
j j j j
= p j |A ihA i |B ihB i, (7.293)
j
P
where j p j = 1, 0 p j 1. Otherwise, the state is entangled.
As an example let us consider the so-called Werner states,
W x| ih |+{1 x} | ih | + |+ ih+ | + |+ ih+ | + |+ ih+ | . (7.294)

It turns out that the state (7.294) is separable for x < 1/3.
Let us mention that sometimes one falsely associates entanglement with the
violation of the Bell inequalities. In this connection, the Werner states (7.294)
represent an important example: at 1/3 < x < 1/ 2 they are not separable (and
hence are entangled) but do not violate the Bell inequalities.
It is worth noting that the concept of entangled states was actively developed
in quantum information science and formally relates to systems of quantum bits
(qubits) or quantum dits (qudits) which are rather abstract notions. Nowadays
there are several physical systems that play the role of qubits, such as polarization
states of single photons, two-level atoms or ions in traps, etc. Real physical qubits
need to be carefully specified to avoid a contradiction with the formal description.
For instance, the well-known permutation property of identical particles leads to
entanglement since their wave function should be symmetrized, hence the singlet
state of two spin-1/2 particles takes the form
1
= [| i| i | i| i]. (7.295)
2
However, this sort of states can not serve as a resource of quantum information
since it is impossible to perform local operations over the subsystems and there
is no possibility to change the coefficients in the coherent superposition (7.295).
This fact was discussed by Zanardi and Peres [Peres (1993)] but using systems of
identical particles in the protocols of quantum information and quantum commu-
nication seems to be still an open question.
Starting with the pioneering experiments performed by Fry and independently
by Aspect [Grynberg (2010)] with two-photon fluorescence of atoms, entangled
states were actively studied in quantum optics. In particular, the most popular
object was the two-photon state (biphoton) [Klyshko (1998)] created via sponta-
neous parametric down conversion (see Secs. 6.5 and 7.6). Several types of bipar-
tite entangled states were suggested depending on the available degrees of free-
dom under consideration: entanglement between polarization and momentum, po-
larization and frequency, energy and time, frequency and momentum. The partic-
ular type of entanglement is determined by the phase-matching conditions and/or
further transformations performed over the biphotons. For instance, polarization-

momentum entangled state can be achieved by choosing the non-collinear degen-
erate regime of SPDC when two photons with the same frequencies propagate
along different directions , 0 and carry orthogonal polarizations H (horizontal)
or V (vertical):
1
= [|H i|V0 i + ei |V i|H0 i]. (7.296)
2
Sometimes the entangled state can be created by means of post-selection, i.e.
by taking into account only part of the state, for instance using a coincidence
circuit, which measures only events corresponding to incoming photons coincid-
ing in time. Such a scheme exploiting energy-time entanglement was suggested
by Franson. The scheme contained unbalanced Mach-Zehnder interferometers in
both signal and idler channels, which introduced delays exceeding the coherence
length (see Sec. 7.2) of each photon,
!1/2
d2 k
lcoh = ccoh = c | /2 l , (7.297)
d2 p
where l is the length of the crystal generating biphotons and p is the pump laser
frequency. Starting with a separable state of two photons (the same frequency but
different angles of propagations),
1 1
s,i = [|S i s + eis |Li s ] [|S ii + eii |Lii ] = s i , (7.298)
2 2
and using post-selection technique by picking up only photons passing through the
long (L) or short (S ) paths simultaneously, the final state (after the renormalization
caused by the non-unitary operation of post-selection) becomes an entangled one,
1
= [|S i s |S ii + ei(s +i ) |Li s |Lii ], (7.299)
2
where s,i are the phase delays introduced in the signal and idler channels. By
varying the phase delays one can observe the second-order interference in the
coincidence counts, while there is no modulation in the intensities both for the
signal and idler channels. Note that sign + in the phase of (7.299) is typical for
two-photon interference experiments (compare with the typical sign observed
for usual classical interference experiments).
Many other types of entangled states have been studied. We would like to
specially mention some important cases, namely:
bipartite multidimensional systems and the so-called Fedorovs ratio, which
serves as an operational entanglement quantifier;
bright-light entanglement and, in particular, quadrature and polarization entan-

gled states [Bachor (2004)];
multi-particle entanglement like GHZ and W-states.
A special class of entangled states, cluster states, attracts a great attention

due to one-way quantum computation scheme suggested by Raussendort, Browne,
and Briegel in 2003. Also, a special class of graph states seems to be useful for
quantum computation, quantum error correction, and other applications of multi-
particle entangled states.
Perhaps one of the most stupendous applications of entangled states relates to
quantum key distribution, where entangled photon pairs are used for establishing
a secret key between remote parties both in free space and using single-mode
fibres. Another field under an unquenchable interest is testing the foundations of
the quantum theory, where the use of multidimensional entangled systems gives
unexpected results with respect to the case of bipartite systems.

7.6 Statistics of photons and photoelectrons
Let us consider in more detail the statistics of the photon number N (see also
Ref. [Loudon (2000)]). For simplicity, we will mostly consider a single mode.
Below, we will show that the distribution P(N) NN or the moments hN m i
P m
N P(N) can be experimentally obtained from the statistics of the number of
photocounts, i.e., the number of electrons released by light from the photocathode
of a PMT during some sample time T . Such methods form the base for the optical
mixing spectroscopy [Cummins (1974)] where the traditional spectral analysis of
light is replaced by the statistical analysis of photocurrent at the PMT output.
7.6.1 Photon statistics

The distribution P(N) at a fixed time moment for an arbitrary instantaneous state
of the mode is determined by the general rule,
P(N) = Tr{|NihN|}, (7.300)
where |NihN| P(N) is the projection operator. Assuming in (7.272) f = N =

a a, we obtain the z-representation for NN ,
Z
2
P(N) = d2 zP(z)|z|2N e|z| /N! (7.301)
Further, using (7.270), one can write P(N) as an infinite series of normally ordered
moments of orders m 1 N,

1 1 X (1)k (N+k)
P(N) = h: N N eN :i = G . (7.302)
N! N! k=0 k!
Thus, the photon-number distribution has the form of a Poissonian one, N e /N!,
with the random parameter N, which requires additional quantum averaging
with normal ordering.
Above, we have considered two examples of P(N) distributions: the Poisso-
nian one, (7.224), for a coherent state, and the geometric one, (7.279), for a chaotic
(thermal) state. These are single-parameter distributions: they are fully character-
ized by, for instance, the first moment, hNi. One can show (see, for instance,
Ref. [Loudon (2000)]) that for a single-mode laser much above the oscillation
threshold, P(N) is Poissonian and for a laser below the threshold, it is geometric.
Near the threshold, the distribution has a shape that is intermediate between these
two limiting cases. This distribution, in the simplest models, is determined by two
parameters, for instance, hNi and the excess inversion above the threshold.
It is useful to consider another type of states, namely, an incoherent mixture
of the vacuum, |0i, and the K-photon state, |Ki. Then, P(N) differs from zero only
at two points,
P(N) = P(0)N0 + P(K)NK . (7.303)
Hence, hNi = KP(K), and all values can be expressed in terms of a single param-
eter, the mean photon number,
P(K) = hNi/K, P(0) = 1 P(K), (7.304)
hN m i = K m P(K) = K m1 hNi, 0 0 N 0 K. (7.305)
In the limiting case hNi = K, we obtain a pure energy state with K photons;
however, it is interesting to consider the more realistic case hNi K. Such states
can be generated via a K-photon decay of a single excited atom into one mode.
Repetition of this process in time leads to the appearance of K-photon light,
radiation consisting of K-photon groups. Two-photon light can be also generated
via the spontaneous (hNi 1) parametric down-conversion of usual light, in
which photons have Poissonian or Bose-Einstein distribution.
Figure 7.21 shows the plots of all three distributions we have considered.
Below, it will be shown that the termination of the K-photon distribution at
N > K leads to the effect of photon anti-bunching for hNi > K 1, while at
K > 1, hNi K 1, it leads to the effect of super-bunching. (Conventionally, the
Poissonian distribution is considered as having no bunching.)
Fig. 7.21 Typical photon-number distributions: (a) geometric; (b) Poissonian; (c) two-photon mixed.
The dashed line shows a Gaussian function with the mean value 9.5 (see (7.320)).
Often, calculations and interpretation of the experiments are much simplified

by passing, from usual moments hN m i, to factorial moments,
G(m) h: N m :i = hN(N 1) . . . (N m + 1)i, (7.306)
and their generating functions,
X
Q(x) (1 + x)N P(N) = h: e xN :i. (7.307)
N=0
The last equality follows from (7.302). Note that Q(x) differs from the he xN i
function, whose derivatives yield usual moments. If Q(x) is known, the factorial
moments and the distribution P(N) can be easily found by differentiating it at
points x = 0 and x = 1. Indeed, it follows from the definitions that
G(m) = Q(m) (0), (7.308)
P(N) = Q(N) (1)/N!, (7.309)

where
Q(m) (x) dm Q(x)/dxm . (7.310)
By substituting into (7.307) one-by-one the distributions (7.224), (7.279),

(7.303), we find the generating functions for the three considered types of os-
cillator states,
Qz (x) = e xhNi , (7.311)
QT (x) = 1/(1 xhNi), (7.312)
QK (x) = 1 + hNi[(1 + x)K 1]/K. (7.313)
Hence, according to the rule (7.308), we immediately obtain

G(m)
z = hNim , G(m) m
T = m!hNi ,
(7.314)
G(m)
K = hNi(K 1) . . . (K m + 1), 1 < m 0 K.
An important property of the generating functions (GFs) follows from the def-
P
inition (7.307): the GF Q(x) of a sum N = Ni of independent integers Ni is
equal to the product of the GFs for these integers,
Y
Q(x) = Qi (x). (7.315)
i
For instance, let a set of M similar atoms independentlyy emit K-photon light;
then, using (7.313) and (7.315), we find
M
Y
Q(x) = {1 + hNi i[(1 + x)K 1]/K} exp{hNi[(1 + x)K 1]/K}, (7.316)
i=1
P
where we assumed M and hNi i 0 with a finite hNi hNi i. In the case
K = 1, (7.316) becomes (7.311), i.e., weak single-photon radiation from a large
number of independent atoms results in a Poissonian distribution (in the absence
of interference). At K > 1, (7.316) describes Poissonian distribution for groups of
K photons. For instance, in the case of two-photon radiation,
Q(x) = exp[(2x + x2 )], hNi/2,
(7.317)
P(2N) = N e /N!, P(2N + 1) = 0.
Apparently, this example also includes the case where M is the number of inde-
pendent modes and N is the total number of photons in these modes.
Consider another example of applying the composition rule (7.315). Let us
focus on the total photon number N in M independent modes with the geomet-
ric distributions and the same mean photon numbers, hNk i = hNi/M (the
y Here we neglect interference, which one can do in the case of multi-mode detectors with a large
detection volume (Sec. 7.2).

degeneracy factor). The generating function, according to (7.312) and (7.315), is

M
Y
Q(x) = 1/(1 xhNk i) = 1/(1 x) M . (7.318)
k=1
Hence, at M 1, x 1, with finite hNi we once again obtain a Poissonian

distribution,
Q(x) e xhNi (M 1). (7.319)
Thus, the total number of photons in a large number of independent modes
with geometric distributions tends to a Poissonian distribution. Note that at hNi
1, a discrete Poissonian distribution can be approximated by a normal distribution
(Fig. 7.21(b)) with the first moment and variance coinciding,
p
P(N) exp{(N hNi)2 /2hNi}/ 2hNi. (7.320)
Relative fluctuations become small at N 1,
N/hNi = hNi1/2 = (~/hEi)1/2 1. (7.321)
Thus, for high mean energy and large mode number we have obtained a Gaussian
distribution, according to the central limit theorem and classical thermodynamics.
In the general case, (7.318) leads to the Pascal distribution,
! !
(k + M 1)! k 1 k1
G(k) = = hNik 1 + ... 1+ ,
(M 1)! M M
(7.322)
(N + M 1)! N
P(N) = .
N!(M 1)! (1 + ) M+N
Hence, assuming k = 2, we find that the relative fluctuations decrease with the
growth of the number of modes,
!1/2
N 1 1
= + . (7.323)
hNi hNi M
7.6.2 Photon bunching and anti-bunching

Among various types of instantaneous states of a quantum oscillator, the coherent
ones are special because, in a certain sense, they are most close to the state of a
classical oscillator with fixed coordinate and velocity. According to (7.224), the
distribution of the photon number N for a coherent state is Poissonian, so that the
variance coincides with the mean photon number,
hN 2 iz = hNi = |z|2 . (7.324)
Thus, the N distribution for a coherent quantum ensemble coincides with the
distribution of chaotically scattered classical particles over a set of cells, either in
space or in time. For an arbitrary state, the variance hN 2 i can certainly differ
from hNi; in this case, for hN 2 i > hNi one speaks of photon bunching, while
for hN 2 i < hNi, of photon anti-bunching. These effects can be quantitatively
described in terms of the normalized second factorial moment,
h: N 2 :i hN 2 i hNi hN 2 i hNi
g 2
= 2
= 1+ . (7.325)
hNi hNi hNi2
For a Poissonian distribution, g = 1, in the case of bunching g > 1, and in
the case of anti-bunching g < 1. Because the rate of a stimulated two-photon
transition scales as h: N 2 :i, g determines the ratio of two-photon efficiencies for
a given field and for a coherent field with the same mean energy. Note that if we
ignore the normal ordering in the definition of g (7.325), then g will be always
greater than a unity,
gclass hN 2 i/hNi2 = 1 + hN 2 i/hNi2 > 1. (7.326)
In a chaotic (thermal) state, there is always bunching, since, according to

(7.314), for the geometric distribution,
hN 2 iT = hNi(1 + hNi), (7.327)
and gT = 2. The variance is above hNi due to the relatively slow decay of the ge-
ometric (exponential) distribution (Fig. 7.21), which makes groups of N pho-
tons with N , hNi occur more often than in the case of a Poissonian distribution.
As a result, relative fluctuations of the photon number in a thermal state tend at
hNi 1 to the unity (and not to zero, as for a coherent state),
(N/hNi)T = (1/hNi + 1)1/2 1. (7.328)
Sometimes one says that the first and the second terms in (7.327) and (7.328)
correspond, respectively, to the corpuscular and wave sides in the wave-particle
duality of a photon, and that the bunching effect (i.e., the second term) confirms
that the photons have a tendency to joining in groups. This terminology masks
the fact that the variance of some observable f characterizes the state, and not
the properties of f . This approach also ignores the presence of states with anti-

bunching, i.e., with the relative fluctuations less than 1/ hNi and with two-photon
efficiency less than the one for a coherent state with the same mean energy.
An evident example of an anti-bunched state are photon-number states. In a
K-photon state, the fluctuations of N are absent, P(N) = NK , so that hN m i = hNim
and g = 1 K 1 < 1.
Fig. 7.22 A typical distribution of 10 photons in time in the cases of (a) chaotic light (bunching);
(b) coherent light; (c) single-photon light (anti-bunching); (d) two-photon light (super-bunching).
The above-considered mixture of the vacuum and the K-photon state (7.303)
also manifests anti-bunching under the condition hNi > K 1. Indeed, it follows
from (7.305) that
gK = (K 1)/hNi. (7.329)
Hence, single-photon decay of separate atoms turns the field into an anti-bunched
state, g = 0, which was observed in the resonance fluorescence (Sec. 5.2) of
sodium atomic beam [Paul (1977)]. Certainly, in experiment the averaging runs
not over the ensemble of experimental setups but over time.
This effect can be explicitly explained by the fact that the decay of a single
atom cannot create two photons; therefore, the emitted photons are always sepa-
rated by a certain time interval required for a second excitation and de-excitation
of the same atom or the next atom in the beam (Fig. 7.22(c)). Recall that our
single-mode theory relates only to time intervals T much less than ; therefore,
P(N) = 0 for N > 1.
Radiation with anti-bunching (single-photon light) can also emerge as a re-
sult of multi-photon absorption (Sec. 6.4) of usual (chaotic or coherent) radia-
tion. The reason is that K-photon absorption, apparently, influences only the tail
of the P(N) distribution at N 1 K, which leads to the rarefication of photon groups
and the reduction of fluctuations in the initial radiation. On the contrary, the sat-
uration effect in single-photon absorption (Sec. 4.3) makes fluctuations more pro-
nounced, i.e., leads to photon bunching. This forms the base for a method of
obtaining very short (picosecond)z pulses in mode-locked lasers.
In the case of weak multi-photon light, K > 1, hNi K 1, Eq. (7.329)
yields g 1 (Fig. 7.23). This effect can be called photon super-bunching. It
z Editors note: at present, tens of femtoseconds are achieved.
Fig. 7.23 The bunching parameter g as a function of the mean photon number hNi in the cases of
single-photon light (1), two-photon light (2), three-photon light (3), coherent light (4), and chaotic
light (5). In range I, there is photon anti-bunching, in range II, bunching, in range III, super-bunching.
can be explained by the existence of regular groups of K photons separated by

large time intervals (Fig. 7.22(d)). A convenient way of obtaining directed two-
photon light is parametric down-conversion (Sec. 6.5) in piezoelectric crystals.
Below, it will be shown (see (7.356)) that super-bunching can be used for the
absolute (reference-free) measurement of the quantum efficiency of photodetec-
tors [Klyshko (1980)].
States of field with anti-bunching and the above-mentioned squeezed states
attract much attention nowadays, similarly to the way the bunching effect was very
popular soon after its discovery by Hanbury Brown and Twiss in 1955. The anti-
bunching effect, observed in 1977, is considered as disproving the semi-classical
radiation theory (see (7.326)). Indeed, according to this theory, a single-mode
laser with stabilized intensity should create photocurrent with minimal possible
fluctuations, which, similarly to the shot noise, should be Poissonian due to the
equivalence of all time moments. Thus, it seems to be impossible to create a more
uniform time distribution of photoelectrons than a Poissonian one. In contrary,
from the photon viewpoint, a flux of photons that are equidistant in time (obtained,
for instance, via a periodic excitation of a single atom) will cause an equidistant
flux of photoelectrons, provided that the quantum efficiency is high enough.
It is typical that in states with g < 1, for instance, K-photon ones, quasi-
probabilities P(z) have huge singularities or take negative values [Glauber (1965);
Perina (1972)], which allows one to consider such states as essentially non-
classical, i.e., having no classical analogues.
Consider now the quantum interpretation of interference experiments like the

Young one and the Hanbury BrownTwiss one (Sec. 7.2). While quantizing the
field, it is now convenient to expand it not over plane waves but over orthogonal
eigenfunctions un (r) of the corresponding boundary problem, which takes into
account the existence of screens with slits and semi-transparent mirrors. In the
simplest case of a monochromatic field and single-mode detectors, all field can be
represented as a single mode,
E(r, t) = u(r)a(t) + u (r)a (t) E (+) + E () , (7.330)
where a(t) = aeit is the photon annihilation operator in this mode.
The counting rate of a single-photon detector placed at point r1 scales as the
mean intensity,
G(1) () (+) 2
1 hE 1 E 1 i = |u1 | hNi, (7.331)
where the subscript 1 substitutes the argument r1 , N a a, and the averaging runs
over the initial state of the field. We stress that u(r) is the solution to the classical
wave equation, it describes the propagation of classical waves, their diffraction and
interference. Thus, the concepts of a photon and a wave by no means contradict
each other, and here we do not need to speak about the wave-particle duality.
The operators a, a , N do not depend on the coordinate; therefore, a photon is an
elementary excitation of the whole field and the question which slit did the photon
go through? is meaningless. The space structure of the field is determined by the
function u(r), its square gives the probability to discover a photon at an arbitrary
point r; therefore, it plays the role of the wavefunction of a photon.
The corpuscular properties of a photon are only revealed through detection,
when the energy of the whole field, ~ (in the case |t0 i = |1i), gets focused at
a single point. This is manifested most clearly while observing flashes on the
screen of an image intensifier. Note that the duality of wavepackets, discussed
quite often, is not typical for only quantum mechanics; the same property is pos-
sessed by classical waves.
Further, correlation of the counts of m single-photon detectors placed at points
r1 , . . . , rm scales as the normally ordered moment of order m,
G(m) () () (+) (+) 2 m
1...m hE 1 . . . E m E m . . . E 1 i = |u1 . . . um | h: N :i. (7.332)
The first factor here describes the influence of the spatial positions of the detectors.
For the case m = 2, this result was obtained in Sec. 7.2 from a simple model
assuming the binomial distribution of photons at points r1 , r2 .
Let, for instance, two-photon light be incident on an intensity interferometer
(Fig. 7.13); then, according to (7.314), the readings of the correlator will scale as
G(2) 2
12 = |u1 u2 | hNi,
while the mean counting rate will scale as G(1)

n . Hence, the relative correlation is
g(2) (2) (1) (1)

12 = G 12 /G 1 G 2 = 1/hNi. (7.333)
In the case of a pure two-photon state, hNi = 2, and the so-called negative corre-
lation, or anti-correlation, takes place, with g(2)
12 < 1 (Fig. 7.12).
Above, for simplicity we discussed the statistics of a single mode; however,
many conclusions can be generalized to the case of a multi-mode field. Indeed, in
spontaneous two-photon emission (Sec. 6.2) and in spontaneous parametric down-
conversion (Sec. 6.5), photons in pairs usually belong to different modes, k1 and
k2 , which differ both in frequency and in direction. Then, a small contribution
from the state |1i1 |1i2 is added to the vacuum, which provides the equality between
the moments,
hN1 N2 i = hN1 i = hN2 i.
As a result, at hNi i 1, the probability of discovering two photons is much

greater than the product of single-photon probabilities,
g(k1 , k2 ) hN1 N2 i/hN1 ihN2 i = 1/hNi i 1. (7.334)
This inequality, which can also be interpreted as the super-bunching effect (com-
pare with (7.329) at K = 2), was experimentally confirmed by measuring the rate
of coincidences between two PMTs.
7.6.3 Statistics of photoelectrons

Let Vdet Vcoh , then the radiation incident on a PMT photocathode can be consid-
ered as single-mode (Sec. 7.2). Following Scully [Arecchi (1974)], from explicit
combinatorics considerations we will now show (see also Refs. [Loudon (2000);
Klauder (1968); Glauber (1965); Perina (1972)]) that the Mandel formula (7.71)
maintains its form even in the framework of the quantum theory if I is replaced
by N, where is the PMT quantum efficiency and N is the operator of the photon
number in the detection volume (equal to the operator of the photon number in
one mode times the factor Vdet /Vcoh ).
Consider first a field in a pure Fock state with the photon number N incident
on a photodetector (PMT). Let the probability of registering a single photon be
(0 0 0 1), then the probability of registering any m photons out of their total
number N 1 m, apparently, is determined by the Bernoulli binomial distribution
(Fig. 7.24),
P(m|N) = C Nm m (1 )Nm . (7.335)

Fig. 7.24 Distributions of the photon number, P(N), and the photoelectron number, P(m), in the case
of the detector efficiency 0.3 and the field being in a four-photon Fock state.
Hence, in the case of an arbitrary state of the field, we find the relation between
the distribution of photons, P(N) NN , and the distribution of photoelectrons (or,
in other words, photocounts),

X
P(m) = P(m|N)P(N). (7.336)
N=m
In the case = 1, we have P(m|N) = mN , and the distributions of electrons and

photons coincide. At < 1, the binomial transformation (7.336) adds additional
stochasticity, distorts P(N) and complicates solving the inverse problem, which is
the determination of the field statistics from the measured photocount statistics.
(Using (7.346), one can show that P(N) can be formally expressed in terms of
P(m) through a relation similar to (7.336) but with replaced by 1/.)
Note that if is understood as the probability of a photon survival after light
passing a layer of matter with single-photon absorption, then m has the meaning
of the photon number at the output of the layer, while N is the input number
of photons. Then, = eI , where is the absorption coefficient and l is the
layer thickness. Thus, Eq. (7.336) and all its corollaries are also applicable to the
transformation of the field statistics caused by linear absorption of the field by
cold matter.
Let us pass from the N-representation to the z-representation using (7.301).
As a result, (7.336) leads to the quantum analogue of the Mandel formula (7.71),
which gives the photoelectron distribution in terms of the quasi-probability P(z)

of a z state,
Z
2
P(m) = d2 zP(z)(|z|2 )m e|z| /m! (7.337)
This relation, after taking into account (7.270), can be represented in an invariant
form,
P(m) = h: (N)m eN :i/m!, (7.338)
where N = a a. Thus, the distribution of photoelectrons is of a Poissonian form

but has additional quantum averaging including the operation of normal ordering.
Note that the obtained expression differs from the photon distribution (7.302) only
by the replacement of N by N.
The integral transformation (7.337)can be easily realized for the cases of co-
herent and chaotic states. Then, it turns out that the functional forms of the dis-
tributions for photons and photoelectrons in these two cases coincide. Indeed,
assuming in (7.337) P(z) = (2) (z z1 ), we find
Pz (m) = hmim ehmi /m!; (7.339)
further, it follows from (7.337) and (7.280) that
PT (m) = hmim /(1 + hmi)m+1 , (7.340)
where hmi = hNi.

Recall that P(m) can be also understood as the photon distribution at the output
of a cold layer with single-photon absorption ( = eI ). Thus, linear absorption at
T = 0K does not change the form of the photon distribution in the cases of chaotic
and coherent incident radiation (in contrast to N-photon absorption, see (??)).
Absorption or amplification at T , +0 is accompanied by chaotic spontaneous
emission, which changes the shape of the distribution.
The obtained relations (7.336), (7.337), (7.338) between the distributions of
photons and photoelectrons are rather complicated. The relations between the
generating functions and the factorial moments are much simpler,
m(k) = k N (k) , (7.341)
where
m(k) hm(m 1) . . . (m k + 1)i, (7.342)
N (k) h: N k :i = hN(N 1) . . . (N k + 1)i. (7.343)

Here, the functions of m are averaged using the discrete distribution P(m),

X
h f (m)i f (m)P(m). (7.344)
m=0
The quantum mean value, h f (N)i, has a similar form in the N representation.
Equation (7.341) can be easily obtained using the formalism of generating
functions for photons, Q phot (x) (see (7.307)), and electrons,

X
Qel (x) (1 + x)m P(m). (7.345)
m=0
By substituting (7.338), we find
Qel (x) = Q phot (x) = h: e xN :i. (7.346)
Hence, with an account for (7.308), we come to (7.341).

From (7.341), we easily obtain relations between usual moments,
hmi = hNi, (7.347)
hm2 i = hNi + 2 (hN 2 i hNi), (7.348)
and between the variances (compare with (7.78)),
m2 = hmi(1 ) + 2 hN 2 i. (7.349)
Thus, to the usual shot noise of the photocurrent, hmi, photon noise, hN 2 i,
is added with the weight 2 (which seems natural), but simultaneously, the term
2 hNi is subtracted (which is surprising for semi-classical theory). In the case of
photon anti-bunching, hN 2 i < hNi, so that the photocurrent noise is less than the
Poissonian, i.e., electrons appear with some regularity, repulse each other.
Note that, according to (7.341), the bunching parameters (as well as all nor-
malized factorial moments) of photons and electrons coincide,
g = m(2) /hmi2 = N (2) /hNi2 .
If N is understood in these equations as the total number of photons in M

independent modes, then the equations will describe the statistics of an M-mode
detector. For instance, in (7.315) and (7.346) we find that
M
Y
Qel (x) = Qk (x), (7.350)
k=1
where Qk (x) is the generating function for the kth mode.

In the case of modes with a geometric (thermal) distribution of photons,

(7.312) leads to (compare with (7.318))
Qel (x) = 1/(1 x) M , (7.351)
where = hNi/M is the mean number of photons in one mode.
At M 1 and x 1, this function tends to (compare with (7.319))
Qel (x) e xM = e xhmi . (7.352)
Hence, a multi-mode detector registering a thermal field produces photocounts
with Poissonian statistics and does not manifest bunching (similarly to a multi-
mode field, see (7.319)).
For a finite number of modes M, from (7.351) we find the distribution and
factorial moments that coincide with (7.322) after the replacement of by , or
hNi by hmi. In particular, from (7.323) we find relative fluctuations,
m/hmi = (1/hmi + 1/M)1/2 . (7.353)
Similarly, replacing x by x in (7.317), we find the generating function for a
multi-mode detector registering two-photon light,
Qel (x) = exp[hmi(x + x2 /2)]. (7.354)
It is important that at , 1, this function differs from (7.317), which opens an
interesting possibility of absolute (reference-free) measurement of the PMT quan-
tum efficiency . Indeed, from (7.354) we find that
hm2 i = hmi(hmi + 1 + ), (7.355)
or
= hm2 i/hmi 1 = hmi(g 1). (7.356)
Thus, by measuring the mean value and variance (or the bunching parameter)
of the photocounts one can find the quantum efficiency.

7.7 Interaction of an atom with quantized field
So far, we have been calculating the probabilities of quantum transitions in the

framework of the semiclassical approach, where the spontaneous transitions were
described not rigorously, with the help of additional rules or analogies introduced
without sufficient justification. In the present section, we will fill this gap. It will
be shown that the interaction between two stationary quantum systems can be con-
veniently described phenomenologically in terms of normally and anti-normally
ordered correlation functions (CFs). These functions have more direct relation
to the energy exchange than the symmetrized CFs that are commonly used. We
will also consider the symmetry of CFs, the relations between different CFs, and
the dependence of CFs on the microscopic parameters of the systems and their
Greens functions, i.e., response to coherent perturbations.
7.7.1 Absorption and emission probabilities

Consider the interaction of a single atom with quantized field. Let the atom and
the field be independent at the initial time moment, |t0 i = |mi|ii |mii, where |mi,
|ii are the initial states of the atom and the field, respectively. In the first order
of the perturbation theory, the amplitude cn f (t) of the transition into some state
|n f i is determined by the matrix element of the interaction operator, hn f |V|mii
(Sec. 2.1). In the dipole approximation,
Z t
1
cn f = dt0 hn f |d(t0 ) E(t0 )|mii, (7.357)
i~ t0
where the operators are considered in the interaction picture, i.e., without the ac-
count for the perturbation.
Let us split these operators into the positive- and negative-frequency parts. At
t t0 T 1/, fast oscillating (with approximately twice the mean frequency
) products d(+) E(+) and d() E() have no contribution into the integral (7.357);
therefore, one can write
V(t) d() (t) E(+) (t) + d(+) (t) E() (t). (7.358)
This is the rotating-wave approximation, which we already discussed in Sec. 2.2.
Further, assume that the initial and final states of the atom and/or the field
are energy ones, then the first term in (7.358) gives a nonzero contribution only
for a quantum transmitted from the field to the atom, while the second one, only
for a quantum transmitted from the atom to the field. Hence, the probability of a
transition with absorption is
Z t
P (n f |mi) = ~ 2 (+) 0 () 00
dt0 dt00 dmn (t )dnm (t )Ei() 0 (+) 00
f (t )E f i (t ), (7.359)
t0
(+) ()
where we have used the equality fmn = ( fnm ) and assumed, for simplicity, that
the vectors d and E are parallel. In the case of emission, apparently, the super-
scripts (+) and () in (7.359) should be interchanged. Note that the approximation
(7.358) does not influence the transition probability in the case where one of the
(+) (+)
interacting systems is in an energy state, since the products of the form fmn fnm
are zero then.
If we are not interested in the final state of the system, (7.359) should be
summed over all possible states |n f i. These states form a complete set; therefore,
the total probability of the transition up will be (compare with (2.82))
Z t
2
P = ~ dt0 dt00 F () (t0 , t00 )G(+) (t0 , t00 ), (7.360)
t0
where
F () (t0 , t00 ) hd(+) (t0 )d() (t00 )i, G(+) (t0 , t00 ) hE () (t0 )E (+) (t00 )i (7.361)
are, respectively, the anti-normally ordered CF for the dipole moment of the atom
and the normally ordered CF for the field at the initial state (which, apparently,
can be mixed as well).
Similarly, the probability of the emission of a quantum is determined by the
square of (7.357) with only the second term in (7.358) taken into account,
Z t
P = ~2 dt0 dt00 F (+) (t0 , t00 )G() (t0 , t00 ). (7.362)
t0
Thus, the probability of an energy quantum transfer from one quantum sys-
tem to another one is determined by the product of the unperturbed correlation
functions, the normal one for the emitting system and the anti-normal one for the
absorbing system (provided that the initial states are independent and at least one
of them is not coherent). The last condition is to exclude coherent interactions,
which depend on the phases of the states.
The mean variation of the total photon number within time T , in the second
order of the perturbation theory, is equal to the difference between (7.362) and
(7.360),
Z t
N = ~2 dt0 dt00 (F (+)G() F ()G(+) ). (7.363)
t0
In the ground states, normal CFs are equal to zero; therefore, the spontaneous
emission of an atom is determined by F (+) (this conclusion was already used in
Chapter 5), while the probability of a cold detector registering a photon is deter-
mined by G(+) (Secs. 7.2, 7.6).
Note that the obtained result (7.363) is valid for any systems with the interac-
P
tion energy of the form fi gi .
7.7.2 Spontaneous emission

According to (7.362), the probability of a spontaneous transition is
Z t
2
P sp = ~ dt0 dt00 F (+) )Gvac , (7.364)
t0
where Gvac is the anti-normally ordered CF of the field in the vacuum state.
According to (7.131),
X X 0 00
Gvac (t0 , t00 ) = ck0 ck00 h0|ak0 (t0 )ak00 (t00 )|0i = c2k eik (t t ) , (7.365)
k0 k00 k
where c2k = 2~k /L . Note that regardless of the state of the field,
3
G() (t0 , t00 ) = G(+) (t00 , t0 ) + Gvac (t0 , t00 ). (7.366)

If only two levels are taken into account (see (5.30)),
d(+) (t) = d0 12 ei0 t , d() (t) = d0 21 ei0 t , (7.367)
12 21
where d0 d12 , d11 = d22 0, 0 21 > 0, = ( ) |1ih2|. Since
kl mn kn mn
= lm , h i = nm , (7.368)
it follows that
0 00
F (+) (t0 , t00 ) = |d0 |2 ei0 (t t ) 22 ,
0 00 (7.369)
F () (t0 , t00 ) = |d0 |2 ei0 (t t ) 11 .
Thus, the correlation functions of a two-level atom scale as the populations
of the two levels: the normally ordered one, as the upper-level population, the
anti-normally ordered one, as the lower-level population.
With an account for (7.365) and (7.369), Eq. (7.364) takes the form
X Z
00 0
P sp = ~2 |d0 |2 22 c2k sin2 k dt0 dt00 ei(k 0 )(t t ) ,
k
where k is the angle between k and d0 . At t, t0 , the integral yields
[2(k 0 )]2 = 2T (k 0 ). (7.370)
Passing to integration over modes (see (7.100)), we obtain
P |d0 |2 22
Z
W sp = d3 kk sin2 k (k 0 ). (7.371)
T 2~
The integral of sin2 k over all directions equals 8/3; as a result, at 11 = 1 we
again obtain the familiar equation for the probability of a spontaneous transition
of an excited two-level atom,
W sp = (4k03 /3~)|d0|2 , (7.372)
where k0 0 /c.
Spontaneous emission from a three-level system has been already considered
in Sec. 5.2, where it was shown that P sp (T ) oscillates with the frequency 32
corresponding to the splitting between the two upper levels (the quantum beats
effect with the coherent initial state of the atom). Also, recall that the spontaneous
emission of N two-level atoms leads to the superradiance effect in which W sp is
increased by a factor of N 2 (Sec. 5.3).
7.7.3 Interaction of stationary systems

If the initial unperturbed states of both interacting systems are stationary, then all
CFs in (7.363) depend only on the difference of the integration variables, t0 t00
. (This assumption excludes from consideration the quantum beats, see Sec. 5.2,
and the coherent states, see Sec. 7.5.) If, in addition, the observation time tt0 T
is much greater than the correlation times of the atom and the field, then the double
integral in (7.363) scales as T , and one can introduce time-independent rates of
up and down transitions, W P/T .
Let us define the correlation functions of a single variable,
()
F() F () (t, t + ) = hd() (0)d() ()i = F ,
(7.373)
G() () () () ()
G (t, t + ) = hE (0)E ()i = G ,
then (7.363) becomes

Z T
N = ~2 d(T )[F(+)G() () (+)
F G + ( )]. (7.374)
0
If T is much greater than the correlation time of the atom and the field, then it
follows from (7.374) that
Z
W = ~2 d(F(+)G() () (+)
F G ), (7.375)

where W (P P )/T and the sign of W determines the direction of the quantum
transfer (from the atom to the field at W > 0).
Our initial model describes the time evolution of the states of the atom and
the field. Angle brackets in (7.373) denote averaging over the ensemble of experi-
ments with the same initial conditions and different . If, however, we assume that
both systems are in contact with their thermostats (Fig. 7.25), which continuously
restore and maintain the initial stationary states, we will obtain an ergodic model.
Then, (7.375) describes a continuous flux of energy quanta transmitted from one
thermostat to the other one through the atom-field system.
Fig. 7.25 Stationary interaction between an atom and a field. The atom-field coupling is often the
bottleneck for the energy exchange between the thermostats T 1 , T 2 .
Let us now write the CF (7.373) in terms of the microscopic parameters of the
systems. Averaging with the help of stationary density matrices (not necessarily
equilibrium ones) yields
X
G(+)
= c2k hNk ieik ,
k
X
G()
= c2k (hNk i + 1)eik ,
k
X (7.376)
F(+) = |dmn |2 mm eimn ,
m>n
X
F() = |dmn |2 nn eimn ,
m>n
where hNik hak ak i are mean photon numbers in the modes and nn are relative
populations of the atomic levels. After substituting (7.376) into (7.375), the rate
of quantum exchange takes the form
X
W = 2~2 c2k |dmn |2 (k mn )[(hNk i + 1)mm hNk inn ]. (7.377)
k,m>n
The three terms of the last factor correspond to the three types of transitions be-
tween each pair of levels according to Einstein (Chapter 2).
By inserting into (7.377) the energy of a quantum ~k , apparently, we find the
power of emission or absorption,
X
P = (2)2 L3 2k |dmn |2 (k mn )(mm nm hNk i), (7.378)
k,m>n
where nm nn mm is the population difference. This equation at hNk i = 0

describes the spontaneous emission of an atom into the vacuum and at mm = 0,
the radiation heating of a cold atom.
Note that the normally and anti-normally ordered CFs introduced by (7.373)
are not real and do not possess a certain parity (in contrast to classical CFs). How-
ever, one can construct even (symmetrized) and odd (anti-symmetrized) combina-
tions,
1
F(s) ReF = hd0 d + d d0 i = F(s)
,
2
(7.379)
(a) 1 (a)
F ImF = hd0 d d d0 i = F ,
2i
and similar functions G(s) (a)
, G for the field. Here, we have also introduced the
total CF F , equal to the sum of normally and anti-normally ordered CFs,
F hd0 d i = hd0() d(+) + d0(+) d() i = F(+) + F() = F

. (7.380)
In order to find the inverse transformation (express F() in terms of F ), it is nec-

essary to use the spectral expansion.
Often, it is only the symmetric combination F(s) that is used, and it is called
the CF. Below, it will be shown that the antisymmetric combination F(a) is closely
connected with the response of the system to a coherent perturbation, i.e., with
its susceptibility or the Greens function. Note that in experimental optics, one
usually deals with non-coherent systems, so that, according to (7.375), CF F()
provide a more direct description of the observed effects.
From the definitions (7.379) and (7.376), we find the microscopic formulas,
X
G(s)
= c2k (2hNk i + 1) cos(k ), (7.381)
k
X
G(a)
= c2k sin(k ) = ImGvac
, (7.382)
k
X
F(s) = |dmn |2 (mm + nn ) cos(mn ), (7.383)
m>n
X
F(a) = |dmn |2 nm sin(mn ). (7.384)
m>n
Below, it will be shown (see (7.390), (7.394)) that in the case of a linear or equi-
librium system, the only difference between various types of CFs is the different
contributions of zero-point fluctuations into these functions.
7.7.4 Spectral representation

The Fourier transforms of stationary CFs are called the spectral densities (of the
fluctuations of the corresponding observables, such as, for instance, the electric
field or the dipole moment of an atom). It is clear from definitions (7.373) that
the Fourier transform F(+) of the normally ordered CF F(+) is nonzero only at
> 0, while the Fourier transform of the anti-normally ordered CF, only at < 0.
Therefore, they can be combined into a single function F , the Fourier transform
of F ,
Z
F dei F /2, F = F = F(s) + iF(a) = F(+) + F() 1 0,
(7.385)
F() = F (), F(s) = (F + F )/2, F(a) = (F F )/2i,
and similarly for the field CFs G()

. Here, we have also introduced the Fourier
transforms of symmetric and anti-symmetric real CFs, which have the following
properties:
(s) (a)
F(s) = F(s) = F , F(a) = F(a) = F . (7.386)
Substituting into (7.375) the inverse transformation,

Z
F() = dei F , (7.387)
0
with the factor ~, we obtain the spectral expansion of the power,

2 2
Z Z
P= d(FG FG ) = dF G , (7.388)
~ 0 ~
where all spectral CFs are positive. Thus, we have separated positive and nega-
tive terms, corresponding to emission and absorption. If we pass to the Fourier
transforms of symmetric and anti-symmetric CFs, (7.388) takes the form
4
Z
P= d(F(s)G(a) (a) (s)
F G ), (7.389)
i~ 0
where such separation is absent.
The explicit expressions for the spectral CFs in terms of the microscopic pa-
rameters can be easily found through the Fourier transformation of (7.376) and
(7.381)(7.384),
X
G(+)
= c2k hNk i( k ),
k
X
G()
= c2k (hNk i + 1)( + k ),
k
!
X 1
G(s)
= c2k hNk i + [( k ) + ( + k )],
k
2
iX 2
G(a)
= c [( k ) ( + k )],
2 k k
X (7.390)
F(+) = |dmn |2 mm ( mn ),
m>n
X
F() = |dmn |2 nn ( + mn ),
m>n
1X
F(s) = |dmn |2 (mm + nn )( + mn ),
2 mn
1 X
F(a) = |dmn |2 nm ( + mn ).
2i mn
Comparing the last expression with Eq. (4.57) for the linear susceptibility (polaris-
ability) of a single atom at = 0, one can see that the anti-symmetric spectral
CF of a stationary system coincides (up to the factor 2/i~) with the imaginary
part of the susceptibility of the system with respect to a coherent excitation,
00 = (2/i~)F(). (7.391)
The response function and its Fourier transform are called the retarded part
of the Greens function of the system. Note that in order to determine the suscep-
tibility of matter, experimentally or theoretically, one requires the field to be in a
coherent quantum state (Sec. 7.5). And vice versa, the definition of the suscepti-
bility of the vacuum (Sec. 4.1) is based on the coherent state of the matter. From
the comparison between (7.390) and (7.365), it follows that G(a) vac
= (ImG ) , i.e.,
the role of in the case of the field is played by the anti-normally ordered CF of
the vacuum.
Recall that 00 , in its turn, unambiguously determines 0 (Sec. 4.1). Thus,
the kinetic parameters of a stationary system are in one-to-one correspondence
with its unperturbed fluctuation characteristics.aa In the case of equilibrium sys-
tems, this relation can be inverted, so that each one of the four CFs of the system
determines the other ones.
7.7.5 Equilibrium systems. FDT

At thermodynamic equilibrium, we have Plancks distribution for the photons and
Boltzmanns distribution for the populations,
hNk i = N(k ), mm = nm N(mn ), (7.392)
where
" #1
~
N() = exp 1 N . (7.393)
T
Due to the functions in (7.390), the temperature factors of N can be removed
from the sum; as a result, all CFs can be expressed in terms of each other or in
terms of the susceptibility imaginary part. For instance, in the case of an atom,
~
F(+) = N 00 ,

() ~ ~
F = (N + 1)00 = F(+) exp , (7.394)
T
!
~ 1 00
F(s) = N + ,
2
aa Such relations are called the Kubo formulas, see, for instance, Ref. [Zubarev (1971)].
where equals unity at > 0 and zero at < 0. The last equation in (7.394)
is called the fluctuation-dissipation theorem (FDT) or the Nyquist-Callen-Welton
theorem.
Let the atom and the field be in equilibrium states with the temperatures T 1
and T 2 , respectively. Then, according to (7.394) and similar equations for field
CFs, (7.388), (7.389) take the form
Z
P = 4i d00 G(a)
(N1 N2 ), (7.395)
0
1
where Nn [exp(~/T n ) 1] . If the temperatures are close, the last factor
in (7.395) scales as T 1 T 2 , then the ratio P/|T 1 T 2 | determines, through 00 ,
the heat conductivity of the atom-field link, which provides the heat exchange
between the two thermostats.
Here, we only considered single-quantum transitions and linear susceptibility,
which are described in the first orders of the perturbation theory. A similar consid-
eration can be performed for multi-quantum transitions, nonlinear susceptibilities,
and higher-order CFs. The corresponding generalizations of the FDT, obtained by
Efremov and Stratonovich, are described in short in Ref. [Klyshko (1980)].
In conclusion, let us mention that the interaction between an atom and a field
leads not only to the energy exchange but also to a certain shift of the energy
levels. In the case of the vacuum state of the field, this shift leads to a variation of
the eigenfrequencies by a value about 109 Hz (for the hydrogen atom). This effect
is called the Lamb shift (see, for instance, Ref. [Allen (1975)]). If the field is in
an excited state, the shift is called the Stark effect in alternating field (5.48). This
phenomenon has to be taken into account in many problems of laser spectroscopy.
Bibliography
Akhmanov, S. A. and Khokhlov, R. V. (1964). Problems of nonlinear optics (VINITI,

Moscow) [in Russian].
Akhmanov, S. A. and Chirkin, A. S. (1971). Statistical Phenomena in Nonlinear Optics
(MGU, Moscow) [in Russian].
Akhmanov, S. A., Dyakov, Yu. E., and Chirkin, A. S. (1981). Introduction to statistical
radiophysics and optics (Nauka, Moscow) [in Russian].
Akhmanov, S. A. and Koroteev, N. I. (1981). Methods of nonlinear optics in the spec-
troscopy of scattered light (Nauka, Moscow) [in Russian].
Aleksandrov, E. B. (1972). Optical manifestation of interference of non-degenerate atomic
states, Sov. Phys. Usp. 15, pp. 436451.
Allen, L., and Eberly, J. H. (1975). Optical Resonance and Two-level Atoms (John Wiley,
New York).
Andreev, A. V., Emelyanov V. I., Ilinskii, Yu. A. (1980). Collective spontaneous emission
(Dicke superradiance), Sov. Phys. Usp. 23, pp. 493-514.
Apanasevich, P. A. (1977). Foundations of the theory of interaction of light with matter
(Nauka i Tekhnika, Minsk) [in Russian].
Arecchi, F. T., Scully, M. O., Haken, H., and Weidlich, W. (1974). Quantum Fluctuations
of Laser Radiation (Mir, Moscow) [Russian translation].
Barachevskii, V. A., Lashkov, G. I., and Tsekhomskii, V. A. (1977). Photochromism and
its applications (Khimiya, Moscow) [In Russian].
Bertin, F. (1971). Principles of Quantum Electronics (Russ. transl.) (Mir, Moscow) [in
Russian].
Bespalov, V. I. (ed.) (1979). Optical Phase Conjugation in Nonlinear Media (Institute of
Applied Physics, Gorky) [In Russian].
Bloembergen, N. (1965). Nonlinear Optics (W. A. Benjamin, Inc., New York).
Bloembergen, N. (ed.) (1977). Nonlinear Spectroscopy (North-Holland, Amsterdam).
Braginsky, V. B. and Vorontsov, Yu. I. (1975). Quantum-mechanical limitations in
macroscopic experiments and modern experimental technique, Sov. Phys. Usp. 18,
pp. 644650.
Braginsky, V. B. and Vorontsov, Yu. I., Thorne, K. S. (1980). Quantum Nondemolition
Measurements, Science 209, pp. 547557.
Butylkin, V. S., Kaplan, A. E., Yakubovich, E. I., Khronopulo, Yu. G. (1977). Resonant
337
Interactions of Light with Matter (Nauka, Moscow) [In Russian].

Chang, R. K. and Furtak, T. (eds.) (1981). Surface Enhanced Raman Scattering (Plenum,
New York).
Cummins, H. Z. and Pike, E. R. (eds.) (1974). Photon-correlation and light-beating
spectroscopy (Proc. NATO Advanced Study Group) (Plenum Press, New York).
Delone, N. B., Krainov, V. P., Khodovoi, V. A. (1975). The two-level system in a strong
light field, Sov. Phys. Usp. 18, pp. 750-755.
Delone, N. B. and Krainov, V. P. (1978). Atom in a strong light field (Atomizdat, Moscow)
[in Russian]; (1985). Atoms in Strong Light Fields (Springer-Verlag, Heidelberg).
Dmitriev, V. G. and Tarasov, L. V. (1982). Applied nonlinear optics (Radio i Svyaz,
Dunskaya, I. M. (1981). Rise of quantum electronics (Nauka, Moscow) [in Russian].
Elyutin, P. V. (1982). Theoretical foundations of quantum radiophysics (MGU, Moscow)
[in Russian].
Fabelinsky, I. L. (1965). Molecular Scattering of Light (Nauka, Moscow) [In Russian];
(1968). (Plenum Press, New York).
Fain, V.M. (1972). Photons and non-linear media (Sov. Radio, Moscow) [in Russian].
Ginzburg, V. L. (1983). The nature of spontaneous radiation, Sov. Phys. Usp. 26, pp. 713-
719.
Glauber, R. J. (1965). in Quantum Optics and Electronics, edited by C. D. A. Blandin and
C. Cohen-Tannoudji (Gordon & Breach, New York).
Gorelik, G. S. (1948). On the demodulation analysis of light, Usp. Phys. Nauk 34, p. 321
[In Russian].
Grishanin, B. A. (1981). Quantum electronics for radio physicists (MGU, Moscow) [in
Russian].
Haken, H. (1977). Synergetics (Springer-Verlag, Berlin).
Hanbury Brown, R. (1971). Measuring the angular diameters of stars, Contemp. Phys. 12,
pp. 357377.
Kaczmarek, F. (1981). Introduction to laser physics (Russ. transl.) (Mir, Moscow) [in
Russian]; (1978). (PWN, Warsaw) [in Polish].
Karlov, N. V. (1992). Lectures on quantum electronics (Translated from Russian by
E. Yankovsky) (Mir, Moscow).
Klimontovich, Yu. L. (1966). Quantum light generators (Prosveshchenie, Moscow) [in
Russian].
Khanin, Y. (1975). Dynamics of quantum oscillators (Sov. Radio, Moscow) [in Russian];
Khanin, Y. (1995) Principles of Laser Dynamics (Elsevier, Amsterdam).
Kielich, S. (1980). Nonlinear molecular optics (Russ. transl.) (Nauka, Moscow) [in Rus-
sian]; (1977). (PWN, Warsaw) [in Polish].
Klauder, J. R. and Sudarshan, E. C. G. (1968). Fundamentals of Quantum Optics (W. A.
Benjamin, Inc., New York).
Klimontovich, Yu. L. (1980). Kinetic Theory of Electromagnetic Processes (Nauka,
Moscow) [In Russian]; (1982). (Springer, Heidelberg).
Klimontovich, Yu. L. (1982). Statistical Physics (Nauka, Moscow) [In Russian]; (1986).
(Harwood, New York).
Klyshko, D. N. (1980). Photons and Nonlinear Optics (Nauka, Moscow) [In Russian];
(1988). (Gordon & Breach, New York).
Bibliography 339
Landau, L. D. and Lifshitz, E. M. (1964). Statistical Physics (Nauka, Moscow) [in

Russian]; (1980). 3rd ed. (Pergamon Press, Oxford).
Landau, L. D. and Lifshitz, E. M. (1973). Field Theory (Nauka, Moscow) [in Russian];
(1975). (Pergamon Press, Oxford).
Landau, L. D. and Lifshitz, E. M. (1982). Electrodynamics of Continuous Media (Nauka,
Moscow) [in Russian]; (1984). 2nd ed. (Pergamon Press, Oxford).
Lax, M. (1968). Fluctuation and Coherence Phenomena in Classical and Quantum Physics
(Gordon & Breach, New York).
Letokhov, V. S. and Chebotaev, V. P. (1975). Principles of nonlinear laser spectroscopy
(Nauka, Moscow) [In Russian].
Letokhov, V. S. (1983). Nonlinear Selective Photoprocesses in Atoms and Molecules
(Nauka, Moscow) [in Russian].
Loudon, R. (2000). The Quantum Theory of Light, 3rd edn. (Oxford University Press,
Oxford).
Louisell, W. H. (1964). Radiation and Noise in Quantum Electronics (McGraw-Hill, New
York).
Macomber, J. D. (1976). The Dynamics of Spectroscopic Transitions (John Wiley, New
York).
Maitland, B. and Dunn, M. (1969). Laser Physics (North Holland, Amsterdam).
Manykin, E. A., and Samartsev, V. V. (1984). Optical Echo Spectroscopy (Nauka, Moscow)
[In Russian].
Migulin, V. V., Medvedev, V. I., Mustel, E. R., and Parygin, V. N. (1978). Fundamentals
of the oscillation theory (Nauka, Moscow) [in Russian].
Minogin, V. G. and Letokhov, V. S. (1986). Laser radiation pressure on atoms (Nauka,
Moscow) [In Russian].
Nye, J. F. (1957). Physical Properties of Crystals (Clarendon Press; Oxford University
Press, Oxford).
Pantell, R. and Puthoff, H. (1969). Fundamentals of Quantum Electronics (John Wiley,
New York).
Paul, H. (1980). Photon counting as a tool for basic research, Soviet Journal of Quantum
Electronics 7, pp. 1437.
Perina, J. (1972). Coherence of Light (Van Nostrand Reinhold, London).
Piekara, A. (1973). A new face of optics: Introduction to quantum electronics and nonlinear
optics (Russ. transl.) (Sov. Radio, Moscow) [in Russian]; (1968). (PWN, Warsaw)
[in Polish].
Rabinovich, M. I., Trubetskov, D. I. (1989). Oscillations and waves in linear and nonlinear
systems (Kluwer-Academic Publ., Amsterdam).
Rytov, S. M. (1976). Introduction to statistical radiophysics. Part 1. (Nauka, Moscow) [in
Russian].
Schubert, M. and Wilhelmi, B. (1973). Introduction to Nonlinear Optics (Mir, Moscow)
[Russian translation]; Einfuhrung in die nichtlineare Optik (Teubner, Leipzig) [in
German]; (1986). Nonlinear Optics and Quantum Electronics (Wiley, New York).
Silin, V. P., Rukhadze, A. A. (1961). Electromagnetic properties of plasma and plasma-like
media (Gosatomizdat, Moscow) [in Russian].
Steinfeld, J. I. (ed.) (1978). Laser and Coherence Spectroscopy (Plenum Press, New York).
Strakhovskii, G. M., Uspenskii, A. V. (1979). Foundations of quantum electronics

(Vysshaya Shkola, Moscow) [in Russian].
Svelto, O. (2010). Principles of Lasers (Springer, New York).
Tarasov, L. V. (1976). Physical foundations of quantum electronics (Sov. Radio, Moscow)
[in Russian].
Tarasov, L. V. (1981). Physics of processes in coherent optical radiation sources (Radio i
Svyaz, Moscow) [in Russian].
Vinogradova, M. B., Rudenko, O. V., and Sukhorukov, A. P. (1979). Wave theory (Nauka,
Walls, D. F. (1983). Squeezed states of light, Nature 306, pp. 141146.
Walther, H. (ed.) (1976). Topics in Applied Physics, Volume 2: Laser Spectroscopy of Atoms
and Molecules (Springer-Verlag, Berlin-Heidelberg-New York).
Yariv, A. (1976). Introduction to Optical Electronics, 2nd ed. (Holt, Rinehart & Winston,
New York).
Yariv, A. (1989). Quantum Electronics, 3rd ed. (John Wiley, New York).
Zeldovich, Ya. B., Pilipetsky, N. F., and Shkunov, V. V. (1985). Principles of Phase
Conjugation (Nauka, Moscow) [In Russian]; (1985). (Springer Verlag, Berlin).
Zernike, F., Midwinter, J. E. (1973). Applied Nonlinear Optics (Wiley, New York).
Zhabotinskii, M. E. (ed.) (1969). Kvantovaya elektronika. Malenkaya entsiklopediya
(Small Encyclopedia of Quantum Electronics) (Sov. Entsiklopediya, Moscow) [in
Russian].
Zubarev, D. N. (1971). Nonequilibrium Statistical Thermodynamics (Nauka, Moscow) [In
Russian]; (1974). (Consultant Bureau, New York).
Bibliography expanded by the Editors:
Agrawal, G. (2007) Nonlinear Fiber Optics, 4th ed. (Academic Press, San Diego).
Bachor, H.-A. and Ralf, T. (2004). A Guide to Experiments in Quantum Optics (Wiley-VCH
Verlag GmbH & Co. KGaA, Weinheim).
Bouwmeester, D., Ekert, A., Zeilinger, A. (Eds.) (2000). The Physics of Quantum
Information (Springer, Berlin).
Boyd, R. (2008) Nonlinear optics, 3rd ed. (Academic Press, San Diego).
Bruss, D. (2002) Characterizing entanglement, Journ. of Math. Phys., 43, p. 4237.
Goodman, J. (1985). Statistical Optics (McGraw-Hill, New York).
Grynberg, G., Aspect, A., Fabre, C. (2010). Introduction to quantum optics: From the
semiclassical approach to quantized fields (Cambridge University Press, New York).
Klyshko, D. N. (1994). Quantum optics: quantum, classical, and metaphysical aspects,
Phys. Usp. 37, pp. 10971122.
Klyshko, D. N. (1998). Basic quantum mechanical concepts from the operational view-
point, Phys. Usp. 41, pp. 885922.
Leonhardt, U. (1997). Measuring the quantum state of light (Cambridge University Press,
Cambridge).
Mandel, L. and Wolf, E. (1995). Optical Coherence and Quantum Optics (Cambridge
University Press, Cambridge).
Bibliography 341
Milloni, P, Eberly, J. (2009) Laser Physics (Wiley, New Jersey).

Nielsen, M. A. and Chuang I. L. (2000). Quantum Computation and Quantum Information
(Cambridge University Press, Cambridge).
Peres, A. (1993). Quantum Theory: Concepts and Methods (Kluwer, Amsterdam).
Schleich, W. P. (2001). Quantum Optics in Phase Space (Wiley, New York).
Scully, M. O. and Zubairy, M. (1997). Quantum Optics (Cambridge University Press,
Cambridge).
Schrodinger, E. (1935). Die gegenwartige Situation in der Quantenmechanik, Naturwis-
senschaften 23, p. 807812.
Walls, D.F. and Milburn, G. J. (1994). Quantum Optics (Springer)
Yamamoto, Y. and Imamoglu A. (1999). Mesoscopic quantum optics. (Wiley, New York).

Index
absorption, 70, 76 equations, 96, 101, 102, 105, 108, 127

coefficient, 4, 15, 26, 29, 30, 115, 162, sphere, 99, 115, 129, 132
166, 175, 194, 240242, 244 vector, 99, 102, 108, 112, 117, 128,
resonance, 29, 228 130, 132, 133
amplification brightness
bandwidth, 30 temperature, 39
coefficient, 4, 31, 194, 221, 226, 232,
239, 240, 243, 247 canonical
analytic field variables, 278
signal, 253 causality, 69
analytical classical
solution, 110 theory of dispersion, 77
anharmonic coherence
oscillator, 154 volume, 260
anharmonicity coherent
electrocaloric, 164, 166, 177 states, 127, 129, 294, 297, 302, 308,
orientation, 166, 168 310, 331
Raman, 157, 161, 199 commutation
striction, 152 relations, 285, 287
temperature, 162, 164166, 201 correlation
anti-normally ordered functions, 33, 44, 56, 95, 256, 298, 307,
correlation functions, 330, 332, 333, 308, 328331
335 cross-section, 27
anti-Stokes
Raman degeneracy, 30, 31, 52, 56
scattering, 181 factor, 40, 237, 256, 260, 261, 308, 318
atom of the levels, 31, 32
optics, 11 density matrix
more general definition, 47
backward dipole
interaction, 216, 234 approximation, 21, 33, 79, 116, 146,
Bloch 148, 283285, 328
343
Diracs hyper-parametric
notation, 289 scattering, 12, 221, 228
dispersion, 26, 39, 140, 144
dependence, 12, 77, 81, 83, 8589, 159, intensity
161, 169, 180182, 216, 227 correlation, 266, 267, 269
frequency-angular, 179 interference, 271, 272
induced, 180, 191, 207 interaction
law, 7577, 86, 87, 184, 189 Hamiltonian, 283, 284
linear and nonlinear, 181 of stationary systems, 331
negative, 85 operator, 57, 328
spatial, 72, 76, 139 picture, 6264, 170, 328
theory, 5, 77, 154, 207 inversion
Doppler-free saturation, 196 in semiconductors, 55
Doppler-free spectroscopy, 195, 197 isolated resonance, 82, 83
Einsteins kinetic equations, 28, 61, 90, 95, 96, 102,

coefficient, 48, 62, 240, 332 108, 162, 177, 237
Einsteins B coefficient, 35 for the density matrix, 59, 79, 91
emission for the mean values, 95
of a dipole, 116 Kirchhoff law, 241
spontaneous, 59, 62, 116, 118123,
128132, 221, 238, 240, 243, laser
247, 259, 325, 329, 330, 332 physics, 10
stimulated, 2, 3, 6, 7, 136, 161, 181 spectroscopy, 229, 336
energy states, 291, 292 lasers, 8
entangled states, 12, 13, 310, 312314 lasing condition, 3
equilibrium, 112 light pressure, 75, 148152
distribution, 4 force, 152
matter, 167 linear approximation, 20, 79, 158, 185,
medium, 3 192, 208, 287
populations, 30, 51, 61, 90, 244
radiation, 35, 36, 40, 48, 62, 70, 86, 87, magnetic resonance, 7, 9, 104106, 127
121, 241, 246, 261 electron, 251
state, 308 electronic, 89
systems, 56, 112, 115, 131, 335, 336 nuclear, 61, 89, 128
Mandel formula, 262, 263, 265, 266, 323,
far field, 210, 288 325
FDT, 70, 86, 95, 160, 178, 239, 335, 336 Manley-Rowe relation, 187
feedback, 3, 7, 199, 215, 216, 247249 masers, 13, 6, 7
finite level widths, 24 mixed state, 4547, 49, 105, 121, 291,
four-wave interactions, 226, 232 305, 307, 311, 312
frequency multi-photon
up-conversion, 213, 221 transition
probability, 173
Heisenberg
picture, 44, 63, 64, 95, 118, 287 natural
Index 345
bandwidth, 38, 94, 121 amplification, 214, 215, 224, 225, 230,
broadening, 29, 38, 39, 60 303
non-equilibrium coefficient, 215
matter, 242 amplifier, 220222, 244
radiation, 41, 261, 309 approximation, 213
state, 52, 56 down-conversion, 12, 220, 227, 312,
system, 53 315, 321, 323
systems, 52, 56, 131 spontaneous, 135
thermodynamics, 239 fluorescence, 220
non-parametric oscillation, 214, 224, 225
processes, 135, 136, 179, 180, 182, oscillator, 135, 215
187, 191, 197 processes, 135, 136, 158, 179182,
nonclassical 187189, 199, 207209, 213,
light, 11 219, 226228, 230, 235
nonlinear scattering, 220
absorption, 136, 180, 191 super-fluorescence, 224
dispersion, 180 permutation symmetry, 187
optics, 10 permutative symmetry, 141, 142
spectroscopy, 136, 197, 228, 229 perturbation theory, 15, 19, 20, 23, 25, 32,
susceptibilities, 10, 138, 142, 146, 159, 33, 51, 58, 62, 64, 79, 89, 108, 111,
336 120, 172, 173, 263, 284, 293, 328, 329,
normally ordered 336
characteristic function, 307 photon
correlation functions, 328330, 332, anti-bunching, 262, 270, 315, 318320,
333 326
emission, 118 bunching, 262, 265, 266, 269, 309,
fluctuations, 40 318320
moments, 118, 119, 122, 286, 297, 306, echo, 131
315, 322 kinetics, 28
operators, 295, 306, 307 statistics, 314
number of photons, 39 physics of lasers, 9
nutation, 100, 109, 110, 112114, 126, polaritons, 85, 86, 197, 223, 226
127 population
inversion, 2, 3, 68, 37, 46, 52, 56, 80,
observables, 43, 46, 64, 95, 96, 102, 286, 84, 94, 112, 114, 131, 152,
288, 293, 333 194, 199
one-dimensional kinetics, 28, 36
approximation, 29, 182, 210, 217 populations
equations, 189, 191 in semiconductors, 53
model, 155, 185, 207, 212, 234, 235 of the levels, 3, 4, 8, 17, 18, 20, 2830,
oscillator 32, 36, 37, 43, 5052, 54, 55,
strength, 81, 150 6062, 67, 79, 9092, 96, 98,
oscillator strength, 79 104, 105, 107, 119, 120, 123,
152, 165, 172, 175, 197
paramagnetic resonance, 94, 105, 251
parametric quantum
beats, 121124, 197, 330, 331 fluorescence, 124126, 152, 320

theory of dispersion, 5, 79 interaction, 27
parametric
radio processes, 188
spectroscopy, 6 parametric processes, 179
Raman Q-factor, 82
absorption, 159 Raman
amplification, 158, 159, 191, 197199 scattering, 180
coefficient, 226, 232 scattering, 125
amplification spectroscopy, 229 susceptibilities, 142, 226
amplifier, 221 susceptibility, 232
anharmonicity, 157, 161, 199 transitions, 27
coherent anti-Stokes scattering, 226
induced dispersion, 159 saturation, 4, 6, 29, 53, 54, 61, 89, 9194,
inverse effect, 180, 229 107, 111, 112, 114, 127, 152, 180, 191,
inverse scattering, 201 192, 194, 195, 238, 247, 252, 320
laser, 199 factor, 104
nonlinearity, 177 lineshape, 92
resonance, 201 spectroscopy, 94
resonances, 198 scattering
scattering, 125127, 137, 157, 159,
by polaritons, 223, 225, 226
173, 174, 176, 180, 197, 229
matrix, 219
by polaritons, 223
Schrodinger
spectroscopy, 175, 229
picture, 44, 63
susceptibility, 142
second harmonic
two-photon transition, 136
generation, 181, 217
random intensity, 254
second-order
relaxation, 46, 8, 15, 28, 29, 32, 34, 37,
43, 48, 5860, 62, 79, 103, 105107, coherence, 267, 270
113, 114, 127, 131, 152, 167, 168, 193, parameters, 267, 271, 272
244 correlation function, 271
longitudinal, 59, 61, 152 correlation functions, 267, 270
model, 41, 102 interference, 271, 313
rate, 90 moments, 33
time, 41, 56, 58, 59, 61, 90, 95, 96, 102, self-focusing, 154, 166, 181, 183, 185,
107, 112, 113, 116, 163166, 191, 201, 203, 206, 207, 227, 235
168, 194, 233 length, 203, 206
transverse, 59, 60, 120 self-induced transparency, 107, 114, 137
resonance, 23, 24, 27, 30, 72, 79, 8183, spatial
85, 87, 92, 93, 110, 112, 152, 156, 168, coherence, 259, 267, 273
173175, 180182, 192195, 197, 228, dispersion, 72, 76, 85, 89
229, 244, 249 spectral
absorption, 29 brightness, 39
condition, 21 field density, 35
conditions, 89 representation, 139, 333
dispersion, 159 spontaneous
Index 347
emission, 5, 15, 59, 62, 116, 118123, symmetry breaking, 166

128, 130132, 238, 240, 243,
247, 259, 329, 330, 332 temporal coherence, 257, 272
four-wave mixing, 12 thermodynamic approach, 72, 143
parametric down-conversion, 12, 135, three-wave interaction, 212, 234
221, 224, 312, 315, 323 transition
radiation, 246 cross section, 92
Raman scattering, 157 cross-section, 27, 29, 247
scattering, 162, 166, 181, 199 probability, 24, 15, 18, 2124, 27, 32,
superradiance, 130 33, 38, 43, 48, 58, 108, 111,
transitions, 27, 29, 35, 3741, 60, 62, 117, 152, 174, 175, 239,
116118 327330
squeezed states, 12, 125, 302304, 321 rate, 23, 25, 26, 28, 34, 35, 37, 9294,
statistics 111, 116, 175, 331
of photocounts, 262 transparent matter, 74, 141, 142, 172, 177,
of photoelectrons, 239, 323 220
of photons, 239
stimulated unperturbed atom, 16, 176
absorption, 22
Compton effect, 151 von Neumann
effects, 108, 158, 227 entropy, 311
emission, 2, 3, 6, 7, 22, 136, 161, 181 equation, 57, 58, 6264, 108
Mandelshtam-Brillouin scattering, 154, principle, 145
161, 198, 232
oscillations, 18
parametric
down-conversion, 224
parametric down-conversion, 224, 225
precession, 109, 128
Raman
scattering, 158, 180, 199, 232
Rayleigh-wing scattering, 166, 168
scattering, 162, 164, 191, 199, 206, 235
temperature scattering, 163
transition, 286, 319
transitions, 35, 15, 18, 22, 24, 3235,
3741, 55, 61, 62, 93, 120,
152, 176, 240
superfluorescence, 246, 247, 309
superradiance, 18, 107, 120, 127133, 330
state, 114, 130
symmetry, 21, 34, 54, 68, 74, 80, 90, 137,
141, 142, 144, 145, 157, 172, 187, 188,
219, 311, 328
of susceptibilities, 73, 188
of the medium, 68, 144, 218, 226

Optics and Quantum Electronics - Research Laboratory of Electronics PDF

Uploaded by

Copyright:

Available Formats

Optics and Quantum Electronics - Research Laboratory of Electronics PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Optics and Quantum Electronics - Research Laboratory of Electronics PDF

Uploaded by

Copyright:

Available Formats

Physical Foundations of by David Klyshko

7930.9789814324502-tp.indd 2 3/23/11 3:11 PM

7930.9789814324502-tp.indd 1 3/23/11 3:11 PM

British Library Cataloguing-in-Publication Data

PHYSICAL FOUNDATIONS OF QUANTUM ELECTRONICS BY DAVID KLYSHKO

Alvin - Physical Foundations of Quantum.pmd 1 3/1/2011, 9:45 AM

vi Physical Foundations of Quantum Electronics

A large number of general guides in quantum electronics have already

Below, we present the translation of a book by David Klyshko (19292000), which

This page intentionally left blank

List of Notation and Acronyms

a, transverse size, cm; photon annihilation operator

x Physical Foundations of Quantum Electronics

P, polarization, (erg/cm3)1/2 ; probability

List of Notation and Acronyms xi

, circular frequency, rad/s

This page intentionally left blank

List of Notation and Acronyms ix

2. Stimulated Quantum Transitions 15

xiv Physical Foundations of Quantum Electronics

2.1.2 Atom in an alternating field . . . . . . . . . . . . . . . 18

3. Density Matrix, Populations, and Relaxation 43

3.2.2 Two-level system and the negative temperature . . . . . 52

4. The Susceptibility of Matter 67

xvi Physical Foundations of Quantum Electronics

4.4.7 Magnetic resonance . . . . . . . . . . . . . . . . . . . 104

5. Non-Stationary Optics 107

6. Nonlinear Optics 135

6.2.10 Probability of multi-photon transitions . . . . . . . . . 173

7. Statistical Optics 237

xviii Physical Foundations of Quantum Electronics

7.1.5 Gain and bandwidth of a cavity amplifier . . . . . . . . 248

7.7.4 Spectral representation . . . . . . . . . . . . . . . . . 333

Quantum electronics studies the interaction of electromagnetic field with matter

2 Physical Foundations of Quantum Electronics

1.1 Basic notions of quantum electronics

1.1.1 Stimulated emission

1.1.2 Population inversion

1.1.3 Feedback and the lasing condition

4 Physical Foundations of Quantum Electronics

unit time, W, and the number of active particles, N:

1.1.4 Saturation and relaxation

Even in the absence of incident radiation or other external influence, an excited

1.2 History of quantum electronics

Quantum electronics can be considered as a new chapter in the theory of light

6 Physical Foundations of Quantum Electronics

1.2.1 First steps

1.2.2 Radio spectroscopy

8 Physical Foundations of Quantum Electronics

(2010)], making a gamma-ray laser is still a challenge.

1.3 Recent progress in quantum electronics (added by the Editors)

1.3.1 Physics of lasers

10 Physical Foundations of Quantum Electronics

1.3.2 Laser physics

1.3.3 New trends in nonlinear optics

1.3.4 Atom optics

1.3.5 Optics of nonclassical light

12 Physical Foundations of Quantum Electronics

types of nonclassical light have been generated. In addition to single-photon and