Laszlo Solymar, Ekaterina Shamonina - Waves in Metamaterials-OUP (2009) PDF

Waves in Metamaterials
This page intentionally left blank

Waves in Metamaterials
L. Solymar
Imperial College, London
E. Shamonina
University of Erlangen-Nürnberg
1
3
Great Clarendon Street, Oxford OX2 6DP
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With oﬃces in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
c L. Solymar and E. Shamonina 2009

The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First Published 2009
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloging in Publication Data
Data available
Printed in Great Britain
on acid-free paper by
CPI Antony Rowe, Chippenham, Wiltshire
ISBN: 978–0–19–921533–1
1 3 5 7 9 10 8 6 4 2
To Oscar and Sasha
Preface
Metamaterials is a new subject. The term ‘metamaterial’ could have

been coined at any time in the last hundred years or so (Greek prefixes
have always been fashionable) meaning that it is something well beyond
ordinary materials, but, apparently, nobody did so. The birth of the
term coincides with the birth of the subject. It appears in the first
paper (Smith et al., 2000) that showed the existence of an artificial
material that can have both negative permeability and permittivity. The
roots go back as far as the nineteenth century with a paper by Bose
(1898) concerned with the effect of a man-made twisted structure on
the polarization of an electromagnetic wave.
So what are metamaterials? There is no consensus. A brief and quite
good definition is ‘artificial media with unusual electromagnetic proper-
ties’. This is good because it captures the spirit of the subject. On the
other hand, if we accept this definition then the burden only goes back
on the meaning of the word ‘unusual’. What is usual to one man/woman
might be unusual to another one. And anyway, is the definition general
enough? Are we always concerned with artificial media? Not really. At-
tempts to produce a flat lens with subwavelength resolution rely mostly
on silver films, not an artificial material. The unfortunate fact is that
no definition exists that would be universally applicable or universally
accepted. Thus, the best we can do is to enumerate the topics that we
have included in this book and that, at least in our definition, belong to
the subject of metamaterials.
Chapter 1 is just a summary of the basic concepts and basic equations
that appear later in the book. Chapter 2 is an attempt to give both an
introduction and an overview. It is only from Chapter 3 that we start
to go into more detail. Chapter 3 is concerned with plasmas and par-
ticularly surface plasma waves. Although many in the field might think
that plasmas have only marginal relevance to metamaterials we believe
that they occupy a central position both in the theoretical formulations
and in establishing a physical picture. Growing exponential waves, for
example, (often referred to as amplification) make no sense unless we
refer to the excitation of surface plasma waves. Chapter 4 is divided
into two parts. One is a gallery of various small resonators (we describe
more than three dozen) suitable for components in a metamaterial. The
second part concentrates on the very popular element known as the split-
ring resonator and determines its electrical properties by deriving a dif-
ferential equation. Chapter 5 discusses subwavelength imaging, which is
still a controversial subject. We attempt to give a fair summary of the
viii Preface
achievements and of some of the misconceptions. Chapter 6 discusses

waveguides, mainly the effect of metamaterials on their cutoff frequency
but, in addition, various applications involving metamaterials are also
discussed. Chapters 7 and 8 are mostly concerned with our own work
on magnetoinductive waves. Chapter 9 is a kind of last-ditch attempt
to discuss phenomena that did not fit into earlier treatments. It dis-
cusses seven different topics that include nanoparticles and invisibility.
In Chapter 10 we look at the subject of metamaterials asking the ques-
tion how it started and how it became so popular. We have a large
number of appendices, most of them serving their traditional purpose;
to offer a mathematical derivation that readers might find too time con-
suming to do themselves or make some comments that are of interest
only to a specialized audience.
Having discussed what is included we should also say what is not in-
cluded. We are aware that we have given very little attention to photonic
bandgap materials. Our main argument for this omission is that they
are not really part of metamaterials. Both the physics and the mathe-
matical treatments are quite different. The ideal metamaterial contains
many elements per free-space wavelength, whereas photonic bandgap
materials are traditionally based on Bragg interaction and on a large
difference between the dielectric constants. Two further omissions are
chiral elements and non-linear effects. We felt that including them would
be too much of a diversion from our main aim, to give a fairly unified
treatment of the phenomena associated with metamaterials. We made
an exception with parametric amplification of magnetoinductive waves
which is included in Chapter 8.
After the content it is customary to say what the expected audience is
and who in our opinion might benefit from reading the book. We believe
it would be perfectly feasible for a research-oriented undergraduate to
take this book in hand during his/her last year of study, look at it,
turn the pages, read it here and there, draw a few sketches, make a few
calculations on the back of an envelope and put it (or not, as the case may
be) among those research subjects he/she might be interested to pursue
in the future. The attraction for graduate students is to learn about
the subject as a whole beyond their narrower speciality. Experts might
buy the book for the library of their institution for the benefit of the
graduate students but will they read it themselves? A good proportion
of the experts are doers and not readers. They may not want to have
their ideas tainted by learning about other people’s feats. There are
many though who are interested in finding out what they might have
missed and are willing to delve into a book just in case they come across
something new. And there is one more potential market for a book in
a fast growing field. Those who feel that their decade (or two?) -old
involvement with a particular field no longer leads to tangible results,
and are thinking of changing discipline. This book might facilitate the
choice for them.
We had no policy on references. It was clear from the beginning
that we would not be able to refer to all the papers written about the
ix
subject or relevant to the subject. A conservative estimate would make

their number exceed 1500. The way we proceeded was to write about a
particular topic and choose the references as they came up. Our total
number is less than 500. We are bound to have missed important papers.
We offer our apologies.
Concerning presentation we have to say a few words about acronyms.
We don’t like them, although, we must admit, we are responsible for
some of them. It is not particularly surprising that in an acronym-mad
world many of us in the metamaterial community have been affected by
the disease. Just to show how far the disease has spread we enumerate
in Appendix A the acronyms we could find after a brief search. The
use of acronyms is probably counter-productive. It is supposed to save
time. More likely it leads to occasional incomprehension and miscom-
prehension. We believe authors of technical books should refrain from
proliferating them. We shall make a conscious effort not to use any
of the new ones invented specifically for the benefit of metamaterials
with a few exceptions: SRR for split-ring resonator, SPP for surface
plasmon–polariton and MI for magnetoinductive.
We have lots of people to thank for getting to the end of this journey.
We were lucky to have plenty of support in our efforts. Our thanks go
first of all to Lesha Sydoruk who helped us at every stage of the book
by reading every section, finding mistakes and inaccuracies, and making
numerous proposals for improvement. We also wish to acknowledge the
help we received from our close collaborators, David Edwards, Frank
Hesmer, Victor Kalinin, Anya Radkovskaya, Klaus Ringhofer, Misha
Shamonin, Chris Stevens, Rich Syms, Zhenya Tatartschuk, Ian Young
and Sasha Zhuromskyy. For discussions on a number of subjects we are
obliged to P. Belov, N. Engheta, M. Freire, M. Gorkunov, M. Kafesaki,
A. F. Koenderink, M. Lapine, R. Marques, S. A. Tretyakov, V. Veselago
and M. Wiltshire. Special thanks are due to those whom we asked for
original figures and who so warmly responded: K. Aydin, P. Belov, P.
Berini, S. R. J. Brueck, W. Cai, C. Caloz, H. Chen, S. A. Cummer,
G. V. Eleftheriades, N. Engheta, M. J. Freire, O. Hess, S. Hrabar, M.
Kafesaki, A. F. Koenderink, R. Marques, O. J. F. Martin, J. B. Pendry,
I. Smolyaninov, D. Schurig, V. M. Shalaev, C. M. Soukoulis, R. R. A.
Syms, S. A. Tretyakov, K. L. Tsakmakidis, W. H. Weber, T. Zentgraf,
X. Zhang and J. Zhou.
Finally, L.S. would like to thank, and thank profusely, his wife Mar-
ianne, who was willing to put up with the long hours he spent on this
book in violation of the basic principle that retired people should sit on
their laurels and work no longer. E.S. thanks her family for their help,
support and encouragement and Sasha for being so generous in letting
his mum work and for being such a good boy.
Contents
1 Basic concepts and basic equations 1

1.1 Introduction 1
1.2 Newton’s equation and electrical conductivity 2
1.3 Maxwell’s equations, fields and potentials 3
1.4 The wave equation and boundary conditions 4
1.5 Hollow metal waveguides 6
1.6 Refraction at a boundary: Snell’s law
and the Ewald circle construction 8
1.7 Fermat’s principle 11
1.8 The optical path and lens design 12
1.9 The effective ε in the presence of a current 14
1.10 Surface waves 14
1.11 Plane wave incident upon a slab 16
1.12 Dipoles 17
1.13 Poynting vector 18
1.14 Radiation resistance 18
1.15 Permittivity and permeability tensors 19
1.16 Polarizability 20
1.17 Working with tensors 21
1.18 Dispersion: forward and backward waves 21
1.19 Mutual impedance and mutual inductance 23
1.20 Kinetic inductance 24
1.21 Four-poles: impedance and chain matrices 25
1.22 Transmission line equations 26
1.23 Waves on four-poles 28
1.24 Scattering coefficients 30
1.25 Fourier transform and the transfer function 31
2 A bird’s-eye view of metamaterials 35

2.1 Introduction 35
2.2 Natural and artificial materials 35
2.3 Determination of the effective permittivity/dielectric
constant in a natural material 40
2.4 Effective plasma frequency of a wire medium 42
2.5 Resonant elements for metamaterials 44
2.6 Loading the transmission line 45
2.6.1 By resonant magnetic elements in the form
of LC circuits 46
2.6.2 By metallic rods 48
xii Contents
2.6.3 By a combination of resonant magnetic

elements and metallic rods 49
2.7 Polarizability of a current-carrying resonant loop:
radiation damping 50
2.8 Effective permeability 52
2.9 Dispersion equation of magnetoinductive waves
derived in terms of dipole interactions 55
2.10 Backward waves and negative refraction 56
2.11 Negative-index materials 58
2.11.1 Do they exist? 58
2.11.2 Terminology 59
2.11.3 Negative-index lenses 60
2.11.4 The flat-lens family 61
2.11.5 Experimental results and numerical simulations 63
2.11.6 Derivation of material parameters
from reflection and transmission coefficients 65
2.12 The perfect lens 66
2.12.1 Does it exist? 66
2.12.2 The ideal situation, εr = −1 and µr = −1 67
2.12.3 The periodic solution 69
2.12.4 The electrostatic limit: does it exist? 70
2.12.5 Far field versus near field: Veselago’s lens
versus Pendry’s lens 70
2.13 Circuits revisited 73
3 Plasmon–polaritons 75
3.1 Introduction 75
3.2 Bulk polaritons. The Drude model 77
3.3 Surface plasmon–polaritons. Semi-infinite case,
TM polarization 79
3.3.1 Dispersion. Surface plasmon wavelength 79
3.3.2 Effect of losses. Propagation length 83
3.3.3 Penetration depth 85
3.3.4 Field distributions in the lossless case 88
3.3.5 Poynting vector: lossless and lossy 90
3.4 Surface plasmon–polaritons on a slab: TM polarization 92
3.4.1 The dispersion equation 92
3.4.2 Field distributions 97
3.4.3 Asymmetric structures 100
3.5 Metal–dielectric–metal and periodic structures 102
3.6 One-dimensional confinement: shells and stripes 103
3.7 SPP for arbitrary ε and µ 106
3.7.1 SPP dispersion equation for a single interface 106
3.7.2 Domains of existence of SPPs
for a single interface 108
3.7.3 SPP at a single interface to a metamaterial:
various scenarios 110
3.7.4 SPP modes for a slab of a metamaterial 116
Contents xiii
4 Small resonators 119

4.1 Introduction 119
4.2 Early designs: a historical review 120
4.3 A roll-call of resonators 126
4.4 A mathematical model and further experimental results 144
4.4.1 Distributed circuits 144
4.4.2 Results 148
4.4.3 A note on higher resonances 152
5 Subwavelength imaging 155

5.2 The perfect lens: controversy around the concept 157
5.2.1 Battle of wits 157
5.2.2 Non-integrable fields 158
5.2.3 High spatial frequency cutoff 158
5.3 Near-perfect lens 159
5.3.1 Introduction 159
5.3.2 Field quantities in the three regions 160
5.3.3 Effect of losses: Transfer function, cutoff,
electrostatic limit 162
5.3.4 Lossless near-perfect lens with εr ≃ −1, µr ≃ −1 168
5.3.5 Near-perfect? Near-sighted! 172
5.3.6 General cutoff frequency relationships 174
5.3.7 Effect of discretization in numerical simulations 175
5.4 Negative-permittivity lens 176
5.4.2 Dependence on thickness 176
5.4.3 Field variation in the lens 178
5.4.4 Other configurations: compression 179
5.4.5 The electrostatic approximation revisited 180
5.4.6 Experimental results 183
5.5 Multilayer superlens 188
5.6 Magnifying multilayer superlens 192
5.7 Misconceptions 196
6 Phenomena in waveguides 199

6.2 Propagation in cutoff waveguides 199
6.3 Filters in coplanar and microstrip waveguides 204
6.4 Tunnelling 206
6.5 Phase shifters 208
6.6 Waveguide couplers 209
6.7 Imaging in two dimensions: transmission-line approach 210
7 Magnetoinductive waves I 213

7.2 Dispersion relations 214
7.3 Matching the transmission line 217
xiv Contents
7.4 Excitation 217

7.5 Eigenvectors and eigenvalues 218
7.6 Current distributions 220
7.7 Poynting vector 224
7.8 Power in a MI wave 225
7.9 Boundary reflection and transmission 225
7.10 Tailoring the dispersion characteristics:
biperiodic lines 227
7.11 Experimental results 231
7.12 Higher-order interactions 236
7.13 Coupled one-dimensional lines 238
7.14 Rotational resonance 242
7.15 Applications 243
7.15.2 Waveguide components 244
7.15.3 Imaging 248
7.15.4 Detection of nuclear magnetic resonance 249
8 Magnetoinductive waves II 251

8.1 MI waves in two dimensions 251
8.1.2 Dispersion equation, group velocity,
power density 252
8.1.3 Reflection and refraction 254
8.1.4 Excitation by a point source:
reflection and diffraction 258
8.1.5 Spatial resonances in hexagonal lattices 259
8.1.6 Imaging 265
8.2 MI waves retarded 266
8.2.2 Dispersion equation 268
8.2.3 The nature of the dispersion equation 269
8.2.4 A 500-element line 270
8.2.5 Conclusions 276
8.3 Non-linear effects in magnetoinductive waves 277
8.3.2 Phase matching 278
8.3.3 Theoretical formulation of amplification
for the single array 280
8.3.4 Theoretical formulation for the coupled arrays 285
8.3.5 MRI detector 287
9 Seven topics in search of a chapter 289

9.2 Further imaging mechanisms 290
9.2.1 Parallel sheets consisting of resonant elements 290
9.2.2 Channelling by wire structures 291
9.2.3 Imaging by photonic crystals 295
Contents xv
9.3 Combinations of negative-permittivity

and negative-permeability layers 296
9.4 Indefinite media 297
9.5 Gaussian beams and the Goos–Hanchen shift 299
9.6 Waves on nanoparticles 302
9.7 Refractive index close to zero 306
9.7.2 Wavefront conversion 307
9.7.3 Effect of low phase variation 308
9.8 Invisibility and cloaking 309
10 A historical review 315

10.2 Forerunners 317
10.2.1 Effective-medium theory 317
10.2.2 Negative permittivity 317
10.2.3 Negative permeability 318
10.2.4 Plasmon–polaritons 318
10.2.5 Backward waves 318
10.2.6 Theory of periodic structures 319
10.2.7 Resonant elements
small relative to the wavelength 319
10.2.8 Chiral materials 320
10.2.9 Faster than light 320
10.2.10 Frequency filters made of periodically
arranged resonant elements 320
10.2.11 Slow-wave structures 320
10.2.12 Waves arising from nearest-neighbour
interactions 321
10.2.13 Superdirectivity, superresolution,
subwavelength focusing and imaging 321
10.2.14 Inverse scattering 321
10.2.15 Bianisotropy 322
10.2.16 Photonic bandgap materials 322
10.2.17 Waves on nanoparticles 322
10.3 ... and the subject went on and flourished... 323
A Acronyms 325
B Field at the centre of a cubical lattice

of identical dipoles 327
C Derivation of material parameters from

reflection and transmission coefficients 329
D How does surface charge appear

in the boundary conditions? 331
E The Brewster wave 333

xvi Contents
F The electrostatic limit 335

F.1 Single interface 335
F.2 Symmetric slab 336
G Alternative derivation of the dispersion equation

for SPPs for a dielectric–metal–dielectric structure:
presence of a surface charge 339
H Electric dipole moment induced by a magnetic field

perpendicular to the plane of the SRR 343
I Average dielectric constants of a multilayer structure 345
J Derivation of mutual inductance between two

magnetic dipoles in the presence of retardation 347
References 349
Index 381
Basic concepts and basic
equations
1
1.1 Introduction 1.1 Introduction 1
1.2 Newton’s equation and
We shall make an attempt in this chapter to introduce most of the basic electrical conductivity 2
concepts needed later in the book and in a form upon which we can 1.3 Maxwell’s equations,
build in later chapters. Some will be very basic indeed, like fields and fields and potentials 3
potentials, some others will be closer to practical cases, e.g. the inci- 1.4 The wave equation and
dence of a plane wave upon a dielectric slab. Some will represent useful boundary conditions 4
artifices like electric and magnetic dipoles, leading from there to the 1.5 Hollow metal waveguides 6
introduction of the concept of polarizability: how large are the dipole 1.6 Refraction at a boundary:
moments excited by various fields. Boundary conditions and boundary Snell’s law and the Ewald
circle construction 8
refraction problems must of course be mentioned because they will ap-
pear repeatedly in the analysis of metamaterial properties. The Ewald
1.8 The optical path and lens
circle construction will then come in as a useful technique. Dispersion design 12
is again one of the basic concepts. In any wave phenomenon one must 1.9 The effective ε in the
know the relationship between frequency and wave number.1 Talking presence of a current 14
about dispersion necessitates the introduction of forward and backward 1.10 Surface waves 14
waves. The latter were hardly mentioned in the past, their properties 1.11 Plane wave incident
rarely emphasized in studies of electromagnetic waves, but they have upon a slab 16
lately risen to fame due to their role in metamaterials. Another de- 1.12 Dipoles 17
parture from usual introductions is the prominence of circuit theory, a 1.13 Poynting vector 18
subject often neglected in physics syllabuses. Very often we shall have 1.14 Radiation resistance 18
to look at phenomena both from the point of view of fields and cir- 1.15 Permittivity and
cuits. Sometimes the two explanations reinforce each other, sometimes permeability tensors 19
we shall resort to only one of them because the other one may be too 1.16 Polarizability 20
unwieldy. Acquaintance with Fourier analysis is of course a basic re- 1.17 Working with tensors 21
quirement whether one is concerned with the temporal or the spatial 1.18 Dispersion: forward
regime. In more general terms it means working in reciprocal space but and backward waves 21
we shall do that only sparingly, mostly in the one-dimensional context. 1.19 Mutual impedance and
The subject of metamaterials can be highly mathematical, particu- mutual inductance 23
larly when radiation effects are included. Our aim, as mentioned in 1.20 Kinetic inductance 24
the Preface, is to offer a treatment that can be happily studied by a 1.21 Four-poles: impedance
and chain matrices 25
final-year undergraduate so we shall try to keep it simple. Some mathe-
matics is of course unavoidable. There are Maxwell’s equations to start 1.22 Transmission line
equations 26
with, which involve a fair amount of vector analysis. So we shall present
1.24 Scattering coefficients 30
1 Engineers prefer to call it propagation coefficient, a usage we shall sometimes prac-
1.25 Fourier transform and
tise when it sounds more appropriate in the context.
the transfer function 31
2 Basic concepts and basic equations
the equations but they will not be called upon very often. The wave
equation and some simple theory of differential equations is of course
necessary. Some tensors will also be mentioned, but only at an elemen-
tary level: no fancy co-ordinate transformations, only a description in
terms of matrices. There is obviously not enough space to derive all the
equations needed. The question is which equations to derive, or at least
give hints about the derivation, and which ones to postulate. We hope
the compromises chosen will be acceptable.
The aim of this chapter is to serve partly as an introduction to con-
cepts and equations, and partly as a reference library to which the reader
can turn when the need arises.
1.2 Newton’s equation and electrical

conductivity
The most basic equation, historically or otherwise, is the equation of
motion postulated by Newton. It comes into electrical studies when a
charged particle is affected by an electric field. For an electron of mass
m, and charge e, the equation can be written as

dv v
m + = eE , (1.1)
dt τ
where v is the velocity, E is the electric field and t is time. Note that
we have added a damping term with a relaxation time, τ . In this book,
with very few exceptions, we shall assume temporal variation in the
2
Unfortunately, the same harmonic form2 exp(j ωt) where ω is the frequency. With this assumption the
time variation is denoted in four differ- velocity may be expressed from eqn (1.1) in the form
ent ways in the literature. The expo-
nent may be −i ωt, i ωt, j ωt and −j ωt. e E
Admittedly, one rarely sees the second v= , (1.2)
and the fourth one but the other two m 1
jω +
are equally popular, −i ωt with physi- τ
cists, and j ωt with electrical engineers. and the current density as
We shall adopt here j ωt for the reason
that circuit quantities will often appear
N e2 τ E
and then, surely, the reactance of an in- J = N ev = , (1.3)
ductor or a capacitor should be denoted m 1 + j ωτ
by j ωL and 1/j ωC and not by −i ωL which may also be written as
and −1/i ωC.
J = σE , (1.4)
where
σ0 N e2 τ
σ= , σ0 = , (1.5)
1 + j ωτ m
and σ0 is the electrical conductivity. Note that we have abandoned here
the time-honoured notation for conductivity that is simply σ instead of
σ0 . The reason is that we wish to reserve σ, a complex quantity, for
giving the relationship between the current density and the electric field
that will be often needed in the form of eqn (1.5).
1.3 Maxwell’s equations, fields and potentials 3
1.3 Maxwell’s equations, fields and

potentials
Maxwell’s equations in the differential form are given below
∇×H = J + j ωD , (1.6)
∇×E = −j ωB , (1.7)
∇·D = ρ, ∇· B = 0, (1.8)
D = εE , B = µH . (1.9)
The electric and magnetic fields may also be expressed in the form
∂A
E = −∇ϕ − and H = ∇ × A. (1.10)
∂t
In eqns (1.6)–(1.9) the quantities printed with bold letters are vectors, H
is the magnetic field, B is the magnetic flux density, D is the electric flux
density, also called the dielectric displacement, J is the current density, ρ
is the charge density, A is the vector potential, ϕ is the scalar potential,
µ and ε are the material constants, permeability and permittivity. ∇ is
a differential operator defined as
∂ ∂ ∂
∇= ix + iy + iz , (1.11)
∂x ∂y ∂z
where ix , iy and iz are unit vectors in the x, y and z directions.
Two further relationships often referred to in the book, are Faraday’s
and Ampere’s laws, the former relating the temporal derivative of Φ, the
magnetic flux, to V , the induced voltage, as
V = −j ωΦ , (1.12)
and the latter relating the line integral of the magnetic field to I, the
enclosed current:
Z
Hds = I , (1.13)
where ds is the line element vector. The third important relationship we

shall need is that between the vector potential and the current density.
It is of the form3 3
Note that this is valid only in the low-
frequency limit that ignores retardation
µ0
Z
J that is the finite velocity of the electro-
A= dτ , (1.14) magnetic waves, but this will suffice for
4π r
most of the book. Retardation will be
where dτ is the volume element, the integration is over the region in introduced in Section 8.2.
which J is finite, and r is the distance between the point where the
current flows and where the vector potential is evaluated.
Let us now return to the permittivity and permeability that will often
be discussed in the book. They can be written as
ε = ε0 εr and µ = µ0 µr , (1.15)
where εr and µr are the relative permittivity (also called dielectric con-
stant or dielectric function) and the relative permeability. In the early
times when Gaussian units were in fashion ε0 and µ0 were taken as
unity. With the emergence of SI (System International) units they have
the numerical values of
ε0 = 8.85 × 10−12 A s V−1 m−1 , µ0 = 4π × 10−7 V s A−1 m−1 . (1.16)
In vacuum εr = µr = 1. In a material, they are different from unity

due to the appearance of P and M, the electric and magnetic polariza-
tions, respectively. In a material containing N particles (atoms, mole-
cules, artificially inserted elements) we can write
P = Np and M = Nm , (1.17)
where p and m are the electric and magnetic dipole moments defined as
p = qd and m = µ0 ISip . (1.18)

q is charge and d is a vector connecting the negative and positive charges
(a) (b) infinitesimally close to each other (Fig. 1.1(a)). The magnetic dipole is
I equivalent to an infinitesimal loop in which a current I flows. S is the
q area of the loop. The direction of the magnetic dipole moment, given by
–q the unit vector ip , is perpendicular to the plane of the loop (Fig. 1.1(b)).
The electric and magnetic dipole moments may already be in the
Fig. 1.1 (a) Electric and (b) magnetic material or may come about as a response to the incident field. The
dipole electric and magnetic flux densities may then be written as
D = ε0 E + P (1.19)
and
B = µ0 H + M . (1.20)
The usual problem, both for natural materials and metamaterials, is
to find the relationship between the fields and the polarizations and
determine from that the values of εr and µr . In natural materials this
is a serious problem because, even in the linear case, we have only a
very approximate idea of what is going on at the atomic or molecular
level. For man-made materials the difficulties are somewhat easier to
overcome because we have a better knowledge of the properties of the
elements inserted.
1.4 The wave equation and boundary

conditions
The equation we shall need often is the wave equation. It may be ob-
tained from Maxwell’s equations in the form
1.4 The wave equation and boundary conditions 5
∇2 E + k 2 E = 0 , (1.21)
where
k 2 = ω 2 µε . (1.22)
In Section 1.2 we started with the assumption of harmonic time varia-
tion. We add to it now harmonic variation in space as well. The solution
of eqn (1.21) is assumed in the form
E = E0 e −j k · r , (1.23)
where k is the wave vector that indicates the direction of propagation, r
is a position vector and E0 is a constant vector perpendicular to k. This
is called a plane wave because none of the variables change in a plane
perpendicular to k.
The magnitude of the wave vector k = |k| is known as the wave
number by physicists and as the complex propagation coefficient by en-
gineers. Substituting eqn (1.23) into the wave equation and separating
the variables we find that a solution exists when the equation
kx2 + ky2 + kz2 = k 2 (1.24)

is satisfied. Note that k may be complex. Its real and imaginary parts
will be denoted by
k = β − jα, (1.25)
and they will be called the propagation coefficient, β, and the attenua-
tion coefficient, α. Assuming propagation in the z direction and taking
the electric field as a scalar, eqn (1.23) may be written as
E = E0 e −j βz e −αz , (1.26)
i.e. the wave declines exponentially in the direction of propagation as
may be expected. Although losses are indispensable if we want an ac-
curate answer, very often we can find the main features of a physical
phenomenon by neglecting losses. In those cases we shall make no dis-
tinction between k and β.
In this plane-wave solution the magnetic field is perpendicular both
to the electric field and to the wave vector, and its magnitude is equal
to
r
E µ0
H= , η0 = , (1.27)
η0 ε0
where η0 is the free-space impedance equal to 120π ohm. Note that the
vectors E, H and k are a right-handed set. It has become fashionable to
refer to media where this relationship is satisfied as right-handed media
in contrast to left-handed media to be discussed in Section 2.11.
The velocity of propagation in free space is equal to c, the velocity of

light. If the propagation is in a medium with material constants ε and
µ then the velocity of propagation is
ω c
v= = √ , (1.28)
k µr εr
and the relationship between the electric and magnetic field becomes
r
E µr
H= and η = η0 (1.29)
η εr
is called the impedance of the medium. We may now define the index
of refraction as
√
n= µr εr , (1.30)
which will probably come as a surprise to many in optics because they
learned to ignore permeability. In fact, in metamaterials µr can be
different from unity even in the optical region, therefore eqn (1.30) is
the proper definition of the index of refraction. In general, it is a measure
of the optical density of the medium. The denser the medium the lower
is the velocity of propagation.
When waves are incident from one medium upon another then one
would expect some quantities to be continuous across the boundary.
These are, as may be found in any textbook on electromagnetism, the
tangential components of the electric and magnetic fields and the normal
4
This formulation ignores surface cur- components of the electric and magnetic flux densities.4 If H1 , E1 are
rents and surface charges which rarely the fields in medium 1 of material constants µ1 , ε1 , and H2 , E2 , µ2 ,
appear in this book. They can be in-
troduced for surface plasma waves (see
ε2 are the corresponding quantities in medium 2 then the conditions in
Section 3.3.1) mathematical form are
H1t = H2t , E1t = E2t , ε1 E1n = ε2 E2n , µ1 H1n = µ2 H2n , (1.31)
where subscripts t and n stand for the tangential and normal compo-
nents. If medium 2 is a metal, it is nearly always (surface plasma waves
are again exceptions) taken as having infinite conductivity, which means
that the tangential component of the electric field must vanish in medium
2.
1.5 Hollow metal waveguides

Can electromagnetic waves propagate inside a metallic pipe? The ex-
periment to conduct is to raise such a pipe to one of our eyes. If we can
see through it that’s a clear proof for the propagation of electromagnetic
waves. It is true, however, that if we take an ordinary pipe (say a cop-
per pipe used for central heating) its diameter is enormous compared
with the wavelength of light, whatever its colour might be. Therefore,
we should qualify the previous statement by saying that electromagnetic
waves can propagate in a metallic pipe, provided it has a diameter large
1.5 Hollow metal waveguides 7
enough relative to the wavelength. How large should the dimensions be x

to enable a wave to propagate? For that we need a moderate amount of
mathematics. The pipe we shall analyze is a rectangular one as shown
in Fig. 1.2. In technical terms it is called a waveguide because it can
guide waves. z
b
We shall now assume that only three field components are present, the
x component of the electric field5 and the y and z components of the y
a
magnetic field, and we further assume that the electric field is indepen-
dent of the x co-ordinate. The relevant equations to satisfy are those of
Maxwell: eqns (1.6) and (1.7). With the field quantities assumed they Fig. 1.2 A rectangular metallic
may be written as waveguide
5
∂Hz ∂Hy When the electric field has compo-
− = j ωε0 Ex , (1.32) nents only in the transverse plane while
∂y ∂z
propagating in a waveguide, the wave
∂Ex is referred to as a transverse electric or
= −j ωµ0 Hy , (1.33)
∂z TE wave.
∂Ex
= j ωµ0 Hz . (1.34)
∂y
Eliminating Hy and Hz from the above equation we end up with the
wave equation in the form
∂ 2 Ex ∂ 2 Ex
2 + + k02 Ex = 0 , (1.35)
∂y ∂z 2
√
where k0 = ω ε0 µ0 . The propagation along the waveguide in the z di-
rection may be described by the function exp(−j kz z) as for plane waves.
But there is a new feature now. The wave is inside a waveguide. As we
learned in the previous section the tangential component of the electric
field must vanish at a metal boundary. These boundary conditions are
clearly satisfied on the x = 0 and x = b surfaces since the chosen electric
field is perpendicular to those surfaces. But the Ex component must
vanish at y = 0 and y = a. It would vanish if those boundaries coin-
cided with the nodes of a standing wave. Hence, it is logical to assume
that the variation in the y direction will be of the form sin(ky y), and if
we choose the nearest nodes then we need to take ky = π/a. Hence, the
solution for the electric field is
Ex = E0 sin(ky y) e −j kz z . (1.36)
Substituting the above equation into eqn (1.35) we find the relation-
ship
π 2
kz2 = k02 − . (1.37)
a
2
Clearly, there is propagation in the z direction when kz is positive, and
exponential decay when kz2 is negative. The limiting case is when kz = 0,
which occurs when
2
2π π 2
k02 = = or λ = 2a . (1.38)
λ a
The message is clear. As the size of the waveguide shrinks there is

a point at which propagation is no longer possible, and that happens
when the waveguide is half-wavelength wide. If a < λ/2 then no wave
can propagate. The waveguide is referred to as a cutoff waveguide.
1.6 Refraction at a boundary: Snell’s law

and the Ewald circle construction
The next basic concept we shall present is refraction. A wave incident
upon a boundary is partly reflected and partly refracted. There is not
much point in solving the general case when the fields can have arbitrary
polarizations. It is unwieldy and has little practical significance anyway.
We shall restrict generality and consider only a two-dimensional situa-
tion when there is no variation in the y direction and the wave is incident
at an angle θ1 . The plane of incidence is the xz plane. In Fig. 1.3(a)
the electric, whereas in Fig. 1.3(b) it is the magnetic polarization that
is perpendicular to the plane of incidence. The former is known as a
TE and the latter as a TM wave. TE and TM still stand for transverse
electric and transverse magnetic but it means now transverse6 to the
plane of incidence.
In both cases the waves incident upon the boundary are partly re-
flected (at an angle θ1 ) and partly refracted (at θ2 ) as shown in Fig. 1.3.
The aim is to find the reflection and transmission coefficients and the
relationship between θ1 and θ2 .
We shall do the derivation for a TM wave and write the y component
of the magnetic field in medium 1 as
Hy1 = A e −j (kz1 z + kx1 x) + B e j (kz1 z − kx1 x) , (1.39)

where A and B are the amplitudes of the incident and reflected waves,
and kz1 and kx1 are the z and x components of the wave vector satisfying
the condition
2 2
kz1 + kx1 = k12 = ω 2 µ1 ε1 . (1.40)
The electric field will have z and x components as follows from
eqn (1.6). For matching we shall need only the tangential component,
which is Ex1 . It may be obtained in the form
1 ∂Hy1
Ex1 = −
j ωε1 ∂z
6 This terminology is not a fortunate one. For plane waves both the electric and the magnetic fields are in the transverse
plane, perpendicular to the direction of propagation. They are called, and should be called, transverse electromagnetic or TEM
waves. This is in agreement with the definition in the previous section where a mode in a waveguide is called a TE mode,
provided it has only transverse electric fields. Calling waves, as we do in this section and many times later, TE or TM waves
depending which field is perpendicular to the plane of incidence can only lead to confusion, but the notation is so widespread
in the literature that we reluctantly accept it. There is actually another set of notations, used often by physicists, which is
based on the direction of the electric field vector, E. It is called p polarization when E is parallel and s polarization when it is
perpendicular to the plane of incidence (s coming from the German word senkrecht).
1.6 Refraction at a boundary: Snell’s law and the Ewald circle construction 9
Fig. 1.3 Refraction of a TE (a) and a TM (b) wave
kz1 h
A e −j (kz1 z + kx1 x) − B e j (kz1 z − kx1 x) .
i
=
ωε1
(1.41)
In medium 2 only one wave will propagate, the transmitted wave,

because it is assumed to be infinitely wide. Denoting its amplitude by C
the tangential components of the electric and magnetic fields in medium
2 are
Hy2 = C e −j (kz2 z + kx2 x) (1.42)

and
kz2
Ex2 = C e −j (kz2 z + kx2 x) . (1.43)
ωε2
According to our boundary conditions (eqn (1.31)) we need to match
both field components at the boundary z = 0. It is immediately obvious
that matching is only possible if the fields vary in the same manner in
the x direction, which requires that
kx1 = kx2 , (1.44)

meaning that the phase velocity along the x direction must be the same
on both sides of the boundary. It follows from the geometry that
7
It was first published by Rene
kx1 = k1 sin θ1 and kx2 = k2 sin θ2 , (1.45) Descartes in his La Dioptrique in 1637
but, apparently, a contemporary of his,
whence eqn (1.44) modifies to Villebrord van Roijen Snell, discovered
it before him but failed to publish it. In
√ √ the English language world it is known
µ1 ε1 sin θ1 = µ2 ε2 sin θ2 , (1.46)
as Snell’s law, in France as la loi de
which, remembering the definition of eqn (1.30), is known as Snell’s Descartes. In many other countries, in
a spirit of compromise, it is called the
law,7 feared and respected by generations of schoolchildren for at least
Snell–Descartes law.
a couple of centuries.
B B
P
P Q
k1b
k1
k2 kɫ k1a
O O
ș1 ș 2 k2a
B’ B’
(a) (b)
Fig. 1.4 Wave vector diagram at the boundary of two media. (a) k1 < k2 and (b) k1 > k2
Further conditions of matching are

kz1 kz2
A+B =C and (A − B) = C, (1.47)
ε1 ε2
whence the reflection and transmission coefficients are obtained as
B 1 − ζe C 2
R= = and T = = , (1.48)
A 1 + ζe A 1 + ζe
where
ε1 kz2
ζe = . (1.49)
ε2 kz1
This is a rather unusual form of the well-known reflection and trans-
mission coefficients but convenient for the present book. If instead of a
TM wave we assume a TE wave with the incident electric field in the
y direction (Fig. 1.3(a)) then the derivation of the reflection and trans-
mission coefficients follows the same pattern and the resulting equations
are identical with those of eqn (1.48), but ζe needs to be replaced by
µ2 kz1
ζm = . (1.50)
µ1 kz2
The condition that the velocity of the wave must be the same on both
sides of the boundary leads to a simple construction method for finding
the direction of the refracted wave. It is to be recommended not only
for its simplicity but also, and mainly, because it gives new insight into
the phenomenon, particularly useful when discussing the refraction of
backward waves.
The loci of the wave vectors on the two sides of the boundary are
represented by concentric circles of radii k1 and k2 . They show all the
Boundary
Medium 1 Medium 2
xB B Land Sea
P ș2
ș1
zA
xP
A
xA
zA z=0 zB
(a) (b)
Fig. 1.5 Fermat’s principle for the refraction problem. (a) Geometrical scheme. (b) Practical example ‘damsel in distress’
possible directions for the wave vectors k1 and k2 . The geometrical

construction is shown in Fig. 1.4(a). We draw the wave vector k1 cor-
responding to its angle of incidence, θ1 . On the diagram k1 = OP. Now
drop a perpendicular from P to the boundary BB′ . This line intersects
the circle of radius k2 at Q. Hence k2 = OQ will be the direction of wave
propagation in medium 2 since it may be seen to satisfy the condition
posed by eqn (1.45). The angle of refraction will be θ2 .
To show the usefulness of the construction let us assume that k1 > k2 .
It may now be clearly seen in Fig. 1.4(b) that for an incident wave vector
of k1a a wave can propagate in medium 2 with wave vector k2a , whereas
no wave can propagate in medium 2 for the incident wave vector k1b .
The perpendicular drawn from P to BB′ does not intersect the inner
circle but intersects the outer circle yielding the wave vector k1c , which
is that of the reflected wave. It is clearly the case of total internal
reflection.
1.7 Fermat’s principle

Now we attack the boundary refraction problem from yet another angle
using Fermat’s principle of minimum time. This will tell us how to
get from point A in medium 1 (see Fig. 1.5(a)) to point B in medium 2.
The problem is also known as that of the damsel in distress (Fig. 1.5(b)).
Assume that medium 1 is land and medium 2 is the sea. A handsome
young man (the story comes from the times when sex equality was not
uppermost in the minds of storytellers) standing on firm land in point A
perceives that a young lady at point B is to be overcome by dangerous
waves. He is of a very generous nature and immediately decides to go

to the rescue of the lady but there is a problem. To which point on
the shore should he run? That surely depends on the relative speed he
can muster on land and sea. What he wants is to minimize the time.
Mathematically his problem appears as
AP PB
+ = Min! (1.51)
vland vsea
With the co-ordinates defined in Fig. 1.5(a) the time needed to reach
point B is
q q
(xP − xA )2 + zA2 2
(xB − xP )2 + zB
T = + . (1.52)
vland vsea
We find the optimum time with differentiation by the unknown xP
dT xP − xA 1
=0 = q
dxP v
2 land
(xP − xA )2 + zA
xB − xP 1
−q , (1.53)
2 vsea
(xB − xP )2 + zB
which may be seen to reduce to

sin θ1 sin θ2
= . (1.54)
vland vsea
If instead of the speed of running and swimming we substitute the
speed of an electromagnetic wave in medium 1 as c/n1 and in medium
2 as c/n2 then, yet again, we obtain Snell’s law as
n1 sin θ1 = n2 sin θ2 . (1.55)
1.8 The optical path and lens design

Next, we ask about the phase of the electromagnetic wave as it travels
from A to B via P in Fig. 1.5(a). The phase, as we know, is given by
k · r, which, for the present case, comes to
2π
ϕtotal = k1 AP + k2 PB = (n1 AP + n2 PB) . (1.56)
λ
The total path is AP + PB but that is not the quantity of interest.
It is clear from eqn (1.56) that what we need is n1 AP + n2 PB, which
we call the optical path. In general, the total optical path is equal to
the geometrical path in each medium multiplied by the corresponding
refractive index.
If the optical paths of various trajectories are equal then the phases
of the arriving rays must also be equal. If we want to convert a wave
emanating from a point to a plane wave then we must make sure that
1.8 The optical path and lens design 13
Fig. 1.6 (a) Lens design using the con-

(a) (b) cept of optical path, (b) lens contours
all the optical paths are equal when the plane wavefront is reached.
Traditionally it is the job of a lens to realize this conversion. So, let us
find the contour of a lens that will do it.
We shall assume a 2D geometry with the rays emanating from point O
and work out the optical path of a general ray travelling in the θ direction
(see Fig. 1.6(a)) and that of the axial ray when they reach the P1 P2
wavefront. The former one propagates in free space hence its optical
path is OP1 , whereas the latter one propagates in free space from O to
C and in the lens from C to P2 . Hence, the equality of optical paths
demands
OP1 = OC + nCP2 , (1.57)
which may be written in polar co-ordinates as
r = d + n(r cos θ − d) , (1.58)
whence
d(1 − n)
r= . (1.59)
1 − n cos θ
This may be recognized as a hyperbola in polar co-ordinates. It has
an asymptote at
1
cos θ = . (1.60)
n
Transformation to normalized Cartesian co-ordinates,
r r
z= cos θ and x= sin θ (1.61)
d d
yields the more familiar equation of a hyperbola8 8
2D lens contours are generally known
as circles. In fact, the above equation
2 of a hyperbola for low angles is approx-
x2

n 1 imately the same as that of a circle.
z− − 2 = . (1.62)
n+1 n −1 (n + 1)2
For n = 1.5 and 2 the contours are plotted in Fig. 1.6(b). As usual, it
has a convex shape.
1.9 The effective ε in the presence of a

current
We have already mentioned several times the relationship of the current
density to other variables like the electric field or the vector potential
but the wave solutions presented so far ignored the possible existence of
a current. The aim of this section is to take the current into account in
a simple way by defining an effective dielectric constant. The way to do
this is to rewrite the right-hand side of eqn (1.6) in the form
J + j ωεE = j ωεeff E , (1.63)

with which εeff is defined. But J is related to E by eqn (1.4), whence
we find
σ0 1
εeff = ε + . (1.64)
j ω 1 + j ωτ
In the low-frequency limit we find
σ0
εeff = ε − j . (1.65)
ω
This is the form used for lossy dielectrics. In the high-frequency limit,
taking ε = ε0 ,
!
ωp2
εeff = ε0 1 − 2 , (1.66)
ω
where
N e2
ωp2 = (1.67)
ε0 m
is known as the plasma frequency. Equation (1.66) is the dielectric
constant of an ideal plasma. Interestingly, it is negative below the plasma
frequency and positive above the plasma frequency.
Note that eqn (1.22) still applies. We only need to replace ε by εeff .
When εeff is complex, k will also be complex, implying both propagation
and attenuation. When εeff is a negative real number then k is purely
imaginary. There is no propagation. The wave declines without any
change in the phase.
1.10 Surface waves

We have seen what happens when a wave is incident from one medium
upon another one. Let us now explore another aspect of wave propaga-
tion related to a boundary between two media, namely that the wave
can stick to the boundary. When it does so it is called a surface wave.
Let us see the conditions under which these waves may exist.
Assuming a TM wave as in Section 1.6 we have already got the expres-
sions for Hy and Ex on both sides of the boundary. The new feature is
1.10 Surface waves 15
that the waves can propagate along the boundary but their amplitudes Boundary
decline exponentially away from the boundary as shown in Fig. 1.7.
Medium 1 Medium 2
This can happen when kx is sufficiently high so that kz1 and kz2 are
imaginary and must be replaced by −j κ1 and −j κ2 , where κ1 and κ2
are real. Hence, the propagation coefficient in the x direction is obtained x
Field amplitude
as
z
kx2 = k12 + κ21 = k22 + κ22 . (1.68)
We are looking for a wave that can exist on the surface without an
Fig. 1.7 Variation of a surface wave
input.9 Hence, we can take A = 0 but B is taken as finite. It is the amplitude at a boundary
amplitude of the wave that declines away from the boundary in the
negative z direction. The equations for the magnetic and electric fields 9
In more pretentious language it means
are then that we are looking for eigensolutions.
Hy1 = B e κ1 z e −j kx x ,
−κ1
Ex1 = B e κ1 z e −j kx x ,
j ωε1
−kx
Ez1 = B e κ1 z e −j kx x (1.69)
ωε1
in medium 1 and
Hy2 = C e −κ2 z e −j kx x ,
κ2
Ex2 = C e −κ2 z e −j kx x ,
j ωε2
−kx
Ez2 = C e −κ2 z e −j kx x (1.70)
ωε2
in medium 2. In order to satisfy the boundary conditions we need to
match Hy and Ex at z = 0, which yields10 10
Note that the boundary condition
ε1 Ez1 = ε2 Ez2 is automatically satis-
κ1 κ2 fied, provided eqn (1.71) holds.
B=C and − = . (1.71)
ε1 ε2
Note that we have taken both κ1 and κ2 to be positive. Hence,
eqn (1.71) can be satisfied only when ε2 is negative. This is a con-
dition for the existence of a surface wave. It is interesting to note at this
stage that eqn (1.71) is equivalent to the condition that
ζe = −1 . (1.72)
Another condition to satisfy is given by eqn (1.68). We may then
substitute κ1 and κ2 from eqn (1.68) into eqn (1.71) and obtain, after
some algebraic operations, a relationship between kx and ω,
r
ω εr1 εr2
kx = . (1.73)
c εr1 + εr2
We want the wave to travel along the surface, hence kx must be real.
With εr2 < 0 we may then write the condition to satisfy as
Fig. 1.8 Schematic presentation of

wave propagation in three media when
a wave is incident from medium 1
εr1 εr2 < 0 and εr1 + εr2 < 0 . (1.74)
This is actually the situation pertaining to an ideal plasma that can

have a negative dielectric constant, as shown in the previous section.
The waves are usually referred to as surface plasmons, which is really
a misnomer because the phenomenon is purely classical. It is as if we
would always talk about photons instead of electromagnetic waves. We
would of course be right but there is not much point in using quantum-
mechanical terminology when the classical would do.
1.11 Plane wave incident upon a slab

In Section 1.4 we investigated the reflection and transmission of a plane
wave when incident upon another medium. We shall now add one more
boundary, as shown in Fig. 1.8. For simplicity, media 1 and 3 are as-
sumed to have the same material constants and the input wave is a TM
wave, i.e. the magnetic field has only a z component.
The major difference from the previous study is that in addition to the
transmitted wave there will now be a reflected wave as well in medium
2. The amplitudes, as may be seen in Fig. 1.8, are denoted by A, B,
C, D and F . The technique of solution is the same as for the single
boundary but now we need to match the tangential components of the
electric and magnetic fields at both boundaries, at z = 0 and at z = d.
This leads to four equations from which the unknown amplitudes may be
determined. After a number of algebraic operations we find the reflection
and transmission coefficients as
B 2j (1 − ζe2 ) sin(kz2 d)
R= = , (1.75)
A (1 + ζe )2 e j kz2 d − (1 − ζe )2 e −j kz2 d
F 4ζe
T = = , (1.76)
A (1 + ζe ) e z2 − (1 − ζe )2 e −j kz2 d
2 j k d
C 2(1 + ζe )
= , (1.77)
A (1 + ζe ) e z2 d − (1 − ζe )2 e −j kz2 d
2 j k
1.12 Dipoles 17
D 2(ζe − 1)
= , (1.78)
A (1 + ζe ) e z2 d − (1 − ζe )2 e −j kz2 d
2 j k
and, as before, ζe should be replaced by ζm for an incident TE wave.
1.12 Dipoles
The microscopic electric and magnetic dipole moments, p and m, were
introduced in Section 1.3 for determining the macroscopic electric and
magnetic polarizations, P and M. However, dipoles are useful in many
other contexts partly because they can provide good approximate models
to describe complicated electromagnetic phenomena but also because
they can form the basis of actual devices like the electric dipole antenna.
We shall present here the equations relating the dipole moments to
fields in a compact vectorial form that physicists prefer. It might look
unfamiliar to electrical engineers:
3(r · p)r − r2 p 2

1 2 r p − (r · p)r
E= (1 + j k0 r) + k0 e −j k0 r ,
4πε0 r5 r3
(1.79)
with k0 = 2π/λ. The relationship between the magnetic dipole moment
and the magnetic field takes an entirely analogous form: p should be
replaced by m and ε0 in front of the square bracket by µ0 .
One works usually in spherical co-ordinate systems (z, θ, ϕ) and the
usual assumption is that the dipole moments (whether electric or mag-
netic) point in the z direction. To find the components of the electric
field we need then the unit vector in the r and θ directions:
ir = ix sin θ cos ϕ + iy sin θ sin ϕ + iz cos θ , (1.80)

iθ = ix cos θ cos ϕ + iy cos θ sin ϕ − iz sin θ . (1.81)
Substituting eqns (1.80) and (1.81) into eqn (1.79) we obtain the electric-
field components in the form
2p cos θ
Er = (1 + j k0 r) e −j k0 r , (1.82)
4πε0 r3
p sin θ
Eθ = (1 + j k0 r − k02 r2 ) e −j k0 r , (1.83)
4πε0 r3
Eϕ = 0, (1.84)
and the corresponding magnetic field is
j ωp sin θ
Hϕ = (1 + j k0 r) e −j k0 r . (1.85)
4πr2
The analogous equations for a magnetic dipole are
m cos θ
Hr = (1 + j k0 r) e −j k0 r , (1.86)
2πµ0 r3
m sin θ
Hθ = (1 + j k0 r − k02 r2 ) e −j k0 r , (1.87)
4πµ0 r3
Hϕ = 0, (1.88)
and the corresponding electric field is

j ωm sin θ
Eϕ = (1 + j k0 r) e −j k0 r . (1.89)
4πr2
1.13 Poynting vector

The derivation of Poynting’s theorem can be found in most textbooks
on electromagnetic theory. It relates the movement of electromagnetic
power to the temporal variation of stored energy. The theorem is usually
presented both in integral and differential form. The essential thing for
our purpose is the complex Poynting vector itself,
1
S= Re (E × H∗ ) , (1.90)
2
where the star denotes complex conjugate. The Poynting vector has
the dimensions of power per unit surface; it gives the magnitude and
direction of power flow. It is more popular in some branches of electricity
than in others. It does not very often appear in metamaterial studies.
Two recent books on metamaterials we have looked at do not even have
entries for the Poynting vector in the Index. In fact, the Poynting vector
played a pivotal role at one of the turning points of materials research.
It was shown by Veselago (1968) that for an isotropic material in which
both the permittivity and permeability are negative the direction of the
Poynting vector is opposite to k, the direction of propagation.
The Poynting vector has, however, other significance too in the the-
ory of metamaterials. Once we are concerned with field phenomena on
a scale small relative to the wavelength then the best guide to under-
standing the physics is the streamlines of the Poynting vector. They will
tell us where the power originates from, how it propagates and where
it is absorbed. Often, it is the only vectorial quantity that can give
us guidance because when the electric and magnetic fields are ellipti-
cally polarized the field lines can no longer be shown in simple graphical
terms.
1.14 Radiation resistance

We have seen in Section 1.11 that the field decay away from the dipole
may be described by 1/r3 , 1/r2 and 1/r terms. The last one, which has
the slowest decay, will be the one that survives a long distance away. It
is the term responsible for the power radiated out. How large is that
1.15 Permittivity and permeability tensors 19
power? For an electric dipole the Poynting vector may be calculated

from eqns (1.82) and (1.83) as
2
1 1 kId
S = Re (Eθ Hϕ∗ ) = η0 sin2 θ ir , (1.91)
2 2 4πr
where the relationship
j ωp
I = j ωq = (1.92)
d
has been used. Equation (1.91) gives the power density due to radiation
at a point a distance r away from the dipole. The total power can be
obtained by integrating the power density over the sphere of radius r.
Performing the integration we obtain
1
P = η0 (kId)2 . (1.93)
12π
We can now introduce the concept of radiation resistance. It is a useful
artifice responsible for radiated power. The radiated power may be re-
garded as that absorbed by the radiation resistance. Hence, its definition
comes from the equation
1
Rs I 2 = P . (1.94)
2
From eqns (1.93) and (1.94) we then obtain
2
2π d
Rs = η0 . (1.95)
3 λ
The radiation resistance of a magnetic dipole may be obtained by similar
arguments as
4
π 2πr0
Rs = η0 . (1.96)
6 λ
1.15 Permittivity and permeability tensors

The relationships B = µH and D = εE imply that the vectors B and
H and, similarly, D and E, point in the same direction because µ and ε
are scalar. However, for certain crystals and for certain metamaterials
the relationship between the components of B and H, and D and E,
may be in the form of a matrix. For the magnetic quantities it takes the
form
    
Bx µxx µxy µxz Hx
 By  = µ0  µyx µyy µyz   Hy  , (1.97)
Bz µzx µzy µzz Hz
which means that if there is a magnetic field directed along a particular
co-ordinate axis it may lead to flux densities that have components along
all three axes. Note that the nine components are not independent
of each other: the matrix is symmetric. When the permeability is in

this form we call it the permeability tensor. Hence, we may write the
relationship between B and H in the form
B = µ0 µr H , (1.98)
where µr denotes now the tensor in eqn (1.97). Analogously, the per-
mittivity may also be a symmetric tensor in both natural materials and
in metamaterials.
1.16 Polarizability
Let us take an element and ask the question whether electric or magnetic
dipole moments can be induced in it by an electric or magnetic field.
Assuming the relationships to be linear we may write them as
p = αe E and m = αm H , (1.99)
where αe and αm are the electric and magnetic polarizabilities. Equa-
tion (1.99) implies that the dipole moments will point in the same direc-
tion as the fields that induce them. We can, however, write more general
relationships if we allow a field in one direction to induce a dipole mo-
ment in another direction. Then, the polarizabilities become tensors and
eqn (1.99) may be rewritten as
p = αe E and m = αm H , (1.100)
where αe and αm are symmetric tensors. However, this is still not general
enough because an electric field might induce a magnetic dipole moment
and, vice versa, a magnetic field might induce an electric dipole moment.
Acknowledging the possibility of such cross-polarization we write the
general relationship as
p = αee E + αem H , (1.101)

me mm
m = α E+α H, (1.102)
where new notations have been introduced. Now the electric and mag-
netic polarizability tensors are denoted by the superscripts ee and mm,
respectively, indicating that the electric field induces an electric di-
pole and the magnetic field induces a magnetic dipole. The cross-
polarizability tensors αem and αme are related to each other (the so-
called Onsager relations) as
αem = −(αme )T , (1.103)

where the superscript T on a matrix means that it is transposed. The
notations are a little clumsy. If, for example, a magnetic field in the z
direction induces an electric dipole moment in the y direction then the
relationship is written as
1.17 Working with tensors 21
py = αem
yz Hz . (1.104)
There are also other notations in the literature but they don’t seem to
be simpler either.
1.17 Working with tensors

The last two sections have been concerned with tensors. They might look
a little intimidating to those who come across them for the first time.
Physically, all they mean is that the properties of certain materials are
different in different directions. Mathematically, it needs a few matrix
multiplications, that’s all. To show that it is not difficult to manage
tensors we shall derive here the wave equation for a case that will be
further treated in Chapter 9.
We shall assume a TE wave propagating in the z direction and find
a solution for the case when both the permittivity and permeability are
diagonal tensors, and the field components are Ey , Hx and Hz . We
need just one simple matrix multiplication and we can then write the
right-hand sides of Maxwell’s equations (1.6) and (1.7) in the form
D = εyy Ey iy and B = µxx Hx ix + µyy Hy iy . (1.105)

The curls on the left-hand sides of eqns (1.6) and (1.7) may be worked
out by the usual methods, giving rise to the equations
∂Ey
= j ωµ0 µxx Hx , (1.106)
∂z
∂Ey
= −j ωµ0 µzz Hz , (1.107)
∂x
∂Hx ∂Hz
− = j ωε0 εyy Ey , (1.108)
∂z ∂x
from which the wave equation can be derived as
∂ 2 Ey ∂ 2 Ey ω2
µzz 2 + µxx 2 + 2 εyy µxx µzz Ey = 0 , (1.109)
∂z ∂x c
which is the same kind of second-order differential equation as the wave
equation (1.21).
1.18 Dispersion: forward and backward

waves
The term dispersion comes from the observation that white light will
be decomposed into its constituents when incident on a prism in which
the velocity depends on frequency. So, when we plot a dispersion curve
it should be, strictly speaking, velocity versus frequency. However, the
accepted presentation is frequency against wave number, often called the
ω–k diagram. The dependence of ω on k is then given by the dispersion

equation.
Have we already come across dispersion equations? Yes, of course.
The simplest one is that of eqn (1.28), ω = kv, where v is the phase
velocity. This gives a linear relationship between frequency and wave
number, provided the velocity is independent of frequency as it would
be in free space. Hence, the ω = kc curve shows no dispersion, the rela-
tionship is called dispersionless. On the other hand, the relationship we
have come across for an ideal plasma in Section 1.9 does show disper-
sion. From eqns (1.22) and (1.66) we obtain, after a moderate amount
of algebra,
ω 2 = k 2 c2 + ωp2 . (1.110)
It follows from this equation that there is no propagation for ω < ωp and
that for large enough ω we again obtain the linear relationship ω = kc.
The phase velocity is the velocity with which a single-frequency wave
travels. But of course a single-frequency wave does not carry any infor-
mation. The velocity of a group of frequencies that do carry information
is the group velocity defined as
refractive index
dω
vg = . (1.111)
dk
A wave with positive group velocity is called a forward wave (phase
and energy move in the same direction) and one with a negative group
velocity is a backward wave. The distinction between forward and back-
ward waves is quite fundamental and particularly important in the the-
frequency ory of metamaterials, as we shall see later in the book.
Under the heading of dispersion it is desirable to mention anomalous
Fig. 1.9 Anomalous dispersion dispersion as well. It is a term from the nineteenth century when ab-
sorption spectra of various materials were studied in the optical region.
The variation of refractive index close to the absorption peak took the
shape shown in Fig. 1.9. There is clearly a region where dn/dω < 0 tak-
ing its smallest value at the resonant frequency. Does it mean that the
corresponding group velocity is negative? Not necessarily. Considering
the definition of the refractive index in Section 1.4 as n = ck/ω we find

dn c k dk
= − + , (1.112)
dω ω ω dω
whence
dω c
= , (1.113)
dk dn
n+ω
dω
i.e. the condition for the group velocity to be negative is not only that
dn/dω must be negative but also the slope of the n versus ω curve must
be sufficiently large.
1.19 Mutual impedance and mutual inductance 23
I1 I1
loop 1
loop 1 loop 2
loop 2
(a) (b)
Fig. 1.10 Magnetic coupling for two loops is positive in the axial case (a) and negative in the planar case (b)
1.19 Mutual impedance and mutual

inductance
The term ‘impedance’ has already been used in Section 1.4 when we
introduced the free-space impedance, defined as the ratio of the electric
to magnetic field in a plane wave. Nevertheless, most people would
regard it a circuit concept, related to inductances, capacitances and
resistances. The mutual impedance belongs to the same category: it
could be claimed both by circuits and fields. Its definition is usually in
terms of voltages and currents, which implies that the claim of circuit
people might be stronger. Mutual impedance between two elements is
defined by the ratio of the voltage in element 2 to the current in element 1
that induced it. It has great significance in the theory of antenna arrays.
Our main interest here is in magnetic coupling and, consequently, in
mutual inductance. The technique of determining mutual inductance is
fairly simple in principle. As an example, we shall go through the steps
for two circular loops shown in Fig. 1.10.
Let us assume that a low-frequency current I1 flows in loop 1 that, true
to circuit theory, is constant everywhere in the wire. The corresponding
vector potential can be found from eqn (1.14) and the magnetic field
from eqn (1.10). Having found the perpendicular magnetic field over
the area of loop 2 we can find the total flux threading the loop. The
mutual inductance between loops 1 and 2 is then defined by the equation
Φ2 = M21 I1 . (1.114)
Note that the mutual inductance can be complex if the distance be-
tween the elements becomes comparable with the wavelength. For most
of the book we shall be concerned with physical situations in which the
mutual inductance is real, but it may still be positive or negative. A sim-
ple illustration in Figs. 1.10(a) and (b) shows the difference between the
two cases. The mutual inductance is positive if the magnetic field lines
Fig. 1.11 RL circuit (a) and the cur-

rent build-up when a step voltage is ap-
plied (b)
cross the two loops in the same direction (we call this the axial config-
uration), and the mutual inductance is negative (planar configuration)
when the magnetic field lines are in the opposite direction.
1.20 Kinetic inductance

Let us first quote the well-known relationship between current and volt-
age due to the presence of an inductance, L, and a resistance, R, in the
circuit shown in Fig. 1.11(a)
dI
V =L + RI , (1.115)
dt
and the solution for the case when a step voltage is imposed upon the
circuit

V0 t L
I= 1 − exp − ; τ= . (1.116)
R τ R
This is shown in Fig. 1.11(b). The current may be seen to be delayed
with a time constant τ .
We may now ask the question: is there anything else in an electrical
circuit that causes delay? Clearly, in order to have a current, charge
carriers must be accelerated, and it takes time to accelerate particles of
finite mass. Hence, the current will necessarily lag behind the voltage
causing its rise. The basic equation is of course the equation of motion
with which we started this chapter. Let us take a piece of conducting
material of length l and cross-section S. The voltage and current may
then be expressed with the aid of the electric field and electron velocity
as
11
It is only a minority of physicists
and engineers, mainly those concerned V = El and I = N eSv , (1.117)
with low-temperature work, who are
familiar with the concept of kinetic which, substituted into eqn (1.1) leads to
inductance. As may be seen from
the derivation in this section, it is a lm dI I
trivial concept, which, actually, found V = + . (1.118)
its way into some undergraduate text- N e2 S dt τ
books, see, e.g., Solymar (1984). Comparing now eqn (1.115) with the above equation we may define a
‘kinetic’ inductance11 and a ‘kinetic’ resistance with the relationships
1.21 Four-poles: impedance and chain matrices 25
lm lm
Lk = and Rk = . (1.119)
N e2 S N e2 Sτ
Looking at the expression of the kinetic resistance it becomes clear
that it is nothing else but the ordinary resistance. On the other hand,
the expression for the kinetic inductance is entirely new. As the cross-
section of the conductor gets smaller the kinetic inductance becomes
comparable with the magnetic inductance. It could even become the
dominant inductance, as for example in circuits containing nanorods.
Since the plasma frequency (defined by eqn (1.67)) will often appear in
this book we shall rewrite the expression for the kinetic inductance in
the form
l
Lk = , (1.120)
πr ε0 ωp2
2
where r is the radius of a conductor of circular cross-section.
1.21 Four-poles: impedance and chain

matrices
Iin Iout
Four-poles are undeniably circuit quantities but they are also suitable
for describing the propagation of waves as will emerge in the next few Vin Vout
sections. We shall mostly follow the excellent and instructive treatment
of Brillouin (1953).
Let us start with the basics. A four-pole is a black box from which
four wires hang out as shown in Fig. 1.12. The input is on the left- Fig. 1.12 A four-pole
hand side and the output is on the right-hand side.We shall use the
eminently logical notations Vin and Iin for the input voltage and current,
and Vout and Iout for the output voltage and current. A four-pole is
characterized by relating two of the above variables to the other two. The
most frequently used form of these relations is in terms of impedances,
as follows
Vin = Z11 Iin − Z12 Iout ,

Vout = Z21 Iin − Z22 Iout , (1.121)
which may be written in terms of vectors and matrices as 12

The principle of reciprocity for a gen-
eral set of circuit elements consisting
V = ZI , (1.122) of many branches can be formulated as
follows: If a voltage generator V , ap-
where plied to the ith branch, drives a current
I in the kth branch, then the same cur-
rent I is obtained in the ith branch if
Vin Z11 −Z12 Iin the voltage is applied to the kth branch.
V= , Z= , I= . (1.123) For a proof see, e.g., Simonyi (1963). A
Vout Z21 −Z22 Iout
consequence of this principle is that for
a four-pole described by eqn (1.121) the
Due to reciprocity12 Z12 = Z21 . Note that in the usual represen- impedances Z12 and Z21 must be equal.
tations Z12 and Z22 have positive signs and the output current points
Iin Iout
Vin 1 2 N Vout
Fig. 1.13 A chain of N four-poles
inwards. They are taken negative in eqn (1.123) to correspond to the

case when the output current points outwards, which is more suitable
for our purpose because it allows the chain matrix representation that
relates the output quantities to the input quantities with the aid of the
chain matrix b:

Vout b11 b12 Vin
= . (1.124)
Iout b21 b22 Iin
The reciprocity condition may be shown with a little algebra to take
the form
b11 b22 − b12 b21 = 1 . (1.125)

The advantage of the representation of eqn (1.124) is that input/output
relations for a chain of four-poles (Fig. 1.13) may be simply written as
Iin Z Iout ! N
!
(N ) (1)
Vout Y
(i) Vin
(N ) = b (1) , (1.126)
Iout Iin
Vin Y Vout l=1
i.e. the resultant chain matrix is the product of the individual chain
matrices.
So far we have been concerned with four-poles characterized by their
Fig. 1.14 Circuit with a series im- matrix representation without specifying what the matrix elements are.
pedance and a shunt admittance To get nearer to a practical case we shall choose a simple circuit as shown
in Fig. 1.14 with a series impedance Z and a shunt admittance Y .
Kirchhoff’s law for the circuit yields the two equations
Iin − Iout
Vin = Iin Z + Vout and Vout = . (1.127)
Y
13
For those in the habit of scrutiniz- A little algebraic manipulation then gives the elements of the b matrix13
ing dimensions it may be disconcerting as
to look at eqn (1.128) because differ-
ent elements of the matrix have differ-
ent dimensions, but that follows from
the definition. It is strange when first b11 = 1, b12 = −Z, b21 = −Y, b22 = 1 + Y Z . (1.128)
meeting them; and it takes time to get
used to.
1.22 Transmission line equations
We shall reproduce Fig. 1.14 in Fig. 1.15 with the difference that now
the circuit represents a dz section of a transmission line and Zu and Yu
are per unit length impedance and admittance, respectively.
Equation (1.127) is still valid, now yielding
1.22 Transmission line equations 27
Zu
I(z) I(z + dz)
V(z) Yu V(z + dz) Fig. 1.15 Circuit with a series im-

pedance and a shunt admittance of a
dz section of a transmission line
V (z + dz) − V (z) = −Zu I(z)dz (1.129)

and
I(z + dz) − I(z) = −Yu V (z + dz)dz , (1.130)

whence one can derive the differential equations
dV dI
= −Zu I and = −Yu V . (1.131)
dz dz
Differentiating either of the above differential equations and substi-
tuting into the other one will yield the wave equation. For the voltage
it takes the form
d2 V
+ Yu Zu V = 0 . (1.132)
dz 2
Comparing the above equation with eqn (1.21) we find that the two
are identical if we take for the propagation constant
k 2 = Yu Zu . (1.133)
Now let us take the values
Zu = j ωLu and Yu = j ωCu , (1.134) 2rw
in which case eqn (1.133) reduces to

d
2 2
k = ω Lu Cu . (1.135)
We may now take as an example a two-wire transmission line as shown
in Fig. 1.16. When d ≫ rw (d is the separation of the lines and rw is
Fig. 1.16 A two-wire transmission line
the radius of the wire) the per unit inductance and capacitance are (see,
e.g., Ramo et al. 1965)

µ0 d πε0
Lu = ln and Cu = . (1.136)
π rw d
ln
rw
Substituting eqn (1.136) into eqn (1.135) we find
ω
k= . (1.137)
c
This means that the velocity of propagation in a two-wire transmission
line is equal to the velocity of light. A wave along a transmission line is
of course not a plane wave but Fig. 1.15 and eqn (1.135) are also capable
of providing the characteristics of a plane wave if we introduce
Lu = µ0 and Cu = ε0 , (1.138)
which would lead again, in a different manner, to eqn (1.137).
1.23 Waves on four-poles

Figures 1.14 and 1.15 showed the same kind of simple four-pole and we
managed to obtain from that the wave equation in the same form as
eqn (1.21). The present section aims at more generality. A four-pole
can have a different structure and still support waves.
A characteristic of a wave is that for the same interval the phase of
the wave always changes by the same amount. Thus, if we have a chain
of four-poles the phase change between the output and input quantities
should be the same factor, exp(−j ka), where a can be regarded as the
physical length of a unit. It follows, hence, that
Vout = e −j ka Vin , (1.139)

and
Iout = e −j ka Iin . (1.140)

Substituting the above quantities into eqn (1.124) we find that a so-
lution exists when
2 cos ka = b11 + b22 . (1.141)

This is now an entirely general dispersion equation. If we know the
chain matrix of a chain of identical four-poles (in fact, we need only
the main diagonal elements) we have immediately the corresponding
dispersion relation.
Let us take as an example a circuit with a series impedance Z and a
shunt admittance, Y . The main diagonal elements of the corresponding
chain matrix are then given by eqn (1.128). Substituting them into
eqn (1.141) we find the dispersion relation
YZ
cos ka = 1 + , (1.142)
2
which may be written in the alternative form
ka
4 sin2 = −Y Z . (1.143)
2
If Z = j ωL and Y = j ωC, then from the above two equations we obtain
1 ka
cos ka = 1 − ω 2 LC , 4 sin2 = ω 2 LC . (1.144)
2 2
Note that the assumptions were practically the same as in the previous
section when discussing the transmission line equations. Nevertheless,
Fig. 1.17 (a) Section of a transmis-

sion line with a series capacitance and
a shunt inductance and (b) the disper-
sion curve of a chain of such sections
eqns (1.135) and (1.144) are different. The reason is that in Fig. 1.15
we already assumed that the line is continuous, whereas our chain of
four-poles implies a set of discrete elements. We can, of course, still
affect the conversion from a discrete to a continuous line by assuming
that ka ≪ 1, in which case eqn (1.144) reduces to
(ka)2 = ω 2 LC , (1.145)
which is identical with eqn (1.135) considering that Lu = L/a and Cu =
C/a.
We have now proven that our last two models may lead to the same
result. We should notice, however, that we have now a wave solution
for any kind of four-poles. For our second example we shall take the
four-pole of Fig. 1.17(a) where
1 1
Z= and Y = . (1.146)
j ωC j ωL
The corresponding dispersion equation is
1
cos ka = 1 − 2 . (1.147)
2ω LC
plotted in Fig. 1.17(b) for ka in the range 0 to π. The phase velocity,
ω/k, is always positive and the group velocity, dω/dk is always negative.
This is our first example of a backward wave.
Our third example is quite a different four-pole, shown in Fig. 1.18(a).
We have separated here the mutual inductance, M , and the self-induc-
tance, L, as it leads to simpler mathematics. Its chain matrix can be
found quite simply in the form
 
0 −j ωM
 
b= 2  ,
 (1.148)
1 L ω
1 − 02

−
−j ωM M ω
√
where ω0 = 1/ LC. The dispersion equation derived from the above
parameters, solving this time for ω is
Fig. 1.18 (a) An equivalent circuit for magnetoinductive waves and (b) its dispersion curves for positive and negative coupling
ω0
ω= r . (1.149)
2M
1+ cos(ka)
L
The corresponding dispersion curves for 2M/L = 0.1 and −0.1 are
shown in Fig. 1.18(b). It may be seen that we have a forward wave for
positive M and a backward wave for negative M . Both waves will often
appear in this book under the name of magnetoinductive waves.
1.24 Scattering coefficients

The scattering coefficients are defined in terms of wave amplitudes. We
are still talking about four-poles, as shown schematically in Fig. 1.12,
but instead of voltages and currents the quantities related to each other
are the amplitudes of the waves pointing inwards, A1 at port 1 and A2
at port 2, and the amplitudes of the waves propagating outwards, B1 at
port 1 and B2 at port 2. The relationship between them is given by the
scattering matrix, S, in the form

B1 S11 S12 A1
= , (1.150)
B2 S21 S22 A2
where, due to reciprocity, S12 = S21 . When there is an incident wave at
port 1 and the output waveguide is matched, so that A2 = 0, then
B1 = S11 A1 and B2 = S21 A1 . (1.151)

It may be seen from eqn (1.151) that under these conditions S11 and
S21 are the reflection and transmission coefficients. It follows then that
for a lossless line
|S11 |2 + |S22 |2 = 1 . (1.152)

Note that in the large majority of measurements, aimed at evaluating
device performance, the S11 and S12 coefficients are measured. These are
the values that those more interested in what is going on will compare
with theoretical results.
1.25 Fourier transform and the transfer

function
Our main interest in this book is in spatial in contrast to temporal
variation which is nearly always taken in the form exp(j ωt). On the
other hand, the spatial information will be of interest in both the actual
space domain and in the spatial frequency domain. The relationship
between the two is given by the Fourier transform. Let us take a complex
two-variable function g(x, y). Its Fourier transform is given by
Z
F (g) = G(fx , fy ) = g(x, y) e −j 2π(fx x + fy y) dxdy . (1.153)
The variables fx and fy are spatial frequencies and G(fx , fy ) is the

amplitude of the fx , fy pair of spatial frequencies. In intuitive terms we
may say that by adding up many sinusoidals of various frequencies and
amplitudes the g(x, y) function can be reproduced. Note that the spatial
frequencies are related simply to the previously defined wave number as
kx = 2πfx and ky = 2πfy .
The problem usually arising is to find the variation in x and y of some
function at the plane z2 when the function is known at the plane z1 . The
space between z1 and z2 is filled by some medium (Fig. 1.19). We can
then investigate what happens to a particular pair of spatial frequencies
of complex amplitude g(fx , fy ) when it traverses the medium between
z1 and z2 . In a given case this may turn out to be a very difficult z = z1 z = z2
problem to solve but it is simple in principle. All that can happen to a
particular pair is that at the exit it will have a different amplitude and x
a different phase. The function that tells us what happens to all the
spatial frequencies is called the transfer function T (fx, fy ). So, how can
we find the spatial distribution at the plane z2 ? z
We may denote the function (say the tangential component of the elec-
tric field) varying in the plane z1 as a function of x and y by g1 (x, y)|z=z1 . Fig. 1.19 An object at z1 imaged at z2
Its spatial harmonics (which together we may call the Fourier spectrum)
are given by its Fourier transform G1 (fx , fy ). When traversing the
medium each one of the Fourier components will undergo some change
corresponding to the transfer function T (fx , fy ). Hence, the Fourier
spectrum at z2 is T (fx, fy )G(fx , fy ). But what we are interested in is
not the Fourier spectrum but the spatial variation of the field. We can
find that by taking the inverse Fourier transform. Hence
Z
g(x, y)|z=z2 = T (fx , fy )G(fx , fy ) e j 2π(fx x + fy y) dfx dfy . (1.154)
If there are a number of media with different properties between z1

and z2 then the transfer function of each one of them must be multiplied
together similarly to chain matrices in the previous section.
As an example, let us find the transfer function of a wave in free space
propagating at an angle θ1 relative to the z axis. As we have seen before,
the phase varies in the z and x directions as exp[−j (kz z + kx x)] where
kz2 + kx2 = k 2 = ω 2 µε . (1.155)

The period along the x axis is determined by kx , it is actually equal to
2π/kx , similarly to the temporal period, which is equal to T = 2π/ω.
Hence, in analogy, we may call kx the spatial frequency. Its amplitude
at z = 0 is equal to 1. When the wave propagates from z = 0 to z = d1
its complex amplitude will be exp(−j kz d1 ) where kz is related to kx by
eqn (1.155). In this example, clearly, only the phase has changed. The
transfer function is
p
T (kx ) = e −j ω 2 µ0 ε0 − kx2 d1 . (1.156)
Our next example is a little more complicated. The space of interest
is now from z = 0 to z = d1 + d2 + d3 as shown in Fig. 1.20. It is free
space from z = 0 to z = d1 and from z = d1 + d2 to z = d1 + d2 + d3 .
From z = d1 to z = d1 + d2 it is a medium with material constants ε, µ.
A plane wave is incident at z = 0 and we detect it at z = d1 + d2 + d3 .
What will be the total transfer function? We have found the transfer
function in free space in the previous example. So, next we must find
it for the slab. How can we determine its transfer function? It is a fair
amount of work but as it happens we have already done all the work
in Section 1.10 but we just did not call it a transfer function. But if
we look at eqn (1.76) and realize that kz1 and kz2 are related to the
spatial frequency kx1 = kx2 = kx then it can be recognized as a transfer
function in disguise. Hence, the total transfer function is
Tt (kx ) = T1 (kx )T2 (kx )T3 (kx ) , (1.157)

and the individual transfer functions are
p
T1 (kx ) = e −j ω 2 µ0 ε0 − kx2 d1 , (1.158)
4ζe
T2 (kx ) = , (1.159)
(1 + ζe ) e j
2 kz2 d2 − (1 − ζe )2 e −j kz2 d2
p
T3 (kx ) = e −j ω 2 µ0 ε0 − kx2 d3 . (1.160)
Equations (1.158)–(1.160) look complicated but they can simplify much

in certain special cases. The possibility of a ‘perfect lens’ does follow
from them as will be discussed in Section 2.12.
We have not so far asked any questions about the range of spatial
frequencies. From the analogy with temporal frequency we may assert
that it can take any value from zero to infinity. Obviously the smaller
Fig. 1.20 Imaging in a three-media

configuration
the details of a spatial function the higher the spatial frequencies that
can reproduce those details. Note, however, that eqn (1.155) must be
satisfied. When the spatial frequency is high enough to satisfy the in-
equality
√
kx > ω µε , (1.161)
then, according to eqn (1.155), kz , the wave number responsible for
propagation in the z direction, becomes imaginary. The wave declines
exponentially, as we saw for surface waves in Section 1.9, but the ex-
pressions derived are still valid.
A bird’s-eye view of
metamaterials
2
2.2 Natural and artificial
The aim of this chapter is to provide an introduction to the subject. It is materials 35
called a bird’s-eye view because it looks at the subject of metamaterials, 2.3 Determination of the
admittedly superficially, from above, pausing at certain views that the effective permittivity/
authors like to share with the reader and making some relevant com- dielectric constant
in a natural material 40
ments without going into very much detail. Physical concepts, most of
2.4 Effective plasma frequency
them already introduced in Chapter 1, will be further developed. The of a wire medium 42
mathematics will be kept to an absolute minimum. It will mostly be
2.5 Resonant elements
simple algebra. for metamaterials 44
Who might benefit from this survey? Well, the authors have certainly 2.6 Loading the transmission
benefited from writing it because they were forced to pronounce judg- line 45
ment on the relative simplicity of the various parts of the subject and 2.7 Polarizability of a
had to decide on where to start, how to proceed and what to include. For current-carrying resonant
a beginner, who first comes into contact with the subject, there might be loop: radiation damping 50
too many new concepts. Even then it is hoped that some concepts will be 2.8 Effective permeability 52
picked up, stored in the brain, and will later act as catalysts facilitating 2.9 Dispersion equation of
magnetoinductive waves
the absorption of further information. For a research student who has derived in terms of
already acquired some familiarity with the fundamentals it might serve dipole interactions 55
as a reinforcement of existing knowledge. For unbelievers who question 2.10 Backward waves
the correctness of all new ideas wherever they come from, until properly and negative refraction 56
checked, this chapter might offer new things to worry about. For those 2.11 Negative-index materials 58
who are thoroughly familiar with all the basic tenets of modern research 2.12 The perfect lens 66
in metamaterials this might still be suitable for bedtime reading. 2.13 Circuits revisited 73
This chapter makes no claim to rigour, which will be sacrificed at
the altar of simplicity. We shall not be concerned with priorities either.
When we just want to get through a large number of different concepts
it is not worth pausing and telling the reader who did what and when
and it is particularly difficult to reconcile rival claims to priority. We
shall therefore quote relatively few references, but that would not, of
course, apply to other chapters.
2.2 Natural and artificial materials

We all know what natural materials are. They are made up by lots and
lots of small elements like atoms and molecules. Some of these materials
36 A bird’s-eye view of metamaterials
a are amorphous, meaning that all those elements are heaped upon each
other in a random manner, others are crystalline, which means that they
arrange themselves into some regular periodical pattern.
Our main interest is in the interplay of waves and materials restricted
to classical physics. The key parameter is a/λ, where a is the distance
between elements in the material and λ is the free-space wavelength.
For simplicity, let us assume that the elements arrange themselves in a
regular cubic lattice, the same in all three directions, as may be seen in
Fig. 2.1. We may now look at two cases: the wavelength is comparable
with a or much larger than a. In the first case the Bragg effect comes
into play. The simplest example is shown schematically in Fig. 2.2. An
electromagnetic wave may be seen to be incident perpendicularly at a
lattice. The wave propagating then from row 1 to row 2 will cover a
Fig. 2.1 Cubic lattice path a. The part of the wave that is reflected by row 2 will have covered
a an additional distance a when arriving back at row 1. When a happens
to be equal to one half of a wavelength then the waves reflected by all
the rows will have the same phase and will reinforce each other. If there
are many rows, and there are indeed many of them in a crystal, then
incident most of the incident power may be reflected. This effect is at the basis
wave of X-ray and electron diffraction in crystals.
When the wavelength is much larger than the lattice period then no
such dramatic effect occurs, but it is nonetheless significant. There may
not be major reflection or diffraction but the electromagnetic wave is still
considerably affected when it enters a material. We may then ignore the
details and pretend that there is no discrete structure: the material is
1 2 3 homogeneous and continuous. The aim is then to find some effective
parameters like electric permittivity and magnetic permeability. This is
Fig. 2.2 Electromagnetic wave incident
known as the effective-medium approximation. Summarizing, there is
normally at a lattice the Bragg effect, when the distance between the elements is comparable
with the wavelength, and there is effective-medium response when that
distance is much less than the wavelength.
Now let’s think of artificial materials in which atoms and molecules
are replaced by macroscopic, man-made, elements. Let’s not worry for
the moment how the elements remain in their allotted space. That may
not be always obvious but we can safely assume that we have complete
freedom in choosing both the elements and the distance between them.
Now, all dimensions are bigger than in natural materials but the division
into the above two categories is still valid. When the separation between
the elements is comparable with the wavelength we have the Bragg effect,
and when the separation is much smaller than the wavelength we can
resort to effective-medium theory. In the former case we talk about
photonic bandgap materials and in the latter case about metamaterials.
Can we have a better definition of metamaterials? Not easily. There
is broad agreement on what the subject is about but not about all the
details. It would need a fairly long description accompanied by a number
of examples to be more precise. We shall give here two definitions in
current use.
1. Metamaterials are engineered composites that exhibit superior prop-
erties not found in nature and not observed in the constituent materials.
2. A metamaterial is an artificial material in which the electromagnetic
properties, as represented by the permittivity and permeability, can be
controlled. It is made up of periodic arrays of metallic resonant elements.
Both the size of the element and the unit cell are small relative to the
wavelength.
Definition 1 is too general, whereas definition 2 is not general enough:
neither of them mention applications and do not even pay homage to
negative refraction. We shall make no attempt here to give a more com-
prehensive definition. Perhaps definition 2 could be made more general
by adding that control, among other things, means that it is possible to
achieve simultaneously negative permittivity and negative permeability
at the same frequency, which will then lead to negative-index media and
to negative refraction. It is not easy to find a definition that would sat-
isfy everybody. For a discussion of the difficulties of a proper definition
and for many other ideas on the subject see Sihvola (2007).
We have talked about natural and artificial materials and their rela- a
tionship to transverse electromagnetic waves. But electromagnetic waves
n-1 n n+1
are not the only ones that should be considered. In a crystal, for ex-
ample, the atoms and molecules may move relative to each other. They
cannot move far away because there are some restoring forces. One of
the manifestations of these motions and of the forces opposing them is
the emergence of acoustic waves. xn-1 xn xn+1
Let us take a one-dimensional chain of atoms in which the elements are
at a distance a from each other at rest. Figure 2.3 shows schematically Fig. 2.3 1D chain of atoms showing
the positions of three of the atoms at xn−1 , xn and xn+1 . Three elements displacement from the quiescent posi-
are usually sufficient when we can get away with an approximation that tion
takes into account only nearest neighbours (see, e.g., Brillouin 1953;
Dekker 1965). Note that all the displacements are in the longitudinal
direction, which makes the problem conceptually simpler. How can we
work out the net force? If xn+1 > xn then there will be a force on the
atom xn wanting to move it to the right. Conversely, if xn > xn−1 then
Za
there will be a force to the left. A proper mathematical formulation
using Newton’s equation (eqn (1.1)) followed by the assumption of a
wave solution will yield the dispersion equation for acoustic waves (Fig.
2.4). It shows the relation between frequency, ω, and wave number,
k. The uppermost frequency ωa at which acoustic waves can propagate
occurs at ka = π or λa = a/2, where λa is the acoustic wavelength. At
frequencies above ωa acoustic waves of the kind we investigated cannot
propagate. The band up to ωa is the pass band and above it, where the 0
0 S
wave cannot propagate, is the stop band. ka
To give another example plasma waves may also propagate in a natural
material. Take sodium for example. It is a metal in which numerous Fig. 2.4 Dispersion curve of acoustic
electrons float in a pool and are compensated by positive ions. In the waves
simplest case all the electrons undergo simple harmonic motion in the
direction of their propagation. If the frequency is high enough, above the
plasma frequency somewhere in the ultraviolet, these electrons can move
quite freely. We should also mention spin waves in magnetic materials
in which the direction of the magnetic moment changes from element to

element.
What we wish to say is that our attention should not be restricted
to the effect of a material upon the propagation of transverse electro-
magnetic waves. There can be waves in a natural material due to the
interaction between the particles, and these waves may not exhibit any
electrical or magnetic phenomena as in our example of acoustic waves.
Now back to metamaterials. Can there be waves on the elements
Fig. 2.5 Capacitively loaded loop that are unrelated to transverse electromagnetic waves? There can be.
In fact, a considerable part of this book (Chapters 7 and 8) will be
concerned with one of these waves that we called magnetoinductive (MI)
waves (Shamonina et al., 2002a). They have already been introduced in
Section 1.23 as the waves propagating on the chain of four-poles shown
in Fig. 1.18. The simplest elements that will propagate this wave are
capacitively loaded metallic loops shown schematically in Fig. 2.5.
The resonant frequency of the elements, which can be simply regarded
as LC circuits, is
n-1 n n+1 1
ω0 = √ , (2.1)
LC
where L is the inductance of the loop and C is the capacitance of the
loading capacitor. Considering again nearest-neighbour interactions1 we
a a
show three such elements separated by a distance a from each other in
Fig. 2.6.
It may be seen that the magnetic field created by element n will also
Fig. 2.6 Magnetic nearest-neighbour thread elements n−1 and n+1. Clearly, there is some magnetic coupling
interactions in a metamaterial chain of between the elements. Can it lead to waves? It can. The mathematical
identical elements formulation may be based on Kirchhoff’s voltage equation. It says that
the total voltage in a closed circuit must be zero. When we consider only
three currents, In−1 , In and In+1 , then the total voltage in circuit n will
have three contributions: (i) the self-voltage, equal to In Z0 (where Z0 is
the self-impedance of the element), (ii) the voltage induced by element
n − 1 that is equal to j ωM In−1 (where M is the mutual inductance
between elements n − 1 and n) and (iii) a contribution from element
n + 1 that is j ωM In+1 . Hence, the relevant equation is
Z0 In + j ωM (In−1 + In+1 ) = 0. (2.2)

Next, assume a wave solution in the form
In = I0 e −j nka . (2.3)
We obtain the dispersion equation for MI waves as
1 This is now the second time in this section that we mention nearest-neighbour interaction. It is not something we would
readily associate with waves. It makes no sense for water waves and even less for electromagnetic waves in vacuum. There are
no neighbours in vacuum. On the other hand, it would not be difficult to develop a physical picture of wave propagation based
on nearest neighbours if we give a little thought to it. Imagine, for example, a large number of houses next to each other in a
street in which nearest neighbours can talk to each other over the fence. An interesting piece of news could certainly reach the
last house in the street by propagating along the row of houses via nearest-neighbour interaction.
Z0
2M
1 L
Z
Z0
Z0
2M
1 L
S/2 S Fig. 2.7 Dispersion of magnetoinduc-
tive waves with positive magnetic cou-
ka pling
ω2

L 1 − 02 + 2M cos(ka) = 0, (2.4)
ω
where we took into account that
1
Z0 = j ωL + . (2.5)
j ωC
Note that eqn (2.4) is identical with eqn (1.149). The corresponding dis-
persion equations have already been plotted in Fig. 1.18. It is replotted
in Fig. 2.7 for 2M/L > 0 for a more pervasive examination.
It shows some similarity to the dispersion curve of acoustic waves
at least in the sense that the group velocity, dω/dk, is always positive
and at the band edge the group velocity is zero, as it is for all waves
on discrete structures. Note also that there is a lower cutoff frequency
below which the MI wave cannot propagate. The pass band is within
the range
ω0 ω0
r <ω< r . (2.6)
2M 2M
1+ 1−
L L
There is no reason of course that the coupling between the metama-
terial elements has to be magnetic. It can be electric. Well before the
advent of metamaterials an experiment, shown in Fig. 2.8(a), was per-
formed by Shefer (1963), using a set of metallic rods (Fig. 2.8(b)). One
of the horns is a transmitter of microwaves, the other horn is a receiver,
and a wave travels along the rods from one horn to the other horn due
to electric coupling. Typical dimensions in the experiments were l = 12
mm, d = 1 mm and a distance between the rods of 5 mm. They found
good transmission between the horn antennas at around the frequency
of 1.2 GHz. A more recent example of wave propagation by rods on a
substrate (Hohenau et al., 2005), at a frequency five orders of magnitude
higher (360 THz), is shown in Fig. 2.8(c), where the element dimensions
are 800 nm × 80 nm × 50 nm. The distance between the elements is
320 nm.
Height (nm)

topography
y (µm)

0
l

í
0 2 4
z (µm)
(a) (b) (c)
Fig. 2.8 Electric coupling on a chain of rods. (a) Schematic of a setup and (b) dimensions of the rod used in the experiment
by Shefer (1963). Copyright c 1963 IEEE. (c) Topography of the setup used by Hohenau et al. (2005). Copyright c 2005
EDP Sciences
2.3 Determination of the effective

permittivity/dielectric constant in a
natural material
The principles are quite simple. They are related to the interdependence
of three vectors: the electric field, E, the electric flux density, D, and
the electric polarization, P. In free space, in SI (Systeme International)
units this can be written as
D = ε0 E + P , (2.7)
as has already been given in eqns (1.9) and (1.19).
The central question is the relationship between E and P in a ma-
terial in which electric dipoles appear in response to an electric field.
Let us consider an element inside a cubic material at the centre of a
rectangular co-ordinate system (0, 0, 0) and apply an electric field, Eext ,
to the material parallel to the z co-ordinate. What will be the electric
field at our chosen element? One’s first thought is that the electric field
will be equal there to the external field. This would indeed be the case
if the electric field would not cause the positive and negative charges to
separate. But the charges do separate. Each dipole will then contribute
to the electric field at (0, 0, 0). We shall call the field there Eloc , the
local field, which is equal to
Eloc = Eext + Edipole . (2.8)

The usual technique for determining Edipole is available in practically
every textbook on solid state physics (see, e.g., Kittel 1953). One takes
Fig. 2.9 A sphere within a cubic lat-
tice used for averaging of macroscopic a sphere (see Fig. 2.9) inside the material that is large microscopically
parameters in the sense that it contains many elements but small macroscopically,
meaning that it is small relative to the electromagnetic wavelength. So
the material is divided into two regions: inside the sphere and outside
the sphere.
Since the radius of the sphere is macroscopically small we can use
the static formulae for the electric field of a dipole, the 1/r3 terms of
2.3 Effective permittivity in a natural material 41
eqns (1.82) and (1.83). When we sum up the contributions at the point
(0, 0, 0) in a cubic lattice from every element inside the sphere it yields
zero electric field as shown in Appendix B. So, after all, the electric
field is equal there to the external field. Is it? Not really, because this
was the contribution of the elements inside the sphere. We still need
to consider the effect of the elements outside the sphere. Now we can
no longer say that the effect will be zero because for elements far away
we also need to take into account the slowly declining radiation field.
It is a tremendously difficult problem. For a recent analysis see Belov
and Simovski (2005a). Here, we shall disregard all those difficulties
and follow the time-honoured method of accounting for the effect of
the elements far away by charges induced on the surface of the sphere.
The electric field at the centre may then be determined by a simple
integration from the charges far away as equal to P/3ε0 , which gives for
the local field
P
Eloc = Eext + . (2.9)
3ε0
But the polarization P at (0, 0, 0) is proportional to the local electric
field there
P = N αe Eloc , (2.10)
where αe is the atomic/molecular polarizability discussed in Section 1.16.
It tells us how effective the local electric field is in producing an electric
polarization. From eqns (2.9) and (2.10) we obtain
N αe
P = Eext . (2.11)
N αe
1−
3ε0
We may now rely on the definition of εeff as
D = εeff Eext = ε0 Eext + P , (2.12)

to find
N αe
εeff = ε0 + . (2.13)
N αe
1−
3ε0
We may also define the relative permittivity (or relative dielectric con-
stant) as
εeff N αe /ε0
εr = =1+ . (2.14)
ε0 N αe
1−
3ε0
Alternatively, we can solve eqn (2.14) for N αe , which may be expressed
as
εeff − ε0 εr − 1
N αe = 3ε0 = 3ε0 . (2.15)
εeff + 2ε0 εr + 2
This is the well-known Clausius–Mossotti equation. The merit of

eqns (2.13)–(2.15) is that they relate the polarizability to the effective
dielectric constant.
Our interest in this section is in the electronic contribution to the
dielectric constant. For natural materials there are other contributions
too, e.g. orientational polarization, but they do not occur for metama-
terials, and our aim is to focus upon such effects that occur both in
natural and in metamaterials.
The above analysis has been done for finding the effective permittivity.
The method is equally valid for finding the effective permeability of a
natural material, and, as we shall see later, of a metamaterial too. The
only difference is then that the electric polarizability needs to be replaced
by the magnetic polarizability.
2.4 Effective plasma frequency of a wire

medium
The plasma frequency of a natural material and its relationship to the
dielectric constant was given in Section 1.9. It was also made clear in
that section that there is no propagation when the dielectric constant
is negative. Plasmas in a solid are well behaved. They are confined
within the material. On the other hand, plasmas in general are difficult
to control. In order to study their properties related to the re-entry
2
One of the interesting questions stud- problem2 plasma simulation was a fashionable topic some time ago. One
ied was the effect of rocket exhaust such simulation was by a lattice of metallic wires as shown in Fig. 2.10(a).
upon the radiation of re-entry vehicle
antennas.
It was shown by Rotman (1962) that they behave as a plasma with a
much reduced plasma frequency.
For determining the plasma frequency we shall use a very simple model
that gives a good approximation. The calculation is based on the rela-
tionship between current and electric field within a cubic unit cell of side
a. Note that a is then the distance between the elements and, also, a is
the length of the rod within the unit cell. Let us first find the current
in a thin piece of wire of length a and radius rw . An incident electric
field of E parallel to the wire will yield a current according to Ohm’s
law equal to
Ea
I= , (2.16)
Zw
where
Zw = Rw + j ωLw (2.17)
is the impedance of the wire. Next, we shall find the average current
density in the unit cell that has an area of a2 . It is
E
Jav = . (2.18)
(Rw + j ωLw )a
2.4 Effective plasma frequency of a wire medium 43
S21 [dB]
Fig. 2.10 (a) Schematic presentation

of a rodded medium. (b) Scattering co-
efficient S21 for samples (rw = 30 µm,
a = 6 mm) with an increasing number
of layers: 5 layers (dot-dashed line), 10
layers (dashed line), 15 layers (dotted
line) and 20 layers (solid line). The ref-
a erence without a sample is also shown
Frequency [GHz] (long-dashed line). From Gay-Balmaz
et al. (2002). Copyright c 2002 Amer-
(a) (b) ican Institute of Physics
Having found the relationship between the electric field and the cur-
rent density we can follow the method outlined in Section 1.9 to find the
effective dielectric constant as
1
εr = 1 + ε0 (Rw + j ωLw )a . (2.19)
jω
Defining now
1
ωp2 = , (2.20)
ε0 aLw
with ωp being an effective plasma frequency,3 we may rewrite eqn (2.19)
as
ωp2 eff
εr = 1 + , (2.21)
jω
ω2 −
τw
where losses are characterized by the time constant
Lw
τw = . (2.22)
Rw
The expressions for the resistance and inductance (see, e.g., Grover 1981)
are
3 Itis always an advantage to look at a phenomenon from different angles. The effective plasma frequency in eqn (2.20) is
written by Pendry et al. (1996) in the form
e2 Neff
ωp2 = .
ε0 meff
It is the same as eqn (1.67) but instead of the actual electron density, N , and actual electron mass, m, the physical quantities
appearing in the above equation are Neff , the average electron density in the unit cell, and meff , the effective mass of the
electron defined so that eqn (2.20) should be satisfied. According to Pendry et al. the reduction of plasma frequency may be
interpreted as due partly to a decrease in the effective electron density and partly to an increase in the effective mass, and that
increase may amount to four orders of magnitude. It is an interesting proposal but there is some inconsistency in it. If the
mass of the electron has increased reducing thereby the plasma frequency then the increased mass must have also reduced the
conductivity. A decrease in conductivity by four orders of magnitude would make the losses enormous and there is no evidence
of that. For a different criticism of the introduction of an effective mass see also Pokrovsky and Efros (2002c).

a µ0 a 2a 3
Rw = 2 and Lw = ln − . (2.23)
πrw σ0 2π rw 4
As an example, let us take a = 6 mm, rw = 0.03 mm, and σ0 = 5.8 ×
107 S/m for copper. The resultant plasma frequency may be calculated
from eqns (2.20) and (2.23) to be 8.73 GHz, well in the microwave region.
For copper at room temperature the time constant is τw = 2.24 × 10−8 s,
which makes the factor ωτw = 1230. For these parameters it is large
enough to ignore losses.
For a detailed experimental study see Gay-Balmaz et al. (2002). With
the above choice of rw = 0.03 mm and a = 6 mm the transmission
4
(S21 ) through N layers (each layer consisted of 39 wires clamped in
As mentioned before, a metamaterial
a groove in a substrate) as a function of frequency is plotted in Fig.
can be regarded a proper material only
if there is a sufficient number of ele- 2.10(b). The transition occurs at about 9.2 GHz, which is in fairly good
ments within a free-space wavelength. agreement with the theoretical value of 8.73 GHz calculated above. The
At 9.2 GHz the wavelength is 3.26 cm long dashed line in Fig. 2.10(b) is the reference level measured when no
so there are about 5 layers per wave-
length. Is that enough? From this set
wires are present. The curves with dot-dash, dashed, dotted and solid
of experiments we could conclude that lines correspond to N = 5, 10, 15 and 20 layers. It may be seen that
yes, five is enough but that is of course 5 layers already cause considerable attenuation but in order to have a
not a general proof. We believe, how- sharp transition 15 layers are needed. Increasing the number of layers
ever, that such a number would suffice
for many practical applications.
to 20 makes hardly any difference.4
2.5 Resonant elements for metamaterials

The wire elements in the previous section are not resonant. They are use-
ful because they can provide negative dielectric constant at frequencies
below the effective plasma frequency, which can be adjusted by choosing
the thickness of the wire and the density of the wire mesh. However,
b most metamaterial elements are resonant and then the problem arises
D Ȝg/2 how to make them small. It is not trivial to satisfy the requirement for
a
the elements to be resonant and at the same time to be small relative
(a) (b) to the wavelength.
When we think of an electromagnetic resonator the one first coming to
Fig. 2.11 (a) Fabry–Perot resonator. mind is probably the Fabry–Perot resonator used in lasers. It consists of
(b) Microwave resonator two parallel mirrors a distance D apart (Fig. 2.11(a)). Resonance occurs
when D is equal to an integral (very large) number of wavelengths. It
is then easy to imagine a wave trapped between the two mirrors just
bouncing back and forth between them. A microwave resonator (Fig.
2.11(b)) with dimensions a, b and λg may be as small as half a guide
wavelength, which still enables the waves to bounce between the metal
walls but it is still far too big. If we want a resonator small relative to the
wavelength that can be easily realized by lumped circuit elements then
all we need is an inductance L and a capacitance C. With a lumped
inductance and a lumped capacitance the size of the resonant circuit
can be very small relative to the wavelength. Those circuit elements,
however, will not do because they cannot easily couple to electric or
magnetic fields. One could, however, use a lumped capacitor that can
be very compact, and for the inductance a loop that can couple to a
magnetic field whether it comes from an incident electromagnetic wave
or from currents flowing in neighbouring elements.
Are there other resonant elements in the same category? There are
plenty of them starting, probably, with the re-entrant cavity used in
klystron amplifiers going back more than half a century (see, e.g., Hansen
1939). We shall present a gallery of small resonators in Chapter 4. For
the time being we shall concentrate on one resonator, a member of the
family of split-ring resonators (SRRs), which has become very popular in
the last decade (Pendry et al., 1999). It consists of two concentric split
rings with gaps on opposite sides. Two realizations with small pipes and
in printed circuit form are shown in Figs. 2.12(a) and (b). The third one
(Fig. 2.12(c)) is the so-called complementary split-ring resonator where
metal replaces air and vice versa (see Baena et al. 2005a).
At first sight the physical phenomena governing the operation of a
SRR are quite complicated. Each ring has a self-inductance, there is Fig. 2.12 Split-ring resonators (a) as
a mutual inductance between them, there is a capacitance between the pipes, (b) in printed circuit form, and
rings and there are gap capacitances at the splits. If one wants to take (c) as a complementary variety
into account all these factors then it is difficult indeed to determine its
properties. It turns out, however, that a simplified physical picture can
lead to an excellent approximation (Marques et al., 2002c). First, ig-
nore the gap capacitances on the basis that they are small and they are
unlikely to have a major influence on the flow of currents. Secondly, ig-
nore the mutual inductance. In the third place, take the self-inductance
equal to the average self-inductance of the two rings. In the fourth place,
consider the two inter-ring capacitances between the splits as being con-
nected in series. We may then put these assumptions into mathematical
form. Take the average radius of the SRR to be equal to r0 , the average
inductance of the two rings equal to L and the inter-ring capacitance per
unit length equal to Cpu . Then, the capacitance of a half-ring is equal
to
Chalf−ring = πr0 Cpu , (2.24)

and the total capacitance of the DSDR is equal to
1 1
C= Chalf−ring = πr0 Cpu , (2.25)
2 2
whence the resonant frequency is
5
In fact, the benefits of loading a trans-
2 mission line have been known well be-
ω0 = p . (2.26) fore Brillouin’s work. A patent filed
πr0 LCpu
by Pupin in 1900 stipulated that the
transmission of telephone signals can
2.6 Loading the transmission line be much improved if the line is peri-
odically loaded by inductances. The
patent was bought by American Tele-
As we have seen in Section 1.21, a four-pole can be regarded as a ba- phone and Telegraph for $ 185,000 +
sic unit of a transmission line and it can even represent a plane wave royalties. By 1911 this technique en-
propagating in free space. We have also discussed the loading of the abled AT&T to extend their telephone
lines from New York to Denver.
transmission line based on Brillouin’s (1953) book.5 Can we use that
approach in our quest to unravel the properties of metamaterials? Well,

we can argue that in the presence of our metamaterial elements the plane
wave is not free to propagate. In other words it can only propagate in a
jerky manner: it bumps regularly into obstacles presented by the meta-
material elements. It propagates a little, then it meets an element by
which it is affected, it propagates further, it meets the next element by
which it is again affected, and so on, and so on. We call this the loading
of the transmission line or of free space. In order to proceed we need to
know two things about the element: what is its circuit equivalent, and
how is it coupled to the plane wave?
2.6.1 By resonant magnetic elements in the form of

LC circuits
Let us assume a plane wave incident perpendicularly upon a cubic lattice
H E of capacitively loaded loops as shown in Fig. 2.13. The loops are in the
k xy plane. The propagation direction of the plane wave is along the x
z y axis. The polarization of the magnetic field is in the z direction so that
it can interact with the loops. The coupling between the magnetic field
x and the resonant loop can be represented by a mutual inductance M ′ .
An approximate value of that can be obtained from the following con-
Fig. 2.13 Cubic lattice of capacitively siderations (Syms et al., 2005b). The voltage induced in the resonant
loaded loops element can be expressed as j ωM ′ It , where It is the current flowing in
the transmission line. The same voltage can be expressed by field quanti-
L C ties as j ωπr02 µ0 Ht , where Ht is the magnetic field of the electromagnetic
wave. Noting further that for the unit cell Ht = It /a we find
πr02 µ0
M’ Lt M′ = . (2.27)
a
Ct The equivalent circuit of the loaded transmission line in terms of four-
poles may be seen in Fig. 2.14.
We can find the dispersion equation from there by the method outlined
previously. First, we need to find the chain matrix of the ‘load’, i.e. that
Fig. 2.14 The four-pole equivalent of a of the LC circuit coupled by M ′ to the transmission line. A little algebra
transmission line loaded by a resonant yields
loop
bL11 = 1 , bL21 = 0 , bL22 = 1 ,

j ωM ′2 1 1
bL12 = , ω02 = . (2.28)
L ω02 LC
1− 2
ω
The elements of the chain matrix for the Lt Ct transmission line are given
by eqn (1.128). We then need to multiply the chain matrix of the load
with the chain matrix of the transmission line, and then take the main
diagonal elements to obtain the dispersion equation from eqn (1.141).
After a fair amount of algebra we find
0.1
RHS of eqn (2.29)
1.4
0
0
Z0 Zq Z/Z 1.2
Zq
1 Z0

0.95 1 1.05 1.1 0 0.05 0.1
Z/Z 0 ka/S
(a) (b)
Fig. 2.15 Loading by rings. (a) Plot of the RHS of eqn (2.29) against ω/ω0 and (b) dispersion curve
2 2
ka ω 2 ω − ωq
4 sin2 = (1 − q 2 ) 2 2 , (2.29)
2 ωt ω − ω02
where
M ′2 1 ω02
q2 = , ωt2 = and ωq2 = . (2.30)
LLt Lt Ct 1 − q2
The left-hand side can vary only between 0 and 4, corresponding to
ka = 0 and ka = π. Hence, there is solution of eqn (2.29) only for those
values of the right-hand side that vary within the same limits. For values
of q 2 = 0.1, and ω0 /ωt = 0.1 we plot the right-hand side as a function
of ω/ω0 showing the range for which a solution exists. It may be seen
(Fig. 2.15(a)) that with good approximation there is solution between 0
and ω0 and again above ωq . There is no solution for ω between ω0 and
ωq . The corresponding dispersion curve is shown in Fig. 2.15(b). We
can say that there are pass bands between 0 and ω0 and above ωq up to
a frequency comparable with ωt , and there is a stop band between ω0
and ωq .
If we want to regard the medium made up by these resonant magnetic
elements as a continuous one then we can replace sin(ka/2) by ka/2.
Equation (2.29) then takes the form
2 2
ω 2 ω − ωq
k 2 = (1 − q 2 ) . (2.31)
c2 ω 2 − ω02
The main, and obvious, change is that in the continuous limit the sepa-
ration of the elements, a, no longer appears in the dispersion equation.
It is also clear now that the changes from pass band to stop band and
vice versa occur exactly at the frequencies ω0 and ωq . It may also be
seen that the frequency range extends to infinity.
Lt 1.0
Z/Zp
Ct Lw
0.7
0 0.05 0.1
ka/S
(a) (b)
Fig. 2.16 Loading by rods. (a) Four-pole equivalent and (b) dispersion curve
2.6.2 By metallic rods

A thin, short metallic rod of radius r and length l can be characterized
by a self-inductance Lw . Its value has already been given in eqn (2.23).
Such a rod may interact with the electric field of an input electromag-
netic wave. It will shunt it. Hence, the ‘load’ is here an inductance
across the transmission line. The loaded transmission line is shown in
Fig. 2.16(a). We may now follow the method of the previous subsection
for finding the dispersion equation by multiplying the chain matrices of
load and transmission line but it is much simpler just to look at the
four-pole of Fig. 2.16(a) and find the elements of the chain matrix from
there. The general relationship was already given by eqn (1.128). All
we need to notice is that Y is now equal to the parallel combination of
the capacitance Ct and the inductance Lw hence
!
1 ωp2 1
Y = j ωCt + = j ωCt 1− ; ωp = p . (2.32)
j ωLw ω2 Lw Ct
With Z = j ωLt remaining unchanged we can find the main diagonal

elements of this chain matrix, and from that the dispersion equation in
the form
ka ω 2 − ωp2
4 sin2 = . (2.33)
2 ωt2
Clearly there is no q
solution when ω < ωp but a solution exists up to
the frequency, ω = 4ωt2 + ωp2 . The corresponding dispersion curve is
shown in Fig. 2.16(b).
1.4
L C
Zp
Z/Z0
1.2
M’ Lt
Zq
Ct Lw 1 Z0
0 0.05 0.1
ka/S
(a) (b)
Fig. 2.17 Loading by resonant loops and rods. (a) Four-pole equivalent. (b) Dispersion curve
2.6.3 By a combination of resonant magnetic

elements and metallic rods
We shall now combine into one circuit both the magnetic element and
the metallic rod as shown in Fig. 2.17(a). The technique for finding
the elements of the chain matrix is still the same. The chain matrix
of the magnetic load should be multiplied with the chain matrix of the
rod-loaded transmission line. The resultant equation is
ka ω 2 − ωp2 ω 2 − ωq2
4 sin2 = (1 − q 2 ) . (2.34)
2 ωt2 ω 2 − ω02
The above equation is very similar to eqn (2.29). The only difference
is that in the first term ω 2 should be replaced by ω 2 − ωp2 . What kind
of dispersion characteristics could we expect from eqn (2.34)? It will
depend on the relative values of the three characteristic frequencies, ωp ,
ω0 , and ωq . We know that ωq is larger than ω0 but ωp could be anywhere
depending on the parameters of the rods chosen. Let us choose it to be
larger than ωq . Then, for ω < ω0 we have a stop band, for ω0 < ω < ωq
we have a pass band that happens to be a backward-wave region, for
ωq < ω < ωp we have a stop band, and finally for ω > ωp we have again
a pass band. The dispersion curve for ωp = 1.2 ωq is plotted in Fig.
2.17(b).
The appearance of pass and stop bands is summarized in the diagram
of Fig. 2.18 for the three cases: Rods alone, resonant magnetic elements
alone, rods and magnetic elements combined. Let us just concentrate
our attention on the frequency region between ω0 and ωq in the three
cases. In case 1 that frequency region is a stop band. In case 2 it is
again a stop band, but in case 3 it is a pass band. This is a significant
conclusion. For rods alone or for magnetic elements alone it is a stop
band, but when the two are combined the stop band turns into a pass
band. The rules are quite easy and quite interesting. Two pass bands
Fig. 2.18 Pass and stop bands for rods,

rings and a combination of both. From
Syms et al. (2005a). Copyright c 2005
American Institute of Physics
make a pass band, a pass band and a stop band make a stop band,
and two stop bands make a pass band. According to these rules we can
assign +1 to a pass band and −1 to a stop band. Then 1 × 1 = 1,
1 × (−1) = −1, (−1) × 1 = −1 and (−1) × (−1) = 1. These are in
fact the same rules that will arise later when discussing negative-index
materials.
2.7 Polarizability of a current-carrying

resonant loop: radiation damping
z H
y We shall now find the magnetic polarizability in the simple case of a
loop. More precisely we shall look for the relationship between the z
x component of a spatially constant magnetic field and the induced mag-
netic moment when the loop is in the xy plane (Fig. 2.19).
It is a simple problem because the magnetic field does not change
Fig. 2.19 Resonant loop in a magnetic over the area of the loop and we also know from the geometry that the
field induced magnetic dipole moment will have a z component only. We can
find it by first determining the flux threading the loop, equal to µ0 SH,
then the voltage excited in the loop as being −j ωµ0 SH, whence the
current is
j ωµ0 SH
I=− , (2.35)
Z
where Z is the impedance of the resonant element,
1
Z = j ωL + +R, (2.36)
j ωC
and losses in the form of a resistance have also been taken into account.
The induced magnetic moment is then (see eqn (1.18))
2.7 Polarizability of a current-carrying resonant loop: radiation damping 51
j ωµ20 S 2 H
m = µ0 SI = − , (2.37)
Z
whence according to the definition of eqn (1.100)
j ωµ20 S 2
αm = − . (2.38)
Z
If we want to think in more general terms then we can regard the above
polarizability as the αmm
zz component of the polarizability tensor relating
the z component of the magnetic dipole moment to the magnetic field
applied in the z direction.
Note that in the absence of losses αm is real. It becomes complex when
R is added to the impedance. However, this is not the only source of loss.
Power can turn not only into heat but it can also be lost by radiation.
In Section 1.14 we asked the question whether we can take into account
the radiated power by assigning to the element a resistance that we
called the radiation resistance. The derivation was done for an electric
dipole but the value of the radiation resistance for a magnetic dipole was
also quoted (see eqn (1.96)). A small loop is of course equivalent to a
magnetic dipole, hence we can take care of the radiation loss by adding
4
π 2πr0
Rs = η0 (2.39)
6 λ
to the ohmic resistance. For most metamaterials we do not need to
consider the radiation resistance because all dimensions are small relative
to the wavelength. But when the length of the line is larger than the
free-space wavelength, as will be the case in Section 8.2, then radiation
effects must be taken into account, which we can do by adding the
radiation resistance to the ohmic resistance. Under these conditions it
is preferable to work in terms of the inverse of magnetic polarizability,
which will then take the form

1 1 1 1
= + + , (2.40)
αm αm lossless αm ohmic loss αm radiation loss
where
ω02

1 fr L
= − ; fr = 1 − , (2.41)
αm lossless µ20 S 2 ω2

1 L ωL
= −j ; Q= , (2.42)
αm ohmic loss Qµ20 S 2 R
where Q is the quality factor. For the third term a little algebra will
yield
k3

1
=j 0 . (2.43)
αm radiation loss 6 πµ0
The above expression is called radiation damping by physicists. It is the

physicists’ measure of the radiation due to a magnetic dipole. Antenna
engineers don’t like it and don’t use it. It is one of the examples when
physicists and engineers look differently at the same phenomenon.
For completeness, we shall give below the expression for the radiation
damping associated with an electric dipole, which can be determined in
an analogous manner,
k3

1
=j 0 . (2.44)
αe radiation loss 6πε0
2.8 Effective permeability

We have already found effective permittivities in Section 2.3. We shall
now make an attempt to find the effective permeability in some simple
cases. In our first attempt we make the assumption that the local field
is equal to the externally applied field, i.e. we completely disregard the
effect of all the other elements. We assume now a cubic lattice of loops as
shown in Fig. 2.13. The magnetic field of the incident plane wave is in the
z direction and the loops are in the xy plane. Then, the magnetization
due to the effect of the incident field upon the elements is
Mm = N m = N αm H . (2.45)
We shall now find the effective relative permittivity of the medium per-
pendicular to the plane of the element (according to the notations of
Section 1.16 this is the µzz component of the permeability tensor) from
the definition
B µ0 H + Mm Mm
µr = = =1+ . (2.46)
µ0 H µ0 H µ0 H
With the aid of eqn (2.45), and the definition of magnetic polarizability,
eqn (2.38), we find
µ N S2
µr = 1 − 0 (2.47)
j
L fr −
Q
or
F
µr = 1 − , (2.48)
j
fr −
Q
which is often regarded as the standard form in the literature. F is
defined as
µ0 N S 2
F = . (2.49)
L
Another form of eqn (2.47) for the lossless case may be obtained with a
little algebra as
2.8 Effective permeability 53
ω 2 − ωF2
µr = (1 − F ) , (2.50)
ω 2 − ω02
√
where ωF = ω0 / 1 − F . It may be seen from eqn (2.50) that in the
absence of losses there is a pole at ω0 and a zero at ωF . The variation
of µr with frequency is plotted schematically in Fig. 2.20.
The interesting thing is that between the pole and the zero the per-
meability is negative. How wide is the range of negative permeability?
It can be easily calculated when F is small. Then,
Pr
1 F
∆ωneg.perm. = ω0 √ − 1 ≃ ω0 . (2.51)
1−F 2
In our second model we shall take into account that all the other elements
will also contribute to the flux at element n. The total flux is then due 1
to the applied field plus the flux provided by all the other elements,
Z0 ZF Z
X
Φ = µ0 SH + I Mnn′ , (2.52)
where Mnn′ is the mutual inductance between element n and n′ and the
current I is assumed to be the same in all the elements. The correspond-
ing current must then satisfy the equation Fig. 2.20 Frequency variation of per-
meability
−j ω X
I= µ0 SH + I Mnn′ , (2.53)
Z
whence we may determine the current, from the current the magnetic
dipole density, and from that the effective permeability. We obtain fi-
nally
F
µr = 1 − , (2.54)
j
fr + ∆fr −
Q
where
1X
∆fr = Mnn′ . (2.55)
L
The difference between our first and second model is that a new term,
∆fr , enters the denominator that involves all the mutual inductances.
How can we find that term for a cubic lattice? We have met this problem
in a somewhat different form in Section 2.3. There, we were concerned
with the total electric field at an element due to all other elements. That
sum was shown to be zero in Appendix B. The same applies here. If
the total magnetic field is zero then the mutual inductance must also be
zero. For other lattice configurations the additional term represents a
small shift in the position of the pole where the effective permeability
tends to infinity.
Our third model for finding the effective permeability is based on
eqn (2.31) that gives the dispersion equation for loading by magnetic
elements in the continuous limit. Clearly, the dispersion equation and
the effective permeability are closely connected to each other. For a
plane wave propagating in a medium with a relative permeability µr the

relationship between ω and k (see Section 1.4) is
√
ω µr
k= . (2.56)
c
The corresponding relationship in the continuous approximation be-
tween ω and k for a magnetic element yields
ω 2 − ωq2
µr = (1 − q 2 ) , (2.57)
ω 2 − ω02
which is of the same form as eqn (2.50) but q 2 appears instead of F .
The two expressions would be exactly the same if F were to agree with
q 2 . In fact, if we look at eqns (2.27), (2.30) and (2.49) we shall find that
they do agree. Hence, our third model gives the same result as the first
one.
Our fourth model is the oldest of them all, the one that has been
applied to finding the material constants for natural materials for well
over a century, the Clausius–Mossotti model. We derived it in Section 2.3
for the effective permittivity, but the same equation is valid of course
in the magnetic case. We only need to substitute magnetic polarization
for electric polarization in eqn (2.14). We find for the lossless case

2F
ω2 1 − − ω02
3
µr = . (2.58)
2 F 2
ω 1+ − ω0
3
It may be seen that the shape of the µr curve has hardly changed but
the region of negative permeability has shifted p towards lower frequen-
cies.√The pole hasp moved from ω0 to ω0 / 1 + F/3 and the zero from
ω0 / 1 − F to ω0 / 1 + 2F/3. It follows from the above relations that
for small values of F the width of the negative permeability region has
not changed. It is ω0 F/2.
The fifth model is that of Gorkunov et al. (2002) who essentially repeat
the derivation of the Clausius–Mossotti equation but include the effect
of mutual inductances. Then, as we may guess from our second model
the additional term ∆fr appears. To be exact we need to add ∆fr both
to 2F/3 and to F/3 in eqn (2.58).
Losses. Although losses were taken into account in deriving eqn (2.48)
we disregarded them later for simplicity. In the general case when µr is
complex we use the notation
µr = µ′ − j µ′′ . (2.59)
This is plotted in Fig. 2.21 in the vicinity of the resonant frequency for
Q = 100, 1000 and 10 000 from eqn (2.48) with F = 0.1. As may be
expected, the real part of the permeability no longer tends to infinity
and the maximum µ′ achievable is reduced to 2, 3.6 and 6.1 for the three
values of the quality factor. The imaginary part of µr can be quite large.
Its maximum may be larger than that of the real part.
2.9 Dispersion equation of magnetoinductive waves derived in terms of dipole interactions 55
8 Re(Pr ) 8 Re(Pr ) 8 Re(Pr )

Im(Pr ) Im(P ) Im(P )
Re(Pr ), Im(P r)
r r

0 0 0

0.8 1 1.2 0.8 1 1.2 0.8 1 1.2
ZZ 0 ZZ 0 ZZ 0
(a) (b) (c)
Fig. 2.21 Frequency variation of the real and imaginary parts of permeability for (a) Q = 100, (b) Q = 1000, (c) Q = 10 000
2.9 Dispersion equation of

magnetoinductive waves derived in
terms of dipole interactions
We shall now look again at the linear array of loops shown schematically
in Fig. 2.6, but assume this time that there is an infinite number of
elements, and derive once more the dispersion equation in a different
manner. It is more general in the sense that the interactions are not
limited to nearest neighbours but less general in another sense that our
loops will be replaced by magnetic dipoles, true only if the loops are
small enough and two loops are not too close to each other. This is
a technique used by theoretical physicists who like to start with simple
concepts so that they can later add to them lots of complications of their
own liking.
From the definition of the magnetic moment and magnetic polariz-
ability, as applied to element 0, we may write
m0 = αm H , (2.60)
where H comes from all the other dipoles. The element positioned at a
distance na away for n > 0 will give a contribution
(Hn )r = mn f (n) , (2.61)

where
1 + j k0 na −j k0 na
e
f (n) = . (2.62)
2πµ0 n3 a3
Hence, the contribution of all the elements from n = 1 to ∞ will amount
to
X
Htotal = H(n > 0) + H(n < 0) = (mn + m−n ) f (n) . (2.63)
Next comes the wave assumption
mn = m0 e −j nka , (2.64)
which, substituted into eqn (2.63), leads to the equation
(1 + j k0 na) e −jk0 na
∞
1 1 X
= 3 cos(kna) . (2.65)
αm πµ0 a n=1 n3
The right-hand side of eqn (2.65) is known as the interaction function.
We shall denote it by IF . Hence, the general form of the dispersion
equation is
αm IF = 1 . (2.66)
In what sense is eqn (2.66) a dispersion equation? In the same sense
as the previous dispersion equations. IF depends on ω and k and the
polarizability depends on ω, hence eqn (2.66) relates ω and k to each
other.
This is now the third time that we derive a dispersion equation for a
set of resonant loops. The derivation in Section 1.23 was based purely
on circuit concepts, whereas the one in Section 2.2 was based on nearest-
neighbour coupling and Kirchhoff’s law. It was also assumed there that
all dimensions are small relative to the free-space wavelength. Hence, if
we want eqn (2.66) to reduce to eqn (2.4) we have to take the lossless
case, assume that k0 a ≪ 1 and secondly that the infinite sum should be
reduced to n = 1. We then obtain
ω2 πr4

L 1 − 02 + µ0 30 cos ka = 0 . (2.67)
ω a
Now we are nearly there. Remember that in the derivation used in
6
the present section we assumed magnetic dipoles instead of loops and
For nearest neighbours and small
we found the magnetic field due to a loop in the dipole approximation.
wavelength the radial magnetic field
takes the form Thus, if we further note that in this approximation the mutual induc-
m
tance between two loops a distance a apart in the axial configuration
H=
2πµ0 a3
. (2.68) is6
Considering further that m = µ0 πr02 I,
that the flux Φ is µ0 Hπr02 and the mu-
µ0 πr04
M= , (2.69)
tual inductance is defined as M = Φ/I 2a3
we can find M in the form of eqn (2.69).
then eqns (2.67) and (2.4) may be seen to be identical.
2.10 Backward waves and negative

refraction
Backward waves, as mentioned before, have phase and group velocities
in opposite directions. Does it make a difference whether the waves
are forward or backward when it comes to refraction at a boundary?
Let’s start with two isotropic media 1 and 2, both of which can support
2.10 Backward waves and negative refraction 57
vg
vp
Q P R
B’
(a) (b)
vg2
vg2 Fig. 2.22 Refraction problem. (a) In

vg1 vg1 a backward wave phase and group ve-
locities are opposed to each other. (b)
Ewald circle for two media with k2 >
k1 . (c) Refraction of a backward wave
into a backward wave. (d) Refraction of
(c) (d) a backward wave into a forward wave
backward waves at a particular frequency, and assume that the wave

number k2 is larger than k1 . If wave 1 is incident upon the boundary
then its group velocity must be in the first quarter, and, consequently,
its phase velocity will be in the third quarter, as shown in Fig. 2.22(a).
Next, we draw the Ewald circle with radii k1 and k2 in Fig. 2.22(b) in
the same way as in Section 1.6. The wave vector of the input wave is
k1 = OP. If we want to satisfy the boundary condition that the phase
velocities along the boundaries (denoted by BB′ in Fig. 2.22(b)) match
then the possible wave vectors are either OQ or OR. However, the group
velocity of the wave in medium 2 must point away from the boundary,
i.e. it must be in the first quarter. Hence, the only solution is k2 = OQ.
If we look at the direction of the group velocities at the boundary (Fig.
2.22(c)) then we can see that nothing interesting has happened. The
refraction of backward waves is of the same kind as the refraction of
forward waves.
What about media that support different types of waves, say the wave
in medium 1 is a backward wave but the wave in medium 2 is a forward
wave. If medium 1 is the backward-wave medium then the Ewald circle
construction of Fig. 2.22(b) is still valid: OP is still the wave vector in
medium 1 and OQ and OR are still the two possibilities for k2 that satisfy
the boundary condition. The difference is that the wave in medium 2
is a forward wave therefore the group velocity is in the same direction
as OR and that is now the only solution. The group velocities for this
case are shown in Fig. 2.22(d). Now there is something interesting! In

contrast to the usual case, taught in all science courses, the refraction is
in the negative direction. In current terminology we talk about negative
refraction.
The possibility of negative refraction has been known for a very long
time (see the historical review in Chapter 10) but it became a subject
of intense study only quite recently after Smith et al. (2000) discovered
Veselago’s paper on negative refractive index.
2.11 Negative-index materials

2.11.1 Do they exist?
The possibility of negative-index materials was broached by Veselago in
a paper written in Russian in 1967 and published in English in 1968. It
lay dormant for many years, until Smith et al. (2000) discovered it.
We have seen (eqn (1.30)) earlier the expression for the refractive
index in terms of the relative permittivity and permeability,
√
n= εr µr . (2.70)
When both µr and εr are positive then there is no problem, everything
is familiar apart from the cry of those experienced in optical phenomena
who will very likely ask what µr is doing in that equation. That objection
can be easily overcome by asserting that the relative permeability may
indeed be different from unity for metamaterials. What happens when
only ε is negative? That is well known. That happens in plasmas, we
talked about this in Sections 1.9 and 2.4. But Veselago asked a more
daring question: What happens when both µr and εr are negative? He
discusses the possible responses as quoted below:
‘The situation can be interpreted in various ways. First, we may
admit that the properties of a substance are actually not affected by a
simultaneous change of the signs of ε and µ. Second, it might be that
for ε and µ to be simultaneously negative contradicts some fundamental
law of nature, and therefore no substance with ε < 0 and µ < 0 can
exist. Finally, it could be admitted that substances with negative ε and
µ have some properties different from those of substances with positive
ε and µ.’
Veselago then puts forward the third explanation. He goes on to show
the consequences of negative material constants straight from Maxwell’s
equations. Assuming a plane wave propagating in a medium with mate-
rial constants ε and µ in the form exp(−j k · r) eqns (1.6) and (1.7) take
the form
k × H = ωεE and k × E = −ωµH . (2.71)

It may be seen from the above equations that it makes a difference
whether the material constants are both positive or both negative. In
the former case the vectors E, H and k constitute a right-handed set,
whereas for negative ε and µ we have a left-handed set. The wave vector
k tells us the direction of the phase velocity, the Poynting vector tells us
the direction of the group velocity. If the two are in opposite directions
we have a backward-wave material with all that implies. Thus, negative
refraction at the boundary of two materials, one having positive material n2 = 0.17
constants and the other negative ones, follows immediately. But there n2 = 0.2
is an alternative explanation. We may argue that the square root in n2 = 0.3
eqn (2.70) can be positive or negative. It is sensible to take it positive n2 = 1
when the material constants are both positive and take it negative when
both material constants are negative. But that will have an influence on n2 = -1
ș1=10°
Snell’s law (eqn (1.46)), n2 = -0.3
n2 = -0.2
n1 sin θ1 = n2 sin θ2 . (2.72) n2 = -0.17
Let us take now medium 1 as free space, n1 = 1, and see the direc-
tion of the refracted wave (Fig. 2.23) when medium 2 has a large range
Fig. 2.23 Refraction at a boundary for
of refraction indices, which may be smaller than 1 and may take any various values of n2
negative value. The angle of refraction is 90◦ when n2 = sin θ1 (if n2 is
even smaller then total internal reflection occurs in medium 1). As n2
increases from this value below unity up to infinity the refracted angle ș1 ș1
declines from 90◦ to 0◦ . Note that the angle of refraction is the same
for n2 = −∞ as for n2 = ∞. Now, as n increases from minus infin-
ity to − sin θ1 the angle of refraction declines from 0◦ to −90◦ . If n2
is between − sin θ1 and 0 then there is again total internal reflection.
ș1 ș1
Clearly, negative n2 implies negative refraction. It may be worth reiter-
ating at this stage that we got negative refraction in two different ways.
In Section 2.10 from the properties of backward waves and in the present z
section from the concept of negative refractive index.
The most striking example of what we can do with a negative-index 0 d/2 3d/2 2d
material is Veselago’s flat lens. For n = −1 the angle of refraction is
equal to the negative angle of incidence hence all the rays emanating
Fig. 2.24 Negative refraction for n =
from a line source will be refocused inside the material and brought −1 (Veselago’s flat lens)
to another focus outside the material as shown in Fig. 2.24. The lens
thickness is d. A point source at a distance d/2 in front of the lens is
ȝr
reproduced at a distance d/2 behind the lens.
ENG DPS
2.11.2 Terminology
İr
Having realized that negative material constants may be interpreted
as having negative refractive index, and that the consequence is a left- DNG MNG
handed E, H, k relationship, Veselago called these materials left-handed.
In the many papers that followed Smith et al.’s (2000) rediscovery quite
a number accepted this terminology and referred to these materials as
Fig. 2.25 Terminology for materials
left-handed media (LHM) and to those with positive material constants with various signs of permittivity and
as right-handed media or (RHM). There is, however, a vocal minority permeability after Engheta et al. (2005)
unhappy with this description. They argue that left-handedness and
right-handedness had been terms widely used before in chiral materials
referring to the direction of chirality.
A different terminology related to signs of permittivity and perme-

ability was proposed by Engheta et al. (2005) and may be seen in Fig.
2.25. Materials with both constants positive are double-positive (DPS)
materials, with only ε negative they are ENG materials, with only ε
positive they are MNG materials and with both of them negative they
are double-negative (DNG) materials. Also, ENG and MNG materials
are sometimes referred to as SNG or single-negative materials.
Other terms used are NRI for negative refractive index media, NIM
for negative-index media. The one we prefer is BW (Lindell et al., 2001)
meaning backward-wave media, which is a generic term. All negative-
index media are backward-wave media, but there are plenty of backward-
wave media that, mainly because they are one-dimensional structures,
do not qualify as negative-index media.
A 2.11.3 Negative-index lenses

B
The possibility of negative-index materials can lead to the design of new
r1 families of lenses. The design procedure is the same as in Section 1.8
and the same equation applies. We shall reproduce Fig. 1.6(a) in Fig.
T 2.26 and eqn (1.62) in the equation below in normalized Cartesian co-
O d1 C G ordinates:
2
y2

n 1
x− − 2 = . (2.73)
n+1 n −1 (n + 1)2
Let us now see what the contours are if n can take any value from
n = −∞ to ∞. When n is very large the hyperbola is close to the
Fig. 2.26 Ray picture for a lens straight line x = d (see Fig. 2.27(a) for n = 10). For n = 1.5 the
hyperbola is shown in Fig. 2.27(b). As n → 1 the hyperbola exists
for smaller and smaller angular regions (see Fig. 2.27(c) for n = 1.05)
degenerating eventually into the y = 0 line. What happens when n
becomes less than unity? It means that the phase velocity of the wave is
larger than the velocity of light. That is fine, that does not violate any
physical principle. In fact, such a lens was realized in the 1960s by Kock
(1964) using hollow metallic waveguides in which the phase velocity can
be arbitrarily high. But let us continue our quest. As n becomes less
than unity there is no longer an asymptote, the hyperbola turns into
an ellipse, as may be seen in Fig. 2.27(d) for n = 0.95, the contour still
being quite close to the y = 0 line. As n declines further towards zero
the curvature of the ellipse declines and the contour tends towards a
circle (see Fig. 2.27(e) for n = 0.01). As n becomes negative nothing
radical happens. The contour for n = −0.01 is hardly different from
that at n = 0.01, only the ellipse becomes elongated in the y direction.
At n = −1 the contour switches back to a hyperbola (see Fig. 2.27(f))
but the asymptote occurs at θ > π/2, so it is irrelevant for the shape of
the contour. As n decreases further towards −∞ the hyperbola becomes
flatter (see Fig. 2.27(g) for n = −10) finally reaching the x = 1 line, this
time from the left. As we have seen in Fig. 2.23 it makes no difference
whether the index of refraction is positive infinity or negative infinity.
(a) (b) (c)
Fig. 2.27 Contours of negative-index

(d) (e) (f) (g) lenses. n = 10, 1.5, 1.05, 0.95, 0.01,
−1, −10(a)–(g)
With the advent of negative-index materials new possibilities were

born for the design of lenses. In fact, such lenses may have superior
properties (smaller aberrations) to positive-index lenses, as pointed out
by Schurig and Smith (2004).
2.11.4 The flat-lens family

We have seen in the previous section that the contour of a lens may tend
towards a straight line as the index of refraction tends to infinity. In the
limit that would be a flat lens but it would be of little use. If nothing else
high refractive index leads to high reflection, which makes it unsuitable
for practical applications. However, a flat lens working on somewhat
different principles was proposed by Veselago (1968). He showed that
for a particular geometry a flat lens with n = −1 will bring a diverging
beam to a converging beam, as shown in Fig. 2.24. The new feature was
the presence of an internal focus. Note that the optical path from the
external focus to the internal focus is zero. r1
We may now ask the question whether this lens is unique, whether it r2
T
is the only example of a lens with an internal focus? There is no reason
O1 O2 O3
in principle why one should not be able to realize an internal focus with
d1 d2
a positive refractive index but the lens would be very thick. If we want
a reasonably thin lens we need to design another set of lenses that this
time will have an internal focal point.
The variation of the required contour with angle can be found with
the aid of the sketch in Fig. 2.28 where both contours are shown. Due to Fig. 2.28 Ray picture of a lens having
symmetry, it is sufficient to look at the problem between the points O1 an internal focus
and O2 representing the two foci. The equality of optical paths demands
that
r1 + nr2 = d1 + nd2 . (2.74)
A second relationship between r1 and r2 follows from the geometry

(a) (b) (c) (d) (e)
Fig. 2.29 Contours of lenses with internal foci. n = −0.2, −0.8, −1.2, −2, −4 (a)–(e)
r22 = r12 + (d1 + d2 )2 − 2r1 (d1 + d2 ) cos θ . (2.75)

If we want to have the same type of lens as Veselago’s flat lens then we
should assume that the optical path from focus to focus is zero, i.e.
d1 + nd2 = 0 . (2.76)
Then, the equation in polar co-ordinates reduces to
2
r1 r1
(n + 1) − 2n cos θ + (n − 1) = 0 . (2.77)
d1 d1
It may again be converted into normalized Cartesian co-ordinates yield-
ing the equation
2
n 1
x− + y2 = , (2.78)
n+1 (n + 1)2
which is clearly that of a circle centred at (n/(n + 1), 0). In the limit
of n → −1 it becomes a straight line, i.e. we obtain the flat lens as a
special case.
Next, let us find the contours for various values of n. Clearly, the
condition of zero optical path posed by eqn (2.76) can only be satisfied
if n < 0. Thus, the range of interest is for n between zero and −∞.
When n is small and negative the radius of the circle is close to unity,
but the internal focus is quite far away from the front surface because
negative index material
d2 = −d1 /n. The contour and the focusing mechanism are shown in
Fig. 2.29(a) for n = −0.2. The internal focus is there but the lens is
far too thick. As n tends towards −1 the lens flattens, as may be seen
in Fig. 2.29(b) for n = −0.8. At n = −1 the lens is flat. At this point
F2 F1 P the centre of the circle switches from minus infinity to plus infinity. As
n decreases below n = −1 the radius of the circle decreases, as may be
seen in Fig. 2.29(c) for n = −1.2. For even lower values, n = −2 and
−4 the full lens is shown in Figs. 2.29(d) and (e). It may be seen that
the lens gets smaller as n declines. As n tends towards minus infinity
conducting cylinder
the lens becomes a circle but its radius declines to zero.
An interesting variation on this theme is the reflector of Lagarkov
Fig. 2.30 A reflector that brings a par- and Kissel (2001) also in a 2D geometry. The reflector consists of a
allel beam to a focus. From Lagarkov layer of negative-index material upon a conducting cylinder as shown in
and Kissel (2001). Copyright c 2001 Fig. 2.30. An incident parallel beam refracts in the negative direction
Springer Science + Business Media reaching the conducting (and therefore reflecting) cylinder at P from
which it is reflected, going through the focal point F1 inside the material
and after a further negative refraction on the air/negative-index material
boundary it comes to a second focus at F2 . A possible application of this
reflector is for a microwave antenna. Inverting the path of propagation
a line source positioned at F2 would produce a parallel beam.
We have looked in this section at another family of lenses made possi-
ble by the existence of negative-refractive-index materials. It gives new
options for designing lenses.
2.11.5 Experimental results and numerical

simulations
It took 33 years to confirm experimentally Veselago’s theory but then
it started an avalanche. The credit goes to the group at the University
of California, San Diego headed by David Smith. They were the first
to set out to find a negative-index material, they performed the crucial
experiment and interpreted their results in terms of negative permittivity
and negative permeability. We have seen that it is possible to have a E
material, consisting of metallic rods, which gives negative ε, and it is also
possible to produce negative µ by SRRs. The idea of Smith et al. (2000)
k H
was simply to superimpose in one structure the two different kinds of
elements. A schematic representation of the unit cell is shown in Fig.
2.31.
The experiments were performed with the wave confined between par-
Fig. 2.31 Unit cell consisting of a SRR
allel metal plates and incident perpendicularly upon arrays of those unit and of a metallic rod. From Smith
cells. In order to couple to the SRRs the magnetic field had to be per- et al. (2000). Copyright c 2000 by the
pendicular to the plane of the SRR, and in order to couple to the rods American Physical Society
the electric field had to be parallel with the rods, and of course the di-
rection of propagation was perpendicular to the electric and magnetic
fields as shown in Fig. 2.31.
The crucial experimental results are shown in Fig. 2.32. If only
the SRRs are present there is a stop band between the frequencies of
Transmitted power (dBm)
0
(roughly) 4.7 and 5.2 GHz. In the stop band the attenuation increases –10
from 2 dB to about 35 dB. When the rods are also included then the
–20
stop band turns into a pass band. The attenuation declines from 50 dB
to about 32 dB. Note that the attenuation is very high even in the pass –30
band because there is considerable power absorption by the rods. How- –40
ever, the stop band turning into a pass band proves that a material with –50
both material constants negative can propagate electromagnetic waves. 4.5 5 5.5 6 6.5 7
We shall present here another set of experimental results by Li et al. Frequency (GHz)
(2003). The negative-index material consisted of a similar combination
of SRRs and rods with the difference that there were two rods in the unit
Fig. 2.32 Transmission as a function
cell. The sample was placed at the focal point of a lens-compensated of frequency for SRRs (solid line) and
horn antenna with a similar antenna as the receiver. The experimental for SRRs + rods (dotted line). From
setup is shown in Fig. 2.33(a) and the results in Fig. 2.33(b). This time, Smith et al. (2000). Copyright c 2000
by the American Physical Society
the S-parameters were measured (see Section 1.24 for their definition),
i.e. the reflection in the transmitter and the transmission in the receiver.
We find again that a pass band appears in the frequency range (from
1.0
B D
Magnitude of S21 and S11

0.8
0.6
Experiment
M Simulation
0.4
W C
H L S
0.2
A
0
10 12 14 16 18
Frequency (GHz)
(a) (b)
Fig. 2.33 (a) A negative-index material between two lens-compensated horns. (b) Experimental results (solid lines) for
transmission (A) and reflection (B). Numerical simulations (dotted lines) for transmission (C) and reflection (D). From Li et al.
(2003). Copyright c 2003 American Institute of Physics
Normalized power (linear scale)
1
Microwave absorber LHM
Teflon
Detector 0.8
0.6
0.4
T
0.2
Sample 0
–90 –60 –30 0 30 60 90
Angle from normal (deg)

(a) (b) (c)
Fig. 2.34 (a) Experimental setup for measuring refraction, (b) Left-handed material realized by SRRs and rods, (c) measured
c 2001 AAAS
results. From Shelby et al. (2001a). Copyright
13.2 to 14.2 GHz) where both material parameters are negative. Curves
A and B show the measured values of S21 and S11 , whereas curves C
and D are derived from numerical simulation. The package used was
7
High Frequency Structure Simulation, HFSS,7 a frequency domain Maxwell’s equations solver. The sample
Ansoft Inc., Pittsburgh, 2002 was simulated by a unit cell with periodic boundary conditions. The
agreement may be seen to be extremely good.
The crucial experiment concerning negative refraction was done by
Shelby et al. (2001a). The sample made up by a combination of SRRs
and rods and both produced on printed circuit boards is shown in Fig.
2.34(a). The experimental setup with which they measured the angle
of refraction may be seen in Fig. 2.34(b). The arrays of SRRs and rods
were arranged in the shape of a prism. Microwaves at a frequency of
10.5 GHz were confined vertically by parallel metal plates and laterally
by absorbers. They were incident upon the back of the prism as shown
(a) (b)
Fig. 2.35 Numerical simulations of the refraction of a finite beam by a wedge, (a) εr = 2.2, µr = 1, (b) εr = −1, µr = −1.
From Kolinko and Smith (2003). Copyright c 2003 Optical Society of America. For coloured version see plate section
by the thick black lines. After propagating through the prism the waves
were refracted at an angle of −61◦ corresponding to an effective refrac-
tive index of −2.7. As a control experiment the prism was replaced by
another prism of the same shape but made of Teflon, which has a refrac-
tive index of 1.4. It duly refracted in the positive direction with an angle
corresponding to n = 1.4. The measured beam profiles for Teflon (dot-
ted lines) and for the composite material (solid lines) are shown in Fig.
2.34(c) normalized to unity. The actual measured peaks are of course
quite different because the composite material has much higher losses.
We shall show here simulation of negative refraction by Kolinko and
Smith (2003). A finite beam is incident on a wedge-shaped material that
in the first case (Fig. 2.35(a)) has material parameters εr = 2.2, µr = 1,
and in the second case (Fig. 2.35(b)) εr = −1 and µr = −1. Similarly to
the experimental results shown in Fig. 2.34(c) there is positive refraction
in the first case and negative refraction in the second case. The numerical
package used was HFSS. The finite elements were tetrahedrons. For
convergence, as many as 100 000 elements were required.
2.11.6 Derivation of material parameters from

reflection and transmission coefficients
We shall now be concerned with the inverse problem. Can the values of
ε and µ be deduced from simple measurements, in particular when the
S11 and S21 scattering coefficients are measured for perpendicular inci-
dence? A simple argument suggests that it can be done. The unknowns
are two complex quantities, so they can be determined from the two
complex quantities measured. The analytical forms for the transmission
and reflection coefficients are given by eqns (1.76) and (1.75). Changing
the notation from exponential to trigonometric ones we then obtain for
perpendicular incidence

1 1 j kd j 1
= e = cos(nkd) + η+ sin(nkd) , (2.79)
T′ T 2 η
where η is the impedance of the medium. We also find that

R 1 1
=j η− sin(nkd) . (2.80)
T′ 2 η
√
p of ε and µ we have here the intermediate variables n = εµ and
Instead
η = µ/ε. The technique is then to express cos(nkd) with the aid of R
and T ′ and then find ε and µ from that. The expressions can be found
in the paper by Smith et al. (2000). The main problem is that cos(nkd)
is multivalued, which makes the problem a little unwieldy, but soluble.
The authors were capable of recovering both negative ε and µ, although
their meaning for an insufficient number of elements per wavelength is
controversial. For a discussion see Appendix C.
2.12 The perfect lens

2.12.1 Does it exist?
It is an old subject, going back at least for half a century (Toraldo
di Francia, 1953), how one can beat the classical limit of resolution.
There are essentially two approaches: one uses the far field and relies on
changing the field distribution in the aperture of the lens, the second one
is based on near fields. The first attempt at high-resolution near-field
8
There is actually a much earlier pa- imaging was made by Ash and Nicholls (1972).8 The idea was to make
per recently discovered, in which Synge use of the field leaking out of a microwave cavity through a small hole.
(1928) proposed to improve resolution
in a similar manner. His aim was to
If an object with a structure somewhat larger than the hole is scanned
look at a stained biological specimen, in front of the hole then the resonant frequency of the cavity depends on
ground and planed so that its surface the relative position of the object. By monitoring the resonant frequency
does not diverge from a plane by more it turned out to be possible to obtain information about the structure
than 10 nm. A further requirement
is an opaque plate polished to similar with a resolution of λ/60, close to that of the size of the hole. Their work
planeness and in which there is a hole initiated the whole new field of scanning near-field optical microscopy.
of 10 nm diameter. The object can be An entirely new idea of near-field imaging came with a proposal by
scanned by illuminating the hole with a Pendry (2000). He calculated that the flat lens of Veselago (which re-
high intensity light source and moving
the plate in small steps. He envisaged quired a refractive index of minus unity) will be able to image an object
that these steps might be as small as 10 with infinite resolution, provided εr = −1 and µr = −1. The positions
nm. of the object, lens and image correspond to those of Veselago shown in
Fig. 2.24.
Infinite resolution means perfect imaging. Perfect imaging means that
every single detail of the object is reproduced in the image (including
both propagating and evanescent components). In terms of a spatial fre-
quency (discussed in Section 1.25) it means that the spatial frequency
spectrum of the image (including both propagating and evanescent com-
ponents) will be identical with the spatial frequency spectrum of the ob-
ject. In terms of a transfer function it means that the transfer function is
flat. Entirely flat. It is the same for every spatial frequency component.
Is that possible? Not really. A limit will be set, if by nothing else, then
by the period of the negative-index material (Haldane, 2002). If we can
make metamaterial elements of the size of 100 nm and if the distance
between them is also 100 nm then there would be a chance of making
a lens with a resolution approaching 100 nm. At the time of writing
it does not seem to be likely that such a lens could be made and such
resolution could be obtained, but it is possible in principle. Another
chance is obtained with a material in which only the dielectric constant
is negative. That will not yield a flat transfer function but it would be
flat enough for many purposes, and it would have the great advantage
that natural materials (e.g. silver) with that property exist. The period
in that material will be of the order of one tenth of a nanometer, thus,
at least on that account, the resolution could be extremely high. We
shall discuss the details of this mechanism in Chapter 5. In the present
section we shall show only a few representative examples.
2.12.2 The ideal situation, εr = −1 and µr = −1

Let us first look at the situation when the refractive index is exactly
minus unity in the slab. We found the field distribution in the slab in
Section 1.11 and talked about the spatial frequency spectrum in Sec-
tion 1.25. We shall need both for the analysis. The geometry of the lens
(the Veselago geometry) is plotted in Fig. 2.24. The object is located
at z = 0, the flat lens is between z = d/2 and z = 3d/2, and the im-
age plane is at z = 2d. In order to find the total transfer function we
need to find their value in all three intervals separately. But that has
already been done. Equations (1.158)–(1.160) provide the three transfer
functions. To apply those to the geometry of Fig. 2.24 we only need to
substitute d2 = d and d1 = d3 = d/2 into those equations. Assuming
further an incident TM wave we still have to find the value of ζe from
eqn (1.49). Note that it is necessary to be a little more specific than for
the Veselago lens. For that to work it was sufficient to take n = −1. For
the perfect lens we need both εr and µr to be equal to −1.
We shall now take a particular spatial frequency kx and work out
the total transfer function. Note that in order to satisfy the boundary
conditions the x component of the wave vector is the same in all three
media, so that kx = kx1 = kx2 = kx3 . The propagation coefficient in the
z direction, kz , is given in medium 1 (ε1 = ε0 , µ1 = µ0 ) as
s
ω2
kz1 = − kx2 . (2.81)
c2
In medium 2 (ε2 = −ε0 , µ2 = −µ0 ) we find
s
ω2
kz2 = − kx2 . (2.82)
c2
Since medium 3 is the same as medium 1 it follows that kz3 = kz1 .
Thus, the values of kz are identical in all three media. As a result, from
eqn (1.49)
ζe = −1. (2.83)
Substituting these values into eqns (1.158)–(1.160) we find
j kz d j kz d
− −
T1 = e 2 , T2 = e j kz d , T3 = e 2 , (2.84)
and, consequently,
T1 T2 T3 = 1 . (2.85)
The total transfer function is flat, but not only flat it is actually unity
for all values of the spatial frequency for which there is propagation.
The physics is quite simple. The amplitudes are always the same
because we neglected losses. The phase goes forward in medium 1 by
kz d/2, it goes backward in medium 2 by kz d (remember, it is a backward
wave) and it goes forward again in medium 3 by kz d/2. The total phase
change is zero. Thus, the transfer function is flat for all propagating
waves. This is remarkable but it is still far from the perfect lens. A
perfect lens should reproduce not only those spatial frequencies for which
there is propagation but also those for which the waves are evanescent.
This occurs when the spatial frequency is larger than k0 , i.e. the details
of the object are smaller than the free-space wavelength. This is when
we talk about subwavelength imaging.
The waves are evanescent in all three media when kx > k0 . Then kz1
and kz2 become imaginary (as in Section 1.10) and should be replaced
by −j κ1 and −j κ2 . Then, ζe takes the form
εr2 κ2
ζe = . (2.86)
εr2 κ1
We can then still use the condition ζe = −1, in which case eqn (2.84)
modifies to
κ1 d κ1 d
− −
T1 = e 2 , T 2 = e κ1 d , T3 = e 2 . (2.87)
Remarkably, once more,
z=0 T1 T2 T3 = 1 . (2.88)
The variation of Hy is plotted in Fig. 2.36. It may be seen that the
magnetic field is the same at z = 2d as at z = 0, and that applies to every
Fig. 2.36 Evolution of an evanescent
component single spatial frequency. The conclusion is that the transfer function is
unity whether the wave is propagating or evanescent. However large kx
is, a slab having the material constants εr2 = −1 and µr2 = −1 will
perfectly reproduce the object.
It is easy to understand the physics of phase cancellation. It is also
easy to understand (one might even say it’s trivial) that an evanescent
5 7 9 11
z=0 6d
Fig. 2.37 Multilayer superlens
wave declines exponentially in media 1 and 3. What is difficult to un-

derstand is how one can have a growing wave in medium 2. The usual
expression in the literature is that ‘the wave is amplified’. There is no
doubt that the amplitude of the wave increases within the slab in the
+z direction but ‘amplification’ is too strong a word. It makes one think
that the amount of power in the beam continuously increases, which is
impossible because there is no source of power anywhere. So what is the
explanation?
In general, under evanescent conditions, in the slab there will be a for-
ward propagating wave that declines in the +z direction and a reflected
wave that declines in the −z direction. For ζe = −1 it turns out that
C = 0 (see eqn (1.77)), i.e. the wave that is supposed to decline in the
direction of the group velocity is not present. The only wave present in
the slab is the one that has amplitude D and declines from the bound-
ary at z = 3d/2 in the −z direction. And, of course, the wave declines
away from that boundary in the positive z direction as well. Where did
we see waves declining in both directions from a boundary? It was in
Section 1.10 where we introduced the concept of a surface wave. The
role of the wave growing in the z direction is to excite the surface wave
at the rear boundary.
The problem is of course to realize a material with εr = −1 and
µr = −1 and to use sufficiently small elements to realize significant
subwavelength imaging. This has not been done so far and the chances
do not seem to be very good at the time of writing that it will be done
in the near future. However, there are natural materials for which εr is
negative, silver and gold were examples given before.
We have so far disregarded losses. In their presence, as may be ex-
pected, the transfer function cannot be flat for all values of kx . There
must be a cutoff, as discussed in detail in Chapter 5.
2.12.3 The periodic solution

Looking at Fig. 2.36 it is not difficult to come to the idea (proposed
by Shamonina et al. (2001)) that a periodic solution must exist. If we
create an array by placing further identical slabs at periods of d (see Fig.
2.37) the total transfer function will still be unity. The situation is not
quite the same when µr = 1 and only εr = −1 but it is still true that a
multilayer lens gives a much better image than a single-layer lens of the
same total thickness. This is of great practical significance, as will be
further discussed in Section 5.5.
2.12.4 The electrostatic limit: does it exist?

In the electrostatic limit kx ≫ k0 , hence eqns (2.81) and (2.82) reduce
to
kz1 = kz2 = kz3 = −j kx . (2.89)

When εr = −1 it follows then that ζe = −1 and the derivation in Sec-
tion 2.12.2 yields a flat transfer function, provided the incident wave is
TM polarized. Is this correct? Does it apply to large values of kx ? One
could argue that as kx increases the approximation given by eqn (2.89)
becomes better and better. This is true, but the trouble is elsewhere.
Let us look at the transfer function of the slab. The denominator of
eqn (1.76) is of the form
(1 + ζe )2 e j kz2 d − (1 − ζe )2 e −j kz2 d . (2.90)

If ζe is exactly equal to −1 then the first term in the above expression
is zero and we do get the flat transfer function. Alas, the small devia-
tion of |kx | from |kz2 | that leads to ζe being slightly different from −1
is responsible for the large difference in the value of this term because
the accompanying exponential factor, exp(kx d), may become very large.
Hence, both terms need to be taken into account and the transfer func-
tion is no longer flat, as was pointed out by Shamonina et al. (2001). At
certain values of kx the transfer function suddenly declines. For further
details and curves see Chapter 5.
So, does the electrostatic limit exist? It does not. It is not true that
the transfer function can be flat when only εr is equal to −1. However,
as shown in Chapter 5, the approximation works in a number of well-
defined cases up to a certain value of kx .
2.12.5 Far field versus near field: Veselago’s lens

versus Pendry’s lens
It is easy to describe the operation of Veselago’s lens. All one needs
to know is geometrical optics and the laws of refraction and reflection.
For n = −1 Snell’s law tells us that a wave incident at an angle θ
will be refracted at an angle of −θ. The corresponding ray picture has
already been shown in Fig. 2.24. All dimensions are large relative to the
wavelength.
Pendry’s lens also has n = −1 but it is also important what the ma-
terial parameters are. Both the relative permittivity and the relative
permeability must be equal to −1. Thisp has the further advantage that
the impedance of the medium ηm = η0 µr /εr is equal to the impedance
of free space η0 . Consequently, Pendry’s lens is matched. None of the
í í í30 í í

z/O
z/O

í í

x/O x/O
(a) (b)

max
k x /k :

2
|E x |

í
z/O
(c)
Fig. 2.38 (a), (b) Poynting vector streamlines for a flat lens with d/λ = 500 taking only propagating components into account.
Object half-width is (a) one wavelength and (b) five wavelengths. (c) Evolution of the field across the lens with d/λ = 0.5 and
for various number of evanescent components included. Object half-width is 0.1 λ
incident waves are reflected. However, this is only a minor difference

between the two lenses. After all, Veselago’s lens could also have both
material parameters equal to −1, which would still lead to n = −1. The
major difference is that Veselago’s lens relies on geometrical optics, and
hence on the far field, whereas Pendry’s lens is designed for subwave-
length imaging, and must therefore be concerned with the near field.
Let us see a few examples. We shall take the object in the form of a
Gaussian f (x) = exp(−x2 /τ 2 ) and assume to start with that the object
size (half-width of the Gaussian) is equal to one wavelength and the
width of the lens is d = 500 λ. The object and image planes are, as
usual, at a distance of d/2 in front of and behind the lens. We could
of course use rays to illustrate the operation of the lens but let us use
instead the streamlines of the Poynting vector to see how the power
moves from object to image. The streamline picture (Fig. 2.38(a)) is
exactly the same as the ray picture that we, and everyone else, showed
before. The magnitude of the Poynting vector, presented by the gray

scale, declines in the lateral direction, as it should. Note that when the
amplitude is below the maximum by a factor of 10−10 the scale becomes
completely black. What about evanescent waves? Surely when they
travel a distance of 250 wavelengths they are entirely negligible. For
a value of kx just above the propagating spectrum (say, 1.0001k0) the
value of exp(−κ1 d/2) comes to 10−682 , very small indeed.
Next, we shall choose the half-width of the object five times as large.
The streamlines of the Poynting vector still look the same (see Fig.
2.38(b)) but they are more concentrated because a bigger object gives a
narrower beamwidth. It may be clearly seen in both Figs. 2.38(a) and
(b) that there is a nice internal focus.
Next, let us choose a subwavelength object of half-width 0.1λ and
a lens of d = 0.5 λ and plot in Fig. 2.38(c) |Ex |2 along the z axis.
When we consider propagating waves only 0 < kx < k0 (solid line) there
is a maximum (a rather gentle maximum) in the middle of the slab
corresponding to an internal focus. Let us include now in the spectrum
evanescent components in the range k0 < kx < 1.2k0 (dashed-dotted
line) and in the range k0 < kx < 1.3k0 (dashed line). The maximum
may be seen to shift towards the rear surface. When the upper limit of
kx is chosen up to a value of 1.4 k0 (dotted line) then the maximum is
clearly at the rear surface. The internal focus disappears, but without
influencing the image. In all four cases the image invariably appears in
the image plane. If we include more and more evanescent components
the maximum at the rear surface rises rapidly but the position of the
image is not affected.
Our tentative conclusion is that geometrical optics is valid in the far-
field lens and we can see a corresponding internal focus, whereas the
internal focus disappears in the near-field lens as more and more evanes-
cent waves are added. Is this true? No, it is not. The weak point in
the argument was when we said above about the evanescent waves that
‘Surely when they travel a distance of 250 wavelengths they are entirely
negligible’. Yes, they are very small indeed but we forgot to say that
when those evanescent waves enter the lens they will grow by a factor
exp(κ2 d). Thus, the net growth is enormous.
The growth of the incident evanescent waves is exactly what makes
Pendry’s lens a perfect lens but the same applies whether the lens is
in the far field or in the near field. In principle, there is no difference
between the two lenses. The simple calculations above highlight the
problem with the lossless perfect lens. It cannot possibly exist. However
low are the amplitudes of the evanescent waves incident upon the lens
they will grow in the lens to astronomical figures. The remedy is to
include losses. In the presence of losses the growth in the lens is checked
and near-perfect resolution might be achievable. We shall say a lot more
about this in Chapter 5 concerned with subwavelength imaging.
2.13 Circuits revisited 73
RL
RL
LR
ka S ka S
(a) (b) (c)
Fig. 2.39 Circuit analogue of a left-handed medium (a). Its dispersion curve when ωRL 6= ωLR (b) and ωRL = ωLR (c)
2.13 Circuits revisited

Section 2.6 was devoted to the study of loaded transmission lines. The
main aim was to show in circuit terms how particular loads can lead
to stop bands and pass bands, anticipating the introduction of positive
and negative material constants. Take, for example, Fig. 2.14 where a
resonant LC circuit is coupled by a mutual inductance M ′ to a trans-
mission line. It is a realistic, though approximate, description of what
happens when a plane wave is incident on a lattice of SRRs. Similarly,
Fig. 2.17(a) is a good description in terms of loaded transmission lines
of a lattice of combined SRRs and metallic rods. The primary aim was
to find circuit equivalents of actual physical configurations.
However, once it is conceived that left-handed media can be modelled
by circuits propagating backward waves there is no reason any longer for
the circuits to imitate the physical phenomena occurring in left-handed
media. If a simple non-resonant loading circuit can produce the same
type of dispersion characteristics that would be equally good in building
an analogue of a left-handed medium. In fact, it would be better because
there would be more flexibility in choosing the parameters. Such a simple
circuit in the form of Fig. 2.39(a) was proposed by Eleftheriades et al.
(2003). Here, Lt and Ct are the inductance and capacitance of the
transmission line per unit length, a is the length of the transmission line
section, Lsh is the shunt inductance and Cs is the series capacitance with
which the transmission line is loaded.9 9
It is rather confusing that the same
An identical circuit, but with a different reasoning, was proposed by notations L and C mean sometimes in-
ductances and capacitances proper and
Sanada et al. (2004). They start with the statement that a left-handed some other time inductances and ca-
material may be represented by a series capacitor, CL , and a shunt pacitances per unit length. Alas, these
inductor, LL (for circuit and dispersion curve see Fig. 1.17). But in any notations are so widespread in the lit-
practical realization, they argue, there will be a series inductance LR , erature that the only remedy is to call
attention to them each time they ap-
associated with the capacitor, and a shunt capacitance CR , associated pear.
with the inductor. It is clear from the notations that LR and CR on
their own represent a right-handed line and CL and LL a left-handed
line. The authors called this general line a composite right/left-handed
transmission line.
The two models are of course identical, provided we change the nota-
tions to
CL = Cs , LL = Lsh , CR = Ct a , LR = L t a . (2.91)
The dispersion equation may be obtained with the aid of eqn (1.142) by
noting that

1 1
Z = j ωLR − and Y = j ωCR − , (2.92)
ωCL ωLL
yielding
ka (ω 2 − ωRL
2
)(ω 2 − ωLR
2
)
4 sin2 = −ZY = 2 2 , (2.93)
2 ω ωRR
where
1 1 1
ωRR = p ; ωRL = p ; ωLR = p . (2.94)
CR LR CR LL CL LR
The corresponding dispersion curve is plotted schematically in Fig.
2.39(b). There are two branches: the upper one is a forward wave and
the lower one is a backward wave. An interesting special case arises
when ωRL = ωLR , i.e. the series resonant circuit and the shunt resonant
circuit have the same resonant frequencies. In that case the upper and
lower curves have a common point at ka = 0, ω = ωRL (see Fig. 2.39(c))
where the group velocities are non-zero.
For some practical realizations of the equivalent circuits in the mi-
crowave region see Chapter 6.
Plasmon–polaritons
3
3.2 Bulk polaritons.
The Drude model 77
Plasmas is an old and respectable subject. It was started by Langmuir1
3.3 Surface plasmon–polaritons.
in the 1920s. Its properties have been studied ever since but not always
Semi-infinite case,
with the same vigour. Topics rose in response to new applications, and TM polarization 79
declined on reaching saturation or realizing that the chances of the envis- 3.4 Surface plasmon–polaritons
aged solution are fast receding. The development of radio broadcasting on a slab: TM polarization 92
led to the discovery of the Earth’s ionosphere that reflects radio waves 3.5 Metal–dielectric–metal
and is responsible for reception of radio signals when the transmitter and periodic structures 102
is over the horizon. Hannes Alfven’s theory of magnetohydrodynam- 3.6 One-dimensional
ics (1940) that treated plasma as a conducting fluid has been employed confinement: shells
and stripes 103
to investigate sunspots, solar flares, the solar wind, star formation and
other topics in astrophysics. The interest in thermonuclear fusion came
in the wake of the hydrogen bomb. At one time it was believed that
fusion, assisted by plasmas, is round the corner. The belief turned out
to be not well founded. Later, the invention of lasers opened the new
chapter of laser plasma physics with some promise of fusion but that has 1
A transparent liquid that remains
not been realized either. High-energy physicists hope to use plasma ac- when blood is cleared of various corpus-
celeration techniques to dramatically reduce the size and cost of particle cles was named plasma (after the Greek
word that means ‘formed, jelly-like’) by
accelerators. Among the applications that have come off are microscopy
a Czech medical scientist Purkinje. The
and sensing (see, e.g., Yeatman 1996; Homola 2003; Barnes 2006). Nobel prize-winning American chemist
We have enumerated only a fraction of plasma phenomena that could Irving Langmuir first used this term to
be discussed. The subject is not only old and respectable: it is also very describe an ionized gas in 1927: Lang-
muir was reminded of the way blood
big. Our interest is limited to plasmas on surfaces, the kind that was plasma carries red and white corpus-
presented very briefly in Section 1.10. They are important for meta- cles by the way an electrified fluid car-
material applications that belong to two categories. One is their role ries electrons and ions. This analogy
in the subwavelength manipulation of images. If µr = 1 and εr = −1 is, however, slightly odd: blood plasma
is blood that is free from blood cor-
then the plasma resonances will limit the range of spatial frequencies puscles; electric plasma would lose its
for which the transfer function is approximately flat. This was already properties if we were to get rid of free
briefly discussed in Section 2.12.4. The second relationship with meta- charges in it.
materials is via elements that may exhibit plasma resonances at high
enough frequencies. They are mainly nanosized rods and rings and their
combinations, reminiscent of electric dipole and loop antennas operating
at radio frequencies.
There is, however, a third category that hardly exists at the moment.
Nanoscale geometry, for example, allows us to change plasma properties
and understanding of this mechanism would allow us to design metama-
terial elements with tailored properties. This is just one example that
may or may not turn out to be feasible but there are many others. We
76 Plasmon–polaritons
believe that there is such a wealth of plasma phenomena available that

lots of new metamaterial applications are bound to come. And that we
regard as an equally important motive for delving into the multifarious
physical phenomena.
The simplest plasma wave is a density wave of mobile electrons in
the background of immobile positive charges. They interact with one
another via Coulomb forces. The responsiveness of a plasma is a direct
consequence of the mobility of its constituents, and occurs only when
the charged particles that comprise it are relatively free. Examples of
systems that show plasma-like behaviour include gas discharges, elec-
trons and holes in semiconductors, and, what is of particular interest for
us, free electrons in metals. The dispersion of a plasma wave is quite an
unusual one (see, e.g., Solymar and Walsh 2004),
ω(k) = ωp = const . (3.1)

ω, being a constant independent of k, means that only a single value of
the frequency, the plasma frequency, is allowed, while the wave number
can be arbitrary. Does this equation, which gives a straight horizon-
tal line on an ω–k diagram, show a dispersive behaviour? The answer
2
If we plot a dispersion curve and we is yes.2 Of course not all charged-particle systems show such behav-
find that the frequency is independent iour. An ionic crystal, e.g. sodium chloride, is made up of positively and
of the wave number then it is tempting
to call it a non-dispersive wave—and
negatively charged Na+ and Cl− ions. The ions are so tightly bound
some authors indeed do that. On the to one another by electrostatic forces that they cannot move about
other hand, if we remember the original freely; this system supports acoustic rather than plasma waves. And
aim of the dispersion curves to show the yet both plasma waves and acoustic waves show remarkable similari-
change of wave velocity with frequency
then it is quite obvious that a horizon- ties in their ability to interact with electromagnetic waves, giving rise
tal line is very dispersive. The phase to hybrid modes. And that brings us to the point where we have to
velocity ω/k varies strongly as a func- say something about terminology. As an electromagnetic wave propa-
tion of k. The only dispersionless curve gates through a polarizable medium, the polarization it induces modifies
is a straight line that goes through the
origin. the wave and the electromagnetic wave becomes coupled to the induced
polarization. We could call this a hybrid mode because its properties de-
pend both on those of the electromagnetic wave and also of the medium.
The problem is then that apart from vacuum we nearly always have hy-
brid modes because the properties of the medium can very rarely be
completely disregarded. The argument may then be used that we call
them hybrid modes only when the effect of the medium is significant.
This is apparently what happened to acoustic and plasma waves. When
they significantly alter the properties of electromagnetic waves they ac-
quire the postfix ‘polariton’. And if that does not sound sufficiently pre-
3
Admittedly classical physics fails tentious their name is also upgraded. They change from the classical3
when the propagation length of the ‘plasma waves’ and ‘acoustic waves’ to ‘plasmons’ and ‘phonons’ ending
mode and the wavelength are compara-
ble with atomic distances (Yang et al.,
up eventually as plasmon–polaritons and phonon–polaritons. To make
1991) but we are very far from that things even more complicated we shall have to distinguish between bulk
limit. plasmon–polaritons that propagate inside a homogeneous medium, and
surface plasmon–polaritons that stick to a surface. Most physicists refer
to them nowadays by the acronym SPP.
We shall first look at bulk plasmon–polaritons in Section 3.2 on the
basis of the Drude model, and then find the conditions for the existence
3.2 Bulk polaritons. The Drude model 77
of surface plasmon–polaritons along a single interface in Section 3.3. We

shall discuss there the dispersion characteristics with particular reference
to losses and also the behaviour of the Poynting vector, how it can
predict whether a wave is of the forward or of the backward variety.
In Section 3.4 a more complicated situation is investigated, the plasma
properties of a metal slab embedded in a dielectric. There are then two
brief sections, Sections 3.5 and 3.6, the former on metal–dielectric–metal
slabs, and the latter on one-dimensional structures like a metal sheet of
infinite length but finite width. In Section 3.7 we return to a specifically
metamaterial theme: the surface modes that may exist for arbitrary
values of the permittivity and permeability.
3.2 Bulk polaritons. The Drude model

We have actually encountered bulk plasmon–polaritons, without calling
them so, in Section 1.9. There, we discussed the dielectric function of a
lossless isotropic medium in the presence of a current,
ωp2
εr = 1 − , (3.2)
ω2
known as the Drude model for a free-electron gas. In the presence of
losses the above expression modifies to
ωp2
εr = 1 − , (3.3)
ω (ω − jγp )
with the damping constant, γp = 1/τ , vanishing if the collision time,
τ , becomes infinite. In most metals, the plasma frequency is in the
ultraviolet,4 making them shiny in the visible range as all the incident 4
Taking, for instance, the electron den-
light is reflected. In doped semiconductors, the plasma frequency is sity in a typical metal as Ne = 6 × 1028
m−3 we can calculate that ωp = 2π ×
usually in the infra-red.
2.2×1015 s−1 , which corresponds to the
The asymptotic and limiting cases for eqn (3.2) are wavelength λp = 2πc/ωp = 136 nm,
 which lies in the ultraviolet.
 εr → −∞ as ω→0
εr (ωp ) = 0 . (3.4)
εr → 1 as ω→∞

What are the implications of eqn (3.2) for electromagnetic wave propaga-
tion in such a medium? An electromagnetic wave of the form
exp [j (ωt − k · r)] no longer propagates without dispersion. Its wave
number has to satisfy the condition
s s
2
ω√ ω ω p ωp2
k= εr µr = 1 − 2 = k0 1 − 2 , (3.5)
c c ω ω
resulting in a dispersion equation
ω 2 k02 = ω 2 k 2 + ωp2 k02 . (3.6)

3.5
3
light line
2.5
bulk Z/k=c
Z/Zp
2 plasmon−
polariton
1.5
1
plasmon
0.5 Z=Zp
Fig. 3.1 Dispersion of bulk plasmon– 0

0 1 2 3 4
polaritons, of plasmons and of the elec- k/kp
tromagnetic wave in vacuum
It is plotted in Fig. 3.1 as a function of k/kp (where kp = ωp /c), together

with the ‘light line’ k = ω/c of the electromagnetic wave in vacuum and
with the horizontal dispersion line ω = ωp .
The dispersion curve in Fig. 3.1 may be seen as resulting from the
crossing of the bulk plasmon line ω = ωp and the light line k = ω/c. A
possible way of interpreting the dispersion equation (3.6) is that there is
no longer a pure electromagnetic and a pure plasma wave. Instead, the
plasma wave interacts strongly with light, resulting in our bulk plasmon–
polariton. The asymptotic and limiting cases for the dispersion are

ω → ωp as k→0
. (3.7)
ω → kc as k→∞
At high frequencies the bulk mode lies close to the light line and is light-
like; at low wave numbers it moves away from the light line until the cut-
off plasma frequency is reached. Bulk plasmon–polaritons of frequency
5
There is a stop band for ω < ωp . below the plasma frequency cannot propagate5 in plasma because its free
Note a general feature (Tilley, 1988) electrons screen the electric field of the light. Bulk plasmon–polaritons
that surface modes can be found within
the bulk mode stop bands. It is in this
of frequency above the plasma frequency can propagate through plasma
range that our other type of plasmon– because the electrons cannot respond fast enough to screen it.
polariton, the SPP, can propagate. Since many of the properties of plasmon–polaritons are similar to
those of phonon–polaritons it may be worthwhile to include here a brief
note on the latter. The interaction is then between the electromagnetic
wave and the optical branch (optical phonons) of the acoustic wave in
an ionic crystal. There, the dielectric function has a form
2
ωL2 − ωT
2

ωT
εph = ε∞ + 2 (ε0 − ε∞ ) = ε∞ 1 + 2 , (3.8)
ωT − ω 2 ωT − ω 2
6
Note that ε0 in eqn (3.8) is not the with ωT p the so-called TO (transverse optical) phonon frequency and
free-space permittivity but the value of ωL = ωT ε0 /ε∞ the LO (longitudinal optical) phonon frequency (see,
the dielectric constant at ω = 0. e.g., Kittel 1963).6 The asymptotic and limiting cases here are
3.3 Surface plasmon–polaritons. Semi-infinite case, TM polarization 79

 εph → ±∞ as ω → ωT
εph (ωL ) = 0 . (3.9)
εph → ε∞ as ω→∞

√
The dispersion equation is then k = ω εph µ. The frequency at which
k = 0 is known as the Reststrahl frequency (Fox, 2001). Note further
that in the frequency range ωT < ω < ωL the dielectric constant is
negative7 and the wave vector is imaginary, meaning a stop band for 7
An example is SiC, a polar material.
the bulk phonon–polariton. It may be instructive to note that formally It was used for subwavelength imaging
in the negative dielectric constant re-
the plasmon–polariton dispersion equation (3.5) is a special case of the gion (see Section 5.4.6) by Korobkin
phonon–polariton dispersion equation with ε∞ = 1, ωL = ωp and ωT = et al. 2006a; Korobkin et al. 2006b.
0.
3.3 Surface plasmon–polaritons.

Semi-infinite case, TM polarization
The electromagnetic field of SPPs at a dielectric–metal interface is ob-
tained from the solution of Maxwell’s equations in each medium, and z
the associated boundary conditions. The latter express the continuity
of the tangential components of the electric and magnetic fields across Metal E
the interface. A further obvious condition is that the fields must decline +++ +++ x
away from the boundary and vanish infinitely far away. H
3.3.1 Dispersion. Surface plasmon wavelength Dielectric

We consider now a plane interface z = 0 between two semi-infinite media,
the first one being vacuum, air or a dielectric with the permittivity, ε1 ,
Fig. 3.2 SPP mode. A sketch of charge
taken here as frequency-independent; the second one being a metal with and electric-field distributions at the
the permittivity, ε2 , described by the Drude model (eqn (3.2)). We metal–dielectric boundary
shall first look at the case of TM polarization. Recalling the results
from Section 1.10 the conditions for the existence of a surface mode
propagating along the surface in the x direction8 are given as 8
For εr1 + εr2 < 0, µr1 = µr2 = 1
a TM surface mode with components
Ex , Ez and Hy and propagating in the
ε1 ε2 < 0 and ε1 + ε2 < 0 . (3.10) x direction is the only option. Explicit
The first of these conditions, requiring that the dielectric constant is calculations show that there is no prop-
agation for this mode in the y direction,
changing sign at the boundary, has a simple meaning (Barnes, 2006). A nor is there a surface TE mode carrying
surface mode involves charges accumulated at the boundary between the Hx , Hz and Ey components.
metal and the dielectric. The corresponding electric field that is built up
would originate on positive charges and terminate on negative charges
and do so, in fact, in both medium 1 and medium 2. This is shown
schematically in Fig. 3.2. That means that in the presence of surface 9
We have just described the role of sur-
charges the components of the electric field normal to the surface must
face charges and yet, when it comes to
have opposite signs. the boundary conditions, we ignore the
A boundary condition for the normal component of the displacement surface charge. This apparent contra-
field is9 diction is resolved in Appendix D.
ε1 E1z = ε2 E2z , (3.11)

which is another way of saying that for a surface mode to exist dielec-
tric constants in the two media must have different signs. The second
condition of eqn (3.10) follows from eqn (1.73), which is given below
r
εr1 εr2
kx = k0 . (3.12)
εr1 + εr2
It ensures that kx is real so that the mode can propagate along the
surface. It is a rather innocent-looking equation but it contains a large
amount of information about the properties of SPPs at a single inter-
face. Since εr2 is a function of frequency, eqn (3.12) is the dispersion
equation, i.e. it provides the relationship between the frequency and the
wave number. Inserting eqn (3.2) from the lossless Drude model into
eqn (3.12) we find
kx2 (1 + εr1 )k02 − kp2 = k02 εr1 (k02 − kp2 ) ,

(3.13)
where
ωp 2π
kp = = . (3.14)
c λp
Equation (3.13) may then be solved for k0 to obtain the dispersion equa-
tion in terms of ω. It looks rather complicated so we shall give here the
solution only for εr1 = 1
s
2
ω p ωp4
ω2 = + c2 kx2 ± + c4 kx4 , (3.15)
2 4
10
The positive sign in eqn (3.15) also which is not too long. Taking the negative sign10 we shall plot in Figs.
has physical meaning and yields the 3.3(a) and (b) the dispersion curves at an interface metal–vacuum (εr1 =
Brewster wave discussed in Appen-
dix E.
1) and metal–glass (εr1 = 2.25).
The dispersion curve shows that at low frequencies the surface mode
lies close to the light line and is light-like. As frequency increases, the
mode moves away from the light line, gradually approaching an asymp-
totic limit. This occurs when the permittivities in the two media are
of the same magnitude but opposite sign, thus producing a pole in the
dispersion equation (3.12) (Barnes, 2006). The asymptotes are
√
kx → ω

c εr1 as ω→0
, (3.16)
kx → ∞ as ω → ωs
where
ωp
ωs = √ (3.17)
1 + εr1
is the upper cutoff frequency for the surface mode. For vacuum it is
ωp
ωs = √ . (3.18)
2
Medium 2 with negative permittivity is often called the surface-active
medium, while medium 1 with positive permittivity is called the surface-
passive medium. The reason is quite simple. The terms express the fact
2 2
Z/k=c
bulk light line bulk
1.5 plasmon− Z/k=c 1.5 plasmon- 1/ 2
polariton polariton Z/k=c/Hr1
Z=Zp
Z/Zp Z=Zp
Z/Zp
1 1
Z=Z
s
Z=Z
s
0.5 surface 0.5
plasmon− surface
polariton plasmon-
polariton
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
k/kp
k/k
p
(a) (b)
Fig. 3.3 Dispersion of SPPs (solid line) at the boundary vacuum–metal (a) and at the boundary glass–metal (b). Also shown
is the dispersion curve of the bulk mode with ky = kz = 0 (thin solid line), the light line in vacuum and in the dielectric
(dashed lines), the dispersions of bulk (dotted line) and surface plasmons (dashed-dotted line)
that it is charges in medium 2 that lead to electric-field oscillations at

the boundary (Abeles, 1986; Tilley, 1988).
Note that the SPP dispersion lies in a region to the right of the light
lines in medium 1, where the kx values are larger than the wave number
of the propagating electromagnetic wave
ω√
kx > εr1 , (3.19)
c
and therefore this mode is non-radiative, or bound (Burke et al., 1986).
This means that this wave has no propagating component in the z direc-
tion; z components of the k vector are purely imaginary. There are two
important consequences. First, the surface mode on such a flat interface
cannot decay by emitting a photon, and, conversely, it cannot be excited
directly by an incident plane wave. An incident plane wave can never
have a wave vector parallel to the interface large enough to couple to
this surface mode. In order to couple to the SPP, both k and ω must
match. In engineering terms one could say that both frequency and ve-
locity must agree, whereas physicists would talk about conservation of
energy and momentum. One may use a prism-coupling scheme based
on the method of attenuated total reflection, which would provide an
evanescent component with imaginary kz and sufficiently large kx . Best
known are the Kretschmann and the Otto configurations (Kretschmann
and Raether, 1968; Otto, 1968). Another consequence is that this mode
is capable of interacting with the evanescent part of a spectrum from
a small subwavelength object, providing an important mechanism for
subwavelength imaging lying at the heart of the concept of the ‘perfect
lens’ (see Chapter 5 for a detailed analysis).
According to the inequality (3.19) the SPP wavelength is always smaller
polariton wavelength λSPP [m]
−6
10
light
line
−7
10 −7 −6
Fig. 3.4 Dispersion curve for the 10 λp λ 10
s
metal–vacuum interface free space wavelength λ [m]
than the free-space wavelength. It would be desirable at this stage

to abandon the normalized quantities ω/ωp and k/kp and find the ac-
tual values of the various characteristics of the waves, starting with the
plasma frequency. For silver, the value of the plasma frequency was
determined experimentally by McAlister and Stern (1963) from the dip
in the transmission of electromagnetic waves through a thin film. They
measured the plasma frequency in terms of electron volt units, giving a
value of 3.8 eV, which in SI units corresponds to 918 THz or a wave-
length of 326 nm. According to Pendry (2000) the plasma frequency is
approximately 9 eV, corresponding to 2183 THz, a big difference. The
main problem is of course that the Drude model, assumed so far is not
valid for frequencies close to the plasma frequency. When ω = ωp the
dielectric constant is not equal to 1 but to a value of ε0 (a horrendous
deviation from the meaning of ε0 in the SI system) due to interband
transitions. If we follow Pendry and take its value as 5.7 then εr is equal
to −1 at a wavelength of 359 nm. When coming to imaging by a sil-
ver lens then this is the wavelength range at which successful imaging
was obtained (see, e.g., Melville and Blaikie 2005). Can we save the
Drude model without contradicting experimental results? We can do it
by choosing the plasma frequency so that the Drude relation remains ap-
11
A technique often used (see, e.g., proximately true11 in the critical region around εr = −1. This choice led
Gray and Kupka 2003). us to fp = 1200 THz. We can now plot (Fig. 3.4) the wavelength of the
SPP mode λSPP = 2π/kx versus free-space wavelength λ = 2πc/ω. As
the free space wavelength approaches the asymptotic value λs = 2πc/ωs
the SPP wavelength becomes much shorter than the free-space wave-
length, opening up the possibility for subwavelength manipulation of
the fields. For larger values of the free-space wavelength, λSPP is only
marginally smaller than λ.
In view of the dispersion equations derived we should say here a few
more words about terminology. The logical thing would be to call the
waves surface plasmon–polaritons for small kx when the dispersion curve
is close to that of electromagnetic waves in free space and it is some kind
Z Z
1 s p
−2
permittivity H’, − H"
−H"
−4
H’
−6
−8
Fig. 3.5 The real and imaginary parts

−20 of the dielectric function for a lossy
0 0.5 1 1.5 2
Z/Zp metal according to the Drude model
(γp /ωp = 0.01)
of hybrid wave, but call them surface plasma waves for large kx . Alas,
logic is not an influential factor when it comes to terminology. We shall
accept the majority view, accept the term surface plasmon–polariton,
and keep on referring to them as SPPs. This may also be the place
where we can mention the electrostatic limit, which means obtaining
the dispersion equation by solving Laplace’s equation instead of the full
apparatus of Maxwell’s equations. This is permissible for large kx . For
more details see Appendix F.
3.3.2 Effect of losses. Propagation length

If we take losses of medium 2 into account, we should expect that a
surface wave bound to the interface would propagate along the surface
a finite distance, until its energy is dissipated as heat in the metal.
In eqn (3.12), in the presence of losses, the right-hand side is a complex
quantity, and so is the left-hand side, meaning that kx will have an
imaginary part as well. We need to write
kx = kx′ − j kx′′ , (3.20)

and the complex relative permittivity of the metal as
εr2 = ε′ − j ε′′ . (3.21)

We obtain for its real and imaginary parts (Boardman, 1982; Berini,
2000a)
ωp2 ωp2 γp
ε′ = 1 − , ε′′ = . (3.22)
ω 2 + γp2 ω ω 2 + γp2
The corresponding values of kx′ and kx′′ may be determined from
eqn (3.12) by substituting into it the complex value of the dielectric
0.7 0.7
0.6 0.6
Re(k ) Re(k )
x x
0.5 Im(k x) 0.5
Im(k x)
0.4 light line 0.4 light line
p
p
Z=Z Z/Z
Z/Z
s Z=Z
s
0.3 0.3
02 0.2
0.1 0.1
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
k /k kx/kp
x p
(a) (b)
Fig. 3.6 Dispersion for lossy metal. γp /ωp = 0.01 (a) and 0.1 (b)
constant. The real and imaginary parts of εr2 are plotted in Fig. 3.5 as
a function of frequency for γp = 0.01ωp.
The dispersion curves in the presence of losses may now be calculated
with the aid of eqn (3.12). They are plotted in Figs. 3.6(a) and (b). The
normalized horizontal co-ordinate is kx′ c/ωp = kx′ /kp for the propagation
coefficient (solid lines) and kx′′ c/ωp for the attenuation. For the lossless
case γp = 0 there is no attenuation in the pass band between 0 and
ωs . For γp /ωp = 0.01 the attenuation becomes significant just below
ωs but it declines sharply as the frequency decreases (Fig. 3.6(a)). The
same is true for γp /ωp = 0.1 but the effect is more substantial (see
Fig. 3.6(b)). All the information about propagation and attenuation is
properly contained in Figs. 3.6(a) and (b) but this is not the way most
people in the art of plasmas like to talk about attenuation. They prefer
to use the measure of propagation length, the distance the SPP travels
before its intensity is reduced by a factor of e. The relationship between
the propagation length and the attenuation coefficient is LSPP = 1/2kx′′
(the factor 2 is there because propagation length refers to intensity,
whereas we have used kx′′ for the attenuation of the wave amplitude).
For this reason we also plot the propagation length in Fig. 3.7 for the
same set of loss parameters. It may be seen that in the visible region
(with our choice of ωp ) it falls roughly between 0.2ωp and 0.4ωp. The
largest value of the propagation length for γp = 0.01 ωp is about 10 µm,
a very small length. On the other hand, it is still large relative to both
the free-space wavelength and the SPP wavelength. So there would be
about 20 wavelengths available for manipulation of information and that
might be enough for practical application.
−2
10
propagation length [m]
−4
10
−6
10 Jp/Zp= 0.01
Jp/Zp= 0.1
free O
OSPP
−8
10 −6 −5
10 10 Fig. 3.7 Propagation length. fp =
free space wavelength O [m] ωp /2π = 1200 THz
3.3.3 Penetration depth

The penetration depth into the dielectric, δd , or into the metal, δm , is
the distance in the z direction from the surface over which the intensity
of the field is reduced by a factor of e. It follows immediately that the
penetration depth is inversely proportional to the imaginary part of the
z component of the wave vector,
1 1
δd = ′′ , δm = ′′ . (3.23)
2kz1 2kz1
In Section 1.10 we have already calculated the values of the normal
components of the k vector in both media. The relationship is
q q
kz1 = εr1 µr1 k02 − kx2 , kz2 = εr2 µr2 k02 − kx2 . (3.24)
Disregarding losses we have seen already that the normal components
are purely imaginary, which means that the field amplitudes decay away
from the surface,
kz1 = −j κ1 , kz2 = −j κ2 . (3.25)

A number of important conclusions may be drawn. For the lossless
case it may be shown that

δm εr1
≃ . (3.26)
δd εr2
Since in the frequency range of surface plasmon–polaritons |εr2 | ≥ εr1
(see eqns (3.10)), this means that the penetration depth in the surface
active medium, metal, is always smaller than in the dielectric. At low
frequencies, ω ≪ ωs , the penetration depth in the metal is much smaller
than in the dielectric, whereas in the high-frequency limit with ω → ωs
the penetration depths in the two media become comparable. Another
δair δair
−4 −4
10 δ 10 δ
metal metal
skin depth skin depth
free space λ free space λ
penetration [m]
λ
SPP penetration [m] λ
SPP
−6 −6
10 10
−8 −8
10 10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
ω/ωp ω/ωp
(a) (b)
Fig. 3.8 Penetration depth for (a) γp = 0, (b) γp = 0.1ωp
interesting point is their very different asymptotic behaviour at low fre-

quency,
λp λ2
δm ≃ , δd ≃ . (3.27)
4π 4πλp
The penetration depth in the metal in the low-frequency limit is prac-
tically frequency-independent, whereas the penetration depth in the di-
electric is strongly frequency-dependent and can be quite large in the
infra-red (Abeles, 1986).
Barnes et al. (2003) in their review pointed out the importance for the
SPP-based nanocircuitry of these three characteristic length scales, the
propagation length, the penetration depth into the dielectric and into
the metal. The propagation length of the SPP mode, LSPP , is usually
determined by the loss in the metal. For a low-loss metal, for example,
silver, at a wavelength of 500 nm it is as large as 20 mm. The prop-
agation length sets the upper size limit for any photonic circuit based
on SPPs. The penetration depth into the dielectric material, δd , is typi-
cally of the order of one half the wavelength of light involved and dictates
the maximum height of any individual features, and thus components,
that might be used to control SPPs. The ratio of LSPP /δd thus gives
one a measure of the number of SPP-based components that may be
integrated together. The penetration depth into the metal, δm , deter-
mines the minimum feature size that can be used; this is between one
and two orders of magnitude smaller than the wavelength involved, thus
highlighting the need for good control of fabrication at the nanometer
scale.
Another important point is that the penetration depth into the metal
gives an idea of what thickness is required to allow coupling between the
modes on the two surfaces of a thin metallic film. As shown later in Fig.
3.17(b) for a certain set of parameters, and as will be discussed in much
1 1
10 10
0 free space λ 0 free space λ

10 10
penetration δ/λ
penetration δ/λ
λ λSPP
SPP
δ δ
−1
air −1
air
10 10
skin depth skin depth
δmetal δmetal
−2 −2
10 10
−3 −3
10 10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
ω/ωp ω/ωp
(a) (b)
Fig. 3.9 Penetration depth normalized to the free-space wavelength for (a) γp = 0, (b) γp = 0.1ωp
more detail in Chapter 5, a film of 60 nm thickness at a wavelength

around 360 nm may be regarded as too thick for an effective coupling
between the two surfaces. The subwavelength imaging mechanism relies
on this coupling and therefore very thin films are essential there.
The penetration depth into the dielectric and into the metal are plot-
ted in Figs. 3.8(a) and (b) as a function of ω/ωp for the lossless case,
and for γp = 0.1ωp within the range of the SPPs from 0 to ωs . The
free-space wavelength and the SPP wavelength are plotted as well for
comparison. The penetration depth in the metal exhibits a remarkably
constant value for a wide range of frequencies and there is not much
difference whether losses are present or not. There is though a differ-
ence at very low frequencies where the penetration depth increases with
decreasing frequency. At this stage it is interesting to compare the pen-
etration depth with the skin depth (see, e.g., Solymar 1984), well known
from undergraduate studies. Both the penetration depth and the skin
depth stand for the same thing: how far the field will penetrate into the
metal. The difference between them is that the usual measure of the
skin depth is for perpendicular incidence, implying a kx = 0, whereas
kx is finite in the SPP case. As may be seen in Fig. 3.8, the deviation of
the penetration depth from the skin depth is quite small. The curves of
Fig. 3.8 are replotted in Fig. 3.9 but now normalized to the free-space
wavelength.
The penetration depth into the dielectric is lower than λSPP at higher
frequencies and larger than λSPP at lower frequencies. That increase,
relative to the free-space wavelength, arises because at lower frequencies
the metal is a better conductor and the dispersion of the SPP is closer
to the light line and is less confined to the surface.
It is interesting to note that the penetration depth into the dielectric
for a localized mode associated with nanoparticles can become much
1 0
10
normalized eld at the boundary

normalized eld at the boundary
0.8
0.6 Ez1 −2
10
Ez2
0.4 Ex
Ez1
−4
Ez2
0.2 10
Ex
0 −6 −5 −4 −6 −5 −4
10 10 10 10 10 10
free space wavelength O [m] free space wavelength O [m]
(a) (b)
Fig. 3.10 Electric-field components at the metal–air interface
smaller, of the order of 10 nm (Whitney et al., 2005; Barnes, 2006). This

is attributed to the divergence of the electric field around the tips of such
structures. The larger the localization of the field near the metal surface,
the larger is the field enhancement (Barnes, 1998). This enhanced field
makes SPP modes sensitive to changes at the surface and opens up
possibilities to use SPP modes for sensor applications (Barnes, 2006;
Homola, 2003).
We can conclude from looking at Figs. 3.8 and 3.9 that as the free-
space wavelength decreases both the propagation length and the pene-
tration depth into the dielectric decrease. As ω approaches ωs , the wave
penetrates to equal extents into both the dielectric and the metal, and
as this happens the propagation length declines as well. So the correla-
tion between confinement and attenuation may be roughly expressed as
follows: the smaller the fraction of the wave in the metal, the smaller
is the attenuation. We will return to this question when discussing the
case of two interfaces.
3.3.4 Field distributions in the lossless case

Next, we shall look at field distributions at the boundary. As seen before
there is only one component of the magnetic field, Hy , along the y axis,
while the electric field has two components, Ex and Ez in the sagittal
plane that is defined by the normal to the surface (the z axis) and the
direction of propagation (the x axis).
The field in the z direction is evanescent, reflecting the bound non-
radiative nature of the SPP. No power is transferred in the normal direc-
tion away from the surface. The Hy and Ex components are continuous
at the boundary, whereas the Ez component changes sign and can have
a large discontinuity in its magnitude, depending on the ratio εr2 /εr1 .
1
Ex
Ez
normalized eld
Hy
−1
−200 0 200 −100 0 100 −20 0 20
z [nm] z [nm] z [nm]
(a) (b) (c)
Fig. 3.11 Field amplitudes as a function of z for (a) ω/ωp = 0.25, (b) ω/ωp = 0.5, (c) ω/ωp = 0.7
The ratio of the transverse and longitudinal components of the electric

field in the dielectric medium may be obtained from eqns (1.69) and
(3.12) as
s
ωp2 − ω 2
r
Ez1 −εr2
=j =j . (3.28)
Ex εr1 εr1 ω 2
The transverse component is dominant in the electric field of SPPs with

small wave numbers and low frequencies (light-like case) for which the
dispersion curve is close to the dielectric light line. When kx ≫ k0
then the transverse and longitudinal components are comparable. They
are equal when kx → ∞. The two components of the electric field at
the boundary as a function of frequency, normalized to its maximum,
are plotted in Fig. 3.10(a). They are replotted in Fig. 3.10(b) using a
logarithmic scale in order to show how much smaller Ex is for most of
the range.
The decay of the fields is faster in the metal than in the dielectric as
follows from |κ2 | being always larger than |κ1 |. In Figs. 3.11(a)–(c) we
show the decay of Ez , |Ex | and Hy as a function of z in both media √ for
ω/ωp = 0.25, 0.5 and 0.7. It may be seen that as ω tends towards ωp / 2
the decay in air becomes relatively stronger and the x and z components
of the electric field become equal. The absolute decay strongly increases
as ω increases, which may be seen by realizing that the scales in the
three figures are different.
The amplitude of the magnetic field in the xz plane is shown in Figs.
3.12(a)–(c) by the gray-scale contrast for ω/ωp = 0.25, 0.5 and 0.7. The
magnetic field may be seen to be more confined at ω/ωp = 0.7. Notice
again the change of scale. The SPP wavelength is strongly reduced as
ω approaches ωs . The electric field lines are also plotted in the same
figure.
−1 −0.5 0 0.5 1
250 100 20
z [nm]
z [nm]
z [nm]
0 0 0
−250 −100 −20

−500 −250 0 250 500 −200 −100 0 100 200 −40 −20 0 20 40
x [nm] x [nm] x [nm]
(a) (b) (c)
Fig. 3.12 Magnetic-field (contour plot) and electric-field (field lines) distribution at the metal–air boundary for (a) ω/ωp = 0.25,
(b) ω/ωp = 0.5, (c) ω/ωp = 0.7
30 20
200 80 15
20
10
100 40
10
5
z [nm]
z [nm]
z [nm]
z [nm]
0 0 0 0
−5
−10
−100 −40
−10
−20
−80 −15
−200
−30 −20
x=0 x=0 x=0 x=0

(a) (b) (c) (d)
Fig. 3.13 Poynting vector in the vicinity of the metal–air interface for (a) ω/ωp = 0.25, (b) ω/ωp = 0.5, (c) ω/ωp = 0.685, (d)
ω/ωp = 0.7. Lossless case
3.3.5 Poynting vector: lossless and lossy

Knowing all field components we can easily calculate the magnitude
and direction of the energy flow in the two media using the definition
for the Poynting vector (Section 1.13). Without losses, there can be no
Poynting vector component across the interface. Along the interface the
x component of the Poynting vector may be written as
1
Sx = − Re Ez Hy∗ .

(3.29)
2
With the aid of eqns (1.69) and (1.70) it takes the form
kx kx
Sx1 = |B|2 e 2κ1 z and Sx2 = |B|2 e −2κ2 z , (3.30)
2ωεr1 2ωεr2
30 20
200 80 15
20
10
100 40
10
5
z [nm]
z [nm]
z [nm]
z [nm]
0 0 0 0
−5
−10
−100 −40
−10
−20
−80 −15
−200
−30 −20
x=0 x=0 x=0 x=0

(a) (b) (c) (d)
Fig. 3.14 Poynting vector in the vicinity of the metal–air interface for (a) ω/ωp = 0.25, (b) ω/ωp = 0.5, (c) ω/ωp = 0.685, (d)
ω/ωp = 0.7. Lossy case
where Sx1 and Sx2 are the x components of the Poynting vector in the
dielectric and in the metal, respectively. Taking εr1 = 1 the distribution
of the Poynting vector is shown in Figs. 3.13(a)–(d) both in air and in
the metal for ω/ωp = 0.25, 0.5, 0.685 and 0.7. It is interesting to note
that practically all the power is flowing in air at ω/ωp = 0.25. The
corresponding wave vector is close to the light line. As ω/ωp increases
some of the power flows in the metal, oddly enough in the opposite
direction. In the vicinity of the asymptote at ω/ωp = 0.7 the flow of
power flowing in the positive and negative directions is nearly the same.
There is very little net power flowing. If we want to find the net power of
the SPP moving in the positive x direction we need to integrate Sx1 for
z from −∞ to zero, and Sx2 from z = 0 to z = ∞. The integration can
be easily performed. The final result after some algebra can be found as
kx ε2r2 − ε2r1
Net power = |B|2 . (3.31)
4ωκ1 ε2r1 ε2r2
When εr1 = |εr2 | there is no net power flow at all. Otherwise, the net
power is in the forward direction. We have a forward wave as follows
anyway from the dispersion curve.
Why is there no Poynting vector component in the z direction? Both
Ex and Hy are non-zero hence the normal component of the Poynting
vector
1
Re Ex Hy∗

Sz = (3.32)
2
should exist. It does not for the reason that Ex and Hy are 90 degrees
out of phase, and their product has zero real part.12 However, in the 12
The same picture is known to occur
presence of losses (γp = 0.01ωp) Sz must be finite. The Poynting vector in the case of total internal reflection.
−1 −0.5 0 0.5 1
500 200 40
z [nm]
z [nm]
z [nm]
0 0 0
−500 −200 −40

−1000 −500 0 500 1000 −400 −200 0 200 400 −80 −40 0 40 80
(a) (b) (c)
Fig. 3.15 Magnetic-field (contour plot) and electric-field (streamlines) distribution at the metal–air boundary. Lossy case
must cross the boundary between the dielectric and the metal in order to
replenish the absorption occurring in the metal. This is shown in Figs.
3.14(a)–(d) for the same set of frequencies. An obvious consequence of
losses is that the wave attenuates as it propagates in the x direction.
Figures 3.15(a) and (b) show again (as in Fig. 3.12) the contour plots of
Hy in the xz plane, and the field lines of the electric field. The strong
decline of the magnetic field at ω = 0.7ωp may be clearly seen. There
is a bright spot between x = −80 nm and −40 nm but one period away
the intensity has so much declined that the maximum is hardly visible.
Note also that the field lines, just as the streamlines of the Poynting
vector, are slanted.
3.4 Surface plasmon–polaritons on a slab:

TM polarization
So far we have had a single interface between two semi-infinite media.
Now we consider a metallic slab of dielectric constant ε2 and thickness
d in the z direction (see Fig. 3.16), and infinitely large in the other two
Dielectric
directions. It is sandwiched between two homogeneous isotropic semi-
Metal d infinite media of dielectric constant ε1 . If the two surfaces are far away
from each other then each one is unaware of the existence of the other
Dielectric
one and no new phenomena occur. However, if the slab is sufficiently
thin then the SPP at one of the surfaces ‘feels’ the existence of the SPP
at the other surface. A ‘repulsion of levels’ occurs, and the dispersion
Fig. 3.16 Metallic slab surrounded by
curves for the SPP localized at each of the two interfaces become split
two identical semi-infinite dielectrics
due to the interaction of these waves (Zayats et al., 2005).
3.4.1 The dispersion equation

We have already considered the oblique incidence of a TM wave upon a
slab in Section 1.11 and have already looked upon the transfer function
with ‘perfect imaging’ in mind. Our interest now is to derive the disper-
2
1 10
kp d=0.25
0.8
0.6 1
Hr =-1
kx /k p
10
Z/Z p
0.4
kp d=0.5
kp d=1
02
0 Hr =-2
10
0
0 2 4 6 8 10
0 50 100 150
kx /k p
d [nm]
(a) (b)
Fig. 3.17 (a) Dispersion curves for thin slabs for kp d = 0.25, 0.5 and 1. (b) kx versus thickness for two different values of ω
corresponding to εr = −1 and −2
sion equation. Like in the single-interface case we wish to find a solution

with non-zero components Hy , Ex and Ez . The relationship between the
input wave with amplitude A and the output wave with amplitude F is
given by eqn (1.76). We have an eigensolution when there is an output
without an input. This occurs when the denominator of eqn (1.76) is
zero,13 13
The presence of a pole in expressions
for field amplitudes is a typical feature
(1 + ζe )2 − e −2κ2 d (1 − ζe )2 = 0 ,
of an eigenmode of a system. An alter-
(3.33) native derivation, based on Maxwell’s
where ζe = κ2 ε1 /(κ1 ε2 ). The two solutions are equations, is given in Appendix G.
(1 + ζe ) = ± e −κ2 d (1 − ζe ) , (3.34)
representing two modes guided by the metallic slab (in which the wave,
now aware of both surfaces, travels with one single velocity), one sym-
metric and one antisymmetric with respect to their field distributions.
As far as the dispersion curve is concerned it may be expected that the
unperturbed dispersion curve (i.e. the one for the single interface) will
split. The two branches of the dispersion equation given by eqn (3.34)
can be expressed in more concise form as
κ2 d
ζe = − tanh , (3.35)
2
and
κ2 d
ζe = − coth . (3.36)
2
These transcendental equations can be easily solved numerically. As
shown above there are two solutions, first reported by Oliner and Tamir
(1962). We shall denote the upper branch of the solution by ω (+) and
0.8
0.6
Z/Z p
0.4
0.2
0
Fig. 3.18 Electrostatic approximation 0 2 4 6 8 10
kx/k p
and full solution for a slab. kp d = 0.25
the lower branch by ω (−) . In Fig. 3.17(a) we plot the upper and lower
branches for εr1 = 1 and for three different thicknesses (kp d = 0.25, 0.5
and 1) together with the solution for the single interface (d → ∞) that
is plotted with the dotted line. The meaning of the adjectives ‘upper’
and ‘lower’ make sense because they are above and below the dotted
line. For large kx both curves tend towards ωs , the same limit as for the
single interface. For small kx both types of dispersion curves tend to
the light line. We also find, not unexpectedly, that the split between the
modes becomes smaller as the slab becomes thicker. The propagation
coefficient is plotted in Fig. 3.17(b) as a function of slab thickness for
two different values of ω corresponding to εr = −1 and −2. It may be
seen that for a large enough thickness only one mode may exist, or even
14
It makes good sense that the surface none. The two surfaces are then uncoupled.14
modes that exist on a single interface at It is always nice to find analytical approximations. They give a sense
a given ω will split when the two sur-
faces are coupled. It is, however, less
of security. We know what is going on. The first one presented is
obvious that in the presence of coupling the electrostatic approximation that we have already come across in
the so far prohibited territory between Section 2.12.4. It is described in Appendix F. For the present case we
ωs and ωp may also be inhabited by find the two branches of the dispersion equation in the form
a new mode, the upper branch of the
dispersion curve. Physical considera-
ωp2
1 ± e −kx d .

tions would suggest that such a mode is
ω2 = (3.37)
permissible because the basic condition 2
εr2 < 0, is still valid, allowing charges
to accumulate at the surface. For thin enough slabs the electrostatic approximation turns out to be
extremely good for the lower branch, as may be seen in Fig. 3.18. It is
good too for the upper branch until it reaches the light line but then
it fails miserably. It cannot possibly follow the light line. This is of
course not surprising. When the wave velocity is close to the velocity of
light then nobody would expect an electrostatic approximation to work.
Interestingly, the approximation for the lower branch is still quite good
in the vicinity of the light line, although for sufficiently low values of kx
the approximation is bound to fail. How the approximation deteriorates
for the lower branch as well as for low values of kx may be seen in Figs.
3.19(a)–(c) for three different thicknesses. The approximate curve is
denoted by dashed lines. It may be seen that firstly the approximation
crosses the light line and secondly that the approximation becomes worse
as the thickness increases.
0.5
Z(−) (electrostatic) light
0.4 light line
light
line Z(−) (electrostatic) line
0.3 Z(−) (electrostatic)
Z/Z p
(−)
Z
0.2 Z(−)
Z(−)
0.1
0
0 0.2 0.4 0.6 0.8 10 0.2 0.4 0.6 0.8 1 0 02 0.4 0.6 0.8 1
kx/k p kx/k p kx/k p
(a) (b) (c)
Fig. 3.19 (a) Electrostatic approximation and full solution for a slab. kp d = 0.25, 0.5, 1 (a)–(c)
Two other approximations for small kx were first given by Economou

(1969). The derivation is given below. For small enough kx we may
expect the solution to be close to the light line, i.e. we may assume
kx2 = k02 (1 + δ 2 ) , (3.38)

2
where δ ≪ 1. It follows then that
κ2 = kp and κ1 = δk0 . (3.39)

We may then determine δ by substituting eqn (3.39) into eqn (3.35)
yielding

ω kp d
δ= coth , (3.40)
ωp 2
which leads to the approximate dispersion equation
" #
ω2

2 2 2 kp d
kx = k0 1 + 2 coth . (3.41)
ωp 2
If δ → ∞, i.e. we are back to the single interface, then the approximate

curve is
ω2 ω4
kx2 = + . (3.42)
c2 c2 ωp2
Note that eqn (3.41) would further simplify when kp d/2 ≪ 1 to
ω2 4 ω4
kx2 = 2 + 2 4 . (3.43)
c d ωp
It can be immediately seen that the smaller d is the farther is the dis-
persion equation from the light line for a given ω. Next, we shall look
for an approximation for the upper branch. We shall assume again that
)
Z)
Z
w
w
(lo
(lo
1
)

Z (+
Z | d
Z(+)
ne
0.8
li
Z|
ht
d
lig
0.6
Z)
Z/Z p
() (low
Z Z( )
0.4
0.2
0
Fig. 3.20 Full solution and various ap-
0 0.5 1 1.5 2
proximations for a slab. kp d = 0.5
kx/k p
kx is small and close to the light line so that we can start again with
eqn (3.38). Then, using the same technique we find

ω kp d
δ= tanh , (3.44)
ωp 2
leading to
" #
ω2

kp d
kx2 = k02 1 + 2 tanh2 , (3.45)
ωp 2
or after expanding the tanh function to
ω2 ω 4 d2
+ kx2 =. (3.46)
c2 4c4
It may now be seen that the smaller d is the nearer is this branch of the
dispersion equation to the light line.
Using the above-derived approximations for ω (−) and ω (+) we plot
in Fig. 3.20 the approximate and exact curves for kp d = 0.5. The ap-
proximation is very good for small kx up to about ω/ωp = 0.35 and
deteriorates afterwards.
So far we have looked at the lossless case but of course attenuation is
equally important if we have practical applications in mind. We need
to find not only the ω versus kx′ curve but also ω versus kx′′ . How will
kx′′ enter our equations? It will be through κ2 , which in the presence of
losses will take the form
κ22 = (kx′ − j kx′′ )2 + k02 (ε′ − j ε′′ ) . (3.47)

15
We used an iterative technique first
We need to substitute κ2 from above into eqns (3.33) and (3.34) taking
assuming the value of kx′ from the loss-
less solution and finding kx′′ , and then good care that ζe depends both on κ2 and on εr2 . We are not aware
for this value of kx′′ finding the corre- of any analytical approximations for this case. We need to resort to
sponding value of kx′ , etc. numerical solutions.15
Z(+)
Z(−)
0.6
k d=0.25
p
0.4 k d=0.5
p
p
Z/Z
kpd kpd
kpd=1 kpd=1
0.2
k d=0.5
p
kpd=0.25
−5 0 −5 0
10 10 10 10
kx"/kp k "/k
x p
(a) (b)
Fig. 3.21 Attenuation coefficient versus frequency for ω (+) -branch (a) and ω (−) -branch (b)
We shall not replot the frequency against normalized propagation con-

stant curve—which remains practically the same. The frequency against
normalized attenuation coefficient is plotted in Figs. 3.21(a) and (b) for
the ω (+) and the ω (−) branches, respectively, for a range of normalized
slab thicknesses. It may be seen that the waves in the upper branch are
much less attenuated than the waves in the lower branch. In Fig. 3.22 we
shall return to the concept of propagation length. It is plotted against
the free-space wavelength for the ω (+) branch. It may be clearly seen
that as the thickness declines, the propagation length increases. The
reason why attenuation is lower for the upper branch is far from being
obvious. We shall show first the field distributions and then discuss in
quite some detail the physical reasons why the upper branch deserves
the epithet of ‘long range’.
3.4.2 Field distributions

Let us now see the field distributions at one of these thicknesses (kp d =
0.25) for ω/ωp = 0.3, 0.5 and 0.7. They are plotted in Figs. 3.23(a)–
(c). Let us first look at the symmetry. For the upper mode Ez and
Hy (and, consequently, Sx ) are symmetric as a function of z, whereas
Ex is antisymmetric. For the lower mode Ez and Hy are antisymmetric
(consequently, Sx is symmetric) and Ex is symmetric. It is also true, as
we have already known for the single interface, that the decay away from
the interfaces becomes stronger as ω/ωp increases. It also follows from
the figures that Sx is positive in the dielectric and flows in the opposite
direction in the metal. As ω increases towards ωs the net power flow
tends to zero. We need to emphasize that for a given ω up to ωs we
had two different values of kx belonging to two different branches of
0
10 d=10nm 20nm
60nm
propagation length [m]
−2 100nm
10
d
−4
10
free-space O f
−6
10
Fig. 3.22 Propagation length versus
0 2 4 6 8 10
O f [Pm]
free-space wavelength
the dispersion curve. Above ωs the situation is different. Although

for a given value of ω there are still two values of kx , they happen to
belong to the same branch, to the upper branch. The field distribution
corresponding to ω/ωp = 0.735 is shown in Fig. 3.23(d). The symmetries
are now the same at both values of kx : the fields Ez , Hy and the Poynting
vector Sx are symmetric and Ex is antisymmetric. As we may also
expect, the decay of the fields away from the metal is steeper for the
higher kx . A good look at Sx will show (and numerical integration will
prove it) that the total power carried is now in the opposite direction.
The wave is a backward wave.
As we have seen, a field distribution is either symmetric or antisym-
metric. So terminology should be easy: those that have symmetric dis-
tribution should be called symmetric modes and denoted by s, and those
with antisymmetric distribution could be called antisymmetric (or asym-
metric) modes and denoted by a. The problem is which field component
should we refer to? Is it Ez , the field component normal to the direction
of propagation, or Ex the field component in the direction of propaga-
tion? Opinions have been divided. Economou (1969) refers to the lower
branch, ω (−) , as that with symmetric oscillation and to the upper branch
as that with antisymmetric oscillation. For him it is the charge distri-
bution, and the effect of the tangential component of the electric field
on it, that is of primary importance. Burke et al. (1986) call the upper
branch symmetric and the lower branch antisymmetric. According to
Welford (1988), in the upper branch the two surface waves propagate
π out of phase and hence should be referred to as the antisymmetric
mode. In the branch with the lower ω the surface waves propagate in
phase and should therefore be referred to as the symmetric mode. Burke
et al. (1986) attributed the conflict in notations to the different back-
ground of the authors. The terminology in which the antisymmetric
mode has a zero in its transverse electric field inside the film comes from
integrated optics and is just opposite to the solid-state version, which
Z (+) Ex Z (−) Z (+) Z (−)

Ex Ex Ex
Ez Ez
Ez Ez
Hy Hy Hy
Hy
Sx Sx Sx
Sx
(a) (b)
Z (+) Z (−) Z (+) Z (+)

Ex Ex Ex Ex
Ez Ez
Ez Ez
Hy Hy
Hy Hy
Sx Sx
Sx Sx
(c) (d)
Fig. 3.23 Field distributions of ω (+) and ω (−) -branches for four typical frequencies (a)–(d) ω/ωp = 0.3, 0.5, 0.7, 0.735
deals with the symmetries of the charge distribution and therefore with 16
Maier et al. (2005) use a notation
the longitudinal components of the electric field. By now it has been somewhat at variance with the usual
largely accepted16 that the mode in which the normal component of the notions of symmetry and antisymme-
electric field is symmetric is denoted by s and that in which the normal try. They refer to the ω (+) -mode
as ‘antisymmetric (sb ) mode’ and to
component is antisymmetric is called a. In addition, when the waves are the ω (−) -mode as the ‘symmetric (ab )
bound the modes are referred to as sb and ab and when they are leaky17 mode’.
as sl and al .
If the metal is lossless the modes are lossless. If the metal is lossy 17
Leaky waves, as the name implies,
the modes are lossy, so much is clear. But which one is lossier? The leak out power, which is good for anten-
calculations show that the sb mode is less lossy, meriting the description nas but not for guiding the wave. Here,
we are concerned only with bound
of ‘long-range surface plasmon’. We shall show curves later but first let modes so, with regret, we shall not dis-
us try to give a physical explanation. cuss the properties of leaky waves.
One explanation may be based on the fact that the ω (+) -mode is nearer
to the light line. If it is nearer it resembles more an electromagnetic wave.
Electromagnetic waves travel in vacuum with no attenuation, hence the
upper branch will be the one that has the lower attenuation.
Mansuripur et al. (2007) offers a somewhat more complicated expla-

nation for the loss mechanism. With the even mode, the field component
Ez has the same sign on the boundaries of the slab; therefore at a given
point along x, the electrical charges on the upper and lower surfaces
have opposite signs. Inside the metallic slab, the field component Ez —
reduced by a factor εr2 /εr1 relative to the Ez immediately outside—
helps move the charges back and forth between the two surfaces. The
slab being thin, the transport distance is short; hence the correspond-
ing electrical current is small. In contrast, the charges of the odd mode
have the same sign on opposite sides of the slab. Consequently, positive
and negative charges must move in the x direction during each period
of oscillation. The travel distance is now of the order of the SPP wave-
length, which is greater than the slab thickness. Therefore, the current
densities of the odd mode are relatively large, leading to correspondingly
large losses.
A fairly similar reasoning is due to Zayats et al. (2005) giving the Ex
component as being responsible for loss. For the ω (+) mode Ex vanishes
at the midplane of the slab, while it is a maximum at that plane for
the ω (−) mode. The mode with the smaller fraction of the electric field
inside the metal causes less dissipation.
Barnes (2006) explains the loss mechanism as follows. The ω (+) mode
propagates in such a way that less of the power is carried in the metal
than in the case of a single-interface mode. This reduces the effect of
loss. Propagation lengths of the order of centimeters may be achieved.
There is though a price to pay for such long propagation: the mode is not
localized at the surface, it extends further into the dielectric, resembling
the transverse plane electromagnetic wave.
Maybe the best explanation is based on the penetration depth in the
metal. It may be seen in Figs. 3.23(a)–(c) that for a given frequency
the antisymmetric mode at the higher kx has a fast decay away from
the metal surface. The symmetric mode has a slower decay, hence more
field is in the dielectric, consequently, it is less lossy.
Another consideration is confinement of the fields. The more they are
confined, the larger they are and that’s a good thing for applications
in which high fields are required. On the other hand, as the arguments
above have shown, confinement leads to higher losses. Hence, some
compromise between the two must always be made. For a discussion of
the trade-off see Berini (2006).
H3
3.4.3 Asymmetric structures
Dielectric
We use here the terms symmetric and asymmetric in yet another sense,
H2 Metal d referring this time to structures. A symmetric structure is when the
H1 Dielectric dielectrics on the two sides of the metal slab are identical and it is
asymmetric when the dielectrics are different (see Fig. 3.24).
The asymmetric structure was investigated in some of the early papers
Fig. 3.24 Metallic slab surrounded by (Sarid, 1981; Wendler and Haupt, 1986; Burke et al., 1986; Yang et al.,
different dielectric media 1991) but received much less attention than the symmetric case. If there
103 Ag
O=632.8 nm a=10 nm
17 nm
30 nm
102 50 nm
Propagation length L (cm)
101
100
10–1
10–2
Fig. 3.25 Propagation length of the
long-range SPP vs. dielectric constants
εr1 of the superstrate and εr3 of the
1.9 2.0 2.1 2.2 2.3 2.4 substrate for different thicknesses a of
the Ag layer. From Wendler and Haupt
H r1 H r3 = 2.1211) H r1 H r3 H r3 H r1 = 2.1211) (1986). Copyright c 1986 American
Dielectric constants Institute of Physics
is only a small difference in the dielectric constant then we can still talk
about quasi-symmetric and quasi-antisymmetric modes but we find that
the field confinement is very asymmetric. The quasi-symmetric mode is
localized at the boundary between the metal and the dielectric with the
lower dielectric constant, whereas the quasi-antisymmetric mode is lo-
calized at the higher dielectric constant boundary. The quasi-symmetric
mode inherits the mantle of the sb mode: it is still the long-range mode,
it still has a lower attenuation and, as may be expected, the attenua-
tion decreases as the slab thickness decreases. The new phenomenon is
that the quasi-symmetric mode has a cutoff thickness below which no
propagation is possible. This is bound to occur (Berini, 2001) because
the sb mode no longer has the chance quietly to convert into the TEM
electromagnetic wave as d → 0 because the different dielectrics on the
two sides do not permit it. The quasi-antisymmetric mode remains the
short-range mode, its attenuation increases with decreasing slab thick-
ness, it has not got a cutoff thickness.
For detailed calculations of the propagation length in asymmetric
structures see Wendler and Haupt (1986). They investigate theoreti-
cally thin silver slabs at the He-Ne wavelength of 632.8 nm with a di-
electric constant of εr2 = −18−j 0.47 obtained from Johnson and Christy
(1972). They first assume that both dielectrics have identical dielectric
constants of εr1 = εr3 = 2.1211, then change the dielectric constant of
one of them within a range of about 20%. They find that the propaga-
tion length first remains unchanged and then suddenly increases. Their
results, quite spectacular, are shown in Fig. 3.25.

In a further study Zervas (1991) showed that long-range SPPs are
supported even by highly asymmetric configurations.
3.5 Metal–dielectric–metal and periodic

structures
It is now an easy exercise to derive dispersion for the surface modes for
a thin dielectric layer with two semi-infinite metallic media adjacent to
it. The dispersion equations are very similar to those of eqns (3.35) and
(3.36) and can be derived in an analogous manner. In the final result
the only difference is that κ2 is being replaced by κ1 in the arguments
of the tanh and coth functions, i.e. the equations are given as
κ1 d
ζe = − tanh , (3.48)
2
and
κ1 d
ζe = − coth . (3.49)
2
As before, we shall try to find some approximations. For the upper
branch ω (+) we can take the argument of the coth function small and
then without further approximations we find
2 " 2 # " ω 2 k d 2
#
2 ω ω p
4
p p
kx2 = 1− − − . (3.50)
d ωp ω ω 2
The frequency at which kx = 0 may be obtained from the above equation

as
ω 1
= 1 − (kp d)2 , (3.51)
ωp 8
i.e. the dispersion curve cuts the vertical axis just below the bulk plasma
frequency.
We may again resort to the electrostatic approximation that turns out
to be identical with that (eqn (3.36)) already derived for the dielectric–
metal–dielectric structure.
The dispersion curves calculated from eqns (3.48) and (3.49) are plot-
ted in Figs. 3.26(a) and (b) together with the electrostatic approximation
and that given by eqn (3.50) for kp d = 0.25 and 0.5. The first thing to
notice is that the ω (+) branch crosses the light line. The wave can prop-
agate at phase velocities higher than the velocity of light. The ω (−)
branch behaves as it did for the dielectric–metal–dielectric structure. It
tends to ωs for kx → ∞ and to the light line as kx → 0. The electrostatic
approximation is quite good for the ω (−) mode and even better for the
ω (+) mode. The low kx approximation, not surprisingly, deteriorates as
kx increases.
Z)
(low−
1 1
Z (−)
Z (+) (e
l.-stat.) Z (+) (e
l.-stat.)
e
Z (+) (lo Z (+)
e
0.8 0.8
light lin
light lin
w−Z) Z (+)
Z |d at .)
l.-st Z (+) (lo
− Z)
0.6 Z(−) 0.6 (−) (e w−Z)

at .) Z
l.-st
(low
p
(−) (e
Z Z(−)
/
/
Z (−)
0.4 0.4
0.2 02
0
0 2 4 6 8 10 0 2 4 6 8 10
kx/kp kx/kp
(a) (b)
Fig. 3.26 Dispersion curves for a metal–dielectric–metal structure. Full solutions and approximation for (a) kp d = 0.25 and
(b) kp d = 0.5
Economou (1969) goes on investigating two and more metal slabs

between dielectrics, including the generalization to a periodic structure
(see Fig. 3.27(a)). More and more layers in the structure lead to multiple
splitting, i.e. to the appearance of more and more dispersion curves with
the final result that both the ω (+) and the ω (−) branches widen into
bands, as shown schematically in Fig. 3.27(b).
3.6 One-dimensional confinement: shells

and stripes
All the problems we have investigated so far have been concerned with
two-dimensional confinement of the surface waves, to a single surface
or to multiple surfaces. Clearly, for any practical application we need
a one-dimensional structure to take information from point A to point
B. Such a structure was analyzed by Al-Bader and Imtar (1992). They
considered a cylindrical metallic shell of inner radius a, and thickness
t embedded in a dielectric or in two different dielectrics. They set up
the differential equations for the fields of a TM mode and solved them
numerically for silver at the He-Ne wavelength of 633 nm. They pre-
sented their results in terms of a mode index ne = n′e − j n′′e where the
complex propagation coefficient is given as k = k0 ne . Their results were
similar to those found for planar structures to which they reduced in the
limit of infinite radius. They were also interested in the evolution of the
TM01 fibre mode into surface plasmons, the model they considered was
a dielectric cylinder embedded in metal. They found that the surface
2
2 k
ck
C
2
p II a
2
Metal
Dielectric p
I II b
p/2
Cmax
IV
Cmin
III
O kp k
(a) (b)
Fig. 3.27 Multilayer metal–dielectric structure. (a) Sketch of the configuration. (b) Dispersion relations. From Economou
c 1969 by the American Physical Society
(1969). Copyright
mode appeared for sufficiently large radius.

The physics for cylindrical structures is quite similar to that of planar
structures. However, a host of new phenomena appear when the infinite
slab is turned into the one-dimensional metal stripe of width w and
thickness t, as shown in Fig. 3.28(a). They have received considerable
attention lately (Berini, 1999; Berini, 2000a; Berini, 2000b; Berini, 2001;
Berini, 2006; Al-Bader, 2004) due to their promise of signal processing
in the visible and infra-red region. They solved the relevant partial
differential equations subject to the boundary conditions by the method
of lines. The main difference they found relative to the 2D case was
the variety of field distributions that could occur. Simple TM modes
no longer exist. All six field components must be present in all modes.
The question then arises as to what nomenclature to use to identify the
modes. The obvious start (Berini, 1999; Berini, 2000a) is with the sb and
ab modes of 2D structures because the symmetry of the 1D structure
will again permit symmetric and antisymmetric modes, in fact there are
two symmetry axes in the x and y directions of Fig. 3.28(a). When
w/t ≫ 1 then the main transverse electric field is Ez whose symmetry
with respect to the y and z axes is reflected in the use of the subscripts
ss, sa, as, aa. A further superscript is then used to track the number
of extrema observed in the spatial distribution of Ez along the y axis,
and a second superscript would describe the extrema in the z direction
(likely to exist but they have not been found so far). They are all bound
modes, hence the subscript b is also added as for the 2D case. The
new feature is that the higher-order modes have a cutoff width below
z
H x
t H y
w
(a)
0
10
3.6
sa0b , aa0b
10 –1
3.2
aa 2b , sa1b , a b
EE0
DE 0
2.8 sa0b , aa0b –2

10
ss0b sb
2.4 aa 2b sa1b a
b
10 –3
sb
2 1 as2b sb
ss0b as 0b ssb as0b ss1b asb2
10 –4
0 0.04 0.08 0.12 0.16 0.2 0 0.04 0.08 0.12 0.16 02
Thickness t (Pm) Thickness t (Pm)
(b) (c)
Fig. 3.28 Metallic stripe. (a) Sketch of the geometry. (b), (c) Dispersion with thickness of the first six modes. Also shown the
ab and sb modes for w → ∞ for comparision. From Berini (2000b). Copyright c 2000 by the American Physical Society
which they cannot propagate and some of the modes also have a cutoff
thickness.
For a stripe width of w = 0.5 µm the normalized propagation and
attenuation coefficients are plotted in Figs. 3.28(b) and (c) for the first
six modes as a function of stripe thickness. The sb and ab modes of
the 2D structure are also shown for comparison. As may be expected,
the antisymmetric modes have higher attenuation than the symmetric
modes. The most relevant conclusion is that the 1D stripe geometry
is in no way worse than the amply analyzed 2D geometry. In fact, it
may be much better, at least in theory. The attenuation of two of the
higher modes may be seen to decrease rapidly before they reach cutoff.
Presumably, it happens at the expense of confinement. The ss0b mode
has probably more practical significance. If we compare it with the sb
mode at the lowest thicknesses shown, below 20 nm, we find that the
attenuation of the stripe mode is very considerably below that of the 2D
planar mode, and the attenuation can be further reduced by decreasing
the stripe width.
Stripes having different dielectrics for substrates and superstrates have
also been investigated by Berini (2001). Small differences in the upper
and lower dielectric constants have been found to lead to larger differ-
ences in propagation properties. There has been no reply given as yet to
the question posed by the very large propagation lengths (see Fig. 3.25)
obtained theoretically by Wendler and Haupt (1986), whether they can
be realized in stripe form.
3.7 SPP for arbitrary ε and µ

Looking at the surface modes at interfaces of two media we have so far
assumed that one of the media is a dielectric with µ = 1 and ε being
positive and frequency-independent and that the other one is a metal.
The condition of existence of a surface TM mode at a single metal–
dielectric interface was given as
εr2 /εr1 < −1 , (3.52)

which, assuming the Drude model for a metal, is satisfied below the
surface plasma frequency,
ωp
ω< √ . (3.53)
1 + εr1
In the case of a metal slab, the SPP modes excited at the two surfaces
interact, the dispersion splits into two, one below and another one above
the unperturbed dispersion curve. The resulting condition for the exis-
tence of surface modes is much more relaxed. In the limit of an infinitely
thin slab it is
εr2 /εr1 < 0, (3.54)

meaning that
sign(εr2 ) 6= sign(εr1 ) , (3.55)

and the upper SPP branch can reach up to the bulk plasma frequency
ω < ωp . (3.56)
3.7.1 SPP dispersion equation for a single interface

We will now generalize these results to the case when ε and µ can take
arbitrary values. At this stage we are not concerned with the question
of how or whether these values can be realized. For an early treatment
see Ruppin (2000). The derivations to follow are based on the work of
Darmanyan et al. (2003). The mathematical problems of solving the
field equations subject to the boundary conditions are not particularly
difficult. There is, however, a problem with terminology. Allowing now
the possibility of negative permeability what should we call them? Some
18
The analogy is not very good because people talk about magnetic plasmons,18 which gives the opportunity to
there are no magnetic charges. call them electric and magnetic surface plasmon–polaritons but that is
quite a mouthful and it is unlikely that it would catch on. So we just
decided to stick to the original name and in spite of the changes in
permeability we shall still call them surface plasmon–polaritons.
Having done the derivation for the case when the permittivity could
take negative values it is an easy task to derive the dispersion equations
for the surface waves in the general case. The approach is the same.
The results are though more far reaching and will have significant impli-
cations for the imaging mechanism of the ‘perfect’ lens to be discussed
in Chapter 5.
We disregard losses for simplicity. They can be included later if
needed. Looking for surface-wave solutions declining exponentially we
require the z components of the k vector to be purely imaginary so that
both
q
κ1 = kx2 − εr1 µr1 k02 (3.57)
and
q
κ2 = kx2 − εr2 µr2 k02 (3.58)
are to be positive and real. The dispersion equation is formally the same
as in the case of a metal. It may be obtained from the poles of the field
solutions. For the TM polarization it is
ζe + 1 = 0 (TM case) (3.59)

or
κ2 εr1
= −1 (TM case) , (3.60)
κ1 εr2
whereas for the TE polarization we obtain
ζm + 1 = 0 (TE case) (3.61)

or
κ1 µr2
= −1 (TE case) . (3.62)
κ2 µr1
Inserting eqns (3.57) and (3.58) into eqn (3.60) we obtain for the TM
case
(kx2 − εr2 µr2 k02 )ε2r1 = (kx2 − εr1 µr1 k02 )ε2r2 (TM case) , (3.63)
leading to the dispersion equation
εr1 εr2 µr1 εr2 − µr2 εr1

kx2 = k02 (TM case) . (3.64)
εr1 + εr2 εr2 − εr1
Similarly for the TE case, inserting eqns (3.57) and (3.58) into eqn (3.62)
we obtain the dispersion equation
µr1 µr2 εr1 µr2 − εr2 µr1

kx2 = k02 (TE case) . (3.65)
µr1 + µr2 µr2 − µr1
For the z components of the k vector in the TM case we obtain by using

eqns (3.57), (3.58) and (3.64)
εr1 εr2 µr1 εr1 − µr2 εr2

κ21 = k02 ε2r1 (TM case) (3.66)
εr1 + εr2 ε2r2 − ε2r1
and
κ22 = κ21 ε2 (TM case), (3.67)

and in the TE case by using eqns (3.57), (3.58) and (3.65)
µr1 µr2 εr1 µr1 − εr2 µr2

κ21 = k02 µ2r1 (TE case) (3.68)
µr1 + µr2 µ2r2 − µ2r1
and
κ22 = κ21 µ2 (TE case) , (3.69)

where ε = εr2 /εr1 and µ = µr2 /µr1 .
3.7.2 Domains of existence of SPPs for a single

interface
It follows immediately from eqn (3.60) that, as both κ1 and κ2 are to
be real and positive, for TM modes to exist the condition
εr2 εr1 < 0 (TM case) (3.70)

must be satisfied. For TE modes, as follows from eqn (3.62) the condition
is
µr2 µr1 < 0 (TE case) . (3.71)

The result for the TM polarization makes good sense. We are talking
here about the electric field with its Ez component changing sign across
the boundary, a condition compatible with accumulation of charges at
the surface, as we discussed earlier for metals. It is important to note
that this condition holds for any value of µ: surface TM modes exist
only if ε changes sign at the boundary.
The same argument could be applied for the TE polarization, but it
is somewhat difficult to provide a simple physical interpretation of the
requirement that µ1 and µ2 must be of different sign for a TE surface
mode to exist. The reason why it is difficult to picture what is going
on is that in the case of negative permeability we cannot appeal to our
common-sense expectations. Negative-ε materials have received much
more attention in the past and the condition εr1 εr2 < 0 for the existence
of TM surface modes has been known ever since Fano’s article over sixty
years ago. It is quite different with negative-µ materials. All such mate-
rials, constructed so far, have negative permeability within a frequency
band and not below a certain frequency, as it happens with negative-

permittivity materials. Also, we cannot think of surface charges. Instead
we must consider surface currents, a less familiar concept, and it is par-
ticularly difficult to think about it when the structure is discrete.
Returning to the dispersion equations and taking into account that
for TM polarization εr2 < 0, eqns (3.64) and (3.66) can be rewritten as
|ε|(|ε| + µ)
kx2 = k02 εr1 µr1 (TM case) (3.72)
ε2 − 1
and
1 + |ε|µ
κ21 = k02 εr1 µr1 (TM case) . (3.73)
ε2 − 1
Similarly, taking into account that for TE polarization µr2 < 0, eqns (3.65)
and (3.67) can be rewritten as
|µ|(|µ| + ε)
kx2 = k02 εr1 µr1 (TE case) (3.74)
µ2 − 1
and
1 + |µ|ε
κ21 = k02 εr1 µr1 (TE case) . (3.75)
µ2 − 1
The TM and TE surface modes of eqns (3.72) and (3.74) exist if kx2 , κ21
and κ22 are real and positive. In the case of TM polarization, it follows
from eqn (3.73) that κ21 is positive if either
1
|ε| > 1 and µ>− (TM case) (3.76)
|ε|
or
1
|ε| < 1 and µ<− (TM case) . (3.77)
|ε|
Similarly, for TE polarization, the condition κ21 > 0 in eqn (3.75) is
fulfilled if either
1
|µ| > 1 and ε>− (TE case) (3.78)
|µ|
or
1
|µ| < 1 and ε<− (TE case). (3.79)
|µ|
If κ21 is positive then κ22 is positive as well, see eqns (3.67) and (3.69).
Similar considerations for the positiveness of kx2 (eqns (3.64) and (3.74)),
do not provide any further constraints for the existence of the surface
modes. The domains of existence of the TM mode and of the TE mode
are plotted in Fig. 3.29 in the plane (ε, µ).
The horizontal dashed line µ = 1 corresponds to the case of the second
medium being a metal with a TM surface mode for ε < −1 (εr2 < −εr1 )
being the only possibility.
4
TM
2
µ=1
1
µ=µ /µ
2
0
TE
−2
−4 TE
Fig. 3.29 Domains of TM and TE TM

modes for an interface dielectric– −6
−6 −4 −2 0 2 4 6
metamaterial with arbitrary values of ε=ε /ε
permittivity and permeability 2 1
It is worth mentioning that there can be no surface modes at the

boundary of two left-handed media with εr1 , εr2 , µr1 , µr2 all being neg-
ative (Darmanyan et al., 2003) (this corresponds to the first quadrangle
of the plane ε, µ).
It is important to note that if, in a certain frequency range, in medium
2 both εr2 and µr2 are negative, i.e. if medium 2 is a left-handed one
then a transverse electromagnetic (bulk) wave can propagate with the
dispersion equation
ω√
εr2 µr2 .
k= (3.80)
c
For a surface mode (with fields exponentially decaying away from the
interface) to exist, its wave vector must be larger than those of the bulk
modes in either media, resulting in the condition
n ω√ ω√ o
kx > max k0 = εr1 µr1 , k = εr2 µr2 . (3.81)
c c
This is a new feature that we have not met in the case of a metal where
surface modes were only found in the stop bands of the bulk modes.
In the case of a left-handed metamaterial with both ε2 and µ2 being
negative a bulk and a surface mode can coexist at the same frequency.
If their dispersion curves ω(k) intersect, only those parts of the surface
mode solutions that lie to the right of the bulk modes are meaningful.
3.7.3 SPP at a single interface to a metamaterial:

various scenarios
The results of the previous section are general and can be used to look
at possible SPP modes for any values of ε and µ. There is an obvious
symmetry regarding ε and µ on one side and TM and TE modes on the
other side. It can be easily seen that the results for TM and TE SPP
Table 3.1 Numerical values for the frequencies at which surface modes start or stop.
F = 0.56
ω0 ω|εµ=1 ω|µ=−1 ω|µ=0

ωp ωp ωp ωp (i)TM (ii)TE (ii)TM (iii)TM (iii)TE
forw. backw. backw. forw. forw.
0.4 0.52 0.47 0.6 yes yes — yes —

0.6 0.71 0.71 0.9 yes — — — —
0.8 0.86 0.94 1.2 yes — yes — yes
1.0 — 1.2 1.5 yes — — — yes
4 ε
µ
εµ=1
2
µ=0
ε, µ
0
ω
0
ε=0
−2 µ=−1
ε=−1
−4
Fig. 3.30 Frequency dependence of
0.2 0.6 1 1.4 1.8
ω/ωp
permittivity and permeability for F =
0.56, ω0 = 0.4 ωp
modes transform into each other by the mutual substitution ε ↔ µ. As

we will see in this section the symmetry is broken when εr2 and µr2 have
different types of frequency dependence.
Here, we will look at the dispersion curves of the possible surface
modes assuming that the dielectric constant of medium 2 follows the
Drude model,
ωp2
,
εr2 = 1 − (3.82)
ω2
and that the permeability has a resonant behaviour that can be described
as (see Section 2.8)
F ω2
µr2 = 1 − . (3.83)
ω 2 − ω02
These forms of εr2 and µr2 were frequently used in describing the re-
sponse of metamaterials comprising rods (Pendry et al., 1996) and SRRs
(Pendry et al., 1999). The choice of the parameters, of the plasma fre-
quency, ωp , of the magnetic resonance frequency ω0 and of the filling
factor, F , has a crucial effect on the behaviour of the metamaterial, in
1.6
6
ne
1.4
li
ht
lig
12 4
(i) TM
1 H=0
2
p
1
Z/Z
P=P /P
0.8 (iii) TM (iii) TM
2
H=−1 Q
C
0.6 P=0 0
HP=1 (ii) TE
P=−1
1 (ii) TE B
0.4 Z A
(i) TM 0
−2
02
0 P
0 0.5 1 1.5 2 2.5 3 −4
−8 −6 −4 −2 0 2
kx/kp H=H2/H1
(a ) (b )
Fig. 3.31 SPP dispersion for single interface. F = 0.56, ω0 = 0.4 ωp . (a) ω–kx diagram. (b) µ–ε diagram
general, and on the character of the surface modes, in particular. Let us

recall the most important asymptotes and limits for permeability and
permittivity,


 εr2 → 1 for ω→∞



 ε r2 > 0 for ω > ωp
εr2 : ε r2 < 0 for ω < ωp (3.84)
ω

 r2
 ε = −1 for ω = √p

 2
εr2 → −∞ for ω→0

and
µr2 → 1 − F for ω→∞


ω > √ ω0



 µr2 > 0 for
1−F


ω0 < ω < √ ω0

µ <0 for


 r2

√ 1−F

µr2 : . (3.85)
µr2 = −1 for ω = √ω0 2
2−F







 µr2 → ∓∞ for ω → ω0 ± 0
 µ
 r2

 >0 for ω < ω0
µr2 →1 for ω→0
We will now give examples of four quite different scenarios for SPP
modes using typical sets of parameters. In all examples we keep F =
0.56 and choose for ω0 /ωp values of 0.4, 0.6, 0.8 and 1. In all four
cases,
√ the frequency at which εr = −1 is the surface plasmon frequency,
ωp / 2 ≃ 0.71ωp and the frequency at which εr = 0 is of course the
plasma frequency, ωp . Table 3.1 gives numerical values for the frequen-
cies of interest from eqn (3.85) in the four cases.
1.6 6
ne
1.4
li
ht
lig
4
1.2
(i) TM
1 H=0
P=0 2
1
Z/Zp
P=P /P
0.8
2
H=P=−1, HP=1 Q
0.6 Z 0
0
(i) TM
0.4 A
−2
0.2
P
0 −4
0 0.5 1 1.5 2 2.5 3 −8 −6 −4 −2 0 2
k /k
x p H=H /H
2 1
(a ) (b )
Fig. 3.32 SPP dispersion for single interface. F = 0.56, ω0 = 0.6 ωp . (a) ω–kx diagram. (b) µ–ε diagram
Case (a) ω0 /ωp = 0.4. The variations of ε(ω) and µ(ω) as a function of
frequency are shown in Fig. 3.30. The question we are asking now is what
kind of surface modes may exist for this set of parameters. For simplicity
we shall take εr1 = µr1 = 1, although the equations are valid for the case
of a general lossless dielectric. It can be shown from eqns (3.76)–(3.79)
that there are three SPP modes, as may be seen in Fig. 3.31(a). The
first one is a TM mode that exists in the range 0 < ω < ω0 . Thus, the
presence of the magnetic resonance limits the range of the TM mode.
Instead of moving up to ωs it stops now at ω0 . In other words, the upper
frequency limit occurs where µr2 turns negative instead of εr2 = −1. The
second mode is a TE one that starts at the light line at a value of ω
where εr2 µr2 = 1 and tends asymptotically to the value of ω where
µr2 = −1. It is a backward wave. The third mode is again a TM one.
It starts at the same point as the TE mode but it moves upwards to
tend asymptotically to the εr2 = −1 line. Not shown in the figure is the
bulk mode that propagates for the range of frequencies for which both
εr2 and µr2 are negative.
Let us now return to our diagram of Fig. 3.29 showing the areas in
the εµ plane in which surface modes are possible. Each of our three
modes may be represented by a curve in this plane replotted in Fig.
3.31(b). For the lower TM mode µ is positive and ε is negative. As ω
varies from 0 to ω0 the curve denoted by (i) is described. As ω tends to
ω0 we know that the permeability tends to infinity and then suddenly
changes to minus infinity. Up to µ = ∞ there is a TM mode, but when
the permeability switches to −∞ the surface mode is no longer there.
As the frequency increases further the permeability now increases from
minus infinity. At a certain value of the frequency it reaches the point
P and then traverses a number of regions between P and Q. In the PA
region, as may be seen from the shading, there cannot be surface modes.
1.6 6
ne
1.4
li
ht
lig
P=0 4
1.2
(i) TM
1 H=0
P=−1 2
1
HP=1 (iii) TE
Z/Zp
P=P /P
0.8 Z0
2
(ii) TM Q
H= − 1
0.6 0
(i) TM
B
0.4
−2
0.2 (ii) TM A (iii) TE
0 P
−4
0 0.5 1 1.5 2 2.5 3 −8 −6 −4 −2 0 2
k /k
x p H=H /H
2 1
(a ) (b )
Fig. 3.33 SPP dispersion for single interface. F = 0.56, ω0 = 0.8 ωp . (a) ω–kx diagram. (b) µ − ε diagram
The TE region is entered at A and this mode exists until the frequency
is high enough to reach B. This is the mode we have denoted by (ii).
Increase ω again and we move from B to C, which is in the TM mode
region corresponding to mode (iii). The curve continues beyond C as ω
increases but no further surface modes are possible. It is interesting to
note that this novel presentation in terms of geometrical loci in the ε,µ
plane is not only capable of accounting for all the modes but also makes
it clear how the various surface modes arise.
Next, let us consider what happens when ω0 increases, i.e. the mag-
netic resonance moves up in frequency. It may be shown that both the
TE mode and the upper TM mode we have seen in Fig. 3.31(b) become
flatter and then disappear altogether. For case (b), when ω0 /ωp = 0.6
is taken, only the lower TM mode survives, as shown in the dispersion
curve of Fig. 3.32(a). The geometrical locus of this TM mode in the
εµ plane resembles the one in Fig. 3.31(b), but has shifted to the right,
as illustrated in Fig. 3.32(b). What happened to the other two modes?
Could we get our answer from the geometrical loci? What happened is
that the PQ curve also moved to the right. At this particular value of
ω0 /ωp = 0.6 it crosses the ε = −1, µ = −1 point at A without ever
crossing the area in which either surface mode exists.
When ω0 increases further there is an inversion of the upper TE and
TM modes. First comes a backward TM mode and further up (although
starting again at the same point) a forward TE mode. These changes
are indicated by the next set of curves taken for ω0 /ωp = 0.8 (Fig.
3.33). Note that ω0 > ωs , hence the magnetic resonance does not affect
the lower TM mode. Its range is, as in the case when µ is frequency-
independent, between ω = 0 and ω = ωs . The dispersion curves of the
three modes are plotted in Fig. 3.33(a) and the corresponding curves in
the ε,µ plane in Fig. 3.33(b). It may be clearly seen that in the interval
1.6 6
P=0
ne
1.4
li
ht
lig
(iii) TE 4
1.2 P=−1
H=0 (i) TM
1 Z
0
2
1
Z/Zp
P=P /P
0.8
2
Q
H=−1
(i) TM 0
0.6
A
0.4
−2
0.2 (iii) TE
0 P
−4
0 0.5 1 1.5 2 2.5 3 −8 −6 −4 −2 0 2
kx/kp H=H /H
2 1
(a ) (b )
Fig. 3.34 SPP dispersion for single interface. F = 0.56, ω0 = ωp . (a) ω–kx diagram. (b) µ–ε diagram
PA we have a TM mode, between A and B a TE mode, and no surface

mode beyond B.
In case (d) we have chosen ω0 /ωp = 1. There is now no region for
which µ and ε are simultaneously negative. The lower TM mode, as
may be expected, remains unchanged. The upper TM mode has disap-
peared and only one TE mode remains in the frequency region where the
permeability is negative. The dispersion curve is shown in Fig. 3.34(a)
and the geometrical locus of the modes in Fig. 3.34(b). The region PA
is now responsible for the upper TE mode and no surface mode exists
beyond A.
We may summarize at this point the rules governing the appearance
of the upper TM and TE modes. The upper TM mode exists in the
frequency range between the values that give εµ = 1 and that giving
ε = −1. Depending on which one is higher for the particular value of
ω0 /ωp it is a forward wave or a backward wave. When those two limits
take identical values the upper TM mode disappears. The upper TE
mode originates at the same point as the upper TM mode and tends
asymptotically to the line µ = −1. Depending again on the relative
positions of these two points the TE wave is a backward wave or a
forward wave or it just vanishes when it changes from one into the other
one.
Concerning the number of modes for a given value of ω0 /ωp they vary
between one and three. Specifically, there is only one mode for case (b),
two modes for case (d), and three modes for cases (a) and (d). Case (a)
has been considered by Ruppin (2001). Case (b) includes the perfect
lens condition ε = µ = −1 at ω = ωs .
Note that the total number of modes is dictated by the chosen form
for the frequency dependence of ε and µ that result in various ways of
crossing the regions in which surface modes are allowed. A different
1 H
line
(+)
(iii) TM Z
light
0.8
H
(−)
(+) Z
0.6 (ii) TE Z
p
ZZ
(a)
P
Z(−)
0.4 Z0
(+)
Z
(−)
(i) TM Z
02
0
0 1 2 3 4 5
kx/kp
1 H
line
light
0.8 (+)
(iii) TM Z
H
Z(−)
0.6
(+) (b)
ZZp
(ii) TE Z
(−) P
Z
0.4 Z0
Z(+) (i) TM Z(−)
0.2
Fig. 3.35 SPP dispersion for a slab. 0

0 1 2 3 4 5
F = 0.56, ω0 = 0.4 ωp . kp d = 0.25 (a) kx/kp
and 1 (b)
frequency dependence for ε and µ (e.g. with multiple resonances) can

result in a smaller or larger number of modes. The results of Section 3.7.2
summarized in Fig. 3.29 would still apply.
A number of authors looked at different possibilities for the depen-
dence of ε and µ on frequency. The results reported by Ruppin 2000;
Darmanyan et al. 2003 can be reproduced with our model.
3.7.4 SPP modes for a slab of a metamaterial

The story is the same again as for the epsilon-negative-only medium.
If the slab is sufficiently thin the modes on each side are coupled to
each other. As a consequence, the number of branches in the dispersion
equation of a slab will double in comparison to the single interface. Each
mode splits into a symmetric and an antisymmetric branch. We take
here only two examples, case (a) and case (b) of the previous section,
i.e. for ω0 = 0.4ωp and ω0 = 0.6ωp. The resulting dispersion curves for
1 H
light line
(+)
0.8 (iii) TM Z
(ii) TE
Z(+)
H P
Z( ) (−)
Z Z0
0.6
ZZp
(+) (a)
Z
0.4 (i) TM Z
(−)
02
0
0 2 4 6 8 10
kx/kp
1 H
light line
(ii) TE (iii) TM
}
Z(+) Z(−) (+)

Z
0.8 Z(−)
H P
0.6 Z0
ZZ p
Z(+) (b)
0.4 Z(−)
}
(i) TM
02
0 Fig. 3.36 SPP dispersion for a slab.

0 2 4 6 8 10
k /k
F = 0.56, ω0 = 0.6 ωp . kp d = 0.25 (a)
x p and 1 (b)
kp d = 0.25 and 1.0 are shown in Figs. 3.35(a) and (b) and Figs. 3.36(a)
and (b). Comparing them to the unperturbed solutions for the single
interface shows some striking features in case (b). Apart from the low-ω
TM, which splits into two, two further modes, a backward and a forward
one, appear that asymptotically approach the miraculous frequency at
which εr = µr = −1. The higher kx is, the closer are the two modes (one
above and one below) to the single interface curve. Due to the different
character of the frequency dependence of ε and µ, the split between the
TE modes looks different from the split between the TM modes (it is
actually larger for TM). We will return to this case later when discussing
subwavelength imaging properties of the perfect lens.
The asymmetric structure when the two media surrounding the slab
are different were first considered by Ruppin (2001) and later generalized
by Tsakmakidis et al. (2006).
We finish this section by mentioning an idea due to Oulton et al.
(2008). The authors note that, due to losses and stringent fabrication
requirements, practical SPP waveguides have not succeeded in produc-
ing field confinement beyond that of dielectric waveguides. The new idea
is to form a hybrid waveguide between a cylindrical dielectric nanowire

and a metallic plate. In that case much of the power will propagate in
the space between the nanowire and the plate, reducing thereby the
losses, and confinement will be determined by the distance between
them, which can be made very small by available semiconductor fab-
rication techniques. For further comments on this waveguide see Maier
(2008).
Small resonators
4
4.2 Early designs:
a historical review 120
The kind of resonators we need for metamaterials has already been dis-
cussed in Section 2.5. They have to be small relative to the wavelength
and at the same time should be accessible to electric and/or magnetic 4.4 A mathematical model
and further experimental
fields. The elements that were proved useful for showing the existence results 144
of negative material parameters and negative refraction (Smith et al.,
2000; Shelby et al., 2001a) were metallic rods and split-ring resonators.
The combination was new. It is true, however, that those elements and
all those introduced later were the products of a long line of develop-
ment. This development is of interest on its own account because it tells
us how the practical need of the manipulation of RF and microwave
signals led to a variety of approaches that still give inspiration when
new applications arise or simply when one wishes to improve an earlier
design.
The aim of this chapter is to present most of the small resonators
proposed up to the time this manuscript was sent to the publisher. It
provides a fairly comprehensive list. Obviously, there is no space to go
into much detail but we shall make an attempt to emphasize the main
features of the design, and of course, in many cases it is sufficient just
to look at the figures to appreciate how the resonance comes about and
what function the resonator might fulfil. The order of presentation will
be chronological at first. In Section 4.2 we shall introduce small res-
onators, starting with the invention of the re-entrant cavity in 1939.
For decades afterwards the main application of small resonators was for
providing crucial components in microwave tubes, for detecting nuclear
magnetic and electron spin resonance, and for microwave filters. The
emphasis changed around 1999–2000. Since then, most of the research
on small resonators has been directed towards realizing negative mater-
ial parameters: negative permittivity and negative permeability. Those
new developments will be presented in Section 4.3, including the at-
tempts to reach higher and higher frequencies. These two sections will
provide an extensive list, a kind of catalogue, with relatively little discus-
sion and without mathematical models, but showing some experimental
results as well. In Section 4.4 the mathematical model of Shamonin
et al. 2004; Shamonin et al. 2005 is introduced. In contrast to other
analyses, the equivalent circuit of the split-ring resonator is presented
there in terms of distributed instead of the usual lumped circuits. It is
shown there that the current and voltage distributions may be obtained
120 Small resonators
Volume within which Capacitance

electron beam is modulated
Fig. 4.1 Re-entrant cavity Inductance
by solving a second-order differential equation, and the solution will also

yield the resonant frequency. Theoretical predictions of the resonant fre-
quency will be compared with the experimental results of Radkovskaya
et al. (2005) and also with those obtained by a numerical package. In
Section 4.5 we have a brief look at the neglected topic of higher-order
resonances. We shall compare some of the predictions of the distributed
circuit model with those of simulations.
4.2 Early designs: a historical review

The need for small resonators arose at the design of the first microwave
tubes. In a klystron, physical space was needed for modulating the
1
There had to be an interaction space electron beam1 but at the same time the resonator had to be small
where the field affected the motion of (tubes need to be small) and it had to be resonant for using efficiently
the electrons and vice versa when the
electrons delivered power to the field.
the electromagnetic power. The main requirement of the design was
to ensure a space in which a longitudinal electric field could interact
with a drifting electron beam. The resonator designed (Hansen, 1939;
Hansen and Richtmeyer, 1939; Whinnery and Jamieson, 1944), called a
re-entrant cavity, is shown in Fig. 4.1. The resonance is now not based
on the bouncing of waves between reflectors but comes about because
parts of the single metallic structure can be identified as an inductance
and as a capacitance, respectively. The capacitance is given by the
opposing metallic grids and the inductance is provided by the rest of the
cavity. Clearly, the resonant frequency can be lowered by making the
opposing metal surfaces closer to each other and increasing thereby the
capacitance. A recent realization of a re-entrant cavity is that of White
et al. (2005) who designed a tunable resonator in the GHz region having
a diameter of about λ/10.
The classic example of a small resonator is that of the cavity mag-
netron that played such prominent role in Second World War radar.
For an account by the inventors see Boot and Randall (1976). The
magnetron cavity (three adjacent ones are shown in Fig. 4.2(a)) is in
a sense the ancestor of all small resonators we use today: (i) it is an
open structure, (ii) it has a gap through which it can interact with ex-
ternal electric fields, (iii) it is small relative to the wavelength, (iv) its
resonant frequency is determined by the capacitance and inductance of
the resonator. The capacitance is only partly due to the gap; there are
SEGMENT CAVITY
Fig. 4.2 (a) Three adjacent magnetron

(a ) (b) cavities. (b) Electric-field lines.
l
C
1
C ZL RB
S [
D
D0 Zn Z1
(a) (b)
Fig. 4.3 Shielded symmetrical slotted-tube line (a). Matching and tuning of the slotted tube resonator (b). From Schneider
and Dullenkopf (1977). Copyright c 1977 American Institute of Physics
contributions from other parts of the resonator as well as follows from (a) (b)
the field pattern of Fig. 4.2(b). The inductance is due to the horizontal
current flowing in the cylinder.
Our first two examples were on microwave tubes. Interestingly, the
need for small resonators also arose in the detection of nuclear mag-
netic resonance as the resonant frequency moved towards the hundreds
of MHz region. A resonator with bouncing waves would have been too
large, whereas the traditionally used solenoid plus lumped capacitor no Fig. 4.4 Loop gap resonators
longer worked satisfactorily. The solution proposed by Schneider and
Dullenkopf (1977) is shown schematically in Fig. 4.3(a). It is a cylin-
drical structure called the slotted-tube resonator. As may be seen, it
bears a strong resemblance to the magnetron cavity. The main differ-
ence is that the inner tube is split so that the upper and lower parts of
the inner tube can serve as a transmission line similarly to a strip line.
The outer cylinder provides a shield. Combining these field concepts
with circuit considerations the authors used a matching capacitance C1
r0 t
w
l
(a) (b)
Fig. 4.5 Split-ring resonator. (a) the simplest version and (b) the double-ring version. From Hardy and Whitehead (1981).
c 1981 American Institute of Physics
Copyright
and a capacitance C to reduce the length of line for resonance as shown

in Fig. 4.3(b). The same type of resonator for electron spin resonance
measurements in the range of 1–10 GHz was proposed by Mehring and
Freysoldt (1980). Tuning was achieved by inserting a dielectric. A very
similar design by Froncisz and Hyde (1982) was called a loop gap res-
onator and the same terminology was used by Pfenninger et al. (1988)
and Ghim et al. (1996). The essential features of their resonators are
shown in Figs. 4.4(a) and (b). That of Pfenninger et al. (1988), the so-
called bridged loop gap resonator, has an additional capacitance due to
two surfaces opposite the two gaps. In the design of Ghim et al. (1996)
two circles of slightly different radii overlap, producing the capacitance
needed for the resonance.
2
This is the first time that the term A variation on the slotted-tube resonator is the split-ring resonator2
split-ring resonator is mentioned in the of Hardy and Whitehead (1981) shown in Figs. 4.5(a) and (b). Now the
literature. It consists of a double ring
but only the inner ring is split, whereas
metallic tube is split on one side only. The split tube is a resonator on
in the split-ring resonator of Pendry its own, conceptually exactly the same as the magnetron cavity but the
et al. (1999), so popular nowadays, geometry of the gap is different, the contribution of the gap to the capac-
both rings are split. In the terminol- itance is higher. The outer cylinder is for confining the magnetic field in
ogy of Shamonin et al. 2004; Shamonin
et al. 2005 the resonator of Hardy and the annular region. Then, the magnetic flux in the outer region can be
Whitehead (1981) is called a singly split taken to be equal to that in the inner region providing the relationship
double ring.
BS2 = −S1 B0 (4.1)
holds, where
S1 = πr02 , S2 = π[R2 − (r0 + w)2 ] . (4.2)

w is the thickness of the inner tube and R and r0 are the inner radii
of the outer and inner tubes, respectively. Then, from the equality of
magnetic and electric energy they determine the resonant frequency as
r r
S1 t c
ω0 = 1 + w , (4.3)
S2 π r0
(a) (b) (c)
Fig. 4.6 Double circular elements.

From Kostin and Shevchenko 1993;
Kostin and Shevchenko 1994
where t is the gap width. They found the above expression to be accurate
within about 20%. For an extension of the frequency range of the split-
ring resonator to 4 GHz see Momo et al. (1983).
The aims of Kostin and Shevchenko 1993; Kostin and Shevchenko 1994
were somewhat different. They wanted to realize an artificial medium (a) (b)
using metallic elements that may yield a high value of the imaginary
part of the permeability. They wanted high magnetic losses and at the
same time they wished to minimize the amount of metal, which made
them use thin films. Their first design was a non-resonant metallic loop.
Later ones were resonant solutions combining in slightly different ways
(c) (d)
capacitances and inductances, as shown in Figs. 4.6(a)–(c).
An element combining an electric and a magnetic dipole was proposed
by Saadoun and Engheta (1992). The corresponding media were called Fig. 4.7 Ω-particles. From Engheta
by the authors omega or pseudochiral media. The reason for the first et al. (2002)
designation is obvious, the elements are shaped in the form of the Greek
letter, Ω, as may be seen in Fig. 4.7(a) (some related elements are shown
in Figs. 4.7(b)–(d)). The second description is due to the fact that
although the element possesses no handedness, it couples to each other
electric and magnetic fields with a phase difference of 90 degrees. The
general relationship is formulated3 by the authors as 3
This is a somewhat different form of
the tensorial relationships given in Sec-
1 tions 1.15 and 1.16.
D = εE + Ωem B and H= B + Ωme E , (4.4)
µ
where ε, µ, Ωem and Ωme are the permittivity, permeability and coupling
tensors, respectively. ε and µ are diagonal tensors, whereas
Ωem = j Ωyz and Ωme = j Ωzy . (4.5)

A generalization of the element by Saadoun and Engheta (1994) to one
including a small transmission line between the stems and the loop is
shown in Fig. 4.7(b). The effective permittivity and permeability of
the medium, as calculated by them, are shown in Figs. 4.8(a) and (b).
The remarkable fact is that both materials constants turn negative at
the same frequency and remain negative in a certain frequency band.
This is actually the medium Veselago (1968) was looking for but neither
Saadoun and Engheta nor anybody else who came across their paper
was interested at the time to make use of it. Saadoun and Engheta had
actually some doubts whether their circuit model was valid at such high
frequencies. Some nine years later an analytical study by Simovski and
He (2003) confirmed that both material parameters can be negative in a
certain frequency band. Experimental proof that a material containing
such elements can exhibit negative refraction was provided by Ran et al.
Fig. 4.8 Effective permittivity and permeability of an Ω-particle medium. From Saadoun and Engheta (1994)
z z
P a y a y
Fig. 4.9 One-turn helices. (a) Right- P

handed helix. (b) Left-handed helix.
x x
From Bahr and Clausing (1994). Copy-
c 1994 IEEE
right (a) (b)
(2004). It is a rather curious fact that the omega particle, in spite of its
obvious suitability as a metamaterial element, has been eclipsed by the
SRR plus rod combination.
A chiral element with actual handedness was analyzed by Bahr and
Clausing (1994). The one-turn helix (Figs. 4.9(a) and (b)) was either
right-handed or left-handed. A further chiral element similar to that of
z0 Saadoun and Engheta (1994) was investigated by Tretyakov et al. (1996).
The analysis was based on the available theory of both linear and loop
antennas (King, 1969). Their basic element is shown in Fig. 4.10. It
y0 consists again of a combination of an electric and a magnetic dipole but
A this time the perpendicular to the loop is in the direction of the electric
a f
A1
x0 dipole. The excitation of dipoles (electric and magnetic) in terms of fields
l (electric and magnetic) mediated by polarizability tensors is summarized
in Table 4.1. It shows which tensors play a role for particular excitations.
The terminology is the same as that presented in Section 1.16.
Let us take as an example a plane wave incident upon such an ele-
Fig. 4.10 Geometry of a chiral particle. ment with electric-field polarization in the x direction and magnetic-field
From Tretyakov et al. (1996). Copy- polarization in the z direction. The corresponding tensor elements are
c 1996 IEEE
right αmm em me ee
zz , αzz , αyz , and αxx . The corresponding physics is quite straight-
forward. A magnetic field in the z direction sets up a current flowing
in the loop that gives rise to a magnetic moment in the z direction and
Table 4.1 Effective dipole moments for different exciting field directions. From
c 1996 IEEE
Tretyakov et al. (1996). Copyright
Field orientation Effective dipole moments Polarizability components
pz
m
E αee
zz
py αme
zz
αee
yz
pz
m
H αem
zz
py αmm
zz
αem
yz
pz
E m
αee
yy
py αme
zy
αee
zy
E αee
px xx
to an electric dipole moment in the y direction. But the same current

flows also in the electric dipole and sets up therefore an electric dipole
moment in the z direction. Finally, the electric field in the x direction
sets up an electric dipole moment in the same direction. Note that for
αme ee
yz and αxx to be finite the current in the loop must vary as a function
of the angle, i.e. the current in the loop cannot be uniform, which means
that the loop cannot be infinitesimally small.
Another version of the magnetron cavity realized in thin-film form
was used by Hong and Lancaster 1996a; Hong and Lancaster 1996b;
Hong and Lancaster 1999; Hong 2000 in designing microwave filters.
Two of their typical filter configurations may be seen in Figs. 4.11(a)
and (b), showing a four-pole and a six-pole design, respectively. In the
same spirit, Hong and Lancaster (1998) produced yet another microwave
resonator for filter applications that was U-shaped, as shown in Fig.
4.11(c). They called them hairpin resonators.
The elements that have been discussed so far were designed either for
some specific purpose (e.g. for a filter or for the measurement of nuclear
magnetic resonance) or as potential elements in some, not very well
specified, artificial material. A wide variety of resonators had already
been designed and measured. Any new resonator was bound to be some
modified version of an existing one.
Fig. 4.11 Microwave filters in thin-

film form. (a) and (b) From Hong and
Lancaster (1996a). Copyright c 1996
IEEE. (c) From Hong and Lancaster
(1998). Copyright c 1998 IEEE (a) (b) (c)
4.3 A roll-call of resonators

As we have seen before the term ‘split-ring resonator’ was coined by
Hardy and Whitehead (1981) referring to a design in which the inner
ring is split, and the same name was retained by Pendry et al. (1999)
whose resonator (see Fig. 2.12) bore close resemblance to resonators
already shown in Figs. 4.3(a), 4.4(a), 4.5(b) and 4.6(a). There was,
Fig. 4.12 Swiss roll
however, a difference. The design of Pendry et al. (1999) was specifically
aimed at producing an effective negative permeability. The authors also
introduced a new embodiment of the swiss roll (Fig. 4.12) that was used
earlier for generating homogeneous magnetic fields (Sasaki et al., 1995).
The subject had not been called metamaterials as yet but there was
already a central focus: how to achieve negative material parameters
(a)
and how to demonstrate negative refraction? It turned out that prac-
tically all conceivable magnetic resonators were capable of producing
negative permeability, provided the magnetic field could have access to
the element.
Next, we shall show three magnetic resonators investigated by Baena
et al. (2005a). The first one (Fig. 4.13(a)) is the celebrated split-ring
resonator,4 the other two are of fairly similar design. Figure 4.13(b)
shows a quadruply split double ring, and that in Fig. 4.13(c) is a some-
(b) what distorted spiral (see also Baena et al. 2004). For the SRR of
Pendry et al. (1999), as mentioned already in Section 2.5, the induc-
tance can be regarded as Lav , the average of that of the inner and outer
rings. Also, the two inter-ring capacitors (those between the splits) are
in series resulting in the equivalent circuit of Fig. 4.13(a), giving a resul-
tant capacitance of C0 /4, where C0 is the total inter-ring capacitance. In
the quadruply split double ring of Fig. 4.13(b) there are four inter-ring
(c) sections that can be regarded as being responsible for four capacitances
in series, each of them equal to C0 /4. In the spiral of Fig. 4.13(c) there is
no gap, hence the total capacitance is equal to C0 . Clearly, for the same
Fig. 4.13 Magnetic resonators and physical size the lowest resonant frequency can be achieved with the spi-
their equivalent circuits: SRR (a),
quadruply split double ring (b) and spiral resonator. We shall quote here two of their results for the SRR and
ral resonator (c). From Baena et al. the quadruply split double ring for resonators having the dimensions,
(2005a). Copyright c 2005 IEEE average radius = 3.55 mm, metal width = inter-ring separation = 0.3
mm. Resonant frequencies calculated from this theory gave 3.33 GHz
4
From now on the term split-ring res- for the SRR and, as expected, 6.66 GHz for the quadruply split double
onator, abbreviated as SRR, will be re- ring. The corresponding measured resonant frequencies were 3.40 and
served for the resonator shown in Fig. 6.77 GHz, showing excellent agreement.
4.13(a)
A resonator that may be regarded as the dual of the SRR was also
a a
CM
c
ext
CG LC
Fig. 4.14 Complementary split-ring
d resonator (a) and its equivalent cir-
cuit (b). From Beruete et al. (2006).
Copyright c 2006 American Institute
CSRR of Physics
proposed by Baena et al. (2005b) and used as an element in an array by

Beruete et al. (2006). It is shown in Fig. 4.14 with the equivalent circuit. (a)
The additional capacitor CM is for coupling to the next element. It may
be seen that metal and air have been interchanged, hence the name of
complementary split-ring resonator (already introduced in Section 2.5
as an example of a resonant element). For another study of rings and
(b)
their equivalent circuits see Rogla et al. (2007).
Another set of thin-film split-ring resonators were investigated by Ay-
din et al. (2005). The first three (Figs. 4.15(a)–(c)) are single rings
split one, two and four times. The fourth resonator is a quadruply split
(c)
double ring, the same type as shown in Fig. 4.13(b) and the fifth one
is another double ring split eight times with the gaps lined up. The
basic structure of the SRR has dimensions external radius = 3.6 mm,
gap width = inter-ring separation = 0.2 mm, metal width = 0.9 mm.
With these parameters the resonant frequency of the SRR is 3.63 and (d)
3.60 GHz obtained by measurement and simulation, respectively. The
resonance curves for gap widths equal to 0.2, 0.3, 0.4 and 0.5 mm are
shown in Fig. 4.16, the measured results in Fig. 4.16(a) and the sim-
ulated ones in Fig. 4.16(b). The agreement may be seen to be good (e)
for the resonant frequencies, differs somewhat for the depth of the reso-
nances, and differs considerably for the widths of the resonance curves.
Apparently, the simulated results show wider resonance curves, indicat-
ing higher losses. For the resonant frequencies the conclusion is clear: it Fig. 4.15 Magnetic resonators: sin-
increases as the gap increases, which follows of course from the fact that gle rings split one, two and four times
(a)–(c), quadruply split double ring (d),
larger gap means smaller capacitance. double ring split eight times (e). From
Dependence of the resonance curves on ring separation is shown in Aydin et al. (2005)
Figs. 4.17(a) and (b) for measured and simulated results, respectively.
Again, the resonant frequency increases as the separation increases, and
the reason is again that increased separation reduces the inter-ring ca-
pacitance and leads therefore to higher resonant frequency. The agree-
ment between measured and simulated results is again good for the
resonant frequencies, reasonable for the resonance dip and less good for
the width of the resonant curves.
Next, we shall look at a comprehensive numerical study carried out
by Kafesaki et al. (2005) in which the authors explore a wide frequency
range. The resonator investigated is shown in Fig. 4.18 (dimensions
(a) (b)
0 0
–5 –5
Transmission (dB)
Transmission (dB)
–10 –10
–15 d = 0.2 mm –15 d = 0.2 mm

d = 0.3 mm d = 0.3 mm
–20 d = 0.4 mm –20 d = 0.4 mm
d = 0.5 mm d = 0.5 mm
–25 –25
3.4 3.5 3.6 3.7 3.8 3.9 4.0 4.1 4.2 3.4 3.5 3.6 3.7 3.8 3.9 4.0 4.1 4.2
Frequency (GHz) Frequency (GHz)
Fig. 4.16 Transmission spectra of individual SRRs with d, the width of the gap, as a parameter, obtained by (a) experiment
and (b) simulation. From Aydin et al. (2005)
(a) (b)
0 0
Transmission (dB)
Transmission (dB)
–5 –5
–10 –10
t = 0.2 mm –15 t = 0.2 mm
–15
t = 0.3 mm t = 0.3 mm
t = 0.4 mm t = 0.4 mm
–20 t = 0.5 mm –20 t = 0.5 mm
–25 –25
3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8 3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8
Fig. 4.17 Transmission spectra of individual SRRs with t, the inter-ring separation, as a parameter, obtained by (a) experiment
and (b) simulation. From Aydin et al. (2005)
l = 1.8 mm, g = t = w = 0.2 mm). The transmission of a plane wave

through this resonator is then determined by a numerical package. The
relevant resonance curves are shown in Fig. 4.19(a) for the case when
the electric field is parallel with the gaps. There are three resonances at
around 22, 50 and 70 GHz. These results may then be compared with
those exhibited in Fig. 4.19(b) that are calculated for the inner and outer
rings separately. If we reduce the inter-ring separation both resonances
tend to move towards lower frequencies: the upper resonance becomes
weaker and the lower resonance stronger. At a value of t = 0.05 mm the
upper resonance practically disappears and the lower resonance shifts
to 15 GHz (Tatartschuk, 2007). The single resonance at 15 GHz or the
two resonances at 22 and 50 GHz are magnetic resonances in the sense
that they are caused by circulating currents that then produce magnetic
dipoles. Alternatively, they could be regarded as LC resonances in the
sense that they are caused by the inductances and capacitances of the
g l
w Fig. 4.18 Double-ring SRR. g = t =
t
w = 0.2 mm; metal depth = 0.2 mm.
The SRR side l = 1.8 mm. From Kafe-
c 2005
saki et al. (2005). Copyright
k IOP Publishing Ltd

T

outer
inner
Fig. 4.19 Transmission (T , in dB) ver-
double
sus frequency (in GHz) for the double-
ring SRR shown in Fig. 4.18(a), and
10 20 30 40 50 60 70 10 20 30 40 50 60 70 for its isolated outer and inner ring res-
Frequency (GHz) Frequency (GHz) onators (b). The background mater-
ial is air. From Kafesaki et al. (2005).
(a) (b) Copyright c 2005 IOP Publishing Ltd
SRR (cf. the physical arguments advanced in Section 2.5 for the resonant
frequency of a SRR).
We may then ask the question what the third resonance at 70 GHz
(Fig. 4.19) is due to? It seems very likely to be an electrical reso-
nance akin to that of an electric dipole antenna. It needs to be recalled
that a centre-fed (or short-circuit) electric dipole antenna has resonances
when its length is approximately equal to an integral multiple of half-
wavelengths. More accurately, the resonant length of an infinitely thin
electric dipole is a little below half-wavelength and the resonant length
is further reduced as the diameter to length ratio is increasing. 70 GHz
corresponds to a free-space wavelength of 4.3 mm. Thus, a side of the
SRR in Fig. 4.18 being equal to 1.8 mm is about 0.42 wavelength. So
the electrical resonance hypothesis is plausible.
Kafesaki et al. (2005) also determined the transmission of a plane wave
for three further orientations of the SRR, as illustrated in Fig. 4.20(a).
In configuration (A) the electric field is perpendicular to the gap. The
corresponding resonances are shown in Fig. 4.20(b). Four resonances
may now be seen. The two magnetic resonances are at the same fre-
quencies, as may be expected, since the magnetic field is perpendicular
to the plane of the SRR. The physical mechanism for the two resonances
E
In-plane incidence Normal incidence
(A)
k
H
E
T
(B)
k
H (A) (B)
E (C)

(C)
k Frequency (GHz) Frequency (GHz)
H
(a) (b) (c)
Fig. 4.20 Plane-wave transmission. (a) Configurations (a). Transmission spectra for in-plane incidence (b) and for normal
c 2005 IOP Publishing Ltd
incidence (c). From Kafesaki et al. (2005). Copyright
at 35 and 58 GHz is not clear, it needs further analysis. In configuration

(B) the magnetic field is in the plane of the SRR, hence it cannot induce
a current. There are no magnetic resonances as shown clearly in Fig.
4.20(c) but there is an electrical resonance at 57 GHz (solid line) that
must be the same type as in configuration (A). In configuration (C) there
are four resonances, as plotted with dashed lines in Fig. 4.20(c). The
magnetic resonances at about 20 and 50 GHz have reappeared in spite of
the fact that the magnetic field is not perpendicular to the plane of the
SRR. The resonances are caused by the electric field that, in this case,
is parallel with the gap. The physical mechanism is that the electric
field sets up an electric dipole moment that then causes currents to flow
in both the inner and the outer ring, so the conditions are right again
for a magnetic resonance. This is certainly one way of looking at the
physics but it is probably much better to consider how a magnetic field
y can set up an electric dipole and then invoke reciprocity. This is done
x in Appendix H. The other two resonances are slightly shifted relative to
z (a) (b) those shown in Fig. 4.20(b).
We have talked about magnetic and electric resonances. The lower
Fig. 4.21 SRR (a) and full ring (b) ones were regarded magnetic and the higher ones electrical. A better
configurations. From Aydin and Ozbay criterion is the one used by Koschny et al. 2004a; Koschny et al. 2004b,
(2007b). Copyright c 2007 Wiley- and Aydin and Ozbay 2006; Aydin and Ozbay 2007b. The essential phys-
VCH Verlag GmbH & Co. KGaA
ical mechanism in the magnetic resonance of SRRs is the non-uniform
distribution of currents along the inner and outer rings. If those currents
could be made uniform the magnetic resonances would be absent. We
can indeed have uniform distribution of currents by removing the gaps,
and we are then left with two full concentric rings. In the experiments
of Aydin and Ozbay (2007b) the transmission was measured both for a
SRR and for a full ring (called by the authors a closed-ring resonator
(a) (b)
0 0
Transmission (dB)
–5 –5
–10 –10
–15 –15
–20 SRR –20 SRR
–25 CRR CRR
–25
3 4 5 6 7 8 9 10 11 12 13 3 4 5 6 7 8 9 10 11 12 13
Fig. 4.22 Frequency response of an individual SRR and CRR (full ring) obtained from (a) experiments and (b) simulations.
From Aydin and Ozbay (2007b). Copyright c 2007 Wiley-VCH Verlag GmbH & Co. KGaA
–10 0
30
Transmission (dB)
60
–20 90
E
H
–30
k
–40
Fig. 4.23 Transmission spectra for
oblique incidence. From Aydin and
3 4 5 6 7 Ozbay (2007b). c 2007
Copyright
Frequency (GHz) Wiley-VCH Verlag GmbH & Co. KGaA
and abbreviated by CRR) in the configuration when the electric field is

in the y direction and the magnetic field in the z direction (see Figs.
4.21(a) and (b)). The dimensions of the ring were: outer radius 3.6 mm,
inter-ring separation and the gap 0.2 mm and the metal width 0.9 mm.
The experimental results are shown both for the SRR and the full ring
in Fig. 4.22(a) and the corresponding simulations in Fig. 4.22(b). There
are three resonances for the SRR and only one for the full ring. The
agreement between the measured and simulated results may be seen to
be very close. Classification is now easy: the resonance that is com-
mon for the SRR and full ring is an electrical one, the two others are
magnetic.
Aydin and Ozbay (2007b) also measured the transmission of micro-
waves incident at an angle upon a 3D array of SRRs in the configu-
ration shown in the inset of Fig. 4.23. The electric field is in the y
direction parallel with the gaps and the direction of propagation varies
between 0◦ and 90◦ . At 0◦ the magnetic field is perpendicular to the
dc g
w
Lc d
D w r2
b h r3 r1
g r4
f
b s t e dc
(a) (b) (c) (d) (e)
w t
r r
l s
g
(f ) (g) (h) (i) (j)
Fig. 4.24 A variety of resonators (a) by O’Brien and Pendry (2002), copyright c 2002 IOP Publishing Ltd, (b) and (c) by
Guo et al. (2005), copyright c 2005 IEEE, (d) by Hsu et al. (2004), copyright c 2004 American Institute of Physics, (e) by
c 2006 Optical Society of America, (f) and (g) by Radkovskaya et al. (2007b), copyright
Bulu et al. (2005a), copyright c 2007
Wiley-VCH Verlag GmbH & Co. KGaA, (h), (i) and (j) by Kafesaki et al. (2005), copyright c 2005 IOP Publishing Ltd
plane of the SRR causing, as mentioned several times before, a magnetic

t
resonance. There is large absorption in the corresponding stop band. As
– the direction of the wave propagation declines from 90◦ the component
++ +–+ + –
–– c of the magnetic field perpendicular to the SRR plane is becoming smaller
rext
r0 and smaller, and correspondingly the resonant dip becomes less and less
deep. At 0◦ incidence the magnetic field no longer threads the SRR; the
++ magnetic resonance is completely absent, the stop band disappears.
+
– – –– + A number of other resonators of interest are shown in Fig. 4.24. They
++ ε
look quite different but they all obey the basic rule: loops (mostly bro-
ken) to provide the inductance, and surfaces close to each other to pro-
h vide the capacitance.
ε t The modified split-ring resonator of Marques et al. (2002b), rechris-
tened later as the broadside-coupled split-ring resonator (Marques et al.,
2003), is of particular importance because it has no bianisotropy: there
Fig. 4.25 Broadside-coupled split-ring
is no magneto-electric coupling. The element is shown schematically in
resonator. From Marques et al. (2003).
Copyright c 2003 IEEE Fig. 4.25. It consists of two split rings in which splits are 180 degrees
apart. The rings are then placed parallel to each other upon opposite
w sides of a dielectric substrate. It has the further advantage that the
g
opposing ring surfaces can be made large, the dielectric substrate can
z be made thin, so that the resonant frequency can be much higher than
for the usual SRRs. A similar design, elements parallel to each other
a
with splits opposite, was suggested by Maslovski et al. (2005). It differs
from Marques et al.’s broadside-coupled split-ring resonator by having a
b d large number of elements lined up, as may be seen in Fig. 4.26. For their
resemblance to a solenoid the authors called them metasolenoids. Two
Fig. 4.26 Metasolenoid. From
interesting designs with symmetry in the horizontal plane and showing
Maslovski et al. (2005) some new aspects of SRRs, are due to Gay-Balmaz and Martin (2002),
and to Chen et al. (2006). They may be seen in Figs. 4.27(a) and (b),
p C2
C4
I C1
C6
C6
z
y C3
p C5
x
C1
Fig. 4.27 Isotropic resonators. (a)
From Gay-Balmaz and Martin (2002).
of Physics.(b) From Chen et al. (2006).
(a) (b) of Physics
Fig. 4.28 Capacitively loaded resonators. (a) From Wiltshire et al. (2003b). (b) From Syms et al. (2006b)
respectively.
A simple way of building a small resonator to which magnetic fields
have access is to take a metallic loop and insert a lumped capacitance.
This was done by Wiltshire et al. (2003b) who wound two turns of 1
mm diameter copper wire on a dielectric rod (9.6 mm diameter) and
tuned the frequency to 60 MHz by inserting a capacitor (nominally 100
pF) between the ends of the wire as shown in Fig. 4.28(a). Syms et al.
(2006b) used rectangular coils on one (design A) or both sides (design
B) of a PCB as shown in Fig. 4.28(b). A resonator that the authors
called an open split-ring resonator was used by Martel et al. (2004) as
a component in a microstrip line for realizing a filter. It is shown with
its equivalent circuit in Figs. 4.29(a) and (b).
It is also possible to augment the inter-ring capacitance in a SRR
by adding an external capacitor (Aydin et al., 2005). In one of their
examples the resonant frequency of the SRR was 3.63 GHz, which they
could change to 2.87 GHz and to 1.63 GHz by inserting capacitors of 0.1
OSRR
+ – – +
Top view
Ground plane window
Bottom view
(a)
R L C
Z0, Hef Z0, Hef
d
Fig. 4.29 Open split-ring resonator.
From Martel et al. (2004) (b)
e pF and 0.8 pF, respectively, across the outer gap. In another publication
d Aydin and Ozbay (2007a) reported a reduction of the resonant frequency
from 3.82 GHz to 0.99 GHz by using a C = 2.2 pF capacitor in the
e gap. The authors also attempted to reduce the resonant frequency by
inserting capacitors between the rings. They succeeded, but found that
(a) it was a less efficient way of reducing the resonant frequency. Another
motivation for inserting a capacitance into a SRR is to tune the resonant
e frequency. This can be done with the aid of a varactor diode as shown
by Gil et al. 2004; Gil et al. 2006. They were able to tune their element
d from about 2.5 GHz to 3.1 GHz by applying a voltage to the diode in
the range of 0 to 30 V. The element is designed with the need in mind
to accommodate the diode and to facilitate its biasing. The two models
e g
considered, both elongated, are shown in Figs. 4.30(a) and (b).
All the elements discussed so far had anisotropic properties. If the
(b)
aim is to construct an isotropic medium then the unit cell should display
some 3D symmetry. Examples are the unit cube with six SRRs on its
Fig. 4.30 Capacitively loaded split surface (see Fig. 4.31(a)) as proposed by Pendry et al. (1999) and further
rings. From Gil et al. (2006). Copy- investigated by Baena et al. (2006), and the 3D symmetric element of
c 2006 IEEE
right
Gay-Balmaz and Martin (2002) (Fig. 4.31(b)). A fully symmetric design
by Soukoulis et al. (2006) and Padilla (2007) is based on quadruply split
(a) (b) (c)
c 1999 IEEE. (b) From Gay-Balmaz and Martin

Fig. 4.31 Isotropic resonators. (a) From Pendry et al. (1999). Copyright
(2002). Copyright c 2002 American Institute of Physics. (c) From Padilla (2007). Copyright
c 2007 Optical Society of
America
single rings as shown in Fig. 4.31(c).

The metamaterial concept has proved useful for inspiring applications
both in the RF and in the microwave region. It is certainly a desirable
thing to manipulate fields on a scale much smaller than the wavelength,
and there is no difficulty doing that up to the microwave region but
should we carry on and explore the possibilities at higher frequencies as
well? Is there a need for it? Yes, if we are envisaging optical circuits
as proposed, for example, by Engheta et al. (2005). How can we realize
higher-frequency resonators? Simply by reducing the dimensions of the
resonator. It was indeed found by Yen et al. (2004), Moser et al. (2005),
and Xu et al. (2006) that they could increase the resonant frequency up
to several THz by making the same sort of split-ring resonators smaller,
reducing the overall size of the element below 30 µm.
Could SRRs be made even smaller? Should we rely on SRRs anyway?
At the time of writing it is not clear which kind of element will win the
race towards the highest frequencies. There is, however, a strong argu-
ment in favour of the single ring, the magnetron-type open resonator,
since for the same size it has a higher resonant frequency. Indeed the
use of this type of resonators enabled Gundogdu et al. (2006), Zhang
et al. (2005b), Linden et al. (2004), Enkrich et al. (2005a) and Klein
et al. (2006) to reach 6, 65, 100, 250 and 330 THz, respectively.
We have to add here that Enkrich et al. (2005a) investigated a range
of open resonators as shown in Fig. 4.32, where going from right to left
the ‘U’ is filled in with metal. The metal square on the left is known to
give a plasma resonance that hardly changes with the ‘filling’ but the 5
We know from previous studies of the
magnetic resonance keeps on increasing to about 100 THz as the ‘filling’
Full SRR that there is no need for any
becomes smaller and the capacitance increases. A detailed investigation kind of splits to obtain electrical res-
of the resonant frequency of the U-shaped element was performed by onances (King, 1969; Tretyakov et al.,
Rockstuhl et al. (2006). They obtained plasma resonances5 up to several 1996) and that applies also to the opti-
cal range. Aizpurua et al. (2003) mea-
hundred THz. For an incident plane wave in which the electric field is sured plasma resonances in a gold ring
across the gap and for a particular resonance the same component of of 60 nm radius in the frequency range
the electric field is plotted in Figs. 4.33(a) and (b) and compared with of 200–300 THz.
the corresponding distributions in a short, thin metallic rod. The plots
200 nm
2.8
2.4
Wavelength [mm]
2.0
Fig. 4.32 Measured (crosses) and cal-

culated (squares) spectral positions of 1.6
the magnetic resonance (black) and the
electric resonance (gray) as a function
of the depth u (from left to right: u = 0, 1.2
30, 85, 160 and 210 nm). The corre-
sponding five SRR designs are depicted
at the top. From Enkrich et al. (2005a).
0 50 100 150 200
Copyright c 2005 Wiley-VCH Verlag
GmbH & Co. KGaA U [nm]
offer good physical insight into the emergence of the resonances.

Can even higher frequencies be achieved? Enkrich et al. (2005b)
reached 370 THz. Is there a limit? This question was addressed in
a number of publications by Ishikawa et al. (2005), Zhou et al. (2005a),
Marques and Freire (2005) and Tretyakov (2007). All agree that the lim-
itation comes about due to the inertia of the electrons, and all agree that
the limit for magnetic resonance comes at about 350 THz. In order to
reach this conclusion the concept needed is that of kinetic inductance dis-
cussed in Section 1.20, which has a very simple expression. Marques and
Freire argue that the SRR might still be the preferred candidate as we
try to reach higher frequencies. Tretyakov also comes to the conclusion
that the limit is about 350 THz for magnetic resonance, but he shows
c 2006 Springer Science +

Fig. 4.33 Electric-field distribution at the resonance. From Rockstuhl et al. (2006). Copyright
Business Media. For coloured version see plate section
d
E
w l D
H
k (a)
600
500
400
Frequency (THz)
300
200 1-cut (Simulation)

2-cut (Simulation)
4-cut (Simulation)
100 1-cut (LC-model)
2-cut (LC-model)
4-cut (LC-model)
0
0 5 10 15 20 25 30
–1
1/a (Pm )
(b)
Fig. 4.34 The scaling of the simulated magnetic resonance frequency as a function of the linear size a of the unit cell for
the 1-, 2-, and 4-cut SRRs (solid lines with symbols). Up to the lower terahertz region, the scaling is linear. The maximum
attainable frequency is strongly enhanced with the number of cuts in the SRR ring. The hollow symbols as well as the vertical
line at 1/a=17.9 µm−1 indicate that µ < 0 is no longer reached. The non-solid lines show the scaling of the magnetic resonance
frequency calculated through the LC circuit model. From Zhou et al. (2005a). Copyright c 2005 by the American Physical
Society. For coloured version see plate section
analytically that plasma resonances may occur as long as the operating

frequency is below the plasma frequency. Calculations by Sondergaard
and Bozhevolnyi (2007) indicated that in a silver strip of 10 × 20 nm
dimensions resonant frequencies as high as 830 THz may be obtained.
We shall show here some curves showing the saturation of the magnetic
resonance frequency by Zhou et al. (2005a). They consider three config-
urations, as shown in Fig. 4.34(a): the open resonator, the doubly split
single ring and the quadruply split single ring. The resonant frequencies
coming from their simulations are plotted in Fig. 4.34(b) against inverse
size. It may be seen that the resonant frequency in all three cases satu-
rates as the size becomes smaller and smaller. As may be expected, the
elements having more splits have higher resonant frequencies.
An interesting and rather unusual set of resonators was realized by
Zhang et al. (2005c). Their design is shown in Figs. 4.35(a) and (b). The
W do
z tAu-top
Au
H ZnS
ZnS Structure
tc
Substrate tAu Image
ZnS
(a) (b)
Fig. 4.35 Staple structure. (a) Schematic of the nanostructures. The light is incident from the top: for TM polarization, H
is in the y direction, perpendicular to the loop. (b) The equivalent LC circuit formed between the top staple structure and
its image in the metal. The dashed lines are the reference planes for the effective permeability calculation. From Zhang et al.
(2005c). Copyright c 2005 by the American Physical Society
Table 4.2 Sample geometrical parameters (all dimensions

in nm). From Zhang et al. (2005b)
Sample W H dc tAu tAu−top tc
A 130 280 190 ∼ 15 ∼ 30 80

B(D) 130 180 190 ∼ 15 ∼ 30 80
C 130 280 90 ∼ 15 ∼ 30 80
bold lines show contours by gold. The dielectric between the elements
and the reflector is ZnS. The authors call the resonators, quite aptly,
‘staples’. Taken together with their mirror image it is easy to see that
they possess inductances and capacitances. They were produced in 2D
arrays in the x and z directions.
The parameters for three structures realized are given in Table 4.2.
The experimental and simulation results for reflection at normal inci-
dence are plotted in Figs. 4.36(a) and (b). A large dip may be clearly
seen for configuration A at a frequency of about 45 THz for an incident
TM wave (magnetic polarization in the y direction). For configurations
B and C the dip (for the same polarization) occurs at higher frequen-
cies. The reason is clearly that for configuration B the height of the
‘staple’ H is smaller hence the inductance is smaller, and for configu-
ration C the flange is smaller, yielding reduced capacitance. Note that
the dip occurs only for the TM polarization when the magnetic field can
access the open loops. As may be seen in Fig. 4.36(a), the reflection is
frequency-independent for the TE polarization. The effective permeabil-
ity determined from their model is plotted in Fig. 4.36(c). A negative
permeability region for configuration C around 64 THz may be seen.
Clearly, the idea of having a reflector leads to interesting physics but
it makes it impossible to measure transmission. Absorption and effec-
100
C,TE
80
D,TM
60
(a)
40
A,TM C,TM
Reflectance (%)
20
Experiment
0
80
60
(b)
40
20 Fig. 4.36 Experimental (a) and simu-

Simulation lation (b) results for the reflectivity as
0
a function of frequency for three differ-
Effective Permeability
ent structures of Table 4.2. The res-

3 onant frequency is a function of the
nanostructure parameters and is inde-
2 pendent of array period, which is the
(c) same for all three structures. The res-
1 onance is observed only for TM polar-
Re(Peff ) ization, i.e. incident magnetic field cou-
0 pled into the inductive structure. (c)
Im(Peff ) Effective permeability extracted from
–1
the model for sample C. From Zhang
30 40 50 60 70 80 et al. (2005b). Copyright c 2005 by
Frequency (THz) the American Physical Society
tive permeability would be of interest for waves propagating in the x

direction but those were not available in the paper.
Having considered split-ring resonators, open resonators, various U
geometries, the ‘partially filled U’ and the ‘staples’ we shall next discuss
a very promising candidate for high-frequency resonance: the short-rod
pairs (see Figs. 4.37(a) and (b)). It started with sets of metal rods
investigated both theoretically Lagarkov and Sarychev (1996) and ex-
perimentally Lagarkov et al. (1998) in the context of the percolation
problem.6 Another approach to the use of small metallic components 6
A dielectric suddenly becomes con-
was that of Svirko et al. (2001) who proposed pairs of metallic stripes at ductive when the density of metallic in-
clusions reaches a critical value.
an angle to each other (see Fig. 4.38) for chiral applications in the optical
region. Renewed interest came when it was realized (Podolskiy et al.,
2002; Podolskiy et al., 2003; Panina et al., 2002) that a pair of short
metallic rods will have resonances below the half-wavelength associated
with dipoles at lower frequencies. For a set of experimental results on
resonances using short metal pillars see Grigorenko et al. (2005).
One may argue that short-rod pairs do not appreciably differ from the
open resonators discussed by Linden et al. 2004; Enkrich et al. 2005a;
Rockstuhl et al. 2006; Sarychev et al. 2006. They have both plasma and
LC type of resonances. The plasma resonance follows from the fact that
E H
k
d
b2
1
2b
Fig. 4.37 Short-rod pair. (a) Single

element. (b) Elements arranged into a
medium (a) (b)
Inductive Coupling
l1
φ
Fig. 4.38 An element of the bilay-
ered planar chiral metallic microstruc-
ture with inductive cross-layer cou- (a2 + b2) 1/2 a
pling. Two subwavelength metallic z
stripes of size d×∆ are mutually shifted d
in the xy plane on distance b and spaced l2
by a dielectric layer of thickness a (not
shown). From Svirko et al. (2001). y
Copyright c 2001 American Institute ∆ x
of Physics
the frequency is high enough to be not too far from the metal’s plasma
frequency, and for the LC resonance the following simple argument can
be presented. The two rods of Fig. 4.39 have self-inductances, there is a
mutual inductance between them and there is a capacitance between the
I rods, analogously to a split-ring resonator. There are conduction cur-
rents (denoted by I flowing in the opposite directions, and displacement
I currents as well flowing between the rods.
Another illustration of the transition7 from the open resonator to the
Fig. 4.39 Conduction current in a short-rod pair due to Dolling et al. (2005) is shown in Fig. 4.40.
short-rod pair A set of short-rod pairs turn out to be suitable not only to provide
7
The U-shaped resonator, second from negative permeability but negative permittivity as well. This was shown
the left, is reminiscent of the hair- theoretically by Podolskiy et al. (2002) and is illustrated in Figs. 4.41(a)
pin resonator of Hong and Lancaster and (b). The corresponding refractive index is plotted in Fig. 4.41(c).
(1998). Experimental results by Shalaev et al. (2005) and Kildishev et al. (2006)
exhibited negative indices of refraction at a frequency of 200 THz but
other experimenters (Dolling et al., 2005; Garwe et al., 2006) did not
succeed in obtaining a negative index in that frequency range.
Negative index due to a structure of short wire pairs was found by
Fig. 4.40 Transition from the open res- Zhou et al. (2006) in a narrow frequency range near 14 GHz. A design
onator to the short-rod pair. From
c
by Zhang et al. (2005a), shown in Fig. 4.42 (top) relies, instead of short
Dolling et al. (2005). Copyright
2005 Optical Society of America rods, on short but wider surfaces facing each other. This structure is
then responsible for the magnetic resonance. Rods reappear to provide
(a) (b) (c)
Fig. 4.41 Permeability (a), permittivity (b) and refractive index (c) for the short-rod-pair medium. From Podolskiy et al.
(2002)
Simplified magnetic structure
Fig. 4.42 Transition from short-rod

pairs to a grid structure. From Zhang
et al. (2005a). Copyright c 2005 Op-
Negative index material tical Society of America
the electrical resonance but in a somewhat different form, as may be seen

in Fig. 4.42 (bottom). Simulations by the authors predict the existence
of a frequency region around 150 THz that has both negative permittiv-
ity and negative permeability. The structure is becoming increasingly
popular and is often called a fishnet.
Zhang et al. (2005a) also succeeded in demonstrating experimentally
a negative index in the same frequency range. They used an entirely
different structure as shown in Figs. 4.43(a) and (b), where a dielec-
tric layer between two perforated gold films can be seen. The authors
measured the phases and amplitudes of the reflected and transmitted
waves in response to a perpendicularly incident beam, and then used
the algorithm of Smith et al. (2002) to find the negative index.
Further advance was reported recently by Valentine et al. (2008) based
on the theoretical predictions of Zhang et al. (2006). The idea is to
use a multilayer structure shown schematically in Fig. 4.44(a). The
scanning electron microscope image may be seen in Fig. 4.44(b). The
structure consists of alternating layers (21 altogether) of 30 nm silver
(Ag) and 50 nm magnesium fluoride (MgF2 ). The size of the unit cell
is p = 860 nm, and the other dimensions are a = 565 nm and b = 265
nm. The variation of the index of refraction was investigated in a series
E k
Fig. 4.43 Top: Schematic of the mul-

tilayer fishnet structure consisting of
an Al2 O3 dielectric layer between two
Au films perforated with a square ar-
ray of holes (838 nm pitch; 360 nm
diameter) atop a glass substrate. For
the specific polarization and propaga-
tion direction shown, the active regions
for the electric (dark regions) and mag-
netic (hatched regions) responses are
indicated. Bottom: SEM picture of the
fabricated structure. From Zhang et al.
(2005a). Copyright c 2005 Optical So-
ciety of America. For coloured version
see plate section
a b
a
b
Fig. 4.44 Diagram (a) and SEM image
(b) of fabricated fishnet structure. The p
inset shows a cross-section of the pat-
tern taken at a 45◦ angle. From Valen- Ag
tine et al. (2008). Copyright c 2008 1 µm
MgF 2
Nature Publishing Group
of experiments, in which an input optical beam was incident upon the

material shaped in the form of a prism. In a control experiment the
prism was removed letting the beam through a window. The position
of the beam, as may be expected, turned out to be independent of the
wavelength of the input beam as shown in Fig. 4.45(a). In the presence of
the prism the beam was displaced by increasing amounts as shown in Fig.
4.45(b). From the amount of displacement it was possible to determine
the index of refraction. It is interesting to note that at a wavelength
of 1465 nm the refractive index is close to zero, which means that the
optical path through the prism is everywhere the same (i.e. zero), hence
the prism loses its ability to displace the beam. The variation of the
refractive index as a function of wavelength is plotted in Fig. 4.45(c).
The measurement results are denoted by circles and the corresponding
error bars. The theoretical curve (continuous line) was obtained by using
the rigorous coupled-wave analysis (Moharam et al., 1995).
Plasma resonances of small 2D particles of silver were investigated nu-
merically by Kottmann et al. 2000b; Kottmann et al. 2000a; Kottmann
Window Prism Frequency (THz)

250 235 220 205 190 175
1,200 nm 1.0
n=1 n=1 n=0
1,350 nm 0.5
Index of refraction
0
1,475 nm
–0.5
1,590 nm
–1.0
1,700 nm
–1.5
1,200 1,400 1,600 1,800
–5 0 5 10 –5 0 5 10
Wavelength (nm)
Position (mm) Position (mm)
(a) (b) (c)
Fig. 4.45 Experimental results and FDTD simulations. Fourier-plane images of the beam for the window (a) and prism sample
(b) for various wavelengths. The horizontal axis corresponds to the beam shift d, and positions of n = 1 and n = 0 are denoted
by the white lines. The image intensity for each wavelength has been normalized for clarity. (c) Measurements (circles wuth
error bars after four measurements) and simulation (solid line) of the fishnet refractive index. From Valentine et al. (2008).
Copyright c 2008 Nature Publishing Group. For coloured version see plate section
and Martin 2001; Kottmann et al. 2001 both of regular and non-regular
shape. They concluded that there are many more resonances for a non-
regular shape, as shown in Fig. 4.46 where scattering cross-section is
plotted against wavelength for plane-wave incidence. The particle is a
right-angled triangle of 10 nm base and 20 nm side. The polarization of
the incident electric field is in the plane of the figure perpendicular to the
arrow in the inset. As many as five resonances may be seen, whereas an
ellipse has only one resonance. The electric-field distribution is shown
by a colour code in Fig. 4.47 at the main resonance of 458 nm. Normal-
ization is to the amplitude of the incident field. The amplitude remains
ten times that of the incident field at a 10 nm distance from the sharp
corner.
The effect of the coupling between two cylinders of 25 nm radii made
of silver was also investigated (Kottmann et al., 2000a) as a function of
the distance between cylinders. A plane wave in the direction of the axis
of symmetry was incident with the electric field perpendicular to the axes
of the cylinders. The resonant wavelengths were found to be 350 nm,
358 nm, 368 nm, 380 nm and 404 nm for distances between the cylinders
of 50 nm, 20 nm, 10 nm, 5 nm and 2 nm. A further radiation-induced
resonance was found by Kottmann and Martin (2001).
λ=458nm
Scattering cross section [nm]
λ=392nm
101
λ=369nm
100
λ=329nm
10–1
Fig. 4.46 Scattering cross-section ver-

sus wavelength for a triangle particle. 10–2
From Kottmann et al. (2000b). Copy-
c 2000 Optical Society of Amer-
300 400 500 600
right
ica Wavelenght [nm]
Fig. 4.47 Field map of the electric

field at the main resonance for a tri-
angle particle. From Kottmann et al.
(2000b). Copyright c 2000 Optical So-
ciety of America. For coloured version
see plate section
4.4 A mathematical model and further

experimental results
4.4.1 Distributed circuits
Once the size of a resonant element is small relative to the wavelength the
chances of a successful representation in terms of circuits becomes high.
This was indeed regarded by many authors as an aid to understanding
the physics. Most of the representations were in terms of lumped ele-
ments (see, for example, Baena et al. 2005b). Our aim in this section is
to introduce a distributed circuit model on the lines of Shamonin et al.
2004; Shamonin et al. 2005 to describe the properties of the split-ring
resonator, shown once more in Fig. 4.48(a). The physical relationships
to be embodied in the model are as follows: (i) Conduction currents I1
and I2 in the outer and inner rings vary as a function of the azimuthal
angle ϕ, and are complemented by displacement currents in the gaps
in order to form closed circuits, (ii) the currents in the outer and inner
rings are coupled to each other by a mutual inductance, M , (iii) a dis-
g1 d
r 10 G1 dM
L1 dM
I 1 (M) + - I 1 (M+dM)
~
(a) Ic
g2 V(M) M dM C dM V(M+dM) (b)
+ - ~
w I 2 (M) L dM G dM I 2 (M+dM)
h 2 2
w
y y
x x
Cg1
G1 dM
~
~ M dM ~
~ L1 dM
~ ~
G2 dM
L2 dM C dM
~
~
(c)
dM
~
~
~ Cg2 ~
~
~ ~
~
Fig. 4.48 Split-ring resonator. (a) A schematic view. (b) The equivalent circuit of a dϕ element. (c) The equivalent circuit
c 2004 American Institute of Physics, and Shamonin et al.
of the entire resonator. From Shamonin et al. (2004), copyright
(2005), copyright c 2005 Wiley-VCH Verlag GmbH & Co. KGaA
placement current, Ic , flows between the outer and inner rings, (iv) the
inner and outer rings have inductances L1 and L2 and resistances R1 and
R2 , respectively, (v) the capacitances across the gaps in the outer and
inner rings are Cg1 and Cg2 , respectively, (vi) a spatially constant and
in time harmonically varying magnetic field will induce voltages both in
the outer and inner rings.
We shall now write Kirchhoff’s equations for a dϕ element of the SRR
shown in Fig. 4.48(b). The current equation for the upper line is
I1 (ϕ + dϕ) = I1 (ϕ) − Ic (ϕ) , (4.6)

where
Ic (ϕ) = j ωCV (ϕ), (4.7)

V (ϕ) is the voltage across the ring and C is the inter-ring capacitance
per unit radian. The current equation for the lower line is
I2 (ϕ + dϕ) = I2 (ϕ) + Ic (ϕ) . (4.8)

From eqns (4.6)–(4.8) we obtain the differential equations
dI1 dI2
= −j ωCV and = j ωCV, (4.9)
dϕ dϕ
whence the physically obvious relationship
I1 + I2 = I0 (4.10)
follows. I0 is a constant that will have to be determined from the bound-
ary conditions. We can similarly write the voltage equation in the form
F1 − F2
V (ϕ + dϕ) = V (ϕ) + I1 Z1 − j ωM I1 − I2 Z2 + j ωM I2 + , (4.11)
2π
where
Z1 = R1 + j ωL1 , Z2 = R2 + j ωL2 . (4.12)

F1 and F2 are the induced voltages in the outer and inner rings, and are
equal to
F1 = j ωµ0 Hπr12 and F2 = j ωµ0 Hπr22 , (4.13)

where r1 and r2 are the inner radii of the outer and inner rings. Note
further that the symbols in eqns (4.9)–(4.12), C, R1 , R2 , L1 , L2 and M
all refer to values per radian.
Expressing I2 in terms of I1 from eqn (4.10) we can derive a differential
equation from eqn (4.11) in the form
dV F1 − F2
− = I1 (Z1 + Z2 − 2j ωM ) − I0 (Z2 − j ωM ) − . (4.14)
dϕ 2π
We have now two first-order differential equations in I1 and V from
which we may obtain a second-order differential equation with constant
coefficients in either variable that can be solved in terms of constants
and trigonometric functions. To continue we need the full equivalent
circuit shown in Fig. 4.48(c) which will help to formulate the boundary
conditions.
Next, let us look in more detail at what happens at ϕ = 0 (note
that in order to exploit symmetry our azimuthal co-ordinate system
runs from −π to 0, and from 0 to π. According to our model the gap
capacitor Cg1 is exactly at the position ϕ = 0 and has zero spatial
extent. There is, however, bound to be a voltage drop across it due to
the displacement current flowing through it. Consequently, the inter-ring
voltage cannot be continuous at ϕ = 0. The current must be continuous
but not necessarily differentiable. It follows then that we need to solve
the differential equations separately for the −π to 0, and for the 0 to π

regions. The corresponding functions will be called V− , I1− , V+ , I1+ ,
respectively. The boundary conditions will then take the form: first for
the currents
I1+ (0) = I1− (0) , I2+ (0) = I2− (0) ,

I1+ (π) = I1− (−π) , I2+ (π) = I2− (−π) , (4.15)
and for the voltages
j ωCg1 (V− (0) − V+ (0)) = I1 (0) ,

j ωCg2 (V− (−π) − V+ (π)) = I2 (π) . (4.16)
We have taken care of the currents and voltages but not, as yet, of the
excitation. For that we shall have to write Kirchhoff’s voltage law either
along the inner or the outer ring, which will include the excitation. The
condition for the total voltage to be zero along the outer ring may then
be written as
Z0 Zπ
 
I1 (0)
(Z1 +j ωM )  I1− dϕ + I1+ dϕ −2j ωM πI0 + = F1 . (4.17)
j ωCg1
−π 0
In principle everything is easy. We are concerned only with differen-

tial equations of constant coefficients, and the boundary conditions are
also constants. Hence, the analytic solution is straightforward although
rather laborious, and the resulting expressions are rather long. We shall
not show them here but will plot a number of relevant curves later.
Next, we shall look at the eigensolutions that occur in the absence
of excitation. The linear equations for the voltages and currents are
then homogeneous, and the condition for a solution to exist is that the
determinant has to vanish. The resulting characteristic equation then
takes the form (Shamonin et al., 2004),
κ sin κπ[4πκ2 − πγ1 γ2 − 2γ1 ν2 − 2γ2 ν1 ]

+ cos κπ[−2πκ2 (γ1 + γ2 ) + γ1 γ2 (ν1 + ν2 )] = −2γ1 γ2 ν12 , (4.18)
where
C C
κ2 = ω 2 Leq C , ,
γ1 = γ2 = ,
Cg1 Cg2
(L1 − M )2 (L2 − M )2
ν1 = , ν2 = ,
L L − M2 L1 L2 − M 2
√1 2
ν12 = ν1 ν2 , Leq = L1 + L2 − 2M . (4.19)
We shall look at the solution of this equation for a set of parameters

further below. First, let us find an approximate solution for the low-
est resonant frequency, i.e. for the lowest value of κ. We can get that
approximate solution from eqn (4.19) by assuming that κ is small. Ex-
panding then sin κπ and cos κπ in a series and neglecting terms in κ4
and higher, eqn (4.19) reduces to
1
ω2 = . (4.20)
π2
(L1 + L2 )C + 2πL1 Cg1 + 2πL2 Cg2
2
The remarkable thing is that Leq (the only term that contains the mutual
inductance M ) does not appear in eqn (4.20), i.e. the lowest resonant
frequency is independent of the mutual inductance and depends only on
the individual inductances. Since in an SRR the radii of the outer and
inner rings are close to each other we can make the further approximation
that L1 = L2 = L, in which case eqn (4.20) takes the form
Lt
1
ω2 = π . (4.21)
2πL C + Cg1 + Cg2
2
Note that L and C are per radian. Thus, if we want to use the total
Ct/2 Ct/2 inductance and the total inter-ring capacitance of the SRR then it is
Lt = 2πL and Ct = 2πC with which eqn (4.21) takes the final form
Cg1 1
ω2 = . (4.22)
Ct
Lt + Cg1 + Cg2
Cg2 4
The conclusion is now clear. The lowest resonant frequency of a SRR
is given by the equivalent circuit shown in Fig. 4.49, a generalization
Fig. 4.49 Equivalent circuit of that of Fig. 4.13(a). The inter-ring capacitance between two splits is
Ct /2. These two capacitances in series are then connected in parallel
with the two gap capacitances. Note that eqn (4.22) is a generalization of
the expression given by eqn (2.26), derived intuitively by Marques et al.
(2002b). It now includes the gap capacitances as well and, interestingly,
8
Having set up a model for a SRR we in a very simple manner.8
can easily extend the same principles
to the modelling of a singly split dou-
ble ring that also consists of two con- 4.4.2 Results
centric rings but differs from the SRR
by having only one of the rings split. The main interest is in the lowest resonant frequency because the aim
In fact, the inner ring being split is an is to produce a medium describable by effective material parameters.
old design due to Hardy and Whitehead The smaller the elements the more applicable effective-medium theory
(1981) that has already been shown in
Fig. 4.5. The analysis, quite similar to is. We shall now look at a few examples. A schematic drawing of the
that presented above, was performed by SRR has already been shown in Fig. 4.48(a). We shall now choose the
Shamonin et al. (2004). The conclu- dimensions as follows: Its height is h = 5 mm. The external radius of
sions were similar too. The main dif-
the outer ring is re1 = 11 mm and the wall thickness is 1 mm. The gaps
ference relative to the SRR was that
having only one split the resonance fre- are 1 mm both in the inner and in the outer ring. The wall thickness of
quency for the same parameters was the inner ring is 0.8 mm. We shall look at three examples in which the
higher. dimensions enumerated above are all identical but the external radius of
the inner ring takes three different values, namely re2 = 7.5, 8.5 and 9
Table 4.3 Parameters and resonant frequencies of split-ring resonators with variable inner-ring
radius
re2 [mm] L2 [nH/rad] M [nH/rad] C [pF/rad] f0 [GHz] f1 [GHz] f2 [GHz]
7.5 2.74 1.99 0.239 1.50 3.88 7.97

8.5 3.34 2.70 0.382 1.19 3.80 7.72
9 3.69 3.49 0.550 0.97 4.50 8.96
mm. Having specified the elements we need to determine the parameters

that come into the analytical theory presented above: the capacitances
and inductances. They can be determined with the aid of formulae and
tables available in the literature (see, e.g., Grover 1981; Hammond and
Sykulski 1994). The inductance of the outer ring is found as L1 = 4.90
nH/rad, and the gap capacitances as Cg1 = 0.106 pF and Cg2 = 0.092
pF. The other parameters that depend on re2 are given in the second to
fourth columns of Table 4.3, all per radian. Having obtained the values of
all the parameters we can determine the resonant frequencies by solving
numerically eqn (4.18) for ω. The equation, being transcendental, has
an infinite number of solutions. The first three resonant frequencies
obtained are given in columns 5–7 of Table 4.3. As expected, the lowest
resonant frequency occurs for case 3 where the inter-ring separation is
minimum and the inter-ring capacitance is maximum.
A qualitative explanation in terms of circuits has already been given
for the fundamental resonance. For the first and second higher-order res-
onance the physical explanation is that an electromagnetic wave travels
round in the inter-ring space. The mechanism is somewhat similar to
that in a strip-line resonator (Chang, 1996) but in the present case they
are capacitively loaded. The result is that the first higher-order reso-
nance occurs at a wavelength considerably less than the perimeter. On
the other hand, the second higher-order resonance has a frequency twice
as large as the first-order resonance, showing clearly that wave propa-
gation plays a dominant role.
To check these predictions and to have a better understanding of the
physics we shall plot the variation of the voltages and currents as a
function of angle in Fig. 4.50 as follows from our theory. The parameters
are those given above for re2 = 7.5 mm. The corresponding resonant
frequencies are given in Table 4.3 as f0 = 1.5 GHz, f1 = 3.88 GHz and
f2 = 7.97 GHz. The frequencies at which the variables are plotted are
0.05, 1.5, 2.5, 3.02, 3.88 and 7.97 GHz corresponding to Figs. 4.50(a)–(f).
Let us look at the currents first (upper figures). Our normalization is
to the sum of the two currents: we shall take I0 = 1. At 0.05 GHz the
linear approximation may be seen to be quite good. The surprising thing
is that the approximation is still good at the lowest resonant frequency
of 1.5 GHz and it is still not too bad at a frequency of 3.02 GHz, al-
though by then the currents are more like sinusoidals. The major change
1 1
current
current
0.5 0.5
0 (a) 0 (b)
1 1
voltage
voltage
0 0
−1 −1
−180 −90 0 90 180 −180 −90 0 90 180
M [deg] M [deg]
1 1
current
current
0.5 0.5
0 (c) 0 (d)
1 1
voltage
voltage
0 0
−1 −1
−180 −90 0 90 180 −180 −90 0 90 180
M [deg] M [deg]
4
current
current
2 5
0 0
−2 −5
1 (e) 1 (f)
voltage
voltage
0 0
−1 −1
−180 −90 0 90 180 −180 −90 0 90 180
M [deg] M [deg]
Fig. 4.50 Voltage and current distributions. fr = 50 MHz, 1.5, 2.5, 3.02, 3.88, 7.97 GHz (a)–(f)
in the current distribution occurs at the first higher resonance of 3.88

9
We are actually slightly off resonance GHz.9 Note that the currents are now in the opposite direction, a clear
because in the absence of losses all the indication that the physics has changed. We have moved away from the
amplitudes would be infinitely large at
the exact resonant frequency.
LC resonance to a resonance that is more akin to those in transmission
lines. At f2 , the second higher resonance, as we may expect, the current
has a double period.
The voltages, plotted in the lower figures, tell a similar story. Note
that normalization is now to the maximum amplitude. The inter-ring
voltage is supposed to be constant according to the qualitative explana-
tion. In fact, it is somewhat different from constant even at the lowest
frequency investigated. On the other hand, we may see that at 50 MHz
and even at 1.5 GHz, the lowest resonant frequency, the voltage jumps
at the two gaps (at 0◦ and at 180◦) from positive maximum to negative
maximum and vice versa. At 2.5 GHz the voltage jump at 0◦ is smaller
10
This particular value of frequency and it is entirely absent at 3.02 GHz.10 After 3.02 GHz the voltage jump
was chosen with the explicit aim of find- reverses. At the first higher resonance the voltage period halves and it
ing a continuous variation of voltage
with just one discontinuity at 180◦ .
halves again at the second higher resonance.
Next, we shall discuss a set of experiments by Radkovskaya et al.
A1 A2 B1 B2
re Fig. 4.51 Schematic representation of

ri the elements from which resonators are
h constructed. From Radkovskaya et al.
(2005). Copyright c 2005 Wiley-VCH
g Verlag GmbH & Co. KGaA
Table 4.4 Values of measured and calculated resonant fre-

quencies in GHz for the various resonators. From Rad-
kovskaya et al. (2005)
resonator experimental analytical numerical
B1 1.74 1.77
B2 2.59 2.67
B1 A2 1.89 2.17 1.94
A1 B2 2.78 2.89 2.84
B1 B2 1.44 1.50 1.438
(2005) in which four copper rings (Fig. 4.51) were used in different com-
binations. The dimensions of the rings may be recognized as those we
used in our earlier theoretical calculations with re2 = 7.5 mm. Out of the 11
For the same sets of resonators fur-
four rings it is possible to construct five resonators, as shown schemat- ther results (measuring the quality fac-
ically in Table 4.4. B1 A2 is the notation for the full ring A2 placed tor, Q) were obtained by Hao et al.
concentrically inside the split ring B1 . Similarly, A1 B2 means the split 2005b; Hao et al. 2005a.
ring B2 inside the full ring A1 , and B1 B2 means the smaller split ring
B2 inside the larger split ring B1 with splits at opposite sides. The last
one is of course a SRR. 12
MICRO-STRIPES is a registered
The purpose of the experiments was to measure the resonant frequen- trademark of Flomerics Ltd., Surrey,
cies of all four configurations.11 They are shown in column 3 of Table 4.4, UK.
whereas the resonant frequencies obtained by the MICRO-STRIPES12
numerical package are in column 5. As expected, B2 , the smaller split
ring has a higher resonant frequency than B1 . The smaller the rings
−2 −1 0 the higher the resonant frequencies. One would indeed expect simple in-
verse scaling of the resonant frequency with the radii. Note from Table
4.4 that the ratio of the inner radius of ring B1 to that of B2 is 1.49,
and the inverse ratio of the corresponding resonant frequencies is also
1.49. It may also be seen that in the configurations B1 A2 and A1 B2
the resonant frequency is determined by the size of the split ring. The
(a)
ratio of the corresponding resonant frequencies of 2.78 GHz and 1.89
GHz is 1.47, very close to the ratio of the radii. Thus, the influence of
the full inner and outer rings is merely to increase a little the resonant
frequency. The lowest resonant frequency is 1.44 GHz, obtained by the
SRR. The agreement between the experimental and simulation results is
quite close. The maximum deviation is for the B2 case, which amounts
to about 3%.
We have no analytic expressions for the resonance frequencies of the
single rings B1 and B2 . The characteristic equation (eqn (4.18)) makes
it possible to determine the resonant frequency, as follows from our dis-
tributed circuit model. The resonant frequency obtained from that is
(b) 1.50 GHz, in contrast to the measured 1.44 GHz. We have not shown the
analytic expressions for the singly split double rings of the A1 B2 and the
A2 B1 configurations, but they are available in Shamonin et al. (2004).
The resonant frequencies calculated from there are given in Table 4.4 as
2.17 and 2.89 GHz, which are still reasonable approximations.
For the resonant frequency of the SRR three more approximate ex-
pressions are available, that of eqn (4.22), and those of Pendry et al.
(1999) and Sauviac et al. (2004). We shall give here the expression of
Pendry et al. (1999) which is in the simple form of
r
c 3d
ωr = , (4.23)
(c) r π2 r
where d = ri1 − re2 , r = (ri2 − re2 )/2. The above equation yields a
resonant frequency of 1.62 GHz, far too high, whereas the expression
given by Sauviac et al. (2004) yields 1.18 GHz. On the other hand,
eqn (4.22) yields quite a good approximation of 1.35 GHz. For the
A1 B2 configuration there is also a simple analytical formula as given by
eqn (4.3). It gives a resonant frequency of 6.4 GHz, quite far from the
Fig. 4.52 Electric-field distributions at measured 2.78 GHz. The probable reason for the large discrepancy is
the fundamental (a), the first (b) and that they attributed too much importance to the gap capacitance, which
the second higher resonance (c). Po- was indeed quite large in their design shown in Fig. 4.5.
sitions of the voltage zeros are shown
by white dashed lines. From Hesmer
(2008). For coloured version see plate 4.4.3 A note on higher resonances
section
In Table 4.3 we gave the analytical results both for the fundamental
and for the higher-order resonances. We have also determined them by
13
CST MICROWAVE STUDIO is simulations using the numerical package CST MICROWAVE STUDIO13
a registered trademark of CST– (Hesmer, 2008). It is reasonable to expect good agreement for the fun-
Computer Simulation Technology AG,
Darmstadt, Germany
damental resonance, but the distributed circuit model is less likely to
be valid for the higher-order resonances. The analytical results from
Table 4.3 are 1.50, 3.88 and 7.97 GHz. Our numerical simulations gave
1.46, 3.42 and 7.68 GHz, which shows larger discrepancies for the higher
orders.
The simulations also provided the field distributions at the resonant
frequencies. They are shown in Fig. 4.52. It may be immediately seen
that the field distribution has no zeros for the fundamental resonance,
two zeros for the first higher-order resonance and four zeros for the sec-
ond higher-order resonance. Looking at Figs. 4.50(e) and (f) we can see
that the number of zero points is the same as in the analytical solution.
We can also compare the positions of the zeros. The analytical calcu-
lations give (they are symmetric around the zero angle so we shall give
only the positive values) for f1 74.2◦ , and 26.2◦ and 147◦ for f2 . In
contrast the simulations yield 62◦ for the first and 19.5◦ and 145◦ for
the second higher resonance. Not a close agreement, but all are in the
same ballpark.
Subwavelength imaging
5
5.2 The perfect lens:
The subject of metamaterials started with Veselago’s (1967) introduc- controversy around
tion of negative refractive index and his simultaneous proposal for a flat the concept 157
lens. We have already discussed the flat-lens family in Section 2.11.4. 5.3 Near-perfect lens 159
Its characteristic feature is the existence of a focus inside the lens. The 5.4 Negative-permittivity
lens 176
basic principle of its operation has already been shown in Fig. 2.24; for
convenience it is shown again in Fig. 5.1. The essential requirement is
that the index of refraction is n = −1. The ray trajectories follow then 5.6 Magnifying multilayer
superlens 192
in the geometrical optics approximation. A focal point is brought to a
focal point and an object in the object plane is reproduced perfectly in
the image plane. This is an obvious advantage of the flat lens. It has no
optical axis. A simple illustration in Fig. 5.1 shows an object consisting
of three points in the object plane that is perfectly reproduced. At the
same time we need to mention a serious disadvantage of the flat lens.
It can work only at a single frequency because the n = −1 condition is
very strongly frequency-dependent. In traditional imaging language this
means that the flat lens has very large chromatic aberration.
As follows from the construction in Fig. 5.1, Veselago’s lens is perfect
when we consider propagating waves only. But, if the object is small
relative to the wavelength it will have both propagating and evanescent
components in its Fourier spectrum. The image is not perfect because
the subwavelength features have not been reproduced. It was shown by
Pendry (2000) that a flat lens in the Veselago geometry can reproduce
ș1 ș1
ș1 ș1
z
Fig. 5.1 Imaging three point sources
0 d/2 3d/2 2d with Veselago’s flat lens
156 Subwavelength imaging
1
There is some problem here with ter- subwavelength features on account of the evanescent waves growing1
minology. In Section 2.12 we have al- in the lens. This was already discussed in Section 2.12. Here, we shall
ready objected to the term ‘amplifica-
tion’ because there is no source of ad-
assume that the reader has already read that section, and is also familiar
ditional power there. Strictly speak- with surface plasmons discussed in Chapter 3. Later in this chapter we
ing we should not talk about growing shall of course look at subwavelength imaging from a number of different
evanescent waves either. To evanesce angles but here we shall start with a brief summary in the form of
means to fade away. A growing evanes-
cent wave is a contradiction in terms.
questions and answers.
So does it grow or does it evanesce? Firstly, what is the difference between Veselago’s flat lens and Pendry’s
It can’t do both. The correct descrip- perfect lens? The geometry is the same but not the parameters of the
tion would be the growth of the ampli- lens. Veselago’s lens will work whenever the index of refraction is n =
tude of the non-propagating waves in
the +z direction. Unfortunately, that’s −1. For Pendry’s perfect lens it is required for both material parameters,
quite a mouthful and goes against cur- εr and µr , to be equal to −1. Pendry’s lens is always matched. If
rent practice. Thus, with regret, we in Veselago’s lens the n = −1 condition is obtained by both material
shall often refer to growing evanescent parameters being equal to −1 then for a lossless lens there is no difference
waves.
between the Veselago and Pendry variety.
Secondly, why is the total phase change from object to image in
Pendry’s lens equal to zero? Because the wave in the lens is a back-
ward wave that has the same phase velocity as the forward waves from
object to lens, and from lens to image. Since the path in the lens is
equal to the path in air the total phase change is zero.
Thirdly, why is the total amplitude of a non-propagating space har-
monic from object to image equal to unity? This is because the rate of
increase in the lens is equal to the rate of decrease in air.
Fourthly, is a flat transfer function possible in practice? It is not
possible in practice and actually not in theory either because it implies
infinitely high fields.
In the fifth place, what will prevent the transfer function from be-
ing flat? Losses, because the growing non-propagating waves will be
attenuated and therefore perfect compensation of amplitudes becomes
impossible. Also, any deviation from the ideal conditions will cause the
appearance of a cutoff in the transfer function.
In the sixth place, why is there a chance to build a subwavelength lens
from a material in which only εr or only µr is equal to −1? It follows
from the properties of TM waves that their amplitudes depend strongly
only on ε and vice versa for TE waves.
In the seventh place, why does the perfect lens need both material
parameters to be −1 if with the right polarization the amplitudes depend
strongly only on one of the material parameters? The reason is that the
phase velocity of propagating waves and the rate of decay (or growth)
of non-propagating waves depend on both material parameters.
In the eighth place, why do we have growing non-propagating waves
in the epsilon-negative-only lens? Because surface plasmon–polaritons
(SPPs) are excited at the far boundary and they cannot be excited by
a propagating wave owing to their shorter wavelength.
In the ninth place, why is it necessary for the epsilon-negative-only
lens to be thin? This is because the SPP on the rear surface can only
be excited by an input wave if the front and rear surfaces are strongly
coupled, and they are strongly coupled only when they are near to each
5.2 The perfect lens: controversy around the concept 157
other.
In the tenth place, why is a multilayer lens preferable to a single-layer
lens? Because the SPP interaction between the front and rear surfaces is
still active for consecutive thin lenses, and at the same time the distance
between the object and image increases.
The above is a rather rough and very concise summary of what has
been said so far in this book about imaging by a flat lens. The rest of
the chapter will of course go into a lot of details. Section 5.2 will be
concerned with history, with the controversy that arose in the wake of
the publication of Pendry’s paper (2000). The most important part of
this chapter is probably Section 5.3, which will describe in great detail
what happens to the transfer function when the parameters deviate from
the ideal ones. Also in the same section, we shall establish relationships
between phenomena in the perfect lens and the physics of SPPs. In
Section 5.4 we shall investigate the quality of images under conditions
when only the permittivity is equal to −1. Section 5.5 is concerned
with the improvements a multilayer lens can offer. Section 5.6 discusses
magnification. Section 5.7 is an attempt to show and briefly discuss
some of the widely held beliefs that may not stand closer scrutiny.
5.2 The perfect lens: controversy around

the concept
Pendry’s paper on the perfect lens has been the main driving force in
the rapid development of the field of metamaterials.2 It literally stirred 2
Published in October 2000, this paper
up the scientific community and caused an avalanche of papers devoted was quoted over 2000 times in the less
than eight years that passed since.
to various aspects of the concept of the perfect lens. At first, the con-
cept was challenged by a large number of authors who questioned the
correctness of the approach and looked at flaws in Pendry’s arguments.
The controversy around the idea of the perfect lens has had many facets.
5.2.1 Battle of wits

The possibility of negative refraction and that of the perfect lens were
attacked afterwards in several publications (see, e.g., Williams 2001;
Garcia and Nieto-Vesperinas 2002; Valanju et al. 2002; Pokrovsky and
Efros 2002c). An observer could witness a battle of research papers with
quick-witted titles, which showed that scientists are good at puns and at
repartees: ‘Negative reaction to negative refraction’ (Cartlidge, 2002),
‘Wave refraction in negative-index media: always positive and very in-
homogeneous’ (Valanju et al., 2002), ‘Perfect lens in a non-perfect world’
(Kik et al., 2002), ‘Near-sighted lens’ (Podolskiy and Narimanov, 2005),
‘Entering the Negative Age’ (Pendry, 2001), ‘Optics: positively negative’
(Pendry, 2003a). ‘Perfect lenses made with left-handed materials: Al-
ice’s mirror?’ (Maystre and Enoch, 2004). For a while, the controversy
raged unabated, the ‘perfect lens’ was demoted to a ‘near-perfect lens’
(Ramakrishna and Pendry, 2002). Fairly soon the tone of the papers
changed: The goal was no longer solely to prove that the perfect lens
concept has flaws (it surely does, as nothing is ever perfect in nature),
but rather to understand the underlining physical mechanisms of sub-
wavelength imaging and to explore the possibilities offered by metama-
terial near-field lenses. Most researchers agree by now that the concept
of subwavelength imaging by a flat lens is both novel and useful.
5.2.2 Non-integrable fields

The concern about non-integrable quantities was raised, among oth-
ers, by ‘t Hooft 2001; Pokrovsky and Efros 2002a; Pokrovsky and Efros
2002b; Garcia and Nieto-Vesperinas 2002 and Haldane 2002. The essence
of the problem is as follows. Considering wave propagation from an
object plane via air, metamaterial slab, and air again, the field is de-
composed into Fourier components with kx values ranging from zero
to infinite values. The total electrical field at each point in space is
calculated by performing integration (or summation in the discretized
formulation of the problem) over all possible wave vectors. The trans-
fer function from the object plane to the image plane being flat means
that all evanescent (non-propagating) components, no matter how large
their kx vectors, are all growing inside the lens. As already described
in Section 2.12.2, for the case εr = µr = −1, the kz components of the
wave vectors in the air and in the metamaterial are purely imaginary for
sufficiently large values of kx . This turns a square-integrable incoming
wave into a non-integrable one. The mathematics is quite simple. We
consider a square-integrable field with Fourier components Ex (kx ) at
the source plane z = 0. Each wave component that declined by a factor
exp(−|kz |d/2) before reaching the entrance surface of the metamaterial
is then growing in the slab and reaches at the rear surface z = 3d/2
a value of exp(|kz |d) times higher than in the input signal. A sum-
mation over all Fourier components yields an integral that diverges for
large wave vectors. This problem of non-integrable quantities occurs,
however, solely for the case with εr and µr being precisely equal to −1.
For any other combination of εr and µr the remedy is provided by the
high-frequency cutoff.
5.2.3 High spatial frequency cutoff

Common sense suggests that the divergences mentioned above can be
cured by losses. Due to losses the amplitudes of the various space har-
monics are not faithfully reproduced causing a high spatial frequency
cutoff. An immediate consequence of the existence of this high-frequency
cutoff is that the image is no longer perfect, as the spatial components
above the cutoff die out before having a chance to reach the image plane.
Are there any other reasons for a cutoff frequency? We shall mention
some of the factors below.
Periodic structure. A constraint that comes immediately to mind
is that the metamaterial is not homogeneous as discussed already in
Section 2.12. It is a heterogeneous structure composed of unit cells

with finite dimensions. There must be a length scale for any practical
realization of a metamaterial below which the homogeneous description
is no longer applicable. Thus, components with too large kx values above
the characteristic value in the reciprocal space related to the periodicity
of the material will be subject to a cutoff. If the period is a, there can be
no chance whatsoever to reproduce a Fourier component at kx = 2π/a.
Dispersion. The material constants ε and µ are bound to be dispersive
in the region where they are negative. We know further (Kramers–
Kronig relations, see, e.g., Landau and Lifschitz 1984) that any variation
of the real parts of ε and µ with frequency is accompanied by variation
in the imaginary part.
Deviations from the ideal conditions. The three reasons for the exis-
tence of a high spatial frequency cutoff enumerated above are perhaps
not trivial but with a little thought devoted to the problems arising they
are quite predictable. The effect of deviations from ideal conditions of
εr = −1 and µr = −1 is less obvious but, as will be shown later, it also
leads to the appearance of a cutoff frequency.
Build-up time. The starting point of Gomez-Santos’ (2003) analysis
was the belief that the concept of the perfect lens can be rescued. He
follows Haldane (2002) in attributing a crucial role to the surface plas-
mons and particularly to their near degeneracy as kx → ∞. Working
in the time domain and using a model of coupled oscillators he proves
that high resolution is obtained at the price of delay. Smaller details
take longer and longer time to develop. Infinitely high resolution needs
infinitely long time for its realization.
Only one material parameter being equal to −1, say εr = −1, µr = 1.
If only ε is negative and µ is positive then, as mentioned earlier, the rate
of decay in air is not the same as the rate of growth in the lens, leading
to a cutoff in the transfer function. The claim (Pendry, 2000) that for a
lossless lens with εr = −1, µr = 1 perfect imaging is still possible for an
incident TM wave is not valid in general.
5.3 Near-perfect lens

5.3.1 Introduction
Let us sum up the conclusions about the perfect lens we have reached
so far. The new feature we have learned is that when the ideal lens
(εr = µr = −1) is excited by an incident field from a source with sub-
wavelength features, the ‘perfect lens’ structure would not provide at
once a perfect image in the image plane. Instead, as time goes by, a
field distribution corresponding to a solution with continuously improv-
ing resolution would build up. The higher the desired resolution, the
more patience (and more energy from the source field) is required.
So there is one more reason why the perfect lens is an idealization that
can never be reached. The next question that can be legitimately asked
is how well a near-perfect lens would operate whose material parameters
differ slightly from the ideal constellation. In other words to what ex-
tent imaging beyond the diffraction limit is possible using a fabricated
material and to what extent the process can be simulated by numerical
methods that inevitably approximate the ideal situation.
To familiarize ourselves with the operation of a homogeneous meta-
material slab acting as a near-perfect lens we will from now on consider
only the stationary state. It follows from our previous discussion in
Section 5.2 that we can do it with a clear conscience: a steady state
would exhibit a high-frequency cutoff ensuring that all fields are square-
integrable.
We shall now investigate the functioning of the lens by making use
of various tools of analysis. Firstly, the transfer function provides the
information on how well (or how badly) the input field in the plane
z = 0 is reproduced in the image plane z = 2d. Secondly, the reflection
coefficient (identically zero in the perfect-lens idealization) when it is
different from zero indicates the existence of a surface mode at the front
surface. Thirdly, variation of individual Fourier components across the
slab helps to visualize the mechanism of what is going on inside the slab.
This section will be divided into a number of subsections. In Sec-
tion 5.3.2 we shall provide the equations needed. The effect of losses
on the transfer function, and in particular on the cutoff frequency, is
treated in Section 5.3.3, then in Section 5.3.4 we shall return to the loss-
less case and investigate the effect of deviations from the ideal values of
εr = µr = −1 followed by an investigation of the deviations from the
ideal refractive index of n = −1, and finally the case is examined when
ε and µ vary in such a manner that their product remains the same,
i.e. the refractive index does not change. In Section 5.3.5 the effect of
lens thickness is investigated in the presence of losses. The influence
of the various kinds of deviations is summarized in Section 5.3.6. An
interesting and important point is raised in Section 5.3.7 that may cheer
up those sceptical about the usefulness of numerical simulations. The
problem seems to be an inherent mismatch in the constitutive parame-
ters of the slab and the surrounding medium that may lead to incorrect
results.
5.3.2 Field quantities in the three regions

We need the expressions for all components of the field amplitudes. We
consider for simplicity TM polarization keeping in mind that similar
expressions can be easily derived for TE polarization. We have already
derived the solution to Maxwell’s equations in a slab in Chapter 1 and
discussed eigenmodes of the solution, SPPs, in Chapter 3. We quote
here expressions needed for our analysis.
The fields from a source in the plane z = 0 are expanded in a Fourier
series over propagating and evanescent waves, i.e. those with kx ≤ k0
and with kx > k0 , all of which are proportional to exp(j ωt) in the steady
state,
A(kx ) e −j kx x .
X
Hy (x, z = 0) = (5.1)
kx
The field expansion of the full solution everywhere in the xz plane takes
the form
A e −j kz1 z + B e j kz1 z e −j kx x
X
Hy1 (x, z) =
kx
(medium 1) ,
C e −j kz2 z + D e j kz2 z e −j kx x
X
Hy2 (x, z) =
kx
(medium 2) ,
F e −j kz3 z − j kx x
X
Hy3 (x, z) =
kx
(medium 3) . (5.2)
A = A(kx ) and B = B(kx ) are the amplitudes of the incident and
reflected waves of the Fourier component kx in medium 1, C = C(kx )
and D = D(kx ) are the corresponding amplitudes of the transmitted
and reflected waves in medium 2 (slab) and F = F (kx ) is the amplitude
of the transmitted wave in medium 3. The corresponding components
of the electric field are
X kz1
A e −j kz1 z − B e j kz1 z e −j kx x

Ex1 =
ωε1
kx
(medium 1) ,
X kz2
C e −j kz1 z − D e j kz1 z e −j kx

Ex2 =
ωε2
kx
(medium 2) ,
X kz3
Ex3 = F e −j kz3 z − j kx x
ωε3
kx
(medium 3) (5.3)
and
X kx
A e −j kz1 z + B e j kz1 z e −j kx x

Ez1 = −
ωεr1
kx
(medium 1) ,
X kx
C e −j kz1 z + D e j kz1 z e −j kx x

Ez2 = −
ωεr2
kx
(medium 2) ,
X kx
Ez3 = − F e −j kz3 z − j kx x
ωεr3
kx
(medium 3) . (5.4)
We wish to see what impact minor deviations from the ideal perfect
lens constellation would have, therefore medium 1 and 3 are assumed to
be identical (in the simplest case with εr1 = εr3 = 1, µr1 = µr3 = 1). The
amplitudes B, C, D, F are given by eqns (1.75)–(1.78). In particular,
the transfer function for the distance 2d is
F −j kz1 d
T = e
A
4ζe e −j kz1 d
= , (5.5)
(1 + ζe )2 e j kz2 d − (1 − ζe )2 e −j kz2 d
and the reflection from the front surface of the slab is
B 2j (1 − ζe2 ) sin(kz2 d)
R= = . (5.6)
A (1 + ζe )2 e j kz2 d − (1 − ζe )2 e −j kz2 d
Note that the presence of a pole in the expressions for transmission
and reflection indicates excitation of a SPP mode at the frequency ω
considered.
Next, we shall go through a number of examples of minor perturba-
tions in the perfect-lens parameters.
5.3.3 Effect of losses: Transfer function, cutoff,

electrostatic limit
Losses will be characterized by assigning imaginary parts to the values
of permittivity and permeability
εr2 = −1 − j ε′′ , µr2 = −1 − j µ′′ , (5.7)

with ε′′ , µ′′ ≪ 1. As it happens, the transfer function behaves quite
differently in the cases when (i) kx is large but not too large and (ii)
kx → ∞.
The electrostatic limit has been briefly discussed in Section 2.12.4.
The essential part of the argument was that with good approximation
kz1 = kz2 = kz3 = −j kx , (5.8)

and expression (5.5) for the transfer function can roughly be approxi-
mated as
4ζe e −kx d
T = . (5.9)
(1 + ζe )2 e kx d − (1 − ζe )2 e −kx d
Consequently, the first term in the denominator of eqn (5.5) is regarded
as being close to zero. If kx is not too large but still kx ≫ k0 then the
second term of the denominator in eqn (5.9) dominates, leading to a
transfer function
4 e −kx d
T ≃ =1 (5.10)
4 e −kx d
that is entirely flat. But this is only true for this particular range of
not too large values of kx . The situation is quite different in the limit
kx → ∞. The second term in the denominator of the transfer function,
which declines exponentially with kx , becomes smaller than the first term
and eventually vanishes in the limit kx → ∞. The resulting transfer
function in the limit of high kx values is thus given by
4 e −kx d 4 e −2kx d
T ≃ ≃− . (5.11)
(1 + ζe )2 e kx d (ε′′ )2
Thus, in the presence of loss the transfer function declines exponentially
as kx approaches infinity.
The value of kx at which the transfer function (eqn (5.5)) stops being
flat and starts to decline with kx is called the cutoff frequency and can
be estimated from the condition that the first term in the denominator
(which is the dominant one at not too large kx ) and the second term in
the denominator (which dominates as kx → ∞) are equal (so that the
magnitude of the total transfer function reduces to 1/2),
(1 + ζe )2 ≃ (1 − ζe )2 e −2kx d , (5.12)
yielding the cutoff value of the spatial frequency at
1 2
kxcutoff ≃ ln ′′ . (5.13)
d ε
This approximate expression gives already an idea about how much
the deviation of the metamaterial parameters from the ideal situation
εr = µr = −1 and the thickness of the metamaterial slab as well affect
the cutoff frequency. Note that loss in permeability does not enter the
expression for the cutoff frequency in the electrostatic limit. It would of
course appear for TE polarization. We shall discuss this a little later.
A more general formulation can also be easily derived (see Podolskiy
and Narimanov 2005). Introducing
s
kx2
χ= − 1, (5.14)
k02
we rewrite κ1 and κ2 as
κ1 = k0 χ , (5.15)
ε′′ + µ′′

κ2 ≃ k0 χ 1 − j , (5.16)
2χ2
so that
ε′′ (1 + 2χ2 ) + µ′′

ζe ≃ −1 + j , (5.17)
2χ2
′′ 2
ε (1 + 2χ2 ) + µ′′
(1 + ζe )2 ≃ − , (5.18)
2χ2
(1 − ζe )2 ≃ 4. (5.19)
The transfer function (eqn (5.5)) takes the form

1
T ≃ 2 . (5.20)
ε (1 + 2χ2 ) + µ′′
′′

e 2k0 χd +1
4χ2
A direct comparison with the exact solution eqn (5.5) proves that this
approximate expression for the transfer function fails for small kx in
the vicinity of k0 , but it does indeed provide excellent results for the
cutoff frequency as was proven by Podolskiy and Narimanov (2005).
The cutoff frequency can be found again from the condition that the
transfer function halves there,
2
ε′′ (1 + 2χ2 ) + µ′′

e 2k0 χd = 1. (5.21)
4χ2
Contrary to the electrostatic limit, now both losses in permeability and
permittivity appear in the expression that determines the cutoff fre-
quency. In particular, if only permittivity is lossy, µ′′ = 0, we find
′′ 2
2k0 χd ε (1 + 2χ2 )
e = 1, (5.22)
4χ2
and in addition, assuming kx /k0 ≫ 1, so that χ ≃ kx /k0 , the cutoff
frequency agrees with the value predicted in the electrostatic limit,
1 2
kxcutoff =
ln ′′ , (5.23)
d ε
with logarithmic sensitivity to losses and an inverse dependence on the
thickness. If only permeability is lossy, ε′′ = 0, the cutoff is given by
" 2 #
4 kxcutoff

cutoff 1
kx = ln ′′ , (5.24)
d µ k0
showing a weaker sensitivity to losses.
Case A. Lossy permittivity

Figures 5.2 to 5.5 illustrate in a number of examples the deterioration of
the transfer function with loss. The parameters are d/λ, the thickness of
the slab in terms of wavelength, and the imaginary part of the dielectric
constant that we denote here by δ. The exact solution is plotted by
a solid line and the electrostatic approximation by a dotted line. For
1
10 1.1
0 1.05 −5
10
transfer function |T|

without slab G=10
−1
1
10 −5
G=10
−3 0.95
−2 G=10
10 G=0.1 G=0.1 −3
0.9 G=10
without slab
−3
10 0.85
−4 0.8
10 0 1 0 1
10 10 10 10
k x /k 0 k x /k 0
(a) (b)
Fig. 5.2 Transfer function of a near-perfect lens. d/λ = 0.1, losses only in ε, δ = 10−5 , 10−3 and 0.1. (a) Exact solution
(solid curves) and electrostatic approximation (dotted curves) are shown together with the transfer function for the case when
the slab is removed and the total distance from object to image plane is still the same, 2d (dashed curve). Note that curves
obtained within the electrostatic limit coincide with the exact solution. (b) Detailed view of a part of the graph; only exact
solutions shown
1
10
0
10
−5
G=10
−1 −3
10 G=10
−2 G=0.1
10
without slab Fig. 5.3 Same as in Fig. 5.2(a), losses
−3
10 only in ε, but for d/λ = 0.5. Exact
solution (solid curves) and electrostatic
approximation (dotted curves). Elec-
−4
10 trostatic limit is good for large kx and
0 1
10 10 for small loss. It is less accurate for a
kx /k 0 loss of δ = 0.1
comparison, the transfer function for the case when the slab is removed
is plotted as well (dashed line). In Fig. 5.2(a) a thin slab is taken,
d/λ = 0.1. For a small amount of loss, δ = 10−5 , the transfer function
exhibits a cutoff at kxcutoff /k0 = 100. This means a slab of 0.1λ thickness
with εr = −1−j 0.00001 and µr = −1 would reproduce accurately objects
with a resolution of λ/100. That is pretty good. Increasing the loss
reduces the resolution. For δ = 10−3 the cutoff frequency reduces to
kxcutoff = 10k0 , and for δ = 0.1 to kxcutoff = 3k0 . The resolution of the
slab with δ = 10−3 is therefore λ/10 and for δ = 0.1, which is not an
unrealistic value, only λ/3.
Is such a device better than empty space? Certainly yes; the transfer
function is flatter and more evanescent components can make it to the
image plane even for such a high loss of δ = 0.1. The situation is quite
different for propagating components with kx < k0 . After propagation
1
10
0
10 −5
G=10
−1 −3
10 G=10
G=0.1
−2
10 without slab
Fig. 5.4 Same as in Figs. 5.2(a) −3

10
and 5.3, losses only in ε, but for
d/λ = 1. Exact solution (solid curves)
−4
and electrostatic approximation (dot- 10
0 1
ted curves). Electrostatic limit is only 10 10
reliable for large kx kx /k 0
1
10
−5
G=10
0
10 −3
G=10
G=0.1
−1
10 without slab
−2
10
−3
10
Fig. 5.5 Same as in Figs. 5.2(a), 5.3
−4
and 5.4, losses only in ε, but for d/λ = 10
0 1
2. Electrostatic limit is no longer a re- 10 10
liable approximation kx /k 0
through a lossy slab their amplitude decays, naturally, so the propa-

gating part of the spectrum suffers, while the evanescent part of the
spectrum actually improves in comparison to the empty-space case. To
make this clear in Fig. 5.2(b) only the relevant part of the transfer func-
tion from kx /k0 = 0.3 to 20 is plotted so that small details can be seen.
Does the electrostatic approximation work in this case? Yes, it does, the
exact transfer function and the one obtained in the electrostatic approx-
imation practically coincide (see Fig. 5.2(a)), and the cutoff frequency
can be estimated from the approximate expression (eqn (5.13)).
The transfer function in Figs. 5.3 to 5.5 plotted for thicker slabs,
d/λ = 0.5, 1 and 2, can be seen to deteriorate increasingly faster with
loss. For d/λ = 0.5 the cutoff frequency kxcutoff is 5k0 for δ = 10−5 ,
it is 3.5k0 for δ = 10−3 and just over k0 for δ = 0.1. For a thicker
slab the cutoff frequency is quite close to k0 so that only marginally
subwavelength resolution can be expected, and, in addition, the slab is
getting more and more opaque for propagating components that lose
more and more energy as the slab thickness is increasing. For d/λ ≥ 0.5
the slab with δ = 0.1 does not seem to give much advantage over empty
space. For d/λ = 2, even with small losses, δ = 10−5 the cutoff of
1
10
0
10
−5
G=10
−1 −3
10 G=10
G=0.1
−2
10
without slab
−3
10
Fig. 5.6 Same as in Fig. 5.3, but with

−4
10 losses in µ instead of ε. Electrostatic
0 1
10 10 limit (dotted horizontal line) fails com-
kx /k 0 pletely
kxcutoff /k0 = 1.5 does not make the imaging device look feasible.
As far as the electrostatic limit is concerned, it becomes, as thickness
increases, less and less accurate (Figs. 5.3–5.5). For d/λ = 0.5 the elec-
trostatic limit can be seen to differ considerably from the exact solution
for δ = 0.1. For d/λ = 1 it occurs at δ = 10−3 and for d/λ = 2 the
electrostatic limit δ = 10−5 is no longer capable of reproducing the real
situation.
Case B. Lossy permeability. TM polarization

In the previous example we assumed that only the permittivity is lossy.
What happens if we introduce loss to the permeability instead, taking
µr = −1 − j δ and εr = −1? Figure 5.6 shows the transfer function for
the 0.5λ thick slab with δ = 10−5 , 10−3 and 0.1, with the electrostatic
limit and with the transfer function of empty space. The tendency is the
same, but the cutoff frequency is less sensitive to the loss in permeability
as can be clearly seen by comparing Fig. 5.6 to Fig. 5.3. The cutoff
frequency kxcutoff is 6k0 for δ = 10−5 , it is 4k0 for δ = 10−3 and 2k0
for δ = 0.1. Note that the electrostatic limit still predicts an ideal flat
transfer function. It is not particularly surprising. We should remember
that we consider TM polarization. The only place where µ occurs is in κ1
and κ2 . By neglecting any deviation in κ1 and κ2 from kx we disregard
any deviations of µr from minus unity. The conclusion follows: the
transfer function for TM polarization in the electrostatic approximation
is unaware of any variation in µ and is bound to fail providing unphysical
and unreliable results. This conclusion would apply to any value of µ.
By employing the electrostatic limit we disregard the effect of µ. Are
we allowed to do it? Obviously, we are not: as clearly follows from Fig.
5.6 even minor losses result in a cutoff and this cutoff goes missing in
the electrostatic approximation.
The electrostatic approximation gives simpler formulae and simpler
physics. Unfortunately, it may lead to wrong conclusions. We firmly
believe that there is still a need to clarify the situation. We will address
this important issue later in this chapter when discussing properties of

the silver superlens.
The same would apply in the case of TE polarization to losses intro-
duced to permittivity. The transfer function in that magnetostatic limit
would not account for any change in ε. Clearly, a lens with losses in
permittivity would not be correctly described in the magnetostatic limit
for incident TE polarized light.
5.3.4 Lossless near-perfect lens with εr ≃ −1,

µr ≃ −1
Let us now look at other possible scenarios for a near-perfect lens, this
3
As soon as we go off the perfect lens time disregarding losses3 and introducing small variations to the real
frequency that provides εr = −1, µr = parts of εr and µr from being minus unity.
−1, surface plasmon–polaritons are ex-
cited corresponding to the poles in the
transfer function resulting in infinitely Case C. εr = −(1 ± δ), µr = −1
large amplitudes. Some authors, in or-
der to facilitate numerical calculations, This case is analogous to that considered first by Smith et al. (2003) who
introduce some loss, with the argument for TE polarization assumed the real part of the permeability slightly
that if it is small enough it would not
have any influence on the position of different from the ideal value, keeping the real part of the permittivity
the cutoff frequency (it is true when the at −1. Starting again from the expression (5.5) the transfer function
imaginary parts in ε and µ are much can easily be shown to simplify to
smaller than the perturbations in their
respective real parts). 1
T ≃ 2 . (5.25)
δ 2kx d
e −1
4
The cutoff frequency can be obtained as
1 2
kxcutoff ≃ ln . (5.26)
d |δ|
The transfer function (eqn (5.5)) plotted in Figs. 5.7(a) and (b) for a
number of values of δ (both positive and negative) has some common
features with previously shown results for the lossy near-perfect lens.
The cutoff is inversely proportional to the thickness and depends loga-
rithmically on the deviation from εr = µr = −1.
The new feature is that the transfer function exhibits resonances, one
resonance in Fig. 5.7(a) and two in Fig. 5.7(b). The origin of these
resonances is the excitation of eigenmodes of the slab, the SPP modes.
The direct excitation of the SPP resonances is, especially if losses are
low, undesirable for imaging. The corresponding kx components will be
disproportionately represented in the image. Yet, the existence of these
surface modes for imaging is essential, as the recovery of the evanes-
cent modes can be seen as the result of driving the surface modes off
resonance.
Shouldn’t we expect to see always two resonances, one close to the
light line, and another one at high kx ? Not necessarily. Looking at
Figs. 3.35 and 3.36 helps to understand this. For those frequencies for
which both εr2 and µr2 are negative and their product is larger than
1 1
10 10
0 0
10 10

H r = -1.00001 Hr = -0 99999
−1 H r = −1.001 −1
10 10 Hr = −0 999
H r = −1.1 Hr = −0.9
−2 −2
10 10
−3 −3
10 10
−4 −4
10 0 1 10 0 1
10 10 10 10
k x /k 0 k x /k 0
(a) (b)
Fig. 5.7 Effect of deviations in the real part of permittivity. (a) εr = −1 − δ, µr = −1. Only one resonance can be seen. (b)
εr = −1 + δ, µr = −1. Two resonances are excited. Cut-off is sensitive to variations in ε. The electrostatic limit is working
well
unity, there is only one surface plasmon–polariton mode excited by the

TM polarized light, namely the ω (−) mode. Mathematically, the solu-
tion for the ω (+) branch happens to lie to the left of the ‘left-handed
light line’, kx2 = εr2 µr2 k02 . This means that for this part of the spectrum
evanescent components in vacuum turn into propagating ones inside the
slab without fulfilling the condition for a resonant excitation of surface
modes. Mathematically, although there is a singularity in the denomi-
nator at a value kx close to k0 from the ω (+) branch close to the light
line, this value of kx corresponds to κ2 = 0 (transition frequency from
the propagating to evanescent components inside the slab). Taking the
limit κ2 → 0 in eqn (5.5), it can be shown that the transfer function is
finite and continuous there. Note that this is quite a special situation.
Figure 5.7(b) shows what happens if εr is slightly larger than −1. In this
case the refractive index is larger than −1 and there are two resonances
in the transfer function. Note that the electrostatic limit, that in this
example works well for large values of kx , is not capable of describing
the first resonance close to k0 .
There is a direct connection between the achievable resolution and the
SPP resonances. The flat bit of the transfer function stretches up to the
value of large kx corresponding to the excitation of a SPP mode. It is
the position of this resonance that ultimately determines the resolution
of the slab. It is the excitation of these short waves that enables the
recovery of the subwavelength part of the spectrum of an object. The
cutoff can therefore be, alternatively, attributed to the value of kx for
the second resonance, instead of the condition |T | = 1/2.
Similar results can be obtained for deviations in µr , in both directions
away from the value −1, see Fig. 5.8. In this case the transfer function
can be shown to take the form
1 1
10 10
0 0
10 10

P r = -1.00001 P r = -0.99999
−1 −1 P r = −0.999
10 P r = −1.001 10
P r = −1.1 P r = −0 9
−2 −2
10 P r= 2 10 Pr = 0
Pr = 1
−3 −3
10 10
−4 −4
10 0 1 10 0 1
10 10 10 10
k x /k 0 k x /k 0
(a) (b)
Fig. 5.8 Effect of deviations in the real part of permeability. (a) εr = −1, µr = −1 − δ. Only one resonance can be seen. (b)
εr = −1, µr = −1 + δ. Two resonances are excited. Thickness d/λ = 0.1. Note that the electrostatic limit (dotted line) fails
completely
1
T ≃ 2 4 . (5.27)
δ k0 2kx d
− e +1
4 kx
The cutoff frequency can be obtained as
" 2 #
1 4 kxcutoff
kxcutoff = ln . (5.28)
d |δ| k0
Again, if µr < −1 there will be only one surface mode that can exist,
and if µr > −1, there can be two resonances. Note that the effect of
variations in µ is weaker than the effect of variations in ε, in agreement
with the cutoff expressions (5.26) and (5.28).
An interesting feature of the transfer function is that its magnitude
can be larger than unity, not only at the two resonances but also every-
where in-between, the plateau of the transfer function means the evanes-
cent part of the spectrum gained in amplitude as compared to its values
coming from the object. The question is whether or not there is anything
wrong with energy conservation (see also Section 5.7). The electrostatic
limit fails as expected. It does not know about variations in µ.
Figures 5.9(a) and (b) show that a thicker slab fails to be a good lens;
the transfer function deteriorates too quickly. The cutoff is too short
and no reasonable resolution can be obtained, even with variations as
small as δ = 10−5 ! In addition, in the case of variations in ε only, the
electrostatic limit begins to fail as the cutoff moves to smaller values of
kx .
1 1
10 10
0 0
10 10

H r = -1.00001 P r = -1.00001
−1 H r = −1.001 −1 P r = −1.001
10 10
H r = −1.1 P r = −1.1
−2 −2
10 10
−3 −3
10 10
−4 −4
10 0 1 10 0 1
10 10 10 10
k x /k 0 k x /k 0
(a) (b)
Fig. 5.9 (a) Same as in Fig. 5.7(b) but for thickness d/λ = 2. The only good of the electrostatic limit is that it also predicts
a cutoff, but it fails to give any reasonable quantitative agreement with the full solution. The cutoff is less than 2k0 even for
a deviation from εr = −1 as small as 10−5 . (b) Same as in Fig. 5.8(b) but for thickness d/λ = 2. The electrostatic limit fails.
The cutoff is less than 2k0 even for a deviation as small as 10−5 from µr = −1
Case D. Small deviation in refractive index

It is also possible to look only at the deviation in the refractive index.
This was analyzed by Merlin (2004) who assumed the refractive index
in the form
√
n=− 1−σ (5.29)
and obtained analytical expressions for σ ≪ 1. In agreement with other
studies (Haldane, 2002; Gomez-Santos, 2003) he stressed the importance
of the excitation of surface plasmon–polaritons as kx → ∞. He found
that depending on the sign of σ the near-field distribution is either sym-
metric or antisymmetric.
Case E. The refractive index remains at −1 but both ε and µ

undergo small variations
Smith et al. (2003) and Merlin (2004) approached the perfect lens by
√
perturbing the refractive index n = εr µr and therefore the perfect lens-
ing effect is lost for propagating components. Lu et al. (2005), Cui et al.
(2005) and French et al. (2006) proposed a different route of addressing
deviations from the perfect-lens scenario by introducing deviations in ε
and µ correlated in such a way that their product remains unity. This
can be achieved by taking
1
, εr = −
µr = −(1 ± δ) , (5.30)
1±δ
with δ being real. Any modifications in resolution are therefore not due
to loss effects within the lens.
Figures 5.10(a) and (b) show the transfer function for d/λ = 0.1 and
2, respectively. But does this configuration bring advantages relative
1 1
10 10
0 0
10 10

−5 −5
G=10 G=10
−1 −3 −1 −3
10 G=10 10 G=10
G=0.1 G=0.1
−2 −2
10 G=1 10
−3 −3
10 10
−4 −4
10 0 1 10 0 1
10 10 10 10
k x /k 0 k x /k 0
(a) (b)
Fig. 5.10 Keeping refractive index n = −1. Deviations in ε and µ correlated. Thickness d/λ = 0.1 (a) and 2 (b). Electrostatic
limit works only for a thin slab (a); it is too inaccurate for a thick slab (b)
to the cases when only one material parameter is different from that of
the perfect lens? The transfer function for the propagating components
is, apparently, better than in previous examples. The reason is quite
obvious. The refractive index n being minus unity means merely that
the propagating components exhibit negative refraction at an angle equal
to the angle of incidence. So refraction is at the desired angle but the
reflection coefficient is no longer zero, as for the perfect transfer function,
because the impedance of the negative-index medium,
r r r
µ µ0 µ0
η= = (1 ± δ) 6= = η0 , (5.31)
ε ε0 ε0
differs from that of free space. The good news is that the electrostatic
limit seems to work. At least as long as the cutoff frequency is large
relative to k0 . Remember, the electrostatic limit implies that both κ2
and κ1 are equal and can be approximated by kx . In the case considered
κ1 and κ2 are indeed equal to each other. But the other assumption,
that they can be approximated by kx , is only valid for kx ≫ k0 . So
when the cutoff is too low—and this is, for instance, the case for the
curve with δ = 1, e.g. εr = −2, µr = −0.5—the electrostatic limit fails
yet again. This failure of the electrostatic limit is even more obvious for
a thicker slab (see Fig. 5.10(b)). As far as resolution is concerned, just
as in previous examples, it deteriorates logarithmically with |δ| and is
inversely proportional to the slab thickness.
4
Many authors like to refer to this fact
5.3.5 Near-perfect? Near-sighted!
as an ‘uncertainty principle’ using the
quantum-mechanical language to de-
Suppression of the high-frequency wave components limits the resolu-
scribe the fact that the narrower the tion of the lossy slab acting as an imaging system since the spatial size
object the broader is its Fourier spec- of the image (say, its half-width ∆) and the spectral width of the cor-
trum. responding spectrum, δ, are inversely proportional to each other.4 The
12 12
exact exact
low-loss approx. low-loss approx.
10 10
−5
−5
H’’=10
P’’=10
0
k x /k 0
cut−o k x /k
8 8
−3 −3
P’’=10 H’’=10
6 6
cut−o
P’’=0.1
4 4 H’’=0.1
2 2
0 0.5 1 1.5 2 0 0.5 1 1.5 2

thickness d/O thickness d/O
(a) (b)
12
exact
low-loss approx.
10
cut−o k x /k 0
8 −5
H’’=P’’=10
6 −3
H’’=P’’=10
4 H’’=P’’=0.1
0 0.5 1 1.5 2
thickness d/O
(c)
Fig. 5.11 Near-sighted lens. Cut-off frequency versus slab thickness. Losses in µ only (a), in ε only (b) and in both ε and µ
(c). Solid line: exact calculation, circles: low-loss approximation from Podolskiy and Narimanov (2005)
strong dependence of the cutoff and of the resulting resolution on the

slab thickness means that there is a certain maximum thickness allowed
to achieve a specific resolution.
Estimating roughly the resolution as being inversely proportional to
the cutoff frequency, ∆ = 2π/kxcutoff , it follows after some algebra from
eqn (5.21) that
" #
2
λ
4 −1
∆ ∆
d= s 2 ln
"
2
# . (5.32)
∆ ′′ λ ′′
2π 1 − ε 2 −1 +µ
λ ∆
The dependence of the cutoff frequency on the slab thickness is plot-
ted in Figs. 5.11(a)–(c) for various amounts of loss. By comparing
Fig. 5.11(a) and Fig. 5.11(b) it can be seen that, in agreement with
eqns (5.23) and (5.24), the cutoff frequency for thinner slabs is higher
for lossy µ than for lossy ε. As thickness increases, even small losses
either in ε (Fig. 5.11(a)) or in µ (Fig. 5.11(b)) or in both ε and µ (Fig.
5.11(c)) eventually bring the cutoff down to k0 . Note that the defini-
tion chosen for the cutoff frequency assumes that the transfer function
reaches the magnitude of 1/2. For any loss, there is a critical thickness
for which the definition of a cutoff becomes meaningless as |T | < 1/2
5
Note that the plots have been calcu- for any kx (noticeable in each of Figs. 5.11(a)–(c)5 as an abrupt end for
lated from the exact expressions but curves with the largest value of loss). An important conclusion follows:
the results calculated from the approx-
imation of Podolskiy and Narimanov
while a thin lens of d ≪ λ can provide subwavelength resolution, the
(2005) are practically identical. resolution of a lossy far-field lens with d ≥ λ does not perform better
than usual optical devices (Podolskiy and Narimanov, 2005). Thus, the
expression for the resolution limit of a LHM-based lens proves that the
area of its subwavelength performance is usually limited to the near-field
6
Podolskiy and Narimanov (2005) in- zone.6
troduced as ‘near-sighted’ the flat near- We can put it in a different way. The consequence of the extreme
perfect lens, a catchy name that has
often been used. Interestingly, it was
dependence of the resolution on the deviation from the perfect-lens con-
Narimanov’s group that later and in dition is that in order to achieve a cutoff frequency not lower than 10
parallel to Engheta’s group proposed k0 , a slab of 0.1 λ thickness would require a loss level of about 0.002,
the concept of a ‘far-sighted’ cylindri- but a slab of 0.67 λ would tolerate a loss of no greater than 6 × 10−19
cal lens described below in Section 5.6
(see also Jacob et al. 2006; Salandrino
(see also Smith et al. 2003; French et al. 2006).
and Engheta 2006).
5.3.6 General cutoff frequency relationships

We have seen that any deviation from the ideal conditions causes the
appearance of a cutoff in the spatial frequency spectrum. This was shown
in a number of figures. A general feature found was that the cutoff was
higher when µ deviated from its ideal value than when ε did so. However,
for deviations in the same material parameter it is not obvious which
deviation is more harmful. Is it the real part or the imaginary part?
Or possibly what matters is the deviation from the n = −1 condition?
The answer may be obtained from Fig. 5.12 where the normalized cutoff
frequency is plotted against the logarithm of the deviation δ for the case
when d/λ = 0.1, i.e. when the lens is thin enough for good resolution. It
turns out that the main difference is whether it is ε that changes or not.
The two cases differ quite substantially. But apart from that it hardly
matters what kind of deviation occurs. All deviations seem to have
identical effects. Comparing expressions (5.23) and (5.26) it becomes
obvious that deviations in either the real or the imaginary part of the
permittivity would result in the same resolution. The same would apply
7
For more details see Chapter 3 and to TE polarization and variations in the permeability, respectively.
also discussions on SPPs for the sil- The relationship between the higher SPP resonance and the cutoff in
ver lens later in this chapter. ω (−) is spatial frequency has been noted before. We have done the calculations
shown there to be the higher kx reso-
nance that determines the cutoff only for all the deviations investigated before, in order to find out how close
if εr < −1, µr = 1, whereas it is ω (+) the relationship is. It turns out to be so close that it is not worth plotting
for −1 < εr < 0, µr = 1. the results. It is though worth pointing out that the resonance is not
necessarily provided by the ω (−) branch. It can be the ω (+) branch.7
60
Hr =−1−G, Pr =−1
Hr =−1−jG, Pr =−1
50 Hr =−1−G, Hr Pr =1
Hr =−(1+G)-1, Hr Pr =1
40 Hr =Pr =−1−G
0
Hr =−1, Pr =−1−G
cut−o k /k
x
30 Hr =−1, Pr =−1−jG
20 Fig. 5.12 Relative cutoff frequency

against the logarithm of the deviation δ
from the ideal conditions. εr = −1 − δ,
10 µr = −1 (circles); εr = −1 − j δ, µr =
−1 (crosses); εr = −1 − δ, εr µr = 1
(squares); εr = −1/(1 + δ), εr µr = 1
0 (pluses); εr = µ = −1 − δ (dots);
−10 −8 −6 −4 −2 0 εr = −1, µr = −1 − δ (diamonds);
log(G) εr = −1, µr = −1 − j δ (stars)
5.3.7 Effect of discretization in numerical

simulations
In the wake of the availability of various numerical packages simulations
have become very fashionable. It is much less effort to do the simulation
than to perform the corresponding experiment. A sign of this way of
thinking is the introduction of the term ‘numerical experiment’. The
term is justified in the sense that by solving Maxwell’s equations we
can get the truth, the whole truth and nothing but the truth. That is
the principle anyway. The reality is often different. Just as with real
experiments one needs green fingers to perform numerical experiments.
The expertise needed is usually acquired by devoting a long time to the
resolution of mysteries inherent in all numerical packages. Paul et al.
(2001) found that the resolution of the perfect lens is not better than
the diffraction limit. Ziolkowski and Heyman (2001) maintained that
no superlens effect occurs when the lens medium is dispersive, which of
course it is bound to be. In a later publication Chen et al. (2005) inves-
tigate the ability of the finite-difference time-domain method to model
a perfect lens. They show that because of the frequency-dispersive na-
ture of the medium and the time discretization, an inherent mismatch in
the constitutive parameters exists between the slab and its surrounding
medium. This mismatch in the real part of the permittivity and per-
meability is shown to have the same order of magnitude as the losses
typically used in numerical simulation. When the LHM slab is taken as
lossless this mismatch is shown to be the main factor contributing to the
image resolution loss of the slab.
Koschny et al. (2006) investigate the transfer function of the dis-
cretized perfect lens in finite-difference time-domain (FDTD) and trans-
fer matrix method (TMM) simulations; the latter allow us to eliminate
the problems associated with the explicit time dependence in FDTD sim-
ulations. They find that the finite discretization mesh acts like imaginary
deviations from µr = εr = −1 and leads to a cross-over in the transfer
function from constant values to exponential decay around kxcutoff lim-
iting the attainable super-resolution. A qualitative model is proposed
that is capable of describing the impact of the discretization. kxcutoff is
found to depend logarithmically on the mesh constant, in qualitative
agreement with the transfer matrix method simulations.
5.4 Negative-permittivity lens

5.4.1 Introduction
It should have become clear from previous discussions that it is not nec-
essary at all for a material to have negative refractive index to provide
subwavelength resolution. The most prominent example is a slab of silver
that supports SPP modes with high kx values due to its negative per-
mittivity. In Engheta’s (2002) notation this is called an epsilon-negative
material and abbreviated as ENG. Such a slab can be used to replicate
field distribution from the objects with subwavelength features for TM
polarization. Of course the images produced cannot be expected to be
perfect: propagating modes emanating from a source do not contribute
to the image because inside the slab they turn into evanescent waves
(the k vector is purely imaginary as k 2 = ω 2 εµ < 0) so they can barely
8
Losses will of course contribute to the make their way to the image plane.8
decay but the major factor is reflection. As far as the evanescent part of the spectrum is concerned, perfect
If the slab is not too thick, the wave
can tunnel through the slab, emerging
imaging requires a flat transfer function for all values of kx . For sil-
at the output surface but with much ver, the obstacles to achieving that aim, SPP resonances causing the
reduced amplitude. high spatial frequency cutoff, cannot be disregarded and would never
disappear as would be in the case of the perfect lens εr = µr = −1
(Shamonina et al., 2001). Retardation effects (i.e. not assuming the
electrostatic limit) were also investigated by Shen and Platzman 2002;
Ramakrishna et al. 2003; Jiang and Pike 2005.
5.4.2 Dependence on thickness

The lens, being near-sighted, applies to the ‘epsilon-negative-only’ case
as well. Thinner lenses are better than thicker ones. An example is
shown in Fig. 5.13 where the material is assumed slightly lossy: εr =
−1 − j 0.01. The object positioned at z = 0 is of the shape of a Gaussian.
The y component of the magnetic field varies as

x 2
Hy = exp − . (5.33)
τ
In our example τ is chosen as λ/15 so that the half-width of the Gaussian
is equal to 30 nm. The aim is to look at an object that is small relative
to the wavelength but not too small.
The object, the image and the transfer function are shown in Figs.
5.13(a)–(d) for thicknesses of d = 60, 40, 20 and 10 nm, respectively.
object and image transfer function

10
1
|E |
|T|
x
1 (a)
0 0.1
10
1
|E |
|T|
x
1 (b)
0 0,1
10
1
|E |
|T|
x
1 (c)
0 0.1
10
1
|Ex|
|T|
1 (d)
0 0.1
−50 0 50 1 10 100
x [nm] kx/k0
Fig. 5.13 Imaging for a low-loss (ε′′ = 10−2 ) silver slab of various thicknesses. The object (dashed lines) and image (solid
lines) are shown in the left-hand column. The amplitude of the transfer function is shown in the right-hand column. d = 60,
40, 20, 10 nm (a)–(d)
It may be seen that at 60 nm thickness (twice the half-width of the

Gaussian) the transfer function looks entirely unsuitable for imaging and
the image is so poor that the original Gaussian cannot be recognized.
For d = 40 nm the situation is better. The transfer function has not got
any flat parts but at least the cutoff frequency is further away. Now, with
a little imagination one can see some distorted Gaussian. Proceeding to
a thickness of d = 20 nm the transfer function is now flat for a range of
spatial frequencies and we have a good Gaussian. Reducing the thickness
further to a value of 10 nm the flat part of the transfer function becomes
wider and the image practically coincides with the object.
The conclusion may now be drawn that for good imaging the thick-
ness of the lens should be less than the width of the object. Does this
conclusion apply to the case when losses are much higher? In our next
example we shall take the permittivity as εr = −1 − j 0.4 and look at
the imaging of the same Gaussian variation of the magnetic field for the
same four thicknesses. A brief look at Fig. 5.14 reveals that the situa-
tion has considerably deteriorated. Thicknesses of d = 60 and 40 nm are
obviously not suitable but even for d = 20 nm (which is less than the
width of the object) the resulting image is much wider than the object.
For d = 10 nm there is some flat region in the transfer function and the
image is quite good. The conclusion is clearly that losses are bad. High
object and image transfer function

10
1
|E |
|T|
x
1 (a)
0 0.1
10
1
|Ex|
|T|
1 (b)
0 0,1
10
1
|Ex|
|T|
1 (c)
0 0.1
10
1
|Ex|
|T|
1 (d)
0 0.1
−50 0 50 1 10 100
x [nm] kx/k0
Fig. 5.14 Same as Fig. 5.13 but for ε′′ = 0.4
losses lead to further deterioration in image quality.
5.4.3 Field variation in the lens

As discussed in Chapter 2 the field grows exponentially in the ideal lens.
For the ‘epsilon-negative-only’ case this is not true in general. Let us
investigate it in an example. We know that we have good imaging for
low losses when the lens thickness is d = 20 nm. The transfer function
is that in the third row of Fig. 5.13, which is replotted in Fig. 5.15.
The upper SPP resonance is now at kx = 12k0 . We shall look at three
subwavelength components, kx /k0 = 5, 10 and 15 denoted by the points
A, B and C. The corresponding field variations (maximum normalized
to unity) are shown in Figs. 5.15(b)–(d). At A, in the flat section of
the transfer function, the field variation is the same as in the ideal lens
consisting of decay, growth, decay. At B, close to the SPP resonance,
there is no pure growth inside the lens: the field first slightly decays. At
C, beyond the SPP resonance, the field has two clear maxima inside the
lens, the larger one at the input surface. It is clear now that the image
reproduction does not in general proceed as predicted by the ideal case
but it still applies to lower values of kx .
The explanation may again be provided in terms of SPPs. The transfer
function is flat as long as the rear surface is excited strongly enough.
Beyond the cutoff, kx d becomes too large, so that the two surfaces are
(a) (b)
2 1
A: k =5k
x 0
|Ex(kx)|
1.5
A B 0.5
|T|
1
C
0.5 0
0 1
0 5 10 15 20 B: k =10k
x 0
k /k
|Ex(kx)|
x 0
0.5
1
C: kx=15k0
|Ex(kx)|
0.5
0
0 10 20 30 40
z [nm]
Fig. 5.15 Transfer function (a) and the evolution of individual Fourier components for a number of spatial frequencies indicated
by the points A, B and C (b)
effectively decoupled and the imaging mechanism collapses as discussed

before.
5.4.4 Other configurations: compression

Is the d/2–d–d/2 configuration the only one suitable for imaging with a
negative-index material? Obviously not. The geometrical optics focus is
still there if the lens in the Veselago geometry is shifted sideways keeping
the positions of the object and image fixed as shown in Fig. 5.16. The
distances are now d1 , d, and d − d1 but the total distance from object to
image is still equal to 2d. The paths are the same in air and in the lens,
hence the total phase shift is zero for propagating waves and the total
amplitude is unity for non-propagating waves satisfying the conditions
for the perfect lens. We are not aware of any detailed investigation of
z
what happens in this lens when conditions differ from the ideal ones.
On the whole it seems unlikely that this unsymmetric geometry would 0 2d
d1 d+d1
have any advantages over the symmetric geometry.
If our aim is not to faithfully reconstruct but only to process some
spatial information in the spatial frequency region then the field is open, Fig. 5.16 Veselago’s flat lens: lens
any configuration is possible. We shall show here only one example that shifted relative to the object. Image at
we reported in our early paper (Shamonina et al., 2001) considering the the same position
compression of the signal. It is possible to do that by investigating (in
1
full
2 input
10 ES
full
0.8
ES
air
2
intensity |Ex|
0 0.6
10
0.4
−2
10
0.2
−4
10 0
0 10 20 30 40 50 −80 −60 −40 −20 0 20 40 60 80
kx/k0 x [nm]
(a) (b)
Fig. 5.17 (a) The transfer function and (b) the image for the 5-18-5 nm configuration
our case it was merely by trial and error) how altering the geometry
affects the transfer function. We shall show here the object, the image
and the transfer function in Fig. 5.17 for the 5–18–5 configuration, all
dimensions in nm. Interestingly, the transfer function between the two
SPP resonances increases. If higher values of kx dominate that means
that the image is made up by higher harmonics and, consequently, it
may be narrower than the object. Compression has been achieved.
5.4.5 The electrostatic approximation revisited

Earlier in this chapter we have seen a number of examples where the
electrostatic approximation is compared with the exact result. Some-
times it is good, sometimes reasonably good and sometimes outright
bad. In the present section we shall re-examine the approximation re-
lated to the first publication of the perfect lens (Pendry, 2000). In Fig.
5.18 we reproduce the original figure from Phys. Rev. Lett. The object
consists of a pair of step functions 14 nm wide centred at x = ±33 nm.
The material constants are µr = 1 and εr = −1 − 0.4j. The thickness
of the lens is d = 40 nm and the object–lens and lens–image distances
are 20 nm, i.e. it follows the Veselago geometry. The image (intensity,
not amplitude) is then plotted by Pendry as shown in Fig. 5.18(c). The
two peaks are nicely resolved. The image is also shown in the absence
of the lens that has only a central maximum. This figure is quite well
known. Unfortunately, at some stage a numerical mistake must have
been committed. The correct figure for the same parameters is shown in
Fig. 5.19(a). The image in the absence of the lens looks similar but not
in the presence of the lens. The two peaks are no longer resolved either
in the electrostatic approximation or in the exact calculation. We can,
however, resolve the two peaks in the object if we reduce all the dimen-
sions by a factor of two, i.e. taking a 20-nm thick lens, as may be seen
40nm
80nm
Object Silver Image

plane slab plane
z-axis
(a)
Object Image
intensity - |V | 2 intensity - |V | 2
Image with
silver slab
Image without
silver slab
-100 0 +100 -100 0 +100

x-axis (nanometers) x-axis (nanometers)
(b) (c)
Fig. 5.18 Reproduction of Fig. 2 of Pendry’s paper on the perfect lens (2000). (a) Plan view of the new lens in operation. A
quasi-electrostatic potential in the object plane is imaged by the action of a silver lens. (b) The electrostatic field in the object
plane. (c) The electrostatic field in the image plane with and without the silver slab in place. The reconstruction would be
perfect were it not for finite absorption in the silver. Copyright
in Fig. 5.19(b). The same conclusion can be drawn from the transfer
functions plotted in Figs. 5.20(a) and (b). The one for the 40-nm lens
is much narrower. There is roughly a factor of two between the cutoff
frequencies.
The numerical mistake is unfortunate but that is not the point we wish
to make. It is the electrostatic approximation that we wish to examine
more closely. Pendry based the electrostatic approximation on the fact
that in the examples investigated all dimensions were much smaller than
the electromagnetic wavelength. The question is whether small dimen-
sions in themselves are sufficient for the accuracy of the electrostatic
approximation to the exact result. Pendry certainly came to the con-
clusion that for a thin lens the electrostatic approximation, which in the
lossless case predicts a flat transfer function and perfect reconstruction
of the image, is correct. This is suggested both in the figure caption of
his Fig. 2 (our Fig. 5.18) and in the text where he writes: ‘Evidently, only
the finite imaginary part of the dielectric function prevents ideal recon-
struction’. This statement can be checked. Comparing the electrostatic
approximation with the exact result in Fig. 5.19(a) we can see that the
plots are similar. The same is true if we look at the transfer functions:
0.1 0.18
"20−40−20" "10−20−10"
object object ES
0.08
full
ES
2
2
intensity |Ex|
intensity |E |
x
0.06
full
0.09
0.04
air
air
0.02
0 0
−100 −50 0 50 100 −100 −50 0 50 100
x [nm] x [nm]
(a) (b)
Fig. 5.19 The object (Pendry’s double step function shown as gray bars), the image from the full solution (solid line) and from
the electrostatic approximation (dashed line) and the image in the absence of the slab (dashed-dotted line). Silver slab with
εr = −1 − j 0.4 of the thickness d = 40 nm (a) and d = 20 nm (b)
the electrostatic approximation is quite close to the exact result, see Fig.
5.20(a). Thus, the claim that the electrostatic approximation is applica-
ble might be acceptable. However, if we think of the physical reason for
the growth of the non-propagating waves, namely the excitation of SPP
resonances, then this argument cannot be correct. It is incorrect on two
counts. In the absence of losses the transfer function will tend to infin-
ity at the value of kx that satisfies the resonance condition. Secondly,
the resonance is always followed by a cutoff. The cutoff will limit the
resolution but the resonance will do something worse: it will lead to an
image in which the spatial harmonic causing the resonance dominates.
In the next example we shall show what happens when the object is
the same and the thickness of the lens is the same but we shall drastically
reduce losses by taking ε′′ = 10−4 . The field distribution in the image,
and the transfer function are shown in Fig. 5.21. The electrostatic ap-
proximation yields a very good reproduction of the object, much better
than the one achieved for ε′′ = 0.4. Alas, the results calculated from
the exact solution show a much wider image caused not so much by the
cutoff but by the SPP resonance. There is also considerable discrepancy
now between the cutoff points of the approximate and exact results.
We may thus say categorically that the electrostatic approximation
fails when the losses are zero or very small. For high enough losses the
electrostatic approximation works reasonably well. High losses are good
because they blunt the resonances.
The mathematical reason for the failure of the electrostatic approx-
imation for ‘epsilon-negative-only’ materials (first pointed out by Sha-
monina et al. (2001)) was already given in Section 2.12.4. It comes down
to the case of which one of two small quantities is negligible.
Whether the electrostatic approximation works or not is not an im-
portant matter. One can always rely on the exact expression. The
2 2
10 10
full full
ES ES
1 1
10 10
0 0
|T|
|T|
10 10
−1 −1
10 10
−2 −2
10 10
0 5 10 15 20 0 5 10 15 20
k /k k /k
x 0 x 0
(a) (b)
Fig. 5.20 (a) Transfer function for the same object for a slab of silver, εr = −1 − j 0.4. Thickness of lens (a) d = 40 nm and
(b) d = 20 nm
2
1.6 10
full
1.4 ES
1
1.2 10
full
1
|E |2
0
|T|
0.8 10
x
0.6
−1
0.4 10
0.2
ES −2
0 10
−200 −100 0 100 200 0 5 10 15 20
x [nm] kx/k0
(a) (b)
Fig. 5.21 (a) Image and (b) transfer function for a slab of silver, εr = −1 − j 10−4 . Lens thickness d = 40 nm
effect of the SPP resonance on the image is not necessarily significant

either. See, for example, Fig. 5.13(c) where the sidelobes are kept low
in spite of the SPP resonances. The reason we devoted a whole section
to the accuracy of the electrostatic approximation is that those inaccu-
rate statements were published in a paper that, at the time of writing,
had over 2000 citations. It is an influential paper, the statements made
there have gained wide currency: it needed a correction. We shall also
mention it in Section 5.7, the last section of this chapter concerned with
widely held beliefs that do not stand closer scrutiny.
5.4.6 Experimental results

Theoreticians pride themselves on being able to deliver a new theory
in the course of a single afternoon, provided it is rainy and they have
R
(a) Z (b)
L
Y R
30 X
4
L
2 R
HF oscillator
1 9 L
3
10.5
HF
receiver
90
Y
9 2 4
1
3
Fig. 5.22 Schematic representation of X
the experimental setup. From La- 105
c
garkov and Kissel (2005). Copyright 11
2005 by the American Physical Society
nothing better to do. To prepare a new experiment takes longer. This

is the principal reason why it took nearly four years before experimen-
tal papers on the superlens came out. The first paper concerned with
imaging in a negative-index material was written by Lagarkov and Kissel
(2005). The experiments were performed with microwaves at around a
frequency of 1.7 GHz using the experimental setup shown in Fig. 5.22.
The negative-index material was similar to that of Shelby et al. (2001b)
but instead of SRRs the authors used spirals. The object was made
up by two vertical half-wave dipoles 30 mm apart. The received power
was measured by a similar dipole in the image plane. Measurements
conducted in the absence and presence of the lens showed that the two
transmitting dipoles λ/6 apart could be resolved with the aid of the
lens, whereas no feature was seen in the absence of the lens. Similar ex-
periments at frequencies around 3.5 GHz were reported by Aydin et al.
(2007) using the rod+SRR structure of Shelby et al. (2001a). They were
able to resolve two dipole sources separated by a distance of λ/8.
All other reports we know of were concerned with imaging at much
shorter wavelengths from the infra-red to the ultraviolet region relying
on epsilon-negative-only materials. Most of the experiments were done
in three groups, those of Blaikie, Zhang and Hillebrand and Shvetz. The
first attempt to use silver as the lens material was made by Melville et al.
(2004). The lens chosen was 120-nm thick, unsuitable for subwavelength
imaging but served as a practice run. The authors were able to show that
the image was better in the presence than in the absence of the lens. In
the next paper (Melville and Blaikie, 2005) the lens configuration was
(a) (b) H
UV exposure E
Conformable mask
UV exposure
Tungsten Conformable glass
Conformable
mask mask support

Tungsten Conformable glass
Silver lens
mask Dielectric spacer (PMMA)
mask
mask support
Planar lens (silver)
Vacuum contact Dielectric spacer (PMMA or SiO2)
Resist
Resist stack
Vacuum contact
BARC
Resist
Resist stack
Silicon substrate BARC
z
Silicon substrate
Y
x
Fig. 5.23 Schematic representation of two experimental setups comparing hard-contact and flat-lens lithography. From Melville
c 2006 Elsevier Ltd
et al. (2006). Copyright
PMMA9 –Ag–SiO2 with thicknesses 25–50–10 nm. The wavelength of 9 poly-methyl metachrylate
operation was 365 nm. The patterns that were imaged included isolated
lines, line pairs, gratings and arrays of dots with some feature sizes
smaller than 100 nm. Gratings with periods down to 145 nm were
resolved.
In a later paper by Melville et al. (2006) the relative merits of hard-
contact lithography (often referred to as evanescent near-field optical
lithography) and flat-lens lithography were explored. The aim was to
project an image into the photoresist. Schematic drawings of the two
somewhat different experimental setups are shown in Figs. 5.23(a) and
(b). They found that above the diffraction limit both hard-contact litho-
graphy and the silver lens perform in a similar manner. Below the dif-
fraction limit the hard/contact lithography produces clearer images in
the resist. The proposed explanation is that it is due to losses in the sil-
ver and to (insufficient) quality of the silver deposition. Both single- and
double-layer lenses were examined by Melville and Blaikie (2006). They
found that for the same total thickness of silver, the resolution limit is
qualitatively better for the double-layer stack. However, pattern fidelity
is reduced in the double-layer experiments because of increased surface
roughness. The same group was also concerned with modelling the sil-
ver superlens (Blaikie and McNab, 2002; Melville and Blaikie, 2007) in
order to assist the experimental work.
In another set of experiments with a silver lens by Fang et al. (2005)
the object was an array of 60-nm wide slots of 120-nm pitch next to
the word NANO inscribed onto a chromium10 screen (see Fig. 5.24). 10
Note that the plasma frequency of
The illumination coming from below is at a wavelength of 365 nm. The chromium is sufficiently far away from
that of silver so as not to interfere with
silver lens is separated from the object by a layer of PMMA and the the superlensing effect.
image is projected into a layer of photoresist. Detection is achieved by
PR
Ag
PMMA
Quartz Cr
Fig. 5.24 Schematic representation

of an experiment imaging the word
‘NANO’. From Fang et al. (2005).
c 2005 AAAS
Copyright 365 nm Illumination
converting the image into topographic modulations and is then mapped

using atomic force microscopy. The same group (Lee et al., 2005) re-
ported further improvement in their experimental procedure achieving
the image of a 50-nm half-pitch object at λ/7 resolution.
Webb-Wood et al. (2006) performed frequency-dependent near-field
scanning optical microscopy measurements of plasmon-mediated near-
field focusing using a 50-nm thick Au film. In these studies the tip aper-
ture of a probe acts as a localized light source, while the near-field image
formed by the metal lens is detected in situ using nanoscale scatterers
placed in the image plane. By scanning the relative position of object
and probe, the near-field image generated by the lens is resolved. Scans
performed at different illumination frequencies reveal an optimum near-
field image quality at frequencies close to the SPP resonant frequency.
A different material, SiC, in a different wavelength spectrum, mid-
infra-red, was utilized by Korobkin et al. 2006a; Korobkin et al. 2006b
for producing subwavelength imaging based on earlier work of their
group (Shvets, 2003; Shvets and Urzhumov, 2004). SiC is a polar mate-
rial. The analogy between plasmon–polaritons and phonon–polaritons
was briefly discussed in Section 3.2 and the dependence of its dielectric
constant on frequency was given in eqn (3.8). It may be seen there
that its dielectric constant is negative in the Reststrahlen band between
the frequencies of the transverse and longitudinal optical phonons. If
the superlens is made up by a SiO2 –SiC–SiO2 sandwich then the di-
electric constant of the SiC slab should match that of SiO2 , which is
εd = 4. Very conveniently the dielectric constant of SiC is equal to −4
at λ = 11µm and that wavelength, very conveniently again, can be pro-
duced by a tunable CO2 laser. The lens itself is of conventional design
with thicknesses of 200–400–200 nm but the detection mechanism is not
SiO2 SiC SiO2

200 nm 400 nm 200 nm
Object slits Diagnostic slits

500 nm Au or Ag 500 nm, Au or Ag
2.5 µm period 2.5 µm period
Fig. 5.25 Schematic representation of an imaging experiment with slits for detection. From Korobkin et al. (2006a). Copyright
c 2006 Optical Society of America

conventional (see Fig. 5.25).

There are now additional metallic sheets on both sides of the lens. A
set of periodic slits in the front metal sheet constitutes the object and
another set of slits of the same period serve for detecting the image.
Two kinds of detecting slits are used. The first one (seen in Fig. 5.25) is
spatially in-phase with the object slits and a second one (not shown) is
produced so that the detecting slits are half-way in between the object
slits. The measurement is done with an electromagnetic wave incident
perpendicularly upon the lens. Away from the wavelength at which
the superlens conditions are satisfied, it makes no difference where the
detecting slits are. The output is the same whichever configuration is
used. On the other hand, under superlens conditions each of the object
slits are imaged exactly at the place where the image slits are, yielding a
high output. The output for both slit configurations was measured as a
function of wavelength. It was found that under superlens conditions the
output was higher by several orders of magnitude than at wavelengths
further away, an excellent proof for the existence of the superlensing
effect in SiC.
Further experimental evidence was provided by Taubner et al. (2006).
Their experimental setup is shown in Fig. 5.26. There, the SiO2 –SiC–
SiO2 lens has slightly different thicknesses, 220–440–220 nm, the objects
are holes of about 1 µm diameter in a 60-nm thick gold film, and the
detection is by a probe. The authors were able to resolve these holes at
a wavelength of 10.85 µm.
An entirely new idea for a superlens was reported by Liu et al. (2007a).
Instead of using a flat silver lens the outer boundary was formed into
a subwavelength grating. It served to convert the incident evanescent
waves into propagating waves that could then be magnified by a con-
ventional microscope. The authors were able to resolve two nanowires
separated by 70 nm at a wavelength of 377 nm.
Fig. 5.26 Schematic representation of

imaging holes by a SiC experiment.
From Taubner et al. (2006). Copyright
c 2006 AAAS
5.5 Multilayer superlens

We have already mentioned the multilayer lens in Section 2.12.3 and
illustrated it in Fig. 2.37. When only εr is equal to minus unity then
the image is no longer perfect but a multilayer configuration can still
significantly improve the imaging abilities of the flat lens. Although
the idea was proposed as early as 2001 (Shamonina et al., 2001), the
next lot of papers came only in 2003 (Pendry and Ramakrishna, 2003;
Ramakrishna et al., 2003) and 2004 (Gao and Tang, 2004; Tang and
Ex (z = d) Gao, 2004). However, by now (writing in early 2008) there is a consensus
image that the most exciting application, to have pre-magnification by a flat
lens followed by a conventional optical microscope, would be impossible
H<0 without a multilayer configuration. The multilayer lens is now part
H>0 of conventional wisdom. We shall return to the magnifying variety in
Section 5.6. Here we shall discuss the original idea.
The rationale for the operation of the multilayer lens can be deduced,
z
as in many previous cases, from the properties of SPPs. We know that
the excitation of coupled SPPs at both interfaces of the silver slab is
a necessary condition for the near-field superlens. The excitation of
coupled SPPs ceases, however, if the slab is not thin enough. A silver slab
of 60 nm would, e.g., be too thick to act as a superlens at the wavelength
Ex (z = 0) of 360 nm. It is, of course, desirable for the image to be transferred over
object a larger distance. The solution is offered in the form of a multilayer
metamaterial made from a series of slabs with negative permittivity
x separated by layers with positive permittivity (see Fig. 5.27).
In our first example we shall look at the advantage of having more than
Fig. 5.27 Multilayer superlens one layer in the superlens while the total width of the active material
remains constant. The object in all four cases shown in Fig. 5.28 is a
Gaussian of 14 nm half-width, and the wavelength of operation is 365
nm. For the imaginary part of the dielectric constant we shall choose
ε′′ = 0.1. This value comes as a compromise. If losses are too low, as
explained before, the image quality may be poor on account of the SPP
resonance. If losses are too high they will cause a cutoff at a low spatial
frequency that would again adversely influence the image quality. The
80 80 80
60 60 60
z [nm]
40 40 40
20 20 20
0 0 0
−40 0 40 −40 0 40 −40 0 40
1 1 1
0.8 0.8 0.8

|Ex | 2
0.6 0.6 0.6
0.4 0.4 0.4
0.2 02 0.2
0 0 0
−40 −20 0 20 40 −40 −20 0 20 40 −40 −20 0 20 40
(a) (b) (c)
Fig. 5.28 Superlens of the overall thickness of 80 nm. Top: Poynting vector streamlines. Bottom: object (dashed line) and
image (solid line). (a) Multilayer lens of four 10-nm silver layers separated by 10-nm layers of air. (b) Two 20-nm silver layers
separated by 20-nm air layer. (c) Single lens of 40-nm thick silver slab. εr = −1 − j 0.1
upper figures show the streamlines of the Poynting vector. As shown

already in Fig. 2.38(c), the focus moves to the outer boundary of the
slab as soon as we take into account evanescent waves. This is confirmed
by Fig. 5.28(c) where a single silver layer of 40 nm is considered. For
two silver layers (Fig. 5.28(b)) the Poynting vector diverges-converges- 11
The concept of Poynting vector op-
diverges-converges-diverges, and the same phenomenon occurs for the tics was introduced in a somewhat dif-
four-layer case shown in Fig. 5.28(a). As stressed at several places in ferent context by Russell (1984).
this book we should not underestimate the significance of the Poynting
vector streamlines. Veselago’s construction of the flat lens relied on
12
the concept of rays, clearly unsuited for near-field phenomena. When This happens because at the fre-
quency close to the plasmon resonance,
evanescent waves are included then the ray picture is bound to fail. dispersion of SPPs approaches high
Its role is taken over by the streamlines of the Poynting vector.11 The kx values with the group velocity ap-
streamlines converging and diverging offer an excellent physical picture proaching zero. The net power flow
of how the image is carried from layer to layer.12 along the boundary is therefore close to
zero. This comes about as the power
The lower figures in Fig. 5.28 show the fidelity of reproduction in the flow outside the slab diverging away
three cases. In the configuration of four 10-nm thick silver layers (Fig. from the small object is compensated
5.28(a)) the imaging is good, although there is a slight deterioration in by the power flow inside the slab.
1
10 0.5
1 layer
|Ex|2
1 layer
0
2 layers 0.2
2 layers
|E |2
0
|T|
x
10
4 layers
0
0.06 4 layers
|E |2
x
−1
10 0
0 1 −100 −50 0 50 100
10 10 x [nm]
kx/k0
(a) (b)
Fig. 5.29 Superlens with one, two and four 10-nm thick layers of silver. Transfer function and image distribution
quality. For two silver layers of 20 nm each (Fig. 5.28(b)) the image
becomes wider, and even wider when there is only one single silver layer
of 40 nm thickness (Fig. 5.28(c)). The above example offers a clear proof
for the efficacy of the multilayer lens both by the Poynting vector picture
and by the improved fidelity of reproduction.
There can be no doubt, however, that the multilayer lens is not a
magic cure for all the ills of the silver lens. The image quality is bound
to deteriorate with the number of layers. We shall show here a further
example that considers not only the fidelity of reproduction but resolu-
tion as well, and look at the image after 1, 2 and 4 layers. The object
consists of the two step functions 50 nm apart shown already in Fig.
5.18(b). For the same amount of loss as in the previous example we plot
the transfer function and the image intensity in Figs. 5.29 and 5.30 for
d = 10 nm and 20 nm, respectively. For d = 10 nm there is a slight
deterioration of the image as the number of layers increases from 1 to 4.
For d = 20 nm the two peaks can be clearly resolved for 1 layer, there
is some deterioration for 2 layers and the object is unrecognizable when
there are 4 layers. The conclusions are the same as before. For good
imaging, thin layers are needed.
We have now covered the major principles upon which multilayer
lenses work. For further discussions see Feng et al. 2005; Feng and
Elson 2006; Dorofeenko et al. 2006; Webb and Yang 2006; Melville and
Blaikie 2007.
Another strand of thought that could lead to multilayer imaging comes
from traditional treatments of finding the effective dielectric constant
(for a review see, e.g., Bergman 1978). In its electrical engineering con-
text see the detailed analysis by Wait (1962). In optics they appear
in the classical work of Born and Wolf (1975). They are called strati-
fied media described by transfer matrices and used mainly as filters and
matching elements. For small modulation of the dielectric constant they
1
10
0.12 1 layer
|E |2
x
0
1 layer
0.03 2 layers
2 layers
2
|Ex|
0
|T|
10
4 layers
0
0.01 4 layers
|Ex|2
−1
10 0
0 1 −100 −50 0 50 100
10 10 x [nm]
kx/k0
(a) (b)
Fig. 5.30 Superlens with one, two and four 20-nm thick layers of silver. Transfer function and image distribution
provide holographic filters (see, e.g., Solymar and Cook 1981; Solymar
et al. 1996), and they also appear in photonic bandgap materials in
their 1D version13 (Scalora et al., 1998). 13
Scalora et al. (1998) were the first
In the simple case of a multilayer made up by isotropic materials of to realize that the transparency of a
stack of metal–dielectric layers could
equal thickness the derivation of the effective dielectric constants is given be much higher than that of a single
in Appendix I. It results in an anisotropic material with longitudinal and one. They also introduced the term
transverse dielectric constants given as ‘resonance-enhanced tunnelling’.
1 −1 1
ε−1 ε1 + ε−1

z = 2 and εt = (ε1 + ε2 ) . (5.34)
2 2
The idea is that for sufficiently thin materials simple averaging works.
For εr1 = 1 and εr2 = −(1 + δ) the effective dielectric constants,
2
εz ≃ and εt ≃ −δ , (5.35)
δ
turn out to be of opposite sign. Taking εr1 = 1 and εr2 = −1 we find
the interesting outcome that
εz = ∞ and εt = 0 . (5.36)
The consequence is that an image pasted on the front surface will appear
at the back surface. It is an interesting concept worth mentioning but it
is a poor approximation. A layer thickness of 20 nm (for which imaging
is shown in Fig. 5.30) is already too thick for the approximation to work.
For further discussions see Wood et al. (2006).
14
The difference between the two methods may be best seen by looking Although we have a healthy scep-
ticism concerning the results of many
at the Poynting vector. The above effective-medium approximation gives simulations we are happy with this one
a straight line, whereas the one based on the transfer matrices leads to because it comes from a very reliable
the diverging-converging picture of Fig. 5.28. The result of a numerical source (our group) and provides further
simulation14 by CST MICROWAVE STUDIO (Tatartschuk, 2008) is proof for the correctness of the physical
picture presented in Fig. 5.28.
shown in Fig. 5.31.
−2.6
100
−2.8
80
−3
60
−3.2
Y [nm]
−3.4
40
−3.6
20
−3.8
0
−4
−20 −4.2
−60 −40 −20 0 20 40 60
X [nm]
Fig. 5.31 Numerical simulation of the streamlines and amplitude of the electric field for a multilayer silver lens. From
Tatartschuk (2008). For coloured version see plate section
5.6 Magnifying multilayer superlens

In Fig. 5.31 we have seen that in an epsilon-negative multilayer structure
the spatial information is preserved by the Poynting vector converging
in the negative-epsilon material and diverging in the dielectric. With
proper design it can be ensured that the reproduction of the subwave-
length image is still good. What could we expect if instead of the mul-
tiple planar slabs we have multiple cylindrical annuli? Common sense
suggests that the image will simply expand according to the rules of geo-
metrical optics. Numerical simulations using CST MICROWAVE STU-
DIO confirm this expectation, as may be seen in Fig. 5.32 (Tatartschuk,
2008). Taking two objects some distance apart very close to the surface
of the first annulus the variation of the Poynting vector streamlines may
be seen to follow a regular pattern. The image is magnified. Again,
common sense suggests that the magnification is equal to the ratio of
the radii of the outer ring to the inner ring.
The idea that a cylindrical lens may give a magnified image was first
proposed by Pendry and Ramakrishna (2002). They used conformal
transformation from rectangular to cylindrical co-ordinates to find the
equivalent of the perfect flat lens in cylindrical form. A somewhat dif-
ferent derivation of the magnification effect is due to Pendry (2003b). In
400
−4.5
200
−5
0
Y [nm]
−5.5
−200
−6
−400
−6.5
−600
−7
−600 −400 −200 0 200 400 600
X [nm]
Fig. 5.32 Magnifying superlens. Numerical simulation of Poynting vector streamlines emanating from two subwavelength
objects. εr = −1 − j 0.4. From Tatartschuk (2008). For coloured version see plate section
that solution, however, the permittivity and permeability had to change

as a function of space in order to provide a perfect image.
The possibility of using a cylindrical multilayer structure for magni-
fying the image was proposed by Salandrino and Engheta (2006). A
theory of the magnification process for a cylindrical multilayer lens in
terms of higher-order scattering of cylindrical waves was formulated by
Jacob et al. (2006). They also found the 2D dispersion equation for
the cylindrical geometry. The isofrequency curves, plotted in the kr
(wave vector in the radial direction), kθ (wave vector in the azimuthal
direction) plane, turn out to be hyperbolas. We may remember from
Section 5.5 that a planar multilayer material in the effective-medium
approximation has diagonal elements of the opposite sign in the permit-
tivity tensor, and that leads to a hyperbolic dispersion equation. It is
not unreasonable to expect that the same kind of relationship is valid
in a multilayer cylindrical structure as well. See the analysis by Jacob
et al. (2006) for the cylindrical case, and also Sections 8.1.2 and 9.4 for
similar dispersion curves.
A practical realization of the magnifying lens with subwavelength res-
olution has been achieved by Liu et al. (2007b). The lens consisted
(a) (b)
Fig. 5.33 (a) Schematic of magnifying optical hyperlens and numerical simulation of imaging of subdiffraction-limited objects.
(b) An arbitrary object ‘ON’ imaged with subdiffraction resolution. Line width of the object is about 40 nm. The hyperlens is
made of 16 layers of Ag-Al2 O3 . From Liu et al. (2007b). Courtesy of Prof. Xiang Zhang of University of California at Berkeley.
Copyright c 2007 AAAS. For coloured version see plate section
of alternate cylindrical layers of Ag and Al2 O3 deposited on a half-

cylindrical cavity (see Fig. 5.33). There were 16 layers of both materials
with thicknesses of 35 nm each. Finally, a 50-nm thick chromium layer
is deposited upon the last layer of the lens. The object is the two letters
ON inscribed in the chromium layer. The smallest feature, the linewidth,
is 40 nm and the lines are 150 nm apart. In the magnified image that
spacing becomes 350 nm. The wave illuminating the object has a wave-
length of 356 nm. It is the wavelength where the dielectric constant of
15
See discussion in Section 3.3. silver15 is equal to −1. The spacing of 350 nm is close to the wavelength
of the incident wave, hence the output image can be magnified by a
conventional microscope. This was indeed the aim of the authors. The
hope is that lenses of this type will be available in the future with higher
initial magnification making it possible to view subwavelength objects
by a conventional microscope. The main limitation is that the object
has to be very close to the first layer of the lens. A generalization may
also be possible. One may argue that if such a magnifying system can be
realized by a cylindrical lens then a spherical lens should also be within
the realm of possibilities, although in that case the polarization of the
Regular microscope objective
0.247
Magnifying
superlens
‘‘Super’’ ~70 nm
Plasmon illumination
Z (a.u.)
images
Plasmon
Phase-matching structure Sample rays –0.041
Glass substrate Gold film

Laser illumination
0.057 0.151
X (a.u.)
(a) (b)
Fig. 5.34 (a) Schematic of the magnifying superlens integrated into a conventional microscope. The plasmons generated by
the phase-matching structure illuminate the sample positioned near the centre of the superlens. The lateral distance between
the images produced by the alternating layers of materials with positive and negative refractive index grows with distance along
the radius. The magnified images are viewed by a regular microscope. (b) The cross-section of the optical image indicates
resolution of at least 70 nm or ∼ λ/7. Z, optical signal; X, distance. From Smolyaninov et al. (2007). Copyright c 2007
AAAS
wave scattered by the object might adversely influence the fidelity of

reproduction.
Another realization of image magnification by a multilayer cylindrical
lens is due to Smolyaninov et al. (2007). Their experimental setup is
shown in Fig. 5.34(a). The lens consists of cylindrical annuli of PMMA
deposited on a gold film. The wave propagating in the lens is a SPP
excited by a laser at a wavelength of 495 nm. In what sense is this a
lens? The principle is still the same as in the previous example in which
layers of positive and negative ε alternated. Considering the properties
of SPPs it is possible to assign to them an index of refraction on the
basis that by how much is the velocity of the SPP reduced. We have
two regions: (i) the gold film on its own, and (ii) the gold film covered
by PMMA. For the gold film on its own the SPP resonance is above the 16
Remember the SPP resonant fre-
frequency of excitation. The SPP is a forward wave with a positive index √
quency is ωs = ωp / 1 + εd , where εd
of refraction, n1 . For the PMMA-covered gold film the SPP resonance16 is the relative dielectric constant of the
is below the frequency of excitation hence the wave propagating in this adjoining dielectric.
region is a backward wave to which it is possible to assign a negative
index of refraction, n2 . Now the condition for imaging action is that
n1 d1 = −n2 d2 , (5.37)
where d1 and d2 are the widths of regions 1 and 2. The parameters were
so chosen by the authors that the above relationship is satisfied at the
wavelength of excitation.
The objects were 2 or 3 rows of PMMA dots in the radial direction
positioned near the centre of the lens. For the three-row case the distance
Regular microscope
objective
Dielectric droplet
acting as a mirror
Surface plasmon-
Sample produced image
Focal point
Gold film
Glass prism
Fig. 5.35 Magnifying superlens. From
Smolyaninov et al. (2005). Copyright
c 2005 by the American Physical So- Laser illumination
ciety
between them at the input of the lens was 70 nm. Upon illumination
they gave rise to three divergent SPP ‘rays’. The field distribution at
the output of the lens was then large enough (10 times as large as at
the object) to be further magnified by a conventional microscope. The
microscope output showing three distinct peaks is plotted in Fig. 5.34(b).
An entirely different approach to image magnification was proposed
by Smolyaninov et al. (2005) based on earlier work (Smolyaninov, 2003;
Smolyaninov and Davis, 2004). The essential ingredient is a parabolic-
shaped glycerine droplet on a metal film as shown in Fig. 5.35. The
waves involved are SPPs excited by laser light. In the experiments, light
of 502 nm wavelength was used at which the SPP wavelength is about
70 nm giving an effective index of refraction of about 7. Imaging is due
to reflection of the SPPs by the parabolic boundary. An object placed
in the vicinity of the focal point will have an image as shown in Fig.
5.35. The authors successfully imaged arrays of nanometric pinholes in
the gold film. The field distribution is then viewed from above by a
conventional microscope.
5.7 Misconceptions
Whatever our chosen field is we like to understand it in simple qualitative
terms. What can we say about the superlens? It gives a perfect image
but a perfect image cannot be obtained in practice due to losses, im-
perfections, time delay and the granular nature of the metamaterial. It
is easy to understand the consequence of granular structure. Surely, we
cannot obtain a resolution that is better than the period in the meta-
material. The influence of other effects is less obvious. Some of the
relationships are quite complicated and superficial references to them
may not be adequate. The misconceptions, quoted below, may only be
entertained by a small minority but we think it is worth including them
in this section.
Let us first look at the superlens with εr = µr = −1.
Believed by many: The physical picture of imaging shown, e.g., in Fig.

2.38(a) with the aid of geometrical optics is still approximately true in
the presence of evanescent waves.
Not true, not even approximately. There is no internal focus when
the object is subwavelength. The Poynting vector streamlines converge
upon the outer boundary of the lens.
Believed by many: The superlens not only images but focuses as well.
A focus means that the intensity is high in a small region around it,
and in whichever direction we move away from the focus the intensity
declines. This does not occur in the superlens. If, for example, we
measure the intensity along the axis beyond the outer boundary of the
lens it turns out to be a monotonically declining function. It has no
maximum. It has no focal point.
Believed by many: Evanescent waves do not carry power.
Indeed, they do not carry power in the lossless case but they do in the
presence of losses because there is then a phase difference between the
coefficients of the various evanescent waves.
Believed by many: The presence of evanescent waves does not influence
the flow of power.
Not true. The streamlines of the Poynting vector are quite different
when the evanescent waves are taken into account. There are then vor-
tices that may be closed inside the negative-index material but they may
also extend into the neighbouring positive-index medium.
Beliefs that would not stand closer scrutiny extend also to the silver
lens, i.e. when only the relative dielectric constant is equal to minus
unity and the relative permeability remains at plus unity.
Believed by many: Resolution of the silver superlens is merely re-
stricted by losses. The image is perfect without loss.
Not true. There is a cutoff spatial frequency even in the absence of
losses.
Believed by many: The smaller the loss in a silver lens the more faith-
fully is the image reproduced.
Not true. If losses are absent or too small the transfer function exhibits
resonances due to the excitation of the SPPs. If the amplitude of the
resonance peak is large, then the image is not good at all: it shows only
the oscillatory field distribution of the corresponding SPP mode.
Believed by many: The electrostatic approximation is good if losses
are small.
Not true. The SPP resonances need to be damped for the electrostatic
approximation to apply. Hence, losses have to be sufficiently large.
Phenomena in waveguides
6
6.2 Propagation
The term waveguide is often used in a general sense meaning a struc- in cutoff waveguides 199
ture that guides a wave from point A to point B. Some waveguides are 6.3 Filters in coplanar and
open; the fields may extend to a considerable distance away or they may microstrip waveguides 204
be closed, confining the fields to the interior of the waveguide. Both 6.4 Tunnelling 206
varieties will be discussed in the present chapter. For experiments on 6.5 Phase shifters 208
metamaterials it is an advantage to confine the waves. This was the 6.6 Waveguide couplers 209
reason, for example, why in the first experiments showing negative re- 6.7 Imaging in two
fraction (Shelby et al., 2001a), the waves propagated between metal dimensions:
plates (see Fig. 2.34). transmission-line
approach 210
The waveguides discussed fall into two categories: they have either
some special features or have device applications. In Section 6.2 the
special feature is that under certain well-defined circumstances waves
may propagate in a hollow waveguide (see Section 1.5) below their cutoff
frequency. In Section 6.3 we shall present filters using metamaterial
elements in coplanar and strip waveguides. Section 6.4 is concerned
with phase shifters, Section 6.5 with couplers. In Section 6.6 we look
in some more detail at the 2D experiments of Grbic and Eleftheriades
(2004) where microstrip waveguides loaded by lumped elements are used
for demonstrating subwavelength imaging.
6.2 Propagation in cutoff waveguides

We have learned in Section 1.5 that electromagnetic waves can propagate
in a hollow metallic waveguide, provided the dimensions of the waveguide
are large enough. The question may then be asked: how will the waves be
affected if we insert in the waveguide an array of small resonators? In the
metamaterial context the first experiments were conducted by Marques
et al. (2002a). The experimental setup is shown in Figs. 6.1(a) and (b).
The waveguide has a square cross-section of a × a where a = 6 mm. The
corresponding cutoff frequency (see eqn (1.38)) is 25 GHz. The SRRs
inserted have diameters of 5.6 mm and a resonant frequency of about
6 GHz. With a distance of a between them they nearly touch each other.
The input and output are via standard coaxial to rectangular waveguide
junctions.
The experimental results for a waveguide of 36 mm length housing
nine elements are shown in Fig. 6.2 where transmission between input
and output is plotted against frequency. It may be clearly seen that
200 Phenomena in waveguides
(a) (b)
Fig. 6.1 (a) The SRR-loaded square waveguide. (b) Sketch of the experimental setup. From Marques et al. (2002a). Copyright

Fig. 6.2 Measured transmission coef-

ficient for l = 36 mm. From Marques
et al. (2002a). Copyright c 2002 by
the American Physical Society
there is a transmission peak in the frequency region between about 5.8

and 6.5 GHz. This is a remarkable result. In this region the waveguide
is well below cutoff and yet there is transmission through it owing to the
effect of the SRRs. The pass band is more than 60 dB above the noise
level that characterizes the cutoff waveguide.
The experiments were repeated by Hrabar et al. (2005), also using
square waveguides and SRRs with a resonant frequency of 7.8 GHz.
Their experimental results are shown in Figs. 6.3(a) and (b), where the
scattering coefficient S21 (as discussed in Section 1.24, |S21 |2 gives the
output power relative to the input power) is plotted against frequency.
The results for waveguide A (dimensions, a = 35 mm and b = 15 mm,
cutoff frequency 4.3 GHz) are shown in Fig. 6.3(a). In that case, the
cutoff frequency is below the element’s resonant frequency and the pres-
ence of the SRRs appears in a stop band. Note that the SRRs also
influence the cutoff frequency, which moved towards higher frequencies.
Waveguide B has a cutoff frequency of about 12.5 GHz, well above the
resonant frequency of the element. Now, the presence of the SRRs ap-
pears again in an upward shift of the cutoff frequency, and, more sig-
nificantly, in a pass band, as shown in Fig. 6.3(b); the same result as
obtained by Marques et al. (2002a).
The theoretical description of these phenomena is quite straightfor-
ward, provided we realize that the effective permeability may be a tensor
(see Section 1.15). For a SRR this is discussed in Section 2.8 where the
0 0
10 10
Backward wave
20 pass band
20
30 30
40 Waveguide Cut Off 40
S21(dB)
Waveguide cut off
S21(dB)
50 50
60 Stop band with negative transversal 60
70 permeability 70
80 80
90 90
100 100
6 7 8 9 10 11 12 13 14 15 6 7 8 9 10 11 12 13 14 15
(a) (b)
Fig. 6.3 Measured transmission coefficient. There is (a) a stop band when the cutoff frequency of the waveguide is below
the elements’ resonant frequency; (b) a pass band when the cutoff frequency of the waveguide is above the elements’ resonant
frequency. From Hrabar et al. (2005). Copyright c 2005 IEEE
effective permeability is determined in the direction perpendicular to the

plane1 of the SRR. 1
Unfortunately, it happens sometimes
The corresponding theory (Kondrat’ev and Smirnov, 2003; Hrabar that customary notations clash and
therefore at least one of them has to be
et al., 2005) follows from that introduced in Section 1.5 considering fur- modified. The direction of anisotropy
ther that the relative magnetic permeability is a tensor (see eqn (1.97)), for an SRR is usually taken in the z di-
which may now be written as rection, which is the notation used in
Section 2.8. We have also chosen the
z axis as the direction of wave propa-
 
1 0 0
gation. In order to resolve the conflict
µ = µ0  0 µyy 0  , (6.1) we take in this chapter the y direction
0 0 1 as perpendicular to the plane of SRRs.
Hence, the anisotropic term in eqn (6.1)
where µyy is the relative permeability in the y direction. Equations (1.32)– is referred to as µyy .
(1.34) may now be rewritten as follows
∂Hz ∂Hy
− = j ωε0 Ex , (6.2)
∂y ∂z
∂Ex
= −j ωµ0 µyy Hy , (6.3)
∂z
∂Ex
= j ωµ0 Hz . (6.4)
∂y
The above set of differential equations are still linear with constant
coefficients, hence they can be solved without difficulty. Using again the
assumption of a wave solution and satisfying the boundary conditions
on the waveguide wall we end up with the relationship
π 2
kz2 = µyy k02 − . (6.5)
a
This equation differs from that of eqn (1.37) by having µyy on the
right-hand side. But that makes all the difference. We can now explain
the experimental results shown in Figs. 6.2 and 6.3. Above the cutoff
frequency k 2 − (π/a)2 > 0, hence µyy being negative will result in kz2
being negative and, consequently, in a stop band. Below the cutoff
2
frequency k 2 − (π/a) < 0, hence µyy being negative results in kz2 being
positive, and, consequently, in a pass band.
The original explanation provided by Marques et al. (2002a) is some-
what different. They quoted Rotman (1962) who simulated plasmas
not only by metallic rods (discussed in Section 2.4) but also by cutoff
waveguides. The claim was that a cutoff waveguide behaves analogously
to a plasma below its plasma frequency. Hence, Marques et al. argued, if
a set of elements, capable of producing negative permeability, is inserted
into a cutoff waveguide that should lead to propagation, the situation
being equivalent to a medium having both negative ε and negative µ.
This argument can be made quantitative by assuming, with Rotman,
that a cutoff waveguide could be regarded as having an effective relative
dielectric constant of
2 ω 2
λ p eff
εr eff = 1 − = 1− , (6.6)
2a ω
where
cπ
ωp eff = (6.7)
a
is the effective plasma frequency.
In general, the propagation coefficient of an electromagnetic wave in
a homogeneous medium can be written as
ω2
kz2 = µr εr , (6.8)
c2
where the wave happens to propagate in the z direction.
Using the definition of the effective relative dielectric constant in the
form of eqn (6.6) we find that
" 2 #
λ π 2
2 2 2
kz = k0 µr 1 − = µr k0 − , (6.9)
2a a
which is of the same form as eqn (6.5). Thus, we may conclude that the
heuristic derivation in terms of the effective dielectric constant leads to
the same result as the one based on the field equations. The advantage
of the heuristic derivation is that it links the phenomena observed in
waveguides to those in left-handed materials allowing the possibility that
both ε and µ may be negative.
Note, however, that eqn (6.9) does not represent the whole truth. It
has µr instead of µyy . Looking at eqn (6.9) one might come to the con-
clusion that propagation is possible in a cutoff waveguide filled with an
isotropic negative-permeability material. And that would be the wrong
conclusion. Equation (6.5) suggests that in order to have propagation
(a) (b)
(c) (d)
Fig. 6.4 Transformation of the waveguide problem using the image principle. From Belov and Simovski (2005b). Copyright

we need specifically a negative transverse permeability. For that reason

it is better to rely on the derivation offered at the beginning of this
section.
It has been argued by Belov and Simovski (2005b) that neither of the
previously offered explanations represent the whole truth. Their model
is quite different. First, they resort to a simplification. Instead of talk-
ing about actual resonant elements, e.g. SRRs or some other varieties,
they assume that the waveguide is filled with resonant arrays of either
magnetic dipoles or short electric dipoles. The latter are made resonant
by loading them with inductances (their properties were investigated in
detail by Tretyakov et al. (2003)). Both types of dipoles can have po-
larizations either in the axial or in the transverse directions. The model
is based on mirror images in the waveguide walls as may be seen in
Figs. 6.4(a)–(d). Thus, instead of a waveguide problem they solve the
problem of propagation in an infinite three-dimensional lattice. They
derive the relevant dispersion relationships, i.e. the variation of wave
number against frequency. It is essentially the same kind of approach
as that of Shamonina et al. (2002b) for MI waves in 3D that was based
on nearest-neighbour interaction. However, the treatment of Belov and
Simovski is much more rigorous. They include interaction between all
elements, and also include radiation effects, i.e. take into account the
dipole fields varying with the inverse of the distance (see Section 1.12).
Their conclusion is that all four configurations may lead to pass bands
in cutoff waveguides.
It should be mentioned that cutoff waveguides that can propagate
electromagnetic waves have potential for applications because their di-
mensions may be much smaller than waveguides working above cutoff.
However, devices in cutoff waveguides are in no sense new. In the 1960s
and 1970s it was a hot subject. The research resulted in a number of
microwave devices: not only broadband filters for which it is most suit-
able but also circulators, frequency mixers, multipliers, phase shifters,
etc. The main aim was to reduce weight and, often, trade weight for
loss. For a review see Craven (1972).
Finally, we need to face the undisputable truth that the explanation
of propagation in a cutoff waveguide in terms of negative transverse
permeability is not new either. The same explanation was suggested by
Thompson (1955) whose paper in Nature covered both theory and exper-
iment. The experiments were conducted in a circular metallic waveguide
loaded by a longitudinally biased ferrite rod. The phenomenon was fur-
ther investigated by Thompson (1963).
Hollow circular waveguides with perfectly conducting walls and loaded
along their axes by a periodic array of thin dielectric (ε > 0) and metallic
(ε < 0) layers, were investigated by Govyadinov and Podolskiy (2006)
in the infra-red region. Note that the anisotropy is this time in the
axial direction. The authors showed that it was possible to obtain both
positive and negative effective indices of refraction. When n < 0 then
propagation is only possible when the waveguide radius is sufficiently
small. Hence, if the radius of the waveguide (with the load inside) is
gradually reduced waves can still propagate and the result is high power
compression.
6.3 Filters in coplanar and microstrip

waveguides
We have discussed a number of very interesting phenomena in hollow
metallic waveguides in the previous section. When it comes to cut-
w s w
off waveguides the size is certainly smaller but unlikely to be small
enough for practical applications under present conditions when one of
the requirements is compatibility with planar technology. A waveguide
particularly suitable for this purpose is the coplanar waveguide shown
schematically in Fig. 6.5. It consists of a centre conductor (signal strip)
Fig. 6.5 Coplanar waveguide geometry
and ground planes on both sides. Loading this waveguide by series ca-
pacitors and shunt inductors Grbic and Eleftheriades (2002) realized
a backward-wave line that, they showed, could radiate in the reverse
direction.
A series of studies concerned with microwave filters were carried out by
a team from three Spanish Universities, those of Navarra, Barcelona and
Sevilla (Martin et al., 2003a; Martin et al., 2003b; Falcone et al., 2004a;
Falcone et al., 2004b; Falcone et al., 2004c; Falcone et al., 2004d; Mar-
tel et al., 2004; Garcia-Garcia et al., 2004; Baena et al., 2005a). They
6.3 Filters in coplanar and microstrip waveguides 205
G
w Fig. 6.6 Layout of the SRR loaded
coplanar waveguide. From Martin et al.
(2003a). Copyright c 2003 American
Institute of Physics
(a) (b) (c)
Fig. 6.7 Photograph of a four-element filter based on complementary SRRs. (b) Simulated and (c) measured S11 (thin line)
c 2004 IEEE
and S21 (thick line) coefficients. From Falcone et al. (2004a). Copyright
loaded transmission lines (mainly in the form of coplanar waveguides)

by SRRs and rods and relied on the previously established experimen-
tal results that SRRs on their own provided a stop band and the stop
band turned into a pass band when short metallic rods were added to
the structure. In one of the realizations (Martin et al., 2003a) SRRs are
placed symmetrically upon the reverse side of the substrate of a coplanar
waveguide, as shown in Fig. 6.6. Note further that at the centre of the
SRRs a thin metal wire connects the signal strip to ground. In this con-
figuration the magnetic field is perpendicular to the plane of the SRRs
and the electric field is between the signal and ground conductors. In
the experiments both transmission and reflection were measured. Simi-
larly to the experiments of Shelby et al. (2001a) they found that in the
range around 8 GHz, which corresponds to the resonant frequency of
the SRRs, there is a pass band when the wires are present, and a stop
band when the wires are removed.
In the study by Falcone et al. (2004a) complementary SRRs (see Sec-
tion 4.3) were used in producing a four-element filter in a microstrip
waveguide (Fig. 6.7(a)). Their simulation and experimental results for
reflection (S11 ) and transmission (S21 ) are remarkably close to each
other, as may be seen in Figs. 6.7(b) and (c). In another similar de-
velopment Falcone et al. (2004b) used spiral resonators (see Section 4.3)
instead of SRRs. Their advantage is that for the same resonant fre-
quency they have a smaller size. We should also note that in order to
achieve a wider bandwidth Martin et al. (2003a) used SRRs tuned to
slightly different frequencies.
Fig. 6.8 A metamaterial-filled square

waveguide sandwiched between two z
empty waveguides operating below cut-
off and two input/output waveguides a
above cutoff. The length of each 1
waveguide section is li . One side of the ε ε0 Metamaterial ε0 ε T
R
waveguide is a = 24 mm. From Baena
et al. (2005c). Copyright c 2005 by y
1 2 x 3 4 5
the American Physical Society
6.4 Tunnelling
Tunnelling is a well-known phenomenon in a cutoff waveguide. It means
that in spite of the wave being evanescent a certain proportion of the
input power can get through to the output. As we know, there can be a
growing wave in a negative-index material. Would the same thing apply
in a cutoff waveguide? The analysis was done by Baena et al. (2005c).
A schematic representation of the waveguide is shown in Fig. 6.8. As
may be seen, there are 5 sections. The waveguide dimensions are such
that in the empty sections, 2 and 4, the wave cannot propagate. In
sections 1 and 5, both taken as infinitely long, the waveguide is above
cutoff by choosing ε to be high enough. In section 3 there is a negative-
index material. The length of each section is li . The electromagnetic
wave is incident from section 1 in the TE10 mode. Since the size of the
waveguide is the same everywhere the TE10 mode remains unchanged
in all the other sections.
It is shown that in the lossless case for isotropic materials the condi-
tions for perfect tunnelling (100% transmission) are as follows:
l2 α2 + l3 α3 + l4 α4 = 0 (6.10)
and
Z2 = Z3 = Z4 , (6.11)
where Zi is the impedance of the ith section and αi is the attenuation

coefficient. This is of course attenuation under cutoff conditions. It does
not imply any ohmic loss. In fact, in order to satisfy eqn (6.10), α3 must
be negative, which means that the wave is growing in section 3, and
l3 = l 1 + l 2 . (6.12)
To satisfy eqn (6.11) we have to choose the materials constants in section

3 as
ε3 = −ε0 and µ3 = −µ0 . (6.13)
It may now be noticed that the conditions for perfect tunnelling are the
same as for perfect imaging.
6.4 Tunnelling 207
2.5 5
1 2 3 4 5 1 2 3 4
2.0 4
Field amplitude
Field amplitude
1.5 3
1.0 2
0.5 1
0.0 0
–10 –5 0 5 10 15 –10 –5 0 5 10 15
y (cm) y (cm)
(a) (b)
Fig. 6.9 Field distribution inside the structure of Fig. 6.8 when section 3 is filled by an isotropic metamaterial with εr = µr = −1.
(a) Perfect tunnelling when l3 = 2l2 = 2l4 . (b) Field distribution when l4 → ∞. Dashed lines show the field amplitude when
section 3 is empty. From Baena et al. (2005c). Copyright c 2005 by the American Physical Society
The distribution of the electric field in the waveguide under these con-
ditions was calculated by Baena et al. (2005c). The geometrical dimen-
sions are a = 24 mm, l3 = 30 mm and the frequency is 5 GHz. We have
the conventional tunnelling situation when ε2 = ε3 = ε4 . Then, there
is a standing-wave pattern in section 1, and the wave declines beyond
that, as shown in Fig. 6.9(a) (dashed line). Under optimum conditions
(continuous line) there is now no standing-wave pattern and the input
amplitude may be seen to be the same as the output amplitude, the
sign of perfect tunnelling. Any deviation from the optimum conditions
leads to rapid degradation of the transmitted power. In Fig. 6.9(b) the
field distribution is shown for two different conditions: when section 3 is
empty (dashed line) and when section 4 extends to infinity (continuous
line). As may be expected, there is an exponential decay in both cases,
but in the latter one the growing wave in section 3 is still there.
A somewhat analogous situation was investigated by Alu and Engheta
(2003) in free space for a periodic array of ε-negative materials paired
with µ-negative materials. Among others they found conditions for per-
fect tunnelling.
Perfect tunnelling in free space through a negative-positive-negative
permittivity medium was investigated by Zhou et al. (2005b) both exper-
imentally and by simulation. They found that perfect tunnelling occurs
at two frequencies at which there is an increase in the magnetic field.
They showed also that the perfect tunnelling effect is insensitive to the
input angle of the incident wave.
A related problem, perfect tunnelling under conditions of frustrated
total internal reflection for pairs of metamaterials was investigated by
Zhou and Hu (2007).
6.5 Phase shifters

A transmission line loaded by a shunt inductance and a series capacitor
was discussed in Section 2.13 and is shown schematically in Fig. 2.39(a).
The circuits were considered for applications as phase shifters by Antoni-
ades and Eleftheriades (2003) in the symmetric form shown in Fig. 6.10.
The corresponding dispersion curve was already plotted in Fig. 2.39(b),
exhibiting a stop band between ω1 and ω2 . In order to address phase
2C 2C
0 0 shifts around zero the stop band needs to be eliminated. This occurs
Z Z when the shunt circuit and the parallel circuit have the same resonant
frequencies. The phase shift per unit cell is then given by
L0
p 1p
ka = ω Lt aCs − Lsh Ct a , (6.14)
ω
a
i.e. a backward phase shift is subtracted from a forward phase shift.
Such phase shifters consisting of two and four stages were realized by
Fig. 6.10 1D phase shifter unit cell Antoniades and Eleftheriades (2003) using coplanar waveguides. The
advantage relative to transmission-line phase shifters is that phase shifts
can be not only positive but also negative or zero. This is a major
advantage if small phase shifts around zero are required. Moreover, the
phase incurred is independent of the length of the structure. For further
work on phase shifters see Islam and Eleftheriades 2004; Antoniades and
Eleftheriades 2005; Abdalla et al. 2005; Eleftheriades and Islam 2007;
Eleftheriades and Balmain 2005; Caloz and Itoh 2006.
There is further discussion of the properties of phase shifters using
backward waves by Nefedov and Tretyakov (2005). They point out that a
metamaterial phase shifter of −10◦ is bound to have a broader band than
a transmission-line phase shifter of 350◦ on account of its shorter length.
However, they prove that any combination of forward- and backward-
wave sections that produce a positive phase shift is less broadband than
a transmission-line phase shifter producing the same phase shift. They
further show that dispersion can be reduced by lines exhibiting positive
anomalous dispersion (see Section 1.18).
This is probably the best place to introduce the proposal by Engheta
(2002) for a thin subwavelength cavity resonator. It consists of a cavity
into which two materials of thicknesses d1 and d2 are placed side-by-
side. One of the materials has a conventional positive index and the
other one a negative index. It is shown there that the negative-index
material in the cavity may serve as a phase compensator, leading to a
resonant frequency that depends not on the sum of the thicknesses but
on their ratio. Thus, in principle, the cavity can be arbitrarily small. It
is further shown that under certain approximations the ratio of the two
thicknesses is independent of the dielectric constants and depends only
on the permeabilities as

d1 µ2
= . (6.15)
d2 µ1
6.6 Waveguide couplers 209
m strip line Interdigital capacitor m strip line

(input) (output)
Shorted (via) stub inductor Fig. 6.11 Layout of the unit cell of
the microstrip left-handed transmission
line. From Caloz et al. (2004). Copy-
Unit cell c 2004 IEEE
right
Fig. 6.12 A 0-dB (9-cell) edge-coupled

directional coupler comprising of two
interdigital/stub composite right-left-
handed transmission line. From Caloz
c 2004 IEEE
6.6 Waveguide couplers

A left-handed transmission line in terms of a series capacitor and a
shunt inductance, incorporated in a microstrip line, was realized in the
form shown in Fig. 6.11 (Liu et al., 2002; Caloz et al., 2004; Caloz 2C
0
and Itoh, 2006). The series capacitor is an interdigital one in order to 2C
increase the capacitance. The shunt inductance is implemented by a stub 0 2C
0
connected to the ground plane. However, the authors argue that neither
the capacitance nor the inductance can be regarded as pure. Both will
2C y
be accompanied by parasitic reactances that will constitute elements of 0 L0
x
a right-handed transmission line. The parasitic inductance LR is caused
by the current flowing along the digits of the capacitor and the parasitic
capacitance CR is due to the electric field between the metal structure a z
forming the elements and the ground plane. This is the justification for
the use of the equivalent circuit shown already in Section 2.13 in which
there is a series resonant circuit consisting of the series capacitance CL Fig. 6.13 2D realization of the 1D cir-
and the parasitic inductance LR , and a shunt resonant circuit consisting cuit of Fig. 6.10
of an inductance LL in parallel with the parasitic capacitance CR . When
two such lines are laid sufficiently close to each other (see Fig. 6.12) they
are coupled via a mutual inductance and a coupling edge capacitance.
The result is a waveguide coupler that connects the various ports to each
Fig. 6.14 The left-handed planar

transmission-line lens. The unit cell of
the left-handed (loaded) grid is shown
in the top inset, that of the right-
handed (unloaded) grid is shown in the
bottom inset. From Grbic and Eleft-
heriades (2004). Copyright c 2004 by
the American Physical Society. For
coloured version see plate section
other. For a detailed description of this coupler, for its design, theory
of operations and measurement results see Caloz et al. 2004; Caloz and
Itoh 2006.
6.7 Imaging in two dimensions:

transmission-line approach
One-dimensional representations of negative-index materials in terms of
loaded transmission lines were discussed in Sections 2.6 and 2.13. They
were suitable for finding dispersion characteristics and demonstrating
the backward-wave properties of metamaterials. However, if the aim is
3 to demonstrate negative refraction or subwavelength focusing, it is nec-
essary to resort to no less than two dimensions. The generalization from
Normalized field level
2.5
2
one to two dimensions is quite straightforward. The two transmission
lines are then rectangular to each other and loading is introduced in
1.5 both directions as shown in Fig. 6.13, which is a direct generalization
1 of the 1D circuit of Fig. 6.10 to two dimensions. A number of aspects
0.5 of these 2D loaded transmission lines were investigated by Eleftheriades
and coworkers in a series of papers (Eleftheriades et al., 2002; Grbic and
0
0 5 10 15 20 25 30 Eleftheriades, 2003b; Grbic and Eleftheriades, 2003c; Grbic and Eleft-
Cell number (column) heriades, 2003d; Grbic and Eleftheriades, 2004; Iyer et al., 2003).
We shall describe here in some detail one of their imaging experiments
Fig. 6.15 The measured vertical elec- (Grbic and Eleftheriades, 2004) in which microstrip lines (dielectric con-
tric field above row 0 at 1.057 GHz. stant, 3.0, substrate thickness 1.52 mm) are loaded by 2-pF capacitors
The vertical dashed lines identify the in parallel, and by 18-nH inductors in series. The size of the unit cell
source (column 0) and image (column
is equal to 8.4 × 8.4 mm. There are 19 × 5 unit cells for the loaded
10) planes, while the vertical solid
lines identify the interfaces of the left- transmission line (see Fig. 6.14 for the experimental setup), and 19 × 12
handed planar slab. The growth of unit cells on each side for the bare (unloaded) transmission lines that
the evanescent waves in the left-handed are analogous to right-handed media. The effective propagation coeffi-
lens is clear. From Grbic and Eleftheri-
c 2004 by the
cients of the loaded and unloaded media were designed to be equal in
ades (2004). Copyright
American Physical Society magnitude but opposite in sign at 1.00 GHz. In the experiment the
first unloaded medium is excited by a vertical monopole fed by a coax-
6.7 Imaging in two dimensions: transmission-line approach 211
Fig. 6.16 The measured vertical electric field detected 0.8 mm above the surface of the entire structure at 1.057 GHz. The
plot has been normalized with respect to the source amplitude (linear scale). From Grbic and Eleftheriades (2004). Copyright
c 2004 by the American Physical Society. For coloured version see plate section

ial cable through the ground plane. The monopole attaches the inner
conductor of the coaxial cable to the unloaded medium, while the outer
conductor of the coaxial cable is attached to the ground plane. The
excitation is in the central row 2.5 unit cells away from the interface.
The image is then supposed to be 2.5 unit cells away on the other side
of the loaded medium. The vertical electric field is detected 0.8 mm
above the surface using a short vertical probe connected to a network
analyzer. Its value above row 0 is plotted in Fig. 6.15 at the optimum
frequency of 1.057 GHz. The source plane at column 0 and the image
plane at column 10 are denoted by dotted lines, whereas the boundaries
of the loaded line are denoted by thick continuous lines. The value of
the field may be seen to be the same at the source and at the image. A
two-dimensional plot of the electric field is shown in Fig. 6.16. As may
be expected, the field is highest at the output interface. The lateral size
of the image was found to be equal to 0.21 effective wavelengths, in con-
trast to the 0.36 wavelengths that is the diffraction-limited value. This is
the first time that the classical limit was beaten by a flat negative-index
lens. It was made possible by the small size of the loading inductors and
capacitors, by the fact that the non-resonant structure was less lossy
than its resonant counterparts and finally because it was relatively easy
to produce an isotropic backward-wave medium.
Magnetoinductive waves I
7
Magnetoinductive (MI) waves have come about as a by-product of the
7.3 Matching the
research on metamaterials. The magnetic elements used in the first transmission line 217
realization of negative refraction by Shelby et al. (2001a) were split-ring 7.4 Excitation 217
resonators that could be modelled (Marques et al., 2002c) as LC circuits, 7.5 Eigenvectors and
provided the dimensions are small relative to the free-space wavelength. eigenvalues 218
We shall also assume, as we have done so far, that the separation of 7.6 Current distributions 220
the elements is also much smaller than the wavelength. This is often 7.7 Poynting vector 224
referred to as the quasi-static approximation. 7.8 Power in a MI wave 225
The simplest realization of an LC circuit as a metamaterial element 7.9 Boundary reflection
is a capacitively loaded loop, shown schematically in Fig. 7.1(a). Two and transmission 225
such loops close to each other are coupled to each other due to the mag- 7.10 Tailoring the dispersion
netic field of one loop threading the other loop and inducing a current characteristics:
in it (Fig. 7.1(b)). The presence of such coupling leads to waves that biperiodic lines 227
were called MI waves by Shamonina et al. (2002a). They belong to the 7.11 Experimental results 231
category of slow waves that propagate at a velocity less than that of 7.12 Higher-order
interactions 236
light.
7.13 Coupled one-dimensional
The properties of magnetoinductive waves have received considerable lines 238
attention since they were first proposed (Shamonina et al., 2002a). A
7.14 Rotational resonance 242
detailed study of the dispersion characteristics was carried out in Sha-
monina et al. 2002b, some other aspects of the theory were treated in
Syms et al. 2005a; Syms et al. 2005b; Sydoruk et al. 2005, experimental
results and comparison between experiments and theory were given in
Wiltshire et al. 2003b; Wiltshire et al. 2004b; Sydoruk et al. 2006;
Syms et al. 2006b; Sydoruk et al. 2007b; Syms et al. 2007a, devices
in MI waveguides were reported in Shamonina and Solymar 2004; Syms
et al. 2005c; Syms et al. 2006a and review papers were published in
Shamonina and Solymar 2006; Shamonina 2008. (a) (b)
Further properties of MI waves concerned with 2D effects and retarda-
tion will be presented in Chapter 8. Note that we have already discussed Fig. 7.1 (a) Capacitively loaded loop.
some simple aspects of MI waves earlier in this book in Section 1.23, (b) Two magnetically coupled elements
among wave solutions on four-poles (Brillouin, 1953): an example was
given concerned with magnetic coupling within one resonant four-pole.
The result was a dispersion equation of the form
ω0
ω= , (7.1)
2M
1+ cos(ka)
L
where ω0 is the resonant frequency, L is the inductance and M is the
214 Magnetoinductive waves I
M M
Vn-2 Vn-1 Vn Vn+1 Vn+2

Fig. 7.2 Equivalent circuit of an array
n-2 n-1 n n+1 n+2
of capacitively loaded loops
mutual inductance. The same dispersion equation was also derived in

Section 2.2 by applying Kirchhoff’s voltage law to magnetically coupled
resonant loops assuming that there is coupling only between nearest
neighbours. This constitutive (recurrent) relation was given in the form
Z0 In + j ωM (In−1 + In+1 ) = 0 , (7.2)

where Z0 is the self-impedance of the loop and In is the current in the
nth loop as shown in Fig. 7.2.
Our aim in the present chapter is further to investigate the properties
of a one-dimensional array of capacitively loaded loops. In Sections 7.2–
7.10 we shall assume nearest-neighbour coupling and give some illus-
trations how such an array may work as a transmission line. We shall
discuss the dispersion relations in Section 7.2, matching by a terminal
impedance in Section 7.3, the problem of excitation in Section 7.4, eigen-
values and eigenvectors in Section 7.5, the distribution of current along
the line in Section 7.6, the Poynting vector in Section 7.7, the definition
of power in Section 7.8, reflection and transmission at a boundary in
Section 7.9, and the tailoring of the dispersion curve in Section 7.10. Ex-
perimental results and comparisons between theory and experiments are
presented in Section 7.11. In Section 7.12 we shall abandon the nearest-
neighbour approximation and assume that, in general, each element of
the line is coupled to all the other elements, and investigate its effect
on the dispersion curve and on the current distribution. Pseudo-one-
dimensional cases (when the theoretical treatment requires only some
minor modification of one-dimensional theory) will be presented in Sec-
tions 7.13 and 7.14, and applications in Section 7.15.
7.2 Dispersion relations

As mentioned before the dispersion relations were derived in Chapters 1
and 2 in two different manners for the simple case when losses are absent.
We shall now add losses, i.e. assume the self-impedance in the form
1
Z0 = j ωL + +R, (7.3)
j ωC
where R is the resistance. The wave assumption is still taken in the form
(see eqn (2.3))
In = I0 e −j kna , (7.4)
N
N-1
z
3
2 3 N-1 N
1 1 2
y r0 Zt
x x
r0
a a
V1 V1 Zt
(a) (b)
Fig. 7.3 (a) Axial and (b) planar array
but now k is complex
k = β − jα, (7.5)
where β is the propagation constant and α is the attenuation coeffi-
cient. The dispersion equation (7.1) may then be separated into real
and imaginary parts yielding
ω02
1− + κ cos(βa) cosh(αa) = 0, (7.6)
ω2
1
− κ sin(βa) sinh(αa) = 0, (7.7)
Q
where κ is the coupling coefficient equal to 2M/L and Q is the quality 1
The simplicity of derivation and the
factor of the resonant circuit equal to ωL/R. Note that M may be ease with which such arrays can be
constructed (see Section 7.4) makes it
positive or negative, as discussed in Section 1.23. It is positive when the
likely that MI waves will be included
array is axial (see Fig. 7.3(a)) and negative in the planar configuration1 in the near future in the undergradu-
(Fig. 7.3(b)). ate syllabus of both physics and elec-
If losses are small enough (Q is high enough) then trical engineering courses. It would also
help to demystify the concept of back-
ward waves. As things stand the only
cosh(αa) = 1 , sinh(αa) = αa , (7.8) place where backward waves make an
appearance in the physics syllabus is in
which means that the dispersion equation for the phase change per ele- the derivation and interpretation of the
ment remains the same, and the losses per element are given as optical branch of acoustic waves. But
even then the emphasis is on the di-
1 atomic lattice and on the interaction
αa = . (7.9) with incident electromagnetic waves.
κQ sin(βa)
The backward-wave character, if at all,
It may indeed be expected that losses decline as the coupling coeffi- is only casually mentioned. In circuit
cient and the Q of the circuit increase. Dependence on βa may also be theory backward waves may appear (as
anticipated, that attenuation should be minimum at the resonant fre- in Section 1.23) but they are usually as-
sociated with lumped inductors and ca-
quency and should increase towards the band edge where βa = 0 or π. pacitors and not as waves in which the
The approximation, however, breaks down right at the band edge where magnetic and electric fields vary along
the attenuation should not be infinitely large. We can get the correct the direction of propagation. A look at
result by solving numerically eqns (7.6) and (7.7). the MI transmission line and the corre-
sponding dispersion curves would con-
We have already plotted in Fig. 1.18 the dispersion equations for κ = vince students that there is nothing ex-
±0.1 for the lossless case. We shall replot them in Figs. 7.4 and 7.5 for otic about backward waves.
Q = 40, 100 and 1000 including both the phase change and attenuation.
1.1 1.1
1.05 1.05
Q=40
0
0
Q=100
Z/Z
1 Z/Z 1
Q=1000
0.95 0.95
0.9 0.9
0 0.5 1 0 0.1 0.2 0.3 0.4
Ea/S Da/S
Fig. 7.4 Dispersion for an axial array. κ = 0.1
1.1 1.1
1.05 1.05
Q=40
0
0
Q=100
Z/Z
Z/Z
1 1
Q=1000
0.95 0.95
0.9 0.9
0 0.5 1 0 0.1 0.2 0.3 0.4
Ea/S Da/S
Fig. 7.5 Dispersion for a planar array. κ = −0.1
It may be seen that (i) the attenuation sharply increases as the quality
factor declines, and (ii) both phase change and attenuation vary rapidly
near the band edge.
It needs to be noted that the coupling coefficient is not necessarily
small. In principle it may be as high as 2. In practice, the highest value
measured so far (Syms et al., 2006b) is 1.5 in the axial configuration in
which the elements can be quite closely packed and −0.7 (Syms, 2006)
in the planar configuration. For κ = 1.5 and Q = 100 the dispersion
curves look quite different, as may be seen in Fig. 7.6. There is still
a lower cutoff frequency but the bandwidth now extends to arbitrarily
high frequencies with an asymptote at

1
βa = arccos , (7.10)
κ
which in the present case comes to βa = 0.73 π.

7.3 Matching the transmission line 217
1.5
ZZ0
0.5
0
0 0.5 1 Fig. 7.6 Dispersion for an axial array
Ea/S in the case of a strong coupling. κ = 1.5
7.3 Matching the transmission line

It is quite clear from what we have done so far that MI waves can prop-
agate along an array whether it is axial or planar. We may therefore
regard such arrays as transmission lines having somewhat unusual dis-
persion relations. We know that for all transmission lines a terminal
impedance exists that can absorb all the incident power. When this
happens we talk about matching. In other words, in a matched trans-
mission line there is a wave travelling from source to load but there are
no reflected waves. How can we find this terminal impedance for a MI
transmission line? Let’s go back to eqn (7.2), which gives the relation-
ship between currents In−1 , In and In+1 . This is true everywhere in the
line except at the first and the last element. If the last element is the
N th then it can see a neighbour at the site N − 1 but the element at
N + 1 is missing. We may then pose the question: can we substitute an
impedance for the missing (N + 1)th element? Let’s call this impedance
ZT that is inserted in the N th element. Hence, Kirchhoff’s law for the
N th element takes the form (Shamonina et al., 2002b)
(Z0 + ZT )IN + j ωM IN −1 = 0 . (7.11)

We want a travelling wave so that IN and IN −1 are related to each
other by the factor exp(−j ka), whence ZT may be obtained from
eqn (7.11), with the aid of the dispersion equation, as
ZT = j ωM e −j ka . (7.12)
It turns out that the terminal impedance is not a real constant, as for a
coaxial line for example, but it is complex and frequency-dependent.
7.4 Excitation
The dispersion equation is derived on the assumption that there is no
external excitation. As it happens, it is quite easy to include possible
excitation by ideal voltage sources in any of the elements. The excita-

tion may come from an incident plane wave or from separate voltage
sources. The general relationship between the applied voltages and the
resulting currents is still described by Kirchhoff’s law but now in the
form (Shamonina et al., 2002b)
V = ZI . (7.13)
The above equation may also be regarded as a generalized Ohm’s law.
V and I are N -dimensional vectors
V = (V1 , V2 , . . . Vn , . . . VN ) , I = (I1 , I2 , . . . In , . . . IN ) , (7.14)
and Z is the N × N impedance matrix
 
Z0 j ωM 0 0 ... 0 0

 j ωM Z0 j ωM 0 ... 0 0 

Z=
 ... ... ... ... ... ... ... .
 (7.15)
 0 0 0 . . . j ωM Z0 j ωM 
0 0 0 ... ... j ωM Z0
This is clearly a tri-diagonal matrix. The main diagonal elements are all
Z0 and the elements next to them (left, right, up or down) are all j ωM .
If the line is terminated by an impedance ZT then the last element in
the matrix (N th row, N th column) should be replaced by Z0 + ZT .
If the voltage excitation vector is known the current may be obtained
by inverting the Z matrix,
I = Z −1 V . (7.16)
Very often, only the first element is excited. In which case only the first
component of the voltage vector is finite and all the others are zero.
7.5 Eigenvectors and eigenvalues

To determine the natural modes of the system we need to work out
the eigenvectors and eigenvalues, which we shall do here for the lossless
case. The eigenvectors for a line consisting of N elements may be easily
guessed from the physics. We may postulate that there are elements at
sites zero and N + 1 but the current is zero there. We may also assume
that the currents vary sinusoidally between the two ends. Hence, the
elements of the lth eigenvector may be written in the form (Shamonina
et al., 2002b)

nlπ
In(l) = I(N ) sin (n = 0, 1, 2, . . . N, N + 1). (7.17)
N +1
The value of I(N ) can be found from the orthonormality condition
7.5 Eigenvectors and eigenvalues 219

(l) (m) 0 if l 6= m
I ·I = , (7.18)
1 if l=m
yielding
2
I(N ) = I0 . (7.19)
N +1
The corresponding eigenvalues may be obtained from their definition
ZI(l) = λl I(l) , (7.20)
as

lπ
λl = Z0 + 2j ωM cos . (7.21)
N +1
Having got the eigenvectors and eigenvalues we can find, in closed
form, the response to a general excitation. A voltage vector V of dimen-
sion N can be expanded in terms of the eigenvectors as
N
X
V= µl I(l) . (7.22)
l=1
The unknown coefficients µl can be determined from the orthonor-

mality condition as
µl = V · I(l) . (7.23)
Considering further that a matrix can be expanded in terms of its

eigenvectors and eigenvalues we obtain for the current vector
N
X µl
I= I(l) . (7.24)
λl
l=1
It may be seen from the above equation that in the general case all
the eigenvectors are excited. In order to excite a single mode the cor-
responding eigenvalue must be close to zero.2 From eqn (7.21) the lth 2
When the eigenvalue is exactly zero
eigenvalue is zero when the current amplitude will be infinitely
large. This is a consequence of neglect-
ing losses.
lπ
Z0 + 2j ωM cos = 0. (7.25)
N +1
But this is nothing else than our dispersion equation. A single mode,
i.e. a single value of ka, can be excited for a discrete set of frequencies.
For N = 5 the normalized set of eigenvectors are plotted in Fig. 7.7(a).
For the axial configuration and for κ = 0.1 the corresponding values of
ka are shown on the dispersion curve of Fig. 7.7(b).
1
0
−1
1
0
−1
n
1
current I
0
1.1
−1
1
1.05
0
0
1
/
−1
1
0.95
0
−1 0.9
1 2 3 4 5 0 0.5 1
Fig. 7.7 (a) Eigenvectors and (b) the element number n a/
corresponding values of ka for a 5-
element array (a) (b)
7.6 Current distributions

We have in the last two sections derived the current distribution in
response to a voltage excitation in two different ways: by inverting
the impedance matrix (eqn (7.16)) and by relying on eigenvectors and
eigenvalues (eqn (7.24)). In the present section we shall use the former
method that we have found numerically more convenient, particularly
for the lossy case. As examples, we shall show current distributions
on arrays of resonant loops for three typical cases: (i) travelling waves,
(ii) resonances occurring for standing waves and (iii) evanescent current
distributions with implications for near-field pixel-to-pixel imaging (Sec-
tion 7.15.3). In each of the following examples we will assume that only
one or two of the loops are excited by an external voltage source and
that the currents in all other loops are induced via the magnetoinductive
interactions.
(i) Travelling waves
If the line with a finite number of resonant loops is terminated by its
matching terminal impedance (eqn (7.12)) then a single travelling MI
wave propagates along the line. As an example, we choose an axial array
with the parameters N = 31, κ = 0.1, Q = 100, with the first element
being driven by a voltage V1 and with the last element being terminated
with the matching terminal impedance. The current distributions in the
complex plane for three different frequencies, ω/ω0 = 0.9757, 1, 1.0262,
are shown in Fig. 7.8. The corresponding values of the propagation
constant, βa = π/3, π/2, 2π/3 and of the attenuation constant, αa =
0.12, 0.1, 0.12, are provided by the dispersion equations (7.6) and (7.7)
(see also Fig. 7.4). It may be seen from Fig. 7.8 that the phase angle of
the current varies from element to element by the corresponding value
of βa and that the amplitude of the current changes exponentially as
1 1 1
01 0.1 0.1
0.01 0.01 0.01

Im(In )
Im(In )
Im(In )
Re(I n) Re(I n) Re(I n)

(a) (b) (c)
Fig. 7.8 Travelling wave. Complex amplitude of currents along the axial array, normalized to that at element 1, at ω/ω0 =
0.9757, 1 and 1.0262 (a)–(c)
1 1
0.1 01
0.01 0.01
Im(In )
Im(In )
Im(In )
Re(I n) Re(I n) Re(I n)

(a) (b) (c)
Fig. 7.9 Travelling wave. Complex amplitude of currents along the planar array, normalized to that at element 1, at ω/ω0 =
0.9757, 1 and 1.0262 (a)–(c)
exp(−αa). Note that the radial co-ordinate is on a logarithmic scale

covering three orders of magnitude. All the currents can be seen to lie
on an angular spiral (in the lossless case the spiral would turn into a
polygon). The higher the frequency, the larger the value of βa, the more
tightly wound is the spiral. This is a consequence of the forward nature
of the axial array.
Figure 7.9 shows another example, current distributions for a planar
line with κ = −0.1 and for the same three values of ω/ω0 . The corre-
sponding phase change is now βa = 2π/3, π/2, π/3 and the attenuation
constant αa = 0.12, 0.1, 0.12. The current distributions look similar
to those in Fig. 7.8, they follow a spiral. There are though two major
differences in comparison with the axial case. (i) the sense of rotation
of the spiral has changed and (ii) the higher the frequency the smaller
the phase change from element to element, the less tightly wound is the
|I |
1
0
1
|I |
2
0
1
|I |
3
0
1
|I |
4
0
1
|I |
5
Fig. 7.10 Resonances. Normalized

current amplitudes as a function of fre-
quency for a five-element axial struc- 0
0.95 1 1.05
ture with element 1 being excited. Q =
Z/Z0
100. κ = 0.1
spiral. This is a consequence of the backward nature of the planar array.

(ii) Resonances
If the line is not terminated by its matching impedance, reflections of
1
MI waves from the line’s ends take place, resulting in a standing-wave
|I1|
pattern for the currents. Standing waves that have an integral number of
0 half-periods correspond to the eigenvectors of the system (see eqn (7.17)
1 and Fig. 7.7) which manifest themselves as resonances. In general, a
line consisting of N elements can have up to N resonances at N discrete
|I2|
frequencies given by eqn (7.25). It is clear that all the resonances are
0
1 within the pass band of the MI waves since the argument of the cosine
function can never exceed π.
|I3|
Whether or not any particular resonant mode can be excited depends

0 on the symmetry of the problem. For example, an antisymmetric mode
1
can never be excited under a symmetric excitation and vice versa. To
|I4|
illustrate this we consider again the five-element axial line (for which
0 the eigenvectors and the corresponding resonances are shown in Fig.
1
7.7). Figure 7.10 shows the relative amplitudes of the currents in all five
|I5|
elements as a function of frequency for the case when the first current is
0 being driven by an external voltage. There are five resonant frequencies.
0.95 1 1.05 Note that the third resonance is missing in elements 2 and 4 and the
Z/Z0
second and fourth resonances are missing in element 3.
We can selectively exclude some of the resonances from being excited
Fig. 7.11 Resonances: symmetric exci- by choosing either a symmetric or an antisymmetric excitation. Figure
tation. Normalized current amplitudes 7.11 shows the currents for the case of a symmetric excitation, when
as a function of frequency for a five- the central loop is excited by an external voltage. Due to the symmetry
element axial structure with element 3 argument it is quite obvious why the second and the fourth resonances
being excited. Q = 100. κ = 0.1
that correspond to the antisymmetric eigenmodes are missing here.
Finally, Fig. 7.12 shows the currents for the case of an antisymmetric
1
|I1|
0
1
|I2|
0
1
|I3|
0
1
|I4|
0
1
Fig. 7.12 Resonances: antisymmetric
|I5|
excitation. Normalized current ampli-

tudes as a function of frequency for
0 a five-element axial structure with el-
0.95 1 1.05
ements 1 and 5 being excited in an-
Z/Z0
tiphase. Q = 100. κ = 0.1
1 1
n
n
current I
current I
0 0
−1 −1 Fig. 7.13 Evanescent magnetoinduc-

10 20 30 40 10 20 30 40 tive waves for a lossless 41-element axial
element number n element number n line. κ = 0.1. ω/ω0 = 0.95 (a) and 1.06
(a) (b) (b)
excitation, when the first loop and the last loop are excited by external
voltages in antiphase. In this case the central loop is not excited at all,
and only the second and the fourth resonances are present.
(iii) Excitation outside the pass band
Excitation of an array outside the pass band results in evanescent
magnetoinductive waves. As follows from the dispersion equation (7.1)
there are two different branches of evanescent waves. For the branch
with β = 0 currents in all elements are in phase, while for the branch
with βa = π currents in neighbouring elements are always in antiphase.
We take as an example a lossless 41-element axial line (which is long
enough in order to disregard reflections from the unmatched ends) with
the central element being excited. Figure 7.13 shows the current distri-
butions for two values of the frequency, one from the lower and another
one from the upper stop band, which correspond to αa = 0.2. Obviously
1
0
P1
x/r
0 P2
−1
Fig. 7.14 Poynting vector streamlines
for a 5-element array. From Shamon- −1 0 1 2 3 4 5
c 2002
ina et al. (2002b). Copyright
z/r
American Institute of Physics 0
the current amplitudes decay strongly as we move away from the ele-
ment that is driven by an external source. This result has an important
consequence. When operating in the stop band, a magnetoinductive
line is able to replicate any pattern of excitation by the ‘pixel-to-pixel’
mechanism. This significant property is crucial for the design of the
magnetoinductive near-field lens (see Section 7.15.3).
7.7 Poynting vector

One of the advantages of MI waves is that their properties can be eas-
ily understood both in their circuit and field representations. Currents
flowing in loops create magnetic fields. They can be calculated by first
determining the vector potential from the current (assumed to be con-
stant along the periphery) with the aid of eqn (1.14) and the magnetic
3
The derivation is available in a num- field from the vector potential (eqn (1.10)).3 Having got the magnetic
ber of textbooks (see, e.g., Landau and field the electric field can be obtained from Maxwell’s equations. Having
Lifschitz 1984). The vector potential
is obtained in the form of elliptic func-
both the electric and magnetic fields it is then possible to determine the
tions. Poynting vector.
The starting point is the excitation that is assumed to be done by a
voltage source applied to the first element. From that, and from the in-
verted impedance matrix, the currents can be found. Next, the magnetic
and electric fields need to be calculated from each of the loop currents,
and then they have to be added vectorially to find the total electric and
magnetic fields in a given point. There is then enough information to
determine the direction of the Poynting vector at each point. The cal-
culation was performed both for a 25-element (Shamonina et al., 2002a)
and for a 5-element line (Shamonina et al., 2002b). The Poynting vector
streamlines for the latter case are shown in Fig. 7.14. It may be clearly
seen that the streamlines originate on the first loop and reach, by follow-
ing a variety of paths, the last loop, which is terminated by a matched
impedance. Note that the streamlines approaching the load from the
inside and outside are separated from each other by a so-called P-point,
where the Poynting vector is zero.
7.8 Power in a MI wave 225
7.8 Power in a MI wave

One way of determining the power in a MI wave is to take a cross-
section somewhere between two elements and integrate the Poynting
vector. This can be done only numerically, which would in no way help
in building a physical picture. We shall instead resort to an alternative
analytical method in which the power is calculated by multiplying the
stored energy per unit distance by the group velocity.
The stored energy may be calculated from those stored in the induc-
tance of the loop and in the capacitance, to which the mutual energy due
to coupling between the elements should be added. The full expression
is of the form (Syms et al., 2005b)
1 1 1
W = L|In |2 + C|Vn |2 + M (In In−1
∗ ∗
+ In In+1 ). (7.26)
2 2 2
Note that Vn here is not the voltage applied to the nth element but the
voltage across the capacitor of the nth element, so it is related to In as
In = j ωCVn . (7.27)
Assuming again a travelling wave and substituting the values of In−1
and In+1 in terms of In and making use of the dispersion equation the
expression for the stored energy simplifies to
ω02
W = L|In |2 . (7.28)
ω2
The group velocity may be obtained from the dispersion equation (7.1)
as
!
dω d ω0 ω0 a ω0 3
vg = = p = k sin(ka) , (7.29)
dk dk 1 + κ cos(ka) 2 ω
whence the power can be found as

1
P = W vg = ωM |I0 |2 sin(ka) . (7.30)
2
Note that no power can be transferred at the band edges and optimum
transfer is at the resonant frequency when ka = π/2.
7.9 Boundary reflection and transmission

Let us assume that two MI transmission lines, say both of them in
the axial configuration, are joined together as shown in Fig. 7.15. The
distance between the elements is a1 in line 1, a2 in line 2 and ab across
the boundary. The mutual inductances are M1 in line 1, M2 in line 2
and Mb between the last element of line 1 and the first element of line
2. The elements are numbered in such a way that element 0 is the last
element of line 1 and element 1 is the first element of line 2.
line 1 line 2
V1 Zt
Fig. 7.15 Two magnetoinductive lines
joined together a1 ab a2
Assume further that a MI wave is incident from the left. In general,

when a wave is incident from one medium upon another medium it may
be expected that part of the wave will be reflected back into medium 1
and part of it will propagate in medium 2. Hence, the current distribu-
tions in lines 1 and 2 will be taken in the form (Syms et al., 2005b)
In = I00 e −j n(ka)1 + R e j n(ka)1 , n ≤ 0

h i
(7.31)
and
In = I00 T e −j n(ka)2 , n > 0, (7.32)

where I00 is a constant, R and T are the reflection and transmission
coefficients, and (ka)1 and (ka)2 are the phase change per element in
media 1 and 2, respectively.
We may now write Kirchhoff’s law for elements 0 and 1, the elements
on the opposite sides of the boundary, in the form
Z01 I0 + j ωM1 I−1 + j ωMb I1 = 0 (7.33)

and
Z02 I1 + j ωM2 I2 + j ωMb I0 = 0 . (7.34)

Substituting eqns (7.31) and (7.32) into eqns (7.33) and (7.34) the two
unknowns R and T may be determined as follows
Mb2 e −j (ka)2 − M1 M2 e −j (ka)1

R = , (7.35)
M M e j (ka)1 − M 2 e −j (ka)2
1 2 b
2j M1 Mb sin [(ka)1 ]
T = . (7.36)
M1 M2 e j (ka)1 − Mb2 e −j (ka)2
In the special case when M1 = Mb = M2 eqns (7.35) and (7.36) reduce

to
e −j (ka)2 − e −j (ka)1
R = , (7.37)
e j (ka)1 − e −j (ka)2
2j sin [(ka)1 ]
T = . (7.38)
e j (ka)1 − e −j (ka)2
7.10 Tailoring the dispersion characteristics: biperiodic lines 227
This situation can arise only when the sole difference between the two
media is that the self-impedances (and hence the dispersion curves) are
different.
Note that eqn (7.37) appears in Tretyakov’s book (2003) but the un-
derlying physics is quite different in the two cases. Here, we are con-
cerned with the reflection and transmission of waves propagating in dif-
ferent periodic media, whereas Tretyakov’s expression is valid when a
plane electromagnetic wave is incident upon a periodic medium and
higher-order modes at the boundary can be disregarded. It is a coinci-
dence that the expressions are identical.
A further simplification can be obtained by considering the continuous
limit when (ka)1 , (ka)2 ≪ 1. Then, the reflection and transmission
coefficients reduce to
k1 − k2 2k1
R= and T = . (7.39)
k1 + k2 k1 + k2
The above expressions look quite familiar, occurring for example when
Schrodinger’s equation is solved for an electron wave incident upon a
potential barrier (see, e.g., Solymar and Walsh 2004). There are no
periodic media in that case, simply a wave incident from a medium with
propagation constant k1 upon a medium with propagation constant k2 .
For the reflection and transmission coefficients to make sense it is a
necessary condition that the power flow should be the same in lines 1
and 2. The power in lines 1 and 2 may be written as
1
1 − |R|2 ωM1 sin(ka)1 ,

P1 = (7.40)
2
1 2
P2 = |T | ωM2 sin(ka)2 . (7.41)
2
Substituting eqns (7.34) and (7.35) into the above equations for power
it may be shown, using a fair number of algebraic operations, that P1 =
P2 , i.e. the power across the boundary is conserved.
7.10 Tailoring the dispersion

characteristics: biperiodic lines
For small coupling coefficients, which is usually the case, the pass band
of MI waves is narrow. However, in many cases this may not be de-
sirable. Ideally, we would like to tailor the dispersion characteristics
to any requirement. We might, for example, wish to realize two pass
bands instead of one. Another example, to which we shall return in
Chapter 8 where non-linear relations are discussed, is parametric ampli-
fication (see, e.g., Sydoruk et al. 2007a). In that case we have a signal
wave that has a propagation coefficient β at a frequency ω. The aim is
to amplify this signal wave with the aid of a pump wave propagating
on the same structure at twice the frequency and with a propagation
(2f/m) 1/2
frequency
(2f/M)1/2
M f m f M
0 0.5 1
2a propagation constant 2Ea/S
(a) (b)
Fig. 7.16 (a) Diatomic chain of atoms and (b) the phonon dispersion curve with two branches
coefficient 2β. The two waves are then in synchronism because they
propagate with the same phase velocity
ω 2ω
vp = = , (7.42)
β 2β
and the signal wave can be amplified by transferring power from the
pump wave.
In this section our aim is to show that we have some freedom over
the dispersion characteristics and, in particular, we can realize the syn-
chronism condition. The way to do it is suggested by the close analogy
between MI waves and acoustic waves. It is well known that the disper-
sion characteristics of diatomic solids differ greatly from those having
identical elements (see, e.g., Brillouin 1953). The consequence of having
a material in which two different atoms of different masses alternate (e.g.
NaCl) is the appearance of a new band, known as the optical branch, in
the dispersion characteristics. A sketch of the dispersion characteristics
of such a diatomic solid (Brillouin, 1953) is shown in Fig. 7.16.
We have now two pass bands. When it comes to acoustic waves in a
solid we have very little freedom. The two different masses are provided
by Nature and we have very little control over the way such masses
arrange themselves. However, when metamaterial elements are put next
to each other we can build every single element as we wish.
There are two obvious ways of achieving double periodicity: (i) To
change some parameter of the element (L or C resulting in a change of
resonant frequency) and (ii) to vary the distance between the elements,
which means that there will be two different mutual inductances. These
possibilities are shown schematically in Figs. 7.17(a) and (b) for the pla-
nar configuration and in Figs. 7.17(c) and (d) for the axial configuration.
Let us look at such a biperiodic line and apply Kirchhoff’s equations
to elements 2n and 2n + 1 (see Fig. 7.18). We find
Z01 I2n + j ωM1 I2n−1 + j ωM2 I2n+1 = 0 (7.43)

and
Z02 I2n+1 + j ωM2 I2n + j ωM1 I2n+2 = 0 . (7.44)

7.10 Tailoring the dispersion characteristics: biperiodic lines 229
(a) (b)
Fig. 7.17 Schematic representation of
planar (a) and (b) and axial (c) and
(d) biperiodic configurations. (a), (c)
Resonant frequency varies from element
to element. (b), (d) Distance varies
between neighbouring elements. From
Sydoruk et al. (2005). Copyright c
(c) (d) 2005 American Institute of Physics
a1 a2
M1 M2
Z01 Z02 Z01 Z02

Fig. 7.18 Schematic view of a biperi-
2n-1 2n 2n+1 2n+2 odic chain of metamaterial elements
We shall assume now propagating solutions both for the even- and for
the odd-numbered elements
I2n = A2 e −j k2n(a1 + a2 ) (7.45)

and
I2n+1 = A1 e −j k(2n + 1)(a1 + a2 ) . (7.46)

Substituting eqns (7.45) and (7.46) into eqns (7.43) and (7.44) we find
after some algebraic operations the dispersion equation
r
Z01 Z02
− − (M1 − M2 )2
k(a1 + a2 ) 1 ω 2
cos = p . (7.47)
2 2 M1 M2
Note that eqn (7.47) reduces to eqn (7.1) when Z01 = Z02 = Z0 and
M1 = M2 = M , as it should.
Let us now look at two practical examples, one for the planar, and
one for the axial configuration. The parameters chosen are as follows:
r0 = 10 mm, wire thickness dw = 2 mm, L = 33 nH, C1 = 208 pF,
C2 = 177 pF, ω01 = (LC1 )−1/2 = 0.95 ω0, ω02 = (LC2 )−1/2 = 1.05 ω0 ,
ω0 /2π = 63.87 MHz.4 For the axial configuration we choose a = 10 4
The reason for this choice is that 63.87
mm resulting in M/L = 0.149 and for the planar case a = 20.5 mm, MHz corresponds to the magnetic reso-
nance frequency of a proton for a mag-
corresponding to M/L = −0.104. Losses are taken into account by the netic field of 1.5 T, one of the possible
quality factor, Q = 150. The dispersion curves with both propagation choices for magnetic resonance imag-
and attenuation constants are shown in Figs. 7.19 and 7.20 for the axial ing.
and planar cases, respectively.
ZZ0
1
1.2
0 Ea/S 1
Fig. 7.19 Dispersion curve for the 'optical’ branch

ZZ0
biperiodic axial configuration. Cur-

rents within a period are in antiphase in
1
the upper (‘optical’) and in phase in the
lower (‘acoustic’) branch. Inset shows 'acoustic’ branch
the dispersion curve for the singly pe-
riodic axial configuration with the res-
onant frequency ω0 /2π = 63.87 MHz
and coupling between nearest neigh-
bours 2M/L = 0.149. From Sydoruk 0.8
0 0.5 1 0.1 1
ican Institute of Physics 2Ea/S 2Da/S
ZZ0
1
1.2
0 Ea/S 1
'acoustic’ branch
Fig. 7.20 Dispersion curve for the
ZZ0
biperiodic planar configuration. Cur-

rents within a period are in phase 1
in the upper (‘acoustic’) and in an-
tiphase in the lower (‘optical’) branch. 'optical’ branch
Inset shows the dispersion curve for
the singly periodic planar configuration
with the resonant frequency ω0 /2π =
63.87 MHz and coupling between near-
est neighbours 2M/L = −0.104. From 0.8
0 0.5 1 0.1 1
Sydoruk et al. (2005). Copyright c
2005 American Institute of Physics 2Ea/S 2Da/S
It may be immediately seen that the major distinction between the

axial and planar cases, that one gives a forward wave and the other one
a backward wave, is no longer there. In both cases the lower branches
are forward waves and the upper branches are backward waves. There
is, however, a difference if we consider the phases of the currents within
a pair constituting the unit cell. For the axial line the currents of the
neighbouring elements in the upper branch are in antiphase and the
currents in the lower branch are in phase (Fig. 7.19). Using the analogy
with the diatomic model we can refer to the upper branch as ‘optical’
and to the lower branch as ‘acoustic’. For the planar line the situation
is reversed. The currents in the lower branch are now in antiphase, thus
this is the one we should, strictly speaking, call ‘optical’ (Fig. 7.20).
Let us next consider parametric amplification and the problem of en-
suring synchronism between the signal wave and the pump wave. We
choose the signal frequency at ω0 /2π = 63.87 MHz. The propagation
constant may be chosen at 2βa = π/3 and the propagation constant of
2.8
(2k, 2ω )
0
2
0
ω/ω
1.5
Fig. 7.21 Dispersion curve for the
(k, ω ) biperiodic array permitting the propa-
0
1 gation of both signal (k, ω) and pump
(2k, 2ω) waves required for paramet-
0.8
ric amplification. From Sydoruk et al.
1/3 2/3 1 c 2005 American
(2005). Copyright
2βa/π
the pump wave as 2βa = 2π/3 at the frequency 2ω0 /2π = 127.74 MHz.
These requirements can be satisfied with an axial structure where the
distance between the elements is a = 0.5ro (M = 0.336L), C1 = 164 pF,
C2 = 56 pF, L = 33 nH. The corresponding dispersion curve is shown in
Fig. 7.21. It can be seen that the condition of synchronism is satisfied
and both the signal and the pump waves can propagate in the system.
So far we have talked about an infinitely long biperiodic line. Does a
terminal impedance exist that matches the line, that makes it possible to
have a single travelling wave? Since the line consists now of two different
kinds of elements it may be expected that we shall need two terminal
impedances, one to be inserted into the last, and the other one into the
last-but-one element. This may be shown to be the case. The values of (a)
the two terminal impedances may be found from a calculation similar
to that in Section 7.3 as
M1,2 Z01,02
ZT (1, 2) = − . (7.48) (b)
M1,2 + M e j k(a1 + a2 )
2,1
7.11 Experimental results (c)
The first experimental results were obtained (Wiltshire et al., 2003b) not
long after the derivation of the dispersion equation (Shamonina et al.,
Fig. 7.22 Capacitively loaded loops.
2002a). The basic element of the line, a capacitively loaded loop was re- Photographs of single element (a) and
alized by winding two turns of 1-mm diameter copper wire on a dielectric fragments of an axial line (b) and a
rod (diameter 9.6 mm), and tuned to the desired frequency of 60 MHz by planar line (c). From Wiltshire et al.
inserting a capacitor (nominally 100 pF) between the ends of the wire, (2003b). Copyright c 2003 IEE
as shown in Fig. 7.22(a). Two lines were assembled from these elements,
an axial line in which 32 elements were arranged along the axis of a di-
electric rod, spaced by their diameter (Fig. 7.22(b)) and a planar line in
which 15 elements were placed side-by-side (Fig. 7.22(c)). The resonant
frequency and the quality factor were determined as f0 = (61.4 ± 0.4)
MHz and Q = 48 ± 5.
12 0
phase/ S [rad]
amplitude [dB]
8 −20
4 −40
0 −60
10 20 30 10 20 30
element number element number
(a) (b)
Fig. 7.23 Phase (a) and amplitude variation (b) along axial structure at 61 MHz. From Wiltshire et al. (2003b). Copyright
c 2003 IEE

64 64
frequency f [MHz]
frequency f [MHz]
60 60
56 56
0 0.5 1 0 0.2 0.4
Ea/S Da/S
(a) (b)
Fig. 7.24 Dispersion relationship for axial line. (a) Propagation constant and (b) attenuation constant versus frequency. From
c 2003 IEE
Wiltshire et al. (2003b). Copyright
The overall transmission characteristics were measured with a network

analyzer at 401 frequency points in the range 50–70 MHz by placing the
transmitter loop near to the first element and the receiver loop near to
the last element. The measurements showed clearly the existence of a
pass band flanked by two stop bands.
The dispersion characteristics were measured using the same setup,
with the input loop next to the first element and the output loop over
the rod adjacent to each successive element in turn, and the phase and
amplitude of the output wave measured at each element for the same
401 frequencies. A typical behaviour of the phase and amplitude against
element position at 61 MHz is shown by circles in Figs. 7.23(a) and
(b). The phase variation may be seen to be substantially linear. The
amplitude variation, plotted on a logarithmic scale, is also linear, apart
from the end of the line where there is some variation due to a reflected
wave (note that no attempt was made to match the line). The theoretical
values are shown by solid lines. The agreement is quite good.
66 66
frequency f [MHz]
62 frequency f [MHz] 62
58 58
0 0.5 1 0 0.1 0.2
Ea/S Da/S
(a) (b)
Fig. 7.25 Dispersion relationship for planar line. (a) Propagation constant and (b) attenuation constant versus frequency.
c 2003 IEE
From Wiltshire et al. (2003b). Copyright
SSRRs
Microstrips
d1
d2
s
h1 h2 w Fig. 7.26 (a) Photograph and sketch of
Ground the planar MI transducer. (b) sketch of
Dielectric plane the element. From Freire et al. (2004).
substrate Copyright c 2004 American Institute
(a) (b) of Physics
Having obtained the phase and amplitude variation from element to

element at all the frequencies we can now deduce the dispersion curves
versus βa/π and αa/π. These are shown in Figs. 7.24 and 7.25 for the
axial and planar configurations, respectively. The theoretical curves,
solid lines, are obtained from the parameters of the two lines and the
measured value of the quality factor. The agreement is very good for
the phase variation and quite reasonable for the attenuation.
Propagation of MI waves at a much higher frequency (in the range
of 3.5 to 5 GHz) were measured by Freire et al. (2004) using a MI
transmission line between two microstrip lines as shown in Fig. 7.26(a).
The shape of the individual elements is a variation on the split-ring
resonator, as may be seen in Fig. 7.26(b). The high length to width
ratio was chosen to strengthen the coupling between adjacent elements.
They found a pass band centred around 4.5 GHz. The total transmission
loss was close to 3 dB, i.e. the loss per element was 0.6 dB, in contrast
to about 1.4 dB per element that follows from Fig. 7.23(b).
c 2006 IOP Publishing

Fig. 7.27 Single-sided (a) and double-sided PCB element (b). From Syms et al. (2006b). Copyright
Ltd
2.0
PCB thickness
1.5
1.0
κ
Single-sided
Double-sided
0.5
Fig. 7.28 Variation of the coupling co- 0.0

efficient κ with the element spacing a.
0 5 10 15 20
From Syms et al. (2006b). Copyright
c 2006 IOP Publishing Ltd a (mm)
Since low attenuation is a necessary condition for any practical appli-

cation for signal processing further attempts were made by Syms et al.
(2006b) to lower the attenuation. The solutions chosen may be seen in
Fig. 7.27. In Fig. 7.27(a) the turns made of copper are only on one side
of a printed circuit board (PCB), whereas in Fig. 7.27(b) the inductor is
double-sided. There are pads in both cases for adding capacitors to the
inductors in order to bring the resonant frequency to the desired value.
As may be expected, the coupling coefficient is higher for the double-
sided case as shown in Fig. 7.28. It is worth noting that by placing the
elements close to each other in the axial configuration a coupling coeffi-
cient as high as 1.5 can be achieved, which makes wide-band operation
possible.
The experimental setup using double-sided elements is shown in Fig.
7.29(a). The measured values of S21 as a function of frequency for a 30-
element line and for six different element spacings may be seen in Fig.
7.29(b). For the smallest spacing of 2.5 mm between the elements the
lowest value of S21 is about 4.4 dB. The coupling losses were measured as
0.4 dB, both at the input and at the output. This low value was achieved
by introducing matching elements between the first (last) element of
the line and the network analyzer. Deducting the coupling losses the
(a)
Axial No. of elements

spacing
(b) (c)
Fig. 7.29 (a) Experimental arrangement of MI waveguides; (b) experimental frequency variation of S21 for waveguides based
on double-sided coils and a fixed number of elements (30) and different element spacing, and (c) corresponding result for a fixed
element spacing (2.5 mm) and different numbers of elements. From Syms et al. (2006b). Copyright c 2006 IOP Publishing
Ltd
attenuation per element comes to the low figure of 0.12 dB per element.
The dependence on the number of elements of the S21 versus frequency
curve is shown in Fig. 7.29(c). It may be seen that it is feasible to set
up long lines.
All the experiments mentioned so far were on lines consisting of iden-
tical elements. We shall now report experimental results measured on
biperiodic lines by Radkovskaya et al. (2007a). The loop used in the
experiments, we shall call it a split pipe, is shown in Fig. 7.30(a). It
became a resonant element at a desired frequency when loaded by a ca-
pacitor. Two sets of elements were used, one loaded by a capacitor of
330 pF, and the other one by a capacitor of 680 pF, yielding resonant
frequencies of 46.21 MHz and 32.46 MHz. The two sets were interleaved
to produce a biperiodic line as shown in Fig. 7.30(b) for the axial and in
Fig. 7.30(c) for the planar configuration where the positions of the trans-
mitter and receiver coils are also shown. The measurement technique
was the same as described earlier in this section.
The measured and theoretical curves of ω versus βa are shown in Figs.
7.31(a) and (b) for the axial and planar configurations, respectively. The
Receiver coil
Transmitter
coil
w Receiver
coil
r
h
g
Transmitter coil
(a) (b) (c)
Fig. 7.30 Biperiodic arrays of capacitively loaded split pipes. Schematic representations of (a) element dimensions, (b) axial
and (c) planar configurations with measuring coils. From Radkovskaya et al. (2007a). Copyright c 2007 IEE
50 50
frequency [MHz]
frequency [MHz]
45 45
axial array planar array
40 theory 40 theory
experiment experiment
35 35
30 30
0 0.5 1 0 0.5 1
Ea/ S Ea/ S
(a) (b)
Fig. 7.31 Dispersion characteristics for the axial (a) and planar (b) biperiodic structure. From Radkovskaya et al. (2007a).
c 2007 IEE
Copyright
agreement is remarkably good for the planar line and quite good for the
axial line. In particular, it should be emphasized that for the biperiodic
line there is hardly any difference between the dispersion curves of the
axial and planar configurations as predicted by eqn (7.47).
7.12 Higher-order interactions

We have so far considered nearest-neighbour interaction only. This is
usually a good approximation when there is fast decay of the fields away
from the element. By fast decay we mean cubic decay as would be
the case for elements that can be regarded static magnetic dipoles. The
question nevertheless arises: what kind of modifications in the properties
of MI waves would be caused by taking higher-order interactions into
account? The generalization to higher interactions is straightforward.
In eqn (7.2) Kirchhoff’s voltage law is applied to element n assuming
nearest-neighbour interaction only. If elements further away may also
induce voltages then eqn (7.2) will modify to
7.12 Higher-order interactions 237
1.4
N=1
N=2
N=5
12
N=20
ZZ0
Fig. 7.32 Dispersion characteristic of a

0.8 MI waveguide assuming a cubic decay
of the coupling constant with distance
0 0.5 1 and including interactions up to N = 1,
Ea/S 2, 5 and 20
1.4
N=1
N=2
12 N=5
N=20
ZZ0
0.8
Fig. 7.33 Same as in Fig. 7.32 but as-
0 0.5 1 suming a quadratic decay of the cou-
Ea/S pling constant with distance
∞
X
Z0 In + j ω Mm (In+m + In−m ) = 0 , (7.49)
m=1
where Mm is the mutual inductance between two elements a distance
ma from each other. Assuming again a wave solution of the form of
eqn (7.4) the dispersion equation can be derived as
∞
ω02 X
1− + κn cos(nkd) = 0 , (7.50)
ω 2 n=1
where κn = 2Mn /L and only the lossless case is considered. How large
is the influence of higher-order couplings? Assuming cubic decay of the
mutual inductance with distance and κ1 = 0.5 the dispersion equation
is plotted in Fig. 7.32 for N = 1, 2, 5 and 20. The effect of higher
interactions may be seen to be small. For finite-size elements the cubic
decay is not a good approximation. The actual decay may be closer to a
quadratic one. In Fig. 7.33 we plot again the dispersion curve for N = 1,
2, 5 and 20 and κ1 = 0.5. For this lower decay, as expected, the effect
of higher orders is more significant.
Can we say anything in more general terms about the dispersion
curve? The answer is yes, when κn can be expanded into a series in
M2 M2
M4 M4
Fig. 7.34 Schematic representation of M M M
the coupling between two lines of reso-
nant magnetic metamaterial elements. M3 M3
Mutual inductances M1 , M2 , M , M3
and M4 between the nearest neighbours M1 M1
are shown by arrows. From Sydoruk
c 2006 by the
American Physical Society (n − 1)th nth (n + 1)th
inverse powers of the distance,

∞
X
κn = cm (na)−m , (7.51)
m=2
where the coefficients cm need to be determined. With the aid of
eqn (7.51) we find the dispersion equation (7.50) reduces to the form
∞ ∞
ω02 X
−m
X
1− + c m a n−m cos(nka) = 0 . (7.52)
ω 2 m=2 n=1
Note that the summation over n can be expressed in terms of Lim , a set
of special functions called polylogarithms (Lewin, 1981). The relation is
∞
1h
Lim e j ka + Lim e −j ka .
X i
n−m cos(nka) = (7.53)
n=1
2
For our purpose the most important property of this special function
is that it is monotonic, from which it follows that the dispersion curve,
provided the expansion of κn is possible in the form of eqn (7.51), is also
monotonic.
The generalization to higher-order interactions is also straightforward
for the case when there is excitation by a set of voltages. The mathe-
matical form is still given by eqn (7.13) as
V = ZI , (7.54)
but Z is no longer a tri-diagonal matrix. The main diagonal elements
are still equal to Z0 but the off-diagonal elements are now equal to
Zij = j ωM|i−j| , i 6= j . (7.55)

For a given set of voltage excitations we may still obtain the current
distribution by inverting the relation in eqn (7.13).
7.13 Coupled one-dimensional lines

We shall now look at the case (Sydoruk et al., 2006) when two lines,
both capable of propagating MI waves, are coupled to each other, as
shown schematically in Fig. 7.34.
We shall describe their properties by first assuming nearest-neighbour

interaction. However, in contrast to a one-dimensional line, there are
now not two but five nearest neighbours. The notations for the mu-
tual inductances are shown in Fig. 7.34 as M1 (between neighbouring
elements in line 1), M2 (between neighbouring elements in line 2), M M1 M1
(between the nth element in line 1 and the corresponding nth element in h M
line 2), M3 (between the (n − 1)th element in line 1 and the nth element
in line 2), and M4 (between the nth element in line 1 and the (n − 1)th
M1 M1
element in line 2). a
Kirchhoff’s voltage equations written for element n in line 1 and the (a)
corresponding element n in line 2 are as follows,
Z01 In + j ωM1 (In−1 + In+1 ) + j ωM Jn + j ωM4 Jn−1 + j ωM3 Jn+1 = 0

(7.56)
and
M2 M2
Z02 Jn + j ωM2 (Jn−1 + Jn+1 ) + jωM In + j ωM3 In−1 + j ωM4 Jn+1 = 0 ,
(7.57) h -M3 M3
where In and Jn are the currents in lines 1 and 2, respectively. Assuming M1 M1
further wave solutions in both lines 1 and 2 eqns (7.56) and (7.57) yield a
the dispersion equation in analytic form (Sydoruk et al., 2006). We (b)
shall show here only the results for two different coupling arrangements,
as shown in Figs. 7.35(a) and (b), where the mutual inductances taken
into account are also shown. The first one shows two planar lines placed Fig. 7.35 (a) Configurations of bi-
atomic metamaterial structures (a) pla-
above each other, a simple enough situation, but even then we find the nar lines above each other. (b) Planar
unusual coupling relationship that the intraline mutual inductances are and axial lines above each other. The
negative, whereas the interline mutual inductances are positive. Figure non-zero mutual inductances are shown
7.35(b) shows a planar line coupled to an axial line. The interesting by arrows. From Sydoruk et al. (2006).
Copyright c 2006 by the American
feature is now that the upper line carries a forward wave, the lower line Physical Society
a backward wave and the elements just above each other are not coupled.
What kind of dispersion equation curves would we expect for the two
coupled planar lines? A single planar line supports backward waves. If
the coupling between the two planar lines is small (the two lines are at
a distance of 20 mm from each other) then we may expect a small split
in the dispersion curve. If the distance between the lines is less (10 mm)
then we may expect a bigger split.
Experiments on two coupled lines were conducted by Sydoruk et al.
(2006) using the split-pipe elements shown in Fig. 7.30(a) used in the
experiments of Radkovskaya et al. (2007b). The loading capacitor had
a capacitance of C = 330 pF, which led to a resonant frequency of
46.21 MHz. For the planar lines (Fig. 7.35(a)) the experimental and
theoretical results are shown in Figs. 7.36(a) and (b) for line separations
of h = 20 mm and 10 mm. As expected, there is a small split for the
large separation and a large split for the smaller separation.
The interesting result is that when the split is large we have effectively
two pass bands with a stop band between them. The same elements were
used for the coupled line shown in Fig. 7.35(b), where a backward wave is
55 55
frequency [MHz]
frequency [MHz]
50 50
45 45
40 40
0 0.5 1 0 0.5 1
Ea/S Ea/S
(a) (b)
Fig. 7.36 Dispersion characteristics of coupled planar lines with h = 20 mm (a) and 10 mm (b). Theory (solid lines) and
experiment (circles and squares). From Sydoruk et al. (2006). Copyright
50 50
frequency [MHz]
frequency [MHz]
46 46
42 42
0 0.5 1 0 0.5 1
Ea/S Ea/S
(a) (b)
Fig. 7.37 Dispersion characteristics of coupled planar–axial lines with h = 30 mm (a) and 15 mm (b). Theory (solid lines)
and experiment (circles and squares). From Sydoruk et al. (2006). Copyright c 2006 by the American Physical Society
coupled to a forward wave. The theoretical and experimental dispersion

curves are shown in Figs. 7.37(a) and (b). As may be expected, we have
a combination of forward and backward waves. Notice that there is a
stop band around the resonant frequency. For smaller coupling (h = 30
mm) the stop band is smaller, for larger coupling (h = 15 mm) the stop
band is larger. The agreement between theory and experiment is very
good, although we have a few spurious experimental points that cannot
be accounted for by the theory.
Having looked at the dispersion curves we shall return to the coupled
lines with a different question in mind. How large is the transmission,
i.e. what is the value of S21 , between the first element of line 1 and the
last element of line 2 (see Fig. 7.38(a) for the experimental arrangement)
and how does it vary as the two lines are shifted (Fig. 7.38(b)) relative to
each other? The theoretical calculations have been done by Radkovskaya
et al. (2007b) by assuming interaction between any two elements, i.e.
relying on the generalized Ohm’s law presented in eqn (7.13). This
a Out a Out
h h
In In
(a) (b)
Fig. 7.38 Schematic representation of the (a) unshifted and (b) half-a-period shifted coupled lines. The first element of the
lower array is excited by a transmitting coil and the signal in the last element of the upper array is measured by a receiving
coil. From Radkovskaya et al. (2007b). Copyright c 2007 Wiley-VCH Verlag GmbH & Co. KGaA
50 50 80
49 49
Frequency [MHz]
Frequency [MHz]
40
48 48
47 47
0
46 46
45 45 −40
44 44
0 1 2 3 4 0 1 2 3 4
Relative shift, '/a Relative shift, '/a
(a) (b)
Fig. 7.39 Contour plots of transmission between the split-pipe arrays as a function of frequency and shift, (a) experiment, (b)
theory. From Radkovskaya et al. (2007b). Copyright c 2007 Wiley-VCH Verlag GmbH & Co. KGaA. For coloured version see
plate section
necessitated the determination of the mutual inductances between any

two elements. It was done by assuming each element to be represented
by a filamentary loop. If we know all the elements of the N ×N Z matrix
then the current distribution may be obtained by inverting numerically
the Z matrix. The experiments were performed and comparisons with
theoretical results were made by Radkovskaya et al. (2007b) as shown
in Figs. 7.39(a) and (b). The amplitude of S21 is colour-coded. The
variables are the shift between the elements (this is done for four full
periods) on the horizontal axis and the frequency on the vertical axis.
The agreement between theory and experiment may be regarded as very
good. Note that maximum transmission occurs when the shift is a half-
period and minimum transmission when the elements are just above each
other. The results have an interesting implication for coupled lines. As
far as we know this is the first time ever that by reducing the coupling
between opposite elements in coupled lines the transfer of power actually
increases. The physical reason can be attributed to the dispersion curve
shown in Fig. 7.36(b). When the coupling between the two lines is strong
a stop band appears and total transmission is minimum.

z [cm]

rotating
dipole
í
Fig. 7.40 Schematic view of rotational í

resonator composed of 24 capacitively
loaded loops. From Solymar et al.
c 2006 American
x [cm]
(2006). Copyright
y [cm] í í
7.14 Rotational resonance

We have discussed so far MI wave propagation along a line and analyzed
what happens when two lines are coupled to each other. We shall now
consider the case when the MI wave travels round and round on a ring
structure of resonant elements, as shown in Fig. 7.40.
We call the resulting resonance a rotational resonance (Solymar et al.,
2006). Such a construction is known in microwaves as a strip-line ring
resonator (Chang, 1996). It operates on the principle that the total
phase shift that a wave accumulates round a closed path should be an
integral multiple of 2π. An alternative description is that the resonance
occurs when the circumference of the circle is equal to an integral number
of wavelengths of the MI wave. Another analogy is with the cavity
magnetron that is operated by coupling power from a circulating electron
beam to an electromagnetic wave propagating round a ring of coupled
cavity resonators.
The general dispersion equation, with all interactions taken into ac-
count, is given by eqn (7.50). The difference is that in the ring structure
there are only a finite number of interactions, depending obviously on
the number of elements. As it happens, there is a slight difference in the
dispersion equation depending on whether the number of elements con-
stituting the ring is odd or even. The dispersion equations can be found
easily with the aid of a few algebraic operations. For even numbers it is
N −1
ω02 κN −j N ka X
1− 2 + e + κn cos(nka) = 0 , (7.58)
ω 2 n=1
where the number of elements is equal to 2N , κn = 2Mn /L and Mn is

the mutual inductance between two elements n neighbours apart. For
an odd number of elements the dispersion equation is
Table 7.1 Rotational resonances ωn /ω0 for a 9-element ring. From Soly-
mar et al. (2006)
resonance number n nearest-neighbour case all-interaction case
0 1.069 1.085
1 1.052 1.049
2 1.011 1.005
3 0.970 0.969
4 0.946 0.949
N
ω2 X
1 − 02 + κn cos(nka) = 0 , (7.59)
ω n=1
where the number of elements is equal to 2N + 1. As an example we

shall choose a 9-element ring. The elements are assumed to be circular
resonant loops nearly touching each other. The corresponding positions
of the rotational resonances are given in Table 7.1 both in the nearest-
neighbour and in the all-interaction cases for a quality factor, Q = 100.
It may be seen that the more accurate calculations show a small shift in
the resonant frequencies.
7.15 Applications
7.15.1 Introduction
Waves propagating on coupled resonant structures had applications in
microwave tubes and in linear accelerators (Bevensee, 1964). In the
metamaterial context at the time of writing we can only talk about po-
tential applications. To that category belongs the early work of Wiltshire
et al. (2001) who conducted, with the aid of swiss rolls, magnetic infor-
mation from an MRI machine to a detector. An application as a delay
line was envisaged by Freire et al. (2004). Their device was shown in Fig.
7.26(a). The information travelled from one strip line to the other strip
line via a MI wave. The maximum delay time measured was about 6 ns.
Phase shifters were designed by Nefedov and Tretyakov (2005). They
made good use of the fact that lines with positive phase shift (those
that support forward waves) and also with negative phase shift (those
that support backward waves) are simultaneously available.5 Near-field 5
That arrangement can lead to shorter
imaging was demonstrated by Freire and Marques (2005). Later, it was phase shifters. If, for example, a phase
shifter of -20◦ is required it can be a
shown by the same group (Mesa et al., 2005) that the imaging strongly short one if negative phase shift is avail-
depends on the characteristics of the receiver. able. On the other hand, if one has to
In this section we shall investigate three kinds of applications in a little rely on a positive phase shift then the
more detail. Section 7.15.2 will describe the signal processing aspects phase shifter must be 17 times larger to
offer a phase shift of 340◦ .
of MI waves that require various waveguide components. Section 7.15.3
will discuss potential applications for imaging with some experimental
results showing the quality of imaging obtained. Finally, Section 7.15.4
1.0 1.0
Transmitted
Normalized power
0.8 Transmitted 0.8
0.6 P Normalized power 0.6

0.4 0.4
0.2 Reflected 0.2 Reflected

0 0
0.90 0.95 1.00 1.05 1.10 1.15 0 1 2 3 4
ZZ 0 P
(a) (b)
Fig. 7.41 Magnetoinductive waveguide mirror. (a) Variation of |R|2 and |T |2 with ω/ω0 for κ = 0.2 and different values of µ.
(b) Variation of |R|2 and |T |2 with µ for ω = ω0 . From Syms et al. (2006a). Copyright
c 2006 IEE
will resurrect the rotational resonance of the previous section and discuss
briefly its potential applications in magnetic resonance measurements.
7.15.2 Waveguide components

The design of various MI waveguide components was treated in Shamon-
ina and Solymar 2004; Syms et al. 2005c; Syms et al. 2006a. We shall
present here some of the components discussed by Syms et al. (2006a).
The first one is a frequency-dependent reflector that we have already
analyzed in Section 7.9 as a boundary-reflection problem. The setup
may be seen in Fig. 7.15. Two one-dimensional lines may be seen to be
coupled to each other by a mutual inductance Mb across the boundary.
Let us now simplify the problem and say that the two lines are identical,
so that the only new parameter is Mb . Then, eqns (7.35) and (7.36) lead
to the power reflection and transmission coefficients
(µ2 − 1)2 4µ2 sin2 (ka)

R2 = and T2 = , (7.60)
D D
where
Mb
D = 1 + µ4 − 2µ2 cos(2ka) and. (7.61)
µ=
M
The normalized power for κ = 0.2 and µ = 0.9, 0.8 and 0.7 is shown in
Fig. 7.41(a) as a function of frequency over the range corresponding to
the pass band of the MI wave. The curves are slowly varying across the
band. Towards the band edge, where ka = 0 or π, the power reflection
coefficient tends to unity: the discontinuity reflects everything. Near the
centre of the band the transmission is high but steadily reduces with µ.
In this region we may obtain the approximation
µ2 − 1 2µ
R= and T = . (7.62)
µ2 + 1 µ2 + 1
–2 –1 0 1 2
Zm Z Zm Z Zm 1 Z Zm 2 Z Zm Z Zm
Transmitted
Incident + Reflected
(a)
1.0 1.0
Normalized power
0.8 0.8
Transmitted power
P (=P )

0.6 0.4 0.6
0.2
0.4 0.4
02 0.2
0 0
0.90 0 95 1.00 1.05 1.10 1.15 0.90 0.95 1.00 1.05 1.10 1.15
(b) (c)
Fig. 7.42 (a) Magnetoinductive waveguide Fabry–Perot resonator. (b) Variation of |T |2 with ω/ω0 for κ = 0.2 and different
values of µ. (c) Variation of |T |2 with ω/ω0 for two-loop cavity with κ = 0.2 and µ = 0.2. From Syms et al. (2006a). Copyright
c 2006 IEE
Figure 7.41(b) shows the variation of R2 and T 2 with µ obtained from

the above equation. These results are independent of ka and show that
a reflector with reasonable broadband performance may be constructed
by a slight variation in the spacing at the junction between two lines.
As known in optics, two collinear reflectors make up a Fabry–Perot
resonator and that applies to MI waves as well. This is achieved by
inserting an additional reflector as shown in Fig. 7.42(a). The mutual
inductance between the elements is taken as M , with the exception of
those between elements −1 and 0, and between 0 and +1, which are
denoted by M1 and M2 , respectively. We may then write Kirchhoff’s
equations for elements −1, 0 and 1 as follows
Z0 I−1 + j ωM1 I0 + j ωM I−2 = 0, (7.63)

Z0 I0 + j ωM2 I1 + j ωM1 I−1 = 0, (7.64)
Z0 I1 + j ωM I2 + j ωM2 I0 = 0. (7.65)
The current in the uniform line to the left of element −1 may be

assumed in the form of an incident and a reflected wave, and as a trav-
eling wave to the right of element 1. There are three unknowns then,
the reflection coefficient, the transmission coefficient and the current in
Uniform Periodically varying Uniform

/=2a
0 1 2 N N+1 N+2
Zm Zm Zm 1 Zm 2 Zm 1 Zm 2 Zm 1 Zm 2 Zm Zm
Transmitted
Incident + Reflected
(a)
1.0 1.0
Reflected
0.8 0.8 Uniform
Normalized power
Reflected power
0.6 Tapered
0.6
0.4 0.4
02 0.2
Transmitted
0 0
0 2 4 6 8 10 12 14 16 18 20 0 90 0.95 1.00 1.05 1.10 1 15
Number of periods ZZ
(b) (c)
Fig. 7.43 (a) Magnetoinductive waveguide Bragg grating. (b) Variation of |R|2 and |T |2 at ω = ω0 , with the number of periods
for a Bragg grating with κ = 0.2 and µ1 = 1.1, µ2 = 0.9. (c) Variation of |R|2 with ω/ω0 for a similar grating containing
20 periods (full line) and corresponding result for grating with raised cosine taper (dashed line). From Syms et al. (2006a).
Copyright c 2006 IEE
element 0. They can be obtained from eqns (7.63)–(7.65) and the an-
alytical solution may be found in Syms et al. (2006a). Choosing again
the coupling coefficient between nearest neighbours in the uniform lines,
κ = 0.2 and µ1 = M1 /M = µ2 = M2 /M equal to 0.2 and 0.4 the nor-
malized transmitted power is plotted in Fig. 7.42(b). Clearly, there is a
transmission peak around the resonant frequency.
If we insert additional resonant loops between elements 0 and 1 in
Fig. 7.42(a) then we have additional transmission peaks. The transmit-
ted power as a function of frequency for a two-loop cavity is shown in
Fig. 7.42(c) for the same set of parameters. There are now two narrow
transmission peaks.
Following further the optical analogy we should be able to have large
reflection in a certain frequency band if a large number of small re-
flections add coherently. These are called Bragg reflectors in optics. We
can achieve a large number of small reflections if the mutual inductances
undergo a small but periodic variation along the line (see Fig. 7.43(a),
where only one period is shown). The period is Λ = 2a. How many peri-
2
Zm –1 N+2
Port 3 (I3 , n) Zm Zm
Z Port 2 (I’n )
1 Z N+1 Z
Zm 0
Zm Zm
Z Z +1 Z
Transmitted N
Port 1 (I1 , n) R2 Zm Zm T2
Z m3
–2 –1 0 Z Zm Z Zm Z
Zm Z Zm Z Z m1 Z Z’m Z’m Z’m
Z m2 1 R1 Z Zm Z Zm Z
Transmitted Zm Zm T1
Incident + Reflected Z
2 Z Z
Zm Zm Zm
Port 2 (I2 , n) Z Port 1 (I n) Z
Z
Zm Zm Zm
(a) (b)
Fig. 7.44 (a) Magnetoinductive waveguide 3-port power splitter. (b) Magnetoinductive waveguide directional coupler. From
c 2006 IEE
Syms et al. (2006a). Copyright
ods do we need for large reflection? The answer is given in Fig. 7.43(b),
where the dependence of the reflection and transmission coefficients is
plotted as a function of the number of periods for µ1 = 1.1 and µ2 = 0.9.
It may be seen that to reach saturation (that is 100% reflection) about
20 periods are needed. Next, we calculate the reflected power as a func-
tion of frequency for the same 20 periods and the same parameters. It
may be seen in Fig. 7.43(c) that close to total reflection occurs within a
narrow band centred on the resonant frequency of the element.
Have we seen similar things before? If we look at the alternating values
of mutual inductance in Fig. 7.43(a) it should remind us of the biperiodic
lines of Figs. 7.17(a)–(d). In Section 7.10 we looked at this problem and
came to the conclusion (an infinite line was assumed there) that for
a biperiodic line there is a stop band in the middle of the dispersion
characteristics. In the present section we have a finite biperiodic line
but the conclusion is the same. Instead of saying that there is a stop
band we say now that we have nearly perfect reflection within a certain
band due to Bragg reflection—and that’s the same thing.
Another useful device may be seen in Fig. 7.44(a). It is shown in Syms
et al. (2006a) that by judicious choice of M1 , M2 and M3 any desired
power ratio between the two output lines can be achieved without any
reflection in the input line.
Our final example is a four-port device shown schematically in Fig.
7.44(b). The free parameters are M ′ and N (note that Fig. 7.44(b) shows
only one extra element in the coupling region), the mutual inductance
and the number of elements in the coupling region. The aim is to direct
desired powers to the two outputs. The analysis may be found in Syms
et al. (2006a).
1’ 2’ 3’ 4’
1 2 3 4
(a)
a receiver
h
... -3 -2 -1 0 1 2 3 ...
transmitter
(b)
Fig. 7.45 (a) Near-field pixel-to-pixel a receiver

imaging scheme. Schematic repre- h/4
sentation of (b) the single-layer and h/2
(c) the double-layer magnetoinductive h/4
lens. From Sydoruk et al. (2007b). transmitter
of Physics (c)
7.15.3 Imaging
We shall discuss here a particular type of imaging under conditions when
all the dimensions are small relative to the wavelength. It is a pixel-by-
pixel imaging that simply translates the object. An example is shown in
Fig. 7.45(a) where the object, consisting of points 1, 2, 3, 4 is translated
along the dotted lines to 1′ , 2′ , 3′ , 4′ , a distance h away. In order to
simplify the problem we shall look at one object point only and represent
it by a small non-resonant transmitter coil at the point x = 0, y = 0. The
imaging is tested by moving a small, non-resonant receiver coil along the
line y = h. If the received power has a narrow maximum in the vicinity
of the point x = 0, y = h then we can regard it as an image. However,
if all we have is a small transmitter coil and a small receiver coil then
the maximum along the y = h line will be wide, corresponding to the
field distribution of the small transmitter coil. How could we make it
sharper? Let’s insert a MI waveguide between the transmitter and the
receiver, as shown in Fig. 7.45(b). We may now claim that the field at
the point x = 0, y = h will be higher due to the coupling of element
0 both to the transmitter and to the receiver. This claim, however, is
not necessarily correct. Inserting the MI waveguide will not, in general,
make the field opposite the transmitter more concentrated because there
will be a MI wave propagating in both directions away from element zero
spreading the power in the x direction.
The remedy is to have a MI wave in the y direction but suppress
it in the x direction. How can we do this? We have actually done
so in Section 7.13. When we have two coupled lines and the elements
are above each other then there is no power transfer along the coupled
lines, provided the coupling is high enough. This conclusion can be
0
60 60
í

Frequency f [MHz]
Frequency f [MHz]
í

í
40 40
í

30 30 í
í í
Normalized distance x/a Normalized distance x/a
(a) (b)
Fig. 7.46 Near-field imaging for the double lens with h = 10 mm. Magnetic-field distribution in the image plane versus
c 2007 American Institute
frequency (contour plot). Experiment (a) and theory (b). From Sydoruk et al. (2007b). Copyright
of Physics. For coloured version see plate section
drawn from Fig. 7.36(b), which shows that there is a stop band at ω0
for h = 10 mm, and also from Fig. 7.39(a), showing no power transfer
for the unshifted case. Thus, power can be transferred in the y direction
but now it cannot spread in the x direction.
The relevant experiment was done (Sydoruk et al., 2007b) with two
coupled lines consisting of split-pipes and separated from each other by
h/2. The elements are arranged as in Fig. 7.38 but the measurement is
done now by placing the transmitter below the element in the middle
at a distance h/4, while the receiver moves above the upper line at a
distance h/4, as shown schematically in Fig. 7.45(c). The experimental
and theoretical results are shown in Figs. 7.46(a) and (b) for a distance
of h = 20 mm. The horizontal axis shows the displacement of the re-
ceiver, whereas the frequency is on the vertical axis. The measured field
strength is colour-coded. There is remarkable agreement between theory
and experiment, both showing that an image exists in the vicinity of the
resonant frequency at which sideways propagation of MI waves is pro-
hibited. The image is translated in the present case by 2h. Translation
further away can be realized by inserting further lines. The essential
criterion is that sideways propagation of MI waves must be prohibited.
7.15.4 Detection of nuclear magnetic resonance

We shall discuss here the application of MI waves to the detection of
nuclear magnetic resonance. It needs to be remembered at this stage
that there are three distinct resonance phenomena at play: the resonance
of the element at ω0 , the rotational resonances occurring in the ring
structure and the nuclear magnetic resonances of the various nuclei. The
frequency of nuclear magnetic resonance is known to be proportional to

z [cm]

rotating
dipole
í
í

x [cm]
Fig. 7.47 Bilayered rotational res-
onator
y [cm] í í
the applied dc magnetic field. For a magnetic field of 1 T the resonant

frequency of protons (the one most often examined in medical imaging)
is 41 MHz, which is just in the range of frequencies at which MI waves
can easily propagate and for which most of the experiments reported in
this chapter have been done.
Under resonance conditions the nuclear dipoles precess, producing a
rotating magnetic field. The simplest detecting mechanism consists of
a single coil in which the rotating magnetic field induces a voltage. In
order to have a more sensitive detecting mechanism the best chance is to
create synchronism between the rotating magnetic field and the MI wave
propagating round the ring structure of Fig. 7.40. This occurs when the
phase velocity of the rotating field at a radius R is equal to the phase
velocity of the MI wave at rotational resonance. It is easy to show that
for the synchronism condition to be satisfied the frequency of rotational
resonance must agree with the nuclear magnetic resonance.
Some aspects of this detecting mechanism were analyzed in Solymar
et al. (2006). The power extracted from a single element was compared
with that that can be extracted from an N -element synchronous ring by
an optimized load impedance and by optimal matching. It was shown
that in the latter case N times the power from a single ring can be
obtained.
A further advantage of the synchronous detection scheme is that the
circulating MI wave can be amplified by parametric amplification at the
expense of a pump wave. In order to do so a second ring is needed (see
Fig. 7.47) to couple magnetically to the first one, furthermore, the dis-
persion characteristics need to be tailored so that both the fundamental
and the second temporal harmonic can propagate and, of course, a non-
linear element like a varactor diode is needed to couple the fundamental
frequency to the pump wave.
For a practical realization see Syms et al. (2008) where experiments
with a three-frequency parametric amplifier are shown.
Magnetoinductive waves II
8
8.1 MI waves in two dimensions 8.1 MI waves
in two dimensions 251
8.1.1 Introduction 8.2 MI waves retarded 266
8.3 Non-linear effects in
A one-dimensional treatment was sufficient to present most of the prop- magnetoinductive waves 277
erties of MI waves: how they propagate, how they attenuate, how they
are reflected, how to find a matching impedance, what is the power
density and how they are coupled to each other. However, for some
other properties, e.g. refraction, a two-dimensional theory is needed. In
the theory of metamaterials refraction occupies a central position. One
might even say that the subject started with the discovery of Veselago’s
paper (1968) by Smith et al. (2000). The new physical concept intro-
duced, as mentioned several times in this book, was negative refractive
index and negative refraction. We may now ask the question whether
MI waves can exhibit negative refraction? For that we need to have a
formulation of MI wave propagation in a 2D medium and we have to
find out what happens at the boundary of two periodic materials both
of them capable of propagating MI waves.
Since metamaterial elements often have circular shapes, one of the
popular 2D structures is a hexagonal one. We shall investigate their
spatial resonances and imaging properties that were the subject of ex-
perimental work by Wiltshire et al. 2003a; Wiltshire et al. 2004a.
The theoretical treatment does not differ much from the one-dimen-
sional one. The physics has not changed. The MI wave propagates in 1 2 3 4
the same way due to the coupling between the elements. The nearest-
neighbour approximation may still be used but the number of nearest
neighbours is now four for the square configuration and six for the hexag- (a)
onal one. The dispersion equations will look different and the frequency
against wave-vector diagrams must also be presented in a different man- 1 2
ner. However, if we consider the treatment in terms of currents, applied
3 4
voltages and the mutual impedance matrix then there is no qualitative
difference. (b)
In eqn (7.16) the current flowing in the nth element is related to the
voltage excitation in all the other elements. The relationship is via the Fig. 8.1 Four capacitively loaded loops
inverse mutual impedance matrix arranged (a) in a line, (b) in two rows
I = Z−1 V. (8.1)
The impedance matrix as presented in eqn (7.15) is a tri-diagonal
matrix because the treatment there was restricted to nearest-neighbour
252 Magnetoinductive waves II
0.9
0.9
2

08
1.
k y a/S
1

1
1.08
1
y a í
2
Fig. 8.2 Capacitively loaded loops in 2
0.9
1
a .9
the planar configuration (a) geome- í 0
try, (b) normalized dispersion (isofre- x í í
quency) curves for a = 2.25r0 . From k xa/S
c 2005
Syms et al. (2005b). Copyright
(a) (b)
EDP Sciences
interaction in one dimension, but in the general case, when we consider

every element coupled to every other element, the two-dimensional case
looks just the same. This is illustrated in a simple example. Two sets
of four elements are shown in Fig. 8.1, the first one is a one-dimensional
array, the second one is a two-dimensional one. The impedance matrix
for both of them may be written in the form
 
Z11 Z12 Z13 Z14
 Z21 Z22 Z23 Z24 
 Z31 Z32 Z33 Z34  . (8.2)
 
Z41 Z42 Z43 Z44

Of course the values of the mutual inductance between element 1 and 4
for example would be quite different for the two configurations but the
mathematical formulation would be identical.
8.1.2 Dispersion equation, group velocity, power

density
It is rarely necessary to consider interaction between all the particles.
For most purposes nearest-neighbour interaction is sufficient. Kirch-
hoff’s law for element (n, m) in a square lattice may be written as
Z(ω)In,m +j ωMx (In+1,m +In−1,m )+j ωMy (In,m+1 +In,m−1 ) = 0, (8.3)
where In,m is the current in the element located at the nth row and
mth column, Mx and My are the mutual inductances in the x (horizon-
tal) and y (vertical) directions, respectively, and, as before, Z(ω) is the
self-impedance of the elements. For the planar configuration (see Fig.
8.2(a)), Mx = My and they are both negative. For the planar–axial case
shown in Fig. 8.3(a) Mx is of course still negative, but My is positive
and under the present conditions (element spacing being the same in
both directions) its value is smaller than |Mx |.
Assuming the current in the form

1.05
9
1.0
1.0 1.01
1.01 5

0.97
k y a/S
.93 0.97
0 0.93
1.01 1.01
y a í
0.97 7
1.05 0.9
Fig. 8.3 Capacitively loaded loops
a 1.01 1.05 1.01
í in the planar–axial configuration (a)
x í í geometry, (b) normalized dispersion
k xa/S (isofrequency) curves for a = 2.25r0 .
From Syms et al. (2005b). Copyright
(a) (b) c 2005 EDP Sciences
In,m = I0 e −j (nkx a + mky a) , (8.4)

where kx and ky are the x and y components of the wave vector, k, and
I0 is a constant, and substituting eqn (8.4) into eqn (8.3) we obtain the
dispersion equation in the form
ω
= A−1/2 ; A = 1 + κx cos(kx a) + κy cos(ky a), (8.5)
ω0
where κx and κy are the coupling coefficients defined as κx,y = 2Mx,y /L.
For our purpose the most useful presentation of the equation is in the
form of the ω/ω0 = constant curves where ω0 = (LC)−1/2 in Fig. 8.2(b)
for the planar configuration (κx = κy = −0.106) and in Fig. 8.3(b) for
the planar–axial configuration (κx = −0.106, κy = 0.066). In both cases
the separation of the elements is 2.25r0 , where r0 is the external radius
of the SRR. When kx a, ky a ≪ 1 the curves in Fig. 8.2(b) may be shown
to be circles, whereas those in Fig. 8.3(b) are hyperbolae.
In the 2D case the group velocity is given by the gradient of the ω
versus k curves plotted in Figs. 8.2(b) and 8.3(b). Mathematically, they
may be obtained from eqn (8.5) as
aω0 −3/2
vg = A [κx sin(kx a)ix + κy sin(ky a)iy ] , (8.6)
2
where ix and iy are unit vectors in the x and y directions, respectively.
For the planar case, when κx = κy and the arguments of both sine
functions are small, the group velocity may be seen to be in a direction
opposite to the phase velocity—a clear case of a backward wave. The
relationship is more complicated for the planar–axial configuration, as
will be discussed in the next section.
The direction of power flow is given by the group velocity and we may
again obtain the power density by multiplying the group velocity by the
stored energy per unit surface
1
S= vg Es . (8.7)
2
The energy stored in any of the elements is given by the sum of the en-
ergies in the inductance, the capacitance and in the mutual inductances
relating to nearest neighbours. In terms of a single element (m, n) it is
given by
|In,m |2

1 2 ∗ ∗

Es = 2 L|In,m | + ωC
+ Mx In,m In−1,m + In+1,m
2a
∗ ∗

+ My In,m In,m−1 + In,m+1 . (8.8)
With our wave assumption for the current (eqn (8.4)), and using the
expression of the group velocity given by eqn (8.6), the power density
may be written as follows
1
S= ω|I0 |2 [Mx sin(kx a)ix + My sin(ky a)iy ] , (8.9)
2
i.e. it is independent of the circuit parameters L and C. The condition
for ω to be in the pass band must of course be satisfied.
8.1.3 Reflection and refraction

We shall now look at reflection and refraction of a MI wave at the bound-
ary of two different media. Reflection and transmission at perpendicular
incidence has already been considered in Section 7.9. We shall now as-
sume that the MI wave is incident at an angle, so we shall be able to
determine refraction as well.
Let us assume two semi-infinite 2D media lying on each side of the
x = 0 line, which can support MI waves and have the same regular
rectangular lattice with a lattice constant a. We shall consider here three
examples. In the first case the elements in both media are in the planar
configuration, medium 1 having the same parameters as in Fig. 8.2(b),
whereas medium 2 differs from medium 1 by having a slightly different
capacitance and, consequently, a slightly different resonant frequency
that we shall take as ω02 = 1.03 ω01, see Fig. 8.4. Assuming further that
kx a and ky a are small relative to unity the constant frequency curves
are circles with good approximation. They are shown in Fig. 8.5(a) for
ω/ω01 = 1.11.
It needs to be noted that we have here a rather unusual situation.
Normally, positive refraction is due to forward waves propagating in
both media. In the present case both media support backward waves.
This does not lead to any complications but it means that if we want
a wave incident at a positive angle (i.e. the group velocity to be in the
first quarter) we have to choose kx1 and ky1 in medium 1, to be in the
third quarter.
The boundary condition to be satisfied is that the phase velocities
along the boundary must be the same on both sides (see discussion and
the Ewald circle construction in Fig. 1.4) hence ky2 = ky1 and kx2 is
given by the construction shown in Fig. 8.5(a). The corresponding group
medium 1 medium 2
m+1
m-1
Fig. 8.4 Two media consisting of ca-
pacitively loaded loops in the planar
configuration having different resonant
n=0 n=1 frequencies

medium 1 medium 2 medium 1 medium 2
k y a/S
k y a/S

í í
í í

k xa/S k xa/S
(a) (b)
Fig. 8.5 Refraction at the boundary of two media. Planar configuration with ω02 /ω01 = 1.03. (a) The loci ω/ω01 = 1.11 and
corresponding phase velocity vectors; (b) directions for the group velocity vectors, assuming (kx1 a, ky1 a) = (−0.18, −0.12)π
and (kx2 a, ky2 a) = (−0.38, −0.12)π. From Syms et al. (2005b). Copyright c 2005 EDP Sciences
velocities, very closely opposite to the chosen wave vectors, are shown in
Fig. 8.5(b). It may be seen from Fig. 8.5(a) that the construction can be
performed for all possible values of kx1 , i.e. at that particular frequency
a refracted wave exists for any incident wave.
In our next example we shall take the same planar configuration on
both sides of the boundary but now the resonant frequency is assumed to
be smaller in medium 2. We take ω02 = 0.97ω01. The ratio of the radii
is then reversed as shown in Fig. 8.6(a). Using the same construction as
before we find that kx2 > kx1 and refraction is now pointing away from
the perpendicular, as may be seen in Fig. 8.6(b), where the correspond-
ing group velocities are shown at the boundary of the two media. It may
also be seen in Fig. 8.6(a) that no refraction is possible for a range of
incident angles. This is the case of total internal reflection.
In our third example we choose planar–axial configurations on both
sides with all the parameters being the same but the orientation in
medium 2 is at right angle to that in medium 1, as shown in Fig. 8.7(a).
Note that the lattice in medium 2 is shifted by one half of the lattice
constant relative to that in medium 1, which ensures increased magnetic

medium 1 medium 2
medium 1 medium 2
k y a/S
k y a/S

í í
í í

k xa/S k xa/S
(a) (b)
Fig. 8.6 Refraction at the boundary of two media. Planar configuration. ω02 /ω01 = 0.97. (a) the loci ω/ω01 = 1.08 and
corresponding phase velocity vectors; (b) directions for the group velocity vectors, assuming (kx1 a, ky1 a) = (−0.33, −0.19)π
and (kx2 a, ky2 a) = (−0.04, −0.12)π. From Syms et al. (2005b). Copyright c 2005 EDP Sciences
coupling between the two media.

Our aim is now the same as in the previous examples: we wish to
have an incident wave with group velocity in the first quarter. Let
us remember that the group velocity is the gradient vector of the ω =
constant curves. Hence, any ω = constant curve in the second quarter of
the dispersion diagram, shown in Fig. 8.3(b) would qualify. Let us choose
the ω/ω01 = 1.01 curve plotted in Fig. 8.7(b) with small arrows showing
the direction of the group velocity. The relevant dispersion curve for
the same value of ω/ω01 in medium 2 is rotated by 90 degrees and is
therefore in the first quarter, as shown also in Fig. 8.7(b). It may be seen
that there is only a very limited range of incident wave vectors for which
refraction exists. The effect could be used for switching, modulation
or spatial filtering, for example. Choosing a value of ky a = 0.4π, and
adhering again to the rule that the phase velocities must agree across the
boundary, the group velocities in the two media are shown in Fig. 8.7(c).
There is negative refraction. Note that in the present case the angle
between phase and group velocities is slightly less than 90 degrees in both
media, so the waves on both sides would qualify as forward waves. In
fact, we could make the angles between phase and group velocity further
decrease by choosing ω/ω0 = 1 or increase the angle (and going thereby
into the region where the waves on both sides are backward waves) by
choosing ω/ω0 = 1.01. This is similar to the conclusions reached by Luo
et al. (2002b) in the sense that negative refraction may occur without
the presence of backward waves but in our case the refraction angle is a
strongly varying function of the incident angle.
We could also change the angle of negative refraction in medium 2
by leaving medium 1 unchanged and rotating the orientation of the
loops in medium 2, relative to medium 1, by less than 90 degrees. In
addition, we could considerably influence the dispersion characteristics
of MI waves in medium 2 by changing the resonant frequency and the
medium 1 medium 2
m+1
m+1/2
m
m-1/2
m-1
n=0 n=1
x
(a)
1 1
medium 1 medium 2 medium 1 medium 2
k y a/S
k y a/S
0.5 0.5
0 0
í -0.5 0 0 0.5 1
k xa/S k xa/S
(b) (c)
Fig. 8.7 Refraction at the boundary of two media. Planar–axial configuration with different orientations of the loops in
medium 1 and medium 2. (a) The geometry of the loops; (b) the loci ω/ω01 = 1.01 and directions for group velocity vectors;
c 2005
(c) directions for the group velocity vectors for (kx1 a, ky1 a) = (−0.39, −0.42)π. From Syms et al. (2005b). Copyright
EDP Sciences
coupling coefficients. In fact, any of the four combinations may lead to

negative refraction: (i) both waves forward, (ii) both waves backward,
(iii) incident wave in medium 1 forward, refracted wave in medium 2
backward and (iv) incident wave in medium 1 backward, refracted wave
in medium 2 forward.
Most efforts in the literature to find negative refraction have been
aimed at electromagnetic waves incident from free space upon a periodic
medium. In our case both media are periodic. The MI wave, quite
obviously, could not be incident from free space because it can exist
only in certain periodic media. It needs to be noted that we have a
large amount of freedom choosing the dispersion characteristics in both
media, thereby allowing a large variety of refractive angles, positive or
negative, to be realized.
Next, we shall derive the reflection and transmission coefficients for
element number n y
Fig. 8.8 Contour plot of currents in a −5

2D array of resonators showing reflec- −25 −20 −15 −10 −5 0 5 10 15 20 25
tion of a magnetoinductive wave. For
coloured version see plate section element number n x

element number n y

í í í
í í í

element number nx element number nx element number nx
(a) (b) (c)
Fig. 8.9 Contour plot of currents in a 2D array of resonators showing diffraction on a defect. For coloured version see plate
section
the wave incident at an interface between two media. The technique is

the same as in Section 7.9. The current in medium 1 is written as the sum
of the incident and reflected waves, and in medium 2 as a transmitted
wave. We can then demand that the currents across the boundary satisfy
the recursion equations as in Section 7.9. The equations turn out to be
the same as eqns (7.35) to (7.39), we just need to replace k by kx and
the mutual inductances by their values in the x direction perpendicular
to the boundary.
8.1.4 Excitation by a point source: reflection and

diffraction
Up to now we have assumed that MI waves propagate in a 2D medium
with plane wavefronts. We shall now look at the propagation of MI
waves when one of the elements is excited by a temporally varying mag-
netic field. We wish to see how the MI wave originating in a single point
is reflected by a boundary and diffracted by an obstacle. In other words
we wish to show that MI waves are no different from other kinds of
waves physicists love and cherish. The aim is to find the current distri-
bution. The solution is provided by the matrix equation (8.1) in which
the voltage vector has one non-zero element.
Our first example is a 2D sheet of 15 × 60 elements in a rectangular
lattice. We assume that the element in the lower left-hand corner is
excited. Having 900 elements necessitates the inversion of a 900 by

900 matrix that is well within the capabilities of the relevant Matlab
program. The resultant distribution of the currents is shown in Fig. 8.8 y
by coloured contour plots. The excitation and reflection of the wave may
be clearly seen.
In our second example a square sheet of 31 × 31 elements is considered
that is excited at the centre. The regularity of the square lattice is broken
by a missing element. Figures 8.9(a)–(c) show the wave pattern of the x
MI wave for three different positions of the missing element. The MI (a)
wave is clearly diffracted.
b2
8.1.5 Spatial resonances in hexagonal lattices a2
Next, we shall look at the propagation of MI waves in hexagonal lattices a1
and in particular at spatial resonances that were experimentally investi-
gated by Wiltshire et al. (2004a). The hexagonal lattice and the corre- b1
sponding co-ordinate system is shown in Figs. 8.10(a) and (b). Consid-
ering only nearest-neighbour interactions Kirchhoff’s law for the voltage
in element (n, m) may be written as (b)
ZIn,m + j ωM (In,m−1 + In,m+1 + In−1,m Fig. 8.10 (a) Lattice of resonant el-
ements with hexagonal arrangement.
+In+1,m + In−1,m+1 + In+1,m−1 ) = 0, (8.10) (b) (a1 , a2 ) direct vectors of the hexag-
onal lattice and (b1 , b2 ) reciprocal vec-
where Im,n is the current in the (m, n) element, Z is the self-impedance tors
assumed again as lossless, M is the mutual inductance between nearest
neighbours. The position of an element is given by the radius vector
rn,m = na1 + ma2 , (8.11)

where a1 and a2 are the direct vectors of the lattice, shown in Fig.
8.10(b). In the x, y co-ordinate system they can be written as
a1 = ix a, a2 = a(ix cos 60◦ + iy sin 60◦ ), (8.12)

where a is the spacing between the elements and ix and iy are unit vectors
in the x and y directions, respectively. The corresponding vectors of the
reciprocal lattice, b1 and b2 may be obtained by requiring the following
conditions be satisfied
a1 · b1 = 1, a1 · b2 = 0,
(8.13)
a2 · b1 = 0, a2 · b2 = 1.
We shall now look for the solution of eqn (8.10) in the form
In,m = I0,0 e −j k · rn,m . (8.14)

I0,0 is a constant and k is the wave vector that may be expressed in
terms of the reciprocal lattice vector as
k = 2 π(f1 b1 + f2 b2 ), (8.15)
k y a/S
Fig. 8.11 2D normalized dispersion

(isofrequency) curves for a hexago-
nal lattice. The boundaries of the
first Brillouin zone are shown by bold
lines. From Zhuromskyy et al. (2005a).
Copyright c 2005 Optical Society of
America k x a/S
where f1 and f2 are constants.

Substituting eqn (8.14) into eqn (8.10) we obtain the dispersion equa-
tion in the form
ω 1
=p , (8.16)
ω0 1 + κ[cos(2 πf1 ) + cos(2 πf2 ) + cos(2 πf1 − 2 πf2 )]
√
where ω0 = 1/ LC is the resonant frequency of the element and κ =
2M/L is the coupling coefficient. The ω/ω0 = constant curves as func-
tions of the wave vector are plotted in Fig. 8.11. The pass band is found
to extend from 0.93ω0 to 1.21ω0. Note that the waves are backward
waves with phase and group velocities in opposite directions.
In the examples to follow in the presence of an excitation we shall use
eqn (8.1) to determine the current distribution. First, however, we shall
attempt to find an approximate solution that will lead us to a mathemat-
ically familiar territory and offer a clear physical picture. We may expect
that the propagation of MI waves, similarly to the propagation of most
other waves, can be mathematically described by a second-order partial
differential equation. We shall therefore convert our difference equation
(eqn (8.10)) into a differential equation. We can do that when the wave
vectors are sufficiently small or in other words when the wavelength of
the MI wave is much larger than the element spacing. Consequently
we shall introduce the ν, µ co-ordinate system in the directions a1 and
a2 with the continuous variables ν = na and µ = ma, and replace the
discrete function In,m by the continuous function I(ν, µ). A change in
the subscript by unity would then be equivalent to a change of the con-
tinuous variable by a, which is then regarded as an elementary change.
To convert all the terms in eqn (8.10) into continuous variables we need
to expand the current into a Taylor series as follows

∂ ∂
I(ν + ∆ν, µ + ∆µ) = 1 + ∆ν + ∆µ
∂ν ∂µ
∂2 ∂2

1
+ (∆ν)2 2 + 2∆ν∆µ
2 ∂ν ∂ν∂µ
2
∂
+(∆µ)2 2 I(ν, µ), (8.17)
∂ν
where ∆ν and ∆µ are small deviations from ν and µ. With the aid of
eqn (8.17) we may now convert eqn (8.10) into the differential equation
∂2I ∂2I ∂2I

1 Z
− + + 6 + I = 0. (8.18)
∂ν 2 ∂ν∂µ ∂ν 2 2d2 jωM
A further transformation to the x, y co-ordinate system yields the rela-
tionship
∂2 ∂2 ∂2 3 ∂2 ∂2

− + = + . (8.19)
∂ν 2 ∂ν∂µ ∂ν 2 4 ∂x2 ∂y 2
So, we end up with the familiar wave equation
∂2I ∂2I
2 + + k 2 I = 0, (8.20)
∂x ∂y 2
where
ω2

2 4 1
k = 2 1+ 1− 2 , (8.21)
a 3κ ω0
and k = |k|. It may be easily shown that the dispersion equation (8.16)
reduces to eqn (8.21) when ka ≪ 1.
A clear advantage of having the wave equation is that, at least for
certain geometries, we know the solutions from past experience. For
rectangular boundaries, possible solutions are
I = I0 sin(kx x) sin(ky y), (8.22)

where I0 is a constant and kx and ky are the x and y components of the
wave vector. For a circular boundary there are circularly symmetrical
solutions of the form
I = I0 J0 (kr), (8.23)
where J0 is the zero-order Bessel function of the first kind and r is the
distance from the centre of the circular structure of the elements.
If we choose a frequency and know which elements are excited (our
formulation allows all of them to be separately excited) we can determine
the current distribution from eqn (8.1). The result might be in the form
of odd-looking current distributions because several wave vectors may
coexist at a particular frequency. It is only at the spatial resonance
that no more than one wave vector survives. Spatial resonances are
characteristic to all wave phenomena. They exhibit the same behaviour
whether they occur in vibrating membranes, organ pipes or Fabry–Perot
resonators. The boundary condition to be satisfied in the present case
is that the current must vanish at the boundary.
0
20
y [cm]
0 20
20
-40
20 0 20 20 0 20 20 0 20
x [cm] x [cm] x [cm]
(a) (b) (c)
Fig. 8.12 Numerically obtained current distributions for circular boundary conditions at ω/ω0 = (a) 1.207, (b) 1.192, (c) 1.167.
Black dots show positions of the elements. From Zhuromskyy et al. (2005a). Copyright c 2005 Optical Society of America.
For coloured version see plate section
Array of circular shape. For a circle of radius R the condition for

spatial resonance is
kR = ρi , (8.24)
where ρi is the ith root of J0 .
We shall now look at the phenomenon of spatial resonance with the
aid of a few examples. At this stage we need to commit ourselves as
to the distance between the elements and the resonant frequency. The
loop radius, the wire diameter and the distance between the elements
are taken as 10 mm, 2 mm and 22.5 mm, and the resonant frequency
as 21.5 MHz. The inductance of the loop may then be determined from
standard formulae (Grover, 1981) that give L = 33 nH. The correspond-
ing capacitance can be determined from the resonant frequency as 1.66
pF. The mutual inductance between two neighbouring elements is 1.75
nH, yielding κ = −0.106. Owing to the hexagonal arrangement it is not
possible for all the elements to lie exactly on a circular boundary. With
our choice of 361 elements the deviation from the circular boundary is
quite small, as may be appreciated by looking at Figs. 8.12(a)–(c).
Knowing the geometry the frequencies of the first 3 spatial resonances
may be determined from eqn (8.24) as ω/ω0 = 1.207, 1.187, 1.155. Nu-
merical calculations based on the known values of the mutual impedance
matrix yield 1.207, 1.191, 1.165, a very good approximation. The ana-
lytical results are of interest because they give good approximation for
low values of k and, of course, they give an immediate idea of what
the current distribution looks like. For the general case, however, it is
more accurate to rely on the exact solution of the discrete problem that
is based on the inversion of the impedance matrix. Note that for the
numerical determination of the current distribution we need an excita-
tion. We assume that the central element out of the 361 is excited and
then proceed with the numerical solution. The numerically determined
current distributions are shown by a colour code in Figs. 8.12(a)–(c).
Normalization in each figure is to the maximum value within the figure.
0
(a) (b)
40 40
y [cm]
y [cm] -20
20 20
-40
0 20 40 0 20 40
x [cm] x [cm]
0
(c) (d)
40 40
y [cm]
y [cm]
-20
20 20
-40
0 20 40 0 20 40
x [cm] x [cm]
Fig. 8.13 Numerically obtained current distributions for rectangular boundary conditions, (a) non-resonant distribution for
asymmetric excitation, (b)–(d) resonant excitation at ω/ω0 = 1.207, 1.202 and 1.201, respectively. Black dots show positions
of the elements. From Zhuromskyy et al. (2005a). Copyright c 2005 Optical Society of America. For coloured version see
plate section
As may be expected, at the first spatial resonance there is zero current

only at the boundary. For the second and third spatial resonances the
currents are approximately zero at one and two radii, respectively. An
interesting feature of Fig. 8.12(c) is that in spite of the close-to-circular
boundary the hexagonal nature of the element geometry re-establishes
itself further away from the centre.
Array of rectangular shape. We shall assume the array to extend from
0 to Dx in the x direction and from 0 to Dy in the x direction. In view
of eqn (8.22) the wave vector components leading to spatial resonances
are given by the relationships
kx Dx = px π and ky Dy = py π, (8.25)
where px and py are integers.
We shall arrange now the hexagonal lattice in a 19 × 19 square geom-
etry. As in the previous example we shall compare the analytical and
numerical values for the frequencies of 3 spatial resonances (fundamental
plus two of the second order) shown in Figs. 8.13(b)–(d). The resonant
frequencies are ω/ω0 = 1.207, 1.202 and 1.201 from the numerical solu-
tions. There is some ambiguity in the analytical expression: due to the
jagged boundaries Dx cannot be exactly defined. The values we obtain
for the resonant frequencies from the analytical solution are 1.207, 1.202
and 1.202. The last two figures agree because in theory the two current
0
20 (a) (e) 20
y [cm]
0 0 -20
20 20
-40
20 0 20 20 0 20
0
20 (b) (f) 20
y [cm]
0 0 -20
20 20
-40
20 0 20 20 0 20
0
20 (c) (g) 20
y [cm]
0 0 -20
20 20
20 0 20 20 0 20 -40
0
Fig. 8.14 Magnetic-field distribution 20 (d) (h) 20
at four resonant frequencies: (a) and
(e) ω/ω0 = 1.206, (b) and (f) ω/ω0 =
y [cm]
1.186, (c) and (g) ω/ω0 = 1.154, (d) 0 0 -20

and (h) ω/ω0 = 1.114; normal com-
ponent (a)–(d) and tangential compo-
nent (e)–(h). From Zhuromskyy et al. 20 20
(2005a). Copyright c 2005 Optical So- -40
ciety of America. For coloured version 20 0 20 20 0 20
see plate section x [cm] x [cm]
distributions are identical, only rotated by 90 degrees. We may claim

again good agreement between the analytical and numerical results.
Note that the establishment of the spatial resonances depends little
on the exact position of the excitation as long as it is not at an expected
current minimum. For the present rectangular case the excitation is
taken one quarter of the way along one of the diagonals. We can obtain
with this excitation all the spatial resonances, but the current pattern
will of course depend on the position of the excitation if we are not at
a spatial resonance. At a frequency of ω/ω0 = 1.208 we do have an
asymmetric pattern characteristic to the excitation, as may be seen in
Fig. 8.13(a).
For hexagonal boundaries we have a chance to compare the theoreti-
cal results with a set of experimental ones measured by Wiltshire et al.
(2004a). The experiments were performed on a 2D array of 271 ‘swiss
rolls’. The array was centrally excited on one side of the structure and
0
10 (a) (b) 10
y [cm]
0 0
Fig. 8.15 The normal component of
magnetic field at ω/ω0 = (a) 0.98 and
10 10 (b) 1.01. From Zhuromskyy et al.
25 (2005a). Copyright c 2005 Optical So-
10 0 10 10 0 10 ciety of America. For coloured version
x [cm] x [cm] see plate section
the axial (perpendicular to the plane of the elements) and radial com-
ponents of the magnetic field were measured on the other side. The
measured 2D distribution of these two components of the magnetic field
are given in their Figs. 4–7 for the first four spatial resonances. Unfor-
tunately, we have not been able to get hold of these figures. The reader
might want to look at the original publication. Our theoretical results
plotted in Fig. 8.14 display practically the same spatial variation for
both components of the magnetic field as in the experiments.
8.1.6 Imaging
We have already discussed imaging with MI waves (Sydoruk et al.,
2007b), both experimentally and theoretically, in Section 7.15.3 where
the ‘lens’ consisted of two coupled 1D lines made up by split pipes. It
was a pixel-by-pixel imaging that could be best described as channelling
the spatial information across the imaging device. The channelling oc-
curred under conditions when the propagation of the MI wave was for-
bidden along the length of the coupled lines. Much earlier, Wiltshire
et al. (2003a) successfully imaged an object with the aid of a hexag-
onally arranged array of swiss rolls consisting of 271 elements. The
authors excited the 2D resonant structure by placing an M-shaped wire
antenna below the structure and measuring the axial component of the
magnetic field on the top. The experimentally obtained image is shown
in their Fig. 4. The image we have calculated from our simple model (in
which we consider capacitively loaded rings instead of swiss rolls) may
be seen in Figs. 8.15(a) and (b) for ω/ω0 = 0.98 and 1.01, respectively.
The agreement with the experimental results of Wiltshire et al. (2003a)
is good. However, as the frequency increases beyond the resonant fre-
quency of the element the image quickly deteriorates. At ω/ω0 = 1.01
the object is unrecognizable.
Imaging with MI waves using two parallel planes of broadside-coupled
SRRs was carried out by Freire and Marques (2005). Their experimen-
tal setup is shown in Fig. 8.16(a). The two planes have areas of 7 × 7
cm2 . They are separated from each other by a foam slab of thickness
d = 4 mm. The substrate thickness is h = 0.254 mm and the dielec-
tric permittivity is εr = 10. The transmitting and receiving antennas
are square loops of 1 cm2 . The field strength was measured by moving
Fig. 8.16 (a) Experimental setup for imaging with magnetoinductive waves. (b) The magnitude of the transmission coefficient
between the input and output antennas at a frequency of 3.23 GHz. From Freire and Marques (2005). Copyright c 2005
American Institute of Physics. For coloured version see plate section
the receiving antenna in the image plane. The experimental results are
given in Fig. 8.16(b). A focal region may be clearly seen where the field
strength is higher than in the surrounding region. The question whether
it is a proper focus is discussed by Mesa et al. (2005). A proper focus is
one that is measured by an ideal receiving device that does not interact
with the elements of the lens. The authors conclude that the focus mea-
sured by Freire and Marques (2005) was due to that kind of interaction
1
A generalization of the principle was i.e. not a proper focus.1 Further imaging work using MI waves in similar
presented by Sydoruk et al. (2007d). configuration was done later by Freire and Marques (2006), using this
The authors showed that such ‘im-
proper’ focus can be placed in any de-
time low-impedance receivers in order to avoid the interaction. They
sired position for any transmitter–lens showed both experimentally and theoretically (theory based on that of
configuration by appropriate matching Maslovski et al. (2004)) that good imaging can be achieved. A further
of the receiver. generalization of the problem was done by Marques and Freire (2005).
They showed theoretically that subdiffraction imaging devices cannot
produce focusing of power into three-dimensional spots of subdiffraction
size.
8.2 MI waves retarded

8.2.1 Introduction
All the MI waves investigated so far could be termed low-frequency waves
or perhaps quasi-static waves. In either case, one envisages waves that
propagate on a line, which may attenuate due to ohmic losses but do
not lose power by radiation. The dispersion characteristics were deter-
mined entirely by disregarding radiation. If the dispersion equation is
to be found in the presence of radiation then all the assumptions made
so far need to be reassessed. The most often used approximation was
that of nearest-neighbour interaction. The tacit assumption is then that
the field strength declines so rapidly away from an element that it is
too small to matter at the position of a neighbour further away. In the

presence of radiation this assumption cannot be valid. But what about
our calculation of the magnetic field via the vector potential? Is that
valid? Let us recall how we did that. We used eqn (1.14) that cannot
cater for radiation effects, and we assumed a uniform current distribu-
tion around the loop. The magnetic field obtained in this way declined
rapidly away from the loop (in the approximation of a small loop it can
be regarded as a magnetic dipole and the field decays with the third
power of the distance). If the free-space wavelength is comparable with
some dimension of the MI waveguiding structure then we can no longer
argue in favour of these approximations.
Let us start with the weakest constraint. What if the total length of
the guiding structure is comparable with the wavelength? Can it hap-
pen? Not at the frequencies used for magnetic resonance imaging, one
of the candidates for the application of MI waves. For a large magnetic
field of 2 T the proton magnetic resonant frequency is still only about
100 MHz, corresponding to a wavelength of 3 m, which is still large
enough to ignore, unless the line is very long. However, if we consider
phenomena at microwaves with free-space wavelengths of less than 10
cm and element sizes of about 10 mm, then one feels radiation effects
might matter. Then, the vector potential can no longer be calculated
from the current as given by eqn (1.14). We need to take into account
that the effect of the current is retarded. To reach point r2 from r1 a
time t = (r2 − r1 )/c is required. The expression for the vector potential
becomes
µ0 e −j k0 |r2 − r1 |
Z
A(r2 , t) = J(r1 ) dτ , (8.26)
4π |r2 − r1 |
where the current flows in point r1 and the vector potential is calculated
in point r2 . Hence, the mutual inductance between two loops must be
calculated from the above equation, increasing considerably the numer-
ical work.
If the size of the loop is comparable with the wavelength (say the
diameter of the loop is equal to λ/4, that may very well happen in
the optical region) then there are further complications. The current
distribution can no longer be regarded as uniform and one needs to resort
to the integral equations of antenna theory (see, e.g., King 1969) to find
the current as a function of azimuthal angle. We shall certainly not go
that far. In the present section we shall make a compromise: assume
a uniform current in the element but include retardation in calculating
the field quantities. As a result, the fields far away from the element
will decay as the inverse of the distance, which means that for a long
array it is necessary to include the effect of radiation when calculating
the interaction between any two elements in the array. The concept of
mutual inductance may still be used but it has to be calculated in a
different manner and, due to the phase delay it turns out to be complex.
In Section 8.2.2 we shall derive the dispersion equation for an infi-
nitely long line that now includes retardation, and compare the result
with that obtained by the quasi-static approximation. In Section 8.2.3

we make an attempt to clarify the usefulness and limitations of the
dispersion-equation approach. In Section 8.2.4 we shall return to the
finite-line problem. Two excitations will be investigated for the full
range of frequencies: (i) the first element excited, a situation we have
examined before in the quasi-static approximation, and (ii) all elements
excited with the value of k prescribed. Conclusions will be drawn in
Section 8.2.5.
8.2.2 Dispersion equation

We shall now include retardation in the dispersion equation, but to
simplify the matter, and to have a complete analytical formulation, we
shall replace our loops by magnetic dipoles. We have done so before
in Section 2.9 for the axial configuration, where we relied on the con-
cept of polarizability to derive the dispersion equation in the form (see
eqn (2.66))
1
= IF , (8.27)
αm
where IF is the interaction function. We shall now do the derivation
for the planar case where the interaction between the elements is due
to the Hθ component given by eqn (1.87) where we need to take θ =
90◦ . The polarizability for the lossless case is still the same as given by
eqn (2.41) but now we need to include radiation damping (see eqn (2.43))
in the expression for polarizability, and of course the interaction function
must include the term declining with the inverse distance responsible for
radiation. Formally, the dispersion equation is the same as in Section 2.9.
It can be written in the form
∞
X
1 = αm f (n) cos(kna), (8.28)
n=1
where now
ω02 k3

1 L
= 2 2 1− 2 +j 0 , (8.29)
αm µ0 S ω 6 πµ0
and
k03

1 j 1
f (n) = − − e −j k0 na . (8.30)
2 πµ0 nk0 a (nk0 a)2 (nk0 a)3
Note that the dispersion equation as given by eqns (8.28)–(8.30) is com-

plex. It has a real part and an imaginary part. Unlikely as it seems, it
may be shown (first realized by Simovski et al. (2005) on analyzing an
infinite set of nanoparticles) that the imaginary parts cancel, provided
k > k0 . Having assumed no ohmic losses the equation being real im-
plies that there are no radiation losses either. Can one offer a physical
1.03
Ns = 10 Ns = 100 Ns = 1000
1.02
ZZ0
1.01
Fig. 8.17 Dispersion curves derived

1 from eqns (8.28)–(8.30) summed for
0 0.2 0 0.2 0 0.2 the first Ns terms. From Zhuromskyy
ka/S (2008)
explanation? One may argue that if the line is infinitely long it cannot
radiate for the reason that there is no space left into which it could radi-
ate. However, this argument does not apply to the part of the dispersion
characteristics to the left of the light line. The imaginary part does not
cancel then: the dispersion equation predicts decay of amplitude due to
radiation. These waves belong to the category of leaky waves (see, e.g.,
Hessel and Oliner 1965; Marcuvitz 1956), an interesting subject with a
large literature, but beyond the scope of the present book.
Equation (8.28) is in an analytical form but the infinite series still
needs to be summed numerically up to a certain number of elements. It
can be shown that the series is convergent but converges rather slowly
in the vicinity of the light line. To demonstrate that convergence nu-
merically we need a few parameters, which we shall choose as L = 33
nH, ω0 /(2 π) = 0.96 GHz, a = 25 mm, r0 = 10 mm. The dispersion
curves corresponding to N = 10, 100 and 1000 elements are shown in
Fig. 8.17. Taking now N = 1000 we can next examine the difference
between the quasi-static and the full dispersion equation for the lossless
case. For the previous set of parameters it is shown in Fig. 8.18. The
main difference is that for the quasi-static case there is no dip at the
light line. For larger values of k there is hardly any difference.
8.2.3 The nature of the dispersion equation

As we have said many times before the dispersion equation relates the
frequency to the wave number. If we know one of them we can find
the other one from the dispersion equation. In the derivation above we
managed to show that as long as we take into account all the elements
up to infinity and k > k0 , all the terms in the dispersion equation are
real: there is no radiation loss. What about ohmic losses? There are
no difficulties formally to include losses. The wave then declines as it
propagates, which is taken into account by making k complex. But then
the concept of an infinite number of elements becomes a little unrealistic.
Surely, there is no power left infinitely far away from the excitation,
N s= 1000
1.02

ZZ quasi-static
retarded
Fig. 8.18 Comparison between the light line

retarded and quasi-static dispersion 0.98
curves for Ns = 1000. From Zhurom- 0.2 0.4 0.6 0.8 1
skyy (2008) ka/S
and, anyway, where is the line excited? At minus infinity? At zero?

The dispersion equation does not know about it. So we are in trouble
with the infinitely long line if we want to include losses. Is there a way
out? In principle there is. Instead of specifying the frequency we could
specify the wave number, i.e. impose a periodic spatial variation upon
the infinitely long line. For the lossless case the physical picture does not
change. To the chosen k there will be a corresponding ω, meaning that
the wave amplitude will vary harmonically at every point in the line.
The physical picture is still OK for the lossy case. After being excited
at every element of the infinite line the wave amplitude declines as a
function of time everywhere in space while it oscillates at the frequency
ω. Mathematically, the decay appears by assigning an imaginary ω to
the real k. For a given set of parameters and for a given real k we can
solve the dispersion equation and obtain from it both the real and the
imaginary part of ω. So the physical picture is consistent. But this
approach also has some drawbacks. First, in any practical situation
we are interested in Im(k) (we want to know how the wave declines
as it propagates), and not in Im(ω), and also, in practice, we impose
the frequency and not the wave number. The third drawback is the
infinite number of elements in the model: it is difficult experimentally
to realize it. So neither of the models discussed above are perfect. The
question arises: has it been a worthwhile exercise to derive the dispersion
equation? The answer is: certainly, yes. The dispersion equation gives
a good idea of what is going on and what can be expected, but if we
want to know what happens in a practical situation we need to look at
a finite line and do the analysis for the actual excitation. We shall do
this in the next section.
8.2.4 A 500-element line

Our aim is to find the current distribution for particular excitations, and
in addition, to see how a kind of dispersion equation can be deduced
from the data obtained and how close the values of ω and k are to those
predicted by the dispersion equation derived in Section 8.2.2. We also

need to choose the length of the line. The longer the line the easier it
is to find the Fourier components of the current. In a paper concerned
with the effect of retardation on magnetoinductive waves (Zhuromskyy
et al., 2005b) a 100-element line was chosen and analyzed. Below we shall
repeat the exercise for a larger number of elements. A good compromise
between accuracy and the ease of a numerical solution resulted in 500
elements.
The equation that will yield the current distribution for a given ex-
citation is eqn (8.1), the generalized Ohm’s law. There is no difficulty
calculating the mutual inductance between any two elements. It has
been done a number of times before in this book. However, in this
section we chose to simplify the problem by assuming magnetic dipoles
instead of small loops. Consequently, the mutual inductance will be
given by the simple expression (see Appendix J)
µ0 πr04
M= (1 + j k0 a − k02 a2 ) e −j k0 a . (8.31)
4a3
A further problem is the choice of resistances. We have decided to
regard the ohmic resistance as a parameter that follows from the assumed
value of Q, which we shall often vary within a wide range to illustrate the
effect of losses. What about the radiation resistance? Haven’t we proven
that the line does not radiate so we need not bother about radiation
resistance? Yes, that is true to the right of the light line and for an
infinite number of elements. Once the number of elements is finite we
are not entitled to disregard radiation. Hence, we shall include the
radiation resistance of a small loop that was given by eqn (1.96) and
repeated below
4
π 2 πr0
Rs = η0 . (8.32)
6 λ
For the parameters given above this radiation resistance turns out to
be 0.32 ohm. It must be added to the ohmic resistance so the effective
Q will decline.
Having got all the parameters of the impedance matrix we are ready to
determine the current distribution by inverting the impedance matrix.
Excitation by the first element. This means that in the voltage vector
V the first element is taken as 1 V and the further 499 elements as
zero. The current distribution may now be calculated for a given value
of frequency that we shall choose to be f = 1.01 f0.
The modulus and phase of the current flowing in the nth element are
plotted in Figs. 8.19(a) and (b) for Q = 100. The current, as may be ex-
pected, declines roughly exponentially. The phase variation looks rather
odd, quite different from phase variations one usually encounters. It
may be approximated by two straight lines: up to the first 15 elements
the phase increases linearly along the line and then the phase variation
reverses and declines linearly up to the end of the line. The increasing
variation is a sign of a backward wave, whereas the decreasing phase
0.2
current magnitude
10
phase [rad]
0.1
0
Fig. 8.19 (a) The magnitude of the 0

current and (b) its phase as a function
10 20 30 40 50 10 20 30 40 50
of element number for a quality factor element number element number
of 100. From Zhuromskyy (2008) (a) (b)
1 1
forward wave
normalized current
normalized current
backward wave
0
0
real part real part

Fig. 8.20 (a) The real and (b) the imaginary part imag. part
imaginary part of the current as a
function of element number showing 5 10 15 20 35 40 45 50
whether a wave is forward or backward. element number element number
From Zhuromskyy (2008) (a) (b)
variation indicates a forward wave. The explanation for this rather odd
behaviour is that two waves are present: a backward wave and a forward
wave, both excited at element 1. The backward wave is excited with a
higher amplitude, hence it dominates for the first 15 elements. How-
ever, this backward wave is lossy and has a higher attenuation than the
forward wave. After propagating for 15 elements the two waves are of
about equal amplitude. After the 15th element the forward wave dom-
inates, and hence the decreasing phase variation. A rough calculation
would give for ka (i.e. for the phase change per element) −0.22 π for the
backward wave and 0.16 π for the forward wave. Note that the power
moves from the excited first element to the end of the line corresponding
to a positive group velocity, which applies both to the backward and to
the forward wave. But, as the name implies, the backward wave has a
phase velocity pointing towards the source.
An alternative way to gain an intuitive feel for the periodic variation
of the current is to plot its real and imaginary part. These are shown in
Figs. 8.20(a) and (b) for n = 1 to 20 and for n = 30 to 50, respectively.
These figures confirm the conclusions drawn from Fig. 8.19(b). At the
beginning of the line one can see only the backward wave. By element 15
it declines sufficiently so that the forward wave, which is less attenuated,
can be seen. Figures 8.20(a) and (b) may, clearly, provide the value of
0.16
|Fourier amplitude|
|Fourier amplitude|
0.8
0.12
0.4
0.08
0.04 0 Fig. 8.21 Fourier spectrum as a func-

0 200 400 0 200 400 tion of the components for (a) f /f0 =
n n 1.03 and (b) f /f0 = 1.00. From
(a) (b) Zhuromskyy (2008)
1
Q= 5000
0.5
0
1
Q= 2500
0.5
amplit ude
0
1
Q= 1000
0.5
0
1
Q= 100
0.5
0 Fig. 8.22 Decay of current amplitude
0 100 200 300 400 500 as a function of element number. From
element number Zhuromskyy (2008)
ka from the periodic behaviour of the curves.

The above two methods of extracting the information have been shown
for their intuitive value. There is of course a more rigorous way of
obtaining the spatial spectrum of the waves. Take the complex value
of the current for each element from n = 1 to 500 and find its Fourier
components. The results of that exercise (using Matlab program fft )
are shown in Figs. 8.21(a) and (b) for f /f0 = 1.00 and 1.03. The
value of ka can be obtained from these diagrams by the following rule.
ka = −2 π(n − 1)/499 when n > 250.5 and ka = 2 π(501 − n)/499
otherwise. The peaks both for the backward wave and for the forward
wave may be clearly seen. The spectra are broad due to losses.
We can repeat the exercise for a number of Q values. As Q increases
losses decrease and the amplitude decline is more gradual. This is shown
in Fig. 8.22 for Q = 100, 1000, 2500, 5000. The phase against element
number also depends strongly on the value of Q, as shown in Fig. 8.23.
We can see again that as losses are lower the backward wave dominates
further away in the line.
Next, we shall find the Fourier coefficients of the current distribution
(Figs. 8.24(a)–(d)) presented by a contour code for four values of Q:
600 Q = 5000
Q = 4000
Q = 3000
400 Q = 2000
Q = 1000
Q = 100
200
phase
−200
Fig. 8.23 Variation of the phase of the −400

current as a function of element num- 0 100 200 300 400 500
ber. From Zhuromskyy (2008) element number
100, 1000, 10 000 and infinity. For a given value of Q we find a spread
in the corresponding k values. The amount of the spread depends, as
may be expected, on the value of Q. For Q = 10 000 and infinity the
lines representing the dispersion curve are quite narrow so they may
be taken as the ‘true’ dispersion curve. For Q = 10 000 there is a just
discernible curve in the range 1 < ka/π < 2. For Q = ∞ the curve
is clearly visible. The reason is that without a matched load there is a
nearly perfect reflection from the end of the array, hence, in the presence
of low ohmic losses, the k values are about the same in both directions.
The spread in the wave vector was shown in order to appreciate how
narrow (or wide) is the Fourier spectrum. Taking the maxima of the
spectrum, an unambiguous dispersion curve is obtained for each value
of Q as shown in Figs. 8.25(a)–(d). The new feature shown is the conver-
gence of the curve to the light-line for Q = 10 000 and ∞, which means
that another solution exists (represented by a local maximum) as well.
It could not be seen in Figs. 8.24(c) and (d) because their values were
too small. This is in line with the conclusions arrived at earlier in this
section. There are two waves excited, a backward wave and a forward
wave, the latter one propagating close to the velocity of light.
The dispersion curves shown in Figs. 8.25(a)–(d) have been found
by exciting the first element. Would we find the same curves if more
elements were excited? We tried exciting the first few elements (up to
5); the dispersion curves found were no different.
Imposing the wave number. In the example given above the temporal
frequency ω was imposed and we looked at the spread in the wave num-
ber. There is another obvious possibility: to impose the wave number
and look at the spread in the temporal frequency. We found similar
results to those of Figs. 8.24(a)–(d). We shall not show the spread here.
We shall give only one value of ω for one value of k. Our criterion for
finding that single value of ω is based on the concept of the input power
1.03 1 1.03 10
1.02 0.8 1.02 8
1.01 0.6 ω/ω0 1.01 6

ω/ω0
1 0.4 1 4
0.99 0.2 0.99 2
0.98 0 0.98 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
ka/π ka/π
(a) (b)
1.03 1.03 100

70
1.02 60 1.02 80
50
ω/ω0
ω/ω0
1.01 1.01 60
40
1 30 1 40
20
0.99 0.99 20
10
0.98 0 0.98 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
ka/π ka/π
(c) (d)
Fig. 8.24 Magnitude and distribution of the Fourier component plotted in the normalized ω − ka plane for various values of
the quality factor, (a)–(d) Q = 100, 1000, 10 000 and ∞. From Zhuromskyy (2008)
500
X
Pin = Vi Ii∗ . (8.33)
i=1
The frequency at which the input power is maximum is then regarded

as a point on the dispersion curve. For Q = 1000 the dispersion curve is
plotted in Fig. 8.26. Curves are not shown for other values of Q because
they are very close to that in Fig. 8.26. Apparently, losses have little
influence on the ω–ka relationship, presumably because the excitation
occurs all along the line.
One would hope that exciting the first element or exciting all of them
in a periodic manner will lead more or less to the same dispersion curve
for the same value of Q. We can check this assertion by comparing, say,
the curves in Fig. 8.25(c) with that of Fig. 8.26. This is shown in Fig.
8.27 where both are plotted. The agreement may be seen to be very
1.03 1.03
1.02 1.02
1.01 1.01
ω/ω0
ω/ω0
1 1
0.99 0.99
0.98 0.98
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
ka/π ka/π
(a) (b)
1.03 1.03
1.02 1.02
1.01 1.01
ω/ω0
ω/ω0
1 1
0.99 0.99
0.98 0.98
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
ka/π ka/π
(c) (d)
Fig. 8.25 Dispersion curves derived from the maximum of the Fourier spectrum, (a)–(d) Q = 100, 1000, 10 000 and ∞. From
Zhuromskyy (2008)
good to the right of the light line. The two curves coincide within the
thickness of the line. To the left of the light line they differ considerably.
We cannot say which is correct, they are just the results of different
excitations but there is no doubt that the one obtained from imposed
wave number is closer to the dispersion curve of the infinite line. The
information about the solution at the light line comes from the first
element excitation. The reason is technical. When ω is imposed we can
find a value of k close to the light line but we cannot impose k values
infinitely close to each other.
8.2.5 Conclusions
We cannot claim to be able to draw clear conclusions. When the line is
excited in a certain manner then we can predict what happens whether it
is ω or k that is imposed. We have shown that in the planar configuration
two waves are excited: a high attenuation backward wave and a low-
attenuation forward wave that travels practically at the velocity of light.
1.03
1.02
1.01
ZZ
0.99
Fig. 8.26 Dispersion curve for Q =

0.98 1000 derived from the criterion of max-
0 0.2 0.4 0.6 0.8 1
imum input power. From Zhuromskyy
ka/S (2008)
1.03
1.02
1.01
ZZ
0.99
Fig. 8.27 Dispersion curves for Q =
1000 derived from the criterion of
0.98 Fourier spectrum maximum (solid
0 0.2 0.4 0.6 0.8 1
lines) and input power maximum (dot-
ka/S ted lines). From Zhuromskyy (2008)
We can find current distributions and from these, if needed, we can find
the radiation patterns. The ambiguity comes in when we consider the
dispersion curve for infinite number of elements. It is not something that
could ever be measured and for the lossy case its significance is rather
limited.
8.3 Non-linear effects in magnetoinductive

waves
8.3.1 Introduction
We have looked at a number of properties of magnetoinductive waves:
we know about the interaction between resonant elements, the disper-
sion curve, propagation and attenuation, stop bands and pass bands, the
interaction between arrays, and various applications. So far they were
all linear. Can we extend our analysis to the non-linear case? Exper-
imentally it is quite easy. Since in most cases the basic building block
was the capacitively loaded loop all we need to do is to insert an element

whose capacity varies with the voltage. Such an element is a varactor
diode. As an approximation we shall assume that the variable part of
the capacitance is a linear function of the voltage
(nl)
C = C0 (1 + γV ) , (8.34)
where C0 and γ are constants, and the second term is assumed to be
small relative to the first term. What do we want to do with this non-
linear diode? The application we have in mind is parametric amplifica-
tion. It could serve for amplifying a signal with little noise added, and
it could, in general, be used for loss compensation.
R Parametric amplifiers were invented in the 1930s. For a long time they
were no more than scientific curiosities but came into their own when,
owing to their low-loss properties, they were used as pre-amplifiers in
L satellite communications. With the advent of coherent optics they also
Vs played a role in producing practical oscillators (see, e.g., Solymar and
Walsh 2004). The basic circuit is shown in Fig. 8.28. We have an RLC
circuit in which the value of the capacitance is changing at twice the
C =C0(1įcos2Ȧst) signal frequency of ωs . In general, there are three frequencies the signal
at ωs , the pump at ωp and the idler at ωi . For our purpose it is perfectly
adequate to omit the idler and to use the pump at 2 ωs .
The aim is to amplify waves that propagate in a line. In our case they
Fig. 8.28 RLC circuit in which the happen to be magnetoinductive waves, but whatever type the waves to
value of the capacitance is changing at be amplified are they need a pump wave, with which they can interact
twice the signal frequency of ωs and which travels along with the same phase velocity. Thus, besides the
condition of ωp = 2 ωs , we also need to satisfy the condition βp = 2 βs .
In that case the signal velocity vs = ωs /βs will be equal to the pump
velocity vp = ωp /βp . This is the so-called phase-matching condition, the
subject of Section 8.3.2.
8.3.2 Phase matching

In most of the examples we have seen so far the magnetoinductive wave
was confined to a narrow band of frequencies. The corresponding values
of the coupling coefficient were not much larger than |κ| = 0.1. Such
a band is clearly too narrow for any chance of satisfying the phase-
matching condition. We need a much larger |κ|. How large? We shall
find it by substituting the phase-matching conditions into the dispersion
equation of magnetoinductive waves, which yield
ωs2 1
2 = 1 + κ cos(β a) (8.35)
ω0 s
for the signal wave and
4ωs2 1
= (8.36)
ω02 1 + κ cos(2 βs a)
for the pump wave. We shall choose ωs /(2 π) = 63.87 MHz, and the
mutual inductance as M = 0.45L (i.e. κ = 0.9). We can then solve
(2f s ,2Es )
frequency f [MHz]
frequency f [MHz]
(2f s ,2Es )
(f s ,E s )
(f s ,E s )
Fig. 8.29 Phase matching in (a) single

axial and (b) two coupled planar arrays
supporting MI waves. From Sydoruk
c 2007 Wiley-
et al. (2007a), copyright
propagation constant Ea/S propagation constant Ea/S VCH Verlag GmbH & Co. KGaA, and
Sydoruk et al. (2007c), copyright c
(a) (b) 2007 IOP Publishing Ltd
eqns (8.35) and (8.36) for the resonant frequency and propagation con-
stant, giving, respectively, ω0 /(2 π) = 73.44 MHz and βs a = 0.38 π. For
elements with inductance L = 50 nH this value of the resonant frequency
can be achieved by loading the elements by capacitances of C0 = 94 pF.
The question arises whether we can have such a high coupling coeffi-
cient. According to recent experimental work (Syms et al., 2006b; Syms
et al., 2007a) this is indeed possible, provided the elements in the axial
configuration are sufficiently close2 to each other. So, the possibility is 2
In fact, coupling coefficients as high
there. If the sole purpose is to amplify a magnetoinductive wave then as κ = 1.5 have been reported (Syms
et al., 2006b).
such an axial array could do it. For the elements given above the disper-
sion curve is shown in Fig. 8.29(a). It can be seen that both the signal
and the pump waves can propagate and are phase matched.
An alternative solution is to rely on two coupled arrays (Sydoruk
et al., 2006). The dispersion curve for such coupled arrays has al-
ready been shown in Fig. 7.36 for the planar configuration. It consists
of two branches, hence the phase-matching condition can be satisfied
with small values of the coupling coefficient. We shall choose again
ωs /(2 π) = 63.87 MHz and L = 50 nH for the signal frequency and
the inductance of the loop. We take two identical arrays with sepa-
ration of h = 10 mm between them (see Fig. 7.35(a)). This time the
upper array is tuned to the signal frequency and the lower array to the
pump frequency. The distance between the elements within the same
array is a = 24 mm. Hence, the mutual inductances can be calculated3 3
Note that in Section 7.13 several
as M = 0.15L and M1 = −0.05L. These values are close to those mutual inductances were considered.
In the present section, for simplicity,
investigated experimentally in Section 7.13. For choosing the propaga- we use the approximation that only
tion coefficients we have quite a lot of freedom. We shall take them as nearest-neighbour coupling needs to be
βs = π/(3a) and βp = 2 π/(3a). Two parameters that are left undefined taken into account.
are the capacitances of the elements C1,2 . They are to be determined
from the condition that the two phase-matched waves satisfy the dis-
persion relation. For the two coupled waveguides it is given (Sydoruk
et al., 2006) as
C (nl)=C0(1+JU) C (nl)=C0(1+JU) C (nl)=C0(1+JU)
M
L/2
L/2 L/2 L/2
... L/2 L/2
Fig. 8.30 An equivalent circuit of a sin- U
gle array. The capacitance is voltage-
dependent, leading to parametric inter-
action between the signal and the pump R R R ZT

1
j ωL + + R1 + 2j ωM1 cos ka
j ωC1

1
× j ωL + + R2 + 2j ωM1 cos ka = −ω 2 M 2 . (8.37)
j ωC2
The above equation must be satisfied both for ωs , βs and for 2 ωs , 2 βs .
From the solution we find then that C1 = 130 pF and C2 = 30 pF. The
corresponding resonant frequencies are ω01 /(2 π) = 62.49 MHz for the
upper array and ω02 /(2 π) = 129.15 MHz for the lower one. The two
branches of the dispersion curve showing phase matching are plotted in
Fig. 8.29(b).
8.3.3 Theoretical formulation of amplification for

the single array
The equivalent circuit of a non-linear magnetoinductive waveguide is
shown in Fig. 8.30. There is a voltage source in the first element and
a matching impedance, both for the signal and pump waves (see Sec-
tion 7.3), in the last element.
The voltage in the nth element across the non-linear capacitor is taken
as
1h
us (n) e j νs (n) + c.c. ,
i
Us (n) = (8.38)
2
1h
up (n) e 2j νs (n) + c.c. ,
i
Up (n) = (8.39)
2
where
νs (n) = ωs t − nβs a , (8.40)

and c.c. denotes complex conjugate. Similarly, we can write for the
current in the nth element. Note that when relations are non-linear we
need to exercise great care with complex notations. The intermediate
variables us , up can be complex, but we ensure, by taking the complex
conjugate, that the physically significant quantities, the signal and pump
voltages are real.
The current flowing through the capacitor is the temporal derivative

of the charge across the capacitor, hence it may be written in the general
form
d(CU )
In = . (8.41)
dt
The problem is now that the capacitor, being voltage-dependent (see
eqn (8.34)), the current will have a number of harmonics, even if we
can assume (as we did in eqns (8.39) and (8.40)) that only the signal
and pump voltages will be present across the capacitor. We shall follow
here the time-honoured method and simplify the problem by further
assuming that the currents will also have only those two frequencies:
1h
is (n) e j νs (n) + c.c.
i
Is (n) = (8.42)
2
and
1h
ip (n) e 2j νs (n) + c.c. .
i
Ip (n) = (8.43)
2
The relationship between the voltages and currents will be determined
by eqn (8.41). The linear part of the problem is trivial but the non-linear
part is less so. Let us evaluate the term γV 2 when the signal and pump
voltages are simultaneously present. We find
γh i2
γV 2 = us (n) e j νs (n) + up (n) e 2j νs (n) + c.c. . (8.44)
4
It can be immediately seen that the term varying at the pump frequency
will take the form
γ 2
u (n) e 2j νs (n) . (8.45)
4 s
It is a little more difficult to recognize the term varying at the signal
frequency. It is
γ
up u∗s (n) e j νs (n) , (8.46)
2
where the star is our other notation for the complex conjugate. There are
13 more terms in eqn (8.44) but none of them are at these frequencies.
Adding now the current due to the linear term in eqn (8.41) to those in
eqns (8.45) and (8.46) we find the signal and pump currents as
is (n) = j ωs C0 [us (n) + γup (n)u∗s (n)] (8.47)

and
h γ i
ip (n) = 2j ωs C0 up (n) + u2p (n) . (8.48)
2
One more relationship that needs to be satisfied is of course Kirchhoff’s
voltage law that we have quoted a number of times and shall write down
below but this time we need to keep the temporal derivatives,
d dI(n)
M [I(n + 1) + I(n − 1)] + L + RI(n) + U (n) = 0 . (8.49)
dt dt
The above equation relates nearest neighbours to each other involving
the (n − 1)th, the nth and the (n + 1)th element. The equation needs
to be modified when it is written for the first and the last element in
the array that have only one neighbour. The technique is described in
Section 7.3 for finding the matching impedance. The present case is a
little more complicated because the line needs to be matched at both
the signal and the pump frequency so that
ZT (ωs ) = j ωs M e −j βs a
. (8.50)
ZT (2ωs ) = 2j ωs M e −2j βs a
Performing the rather tedious algebraic operations the final result may
be presented in matrix form (Sydoruk, 2007) as follows

(s) 1
 Z −j E us + γZ(s) qs = us0


ωs C0
. (8.51)
(p) 1 γ
 Z −j E up + Z(p) qp = up0


2ωs C0 2
Here, E is an N × N identity matrix,
 R 
L+ M e −j βs a ... 0
 j ωs 
 R .. 
M e βs a
j L+ ... .
 
j ωs
 
 
Z (s)
= j ωs 
 .. .. .. 
 . . . 0 

 .. .. .. 
. . . M e −j βs a
 
 
 R + ZT (ωs ) 
0 ... M e j βs a L+
j ωs
(8.52)
and
 R 
L+ M e −2j βs a ... 0
 2j ωs 
 R ..
M e βs a
2j L+ ... .
 
2j ωs
 
 
Z(p)
= 2j ωs 
 .. .. .. .

 . . . 0 
 .. .. .. 
. . . M e −2j βs a
 
 
 R + ZT (2ωs ) 
0 ... M e 2j βs a L+
2j ωs
(8.53)
us0 = (us0 , 0, . . . , 0)T and up0 = (up0 , 0, . . . , 0)T (the superscript T

means transpose of a vector). Here, us0 and up0 characterize the ampli-
tudes of, respectively, the signal and pump sources in the first element of
the array. The vector qs has elements qsn = up (n)u∗s (n), and the vector
qp , the elements qpn = u2s (n).
Equation (8.49) constitutes a system of difference equations in space
that can be solved numerically for the signal and pump voltages for a
given input. For an array of N elements a system of 2N equations should
be solved. It is rather tiresome to do so but if the elements are different
that is the only way to solve the problem. For identical elements it is
usually preferable to convert the problem into a continuous one and turn
the difference equations into differential equations. Provided the change
of variables from element to element is small, this can be done. In fact,
we have already done so in Section 8.1.5. In the present section the
technique will be somewhat different because we don’t want to impose
the restriction that the phase change must be small from element to
element. We only wish to assume that us (n) will have a value close to
us (n − 1) and to us (n + 1). In that case we can convert the discrete
problem into a continuous one by using the following relationships
dus,p (n)
z = (n − 1)a and us,p (n ± 1) = us,p (n) ± a . (8.54)
dz
As an illustration, let us convert the expression for the signal wave
j ωs M us (n + 1) e −j βs a + j ωs M us (n − 1) e j βs a (8.55)
into continuous form. We find that it comes to

dus dus
j ωs M e −j βs a + us (z) − a
us (z) + a e j βs a
dt dt

dus
= j ωs M us (z) + a sin(βs a) . (8.56)
dt
Following the same technique and assuming that losses are weak and the
non-linearity of the capacitance γ is small we end up (Sydoruk, 2007)
with the differential equations
dus (z)
= −j gs up (z)u∗s (z) − αs us (z)
dz
, , (8.57)
dup (z)
= −j gp u2s (z) − αp up (z)
dz
where
γ γ
gs = and gp = (8.58)
2aωs2 C0 M sin βs a 16aωs2 C0 M sin 2βs a
characterize non-linearity and
signal signal
amplitude [V]
amplitude [V]
pump pump
Fig. 8.31 Parametric amplification in a

single array: (a) without and (b) with
losses. From Sydoruk et al. (2007a).
Copyright c 2007 Wiley-VCH Verlag element number element number
GmbH & Co. KGaA (a) (b)
maximum gain [dB]
maximum position
Fig. 8.32 Maximum gain achieved and

maximum gain
position of the signal maximum for dif- maximum position
ferent amplitudes of the signal source.
From Sydoruk et al. (2007a). Copy-
c 2007 Wiley-VCH Verlag GmbH
right
& Co. KGaA signal source amplitude [mV]
R R
αs = and αp = (8.59)
2aωs M sin βs a 4aωs M sin 2βs a
are the attenuation coefficients. Note that eqns (8.57) are analogous
to the corresponding equations for parametric amplification and second-
harmonic generation well known in non-linear optics (Shen, 1984; Bloem-
bergen, 2005).
For a numerical example we shall take the parameters given above
and γ = 0.1 V−1 , up0 = 0.25 V and us0 = 0.5 mV. In the absence of
losses there is a periodic exchange of power between the signal and pump
waves, as shown in Fig. 8.31(a). Both waves decline in the presence of
loss as shown in Fig. 8.31(b) for Q = 250. In a practical case of course the
pump wave is much stronger than the signal wave. The amplification
against signal input amplitude for that case is shown in Fig. 8.32 for
us0 = 1 µV–100 mV. It may be seen that the amplification is constant
up to a certain signal amplitude and then starts to decline.
C1(nl)=C1(1+JU) C1(nl)=C1(1+JU) C1(nl)=C1(1+JU)
M1
L/2
L/2 L/2 L/2
... L/2 L/2
U R1 R1 R1 ZT1
M M M
C2 C2 C2 Fig. 8.33 Equivalent circuit of two cou-
pled arrays. The capacitance of the top
L/2 array is voltage-dependent, leading to
L/2 L/2 L/2 L/2 L/2 parametric interaction between the sig-
... nal and the pump. From Sydoruk et al.
V R2 R2 R1 ZT2
M1 (2007c). Copyright c 2007 IOP Pub-
lishing Ltd
top waveguide bottom waveguide
pump
amplitude [V]
pump
signal
Fig. 8.34 Periodic exchange of power

between the signal and pump in the
signal general case, top (a) and bottom (b)
waveguides. U0 = 10 mV, V0 =
125 mV, γ = 0.1 V−1 . Losses are not
element number element number included. From Sydoruk et al. (2007c).
(a) (b) Copyright c 2007 IOP Publishing Ltd
8.3.4 Theoretical formulation for the coupled

arrays
As indicated in Section 8.3.2 an alternative solution for achieving phase
matching is to use two coupled arrays. This has two main advantages:
firstly, more parameters give more freedom to design and a solution is
possible in terms of planar arrays, which give a chance of further inter-
action with external agents. The equivalent circuit is shown in Fig. 8.33.
The signal input is in the upper array and that of the pump in the lower
array. However, due to the coupling the signal and the pump propagate
in both arrays in spite of the fact that (remember from Section 8.3.2)
the upper array is tuned to the signal frequency and the lower array
to the pump frequency. A non-linear element is clearly necessary for
parametric amplification, but it turns out that it is sufficient to insert a
non-linear element in the upper array only.
The mathematical formulation is entirely analogous to that of a single
array. We can derive a discrete set of equations in matrix form and it
is also possible to assume continuity of the variables and formulate the
problem in terms of partial differential equations. For the details of the
derivations see Sydoruk et al. (2007c). Here, we shall only formulate the
top waveguide bottom waveguide
amplitude [V]
(b)
high/low amplification
maximum gain [dB]
element number
amplitude [V]
no amplification
(c)
signal
max. gain pump
max. position
quality factor
(a)
amplitude [V]
(d)
element number element number
Fig. 8.35 Maximum signal gain achieved (dotted line) and the position of the signal maximum (solid line) depending on the
amount of loss (a) and the distribution of the signal (solid lines) and pump (dashed lines) amplitudes along the waveguides for
(b) Q = 1000, (c) Q = 400, and (d) Q = 150. Other parameters are as in Fig. 8.34. From Sydoruk et al. (2007c). Copyright
c 2007 IOP Publishing Ltd
problem and show some results.

It is convenient to write the equations in terms of the voltages across
the capacitors of the nth elements in both arrays
1h
us (n) e j (ωs t − nβs a) + c.c.
i
Us (n) =
2
1h
up (n) e 2j (ωs t − nβs a) + c.c.
i
Up (n) =
2
. (8.60)
1h j (ω s t − nβ s a)
i
Vs (n) = vs (n) e + c.c.
2
1h
vp (n) e 2j (ωs t − nβs a) + c.c.
i
Vp (n) =
2
Here, us (n) and up (n) are the amplitudes of, respectively, the signal and
pump waves in the upper array, and vs (n) and vp (n) are their amplitudes
in the lower array. The corresponding currents flowing through the
elements of the upper and lower arrays are, respectively,
(nl)
d[C1 (Us (n) + Up (n))]
In (n) =
dt
. (8.61)
d[Vs (n) + Vp (n)]
Jn (n) = C2
dt
Neglecting higher harmonics the currents can be written separately at
the signal and pump frequencies as
j ωs C1
Is (n) = [us (n) + γup (n)u∗s (n)] e j (ωs t − nβs a) + c.c.
2
j ωs C1
2up(n) + γu2s (n) e 2j (ωs t − nβs a) + c.c.

Ip (n) =
2
. (8.62)
amplitude [PV]
j ωs C2
Js (n) = vs (n) e j (ωs t − nβs a) + c.c.
2
Jp (n) = j ωs C2 vp (n) e 2j (ωs t − nβs a) + c.c.
The next steps are to substitute the currents and voltages into the dif- with amplification
ference equations of the magnetoinductive waves taking into account the without amplification
excitation and matched terminal impedances, turn them into differential
equations and solve the differential equations numerically. element number
A set of results showing the periodic exchange of energy between the
signal and pump waves is shown in Fig. 8.34. The gain against signal Fig. 8.36 Loss compensation by means
amplitude is plotted in Fig. 8.35(a), whereas Figs. 8.35(b)–(d) show of parametric amplification. The am-
the variation of signal and pump amplitudes as a function of element plitude of the signal in the ampli-
number. An interesting conclusion is that in the upper array the signal fied case (solid line) is almost con-
stant along the waveguide, whereas
amplitude may well exceed that of the pump. The pump amplitude is it declines exponentially in the non-
of course still high in the lower array. amplified case (dash-dotted) line. U0 =
A possible use of parametric amplification is for loss compensation. 1 µV, V0 = 50 mV, Q = 500, γ = 0.1
V−1 . From Sydoruk et al. (2007c).
Taking Q = 500 and an appropriate choice of parameters the signal c 2007 IOP Publishing Ltd
Copyright
amplitude in the presence and absence of loss is shown in Fig. 8.36. It
may be seen that the amplitude of the signal can be kept constant with
good approximation.
8.3.5 MRI detector

At the time of writing experimental work is going on in Syms’ group
at Imperial College on a detector (see Fig. 8.37) based on parametric
amplification of MI waves (Syms et al., 2007b; Syms et al., 2008). The
elements are spirals loaded by lumped capacitors to make them resonant
at a required frequency. The signal wave is excited by rotating nuclear
dipoles and travels (on the lower set of elements) in synchronism with
Fig. 8.37 Experimental setup of a

magnetoinductive ring resonator. From
Syms et al. (2008). For coloured version
see plate section
them under conditions of rotational resonance. In contrast to the two-

wave amplifiers discussed in the previous sections, this is a three-wave
amplifier. The signal and pump waves both propagate in the planar
configuration. The idler wave is realized by lumped circuits.
Seven topics in search of a
chapter
9
9.2 Further imaging
We have tried to be as comprehensive as possible in our treatment of mechanisms 290
the various phenomena of metamaterials. It is, however, inevitable that 9.3 Combinations of
some topics just do not fit into the general framework. We introduced negative-permittivity and
imaging by negative-index materials in Chapter 2 and discussed it later negative-permeability
layers 296
in Chapter 5 in considerable detail. Then, there was imaging by MI
waves both in Chapters 7 and 8. Nevertheless, there are still some
9.5 Gaussian beams and
aspects of imaging that are important and need to be included. They the Goos–Hanchen shift 299
will be discussed in Section 9.2.
Although multilayers were discussed in Section 5.5 and have also been
9.7 Refractive index
treated in a different context in Section 8.1.2 there is still a combination close to zero 306
of layers (negative permittivity and negative permeability in neighbour- 9.8 Invisibility and cloaking 309
ing layers) that did not seem to fit into any of the discussions. We
describe them briefly in Section 9.3.
Focusing by indefinite media is an odd phenomenon because it hap-
pens only under specific circumstances when both a certain type of
anisotropy and a periodic structure are present. The physics is simi-
lar to that of MI waves (see construction of the output beam in Fig.
8.7) but there we were concerned with only a single plane wave incident
at a particular angle, whereas for focusing with indefinite media the in-
cident beam is taken as divergent. The mathematical treatments are
also quite different so we decided to treat indefinite media separately in
Section 9.4.
Another interesting phenomenon is the Goos–Hanchen shift that stip-
ulates that at reflection a finite beam appears to be shifted laterally. It
is well known for positive-index materials. We shall give in Section 9.5
a summary of what happens when one of the media has negative in-
dex, and discuss briefly how the beam is guided when a negative-index
material is surrounded by two positive-index materials.
Section 9.6 is on waves in nanoparticles. It is a little doubtful whether
it should be included here because nanoparticles-cum-nanophotonics is
a vast subject on its own. We have decided to include here those aspects
that are akin to magnetoinductive waves.
Section 9.7 is devoted to phenomena occurring when the permittivity 1
Similar phenomena will of course oc-
is close to zero.1 The main effect is high phase velocity, well above the cur when the permeability is near to
velocity of light. The consequences can be described by saying that those zero.
290 Seven topics in search of a chapter
r r r2 r2 r r
Fig. 9.1 The classical imaging arrange-

ment of the ‘perfect lens’. The region
between z = d/2 and z = 3d/2 is usu-
ally a negative-index material. How-
ever, this figure may serve to illustrate
the presence of two plates that should
be resonant and coupled
in a hurry do not pay attention to small obstacles. They just sweep

past them. Finally, we shall come to a late entry to the field: cloaking
and invisibility. The basic principles will be discussed and some of the
proposed solutions illustrated in Section 9.8.
On the whole, this chapter differs from the others in the sense that
there will be very little mathematics. The emphasis will always be on
the multifarious physical phenomena, a hallmark of the subject of meta-
(a) materials.
9.2 Further imaging mechanisms

9.2.1 Parallel sheets consisting of resonant elements
We shall reproduce here in Fig. 9.1 once more the basic configuration
of the flat lens suitable for subwavelength imaging (Pendry, 2000). The
lens consisting of a material with εr = µr = −1 is between z = d/2
and z = 3d/2, the object is at z = 0 and the image is obtained at
(b) z = 2d. Now let us imagine that the two parallel lines in Fig. 9.1 are not
the boundaries of a negative-index material but represent two plates,
each one consisting of two-dimensional arrays of resonators. Would we
Fig. 9.2 (a) Resonant element made still have subwavelength imaging? In Sections 5.3 and 5.4 we repeat-
of thin wire of a length close to half- edly called attention to the relationship between the growing evanescent
wavelength. (b) Two small elements in
foam holders on each side with a dipole waves and the SPP resonances. Is it possible that the negative-index
antenna as a source. From Maslovski material is only there to provide two surfaces that can propagate SPPs?
et al. (2004). Copyright c 2004 Amer- Rao and Ong (2003) concluded on the basis of an FDTD (finite-difference
ican Institute of Physics time domain) numerical analysis that their results could be explained by
a simple model that relies on coupling between two resonant entities a
certain distance apart. Maslovski et al. (2004) went further by propos-
ing that for subwavelength imaging it is a sufficient condition to have
two coupled resonant sheets. For simplicity, they chose resonant lines
instead of resonant sheets in their experiments. It was accomplished
by placing the experimental apparatus between two conducting sheets a
distance of 25 mm apart.
The element chosen by Maslovski et al. (2004) is shown in Fig. 9.2(a).
It is made of thin wire. Its total length is 25 mm. It was found to be
resonant at a frequency of 5 GHz, which is about right considering that
the half-wavelength is 30 mm. Two such elements were placed in foam
Fig. 9.3 Photograph of the experimen-

tal setup. The foam holders are be-
tween two planes made by a dense mesh
of conducting strips. From Maslovski
ican Institute of Physics
holders at x = −105 mm and x = −65 mm, a distance of 40 mm apart,

shown in Fig. 9.2(b). The object was a horizontal electric dipole placed
at x = −125 mm, i.e. 20 mm behind the first set of elements. A pho-
tograph of the measuring apparatus may be seen in Fig. 9.3. The two
conducting planes are realized by dense meshes. The upper mesh was Second particle
made weakly penetrable to the fields in order to allow the measurement
of the field distribution by a probe positioned at the top of the mesh. First particle
Note that the electric dipole is parallel with the meshes, and the distance Image
between the meshes is less than half-wavelength. Consequently, the di- Source
pole can excite only evanescent waves. The wave amplitudes measured
just above the mesh are shown in Fig. 9.4. They may be seen to follow
the expected pattern. The field declines from the source to the position
of the first line, grows afterwards until it reaches the second line, and
declines again beyond the second line. Its value at the image position,
20 mm behind the second line, is the same as at the source. Further -180 -120 -60 0
experiments with more elements yielded similar conclusions. x [mm]
The essential requirement is, apparently, the presence of two coupled
resonant lines. In the experiments of Freire and Marques (2005, 2008), Fig. 9.4 Field amplitude variation
discussed in Section 8.1.6, the resonant sheets supported MI waves, along the axis. Key positions on the
hence the mechanism was very likely the same as that of Maslovski et al. plot are indicated by arrows. From
c
Maslovski et al. (2004). Copyright
(2004), with channelling effects of the type discussed in Section 7.15.3
2004 American Institute of Physics
playing a role.
9.2.2 Channelling by wire structures

The interaction between wire structures and electromagnetic waves have
been known for a long time. See the early papers of Kock 1964; Macfar-
lane 1946; Cohn 1946; Brown 1950; Brown 1953; Lewis and Casey 1952;
Casey and Lewis 1952, the still influential paper of Rotman (1962) and
those of Pendry et al. 1996; Pendry et al. 1998 in which the wires’
plasma properties were rediscovered. Wires were of course used in the
(a)
(b) (c) (d)
Fig. 9.5 (a) Horizontal stack of a square lattice of thin wires. Length chosen so as to satisfy the Fabry–Perot condition of
resonance. (b) Image of letter P 2.5 mm in front of the object obtained by numerical simulation. (c) Image 2.5 mm behind
the lens obtained by numerical simulation. (d) Electric-field distribution in the image plane measured by near-field scanning.
From Belov et al. (2006a). Copyright c 2006 by the American Physical Society. For coloured version see plate section
experiments of Shelby et al. (2001a) to produce the negative permittiv-

ity, while the negative permeability was provided by SRRs.
In this section we shall discuss how to use a wire structure for chan-
nelling the spatial information from the object to the image. In other
words this is pixel-by-pixel imaging. The basic idea is quite simple
(Belov et al., 2006a; Zhao et al., 2006; Zhao et al., 2007). Wires arranged
as in Fig. 9.5(a) will carry a wave with p polarization along the wires.
The resolution is limited by the period. The main problem is reflection
from the surfaces, which is quite significant for thicker slabs. The solu-
tion is to choose thicknesses (integral multiples of the half-wavelength)
that give rise to Fabry–Perot resonance. The wire structure chosen both
for numerical simulation and for experiment had wires of 1 mm radius,
the period was a = 10 mm, there were 21 × 21 wires, and the slab thick-
ness was 15 cm, a half-wavelength at the frequency of 1 GHz. The object
was in the form of the letter P, placed at 5 mm in front of the lens.
The images at 2.5 mm in front of the lens (Fig. 9.5(b)) and 2.5 mm
behind the lens (Fig. 9.5(c)) were obtained by the CST MICROWAVE
STUDIO package. It may be seen that transmission though the lens does
not lead to significant distortion. The results were further confirmed by

experiments. The object was an antenna in the form of the letter P fed
by a coaxial cable and located 3 mm away from the wires. The receiver
was a short piece of wire connected to a coaxial cable. The electric field
was scanned at 5 mm behind the lens. At the best frequency of 0.98
GHz the measured image is shown in Fig. 9.5(d).
Similar experiments by a linearly expanding set of wires were con-
ducted by Ikonen et al. (2007b) using the letter M as the object. The
pattern was found to be distorted at the output but still recognizable.
Thanks to the radial expansion a magnification by a factor of 3 was
achieved.
Wire structure experiments with a meander line as the object were
performed by Belov et al. (2006b). They showed (in agreement with the
theoretical predictions of Belov and Silveirinha (2006)) that a resolution
of λ/15 can be achieved irrespective of the shape and complexity of the
object.
Similar but much smaller wire structures suitable for the infra-red and
optical range were proposed by Ono et al. (2005) and Silveirinha and
Engheta (2007). For a non-local homogenization model see Silveirinha
and Engheta (2006).
A quite different idea for imaging was conceived by Belov et al. (2005).
The wire structure consisting of 14 layers is now perpendicular to the di-
rection of propagation as shown in Figs. 9.6(a) and (b). Note in the inset
that the vertical rods are capacitively loaded (analyzed earlier by Belov
et al. (2002)). The isofrequency curves corresponding to the orientation
of the wire structure shown in Fig. 9.6(b) may be seen in Fig. 9.7. Note
that at a certain value of the longitudinal wave vector the contour is flat
within a wide range of the transverse wave vector. The incident wave
vectors are represented by the half-circle on the left of the diagram. As
usual, the criterion of refraction is that tangential components of the
wave vectors must agree outside and inside the material. This means
that the transverse component of k(1) (incident wave vector) is identical
with the transverse component of q(1) (wave vector of the propagating
wave inside the wire structure). If the isofrequency curve is flat then all
the propagating waves inside the wire structure will have identical group
(1)
velocities vg pointing in the same (horizontal) direction—and that is
the condition of perfect channelling. Interestingly, the group velocity
will be roughly the same for a range of evanescent waves as well because
(2)
vg corresponding to the input evanescent wave of k(2) differs very lit-
(1)
tle from vg . Hence, the evanescent waves in air turn into propagating
waves in the wire structure. At the output of the structure the inverse
phenomenon takes place: the propagating wave turns into an evanescent
wave. Thus, the evanescent waves representing subwavelength informa-
tion can be reproduced. It is again an advantage to have thicknesses
corresponding to Fabry–Perot resonance. They are shown by vertical
lines in Fig. 9.7.
The results of a numerical simulation may be seen in Fig. 9.8. The
object is a point course located at a distance of d in front of the structure.
(a) (b)
Fig. 9.6 (a) Square lattice of capacitively loaded vertical wire structure, (b) view from above. From Belov et al. (2005).
Copyright
The focal spot was calculated as equal to λ/6.

It needs to be noted that there are no growing evanescent waves.
Thus, if the object is farther away from the input surface then it will
decline before it turns into a propagating wave inside the structure.
There is no mechanism to restore the evanescent waves to their original
amplitude. There are no growing waves and there are no SPPs either.
The mechanism of subwavelength imaging in this case is entirely different
from that proposed by Pendry (2000).
The 1D metal–dielectric structure was already discussed in Section 5.5.
It was claimed there that the structure is, under certain circumstances,
equivalent to an anisotropic structure with permittivities zero in the
transverse direction and infinitely large in the axial direction (see Ap-
pendix I). Its imaging properties were discussed in more detail by Belov
and Zhao (2006) and Li et al. (2007). The qualitative mechanism invoked
was channelling, which applied both to the propagating and evanescent
components. The length of the device could be arbitrarily long in the
lossless case. High transmission amplitude was ensured by making the
length of the device equal to an integral multiple of the half-wavelength,
i.e. by constructing a Fabry–Perot resonator. In one of their designs,
Belov and Zhao (2006) used an anisotropic high positive permittivity
metamaterial that was proposed earlier by Shen et al. (2005).
Finally, we wish to mention that the wire structures discussed in this
section (in particular the one where the wires are parallel to the in-
terface) could also be regarded as examples of photonic crystals. We
included them here owing to their significance for metamaterials. Pho-
tonic crystals themselves are beyond the scope of the present book. We
shall make only a few brief comments on them in the next section.
Fabri-Perot resonances
Normalized transverse component of wave vector, qt b/(2S)
evanescent modes
propagating modes
evanescent modes
Fig. 9.7 Iso-frequency curves in the

wave-vector plane. Lines of Fabry–
Perot resonances also shown. From
c 2005
Belov et al. (2005). Copyright
Normalized longitudinal component of wave vector, ql b/(2S)
by the American Physical Society
9.2.3 Imaging by photonic crystals

Photonic crystals rose to fame as periodic (usually dielectric) structures
capable of reflecting electromagnetic waves coming from any directions,
hence capable of trapping electromagnetic waves. Their operation is
based on a generalization of the principles upon which reflection holo-
grams works: total reflection is built up from elementary reflections
satisfying the Bragg conditions. Hence, their period is comparable with
the wavelength. Unfortunately, their design is much more complicated
because the periodic structures of photonic crystals are not minor per-
turbations of the background. The contrast must be high, e.g. the back-
ground material for a periodic set of holes may have a relative dielectric
Fig. 9.8 Distribution of the electric

field showing the source and the image.
From Belov et al. (2005). Copyright c
2005 by the American Physical Society.
For coloured version see plate section
constant as high as 10.

As a response to the advent of metamaterials those working on pho-
tonic crystals were keen to show that they can be made suitable both
for negative refraction and imaging. Whether the subject has thereby
become part of metamaterials is disputable, largely a matter of defini-
tion. In a recent book edited by Engheta and Ziolkowski (2006) entitled
metamaterials about one half of the book is devoted to electromagnetic
bandgap materials which is just another phrase for photonic crystals. As
explained in the Preface we believe that both the physical description
and the mathematical formulation of photonic crystals differ so much
from metamaterials that it is difficult to give a unified description that
applies to both of them. Hence, in this book we have not so far men-
tioned photonic crystals. Nonetheless, we shall say a few words about
them here and give a few references.
Notomi (2000) pointed out that at frequencies close to the bandgap
it is possible to define an index of refraction for waves propagating in
photonic crystals. Subsequently, it was shown theoretically that other
configurations are also possible and the concept of negative index is
not necessary (Luo et al., 2002a; Luo et al., 2002b; Luo et al., 2002c).
Experimental confirmation of imaging was provided by Parimi et al.
2003; Parimi et al. 2004.
9.3 Combinations of negative-permittivity

and negative-permeability layers
We have mentioned several times multilayer structures in different con-
texts. In each case the aim was to produce properties superior to those of
single layers. The choice in each case was a metal–dielectric structure, or
in other words combinations of layers with negative and positive permit-
tivities. A different combination of layers, namely an ε < 0, µ > 0 layer
combined with an ε > 0, µ < 0 layer, was proposed by Fredkin and Ron
(2002). They showed that such a sandwich may behave as a negative-
index material with phase and group velocities in opposite directions and
may also exhibit negative refraction. The authors also found solutions
with positive effective index with bandgaps associated with the period
of the structure. A more detailed analysis of the same problem is due to
Alu and Engheta (2003) who showed that the combination had a number
of remarkable properties, like resonance and perfect tunnelling. To show
the physics of the perfect tunnelling mechanism they plotted Poynting
vector streamlines. The streamlines were highly curved in both media
but the joint effect was a 100% transmission of the electromagnetic wave
incident at a particular angle as if the bilayer was not there. A similar
analysis with similar conclusions was performed by Wang et al. (2006).
A version that uses indefinite media anisotropic layers (see Section 9.4)
and is suitable for imaging was proposed by Schurig and Smith (2005).
Experimental realization of imaging with the aid of loaded anisotropic
transmission lines was achieved by Feng et al. (2007). They also con-
cluded that imaging by the combination of negative-permittivity and

negative-permeability layers was less sensitive to losses than Pendry’s
‘perfect lens’.
9.4 Indefinite media

In a search for media exhibiting negative refraction, Lindell et al. (2001)
came to the conclusion that uniaxially anisotropic media are suitable for
the purpose, as could be deduced from the dispersion equation. They
showed that under certain circumstances it is sufficient to have just
one of the material parameters as negative in order to achieve negative
refraction. The concept was further developed by Smith and Schurig
2003; Schurig and Smith 2005 and Smith et al. 2004b. They referred
to the media, in which the diagonal elements of the material parameter
tensors may be of opposite sign, as indefinite media. These media have
some interesting properties that we shall investigate in this section.
The assumption is that the material parameters are given by diagonal
tensors, the electric field has only a component in the y direction and
the direction of propagation is the z direction. The corresponding wave
equation was obtained in Section 1.17. We may obtain the dispersion
equation from this by assuming the electric field in the form
Ey = E0 e −j (kz z + kx x) , (9.1)
and substituting it into the wave equation. The dispersion equation may
then be obtained as
k2

kz2 = µxx εyy k02 − x . (9.2)
µzz
The question of interest is whether the wave is propagating (kz2 > 0) or
evanescent (kz2 < 0). The boundary between the two cases may be easily
obtained from eqn (9.2). It occurs at kc , a critical value of the spatial
frequency, when
√
kx = kc = k0 εyy µzz . (9.3)
In eqn (9.2) there are two parameters, µxx εyy and µxx /µzz . We can
distinguish four cases:
(i) µxx εyy > 0 µxx /µzz > 0 , (9.4)

(ii) µxx εyy > 0 µxx /µzz < 0 , (9.5)
(iii) µxx εyy < 0 µxx /µzz > 0 , (9.6)
2
The authors called these four cases
(iv) µxx εyy < 0 µxx /µzz < 0 . (9.7) cutoff, anti-cutoff, never cutoff and al-
ways cutoff (a terminology shared by Li
It follows from eqn (9.2) that in case (i) there is propagation below et al. (2007)). Although the terminol-
kc , in case (ii) there is propagation above kc , in case (iii) there is prop- ogy is entirely appropriate we did not
agation for all values of the spatial frequency, and in case (iv) there is adopt it because we did not want to in-
no propagation for any value of the spatial frequency.2 troduce new definitions.
Case (iii) is particularly interesting. Propagation for all values of the

spatial frequency means that waves that are evanescent in air (i.e. for
which kx > kc ) can propagate in this medium. The evanescent waves
do not grow but do not decline either if the medium is lossless. Does
such a medium exist? It does. The material parameters of a set of SRRs
satisfy eqn (9.5). The longitudinal component of the permeability tensor
is negative, whereas µxx and εyy are positive. If, in addition, µxx = 1,
εyy = 1 and µzz = −1, then we obtain the nice hyperbola
kz2 − kx2 = k 2 , (9.8)

plotted in Fig. 9.9.
What is this medium good for? Smith et al. (2004b) proposed that
it would bring a diverging beam to a partial focus. Why? Let us draw
an input wave vector from the origin to the hyperbola. When it is
incident upon our indefinite medium the direction of the group velocity
kx
is obtained by drawing perpendiculars to the ω = const line, as has
already been explained in Section 8.1. The different focusing properties
of Veselago’s lens and that relying on the properties of indefinite media
are shown in Fig. 9.10. Figure 9.10(a) is the well-known ray diagram for
Veselago’s flat lens, whereas Fig. 9.10(b) was calculated by the authors
from the dispersion curves. It may be seen that the rays converge upon
a reasonably good focal point at the back of the slab. The electric-field
kz distribution in the slab given by the numerical package HFSS is given
in Fig. 9.11(a).
The experimental proof was provided by Smith et al. (2004b) in a
Fig. 9.9 A hyperbolic isofrequency 2D configuration in the microwave region at a frequency of 11.3 GHz,
curve in the 2D wave vector plane. Ar- known from previous experiments to yield a value close to −1 for the
rows show the incident and refracted µzz component. The elements themselves (SRRs) and the measuring
‘rays’
apparatus were the same as in the experiments of Shelby et al. (2001a).
The waves propagated in between parallel plates. All the SRRs were
oriented in the same direction with their axes perpendicular to the z
direction, the direction of propagation. The metamaterial was made up
by strips of three unit cells high and 32 unit cells long. The structure
was illuminated by a line current 2 cm away. The experimental results
measured just outside the slab are displayed in Fig. 9.11(b). There seems
to be good correlation between the experimental and numerical results,
although they are not shown in great detail.
We can now go back to Section 8.1 concerned with refraction of mag-
netoinductive waves. The configuration of interest is the one illustrated
in Fig. 8.3(b). It is the dispersion characteristics of the planar–axial
configuration of capacitively loaded loops shown in Fig. 8.3(a). The
similarity with the indefinite media dispersion comes from the fact that
the mutual inductance is positive in the axial direction and negative in
the planar direction. The dispersion curves for the magnetoinductive
waves are hyperbolic allowing the same construction as shown above in
Fig. 9.9. Thus, the whole argument about possible focusing could have
been presented in Chapter 8. We postponed the discussion to the present
section for the reason that we can show here the experimental results
(a) (b)
Fig. 9.10 A ray diagram showing the

x trajectories of rays emerging from a
point source for a slab of (a) negative-
index and (b) indefinite medium. From
z Smith et al. (2004b). Copyright c 2004
American Institute of Physics
of Smith et al. (2004b) thus lending more credence to the qualitative

argument.
Note that the arguments have been conducted in terms of rays that
have limited validity when the dimensions of the lens are comparable
with the wavelength. However, for this particular arrangement of SRRs
it is worth considering an alternative approach relaying on magnetoin-
ductive waves and on the generalized Ohm’s law
V = ZI , (9.9)
discussed several times in this book. Each element is then excited by
the current source and the currents in the elements may be determined
by assuming that each element is coupled to every other element. Such
an approach was attempted by Kozyrev et al. (2007). They were able
to find the focus in a narrow frequency band but no regular pattern at
different frequencies.
The general question of interest is when to use effective-medium theory
and when to resort to solution in terms of eigenwaves. The latter is more
accurate, but feasible only when the number of elements are not too large
(the limit posed by computational difficulties is around 104 ). For mag-
netically coupled elements like SRRs or capacitively loaded loops these
waves are magnetoinductive waves. There are of course other eigenwaves
as well, like the electroinductive waves of Beruete et al. (2006) and waves
propagating on an array of dipoles to be discussed in Section 9.6 con-
cerned with nanoparticles.
9.5 Gaussian beams and the

Goos–Hanchen shift
Gaussian beams propagating through a material with negative ε have
already been discussed in Chapter 5, where it was shown how an image
degrades (Shamonina et al., 2001). In negative-index materials the re-
Fig. 9.11 (a) Numerical simulation of

the distribution of the electric field.
The slab indicated by the solid lines has
εr = −1 and a diagonal permeability
tensor for which the longitudinal com-
ponent µz = −1 and µx = µy = 1. The
slab is 16 cm long, with a line source
placed 2 cm from the slab. The slab
thickness is 4 cm. (b) Experimentally
obtained field distribution in the plane
starting 4 cm away from the output of
the slab. From Smith et al. (2004b).
of Physics. For coloured version see
plate section (a) (b)
fraction and reflection of a Gaussian beam was treated by Kong et al.

2002a; Kong et al. 2002b.
A particularly interesting reflection of a finite beam under conditions
of total internal reflection is known as the Goos–Hanchen shift (Goos and
Hanchen, 1947). The reflected beam suffers a positive shift, as shown
n1>0 schematically in Fig. 9.12(a).
For an incident plane wave in the TM configuration we derived the
n2n1 positive lateral shift reflection coefficient in Section 1.6 in the form
(a) 1 − ζe
R= , (9.10)
1 + ζe
where ζe is dependent on the propagation coefficients on both sides of
the boundary. Since all finite beams may be constructed by the super-
position of infinite beams this means that the reflection coefficient is
n1>0
different for each plane-wave component. When the reflected beam is
reconstructed from its elements it turns out that the beam still keeps its
n2<0<n1 negative lateral shift shape but it is laterally shifted in the positive direction. This applies to
the normal case, that is when the refractive index on both sides of the
(b)
boundary is positive.
When the beam is incident from a positive-index material upon a
Fig. 9.12 Schematic representation of negative-index material then the same model predicts a negative shift
(a) a positive and (b) a negative lat- (Fig. 9.12(b)). This was shown practically simultaneously by Berman
eral shift of an incident beam upon re- (2002) and Lakhtakia (2003) both of them formulating the problem an-
flection from (a) a positive- and (b) alytically. A similar analytical approach was used by Ziolkowski 2003a;
a negative-index material. From Zi-
olkowski (2003a). Copyright c 2003 Ziolkowski 2003b who also did numerical calculations with the aid of the
Optical Society of America FDTD (finite-difference time domain) package. The frequency depen-
(a)
Fig. 9.13 Intensity distribution for a

Gaussian beam incident at 40◦ from a
medium with εr = 9, µr = 1 upon a
slab of (a) εr = 3, µr = 1 and (b) εr =
−3, µr = −1. From Ziolkowski (2003a).
Copyright c 2003 Optical Society of
America. For coloured version see plate
(b) section
dence of the permittivity and permeability, needed for this package, was
taken in the functional form of the Drude model.3 A Gaussian beam of 3
To become negative at a threshold
1-cm waist is incident at a frequency of 30 GHz from medium 1 upon frequency is a good approximation for
the permittivity. However, the known
medium 2. The material parameters of medium 1 are εr = 9 and µr = 1 negative-permeability materials do not
giving a refractive index of n1 = 3. Medium 2 is assumed to be a pos- follow this frequency dependence. The
itive index medium with εr = 3 and µr = 1 in the first case, and a permeability is negative within a band,
negative index medium with εr = −3 and µr = −1 in the second case. and not below a threshold value, as dis-
cussed many times in this book (see,
The incident angle is chosen to be 40◦ above the critical angle of 35.26◦, e.g., Section 2.8). Nevertheless, it
hence the beam in both cases will suffer total internal reflection. The re- seems likely that the model is good
flected and scattered beams yielded by the numerical package are shown enough to find the main features of the
in Figs. 9.13(a)) and (b), respectively. The direction of scattering may incident, scattered and reflected beams.
be immediately seen. For the positive-index material it is in the positive
direction and for the negative-index material it is in the negative direc-
tion. The actual shift is too small to appreciate it just by looking at the
figures. Ziolkowski (2003a) gives values of 32 cells for the positive shift
and 33 cells for the negative shift, both very close to the predictions of
his analytical model. The simulation space was 520 × 1040 cells. The
size of a cell was taken as 0.1 mm.
A large Goos–Hanchen shift was predicted theoretically by Shadrivov
et al. (2003) by having an extra layer between the positive- and negative-
index materials that can bring about the resonant excitation of surface
polaritons.
An interesting consequence of the negative Goos–Hanchen shift is that
in a negative-index waveguide, depending on the parameters of the three
media, propagation may be forward or backward, as shown in Figs.
x 13
A B
2
C D
x 12
(a)
x 13
2 A B A
B
x 12
(b)
x 13
Fig. 9.14 A waveguide consisting of 2 A B
a negative-index material bounded by x 12
two positive-index materials. Parame-
ters are chosen so that (a) the wave
is forward, (b) the wave is backward,
and (c) the wave is stationary. From
Tsakmakidis et al. (2007). Copyright
c 2008 Nature Publishing Group. For (c)
4 9.14(a) and (b) on the basis of a ray picture. It follows then (Tsak-
The authors called the resulting shape
an optical ‘clepsydra’, which according makidis et al., 2006; Tsakmakidis et al., 2007) that for some parameters
to the Oxford Dictionary is an ancient in between, the rays may bite their own tails: the propagation is neither
time-measuring device presumably sim- forward nor backward, it is stationary,4 as shown in Fig. 9.14(c). The
ilar to an hourglass. The pattern could
conclusion may then be drawn that such a device may stop and store
also be described as biconical.
light.
9.6 Waves on nanoparticles

It is doubtful whether the present section should be part of a book on
waves in metamaterials. Nanoelectronics-cum-nanophotonics is a new
subject that has made enormous strides in the last few years. The idea is
to manipulate surface plasmons with a view to optical signal processing.

Obviously we shall not be able to do more than scratch the surface.
We wish to scratch the surface at that particular point where analogies
emerge. In spite of the enormous difference in dimensions and in the
shape of the elements it turns out that waves on nanoparticles are very
similar to magnetoinductive waves. They are both based on coupling
between elements, they can both be treated by a method in which only
the near field and only nearest-neighbour interactions are taken into
account, and for both cases the theory can be formulated by taking into
account interactions between any two elements and determining that
interaction by including radiation terms as well. In fact, the ease with
which experiments on magnetoinductive waves can be realized might
serve as a testing ground for ideas emerging in the field of nanophotonics.
By nanoparticles we mean small metallic spheres, although other
shapes are not excluded either. We actually mentioned small metallic
triangles by Kottmann et al. 2000b; Kottmann et al. 2000a; Kottmann
and Martin 2001; Kottmann et al. 2001 and full rings by Aizpurua et al.
(2003) of the order of tens of nanometers in Chapter 4 among small res-
onant elements. A chain of spherical nanoparticles might consist of 50
elements, the diameter may be 50 nm and the distance between the el-
ements 75 nm. These were actually the parameters used by Quinten
et al. (1998) when they first investigated waves propagating along these
elements. Their motivation was to prove the potential of waveguides at
optical frequencies that are small relative to the wavelength and thus
they may be instrumental in founding a new type of integrated optics.
Other aspects of these waves, like switching (Brongersma et al., 2000),
splitting between longitudinal and transverse modes (Maier et al., 2002),
pulse propagation (Maier et al., 2003a), detection of electromagnetic en-
ergy (Maier et al., 2003b), multipoles (Park and Stroud, 2004), effect on
a channel waveguide (Quidant et al., 2004), coupling and 2D represen-
tations (Maier et al., 2005) have also been discussed.
Why can waves propagate on nanoparticles at all? The reason is the
presence of free electrons in the metal that can form electric dipoles,5 5
Multipoles as well but they only mat-
and two dipoles can interact in the same manner as two loops. We ter when the elements are extremely
close to each other so they rarely come
have actually referred to experiments on dipole arrays (see Fig. 2.8) into consideration.
conducted half a century ago (Shefer, 1963). In that case the wave
on the dipoles was excited by a horn antenna and it was detected by
another horn. Those dipoles could carry waves, although they were not
resonant. To have them resonant is even better and nanoparticles are
resonant indeed, belonging to the family of plasma resonances known
also as Mie resonances (see, e.g., Stratton 1941; Kreibig and Vollmer
1995).
Let us first work out their resonant frequency. It may be obtained
from a combination of the quasi-static model of Clausius–Mossotti with
the Drude model. The polarizibility of a metallic sphere is given by
εr (ω) − 1
αe = d3 . (9.11)
εr (ω) + 2
At resonance, the polarizability tends to infinity, i.e. it occurs when

εr = −2. Then, using the lossless Drude model
ωp2
εr (ω) = 1 − , (9.12)
ω2
we find that the resonant frequency is
ωp
ω0 = √ . (9.13)
3
The polarizability of the metallic sphere is then obtained by substituting
eqn (9.13) into eqn (9.11). We find
ω2

1 1
= 3 1− 2 . (9.14)
αe d ω0
Since in this section we intend to include radiation effects we should
add to the above equation the radiation damping term derived in Sec-
tion 2.7 for electric dipoles
k03

1
Im = . (9.15)
αe 6 πε0
How will this metallic sphere respond to an electric field? It will set
up an electric dipole. By definition, the dipole moment of the electric
dipole (see Section 1.16) is related to the electric field as
p = αe E . (9.16)
Let us now consider a linear chain consisting of N metallic nanospheres.
The aim is now the same as in Sections 2.9 and 8.2.2, where we derived
the dispersion characteristics of magnetoinductive waves based on the
concept of polarizability. We need two equations, the effect of the elec-
tric field on the polarization of the nth element, and the electric field
at the nth element produced by all the other elements known as IF , the
interaction function. The electric field due to an electric dipole has al-
ready been given in closed form in Section 1.12. Assuming that we have
N dipoles with an interelement spacing of a the field at the nth dipole
is given by summing up the individual contributions. From the require-
6
In eqn (9.17) we used our notation in- ment that αe IF = 1 the dispersion equation was obtained by Weber and
stead of that of Weber and Ford and
also changed to SI units. The disper-
Ford (2004) as
sion equation of Simovski et al. (2005)
is identical but, as mentioned in Sec-
k03 X

1 1 j 1
tion 8.2.2, they show that the imagi-
= − − e −j k0 na (9.17)
nary part of the equation cancels, and αe 2 πε0 nk0 a (nk0 a)2 (nk0 a)3
they also sum up some of the infinite
series.
for the electric dipoles transverse to the direction of propagation.6 It may
be seen that eqn (9.17) is entirely analogous to eqns (8.28)–(8.30) only
the magnetic polarizability should be changed to the electric polarizabil-
ity and µ0 to ε0 . This is not really surprising because in Section 8.2.2 we
used the magnetic dipole approximation to MI waves and in this section
the metallic nanospheres are regarded as electric dipoles.
Transverse Modes
1.05
Light line
Quasistatic
1.00 lossless metal
Real(Z/Z0 )
0.95
Full retarded solution

for lossless metal
0.90
Full retarded solution
with lossy metal
Fig. 9.15 Dispersion curves of
nanoparticle chains with and without
retardation in the transverse mode.
0.0 0.5 1.0 1.5 2.0 2.5 3.0 From Weber and Ford (2004). Copy-
c 2004 by the American Physical
right
kd Society
Weber and Ford also proceeded to compare the quasi-static case with
that containing full retardation. This is shown in Fig. 9.15, which is
analogous to Fig. 8.18. It needs to be noted that Weber and Ford find
the retarded solution not from the infinite series but from a finite num-
ber of elements with an imposed wave vector and assuming complex ω.
For the transverse case, when dipoles are perpendicular to the direc-
tion of propagation, the dispersion curves have a dip at the light line
not present in the longitudinal case, as we have already noted in the
previous chapter.
Another attempt at finding the dispersion equation was made by
Koenderink and Polman (2006). They also solved the dispersion equa-
tion for k imposed and working in the complex ω region. Their calcula-
tions took into account both ohmic and radiation losses. Their results
differ considerably from those of Weber and Ford in the vicinity of the
light line as shown in Fig. 9.16. The two branches, to the right and to
the left of the light line do not cross, which is, apparently, not unusual
for polaritons. Their dispersion curve has zeros only at the band edges,
in contrast to those of Simovski et al. (2005) that yield zero group ve-
locity at a particular value of ka within the Brillouin zone. Since in
both models there is some numerical work involved it is too early to say
whether the two models contradict each other.
5
Re( ) (1015 rad/s)
0
Fig. 9.16 Dispersion curves of
nanoparticle chains by Koenderink and 5
Polman (2006). Copyright c 2006
by the American Physical Society.
Thin solid (dotted) line, quasi-static 0
dispersion curve of the longitudinal
(transverse) mode. Squares (circles), 5 ine
ht l
infinite chain dispersion curves of the
Lig
longitudinal (transverse) mode. Full

circles, transverse mode dispersion
curve from ten elements according to 0 0.25 0.50 0.75 1
Weber and Ford (2004). Copyright c
2004 by the American Physical Society k (units /d)
Finally, we wish to mention a perturbation analysis by Citrin (2006)

who solved the dispersion equation with the aid of Clausen functions
to which the infinite sums reduce. He also considered the possibility
of reducing attenuation by embedding the nanoparticle chain in a gain
medium.
9.7 Refractive index close to zero

9.7.1 Introduction
We know that the refractive index is given by
√
n= εr µr . (9.18)
We have also seen that both ε and µ may have very small values so
we may end up with a very small refractive index. The consequences
were explored in a theoretical paper by Ziolkowski (2004). He assumed
that both ε and µ are zero but in such a manner that the characteristic
impedance η = (µ/ε)1/2 remains the same as for the adjoining medium.
He made the point that a zero-index medium has a static character in
space but the field magnitudes vary in time. He performed an FDTD
numerical study of a zero-index material flanked by two media of the
same characteristic impedance. He showed that during the rather long
transients the electric field does vary as a function of space. But when
the steady state is reached the field distribution in the adjoining finite-
index materials is the usual sinusoidal variation as a function of space
at a given moment in time but in the zero-index material there is no
spatial variation at all.
It would be difficult to produce a zero-index material, particularly
with the right characteristic impedance, but a low-index metamaterial
is well within the practical possibilities. The prominent effect would then
be on the refraction properties of the material. Let us remember Snell’s
9.7 Refractive index close to zero 307
law and the relationship between propagation constant and frequency.

They are
ω
n1 sin θ1 = n2 sin θ2 and k = n . (9.19)
c 3º
Let us explore the implications of both relationships. We have already
had a good look at refraction in Section 2.11. We found that for an
incident angle of θ1 from medium 1 the refracted ray can propagate in
ș2
any direction from +90◦ to −90◦, depending on the refractive index of
medium 2 (see Fig. 2.23), and of course for certain values of n2 there
can be total internal reflection. Now, the more interesting case arises
when medium 2 has a refractive index of unity and that of medium 1 is
0
close to zero, say, 0.05. Then, for an incident angle of θ1 , the refracted 0 90º
angle is given by ș1
(a)
sin θ2 = 0.05 sin θ1 . (9.20)
Plotting θ2 in Fig. 9.17(a) as a function of θ1 we can see that θ2 varies z

only between 0 and 3◦ , while θ1 covers the whole angular range from 0 x
to 90◦ . We shall discuss a potential application in Section 9.7.2. vacuum
Let us now look at the propagation coefficient. If the index of refrac-
tion is close to zero then k tends to zero as well and the wavelength metamaterial
in the medium tends to infinity. A low index means that the phase of
the electromagnetic field in this material changes very slowly. This is
vacuum
a major departure from normal practice. We shall discuss some of the
potential applications in Section 9.7.3.
(b)
9.7.2 Wavefront conversion Fig. 9.17 (a) Angle of refraction ver-

Let us assume now a source of electromagnetic waves embedded in a sus angle of incidence for waves incident
from a metamaterial with n = 0.05 in-
near-zero-index material close to the surface of a material with refraction cident upon vacuum with n = 1. (b)
index of 1. Figure 9.17(a) has shown the variation of θ2 with θ1 when Refraction of rays for the same two me-
n2 = 1 and n1 = 0.05. In Fig. 9.17(b) the corresponding ray picture is dia
shown. The output rays are very close to being parallel, which implies
a good directional radiation pattern.
The effect was demonstrated experimentally by Enoch et al. (2002)
at a frequency of 14.5 GHz using six sheets of a metallic grid as the
metamaterial. The source was a monopole antenna, the period in the
lattice was 5.8 mm and the total size 226 mm. The authors achieved a
directivity of 372.
The effect is interesting. Its suitability for producing a directional
antenna has been proven. It is unlikely, however, that it would be com-
petitive with existing antennas for two reasons. One is the problem of
matching. There is a large mismatch in the impedance of the two media
that is not easy to overcome. It is also doubtful that the directivity
achieved is particularly good by the standards of an antenna engineer.
The size of this antenna is 10.91 wavelengths in both directions. Accord-
ing to the well-known expression for the directivity of an antenna with
d PEC walls
a2
h
Fig. 9.18 Geometry of a 2D parallel- Hinc

plate waveguide with walls of perfect a1
electrical conductors and a 180◦ bend,
filled with a permittivity near zero ma-
terial. From Silveirinha and Engheta Einc
(2006). Copyright c 2006 by the
American Physical Society x=y=0
a plane wavefront, D = 4πA/λ2 , where A is the area, this directivity is

1497, way above 372.
The experiments of Bulu et al. (2005b) may or may not belong to this
category. They aimed to produce a directional antenna by a 3D array of
SRRs excited by a monopole antenna inside the structure. They report
increased directivity in the vicinity of the resonant frequency but the
corresponding value of µeff (needed if the mechanism is the same) is not
known. They compare the measured beamwidth (18◦ ) in the H plane
with that of a rectangular aperture of the same size (3.8 λ2 ). They find
that the measured beamwidth is smaller than the beamwidth obtained
by the uniform aperture. This part of their result may be controversial
because one needs a superdirective distribution in order to beat the
uniform one. It is also difficult to explain their measured radiation
pattern in the E plane. Probably further work is required to clarify the
radiation mechanism.
9.7.3 Effect of low phase variation

If the phase varies very slowly then it will ignore obstacles and dis-
continuities. An example, given by Silveirinha and Engheta (2006) is
shown in Fig. 9.18. Two 2D parallel plate waveguide are joined by an
abrupt 180◦ bend that is filled with an ε-near-zero material. Under
normal conditions a wave incident from either of the waveguides would
suffer high reflection. The authors showed using the numerical package
of CST MICROWAVE STUDIO that at certain frequencies near-perfect
transmission may be obtained.
7
Note the similarity between these In another example, Silveirinha and Engheta (2007) investigated the
simulations and the experimental re-
sults on propagation in waveguides con-
transmission through a double bend filled with split-pipe resonators in
ducted by Marques et al. (2002b) and which the slots are filled with a dielectric (see Fig. 9.19) so arranged that
Hrabar et al. (2005), discussed in Chap- the resulting effective index is near zero at a particular frequency. The
ter 6. authors have shown that near-perfect transmission may be achieved.7
Region 1
Region 3
PEC ring + dicl. gap
Region 2
s Fig. 9.19 Geometry of the 3D rectan-
gular metallic waveguide. The H-plane
width, s, is chosen so that region 3 be-
haves as a metamaterial with index of
refraction close to zero. Regions 1 and
3 have unity indices of refraction. From
Silveirinha and Engheta (2007). Copy-
right c 2007 by the American Physical
Society
^r
^
w ^
u
a
ac
e, m Fig. 9.20 Cross-section of a spherical
scatterer composed of two concentric
layers of different isotropic materials.
ec, mc From Alu and Engheta (2005). Copy-
c 2005 by the American Physical
right
e0, m0 Society
9.8 Invisibility and cloaking

Is it science fiction? The invisible man of H. G. Wells goes back over a
century. In more modern times Harry Potter, a budding magician, was
also able to acquire an invisibility cloak. Can it really be realized? If it
can, one could expect quite substantial orders from the various Ministries
of Defence. No general would want to be caught without some kind of
invisibility gadget.
In science, in contrast to science fiction, the problem was broached
recently by Alu and Engheta (2005) followed by a plethora of articles
by popular science writers (a good title among many others was: ‘Invis-
ibility cloaks are in sight’). The news even reached the general public
via many of the daily papers. A number of scientific publications fol-
lowed (Lee et al., 2005; Pendry et al., 2006; Cummer et al., 2006; Leon-
hardt, 2006; Leonhardt and Philbin, 2006; Leonhardt and Philbin, 2007;
Schurig et al., 2006; Silveirinha and Engheta, 2006; Alu and Engheta,
2006; Alu and Engheta, 2007a; Alu and Engheta, 2007b; Cai et al., 2007;
Uslenghi, 2007). So what are the chances? It is early days; too early to
7 -9 6 -3
x 10 x 10
6 5
5 a c = O /100 a c = O /10
Q s /O
2
Q s /O
4
2
4
3
3
2
2
1 1
0 0
0.5 06 07 08 09 10 05 06 07 08 09 10
a/a c a/a c
12
10
08 a c = O /5
Q s /O
2
06
04
02
00
00 02 04 06 08 10
a/a c
Fig. 9.21 Normalized scattering cross-section for a spherical element with three different sizes of the outer radius of the cover.
From Alu and Engheta (2005). Copyright c 2005 by the American Physical Society
say whether anything will come out of it.8

There is no doubt that the questions asked are interesting and the
arguments, that such effects are not impossible, quite convincing. We
shall summarize here briefly the basic principles and the state of affairs.
The basic idea of Alu and Engheta (2005) follows from studies on
scattering. How do we see an object? We know it is there because the
light scattered by it enters our eyes. If that scatter could be cancelled we
8 The desired effect might be qualitatively similar to that of superdirectivity in the

theory of antennas, which predicts that arbitrarily high directivities are possible in
principle. It was shown that superdirectivity is compatible with Maxwell’s equations
(Oseen, 1922), but the practical difficulties of realizing those antennas have been
enormous. They work in a very narrow frequency band, and their performance
is extremely sensitive to tolerances in realizing the required current distribution.
Similarly, perfect (or near-perfect) cloaking may be possible in principle but it will
work only in a very narrow frequency band and tolerance sensitivity could turn out to
be prohibitive. At the time of writing there is not much information about the effect
of non-ideal conditions (see though Cummer et al. (2006) later in this section). If
we do not quite succeed in making things invisible, what then? Can we measure the
degree of invisibility? If we say an object has become half as visible as it was before
we put on the cloak, what do we mean by this? In this respect superdirectivity comes
out better. We know what it means. Any increase in directivity over the classical
limit is welcome, whereas invisibility has to be very close to 100% to be of any value.
R1 R2
Fig. 9.22 Poynting vector streamlines

for an incident plane wave in a 2D con-
figuration. Cloak is designed so that
the electromagnetic power moves round
it. From Pendry et al. (2006). Copy-
right c 2005 AAAS
would not be able to see the object. Theoretical proof for this hypoth-
esis was provided by considering a spherical scatterer having materials
constants ε and µ. The scatter from this sphere is to be cancelled by
a coating with material constants εc and µc , shown in Fig. 9.20. The
mathematics is rather complicated: the scattered field must be writ-
ten as a sum of discrete spherical harmonics and it turns out that the
largest term (the dipolar one) can be cancelled. The final results can
be presented in a simple manner. In Fig. 9.21 the normalized scattering
cross-section is plotted as a function of a/ac , where a is the radius of the
sphere to be hidden and ac is the radius of the sphere with the coating
added. The parameters chosen are ε = 4ε0 , εc = −3ε0 , µ = µc = µ0 .
Note that the coating must have negative permittivity. It may be seen
from Fig. 9.21 that there is very good cancellation at a particular value
of the coating thickness when the object is well in the subwavelength
region (ac = λ0 /100 and λ0 /10, where λ0 is the free-space wavelength)
but much worse when the external radius is a fifth of the wavelength.
Thus, the idea works for subwavelength structures but, apparently, loses
its validity for larger objects. It is also true that, with scattering being
shape-dependent, each object requires a separate design for scattering
cancellation.
A different idea was proposed by Leonhardt (2006) and Pendry et al.
(2006). An object is made invisible because the rays of light do not
go through it but go around it, as shown schematically in Fig. 9.22 for
a simple case. Leonhardt (2006) starts with a disclaimer. He quotes
Nachman (1988) who proved that the inverse scattering problem has a
unique solution. If we measure the scattered field in all directions we
can uniquely determine the spatial variation of ε and µ that caused the
scattering. Hence, from our measurements we should always be able to
determine the distribution of the material parameters whether there is
a cloak there or not. Hence, there cannot be perfect invisibility. If our
measurements show that there is nothing there then the only possibil-
ity is that there is indeed no more than empty space there. However,
if we are allowed to talk about rays instead of waves then the situa-
tion changes. Waves diffract, rays are willing to bend if the index of
refraction varies spatially but they do not diffract. This is, of course,
the geometrical optics approximation. In that approximation, perfect
(a) (b)
electric field (arb. units)

(c) (d)
Fig. 9.23 Electric-field distribution for an incident plane wave in the vicinity of a perfectly conducting shell and a cloak. (a)
ideal parameters, (b) with a loss tangent of 0.1, (c) with an 8-layer approximation to the desired distribution of the material
parameters, (d) with a simplified cloak in which only µr is varying spatially. From Cummer et al. (2006). Copyright c 2006
by the American Physical Society. For coloured version see plate section
invisibility is possible in principle. If we know the desired trajectories

the required variation of the index of refraction may be obtained by a
conformal transformation in the 2D case. The bandwidth would be wide
if the index of refraction were to be independent of frequency. But it is
not independent, and that would limit the bandwidth. In fact, in order
to go round an object, the rays that do that must propagate faster than
the velocity of light (phase velocity only!). That is easy to realize; all
we need is a refractive index less than unity. But then there is strong
dispersion, hence a limitation of the bandwidth.
9
The COMSOL Multiphysics finite- A full wave numerical simulation9 of the 2D cylindrical problem, aimed
element-based electromagnetics solver at investigating deviations from the ideal case (Fig. 9.22) was performed
was used since it allows the specifica-
tion of both anisotropy and inhomo-
by Cummer et al. (2006). The wave incident was a TE wave at a fre-
geneity in the material parameters. quency of 2 GHz in a 2D configuration. The object that the rays are
supposed to circumnavigate is a thin shell of 0.2 m diameter made of a
perfect electric conductor (PEC). It was surrounded by the cloak with
10 outer diameter of 0.4 m in which the permittivity and permeability ten-
One feels, however, that it is much
easier for a ray to propagate round sors varied in a prescribed manner. The simulations were performed
the shell when the shell is impenetra- with the shell present but the authors claim that similar results were
ble anyway. A more testing experiment obtained in the absence of the shell.10 The results are displayed in Fig.
would use a dielectric cylinder as the
9.23 where the electric-field distribution is shown at a particular mo-
object to be hidden.
ment in time. The ideal situation may be seen in Fig. 9.23(a). The field
distribution appears to be unperturbed even in the vicinity of the cloak.
Figure 9.23(b) shows the effect of losses. For a loss tangent of 0.1 there
Fig. 9.24 2D microwave cloaking

structure with a plot of the material pa-
rameters implemented. µr (red line) is
multiplied by a factor of 10 for clarity.
µθ (green line) = 1, εz = 3.423. The
SRRs of cylinder 1 (inner) and cylin-
der 10 (outer) are shown in expanded
schematic form. From Schurig et al.
(2006). Copyright c 2006 AAAS. For
is quite a lot of distortion of the forward scattered wave, indicating that

loss may be the greatest obstacle to be overcome. For a layered solution
consisting of eight discrete homogeneous layers (Fig. 9.23(c)) the field
distribution looks good again. Figure 9.23(d) shows the case when in-
stead of the ideal materials distribution a simpler one, which is easier to
realize, is chosen. Again, the distortion of the field distribution does not
look bad. As said before, these are early times. We have no objective
criterion for the measure of invisibility but it is gratifying to know that
the results look encouraging.
An experimental investigation of cloaking was carried out by Schurig
et al. (2006) in a 2D geometry (the waves were confined between parallel
metallic plates) at a frequency of 8.5 GHz. The inner shell is made of
copper and the cloak by a set of thin cylindrical metamaterial shells. The
element used is shown in the inset of Fig. 9.24. It is the same as that
of Fig. 4.24(b) apart from a little rounding of the corners. There are 10
discrete layers. The design value of the permittivity is εz = 3.423 εθ . The
permeabilities µr and µθ are functions of the radius. The experimental
results in the absence and in the presence of cloaking are shown in Figs.
9.25(a) and (b). It may be seen that the cloak helps to hide the copper
shell. The constant electric field lines on the right are straighter in the
presence of the cloak, hence the object is then less visible.
A similar cloak for a 2D geometry was proposed by Cai et al. (2007)
for TM polarization in which only the permittivity needed to be varied.
In their design the cloak was made of a dielectric host into which metal
wires were embedded, as seen in Fig. 9.26.
(a) (b)
Fig. 9.25 Experimental field distribution for a copper cylinder (to be hidden) and a ten-layer cloak made up by resonating
c 2006 AAAS. For
elements. (a) in the absence and (b) in the presence of the cloak. From Schurig et al. (2006). Copyright
dielectric host
metal wires
Fig. 9.26 Section of the cloak contain-

ing metallic wires for the radial change
of the dielectric constant, suitable for r1
r2
the optical region. From Cai et al.
(2007). Copyright c 2007 Nature Pub-
lishing Group
In conclusion, we can only repeat what was said before. This is an

interesting research topic; whether it will lead to substantial degrees
of invisibility remains to be seen (well, as much as one can ever see
invisibility).
A historical review
10
It is probably too early, the subject is too young and not very well
defined, for writing an all-embracing review of the history of metamate- 10.1 Introduction 315
rials. Our aim here is much more modest. We shall make am attempt at 10.2 Forerunners 317
finding the various antecedents leading to the birth of the subject and 10.3 ... and the subject went on
and flourished... 323
enumerate the seminal papers instrumental in establishing the initial
momentum. A large number of closely related topics will be described
that contributed to the process of cross-fertilization and led to sustained
interest and further rapid development.
10.1 Introduction
Some births are more difficult than others. Pallas Athene is reputed
to have sprung out of Zeus’ head, and Aphrodite from the foam of the
sea. Other births have been known to be less instantaneous and to have
required the simultaneous efforts of a number of parents. Metamateri-
als belong to that category. It had a long gestation period and many
contributors. Before going into specifics and enumerating the various
prior influences it might be worth having a broader look at Physics that
has often been accorded a privileged position in the ranks of natural sci-
ence. We may even go further and quote Ernest Rutherford saying that
‘Physics is the only science: the rest is stamp collecting.’ This is not
quite true in the twenty-first century. Many other disciplines have been
using rigorous criteria for proving their theses. Nonetheless, we could
claim with good conscience that Physics is still the discipline farthest
away from the art of stamp collecting.
When we come to physics-in-the-making it is no longer a rational
subject. We do not know which hypotheses are useful and which are
useless. We do not know whether any particular approach to solving a
problem is feasible or not. We do not know the direction that further
research should take. We can only guess. It is necessary of course to
say nowadays that the new research we have in mind will benefit society
but we have no criteria to determine what benefits society and what
does not. Let us quote Rutherford again. In 1933, in his speech to the
British Association for the Advancement of Science, he claimed: ‘We
cannot control atomic energy to an extent which would be of any value
commercially, and I believe we are not likely ever be able to do so.’
Atomic energy provides as much as three quarter of French electrical
energy generation at the moment so, at least in the opinion of some
French authorities, the continuing research on atomic energy bore fruits.
316 A historical review
According to some other opinions, coming with hindsight, research on

atomic energy should have been banned from the very beginning. We
just do not know.
The sole purpose of the arguments presented so far is to emphasize how
difficult is to know the direction research should take. So how does a new
research topic arise? What will determine whether it fails or flourishes?
Coming now to our specific question: why is there a new research direc-
tion in Physics that is known under the generic term of metamaterials
(or electromagnetic metamaterials if we wish to exclude their association
with other branches of Physics)? How has it come about? Has it been
suggested by those in high administrative positions? Has it come about
by grass-root pressure? Have there been some random factors at play?
Have there been some seminal papers opening the floodgates? Yes, we
can say with some certainty that there were a number of seminal papers.
How many? Four, we think. The first one that needs to be included is
the one by Viktor Veselago (1968) who asked the speculative question:
what happens in a material when both the electric permittivity and the
magnetic permeability are negative? It would lead both to a backward
wave and to a negative refractive index, he suggested. The latter was,
no doubt, a suggestion that should have grabbed the headlines. For cen-
turies people believed in a positive index of refraction and suddenly, it
was suggested that it might be negative. In addition, Veselago claimed
that radiation emerging from a point source on one side of a negative-
index slab could be brought to a focus on the other side. He postulated
a flat lens, a possibility never considered before. His paper was pub-
lished in 1967 in Russian, and a year later in English. Did it open the
floodgates? It did not. Strangely enough it lay dormant until discovered
1
There was actually an earlier paper by Smith et al. (2000) three decades later.1 They designed an artificial
written by Sivukhin, published in 1957, material that, in the same frequency band, could produce both negative
which could have been discovered in-
stead. Sivukhin considers the case
permittivity and negative permeability. They showed experimentally
when both materials constants are neg- that electromagnetic waves cannot propagate in a medium in which one
ative and concludes that such medium of the materials constant is negative but propagation is restored when
would propagate a backward wave. they are both negative. But if they are both negative then, according to
Veselago, the refractive index is negative and hence they should be able
to observe negative refraction. And that was indeed what they did next
(Shelby et al., 2001a). They sent an electromagnetic wave at a frequency
of 10.5 GHz upon a negative-index prism and found that refraction was
in the expected but highly unusual direction. It was on the other side
of the perpendicular to the boundary. The angle of refraction obeyed
Snell’s law, provided the index of refraction was taken negative.
Would the papers Smith et al. 2000 and Shelby et al. 2001a have
been sufficient to open the floodgates? We do not know, but probably
not. The existence of negative refractive index was no doubt sensational
(much more sensational than talking about negative permittivity and
permeability) but it lacked practical applications. That missing require-
ment was provided by the fourth seminal paper, Pendry’s perfect lens
(2000). This was not about beating the classical limit. There have been
many publications on that. This was about the possibility of perfect
imaging, of being able to reproduce not only the travelling waves asso-
ciated with the object but all the evanescent waves as well. The four
papers taken together acted as a catalyst. Papers started to pour in.
Metamaterials was not an accepted term yet but everyone knew which
the fundamental papers were. People agreed and disagreed. The large
majority agreed, with a few dissenting voices.
Our aim in this chapter is not to review the birth of the subject in
any detail. It is rather to point out a number of crucial springboards
that kept up the momentum after the initial upsurge in popularity. The
seminal papers were a necessary condition for launching the subject,
the springboards were responsible for the continued interest. In the
next section we shall enumerate a selection of topics that were alive and
flourishing at the time of the upsurge and could be regarded as various
forerunners of the subject of metamaterials.
10.2 Forerunners
10.2.1 Effective-medium theory
An early success of materials science was to be able to describe in
terms of macroscopic quantities the response of a real material (one
that consists of atoms and molecules) to electric and magnetic fields.
Every textbook in solid state physics devotes some attention to early
attempts at homogenization and provides an expression for the permit-
tivity (or permeability) of a material, known as the Clausius–Mossotti
equation. These attempts, although made in the nineteenth century
(Mossotti, 1850; Clausius, 1879), still give inspiration today. The theory
of Gorkunov et al. (2002) relies on similar arguments. Another example
is the paper by Belov and Simovski (2005a) that discusses homogeniza-
tion in metamaterials including the radiation term. The arguments are
presented as generalizations of the Clausius–Mossotti equation.
It is also worth mentioning an early example by Lewin (1947) of the
calculation of the effective permittivity and permeability of a medium
loaded with spherical particles. The approach was taken up in the meta-
material context by Holloway et al. (2003).
10.2.2 Negative permittivity

The plasma frequency was defined, and plasma oscillations were found
experimentally by Tonks and Langmuir in 1929. Considering then the
Drude model, it was shown (see, e.g., Jackson 1967) that the permit-
tivity associated with a plasma can be negative below the plasma fre-
quency. This was a simple calculation based on Maxwell’s equations and
on the equation of motion. The aim was to simplify the treatment by
introducing a macroscopic quantity that could explain a host of physical
phenomena, the transparency of alkali metals for example at ultraviolet
wavelengths.
The need of radar technology for higher-permittivity low-loss mate-

rials resulted in the development of artificial dielectrics (for a compre-
hensive survey see Collin (1991)). One of the structures studied was
an array of thin wires that were shown to have an effective plasma fre-
quency by Brown (1953). Later motivation came from the desire to
simulate plasmas in order to have more insights into problems like the
effect of rocket exhaust upon the radiation of re-entry vehicle antennas.
The comprehensive paper of Rotman (1962) was a result of that inves-
tigation. Similar calculations reaching similar conclusions were done by
Pendry et al. (1996) a quarter of a century later. The choice of thin wires
for producing negative permittivity by Smith et al. (2000) and Shelby
et al. (2001a) was inspired by Rotman (1962) and Pendry et al. (1996).
10.2.3 Negative permeability

In his search for negative refractive index materials Veselago (1968) was
thinking about gaseous or solid-state plasmas (Chynoweth and Buchs-
baum, 1965) (apparently, unaware of the potential of thin wires) for neg-
ative permittivity. He also proposed anisotropic gyroscopic substances
in which, due to off-diagonal elements in the permeability tensor, the
effective permeability may be negative for one of the two circular po-
larizations of an electromagnetic wave. Interestingly, unknown to Vese-
lago, the existence of negative permeability was already shown in such
a material (a ferrite with an applied dc magnetic field placed in a cir-
cular hollow metallic waveguide) by Thompson 1955; Thompson 1963 a
decade earlier. More recent experiments on a metamaterial array in a
hollow waveguide by Marques et al. (2002a) yielded similar results.
10.2.4 Plasmon–polaritons
The interaction of plasmas with electromagnetic waves led both to a
new term, plasmon–polaritons and to a variety of phenomena both in
the bulk and on surfaces (see, e.g., Cottam and Tilley 1988; Mahan and
Obermair 1969). In fact, the so-called amplification of evanescent waves
described by Pendry (2000) is due to the excitation of a surface plasmon–
polariton on the far surface of the subwavelength lens. Recent studies
by Kempa et al. (2005) and by Belov and Simovski (2006) that have
their inspiration in the field of metamaterials are based on the earlier
work of Mahan and Obermair (1969).
10.2.5 Backward waves

Backward waves were discussed as early as the beginning of the twentieth
century. Schuster (1904) showed that negative refraction takes place at
the boundary of two media, one supporting a forward wave and the
other one a backward wave. Four decades later, Mandelshtam (1945)
discussed the same idea but the theory and practice of backward waves
remained on the whole of little interest to physicists. Most of the work
on backward waves was done after the Second World War by electronic
engineers concerned with device applications. A number of successful
devices like backward-wave oscillators, amplifiers (Beck, 1958; Hutter,
1960) and antennas (Walter, 1965) made their appearance.
Veselago (1968) noted in his original paper the backward-wave charac-
ter of the electromagnetic waves in negative refractive index materials,
which he called left-handed media. Lindell et al. (2001) proposed that
to avoid confusion with chiral materials it would be more logical to in-
troduce the term backward-wave media. Unfortunately, their proposal
was not generally accepted. At the time of writing a plethora of terms
prevail.
10.2.6 Theory of periodic structures

We may say that the theories of Mossotti (1850) and Clausius (1879)
were already part of the theory of periodic structures. The subject
was just becoming popular towards the end of the nineteenth century
(see, for example, Lord Rayleigh’s article (1892) on the properties of
a regular array). Man-made periodical structures also made their ap-
pearance. The first one was probably that of Lippmann (1894) who
produced a standing-wave structure by interference techniques in a pho-
tographic emulsion, and used it for producing colour photographs. The
major part of the theory was developed in the first three decades of the
twentieth century related to X-ray diffraction by crystals. Some of the
pioneering papers were written by Friedrich et al. (1912), Darwin (1914)
and Ewald (1916). A comprehensive treatment devoted entirely to the
theory of wave propagation in periodic structures was formulated some-
what later by Brillouin (1953). Part of this theory, that of periodically
loaded transmission lines, was found particularly useful for describing
wave propagation in metamaterials (see, e.g., Eleftheriades et al. 2002;
Grbic and Eleftheriades 2003a; Caloz and Itoh 2004; Lai et al. 2004;
Syms et al. 2005a; Engheta et al. 2005).
10.2.7 Resonant elements small relative to the

wavelength
Metallic boxes of the order of the wavelength were well known as mi-
crowave resonators. The problem of producing smaller resonators arose
when space was at a premium and there were specific requirements like
providing a narrow interaction region for electrons. The first such el-
ement was probably the re-entrant cavity used in klystrons (Hansen,
1939) but many others were developed later, examples being the split-
ring resonator of Hardy and Whitehead (1981) and the loop-gap res-
onator of Froncisz and Hyde (1982). The split-ring resonator of Pendry
et al. (1999) used in the experiments of Smith et al. (2000) and Shelby
et al. (2001a) was a variation on the same theme.
10.2.8 Chiral materials

They affect the polarization of the incident electromagnetic wave. The
first artificially produced chiral material was probably that of Bose
(1898). The first report of negative permeability by Thompson (1955)
was based on experiments involving ferrites. A microstructured chiral
element was proposed by Svirko et al. (2001). A more general proposal
for a chiral route to metamaterials was made by Pendry (2004). Another
chiral element of interest was studied recently by Bai et al. (2007).
10.2.9 Faster than light

The first lens based on rays going faster than the velocity of light was
realized by Kock (1964). He used hollow metal waveguides that can
propagate waves at any phase velocity up to infinity, obtaining lenses
that were concave for converging beams and convex for diverging beams.
Metamaterials can also reach such phase velocities by having a dielec-
tric constant less (or much less) than unity. Silveirinha and Engheta
(2006) showed how the waves in such a material can go round obsta-
cles, while Pendry et al. (2006) laid down design rules for low dielectric
constant, inhomogeneous metamaterial cloaks that can make an object
inside invisible.
10.2.10 Frequency filters made of periodically

arranged resonant elements
These represented a class of frequency filters particularly suitable for mi-
crowaves. They are briefly described by Atabekov (1965) in his textbook
of 1965. In their more modern form they were reported by Hong and
Lancaster 1996a; Hong and Lancaster 2000. The experimental work of
Martin et al. (2003b) was also directed at producing filters. They used
split-ring resonators inserted into coplanar waveguides.
10.2.11 Slow-wave structures

This was the term employed for a particular set of periodic structures
used in microwave tubes. The structures slowed down the electromag-
netic wave so that they could interact with electron beams drifting at
the same velocity (Beck, 1958; Hutter, 1960; Bevensee, 1964; Silin and
Sazonov, 1966). Similar kinds of structures were also used in particle
accelerators (Knapp et al., 1965) and in magnetic resonance imaging,
as realized by bird-cage resonators (Leifer, 1997). In the metamater-
ial context these waves were resurrected by Shamonina et al. 2002a;
Shamonina et al. 2002b by introducing magnetoinductive waves. For
an application of the interaction between precessing dipoles and magne-
toinductive waves see Solymar et al. (2006).
10.2.12 Waves arising from nearest-neighbour

interactions
The classical example of such waves, acoustic waves including the opti-
cal branch, have been discussed in most textbooks on solid state physics
and periodic structures (see, e.g., Brillouin 1953). They were derived for
coupled waveguides by Syms 1986; Syms 1987, for coupled optical res-
onators by Yariv et al. (1999), for coupled nanoparticles by Brongersma
et al. (2000), another set of forerunners of magnetoinductive waves.
10.2.13 Superdirectivity, superresolution,

subwavelength focusing and imaging
It was believed for a long time that for the radiation of a sharp beam
it is necessary to have a large aperture. Similarly, both a large aperture
and high illumination angle were regarded as necessary conditions for
producing a sharp focus by a lens. The latter was (and still is) known
as the Rayleigh criterion, giving the classical limit that the focal region
must be of the order of the wavelength. The theorem that claims that
an arbitrarily sharp beam can be produced by a finite aperture is due to
Oseen (1922). The resulting radiation pattern is known as an example
of superdirectivity. The term superresolution was introduced by Toraldo
di Francia (1952) who offered practical methods for tailoring the aper-
ture distribution in order to beat the classical limit. High-resolution
near-field imaging was first proposed by Ash and Nicholls (1972) who
relied on the near-field leaking from a microwave cavity through a small
hole. Near-field imaging in the optical region is now a major subject
running under the acronym of SNOM (scanning near-field optical mi-
croscopy), see, e.g., Paesler and Mayer (1996). In the metamaterials
context Pendry’s proposal for a ‘perfect’ lens (Pendry, 2000) also relied
on the near field but the configuration was quite different and no scan-
ning was needed. There were though objections from people who disliked
one or other aspect of the proposal but those doubts were quickly dis-
posed of. The idea of subwavelength imaging was already in the air, and
Pendry’s proposal could very soon become part of conventional wisdom.
10.2.14 Inverse scattering

This is the problem of finding the scattering medium or the source when
the scattered fields are known. For an early paper see Imbriale and Mit-
tra (1970). For a more detailed account see Colton and Kress (1992). In
the metamaterials context the problem was to find the real and imag-
inary parts of the index of refraction by measuring the complex reflec-
tion and transmission coefficients of a slab of negative-index material,
see, e.g., Smith et al. (2002). One of the difficulties is that for the
same scattering coefficients there may be multiple solutions. There are
also questions whether the method can work when the number of ele-
ments per wavelength is not sufficiently large (Efros, 2004; Simovski and
Tretyakov, 2007).
10.2.15 Bianisotropy
The main questions arising are the properties of the elements and their
collective effect upon the propagation of electromagnetic waves; exactly
the same as those arising in the study of metamaterials. We could
actually argue that the subject of metamaterials simply swallowed that
of bianisotropy and related topics like chiral and bi-isotropic media. The
introduction of the omega particle by Saadoun and Engheta (1992),
treatment of chiral scatterers by Tretyakov et al. (1996) or more general
books like that of Lindell et al. (1994) have been very specific forerunners
of later studies. Examples are the study of the bianisotropic character
of split-ring resonators by Marques et al. (2002c) and the consideration
of chiral media for negative refraction by Pendry (2004).
10.2.16 Photonic bandgap materials

The subject was founded in the late 1980s (Yablonovitch, 1987; John,
1987) concerned with confining electromagnetic radiation within a cer-
tain bandwidth inside an artificially produced material. The term bandgap
came from analogy with electron confinement to specific bands in a solid.
The structure of the material had to be such as to produce Bragg re-
flection in every direction. A change towards concepts in metamaterials
came with the work of Notomi (2000) who showed that refraction-like
behaviour is possible close to the bandgap. Further moves towards meta-
materials came later with the aim of showing negative refraction and
imaging (see, for example, Luo et al. 2002a; Lu et al. 2005).
10.2.17 Waves on nanoparticles

This topic may be regarded as another example of wave propagation in
slow-wave structures because resonant elements placed closely to each
other relative to the wavelength are involved. The element in this case
(Quinten et al., 1998; Brongersma et al., 2000) is a small metallic particle
of the
√ order of tens of nanometers whose resonant frequency is equal to
ωp / 3, where ωp is the plasma frequency of the metal. The motivation
is to produce optical circuits. In fact, this nanoparticle array is quite
analogous to the array of capacitively loaded loops (Shamonina et al.,
2002a; Shamonina et al., 2002b) if retardation terms are included in the
latter. The equations are practically the same, although the frequencies
of operation differ by nine orders of magnitude. In the metamaterials
context Zhuromskyy et al. (2005b) determined the dispersion character-
istics of magnetoinductive waves including the effect of radiation while
chains of nanoparticles were studied, e.g., by Alu and Engheta (2006)
and Alitalo et al. (2006).
10.3 ... and the subject went on and flourished... 323
10.3 ... and the subject went on and

flourished...
For those interested in any of the seventeen topics mentioned above (and
a few more we must have missed) the advent of negative refractive index,
negative refraction and ‘perfect’ imaging provided an immediate chal-
lenge. Will these physical phenomena lead to new physics and possibly
to some new applications? That was the question asked by many. The
beauty of this new field was that it had so many different aspects and
a very low barrier to entry. Reading the four seminal papers Veselago
1968; Smith et al. 2000; Shelby et al. 2001a; Pendry 2000 anyone doing
research (or just having an interest) in any of the topics mentioned must
have been tempted to become a metamaterialist (a term that has not
been coined yet but would be quite an appropriate one).
We know what happened. Scores of people entered the field, contri-
butions poured in leading to an exponential increase in the usual mea-
sures of research activities (scientific papers and books (Tretyakov, 2003;
Eleftheriades and Balmain, 2005; Itoh and Caloz, 2005; Engheta and
Ziolkowski, 2006; Marques et al., 2008; Markos and Soukoulis, 2008),
number of groups involved, number of citations, grants received, articles
in popular science journals, etc.) bringing forth the necessity of baptism
(Sihvola, 2007). By now there is a consensus that all the phenomena
associated somehow with wave propagation on resonant elements and
with negative refraction should be known under the generic term of
metamaterials.
And the subject went on and flourished.
Acronyms
A
AANR (all-angle negative refraction)
BC-SRR (broadside-coupled split-ring resonator)
CDE (coupled dipole equations)
CLS (capacitively loaded strips)
CMM (composite metamaterial)
CRLH (composite right/left-handed)
CRR (closed-ring resonator)
CSRR (complementary split-ring resonator)
DNG (double negative)
DNM (double-negative material)
DPS (double positive)
DSDR (doubly split double ring)
DSR (double-spiral resonator)
DSRR (deformed split-ring resonator)
D SRR (double-slit split-ring resonator)
EC-SRR (edge-coupled split-ring resonator)
EEMR (electric excitation of magnetic resonance)
EFC (equal-frequency contour)
ENG (epsilon negative)
ENZ (epsilon near zero)
EVL (epsilon very large)
LH (left handed)
LHH (left-handed heterostructure)
LHM (left-handed media)
MI (magnetoinductive)
MIW (magnetoinductive wave)
MM (metamaterial)
MNG (mu negative)
MNP (metal nanoparticle)
MPP (magnetic plasmon–polariton)
MTM (metamaterial)
NB SRR (non-bianisotropic split-ring resonator)
NFPL (near-field perfect lens)
NIM (negative-index material)
NIR (negative-index refraction)
NMPM (negative-magnetic-permeability medium)
NNA (nearest-neighbour approximation)
NR (negative refractive)
NRI (negative refractive index)
326 Acronyms
NRIM (negative refractive index material)

NRM (negative-refraction medium)
OSRR (open split-ring resonator)
PLH (purely left handed)
PRH (purely right handed)
PRI (positive refractive index)
RCE (resonant conducting element)
RH (right handed)
RHM (right-handed medium)
SLPW (square-lattice photonic waveguide)
SNG (single negative)
SRR (split-ring resonator)
SSDR (singly split double ring)
SSSR (singly split single ring)
TLPW (triangular-lattice photonic waveguide)
VM (Veselago material)
Field at the centre of a
cubical lattice of identical
dipoles
B
In the usual derivation of the Clausius–Mossotti formula (given in Sec-
tion 2.3) it is claimed within the validity of the quasi-static approxima-
tion that the field (electric or magnetic) due to a cubic lattice of dipoles
(electric or magnetic) within a spherical volume is zero at the centre of
the sphere. A similar problem arises in Section 2.8 where the mutual
inductances need to be summed.
We show here the derivation for a magnetic dipole starting with the
expression (see Section 1.12)
1 2

H= 5 3(r · m)r − r m , (B.1) m
4πµ0 r
which is the magnetic equivalent of the electric dipole expression given
r
in eqn (1.79) valid for the static case. The vector r connects the point
where the magnetic dipole is, to the point of observation, which is the
(0, 0, 0) point. m is the magnetic dipole moment, which is assumed to
have only a z component (Fig. B.1). (0,0,0)
Next, we shall find from eqn (B.1) the Cartesian components of the
magnetic field at the (0, 0, 0) point in the r, θ, ϕ spherical co-ordinate Fig. B.1 Geometry for calculating the
system, where r is the radial, θ is the elevation and ϕ is the azimuthal magnetic field at the point (0, 0, 0) due
co-ordinate. We obtain to a magnetic dipole at the general
point (r, θ, ϕ)
3mz
Hx = sin θ cos θ cos ϕ , (B.2)
4πµ0 r3
3mz
Hy = sin θ cos θ sin ϕ , (B.3)
4πµ0 r3
mz
Hz = (3 cos2 θ − 1) . (B.4)
4πµ0 r3
We wish to find the magnetic field at (0, 0, 0) due to all the magnetic
dipoles within a sphere of radius R. For each component we need then to
sum the contribution from each element. In a number of textbooks (see,
e.g., Jackson 1967) the field at (0, 0, 0) is summed in a rectangular co-
ordinate system, which is the logical choice for a cubic lattice. We shall,
in a less rigorous manner, keep the spherical co-ordinates and instead of
328 Field at the centre of a cubical lattice of identical dipoles
summation integrate over the spherical volume. This method will easily
lead to the answer that all three components are zero. The integration
is of the form
1 R π 2π
Z Z Z
hHi i = Hi dV , (B.5)
V 0 0 0
where i = x, y, z, V is the volume of the sphere and
dV = r2 sin θdrdθdϕ . (B.6)

It may be immediately seen that hHx i and hHy i must be zero because
sin ϕ and cos ϕ integrate out to zero. For hHz i an elementary integration
in θ needs to be performed that shows again that the total field is zero
in the z direction as well.
If the total field in the z direction is zero at the position of the dipole
at (0, 0, 0) the sum of all the mutual inductances within the sphere must
be zero as well. Hence, the summation term in eqn (2.53) is zero for the
cubic lattice. It turns out to be finite for other lattice configurations.
Derivation of material
parameters from reflection
and transmission
coefficients
C
It was claimed by Koschny et al. (2004d) that both the real and the
imaginary parts of the material parameters ε and µ can be deduced from
measurements of the complex S parameters, S11 and S21 . They further
pointed out that when the permeability has a resonant behaviour (e.g.
for SRRs) then the permittivity will have an antiresonant behaviour: the
signs of ε′′ and µ′′ will be of the opposite sign, and vice versa when the
permittivity is resonant. The authors argued that although the wrong
sign of the imaginary part of a material parameter leads to negative loss,
the total loss, when both ε′′ and µ′′ are taken into account will still be
positive, so there is no contradiction.
Opposite signs of ε′′ and µ′′ did not seem to bother many of the
authors. It is found acceptable in a number of publications (Smith
and Schurig, 2003; O’Brien and Pendry, 2002; O’Brien et al., 2004;
Smith et al., 2004a; Huang et al., 2004; Katsarakis et al., 2004; Kat-
sarakis et al., 2005). However, the approach was criticized by Depine
and Lakhtakia (2004) and Efros (2004) to which Koschny et al. (2004c)
replied. For a reworking and refining of the retrieval method see Smith
et al. (2005).
The procedure will always give the correct material parameters in
the sense that a plane wave incident upon that lattice from that par-
ticular direction and having those material parameters would yield the
same scattering coefficients. But that is all. If the angle of incidence is
different or the lattice has more or fewer elements then those material
parameters no longer lead to the correct result.
The difficulties due to the insufficient number of elements per wave-
length were known to Drude (1959). The approximation of sharp bound-
aries is then no longer tenable because the phase shift of the fields over
the lattice period cannot be neglected, the position of the boundary
cannot be unambiguously defined. Hence, we do not know the plane
at which the values of the material parameters suddenly change from
that of one medium to that of the other. Drude’s solution was to intro-
duce a thin transition layer in which the material parameters gradually
330 Derivation of material parameters from reflection and transmission coefficients
change. The Drude theory was further developed by Simovski et al.

2000a; Simovski et al. 2000b. In the paper by Simovski and Tretyakov
(2007) the problem of homogenization for insufficient number of ele-
ments (but still far from the Bragg condition) is resolved by introducing
a refined method for extracting the material parameters and introducing
two transition layers. The values of parameters derived in this way are
valid in whatever form (propagating, evanescent or wave packet) and at
whatever angle the waves are incident. Examples of this procedure for
interacting loaded wires have been presented by Ikonen et al. (2007a).
How does surface charge
appear in the boundary
conditions?
D
Let the value of the dielectric displacement be D1 and D2 in the adjoin-
ing media 1 and 2. The boundary condition to satisfy is then
Dn1 − Dn2 = 0 , (D.1)

where the subscript n refers to the component normal to the boundary.
We have shown in Section 1.9 that in a plasma we can define an effective
dielectric constant in the form
!
ωp2
εeff = ε0 1 − 2 . (D.2)
ω
If medium 1 is a plasma and medium 2 is a dielectric of dielectric
constant εr2 then eqn (D.1) may be written as
!
ωp2
ε0 1 − 2 En1 + ε0 εr2 En2 = 0 , (D.3)
ω
where En1 and En2 are the normal components of the electric field in
the two media. The above equation may be rewritten in the form
ωp2
ε0 En1 + ε0 εr2 En2 = ε0 En1 . (D.4)
ω2
The aim of this Appendix is to point out the different physical inter-
pretations of eqns (D.3) and (D.4). According to eqn (D.4) the difference
of the normal components of the dielectric displacement in the two me-
dia is equal to the surface charge density which may be shown to be
equal to ε0 (ωp2 /ω 2 )En1 . Equation (D.3) tells a different story. On the
left-hand side the dielectric constant used is the effective dielectric con-
stant and there is nothing on the right-hand side. Hence, we no longer
need to talk of surface charge.
Thus, all depends on the formulation of the problem. If we say that the
plasma’s relative dielectric constant is unity then we need to introduce
the surface charge in the boundary condition. If we use the effective
dielectric constant (as in eqn (D.2)) then we don’t need to worry about
surface charge at all.
The Brewster wave
E
Taking the positive sign in eqn (3.15) we find the dispersion curve of the
Brewster (Boardman, 1982; Welford, 1991) wave plotted in Fig. E.1(a)
together with
√ the surface plasmon dispersion curve that extends from 0
to ωs = ωp / 2. The upper branch looks similar but not the same as the
bulk plasmon dispersion curve. The curve intersects the y axis at the
same point, ωp , but the asymptote
√ is different. As ω → ∞ we find that
in the present case kx → k0 / 2, in contrast to kx → k0 for the bulk
wave.
The properties of this wave stem from the condition that it needs to
be incident at the Brewster angle. There is then no reflected wave and
we could look at the wave as moving along the dielectric–metal interface
with a propagation coefficient kx . Note that this is not a proper surface
wave. It radiates. It may be called a radiative wave or a leaky surface
wave. It refracts into the metal and propagates in the metal above
the plasma frequency. The relative dielectric constant of the metal is
between 0 and 1.
The condition for no reflection may be obtained from eqn (1.48) as
kz2 ε1
ζe = = 1. (E.1)
kz1 ε2
This is an unusual formulation of the condition of no reflection. Text-
books give it as
r
ε2
tan ϕ1 = , (E.2)
ε1
where ϕ1 is the angle of incidence. Since it is not obvious that the two
expressions are identical we shall show it below. Using the relationships
kz1 = k1 cos ϕ1 and kz2 = k2 cos ϕ2 , (E.3)

we find
kz2 ε1 cos ϕ2
ζe = = , (E.4)
kz1 ε2 sin ϕ1
where ϕ2 is the angle of refraction. But from Snell’s law
r
ε1
cos ϕ2 = 1 − sin2 ϕ1 = sin ϕ1 , (E.5)
ε2
with which ζe = 1.
334 The Brewster wave
2 2 2
Brewster
mode
ZZp
1 1 1
SPP mode
0 0 0
0 1 2 3 4 0 1 2 3 4 0 1 2 3 4
kx /k p kx /k p kx /k p
(a) (b) (c)
Fig. E.1 Dispersion curve of the surface wave. (a) no loss; Brewster mode for ω > ωp and SPP mode for ω < ωs . (b) γp = 0.01.
√
(c) γp = 0.1. Asymptotes: the light line ω/k = c (dotted line) and ω/k = c 2 (dashed-dotted line)
It needs to be noted that kz1 and kz2 are now both positive and
real, and the same applies to the dielectric constants. Can the Brewster
wave be regarded as a surface wave? The justification is that the same
pattern moves along the surface undistorted by reflection. Alternatively
we may argue that the two different types of waves leading to the same
dispersion equation may be construed as one being obtained from the
condition that there is no incident wave (Section 1.10), and the other
one (in the present appendix) demanding that the reflected wave should
vanish.
Losses of surface plasmons have already been considered in
Section 3.3.2 and plotted in Fig. 3.6(a) and (b). We shall replot them in
Figs. E.1(b) and (c) for γp = 0.01ωp and 0.1ωp including this time the
upper branch. It may be seen that the effect of losses is to connect the
lower branch and the upper branch. The stop band between ωs and ωp
has now become a rather odd-shaped pass band (Arakawa et al., 1973;
Alexander et al., 1974; Kovener et al., 1976) showing a backward wave
between points where the group velocity gives the (physically impossi-
ble) value of infinity. The mode has very high attenuation, so it has no
practical significance. The upper branch lying to the left of the light
line is not much affected by loss but, as mentioned above, it is radiative.
We shall not consider it further since our interest is in guided waves.
The electrostatic limit
F
The electrostatic limit is sometimes invoked when a proper (meaning all
of Maxwell’s equations) formulation seems too daunting or one is after
a simple analytical solution. But very often it is part of a gradualist
approach. One may argue: An electrostatic approach is easier than a
proper one. If it leads to a solution that makes good sense physically and
if, in addition, it gives reasonable agreement with experimental results
then one might call off the chase for better results and has every right to
be contented. In the absence of experimental results there is, however,
a danger that what makes good physical sense is not necessarily true.
An example is Pendry’s contention that for the ‘perfect lens’ to operate
there is no need to have a negative permeability; it is sufficient to have
εr = −1. This is discussed in more detail in Chapter 5.
What do we mean by the electrostatic limit? It is a kind of hotch-
potch. We say that things vary with time but not a lot, so we are entitled
to ignore the time derivatives of the electric and magnetic fields. Having
neglected time derivatives the electric and magnetic fields got uncoupled
so we can rely on electric quantities alone. Our starting equations are
∇ × E = 0 , ∇ · D = 0 , D = εE . (F.1)
If the curl of the electric field is zero then it can be expressed as the
gradient of a scalar function
E = −∇ϕ . (F.2)
Then, if the dielectric constant is independent of space the equation to
solve is Laplace’s equation
∇2 ϕ = 0 . (F.3)
F.1 Single interface

We shall now attempt the solution under the above approximations of
the problem of surface waves propagating along the interface of two
different media (Fig. F.1). The solutions in media 1 and 2 are assumed
in the form
ϕ1 = A e −j kx x + κ1 z (F.4)
and
336 The electrostatic limit
ϕ2 = B e −j kx x − κ2 z . (F.5)
z The assumptions are the same sort as in Section 1.10 where we discussed
r2
x surface waves. The waves are declining away from the interface in both
r1 media. The difference is that instead of solving the wave equation we
now solve Laplace’s equation. The time variation in the form of exp(j ωt)
is tacitly assumed but made no use of.
Substituting our assumed solution into Laplace’s equation we readily
Fig. F.1 Two semi-infinite dielectrics obtain
kx2 = κ21 = κ22 . (F.6)
r1
Next, we need to satisfy the boundary conditions at the interface that
z both the function ϕ and its derivative in the x direction must be con-
d r2 tinuous across the boundary, and that, in addition,
x
r1
dϕ1 dϕ2
εr1 = ε r2 . (F.7)
dz z=0 dz z=0
The result is that the electric fields decay nicely away from the interface
Fig. F.2 Thin dielectric slab between
at the same rate and the condition
two identical semi-infinite dielectrics
εr1 = −εr2 (F.8)
needs to be satisfied. This last condition may be seen to be much more
restrictive than those given in Section 3.3.1.
In order to find the dispersion equation ω(kx ) we have to substitute
1
This is, of course, a contradiction be- the frequency dependence of εr2 into eqn (F.8), which is1 equal to
cause this particular dependence on the
frequency comes about by taking into ωp2
account the time derivative of the elec- εr2 = 1 − (F.9)
tric field that we were supposed to have ω2
neglected. Nevertheless, these approxi- into eqn (F.8) leading to
mations very often lead to good results.
ωp
ω= √ , (F.10)
1 + εr1
the equation (3.17) quoted in Section 3.3.1. It is a straight horizontal
line that gives good approximation, provided kx is large enough.
F.2 Symmetric slab

We shall now go one step further and find the dispersion equation of
surface waves when medium 2 with dielectric constant εr2 is sandwiched
between two semi-infinite media of dielectric constant εr1 (Fig. F.2). The
solutions in the three media are assumed in the form
ϕ1 = B e −j kx x + κz , (F.11)
ϕ2 = C e −j kx x − κz + D e −j kx x + κ(z − d) , (F.12)
ϕ3 = F e −j kx x − κ(z − d) . (F.13)
F.2 Symmetric slab 337
Again, Laplace’s equation is there to be solved. The boundary conditions

to be satisfied are the same as in the previous case. The ϕ functions, their
derivatives and the normal component of the dielectric displacement
must be continuous across the boundaries. It is a rather long slog to
find the condition under which a solution exists but eventually it is
found as
εr1 + εr2
= ± e −kx d . (F.14)
εr1 − εr2
Substituting again for εr2 from eqn (F.9) we obtain the dispersion equa-
tion as
1 ± e −kx d
ω 2 = ωp2 , (F.15)
(εr1 + 1) ± (εr1 − 1) e −kx d
which in air simplifies to
ω 2 = ωp2 1 ± e −kx d ,

(F.16)
the expression quoted in eqn (3.37). The approximation is very good
for large kx . How good it is otherwise is discussed in Section 3.4.1 and
shown in Figs. 3.18 and 3.19.
Alternative derivation of
the dispersion equation for
SPPs for a
dielectric–metal–dielectric
structure: presence of a
surface charge
G
We derived the dispersion equation of a metal (plasma) slab by assuming
infinitely large transmission, i.e. taking the denominator of eqn (1.76)
equal to zero. In this appendix we derive the same dispersion equation
in a somewhat different manner. It is not a radically different approach:
it just puts the emphasis on the fields instead on the reflection and
transmission coefficients as it was done before. In addition, this approach
will easily lead to an analytic expression for the surface charge.
We have already given the field quantities both in the dielectric and in
the metal in Section 1.10. That was for a single interface. We shall write
them below for the case when the dielectric extends from z = −∞ to z H1
z = −d/2, and again from z = d/2 to z = ∞, the metal being between z d/2
x H2
then in the range z = −d/2 to z = d/2 (see Fig. G.1). We shall restrict z d/2
here generality and assume a solution in which the transverse electric H1
field is a symmetric function of z. The field quantities, as solutions of
Maxwell’s equations, may then be obtained for the TM mode as follows:
In the dielectric for z < −d/2
Fig. G.1 Dielectric–metal–dielectric
structure

d
κ1 z +
Hy1 = Hd e 2 e −j kx x , (G.1)

d
κ1 κ1 z +
Ex1 = − Hd e 2 e −j kx x , (G.2)
j ωε1

d
kx κ1 z +
Ez1 = − Hd e 2 e −j kx x , (G.3)
ωε1
where Hd is a constant. In the metal for −d/2 < z < d/2
340 Dispersion equation for SPPs: presence of a surface charge
Hy2 = Hm cosh(κ2 z) e −j kx x , (G.4)

κ2
Ex2 = − Hm sinh(κ2 z) e −j kx x , (G.5)
j ωε2
kx
Ez2 = − Hm cosh(κ2 z) e −j kx x , (G.6)
ωε2
where Hm is a constant. Note that we have adhered here to the notations
of Section 1.10 where the dielectric is referred to as medium 1 and the
metal as medium 2. In eqns (G.5) and (G.6) ε2 is the effective dielectric
constant of the plasma which for completeness we shall give again below.
It is
!
ωp2
ε2 = ε0 1 − 2 . (G.7)
ω
Next, let us satisfy the boundary conditions for the tangential com-
ponents of the electric and magnetic fields, i.e. match Hy and Ex at
z = −d/2. We obtain

κ2 d
Hd = Hm cosh (G.8)
2
and

κ1 Hd κ2 κ2 d
− = Hm sinh , (G.9)
ε1 ε2 2
whence the dispersion equation is

κ2 d ε 1 κ2
coth + = 0, (G.10)
2 ε 2 κ1
in agreement with eqn (3.36).
It was already pointed out in Appendix D that when we introduce an
effective dielectric constant there is no need for taking into account the
surface charge. All we need is to satisfy the boundary conditions using
the effective value of the dielectric constant. However, if we wish to
determine the surface charge it can be easily obtained from the equation
̺s = ε2 Ez2 − ε1 Ez1 , (G.11)

which with the aid of eqns (G.3) and (G.6) comes to
ωp2

kz κ2 d
̺s = Hm 2 cosh . (G.12)
ω ωp − ω 2 2
The surface charge may also be obtained in a less formal manner by
invoking the underlining physics. Clearly, the surface charge will wax
when the transverse current feeds it and will wane when the transverse
current leads it away. Hence, the temporal rate of change of the surface
current density must be equal to the transverse current, or
341
j ω̺s + Jz = 0 . (G.13)
The relationship between the current density and the electric field
was one of the first few equations derived (see Section 1.2). Substituting
that into eqn (G.13) and solving it for ̺s we find that it agrees with that
obtained from eqn (G.12).
Talking of surface charge it is worth mentioning that if a transverse dc
voltage is applied across a metal–dielectric sandwich then a dc surface
charge will reside at the boundary. As a consequence there will be an
ac surface current as well which will alter the boundary conditions for
the magnetic field. For experiments and a discussion see Batke and
Heitmann 1984; Tilley 1988.
Electric dipole moment
induced by a magnetic field
perpendicular to the plane
of the SRR
H
A magnetic field will set up a current in the SRR, more correctly it will
set up currents both in the inner and outer rings that vary as a function
of angle. Consequently, there will be a voltage distribution between
the two rings that will lead to charges on the inner and outer surfaces.
Alternatively, we may argue that a necessary consequence of the inter-
ring capacitance are charges on the outer surface of the inner ring and
on the inner surface of the outer ring, as shown in Fig. H.1. The dipole
moment at a point ϕ is equal to qd, where q is the charge and d is the
separation of the rings. Due to symmetry only the x component of the
dipole moment will be different from zero, which is given by
px = qd sin ϕ . (H.1)
Hence, Px , the total dipole moment, is its integral between 0 and π
that needs to be multiplied by two because the other side of the ring y M
contributes the same amount (see Fig. H.1). Thus + _ + _
x + _ + _
z _ + _
Zπ +
Px = 2 px dϕ . (H.2)
0
Using further the q = CV relationship (C is the inter-ring capaci-

tance) and knowing the variation of the voltage between the rings as a
function of angle (the trigonometric functions discussed in Section 4.4) Fig. H.1 Charge distribution in a SRR
the integration can be performed. We also know the variation of the

voltage along the rings that is related to the value of Hz that induces
the currents. Thus, at the end we can relate the input magnetic field to
the electric dipole moment created. After completing all the algebraic
operations we obtain for the relevant element of the polarizability tensor
−1
ω02

αem
yz = −2j ωCπr03 1− 2 . (H.3)
ω
Let us recall what the above notation means: m and z stand for the
344 Electric dipole moment induced by a magnetic field perpendicular to the plane of the SRR
magnetic field in the z direction, and e and y mean that an electric

dipole is created in the y direction.
Calculations by Marques et al. (2002c), based on a simplified physical
picture, yield an expression very similar to eqn (H.3). The only difference
is that ω is replaced by ω02 /ω, which is practically the same as eqn (H.3)
because we are usually talking of a narrow-band device.
Average dielectric
constants of a multilayer
structure
I
Let us assume a 1D multilayer structure (as in Fig. 5.27) consisting of
thin layers of two materials alternating with dielectric constants of ε1 and
ε2 . Provided the layers are thin enough, we can regard this multilayer
structure as a single layer of a homogeneous but anisotropic material.
What will be then the values of the corresponding dielectric constants
in the two distinguishable directions, the axial and the transverse? In
both cases we shall rely on the boundary conditions to find some average
values.
In the axial, z, direction it is the normal component of the dielectric
displacement, Dz , that is conserved across the boundaries. The corre-
sponding values of E1z and E2z may be obtained as
Dz Dz
E1z = and E2z = . (I.1)
ε1 ε2
And the average electric field is
1 −1
ε1 + ε−1

hE1z i = 2 Dz , (I.2)
2
whence we can define the axial component of the permittivity tensor as
1 −1
ε−1 ε1 + ε−1

z = 2 . (I.3)
2
For the transverse case it is the tangential component of the electric field
that is conserved across the boundary. Hence, we may obtain the values
of the dielectric displacement as
D1t = ε1 Et and D2t = ε2 Et . (I.4)

Then, the average value of the dielectric displacement is
1
(ε1 + ε2 )Et ,
hDt i = (I.5)
2
whence the transverse component of the permittivity tensor is
1
εt = (ε1 + ε2 ) . (I.6)
2
Derivation of mutual
inductance between two
magnetic dipoles in the
presence of retardation
J
In Section 2.9 we determined the mutual inductance between two mag-
netic dipoles for the static case in the axial configuration. We shall
consider here the planar configuration and include retardation effects.
The starting point is the Hθ component as given by eqn (1.87)
m sin θ
Hθ = (1 + j k0 r − k02 r2 ) e −j k0 r . (J.1)
4πµ0 r3
To find the magnetic field of dipole 1 at the position of dipole 2 at a
distance a away we need to take θ = 90◦ . The flux exciting the second
dipole (regarded now as a small loop) is
Φ = πr02 µ0 Hθ . (J.2)
By definition
Φ
M= and m = µ0 πr02 I , (J.3)
I
where I is the current flowing in the loop. Hence, the mutual inductance
is
(πr02 µ0 )2 Hθ πµ0 r04

M= = (1 + j k0 r − k02 r2 ) e −j k0 r , (J.4)
m 4r3
a complex quantity.
References
Abdalla, M. A. Y., Phang, K., and Eleftheriades, G. V. (2005). A 0.13-

micron CMOS phase shifter using tunable positive/negative refractive
index transmission lines. IEEE Microw. Wireless Comp. Lett., 16, 705–
707.
Abeles, F. (1986). Surface plasmon (SEW) phenomena, Volume Elec-
tromagnetic surface excitations of Springer series on wave phenomena.
Springer Verlag.
Aizpurua, J., Hanarp, P., Sutherland, D. S., Kall, M., Bryant, G. W.,
and Garcia de Abajo, F. J. (2003). Optical properties of gold nanorings.
Phys. Rev. Lett., 90, 057401–1–4.
Al-Bader, S. J. (2004). Optical transmission on metallic wires-
fundamental modes. IEEE J. Quant. Electron., 40, 325–329.
Al-Bader, S. J. and Imtar, M. (1992). Azimuthally uniform surface-
plasma modes in thin metallic cylindrical shells. IEEE J. Quant. Elec-
tron., 28, 525–532.
Alexander, R. W., Kovener, G. S., and Bell, R. J. (1974). Dispersion
curves for surface electromagnetic curves with damping. Phys. Rev.
Lett., 32, 154–157.
Alitalo, P., Simovski, C., Viitanen, A., and Tretyakov, S. (2006). Near-
field enhancement and subwavelength imaging in the optical region
using a pair of two-dimensional arrays of metal nanospheres. Phys.
Rev. B , 74, 235425–1–6.
Alu, A. and Engheta, N. (2003). Pairing an epsilon-negative slab with a
mu-negative slab: resonance, tunneling and transparency. IEEE Trans.
Ant. Prop., 51, 2558–2571.
Alu, A. and Engheta, N. (2005). Achieving transparency with plas-
monic and metamaterial coatings. Phys. Rev. E , 72, 016623–1–9.
Alu, A. and Engheta, N. (2006). Theory of linear chains of metamater-
ial/plasmonic particles as subdiffraction optical nanotransmission lines.
Phys. Rev. B , 74, 205436–1–18.
Alu, A. and Engheta, N. (2007a). Dispersion characteristics of meta-
material cloaking structures. In Proc. 1st Int. Congr. on Advanced
Electromagnetic Materials in Microwaves and Optics (Metamaterials
2007), Rome, pp. 482–485.
Alu, A. and Engheta, N. (2007b). Plasmonic materials in transparency
and cloaking problems: mechanism, robustness, and physical insight.
Opt. Exp., 15, 3318–3332.
350 References
Antoniades, M. A. and Eleftheriades, G. V. (2003). Compact linear

lead/lag metamaterial phase shifters for broadband applications. IEEE
Ant. Wireless Prop. Lett., 2, 103–106.
Antoniades, M. A. and Eleftheriades, G. V. (2005). A broadband se-
ries power divider using zero-degree metamaterial phase-shifting lines.
IEEE Microw. Guided Wave Lett., 15, 808–810.
Arakawa, E. T., Williams, M. W., Hamm, R. N., and Ritchie, R. H.
(1973). Effect of damping on surface plasmon dispersion. Phys. Rev.
Lett., 31, 1127–1129.
Ash, E. A. and Nicholls, G. (1972). Super-resolution aperture scanning
microscope. Nature, 237, 510–512.
Atabekov, G. I. (1965). Linear network theory. Pergamon Press, Ox-
ford.
Aydin, K., Bulu, I., Guven, K., Kafesaki, M., Soukoulis, C. M., and
Ozbay, E. (2005). Investigation of magnetic resonances for different
split-ring resonator parameters and designs. New J. Phys., 7, 168–1–
15.
Aydin, K., Bulu, I., and Ozbay, E. (2007). Subwavelength resolution
with a negative-index metamaterial superlens. Appl. Phys. Lett., 90,
254102–1–3.
Aydin, K. and Ozbay, E. (2006). Identifying magnetic response of split-
ring resonators at microwave frequencies. Opto-Electron. Rev., 714,
193–199.
Aydin, K. and Ozbay, E. (2007a). Capacitor-loaded split ring res-
onators as tunable metamaterial components. J. Appl. Phys., 101,
024911–1–5.
Aydin, K. and Ozbay, E. (2007b). Experimental and numerical analyses
of the resonances of split ring resonators. phys. stat. sol. (b), 244, 1197–
2001.
Baena, J. D., Bonache, J., Martin, F., Marques, R., Falcone, F.,
Lopetegi, T., Laso, M. A. G., Garcia-Garcia, J., Portillo, M. F., and
Sorolla, M. (2005a). Equivalent-circuit models for split-ring resonators
and complementary split-ring resonators coupled to planar transmis-
sion lines. IEEE Trans. Microw. Theory Tech., 53, 1451–1461.
Baena, J. D., Jelinek, L., and Marques, R. (2005b). Reducing losses and
dispersion effects in multilayer metamaterial tunnelling devices. New
J. Phys., 7, 166–1–13.
Baena, J. D., Jelinek, L., Marques, R., and Medina, F. (2005c).
Near-perfect tunnelling and amplification of evanescent electromagnetic
waves in a waveguide filled by a metamaterial: Theory and experiment.
Phys. Rev. B , 72, 075116–1–8.
Baena, J. D., Jelinek, L., Marques, R., and Zehentner, J. (2006). Elec-
trically small isotropic three-dimensional magnetic resonators for meta-
material design. Appl. Phys. Lett., 88, 134108–1–3.
References 351
Baena, J. D., Marques, R., Medina, F., and Martel, J. (2004). Artificial
magnetic metamaterial design by using spiral resonators. Phys. Rev.
B , 69, 014402–1–5.
Bahr, A. J. and Clausing, K. (1994). An approximate model for artifi-
cial chiral media. IEEE Trans. Ant. Prop., 42, 1592–1599.
Bai, B., Svirko, Y., Turunen, J., and Vallius, T. (2007). Optical activity
in planar chiral metamaterials: Theoretical study. Phys. Rev. A, 76,
023811–1–12.
Barnes, W. L. (1998). Fluorescence near interfaces: The role of pho-
tonic mode density. J. Mod. Opt., 45, 661–669.
Barnes, W. L. (2006). Surface plasmon-polariton length scales: a route
to sub-wavelength optics. J. Opt. A: Pure Appl. Opt., 8, S87–S93.
Barnes, W. L., Dereux, A., and Ebbesen, T.W. (2003). Surface plasmon
subwavelength optics. Nature, 424, 824–830.
Batke, E. and Heitmann, D. (1984). Rapid-scan Fourier transform
spectroscopy of 2-D space charge layers in semiconductors. Infrared
Phys., 24, 189–197.
Beck, A. H. W. (1958). Space charge waves and slow electromagnetic
waves. Pergamon Press, New York.
Belov, P. A., Hao, Y., and Sudhakaran, S. (2006a). Subwavelength
microwave imaging using an array of parallel conducting wires as a
lens. Phys. Rev. B , 73, 033108–1–4.
Belov, P. A., Hao, Y., Sudhakaran, S., Alomainy, A., and Hao, Y.
(2006b). Experimental study of the subwavelength imaging by a wire
medium slab. Appl. Phys. Lett., 89, 262109–1–3.
Belov, P. A. and Silveirinha, M. G. (2006). Resolution of subwavelength
transmission devices formed by a wire medium. Phys. Rev. E , 73,
056607–1–9.
Belov, P. A. and Simovski, C. R. (2005a). On homogenization of elec-
tromagnetic crystals formed by uniaxial resonant scatterers. Phys. Rev.
E , 72, 026615–1–15.
Belov, P. A. and Simovski, C. R. (2005b). Subwavelength metallic
waveguides loaded by uniaxial resonant scatterers. Phys. Rev. E , 72,
036618–1–11.
Belov, P. A. and Simovski, C. R. (2006). Boundary conditions for
interfaces of electromagnetic crystals and the generalized ewald-oseen
extinction principle. Phys. Rev. B , 73, 045102–1–14.
Belov, P. A., Simovski, C. R., and Ikonen, P. (2005). Canalization of
subwavelength images by electromagnetic crystals. Phys. Rev. B , 71,
193105–1–4.
Belov, P. A., Simovski, C. R., and Tretyakov, S. A. (2002). Two-
dimensional electromagnetic crystals formed by reactively loaded wires.
Phys. Rev. E , 66, 036610–1–7.
Belov, P. A. and Zhao, Y. (2006). Subwavelength imaging at optical
frequencies using a transmission device formed by a periodic layered
352 References
metal-dielectric structure operating in the canalization regime. Phys.

Rev. B , 73, 113110–1–4.
Bergman, D. J. (1978). The dielectric constant of a composite material
- a problem in classical physics. Phys. Rep., 43, 377–407.
Berini, P. (1999). Plasmon-polariton modes guided by a metal film of
finite width. Opt. Lett., 24, 1011–1013.
Berini, P. (2000a). Plasmon-polariton modes guided by a metal film of
finite width bounded by different dielectrics. Opt. Exp., 7, 329–335.
Berini, P. (2000b). Plasmon-polariton waves guided by thin lossy metal
films of finite width: Bound modes of symmetric structures. Phys. Rev.
B , 61, 10484–10503.
Berini, P. (2001). Plasmon-polariton waves guided by thin, lossy metal
films of finite width: bound modes of asymmetric structures. Phys.
Rev. B , 63, 125417–1–15.
Berini, P. (2006). Figures of merit for surface plasmon waveguides.
Opt. Exp., 14, 13030–13042.
Berman, P. R. (2002). Goos-Hanchen shift in negatively refractive
media. Phys. Rev. E , 66, 067603–1–3.
Beruete, M., Freire, M. J., Marques, R., Falcone, F., and Baena, J. D.
(2006). Electroinductive chains of complementary metamaterial ele-
ments. Appl. Phys. Lett., 88, 083503–1–3.
Bevensee, R. M. (1964). Electromagnetic slow wave systems. John
Wiley and Sons Inc., New York.
Blaikie, R. J. and McNab, S. J. (2002). Simulation study of ‘perfect
lenses’ for nearfield optical nanolithography. Microelectron. Eng., 61-
62, 97–103.
Bloembergen, N. (2005). Nonlinear optics. World Scientific Publishing,
Singapore.
Boardman, A. D. (ed.) (1982). Electromagnetic surface modes. Wiley,
Chichester.
Boot, H. A. H. and Randall, J. T. (1976). Historical notes on the cavity
magnetron. IEEE Trans. ED, 2, 724–729.
Born, M. and Wolf, E. (1975). Principles of optics (5th edn). Pergamon
Press, Oxford.
Bose, J. C. (1898). On the rotation of plane of polarization of electric
waves by a twisted structure. Proc. Royal Soc., 63, 146–152.
Brillouin, L. (1953). Wave propagation in periodic structures. Dover,
New York.
Brongersma, M. L., Hartman, J. W., and Atwater, H. A. (2000). Elec-
tromagnetic energy transfer and switching in nanoparticle chain arrays
below the diffraction limit. Phys. Rev. B , 62, R16356–R16359.
Brown, J. (1950). The design of metallic delay dielectric. Proc.
IEE , 97(Part III), 45–48.
References 353
Brown, J. (1953). Artificial dielectrics having refractive indices less

than unity. Proc. IEE , 100(pt. 4), Monograph No. 62R, 51–625.
Bulu, I., Caglayan, H., Aydin, K., and Ozbay, E. (2005b). Compact
size highly directive antennas based on the SRR metamaterial medium.
New J. Phys., 7, 223–1–10.
Bulu, I., Caglayan, H., and Ozbay, E. (2005a). Experimental
demonstration of subwavelength focusing of electromagnetic waves by
labyrinth-based two-dimensional metamaterials. Opt. Lett., 31, 814–
817.
Burke, J.J., Stegeman, G. I., and Tamir, T. (1986). Surface-polariton-
like waves guided by thin, lossy metal films. Phys. Rev. B , 33, 5186–
5201.
Cai, W., Chettiar, U. K., Kildishev, A. V., and Shalaev, V. M. (2007).
Optical cloaking with Metamaterials. Nature Photonics, 1, 224–227.
Caloz, C. and Itoh, T. (2004). Transmission line approach of left-
handed (LH) materials and microstrip implementation of an artificial
LH transmission line. IEEE Trans. Ant. Prop., 52, 1159–1166.
Caloz, C. and Itoh, T. (2006). Electromagnetic metamaterials: Trans-
mission line theory and microwave applications. Wiley-Interscience,
Hoboken, NJ.
Caloz, C., Sanada, A., and Itoh, T. (2004). A novel composite right-
/left-handed coupled-line directional coupler with arbitrary coupling
level and broad bandwidth. IEEE Trans. Microw. Theory Tech., 52,
980–992.
Cartlidge, E. (2002). Negative reaction to negative refraction. Phys.
World , 8, 8–9.
Casey, J. P. and Lewis, E. A. (1952). Interferometer action of a parallel
pair of wire gratings. J. Opt. Soc. Am., 42, 971–977.
Chang, K. (1996). Microwave ring circuits and antennas. Wiley, New
York.
Chen, H., Ran, L., Huangfu, J., Grzegorczyk, T. M., and Kong, J. A.
(2006). Equivalent circuit model for left-handed metamaterials. J.
Appl. Phys., 100, 024915–1–6.
Chen, J. J., Grzegorczyk, T.M., Wu, B.-I., and Kong, J.A. (2005).
Limitations of FDTD in simulation of a perfect lens imaging system.
Opt. Exp., 13, 10840–10845.
Chynoweth, A. G. and Buchsbaum, S. J. (1965). Solid-state plasma
(Characteristics of solid state plasma conditions necessary for existence
of plasma in semiconductor or semimetal). Phys. Today, 18, 26–30, 32–
34, 36–37.
Citrin, D. S. (2006). Plasmon-polariton transport in metal-nanoparticle
chains embedded in a gain medium. Opt. Lett., 31, 98–100.
Clausius, R. J. E. (1879). Die mechanische Behandlung der Electricitat,
Volume Abschnitt III. F. Vieweg, Braunschweig.
354 References
Cohn, S. B. (1946). Analysis of the metal strip delay structure. J.

Appl. Phys., 20, 257–262.
Collin, R. E. (1991). Field theory of guided waves. Oxford University
Press, Oxford.
Colton, D. and Kress, R. (1992). Inverse acoustic and electromagnetic
scattering theory. Springer Verlag, Berlin.
Cottam, M. G. and Tilley, D. R. (1988). Introduction to surface and
superlattice excitations. Cambridge University Press, Cambridge.
Craven, G. (1972). Slim guide microwave components. Electrical Com-
munication, 47, 245–258.
Cui, T. J., Cheng, Q., Lu, W. B., Jiang, Q., and Kong, J.A. (2005). Lo-
calization of electromagnetic energy using a left-handed-medium slab.
Phys. Rev. B , 71, 045114–1–11.
Cummer, S. A., Popa, B.-I., Schurig, D., Smith, D. R., and Pendry, J.
(2006). Full wave simulations of electromagnetic cloaking structures.
Phys. Rev. E , 74, 036621–1–5.
Darmanyan, S. A., Neviere, M., and Zakhidov, A. A. (2003). Surface
modes at the interface of conventional and left-handed media. Opt.
Commun., 225, 233–240.
Darwin, C. G. (1914). The theory of X-ray reflection. Philos. Mag., 27,
Part I (6th Series). 315–333, Part 2. 675–690.
Dekker, A. J. (1965). Solid state physics. Macmillan, London.
Depine, R. A. and Lakhtakia, A. (2004). Comment on ‘Resonant and
antiresonant frequency dependence of the effective parameters of meta-
materials’. Phys. Rev. E , 70, 048601–1.
Dolling, G., Enkrich, C., Wegener, M., Zhou, J. F., Soukoulis, C. M.,
and Linden, S. (2005). Cut-wire pairs and plate pairs as magnetic
atoms for optical metamaterials. Opt. Lett., 30, 3198–3200.
Dorofeenko, V., Lisyansky, A. A., Merzlikin, A. M., and Vinogradov,
A. P. (2006). Full-wave analysis of imaging by the Pendry–Ramakrishna
stackable lens. Phys. Rev. B , 73, 235126–1–4.
Drude, P. (1959). The theory of optics (3rd edn). Dover, London.
Economou, E. N. (1969). Surface plasmons in thin films. Phys.
Rev., 182, 539–554.
Efros, A. L. (2004). Comment II on ‘Resonant and antiresonant fre-
quency dependence of the effective parameters of metamaterials’. Phys.
Rev. E , 70, 048602–1.
Eleftheriades, G. V. and Balmain, K. G. (2005). Negative refraction
metamaterials: Fundamental principles and applications. John Wiley
and Sons Inc., New York.
Eleftheriades, G. V. and Islam, R. (2007). Minaturized microwave
components and antennas using negative-refractive index transmission-
line (nri-tl) metamaterials. Metamaterials, 1, 53–61.
References 355
Eleftheriades, G. V., Iyer, A. K., and Kremer, P. C. (2002). Planar neg-

ative refractive index media using periodically LC loaded transmission
lines. IEEE Trans. Microw. Theory Tech., 50, 2702–2712.
Eleftheriades, G. V., Siddiqui, O., and Iyer, A. K. (2003). Transmission
line models for negative refractive index media and associated imple-
mentation without excess resonators. IEEE Microw. Wireless Comp.
Lett., 13, 31–33.
Engheta, N. (2002). An idea for thin subwavelength cavity resonators
using metamaterials with negative permittivity and permeability. IEEE
Ant and Wireless Prop. Lett., 1, 10–13.
Engheta, N., Nelatury, S. R., and Hoorfar, A. (2002). The role of geom-
etry of inclusions in forming metamaterials with negative permittivity
and permeability. In XXVII URSI-GA, Maastricht, Netherlands.
Engheta, N., Salandrino, A., and Alu, A. (2005). Circuit elements at
optical frequencies: Nanoinductors, nanocapacitors, and nanoresistors.
Phys. Rev. Lett., 95, 095504–1–3.
Engheta, N. and Ziolkowski, R. W. (ed.) (2006). Metamaterials:
Physics and engineering explorations. John Wiley and Sons Inc., New
York.
Enkrich, C., Perez-Willard, F., Gerthsen, D., Zhou, J., Koschny, T.,
Soukoulis, C. M., Wegener, M., and Linden, S. (2005a). Focused-ion-
beam nano-fabrication of near infrared magnetic metamaterials. Adv.
Mater., 17, 2547–2549.
Enkrich, C., Wegener, M., Linden, S., Burger, S., Zschiedrich, L.,
Schmidt, F., Zhou, J. F., Koschny, T., and Soukoulis, C. M. (2005b).
Magnetic metamaterials at telecommunication and visible frequencies.
Phys. Rev. Lett., 95, 203901–1–4.
Enoch, S., Tayeb, G., Sabouroux, P., Guerin, N., P., and Vincent
(2002). A metamaterial for directive emission. Phys. Rev. Lett., 89,
213902–1–4.
Ewald, P. P. (1916). Zur Begründung der Kristalloptik. Ann. d.
Phys., 49, 1–38 (Part 1). 1916; 49: 117–143 (Part 2). 1917; 54: 519–597
(Part 3).
Falcone, F., Lopetegi, T., Baena, J. D., Martin, F., Marques, R., and
Sorolla, M. (2004a). Effective negative ε stopband microstrip lines
based on complementary split ring resonators. IEEE Microw. Wireless
Comp. Lett., 14, 280–282.
Falcone, F., Martin, F., Bonache, J., Laso, M. A. G., Garcia-Garcia,
J., Baena, J. D., Marques, R., and Sorolla, M. (2004b). Stop-band
and band-pass characteristics in coplanar waveguides coupled to spiral
resonators. Microw. Opt. Technol. Lett., 42, 386–388.
Falcone, F., Martin, F., Bonache, J., Marques, R., Lopetegi, T., and
Sorolla, M. (2004d). Left handed coplanar waveguide band pass filters
based on bi-layer split ring resonators. IEEE Microw. Wireless Comp.
Lett., 14, 10–12.
356 References
Falcone, F., Martin, F., Bonache, J., Marques, R., and Sorolla, M.
(2004c). Coplanar waveguide structures loaded with split-ring res-
onators. Microw. Opt. Technol. Lett., 40, 3–6.
Fang, N., Lee, H., Sun, C., and Zhang, X. (2005). Sub-diffraction-
limited optical imaging with a silver superlens. Science, 308, 534–537.
Feng, S. and Elson, J. M. (2006). Diffraction-suppressed high-resolution
imaging through metallodielectric nanofilms. Opt. Exp., 14, 216–221.
Feng, S., Elson, J. M., and Overfelt, P. L. (2005). Optical properties of
multilayer metal-dielectric nanofilms with all-evanescent modes. Opt.
Exp., 13, 4113–4124.
Feng, Y., Zhao, J., Teng, X., Chen, Y., and Jiang, T. (2007). Sub-
wavelength imaging with compensated anisotropic bilayers realized by
transmission-line metamaterials. Phys. Rev. B , 75, 155107–1–6.
Fox, M. (2001). Optical properties of solids. Oxford University Press,
Oxford.
Fredkin, D. R. and Ron, A. (2002). Effectively left-handed (negative
index) composite material. Appl. Phys. Lett., 70, 1753–1755.
Freire, M. J. and Marques, R. (2005). A planar magnetoinductive lens
for 3D subwavelength imaging. Appl. Phys. Lett., 86, 182505.
Freire, M. J. and Marques, R. (2006). Near-field imaging in the mega-
hertz range by strongly coupled magnetoinductive surfaces: Experi-
ment and ab initio analysis. J. Appl. Phys., 100, 063105–1–9.
Freire, M. J. and Marques, R. (2008). Optimizing the magnetoinduc-
tive lens: Improvement, limits, and possible applications. J. Appl.
Phys., 103, 013115–1–7.
Freire, M. J., Marques, R., Medina, F., Laso, M. A. G., and Mar-
tin, F. (2004). Planar magnetoinductive wave transducers: theory and
applications. Appl. Phys. Lett., 85, 4439–4441.
French, O. E., Hopcraft, K. I., and Jakeman, E. (2006). Perturbation
on the perfect lens: the near-perfect lens. New J. Phys., 8, 271–1–12.
Friedrich, W., Knipping, P., and von Laue, M. (1912). Interferenz-
Erscheinungen bei Rontgenstrahlen. Sitzungsberichte der (Kgl.) Akad.
der Wiss. Bayer , 42, 303–322.
Froncisz, W. and Hyde, J. S. (1982). The loop-gap resonator: A new
microwave lumped circuit ESR sample structure. J. Magn. Res., 47,
515–521.
Gao, L. and Tang, C. J. (2004). Near-field imaging by a multi-layer
structure consisting of alternate right-handed and left-handed materi-
als. Phys. Lett. A, 322, 390–395.
Garcia, N. and Nieto-Vesperinas, M. (2002). Left-handed materials do
not make a perfect lens. Phys. Rev. Lett., 88, 207403–1–4.
Garcia-Garcia, J., Martin, F., Falcone, F., Bonache, J., Gil, I.,
Lopetegi, T., Laso, M. A. G., Sorolla, M., and Marques, R. (2004).
Spurious passband suppression in microstrip coupled line band pass fil-
References 357
ters by means of split ring resonators. IEEE Microw. Wireless Comp.

Lett., 14, 416–418.
Garwe, F., Rockstuhl, C., Etrich, C., Hubner, U., Bauerschafer, U.,
Setzpfandt, F., Augustin, M., Pertsch, T., Tunnermann, A., and Led-
erer, F. (2006). Evaluation of gold nanorod pairs as a potential negative
index material. Appl. Phys. B , 84, 139–148.
Gay-Balmaz, P., Maccio, C., and Martin, O. J. F. (2002). Microwire
arrays with plasmonic response at microwave frequencies. Appl. Phys.
Lett., 81, 2896–2898.
Gay-Balmaz, P. and Martin, O. J. F. (2002). Efficient isotropic mag-
netic resonators. Appl. Phys. Lett., 81, 939–941.
Ghim, B. T., Rinard, G. A., Quine, R. W., Eaton, S. S., and Eaton,
G. R. (1996). Design and fabrication of copper-film loop-gap resonators.
J. Magn. Res. Ser. A, 120, 72–76.
Gil, I., Bonache, J., Garcia-Garcia, J., and Martin, F. (2006). Tunable
metamaterial transmission lines based on varactor-loaded split-ring res-
onators. IEEE Trans. Microw. Theory Tech., 54, 2665–2674.
Gil, I., Garcia-Garcia, J., Bonache, J., Martin, F., Sorolla, M., and
Marques, R. (2004). Varactor-loaded split ring resonators for tunable
notch filters at microwave frequencies. Electron. Lett., 40, 1347–1348.
Gomez-Santos, G. (2003). Universal features of the time evolution of
evanescent modes in a left-handed perfect lens. Phys. Rev. Lett., 70,
077401–1–4.
Goos, F. and Hanchen, H. (1947). Ein neuer and fundamentaler Ver-
such zur Totalreflexion. Ann. Phys. Lpz., 1, 333–346.
Gorkunov, M., Lapine, M., Shamonina, E., and Ringhofer, K. H.
(2002). Effective magnetic properties of a composite material with
circular conductive elements. Eur. Phys. J. B, 28, 263–269.
Govyadinov, A. A. and Podolskiy, V. A. (2006). Metamaterial photonic
funnels for subdiffraction light compression and propagation. Phys.
Rev. B , 73, 155108–1–5.
Gray, S. K. and Kupka, T. (2003). Propagation of light in metallic
nanowire arrays: Finite-difference time-domain studies of silver cylin-
ders. Phys. Rev. B , 68, 045415–1–11.
Grbic, A. and Eleftheriades, G. V. (2002). Experimental verification of
backward wave radiation from a negative refractive index metamaterial.
J. Appl. Phys., 92, 5930–5935.
Grbic, A. and Eleftheriades, G. V. (2003a). Dispersion analysis of a
micro-strip-based negative refractive index periodic structure. IEEE
Microw. Wireless Comp. Lett., 13, 155–157.
Grbic, A. and Eleftheriades, G. V. (2003b). Growing evanescent
waves in negative-refractive-index transmission-line media. Appl. Phys.
Lett., 82, 1815–1817.
358 References
Grbic, A. and Eleftheriades, G. V. (2003c). Negative refraction, grow-

ing evanescent, and sub-diffraction imaging in loaded transmission-line
metamaterials. IEEE Trans Microw. Theory Tech., 51, 2297–2305.
Grbic, A. and Eleftheriades, G. V. (2003d). Periodic analysis of a 2-d
negative refractive index transmission line structure. IEEE Trans. Ant.
Prop., 51, 2604–2611.
Grbic, A. and Eleftheriades, G. V. (2004). Overcoming the diffrac-
tion limit with a planar left-handed transmission-line lens. Phys. Rev.
Lett., 92, 117403–1–4.
Grigorenko, A. N., Geim, A. K., Gleeson, H. F., Zhang, Y., Firsov,
A. A., Khruschev, I. Y., and Petrovic, J. (2005). Nanofabricated media
with negative permeability at visible frequencies. Nature, 438, 335–
338.
Grover, F. W. (1981). Inductance calculations: Working formulas and
tables. Instrument Society of America, Research Triangle Park, N.C.
Gundogdu, T. F., Tsiapa, I., Kostopoulos, A., Konstantinidis, G., Kat-
sarakis, N., Penciu, R. S., Kafesaki, M., Economou, E. N., Koschny, T.,
and Soukoulis, C. M. (2006). Experimental demonstration of negative
magnetic permeability in the far infrared frequency region. Appl. Phys.
Lett., 89, 084103–1–3.
Guo, Y., Goussetis, G., Feresidis, A. P., and Vardaxoglu, J. C. (2005).
Efficient modelling of novel uniplanar left-handed metamaterials. IEEE
Trans. Microw. Theory Tech., 53, 1462–1468.
Haldane, F. D. M. (2002). Electromagnetic surface modes at inter-
faces with negative refractive index make a ‘not-quite-perfect’ lens.
arxiv:cond-mat/0206420.
Hammond, P. and Sykulski, J. (1994). Engineering electromagnetism:
Physical processes and computation. Oxford University Press, Oxford.
Hansen, W. W. (1939). On the resonant frequency of closed concentric
lines. J. Appl. Phys., 10, 39–45.
Hansen, W. W. and Richtmeyer, R. D. (1939). On resonators suitable
for klystron oscillators. J. Appl. Phys., 10, 189–199.
Hao, T., Stevens, C. J., and Edwards, D. J. (2005a). Simulations
and measurements for 1D metamaterial elements. In Proc. SPIE-COO
Conf., Warsaw, Poland.
Hao, T., Stevens, C. J., and J.Edwards, D. (2005b). Optimisation of
metamaterials by Q factor. Electron. Lett., 41, 653–654.
Hardy, W. H. and Whitehead, L. A. (1981). Split-ring resonator for
use in magnetic resonance from 20-2000 MHz. Rev. Sci. Instrum., 52,
213–216.
Hesmer, F. (2008). unpublished.
Hessel, A. and Oliner, A. A. (1965). A new theory of Wood’s anomalies
on optical gratings. Appl. Opt., 4, 1275–1297.
Hohenau, A., Krenn, J. R., Schider, G., Ditlbacher, H., Leitner, A.,
Aussenegg, F. R., and Schaich, W. L. (2005). Optical near-field of mul-
References 359
tipolar plasmons of rod-shaped gold nanoparticles. Europhys. Lett., 69,

538–543.
Holloway, C. L., Kuester, E. F., Baker-Jarvis, J., and Kabos, P. (2003).
A double negative (DNG) composite medium composed of magnetodi-
electric spherical particles embedded in a matrix. IEEE Trans. Ant.
Prop., 51, 2596–2603.
Homola, J. (2003). Present and future of surface plasmon resonance
biosensors. Anal. Bioanal. Chem., 377, 528–239.
Hong, J. S. (2000). Couplings of asynchronously tuned coupled mi-
crowave resonators. IEE Proc. Microw. Ant. Prop., 147, 354–358.
Hong, J. S. and Lancaster, M. J. (1996a). Couplings of microstrip
square open loop resonators for cross-coupled planar microwave filters.
IEEE Trans. Microw. Theory Tech., 44, 2099–2109.
Hong, J. S. and Lancaster, M. J. (1996b). Theory and experiment of
novel microstrip slow-wave open-loop resonator filters. IEEE Trans.
Microw. Theory Tech., 45, 2358–2365.
Hong, J. S. and Lancaster, M. J. (1998). Cross-coupled microstrip hair-
pin resonator filters. IEEE Trans. Microw. Theory Tech., 46, 118–122.
Hong, J. S. and Lancaster, M. J. (1999). Aperture-coupled microstrip
open-loop resonators and their applications to the design of novel mi-
crostrip bandpass filters. IEEE Trans. Microw. Theory Tech., 47, 1848–
1855.
Hong, J. S. and Lancaster, M. J. (2000). Microstrip filters for
RF/microwave applications. Wiley-Interscience, New York.
Hrabar, S., Bartolic, J., and Sipus, Z. (2005). Waveguide miniaturiza-
tion using uniaxial negative permeability metamaterial. IEEE Trans.
Ant. Prop., 53, 110–119.
Hsu, Y.-J., Huang, Y.-C., Lih, J.-S., and Chern, J.-L. (2004). Elec-
tromagnetic resonance in deformed split ring resonatrs of left-handed
meta-materials. J. Appl. Phys., 96, 1979–1982.
Huang, K. C., Povinelli, M. L., and Joannopoulos, J. D. (2004). Nega-
tive effective permeability in polaritonic photonic crystals. Appl. Phys.
Lett., 85, 543–545.
Hutter, R. G. E. (1960). Beam and wave electronics in microwave tubes.
Van Nostrand, Princeton, NJ.
Ikonen, P., Saenz, E., Gonzalo, R., Simovski, C., and Tretyakov, S.
(2007a). Mesoscopic effective material parameters for thin layers mod-
eled as single and double grids of interacting loaded wires. Metamate-
rials, 1, 89–105.
Ikonen, P., Simovski, C., Tretyakov, S., Belov, P., and Hao, Y. (2007b).
Magnification of subwavelength field distributions at microwave fre-
quencies using a wire medium slab operating in the canalization regime.
Appl. Phys. Lett., 91, 104102–1–3.
Imbriale, W. A. and Mittra, R. (1970). The two-dimensional inverse
scattering problem. IEEE Trans. Ant. Prop., 18, 633–642.
360 References
Ishikawa, A., Tanaka, T., and Kawata, S. (2005). Negative magnetic

permeability in the visible light region. Phys. Rev. Lett., 95, 273401–
1–4.
Islam, R. and Eleftheriades, G. V. (2004). Phase-agile branch-line cou-
plers using metamateril lines. IEEE Microw. Wireless Comp. Lett., 14,
340–342.
Itoh, T. and Caloz, C. (2005). Electromagnetic metamaterials: Trans-
mission line theory and microwave applications. John Wiley and Sons
Inc., New York.
Iyer, K. A., Kremer, P. C., and Eleftheriades, G. V. (2003). Exper-
imental and theoretical verification of focusing in a large periodically
loaded transmission line negative refractive index metamaterial. Opt.
Exp., 11, 696–708.
Jackson, J. D. (1967). Classical electromagnetics. John Wiley and Sons,
New York.
Jacob, Z., Alekseyev, L. V., and Narimanov, E. (2006). Optical hy-
perlens: far-field imaging beyond the diffraction limit. Opt. Exp., 14,
8247–8256.
Jiang, S. H. and Pike, R. (2005). A full electromagnetic simulation
study of near-field imaging using silver films. New J. Phys., 7, 169–1–
20.
John, S. (1987). Strong localization of photons in certain disordered
dielectric superlattices. Phys. Rev. Lett., 58, 2486–2489.
Johnson, P. B. and Christy, R. W. (1972). Optical constants of the
nobel metals. Phys. Rev. B , 6, 4370–4379.
Kafesaki, M., Koschny, T., Penciu, R. S., Gundogdu, T. F., Economou,
E. N., and Soukoulis, C. M. (2005). Experimental demonstration of
negative magnetic permeability in the far infrared frequency region. J.
Opt. A: Pure Appl. Opt., 7, S12–S22.
Katsarakis, N., Konstantinides, G., Kostopoulos, A., Penciu, R. S.,
Gundogdu, T. F., Kafesaki, M., and Economou, E. N. (2005). Magnetic
response of split-ring resonators in the far infrared frequency region.
Opt. Lett., 30, 1348–1350.
Katsarakis, N., Koschny, T., Kafesaki, M., Economou, E. N., and Souk-
oulis, C. M. (2004). Electric coupling to the magnetic resonance of split
ring resonators. Appl. Phys. Lett., 84, 2943–2945.
Kempa, K., Ruppin, R., and Pendry, J. B. (2005). Electromagnetic
response of a point-dipole crystal. Phys. Rev. B , 72, 205103–1–6.
Kik, P. G., Maier, S. A., and Atwater, H. A. (2002). The perfect lens
in a non-perfect world. In Proc. Progress in Electromagnetics Research
Symposium (PIERS), Cambridge, MA, USA.
Kildishev, A. V., Cai, E., Chettiar, U. K., Yuan, H.-K., Sarychev,
A. K., Drachev, V. P., and Shalaev, V. M. (2006). Negative refractive
index in optics of metal-dielectric composites. J. Opt. Soc. Am. B, 23,
423–433.
References 361
King, R. W. P. (1969). The loop antenna for transmission and recep-

tion, Volume Antenna Theory - Part I. McGraw Hill Co., New York.
Kittel, C. (1953). Introduction to solid state physics. John Wiley, New
York.
Kittel, C. (1963). Quantum theory of solids. John Wiley, New York.
Klein, M. W., Enkrich, C., Wegener, M., Soukoulis, C. M., and Linden,
S. (2006). Single-split split-ring resonators at optical frequencies: limits
of size scaling. Opt. Lett., 31, 1259–1261.
Knapp, B. C., Knapp, E. A., Lucas, G. J., and Potter, J. M. (1965).
Resonantly coupled accelerating structures for high-current proton
Linacs. IEEE Trans. Nuclear Science, 12, 159–165.
Kock, W. E. (1964). Metal lens antennas. Proc. IRE , 34, 828–836.
Koenderink, A. F. and Polman, A. (2006). Complex response and
polariton-like dispersion splitting in periodic metal nanoparticle chains.
Phys Rev. B , 74, 033402–1–4.
Kolinko, P. and Smith, D. R. (2003). Numerical study of electromag-
netic waves interacting with negative index materials. Opt. Exp., 11,
640–648.
Kondrat’ev, I. G. and Smirnov, A. I. (2003). Comment on ‘Left-handed
media simulation and transmission of electromagnetic waves in sub-
wavelength SRR loaded waveguide’. Phys. Rev. Lett., 91, 29401–1.
Kong, J. A., Wu, B.-I., and Zhang, Y. (2002a). Lateral displacement
of a Gaussian beam reflected from a grounded slab with negative per-
mittivity and permeability. Appl. Phys. Lett., 80, 2084–2086.
Kong, J. A., Wu, B.-I., and Zhang, Y. (2002b). A unique lateral dis-
placement of a Gaussian beam transmitted through a slab with neg-
ative permittivity and permeability. Microw. Opt. Technol. Lett., 33,
136–139.
Korobkin, D., Urzhumov, Y., and Shvets, G. (2006a). Enhanced near-
field resolution in midinfrared using metamaterials. J. Opt. Soc. Am.
B , 23, 468–478.
Korobkin, D., Urzhumov, Y., Zorman, C., and Shvets, G. (2006b). Far-
field detection of the super-lensing effect in the mid-infrared: theory
and experiment. J. Mod. Phys., 52, 2351–2364.
Koschny, T., Kafesaki, M., Economou, E. N., and Soukoulis, C. M.
(2004a). Effective medium theory of left-handed materials. Phys. Rev.
Lett., 93, 107402–1–4.
Koschny, T., Kafesaki, M., Penciu, R., Katsarakis, N., Soukoulis, C. M.,
and Economou, E. N. (2004b). Qualitative behaviour of left-handed
materials and related metamaterials. In Proc. Progress in Electromag-
netics Research Symposium (PIERS), Pisa, Italy.
Koschny, T., Markos, P., Smith, D. R., and Soukoulis, C. M. (2004c).
Reply to Comment on ‘Resonant and antiresonant frequency depen-
dence of the effective parameters of metamaterials. Phys. Rev. E , 70,
048603–1.
362 References
Koschny, T., Markos, P., Smith, D. R., and Soukoulis, C. M. (2004d).

Resonant and antiresonant frequency dependence of the effective para-
meters of metamaterials. Phys. Rev. E , 68, 065602–1–4.
Koschny, Th., Moussa, R., and Soukoulis, C. M. (2006). Limits on the
amplification of evanescent waves of left-handed materials. J. Opt. Soc.
Am. B , 23, 485–489.
Kostin, M. V. and Shevchenko, V. V. (1993). Theory of artificial mag-
netic substances based on ring currents. Sov. J. Commun. Techn. Elec-
tron., 38, 78–83.
Kostin, M. V. and Shevchenko, V. V. (1994). Artificial magnetics based
on double circular elements. In Proc. Bianisotropics’94, CHIRAL ‘94
3rd Int. Workshop on Chiral, Bi-isotropic and Bi-anisotropic Media,
(ed. F. Mariotte and J. P. Parneix), Perigueux, France, pp. 49–56.
Kottmann, J. P. and Martin, O. J. F. (2001). Plasmon resonant cou-
pling in metallic nanowires. Opt. Exp., 8, 655–663.
Kottmann, J. P., Martin, O. J. F., Smith, D. R., and Schultz, S.
(2000a). Field polarization and polarization charge distributions in
plasmon resonant nanoparticles. New J. Phys., 2, 27.1–27.9.
Kottmann, J. P., Martin, O. J. F., Smith, D. R., and Schultz, S.
(2000b). Spectral response of planar resonant nanoparticle with a non-
regular shape. Opt. Exp., 6, 213–219.
Kottmann, J. P., Martin, O. J. F., Smith, D. R., and Schultz, S. (2001).
Non-regularly shaped plasmon resonant nanoparticle as localized light
source for near field microscopy. J. Microscopy, 202, 60–65.
Kovener, G. S., Alexander, R. W., and Bell, R. J. (1976). Surface
elecromagnetic waves with damping, I. Isotropic media. Phys. Rev.
B , 14, 1458–1464.
Kozyrev, A. B., Qin, C., Shadrivov, I. V., Kivshar, Y. S., Chuang,
I. L., and van der Weide, D. W. (2007). Wave scattering and splitting
by magnetic metamaterials. Opt. Exp., 15, 11714–11722.
Kreibig, U. and Vollmer, M. (1995). Optical properties of metal clusters.
Springer, Berlin.
Kretschmann, E. and Raether, H. (1968). Radiative decay of non radia-
tive surface plasmons excited by light (surface plasma waves excitation
by light and decay into photons applied to nonradiative modes). Z.
Naturf. A, 23, 2135–2136.
Lagarkov, A. N. and Kissel, V. N. (2001). Electrodynamic proper-
ties of simple bodies made of materials with negative permeability and
negative permittivity. Doklady Phys., 46, 163–165.
Lagarkov, A. N. and Kissel, V. N. (2005). Near-perfect imaging in
a focusing system based on a left-handed-material plate. Phys. Rev.
Lett., 92, 077401–1–4.
Lagarkov, A. N., Matytsin, S. M., Rozanov, K. N., and Sarychev,
A. K. (1998). Dielectric properties of fiber-filled composites. J. Appl.
Phys., 84, 3806–3814.
References 363
Lagarkov, A. N. and Sarychev, A. K. (1996). Electromagnetic prop-

erties of composites containing elongated conducting inclusions. Phys.
Rev. B , 53, 6318–6336.
Lai, A., Caloz, C., and Itoh, T. (2004). Composite right/left-handed
transmission line metamaterials. IEEE Microw. Mag., 5, 34–50.
Lakhtakia, A. (2003). On planewave remittances and Goos-Hanchen
shifts of planar slabs with negative real permittivity and permeability.
Electromagnetics, 23, 71–75.
Landau, L. D. and Lifschitz, E. M. (1984). Electrodynamics of contin-
uous media. Pergamon Press, Oxford.
Lee, H., Xiong, Y., Fang, N., Srituravanich, W., Durant, S., Ambati,
M., Sun, C., and Zhang, X. (2005). Realization of optical superlens
imaging below the diffraction limit. New J. Phys., 7, 255–1–16.
Leifer, M. C. (1997). Resonant modes of a birdcage coil. J. Magn.
Res., 124, 51–60.
Leonhardt, U. (2006). Optical conformal mapping. Science, 312, 1777–
1780.
Leonhardt, U. and Philbin, T. G. (2006). General relativity in electrical
engineering. New J. Phys., 8, 247–1–18.
Leonhardt, U. and Philbin, T. G. (2007). General relativity in electrical
engineering. In Proc. 1st Int. Congr. on Advanced Electromagnetic
Materials in Microwaves and Optics (Metamaterials 2007), Rome, pp.
478–481.
Lewin, L. (1947). The electrical constants of a material loaded with
spherical particles. Proc. IEE , 94, 65–68.
Lewin, L. (1981). Polylogarithms and associated functions. North Hol-
land, New York.
Lewis, E. A. and Casey, J. P. (1952). Electromagnetic reflection and
transmission by gratings of resistive wires. J. Appl. Phys., 23, 605–608.
Li, K., McLean, S. J., Greegor, R. B., Parazzoli, C. G., and Tanielian,
M. H. (2003). Free-space focused-beam characterisation of left-handed
materials. Appl. Phys. Lett., 82, 2535–2537.
Li, X., He, S., and Jin, Y. (2007). Subwavelength focusing with a
multilayered Fabry–Perot structure at optical frequencies. Phys. Rev.
B , 75, 045103–1–7.
Lindell, I. V., Sihvola, A. H., Tretyakov, S. A., and Viitanen, A. J.
(1994). Electromagnetic waves in chiral and bi-isotropic media. Artech
House, Norwood, MA.
Lindell, I. V., Tretyakov, S. A., Nikoskinen, K. I., and Ilvonen, S.
(2001). BW media - media with negative parameters, capable of sup-
porting backward waves. Microw. Opt. Technol. Lett., 31, 129–133.
Linden, S., Enkrich, C., Wegener, M., Zhou, J., Koschny, T., and Souk-
oulis, C. M. (2004). Magnetic response of metamaterials at 100 THz.
Science, 306, 1351–1353.
364 References
Lippmann, G. (1894). Sur la theories de la photographie des couleurs

simples et composee par la methode interferentielle. J. Physique, 3,
97–107.
Liu, L., Caloz, C., Chang, C.-C., and Itoh, T. (2002). Forward coupling
phenomena between artificial left-handed transmission lines. J. Appl.
Phys., 92, 5560–5565.
Liu, Z., Durant, S., Lee, H., Picus, Y., Fang, N., Xiong, Y., Sun, C.,
and Zhang, X. (2007a). Far-field optical superlens. Nano Lett., 7,
403–408.
Liu, Z., Lee, H., Xiong, Y., Sun, C., and Zhang, X. (2007b). Far-
field optical hyperlens magnifying sub-diffraction-limited objects. Sci-
ence, 315, 1686–1686.
Lu, Z., Murakowski, J. A., Schuetz, C. A., Shi, S., Schneider, G. J.,
and Prather, D. W. (2005). Three-dimensional subwavelength imaging
by a photonic-crystal flat lens using negative refraction at microwave
frequencies. Phys. Rev. Lett., 95, 153901–1–4.
Luo, C., Johnson, S. G., and Joannopoulos, J. D. (2002a). All-angle
negative refraction in a three-dimensionally periodic photonic crystal.
Appl. Phys. Lett., 81, 2352–2354.
Luo, C., Johnson, S. G., Joannopoulos, J. D., and Pendry, J. B. (2002b).
All-angle negative refraction without negative effective index. Phys.
Rev. B , 65, 201104–1–4.
Luo, C., Johnson, S. G., Joannopoulos, J. D., and Pendry, J. B. (2002c).
Subwavelength imaging in photonic crystals. Phys. Rev. B , 68, 045115–
1–15.
Macfarlane, G. G. (1946). Surface impedance of an infinite parallel wire
grid at oblique angles of incidence. J. IEE Part IIIA, 93, 1523–1527.
Mahan, G. D. and Obermair, G. (1969). Polaritons at surfaces. Phys.
Rev., 183, 834–841.
Maier, S. A. (2008). The best of both worlds. Nature Photonics, 2,
460–461.
Maier, S. A., Friedman, M. D., Barclay, P. E., and Painter, O. (2005).
Experimental demonstration of fiber-accessible metal nanoparticle plas-
mon waveguides for planar energy guiding and sensing. Appl. Phys.
Lett., 86, 071103–1–3.
Maier, S. A., Kik, P. G., and Atwater, H. A. (2002). Observation of
coupled plasmon-polariton modes in Au nanoparticle chain waveguides
of different lengths: Estimation of waveguide loss. Appl. Phys. Lett., 87,
1714–1716.
Maier, S. A., Kik, P. G., and Atwater, H. A. (2003a). Optical pulse
propagation in metal nanoparticle chain waveguides. Phys. Rev. B , 67,
205402–1–5.
Maier, S. A., Kik, P. G., Atwater, H. A., Meltzer, S., Hatrel, E., Koel,
B. E., and Requicha, A. G. (2003b). Local detection of electromag-
netic energy transport below the diffraction limit in metal nanoparticle
plasmon waveguides. Nature Mater., 2, 229–232.
References 365
Mandelshtam, L. I. (1945). Group velocity in crystalline arrays. Zh.

Eksp. Teor. Fiz., 15, 475–478. [Complete Collected Works (Akad. Nauk
SSSR, Moscow, 1947) Vol. 2, p. 334; Complete Collected Works (Akad.
Nauk SSSR, Moscow, 1950), Vol. 5, pp. 419, 465].
Mansuripur, M., Zakharian, A. R., and Moloney, J. V. (2007). Surface
plasmon polaritons on metallic surfaces. Opt. Photon. News, 18, 44–49.
Marcuvitz, N. (1956). Of field representations in terms of leaky modes
or eigenmodes. IRE Trans. Ant. Prop., 4, 192–194.
Markos, P. and Soukoulis, C. M. (2008). Wave propagation: from elec-
trons to photonic crystals and left-handed materials. Princeton Univer-
sity Press, Princeton.
Marques, R. and Freire, M. (2005). On the usefulness of split ring
resonators for magnetic metamaterial design at infrared and optical
frequencies. In IEEE MELECON, Benalmadena (Malaga), Spain, pp.
222–224.
Marques, R., Martel, J., Mesa, F., and Medina, F. (2002a). Left-
handed-media simulation and transmission of em waves in subwave-
length split-ring-resonator loaded metallic waveguides. Phys. Rev.
Lett., 89, 183901–1–4.
Marques, R., Martel, J., Mesa, F., and Medina, F. (2002b). A new
2D isotropic left-handed metamaterial design: theory and experiment.
Microw. Opt. Technol. Lett., 36, 405–408.
Marques, R., Martin, F., and Sorolla, M. (2008). Metamaterials with
negative parameters: Theory, design and microwave applications. John
Wiley and Sons Inc., New York.
Marques, R., Medina, F., and Rafii-El-Idrissi, R. (2002c). Role of
bianisotropy in negative permeability and left-handed metamaterials.
Phys. Rev. B , 65, 144440–1–6.
Marques, R., Mesa, F., Martel, J., and Medina, F. (2003). Comparative
analysis of edge- and broadside-coupled split ring resonators for meta-
material design - theory and experiments. IEEE Trans. Ant. Prop., 51,
2572–2581.
Martel, J., Marques, R., Falcone, F., Baena, J. D., Medina, F., Martin,
F., and Sorolla, M. (2004). A new LC series element for compact
bandpass filter design. IEEE Microw. Wireless Comp. Lett., 14, 210–
212.
Martin, F., Bonache, J., Falcone, F., Sorolla, M., and Marques, R.
(2003a). Split ring resonator-based left-handed coplanar waveguide.
Appl. Phys. Lett., 83, 4652–465.
Martin, F., Falcone, F., Bonache, J., Marques, R., and Sorolla, M.
(2003b). Miniature coplanar waveguide stop band filters based on
multiple tuned split ring resonators. IEEE Microw. Wireless Comp.
Lett., 13, 511–513.
Maslovski, S., Ikonen, P., Kolmakov, I., Tretyakov, S., and Kaunisto,
M. (2005). Artificial magnetic materials based on the new magnetic
particle: metasolenoid. Progr. Electromagn. Res., 54, 61–81.
366 References
Maslovski, S., Tretyakov, S., and Alitalo, P. (2004). Near-field enhance-

ment and imaging in double planar polariton-resonant structures. J.
Appl. Phys., 96, 1293–1300.
Maystre, D. and Enoch, S. (2004). Perfect lenses made with left-handed
materials: Alice’s mirror? J. Opt. Soc. Am. A, 21, 122–131.
McAlister, A. J. and Stern, E. A. (1963). Plasma resonance absorption
in thin metal films. Phys. Rev., 132, 1599 – 1602.
Mehring, M. and Freysoldt, F. (1980). A slotted tube resonator (STR)
for pulsed ESR and ODMR experiments. J. Phys. E. Sci. Instrum., 13,
894–895.
Melville, D. O. S. and Blaikie, R. J. (2005). Super-resolution imaging
through a planar silver layer. Opt. Exp., 13, 2127–2134.
Melville, D. O. S. and Blaikie, R. J. (2006). Experimental comparison
of resolution and pattern fidelity in single- and double-layer planar lens
lithography. J. Opt. Soc. Am. B , 23, 461–467.
Melville, D. O. S. and Blaikie, R. J. (2007). Analysis and optimiza-
tion of multilayer silver superlenses for near-field optical lithography.
Physica B , 394, 197–202.
Melville, D. O. S., Blaikie, R. J., and Alkaisi, M. M. (2006). A com-
parison of near-field lithography and planar lens lithography. Current
Appl. Phys., 6, 415–418.
Melville, D. O. S., Blaikie, R. J., and Wolf, C. R. (2004). Submicron
imaging with a planar silver lens. Appl. Phys. Lett., 84, 4403–4405.
Merlin, R. (2004). Analytical solution of the almost-perfect-lens prob-
lem. Appl. Phys. Lett., 84, 1290–1292.
Mesa, F., Freire, M. J., Marques, R., and Baena, J. D. (2005). 3D
superresolution in metamaterial slab lenses: Experiment and theory.
Phys. Rev. B , 72, 235117–1–5.
Moharam, M. G., Grann, E. B., Pommet, D. A., and Gaylord, T. K.
(1995). Formulation for stable and efficient implementation of the rig-
orous coupled-wave analysis of binary gratings. J. Opt. Soc. Am. B , 12,
1068–1076.
Momo, F., Sotgin, A., and Zonta, R. (1983). On the design of a split
ring resonator for ESR spectroscopy between 1 and 4 GHz. J. Phys.
E. Sci. Instrum., 16, 43–46.
Moser, H. O., Casse, B. D. F., Wilhelmi, O., and Saw, B. T. (2005).
Terahertz response of a microfabricated rod-split-ring-resonator elec-
tromagnetic metamaterial. Phys. Rev. Lett., 94, 063901–1–4.
Mossotti, O. F. (1850). Discussione analitica sull’influenza che l’azione
di un mezzo dielettrico ha sulla distribuzione dell’elettricita alla super-
ficie di piu corpi elettrici disseminati in esso. Memorie di Matematica
e di Fisica della Societa Italiana delle Scienze, Modena, XXIV, Parte
seconda, 49–74.
Nachman, A. I. (1988). Reconstructions from boundary measurements.
Ann. Math., 128, 531–576.
References 367
Nefedov, I. S. and Tretyakov, S. A. (2005). On potential applications

of metamaterials for the design of broadband phase shifters. Microw.
Opt. Technol. Lett., 45, 98–103.
Notomi, M. (2000). Theory of light propagation in strongly modu-
lated photonic crystals: Refraction-like behavior in the vicinity of the
photonic band gap. Phys. Rev. B , 62, 10696–10705.
O’Brien, S., McPeake, D., Ramakrishna, S. A., and Pendry, J. B.
(2004). Near-infrared photonic band gaps and nonlinear effects in neg-
ative magnetic metamaterials. Phys. Rev. B , 69, 241101–1–4.
O’Brien, S. and Pendry, J. B. (2002). Magnetic activity at infrared fre-
quencies in structured photonic crystals. J. Phys. Condens. Matter , 14,
6383–6394.
Oliner, A. A. and Tamir, T. (1962). Backward waves on isotropic
plasma slabs. J. Appl. Phys., 33, 231–233.
Ono, A., Kato, J., and Kawata, S. (2005). Subwavelength optical imag-
ing through a metallic nanorod array. Phys. Rev. Lett., 95, 267407–1–4.
Oseen, C. W. (1922). Die Einsteinsche Nadelstichstrahlung und die
Maxwellschen Gleichungen. Ann. d. Phys., 69, 202–204.
Otto, A. (1968). Excitation of nonradiative surface plasma waves in
silver by the method of frustrated total reflection. Z. Phys., 216, 398–
410.
Oulton, R. F., Sorger, V. J., Genov, D. A., Pile, D. F. P., and Zhang, X.
(2008). A hybrid plasmonic waveguide for subwavelength confinement
and long-range propagation. Nature Photonics, 2, 496–500.
Padilla, W. J. (2007). Group theoretical description of artificial mag-
netic metamaterials. Opt. Exp., 15, 1639–1646.
Paesler, M. and Mayer, P. (1996). Near-field optics: Theory, instru-
mentation and applications. John Wiley and Sons Inc., New York.
Panina, L. V., Grigorenko, A. N., and Makhnovskiy, D. P. (2002).
Optomagnetic composite medium with conducting nanoelements. Phys.
Rev. B , 66, 155411–1–17.
Parimi, P. V., Lu, W. T., Vodo, P., Sokoloff, J., Derov, J. S., and Srid-
har, S. (2004). Negative refraction and left-handed electromagnetism
in microwave photonic crystals. Phys. Rev. Lett., 92, 127401–1–4.
Parimi, P. V., Lu, W. T., Vodo, P., and Sridhar, S. (2003). Photonic
crystals: Imaging by flat lens using negative refraction. Nature, 426,
404–1.
Park, S. Y. and Stroud, D. (2004). Surface-plasmon dispersion relations
in chains of metallic nanoparticles: An exact quasistatic calculation.
Phys. Rev. B , 69, 125418–1–7.
Paul, J., Christopoulos, C., and Thomas, D.W.P. (2001). Time-domain
modelling of negative refractive index material. Electron. Lett., 37,
912–913.
Pendry, J.B., Schurig, D., and Smith, D. R. (2006). Controlling elec-
tromagnetic fields. Science, 312, 1780–1782.
368 References
Pendry, J. B. (2000). Negative refraction makes a perfect lens. Phys.

Rev. Lett., 85, 3966–3969.
Pendry, J. B. (2001). Electromagnetic materials enter the negative age.
Phys. World , 14, 47.
Pendry, J. B. (2003a). Optics: positively negative. Nature, 423, 22–23.
Pendry, J. B. (2003b). Perfect cylindrical lenses. Opt. Exp., 11, 755–
760.
Pendry, J. B. (2004). A chiral route to negative refraction. Sci-
ence, 306, 1353–1355.
Pendry, J. B., Holden, A. J., Robbins, D. J., and Stewart, W. J. (1998).
Low frequency plasmons in thin-wire structures. J. Phys. Condens.
Matter , 10, 4785–4809.
Pendry, J. B., Holden, A. J., Robbins, D. J., and Stewart, W. J. (1999).
Magnetism from conductors and enhanced nonlinear phenomena. IEEE
Pendry, J. B., Holden, A. J., Stewart, W. J., and Youngs, I. (1996).
Extremely low frequency plasmons in metallic mesostructures. Phys.
Rev. Lett., 76, 4773–4776.
Pendry, J. B. and Ramakrishna, S. A. (2002). Near field lenses in two
dimensions. J. Phys. Condens. Matter , 14, 8463–8479.
Pendry, J. B. and Ramakrishna, S. A. (2003). Refining the perfect lens.
Physica B - Condens. Matter , 338, 329–332.
Pfenninger, S., Forrer, J., and Schweiger, A. (1988). Bridged loop-gap
resonator: A resonant structure for pulsed ESR transparent to high-
frequency radiation. Rev. Sci. Instrum., 59, 752–760.
Podolskiy, V. and Narimanov, E. (2005). Near-sighted superlens. Opt.
Lett., 30, 75–77.
Podolskiy, V., Sarychev, A. K., and Shalaev, V. M. (2002). Plasmon
modes in metal nanowires and left-handed materials. J. Nonlin. Opt.
Phys. Mater., 11, 65–74.
Podolskiy, V., Sarychev, A. K., and Shalaev, V. M. (2003). Plasmon
modes and negative refraction in metal nanowire composites. Opt.
Exp., 11, 735–745.
Pokrovsky, A. L. and Efros, A. L. (2002a). Diffraction in left-handed
materials and theory of Veselago lens. arxiv/cond-mat/0202078.
Pokrovsky, A. L. and Efros, A. L. (2002b). Diffraction theory and
focusing of light by left-handed materials. arxiv/cond-mat/0207534.
Pokrovsky, A. L. and Efros, A. L. (2002c). Electrodynamics of metallic
photonic crystals and the problem of left-handed materials. Phys. Rev.
Lett., 89, 093901–1–4.
Quidant, R., Girard, C., Weeber, J.-C., and Dereux, A. (2004). Tai-
loring the transmittance of integrated optical waveguide with short
metallic nanoparticle chains. Phys. Rev. B , 69, 085407–1–7.
References 369
Quinten, M., Leitner, A., Krenn, J. R., and Aussenegg, F. R. (1998).

Electromagnetic energy transport via linear chains of silver nanoparti-
cles. Opt. Lett., 23, 1331–1333.
Radkovskaya, A., Shamonin, M., Stevens, C. J., Faulkner, G., Edwards,
D. J., Shamonina, E., and Solymar, L. (2005). Resonant frequencies
of a combination of split rings: experimental, analytical and numerical
study. Microw. Opt. Technol. Lett., 46, 473–476.
Radkovskaya, A., Sydoruk, O., Shamonin, M., Stevens, C. J., Faulkner,
G., Edwards, D. J., Shamonina, E., and Solymar, L. (2007a). Experi-
mental study of a bi-periodic magnetoinductive waveguide: comparison
with theory. IET Microw. Ant. Prop., 1, 80–83.
Radkovskaya, A., Sydoruk, O., Shamonin, M., Stevens, C. J., Faulkner,
G., Edwards, D. J., Shamonina, E., and Solymar, L. (2007b). Transmis-
sion properties of two shifted magnetoinductive waveguides. Microw.
Opt. Technol. Lett., 49, 1054–1058.
Ramakrishna, S. A. and Pendry, J. B. (2002). The asymmetric lossy
near-perfect lens. J. Mod. Opt., 49, 1747–1762.
Ramakrishna, S. A., Pendry, J. B., Wiltshire, M. C. K., and Stewart,
W. J. (2003). Imaging the near field. J. Mod. Opt., 50, 1419–1430.
Ramo, S., Whinnery, J., and van Duzer, T. (1965). Fields and waves
in communication electronics. Wiley, New York.
Ran, L., Huangfu, J., Chen, H., Li, Y., Zhang, X., Chen, K., and Kong,
J. A. (2004). Microwave solid-state left-handed material with a broad
bandwidth and an ultralow loss. Phys. Rev. B , 70, 073102–1–3.
Rao, X. S. and Ong, C. K. (2003). Amplification of evanescent waves
in a lossy left-handed material slab. Phys. Rev. B , 68, 113103–1–4.
Rayleigh, Lord (1892). On the influence of obstacles arranged in rec-
tangular order upon the properties of a medium. Philos. Mag., 34,
481–502.
Rockstuhl, C., Zentgraf, T., Guo, H., Liu, N., Etrich, C., Loa, I.,
Szassen, K., Kuhl, J., Lederer, F., and Giessen, H. (2006). Resonances
of split-ring-resonator metamaterials in the near infrared. Appl. Phys.
B , 84, 219–227.
Rogla, J. L., Carbonell, J., and Boria, V. (2007). Study of equivalent
circuits for open-ring and split-ring resonators in coplanar waveguide
technology. IET Microw. Ant. Prop., 1, 170–176.
Rotman, W. (1962). Plasma simulation by artificial dielectrics and
parallel-plate media. IRE Trans. Ant. Prop., 10, 82–95.
Ruppin, R. (2000). Surface polaritons of a left-handed medium. Phys.
Lett. A, 277, 61–64.
Ruppin, R. (2001). Surface polaritons of a left-handed material slab.
J. Phys. Condens. Matter , 1, 1811–1819.
Russell, P. St. J. (1984). Novel thick grating beam-squeezing device in
Ta2 O5 corrugated planar waveguide. Electron. Lett., 20, 72–73.
370 References
Saadoun, M. M. I. and Engheta, N. (1992). A reciprocal phase shifter

using novel pseudochiral or omega medium. Microw. Opt. Technol.
Lett., 5, 184–188.
Saadoun, M. M. I. and Engheta, N. (1994). Bianisotropic and bi-
isotropic media and applications. Progr. Electromagn. Res., 9, 351–397.
Salandrino, A. and Engheta, N. (2006). Far-field subdiffraction optical
microscopy using metamaterial crystals: Theory and simulations. Phys.
Rev. B , 74, 075103–1–5.
Sanada, A., Caloz, C., and Itoh, T. (2004). Characteristics of the
composite right/left-handed transmission lines. IEEE Microw. Wireless
Comp. Lett., 14, 68–70.
Sarid, D. (1981). Long-range surface-plasma waves on very thin metal
films. Phys. Rev. Lett., 47, 1927–1930.
Sarychev, A. K., Shvets, G., and Shalaev, V. M. (2006). Magnetic
plasmon resonance. Phys. Rev. E , 73, 036609–1–10.
Sasaki, T., Iwata, K., Otsuka, H., and Itoh, I. (1995). Generation of
homogeneous high magnetic fields within superconducting ‘Swiss Roll’.
Cryogenics, 35, 339–343.
Sauviac, B., Simovski, C. R., and Tretyakov, S. A. (2004). Double
split-ring resonators. analytical modeling and numerical simulations.
Electromagn., 24, 317–338.
Scalora, M., Bloemer, M. J., Pethel, A. S., Dowling, J. P., Bowden,
C. M., and Manka, A. S. (1998). Transparent, metallo-dielectric, one-
dimensional, photonic band-gap structures. J. Appl. Phys., 83, 2377–
2383.
Schneider, S. and Dullenkopf, P. (1977). Slotted tube resonator: a new
NMR probe head at high observing frequencies. Rev. Sci. Instrum., 48,
68–73.
Schurig, D., Mock, J. J., Justice, B. J., Cummer, S. A., Pendry, J. B.,
Starr, A. F., and Smith, D. R. (2006). Metamaterial electromagnetic
cloak at microwave frequencies. Science, 314, 977–980.
Schurig, D. and Smith, D. R. (2004). Negative index lens aberrations.
Phys. Rev. E , 70, 065601–1–4.
Schurig, D. and Smith, D. R. (2005). Sub-diffraction imaging with
compensating bilayers. New J. Phys., 7, 162–1–15.
Schuster, A. (1904). An introduction to the theory of optics. Edward
Arnold, London.
Shadrivov, I. V., Zharov, A. A., and Kivshar, Y. S. (2003). Giant Goos-
Hanchen effect at the reflection from left-handed metamaterials. Appl.
Phys. Lett., 83, 2713–2715.
Shalaev, V. M., Cai, W., Chettiar, U. K., Yuan, H.-K., Sarychev, A. K.,
Drachev, V. P., and Kildishev, A. V. (2005). Negative index of refrac-
tion in optical metamaterials. Opt. Lett., 30, 3356–3358.
Shamonin, M., Shamonina, E., Kalinin, V. A., and Solymar, L. (2004).
Properties of a metamaterial element: Analytical solutions and nu-
References 371
merical simulations for a singly split double ring. J. Appl. Phys., 95,
3778–3784.
Shamonin, M., Shamonina, E., Kalinin, V. A., and Solymar, L. (2005).
Resonant frequencies of a split-ring resonator: analytical solutions and
numerical simulations. Microw. Opt. Technol. Lett., 44, 133–136.
Shamonina, E. (2008). Slow waves in magnetic metamaterials: history,
fundamentals and applications. phys. stat. sol. b, 245, 1471–1482.
Shamonina, E., Kalinin, V. A., Ringhofer, K. H., and Solymar, L.
(2001). Imaging, compression and Poynting vector streamlines for neg-
ative permittivity materials. Electron. Lett., 37, 1243–1244.
(2002a). Magneto-inductive waveguide. Electron. Lett., 38, 371–372.
(2002b). Magnetoinductive waves in one, two, and three dimensions.
J. Appl. Phys., 92, 6252–6261.
Shamonina, E. and Solymar, L. (2004). Magneto-inductive waves sup-
ported by metamaterial elements: components for a one-dimensional
waveguide. J. Phys. D: Appl. Phys., 37, 362–367.
Shamonina, E. and Solymar, L. (2006). Properties of magnetically
coupled metamaterial elements. J. Magn. Magn. Mater., 300, 38–43.
Shefer, J. (1963). Periodic cylinder arrays as transmission lines. IEEE
Shelby, R. A., Smith, D. R., Nemat-Nasser, S. C., and Schultz, S.
(2001b). Microwave transmission through a two-dimensional, isotropic,
left-handed metamaterial. Appl. Phys. Lett., 78, 489–491.
Shelby, R. A., Smith, D. R., and Schultz, S. (2001a). Experimental
verification of a negative index of refraction. Science, 292, 77–79.
Shen, J. T., Catrysse, P. B., and Fan, S. (2005). Mechanism for design-
ing metallic metamaterials with a high index of refraction. Phys. Rev.
Lett., 94, 197401–1–48.
Shen, J. T. and Platzman, M. (2002). Near-field imaging with negative
dielectric constant lenses. Appl. Phys. Lett., 80, 3286–3288.
Shen, Y. R. (1984). Principles of nonlinear optics. Wiley, New York.
Shvets, G. (2003). Applications of surface plasmon and phonon polari-
tons to developing left-handed materials and nano-lithography. Proc.
SPIE , 5221, 124–132.
Shvets, G. and Urzhumov, Y. (2004). Engineering the electromagnetic
properties of periodic nanostructures using electrostatic resonances.
Phys. Rev. Lett., 93, 243902–1–4.
Sihvola, A. (2007). Metamaterials in electromagnetics. Metamateri-
als, 1, 2–11.
Silin, R. A. and Sazonov, V. P. (1966). Slow wave structures. Sovetskoe
Radio, Moscow. in Russian.
372 References
Silveirinha, M. and Engheta, N. (2006). Tunnelling of electromagnetic

energy through subwavelength channels and bends using ε-near-zero
materials. Phys. Rev. Lett., 97, 157403–1–4.
Silveirinha, M. and Engheta, N. (2007). Design of matched zero-index
metamaterials using nonmagnetic inclusions in epsilon-near-zero media.
Phys. Rev. B , 75, 075119–1–10.
Simonyi, K. (1963). Foundations of electrical engineering. Pergamon
Press, Oxford.
Simovski, C. R. and He, S. (2003). Frequency range and explicit ex-
pressions for negative permittivity and permeability for an isotropic
medium formed by a lattice of perfectly conducting omega particles.
Phys. Lett. A, 311, 254–264.
Simovski, C. R., Popov, M., and He, S. (2000a). Dielectric properties
of a thin film consisting of a few layers of molecules or particles. Phys.
Rev. B , 62, 13718–1–13.
Simovski, C. R. and Tretyakov, S. A. (2007). Local constitutive para-
meters of metamaterials from an effective-medium perspective. Phys.
Rev. B , 75, 195111–1–10.
Simovski, C. R., Tretyakov, S. A., Sihvola, A. H., and Popov, M
(2000b). On the surface effect in thin molecular or composite layers.
Eur. Phys. J.: Appl. Phys., 9, 195–204.
Simovski, C. R., Viitanen, A. J., and Tretyakov, S. A. (2005). Res-
onator mode in chains of silver spheres and its possible application.
Phys. Rev. E , 72, 066606–1–10.
Sivukhin, D. V. (1957). The energy of electromagnetic waves in dis-
persive media. Opt. Spectrosc., 3, 308–312.
Smith, D. R., Kolinko, P., and Schurig, D. (2004a). Negative refraction
in indefinite media. J. Opt. Soc. Am. B , 21, 1032–1043.
Smith, D. R., Padilla, W. J., Vier, D. C., Nemat-Nasser, S. C., and
Schultz, S. (2000). Composite medium with simultaneously negative
permeability and permittivity. Phys. Rev. Lett., 84, 4184–4187.
Smith, D. R., Schultz, S., Markoš, P., and Soukoulis, C. M. (2002).
Determination of effective permittivity and permeability of metamate-
rials from reflection and transmission coefficients. Phys. Rev. B , 65,
195104–1–5.
Smith, D. R. and Schurig, D. (2003). Electromagnetic wave propagation
in media with indefinite permittivity and permeability tensors. Phys.
Rev. Lett., 90, 077405–1–4.
Smith, D. R., Schurig, D., Mock, J. J., Kolinko, P., and Rye, P. (2004b).
Partial focusing of radiation by a slab of indefinite media. Appl. Phys.
Lett., 84, 2244–2246.
Smith, D. R., Schurig, D., Rosenbluth, M., Schultz, S., Ramakrishna,
S. A., and Pendry, J. B. (2003). Limitations on subdiffraction imaging
with a negative refractive index slab. Appl. Phys. Lett., 82, 1506–1508.
References 373
Smith, D. R., Vier, D. C., Koschny, Th., and Soukoulis, C. M. (2005).

Electromagnetic parameter retrieval from inhomogeneous Metamateri-
als. Phys. Rev. E , 71, 036617–1–11.
Smolyaninov, I.I. (2003). Surface plasmon toy model of a rotating black
hole. New J. Phys., 5, 147–1–8.
Smolyaninov, I.I. and Davis, C. C. (2004). Linear and nonlinear op-
tics of surface-plasmon whispering-gallery modes. Phys. Rev. B , 69,
205417–1–7.
Smolyaninov, I.I., Elliott, J., Zayats, A., and Davis, C. C. (2005). Far-
field optical microscopy with a nanometer-scale resolution based on the
in-plane mage magnification of the surface plasmon polaritons. Phys.
Rev. Lett., 94, 057401–1–4.
Smolyaninov, I.I., Hung, Y.-J., and Davis, C. C. (2007). Magnifying
superlens in the visible frequency range. Science, 315, 1699–1701.
Solymar, L. (1984). Lectures on electromagnetic theory. Oxford Uni-
versity Press, Oxford.
Solymar, L. and Cook, D. J. (1981). Volume holograms and volume
gratings. Academic Press, London.
Solymar, L. and Walsh, D. (2004). Electrical properties of materials
(7th edn). Oxford University Press, Oxford.
Solymar, L., Webb, D. J., and Grunnet-Jepsen, A. (1996). The physics
and applications of photorefractive materials. Clarendon Press, Oxford.
Solymar, L., Zhuromskyy, O., Sydoruk, O., Shamonina, E., Young,
I. R., and Syms, R. R. A. (2006). Rotational resonance of magne-
toinductive waves: Basic concept and application to nuclear magnetic
resonance. J. Appl. Phys., 99, 123908–1–8.
Sondergaard, T. and Bozhevolnyi, S. (2007). Slow-plasmon resonant
nanostructures: Scattering and field enhancements. Phys. Rev. B , 75,
073402–1–4.
Soukoulis, C. M., Kafesaki, M., and Economou, E. N. (2006). Negative-
index materials: New frontiers in optics. Adv. Mater., 18, 1941–1952.
Stratton, J. A. (1941). Electromagnetic theory. McGraw Hill, New
York.
Svirko, Y., Zheludev, N., and Osipov, M. (2001). Layered chiral metal-
lic microstructures with inductive coupling. Appl. Phys. Lett., 58, 498–
500.
Sydoruk, O. (2007). Tailoring the properties of metamaterials for linear
and nonlinear applications. Ph. D. thesis, University of Osnabrueck.
Sydoruk, O., Kalinin, V. A., and Shamonina, E. (2007a). Parametric
amplification of magnetoinductive waves. phys. stat. sol. b, 244, 1176–
1180.
Sydoruk, O., Radkovskaya, A., Zhuromskyy, O., Shamonina, E., Sha-
monin, M., Stevens, C. J., Edwards, D. J., Faulkner, G., and Solymar,
L. (2006). Tailoring the near-field guiding properties of magnetic meta-
374 References
materials with two resonant elements per unit cell,. Phys. Rev. B , 73,
224406–1–12.
Sydoruk, O., Shamonin, M., Radkovskaya, A., Zhuromskyy, O., Sha-
monina, E., Trautner, R., Stevens, C. J., Faulkner, G., Edwards, D. J.,
and Solymar, L. (2007b). A mechanism of subwavelength imaging with
bi-layered magnetic metamaterials: theory and experiment. J. Appl.
Phys., 101, 073903–1–8.
Sydoruk, O., Shamonina, E., and Solymar, L. (2007c). Parametric
amplification in coupled magnetoinductive waveguides. J. Phys. D, 40,
68796887.
Sydoruk, O., Shamonina, E., and Solymar, L. (2007d). Tayloring of the
subwavelength focus. Microw. Opt. Technol. Lett., 49, 2228–2231.
Sydoruk, O., Zhuromskyy, O., Shamonina, E., and Solymar, L. (2005).
Phonon-like dispersion curves for magnetoinductive waves. Appl. Phys.
Lett., 87, 072501–1–3.
Syms, R. R. A. (1986). Perturbation analysis of nearly uniform coupled
waveguide arrays. Appl. Opt., 25, 2988–2995.
Syms, R. R. A. (1987). Approximate solution of eigenmode problems
for layered coupled waveguide arrays. IEEE J. Quant. Electron., 23,
525–532.
Syms, R. R. A. (2006). private communication.
Syms, R. R. A., Shamonina, E., Kalinin, V., and Solymar, L. (2005a). A
theory of metamaterials based on periodically loaded transmission lines:
Interaction between magnetoinductive and electromagnetic waves. J.
Appl. Phys., 97, 064909–1–6.
Syms, R. R. A., Shamonina, E., and Solymar, L. (2005b). Positive and
negative refraction of magnetoinductive waves in two dimensions. Eur.
Phys. J. B , 46, 301–308.
Syms, R. R. A., Shamonina, E., and Solymar, L. (2006a). Magneto-
inductive waveguide devices. IEE Proc. Microw. Ant. Prop., 153, 111–
121.
Syms, R. R. A., Solymar, L., and Shamonina, E. (2005c). Absorbing
terminations for magneto-inductive waveguides. IEE Proc. Microw.
Ant. Prop., 152, 77–81.
Syms, R. R. A., Solymar, L., and Young, I. R. (2008). Three-frequency
parametric amplification in magneto-inductive ring resonators. Meta-
materials, 2, 122–134.
Syms, R. R. A., Sydoruk, O., Shamonina, E., and Solymar, L. (2007a).
Higher order interactions in magnetoinductive waveguides. Metamate-
rials, 1, 44–51.
Syms, R. R. A., Young, I. R., and Solymar, L. (2006b). Low-loss mag-
netoinductive waveguides. J. Phys. D: Appl. Phys., 39, 1945–1951.
Syms, R. R. A., Young, I. R., and Solymar, L. (2007b). Paramet-
rically amplified magneto-inductive ring resonators. In Proc. 1st Int.
References 375
Congr. on Advanced Electromagnetic Materials in Microwaves and Op-

tics (Metamaterials 2007), Rome, pp. 40–43.
Synge, E. H. (1928). A suggested method for extending microscpoe
resolution into the ultra-microscopic region. Philos. Mag., 6, 356–362.
‘t Hooft, G. W. (2001). Comment on ‘Negative refraction makes a
perfect lens’. Phys. Rev. Lett., 87, 249701–1.
Tang, C. J. and Gao, L. (2004). Surface polaritons and imaging proper-
ties of a multi-layer structure containing negative-refractive-index ma-
terials. J. Phys. Condens. Matter , 16, 4743–4751.
Tatartschuk, E. (2007). unpublished.
Tatartschuk, E. (2008). unpublished.
Taubner, T., Korobkin, D., Urzhumov, Y., Shvets, G., and Hillenbrand,
R. (2006). Near-field microscopy through a SiC superlens. Science, 313,
1595–1595.
Thompson, G. H. B. (1955). Unusual waveguide characteristics asso-
ciated with the apparent negative permeability obtainable in ferrites.
Nature, 175, 1135–1136.
Thompson, G. H. B. (1963). Backward waves in longitudinally mag-
netized ferrite filled guides. Volume 6, Electromagnetic theory and
antennas, E. C. Jordan, editor of International series of monographs
on electromagnetic waves. Pergamon Press. Proc. Symp. Copenhagen,
Denmark, June 1962.
Tilley, D. R. (1988). Basic surface plasmon theory, Volume Surface
Plasmon-Polaritons of IOP Short Meetings Series, pp. 1–23. IOP.
Tonks, L. and Langmuir, I. (1929). Oscillations in ionised gases. Phys.
Rev., 33, 195–211.
Toraldo di Francia, P. (1952). Nuovo pupille superresolventi. Atti Fond.
Girogio Ronchi, 7, 366–372.
Toraldo di Francia, P. (1953). Directivity, supergain and information.
Proc. IRE , 3, 25–26.
Tretyakov, S. (2007). On geometrical scaling of split-ring and double-
bar resonators at optical frequencies. Metamaterials, 1, 40–43.
Tretyakov, S., Nefedov, I., Sihvola, A., Maslovski, S., and Simovski, C.
(2003). Waves and energy in chiral nihility. J. Electromagn. Waves
Appl., 17, 695–706.
Tretyakov, S. A. (2003). Analytical modeling in applied electromagnet-
ics. Artech House, Norwood, MA.
Tretyakov, S. A., Mariotte, F., Simovski, C. R., Kharina, T. G., and
Heliot, J.-P. (1996). Analytical antenna model for chiral scatterers:
comparison with numerical and experimental data. IEEE Trans. Ant.
Prop., 44, 1006–1014.
Tsakmakidis, K. L., Boardman, A. D., and Hess, O. (2007). ‘Trapped
rainbow’ storage of light in metamaterials. Nature, 450, 397–401.
376 References
Tsakmakidis, K. L., Klaedtke, A., Aryal, D. P., Jamois, C., and Hess,
O. (2006). Single-mode operation in the slow-light regime using oscil-
latory waves in generalized left-handed heterostructures. Appl. Phys.
Lett., 89, 201103–1–3.
Uslenghi, P. L. E. (2007). Invisible metamaterial trenches. In Proc. 1st
Int. Congr. on Advanced Electromagnetic Materials in Microwaves and
Optics (Metamaterials 2007), Rome, pp. 486–487.
Valanju, P. M., Walser, R. M., and Valanju, A. P. (2002). Wave refrac-
tion in negative-index media: always positive and very inhomogeneous.
Phys. Rev. Lett., 88, 187401–1–4.
Valentine, J., Zhang, S., Zentgraf, T., Ulin Avila, E., Genov, D. A.,
Bartal, G., and Zhang, X. (2008). Three-dimensional optical metamate-
rial with a negative refractive index. Nature, doi:10.1038/nature07247,
1–5.
Veselago, V. G. (1968). The electrodynamics of substances with simul-
taneously negative values of ε and µ. Sov. Phys. Usp., 10, 509–514.
(translated from Usp. Fiz. Nauk 92, 517, 1967).
Wait, J. R. (1962). Electromagnetic waves in stratified media. MacMil-
lan, New York.
Walter, C. H. (1965). Traveling wave antennas. McGraw Hill, New
York.
Wang, S., Tang, C., Pan, T., and Gao, L. (2006). Effectively nega-
tively reflective material made of negative-permittivity and negative-
permeability bilayer. Phys. Lett. A, 351, 391–397.
Webb, K. J. and Yang, M. (2006). Subwavelength imaging with a
multilayer silver film structure. Opt. Lett., 31, 2130–2132.
Webb-Wood, G., Ghoshal, A., and Kik, P. G. (2006). In situ exper-
imental study of a near-field lens at visible frequencies. Appl. Phys.
Lett., 89, 193110–1–3.
Weber, W. H. and Ford, G. W. (2004). Propagation of optical excita-
tions by dipolar interactions in metal nanoparticle chains. Phys. Rev.
B , 70, 125429–1–8.
Welford, K. (1988). The method of attenuated total reflection, Volume
Surface plasmon-polaritons of IOP Short Meetings Series, pp. 25–78.
Welford, K. (1991). Surface plasmon-polaritons and their uses. Opt.
Quant. Electron., 23, 1–27.
Wendler, L. and Haupt, R. (1986). Long-range surface plasmon-
polaritons in asymmetric layer structures. J. Appl. Phys., 59, 3289–
3291.
Whinnery, J. R. and Jamieson, H. W. (1944). Equivalent circuits for
discontinuities in transmission lines. Proc. IRE , 32, 98–114.
White, J., White, C. J., and Slocum, A. H. (2005). Octave-tunable
miniature RF resonators. Microw. Wireless Comp. Lett., 15, 793–795.
Whitney, A. V., Elam, J. W., Zou, S., Zinovev, A. V., Stair, P. C.,
Schatz, G. C., and van Duyne, R. P. (2005). Localized surface plasmon
References 377
resonance nanosensor: A high-resolution distance-dependence study us-

ing atomic layer deposition. J. Phys. Chem. B, 109, 20522–20528.
Williams, J. M. (2001). Some problems with negative refraction. Phys.
Rev. Lett., 87, 249703–1.
Wiltshire, M. C. K., Hajnal, J. V., Pendry, J. B., Edwards, D. J.,
and Stevens, C. J. (2003a). Metamaterial endoscope for magnetic field
transfer: near field imaging with magnetic wires. Opt. Exp., 11, 709–
715.
Wiltshire, M. C. K., Hajnal, J. V., Pendry, J. B., and Edwards, D. J.
(2004a). RF field transmission through Swiss Rolls - an anisotropic
magnetic metamaterial. In Proc. 27th ESA Antenna Technology Work-
shop on Innovative Periodic Antennas: Electromagnetic Bandgap, Left-
handed Materials, Fractal and Frequency Selective Surfaces, Santiago
de Compostela, Spain.
Wiltshire, M. C. K., Pendry, J. B., Young, I. R., Larkman, J.,
Gilderdale, D. J., and Hajnal, J. V. (2001). Microstructured magnetic
materials for radio frequency operation in magnetic resonance imaging
(MRI). Science, 291, 849–851.
Wiltshire, M. C. K., Shamonina, E., Young, I. R., and Solymar, L.
(2003b). Dispersion characteristics of magneto-inductive waves: com-
parison between theory and experiment. Electron. Lett., 39, 215–217.
Wiltshire, M. C. K., Shamonina, E., Young, I. R., and Solymar, L.
(2004b). Experimental and theoretical study of magneto-inductive
waves supported by one-dimensional arrays of ‘Swiss Rolls’. J. Appl.
Phys., 95, 4488–4493.
Wood, B., Pendry, J. B., and Tsai, D. P. (2006). Directed subwave-
length imaging using a layered metal-dielectric system. Phys. Rev.
B , 74, 115116–1–8.
Xu, X., Quan, B., Gu, C., and Wang, L. (2006). Biansotropic response
of microfabricated metamaterials in the terahertz region. J. Opt. Soc.
Am. B , 23, 1174–1180.
Yablonovitch, E. (1987). Inhibited spontaneous emission in solid-state
physics and electronics. Phys. Rev. Lett., 58, 2059–2062.
Yang, F., Sambles, J. R., and Bradberry, G. W. (1991). Long-range
surface modes supported by thin films. Phys. Rev. B , 44, 5855–5872.
Yariv, A., Xu, Y., Lee, R. K., and Scherer, A. (1999). Coupled-
resonator optical waveguide: a proposal and analysis. Opt. Lett., 24,
711–713.
Yeatman, E. (1996). Resolution and sensitivity in surface microscopy
and sensing. Biosens. Bioelectron., 6, 635649.
Yen, T. J., Padilla, W. J., Fang, N., Vier, D. C., Smith, D. R., Pendry,
J. B., Basov, D. N., and Zhang, X. (2004). Terahertz magnetic response
from artificial materials. Science, 303, 1494–1496.
Zayats, A. V., Smolyaninov, I. I., and Maradudin, A. A. (2005). Nano-
optics of surface plasmon polaritons. Phys. Rep., 408, 131–314.
378 References
Zervas, M. N. (1991). Surface plasmon-polariton waves guided by thin

metal films. Opt. Lett., 16, 720–722.
Zhang, S., Fan, W., Malloy, K. J., Brueck, S. R. J., Panoiu, N. C., and
Osgood, R. M. (2005a). Near-infrared double-negative metamaterials.
Opt. Exp., 13, 4922–4930.
Zhang, S., Fan, W., Minhas, B. K., Frauenglass, A., Malloy, K. J.,
Osgood, R. M., and Brueck, S. R. J. (2005b). Midinfrared resonance
magnetic nanostructures exhibiting a negative permeability. Phys. Rev.
Lett., 94, 037402–1–4.
Zhang, S., Fan, W., Panoiu, N. C., Malloy, K. J., Osgood, R. M., and
Brueck, S. R. J. (2005c). Experimental determination of near-infrared
negative-index metamaterials. Phys. Rev. Lett., 95, 137404–1–4.
Zhang, S., Fan, W., Panoiu, N. C., Malloy, K. J., Osgood, R. M., and
Brueck, S. R. J. (2006). Optical negative-index bulk metamaterials
consisting of 2D perforated metal-dielectric stacks. Opt. Exp., 14, 6778–
6787.
Zhao, Y., Belov, P. A., and Hao, Y. (2006). Spatially dispersive finite-
difference time-domain analysis of sub-wavelength imaging by the wire
medium slabs. Opt. Exp., 14, 5154–5167.
Zhao, Y., Belov, P. A., and Hao, Y. (2007). Amplification of evanescent
spatial harmonics and subwavelength imaging inside of a wire medium
slab. In Proc. 1st Int. Congr. on Advanced Electromagnetic Materials
in Microwaves and Optics (Metamaterials 2007), Rome.
Zhou, J., Koschny, T., Kafesaki, M., Economou, E. N., Pendry, J. B.,
and Soukoulis, C. M. (2005a). Saturation of the magnetic response
of split-ring-resonators at optical frequencies. Phys. Rev. Lett., 95,
223902–1–4.
Zhou, J., Zhang, L., Tuttle, G., Koschny, T., and Soukoulis, C. M.
(2006). Negative index materials using simple short wire pairs. Phys.
Rev. B , 73, 041101(R)–1–4.
Zhou, L., Wen, W., Chan, C. T., and Sheng, P. (2005b). Electromag-
netic wave tunneling through negative-permittivity media with high
magnetic fields. Phys. Rev. Lett., 94, 243905–1–4.
Zhou, X. and Hu, G. (2007). Total transmission condition for photon
tunnelling in a layered structure with metamaterials. J. Opt. A: Pure
Appl. Opt., 9, 60–65.
Zhuromskyy, O. (2008). unpublished.
Zhuromskyy, O., Shamonina, E., and Solymar, L. (2005a). 2D meta-
materials with hexagonal structure: spatial resonances and near field
imaging. Opt. Exp., 13, 9299–9309.
Zhuromskyy, O., Shamonina, E., and Solymar, L. (2005b). Effect of
radiation on dispersion of magneto-inductive waves in a metamaterial.
Proc. SPIE , 5955, 595506–1–6.
Ziolkowski, R. W. (2003a). Pulsed and CW Gaussian beam interactions
with double negative metamaterial slabs. Opt. Exp., 11, 662–681.
References 379
Ziolkowski, R. W. (2003b). Pulsed and CW Gaussian beam interactions

with double negative metamaterial slabs: Errata. Opt. Exp., 11, 1596–
1597.
Ziolkowski, R. W. (2004). Propagation in and scattering from a
matched metamaterial having a zero index of refraction. Phys. Rev.
E , 78, 046608–1–12.
Ziolkowski, R. W. and Heyman, E. (2001). Wave propagation in me-
dia having negative permittivity and permeability. Phys. Rev. E , 64,
056625–1–15.
Index
Acoustic waves, 37–39, 76, 228, 321 Clausius-Mossotti, 42, 54, 303, 317, 327
Acronyms, 325 Cloaking, 290, 309–314
Ag, 101, 141–142, 185–187, 194 Complementary split-ring resonator, 45, 127, 205
Al2 O3 , 142, 194 Conductivity, 2, 6, 43
Anisotropic, anisotropy, 134, 191, 201–204, 289, 294–297, Composite right/left-handed structure, 73, 209
312, 318, 344–345 Compression, 179–180, 204
Anomalous dispersion, 22, 208 Confinement, 88, 100–106, 117–118, 322
Antisymmetric, 93, 97–105, 116, 171, 222–223 Continuous line, 29, 142, 207
Artificial materials, 35–40 Coplanar waveguide, 204–208, 320
Asymmetric structures, 100–101 resonant loops, 214
Attenuation coefficient, 5, 84, 97, 105, 206, 215, 298 Coupled
Au, 138, 142, 186–187 lines, 238–241, 248–249, 265, 279, 285, 291
Axial configuration, 24, 56, 216–219, 225–230, 234, nanoparticles, 321
253–257, 268, 279, 298, 347 oscillators, 159
resonant sheets, 290
Backward waves, 1, 10, 21–22, 56–59, 73, 208, 215, 230, resonators, 242, 321
239–243, 254–256, 318-319 surface plasmon–polaritons, 188
Backward-wave media, 60, 319 wave analysis, 142
Bandgap, 36, 191, 296, 322 waveguides, 279, 321
Bianisotropy, 132, 322 Coupling coefficient, 215–216, 227, 234, 246, 253, 257, 260,
Biperiodic lines, 227–229, 235, 247 278–279
Boundary conditions, 1, 4–9, 15, 64–67, 79, 104–106, Cross-layer coupling, 140
146–147, 201, 262–263, 331, 336–341, 345 Current vector, 219
Bragg Cutoff 169
condition, 295, 330 frequency, 173
effect, 36 waveguide, 201
grating, 246
reflection, 247, 322 Delay line, 243
reflector, 246 Descartes’ law, 9
Brewster wave, 80, 333–334 Diatomic chain, 228
Brillouin zone, 260, 305 Dielectric droplet, 196
Broadside-coupled split ring, 132, 265 Dielectric–metal–dielectric, 339
Build-up time, 159 Dipole, 17–19, 327
Bulk polaritons, 77–79 antenna, 290
approximation, 56
Capacitance, electric, 1, 20, 40, 51, 76, 124–125, 129–130, 203, 291,
gap, 45, 122, 127, 145–152 303–304
inter-ring, 45, 126–127, 133, 145–149, 343 field, 203
voltage dependent, 281, 285 field equations, 343
Capacitively loaded half-wave, 184
loop, 231 interactions, 55–56
rod, 293–294 magnetic, 1, 4, 19, 50–56, 123–124, 128, 203, 236,
split ring, 134, 236, 265 267–271, 327
Cavity, moment, 20, 51, 125, 304
magnetron, 120, 242, 352 nuclear, 250, 287
re-entrant, 45, 120, 319 rotating, 242, 250
resonator, 208, 242 Directional coupler, 209, 247
Chain matrices, 25–26, 32, 48 Discretization, effect of, 175–176
Chiral, chirality, 59, 123–124, 139–140, 319–322 Dispersion, 1, 21–23, 76–87, 92–96, 102–106, 159, 189, 208,
Chromium, 185, 194 217, 256, 298, 312
382 Index
characteristics, 49, 73, 77, 210–213, 227–231, 236–240, Gaussian beam, 301–302
247–250, 256–257, 266–269, 298, 304, 322 Generalized Ohm’s law, 218, 240, 271, 299
curve, 29–30, 39, 47–49, 73–84, 89–92, 98, 102–103, 116, Geometrical optics, 70–72, 155, 179, 192, 197, 311
193, 208, 214, 233–241, 252, 256, 260, 269–270, Gold, 69, 135–141, 187, 195–196
274–280, 298, 305–306, 333–334 Goos–Hanchen shift, 289, 299–302
equation, 28–29, 37–39, 46–49, 53–56, 74–83, 106–116, Group velocity, 22, 29, 39, 57–59, 69, 189, 225, 252–257,
193, 213–225, 231, 237–242, 251–254, 260–261, 272, 293, 298, 305, 334
266–270, 278, 297–298, 304–306, 334–341 Growing wave, 69, 206–207, 294
relation, 28, 103, 203, 214–217, 232–233, 279
Displacement current, 140, 144, 146 Hairpin resonator, 125, 140
Distributed circuits, 144 Helix, 124
Double circular elements, 123 Hexagonal lattice, 259–265
Double ring, 122, 126–129 Higher
quadruply split, 126–127 order interaction, 236–238
singly split, 122, 148, 152 resonance, 150–153
Drude model, 76–83, 106, 111, 301–304, 317 Historical review, 58, 120–126, 315–323
Hybrid modes, 76
Effective Hyperbolic
medium, 36, 148, 191–193, 299, 317 dispersion relation, 193, 298
permeability, 36, 52–55, 138–139, 200–201, 318 isofrequency curve, 298
permittivity, 14, 36, 40–42, 54, 123–124, 317 Hyperlens, 194
plasma frequency, 42–44, 202, 318
refractive index, 65 Imaging,
Eigensolution, 15, 93, 147 by channelling, 265, 291–294
Eigenvector, 214, 218–220, 222 near-field, 66, 249, 321
Electric resonance, 130, 136 of M-shaped wire, 265
Electroinductive waves, 299 of magnetic resonance, 229, 267, 320
Electrostatic approximation, 94–95, 102, 164–167 perfect, 66, 92, 159, 206, 323
failure of, 180-183, 197 subwavelength, 68–72, 79–81, 87, 117, 155–199, 290, 294,
Electrostatic limit, 70, 83, 162, 165–167, 170–171, 335 321
Equivalent circuit, 46, 74, 126–127, 133, 145–148, 209, 214, Impedance matrix, 218–220, 224, 251–252, 262, 271
280, 285 Indefinite media, 289, 296–298
Evanescent wave, 72, 156, 160, 176, 187–189, 197, 210, Inductance,
223, 290–294, 298, 317–318 kinetic, 24–25, 136
Ewald circle, 1, 8–11, 57, 254 mutual, 23-24, 29, 38, 45–46, 53–56, 73, 140, 144, 148,
Excitation of 209, 214, 225–228, 237–247, 252–254, 258–262,
loaded transmission lines, 210–211 267, 271, 278–279, 298, 327–328, 347
magnetoinductive waves, 214–224, 238, 251, 259–264, self-, 29, 45–48, 140
268-271, 276, 287 Inductive coupling, 140
surface plasmon–polaritons, 162, 168–171, 182, 188, Inhomogeneous, inhomogeneity, 157, 312, 320
195–197, 301, 318 Interaction,
higher order, 236–238
Fermat’s principle, 11–12 nearest-neighbour, 38, 203, 236–239, 252, 259, 266, 303,
Ferrite, 204, 318–320 321
Filling factor, 111 Interdigital, 209
Filters, 119, 125–126, 190–191, 199, 204–206, 320 Inter-ring capacitance, 45, 126–127, 133, 145–149, 343
Flat lens, 59–67, 71, 155–158, 179, 185–192, 290, 298, 316 Internal focus, 61–62, 72, 197
Focus, 59–63, 72, 155, 179, 186–189, 197, 210, 266, 289, Inverse scattering, 311, 321
298–299, 316, 321 Invisibility, 290, 309–314
Forward waves, 57, 156, 230, 243, 254–256 Isotropic resonator, 133–135
Fourier
analysis, 1, 31 Kinetic
coefficient, 273 inductance, 24–25, 136
component, 158–161, 179, 271–275 resistance, 25
plane, 143 Kramers–Kronig relations, 159
spectrum, 155, 172, 273–277 Kretschmann configuration, 81, 362
transform, 31–33
Four-poles, 25–30, 38, 45–49, 125, 213 Laplace’s equation, 83, 335–337
Full ring, 130–131, 151, 303 Leaky wave, 99, 269
Left-handed materials, 5, 59, 64, 73, 110, 157, 202
Gap capacitance, 45, 122, 127, 145–152 transmission line, 209–210
Index 383
Lens, Material parameters, 64–65, 70, 119, 123–126, 148,

cylindrical, 174, 192–195 156–159, 163, 297–301, 311–313, 329–330
design, 12–13 Matrix,
flat, 59–67, 71, 155-158, 179, 185–192, 290, 298, 316 chain, 26–29, 46–49
near-perfect, 157–175 impedance, 218–220, 224, 251–252, 262, 271
perfect, 32, 66–72, 81, 107, 115–117, 155–165, 169–181, Maxwell’s equations, 1–4, 21, 58, 64, 79, 83, 93, 160, 175,
290, 297, 316, 321, 335 224, 310, 317, 335, 339
silver, 82, 174, 181, 185–192, 197 Metal–dielectric, 79, 102–106, 191, 294, 296, 341
Veselago’s, 70–73, 155–156, 298 Metal–dielectric–metal, 77, 102–103
Loading, 38, 45–53, 73, 203–204, 210–211, 239, 279 Metallic stripe, 105–106, 139–140
Loop, Metasolenoid, 132
capacitively loaded, 38, 46, 133, 213–214, 231, 242, Microstrip, 133, 199, 204-205, 209–210, 233
251–255, 278, 298–299, 322 Mirror images, 203
gap resonator, 121–122, 319 Misconceptions, 196–197
resonant, 46–52, 56, 214, 220, 243–246 Multilayers, 141–142, 157, 188-196, 289, 296, 345
Losses, 5, 43–44, 50–54, 65–72, 77, 83–92, 100, 107, Multipole, 303
117–118, 123, 127, 150, 156–160, 167–178, Mutual impedance, 23, 251, 262
182–185, 196-197, 214–215, 219, 229, 234, Mutual inductance, 23-24, 29, 38, 45–46, 53–56, 73, 140,
266–275, 283–285, 297, 305, 312, 334 144, 148, 209, 214, 225–228, 237–247, 252–254,
Lossy permeability, 167–168 258–262, 267, 271, 278–279, 298, 327–328, 347
Lossy permittivity, 164–167 in the presence of retardation, 347
Magnetic resonance, 111–114, 121, 125–140, 229, 244, Nanorod, 25

249–250, 267, 320 Nanoparticle, 87, 268, 289, 299–306, 321–322
imaging, 229, 267, 320, 377 Nanophotonics, 289, 302–303
Magnetoinductive (MI) wave Nearest-neighbour interaction, 38, 203, 236–239, 252, 259,
amplification of, 227–231, 250, 278–280, 285–287, 318 266, 303, 321
applications of, 243–249 Near-field imaging, 66, 249, 321
axial array/configuration/line, 24, 56, 215–239, 268, 279 Near-perfect lens, 157–175
planar array/configuration/line, 23–24, 103–105, 140, Near-zero index, 307
215–217, 221–222, 228–240, 252–256, 268, Negative index lens/material/medium/prism, 37, 50,
276–279, 285–288, 347 58–67, 140–141, 172, 179, 184, 195–197,
two-dimensional array, 258–265 206–211, 289–290, 296–302, 316, 321
biperiodic lines, 227–231, 235–236, 247 Negative permittivity, 37, 80, 119, 140–141, 176–188, 207,
current distribution, 214, 220–226, 238–241, 258–263, 289–292, 296–297, 311, 316–318
267–273, 277, 310 Negative permeability, 37, 53–54, 63, 106–108, 119, 126,
directional coupler, 247 138–141, 202, 289–292, 296–297, 301, 316–320,
eigenvector, 214, 218–222 335
eigenvalue, 214, 218–220 Negative refraction, 37, 56–59, 63–65, 119, 123–126, 157,
172, 199, 210–213, 257, 296–297, 316–318,
excitation, 214–224, 238, 251, 259–264, 268–271, 276,
322–323
287
Newton’s equation, 2, 37
hexagonal lattice, 259–263
Nonlinear element/diode/effects, 250, 278
imaging, 220, 229, 243, 248–251, 266–267
Non-integrable fields, 158
matching, 217
Nuclear magnetic resonance, 121, 125, 249–250
negative refraction, 256–257
Numerical simulation, 42, 63–65, 120, 128–131, 137–143,
nonlinear effects, 277–287
152–153, 160, 175–176, 191–194, 205–207,
parametric amplification, 227–231, 250, 278, 284–287 292–293, 300–301, 308, 312
phase matching, 278–280
power, 225, 253–254 Ohm’s law, generalized, 218, 240, 271, 299
Poynting vector, 214, 224–225 Omega particle, 124, 322
reflection and refraction, 254–258 Open split ring resonator, 133–134
retarded, 266–277 Optical
rotational resonance, 242–244, 249–250, 288 circuit, 135, 322
spatial resonance, 251, 259–265 path, 12–13, 61–62, 142
transducer, 233 phonons, 78, 186
two-dimensional, 251, 265 Otto configuration, 81
Magnetron cavity, 120, 242
Magnifying multilayer superlens, 192–196 Parametric amplification, 227–231, 250, 278, 284–287
Matching, 217 Parasitic capacitance, 209
384 Index
Pass band, 37–39, 47–50, 63, 73, 84, 200–205, 222–223, Pump wave, 227–231, 250, 278–280, 284–288
227–228, 232–233, 239, 244, 254, 260, 278, 334
Penetration depth, 85–88, 100 Quality factor, 51–54, 151, 215–216, 229–233, 243,
Percolation, 139 272–275, 286
Perfect imaging, 66, 92, 159, 206, 323 Quartz, 186
Perfect lens, 32, 66–72, 81, 107, 115–117, 155–165, Quasi-static, 213, 266–270, 303–306, 327
169–181, 290, 297, 316, 321, 335
Perfect tunnelling, 206–207, 296 Radiation damping, 50–52, 268, 304
Periodic structure, 102–103, 158, 236, 290, 295, 319–321 Radiation loss, 51–52, 268–269, 305
Permeability Radiation resistance, 18–19, 51, 271
effective, 42, 52–55, 138–139, 200–201, 318 Reciprocity, 25–26, 30, 130
lossy, 167–168 Rectangular waveguide, 199
negative, 37, 53–54, 63, 106–108, 119, 126, 138–141, 202, Re-entrant cavity, 45, 120, 319
289–292, 296–297, 301, 316–320, 335 Reflection coefficient, 65, 160, 172, 244–245, 300
transverse, 204 Refractive index,
tensor, 19–20 deviation in, 169–174
Permittivity near zero, 307
effective, 40–42, 54, 123–124, 317 Resolution, 66–67, 72, 159, 165–176, 182–186, 190–197,
292–293, 321
lossy, 164–167
Reststrahl, 79, 186
negative, 37, 80, 119, 140, 141, 176–188, 207, 289–292,
Resonance,
296–297, 311, 316–318
electric, 130, 136
tensor, 19–20
enhanced tunnelling, 191
Phase shifter, 199, 204, 208–209, 243
higher, 150–153
Phase matching, 195, 278–280, 285
LC, 128, 140, 150
Phase velocity, 9, 22, 29, 57–60, 76, 156, 228, 250–256,
magnetic, 111–114, 121, 125–140, 229, 244, 249–250,
272, 278, 289, 312, 320
267, 320
Phonon, 76–78, 228
of triangle particle, 143–144, 303
Phonon–polariton, 76–79, 186
plasma, 75, 135–142, 303
Photonic
rotational, 242–244, 249–250, 288
bandgap, 36, 191, 322
spatial, 251, 259–265
circuit, 86
Resonant
crystals, 294–296 elements, 37, 44-45, 203, 242, 259, 277, 290–291, 303,
Photoresist, 185 319–323
Planar array/configuration/line, 23–24, 103–105, 140, loops, 46–51, 56, 214, 220, 243–246
215–217, 221–222, 228-240, 252–256, 268, Resonator,
276–279, 285–288, 347 complementary split ring, 45, 127
Planar–axial configuration, 240, 252–257, 298 hairpin, 125, 140
Plasma, 75 isotropic, 133–135
asymmetric structures, 100–101 open, 135
frequency, 14, 25, 37, 42–44, 76–78, 82, 102, 106, spiral, 126, 205
111–112, 137–140, 185, 202, 317–318, 322, 333 small, 119–154, 199
resonance, 75, 135–142, 303 split ring, 45, 63–65, 73, 111, 124–137, 145–152, 184,
simulation, 42 199–205, 233, 253, 265, 292, 298–299, 308, 313,
solid state, 318 329, 343–344
wave, 6, 37, 76–78, 83 Retardation, 3, 176, 213, 267–271, 305, 322, 347
Plasmon–polaritons, 76–118, 156–162, 168–169, 174–189, Right-handed, 5, 59, 73, 124, 209–210
195–197, 290, 294, 334, 339–340 Rods, 25, 39–40, 48–50, 63–64, 73–75, 111, 119, 139–140,
Polarizability, 1, 20–21, 41–42, 50–56, 124–125, 268, 304, 204–205, 293
343 Rotational resonance, 242–244, 249–250, 288
Polylogarithm, 238, 363
Poly-methyl metachrylate (PMMA), 185–186, 195 Scaling, 137, 152
Poynting vector, 18–19, 59, 72, 77, 90–92, 98, 190–193, Scattering, 193, 301, 311
197, 214, 224–225, 296 coefficients, 30–31, 43, 65, 200, 321, 329
optics, 189 cross section, 143–144, 310–311
streamlines, 71, 189, 311 inverse, 311, 321
Printed circuit board (PCB), 64, 234 Short-rod pairs, 139–141
Propagation constant (see also wave number), 27, 97, 215, Signal processing, 104, 234, 243, 303
220, 227–233, 279, 307 Signal wave, 227–230, 278, 283–287
Propagation length, 76, 83–88, 97–101, 106 Silver, 67–72, 82, 86, 101–103, 137, 141–143, 174–177,
Pseudochiral media, 123 181–197
Index 385
Silver lens, 82, 174, 181, 185–192, 197 Thin layers, 190, 345
Simulation, 42, 63–65, 120, 128–131, 137–143, 152–153, Total internal reflection, 11, 59, 207, 255, 300–301, 307
160, 175–176, 191–194, 205–207, 292–293, Transducer, 233
300–301, 308, 312 Transfer function, 31–33, 66–70, 75, 92, 156–183, 190–191,
Single interface, 77–80, 92–100, 106–117, 335–339 197
SiC, 79, 186–188 Transmission,
SiO2 , 185–187 coefficient, 8–10, 16, 30, 65, 200–201, 226–227, 244–247,
Slab, 257, 266, 321, 329–330, 339
plane wave incident upon, 16–18 line, 26–29, 45–50, 73–74, 121–123, 150, 205–209
reflection by, 16 left-handed, 209–210
silver, 101, 177, 181–182, 188–189 spectra, 128–131
surface plasmon–polaritons on, 92–102, 116–118 Transverse
transfer function of, 16, 70 electric mode (TE), 7–10, 17, 21, 79, 107–117, 138–139,
Slotted-tube resonator, 121–122 156, 160–163, 168, 174, 312
Skin depth, 86–87 magnetic mode (TM), 7–10, 15–16, 67–70, 79–117,
Slow wave structures, 320–322 138–139, 156–160, 167–169, 176, 300, 313, 339
Small resonators, 119–154, 199 permeability, 203–204
Snell’s law, 8–12, 59, 70, 306, 316, 333 Triangle particle, 143–144, 303
Spatial frequency, 31–33, 66–68, 158–159, 163, 174–179, Tungsten mask, 185
188, 197, 297–298 Tunnelling, 191, 206–207
Spatial resonance, 251, 259–265 perfect, 206–207, 296
Spherical scatterer, 309–311
Varactor diodes, 134, 250, 278
Split pipe, 235–241, 249, 265, 308
Veselago’s lens, 70–73, 155–156, 298
Split-ring resonator (SRR)
Via, 133
broadside coupled, 132, 265
Voltage vector, 218–219, 258, 271
complementary, 45, 127, 205
edge coupled, 209 Wave,
mathematical model of, 144–152 Brewster, 80, 333–334
Square waveguide, 200, 206 equation, 2–7, 21, 27–28, 261, 297, 336
Staple structure, 138–139 four-pole, 28–30, 38, 213
Stop band, 37, 47–50, 63, 73, 78–79, 110, 132, 200–208, idler, 278, 288
223–224, 232, 239–241, 247–249, 277, 334 leaky, 99, 269
Stored energy, 18, 225, 253 nanoparticle, 87, 268, 289, 299–306, 321–322
Stratified media, 190 number (see also propagation constant), 1, 5, 21–22,
Stripe, 103–106, 139–140 31–33, 37, 57, 76–81, 89, 203, 269–270, 274–276
Subwavelength imaging, 68–72, 79–81, 87, 117, 155–199, pump, 227–231, 250, 278–280, 284–288
290, 294, 321 signal, 227–231, 283–287
Superdirective, superdirectivity, 308–310, 321 vector diagram, 10, 251
Superlens (see also perfect lens), 69, 168, 175, 184–197 Wavefront conversion, 307–308
Superresolution, 321 Waveguide,
Surface plasmon–polariton (SPP), 76–118, 156–162, components, 244–248
168–169, 174–189, 195–197, 290, 294, 334, coplanar, 204–208, 320
339–340 couplers, 209–210, 247
Surface waves, 14–16, 33, 69, 83, 98, 103, 107, 333–336 hollow, 199
Swiss roll, 126, 243, 264–265 microstrip, 133, 199, 204–205, 209–210, 233
rectangular, 199
Tensor, 2, 297 square, 200, 206
permeability, 19–20, 52, 123, 200–201, 298–300, 312, 318 Wire medium, 42–44
permittivity, 19–20, 123, 193, 312
polarizability, 20–21, 51, 124, 343–345 ZnS, 138
Authors
Laszlo Solymar was born in 1930 in Budapest. He is Emeritus Profes-

sor of Applied Electromagnetism at the University of Oxford and Visit-
ing Professor and Senior Research Fellow at Imperial College, London.
He graduated from the Technical University of Budapest in 1952 and re-
ceived the equivalent of a Ph.D in 1956 from the Hungarian Academy of
Sciences. In 1956 he settled in England where he worked first in indus-
try and later at the University of Oxford. He did research on antennas,
microwaves, superconductors, holographic gratings, photorefractive ma-
terials, and metamaterials. He has held visiting professorships at the
Universities of Paris, Copenhagen, Osnabrück, Berlin, Madrid and Bu-
dapest. He published 8 books and over 250 papers. He has been a Fellow
of the Royal Society since 1995. He received the Faraday Medal of the
IEE in 1992.
Ekaterina Shamonina was born in 1970 in Twer, Russia. She is Pro-

fessor in Advanced Optical Technologies at the University of Erlangen-
Nürnberg. She graduated in 1993 in Physics at the Moscow State Uni-
versity and received her doctorate in 1998 from the University of Os-
nabrück, Germany. She was a visiting scientist at the University of
Campinas, Brazil in 1996 and 1998. In 2000 she was awarded the 7-year
Emmy Noether Fellowship from the German Research Council (Deutsche
Forschungsgemeinschaft). She spent the first leg of the fellowship, 2000-
2002 at the University of Oxford. After further six months at Imperial
College, London she returned to the University of Osnabrück where she
built up a research group working on Metamaterials. She completed
her habilitation in Theoretical Physics in 2006. Her main research areas
apart from metamaterials have been amorphous semiconductors, pho-
torefractive materials, antennas and plasmonics. She published over 70
research papers. She was awarded the Hertha-Sponer Prize 2006 of the
German Physical Society.

Laszlo Solymar, Ekaterina Shamonina - Waves in Metamaterials-OUP (2009) PDF

Uploaded by

Copyright:

Available Formats

Laszlo Solymar, Ekaterina Shamonina - Waves in Metamaterials-OUP (2009) 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

Laszlo Solymar, Ekaterina Shamonina - Waves in Metamaterials-OUP (2009) PDF

Uploaded by

Copyright:

Available Formats

What is the definition of metamaterials provided in the text?

What is the definition of metamaterials provided in the text?

What are some examples of early work related to metamaterials mentioned in the preface?

What are some examples of early work related to metamaterials mentioned in the preface?

Waves in Metamaterials

This page intentionally left blank

Metamaterials is a new subject. The term ‘metamaterial’ could have

achievements and of some of the misconceptions. Chapter 6 discusses

subject or relevant to the subject. A conservative estimate would make

1 Basic concepts and basic equations 1

2 A bird’s-eye view of metamaterials 35

2.6.3 By a combination of resonant magnetic

4 Small resonators 119

5 Subwavelength imaging 155

6 Phenomena in waveguides 199

7 Magnetoinductive waves I 213

7.4 Excitation 217

8 Magnetoinductive waves II 251

9 Seven topics in search of a chapter 289

9.3 Combinations of negative-permittivity

10 A historical review 315

B Field at the centre of a cubical lattice

C Derivation of material parameters from

D How does surface charge appear

E The Brewster wave 333

F The electrostatic limit 335

G Alternative derivation of the dispersion equation

H Electric dipole moment induced by a magnetic field

I Average dielectric constants of a multilayer structure 345

J Derivation of mutual inductance between two

1.2 Newton’s equation and electrical

1.3 Maxwell’s equations, fields and

where ds is the line element vector. The third important relationship we

ε0 = 8.85 × 10−12 A s V−1 m−1 , µ0 = 4π × 10−7 V s A−1 m−1 . (1.16)

In vacuum εr = µr = 1. In a material, they are different from unity

p = qd and m = µ0 ISip . (1.18)

1.4 The wave equation and boundary

kx2 + ky2 + kz2 = k 2 (1.24)

The velocity of propagation in free space is equal to c, the velocity of

H1t = H2t , E1t = E2t , ε1 E1n = ε2 E2n , µ1 H1n = µ2 H2n , (1.31)

1.5 Hollow metal waveguides

enough relative to the wavelength. How large should the dimensions be x

The message is clear. As the size of the waveguide shrinks there is

1.6 Refraction at a boundary: Snell’s law

Hy1 = A e −j (kz1 z + kx1 x) + B e j (kz1 z − kx1 x) , (1.39)

Fig. 1.3 Refraction of a TE (a) and a TM (b) wave

In medium 2 only one wave will propagate, the transmitted wave,

Hy2 = C e −j (kz2 z + kx2 x) (1.42)

kx1 = kx2 , (1.44)

Further conditions of matching are

possible directions for the wave vectors k1 and k2 . The geometrical

1.7 Fermat’s principle

waves. He is of a very generous nature and immediately decides to go

which may be seen to reduce to

n1 sin θ1 = n2 sin θ2 . (1.55)

1.8 The optical path and lens design

Fig. 1.6 (a) Lens design using the con-

OP1 = OC + nCP2 , (1.57)

which may be written in polar co-ordinates as