This action might not be possible to undo. Are you sure you want to continue?
Physics of negative refractive index materials
This article has been downloaded from IOPscience. Please scroll down to see the full text article.
2005 Rep. Prog. Phys. 68 449
(http://iopscience.iop.org/00344885/68/2/R06)
Download details:
IP Address: 202.3.77.11
The article was downloaded on 22/05/2010 at 14:05
Please note that terms and conditions apply.
View the table of contents for this issue, or go to the journal homepage for more
Home Search Collections Journals About Contact us My IOPscience
INSTITUTE OF PHYSICS PUBLISHING REPORTS ON PROGRESS IN PHYSICS
Rep. Prog. Phys. 68 (2005) 449–521 doi:10.1088/00344885/68/2/R06
Physics of negative refractive index materials
S Anantha Ramakrishna
Department of Physics, Indian Institute of Technology, Kanpur 208 016, India
Email: sar@iitk.ac.in
Received 3 September 2004
Published 18 January 2005
Online at stacks.iop.org/RoPP/68/449
Abstract
In the past few years, new developments in structured electromagnetic materials have given
rise to negative refractive index materials which have both negative dielectric permittivity
and negative magnetic permeability in some frequency ranges. The idea of a negative
refractive index opens up new conceptual frontiers in photonics. One muchdebated example
is the concept of a perfect lens that enables imaging with subwavelength image resolution.
Here we review the fundamental concepts and ideas of negative refractive index materials.
First we present the ideas of structured materials or metamaterials that enable the design
of new materials with a negative dielectric permittivity, negative magnetic permeability and
negative refractive index. We discuss how a variety of resonance phenomena can be utilized
to obtain these materials in various frequency ranges over the electromagnetic spectrum. The
choice of the wavevector in negative refractive index materials and the issues of dispersion,
causality and energy transport are analysed. Various issues of wave propagation including
nonlinear effects and surface modes in negative refractive materials (NRMs) are discussed. In
the latter part of the review, we discuss the concept of a perfect lens consisting of a slab of
a NRM. This perfect lens can image the farﬁeld radiative components as well as the near
ﬁeld evanescent components, and is not subject to the traditional diffraction limit. Different
aspects of this lens such as the surface modes acting as the mechanism for the imaging of
the evanescent waves, the limitations imposed by dissipation and dispersion in the negative
refractive media, the generalization of this lens to optically complementary media and the
possibility of magniﬁcation of the nearﬁeld images are discussed. Recent experimental
developments verifying these ideas are brieﬂy covered.
(Some ﬁgures in this article are in colour only in the electronic version)
00344885/05/020449+73$90.00 © 2005 IOP Publishing Ltd Printed in the UK 449
450 S A Ramakrishna
Contents
Page
1. Introduction 452
1.1. Objectives 452
1.2. Basic concepts and deﬁnitions 453
2. Materials with negative refractive index 455
2.1. Negative dielectric materials 456
2.1.1. Metals and plasmons at optical frequencies 456
2.1.2. Wiremesh structures as low frequency negative dielectrics 458
2.2. Materials with negative magnetic permeability 463
2.2.1. A stack of metal cylinders 463
2.2.2. The split ring resonator 464
2.2.3. Isotropic negative magnetic media 465
2.2.4. Wave dispersion in a resonant magnetic medium 466
2.2.5. The Swiss roll structure at radiofrequencies 467
2.2.6. Scaling to high frequencies 467
2.2.7. Other approaches 470
2.3. Negative refractive index materials 471
2.4. Homogenization and effective macroscopic material parameters 473
2.5. Causality and energy density in negative refractive media 476
3. Propagation of radiation in negative refractive media 477
3.1. Choice of the wavevector 477
3.2. Some effects of a reversed wavevector 479
3.2.1. The modiﬁed Snell’s law of refraction 479
3.2.2. The reversed Doppler shift 480
3.2.3. An obtuse angle cone for Cerenkov radiation 481
3.2.4. Reversed Goos–H¨ anchen shift 481
3.3. Phase velocity, group velocity and energy ﬂow 482
3.4. Backward wave structures and transmission lines 484
3.5. Negative refraction effect in photonic crystals 485
3.6. Nonlinear effects in negative refractive media 486
3.7. Surface electromagnetic modes in NRMs 488
4. The perfect lens 490
4.1. Nearﬁeld information and the diffraction limit 490
4.2. Pendry’s proposal 492
4.2.1. The quasistatic limit and the silver lens 494
4.2.2. The role of surface plasmons 495
4.2.3. The asymmetric lens 496
4.3. Limitations in real materials and imperfect NRMs 497
4.4. Numerical simulations and time evolution 499
5. Designing superlenses 501
5.1. Overcoming the limitations of real materials 501
5.2. The generalized perfect lens theorem 505
Physics of negative refractive index materials 451
5.3. The perfect lens in other geometries 506
5.3.1. Cylindrical lenses 507
5.3.2. A spherical lens 508
5.3.3. A perfect twodimensional corner lens 510
6. Experimental evidence for NRMs and superlenses 512
6.1. Experiments on negative refractive index 512
6.2. Photonic crystals and the Veselago ﬂat lens 513
6.3. Demonstrating superlenses 514
7. Conclusions 514
Acknowledgments 516
Appendix. Proof of the generalized perfect lens theorem 516
References 517
452 S A Ramakrishna
1. Introduction
In optics, the refractive index of a material is conventionally taken to be a measure of the
‘optical density’ and is deﬁned as
n =
c
v
, (1.1)
where c is the speed of light in vacuum and v is the speed of an electromagnetic plane wave in
the medium. From Maxwell’s equations the refractive index is given by the Maxwell relation,
n
2
= εµ, (1.2)
where ε is the relative dielectric permittivity and µ is the relative magnetic permeability of
the medium. Usual optical materials have a positive ε and µ, and n could easily be taken as
√
εµ without any problems. Although it was realized that the refractive index would have
to be a complex quantity to account for absorption and even a tensor to describe anisotropic
materials, the question of the sign of the refractive index did not arise. In 1967, Veselago [1]
ﬁrst considered the case of a mediumthat had both negative dielectric permittivity and negative
magnetic permeability at a given frequency and concluded that the medium should then be
considered to have a negative refractive index (i.e. the negative square root, n = −
√
εµ, had
to be chosen). Although Veselago went on to point out several interesting effects in NRMs,
such as a modiﬁed Snell’s law of refraction, a reversed Doppler shift and an obtuse angle for
Cerenkov radiation, in such media, his result remained an academic curiosity for a long time
as real materials with simultaneously negative ε and µ were not available. However, in the
last few years, theoretical proposals [2, 3] for structured photonic media whose ε and µ could
become negative in certain frequency ranges were developed experimentally [4, 5], and this
has brought Veselago’s result into the limelight. The striking demonstration by Pendry [6]
that NRMs can be used to make perfect lenses with resolution capabilities not limited by the
conventional diffraction limit has given an enormous boost to the interest in NRMs. This ﬁeld
has become a hot topic of scientiﬁc research and debate over the past four years.
1.1. Objectives
The physics of NRMs can be extremely counterintuitive. ‘Standard’ notions and intuitions of
electromagnetic theory can be misleading, and unexpected situations arise with NRMs. This
has mostly (for example the problems of the perfect lens) been responsible for the initial severe
criticismand debate [7–9] as well as the later explosion of research in this area. It is imperative
now to develop new intuitions and understanding to deal with these materials. Although many
of the developments are only recent, the basic conceptual issues have nowbeen widely debated
and reasonably clariﬁed. At the time of writing the theoretical foundations have been laid and
several experiments verifying the concepts have been performed, placing the ﬁeld on a sound
footing. Thus it appears to be the correct moment to review the conceptual issues of negative
refractive index materials.
Shorter reviews have been published elsewhere [10–12] which cover certain aspects of the
topic, or are at a semipopular level [13, 14]. Thus this will not be the ﬁrst review nor do we
expect it to be the last, given the rapid developments in this ﬁeld. But the hope is that it will
serve as a good reference point for the conceptual ideas involving negative refractive index
materials and the problemof the perfect lens, and as a good starting point for a newcomer to get
into this area of research. Given the exponentially rising number of publications, it will not be
possible to reviewor cite all of them. We will only concentrate here on the main developments.
Physics of negative refractive index materials 453
1.2. Basic concepts and deﬁnitions
The dielectric constant (ε) and the magnetic permeability (µ) characterize the macroscopic
response of a homogeneous medium to applied electric and magnetic ﬁelds. These are
macroscopic parameters because one usually only seeks timeaveraged and spatiallyaveraged
responses averaged over sufﬁciently long times and sufﬁciently large spatial volumes. All
that survive the averaging in macroscopic measurements are the frequency components of the
individual (atomic or molecular) oscillators driven by the external ﬁelds. This idea can now
be extended to a higher class of inhomogeneous materials where the inhomogeneities are on
lengthscales much smaller than a wavelength of the radiation but can be large compared with
atomic or molecular lengthscales. The radiation then does not resolve these individual meso
structures, but responds to the (atomically) macroscopic resonances of the structure. Such
materials have been termed metamaterials [13, 15] and can be characterized by macroscopic
parameters such as ε and µ that deﬁne their response to applied electromagnetic ﬁelds, much
like homogeneous materials. Metamaterials, in some sense, can be strictly distinguished
from other structured photonic materials, i.e. photonic crystals or photonic bandgap materials
[16, 17]. In the photonic crystals or bandgap materials the stops bands or bandgaps arise as
a result of multiple Bragg scattering in a periodic array of dielectric scatterers. In fact, the
periodicity of the structure here is of the order of the wavelength, and hence homogenization
in this sense cannot be carried out. In metamaterials the periodicity is by comparison far less
important and all the properties mainly depend on the single scatterer resonances.
All causal materials are dispersive, i.e. the dielectric permittivity and the magnetic
permeability are, in general, complex functions of the frequency. This is because the
polarizations in such media do not respond instantaneously to the applied ﬁelds but depend
on the history of the applied ﬁelds. This becomes particularly visible when the applied ﬁelds
have frequencies close to a resonant frequency of the material oscillators. In fact, the real
parts and the imaginary parts of the material parameters (ε(ω) and µ(ω)) are related to each
other by the famous Kramers–Kronig relations [18]. Noting that the imaginary parts of ε
and µ relate directly to the absorption of electromagnetic radiation in the material, we realize
that dispersion and dissipation in thermodynamic media always accompany each other. As
with any resonance, the response follows the applied ﬁeld at frequencies below the resonance,
and above the resonance the response is antiphased with respect to the applied ﬁeld. If now
the resonance can be made sharp enough, it will be possible to drive the real parts of the
effective ε or µ negative. This underdamped, overscreened response is responsible for the
negative material parameters. There is no fundamental objection to the real parts of the ε and
µ being negative [19]. In media at thermodynamic equilibrium, there is, however, a restriction
that the imaginary parts of the corresponding ε or µ are not to be negative. This is so that the
total absorbed energy in a volume V of the medium,
_
V
d
3
r
_
∞
−∞
ω[Im(ε(ω))[
¯
E(¯ r, ω)[
2
+ Im(µ(ω))[
¯
H(¯ r, ω)[
2
]
dω
2π
, (1.3)
is positive deﬁnite [19]. Thus, media withnegative ε andµare causal, but necessarilydispersive
and dissipative.
Although most materials exhibiting a good electric response can be found at almost any
frequency from radiofrequencies to the ultraviolet frequencies, the magnetic response of
most materials is limited to low microwave frequencies. Magnetic polarization usually results
from either unpaired electron spins or orbital electron currents, and the collective excitations
of these usually tend to occur at low frequencies. Some ferromagnetic, ferrimagnetic and
antiferromagnetic materials exhibit some magnetic activity at even frequencies of hundreds of
gigahertz [20]. But these are rare and usually have narrowbandwidths. But nowthe possibility
454 S A Ramakrishna
> 0, > 0
Many metals (UV  Optical)
Thin wires structures (GHz)
Evanescent decaying waves
E
H
k
k
E
H
S
S
Right handed propagating waves
Ordinary optical materials
Evanescent decaying waves
Structured materials
Magnetic plasma ?
> 0, < 0
materials (upto GHz)
Some natural magnetic
Artificial Meta–materials
Negative refractive index
< 0, < 0
Left handed propagating waves
Electrical Plasma
< 0, > 0
Figure 1. A schematic showing the classiﬁcation of materials based on the dielectric and magnetic
properties. The wavy lines represent materials that allow propagating waves, and the axes set in
quadrants 1 and 3 show the right and lefthanded nature of E, H and k vectors. The waves in
quadrants 2 and 4 decay evanescently inside the materials, which is depicted schematically. S is
the Poynting vector.
of artiﬁcially structuring materials at micro and nanoscales can enable us to generate a
variety of metamaterials with magnetic activity at almost any frequency we wish—from a
few hertz [21] to nearinfrared frequencies [22] including the radio frequencies [3, 23, 24], the
microwave frequencies [3, 4, 25] and the terahertz frequencies [27, 28] in between. Thus even
magnetic activity, let alone negative magnetic permeability, is special at high frequencies.
Consider the Maxwell’s equation for a plane harmonic wave exp[i(k · r −ωt )]:
k E = ωµ
0
µH, (1.4)
k H = −ωε
0
εE, (1.5)
where Eand Hare the electric and magnetic ﬁelds. If we take a mediumwith negative real parts
of ε and µ with the imaginary parts being small (negligibly small for the time being) at some
frequency (ω), then we realize that the vectors E, Hand k will now form a lefthanded triad. It
is for this reason that such materials are also popularly termed lefthanded materials, although
this does risk a confusion with the terminology of chiral optical materials. Using the deﬁnition
that the wavevector k = nω/c ˆ n, where ˆ n is the unit vector along E H, it appears that the
refractive index in such media with Re(ε) < 0 and Re(µ) < 0 is also negative. Such materials
permit propagating waves with a reversed phase vector (k)
1
compared with media with only
one of Re(ε) or Re(µ) negative which do not allow any propagating modes (all waves decay
evanescently in such media from the point of injection as k
2
< 0). Note that in the case of the
NRMs, the Poynting vector (S = EH) and the phase vector, k, are antiparallel. In ﬁgure 1,
we show the four quadrants in the Re(ε)–Re(µ) plane into which we can conveniently classify
electromagnetic materials. The behaviour of the waves in each of the quadrants is qualitatively
different: materials that fall in the ﬁrst quadrant allow the usual righthanded electromagnetic
propagating waves, the materials that fall in the second and fourth quadrants do not allow any
propagating waves inside them (all electromagnetic radiation is evanescently damped in these
media) and the NRMs that fall in the third quadrant allowlefthanded propagating waves inside
them. This follows from the dispersion for the wavevector
k · k = εµ
ω
2
c
2
. (1.6)
1
This has prompted some authors to call these backward wave media [29] or negative phase velocity media [10].
Some authors have preferred to term such media double negative media, referring to the negative ε and µ [30].
Physics of negative refractive index materials 455
Making Re(ε) or Re(µ) negative not only makes electromagnetic radiation decay
exponentially inside them, but also renders the materials rather very special. For example,
we know that metals whose dielectric permittivity is dominated by the plasmalike response
of the free electron gas,
ε(ω) = 1 −
ω
2
p
ω(ω + iγ )
, (1.7)
where ω
p
is the bulk plasma frequency and γ is the damping constant, have negative dielectric
permittivity at UV–optical frequencies. These materials can support a host of resonant states
localized at their surfaces, known as the surface plasmons [31, 32]. These surface plasmons
can resonantly interact with radiation on structured surfaces, giving rise to a wide variety of
novel optical phenomena [33] and have given rise to the newﬁeld of plasmonics [34]. This is a
generic effect of having a negative dielectric permittivity, and materials with negative magnetic
permeability can also be expected to support the analogous surface plasmons of a magnetic
nature!
These resonant states can affect the propagation of the electromagnetic radiation in a
very profound manner and are deeply involved in the mechanism of the perfect lens [6] made
of an NRM. Such a lens is not limited in its resolution to superwavelength lengthscales
according to the standard principles of diffraction theory [35]. In fact this has been one of the
most fundamental new effects associated with NRMs thus far and has been a topic of intense
debate [8, 30] and much activity. This issue will be treated in detail in the latter sections of
this review.
We will now proceed to examine all these concepts in detail. The review can be broadly
split into two parts: sections 2 and 3, dealing with the general properties of negative refractive
index media, and sections 4 and 5, dealing with the speciﬁc problem of the perfect lens that
can be formed from negative refractive index media. The experimental efforts in this area are
covered in section 6.
2. Materials with negative refractive index
There is nothing fundamentally objectionable about negative material parameters such as
the dielectric permittivity and the magnetic permeability. But ﬁnding a material that has
these properties is a more difﬁcult task and it was because of the lack of a material that
exhibited negative values of these parameters at a common frequency band that Veselago’s
early result did not arouse much interest for a long time. Pendry’s seminal work on structured
materials which display negative dielectric permittivity [2, 43] and magnetic permeability [3],
along with the technological capability to make structures at microscopic lengthscales,
has made it possible and interesting to create experimentally such materials at a variety of
wavelengths. All the NRMs that have been developed today are obtained by putting together
two structured materials that show separately a negative dielectric permittivity and negative
magnetic permeability. Although it is not immediately obvious, the resulting composite
structure possesses a negative refractive index. In this section, we will ﬁrst separately discuss
the negative dielectric materials and the negative magnetic materials before we discuss the
composite that behaves as an NRM. We will then address the issues of homogenization
and causality that confront us in the context of such materials. In general, we will strictly
not consider photonic bandgap materials, which can display some of the effects of negative
refraction [36] at wavelengths which are of the same order of the lengthscales of the structures
in the metamaterials, as negative refractive index materials. There are signiﬁcant problems
of homogenization that are involved here.
456 S A Ramakrishna
2.1. Negative dielectric materials
2.1.1. Metals and plasmons at optical frequencies. At optical frequencies many metals have
a negative dielectric permittivity when the conduction electrons in the metals can be assumed
to be reasonably free in a background of static positive ion cores, the overall system being
charge neutral. It is this plasmalike behaviour that is responsible for a negative dielectric
permittivity at frequencies less than the plasma frequency. Consider the equation of motion
for an electron in an applied time harmonic electromagnetic ﬁeld with angular frequency ω
0
,
m¨ r + mγ ˙ r = −eEexp(−iω
0
t ), (2.1)
where m is the mass of the electron and mγ is a phenomenogical damping (viscous) force
on the electron due to all inelastic processes. It is assumed that the wavelength of light is
large compared with the distance travelled by the electron, so that it effectively sees a spatially
constant ﬁeld and the velocities involved are sufﬁciently low so that the magnetic ﬁeld can be
neglected. This yields a polarization per volume in the medium of
P = (ε −1)ε
0
E = −ner = −
ne
2
/mE
ω(ω + iγ )
, (2.2)
where n is the number density of the conduction electrons and each electron is assumed to
contribute independentlytothe polarization. Fromthis we obtainthe relative dielectric constant
of equation (1.7), where ω
p
= ne
2
/ε
0
m is the plasma frequency. The dielectric permittivity is
negative up to ω
p
and the plasma shields the interior from electromagnetic radiation. Above
ω
p
the medium behaves as an ordinary positive dielectric. This is the theory of Drude and
Lorentz for the dispersion characteristics of a plasma [18, 35]. Substituting the expression for
the dielectric constant into Maxwell’s equation, we obtain the dispersion for light in a metal:
k
2
c
2
+ ω
2
p
= ω
2
. (2.3)
Thus, the waves below the plasma frequency correspond to the negative energy solutions, and
above the plasma frequency, it appears as if the transverse modes of light had a ﬁnite rest mass
of m
0
= ¯ hω
p
/c
2
.
The plasma frequency also has a physical manifestation as a collective excitation of the
electron gas. If the electron gas is displaced by a small distance in a ﬁnite metal, the excess
negative (electronic) charge and the positive (background) charge that accumulate at the edges
provide a restoring force proportional to the displacement and cause the entire electron gas to
oscillate at frequency ω
p
[37]. Hence ω
p
is known as the plasmon frequency. This is the
dispersionless plasmon mode that occurs at ω = ω
p
.
As mentioned before, rendering the dielectric constant negative makes it possible for the
surface to support resonances called the surface plasmons [32]. Consider the interface (z = 0
plane) between a metal with a dielectric constant ε
−
and a positive dielectric medium(vacuum)
with dielectric permittivity ε
+
(=+1) as shown in ﬁgure 2. Now the ﬁelds for a ppolarized
light are
E = E
0
exp[i(k
x
x + k
y
y −ωt ) −κ
z+
z] ∀z > 0, (2.4)
E = E
0
exp[i(k
x
x + k
y
y −ωt ) + κ
z−
z] ∀z < 0, (2.5)
where k
2
x
+ k
2
y
−κ
2
z±
= ε
±
µω
2
/c
2
are good solutions to the Maxwell equations when
κ
z+
ε
+
+
κ
z−
ε
−
= 0. (2.6)
These are collective excitations of electrons with the charges displaced parallel to the (real
part of the) wavevector. Thus we have a mode with longitudinal components whose ﬁelds
Physics of negative refractive index materials 457
z
x
y
k
x
–k
x
+ +
_ _ __ _ _
+
–
+ +
Figure 2. A schematic showing the surface plasmon on an interface between a negative and a
positive dielectric media and the associated charge density ﬂuctuations. The exponential decay
of the ﬁelds normal to the surface and the propagating nature along the surface are depicted
schematically.
L C C
–
Figure 3. A capacitor and an inductor form a resonant circuit that can oscillate at ω
0
= 1/
√
LC.
A capacitor ﬁlled with a negative dielectric has negative capacitance, acts as an inductor and can
resonate with another usual capacitor.
decay exponentially into the metal and the dielectric on either side of the interface. This
charge density wave thus lives on the surface of the metal and is called a surface plasmon.
Equation (2.6) yields the dispersion
k
x
=
ω
c
_
ε
+
ε
−
ε
+
+ ε
−
_
1/2
. (2.7)
For a metal with ε
−
(ω) given by equation (1.7) and setting ε
+
= 1 for vacuum, we note that
k
x
> ω/c for the surface plasmon (see ﬁgure 22). Thus it will not be possible to excite the
surface plasmon on a perfectly ﬂat surface using propagating modes of light, and a coupling
mechanism will have to be provided by mechanisms such as the surface roughness, a grating
structure or a dielectric coupler to vacuum such as a hemisphere or a prism. Complicated
surface structures can generate a variety of surface plasmon dispersions, and this has given
rise to new optical phenomena such as the extraordinary transmission of light through sub
wavelength sized holes [38]. Next note that for reasonably large wavevectors the surface
plasmons are nearly degenerate with a frequency of ω
p
/
√
2. This is the reason why the
interaction of the radiation with the resonant surface plasmons is very rich and difﬁcult to
model: the radiation couples with the metal surface at all lengthscales. This remains true down
to a nanometre or so when nonlocal dispersive effects kick in due to inadequate electronic
screening inside the metal.
The negative dielectric constant is central tothe nature of resonant interactions of structured
metal surfaces with radiation. Pendry [33] provides a very insightful explanation for this. We
know that a capacitor can be formed by two parallel conducting plates with an insulating
dielectric placed in between. Filling up the gap with a negative dielectric material instead
would nowlead to a capacitor with negative capacitance, which is the same as being an inductor.
Thus two capacitors in a circuit, one ﬁlled with a positive dielectric and the other ﬁlled with
a negative dielectric material, can become resonant (see ﬁgure 3). It is as if in a structured
458 S A Ramakrishna
metallic structure the metal pieces form the capacitors ﬁlled by the negative dielectric material
and the holes in between form the capacitors ﬁlled with the positive dielectric material. The
resulting system is a very complicated circuit with many L–C resonances.
For many metals, the plasma frequency is at ultraviolet frequencies and γ is small
compared with ω
p
(see [32]). Thus we have many examples of materials with a negative
dielectric constant at optical frequencies. However, the dissipation in most metals is large, and
we have trouble when we try to extend this behaviour to lower frequencies where ω ∼ γ . Then
it can hardly be claimed that the dielectric constant is an almost real and negative number. The
dissipation dominates all phenomena and we no longer have the behaviour of a good plasma.
It must be pointed out that negative dielectric permittivity can also be obtained in more
ordinary dielectric media with bound charges, within a frequency band above a resonance
frequency. Consider a medium in which the electrons are bound to positive nuclei and its
response to an applied electromagnetic ﬁeld. For small displacements of the electrons there
is a restoring force on the electron, −mω
2
0
r. Including this term in equation (2.1), one obtains
the resonant Lorentz dielectric permittivity,
ε = 1 +
ne
2
/(ε
0
m)
ω
2
0
−ω
2
−iγ ω
, (2.8)
where n is now the total density of bound electrons (here it is assumed simplistically that all
the electrons are bound with the same strength). Setting ω
0
= 0, we immediately obtain the
Drude form for ε for a good plasma. Note now that the dielectric constant can be negative in
a small frequency range above ω
0
if the resonance is sharp enough. In fact, one can make use
of resonant effects such as electromagnetically induced transparency or similar effects based
on atomic coherences [39, 40] to drive the dielectric constant negative over selected small
frequency ranges.
2.1.2. Wiremesh structures as low frequency negative dielectrics. Metaldielectric
composites have long been studied for their rich electrodynamic response (see [41] and
references therein). For example, composites of randomly oriented long conducting ﬁbres
have been known to exhibit very high values of permittivity even at low concentrations [42].
Effective mediumtheories have been developed to describe these systems when the wavelength
of the incident radiation is much larger than the intrinsic lengthscales of the structure.
However, the radiation usually only probes the end surfaces of the metallic structures, and
it is hard to make it penetrate well into the bulk of the structure for the appearance of a three
dimensional effective mediumto hold true in many cases. Pendry et al [2, 43] and Sievenpiper
et al [44] independently demonstrated that metallic wiremesh structures have a lowfrequency
stop band from zero frequency up to a cutoff frequency which they attributed to the motion of
electrons in the metal wires. The metamaterial of Pendry et al, consisting of very thin wires,
is structured on truly subwavelength lengthscales and can be effectively homogenized, while
it is not clear that this holds for the structure of Sievenpiper. The low frequency stop band
can be attributed to an effective negative dielectric permittivity and provides us with a way of
obtaining negative dielectrics at even microwave frequencies. The idea is to obtain negative
permittivity materials at low frequencies by literally doping the vacuum with metal.
Consider an array of inﬁnitely long, parallel and very thin metallic wires of radius r placed
periodically at a distance a in a square lattice with a ¸ r as shown in ﬁgure 4. The electric
ﬁeld is considered to be applied parallel to the wires (along the Z axis) and we will work
in the quasistatic limit when the wavelength of the radiation λ ¸ a ¸ r. The electrons
are conﬁned to move within the wires only, which has the ﬁrst effect of reducing the effective
electron density as the radiation cannot sense the individual wire structure but only the average
Physics of negative refractive index materials 459
Z
X
Y
a
a
r
Figure 4. An array of inﬁnitely long thin metal wires of radius r and a lattice period of a behaves
as a low frequency plasma for the electric ﬁeld oriented along the wires.
charge density. The effective electron density is immediately seen to be
n
eff
=
πr
2
a
2
n, (2.9)
where n is the actual density of conduction electrons in the metal.
There is a second equally important effect to be considered. The thin wires have a large
inductance, and it is not easy to change the currents ﬂowing in these wires. Thus, it appears as
if the charge carriers, namely the electrons, have acquired a tremendously large mass. To see
this, consider the magnetic ﬁeld at a distance ρ from a wire. On an average we can assume
a uniform D ﬁeld within the unit cell. But the current density is not uniform, leading to a
nonzero nonuniform magnetic ﬁeld that is large close to the wires which contributes to most
of the ﬂux. By symmetry there is a point of zero ﬁeld in between the wires and hence we can
estimate the magnetic ﬁeld along the line between two wires as
H(ρ) =
ˆ
φI
2π
_
1
ρ
−
1
a −ρ
_
. (2.10)
The vector potential associated with the ﬁeld of a single inﬁnitely long currentcarrying
conductor is nonunique unless the boundary conditions are speciﬁed at deﬁnite points. In
our case, we have a periodic medium that sets a critical length of a/2. We can assume that the
vector potential associated with a single wire is
A(ρ) =
ˆ zI
2π
ln
_
a
2
4ρ(a −ρ)
_
∀ρ <
a
2
, (2.11)
= 0 ∀ρ >
a
2
. (2.12)
This choice avoids the vector potential of one wire overlapping with another wire and thus the
mutual induction between the adjacent wires is addressed to some extent. Noting that r _ a
460 S A Ramakrishna
by about three orders of magnitude in our model and that the current I = πr
2
nev, where v is
the mean electron velocity, we can write the vector potential as
A(ρ) =
µ
0
πr
2
nev
2π
ln
_
a
ρ
_
ˆ z. (2.13)
This is a very good approximation in the mean ﬁeld limit. We have considered only two wires,
and the lattice actually has fourfold symmetry. The actual deviations from this expression are
much smaller than in our case. We note that the canonical momentum of an electron in an
electromagnetic ﬁeld is p + eA. Thus assuming that the electrons ﬂow on the surface of the
wire (assuming a perfect conductor), we can associate a momentum per unit length of the
wire of
p = πr
2
neA(r) =
µ
0
π
2
r
4
n
2
e
2
v
2π
ln
_
a
r
_
= m
eff
πr
2
nv (2.14)
and thus an effective mass of
m
eff
=
µ
0
πr
2
ne
2
2π
ln
_
a
r
_
, (2.15)
for the electron.
Thus, assuming a longitudinal plasmonic mode for the system, we have
ω
p
=
n
eff
e
2
ε
0
m
eff
=
2πc
2
a
2
ln(a/r)
(2.16)
for the plasmonfrequency. We note that a reducedeffective electrondensityanda tremendously
increased effective electronic mass would immediately reduce the plasmon frequency for this
system. Typicallyone canchoose r = 1 µm, a = 10 mmandaluminiumwires (n = 10
29
m
−3
),
which gives an effective mass of
m
eff
= 2.67 10
−26
kg, (2.17)
that is almost 15 times that of a proton, and a plasma frequency of about 2 GHz! Thus, we
have succeeded in obtaining a negative dielectric material at microwave frequencies.
Note that the ﬁnal expression for the plasma frequency in equation (2.16) is independent
of the microscopic quantities such as the electron density and the mean drift velocity. It only
depends on the radius of the wires and the spacing, suggesting that the entire problem can be
recast in terms of the capacitances and inductances of the problem. This approach has been
taken [45, 46] and we present it below for the sake of completeness. Consider the current
induced by the electric ﬁeld along the wires, related by the total inductance (self and mutual)
per unit length (L):
E
z
= +iωLI = iωLπr
2
nev, (2.18)
noting that the polarization per unit volume in the homogenized medium is
P = −n
eff
er =
n
eff
ev
iω
= −
E
z
ω
2
a
2
L
, (2.19)
where n
eff
= πr
2
n/a
2
as before. We can estimate the inductance, L, by calculating the
magnetic ﬂux per unit length passing through a plane between the wire and the point of
symmetry between itself and the next wire where the ﬁeld is zero:
φ = µ
0
_
a/2
r
H(ρ) dρ =
µ
0
I
2π
ln
_
a
2
4r(a −r)
_
. (2.20)
Physics of negative refractive index materials 461
Z
X
Y
Figure 5. A threedimensional lattice of thin conducting wires behaves like an isotropic low
frequency plasma.
Noting = LI, and the polarization P = (ε − 1)ε
0
E
z
, where ε is the effective permittivity,
we obtain in the limit r _ a,
ε(ω) = 1 −
2πc
2
ω
2
a
2
ln(a/r)
, (2.21)
which is identical to the relation obtained in the plasmon picture. However, we lose the physical
interpretation of a low frequency plasmonic excitation here.
It is simple to add the effects of ﬁnite conductivity in the wires which had been neglected
in the above discussion. The electric ﬁeld and the current would now be related by
E
z
= iωLI + σπr
2
I, (2.22)
where σ is the conductivity, which modiﬁes the expression for the dielectric permittivity to
ε = 1 −
ω
2
p
ω(ω + i(ε
0
a
2
ω
2
p
/πr
2
σ))
. (2.23)
Thus the ﬁnite conductivity of the wires contributes to the dissipation, showing up in the
imaginary part of ε. For aluminium, the conductivity is σ = 3.65 10
7
−1
m
−1
that yields
a values of γ = 0.1ω
p
which is comparable to the values in real metals [32]. Thus the low
frequency plasmon is sufﬁciently stable against absorption to be observable.
In our above discussion, we considered only wires pointing in the zdirection. This makes
the medium anisotropic with negative ε only for waves with the electric ﬁeld along the z
direction. The medium can be made to have a reasonably isotropic response by considering
a lattice of wires oriented along the three orthogonal directions as shown in ﬁgure 5. In the
limit of large wavelengths, the effective medium appears to be isotropic as the radiation fails
to resolve the underlying cubic symmetry yielding a truly threedimensional low frequency
plasma. For very thin wires, the polarization in the direction orthogonal to the wires is small
and can be neglected. Thus the waves only sense the wires parallel to the electric ﬁeld and
correspondingly have a longitudinal mode. The effects of the connectivity of the wires along
different directions at the edges of the unit cell have also been examined [43]. A wire mesh
with nonintersecting wires was also shown to have a negative ε at low frequencies below the
plasma frequency, but had strong spatial dispersion for the transverse modes above the plasma
frequency. This issue of spatial dispersion has been studied more recently by Belov et al [46].
Some objections were raised to Pendry’s original propositions. Mikhailov [47] assumed
the results for a onedimensional plasmon polarition on a metal wire and neglected the essential
462 S A Ramakrishna
threedimensional nature of the problem[48]. Walser et al [49] incorrectly assumed the Drude
form of the dielectric constant of the metal for the dielectric response of a single metallic wire
as well and claimed that dissipation would damp out all phenomena. They also opined that
the interpretation of a heavy electronic effective mass was due to the gauge dependent vector
potential which could be changed at will. However, their results completely neglect the fact
that there are two lengthscales of the problem and that the wires are very thin compared with
the wavelength. Thus no homogenization could be carried out. Further, the claim that any
gauge term could be added to the vector potential without affecting the problem is incorrect as
that would change the electric ﬁeld and the plasmon would appear to have a position dependent
mass [50]. The use of the Coulomb gauge does spoil the manifest covariance, but insofaras
the velocities of the charges involved are small compared with c, the speed of light in vacuum,
it is the most suitable gauge.
The effects of having continuous metal wires have been studied in [43], where it was shown
that the response of wires systems with periodic sections of the wires removed was more akin
to that of periodically placed interacting dipoles [51]. Lowfrequency transverse modes appear
well below the effective plasma frequency for the corresponding continuous wire structure,
but propagating modes appeared above them. The additional capacitance between the edges
of the wires and the selfcapacitance of the edges can be considered easily in our model as a
term additional to the impedance in equation (2.22). The additional capacitance gives rise to
a L–C resonance at a ﬁnite frequency, and one has a resonant Lorentz form of the dispersion
for the dielectric permittivity. For a sufﬁciently sharp resonance and low wire resistivity, one
can again obtain a negative dielectric permittivity within a frequency band above the resonant
frequency. This response for periodic ﬁnite wire strips has also been considered by Koschny
et al in [52], where it was also shown that the corresponding magnetic permeability, however,
showed an antiresonant dispersion with a negative imaginary part of µ when the real part
of ε was negative. We will discuss this latter aspect later. Makhnovskiy and Panina [53]
have considered the possibility of using the magnetoimpedance effect to change the surface
impedance of thin ferromagnetic wires using an applied magnetic ﬁeld to generate negative
dielectric media whose plasma frequency can be tuned by the magnetic ﬁeld. Again the
response of short wires gave rise to a resonant effective permittivity, with a ﬁnite frequency
band for negative permittivity. Simovsky and He [54] considered the response of an array
of shaped particles and found that the array showed the resonant dielectric response of
periodically cut wires.
Dense wire media have been of interest in the electrical engineering community for a
long time for artiﬁcial impedance surfaces [55, 56]. But they were usually considered when
the wavelength was comparable with the period of the lattice and the radius of the wires.
A transmission line model [57] gives a result of
ε =
ε
0
(ka)
2
_
cos
−1
_
cos(ka) +
π
ka ln(a/(2πr))
sin(ka)
__
2
, (2.24)
which was shown to agree with our expression for a low frequency plasmonic medium in
the limit of very thin wires [45]. Pitarke et al [58] have considered the photonic response
of metallic cylinders embedded in a dielectric host and have agreement with the classical
Maxwell–Garnet result when the wavelength is at least twice as large as the spacing between
the cylinders. However, these cases do not ideally fall in the category of effective media. The
scattering turns out to be quite important at higher frequencies.
The photonic response of superconducting cylinders made of high T
c
materials has been
considered [59], and it was shown that the system has a low frequency cutoff that is much
smaller than that of the bulk superconductor. The role of a plasmon in a superconductor as
Physics of negative refractive index materials 463
y
z
x
r
d
(c) (a)
(b)
H
r
d
−
−−
−
−
−
+ +
+
+
+
−
−
+
+
+
+
−
− −
−
−
−
+
+ +
+ +
+
Figure 6. (a) An applied magnetic ﬁeld along the axes of a stack of conducting cylinders induces
circumferential currents that shield the interior. The resulting effective medium is diamagnetic.
(b) The splitring structure: the capacitance across the rings nowcauses the structure to be resonant.
(c) The Swiss roll structure: more capacitance for lower frequency operation.
a massive Higgs boson has been stressed by Anderson [60]. Pendry [2, 43] has pointed out
the importance of the fact that in such thin wire superconductor structures, the frequency of
the plasmon could well be smaller than the superconducting gap, making such media of great
interest in a fundamental manner.
2.2. Materials with negative magnetic permeability
As mentioned before, the magnetic activity in most materials tends to tail off at high frequencies
of even a few gigahertz. So it is indeed a challenge even to obtain magnetic activity, let alone
negative magnetic permeability, at microwave frequencies and beyond. In fact, Landau and
Lifschitz [19] give a very general argument as to howthe magnetic activity arising fromatomic
orbital currents should be negligible at optical frequencies if one could neglect the polarization
currents. In the case of nanometallic structures, the magnetic moments of induced real and
displacement current distributions can actually contribute to an effective magnetization if the
electric polarizability of the corresponding medium is small.
2.2.1. A stack of metal cylinders. Most materials have a natural tendency to be diamagnetic
as a consequence of Lenz’s law. Consider the response of a stack of metallic cylinders to an
incident electromagnetic wave with the magnetic ﬁeld along the axis of the cylinders as shown
in ﬁgure 6. The cylinders have a radius r and are placed in a square lattice with period a. The
oscillating magnetic ﬁeld along the cylinders induces circumferential surface currents which
tend to generate a magnetization opposing the applied ﬁeld. The axial magnetic ﬁeld inside
the cylinders is
H = H
0
+ j −
πr
2
a
2
j, (2.25)
where H
0
is the applied magnetic ﬁeld and j is the induced current per unit length of the
cylinder. The third term is due to the depolarizing ﬁeld which is assumed to be uniform as the
cylinders are inﬁnitely long. The emf around the cylinder can be calculated from Lenz’s law
and is balanced by the Ohmic drop in potential:
iωµ
0
πr
2
_
H
0
+ j −
πr
2
a
2
j
_
= 2πrρj, (2.26)
464 S A Ramakrishna
where time harmonic ﬁelds are assumed and ρ is the resistance per unit length of the cylinder
surface. The frequencies are assumed to be sufﬁciently low so as to have only a small skin
effect. Now we homogenize the system of cylinders by adopting an averaging procedure
(discussed in section 2.4) that consists of averaging the magnetic induction, B, over the area
of the unit cell while averaging the magnetic ﬁeld, H, over a line along the edge of the unit
cell. The averaged magnetic ﬁeld is
B
eff
= µ
0
H
0
, (2.27)
while the averaged H ﬁeld outside the cylinders is
H
eff
= H
0
−
πr
2
a
2
j. (2.28)
Using the above, we obtain the effective relative magnetic permeability:
µ
eff
=
B
eff
µ
0
H
eff
= 1 −
πr
2
/a
2
1 + i2ρ/(µ
0
ωr)
. (2.29)
Thus, the real part of µ
eff
is always less than 1 (diamagnetic) and greater than 0 here. This
diamagnetic screening effect has been known for some time for superconducting cylindrical
shells [61], and a diamagnetic effective medium is also obtained with percolation metallo
dielectric composites [41].
2.2.2. The split ring resonator. In the case of the cylinders, the induced currents made it
appear as if magnetic monopoles were ﬂowing up and down the cylinders. But the problemwas
that the systemonly had an inductive response (the monopoles equivalently had no inertia). By
introducing capacitive elements into the system, a rich resonant response can be induced [3].
Consider an array of concentric cylindrical metallic shells with a gap in them as shown in
ﬁgure 6(b). This has become well known subsequently as the split ring resonator—SRR for
short. The SRRs are the basis of most of the metamaterials exhibiting negative magnetic
permeability today [4, 5, 25–27, 62–64].
The SRR works on the principle that the magnetic ﬁeld of the electromagnetic radiation
can drive a resonant L–Ccircuit through the inductance, and this results in a dispersive effective
magnetic permeability. The induced currents ﬂow in the directions indicated in ﬁgure 6, with
charges accumulating at the gaps in the rings. The large gap in each ring prevents the current
from ﬂowing around in a single ring, and the circuit is completed across the small capacitive
gap between the two rings. Assuming that the gap (d) is very small compared with the radius
(r) and that the capacitance due to the large gaps in any single ring is negligible, we balance
the emf around the circuit as
−iωµ
0
πr
2
_
H
0
+ j −
πr
2
a
2
j
_
= 2πrρj −
j
iωC
, (2.30)
where the effective capacitance C = ε
0
επr/(3d) and ε is the relative dielectric permittivity
of the material in the gap. It is assumed that the potential varies linearly with the azimuthal
angle around the ring. Proceeding as before, we obtain the effective permeability as
µ
eff
= 1 −
πr
2
/a
2
1 −(3d/µ
0
ε
0
επ
2
ω
2
r
3
) + i(2ρ/µ
0
ωr)
= 1 +
f ω
2
ω
2
0
−ω
2
−iω
(2.31)
and thus we have a resonant form of the permittivity with a resonant frequency of
ω
0
=
_
3d
µ
0
ε
0
επ
2
r
3
_
1/2
(2.32)
Physics of negative refractive index materials 465
4 5 6 7 8 9
Frequency (GHz)
5
0
5
R
e
(
)
= 0.2 Ω/m
= 10 Ω/m
= 1.0 Ω/m
4 5 6 7 8 9
Frequency (GHz)
0
4
8
12
I
m
(
)
ρ
ρ
ρ
= 0.2 Ω/m
= 10 Ω/m
= 1.0 Ω/m
ρ
ρ
ρ
Figure 7. The behaviour of (a) the real and (b) the imaginary parts of the effective magnetic
permeability for a system of split cylinders with r = 1.5 mm, d = 0.2 mm and the magnetic
ﬁeld along the axis for different values of ρ. The system has negative magnetic permeability for
6.41 < ω < 7.56 GHz. As the resistance increases, the response of the system tends to tail off.
that arises fromthe L–Cresonance of the system. The factor f = πr
2
/a
2
is the ﬁlling fraction
of the material. For frequencies larger than ω
0
, the response is out of phase with the driving
magnetic ﬁeld and µ
eff
is negative up to the ‘magnetic plasma’ frequency of
ω
m
=
_
3d
(1 −f )µ
0
ε
0
επ
2
r
3
_
1/2
, (2.33)
assuming the resistivity of the material is negligible, and it is seen that the ﬁlling fraction plays
a fundamental role in the bandwidth over which µ < 0. The dielectric permittivity of the
embedding medium, ε, can be used to tune the resonant frequency, which is also borne out in
the numerical calculations of [65]. A ﬁnite resistivity, in general, broadens the peak, and in the
case of very resistive materials the resonance is so highly damped that the region of negative
µ can disappear altogether as shown in ﬁgure 7.
The resonance frequency and the negative µband can be varied by changing the inductance
(area) of the loop and the capacitance (the gap width, d, or the dielectric permittivity, ε, of the
material in the capacitive gap) of the system. For typical sizes of r = 1.5 mm, a = 5 mm,
d = 0.2 mm, we have a resonance frequency of ω
0
= 6.41 GHz, and a ‘magnetic plasma
frequency’ ω
m
= 7.56 GHz. The dispersion in µ
eff
is shown in ﬁgure 7. We note that
µ
eff
. 1 − πr
2
/a
2
asymptotically at large frequencies. This is due to the assumption of a
perfect conductor in our analyses. Let us also note that µ
eff
can attain very large values on the
low frequency side of the resonance. Thus, the effective medium of SRR is going to have a
very large surface impedance Z =
√
µ
eff
/ε
eff
. Such large impedances have also been obtained
by capacitively loaded structured surfaces [67, 68].
2.2.3. Isotropic negative magnetic media. In our discussion till now, we assumed a mediumof
inﬁnitely long split cylinders which shows magnetic properties when the magnetic ﬁeld is along
the axes of the cylinders. Obviously the system does not show any magnetic activity when the
magnetic ﬁeld is oriented perpendicular to the cylinders, rendering the medium uniaxial. To
overcome this limitation, a threedimensional photonic material was proposed [3] which would
have a reasonably isotropic response. The essential building blocks of this system are shown
in ﬁgure 8(a) which consists of interleaved orthogonal planes of planar SRRs. Assuming that
the magnetic ﬂux through the stacked split rings remains the same as for a cylinder, this system
would have the same magnetic response in all the orthogonal directions, and the effective
magnetic permeability has been calculated as
µ
eff
(ω) =
πr
2
/a
2
1 −(3l/µ
0
ε
0
π
2
ω
2
r
3
C) + i(2lρ/µ
0
ωr)
, (2.34)
466 S A Ramakrishna
Figure 8. (a) The unit cell of a threedimensional SRR medium with isotropic magnetic response.
(b) The threedimensional SRR scatterer proposed in [63] (reprinted with permission from [63].
Copyright 2002 American Institute of Physics).
where C = ε
0
/π ln(2τ
c
/d) is the capacitance between two parallel sections of planar rings of
thickness τ
c
. Obviously if the rings are very thin, the capacitance across the ends of the single
rings (which we neglect) can become comparably large. In the limit of an effective medium,
the response of this medium would be very isotropic. For this assumption of no ﬂux leakage to
hold, the stacking distance, l, of the rings have to be much smaller than the radius, r, of the ring
itself, which is not satisﬁed in the actual medium. Hence the actual resonant frequency when
the ﬂux leakage is accounted for (which is nontrivial) can be expected to differ somewhat
from this expression, but it is not otherwise expected to give rise to large differences in the
functioning.
Balmaz and Martin [63] have introduced a novel magnetic scatterer which consists of two
orthogonal intersecting SRRs as shown in ﬁgure 8(b). This has an isotropic response for any
wave incident on it normal to the plane of the rings. Balmaz and Martin discuss the various
orientations of the rings and conclude that the rings can only intersect along symmetric points
for an isotropic response. Three orthogonal nonintersecting SRRs is a possible candidate
as an isotropic scatterer, but unfortunately the SRRs then tend to have nondegenerate resonant
frequencies or different Q factors. An array of twodimensional isotropic scatterers with
randomorientations canbe expectedtobehave reasonablyas aneffective mediumwithisotropic
response.
2.2.4. Wave dispersion in a resonant magnetic medium. Given the generic form of the
dispersion in a SRR medium (or any of the resonant magnetic media to be discussed below),
one can obtain the dispersion relation between k, the wavevector, and ω, the frequency, for a
wave. We showschematically this generic relationship in ﬁgure 9. One can see that the medium
allows two degenerate waves (polarizations) with a gap for frequencies ω
0
< ω < ω
m
, where
ω
0
is the resonance frequency and ω
m
is the magnetic plasma frequency. This is the region
where the magnetic permeability is negative and all waves in the medium are evanescent. This
also gives rise to the idea that a bandgap (say in a photonic crystal) can be ascribed to negative
µ just as one can ascribe it to an imaginary refractive index or negative ε [69]. Of course, the
origin of the gap would depend on the symmetries of the ﬁelds in question.
One also notes that the effective medium can support a longitudinal bulk magnetic plasma
mode at ω
m
. The induced currents in the system give the appearance of magnetic monopoles
sloshing up and down the cylinders. Interestingly the system should also be able to sustain
the magnetic equivalent of surface plasmon modes at an interface with vacuum (say) when
Re(µ) = −1 just like for the usual surface plasmons at an interface between positive and
negative dielectric media. These will couple to spolarized light with a nonzero component
of the magnetic ﬁeld normal to the surface.
Physics of negative refractive index materials 467
Figure 9. The dispersion for a wave in an SRRmedium. The frequency and wavevector are scaled
with respect to the resonance frequency, ω
0
, and k
0
= ω
0
/c. Propagation is not possible in the
frequency gap where µ < 0 as shown.
2.2.5. The Swiss roll structure at radiofrequencies. For lower radiofrequencies, the
capacitance of the system can be made large instead of increasing the size of the loop
(inductance). This is both convenient and increases the validity of our assumption of an
effective medium as well. This can be achieved by rolling up a metal sheet in the form of
a cylinder with each coil separated by an insulator of thickness d as shown in ﬁgure 6(c).
This structure has also been popularly called the Swiss roll structure. The current loop is now
completed through the differential capacitance across the space between the metal sheets as
shown. As before, the effective magnetic permeability for a system of such structures can be
calculated as
µ
eff
= 1 −
πr
2
/a
2
1 −(dc
2
/2π
2
r
3
(N −1)ω
2
) + i(2ρ/µ
0
ωr(N −1))
, (2.35)
where N is the number of coils in the structure and it is assumed that the total thickness of the
wound layers Nd _ r the cylinder radius. This system also has the same generic resonance
formof the permeability with frequency as the SRRstructures, but nowthe resonance frequency
occurs at a much smaller frequency owing to the larger capacitance of the structure. This has
seen application in magnetic resonance imaging (MRI) at radiofrequencies as magnetic ﬂux
tubes in the region where the effective magnetic permeability assumes very large values on the
lower frequency side of the resonance [23, 24].
2.2.6. Scaling to high frequencies. Maxwell’s equations appear to suggest that one can
scale the phenomena to higher frequencies by simply scaling down the corresponding length
scales. Terahartz operation of SRRs has been demonstrated experimentally [27]. However,
the main problem in scaling to higher infrared and optical frequencies is that metals no longer
behave as perfect conductors and electromagnetic ﬁelds penetrate considerably into the metal.
This means that the dispersive nature of the metals must be taken into account for scaling to
higher frequencies. Another cause of concern is the technological ability to make the resonant
structures necessary at the small lengthscales.
Keeping this in mind, a single ring with two symmetric splits providing a capacitive gap
and suitable for operation in the terahertz region was proposed [28] and is shown in ﬁgure 10(a).
468 S A Ramakrishna
0 30 60 90 120 150
Frequency (THz)
1
0.5
0
0.5
1
N
o
r
m
a
l
i
z
e
d
w
a
v
e
v
e
c
t
o
r
light line
Silver rods
SRR columns
50 60 70 80 90 100
Frequency (THz)
2
0
2
4
6
(
)
Re()
Im()
50 60 70 80 90 100
Frequency (THz)
2
1
0
1
2
3
(
)
Re()
Im()
(b)
(c)
(d)
(a)
c
d
c
L
c
a
b
Figure 10. (a) The single ring with a double split with a = 600 nm, b = 312 nm, τ
c
= 24 nm,
d
c
= 24 nm and L
c
= 144 nm used for the calculations. (b) The photonic band structure of this
systemof columns for the spolarization (with Halong the cylindrical axes). The light line (· · · · · ·)
and the bands for silver columns with the dame ﬁlling fraction are shown for comparison. The
bandgap at about 75 THz arises due to a negative µ. (c) The real and imaginary parts of µ and
(d) the real and imaginary part of ε. All the data are taken from [28].
The band structure and the transmission for inﬁnitely long split cylinders of silver obtained
using a photonic bandstructure calculation using the transfer matrix method [70] is shown
in ﬁgure 10(b). The magnetic ﬁeld is assumed to be along the axis of the split cylinders.
A plasma form for the dielectric function of silver, ε(ω) = ε
∞
− ω
2
p
/[ω(ω + iγ )], was used
with the empirical values [71] of ¯ hω
p
= 9.013 eV, γ = 0.018 eV and ε
∞
= 5.7. Note the
presence of a frequency gap that arises due to the negative effective µ of the structure at about
75–80 THz, while a simple circuit theory predicts a resonance frequency of 104 THz for the
L–C resonance. Clearly the assumptions of a perfect conductor cannot be made at these high
frequencies. Using the method of Smith et al [72], the effective ε and µ were recovered from
the emergent quantities, viz, the complex transmission and reﬂection coefﬁcients, and are
shown in ﬁgure 10. Clearly there is a frequency band with Re(µ) < 0, although surprisingly
Im(ε) < 0 near the resonance and Re(ε) disperse the other way around. This behaviour,
analogous to the case of the cutwire system discussed before, was also found by Koschny
et al [52], who call it antiresonant behaviour, and seems to be generic to metamaterials with
one negative material parameter.
Let us analyse this system to gain an insight into the high frequency scaling properties.
Assuming a strong skin effect, i.e. that the thickness, τ
c
, of the metal shells is smaller than
the skin depth (∼20 nm at 100 THz), we can write the displacement current per unit length
j
φ
= −iωε
0
ε
m
E
φ
. Thus the potential drop across each half of the ring is
V
r
=
_
π
0
E
φ
r dφ =
πrj
φ
−iωε
0
ε
m
(2.36)
Physics of negative refractive index materials 469
and the potential drop across each capacitive gap (of arm length L
c
) is
V
c
=
1
C
_
I (t ) dt =
j
φ
τ
c
d
c
−iωε
0
εL
c
. (2.37)
Using the fact that the total emf induced around the loop is
_
E
φ
dl = iωµ
0
H
int
ds, we can
equate the potentials as
2V
r
+ 2V
c
= iωµ
0
πr
2
H
int
. (2.38)
Using Ampere’s law, the magnetic ﬁelds inside and outside the split cylinders can be related as
H
int
−H
ext
= j
φ
τ
c
, (2.39)
yielding the relation that
H
ext
H
int
= 1 −
µ
0
ε
0
ω
2
πr
2
τ
c
[2πr/ε
m
+ 2τ
c
d
c
/(εL
c
)]
. (2.40)
The averaged magnetic induction is B
eff
= (1 − f )µ
0
H
ext
+ f µ
0
H
int
, where f = πr
2
/a
2
is
the ﬁlling factor. Averaging over a line lying entirely outside the cylinders for the magnetic
ﬁeld H
eff
= H
ext
, we obtain the effective magnetic permeability as
µ
eff
=
B
eff
µ
0
H
eff
= 1 +
f ε
0
µ
0
ω
2
πr
2
[(2πr/(ε
m
τ
c
) + 2d
c
/(εL
c
)] −ε
0
µ
0
ω
2
πr
2
τ
c
. (2.41)
Assuming that ε
m
. −ω
2
p
/[ω(ω + iγ )] for frequencies in the infrared and optical regions, we
obtain the same generic form for the effective permeability,
µ
eff
(ω) = 1 +
f
/
ω
2
ω
2
0
−ω
2
−iω
, (2.42)
with the resonant frequency, effective damping and the effective ﬁlling fraction
ω
2
0
=
1
(L
i
+ L
g
)C
, =
L
i
L
g
+ L
i
γ, f
/
=
L
g
L
g
+ L
i
f, (2.43)
where C = ε
0
εL
c
/2d
c
is the effective capacitance of the structure, L
g
= µ
0
πr
2
is the
geometrical inductance and L
i
= 2πr/(ε
0
τ
c
ω
2
p
) is an additional inductance that shows up.
Noting that the plasma frequency is ω
2
p
= ne
2
/(ε
0
m), we see that the additional inductance is
entirely due to the electronic mass and L
i
can hence be termed the inertial inductance.
The presence of this additional inductance can be explained by noting that at high
frequencies the currents are hardly diffusive and almost ballistic because the distance through
which the electron moves within a period of the wave becomes comparable with the mean free
path in the metal. This means that if the frequencies are too high, the electrons can hardly
be accelerated and the response falls. The mass of the electron contributes additionally to the
inductance
2
. The effective damping factor also becomes much larger as the size of the ring
is reduced. This is due to the fact that the proportion of energy in the ballistic motion of the
electrons increases as the size is reduced and the resistive losses are then very effective indeed.
Thus even if the size of the ring were negligible, the inertial inductance would still
be present, preventing scaling to higher frequencies. This effect has also been discussed
in connection with using superconducting SRRs in the microwave region [73]. The large
increase in the damping as the dimensions are scaled down broadens the resonance and the
permeability does not disperse violently now, and the region of negative permeability vanishes
altogether. The increase in damping is a matter of great concern for operation at optical
2
The current j = nev ∼ ne(−iωeE/m), where E is the applied ﬁeld. Then the potential drop is V ∼ m/
(ne
2
)∂j/∂t , implying an inductance that is proportional to the electronic mass.
470 S A Ramakrishna
−
−
−
−
−
−
+
+
+
+
−
−
−
−
−
−
+
+
+
+
−
+
+
+
+
+
+
+
+
+
+
+
+
+
+
−
−
−
−
−
_
_
−
−
−
−
−
−
+
+
+
+
+
−
−
−
−
−
−
−
− +
+
+
(a) (b) (c)
Figure 11. The SRRs develop an electric polarization too although driven by a magnetic ﬁeld. This
bianisotropic behaviour is maximal for (a) a single split ring which develops a large dipole moment.
(b) The original SRR with two splits is also bianisotropic as it develops an electric dipole. The
ﬁeld lines due to the charges are also shown by thin lines. (c) The single ring with two symmetric
splits only develops a quadrupole moment and is hardly bianisotropic.
frequencies [74]. It has been shown that the response of the SRR with two splits tends to
tail off in the infrared region (λ ∼ 5 µm) [28]. By adding more capacitive gaps to lower
the net capacitance and adjusting the dielectric constant of the embedding medium, it has
been numerically demonstrated that a medium of SRR with four splits can have a negative
µ at a telecommunications wavelength of about 1.5 µm [22]. Panoiu and Osgood [75] have
shown that the original SRR, suitably scaled down, can also operate at infrared frequencies.
Panina et al [76] have proposed the use of pairs of parallel metal sticks (with lengths ∼100 nm)
periodically embedded in a dielectric mediumas an effective mediumwith negative magnetism
at infrared frequencies. Hsu et al [64] have claimed resonance with the original SRR at a
wavelength of 10.6 µm. But the size of the SSR was 10.48 µm and the assumption of an
effective medium would hardly be valid.
2.2.7. Other approaches. One might ask then if the main ingredients of a resonant magnetic
medium are just a capacitance and an inductance, then why a single capacitive gap cannot
work as well. In fact, SRR media of rings with a single split [76–78], shaped particles [54],
a double ring with a single split in one ring [79] and deformed split rings [66] have been
proposed as candidates for a resonant magnetic medium. The point is that we would like to
have as low an electrical polarizability as possible. With a single split ring, a large electric
dipole moment would be generated across the capacitive gap (see ﬁgure 11(a)) and this could
well dominate over the weaker magnetic dipole moment generated in the ring. When there are
two splits present, the dipole moments across opposite ends cancel each other and one only
gets a weak electric quadrupole moment (see ﬁgure 11(c)) whose effects would be expected
to be much weaker than those of the magnetic dipole moment. Most of the magnetic media
that simultaneously generate an electric dipole moment are bianisotropic media [80], i.e. the
constitutive relations are
D = εE + αH, B = βE + µH, (2.44)
where α and β are the bianisotropy coefﬁcients. The –shaped particles, ﬁrst introduced as
the components of a bianaisotropic medium [81], when arranged in a periodic lattice also have
a resonant ε and µ and can behave as negative magnetic materials [54]. Even the original SRR
medium [3] is bianisotropic [82] as there is an electric dipole moment that develops across the
capacitive gaps (see ﬁgure 11(b)), and this can also be driven by an electric ﬁeld [62]. Thus the
SRR is oriented such that the magnetic ﬁeld is normal to the plane of the SRR and the electric
ﬁeld is along the SRR and can be driven by the electric ﬁeld and the magnetic and electric
resonances can overlap. This can be effectively resolved, of course, by rotating adjacent
Physics of negative refractive index materials 471
(b) (a)
Figure 12. The left panel shows a photograph of the composite of thin wire arrays and SRRs
that shows a negative refractive index made by the San Diego group (courtesy D R Smith). The
right panel shows the transmission across a sample of thin wire arrays (· · · · · ·), SRRs (   ) and
a composite of thin wire arrays and SRRs (——) as measured by the San Diego group (adapted
from the data of [4]). The enhanced transmission in the latter case within the negative µ band is
interpreted as evidence of a negative refractive index.
SRRs in the plane by 180˚ and the corresponding electric dipole moments would cancel. The
symmetry of the single ring with two symmetrically placed capacitive gaps renders this less
bianisotropic and electrically less active.
In fact, any resonance that can be driven by the magnetic ﬁeld of radiation can be used.
The use of the magnetic Mie scattering resonances of a dielectric cylinder for generating a
negative magnetic mediumhas also been proposed[83]. Onlythe ﬁrst monopolar magnetic Mie
resonance (with the applied magnetic ﬁeld along the cylinders) can contribute to the effective
magnetic polarizability upon homogenization, and the dielectric constant of the cylinders
needs to be large to reduce the resonant frequencies to within the ﬁrst Bragg band. The use of
ferroelectric cylinders with very large values of ε at gigahertz frequencies has been proposed.
The possibility of a medium with a very low frequency (∼50 Hz) negative magnetic meta
material has also been proposed [21]. This consists of a periodic lattice of current carrying
wires, with alternate wires carrying currents in opposite directions. Displacements of these
wires magnetize the medium. The tension in the wires and the mutual repulsion between them
give a restoring force, causing them to oscillate at low frequencies. In a process very similar to
that in an electrical resonant dipole medium, the dispersive µ can become negative in a band
of frequencies above the resonance frequency where the system mimics a plasma of magnetic
monopoles.
2.3. Negative refractive index materials
We will follow the rule that any medium with Re(ε) < 0 and Re(µ) < 0 simultaneously will
be considered to have a negative refractive index. A rigorous justiﬁcation of this is presented
in section 3.1. Smith et al [4] proposed that combining in a composite the thin wire medium
with ε < 0, and the SRR medium with µ < 0 would yield an effective medium with n < 0.
Admittedly it is not obvious that the media would have a negative n. But the calculations
and experimental measurements of Smith et al [4, 84] were very suggestive of it. Their
uniaxial composite shown in ﬁgure 12(a) consisted of wires of 0.8 mm thickness and SRRs
with ω
0
= 4.845 GHz. Their numerical calculations showed that if thin wires were introduced
into a SRRmedium, a passband occurs within the bandgap of negative µfor radiation with the
electric ﬁeld along the wires and the magnetic ﬁeld normal to the plane of the SRR. Combining
472 S A Ramakrishna
the dispersions for the plasmalike dielectric mediumand a resonant magnetic medium, we get
k
2
=
εµω
2
c
2
=
(ω
2
−ω
2
p
)(ω
2
−ω
2
m
)
c
2
(ω
2
−ω
2
0
)
, (2.45)
where ω
p
and ω
m
are the electric and magnetic plasma frequencies, respectively, and ω
0
is
the magnetic resonance frequency (ω
m
= ω
0
/
√
(1 −f )). If ω
p
is the largest of them, then
we obtain a passband in the region ω
0
< ω < ω
m
. The numerical calculations suggested
that the thin wires and the SRRs functioned independently and the composite had a passband
with a negative refractive index. The experimental results on transmission through waveguides
ﬁlled with a medium of thin wires, a medium of SRRs only and a composite medium with
both thin wires and SRRs is shown in ﬁgure 12(b). The transmission is in the stopband due to
the negative µ for SRRs and there is enhanced transmission within this bandwidth when thin
wires are introduced into the SRR medium. This was taken as proof that the medium was a
NRM. The San Diego group went on to show [5] that a prism made of such a composite NRM
refracted microwaves in the direction opposite to the normal direction compared with a prism
of a positive refractive material such as Teﬂon
3
.
The samples in the original San Diego experiments had a small transmission, and it was
claimed that the observed effects could be due to just absorption [9]. Numerical photonic
calculations [65, 85] showed that there was indeed a negative index band which overlapped
reasonably with the negative µ frequency band, but indicated that there was considerable
absorption in these media due to Ohmic losses and dissipation in the dielectric board. Other
calculations [26] showed that the losses were primarily due to Ohmic losses in the metal rings
and showed that the losses were somewhat smaller. The San Diego experiments also met
with criticism as they were conducted in a waveguide and there was speculation about leaky
waveguide modes. Now these experiments have been repeated with large samples in free
space [25, 26] where both the transmission and reﬂection were measured, and the absorption
was found to be sufﬁciently small to unambiguously demonstrate the negative refraction effect
and to be consistent with calculations. These experiments have unequivocally demonstrated
the existence of a negative refractive index.
Although there is both experimental and numerical evidence that the composite of
thin wires and SRRs has a negative refractive index band, it is not a straightforward
conclusion. That the two structures do not interact and interfere with each other’s functioning
is questionable. Pokrovsky and Efros [86] showed that embedding a thin wire array in
a homogeneous medium with µ < 0 does not produce an NRM and the system has no
propagating modes. First, in a metamaterial, one cannot assume a homogeneous ε or µ due
to the other parts of the composite structure. Second, as long as the thin wires are not placed in
regions where the highly inhomogeneous magnetic ﬁelds associated with the SRRs are present
(along the axis of the SRR) and the SRR planes are placed such that they are at the points of
symmetry between the wires (where the magnetic ﬁelds associated with the wires are minimal),
the interference can be reduced greatly [87]. Third, we note that the magnetic ﬁelds due to
the wires fall off rather rapidly with distance from the wires and should not affect the SRRs
badly. Then the quasistatic responses derived here for ε and µ would be valid in the negative
refractive index band, and the SRRs and the thin wires would function independently as if in
vacuum. Thus the relative placement of the components could be crucial and might account
for the differences reported in numerical and experimental data. Aclear demonstration that the
ﬁelds due to the two structures do not interfere with the functioning of each other is yet to be
made. It has been shown that although the quasistatic interaction between the SRRand the thin
3
This is due to the fact that radiation refracts to the other side of the normal at the interface between positive and
negative media (see section 3.2.1).
Physics of negative refractive index materials 473
wire array is small, there are residual interactions between the two subsystems which lead to
both a shift in the resonant frequency [88] and an increased dissipation in the structure [89].
Further there are problems associated with a simplistic approach to homogenization of the
composite at subresonance frequencies where µ is positive and large, and ε due to the thin
wires is negative. Then the wave is highly evanescent and homogenization may not be easily
possible [88].
Alternative proposals for NRMs using other systems have been put forward. The use of
alternatively oriented shaped particles, where the magnetic resonance frequency (for the
magnetic ﬁeld normal to the loop) and the electric dipole resonance (for the electric ﬁeld along
the legs) coincide, has been proposed [54]. Panina et al [76] have proposed the use of oriented
paired parallel metal sticks as the constituents of the NRM. Deformed SRRs have similarly been
used to simultaneously generate an electric and a magnetic resonance for orthogonal magnetic
and electric ﬁelds [66]. Marqu´ es et al [90] have demonstrated that a waveguide ﬁlled with a
negative µ medium behaves as an NRM for frequencies below the cutoff frequency. It is well
known that a waveguide below the cutoff behaves as a onedimensional plasma and the split
rings when placed within it make it behave as an NRM.
2.4. Homogenization and effective macroscopic material parameters
In our discussion, it was assumed that media with highly inhomogeneous inclusions could be
described by only two complex parameters, ε and µ insofaras the electromagnetic interactions
are concerned. Central to this assumption was the fact that the size or lengthscales of the
inclusions are very small compared with the free space wavelength of the radiation. Then the
radiation encounters a coarsegrained spatial average of the polarizabilities of the constituent
structures of the medium. In fact, there are two levels of averaging involved here. The
structural units of the metamaterial are assumed to be sufﬁciently large on a molecular scale,
i.e. microscopically large, to be described by the bulk dielectric permittivity and magnetic
permeability of the constituent materials. Thus we have the Maxwell’s equations
∇ · D = 0, ∇ E = −
∂B
∂t
, (2.46)
∇ · B = 0, ∇ H =
∂D
∂t
, (2.47)
where D = ε
0
ε(r)E and B = µ
0
µ(r)H, and ε and µ are the material parameters of the
constituent materials. This is the ﬁrst level of homogenization. However, the structures in
question are sufﬁciently small compared with the lengthscale over which the applied ﬁelds
vary for only the ﬁelds due to the ﬁrst few multipoles of the charge and current distributions
induced in the structures to contribute to the polarization. In other words, the ﬁne structure
of the charge and current distributions is not discernible, but only a few averages such as
the corresponding dipolar ﬁelds or (rarely) the quadrupolar ﬁelds can be resolved. Thus we
have the averaged ﬁelds,
¸D) = ε
0
˜ ε
eff
· ¸E), ¸B) = µ
0
˜ µ
eff
· ¸H), (2.48)
that determine the effective dielectric permittivity and the magnetic permeability of the bulk
metamaterial which are second ranked tensors in general.
The Maxwell–Garnett approximation, which is essentially the quasistatic generalization
of the Clausius Mossotti electrostatic approximation, has been used widely to calculate the
bulk electromagnetic properties of inhomogeneous materials [91]. The component with the
largest ﬁlling factor is considered to be a host in which other components are embedded,
and the ﬁeld induced in the uniform host by a single inclusion (assumed to be spherical or
474 S A Ramakrishna
a
X
Z
a
a
Figure 13. The procedure for averaging the internal microscopic ﬁelds consists of averaging the
E and H ﬁelds over the edges of the unit cell of the simple cubic lattice, while the D and B ﬁelds
are averaged over a face S
x
for the xcomponent and so on.
elliptical) is calculated and the distortion of this ﬁeld due to the other inclusions is calculated
approximately. The Maxwell–Garnett approach incorporates the distortions due to the dipole
ﬁeld on an average and has been very successful in describing the properties of dilute random
inhomogeneous materials. In our treatment we have in some sense used the Maxwell–Garnett
type approximation but, however, assumed uniform depolarizing ﬁelds that simpliﬁed the
calculations. There are other effective medium theories [92, 93] for high ﬁlling fractions
that include more residual interactions or higher multipole effects. We should also note that
there are more mathematically rigorous homogenization theories [94], which analyse the
ﬁelds in an asymptotic expansion in terms of the microscopic lengthscale associated with
the inhomogeneities, which is however beyond the scope of this review.
In the case of periodically structured materials, the periodicity, ﬁrst of all, implies that
the averaging need not be over orientational and density ﬂuctuations. We will now discuss
an averaging procedure of Pendry [3] that was ﬁrst used to discretize Maxwell’s equations
on a lattice [17]. Consider the unit cell of the lattice (assumed to be simple cubic) shown in
ﬁgure 13. The Maxwell equations can be recast in their integral form:
_
C
H· dl = −
_
S
D · dσ, (2.49)
_
C
E · dl = −
_
S
B · dσ, (2.50)
where the curve C encloses the surface S, which itself suggests an averaging procedure. The
averaged ﬁelds, E
eff
and H
eff
, are deﬁned by averaging Eand Halong the sides of the unit cell:
E
(x)
eff
=
1
a
_
(a,0,0)
(0,0,0)
E
x
dx, E
(y)
eff
=
1
a
_
(0,a,0)
(0,0,0)
E
y
dy, E
(z)
eff
=
1
a
_
(0,0,a)
(0,0,0)
E
z
dz, (2.51)
H
(x)
eff
=
1
a
_
(a,0,0)
(0,0,0)
H
x
dx, H
(y)
eff
=
1
a
_
(0,a,0)
(0,0,0)
H
y
dy, H
(z)
eff
=
1
a
_
(0,0,a)
(0,0,0)
H
z
dz. (2.52)
Similarly, D
eff
and B
eff
are deﬁned by averaging them over the faces of the unit cell:
D
(x)
eff
=
1
a
_
S
x
D
x
dσ
x
, D
(y)
eff
=
1
a
_
S
y
D
y
dσ
y
, D
(z)
eff
=
1
a
_
S
z
D
z
dσ
z
, (2.53)
B
(x)
eff
=
1
a
_
S
x
B
x
dσ
x
, B
(y)
eff
=
1
a
_
S
y
B
y
dσ
y
, B
(z)
eff
=
1
a
_
S
z
B
z
dσ
z
, (2.54)
where S
x
is the face in the Y–Z plane, S
y
is the face lying in the Z–X plane and S
z
is the face
along the X–Y plane. The averaged ﬁelds enable us to determine ε
eff
and µ
eff
componentwise.
Physics of negative refractive index materials 475
There is only one restriction in this procedure: that the unit cell should not intersect any of the
structures contained within, so that the continuity of the parallel components of E
eff
and H
eff
across the surfaces is maintained. This procedure was adopted in section 2.2 to determine the
effective magnetic permeability.
The effective medium parameters determined from any such averaging process should
be unique and independent of the manner of the averaging process. Likewise, it should give
unique mediumresponses to applied ﬁelds, and it should be possible to uniquely determine the
scattered ﬁelds for the metamaterial. In fact, experiments usually access only the emergent
quantities such as the reﬂection and transmission through (say) a slab of the effective medium
and not the internal microscopic ﬁelds. Smith et al [72] inverted the classical relations for the
reﬂection and transmission coefﬁcients from a slab of homogeneous material of thickness d,
T =
_
cos(nk
z
d) −
i
2
_
Z +
1
Z
_
sin(nk
z
d)
_
−1
e
−ik
z
d
, (2.55)
R =
−i
2
_
Z −
1
Z
_
sin(nk
z
d)T e
ik
z
d
, (2.56)
for the refractive index, n, and the impedance, Z, as
n = ±cos
−1
_
1 −r
2
−t
2
2t
_
, (2.57)
Z = ±
_
(1 + r)
2
−t
2
(1 −r)
2
−t
2
_
1/2
, (2.58)
where r = R and t = T e
ik
z
d
, from which one obtains the dielectric permittivity ε = n/Z and
the magnetic permeability µ = nZ. The multivalued nature of the trigonometric functions
gives rise to an ambiguity in n and Z which, however, can be resolved by determining them for
several thicknesses, d, of the slab and requiring that Re(Z) > 0 and Im(n) > 0 for causal and
absorptive media (see section 3.1). It is then a matter of great reassurance that the effective
medium parameters derived from the emergent quantities coincide with the effective medium
parameters obtained from a homogenization of the microscopic internal ﬁelds [28, 72]. In
these processes, there is always an inbuilt ambiguity in the phase of the wave corresponding to
the unit cell size which would always affect the effective medium parameters to a small extent.
That would be negligible in the limit of a very small unit cell size.
As has been pointed out, the metamaterial of cut wires with Re(ε) < 0 has a counter
intuitive Im(µ) < 0 [52]. Similarly, metamaterials with Re(µ) < 0 have Im(ε) < 0 for
both an SRR medium [28, 52] and a medium of dielectric cylinders [83]. Although surprising,
this does not violate any fundamental requirement as long as the passive metamaterial is
only dissipative, i.e. the righthand side of equation (1.3) is positive deﬁnite. These negative
imaginary parts of ε and µ have been explained in part as a result of the ﬁniteness of the
unit cell [52]. One should, however, note that many of the metamaterials are intrinsically
bianisotropic, at least weakly, and a simple interpretation of the reﬂection and transmission of
the metamaterial slab in terms of an isotropic or anisotropic material is not strictly correct.
The reasons for this effect are yet to be completely understood.
It should be noted here that we do not consider a photonic crystal or a bandgap material
to be a metamaterial at high frequencies when the lattice period is comparable with or smaller
than the wavelength in free space. In this limit, the radiation can discern the individual
structures and scattering dominates—there is a splitting of the wave with large phase shifts.
In the metamaterial limit of low frequencies, the scattered ﬁelds corresponding to the small
scatterers have very large wavevectors, implying that the scattered ﬁelds are evanescent and
476 S A Ramakrishna
do not propagate. This is not true in the case of photonic bandstructure materials, where
momentum can be transferred between the propagating modes of the radiation and the crystal.
2.5. Causality and energy density in negative refractive media
It must be noted that dispersionless material parameters ε < 0 or a µ < 0 cannot exist.
Negative ε or µ for static ﬁelds would, for example, imply that the energy density for static
ﬁelds,
E
static
=
ε
0
ε
2
E
2
s
+
µ
0
µ
2
H
2
s
< 0, (2.59)
is negative, which is clearly not possible. Negative ε and µ are resonant effects and they are
necessarily dispersive and dissipative. A nondispersive NRM would imply, for example, that
time runs backwards for light in the medium [110].
The material parameters of a causal medium should satisfy the Kramers–Kronig (KK)
relations [18, 95]. These are relations between the real and imaginary parts of the response
functions:
Re(ε(ω)) −1 =
1
π
P
_
∞
−∞
Im(ε(ω
/
)) dω
/
ω
/
−ω
, (2.60)
Im(ε(ω)) = −
1
π
P
_
∞
−∞
[Re(ε(ω
/
)) −1] dω
/
ω
/
−ω
, (2.61)
Re(µ(ω)) −1 = −
1
π
P
_
∞
−∞
Im(µ(ω
/
)) dω
/
ω
/
−ω
, (2.62)
Im(µ(ω)) =
1
π
P
_
∞
−∞
[Re(µ(ω
/
)) −1] dω
/
ω
/
−ω
, (2.63)
where P is the Cauchy principal value, i.e. the real and imaginary parts are Hilbert transforms of
each other. These relations hold as long as the response functions are analytic in the upper half
plane and poles of the functions are in the lower half plane. If the poles occur only in the upper
half frequency plane, for example an amplifying medium or for nonlinear fourwave mixing,
then too the Hilbert transforms hold, but the sign of the transform becomes negative [96].
The case of meromorphic response functions when poles occur in both the upper and lower
plane [97] has been discussed in [98], which is important for metamaterials with negative
material parameters.
The generic forms of the dispersion for a resonant dielectric and magnetic media satisfy
the KK relations, and they are causal insofaras the homogenization is valid. For the plasma
of equation (1.7) a term of 4πσ/ω should be added to the righthand side of equation (2.61)
due to the pole at zero frequency. However, note that the integrals in the KK relations involve
frequencies all the way up to inﬁnity and it is clear that the effective medium theories will
break down at high frequencies. This never really affects us in the case of common optical
media where the macroscopic material response functions hold almost down to the level of a
few atomic distances. Thus the very high frequency limit is never really probed. In the case of
metamaterials, usually the wavelength is larger than the periodicity by only one or two orders
of magnitude, and this high frequency cutoff, when the homogenization becomes invalid, is
easily accessed. Thus the KKrelations should be applied cautiously to metamaterials keeping
this in mind. Further, due to the ﬁniteness of the structures the metamaterials have highly non
local responses, implying that the dielectric permittivity and the magnetic permeability also
disperse spatially at high frequencies (before the breakdown of homogenization). Generalized
KK relations for the spatially dispersive materials are available [99] and may be made use of.
Physics of negative refractive index materials 477
It is nontrivial to identify a positive deﬁnite quantity as the electromagnetic energy density
in dispersive media [19,100,101]. The dissipation in the mediumdepends on the time histories
of the ﬁelds and the medium polarizations. Assuming narrow band radiation and making a
Taylor series expansion about the carrier frequency and retaining only the linear term, it is
possible to write down an expression for the energy density as [18, 19]
E =
1
2
_
∂(ωε)
∂ω
ε
0
E
2
+
∂(ωµ)
∂ω
µ
0
H
2
_
(2.64)
in the socalled transparency regions, for frequencies far enough away froma resonance where
absorption is small. Thus a sufﬁcient condition for a positive energy density under these
approximations is ∂(ωε)/∂ω > 0 and ∂(ωµ)/∂ω. This is satisﬁed in the regions of normal
dispersion far from the resonances. A general and exact expression for the energy density has
been derived in [101],
E(t ) =
ε
0
2[E(t )[
2
+
µ
0
2[H(t )[
2
+
_
+∞
−∞
dω
_
¸
¸
¸
¸
α
E
(ω)
_
t
−∞
dτ e
iωτ
E(τ) + α
H
(ω)
_
t
−∞
dτ e
iωτ
H(τ)
¸
¸
¸
¸
2
_
, (2.65)
where α
∗
E
(ω)α
E
(ω) = ωImε(ω) and α
∗
H
(ω)α
H
(ω) = ωImµ(ω). This is positive deﬁnite
and applicable to the case where ε and µ can become negative as well. Although important
from a fundamental point of view, this expression is rather difﬁcult to use due to integrals over
the time history of the ﬁelds. Again it is emphasized that the KK relations have been used to
obtain this energy density and it is valid only for the low frequency regime when the effective
medium parameters themselves are valid.
3. Propagation of radiation in negative refractive media
3.1. Choice of the wavevector
As discussed in section 1.2, the reversal of the signs of both ε and µ in Maxwell’s equations,
k E = ωµ
0
µH, (3.1)
k H = −ωε
0
εE, (3.2)
implies that the phase vector k has a reversed sign, i.e. it is opposite to the Poynting vector
S = E H. We can take this to imply that the refractive index n =
√
εµ is negative. Thus,
the phase accumulated in propagating a distance is φ = −nω/c.
Writing ε = ε
/
+ iε
//
, and µ = µ
/
+ iµ
//
, we note that the medium is absorbing if ε
//
> 0,
µ
//
> 0 and amplifying if ε
//
< 0, µ
//
< 0. Let us consider a plane wave exp(ink
0
x) in an
absorbing NRM where the refractive index n = ±[(ε
/
µ
/
− ε
//
µ
//
) + i(ε
/
µ
//
+ µ
/
ε
//
)]
(1/2)
.
±[ε
/
µ
/
+ i/2(ε
/
µ
//
+ µ
/
ε
//
)/(ε
/
µ
/
)] for small absorption. Now the wave should decay in
amplitude as it propagates in the dissipative medium and that is governed by the sign of
Im(k
z
). For the case ε
/
< 0 and µ
/
< 0, this condition demands that the negative sign of the
square root be chosen. Smith and Kroll [102] considered the problem of a radiating current
sheet in a NRM and found that power is radiated away from the current sheet if the impedance
Z =
_
µ(ω)
ε(ω)
=
µ
n
(3.3)
was positive. For positive media, µ > 0 and n > 0 and the power is radiated away. But if
µ < 0 in an NRM, then n < 0. This has been contested by some authors [7, 103]. Walser
478 S A Ramakrishna
and coworkers [7] claimed that the wave propagation in NRMs was very inhomogeneous and
a negative refractive index could not be claimed. Pokrovsky and Efros [103] argued that the
refractive index is ambiguous as it does not enter the Maxwell equations and that only the
group velocity is negative in NRMs. For complex ε, µ and
¯
k, the wave of course becomes
inhomogeneous. However, there does exist a very general problem of choosing the sign of the
square root for determining the wavevector,
[k[
2
= k
2
x
+ k
2
y
+ k
2
z
= εµ
ω
2
c
2
, (3.4)
in such homogeneous media.
This problem can be formally treated [104] by considering the most general properties
of the wavevector in the complex plane. Without loss of generality, we consider an
electromagnetic wave with a wavevector [k
x
, 0, k
z
] to be incident from vacuum on the
left (−∞ < z < 0) on a semiinﬁnite medium (∞ > z > 0) with an arbitrary value of ε
and µ. Due to xinvariance, k
x
is preserved across the interface. The zcomponent of the
wavevector, k
z
, however, has to be obtained from the dispersion relation,
k
z
= ±
_
εµ
ω
2
c
2
−k
2
x
, (3.5)
where a physical choice has to be made for the sign of the square root. Now the waves in
medium 2 could be propagating [k
2
x
< Re(εµω
2
/c
2
)] or evanescent [k
2
x
> Re(εµω
2
/c
2
)].
Further the media could be absorbing or amplifying, depending on the sign of Im(εµ) in
equation (3.5). This enables us to divide the complex plane for Z = k
2
z
into the four quadrants
shown in ﬁgure 14. The waves corresponding to quadrants 1 and 4 have a propagating nature,
and the waves corresponding to quadrants 2 and 3 are evanescent. Crucially, we note that
there is a branch cut in the complex plane for
√
Z = k
z
and one cannot analytically continue
the behaviour of the waves across this branch cut. This branch cut divides the Riemann surface
into two sheets in which the two different signs for the square root will have to be taken. The
different regions in the Z = k
2
z
plane are mapped into the different physical regions of the
√
Z
plane, depending on k
x
, ε and µ. For absorbing media, the wave amplitude at the inﬁnities has
to obviously disappear. For amplifying media, one has to be more careful. The only conditions
are that evanescently decaying waves remain decaying, propagating ones remain propagating
and no information can ﬂow in from the inﬁnities. This ensures that the nearﬁeld features of
a source cannot be probed at large distances merely by embedding the source in an amplifying
medium. Due to the above reasons we will inconveniently choose the branch cut along the
negative imaginary axis as shown in ﬁgure 14. Hence our range for the argument θ of k
2
z
becomes −π/2 < θ < 3π/2 for the ﬁrst Riemann sheet and 3π/2 < θ < 7π/2 for the second
Riemann sheet, corresponding to the two signs of the square root, k
z
= ±
√
Z = [Z[
1/2
e
iθ/2
and [Z[
1/2
e
iπ+iθ/2
, respectively. The complex plane for k
z
=
√
Z with the corresponding eight
regions in the two Riemann sheets is shown in ﬁgure 14. Note that although the position of
the branch cut in the k
2
z
plane is only a convenient choice, the ﬁnal regions in k
z
for various
media is ﬁxed by causality.
Regions 1 and 8 correspond to propagating waves in conventional positive media that are
absorptive or amplifying, respectively. Regions 6 and 7 correspond to growing evanescent
waves that blowup at the inﬁnities, which are unphysical in a semiinﬁnite medium. Decaying
evanescent waves fall within region 2 if ε
/
µ
//
+ ε
//
µ
/
> 0 and in region 3 if ε
/
µ
//
+ ε
//
µ
/
< 0,
regardless of ε
/
and µ
/
. Note that the Poynting vector points away from the source (interface)
if the medium is absorptive overall, and actually towards the source if the medium is overall
amplifying, ε
/
< 0 and µ
/
< 0. For the case of evanescent waves in amplifying media, our
choice results in a Poynting vector that points towards the source (interface in this case). This,
Physics of negative refractive index materials 479
Re (k )
z
2
Im(k )
z
2
+ > 0
, , ,, , ,,
+ < 0
, , , , , ,,
+ +
+
+
+ > 0
, , ,, , ,,
+ > 0
, , ,, , ,,
+ < 0
, , , , , ,,
+ < 0
, , , , , ,,
+ < 0
, , , , , ,,
Re (k )
z
+ < 0
, , , , , ,,
Im(k )
z
+ > 0
, , ,, , ,,
+ < 0
, , , , , ,,
(a)
(b)
Evanescent waves
medium
Propagating waves
medium
Propagating waves
Growing
Evanescent waves
Decaying
Physically inaccessible
in semi infinite media
Absorbing
medium
Propagating waves
Positive refractive
medium
Amplifying
Propagating waves
Propagating waves Evanescent waves
Amplifying
Positive refractive
Absorbing
Propagating waves
Negative refractive
Evanescent waves
Negative refractive
Figure 14. (a) The complex plane for k
2
z
showing the different regions for the propagating and
evanescent waves. The branch cut along the negative imaginary axis for the square root is shown.
(b) The complex plane for k
z
=
√
Z with the corresponding eight regions in the two Riemann
sheets for k
2
z
. The evanescent regions could be absorbing or amplifying as explained in the text,
depending on the sign of ε
and µ
. The sign of the square root for the region is shown enclosed in
a small circle.
however, does not violate causality as the Poynting vector/energy ﬂow decays exponentially to
zero at inﬁnity and no information ﬂows in fromthe inﬁnities. This counterintuitive behaviour
does not imply that source has turned into a sink—rather it indicates that there would be a
large (inﬁnitely large for unsaturated linear gain) accumulation of energy density (intense local
ﬁeld enhancements) near a source. Now propagating waves in media with ε
< 0 and µ
< 0
simultaneously are included in regions 4 and 5, depending on whether ε
µ
+ ε
µ
< 0 or
ε
µ
+ε
µ
> 0, respectively, corresponding to absorbing or amplifying media. In both cases,
the negative sign of the square root needs to be chosen, justifying calling these media with
both ε < 0 and µ < 0 negative refractive index media. In the case of normal incidence,
the sign of the wavevector and the sign of the refractive index are the same. The quantity
ε
µ
+ ε
µ
determines the energy ﬂow which has also been recently noted by Depine and
Lakhtakia [105]. In dissipative media Im(k
z
) < 0 for propagating waves, which reduces to
Im(n) > 0 for normal incidence as noted in [106]. Thus, one can more reasonably talk of
NRMs as media in which radiation has a negative phase velocity rather than a negative group
velocity.
3.2. Some effects of a reversed wavevector
Now let us look at some of the effects of a reversed wavevector in some detail. Veselago [1]
has discussed some of these effects in his original paper
4
.
3.2.1. The modiﬁed Snell’s law of refraction. The refraction of radiation at the interface of a
positive medium and a negative medium has been one of the most discussed issues. Consider
the electromagnetic plane wave incident on an NRM from a positive medium with a wave
vector (k
x
, 0, k
z+
). Fromthe discussion in section 3.1, it is clear that the continuity of the ﬁelds
and Maxwell’s equations require the transmitted wavevector to be (k
x
, 0, k
z−
), where k
z−
is
the negative square root in equation (3.5). Noting that Poynting vector is oriented exactly
opposite to the phase vector, k, in the NRM, we realize that the ray representing the energy
4
In addition to the effect discussed here, Veselago has also suggested a reversed radiation pressure in NRMs. The
problem of radiation pressure even in ordinary dispersive media is controversial [107] and we will not discuss this
problem in structured NRMs and metamaterials here.
480 S A Ramakrishna
vectors
NRM
Wave
Positive media
Energy flow
Rays
Negative
refraction
Positive
refraction
Image Object
plane plane
n = –1
d/2 d/2 d
(a) (b) (c)
Figure 15. (a) Schematic depiction of the negative refraction at an interface between a positive
mediumand an NRM. The energy ﬂowand the phase vectors are in opposite directions in the NRM.
(b) The negative refraction effect through a prism. (c) The Veselago lens consisting of a slab of
NRM that can focus a point source from one side of the slab to the other side.
ﬂow corresponding to the refracted wave should now lie on the other side of the normal as
shown in ﬁgure 15(a). This negative angle can also be intuitively inferred from Snell’s law,
n
+
sin θ
+
= n
−
sin θ
−
, (3.6)
where n
−
< 0 ⇒ θ
−
< 0. The modiﬁcation of the Snell’s lawis counter to a very fundamental
concept in optics of isotropic media that, depending on how dense the second medium is, the
refracted wave can bend as close to the normal as possible, but never cross it. In this case,
the refracted beam actually emerges on the other side of the normal
5
. Note, however, that the
angle of the reﬂected beam remains unaffected. Figure 15(b) shows the negative refraction
through a prism which was used in the experiments of [5, 25].
As a direct consequence of this we have the result that a ﬂat slab of an NRM
6
with n = −1
can focus a source as shown in ﬁgure 15(c). This lens is different from the usual optical lens
with curved surfaces in that it does not focus rays from inﬁnity to a point. These rays would
simply pass through unaffected. On the other hand, rays emanating froma point source located
at a distance d
1
on one side of the slab would be refocused to a point at a distance d
2
on the
other side, provided that d
1
+ d
2
= d, where d is the thickness of the slab. In fact, the total
phase shift accumulated by any wave in going from the source point to the image point is zero
as a consequence of the negative phase vector inside the NRM
7
. This is in contrast to the phase
corrections that are made to different rays in a conventional lens. Consider now the behaviour
of traditional curved lenses made of NRMs. Note that the phase accumulated due to traversal
through the NRM is negative. Hence a convex lens causes a plane wave to diverge while a
concave lens causes the plane wave to become convergent.
3.2.2. The reversed Doppler shift. Assume a moving source emitting radiation at a frequency
ω in an NRM with a velocity v with respect to the medium. The frequency measured by a
detector in the frame of the NRM is
ω
/
= γ (ω + k · v), (3.7)
where γ = (1 − v
2
/c
2
)
−1/2
is the relativistic factor. Noting that [k[ = nω/c, for emission
along the direction of the motion of the source in the NRM (say, with n = −1)
ω =
_
c −v
c + v
, (3.8)
5
Note that negative refraction is possible in anisotropic media as well [108, 109].
6
This is also called a Veselago lens today.
7
In a sense, the phase unfolds inside the NRM and it is almost as if time runs backwards [110].
Physics of negative refractive index materials 481
A B C
A B C
S
S
k
k
v v
Figure 16. The Cerenkov radiation cone angle is modiﬁed for a reversed phase vector. There is
constructive interference for emission only in one direction. The phases of the wavefront emitted
by the particle at A, Band Chave to be identical. The case on the left represents Cerenkov radiation
in a positive medium with positive k while the ﬁgure on the right shows the emission in a medium
with negative k.
1
n
2
<n n 
2
< n
1
n
1 n
1
2
< 0
< 0
2
> 0
> 0
2
2
−∆
∆
(b)
(a)
Figure 17. Schematic showing the transverse displacement of a beamof radiation undergoing total
internal reﬂection. The GH shift is negative for reﬂection from NRMs.
i.e. the frequency measured by a detector would be smaller when the source is moving towards
it! This is counter to the frequency increase that we would get in a normal medium. Again it
is the reversed phase vector in NRMs that is responsible for this reversed Doppler shift.
3.2.3. An obtuse angle cone for Cerenkov radiation. Acharged particle moving at relativistic
speeds through a medium emits Cerenkov radiation in a cone of angles if its speed exceeds
that of light in the medium. The angle of this cone is deﬁned by the condition that the
emitted radiation interferes constructively only in that direction while destructively interfering
in all others. This condition is satisﬁed (see ﬁgure 16) along an acute angle deﬁned by
cos θ = c/(nv). This angle, however, becomes obtuse for constructive interference in NRMs
due to the reversed phase vector as is clear from ﬁgure 16, i.e. θ → π − θ. In reality, due
to the dispersion in the material parameters of the NRM, there will be Cerenkov radiation in
both the forward and backward directions for different frequencies [111]. This modiﬁed cone
for Cerenkov radiation has been numerically demonstrated for photonic crystals that exhibit
the negative refraction effect [112].
3.2.4. ReversedGoos–H¨ anchenshift. Whenlight is incident uponanopticallyrarer medium2
from an optically denser medium 1 (the refractive indices n
1
> n
2
), there is a critical angle of
incidence, θ
c
, given by
sin θ
c
=
_
ε
2
µ
2
ε
1
µ
1
, (3.9)
beyond which the wave in medium 2 is evanescent and the radiation is completely reﬂected
back into medium1. For a beamof light with a ﬁnite transverse extent, this is also accompanied
482 S A Ramakrishna
by a shift of the beamin the transverse direction as shown in ﬁgure 17. This shift occurs usually
along the parallel component of the wavevector and was experimentally ﬁrst demonstrated
by Goos and H¨ anchen [113]. It turns out that the Goos–H¨ anchen (GH) shift is also negative
for reﬂection from NRMs [114–117].
Artmann showed [118] that for angles of incidence sufﬁciently away from the critical
angle and the grazing angle, the beam is displaced by
δ
gh
= −
∂
r
∂k
x1
, (3.10)
where
r
is the phase shift for the reﬂected beam and k
x1
is the parallel component of
the wavevector in medium 1. Now consider the reﬂection coefﬁcient for ppolarized
radiation:
r =
k
z1
ε
1
−k
z2
ε
2
k
z1
/ε
1
+ k
z2
/ε
2
. (3.11)
We note that for total internal reﬂection, k
z2
→ iκ
z
is imaginary. Now if we simultaneously
make negative the signs of µ
2
and ε
2
, κ
z
remains unchanged. However the reﬂectivity from
the NRM medium r
−
= 1/r
+
= e
−i
r
, where r
+
= e
i
r
is the coefﬁcient of reﬂection from
a medium with positive ε and µ. Clearly the GH shift will be negative. A similar analysis
holds for the spolarized radiation with µ replacing the corresponding ε in our analysis. Note
that it is actually sufﬁcient to make only ε < 0 for the ppolarization and µ < 0 for the
spolarization to obtain a negative GH shift. It has been shown that the negative GH shift can
be very large due to surface plasmon states [119], and negative GHshifts have also been shown
for layered systems of positive and negative media [120]. The GH shift does not, however,
violate causality, and a pulse would take a ﬁnite amount of time to undergo the transverse shift.
This effect is primarily due to the evanescent ﬁelds in the second medium.
3.3. Phase velocity, group velocity and energy ﬂow
The wavevector k in an isotropic NRM points opposite to the direction of the Poynting vector
S = E H which gives rise to the negative refraction effect. Thus it is clear that the phase
velocity v
φ
= k/ωis negative. However, the energy ﬂowis along the Poynting vector and away
from a source. The group velocity, deﬁned as v
g
= ∇
k
ω(k), is usually along the direction of
the Poynting vector for reasonably narrowband pulses and thus should be oriented opposite to
the phase velocity. We do note that the very concept of a group velocity can fail for broadband
pulses which can be distorted easily due to dispersion [121]. We note that the phase velocity
and the Poynting vector do point in different directions in anisotropic media and they can also
point in opposite directions in anisotropic media [108]. Hence this point need not surprise us
greatly.
However, the negative refraction effect and the group velocity in an NRM have been the
subject of some interesting debate [7,122]. Valanju et al [7] analysed the interference fronts of
two waves with slightly differing frequencies and found that the interference fronts refracted
in the positive sense (see ﬁgure 18). Misinterpreting the normal of these interference fronts to
be along the group velocity, it was claimed that the refraction at an interface between a positive
medium and an NRM was positive. First of all, the group velocity in a homogeneous isotropic
medium can be expressed as [122]
v
g
= ∇
k
ω(k) = p
k
k
dω(k)
dk
, (3.12)
where p = ±1, depending on the choice of sign of the square root as the frequency does not
depend on the direction due to isotropy. This means that the group velocity can only be parallel
Physics of negative refractive index materials 483
(a) (b)
Figure 18. The interference patterns of two superimposed waves with slightly differing frequencies
refracting at an interface between negative and positive refractive media. (a) The case of inﬁnite
plane waves when the interference fronts appear to refract positively. (b) A schematic picture
of the refraction of beams with ﬁnite transverse size at slightly differing frequencies. Although
the interference patterns appear to refract positively, the beams (energy ﬂow) are seen to refract
negatively. The sidewise motion of the interference fronts is not apparent in the case of inﬁnitely
extended plane waves.
(in a positive medium) or antiparallel (in an NRM) to the phase velocity. The point is that the
velocity of the interference fronts would coincide with the group velocity only if the waves
have wavevectors in the same direction with slightly differing magnitudes (frequencies). Due
to dispersion in ε(ω) and µ(ω) in the NRM, the two waves with differing frequencies would
refract at slightly differing angles. Now the constitutive waves of the interference patterns in
the NRM no longer point in the same direction and the corresponding interference front no
longer propagates along the direction of the group velocity. In fact, a simulation using Gaussian
beams [123] with slightly differing frequencies clearly showed that the beam suffers negative
refraction while the normal of the interference fronts points in a completely different direction
that would correspond to a positive angle of refraction. In fact this is true of any wave packet
that is localized in the transverse direction [124,125] and this motion is not apparent in the case
of plane waves due to their inﬁnite transverse extent. The numerical calculations of Caloz et al
[126] showed that the refracted power in a waveguide bend ﬁlled with an NRM was along the
angle for negative refraction. Pacheco et al [127] analytically calculated the Poynting vector
for the refraction of a multifrequency plane wave and found that the power ﬂow was indeed
along the angle for negative refraction. Interestingly an energy velocity v
E
= S/u can be
deﬁned, where S is the Poynting vector and u is the average electromagnetic energy density.
As pointed out, it is difﬁcult to deﬁne u in a dispersive medium, and this quantity has not been
well investigated for an NRM yet.
Foteinopoulou et al [128] have shown through numerical ﬁnite difference time domain
(FDTD) calculations that the negative refraction of a beam experiences a long delay at the
interface but the beam refracts in the NRM along the negative refraction angle. The FDTD
calculations of Cummer [176] have also indicated that there is indeed a discontinuity in the
phase front of the wave at the interface, but the phase discontinuity was shown to be causal.
The group delay of a pulse propagating through a slab of NRM is also of great interest. For
narrowband pulses, the group delay time is τ
φ
= (∂φ)/(∂ω)[
ω
c
, where φ(ω) is the phase of
the transmission or reﬂection coefﬁcient and ω
c
is the carrier frequency of the pulse within the
group velocity description. Dutta Gupta et al [129] have shown that the phase delay time for
transmission across a slab of NRM is usually large (subluminal) and note interestingly that
superluminal delay times can be obtained in the frequency region of plasmalike behaviour,
ω
m
< ω < ω
p
.
484 S A Ramakrishna
X
Y
Z
i +di y y
i
y
i +di
x x ix
Z/2
Z/2
Z/2
Z/2
Y
L
L
L
C
C
C
(b)
(a)
Figure 19. A transmission line network with distributed impedance and admittance. Choice of (a)
for the inductances gives the usual transmission line with positive propagation constant, while (b)
yields a backward wave structure in which the phase and group velocities are opposite.
3.4. Backward wave structures and transmission lines
There is another completely independent approach based on the theory of transmission lines
to making NRMs [130–132]. This takes advantage of the fact that dielectric and magnetic
materials can be modelled using distributed L–C networks. At radiofrequencies, this can be
easily achieved by using lumped elements.
Consider the twodimensional distributed network shown in ﬁgure 19. The system has an
associated series impedance (Z) and shunt admittance (Y) per unit length. Concentrating on
one section in the medium, one can write the equations for the potentials across the section,
∂V
∂x
= ZI
x
,
∂V
∂y
= ZI
y
(3.13)
and using Kirchoff’s law, one can equate the currents in and out of the node,
∂I
x
∂x
+
∂I
y
∂y
= YV. (3.14)
Eliminating the currents, we obtain the Helmholtz equation:
∂
2
V
∂x
2
+
∂
2
V
∂y
2
+ β
2
V = 0, (3.15)
where β
2
= −ZY = −Z/Z
/
, Z
/
= Y. As usual we have the same problem of a physical
choice of the sign for the propagation constant, which will now depend on the impedances
Z and Z
/
. Let us note that for a TE ﬁeld (where the electric ﬁeld E
z
is perpendicular to the
wavevector) in a waveguide in the X–Y plane, we have the Maxwell’s equations
∂E
z
∂x
= −iωµH
y
,
∂E
z
∂y
= iωµH
x
, (3.16)
∂H
y
∂x
−
∂H
x
∂y
= −iωεE
z
. (3.17)
Mapping the voltage V → E
z
, the currents I
x
→ −H
y
and I
y
→ H
x
, we identify the
equivalent impedances of the material parameters
Z = −iωµ and Z
/
=
1
−iωε
. (3.18)
If we choose Z = −iωL and Z
/
= −1/iωC, where L and C are the inductance per unit
length and capacitance per unit length as in ﬁgure 19(a), we obtain the propagation constant
β = ω
√
LC = ω
√
εµ. The phase velocity for the wave in the distributed network is
Physics of negative refractive index materials 485
v
φ
= ω/β = 1/
√
LC = 1/
√
εµ and the group velocity v
g
= (∂ω)/(∂β) turns out to be
the same. This is the case of usual positive media where the phase velocity and the group
velocity are in the same direction. In fact L = µ
0
and C = ε
0
yield us the equivalent circuit
for vacuum with a characteristic impedance for the network of Z
0
=
√
µ
0
/ε
0
. 377 .
This implies that the conﬁguration in ﬁgure 19(b) has Z = 1/ −iωC and Z
/
= −iωL and
thereby the propagation constant is
β = −
1
ω
√
LC
, (3.19)
where the negative sign of the square root has been chosen. This is so that the group velocity
(which is along the Poynting vector),
v
g
= +ω
2
√
LC, (3.20)
is positive. Thus the phase velocity v
φ
= −ω
2
√
LC is opposite in this mediumand we do have
a backward wave in the distributed network. Thus the distributed LC network in the highpass
conﬁguration behaves as a waveguide ﬁlled with an NRM.
Now we note that any practical implementation of such a backward wave structure would
involve lumped L–C elements in a network, which would bring in its own associated length
scale, at which we periodically or randomly insert the lumped elements, and its corresponding
band structure. To justify our assumption of a homogeneous medium we would require that
the associated periodicity be much smaller than the wavelength of the radiation. The effective
dielectric permittivity and the magnetic permeability at low frequencies for the periodically
loaded medium can be approximated as [132]
ε
eff
= ε −
1
ω
2
Ld
, µ
eff
= µ −
1
ω
2
Cd
, (3.21)
where ε and µ are due to the distributed capacitances and inductances due to the short
interconnecting segments of length d between the lumped elements, acting as a host medium.
At high frequencies, the propagation delays dominate and the medium is a positive index
mediumwith n =
√
εµ. On the other hand, at lowfrequencies, the loading elements dominate
and we have an NRM with a negative phase velocity and a refractive index of 1/ωd
√
LC.
Such backward wave structures show all the hallmarks of an NRM such as a modiﬁed
Snell’s law at an interface between a forward wave and backward structures. These have a
great advantage of a large bandwidth.
3.5. Negative refraction effect in photonic crystals
Although photonic crystal and bandstructure materials strictly cannot be considered to
be homogeneous metamaterials, it has been pointed out that there is a very interesting
negative refraction effect that can take place in photonic bandgap (PBG) media [36, 133].
The constituents of such a PBG medium are materials with positive refractive index, but a
propagating ray incident on this mediumappears to undergo negative refraction at the interface.
This effect can usually be dismissed as a diffraction grating effect if it occurs only for certain
wavevectors. But there is a regime when this happens to all propagating waves within a certain
frequency band, referred to as all angle negative refraction [36].
The secret to the allangle negative refraction effect is that for strongly modulated PBG
media the equifrequency surfaces in the kspace can become completely smooth convex
surfaces around certain points of symmetry, for example the M point for a square lattice
(see ﬁgure 20) for a band of frequencies. Such situations typically occur near the band edges,
and the size of the convex equifrequency surface reduces with increasing frequency. This
486 S A Ramakrishna
Γ
M
X
Figure 20. Schematic of the equifrequency surface for allangle negative refraction for a square
lattice. The dotted line is normal to the interface and shows the conservation of the parallel wave
vector. The thin lines show the wavevector, while the thick lines which are the normals to the
equifrequency surface denote the group velocities.
means that the group velocity is along the inward normal to these surfaces. If this situation
results in the lowest photonic band, then the incoming wave couples to a single Bloch wave in
the photonic crystal. The resulting diffraction will appear to be in the negative refraction sense.
A schematic is shown in ﬁgure 20 for the refraction process in the case of a square lattice. The
equifrequency surface is a convex surface around the M point and the incoming wave couples
to a single Bloch wave in the negative refraction direction. There is a crucial condition that this
equifrequency surface should be convex everywhere and have a larger size than a free space
dispersion circle, so that every incoming wave couples to a Bloch wave. Note, however, that
the group velocity and the phase velocity in the PBG medium are never antiparallel as in the
case of an isotropic NRM and k · (∂ω)(∂k) 0 in the entire band. These frequency regions,
however, do have a negative effective mass. Using a PBG medium of airholes in a dielectric,
an allangle negative refraction and a ﬂat Veselago lens were demonstrated numerically by
Luo et al in two dimensions [36] and three dimensions [134].
Note, however, that the equifrequency surface of the PBG medium need not be very
spherical or circular. This leads to the fact that most of the radiation may couple mainly along
one direction in photonic crystal (along [110] in ﬁgure 20). Thus the radiation might appear
to stream along one direction [135, 136]. This would render the image that should be formed
inside the NRM in the Veselago lens unobservable. Thus the negative refraction effect in the
case of PBG materials can have an interpretation entirely in terms of bandstructure effects
without resorting to the idea of a negative refractive index. Further, although the concept of
an effective refractive index might be useful near the band edges particularly with the lowest
frequency bands, homogenization in the sense of metamaterials is difﬁcult to achieve here.
Although the free space wavevector may be a few times larger than the unit cell of these
PBG materials, the energy transport in these materials is still very sensitive to the periodicity
and structural arrangements. Therefore, the propagation can be mainly considered as due to
Bragg scattering, rather than due to a negative refractive index. But the PBG media which
have allangle negative refraction can be used to make unique photonic devices that can tailor
the propagation.
3.6. Nonlinear effects in negative refractive media
One of the main aspects of the metamaterials is that the resonant nature of the effects makes
the local microscopic ﬁelds in them highly inhomogeneous and fairly intense. There can
Physics of negative refractive index materials 487
be immense enhancement of the local ﬁelds in these systems and the potential for nonlinear
applications was pointed out in the very ﬁrst paper on the SRR structure [3]. In fact, it is the
intense magnetic ﬁelds in the vicinity of the wires in the case of the thin wire structures that
makes them work as a low frequency plasma. SRRs also similarly concentrate the electric
ﬁelds in the tiny capacitive gaps, and this leads to enormously enhanced local electric ﬁelds.
Consider the single ring SRR with two splits. The electric ﬁeld across the capacitive gaps
is E
c
= V
c
/d
c
, where V
c
is given by equation (2.37), yielding us
E
c
(ω
0
) =
µ
0
ω
2
p
rτ
c
2γ L
c
. (3.22)
For the incident radiation, the energy is equally distributed between the electric and magnetic
ﬁelds. Hence we obtain an enhancement factor of
Q =
1
2
1/2ε
0
[E
c
(ω
0
)[
2
1/2µ
0
[H
ext
(ω
0
)[
2
=
1
8c
2
_
rτ
c
ω
2
p
2γ L
c
_
2
(3.23)
for the energy stored in the capacitive gaps at resonance, where the factor of 2 accounts for the
fact that we have two gaps in our system. Typical numbers for this factor range from 10
4
to
10
6
[3, 28].
Regardless of the frequency of interest, the nonlinear Kerr effect in a dielectric is always
possible [137]. This is a typical nonlinearity where the refractive index of the dielectric depends
on the electromagnetic ﬁelds as
n =
√
ε = n
0
+ n
2
I, (3.24)
where I = 1/2(ε/µ)
2
ε
0
c[E[
2
is the intensity of light. The sign of n
2
can be positive or negative,
and it is called a focusing or defocusing nonlinearity, respectively. Suppose the capacitive gap
in the SRR is ﬁlled with a Kerr nonlinear dielectric; then large changes can be expected in
response to the applied electromagnetic ﬁelds by virtue of the enhanced electric ﬁelds in the
gaps. The capacitance of the SRR, and hence the resonant frequency, now becomes a function
of the incident ﬁeld strengths. Consider a metamaterial of SRR with a nonlinear dielectric in
the capacitive gaps. A relation between the incident ﬁeld strength and the nonlinear resonance
frequency can be obtained in a quasistatic calculation as [22, 138]
[H
ext
[
2
=
Z
d
n
2
d
2
c
4n
2
L
2
c
ω
2
0
(1 −x
2
)[(x
2
−
2
)
2
+
2
/ 2
]
2
x
6
, (3.25)
where = ω/ω
0
, x = ω
NL
/ω
0
,
/
= /ω
0
, ω
0
is the resonance frequency for the SRR
embedded in the linear material and ω
NL
is the resonance frequency for the SRR imbedded in
the nonlinear material. The plot of the nonlinear resonance frequency against the ﬁeld strength
is shown in ﬁgure 21 and the bistable behaviour is reasonably stable against dissipation. The
material now switches between a negative magnetic medium (µ < 0, ε > 0) with high
reﬂectivity and a positive medium (µ > 0, ε > 0) which can almost be transparent, depending
on the incident intensity. Use of this mechanism to switch the behaviour of the metamaterial
of thin wires and SRR composites from a negative refractive index to a positive index or
plasmalike behaviour has been suggested [138].
The intense ﬁeld enhancements in metamaterials make this a very fertile ground for
nonlinear optics. Phase matching conditions which are crucial in nonlinear optics can be met
in a different manner in NRMs, leading to qualitative changes in the nonlinear responses. This
has been pointed out in the context of second harmonic generation and stimulated Raman
scattering [139]. The intense local ﬁeld enhancements also make the SRR very sensitive to
any absorbing molecular species in the regions of large ﬁelds, indicating the potential for use
in molecular probes. Ultrashort pulse propagation in these media would be interesting in view
of the very large dispersion in these media.
488 S A Ramakrishna
0 10 20 30 40 50
field strength (kA/m)
1
1.05
1.1
0
N
L
/
0
180nm
24nm
24nm
300nm
Figure 21. The nonlinear resonant frequency versus the ﬁeld strength for two values of the
dissipation rate: ﬁlled circles are for γ of silver and the open circles for 3γ . The inset shows
the SRR structure with the relevant dimensions. The inner box shows the nonlinear medium just
enclosing the SRR structure.
3.7. Surface electromagnetic modes in NRMs
Any material with a negative material permittivity and permeability permits the existence of
surface modes on the interfaces with positive media. NRMs permit both p and spolarized
surface modes, and this topic has received some attention recently [140–147].
Let us ﬁrst examine the nature of the surface plasmons (for ppolarized incident light) at a
single ﬂat interface between semiinﬁnite positive and negative media. The condition for the
existence of a surface plasmon at the interface is [32]
k
z1
ε
+
+
k
z2
ε
−
= 0, (3.26)
which gives the dispersion relation
k
x
=
ω
c
_
ε
−
(ε
−
−µ
−
)
ε
2
−
−1
_
1/2
, (3.27)
assuming ε
+
= 1 and µ
+
= 1 for the positive medium. For the sake of simplicity we will
assume the causal plasmalike forms ε
2
= 1 − ω
2
p
/ω
2
and µ
2
. 1 − ω
2
mp
/ω
2
for the NRM.
We omit dissipation to obtain the dispersion. The corresponding dispersion for the plasmon
mode is plotted in ﬁgure 22. We can see that the plasmon dispersions take different forms
for ω
mp
> ω
p
(the upper curve) and ω
mp
< ω
p
(the lower curve). At large k
x
, the plasmon
frequency tends to the electrostatic limit of ω
p
/
√
2 from either above (ω
mp
> ω
p
) or (below
ω
mp
< ω
p
). At small k
x
, the surface plasmon ﬁrst appears at the lightline (ω = ck
x
). Note
that the situation for the spolarized mode is just the other way around: the lower curve would
describe the spolarized mode for ω
mp
< ω
p
and the upper curve for ω
mp
> ω
p
. In both cases,
the plasmon whose dispersion is given by the upper curve has a negative group velocity in
the transverse direction. Similar results using a resonant form for the magnetic permeability
have been obtained in [140, 143]. Then there are two branches to the ppolarized modes for
ω
mp
< ω
p
, one below the magnetic resonant frequency when µ > 0, ε < 0 and similarly one
more above the magnetic plasma frequency.
Next let us examine the nature of the surface modes in a thin slab of NRM. As is well
known, the two degenerate surface plasmons at the two interfaces get coupled for a thin slab
and gives rise to coupled slab modes: a symmetric and an antisymmetric mode, also called slab
Physics of negative refractive index materials 489
0.0 1.0 2.0 3.0 4.0
k
x
/k
p
0.0
0.2
0.4
0.6
0.8
1.0
mp
= 0.0
mp
= 0.2
p
mp
= 2
p
ll
(1)
p
/
Figure 22. The dispersion for the surface plasmon at an interface between a negative medium and
a positive medium when the effects of dispersion are included. The frequency axis is scaled with
respect to the bulk plasmon frequency (ω
p
) and the k
x
axis is scaled with respect to k
p
= w
p
/c.
The light line ω = ck
x
and the ω = ω
s
= ω
p
/
√
2 line are given by the dotted lines. The dispersion
inside the inﬁnite negative medium, ω = ck
x
/
√
ε
2
µ
2
with ω
mp
= 2ω
p
, is shown by the chain
curve (1).
plasmon polariton modes. The condition for these resonances (for ppolarized light) are [32]
tanh
_
k
(2)
z
d
2
_
= −
ε
2
k
(1)
z
k
(2)
z
, (3.28)
coth
_
k
(2)
z
d
2
_
= −
ε
2
k
(1)
z
k
(2)
z
, (3.29)
where k
(j)
z
=
_
k
2
x
−ε
j
µ
j
ω
2
/c
2
. Using the above conditions, the dispersion relation for
the coupled surface modes can be obtained and we show them in ﬁgure 23, where again we
have assumed the dispersive forms ε(ω) = 1 − ω
2
p
/ω
2
and µ
2
= 1 − ω
2
mp
/ω
2
. We ﬁnd
that the ﬁnite frequency changes the dispersion from the electrostatic limit (where it is given
by ω = ω
p
_
1 ±exp(−k
x
d)/
√
2). The dispersion relations are qualitatively different for
ω
mp
< ω
p
(or µ
2
(ω
s
) > −1) and ω
mp
> ω
p
(or µ
2
(ω
s
) < −1). Physically, the behaviour of
both the symmetric and the antisymmetric modes, at large k
x
, have to tend to the uncoupled
plasmon dispersion for a single surface. Ruppin [141] has calculated the dispersions of these
slab modes using a resonant form for µ and again obtained two branches. Interestingly we can
also have waveguide modes in the slab when ε < 0, µ < 0 and the refractive index of the slab
is higher than that of vacuum as shown in ﬁgure 23. These new waveguide modes are absent
in slabs with only one of ε or µ negative.
The surface plasmon resonances can have a great effect on the scattering properties of
materials and on the radiation in the vicinity of these surfaces. The dispersion of the surface
modes is a function of the geometry of the surface. Ruppin [144] has derived the Mie scattering
490 S A Ramakrishna
0.0 2.0 4.0 6.0 8.0 10.0
k
x
/k
p
0.0
0.2
0.4
0.6
0.8
1.0
p
k
p
d=0.3
light modes
0.0 2.0 4.0 6.0 8.0 10.0
k
x
/k
p
0.0
0.2
0.4
0.6
0.8
1.0
k
p
d=0.3
k
p
d=0.5
(a)
(b)
p x x p
/
p
/
Figure 23. The dispersion for the coupled surface plasmons when the effects of retardation are
included. The frequency axis is scaled with respect to the bulk plasmon frequency (ω
p
) and the k
x
axis is scaled with respect to k
p
= w
p
/c. The lightline ω = ck
x
and the ω = ω
p
/
√
2 line are
given by (· · · · · ·). (a) µ = 1 − ω
2
mp
/ω
2
with ω
mp
= 2ω
p
. The dispersion in the inﬁnite negative
medium, ω = ck
x
/
√
ε
2
µ
2
, is shown by (— · —). The part of the lower branch with the . symbol
shows the nonradiative light modes within the slab. (b) µ
2
= +1.0.
coefﬁcients for an NRM sphere and obtained new resonances corresponding to the negative
magnetic permeability. Similarly the Mie scattering properties of a cylinder [148] made of an
NRMhave revealed the presence of newresonances in the NRMfrequency band. Klimov [145]
has also discussed the new TM scattering resonances due to a negative µ as well as the
whispering gallery modes due to a negative refractive index in the scattering properties of an
NRM sphere. The substantial changes to the spontaneous decay rates in the vicinity of an
NRM has also been pointed out.
4. The perfect lens
One of the most debated consequences of negative refraction is the possibility of a perfect lens
whose image resolution is not subject to the limitations of the traditional diffraction limit of a
wavelength. Indeed the proposal for a perfect lens [6] has provoked heated debates and caught
the imagination of the scientiﬁc community at large. It turns out that Veselago’s ﬂat slab of
NRM with ε = −1 and µ = −1 not only focuses the propagating radiations as shown by a
ray picture, but also focuses the evanescent nonpropagating components of radiation that are
usually conﬁned to the immediate vicinity of a source. The latter components are associated
with the spatial features of the source at a subwavelength lengthscale and the focusing action
for the evanescent waves is accomplished through the surface states that reside on the NRM
surfaces. This result has now been generalized to several conﬁgurations of NRMs [149].
It turns out that the focusing of the evanescent components is quite sensitive to imperfections
in the NRM, and this does somewhat curtail the subwavelength focusing capabilities of the
lens. Subwavelength image resolution, to quite some extent, is possible even with these
imperfect lenses, which we will henceforth refer to as superlenses. In this section, we will
discuss this phenomenon of focusing the nearﬁeld components and will address several of
the muchdebated issues.
4.1. Nearﬁeld information and the diffraction limit
One principle that is taught in optics courses even at the freshman level is the diffraction limit,
that we can never resolve features in an image to better than the order of a wavelength [35].
First let us reexamine it here. Consider an extended source that emits or scatters radiation.
Physics of negative refractive index materials 491
y
z
x
1
2
∆y1
∆y2
(a) (b)
Figure 24. Radiation is scattered and emitted by an object. The angle at which light comes out
depends on the spatial features of the source. A periodic object with large spatial period scatters
the light through small angles (a), while the light comes out at large angles for a periodic object
with small spatial periods (b).
For the time being assume it to be a planar object on the z = 0 plane and the optical axis for
imaging along the z axis as shown in ﬁgure 24. Let E(x, y, 0) be the associated electric ﬁeld.
Now the electromagnetic ﬁelds over all space can be written down as
E(x, y, z; t ) =
_
1
2π
_
2
_
k
x
_
k
y
dk
x
dk
y
E(k
x
, k
y
) exp[i(k
x
x + k
y
y + k
z
z −ωt )], (4.1)
where ω is the frequency of the radiation and
E(k
x
, k
y
) =
_
x
_
y
dx dyE(x, y, 0) exp[−i(k
x
x + k
y
y)] (4.2)
is the Fourier transform of the spatial variation of the source. The Maxwell equations impose
the condition that
k
2
x
+ k
2
y
+ k
2
z
= ε
0
µ
0
ω
2
c
2
= k
2
0
(4.3)
in free space. Note that k
x
and k
y
represent the Fourier components of the spatial variation in
the source in the corresponding directions.
Now k
x
= 2π/
x
and k
y
= 2π/
y
, where
x
and
y
are some spatial periods of the
variation. The source can have intensity or ﬁeld variations over arbitrarily small distances.
There is no restriction on that. As an example, consider the case of isolated molecules on
a surface emitting radiation, in which case the
x,y
can literally be on atomic lengthscales.
The corresponding transverse wavevectors k
x
and k
y
will then be very large, in which case
the dispersion equation, equation (4.3), cannot be satisﬁed without making the associated k
z
imaginary. Thus the waves with large k
x
> k
0
and k
y
> k
0
are evanescent and decay in
amplitude exponentially away from the source z = 0 plane. These are the nearﬁeld modes of
the radiation of the source and have a decay length of 1/k
x,y
< λ/2π. In conventional imaging
systems, these modes are never detected as a consequence. Hence the corresponding source
information about the fast varying features of the source is missing. The largest propagating
wavevector that contributes to the image is k
0
(assuming unit numerical aperture), and we
obtain the corresponding spatial frequency of the wavelength, λ, to be the minimum limit on
the spatial resolution in the image.
However note that this limitation can be sidestepped if the nearﬁeld modes of the source
can be sampled. This is precisely what is done in nearﬁeld optical microscopy [150, 151],
where a ﬁne ﬁbre tip is used to couple the nearﬁeld modes into the propagating modes in the
ﬁbre and detected. This does not violate any fundamental principle (such as the Heisenberg
492 S A Ramakrishna
uncertainty principle [152]). Although the transverse image resolution might be very small,
the corresponding transverse component of the wavevector is arbitrarily large. The essential
higher dimensional (>1) nature of the problem makes this possible [153].
4.2. Pendry’s proposal
Consider Veselago’s lens consisting of a slab of NRMof thickness d with ε = −1 and µ = −1
and surrounded by vacuumas shown in ﬁgure 25. Consider the source to be at z = 0 (the object
plane): we wish to calculate the ﬁelds at z = 2d (the image plane). First let us calculate the
reﬂection and transmission coefﬁcients for a plane wave incident on the NRM slab
8
: writing
the solutions in terms of the partial reﬂection (r
jk
∀j, k = 1, 2, 3) and transmission coefﬁcients
(t
jk
∀ j, k = 1, 2, 3) at the interfaces (see ﬁgure 25), we ﬁnd
T =
t
21
t
32
e
ik
z2
d
1 −r
12
r
21
e
2ik
z2
d
, (4.4)
R =
r
21
+ r
32
e
2ik
z2
d
1 −r
12
r
21
e
2ik
z2
d
. (4.5)
In this geometry, we have independent solutions for the ppolarized light (E ﬁeld in the plane of
reﬂection) and the spolarized light (H ﬁeld in the plane of polarization). The partial reﬂection
and transmission Fresnel coefﬁcients for the spolarized light are
r
21
=
(k
z1
/µ
+
) −(k
z2
/µ
−
)
(k
z1
/µ
+
) + (k
z2
/µ
−
)
, r
12
= −r
21
, (4.6)
r
32
=
(k
z2
/µ
−
) −(k
z3
/µ
+
)
(k
z2
/µ
−
) + (k
z3
/µ
+
)
, (4.7)
t
21
=
2(k
z1
/µ
+
)
(k
z1
/µ
+
) + (k
z2
/µ
−
)
, t
32
=
2(k
z2
/µ
−
)
(k
z2
/µ
−
) + (k
z3
/µ
+
)
, (4.8)
where µ
+
and µ
−
are the permeability of free space and the NRM, respectively. Similarly, the
partial reﬂection and transmission Fresnel coefﬁcients for the ppolarized light are
r
21
=
(k
z1
/ε
+
) −(k
z2
/ε
−
)
(k
z1
/ε
+
) + (k
z2
/ε
−
)
, r
12
= −r
21
, (4.9)
r
32
=
(k
z2
/ε
−
) −(k
z3
/ε
+
)
(k
z2
/ε
−
) + (k
z3
/ε
+
)
, (4.10)
t
21
=
2(k
z1
/ε
+
)
(k
z1
/ε
+
) + (k
z2
/ε
−
)
, t
32
=
2(k
z2
/ε
−
)
(k
z2
/ε
−
) + (k
z3
/ε
+
)
, (4.11)
where ε
+
and ε
−
are the dielectric permittivity of free space and the NRM, respectively. The
zcomponents of the wavevector are
k
z1
=
_
ε
+
µ
+
ω
2
c
2
−k
2
x
−k
2
y
= k
z3
, (4.12)
k
z2
= ±
_
ε
−
µ
−
ω
2
c
2
−k
2
x
−k
2
y
, (4.13)
where we have to choose the component of the wavevector k
z2
in accordance with the
arguments of section 3.1. For propagating waves we will choose the negative sign, and for
evanescent waves we will choose the positive sign.
8
This can be done in several ways: one can use partial reﬂections from the interfaces and sum them up. Or one can
take the eigensolutions for the ﬁelds inside and outside the slab, and apply appropriate boundary conditions on the
ﬁelds at the interfaces. The same solution results, irrespective of the manner of calculation.
Physics of negative refractive index materials 493
t
21
t
12
r
1 2
t
32
r
2 3
z=0 z=d/2 z=3d/2 z=2d
=1
=1
=1
=1
=–1
=–1
object plane image plane
Figure 25. The perfect lens systemconsisting of a slab of NRM. The object, at a distance d/2 from
the surface of the NRM, is focused on the other side of the slab. The restoration of the amplitude of
an evanescent component is depicted schematically. The partial reﬂection/transmission coefﬁcients
across the boundaries are shown.
When the NRM has ε
−
= −ε
+
= −1 and µ
−
= µ
+
= −1, we obtain trivially for
propagating waves that k
z2
= −k
z1
, and t
jk
= 1 and r
jk
= 0 due to the matched impedance.
Hence for the slab
lim
ε
−
→−1
µ
−
→−1
T = e
−ik
z1
d
, lim
ε
−
→−1
µ
−
→−1
R = 0 (4.14)
and this clearly shows that the total phase change for propagation from the object plane
(at z = 0) to the image plane (at z = 2d) is zero. Thus we have the same results for the
propagating waves as those yielded by a ray analysis.
Next consider the evanescent waves with k
z1
= i
_
k
2
x
+ k
2
y
−ε
+
µ
+
ω
2
/c
2
= iκ
z
: then
k
z2
= k
z1
and the partial coefﬁcients t
jk
and r
jk
diverge. However, the transmission and
reﬂection coefﬁcients for the slab are still well deﬁned in this limit,
lim
ε
−
→−1
µ
−
→−1
T = e
+κ
z
d
, lim
ε
−
→−1
µ
−
→−1
R = 0, (4.15)
i.e. the slab actually increases exponentially the amplitude of the evanescent wave at the same
rate by which it decays in free space. Thus the net amplitude change at the image plane
(at z = 2d) from the object plane at z = 0 is zero (see ﬁgure 25. Thus we have that result
the not only does the Veselago lens cancel the phase accumulation for the propagating waves,
but it also restores the amplitudes of the evanescent components, bringing them both to a focus
at the image plane. Further, the reﬂection is zero due to the impedance matching. Thus we
have a lens that is unaffected by the diffraction limit as it includes all the components of the
nearﬁeld too and hence it is a perfect lens. This result was demonstrated by Pendry in his now
classic paper [6].
Note that the total transmission and reﬂection coefﬁcients for the slab are well deﬁned
in spite of being the sum of a geometric series (of the partial waves [6]) whose terms can
have magnitude greater than unity, and can even diverge in the case of evanescent waves. The
calculation has been questioned on these grounds [154], but can be justiﬁed on grounds of
analytic continuity: that the validity of the sum of an inﬁnite series transcends the divergence
of the series itself, provided that the sum was carried out to include all inﬁnite terms and there
494 S A Ramakrishna
are no approximations [155]. In fact, we could have obtained the solution for the evanescent
waves directly from that of the propagating waves by analytic continuation, k
z
→ iκ
z
. Further
note that the transfer matrix for the slab [35] is actually independent of the sign of the wave
vector. That means that the ﬁnal solution is also independent of the sign of the wavevector,
but we choose the sign of the square root in accordance with section 3.1 for consistency.
The process of restoring the amplitudes of the evanescent components, henceforth called
ampliﬁcation by us
9
does not really involve any energy transport, as the Poynting vector
associated with the evanescent waves in a lossless medium is strictly zero (see [156] for an
elementary review). These are the steady state solutions of Maxwell’s equations and give the
ﬁeld distributions essentially at inﬁnite time. The large associated energy density in the NRM
is obtained from the source itself and is built up over some time.
The perfect lens solutions are the exact conditions for the surface plasmons of both electric
andmagnetic nature ona surface of a semiinﬁnite NRM. The divergence of the partial reﬂection
and transmission coefﬁcients for the interfaces was just a manifestation of these resonances.
The denominator of these coefﬁcients going to zero is the condition for the homogeneous
equation to have a nontrivial solution. It is these resonances that are responsible for the
restoration or ampliﬁcation of the evanescent nearﬁelds of the source. This is a generic
effect of a localized resonance for the Helmholtz equation—the transmission and reﬂection
coefﬁcients diverge. In fact, this ampliﬁcation of the incident evanescent waves happens for
the Schr¨ odinger equation as well. Consider two identical separated localized potential wells
(say a δ or a square well) in one dimension with the incident evanescent wave having the energy
of one of the bound states of the wells. The evanescent wave ampliﬁes in the region between
the two potentials. It has been pointed out that the surface states on photonic bandgap media
in the ﬁrst Brillouin zone can also be used for the imaging of the nearﬁelds [157].
In the limit of large wavevectors (k
x
→ ∞) the ampliﬁcation of these evanescent waves
leads to a divergence in the electromagnetic energy [8, 158]. This a manifestation of the
pathological nature of the perfect lens and is because of the lack of a large momentum
cutoff [159]. In principle, however, there is always a cutoff with metamaterials that
corresponds to the inverse of the lengthscale of the structure of the metamaterial, k
c
∼ 1/a.
The ideal lens conditions, ε
−
= −1 and µ = −1, are very singular. All calculations should be
carried out by adding an inﬁnitesimal iδ to ω, and the limit δ → 0 should be taken to preserve
causality. Thus, the unphysical divergences go away when one has even an inﬁnitesimal level
of absorption in the NRM, i.e. ε
−
= −1 + iδ and µ
−
= −1 + iδ, which essentially sets its
own large momentum cutoff. This will be investigated in more detail in section 4.3. In fact,
the absorption in the material limits the ability to amplify the evanescent waves, particularly
those associated with large transverse wavevectors and hence the resolving capabilities of the
NRM lens [8, 160].
4.2.1. The quasistatic limit and the silver lens. We note that the perfect lens works when
the conditions on both the permittivity and the permeability, ε
−
= −ε
+
and µ
−
= −µ
+
, are
satisﬁed. From our discussion in section 2, it is clear that to generate media with both ε and
µ negative at the same frequency is quite difﬁcult. Particularly at optical frequencies, this
problem is more acute due to the lack of magnetic materials. It is in the optical and ultraviolet
frequencies that the prospect of imaging the nearﬁeld modes becomes most exciting. Thus
we would be severely handicapped by the lack of materials.
However, there is one great simpliﬁcationthat canbe afforded, as pointedout byPendry[6].
If we consider the case where all lengthscales in the problem are much smaller than the
9
Note that the solution ampliﬁes with distance and not with time. This ampliﬁcation is distinct from the more usual
ampliﬁcation related to laser gain or ampliﬁers in electronics, where there is an increase in energy.
Physics of negative refractive index materials 495
Figure 26. A thin ﬁlm of silver can act as a nearﬁeld lens. The conﬁguration used is shown on the
lefthand side. The two slits, placed 100 nm apart (subwavelength distance), are clearly resolved
by the lens while they cannot be resolved when at the same distance in vacuum. The calculations
have been performed in the electrostatic limit, and a value of ε
−
= −1 + i0.4 has been used for
silver.
wavelength, thenwe have k
x,y
¸ k
0
= ω/c andk
z
. k
x
inboththe positive andnegative media.
This is the extreme nearﬁeld limit or the quasistatic limit. In this limit, the Fresnel coefﬁcients
for the spolarization given by equations (4.6)–(4.8) become independent of the dielectric
constants and the Fresnel coefﬁcients for the ppolarization given by equations (4.9)–(4.11)
become independent of the magnetic permeabilities. Thus, a slab with negative magnetic
permeability (µ
−
= −µ+) would act as a nearﬁeld lens for the spolarized waves. Similarly
a slab with negative dielectric permittivity (ε
−
= −ε
+
) would act as a nearﬁeld lens for the
ppolarized waves. Thus we can use a slab of metal as a nearﬁeld lens for the ppolarized
radiation at a frequency when ε
−
= −1. Noting that dissipation affects the capability to
resolve large wavevectors, one chooses a highly conducting metal such as silver. The image
of two slits separated by a small distance imaged by a slab of silver is shown in ﬁgure 26 in
this limit. The subwavelength resolution capabilities are clearly visible. The resolution is,
however, limited by the large levels of dissipation in silver (at UV–visible frequencies, the
dielectric permittivity of silver is reasonably described by ε(ω) . 5.7 − 9
2
/ω
2
+ i0.4, ω in
electronvolts).
4.2.2. The role of surface plasmons. As has been pointed out, the role of the surface plasmons
is crucial to the action of the perfect lens [6, 142, 158, 159]. The perfect lens conditions
ε
−
= −ε
+
are exactlythe conditions for the surface plasmons ona metal surface and µ
−
= −µ
+
are similarly the conditions for the spolarized surface modes of a magnetic nature. In the
special case when both these conditions are satisﬁed at a single frequency ω, the conditions
for the surface modes on a semiinﬁnite NRM,
_
k
2
x
−ε
+
µ
+
ω
2
/c
2
ε
+
+
_
k
2
x
−ε
−
µ
−
ω
2
/c
2
ε
−
= 0, (4.16)
_
k
2
x
−ε
+
µ
+
ω
2
/c
2
µ
+
+
_
k
2
x
−ε
−
µ
−
ω
2
/c
2
µ
−
= 0, (4.17)
hold for all k
x
and all the surface plasmon modes (of both polarizations) become degenerate
at ω. These dispersionless modes have a zero group velocity along the surface. The total ﬁeld
can be written as a sum of the ﬁelds due to the source, and the ﬁelds of the surface mode
496 S A Ramakrishna
Figure 27. The asymmetric lens system consists of a slab of NRM with dielectric media with
different ε on either side. The ﬁgure shows the three cases of (top panel) ε
1
= ε
3
= −ε
2
—the
perfect lens, (middle panel) matching the perfect lens conditions on the farside, ε
3
= −ε
2
only,
and (bottom panel) matching the perfect lens conditions on the nearside, ε
1
= −ε
2
only.
excitations for any wave vector iκ
z
,
E(z) = Ae
−κ
z
[z[
+ B e
−κ
x
[z−d
1
[
+ C e
−κ
z
[z−d
2
[
, (4.18)
where A = 1 is the evanescent ﬁeld of the source at z = 0, B = −e
−κ
z
d
1
is the contribution
from the surface plasmon at the left surface z = d
1
and C = e
−κ
z
(d
2
−2d)
is the ﬁeld of the
surface plasmon at the right surface z = d
2
.
10
The presence of the other interface detunes the
surface plasmon resonance on each interface so that they are excited to the correct degree so
as to make the ﬁelds of the surface plasmon exactly cancel the incident ﬁelds for z > d
1
. On
the left side of the slab, the ﬁelds of the two surface plasmons again exactly cancel (for zero
reﬂectivity). It is this coherent action of the surface plasmons that is responsible for the perfect
lens action.
4.2.3. The asymmetric lens. It is actually not necessary that the perfect lens conditions of
ε
−
= −ε
+
, µ
−
= µ
+
(4.19)
be satisﬁed at both the interfaces to enable the ampliﬁcation of evanescent waves. Consider
the geometry shown in ﬁgure 27 with the negative medium between two positive media of
different permittivity and permeability. It is sufﬁcient if the plasmon conditions on both ε and
µ are satisﬁed at either of the interfaces. In either case we have the transmission coefﬁcient
(for the ppolarization) through the slab to be
T =
2(k
z1
/ε
1
)
(k
z3
/ε
3
) + (k
z1
/ε
1
)
exp(−ik
z2
d), (4.20)
clearly showing the phase reversal for the propagating wave and ampliﬁcation for the
evanescent modes. In this asymmetric case, however, the ﬁeld strength at the image plane
differs from the object plane by a constant factor, and thus the image intensity is changed.
Further, it is easily veriﬁed that wavevectors with different k
x
will refocus at slightly different
positions except in the quasistatic limit k
x
→ ∞. Thus there is no unique image plane and
the focus of this lens is aberrated. There is also an impedance mismatch, and the reﬂection
coefﬁcient is clearly nonzero. Hence, we term the asymmetric slab a nearperfect lens [142].
10
d
1
= d/2 and d
2
= 3d/2 in our case. But it is true as long as d
2
+ d
1
= 2d.
Physics of negative refractive index materials 497
A similar result holds for the spolarized wave incident on the slab. We note that while the
transmission is the same in both the cases, the spatial variation of the ﬁeld is completely
different as shown in ﬁgure 27. The reﬂection in the case of ε
2
= −ε
1
is also ampliﬁed.
The presence of a large reﬂectivity has serious consequences for the use of a lens for
nearﬁeld imaging applications as it would disturb the object ﬁeld. However, this idea does
render the technology of making these lenses easier. Particularly in the case of the silver lens, it
makes possible the use of a deposited silver thin ﬁlmon a solid dielectric substrate, which is far
preferable to a freestanding slab of silver with a subwavelength thickness. The technology
for the deposition of these ﬁlms on substrates is well established and the deposited ﬁlm will
certainly be more robust.
4.3. Limitations in real materials and imperfect NRMs
It should be emphasized that there is an inherent large momentum cutoff in all real materials,
and an inﬁnite resolution as our discussion above indicates is actually not possible. In meta
materials, it is clear that once the radiation can probe the structural details, effective medium
parameters make no sense. Thus there is a natural cutoff on the transverse wavevector of
k
x
< k
c
= 2π/a, where a is the lengthscale of periodicity of the metamaterial structures. In
the case of metamaterials, the boundaries of the slab are also indeterminate on the lengthscale
of a unit cell, whichcanalsoleadtosome extra dispersiononthe surface modes, therebycausing
some degradation of the image [161]. Even with metals, the dielectric constant becomes
dependent on the wavevector at large wavevectors and becomes spatially dispersive. For
example, the Lindhard form [162] for the dielectric permittivity of a metal at small wave
vectors k _ k
F
where k
F
is the wavevector at the Fermi surface, yields, within the random
phase approximation (RPA),
ε(ω, k) = 1 −
ω
2
p
ω
2
1
1 −3/5(k · v
f
/ω)
2
, (4.21)
which shows a quadratic wavenumber dependence. Although the RPA holds until there is
sufﬁcient screening and is valid for a lengthscale of a few atomic sizes, given the spatial
dispersion, we would not be aiming to resolve atoms or small molecules with visible or IR
light.
Much before this limit sets in, there are other mechanisms that limit the ability of this lens
to focus the waves with large transverse wavevectors. Foremost is the effect of absorption
which is maximal in the regions of large electromagnetic ﬁelds. The exponential growth of
the evanescent waves implies extremely large ﬁelds at the latter interface. The very large
absorption that would occur implies that the source would no longer be able to maintain the
energy supply and the ampliﬁcation of the very large wavevectors would not be possible. Thus
absorption sets its own large wavevector cutoff. The larger the absorption, the smaller the
corresponding cutoff wavevector. However, it does not imply that the ampliﬁcation process
would breakdown completely even in the presence of inﬁnitesimal absorption as has been
claimed [8].
In general, the perfect lens effect turns out to be rather sensitive to any deviations from
the lens conditions ε
−
= −ε
+
and µ
−
= −µ
+
[15, 142, 163–167]. The effects of absorption
can be included by including the imaginary part of ε
−
and µ
−
in the calculations. Thus they
represent a deviation from these conditions in the imaginary parts. For the silver lens, this is
even more acute as µ
−
= 1 represents a very large deviation from the perfect lens conditions.
Consider the transmission coefﬁcient for the ppolarized light across the slab,
T
p
(k
x
) =
4(k
z1
/ε
+
)(k
z2
/ε
−
) exp(ik
z2
d)
(k
z1
/ε
+
+ k
z2
/ε
−
)
2
−(k
z1
/ε
+
−k
z2
/ε
−
)
2
exp(2ik
z2
d)
. (4.22)
498 S A Ramakrishna
0 10 20 30
k
x
/k
0
10
3
10
2
10
1
10
0
10
1
10
2

T

(1)
(2)
(3)
(4)
(5)
0 10 20 30 40
k
x
/k
0
10
3
10
2
10
1
10
0
10
1
10
2

T

(1)
(2)
(3)
(4)
Figure 28. The absolute value of the transmitted ﬁeld at the image plane [T exp(ik
(1)
z
d)[, for the
ppolarization as a function of k
x
, showing the transmission resonances caused by the resonant
excitations of the slab plasmon polaritons. On the left panel, µ = +1 and (1) ε = −1 + i0.0,
k
0
d = 0.35; (2) ε = −1 + i0.01, k
0
d = 0.35; (3) ε = −0.9 + i0.01, k
0
d = 0.35;
(4) ε = −0.9 + i0.01, k
0
d = 0.5 and (5) ε = −1.1 + i0.01, k
0
d = 0.35. On the right panel,
(1) Re ε = −1 and (1) Im(ε) = 0, µ = −1.1, k
0
d = 0.35; (2) Im(ε) = 0.01, µ = −1.1,
k
0
d = 0.35; (3) Im(ε) = 0.01, µ = −1.1, k
0
d = 0.5; (4) Im(ε) = 0, µ = −2, k
0
d = 0.35.
The absorption broadens and softens the resonances. Deviations in ε affect the ppolarization more
than the deviations in µ.
Under the perfect lens conditions, the ﬁrst term in the denominator goes to zero for the
evanescent waves and the exponential in the second term decays faster than the exponential in
the numerator. However, if there was a mismatch of the material parameters (ε
−
= −1 + δε
and ε
+
= +1, say), then the ﬁrst term is no longer zero. In the quasistatic limit (k
x
¸ ω/c)
we ﬁnd that the two terms in the denominator are approximately equal when
k
x
d = −ln
¸
¸
¸
¸
δε
2
¸
¸
¸
¸
. (4.23)
For larger wave vectors, the ﬁrst termis larger than the second and hence the exponential decay
in the numerator asserts itself. We can take this to be the largest wavevector for which there
is an effective ampliﬁcation.
We can deﬁne the subwavelength resolution as the ratio of the optical wavelength to the
linear size (
min
) of the smallest resolved feature, and we obtain
res =
λ
min
=
−ln [δε/2[
2π
λ
d
. (4.24)
Thus the resolution depends logarithmically on the deviations of the material parameters and
inversely with the width of the slab, implying that the perfect lens effect is very sensitive
indeed to the material imperfections. The deviation, δε, could be in the imaginary part also,
with the same expression for the image resolution. Hence for our silver lens with a thickness
of onetenth the wavelength, we obtain a minimum resolved feature of about λ/2.5 from
this expression. Some extra subwavelength information may be available from the smaller
contributions at larger wavevectors.
The detailed effects of absorption and the deviations in the real part of µ and ε which
introduce retardation effects are, however, quite different. The absorption mainly causes an
exponentially decaying transmission for large wavevectors. The deviations in the real part on
the other hand allow resonant excitations of the slab plasmon modes discussed in section 3.7.
These resonant excitations of the slab would imply a very large transmission (for evanescent
waves T > 1 is allowed) as shown in ﬁgure 28. The direct excitation of these surface modes for
Physics of negative refractive index materials 499
imaging applications is undesirable, as these resonances will be disproportionately represented
in the image; yet, the existence of these resonances is essential, as the recovery of the evanescent
modes can be seen as the result of driving the surface plasmon far offresonance. Absorption
actually damps these transmission resonances and makes possible the very lensing action with
some subwavelength resolution capabilities in the case of the silver lens with µ = +1. The
perfect lens solution in the absence of losses is very special, and can lead to paradoxical
interpretations. The Poynting vector at the location of a point focus is strictly illdeﬁned.
However, the presence of a small and ﬁnite absorption makes the focus a small blob instead
and the Poynting vector also becomes well deﬁned. In the case of the asymmetric lens, we
note that the resolution works out to be
res =
λ
0
λ
min
= −
ln [ε
/
2
/2ε
3
[λ
0
4πd
(4.25)
(assuming ε
3
¸ ε
/
2
, and ε
3
¸ ε
1
), in the favourable case of low reﬂection. In the limit of large
ε
3
the resolution is actually enhanced [142].
Finally, one could ask if a ﬁnite aperture for the slab of the NRM would affect the near
ﬁeld imaging process. To a ﬁrst approximation, a slab with ﬁnite transverse width would
prevent some propagating waves at large angles to the imaging axis from being lost. But the
transport of the evanscent components will not be affected in the same manner and contribute
substantially to the image. Thus some information is lost due to the loss of some propagating
modes, but the nearﬁeld information is still preserved. The ﬁnite width of the slab, however,
can have other effects. The very large density of states near the corners can distort the near
ﬁeld image but will not affect it unless the transverse slab size becomes comparable with the
size of the image features. The plasmons on the surface of the ﬁnitely wide slab can form
standing waves along the transverse directions with the ﬁnite slab acting as a cavity [168] and
signiﬁcantly distort the image formation. But such cavity effects will be extremely limited in
the presence of absorption.
4.4. Numerical simulations and time evolution
After the demonstration of the perfect lens effect by Pendry, there were several attempts to
model this effect numerically. Some calculations obtained focusing [169, 170] while some
others [30] did not show any focusing at all and yet others obtained focusing but not sub
wavelength image resolution [171, 172]. In view of the controversy surrounding the effect at
that time, it was felt that some degree of numerical veriﬁcation based on exact simulations of
Maxwell’s equations was necessary. On afterthought though, one was attempting to verify an
exact analytical solution with numerical simulations which are necessarily approximate due
to the ﬁnite differencing.
Cummer [173] pointed out that the discrete ﬁelds do not obey the same dispersion relations
as the continuous ﬁelds do. For example, consider the discretization of the ﬁelds in the transfer
matrix method [17, 70]. The dispersion for free space along one of the lattice directions (for a
cubic lattice) for the discretized equations is
ω
2
(k
j
) =
c
2
a
2
4 sin
2
_
1
2k
j
a
_
. c
2
k
2
j
_
1 +
1
2k
2
j
a
2
_
, (4.26)
where j = x, y, z anda here denotes the lattice spacingof the discretizationgrid. Evidentlythis
converges to the continuum free space dispersion when k
j
a _ 1, i.e. when the discretization
is indeed very small compared with the inverse wavevector and they become identical only in
the limit of the discrete intervals going to zero. Thus any discrete method has a wavenumber
dependence and an inbuilt higher momentum cutoff that automatically precludes the perfect
500 S A Ramakrishna
focus. In the case of the perfect lens, as has been pointed out, the focusing is very sensitive to
meeting the prescribed conditions on ε = −1 and µ = −1. The different dispersion for the
discrete ﬁelds is equivalent to having slightly different material parameters which thus leads
to quite imperfect focusing as some have reported. Particularly for large wavevectors, the
different dispersion is acutely felt, and it was found to be only remedied by resorting to a larger
degree of discretization [149]. The level of subwavelength focusing that can be obtained by
a numerical calculation is a function of the ﬁniteness of the differencing scheme even under
the perfect conditions of no dissipation.
Cummer [173] also pointed out that the resonant surface plasmon modes at the interfaces
for n = −1 are always excited by any causal incident wave or pulse with a ﬁnite bandwidth.
In an FDTD calculation, these resonances at the two interfaces would ring indeﬁnitely, and
dissipation is always introduced into the material parameters in any such calculation to reach a
steady state in a reasonable amount of time. This probably explains the report of [30], where it
was found that the FDTD simulations never reached a steady behaviour for nonlossy NRMs.
The loss also damps out the ampliﬁcation for large wavevectors, and the image resolution
has been shown to be consistent with the estimate in section 4.3. These conclusions have
been conﬁrmed by FDTD calculations [174, 175] which used different boundary conditions to
simulate evanescent waves.
As brieﬂy noted before, the temporal evolution of the perfect focus is interesting as the
focus reﬁnes in time, and under ideal conditions reﬁnes to a point focus at inﬁnite time, which
is what we obtain with the timeharmonic solutions. Acalculation based on Laplace transform
methods [163] indicated that for a source that is sharply switched on at t = 0, the lens forms
a steady image only at times that are of the order of the absorption time (t ∼ 1/Im(εµ)ω).
GomezSantos [158] has pointed out that the resonant excitation of the plasmons on the
two surfaces is identical to the problem of coupled identical oscillators with an external force
applied on one of the oscillators. The equations of motion for such a system of oscillators are
¨ x
l
+ γ ˙ x
l
ω
2
0
x
l
+ k
c
x
r
= f (t ), (4.27)
¨ x
r
+ γ ˙ x
r
ω
2
0
x
r
+ k
c
x
l
= 0, (4.28)
where x
l
and x
r
are the amplitudes for the left and right oscillators and k
c
is the coupling
between them. First note that forcing the system at ω
0
with no damping makes only x
r
non
zero in the steady state. The left oscillator does not move at all and all the excitation energy
is passed on to the right oscillator. This corresponds to the zero reﬂection in our case. In the
steady state the solutions for a timeharmonic excitation f
0
exp(−iωt ) are
x
l
=
(ω
2
0
−ω
2
+ iγ ω)f
0
(ω
2
0
−ω
2
+ iγ ω)
2
−k
4
c
, x
r
=
−k
2
c
f
0
(ω
2
0
−ω
2
+ iγ ω)
2
−k
4
c
(4.29)
and it can be seen for the case of nonzero dissipation at resonance that the left oscillator is
also excited. For large dissipation the right oscillator is no longer excited, which corresponds
to the noneffective ampliﬁcation in the case of the NRM slab. From this model we can
further consider the temporal evolution of the imaging process by considering the rate at
which energy is transferred between the two surfaces. Note that the slab plasmon polariton
modes are decoupled from the singe plasmon frequency. Thus a source with a sharp onset has
a frequency bandwidth and would excite them. The transmitted ﬁeld across the lossless slab
then has an amplitude [158]
E
t
(t ) = (t )A(t ) exp(−iω
0
t ) exp(k
x
d) (4.30)
in the quasistatic limit and A(t ) . 1/2(ω
kx
t )
2
, where ω
kx
is the frequency spacing
between the two slab plasmon polariton modes at a wavevector k
x
. As seen in section 3.7, the
Physics of negative refractive index materials 501
spacing between the slab plasmon modes monotonically decreases and is exponentially small
at k
x
. Thus the large wavevector components are very small at short times and a cutoff time
can be deﬁned by ω
kx
t
kx
∼ 1 for every wavevector k
x
before which it does not appreciably
contribute to the image formation. Thus the image is initially illdeﬁned and becomes better
and better with the passage of time until it reaches the optimal resolution that is possible, given
the levels of absorption in the system. Also note that without absorption, the slab plasmon
polariton modes would ring indeﬁnitely, never allowing the formation of a steady image.
5. Designing superlenses
As discussed in the previous section, the perfect lens is in principle possible, if we had
perfect negative refractive index materials with the speciﬁed material parameters. Inspite
of the imperfect NRMs, subwavelength resolution is still possible to some extent, and we
call such lenses with some degree of subwavelength image resolution superlenses. It turns
out that the ﬂat slab lens of Veselago is only one lens of a whole class of perfect lenses or
superlenses that are possible. Negative refractive media are the optical analogues of anti
matter, in the sense that passage through NRMs nulliﬁes the effect of passing through positive
media with ε and µ of equal magnitude [149] for radiation. This concept of complementarity
has led to the generalization of the Veselago lens to complementary media [13, 149] as well
as other geometries including the cylindrical [177, 178] and the spherical [179] geometries.
Interestingly, a change in the geometry enables magniﬁcation of even the nearﬁeld image. We
will discuss these ideas in this section.
Any deviations from these lead to a drastic reduction in the possible subwavelength
resolution of the lens. In fact, it is absorption that is the main culprit responsible for reducing
the resolution, and as we sawin the case of the silver lens, we just obtain some subwavelength
image information. Note that it is the ratio λ/d (wavelength to slab thickness) that dominates
the resolution, the logarithm term being a relatively weakly varying function. For example, if
λ/d = 3, we ﬁnd that to achieve a resolution factor of 10, δε must be no greater than ∼10
−11
!
However, for λ/d = 10, δε can be as large as ∼0.002 and we can still achieve the same
resolution. Thus, minimizing the effects of absorption is crucial in order to make these lenses
work. Restructuring the lens to reduce these effects is possible and we ﬁrst discuss some
of the strategies that have been reported. But eventually it does appear that the composite
structures should incorporate media with an active gain (as in a laser) in order to counter
the dissipation [14, 180, 181].
5.1. Overcoming the limitations of real materials
The main limitation of the present day NRMs is the large amount of dissipation in them. The
large ampliﬁcation of the nearﬁeld modes causes large amounts of dissipation, which does
not allow restoration of the nearﬁeld modes. As mentioned at the end of section 4.3, just
by choosing the asymmetric conﬁguration with a large dielectric constant on the side of the
image we can increase the resolution of the slab lens [142]. Although it appears similar to the
concept that is used in immersion lens nearﬁeld imaging [183], in this case it is more related
to the fractional deviation from the perfect lens conditions that is reduced when the magnitude
of the real part of the dielectric constant is very large. In fact, it has also been shown [142] that
working at a slightly different frequency compared with when ε = −1 can lead to the resonant
excitation of slab plasmons which can also help increase the maximum possible resolution.
Solymar and coworkers [164] have considered that moving the image plane closer to the lens
can actually compress the image.
502 S A Ramakrishna
d/5 d/5
d/10
z=0 z=2d
ε
−
ε
+
δ
z=0 z=2d
y
z
x
Figure 29. Chopping the slab of NRM into thin slices and distributing them around does not affect
the image. The ﬁeld distributions for a system of ﬁve such pieces is depicted. If this process of
slicing into thinner and thinner layers is continued, we end up with a very unusual anisotropic
effective medium as δ → 0.
One can reduce the effects of absorption by reducing the slab thickness. Then the
evanescent ﬁelds amplify to much smaller extents, reducing the absorption and enhancing
the image resolution. But then the distance over which we transfer the image is much smaller.
We note that in the ideal lossless case, we can just take the original thick slab, chop it into thin
layers and distribute it around between the source and image planes without affecting the result
of perfect imaging [164, 182]. The perfect impedance matching makes this possible and the
evanescent ﬁelds amplify within the NRM layers and decay in the positive layers as shown in
ﬁgure 29. This does, however, make an enormous difference to the dissipation in the system.
As the ﬁelds nowhere amplify to the same extent as they would have to in the single slab,
the corresponding dissipation is much lower. We note that the source–lens edge and the lens
edge–image plane distance have considerably reduced, and in the limit of the layer thickness
becoming extremely small, the layered system merely transfers the image of the source from
one edge to the other in the sense of an optical ﬁbre bundle.
In the limit of a very small layer thickness and a large number of layers (δ → 0 as shown
in ﬁgure 29(b)), the entire layered system may be regarded as an effective medium. If one
considers the response of the system for an electric ﬁeld applied parallel to the layers, one
obtains
¸D
x
) =
ε
0
2
(ε
+
E
x
+ ε
−
E
x
) = ε
0
ε
xx
E
x
, (5.1)
where the tangential component of the E ﬁeld is used. Similarly, considering the response of
a ﬁeld applied normal to the layers,
¸E
z
) =
1
2ε
0
(ε
−1
+
D
z
+ ε
−1
−
D
z
) = (ε
0
ε
zz
)
−1
D
z
, (5.2)
where the continuity of the normal component of the Dﬁeld is implied. Under the perfect lens
conditions, we have the effective medium dielectric permittivity components
ε
xx
=
(ε
+
+ ε
−
)
2
→ 0, (5.3)
ε
zz
=
2
(ε
−1
+
+ ε
−1
−
)
→ ∞ (5.4)
and the effective mediumis an unusually anisotropic medium. Similar relations can be derived
for the magnetic permeability using the continuity of H

and B
⊥
. Noting that the wave
Physics of negative refractive index materials 503
propagation for the ppolarized radiation in such a uniaxially anisotropic medium has the
dispersion
k
2
x
+ k
2
y
ε
zz
+
k
2
z
ε
xx
= ε
0
µ
yy
ω
2
c
2
, (5.5)
we have k
z
= 0 as the only solution. Thus every wave passes through this anisotropic material
without a change in amplitude or phase [182]. This anisotropic material is equivalent to a
system of perfectly conducting wires along the layer normal embedded a medium with ε = 0.
In the static limit and no dissipation, this corresponds to inﬁnitely thin wires connecting the
two end faces point to point. Thus, it acts as an optical ﬁbre bundle, but unlike a conventional
optical ﬁbre bundle, it also transfers the nearﬁeld information. When there is dissipation, it
merely corresponds to a ﬁnite ﬁbre thickness and, consequently, a ﬁnite resolution. A slab and
layered structure of positive media and anisotropic media with indeterminate permittivity and
permeability tensors (i.e. all the components do not have the same sign) has also been shown
to work as a layered lens with partial refocusing [184, 185].
We can write the transmission through such a layered system of thickness 2d in the quasi
static limit as [182]
T
p
.
1
cos(iε
//
−
k
x
d) + (1/2)(ε
+
+ ε
−1
+
) sin(iε
//
−
k
x
d)
, (5.6)
where the dissipation (represented by ε
//
−
= Im(ε
−
)) in the NRM layers has been taken into
account. Note now that the resolution of this layered lens is
res =
λ
2πε
−//
(5.7)
and the resolution depends inversely on the absorption rather than the logarithm of δε, which
varies very slowly. Thus the scope for improvement here by making the NRM less dissipative
is much higher.
We note here that the study of layered structures with alternating layers of positive media
and NRMs has been of considerable interest. The anomalous tunnelling across such media has
also been considered in [186]. Such stacks can have a newkind of bandgap which corresponds
to a zero refractive index (n
−
d
−
+ n
+
d
+
= 0) [187]. Note that such zero index bandgaps can
also occur for large layer thicknesses and are not related to our effective medium theory, above
which there is also a perfect impedance matching. In fact, the layers in the zero index band
gap material should not be impedance matched. Interestingly the bandgap of these zero index
layered systems is independent of periodicity and is robust against disorder. Other interesting
properties of such layered structures such as omnidirectional and large bandgaps [188–190],
defect modes [188, 191], Bragg reﬂectors [192] and beam shaping [193], among others, have
been considered.
Ultimately the resolution of the layered lens is also limited by dissipation. The point is
that we need to have a supermetal that shows no dissipation. Another approach would be to
put in by optical ampliﬁcation that which is lost by dissipation [180]. Consider the perfect
lens conditions with a lossy NRM: we see the possibility that if
Im(ε
−
) > 0, Im(µ
−
) > 0, (5.8)
then the condition may still be satisﬁed, provided that
Im(ε
+
) < 0, Im(µ
+
) < 0. (5.9)
In other words, the positive mediumshould be optically amplifying (as in a laser gain medium)
in order to counter the effects of absorption in the negative medium. Then the perfect lens
504 S A Ramakrishna
k
x
/k
o
X
Figure 30. The transmission function (left) and the electromagnetic ﬁeld intensity at the image
plane (right) obtained (a) with a single slab of silver of 40 nm thickness, (b) when the slab is split
into eight thin layers of δ/2 = 5 nm thicknesses. (c) Layered silverdielectric stack with optical
gain and δ/2 = 5 nm and (d) layered silverdielectric stack with optical gain and δ/2 = 10 nm.
ε
±
= ±1 ∓i0.4 in (c) and (d).
conditions would be ‘perfectly’ met and the perfect image would result even with lossy
NRM. Thus, the use of optical ampliﬁcation is implied in the perfectlens condition itself.
Keeping with standard practice, we will subsume all the processes relating to the ampliﬁcation
(or absorption) of the electromagnetic wave into the negative (or positive) imaginary part of
the dielectric constant.
More realistically, we can have optical ampliﬁcation mainly for the electric dipole
transitions. The incorporation of gain through the magnetic permeability would necessarily
be through electronic ampliﬁers in the metamaterials for the microwave and rf frequencies.
We will consider here the case of thin layers of silver with positive amplifying dielectric media
in between. To satisfy the perfect lens conditions in both the real and the imaginary parts for
the ppolarized light, we would require
ε
+
= ε
/
+
−iε
//
+
, ε
−
= −ε
/
+
+ iε
//
+
. (5.10)
Then the perfect lens conditions would be perfectly met and the perfect image would result, at
least in the quasistatic limit. This can, for example, be accomplished by using a semiconductor
laser material such as GaN or AlGaAs for the positive medium and silver for the negative
medium. Using blue/ultraviolet (UV) light to pump the AlGaAs, one can make the AlGaAs
now optically amplifying in the red region of the spectrum, where one can satisfy the perfect
lens condition for the real parts of the dielectric constant. By adjusting the pump laser intensity,
the imaginary part of the positive gain medium can be tuned. The imaging can now be carried
out in the red. Of course, it would be possible to use other materials and correspondingly
different wavelengths of light. Alternatively one could also use other high gain processes such
as Raman gain for this purpose. As a note of caution, we note that in the presence of the
intense ﬁeld enhancements that are expected in this system the gain is likely to get saturated.
The largest ﬁeld enhancements will be for the largest transverse wavevectors. If we make
the layers very thin, the local ﬁeld enhancements will not be as intense and the gain might
not get completely bleached. In general, however, we do expect that the effective gain will
be somewhat reduced and the corresponding enhancements of the image resolution will be
smaller.
As a proof of the principle, the transmission function for the layered lens, with and without
ampliﬁcations, andthe images of twocloselyspacedsources as resolvedbythe respective lenses
are shown in ﬁgure 30. For comparison, we also show the case of the original single slab of
Physics of negative refractive index materials 505
d d
d d
(a) (b)
Figure 31. A pair of complementary media nullify the effect of each other for the passage of light.
The light and dark regions indicate opposite refractive indices. The path for the propagating modes
in the media will not be straight lines in general. The overall effect is as if the space of thickness
2d did not exist for an observer on the righthand side.
silver as the lens (solid line and δ/2 = 40 nm) and a layered but gainless system. The two
peaks in the image for the single slab can hardly be resolved, while they are clearly resolved in
the case of the layered system with no gain. The improvement in the image resolution for the
layered system with gain over the corresponding gainless systems is obvious, with the sharp
edges of the slits becoming visible. Interestingly, the transmittance of the layered lens with
gain does not decay rapidly with increase in the number of layers and a total stack thickness
of even a few wavelengths [180]. Although the transmission function is not constant with the
wavevector due to the layer plasmon resonances, we emphasize that the high spatial frequency
components are transferred across. Knowledge of the transmission function of the lens would
therefore enable one to recover a clean image from the observed image.
5.2. The generalized perfect lens theorem
The negative refractive slab can also be considered as optical antimatter for light in the sense
it cancels out the effects of propagation through an equal amount of positive index media
(see ﬁgure 31). This cancellation is applicable to the phase changes for propagating waves and
the amplitude changes for the evanescent waves. Then it would appear for the world ahead
of the slab that optically a thickness of 2d had been removed. In fact, the perfect focusing
action occurs for more general conditions than a homogeneous slab with n = −1. It happens
even when the dielectric permittivity and the magnetic permeability vary along the transverse
directions, provided that the variation is just the same but with opposite signs:
ε
1
= +ε(x, y), µ
1
= +µ(x, y), ∀0 < z < d, (5.11)
ε
2
= −ε(x, y), µ
2
= −µ(x, y), ∀d < z < 2d, (5.12)
where ε(x, y) and µ(x, y) are some arbitrary functions of x and y and can take positive or
negative values (see ﬁgure 31(b) for a schematic picture). This is the generalized lens theorem
of [149] which we will merely state here and the proof of which we will give in appendix.
Consider the most general dielectric permittivity and magnetic permeability tensor for
spatial region 1 (0 < z < d):
˜ ε
1
=
_
_
ε
1xx
ε
1xy
ε
1xz
ε
1yx
ε
1yy
ε
1yz
ε
1zx
ε
1zy
ε
1zz
_
_
, ˜ µ
1
=
_
_
µ
1xx
µ
1xy
µ
1xz
µ
1yx
µ
1yy
µ
1yz
µ
1zx
µ
1zy
µ
1zz
_
_
. (5.13)
506 S A Ramakrishna
Similarly, for region 2 (d < z < 2d)
˜ ε
2
=
_
_
ε
2xx
ε
2xy
ε
2xz
ε
2yx
ε
2yy
ε
2yz
ε
2zx
ε
2zy
ε
2zz
_
_
, ˜ µ
2
=
_
_
µ
2xx
µ
2xy
µ
2xz
µ
2yx
µ
2yy
µ
2yz
µ
2zx
µ
2zy
µ
2zz
_
_
. (5.14)
Note that the tensorial components can be speciﬁed functions of (x, y). It can be shown
(see appendix for the proof) that the ﬁelds to the left and right of the interface between the two
slabs are exactly identical but in opposite order as long as the two slabs have complementary
material parameters:
˜ ε
1
=
_
_
ε
1xx
ε
1xy
ε
1xz
ε
1yx
ε
1yy
ε
1yz
ε
1zx
ε
1zy
ε
1zz
_
_
, ˜ µ
1
=
_
_
µ
1xx
µ
1xy
µ
1xz
µ
1yx
µ
1yy
µ
1yz
µ
1zx
µ
1zy
µ
1zz
_
_
(5.15)
and
˜ ε
2
=
_
_
−ε
1xx
−ε
1xy
+ε
1xz
−ε
1yx
−ε
1yy
+ε
1yz
+ε
1zx
+ε
1zy
−ε
1zz
_
_
, ˜ µ
2
=
_
_
−µ
1xx
−µ
1xy
+µ
1xz
−µ
1yx
−µ
1yy
+µ
1yz
+µ
1zx
+µ
1zy
−µ
1zz
_
_
. (5.16)
Then the two complementary media have an optical sum of zero. This result can be further
generalized to any region that is mirror antisymmetric about a given plane [149]. Such
regions correspond to effectively optical null regions. These results have also been veriﬁed
numerically by considering speciﬁc examples with spatially varying complementary media
with mirror antisymmetry [149].
This theorem, combined with the possibility of using coordinate transformations [194]
to map the slab geometries into other geometries [149], allows us to generate a wide class
of arrangements, all of which will exhibit the property of transferring images of sources in a
perfect sense.
5.3. The perfect lens in other geometries
Traditionally lenses have been used to project images and more importantly magnify or
demagnify the image of a source. The slab of NRM that is a perfect lens maps every point on
the object plane to another point on the image plane. But the size of the image is identical to that
of the source. This is due to the invariance in the transverse direction, and the transverse wave
vector (k
x
, k
y
) is preserved in the refraction process. The curved surfaces of a conventional
lens break this translational symmetry in the transverse direction and enable the change in
the size of the image. In general, to obtain magniﬁcation and demagniﬁcation of the images,
the translational symmetry would have to be broken and curved surfaces will need to be
employed. The perfect lens action in the case of the NRM slab was crucially dependent on the
neardegeneracy of the surface plasmons in the case of the slab, and curved surfaces, in general,
have a completely different dispersion for the surface plasmons. Only in the quasistatic limit
does a cylindrical surface with negative, homogeneous ε have a nearly degenerate plasmon
dispersion, and a cylindrical shell in two dimensions with ε = −1 was shown to act as a lens,
transferring in and out the image of a source [177]. In fact, the use of conformal mapping
on the solutions to the Laplace equation (in the quasistatic limit) for the slab was shown to
yield an entire class of nearﬁeld lenses in two dimensions. In the more general case, one uses
the technique of coordinate transformations to map the solutions of the slab geometry into
other geometries. Note that a distortion of space results in a change of the ε and µ tensors
in general. Thus in many cases, the transformed geometry would involve spatially varying
(heterogeneous) medium parameters. Here we will present three cases of a cylindrical lens,
Physics of negative refractive index materials 507
z
y
x
a
2
a
0
a
1
a
3
Figure 32. A spherical shell with negative ε
−
(r) ∼ −1/r and µ
−
(r) ∼ −1/r images a source
located inside the shell into the external region. The media outside have a positive refractive index,
but ε
−
(r) ∼ 1/r and µ
−
(r) ∼ 1/r. The ampliﬁcation inside the spherical shell of the otherwise
decaying ﬁeld is schematically shown.
a spherical lens and a cornerlens which focuses the source back on itself. The lenses in the
curved geometries result in a magniﬁcation of the image including the nearﬁeld evanescent
modes and the cornerlens is interesting as an open resonator.
5.3.1. Cylindrical lenses. The conformal transformation
z
/
= ln z (5.17)
was shown [177] to map the NRM slab into a cylindrical shell (see ﬁgure 32 and ignore the
dependence along the Z axis). Note that the conformal transformation is valid for the Laplace
equation and the solutions valid in the extreme nearﬁeld approximation. In general we can
transform Maxwell’s equations for the slab lens of an NRM, into a cylindrical shell with ε
−
(r)
and µ
−
(r) whose inner and outer radii are a
1
and a
2
, embedded in a positive medium. As it
has been shown [194], if we deﬁne the new coordinates q
1
(x, y, z), q
2
(x, y, z) and q
3
(x, y, z),
then in the new frame, the material parameters and ﬁelds are given by
˜ ε
i
= ε
i
Q
1
Q
2
Q
3
Q
2
i
, ˜ µ
i
= µ
i
Q
1
Q
2
Q
3
Q
2
i
, (5.18)
˜
E
i
= Q
i
E
i
,
˜
H
i
= Q
i
H
i
, (5.19)
where
Q
2
i
=
_
∂x
∂q
i
_
2
+
_
∂y
∂q
i
_
2
+
_
∂z
∂q
i
_
2
. (5.20)
Now consider the transformation to cylindrical coordinates
x = r
0
e
/
0
cos φ, y = r
0
e
/
0
sin φ, z = Z, (5.21)
so that
Q
=
r
0
0
e
/
0
, Q
φ
= r
0
e
/
0
, Q
Z
= 1. (5.22)
508 S A Ramakrishna
We obtain the transformed dielectric permittivity and magnetic permeability to be
˜ ε
=
0
ε
, ˜ ε
φ
=
−1
0
ε
φ
, ˜ ε
Z
=
r
2
0
0
exp
_
2
0
_
ε
z
, (5.23)
˜ µ
=
0
µ
, ˜ µ
φ
=
−1
0
µ
φ
, ˜ µ
Z
=
r
2
0
0
exp
_
2
0
_
µ
z
. (5.24)
Choosing the scale factor
0
= 1 and
ε
r
= µ
r
= +1, ε
φ
= µ
φ
= +1, ε
z
= µ
z
= +
1
r
2
, ∀ r < a
1
, (5.25)
ε
r
= µ
r
= −1, ε
φ
= µ
φ
= −1, ε
z
= µ
z
= −
1
r
2
, ∀ a
1
< r < a
2
, (5.26)
ε
r
= µ
r
= +1, ε
φ
= µ
φ
= +1, ε
z
= µ
z
= +
1
r
2
, ∀ a
2
< r, (5.27)
where r = r
0
exp(), we obtain
˜ ε
= ˜ µ
= +1, ˜ ε
φ
= ˜ µ
φ
= +1, ˜ ε
z
= ˜ µ
z
= +1, ∀ <
0
ln
_
a
1
r
0
_
, (5.28)
˜ ε
= ˜ µ
= −1, ˜ ε
φ
= ˜ µ
φ
= −1, ˜ ε
z
= ˜ µ
Z
= −1, ∀
0
ln
_
a
1
r
0
_
< <
0
ln
_
a
2
r
0
_
,
(5.29)
˜ ε
= ˜ µ
= +1, ˜ ε
φ
= ˜ µ
φ
= +1,
˜ ε
z
= ˜ µ
Z
= +1, ∀
0
ln
_
a
2
r
0
_
< (5.30)
in the , φ, Z plane, which is identical to the situation of the Veselago lens. Hence the new
systemmust accordinglyact as a cylindrical lens andtransfer images inandout of the cylindrical
lens. The image will be formedonthe surface a
3
= a
0
(a
2
/a
1
)
2
andthere will be a magniﬁcation
of the image by the factor
M=
_
a
2
a
1
_
2
. (5.31)
Note that these cylindrical lens are also shortsighted in the same manner as the slab lens.
They can focus sources from inside to the outside only when a
2
1
/a
2
< r < a
1
, and the other
way around from outside to the inner world, when the source is located in a
2
< r < a
2
2
/a
1
.
They cannot focus outside this region. Further note that for the optical axis along the
direction, the generalized lens theorem predicts that the variations in the transverse φ, Z
directions are irrelevant. Hence the medium parameters in general could also be arbitrary
functions of φ and Z.
In general, a whole class of lenses can be generated in two dimensions using the
transformation techniques [177, 178].
5.3.2. A spherical lens. The transformation technique can be similarly used to map the slab
lens into spherical coordinates (r = r
0
e
/
0
, θ, φ), and using the generalized lens theorem,
a spherical shell of NRM with ε
−
(r) ∼ −1/r and µ
−
(r) ∼ −1/r embedded in a positive
medium with ε
+
(r) ∼ +1/r and µ
+
(r) ∼ +1/r will act as a spherical lens projecting sources
in and out of the shell [149]. For this speciﬁc case, we will present a more conventional proof
in terms of the TE and TM modes in the spherical geometry [179].
Physics of negative refractive index materials 509
Consider the spherically symmetric system shown in ﬁgure 32. Under the circumstances
of spherical symmetry, it is sufﬁcient to specify the quantities (r · E) and (r · H), which will
constitute a full solution to the problem. Consider now the TM polarized modes r · H = 0,
implying that only the electric ﬁelds have a radial component E
r
. Operating on Maxwell’s
equation with ∇,
∇ ∇ E = iωµ
0
∇ [µ(r)H] =
ω
2
c
2
µ(r)ε(r)E + iω
∇µ(r)
µ(r)
∇ E (5.32)
and we have
∇ · D = ∇ · [ε(r)E] = ∇ε(r) · E + ε(r)∇ · E = 0. (5.33)
And if we assume ε(r) = ε(r) and µ(r) = µ(r), we have
∇ · E = −
ε
/
(r)
rε(r)
r · E = −
ε
/
(r)
rε(r)
(rE
r
). (5.34)
We note the following identities for later use:
∇ ∇ E = ∇(∇ · E) −∇
2
E, (5.35)
∇
2
(r · E) = r · ∇
2
E + 2∇ · E. (5.36)
Using the fact that ε only depends on r, we also note that
r · ∇(∇ · E) = −
∂
∂r
_
ε
/
(r)
ε(r)
(rE
r
)
_
+
_
ε
/
(r)
ε(r)
E
r
_
. (5.37)
Using the above four equations, we can obtain the equation for E,
∇
2
(rE
r
) +
∂
∂r
_
ε
/
(r)
ε(r)
(rE
r
)
_
+
ε
/
(r)
rε(r)
(rE
r
) + ε(r)µ(r)
ω
2
c
2
(rE
r
) = 0, (5.38)
which is separable, and the spherical harmonics are a solution to the angular part. Hence the
solution is (rE
r
) = U(r)Y
lm
(θ, φ), where the radial part, U(r), satisﬁes
1
r
2
∂
∂r
_
r
2
∂U
∂r
_
−
l(l + 1)
r
2
U +
∂
∂r
_
ε
/
(r)
ε(r)
U
_
+
ε
/
(r)
rε(r)
U + ε(r)µ(r)
ω
2
c
2
U = 0. (5.39)
If we choose ε(r) = αr
p
and µ(r) = βr
q
, we can have a solution U(r) ∼ r
n
and we get
[n(n + 1) −l(l + 1) + p(n −1) + p]r
n−2
+
αβω
2
c
2
r
p+q+n
= 0, (5.40)
implying p + q = −2 and
n
±
=
1
2
_
−(p + 1) ±
_
(p + 1)
2
+ 4l(l + 1) −
4αβω
2
c
2
_
. (5.41)
Hence the general solution can be written as
E
r
(r) =
l,m
[n
+
A
lm
r
n
+
−1
+ n
−
B
lm
r
n
−
−1
]Y
lm
(θ, φ) (5.42)
and a similar solution can be obtained for the TE modes with r · E = 0.
Now assuming an arbitrary source at r = a
0
, we can now write down the electric ﬁelds of
the TM modes in the different regions for the negative spherical shell of ﬁgure 2 as
E
(1)
(r) =
l,m
[n
+
A
(1)
lm
r
n
+
−1
+ n
−
B
(1)
lm
r
n
−
−1
]Y
lm
(θ, φ), a
0
< r < a
1
, (5.43)
E
(2)
(r) =
l,m
[n
+
A
(2)
lm
r
n
+
−1
+ n
−
B
(2)
lm
r
n
−
−1
]Y
lm
(θ, φ), a
1
< r < a
2
, (5.44)
E
(3)
(r) =
l,m
[n
+
A
(3)
lm
r
n
+
−1
+ n
−
B
(3)
lm
r
n
−
−1
]Y
lm
(θ, φ), a
2
< r < ∞ (5.45)
510 S A Ramakrishna
and similarly for the magnetic ﬁelds. Note that B
(1)
lm
correspond to the ﬁeld components of the
source located at r = a
0
. For causal solutions A
(3)
lm
= 0. Nowthe tangential components of the
magnetic ﬁelds and the normal components of the displacement ﬁelds have to be continuous
across the interfaces. Under the conditions p = −1, q = −1, ε
+
(a
1
) = −ε
−
(a
1
) and
ε
+
(a
2
) = −ε
−
(a
2
), we have
A
(1)
lm
= 0, (5.46)
A
(2)
lm
=
_
1
a
2
1
_
√
l(l+1)−αβω
2
/c
2
B
(1)
lm
, B
(2)
lm
= 0, (5.47)
B
(3)
lm
=
_
a
2
2
a
2
1
_
√
l(l+1)−αβω
2
/c
2
B
(1)
lm
. (5.48)
The lenslike property of the system becomes clear by writing the ﬁeld outside the spherical
shell as
E
(3)
r
=
1
r
_
a
2
2
a
2
1
r
_
√
l(l+1)−αβω
2
/c
2
B
(1)
lm
Y
lm
(θ, φ). (5.49)
Hence, apart from a scaling factor of 1/r, the ﬁelds on the sphere r = a
3
= (a
2
2
/a
2
1
)a
0
are
identical to the ﬁelds on the sphere r = a
0
. We also have a spatial magniﬁcation in the image
by a factor of a
2
2
/a
2
1
. As with the cylindrical lens, this spherical lens will be able to image
sources which are also located only within in a ﬁnite distance of the spherical layer. Note
that the positive medium outside the shell is also required to have spatially varying material
parameters ∼1/r. In the presence of ﬁnite dissipation, using the ideas of the asymmetric
lens [142], we can terminate the spatially varying media at some ﬁnite but large distance. Let
us note a couple of points about the above perfect lens solutions in the spherical geometry. First,
for r > a
3
, i.e. points outside the image surface, the ﬁelds appear as if the source were located
on the spherical image surface (r = a
3
). However, this is not true for points a
2
< r < a
3
within the image surface. Second, note that our imaging direction is along r, and ε and µ can
nowbe arbitrary functions of θ and φ, with only the condition of complementarity between the
negative and positive regions. Third, given that ε
−
(a
2
) = −ε
+
(a
2
), we have the perfect lens
solutions if and only if n
+
= −n
−
, which implies that p = −1 in equation (5.41). Although
the solutions given by equation (5.42) occur in any medium with εµ ∼ 1/r
2
, the perfect lens
solutions only occur for ε ∼ µ ∼ 1/r. Here we have written down the solutions for the TM
modes. The solutions for the TE modes can be similarly obtained.
Again there is a considerable simpliﬁcation that is possible in the extreme nearﬁeld
limit [179]. Then the imaging for the TM modes becomes independent of µ and only requires
ε
−
∼ −1/r
2
. Similarly the TE modes become independent of ε, and the only condition is that
µ
−
∼ −1/r
2
. Further the constraint that the positive media outside also be spatially varying
can be dropped. In this limit it can be shown that the largest multipole that can be resolved is
again limited by the dissipation in the NRM and is approximately [179]
l
max
.
ln{3ε
1
ε
3
/[ε
i
(a
1
)ε
i
(a
2
)]]
2 ln(a
2
/a
1
)
, (5.50)
where ε
1
= −Re(ε
−
(a
1
)) and ε
3
= −Re(ε
−
(a
2
)) for the perfect lens conditions and
ε
i
(r) ∼ 1/r
2
is the imaginary part of the permittivity for the negative medium shell.
5.3.3. A perfect twodimensional corner lens. Consider an inﬁnitely extended corner of
NRM with n = −1 as shown in ﬁgure 33(a). A ray diagram shows that a source located in the
Physics of negative refractive index materials 511
n=–1
n=–1
n=–1
n=–1
n=–1
n=–1 n=–1
n=–1
n=–1
n=–1
n=–1
y
x
y
l l
φ
(a)
(b)
(a) (b)
x
Figure 33. (a) A corner of NRM with n = −1 refocuses a source. This corner lens can be shown
to be perfect by mapping it to a periodic layered stack. (b) A double corner of NRM refocuses
the source back onto itself and creates an image in the diagonally opposite quadrant. This focus is
also perfect in that it includes the nearﬁeld components and can be shown by the mapping into a
periodically layered stack as shown.
second quadrant can be focused into the fourth quadrant. In fact combining two such corners
as shown in ﬁgure 33(b) makes it even more exotic: classical rays emanating from a source
placed in any one of the quadrants are returned to the source point [195]. Using the generalized
lens theorem, a more rigorous result can be proved, namely that the focusing with these corner
lenses is true for the evanescent components as well and is perfect!
Consider the periodic arrangements of slabs shown beside the corners with material
parameters
ε(φ) = µ(φ) =
_
_
_
+1 −
π
2
< φ < π,
−1 −π < φ < −
π
2
(5.51)
in the ﬁrst case (a) of a single corner and
ε(φ) = µ(φ) =
_
_
_
+1 −
π
2
< φ < 0 and
π
2
< φ < π,
−1 −π < φ < −
π
2
and 0 < φ <
π
2
(5.52)
in the second case (b) of the double corner lens. Using the transformation as in the case of the
cylindrical lens,
x = r
0
e
/
0
cos φ, y = r
0
e
/
0
sin φ, z = Z, (5.53)
gives us
˜ ε
= ˜ µ
= ˜ ε
φ
= ˜ µ
φ
=
_
+1 ˜ ε
Z
= ˜ µ
Z
= r
2
0
e
2/
0
−1 ˜ ε
Z
= ˜ µ
Z
= −r
2
0
e
2/
0
(5.54)
in the respective positive and negative quadrants. The conditions of complementarity are
satisﬁed. The dependence along the r axis is irrelevant as it is orthogonal to the imaging axis
along φ. Hence we conclude that the images in both the cases of the single corner (a) and the
double corner (b) will be perfect.
Note that case (b) is unusually singular as each source images on each other and the ﬁelds
grow indeﬁnitely in time. It has also been shown that all the surface modes of this system are
degenerate and the density of modes at the double corner is very large [149]. Such a system
can form a very interesting open resonator in which light can be trapped for long times.
512 S A Ramakrishna
6. Experimental evidence for NRMs and superlenses
This ﬁeld has been one in which theoretical developments have led the experimental
activities. NRMs had been predicted by Veselago in 1967. It was the theoretical recipes
of Pendry et al [2, 3, 43] for designing negative material parameters that made the real
developments possible. But the main catalyst for the sudden increase in the interest in NRMs
came from the experimental demonstration of NRMs by the San Diego group in 2000 [4, 5].
These have now been followed up by several experimental groups. But on the whole, the
growth of experimental activity has been rather slow compared with the almost frenetic pace
in the theoretical studies. Part of the reason could be that the experimental samples are not easy
to fabricate, particularly the microstructures for high frequency operation. This is also possibly
the reason why most of the experiments have been performed at microwave frequencies. But
there is enough experimental evidence to place many of the theoretical ideas on a very safe
footing. In this section,we will brieﬂy cover the main experimental activities.
6.1. Experiments on negative refractive index
The ﬁrst experiments on the plasmonic nature of thin wire arrays were reported in [43], where
the thin wire arrays were made using spark chamber technology. These consisted of sheets of
polysterene rohacell sheets with 20 µm goldplated tungsten wires laid on them with a 5 mm
spacing, and these sheets were stacked up, with alternate sheets rotated by 90˚. This stack
would be expected to have two diagonal components of the dielectric tensor negative below
the plasma frequency. The measured reﬂectivity and transmission of these samples indicated
a plasma frequency of about 9 GHz, which could be reduced to 6 GHz by diluting the density
of the wires by half, thus conﬁrming the theoretical prediction of a low frequency plasmonic
medium. Other experiments [196] have used a looplike structure to increase the inductance
of thicker wires and obtained a low frequency cutoff medium.
There have been several experimental efforts at obtaining negative magnetic materials after
the San Diego experiments [4] brought them into the limelight. There have been studies of
the single scatterer response of these SRR media [62]. There have been attempts at producing
isotropic SRR media by combining three orthogonal SRRs into a single threedimensional
scatterer [63]. Marqu´ es et al [198] have implemented a modiﬁed SRR structure where
the two planar split rings are capacitively coupled not through their edges but through the
broadside by depositing equal sized split rings on either side of an insulating board. This
ameliorates the bianisotropy of the SRRs as well. Wiltshire et al [23, 24] have experimentally
implemented the Swiss roll design for operation at radiofrequencies. The dispersion of
µ(ω) has been measured and found to match accurately with the theoretical predictions.
By using the large effective magnetic permeability of these Swiss rolls at frequencies just
below the resonance frequency, these media have been used as effective ﬂux tubes for the
radiofrequency magnetic ﬁelds and the potential for MRI has been demonstrated. Very
recently the magnetic properties of microscale SRRs has been demonstrated at terahertz
frequencies [27].
The San Diego experiments [4, 5] combined the planar SRRs and the thin wire arrays to
make the world’s ﬁrst NRM (see ﬁgure 12). Transmission experiments were ﬁrst carried out
on uniaxial SRR media, which consisted of planar SRRs of copper rings printed on circuit
boards, placed in a microwave guide, and showed a low transmission band for radiation
with the magnetic ﬁeld along the SRR axis. This could be attributed to the negative µ at
frequencies above the LC resonance of the SRRs. Combining this with 0.8 mm wires placed
symmetrically between the SRRs with the electric ﬁeld along the thin wire resulted in an
Physics of negative refractive index materials 513
enhanced transmission within the negative µ band compared with the transmission of the
bare wire media, indicating a passband due to a negative refractive index. A prism made
with a similar twodimensional isotropic NRM was shown to exhibit the negative refraction
effect [5]. These experiments have, however, been criticized for the low transmittivity of the
samples [9]. Several reasons can qualitatively account for the low transmittivity including
the dissipation in the circuit boards, the large wire diameter and the corresponding large
impedance mismatch. But it certainly does not negate the conclusion of a negative refractive
index.
The experiments of Parazzoli et al [25,26] have provideddecisive experimental evidence of
a negative refractive index in composite samples of thin wire arrays and SRRs. These
experiments used large sized threedimensional isotropic samples and were conducted in free
space rather than in waveguides. These experiments used thin wires of 0.25 mmthickness and a
lowabsorption substrate (Rogers 5880) and measured both the reﬂected and transmitted power.
The insertion loss was shown to be smaller (about −1 dBcm
−1
) [26] and the contribution of
the conduction losses in the SRRs appears to dominate. The issue of loss in these NRM
is complex, and a clear verdict on the exact level of losses is yet to emerge. A prism
made of this medium was shown to again refract microwaves negatively [25], where it
was also numerically shown that the angle of refraction was insensitive to the degree of
absorption in the sample. The transmission passband for the composite SRR and thin wire
medium has also been obtained experimentally by other groups [197]. Marqu´ es et al [90]
have shown that placing SRRs within a waveguide operating below the cutoff also creates
a passband below the cutoff frequency which can be attributed to a negative refractive
index.
Loaded L–C backward wave transmission lines have been implemented by Eleftheriades
et al [131, 132] and Liu et al [130]. The transmission line approach has shown that NRM
structures with a low loss and wide bandwidth are possible, particularly at microwave
frequencies, something that the other resonant structures are not capable of. Eleftheriades
et al experimentally implemented twodimensional loaded L–C backward transmission lines
and have demonstrated the negative refraction effect [131]. They demonstrated that an
image placed in a positive medium near an interface with the NRM would create an image
inside the NRM in the sense of the Veselago lens. Liu et al [130] have demonstrated
the feasibility of a forward microwave coupler between two backward wave structures
using a very short coupling length. This becomes possible due to the inverse dependence
of the wavevector on the frequency (β = −1/(ω
√
LC)). This idea has also been
extended to coupling between ordinary transmission lines and backward wave transmission
lines [199].
6.2. Photonic crystals and the Veselago ﬂat lens
There has been extensive activity on the negative refraction effect with photonic crystals.
Cubukcu et al [200] have shown the negative refraction effect in such media by using
the reversed transverse shift a beam undergoes upon oblique transmission through a slab
compared with normal media. Using a twodimensional array of alumina rods, a Gaussian
beam of microwaves at about 15 GHz was shown to undergo a reversed transverse shift.
Parimi et al [201] have demonstrated the ﬂat lens imaging, again with a twodimensional
photonic crystal of cylindrical alumina rods, at about 9 GHz. More recently this effect has
been demonstrated at telecommunications infrared frequencies using even a low contrast two
dimensional photonic crystal [202]. These measurements are consistent with the theoretical
predictions of Luo et al [36]. The imaging with a ﬂat lens has also been extended to sonic
514 S A Ramakrishna
waves, and phononic crystals have also been shown to focus sound in a similar manner [203].
The complete generality of this effect becomes clear in the recent ﬂat lens imaging by liquid
surface waves where a twodimensional lattice of rigid rods was used to act as a crystal for the
surface waves.
6.3. Demonstrating superlenses
By superlenses, we mean lenses with subwavelength resolution enabled by NRMs. Super
lenses and subwavelength focusing capabilities have enormous implications for nearﬁeld
imaging and nearﬁeld optical lithography. The interest in it has been enormous. The sub
wavelength resolving capabilities of the superlenses stem from the way the evanescent near
ﬁeldcomponents are transmittedacross it: these components are ampliﬁedinmagnitude. There
is now deﬁnite experimental evidence to conﬁrm the essentials of the theory of superlenses.
But more reﬁned measurements are expected in the near future.
Grbic and Eleftheriades [205] have experimentally noted superlens behaviour with a
planar NRMlens made out of twodimensional transmission lines. Subwavelength image res
olutions at about 1 GHz were clearly measured with the capacitively loaded transmission grid
kept between the source and detector, compared with the case of diffraction limited patterns.
The measured electric ﬁelds also clearly showed the ampliﬁcation of the evanescent waves
across the slab. The subwavelength image resolution was limited only by the levels of absorp
tion in the NRM. Thus, most of the elementary features of the theory have been clearly veriﬁed.
Liu et al [204] used a reversed attenuated total reﬂection (RATR) experiment to directly
measure the transmitted ﬁelds across thin ﬁlms of silver. Using the surface roughness to linearly
couple a propagating mode into evanescent surface modes on a silver surface, the transmitted
evanescent ﬁelds of the surface plasmon modes were coupled out using a hemisphere on the
other side of the silver ﬁlm. The measured transmittance increased exponentially with ﬁlm
thickness up to some maximum and then decayed quickly with a further increase in thickness
which was caused by the increased dissipation in the ﬁlm. The behaviour of the transmittance
was quantitatively predicted by Pendry’s expressions and the material parameters of silver [71]
and has offered indirect evidence of the capability for superlensing of the silver lens. More
recently, submicron imaging using the silver slab lens has also been demonstrated [206] in
the context of optical lithography. The images were obtained using a 120 nm thick slab of
silver which, however, precluded the possibility of subwavelength image resolution due to
losses and retardation. The images with lower resolution were obtained which, however, were
demonstrated to be superior to the traditional proximity imaging techniques. At the time of
writing, subwavelength image resolution at optical frequencies using the silver lens is yet to
be demonstrated.
7. Conclusions
In conclusion, we have reviewed the recent developments in the area of negative refractive index
materials. Although predicted a long time ago, in 1967, the practical creation of these materials
had to wait for recipes to selectively make materials with a negative dielectric permittivity and
magnetic permeability. We have examined the problem of making materials with negative
material parameters which are mostly structured composites. The design of metamaterials
that show a negative refractive index is an extreme form of electromagnetic engineering. The
negative material parameters mostly result from an overscreened, underdamped response
due to the structural resonances. A negative refractive index is obtained by interleaving two
structures that individually show a negative ε or negative µ. It should be noted that the sizes of
Physics of negative refractive index materials 515
the structures in these materials are usually much smaller than the wavelength of radiation that
enables them to be treated effectively as homogeneous media. The simpliﬁcation that results
is considerable as the complex internal microscopical fast varying ﬁelds become irrelevant
and need not concern us anymore. There are still very signiﬁcant issues of nonlocal ﬁelds,
bianisotropy and homogenization that need to be carefully looked at. There is now sufﬁcient
experimental evidence for a negative refractive index and the associated effects. The high
degree of absorption in these materials, primarily arising due to conduction losses in the
metallic structures, is a cause for concern, and one of the main challenges remaining in this
area is developing newdesigns for metamaterials with smaller losses. But on the whole NRMs
have to be accepted as today’s reality. They are very dispersive and dissipative, but very real
nevertheless. Calculations indicate that scaling to high frequencies, up to the nearinfrared and
perhaps even the optical frequencies, is possible. The fundamental limitation comes from the
ﬁnite electronic mass as real currents have to ﬂow in these structures.
Several new effects that are possible in NRMs have been discussed. The immensely
dispersive nature of these materials can create many newsurprises. These materials can support
a host of surface modes and have a large potential in the newly developing area of plasmonics.
The resonances in the metamaterials can cause intense local ﬁeld enhancements. This makes
it a new fertile ground for nonlinear phenomena. The study of nonlinear metamaterials is
only starting now, but surprises will almost certainly abound. We have discussed the backward
wave structures using transmission lines that can act as NRMs at microwave frequencies and
radiofrequencies and the photonic bandgap materials that can show the negative refraction
effect. There are considerable problems of homogenization in the photonic bandgap materials,
and they are perhaps better described by the band theory than by homogeneous negative
material parameters. But these PBG materials that show the allangle negative refraction
have a tremendous technological potential due to the low degree of absorption in them, and
some of the earliest technical innovations and applications will possibly occur using them.
The perfect lens that can be made from a slab of NRM is perhaps the most dramatic
of the new effects that are possible. We have discussed in detail this effect where both the
propagating and the evanescent nearﬁeld modes participate in the image formation. We have
considered howthe excitation of surface plasmon modes on the surface of the NRMs are crucial
to restoration of the amplitudes of the evanescent modes. The dissipative and dispersive nature
of the NRM, however, spoils the perfect focus and imposes limits on the maximum extent of
subwavelength image resolution that we can obtain. We have discussed the generalization
of this effect to spatially varying media and the concept of optically complementary media
that behave in a sense like optical antimatter. This generalization, along with a method of
coordinate transformations, enables us to generate classes of surfaces with degenerate surface
plasmons and perfect lenses in different geometries. Several new conﬁgurations of the super
lenses that include the nearﬁeld have been presented. The immense potential of the perfect lens
effect for nearﬁeld imaging, electromagnetic couplers and waveguides in photonic devices
cannot be underemphasized. NRM enable us to manipulate the nearﬁeld radiation which up
to now was untouchable.
The developments in the past fewyears have only offered us a glimpse of what is possible.
NRMare newelectromagnetic materials with no conventional analogues in many cases. Many
of the intuitions of conventional electromagnetism fail here and one needs to develop new
physical intuition here. Further developments will crucially depend on our ability to make
better quality NRM, particularly from the point of view of low losses and large bandwidths.
NRM can enable new nearﬁeld technologies when coupled with innovations in antennae,
integrated circuits, waveguide technology and photonic circuits that can outperform their
conventional counterparts.
516 S A Ramakrishna
Acknowledgments
I would like to acknowledge with thanks Professor John Pendry for encouragement and
enlightening discussions. I would also like to thank him and Imperial College London, and
Professor NKumar and the Raman Research Institute, Bangalore, for hospitality during part of
the time when this reviewwas written. Much of my knowledge of negative index materials has
come from discussions with Olivier Martin, David Smith, N Kumar, Stephen O’Brien, Mike
Wiltshire, Martin McCall and Sebastien Guenneau. I would like to thank Harshawardhan
Wanare and Wayne Williams for critically reading the manuscript.
Appendix. Proof of the generalized perfect lens theorem
Here we present the proof of section 5.2, that two media with complementary optical behaviour
amount effectively to an optical null space as shown in ﬁgure 31. Consider the two slabs with
the spatiallyvaryingdielectric permittivitytensors andmagnetic permeabilitytensors described
by equations (5.13) and (5.14), respectively. Now the electromagnetic ﬁelds have to satisfy
Maxwell’s equations:
∇ E = iωµ
0
˜ µH, ∇ H = −iωε
0
˜ εE. (A1)
Now decomposing the ﬁelds into the Fourier components in the two slabs,
E
1
(x, y, z) = exp(ik
1z
z)
k
x
,k
y
exp[i(k
x
x + k
y
y)]
_
_
E
1x
(k
x
, k
y
)
E
1y
(k
x
, k
y
)
E
1z
(k
x
, k
y
)
_
_
, (A2)
E
2
(x, y, z) = exp(ik
2z
z)
k
x
,k
y
exp[i(k
x
x + k
y
y)]
_
_
E
2x
(k
x
, k
y
)
E
2y
(k
x
, k
y
)
E
2z
(k
x
, k
y
)
_
_
, (A3)
where the Bloch conditions can be assumed. Substituting these ﬁelds into Maxwell’s equations
and separating the components, we have
k
y
E
1z
(k
x
, k
y
) −k
1z
E
1y
(k
x
, k
y
) = −ωµ
0
k
/
x
,k
/
y
[µ
1xx
(k
x
, k
y
; k
/
x
, k
/
y
)H
1x
(k
/
x
, k
/
y
)
+µ
1xy
(k
x
, k
y
; k
/
x
, k
/
y
)H
1y
(k
/
x
, k
/
y
) + µ
1xz
(k
x
, k
y
; k
/
x
, k
/
y
)H
1z
(k
/
x
, k
/
y
)], (A4)
k
1z
E
1x
(k
x
, k
y
) −k
x
E
1z
(k
x
, k
y
) = −ωµ
0
k
/
x
,k
/
y
[µ
1yx
(k
x
, k
y
; k
/
x
, k
/
y
)H
1x
(k
/
x
, k
/
y
)
+µ
1yy
(k
x
, k
y
; k
/
x
, k
/
y
)H
1y
(k
/
x
, k
/
y
) + µ
1yz
(k
x
, k
y
; k
/
x
, k
/
y
)H
1z
(k
/
x
, k
/
y
)], (A5)
k
x
E
1y
(k
x
, k
y
) −k
y
E
1x
(k
x
, k
y
) = −ωµ
0
k
/
x
,k
/
y
[µ
1zx
(k
x
, k
y
; k
/
x
, k
/
y
)H
1x
(k
/
x
, k
/
y
)
+µ
1zy
(k
x
, k
y
; k
/
x
, k
/
y
)H
1y
(k
/
x
, k
/
y
) + µ
1zz
(k
x
, k
y
; k
/
x
, k
/
y
)H
1z
(k
/
x
, k
/
y
)] (A6)
and
k
y
H
1z
(k
x
, k
y
) −k
1z
H
1y
(k
x
, k
y
) = ωε
0
k
/
x
,k
/
y
[ε
1xx
(k
x
, k
y
; k
/
x
, k
/
y
)E
1x
(k
/
x
, k
/
y
)
+ε
1xy
(k
x
, k
y
; k
/
x
, k
/
y
)E
1y
(k
/
x
, k
/
y
) + ε
1xz
(k
x
, k
y
; k
/
x
, k
/
y
)E
1z
(k
/
x
, k
/
y
)], (A7)
Physics of negative refractive index materials 517
k
1z
H
1x
(k
x
, k
y
) −k
x
H
1z
(k
x
, k
y
) = ωε
0
k
/
x
,k
/
y
[ε
1yx
(k
x
, k
y
; k
/
x
, k
/
y
)E
1x
(k
/
x
, k
/
y
)
+ε
1yy
(k
x
, k
y
; k
/
x
, k
/
y
)E
1y
(k
/
x
, k
/
y
) + ε
1yz
(k
x
, k
y
; k
/
x
, k
/
y
)E
1z
(k
/
x
, k
/
y
)], (A8)
k
x
H
1y
(k
x
, k
y
) −k
y
H
1x
(k
x
, k
y
) = ωε
0
k
/
x
,k
/
y
[ε
1zx
(k
x
, k
y
; k
/
x
, k
/
y
)E
1x
(k
/
x
, k
/
y
)
+ε
1zy
(k
x
, k
y
; k
/
x
, k
/
y
)E
1y
(k
/
x
, k
/
y
) + ε
1zz
(k
x
, k
y
; k
/
x
, k
/
y
)E
1z
(k
/
x
, k
/
y
)]. (A9)
Now consider the substitution
E
2x
(k
x
, k
y
) = E
1x
(k
x
, k
y
), E
2y
(k
x
, k
y
) = E
1y
(k
x
, k
y
), E
2z
(k
x
, k
y
) = −E
1z
(k
x
, k
y
),
(A10)
H
2x
(k
x
, k
y
) = H
1x
(k
x
, k
y
), H
2y
(k
x
, k
y
) = H
1y
(k
x
, k
y
), H
2z
(k
x
, k
y
) = −H
1z
(k
x
, k
y
)
(A11)
and
k
2z
= −k
1z
, (A12)
˜ ε
2
(k
x
, k
y
; k
/
x
, k
/
y
) =
_
_
−ε
1xx
(k
x
, k
y
; k
/
x
, k
/
y
) −ε
1xy
(k
x
, k
y
; k
/
x
, k
/
y
) +ε
1xz
(k
x
, k
y
; k
/
x
, k
/
y
)
−ε
1yx
(k
x
, k
y
; k
/
x
, k
/
y
) −ε
1yy
(k
x
, k
y
; k
/
x
, k
/
y
) +ε
1yz
(k
x
, k
y
; k
/
x
, k
/
y
)
+ε
1zx
(k
x
, k
y
; k
/
x
, k
/
y
) +ε
1zy
(k
x
, k
y
; k
/
x
, k
/
y
) −ε
1zz
(k
x
, k
y
; k
/
x
, k
/
y
)
_
_
,
(A13)
˜ µ
2
(k
x
, k
y
; k
/
x
, k
/
y
) =
_
_
−µ
1xx
(k
x
, k
y
; k
/
x
, k
/
y
) −µ
1xy
(k
x
, k
y
; k
/
x
, k
/
y
) +µ
1xz
(k
x
, k
y
; k
/
x
, k
/
y
)
−µ
1yx
(k
x
, k
y
; k
/
x
, k
/
y
) −µ
1yy
(k
x
, k
y
; k
/
x
, k
/
y
) +µ
1yz
(k
x
, k
y
; k
/
x
, k
/
y
)
+µ
1zx
(k
x
, k
y
; k
/
x
, k
/
y
) +µ
1zy
(k
x
, k
y
; k
/
x
, k
/
y
) −µ
1zz
(k
x
, k
y
; k
/
x
, k
/
y
)
_
_
(A14)
which solves Maxwell’s equations in region 2 inside the slab and also matches the ﬁelds across
the boundaries of the slabs. Hence we have the ﬁeld
E(x, y, z = 2d) = E(x, y) exp(−ik
z
d) = E(z, y, z = 0). (A15)
Thus the ﬁelds are exactly repeated in region 2 but in exactly the opposite order. This completes
the proof of the theorem.
References
[1] Veselago V G 1967 Usp. Fiz. Nauk 92 517
Veselago V G 1968 Sov. Phys.—Usp. 10 509
[2] Pendry J B, Holden A J, Stewart W J and Youngs I 1996 Phys. Rev. Lett. 76 4773
[3] Pendry J B, Holden A J, Robbins D J and Stewart W J 1999 IEEE Trans. Microwave Theory Tech. 47 2075
[4] Smith D R, Padilla W J, Vier D C, NematNasser S C and Schultz S 2000 Phys. Rev. Lett. 84 4184
[5] Shelby R A, Smith D R and Schultz S 2001 Science 292 77
[6] Pendry J B 2000 Phys. Rev. Lett. 85 3966
[7] Valanju P M, Walser R M and Valanju A P 2002 Phys. Rev. Lett. 88 187401
[8] Garcia N and NietoVesperinas M 2002 Phys. Rev. Lett. 88 207403
[9] Garcia N and NietoVesperinas M 2002 Opt. Lett. 27 885
[10] McCall M, Weiglhoffer W S and Lakhtakia A 2002 Eur. J. Phys. 23 353
[11] Markos P and Soukoulis C M Preprint condmat/0212136
[12] Shen J Q 2004 Preprint condmat/0402213
[13] Pendry J B 2004 Contemp. Phys. 45 191
518 S A Ramakrishna
[14] Pendry J B and Smith D R 2004 Phys. Today 57 37
[15] Shen J T and Platzmann P M 2002 Appl. Phys. Lett. 80 3286
[16] Yablonovitch E 1987 Phys. Rev. Lett. 58 2059
Yablonovitch E 1993 J. Phys.: Condens. Matter 5 2443
[17] Pendry J B 1996 J. Phys.: Condens. Matter 8 1085
[18] Jackson J D 1999 Classical Electrodynamics (Singapore: Wiley)
[19] Landau L D, Lifschitz E M and Pitaevskii L P 1984 Electrodynamics of Continuous Media 2nd edn (Oxford:
Pergamon)
[20] Kotthaus J P and Jaccarino V 1972 Phys. Rev. Lett. 28 1649
Grunberg P and Metawe F 1977 Phys. Rev. Lett. 39 1561
Sandercock J R and Wettling W 1979 J. Appl. Phys. 50 7784
Camley R E and Mills D L 1982 Phys. Rev. B 26 1280
Remer L, Luthi B, Sauer H, Geick R and Camley R E 1986 Phys. Rev. Lett. 56 2752
Dumelow T, Camley R E, Abraha K and Tilley D R 1998 Phys. Rev. B 58 897
[21] Pendry J B and O’Brien S 2002 J. Phys.: Condens. Matter 14 7409
[22] O’Brien S, MacPeake D, Ramakrishna S A and Pendry J B 2004 Phys. Rev. B 69 241101
[23] Wiltshire M C K, Pendry J B, Young I R, Larkman D J, Gilderdale D J and Hajnal J V 2001 Science 291 848
[24] Wiltshire M C K, Hajnal J V, Pendry J B and Edwards D J 2003 Opt. Express 11 709
[25] Parazzoli C G, Greegor R B, Li K, Kontenbah B E C and Tanielian M H 2003 Phys. Rev. Lett. 90 107401
[26] Li K, McLean S J, Greegor R B, Parazzoli C G and Tanielian M H 2003 Appl. Phys. Lett. 82 2535
[27] Yen T J, Padilla WJ, Fang N, Vier DC, Smith DR, Pendry J B, Basov DNand Zhang X2004 Science 303 1494
[28] O’Brien S and Pendry J B 2002 J. Phys.: Condens. Matter 14 6383
[29] Lindell I V, Tretyakov S A, Nikoskinen K I and Ilvonen S 2001 Microwave Opt. Tech. Lett. 31 129
[30] Ziolkowski R W and Heyman E 2001 Phys. Rev. E 64 056625
[31] Ritchie R H 1957 Phys. Rev. 106 874
[32] Raether H 1980 Excitation of Plasmons and Interband Transitions by Electrons (Berlin: Springer)
[33] Pendry J B 2001 Proc. NATO Advanced Study Inst. (Crete, NATO ASI Series) ed C M Soukoulis (Dordrecht:
Kluwer)
[34] Barnes W L, Dereux A and Ebbesen T W 2003 Nature 424 824
[35] Born M and Wolf E 1989 Principles of Optics (Oxford: Pergamon)
[36] Luo C, Johnson S G, Joannopoulos J D and Pendry J B 2002 Phys. Rev. B 65 201104
[37] Pines D and Bohm D 1952 Phys. Rev 85 338
Bohm D and Pines D 1953 Phys. Rev. 92 609
[38] Ebbesen T W, Lezec H J, Ghaemi H F, Thio T and Wolff P A 1998 Nature 391 667
MartinMoreno L et al 2004 Phys. Rev. Lett. 86 1114
MartinMoreno L, GarciaVidal F J, Lezec H J, Degiron A and Ebbesen T W 2003 Phys. Rev. Lett. 90 167401
[39] Harris S E, Field J E and Imamoglu A 1990 Phys. Rev. Lett. 64 1107
Scully M O 1992 Phys. Rep. 219 191
Harris S E 1997 Phys. Today 50 36
[40] Scully M O and Zubairy M S 1997 Quantum Optics (Cambridge: Cambridge University Press)
[41] Sarychev A K and Shalaev V M 2000 Phys. Rep. 335 275
[42] Lagarkov A N and Sarychev A K 1996 Phys. Rev. B 53 6318
Matitsine S M, Hock K M, Liu L, Gan Y B, Lagarkov A N and Rozanov K N 2003 J. Appl. Phys. 94 1146
[43] Pendry J B, Holden A J, Robbins D J and Steward W J 1998 J. Phys.: Condens. Matter 10 4785
[44] Sievenpiper D F, Sickmiller M E and Yablonovitch E 1996 Phys. Rev. Lett. 76 2480
[45] Maslovsky S I, Tretyakov S A and Belov P A 2002 Microwave Opt. Tech. Lett. 35 47
[46] Belov P A, Marque´ s R, Maslovski S I, Nefedov I S, Silveirinha M, Simovski C R and Tretyakov S A 2003
Phys. Rev. B 67 13103
[47] Mikhailov S A 1997 Phys. Rev. Lett. 78 4135
[48] Pendry J B 1997 Phys. Rev. Lett. 78 4136
[49] Walser R M, Valanju A P and Valanju P M 2001 Phys. Rev. Lett. 87 119701
[50] Pendry J B 2002 private communication
[51] van Coeverden D V, Sprik R, Tip A and Lagendijk A 1996 Phys. Rev. Lett. 77 2412
[52] Koschny T, Markos P, Smith D R and Soukoulis C M 2003 Phys. Rev. E 68 065602
[53] Makhnovsky D P and Panina L V 2003 J. Appl. Phys. 93 4120
[54] Simovsky C R and He S 2003 Phys. Lett. A 311 254
[55] Rotman W 1962 IRE Trans. Antennas Propagat. AP10 82
[56] King R J, Thiel D V and Park K S 1983 IEEE Trans. Antennas Propag. AP31 471
Physics of negative refractive index materials 519
[57] Brown J 1960 Prog. Dielectr. 2 195
[58] Pitarke J M, GarciaVidal F J and Pendry J B 1998 Phys. Rev. B 57 15261
[59] Takeda H and Yoshino K 2003 Phys. Rev. B 67 245109
[60] Anderson P W 1963 Phys. Rev. 130 439
[61] Kittel C, Fahy S and Louie S G 1988 Phys. Rev. B 37 642
[62] GayBalmaz P and Martin O J F 2002 J. Appl. Phys. 92 2929
[63] GayBalmaz P and Martin O J F 2002 Appl. Phys. Lett. 81 939
[64] Hsu A c, Cheng Y K, Chen K H, Chern J L, Wu S C, Chen C F, Chang H, Lien Y H and Shy J T 2004 Japan.
J. Appl. Phys. 43 L176
[65] Markos P and Soukoulis C M 2002 Phys. Rev. B 65 033401
[66] Huang Y C, Hsu YJ, Lih J S and Chern J L 2004 Japan. J. Appl. Phys. 43 L190
[67] Sievenpiper D, Zhang L, Braos R F J, Alex´ opolous N G and Yablonovitch E 1999 IEEE Trans. Microwave
Theory Tech. 47 2059
[68] Broas R F J, Sievenpiper D and Yablonovitch E 2001 IEEE Trans. Microwave Theory Tech. 49 1262
[69] Enoch S, Tayeb G and Maystre D 1999 Opt. Commun. 161 171
Gralak B, Enoch S and Tayeb G 2000 J. Opt. Soc. Am. 17 1012
[70] Pendry J B and MacKinnon A 1992 Phys. Rev. Lett. 69 2772
Pendry J B 1994 J. Mod. Opt. 41 209
[71] Johnson P B and Christy R W 1972 Phys. Rev. B 6 4370
[72] Smith D R, Schultz S, Markos P and Soukoulis C M 2002 Phys. Rev. B 65 195104
[73] Kumar N 2003 Proc. Asia Academic Seminar (Hyderabad) ed P R Rao et al (International Advanced Research
Centre for Powder Metallurgy & New Materials and Japan Society for Promotion of Science) p 33
[74] Dimmock J O 2003 Opt. Express 11 2397
[75] Panoiu N C and Osgood R M 2003 Opt. Commun. A 223 331
[76] Panina L V, Grigorenko A N and Makhnovskiy D P 2002 Phys. Rev. B 66 155411
[77] Hrabar S, Eres Z and Bartolic J 2002 Proc. Eur. Micowave Conf. (Milan) pp 327–30
[78] Shamonina E, Kalinin V A, Ringhofer K H and Solymar L 2002 J. Appl. Phys. 92 6252
[79] Shamonin M, Shamonina E, Kalinin V and Solymar L 2004 J. Appl. Phys. 95 3778
[80] Kong J A 1990 Electromagnetic Wave Theory (New York: WileyInterscience)
[81] Saadoun M M I and Engheta N 1992 Microwave Opt. Tech. Lett. 5 184
[82] Marques R, Medina F and RaﬁiElIdrissi R 2002 Phys. Rev. B 65 144440
[83] O’Brien S and Pendry J B 2002 J. Phys.: Condens. Matter 14 4035
[84] Shelby R A, Smith D R, NematNasser S C and Schultz S 2001 Appl. Phys. Lett. 78 489
[85] Markos P and Soukoulis C M 2002 Phys. Rev. E 65 036622
[86] Pokrovsky A L and Efros A L 2002 Phys. Rev. Lett. 89 093901
[87] Marqu´ es R and Smith D R 2004 Phys. Rev. Lett. 92 059401
[88] Simovski C R, Belov P A and He S 2004 IEEE Trans. Antennas Propag. 51 2582
[89] Simovski C R and Sauviac B 2004 Preprint condmat/0403047
[90] Marqu´ es R, Martel J, Medina F and Mesa F 2002 Phys. Rev. Lett. 89 183901
[91] Bergman D J and Stroud D 1992 Solid State Phys. 46 147
[92] Stroud D 1975 Phys. Rev. 12 3368
[93] Hunderi O 1973 Phys. Rev. B 7 3419
[94] Milton G W 2002 The Theory of Composites (Cambridge: Cambridge University Press)
[95] Kramers H A 1929 Phys. Z. 30 522
de Kronig R 1942 Natuurk 9 402
[96] Kircheva P P and Hadjichristov G B 1994 J. Phys. B: At. Mol. Opt. Phys. 27 3781
[97] Tokunaga E, Terasaki A and Kobayashi T 1993 Phys. Rev. A 47 R4581
[98] Peiponnen K E, Vartiainen E M and Saarinen J J 2004 Phys. Rev. A 69 043818
[99] Ginzburg V L and Meiman N N 1964 Sov. Phys.—JETP 19 169
[100] Diener G 1997 Phys. Lett. A 235 118
[101] Glasgow S, Ware M and Peatross J 2001 Phys. Rev. E 64 046610
[102] Smith D R and Kroll N 2000 Phys. Rev. Lett. 85 2933
[103] Pokrovsky A L and Efros A L 2002 Solid State Commun. 124 283
[104] Ramakrishna S A and Martin O J F 2004 Preprint condmat/0402570
[105] Depine R A and Lakhtakia A 2004 Microwave Opt. Tech. Lett. 41 315
[106] Wang J and Lakhtakia A 2002 Microwave Opt. Tech. Lett. 33 465
[107] Loudon R 2002 J. Mod. Opt. 49 821
[108] Lakhtakia A and McCall M W 2004 Preprint condmat/0402001
520 S A Ramakrishna
[109] Zhang Y, Fleugel B and Mascarenhas A 2003 Phys. Rev. Lett. 91 157404
[110] Pendry J B 2000 Phys. World 13 27
[111] Lu J, Grzegorczyk T M, Zhang Y, Pacheco J, Wu B, Kong J A and Chen M 2003 Opt. Express 11 723
[112] Luo C, Ibanescu M, Johnson S G and Joannopoulos J D 2003 Science 299 368
[113] Goos F and H¨ anchen H 1947 Ann. Phys. Lpz. 1 333
[114] Kong J A, Wu B I and Zhang Y 2002 Appl. Phys. Lett. 80 2084
[115] Berman P R 2002 Phys. Rev. E 66 067603
[116] Lakhtakia A 2002 Electromagnetics 23 71
Lakhtakia A 2002 Preprint condmat/0305133
[117] Ziolkowski R W 2003 Opt. Express 11 662
[118] Artmann K 1948 Ann. Phys. Lpz. 2 87
[119] Shadrivov I V, Sokhorukov A A and Kivshar Y S 2003 Appl. Phys. Lett. 82 3820
[120] Shadrivov I V, Zharov A A and Kivshar Y S 2003 Appl. Phys. Lett. 83 2713
[121] Landauer R and Martin Th 1994 Rev. Mod. Phys. 66 217
[122] Pendry J B and Smith D R 2003 Phys. Rev. Lett. 90 029703
[123] Smith D R, Schurig D and Pendry J B 2002 Appl. Phys. Lett. 81 2713
[124] Lu W T, Sokoloff J B and Sridhar S 2002 Preprint condmat/0207689
[125] Lu W T, Sokoloff J B and Sridhar S 2004 Phys. Rev. E 69 026604
[126] Caloz C, Chang C C and Itoh T 2001 J. Appl. Phys. 90 5483
[127] Pacheco J, Grzegorczyk T M, Wu B I, Zhang Y and Kong J A 2002 Phys. Rev. Lett. 89 257401
[128] Foteinopoulou S, Economou E N and Soukoulis C M 2003 Phys. Rev. Lett. 90 107402
[129] Dutta Gupta S, Arun R and Agarwal G S 2004 Phys. Rev. B 69 113104
[130] Liu L, Caloz C, Chang CC and Itoh T 2002 J. Appl. Phys. 92 5560
[131] Eleftheriades G V, Iyer A K and Kremer P C 2002 IEEE Trans. Microwave Theory Tech. 50 2702
[132] Grbic A and Eleftheriades G V 2002 J. Appl. Phys. 92 5930
[133] Notomi M 2000 Phys. Rev. B 62 10696
[134] Luo C, Johnson S G and Joannopoulos J D 2002 Appl. Phys. Lett. 81 2352
[135] Chien HT, Tang HT, Kuo CH, Chen CC and Ye Z 2003 Preprint condmat/0312582
[136] Kuo CH and Ye Z 2003 Preprint condmat/0310423
Kuo CH and Ye Z 2003 Preprint condmat/0312288
[137] Shen Y R 1984 The Principles of NonLinear Optics (New York: Wiley)
[138] Zharov A A, Shadrivov I V and Kivshar Y S 2003 Phys. Rev. Lett. 91 037401
[139] Agranovich V M, Shen Y R, Baughman R H and Zakhidov A A 2004 Phys. Rev. B 69 165112
[140] Ruppin R 2000 Phys. Lett. A 277 61
[141] Ruppin R 2001 J. Phys.: Condens. Matter 13 1811
[142] Ramakrishna S A, Pendry J B, Smith D R, Schurig D and Schultz S 2002 J. Mod. Opt. 49 1747
[143] Darmanyan S A, Nevi´ ere M and Zakhidov A A 2003 Opt. Commun. 225 233
[144] Ruppin R 2000 Solid State Commun. 116 411
[145] Klimov V V 2002 Opt. Commun. 211 183
[146] Cory H and Barger A 2003 Microwave Opt. Tech. Lett. 38 392
[147] Alu A and Engheta N 2002 Microwave Opt. Tech. Lett. 35 460
[148] Kuzmiak V and Maradudin A A 2002 Phys. Rev. B 66 045116
[149] Pendry J B and Ramakrishna S A 2003 J. Phys.: Condens. Matter 15 6345
[150] Greffet J J and Carminati R 1997 Prog. Surf. Sci. 56 133
[151] Pohl D and Courjon D 1993 NearField Optics (Dordrecht: Kluwer)
[152] Williams J M 2001 Phys. Rev. Lett. 87 249703
[153] Pendry J B 2001 Phys. Rev. Lett. 87 249704
[154] ’t Hooft G W 2001 Phys. Rev. Lett. 87 249701
[155] Pendry J B Phys. Rev. Lett. 87 249702
[156] Ramakrishna S A and Armour A D 2003 Am. J. Phys. 71 562
[157] Luo C, Johnson S G, Joannopoulos J D and Pendry J B 2003 Phys. Rev. B 68 045115
[158] GomezSantos G 2003 Phys. Rev. Lett. 90 077401
[159] Haldane F D M Preprint condmat/0206420
[160] Pendry J B 2003 Phys. Rev. Lett. 91 099701
[161] Feise M W, Bevelacqua P J and Schneider J B 2002 Phys. Rev. B 66 035113
[162] Ashcroft N W and Merimin N D 1976 Solid State Physics (Philadelphia: Saunders College)
[163] Smith D R, Schurig D, Rosenbluth M, Schultz S, Ramakrishna S A and Pendry J B 2003 Appl. Phys. Lett.
82 1506
Physics of negative refractive index materials 521
[164] Shamonina E, Kalinin V A, Ringhofer K H and Solymar L 2001 Electron. Lett. 37 1243
[165] Fang N and Zhang X 2003 Appl. Phys. Lett. 82 161
[166] Ye Z 2003 Phys. Rev. B 67 193106
[167] Merlin R 2004 Appl. Phys. Lett. 84 1290
[168] Chen L, He S and Shen L 2004 Phys. Rev. Lett. 92 107404
[169] Paul J, Christopoulos C and Thomas D W P 2001 Electron. Lett. 37 912
[170] Kolinko P and Smith D R 2003 Opt. Express 11 640
[171] Loschialpo P F, Smith D L, Forester D W, Rachford F J and Schelleng J 2003 Phys. Rev. E 67 025602
[172] K¨ ark¨ ainnen M K and Maslovski S A 2002 Microwave Opt. Tech. Lett. 37 4
[173] Cummer S A 2002 Appl. Phys. Lett. 82 1503
[174] Rao X S and Ong C K 2003 Phys. Rev. B 68 113103
[175] Rao X S and Ong C K 2003 Phys. Rev. E 68 067601
[176] Cummer S A 2003 Appl. Phys. Lett. 82 2008
[177] Pendry J B and Ramakrishna S A 2002 J. Phys.: Condens. Matter 14 8463
[178] Pendry J B 2003 Opt. Express 11 755
[179] Ramakrishna S A and Pendry J B 2004 Phys. Rev. B 69 115115
[180] Ramakrishna S A and Pendry J B 2003 Phys. Rev. B 67 201101
[181] Tretyakov S A 2001 Microwave Opt. Tech. Lett. 31 169
[182] Ramakrishna S A, Pendry J B, Wiltshire M C K and Stewart W J 2003 J. Mod. Opt. 50 1419
[183] Mansﬁeld S M and Kino G S 1990 Appl. Phys. Lett. 57 2615
[184] Smith D R and Schurig D 2003 Phys. Rev. Lett. 90 077405
[185] Smith D R, Schurig D, Mock J J, Kolinko P and Rye P 2004 Appl. Phys. Lett. 84 2244
[186] Zhang Z M and Fu C J 2002 Appl. Phys. Lett. 80 1097
[187] Li J, Zhou L, Chan C T and Sheng P 2003 Phys. Rev. Lett. 90 083901
[188] Jian H, Chen H, Li H, Zhang Y and Zhu S 2003 Appl. Phys. Lett. 83 5386
[189] Bria D, DjafariRouhani B, Akjouj A, Dobrzynski L, Vigneron J P, El Boudouti E H and Nougaoui A 2004
Phys. Rev. E 69 066613
[190] Nefedov I S and Tretyakov S A 2002 Phys. Rev. E 66 036611
[191] Feise M W, Shadrivov I S and Kivshar Y S 2004 Preprint condmat/0402087
[192] Ruppin R 2003 Microwave Opt. Technol. Lett. 38 494
[193] Shadrivov I S, Sukhorukov A A and Kivshar Y S 2003 Appl. Phys. Lett. 82 3820
[194] Ward A J and Pendry J B 1996 J. Mod. Opt. 43 773
[195] Notomi M 2002 Opt. Quant. Electron 34 133
[196] Smith D R, Vier D C, Padilla W, NematNasser S C and Schultz S 1999 Appl. Phys. Lett. 75 1425
[197] Ozbay E, Aydin K, Cubukcu E and Bayindir M 2003 IEEE Trans. Antennas Propag. 51 2592
[198] Marqu´ es R, Mesa F, Martel J and Medina F 2003 IEEE Trans. Antennas Propag. 51 2572
[199] Caloz C and Itoh T 2004 IEEE Microwave Wirel. Components Lett. 14 31
[200] Cubukcu E, Aydin K, Ozbay E, Foteinopoulou S and Soukoulis C M 2003 Nature 423 604
[201] Parimi P V, Lu W T, Vodo P and Sridhar S 2003 Nature 426 404
[202] Berrier A, Mulot M, Swillo M, Qiu M, Thylen L, Talneau A and Anand S 2004 Phys. Rev. Lett. 93 073902
[203] Yang S, Page J H, Liu Z, Cowan M L, Chan C T and Sheng P 2004 Phys. Rev. Lett. 93 024301
[204] Liu Z, Fang N, Yen TJ and Zhang X 2003 Appl. Phys. Lett. 83 5184
[205] Grbic A and Eleftheriades G V 2004 Phys. Rev. Lett. 92 117403
[206] Melville D O S, Blaikie R J and Wolf C R 2004 Appl. Phys. Lett. 84 4403
INSTITUTE OF PHYSICS PUBLISHING Rep. Prog. Phys. 68 (2005) 449–521
REPORTS ON PROGRESS IN PHYSICS doi:10.1088/00344885/68/2/R06
Physics of negative refractive index materials
S Anantha Ramakrishna Department of Physics, Indian Institute of Technology, Kanpur 208 016, India Email: sar@iitk.ac.in Received 3 September 2004 Published 18 January 2005 Online at stacks.iop.org/RoPP/68/449
Abstract In the past few years, new developments in structured electromagnetic materials have given rise to negative refractive index materials which have both negative dielectric permittivity and negative magnetic permeability in some frequency ranges. The idea of a negative refractive index opens up new conceptual frontiers in photonics. One muchdebated example is the concept of a perfect lens that enables imaging with subwavelength image resolution. Here we review the fundamental concepts and ideas of negative refractive index materials. First we present the ideas of structured materials or metamaterials that enable the design of new materials with a negative dielectric permittivity, negative magnetic permeability and negative refractive index. We discuss how a variety of resonance phenomena can be utilized to obtain these materials in various frequency ranges over the electromagnetic spectrum. The choice of the wavevector in negative refractive index materials and the issues of dispersion, causality and energy transport are analysed. Various issues of wave propagation including nonlinear effects and surface modes in negative refractive materials (NRMs) are discussed. In the latter part of the review, we discuss the concept of a perfect lens consisting of a slab of a NRM. This perfect lens can image the farﬁeld radiative components as well as the nearﬁeld evanescent components, and is not subject to the traditional diffraction limit. Different aspects of this lens such as the surface modes acting as the mechanism for the imaging of the evanescent waves, the limitations imposed by dissipation and dispersion in the negative refractive media, the generalization of this lens to optically complementary media and the possibility of magniﬁcation of the nearﬁeld images are discussed. Recent experimental developments verifying these ideas are brieﬂy covered. (Some ﬁgures in this article are in colour only in the electronic version)
00344885/05/020449+73$90.00
© 2005 IOP Publishing Ltd
Printed in the UK
449
450
S A Ramakrishna
Contents
1. Introduction 1.1. Objectives 1.2. Basic concepts and deﬁnitions 2. Materials with negative refractive index 2.1. Negative dielectric materials 2.1.1. Metals and plasmons at optical frequencies 2.1.2. Wiremesh structures as low frequency negative dielectrics 2.2. Materials with negative magnetic permeability 2.2.1. A stack of metal cylinders 2.2.2. The split ring resonator 2.2.3. Isotropic negative magnetic media 2.2.4. Wave dispersion in a resonant magnetic medium 2.2.5. The Swiss roll structure at radiofrequencies 2.2.6. Scaling to high frequencies 2.2.7. Other approaches 2.3. Negative refractive index materials 2.4. Homogenization and effective macroscopic material parameters 2.5. Causality and energy density in negative refractive media 3. Propagation of radiation in negative refractive media 3.1. Choice of the wavevector 3.2. Some effects of a reversed wavevector 3.2.1. The modiﬁed Snell’s law of refraction 3.2.2. The reversed Doppler shift 3.2.3. An obtuse angle cone for Cerenkov radiation 3.2.4. Reversed Goos–H¨ nchen shift a 3.3. Phase velocity, group velocity and energy ﬂow 3.4. Backward wave structures and transmission lines 3.5. Negative refraction effect in photonic crystals 3.6. Nonlinear effects in negative refractive media 3.7. Surface electromagnetic modes in NRMs 4. The perfect lens 4.1. Nearﬁeld information and the diffraction limit 4.2. Pendry’s proposal 4.2.1. The quasistatic limit and the silver lens 4.2.2. The role of surface plasmons 4.2.3. The asymmetric lens 4.3. Limitations in real materials and imperfect NRMs 4.4. Numerical simulations and time evolution 5. Designing superlenses 5.1. Overcoming the limitations of real materials 5.2. The generalized perfect lens theorem
Page 452 452 453 455 456 456 458 463 463 464 465 466 467 467 470 471 473 476 477 477 479 479 480 481 481 482 484 485 486 488 490 490 492 494 495 496 497 499 501 501 505
Demonstrating superlenses 7. Proof of the generalized perfect lens theorem References 506 507 508 510 512 512 513 514 514 516 516 517 .3. Experimental evidence for NRMs and superlenses 6. Conclusions Acknowledgments Appendix. A perfect twodimensional corner lens 6.3. Experiments on negative refractive index 6.3.2.3.Physics of negative refractive index materials 451 5. Cylindrical lenses 5.2.1.1.3.3. A spherical lens 5. The perfect lens in other geometries 5. Photonic crystals and the Veselago ﬂat lens 6.
the refractive index of a material is conventionally taken to be a measure of the ‘optical density’ and is deﬁned as n= c . Although Veselago went on to point out several interesting effects in NRMs. This ﬁeld has become a hot topic of scientiﬁc research and debate over the past four years. n = − εµ. or are at a semipopular level [13. such as a modiﬁed Snell’s law of refraction. 14]. In 1967. it will not be possible to review or cite all of them. From Maxwell’s equations the refractive index is given by the Maxwell relation. Given the exponentially rising number of publications. and unexpected situations arise with NRMs. . and as a good starting point for a newcomer to get into this area of research. Although many of the developments are only recent. placing the ﬁeld on a sound footing. given the rapid developments in this ﬁeld. 5]. Thus this will not be the ﬁrst review nor do we expect it to be the last. We will only concentrate here on the main developments. in such media. The striking demonstration by Pendry [6] that NRMs can be used to make perfect lenses with resolution capabilities not limited by the conventional diffraction limit has given an enormous boost to the interest in NRMs. and n could easily be taken as √ εµ without any problems. But the hope is that it will serve as a good reference point for the conceptual ideas involving negative refractive index materials and the problem of the perfect lens. Although it was realized that the refractive index would have to be a complex quantity to account for absorption and even a tensor to describe anisotropic materials. Veselago [1] ﬁrst considered the case of a medium that had both negative dielectric permittivity and negative magnetic permeability at a given frequency and concluded that the medium should then be √ considered to have a negative refractive index (i. Introduction In optics. n2 = εµ. However. At the time of writing the theoretical foundations have been laid and several experiments verifying the concepts have been performed. theoretical proposals [2.1. 3] for structured photonic media whose ε and µ could become negative in certain frequency ranges were developed experimentally [4. the negative square root. in the last few years. Shorter reviews have been published elsewhere [10–12] which cover certain aspects of the topic. had to be chosen). (1. v (1. ‘Standard’ notions and intuitions of electromagnetic theory can be misleading.2) where ε is the relative dielectric permittivity and µ is the relative magnetic permeability of the medium.452 S A Ramakrishna 1. and this has brought Veselago’s result into the limelight. Objectives The physics of NRMs can be extremely counterintuitive. Usual optical materials have a positive ε and µ. 1. It is imperative now to develop new intuitions and understanding to deal with these materials.e. his result remained an academic curiosity for a long time as real materials with simultaneously negative ε and µ were not available. the basic conceptual issues have now been widely debated and reasonably clariﬁed. Thus it appears to be the correct moment to review the conceptual issues of negative refractive index materials. a reversed Doppler shift and an obtuse angle for Cerenkov radiation. the question of the sign of the refractive index did not arise.1) where c is the speed of light in vacuum and v is the speed of an electromagnetic plane wave in the medium. This has mostly (for example the problems of the perfect lens) been responsible for the initial severe criticism and debate [7–9] as well as the later explosion of research in this area.
the response follows the applied ﬁeld at frequencies below the resonance. complex functions of the frequency. can be strictly distinguished from other structured photonic materials. This underdamped. ferrimagnetic and antiferromagnetic materials exhibit some magnetic activity at even frequencies of hundreds of gigahertz [20]. photonic crystals or photonic bandgap materials [16. All that survive the averaging in macroscopic measurements are the frequency components of the individual (atomic or molecular) oscillators driven by the external ﬁelds. These are macroscopic parameters because one usually only seeks timeaveraged and spatiallyaveraged responses averaged over sufﬁciently long times and sufﬁciently large spatial volumes. 15] and can be characterized by macroscopic parameters such as ε and µ that deﬁne their response to applied electromagnetic ﬁelds. i. and above the resonance the response is antiphased with respect to the applied ﬁeld. Metamaterials. but responds to the (atomically) macroscopic resonances of the structure. If now the resonance can be made sharp enough. in general. This becomes particularly visible when the applied ﬁelds have frequencies close to a resonant frequency of the material oscillators. This idea can now be extended to a higher class of inhomogeneous materials where the inhomogeneities are on lengthscales much smaller than a wavelength of the radiation but can be large compared with atomic or molecular lengthscales.Physics of negative refractive index materials 453 1. a restriction that the imaginary parts of the corresponding ε or µ are not to be negative.e. This is so that the total absorbed energy in a volume V of the medium. it will be possible to drive the real parts of the effective ε or µ negative. much like homogeneous materials. the magnetic response of most materials is limited to low microwave frequencies.2. and hence homogenization in this sense cannot be carried out. In the photonic crystals or bandgap materials the stops bands or bandgaps arise as a result of multiple Bragg scattering in a periodic array of dielectric scatterers. In fact. the periodicity of the structure here is of the order of the wavelength. overscreened response is responsible for the negative material parameters. Although most materials exhibiting a good electric response can be found at almost any frequency from radiofrequencies to the ultraviolet frequencies. the dielectric permittivity and the magnetic permeability are. we realize that dispersion and dissipation in thermodynamic media always accompany each other. ω)2 ] . 2π −∞ V is positive deﬁnite [19]. Magnetic polarization usually results from either unpaired electron spins or orbital electron currents. Some ferromagnetic. Basic concepts and deﬁnitions The dielectric constant (ε) and the magnetic permeability (µ) characterize the macroscopic response of a homogeneous medium to applied electric and magnetic ﬁelds. however. in some sense. ∞ dω (1. 17]. There is no fundamental objection to the real parts of the ε and µ being negative [19].3) d3 r ω[Im(ε(ω))E(r. Such materials have been termed metamaterials [13.e. As with any resonance. In media at thermodynamic equilibrium. In metamaterials the periodicity is by comparison far less important and all the properties mainly depend on the single scatterer resonances. but necessarily dispersive and dissipative. the real parts and the imaginary parts of the material parameters (ε(ω) and µ(ω)) are related to each other by the famous Kramers–Kronig relations [18]. In fact. But these are rare and usually have narrow bandwidths. there is. Thus. The radiation then does not resolve these individual mesostructures. But now the possibility . Noting that the imaginary parts of ε and µ relate directly to the absorption of electromagnetic radiation in the material. ω)2 + Im(µ(ω))H (r. This is because the polarizations in such media do not respond instantaneously to the applied ﬁelds but depend on the history of the applied ﬁelds. i. All causal materials are dispersive. and the collective excitations of these usually tend to occur at low frequencies. media with negative ε and µ are causal.
The behaviour of the waves in each of the quadrants is qualitatively different: materials that fall in the ﬁrst quadrant allow the usual righthanded electromagnetic propagating waves. . Such materials permit propagating waves with a reversed phase vector (k)1 compared with media with only one of Re(ε) or Re(µ) negative which do not allow any propagating modes (all waves decay evanescently in such media from the point of injection as k 2 < 0). A schematic showing the classiﬁcation of materials based on the dielectric and magnetic properties. we show the four quadrants in the Re(ε)–Re(µ) plane into which we can conveniently classify electromagnetic materials. S is the Poynting vector. Consider the Maxwell’s equation for a plane harmonic wave exp[i(k · r − ωt)]: k × E = ωµ0 µH. >0 S H k Electrical Plasma Evanescent decaying waves Many metals (UV .4) k × H = −ωε0 εE. although this does risk a confusion with the terminology of chiral optical materials. the Poynting vector (S = E × H) and the phase vector.Optical) Thin wires structures (GHz) S k E H Ordinary optical materials Right handed propagating waves > 0. <0 < 0. Using the deﬁnition that the wavevector k = nω/cn.6) c 1 This has prompted some authors to call these backward wave media [29] or negative phase velocity media [10]. and the axes set in quadrants 1 and 3 show the right. which is depicted schematically. H and k vectors. k. It is for this reason that such materials are also popularly termed lefthanded materials.5) where E and H are the electric and magnetic ﬁelds. 23. is special at high frequencies. 25] and the terahertz frequencies [27. (1. 24]. >0 > 0. Some authors have preferred to term such media double negative media. it appears that the ˆ ˆ refractive index in such media with Re(ε) < 0 and Re(µ) < 0 is also negative. are antiparallel. The wavy lines represent materials that allow propagating waves. This follows from the dispersion for the wavevector ω2 k · k = εµ 2 . H and k will now form a lefthanded triad. of artiﬁcially structuring materials at micro. the materials that fall in the second and fourth quadrants do not allow any propagating waves inside them (all electromagnetic radiation is evanescently damped in these media) and the NRMs that fall in the third quadrant allow lefthanded propagating waves inside them. 28] in between. the microwave frequencies [3.and nanoscales can enable us to generate a variety of metamaterials with magnetic activity at almost any frequency we wish—from a few hertz [21] to nearinfrared frequencies [22] including the radio frequencies [3. In ﬁgure 1. (1. 4. Thus even magnetic activity. If we take a medium with negative real parts of ε and µ with the imaginary parts being small (negligibly small for the time being) at some frequency (ω). <0 Negative refractive index Left handed propagating waves Artificial Meta–materials Magnetic plasma ? Structured materials Evanescent decaying waves Some natural magnetic materials (upto GHz) Figure 1. The waves in quadrants 2 and 4 decay evanescently inside the materials.454 E S A Ramakrishna < 0. Note that in the case of the NRMs. referring to the negative ε and µ [30].and lefthanded nature of E. let alone negative magnetic permeability. then we realize that the vectors E. (1. where n is the unit vector along E × H.
Although it is not immediately obvious. There are signiﬁcant problems of homogenization that are involved here. (1. which can display some of the effects of negative refraction [36] at wavelengths which are of the same order of the lengthscales of the structures in the metamaterials. ε(ω) = 1 − 2 ωp ω(ω + iγ ) .Physics of negative refractive index materials 455 Making Re(ε) or Re(µ) negative not only makes electromagnetic radiation decay exponentially inside them. Pendry’s seminal work on structured materials which display negative dielectric permittivity [2. we know that metals whose dielectric permittivity is dominated by the plasmalike response of the free electron gas. The review can be broadly split into two parts: sections 2 and 3. But ﬁnding a material that has these properties is a more difﬁcult task and it was because of the lack of a material that exhibited negative values of these parameters at a common frequency band that Veselago’s early result did not arouse much interest for a long time. These materials can support a host of resonant states localized at their surfaces. 2. 30] and much activity. as negative refractive index materials.7) where ωp is the bulk plasma frequency and γ is the damping constant. we will ﬁrst separately discuss the negative dielectric materials and the negative magnetic materials before we discuss the composite that behaves as an NRM. and sections 4 and 5. have negative dielectric permittivity at UV–optical frequencies. dealing with the speciﬁc problem of the perfect lens that can be formed from negative refractive index media. . In fact this has been one of the most fundamental new effects associated with NRMs thus far and has been a topic of intense debate [8. but also renders the materials rather very special. known as the surface plasmons [31. In general. giving rise to a wide variety of novel optical phenomena [33] and have given rise to the new ﬁeld of plasmonics [34]. Such a lens is not limited in its resolution to superwavelength lengthscales according to the standard principles of diffraction theory [35]. We will now proceed to examine all these concepts in detail. and materials with negative magnetic permeability can also be expected to support the analogous surface plasmons of a magnetic nature! These resonant states can affect the propagation of the electromagnetic radiation in a very profound manner and are deeply involved in the mechanism of the perfect lens [6] made of an NRM. we will strictly not consider photonic bandgap materials. Materials with negative refractive index There is nothing fundamentally objectionable about negative material parameters such as the dielectric permittivity and the magnetic permeability. The experimental efforts in this area are covered in section 6. has made it possible and interesting to create experimentally such materials at a variety of wavelengths. For example. This is a generic effect of having a negative dielectric permittivity. 43] and magnetic permeability [3]. In this section. We will then address the issues of homogenization and causality that confront us in the context of such materials. All the NRMs that have been developed today are obtained by putting together two structured materials that show separately a negative dielectric permittivity and negative magnetic permeability. This issue will be treated in detail in the latter sections of this review. 32]. dealing with the general properties of negative refractive index media. the resulting composite structure possesses a negative refractive index. along with the technological capability to make structures at microscopic lengthscales. These surface plasmons can resonantly interact with radiation on structured surfaces.
456 S A Ramakrishna 2.1.5) + 2 ky − = ε± µω /c are good solutions to the Maxwell equations when κz+ κz− + = 0. ε+ ε− 2 κz± 2 2 (2. ∀z < 0. Hence ωp is known as the plasmon frequency. As mentioned before. where ωp = ne2 /ε0 m is the plasma frequency. rendering the dielectric constant negative makes it possible for the surface to support resonances called the surface plasmons [32]. and above the plasma frequency.2) where n is the number density of the conduction electrons and each electron is assumed to contribute independently to the polarization. This yields a polarization per volume in the medium of P = (ε − 1)ε0 E = −ner = − ne2 /mE .6) These are collective excitations of electrons with the charges displaced parallel to the (real part of the) wavevector. so that it effectively sees a spatially constant ﬁeld and the velocities involved are sufﬁciently low so that the magnetic ﬁeld can be neglected. Substituting the expression for the dielectric constant into Maxwell’s equation.7). Metals and plasmons at optical frequencies. This is the theory of Drude and Lorentz for the dispersion characteristics of a plasma [18. r (2. we obtain the dispersion for light in a metal: 2 k 2 c 2 + ωp = ω 2 . the waves below the plasma frequency correspond to the negative energy solutions. Consider the equation of motion for an electron in an applied time harmonic electromagnetic ﬁeld with angular frequency ω0 .4) (2. Consider the interface (z = 0 plane) between a metal with a dielectric constant ε− and a positive dielectric medium (vacuum) with dielectric permittivity ε+ (=+1) as shown in ﬁgure 2. it appears as if the transverse modes of light had a ﬁnite rest mass of m0 = hωp /c2 . Above ωp the medium behaves as an ordinary positive dielectric. (2. It is assumed that the wavelength of light is large compared with the distance travelled by the electron. If the electron gas is displaced by a small distance in a ﬁnite metal. ¯ The plasma frequency also has a physical manifestation as a collective excitation of the electron gas. From this we obtain the relative dielectric constant of equation (1.3) Thus. The dielectric permittivity is negative up to ωp and the plasma shields the interior from electromagnetic radiation. (2. ω(ω + iγ ) (2. At optical frequencies many metals have a negative dielectric permittivity when the conduction electrons in the metals can be assumed to be reasonably free in a background of static positive ion cores. the overall system being charge neutral. Now the ﬁelds for a ppolarized light are E = E0 exp[i(kx x + ky y − ωt) − κz+ z] E = E0 exp[i(kx x + ky y − ωt) + κz− z] where 2 kx ∀z > 0. This is the dispersionless plasmon mode that occurs at ω = ωp . It is this plasmalike behaviour that is responsible for a negative dielectric permittivity at frequencies less than the plasma frequency. Negative dielectric materials 2. 35]. the excess negative (electronic) charge and the positive (background) charge that accumulate at the edges provide a restoring force proportional to the displacement and cause the entire electron gas to oscillate at frequency ωp [37]. Thus we have a mode with longitudinal components whose ﬁelds .1) where m is the mass of the electron and mγ is a phenomenogical damping (viscous) force on the electron due to all inelastic processes.1.1. ˙ m¨ + mγ r = −eE exp(−iω0 t).
We know that a capacitor can be formed by two parallel conducting plates with an insulating dielectric placed in between.Physics of negative refractive index materials 457 x y z + ++ __ ++ __ – –kx kx __ Figure 2. which is the same as being an inductor. Equation (2. A capacitor and an inductor form a resonant circuit that can oscillate at ω0 = 1/ LC. Complicated surface structures can generate a variety of surface plasmon dispersions. and a coupling mechanism will have to be provided by mechanisms such as the surface roughness.6) yields the dispersion kx = ω ε+ ε− c ε+ + ε − 1/2 .7) and setting ε+ = 1 for vacuum. A schematic showing the surface plasmon on an interface between a negative and a positive dielectric media and the associated charge density ﬂuctuations. Filling up the gap with a negative dielectric material instead would now lead to a capacitor with negative capacitance. decay exponentially into the metal and the dielectric on either side of the interface. This remains true down to a nanometre or so when nonlocal dispersive effects kick in due to inadequate electronic screening inside the metal. It is as if in a structured .7) For a metal with ε− (ω) given by equation (1. Thus it will not be possible to excite the surface plasmon on a perfectly ﬂat surface using propagating modes of light. we note that kx > ω/c for the surface plasmon (see ﬁgure 22). A capacitor ﬁlled with a negative dielectric has negative capacitance. can become resonant (see ﬁgure 3). (2. Pendry [33] provides a very insightful explanation for this. This is the reason why the interaction of the radiation with the resonant surface plasmons is very rich and difﬁcult to model: the radiation couples with the metal surface at all lengthscales. The exponential decay of the ﬁelds normal to the surface and the propagating nature along the surface are depicted schematically. The negative dielectric constant is central to the nature of resonant interactions of structured metal surfaces with radiation. Next note that for reasonably large wavevectors the surface √ plasmons are nearly degenerate with a frequency of ωp / 2. This charge density wave thus lives on the surface of the metal and is called a surface plasmon. and this has given rise to new optical phenomena such as the extraordinary transmission of light through subwavelength sized holes [38]. acts as an inductor and can resonate with another usual capacitor. Thus two capacitors in a circuit. – L C C √ Figure 3. one ﬁlled with a positive dielectric and the other ﬁlled with a negative dielectric material. a grating structure or a dielectric coupler to vacuum such as a hemisphere or a prism.
Wiremesh structures as low frequency negative dielectrics.458 S A Ramakrishna metallic structure the metal pieces form the capacitors ﬁlled by the negative dielectric material and the holes in between form the capacitors ﬁlled with the positive dielectric material. Including this term in equation (2. The idea is to obtain negative permittivity materials at low frequencies by literally doping the vacuum with metal. However. ε =1+ 2 ω0 ne2 /(ε0 m) . while it is not clear that this holds for the structure of Sievenpiper. 2. and it is hard to make it penetrate well into the bulk of the structure for the appearance of a threedimensional effective medium to hold true in many cases. within a frequency band above a resonance frequency. composites of randomly oriented long conducting ﬁbres have been known to exhibit very high values of permittivity even at low concentrations [42]. we immediately obtain the Drude form for ε for a good plasma.43] and Sievenpiper et al [44] independently demonstrated that metallic wiremesh structures have a low frequency stop band from zero frequency up to a cutoff frequency which they attributed to the motion of electrons in the metal wires. For many metals. and we have trouble when we try to extend this behaviour to lower frequencies where ω ∼ γ . Note now that the dielectric constant can be negative in a small frequency range above ω0 if the resonance is sharp enough. Pendry et al [2. However. Setting ω0 = 0. The electrons are conﬁned to move within the wires only. The resulting system is a very complicated circuit with many L–C resonances. It must be pointed out that negative dielectric permittivity can also be obtained in more ordinary dielectric media with bound charges. 40] to drive the dielectric constant negative over selected small frequency ranges. − ω2 − iγ ω (2. is structured on truly subwavelength lengthscales and can be effectively homogenized. For small displacements of the electrons there 2 is a restoring force on the electron. In fact. The metamaterial of Pendry et al. Consider a medium in which the electrons are bound to positive nuclei and its response to an applied electromagnetic ﬁeld. one obtains the resonant Lorentz dielectric permittivity. Consider an array of inﬁnitely long.8) where n is now the total density of bound electrons (here it is assumed simplistically that all the electrons are bound with the same strength). which has the ﬁrst effect of reducing the effective electron density as the radiation cannot sense the individual wire structure but only the average .1. the dissipation in most metals is large. the radiation usually only probes the end surfaces of the metallic structures. Thus we have many examples of materials with a negative dielectric constant at optical frequencies. parallel and very thin metallic wires of radius r placed periodically at a distance a in a square lattice with a r as shown in ﬁgure 4. The low frequency stop band can be attributed to an effective negative dielectric permittivity and provides us with a way of obtaining negative dielectrics at even microwave frequencies. Metaldielectric composites have long been studied for their rich electrodynamic response (see [41] and references therein). The electric ﬁeld is considered to be applied parallel to the wires (along the Z axis) and we will work in the quasistatic limit when the wavelength of the radiation λ a r. the plasma frequency is at ultraviolet frequencies and γ is small compared with ωp (see [32]). −mω0 r. one can make use of resonant effects such as electromagnetically induced transparency or similar effects based on atomic coherences [39. Effective medium theories have been developed to describe these systems when the wavelength of the incident radiation is much larger than the intrinsic lengthscales of the structure.2. The dissipation dominates all phenomena and we no longer have the behaviour of a good plasma.1). consisting of very thin wires. For example. Then it can hardly be claimed that the dielectric constant is an almost real and negative number.
In our case. Noting that r a . The effective electron density is immediately seen to be neff = πr 2 n. We can assume that the vector potential associated with a single wire is A(ρ) = a2 zI ˆ ln 2π 4ρ(a − ρ) a =0 ∀ρ > . charge density. 2 ∀ρ < a . it appears as if the charge carriers. and it is not easy to change the currents ﬂowing in these wires.12) This choice avoids the vector potential of one wire overlapping with another wire and thus the mutual induction between the adjacent wires is addressed to some extent. have acquired a tremendously large mass.10) The vector potential associated with the ﬁeld of a single inﬁnitely long currentcarrying conductor is nonunique unless the boundary conditions are speciﬁed at deﬁnite points. On an average we can assume a uniform D ﬁeld within the unit cell. There is a second equally important effect to be considered. An array of inﬁnitely long thin metal wires of radius r and a lattice period of a behaves as a low frequency plasma for the electric ﬁeld oriented along the wires. Thus. namely the electrons. leading to a nonzero nonuniform magnetic ﬁeld that is large close to the wires which contributes to most of the ﬂux.Physics of negative refractive index materials 459 a Z a r X Y Figure 4. 2 (2. a2 (2. But the current density is not uniform.9) where n is the actual density of conduction electrons in the metal. we have a periodic medium that sets a critical length of a/2.11) (2. (2. By symmetry there is a point of zero ﬁeld in between the wires and hence we can estimate the magnetic ﬁeld along the line between two wires as H(ρ) = ˆ φI 2π 1 1 − ρ a−ρ . consider the magnetic ﬁeld at a distance ρ from a wire. To see this. The thin wires have a large inductance.
a = 10 mm and aluminium wires (n = 1029 m−3 ). It only depends on the radius of the wires and the spacing. We note that the canonical momentum of an electron in an electromagnetic ﬁeld is p + eA. 2π r (2. Typically one can choose r = 1 µm. Thus. assuming a longitudinal plasmonic mode for the system.460 S A Ramakrishna by about three orders of magnitude in our model and that the current I = π r 2 nev. where v is the mean electron velocity.19) (2.15) µ0 π 2 r 4 n 2 e 2 v a ln = meff π r 2 nv 2π r (2. This approach has been taken [45.20) . L. Consider the current induced by the electric ﬁeld along the wires. by calculating the magnetic ﬂux per unit length passing through a plane between the wire and the point of symmetry between itself and the next wire where the ﬁeld is zero: φ = µ0 r a/2 H (ρ) dρ = µ0 I a2 ln . The actual deviations from this expression are much smaller than in our case. noting that the polarization per unit volume in the homogenized medium is P = −neff er = Ez neff ev =− 2 2 .18) where neff = πr 2 n/a 2 as before.16) is independent of the microscopic quantities such as the electron density and the mean drift velocity. ˆ (2. and the lattice actually has fourfold symmetry.13) This is a very good approximation in the mean ﬁeld limit. related by the total inductance (self and mutual) per unit length (L): Ez = +iωLI = iωLπ r 2 nev. We note that a reduced effective electron density and a tremendously increased effective electronic mass would immediately reduce the plasmon frequency for this system. and a plasma frequency of about 2 GHz! Thus. 46] and we present it below for the sake of completeness. we can write the vector potential as A(ρ) = a µ0 πr 2 nev ln 2π ρ z.14) for the electron. suggesting that the entire problem can be recast in terms of the capacitances and inductances of the problem. iω ω a L (2. We can estimate the inductance. we have succeeded in obtaining a negative dielectric material at microwave frequencies.17) that is almost 15 times that of a proton. We have considered only two wires. Thus assuming that the electrons ﬂow on the surface of the wire (assuming a perfect conductor). 2π 4r(a − r) (2.16) for the plasmon frequency. Note that the ﬁnal expression for the plasma frequency in equation (2. (2. which gives an effective mass of meff = 2.67 × 10−26 kg. we can associate a momentum per unit length of the wire of p = πr 2 neA(r) = and thus an effective mass of meff = µ0 πr 2 ne2 a ln . we have ωp = neff e2 2πc2 = 2 ε0 meff a ln(a/r) (2.
we obtain in the limit r a. Thus the waves only sense the wires parallel to the electric ﬁeld and correspondingly have a longitudinal mode. The effects of the connectivity of the wires along different directions at the edges of the unit cell have also been examined [43]. we lose the physical interpretation of a low frequency plasmonic excitation here.65 × 107 −1 m−1 that yields a values of γ = 0. the conductivity is σ = 3. the effective medium appears to be isotropic as the radiation fails to resolve the underlying cubic symmetry yielding a truly threedimensional low frequency plasma. It is simple to add the effects of ﬁnite conductivity in the wires which had been neglected in the above discussion. Noting = LI . Thus the low frequency plasmon is sufﬁciently stable against absorption to be observable. This issue of spatial dispersion has been studied more recently by Belov et al [46]. Mikhailov [47] assumed the results for a onedimensional plasmon polarition on a metal wire and neglected the essential . The electric ﬁeld and the current would now be related by Ez = iωLI + σ πr 2 I. For very thin wires. A threedimensional lattice of thin conducting wires behaves like an isotropic low frequency plasma. where ε is the effective permittivity.21) ω a ln(a/r) which is identical to the relation obtained in the plasmon picture. This makes the medium anisotropic with negative ε only for waves with the electric ﬁeld along the zdirection. and the polarization P = (ε − 1)ε0 Ez . we considered only wires pointing in the zdirection.23) 2 ω(ω + i(ε0 a 2 ωp /π r 2 σ )) Thus the ﬁnite conductivity of the wires contributes to the dissipation. In the limit of large wavelengths. but had strong spatial dispersion for the transverse modes above the plasma frequency. the polarization in the direction orthogonal to the wires is small and can be neglected. showing up in the imaginary part of ε. However. (2. Some objections were raised to Pendry’s original propositions.1ωp which is comparable to the values in real metals [32]. which modiﬁes the expression for the dielectric permittivity to 2 ωp ε =1− . In our above discussion. (2.22) where σ is the conductivity. (2. 2πc2 ε(ω) = 1 − 2 2 .Physics of negative refractive index materials 461 Z Y X Figure 5. A wire mesh with nonintersecting wires was also shown to have a negative ε at low frequencies below the plasma frequency. For aluminium. The medium can be made to have a reasonably isotropic response by considering a lattice of wires oriented along the three orthogonal directions as shown in ﬁgure 5.
(2. Further. the speed of light in vacuum. This response for periodic ﬁnite wire strips has also been considered by Koschny et al in [52]. The effects of having continuous metal wires have been studied in [43]. They also opined that the interpretation of a heavy electronic effective mass was due to the gauge dependent vector potential which could be changed at will. Thus no homogenization could be carried out. Low frequency transverse modes appear well below the effective plasma frequency for the corresponding continuous wire structure. however. A transmission line model [57] gives a result of ε= ε0 π cos−1 cos(ka) + sin(ka) (ka)2 ka ln(a/(2π r)) 2 . where it was also shown that the corresponding magnetic permeability. and it was shown that the system has a low frequency cutoff that is much smaller than that of the bulk superconductor. their results completely neglect the fact that there are two lengthscales of the problem and that the wires are very thin compared with the wavelength. The photonic response of superconducting cylinders made of high Tc materials has been considered [59]. one can again obtain a negative dielectric permittivity within a frequency band above the resonant frequency. However. But they were usually considered when the wavelength was comparable with the period of the lattice and the radius of the wires. Makhnovskiy and Panina [53] have considered the possibility of using the magnetoimpedance effect to change the surface impedance of thin ferromagnetic wires using an applied magnetic ﬁeld to generate negative dielectric media whose plasma frequency can be tuned by the magnetic ﬁeld. 56]. Dense wire media have been of interest in the electrical engineering community for a long time for artiﬁcial impedance surfaces [55. The use of the Coulomb gauge does spoil the manifest covariance. Pitarke et al [58] have considered the photonic response of metallic cylinders embedded in a dielectric host and have agreement with the classical Maxwell–Garnet result when the wavelength is at least twice as large as the spacing between the cylinders. but propagating modes appeared above them. The additional capacitance gives rise to a L–C resonance at a ﬁnite frequency. The additional capacitance between the edges of the wires and the selfcapacitance of the edges can be considered easily in our model as a term additional to the impedance in equation (2. showed an antiresonant dispersion with a negative imaginary part of µ when the real part of ε was negative. and one has a resonant Lorentz form of the dispersion for the dielectric permittivity. these cases do not ideally fall in the category of effective media. with a ﬁnite frequency band for negative permittivity. but insofaras the velocities of the charges involved are small compared with c. We will discuss this latter aspect later. The role of a plasmon in a superconductor as . Again the response of short wires gave rise to a resonant effective permittivity. where it was shown that the response of wires systems with periodic sections of the wires removed was more akin to that of periodically placed interacting dipoles [51]. However. For a sufﬁciently sharp resonance and low wire resistivity. it is the most suitable gauge. Walser et al [49] incorrectly assumed the Drude form of the dielectric constant of the metal for the dielectric response of a single metallic wire as well and claimed that dissipation would damp out all phenomena. Simovsky and He [54] considered the response of an array of shaped particles and found that the array showed the resonant dielectric response of periodically cut wires. The scattering turns out to be quite important at higher frequencies.462 S A Ramakrishna threedimensional nature of the problem [48].24) which was shown to agree with our expression for a low frequency plasmonic medium in the limit of very thin wires [45]. the claim that any gauge term could be added to the vector potential without affecting the problem is incorrect as that would change the electric ﬁeld and the plasmon would appear to have a position dependent mass [50].22).
In the case of nanometallic structures. Landau and Lifschitz [19] give a very general argument as to how the magnetic activity arising from atomic orbital currents should be negligible at optical frequencies if one could neglect the polarization currents. (2.2. So it is indeed a challenge even to obtain magnetic activity.Physics of negative refractive index materials z y x 463 + − − − ++++ − − H r − − + − +++ d (b) r d (a) (c) − −− − − ++ ++ + + Figure 6. Pendry [2. A stack of metal cylinders. a massive Higgs boson has been stressed by Anderson [60]. the magnetic moments of induced real and displacement current distributions can actually contribute to an effective magnetization if the electric polarizability of the corresponding medium is small. the frequency of the plasmon could well be smaller than the superconducting gap. (a) An applied magnetic ﬁeld along the axes of a stack of conducting cylinders induces circumferential currents that shield the interior.2. making such media of great interest in a fundamental manner. (b) The splitring structure: the capacitance across the rings now causes the structure to be resonant. the magnetic activity in most materials tends to tail off at high frequencies of even a few gigahertz. The cylinders have a radius r and are placed in a square lattice with period a. let alone negative magnetic permeability. at microwave frequencies and beyond.25) a where H0 is the applied magnetic ﬁeld and j is the induced current per unit length of the cylinder. Most materials have a natural tendency to be diamagnetic as a consequence of Lenz’s law. 2.26) a . The emf around the cylinder can be calculated from Lenz’s law and is balanced by the Ohmic drop in potential: πr 2 iωµ0 πr 2 H0 + j − 2 j = 2π rρj. Materials with negative magnetic permeability As mentioned before. (2. 43] has pointed out the importance of the fact that in such thin wire superconductor structures. The third term is due to the depolarizing ﬁeld which is assumed to be uniform as the cylinders are inﬁnitely long. The resulting effective medium is diamagnetic. Consider the response of a stack of metallic cylinders to an incident electromagnetic wave with the magnetic ﬁeld along the axis of the cylinders as shown in ﬁgure 6. (c) The Swiss roll structure: more capacitance for lower frequency operation. 2. In fact.1. The axial magnetic ﬁeld inside the cylinders is πr 2 H = H0 + j − 2 j. The oscillating magnetic ﬁeld along the cylinders induces circumferential surface currents which tend to generate a magnetization opposing the applied ﬁeld.
The frequencies are assumed to be sufﬁciently low so as to have only a small skin effect. This has become well known subsequently as the split ring resonator—SRR for short. By introducing capacitive elements into the system. the induced currents made it appear as if magnetic monopoles were ﬂowing up and down the cylinders. 25–27. The SRRs are the basis of most of the metamaterials exhibiting negative magnetic permeability today [4. and a diamagnetic effective medium is also obtained with percolation metallodielectric composites [41]. The induced currents ﬂow in the directions indicated in ﬁgure 6.27) (2. Consider an array of concentric cylindrical metallic shells with a gap in them as shown in ﬁgure 6(b). over a line along the edge of the unit cell. B. a rich resonant response can be induced [3]. 62–64]. =1− µ0 Heff 1 + i2ρ/(µ0 ωr) (2. Assuming that the gap (d) is very small compared with the radius (r) and that the capacitance due to the large gaps in any single ring is negligible. The large gap in each ring prevents the current from ﬂowing around in a single ring.29) Thus. we balance the emf around the circuit as −iωµ0 πr 2 H0 + j − πr 2 j a2 = 2π rρj − j . Proceeding as before. 5. The SRR works on the principle that the magnetic ﬁeld of the electromagnetic radiation can drive a resonant L–C circuit through the inductance.32) . while the averaged H ﬁeld outside the cylinders is πr 2 j.464 S A Ramakrishna where time harmonic ﬁelds are assumed and ρ is the resistance per unit length of the cylinder surface. This diamagnetic screening effect has been known for some time for superconducting cylindrical shells [61]. we obtain the effective relative magnetic permeability: Heff = H0 − µeff = Beff π r 2 /a 2 . 2. The averaged magnetic ﬁeld is Beff = µ0 H0 . the real part of µeff is always less than 1 (diamagnetic) and greater than 0 here. over the area of the unit cell while averaging the magnetic ﬁeld. In the case of the cylinders.28) (2. The split ring resonator. Now we homogenize the system of cylinders by adopting an averaging procedure (discussed in section 2.2.31) and thus we have a resonant form of the permittivity with a resonant frequency of ω0 = (2. H . and this results in a dispersive effective magnetic permeability.2.30) where the effective capacitance C = ε0 επ r/(3d) and ε is the relative dielectric permittivity of the material in the gap. with charges accumulating at the gaps in the rings. a2 Using the above. But the problem was that the system only had an inductive response (the monopoles equivalently had no inertia). and the circuit is completed across the small capacitive gap between the two rings.4) that consists of averaging the magnetic induction. we obtain the effective permeability as µeff = 1 − πr 2 /a 2 f ω2 =1+ 2 1 − (3d/µ0 ε0 επ 2 ω2 r 3 ) + i(2ρ/µ0 ωr) ω0 − ω2 − i ω 3d µ0 ε0 επ 2 r 3 1/2 (2. iωC (2. It is assumed that the potential varies linearly with the azimuthal angle around the ring.
As the resistance increases. and a ‘magnetic plasma frequency’ ωm = 7. d = 0.5 mm.2 mm and the magnetic ﬁeld along the axis for different values of ρ.0 Ω/m 8 0 Im( ) 4 5 4 5 7 6 Frequency (GHz) 8 9 0 4 5 7 8 6 Frequency (GHz) 9 Figure 7. d. Let us also note that µeff can attain very large values on the low frequency side of the resonance. we have a resonance frequency of ω0 = 6. ε. broadens the peak. and it is seen that the ﬁlling fraction plays a fundamental role in the bandwidth over which µ < 0. and in the case of very resistive materials the resonance is so highly damped that the region of negative µ can disappear altogether as shown in ﬁgure 7. a threedimensional photonic material was proposed [3] which would have a reasonably isotropic response. To overcome this limitation. The resonance frequency and the negative µ band can be varied by changing the inductance (area) of the loop and the capacitance (the gap width. This is due to the assumption of a perfect conductor in our analyses. The dielectric permittivity of the embedding medium. A ﬁnite resistivity.Physics of negative refractive index materials ρ = 0. which is also borne out in the numerical calculations of [65].33) (1 − f )µ0 ε0 επ 2 r 3 assuming the resistivity of the material is negligible.56 GHz.2 Ω/m ρ = 10 Ω/m ρ = 1. Assuming that the magnetic ﬂux through the stacked split rings remains the same as for a cylinder. that arises from the L–C resonance of the system. Thus.0 Ω/m 5 Re( ) ρ = 10 Ω/m ρ = 1.41 < ω < 7. or the dielectric permittivity. and the effective magnetic permeability has been calculated as π r 2 /a 2 µeff (ω) = . For frequencies larger than ω0 . can be used to tune the resonant frequency.2.34) 2 ω2 r 3 C) + i(2lρ/µ ωr) 1 − (3l/µ0 ε0 π 0 . For typical sizes of r = 1. (2. We note that µeff 1 − πr 2 /a 2 asymptotically at large frequencies.56 GHz.3. the effective medium of SRR is going to have a √ very large surface impedance Z = µeff /εeff . rendering the medium uniaxial. 68]. The system has negative magnetic permeability for 6.5 mm. ε. d = 0. the response of the system tends to tail off.41 GHz. of the material in the capacitive gap) of the system. (2. The behaviour of (a) the real and (b) the imaginary parts of the effective magnetic permeability for a system of split cylinders with r = 1. the response is out of phase with the driving magnetic ﬁeld and µeff is negative up to the ‘magnetic plasma’ frequency of 1/2 3d . this system would have the same magnetic response in all the orthogonal directions. Such large impedances have also been obtained by capacitively loaded structured surfaces [67. The dispersion in µeff is shown in ﬁgure 7. in general. Obviously the system does not show any magnetic activity when the magnetic ﬁeld is oriented perpendicular to the cylinders. Isotropic negative magnetic media.2 Ω/m 465 12 ρ = 0. we assumed a medium of inﬁnitely long split cylinders which shows magnetic properties when the magnetic ﬁeld is along the axes of the cylinders. In our discussion till now. ωm = 2. The essential building blocks of this system are shown in ﬁgure 8(a) which consists of interleaved orthogonal planes of planar SRRs.2 mm. a = 5 mm. The factor f = π r 2 /a 2 is the ﬁlling fraction of the material.
of the ring itself. In the limit of an effective medium. Interestingly the system should also be able to sustain the magnetic equivalent of surface plasmon modes at an interface with vacuum (say) when Re(µ) = −1 just like for the usual surface plasmons at an interface between positive and negative dielectric media. 2.2. Three orthogonal nonintersecting SRRs is a possible candidate as an isotropic scatterer. which is not satisﬁed in the actual medium. and ω. One also notes that the effective medium can support a longitudinal bulk magnetic plasma mode at ωm . Wave dispersion in a resonant magnetic medium. Given the generic form of the dispersion in a SRR medium (or any of the resonant magnetic media to be discussed below). Balmaz and Martin discuss the various orientations of the rings and conclude that the rings can only intersect along symmetric points for an isotropic response.466 S A Ramakrishna Figure 8. One can see that the medium allows two degenerate waves (polarizations) with a gap for frequencies ω0 < ω < ωm . the response of this medium would be very isotropic. An array of twodimensional isotropic scatterers with random orientations can be expected to behave reasonably as an effective medium with isotropic response. the frequency. (b) The threedimensional SRR scatterer proposed in [63] (reprinted with permission from [63]. the stacking distance. This has an isotropic response for any wave incident on it normal to the plane of the rings. . Obviously if the rings are very thin. the wavevector. for a wave. where C = ε0 /π ln(2τc /d) is the capacitance between two parallel sections of planar rings of thickness τc . We show schematically this generic relationship in ﬁgure 9. but it is not otherwise expected to give rise to large differences in the functioning. where ω0 is the resonance frequency and ωm is the magnetic plasma frequency. r. This also gives rise to the idea that a bandgap (say in a photonic crystal) can be ascribed to negative µ just as one can ascribe it to an imaginary refractive index or negative ε [69]. Balmaz and Martin [63] have introduced a novel magnetic scatterer which consists of two orthogonal intersecting SRRs as shown in ﬁgure 8(b). These will couple to spolarized light with a nonzero component of the magnetic ﬁeld normal to the surface. the origin of the gap would depend on the symmetries of the ﬁelds in question.4. l. but unfortunately the SRRs then tend to have nondegenerate resonant frequencies or different Q factors. This is the region where the magnetic permeability is negative and all waves in the medium are evanescent. of the rings have to be much smaller than the radius. Copyright 2002 American Institute of Physics). The induced currents in the system give the appearance of magnetic monopoles sloshing up and down the cylinders. For this assumption of no ﬂux leakage to hold. (a) The unit cell of a threedimensional SRR medium with isotropic magnetic response. one can obtain the dispersion relation between k. Hence the actual resonant frequency when the ﬂux leakage is accounted for (which is nontrivial) can be expected to differ somewhat from this expression. the capacitance across the ends of the single rings (which we neglect) can become comparably large. Of course.
The frequency and wavevector are scaled with respect to the resonance frequency. For lower radiofrequencies. This system also has the same generic resonance form of the permeability with frequency as the SRR structures. Terahartz operation of SRRs has been demonstrated experimentally [27]. The dispersion for a wave in an SRR medium. 2. ω0 .2.35) 1 − (dc2 /2π 2 r 3 (N − 1)ω2 ) + i(2ρ/µ0 ωr(N − 1)) where N is the number of coils in the structure and it is assumed that the total thickness of the wound layers N d r the cylinder radius. Maxwell’s equations appear to suggest that one can scale the phenomena to higher frequencies by simply scaling down the corresponding lengthscales. the effective magnetic permeability for a system of such structures can be calculated as π r 2 /a 2 µeff = 1 − . As before. This has seen application in magnetic resonance imaging (MRI) at radiofrequencies as magnetic ﬂux tubes in the region where the effective magnetic permeability assumes very large values on the lower frequency side of the resonance [23.5. This structure has also been popularly called the Swiss roll structure. and k0 = ω0 /c. This can be achieved by rolling up a metal sheet in the form of a cylinder with each coil separated by an insulator of thickness d as shown in ﬁgure 6(c). the capacitance of the system can be made large instead of increasing the size of the loop (inductance). Scaling to high frequencies. . However. 2. Keeping this in mind. 24]. The current loop is now completed through the differential capacitance across the space between the metal sheets as shown. (2.6.2. This is both convenient and increases the validity of our assumption of an effective medium as well. but now the resonance frequency occurs at a much smaller frequency owing to the larger capacitance of the structure. a single ring with two symmetric splits providing a capacitive gap and suitable for operation in the terahertz region was proposed [28] and is shown in ﬁgure 10(a). The Swiss roll structure at radiofrequencies. Propagation is not possible in the frequency gap where µ < 0 as shown.Physics of negative refractive index materials 467 Figure 9. the main problem in scaling to higher infrared and optical frequencies is that metals no longer behave as perfect conductors and electromagnetic ﬁelds penetrate considerably into the metal. This means that the dispersive nature of the metals must be taken into account for scaling to higher frequencies. Another cause of concern is the technological ability to make the resonant structures necessary at the small lengthscales.
7. Using the method of Smith et al [72]. that the thickness. The magnetic ﬁeld is assumed to be along the axis of the split cylinders. dc = 24 nm and Lc = 144 nm used for the calculations. τc . b = 312 nm. (c) The real and imaginary parts of µ and (d) the real and imaginary part of ε. while a simple circuit theory predicts a resonance frequency of 104 THz for the L–C resonance.468 1 Normalized wavevector S A Ramakrishna (b) 0. was used with the empirical values [71] of hωp = 9. although surprisingly Im(ε) < 0 near the resonance and Re(ε) disperse the other way around. Note the ¯ presence of a frequency gap that arises due to the negative effective µ of the structure at about 75–80 THz. All the data are taken from [28]. (a) The single ring with a double split with a = 600 nm. The bandgap at about 75 THz arises due to a negative µ. of the metal shells is smaller than the skin depth (∼20 nm at 100 THz).5 a b Lc c 0 0. Thus the potential drop across each half of the ring is Vr = 0 π Eφ r dφ = πrjφ −iωε0 εm (2. i. Clearly the assumptions of a perfect conductor cannot be made at these high frequencies. τc = 24 nm. the complex transmission and reﬂection coefﬁcients. (b) The photonic band structure of this system of columns for the spolarization (with H along the cylindrical axes). The light line (· · · · · ·) and the bands for silver columns with the dame ﬁlling fraction are shown for comparison. Assuming a strong skin effect. Clearly there is a frequency band with Re(µ) < 0. was also found by Koschny et al [52].36) ( ) 2 . we can write the displacement current per unit length jφ = −iωε0 εm Eφ . viz.5 light line Silver rods SRR columns dc (a) 1 0 30 90 120 60 Frequency (THz) Re( ) Im( ) 150 3 6 (c) 4 ( ) 2 1 0 1 (d) 0 2 50 60 70 80 90 Frequency (THz) 100 Re( ) Im( ) 2 50 60 70 80 90 Frequency (THz) 100 Figure 10. and are shown in ﬁgure 10. The band structure and the transmission for inﬁnitely long split cylinders of silver obtained using a photonic bandstructure calculation using the transfer matrix method [70] is shown in ﬁgure 10(b). analogous to the case of the cutwire system discussed before. Let us analyse this system to gain an insight into the high frequency scaling properties.018 eV and ε∞ = 5. 2 A plasma form for the dielectric function of silver.e. This behaviour. ε(ω) = ε∞ − ωp /[ω(ω + iγ )]. who call it antiresonant behaviour. and seems to be generic to metamaterials with one negative material parameter. γ = 0. the effective ε and µ were recovered from the emergent quantities.013 eV.
. yielding the relation that Hext µ 0 ε0 ω 2 π r 2 τc =1− . implying an inductance that is proportional to the electronic mass. f = f. we see that the additional inductance is entirely due to the electronic mass and Li can hence be termed the inertial inductance. we obtain the effective magnetic permeability as µeff = Beff f ε 0 µ0 ω 2 π r 2 =1+ . Lg = µ0 π r 2 is the 2 geometrical inductance and Li = 2π r/(ε0 τc ωp ) is an additional inductance that shows up. where E is the applied ﬁeld. This is due to the fact that the proportion of energy in the ballistic motion of the electrons increases as the size is reduced and the resistive losses are then very effective indeed. effective damping and the effective ﬁlling fraction Lg 1 Li 2 ω0 = γ. Thus even if the size of the ring were negligible. where f = π r 2 /a 2 is the ﬁlling factor. This effect has also been discussed in connection with using superconducting SRRs in the microwave region [73].42) with the resonant frequency. preventing scaling to higher frequencies. The effective damping factor also becomes much larger as the size of the ring is reduced. the magnetic ﬁelds inside and outside the split cylinders can be related as (2. we can (2.39) Using Ampere’s law.41) 2 Assuming that εm −ωp /[ω(ω + iγ )] for frequencies in the infrared and optical regions. Hint [2πr/εm + 2τc dc /(εLc )] (2. 2 Noting that the plasma frequency is ωp = ne2 /(ε0 m). µ0 Heff [(2π r/(εm τc ) + 2dc /(εLc )] − ε0 µ0 ω2 π r 2 τc (2.38) (2. The mass of the electron contributes additionally to the inductance2 . I (t) dt = C −iωε0 εLc Using the fact that the total emf induced around the loop is equate the potentials as 2Vr + 2Vc = iωµ0 πr 2 Hint . µeff (ω) = 1 + 2 ω0 f ω2 . and the region of negative permeability vanishes altogether. This means that if the frequencies are too high. Hint − Hext = jφ τc .Physics of negative refractive index materials 469 and the potential drop across each capacitive gap (of arm length Lc ) is 1 jφ τc dc Vc = .43) where C = ε0 εLc /2dc is the effective capacitance of the structure. The presence of this additional inductance can be explained by noting that at high frequencies the currents are hardly diffusive and almost ballistic because the distance through which the electron moves within a period of the wave becomes comparable with the mean free path in the metal. . Then the potential drop is V ∼ m / (ne2 )∂j/∂t. the inertial inductance would still be present. we obtain the same generic form for the effective permeability.37) Eφ dl = iωµ0 Hint ds. = (Li + Lg )C Lg + L i Lg + L i (2. − ω2 − i ω (2. Averaging over a line lying entirely outside the cylinders for the magnetic ﬁeld Heff = Hext . The increase in damping is a matter of great concern for operation at optical 2 The current j = nev ∼ ne(−iωeE/m).40) The averaged magnetic induction is Beff = (1 − f )µ0 Hext + f µ0 Hint . The large increase in the damping as the dimensions are scaled down broadens the resonance and the permeability does not disperse violently now. the electrons can hardly be accelerated and the response falls.
frequencies [74]. The ﬁeld lines due to the charges are also shown by thin lines. when arranged in a periodic lattice also have a resonant ε and µ and can behave as negative magnetic materials [54].2. a double ring with a single split in one ring [79] and deformed split rings [66] have been proposed as candidates for a resonant magnetic medium. By adding more capacitive gaps to lower the net capacitance and adjusting the dielectric constant of the embedding medium. (b) The original SRR with two splits is also bianisotropic as it develops an electric dipole. One might ask then if the main ingredients of a resonant magnetic medium are just a capacitance and an inductance.48 µm and the assumption of an effective medium would hardly be valid. SRR media of rings with a single split [76–78]. Panina et al [76] have proposed the use of pairs of parallel metal sticks (with lengths ∼100 nm) periodically embedded in a dielectric medium as an effective medium with negative magnetism at infrared frequencies. Other approaches. (c) The single ring with two symmetric splits only develops a quadrupole moment and is hardly bianisotropic. The –shaped particles. Thus the SRR is oriented such that the magnetic ﬁeld is normal to the plane of the SRR and the electric ﬁeld is along the SRR and can be driven by the electric ﬁeld and the magnetic and electric resonances can overlap. then why a single capacitive gap cannot work as well. can also operate at infrared frequencies. Panoiu and Osgood [75] have shown that the original SRR. The point is that we would like to have as low an electrical polarizability as possible.5 µm [22]. (2. suitably scaled down. i. a large electric dipole moment would be generated across the capacitive gap (see ﬁgure 11(a)) and this could well dominate over the weaker magnetic dipole moment generated in the ring. When there are two splits present. The SRRs develop an electric polarization too although driven by a magnetic ﬁeld. Hsu et al [64] have claimed resonance with the original SRR at a wavelength of 10. But the size of the SSR was 10.6 µm. and this can also be driven by an electric ﬁeld [62]. With a single split ring. Most of the magnetic media that simultaneously generate an electric dipole moment are bianisotropic media [80]. of course. This can be effectively resolved. 2. shaped particles [54].44) where α and β are the bianisotropy coefﬁcients. ﬁrst introduced as the components of a bianaisotropic medium [81].470 S A Ramakrishna (a) −− + −−− − +++ + ++ + − − (b) (c) − − ++ −− + − + + − − _ _ −− − − − − − − − − − ++ + + + + + + + + ++ − − − − − − ++ ++ + Figure 11. the dipole moments across opposite ends cancel each other and one only gets a weak electric quadrupole moment (see ﬁgure 11(c)) whose effects would be expected to be much weaker than those of the magnetic dipole moment. it has been numerically demonstrated that a medium of SRR with four splits can have a negative µ at a telecommunications wavelength of about 1. This bianisotropic behaviour is maximal for (a) a single split ring which develops a large dipole moment. In fact. the constitutive relations are D = εE + αH. It has been shown that the response of the SRR with two splits tends to tail off in the infrared region (λ ∼ 5 µm) [28]. Even the original SRR medium [3] is bianisotropic [82] as there is an electric dipole moment that develops across the capacitive gaps (see ﬁgure 11(b)). by rotating adjacent . B = βE + µH.e.7.
Their numerical calculations showed that if thin wires were introduced into a SRR medium. causing them to oscillate at low frequencies. SRRs (. The possibility of a medium with a very low frequency (∼50 Hz) negative magnetic metamaterial has also been proposed [21]. The enhanced transmission in the latter case within the negative µ band is interpreted as evidence of a negative refractive index. any resonance that can be driven by the magnetic ﬁeld of radiation can be used. The right panel shows the transmission across a sample of thin wire arrays (· · · · · ·). SRRs in the plane by 180˚ and the corresponding electric dipole moments would cancel. the dispersive µ can become negative in a band of frequencies above the resonance frequency where the system mimics a plasma of magnetic monopoles. Combining . Admittedly it is not obvious that the media would have a negative n. Only the ﬁrst monopolar magnetic Mie resonance (with the applied magnetic ﬁeld along the cylinders) can contribute to the effective magnetic polarizability upon homogenization. 2. In a process very similar to that in an electrical resonant dipole medium. 84] were very suggestive of it. Their uniaxial composite shown in ﬁgure 12(a) consisted of wires of 0. In fact.3..Physics of negative refractive index materials (a) (b) 471 Figure 12.. The use of ferroelectric cylinders with very large values of ε at gigahertz frequencies has been proposed. and the dielectric constant of the cylinders needs to be large to reduce the resonant frequencies to within the ﬁrst Bragg band.845 GHz. The left panel shows a photograph of the composite of thin wire arrays and SRRs that shows a negative refractive index made by the San Diego group (courtesy D R Smith). and the SRR medium with µ < 0 would yield an effective medium with n < 0. This consists of a periodic lattice of current carrying wires. Displacements of these wires magnetize the medium.1.8 mm thickness and SRRs with ω0 = 4. Negative refractive index materials We will follow the rule that any medium with Re(ε) < 0 and Re(µ) < 0 simultaneously will be considered to have a negative refractive index. But the calculations and experimental measurements of Smith et al [4. Smith et al [4] proposed that combining in a composite the thin wire medium with ε < 0.) and a composite of thin wire arrays and SRRs (——) as measured by the San Diego group (adapted from the data of [4]). with alternate wires carrying currents in opposite directions. The symmetry of the single ring with two symmetrically placed capacitive gaps renders this less bianisotropic and electrically less active. A rigorous justiﬁcation of this is presented in section 3. The tension in the wires and the mutual repulsion between them give a restoring force. a passband occurs within the bandgap of negative µ for radiation with the electric ﬁeld along the wires and the magnetic ﬁeld normal to the plane of the SRR. The use of the magnetic Mie scattering resonances of a dielectric cylinder for generating a negative magnetic medium has also been proposed [83].
and the absorption was found to be sufﬁciently small to unambiguously demonstrate the negative refraction effect and to be consistent with calculations. Then the quasistatic responses derived here for ε and µ would be valid in the negative refractive index band.2. but indicated that there was considerable absorption in these media due to Ohmic losses and dissipation in the dielectric board. it is not a straightforward conclusion. 85] showed that there was indeed a negative index band which overlapped reasonably with the negative µ frequency band. the interference can be reduced greatly [87]. The San Diego group went on to show [5] that a prism made of such a composite NRM refracted microwaves in the direction opposite to the normal direction compared with a prism of a positive refractive material such as Teﬂon3 . These experiments have unequivocally demonstrated the existence of a negative refractive index. as long as the thin wires are not placed in regions where the highly inhomogeneous magnetic ﬁelds associated with the SRRs are present (along the axis of the SRR) and the SRR planes are placed such that they are at the points of symmetry between the wires (where the magnetic ﬁelds associated with the wires are minimal). and ω0 is √ the magnetic resonance frequency (ωm = ω0 / (1 − f )). in a metamaterial. Although there is both experimental and numerical evidence that the composite of thin wires and SRRs has a negative refractive index band. That the two structures do not interact and interfere with each other’s functioning is questionable. The San Diego experiments also met with criticism as they were conducted in a waveguide and there was speculation about leaky waveguide modes. then we obtain a passband in the region ω0 < ω < ωm . Third. The experimental results on transmission through waveguides ﬁlled with a medium of thin wires. This was taken as proof that the medium was a NRM.1). we get k2 = 2 2 (ω2 − ωp )(ω2 − ωm ) εµω2 . If ωp is the largest of them. and it was claimed that the observed effects could be due to just absorption [9]. a medium of SRRs only and a composite medium with both thin wires and SRRs is shown in ﬁgure 12(b). Other calculations [26] showed that the losses were primarily due to Ohmic losses in the metal rings and showed that the losses were somewhat smaller. respectively. we note that the magnetic ﬁelds due to the wires fall off rather rapidly with distance from the wires and should not affect the SRRs badly. A clear demonstration that the ﬁelds due to the two structures do not interfere with the functioning of each other is yet to be made. The numerical calculations suggested that the thin wires and the SRRs functioned independently and the composite had a passband with a negative refractive index. The samples in the original San Diego experiments had a small transmission. First. 26] where both the transmission and reﬂection were measured. It has been shown that although the quasistatic interaction between the SRR and the thin 3 This is due to the fact that radiation refracts to the other side of the normal at the interface between positive and negative media (see section 3. Second. Thus the relative placement of the components could be crucial and might account for the differences reported in numerical and experimental data. Numerical photonic calculations [65. Pokrovsky and Efros [86] showed that embedding a thin wire array in a homogeneous medium with µ < 0 does not produce an NRM and the system has no propagating modes. . Now these experiments have been repeated with large samples in free space [25. and the SRRs and the thin wires would function independently as if in vacuum.45) where ωp and ωm are the electric and magnetic plasma frequencies. The transmission is in the stopband due to the negative µ for SRRs and there is enhanced transmission within this bandwidth when thin wires are introduced into the SRR medium.472 S A Ramakrishna the dispersions for the plasmalike dielectric medium and a resonant magnetic medium. one cannot assume a homogeneous ε or µ due to the other parts of the composite structure. = 2 c2 c2 (ω2 − ω0 ) (2.
This is the ﬁrst level of homogenization. Marqu´ s et al [90] have demonstrated that a waveguide ﬁlled with a e negative µ medium behaves as an NRM for frequencies below the cutoff frequency. It is well known that a waveguide below the cutoff behaves as a onedimensional plasma and the split rings when placed within it make it behave as an NRM. the structures in question are sufﬁciently small compared with the lengthscale over which the applied ﬁelds vary for only the ﬁelds due to the ﬁrst few multipoles of the charge and current distributions induced in the structures to contribute to the polarization. However. microscopically large.4. Homogenization and effective macroscopic material parameters In our discussion. has been proposed [54]. Then the wave is highly evanescent and homogenization may not be easily possible [88].e.48) that determine the effective dielectric permittivity and the magnetic permeability of the bulk metamaterial which are second ranked tensors in general. i. and ε due to the thin wires is negative. D = ε0 εeff · E . but only a few averages such as the corresponding dipolar ﬁelds or (rarely) the quadrupolar ﬁelds can be resolved. In fact.Physics of negative refractive index materials 473 wire array is small. 2. where the magnetic resonance frequency (for the magnetic ﬁeld normal to the loop) and the electric dipole resonance (for the electric ﬁeld along the legs) coincide. The Maxwell–Garnett approximation. The structural units of the metamaterial are assumed to be sufﬁciently large on a molecular scale. Thus we have the Maxwell’s equations ∂B ∇ ×E=− ∇ · D = 0. there are two levels of averaging involved here. (2. to be described by the bulk dielectric permittivity and magnetic permeability of the constituent materials. Thus we have the averaged ﬁelds. which is essentially the quasistatic generalization of the Clausius Mossotti electrostatic approximation. Further there are problems associated with a simplistic approach to homogenization of the composite at subresonance frequencies where µ is positive and large. and the ﬁeld induced in the uniform host by a single inclusion (assumed to be spherical or . Then the radiation encounters a coarsegrained spatial average of the polarizabilities of the constituent structures of the medium. . ˜ (2. ε and µ insofaras the electromagnetic interactions are concerned. In other words. and ε and µ are the material parameters of the constituent materials. Deformed SRRs have similarly been used to simultaneously generate an electric and a magnetic resonance for orthogonal magnetic and electric ﬁelds [66]. it was assumed that media with highly inhomogeneous inclusions could be described by only two complex parameters. Central to this assumption was the fact that the size or lengthscales of the inclusions are very small compared with the free space wavelength of the radiation. (2. Alternative proposals for NRMs using other systems have been put forward. has been used widely to calculate the bulk electromagnetic properties of inhomogeneous materials [91]. Panina et al [76] have proposed the use of oriented paired parallel metal sticks as the constituents of the NRM. The component with the largest ﬁlling factor is considered to be a host in which other components are embedded. The use of alternatively oriented shaped particles. ∇ ×H= . ˜ B = µ0 µeff · H .47) ∂t where D = ε0 ε(r)E and B = µ0 µ(r)H. the ﬁne structure of the charge and current distributions is not discernible. there are residual interactions between the two subsystems which lead to both a shift in the resonant frequency [88] and an increased dissipation in the structure [89].46) ∂t ∂D ∇ · B = 0.
52) (x) Heff = Hx dx. a Sx a Sy a Sz (2. are deﬁned by averaging E and H along the sides of the unit cell: (x) Eeff = 1 a 1 a (a. In the case of periodically structured materials.0) (0.0) (z) Eeff = 1 a 1 a (0.a. (0. B · dσ . however. Beff = Bz dσz . Sy is the face lying in the Z–X plane and Sz is the face along the X–Y plane. Beff = By dσy .0) Ex dx. Eeff and Heff . ﬁrst of all. The averaged ﬁelds.474 Z S A Ramakrishna a a X Figure 13.0. S (2. (0.0) (z) Heff = Hz dz. Deff = Dy dσy . 93] for high ﬁlling fractions that include more residual interactions or higher multipole effects. We should also note that there are more mathematically rigorous homogenization theories [94]. .0) (0.0. The Maxwell–Garnett approach incorporates the distortions due to the dipole ﬁeld on an average and has been very successful in describing the properties of dilute random inhomogeneous materials.a) (2. implies that the averaging need not be over orientational and density ﬂuctuations.0. The procedure for averaging the internal microscopic ﬁelds consists of averaging the E and H ﬁelds over the edges of the unit cell of the simple cubic lattice.0) Eeff = Heff = (y) (y) 1 a 1 a (0. Deff and Beff are deﬁned by averaging them over the faces of the unit cell: 1 1 1 (y) (x) (z) Dx dσx . Consider the unit cell of the lattice (assumed to be simple cubic) shown in ﬁgure 13.0) Hy dy.0. There are other effective medium theories [92. The Maxwell equations can be recast in their integral form: H · dl = − C S D · dσ . while the D and B ﬁelds are averaged over a face Sx for the xcomponent and so on.53) (2.0. (0. (0. Deff = Dz dσz . which itself suggests an averaging procedure. The averaged ﬁelds enable us to determine εeff and µeff componentwise.0.0. In our treatment we have in some sense used the Maxwell–Garnett type approximation but. assumed uniform depolarizing ﬁelds that simpliﬁed the calculations. (0. We will now discuss an averaging procedure of Pendry [3] that was ﬁrst used to discretize Maxwell’s equations on a lattice [17].0) Ey dy.49) (2. (0.50) E · dl = − C where the curve C encloses the surface S. which is however beyond the scope of this review.54) where Sx is the face in the Y –Z plane.0) Similarly.0. elliptical) is calculated and the distortion of this ﬁeld due to the other inclusions is calculated approximately.0. Deff = a Sx a Sy a Sz 1 1 1 (y) (x) (z) Beff = Bx dσx .a) Ez dz.51) (2. which analyse the ﬁelds in an asymptotic expansion in terms of the microscopic lengthscale associated with the inhomogeneities.a.0) (a. the periodicity.0.
It should be noted here that we do not consider a photonic crystal or a bandgap material to be a metamaterial at high frequencies when the lattice period is comparable with or smaller than the wavelength in free space. so that the continuity of the parallel components of Eeff and Heff across the surfaces is maintained. In the metamaterial limit of low frequencies. This procedure was adopted in section 2. 72]. and a simple interpretation of the reﬂection and transmission of the metamaterial slab in terms of an isotropic or anisotropic material is not strictly correct. The multivalued nature of the trigonometric functions gives rise to an ambiguity in n and Z which. The reasons for this effect are yet to be completely understood. however. These negative imaginary parts of ε and µ have been explained in part as a result of the ﬁniteness of the unit cell [52]. i. for the refractive index. Although surprising. As has been pointed out. however. Likewise. it should give unique medium responses to applied ﬁelds. note that many of the metamaterials are intrinsically bianisotropic. the radiation can discern the individual structures and scattering dominates—there is a splitting of the wave with large phase shifts. One should.55) (2. In these processes.e. and the impedance. at least weakly. Z.56) sin(nkz d)T eikz d . . of the slab and requiring that Re(Z) > 0 and Im(n) > 0 for causal and absorptive media (see section 3.2 to determine the effective magnetic permeability. Similarly. this does not violate any fundamental requirement as long as the passive metamaterial is only dissipative. implying that the scattered ﬁelds are evanescent and . In this limit. can be resolved by determining them for several thicknesses. and it should be possible to uniquely determine the scattered ﬁelds for the metamaterial.3) is positive deﬁnite. as n = ± cos−1 Z=± 1 − r2 − t2 2t 1/2 .57) (2. the righthand side of equation (1. from which one obtains the dielectric permittivity ε = n/Z and the magnetic permeability µ = nZ. the scattered ﬁelds corresponding to the small scatterers have very large wavevectors. T = cos(nkz d) − R= −i 1 Z− 2 Z 1 i Z+ 2 Z −1 sin(nkz d) e−ikz d . experiments usually access only the emergent quantities such as the reﬂection and transmission through (say) a slab of the effective medium and not the internal microscopic ﬁelds. It is then a matter of great reassurance that the effective medium parameters derived from the emergent quantities coincide with the effective medium parameters obtained from a homogenization of the microscopic internal ﬁelds [28. That would be negligible in the limit of a very small unit cell size. there is always an inbuilt ambiguity in the phase of the wave corresponding to the unit cell size which would always affect the effective medium parameters to a small extent. In fact.1). (2. 52] and a medium of dielectric cylinders [83]. d. n. The effective medium parameters determined from any such averaging process should be unique and independent of the manner of the averaging process. the metamaterial of cut wires with Re(ε) < 0 has a counterintuitive Im(µ) < 0 [52]. metamaterials with Re(µ) < 0 have Im(ε) < 0 for both an SRR medium [28. Smith et al [72] inverted the classical relations for the reﬂection and transmission coefﬁcients from a slab of homogeneous material of thickness d. (2.58) (1 + r)2 − t 2 (1 − r)2 − t 2 where r = R and t = T eikz d .Physics of negative refractive index materials 475 There is only one restriction in this procedure: that the unit cell should not intersect any of the structures contained within.
ω −ω [Re(µ(ω )) − 1] dω . The generic forms of the dispersion for a resonant dielectric and magnetic media satisfy the KK relations. Further. 2 s 2 is negative.63) 1 Re(µ(ω)) − 1 = − P π Im(µ(ω)) = 1 P π ∞ −∞ Im(µ(ω )) dω .60) Re(ε(ω)) − 1 = P π ω −ω −∞ 1 Im(ε(ω)) = − P π ∞ −∞ [Re(ε(ω )) − 1] dω . is easily accessed. which is clearly not possible. A nondispersive NRM would imply. These are relations between the real and imaginary parts of the response functions: ∞ Im(ε(ω )) dω 1 .5. implying that the dielectric permittivity and the magnetic permeability also disperse spatially at high frequencies (before the breakdown of homogenization). This never really affects us in the case of common optical media where the macroscopic material response functions hold almost down to the level of a few atomic distances. for example. which is important for metamaterials with negative material parameters. ω −ω where P is the Cauchy principal value. 95]. note that the integrals in the KK relations involve frequencies all the way up to inﬁnity and it is clear that the effective medium theories will break down at high frequencies. The material parameters of a causal medium should satisfy the Kramers–Kronig (KK) relations [18. Thus the very high frequency limit is never really probed. when the homogenization becomes invalid. Generalized KK relations for the spatially dispersive materials are available [99] and may be made use of. ω −ω ∞ −∞ (2. and they are causal insofaras the homogenization is valid. For the plasma of equation (1. . i. However. Negative ε and µ are resonant effects and they are necessarily dispersive and dissipative.e. This is not true in the case of photonic bandstructure materials.62) (2. usually the wavelength is larger than the periodicity by only one or two orders of magnitude. but the sign of the transform becomes negative [96]. then too the Hilbert transforms hold. (2.61) (2. Causality and energy density in negative refractive media It must be noted that dispersionless material parameters ε < 0 or a µ < 0 cannot exist. the real and imaginary parts are Hilbert transforms of each other. ε0 ε 2 µ0 µ 2 Estatic = (2. imply that the energy density for static ﬁelds. If the poles occur only in the upper half frequency plane.7) a term of 4πσ/ω should be added to the righthand side of equation (2. 2. In the case of metamaterials. where momentum can be transferred between the propagating modes of the radiation and the crystal. for example. for example an amplifying medium or for nonlinear fourwave mixing.61) due to the pole at zero frequency. that time runs backwards for light in the medium [110]. due to the ﬁniteness of the structures the metamaterials have highly nonlocal responses. and this high frequency cutoff.59) E + Hs < 0. Negative ε or µ for static ﬁelds would. These relations hold as long as the response functions are analytic in the upper half plane and poles of the functions are in the lower half plane.476 S A Ramakrishna do not propagate. The case of meromorphic response functions when poles occur in both the upper and lower plane [97] has been discussed in [98]. Thus the KK relations should be applied cautiously to metamaterials keeping this in mind.
3. For the case ε < 0 and µ < 0.100.65) ∗ ∗ where αE (ω)αE (ω) = ω Imε(ω) and αH (ω)αH (ω) = ω Imµ(ω). Thus. The dissipation in the medium depends on the time histories of the ﬁelds and the medium polarizations. for frequencies far enough away from a resonance where absorption is small. Again it is emphasized that the KK relations have been used to obtain this energy density and it is valid only for the low frequency regime when the effective medium parameters themselves are valid. (2. Writing ε = ε + iε . Choice of the wavevector As discussed in section 1. This has been contested by some authors [7.1) (3. This is positive deﬁnite and applicable to the case where ε and µ can become negative as well. it is possible to write down an expression for the energy density as [18. ε0 µ0 E (t) = + 2E(t)2 2H(t)2 +∞ t t 2 + −∞ dω αE (ω) −∞ dτ eiωτ E(τ ) + αH (ω) dτ eiωτ H(τ ) −∞ . For positive media. then n < 0. Propagation of radiation in negative refractive media 3.2. Smith and Kroll [102] considered the problem of a radiating current sheet in a NRM and found that power is radiated away from the current sheet if the impedance Z= µ(ω) µ = ε(ω) n (3.e. the phase accumulated in propagating a distance is φ = −nω /c. µ > 0 and amplifying if ε < 0. µ > 0 and n > 0 and the power is radiated away. Walser . 19] E= 1 ∂(ωε) ∂(ωµ) ε 0 E2 + µ 0 H2 2 ∂ω ∂ω (2. But if µ < 0 in an NRM.101]. Thus a sufﬁcient condition for a positive energy density under these approximations is ∂(ωε)/∂ω > 0 and ∂(ωµ)/∂ω. A general and exact expression for the energy density has been derived in [101]. (3.1. This is satisﬁed in the regions of normal dispersion far from the resonances.64) in the socalled transparency regions.3) was positive. this condition demands that the negative sign of the square root be chosen. i. k × H = −ωε0 εE.Physics of negative refractive index materials 477 It is nontrivial to identify a positive deﬁnite quantity as the electromagnetic energy density in dispersive media [19. µ < 0. 103]. we note that the medium is absorbing if ε > 0.2) implies that the phase vector k has a reversed sign. and µ = µ + iµ . Let us consider a plane wave exp(ink0 x) in an absorbing NRM where the refractive index n = ±[(ε µ − ε µ ) + i(ε µ + µ ε )](1/2) ±[ε µ + i/2(ε µ + µ ε )/(ε µ )] for small absorption. k × E = ωµ0 µH. Assuming narrow band radiation and making a Taylor series expansion about the carrier frequency and retaining only the linear term. the reversal of the signs of both ε and µ in Maxwell’s equations. Now the wave should decay in amplitude as it propagates in the dissipative medium and that is governed by the sign of Im(kz ). We can take this to imply that the refractive index n = εµ is negative. this expression is rather difﬁcult to use due to integrals over the time history of the ﬁelds. it is opposite to the Poynting vector √ S = E × H. Although important from a fundamental point of view.
kz = ± εµ . The zcomponent of the wavevector. we consider an electromagnetic wave with a wavevector [kx . which are unphysical in a semiinﬁnite medium. kz . and actually towards the source if the medium is overall amplifying. Pokrovsky and Efros [103] argued that the refractive index is ambiguous as it does not enter the Maxwell equations and that only the group velocity is negative in NRMs. Without loss of generality. This problem can be formally treated [104] by considering the most general properties of the wavevector in the complex plane. However. The only conditions are that evanescently decaying waves remain decaying. For amplifying media. Regions 1 and 8 correspond to propagating waves in conventional positive media that are absorptive or amplifying. corresponding to the two signs of the square root. Decaying evanescent waves fall within region 2 if ε µ + ε µ > 0 and in region 3 if ε µ + ε µ < 0. our choice results in a Poynting vector that points towards the source (interface in this case). has to be obtained from the dispersion relation. Due to xinvariance. This ensures that the nearﬁeld features of a source cannot be probed at large distances merely by embedding the source in an amplifying medium. 2 2 2 k2 = kx + ky + kz = εµ ω2 . respectively. kx is preserved across the interface. Further the media could be absorbing or amplifying. Hence our range for the argument θ of kz becomes −π/2 < θ < 3π/2 for the ﬁrst Riemann sheet and 3π/2 < θ < 7π/2 for the second √ Riemann sheet. Note that the Poynting vector points away from the source (interface) if the medium is absorptive overall.478 S A Ramakrishna and coworkers [7] claimed that the wave propagation in NRMs was very inhomogeneous and a negative refractive index could not be claimed. This. ε < 0 and µ < 0. Now the waves in 2 2 medium 2 could be propagating [kx < Re(εµω2 /c2 )] or evanescent [kx > Re(εµω2 /c2 )]. however. kz = ± Z = Z1/2 eiθ/2 √ and Z1/2 eiπ +iθ/2 . c2 (3. we note that √ there is a branch cut in the complex plane for Z = kz and one cannot analytically continue the behaviour of the waves across this branch cut. µ and k. (3. kz ] to be incident from vacuum on the left (−∞ < z < 0) on a semiinﬁnite medium (∞ > z > 0) with an arbitrary value of ε and µ. regardless of ε and µ . ε and µ. This branch cut divides the Riemann surface into two sheets in which the two different signs for the square root will have to be taken. respectively.4) in such homogeneous media. depending on kx . Note that although the position of 2 the branch cut in the kz plane is only a convenient choice. Regions 6 and 7 correspond to growing evanescent waves that blow up at the inﬁnities. The waves corresponding to quadrants 1 and 4 have a propagating nature. The complex plane for kz = Z with the corresponding eight regions in the two Riemann sheets is shown in ﬁgure 14. For complex ε. one has to be more careful. the wave of course becomes inhomogeneous. the wave amplitude at the inﬁnities has to obviously disappear.5). the ﬁnal regions in kz for various media is ﬁxed by causality. √ The 2 different regions in the Z = kz plane are mapped into the different physical regions of the Z plane. and the waves corresponding to quadrants 2 and 3 are evanescent. there does exist a very general problem of choosing the sign of the square root for determining the wavevector. For absorbing media. propagating ones remain propagating and no information can ﬂow in from the inﬁnities. This enables us to divide the complex plane for Z = kz into the four quadrants shown in ﬁgure 14.5) c2 where a physical choice has to be made for the sign of the square root. ω2 2 − kx . Crucially. 0. Due to the above reasons we will inconveniently choose the branch cut along the 2 negative imaginary axis as shown in ﬁgure 14. For the case of evanescent waves in amplifying media. depending on the sign of Im(εµ) in 2 equation (3.
.< 0 Physically inaccessible in semi infinite media 2 Figure 14. The quantity ε µ + ε µ determines the energy ﬂow which has also been recently noted by Depine and Lakhtakia [105]. it is clear that the continuity of the ﬁelds and Maxwell’s equations require the transmitted wavevector to be (kx . 0.Physics of negative refractive index materials 479 (a) Im (kz2) (b) . The evanescent regions could be absorbing or amplifying as explained in the text. respectively.5). Re (kz2) + Propagating waves Re (kz ) + Propagating waves Amplifying Evanescent waves Propagating waves Positive refractive Growing . From the discussion in section 3. we realize that the ray representing the energy 4 In addition to the effect discussed here.+ . + . . . + <0 + >0 . .2.. does not violate causality as the Poynting vector/energy ﬂow decays exponentially to zero at inﬁnity and no information ﬂows in from the inﬁnities. Some effects of a reversed wavevector Now let us look at some of the effects of a reversed wavevector in some detail. . the negative sign of the square root needs to be chosen. k.> 0 . .1. depending on whether ε µ + ε µ < 0 or ε µ + ε µ > 0.. . . Veselago [1] has discussed some of these effects in his original paper4 .+ . . In dissipative media Im(kz ) < 0 for propagating waves.1. . kz− ). . .. > 0 . (b) The complex plane for kz = Z with the corresponding eight regions in the two Riemann 2 sheets for kz .< 0 Evanescent waves Propagating waves . This counterintuitive behaviour does not imply that source has turned into a sink—rather it indicates that there would be a large (inﬁnitely large for unsaturated linear gain) accumulation of energy density (intense local ﬁeld enhancements) near a source. + <0 + >0 Decaying Evanescent waves Absorbing Absorbing Positive refractive Negative refractive medium medium + + Propagating waves Propagating waves Amplifying Negative refractive . . Thus. . . The refraction of radiation at the interface of a positive medium and a negative medium has been one of the most discussed issues.2. . medium Evanescent waves medium . corresponding to absorbing or amplifying media. ... . The problem of radiation pressure even in ordinary dispersive media is controversial [107] and we will not discuss this problem in structured NRMs and metamaterials here. The sign of the square root for the region is shown enclosed in a small circle. In both cases. however. + <0 .. . where kz− is the negative square root in equation (3.. depending on the sign of ε and µ . 3. justifying calling these media with both ε < 0 and µ < 0 negative refractive index media.+ . Noting that Poynting vector is oriented exactly opposite to the phase vector. in the NRM. . The branch cut√ along the negative imaginary axis for the square root is shown. one can more reasonably talk of NRMs as media in which radiation has a negative phase velocity rather than a negative group velocity. which reduces to Im(n) > 0 for normal incidence as noted in [106]. 0. . the sign of the wavevector and the sign of the refractive index are the same. . Veselago has also suggested a reversed radiation pressure in NRMs.. 3. The modiﬁed Snell’s law of refraction.. . . (a) The complex plane for kz showing the different regions for the propagating and evanescent waves..< 0 Im (kz) . Now propagating waves in media with ε < 0 and µ < 0 simultaneously are included in regions 4 and 5. .+ . . In the case of normal incidence.. kz+ ). . Consider the electromagnetic plane wave incident on an NRM from a positive medium with a wavevector (kx .
The modiﬁcation of the Snell’s law is counter to a very fundamental concept in optics of isotropic media that. however. Noting that k = nω/c. that the angle of the reﬂected beam remains unaffected. (3. The reversed Doppler shift. ﬂow corresponding to the refracted wave should now lie on the other side of the normal as shown in ﬁgure 15(a). (c) The Veselago lens consisting of a slab of NRM that can focus a point source from one side of the slab to the other side. for emission along the direction of the motion of the source in the NRM (say. On the other hand. rays emanating from a point source located at a distance d1 on one side of the slab would be refocused to a point at a distance d2 on the other side.2. depending on how dense the second medium is. the refracted beam actually emerges on the other side of the normal5 . The energy ﬂow and the phase vectors are in opposite directions in the NRM. This is also called a Veselago lens today. Figure 15(b) shows the negative refraction through a prism which was used in the experiments of [5. n+ sin θ+ = n− sin θ− . . (a) Schematic depiction of the negative refraction at an interface between a positive medium and an NRM. In a sense. 25]. but never cross it.7) where γ = (1 − v /c ) is the relativistic factor. In fact. Hence a convex lens causes a plane wave to diverge while a concave lens causes the plane wave to become convergent. This lens is different from the usual optical lens with curved surfaces in that it does not focus rays from inﬁnity to a point. (b) The negative refraction effect through a prism. 3. 109]. c+v (3. This negative angle can also be intuitively inferred from Snell’s law. the total phase shift accumulated by any wave in going from the source point to the image point is zero as a consequence of the negative phase vector inside the NRM7 .8) Note that negative refraction is possible in anisotropic media as well [108. the phase unfolds inside the NRM and it is almost as if time runs backwards [110].6) where n− < 0 ⇒ θ− < 0. The frequency measured by a detector in the frame of the NRM is ω = γ (ω + k · v ). Consider now the behaviour of traditional curved lenses made of NRMs. where d is the thickness of the slab. Note that the phase accumulated due to traversal through the NRM is negative. In this case. with n = −1) ω= 5 6 7 c−v . These rays would simply pass through unaffected.480 S A Ramakrishna (a) Positive media (b) (c) Object plane Negative refraction Image plane n = –1 NRM Wave vectors Rays Energy flow Positive refraction d/2 d d/2 Figure 15. Note. Assume a moving source emitting radiation at a frequency ω in an NRM with a velocity v with respect to the medium. 2 2 −1/2 (3. As a direct consequence of this we have the result that a ﬂat slab of an NRM6 with n = −1 can focus a source as shown in ﬁgure 15(c). This is in contrast to the phase corrections that are made to different rays in a conventional lens. the refracted wave can bend as close to the normal as possible. provided that d1 + d2 = d.2.
Physics of negative refractive index materials
S k S k
481
A
B
v
C
A
B
v
C
Figure 16. The Cerenkov radiation cone angle is modiﬁed for a reversed phase vector. There is constructive interference for emission only in one direction. The phases of the wavefront emitted by the particle at A, B and C have to be identical. The case on the left represents Cerenkov radiation in a positive medium with positive k while the ﬁgure on the right shows the emission in a medium with negative k.
(a)
>0 2>0
2
(b) −∆
n1
2 2
<0 <0
∆
n1 n 2 < n1
n2 < n 1
Figure 17. Schematic showing the transverse displacement of a beam of radiation undergoing total internal reﬂection. The GH shift is negative for reﬂection from NRMs.
i.e. the frequency measured by a detector would be smaller when the source is moving towards it! This is counter to the frequency increase that we would get in a normal medium. Again it is the reversed phase vector in NRMs that is responsible for this reversed Doppler shift. 3.2.3. An obtuse angle cone for Cerenkov radiation. A charged particle moving at relativistic speeds through a medium emits Cerenkov radiation in a cone of angles if its speed exceeds that of light in the medium. The angle of this cone is deﬁned by the condition that the emitted radiation interferes constructively only in that direction while destructively interfering in all others. This condition is satisﬁed (see ﬁgure 16) along an acute angle deﬁned by cos θ = c/(nv). This angle, however, becomes obtuse for constructive interference in NRMs due to the reversed phase vector as is clear from ﬁgure 16, i.e. θ → π − θ. In reality, due to the dispersion in the material parameters of the NRM, there will be Cerenkov radiation in both the forward and backward directions for different frequencies [111]. This modiﬁed cone for Cerenkov radiation has been numerically demonstrated for photonic crystals that exhibit the negative refraction effect [112]. 3.2.4. Reversed Goos–H¨ nchen shift. When light is incident upon an optically rarer medium 2 a from an optically denser medium 1 (the refractive indices n1 > n2 ), there is a critical angle of incidence, θc , given by ε2 µ 2 sin θc = , (3.9) ε1 µ 1 beyond which the wave in medium 2 is evanescent and the radiation is completely reﬂected back into medium 1. For a beam of light with a ﬁnite transverse extent, this is also accompanied
482
S A Ramakrishna
by a shift of the beam in the transverse direction as shown in ﬁgure 17. This shift occurs usually along the parallel component of the wavevector and was experimentally ﬁrst demonstrated by Goos and H¨ nchen [113]. It turns out that the Goos–H¨ nchen (GH) shift is also negative a a for reﬂection from NRMs [114–117]. Artmann showed [118] that for angles of incidence sufﬁciently away from the critical angle and the grazing angle, the beam is displaced by ∂ r δgh = − , (3.10) ∂kx1 where r is the phase shift for the reﬂected beam and kx1 is the parallel component of the wavevector in medium 1. Now consider the reﬂection coefﬁcient for ppolarized radiation: kz1 ε1 − kz2 ε2 r= . (3.11) kz1 /ε1 + kz2 /ε2 We note that for total internal reﬂection, kz2 → iκz is imaginary. Now if we simultaneously make negative the signs of µ2 and ε2 , κz remains unchanged. However the reﬂectivity from the NRM medium r− = 1/r+ = e−i r , where r+ = ei r is the coefﬁcient of reﬂection from a medium with positive ε and µ. Clearly the GH shift will be negative. A similar analysis holds for the spolarized radiation with µ replacing the corresponding ε in our analysis. Note that it is actually sufﬁcient to make only ε < 0 for the ppolarization and µ < 0 for the spolarization to obtain a negative GH shift. It has been shown that the negative GH shift can be very large due to surface plasmon states [119], and negative GH shifts have also been shown for layered systems of positive and negative media [120]. The GH shift does not, however, violate causality, and a pulse would take a ﬁnite amount of time to undergo the transverse shift. This effect is primarily due to the evanescent ﬁelds in the second medium. 3.3. Phase velocity, group velocity and energy ﬂow The wavevector k in an isotropic NRM points opposite to the direction of the Poynting vector S = E × H which gives rise to the negative refraction effect. Thus it is clear that the phase velocity vφ = k/ω is negative. However, the energy ﬂow is along the Poynting vector and away from a source. The group velocity, deﬁned as vg = ∇k ω(k), is usually along the direction of the Poynting vector for reasonably narrowband pulses and thus should be oriented opposite to the phase velocity. We do note that the very concept of a group velocity can fail for broadband pulses which can be distorted easily due to dispersion [121]. We note that the phase velocity and the Poynting vector do point in different directions in anisotropic media and they can also point in opposite directions in anisotropic media [108]. Hence this point need not surprise us greatly. However, the negative refraction effect and the group velocity in an NRM have been the subject of some interesting debate [7,122]. Valanju et al [7] analysed the interference fronts of two waves with slightly differing frequencies and found that the interference fronts refracted in the positive sense (see ﬁgure 18). Misinterpreting the normal of these interference fronts to be along the group velocity, it was claimed that the refraction at an interface between a positive medium and an NRM was positive. First of all, the group velocity in a homogeneous isotropic medium can be expressed as [122] k dω(k) , (3.12) vg = ∇k ω(k) = p k dk where p = ±1, depending on the choice of sign of the square root as the frequency does not depend on the direction due to isotropy. This means that the group velocity can only be parallel
Physics of negative refractive index materials
(a) (b)
483
Figure 18. The interference patterns of two superimposed waves with slightly differing frequencies refracting at an interface between negative and positive refractive media. (a) The case of inﬁnite plane waves when the interference fronts appear to refract positively. (b) A schematic picture of the refraction of beams with ﬁnite transverse size at slightly differing frequencies. Although the interference patterns appear to refract positively, the beams (energy ﬂow) are seen to refract negatively. The sidewise motion of the interference fronts is not apparent in the case of inﬁnitely extended plane waves.
(in a positive medium) or antiparallel (in an NRM) to the phase velocity. The point is that the velocity of the interference fronts would coincide with the group velocity only if the waves have wavevectors in the same direction with slightly differing magnitudes (frequencies). Due to dispersion in ε(ω) and µ(ω) in the NRM, the two waves with differing frequencies would refract at slightly differing angles. Now the constitutive waves of the interference patterns in the NRM no longer point in the same direction and the corresponding interference front no longer propagates along the direction of the group velocity. In fact, a simulation using Gaussian beams [123] with slightly differing frequencies clearly showed that the beam suffers negative refraction while the normal of the interference fronts points in a completely different direction that would correspond to a positive angle of refraction. In fact this is true of any wave packet that is localized in the transverse direction [124,125] and this motion is not apparent in the case of plane waves due to their inﬁnite transverse extent. The numerical calculations of Caloz et al [126] showed that the refracted power in a waveguide bend ﬁlled with an NRM was along the angle for negative refraction. Pacheco et al [127] analytically calculated the Poynting vector for the refraction of a multifrequency plane wave and found that the power ﬂow was indeed along the angle for negative refraction. Interestingly an energy velocity vE = S/u can be deﬁned, where S is the Poynting vector and u is the average electromagnetic energy density. As pointed out, it is difﬁcult to deﬁne u in a dispersive medium, and this quantity has not been well investigated for an NRM yet. Foteinopoulou et al [128] have shown through numerical ﬁnite difference time domain (FDTD) calculations that the negative refraction of a beam experiences a long delay at the interface but the beam refracts in the NRM along the negative refraction angle. The FDTD calculations of Cummer [176] have also indicated that there is indeed a discontinuity in the phase front of the wave at the interface, but the phase discontinuity was shown to be causal. The group delay of a pulse propagating through a slab of NRM is also of great interest. For narrowband pulses, the group delay time is τφ = (∂φ)/(∂ω)ωc , where φ(ω) is the phase of the transmission or reﬂection coefﬁcient and ωc is the carrier frequency of the pulse within the group velocity description. Dutta Gupta et al [129] have shown that the phase delay time for transmission across a slab of NRM is usually large (subluminal) and note interestingly that superluminal delay times can be obtained in the frequency region of plasmalike behaviour, ωm < ω < ωp .
13) and using Kirchoff’s law. At radiofrequencies. which will now depend on the impedances Z and Z . (3. we obtain the propagation constant √ √ β = ω LC = ω εµ. 3. we identify the equivalent impedances of the material parameters 1 .484 Z Y S A Ramakrishna iy +di Z/2 X y L (a) C L Z/2 ix Z/2 Y iy Z/2 i x +di x (b) C L C Figure 19. we obtain the Helmholtz equation: ∂ 2V ∂ 2V + + β 2 V = 0. A transmission line network with distributed impedance and admittance. ∂x ∂y (3.4. Let us note that for a TE ﬁeld (where the electric ﬁeld Ez is perpendicular to the wavevector) in a waveguide in the X–Y plane. Z = Y . Choice of (a) for the inductances gives the usual transmission line with positive propagation constant.14) where β 2 = −ZY = −Z/Z . (3.16) (3. = iωµHx . the currents Ix → −Hy and Iy → Hx . while (b) yields a backward wave structure in which the phase and group velocities are opposite. The phase velocity for the wave in the distributed network is Z = −iωµ and Z = .17) Mapping the voltage V → Ez .15) ∂V = ZIy ∂y (3. one can write the equations for the potentials across the section. The system has an associated series impedance (Z) and shunt admittance (Y ) per unit length. Consider the twodimensional distributed network shown in ﬁgure 19. ∂V = ZIx . Concentrating on one section in the medium. This takes advantage of the fact that dielectric and magnetic materials can be modelled using distributed L–C networks. ∂x ∂y Eliminating the currents. ∂x 2 ∂y 2 (3.18) −iωε If we choose Z = −iωL and Z = −1/iωC. As usual we have the same problem of a physical choice of the sign for the propagation constant. ∂x ∂Ix ∂Iy + = YV. one can equate the currents in and out of the node. where L and C are the inductance per unit length and capacitance per unit length as in ﬁgure 19(a). we have the Maxwell’s equations ∂Ez ∂Ez = −iωµHy . Backward wave structures and transmission lines There is another completely independent approach based on the theory of transmission lines to making NRMs [130–132]. this can be easily achieved by using lumped elements. ∂x ∂y ∂Hy ∂Hx − = −iωεEz .
but a propagating ray incident on this medium appears to undergo negative refraction at the interface. the loading elements dominate √ and we have an NRM with a negative phase velocity and a refractive index of 1/ωd LC. referred to as all angle negative refraction [36]. (3. Such backward wave structures show all the hallmarks of an NRM such as a modiﬁed Snell’s law at an interface between a forward wave and backward structures.20) vg = +ω2 LC.19) ω LC where the negative sign of the square root has been chosen. at which we periodically or randomly insert the lumped elements. and its corresponding band structure. The constituents of such a PBG medium are materials with positive refractive index. 3. The secret to the allangle negative refraction effect is that for strongly modulated PBG media the equifrequency surfaces in the kspace can become completely smooth convex surfaces around certain points of symmetry. In fact L = µ0 and C = ε0 yield us the equivalent circuit √ for vacuum with a characteristic impedance for the network of Z0 = µ0 /ε0 377 .Physics of negative refractive index materials 485 √ vφ = ω/β = 1/ LC = 1/ εµ and the group velocity vg = (∂ω)/(∂β) turns out to be the same. 133]. Thus the distributed LC network in the highpass conﬁguration behaves as a waveguide ﬁlled with an NRM. and the size of the convex equifrequency surface reduces with increasing frequency. for example the M point for a square lattice (see ﬁgure 20) for a band of frequencies. At high frequencies. To justify our assumption of a homogeneous medium we would require that the associated periodicity be much smaller than the wavelength of the radiation. Now we note that any practical implementation of such a backward wave structure would involve lumped L–C elements in a network. Negative refraction effect in photonic crystals Although photonic crystal and bandstructure materials strictly cannot be considered to be homogeneous metamaterials. the propagation delays dominate and the medium is a positive index √ medium with n = εµ. √ (3. This is the case of usual positive media where the phase velocity and the group velocity are in the same direction. But there is a regime when this happens to all propagating waves within a certain frequency band.21) √ where ε and µ are due to the distributed capacitances and inductances due to the short interconnecting segments of length d between the lumped elements. at low frequencies. which would bring in its own associated lengthscale. The effective dielectric permittivity and the magnetic permeability at low frequencies for the periodically loaded medium can be approximated as [132] εeff = ε − 1 ω2 Ld . This effect can usually be dismissed as a diffraction grating effect if it occurs only for certain wavevectors. µeff = µ − 1 ω2 Cd . Such situations typically occur near the band edges. This implies that the conﬁguration in ﬁgure 19(b) has Z = 1/ − iωC and Z = −iωL and thereby the propagation constant is 1 β=− √ .5. acting as a host medium. √ is positive. This . it has been pointed out that there is a very interesting negative refraction effect that can take place in photonic bandgap (PBG) media [36. On the other hand. (3. This is so that the group velocity (which is along the Poynting vector). These have a great advantage of a large bandwidth. Thus the phase velocity vφ = −ω2 LC is opposite in this medium and we do have a backward wave in the distributed network.
Therefore. The dotted line is normal to the interface and shows the conservation of the parallel wavevector. that the group velocity and the phase velocity in the PBG medium are never antiparallel as in the case of an isotropic NRM and k · (∂ω)(∂k) 0 in the entire band. however. an allangle negative refraction and a ﬂat Veselago lens were demonstrated numerically by Luo et al in two dimensions [36] and three dimensions [134]. rather than due to a negative refractive index. Note. A schematic is shown in ﬁgure 20 for the refraction process in the case of a square lattice.6. however. Note. Using a PBG medium of airholes in a dielectric. homogenization in the sense of metamaterials is difﬁcult to achieve here. This leads to the fact that most of the radiation may couple mainly along one direction in photonic crystal (along [110] in ﬁgure 20). while the thick lines which are the normals to the equifrequency surface denote the group velocities. Although the free space wavevector may be a few times larger than the unit cell of these PBG materials. means that the group velocity is along the inward normal to these surfaces. The resulting diffraction will appear to be in the negative refraction sense. These frequency regions.486 S A Ramakrishna Γ X M Figure 20. But the PBG media which have allangle negative refraction can be used to make unique photonic devices that can tailor the propagation. however. Nonlinear effects in negative refractive media One of the main aspects of the metamaterials is that the resonant nature of the effects makes the local microscopic ﬁelds in them highly inhomogeneous and fairly intense. 3. 136]. Thus the radiation might appear to stream along one direction [135. The equifrequency surface is a convex surface around the M point and the incoming wave couples to a single Bloch wave in the negative refraction direction. the energy transport in these materials is still very sensitive to the periodicity and structural arrangements. Schematic of the equifrequency surface for allangle negative refraction for a square lattice. although the concept of an effective refractive index might be useful near the band edges particularly with the lowest frequency bands. Further. so that every incoming wave couples to a Bloch wave. that the equifrequency surface of the PBG medium need not be very spherical or circular. This would render the image that should be formed inside the NRM in the Veselago lens unobservable. do have a negative effective mass. then the incoming wave couples to a single Bloch wave in the photonic crystal. If this situation results in the lowest photonic band. The thin lines show the wavevector. Thus the negative refraction effect in the case of PBG materials can have an interpretation entirely in terms of bandstructure effects without resorting to the idea of a negative refractive index. the propagation can be mainly considered as due to Bragg scattering. There is a crucial condition that this equifrequency surface should be convex everywhere and have a larger size than a free space dispersion circle. There can .
then large changes can be expected in response to the applied electromagnetic ﬁelds by virtue of the enhanced electric ﬁelds in the gaps. Consider a metamaterial of SRR with a nonlinear dielectric in the capacitive gaps. it is the intense magnetic ﬁelds in the vicinity of the wires in the case of the thin wire structures that makes them work as a low frequency plasma. and it is called a focusing or defocusing nonlinearity. and this leads to enormously enhanced local electric ﬁelds. The electric ﬁeld across the capacitive gaps is Ec = Vc /dc . In fact. 28]. Hence we obtain an enhancement factor of 2 2 1 1/2ε0 Ec (ω0 )2 1 rτc ωp Q= = 2 (3. Typical numbers for this factor range from 104 to 106 [3. (3. The intense local ﬁeld enhancements also make the SRR very sensitive to any absorbing molecular species in the regions of large ﬁelds. The sign of n2 can be positive or negative. A relation between the incident ﬁeld strength and the nonlinear resonance frequency can be obtained in a quasistatic calculation as [22. where Vc is given by equation (2. Suppose the capacitive gap in the SRR is ﬁlled with a Kerr nonlinear dielectric. the energy is equally distributed between the electric and magnetic ﬁelds.25) 2 2x6 4n2 L2 ω0 c = /ω0 . indicating the potential for use in molecular probes. . SRRs also similarly concentrate the electric ﬁelds in the tiny capacitive gaps. leading to qualitative changes in the nonlinear responses. and hence the resonant frequency. ε > 0) with high reﬂectivity and a positive medium (µ > 0. embedded in the linear material and ωNL is the resonance frequency for the SRR imbedded in the nonlinear material. Use of this mechanism to switch the behaviour of the metamaterial of thin wires and SRR composites from a negative refractive index to a positive index or plasmalike behaviour has been suggested [138]. (3. Consider the single ring SRR with two splits. Phase matching conditions which are crucial in nonlinear optics can be met in a different manner in NRMs. depending on the incident intensity.24) where I = 1/2(ε/µ)2 ε0 cE2 is the intensity of light. respectively. where the factor of 2 accounts for the fact that we have two gaps in our system.23) 2 1/2µ0 Hext (ω0 )2 8c 2γ Lc for the energy stored in the capacitive gaps at resonance. 138] 2 Zd n2 dc (1 − x 2 )[(x 2 − 2 )2 + 2 2 ] Hext 2 = .Physics of negative refractive index materials 487 be immense enhancement of the local ﬁelds in these systems and the potential for nonlinear applications was pointed out in the very ﬁrst paper on the SRR structure [3]. yielding us 2 µ0 ωp rτc . This has been pointed out in the context of second harmonic generation and stimulated Raman scattering [139]. Ultrashort pulse propagation in these media would be interesting in view of the very large dispersion in these media. the nonlinear Kerr effect in a dielectric is always possible [137]. The intense ﬁeld enhancements in metamaterials make this a very fertile ground for nonlinear optics. x = ωNL /ω0 .37). ω0 is the resonance frequency for the SRR where = ω/ω0 . Regardless of the frequency of interest. ε > 0) which can almost be transparent. The material now switches between a negative magnetic medium (µ < 0.22) Ec (ω0 ) = 2γ Lc For the incident radiation. This is a typical nonlinearity where the refractive index of the dielectric depends on the electromagnetic ﬁelds as √ n = ε = n0 + n2 I. The plot of the nonlinear resonance frequency against the ﬁeld strength is shown in ﬁgure 21 and the bistable behaviour is reasonably stable against dissipation. (3. The capacitance of the SRR. now becomes a function of the incident ﬁeld strengths.
Note that the situation for the spolarized mode is just the other way around: the lower curve would describe the spolarized mode for ωmp < ωp and the upper curve for ωmp > ωp . the plasmon (the frequency tends to the electrostatic limit of ωp / 2 from either above (ωmp > ωp ) or (below ωmp < ωp ). Similar results using a resonant form for the magnetic permeability have been obtained in [140. and this topic has received some attention recently [140–147]. We omit dissipation to obtain the dispersion.7.05 180nm 300nm 1 0 10 20 30 field strength (kA/m) 40 50 Figure 21.1 / 0 0NL 24nm 24nm 1. ε+ ε− which gives the dispersion relation kx = ω ε− (ε− − µ− ) 2 c ε− − 1 1/2 (3. For the sake of simplicity we will 2 2 assume the causal plasmalike forms ε2 = 1 − ωp /ω2 and µ2 1 − ωmp /ω2 for the NRM. the two degenerate surface plasmons at the two interfaces get coupled for a thin slab and gives rise to coupled slab modes: a symmetric and an antisymmetric mode. Then there are two branches to the ppolarized modes for ωmp < ωp . the plasmon whose dispersion is given by the upper curve has a negative group velocity in the transverse direction. 143]. (3. At large kx . Next let us examine the nature of the surface modes in a thin slab of NRM. The condition for the existence of a surface plasmon at the interface is [32] kz1 kz2 + = 0. As is well known. the surface plasmon ﬁrst appears at the lightline (ω = ckx ).and spolarized surface modes. The nonlinear resonant frequency versus the ﬁeld strength for two values of the dissipation rate: ﬁlled circles are for γ of silver and the open circles for 3γ . Surface electromagnetic modes in NRMs Any material with a negative material permittivity and permeability permits the existence of surface modes on the interfaces with positive media. NRMs permit both p. We can see that the plasmon dispersions take different forms for ωmp > ωp (the upper curve) and ωmp < ωp √ lower curve). The inner box shows the nonlinear medium just enclosing the SRR structure. 3. ε < 0 and similarly one more above the magnetic plasma frequency. Let us ﬁrst examine the nature of the surface plasmons (for ppolarized incident light) at a single ﬂat interface between semiinﬁnite positive and negative media. The inset shows the SRR structure with the relevant dimensions.26) . also called slab . The corresponding dispersion for the plasmon mode is plotted in ﬁgure 22. In both cases.27) assuming ε+ = 1 and µ+ = 1 for the positive medium.488 S A Ramakrishna 1. one below the magnetic resonant frequency when µ > 0. At small kx .
the dispersion relation for the coupled surface modes can be obtained and we show them in ﬁgure 23.Physics of negative refractive index materials 489 1. Interestingly we can also have waveguide modes in the slab when ε < 0. plasmon polariton modes. . have to tend to the uncoupled plasmon dispersion for a single surface. The condition for these resonances (for ppolarized light) are [32] tanh coth (j ) (2) kz d 2 (2) kz d 2 =− =− (1) ε2 kz (2) kz (1) ε2 kz (2) kz .6 p (1) 0. where again we 2 2 have assumed the dispersive forms ε(ω) = 1 − ωp /ω2 and µ2 = 1 − ωmp /ω2 .0 1. Physically. The dispersion relations are qualitatively different for ωmp < ωp (or µ2 (ωs ) > −1) and ωmp > ωp (or µ2 (ωs ) < −1).0 ll 0.8 mp mp = 0.4 / 0. √ The light line ω = ckx and the ω = ωs = ωp / 2 line are given by the dotted lines. The dispersion for the surface plasmon at an interface between a negative medium and a positive medium when the effects of dispersion are included.0 0.29) 2 where kz = kx − εj µj ω2 /c2 .0 2. is shown by the chain curve (1). Ruppin [144] has derived the Mie scattering .0 Figure 22. ω = ckx / ε2 µ2 with ωmp = 2ωp . at large kx . The dispersion √ inside the inﬁnite negative medium. These new waveguide modes are absent in slabs with only one of ε or µ negative. The dispersion of the surface modes is a function of the geometry of the surface.0 = 0. µ < 0 and the refractive index of the slab is higher than that of vacuum as shown in ﬁgure 23. We ﬁnd that the ﬁnite frequency changes the dispersion from the electrostatic limit (where it is given √ by ω = ωp 1 ± exp(−kx d)/ 2).2 0.28) (3. (3. The surface plasmon resonances can have a great effect on the scattering properties of materials and on the radiation in the vicinity of these surfaces. Ruppin [141] has calculated the dispersions of these slab modes using a resonant form for µ and again obtained two branches. the behaviour of both the symmetric and the antisymmetric modes.0 4.0 kx/kp 3.2 p p =2 mp 0. Using the above conditions. The frequency axis is scaled with respect to the bulk plasmon frequency (ωp ) and the kx axis is scaled with respect to kp = wp /c.
In this section.3 kpd=0. The perfect lens One of the most debated consequences of negative refraction is the possibility of a perfect lens whose image resolution is not subject to the limitations of the traditional diffraction limit of a wavelength.0 kx /kp 8. The part of the lower branch with the symbol shows the nonradiative light modes within the slab. (a) µ = 1 − ωmp mp = 2ωp . Nearﬁeld information and the diffraction limit One principle that is taught in optics courses even at the freshman level is the diffraction limit. and this does somewhat curtail the subwavelength focusing capabilities of the lens. is shown by (— · —). to quite some extent.0 6.0 x p 0. The substantial changes to the spontaneous decay rates in the vicinity of an NRM has also been pointed out. The dispersion in the inﬁnite negative √ medium. Subwavelength image resolution.3 0. 4. we will discuss this phenomenon of focusing the nearﬁeld components and will address several of the muchdebated issues.8 p kpd=0.0 10.0 6.0 0. The latter components are associated with the spatial features of the source at a subwavelength lengthscale and the focusing action for the evanescent waves is accomplished through the surface states that reside on the NRM surfaces.0 0. coefﬁcients for an NRM sphere and obtained new resonances corresponding to the negative magnetic permeability.4 0.0 0.0 4.5 / light modes / 2. It turns out that Veselago’s ﬂat slab of NRM with ε = −1 and µ = −1 not only focuses the propagating radiations as shown by a ray picture. (b) µ2 = +1. First let us reexamine it here.0 10. The dispersion for the coupled surface plasmons when the effects of retardation are included.4 0. Klimov [145] has also discussed the new TM scattering resonances due to a negative µ as well as the whispering gallery modes due to a negative refractive index in the scattering properties of an NRM sphere.0 4.490 S A Ramakrishna (a) 1. ω = ckx / ε2 µ2 . The frequency axis is scaled with respect to the bulk plasmon frequency (ωp ) and the kx √ axis is scaled with respect to kp = wp /c.6 0. Indeed the proposal for a perfect lens [6] has provoked heated debates and caught the imagination of the scientiﬁc community at large.2 kpd=0.8 p (b) 1. This result has now been generalized to several conﬁgurations of NRMs [149]. but also focuses the evanescent nonpropagating components of radiation that are usually conﬁned to the immediate vicinity of a source. Consider an extended source that emits or scatters radiation. The lightline ω = ckx and the ω = ωp / 2 line are 2 /ω2 with ω given by (· · · · · ·).0 0.2 0.6 0.0 x p 2. It turns out that the focusing of the evanescent components is quite sensitive to imperfections in the NRM. which we will henceforth refer to as superlenses. .0 kx /k p 8.1.0 Figure 23.0 0. 4.0. is possible even with these imperfect lenses. Similarly the Mie scattering properties of a cylinder [148] made of an NRM have revealed the presence of new resonances in the NRM frequency band. that we can never resolve features in an image to better than the order of a wavelength [35].
Note that kx and ky represent the Fourier components of the spatial variation in the source in the corresponding directions. while the light comes out at large angles for a periodic object with small spatial periods (b). ky ) exp[i(kx x + ky y + kz z − ωt)]. y. The source can have intensity or ﬁeld variations over arbitrarily small distances. Thus the waves with large kx > k0 and ky > k0 are evanescent and decay in amplitude exponentially away from the source z = 0 plane. This is precisely what is done in nearﬁeld optical microscopy [150. z.Physics of negative refractive index materials (a) z y x 491 (b) 1 2 ∆ y1 ∆ y2 Figure 24. In conventional imaging systems.y < λ/2π . These are the nearﬁeld modes of the radiation of the source and have a decay length of 1/kx. these modes are never detected as a consequence. equation (4. kx ky (4. The corresponding transverse wavevectors kx and ky will then be very large. y. to be the minimum limit on the spatial resolution in the image. t) = 1 2π 2 dkx dky E (kx .y can literally be on atomic lengthscales. For the time being assume it to be a planar object on the z = 0 plane and the optical axis for imaging along the z axis as shown in ﬁgure 24. This does not violate any fundamental principle (such as the Heisenberg . where x and y are some spatial periods of the variation. A periodic object with large spatial period scatters the light through small angles (a). where a ﬁne ﬁbre tip is used to couple the nearﬁeld modes into the propagating modes in the ﬁbre and detected. ky ) = x y dx dyE(x. Radiation is scattered and emitted by an object. 151]. cannot be satisﬁed without making the associated kz imaginary. consider the case of isolated molecules on a surface emitting radiation. As an example.2) is the Fourier transform of the spatial variation of the source. 0) be the associated electric ﬁeld. 0) exp[−i(kx x + ky y)] (4. There is no restriction on that. y. Let E(x. and we obtain the corresponding spatial frequency of the wavelength. Now kx = 2π/ x and ky = 2π/ y .3) kx + ky + kz = ε0 µ0 2 = k0 c in free space. The largest propagating wavevector that contributes to the image is k0 (assuming unit numerical aperture).1) where ω is the frequency of the radiation and E (kx . in which case the x. However note that this limitation can be sidestepped if the nearﬁeld modes of the source can be sampled. Now the electromagnetic ﬁelds over all space can be written down as E(x. The Maxwell equations impose the condition that ω2 2 2 2 2 (4. Hence the corresponding source information about the fast varying features of the source is missing. The angle at which light comes out depends on the spatial features of the source. λ.3). in which case the dispersion equation.
k = 1.7) (kz2 /µ− ) + (kz3 /µ+ ) 2(kz1 /µ+ ) 2(kz2 /µ− ) t21 = . (4. 2. 2. k = 1. Although the transverse image resolution might be very small. The essential higher dimensional (>1) nature of the problem makes this possible [153]. respectively. 3) and transmission coefﬁcients (tj k ∀ j. Consider the source to be at z = 0 (the object plane): we wish to calculate the ﬁelds at z = 2d (the image plane).4) 1 − r12 r21 e2ikz2 d r21 + r32 e2ikz2 d R= . . 3) at the interfaces (see ﬁgure 25). and for evanescent waves we will choose the positive sign.5) 1 − r12 r21 e2ikz2 d In this geometry.12) c2 ε− µ − ω 2 2 2 kz2 = ± − kx − ky . (4.6) . For propagating waves we will choose the negative sign. r12 = −r21 .8) (kz1 /µ+ ) + (kz2 /µ− ) (kz2 /µ− ) + (kz3 /µ+ ) where µ+ and µ− are the permeability of free space and the NRM. t32 = .9) . the corresponding transverse component of the wavevector is arbitrarily large. we ﬁnd t21 t32 eikz2 d . (4. we have independent solutions for the ppolarized light (E ﬁeld in the plane of reﬂection) and the spolarized light (H ﬁeld in the plane of polarization). (kz1 /ε+ ) + (kz2 /ε− ) (kz2 /ε− ) − (kz3 /ε+ ) r32 = . The zcomponents of the wavevector are T = ε + µ+ ω 2 2 2 − kx − ky = kz3 . and apply appropriate boundary conditions on the ﬁelds at the interfaces. First let us calculate the reﬂection and transmission coefﬁcients for a plane wave incident on the NRM slab8 : writing the solutions in terms of the partial reﬂection (rj k ∀j.10) (kz2 /ε− ) + (kz3 /ε+ ) 2(kz1 /ε+ ) 2(kz2 /ε− ) t21 = . (4. r12 = −r21 .11) (kz1 /ε+ ) + (kz2 /ε− ) (kz2 /ε− ) + (kz3 /ε+ ) where ε+ and ε− are the dielectric permittivity of free space and the NRM. The same solution results.1. t32 = . (4. Or one can take the eigensolutions for the ﬁelds inside and outside the slab.13) c2 where we have to choose the component of the wavevector kz2 in accordance with the arguments of section 3. The partial reﬂection and transmission Fresnel coefﬁcients for the spolarized light are (kz1 /µ+ ) − (kz2 /µ− ) r21 = (4. respectively.2. (4. Pendry’s proposal Consider Veselago’s lens consisting of a slab of NRM of thickness d with ε = −1 and µ = −1 and surrounded by vacuum as shown in ﬁgure 25. (kz1 /µ+ ) + (kz2 /µ− ) (kz2 /µ− ) − (kz3 /µ+ ) r32 = . irrespective of the manner of calculation. Similarly. (4. kz1 = 8 This can be done in several ways: one can use partial reﬂections from the interfaces and sum them up.492 S A Ramakrishna uncertainty principle [152]). (4. the partial reﬂection and transmission Fresnel coefﬁcients for the ppolarized light are (kz1 /ε+ ) − (kz2 /ε− ) r21 = (4. 4.
When the NRM has ε− = −ε+ = −1 and µ− = µ+ = −1. Note that the total transmission and reﬂection coefﬁcients for the slab are well deﬁned in spite of being the sum of a geometric series (of the partial waves [6]) whose terms can have magnitude greater than unity. Hence for the slab ε− →−1 µ− →−1 lim T = e−ikz1 d . but can be justiﬁed on grounds of analytic continuity: that the validity of the sum of an inﬁnite series transcends the divergence of the series itself. The calculation has been questioned on these grounds [154]. the reﬂection is zero due to the impedance matching. This result was demonstrated by Pendry in his now classic paper [6]. we obtain trivially for propagating waves that kz2 = −kz1 . Thus we have a lens that is unaffected by the diffraction limit as it includes all the components of the nearﬁeld too and hence it is a perfect lens. ε− →−1 µ− →−1 lim R = 0 (4. Further. bringing them both to a focus at the image plane. Thus we have the same results for the propagating waves as those yielded by a ray analysis. However.e. The partial reﬂection/transmission coefﬁcients across the boundaries are shown.15) i.14) and this clearly shows that the total phase change for propagation from the object plane (at z = 0) to the image plane (at z = 2d) is zero. µ− →−1 ε− →−1 lim R = 0. but it also restores the amplitudes of the evanescent components. provided that the sum was carried out to include all inﬁnite terms and there . Thus we have that result the not only does the Veselago lens cancel the phase accumulation for the propagating waves. the slab actually increases exponentially the amplitude of the evanescent wave at the same rate by which it decays in free space. The object. The restoration of the amplitude of an evanescent component is depicted schematically. 2 2 Next consider the evanescent waves with kz1 = i kx + ky − ε+ µ+ ω2 /c2 = iκz : then kz2 = kz1 and the partial coefﬁcients tj k and rj k diverge. (4. is focused on the other side of the slab. The perfect lens system consisting of a slab of NRM. the transmission and reﬂection coefﬁcients for the slab are still well deﬁned in this limit. and tj k = 1 and rj k = 0 due to the matched impedance. and can even diverge in the case of evanescent waves.Physics of negative refractive index materials 493 object plane image plane r21 t2 1 t 12 =1 =1 r3 2 t 32 = –1 = –1 =1 =1 z=0 z=d/2 z=3d/2 z=2d Figure 25. at a distance d/2 from the surface of the NRM. Thus the net amplitude change at the image plane (at z = 2d) from the object plane at z = 0 is zero (see ﬁgure 25. ε− →−1 µ− →−1 lim T = e+κz d .
1 for consistency. . 4. ε− = −ε+ and µ− = −µ+ . The evanescent wave ampliﬁes in the region between the two potentials. the unphysical divergences go away when one has even an inﬁnitesimal level of absorption in the NRM. Further note that the transfer matrix for the slab [35] is actually independent of the sign of the wavevector. there is always a cutoff with metamaterials that corresponds to the inverse of the lengthscale of the structure of the metamaterial. In fact. This a manifestation of the pathological nature of the perfect lens and is because of the lack of a large momentum cutoff [159]. Particularly at optical frequencies.494 S A Ramakrishna are no approximations [155].2. where there is an increase in energy. However. The denominator of these coefﬁcients going to zero is the condition for the homogeneous equation to have a nontrivial solution. The perfect lens solutions are the exact conditions for the surface plasmons of both electric and magnetic nature on a surface of a semiinﬁnite NRM. ε− = −1 + iδ and µ− = −1 + iδ. 158]. the absorption in the material limits the ability to amplify the evanescent waves. but we choose the sign of the square root in accordance with section 3.3.e. kc ∼ 1/a. It has been pointed out that the surface states on photonic bandgap media in the ﬁrst Brillouin zone can also be used for the imaging of the nearﬁelds [157]. this ampliﬁcation of the incident evanescent waves happens for the Schr¨ dinger equation as well. Consider two identical separated localized potential wells o (say a δ or a square well) in one dimension with the incident evanescent wave having the energy of one of the bound states of the wells. These are the steady state solutions of Maxwell’s equations and give the ﬁeld distributions essentially at inﬁnite time. as pointed out by Pendry [6]. The divergence of the partial reﬂection and transmission coefﬁcients for the interfaces was just a manifestation of these resonances. The process of restoring the amplitudes of the evanescent components. If we consider the case where all lengthscales in the problem are much smaller than the 9 Note that the solution ampliﬁes with distance and not with time. as the Poynting vector associated with the evanescent waves in a lossless medium is strictly zero (see [156] for an elementary review). In the limit of large wavevectors (kx → ∞) the ampliﬁcation of these evanescent waves leads to a divergence in the electromagnetic energy [8.1. In fact. The quasistatic limit and the silver lens. The large associated energy density in the NRM is obtained from the source itself and is built up over some time. particularly those associated with large transverse wavevectors and hence the resolving capabilities of the NRM lens [8. In fact. are satisﬁed. ε− = −1 and µ = −1. We note that the perfect lens works when the conditions on both the permittivity and the permeability. this problem is more acute due to the lack of magnetic materials. henceforth called ampliﬁcation by us9 does not really involve any energy transport. 160]. This ampliﬁcation is distinct from the more usual ampliﬁcation related to laser gain or ampliﬁers in electronics. however. From our discussion in section 2. It is these resonances that are responsible for the restoration or ampliﬁcation of the evanescent nearﬁelds of the source. In principle. The ideal lens conditions. That means that the ﬁnal solution is also independent of the sign of the wavevector. All calculations should be carried out by adding an inﬁnitesimal iδ to ω. This will be investigated in more detail in section 4. which essentially sets its own large momentum cutoff. i. are very singular. This is a generic effect of a localized resonance for the Helmholtz equation—the transmission and reﬂection coefﬁcients diverge. It is in the optical and ultraviolet frequencies that the prospect of imaging the nearﬁeld modes becomes most exciting. Thus. Thus we would be severely handicapped by the lack of materials. it is clear that to generate media with both ε and µ negative at the same frequency is quite difﬁcult. we could have obtained the solution for the evanescent waves directly from that of the propagating waves by analytic continuation. there is one great simpliﬁcation that can be afforded. kz → iκz . and the limit δ → 0 should be taken to preserve causality.
4.9)–(4. The calculations have been performed in the electrostatic limit.8) become independent of the dielectric constants and the Fresnel coefﬁcients for the ppolarization given by equations (4. Similarly a slab with negative dielectric permittivity (ε− = −ε+ ) would act as a nearﬁeld lens for the ppolarized waves.11) become independent of the magnetic permeabilities.2. are clearly resolved by the lens while they cannot be resolved when at the same distance in vacuum. ω in electronvolts). then we have kx. ε− 2 kx − ε− µ− ω2 /c2 = 0. In this limit.4 has been used for silver. 159]. The perfect lens conditions ε− = −ε+ are exactly the conditions for the surface plasmons on a metal surface and µ− = −µ+ are similarly the conditions for the spolarized surface modes of a magnetic nature.7 − 92 /ω2 + i0. µ− (4. 142.16) (4.2. Thus. a slab with negative magnetic permeability (µ− = −µ+) would act as a nearﬁeld lens for the spolarized waves. In the special case when both these conditions are satisﬁed at a single frequency ω. The image of two slits separated by a small distance imaged by a slab of silver is shown in ﬁgure 26 in this limit. limited by the large levels of dissipation in silver (at UV–visible frequencies. The role of surface plasmons. and a value of ε− = −1 + i0. the role of the surface plasmons is crucial to the action of the perfect lens [6. one chooses a highly conducting metal such as silver. the conditions for the surface modes on a semiinﬁnite NRM. however. The two slits. Noting that dissipation affects the capability to resolve large wavevectors. The resolution is.Physics of negative refractive index materials 495 Figure 26. the Fresnel coefﬁcients for the spolarization given by equations (4. The conﬁguration used is shown on the lefthand side.y k0 = ω/c and kz kx in both the positive and negative media. These dispersionless modes have a zero group velocity along the surface. Thus we can use a slab of metal as a nearﬁeld lens for the ppolarized radiation at a frequency when ε− = −1.17) hold for all kx and all the surface plasmon modes (of both polarizations) become degenerate at ω. and the ﬁelds of the surface mode . 4. the dielectric permittivity of silver is reasonably described by ε(ω) 5. 158. placed 100 nm apart (subwavelength distance). This is the extreme nearﬁeld limit or the quasistatic limit. As has been pointed out.6)–(4. The subwavelength resolution capabilities are clearly visible. A thin ﬁlm of silver can act as a nearﬁeld lens. wavelength. The total ﬁeld can be written as a sum of the ﬁelds due to the source. 2 kx − ε+ µ+ ω2 /c2 + ε+ 2 kx − ε+ µ+ ω2 /c2 + µ+ 2 kx − ε− µ− ω2 /c2 = 0.
496 S A Ramakrishna Figure 27. excitations for any wave vector iκz . ε1 = −ε2 only. B = −e is the contribution from the surface plasmon at the left surface z = d1 and C = e−κz (d2 −2d) is the ﬁeld of the surface plasmon at the right surface z = d2 .20) exp(−ikz2 d). The asymmetric lens. and (bottom panel) matching the perfect lens conditions on the nearside. and the reﬂection coefﬁcient is clearly nonzero. In either case we have the transmission coefﬁcient (for the ppolarization) through the slab to be 2(kz1 /ε1 ) T = (4. E(z) = A e−κz z + B e−κx z−d1  + C e−κz z−d2  . The ﬁgure shows the three cases of (top panel) ε1 = ε3 = −ε2 —the perfect lens. −κz d1 (4. it is easily veriﬁed that wavevectors with different kx will refocus at slightly different positions except in the quasistatic limit kx → ∞. On the left side of the slab. (kz3 /ε3 ) + (kz1 /ε1 ) clearly showing the phase reversal for the propagating wave and ampliﬁcation for the evanescent modes.2. and thus the image intensity is changed. Consider the geometry shown in ﬁgure 27 with the negative medium between two positive media of different permittivity and permeability. ε− = −ε+ . the ﬁelds of the two surface plasmons again exactly cancel (for zero reﬂectivity).10 The presence of the other interface detunes the surface plasmon resonance on each interface so that they are excited to the correct degree so as to make the ﬁelds of the surface plasmon exactly cancel the incident ﬁelds for z > d1 . 10 d1 = d/2 and d2 = 3d/2 in our case. ε3 = −ε2 only. 4. however. we term the asymmetric slab a nearperfect lens [142]. Further. It is this coherent action of the surface plasmons that is responsible for the perfect lens action. There is also an impedance mismatch. the ﬁeld strength at the image plane differs from the object plane by a constant factor.18) where A = 1 is the evanescent ﬁeld of the source at z = 0. In this asymmetric case. But it is true as long as d2 + d1 = 2d. The asymmetric lens system consists of a slab of NRM with dielectric media with different ε on either side.19) be satisﬁed at both the interfaces to enable the ampliﬁcation of evanescent waves. (middle panel) matching the perfect lens conditions on the farside. Hence. .3. Thus there is no unique image plane and the focus of this lens is aberrated. It is actually not necessary that the perfect lens conditions of µ− = µ+ (4. It is sufﬁcient if the plasmon conditions on both ε and µ are satisﬁed at either of the interfaces.
thereby causing some degradation of the image [161]. the spatial variation of the ﬁeld is completely different as shown in ﬁgure 27. The larger the absorption.3. given the spatial dispersion. the dielectric constant becomes dependent on the wavevector at large wavevectors and becomes spatially dispersive. The effects of absorption can be included by including the imaginary part of ε− and µ− in the calculations. Foremost is the effect of absorption which is maximal in the regions of large electromagnetic ﬁelds. (4. In metamaterials. within the random phase approximation (RPA). Particularly in the case of the silver lens. 142. it does not imply that the ampliﬁcation process would breakdown completely even in the presence of inﬁnitesimal absorption as has been claimed [8]. In the case of metamaterials. The exponential growth of the evanescent waves implies extremely large ﬁelds at the latter interface. Although the RPA holds until there is sufﬁcient screening and is valid for a lengthscale of a few atomic sizes. However.21) ω 1 − 3/5(k · vf /ω)2 which shows a quadratic wavenumber dependence. Limitations in real materials and imperfect NRMs It should be emphasized that there is an inherent large momentum cutoff in all real materials. The presence of a large reﬂectivity has serious consequences for the use of a lens for nearﬁeld imaging applications as it would disturb the object ﬁeld. The technology for the deposition of these ﬁlms on substrates is well established and the deposited ﬁlm will certainly be more robust. which can also lead to some extra dispersion on the surface modes. Consider the transmission coefﬁcient for the ppolarized light across the slab. the perfect lens effect turns out to be rather sensitive to any deviations from the lens conditions ε− = −ε+ and µ− = −µ+ [15.22) (kz1 /ε+ + kz2 /ε− )2 − (kz1 /ε+ − kz2 /ε− )2 exp(2ikz2 d) . and an inﬁnite resolution as our discussion above indicates is actually not possible. (4. where a is the lengthscale of periodicity of the metamaterial structures. The very large absorption that would occur implies that the source would no longer be able to maintain the energy supply and the ampliﬁcation of the very large wavevectors would not be possible. we would not be aiming to resolve atoms or small molecules with visible or IR light. 4(kz1 /ε+ )(kz2 /ε− ) exp(ikz2 d) Tp (kx ) = . the Lindhard form [162] for the dielectric permittivity of a metal at small wavevectors k kF where kF is the wavevector at the Fermi surface. In general. the smaller the corresponding cutoff wavevector. 4. Thus they represent a deviation from these conditions in the imaginary parts. the boundaries of the slab are also indeterminate on the lengthscale of a unit cell. this is even more acute as µ− = 1 represents a very large deviation from the perfect lens conditions. it makes possible the use of a deposited silver thin ﬁlm on a solid dielectric substrate. However. 2 ωp 1 ε(ω. it is clear that once the radiation can probe the structural details. Thus absorption sets its own large wavevector cutoff. For the silver lens. yields. 163–167]. k) = 1 − 2 . For example. this idea does render the technology of making these lenses easier. Even with metals. effective medium parameters make no sense. The reﬂection in the case of ε2 = −ε1 is also ampliﬁed.Physics of negative refractive index materials 497 A similar result holds for the spolarized wave incident on the slab. Thus there is a natural cutoff on the transverse wavevector of kx < kc = 2π/a. there are other mechanisms that limit the ability of this lens to focus the waves with large transverse wavevectors. Much before this limit sets in. We note that while the transmission is the same in both the cases. which is far preferable to a freestanding slab of silver with a subwavelength thickness.
δε. quite different.1 + i0.35. (4) Im(ε) = 0. The direct excitation of these surface modes for kx d = − ln . k0 d = 0.01. µ = −2. The deviations in the real part on the other hand allow resonant excitations of the slab plasmon modes discussed in section 3. These resonant excitations of the slab would imply a very large transmission (for evanescent waves T > 1 is allowed) as shown in ﬁgure 28. Some extra subwavelength information may be available from the smaller contributions at larger wavevectors. (1) Re ε = −1 and (1) Im(ε) = 0. k0 d = 0. However. In the quasistatic limit (kx ω/c) we ﬁnd that the two terms in the denominator are approximately equal when δε . µ = −1. µ = +1 and (1) ε = −1 + i0.01.1.9 + i0. with the same expression for the image resolution. k0 d = 0.35. showing the transmission resonances caused by the resonant excitations of the slab plasmon polaritons. The absolute value of the transmitted ﬁeld at the image plane T exp(ikz d). for the ppolarization as a function of kx .01.35.35.0. implying that the perfect lens effect is very sensitive indeed to the material imperfections. k0 d = 0. k0 d = 0. µ = −1.5. Deviations in ε affect the ppolarization more than the deviations in µ. the ﬁrst term is larger than the second and hence the exponential decay in the numerator asserts itself. say). and we obtain − ln δε/2 λ λ = .35. could be in the imaginary part also.5 and (5) ε = −1. however. (3) Im(ε) = 0. k0 d = 0. (4.35.1.23) 2 For larger wave vectors.01. The absorption mainly causes an exponentially decaying transmission for large wavevectors. k0 d = 0.35. Under the perfect lens conditions. k0 d = 0. (4. The detailed effects of absorption and the deviations in the real part of µ and ε which introduce retardation effects are.1. On the right panel. k0 d = 0.7. (2) Im(ε) = 0. The absorption broadens and softens the resonances.5 from this expression.498 S A Ramakrishna 10 2 10 10 T 10 10 1 0 (1) (2) (3) (4) (5) 10 2 10 10 T 10 10 1 0 (1) (2) (3) (4) 1 1 2 2 10 3 0 10 k x/k 0 20 30 10 3 0 10 20 k x/k 0 30 40 (1) Figure 28. (2) ε = −1 + i0.01. (3) ε = −0. (4) ε = −0.01.24) res = 2π d min Thus the resolution depends logarithmically on the deviations of the material parameters and inversely with the width of the slab. then the ﬁrst term is no longer zero. we obtain a minimum resolved feature of about λ/2. We can deﬁne the subwavelength resolution as the ratio of the optical wavelength to the linear size ( min ) of the smallest resolved feature. if there was a mismatch of the material parameters (ε− = −1 + δε and ε+ = +1. On the left panel. The deviation. µ = −1. We can take this to be the largest wavevector for which there is an effective ampliﬁcation. Hence for our silver lens with a thickness of onetenth the wavelength.9 + i0. the ﬁrst term in the denominator goes to zero for the evanescent waves and the exponential in the second term decays faster than the exponential in the numerator.
In view of the controversy surrounding the effect at that time. In the case of the asymmetric lens. yet. However. 4. the presence of a small and ﬁnite absorption makes the focus a small blob instead and the Poynting vector also becomes well deﬁned. Thus some information is lost due to the loss of some propagating modes. 172]. in the favourable case of low reﬂection. Some calculations obtained focusing [169. In the limit of large ε3 the resolution is actually enhanced [142].25) λmin 4πd (assuming ε3 ε2 . Cummer [173] pointed out that the discrete ﬁelds do not obey the same dispersion relations as the continuous ﬁelds do. The perfect lens solution in the absence of losses is very special. Thus any discrete method has a wavenumber dependence and an inbuilt higher momentum cutoff that automatically precludes the perfect . we note that the resolution works out to be ln ε2 /2ε3 λ0 λ0 res = =− (4. can have other effects. But such cavity effects will be extremely limited in the presence of absorption. it was felt that some degree of numerical veriﬁcation based on exact simulations of Maxwell’s equations was necessary. 170] while some others [30] did not show any focusing at all and yet others obtained focusing but not subwavelength image resolution [171. and ε3 ε1 ). the existence of these resonances is essential. but the nearﬁeld information is still preserved. Numerical simulations and time evolution After the demonstration of the perfect lens effect by Pendry. and can lead to paradoxical interpretations. however. 70]. Absorption actually damps these transmission resonances and makes possible the very lensing action with some subwavelength resolution capabilities in the case of the silver lens with µ = +1. consider the discretization of the ﬁelds in the transfer matrix method [17. y. The plasmons on the surface of the ﬁnitely wide slab can form standing waves along the transverse directions with the ﬁnite slab acting as a cavity [168] and signiﬁcantly distort the image formation. one could ask if a ﬁnite aperture for the slab of the NRM would affect the nearﬁeld imaging process. The very large density of states near the corners can distort the nearﬁeld image but will not affect it unless the transverse slab size becomes comparable with the size of the image features. when the discretization is indeed very small compared with the inverse wavevector and they become identical only in the limit of the discrete intervals going to zero. The ﬁnite width of the slab.e.4. The dispersion for free space along one of the lattice directions (for a cubic lattice) for the discretized equations is ω2 (kj ) = c2 4 sin2 a2 1 2kj a 2 c 2 kj 1 + 1 2 2kj a 2 . as the recovery of the evanescent modes can be seen as the result of driving the surface plasmon far offresonance.26) where j = x. For example. z and a here denotes the lattice spacing of the discretization grid. a slab with ﬁnite transverse width would prevent some propagating waves at large angles to the imaging axis from being lost. as these resonances will be disproportionately represented in the image. Evidently this converges to the continuum free space dispersion when kj a 1. Finally. But the transport of the evanscent components will not be affected in the same manner and contribute substantially to the image. (4. there were several attempts to model this effect numerically.Physics of negative refractive index materials 499 imaging applications is undesirable. i. one was attempting to verify an exact analytical solution with numerical simulations which are necessarily approximate due to the ﬁnite differencing. The Poynting vector at the location of a point focus is strictly illdeﬁned. To a ﬁrst approximation. On afterthought though.
which corresponds to the noneffective ampliﬁcation in the case of the NRM slab. In the case of the perfect lens.30) in the quasistatic limit and A(t) 1/2( ωkx t)2 . 2 4 (ω0 − ω2 + iγ ω)2 − kc xr = 2 −kc f0 2 4 (ω0 − ω2 + iγ ω)2 − kc (4. This probably explains the report of [30]. The equations of motion for such a system of oscillators are ¨ ˙ 2 xl + γ xl ω0 xl + kc xr = f (t). ¨ ˙ 2 xr + γ xr ω0 xr + kc xl = 0. where ωkx is the frequency spacing between the two slab plasmon polariton modes at a wavevector kx . A calculation based on Laplace transform methods [163] indicated that for a source that is sharply switched on at t = 0. This corresponds to the zero reﬂection in our case. and the image resolution has been shown to be consistent with the estimate in section 4. and under ideal conditions reﬁnes to a point focus at inﬁnite time.28) where xl and xr are the amplitudes for the left and right oscillators and kc is the coupling between them. and dissipation is always introduced into the material parameters in any such calculation to reach a steady state in a reasonable amount of time.27) (4.3. Note that the slab plasmon polariton modes are decoupled from the singe plasmon frequency. the different dispersion is acutely felt. (4. As seen in section 3. In the steady state the solutions for a timeharmonic excitation f0 exp(−iωt) are xl = 2 (ω0 − ω2 + iγ ω)f0 . Cummer [173] also pointed out that the resonant surface plasmon modes at the interfaces for n = −1 are always excited by any causal incident wave or pulse with a ﬁnite bandwidth. the focusing is very sensitive to meeting the prescribed conditions on ε = −1 and µ = −1. the . The left oscillator does not move at all and all the excitation energy is passed on to the right oscillator. As brieﬂy noted before. GomezSantos [158] has pointed out that the resonant excitation of the plasmons on the two surfaces is identical to the problem of coupled identical oscillators with an external force applied on one of the oscillators. The transmitted ﬁeld across the lossless slab then has an amplitude [158] Et (t) = (t)A(t) exp(−iω0 t) exp(kx d) (4. the temporal evolution of the perfect focus is interesting as the focus reﬁnes in time.7. and it was found to be only remedied by resorting to a larger degree of discretization [149]. From this model we can further consider the temporal evolution of the imaging process by considering the rate at which energy is transferred between the two surfaces.500 S A Ramakrishna focus. The level of subwavelength focusing that can be obtained by a numerical calculation is a function of the ﬁniteness of the differencing scheme even under the perfect conditions of no dissipation. First note that forcing the system at ω0 with no damping makes only xr nonzero in the steady state. The loss also damps out the ampliﬁcation for large wavevectors. the lens forms a steady image only at times that are of the order of the absorption time (t ∼ 1/Im(εµ)ω). 175] which used different boundary conditions to simulate evanescent waves. where it was found that the FDTD simulations never reached a steady behaviour for nonlossy NRMs. Thus a source with a sharp onset has a frequency bandwidth and would excite them. In an FDTD calculation. These conclusions have been conﬁrmed by FDTD calculations [174.29) and it can be seen for the case of nonzero dissipation at resonance that the left oscillator is also excited. which is what we obtain with the timeharmonic solutions. as has been pointed out. The different dispersion for the discrete ﬁelds is equivalent to having slightly different material parameters which thus leads to quite imperfect focusing as some have reported. Particularly for large wavevectors. For large dissipation the right oscillator is no longer excited. these resonances at the two interfaces would ring indeﬁnitely.
5. a change in the geometry enables magniﬁcation of even the nearﬁeld image. The large ampliﬁcation of the nearﬁeld modes causes large amounts of dissipation. δε must be no greater than ∼10−11 ! However. for λ/d = 10. This concept of complementarity has led to the generalization of the Veselago lens to complementary media [13. and we call such lenses with some degree of subwavelength image resolution superlenses. if we had perfect negative refractive index materials with the speciﬁed material parameters. Any deviations from these lead to a drastic reduction in the possible subwavelength resolution of the lens.3. if λ/d = 3. Negative refractive media are the optical analogues of antimatter. we ﬁnd that to achieve a resolution factor of 10. Note that it is the ratio λ/d (wavelength to slab thickness) that dominates the resolution. δε can be as large as ∼0. which does not allow restoration of the nearﬁeld modes. minimizing the effects of absorption is crucial in order to make these lenses work. in this case it is more related to the fractional deviation from the perfect lens conditions that is reduced when the magnitude of the real part of the dielectric constant is very large. Solymar and coworkers [164] have considered that moving the image plane closer to the lens can actually compress the image. 5. 180.1. the slab plasmon polariton modes would ring indeﬁnitely. 149] as well as other geometries including the cylindrical [177. in the sense that passage through NRMs nulliﬁes the effect of passing through positive media with ε and µ of equal magnitude [149] for radiation. Also note that without absorption. Designing superlenses As discussed in the previous section. just by choosing the asymmetric conﬁguration with a large dielectric constant on the side of the image we can increase the resolution of the slab lens [142]. Although it appears similar to the concept that is used in immersion lens nearﬁeld imaging [183].Physics of negative refractive index materials 501 spacing between the slab plasmon modes monotonically decreases and is exponentially small at kx . As mentioned at the end of section 4. Overcoming the limitations of real materials The main limitation of the present day NRMs is the large amount of dissipation in them. subwavelength resolution is still possible to some extent. In fact. In fact. never allowing the formation of a steady image. Interestingly. the perfect lens is in principle possible. Thus the image is initially illdeﬁned and becomes better and better with the passage of time until it reaches the optimal resolution that is possible. . Thus the large wavevector components are very small at short times and a cutoff time can be deﬁned by ωkx tkx ∼ 1 for every wavevector kx before which it does not appreciably contribute to the image formation. We will discuss these ideas in this section. Thus. 178] and the spherical [179] geometries. it has also been shown [142] that working at a slightly different frequency compared with when ε = −1 can lead to the resonant excitation of slab plasmons which can also help increase the maximum possible resolution. It turns out that the ﬂat slab lens of Veselago is only one lens of a whole class of perfect lenses or superlenses that are possible. it is absorption that is the main culprit responsible for reducing the resolution. given the levels of absorption in the system. Restructuring the lens to reduce these effects is possible and we ﬁrst discuss some of the strategies that have been reported. For example. Inspite of the imperfect NRMs. we just obtain some subwavelength image information. the logarithm term being a relatively weakly varying function.002 and we can still achieve the same resolution. But eventually it does appear that the composite structures should incorporate media with an active gain (as in a laser) in order to counter the dissipation [14. 181]. and as we saw in the case of the silver lens.
we end up with a very unusual anisotropic effective medium as δ → 0. Noting that the wave . Ez = 1 −1 −1 (ε Dz + ε− Dz ) = (ε0 εzz )−1 Dz . the entire layered system may be regarded as an effective medium.2) where the continuity of the normal component of the D ﬁeld is implied. If this process of slicing into thinner and thinner layers is continued.4) and the effective medium is an unusually anisotropic medium. One can reduce the effects of absorption by reducing the slab thickness. Similar relations can be derived for the magnetic permeability using the continuity of H and B⊥ . however. reducing the absorption and enhancing the image resolution. and in the limit of the layer thickness becoming extremely small. chop it into thin layers and distribute it around between the source and image planes without affecting the result of perfect imaging [164. we can just take the original thick slab. the corresponding dissipation is much lower. The ﬁeld distributions for a system of ﬁve such pieces is depicted. 2 2 →∞ εzz = −1 −1 (ε+ + ε− ) εxx = (5. 182]. 2ε0 + (5. Similarly. This does.1) 2 where the tangential component of the E ﬁeld is used. we have the effective medium dielectric permittivity components (ε+ + ε− ) → 0. the layered system merely transfers the image of the source from one edge to the other in the sense of an optical ﬁbre bundle. Under the perfect lens conditions. If one considers the response of the system for an electric ﬁeld applied parallel to the layers. considering the response of a ﬁeld applied normal to the layers. Chopping the slab of NRM into thin slices and distributing them around does not affect the image. (5. Then the evanescent ﬁelds amplify to much smaller extents. make an enormous difference to the dissipation in the system. We note that in the ideal lossless case.3) (5. In the limit of a very small layer thickness and a large number of layers (δ → 0 as shown in ﬁgure 29(b)). We note that the source–lens edge and the lens edge–image plane distance have considerably reduced. one obtains ε0 Dx = (ε+ Ex + ε− Ex ) = ε0 εxx Ex . The perfect impedance matching makes this possible and the evanescent ﬁelds amplify within the NRM layers and decay in the positive layers as shown in ﬁgure 29. As the ﬁelds nowhere amplify to the same extent as they would have to in the single slab. But then the distance over which we transfer the image is much smaller.502 x z y S A Ramakrishna z=0 ε − ε+ d/5 d/5 z=2d d/10 z=0 δ z=2d Figure 29.
In the static limit and no dissipation. the layers in the zero index bandgap material should not be impedance matched. 185].6) −1 cos(iε− kx d) + (1/2)(ε+ + ε+ ) sin(iε− kx d) where the dissipation (represented by ε− = Im(ε− )) in the NRM layers has been taken into account. Other interesting properties of such layered structures such as omnidirectional and large bandgaps [188–190]. 191]. Im(µ+ ) < 0. a ﬁnite resolution. above which there is also a perfect impedance matching. it acts as an optical ﬁbre bundle. c εxx (5. Bragg reﬂectors [192] and beam shaping [193]. In fact. This anisotropic material is equivalent to a system of perfectly conducting wires along the layer normal embedded a medium with ε = 0. Thus every wave passes through this anisotropic material without a change in amplitude or phase [182]. Interestingly the bandgap of these zero index layered systems is independent of periodicity and is robust against disorder. the positive medium should be optically amplifying (as in a laser gain medium) in order to counter the effects of absorption in the negative medium. among others.8) then the condition may still be satisﬁed. which varies very slowly. defect modes [188. A slab and layered structure of positive media and anisotropic media with indeterminate permittivity and permeability tensors (i. it merely corresponds to a ﬁnite ﬁbre thickness and. Another approach would be to put in by optical ampliﬁcation that which is lost by dissipation [180]. it also transfers the nearﬁeld information. Then the perfect lens . Im(ε+ ) < 0. Thus. consequently. The anomalous tunnelling across such media has also been considered in [186]. Thus the scope for improvement here by making the NRM less dissipative is much higher. Im(µ− ) > 0.7) 2πε− and the resolution depends inversely on the absorption rather than the logarithm of δε. Note now that the resolution of this layered lens is λ res = (5. We can write the transmission through such a layered system of thickness 2d in the quasistatic limit as [182] 1 Tp . (5. The point is that we need to have a supermetal that shows no dissipation. Such stacks can have a new kind of bandgap which corresponds to a zero refractive index (n− d− + n+ d+ = 0) [187].9) In other words. all the components do not have the same sign) has also been shown to work as a layered lens with partial refocusing [184. When there is dissipation. Ultimately the resolution of the layered lens is also limited by dissipation. have been considered.Physics of negative refractive index materials 503 propagation for the ppolarized radiation in such a uniaxially anisotropic medium has the dispersion 2 2 kx + ky εzz + 2 kz ω2 = ε0 µyy 2 .e. We note here that the study of layered structures with alternating layers of positive media and NRMs has been of considerable interest. Consider the perfect lens conditions with a lossy NRM: we see the possibility that if Im(ε− ) > 0.5) we have kz = 0 as the only solution. this corresponds to inﬁnitely thin wires connecting the two end faces point to point. provided that (5. Note that such zero index bandgaps can also occur for large layer thicknesses and are not related to our effective medium theory. but unlike a conventional optical ﬁbre bundle. (5.
conditions would be ‘perfectly’ met and the perfect image would result even with lossy NRM. for example. we will subsume all the processes relating to the ampliﬁcation (or absorption) of the electromagnetic wave into the negative (or positive) imaginary part of the dielectric constant. the use of optical ampliﬁcation is implied in the perfectlens condition itself. the transmission function for the layered lens. be accomplished by using a semiconductor laser material such as GaN or AlGaAs for the positive medium and silver for the negative medium.10) Then the perfect lens conditions would be perfectly met and the perfect image would result. (b) when the slab is split into eight thin layers of δ/2 = 5 nm thicknesses. The largest ﬁeld enhancements will be for the largest transverse wavevectors. we note that in the presence of the intense ﬁeld enhancements that are expected in this system the gain is likely to get saturated. we can have optical ampliﬁcation mainly for the electric dipole transitions. and the images of two closely spaced sources as resolved by the respective lenses are shown in ﬁgure 30. If we make the layers very thin. The imaging can now be carried out in the red. it would be possible to use other materials and correspondingly different wavelengths of light. the imaginary part of the positive gain medium can be tuned. The transmission function (left) and the electromagnetic ﬁeld intensity at the image plane (right) obtained (a) with a single slab of silver of 40 nm thickness. the local ﬁeld enhancements will not be as intense and the gain might not get completely bleached. More realistically. (c) Layered silverdielectric stack with optical gain and δ/2 = 5 nm and (d) layered silverdielectric stack with optical gain and δ/2 = 10 nm. We will consider here the case of thin layers of silver with positive amplifying dielectric media in between.504 S A Ramakrishna kx /ko X Figure 30. with and without ampliﬁcations. however. Using blue/ultraviolet (UV) light to pump the AlGaAs. Of course. The incorporation of gain through the magnetic permeability would necessarily be through electronic ampliﬁers in the metamaterials for the microwave and rf frequencies. ε± = ±1 ∓ i0. As a note of caution. In general. To satisfy the perfect lens conditions in both the real and the imaginary parts for the ppolarized light. Thus. This can. one can make the AlGaAs now optically amplifying in the red region of the spectrum. As a proof of the principle. we would require ε+ = ε+ − iε+ . we also show the case of the original single slab of . (5. For comparison. Keeping with standard practice. ε− = −ε+ + iε+ . where one can satisfy the perfect lens condition for the real parts of the dielectric constant.4 in (c) and (d). Alternatively one could also use other high gain processes such as Raman gain for this purpose. we do expect that the effective gain will be somewhat reduced and the corresponding enhancements of the image resolution will be smaller. By adjusting the pump laser intensity. at least in the quasistatic limit.
µ2 = −µ(x. Then it would appear for the world ahead of the slab that optically a thickness of 2d had been removed. The path for the propagating modes in the media will not be straight lines in general. we emphasize that the high spatial frequency components are transferred across.13) . y). Although the transmission function is not constant with the wavevector due to the layer plasmon resonances. y). The two peaks in the image for the single slab can hardly be resolved. the transmittance of the layered lens with gain does not decay rapidly with increase in the number of layers and a total stack thickness of even a few wavelengths [180]. Interestingly.2. ε2 = −ε(x.Physics of negative refractive index materials 505 (a) (b) d d d d Figure 31. silver as the lens (solid line and δ/2 = 40 nm) and a layered but gainless system. The overall effect is as if the space of thickness 2d did not exist for an observer on the righthand side. The improvement in the image resolution for the layered system with gain over the corresponding gainless systems is obvious. µ1 = +µ(x.11) (5. y). ε1zz µ1xx µ1yx µ1 = ˜ µ1zx µ1xy µ1yy µ1zy µ1xz µ1yz . The generalized perfect lens theorem The negative refractive slab can also be considered as optical antimatter for light in the sense it cancels out the effects of propagation through an equal amount of positive index media (see ﬁgure 31).12) where ε(x. It happens even when the dielectric permittivity and the magnetic permeability vary along the transverse directions. while they are clearly resolved in the case of the layered system with no gain. provided that the variation is just the same but with opposite signs: ε1 = +ε(x. The light and dark regions indicate opposite refractive indices. (5. y) are some arbitrary functions of x and y and can take positive or negative values (see ﬁgure 31(b) for a schematic picture). A pair of complementary media nullify the effect of each other for the passage of light. ∀0 < z < d. Knowledge of the transmission function of the lens would therefore enable one to recover a clean image from the observed image. This cancellation is applicable to the phase changes for propagating waves and the amplitude changes for the evanescent waves. µ1zz (5. y). This is the generalized lens theorem of [149] which we will merely state here and the proof of which we will give in appendix. 5. In fact. y) and µ(x. ∀d < z < 2d. Consider the most general dielectric permittivity and magnetic permeability tensor for spatial region 1 (0 < z < d): ε1xx ε1yx ε1 = ˜ ε1zx ε1xy ε1yy ε1zy ε1xz ε1yz . with the sharp edges of the slits becoming visible. the perfect focusing action occurs for more general conditions than a homogeneous slab with n = −1.
the use of conformal mapping on the solutions to the Laplace equation (in the quasistatic limit) for the slab was shown to yield an entire class of nearﬁeld lenses in two dimensions. It can be shown (see appendix for the proof) that the ﬁelds to the left and right of the interface between the two slabs are exactly identical but in opposite order as long as the two slabs have complementary material parameters: ε1xx ε1xy ε1xz µ1xx µ1xy µ1xz µ1 = µ1yx µ1yy µ1yz ˜ (5. all of which will exhibit the property of transferring images of sources in a perfect sense. −ε1zz −µ1xx µ2 = −µ1yx ˜ +µ1zx −µ1xy −µ1yy +µ1zy +µ1xz +µ1yz . Only in the quasistatic limit does a cylindrical surface with negative. and the transverse wavevector (kx . −µ1zz (5. transferring in and out the image of a source [177]. Note that a distortion of space results in a change of the ε and µ tensors in general.14) Note that the tensorial components can be speciﬁed functions of (x. Here we will present three cases of a cylindrical lens. to obtain magniﬁcation and demagniﬁcation of the images. These results have also been veriﬁed numerically by considering speciﬁc examples with spatially varying complementary media with mirror antisymmetry [149]. Thus in many cases. In fact. homogeneous ε have a nearly degenerate plasmon dispersion. allows us to generate a wide class of arrangements. the transformed geometry would involve spatially varying (heterogeneous) medium parameters. This theorem. the translational symmetry would have to be broken and curved surfaces will need to be employed.506 S A Ramakrishna Similarly. ky ) is preserved in the refraction process. The slab of NRM that is a perfect lens maps every point on the object plane to another point on the image plane. The perfect lens action in the case of the NRM slab was crucially dependent on the neardegeneracy of the surface plasmons in the case of the slab. In the more general case. in general. This result can be further generalized to any region that is mirror antisymmetric about a given plane [149]. 5. The perfect lens in other geometries Traditionally lenses have been used to project images and more importantly magnify or demagnify the image of a source. y). µ2zz (5. Such regions correspond to effectively optical null regions. The curved surfaces of a conventional lens break this translational symmetry in the transverse direction and enable the change in the size of the image. combined with the possibility of using coordinate transformations [194] to map the slab geometries into other geometries [149]. and curved surfaces. ˜ ε1zx ε1zy ε1zz µ1zx µ1zy µ1zz and −ε1xx ε2 = −ε1yx ˜ +ε1zx −ε1xy −ε1yy +ε1zy +ε1xz +ε1yz . for region 2 (d < z < 2d) ε2xx ε2xy ε2xz µ2xx µ2 = µ2yx ˜ ε2 = ε2yx ε2yy ε2yz .16) Then the two complementary media have an optical sum of zero. one uses the technique of coordinate transformations to map the solutions of the slab geometry into other geometries. In general. and a cylindrical shell in two dimensions with ε = −1 was shown to act as a lens.15) ε1 = ε1yx ε1yy ε1yz . . ˜ ε2zx ε2zy ε2zz µ2zx µ2xy µ2yy µ2zy µ2xz µ2yz . have a completely different dispersion for the surface plasmons. But the size of the image is identical to that of the source.3. This is due to the invariance in the transverse direction.
Cylindrical lenses.17) was shown [177] to map the NRM slab into a cylindrical shell (see ﬁgure 32 and ignore the dependence along the Z axis). z). The ampliﬁcation inside the spherical shell of the otherwise decaying ﬁeld is schematically shown.22) . The lenses in the curved geometries result in a magniﬁcation of the image including the nearﬁeld evanescent modes and the cornerlens is interesting as an open resonator. Note that the conformal transformation is valid for the Laplace equation and the solutions valid in the extreme nearﬁeld approximation. Qφ = r0 e / 0 . (5. y. A spherical shell with negative ε− (r) ∼ −1/r and µ− (r) ∼ −1/r images a source located inside the shell into the external region.19) ˜ Ei = Qi Ei . y = r0 e / 0 sin φ. 5. QZ = 1. The media outside have a positive refractive index. y. z = ln z The conformal transformation (5. but ε− (r) ∼ 1/r and µ− (r) ∼ 1/r.21) e / 0 .Physics of negative refractive index materials 507 z a3 a2 a1 a0 y x Figure 32. z). (5. As it has been shown [194]. y. embedded in a positive medium.18) (5. Q2 i µi = µi ˜ ˜ Hi = Qi Hi .3.1. where Q2 = i ∂x ∂qi + ∂y ∂qi + ∂z ∂qi . if we deﬁne the new coordinates q1 (x. the material parameters and ﬁelds are given by εi = εi ˜ Q1 Q2 Q3 .20) Now consider the transformation to cylindrical coordinates x = r0 e so that Q = r0 0 / 0 cos φ. 2 2 2 Q1 Q2 Q3 . Q2 i (5. z) and q3 (x. In general we can transform Maxwell’s equations for the slab lens of an NRM. then in the new frame. q2 (x. a spherical lens and a cornerlens which focuses the source back on itself. into a cylindrical shell with ε− (r) and µ− (r) whose inner and outer radii are a1 and a2 . z = Z. (5.
23) µz . and the other 2 way around from outside to the inner world. Z plane. 1 . Hence the new system must accordingly act as a cylindrical lens and transfer images in and out of the cylindrical lens.3. 0 εφ = ˜ µφ = ˜ = 1 and −1 0 εφ .2. The transformation technique can be similarly used to map the slab lens into spherical coordinates (r = r0 e / 0 . They cannot focus outside this region. Z directions are irrelevant. The image will be formed on the surface a3 = a0 (a2 /a1 )2 and there will be a magniﬁcation of the image by the factor M= a2 a1 2 . For this speciﬁc case. a whole class of lenses can be generated in two dimensions using the transformation techniques [177. ˜ ε = µ = −1. 178].30) in the . ˜ ˜ εφ = µφ = −1. r where r = r0 exp( ).27) ˜ ε = µ = +1. εz = µz = − 2 . r2 1 εr = µr = −1. and using the generalized lens theorem. which is identical to the situation of the Veselago lens. εφ = µφ = −1. a spherical shell of NRM with ε− (r) ∼ −1/r and µ− (r) ∼ −1/r embedded in a positive medium with ε+ (r) ∼ +1/r and µ+ (r) ∼ +1/r will act as a spherical lens projecting sources in and out of the shell [149]. 2 They can focus sources from inside to the outside only when a1 /a2 < r < a1 . ˜ ˜ < (5. ˜ ˜ a2 εz = µZ = +1. Further note that for the optical axis along the direction. εφ = µφ = +1. Hence the medium parameters in general could also be arbitrary functions of φ and Z. ˜ ˜ εz = µz = +1.29) ln a1 r0 εφ = µφ = +1. ∀ a2 < r. εz = µz = + εφ = µφ = +1. . . (5. < 0 (5. A spherical lens.24) 0µ µZ = ˜ exp Choosing the scale factor εr = µr = +1. we obtain εφ = µφ = +1. (5.508 S A Ramakrishna We obtain the transformed dielectric permittivity and magnetic permeability to be ε = ˜ µ = ˜ 0ε . ˜ ˜ ∀ < 0 0 ln a1 r0 < . r 1 εr = µr = +1.31) Note that these cylindrical lens are also shortsighted in the same manner as the slab lens. εZ = ˜ 2 r0 0 exp 2 r0 0 2 0 εz . 2 0 (5.25) (5. εz = µz = + 2 . φ. ˜ ˜ ∀ ∀ r < a1 . ∀ a1 < r < a2 . ˜ ˜ εz = µZ = −1. the generalized lens theorem predicts that the variations in the transverse φ. we will present a more conventional proof in terms of the TE and TM modes in the spherical geometry [179]. θ. In general. when the source is located in a2 < r < a2 /a1 .26) (5. (5. −1 0 µφ . φ).28) ln a2 r0 . ˜ ˜ ∀ 0 ln r0 ε = µ = +1. 5. (5.
m (1) (1) [n+ Alm r n+ −1 + n− Blm r n− −1 ]Ylm (θ.39) − U + U + ε(r)µ(r) 2 U = 0. l. (2) (2) [n+ Alm r n+ −1 + n− Blm r n− −1 ]Ylm (θ. And if we assume ε(r) = ε(r) and µ(r) = µ(r). φ).40) [n(n + 1) − l(l + 1) + p(n − 1) + p]r n−2 + 2 p+q+n = 0.33) ∇ · D = ∇ · [ε(r)E] = ∇ε(r) · E + ε(r)∇ · E = 0. r · ∇(∇ · E) = − ∂r ε(r) ε(r) Using the above four equations. which will constitute a full solution to the problem. ω2 ∇µ(r) ∇ × ∇ × E = iωµ0 ∇ × [µ(r)H] = 2 µ(r)ε(r)E + iω ×∇ ×E (5.41) Hence the general solution can be written as Er (r) = l.37) (rEr ) + Er .m [n+ Alm r n+ −1 + n− Blm r n− −1 ]Ylm (θ. Hence the solution is (rEr ) = U (r)Ylm (θ. we also note that ε (r) ∂ ε (r) (5. and the spherical harmonics are a solution to the angular part.m (3) (3) [n+ Alm r n+ −1 + n− Blm r n− −1 ]Ylm (θ. ∂ ε (r) ε (r) ω2 (5. φ).32) c µ(r) and we have (5. we have ε (r) ε (r) r·E=− (rEr ). Operating on Maxwell’s equation with ∇.38) (rEr ) + (rEr ) + ε(r)µ(r) 2 (rEr ) = 0. we can have a solution U (r) ∼ r n and we get αβω2 (5. a1 < r < a2 . r 2 ∂r ∂r r2 ∂r ε(r) rε(r) c If we choose ε(r) = αr p and µ(r) = βr q . ∇ 2 (rEr ) + ∂r ε(r) rε(r) c which is separable. c2 (5. φ).45) E(2) (r) = E(3) (r) = .43) (5. φ). l.42) and a similar solution can be obtained for the TE modes with r · E = 0.44) (5. U (r). we can now write down the electric ﬁelds of the TM modes in the different regions for the negative spherical shell of ﬁgure 2 as E(1) (r) = l.Physics of negative refractive index materials 509 Consider the spherically symmetric system shown in ﬁgure 32. Under the circumstances of spherical symmetry. (5. satisﬁes ∂U 1 ∂ ∂ ε (r) l(l + 1) ε (r) ω2 r2 U+ (5. (5. c r implying p + q = −2 and n± = 1 −(p + 1) ± 2 (p + 1)2 + 4l(l + 1) − 4αβω2 .36) Using the fact that ε only depends on r. φ) (5. a2 < r < ∞ (5. we can obtain the equation for E.m a0 < r < a1 . it is sufﬁcient to specify the quantities (r · E) and (r · H). where the radial part. Consider now the TM polarized modes r · H = 0. implying that only the electric ﬁelds have a radial component Er . ∇ 2 (r · E) = r · ∇ 2 E + 2∇ · E.34) ∇ ·E=− rε(r) rε(r) We note the following identities for later use: (5.35) ∇ × ∇ × E = ∇(∇ · E) − ∇ 2 E. Now assuming an arbitrary source at r = a0 .
(3) Blm = (5. for r > a3 . with only the condition of complementarity between the negative and positive regions.49) 2 r a1 2 2 Hence.e.48) The lenslike property of the system becomes clear by writing the ﬁeld outside the spherical shell as √ l(l+1)−αβω2 /c2 2 1 a2 (1) (3) Er = r Blm Ylm (θ. this is not true for points a2 < r < a3 within the image surface. Then the imaging for the TM modes becomes independent of µ and only requires ε− ∼ −1/r 2 . Note that the positive medium outside the shell is also required to have spatially varying material parameters ∼1/r. given that ε− (a2 ) = −ε+ (a2 ). q = −1. 5. apart from a scaling factor of 1/r. We also have a spatial magniﬁcation in the image 2 2 by a factor of a2 /a1 .3. A perfect twodimensional corner lens. the ﬁelds appear as if the source were located on the spherical image surface (r = a3 ). the ﬁelds on the sphere r = a3 = (a2 /a1 )a0 are identical to the ﬁelds on the sphere r = a0 .46) (1) Blm .50) where ε1 = −Re(ε− (a1 )) and ε3 = −Re(ε− (a2 )) for the perfect lens conditions and εi (r) ∼ 1/r 2 is the imaginary part of the permittivity for the negative medium shell.510 S A Ramakrishna (1) and similarly for the magnetic ﬁelds.3. 2 ln(a2 /a1 ) (5.41). Here we have written down the solutions for the TM modes. However.47) √ l(l+1)−αβω2 /c2 (1) Blm . Now the tangential components of the magnetic ﬁelds and the normal components of the displacement ﬁelds have to be continuous across the interfaces. The solutions for the TE modes can be similarly obtained. and ε and µ can now be arbitrary functions of θ and φ. For causal solutions Alm = 0. the perfect lens solutions only occur for ε ∼ µ ∼ 1/r. we can terminate the spatially varying media at some ﬁnite but large distance. As with the cylindrical lens. we have (1) Alm = 0. Further the constraint that the positive media outside also be spatially varying can be dropped. Although the solutions given by equation (5.42) occur in any medium with εµ ∼ 1/r 2 . note that our imaging direction is along r. Again there is a considerable simpliﬁcation that is possible in the extreme nearﬁeld limit [179]. Third. First. Consider an inﬁnitely extended corner of NRM with n = −1 as shown in ﬁgure 33(a). (2) Alm = √ (5. points outside the image surface. Note that Blm correspond to the ﬁeld components of the (3) source located at r = a0 . φ). Similarly the TE modes become independent of ε. Let us note a couple of points about the above perfect lens solutions in the spherical geometry. (5. this spherical lens will be able to image sources which are also located only within in a ﬁnite distance of the spherical layer. i. using the ideas of the asymmetric lens [142]. (2) Blm = 0. A ray diagram shows that a source located in the . In the presence of ﬁnite dissipation. Under the conditions p = −1. 1 2 a1 2 a2 2 a1 l(l+1)−αβω2 /c2 (5. which implies that p = −1 in equation (5. ε+ (a1 ) = −ε− (a1 ) and ε+ (a2 ) = −ε− (a2 ). and the only condition is that µ− ∼ −1/r 2 . Second. In this limit it can be shown that the largest multipole that can be resolved is again limited by the dissipation in the NRM and is approximately [179] lmax ln{3ε1 ε3 /[εi (a1 )εi (a2 )]} . we have the perfect lens solutions if and only if n+ = −n− .
(a) A corner of NRM with n = −1 refocuses a source. This focus is also perfect in that it includes the nearﬁeld components and can be shown by the mapping into a periodically layered stack as shown. Using the generalized lens theorem. The conditions of complementarity are satisﬁed. namely that the focusing with these corner lenses is true for the evanescent components as well and is perfect! Consider the periodic arrangements of slabs shown beside the corners with material parameters π +1 − < φ < π. (b) A double corner of NRM refocuses the source back onto itself and creates an image in the diagonally opposite quadrant.52) π π −1 −π < φ < − and 0 < φ < 2 2 in the second case (b) of the double corner lens. . Using the transformation as in the case of the cylindrical lens. x = r0 e gives us ε = µ = εφ = µφ = ˜ ˜ ˜ ˜ +1 −1 2 ˜ εZ = µZ = r0 e2 / 0 ˜ 2 εZ = µZ = −r0 e2 / ˜ ˜ 0 / 0 cos φ. 2 ε(φ) = µ(φ) = (5. 2 2 ε(φ) = µ(φ) = (5.51) π −1 −π < φ < − 2 in the ﬁrst case (a) of a single corner and π π +1 − < φ < 0 and < φ < π. y = r0 e / 0 sin φ. (5. It has also been shown that all the surface modes of this system are degenerate and the density of modes at the double corner is very large [149].Physics of negative refractive index materials y 511 (a) x (a) n=–1 φ (b) n=–1 n=–1 (b) y n=–1 n=–1 n=–1 n=–1 x n=–1 l n=–1 n=–1 l n=–1 Figure 33. a more rigorous result can be proved. Hence we conclude that the images in both the cases of the single corner (a) and the double corner (b) will be perfect. In fact combining two such corners as shown in ﬁgure 33(b) makes it even more exotic: classical rays emanating from a source placed in any one of the quadrants are returned to the source point [195]. This corner lens can be shown to be perfect by mapping it to a periodic layered stack. The dependence along the r axis is irrelevant as it is orthogonal to the imaging axis along φ. z = Z. second quadrant can be focused into the fourth quadrant.54) in the respective positive and negative quadrants. Note that case (b) is unusually singular as each source images on each other and the ﬁelds grow indeﬁnitely in time.53) (5. Such a system can form a very interesting open resonator in which light can be trapped for long times.
But there is enough experimental evidence to place many of the theoretical ideas on a very safe footing. placed in a microwave guide. 24] have experimentally implemented the Swiss roll design for operation at radiofrequencies. NRMs had been predicted by Veselago in 1967. 5] combined the planar SRRs and the thin wire arrays to make the world’s ﬁrst NRM (see ﬁgure 12). In this section. The dispersion of µ(ω) has been measured and found to match accurately with the theoretical predictions. Wiltshire et al [23. Other experiments [196] have used a looplike structure to increase the inductance of thicker wires and obtained a low frequency cutoff medium. These consisted of sheets of polysterene rohacell sheets with 20 µm goldplated tungsten wires laid on them with a 5 mm spacing. The measured reﬂectivity and transmission of these samples indicated a plasma frequency of about 9 GHz. Combining this with 0. By using the large effective magnetic permeability of these Swiss rolls at frequencies just below the resonance frequency. the growth of experimental activity has been rather slow compared with the almost frenetic pace in the theoretical studies. where the thin wire arrays were made using spark chamber technology. with alternate sheets rotated by 90˚. Part of the reason could be that the experimental samples are not easy to fabricate. But on the whole. and these sheets were stacked up. and showed a low transmission band for radiation with the magnetic ﬁeld along the SRR axis. This ameliorates the bianisotropy of the SRRs as well. There have been studies of the single scatterer response of these SRR media [62]. The San Diego experiments [4. There have been attempts at producing isotropic SRR media by combining three orthogonal SRRs into a single threedimensional scatterer [63]. 5]. 3. particularly the microstructures for high frequency operation. This stack would be expected to have two diagonal components of the dielectric tensor negative below the plasma frequency. Very recently the magnetic properties of microscale SRRs has been demonstrated at terahertz frequencies [27]. This could be attributed to the negative µ at frequencies above the LC resonance of the SRRs. these media have been used as effective ﬂux tubes for the radiofrequency magnetic ﬁelds and the potential for MRI has been demonstrated.we will brieﬂy cover the main experimental activities.1. These have now been followed up by several experimental groups. This is also possibly the reason why most of the experiments have been performed at microwave frequencies. Transmission experiments were ﬁrst carried out on uniaxial SRR media. which consisted of planar SRRs of copper rings printed on circuit boards.512 S A Ramakrishna 6. Experiments on negative refractive index The ﬁrst experiments on the plasmonic nature of thin wire arrays were reported in [43]. It was the theoretical recipes of Pendry et al [2. But the main catalyst for the sudden increase in the interest in NRMs came from the experimental demonstration of NRMs by the San Diego group in 2000 [4. 43] for designing negative material parameters that made the real developments possible. Marqu´ s et al [198] have implemented a modiﬁed SRR structure where e the two planar split rings are capacitively coupled not through their edges but through the broadside by depositing equal sized split rings on either side of an insulating board. There have been several experimental efforts at obtaining negative magnetic materials after the San Diego experiments [4] brought them into the limelight.8 mm wires placed symmetrically between the SRRs with the electric ﬁeld along the thin wire resulted in an . Experimental evidence for NRMs and superlenses This ﬁeld has been one in which theoretical developments have led the experimental activities. 6. which could be reduced to 6 GHz by diluting the density of the wires by half. thus conﬁrming the theoretical prediction of a low frequency plasmonic medium.
and a clear verdict on the exact level of losses is yet to emerge. These measurements are consistent with the theoretical predictions of Luo et al [36]. where it was also numerically shown that the angle of refraction was insensitive to the degree of absorption in the sample. indicating a passband due to a negative refractive index. The transmission line approach has shown that NRM structures with a low loss and wide bandwidth are possible. They demonstrated that an image placed in a positive medium near an interface with the NRM would create an image inside the NRM in the sense of the Veselago lens.2. These experiments used large sized threedimensional isotropic samples and were conducted in free space rather than in waveguides. been criticized for the low transmittivity of the samples [9].Physics of negative refractive index materials 513 enhanced transmission within the negative µ band compared with the transmission of the bare wire media. Eleftheriades et al experimentally implemented twodimensional loaded L–C backward transmission lines and have demonstrated the negative refraction effect [131]. But it certainly does not negate the conclusion of a negative refractive index. A prism made with a similar twodimensional isotropic NRM was shown to exhibit the negative refraction effect [5]. 6. More recently this effect has been demonstrated at telecommunications infrared frequencies using even a low contrast twodimensional photonic crystal [202]. The issue of loss in these NRM is complex. This becomes possible due to the inverse dependence √ of the wavevector on the frequency (β = −1/(ω LC)). The imaging with a ﬂat lens has also been extended to sonic . particularly at microwave frequencies. Liu et al [130] have demonstrated the feasibility of a forward microwave coupler between two backward wave structures using a very short coupling length. 132] and Liu et al [130]. The experiments of Parazzoli et al [25. Photonic crystals and the Veselago ﬂat lens There has been extensive activity on the negative refraction effect with photonic crystals. The insertion loss was shown to be smaller (about −1 dB cm−1 ) [26] and the contribution of the conduction losses in the SRRs appears to dominate. A prism made of this medium was shown to again refract microwaves negatively [25]. This idea has also been extended to coupling between ordinary transmission lines and backward wave transmission lines [199]. Parimi et al [201] have demonstrated the ﬂat lens imaging. the large wire diameter and the corresponding large impedance mismatch. however. Using a twodimensional array of alumina rods. a Gaussian beam of microwaves at about 15 GHz was shown to undergo a reversed transverse shift. at about 9 GHz.25 mm thickness and a low absorption substrate (Rogers 5880) and measured both the reﬂected and transmitted power. These experiments have. Loaded L–C backward wave transmission lines have been implemented by Eleftheriades et al [131. Several reasons can qualitatively account for the low transmittivity including the dissipation in the circuit boards. something that the other resonant structures are not capable of.26] have provided decisive experimental evidence of a negative refractive index in composite samples of thin wire arrays and SRRs. Cubukcu et al [200] have shown the negative refraction effect in such media by using the reversed transverse shift a beam undergoes upon oblique transmission through a slab compared with normal media. These experiments used thin wires of 0. again with a twodimensional photonic crystal of cylindrical alumina rods. Marqu´ s et al [90] e have shown that placing SRRs within a waveguide operating below the cutoff also creates a passband below the cutoff frequency which can be attributed to a negative refractive index. The transmission passband for the composite SRR and thin wire medium has also been obtained experimentally by other groups [197].
The design of metamaterials that show a negative refractive index is an extreme form of electromagnetic engineering. We have examined the problem of making materials with negative material parameters which are mostly structured composites. underdamped response due to the structural resonances. Using the surface roughness to linearly couple a propagating mode into evanescent surface modes on a silver surface.3. The behaviour of the transmittance was quantitatively predicted by Pendry’s expressions and the material parameters of silver [71] and has offered indirect evidence of the capability for superlensing of the silver lens. The complete generality of this effect becomes clear in the recent ﬂat lens imaging by liquid surface waves where a twodimensional lattice of rigid rods was used to act as a crystal for the surface waves. The images with lower resolution were obtained which. precluded the possibility of subwavelength image resolution due to losses and retardation. we have reviewed the recent developments in the area of negative refractive index materials. More recently. But more reﬁned measurements are expected in the near future. The measured electric ﬁelds also clearly showed the ampliﬁcation of the evanescent waves across the slab. most of the elementary features of the theory have been clearly veriﬁed. 7. however.514 S A Ramakrishna waves. the transmitted evanescent ﬁelds of the surface plasmon modes were coupled out using a hemisphere on the other side of the silver ﬁlm. compared with the case of diffraction limited patterns. in 1967. Superlenses and subwavelength focusing capabilities have enormous implications for nearﬁeld imaging and nearﬁeld optical lithography. It should be noted that the sizes of . submicron imaging using the silver slab lens has also been demonstrated [206] in the context of optical lithography. A negative refractive index is obtained by interleaving two structures that individually show a negative ε or negative µ. Conclusions In conclusion. however. Thus. The subwavelength resolving capabilities of the superlenses stem from the way the evanescent nearﬁeld components are transmitted across it: these components are ampliﬁed in magnitude. subwavelength image resolution at optical frequencies using the silver lens is yet to be demonstrated. The interest in it has been enormous. Demonstrating superlenses By superlenses. There is now deﬁnite experimental evidence to conﬁrm the essentials of the theory of superlenses. The negative material parameters mostly result from an overscreened. Liu et al [204] used a reversed attenuated total reﬂection (RATR) experiment to directly measure the transmitted ﬁelds across thin ﬁlms of silver. Grbic and Eleftheriades [205] have experimentally noted superlens behaviour with a planar NRM lens made out of twodimensional transmission lines. and phononic crystals have also been shown to focus sound in a similar manner [203]. were demonstrated to be superior to the traditional proximity imaging techniques. Subwavelength image resolutions at about 1 GHz were clearly measured with the capacitively loaded transmission grid kept between the source and detector. the practical creation of these materials had to wait for recipes to selectively make materials with a negative dielectric permittivity and magnetic permeability. The measured transmittance increased exponentially with ﬁlm thickness up to some maximum and then decayed quickly with a further increase in thickness which was caused by the increased dissipation in the ﬁlm. Although predicted a long time ago. At the time of writing. The subwavelength image resolution was limited only by the levels of absorption in the NRM. The images were obtained using a 120 nm thick slab of silver which. we mean lenses with subwavelength resolution enabled by NRMs. 6.
electromagnetic couplers and waveguides in photonic devices cannot be underemphasized. The simpliﬁcation that results is considerable as the complex internal microscopical fast varying ﬁelds become irrelevant and need not concern us anymore. We have considered how the excitation of surface plasmon modes on the surface of the NRMs are crucial to restoration of the amplitudes of the evanescent modes. but very real nevertheless. Calculations indicate that scaling to high frequencies. But these PBG materials that show the allangle negative refraction have a tremendous technological potential due to the low degree of absorption in them. There are considerable problems of homogenization in the photonic bandgap materials. along with a method of coordinate transformations. We have discussed in detail this effect where both the propagating and the evanescent nearﬁeld modes participate in the image formation. We have discussed the backward wave structures using transmission lines that can act as NRMs at microwave frequencies and radiofrequencies and the photonic bandgap materials that can show the negative refraction effect. primarily arising due to conduction losses in the metallic structures. integrated circuits. But on the whole NRMs have to be accepted as today’s reality. NRM enable us to manipulate the nearﬁeld radiation which up to now was untouchable. The study of nonlinear metamaterials is only starting now. is possible. The fundamental limitation comes from the ﬁnite electronic mass as real currents have to ﬂow in these structures. Many of the intuitions of conventional electromagnetism fail here and one needs to develop new physical intuition here. however. The immensely dispersive nature of these materials can create many new surprises. The resonances in the metamaterials can cause intense local ﬁeld enhancements. We have discussed the generalization of this effect to spatially varying media and the concept of optically complementary media that behave in a sense like optical antimatter. The dissipative and dispersive nature of the NRM.Physics of negative refractive index materials 515 the structures in these materials are usually much smaller than the wavelength of radiation that enables them to be treated effectively as homogeneous media. The immense potential of the perfect lens effect for nearﬁeld imaging. There are still very signiﬁcant issues of nonlocal ﬁelds. This generalization. bianisotropy and homogenization that need to be carefully looked at. and they are perhaps better described by the band theory than by homogeneous negative material parameters. . NRM are new electromagnetic materials with no conventional analogues in many cases. but surprises will almost certainly abound. The perfect lens that can be made from a slab of NRM is perhaps the most dramatic of the new effects that are possible. spoils the perfect focus and imposes limits on the maximum extent of subwavelength image resolution that we can obtain. particularly from the point of view of low losses and large bandwidths. This makes it a new fertile ground for nonlinear phenomena. and some of the earliest technical innovations and applications will possibly occur using them. The high degree of absorption in these materials. Several new effects that are possible in NRMs have been discussed. Several new conﬁgurations of the superlenses that include the nearﬁeld have been presented. The developments in the past few years have only offered us a glimpse of what is possible. waveguide technology and photonic circuits that can outperform their conventional counterparts. up to the nearinfrared and perhaps even the optical frequencies. NRM can enable new nearﬁeld technologies when coupled with innovations in antennae. is a cause for concern. These materials can support a host of surface modes and have a large potential in the newly developing area of plasmonics. and one of the main challenges remaining in this area is developing new designs for metamaterials with smaller losses. Further developments will crucially depend on our ability to make better quality NRM. enables us to generate classes of surfaces with degenerate surface plasmons and perfect lenses in different geometries. They are very dispersive and dissipative. There is now sufﬁcient experimental evidence for a negative refractive index and the associated effects.
ky [ε1xx (kx . ky . we have ky E1z (kx . k1z E1x (kx .14). ky )E1z (kx . kx . ky . Appendix. ky . kx . ky ) + µ1xz (kx . N Kumar. ky ) = −ωµ0 kx . kx .ky [µ1xx (kx . ky )]. ky )H1x (kx . kx . ky )H1y (kx . ky ) + µ1zz (kx . Martin McCall and Sebastien Guenneau. ky ) + µ1yz (kx . Now the electromagnetic ﬁelds have to satisfy Maxwell’s equations: ∇ × E = iωµ0 µH. ky ) = ωε0 kx . ky ) . kx . z) = exp(ik1z z) exp[i(kx x + ky y)] E1y (kx . kx . ky ) = −ωµ0 kx . kx . ky ) . ky . Stephen O’Brien. ky ) E2x (kx . Consider the two slabs with the spatially varying dielectric permittivity tensors and magnetic permeability tensors described by equations (5. kx . ky )H1y (kx . ky ) E2 (x.ky [µ1yx (kx . ky )H1z (kx .516 S A Ramakrishna Acknowledgments I would like to acknowledge with thanks Professor John Pendry for encouragement and enlightening discussions. ky ) − k1z H1y (kx . ky )H1z (kx . ky )H1z (kx . ky . ky ) − kx E1z (kx . ky . Bangalore.13) and (5. ky . ky )H1y (kx .ky E2z (kx . ky ) (A6) +µ1zy (kx . y. ky )]. ky . Much of my knowledge of negative index materials has come from discussions with Olivier Martin. ky )H1x (kx . ky ) − ky E1x (kx . kx E1y (kx . respectively. ky )E1y (kx . ky )H1x (kx . ky ) = −ωµ0 kx . kx . ky )] and ky H1z (kx . ky . ky )E1x (kx . ky ) − k1z E1y (kx . . kx . ky . kx . ky ) (A7) +ε1xy (kx . Proof of the generalized perfect lens theorem Here we present the proof of section 5. ky ) E1 (x. ky .ky [µ1zx (kx . that two media with complementary optical behaviour amount effectively to an optical null space as shown in ﬁgure 31. ky ) (A2) (A3) where the Bloch conditions can be assumed. y. kx . David Smith. kx . z) = exp(ik2z z) exp[i(kx x + ky y)] E2y (kx . ky .2. Substituting these ﬁelds into Maxwell’s equations and separating the components. E1x (kx . ky ) (A4) +µ1xy (kx . Mike Wiltshire. kx . ky )]. I would like to thank Harshawardhan Wanare and Wayne Williams for critically reading the manuscript. ˜ (A1) Now decomposing the ﬁelds into the Fourier components in the two slabs. I would also like to thank him and Imperial College London. ˜ ∇ × H = −iωε0 εE. and Professor N Kumar and the Raman Research Institute.ky E1z (kx . ky ) + ε1xz (kx . for hospitality during part of the time when this review was written. ky ) (A5) +µ1yy (kx .
NematNasser S C and Schultz S 2000 Phys. Now consider the substitution E2x (kx . ky )E1y (kx . ky ) = H1x (kx . ky )E1z (kx . ky ) = E1y (kx . ky . ky ) = −H1z (kx . Stewart W J and Youngs I 1996 Phys. ky ) − kx H1z (kx . ky . 47 2075 [4] Smith D R. ky . 27 885 [10] McCall M. ky ) −ε1zz (kx . Holden A J. Rev. Lett. kx . ky . kx . Rev. ky . ky . Phys. ky ) (A9) +ε1zy (kx . kx . kx . Hence we have the ﬁeld E(x. 45 191 . kx . ky ) (A14) which solves Maxwell’s equations in region 2 inside the slab and also matches the ﬁelds across the boundaries of the slabs.Physics of negative refractive index materials 517 k1z H1x (kx . ky ) . ky ) (A11) k2z = −k1z . ky . ky . ky ) = ωε0 kx . ky . This completes the proof of the theorem. y. 85 3966 [7] Valanju P M. z = 2d) = E(x. Walser R M and Valanju A P 2002 Phys. kx . ky ) +ε1zy (kx . ky ) = H1y (kx . kx . H2x (kx . Lett. ky ) −µ1xy (kx . and E2y (kx . ky .ky [ε1yx (kx . ky . ky )]. (A12) −ε1xx (kx .ky [ε1zx (kx . Padilla W J. ky . kx . Nauk 92 517 Veselago V G 1968 Sov. ky )]. kx . (A10) H2z (kx . ky . ky )E1y (kx . ky ) + ε1yz (kx . kx . ky ) − ky H1x (kx . kx H1y (kx . ky . References [1] Veselago V G 1967 Usp. Rev. y. ky ) +µ1xz (kx . ky )E1x (kx . ky . ky . kx . 76 4773 [3] Pendry J B. J. (A15) Thus the ﬁelds are exactly repeated in region 2 but in exactly the opposite order. ky ) +µ1zy (kx . Rev. ky ) −ε1yy (kx . E2z (kx . ky ) −µ1yy (kx . z = 0). kx . 84 4184 [5] Shelby R A. ky ) +µ1yz (kx . ky ). kx . ky ). ky )E1x (kx . ky ) −ε1xy (kx . Phys. 88 187401 [8] Garcia N and NietoVesperinas M 2002 Phys. ky ) = E1x (kx . Fiz. Vier D C. kx . ky . 23 353 [11] Markos P and Soukoulis C M Preprint condmat/0212136 [12] Shen J Q 2004 Preprint condmat/0402213 [13] Pendry J B 2004 Contemp. kx . ky . kx . ky ) = −ε1yx (kx . ky ) ε2 (kx . kx . ky ) + ε1zz (kx . ky ) +ε1yz (kx . ky . ky . kx . Lett. Lett. ky ) µ2 (kx . 10 509 [2] Pendry J B. ky . ˜ +ε1zx (kx . ky ) −µ1zz (kx . kx . 88 207403 [9] Garcia N and NietoVesperinas M 2002 Opt. kx .—Usp. ky ). Rev. ky ) ˜ +µ1zx (kx . ky . y) exp(−ikz d) = E(z. Weiglhoffer W S and Lakhtakia A 2002 Eur. kx . Smith D R and Schultz S 2001 Science 292 77 [6] Pendry J B 2000 Phys. kx . ky ) (A13) −µ1xx (kx . ky ) (A8) +ε1yy (kx . ky . kx . kx . kx . Lett. ky ) = ωε0 kx . ky . Lett. ky )E1z (kx . Phys. ky ) = −E1z (kx . ky ). ky . ky ). Microwave Theory Tech. Robbins D J and Stewart W J 1999 IEEE Trans. ky . ky ) +ε1xz (kx . Holden A J. ky ) = −µ1yx (kx . kx . H2y (kx .
50 7784 Camley R E and Mills D L 1982 Phys. E 68 065602 [53] Makhnovsky D P and Panina L V 2003 J. 39 1561 Sandercock J R and Wettling W 1979 J. Rev. Rep. MacPeake D. Rev. Sauer H. 219 191 Harris S E 1997 Phys. Lett. Rev. Rev. Matter 14 7409 [22] O’Brien S. 106 874 [32] Raether H 1980 Excitation of Plasmons and Interband Transitions by Electrons (Berlin: Springer) [33] Pendry J B 2001 Proc. 78 4135 [48] Pendry J B 1997 Phys. Padilla W J. 64 1107 Scully M O 1992 Phys. B 53 6318 Matitsine S M. Matter 14 6383 [29] Lindell I V. Lett. B 67 13103 [47] Mikhailov S A 1997 Phys. Marque´ R. Rev. Matter 10 4785 [44] Sievenpiper D F. 78 4136 [49] Walser R M. Kontenbah B E C and Tanielian M H 2003 Phys. NATO Advanced Study Inst. Pendry J B. Rev. Rev. Lett. Lett. Nefedov I S. Lett. Appl. Thio T and Wolff P A 1998 Nature 391 667 MartinMoreno L et al 2004 Phys. Camley R E. 28 1649 Grunberg P and Metawe F 1977 Phys. Rev. Tretyakov S A. 94 1146 [43] Pendry J B. Today 50 36 [40] Scully M O and Zubairy M S 1997 Quantum Optics (Cambridge: Cambridge University Press) [41] Sarychev A K and Shalaev V M 2000 Phys. Lett. Phys. Antennas Propagat. A 311 254 [55] Rotman W 1962 IRE Trans. Maslovski S I. AP31 471 . Lett. Phys. Rev. Liu L. Fang N. Joannopoulos J D and Pendry J B 2002 Phys. Young I R. 335 275 [42] Lagarkov A N and Sarychev A K 1996 Phys. Lifschitz E M and Pitaevskii L P 1984 Electrodynamics of Continuous Media 2nd edn (Oxford: Pergamon) [20] Kotthaus J P and Jaccarino V 1972 Phys. Tech. Greegor R B. 77 2412 [52] Koschny T. Pendry J B. E 64 056625 [31] Ritchie R H 1957 Phys. B 26 1280 Remer L. Lett. Tech. Luthi B.: Condens. 90 107401 [26] Li K. 86 1114 MartinMoreno L. Lett. Field J E and Imamoglu A 1990 Phys. Rev. Gan Y B. Lett. Lett. Rev. Rev. Robbins D J and Steward W J 1998 J. Lett. Vier D C. NATO ASI Series) ed C M Soukoulis (Dordrecht: Kluwer) [34] Barnes W L. 58 2059 Yablonovitch E 1993 J. 31 129 [30] Ziolkowski R W and Heyman E 2001 Phys. Appl. B 58 897 [21] Pendry J B and O’Brien S 2002 J. Phys. Matter 5 2443 [17] Pendry J B 1996 J. Gilderdale D J and Hajnal J V 2001 Science 291 848 [24] Wiltshire M C K. Sprik R. Phys. Johnson S G. Larkman D J. 76 2480 [45] Maslovsky S I. 90 167401 [39] Harris S E. Nikoskinen K I and Ilvonen S 2001 Microwave Opt.: Condens. Ramakrishna S A and Pendry J B 2004 Phys. Thiel D V and Park K S 1983 IEEE Trans. Lett. Smith D R and Soukoulis C M 2003 Phys. 82 2535 [27] Yen T J. McLean S J. Hajnal J V. Basov D N and Zhang X 2004 Science 303 1494 [28] O’Brien S and Pendry J B 2002 J. 93 4120 [54] Simovsky C R and He S 2003 Phys. Lett. Smith D R. Rev. Tip A and Lagendijk A 1996 Phys. Phys. Express 11 709 [25] Parazzoli C G. Phys. Lezec H J. Rev. Lett. Dereux A and Ebbesen T W 2003 Nature 424 824 [35] Born M and Wolf E 1989 Principles of Optics (Oxford: Pergamon) [36] Luo C. Simovski C R and Tretyakov S A 2003 s Phys. Tretyakov S A and Belov P A 2002 Microwave Opt. Rev.: Condens. Phys. Rev. 35 47 [46] Belov P A. Phys. Rev. 80 3286 [16] Yablonovitch E 1987 Phys. Rev. Rev.518 S A Ramakrishna [14] Pendry J B and Smith D R 2004 Phys. Lett. Ghaemi H F. Rep. Lett. Rev. 92 609 [38] Ebbesen T W. Rev. 56 2752 Dumelow T. Rev 85 338 Bohm D and Pines D 1953 Phys. Holden A J. B 69 241101 [23] Wiltshire M C K. B 65 201104 [37] Pines D and Bohm D 1952 Phys. Phys. Geick R and Camley R E 1986 Phys. Degiron A and Ebbesen T W 2003 Phys. Markos P. Rev. Silveirinha M. Hock K M. Today 57 37 [15] Shen J T and Platzmann P M 2002 Appl. Lett. Pendry J B and Edwards D J 2003 Opt. Li K. AP10 82 [56] King R J.: Condens. Sickmiller M E and Yablonovitch E 1996 Phys. Rev. Parazzoli C G and Tanielian M H 2003 Appl. Abraha K and Tilley D R 1998 Phys. Antennas Propag. Matter 8 1085 [18] Jackson J D 1999 Classical Electrodynamics (Singapore: Wiley) [19] Landau L D. Lezec H J. Lagarkov A N and Rozanov K N 2003 J.: Condens. Greegor R B. Valanju A P and Valanju P M 2001 Phys. (Crete. Phys. 87 119701 [50] Pendry J B 2002 private communication [51] van Coeverden D V. Appl. GarciaVidal F J.
Phys. 2 195 Pitarke J M. Ware M and Peatross J 2001 Phys. Belov P A and He S 2004 IEEE Trans. Rev. Commun. Z. B: At. B 67 245109 Anderson P W 1963 Phys. Rev. Opt. Appl. Martel J. Opt. Lett. B 37 642 GayBalmaz P and Martin O J F 2002 J. B 6 4370 Smith D R.: Condens. 47 2059 Broas R F J. Schultz S. 92 6252 Shamonin M. Phys. 43 L176 Markos P and Soukoulis C M 2002 Phys. Chang H. 43 L190 Sievenpiper D. Rev. Lih J S and Chern J L 2004 Japan. B 65 033401 Huang Y C. B 57 15261 Takeda H and Yoshino K 2003 Phys. 27 3781 Tokunaga E. Rev. Fahy S and Louie S G 1988 Phys. Lett. J. Lett. Tech. A 47 R4581 Peiponnen K E. Chen K H. Tech. Wu S C. 41 209 Johnson P B and Christy R W 1972 Phys. Asia Academic Seminar (Hyderabad) ed P R Rao et al (International Advanced Research Centre for Powder Metallurgy & New Materials and Japan Society for Promotion of Science) p 33 Dimmock J O 2003 Opt. A 69 043818 Ginzburg V L and Meiman N N 1964 Sov. Mod. Lett. Appl. Cheng Y K. Rev. Lett. Rev. B 65 195104 Kumar N 2003 Proc. Express 11 2397 Panoiu N C and Osgood R M 2003 Opt. Matter 14 4035 Shelby R A. Medina F and RaﬁiElIdrissi R 2002 Phys. Rev. 89 183901 Bergman D J and Stroud D 1992 Solid State Phys. Rev. Lett. Lett. Lett. Dielectr. E 65 036622 Pokrovsky A L and Efros A L 2002 Phys. Markos P and Soukoulis C M 2002 Phys. Chen C F. Micowave Conf. Medina F and Mesa F 2002 Phys. Microwave o Theory Tech. 81 939 Hsu A c. Tayeb G and Maystre D 1999 Opt. Enoch S and Tayeb G 2000 J. B 66 155411 Hrabar S. Mod. 89 093901 e Marqu´ s R and Smith D R 2004 Phys. 49 1262 Enoch S. Rev. Rev. Rev. Rev. 46 147 Stroud D 1975 Phys. Tech. Rev. Appl. Lett. NematNasser S C and Schultz S 2001 Appl. Terasaki A and Kobayashi T 1993 Phys. 12 3368 Hunderi O 1973 Phys. Antennas Propag. Eres Z and Bartolic J 2002 Proc. Phys. Ringhofer K H and Solymar L 2002 J. Rev. GarciaVidal F J and Pendry J B 1998 Phys. Vartiainen E M and Saarinen J J 2004 Phys. A 235 118 Glasgow S. E 64 046610 Smith D R and Kroll N 2000 Phys. Rev. Soc. Phys. Mol. Rev. Appl. Shamonina E. 17 1012 Pendry J B and MacKinnon A 1992 Phys. Chern J L. 5 184 Marques R. Phys. 92 2929 GayBalmaz P and Martin O J F 2002 Appl. Appl. J. 92 059401 Simovski C R. 161 171 Gralak B.—JETP 19 169 Diener G 1997 Phys.Physics of negative refractive index materials [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] 519 Brown J 1960 Prog. 95 3778 Kong J A 1990 Electromagnetic Wave Theory (New York: WileyInterscience) Saadoun M M I and Engheta N 1992 Microwave Opt. Phys. Lett. Commun. Lett. Sievenpiper D and Yablonovitch E 2001 IEEE Trans. 124 283 Ramakrishna S A and Martin O J F 2004 Preprint condmat/0402570 Depine R A and Lakhtakia A 2004 Microwave Opt. 69 2772 Pendry J B 1994 J. Kalinin V and Solymar L 2004 J. Opt. (Milan) pp 327–30 Shamonina E. 33 465 Loudon R 2002 J. Phys. Am. Phys. Kalinin V A. Eur. Smith D R. Alex´ polous N G and Yablonovitch E 1999 IEEE Trans. Phys. 30 522 de Kronig R 1942 Natuurk 9 402 Kircheva P P and Hadjichristov G B 1994 J. 49 821 Lakhtakia A and McCall M W 2004 Preprint condmat/0402001 . 85 2933 Pokrovsky A L and Efros A L 2002 Solid State Commun. Opt. 78 489 Markos P and Soukoulis C M 2002 Phys. Lien Y H and Shy J T 2004 Japan. B 65 144440 O’Brien S and Pendry J B 2002 J. 130 439 Kittel C. Phys. Phys. Braos R F J. A 223 331 Panina L V. Hsu YJ. 51 2582 Simovski C R and Sauviac B 2004 Preprint condmat/0403047 e Marqu´ s R. Rev. Rev. Zhang L. B 7 3419 Milton G W 2002 The Theory of Composites (Cambridge: Cambridge University Press) Kramers H A 1929 Phys. Microwave Theory Tech. Rev. Grigorenko A N and Makhnovskiy D P 2002 Phys. 41 315 Wang J and Lakhtakia A 2002 Microwave Opt. Rev.
Phys. A 277 61 Ruppin R 2001 J. Sokhorukov A A and Kivshar Y S 2003 Appl. Kong J A and Chen M 2003 Opt. Wu B. 91 157404 Pendry J B 2000 Phys. Wu B I. E 69 026604 Caloz C. Lett. Fleugel B and Mascarenhas A 2003 Phys. B 62 10696 Luo C. Sci. Phys. 92 5560 Eleftheriades G V. Shen Y R. B 66 035113 Ashcroft N W and Merimin N D 1976 Solid State Physics (Philadelphia: Saunders College) Smith D R. Phys. Sokoloff J B and Sridhar S 2002 Preprint condmat/0207689 Lu W T. 87 249702 Ramakrishna S A and Armour A D 2003 Am. Commun. 80 2084 Berman P R 2002 Phys. Rev. Lett. Smith D R. Chang C C and Itoh T 2001 J. Tang HT. Phys. Rev. Rev. Wu B I and Zhang Y 2002 Appl. Tech. Matter 15 6345 Greffet J J and Carminati R 1997 Prog. Pacheco J. Lett. Lett. Rev. Johnson S G and Joannopoulos J D 2003 Science 299 368 Goos F and H¨ nchen H 1947 Ann. Rev. 2 87 Shadrivov I V. 90 029703 Smith D R. 91 099701 Feise M W. 66 217 Pendry J B and Smith D R 2003 Phys. 90 077401 Haldane F D M Preprint condmat/0206420 Pendry J B 2003 Phys. 38 392 Alu A and Engheta N 2002 Microwave Opt. Grzegorczyk T M. 87 249701 Pendry J B Phys. 81 2352 Chien HT.: Condens. Nevi´ re M and Zakhidov A A 2003 Opt. Grzegorczyk T M. Ibanescu M. Opt. Mod. 49 1747 e Darmanyan S A. 90 107402 Dutta Gupta S. 56 133 Pohl D and Courjon D 1993 NearField Optics (Dordrecht: Kluwer) Williams J M 2001 Phys. Zharov A A and Kivshar Y S 2003 Appl. Lett. 82 3820 Shadrivov I V. Economou E N and Soukoulis C M 2003 Phys. Lett. Schurig D. Appl. B 69 113104 Liu L. Lett.520 [109] [110] [111] [112] [113] [114] [115] [116] [117] [118] [119] [120] [121] [122] [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133] [134] [135] [136] [137] [138] [139] [140] [141] [142] [143] [144] [145] [146] [147] [148] [149] [150] [151] [152] [153] [154] [155] [156] [157] [158] [159] [160] [161] [162] [163] S A Ramakrishna Zhang Y. Lett. Rev. Phys. 81 2713 Lu W T. Rev. Phys. 83 2713 Landauer R and Martin Th 1994 Rev. Kuo CH. Lett. Caloz C. Rev. Microwave Theory Tech. E 66 067603 Lakhtakia A 2002 Electromagnetics 23 71 Lakhtakia A 2002 Preprint condmat/0305133 Ziolkowski R W 2003 Opt. 87 249703 Pendry J B 2001 Phys. B 68 045115 GomezSantos G 2003 Phys. Chen CC and Ye Z 2003 Preprint condmat/0312582 Kuo CH and Ye Z 2003 Preprint condmat/0310423 Kuo CH and Ye Z 2003 Preprint condmat/0312288 Shen Y R 1984 The Principles of NonLinear Optics (New York: Wiley) Zharov A A. Zhang Y. 1 333 a Kong J A. Phys. Schurig D and Pendry J B 2002 Appl. Lett. Mod. Rev.: Condens. 87 249704 ’t Hooft G W 2001 Phys. 92 5930 Notomi M 2000 Phys. 82 1506 . Sokoloff J B and Sridhar S 2004 Phys. Appl. Shadrivov I V and Kivshar Y S 2003 Phys. Arun R and Agarwal G S 2004 Phys. Bevelacqua P J and Schneider J B 2002 Phys. Phys. Phys. 211 183 Cory H and Barger A 2003 Microwave Opt. Express 11 662 Artmann K 1948 Ann. Phys. Rev. Matter 13 1811 Ramakrishna S A. Phys. Phys. Rev. Lpz. Lett. Schultz S. 50 2702 Grbic A and Eleftheriades G V 2002 J. 90 5483 Pacheco J. Lett. Lett. Rev. Pendry J B. Phys. Johnson S G. Joannopoulos J D and Pendry J B 2003 Phys. Lett. Phys. Ramakrishna S A and Pendry J B 2003 Appl. World 13 27 Lu J. Phys. Rev. Lett. Lett. Iyer A K and Kremer P C 2002 IEEE Trans. 35 460 Kuzmiak V and Maradudin A A 2002 Phys. Tech. Johnson S G and Joannopoulos J D 2002 Appl. 71 562 Luo C. Rosenbluth M. Lett. Lpz. B 66 045116 Pendry J B and Ramakrishna S A 2003 J. Baughman R H and Zakhidov A A 2004 Phys. 91 037401 Agranovich V M. Rev. Lett. Lett. Chang CC and Itoh T 2002 J. Rev. 116 411 Klimov V V 2002 Opt. Commun. Rev. Rev. Rev. J. Schurig D and Schultz S 2002 J. Surf. Express 11 723 Luo C. Rev. Lett. Appl. Zhang Y and Kong J A 2002 Phys. B 69 165112 Ruppin R 2000 Phys. 225 233 Ruppin R 2000 Solid State Commun. 89 257401 Foteinopoulou S.
14 31 Cubukcu E. Tech. Components Lett. Rev. Rev. Lett. NematNasser S C and Schultz S 1999 Appl. Kalinin V A. Padilla W. Lu W T. 37 912 Kolinko P and Smith D R 2003 Opt. Lett. Opt.: Condens. Phys. Rev. 90 077405 Smith D R. 84 4403 . Rev. Mod. 84 1290 Chen L. Matter 14 8463 Pendry J B 2003 Opt. Antennas Propag. 37 4 a a Cummer S A 2002 Appl. Lett. Zhang Y and Zhu S 2003 Appl. Wiltshire M C K and Stewart W J 2003 J. Rachford F J and Schelleng J 2003 Phys. Rev. Phys. Phys. 83 5386 Bria D. 84 2244 Zhang Z M and Fu C J 2002 Appl. Lett. Lett. Sukhorukov A A and Kivshar Y S 2003 Appl. Lett. Rev. Lett. 80 1097 Li J. 51 2572 Caloz C and Itoh T 2004 IEEE Microwave Wirel. Lett. Aydin K. Dobrzynski L. Phys. Lett. Opt. Phys. Rev. Electron 34 133 Smith D R. Quant. B 68 113103 Rao X S and Ong C K 2003 Phys. Foteinopoulou S and Soukoulis C M 2003 Nature 423 604 Parimi P V. 82 1503 Rao X S and Ong C K 2003 Phys. Phys. Cowan M L. Qiu M. Phys. Aydin K. 57 2615 Smith D R and Schurig D 2003 Phys. 31 169 Ramakrishna S A. Blaikie R J and Wolf C R 2004 Appl. Express 11 640 Loschialpo P F. Vier D C. E 69 066613 Nefedov I S and Tretyakov S A 2002 Phys. B 67 201101 Tretyakov S A 2001 Microwave Opt. Mod. 82 161 Ye Z 2003 Phys. Vigneron J P. Lett. 82 2008 Pendry J B and Ramakrishna S A 2002 J. Phys. Rev. Li H. Antennas Propag. Rev. Mulot M. B 67 193106 Merlin R 2004 Appl. 93 024301 Liu Z. Lett. Phys. Lett. Pendry J B. Technol. 51 2592 e Marqu´ s R. Lett. Liu Z. 93 073902 Yang S. Smith D L. Talneau A and Anand S 2004 Phys. Lett. Lett. Ringhofer K H and Solymar L 2001 Electron. 37 1243 Fang N and Zhang X 2003 Appl. Lett. E 66 036611 Feise M W. Phys. Shadrivov I S and Kivshar Y S 2004 Preprint condmat/0402087 Ruppin R 2003 Microwave Opt. 75 1425 Ozbay E. Rev. Lett. Page J H. Phys. Swillo M. Cubukcu E and Bayindir M 2003 IEEE Trans. Rev. Schurig D. Lett. Phys. Zhou L. Lett.Physics of negative refractive index materials [164] [165] [166] [167] [168] [169] [170] [171] [172] [173] [174] [175] [176] [177] [178] [179] [180] [181] [182] [183] [184] [185] [186] [187] [188] [189] [190] [191] [192] [193] [194] [195] [196] [197] [198] [199] [200] [201] [202] [203] [204] [205] [206] 521 Shamonina E. Kolinko P and Rye P 2004 Appl. Lett. 92 117403 Melville D O S. 83 5184 Grbic A and Eleftheriades G V 2004 Phys. 50 1419 Mansﬁeld S M and Kino G S 1990 Appl. Lett. Thylen L. Christopoulos C and Thomas D W P 2001 Electron. Lett. Lett. Rev. He S and Shen L 2004 Phys. Rev. 82 3820 Ward A J and Pendry J B 1996 J. Ozbay E. Chen H. DjafariRouhani B. B 69 115115 Ramakrishna S A and Pendry J B 2003 Phys. Express 11 755 Ramakrishna S A and Pendry J B 2004 Phys. Martel J and Medina F 2003 IEEE Trans. Chan C T and Sheng P 2003 Phys. Rev. Vodo P and Sridhar S 2003 Nature 426 404 Berrier A. E 68 067601 Cummer S A 2003 Appl. 43 773 Notomi M 2002 Opt. 38 494 Shadrivov I S. Forester D W. E 67 025602 K¨ rk¨ innen M K and Maslovski S A 2002 Microwave Opt. Tech. El Boudouti E H and Nougaoui A 2004 Phys. Mock J J. Chan C T and Sheng P 2004 Phys. Akjouj A. Fang N. Yen TJ and Zhang X 2003 Appl. Mesa F. 92 107404 Paul J. 90 083901 Jian H. Phys.