Related content

- Quantum optics of spatial transformation

General relativity in electrical engineering media
Ulf Leonhardt and Thomas G Philbin

To cite this article: Ulf Leonhardt and Thomas G Philbin 2006 New J. Phys. 8 247 - Generalized transformation optics of linear
materials
Robert T Thompson, Steven A Cummer
and Jörg Frauendiener

- Invisibility cloaking without superluminal

View the article online for updates and enhancements. propagation
Janos Perczel, Tomáš Tyc and Ulf
Leonhardt

Recent citations
- Entanglement of two two-level atoms
mediated by a cylindrical optical black hole
Mahnaz Tavakoli and Ehsan
Amooghorban

- Generalized transformation design:

Metrics, speeds, and diffusion
Paul Kinsler and Martin W. McCall

- Analysis of a conformal invisible device

Lin Xu et al

This content was downloaded from IP address [Link] on 28/01/2018 at 19:06

 New Journal of Physics The open–access journal for physics

General relativity in electrical engineering

Ulf Leonhardt and Thomas G Philbin
School of Physics and Astronomy, University of St Andrews,
North Haugh, St Andrews KY16 9SS, UK
E-mail: ulf@[Link]
New Journal of Physics 8 (2006) 247
Received 26 July 2006
Published 23 October 2006
Online at [Link]
doi:10.1088/1367-2630/8/10/247

Abstract. In electrical engineering metamaterials have been developed that

offer unprecedented control over electromagnetic fields. Here, we show that
general relativity provides the theoretical tools for designing devices made of
such versatile materials. Given a desired device function, the theory describes the
electromagnetic properties that turn this function into fact. We consider media that
facilitate space-time transformations and include negative refraction. Our theory
unifies the concepts operating behind the scenes of perfect invisibility devices,
perfect lenses, the optical Aharonov–Bohm effect and electromagnetic analogues
of the event horizon, and may lead to further applications.

New Journal of Physics 8 (2006) 247 PII: S1367-2630(06)29513-2

1367-2630/06/010247+18$30.00 © IOP Publishing Ltd and Deutsche Physikalische Gesellschaft
2 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
Contents

1. Introduction 2
2. Theory 3
2.1. Electromagnetism in curved coordinates . . . . . . . . . . . . . . . . . . . . . 3
2.2. Transformation media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3. Examples 6
3.1. Perfect invisibility devices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2. Perfect lenses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.3. Optical Aharonov–Bohm effect . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.4. Artificial black holes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4. Conclusions 12
Acknowledgments 13
Appendix A 13
Appendix B 16
References 17

1. Introduction

Modern metamaterials offer remarkable control over electromagnetic fields [1] with applications
ranging from invisibility devices [1]–[8] to perfect lenses [9]–[15]. Imagine there were no
practical limits on the electromagnetic properties of materials. Given a desired function, how
do we find the design of the device that turns this function into fact? In this paper, we show
that general relativity provides clear recipes for calculating the required material properties.
The practical use of general relativity in electrical engineering may seem surprising: relativity
has been associated with the physics of gravitation [16] and cosmology [17] or, in engineering
[18], has been considered a complication, not a simplification. But the design concept described
here is rooted in a simple idea with a distinguished history: according to Fermat’s principle
[5, 19, 20] light rays follow the shortest optical paths in media; they are effective geodesics, and
general relativity has developed the theoretical tools for fields in curved geometries1 [16]. Our
theory generalizes the concept behind the perfect invisibility devices [1]–[3] to magneto-electric
or moving media and it incorporates negative refraction [9]–[15]. For example, using a simple
pictorial argument we show how to design perfect magnifying lenses. Finally, metamaterials
may be also applied for laboratory analogues of general relativity, in particular for artificial
black holes [21]–[31]. In this way, we unify and generalize a range of physical phenomena that
rely on the geometry of media and the recent opportunities of metamaterials.
Metamaterials are materials with designed properties that stem from structure, not substance,
whereas man-made structures determine the electromagnetic properties, structures that are
smaller than the electromagnetic wavelengths involved. Metamaterials have a long history:
mediaeval ruby glass, for example, is a metamaterial. Ruby glass contains nano-scale gold
colloids that render the glass neither golden nor transparent, but ruby, depending on the size
and concentration of the gold droplets. The colour originates from a resonance of the surface
1
However, as in other analogue models of gravity [21], we only use some aspects of general relativity, in particular
the kinematic aspects, not the dynamic ones.

New Journal of Physics 8 (2006) 247 ([Link]

3 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
plasmons [32] on the metallic droplets. Metamaterials per se are nothing new: what is new is
the degree of control over the structures in the material that generate the desired properties. For
example, in the modern version of negatively refracting ruby glass [33] nano-manufactured pairs
of gold-pillars on a silicon substrate generate finely tuned plasmon resonances where each pair
acts like a designer atom with controllable properties.
Our starting point is not new either: in the early 1920s Gordon [34] noticed that moving
isotropic media appear to electromagnetic fields as certain effective space-time geometries.
Tamm [35, 36] generalized this geometric approach to anisotropic media and briefly applied this
theory [36] to the propagation of light in curved geometries. In 1960 Plebanski [37] formulated
the electromagnetic effect of curved space-time or curved coordinates in concise constitutive
equations. Dielectric media act on electromagnetic fields as geometries and geometries act as
effective media. In 2000, it was understood [38] that the dipole forces of electromagnetic fields
in media appear as the effects of geometries as well; electromagnetic fields act as geometries on
media, a concept used to identify the conditions for the Abraham or the Minkowski momentum in
media [39]. Geometrical ideas have been applied to construct conductivities that are undetectable
by static electric fields [40, 41]. Only very recently [1]–[5] meta-material implementations of
coordinate transformations were considered as engineering tools, ideas we take further here: we
show that general relativity provides the most natural recipes for such engineering applications.

2. Theory

2.1. Electromagnetism in curved coordinates

A geometry is characterized by the space-time measure [42] ds2 = αβ gαβ dxα dxβ , the metric,
where we denote the space-time coordinates by xα with Greek indices running from 0 to 3.
Latin indices indicate the spatial coordinates and run from 1 to 3, whereas x0 = ct describes
time measured in spatial units with c being the speed of light in vacuum. The matrix gαβ , the
metric tensor, may vary as a function of the coordinates, because space-time may be curved or
because curved coordinates are used in flat space. The determinant√ g of gαβ measures the four-
dimensional (4D) volume of an infinitesimal space-time element as −g times the infinitesimal
coordinate volume [16, 42]. We denote the matrix inverse of gαβ by gαβ ; where, as customary
in general relativity [16, 42], the position of the indices indicates that gαβ is co-variant and
gαβ is contra-variant under coordinate transformations. For example, flat space in cylindrical
coordinates ct, r, ϕ, z is described by the metric tensor gαβ = diag(1, −1, −r 2 , −1) with matrix
inverse gαβ = diag(1, −1, −r −2 , −1) and determinant g = −r 2 .
As our starting point, we use the result of general relativity [37, 43], derived in appendix A,
that the free-space Maxwell equations can be written in the macroscopic form [44]

∂D ∂B
∇ · D = ρ, ∇ × H = + j, ∇ · B = 0, ∇ × E = − , (1)
∂t ∂t
or, in Cartesian components,
 ∂Di  ∂Bi
= ρ, = 0, (2)
i
∂xi i
∂xi

New Journal of Physics 8 (2006) 247 ([Link]

4 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
 ∂Hk ∂Di  ∂Ek ∂Bi
ijk = + ji, ijk = − , (3)
jk
∂xj ∂t jk
∂xj ∂t

where ijk is the completely antisymmetric Levi-Civita tensor [16]: in Cartesian components ijk
is 1 for all cyclic permutations of 123 , −1 for all cyclic permutations of 213 and 0 otherwise. The
spatial indices indicate that in this representation Ei and Hi form the components of vectors that
are co-variant under purely spatial transformations, whereas Di and Bi constitute√ contra-variant
√
spatial vectors. The charge density ρ and the current density j are given by −g j 0 and c −g j i
of the four-current j α [42]. In empty but possibly curved space, the electromagnetic fields are
connected by the constitutive equations in SI units [37],
w µ w
D = ε0 ε E + × H, B= H − ×E (4)
c ε0 c2 c
with the symmetric matrices ε and µ and the vector w given as [37]
√
−g ij g0i
ε=µ=− g , wi = . (5)
g00 g00

In dielectric media, the E, B vectors generate electric polarizations and magnetizations that
constitute the D, H fields. The constitutive equations are described here in implicit form, but
one can also write D, H as functions of E and B [18]. We see that empty but curved space-time
manifolds or curved coordinates appear as effective dielectric media. The electric permittivity ε
and the magnetic permeability µ are equal, as in perfect impedance matching [44]. In addition, in
the general case, in which ε and µ are not multiples of the identity, they represent an anisotropic
medium. The w vector describes a magneto-electric coupling between the magnetic and the
electric fields. Some materials are magneto-electric, but the simplest example is a moving medium
[18, 38, 45], because a moving medium responds to the electromagnetic field in locally co-
moving frames and Lorentz transformations from the laboratory frame naturally mix electric and
magnetic fields [18, 45]. Summarizing, one can say that empty space appears as an impedance-
matched anisotropic magneto-electric or moving medium.

2.2. Transformation media

Since empty space appears as a medium, what happens if a medium appears as empty space
[1]–[5]? Consider the following situation: suppose that the constitutive equations (4) of the
medium fit the equations (5) of empty flat space in some curved coordinates. The medium thus
maps the electromagnetism in this virtual empty space to physical space, see, for example,
figure 1. Such media are called transformation media [1]. We call the virtual space
electromagnetic space-time, in analogy to optical space [4]. Imagine we invert the coordinate
transformation from electromagnetic space-time to physical space: the medium would disappear.
For instance, in electromagnetic space-time, light would travel along straight lines, whereas
in physical space light rays follow the curved lines of the transformed coordinate grid.
Of course, this apparent curvature is an illusion, because in theory it is removable by the inverse
coordinate transformation: but, in practice, one can exploit this apparent curvature to create
illusions [1]–[5].

New Journal of Physics 8 (2006) 247 ([Link]

5 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT

Figure 1. Transformation media implement coordinate transformations. The left

figure shows an orthogonal grid of coordinates in electromagnetic space (a slice
of cylindrical coordinates at constant z). The right figure shows the transformed
grid in physical space that corresponds to the invisibility device described by
equation (11) and illustrated in figure 2.

We use primes to denote the geometry and coordinates of electromagnetic space-time.2 For
keeping the theory as flexible as possible, we describe physical space in another set of arbitrary
spatial coordinates xi with spatial metric γij and determinant γ. For example, we may wish to
use cylindrical coordinates r, ϕ, z with spatial metric tensor γij = diag(1, r 2 , 1) and γ = r 2 .
Since the transformation medium exploits differences in the coordinates of physical space and
electromagnetic space-time, the metric γij should differ from the spatial part of the effective
geometry gαβ generated by the medium.
The transformation rules of tensors in general relativity give rise to a simple recipe for
the construction of media that facilitate space-time coordinate transformations. Contra-variant
tensors are transformed as [16, 18, 42]

 ∂xα ∂xβ  
α β
g αβ
= α  β g . (6)
α β 
∂x ∂x

The transformed inverse metric serves as the building block of the dielectric tensors (5) of the
medium. If we wish to express physical space in arbitrary coordinates, we need to consider
a subtlety that appears when we write the divergences in Maxwell’s equations (2) in spatially
co-variant form [42]
√ √
1  ∂ γDi 1  ∂ γBi
√ = ρ, √ = 0. (7)
γ i ∂xi γ i ∂xi

The εij and µij tensors are naturally contra-variant with respect to the background geometry
of physical space, but γ differs from −g, in general. Consequently, for writing the medium as
√
an active coordinate transformation, we need to multiply D and B by γ in the constitutive
equations (4) and re-scale ρ and j accordingly, which is also consistent with the form of the
Levi–Civita tensor in curved coordinates [16] in the curls in Maxwell’s equations (3). However,
2
We use the opposite notation of [1] since, we believe, it is more logical.

New Journal of Physics 8 (2006) 247 ([Link]

6 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
a second subtlety arises from the curls: the coordinate transformation may turn a right-handed
coordinate system into a locally left-handed one. The coordinate directions change handedness
when the product of the eigenvalues of the Jacobian matrix ∂xi /∂xj changes sign, i.e. when the
Jacobian changes from positive to negative. The problem is that the curls (3) implicitly assume
a right-handed system, because ijk changes sign under coordinate transformations from right to
left-handed systems. Consequently, wherever this is the case, we need to invert the sign of ε, µ
and ρ, j. The transformation to a left-handed coordinate system thus corresponds to negative
refraction [14, 15] occurring in what has, for other reasons [15], been fittingly described as
left-handed media.3 Taking all this into account, we arrive at the simple recipe
√
−g gij 1 g0i
ε =µ =∓ √
ij ij
, wi = √ . (8)
γ g00 γ g00

The ∓ sign indicates the handedness: minus for right-handed transformations and plus for locally
left-handed ones. Starting from the inverse metric gαβ in electromagnetic space, the gαβ matrix
is calculated according to the transformation rule (6). The matrix inverse of gαβ gives the metric
tensor gαβ and the inverse of the determinant of gαβ gives g. Equations (8) specify the required
electromagnetic properties that will turn the coordinate transformation into reality. In the special
case of right-handed and purely spatial transformations our recipe agrees with the theory of
[1, 56], see appendix B, where however, instead of the contra-variant ε and µ, mixed tensors
occur with one of the indices lowered by the spatial metric,

εik = µik = εij γjk . (9)
j

Here, we consider more general transformations that may mix space and time and that may be
multivalued. As long as the matrix (6) is single-valued and not explicitly time-dependent such a
device is physically allowed and stationary.
The coordinate transformation encodes the function of the device. Outside of it, the
electromagnetic coordinates agree with the coordinates of physical space: the transformation is
trivial; ε and µ are unity and w vanishes. Inside the device, electromagnetic fields are controlled
as prescribed by the coordinate transformation. If the physical coordinates enclose a hole one
obtains the blueprint of an invisibility device [1]–[5]. We show in the next section that perfect
lensing [9]–[15], the optical Aharonov–Bohm effect [22, 23, 57, 58] and optical black holes
[21]–[31] can be understood as applications of the same idea as well.

3. Examples

3.1. Perfect invisibility devices

Perfect invisibility devices [1]–[3] facilitate coordinate transformations with holes: suppose
that electromagnetic space does not cover the entire physical space; the medium excludes
electromagnetic fields in certain regions, but smoothly fits them to fields outside the device.
3
Negative refraction in left-handed media is not related to the case of negative phase velocities in gravity [46]–[53],
because transformations to left-handed or any other coordinates have no physical significance there, in contrast to
transformations facilitated by media. Note also that the papers [46]–[53] have been criticized [54, 55].

New Journal of Physics 8 (2006) 247 ([Link]

7 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT

Figure 2. Invisibility device. The transformation medium (11) acts as an

invisibility cloak, guiding light around the interior of the cloak without causing
any distortion. The device facilitates the coordinate transformation illustrated in
figure 1.

Therefore, electromagnetic radiation is naturally guided around the excluded regions. The
medium cloaks these regions such that any object placed inside is hidden without revealing
anything to the outside world. Perfect invisibility devices [1]–[3] must employ anisotropic
media, because the inverse scattering problem for waves in isotropic media has unique solutions
[59, 60]. They also require media where the phase velocity of light approaches infinity at the
inside of the cloaking layer, because coordinate transformations with holes rip apart points of
zero volume in electromagnetic space and tear them to finite volumes in physical space, unless
the coordinate transformation is singular. To prove this, consider the determinant of the ε and
µ tensor (8). For an invisibility device [1]–[3] only space is transformed, so w vanishes, g00 is
unity and −g is reduced to the inverse of the determinant of the spatial components gij . For det ε
we obtain from equations (6) and (8) in 3D space in terms of the Jacobian [42]

∂(x1 , x2 , x3 )

detε = (−g)1/2 γ −3/2 = (−g )1/2 γ −3/2 . (10)
∂(x1 , x2 , x3 )
√
At a point of measure zero −g vanishes and so does det ε, the product of the eigenvalues of ε
and µ. Consequently, at least one of the eigenvalues of ε and µ are zero; therefore in at least one
direction of the anisotropic medium the phase-velocity of light diverges near the inside of the
cloak. The speed of light in media can reach high values in narrow frequency ranges or, naturally,
for static fields. On the other hand, invisibility devices that are only perfect within the limits of
geometrical optics [4]–[6] are not subject to such constraints [6].
Figure 2 illustrates a cylindrical invisibility device that stretches the z-axis out in the radial
direction to a full cylinder of radius R1 , compressed within a cylindrical volume of radius R2 ,
New Journal of Physics 8 (2006) 247 ([Link]
8 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
as shown in figure 1. This device is an invisibility cloak of thickness R2 − R1 where anything
placed inside the inner radius R1 is hidden. We obtain from the theory
 
 R2 − R1 R2 r  (R2 − R1 )2 r 2
r = R1 + r ⇒ εj =i
diag , 2 , 1 , (11)
R2 R2 − R1 r R22 r

within R1
r
R2 or, equivalently, 0
r 
R2 . Clearly, close to the lining of the cloak at

r → R1 where r  → 0 the speed of light in the r and z directions diverges, whereas in ϕ direction
the phase velocity tends to zero.

3.2. Perfect lenses

In perfect invisibility devices, electromagnetic space does not cover the entire physical space;
here, we show that perfect lenses correspond to multi-valued electromagnetic space: single points
in electromagnetic space are mapped to multiple points in physical space. Consider in Cartesian
coordinates the transformation x(x ) illustrated in figure 3, whereas all other coordinates are not
changed. In the fold of the function x(x ) a point x in electromagnetic space has three faithful
images in physical space. Obviously, electromagnetic fields at one of those points are perfectly
imaged on to the other: the device is a perfect lens [9]. Such perfect lensing requires left-handed
media with negative ε and µ [14, 15], because inside the device, i.e. inside of the x fold, the
derivative of x(x ) becomes negative and the coordinate system changes handedness. We obtain
from our recipe (8) the compact result
 
dx dx dx
ε = µ = diag , , . (12)
dx dx dx

At the place where the transition from right-handed to left-handed coordinates occurs, the
derivative of x(x ) is undefined and so are ε and µ, but this has no adverse effect in practice [61].
It simply describes the discontinuity at the boundary of the medium where our transformation
x (x) automatically gives the correct boundary conditions [44], because Maxwell’s equations are
satisfied arbitrarily close to the boundary.
Equation (12) shows that when dx /dx is −1 inside the device and +1 outside, we obtain
the standard perfect lens [9]–[15] based on an isotropic left-handed material with ε = µ = −1
inside. Perfect lensing was first analysed [9] as the imaging of evanescent waves in a slab of
negatively refracting material, waves that may carry images finer than the optical resolution
limit [19]. Various aspects of this idea [9] have been subject to a considerable theoretical debate
[62]–[66], but recent experiments [10]–[13] confirmed sub-resolution imaging. Our pictorial
argument leads to a simple intuitive explanation of why such lenses are indeed perfect.
Furthermore, we see that lenses with ε = µ = −1 are not the most general choice. One could use
an anisotropic medium to magnify perfect images, in contrast to the existing examples [10]–[13],
by embedding the source or the image in transformation media with |dx /dx| = 1. Additionally,
our picture turns out [67] to give an intuitive explanation why the Casimir force [68, 69] in media
with negative refraction may turn from attraction to repulsion, possibly to the extend of levitating
objects on quantum zero-point fluctuations [67].

New Journal of Physics 8 (2006) 247 ([Link]

9 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
x′

Figure 3. Perfect lens. Negatively refracting perfect lenses employ transforma-

tion media. The top figure shows a suitable coordinate transformation from the
physical x-axis to the electromagnetic x ; the lower figure illustrates the cor-
responding device. The inverse transformation from x to x is either triple- or
single-valued. The triple-valued segment on the physical x-axis corresponds to
the focal region of the lens: any source point has two images, one inside the lens
and one on the other side. Since the device facilitates an exact coordinate trans-
formation, the images are perfect with a resolution below the normal diffraction
limit [19]: the lens is perfect [9]. In the device, the transformation changes right-
handed into left-handed coordinates. Consequently, the medium employed here
is left-handed, with negative refraction [15].

3.3. Optical Aharonov–Bohm effect

Perfect invisibility devices and perfect lenses exploit transformation media with non-trivial
topology, with excluded regions in physical space or folds in electromagnetic space. Here we
consider an example where the transformation to physical space is multi-valued, but the medium
is single-valued and hence physically allowed: the opticalAharonov–Bohm effect [22, 23, 57, 58].
One can demonstrate this effect with light passing through a water vortex [22] or with slow light
in Bose–Einstein condensates [23]. Note that the effect is related to the Aharonov–Bohm effect
with surface waves [70]–[72] and to the gravitational Aharonov–Bohm effect [73]. The optical
Aharonov–Bohm effect is an example of a transformation medium that mixes space and time,

New Journal of Physics 8 (2006) 247 ([Link]

10 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT

Figure 4. Aharonov–Bohm effect. A fluid vortex generates the optical Aharonov–

Bohm effect described by the coordinate transformation (13). Light, incident from
the right, is Fresnel-dragged by the moving medium: light propagating with the
flow is advanced, whereas light propagating against the current is retarded. The
wave should develop the phase slip shown in the left-hand side figure. However,
although the transformation (13) is exact, physical space-time would be described
in multi-valued coordinates here. Instead of the simple phase slip, the light turns
out to exhibit the characteristic interference pattern illustrated in the right-hand
side figure [74, 75].

and yet the medium is stationary. Consider in cylindrical coordinates the transformation

r z
ct = ct  + aϕ , r= , ϕ = ϕ , z= , (13)
n n
with the constants n and a. We obtain from our theory (8) that the medium is isotropic with
refractive index n and is magneto-electric with wi = (0, a/r, 0), which corresponds to a fluid
vortex with velocity profile u/c = w/(n2 − 1), as we see comparing equation (4) with the
constitutive equations of moving media [18, 45] in lowest relativistic order. Normally, a moving
medium Fresnel-drags light [22], but in the case of a vortex, light rays follow straight lines,
because the transformation (13) changes only time, such that straight rays in electromagnetic
space are mapped on to straight lines in physical space. Similarly, in the original Aharonov–
Bohm effect [74, 75] electron rays are not bent by a magnetic vortex, but they develop a phase
slip in the direction of incidence. In our case, when the light has passed the vortex, the time
change in the transformation (13) results in a phase slip of ±πak, where k is the wavenumber,
depending on whether the light propagates with or against the current, see figure 4. The coordinate
transformation to physical space-time is multi-valued, with a branch cut in the direction of
incidence, resembling the infinitely sheeted Riemann surfaces around a logarithmic branch point
[76], but the medium is single-valued and has the simple physical interpretation of a moving
fluid forming a vortex.

3.4. Artificial black holes

Moving isotropic media generate the effective space-time geometry discovered by Gordon in
1923 [34]. Consider an effectively 1D situation where the medium is moving in x-direction and
the electromagnetic field varies along the x-axis with field vectors pointing orthogonal to x.
New Journal of Physics 8 (2006) 247 ([Link]
11 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
An impedance-matched medium of refractive index n and velocity u is described by the inverse
metric tensor [22, 34, 38]
   
1 0 0 0 1 u/c 0 0
 0 −1 0 u/c u2 /c2 0
+ n −1
2
0 0
gαβ =
0
 . (14)
0 −1 0  1 − u2 /c2  0 0 0 0
0 0 0 −1 0 0 0 0

Both the velocity u and the index n may vary. We show that this effective geometry is generated
from empty space by the coordinate transformation

 c  c n ± u/c
ct = (t− + t+ ), x = (t− − t+ ), t± = t − dx, (15)
2 2 nu ± c
as long as we restrict our attention to fields varying in x-direction. For this, we use the
transformation rule (6), but from un-primed to primed space-time, and find that
 2 2 
αβ n (c − u2 ) n2 (c2 − u2 )
g = diag ,− 2 , −1, −1 . (16)
c2 − n2 u2 c − n2 u2
In electromagnetic space-time, we are free to apply the recipe (8) for assigning the medium
properties as well as in physical space. We find ε = µ = 1 in the directions orthogonal to x
and w = 0: vacuum, which proves our point. Here wave packets are superpositions of waves
propagating in positive or negative direction as functions of either t+ or t− . Substituting for t± the
representations (15) as functions of ct and x, we obtain the electromagnetic waves in physical
space. The transformation (15) describes the relativistic addition theorem of velocities [18, 42]
for the medium and light propagating in the positive or negative direction.
Interesting phenomena occur when the velocity of the medium, u, reaches the speed of light
in the medium, c/n, which is possible in theory and perhaps also in practice. In this case, the
integral in t± develops a logarithmic singularity. Light propagating against the current freezes with
exponentially increasing oscillations at a horizon [77], see figure 5. This horizon is completely
analogous to the event horizon of the black hole if the light is escaping from a superluminal
region and to a white hole if the light is attempting to enter a counter-propagating superluminal
medium [30]. A horizon cuts physical space-time into two distinct regions without possible
communication in one direction. It corresponds to two disconnected branches of multi-valued
electromagnetic space-time, where each branch maps to one of the regions in physical space.
On the other hand, in the case of perfect lensing the folds of electromagnetic space are finite
and connected. Horizons are predicted to exhibit remarkable quantum effects [77], in particular
Hawking radiation [78], that are extremely difficult to observe in astronomy, but may possibly
be demonstrated in laboratory analogues [21]–[31]. Our method suggests a magneto-electric
analogue of the event horizon, perhaps with metamaterials, if we interpret the effective geometry
(14) as generating the constitutive equations (4) expressed in terms of the D and H fields as

(n2 − u2 /c2 )Ey − (n2 − 1)uBz (c2 − n2 u2 )By − (n2 − 1)uEz

Dy = ε0 , Hy = ε0 . (17)
n(1 − u2 /c2 ) n(1 − u2 /c2 )

Here, the functions n and u of the moving medium appear as parameters of a magneto-
electric material at rest. The material establishes horizons at u = c/n without singular dielectric
New Journal of Physics 8 (2006) 247 ([Link]
12 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT

ct

Figure 5. Event horizons. Space-time diagram showing light propagation near

two artificial event horizons, a white-hole horizon at the left-hand side and a
black-hole horizon at the right-hand side. The horizons could be made by a
medium moving from the right-hand side to the left. The places x where the
medium velocity reaches the local speed of light in the medium are the horizons.
At the black-hole horizon the medium turns from subluminal to superluminal
velocity; at the white-hole horizon it returns to subluminal speed. The exponential
confinement near the horizons translates into exponential red shifts of optical
oscillations at the black hole and exponential blue shifts at the white hole.
Magneto-electric media could mimic the moving medium generating the horizons.

properties, in contrast to the previous proposal [24]. General relativity can be put to practical use
in electrical engineering, but electrical engineering may be also used for demonstrating some
elusive effects of general relativity.

4. Conclusions

Perfect invisibility devices [1]–[3], perfect lenses [9]–[15], the optical Aharonov–Bohm effect
[22, 23, 57, 58] and artificial event horizons [21]–[31] are all examples of one unifying concept:
electromagnetic media that facilitate coordinate transformations. Adopting ideas from general
relativity [34]–[43], we developed a concise formalism for finding the properties of metamaterials
that turn a desired function into fact. We extended the previously reported method [1]–[3] to media
that act as space-time transformations and that may exhibit negative refraction and other signs
of multi-valued transformations. The concepts of general relativity lead to practical applications
in electrical engineering.
Of course, here we exploited only some of the aspects of general relativity: we used the
physics of curved space, but not the physics [16] that creates space-time curvature in gravity.
In our case, electromagnetic metamaterials, not masses, create effective geometries. On the
other hand, we also extended some concepts beyond their normal use in general relativity. In
general relativity, coordinate transformations are theoretical techniques to describe the physics
of curved space-time, they have no influence on the physics itself, whereas here we considered

New Journal of Physics 8 (2006) 247 ([Link]

13 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
active transformations generated by real dielectric media. Transformation media do have physical
effects, sometimes dramatic ones.
The most interesting properties of transformation media seem to stem from non-trivial
topologies: spaces with holes in the case of invisibility devices, coordinate folds for negative
refraction, multi-valued transformations for the Aharonov–Bohm effect and multiple sheets of
electromagnetic space-time in the case of artificial event horizons. Our theory can be extended
in at least two directions. So far we have assumed that electromagnetic space-time is empty, but
one could easily incorporate media with non-trivial ε and µ here. We have also assumed that
the mapping between electromagnetic fields in real and fictitious space is exact. In practice, the
accuracy of geometrical optics [19] is often completely sufficient; one could further generalize
our method to include transformations that are only exact within the validity range of geometrical
optics. Optical conformal mapping [4, 5] is such an example. These more general transformation
media act as local transformations of the dispersion relation of light waves. They all share the
same spirit: electromagnetic media act as effective geometries.

Acknowledgments

We thank John Allen, Ildar Gabitov, Andrew Green, Chris Hooley, Natalia Korolkova and Irina
Leonhardt for their help and helpful comments. Our work was supported by the Leverhulme
Trust and the Engineering and Physical Sciences Research Council.

Appendix A

In this appendix, we show that the free-space Maxwell equations in generally covariant form are
equivalent to Maxwell’s equations in a material medium with constitutive equations (4) and (5).
The generally covariant Maxwell equations are [16, 42]

ε0 √
F[µν;λ] = ∂[λ Fµν] = 0, ε0 F;νµν = √ ∂ν ( −gF µν ) = j µ , (A.1)
−g
where Fµν denotes the electromagnetic field tensor, j µ is the four-current, the square brackets
denote antisymmetrization and the semi-colon indicates covariant differentiation. We employ
Einstein’s summation convention over repeated indices. The covariant tensor Fµν contains the E
and B fields in the usual manner:
 
0 Ex Ey Ez
−Ex 0 −cBz cBy
Fµν = −Ey
. (A.2)
cBz 0 −cBx
−Ez −cBy cBx 0

We define a quantity H µν with contravariant indices by

√ √
H µν = ε0 −gF µν = ε0 −ggµλ gνρ Fλρ (A.3)

1
⇒ Fµν = √ gµλ gνρ H λρ , (A.4)
ε0 −g
New Journal of Physics 8 (2006) 247 ([Link]
14 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
and regard H µν as being constructed from D and H fields as follows:
 
0 −Dx −Dy −Dz
Dx 0 −Hz /c Hy /c
H µν = Dy
. (A.5)
Hz /c 0 −Hx /c
Dz −Hy /c Hx /c 0
√
Then, introducing a new four-current J µ = −gj µ , Maxwell’s equations (A.1) can be written

∂[λ Fµν] = 0, ∂ν H µν = J µ , (A.6)

which are the electromagnetic equations in a material medium, described by the constitutive
equations (A.2)–(A.5), with free charge and current densities J µ .
To obtain relations between the vector fields D, H and E, B consider first the components
F0i ; from equations (A.2), (A.4) and (A.5), one obtains
1 c
Ei = √ (gi0 gj0 − gij g00 )Dj − √ [jkl]g0j gik Hl , (A.7)
ε0 −g ε0 −g
where [jkl] denotes the completely antisymmetric permutation symbol [16] with [xyz] = +1.
We simplify this result as follows. The identity
gµλ gλν = δνµ
gives
1
giλ gλ0 = 0 ⇒ gi0 = − 00
gij gj0 , (A.8)
g

1 ij
g0λ gλi = 0 ⇒ gi0 = − g gj0 , (A.9)
g00

gjλ gλi = gj0 g0i + gjk gki = δij . (A.10)

Use of equations (A.8) or (A.9) in equation (A.10) produces the two relations
   
1 i0 k0 1
g − 00 g g
ij
gkj = δj , g gkj −
i ik
gk0 gj0 = δij , (A.11)
g g00
which reveal inverse-related 3 × 3 matrices. In view of equations (A.10) and (A.11), transvection
of (A.7) by gli yields
√
ε0 −g ij ε0 c
Di = − g Ej + [ijk]gj0 Hk , (A.12)
g00 g00
the first of the constitutive equations (4) with (5).
To obtain the second constitutive relation, we employ the tensors dual to Fµν and H µν . This
requires use of the 4D Levi-Civita tensor [16], given by
√ 1
µνλρ = −g[µνλρ], µνλρ = − √ [µνλρ], [0123] = +1. (A.13)
−g
New Journal of Physics 8 (2006) 247 ([Link]
15 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
The dual tensors ∗F µν and ∗Hµν are defined by [16]
∗ µν
F = 21 µνλρ Fλρ ⇒ Fµν = 21 µνλρ ∗F λρ , (A.14)

∗
Hµν = 21 µνλρ H λρ ⇒ H µν = 21 µνλρ ∗H λρ , (A.15)

so they have components

 
0 −cBx −cBy −cBz
∗ µν 1 cBx 0 Ez −Ey 
,
F =√  (A.16)
−g cBy −Ez 0 Ex 
cBz Ey −Ex 0
 
0 Hx /c Hy /c Hz /c
√ −Hx /c 0 Dz −Dy
∗
Hµν = −g 
−Hy /c −Dz
. (A.17)
0 Dx 
−Hz /c Dy −Dx 0

Re-expressed in terms of the dual tensors, the constitutive equations (A.3) and (A.4) read
∗ √
Hµν = ε0 −ggµλ gνρ ∗F λρ , (A.18)

∗ µν 1
F = √ gµλ gνρ∗Hλρ . (A.19)
ε0 −g

Writing out ∗H0i using equations (A.16)–(A.18), one finds

ε0 c2 ε0 c
Hi = − √ (g00 gij − gi0 gj0 )Bj + √ [jkl]gj0 gik El . (A.20)
−g −g

Comparison of this with equations (A.7) and (A.12) shows that

√
−g 1
Bi = − 2 gij Hj − [ijk]gj0 Ek , (A.21)
ε0 c g00 ε0 cg00

which is the second of the constitutive equations (4) with (5).

Clearly, several other relations between D, H, E and B are contained in (A.3) and (A.4), and
(A.18) and (A.19). For example, to express D and H in terms of E and B, as is done in equation (17),
we need only take the time-space components of (A.3), obtaining
√  √
Di = ε0 −g gi0 gj0 − gij g00 Ej − ε0 c −g[jkl]gk0 gij Bl , (A.22)

and the required formulae are equations (A.20) and (A.22).

New Journal of Physics 8 (2006) 247 ([Link]

16 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
Appendix B

In this appendix, we show that the expressions for εij and µij derived in [56] and utilized in [1]
are a special case of the formalism used in this paper. Consider the effective εij and µij resulting
from a transformation of the spatial coordinates of a Minkowski system. From equation (5) we
obtain
√ √
εij = − −ggij ε , µij = − −ggij µ , (B.1)

where we have allowed for general isotropic permittivity and permeability ε = 1, µ = 1 in

electromagnetic space.
Rescale the spatial coordinate basis vectors ∂i so that they form a (in general non-coordinate)
basis ui of vectors of unit length:
1
ui = √ ∂i , g(ui , ui ) = 1 (no summation). (B.2)
−gii

Let ḡij be the components of the metric tensor in the basis ui :

1
ḡij = g(ui , uj ) = √ gij (no summation). (B.3)
gii gjj

We need to compute the triple product u1 · (u2 × u3 ) in the coordinate basis ∂i ; for this we require
the 3D Levi-Civita tensor in this basis:
√ 1
εijk = −g[ijk], εijk = − √ [ijk], (B.4)
−g

where g is both the determinant of the space-time metric and the negative determinant of the
spatial metric gij since g00 = 1, gi0 = 0. Using equations (B.2)–(B.4), we find

u1 · (u2 × u3 ) = − gij ui1 εjkl u2k u3l = −gij ui1 εjkl gkm um n
2 gln u3
1 1 1 1 1
= − g1j √ εjkl gk2 √ gl3 √ =√ √ [jkl]g1j g2k g3l
−g11 −g22 −g33 −g11 g22 g33 −g
√
1 1 −g
=√ √ g= √ . (B.5)
−g11 g22 g33 −g −g11 g22 g33

Using equations (B.3) and (B.5) we write (B.1) as

√ 1
εij = −ε u1 · (u2 × u3 ) −g11 g22 g33 √ ḡij (no summation). (B.6)
gii gjj

√ 1
µij = −µ u1 · (u2 × u3 ) −g11 g22 g33 √ ḡij (no summation). (B.7)
gii gjj
which are the expressions derived in [56]. Note that this form of the theory implicitly contains
the case of negative refraction, because for a transition to a left-handed system the cross products
change sign, but this fact has never been mentioned nor used so far.

New Journal of Physics 8 (2006) 247 ([Link]

17 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
References

[1] Pendry J B, Schurig D and Smith D R 2006 Science 312 1780

[2] Schurig D, Pendry J B and Smith D R 2006 Opt. Express at press (Preprint physics/0607205)
[3] Schurig D, Mock J J, Justice B J, Cummer S A, Pendry J B, Starr A F and Smith D R 2006 Science at press
[4] Leonhardt U 2006 Science 312 1777
[5] Leonhardt U 2006 New J. Phys. 8 118
[6] Hendi A, Henn J and Leonhardt U 2006 Phys. Rev. Lett. 97 073902
[7] Alu A and Engheta N 2005 Phys. Rev. E 72 016623
[8] Milton G W and Nicorovici N-A P 2006 Proc. R. Soc. Lond. A, published online 3 May:
doi:10.1098/rspa.2006.1715
[9] Pendry J B 2000 Phys. Rev. Lett. 85 3966
[10] Grbic A and Eleftheriades G V 2004 Phys. Rev. Lett. 92 117403
[11] Fang N, Lee H, Sun C and Zhang X 2005 Science 308 534
[12] Lee H, Xiong Y, Fang N, Srituravanich W, Durant S, Ambati M, Sun C and Zhang X 2005 New J. Phys. 7 255
[13] Melville D O S and Blaikie R J 2005 Opt. Express 13 2127
[14] Smith D R, Pendry J B and Wiltshire M C K 2004 Science 305 788
[15] Veselago V, Braginsky L, Shkover V and Hafner C 2006 J. Comput. Theor. Nanosci. 3 189
[16] Misner Ch W, Thorne K S and Wheeler J A 1999 Gravitation (New York: Freeman)
[17] Peacock J A 1999 Cosmological Physics (Cambridge: Cambridge University Press)
[18] Van Bladel J 1984 Relativity and Engineering (Berlin: Springer)
[19] Born M and Wolf E 1999 Principles of Optics (Cambridge: Cambridge University Press)
[20] Feynman R P, Leighton R B and Sands M 1983 The Feynman Lectures on Physics. Mainly Mechanics,
Radiation and Heat (Reading, MA: Addison Wesley) chapter 26
[21] Novello M, Visser M and Volovik G E (ed) 2002 Artificial Black Holes (Singapore: World Scientific)
[22] Leonhardt U and Piwnicki P 1999 Phys. Rev. A 60 4301
[23] Leonhardt U and Piwnicki P 2000 Phys. Rev. Lett. 84 822
[24] Reznik B 2000 Phys. Rev. D 62 044044
[25] Schützhold R, Plunien G and Soff G 2002 Phys. Rev. Lett. 88 061101
[26] Leonhardt U 2002 Nature 415 406
[27] Leonhardt U 2002 Phys. Rev. A 65 043818
[28] Schützhold R and Unruh W G 2002 Phys. Rev. D 66 044019
[29] Unruh W G and Schützhold R 2003 Phys. Rev. D 68 024008
[30] Leonhardt U 2003 Rep. Prog. Phys. 66 1207
[31] Schützhold R and Unruh W G 2005 Phys. Rev. Lett. 95 031301
[32] Barnes W L, Dereux A and Ebbesen Th W 2003 Nature 424 824
[33] Grigorenko A N, Geim A K, Gleeson H F, Zhang Y, Firsov A A, Khrushchev I Y and Petrovic J 2005 Nature
438 335
[34] Gordon W 1923 Ann. Phys., Lpz. 72 421 (in German)
[35] Tamm I Y 1924 J. Russ. Phys.—Chem. Soc. 56 248 (in Russian)
[36] Tamm I Y 1925 J. Russ. Phys.—Chem. Soc. 56 1 (in Russian)
[37] Plebanski J 1960 Phys. Rev. 118 1396
[38] Leonhardt U 2000 Phys. Rev. A 62 012111
[39] Leonhardt U 2006 Phys. Rev. A 73 032108
[40] Greenleaf A, Lassas M and Uhlmann G 2003 Physiol. Meas. 24 S1
[41] Greenleaf A, Lassas M and Uhlmann G 2003 Math. Res. Lett. 10 1
[42] Landau L D and Lifshitz E M 1995 The Classical Theory of Fields (Oxford: Butterworth-Heinemann)
[43] Schleich W and Scully M O 1984 General relativity and modern optics Les Houches Session 38 New Trends
in Atomic Physics (Amsterdam: Elsevier)

New Journal of Physics 8 (2006) 247 ([Link]

18 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
[44] Jackson J D 1998 Classical Electrodynamics (New York: Wiley)
[45] Landau L D and Lifshitz E M 1993 Electrodynamics of Continuous Media (Oxford: Butterworth-Heinemann)
[46] Lakhtakia A and Mackay T G 2004 J. Phys. A: Math. Gen. 37 L505
[47] Lakhtakia A and Mackay T G 2005 J. Phys. A: Math. Gen. 38 2545
[48] Lakhtakia A, Mackay T G and Setiawan S 2005 Phys. Lett. A 336 89
[49] Setiawan S, Mackay T G and Lakhtakia A 2005 Phys. Lett. A 341 15
[50] Mackay T G, Lakhtakia A and Setiawan S 2005 New J. Phys. 7 75
[51] Mackay T G, Lakhtakia A and Setiawan S 2005 New J. Phys. 7 171
[52] Mackay T G, Setiawan S and Lakhtakia A 2005 Eur. Phys. J. C 41 1
[53] Mackay T G, Lakhtakia A and Setiawan S 2005 Europhys. Lett. 71 925
[54] McCall M W 2005 J. Phys. A: Math. Gen. 38 2543
[55] McCall M W 2006 J. Mod. Opt. at press
[56] Ward A J and Pendry J B 1996 J. Mod. Opt. 43 773
[57] Hannay J H 1976 Cambridge University Hamilton Prize Essay (unpublished)
[58] Cook R J, Fearn H and Milonni P W 1995 Am. J. Phys. 63 705
[59] Nachman A I 1988 Ann. Math. 128 531
[60] Gbur G 2003 Prog. Opt. 45 273
[61] Leonhardt U 2003 IEEE J. Selected Topics Quantum Electron. 9 102
[62] Minkel J R 2002 Phys. Rev. Focus 9 23
[63] Pendry J B and Smith D R 2003 Phys. Rev. Lett. 90 029703
[64] Valanju P M, Walser R M and Valanju A P 2003 Phys. Rev. Lett. 90 029704
[65] Pendry J B 2003 Phys. Rev. Lett. 91 099701
[66] Nieto-Vesperinas M and Garcia N 2003 Phys. Rev. Lett. 91 099702
[67] Leonhardt U and Philbin T G 2006 quant-ph/0608115
[68] Milonni P W 1994 The Quantum Vacuum (London: Academic)
[69] Lamoreaux S K 1999 Am. J. Phys. 67 850
[70] Berry M V, Chambers R G, Large M D, Upstill C and Walmsley J C 1980 Eur. J. Phys. 1 154
[71] Roux P, deRosny J, Tanter M and Fink M 1997 Phys. Rev. Lett. 79 3170
[72] Vivanco F, Melo F, Coste C and Lund F 1999 Phys. Rev. Lett. 83 1966
[73] Stachel J 1982 Phys. Rev. D 26 1281
[74] Aharonov Y and Bohm D 1959 Phys. Rev. 115 485
[75] Peshkin M and Tonomura A 1989 The Aharonov–Bohm Effect (Berlin: Springer)
[76] Ablowitz M J and Fokas A S 1997 Complex Variables (Cambridge: Cambridge University Press)
[77] Brout R, Massar S, Parentani R and Spindel Ph 1995 Phys. Rep. 260 329
[78] Hawking S M 1974 Nature 248 30

New Journal of Physics 8 (2006) 247 ([Link]