Electromagnetic Fields and Waves: Lecture Notes On
Electromagnetic Fields and Waves: Lecture Notes On
Electromagnetic Fields and Waves: Lecture Notes On
(227-0052-10L)
February 9, 2013
Introduction
The properties of electromagnetic fields and waves are most commonly discussed
in terms of the electric field E(r, t) and the magnetic induction field B(r, t). The
vector r denotes the location in space where the fields are evaluated. Similarly, t
is the time at which the fields are evaluated. Note that the choice of E and B is ar-
bitrary and that one could also proceed with combinations of the two, for example,
with the vector and scalar potentials A and φ, respectively.
The fields E and B have been originally introduced to escape the dilemma of
“action-at-distance’, that is, the question of how forces are transferred between
two separate locations in space. To illustrate this, consider the situation depicted
in Figure 1. If we shake a charge at r1 then a charge at location r2 will respond.
But how did this action travel from r1 to r2 ? Various explanations were developed
over the years, for example, by postulating an aether that fills all space and that
acts as a transport medium, similar to water waves. The fields E and B are pure
constructs to deal with the “action-at-distance’ problem. Thus, forces generated by
?
q1 q2
r1 r2
1
2
electrical charges and currents are explained in terms of E and B, quantities that
we cannot measure directly.
Basic Properties
q v q v
v
Figure 2: The fields E and B depend on the inertial frame. An observer at rest
sees a B field when a charge at velocity v moves by (left). However, an observer
moving at the same speed will experience only an E field.
3
frame. To illustrate this consider the two situations shown in Figure 2 where an
observer is measuring the fields of a charge moving at velocity v. An observer at
rest will measure a B field whereas an observer moving at the same speed as the
charge will only experience only an E field. Why? Because the charge appears to
be at rest from the observer’s point of view.
where c = 2.99792456 .. 108 m/s is the speed of light. As shown in Figure 3, r′ is the
distance between the charge and the observer at the earlier time (t − r′ /c). Simi-
larly, nr′ is the unit vector point from the charge towards the observer at the earlier
time (t − r′ /c). Thus, the field at the observation point ro at the time t depends on
the motion of the charge at the earlier time (t − r′ /c)! The reason is that it takes a
time ∆t = r′ /c for the field to travel the distance r′ to the observer.
Let us understand the different terms in Eq. (2). The first term is proportional
to the position of the charge and describes a retarded Coulomb field. The second
term is proportional to the velocity of the charge. Together with the first term it
describes the instantaneous Coulomb field. The third term is proportional to the
acceleration of the charge and is associated with electromagnetic radiation.
E(ro, t)
r’
nr’ r
q
position at t
position at t- r’/c q
Figure 3: The field at the observation point ro at the time t depends on the motion
of the charge at the earlier time (t − r′ /c).
4
The objective of this course is to establish the theoretical foundations that lead
to Eq. (2) and to develop an understanding for the generation and propagation of
electromagnetic fields.
where rn denotes the position vector of the nth charge and ṙn = vn its velocity. The
total charge and current of the particle are obtained by a volume integration over ρ
and j. In terms of ρ and j, the force law in Eq.(1) can be written as
Z
F(r, t) = [ρ(r, t)E(r, t) + j(r, t) × B(r, t)] dV (5)
V
qn
rn
r
Figure 4: In the microscopic picture, optical radiation interacts with the discrete
charges qn of matter.
5
Pre-Maxwellian Electrodynamics
Let us review and summarize the laws of Ampère (Oersted), Faraday and Gauss,
as introduced in the courses Netzwerke und Schaltungen I & II.
1
Z Z
E(r, t) · n da = ρ(r, t) dV Gauss′ law (Cavendish 1772)
∂V ε0 V
∂
Z Z
E(r, t) · ds = − B(r, t) · n da Faraday′ s law (Faraday 1825)
∂A ∂t A
Z Z
B(r, t) · ds = µ0 j(r, t) · n da Ampere′ s law (Oersted 1819)
∂A A
Z
B(r, t) · n da = 0 No magnetic monopoles (6)
∂V
Vs Vs
µ0 = 4π 10−7 = 1.2566370 10−6 (magnetic permeability)
Am Am
1 As
ε0 = 2
= 8.8541878 10−12 (electric permittivity)
µ0 c Vm
where c = 2.99792458 108 m/s is the vacuum speed of light. Fig. 5 illustrates the
meaning of the four equations (6). The first equation, Gauss’ law, states that the
flux of electric field through a closed surface ∂V is equal to the total charge q
inside ∂V . The second equation, Faraday’s law, predicts that the electric field in-
tegrated along a loop ∂A corresponds to the time rate of change of the magnetic
flux through the loop. Similarly, Ampère’s law, states that the magnetic field inte-
grated along a loop ∂A is equal to the current flowing through the loop. Finally, the
fourth equation states that the flux of magnetic field through a closed surface is al-
ways zero which, in a analogy to Gauss’ law, indicates that there are no magnetic
charges.
a E b
q1 ∂
q3 A B
q2 ∂t
V E ds
c j d
B
A V
B ds
Figure 5: Illustration of (a) Gauss’ law, (b) Faraday’s law, (c) Ampere’s law, and (d)
the non-existence of magnetic charges.
7
Figure 6: Ampère’s law applied to a cube. Opposite faces cancel the magnetic
R
circulation ∂A B(r, t) · ds, predicting that the flux of current through any closed
surface is zero.
Let us consider Ampère’s law for the different end faces of a small cube (c.f.
Fig. 6). 1 It turns out that the magnetic field integrated along the circumference of
R
one end face ( ∂A B(r, t) · ds) is just the negative of the magnetic field integrated
along the circumference of the opposite end face. Thus, the combined flux is zero.
The same is true for the other pairs of end faces. Therefore, for any closed surface
Ampère’s law reduces to
Z
j(r, t) · n da = 0 Kirchhoff I . (7)
∂V
In other words, the flux of current through any closed surface is zero, - what flows
in has to flow out.
Eq. (7) defines the familiar current law of Kirchhoff (Knotenregel). On the other
hand, Kirchhoff’s voltage law (Maschenregel) follows from Faraday’s law if no time-
varying magnetic fields are present. In this case,
Z
E(r, t) · ds = 0 Kirchhoff II . (8)
∂A
The two Kirchhoff laws form the basis for circuit theory and electronic design.
1
An arbitrary volume can be viewed as a sum over infinitesimal cubes.
8
Chapter 1
Maxwell’s Equations
Eq. (7) is a statement of current conservation, that is, currents cannot be generated
or destroyed, the net flux through a closed surface is zero. However, this law
is flawed. For example, let’s take a bunch of identical charges and hold them
together (see Fig. 1.1). Once released, the charges will speed out because of
Coulomb repulsion and there will be a net outward current. Evidently, the outward
current is balanced by the decrease of charge inside the enclosing surface ∂V ,
and hence, Eq. (7) has to be corrected as follows
∂
Z Z
j(r, t) · n da = − ρ(r, t) dV (1.1)
∂V ∂t V
This equation describes the conservation of charge. It’s general form is found in
many different contexts in physics and we will encounter it again when we discuss
the conservation of energy (Poynting theorem).
Because Eq. (7) has been derived from Ampère’s law, we need to modify the
9
10 CHAPTER 1. MAXWELL’S EQUATIONS
latter in order to end up with the correct conservation law of Eq. (1.1). This is where
Maxwell comes in. He added an additional term to Ampère’s law and arrived at
1 ∂
Z Z Z
B(r, t) · ds = µ0 j(r, t) · n da + 2 E(r, t) · n da , (1.2)
∂A A c ∂t A
where 1/c2 = ε0 µ0 . The last term has the form of a time-varying current. Therefore,
ε0 ∂E/∂t is referred to as the displacement current.
We again apply this equation to the end faces of a small cube (c.f. Fig. 6) and,
as before, the left hand side vanishes. Thus,
1 ∂
Z Z
µ0 j(r, t) · n da + 2 E(r, t) · n da = 0 . (1.3)
∂V c ∂t ∂V
Substituting Gauss’ law from Eq. (6) for the second expression yields the desired
charge continuity equation (1.1).
In summary, replacing Ampère’s law in (6) by Eq. (1.2) yields a set of four equa-
tions for the fields E and B that are consistent with the charge continuity equation.
These four equations define what is called Maxwell’s integral equations.
e- e-
-
e -
e - e- e
-
e e- e- e- e-
-
e-
e e- e-
e- e- e-
e-
e-
t=0 t>0
Figure 1.1: Illustration of charge conservation. A bunch of identical charges is
held together at t = 0. Once released, the charges will spread out due to Coulomb
repulsion, which gives rise to a net outward current flow.
1.2. INTERACTION OF FIELDS WITH MATTER 11
So far we have discussed the properties of the fields E and B in free space. The
sources of these fields are charges ρ and currents j, so-called primary sources.
However, E and B can also interact with materials and generate induced charges
and currents. These are then called secondary sources.
where ρ is the charge density associated with primary sources. It is assumed that
these sources are not affected by the fields E and B. On the other hand, ρpol is
the charge density induced in matter through the interaction with the electric field.
It is referred to as the polarization charge density.1 On a microscopic scale, the
electric field slightly distorts the atomic orbitals in the material (see Fig. 1.2). On a
macroscopic scale, this results in an accumulation of charges at the surface of the
material (see Fig. 1.3). The net charge density inside the material remains zero.
P has units of Cb/m2 , which corresponds to dipole moment (Cb / m) per unit vol-
1
The B-field interacts only with currents and not with charges.
a b
- -
+ +
nucleus electron orbital
Figure 1.2: Microscopic polarization. An external electric field E distorts the orbital
of an atom. (a) Situation with no external field. (b) Situation with external field.
12 CHAPTER 1. MAXWELL’S EQUATIONS
ume (m3 ). Inserting Eqs. (1.4) and (1.5) into Gauss’ law yields
Z Z
[ε0 E(r, t) + P(r, t)] · n da = ρ(r, t) dV . (1.6)
∂V V
D = ε0 E + P . (1.7)
An electric field interacting with matter not only gives rise to polarization cur-
rents but also to conduction currents. We will denote the conduction current den-
sity as jcond . Furthermore, according to Ampère’s law, the interaction of matter with
a - b -
+- + - + +-
-- -
+ +
+ +--+
ρpol −ρpol
-
-+ + --
++
no net charge
magnetic fields can induce magnetization currents. We will denote the magneti-
zation current density as jmag . Taken all together, the total current density can be
written as
jtot (r, t) = j0 (r, t) + jcond (r, t) + jpol (r, t) + jmag (r, t) , (1.10)
where j0 is the source current density. In the following we will not distinguish
between source current and conduction current and combine the two as
j is simply the current density due to free charges, no matter whether primary or
secondary. On the other hand, jpol is the current density due to bound charges,
that is, charges that experience a restoring force to a point of origin. Finally, jmag is
the current density due to circulating charges, associated with magnetic moments.
We now introduce (1.10) into Ampère’s modified law (1.2) and obtain
Z
∂
Z
B · ds = µ0 j + jpol + ε0 E + jmag · n da , (1.12)
∂A A ∂t
where we have skipped the arguments (r, t) for simplicity. According to Eqs. (1.9)
and (1.7), the term inside the round brackets is equal to ∂D/∂t. To relate the
induced magnetization current to the B-field we define in analogy to Eq. (1.5)
Z Z
M(r, t) · ds = jmag (r, t) · n da , (1.13)
∂A A
with M being the magnetization. Inserting into Eq. (1.12) and rearranging terms
leads to
Z Z
1 ∂
B − M · ds = j+ D · n da , (1.14)
∂A µ0 A ∂t
The expression in brackets on the left hand side is called the magnetic field
1
H = B−M. (1.15)
µ0
It has units of A/m. The magnitude and the dynamics of magnetization currents
depends on the specific material properties [M = f (B)].
14 CHAPTER 1. MAXWELL’S EQUATIONS
Z Z
D(r, t) · n da = ρ(r, t) dV (1.16)
∂V V
∂
Z Z
E(r, t) · ds = − B(r, t) · n da (1.17)
∂A ∂t A
Z
∂
Z
H(r, t) · ds = j(r, t) + D(r, t) · n da (1.18)
∂A A ∂t
Z
B(r, t) · n da = 0 (1.19)
∂V
The displacement D and the induction B account for secondary sources through
These equations are always valid since they don’t specify any material properties.
To solve Maxwell’s equations (1.16)-(1.19) we need to invoke specific material
properties, i.e. P = f (E) and M = f (B), which are denoted constitutive relations.
For most of this course it will be more convenient to express Maxwell’s equations
in differential form. Using Stokes’ and Gauss’ theorems we can easily transform
Eq. (1.16)-(1.19). However, before doing so we shall first establish the notation
that we will be using.
1.4. MAXWELL’S EQUATIONS IN DIFFERENTIAL FORM 15
Differential Operators
The gradient operator (grad) will be represented by the nabla symbol (∇) and is
defined as a Cartesian vector
∂/∂x
∇ ≡ ∂/∂y . (1.21)
∂/∂z
It can be transformed to other coordinate systems in a straightforward way. Using
∇ we can express the divergence operator (div) as ∇· . To illustrate this, let’s
operate with ∇· on a vector F
∂/∂x Fx
∂ ∂ ∂
∇ · F = ∂/∂y · Fy = Fx + Fy + Fz . (1.22)
∂x ∂y ∂z
∂/∂z Fz
In other words, the divergence of F is the scalar product of ∇ and F.
The theorems of Gauss and Stokes have been derived in Analysis II. We won’t
reproduce the derivation and only state their final forms
Z Z
F(r, t) · n da = ∇ · F(r, t) dV Gauss′ theorem (1.28)
∂V V
Z Z
F(r, t) · ds = [∇ × F(r, t)] · n da Stokes′ theorem (1.29)
∂A A
Using these theorems we can turn Maxwell’s integral equations (1.16)-(1.19) into
differential form.
Applying Gauss’ theorem to the left hand side of Eq. (1.16) replaces the surface
integral over ∂V by a volume integral over V . The same volume integration is
performed on the right hand side, which allows us to write
Z
[∇ · D(r, t) − ρ(r, t)] dV = 0 . (1.30)
∂V
This result has to hold for any volume V , which can only be guaranteed if the
integrand is zero, that is,
∇ · D(r, t) = ρ(r, t) . (1.31)
This is Gauss’ law in differential form. Similar steps and arguments can be applied
to the other three Maxwell equations, and we end up with Maxwell’s equations in
differential form
∇ · D(r, t) = ρ(r, t) (1.32)
∂
∇ × E(r, t) = − B(r, t) (1.33)
∂t
∂
∇ × H(r, t) = D(r, t) + j(r, t) (1.34)
∂t
∇ · B(r, t) = 0 (1.35)
1.4. MAXWELL’S EQUATIONS IN DIFFERENTIAL FORM 17
It has to be noted that it was Oliver Heaviside who in 1884 has first written Maxwell’s
equations in this compact vectorial form. Maxwell had written most of his equa-
tions in Cartesian coordinates, which yielded long and complicated expressions.
Maxwell’s equations form a set of four coupled differential equations for the
fields D, E, B, and H. The components of these vector fields constitute a set of
16 unknowns. Depending on the considered medium, the number of unknowns
can be reduced considerably. For example, in linear, isotropic, homogeneous and
source-free media the electromagnetic field is entirely defined by two scalar fields.
Maxwell’s equations combine and complete the laws formerly established by Fara-
day, Oersted, Ampère, Gauss, Poisson, and others. Since Maxwell’s equations are
differential equations they do not account for any fields that are constant in space
and time. Any such field can therefore be added to the fields.
Let us remind ourselves that the concept of fields was introduced to explain
the transmission of forces from a source to a receiver. The physical observables
are therefore forces, whereas the fields are definitions introduced to explain the
troublesome phenomenon of the “action at a distance”.
∂
∇ · j(r, t) + ρ(r, t) = 0 (1.36)
∂t
After substituting the fields D and B in Maxwell’s curl equations by the expressions
in (1.20) and combining the two resulting equations we obtain the inhomogeneous
wave equations
1 ∂2E
∂ ∂P
∇×∇×E + 2 2
= −µ0 j+ + ∇×M (2.1)
c ∂t ∂t ∂t
1 ∂2H ∂P 1 ∂2 M
∇×∇×H + = ∇ × j + ∇ × − (2.2)
c2 ∂ t2 ∂t c2 ∂ t2
where we have skipped the arguments (r, t) for simplicity. The expression in the
round brackets corresponds to the total current density
∂P
j = j+ + ∇×M , (2.3)
∂t
where j is the source and the conduction current density, ∂P/∂t the polarization
current density, and ∇ × M the magnetization current density. The wave equations
as stated in Eqs. (2.1) and (2.2) do not impose any conditions on the media and
hence are generally valid.
We first consider the solution of the wave equations in free space, in absence of
matter and sources. For this case the right hand sides of the wave equations are
19
20 CHAPTER 2. THE WAVE EQUATION
zero. The operation ∇ × ∇× can be replaced by the identity (1.27), and since in
free space ∇ · E = 0 the wave equation for E becomes
1 ∂2
∇2 E(r, t) − E(r, t) = 0 (2.4)
c2 ∂ t2
with an identical equation for the H-field. Each equation defines three independent
scalar equations, namely one for Ex , one for Ey , and one for Ez .
In the one-dimensional scalar case, that is E(x, t), Eq. (2.4) is readily solved by
the ansatz of d’Alembert E(x, t) = E(x−ct), which shows that the field propagates
through space at the constant velocity c. To tackle three-dimensional vectorial
fields we proceed with standard separation of variables
∂2
T (t) + ω 2 T (t) = 0 (2.7)
∂ t2
ω2
∇2 R(r) + R(r) = 0 . (2.8)
c2
Note that both R(r) and T (t) are real functions of real variables.
where c′ω and c′′ω are real constants and cω = c′ω + ic′′ω is a complex constant. Thus,
according to ansatz (2.5) we find the solutions
In what follows, we will denote cω R(r) as the complex field amplitude and abbrevi-
ate it by E(r). Thus,
E(r, t) = Re{E(r) e−iωt } (2.11)
Notice that E(r) is a complex field whereas the true field E(r, t) is real. The sym-
bol E will be used for both, the real time-dependent field and the complex spatial
part of the field. The introduction of a new symbol is avoided in order to keep the
notation simple. Eq. (2.11) describes the solution of a time-harmonic electric field,
a field that oscillates in time at the fixed angular frequency ω. Such a field is also
referred to as monochromatic field.
To solve for the solutions of the Helmholtz equation (2.12) we use the ansatz
ω2
kx2 + ky2 + kz2 = (2.14)
c2
The left hand side can also be represented by k · bf k = k 2 . For the following we
assume that kx , ky , and kz are real. After inserting Eq. (2.13) into Eq. (2.11) we
find the solutions
E(r, t) = Re{E0 e±ik·r−iωt } (2.15)
which are called plane waves or homogeneous waves. Solutions with the + sign
in the exponent are waves that propagate in direction of k = [kx , ky , kz ] . They
are denoted outgoing waves. On the other hand, solutions with the − sign are
incoming waves and propagate against the direction of k.
22 CHAPTER 2. THE WAVE EQUATION
Although the field E(r, t) fulfills the wave equation it is not yet a rigorous solu-
tion of Maxwell’s equations. We still have to require that the fields are divergence
free, i.e. ∇·E(r, t) = 0. This condition restricts the k-vector to directions perpen-
dicular to the electric field vector (k·E0 = 0). Fig. 2.1 illustrates the characteristic
features of plane waves.
Let us consider a plane wave with real amplitude E0 and propagating in direc-
tion of the z axis. This plane wave is represented by E(r, t) = E0 cos[kz − ωt],
where k = |k| = ω/c. If we observe this field at a fixed position z then we’ll mea-
sure an electric field E(t) that is oscillating with angular frequency f = ω / 2π. On
the other hand, if we take a snapshot of this plane wave at t = 0 then we’ll ob-
serve a field that spatially varies as E(r, t = 0) = E0 cos[k z]. It has a maximum
at z = 0 and a next maximum at k z = 2π. The separation between maxima is
λ = 2π/k and is called the wavelength. After a time of t = 2π/ω the field reads
E(r, t = 2π/ω) = E0 cos[k z − 2π] = E0 cos[k z], that is, the wave has prop-
agated a distance of one wavelength in direction of z. Thus, the velocity of the
wave is v0 = λ/(2π/ω) = ω/k = c, the vacuum speed of light. For radio waves
λ E0
H0 k
Figure 2.1: Illustration of a plane wave. In free space, the plane wave propagates
with velocity c in direction of the wave vector k = (kx , ky , kz ). The electric field
vector E0 , the magnetic field vector H0 , and k are perpendicular to each other.
2.1. HOMOGENEOUS SOLUTION IN FREE SPACE 23
λ ∼ 1 km, for microwaves λ ∼ 1 cm, for infrared radiation λ ∼ 10 µm, for visible
light λ ∼ 500 nm, and for X-rays λ ∼ 0.1 nm, - the size range of atoms. Fig. 2.2
illustrates the length scales associated with the different frequency regions of the
electromagnetic spectrum.
A plane wave with a fixed direction of the electric field vector E0 is termed lin-
early polarized. We can form other polarizations states (e.g. circularly polarized
waves) by allowing E0 to rotate as the wave propagates. Such polarization states
can be generated by superposition of linearly polarized plane waves.
Plane waves are mathematical constructs that do not exist in practice because
their fields E and H are infinitely extended in space and therefore carry an infinite
amount of energy. Thus, plane waves are mostly used to locally visualize or ap-
proximate more complicated fields. They are the simplest form of waves and can
be used as a basis to describe other wave fields (angular spectrum representa-
tion). For an illustration of plane waves go to http://en.wikipedia.org/wiki/Plane wave.
Figure 2.2: Length-scales associated with the different frequency ranges of the
electromagnetic spectrum. From mynasadata.larc.nasa.gov.
24 CHAPTER 2. THE WAVE EQUATION
So far we have restricted the discussion to real kx , ky , and kz . However, this restric-
tion can be relaxed. For this purpose, let us rewrite the dispersion relation (2.14)
as q
kz = (ω 2/c2 ) − (kx2 + ky2 ) . (2.16)
If we let (kx2 + ky2 ) become larger than k 2 = ω 2 /c2 then the square root no longer
yields a real value for kz . Instead, kz becomes imaginary. The solution (2.15) then
turns into
E(r, t) = Re{E0 e±i(kx x+ky y)−iωt } e∓|kz |z . (2.17)
These waves still oscillate like plane waves in the directions of x and y, but they
exponentially decay or grow in the direction of z. Typically, they have a plane of
origin z = const. that coincides, for example, with the surface of an insulator or
metal. If space is unbounded for z > 0 we have to reject the exponentially growing
solution on grounds of energy conservation. The remaining solution is exponen-
tially decaying and vanishes at z → ∞ (evanescere = to vanish). Because of their
exponential decay, evanescent waves only exist near sources (primary or sec-
ondary) of electromagnetic radiation. Evanescent waves form a source of stored
energy (reactive power). In light emitting devices, for example, we want to convert
kx k plane waves
x kx2+k y2 = k2
ϕ kz kx
x evanescent waves
E
To summarize, for a wave with a fixed (kx , ky ) pair we find two different charac-
teristic solutions
Plane waves are oscillating functions in z and are restricted by the condition kx2 +
ky2 ≤ k 2 . On the other hand, for kx2 + ky2 > k 2 we encounter evanescent waves with
an exponential decay along the z-axis. Figure 2.3 shows that the larger the angle
between the k-vector and the z-axis is, the larger the oscillations in the transverse
plane will be. A plane wave propagating in the direction of z has no oscillations in
the transverse plane (kx2 + ky2 = 0), whereas, in the other limit, a plane wave propa-
gating at a right angle to z shows the highest spatial oscillations in the transverse
plane (kx2 + ky2 = k 2 ). Even higher spatial frequencies are covered by evanescent
waves. In principle, an infinite bandwidth of spatial frequencies (kx , ky ) can be
achieved. However, the higher the spatial frequencies of an evanescent wave are,
the faster the fields decay along the z-axis will be. Therefore, practical limitations
make the bandwidth finite.
The condition on the right corresponds to the dispersion relation (2.14). Further-
more, the divergence condition requires that E0 (kn , ωm ) · kn = 0. We have added
the argument (kn , ωm ) to E0 since each plane or evanescent wave in the sum is
characterized by a different complex amplitude.
26 CHAPTER 2. THE WAVE EQUATION
The solution (2.19) assumes that there is a discrete set of frequencies and
wavevectors. Such discrete sets can be generated by boundary conditions, for
example, in a cavity where the fields on the cavity surface have to vanish. In free
space, the sum in Eq. (2.19) becomes continuous and we obtain
Z Z
±ik·r−iωt 3
E(r, t) = Re E0 (k, ω) e dω d k with k · k = ω 2 /c2 (2.20)
k ω
We have replaced the complex amplitude E by Ê since we’re now dealing with an
amplitude per unit frequency, i.e. Ê = lim∆ω→0 [E/∆ω]. We have also included ω in
the argument of Ê since each solution of constant ω has its own amplitude.
which leads us to
Z ∞
E(r, t) = Ê(r, ω) e−iωt dω (2.23)
−∞
This is simply the Fourier transform of Ê. In other words, E(r, t) and Ê(r, ω) form
a time-frequency Fourier transform pair. Ê(r, ω) is also denoted as the temporal
2.2. SPECTRAL REPRESENTATION 27
spectrum of E(r, t). Note that Ê is generally complex, while E is always real. The
inverse transform reads as
∞
1
Z
Ê(r, ω) = E(r, t) eiωt dt . (2.24)
2π −∞
∇ · B̂(r, ω) = 0 (2.28)
Inserting the last expression into Eq. (2.24) yields the temporal spectrum of a
monochromatic wave
1
Ê(r, ω ′) = [E(r) δ(ω ′ − ω) + E∗(r) δ(ω ′ + ω)] . (2.30)
2
R
Here δ(x) = exp[ixt] dt/(2π) is the Dirac delta function. Iif we use Eq. (2.30)
along with similar expressions for the spectra of E, D, B, H, ρ0 , and j0 in Maxwell’s
28 CHAPTER 2. THE WAVE EQUATION
∇ · B(r) = 0 (2.34)
These equations are used whenever one deals with time-harmonic fields. They
are formally identical to the spectral representation of Maxwell’s equations (2.25)–
(2.28). Once a solution for the complex fields is found, the real time-dependent
fields are found through Eq. (2.11).
Detectors do not respond to fields, but to the intensity of fields, which is defined (in
free space) as
ε0
r
I(r) = E(r, t) · E(r, t) , (2.35)
µ0
with h..i denoting the time-average. For a monochromatic wave, as defined in
Eq. (2.29), this expression becomes
1 ε0
r
I(r) = |E(r)|2 . (2.36)
2 µ0
Coherent Fields
Coherence is a term that refers to how similar two fields are, both in time and
space. Coherence theory is a field on its own and we won’t dig too deep here.
Maximum coherence between two fields is obtained if the fields are monochro-
matic, their frequencies are the same, and if the two fields have a well-defined
phase relationship. Let’s have a look at the sum of two plane waves of the same
frequency ω
E(r, t) = Re{ E1 eik1 ·r + E2 eik2 ·r e−iωt } ,
(2.39)
I(x)
x
α β
k1 k2 I1+ I2
E1 E2
Figure 2.4: Interference of two plane waves incident at angles α and β. Left:
Interference pattern along the x-axis for two different visibilities, 0.3 and 1. The
average intensity is I1 + I2 .
30 CHAPTER 2. THE WAVE EQUATION
and let’s denote the plane defined by the vectors k1 and k2 as the (x,y) plane. Then,
k1 = (kx1 , ky1 , kz1 ) = k(sin α, cos α, 0) and k2 = (kx2 , ky2 , kz2 ) = k(− sin β, cos β, 0)
(see Figure 2.4). We now evaluate this field along the x-axis and obtain
1 ε0
r
2
I(x) = E1 eikx sin α + E2 e−ikx sin β (2.41)
2 µ0
1 ε0
r
E1 · E∗2 eikx(sin α+sin β) + E∗1 · E2 e−ikx(sin α+sin β)
= I1 + I2 +
2 µ0
p
= I1 + I2 + ε0 /µ0 Re E1 · E∗2 eikx(sin α+sin β) .
This equation is valid for any complex vectors E1 and E2 . We next assume that the
two vectors are real and that they are polarized along the z-axis. We then obtain
p
I(x) = I1 + I2 + 2 I1 I2 cos[kx(sin α + sin β)] . (2.42)
The cosine term oscillates between +1 and -1. Therefore, the largest and smallest
√
signals are Imin = (I1 + I2 ) ± 2 I1 I2 . To quantify the strength of interference one
defines the visibility
Imax − Imin
η = , (2.43)
Imax + Imin
which has a maximum value of η = 1 for I1 = I2 . The period of the interfer-
ence fringes ∆x = λ/(sin α + sin β) decreases with α and β and is shortest for
α = β = π/2, that is, when the two waves propagate head-on against each other.
In this case, ∆x = λ/2.
Incoherent Fields
Let us now consider a situation for which no interference occurs, namely for two
plane waves with different frequencies ω1 and ω2 . In this case Eq. (2.39) has to be
replaced by
I(x) = I1 + I2 , (2.46)
It has to be emphasized, that the two situations that we analyzed are extreme
cases. In practice there is no absolute coherence or incoherence. Any electromag-
netic field has a finite line width ∆ω that is spread around the center frequency ω.
In the case of lasers, ∆ω is a few MHz, determined by the spontaneous decay
rate of the atoms in the active medium. Thus, an electromagnetic field is at best
only partially coherent and its description as a monochromatic field with a single
frequency ω is an approximation.
The relationship (2.22) indicates that the positive frequency region contains all
the information of the negative frequency region. If we restrict the integration in
Eq. (2.23) to positive frequencies, we obtain what is called a complex analytic
signal Z ∞
+
E (r, t) = Ê(r, ω) e−iωt dω , (2.47)
0
with the superscript ‘+’ denoting that only positive frequencies are included. Sim-
ilarly, we can define a complex analytic signal E− that accounts only for negative
frequencies. The truncation of the integration range causes E+ and E− to become
complex functions of time. Because E is real, we have [E+ ]∗ = E− . By taking the
Fourier transform of E+ (r, t) and E− (r, t) we obtain Ê+ (r, ω) and Ê− (r, ω), respec-
tively. It turns out that Ê+ is identical to Ê for ω > 0 and it is zero for negative
32 CHAPTER 2. THE WAVE EQUATION
frequencies. Similarly, Ê− is identical to Ê for ω < 0 and it is zero for positive fre-
quencies. Consequently, Ê = Ê+ + Ê− . In quantum mechanics, Ê− is associated
with the creation operator ↠and Ê+ with the annihilation operator â.
Chapter 3
Constitutive Relations
Maxwell’s equations define the fields that are generated by currents and charges.
However, they do not describe how these currents and charges are generated.
Thus, to find a self-consistent solution for the electromagnetic field, Maxwell’s
equations must be supplemented by relations that describe the behavior of matter
under the influence of fields. These material equations are known as constitutive
relations.
we find that the lowest-order term depends linearly on E. In most practical situ-
ations the interaction of radiation with matter is weak and it suffices to truncate
the power series after the linear term. The nonlinear terms come into play when
the fields acting on matter become comparable to the atomic Coulomb potential.
This is the territory of strong field physics. Here we will entirely focus on the linear
properties of matter.
1
In some exotic cases we can have P = f [E, H] and M = f [H, E], which are so-called bi-
isotropic or bi-anisotropic materials. These are special cases and won’t be discussed here.
33
34 CHAPTER 3. CONSTITUTIVE RELATIONS
Z∞Z 0
D(r, t) = ε0 ε̃(r−r′, t−t′ ) E(r′ , t′ ) d3 r′ dt′ , (3.2)
−∞ −∞
which states that the response D at the location r and at time t not only depends
on the excitation E at r and t, but also on the excitation E in all other locations
r′ and all previous times t′ . The integrals represent summations over all space
and over all previous times. For reasons of causality (no response before excita-
tion), the time integral only runs to t′ = 0. The response function ε̃ is a tensor of
P
rank two. It maps a vector E onto a vector D according to Di = j ε̃ij Ej , where
{i, j} ∈ {x, y, z}. A material is called temporally dispersive if its response function
at time t depends on previous times. Similarly, a material is called spatially dis-
persive if its response at r depends also on other locations. A spatially dispersive
medium is also designated as a nonlocal medium.
Note that Eq. (3.2) is a convolution in space and time. Using the Fourier trans-
form with respect to both time and space, that is,
Z∞Z∞Z∞Z∞
1
D̂(kx , ky , kz , ω) = D(x, y, z, t) eikx x eiky y eikz z eiωt dx dy dz dt , (3.3)
(2π)4
−∞−∞−∞−∞
where ε is the Fourier transform of ε̃. Note that the response at (k, ω) now only de-
pends on the excitation at (k, ω) and not on neighboring (k′ , ω ′). Thus, a nonlocal
relationship in space and time becomes a local relationship in Fourier space! This
is the reason why life often is simpler in Fourier space.
with µ(ω) being the relative magnetic permeability. Notice that the spectral repre-
sentation of Maxwell’s equations [ (2.25)–(2.28)] is formally identical to the com-
plex notation used for time-harmonic fields [ (2.31)–(2.34)] . Therefore, Eqs. (3.5)
and (3.6) also hold for the complex amplitudes of time-harmonic fields
However, these equations generally do not hold for time-dependent fields E(r, t)!
One can use (3.7) for time-dependent fields only if dispersion can be ignored, that
is ε(ω) = ε and µ(ω) = µ. The only medium that is strictly dispersion-free is vac-
uum.
The linear relationships (3.7) are often expressed in terms of the electric and mag-
netic susceptibilites χe and χm , respectively. These are defined as
3.1.2 Conductivity
It turns out that the conduction current is accounted for by the imaginary part of
ε(ω) as we shall show in the following.
Let us explicitly split the current density j into a source and a conduction cur-
rent density according to Eq. (1.11). Maxwell’s curl equation for the magnetic
field (2.33) then becomes
We now introduce the linear relationships (3.7) and (3.9) and obtain
With the new definition of ε, the wave equations for the complex fields E(r) and
H(r) in linear media are
where k0 = ω/c denotes the vacuum wavenumber. Note that these equations are
also valid for anisotropic media, i.e. if ε and µ are tensors.
If µ is isotropic then we can multiply Eq. (3.13) on both sides with µ. Fur-
thermore, if there are no sources we can drop j0 and obtain the Helmholtz equa-
tion (2.12), but with the difference that now k 2 = k02 εµ, that is,
√
where n = εµ is called index of refraction.
Conductors
The conductivity σ is a measure for how good a conductor is. For example, quartz
has a conductivity of σSiO2 = 10−16 A / V m, and the conductivity of copper is
σAu = 108 A / V m. These values are different by 24 orders of magnitude! There
are hardly any other physical parameters with a comparable dynamic range.
The net charge density ρ inside a conductor is zero, no matter wether it trans-
ports a current or not. This seems surprising, but it directly follows from the charge
conservation (1.36) and Gauss’ law (1.32). Combining the two equations and us-
ing j = σE and D = ε0 εE yields
∂ σ
ρ(t) = − ρ(t) , (3.16)
∂t ε0 ε
Thus, any charge inside the conductor dissipates within a time of Tρ = ε0 ε/σ. For
a perfect conductor, σ → ∞ and hence ρ(t) = 0. For realistic conductors with finite
σ the characteristic time is Tρ ∼ 10−19 s, which is so short that it can be neglected.
When a charge moves through a conductor it undergoes collisions with the lat-
tice. After a collision event, the charge is accelerated by the external field until
it is slowed down by the next collision. For good conductors (copper), the time
between collisions is typically in the order of τ ∼ 10−14 s. The sequence of acceler-
ation and deceleration events results in a finite velocity vd for the charge, called the
drift velocity. The current density due to a charge density moving at finite speed
is j = q vd n, where n2 is the charge density, i.e. the number of charges per unit
volume. The drift velocity vd is proportional to the driving field E and the propor-
tionality constant is called mobility µ3 . Thus, σ = nqµ.
2
Not to be confused with the index of refraction n.
3
Not to be confused with the magnetic permittivity µ.
38 CHAPTER 3. CONSTITUTIVE RELATIONS
and because of j = σ E the same equation holds for the current density j. Note
that we’re using complex equation and that j(r, t) = Re{j(r) exp[−iω t]}.
jz
invariance in y and z we set jz (x) = A exp[B x] and insert into Eq. (3.18). Noting
√ √
that i = (1 + i)/ 2 we obtain
r
−(i−1) x/Ds 2
jz (x) = jz (x = 0) e with Ds = . (3.19)
σµ0 µ ω
The length Ds is called the skin depth. Since |jz (x)/jz (0)| = exp[−x/Ds ] it de-
scribes the penetration of fields and currents into the metal. Evidently, for a perfect
conductor (σ → ∞) the skin depth becomes Ds = 0, that is, all the current is trans-
ported on the surface of the metal. Ds also decreases with increasing frequency
ω, but eventually the result (3.19) becomes inaccurate because the polarization
current ∂P/∂t becomes stronger than the conduction current. In wires of finite
diameter the skin depth also depends on the curvature of the wire. At low frequen-
cies, the conductance of a wire scales with the cross-section of the wire, but at
high frequencies, the current is confined to the surface of the wire and hence the
conductance scales with the circumference.
40 CHAPTER 3. CONSTITUTIVE RELATIONS
Chapter 4
Material Boundaries
41
42 CHAPTER 4. MATERIAL BOUNDARIES
Eqs. (4.1) and (4.2) are only valid in the interior of the regions Di . However,
Maxwell’s equations must also hold on the boundaries ∂Dij . Due to the mate-
rial discontinuity at the boundary between two regions it is difficult to apply the
differential forms of Maxwell’s equations. We will therefore use the corresponding
integral forms (1.16) - (1.19). If we look at a sufficiently small section of the ma-
terial boundary we can approximate the boundary as being flat and the fields as
homogeneous on both sides (Fig. 4.1).
Let us first analyze Maxwell’s first equation (1.16) by considering the infinites-
imal rectangular pillbox illustrated in Fig. 4.1(a). Assuming that the fields are ho-
mogeneous on both sides, the surface integral of D becomes
Z
D(r) · n da = A [n · Di (r)] − A [n · Dj (r)] , (4.3)
∂V
where A is the top surface of the pillbox. The contributions of the sides dA dis-
appear if we shrink the thickness dL of the pillbox to zero. The right hand side of
Eq. (1.16) becomes
Z
ρ(r, t) dV = A dL ρi (r) + A dL ρj (r) , (4.4)
V
4.2. BOUNDARY CONDITIONS 43
Here, σ = dL ρi + dL ρj = dLρ is the surface charge density. Eq. (4.5) is the bound-
ary condition for the displacement D. It states that if the normal component of the
displacement changes across the boundary in point r, then the change has to be
compensated by a surface charge σ0 .
Let us now have a look at Faraday’s law, i.e. Maxwell’s second equation (1.17).
We evaluate the path integral of the E-field along the small rectangular path shown
in Fig. 4.1(b)
Z
E(r) · ds = L [n × Ei (r)] · ns − L [n × Ej (r)] · ns . (4.6)
∂A
The contributions along the normal sections vanish in the limit dL → 0. We have
introduced ns for the unit vector normal to the loop surface. It is perpendicular to n,
the unit vector normal to the boundary ∂Dij . For the right hand side of Eq. (1.17)
we find
∂
Z
− B(r) · ns da = 0 , (4.7)
∂t A
(a) (b)
∂D ij ∂D ij
Dj Dj
Di Di
. dL
n
dA .
n
.
A L
ns
dL
Figure 4.1: Integration paths for the derivation of the boundary conditions on the
interface ∂Dij between two neighboring regions Di and Dj .
44 CHAPTER 4. MATERIAL BOUNDARIES
that is, the flux of the magnetic field vanishes as the area is reduced to zero.
Combining the last two equations we arrive at
We obtain a similar result for the Ampère-Maxwell law (1.18), with the exception
that the flux of the current density does not vanish necessarily. If we allow for the
existence of a current density that is confined to a surface layer of infinitesimal
thickness then
Z
j(r) · ns da = L dL [ji (r) · ns ] + L dL [jj (r) · ns ] . (4.9)
A
Using an equation similar to (4.6) for H and inserting into Eq. (1.18) yields
The surface charge density σ and the surface current density K are confined to
the very interface between the regions Di and Dj . Such confinement is not found
in practice and hence, σ and K are mostly of theoretical significance. The exis-
tence of σ and K demands materials that perfectly screen any charges, such as a
metal with infinite conductivity. Such perfect conductors are often used in theoret-
ical models, but they are approximations for real behavior. In most cases we can
set σ = 0 and K = 0. Any currents and charges confined to the boundary Dij are
adequately accounted for by the imaginary part of the dielectric functions εi and εj .
Note that Eq. (4.13) and (4.14) yield two equations each since the tangent
of a boundary is defined by two vector components. Thus, the boundary condi-
tions (4.11)–(4.14) constitute a total of six equations. However, these six equations
4.3. REFLECTION AND REFRACTION AT PLANE INTERFACES 45
are not independent of each other since the fields on both sides of ∂Dij are linked
by Maxwell’s equations. It can be easily shown, for example, that the conditions for
the normal components are automatically satisfied if the boundary conditions for
the tangential components hold everywhere on the boundary and Maxwell’s equa-
tions are fulfilled in both regions.
Q = Qk + Q⊥ n , (4.15)
and we assume the we can ignore surface charges and currents (σ = 0 and K =
0), we can represent Eqs. (4.11)-(4.14) as
The index ‘1’ specifies the medium in which the plane wave is defined. At the inter-
face z = 0 the incident wave gives rise to induced polarization and magnetization
currents. These secondary sources give rise to new fields, that is, to transmitted
and reflected fields. Because of the translational invariance along the interface
we will look for reflected/transmitted fields that have the same form as the incident
field
E1r (r, t) = Re{E1r eik1r ·r−iωt } E2 (r, t) = Re{E2 eik2 ·r−iωt } , (4.22)
where the first field is valid for z < 0 and the second one for z > 0. The two new
wavevectors are defined as k1r = [kx1r , ky1r , kz1r ] and k1r = [kx2 , ky2 , kz2 ], respec-
tively. All fields have the same frequency ω because the media are assumed to be
linear.
a) z b) z (p)
(s) E2
E2
k2 k2
¡2+2 θ2 ¡2+2 θ2
x x
¡1+1 ¡1+1
(p)
k1 (s) k1 E1r
e1 e1 E1r e1 e1
(s)
E1 (p)
k1r
k1r E1
Figure 4.2: Reflection and refraction of a plane wave at a plane interface. (a)
s-polarization, and (b) p-polarization.
4.3. REFLECTION AND REFRACTION AT PLANE INTERFACES 47
since the reflected field is propagating away from the interface. To summarize, the
three waves (incident, reflected, transmitted) can be represented as
Since the transverse wavenumber kk = [kx2 + ky2 ]1/2 is defined by the angle of
incidence θ1 as q
kk = kx2 + ky2 = k1 sin θ1 , (4.28)
it follows from (4.27) that also kz1 and kz2 can be expressed in terms of θ1 .
If we denote the angle of propagation of the refracted wave as θ2 (see Fig. 4.2),
then the requirement that kk be continuous across the interface leads to kk =
k1 sin θ1 = k2 sin θ2 . In other words,
The incident plane wave can be written as the superposition of two orthogonally
polarized plane waves. It is convenient to choose these polarizations parallel and
perpendicular to the plane of incidence as
(s) (p)
E1 = E1 + E1 . (4.30)
(s) (p)
E1 is parallel to the interface, while E1 is perpendicular to the wavevector k1 and
(s)
E1 . The indices (s) and (p) stand for the German “senkrecht” (perpendicular) and
48 CHAPTER 4. MATERIAL BOUNDARIES
“parallel” (parallel), respectively, and refer to the plane of incidence. Upon reflec-
tion or transmission at the interface, the polarizations (s) and (p) are conserved,
which is a consequence of the boundary condition (4.18).
s-Polarization
For s-polarization, the electric field vectors are parallel to the interface. Expressed
in terms of the coordinate system shown in Fig. 4.2 they can be expressed as
0 0 0
E1 = E(s) E1r = E(s) E2 = E(s)
2 , (4.31)
1 1r
0 0 0
where the superscript (s) stands for s-polarization. The magnetic field is defined
through Faraday’s law (2.32) as H = (ωµ0µ)−1 [k × E], where we assumed linear
and isotropic material properties (B = µ0 µH). Using the electric field vectors (4.31)
we find
(s) (s)
−(kz1 /k1 )E1 (kz1 /k1 )E1r
1 1
H1 = 0 H1r = 0
Z1 Z1
(s) (s)
(kx /k1 )E1 (kx /k1 )E1r
(s)
−(kz2 /k2 )E2
1
H2 = 0 , (4.32)
Z2
(s)
(kx /k2 )E2
r
µ0 µi
Zi = (4.33)
ε0 εi
Notice that we are dealing with an inhomogeneous problem, that is, the re-
sponse of the system (reflection and refraction) depends on the excitation. There-
(s)
fore, there are only two unknowns in the fields (4.31) and (4.32), namely E1r and
(s) (s)
E2 . E1 is the exciting field, which is assumed to be known. To solve for the two
unknowns we require two boundary conditions. However, Eqs. (4.16)–(4.19) con-
stitute a total of six boundary equations, four for the tangential field components
and two for the normal components. However, only two of these six equations are
independent.
The two other tangential boundary conditions (continuity of Ex and Hy ) are trivially
fulfilled since Ex = Hy = 0. The boundary conditions for the normal components,
(4.16) and (4.17), yield equations that are identical to (4.34) and (4.35) and can
therefore be ignored.
(s)
Solving Eqs. (4.34) and (4.35) for E1r yields
(s)
E1r µ 2 k z1 − µ 1 k z2
(s)
= ≡ r s (kx , ky ) , (4.36)
E1 µ 2 k z1 + µ 1 k z2
(s)
with r s being the Fresnel reflection coefficient for s-polarization. Similarly, for E2
we obtain
(s)
E2 2µ2 kz1
(s)
= ≡ ts (kx , ky ) , (4.37)
E1 µ k
2 z1 + µ k
1 z2
where ts denotes the Fresnel transmission coefficient for s-polarization. Note that
according to Eq. (4.27), kz1 and kz2 are functions of kx and ky . kz1 can be expressed
in terms of the angle of incidence as kz1 = k1 cos θ1 .
The procedure for p-polarized fields is analogous and won’t be repeated here. Es-
sentially, p-polarization corresponds to an exchange of the fields (4.31) and (4.32),
50 CHAPTER 4. MATERIAL BOUNDARIES
that is, the p-polarized electric field assumes the form of Eq. (4.32) and the p-
polarized magnetic field the form of Eq. (4.31).The amplitudes of the reflected and
transmitted waves can be represented as
where r s and tp are the Fresnel reflection and transmission coefficients for p-
polarization, respectively. In summary, for a plane wave incident on an interface
between two linear and isotropic media we obtain the following reflection and trans-
mission coefficients 1
µ 2 k z1 − µ 1 k z2 ε 2 k z1 − ε 1 k z2
r s (kx , ky ) = r p (kx , ky ) = (4.39)
µ 2 k z1 + µ 1 k z2 ε 2 k z1 + ε 1 k z2
The sign of the Fresnel coefficients depends on the definition of the electric field
vectors shown in Fig. 4.2. For a plane wave at normal incidence (θ1 = 0), r s and
r p differ by a factor of −1. Notice that the transmitted waves can be either plane
waves or evanescent waves. This aspect will be discussed in Chapter 4.4.
The Fresnel reflection and transmission coefficients have many interesting pre-
dictions. For example, if a plane wave is incident on a glass-air interface (incident
from the optically denser medium, i. e. glass) then the incident field can be to-
tally reflected if it is incident at an angle θ1 that is larger than a critical angle θc .
In this case, one speaks of total internal reflection. Also, there are situations for
which the entire field is transmitted. According to Eqs. (4.39), for p-polarization
this occurs when ε2 kz1 = ε1 kz2 . Using kz1 = k1 cos θ1 and Snell’s law (4.29) we ob-
tain kz2 = k2 [1 − (n1 /n2 )2 sin2 θ1 ]1/2 . At optical frequencies, most materials are not
magnetic (µ1 = µ2 = 1), which yields tan θ1 = n2 /n1 . For a glass-air interface this
occurs for θ1 ≈ 57◦ . For s-polarization one cannot find such a condition. There-
fore, reflections from surfaces at oblique angles are mostly s-polarized. Polarizing
sunglasses exploit this fact in order to reduce glare. The Fresnel coefficients give
1
For symmetry reasons, some textbooks omit the square root term in the coefficient tp . In this
case, tp refers to the ratio of transmitted and incident magnetic field.
4.4. EVANESCENT FIELDS 51
rise to even more phenomena if we allow for two or more interfaces. Examples are
Fabry-Pérot resonances and waveguiding along the interfaces.
To derive the evanescent wave generated by total internal reflection at the sur-
face of a dielectric medium, we refer to the configuration shown in Fig. 4.2. We
choose the x-axis to be in the plane of incidence. Using the symbols defined in
Section 4.3, the complex transmitted field vector can be expressed as
(p)
−E1 tp (kx ) kz2 /k2
(s)
ik x + ik z
E2 =
E1 ts (kx ) e x
z2
, (4.41)
(p)
E1 tp (kx ) kx /k2
which can be expressed entirely by the angle of incidence θ1 using kx = k1 sin θ1 .
Note that we suppressed the harmonic time factor exp(−iωt). With this substitution
the longitudinal wavenumbers can be written as (cf. Eq. (4.27))
p p
kz1 = k1 1 − sin2 θ1 , kz2 = k2 1 − ñ2 sin2 θ1 , (4.42)
where we introduced the relative index of refraction
√
ε1 µ1
ñ = √ . (4.43)
ε2 µ2
52 CHAPTER 4. MATERIAL BOUNDARIES
For ñ > 1, with increasing θ1 the argument of the square root in the expression
of kz2 gets smaller and smaller and eventually becomes negative. The critical angle
θc can be defined by the condition
1 − ñ2 sin2 θ1 = 0 ,
(4.44)
which describes a refracted plane wave with zero wavevector component in the
z-direction (kz2 = 0). Consequently, the refracted plane wave travels parallel to the
interface. Solving for θ1 yields
θc = arcsin[1/ñ] . (4.45)
This equation describes a field that propagates along the surface but decays ex-
ponentially into the medium of transmittance. Thus, a plane wave incident at an
angle θ1 > θc creates an evanescent wave. Excitation of an evanescent wave with
a plane wave at supercritical incidence (θ1 > θc ) is referred to as total internal
reflection (TIR). For the glass/air interface considered above and an angle of in-
cidence of θi = 45◦ , the decay constant is γ = 2.22/λ. This means that already
at a distance of ≈λ/2 from the interface, the time-averaged field is a factor of e
smaller than at the interface. At a distance of ≈2λ the field becomes negligible.
The larger the angle of incidence θi the faster the decay will be. Note that the Fres-
nel coefficients depend on θ1 . For θ1 > θc they become complex numbers and,
consequently, the phase of the reflected and transmitted wave is shifted relative to
the incident wave. This phase shift is the origin of the so-called Goos–Hänchen
shift. Furthermore, for p-polarized excitation, it results in elliptic polarization of the
evanescent wave with the field vector rotating in the plane of incidence.
Figure 4.4: Spatial modulation of the standing evanescent wave along the propa-
gation direction of two interfering waves (x-axis) and the decay of the intensity in
the z-direction. The ordinate represents the measured optical power. From Appl.
Opt. 33, 7995 (1994).
54 CHAPTER 4. MATERIAL BOUNDARIES
For p- and s-polarized evanescent waves, the intensity of the evanescent wave
can be larger than that of the input beam. To see this we set z = 0 in Eq. (4.46)
and we write for an s- and p-polarized plane wave separately the intensity ratio
|E2(z = 0)|2 /|E1(z = 0)|2 . This ratio is equal to the absolute square of the Fresnel
transmission coefficient tp,s . These transmission coefficients are plotted in Fig. 4.5
for the example of a glass/air interface. For p-(s-)polarized light the transmitted
evanescent intensity is up to a factor of 9 (4) larger than the incoming intensity.
The maximum enhancement is found at the critical angle of TIR. The physical rea-
son for this enhancement is a surface polarization that is induced by the incoming
plane wave which is also represented by the boundary condition (4.17). A similar
enhancement effect, but a much stronger one, can be obtained when the glass/air
interface is covered by a thin layer of a noble metal. Here, so called surface plas-
mon polaritons can be excited.
p
6
|tp,s|2(z=0)
2 s
0
0 20 40 60 80
o
angle of incidence 1[ ]
Evanescent fields can be converted into propagating radiation if they interact with
matter. A plane interface will be used in order to create an evanescent wave by TIR
as before. A second parallel plane interface is then advanced toward the first inter-
face until the gap d is within the range of the typical decay length of the evanescent
wave. A possible way to realize this experimentally is to close together two prisms
with very flat or slightly curved surfaces as indicated in Fig. 4.6(b). The evanes-
cent wave then interacts with the second interface and can be partly converted
into propagating radiation. This situation is analogous to quantum mechanical tun-
neling through a potential barrier. The geometry of the problem is sketched in
Fig. 4.6(a).
The fields are most conveniently expressed in terms of partial fields that are
restricted to a single medium. The partial fields in media 1 and 2 are written as a
superposition of incident and reflected waves, whereas for medium 3 there is only
Figure 4.6: Transmission of a plane wave through a system of two parallel inter-
faces. In frustrated total internal reflection (FTIR), the evanescent wave created
at interface B is partly converted into a propagating wave by the interface A of a
second medium. (a) Configuration and definition of parameters. A, B: interfaces
between media 2, 3 and 1, 2, respectively. The reflected waves are omitted for
clarity. (b) Experimental set-up to observe frustrated total internal reflection.
56 CHAPTER 4. MATERIAL BOUNDARIES
a transmitted wave. The propagation character of these waves, i.e. whether they
are evanescent or propagating in either of the three media, can be determined
from the magnitude of the longitudinal wavenumber in each medium in analogy to
Eq. (4.42). The longitudinal wavenumber in medium j reads
q q
kj z = kj2 − kk2 = kj 1 − (k1 /kj )2 sin2 θ1 , j ∈ {1, 2, 3} , (4.48)
√
where kj = nj k0 = nj (ω/c) and nj = εj µj . In the following a layered system with
n2 < n3 < n1 will be discussed, which includes the system sketched in Fig. 4.6.
This leads to three regimes for the angle of incidence in which the transmitted
intensity as a function of the gap width d shows different behavior:
1. For θ1 < arcsin(n2 /n1 ) or kk < n2 k0 , the field is entirely described by propa-
gating plane waves. The intensity transmitted to a detector far away from the
second interface (far-field) will not vary substantially with gapwidth, but will
only show rather weak interference undulations.
2. For arcsin(n2 /n1 ) < θ1 < arcsin(n3 /n1 ) or n2 k0 < kk < n3 k0 the partial field
in medium 2 is evanescent, but in medium (3) it is propagating. At the sec-
ond interface evanescent waves are converted into propagating waves. The
intensity transmitted to a remote detector will decrease strongly with increas-
ing gapwidth. This situation is referred to as frustrated total internal reflection
(FTIR).
3. For θ1 > arcsin (n3 /n1 ) or kk > n3 k0 the waves in layer (2) and in layer (3)
are evanescent and no intensity will be transmitted to a remote detector in
medium (3).
If we chose θ1 such that case 2 is realized (FTIR), the transmitted intensity I(d)
will reflect the steep distance dependence of the evanescent wave(s) in medium
2. However, as shown in Fig. 4.7, I(d) deviates from a purely exponential behavior
because the field in medium 2 is a superposition of two evanescent waves of the
form
c1 e−γz + c2 e+γz . (4.49)
The second term originates from the reflection of the primary evanescent wave
(first term) at the second interface and its magnitude (c2 ) depends on the material
properties.
4.4. EVANESCENT FIELDS 57
transmitted intensity
0.8 (a)
0.6
(b)
0.4
0.2 (c)
0
0 0.2 0.4 0.6 0.8 1
d/
The discussion of FTIR in this section contains most of the ingredients ti un-
derstand the physics of optical waveguides. We will return to this topic towards the
end of this course.
58 CHAPTER 4. MATERIAL BOUNDARIES
Chapter 5
The equations established so far describe the behavior of electric and magnetic
fields. They are a direct consequence of Maxwell’s equations and the properties
of matter. Although the electric and magnetic fields were initially postulated to
explain the forces in Coulomb’s and Ampere’s laws, Maxwell’s equations do not
provide any information about the energy content of an electromagnetic field. As
we shall see, Poynting’s theorem provides a plausible relationship between the
electromagnetic field and its energy content.
If the scalar product of the field E and Eq. (1.34) is subtracted from the scalar
product of the field H and Eq. (1.33) the following equation is obtained
∂B ∂D
H · (∇ × E) − E · (∇ × H) = −H · −E· −j·E. (5.1)
∂t ∂t
Noting that the expression on the left is identical to ∇ · (E × H), integrating both
sides over space and applying Gauss’ theorem the equation above becomes
∂B ∂D
Z Z h i
(E × H) · n da = − H· +E· + j · E dV (5.2)
∂V V ∂t ∂t
Although this equation already forms the basis of Poynting’s theorem, more insight
is provided when B and D are substituted by the generally valid equations (1.20).
59
60 CHAPTER 5. ENERGY AND MOMENTUM
∂ 1
Z Z h i Z
(E × H) · n da + D · E + B · H dV = − j · E dV (5.3)
∂V ∂t V 2 V
1 ∂P ∂E i µ0 ∂M ∂H i
Z h Z h
− E· −P· dV − H· −M· dV.
2 V ∂t ∂t 2 V ∂t ∂t
This equation is a direct conclusion of Maxwell’s equations and has therefore the
same validity. Poynting’s theorem is more or less an interpretation of the equation
above. The left hand side has the general appearance of a conservation law, sim-
ilar to the conservation of charge encountered previously in Eq. (1.1).
1h i
W = D·E+B·H (5.4)
2
S = (E × H) (5.5)
If the medium within V is linear and non-dispersive, the two last terms in
Eq. (5.3) equal zero and the only term accounting for energy dissipation is j · E.
To understand this term, we consider the work done per unit time on a single
charge q. In terms of the velocity v of the charge and the force F acting on it,
the work per unit time is dW/dt = F · v. Using the Lorentz force in Eq. (1) gives
dW/dt = q E · v + q [v × B] · v. Because [v × B] · v = [v × v] · B = 0 we obtain
5.1. POYNTING’S THEOREM 61
In most practical applications we are interested in the mean value of S, that is,
the value of S averaged over several oscillation periods. This quantity describes
the net power flux density and is needed, for example, for the evaluation of radiation
patterns. Assuming that the fields are harmonic in time, linear and non-dispersive,
then the two last terms in Eq. (5.3) disappear. Furthermore, we assume that the
energy density (5.4) only accounts for polarization and magnetization currents that
a re loss-free, that is, all losses are associated with the j·E term. The time average
of Eq. (5.3) then becomes
1
Z Z
hS(r)i·n da = − Re {j∗ (r)· E(r)} dV (5.6)
∂V 2 V
where we have used complex notation. The term on the right defines the mean
energy dissipation within the volume V . hSi represents the time average of the
Poynting vector
1
hS(r)i = Re {E(r) × H∗ (r)} (5.7)
2
In the far-field, that is, far from sources and material boundaries, the electro-
magnetic field can be locally approximated by a plane wave (see Section 2.1.1).
The electric and magnetic fields are in phase, perpendicular to each other, and the
ratio of their amplitudes is constant. hSi can then be expressed by the electric field
alone as
1 1 1 ε0
r
2
hS(r)i = |E(r)| nr = ni |E(r)|2 (5.8)
2 Zi 2 µ0
√
where nr represents the unit vector in radial direction, ni = εi µi is the index of
refraction, and Zi is the wave impedance (4.33).
62 CHAPTER 5. ENERGY AND MOMENTUM
The surface integral of hSi correspond to the total power generated or dissi-
pated inside the enclosed surface, that is,
Z Z
P̄ = hS(r)i·n da = I(r) da (5.9)
∂V ∂V
Let us consider a single dielectric interface that is irradiated by a plane wave under
conditions of total internal reflection (TIR) (c.f. Section 4.4). For non-absorbing
media and for supercritical incidence, all the power of the incident wave is reflected.
One can anticipate that because no losses occur upon reflection at the interface
there is no net energy transport into the medium of transmittance. In order to
prove this fact we have to investigate the time-averaged energy flux across a plane
parallel to the interface. This can be done by considering the z-component of the
Poynting vector (cf. Eq. (5.7))
1
hSiz = Re Ex Hy∗ − Ey Hx∗ ,
(5.10)
2
where all fields are evaluated in the upper medium, i.e. the medium of trans-
mittance. Applying Maxwell’s equation (2.32) to the special case of a plane or
evanescent wave, allows us to express the magnetic field in terms of the electric
field as
ε0 ε
r
k
H= ×E . (5.11)
µ0 µ k
Introducing the expressions for the transmitted field components of E and H into
Eq. (5.10), it is straightforward to prove that hSiz vanishes and that there is no net
energy transport in the direction normal to the interface.
On the other hand, when considering the energy transport along the interface
(hSix ), a non-zero result is found:
2
1 ε2 µ2
r 2
s 2 (s) p 2 (p)
hSix = sin θ1 |t | E1 + |t | E1 e−2γz . (5.12)
2 ε1 µ1
Thus, an evanescent wave transports energy along the surface, in the direction of
the transverse wavevector.
5.1. POYNTING’S THEOREM 63
The two last terms in Eq. (5.3) strictly vanish only in a linear medium with no dis-
persion and no losses. The only medium fulfilling these conditions is vacuum. For
all other media, the last two terms only vanish approximately. In this section we
consider a linear medium with a frequency-dependent and complex ε and µ.
Let us return to the Poynting theorem stated in Eq. (5.2). While the left hand
side denotes the power flowing in or out of the volume V , the right hand side
denotes the power dissipated or generated in the volume V . The three terms on
the right hand side are of similar form and so we start by considering first the
electric energy term E · (∂D/∂t). The electric energy density wE at the time t is
t
∂D(r, t′ ) ′
Z
wE (r, t) = E(r, t′) · dt . (5.13)
−∞ ∂ t′
We now express the fields E and D in terms of their Fourier transforms as E(t′ ) =
Ê(ω) exp[−iωt′ ] dω and D(t′ ) = D̂(ω) exp[−iωt′ ] dω, respectively. In the last ex-
R R
used D̂(−ω ′ ) = D̂∗ (ω ′ ) since D(t) is real (c.f. Eq. (2.22)). Using the linear relation
D̂ = ε0 ε Ê and inserting the Fourier transforms in Eq. (5.13) yields
Z ∞Z ∞ ′ ∗ ′
ω ε (ω ) ′
wE (r, t) = ε0 ′
Ê(ω) · Ê∗ (ω ′ ) ei(ω −ω)t dω ′ dω , (5.14)
−∞ −∞ ω − ω
where we have carried out the differentiation and integration over time and as-
sumed that the fields were zero at t → −∞. For later purposes it is advantageous
to represent the above result in different form. Using the substitutions u′ = −ω and
u = −ω ′ and making use of Ê(−u) = Ê∗ (u) and ε(−u) = ε∗ (u) gives an expression
similar to Eq. (5.14) but in terms of u and u′ . Finally, we add this expression to
Eq. (5.14) and take one half of the resulting sum, which yields
ε0 ∞ ∞ ω ′ ε∗ (ω ′) − ωε(ω)
Z Z
′
wE (r, t) = ′
Ê(ω) · Ê∗ (ω ′ ) ei(ω −ω)t dω ′ dω . (5.15)
2 −∞ −∞ ω −ω
Similar expressions are obtained for the magnetic term H · (∂B/∂t) and the dissi-
pative term j · E in Eq. (5.2).
If ε(ω) is a complex function then wE not only accounts for the energy density
built up in the medium but also for the energy transferred to the medium, such as
64 CHAPTER 5. ENERGY AND MOMENTUM
It can be shown that Eq. (5.17) also holds for quasi-monochromatic fields which
have frequency components ω only in a narrow range about a center frequency ω0 .
Such fields can be represented as
which is known as the slowly varying amplitude approximation. Here, E0 (r, t) is the
slowly varying (complex) amplitude and ω0 is the ’carrier’ frequency. The envelope
E0 spans over many oscillations of frequency ω0 .
d [ω ε′(ω)] D d [ω µ′(ω)] D
E E
W̄ = ε0 E · E + µ0 H·H (5.19)
dω dω
where E = E(r, t) and H = H(r, t) are the time-dependent fields. Notice, that ω is
the center frequency of the spectra of E and H. For a medium with negligible dis-
persion this expression reduces to the familiar W̄ = (1/2) [εo ε′ |E0|2 + µo µ′ |H0|2 ],
5.2. THE MAXWELL STRESS TENSOR 65
which follows from Eq. (5.4) using the dispersion-free constitutive relations. Be-
cause of d(ωε′)/dω > 0 and d(ωµ′)/dω > 0 the energy density is always positive,
even for metals with ε′ < 0.
In this section we use Maxwell’s equations to derive the conservation law for lin-
ear momentum in an electromagnetic field. The net force exerted on an arbitrary
object is entirely determined by Maxwell’s stress tensor. In the limiting case of an
infinitely extended object, the formalism renders the known formulas for radiation
pressure.
The general law for forces in electromagnetic fields is based on the conser-
vation law for linear momentum. To derive this conservation law we will consider
Maxwell’s equations in vacuum. In this case we have D = ε0 E and B = µ0 H.
Later we will relax this constraint. The conservation law for linear momentum is
entirely a consequence of Maxwell’s equations (1.16) - (1.19) and of the Lorentz
force law (5), which connects the electromagnetic world with the mechanical one.
ε0 (∇ × E) × E = (5.21)
∂/∂x(Ex2 − E 2/2) + ∂/∂y(Ex Ey ) + ∂/∂z(Ex Ez )
ε0 ∂/∂x(Ex Ey ) + ∂/∂y(Ey2 − E 2/2) + ∂/∂z(Ey Ez ) − ε0 E ∇·E
2 2
∂/∂x(Ex Ez ) + ∂/∂y(Ey Ez ) + ∂/∂z(Ez − E /2)
↔
= ∇ · [ε0 EE − (ε0 /2)E 2 I] − ρE .
66 CHAPTER 5. ENERGY AND MOMENTUM
where Eq. (1.32) has been used in the last step. The notation EE denotes the
↔
outer product, E 2 = Ex2 + Ey2 + Ez2 is the electric field strength, and I denotes the
unit tensor. A similar expression can be derived for µ0 (∇ × H) × H. Using these
two expressions in Eq. (5.20) we obtain
1 ↔ d 1
∇ · [ε0 EE − µ0 HH − (ε0 E 2 + µ0H 2 ) I] = [E × H] + ρ E + j × B . (5.22)
2 dt c 2
The expression in brackets on the left hand side is called Maxwell’s stress tensor
↔
in vacuum, usually denoted as T. In Cartesian components it reads as
↔ 1 2 2
↔
T = ε0 EE − µ0 HH − (ε0 E + µ0 H ) I = (5.23)
2
ε0 (Ex2 −E 2/2) + µ0 (Hx2 −H 2/2) ε0 Ex Ey + µ0 Hx Hy
ε0 Ex Ey + µ0 Hx Hy ε0 (Ey −E 2/2) + µ0 (Hy2 −H 2/2)
2
ε0 Ex Ez + µ0 Hx Hz ε0 Ey Ez + µ0 Hy Hz
ε0 Ex Ez + µ0 Hx Hz
ε0 Ey Ez + µ0 Hy Hz
After integration of Eq. (5.22) over an arbitrary volume V which contains all sources
ρ and j we obtain
d 1
Z Z Z
↔
∇· T dV = [E × H] dV + [ρ E + j × B] dV . (5.24)
V dt c 2 V V
The last term is recognized as the mechanical force (cf. Eq. (5)). The volume inte-
gral on the left can be transformed to a surface integral using Gauss’s integration
law Z Z
↔ ↔
∇· T dV = T ·n da . (5.25)
V ∂V
∂V denotes the surface of V , n the unit vector perpendicular to the surface, and
da an infinitesimal surface element. We then finally arrive at the conservation law
for linear momentum
d
Z
↔
T (r, t) · n(r) da = Gfield + Gmech (5.26)
∂V dt
Here, Gmech and Gfield denote the mechanical momentum and the field momentum,
respectively. In Eq. (5.26) we have used Newton’s expression of the mechanical
5.2. THE MAXWELL STRESS TENSOR 67
force F = d/dtGmech and the definition of the field momentum (Abraham density)
1
Z
Gfield = 2 [E × H] dV (5.27)
c V
This is the momentum carried by the electromagnetic field within the volume V . It
is created by the dynamic terms in Maxwell’s curl equations. The time-derivative
of the field momentum is zero when it is averaged over one oscillation period and
hence the average mechanical force becomes
Z
↔
hFi = hT (r, t)i · n(r) da (5.28)
∂V
with h...i denoting the time average. Equation (5.28) is of general validity. It allows
the mechanical force acting on an arbitrary body within the closed surface ∂V to
be calculated (see Figure 5.1). The force is entirely determined by the electric
and magnetic fields on the surface ∂V . It is interesting to note that no material
properties enter the expression for the force; the entire information is contained
in the electromagnetic field. The only material constraint is that the body is rigid.
If the body deforms when it is subject to an electromagnetic field we have to in-
clude electrostrictive and magnetostrictive forces. Since the enclosing surface is
∂V
incident field B
F
scattered field
Figure 5.1: The mechanical force F acting on the object B is entirely determined
by the electric and magnetic fields at an arbitrary surface ∂V enclosing B.
68 CHAPTER 5. ENERGY AND MOMENTUM
arbitrary the same results are obtained whether the fields are evaluated at the sur-
face of the body or in the far-field. It is important to note that the fields used to
calculate the force are the self-consistent fields of the problem, which means that
they are a superposition of the incident and the scattered fields. Therefore, prior to
calculating the force, one has to solve for the electromagnetic fields. If the object
B is surrounded by a medium that can be represented accurately enough by the
dielectric constant ε and magnetic susceptibility µ, the mechanical force can be
calculated in the same way if we replace Maxwell’s stress tensor Eq. (5.23) by
↔ 1 ↔
T = [ε0 εEE − µ0 µHH − (ε0 εE 2 + µ0 µH 2 ) I] (5.29)
2
Using Maxwell’s curl equation (1.33) we find for the magnetic field
p n o
H(r, t) = ε0 /µ0 E0 Re [eikz − r e−ikz ] e−iωt ny . (5.31)
E0
H0
nx
nz
ny A
acting on the medium. Using the fields of Eqs. (5.30) and (5.31) we find that the
first two terms in Maxwell’s stress tensor Eq. (5.23) give no contribution to the
radiation pressure. The third term yields
↔ 1 ε0 2
hT (r, t)i · nz = − hε0 E 2 + µ0 H 2 i nz = E [1 + |r|2 ] nz . (5.33)
2 2 0
Using the definition of the intensity of a plane wave I0 = (ε0 /2) c E02, c being the
vacuum speed of light, we can express the radiation pressure as
I0
P = [1 + R] , (5.34)
c
with R = |r|2 being the reflectivity. For a perfectly absorbing medium we have
R = 0, whereas for a perfectly reflecting medium R = 1. Therefore, the radiation
pressure on a perfectly reflecting medium is twice as high as for a perfectly ab-
sorbing medium.
Radiation
In this chapter we will discuss the emission of electromagnetic radiation from ele-
mentary sources. A stationary charge yields a static electric field, but it does not
radiate. Similarly, a uniformly moving charge (a current) yields a static magnetic
field, but it does not radiate. It is the acceleration of charge that gives rise to ra-
diation. The smallest radiating unit is a dipole, an electromagnetic point source.
According to linear response theory, a point source excitation yields the system
response function, from which we can calculate the fields of more complicated
sources by using the superposition principle. The system response function is
also referred to as the Green function.
Before calculating the fields radiated by elementary sources let us discuss an im-
portant mathematical concept, namely the concept of the Green function. Consider
the following general, inhomogeneous equation:
71
72 CHAPTER 6. RADIATION
↔
In a next step, assume that Eq. (6.3) has been solved and that G is known.
Postmultiplying Eq. (6.3) with B(r′ ) on both sides and integrating over the volume
V in which B 6= 0 gives
Z Z
↔
′ ′ ′
L G (r, r ) B(r ) dV = B(r′ ) δ(r − r′ ) dV ′ . (6.4)
V V
The right hand side simply reduces to B(r) and with Eq. (6.1) it follows that
Z
↔
L A(r) = L G (r, r′) B(r′ ) dV ′ . (6.5)
V
If on the right hand side the operator L is taken out of the integral, the solution of
Eq. (6.1) can be expressed as
Z
↔
A(r) = G (r, r′) B(r′) dV ′ . (6.6)
V
Thus, the solution of the original equation can be found by integrating the product
of the dyadic Green function and the inhomogeneity B over the source volume V .
6.2. SCALAR AND VECTOR POTENTIALS 73
strictly valid and special care must be applied if the integrand is not well behaved.
↔
Most often G is singular at r = r′ and an infinitesimal exclusion volume surround-
ing r = r′ has to be introduced. As long as we consider field points outside of the
source volume V , i.e. r 6∈ V , we do not need to consider these tricky issues.
The E and B fields define a total of six functions in space and time. It turns out, that
these fields are not independent and that one needs fewer functions to uniquely
determine the electromagnetic field. The vector potential A and the scalar po-
tential φ constitute a set of only four functions which, depending on the type of
problem, can be reduced to even fewer functions. These potentials are also of key
importance in quantum mechanics.
∂
E(r, t) = − A(r, t) − ∇φ(r, t) (6.7)
∂t
B(r, t) = ∇ × A(r, t) (6.8)
It turns out that these definitions of vector potential A and scalar potential φ are
not unique. If the potentials are replaced by new potentials Ã, φ̃ according to
with χ(r, t) being an arbitrary gauge function, Maxwell’s equations remain unaf-
fected. This is easily seen by introducing the above substitutions in the definitions
of A and φ.
74 CHAPTER 6. RADIATION
Lorenz Gauge
1 ∂2
2
∇ − 2 2 A = −µ0 jtot (6.13)
c ∂t
1 ∂2
2 1
∇ − 2 2 φ = − ρtot (6.14)
c ∂t ε0
1
This has the form of a continuity equation (A is the current density and φ/c2 is the charge
density).
6.2. SCALAR AND VECTOR POTENTIALS 75
Thus, we obtain two decoupled partial differential equations of the same form for
A and φ. Note, that one ends up with the same differential equations by a proper
choice of the gauge function χ (6.9).
The advantage of the Lorenz gauge is that the vectorial differential equation for
A is decoupled into a set of three independent scalar differential equations, that
is, each vector component Ai depends only on the source component jtoti . There
is no mixing of components i ∈ [x, y, z].
Coulomb Gauge
1 ∂2
2 1 ∂φ
∇ − 2 2 A = −µ0 jtot + ∇
c ∂t c2 ∂t
1
∇2 φ = − ρtot . (6.15)
ε0
Here, the scalar potential φ is determined by a Poisson equation, that is, there is no
retardation and φ is an instantaneous function. The Coulomb gauge is mostly used
for problems in quantum optics and is less important for this course. There are
many more gauges, which we won’t discuss here. Among them are the Poincaré
gauge, the Landau gauge, and the Weyl gauge. We will be mostly dealing with the
Lorenz gauge.
Note that by going from E, H to A, φ we reduced the field parameters from six to
four (three per vector and one per scalar). It turns out that the four parameters are
still redundant and that they can be reduced even more. One way is to introduce
the so-called Hertz potential Π(r), which has only three components. The vector
and scalar potentials are related to Π as A = (1/c2) ∂Π/∂t and φ = −∇ · Π, re-
spectively. Using so-called Debye potentials is yet another representation of fields,
but these won’t be discussed here.
76 CHAPTER 6. RADIATION
In this section we will derive the electromagnetic field of a dipole source, the small-
est radiating system. Mathematically, the dipole source corresponds to a delta
excitation, and the response to a delta excitation is the Green function discussed
previously. Any source can be thought of as being composed of individual point
sources with different origins. In other words, any macroscopic source volume can
be chopped up into little infinitesimal cubes, each of which carries a current that is
represented by a delta function.
where [j · ns ]da is the current flowing through the cross-sectional area da. The
product of da and ds defines the infinitesimal source volume dV .
Let us now consider an arbitrary macroscopic current density j(r) that is entirely
contained within the volume V . We can express this current density in terms of a
sum of microscopic point current densities. In terms of the Dirac delta function δ
ns ns
+q +ρ
ds da ds
-q −ρ
Here, j(r′ )δ(r − r′ )dV ′ are elementary point currents. Using Eq. (6.17) we obtain
the current density of an elementary dipole source
∂
j(r, t) = p(t) δ(r − r0 ) (6.20)
∂t
To calculate the fields of a dipole we will use the Lorenz gauge. The appeal of the
Lorenz gauge is its symmetry, that is, there is a scalar wave equation of the form
[∇2 − (1/c2 )∂ 2 /∂t2 ] Ψ = Θ for any of the field components Ax , Ay , Az , φ.
Let us start with deriving the vectorfield of a dipole with a harmonic time depen-
dence. In this case, p(t) = Re{p exp[−iωt]}, with p being a complex amplitude.
Thus, we can use complex notation and the equations for the components of A
become
2
∇ + k 2 Ai (r) = iωµ0 pi δ(r − r′ ) .
(6.21)
where we used Eq. (6.20) for the current density of a dipole field. Let us now define
the scalar Green function as G0 = i Ai/ωµ0 pi . Then, Eq. (6.21) turns into
We have included the origin of the the point source (r′ ) in the argument of G0 to
remind us where the origin of the point source is. In other words G0 (r, r′ ) is the
response at r to to a dipole source at r′ .
In free space, G0 (r, r′) must be point symmetric, because the source depends
only on the radial distance R = |r − r′ | and things don’t change if the coordinate
system is rotated around the origin r = r′ . To solve Eq. (6.22) we will try the
following ansatz
R G0 = a1 eikR + a2 e−ikR , (6.23)
which is a superposition of an outgoing and an incoming wave. After inserting
into Eq. (6.22) and integrating on both sides over a small spherical volume ∆V
centered at R = 0 and with radius ro . We obtain
2 1 1 1
Z Z
2
∇ dV + k dV = (6.24)
∆V R ∆V R a1 + a2
The second term integrates to 2πk 2 ro2 and the first term is calculated as
Z
1 1 nR · nR
Z Z
∇· ∇ dV = ∇ · nR da = − da = −4π . (6.25)
R R R2
∆V ∂∆V ∂∆V
′
′ ei k |r−r |
G0 (r, r ) = (6.26)
4π|r−r′|
where we used the definition of the delta function. We now assume that the ob-
servation point r is outside the source volume described by the coordinate r′ . In
this case, we can swap the sequence of integration and differentiation in Eq. (6.28)
and obtain
Z
2 2
G0 (r, r′ ) jtoti (r′ ) dV ′ = −µ0 jtoti (r) .
∇ + k µ0 (6.29)
V
Thus, the solution for A turns out to be the linear superposition of dipole fields with
different origins r′ and different weights jtot .
Now that we have derived the vector potential A of an oscillation dipole, we find
the magnetic field using B = ∇ × A and the electric field using Maxwell’s equation
E = i(ω/k 2)∇ × B. Skipping the details of the calculation, we find
↔
E(r) = ω 2µ0 G0 (r, r′) p (6.31)
↔
h i
H(r) = −iω ∇× G0 (r, r′) p (6.32)
↔
where we introduced the so-called dyadic Green function G0 defined as
↔
′
↔ 1
G0 (r, r ) = I + 2 ∇∇ G0 (r, r′ ) (6.33)
k
↔
with G0 being the scalar Green function (6.26) and I being the unit tensor. Notice
↔ ↔
that G0 is a tensor. It is straightforward to calculate G0 in the major three coordinate
↔
systems. In a Cartesian system G0 can be written as
ikR − 1 ↔ 3 − 3ikR − k 2 R2 RR
↔
′ exp(ikR)
G0 (r, r ) = 1+ 2 2 I + (6.34)
4πR k R k 2 R2 R2
80 CHAPTER 6. RADIATION
where R is the absolute value of the vector R = r−r′ and RR denotes the outer
product of R with itself. Equation (6.34) defines a symmetric 3×3 matrix
Gxx Gxy Gxz
↔
G0 = Gxy Gyy Gyz , (6.35)
which, together with Eqs. (6.31) and (6.32), determines the electromagnetic field
of an arbitrary electric dipole p with Cartesian components px , py , pz . The tensor
↔
[∇× G0 ] can be expressed as
↔
exp(ikR) k R × I
↔
′ 1
∇× G0 (r, r ) = i− , (6.36)
4πR R kR
↔
where R × I denotes the matrix generated by the cross-product of R with each
↔
column vector of I .
|p| cos ϑ exp(ikr) 2 2 2i
Er = k − , (6.41)
4πε0 ε r k 2 r 2 kr
|p| sin ϑ exp(ikr) 2 1 i
Eϑ = k − −1 , (6.42)
4πε0ε r k 2 r 2 kr
r
|p| sin ϑ exp(ikr) 2 i ε0 ε
Hϕ = k − −1 . (6.43)
4πε0ε r kr µ0 µ
The fact that Er has no far-field term ensures that the far-field is purely trans-
verse. Furthermore, since the magnetic field has no terms in (kr)−3 the near-field
is dominated by the electric field (see Fig. 6.3). This justifies a quasi-electrostatic
consideration.
ϑ r
p
y
ϕ
Figure 6.2: The fields of a dipole are most conveniently represented in a spherical
coordinate system (r, ϑ, ϕ) in which the dipole points along the z-axis (ϑ = 0).
82 CHAPTER 6. RADIATION
1000 1000
transverse field longitudinal field
10 10
1 1
–1
r r–2
Figure 6.3: Radial decay of the dipole’s transverse and longitudinal fields.
The curves correspond to the absolute value of the expressions in brackets of
Eqs. (6.41) and (6.42), respectively. While both the transverse and the longitudi-
nal field contribute to the near-field, only the transverse field survives in the far-
field. Notice that the intermediate-field with (kr)−2 does not really show up for the
transverse field. Instead the near-field dominates for (kr) < 1 and the far-field for
(kr) > 1.
It is instructive to also have a look at the phase of the dipole field since close
to the origin it deviates considerably from the familiar phase of a spherical wave
exp[ikr]. The phase of the field is defined relative to the oscillation of the dipole pz .
In Fig. 6.4 we plot the phase of the field Ez along the x-axis and along the z-axis
(c.f. Fig. 6.2). Interestingly, at the origin the phase of the transverse field is 180◦ out
of phase with the dipole oscillation (Fig. 6.4(a)). The phase of the transverse field
then drops to a minimum value at a distance of x ∼ λ/5 after which it increases
and then asymptotically approaches the phase of a spherical wave with origin at
the dipole (dashed line). On the other hand, the phase of the longitudinal field,
shown in Fig. 6.4(b), starts out to be the same as for the oscillating dipole, but it
runs 90◦ out of phase for distances z ≫ λ. The reason for this behavior is the
missing far-field term in the longitudinal field (c.f. Eq. (6.41). The 90◦ phase shift is
6.3. DIPOLE RADIATION 83
(a) (b)
6π/4 3π/4
5π/4 2π/4
Arg(Ez / pz)
Arg(Ez / pz )
π π/4
3π/4 0
0 1 2 3 4 5 0 1 2 3 4
kx kz
Figure 6.4: Phase of the electric field near the origin. (a) Phase of the transverse
field Ez evaluated along the x-axis. At the origin, the electric field is 180◦ out of
phase with the dipole. The phase drops to a minimum at a distance of x ∼ λ/5. For
larger distances, the phase approaches that of a spherical wave exp[ik r] (dashed
line). (b) Phase of the longitudinal field Ez evaluated along the z-axis. At the origin,
the electric field is in phase with the dipole. At larger distances, the phase is 90◦
out of phase with a spherical wave exp[ikr] (dashed line).
due to the intermediate field represented by the Green function in Eq. (6.39). The
same intermediate field is also responsible for the dip near x ∼ λ/5 in Fig. 6.4(a).
This phase dip is of relevance for the design of multi-element antennas, such as
the Yagi-Uda antennas. It is important to remember that close to the source the
phase of the field does not evolve linearly with distance and that the phase can be
advanced or delayed by small distance variations.
Eq. 5.9)
1
Z
P̄ = Re {E(r) × H∗ (r)}·n da . (6.44)
2 ∂V
Because we chose a spherical surface, the normal vector n is a radial vector and
hence we only need to calculate the radial component of hSi. Using Eqs. (6.42)
and (6.43) we find
1
Z
Re Eϑ Hϕ∗ sin ϑ dϑ dϕ ,
P̄ = (6.45)
2 ∂V
which yields
|p|2 n3 ω 4 |p|2 ω k 3
P̄ = = (6.46)
4πε0ε 3 c3 12πε0 ε
We find that the radiated power scales with the fourth power of the frequency and
that only the far-field of the dipole contributes to the net energy transport.
To determine the radiation pattern we calculate the power P̄ (ϑ, ϕ) radiated into
an infinitesimal unit solid angle dΩ = sin ϑ dϑ dϕ and divide by the total radiated
power P̄
P̄ (ϑ, ϕ) 3
= sin2 ϑ . (6.47)
P̄ 8π
z z
x 10’000
Figure 6.5: Electric energy density outside a fictitious sphere enclosing a dipole
p = pz . (Left) Close to the dipole’s origin the field distribution is elongated along
the dipole axis (near-field). (Right) At larger distances the field spreads transverse
to the dipole axis (far-field).
6.4. DIPOLE RADIATION IN ARBITRARY ENVIRONMENTS 85
Thus, in the far-field most of the energy is radiated perpendicular to the dipole mo-
ment (see Fig. 6.47) and there is no radiation at all in the direction of the dipole.
According to Poynting’s theorem (cf. Eq. 5.6) the radiated power P̄ of any
current distribution with a harmonic time dependence has to be identical to the
rate of energy dissipation dW/dt given by
dW 1
Z
= − Re{j∗ · E} dV, (6.48)
dt 2 V
V being the source volume. The current density j is either a source current that
generates the fields, or a loss current that is associated with thermal losses. Ei-
ther way, j represents both energy sources and energy sinks. If we introduce the
dipole’s current density from Eq. (6.20) we obtain the important result
ω
P̄ = Im {p∗· E(r0)} (6.49)
2
where the field E is evaluated at the dipole’s origin r0 . This equation can be rewrit-
ten in terms of the Green function by using Eq. (6.31) as
ω 3 |p|2 h n↔ o i
P̄ = np · Im G (r0 , r0 ) · np , (6.50)
2 c2 ε 0 ε
with np being the unit vector in the direction of the dipole moment.
At first sight it seems not possible to evaluate Eq. (6.49) since exp(ikR)/R ap-
pears to be infinite at r = r0 . As we shall see this is not the case. We first note that
due to the dot product between p and E we need only to evaluate the component
of E in the direction of p. Choosing p = |p| nz we calculate Ez as
|p| eikR h 2 1 ik i
Ez = k sin2 ϑ + 2 (3 cos2 ϑ − 1) − (3 cos2 ϑ − 1) . (6.51)
4 π ε0 ε R R R
86 CHAPTER 6. RADIATION
Since the interesting part is the field at the origin of the dipole, the exponential term
is expanded into a series [exp(ikR) = 1 + ikR + (1/2)(ikR)2 + (1/6)(ikR)3 + · · · ]
and the limiting case R → 0 is considered. Thus,
dW ω ω |p| 2 n2
3 2
o |p| 2 ω
= lim |p| Im{Ez } = lim k +R (..) + .. = k 3,
dt R→0 2 8 π ε0 ε R→0 3 12 π ε0 ε
(6.52)
which is identical with Eq. (6.46). Thus, Eq. (6.49) leads to the correct result de-
spite the apparent singularity at R = 0.
As illustrated in Fig. 6.6, we decompose the electric field at the dipole’s position
Es
p
E0
Figure 6.6: Illustration of dipole radiation in inhomogeneous environments. The
total field is composed of a primary field E0 directly radiated by the dipole and
a secondary field Es that is emitted by the dipole and then scattered at inhomo-
geneities in the environment.
6.5. FIELDS EMITTED BY ARBITRARY SOURCES 87
as
E(r0) = E0 (r0 ) + Es (r0 ) , (6.53)
where E0 and Es are the primary dipole field and the scattered field, respectively.
Introducing Eq. (6.53) into Eq. (6.49) allows us to split the rate of energy dissipa-
tion P = dW/dt into two parts. The contribution of E0 has been determined in
Eqs. (6.46) and (6.52) as
|p|2 ω 3
P̄0 = k , (6.54)
12π ε0 ε
which allows us to write for the normalized rate of energy radiation
P̄ 6πε0 ε 1
= 1 + 2 3
Im{p∗ · Es (r0 )} (6.55)
P̄0 |p| k
Thus, the change of energy dissipation depends on the secondary field of the
dipole. This field corresponds to the dipole’s own field emitted at a former time. It
arrives at the position of the dipole after it has been scattered in the environment.
According to Eq. (6.31) the E-field can be expressed in terms of a dyadic (ten-
↔
sorial) Green function as E(r) = ω 2 µ0 G0 (r, r′) p, where r′ is the origin of the
dipole. We can rewrite this equation as
Z
↔
2
E(r) = ω µ0 G0 (r, r′′ ) p δ(r′ − r′′ ) dV ′′ (6.56)
V
Using Eq. (6.20) for the current density of a dipole [ j(r′′ ) = −iω p δ(r′ − r′′ )] and
substituting into Eq. (6.56) above, yields
Z
↔
E(r) = i ωµ0 G0 (r, r′ ) j(r′ ) dV ′ (6.57)
V
88 CHAPTER 6. RADIATION
G(r,r')
j(r') E(r)
r' r
V
↔
Figure 6.7: Illustration of the dyadic Green function G (r, r′). The Green function
renders the electric field at the field point r due to a single point source j at the
source point r′ . Since the field at r depends on the orientation of j the Green
function must account for all possible orientations in the form of a tensor.
where j now is an arbitrary current density distribution within the source volume V .
We could have derived Eq. (6.57) also by following a more formal way using the
definition of the Green function as described in Section 6.1. Fig. 6.7 illustrates the
meaning of Eq. (6.57): The volume V is subdivided into infinitesimal units, each
of which occupied by a point source with weight j(r′ ). In a similar way we find the
solution for the magnetic field as
Z
↔
H(r) = [∇× G0 (r, r′)] j(r′ ) dV ′ (6.58)
V
Note that mathematically, the fields E and H above are particular solutions of the
differential equations (2.1) and (2.2). For a complete solution we need to super-
impose the homogeneous solutions, which are solutions of (2.1) and (2.2) with the
right sides being zero. These homogeneous solutions are fields that are present
even in absence of the sources j.
So far we have considered the fields generated by a source that is oscillating har-
monically in time with angular frequency ω. But what if the time dependence is
6.6. SOURCES WITH ARBITRARY TIME-DEPENDENCE 89
arbitrary, for example, a short pulse? In these cases we can employ Fourier trans-
forms, which describe an arbitrary time dependence by a superposition of time
harmonic dependences (see Section 2.2).
where, according to Eq. (6.26), Ĝ0 = exp(i k(ω) |r − r′ |) / (4π|r −r′|), with k(ω) =
n(ω) ω/c. ĵtot denotes the Fourier transform of an arbitrary time-dependent current
density j(r, t), that is,
Z ∞
1
ĵtot (r, ω) = ĵtot (r, t) eiωt dt . (6.60)
2π −∞
1 δ[t − |r − r′ |n/c]
G0 (r, r′, t) = . (6.64)
2 |r−r′|
2
R
exp[ixy]dy = 2πδ[x].
90 CHAPTER 6. RADIATION
Thus, the Green function in time domain is a simple delta function evaluated at the
earlier time t′ = t−nR/c, where t is the current time and R is the distance between
source point r′ and observation point r.
These equations state that the fields A and φ at the location r and time t are deter-
mined by the sources jtot and ρtot at location r′ at the earlier time t−|r−r′ |n/c. The
earlier time is a consequence of the speed of light c: it takes a time of |r − r′ |n/c for
the fields to travel a distance of |r−r′ | in a medium with index of refraction n. Thus,
Maxwell’s equations explain the mysterious “action-at-distance” phenomenon dis-
cussed in the introduction of this course (see Fig. 1). It has to be emphasized that
the index of refraction n is assumed to be dispersion-free, which is an approxima-
tion. The only material that is truly dispersion-free is vacuum (n = 1).
To find the fields E and H we insert the solutions of A and φ into Eqs. (6.7)
and (6.8). The calculation is not straightforward because A and φ depend on the
retarded time t − |r − r′ |n/c. Therefore, an operation on the spatial coordinates
(e.g. ∇×) is implicitly also an operation on time. We will not go through this exer-
cise and only mention that the solution is identical with Eq. (2) if we express ρ and
j with the charge and current densities of a discrete charge (c.f. Eq. 3 and 4). The
result has three terms: the first term depends on the charge, the second term on
the velocity of the charge, and the third term on the acceleration of charge. It is
the latter that is associated with electromagnetic radiation.
6.6. SOURCES WITH ARBITRARY TIME-DEPENDENCE 91
We have calculated the fields of a dipole with harmonic time dependence in Sec-
tion 6.3.2. These fields were expressed in spherical vector components (see
Eq. 6.41 - 6.43). Note that these fields are the complex amplitudes and that the
time-dependent fields are arrived at by multiplying with exp[−iωt] and taking the
real part.
Remember, that Maxwell’s equations for the complex amplitudes of time har-
monic fields (Eq. 2.31 - 2.34) are identical with Maxwell’s equations for the Fourier
transforms of fields with arbitrary time dependence (Eq. 2.25 - 2.28) . Therefore,
the solutions are identical as well. For example, the spectrum of the dipole’s Eϑ
field is
sin ϑ exp[ik(ω) r] 2 1 i
Êϑ (r, ω) = p̂(ω) k (ω) 2 − − 1 , (6.68)
4πε0 ε(ω) r k (ω) r 2 k(ω) r
j(r,t) E(r,t)
Fourier Fourier
^ ^
j(r,ω) n(r,ω) E(r,ω)
Maxwell
To solve this integral we set ω 2 p̂(ω) = ψ̂(ω), where the Fourier transform of ψ̂(ω) is
ψ(t). Using the translation property of Fourier-transforms we obtain
Z ∞
ψ̂(ω) e−iω(t−r/c) dω = ψ(t − r/c) . (6.70)
∞
Thus, it remains to solve for ψ(t):
Z ∞
d2
ψ(t) = ω 2 p̂(ω) e−iωt dω = − 2 p(t) . (6.71)
∞ dt
Putting the pieces together we finally find
sin ϑ 1 d2 p(τ )
f
Eϑ (r, t) = (6.72)
4πε0 c2 r dτ 2 τ =t−r/c
1 d2
sin ϑ 1 1 d
Eϑ (t) = − + + 2 p(τ ) (6.74)
4πε0 r 3 2
c r dτ c r dτ 2 τ =t−r/c
1 d2
r
sin ϑ ε0 1 d
Hϕ (t) = − + p(τ ) (6.75)
4πε0 µ0 c r 2 dτ c2 r dτ 2
τ =t−r/c
We see that the far-field is generated by the acceleration of the charges that con-
stitute the dipole moment. Similarly, the intermediate-field and the near-field are
generated by the speed and the position of the charges, respectively.
6.7. THE LORENTZIAN POWER SPECTRUM 93
d2 d
2
p(t) + γ0 p(t) + ω02 p(t) = 0 . (6.76)
dt dt
The natural frequency of the oscillator is ω0 and its damping constant is γ0 . The
solution for p is n √ 2 2
o
p(t) = Re p0 e−iω0 1−(γ0 /4ω0 ) t eγ0 t/2 . (6.77)
Typically, the damping constant is much smaller than the oscillation frequency
p
(γ0 ≪ ω0 ), which implies 1 − (γ02 /4ω02 ) ≈ 1.
The spectrum Êϑ (ω) detected by the observer is (cf. Eq. (2.24))
Z ∞
1
Êϑ (ω) = Eϑ (t) eiωt dt. (6.78)
2π r/c
Here we set the lower integration limit to t = r/c because the dipole starts emitting
at t = 0 and it takes the time t = r/c for the radiation to propagate to the observa-
tion point. Therefore Eϑ (t < r/c) = 0. Inserting the solution for the dipole moment
from Eq. (6.77) and making use of γ0 ≪ ω0 we obtain after integration
The energy radiated into the unit solid angle dΩ = sinϑ dϑ dϕ is calculated as
∞ ∞ ∞
dW ε0 ε0
Z r Z r Z
2 2 2 2
= I(r, t) r dt = r |Eϑ (t)| dt = 4πr |Êϑ (ω)|2 dω ,
dΩ −∞ µ0 −∞ µ0 0
(6.80)
94 CHAPTER 6. RADIATION
where we applied Parseval’s theorem and used the definition of the intensity I =
p
ε0 /µ0 |Eϑ |2 of the emitted radiation. The total energy per unit solid angle dΩ and
per unit frequency interval dω can now be expressed as
The spectral shape of this function is determined by the expression in the brackets
known as the Lorentzian lineshape function. The function is shown in Fig. 6.9. The
width of the curve measured at half its maximum height is ∆ω = γ0 , and is called
“radiative linewidth.”
Integrating the lineshape function over the entire spectral range yields a value
of πγ0 /2. Integrating Eq. (6.81) over all frequencies and all directions leads to the
totally radiated energy
|p|2 ω04
W = . (6.82)
4πε0 3 c3 γ0
This value is equal to the average power P̄ radiated by a driven harmonic oscillator
divided by the linewidth γ0 (cf. Eq. 6.46).
0.5 γ0
ω
ω0
In situations where the field is not time-harmonic, we we simply replace the com-
plex field E(r) by the spectrum Ê(r, ω) and obtain the time-dependent field by
Fourier transformation (see Section 2.2).
95
96 CHAPTER 7. ANGULAR SPECTRUM REPRESENTATION
in a plane z = const. transverse to the chosen axis. In this plane we can evaluate
the two-dimensional Fourier transform of the complex field E(r) = E(x, y, z) as
ZZ∞
1
Ê(kx , ky ; z) = E(x, y, z) e−i [kx x + ky y] dx dy , (7.2)
4π 2
−∞
Notice that in the notation of Eqs. (7.2) and (7.3) the field E = (Ex , Ey , Ez ) and its
Fourier transform Ê = (Êx , Êy , Êz ) represent vectors. Thus, the Fourier integrals
hold separately for each vector component.
So far we have made no requirements about the field E, but we will assume that
in the transverse plane the medium is homogeneous, isotropic, linear and source-
free. Then, a time-harmonic, optical field with angular frequency ω has to satisfy
Einc
Escatt
z = const.
Figure 7.1: In the angular spectrum representation the fields are evaluated in
planes (z = const.) perpendicular to an arbitrarily chosen axis z.
97
(∇ 2 + k 2 )E(r) = 0 , (7.4)
√
where k is determined by k = (ω/c) n and n = µε is the index of refraction.
Inserting the Fourier representation of E(r) (Eq. 7.3) into the Helmholtz equation
and defining q
kz ≡ (k 2 − kx2 − ky2 ) with Im{kz } ≥ 0 , (7.5)
The ± sign specifies that we have two solutions that need to be superimposed:
the + sign refers to a wave propagating into the half-space z > 0 whereas the −
sign denotes a wave propagating into z < 0. Equation (7.6) tells us that the Fourier
spectrum of E in an arbitrary image plane located at z = const. can be calculated by
multiplying the spectrum in the object plane at z = 0 by the factor exp(±i kz z). This
factor is called the propagator in reciprocal space. In Eq. (7.5) we defined that
the square root leading to kz renders a result with positive imaginary part. This
ensures that the solutions remain finite for z → ±∞. Inserting the result of Eq. (7.6)
into Eq. (7.3) we finally find for arbitrary z
ZZ∞
E(x, y, z) = Ê(kx , ky ; 0) ei [kx x + ky y ± kz z] dkx dky (7.7)
−∞
For the case of a purely dielectric medium with no losses the index of refrac-
tion n is a real and positive quantity. The wavenumber kz is then either real or
imaginary and turns the factor exp(±i kz z) into an oscillatory or exponentially de-
caying function. For a certain (kx , ky ) pair we then find two different characteristic
solutions: plane waves with kx2 + ky2 ≤ k 2 and evanescent waves with kx2 + ky2 > k 2
(see Section 2.1.2).
also referred to as the optical transfer function (OTF) in free space. Remem-
ber that the longitudinal wavenumber is a function of the transverse wavenumber,
i.e. kz = [k 2 − (kx2 + ky2 )]1/2 , where k = n k0 = n ω/c = n 2π/λ. The ± sign indicates
that the field can propagate in positive and/or negative z direction. Equation (7.10)
can be interpreted in terms of linear response theory: Ê(kx , ky ; 0) is the input, Ĥ
is a filter function, and Ê(kx , ky ; z) is the output. The filter function describes the
propagation of an arbitrary spectrum through space. Ĥ can also be regarded as
the response function because it describes the field at z due to a point source at
↔
z = 0. In this sense, it is directly related to the Green’s function G0 .
7.1. PROPAGATION AND FOCUSING OF FIELDS 99
The filter Ĥ is an oscillating function for (kx2 + ky2 ) < k 2 and an exponentially
decreasing function for (kx2 + ky2 ) > k 2 . Thus, if the image plane is sufficiently
separated from the object plane, the contribution of the decaying parts (evanescent
waves) is zero and the integration can be reduced to the circular area (kx2 + ky2 ) ≤
k 2 . In other words, the image at z is a low pass filtered representation of the
original field at z = 0. The spatial frequencies (kx2 + ky2 ) > k 2 of the original field
are filtered out during propagation and the information on high spatial variations
gets lost. Hence, there is always a loss of information on propagating from near-
to far-field and only structures with lateral dimensions larger than
1 λ
∆x ≈ = (7.12)
k 2πn
can be imaged with sufficient accuracy. Here, n is the index of refraction. This
equation is qualitative and we will provide a more detailed discussion later. In
general, higher resolution can be obtained by a higher index of refraction of the
embodying system (substrate, lenses, etc.) or by shorter wavelengths. Theoreti-
cally, resolutions down to a few nanometers can be achieved by using far-ultraviolet
radiation or X-rays.
Let us now determine how the fields themselves evolve. For this purpose we
denote the transverse coordinates in the object plane at z = 0 as (x′ , y ′) and in the
image plane at z = const. as (x, y). The fields in the image plane are described
by the angular spectrum (7.7). We just have to express the Fourier spectrum
Ê(kx , ky ; 0) in terms of the fields in the object plane. Similarly to Eq. (7.2) this
Fourier spectrum can be represented as
ZZ∞
1 ′ ′
Ê(kx , ky ; 0) = E(x′ , y ′, 0) e−i [kx x + ky y ] dx′ dy ′ . (7.13)
4π 2
−∞
After inserting into Eq. (7.7) we find the following expression for the field E in the
image plane z = const.
ZZ∞ ZZ∞
1 ′ ′
E(x, y, z) = E(x , y ; 0) ei [kx (x−x ) + ky (y−y ) ± kz z] dx′ dy ′ dkx dky
′ ′
4π 2
−∞ −∞
This equation describes an invariant filter with the following impulse response
100 CHAPTER 7. ANGULAR SPECTRUM REPRESENTATION
ZZ∞
H(x, y; z) = ei [kx x + ky y ± kz z] dkx dky . (7.15)
−∞
In many optical problems the light fields propagate along a certain direction z and
spread out only slowly in the transverse direction. Examples are laser beam prop-
agation or optical waveguide applications. In these examples the wavevectors
k = (kx , ky , kz ) in the angular spectrum representation are almost parallel to the
z-axis and the transverse wavenumbers (kx , ky ) are small compared to k. We can
then expand the square root of Eq. (7.5) in a series as
q (kx2 + ky2 )
kz = k 1 − (kx2 + ky2 )/k 2 ≈ k − . (7.16)
2k
where E0 is a constant field vector in the transverse (x, y) plane. We have chosen
z = 0 at the beam waist. The parameter w0 denotes the beam waist radius. We
7.1. PROPAGATION AND FOCUSING OF FIELDS 101
4π 2
−∞
w02 −(kx2 +ky2 ) w02
= E0 e 4 , (7.18)
4π
which is again a Gaussian function. We now insert this spectrum into the an-
gular spectrum representation Eq. (7.7) and replace kz by its paraxial expression
in Eq. (7.16)
ZZ∞
w2 2 2 w02
+ 2i zk )
E(x, y, z) = E0 0 ei kz e−(kx +ky )( 4 ei [kx x + ky y] dkx dky , (7.19)
4π
−∞
This equation can be integrated and gives as a result the familiar paraxial repre-
sentation of a Gaussian beam
2 2)
E0 ei kz −
(x +y
2
1
(1 + 2 i z/kw2 )
E(x, y, z) = 2
e w0 0 . (7.20)
(1 + 2 i z/kw0 )
2
w0 − w2ρ (z) 2
E(ρ, z) = E0 e ei [kz−η(z)+kρ /2R(z)] (7.22)
w(z)
It can be shown that the surface defined by this equation is a hyperboloid whose
asymptotes enclose an angle
2
θ = (7.25)
k w0
with the z-axis. From this equation we can directly find the correspondence be-
tween the numerical aperture (NA = n sin θ) and the beam angle as NA ≈ 2n/kw0 .
Here we used the fact that in the paraxial approximation, θ is restricted to small
beam angles. Another property of the paraxial Gaussian beam is that close to
the focus, the beam stays roughly collimated over a distance 2z0 . z0 is called the
Rayleigh range and denotes the distance from the beam waist to where the beam
√
radius has increased by a factor of 2. It is important to notice that along the z-axis
(ρ = 0) the phases of the beam deviate from those of a plane wave. If at z → −∞
the beam was in phase with a reference plane wave, then at z → +∞ the beam will
be exactly out of phase with the reference wave. This phase shift is called Gouy
phase shift. The 180◦ phase change happens gradually as the beam propagates
through its focus. The phase variation is described by the factor η(z) in Eq. (7.23).
The tighter the focus the faster the phase variation will be.
|E|
ρ
1/e
ρ z
w(z)
θ ≈ 2/ kw0
2 z0
Figure 7.2: Illustration and main characteristics of a paraxial Gaussian beam. The
beam has a Gaussian field distribution in the transverse plane. The surfaces of
constant field strength form a hyperboloid along the z-axis.
7.2. FAR-FIELD APPROXIMATION 103
imation is introduced, the field E fulfills no longer Maxwell’s equations. The error
becomes larger the smaller the beam waist radius w0 is. Another important aspect
of Gaussian beams is that they do not exist, no matter how rigorous the theory that
describes them! The reason is that a Gaussian beam profile demands a Gaussian
spectrum. However, the Gaussian spectrum is infinite and contains evanescent
components that are not available in a realistic situation. Thus, a Gaussian beam
must always be regarded as an approximation. The tighter the focus, the broader
the Gaussian spectrum and the more contradictory the Gaussian beam profile will
be. The angular spectrum representation can be used to derive a rigorous descrip-
tion of focussed fields (e.g. Novotny, Principles of Nano-Optics).
In this section we will derive the important result that Fourier Optics and Geomet-
rical Optics naturally emerge from the angular spectrum representation.
We are interested in the asymptotic far-zone approximation of this field, i.e. in the
evaluation of the field in a point r = r∞ at infinite distance from the object plane.
The dimensionless unit vector s in the direction of r∞ is given by
x y z
s = (sx , sy , sz ) = , , , (7.27)
r r r
where r = (x2 + y 2 + z 2 )1/2 is the distance of r∞ from the origin. To calculate the
far-field E∞ we require that r → ∞ and rewrite Eq. (7.26) as
Z Z
kx ky kz
E∞ (sx , sy ) = lim Ê(kx , ky ; 0) ei kr [ k sx + k sy ± k sz ] dkx dky , (7.28)
kr→∞
(kx2 +ky2 )≤k 2
104 CHAPTER 7. ANGULAR SPECTRUM REPRESENTATION
q
where sz = 1 − (s2x + s2y ). Because of their exponential decay, evanescent waves
do not contribute to the fields at infinity. We therefore rejected their contribution and
reduced the integration range to (kx2 + ky2 ) ≤ k 2 . The asymptotic behavior of the
double integral as kr → ∞ can be evaluated by the method of stationary phase.
A clear outline of this method can be found in other textbooks (e.g. Born & Wolf,
Principles of Optics). Without going into details, the result of Eq. (7.28) can be
expressed as
ei kr
E∞ (sx , sy ) = −2πik sz Ê(ksx , ksy ; 0) . (7.29)
r
This equation tells us that the far-fields are entirely defined by the Fourier spectrum
of the fields Ê(kx , ky ; 0) in the object plane if we replace kx → ksx and ky → ksy .
This simply means that the unit vector s fulfills
kx ky kz
s = (sx , sy , sz ) = , , , (7.30)
k k k
which implies that only one plane wave with the wavevector k = (kx , ky , kz ) of
the angular spectrum at z = 0 contributes to the far-field at a point located in the
direction of the unit vector s (see Fig. 7.3). The effect of all other plane waves is
cancelled by destructive interference. This beautiful result allows us to treat the
field in the far-zone as a collection of rays with each ray being characterized by a
particular plane wave of the original angular spectrum representation (Geometrical
optics). Combining Eqs. (7.29) and (7.30) we can express the Fourier spectrum Ê
in terms of the far-field as
ir e−i kr kx ky
Ê(kx , ky ; 0) = E∞ ( , ) , (7.31)
2πkz k k
keeping in mind that the vector s is entirely defined by kx , ky . This expression can
be substituted into the angular spectrum representation (Eq. 7.26) as
ir e−i kr kx ky i [kx x + ky y ± kz z] 1
ZZ
E(x, y, z) = E∞ ( , ) e dkx dky (7.32)
2π k k kz
(kx2 +ky2 )≤k 2
Thus, as long as evanescent fields are not part of our system then the field E and
its far-field E∞ form essentially a Fourier transform pair at z = 0. The only devia-
tion is given by the factor 1/kz . In the approximation kz ≈ k, the two fields form a
perfect Fourier transform pair. This is the limit of Fourier Optics.
7.2. FAR-FIELD APPROXIMATION 105
Lx Ly sin(kx Lx ) sin(ky Ly )
= E0 , (7.33)
π2 kx Lx ky Ly
With Eq. (7.29) we now determine the far-field as
2Lx Ly sin(ksx Lx ) sin(ksy Ly ) ei kr
E∞ (sx , sy ) = −iksz E0 , (7.34)
π ksx Lx ksy Ly r
s
z
z=0
Diffraction refers to the observation that light rays break away from their geometri-
cal paths, which is to say, that the wave nature of radiation becomes relevant. In
this section we will discuss two important regimes of diffraction theory, Fresnel and
Fraunhofer diffraction.
The far-field approximation derived in the previous chapter has its limitations. It
has been assumed that the observation point is at infinite distance from the source
plane. However, how far do we have to go to be approximately at infinity? The best
x’ x z
R
y’ y
z=0 z = const.
k1 k2
n1 n2
x
n1 n2
θ1 θ2
p z
f1 f2
f1 sin θ1 = f2 sin θ2
we have a radiating point source then the image of that source will appear to have
a finite size. This broadening is a direct consequence of spatial filtering. A point in
space is characterized by a delta function that has an infinite spectrum of spatial
frequencies kx , ky . On propagation from the source to the image, high-frequency
components are filtered out. Usually the entire spectrum (kx2 +ky2 ) > k 2 associated
with the evanescent waves is lost. Furthermore, not all plane wave components
can be collected, which leads to a further reduction in bandwidth. The reduced
spectrum is not able to accurately reconstruct the original point source and the
image of the point will have a finite size.
Taken all together, the focal field represented by Eq. (7.32) can be written as
Max[θ
Z 2Z] 2π
ik2 f2 e−i k2 f2
E(ρ, ϕ, z) = E∞ (θ2 , φ) eik2 z cos θ2 eik2 ρ sin θ2 cos(φ−ϕ) sin θ2 dφ dθ2
2π
0 0
(7.42)
110 CHAPTER 7. ANGULAR SPECTRUM REPRESENTATION
Here, we have replaced the distance r∞ between the focal point and the surface
of the reference sphere of the lens by the focal length f2 . We have also limited
the integration over θ2 to the finite range [0 .. Max[θ2 ]] because any lens will have
a finite size. Furthermore, since all fields propagate in the positive z-direction we
retained only the + sign in the exponent of Eq. (7.32).
To evaluate Eq. (7.42) we need to insert the field E∞ of the dipole after it has
been refracted by the lens. To simplify the analysis we will ignore the vectorial
nature of dipole and its fields. Furthermore, we will assume that the imaging sys-
tem can be treated in the paraxial approximation, or small angle limit, for which
sin θ1 ≈ θ1 and sin θ2 ≈ θ2 . Using the far-field term of Eq. (6.42) and ignoring the
angular dependence (scalar point source), the dipole field before refraction at the
lens is
p k12 exp(ik1 f1 )
E1 = − . (7.43)
4πε0 ε1 f1
We trace this field through the lens and then insert it as E∞ into Eq. (7.42) above.
In essence, the dipole field is a field that uniformly illuminates the focusing lens,
that is E∞ (θ2 , φ) ≈ const.. Using the mathematical relation
Z2π
eix cos(φ−ϕ) dφ = 2π J0 (x) , (7.44)
0
we can carry out the integration over φ analytically. Here, J0 is the 0th-order Bessel
θ2
kz
Figure 7.6: Illustration of the substitution (1/kz ) dkx dky = k sin θ dθ dφ. The factor
1/kz = 1/(k cos θ) ensures that the differential areas on the plane and the sphere
stay equal.
7.4. THE POINT-SPREAD FUNCTION 111
function. The final expression for the focal field now contains a single integration
over the variable θ2 . Skipping the constant prefactors we obtain
Max[θ
Z 2]
E(ρ, ϕ, z) ∝ eik2 z cos θ2 J0 (k2 ρ sin θ2 ) sin θ2 dφ dθ2 . (7.45)
0
we find the following result for the intensity in the image plane
2
π 4 p2 NA4
E(ρ, z = 0)2 =
J1 (2π ρ̃) NA ρ
lim 2 , ρ̃ = (7.47)
θmax ≪π/2 ε20 n1 n2 λ6 M 2 (2π ρ̃) Mλ
Here, we have used a normalized coordinate ρ̃, which is expressed in terms of the
magnification M = (n1 /n2 )(f2 /f1 ) and the numerical aperture
NA = n1 sin(Max[θ1 ]) . (7.48)
Note that sin(Max[θ1 ]) = (f2 /f1 ) sin(Max[θ2 ]). The result (7.47) is the point-spread-
function, first derived by Abbe in 1873. Its functional form is given by the term in
Δx M Δx
Figure 7.7: Illustration of the resolution limit. Two simultaneously radiating point
sources separated by ∆r|| in the object plane generate a combined point-spread
function in the image plane. The two point sources are resolved if they can be
distinguished based on their image pattern.
112 CHAPTER 7. ANGULAR SPECTRUM REPRESENTATION
brackets which is known as the Airy function. It tells us that the image of a point
is no longer a point, but a function with a finite width. This width determines the
resolution of an imaging system. In essence, two points in the object plane have
to be separated by more than the width of the PSF in order to be distinguishable.
This is illustrated in Fig. 7.7.
The width of the point-spread function ∆x is usually defined as the radial dis-
30
25
photon count rate (kHz)
20
15
10
0
-1.0 -0.5 0 0.5 1.0
distance (μm)
Figure 7.8: Point-spread function measured with a single molecule point source.
Fluorescence photons emitted by a DiI molecule are collected with a NA = 1.3
objective lens. The center wavelength is λ ≈ 580 nm. The data points correspond
to a horizontal line cut through the center of the fluorescence rate image shown in
the inset. The solid curve corresponds to the Airy function.
7.4. THE POINT-SPREAD FUNCTION 113
tance for which the value of the paraxial point-spread function becomes zero, or
Mλ
∆x = 0.6098 . (7.49)
NA
This width is also denoted as the Airy disk radius. It depends in a simple manner
on the numerical aperture, the wavelength and the magnification of the system.
114 CHAPTER 7. ANGULAR SPECTRUM REPRESENTATION
Chapter 8
8.1 Resonators
Let us consider a rectangular box with sides Lx , Ly , and Lz , as shown in Fig. 8.1.
The interior of the box is filled with a linear and isotropic material characterized
by ε and µ. The walls of the box are perfectly reflecting, that is, the fields are not
able to penetrate into the walls. Furthermore, there are no sources present, which
implies that we’re looking for homogeneous solutions of the wave equation, that is,
solutions of the Helmholtz equation (3.15).
To solve the Helmholtz equation for the x component of the complex electric
field vector E we write
(x)
Ex (x, y, z) = E0 X(x) Y (y) Z(z) , (8.1)
115
116 CHAPTER 8. WAVEGUIDES AND RESONATORS
In order for this equation to hold for any x, y, and z we have to require that each
of the terms is constant. We set these constants to be −kx2 , −ky2 and −kz2 , which
implies
ω2
kx2 + ky2 + kz2 = k 2 = 2 n2 (ω) . (8.3)
c
We obtain three separate equations for X, Y , and Z
∂ 2 X / ∂x2 + kx2 X = 0
∂ 2 Y / ∂y 2 + ky2 Y = 0
∂ 2 Z / ∂z 2 + kz2 Z = 0 , (8.4)
with the solutions exp[±ikx x], exp[±iky y], and exp[±ikz z]. Thus, the solutions for
Ex become
(x)
Ex (x, y, z) = E0 [c1,x e−ikx x + c2,x eikx x ] [c3,x e−iky y + c4,x eiky y ] [c5,x e−ikz z + c6,x eikz z ] .
(8.5)
Electric fields cannot penetrate into a perfect conductor. Therefore, the boundary
condition (4.13) implies Ex (x = 0) = Ex (x = Lx ) = 0, which turns Eq. (8.5) into
(x)
Ex (x, y, z) = E0 sin[n π x/Lx ] [c3,x e−iky y + c4,x eiky y ] [c5,x e−ikz z + c6,x eikz z ] , (8.6)
Lx
Lz z
Ly
Figure 8.1: A resonator with perfectly reflecting walls and side lengths Lx , Ly , and
Lz .
8.1. RESONATORS 117
with n being an integer. Similar solutions are found for Ey and Ez , namely,
(y)
Ey (x, y, z) = E0 [c1,y e−ikx x + c2,y eikx x ] sin[m π y/Ly ] [c5,y e−ikz z + c6,y eikz z ] (8.7)
(z)
Ez (x, y, z) = E0 [c1,z e−ikx x + c2,z eikx x ] [c3,z e−iky y + c4,z eiky y ] sin[l π z/Lz ]. (8.8)
(x)
Ex (x, y, z) = E0 sin[n π x/Lx ] sin[m π y/Ly ] sin[l π z/Lz ]
(y)
Ey (x, y, z) = E0 sin[n π x/Lx ] sin[m π y/Ly ] sin[l π z/Lz ]
(z)
Ez (x, y, z) = E0 sin[n π x/Lx ] sin[m π y/Ly ] sin[l π z/Lz ] , (8.9)
and
n (x) m (y) l (z)
E0 + E0 + E = 0 (8.10)
Lx Ly Lz 0
Using Eq. (8.3) we find the dispersion relation or mode structure of the resonator
n2 m2 l2 2
2 ωnml
π + + = n2 (ωnml ) n, m, l ∈ {0, 1, 2, 3, ..} (8.11)
L2x L2y L2z c 2
Thus, we find that the resonator supports only discrete frequencies ωnml , each as-
sociated with a mode (n, m, l) of the resonator. Note that n is used both for the
index of refraction and for the mode number in x.
Let us now consider a resonator with equal sides, that is, L = Lx + Ly + Lz . In this
case,
2
2 2 2 ωnml L n(ωnml )
n +m +l = . (8.12)
πc
If n, m, and l were real numbers, then this equation defines a sphere of radius
r0 = [omeganml L n(ωnml )/(π c)]. Indeed, for large numbers we can approximate
these numbers by real numbers. Let us now consider a specific mode given by
[n, m, l] and with angular frequency ω = ωnml and count the number of modes N
with frequencies smaller than ω. According to Eq. (8.12) this corresponds to the
118 CHAPTER 8. WAVEGUIDES AND RESONATORS
interior of a sphere with radius r0 , and because n, m, and l are positive numbers
we have to count only 1/8-th of the sphere. Thus,
1 4π 3
N(ω) = r 2. (8.13)
8 3 0
The ‘2’ at the end has been added because for each [n, m, l] there are two solutions
with different polarizations. This follows from Eq. (8.10). Inserting the expression
for r0 we find
ω 3 n3 (ω)
N(ω) = V , (8.14)
3π 2 c3
where V is the volume of the resonator. The number of different resonator modes
in the frequency interval [ω .. ω + ∆ω] becomes
d N(ω) ω 2 n3 (ω)
∆ω = V ∆ω , (8.15)
dω π 2 c3
which states that there a many more modes for high frequencies ω. We now define
the number of modes ρ(ω) per unit volume V and unit frequency ∆ω. We obtain
ω 2 n3 (ω)
ρ(ω) = (8.16)
π 2 c3
which is generally referred to as the density of states (DOS). The number of modes
N in the volume V and in the frequency range [ω1 .. ω2 ] is then calculated as
Z Z ω2
N(ω) = ρ(ω) dω dV . (8.17)
V ω1
where ω0 represents one of the resonance frequencies ωmnl . The spectrum of the
stored energy density becomes
ω02 Wω (ω0 )
Wω (ω) = . (8.20)
4Q (ω − ω0 )2 + (ω0 / 2Q)2
2
with E0 (r, ω0 ) and H0 (r, ω0 ) denoting the unperturbed complex field amplitudes. A
particle with material parameters ∆ε(r), ∆µ(r) constitutes a perturbation and gives
120 CHAPTER 8. WAVEGUIDES AND RESONATORS
rise to a new resonance frequency ω. Maxwell’s curl equations for the perturbed
system read as
Notice that both ∆ε and ∆µ are zero outside of the volume occupied by the per-
turbation. Using ∇ · (A × B) = (∇ × A) · B − (∇ × B) · A we find
We now consider a fictitious spherical surface ∂V at very large distance from the
cavity and integrate Eq. (8.24) over the enclosed volume V (c.f. Fig. 8.2). Using
Gauss’ theorem, the left hand side of Eq. (8.24) becomes
Z
[H · (n × E∗0) + H∗0 · (n × E)] da = 0 (8.25)
∂V
where n is a unit vector normal to the surface ∂V . The above expression vanishes
because of the transversality of the field, i.e. (n × E∗0 ) = (n × E) = 0 on the surface
enclosing surface
∂V
resonator
perturbation
ΔV
We assume that the perturbation has a small effect on the cavity. Therefore we
write as a first-order approximation E = E0 and H = H0 . After performing these
substitutions in Eq. (8.26) we find
For a high-Q resonator the radiation losses are small and the integration volume
V can be taken over the boundaries of the resonator. To evaluate Eq. (8.27) we
first must solve for the fields E0 (r), H0(r) of the unperturbed cavity. Interestingly,
for a weakly-dispersive medium the denominator of Eq. (8.27) denotes the total
energy of the unperturbed cavity (W0 ) whereas the nominator accounts for the
energy introduced by the perturbation (∆W ). Hence, (ω − ω0 )/ω = −∆W/W0 .
An increase of energy by ∆W causes the resonance frequency to red-shift to
ω = ω0 [W0 /(W0 + ∆W )]. A blue-shift is possible by perturbing the cavity volume,
i.e. by removing ∆W from the cavity.
As an example let us consider a planar cavity with perfectly reflecting end faces
of area A and separated by a distance L. The fundamental mode λ = 2L has a
resonance frequency ω0 = πc/L, and the electric and magnetic fields inside the
p
cavity are calculated to be E0 sin[πz/L] and −i ε0 /µ0 E0 cos[πz/L], respectively.
The coordinate z is perpendicular to the surfaces of the end faces. The denomi-
nator of Eq. (8.27) is easily determined to be V ε0 E02 , where V = LA. We place a
spherical nanoparticle with dielectric constant ∆ε and volume ∆V in the center of
122 CHAPTER 8. WAVEGUIDES AND RESONATORS
the cavity and assume that the field is homogeneous across the dimensions of the
particle. The nominator of Eq. (8.27) is calculated to be ∆V ∆ε ε0 E02 and the fre-
quency shift is determined to be (ω − ω0 )/ω = −∆ε ∆V /V . A better approximation
retains the perturbed fields E and H in the nominator of Eq. (8.26). Making use of
the quasi-static solution for a small spherical particle we write E = 3E0 /(2 + ∆ε)
and obtain a frequency shift of (ω − ω0 )/ω = −[3 ∆ε/(2 + ∆ε)] ∆V /V . In both cases
the resonance shift scales with the ratio of resonator and perturbation volumes.
8.2 Waveguides
direction z (TE modes), and 2) waves that have no magnetic field in propagation
direction z (TM modes).
TE Modes
For TE modes the electric field is parallel to the surface of the plates. As an ansatz
we choose a plane wave propagating at an angle θ towards the bottom plate
with k = n(ω) ω/c. The wave reflects at the bottom plate and then propagates
towards the top plate
where the sign change is due to the boundary condition, that is, the total parallel
E-field has to vanish. The sum of the fields becomes
k z
p
k zn = k 2 − n2 [π/d]2 n ∈ {1, 2, .. } (8.33)
As long as kzn is real the field will propagate along the z direction. However, when
nπ/d > k the propagation constant kzn is imaginary and the field will exponential
decay in direction of z, similar to evanescent waves discussed in Section 4.4. Be-
cause k ∝ ω it turns out that the waveguide acts as a high-pass filter. Frequencies
below the cut-off frequency
nπc
ωc = n ∈ {1, 2, .. } (8.34)
d n(ωc )
cannot propagate. In summary, the solutions of the wave equation for a system of
two parallel plates are characterized by a mode number n. We refer to the solu-
tions that have no electric field in propagation direction as T En modes. Solutions
are found only for discrete frequencies ωn and for each mode there is a cut-off fre-
quency below which no propagation is possible. Note, that n(ωc ) is the dispersive
index of refraction of the medium between the two plates.
For propagating fields with ω > ωc , the phase velocity is defined by the phase
factor exp[ikzn z − iωt] as vph = ω/kzn . On the other hand, the energy of the field
and thus any information associated with it is transported at the group velocity
vg = dω/dkzn .
TM Modes
Let us now repeat the analysis for the case where the magnetic field is parallel to
the surface of the plates. Similar to Eq. (8.28) we write
The wave reflects at the bottom plate and then propagates towards the top plate
In contrast to Eq. (8.36) there is no sign change of H0 upon reflection at the bound-
ary. This follows from the boundary conditions (4.11) - (4.14).1 The sum of the
fields becomes
and the boundary conditions at the top interface z = d lead to the condition
In contrast to the quantization condition (8.31) for TE modes, we now also find
solutions for n = 0 (cos(0) 6= 0). Thus,
p
k zn = k 2 − n2 [π/d]2 n ∈ {0, 1, .. } (8.39)
The fundamental mode T M0 has no cut-off frequency whereas all higher order T M
do have a cut-off, similar to T E modes discussed above. The absence of a cut-off
1
The electric field associated with H1 is E1 ∼ E0 [sin θ, 0, cos θ]T , which upon reflection becomes
E2 ∼ E0 [sin θ, 0, − cos θ]T because the parallel E-field at the boundary has to vanish. Consequently,
H0 remains unchanged upon reflection.
1.0
TM0
0.8 TE1
TM1 TE2
0.6 TM2 TE3
kz / k
TM3 TE4
0.4 TM4 TE5
TM5
0.2
0
0 0.5 1.0 1.5 2.0 2.5 3.0
d/λ
for the T M mode is a finding that is not restricted to two parallel plates, but holds
for any waveguide made of two electrically isolated metal electrodes. For example,
a coaxial cable has no cut-off, but a hollow metal waveguide (a pipe) does have a
cut-off.
We now confine the lateral extent of the waveguide modes. The parallel-plate
waveguide then turns into a hollow metal waveguide as illustrated in Fig. 8.5. To
solve for the fields in such a waveguide structure we write
where ∇2t = ∂ 2 /∂x2 +∂ 2 /∂y 2 is the transverse Laplacian and and kt = [kx2 +ky2 ]1/2 =
[k 2 − kz2 ]1/2 is the transverse wavenumber.
Next, we write the electric field in terms of a transverse vector in the (x,y) plane
and a longitudinal vector that points along the z axis, that is,
E = Et + Ez . (8.42)
Lx
y
Ly
where Z is the wave impedance defined in Eq. (4.33). These equations show that
the transverse field components Exxy , Eyxy and Hxxy , Hyxy derive from the longitudinal
field components Ezxy and Hzxy . Thus, it is sufficient to solve for Ezxy and Hzxy . So
far, the discussion was general, not restricted to any particular waveguide geome-
try. We next discuss the particular case of a rectangular hollow metal waveguide,
as illustrated in Fig. 8.5. Such waveguides are commonly used in the microwave
regime, which spans the frequency range of 1-100 GHz.
TE Modes
n2 π 2 m2 π 2
kt2 = [kx2 + ky2 ] = + n, m ∈ {0, 1, 2 .. } , (8.48)
L2x L2y
128 CHAPTER 8. WAVEGUIDES AND RESONATORS
similar to the mode structure of a resonator (Eq. 8.11). Accordingly, the frequen-
cies of the T Enm modes are
s
πc n2 m2
ωnm = + n, m ∈ {0, 1, 2, ..} (8.49)
n(ωnm ) L2x L2y
with n(ωnm ) being the index of refraction. It turns out that n and m cannot both
be zero because the T E00 does not exist. Thus, the lowest frequency mode is the
T E01 or the T E10 mode. While the modes of a parallel-plate waveguide are formed
by the superposition of two plane waves, the fields of a hollow rectangular waveg-
uide follow from the superposition of four plane waves. Note that the condition
of a magnetic field parallel to the waveguide surfaces leads here to T E modes,
whereas in the case of two parallel plates it leads to T M modes.
Let us choose Lx > Ly . In this case, the fundamental mode is the T E01 mode
for which, according to Eqs. (8.43) - (8.46), the fields are determined as
π
Hzxy = H0z cos[ y] (8.51)
Ly
π
Hyxy = i(kz /kt2 ) H0z (π/Ly ) sin[ y] (8.52)
Ly
π
Exxy = −(ikZ/kt2 ) H0z (π/Ly ) sin[ y] (8.53)
Ly
Hxxy xy xy
= Ey = Ez = 0 .
According to Eq. (8.40), the total fields E and H are obtained by multiplying with
exp[−ikz z]. The fields of the T E01 mode are illustrated in Fig. 8.6.
8.2. WAVEGUIDES 129
H
x = Lx / 2
y = Ly / 2
Figure 8.6: Fields of the fundamental T E01 mode. Adapted from Heald & Marion,
Classical Electromagnetic Radiation, Saunders College Publishing, 3rd ed. 1995.
TM Modes
refractive indices n1 and n2 it must be incident from the optically denser medium
(n1 > n2 ) and propagate at an angle larger than the critical angle θc = arctan[n2 /n1 ]
measured from the surface normal. In contrast to metal waveguides, TIR in dielec-
tric waveguides leads to evanescent fields that stretch out into the surrounding
medium with the lower index of refraction n2 (see Fig. 8.7). Thus, the boundary
conditions at the interfaces become more complex.
Optical fibers are axially symmetric, that is, they consist of a dielectric rod of in-
dex n1 (the core) surrounded by a medium of index n2 (the cladding). For claddings
of sufficient thickness the evanescent waves are strongly attenuated at the outer
boundaries and it is reasonable to approximate the cladding radius as being infi-
nite. The fields of an optical waveguide with circular cross section are described
by cylindrical Bessel functions Jn of order n. To keep things simple, we will not
analyze optical fibers with circular cross-sections. Instead, we focus on the di-
electric slab waveguide that consists of a dielectric layer with index n1 sandwiched
between two infinite dielectrics of index n2 < n1 . The mode structure is different
but the physical principles are the same.
The waveguide fields have to satisfy the Helmholtz equation (3.15). Similar to
Section 8.2.1 we will separate the fields into transverse electric (T E) and trans-
verse magnetic (T M) solutions. In the former case the electric field has no vector
component along the waveguide axis z, whereas in the latter case the same holds
for the magnetic field. As illustrated in Fig. 8.7, a waveguide mode can be de-
n2 E4 a
d n1 E3 E2 b
n2 z
E1 c
Figure 8.7: Optical step-index waveguide made of a dielectric with refractive in-
dex n1 surrounded by a dielectric with lower index of refraction n2 . Waveguiding
originates from a sequence of total internal reflections.
8.2. WAVEGUIDES 131
scribed by the superposition of four waves, namely two plane waves inside the
waveguide (0 < x < d), an evanescent wave in the medium above the waveguide
(x > d) and an evanescent wave below the waveguide (x < 0). The evanescent
waves above and below the waveguide ensure that no energy is radiated away
from the waveguide. Thus, the total field E is written in terms of partial fields
E1 .. E4 as
E1 (r) x<0
E(r) = E2 (r) + E2 (r) 0 < x < d (8.55)
E4 (r) x>d
TM Modes
Having defined the partial fields we know have to match them at the boundaries
x = 0 and x = d. The continuity of the parallel components of electric and magnetic
fields lead to
kx2 /k2 −kx1 /k1 kx1 /k1 0 E1
1/Z −1/Z1 −1/Z1 0 E
2 2
= 0
0 kx1 /k1 exp[−ikx1 d] −kx1 /k1 exp[ikx1 d] kx2 /k2 exp[ikx2 d] E3
0 1/Z1 exp[−ikx1 d] 1/Z1 exp[ikx1 d] −1/Z2 exp[ikx2 d] E4
(8.60)
This is a homogeneous system of linear equations, that is, an eigenvalue problem.
The reason why we end up with a homogeneous system of equations is the ab-
sence of an excitation, which means that there are no sources or incident fields.
Thus, we are looking for solutions that the system supports in absence of an ex-
ternal driving force, similar to an undriven harmonic oscillator. A homogeneous
system of equations has solutions only if the determinant of the matrix acting on
the eigenvector [E1 , E2 , E3 , E4 ]T vanishes, that is,
kx2 /k2 −kx1 /k1 kx1 /k1 0
1/Z −1/Z1 −1/Z1 0
2
= 0 (8.61)
0 kx1 /k1 exp[−ikx1 d] −kx1 /k1 exp[ikx1 d] kx2 /k2 exp[ikx2 d]
0 1/Z1 exp[−ikx1 d] 1/Z1 exp[ikx1 d] −1/Z2 exp[ikx2 d]
Writing out the expression for the determinant and arranging terms we obtain
(Z2 kx2 /k2 − Z1 kx1 /k1 ) (Z1 kx1 /k1 − Z2 kx2 /k2 )
1 + exp[2ikx1 d] = 0 , (8.62)
(Z2 kx2 /k2 + Z1 kx1 /k1) (Z1 kx1 /k1 + Z2 kx2 /k2 )
p p
1 + rab (kz ) rbc (kz ) e2ikx1 d = 0 (8.63)
p p
Here, rab and rbc are the Fresnel reflection coefficients for p polarization, as de-
fined in Eq. (4.39). The subscript ‘ab’ indicates that the reflection is measured
between the upper medium (medium a in Fig. 8.7) and the waveguide (medium
b in Fig. 8.7). Similarly, the subscript ‘bc’ is the reflection coefficient measured
8.2. WAVEGUIDES 133
between the waveguide and the lower medium (medium 3 in Fig. 8.7). Note that
r r
q
ω 2 q
ω2
kx1 = k12 − kz2 = µ ε
1 1 − k 2,
z k x2 = k 2
2 − k 2 =
z µ2 ε2 − kz2 , (8.64)
c2 c2
and hence Eq. (8.63) defines the characteristic equation for the eigenvalues kz .
Every solution for kz defines a waveguide mode. It has to be emphasized that the
sign of the square roots in the expressions for kx1 and kx2 has to be chosen such
that the imaginary part is positive. This ensures that the fields decay exponentially
with distance from the waveguide (evanescent fields). The other sign would imply
an exponential increase, which would violate energy conservation.
TE Modes
A similar analysis can be applied for T E polarized fields, for which the electric field
is parallel to the boundaries. The resulting characteristic equation for kz is
s s
1 + rab (kz ) rbc (kz ) e2ikx1 d = 0 (8.65)
1.5
1.4
1.3
kz / k 2
TE1
TE2
1.2 TE3
TE4
TM1
1.1 TE5
TM2
TM3
TM4
0
0 0.5 1.0 1.5 2.0
d/λ
Figure 8.8: Mode structure of a dielectric slab waveguide with n1 = 1.5 (glass) and
n2 = 1 (air). The vertical axis shows the real part of the normalized propagation
constant kz . There is no cut-off for the fundamental modes T M1 and T E1 .
134 CHAPTER 8. WAVEGUIDES AND RESONATORS
p p
with rab and rbc being the Fresnel reflection coefficients for s polarization.
As shown in Fig. 8.8, the mode structure for the dielectric waveguide is similar
to the mode structure of a parallel-plate waveguide in Fig. 8.4. However, due to
the different Fresnel reflection coefficients for T E and T M modes, the curves for
T M and T E modes are no longer the same and split into separate curves. The
fundamental modes T E1 and T M1 have no cut-off and in the limit d/λ → 0 they
become plane waves (kz = k2 ). Similarly, in the limit d/λ → ∞ the modes become
plane waves propagating in the higher dielectric (kz = k1 ).
In many regards the mode structure in optical fibers is similar to the mode structure
of a dielectric slab waveguide. There are, however, also some relevant differences.
First, an optical fiber is a waveguide that confines the electromagnetic field in two
dimensions which, according to Section 8.2.2, leads to two mode indices nm. One
of them specifies the angular dependence of the fields [sin mφ, cos mφ] and the
(1)
other one the radial dependence [Jn (ρ), Hn (ρ)]. Second, besides pure T E and
5
attenuation (dB / km)
0
0.7 0.9 1.1 1.3 1.5 1.7
wavelength (μm)
Figure 8.9: Attenuation in an optical fiber. Lowest propagation losses are obtained
at wavelengths of ∼ 1.3 µm and ∼ 1.5 µm.
8.2. WAVEGUIDES 135
T M modes there are also hybrid modes, which are classified as HE and EH
modes. HE modes have T E flavor whereas the EH modes have more T M char-
acter.
There are two basic versions of optical fibers: gradient index fibers and step-
index fibers. The former has an index of refraction profile that varies gradually
with radius [n = n(ρ)], whereas the latter exhibits an abrupt transitions between
two refractive indices, similar to the slab waveguide. Step-index fibers are the
most commonly used fibers and usually the index difference is very small, that is
(n1 − n2 )/n1 ≪ 1, which is referred to as the weakly guiding condition. Polarized
weakly guided modes are denoted as LPnm modes. A single-mode fiber supports
only the fundamental LP01 mode and all higher-order modes are suppressed by
the cut-off condition. Fig. 8.9 shows the propagation loss in a modern optical fiber.
The losses on the short-wavelength side are due to Rayleigh scattering and the
losses at the long wavelength end are due to infrared absorption. The bumps are
absorption bands of molecular bonds, such as OH − . The lowest attenuation is
obtained near 1.3 and 1.5 µm and is the reason why fiber-optic communication is
operated at these wavelengths.
136 CHAPTER 8. WAVEGUIDES AND RESONATORS
Chapter 9
137
138 CHAPTER 9. CLASSICAL ELECTRON THEORY
Chapter 10
Coherence Theory
139
140 CHAPTER 10. COHERENCE THEORY
Chapter 11
Relativistic Electromagnetics
141
142 CHAPTER 11. RELATIVISTIC ELECTROMAGNETICS
Chapter 12
Antennas
143