UNIVERSITY OF EASTERN FINLAND

DEPARTMENT OF PHYSICS AND MATMEMATICS

MICRO- AND NANOPHOTONICS
Jari Turunen
0 100 200 300 400 500
0
5
10
15
20
25
30
A TALBOT CARPET
Lecture notes
Third Edition (2013)
JOENSUU 2013
Chapter I
Harmonic elds
The basis of all micro-optical design is the ability to govern the propagation of electro-
magnetic waves in free space or in homogeneous media. These waves may originate from
a large number of sources with widely dierent properties: the sources will be discussed
in detail in Chapter 5. The radiation may be essentially monochromatic and spatially
coherent, i.e., capable of producing high-contrast interference fringes in devices such as
Youngs and Michelsons interferometer. The study of such idealized harmonic elds is
the main topic of this Chapter. Real light elds are never harmonic; light can propagate
in the form of short pulses with an appreciable spectral width, or consist of mutually
independent modes. All such elds can, however, be represented as superpositions of
harmonic elds and these superpositions will be discussed in Chapter 2.
A fully electromagnetic approach is taken throughout the text to reect the fact that
light elds are vectorial, even though their description by a single scalar component is
sometimes adequate. We will return on several occasions to conditions under which such
a simplied approach is possible.
1.1 Preliminaries
Let us begin by dening some concepts, notations and conventions on electromagnetic
elds to be used throughout the text.
1.1.1 Complex eld representation
Although electromagnetic elds are real-valued functions of position r and time t, in
linear optics it is mathematically more convenient to use complex functions for their
representation (however, real eld representations must always be used when dealing
with nonlinear functions of the eld). We thus associate with any real eld f
(r)
(r, t) the
complex-analytic signal f (r, t) such that
f
(r)
(r, t) = f (r, t) =
1
2
[f (r, t) +f

(r, t)] = [f (r, t)[ cos arg [f (r, t)] , (1.1)

where denotes the real part, the asterisk denotes complex conjugate, and arg represents
the argument of a complex number.
1
2 1. Harmonic elds
Let us represent the real (square-integrable) eld f
(r)
(r, t) in the form of a temporal
Fourier integral
f
(r)
(r, t) =
_

f
(r)
(r, ) exp (it) d. (1.2)
Since f
(r)
(r, t) is real, the Fourier components
f
(r)
(r, ) =
1
2
_

f
(r)
(r, t) exp (it) dt (1.3)
satisfy the condition
f
(r)
(r, ) =
_
f
(r)
(r, )

(1.4)
indicating that the negative frequency components contain no information that is not
already contained in the positive ones. Let us now dene
f(r, ) = [f(r, )[ exp i arg [f(r, )] =
_
2f
(r)
(r, ) if 0
0 if < 0
(1.5)
and write
f (r, t) =
_

0
f(r, ) exp(it) d. (1.6)
Evidently, Eqs. (1.1) and (1.2) are satised and hence the quantity in Eq. (1.6) represents
the complex analytic signal. Furthermore, we have explicitly
f
(r)
(r, t) =
_

0
[f(r, )[ cos arg [f(r, )] t d. (1.7)
This is a spectral representation of the real eld in terms of its positive frequency com-
ponents 0.
A strictly monochromatic eld oscillates at a single frequency, say, =
0
. Thus the
complex analytic signal takes the form
f (r, t) = f(r,
0
) exp (i
0
t) (1.8)
with
f(r, ) = f(r,
0
)(
0
). (1.9)
The real eld may be written as
f
(r)
(r, t) =
1
2
[f(r,
0
) exp(i
0
t) + f

(r,
0
) exp(i
0
t)]
= [f(r,
0
)[ cos arg [f(r,
0
)]
0
t (1.10)
and its Fourier representation is
f
(r)
(r, ) =
1
2
[f(r,
0
)(
0
) + f

(r,
0
)( +
0
)] . (1.11)
Fields of this form are called harmonic; while strictly harmonic elds can never be encoun-
tered in nature, they can be well approximated by radiation originating from continuous-
wave single-mode lasers.
When considering strictly monochromatic elds, we will often suppress the explicit
dependence of spectral elds quantities f(r,
0
) on
0
to shorten the notation.
1.1 Preliminaries 3
1.1.2 Macroscopic Maxwells equations
Using complex analytic signal representations for all position- and time-dependent quan-
tities, we denote by E(r, t) and B(r, t) the fundamental eld vectors associated with
electromagnetic waves, namely the electric eld and the magnetic induction.
1
We denote
the primary sources of the eld, i.e., the free charge and current densities, by (r, t) and
J(r, t), respectively. Further, the polarization and magnetization of the (macroscopic)
medium are denoted by P(r, t) and M(r, t). Then the elds E(r, t) and B(r, t) satisfy, if
the medium is at rest,
2
Maxwells equations [1]
E(r, t) =

t
B(r, t), (1.12)
B(r, t) =
0
_
J(r, t) +M(r, t) +
0

t
E(r, t) +

t
P(r, t)
_
, (1.13)
E(r, t) =
1
0
[(r, t) P(r, t)] , (1.14)
and
B(r, t) = 0, (1.15)
where
0
is the permittivity and
0
is the permeability of vacuum. According to Eq. (1.13),
B is aected by true currents, magnetization currents, and currents due to the rates
of change of the electric eld and polarization. Similarly, in view of Eq. (1.14), both true
charges and polarization charges contribute to E.
In principle the polarization and magnetization are nonlinear functions of the electric
and magnetic elds. At the high frequencies in the optical region, magnetization is typi-
cally very small even if the eld strength is large (it can often be neglected, or at least a
linear dependence on B can be assumed). However, if intense laser illumination is used,
polarization must generally we written in the form of a series
P(r, t) =
(1)
(r)E(r, t) +
(2)
(r)E
2
(r, t) +
(2)
(r)E
3
(r, t) + . . . , (1.16)
where the (generally position-dependent) constants
(j)
are susceptibilities of dierent
orders. If it is sucient to retain only the rst term in the right-hand-side of Eq. (1.16),
we are in the domain of linear optics; the second and third terms give rise to nonlinear
eects of second and third order, respectively [2]. In this text we restrict the attention to
the linear domain, i.e., we assume that only the linear susceptibility
(1)
diers appreciably
from zero.
Maxwells equations can be simplied (at least formally if not in physical sense) by
introducing two new eld quantities, the electric displacement
D(r, t) =
0
E(r, t) +P(r, t) (1.17)
and the magnetic eld
H(r, t) =
1
0
B(r, t) M(r, t) (1.18)
1
The term induction is of historical origin.
2
Additional terms must be included if the medium is in motion, but we do not consider such situations
in this text.
4 1. Harmonic elds
to describe the eect of the medium. With the aid of these new eld vectors, Eqs. (1.13)
and (1.14) are transformed into the form
H(r, t) = J(r, t) +

t
D(r, t), (1.19)
D(r, t) = (r, t), (1.20)
while Eqs. (1.12) and (1.15) remain unchanged.
In addition to Maxwells equations, we need so-called constitutive relations describe the
eect of the medium in the eld vectors. Such relations generally depend on the frequency
in a much simpler way than they depend on time. Hence it is of advantage to tranform
Maxwells equations into the frequency domain using Fourier integral representations of
the analytic signals in the form of Eq. (1.6). Performing the temporal integrations, we
then have from Eqs. (1.12), (1.19), (1.20), and (1.15) the set
E(r, ) = iB(r, ), (1.21)
H(r, ) = J(r, ) iD(r, ), (1.22)
D(r, ) = (r, ), (1.23)
B(r, ) = 0. (1.24)
If we let the media be temporally dispersive but assume them to be spatially non-dispersive,
the constitutive equations depend on frequency but not on the spatial neighborhood of
the observation point r. It is then customary to write them in the form (applicable to
linear media only)
D(r, ) =
0

r
(r, )E(r, ), (1.25)
H(r, ) =
1
0

1
r
(r, )B(r, ), (1.26)
J(r, ) = (r, )E(r, ), (1.27)
where
r
is the relative permittivity of the medium,
r
is its relative permeability, and
is its electrical conductivity. In anisotropic media, where the optical properties depend
on direction, these scalar constants must be replaced by 3 3 tensors. However, in this
text, we will not consider anisotropic materials.
By applying the constitutive relations (1.25)(1.27), we may cast Eqs. (1.22) and (1.23)
into the form

1
r
(r, )B(r, )

= i
_
/c
2
_

r
(r, )E(r, ), (1.28)
[
r
(r, )E(r, )] = (r, ), (1.29)
where

r
(r, ) =
r
(r, ) + i(r, )/
0
(1.30)
is known as the complex relative permittivity of the medium and
c = (
0

0
)
1/2
(1.31)
is the speed of light in vacuum.
1.1 Preliminaries 5
As already pointed out, we can often (especially at optical frequencies) write
r
(r, ) =
1 since the materials can be considered non-magnetic. This obviously simplies Eq. (1.28)
and it also allows us to eectively consider H instead of B as the basic magnetic eld
vector since they dier only by the constant
0
. We will follow this convention in much of
the rest of the text since it is common practise in optics literature to desribe elds with
the aid of vectors E and H.
1.1.3 Measures of energy and ux
It is a simple matter to show (see Problem 8.1.1) that Maxwells equations (1.12) and
(1.19) lead to
3
_
S
_
E
(r)
(r, t) H
(r)
(r, t)
_
n dS +
_
V
E
(r)
(r, t) J
(r)
(r, t) dV
+
_
V
_
E
(r)
(r, t)

t
D
(r)
(r, t) +H
(r)
(r, t)

t
B
(r)
(r, t)
_
dV = 0 (1.32)
where V is an arbitrary volume, S is its surface, and n is the unit normal vector on
S. This result is known as the energy law for electromagnetic elds. It is exact since it
was derived directly from Maxwells equations. Here we consider real elds, instead of
complex analytic signals associated with them, since nonlinear functions of eld quantities
are involved.
Let us now assume that the eld is either harmonic or quasimonochromatic, meaning
that its frequency spectrum is conned over a narrow band in the neighborhood of a center
frequency
0
. We may then regard the permittivity, permeability, and conductivity in the
constitutive relations (1.25)(1.27) as constants across the frequency interval in question
4
and write
D
(r)
(r, t) =
0

r
(r)E
(r)
(r, t), (1.33)
H
(r)
(r, t) =
1
0

1
r
(r)B
(r)
(r, t), (1.34)
J
(r)
(r, t) = (r)E
(r)
(r, t). (1.35)
Equations (1.33) and (1.34) allow us to write
E
(r)
(r, t)

t
D
(r)
(r, t) =
1
2

t
_
E
(r)
(r, t) D
(r)
(r, t)
_
, (1.36)
H
(r)
(r, t)

t
B
(r)
(r, t) =
1
2

t
_
H
(r)
(r, t) B
(r)
(r, t)
_
. (1.37)
Then the quantity
E
V
(t) =
_
V
[w
e
(r, t) + w
m
(r, t)] dV, (1.38)
3
Using Eq. (1.1) one can immediately see that Maxwells equations are valid for real elds in the same
form as for complex analytic signals.
4
This amounts to the same thing as assuming that the materials are temporally non-dispersive, which
however is not true for real materials over a substantial spectral range: both
r
and are generally
complex and depend on .
6 1. Harmonic elds
where
w
e
(r, t) =
1
2
E
(r)
(r, t) D
(r)
(r, t) (1.39)
and
w
m
(r, t) =
1
2
H
(r)
(r, t) B
(r)
(r, t) (1.40)
may be interpreted (see, e.g., Ref. [3], Sect. 1.1.4) as the the total energy of the eld within
the volume V at time t. Furthermore, the quantities dened in Eqs. (1.38) and (1.39) can
be interpreted as the electric energy density and the magnetic energy density of the eld,
respectively. Consequently, the third term in Eq. (1.32) represents the rate of change
of energy in the volume V . The total current dentity J can be divided into conduction
and convection current densities, and doing this allows one to divide the second term
in Eq. (1.32) into Joule heat losses in a conductor and the rate of work done by the
electromotive forces on charges in the system. The rst term in Eq. (1.32) represents the
ow of energy through the boundary S. The integrand
S(r, t) = E
(r)
(r, t) H
(r)
(r, t) (1.41)
is known as the Poynting vector; its magnitude is often interpreted as the intensity of
the electromagnetic wave at point r and time t, and its direction is taken to represent
the direction of local energy ow. However, it must be stressed that such a pointwise
interpretation has no solid physical basis since only the integral of the Poynting vector
over the closed surface S has an unambiguous physical meaning. The interpretation is
nevertheless reasonable if the electromagnetic eld varies suciently slowly with position.
1.1.4 Response of ideal detectors
Detection of the electromagnetic eld is a complicated process involving the interaction
of the eld with the material in a detector having nite size, temporal response, and
frequency bandwidth. Direct measurement of amplitudes and phases of dierent compo-
nents of the eld vectors is straightforward at millimeter-range and longer wavelengths,
and possible also in the Terahertz (or quasi-optical) region.
5
However, at infrared and op-
tical wavelengths rather special techniques are required to obtain direct information on the
rapidly varying elds [46]. Usually the detector can not follow these rapid uctuations,
and therefore time averaged or time-integrated quantities are measured.
To keep things manageable, we consider an idealized situation in which the eld in
free space is detected by a point detector (no spatial averaging) placed behind an ideal
(tunable) narrow-band frequency lter, which permits one to measure one harmonic eld
component at a time. Therefore, the detected signal is taken to be the time average, for
a harmonic eld, of a quantity such as the electric or magnetic energy density, or the
Poynting vector.
6
5
There is much current interest in this spectral region, usually dened as the wavelength band from
100 m to 1 mm, or frequency band 0.3 3 THz.
6
Real detectors respond mainly to the electric energy density. Note that, even when measuring
ultrafast pulses, each ideal harmonic component lasts temporally forever. In practise, or course, the
intensity of the signal would vanish for any nite pulse energy if such an ideal delta-band lter could be
realized. Therefore we denitely deal with an idealized situation!
1.2 Plane-wave representation of harmonic elds 7
Considering the energy density at some setting frequency
0
, the detector measures
under the above-stated conditions the time average
w
e
(r, t) =

0
2
lim
T
1
2T
_
T
T
E
(r)
(r, t) E
(r)
(r, t) dt (1.42)
where we have used the constitutive relation (1.33) with
r
(r) = 1 since we are measuring
the eld in free space. Using the rst form of Eq. (1.10) we then obtain, after some simple
algebra,
w
e
(r, t) =

0
4
E

(r,
0
) E(r,
0
)
+

0
8
[E(r,
0
) E(r,
0
) +E

(r,
0
) E

(r,
0
)] lim
T
sin (2
0
T)
2
0
T
(1.43)
and the second term on the right-hand-side of Eq. (1.43) vanishes when the limit is taken.
We are therefore left with the observable quantity
w
e
(r, t) =

0
4
[E(r,
0
)[
2
. (1.44)
In a similar fashion, assuming
r
(r) = 1, we may derive
w
m
(r, t) =

0
4
[H(r,
0
)[
2
(1.45)
and
S(r, t) =
1
2
E

(r,
0
) H(r,
0
) , (1.46)
where again denotes the real part. This expression actually holds also in dielectric and
metallic (non-magnetic) media.
1.2 Plane-wave representation of harmonic elds
Throughout the rest of this Chapter we assume that the eld under consideration is strictly
harmonic at frequency
0
(which we will omit for brevity of notation). However, unless
stated otherwise, the medium in which the eld propagates is takes to be non-magnetic
with complex relative permittivity dened in Eq. (1.30). Here the quantities
r
and
may depend on both position and frequency, and they are generally complex. In dielectric
materials = 0 and
r
is nearly real, except close to atomic or molecular resonances of the
material (where electronic transitions between energy levels can take place). In metals,
diers from zero and is predominantly real, at least in the infrared region and at longer
wavelengths. However, in all cases we may dene the complex refractive index
n(r, ) =
_

r
(r, ) = n(r, ) + i(r, ) (1.47)
where n(r, ) and (r, ) are both real functions of position and frequency. Clearly, if
= 0 and
r
is real, n(r, ) =
_

r
(r, ) represents the real refractive index of a dielectric
material (far from resonances).
8 1. Harmonic elds
1.2.1 Harmonic plane waves
Let us consider harmonic elds in a non-magnetic homogeneous medium with n(r,
0
) = n
and
r
(r,
0
) = 1. Substituting from Eqs. (1.21) and (1.26) into Eq. (1.28), we obtain
[E(r)] = k
2
0
n
2
E(r). (1.48)
If we use the vector identity
(E) = ( E)
2
E (1.49)
and assume that the (source) charge density vanishes in the region under consideration,
i.e., (r) = 0 in Eq. (1.23), we see that the electric eld vector must satisfy the Helmholtz
equation

2
E(r) + k
2
0
n
2
E(r) = 0. (1.50)
Here we used Eq. (1.47) and denoted k
0
=
0
/c = 2/
0
, where
0
is the vacuum
wavelength of the monochromatic eld in question. The simplest solution of Eq. (1.50) is
the harmonic plane wave
E(r) = Aexp (ik r) , (1.51)
where A is called the polarization vector and k the wave vector. Substituting this expres-
sion in Eq. (1.50), we see that the wave vector must satisfy the condition
k k = k
2
0
n
2
. (1.52)
If we write, in cartesian wave-vector coordinates, k = (k
x
, k
y
, k
z
) and take = (k
x
, k
y
) as
a real-valued spatial-frequency vector, Eq. (1.52) xes
k
z
=
_
k
2
0
_
n
2

2
_
k
2
x
k
2
y
+ i2k
2
0
n

1/2
= k
zr
+ ik
zi
. (1.53)
Here k
zr
and k
zi
are the real and imaginary parts of k
z
, respectively. Hence we may write
the space-dependent part of the electric eld vector in the form
exp (ik r) = exp [i (k
x
x + k
y
y + k
zr
z)] exp (k
zi
z) , (1.54)
which is a product of rapidly oscillating and real-exponential parts. When considering
elds in the positive half-space, we always choose the branch of the square root that
gives a positive value for k
zr
. In lossy media with > 0 we choose the branch such that
also k
zi
> 0 so that the eld amplitude decreases exponentially in the z-direction. In
amplifying media with < 0 we take k
zi
< 0, which implies that the eld amplitude
increases exponentially with z.
Lossless dielectrics represent an important special case worth separate consideration.
Writing = 0 we obviously have, from Eq. (1.53),
k
z
=
_
k
zr
=
_
k
2
0
n
2
k
2
x
k
2
y
_
1/2
if k
2
x
+ k
2
y
k
2
0
n
2
ik
zi
= i
_
k
2
x
+ k
2
y
k
2
0
n
2
_
1/2
if k
2
x
+ k
2
y
> k
2
0
n
2
(1.55)
when considering plane waves in the positive half-space. Then we have homogeneous waves
with real-valued k
z
and exponentially decaying evanescent waves with purely imaginary
1.2 Plane-wave representation of harmonic elds 9
k
z
. In accordance with the general case, the positive root in the lower branch of Eq. (1.55)
ensures that evanescent waves die out when z increases.
Returning to the general case, the real and imaginary parts of the z-component of the
wave vector k, dened in Eq. (1.53), may be written as
k
zr
= [k
z
[ cos , (1.56)
k
zi
= [k
z
[ sin , (1.57)
where = arg (k
z
). To make the notation more compact, we dene further the normalized
radial spatial frequency f as
f = [[/k
0
=
_
(k
x
/k
0
)
2
+ (k
y
/k
0
)
2
. (1.58)
Noting that k
2
z
= [k
z
[
2
exp (i2), we obtain from Eqs. (1.56)(1.58) the relations
[k
z
[
2
cos(2) = k
2
0
_
n
2

2
f
2
_
, (1.59)
[k
z
[
2
sin(2) = 2k
2
0
n. (1.60)
Squaring and adding these two expressions, we see that
[k
z
[
2
= k
2
0
_
_
n
2

2
f
2
_
2
+ (2n)
2
_
1/2
. (1.61)
Clearly, Eqs. (1.59)(1.61) dene k
z
unambiguously. Further, using well-known trigono-
metric identities sin
2
x = [1 cos(2x)] /2 and cos
2
x = [1 + cos(2x)] /2, we obtain explicit
expressions for the real and imaginary parts of the wave vector:
k
2
zr
=
1
2
k
2
0
_
_
_
n
2

2
f
2
_
2
+ (2n)
2
_
1/2
+
_
n
2

2
f
2
_
_
, (1.62)
k
2
zi
=
1
2
k
2
0
_
_
_
n
2

2
f
2
_
2
+ (2n)
2
_
1/2

_
n
2

2
f
2
_
_
. (1.63)
The square roots are taken according to the criteria stated above: The right-hand-sides
of these expressions are always positive, that of Eq. (1.62) being monotonically increasing
and that of Eq. (1.63) being monotonically decreasing as illustrated in Figs. 1.1 and 1.2.
We therefore always choose a positive square root when calculating k
zr
. A positive sign
is taken for k
zi
in lossy media and a negative one in gain media.
Proceeding to consider the polarization vector, we see from Eq. (1.23) that k A = 0.
This implies that only two of the cartesian components of the polarization vector are
independent; taking these to be A
x
and A
y
we have
A
z
= k
1
z
(k
x
A
x
+ k
y
A
y
) . (1.64)
In view of Eq. (1.21), the magnetic eld vector is of the form
H(r) =
1
k
0
_

0
(k A) exp (ik r) . (1.65)
10 1. Harmonic elds
0 1 2 3 4 5
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
k
z
r
f
(a)
0 1 2 3 4 5
0
1
2
3
4
5
k
z
i
f
(b)
Figure 1.1: (a) Real and (b) imaginary parts of k
z
in lossy media. Solid line:
n = 1.5 + i1.0 (high loss). Dashed line: n = 1.5 + i0.1 (smaller loss). Dotted line:
n = 1.5+i0.01 (nearly dielectric). The dotted line resembles the result of Eq. (1.55).
0 1 2 3 4 5
0.0
0.1
0.2
0.3
0.4
0.5
k
z
r
f
(a)
0 1 2 3 4 5
0
2
4
6
8
k
z
i
f
(b)
Figure 1.2: (a) Real and (b) imaginary parts of k
z
in metal-like media. Solid line:
n = 0.5+i1.0 (poor metal). Dashed line: n = 0.5+i2.0 (rather poor metal). Dotted
line: n = 0.5 + i5.0 (decent metal).
It follows from simple vector algebra that the time-averaged Poynting vector, dened in
Eq. (1.46), is given by
S(r, t) =
1
2k
0
_

0
[A[
2
k . (1.66)
Hence, for the special case of homogeneous plane waves with k = k, the Poynting
vector is parallel to the wave vector. Since (k A) = 0 in view of Eq. (1.24) and
we already saw that k A = 0, the direction of S(r, t) and k is perpendicular to the
oscillation direction of the electric and magnetic elds. Therefore homogeneous plane
electromagnetic waves are often called transverse waves.
Returning again to the general case, all waves with k
zi
,= 0 have surfaces of constant
amplitude in planes z = constant but the surfaces of equal phase = constant are given
1.2 Plane-wave representation of harmonic elds 11
by
k r = k
x
x + k
y
y + k
zr
z = . (1.67)
Apart from the special case k
x
= k
y
= 0 (an attenuated homogeneous wave), these
surfaces are inclined with respect to planes z = constant while still being perpendicular
to the Poynting vector; such waves are called inhomogeneous. In case of purely evanescent
waves, S(r, t) is clearly conned to the xy plane and the surfaces of constant phase are
perpendicular to this plane.
The wavelength of a plane wave in a medium is dened as the perpendicular distance
between two surfaces of constant phase, having a phase dierence equal to 2 radians.
Since we just saw that S(r, t) is perpendicular to the surfaces of constant phase in all
cases, the vector
n =
S(r, t)
[S(r, t)[
=
k
[k[
(1.68)
is the unit surface normal of the plane wave front and the wavelength is
=
2
[k[
. (1.69)
Another important characteristic of plane waves is the eective decay distance, dened as
the distance in which the eld amplitude reduces to 1/e of the original value. Considering
the real-exponential part of Eq. (1.54), this distance is
z
d
=
1
k
zi
. (1.70)
Obviously, for propagating waves, z
d
= .
Let us again rst consider the special case of lossless media. Then, for homogeneous
plane waves, the wavelength is
=

0
n
(1.71)
and for evanescent waves it is
=

0
f
. (1.72)
Thus the wavelength of a homogeneous wave in a lossless medium depends (apart from

0
) only on the refractive index n, but in case of evanescent waves it only depends on the
radial spatial frequency. However, since the transition from homogeneous to evanescent
waves takes place at f = n, there is no jump in at this point. Therefore the wavelengths
of evanescent waves can be arbitrarily small in any medium including vacuum, and
0
,
which is often referred to as the vacuum wavelength of light, actually refers only to
the wavelength of homogeneous waves. We will later nd that this result has extremely
important implications in nanophotonics.
7
The decay distance of purely evanescent waves
is
z
d
=

0
2
_
f
2
n
2
(1.73)
7
While evanescent waves in air can be detected directly by suitable probing techniques to be described
later, direct observation of the wavelength inside a medium is not possible. However, indirect observation
through dierent kinds of interactions is possible. If, e.g., the material is light-sensitive, interference of
two (or more) waves generates a pattern that depends directly on the wavelength of the waves involved.
12 1. Harmonic elds
with f > n. Thus, when f n, the decay distance becomes innite. However, it
reaches subwavelength scale when f >
_
n
2
+ 1/(2)
2
. In vacuum, for example, this
subwavelength decay threshold is f 1.0126, which is only slightly above the evanescent-
wave threshold f = 1. Therefore it can generally be said that evanescent waves do not
contribute signicantly to the eld at distances greater than a few wavelengths.
8
In media with complex refractive index the expressions for the wavelength and the
eective decay distance read as
=

2
0
_
_
(n
2

2
f
2
)
2
+ (2n)
2

1/2
+ (n
2

2
+ f
2
)
_
1/2
(1.74)
and
z
d
=

0

2
_
_
(n
2

2
f
2
)
2
+ (2n)
2

1/2
(n
2

2
f
2
)
_
1/2
. (1.75)
Figures 1.3 and 1.4 illustrate the wavelength and eective decay distance, as a function of
the normalized radial spatial frequency, of plane waves in several dielectric and metallic
media. It is noteworthy that the wavelengths of plane waves in some metals can be
substantially larger than
0
for small values of f. This occurs for metals with n < 1 for
small values of f since, for a homogeneous wave with f = 0, we have =
0
/n.
0 2 4 6 8 10
0.0
0.2
0.4
0.6
0.8
1.0

0
f
(a)
0 1 2 3 4 5
0.0
0.2
0.4
0.6
0.8
1.0
z
d
/

0
f
(b)
Figure 1.3: (a) Wavelength and (b) eective penetration depth of plane waves in
dielectric media with n = 1 (vacuum). n = 1.45 (SiO
2
), and n = 2.2 (TiO
2
).
Figure 1.5 illustrates the dierent types of plane waves. Here we consider the real
(monochromatic) eld at t = 0, i.e.,
E
(r)
(r, 0) = Acos (k
x
x + k
y
y + k
zr
z) exp (k
zi
z) , (1.76)
assuming that A = y and k
y
= 0 so that the electric eld has only a single scalar
component that varies only in the xz plane. In (a) we assume n = 1.45 and k
x
= 0.5k
0
8
The small range of evanescent elds is another property of evanescent waves, which is highly sig-
nicant in nanophotonics. This property implies that observations of evanescent waves must be made at
distances within around one vacuum wavelength of the region where they are created.
1.2 Plane-wave representation of harmonic elds 13
0 1 2 3 4 5
0
1
2
3
4
5

0
f
(a)
0 2 4 6 8 10
0.00
0.01
0.02
0.03
0.04
0.05
z
d
/

0
f
(b)
Figure 1.4: (a) Wavelength and (b) eective decay distance of plane waves in
metals at
0
= 633 nm with n = 1.45 + i7.5 (Aluminium, solid), n = 0.20 + i3.09
(Gold, dashed), n = 3.68 + i3.62 (Chromium, dotted).
to illustrate a homogeneous wave propagating in a dielectric. In (b) and (c) we illustrate
evanescent waves in the same dielectric by choosing k
x
= 1.5k
0
(just above the threshold
of becoming evanescent) and k
x
= 2.5k
0
, respectively; soon after the evanescent-wave
threshold is reached, the eective decay distance in a dielectric becomes less than one
wavelength. Plane waves in metals are illustrated in (d)(f): we assume n = 1.45 + i7.5,
which corresponds roughly to the complex refractive index of aluminium. In (d) we
have k
x
= 0, which generated a decaying homogeneous wave. In (e) and (f) we choose
k
x
= 1.5k
0
and k
x
= 2.5k
0
to illustrate inhomogeneous waves in metal. Clearly, waves in
metals decay appreciably in a scale that is a small fraction of the vacuum wavelength.
1.2.2 Polarization of a plane wave
Let us next look more closely at the concept of polarization or propagating plane waves.
We saw that only two components of the polarization vector A are independent and that
k A = 0. Hence generality is not sacriced if we assume k = z and consider A
x
and
A
y
as the independent components (in this case A
z
= 0). If we set A
x
= a
x
exp (i
x
) and
A
y
= a
y
exp (i
y
), the real representations of the transverse components of the electric
eld read
E
(r)
x
(z, t) = E
rx
= a
x
cos ( +
x
) , (1.77)
E
(r)
y
(z, t) = E
ry
= a
y
cos ( +
y
) , (1.78)
where we have abbreviated the notation and written = k
0
nz . These are in general
parametric equations of an ellipse, dened by the evolution of the tip of the electric eld
vector, which is known as the polarization ellipse.
We may now use standard trigonometric identities and simple manipulations to put
the two equations in a form
E
2
rx
a
2
x
+
E
2
ry
a
2
y

2E
rx
E
ry
a
x
a
y
cos = sin
2
, (1.79)
14 1. Harmonic elds
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
x/
0
z
/

0
(a)
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
x/
0
z
/

0
(b)
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
x/
0
z
/

0
(c)
0 0.5 1 1.5 2
0
0.02
0.04
0.06
0.08
0.1
x/
0
z
/

0
(d)
0 0.5 1 1.5 2
0
0.02
0.04
0.06
0.08
0.1
x/
0
z
/

0
(e)
0 0.5 1 1.5 2
0
0.02
0.04
0.06
0.08
0.1
x/
0
z
/

0
(f)
Figure 1.5: Amplitude of the real eld E
(r)
y
(x, z, 0) for (a) a propagating homoge-
neous plane wave, (b,c) evanescent plane waves, (d) a homogeneous decaying wave
and (e,f) inhomogeneous plane waves. Note the 20 magnied z-scale in the lower
row compared to the upper row. White means unit eld amplitude, the darkest
shade means 1, and the middle shade represents zero amplitude.
where =
y

x
. This expression describes more explicitly a rotated ellipse. Note that,
unlike Eqs. (1.77) and (1.78), the expression (1.79) no longer contains information on the
direction of rotation of the tip along the ellipse. A simpler axpression for tje polarization
ellipse is obtained by rotating the coordinate axes such that they are parallel to the major
and minor axes as shown in Fig. 1.6. In this new coordinate system we have
E
r
= E
rx
cos + E
ry
sin , (1.80)
E
r
= E
rx
sin + E
ry
cos , (1.81)
and the equation of the ellipse is simply
E
2
r
a
2
+
E
2
r
b
2
= 1. (1.82)
Let us now multiply Eq. (1.79) by a
2
x
a
2
y
and Eq. (1.82) by a
2
b
2
and compare the resulting
expressions. We then see that
a
2
x
= a
2
cos
2
+ b
2
sin
2
, (1.83)
1.2 Plane-wave representation of harmonic elds 15
a
2
y
= a
2
sin
2
+ b
2
cos
2
, (1.84)
and also
2a
x
a
y
cos =
_
a
2
b
2
_
sin(2), (1.85)
a
x
a
y
sin = ab. (1.86)
We have therefore related the properties of the ellipse in the two coordinate systems.
x
y

a
x
a
y
a
0
Figure 1.6: The polarization ellipse with major and minor semi-axes of length a
and b, inclined at an angle with respect to the cartesian coordinate axes.
It is useful to derive explicit expressions for the properties of the elliple in the rotated
system to see its main features more clearly. To this end, we dene angles and by
writing
9
a
x
= a
0
sin , a
y
= a
0
cos (1.87)
and
tan
2
= b
2
/a
2
. (1.88)
Adding Eqs. (1.83) and (1.84), we rst have the relations
a
2
+ b
2
= a
2
x
+ a
2
y
= a
2
0
. (1.89)
Further, subtracting Eq. (1.84) from Eq. (1.83) and inserting into Eq. (1.85), we see that
tan(2) =
2a
x
a
y
a
2
x
a
2
y
cos = tan(2) cos . (1.90)
9
The angle is widely used in ellipsometry, along with , to characterize the state of polarization of
a plane wave. In fact these two angles are often referred to as ellipsometric angles: see, e.g., Ref. [8],
noting that the notation there are dierent from those adopted here. We will return to the subject of
ellipsometry in Sect. 3.2.3.
16 1. Harmonic elds
To obtain the last form, we have used the identity tan(2x) = 2 tanx/ (1 + tan
2
x). This
expression gives the polarization azimuth rotation angle, dened in the range 0 < ,
which is the angle between the major axis of the ellipse with respect to the x axis as
illustrated in Fig. 1.6. The ellipticity angle [/4, /4) denes the ratio of the
lengths of the major and minor axes of the ellipse. Using Eqs. (1.88), (1.86), (1.87), and
the identity sin(2x) = 2 tanx/ (1 + tan
2
x) we have
sin(2) =
2a
x
a
y
a
2
x
+ a
2
y
sin = sin(2) sin. (1.91)
The angles and , together with a, clearly dene the orientation, shape, and size of the
polarization ellipse uniquely.
Since the tip of the electric eld vector denes an ellipse, a monochromatic plane wave
is said to be elliptically polarized in general. If sin > 0, also 0 < /2 and the wave
is said to be right-handed elliptically polarized. In this case the tip is seen to rotate in
a clockwise sense if observed from the direction in which the wave propagates. In the
opposite case, i.e., when sin < 0 and /4 < 0, the polarization is left-handed and
the observer sees a counter-clockwise rotation of the tip.
The ellipse is known to have two degenerate states, namely a line and a circle. If = 0
or = , Eq. (1.79) is seen at once to reduce to a line, i.e.,
E
ry
=
a
y
a
x
E
rx
, (1.92)
where the positive sign corresponds to = 0 and the negative sign to = . Then the
light is said to be linearly polarized, having a slope a
y
/a
x
with respect to the x axis. If,
on the other hand, a
x
= a
y
= a
0
and = /2, Eq. (1.79) reduces to the equation of a
circle:
E
2
rx
+ E
2
ry
= a
2
0
. (1.93)
Then wave is said to be right circularly polarized (RCP) if = /2 and left circularly
polarized (LCP) if = /2.
We just saw that any monochromatic plane wave is necessarily elliptically polarized.
Thus light with no preferential polarization direction, i.e., unpolarized light, cannot be
monochromatic. To describe such light, it is necessary to make use of statistical concepts,
and we will return to this later.
It is often convenient to use a shorthand notation for polarization properties of a plane
wave by writing the real electric eld in the form
E
(r)
(z, t) = J exp (i) , (1.94)
where the column vector
J =
_
A
x
A
y
_
=
_
a
x
exp (i
x
)
a
y
exp (i
y
)
_
= exp (i
x
)
_
a
x
a
y
exp (i)
_
(1.95)
(with =
y

x
as before) is known as the Jones vector.
10
It contains all relevant
information about the polarization properties of a plane wave. An arbitrary state of
10
Not to be confused with the electric current density, which was denoted by the same symbol.
1.2 Plane-wave representation of harmonic elds 17
polarization J can be represented as a superposition of two orthogonal states J
1
and J
2
with [J
1
[
2
= [J
2
[
2
= 1 and J

1
J
2
= 0 so that
J = c
1
J
1
+ c
2
J
2
(1.96)
with
c
1
= J

1
J, c
2
= J

2
J. (1.97)
Examples of such orthogonal states are the x and y linearly polarized states
J
1
=
_
1
0
_
, J
2
=
_
0
1
_
(1.98)
with c
1
= A
x
and c
2
= A
y
, as well as the RCP and LCP states
J
1
=
1

2
_
1
i
_
, J
2
=
1

2
_
1
i
_
(1.99)
with c
1
=
1

2
(A
x
iA
y
) and c
2
=
1

2
(A
x
+ iA
y
).
1.2.3 Angular spectrum of plane waves
We saw above that the electric eld vector of any harmonic wave satises, in a homo-
geneous medium, the Helmholtz equation (1.50), and is is easy to show that also the
magnetic eld vector satises precisely the same equation. We now proceed to show that
the general solutions of these equations (and in fact those of Maxwells equations) can be
represented as superpositions of plane waves. To this end, and also later on throughout
this text, it is convenient to introduce certain notations for the scalar components of the
eld vectors. For frequency-domain (and harmonic) elds we write V
j
(r,
0
) = V
j
(r) with
the index j running such that
V
1
= E
x
, V
2
= E
y
, V
3
= E
z
, V
4
= H
x
, V
5
= H
y
, V
6
= H
z
. (1.100)
For time-dependent analytic signals we will analogously write V
j
(r, t) with
V
1
= E
x
, V
2
= E
y
, V
3
= E
z
, V
4
= H
x
, V
5
= H
y
, V
5
= H
z
(1.101)
and a similar convention will be adopted for real eld components V
(r)
j
(r, t).
Obviously, each component V
j
(r) of a harmonic eld satises the same Helmholtz
equation as the eld vectors do. Let us introduce the two-dimensional Fourier integral
representations
V
j
(, z) =
__

A
j
(, z) exp (i ) d
2
(1.102)
with = (x, y) and = (k
x
, k
y
). Then we have, for the functions
A
j
(, z) =
1
(2)
2
__

V
j
(, z) exp (i ) d
2
(1.103)
18 1. Harmonic elds
an ordinary dierential equation
d
2
dz
2
A
j
(, z) + k
2
z
A
j
(, z) = 0, (1.104)
where k
z
is dened in Eq. (1.53). The solutions of this equation are linear combinations of
terms of the form exp (ik
z
z). Considering waves that propagate (or decay) in the positive
z-directions, we disregard the solution with the negative sign and write the general solution
in the form
A
j
(, z) = A
j
(, z
0
) exp (ik
z
z) , (1.105)
where z = z z
0
and z = z
0
is an arbitrary reference plane.
In view of the preceding results, the general solution of our original Helmholtz equation
can be written as
V
j
(, z) =
__

A
j
(, z
0
) exp (ik
z
z) exp (i ) d
2
. (1.106)
This is a continuous superposition of plane waves. If we again consider A
1
= E
x
and
A
2
= E
y
as xed, we have
A
3
(, z
0
) = k
1
z
[k
x
A
1
(, z
0
) + k
y
A
2
(, z
0
)] . (1.107)
in analogy with Eq. (1.64). Writing C
0
= k
1
0
_

0
/
0
, we also have
A
4
(, z
0
) = C
0
k
1
z
_
k
x
k
y
A
1
(, z
0
) +
_
k
2
y
+ k
2
z
_
A
2
(, z
0
)

, (1.108)
A
5
(, z
0
) = C
0
k
1
z
__
k
2
x
+ k
2
z
_
A
1
(, z
0
) + k
x
k
y
A
2
(, z
0
)

, (1.109)
and
A
6
(, z
0
) = C
0
[k
y
A
1
(, z
0
) k
x
A
2
(, z
0
)] (1.110)
in analogy with Eq. (1.65). Using Eq. (1.103) at z = z
0
, we can evaluate A
1
and A
2
if V
1
and V
2
are known at that plane. Then Eqs. (1.106) are applied to obtain the remaining
components A
3
to A
6
, and nally all scalar components of the electromagnetic eld can be
calculated from Eq. (1.106) anywhere in the half-space z > z
0
. In conclusion, we now have
a complete solution of the electromagnetic wave propagation problem in a homogeneous
medium, expressed in terms of (homogeneous and inhomogeneous) plane waves.
The functions A
j
(, z
0
) are known as the angular spectra of the scalar eld components,
and eld expansions in Eq. (1.106) are known as angular spectrum representations of
the associated eld components. This nomenclature arises from the interpretation of the
spatial frequency vector as a vector dening the propagation direction of a homogeneous
plane wave. To see this, we dene polar coordinates (, ) and spherical polar angles (, )
by equations
_
_
_
k
x
= cos = k
0
nsin cos
k
y
= sin = k
0
nsin sin
k
z
= k
0
ncos
(1.111)
and
= [[ =
_
k
2
x
+ k
2
y
_
1/2
= k
0
nsin . (1.112)
1.2 Plane-wave representation of harmonic elds 19
Now the angles and dene uniquely the direction of the wave vector k provided that
the medium is dielectric and sin < 1. Similarly, we will often employ spherical polar
coordinates also for the position vector:
_
_
_
x = cos = r sin cos
y = sin = r sin sin
z = r cos
(1.113)
with
= [[ =
_
x
2
+ y
2
_
1/2
= r sin (1.114)
and
r = [r[ =
_

2
+ z
2
_
1/2
=
_
x
2
+ y
2
+ z
2
_
1/2
. (1.115)
The notations just dened are illustrated in Fig. 1.7, which also shows the unit position
vector s = r/r.
(a) (b)
x
y
z
k
x
k
y
k
z

r s
k
z = z
0

Figure 1.7: Notations used for (a) wave-vector coordinates of a propagating wave
and (b) position coordinates, with denitions of spherical polar angles in dielectric
media.
If we use polar coordinates for spatial frequencies but cartesian components for the
position vector, the angular spectrum representation of the eld components takes the
form
V
j
(x, y, z) =
_

0
_
2
0
A
j
(, , z
0
) exp [i (xcos + y sin ) + ik
z
z] d d (1.116)
where
A
j
(, , z
0
) =
1
(2)
2
__

V
j
(x, y, z) exp [i (xcos + y sin )] dxdy (1.117)
and k
2
z
= k
2
o
n
2

2
, with sign conventions in taking the square root as described above.
If we use polar coordinates also for the transverse position vector, the angular spectrum
20 1. Harmonic elds
representation reads as
V
j
(, , z) =
_

0
_
2
0
A
j
(, , z
0
) exp [i cos( ) + ik
z
z] d d (1.118)
with
A
j
(, , z
0
) =
1
(2)
2
_

0
_
2
0
V
j
(, , z
0
) exp [i cos( )] d d. (1.119)
Finally, if we use spherical polar coordinates for both spatial frequencies and position co-
ordinates, assume a purely dielectric medium, and include only propagating plane waves,
the angular spectrum representation takes the form
V
j
(, , z) = k
2
_
/2
0
_
2
0
sin cos A
j
(, , z
0
)
exp ik [ sin cos( ) + z cos ] dd (1.120)
with
A
j
(, , z
0
) =
1
(2)
2
_

0
_
2
0
V
j
(, , z
0
) exp [ik sin cos( )] d d. (1.121)
Explicit expressions for A
j
and V
j
in terms of A
1
and A
2
are obtained straightforwardly
using Eqs. (1.107)(1.110). We will nd these last results very useful when dealing with
focusing and imaging problems of electromagnetic elds.
1.2.4 Far-zone elds
The angular spectrum representation allows us to propagate all scalar eld components
over any nite distance in a homogeneous medium. The behavior of the eld is, however,
simplied considerably if we proceed suciently far from the source. In fact, we will
see that if r in a homogeneous dielectric with n = n, the radiation eld can be
interpreted as a geometrical (electromagnetic) wave with a well-dened local propagation
direction.
Let us write the right-hand-side of Eq. (1.106) explicitly in terms of the cartesian
coordinates of the wave vector and the unit position vector and make use of the upper
branch of Eq. (1.55). We then have
V
j
(r s) =
__

A
j
(k
x
, k
y
, z
0
) exp [ir (k
x
s
x
+ k
y
s
y
+ k
z
s
z
)] dk
x
dk
y
. (1.122)
where the argument of the exponential term is real. This integral that appears here is of
the general form
F(u) =
__

f(x, y) exp[iug(x, y)] dxdy. (1.123)

In the asymptotic limit u , the integral in Eq. (1.123) can be evaluated by the method
of stationary phase. If g(x, y) is real, its limiting value is (see Sect. 3.3.3 in Ref. [7])
F(u) =
i2f(x
c
, y
c
)
_
[
2
[
exp [iug(x
c
, y
c
)]
u
, (1.124)
1.2 Plane-wave representation of harmonic elds 21
where x
c
and y
c
are the (transverse) coordinates of a critical point (of the rst kind).
These coordinates may be determined by solving the equations

x
g(x, y) =

y
g(x, y) = 0 . (1.125)
Furthermore,
=
_
_
_
1 if >
2
and 0
1 if >
2
and < 0
i if <
2
(1.126)
and
=

2
x
2
g(x, y) [
(xc,yc)
, (1.127)
=

2
y
2
g(x, y) [
(xc,yc)
, (1.128)
=

2
xy
g(x, y) [
(xc,yc)
, (1.129)
where [
(xc,yc)
means evaluation at the critical point.
The integral on the right-hand-side of Eq. (1.122) is precisely of the form of that in
Eq. (1.123) if we make the substitutions u r, (x, y) (k
x
, k
y
), f(x, y) A(k
x
, k
y
),
and g(x, y) g(k
x
, k
y
) with
g(k
x
, k
y
) = s
x
k
x
+ s
y
k
y
+ s
z
k
z
. (1.130)
Let us calculate the partial derivatives of g(k
x
, k
y
). Recalling that k
z
= (k
2
k
2
x
k
2
y
)
1/2
,
where we have denoted k = k
0
n for brevity, and that s is a unit vector, we nd nd that
the coordinates of the critical point are to be determined from equations
s
x
s
z
k
x
k
z
= 0, (1.131)
s
y
s
z
k
y
k
z
= 0. (1.132)
The (only) solution is
k
xc
= ks
x
, k
yc
= ks
y
, k
zc
= ks
z
. (1.133)
Therefore we have
g (k
xc
, k
yc
) = k (1.134)
and
= k
1
_
1 + (s
x
/s
z
)
2

, (1.135)
= k
1
_
1 + (s
y
/s
z
)
2

, (1.136)
= k
1
s
x
s
y
/s
2
z
. (1.137)
Hence
2
= (ks
z
)
2
> 0, which implies that = 1 in Eq. (1.126), and we have
V
j
(r s) = i2ks
z
A
j
(ks
x
, ks
y
, z
0
)U(r) , (1.138)
22 1. Harmonic elds
where
U(r) =
exp (ikr)
r
(1.139)
is the expression for a diverging scalar spherical wave. Hence we have established that each
component of the electromagnetic eld behaves, in the far zone, as a diverging spherical
wave modulated by the angular spectrum of plane waves A
j
(ks
x
, ks
y
, z
0
) associated with
this particular eld component.
If we now use representations of the form of Eq. (1.138) for the transverse components
V
1
and V
2
of the electric eld vector, Eq. (1.107) implies that the longitudinal component
of the electric vector takes the form
V
3
(r s) = i2k [s
x
A
1
(ks
x
, ks
y
, z
0
) + s
y
A
2
(ks
x
, ks
y
, z
0
)] U(r) , (1.140)
Furthermore, in view of Eqs. (1.108)(1.110), the transverse scalar components of the
magnetic eld are given by
V
4
(r s) = i2k
2
C
0
_
s
x
s
y
A
1
(ks
x
, ks
y
, z
0
) +
_
s
2
y
+ s
2
z
_
A
2
(ks
x
, ks
y
, z
0
)

U(r) , (1.141)
V
5
(r s) = i2k
2
C
0
__
s
2
x
+ s
2
z
_
A
1
(ks
x
, ks
y
, z
0
) + s
x
s
y
A
2
(ks
x
, ks
y
, z
0
)

U(r) , (1.142)
and the longitudinal component is
V
6
(r s) = i2k
2
C
0
s
z
[s
y
A
1
(ks
x
, ks
y
, z
0
) s
x
A
2
(ks
x
, ks
y
, z
0
)] U(r) . (1.143)
The time-averaged Poynting vector in the far zone is obtain after somewhat lengthy but
straightforward calculation from the denition in Eq. (1.46), using the eld representation
just derived. The result is (see Problem 8.1.8)
S(r s, t) = J( s)
s
r
2
, (1.144)
where
J( s) = J
0
_
[s
z
A
1
(ks
x
, ks
y
, z
0
)[
2
+[s
z
A
2
(ks
x
, ks
y
, z
0
)[
2
+ [s
x
A
1
(ks
x
, ks
y
, z
0
) + s
y
A
2
(ks
x
, ks
y
, z
0
)[
2

(1.145)
and
J
0
=
n
2
_

0
(2k)
2
. (1.146)
This result has important implications. If we consider the direction of S(r s, t) as the
propagation direction of the eld, we see that this direction is always that of the unit
position vector s of the observation point. On the other hand, we may now safely interpret
the magnitude of S(r s, t) as the optical intensity in a local sense since the eld at
position rs is fully determined by a single plane-wave component in the angular spectrum,
with wave vector k parallel to s.
The quantity J( s) is known as radiant intensity. It expresses the angular distribu-
tion of the radiation pattern in the far zone: using the spherical polar angles dened in
Eq. (1.113) we have
J( s)/J
0
= cos
2

_
[A
1
(ks
x
, ks
y
, z
0
)[
2
+[A
2
(ks
x
, ks
y
, z
0
)[
2

+sin
2
[cos A
1
(ks
x
, ks
y
, z
0
) + sin A
2
(ks
x
, ks
y
, z
0
)[
2
(1.147)
1.2 Plane-wave representation of harmonic elds 23
with s
x
= sin cos and s
y
= sin sin . Hence the radiation pattern in the far zone is
seen to depend critically on the relation between A
1
and A
2
, i.e., on the polarization of
the eld at z = z
0
.
The derivation presented above is not formally valid if n is complex because then the
function g(k
x
, k
y
) dened in Eq. (1.130) is no longer real. However, if we allow the critical
point have complex coordinates, the same procedure can be carried through with the
critical point k
xc
= k
0
ns
x
, k
yc
= k
0
ns
y
and the nal result is the same as Eqs. (1.138)
(1.143), but with k = k
0
n. We will, however, justify numerically in Sect. 1.3.1 that, in
weakly absorbing of amplifying media with [[ n, k = k
0
n can be used in Eq. (1.138)
everywhere except in the expression (1.139), which must be written in the form
U(r) =
exp (ik
0
nr)
r
(1.148)
to take absorption or amplication properly into account.
1.2.5 Polarization of harmonic elds
We saw above that, in the far zone, the electromagnetic eld in a dielectric medium is
a diverging spherical wave modulated by the angular spectrum of the eld at z = z
0
.
The scalar components of the angular spectrum depend only on the spherical polar angles
and or, equivalently, on the angles and , Therefore, the eld can be considered
locally as a plane wave propagating in this precise direction. Hence it is meaningful to
dene the local state of polarization of the eld in the far zone by considering a Jones
vector
J(, ) =
_
A

(, )
A

(, )
_
=

A

(, ) + A

(, ), (1.149)
where

and are unit vectors in the two orthogonal and directions, and we have
dropped the dependence of A

(, ) and A

(, ) on z
0
for brevity.
As indicated by the notation, the complex amplitudes A

= a

exp (i

) and A

=
a

exp (i

) may generally depend on propagation direction. At least in principle, the

values of a

, a

and

can be measured using a goniometric setup to determine the

local state of polarization experimentally. If it is found that a

= 0 for all and , the

angular spectrum is said to be azimuthally polarized. If, on the other hand, a

= 0 for all
and , the angular spectrum is said to be radially polarized. Since the unit vectors

and
are orthogonal and the polarization vector does not have a component in the direction
of s, the angular spectrum may always be expressed as a superposition of its azimuthally
and radially polarized components as indicated in Eq. (1.149). Its representation in the
polar (, , z) coordinate system is (A

, A

, A
z
) with
A

= A

cos , (1.150)
A
z
= A

sin . (1.151)
In the cartesian coordinate system we get
A
1
= A

cos A

sin = A

cos cos A

sin , (1.152)
24 1. Harmonic elds
A
2
= A

sin + A

cos = A

cos sin + A

cos , (1.153)
A
3
= A

sin . (1.154)
In writing these transformations, we have dropped the explicit dependence of angular
spectra on the angles and for brevity. Using the two orthogonal components of the
angular spectrum, the radiant intensity assumes the simple form
J( s) = J
0
cos
2

_
[A

(ks
x
, ks
y
, z
0
)[
2
+[A

(ks
x
, ks
y
, z
0
)[
2

(1.155)
since the interference term vanishes (see Problem 8.1.9).
The polarization state of the eld in any point of space can be determined by inserting
Eqs. (1.152) and (1.152) into the angular spectrum representations of the components V
1
,
V
2
, and V
3
of the electric eld vector. In general the eld is elliptically polarized; the
ellipticity, orientation of the major axis, and the plane dened by the tip of the electric
eld can vary with position. All these properties can, of course, be calculated if the scalar
components V
1
, V
2
, and V
3
are known (see Ref. [3] for details). Some specic polarization
congurations of the eld are of particular interest. For example, we may express the eld
in polar coordinates such that
E(r) = E

(r) +

E

(r) + zE
z
(r), (1.156)
where
E

(r) = V
1
(r) cos + V
2
(r) sin, (1.157)
E

(r) = V
1
(r) sin + V
2
(r) cos , (1.158)
and E
z
(r) = V
3
(r). If it is then found that E

= 0 across some plane (or throughout

the entire space), the eld in called azimuthally polarized in that particular plane (or
globally). Analogously, if E

= 0, the eld is said to be radially polarized. It should

be stressed that radial or azimuthal polarization state of the angular spectrum does not
necessarily imply that the eld itself would be radially or azimuthally polarized, although
examples of elds that behave in this manner can be found (see, e.g., Ref. [9]).
1.2.6 Paraxial elds
Fields emitted by a number of practical light sources, such as many lasers, propagate in
the form of narrow beams. Thus they eectively contain only homogeneous plane waves,
which have appreciable complex amplitudes only within a narrow cone centered at the z
axis. In this situation we may use the rst two non-vanishing terms of the Taylor series
to approximate k
z
as
k
z
k
0
n
[[
2
2k
0
n
. (1.159)
This is known as the paraxial approximation of the wave vector.
Using the approximation in Eq. (1.159) allows us to obtain, from Eq. (1.102), the
general paraxial propagation formula in media of arbitrary refractive index in the form
V
j
(, z) = exp (ik
0
nz)
__

A
j
(, z
0
) exp
_
a(z) [[
2

exp (i ) d
2
(1.160)
1.2 Plane-wave representation of harmonic elds 25
with
a(z) =
iz
2k
0
n
=

n
2
+
2
z
2k
0
+ i
n
n
2
+
2
z
2k
0
. (1.161)
To put this formula in a more commonly appearing form, we us now replace with

in
Eq. (1.103) and insert the result in Eq. (1.160). The spatial-frequency integrations can
be evaluated with the identity
_

exp
_
ax
2
+ bx
_
dx =
_

a
exp
_
b
2
4a
_
, (1.162)
which is valid for complex a and b provided that a > 0. Then
__

exp
_
a(z) [[
2
+ i (

d
2
=

a(z)
exp
_

)
2
4a(z)
_
. (1.163)
Hence, if a(z) > 0, which implies > 0,
V
j
(, z) =
k
0
nexp (ik
0
nz)
i2z
__

V
j
(

, z
0
) exp
_
ik
0
n
2z
(

)
2
_
d
2

. (1.164)
It must be stressed that this Fresnel propagation formula is in principle not valid in gain
media since its derivation required the condition > 0. Thus, when dealing with gain
media, it is safer to use Eq. (1.160).
By expanding the exponential within the integral of Eq. (1.164) we obtain an alterna-
tive form of the Fresnel propagation formula:
V
j
(, z) = V
0j
k
0
nexp (ik
0
nz)
i2z
exp
_
ik
0
n
2z
[[
2
_

__

V
j
(

, z
0
) exp
_
ik
0
n
2z
[

[
2
_
exp
_

ik
0
n
z

_
d
2

. (1.165)
For suciently large values of z, the contribution of the quadratic term within the
integral becomes negligible. We may then approximate the Fresnel propagation formula
with the Fraunhofer propagation formula
V
j
(, z) = V
0j
k
0
nexp (ik
0
nz)
i2z
exp
_
ik
0
n
2z
[[
2
___

V
j
(

, z
0
) exp
_

ik
0
n
z

_
d
2

.
(1.166)
This formula is also obtained from Eq. (1.138) with k = k
o
n if we approximate
r z +
[[
2
2z
(1.167)
in the exponential factor of Eq. (1.139), r z in its denominator, and s
x
x/z,
s
y
y/z in the arguments of A
j
.
According to the Fraunhofer formula, the propagated eld in a purely dielectric medium
with real refractive index is just a scaled Fourier transform of the eld at z = z
0
. This is a
26 1. Harmonic elds
good approximation also in media with low loss, i.e., if n. Then we may approximate
n by n in the latter (quadratic) term in Eq. (1.159). As a result, we may replace n by n
in all but the rst factor of the Fresnel and Fraunhofer formulas.
The paraxial approximation, Eq. (1.159), is valid if k
x
k
z
and k
y
k
z
. Hence, in
view of Eqs. (1.107) and (1.110), the longitudinal components of the electromagnetic eld
vanish to rst approximation. According to Eqs. (1.108) and (1.109), we further have
A
4
(, z
0
) k
0
nC
0
A
2
(, z
0
) (1.168)
and
A
5
(, z
0
) k
0
nC
0
A
1
(, z
0
) (1.169)
Thus H
x
E
y
and H
y
E
x
, and the eld may be called transverse electromagnetic in
the same sense as a plane wave. However, it can be shown numerically that the paraxial
approximation can be reasonably accurate up to eld angles of the order of 30

(see
Sect. 1.3.1). In this case the longitudinal components are no longer negligible. If we
approximate k
z
k
0
n, the angular spectrum of the longitudinal electric eld takes the
form
A
3
(, z
0
) (k
0
n)
1
[k
x
A
1
(, z
0
) + k
y
A
2
(, z
0
)] . (1.170)
To the same degree of approximation, we have
A
6
(, z
0
) (k
0
n)
1
C
0
[k
y
A
1
(, z
0
) k
x
A
2
(, z
0
)] (1.171)
for the angular spectrum of the longitudinal magnetic eld.
1.3 Propagation of some model elds
We will next proceed to consider some special types of elds with substantial practical
importance. This will also allow us to consider some of the approximations made above
in more quantitative terms.
1.3.1 Gaussian elds
Let us assume that, at the plane z = z
0
, the transverse components of the electric eld
are of the Gaussian form
V
j
(, z
0
) = V
0j
exp
_

[[
2
w
2
0
_
= V
0j
exp
_

k
0
n
2z
R
[[
2
_
, (1.172)
where j = 1, 2, V
0j
and w
0
are constants, and we have introduced a parameter
z
R
=
1
2
k
0
nw
2
0
=
n

0
w
2
0
. (1.173)
If we introduce Eq. (1.172) into Eq. (1.103) and use again the integral formula (1.162), it
follows that
A
j
(, z
0
) = A
0j
exp
_

1
4
w
2
0
[[
2
_
= A
0j
exp
_

z
R
2k
0
n
[[
2
_
, (1.174)
1.3 Propagation of some model elds 27
where A
0j
= (w
2
0
/4) V
0j
= z
R
V
0j
/2k
0
n.
The far-zone behavior of the eld can be examined by inserting Eq. (1.174) into
Eq. (1.138) with k = k
0
n. After some manipulation, we then obtain the result
V
j
(r s) =
1
2
k
0

n
2

2
w
2
0
V
0j
cos
exp (k
0
r)
r
exp
_

1
4
k
2
0
_
n
2

2
_
w
2
0
sin
2

_
exp
_
i
_
k
0
nr

2
__
exp [i()] , (1.175)
where
() = arctan
_

n
_

1
2
k
2
0
nw
2
0
sin
2
(1.176)
is the wave front aberration due to absorption, i.e., the departure of the wave front from
a shperical shape that results when ,= 0. It is, however, a simple matter to show
numerically that theis wave aberration is only a small fraction of a radian over the region
where the eld amplitude is signicant, even if [[ is as large as 0.1. Thus, for all practical
purposes, we may set () = 0 in weakly absorbing or amplifying media. Using the same
approximation elsewhere in Eq. (1.175) allows us to write
V
j
(r s) =
1
2
k
0
nw
2
0
V
0j
cos
exp (k
0
r)
r
exp
_

1
4
(k
0
nw
0
)
2
sin
2

_
exp
_
i
_
k
0
nr

2
__
. (1.177)
This amounts to approximating k = k
0
n k
0
n everywhere in Eqs. (1.138), except in U(r).
This justies the comments before Eq. (1.148) for Gaussian elds in weakly absorbing or
amplifying media, and we may expect that the same applies to all well-behaved elds.
The evaluation of the integral in the exact propagation formula (1.102) is possible only
numerically. However, a simple closed-form solution is available in paraxial approxima-
tion. Inserting from Eq. (1.174) into Eq. (1.160) and dening a new parameter
q(z) = z iz
R
(1.178)
we obtain
V
j
(, z) = exp (ik
0
nz)
__

exp
_

iq(z)
2k
0
n
[[
2
_
exp (i ) d
2
. (1.179)
The integral can again be evaluated using the identity (1.162), which gives
V
j
(, z) = V
0j
z
R
iq(z)
exp
_
ik
0
n
_
z +
[[
2
2q(z)
__
, (1.180)
provided (formally) that

_
iq(z)
2k
0
n
_
> 0 (1.181)
or, more explicitly
z +
1
2
k
2
0
w
2
0
_
n
2
+
2
_
> 0. (1.182)
28 1. Harmonic elds
This condition is, of course, valid for all lossy media. In gain media with negative values
of , it implies an upper bound for propagation distances z, which however is very large
for small values of .
A paraxial expression for the eld in the far zone is obtained by inserting from
Eq. (1.172) into the Fraunhofer propagation formula. This results, after some simpli-
cation, into
V
j
(, z) = V
0j
z
R
iz
exp
_
ik
0
n
_
z +
[[
2
2z
__
exp
_

_
z
R
w
0
z
_
2
[[
2
_
. (1.183)
Precisely the same results can are obtained from Eq. (1.180). To this end, we note that
if z [z
R
[, we may write
1
q(z)

1
z
_
1 + i
z
R
z
_

1
z
. (1.184)
If we use the second approximation in the factor in front of the exponential of Eq. (1.180)
and the rst approximation inside the exponential itself, we arrive at Eq. (1.183). It is
again easy to show numerically that if [[ n, the complex z
R
can be replaced by its
real part in Eq. (1.183) without making an appreciable error.
It is worthwhile to consider the result of the Fresnel propagation formula for Gaussian
elds, Eq. (1.180), in more detail. Let us rst consider the case of purely dielectric media,
which is usually a reasonable approximation. Then, if we write
ik
0
n
2q(z)
=
1
w
2
(z)
+
ik
0
n
2R(z)
=
1
w
2
(z)
+
iz
R
w
2
0
R(z)
, (1.185)
a straightforward calculation gives
w(z) = w
0

1 +
_
z
z
R
_
2
(1.186)
and
R(z) = z +
z
2
R
z
, (1.187)
where of course z
R
= k
0
nw
2
0
/2. For large values of z we may approximate
w(z)
w
0
z
z
R
=
2z
k
0
nw
0
=

0
z
nw
0
(1.188)
and
R(z) z. (1.189)
Writing also q(z) = [q(z)[ exp [i(z)] with [q(z)[ =
_
z
2
+ z
2
R
and
(z) = arctan(z/z
R
) (1.190)
we obtain a more explicit form of Eq. (1.180), i.e.,
V
j
(, z) = V
0j
w
0
w(z)
exp [i(z)] exp
_

[[
2
w
2
(z)
_
exp
_
ik
0
n[[
2
2R(z)
_
. (1.191)
1.3 Propagation of some model elds 29
Evidently, w(z) is the characteristic (or 1/e) width of [V (, z)[ and R(z) is the radius of
curvature of the parabolic wave front. The evolution of the Gaussian beam in a dielectric
medium is illustrated in Fig. 1.8. The dashed line illustrates the asymptotic (far-zone)
limit, where the beam width and the radius of curvature both grow linearly with z as
indicated in Eqs. (1.188) and (1.189). The angle
0
is given by

0
= lim
z
w(z)
z
=

0
nw
0
. (1.192)
Furthermore, Eq. (1.188) shows that at z = z
R
the beam width has increased by a factor
of

2, and Eq. (1.189) shows that the radius of curvature at this propagation distance
reaches its minimum value R(z
R
) = 2z
R
.

y
z

0
z
0
w
0
w(z)
R(z)
Figure 1.8: Illustration of the evolution of a Gaussian eld in a homogeneous
dielectric medium.
The Gaussian elds provide us a possibility to investigate more quantitatively some
of the approximations made above. Let us rst consider the distances at which the
asymptotic approximation r is reasonable. The asymptotic formula (1.138) now
gives
V
(a)
j
(r s) = i n(w
0
/
0
) cos exp
_
( nw
0
/
0
)
2
sin
2

exp (i2 nr/

0
)
r/
0
, (1.193)
where we have introduced the spherical eld angle . The exact angular spectrum repre-
sentation for a rotationally symmetric case is obtained from Eq. (1.118) with the aid of
the Bessel-function identity
J
0
(x) =
1
2
_
2
0
exp (ixcos ) d. (1.194)
to carry out the angular integration. The general result is
V
j
(, z) = 2
_

0
A
j
(, z
0
)J
0
() exp (ik
z
z) d. (1.195)
Inserting from Eq. (1.174), we then have
V
j
(r s) =
1
2
w
2
0
_

0
exp
_

1
4
w
2
0

2
_
J
0
(r sin ) exp
_
ir cos
_
k
2
0
n
2

2
_
d. (1.196)
30 1. Harmonic elds
Some representative convergence curves curves of V
(a)
j
(r s) towards V
j
(r s), as r , are
shown in Fig. 1.9. Here we have chosen sin = 0.1, w
0
= 2, n = 1.5, and plotted the
relative amplitude error
A = 1

V
(a)
j
(r s)
V
j
(r s)

(1.197)
as well as the absolute phase error
= arg
_
V
(a)
j
(r s)
_
arg [V
j
(r s)] (1.198)
for several values of . These values were chosen small enough to allow a signicant
eld amplitude at distances up to r = 200. It is seen from the gures that uniform
convergence towards the asymptotic limit is achieved at large distances for all values of
in terms of both eld amplitude and phase.
60 80 100 120 140 160 180 200
0.00
0.01
0.02
0.03
0.04
0.05
(a)
r/
0

A
60 80 100 120 140 160 180 200
0.06
0.05
0.04
0.03
0.02
0.01
0.00
(b)
r/
0

Figure 1.9: Comparison of the exact angular spectrum and stationary-phase propa-
gation techniques. (a) Relative amplitude error and (b) phase error of the asymptotic
result as a function of r for eld angle = arcsin 0.1 if n = 1.5 and = 0 (solid
lines), = 0.001 (dashed lines) and = 0.01 (dotted lines).
The rate of convergence towards the asymptotic limit depends critically on the value
of w
0
. For small values, the asymptotic limit can be achieved (at least for most practical
purposes) at distances of the order of tens of wavelengths or less, as shown in Fig. 1.9.
To obtain more quantitative results, valid in the paraxial domain, we may consider the
evolution of the propagation parameters w(z) and R(z) in more detail. The relative beam
width error
w =
w(z) w
0
z/z
R
w(z)

1
2
_
z
R
z
_
2
(1.199)
is less than 1% if z > 7z
R
. The absolute phase error at [[ = w(z), dened as

1/e
=
_
k
0
n
2z

k
0
n
2R(z)
_
w
2
(z) =
z
R
z
, (1.200)
is less that 2/10 already at distances z > 1.6z
R
. This implies that the eld at those
distances is a good approximation of a spherical wave with a Gaussian amplitude prole,
1.3 Propagation of some model elds 31
with its origin at r = (0, 0, z
0
), in view of a commonly used wave aberration criterion that
the absolute optical path error should be less than, say, /8 over the region where the eld
amplitude diers signicantly from zero. The observations se just made on asymptotic
convergence of relative amplitude and absolute phase are in agreement with the non-
paraxial results presented in Fig. 1.9, which also shows that the absolute phase reaches a
good approximation of its asymptotic value well before the amplitude distribution does
the same.
In Sect. 1.2.6 we did not specify the range of validity of the paraxial approximation.
This is indeed dicult since it depends on the chosen criterion. Let us now have a
closer look at this problem by comparing the exact and paraxial expressions for far-zone
distributions of Gaussian elds in dielectric media, i.e., Eqs. (1.177) and (1.183) with
= 0. If we write them in slightly dierent forms
V
j
(r s) = V
0j
z
R
ir
cos exp (ik
0
nr) exp
_

_
z
R
w
0
_
2
sin
2

_
(1.201)
and
V
j
(, z) = V
0j
z
R
iz
exp
_
ik
0
n
_
z +
[[
2
2z
__
exp
_

_
z
R
w
0
_
2
[[
2
z
2
_
, (1.202)
we see immediately that Eq. (1.202) follows from Eq. (1.201) in the paraxial approximation
(1.167), which also implies cos = z/r 1 and sin = [[/r [[/z. In particular,
we wish to compare these expression at an arbitrary plane z = constant and thereby write
r
2
= [[
2
+ z
2
in Eq. (1.201) so that the exact expression in the far zone reads as
V
j
(, z) =
V
0j
z
R
z
i
_
[[
2
+ z
2
_ exp
_
ik
0
n
_
[[
2
+ z
2
_
exp
_

_
z
R
w
0
_
2
[[
2
[[
2
+ z
2
_
.(1.203)
Hence, view of the exact expressions, the far-eld amplitudes dier from Gaussian both
on the surface of any sphere r = constant and across any plane z = constant. The same
conclusion also hold before the far zone is reached, although analytic expressions are not
available there.
Figure 1.10 shows a comparison of far-zone distributions of two Gaussian elds with
wavelength-scale waist sizes in free space (n = 1) at a distance z = 25
0
, which is
large enough to ensure that the far-zone condition is valid. It can be concluded that the
dierence between the exact and paraxial results is insignicant if w
0
>
0
, i.e., if the
far-zone eld is eectively conned within an angular range < 30

. This divergence-
based esitmate for the validity of the paraxial approximation is useful also for other
well-behaves elds. However, for elds of wider divergence, the error made in using the
paraxial approximation becomes signicant at smaller elds angles, as indicated by the
results of Fig. 1.10 for w
0
=
0
/2.
The absolute phase error encountered in applying the paraxial approximation at a
plane z = constant in the direction dened by eld angle may be expressed as
(z, ) = k
0
nz
_
1 +
1
2
tan
2
cos
1

_
, (1.204)
32 1. Harmonic elds
0 5 10 15 20 25 30 35
0.00
0.02
0.04
0.06
0.08
0.10
0.12
[[/
0
[
V
1
(
[

[
,
z
)
/
V
0
1
[
0 5 10 15 20 25 30
0.0
0.5
1.0
1.5
2.0
2.5
(b)
[

)
Figure 1.10: (a) Comparison of the exact (solid lines) and paraxial (dashed lines)
far-zone eld distributions for Gaussian waves in free space. The narrower pair
corresponds to waist radius w
0
=
0
and the wider pair to w
0
=
0
/2. (b) Compar-
ison of the absolute phase error as a function of eld angle on a plane surface with
z = 10
0
(solid line) and on a spherical surface of radius r = 10
0
(dashed line).
and on a spherical surface r = constant as
(r, ) = k
0
nr
_
cos +
1
2
sin tan 1
_
. (1.205)
These expressions depend linearly on z and r, respectively, which indicates that the
phase error can become arbitrarily large in the far zone and therefore the exact non-
paraxial propagation laws should be used to determine the phase. Figure 1.10 shows
that the errors at z = 25
0
as a function of the eld angle. As discussed above, this
distance is aleady large enough for the far-zone expression to hold for highly divergent
elds. We see that at least for reasonable values of the phase error is still only a
small fraction of a radian; since a closed-form expression that gives the correct phase
is available, there is no necessary reason to use the paraxial formula at large distances
and eld angles. Moreover, we may expect that at smaller distances the paraxial Fresnel
formula for Gaussian beams (and other well-behaved elds) holds for both amplitude and
phase, provided that the divergence is less than 30

. Finally, it is easy to extablish

numerically that if w
0
> 10
0
, propagation to distances of the order of 10
6

0
can be
described by the paraxial formulation without introducing a phase error larger than 1 rad
at a eld angle satisfying sin
1/e
= w
0
/z
R
, where the eld amplitude has dropped to 1/e
of its axial value.
So far we have only considered the transverse (electric) components of the Gaussian
eld. Evidently, in general, the longitudinal components also play a role, which can be
evaluated using the techniques presented in Sections 1.2.3 and 1.2.4. The eect of these
components is included in, e.g., Eq. (1.145) for the radiant intensity. Thus, for example,
if A
2
= 0, we have
J(, )
_
cos
2
+ sin
2
cos
2

_
exp
_

1
2
(k
0
nw
0
)
2
sin
2

_
(1.206)
1.3 Propagation of some model elds 33
and the radiant intensity exhibits rotational non-symmetry, which is due to the longitu-
dinal eld components. Numerical investigation of this expression shows, however, that
the asymmetry is negligible unless w
0

0
or smaller. Such a weak polarization depen-
dence of the radiation pattern is common to elds that contain plane-wave components
propagating mainly at small angles. We will, however, see in the following subsection that
for certain types of elds the eects due to longitudinal eld components can be much
more signicant. We nally stress that these eects do not (completely) vanish in the
paraxial approximation, which in Eq. (1.206) means cos 1 and sin . Thus the
commonly held belief that paraxial approximation is equivalent with scalar modeling of
wave propagation
11
is incorrect.
1.3.2 Conical elds
Field consisting of plane waves with wave vectors conned on a single cone are called
conical elds. Both scalar [10, 11] and electromagnetic [9, 12, 13] conical elds have have
been studied extensively and shown to have many interesting properties and applications,
reviewed in Ref. [14]. We will nd such conical eld useful in a number of examples to
follow, and now proceed to investigate some of their basic properties.
Considering the angular spectrum representation in term,s of the components of the
electric angular spectrum in the orthogonal basis, we obtain the general class of conical
waves by demanding that the angular spectrum vanishes for all but one single value of
the angle , say
0
. Mathematically, we then write
A

(, ) = A

()(
0
) (1.207)
and
A

(, ) = A

()(
0
), (1.208)
thus allowing the radial and azimuthal components A

and A

to vary arbitrarily in
the azimuthal direction. According to Problem 8.1.13, the cartesian components of the
angular spectrum spectrum are then
_

_
A
1
(, )
A
2
(, )
A
3
(, )
A
4
(, )
A
5
(, )
A
6
(, )
_

_
= A

()(
0
)
_

_
cos cos
cos sin
sin
C

0
sin
C

0
cos
0
_

_
+ A

()(
0
)
_

_
sin
cos
0
C

0
cos cos
C

0
cos sin
C

0
sin
_

_
(1.209)
with C

0
= C
0
k = n
_

0
/
0
.
The expressions for the cartesian components of the electric and magnetic eld can now
be obtained by inserting Eq. (1.209) into the angular spectrum representation. Because
of the symmetry of the cone, it will prove convenient to use polar position coordinates
11
By this one usually means that consideration of a single scalar component is sucient for adequate
description of eld propagation in terms of intensity and phase; polarization eects can of course not be
described using a scalar model.
34 1. Harmonic elds
(, ) and employ Eq. (1.120). To achieve a more compact notation, we introduce two
new constants
= k sin
0
, (1.210)
= k cos
0
, (1.211)
which are in fact the radial and longitudinal components of any wave vector on the cone
dened by the angle
0
. In this way we obtain
V
j
(, , z) = exp (iz)
_
2
0
A
j
() exp [i cos( )] d, (1.212)
where the azimuthal angular spectra can be expressed in the form
_

_
A
1
()
A
2
()
A
3
()
A
4
()
A
5
()
A
6
()
_

_
= A

()
_

_
(/k) cos
(/k) sin
/k
C

0
sin
C

0
cos
0
_

_
+ A

()
_

_
sin
cos
0
C

0
(/k) cos
C

0
(/k) sin
C

0
/k
_

_
. (1.213)
The result (1.212) reveals a key property of conical elds. We see that the z-dependence
of all eld components of imaginary-exponential form, separable from the transverse de-
pendence dened by the integral. Therefore all quantities related to energy and ux (the
time-averaged electric and magnetic energy densities and the magnitude of the Poynting
vector) are independent on z. Therefore all conical elds may be called propagation-
invariant elds.
The limit of a conical wave at
0
0 is a plane wave propagating in the direction
of the z-axis,
12
which of course is the simplest example of a propagation-invariant eld.
In the highly paraxial region the longitudinal components of the electric and magnetic
vectors are seen to be negligible with (at least the largest) transverse components, and
thus the eld electromagnetic eld itself is essentially transverse.
Let us examine the nature of conical elds by looking at a certain special case in more
detail. To begin with, since the spherical components of the angular spectrum of any
conical eld are periodic with period 2, we may (quite generally) express them in the
form of Fourier series
A

() =

m=
a
m
exp (im) , (1.214)
A

() =

m=
a
m
exp (im) , (1.215)
with Fourier coecients
a
m
=
1
2
_
2
0
A

() exp (im) d, (1.216)

12
In this limit 0 so that the eld vanishes, in principle. However, this is only a matter of re-
normalization; the energy on the cone vanishes when the cone radius approaches zero, but the functional
form of the eld approaches that of a plane wave.
1.3 Propagation of some model elds 35
a
m
=
1
2
_
2
0
A

() exp (im) d. (1.217)

Let us rst choose the spherical components of the angular spectrum such that A
1
() =
exp (im) and A
2
() = 0. This amount to assuming that the electric eld is polarized
in the xz-plane, the magnetic eld is linearly polarized in y-direction, and in the highly
paraxial limit the electric eld is linearly x-polarized. Moreover, only a single Fourier
component with a xed integer value of m is non-zero in Eqs. (1.216) and (1.217). Then,
if we use the integral formula
_
2
0
exp [i (m xcos )] d = 2i
m
J
m
(x), (1.218)
where J
m
(x) is a Bessel function of the rst kind and order m, we have
V
1
(, , z) = V
0m
exp (iz) J
m
() exp (im) , (1.219)
where V
0m
= 2i
m
. Also V
3
and V
5
are non-zero, but V
3
is negligible in the highly
paraxial limit , and in this limit we also have V
5
V
1
. Hence the properties of the
eld can be studied by inspection of V
1
alone.
In view of Eq. (1.219), the eld V
1
has a helical phase prole for all values of m other
than m = 0. A eld with this property is said to have vortex of order m at the optical
axis = 0. This can be interpreted as a rotation of the electric eld about the optical
axis upon propagation, and such rotation has important mechanical eects. The electric
energy density at the highly paraxial limit is readily seen to be proportional to J
2
m
(),
i.e., to the square of the Bessel function of order m, and therefore elds of this type are
referred to as Bessel elds. Transverse proles of some lowest-order paraxial Bessel elds
are illustrated in Fig. 1.11 and some properties of the fundamental J
0
Bessel eld are
given in Table 1.1.
Table 1.1: Some properties of the ring system associated with the fundamental J
0
Bessel eld (adopted from Ref. [14]).
lobe n:o 1 2 3 4 5 q 1
outer radius/ 2.405 5.520 8.654 11.791 14.931 (q 1/4)
width/ 3.115 3.134 3.137 3.140
energy 4.897 6.187 6.252 6.268 6.274 2
It is seen from Fig. 1.11 that the fundamental Bessel eld has a central maximum
surrounded by fainter concentric rings, while the higher-order Bessel elds are hollow
in the sense that the axial eld amplitude is zero. Both types of elds are useful in
manipulation and trapping of particles. The transverse scale of the eld depends on the
36 1. Harmonic elds
x x x
x x x

y
(a)
(b)
(c)
(d)
(e)
(f)
0
0
0
0
0
0
0
0
0
0
0
0
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
Figure 1.11: Transverse eld distributions of Bessel elds of dierent order. Left:
J
0
eld. Middle: J
1
eld. Right: J
2
eld. Upper row: amplitude [V
1
(x, y)[; white
indicates the maximum of the J
0
eld, to which the other scales are normalized.
Lower row: phase; white and black indicate extreme values of and , respectively
(adopted from Ref. [14]).
parameter , i.e., on the cone angle. As indicated in Table 1.1, the radius of the central
maximum is

0

2.405

0.38

0
nsin
0
. (1.220)
Hence it can be smaller than /2 for large values of
0
, indicating that the eld is localized
in a tube of wavelength-scale diameter. However, while the functional form of V
1
is
independent on the scale, the longitudinal components becomes important for large values
of the cone angle and cannot be neglected (see Problem 8.1.14). This component has a
zero on axis and it exhibits azimuthal nonsymmetry, which causes the electric energy
density of the total eld to be rotationally non-symmetric as well.
Several other special cases of electromagnetic propagation-invariant elds have been
discussed in, e.g., Ref. [14]. For example, a purely azimuthally polarized fundamental
Bessel eld has no longitudinal component and a rotationally symmetric transverse energy
density distribution proportional to J
2
1
, but it does not have a vortex.
1.3.3 Electric dipole radiation
An electric dipole is usually an adequate model for radiation from electromagnetic point
sources, i.e., for sources with dimensions much smaller than the wavelength in a medium.
1.3 Propagation of some model elds 37
A radiating monochromatic dipole is described by a vector potential
13
a(r) = ikpU(r), (1.221)
where k = k
0
n, the constant (real-valued) vector p is known as the electric dipole moment,
and
U(r) =
exp (ikr)
r
(1.222)
is a scalar spherical wave. The magnetic and electric eld vectors can generally be derived
from the vector potential using the formulas [15]
H(r) =
1
0
a(r) (1.223)
and
E(r) =
i
kn
_

0
H(r) =
ic
kn
[a(r)] . (1.224)
Inserting from Eqs. (1.221) and (1.222) into Eqs. (1.223) and (1.224) we obtain (after
some vector algebra)
H(r) =
k
2

0
s p
_
1
1
ikr
_
U(r) (1.225)
and
E(r) =
c
n
_
k
2
( s p) s + [3 s ( s p) p]
_
1
r
2

ik
r
__
U(r), (1.226)
where s is the unit position vector as before.
The electric eld in Eq. (1.226) is seen to consist of a superposition of three terms
with dierent (inverse) powers of r, whereas the magnetic eld in Eq. (1.225) contains
only two such terms. Clearly, only the term proportional to U(r) survives in the far zone,
and we therefore see immediately that the eld there is an outgoing spherical wave with
H(r) =
k
2

0
U(r) s p (1.227)
and
E(r) =
1
n
_

0
H(r) s (1.228)
so that E H s in agreement with the general conclusions of Sect. 1.2.4. In the
near-eld, where r
0
, the term diverging as U(r)/r
2
dominates in Eq. (1.226); since
the magnetic eld does not have a term of this magnitude, the eld in the near zone
of the dipole is predominantly electric and one often speaks of the static zone. In the
intermediate zone, where r and are of the same order of magnitude, all terms must be
retained to obtain an accurate representation of the eld.
We may again use the time-averaged Poynting vector to characterize the intensity and
direction of energy ow in the the far zone. A simple calculation shows that
S(r s, t) [p[
2
sin
2

s
r
2
, (1.229)
13
A capital A is usually used to denote the vector potential in the literature but we use a lower-case
a to avoid confusion with the angular spectrum.
38 1. Harmonic elds
where is the angle between p and the observation direction s. Thus we have the familiar
1/r
2
decay of eld amplitude and we see that the dipole does not radiate in the (positive
and negative) directions of the dipole-moment vector p.
It is straightforward to show (see Problem 8.1.16) that the angular spectrum of the
diverging scalar wave U(r) is given by
A(k
x
, k
y
) =
i
2k
z
(1.230)
and the angular spectrum representation of the scalar diverging spherical wave in the
half-space z > z
0
is therefore of the form
U(r) =
i
2
__

1
k
z
exp [i (k
x
x + k
y
y + k
z
z)] dk
x
dk
y
. (1.231)
This result is known as Weyls formula.
The angular spectrum representation of electric dipole radiation into the positive half-
space z > z
0
can be readily evaluated using Weyls formula. In view of Eqs. (1.221),
(1.222), and (1.231), the vector potential clearly has the representation
a(r) =
__

k
2k
z
p exp [i (k
x
x + k
y
y + k
z
z)] dk
x
dk
y
. (1.232)
Applying Eqs. (1.225) and (1.226), we can calculate the cartesian components of the
magnetic and electric angular spectra. The results can be organized in a compact form
E(r) =
__

A
(e)
p exp [i (k
x
x + k
y
y + k
z
z)] dk
x
dk
y
(1.233)
and
H(r) =
__

A
(m)
p exp [i (k
x
x + k
y
y + k
z
z)] dk
x
dk
y
(1.234)
if we consider p as a column vector and dene matrices A
(e)
(k
x
, k
y
) and A
(m)
(k
x
, k
y
) as
A
(e)
(k
x
, k
y
) =
A
(e)
0
k
z
_
_
k
2
k
2
x
k
x
k
y
k
x
k
z
k
x
k
y
k
2
k
2
y
k
y
k
z
k
x
k
z
k
y
k
z
k
2
k
2
z
_
_
(1.235)
and
A
(m)
(k
x
, k
y
) =
A
(m)
0
k
z
_
_
0 k
z
k
y
k
z
0 k
x
k
y
k
x
0
_
_
, (1.236)
where A
(e)
0
= c/2n and A
(m)
0
= k/2
0
.
It is sometimes necessary to treat also dipole elds propagating towards the negative
half-space. This can be accomplished by replacing k
z
with k
z
in expressions (1.233)
(1.235) and writing [z[ instead of z in the exponentials of the angular spectrum rep-
resentations. In addition, a dipole positioned at an arbitrary location (z
0
, y
0
, z
0
) can be
described by replacing x by x = x x
0
and y by y = y y
0
in the exponentials of the
angular spectrum representations.
References
[1] W. K. H. Panofsky and M. Phillips, Classical Electricity and Magnetism (Addison
Wesley, Reading, MA, 1978).
[2] R. W. Boyd, Nonlinear Optics (Academic Press, Boston, 1992).
[3] M. Born and E. Wolf, Principles of Optics (Pergamon, Oxford, 1980).
[4] C. Iaconis, V. Vong, and I. A. Walmsley, Direct interferometric techniques for
characterizing ultrashort optical pulses, IEEE J. Sel. Top. Quantum Electron. 4,
285294 (1999).
[5] R. Trebino, ed., Frequency-Resolved Optical Gating: the Measurement of Ultrashort
Optical Pulses (Kluwer, 2002).
[6] I. A. Walmsley and C. Dorrer, Characterization of ultrashort electromagnetic
pulses, Adv. Opt. Photon. 1, 308437 (2009).
[7] L. Mandel and E. Wolf, Optical Coherence and Quantum Optics (Cambridge Uni-
versity Press, Cambridge, 1995).
[8] H. G. Tompkins and E. A. Irene, Eds. Handbook of Ellipsometry (William Andrew
Publ., Norwich, NY, 2005).
[9] P. P aakkonen, J. Tervo, P. Vahimaa, J. Turunen, and F. Gori, General vectorial
decomposition of electromagnetic elds with application to propagation-invariant
and rotating elds, Opt. Express 10, 949959 (2002).
[10] J. Durnin, Exact solutions for nondiracting beams. I. The scalar theory, J. Opt.
Soc. Am. A 4, 651654 (1987).
[11] J. Durnin, J. J. Miceli, and J. H. Eberly, Diraction-free beams, Phys. Rev. Lett.
58, 1499-1501 (1987)
[12] J. Turunen and A. T. Friberg, Self-imaging and propagation-invariance in electro-
magnetic elds, J. Eur. Opt. Soc. Part I: Pure Appl. Opt. 2, 5160 (1993).
39
40 REFERENCES
[13] K. Volke-Sepulveda, V. Garces-Chavez, S. Chavez-Cerda, J. Arlt, and K. Dholakia,
Orbital angular momentum of a high-order Bessel light beam, J. Opt. B: Quantum
Semiclass. Opt. 4, S82S89 (2002).
[14] J. Turunen and A. T. Friberg, Propagation-invariant optical elds, in Progress in
Optics, Vol. 54, E. Wolf, ed. (Elsevier, Amsterdam, 2009), 188.
[15] J. D. Jackson, Classical Electrodynamics (Wiley, New York, 1961).