Chapter 5

Electromagnetic waves

5.1 Electromagnetic waves in non-dispersive dielectric media

In this section we discuss some basic properties of electromagnetic

waves.1 Maxwell noticed that sourceless electromagnetic field equations 1
The radiation mechanism is discussed
possess wave solutions. This theoretical prediction was experimentally in a separate chapter.

confirmed by H. Hertz in November 1886. The most important Hertz’s

results were published in 1888.2 2
Ann. Phys. 34 610, Ann. Phys. 36 769,
Ann. Phys. 36 1

Wave equations
Macroscopic sourceless Maxwell’s equations in non-conducting continu-
ous media read

∂0 D + r ⇥ H = 0, (5.1)
r · D = 0, (5.2)
∂0 B + r ⇥ E = 0, (5.3)
r · B = 0, (5.4)

where ∂0 ⌘ 1c ∂t . They can be solved only if there are also given

constitutive relations i.e. relations between fields forming pairs (E, B)
and (D, H). In the simplest case of isotropic linear media those relations
have the form Constitutive relations in linear
1 and isotropic media
D = #E, H= B (5.5)
µ

where # = const and µ = const. Acting with the curl operator “r⇥” on
equations (5.1) and (5.3) one gets

#µ∂0 r ⇥ E + r ⇥ r ⇥ B = 0, (5.6)
| {z } | {z }
∂0 B r(r·B) r2 B

∂0 r ⇥ B + r ⇥ r ⇥ E = 0. (5.7)
| {z } | {z }
#µ∂0 E r(r·E) r2 E
120 lecture notes on classical electrodynamics

It gives

∂2 E = 0, and ∂2 B = 0 (5.8)

where ∂2 is the d’Alembert operator

#µ 2
∂2 ⌘ ∂ r2 . (5.9)
c2 t

Equations (5.8) are just wave equations. They indicate that Maxwell’s
theory may support the existence of electromagnetic waves.3 The 3
Wave equations are resultant equations
parameter (characteristic speed) and thus not all their solutions are solu-
tions of the set of Maxwell’s equations.
c In order to get the electromagnetic wave
v := p (5.10) solution one has to show that solution
#µ
of wave equation solves Maxwell’s equa-
tions.
which appears in the d’Alembert operator depends on properties of
continuous media. Note that for # = 1 = µ the continuous medium
is replaced by an empty space where v = c. The fact that the speed
of electromagnetic waves in dielectric media is lower than the speed
of light in empty space does not violate the Einstein’s postulate. The
slowing effect originates in effective description of dielectric media.
On quantum level photons are absorbed end emitted by atoms. Thus,
speed v cannot be interpreted as real speed of photons – it is rather a
sort of phenomenological parameter.

Longitudinal and transverse components

The electromagnetic potentials ( j, A) also satisfy the wave equation.
Indeed, plugging expressions

E= ∂0 A r j, B = r⇥A (5.11)

into the first pair of Maxwell’s equations (5.1) and (5.2) one gets

(#µ ∂20 r2 )A + r(#µ ∂0 j + r · A) = 0, (5.12)

2
∂0 (r · A) r j = 0. (5.13)

Furthermore, imposing the Lorenz gauge condition Lorenz gauge condition

#µ ∂0 j + r · A = 0 (5.14)

one gets that potentials j and A satisfy the wave equation

∂2 j = 0, ∂2 A = 0. (5.15)

The approach based on the electromagnetic potentials requires further

comments due to some redundancy in description based on electro-
magnetic potentials. The gauge transformations of the potentials

j0 ( x ) = j( x ) ∂0 c ( x ) and A0 ( x ) = A( x ) + rc( x )
 electromagnetic waves 121

preserve the fields E and B. It means that there exists a whole class of
electromagnetic potentials that describe the same physical situation.
We shall split the solution of the equation (5.15) into two components Decomposition on two sets of po-
tentials
( j, A) = ( j, 0) + (0, A) (5.16)
| {z } | {z }
( j1 ,A1 ) ( j2 ,A2 )

and assume that each pair of components satisfies the Lorenz condition
j 0
z}|{ z}|{
#µ ∂0 j1 +r · A1 = 0 ) ∂0 j = 0 (5.17)
#µ ∂0 j2 +r · A2 = 0 ) r·A = 0 (5.18)
|{z} |{z}
0 A

Clearly, the potentials ( j, A) being a superposition of ( j1 , A1 ) and

( j2 , A2 ) satisfy the Lorenz condition as well. We shall assume that each
pair of potentials (5.16) satisfies the wave equation. Potentials ( j1 , A1 )
give

E1 = r j, B1 = 0, (5.19)

whereas potentials ( j2 , A2 ) lead to

E2 = ∂0 A, B2 = r ⇥ A. (5.20)

According to (5.19) the electric field obeys

r ⇥ E1 = 0.

Thus, waves described by ∂2 j = 0 would be longitudinal 4 . On the 4

First, note that we do not claim that
other hand, from (5.20) and the Lorenz condition (5.18) we get such waves do exist in the nature. We
just analyse some consequences of de-
scription of the field in terms of electro-
r · E2 = 0. magnetic potentials. Second, note also
that for electric field given by expression
Hence, the electromagnetic waves being solutions of ∂2 A = 0 would E1 = nE1 (n · x, t) one gets
be transversal. For instance, taking E2 = n0 E2 (n · x, t) with n0 · n = 0, r ⇥ E1 = (êi ⇥ n)∂i E1 (n · x, t)
one gets r · E2 = 0. Thus, the potentials ( j1 , A1 ) represent longitudinal = (êi ⇥ n)ni ∂s E1 (s, t)
degrees of freedom whereas the potentials ( j2 , A2 ) represent transverse = (n ⇥ n)∂s E1 (s, t) = 0,
degrees of freedom. Note, that distinguishing on longitudinal and where s := n · x.
transverse time dependent degrees of freedom is meaningful only for
the electromagnetic potentials. The longitudinal degrees of freedom can
be eliminated by properly chosen gauge transformation. In order to
see it we consider the gauge transformation Elimination of longitudinal de-
grees of freedom
j0 = j ∂0 c, A0 = A + rc. (5.21)

which leaves unchanged the electric field E = E0 . Transformation (5.21)

can be seen as combination of two gauge transformations ( j1 , A1 ) !
( j10 , A10 ) and ( j2 , A2 ) ! ( j20 , A20 ) with c1 ⌘ c ⌘ c2 . The fields E1 and
122 lecture notes on classical electrodynamics

E2 given in terms of ( j1 , A1 ) and ( j2 , A2 ) can be replaced by E10 and

E20 expressed by new potentials ( j10 , A10 ) and ( j20 , A20 ). One gets
E E
z }|1 { z }|2 {
E = [ r j ∂0 0] + [ ∂0 A r0]
E0 E0
z }|1 { z }|2 {
= [ r( j r⇢
∂0 c ) ∂0 (0 + ⇢ c)] + [ ∂0 ( A + rc) r(0 ∂ ⇢c )]
⇢0
= r j0 ∂0 A 0 . (5.22)

The appropriate choice of c(t, x), namely An appropriate choice of the

gauge transformation
j = ∂0 c, ∂2 c = 0 (5.23)

allows to eliminate longitudinal degrees of freedom: r j0 . The condi-

tion ∂2 c = 0 assures that new potentials ( j0 , A0 ) satisfy also the Lorenz
condition. Indeed, one gets

#µ∂0 j0 + r · A0 = #µ∂0 ( j ∂0 c ) + r · ( A + r c )
= #µ∂0 j + r · A ∂2 c .
| {z } |{z}
0 0

Note that the choice j = ∂0 c is compatible with the fact that ∂2 j = 0.

The Lorenz condition (5.14) imposed on the potentials ( j0 , A0 ) reduces
to the form Coulomb gauge condition:
gauge potentials giving trans-
j0 = 0, r · A0 = 0 (5.24)
verse fields
for c given by (5.23).
It restricts wave solutions to transverse waves. It is important to stress
that the gauge fixing (5.23) exist only for some free and time-dependent
fields. For electromagnetic field which is not free the condition (5.24)
is not gauge condition anymore. Instead it is a kind of restriction
which allows to separate out the transversal (radiation) part of the
electromagnetic field. The longitudinal part represents static fields that
do not satisfy (5.24). Note that such decomposition is not invariant under
Lorentz transformations.

5.2 Plane waves

Phase velocity.
A special group of electromagnetic waves is given by waves character-
ized by constant phase surfaces. Such surfaces are solutions of condition

y(t, x) = const. (5.25)

The form of function y determines geometric character of constant

phase surfaces. For instance, any constant phase surface describing
 electromagnetic waves 123

plane wave that propagates in homogeneous dielectric media is given by

solution of equation Plane surface of the constancy of
a phase of a wavefront
y ⌘ n·x vt = const, (5.26)

where n is a constant unit vector that points out in direction of propaga-

tion of the wave. For spherical wave the vector n is replaced by a radial
spherical versor r̂ and for cylindrical wave by a radial versor in cylindrical
coordinates r̂.
The phase velocity v p of a wave is defined as velocity of translocation of
its phase surface. It can be obtained from equation dy = 0 which can be
written in the form

∂t y dt + ry · dx = 0.

Dividing by |ry| one gets

ry dx ∂t y
vp = · = . (5.27)
|ry| dt |ry|

dx ry
The phase velocity is just a projection of dt on a unit vector |ry|
, where
dx
dt is the velocity of the point x belonging to the surface of constant
ry
phase and |ry| is a vector normal to this surface. The phase velocity of
a plane wave with the surface y = const given by (5.26) reads
c
vp = v = p . (5.28)
#µ

Solution of wave equation in 1+1 dimensions

Equations ∂2 E = 0 and ∂2 B = 0 can be represented by a single equation
∂2 F = 0 where F = { E1 , E2 , E3 , B1 , B2 , B3 }. In the case of plane waves
the surface of constant phase depends on a single coordinate. Aligning
one of the axes (e.g. x) of the Cartesian reference frame in direction of
propagation one gets F as function of only two variables (t, x ). Defining
two light-cone coordinates Light-cone coordinates

x± := x ± vt (5.29)

one gets
∂x+ ∂x
∂t = ∂+ + ∂ = v ( ∂ + ∂ ), (5.30)
∂t ∂t
∂x+ ∂x
∂x = ∂+ + ∂ = ∂+ + ∂ , (5.31)
∂x ∂x
∂2 = 4∂+ ∂ . (5.32)

The equation ∂+ ∂ F( x+ , x ) = 0 has general solution which is a

sum of two arbitrary functions, each depending on a single light-cone
variable F = F+ ( x+ ) + F ( x ). In our original coordinates Superposition of two waves
124 lecture notes on classical electrodynamics

F(t, x ) = F+ ( x + vt) + F ( x vt). (5.33)

This solution describes superposition of two waves F ( x vt) and

F+ ( x + vt) that propagate forward and backward along the x-axis.
Let us observe that for spherical waves5 one can also get an explicit 5
Thy are so-called scalar spherical waves.
form of the solution. The wave equation in this case reads The spherical electromagnetic wave must
satisfy the Maxwell’s equations and not
1 2 1 ⇣ 2 ⌘ only the wave equation.
∂ F ∂ r r ∂ r F = 0. (5.34)
v2 t r2
Plugging F(t, r ) = 1r c(t, r ) into the above equation one gets

1 1 2
∂ c ∂2r c = 0, (5.35)
r v2 t

where r2 ∂r F(t, r ) = r∂r c(t, r ) c(t, r ). The function c(t, r ) satisfies

equation ∂+ ∂ c = 0, where x± := r ± vt. Thus the spherical scalar
wave reads Spherical wave for scalar field

c+ (r + vt) c (r vt)
F(t, r ) = + . (5.36)
r r
This solution contains superposition of two spherical waves: the ingoing
c+ and the outgoing c one.

Spectral decomposition
In this section we study a generic plane wave which can be decomposed
on many plane waves with different frequencies w. The frequency of
a single component enters to the solution through one of the factors
cos wt, sin wt, exp ( iwt). Here we shall not focus on polarization
aspects. This subject will be discussed in further part of the present
chapter. We assume that all vectors A (or E) assigned to different
frequencies oscillate in single common direction. Mathematically, such
superposition of waves can be represented by Fourier transform.
Any function f ( x ) of class L1 i.e.
Z •
| f ( x )|dx < • (5.37)
•

which satisfies
1
f (x) = [ f (x 0) + f ( x + 0)] (5.38)
2
at the discontinuity points can be represented by Fourier integral
Z •
1
f (x) = dk F (k)eikx . (5.39)
2p •

The expansion coefficients are given by Fourier transform of f ( x )

Z •
ikx
F (k) ⌘ F [ f ( x )](k) := dx f ( x )e . (5.40)
•
 electromagnetic waves 125

If the function f ( x ) depends on more spacetime coordinates x then

Z
1
f ( x) = d3 x F (k)eik· x , (5.41)
(2p )3 R3
Z
F (k) = d3 k f ( x ) e ik· x
. (5.42)
R3

A very convenient approach to electromagnetic field is based on

its complex-valued representation. The complex version of the elec-
tromagnetic field may be seen as a sort of auxiliary field. The physical
content of the field is encoded in its real or imaginary part. The electro-
magnetic plane wave can be represented by Fourier decomposition on
monochromatic plane waves
Z
1
A(t, x) = d3 k a(t, k)eik· x (5.43)
(2p )3 R3

where
Z
a(t, k) = d3 x A(t, x)e ik· x
. (5.44)
R3

Here A is a complex-valued vector function.6 Since the generic wave is 6

Note, the Fourier coefficients must obey
given by superposition of many waves with different frequencies then relation a⇤ (t, k) = a(t, k) for a real-
valued electromagnetic potential.
one can factorize the time dependence in expansion coefficients7 7
A factor e iw (k)t can be introduced only
iw (k )t
for complex-valued fields. For real-
a(t, k) = a(k)e , (5.45) valued fields it must be replaced by either
cos(w (k )t) or cos(w (k )t).
which depend on wave number k := |k| and wave vector k. The electro-
magnetic potential takes the form
Z
1
A(t, x) = d3 k a ( k ) e i ( k · x w (k)t)
. (5.46)
(2p )3 R3

Let us observe that each monochromatic component must be a solution

of wave equation
⇣ #µ ⌘ w2
∂2t r2 e i ( k · x w (k)t)
=0 ) #µ k2 = 0. (5.47)
c2 c2

The algebraic equality in (5.47) is called dispersion relation and it can

p
be written in terms of wave number ck = #µ w (k ). The characteristic
p
expression #µ is called refraction coefficient and it is usually denoted
by
p
n := #µ.
If the x axis is parallel to the vector k then

a(k) = a(k1 ) (2p )2 d(k2 )d(k3 ).

Consequently, Fourier integral can be written in the form

Z •
1 1x w (|k1 |)t)
A(t, x) = dk1 a(k1 )ei(k . (5.48)
2p •
126 lecture notes on classical electrodynamics

Suppose that the coefficients a(k1 ) vanish outside the interval |k1
k10 | < e, where e is certain small number. If k10 > 0 then the integral
contains only contributions from k1 > 0. In such a case k1 := k ⌘ |k|.8 8
For k10 < 0 analysis is very similar with
k1 = k.

Figure 5.1: The amplitude coefficient

| a(k1 )|.

First few terms of expansion of w (k) in neighborhood of k0 have the

form
✓ ◆
dw
w (k) = w (k0 ) + (k k0 ) +... (5.49)
dk k=k0
Then, defining w0 := w (k0 ) and Group velocity
✓ ◆
dw
vg := (5.50)
dk k=k0

one gets
Z k0 +e
1 w0 t ( k k 0 ) v g t )
A(t, x) = dk a(k)ei(kx (5.51)
2p k0 e
Z k0 +e
1 k0 )( x vg t)
= ei (k0 x w0 t )
dk a(k)ei(k . (5.52)
2p k0 e
| {z }
A0 (t,x )

The expression ei(k0 x w0 t) contains the dominant frequency term and

therefore it is the fastest oscillating function. The amplitude term
A0 (t, x ) assumes constant values on the planes x vg t = const. This
expression defines the profile (envelope) of a wave packet. The velocity
with which the envelope moves is given by (5.50). This velocity is
termed group velocity.
When frequency is a certain linear function of wave number, i.e. Linear dispersion
w = vk, then the phase factor y has the form y = kx vkt. In such a
case the group velocity is equal to the phase velocity
dw ∂t y w
vg = = v, vp = = = v. (5.53)
dk |ry| k
Let us observe that relation w = k v p , where v p = v p (k ) 6= const, leads
to
d(k v p ) dv p dv p
vg = = vp + k = vp l (5.54)
dk dk dl
2p
where wavelength is denoted by l := k . Wavelength
 electromagnetic waves 127

Monochromatic wave in homogeneous dielectrics

We consider an electromagnetic wave with a single value of angular
frequency – monochromatic wave. The wave propagates in homoge-
neous dielectric medium with constant permittivity # and constant
permeability µ. Such wave is described by auxiliary complex fields

E = E0 e i ( k · x wt)
, B = B0 ei(k· x wt)
, (5.55)

where E0 and B0 are some constant complex amplitude vectors. The

fields (5.55) are solutions of the wave equation provided that

w2
n2 = k2 . (5.56)
c2
It is enough to consider k as real-valued vector. In order to establish
what is the mutual orientation of vectors E, B and k one has to plug so-
lutions (5.55) into Maxwell’s equations and solve the resulting algebraic
equations. Considering that

r ei (k· x wt)
= ik ei(k· x wt)

one gets

r · E = i k · E, r ⇥ E = i k ⇥ E, ∂t E = iwE (5.57)

and similarly for B. Electric and magnetic Gauss laws give

r·E = 0 ) k · E0 = 0, (5.58)
r·B = 0 ) k · B0 = 0, (5.59)

whereas from Ampere-Maxwell’s law and Faraday’s law one gets

n2 w
r⇥B ∂t E = 0 ) k ⇥ B0 + n2 E0 = 0, (5.60)
c c
1 w
r ⇥ E + ∂t B = 0 ) k ⇥ E0 B0 = 0. (5.61)
c c
Equations (5.58) and (5.59) imply that vectors E and B are orthogonal
to the wave vector k. Taking into account that the wave number k = n wc
is given by dispersion relation (5.164), one gets from (5.60) and (5.61)
that

1 k
E0 = k̂ ⇥ B0 , B0 = n k̂ ⇥ E0 , k̂ ⌘ . (5.62)
n k

From scalar product of amplitudes one gets

E0 · B0 = [k̂2 E0 · B0 (k̂ · E0 )(k̂ · B0 )] = E0 · B0 (5.63)

which allows us to conclude that

E0 · B0 = 0. (5.64)
128 lecture notes on classical electrodynamics

Thus electric and magnetic field vectors are mutually perpendicular.

The amplitudes of both fields are proportional. It can be seen from
1 1
E0 · E0⇤ = 2
[k̂ ⇥ B0 ] · [k̂ ⇥ B0⇤ ] = 2 [k̂2 ( B0 · B0⇤ ) (k̂ · B0 )(k̂ · B0⇤ )]
n n
which gives

1
| E0 |2 = | B0 |2 . (5.65)
n2

Moreover, the electric and magnetic fields have the same phase. It is
clear from the fact that the amplitude of a given field is a vector product
of two vectors: the amplitude of the other field and the real-valued
vector n k̂. The square of E0 is a complex number and thus it can
be represented in the form E20 = | E0 |2 e 2ij . The choice of the phase
“ 2j” is motivated by convenience. Taking the square of the complex
magnetic field vector one gets

( B0 )2 = n2 (k̂ ⇥ E0 )2 = n2 | E0 |2 e 2ij
) B02 = n2 | E0 |2 e 2ij
.

what proves our statement. Some of the above statements9 do not hold 9
For instance, the equality of phases of
in conducting media where the vector k must be replaced by a complex electric and magnetic field.

vector.

5.3 Polarization of electromagnetic waves

Decomposition on polarization states.

We consider a free electromagnetic field in an empty space. The field
is described by solutions of sourceless Maxwell equations. In order to
eliminate non-physical degrees of freedom we impose the Coulomb
gauge condition Coulomb gauge condition
A0 = 0, r · A = 0.
We shall solve explicitly the condition r · A = 0. The vector potential A
is a solution of the wave equation (∂20 r2 ) A = 0. Since the equation
is linear then the potential can be written in the form of Fourier integral

3 Z
1
A(t, x) = Â d3 k ê a (k) a a (t, k) eik· x , (5.66)
(2p )3 a =1
where ê a (k) are some real-valued and constant vectors that satisfy the
condition of orthogonality

ê a (k) · êb (k) = dab . (5.67)

They are called polarization vectors. We take the vector ê3 (k) as The polarization vectors
k
ê3 (k) := , k 6= 0. (5.68)
|k|
 electromagnetic waves 129

The vector k in quantum field theory is proportional to linear momen-

tum of photons. Since photons with k = 0 do not contribute to total
energy and momentum of the electromagnetic field one can assume
that coefficients a a (t, k) vanish at k = 0 i.e.

a a (t, k = 0) = 0. (5.69)

Another assumption is independence of the polarization vectors on spatial

reflection of k, namely

ê a ( k) = ê a (k) for a = 1, 2. (5.70)

The complex coefficients a a (t, k) are called modes of electromagnetic field.

Since the electromagnetic field is described by real-valued function
then the modes satisfy some additional constraints. The condition Restrictions on modes
A⇤ (t, x) = A(t, x) has its explicit form

3 Z 3 Z
 d3 k ê a (k) a⇤a (t, k) e ik· x
= Â d3 k ê a (k) a a (t, k) eik· x .
a =1 a =1

Substituting variables ki ! ki in the first integral one gets

3 Z 3 Z
 d3 k ê a ( k) a⇤a (t, k) eik· x =  d3 k ê a (k) a a (t, k) eik· x
a =1 a =1

where the sign change dki ! dki has been absorbed in limits of
integration
Z • Z • Z •
dk ! ( dk) = dk.
• +• •
It gives
ê a ( k) a⇤a (t, k) = ê a (k) a a (t, k). (5.71)

Since polarization vectors labeled by a = 1, 2 do not depend on spatial

reflection of the vector k i.e. ê a (k) = ê a ( k), then (5.71) implies that

a⇤a (t, k) = a a (t, k). (5.72)

k
On the other hand, the vector (5.68) satisfies ê3 ( k) ⌘ |k|
= ê3 (k)
which gives
a3⇤ (t, k) = a3 (t, k). (5.73)

The Coulomb condition r · A = 0 takes the form

3 Z
 d3 k ik · ê a (k) a a (t, k) eik· x = 0.
a =1

This condition is satisfied providing that

k · ê a (k) a a (t, k) = 0. (5.74)

130 lecture notes on classical electrodynamics

The condition (5.74) is satisfied for k = 0. Since vectors ê a (k) are

orthonormal and ê3 µ k then for k 6= 0 the condition (5.74) constititues
constraint on a3 (t, k) which is of the form |k| a3 (t, k) = 0. It has a
solution Two polarization states of electro-
a3 (t, k) = 0. (5.75) magnetic waves
We conclude that electromagnetic potentials which are consistent with
the Coulomb gauge condition are parametrized by just two Fourier
amplitudes a1 (t, k) and a2 (t, k). It means that the electromagnetic field
has only two polarization states. Both of them are perpendicular to the
propagation vector k.

Totally polarized electromagnetic wave

In order to study polarization of electromagnetic waves we choose any
point P and look at the electric field E at this point.10 The magnetic 10
We choose x = 0 in order to avoid the
field can be discarded in this analysis because it is not independent. Its additional complex number eik· x .

orientation is determined by (5.62).

The monochromatic wave is totally polarized – spatial orientation
of constant amplitude vector E0 remains unchanged in time. In this
section we shall consider totally polarized waves. We consider complex-
valued11 electric field E = E0 ei(k· x wt) describing monochromatic elec- 11
Auxiliary field
tromagnetic wave. The amplitude vector E0 is complex-valued and
constant. The physical electric field is given by its real part Re( E). The
field E is function of time at any fixed point P.
Gauss law implies that E0 · k = 0. The expression E0 · E0 is a complex
number because the amplitude vector is complex-valued. Following
the previous section we shall parametrize this number as

E0 · E0 = | E0 |2 e 2ij
. (5.76)

The amplitude vector E0 can be parametrized in terms of two real vectors

e1 and e2 in the following way
ij
E0 = (e1 + ihe2 )e , h = ±1. (5.77)

The square of (5.77) reads E0 · E0 = (e1 2 e2 2 + 2ihe1 · e2 )e 2ij . This

expression must be equal to (5.76). It means that e1 and e2 are mutually
perpendicular real vectors which, in addition, are perpendicular to the
wave vector k,

e1 · e2 = 0, e1 · k = 0, e2 · k = 0. (5.78)

Lengths of vectors e1 and e2 , denoted by ea := |e a |, can be expressed

in terms of three parameters: two amplitudes of electric field in certain
reference frame, measured in two orthogonal directions, and a third
parameter – phase shift. Without loss of generality we can choose two
 electromagnetic waves 131

Cartesian versors x̂ and ŷ which are parallel to the plane defined by e1

and e2 . A third versor ẑ is defined as the vector product ẑ = x̂ ⇥ ŷ. The
versors x̂, ŷ, ẑ define laboratory reference frame. The amplitude of electric
field has components

E0 = A eia x̂ + B eib ŷ (5.79)

where A, B, a, b are real-valued quantities. Then

E0⇤ · E0 = A2 + B2 , (5.80)
h i
E0⇤ ⇥ E0 = AB ei( b a)
e i( b a)
ẑ = 2i AB sin(d)ẑ (5.81)

where d := b a 2 [ p, p ]. On the other hand, scalar and vector

product of the electric vector and its complex conjugate can be cast in
the form

E0⇤ · E0 = (e1 ihe2 ) · (e1 + ihe2 ) = e12 + e22 , (5.82)

E0⇤ ⇥ E0 = (e1 ihe2 ) ⇥ (e1 + ihe2 ) = 2ih e1 ⇥ e2 = 2ih e1 e2 ê1 ⇥ ê2
(5.83)

where ê1 ⇥ ê2 is a unit vector. One can always choose both versors ê1
and ê2 in a way that ê1 ⇥ ê2 = ẑ. Comparing (5.80) with (5.82) and
(5.81) with (5.83) one gets

e12 + e22 = A2 + B2 , he1 e2 = AB sin d. (5.84)

Note, that since ea 0 and A, B 0 then h = sgnd. Hence, hd 2 [0, p ].

One gets

(e1 ± e2 )2 = A2 + B2 ± 2AB sin(hd). (5.85)

Sum and difference of square roots of (5.85) gives Figure 5.2: Relations between vectors
q q
1
e1 = A2 + B2 + 2AB sin(hd) + A2 + B2 2AB sin(hd) (5.86)
2

and
q q
1
e2 = A2 + B2 + 2AB sin(hd) A2 + B2 2AB sin(hd) .
2
(5.87)
Expressions (5.86) and (5.87) give, respectively, lengths of major and
minor semi-axes of the polarization ellipse. The orientation of the
ellipse is given by the angle J between vectors x̂ and ê1

x̂ · ê1 = cos J, x̂ · ê2 = sin J (5.88)

ŷ · ê1 = sin J, ŷ · ê2 = cos J. (5.89)
132 lecture notes on classical electrodynamics

The angle J can be determined from the identity

Re[( E0 · ê1 )( E0⇤ · ê2 )] ⌘ 0 (5.90)

which follows from

ij
[e (e1 + ihe2 ) · ê1 ][eij (e1 ihe2 ) · ê2 ] = ihe1 e2 .

Plugging (5.79) into (5.90) one gets

Re[( Aeia x̂ · ê1 + Beib ŷ · ê1 )( Ae ia

x̂ · ê2 + Be ib
ŷ · ê2 )]
id id
= Re[( A cos J + Be sin J )( A sin J + Be cos J )]
2 2 2
= Re[( B A ) sin J cos J + AB(e id
cos J eid sin2 J )]
1 2
= (A B2 ) sin(2J ) + AB cos d cos(2J ) ⌘ 0. (5.91)
2
Equation (5.91) has solution

2AB
tan(2J ) = cos d. (5.92)
A2 B2

Below we study some characteristic cases.

1. Linear polarization
For hd = {0, p }, and consequently cos d = ±1, one gets
p 2AB
e1 = A2 + B2 , e2 = 0, tan(2J ) = ± .
A2 B2
The complex-valued electric field reads
ic
E = (e1 + ihe2 )e , c := wt k · x + j. (5.93)

It gives physical electric field

p
Re[ E] = A2 + B2 cos c ê1 . (5.94)

Electric field vector at any point of space is a periodic function which

oscillates with frequency w. The vector Re[ E] is fixed in direction
of ê1 . The field oscillate in the ê1 = x̂ direction for B = 0 and it
oscillates in the ê1 = ŷ direction for A = 0. In the case A = B the
electric field vector form an angle J = p/4 with x̂ providing that
d = 0 and it forms angle J = p/4 for d = p.

2. Circular polarization
p
Another interesting case is hd = 2 and A = B. In this case

e1 = A, e2 = ± B ⌘ ± A, tan(2J ) = undetermined. (5.95)

The auxiliary, complex-valued, electric field reads

E = A(ê1 ± ih ê2 )(cos c i sin c) (5.96)

 electromagnetic waves 133

whereas the physical electric field has the form

Re[ E] = A[cos c ê1 + h sin c ê2 ] (5.97)

where c := wt k · x + j. A characteristic property of circular

polarization is length preserving rotation of the vector Re[ E] at any
point of space. The rotation has positive helicity (anti-clockwise) for
d = p/2 (h = +1) and negative helicity (clockwise) for d = p/2
(h = 1).

3. Elliptical polarization
If non of cases listed above is present then the electromagnetic wave
has elliptical polarization. The electric field vector rotates and oscillates
simultaneously. Note that elliptically polarized wave Ellipticaly polarized wave as su-
perposition of two linearly polar-
Re[ E] = Re[(e1 + ihe2 )(cos c i sin c)] = e1 ê1 cos c + he2 ê2 sin c ized waves
(5.98)

can be seen as superposition of two linearly polarized waves with polariza-

tion planes being mutually perpendicular. Each linear polarization
can be decomposed on a combination of two circularly polarized
waves. With help of two vectors

ê± := ê1 cos c ± ê2 sin c, (5.99)

one gets

1 1
ê1 cos c = (ê+ + ê ), ê2 sin c = (ê+ ê ). (5.100)
2 2
It allows us to cast the formula (5.98) in the form Ellipticaly polarized wave as su-
perposition of two circularly po-
e1 + he2 e he2
Re[ E] = ê+ + 1 ê . (5.101) larized waves
2 2

One can conclude that each elliptically polarized electromagnetic

wave can be decomposed into two linearly polarized waves with
mutually orthogonal directions of polarization, or alternatively, into
two circularly polarized waves with opposite helicities.

Partially polarized electromagnetic wave

A realistic electromagnetic wave is not perfectly monochromatic. Its
frequencies belong to a narrow interval Dw around some frequency w.
A single monochromatic wave is polarized, however, superposition of
such waves with different polarizations needs a special treatment. At
fixed space point the electric field of such a wave is of the form
iwt
E = E0 ( t ) e , (5.102)
134 lecture notes on classical electrodynamics

where the amplitude E0 (t) is slow-varying function of time. The ampli-

tude vector describes polarization hence polarization of electromagnetic
wave changes slowly with time .
Experimental data of polarized wave contain measurement of the
intensity of the light beam that passes through the polarizing filter. The
intensity of light is a quadratic function of electric field. For this reason
we shall consider only quadratic functions containing components Ei (t)
and E⇤i (t), namely
j 2iwt ⇤j ⇤j
Ei E j = E0i E0 e , E⇤i E⇤ j = E0⇤i E0 e2iwt , Ei E⇤ j = E0i E0 .

Actual values of such quantities are less important than theirs time
average values 12 12
A measure process is made at a certain
time interval.
Z T
1
h f (t)i := dt f (t). (5.103)
T 0

The characteristic time scale in which amplitudes vary and the period
of functions e±2iwt are essentially different. Thus, time averaging over
2p
the interval T w , such that the phase factor oscillates many times ⌦ ↵
whereas the amplitude remains essentially unchanged, vanishes for e±iwt = 0
terms which depends on dominant D frequency
E w. On the other hand,
⌦ i ⇤j ↵ i ⇤j
the average values E E = E0 E0 do not vanish. It means that
properties of partially polarized electromagnetic wave are completely
characterized by the tensor Tensor of absolute intensity
D E
⇤j
Jij := E0i E0 . (5.104)

Since the vector E0 is perpendicular to the wave vector k then Jij has
only four components. We choose the Cartesian x1 , x2 axes perpen-
dicular to the wave vector. The indices in (5.104) run over i, j = {1, 2}.
The trace of this tensor represents intensity of the wave (density of the
energy flux) and it reads

2 D E D E D E
Tr Ĵ = Â Jii = | E01 |2 + | E02 |2 = | E0 |2 . (5.105)
i =1

This quantity is not related to polarization properties of the wave and

therefore tensor containing relative intensities is more adequate than
(5.104). A polarization tensor is defined in the following way Polarization tensor
D E
⇤j
Jij E0i E0
rij := = . (5.106)
Tr Ĵ h| E0 |2 i

It has the following properties:

1. Trr̂ = 1 , r11 + r22 = 1,

 electromagnetic waves 135

2. r̂† = r̂ , r11 , r22 2 R, ⇤ .

r21 = r12
The polarization tensor determinant reads
1 hD ED E D ED Ei
1 ⇤1 2 ⇤2 1 ⇤2 2 ⇤1
det r̂ = E0 E0 ( E0 E0 E0 E0 E0 E0 . (5.107)
h| E0 |4 i
There are two limit cases of partially polarized waves – total polar-
ization and absence of polarization.
For totally polarized electromagnetic waves the amplitude vector is
constant, E0 = const, and its time averaging give the proper vector,
hence Totally polarized wave
⇤j
E0i E0
rij = . (5.108)
| E0 |2
The polarization tensor determinant vanishes in such a case
1 h 1 ⇤1 2 ⇤2 1 ⇤2 2 ⇤1
i
det r̂ = ( E E
0 0 )( E E
0 0 ) ( E E
0 0 )( E E
0 0 ) = 0. (5.109)
| E0 |4
On the other hand, for unpolarized electromagnetic wave (e.g. natural
light beam) the average intensity has the same value in all directions. It Unpolarized wave
gives
D E D E 1D E
E01 E0⇤1 = E02 E0⇤2 = | E0 |2 . (5.110)
2
The components E01 (t) and E02 (t) are not correlated for totally unpolarized
wave. It leads to vanishing of expressions
D E D E
E01 E0⇤2 = 0 = E02 E0⇤1 . (5.111)

Plugging this results into (5.106) one gets the polarization tensor
1
rij = d . (5.112)
2 ij
Its determinant (5.109) has value det r̂ = 14 . Thus, the polarization
tensor determinant vanishes for totally polarized electromagnetic wave
and it equals to 1/4 in absence of polarization. A grade of polarization
P 2 [0, 1] is defined as follows Grade of polarization

1⇣ ⌘
det r̂ = 1 P2 , (5.113)
4
where P = 0 and P = 1 represent, respectively, absence of polarization
and maximal grade of polarization.
A convenient decomposition of polarization tensor consists on its
symmetric Sij and anti-symmetric Aij part. They are defined as follows Polarization tensor decomposi-
tion
1 1
Sij := (r + r ji ) = (rij + rij⇤ ) 2 R, (5.114)
2 ij 2
1 1 i
Aij := (rij r ji ) = (rij rij⇤ ) ⌘ # A 2 I, (5.115)
2 2 2 ij
136 lecture notes on classical electrodynamics

where A 2 R, then

i
rij = Sij # A. (5.116)
2 ij

Polarization tensor for totally polarized wave.

In this section we shall analyse what is the meaning of the polarization
tensor components in the case of totally polarized electromagnetic wave

E = E0 e i ( k · x wt)
= ( E01 x̂ + E02 ŷ)eik· x e iwt
,

where E01 = A eia and E02 = B eib . At given point P described by the
position vector x the expression eik· x is a fixed number. It is irrelevant
for the tensor rij because it is an overall phase factor and therefore it
does not contribute to d = b a. Plugging this components to (5.108)
one gets
" #
1 A2 AB e id
rij =
A2 + B2 AB eid B2
" # " #
1 A2 AB cos d i 0 1 2AB
= 2
sin d
A2 + B2 AB cos d B 2 1 0 | 2 + {z
A B2
| {z } }
A
Sij
(5.117)

The parameter A vanishes for d = {0, p }. It has been shown that

electromagnetic wave is linearly polarized for such values of phase
shift. The corresponding polarization tensor is symmetric, Linear polarization
" #
1 A2 ± AB
rij = 2 , (5.118)
A + B2 ± AB B2

and it takes the following forms

" # " # " #
1 0 0 0 1 1 ±1
rij = rij = rij = (5.119)
0 0 0 1 2 ±1 1

for polarizations aligned with, respectively, the x, y axes and the di-
agonal y = ± x direction. The circular polarization is corresponds to
d = ± p2 and A = B. It gives Circular polarization
" # " #
1 1 0 i 0 1
rij = (±) . (5.120)
2 0 1 2 1 0

The coefficient A has interpretation of degree of circular polarization.

Its extremal values A = 1 and A = +1 correspond to circularly
polarized waves with, respectively, negative and positive helicity.
 electromagnetic waves 137

Stokes parameters
Going back to our observation that the polarization tensor is a 2 ⇥ 2 Her-
mitian matrix we shall make use of the fact that it can be decomposed
on the Pauli matrices
" # " # " #
0 1 0 i 1 0
s1 = , s2 = , s3 =
1 0 i 0 0 1

and the identity matrix 1,

1
r̂ = [1 + x a sa ] . (5.121)
2
Coefficients x a are called Stokes parameters. The polarization tensor
determinant is related to the polarization degree P according to (5.113).
Hence, the equality
" #
1 2 1 1 + x 3 x 1 ix 2 1
(1 P ) = det = [1 (x 12 + x 22 + x 32 )]
4 4 x 1 + ix 2 1 x 3 4

gives
q
P= x 12 + x 22 + x 32 . (5.122)

The expression (5.122) means that all states having the same grade
of polarization form spherical surfaces in three-dimensional space of
Stokes parameters. The states with maximal polarization form the
sphere with P = 1 radius and unpolarized states correspond with the
point at the origin x 1 = x 2 = x 3 = 0.
In order to understand better what is the physical significance of
the Stokes parameters we consider all the states P = 1 and express the
parameters x a in terms of A, B and d. Using properties of the Pauli
matrixces

Tr(sa sb ) = 2dab , Tr(sa ) = 0

one gets from (5.121) the coefficients

x a = Tr(sa r̂). (5.123)

Next, plugging the polarization tensor parametrizd by A, B and d,

" #
1 A2 AB (cos d i sin d)
r̂ = 2 ,
A + B2 AB (cos d + i sin d) B2

into (5.123) one gets

2AB 2AB A2 B2
x1 = cos d, x2 = sin d x3 = .
A2 + B2 A2 + B2 A2 + B2
(5.124)
138 lecture notes on classical electrodynamics

Comparing with decomposition (5.117) and (5.121) one gets x 2 = A,

and thus it describes the grade of circular polarization. For B = 0 one gets
x 1 = x 2 = 0 and x 3 = +1 what correspond to polarization along the
x-axis. Similarly, for A = 0 the Stokes parameters read x 1 = x 2 = 0
and x 3 = 1. This case stands for the wave polarized along the y-axis.
One can conclude that the parameter x 3 describes polarization along the x
and y axes. Finally, in the case A = B and d = 0 the wave is polarized
along the line which form the angle J = p/4 with the x axis. This case
corresponds to the Stokes parameters x 2 = x 3 = 0 and x 1 = +1. For
d = 0 replaced by d = p the parameter x 3 became x 3 = 1 (polarization
along the line which form the angle J = p/4 with the x-axis).

Figure 5.3: The space of Stokes parame-

ters and the meaning of points at the sur-
face of the sphere with the radius P = 1.

It has been shown that totally polarized electromagnetic wave, P = 1,

is characterized by the polarization ellipse with semi-axes e1 and e2 that
satisfy
2AB
e12 + e22 = A2 + B2 , he1 e2 = AB sin d, tan(2J ) = cos d.
A2 B2
These partameters allows us to express the Stokes parameters in the
form

x1 2e1 e2
= tan(2J), x2 = h . (5.125)
x3 e12 + e22

Expressions (5.125) gives relations between parameters of the polarisa-

tion ellipse (its size and orientation) and the Stokes parameters. The
circular polarization component exists if none of the parameters e1 and
e2 is equal to zero. Otherwise, the ellipse degenerates to a segment of
straight line.
 electromagnetic waves 139

q
The parameter x 2 and x 12 + x 32 are invariant under Lorentz trans-
formations.

Decomposition of partially polarized wave on polarized and unpolarized

components
⌦ ↵ (n)
We split the tensor Jij = Ei E⇤ j into two parts: the component Jij
that represent unpolarized electromagnetic wave and the component
( p)
Jij corresponding to totally polarized electromagnetic wave. It has been
shown that the polarization tensor describing unpolarized wave is
proportional to the Kronecker delta. Hence
(n)
(n)
Jij 1 (n) 1 (n)
rij := = d ) Jij = J dij (5.126)
J (n) 2 ij 2

where J (n) ⌘ Tr( Ĵ (n) ). Time averaging of polarized components is

( p) i( p) ⇤ j( p) (n) ( p)
redundant Jij = E0 E0 , hence the expression Jij Jij = Jij
reads
1 (n) i( p) ⇤ j( p)
Jij J dij = E0 E0 . (5.127)
2
h i
i( p) ⇤ j( p)
The matrix E0 E0 has null determinant, see (5.109). It leads to
the equation

1 (n)
det Jij J dij = 0, (5.128)
2

where Jij = Jrij with J ⌘ Tr( Ĵ ). The equation (5.128) allows us to

determine the intensity of unpolarized component J (n)
" #
Jr11 12 J (n) Jr12
det =
Jr21 Jr22 12 J (n)
1 (n)1 (n)
= ( Jr11 J ) J 2 r12 r21
J )( Jr22
2 2
1 1 (n)
= J 2 [r11 r22 r12 r21 ] + ( J (n) )2 J J [r11 + r22 ]
| {z } 4 2 | {z }
det r̂ = 14 (1 P2 ) Trr̂ = 1

1h i
= ( J ( n ) )2 2J J (n) + (1 P2 ) J 2 = 0 (5.129)
4

which gives J (n) = (1 ± P) J. Since J (n) < J, then the physical solution
reads

J ( n ) = (1 P) J. (5.130)

The intensity of the polarized component equals to J ( p) = PJ since

( p) (n)
Jij = Jij Jij . One can easily establish the relation with the Stokes
140 lecture notes on classical electrodynamics

parameters. The only difference is that the polarized component in-

tensity is certain fraction of total intensity i.e. e12 + e22 = A2 + B2 = PJ.
Hence
x1 2e1 e2
= tan(2J), x2 = h . (5.131)
x3 PJ

Decomposition of partially polarized wave on two incoherent elliptically

polarized waves
The eigenvalues l a , a = 1, 2 of the polarisation tensor r̂ are real-valued
because the tensor is Hermitian. The eigenvectors n( a) of the polariza-
tion tensor have the form of two complex versors n⇤( a) · n( a) = 1 that
satisfy equations13 13
L.D. Landau, E.M. Lifshitz, The Classi-
cal Theory of Fields. Volume 2.
( a) ( a)
rij n j = l a ni . (5.132)

⇤( a)
Multiplying this equation by ni and taking sum over i we get

1 D i ⇤ j E ⇤( a) ( a)
⇤( a) ( a) 1 D i ⇤( a) ⇤ j ( a)
E
l a = rij ni =nj E0 E0 ni n j = ( E0 ni )( E0 n j )
J J
1 D i ⇤( a) 2 E
= | E0 ni | > 0.
J

Hence, both eigenvalues are real-valued and positive. The eigenvalues l a

can be parametrized by the polarisation degree P. Indeed, the equation

det[r̂ l1 ] = 0 , l2 Trr̂ l + det r̂ = 0 (5.133)

|{z} | {z }
1 1 (1 P2 )
4

has two solutions l1,2 = 12 (1 ± P).

The eigenvectors are mutually orthogonal. This can be shown as
follows. Multiplying the equation with a = 1 by n⇤(2) and the complex
conjugated equation with a = 2 by n(1)
8
< r n(1) = l n(1) /n⇤(2)
ij j 1 i i
⇤( 2 ) ⇤(2) (1) (5.134)
: r n
⇤ = l2 n /n
ij j i i

and subtracting the resultant equations one gets

(1) ⇤(2) (1) ⇤(2)
(rij r⇤ji ) n j ni = ( l1 l2 ) n i n i .
| {z }
0

Since l1 l2 = P 6= 0 (there is unpolarized cimponent) then n(1) ·

n⇤(2) = 0. It means that the complex eigenvectors form the orthonormal
set

n( a) · n⇤(b) = dab . (5.135)

 electromagnetic waves 141

The matrix whose columns are formed by the eigenvectors is an unitary

matrix
" # " #
1 2 † n⇤(1)
U := n ( ) n ( ) and U := .
n⇤(2)

This definition and the fact that n( a) are eigenvectors of r̂ give

" # " #
† l1 0 l1 0
U r̂ U = , r̂ = U U † (5.136)
0 l2 0 l2

The last expression in (5.136) reads

(1) ⇤(1) (2) ⇤(2)

rij = l1 ni n j + l2 n i n j . (5.137)

A complex amplitude vector can always be chosen in the way that one
of two mutually perpendicular components is real-valued whereas the
other one is purely imaginary-valued. Let e1 and e2 be two ortogonal
real vectors. We consider the following form of first eigenvector

n(1) := e1 ê1 + ihe2 ê2 . (5.138)

The normalization condition n⇤(1) · n(1) = 1 leads to the condition

e12 + e22 = 1. The second eigenvector n(2) is orthogonal to the first
one. Thus, taking n⇤(2) = a⇤ ê1 + b⇤ ê2 one gets a⇤ e1 + ihb⇤ e2 = 0. The
solution of the last condition reads a⇤ = ie2 and b⇤ = he1

n(2) := ie2 ê1 + he1 ê2 (5.139)

where we made use of the fact that n⇤(2) · n(2) = 1 = |a|2 + | b|2 .
The vectors (5.138) and (5.139) describe two identical ellipses (the
same ratio of semi-axes). The major semi-axes of these ellipses form the
angle p/2. The polarization components associated which each ellipse
are incoherent i.e. there are no cross terms in (5.137).
Figure 5.4: The polarization ellipses.

5.4 Energy and momentum flux of electromagnetic waves

The energy density u and the momentum flux density S of the electro-
magnetic field is given by expressions

1
u= [ E · D + H · B] , (5.140)
8p
c
S= [E ⇥ H ] , (5.141)
4p
where all fields are real-valued. We assume that the continuum medium
in which the wave propagates is linear, homogeneous and isotropic. It
leads to linear constitutive realtions D = #E and B = µH.
142 lecture notes on classical electrodynamics

The physical real fields are given by real (or imaginary) parts of
auxiliary complex fields E 2 C, B 2 C, hence
1 1
Re E = ( E + E ⇤ ), Re B = ( B + B ⇤ ). (5.142)
2 2
Consequently, the expressions (5.140) and (5.141) must be replaced by
the following ones

1 1
u= #( Re E)2 + ( Re B)2 , (5.143)
8p µ
c
S= [ Re E ⇥ Re B] . (5.144)
4p µ

Instantaneous values of u and S are less relevant than theirs time aver-
age values hui and hSi for fields that oscillate very quickly. Taking the
monochromatic wave in the representation of complex-valued auxiliary
fields

E = E0 e i ( k · x wt)
, B = B0 ei(k· x wt)
, (5.145)

one gets
D E 1 hD 2 E D Ei 1 1
( Re E)2 = E + 2 h E · E⇤ i + E⇤2 = E · E⇤ = E0 · E0⇤ ,
4 2 2
D E 1 hD 2 E D Ei 1 1
2 ⇤ ⇤2
( Re B) = B + 2 hB · B i + B = B · B = B0 · B0⇤ ,
⇤
4 2 2
where quickly oscillating terms drop out
D E D E D E D E
E2 ⇠ e 2iwt = 0, E⇤2 ⇠ e2iwt = 0.

Similarly, the following expression

1
h Re E ⇥ Re Bi = [h E ⇥ B⇤ i + h E⇤ ⇥ Bi + h E ⇥ Bi + h E⇤ ⇥ B⇤ i]
4 | {z } | {z }
0 0
1 1
= Re [ E ⇥ B⇤ ] = Re [ E0 ⇥ B0⇤ ] (5.146)
2 2
does not contain terms proportional to e±iwt . Thus, time averaged
values of energy density ant the Poynting vector read

1
hui = Re [ E · D ⇤ + H · B⇤ ] (5.147)
16p

c
hSi = Re [ E ⇥ H ⇤ ] . (5.148)
8p
It has been shown that Maxwell’s equations and plane wave ansatz lead
to the following algebraic equations
1 1
E= k̂ ⇥ B, B = n k̂ ⇥ E, | E |2 = | B |2 , (5.149)
n n2
 electromagnetic waves 143

k
where k̂ := |k|
. The average value of the energy density reads

1 1 1 h #µ i 1
hui = # | E |2 + | B |2 = 2
+ 1 | B |2 = | B |2 .
16p µ 16p µ n 8pµ

Similarly, the average value of the Poynting vector has the form
 ✓ ◆
c 1 1 ⇤ c
hSi = Re (k̂ ⇥ B) ⇥ B = Re(| B|2 )k̂
8p n µ 8pµ n
c
= | B|2 k̂.
8pµ n

The magnitude of hSi is proportional to hui, namely

hSi · k̂ c
= = v, (5.150)
hui n

where the proportionality coefficient is equal to the speed of propaga-

tion of electromagnetic wave.

5.5 Reflection and refraction of light at the interface between

two dielectrics

In this section we shall deal with description of electromagnetic wave on

the border of two different homogeneous dielectrics. Such dielectrics
are characterized by electric permittivities # 1 , # 2 and magnetic per-
p
meabilities µ1 and µ2 . The refractive indices read n1 := # 1 µ1 and
p
n2 := # 2 µ2 . The boundary between two spatial regions containing
different dielectric media is called an interface. One of the simplest
solutions is obtained for interface in the form of infinite plane. Let n̂
be a unit vector, normal to the interface and pointing out from first
dielectric (1) to the second dielectic (2). In absence of free charges and
free currents the fields are solutions of sourceless Maxwell’s equations

1
r · D = 0, r⇥H ∂t D = 0, (5.151)
c
1
r · B = 0, r ⇥ E + ∂t B = 0, (5.152)
c
and the boundary conditions

n̂ · ( D2 D1 ) = 0, n̂ ⇥ ( H2 H1 ) = 0, (5.153)
n̂ · ( B2 B1 ) = 0, n̂ ⇥ ( E2 E1 ) = 0. (5.154)

These conditions imply continuity of normal components Dn and Bn

and continuity of tangent components Ht and Et .
We choose the z axis perpendicular to the interface and oriented
in direction of n̂, i.e. ẑ = n̂. The incoming electromagnetic wave,

Figure 5.5: The incidence, reflection and

refraction wave vectors at the interface of
two optical media.
144 lecture notes on classical electrodynamics

characterized by the wave vector k0 , propagates in the medium (1)

We shall denote by k1 the wave vector associated with the reflected
electromagnetic wave in (1) and by k2 the wave vector describing the
transmitted electromagnetic wave in (2). For oblique incidence the
vectors k0 and n̂ are not parallel. In such a case they define a plane of
incidence. The wave vectors k0 , k1 , k2 form angles q0 , q1 and q2 with the
vector n̂. Without loss of generality we choose the x̂ versor parallel to
the plane of incidence (and perpendicular to ẑ). Third versor ŷ is given
by ŷ = ẑ ⇥ x̂.
The incident, reflected and transmitted waves are given by fields

Incident beam E0 = E00 ei(k0 ·r w0 t )

✓ ◆
n1 n1
H0 = k̂0 ⇥ E0 = k̂0 ⇥ E00 ei(k0 ·r w0 t )
,
µ1 µ1
Reflected beam E1 = E01 ei(k1 ·r w1 t )
,
✓ ◆
n n1
H1 = 1 k̂1 ⇥ E1 = 0
k̂ ⇥ E1 ei(k1 ·r w1 t )
,
µ1 µ1 1
Refracted beam E2 = E02 ei(k2 ·r w2 t )
,
✓ ◆
n2 n2 0
H2 = k̂2 ⇥ E2 = k̂2 ⇥ E2 ei(k2 ·r w2 t )
.
µ2 µ2

Angles of incidence, reflection and refraction

All waves must exist simultaneously at the boundary z = 0 which
is the xy-plane. Tangential components of the fields must be equal
on both sides of the interface. The fields satisfy boundary conditions
at the interface z = 0, namely their tangential components must be
continuous on both sides of the boundary

n̂ ⇥ [ E0 + E1 ] = n̂ ⇥ E2 , (5.155)
n̂ ⇥ [ H0 + H1 ] = n̂ ⇥ H2 . (5.156)

Plugging the complex vectors into the boundary conditions (5.155) and
(5.156) we get

n̂ ⇥ [ E00 ei(k0 ·r w0 t )
+ E01 ei(k1 ·r w1 t )
] = n̂ ⇥ E02 ei(k2 ·r w2 t )
, (5.157)
n̂ ⇥ [ H00 ei(k0 ·r w0 t )
+ H10 ei(k1 ·r w1 t )
] = n̂ ⇥ H20 ei(k2 ·r w2 t )
. (5.158)

If the wave amplitudes are constant, the only way that the conditions
(5.157) and (5.158) can be true is phase matching

k0 · r w0 t = k 1 · r w1 t = k 2 · r w2 t. (5.159)

For r = 0 the condition (5.159) gives w0 t = w1 t = w2 t which imply

equality of frequencies w0 = w1 = w2 . We shall denote this frequency The incident, relected and re-
fracted waves have the same fre-
quency w := w0 = w1 = w2
 electromagnetic waves 145

by w. Consequently, (5.159) gives

k0 · r = k1 · r = k2 · r. (5.160)

The vector r restricted to the interface z = 0 has only components x

and y, r = x x̂ + y ŷ.
Without loss of generality we can assume that the incident wave
vector has no y-component (otherwise, one can rotate Cartesian frame
to eliminate the y-component), k0y := 0 i.e. k0 · ŷ = 0. The condition
(5.160) must be true for any x and y. It leads to sequence of equalities
involving coefficients that multiply the variables x and y

k0 · x̂ = k1 · x̂ = k2 · x̂, (5.161)
0 = k1 · ŷ = k2 · ŷ. (5.162)

The second condition (5.162) gives k1y = k2y = 0. It means that the
refraction and reflection wave vectors belong to the plane of incidence. Vectors k0 , k1 and k2 belong to
In terms of angles the plane of incidence.

k a · x̂ = k a cos (q a + p/2) = k a sin q a , a = 0, 1, 2

the condition (5.161) reads

k0 sin q0 = k1 sin q1 = k2 sin q2 . (5.163)

Taking into account the dispersion relation in homogeneous media,

one gets the following sequence of equalities
w k k k
= 0 = 1 = 2. (5.164)
c n1 n1 n2

(5.164) implies k1 = k0 and k2 = nn2 k0 . The first equality in (5.163)

1
implies equality of the incidence and refraction angle Law of reflection

sin q0 = sin q1 ) q0 = q1 (5.165)

which is known as law of reflection.

The second equality in known as Snell’s law or law of refraction Law of refraction

n1 sin q0 = n2 sin q2 . (5.166)

Snell’s law determines the angle of refraction q2 = arcsin( nn12 sin q0 )

in dependence on the angle of incidence. Note that the solution q2
could not exist. This depends on the ratio of coefficients n1 /n2 . For
n2 > n1 the solution q2 exists for any value of the angle of incidence q0 .
On the other hand, for n2 < n1 it is not true. There exists the critical
value of the angle q0 ,

n2
qc = arcsin (5.167)
n1 Figure 5.6: The refraction angle q2 =
arcsin( nn1 sin q0 ) in dependence on the
2
angle of incidence for n1 < n2 and n1 >
n2 .
146 lecture notes on classical electrodynamics

r ⇣ ⌘2
n1
such that cos q2 = 1 n2 sin2 q0 is not real number for q0 > qc .
This subject is discussed in the further part. The relation between angles
of incidence q0 and refraction q2 is shown in Fig.5.6.

Conditions for amplitudes

The amplitudes are solutions of the boundary conditions (5.155) and

(5.156). Since B = n k̂ ⇥ E, or equivalently, E = n1 k̂ ⇥ B one gets that
E and H = µ1 B satisfy relation

1
H= k̂ ⇥ E, E= Z k̂ ⇥ H, (5.168)
Z
q q
µ
where Z ⌘ n =
µ 14 The field amplitudes are
# is called impedance.
14 µ0
For instance, #0 = 377 W. Moreover,
proportional E = Z H. It can be seen by taking square of equations in most of dielectrics it is sufficient to
approximate µ ⇡ µ0 .
(5.168).
The generic monochromatic incident wave can be decomposed into
two components – perpendicular and parallel to the plane of incidence

E a = E a? + E ak , Ha = Ha? + Hak , a = 0, 1, 2. (5.169)

In further sections we study incidence waves which has electric field

perpendicular to the plane of incidence (s-polarized waves) or parallel
to this plane (p-polarized waves). In the last case, the magnetic field
is perpendicular to the plane of incidence. Such plane waves with
linear polarizations are also called transverse electric (TE) and transverse
magnetic (TM).
The ratios of amplitudes of reflected (and refracted) electric field
by the amplitude of the incident field define amplitude coefficients of
reflection and refraction. These coefficients were obtained by A. J.
Fresnel in 1818. Fresnel analyzed oscillations of light in hypothetical
luminiferous ether. The result was obtained before the final formulation Figure 5.7: Augustin-Jean Fresnel, 1788–
1827.
of Maxwell’s equations. Fresnel was the first who understood that light
is a transverse wave.

Perpendicular electric field (s–polarization)

We shall consider the case of incident electric field E0 = E0 ŷ which is

perpendicular to the plane of incidence (transverse electric field TE).15 15
To simplify formulas, we shall omit the
This is also true at the interface z = 0. The amplitudes E01 and E02 can be subscript ?.

determined from boundary conditions at z = 0. Their direction is is not

known beforehand. For simplicity, we shall assume that they have the
same polarization as the incident field i.e. they are perpendicular to the
plane of incidence: E1 = E1 ŷ and E2 = E2 ŷ. The continuity conditions
 electromagnetic waves 147

(5.155) and (5.156) read

n̂ ⇥ ŷ [ E0 + E1 ] = n̂ ⇥ ŷ E2 , (5.170)
Z
n̂ ⇥ (k̂0 ⇥ ŷ) E0 + n̂ ⇥ (k̂1 ⇥ ŷ) E1 = 1 n̂ ⇥ (k̂2 ⇥ ŷ) E2 , (5.171)
| {z } | {z } Z2 | {z }
(n̂·k̂0 )ŷ (n̂·k̂1 )ŷ (n̂·k̂2 )ŷ

where n̂ ⇥ (k̂ a ⇥ ŷ) = (n̂ · ŷ)k̂ a (n̂ · k̂ a )ŷ = (n̂ · k̂ a )ŷ because n̂ · ŷ =
0. Note that n̂ · k̂1 = n̂ · k̂0 .
The conditions (5.170) and (5.171) takes the form

E1 + E2 = E0 , (5.172)
Z2 (n̂ · k̂0 ) E1 + Z1 (n̂ · k̂2 ) E2 = Z2 (n̂ · k̂0 ) E0 . (5.173)

Solving equations (5.172) and (5.173) one gets amplitude coefficients

E1 Z (n̂ · k̂0 ) Z1 (n̂ · k̂2 )

= 2 , (5.174)
E0 Z2 (n̂ · k̂0 ) + Z1 (n̂ · k̂2 )
E2 2Z2 (n̂ · k̂0 )
= . (5.175)
E0 Z2 (n̂ · k̂0 ) + Z1 (n̂ · k̂2 ) Figure 5.8: The electric field vector per-
pendicular to the plane of incidence.
The amplitude coefficients can be parametrized by angles of inci-
dence q0 and refraction q2 . The ratio of impedances reads
Z1 µ n µ sin q0
= 1 2 = 1 , (5.176)
Z2 µ2 n1 µ2 sin q2
where the last equality is true for all angles of incidence q0 providing
that n1 < n2 . Otherwise, it holds only for q0 which is smaller than
the critical angle qc . The value of this angle shall be determined in
Section 5.5. All three wave vectors are real-valued providing that q2 is
a solution. In such a case the vectors read

k̂0 = sin q0 x̂ + cos q0 ẑ,

k̂1 = sin q0 x̂ cos q0 ẑ, (a)
k̂2 = sin q2 x̂ + cos q2 ẑ.

The amplitude coefficients read

E1 Z cos q0 Z1 cos q2
= 2 , (5.177)
E0 Z2 cos q0 + Z1 cos q2
E2 2Z2 cos q0
= , (5.178)
E0 Z2 cos q0 + Z1 cos q2
or equivalently (b)

E1 µ tan q2 µ1 tan q0 µ1 = µ2 sin(q2 q0 ) Figure 5.9: The ratios of amplitudes of

= 2 = , (5.179) electric field perpendicular to the plane
E0 µ2 tan q2 + µ1 tan q0 sin(q2 + q0 ) of incidence for µ1 = µ2 ⇡ µ0 .
E2 2µ2 tan q2 µ1 = µ2 2 cos q0 sin q2
= = . (5.180)
E0 µ2 tan q2 + µ1 tan q0 sin(q2 + q0 )
148 lecture notes on classical electrodynamics

Last equalities in (5.179) and (5.180) were obtained for µ1 = µ2 ⇡ µ0 .

Each amplitude coefficent E1 /E0 and E2 /E0 is considered for two cases:
n1 < n2 and n1 > n2 . Figure.5.9 shows these ratios in dependence on
the angle of incidence for the case µ1 = µ2 .
For n1 < n2 the ratio E1 /E0 (at z = 0) is negative independently on
the value of angle of incidence. It means that the electric field vector of
the wave reflected at the interface separating the material media changes
its phase by p, eip = 1. The absolute electric field amplitude | E2 | is
thus less than the amplitude | E0 |.
Both ratios satisfy the condition E2 /E0 E1 /E0 = 1 which follows
from (5.172). For the special case of normal incidence, q0 = 0, q2 = 0, the
amplitude coefficients simplify to the form
E1 µ n µ1 n2 E2 2µ2 n1
= 2 1 , = , (5.181)
E0 µ2 n1 + µ1 n2 E0 µ2 n1 + µ1 n2
where first equality of (5.176) has been used.
For n1 > n2 the reflected wave and the incident wave have equal
phases. However, the refracted wave exists only for q0 < qc . This
phenomenon, called total internal reflection, is discussed in Section 5.5.

Parallel electric field (p–polarization)

For the case of electric field parallel to the plane of incidence (p-
polarization), the magnetic field is perpendicular to this plane (TM
waves). We assume magnetic field in the form

Ha = Ha ŷ, a = 0, 1, 2.
The electric field is given by expression E = Z k̂ ⇥ H. Plugging these
expressions into continuity the conditions (5.155) and (5.156) one gets
Z2 Figure 5.10: The magnetic field vector
n̂ ⇥ (k̂0 ⇥ ŷ) H0 + n̂ ⇥ (k̂1 ⇥ ŷ) H1 = n̂ ⇥ (k̂2 ⇥ ŷ) H2 , (5.182) perpendicular to the plane of incidence.
| {z } | {z } Z1 | {z }
(n̂·k̂0 )ŷ (n̂·k̂1 )ŷ (n̂·k̂2 )ŷ

n̂ ⇥ ŷ [ H0 + H1 ] = n̂ ⇥ ŷH2 . (5.183)

Taking into account that n̂ · k̂1 = n̂ · k̂0 and plugging the amplitude
ZH = E one gets from (5.182) and (5.183)

(n̂ · k̂0 ) E1 + (n̂ · k̂2 ) E2 = (n̂ · k̂0 ) E0 . (5.184)

Z2 E1 + Z1 E2 = Z2 E0 . (5.185)

The solution of (5.184) and (5.185) gives amplitude coefficients

E1 Z (n̂ · k̂0 ) Z2 (n̂ · k̂2 )

= 1 , (5.186)
E0 Z1 (n̂ · k̂0 ) + Z2 (n̂ · k̂2 )
E2 2Z2 (n̂ · k̂0 )
= . (5.187)
E0 Z1 (n̂ · k̂0 ) + Z2 (n̂ · k̂2 )
 electromagnetic waves 149

The solutions (5.187) and (5.186) can be cast in the form

E1 Z cos q0 Z2 cos q2
= 1 , (5.188)
E0 Z1 cos q0 + Z2 cos q2
E2 2Z2 cos q0
= . (5.189)
E0 Z1 cos q0 + Z2 cos q2

E1 µ sin(2q0 ) µ2 sin(2q2 ) µ1 = µ2 tan(q0 q2 )

= 1 = , (5.190)
E0 µ1 sin(2q0 ) + µ2 sin(2q2 ) tan(q0 + q2 )
where
h i h i
sin(2q0 ) sin(2q2 ) 2 sin 2q0 2 2q2 cos 2q0 +2 2q2 tan(q0 q2 )
= h i h i = . (a)
sin(2q0 ) + sin(2q2 ) 2 sin 2q 0 + 2q 2
cos 2q 0 2q 2 tan(q0 + q2 )
2 2

Plugging Z 1 sin q0
µ
Z2 = µ2 sin q2 into (5.188) one gets
1

The second coefficient (5.189) takes the form

E2 4µ2 cos q0 sin q2
=
E0 µ1 sin(2q0 ) + µ2 sin(2q2 )
µ1 = µ2 2 cos q0 sin q2
= .
sin(q0 + q2 ) cos(q0 q2 )
Figure 5.11 shows the amplitude coefficients for µ1 = µ2 ⇡ µ0 and (b)
Figure 5.11: The amplitude coefficients
for both cases of n1 < n2 and n1 > n2 . For the case n1 < n2 the
for electric field parallel to the plane of
solutions exist for all angles of incidence q0 and for n1 > n2 they exist incidence (TM waves) for µ1 = µ2 ⇡ µ0 .
in a limited range of that angle. There is no reflected wave for the Brew-
ster angle q B .
For n1 < n2 the ratio E10 /E00 is positive for q0 < q B and it is negative
for q0 > q B . The angle q B , called Brewster’s angle, represents a very spe-
cial value of the incidence angle which results in absence of reflected wave.
According to the formula (5.190) (the case of equal magnetic perme-
abilities) an intensity of the reflected beam vanishes (the denominator
tends to infinity) for
p
q B + q2 = .
2
Applying Snell’s law one gets
⇣p ⌘
n1 sin q B = n2 sin q2 = n2 sin q B = n2 cos q B
2
which gives expression for Brewster’s angle in terms of the ratio of
refractive indices
n2
tan q B = . (5.191)
n1
A general electromagnetic wave (for instance natural light beam) Getting polarized light
is a mixture components with different polarisations. If the incident
light beam form Brewster’s angle with normal to the interface then
reflected beam necessarily does not contain light with p-polarization.
Consequently, reflected beam is polarized – its electric field vector is
perpendicular to the plane of incidence (s-polarization). This method is
used to obtain polarized electromagnetic wave.
150 lecture notes on classical electrodynamics

Total reflection
We shall present the formulas that involve q0 and q2 in the form such
that they are sound for n1 > n2 and q0 > qc . The vectors k a belong
to the plane of incidence and thus they do not contain components in
direction of ŷ, k ay := 0. The components normal to the interface are
k an ⌘ k az whereas the tangent components are k at ⌘ k ax where

k ax = k a · x̂ = k a (k̂ a · x̂) = k a cos (p/2 q a ) = k a sin q a

with a = 0, 1, 2. According to our previous results, the components k ax

are equal

k0 sin q0 = k1 sin q1 = k2 sin q2 .

| {z } | {z } | {z }
k0t =k0x k1t =k1x k2t =k2x

The reflected and incoming normal components have opposite values,

k1n = k0n . Taking into account that k22n = k22 k22t = k22 k20t one gets

k2t = k0 sin q0 , (5.192)

" ✓ ◆2 #
n1
k22n = k22 k20 sin2 q0 = k22 1 sin2 q0 . (5.193)
n2

These formulas are valid for any angle q0 and for any value of the ratio
n1 /n2 . Plugging k2t = k2 sin q2 and k2n = k2 cos q2 into (5.192) and
(5.193) and using (5.164) one gets

n2 sin q2 = n1 sin q0 , (5.194)

✓ ◆2
2 n1
cos q2 = 1 sin2 q0 . (5.195)
n2

The expression (5.194) is just Snell’s law. Now, the point is that the
angle q2 may or may not have geometric meaning. The formula (5.195) is
used to determine this variable independently on its interpretation.
The amplitude coefficients are given by ratios of amplitude field
components. Going back to to this subject we write down electric fields
of the incident and transmitted wave

E0 = E00 ei[k0 ·r wt]

= E00 ei[(k0x x+k0z z) wt]

= E00 ei[k0 ( x sin q0 +z cos q0 ) wt]

,
E2 = E02 ei[k2 ·r wt] = E02 ei[(k2x x+k2z z) wt]
= E02 ei[(k0 x sin q0 +k2z z) wt] ,

where r = x x̂ + zẑ, k2x ⌘ k2t = k0t and k2z ⌘ k2n .

The main question is a problem of angles for n1 > n2 and q0 > qc . In
such a case the right hand side of (5.195) is negative and consequently
 electromagnetic waves 151

cos2 q2 < 0. It shows that q2 has no geometric meaning for the considered
case. Thus, we define
"✓ ◆ # "✓ ◆ #
2 2 2 n1 2 2 2 sin q0 2
s := k2n = k2 sin q0 1 = k2 1 . (5.196)
n2 sin qc

We made use of the fact 16 that for q0 ! qc the refraction angle tends 16
In the range q0  qc
to p/2. The Snell’s law gives n1 sin qc = n2 . The formula (5.196) gives
k2z = ±is where
s
✓ ◆2
n1
s = k2 sin2 q0 1. (5.197)
n2
Plugging k2z into E2 one gets

E2 = E02 e sz i [k0 x sin q0 wt]

e , (5.198)

where e+sz is not allowed because it leads to unlimited grow of ampli-

tude for z ! • (what is not consistent with energy conservation).
Penetration depth is defined as Penetration depth
1 1 l2
d := = r (5.199)
s 2p ⇣ n ⌘2
1
n2 sin2 q0 1

where k = 2p/l with l being wavelength. The formula (5.198) shows

that the refracted electromagnetic wave penetrate the second medium,
however, the penetration depth is very short – of the order of magnitude
of wavelength l2 . The refracted wave has phase y = k0 x sin q0 wt.
Hence, its phase velocity reads
✓ ◆
∂t y w c c/n2 sin qc
vp = = = = = v2 .
|ry| k0 sin q0 n1 sin q0 (n1 /n2 ) sin q0 sin q0
The expression v2 = c/n2 stands for phase velocity of plane wave in the
medium with refractive index n2 . Since the phase velocity v p is smaller
than v2 , (qc < q0 ), then we can conclude that (5.198) does not represent
electric field of plane waves. In other words, the wave is not plane. Its
amplitude and phase are not constant on planes perpendicular to the
wave vector k2 . The phase of this wave clearly depends on properties
of both media, qc = arcsin nn2 , and the angle of incidence q0 .
1
Replacing cos q2 in Fresnel formulas (5.177) and (5.188) by
s
✓ ◆2
n1 is
cos q2 = 1 sin2 q0 =
n2 k2
one gets
✓ ◆
E1 Z cos q0 iZ1 (s/k2 )
= 2 = e2ij? , (5.200)
E0 ? Z2 cos q0 + iZ1 (s/k2 )
✓ ◆
E1 Z cos q0 iZ2 (s/k2 )
= 1 = e2ijk . (5.201)
E0 k Z1 cos q0 + iZ2 (s/k2 )
152 lecture notes on classical electrodynamics

Parameters j? and jk are defined as follows

Z1 (s/k2 ) Z2 (s/k2 ) Z1 µ n
tan j? := , tan jk := , ⌘ 1 2.
Z2 cos q0 Z1 cos q0 Z2 µ2 n1
In the case of first formula we get
⇣ ⌘
Z1 (s/k2 )
1+i Z2 cos q0 1 + i tan j? cos j? + i sin j? eij?
⇣ ⌘ = = = ij? = e2ij?
1 i
Z1 (s/k2 ) 1 i tan j? cos j? i sin j? e
Z2 cos q0

and similarly for jk . Since ratios of the amplitudes have absolute values
(a)
|e2ij? | = 1 and |e2ijk | = 1 then magnitudes of reflected and incident
wavs are equal (only their phases are different).
It means that the energy flux of reflected wave is equal to energy
flux of incident wave. Consequently, there is no energy transfer to the
other medium. This is the reason why this phenomenon is called total
reflection. The existence of penetration depth does not contradict the
above result that whole wave is reflected. Experiments show that there
exists small region in the second medium such that the refracted wave
propagates parallel to the surface of interface. Finally, the wave goes
back to the first medium. Total reflection is responsible for formation (b)

of mirages. Figure 5.12: Total internal reflection; (a)

water–air, (b) glass–air.

Energy fluxes of reflected and refracted wave

The energy flux is represented by time average of Poynting vector given
by
c c
hSi = Re [ E ⇥ B⇤ ] = | E|2 k̂. (5.202)
8pµ 8p Z
The normal component of this vector is projection of hSi onto a versor
n̂ normal to the interface, hSin = hSi · n̂. Its tangent component is given
by hSi hSin n̂.
Reflection coefficient is given by the ratio Reflection coefficient
| hS1 i · n̂| | E |2
R := = 1 2, (5.203)
| hS0 i · n̂| | E0 |
where k̂0 · n̂ = k̂1 · n̂ since q1 = q0 . The reflection coefficient R repre-
sents total flux of energy reflected on the interface. This flux can be
split into two parts perpendicular and parallel to the plane of incidence.
Squares of the amplitude coefficients (5.177) and (5.188) give reflection
coefficients R? (for s-polarized component) and Rk (for p-polarized
component)
✓ ◆2 
E1 Z2 cos q0 Z1 cos q2 2
R? = = ,
E0 ? Z2 cos q0 + Z1 cos q2
✓ ◆2 
E1 Z1 cos q0 Z2 cos q2 2
Rk = = .
E0 k Z1 cos q0 + Z2 cos q2
 electromagnetic waves 153

In the case of normal incidence q0 = 0 and for µ1 = µ2 ⇡ µ0 the

reflection coefficient has the form Reflection coefficient for normal
✓ ◆ incidence
n1 n2 2
Rn = . (5.204)
n1 + n2

Transmission coefficient is defined as follows Transmission coefficient

| hS2 i · n̂| Z cos q2 | E2 |2

T := = 1 . (5.205)
| hS0 i · n̂| Z2 cos q0 | E0 |2

The perpendicular and parallel transmission coefficients are propor-

tional to squares of the amplitude coefficients (5.178) and (5.189) and
they read
✓ ◆2
Z1 cos q2 E2 4Z1 Z2 cos q0 cos q2
T? = = ,
Z2 cos q0 E0 ? ( Z2 cos q0 + Z1 cos q2 )2
✓ ◆2
Z cos q2 E2 4Z1 Z2 cos q0 cos q2
Tk = 1 = .
Z2 cos q0 E0 k ( Z1 cos q0 + Z2 cos q2 )2

For normal incidence and for µ1 = µ2 one gets

4n1 n2
Tn = . (5.206)
( n2 + n1 )2
Let us observe that R? + T? = 1 and similarly Rk + Tk = 1. The
energy conservation requires that it must hold for total coefficients

R + T = 1. (5.207)

Reflection and transmission coefficients are shown in Figure 5.13. All

coefficient belong to the interval [0, 1]. In the case n1 > n2 the reflection
coefficient is equal to unity and the transmission coefficient is equal to
zero for q0 qc . Figure 5.13: Reflection and transmission
coefficients.

5.6 Electromagnetic waves in conductors

Constitutive relations in conducting media contain Ohm’s law which

relates the current density J with the electric field E which exists inside
the material. Ohm is a local relation between these two quantities.
We shall study the case of linear and non-perfect conductors (with
dielectric and magnetic properties) characterized by

D = #E, B = µH, J = sE, r = 0, (5.208)

where s stands for electric conductivity and r for density of free electric
charges. A natural extension of a single conductor is a material con-
taining many domains made of different type of conductors separated
by interfaces. We shall consider the homogeneous and non-dispersive
154 lecture notes on classical electrodynamics

medium, # = const, µ = const. Electric and magnetic fields in such a

medium satisfy Maxwell’s equations

#µ 4p
r · E = 0, r⇥B ∂t E sµE = 0 (5.209)
c c
1
r · B = 0, r ⇥ E + ∂t B = 0 (5.210)
c
where (5.208) have been used.

Relaxation time
We shall study propagation of plane electromagnetic waves in non-
perfect conductor. Without loss of generality we can choose the ê3
Cartesian versor as being aligned with direction in which the wave
propagates. It has been shown that electromagnetic fields are transver-
sal in empty space. Now we shall look again at this problem. In order
to get a general solution we shall not assume the perpendicularity of
wave vector and the fields. First of all, we observe that for electric field
which depends on t and spatial component x3 one gets

r · E = êi · (∂i E) = ê3 · (∂3 E),

r ⇥ E = êi ⇥ (∂i E) = ê3 ⇥ (∂3 E).

Thus, electric and magnetic Gauss’ law reads

ê3 · (∂3 E) = 0, ê3 · (∂3 B) = 0, (5.211)

whereas Ampère-Maxwell law and Faraday’s law have the form

#µ 4p
ê3 ⇥ (∂3 B) ∂t E = sµE, (5.212)
c c
1
ê3 ⇥ (∂3 E) + ∂t B = 0. (5.213)
c
Taking a scalar product of ê3 and (5.212) one gets

4ps
ê3 · ∂t E = ê3 · E (5.214)
#

Taking sum of (5.214) multiplied by dt and (5.211) multiplied by dx3

one gets equation
h i 4ps
ê3 · ∂t Edt + ∂3 Edx3 = ê3 · Edt
| {z } #
dE

which can be cast in the form


dE 4ps
ê3 · + E = 0. (5.215)
dt #
 electromagnetic waves 155

In fact, solution of(5.215) represents longitudinal component of the electric

field,Ek := ê3 · E. In a similar way, we get from other two Maxwell’s
equations ê3 · dB
dt = 0. Thus, one gets

dEk 1 dBk #
+ E = 0, = 0, t := (5.216)
dt t k dt 4ps
where relaxation time is denoted by t. First of equations (5.216) has
solution
t
Ek (t, x3 ) = Ek (0, x3 )e t . (5.217)

This solution shows that there could exist time dependent longitudinal
component of the electric field in non-perfect conductors. This compo-
t
nent decreases with time in exponential manner, e t . On the other
hand, magnetic longitudinal component being solution of Maxwell’s
equations is static, Bk = const. Any electromagnetic wave has both,
electric and magnetic fields which depend on time. Thus the longitu-
dinal component of electric field cannot be related with any wave and
therefore all electromagnetic waves in conductors are transversal.

Dispersion relation in conductors

In similarity to the case of non-conductors one can derive the second-
order equations for electric and magnetic fields. Taking rotational
of Ampere’s-Maxwell equation and using remaining equations one
can get equation for magnetic field. Similarly, acting with rotational
on Faraday’s law one gets equation for electric field. The resulting
equations have the form

L E = 0, L B = 0, (5.218)

where linear differential operator L reads

#µ 2 sµ
L := ∂ + 4p 2 ∂t r2 . (5.219)
c2 t c
sµ
The first order differential operator 4p c2 ∂t plays the role of dissipative
term (like in diffusion equation). For s ! 0 the operator L is reduced
to the d’Alembert (wave) operator. We assume the following general
form of solutions describing electromagnetic plane wave in conducting
media

E = E0 e i (k ·r wt)
, B = B0 ei(k·r wt)
. (5.220)

The equation (5.218) gives condition for k. There is only a single

condition for both the electric and magnetic parts, namely Lei(k·r wt) =
0. It gives the following algebraic equation

#µ 2 4psµ
2
w i 2 w + k 2 ei(k·r wt) = 0
c c
156 lecture notes on classical electrodynamics

which must be satisfied for any t and r. It leads to dispersion relation


w2 4ps
k 2 = #µ 1+i . (5.221)
c2 #w

Note, that right hand side of (5.221) is complex. It means that k must
be some complex-valued vector, i.e. k2 = k 2 2 C. The complex number
k can be represented by its real and imaginary parts

k = k + is, k, s 2 R.

Plugging this expression into (5.221) and comparing real and imaginary
parts of both sides one gets

w2 w
k2 s2 = #µ , ks = 2psµ . (5.222)
c2 c2
Combining equations (5.222) one eliminates variables s or k
✓ ◆2
#µ w 2 2 2psµ w
k4 k = 0, (5.223)
c2 c2
✓ ◆2
4 #µ w 2 2 2psµ w
s + s = 0. (5.224)
c2 c2

Both these equations are quadratic in variables k2 , s2 and they have

equal discriminants
✓ ◆2 " ✓ ◆ #
#µw 2 4ps 2
D= 1+ . (5.225)
c2 #w

Physically sound solutions are of the form

2s 3
2 ✓ ◆2
#µ w 4ps
k2 = 4 1+ + 15 , (5.226)
2c2 #w
2s 3
2 ✓ ◆2
#µ w 4ps
s2 = 4 1+ 15 .
2c2 #w
p
Terms with D were discarded because they would give k2 < 0,
2
s < 0 and consequently the variables k and s would be imaginary.
Thus, the final form of the solution reads
2s 31
⇣ ⌘ ✓ ◆2 2
n w 4 4ps 5
k= p 1+ +1 (5.227)
2 c #w
2s 31
✓ ◆ 2
n ⇣w⌘ 4 4ps 2
s= p 1+ 15 . (5.228)
2 c #w
 electromagnetic waves 157

Relation between amplitudes

The field (5.220) where k is a solution of (5.221) solve equations (5.218).
However, the electromagnetic wave must satisfy not only the wave-like
equation but a complete set of Maxwell’s equations. Plugging the
ansatz (5.220) into Maxwell’s equations we get algebraic conditions for
amplitudes E0 and B0 . Thus, we consider an electromagnetic wave
which propagates along the axis x3 . Taking the wave vector in the form
k = (k + is)ê3 one gets
s x3 i (k x3 wt)
E = E0 e e (5.229)
s x3 i (k x3 wt)
B = B0 e e . (5.230)

Electric and magnetic fields depend on spatial coordinate x3 and time

t. Faraday’s law ê3 ⇥ ∂3 E + 1c ∂t B = 0, where ∂3 E = i (k + is) E and
∂t B = iwB results in
c
B0 = (k + is) ê3 ⇥ E0 . (5.231)
w
where complex number k is parametrized by its amplitude |k | and
phase f,
p ⇣s⌘
k = |k |eif , |k | = k 2 + s2 , f = arctan . (5.232)
k
Plugging (5.227) and (5.228) into these formulas one gets
" ✓ ◆ # 14
nw 4ps 2
|k | = 1+ , (5.233)
c #w
✓ ◆
1 4ps
f = arctan . (5.234)
2 #w

Proof. We consider expression

p 1 ✓ ◆2
s u 1 2 4ps
tan f = = p , u ⌘ 1+ . (5.235)
k u+1 #w

Then, applying some trigonometric identities we get

sin(2f) 2 sin f cos f 2 tan f

tan(2f) = = =
cos(2f) cos2 f sin2 f 1 tan2 f
hp i1 hp i1
pu 1 pu 1
2 2
s
2k u +1 u +1
= =2 p =2p p
s 2
1 k 1 puu+11 u+p1
u +1
u +1

p 1
p u 1 2
= ( u + 1) p
u+1
p 2 p 4ps
= [( u) 1]1/2 = u 1= .
#w
158 lecture notes on classical electrodynamics

Finally, we obtain relation between complex electric and magnetic field

amplitudes
" ✓ ◆ # 14
4ps 2
B0 = n 1 + eif ê3 ⇥ E0 . (5.236)
#w

The factor eif is responsible for time lag associated with relative phase Time lag
shift for fields B and E. Since ê3 · E0 = 0 = ê3 · B0 then taking square
root of a square of (5.236) one gets
" ✓ ◆ # 14
4ps 2
| B0 | = n 1 + | E0 |. (5.237)
#w

Limit cases
In this section we study the equation
#µ 2 sµ
∂ E + 4p 2 ∂t E r2 E = 0
c2 t c
looking at some characteristic limit cases. The term c2 ∂2t E which is sec-
#µ

ond order in time derivatives originates in displacement current whereas

sµ
the term 4p c2 ∂t E originates in conduction current. Bothe these terms
contribute to the dispersion relation through expressions
#µ 2 #µ 2 sµ 4ps µ
2
∂t E ! 2
w , 4p 2 ∂t E ! i 2 w.
c c c c
The ratio of absolute values of these two terms determines the phase
shift and it reads
4ps µ
c2
w 4ps
#µ 2 = = tan(2f). (5.238)
w #w
c2
4ps
• Case #w ⌧1

In this case the effect of conduction current is small comparing with

the effect of displacement current. Expanding k in powers of 4ps #w
one gets
2s 31
✓ ◆ 2
n ⇣w⌘ 4 4ps 2 5
k =p 1+ +1
2 c #w
" ✓ ◆ #1
n ⇣w⌘ 1 4ps 2
2
=p 1+ +...+1
2 c 2 #w
" ✓ ◆ #1
1 4ps 2
2
w
=n 1+ +...
c 4 #w
" ✓ ◆ #
w 1 2ps 2
=n 1+ +... (5.239)
c 2 #w
 electromagnetic waves 159

and similarly for s,

2s 31
✓ ◆ 2
n ⇣w⌘ 4 4ps 2
s =p 1+ 15
2 c #w
" ✓ ◆ #1
1 4ps 2
2
n w
=p 1+ + . . . + ( 1)
2 c 2 #w
r
#µ w 1 4ps
= p +...
2 c 2 #w
r
2ps µ
= +..., (5.240)
c #
q
µ
where Z := # . Notice that the imaginary part s = Im (k ) does not
depend on the frequency w. For s ⇡ 0 the phase shift is very small,
f ⇡ 0. Consequently, the fields E and B have approximately the
same phase.
4ps
• Case #w 1

In this limit the conduction current dominates. This is the case of

metals where s/# ⇡ 1018 . The conduction current dominates for
frequencies lower than 1017 Hz (microwaves, radio-frequency, light,
some range of X-ray). It means that the real and imaginary parts are
almost equal,
1p
k⇡s⇡ 2ps µ w.
c
It gives
p 1p
|k | = k 2 + s2 ⇡ 4ps µ w (5.241)
⇣s⌘ c p
f = arctan ⇡ arctan(1) = . (5.242)
k 4
The relative phase shift of fields takes value f = p/4 whereas abso-
lute values of their amplitudes satisfy the proportionality relation
r
4ps
| B0 | ⇡ n | E0 |, (5.243)
#w
q
where n takes values of the order of unity. Since 4ps #w 1 then
| B0 | | E0 |.

Distribution of electric current in conductors

The term ∂t D, that represents the displacement term, is irrelevant

for good conductors. In such a case the equation
4psµ
r2 E ∂t E = 0
c2
160 lecture notes on classical electrodynamics

is equivalent to the following one

4psµ
r2 J ∂t J = 0, (5.244)
c2
where J = sE. (5.244) is a diffusion equation. Since the electric field
depends on time through e iwt , then the current J one can expect
that the current density is proportional to this exponential function,
namely J = J0 (r )e iwt . Plugging this expression into (5.244) we
q
get17 17
In SI units d = 2
sµw .

4psµw
r2 J0 + t 2 J0 = 0, t 2 := i . (5.245)
c2
Considering that
p 1 4ps
i = p (1 + i ) and 1
2 #w

we get

1+i c 1
t= , d= p ⇡ .
d 2psµ w s

We shall look at the example of a conductor which occupies the

region x1 0. The border x1 ⌘ x = 0 is the interface that separates
conducting material and empty space. The electromagnetic wave
propagates along the axis x1 ⌘ x. This case corresponds with normal
incidence of the wave. We assume that the wave is linearly polarized
in direction of the x3 axes and then J0 (r ) = J03 ( x )ê3 . The resulting
equation

d2
+ t 2 J03 ( x ) = 0 (5.246)
d ( x )2

has the solution

x x
J03 ( x ) = J03 (0)eit x = J03 (0)e d ei d . (5.247)

We have to reject the second solution

it x x i xd
e = ed e

because it leads to non-physical behaviour in the conductor x

0. The electric current density vanishes exponentially inside the
conductor. The characteristic length d is called skin depth.
In the case of silver, s 1 = 1.58 ⇤ 10 8 Wm, and wave frequency
1
n = 4 ⇤ 109 Hz which skin depth has value d = ppµsn = 10 6 m
whereas for copper, s 1 = 1.68 ⇤ 10 8 Wm, and UV light with n =
1015 Hz it reads d = 10 9 m.
 electromagnetic waves 161

5.7 Electromagnetic waves in dispersive dielectric media

Dispersive media
The dependence of dielectric permittivity # and magnetic permeability
µ on the electromagnetic wave frequency w is called dispersion. Many
dielectric media, like for instance water, are characterized by permit-
tivity # which is constant for slowly varying electromagnetic field and
varies (differently for different substances) for higher electromagnetic
field frequencies.
The study of high-frequency electromagnetic field in polarizable me-
dia is a very interesting subject. The electromagnetic field which is
periodic in time is also periodic in space. Spatial oscillations are char-
acterized by wavelength l ⇠ wc , where w is the frequency of the field.
The wavelength is getting shorter as frequency w is increasing. For suf-
ficiently high frequencies the corresponding wavelength is comparable
with typical size of atoms represented by atomic scale.18 The macro- 18
A characteristic magnitude of a is 1
scopic description of the electromagnetic field is meaningless for l ⇠ a. Angstrom equals to 10 10 m.

Thus, the macroscopic regime for electromagnetic waves is charcterized

by wavelengths l a. For this reason we are interested in the range
of frequencies for which
• there is a meaningful macroscopic description,

• there are present some new effects caused by dispersion.

The existence of dispersion for most of materials is expected within the
scenario of electronic mechanism. This mechanism is the most rapid
manner of establishment of the electric or magnetic polarization in
matter. For v being a typical velocity of electrons in atoms, the ratio
a/v is a characteristic time which is of the order of relaxation time.
The relaxation time is of the order of characteristic time v ⌧ c. The
wavelength corresponding to these times l ⇠ ac/v is much larger than a
because v ⌧ c. In what follows we shall consider electromagnetic waves
with wavelength l a. This condition, applicable dielectrics, may not
be sufficient in generality. For instance, meta ls in low temperatures
possess region of frequencies in which the macroscopic theory has not
application, although the condition c/w a is satisfied.
Macroscopic Maxwell’s equations in material media have formally
the same form as in empty space
1
r · D = 0, r⇥H = ∂t D,
c
and
1
r · B = 0, r⇥E = ∂t B.
c
Differential equations in such a form have no solution until they are
complemented by relations between fields E, D and B, H i.e. constitutive
162 lecture notes on classical electrodynamics

relations. For static and quase-static fields in homogeneous and isotropic

media these relations are just proportionality relations D = #E, B = µH,
where parameters e and µ are material constants. The constitutive
relations in such a case are local in time. In other words, fields D at
instant of time t is determined bt E taken at the same instant of time.
The same is true for the pair B, H.
Temporal locality of equations is a consequence of the fact that the
electric and magnetic dipoles have enough time for changing their
orientation responding for slowly varying electromagnetic field. The
situation is different for quickly varying electromagnetic field. Since
dipoles cannot immediately change their orientation then there exists
certain time delay in response of the material. In such a case, the electric
permittivity and magnetic permeability are functions of the frequency
w. The dependence on w is different for functions describing electric
and magnetic properties of matter.

Dielectrics
The polarization vector P is defined taking into account neutrality of
R
dielectrics, V d3 x hrb i = 0, where V is the region occupied by dielectric
body and hrb i is volume density of bounded charges. Electrical neu-
trality is assured by the assumption that the polarization divergence is
proportional to the electric charge density, hrb i = r · P where P ⌘ 0.
The electric dipole moment of the body is given by expression
Z Z
3
p= d x r hrb i = d3 xP. (5.248)
V V

It means that the polarisation vector P has interpretation of electric dipole

moment density. The derivation and interpretation of the polarization
1
vector holds for time-varying fields. Thus the vector P = 4p ( D E)
represents the electric polarization independently on dispersion in the
medium.
Fields which oscillates with high frequencies have usually small
amplitudes. For fields which are not too strong the fields D (t, x) and
E(t, x) are related by linear transformation. The most general such
transformation is given by integral
Z t
D (t, x) = E(t, x) + dt0 f (t t 0 ) E ( t 0 , x ),
•

where the function f (t) is determined by properties of the medium. The

upper limit t0 = t of the integral represents causal cut-off. For further
convenience the term E(t, x) has been separated from the integral
expression. 19 Changing the variable of integration 19
Note that absence of the integral is
physically interpret as absence of dielec-
Z t Z 0
tric medium. The adopted way of rep-
t := t t0 , dt0 ! ( dt ), resenting the electric displacement field
• •
allows us to remove the Dirac delta from
the function f (t).
 electromagnetic waves 163

one gets
Z •
D (t, x) = E(t, x) + dt f (t ) E(t t, x). (5.249)
0

One can put the constitutive relation for dispersive media (5.249) in the
similar form to the relation D = #E, namely

D=b
#E (5.250)

where b# is some linear integral operator. The presented formalism has

also application in metals.
In what follows we shall apply the method of Fourier decomposition
of time-dependent field. For a single frequency field
iwt e
E(t, x) = e E(w, x)

the expression (5.249) takes the form

✓Z • ◆
iwt iwt e
D (t, x) = E(t, x) + dt f (t )e e E(w, x) = #(w ) E(t, x)
0 | {z }
E(t,x)
(5.251)
where
Z •
#(w ) = 1 + dt f (t )eiwt . (5.252)
0

Hence, for dingle frequency fields the integral operator b

# takes the form
of proportionality coefficient between fields D and E. This coefficient
depends on properties of the material medium and the field frequency.
The dispersion law #(w ) is a complex-valued function and thus it can be
represented in the form

#(w ) = #0 (w ) + i#00 (w ). (5.253)

where #0 (w ), #00 (w ) 2 R are, respectively, its real and imaginary parts.

The expression (5.252) implies the relation

#( w ) = #⇤ (w ) (5.254)

which gives
#0 ( w ) + i#00 ( w ) = #0 (w ) i#00 (w ).
It follows from the above expression that

# 0 ( w ) = # 0 ( w ), #00 ( w ) = #00 (w ), (5.255)

i.e. the real part of e(w ) is even function of w whereas the imaginary
part is odd. Expanding the real and imaginary part of the function #(w )
in Taylor series at w = 0 one gets that
• •
#0 (w ) = # 0 + Â # 2n w2n , #00 (w ) = # 1 w + Â # 2n+1 w2n+1 ,
n =1 n =1
164 lecture notes on classical electrodynamics

i.e. they contain either even or odd powers of w.

Only the lowest order expansion terms are significant in the limit
w ! 0. Thus, the first term of approximation for the real part # 0 20 20
Note that # 0 is not an electric constant
corresponds with non-dispersive dielectrics. For the imaginary part the in vacuum because in Gaussian units this
constant is equal unit.
lower order term id linear in frequency w.
The function e(w ) is physically sound for metals in the low range
of frequencies. Ampere’s law in dielectric media (without free current)
with this law in non perfect conductors have the form
✓ ◆
4p 1 4p
r⇥H = ∂t D , r⇥H = (sE) (5.256)
c 4p c | {z }
| {z } Jc
Jd

where Jd and Jc are, respectively, displacement and conduction currents.

Thus the expression (5.249) can be formally applied for metals with
substitution ∂t D ! 4psE. It means that
⇣ ⌘
∂t D (t, x) = ∂t #(w )e iwt E
e(w, x) = iw#(w ) E(t, x)
| {z }
4ps

and thus the function w (#) is of the form

4ps
#(w ) = i . (5.257)
w
Hence, the function #(w ) for conductors contains the singular imaginary
term, proportional to w 1 . The subsequent expansion term is a real
constant, however, in the case of metals this constant has no similar
interpretation as # 0 in dielectrics.

The Lorentz model

The Lorentz model is a classical model of polarizability. Typical veloci-
ties v of electrons in atoms are small when comparing with the speed
of light c. It mens that dislocations of electrons in atoms are of order
v/w and thus they are much smaller than the electromagnetic wave-
length c/w. Consequently, the electric field experienced by bounded
electron is approximately uniform in space and can be approximated by
E(t) = E0 e iwt at some point r0 in vicinity of the electron.
We shall consider a classical model in which the electron-nucleus
interaction is modeled by harmonic force F = kx, where x denotes
dislocation of the electron from the position of the equilibrium. By
assumption the electric field is uniform in whole region where the
electron moves. The loses of energy are modeled by the force Fd =
h dx
dt . Thus, the dynamics of the electron is governed by classical
equation of motion

d2 x dx
m = kx h eE(t), (5.258)
dt2 dt
 electromagnetic waves 165

or equivalently

d2 x dx e
+ g + w02 x = E0 e iwt
. (5.259)
dt2 dt m
We shall consider a complex-valued form of the solution, namely x(t) =
x0 e iwt . It gives21 21
This is so-called particular solution of
non homogeneous differential equation.
e E0
x0 = . (5.260)
m (w02 w2 ) igw

Note that electric field E(t) which depend on multiple frequencies

1
R
can be represented by the Fourier integral E(t) = 2p dwe iwt E
e(w ),
e
where E(w ) plays the role of E0 . In such a case the solution x(t) is
1
R
assumed to have the form of Fourier integral x(t) = 2p dwe iwt xe(w ).
The coefficients xe(w ) are of the form

e Ẽ(w )
x̃(w ) = .
m (w02 w 2 ) igw

The electric dipole moment of a single electron reads

p(t) = e x ( t ).

If the material has N such electrons per unit volume than the polariza-
tion vector reads P (t) = N p(t) and thus

Ne2 E(t)
P(t) = = c e E ( t ). (5.261)
m (w02 w2 ) igw

More realistic models assume that different electrons interact dif-

ferently with the electric field. For instance, a dielectric material with
Nm molecules in the unit of volume and Z electrons in each molecule
(N = Nm Z) possesses f j electrons (of total Z electrons) that are charac-
terized by frequencies w j and damping constants g j . Thus the electric
susceptibility is generalized to the form

Nm e2 fj
ce =
m  (w 2 w2 ) ig j
, Â f j = Z. (5.262)
j j j

The function #(w ) reads

w 2p 4pNe2
#(w ) = 1 + , w 2p = . (5.263)
(w02 w2 ) igw m

wgere w p is plasma frequency. The real and imaginary parts of #(w ) have
the form
w 2p (w02 w2 ) w 2p w
#(w ) = 1 + +i . (5.264)
(w02 w 2 ) + g2 w 2 (w 2 w 2 ) + g2 w 2
| {z } | 0 {z }
#0 (w ) #00 (w )
166 lecture notes on classical electrodynamics

In the limit of high frequencies w the function #(w ) is approximated

by expression
w 2p
#(w ) = 1 (5.265)
w2
which tends to unity for w ! •. This approximation is valid for light
atoms in deep UV region and for heavier ones in the region of X-ray
frequencies.

Dispersion of magnetic permeability

In contrast to #(w ), the magnetic permeability µ(w ) loses its physical
meaning for high frequencies. The expression

1
M= (B H)
4p
can be interpreted as magnetic moment density providing that ∂t P can
be neglected. This can be seen from magnetic moment definition
Z
1
m= d3 x r ⇥ h jb i (5.266)
2c V

Ampere’s law in absence of macroscopic conduction currents can be

put in two different ways, namely

1 4p 1
r⇥B = ∂t E + h jb i , r⇥H = ∂t D.
c c c
Subtraction of these equations gives

1 4p
r ⇥ ( B H ) = ∂t ( E D ) + h jb i .
| {z } c | {z } c
4pM 4pP

Thus the current density of bounded charges reads

h jb i = cr ⇥ M + ∂t P. (5.267)

Clearly, this current depends on time derivative of a polarization vector.

Plugging (5.267) into (5.266) and using the fact that M = 0 in empty
space we get
Z Z
1
m= d3 x M + d3 x r ⇥ ∂t P.
V 2c V
This expression shows that M has interpretation of magnetic moment
density only if the term ∂t P is very small comparing with the other
term
∂t P (t) ⌧ cr ⇥ M. (5.268)
For instance, we can take a small body with characteristic size l (e.g.
a sphere of radius l) in oscillating and approximately uniform magnetic
field B ⇡ B0 e iwt .
 electromagnetic waves 167

Integration of (5.268) over small disc localized inside the body and
having radius approximately equal to l gives
Z
da · ∂t P (t) ⇠ l 2 ∂t P(t)
S

and Z I
da · (r ⇥ M ) = dl · M ⇠ lM (t) = lcm H (t).
S C

Thus the condition (5.268) results in the following one

cm c
∂t P(t) ⌧ H ( t ). (5.269)
l
1 # 1
The polarization vector is given by P = 4p ( D E) = 4p E. Taking
# 1 ⇠ 1 (e.g. for teflon # = 2.1) we get

∂ t P ( t ) ⇠ ∂ t E ( t ).

The electric field magnitude can be estimated from Faraday’s law

Z Z
1
dl · E = da · B.
C c S

Choosing the integration region S as a disc of radius l we get integrals

Z Z
1 d l2 l2
dl · E ⇠ lE(t), da · B ⇠ ∂t B(t) ⇡ ∂t H (t)
C c dt S c c
where the approximation of the field B by H follows from the fact that
cm is usually vary small for paramagnetics and diamagnetics. Thus we
have from Faraday’s laws

l
E(t) ⇠ ∂ t H ( t ).
c

Since H ⇡ B0 e iwt , then

l 2 lw 2
∂t P(t) ⇠ ∂t H (t) = H ( t ).
c c
Plugging this expression into (5.268) we get

c m c2
l2 ⌧ . (5.270)
w2

However, l cannot be arbitrary small. The macroscopic character of the

body is given by the condition l a where a is the atomic size. The
2
condition (5.270) does not hold for optical frequencies because cm ⇠ vc2
and w ⇠ va . Thus the right hand side of (5.270) is proportional to a2 .
Consequently, the concept of magnetic permeability is meaningless for
optical frequencies. In such a case µ = 1.
168 lecture notes on classical electrodynamics

The energy of the electromagnetic field in dispersive media

The electromagnetic energy flux is given by Poynting vector
c
S= E ⇥ H. (5.271)
4p
This formula holds for any time-varying electromagnetic field including
the case of dispersive media. The rate of change of energy density ∂t u
is given by divergence of Poynting vector r · S, namely
c
∂t u = r·S = e ∂ (Ej Hk )
4p ⇣ijk i ⌘
c
= eijk ∂i E j H k e jik E j ∂i H k
4p
c
= ( H · (r ⇥ E) E · (r ⇥ H ))
4p | {z } | {z }
1∂ B 1∂ D
c t c t

1
= ( E · ∂ t D + H · ∂ t B ). (5.272)
4p
In non-dispersive dielectric media the constitutive relations D = #E
and B = #H allows us to put energy density in the form

1
u= (#E2 + µH 2 ). (5.273)
8p
Thermodynamic interpretation of energy density (5.273) is the following
one: it is equal to the difference of total energy inside unit volume
with and without electromagnetic field. Such interpretation of the
electromagnetic energy is not possible in presence of dispersion due to
existence of absorption process (energy dissipation).

Dispersion for monochromatic wave

We shall consider the simplest case of monochromatic electromagnetic
field. We assume constancy of the field amplitude which means that
whole energy dissipated inside unit volume in unit of time is compen-
sate by transfer of energy from external source. The average value of
the rate of change of energy density h∂t uit where
Z T
1
h f (t)it := f (t)dt
T 0

represents averaging process on a period represents the amount of heat

per unit of time transferred into unit volume. This quantity can be cast
in the form
1 ⇣ ⌘
Q = h r · Sit = h E · ∂t D it + h H · ∂t Bit , (5.274)
4p
where all the fields are real-valued. This expression is quadratic in
fields. For auxiliary complex fields this expression must be replaced by
 electromagnetic waves 169

the following one

1 ⇣ ⌘
Q= h Re( E) · ∂t Re( D )it + h Re( H ) · ∂t Re( B)it (5.275)
4p

where physical fields are given by real parts of auxiliary fields

1 1
Re( E) = ( E + E ⇤ ), Re( D ) = ( D + D ⇤ ).
2 2

A similar procedure is applied to magnetic fields H and B. The fields

D (t, x) = D0 ( x)e iwt and B(t, x) = B0 ( x)e iwt are monochromatic.
According to (5.251) the constitutive relations read

D (t, x) = #(w ) E(t, x),

B(t, x) = µ(w ) H (t, x).

Their derivatives with respect to time are of the form

iw
∂t Re( D ) = (#E #⇤ E⇤ ) ,
2
iw
∂t Re( B) = (#B #⇤ B⇤ ) .
2

The scalar products in (5.275) take the form

iw
h Re( E) · ∂t Re( D )it = h( E + E⇤ ) · (#E #⇤ E⇤ )it
4
iw h D E D E D E i
= ( # # ⇤ ) | E |2 + # E2 #⇤ E⇤ 2
4 | {z }t | {z }t | {z }t
| E |2 0 0
iw w
= (# #⇤ ) | E|2 = #00 | E|2
4 | {z } 2
2i#00

and
iw w
h Re( H ) · ∂t Re( B)it = (µ µ⇤ ) | H |2 = µ00 | H |2 .
4 | {z } 2
2iµ00

Thus (5.275) reads

w ⇣ 00 2 ⌘
Q= # | E| + µ00 | H |2 . (5.276)
8p

The obtained result shows significance of the imaginary parts of #(w )

and µ(w ). They express the absorption properties of material media.
Since quantity Q is strictly positive, Q > 0, absorption coefficients must
be also positive, #00 > 0 and µ00 > 0. Thus #00 describes electric loses
whereas µ00 describes magnetic loses.
170 lecture notes on classical electrodynamics

Dispersion for non-monochromatic wave

Non-monochromatic fields are represented in the form of Fourier trans-

form
Z •
1 iwt e
E(t, x) = dwe E(w, x), (5.277)
2p •
Z •
1 iwt e (w, x).
D (t, x) = dwe D (5.278)
2p •

Plugging expression (5.277) into (5.249) one gets

Z •
D (t, x) = E(t, x) + dt f (t ) E(t t, x)
0
Z • ✓ Z •
◆
1
= dwe iwt 1 + dt f (t )eiwt e(w, x)
E
2p • 0
Z •
1 iwt e(w, x)
= dwe #(w ) E (5.279)
2p • | {z }
e (w,x)
D

and thus
Z •
i iwt e(w, x).
∂t D (t, x) = dwe w#(w ) E (5.280)
2p •

The real-valued electric field has the form E(t, x) = E⇤ (t, x) where
Z •
1
E⇤ (t, x) = dweiwt E
e⇤ (w, x).
2p •

Changing the variable w ! w in (5.277) one gets

Z •
1
E(t, x) = dweiwt E
e( w, x).
2p •

Comparing both formulas one concludes that the Fourier amplitudes

must satisfy condition

e⇤ (w, x).
e( w, x) = E
E

We assume that non-monochromatic fields vanish sufficiently quickly

for t ! ±•. Total energy dissipated in unit volume is given by the
integral
Z • Z •
dt Q = dt( r · S)
• •
Z •
1
= dt( E · ∂t D + H · ∂t B). (5.281)
4p •

The quantity Q in (5.281) is local in time, Q = ∂t u, which means

that unlike (5.274) it is not average over the period (the field is not
 electromagnetic waves 171

necessarily periodic). The first integral in (5.281) can be cast in the form
Z •
1
dtE · ∂t D =
4p •
Z • Z • Z •
i
dw 0 e i(w +w )t w#(w ) E
0
= dt dw e(w 0 , x)
e(w, x) · E
4p (2p )2 • • •
Z • Z • ✓Z • ◆
i
dte i(w +w )t w#(w ) E
0
= dw dw 0 e(w 0 , x)
e(w, x) · E
4p (2p )2 • • •
| {z }
2pd(w +w 0 )
Z •
i e(w, x) · E
e( w, x)
= dw w#(w ) E
8p 2 • | {z }
e⇤ (w,x)
E
Z •
i e(w, x)|2
= dw w#(w )| E
8p 2 •
Z • Z •
=
i e(w, x)|2 + 1
dw w#0 (w )| E e(w, x)|2 .
dw w#00 (w )| E
8p 2 • 8p 2 •
| {z }
0

First integral vanishes because its integrand is odd. Not that the
expression w is odd whereas #0 ( w ) = #0 (w ) is even and

| Ee( w, x)|2 = Ee( w, x) Ẽ⇤ ( w, x) = | Ee(w, x)|2

| {z } | {z }
e⇤ (w,x)
E e(w,x)
E

is even as well. The magnetic contribution to the integral is obtained in

a similar way. It gives
Z • Z • h i
1 00 e(w, x)|2 + µ00 (w )| H
e (w, x)|2
dt Q = dw w # ( w )| E
• 8p 2 •
Z • h i
1 00 e(w, x)|2 + µ00 (w )| H
e (w, x)|2 .
= dw w # ( w )| E
4p 2 0
(5.282)

where we made use of the fact that the integrand is even function of w,
namely g( w ) = g(w ), and thus the integral over a negative range of
w can be represented the integral over a positive range,
Z 0 Z 0 Z •
dwg(w ) = ( dw ) g( w ) = dwg(w ).
• +• 0

(5.282) shows that coefficients #00 and µ00 are responsible for energy
dissipation in dispersive media. The dissipated energy is converted
into heat which means that Q > 0. Since all frequencies in (5.282) are
positive, then it must hold

#00 > 0, µ00 > 0 (5.283)

for any material. There are no such restriction on the sign of real parts
#0 and µ0 .
Although loses are always present in real materials there are ranges
of frequencies for which they are not significant. They are called
172 lecture notes on classical electrodynamics

transparency regions. In such regions there is possible to introduce the

concept of internal energy of the body in the same sense as in a static
field.

Dispersion for quase-monochromatic wave

For single monochromaic wave there is no steady accumulation of elec-
tromagnetic energy. For this reason we consider a quase-monochromatic
wave i.e. a wave whose components have frequencies in narrow range
about some mean value w0 . The auxiliary complex-valued electromag-
netic field has the form

iw0 t iw0 t
E(t, x) = E0 (t, x)e , H (t, x) = H0 (t, x)e (5.284)

where E0 (t, x) and H0 (t, x) are slowly-varying functions of time. Plug-

ging

1 1
Re( E) = ( E + E ⇤ ), Re( D ) = ( D + D ⇤ ), (5.285)
2 2
1 1
Re( H ) = ( H + H ⇤ ), Re( B) = ( B + B⇤ ),
2 2
into expression

1 ⇣ ⌘
∂t u = Re( E) · ∂t Re( D ) + Re( H ) · ∂t Re( B)
4p

and skipping quickly oscillating terms E · D ⇠ e 2iw0 t , E⇤ · D ⇤ ⇠ e2iw0 t

e.t.c. we get

1
∂t u = ( E · ∂t D ⇤ + E⇤ · ∂t D + H · ∂t B⇤ + H ⇤ · ∂t B) . (5.286)
16p

Quickly oscillating terms ⇠ e±2iw0 t can be formally eliminated by

time averaging over the interval T = 2p w0 . Such averaging does not
change slowly-varying functions h E0 (t, x)it ⇡ E0 (t, x) e.t.c. To obtain
expressions ∂t D and ∂t B we expand the electric field in Fourier series

i ( w0 + w 0 ) t
E(t, x) = Â0 Ee0 (w0 , x)e .
w

Since Ee0 (t, x) is a slowly varying function of time, the frequencies w 0 of

the expansion E0 (t, x) = Âw 0 E e0 (w 0 , x)e iw 0 t are small comparing with
w0 , thus w 0 ⌧ w0 . It leads to the expression
Z •
D (t, x) = E(t, x) + dt f (t ) E(t t, x)
0
✓ Z • ◆
i ( w0 + w 0 ) t e i ( w0 + w 0 ) t
= Â 1+ dt f (t )e E0 ( w 0 , x ) e . (5.287)
w0 | 0
{z }
# ( w0 + w 0 )
 electromagnetic waves 173

Thus temporal derivative is of the form

w#(w )
z }| {
∂t D (t, x) = Â0 e0 (w 0 , x)e i(w0 +w 0 )t
i [(w0 + w 0 )#(w0 + w 0 )] E
w
" #
d(w#(w )) e0 (w 0 , x)e i(w0 +w 0 )t
=Â i w0 # ( w0 ) + 0
w +... E
w0
dw w0
iw0 t iw 0 t
= iw0 #(w0 )e Â0 Ee0 (w 0 , x)e +
w

iw0 t d(w#(w )) iw 0 t
+e
dw Â0 Ee0 (w 0 , x) |( iw 0 )e
{z }
+...
w0 w
d iwt
dt e
!
d(w#) iw0 t
= iw0 #(w0 ) E0 (t, x ) + ∂t E0 (t, x ) + . . . e . (5.288)
dw w0

The quadratic expression is of the form

d(w#)
E⇤ (t, x) · ∂t D (t, x) = iw 0 #(w0 ) E0⇤ · E0 + E ⇤ · ∂ t E0 + . . .
dw w0 | 0 {z }
1 ∂ ( E⇤ · E )
2 t 0 0

Hence, only real-valued terms do not cancel in the sum

d(w#)
E⇤ (t, x) · ∂t D (t, x) + E(t, x) · ∂t D ⇤ (t, x) = ∂t ( E⇤ · E) + . . .
dw w0

where E0⇤ · E0 = E⇤ · E. A similar considerations for magnetic field leads

to expression

d(wµ)
H ⇤ (t, x) · ∂t B(t, x) + H (t, x) · ∂t B⇤ (t, x) = ∂t ( H ⇤ · H ) + . . .
dw w0

Taking the lowest terms of approximation we get

!
1 d(w#) d(wµ)
h∂t uit = ∂t ( E⇤ · E) + ( H ⇤ · H ) . (5.289)
16p dw w0 dw w0

The rate of change of energy density is slow in timescale T describing

the averaging interval. Thus h∂t uit = ∂t huit and it follows from from
(5.289) the expression
!
1 d(w#) ⇤ d(wµ) ⇤
huit = ( E · E) + (H · H) . (5.290)
16p dw w0 dw w0

The scalar product E⇤ · E can be put in the form containing real-valued

fields D E
E ⇤ · E = 2 E2 .
t
Dropping symbols w0 and t one gets expression
✓ ◆
1 d(w#) D 2 E d(wµ) D 2 E
hui = E + H . (5.291)
18p dw dw
174 lecture notes on classical electrodynamics

The formula (5.291) is valid only for electromagnetic fields characterized

by slowly-varying amplitude E0 (t, x). Interruption of electromagnetic
energy supply results in conversion of energy stored in the body (vol-
ume integral of hui) into heat. It means that hui > 0, and thus

d(w#) d(wµ)
> 0, > 0. (5.292)
dw dw

Analytic properties of #(w )

Properties of f (t )
The function f (t ) is represents dispersive properties of material media.
Once given it determines the form of permittivity #(w ). Below we list
main properties of f (t ).

• f (t ) 2 R is finite for all values of t, including t = 0.

The isolation of the term E(t, x) in expression
Z •
D (t, x) = E(t, x) + dt f (t ) E(t t, x) (5.293)
0
Z •
= dt [d(t ) + q (t ) f (t )] E(t t, x)
•

avoids the presence of Dirac delta d(t ).

• For dielectrics f (t ) ! 0 for t ! •.

According to expression
Z t
D (t, x) = E(t, x) + dt0 f (t t 0 ) E ( t 0 , x ),
•

which is just (5.293) for t = t t0 , the remote past t0 ! • cor-

responds with t ! •. For this reason f (t ) for t ! • describes
dielectric media in their very early stage. Actual value of D at t
cannot be strongly influenced by values of the electric field E in the
remote past. This statement is motivated by the fact that physical
mechanism behind (5.293) is a polarisation of media. The func-
tion f (t ) is significantly different from zero for t being of order of
relaxation time.

• For conductors f (t ) 4ps ! 0 for t ! •.

The mathematical formalism developed for dielectrics possesses
applications for conductors despite the physical mechanism behind
(5.293) is different. Although the stationary current does not change
the physical state of conductors, it is formally responsible for the
appearance of the displacement function D. Indeed, comparing the
displacement current with the conduction current

1 ∂D 4p
= sE
c ∂t c
 electromagnetic waves 175

and integrating this expression one gets

Z t Z •
D (t) = 4ps dt0 E(t0 , x) = 4ps dtE(t t, x),
• 0

where the condition that there is no external electric field at the

remote past, E(t0 , x) ! 0 for t0 ! • is used.

In further part we shall discuss properties of function #(w ) which

was defined as Z •
#(w ) = 1 + dt f (t )eiwt .
0

The parameter w was real-valued in our previous analysis. However,

some important properties of the function #(w ) can be obtained using
method of analysis of complex functions. For this reason we shall
assume that argument of the function #(z) is a complex variable, z 2 C
i.e.
z = z0 + iz00 , z0 , z00 2 R.
Properties of #(w ) in the upper
Function #(w ) has the following properties in the upper complex complex half-plane
half-plane.

1. Function #(z) is a single-valued regular function in the upper half-plane. It

means that the function has no singular points in this region. Indeed,
for z00 > 0 (upper half-plane) the expression eizt is proportional to
00
e z t . Finiteness of f (t ) in whole region of integration leads to
convergence of the integral
Z • Z •
z00 t iz0 t
#(z) = 1 + dt f (t )eizt = 1 + dt f (t )e e . (5.294)
0 0

It means that the function is single-valued and finite in this region.

Function #(z) has no singularities at the real axis (z00 = 0), except
the case of metals, where it has simple pole at the origin. A regular
character of #(z) in the upper half-plane is a consequence of causality
principle, implemented by the presence of q (t0 t) in the integral
Z • Z •
0 0 0 0
dt q (t t) f (t t) E(t , x) = dt f (t ) E(t, x).
• 0

The expression (5.294) is meaningless in the lower half-plane because

the integral diverges. For this reason the integral in the region z00 < 0
can be defined as analytic continuation of the expression #(z) given
in the upper half plane. However, in general such defined function
has some singularities. Only in the upper half-plane the function
#(z) well-defined from mathematical point of view and has physical
interpretation.
176 lecture notes on classical electrodynamics

2. Function #(z) is odd at the real z00 = 0 axis and changes its sign at z0 = 0.
Since f (t ) is real-valued, then
Z • ✓ Z • ◆⇤
⇤
#( z⇤ ) = 1 + dt f (t )e iz t = 1 + dt f (t )eizt
0 0
⇤
= # ( z ).
It implies that the real and imaginary parts of #(z) satisfy

# 0 ( z ⇤ ) = # 0 ( z ),
#00 ( z⇤ ) = #00 (z). (5.295)

In particular, #(z) is real-valued, #00 (iz00 ) = 0, at the imaginary z0 = 0

axis
#(iz00 ) = #⇤ (iz00 ).
According to our previous considerations, the imaginary part of
# = #0 + i#00 is positive, #00 > 0, for positive frequencies w > 0. It
means that imaginary part of permittivity is positive, #00 (z0 ) > 0, at
real z0 > 0 semi-axis. The equality (5.295) takes the form #00 ( z0 ) =
#00 (z0 ) < 0. We conclude that #(z0 ) is negative on the negative part
of the real z0 axis. The imaginary part of #(z) changes its sign at
z0 = 0 passing through zero for dielectrics or through infinity for
metals. The point z0 = 0 is the only point at the real axis in which
the function #(z) can vanish.

3. Function #(z) tends to unity for z ! • in the upper-half plane.

w2
It has been shown in the particular case that #(w ) = 1 w2p ! 1 for
w ! • which corresponds with the case of the real z0 axis (z00 = 0).
In generality, #(z) ! 1 for z ! • in any manner in the upper half
plane. It follows directly from the integral
Z •
z00 t iz0 t
#(z) = 1 + dt f (t )e e
0

which vanishes in the limit z00 ! • for z0 = 0. For finite z00 and
0
z0 6= 0 the integral vanishes because of the oscillating factor eiz t .

4. Theorem: Function #(z) has no real values at any finite point localized
in the upper half-plane, except the points at the imaginary axis where it
decreases monotonically

dielectrics : #0 & 1 for z00 : 0 ! •,

metals : • & 1 for z00 : 0 ! •.

This theorem implies that #(z) has no zeros in the upper half-plane.
Proof: Let us consider the integral over closed curve C in the complex
plane z
I
1 dg(z) dz
= N0 N• , (5.296)
2pi C dz g(z) a
 electromagnetic waves 177

where N0 and N• stand, respectively, for number of zeros and

number of poles of the function g(z) a in the region enclosed by C.
Assumptions:

• g(z) is a function which has no poles in the upper half plane (the
pole at z = 0 is allowed),
• a is a real number,
• g0 2 R is represents the value of the function g(z) at the origin
z = 0 i.e. g0 = limz!0 g(z),
• the contour of integration C consists on the real axis z00 = 0 and
semi circle of infinite radius in the upper half-plane,
• the function g(z) vanishes at the infinite semicircle.

The contour C in the complex pane z is mapped on the closed

contour C 0 in the complex plane g such that the point z = 0 is
mapped on g0 2 R, infinite semicircle is mapped on g = 0 and the
real positive/negative semi-axes are mapped on two curves (with
self-crossing points) lying, respectively, in the upper/lower half
planes. The contours are plotted in Figure 5.14. These curves have
no crossing points with the real axis (z00 = 0) except two points g = 0
and g = g0 .
Since g(z) has no poles in the upper half-plane (so is g(z) a) then (a)
the integral (5.296) equals to the number of zeros
I
1 dg
= N0 . (5.297)
2pi C0 g a

There are two possibilities:

• For g0 < • it follows from the form of the contour C 0 that the total
increase of the argument of g(z) a is equal to 2p providing that
a is located inside the contour. Otherwise, it is equal to zero. It
gives N0 = 1 for 0 < a < g0 and N0 = 0 for a > g0 . It means that
g(z) in the upper half-plane takes the value a 2 R only once and it
does not take real values R \[0, g0 ]. We conclude that g(z) cannot (b)
have maximum neither minimum at the imaginary upper semi-axis Figure 5.14: (a) The integration contour
Cat the complex plane z and (b) its image
z0 = 0 (otherwise it would take some values twice). Hence, this C 0 on the complex plane g.
function changes monotonically at the upper imaginary semi-axis
of the complex plane z.
• For g0 = • the function g(z) has a pole at z = 0. The proof
is similar. The only difference is presence of additional small
semicircle around z = 0 in the upper half-plane. The function at
the upper imaginary semi-axis decreases from g = • to g = 0.

Properties of the function #(z) follows from properties of g(z) for

g(z) = #(z) 1.
178 lecture notes on classical electrodynamics

The Kramers-Kronig relations

Analytical properties of the function #(z) in the upper half-plane allows

for obtaining some important relations between their real and imaginary
parts – calculated for points lying on the real axis.
Let g(z) be a complex function that has continuous complex deriva-
tive (Cauchy-Riemann equations are satisfied i.e. ∂ x Re( g) ∂y Im( g) =
0 and ∂ x Im( g) + ∂y Re( g) = 0, where z = x + iy). According to the
integral Cauchy theorem the integral over the border ∂D of an open set
D vanishes,
I
g(z)dz = 0.
∂D

It implies the integral Cauchy formula

I
(
g(z) 2pig(z0 ) for z0 2 D,
dz = (5.298)
∂D z z0 0 for z0 2
/ D.

In the following part we set the function g(z) := #(z) 1 (regular in

the upper half-plane) and denote the pole z0 := w 2 R + .
The integration contour C consists on the real axis (z00 = 0) with
small semicircle(s) in the upper half plane: Cw and C0 (only for metals)
with centers at z = w and z = 0 and the semicircle with infinite radius Figure 5.15: The integration contour C.
C• in the upper half-plane, see Figure 5.15. The function #(z) 1
vanishes at the infinite semicircle so the function (# 1)/(z w ) tends
to zero for z ! • quicker than the function 1/z. Hence, the integral
H #(z) 1
C dz z w converges. Moreover, since # ( z ) has no poles in the upper
half-plane and the point z = w does not belong to region D then
(#(z) 1)/(z w ) is analytic function. In such a case the integral along
C must vanish
I
#(z) 1
dz = 0.
C z w

The integral at the large semicircle C• vanishes. In the case of dielectrics

there is only one small semicircle is Cw . We choose the parametrization
z = w + reij at Cw and take the limit r ! 0,
✓Z Z •◆ Z
r #(z) 1 #(z) 1
lim + dz + lim dz = 0. (5.299)
r !0 • r z w r !0 Cw (r) z w
| {z }
R• #(z) 1
P • z w dz

The second integral can be split into two integrals

Z Z Z
#(z) 1 #(z) #(w ) #(w ) 1
dz = dz + dz
Cw (r) z w Cw (r) z w Cw (r) z w
 electromagnetic waves 179

where
Z Z
#(z) #(w ) #(z) #(w )
dz  dz
Cw (r) z w Cw (r) z w
Z Z
|#(z) #(w )| |#(z) #(w )| i( j+ p )
 |dz| = |re 2 dj |
Cw (r) |z w | Cw (r) |reij |
Z Z 0
= |#(z) #(w )|dj  sup |#(z) #(w )| dj.
Cw (r) x 2Cw (r) p

The function #(z) is continuous in z = w so

!
lim p sup |#(z) #(w )| = 0.
r !0 z2Cw (r)

It leads to vanishing of the integral

Z
#(z) #(w )
lim dz = 0.
r!0 Cw (r) z w

The second integral reads

Z Z
#(w ) 1 ireij dj
lim dz = (#(w ) 1) lim = ip (#(w ) 1).
r!0 Cw (r) z w r!0 Cw (r) reij

Then the equality (5.299) takes the form

Z •
#(z) 1
P dz ip (#(w ) 1) = 0
• z w

and thus
Z •
1 #( x ) 1
#(w ) 1= P dx (5.300)
ip • x w

where the symbol x 2 R emphasizes that the integration is taken along

the real axis. Plugging # = #0 + i#00 we get
Z •
i (#0 ( x ) 1) + i#00 ( x )
(#0 (w ) 1) + i#00 (w ) = P dx.
p • x w

It leads to two equalities

Z • 00
1 # (x)
#0 (w ) 1= P dx, (5.301)
p • x w
Z •
1 #0 ( x ) 1
#00 (w ) = P dx. (5.302)
p • x w

where w, x 2 R. Functions #0 and #00 have defined parity. The relation

#( z⇤ ) = #⇤ (z) gives

# 0 ( x ) = # 0 ( x ), #00 ( x ) = #00 ( x ).
180 lecture notes on classical electrodynamics

The integral in (5.301) can be cast in the form of integral over positive
frequencies
Z • 00 Z 0 Z
• #00 ( x )
# (x) #00 ( x )
P dx = P dx + P dx
• x w • x w 0 x w
Z 0 Z • #00 ( x )
#00 ( x )
=P ( dx ) + P dx
• x w 0 x w
Z •✓ ◆
x w x+w
=P + #00 ( x )dx
0 x2 w 2 x2 w 2
Z •
x#00 ( x )
= 2P dx.
0 x2 w2
In a similar way we transform the integral in (5.302)
Z • 0 Z 0 0 Z • 0
# (x) 1 # (x) 1 1
# (x)
P dx = P dx + P dx
• x w • x w 0 x w
Z 0 0 Z • 0
# (x) 1 # (x) 1
=P ( dx ) + P dx
• x w 0 x w
Z • ✓ ◆
x w x+w
=P 2 2
(#0 ( x ) 1)dx
0 x w x2 w 2
Z • 0
# (x) 1
= 2w P .
0 x2 w 2
It leads to Kramers-Kronig relations Kramers-Kronig relations
Z •
2 x#00 ( x )
#0 (w ) = 1 + P dx, (5.303)
p 0 x2 w2

Z • 0
00 2w # (x) 1
# (w ) = P dx. (5.304)
p 0 x2 w2
The above relations have been derived for dielectrics. In order to get
relations that are valid for metals we observe that #(z) have singular
part i 4ps
w at w ! 0
4ps
#(w ) = i + #̃(w ) (5.305)
w
where #̃(w ) is some regular term. The singular term affects only the
imaginary part #0 . Thus

4ps
#̃0 (w ) = #(w ), #̃00 (w ) = #00 (w ) i .
w
Substituting regular expressions

4ps
#0 ! #̃0 = #0 , #00 ! #̃00 = #00
w
into (5.301) and (5.302) one gets
Z •
0 2 x#00 ( x )
# (w ) = 1 + P dx, (5.306)
p 0 x2 w2
 electromagnetic waves 181

and
Z • 0
4ps 2w # (x) 1
#00 (w ) = P dx. (5.307)
w p 0 x2 w2

The second relation (5.307) follows immediately from (5.302). On the

other hand, the first relation (5.306) requires more careful treatment.
The integral in (5.301) reads
⇣ ⌘
Z • x #00 ( x ) 4ps Z • Z •
x x#00 ( x ) dx
P 2 2
dx = P 2 2
dx 4ps P 2
,
0 x w 0 x w 0 x w2
where
Z • Z •✓ ◆
dx 1 1 1
P = P dx
0 x2 w 2 2w 0 x w x+w
Z R ✓ ◆
1 1 1
= lim P dx
2w R!• 0 x w x+w
✓Z w h Z R ◆
1 1 1
= lim lim + dx
2w R!• h !0 0 x w w +h x w
Z
R dx
1
lim
2w R!• 0 x + w
1 ⇣
= lim lim ln(w h w ) ln( w ) + ln( R w )
2w R!• h !0
⌘ 1
ln(w + h w ) lim [ln( Rw ) ln(w )]
2w R!•
✓ ◆ ✓ ◆
1 h(R w) 1 R+w
= lim lim ln lim ln
2w R!• h !0 h( w) 2w R!• w
✓ ◆
1 R w
= lim ln = 0. (5.308)
2w R!• R+w

Formulas (5.301) and (5.306) are especially important because they

allows to calculate function #0 (w ) if #00 (w ) is known (at least empirically).
The function #00 (w ) can be determined experimentally by the measure
of absorption.

5.8 Waveguides

Waveguides are formed by conducting tubes filled with dielectric mate-

rials. In general such can have magnetic properties. The perpendicular
cross section of the waveguide is not necessary a symmetric figure.
The electric and magnetic field exist only inside the waveguide

E1 ⌘ E, B1 ⌘ B, D1 ⌘ D, H1 ⌘ H
Figure 5.16: A waveguide cross section.
and

E2 = 0, B2 = 0, D2 = 0, H2 = 0.
182 lecture notes on classical electrodynamics

The field intensities and inductions are related by constitutive relations

D = #E and B = µH.
Let n̂ be a vector normal to the waveguide surface pointing out-
side the waveguide. The fields satisfy the boundary conditions at the
waveguide surface S, namely

D · n̂ = 4ps, B · n̂ = 0, (5.309)
4p Figure 5.17: Waveguides
n̂ ⇥ E = 0, n̂ ⇥ H = K. (5.310)
c
Maxwell’s equations imply that electric and magnetic field obey wave
equation ∂2 E = 0 and ∂2 B = 0 where ∂2 := c2 ∂2t r2 . We choose the
#µ

axis x3 as being parallel to the longitudinal axis of the waveguide. The

electric and magnetic field can be considered in the following form
3
E(t, r ) = E( x1 , x2 )ei(±kx wt)
, (5.311)
3
B(t, r ) = B( x1 , x2 )ei(±kx wt)
. (5.312)

We shall adopt convention that, if not written explicitly, vectors E( x1 , x2 ) ⌘

E and B( x1 , x2 ) ⌘ B are denoted by symbols without arguments.
Plugging this expression into wave equation we get equations for
E( x1 , x2 ) and B( x1 , x2 ). The wave operator acting on electric field gives

h #µ 2 i ⇣ 3
⌘
∂2 E(t, r ) = r2? + ∂33 ∂
2 t
E( x1 , x2 )ei(±kx wt)

h i c
3
= r2? + g2 E( x1 , x2 )ei(±kx wt) , (5.313)
Figure 5.18: Coaxial cable
and similarly for the magnetic field, where

w2
g2 := #µ k2 . (5.314)
c2
It gives
h i
r2? + g2 E( x1 , x2 ) = 0,
h i (5.315)
r2? + g2 B( x1 , x2 ) = 0.

Parallel and transversal components

The electric and magnetic field amplitudes E( x1 , x2 ), B( x1 , x2 ) can be
decomposed on two components which are perpendicular and parallel
to the x3 axis,
E(t, r ) = E1 ê1 + E2 ê2 + E3 ê3 ,
| {z } |{z}
E? Ek
(5.316)
B(t, r ) = B ê1 + B ê2 + B3 ê3 .
1 2
| {z } |{z}
B? Bk
 electromagnetic waves 183

Such decomposition is very useful. In some cases (transverse electric

TE and transverse magnetic TM waves) the perpendicular E? , B?
and parallel components satisfy some algebraic relations. It follows
that the perpendicular components can be determined from the parallel
components. The parallel components E3 and B3 are determined by
equations (5.315). We shall present Maxwell’s equations in terms of
perpendicular and parallel components. The gradient operator is split
as follows r = r? + ê3 ∂3 .

• Faraday’s law takes the form

1
r? ⇥ E(t, r ) + ê3 ⇥ ∂3 E(t, r ) = ∂t B(t, r ) . (5.317)
| {z } c | {z }
±ikê3 ⇥ E(t,r ) iwB(t,r )

It follows that expression e i (±kx3

wt) became just multiplicative factor,

then the amplitude vectors E( x1 , x2 ) and B( x1 , x2 ) satisfy equation

w
r? ⇥ E ± ik ê3 ⇥ E = i B.
c
Plugging decomposition (5.316) into the above equation one gets
w
r? ⇥ Ek + r? ⇥ E? ± ik ê3 ⇥ E? = i ( B + B? ). (5.318)
c k
Taking the scalar product of (5.318) and ê3 and considering that
ê3 · (r? ⇥ Ek ) = 0 and ê3 · (ê3 ⇥ E? ) = 0 we obtain

w 3
ê3 · (r? ⇥ E? ) = i B . (5.319)
c
Next, taking the cross product of (5.318) with ê3 and taking into
account relations ê3 ⇥ (r? ⇥ E? ) = 0 and ê3 ⇥ (ê3 ⇥ E? ) = E?
and

ê3 ⇥ (r? ⇥ Ek ) = ê3 ⇥ (ê1 ⇥ ê3 ∂1 E3 + ê2 ⇥ ê3 ∂2 E3 )

= ê3 ⇥ ( ê2 ∂1 E3 + ê1 ∂2 E3 ) = ê1 ∂1 E3 + ê2 ∂2 E3
= r ? E3

we get equation
w
r? E3 ⌥ ik E? = i ê3 ⇥ B? . (5.320)
c

• For Ampere-Maxwell’s law we obtain

#µ
r? ⇥ B(t, r ) + ê3 ⇥ ∂3 B(t, r ) = ∂t E(t, r ) . (5.321)
| {z } c | {z }
±ikê3 ⇥ B(t,r ) iwE(t,r )

Then
w
r? ⇥ Bk + r? ⇥ B? ± ik ê3 ⇥ B? = i#µ ( E + E? ). (5.322)
c k
184 lecture notes on classical electrodynamics

Taking the scalar product of (5.322) and ê3 one gets equation

w 3
ê3 · (r? ⇥ B? ) = i#µ E . (5.323)
c

The cross product yields

w
r? B3 ⌥ ik B? = i#µ ê3 ⇥ E? . (5.324)
c

• Finally for electric and magnetic Gauss law one gets

r? · E? = ⌥ik E3 , r? · B? = ⌥ik B3 . (5.325)

In order to obtain E? we take left multiplication of (5.324) by ê3 Relations between amplitudes
which gives for g2 6= 0
w
ê3 ⇥ (r? B3 ) ⌥ ik ê3 ⇥ B? = i#µ ê3 ⇥ (ê3 ⇥ E? ) . (5.326)
c | {z }
E?

c
Substituting ê3 ⇥ B? = iw (r? E3 ⌥ ik E? ), obtained from (5.320), into
(5.326) and multiplying by i wc we obtain

i h w i
E? = ± k r ? E3 ê3 ⇥ r? B3 , (5.327)
g2 c

where in the last step divide by g2 . This relation is true unless g2 = 0

2
i.e. when #µ wc2 = k2 . Similar result can be obtained for B? . In this case
we take left cross product of (5.320) with ê3
w
ê3 ⇥ (r? E3 ) ⌥ ik ê3 ⇥ E? = i ê3 ⇥ (ê3 ⇥ B? ) . (5.328)
c | {z }
B?

c 3
and substitute ê3 ⇥ E? = iw#µ (r? B ⌥ ik B? ) obtained from (5.324).
Multiplying the result by i#µ wc and and dividing by g2 we obtain

i h 3 w 3
i
B? = ± k r ? B + #µ ê 3 ⇥ r ? E . (5.329)
g2 c

Note that in this case the amplitudes E? and B? are uniquely deter-
mined by components E3 and B3 .

TE and TM waves
The electromagnetic field takes non-zero values only inside the waveg-
uide. The electric field E satisfy the equation (r2? + g2 ) E = 0 (and
similarly for B). Moreover, since the waveguide surface S is a conductor
then tangential component of the electric field must vanish at this surface
 electromagnetic waves 185

E3 |S = 0. Taking the scalar product of the equation (5.324) with the

vector n̂ normal to the waveguide surface we obtain

w
n̂ · (r? B3 ) ⌥ik n̂ · B? = i#µ n̂ · (ê3 ⇥ E? ) . (5.330)
| {z } | {z } c | {z }
∂B3 n̂· B ê3 ·(n̂⇥ E)
∂n

At the surface S where n̂ · B|S = 0 and n̂ ⇥ E|S = 0 the equation (5.330)

implies that

∂B3
= 0.
∂n S

Any solution representing electromagnetic wave that propagates in the

waveguide must satisfy the following boundary conditions

∂B3
n̂ · B|S = 0, n̂ ⇥ E|S = 0, E3 |S = 0, = 0. (5.331)
∂n S

• Transverse electric waves TE is a group of electromagnetic waves such

that E3 = 0 everywhere. The relation between E? and B? for TE
waves follows directly from (5.320) and it reads

w kc
E? = ⌥ ê3 ⇥ B? , B? = ± ê3 ⇥ E? . (5.332)
kc w

The equation (5.329) gives

ik
B? = ± r B3 , (5.333)
g2 ?

where B3 is the solution of (r2? + g2 ) B3 = 0. This solution must

3
satisfy the Neumann boundary condition ∂B∂n |S = 0. The particular
form of the solution depends on the waveguide geometry.

• Transverse magnetic waves TM are solutions such that B3 = 0 every-

where. The equation (5.324) gives

w kc
B? = ±#µ ê3 ⇥ E? , E? = ⌥ ê3 ⇥ B? , (5.334)
kc #µ w

where E? follows from (5.327) and it reads

ik
E? = ± r E3 . (5.335)
g2 ?

The transverse component of the electric field is obtained from

(r2? + g2 ) E3 = 0 with the Dirichlet boundary condition E3 |S = 0.
186 lecture notes on classical electrodynamics

Transverse electromagnetic waves TEM

Transverse electromagnetic waves TEM are solutions such that E3 = 0 and
B3 = 0 everywhere. Equations (5.319) and (5.323) take the form

ê3 · (r? ⇥ E? ) = 0, ê3 · (r? ⇥ B? ) = 0. (5.336)

From (5.325) we have

r? · E? = 0, r? · B? = 0. (5.337)

The electric field in Cartesian coordinates satisfies equations

∂1 E2 ∂2 E1 = 0, ∂1 E1 + ∂2 E2 = 0. (5.338)

Taking derivative of the first equation with respect to x1 and using the
second equation we obtain (∂21 + ∂22 ) E2 = 0. Similarly, taking derivative
with respect to x2 one gets (∂21 + ∂22 ) E1 = 0. Repeating the same steps
with respect to the magnetic field one gets identical equations, thus

r2? E? = 0, r2? B? = 0. (5.339)

Comparing (5.339) and (5.315) one can conclude that

w2
g2 = #µ k2 = 0.
c2
It means that dispersion relation for electromagnetic waves TEM has
the same form as for waves in infinite space
w
k=n . (5.340)
c
The algebraic relations between electric and magnetic fields follow
directly from (5.320) and (5.324). They have the form
1
B = ±n ê3 ⇥ E? , E=⌥ ê3 ⇥ B? . (5.341)
n
Another important property of TEM waves is fact that such waves
cannot exist in hollow waveguides. The expression r? ⇥ E? has only
component parallel to x3 , thus the equation ê3 · (r? ⇥ E? ) = 0 is
equivalent to r? ⇥ E? = 0. However, since E? is a function of two
spatial variables then

r? ⇥ E? = 0 ) E? = r j ( x 1 , x 2 ). (5.342)

Plugging this expression into r? · E? = 0 we obtain Laplace equation

in two dimensions j

r2? j = 0. (5.343)

Since the electric field vanishes on S, then the potential j = const on

S. The Laplace equation has only trivial constant solution in such a
case. Then E? = 0 inside the waveguide. For this reason there are no
electromagnetic TEM waves in hollow waveguides.