2 1.

Maxwells Equations

the receiving antennas. Away from the sources, that is, in source-free regions of space,

1 Maxwells equations take the simpler form:

B
E=
Maxwells Equations D
t

H=
t (source-free Maxwells equations) (1.1.2)

D=0

B=0

The qualitative mechanism by which Maxwells equations give rise to propagating

electromagnetic elds is shown in the gure below.

1.1 Maxwells Equations

Maxwells equations describe all (classical) electromagnetic phenomena:

B
E= For example, a time-varying current J on a linear antenna generates a circulating
t
and time-varying magnetic eld H, which through Faradays law generates a circulating
D
H=J+ electric eld E, which through Amperes law generates a magnetic eld, and so on. The
t (Maxwells equations) (1.1.1)
cross-linked electric and magnetic elds propagate away from the current source. A
D= more precise discussion of the elds radiated by a localized current distribution is given
in Chap. 15.
B=0

The rst is Faradays law of induction, the second is Amperes law as amended by
Maxwell to include the displacement current D/t, the third and fourth are Gauss laws 1.2 Lorentz Force
for the electric and magnetic elds.
The force on a charge q moving with velocity v in the presence of an electric and mag-
The displacement current term D/t in Amperes law is essential in predicting the
netic eld E, B is called the Lorentz force and is given by:
existence of propagating electromagnetic waves. Its role in establishing charge conser-
vation is discussed in Sec. 1.7.
F = q(E + v B) (Lorentz force) (1.2.1)
Eqs. (1.1.1) are in SI units. The quantities E and H are the electric and magnetic
eld intensities and are measured in units of [volt/m] and [ampere/m], respectively. Newtons equation of motion is (for non-relativistic speeds):
The quantities D and B are the electric and magnetic ux densities and are in units of
[coulomb/m2 ] and [weber/m2 ], or [tesla]. D is also called the electric displacement, and dv
m = F = q(E + v B) (1.2.2)
B, the magnetic induction. dt
The quantities and J are the volume charge density and electric current density where m is the mass of the charge. The force F will increase the kinetic energy of the
(charge ux) of any external charges (that is, not including any induced polarization charge at a rate that is equal to the rate of work done by the Lorentz force on the charge,
charges and currents.) They are measured in units of [coulomb/m3 ] and [ampere/m2 ]. that is, v F. Indeed, the time-derivative of the kinetic energy is:
The right-hand side of the fourth equation is zero because there are no magnetic mono-
1 dWkin dv
pole charges. Eqs. (1.3.17)(1.3.19) display the induced polarization terms explicitly. Wkin = mv v = mv = v F = qv E (1.2.3)
The charge and current densities , J may be thought of as the sources of the electro- 2 dt dt
magnetic elds. For wave propagation problems, these densities are localized in space; We note that only the electric force contributes to the increase of the kinetic energy
for example, they are restricted to ow on an antenna. The generated electric and mag- the magnetic force remains perpendicular to v, that is, v (v B)= 0.
netic elds are radiated away from these sources and can propagate to large distances to
1.3. Constitutive Relations 3 4 1. Maxwells Equations

Volume charge and current distributions , J are also subjected to forces in the The next simplest form of the constitutive relations is for simple homogeneous
presence of elds. The Lorentz force per unit volume acting on , J is given by: isotropic dielectric and for magnetic materials:

f = E + J B (Lorentz force per unit volume) (1.2.4) D = E

(1.3.4)
3
where f is measured in units of [newton/m ]. If J arises from the motion of charges B = H
within the distribution , then J = v (as explained in Sec. 1.6.) In this case,
These are typically valid at low frequencies. The permittivity  and permeability
f = (E + v B) (1.2.5) are related to the electric and magnetic susceptibilities of the material as follows:

By analogy with Eq. (1.2.3), the quantity v f = v E = J E represents the power

per unit volume of the forces acting on the moving charges, that is, the power expended
by (or lost from) the elds and converted into kinetic energy of the charges, or heat. It
has units of [watts/m3 ]. We will denote it by:
dPloss
=JE (ohmic power losses per unit volume) (1.2.6)
dV
In Sec. 1.8, we discuss its role in the conservation of energy. We will nd that elec-
tromagnetic energy owing into a region will partially increase the stored energy in that
region and partially dissipate into heat according to Eq. (1.2.6).
 = 0 (1 + )
(1.3.5)
= 0 (1 + m )
The susceptibilities , m are measures of the electric and magnetic polarization
properties of the material. For example, we have for the electric ux density:
D = E = 0 (1 + )E = 0 E + 0 E = 0 E + P (1.3.6)
where the quantity P = 0 E represents the dielectric polarization of the material, that
is, the average electric dipole moment per unit volume. In a magnetic material, we have

B = 0 (H + M)= 0 (H + m H)= 0 (1 + m )H = H (1.3.7)

1.3 Constitutive Relations
where M = m H is the magnetization, that is, the average magnetic moment per unit
The electric and magnetic ux densities D, B are related to the eld intensities E, H via volume. The speed of light in the material and the characteristic impedance are:
the so-called constitutive relations, whose precise form depends on the material in which 
1
the elds exist. In vacuum, they take their simplest form: c= , = (1.3.8)
 
D = 0 E The relative permittivity, permeability and refractive index of a material are dened by:
(1.3.1)
B = 0 H 
rel = = 1 + , rel = = 1 + m , n= rel rel (1.3.9)
0 0
where 0 , 0 are the permittivity and permeability of vacuum, with numerical values:
so that n2 = rel rel . Using the denition of Eq. (1.3.8), we may relate the speed of light
and impedance of the material to the corresponding vacuum values:
0 = 8.854 1012 farad/m
(1.3.2) 1 1 c0 c0
0 = 4 107 henry/m c= = = =
0 0 rel rel rel rel n
    (1.3.10)
The units for 0 and 0 are the units of the ratios D/E and B/H, that is, 0 rel rel rel n
= = = 0 = 0 = 0
 0 rel rel n rel
coulomb/m2 coulomb farad weber/m2 weber henry
= = , = =
volt/m volt m m ampere/m ampere m m For a non-magnetic material, we have = 0 , or, rel = 1, and the impedance
becomes simply = 0 /n, a relationship that we will use extensively in this book.
From the two quantities 0 , 0 , we can dene two other physical constants, namely,
More generally, constitutive relations may be inhomogeneous, anisotropic, nonlin-
the speed of light and the characteristic impedance of vacuum:
ear, frequency dependent (dispersive), or all of the above. In inhomogeneous materials,
 the permittivity  depends on the location within the material:
1 0
c0 = = 3 108 m/sec , 0 = = 377 ohm (1.3.3)
0 0 0 D(r, t)= (r)E(r, t)
1.3. Constitutive Relations 5 6 1. Maxwells Equations

In anisotropic materials,  depends on the x, y, z direction and the constitutive rela- In Sections 1.101.15, we discuss simple models of () for dielectrics, conductors,
tions may be written component-wise in matrix (or tensor) form: and plasmas, and clarify the nature of Ohms law:

Dx xx xy xz Ex J = E (Ohms law) (1.3.15)

Dy = yx yy yz Ey (1.3.11)
In Sec. 1.17, we discuss the Kramers-Kronig dispersion relations, which are a direct
Dz zx zy zz Ez
consequence of the causality of the time-domain dielectric response function (t).
Anisotropy is an inherent property of the atomic/molecular structure of the dielec- One major consequence of material dispersion is pulse spreading, that is, the pro-
tric. It may also be caused by the application of external elds. For example, conductors gressive widening of a pulse as it propagates through such a material. This effect limits
and plasmas in the presence of a constant magnetic eldsuch as the ionosphere in the the data rate at which pulses can be transmitted. There are other types of dispersion,
presence of the Earths magnetic eldbecome anisotropic (see for example, Problem such as intermodal dispersion in which several modes may propagate simultaneously,
1.10 on the Hall effect.) or waveguide dispersion introduced by the conning walls of a waveguide.
In nonlinear materials,  may depend on the magnitude E of the applied electric eld There exist materials that are both nonlinear and dispersive that support certain
in the form: types of non-linear waves called solitons, in which the spreading effect of dispersion is
D = (E)E , where (E)=  + 2 E + 3 E2 + (1.3.12) exactly canceled by the nonlinearity. Therefore, soliton pulses maintain their shape as
Nonlinear effects are desirable in some applications, such as various types of electro- they propagate in such media [1417,919,917].
optic effects used in light phase modulators and phase retarders for altering polariza- More complicated forms of constitutive relationships arise in chiral and gyrotropic
tion. In other applications, however, they are undesirable. For example, in optical bers media and are discussed in Chap. 4. The more general bi-isotropic and bi-anisotropic
nonlinear effects become important if the transmitted power is increased beyond a few media are discussed in [30,96]; see also [57].
milliwatts. A typical consequence of nonlinearity is to cause the generation of higher In Eqs. (1.1.1), the densities , J represent the external or free charges and currents
harmonics, for example, if E = E0 ejt , then Eq. (1.3.12) gives: in a material medium. The induced polarization P and magnetization M may be made
explicit in Maxwells equations by using the constitutive relations:
D = (E)E = E + 2 E2 + 3 E3 + = E0 ejt + 2 E02 e2jt + 3 E03 e3jt +
D = 0 E + P , B = 0 (H + M) (1.3.16)
Thus the input frequency is replaced by , 2, 3, and so on. In a multi-
Inserting these in Eq. (1.1.1), for example, by writing B = 0 (H + M)=
wavelength transmission system, such as a wavelength division multiplexed (WDM) op-
0 (J + D + M)= 0 (0 E + J + P + M), we may express Maxwells equations in
tical ber system carrying signals at closely-spaced carrier frequencies, such nonlinear-
terms of the elds E and B :
ities will cause the appearance of new frequencies which may be viewed as crosstalk
among the original channels. For example, if the system carries frequencies i , i = B
E=
1, 2, . . . , then the presence of a cubic nonlinearity E3 will cause the appearance of the t
frequencies i j k . In particular, the frequencies i + j k are most likely E  P 
to be confused as crosstalk because of the close spacing of the carrier frequencies. B = 0 0 + 0 J + + M
t t (1.3.17)
Materials with a frequency-dependent dielectric constant () are referred to as
1
dispersive. The frequency dependence comes about because when a time-varying elec- E= P)
0
tric eld is applied, the polarization response of the material cannot be instantaneous.
Such dynamic response can be described by the convolutional (and causal) constitutive B=0
relationship:

t We identify the current and charge densities due to the polarization of the material as:
D(r, t)= (t t )E(r, t ) dt (1.3.13)

P
which becomes multiplicative in the frequency domain: Jpol = , P
pol = (polarization densities) (1.3.18)
t

D(r, )= ()E(r, ) (1.3.14) Similarly, the quantity Jmag = M may be identied as the magnetization current
density (note that mag = 0.) The total current and charge densities are:
All materials are, in fact, dispersive. However, () typically exhibits strong depen-
dence on only for certain frequencies. For example, water at optical frequencies has P
Jtot = J + Jpol + Jmag = J + + M
refractive index n = rel = 1.33, but at RF down to dc, it has n = 9. t (1.3.19)
tot = + pol = P
1.4. Negative Index Media 7 8 1. Maxwells Equations

and may be thought of as the sources of the elds in Eq. (1.3.17). In Sec. 15.6, we examine current density; the difference of the normal components of the ux density D are equal
this interpretation further and show how it leads to the Ewald-Oseen extinction theorem to the surface charge density; and the normal components of the magnetic ux density
and to a microscopic explanation of the origin of the refractive index. B are continuous.
The Dn boundary condition may also be written a form that brings out the depen-
dence on the polarization surface charges:
1.4 Negative Index Media
(0 E1n + P1n )(0 E2n + P2n )= s 0 (E1n E2n )= s P1n + P2n = s,tot
Maxwells equations do not preclude the possibility that one or both of the quantities
, be negative. For example, plasmas below their plasma frequency, and metals up to The total surface charge density will be s,tot = s +1s,pol +2s,pol , where the surface
optical frequencies, have  < 0 and > 0, with interesting applications such as surface charge density of polarization charges accumulating at the surface of a dielectric is seen
plasmons (see Sec. 8.5). to be (n is the outward normal from the dielectric):
Isotropic media with < 0 and  > 0 are more difcult to come by [168], although
examples of such media have been fabricated [396]. s,pol = Pn = n P (1.5.2)
Negative-index media, also known as left-handed media, have , that are simulta-
neously negative,  < 0 and < 0. Veselago [391] was the rst to study their unusual The relative directions of the eld vectors are shown in Fig. 1.5.1. Each vector may
electromagnetic properties, such as having a negative index of refraction and the rever- be decomposed as the sum of a part tangential to the surface and a part perpendicular
sal of Snels law. to it, that is, E = Et + En . Using the vector identity,
The novel properties of such media and their potential applications have generated
a lot of research interest [391473]. Examples of such media, termed metamaterials, E = n (E n)+n(n E)= Et + En (1.5.3)
have been constructed using periodic arrays of wires and split-ring resonators, [397]
we identify these two parts as:
and by transmission line elements [430432,452,465], and have been shown to exhibit
the properties predicted by Veselago.
Et = n (E n) , En = n(n E)= nEn
When rel < 0 and rel < 0, the refractive index, n2 = rel rel , must be dened by

the negative square root n = rel rel . Because then n < 0 and rel < 0 will imply
that the characteristic impedance of the medium = 0 rel /n will be positive, which
as we will see later implies that the energy ux of a wave is in the same direction as the
direction of propagation. We discuss such media in Sections 2.13, 7.16, and 8.6.

1.5 Boundary Conditions

The boundary conditions for the electromagnetic elds across material boundaries are
given below:
Fig. 1.5.1 Field directions at boundary.

E1t E2t = 0 Using these results, we can write the rst two boundary conditions in the following
vectorial forms, where the second form is obtained by taking the cross product of the
H1t H2t = Js n
(1.5.1) rst with n and noting that Js is purely tangential:
D1n D2n = s
n (E1 n) n (E2 n) = 0 n (E1 E2 ) = 0
B1n B2n = 0 or, (1.5.4)
n (H1 n) n (H2 n) = Js n n (H1 H2 ) = Js
where n is a unit vector normal to the boundary pointing from medium-2 into medium-1.
The quantities s , Js are any external surface charge and surface current densities on The boundary conditions (1.5.1) can be derived from the integrated form of Maxwells
the boundary surface and are measured in units of [coulomb/m2 ] and [ampere/m]. equations if we make some additional regularity assumptions about the elds at the
In words, the tangential components of the E-eld are continuous across the inter- interfaces.
face; the difference of the tangential components of the H-eld are equal to the surface
1.6. Currents, Fluxes, and Conservation Laws 9 10 1. Maxwells Equations

In many interface problems, there are no externally applied surface charges or cur-
rents on the boundary. In such cases, the boundary conditions may be stated as:

E1t = E2t

H1t = H2t
(source-free boundary conditions) (1.5.5)
D1n = D2n Fig. 1.6.1 Flux of a quantity.

B1n = B2n
Similarly, when J represents momentum ux, we expect to have Jmom = cmom .
Momentum ux is dened as Jmom = p/(St)= F/S, where p denotes momen-
1.6 Currents, Fluxes, and Conservation Laws tum and F = p/t is the rate of change of momentum, or the force, exerted on the
surface S. Thus, Jmom represents force per unit area, or pressure.
The electric current density J is an example of a ux vector representing the ow of the
Electromagnetic waves incident on material surfaces exert pressure (known as ra-
electric charge. The concept of ux is more general and applies to any quantity that
diation pressure), which can be calculated from the momentum ux vector. It can be
ows. It could, for example, apply to energy ux, momentum ux (which translates
shown that the momentum ux is numerically equal to the energy density of a wave, that
into pressure force), mass ux, and so on.
is, Jmom = en , which implies that en = mom c. This is consistent with the theory of
In general, the ux of a quantity Q is dened as the amount of the quantity that
relativity, which states that the energy-momentum relationship for a photon is E = pc.
ows (perpendicularly) through a unit surface in unit time. Thus, if the amount Q
ows through the surface S in time t, then:

Q 1.7 Charge Conservation

J= (denition of ux) (1.6.1)
St
Maxwell added the displacement current term to Amperes law in order to guarantee
When the owing quantity Q is the electric charge, the amount of current through charge conservation. Indeed, taking the divergence of both sides of Amperes law and
the surface S will be I = Q/t, and therefore, we can write J = I/S, with units using Gausss law D = , we get:
of [ampere/m2 ].
The ux is a vectorial quantity whose direction points in the direction of ow. There D
H = J+ =J+ D=J+
is a fundamental relationship that relates the ux vector J to the transport velocity v t t t
and the volume density of the owing quantity:
H = 0, we obtain the differential form of the charge
Using the vector identity
J = v (1.6.2) conservation law:

+ J = 0 (charge conservation) (1.7.1)
This can be derived with the help of Fig. 1.6.1. Consider a surface S oriented per- t
pendicularly to the ow velocity. In time t, the entire amount of the quantity contained
Integrating both sides over a closed volume V surrounded by the surface S, as
in the cylindrical volume of height vt will manage to ow through S. This amount is
shown in Fig. 1.7.1, and using the divergence theorem, we obtain the integrated form of
equal to the density of the material times the cylindrical volume V = S(vt), that
Eq. (1.7.1): 

is, Q = V = S vt. Thus, by denition:

d
J dS = dV (1.7.2)
Q S vt S dt V
J= = = v
St St The left-hand side represents the total amount of charge owing outwards through
When J represents electric current density, we will see in Sec. 1.12 that Eq. (1.6.2) the surface S per unit time. The right-hand side represents the amount by which the
implies Ohms law J = E. When the vector J represents the energy ux of a propagating charge is decreasing inside the volume V per unit time. In other words, charge does not
electromagnetic wave and the corresponding energy per unit volume, then because the disappear into (or created out of) nothingnessit decreases in a region of space only
speed of propagation is the velocity of light, we expect that Eq. (1.6.2) will take the form: because it ows into other regions.
Another consequence of Eq. (1.7.1) is that in good conductors, there cannot be any
Jen = cen (1.6.3) accumulated volume charge. Any such charge will quickly move to the conductors
In this sense, the terms electric and magnetic ux densities for the quantities D, B are somewhat of a surface and distribute itself such that to make the surface into an equipotential surface.
misnomer because they do not represent anything that ows.
1.8. Energy Flux and Energy Conservation 11 12 1. Maxwells Equations

where we introduce a change in notation:

1 1
en = w = |E|2 + |H|2 = energy per unit volume
2 2 (1.8.2)
Jen = P = E H = energy ux or Poynting vector

where |E|2 = E E . The quantities w and P are measured in units of [joule/m3 ] and
[watt/m2 ]. Using the identity (E H)= H E E H, we nd:
Fig. 1.7.1 Flux outwards through surface.
w E H
+ P =  E+ H + (E H)
t t t
Assuming that inside the conductor we have D = E and J = E, we obtain
D B
= E+ H+H EE H
t t
E= D=
J =    
  D B
= H E+ + E H
t t

+ =0 (1.7.3) Using Amperes and Faradays laws, the right-hand side becomes:
t 
with solution:
w
(r, t)= 0 (r)et/ + P = J E (energy conservation) (1.8.3)
t
where 0 (r) is the initial volume charge distribution. The solution shows that the vol-
As we discussed in Eq. (1.2.6), the quantity JE represents the ohmic losses, that
ume charge disappears from inside and therefore it must accumulate on the surface of
is, the power per unit volume lost into heat from the elds. The integrated form of
the conductor. The relaxation time constant rel = / is extremely short for good
Eq. (1.8.3) is as follows, relative to the volume and surface of Fig. 1.7.1:
conductors. For example, in copper,


d
 8.85 1012 P dS = w dV + J E dV (1.8.4)
rel = = = 1.6 1019 sec S dt V V
5.7 107
It states that the total power entering a volume V through the surface S goes partially
By contrast, rel is of the order of days in a good dielectric. For good conductors, the into increasing the eld energy stored inside V and partially is lost into heat.
above argument is not quite correct because it is based on the steady-state version of
Ohms law, J = E, which must be modied to take into account the transient dynamics Example 1.8.1: Energy concepts can be used to derive the usual circuit formulas for capaci-
of the conduction charges. tance, inductance, and resistance. Consider, for example, an ordinary plate capacitor with
It turns out that the relaxation time rel is of the order of the collision time, which plates of area A separated by a distance l, and lled with a dielectric . The voltage between
the plates is related to the electric eld between the plates via V = El.
is typically 1014 sec. We discuss this further in Sec. 1.13. See also Refs. [147150].
The energy density of the electric eld between the plates is w = E2 /2. Multiplying this
by the volume between the plates, Al, will give the total energy stored in the capacitor.
1.8 Energy Flux and Energy Conservation Equating this to the circuit expression CV2 /2, will yield the capacitance C:

1 2 1 1 A
Because energy can be converted into different forms, the corresponding conservation W= E Al = CV2 = CE2 l2 C=
2 2 2 l
equation (1.7.1) should have a non-zero term in the right-hand side corresponding to
the rate by which energy is being lost from the elds into other forms, such as heat. Next, consider a solenoid with n turns wound around a cylindrical iron core of length
Thus, we expect Eq. (1.7.1) to have the form: l, cross-sectional area A, and permeability . The current through the solenoid wire is
related to the magnetic eld in the core through Amperes law Hl = nI. It follows that the
en stored magnetic energy in the solenoid will be:
+ Jen = rate of energy loss (1.8.1)
t 1 1 1 H2 l2 A
W= H2 Al = LI2 = L 2 L = n2
Assuming the ordinary constitutive relations D = E and B = H, the quantities 2 2 2 n l
en , Jen describing the energy density and energy ux of the elds are dened as follows, Finally, consider a resistor of length l, cross-sectional area A, and conductivity . The
voltage drop across the resistor is related to the electric eld along it via V = El. The
1.9. Harmonic Time Dependence 13 14 1. Maxwells Equations

current is assumed to be uniformly distributed over the cross-section A and will have Next, we review some conventions regarding phasors and time averages. A real-
density J = E. valued sinusoid has the complex phasor representation:
The power dissipated into heat per unit volume is JE = E2 . Multiplying this by the
resistor volume Al and equating it to the circuit expression V2 /R = RI2 will give: A(t)= |A| cos(t + )  A(t)= Aejt (1.9.3)
   
V2 E 2 l2 1 l where A = |A|ej . Thus, we have A(t)= Re A(t) = Re Aejt . The time averages of
(J E)(Al)= E2 (Al)= = R=
R R A the quantities A(t) and A(t) over one period T = 2/ are zero.
 
The time average of the product of two harmonic quantities A(t)= Re Aejt and
The same circuit expressions can, of course, be derived more directly using Q = CV, the  jt 
magnetic ux = LI, and V = RI. 
 B(t)= Re Be with phasors A, B is given by (see Problem 1.4):

T
Conservation laws may also be derived for the momentum carried by electromagnetic 1 1 
A(t)B(t) = A(t)B(t) dt = Re AB ] (1.9.4)
elds [41,1293]. It can be shown (see Problem 1.6) that the momentum per unit volume T 0 2
carried by the elds is given by: In particular, the mean-square value is given by:

T
1 1 1 1  1
G=DB= 2
EH= P (momentum density) (1.8.5) A2 (t) = A2 (t) dt = Re AA ]= |A|2 (1.9.5)
c c2 T 0 2 2
Some interesting time averages in electromagnetic wave problems are the time av-
where we set D = E, B = H, and c = 1/ . The quantity Jmom = cG = P /c will
represent momentum ux, or pressure, if the elds are incident on a surface. erages of the energy density, the Poynting vector (energy ux), and the ohmic power
losses per unit volume. Using the denition (1.8.2) and the result (1.9.4), we have for
these time averages:
1.9 Harmonic Time Dependence  
1 1 1
w= Re E E + H H (energy density)
Maxwells equations simplify considerably in the case of harmonic time dependence. 2 2 2
Through the inverse Fourier transform, general solutions of Maxwells equation can be 1 

P = Re E H (Poynting vector) (1.9.6)
built as linear combinations of single-frequency solutions: 2

dPloss 1  
d = Re Jtot E (ohmic losses)
E(r, t)= E(r, )ejt (1.9.1) dV 2
2
where Jtot = J + jD is the total current in the right-hand side of Amperes law and
Thus, we assume that all elds have a time dependence ejt : accounts for both conducting and dielectric losses. The time-averaged version of Poynt-
ings theorem is discussed in Problem 1.5.
E(r, t)= E(r)ejt , H(r, t)= H(r)ejt
The expression (1.9.6) for the energy density w was derived under the assumption
where the phasor amplitudes E(r), H(r) are complex-valued. Replacing time derivatives that both  and were constants independent of frequency. In a dispersive medium, ,
by t j, we may rewrite Eq. (1.1.1) in the form: become functions of frequency. In frequency bands where (), () are essentially
real-valued, that is, where the medium is lossless, it can be shown [168] that the time-
E = jB averaged energy density generalizes to:
 
H = J + jD 1 1 d() 1 d()
w= Re E E + H H (lossless case) (1.9.7)
(Maxwells equations) (1.9.2) 2 2 d 2 d
D=
The derivation of (1.9.7) is as follows. Starting with Maxwells equations (1.1.1) and
B=0 without assuming any particular constitutive relations, we obtain:

In this book, we will consider the solutions of Eqs. (1.9.2) in three different contexts: E H = E D H B J E (1.9.8)
(a) uniform plane waves propagating in dielectrics, conductors, and birefringent me-
dia, (b) guided waves propagating in hollow waveguides, transmission lines, and optical As in Eq. (1.8.3), we would like to interpret the rst two terms in the right-hand side
bers, and (c) propagating waves generated by antennas and apertures. as the time derivative of the energy density, that is,
The ejt convention is used in the engineering literature, and eit in the physics literature. One can
dw
pass from one convention to the other by making the formal substitution j i in all the equations. = E D + H B
dt
1.9. Harmonic Time Dependence 15 16 1. Maxwells Equations

Anticipating a phasor-like representation, we may assume complex-valued elds and Because we assume that () is real (i.e., lossless) in the vicinity of 0 , it follows that:
derive also the following relationship from Maxwells equations: 1   1   1  
    1   1   Re E D = Re E 0 (t) ja0 E 0 (t)+b0 E 0 (t) = b0 Re E 0 (t) E 0 (t) , or,
1 1 2 2 2
Re E H = Re E D Re H B Re J E (1.9.9)  
2 2 2 2    
1   d 1 d 1 d () 0
from which we may identify a time-averaged version of dw/dt, Re E D = b0 |E 0 (t)|2 = |E 0 (t)|2 (1.9.18)
2 dt 4 dt 4 d
dw 1   1  
= Re E D + Re H B (1.9.10) Dropping the subscript 0, we see that the quantity under the time derivative in the
dt 2 2 right-hand side may be interpreted as a time-averaged energy density for the electric
In a dispersive dielectric, the constitutive relation between D and E can be written eld. A similar argument can be given for the magnetic energy term of Eq. (1.9.7).
as follows in the time and frequency domains: We will see in the next section that the energy density (1.9.7) consists of two parts:

one part is the same as that in the vacuum case; the other part arises from the kinetic
D(t)= (t t )E(t )dt  D()= ()E() (1.9.11) and potential energy stored in the polarizable molecules of the dielectric medium.

When Eq. (1.9.7) is applied to a plane wave propagating in a dielectric medium, one
where the Fourier transforms are dened by


can show that (in the lossless case) the energy velocity coincides with the group velocity.
1
(t)= ()ejt d  ()= (t)ejt dt (1.9.12) The generalization of these results to the case of a lossy medium has been studied
2 extensively [168182]. Eq. (1.9.7) has also been applied to the case of a left-handed
The time-derivative of D(t) is then medium in which both () and () are negative over certain frequency ranges. As

argued by Veselago [391], such media must necessarily be dispersive in order to make
D(t)= (t t )E(t )dt (1.9.13) Eq. (1.9.7) a positive quantity even though individually  and are negative.

Analogous expressions to (1.9.7) may also be derived for the momentum density of
where it follows from Eq. (1.9.12) that

a wave in a dispersive medium. In vacuum, the time-averaged momentum density is
1 given by Eq. (1.8.5), that is,
(t)= j()ejt d (1.9.14)  
2 1
G = Re 0 0 E H
Following [168], we assume a quasi-harmonic representation for the electric eld, 2
E(t)= E 0 (t)ej0 t , where E 0 (t) is a slowly-varying function of time. Equivalently, in the For the dispersive (and lossless) case this generalizes to [391,467]
  
frequency domain we have E()= E 0 ( 0 ), assumed to be concentrated in a small 1 k d d
G = Re E H + |E|2 + |H|2 (1.9.19)
neighborhood of 0 , say, | 0 | . Because () multiplies the narrowband 2 2 d d
function E(), we may expand () in a Taylor series around 0 and keep only the
linear terms, that is, inside the integral (1.9.14), we may replace:
 
1.10 Simple Models of Dielectrics, Conductors, and Plasmas
d () 

()= a0 + b0 ( 0 ) , a0 = 0 (0 ) , b0 =  (1.9.15)
d  A simple model for the dielectric properties of a material is obtained by considering the
0
motion of a bound electron in the presence of an applied electric eld. As the electric
Inserting this into Eq. (1.9.14), we obtain the approximation eld tries to separate the electron from the positively charged nucleus, it creates an

1   electric dipole moment. Averaging this dipole moment over the volume of the material
(t) ja0 + b0 (j j0 ) ejt d = ja0 (t)+b0 (t j0 )(t) (1.9.16)
2 gives rise to a macroscopic dipole moment per unit volume.
where (t) the Dirac delta function. This approximation is justied only insofar as it is A simple model for the dynamics of the displacement x of the bound electron is as
used inside Eq. (1.9.13). Inserting (1.9.16) into Eq. (1.9.13), we nd follows (with x = dx/dt):

mx = eE kx mx (1.10.1)
 
D(t) = ja0 (t t )+b0 (t j0 )(t t ) E(t )dt = where we assumed that the electric eld is acting in the x-direction and that there is

a spring-like restoring force due to the binding of the electron to the nucleus, and a
= ja0 E(t)+b0 (t j0 )E(t) friction-type force proportional to the velocity of the electron.
(1.9.17)
 The spring constant k is related to the resonance frequency of the spring via the
= ja0 E 0 (t)ej0 t + b0 (t j0 ) E 0 (t)ej0 t
relationship 0 = k/m, or, k = m20 . Therefore, we may rewrite Eq. (1.10.1) as
 
= ja0 E 0 (t)+b0 E 0 (t) ej0 t
e
To
x + x + 20 x = E (1.10.2)
unclutter the notation, we are suppressing the dependence on the space coordinates r. m
1.11. Dielectrics 17 18 1. Maxwells Equations

The limit 0 = 0 corresponds to unbound electrons and describes the case of good The electric ux density will be then:
conductors. The frictional term x arises from collisions that tend to slow down the 
electron. The parameter is a measure of the rate of collisions per unit time, and D = 0 E + P = 0 1 + () E ()E
therefore, = 1/ will represent the mean-time between collisions.
In a typical conductor, is of the order of 1014 seconds, for example, for copper, where the effective permittivity () is:
= 2.4 1014 sec and = 4.1 1013 sec1 . The case of a tenuous, collisionless,
Ne2
plasma can be obtained in the limit = 0. Thus, the above simple model can describe
()= 0 + 2 m (1.11.4)
the following cases:
0 2 + j
a. Dielectrics, 0 = 0, = 0. This can be written in a more convenient form, as follows:
b. Conductors, 0 = 0, = 0.
c. Collisionless Plasmas, 0 = 0, = 0. 0 2p
()= 0 + (1.11.5)
20 2 + j
The basic idea of this model is that the applied electric eld tends to separate positive
from negative charges, thus, creating an electric dipole moment. In this sense, the
where 2p is the so-called plasma frequency of the material dened by:
model contains the basic features of other types of polarization in materials, such as
ionic/molecular polarization arising from the separation of positive and negative ions
by the applied eld, or polar materials that have a permanent dipole moment. Ne2
2p = (plasma frequency) (1.11.6)
0 m

1.11 Dielectrics The model dened by (1.11.5) is known as a Lorentz dielectric. The corresponding

susceptibility, dened through ()= 0 1 + () , is:
The applied electric eld E(t) in Eq. (1.10.2) can have any time dependence. In particular,
if we assume it is sinusoidal with frequency , E(t)= Eejt , then, Eq. (1.10.2) will have 2p
()= (1.11.7)
the solution x(t)= xejt , where the phasor x must satisfy: 20 2 + j
e For a dielectric, we may assume 0 = 0. Then, the low-frequency limit ( = 0) of
2 x + jx + 20 x = E
m Eq. (1.11.5), gives the nominal dielectric constant:
which is obtained by replacing time derivatives by t j. Its solution is:
2p Ne2
(0)= 0 + 0 = 0 + (1.11.8)
e 20 m20
E
x= 2 m (1.11.1)
0 + j
2 The real and imaginary parts of () characterize the refractive and absorptive
properties of the material. By convention, we dene the imaginary part with the negative
The corresponding velocity of the electron will also be sinusoidal v(t)= vejt , where sign (because we use ejt time dependence):
v = x = jx. Thus, we have:
()=  ()j () (1.11.9)
e
j E
v = jx = 2 m (1.11.2) It follows from Eq. (1.11.5) that:
0 2 + j
0 2p (20 2 ) 0 2p
From Eqs. (1.11.1) and (1.11.2), we can nd the polarization per unit volume P.  ()= 0 + ,  ()= (1.11.10)
(2 20 )2 +2 2 (2 20 )2 +2 2
We assume that there are N such elementary dipoles per unit volume. The individual
electric dipole moment is p = ex. Therefore, the polarization per unit volume will be: Fig. 1.11.1 shows a plot of  () and  (). Around the resonant frequency 0 ,
the real part  () behaves in an anomalous manner, that is, it drops rapidly with
Ne2
E frequency to values less than 0 and the material exhibits strong absorption. The term
P = Np = Nex = 2 m 0 ()E (1.11.3) normal dispersion refers to an  () that is an increasing function of , as is the
0 + j
2
case to the far left and right of the resonant frequency.
1.11. Dielectrics 19 20 1. Maxwells Equations

Fig. 1.11.1 Real and imaginary parts of the effective permittivity (). Fig. 1.11.2 Effective permittivity in a two-level gain medium with f = 1.

Real dielectric materials exhibit, of course, several such resonant frequencies cor- In practice, Eq. (1.11.14) is applied in frequency ranges that are far from any reso-
responding to various vibrational modes and polarization mechanisms (e.g., electronic, nance so that one can effectively set i = 0:
ionic, etc.) The permittivity becomes the sum of such terms:
 Bi 2i  Bi 2
 2 n2 ()= 1 + =1+ (Sellmeier equation) (1.11.15)
Ni ei /mi 0 2
i 2 2 2i
()= 0 + 0 2 (1.11.11) i i
i
i 2 + ji
where , i denote the corresponding free-space wavelengths (e.g., = 2c/). In
A more correct quantum-mechanical treatment leads essentially to the same formula: practice, refractive index data are tted to Eq. (1.11.15) using 24 terms over a desired
 fji (Ni Nj )e2 /m0 frequency range. For example, fused silica (SiO2 ) is very accurately represented over the
()= 0 + 0 (1.11.12) range 0.2 3.7 m by the following formula [156], where and i are in units of
2ji 2 + jji
j>i m:
where ji are transition frequencies between energy levels, that is, ji = (Ej Ei )/, 0.6961663 2 0.4079426 2 0.8974794 2
and Ni , Nj are the populations of the lower, Ei , and upper, Ej , energy levels. The quan- n2 = 1 + + 2 + 2 (1.11.16)
2 (0.0684043)2 (0.1162414) 2 (9.896161)2
tities fji are called oscillator strengths. For example, for a two-level atom we have:

f 2p 1.12 Conductors
()= 0 + 0 (1.11.13)
20 2 + j
The conductivity properties of a material are described by Ohms law, Eq. (1.3.15). To
where we dened: derive this law from our simple model, we use the relationship J = v, where the volume
density of the conduction charges is = Ne. It follows from Eq. (1.11.2) that
N1 N2 (N1 + N2 )e2
0 = 21 , f = f21 , 2p =
N1 + N2 m0 Ne2
E
j
Normally, lower energy states are more populated, Ni > Nj , and the material behaves J = v = Nev = 2 m ()E
0 2 + j
as a classical absorbing dielectric. However, if there is population inversion, Ni < Nj ,
then the corresponding permittivity term changes sign. This leads to a negative imag- and therefore, we identify the conductivity ():
inary part,  (), representing a gain. Fig. 1.11.2 shows the real and imaginary parts
of Eq. (1.11.13) for the case of a negative effective oscillator strength f = 1. Ne2
j j0 2p
The normal and anomalous dispersion bands still correspond to the bands where ()= 2 m = 2 (1.12.1)
the real part  () is an increasing or decreasing, respectively, function of frequency. 0 + j
2 0 2 + j
But now the normal behavior is only in the neighborhood of the resonant frequency,
We note that ()/j is essentially the electric susceptibility considered above.
whereas far from it,the behavior is anomalous.
Indeed, we have J = Nev = Nejx = jP, and thus, P = J/j = (()/j)E. It
Setting n()= ()/0 for the refractive index, Eq. (1.11.11) can be written in the
follows that ()0 = ()/j, and
following form, known as the Sellmeier equation (where the Bi are constants):

 Bi 2i 0 2p ()
n2 ()= 1 + (1.11.14) ()= 0 + = 0 + (1.12.2)
2i 2 + ji 20 2 + j j
i
1.12. Conductors 21 22 1. Maxwells Equations

Since in a metal the conduction charges are unbound, we may take 0 = 0 in where u(t) is the unit-step function. As an example, suppose the electric eld E(t) is a
Eq. (1.12.1). After canceling a common factor of j , we obtain: constant electric eld that is suddenly turned on at t = 0, that is, E(t)= Eu(t). Then,
the time response of the current will be:
0 2p
t
()= (1.12.3)  0 2p  
+ j J(t)= 0 2p e(tt ) Edt = E 1 et = E 1 et
0
The model dened by (1.12.3) is know as the Drude model. The nominal conduc- where = 0 2p / is the nominal conductivity of the material.
tivity is obtained at the low-frequency limit, = 0: Thus, the current starts out at zero and builds up to the steady-state value of J = E,
which is the conventional form of Ohms law. The rise time constant is = 1/. We
0 2p Ne2 saw above that is extremely smallof the order of 1014 secfor good conductors.
= = (nominal conductivity) (1.12.4)
m The building up of the current can also be understood in terms of the equation of
motion of the conducting charges. Writing Eq. (1.10.2) in terms of the velocity of the
charge, we have:
Example 1.12.1: Copper has a mass density of 8.9 106 gr/m3 and atomic weight of 63.54 e
v(t)+v(t)= E(t)
(grams per mole.) Using Avogadros number of 6 1023 atoms per mole, and assuming m
one conduction electron per atom, we nd for the volume density N: Assuming E(t)= Eu(t), we obtain the convolutional solution:

t
6 1023
atoms  e e 
mole 8.9 106 gr  1 electron  = 8.4 1028 electrons/m3 v(t)= e(tt ) E(t )dt = E 1 et
N= gr 0 m m
63.54 m3 atom
mole
For large t, the velocity reaches the steady-state value v = (e/m)E, which reects
It follows that: the balance between the accelerating electric eld force and the retarding frictional force,
that is, mv = eE. The quantity e/m is called the mobility of the conduction charges.
Ne2 (8.4 1028 )(1.6 1019 )2 The steady-state current density results in the conventional Ohms law:
= = = 5.8 107 Siemens/m
m (9.1 1031 )(4.1 1013 )
Ne2
19 31 J = Nev = E = E
where we used e = 1.6 10 , m = 9.1 10 , = 4.1 10 . The plasma frequency
13
m
of copper can be calculated by
A more accurate description of the permittivity properties of metals, especially at

p 1 Ne2 optical and infrared frequencies which are relevant in plasmonic waveguides, requires
fp = = = 2.6 1015 Hz
2 2 m0 the addition of interband terms, generalizing the Drude model to the so-called Drude-
Lorentz model of the form,
which lies in the ultraviolet range. For frequencies such that  , the conductivity
(1.12.3) may be considered to be independent of frequency and equal to the dc value of () f0 2p 
k
fi 2p
Eq. (1.12.4). This frequency range covers most present-day RF applications. For example, =1+ + (Drude-Lorentz model) (1.12.7)
0 j( + j) i=1 i 2 + ji
2
assuming 0.1, we nd f 0.1/2 = 653 GHz. 


Rakic, et al. [163] have tted 11 metals, such as silver, gold, aluminum, copper,
So far, we assumed sinusoidal time dependence and worked with the steady-state
to such an expression with 56 terms, covering a wide range of frequencies and wave-
responses. Next, we discuss the transient dynamical response of a conductor subject to
lengths, 25 THz < f < 1500 THz, or, equivalently, 200 nm < < 12 m. The MATLAB
an arbitrary time-varying electric eld E(t).
function, drude, implements the results of [163],
Ohms law can be expressed either in the frequency-domain or in the time-domain
with the help of the Fourier transform pair of equations: ep = drude(lambda,metal) % Drude-Lorentz model for Silver, Gold, Copper, Aluminum

t
J()= ()E()  J(t)= (t t )E(t )dt (1.12.5) lambda = vector of wavelengths in nanometers
metal = s, g, c, a, for silver, gold, copper, aluminum
where (t) is the causal inverse Fourier transform of (). For the simple model of
ep = complex relative permittivity (same size as lambda)
Eq. (1.12.3), we have:
(t)= 0 2p et u(t) (1.12.6) For some improved models, see Refs. [164167].
1.13. Charge Relaxation in Conductors 23 24 1. Maxwells Equations

1.13 Charge Relaxation in Conductors Denoting the rst two terms by d () and the third by c ()/j, we obtain the
total effective permittivity of such a material:
Next, we discuss the issue of charge relaxation in good conductors [147150]. Writing
(1.12.5) three-dimensionally and using (1.12.6), Ohms law reads in the time domain: c ()
()= d ()+ (effective permittivity) (1.14.2)

t j
2 (tt )  
J(r, t)= p e 0 E(r, t ) dt (1.13.1)
In the low-frequency limit, = 0, the quantities d (0) and c (0) represent the
Taking the divergence of both sides and using charge conservation, J + = 0, nominal dielectric constant and conductivity of the material. We note also that we can
and Gausss law, 0 E = , we obtain the following integro-differential equation for write Eq. (1.14.2) in the form:
the charge density (r, t):

t
t j()= c ()+jd () (1.14.3)
 
(r, t)= J(r, t)= 2p e(tt ) 0 E(r, t )dt = 2p e(tt ) (r, t )dt These two terms characterize the relative importance of the conduction current and

the displacement (polarization) current. The right-hand side in Amperes law gives the
Differentiating both sides with respect to t, we nd that satises the second-order total effective current:
differential equation:
D
(r, t)+(r, t)+2p (r, t)= 0 (1.13.2) Jtot = J + = J + jD = c ()E + jd ()E = j()E
t
whose solution is easily veried to be a linear combination of:
 where the term Jdisp = D/t = jd ()E represents the displacement current. The
2 relative strength between conduction and displacement currents is the ratio:
et/2 cos(relax t) , et/2 sin(relax t) , where relax = 2p  
4 J 
 cond  |c ()E| |c ()|
 = = (1.14.4)
Thus, the charge density is an exponentially decaying sinusoid with a relaxation time  Jdisp  |jd ()E| |d ()|
constant that is twice the collision time = 1/:
This ratio is frequency-dependent and establishes a dividing line between a good
2 conductor and a good dielectric. If the ratio is much larger than unity (typically, greater
relax = = 2 (relaxation time constant) (1.13.3) than 10), the material behaves as a good conductor at that frequency; if the ratio is much

smaller than one (typically, less than 0.1), then the material behaves as a good dielectric.
Typically, p  , so that relax is practically equal to p . For example, using
Example 1.14.1: This ratio can take a very wide range of values. For example, assuming a
the numerical data of Example 1.12.1, we nd for copper relax = 2 = 51014 sec. frequency of 1 GHz and using (for illustration purposes) the dc-values of the dielectric
We calculate also: frelax = relax /2 = 2.61015 Hz. In the limit , or 0, constants and conductivities, we nd:
Eq. (1.13.2) reduces to the naive relaxation equation (1.7.3) (see Problem 1.9).

In addition to charge relaxation, the total relaxation time depends on the time it takes   9
for copper with = 5.8107 S/m and  = 0
J  10
 cond 
for the electric and magnetic elds to be extinguished from the inside of the conductor,  = = 1 for seawater with = 4 S/m and  = 720
 Jdisp   9
as well as the time it takes for the accumulated surface charge densities to settle, the 10 for a glass with = 1010 S/m and  = 20
motion of the surface charges being damped because of ohmic losses. Both of these
times depend on the geometry and size of the conductor [149]. Thus, the ratio varies over 18 orders of magnitude! If the frequency is reduced by a factor
of ten to 100 MHz, then all the ratios get multiplied by 10. In this case, seawater acts like
a good conductor. 

1.14 Power Losses
The time-averaged ohmic power losses per unit volume within a lossy material are
To describe a material with both dielectric and conductivity properties, we may take the given by Eq. (1.9.6). Writing ()=  ()j (), we have:
susceptibility to be the sum of two terms, one describing bound polarized charges and
the other unbound conduction charges. Assuming different parameters {0 , p , } for Jtot = j()E = j ()E +  ()E
each term, we obtain the total permittivity:  
Denoting E  = E E , it follows that:
2

0 2dp 0 2cp   1  2
()= 0 + + dPloss 1
2
d0 2 + jd j(c + j)
(1.14.1) = Re Jtot E =  ()E  (ohmic losses) (1.14.5)
dV 2 2
1.15. Plasmas 25 26 1. Maxwells Equations

Writing d ()= d ()j

d () and assuming that the conductivity c () is real-
valued for the frequency range of interest (as was discussed in Example 1.12.1), we nd

k = ()
by equating real and imaginary parts of Eq. (1.14.2):
It follows that for a plasma:
c () 
 ()= d () ,  ()= 
d ()+ (1.14.6)  1
k = 0 0 1 2p /2 = 2 2p (1.15.2)
c
Then, the power losses can be written in a form that separates the losses due to
conduction and those due to the polarization properties of the dielectric: where we used c = 1/ 0 0 .
If > p , the electromagnetic wave propagates without attenuation within the
dPloss 1  2 plasma. But if < p , the wavenumber k becomes imaginary and the wave gets
= c ()+ E 
d () (ohmic losses) (1.14.7)
attenuated. At such frequencies, a wave incident (normally) on the ionosphere from the
dV 2
A convenient way to quantify the losses is by means of the loss tangent dened in ground cannot penetrate and gets reected back.
terms of the real and imaginary parts of the effective permittivity:

1.16 Energy Density in Lossless Dispersive Dielectrics

 ()
tan = (loss tangent) (1.14.8)
 () The lossless case is obtained from Eq. (1.11.5) by setting = 0, which is equivalent to
where is the loss angle. Eq. (1.14.8) may be written as the sum of two loss tangents, assuming that is far from the resonance 0 . In this case the permittivity is:
 
one due to conduction and one due to polarization. Using Eq. (1.14.6), we have: 2p
()= 0 1 +
20 2
c ()+
d () c ()  () from which it follows that:
tan =  =  + d = tan c + tan d (1.14.9)  
d () d () d () 2p (2 + 20 )
d()
= 0 1 + (1.16.1)
The ohmic loss per unit volume can be expressed in terms of the loss tangent as: d (20 2 )2

 2 Thus, the electric part of the energy density (1.9.7) will be:
dPloss 1
= d ()tan E  (ohmic losses) (1.14.10)  
dV 2 1 d() 1 2p (2 + 20 )
we = |E|2 = 0 |E|2 1 + (1.16.2)
4 d 4 (20 2 )2
1.15 Plasmas This expression can be given a nice interpretation: The rst term on the right is the
energy density in vacuum and the second corresponds to the mechanical (kinetic and
To describe a collisionless plasma, such as the ionosphere, the simple model consid- potential) energy of the polarization charges [169,192]. Indeed, the displacement x and
ered in the previous sections can be specialized by choosing 0 = = 0. Thus, the velocity v = x of the polarization charges are in this case:
conductivity given by Eq. (1.12.3) becomes pure imaginary:
eE/m
x= , v = jx
0 2p 20 2
()=
j The time-averaged mechanical energy (per unit volume) is obtained by adding the
kinetic and potential energies:
The corresponding effective permittivity of Eq. (1.12.2) becomes purely real:   
1 1 1 1
  wmech = Re N m|v|2 + m20 |x|2 = Nm(2 + 20 )|x|2
() p2 2 2 2 4
()= 0 + = 0 1 2 (1.15.1)  
j 1 Nm(2 + 20 )e2 |E|2 /m2 1 2p (2 + 20 )
= = 0 |E|2
4 (20 2 )2 4 (20 2 )2
The plasma frequency can be calculated from p 2
= Ne2 /m0 . In the ionosphere
the electron density is typically N = 1012 , which gives fp = 9 MHz. where we used the denition (1.11.6) of the plasma frequency. It follows that Eq. (1.16.2)
We will see in Sec. 2.6 that the propagation wavenumber of an electromagnetic wave can be written as the sum:
propagating in a dielectric/conducting medium is given in terms of the effective permit- 1 d() 1
we = |E|2 = 0 |E|2 + wmech = wvac + wmech (1.16.3)
tivity by: 4 d 4
1.17. Kramers-Kronig Dispersion Relations 27 28 1. Maxwells Equations

1.17 Kramers-Kronig Dispersion Relations Rearranging terms and canceling a factor of 1/2, we obtain the Kramers-Kronig re-
lation in its complex-valued form:
The convolutional form of Eq. (1.3.13) implies causality, that is, the value of D(r, t) at

1 ( )
the present time t depends only on the past values of E(r, t ), t t. ()= P d (Kramers-Kronig) (1.17.7)
This condition is equivalent to requiring that the dielectric response (t) be a right- j 
sided (causal) function of time, that is, (t)= 0 for t < 0. Then, Eq. (1.3.13) may be The reason for applying this relation to () instead of () is that () falls off
written as ordinary convolution by extending the integration range over all times: sufciently fast for large to make the integral in (1.17.5) convergent, whereas ()

t
tends to the constant 0 .
D(r, t)= (t t )E(r, t )dt = (t t )E(r, t )dt Setting ()= r ()ji () and separating (1.17.7) into its real and imaginary

parts, we obtain the conventional form of the Kramers-Kronig dispersion relations:
Because D(r, t)= 0 E(r, t)+P(r, t), we may dene the time-domain susceptibility

function (t) through: 1 i ( )

r () = P d
(t)= 0 (t)+0 (t) (1.17.1) 

(Kramers-Kronig relations) (1.17.8)
where (t) is the Dirac delta function. Therefore, if (t) is causal, so is (t). The 1 r ( )
i () = P d
polarization is then given by: 

t

P(r, t)= 0 (t t )E(r, t )dt = 0 (t t )E(r, t )dt (1.17.2) Because the time-response (t) is real-valued, its Fourier transform () will sat-
isfy the Hermitian symmetry property ()= (), which is equivalent to the even
In the frequency domain, this becomes multiplicative: P(r, )= 0 ()E(r, ). symmetry of its real part, r ()= r (), and the odd symmetry of its imaginary part,
The Kramers-Kronig relations are the frequency-domain expression of causality and re- i ()= i (). Taking advantage of these symmetries, the range of integration in
late the real and imaginary parts of the susceptibility function (). Here, the functions (1.17.8) can be folded in half resulting in:
(t) and () are Fourier transform pairs:

2  i ( )


r () = P d
1 0 2 2
()= (t)ejt dt  (t)= ()ejt d (1.17.3)
(1.17.9)
2 2 r ( )
i () = P d
The causality condition, (t)= 0 for t < 0, can be expressed in terms of the unit-step 0 2 2
function u(t) in the equivalent manner: There are several other ways to prove the Kramers-Kronig relations. For example,
a more direct way is to state the causality condition in terms of the signum function
(t)= (t)u(t) , for all t (1.17.4) 
sign(t). Indeed, because u(t)= 1 + sign(t) /2, Eq. (1.17.4) may be written in the
Using the property that the Fourier transform of a product of two time functions is equivalent form (t)= (t)sign(t). Then, Eq. (1.17.7) follows by applying the same
the convolution of their Fourier transforms, it follows that Eq. (1.17.4) can be written in frequency-domain convolution argument using the Fourier transform pair:
the equivalent frequency-domain form:
2

sign(t)  P (1.17.10)
1    j
()= ( )U( )d (1.17.5)
2
Alternatively, the causality condition can be expressed as u(t)(t)= 0. This ap-
where U() is the Fourier transform of the unit-step. Eq. (1.17.5) is essentially the proach is explored in Problem 1.12. Another proof is based on the analyticity properties
Kramers-Kronig relation. The function U() is given by the well-known expression: of (). Because of the causality condition, the Fourier integral in (1.17.3) can be re-
stricted to the time range 0 < t < :
1 1
U()= lim =P + () (1.17.6)

0+ j +  j
()= ejt (t)dt = ejt (t)dt (1.17.11)

where P denotes the principal value. Inserting (1.17.6) into (1.17.5), we have: 0

  This implies that () can be analytically continued into the lower half -plane,
1  1
() = ( ) P + ( ) d
 so that replacing by w = j with 0 still gives a convergent Fourier integral
2 j(  )
The right-hand side (without the j) in (1.17.7) is known as a Hilbert transform. Exchanging the roles


1 ( ) 1 of t and , such transforms, known also as 90o phase shifters, are used widely in signal processing for
= P d + ()
2j  2 generating single-sideband communications signals.
1.18. Group Velocity, Energy Velocity 29 30 1. Maxwells Equations

in Eq. (1.17.11). Any singularities in () lie in the upper-half plane. For

 example, the This form preserves the sign of i , that is, ni and i are both positive for absorbing
simple model of Eq. (1.11.7) has poles at = 0 + j/2, where 0 = 20 2 /4. media, or both negative for gain media. The following approximate solution is often
Next, we consider a clockwise closed contour C = C + C consisting of the real axis used, which can be justied whenever ||  1 (for example, in gases):
C and an innite semicircle C in the lower half-plane. Because () is analytic in the  1 1
region enclosed by C, Cauchys integral theorem implies that for any point w enclosed nr ()jnr ()= 1 + ()  1 + nr = 1 + r , ni = i (1.18.3)
2 2 2
by C, that is, lying in the lower half-plane, we must have:
We will see in Chap. 2 that a single-frequency uniform plane wave propagating, say,
in the positive z-direction, has a wavenumber k = n/c = (nr jni )/c j,
 where c is the speed of light in vacuum. Therefore, the wave will have a space-time
1 (w )
(w)= dw (1.17.12) dependence:
2j C w w
ej(tkz) = ej(t(j)z) = ez ej(tz) = eni z/c ej(tnr z/c) (1.18.4)

where the overall minus sign arises because C was taken to be clockwise. Assuming that
The real part nr denes the phase velocity of the wave, vp = / = c/nr , whereas
() falls off sufciently fast for large , the contribution of the innite semicircle
the imaginary part ni , or = ni /c, corresponds to attenuation or gain depending on
can be ignored, thus leaving only the integral over the real axis. Setting w = j and
the sign of ni or i .
taking the limit  0+, we obtain the identical relationship to Eq. (1.17.5):
When several such plane waves are superimposed to form a propagating pulse, we


will see in Sec. 3.5 that the peak of the pulse (i.e., the point on the pulse where all the
1 ( ) 1 1
()= lim 
d = ( ) lim d individual frequency components add up in phase), propagates with the so-called group
0+ 2j + j 2 0+ j(  )+
velocity dened by:
An interesting consequence of the Kramers-Kronig relations is that there cannot
exist a dielectric medium that is purely lossless, that is, such that i ()= 0 for all , d 1 c c
vg = = = = = group velocity (1.18.5)
because this would also require that r ()= 0 for all . d d d(nr ) dnr
nr +
However, in all materials, i () is signicantly non-zero only in the neighborhoods d d d
of the mediums resonant frequencies, as for example in Fig. 1.11.1. In the frequency A group refractive index may be dened through vg = c/ng , or, ng = c/vg :
bands that are sufciently far from the resonant bands, i () may be assumed to be
essentially zero. Such frequency bands are called transparency bands [168]. d(nr ) dnr dnr
ng = = nr + = nr = group refractive index (1.18.6)
d d d

1.18 Group Velocity, Energy Velocity where is the free-space wavelength related to by = 2c/, and we used the
differentiation property that d/d = d/d.
Assuming a nonmagnetic material ( = 0 ), a complex-valued refractive index may be Within an anomalous dispersion region, nr is decreasing rapidly with , that is,
dened by: dnr /d < 0, as in Fig. 1.11.1. This results in a group velocity vg , given by Eq. (1.18.5),

 () that may be larger than c or even negative. Such velocities are called superluminal.
n()= nr ()jni ()= 1 + () = (1.18.1) Light pulses propagating at superluminal group velocities are referred to as fast light
0
and we discuss them further in Sec. 3.9.
where nr , ni are its real and imaginary parts. Setting = r ji we have the condition
 Within a normal dispersion region (e.g., to the far left and far right of the resonant
nr jni = 1 + r ji . Upon squaring, this splits into the two real-valued equations frequency 0 in Fig. 1.11.1), nr is an increasing function of , dnr /d > 0, which
n2r n2i = 1 + r and 2nr ni = i , with solutions: results in vg < c. In specially engineered materials such as those exhibiting electro-
 1/2 magnetically induced transparency, the slope dnr /d may be made so steep that the
(1 + r )2 +2i + (1 + r ) resulting group velocity vg becomes extremely small, vg  c. This is referred to as
nr =
2 slow light.
(1.18.2) We close this section by showing that for lossless dispersive media, the energy ve-
 1/2
(1 + r )2 +2i (1 + r ) locity of a plane wave is equal to the group velocity dened by (1.18.5). This result is
i
ni = sign(i ) = quite general, regardless of the frequency dependence of () and () (as long as
2 2nr
these quantities are real.)
1.19. Problems 31 32 1. Maxwells Equations

We will see in the next chapter that a plane wave propagating along the z-direction 1.2 Prove the vector analysis identities:
has electric and magnetic elds that are transverse to the z-direction and are related by:
 ()= 0
1 ()= 2 + (Greens rst identity)
|H| = |E| , =
 ( )= 2 2 (Greens second identity)
(A)= ()A + A
Moreover the time-averaged energy ux (in the z-direction) and energy density are: (A)= ( )A + A
  A)= 0
(
|E|2 1 d() 1 d() 1 d() 1 d() A)A (
A B = B ( B)
Pz = , w = |E|2 + |H|2 = + 2 |E|2
2 4 d 4 d 4 d d ( A)= ( A)2 A

The energy velocity is dened by ven = Pz /w. Thus, we have: 1.3 Consider the innitesimal volume element xyz shown below, such that its upper half
     lies in medium 1 and its lower half in medium 2 . The axes are oriented such that n = z.
1 w 1 d() 1 d() 1 d()  d()
ven = = + = + Applying the integrated form of Amperes law to the innitesimal face abcd, show that
Pz 2 d d 2  d d
Dx
It is easily veried that the right-hand side can be expressed in terms of the wave- H2y H1y = Jx z + z
 t
number k =  in the form:
In the limit z 0, the second term in the right-hand side may be assumed to go to zero,
   
1 d()  d() d  dk whereas the rst term will be non-zero and may be set equal to a surface current density,
1
ven = + = = = vg1 (1.18.7) that is, Jsx limz0 (Jx z). Show that this leads to the boundary condition H1y H2y =
2  d d d d
Jsx . Similarly, show that H1x H2x = Jsy , and that these two boundary conditions can be
which shows the equality of the energy and group velocities. See Refs. [168182] for combined vectorially into Eq. (1.5.4).
further discussion on this topic.
Eq. (1.18.7) is also valid for the case of lossless negative-index media and implies that
the group velocity, and hence the group refractive index ng = c0 /vg , will be positive,
even though the refractive index n is negative. Writing  = || and = || in this
 
case and noting that = ||/|| and n = ||/ 0 0 , and k = n/c0 , we have:
     
1 1 || d() || d() 1 || d(||) || d(||)
ven = + = +
2 || d || d 2 || d || d
  Next, apply the integrated form of Gausss law to the same volume element and show the
d || 1 d(n) dk
= = = = vg1 boundary condition: D1z D2z = s = limz0 (z).
d c0 d d  
1.4 Show that the time average of the product of two harmonic quantities A(t)= Re Aejt
 
from which we also obtain the usual relationship ng = d(n)/d. The positivity of and B(t)= Re Bejt with phasors A, B is given by:
vg and ng follows from the positivity of the derivatives d()/d and d()/d, as
T
required to keep ven positive in negative-index media [391]. 1 1 
A(t)B(t) = A(t)B(t) dt = Re AB ]
T 0 2

where T = 2/ is one period. Then, show that the time-averaged values of the cross
1.19 Problems  
and dot products of two time-harmonic vector quantities A (t)= Re A ejt and B (t)=
 jt 
Re B e can be expressed in terms of the corresponding phasors as follows:
1.1 Prove the vector algebra identities:
1   1  
A (B C)= B(A C)C(A B) (BAC-CAB identity) B(t) =
A (t)B Re A B , B(t) =
A (t)B Re A B
2 2
A (B C)= B (C A)= C (A B)
|A B|2 + |A B|2 = |A|2 |B|2 1.5 Assuming that B = H, show that Maxwells equations (1.9.2) imply the following complex-
A = n A n + (n A)n (n is any unit vector) valued version of Poyntings theorem:

In the last identity, does it a make a difference whether n A n is taken to mean n (A n) (E H )= jH H E Jtot

, where Jtot = J + jD
or (n A)n?
1.19. Problems 33 34 1. Maxwells Equations

Extracting the real-parts of both sides and integrating over a volume V bounded by a closed 1.10 Conductors and plasmas exhibit anisotropic and birefringent behavior when they are in the
surface S, show the time-averaged form of energy conservation: presence of an external magnetic eld. The equation of motion of conduction electrons in

a constant external magnetic eld is mv = e(E + v B)mv, with the collisional term
1 1 included. Assume the magnetic eld is in the z-direction, B = z B, and that E = x Ex + y Ey
Re[E H ]dS =
Re[E Jtot ] dV
S 2 V 2 and v = x vx + y vy .

which states that the net time-averaged power owing into a volume is dissipated into heat. a. Show that in component form, the above equations of motion read:
For a lossless dielectric, show that the above integrals are zero and provide an interpretation.
e
1.6 Assuming that D = E and B = H, show that Maxwells equations (1.1.1) imply the following vx = Ex + B vy vx
m eB
relationships: where B = = (cyclotron frequency)
e m
vy = Ey B vx vy
B  1  m
Ex + D = Ex E x E2
t x 2
What is the cyclotron frequency in Hz for electrons in the Earths magnetic eld B =
D  1  0.4 gauss = 0.4104 Tesla? [Ans. 1.12 MHz.]
(J B)x + B x = Hx H x H2
t 2
b. To solve this system, work with the combinations vx jvy . Assuming harmonic time-
where the subscript x means the x-component. From these, derive the following relationship dependence, show that the solution is:
that represents momentum conservation:
e
(Ex jEy )
vx jvy = m
Gx
fx + = Tx (1.19.1) + j( B )
t
where fx , Gx are the x-components of the vectors f = E + J B and G = D B, and Tx is c. Dene the induced currents as J = Nev. Show that:
dened to be the vector (equal to Maxwells stress tensor acting on the unit vector x):
0
Jx jJy = ()(Ex jEy ), where ()=
1 + j( B )
Tx = Ex E + Hx H x (E2 + H2 )
2
Ne2
where 0 = is the dc value of the conductivity.
Write similar equations of the y, z components. The quantity Gx is interpreted as the eld m
momentum (in the x-direction) per unit volume, that is, the momentum density. d. Show that the t-domain version of part (c) is:
1.7 Show that the causal and stable time-domain dielectric response corresponding to Eq. (1.11.5)
t

is given as follows: Jx (t)jJy (t)= (t t ) Ex (t )jEy (t ) dt
0
2p
(t)= 0 (t)+0 (t) , (t)= et/2 sin(0 t)u(t) (1.19.2) where (t)= 0 et ejB t u(t) is the inverse Fourier transform of () and
0
u(t) is the unit-step function.

where u(t) is the unit-step function and 0 = 20 2 /4, and we must assume that e. Rewrite part (d) in component form:
< 20 , as is typically the case in practice. Discuss the solution for the case /2 > 0 .
t
 
1.8 Show that Jx (t) = xx (t t )Ex (t )+xy (t t )Ey (t ) dt
the plasma frequency for electrons can be expressed in the simple numerical form: 0
fp = 9 N, where fp is in Hz and N is the electron density in electrons/m3 . What is fp for
t
the ionosphere if N = 1012 ? [Ans. 9 MHz.]  
Jy (t) = yx (t t )Ex (t )+yy (t t )Ey (t ) dt
0
1.9 Show that the relaxation equation (1.13.2) can be written in the following form in terms of
the dc-conductivity dened by Eq. (1.12.4): and identify the quantities xx (t), xy (t), yx (t), yy (t).

1 f. Evaluate part (e) in the special case Ex (t)= Ex u(t) and Ey (t)= Ey u(t), where Ex , Ey
(r, t)+(r, t)+ (r, t)= 0 are constants, and show that after a long time the steady-state version of part (e) will
0
be:
Then, show that it reduces to the naive relaxation equation (1.7.3) in the limit = 1/ 0.
Show also that in this limit, Ohms law (1.13.1) takes the instantaneous form J = E, from Ex + bEy
Jx = 0
which the naive relaxation constant relax = 0 / was derived. 1 + b2
Ey bEx
Jy = 0
1 + b2
1.19. Problems 35 36 1. Maxwells Equations

where b = B /. If the conductor has nite extent in the y-direction, as shown above, Indeed, show that these integrals may be reduced to the following ones, which can be
then no steady current can ow in this direction, Jy = 0. This implies that if an electric found in standard tables of integrals:
eld is applied in the x-direction, an electric eld will develop across the y-ends of the

conductor, Ey = bEx . The conduction charges will tend to accumulate either on the 2 dy 2 y2 dy 1
= = 
right or the left side of the conductor, depending on the sign of b, which depends on 0 1 2y2 cos + y4 0 1 cos + y4
2y 2 2(1 cos )
the sign of the electric charge e. This is the Hall effect and is used to determine the
where sin(/2)= /(20 ).
sign of the conduction charges in semiconductors, e.g., positive holes for p-type, or
negative electrons for n-type. e. Consider the limit of Eq. (1.19.5) as 0. Show that in this case the functions r , i
What is the numerical value of b for electrons in copper if B is 1 gauss? [Ans. 4.3107 .] are given as follows, and that they still satisfy the Kramers-Kronig relations:

g. For a collisionless plasma ( = 0), show that its dielectric behavior is determined from 2p 2p 2p  
Dx jDy =  ()(Ex jEy ), where r ()= P +P , i ()= ( 0 )( + 0 )
0 0 + 2 0
 
2p 1.12 Derive the Kramers-Kronig relationship of Eq. (1.17.7) by starting with the causality condi-
 ()= 0 1
( B ) tion (t)u(t)= 0 and translating it to the frequency domain, that is, expressing it as the
convolution of the Fourier transforms of (t) and u(t).
where p is the plasma frequency. Thus, the plasma exhibits birefringence.
1.13 An isotropic homogeneous lossless dielectric medium is moving with uniform velocity v with
1.11 This problem deals with various properties of the Kramers-Kronig dispersion relations for respect to a xed coordinate frame S. In the frame S moving with dielectric, the constitutive
the electric susceptibility, given by Eq. (1.17.8). relations are assumed to be the usual ones, that is, D  = E  and B  = H  . Using the Lorentz
transformations given in Eq. (K.30) of Appendix K, show that the constitutive relations take
a. Using the symmetry properties r ()= r () and i ()= i (), show that the following form in the xed frame S:
(1.17.8) can be written in the folded form of Eq. (1.17.9).
 0 0
D = E + av (H v E) , B = H av (E + v H) , a=
b. Using the denition of principal-value integrals, show the following integral: 1 v2

d
P =0 (1.19.3)
0 2 2

 
dx 1 a + x
Hint : You may use the following indenite integral: = ln  .
a2 x2 2a
 a x
c. Using Eq. (1.19.3), show that the relations (1.17.9) may be rewritten as ordinary inte-
grals (without the P instruction) as follows:

2  i ( )i ()
r () = d
0 2 2

(1.19.4)
2 r ( )r ()
i () = d
0 2 2

Hint : You will need to argue that the integrands have no singularity at  = .
d. For a simple oscillator model of dielectric polarization, the susceptibility is given by:

2p
() = r ()ji ()=
20 2 + j
(1.19.5)
2p (20 2 ) 2p
= j
(20 2 )2 +2 2 (20 2 )2 +2 2

Show that for this model the quantities r () and i () satisfy the modied Kramers-
Kronig relationships (1.19.4). Hint : You may use the following denite integrals, for
which you may assume that 0 < < 20 :

2 dx 1 2 x2 dx 1
= , =
0 (20 x2 )2 +2 x2 20 0 (20 x2 )2 +2 x2