EECS 434: Principles of Photonics

Electromagnetic Theory of Light, Part 2

Department of Electrical Engineering and Computer Science

University of Michigan
Fall 2023
 Further Reading

Saleh & Teich §5.4–§5.6

 Quiz
Suppose E is a vector eld and φ is a scalar eld. Which of the following is not a legitimate operation?

A. ∇ × ∇φ

B. ∇ ⋅ ∇φ

C. ∇2 E

D. ∇ × ∇ ⋅ E

E. ∇ ⋅ ∇ × E
fi
fi
 Summary of Vector Calculus
❖ Divergence: For a vector eld E(x, y, z) = Ex(x, y, z)x̂ + Ey(x, y, z)ŷ + Ez(x, y, z)z,̂ the divergence is de ned as
∂Ex ∂Ey ∂Ez
∇⋅E= + + . Divergence quanti es the vector eld’s “source” at each point in space.
∂x ∂y ∂z

( ∂z ∂x )
∂Ez ∂Ey ∂Ex ∂Ez ∂Ey ∂Ex
( ∂y ∂z ) ( ∂x ∂y )
❖
Curl: For a vector eld E(x, y, z), the curl is de ned as ∇ × E = − x̂ + − ŷ + − z.̂ The

curl at a point in the eld is represented by a vector whose length and direction correspond to the magnitude and axis of the
maximum circulation.

∂V ∂V ∂V
Gradient: For a scalar eld V(x, y, z), the gradient is de ned as ∇V = x̂ + ŷ + z,̂ where x,̂ y,̂ z ̂ are unit vector along
❖
∂x ∂y ∂z
the three axes. The gradient is the direction in which the function increases most quickly from V, and the magnitude of the
gradient is the rate of increase in that direction.
2 2 2
2 ∂ V ∂ V ∂ V
Laplacian: For a scalar eld V(x, y, z), the Laplacian is de ned as ∇ V = ∇ ⋅ ( ∇V ) = + +
❖
∂x 2 ∂y 2 ∂z 2
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
 Maxwell’s Equations in Free Space
Integral Form Differential Form
Q ρ
∯S
Gauss’s law E ⋅ dA = ∇⋅E= E: Electric eld (V/m)
ε0 ε0
B: Magnetic induction

∯S
Gauss’s law for B ⋅ dA = 0 (webers/m2, Tesla)
∇⋅B=0
magnetism
ρ: Electric charge density
(C/m3)
∂B ∂B
∮C ∬Σ ∂t
Faraday’s law of E ⋅ dl = − ⋅ dA ∇×E=−
induction ∂t J: electric current density
(A/m2)

∬Σ ( ∂t )
Ampère’s circuital ∂E ∂E
∮C
law with B ⋅ dl = μ0 J + μ0ε0 ⋅ dA ∇ × B = μ0 J + μ0ε0
∂t
displacement current
fi
 Maxwell’s Equations in a Medium
Integral Form Differential Form E: Electric eld (V/m)
Qf ρf
∯S
Gauss’s law D ⋅ dA = ∇⋅D= D: Displacement eld
ε0 ε0 (C/m2)

B: Magnetic induction
∯S
Gauss’s law for B ⋅ dA = 0 ∇⋅B=0 (webers/m2, Tesla)
magnetism
H: Magnetic eld (A/m)
∂B ∂B
∮C ∬Σ ∂t
Faraday’s law of E ⋅ dl = − ⋅ dA ∇×E=− ρ: Electric charge density
induction ∂t
(C/m3)

∬Σ ( ∂t )
Ampère’s circuital ∂D J: electric current density
∂D
∮C
law with H ⋅ dl = Jf + ⋅ dA ∇ × H = Jf + (A/m2)
∂t
displacement current
fi
fi
fi
 Constitutive Equations
E: Electric eld (V/m)

D = ε0E + P D: Displacement eld

(C/m2)
1
H= B−M P: Polarization density
μ0 (C/m2)

For isotropic linear materials B: Magnetic induction

(webers/m2, Tesla)

P = χε0E, D = εE = ε0(1 + χ)E = ε0εr E H: Magnetic eld (A/m)

M = χH, B = μH = μ0(1 + χ)H = μ0 μr H M: Magnetization eld
(A/m)
fi
fi
fi
fi
 Wave Equation in Free Space
Per Faraday’s law of induction and Ampère’s circuital law with displacement current, vector elds E and
H are coupled! Let us try to decouple them using our constitutive relations and a mathematical trick. Take
the curl of one curl equation and substitute the other into it.

( ∂t )
∂H
∇ × ( ∇ × E) = ∇ × −μ0
2 ∂( ∇ × H)
∇( ∇ ⋅ E) − ∇ E = − μ0
∂t
In free space there is no charge, so ∇ ⋅ E = 0 on the LHS, per Gauss’s law.
∂( ∇ × H) ∂E
Thus, ∇2 E = μ0 . Also, since ∇ × H = ε0 in free space per Ampère’s circuital law with
∂t ∂t
displacement current, we arrive at the wave equation in free space
2
2 ∂ E
∇ E = μ0ε0 2
∂t

fi
 Wave Equation for E in (x, t)
Compare this with the 1-dimensional equation for transverse waves traveling
in a straight line along z at velocity v:
2 2 2
2 ∂ E d E 1 ∂ E
∇ E = μ0ε0 2 ↔ 2 = 2 2
∂t dz v ∂t
Evidently the velocity is given by:
1 1 8
= μ ε
0 0 . Thus, v = ≈ 3 × 10 m/s—speed of light!
v 2 μ0ε0
 Maxwell’s Conclusion
This velocity is so nearly that of light that it seems we have a strong reason
to conclude that light itself (including radiant heat and other radiations) is
an electromagnetic disturbance in the form of waves propagated through
the electromagnetic eld according to electromagnetic laws.

- James Clerk Maxwell

fi
Possible Solutions of the Vector Wave Equation
2
2 ∂ E
Starting from the wave equation: ∇ E = μ0ε0
∂t 2
Harmonic functions yield solutions to the vector wave equation. Consider cosine and sine functions:
∂2 E
E(r, t) = E01 cos (k ⋅ r − ωt) + E02 sin (k ⋅ r − ωt). Now let us calculate 2 and ∇2 E:
∂t
2
∂ E(r, t) 2
= − ω E(r, t)
∂t 2
∇2 E(r, t) = − k 2 E(r, t).
2
ω 1 2
To satisfy the wave equation, one requires = ≡ c .
k2 μ0ε0
As such, EM waves must obey a simple, but important, dispersion relation that links wave vector, velocity, and
frequency:
ω = ck
 Phase Velocity of Electromagnetic Waves
Because the comparison was with a 1-dimensional wave equation, there is implicitly no variation of
wave amplitude in other directions. The velocity v is therefore the velocity of constant phase fronts
perpendicular to z. This velocity is the phase velocity in free space.
1
vp =
μ0ε0

In a medium, the phase velocity is

1 1 c
vp = = = ,
με μ0ε0 μrεr n

where n is referred to as the refractive index of the material. For non-magnetic material, μr = 1, so that
n = εr .
 Boundary Conditions
Maxwell’s equations can be solved even in regions where ε and μ are discontinuous. We need these
continuity relations, derived from Maxwell’s equations at boundaries, in order to solve for elds in
waveguides and other situations. n̂
Dielectric 1 dl

Σ C
Dielectric 2
S

Divergence theorem Stokes’s theorem

∭V ∯S ∬Σ ∮C
∇ ⋅ FdV = F ⋅ dA ( ∇ × F) ⋅ dA = F ⋅ dl
n̂ ⋅ (B2 − B1) = 0 Since the area Σ can be made in nitesimal length and
n̂ ⋅ (D2 − D1) = σ (surface charge density) ∇ × E = − ∂B/∂t is nite, LHS = 0 for F = E. We have
⇒ B2n = B1n, D2n − D1n = σ E2t = E1t. For F = H, the boundary condition is
H2t − H1t = K (surface current density)
fi
fi
fi
 Boundary Conditions
Interface between two dielectrics

Dielectric 1
B2n = B1n, D2n − D1n = σ
Dielectric 2 E2t = E1t, H2t − H1t = K

Interface between dielectric & conductor

Dielectric 1
En =0
σ K
Conductor =0 =0

There is a concentration of fields at the tip of a conducting needle

resulting from these boundary conditions (because the E field must be
normal to the interface) – the Lightning Rod Effect.
𝑛
𝑡
𝑡
𝑡
𝑛
𝑛
𝑡
𝑡
𝐷
𝐵
𝐸
𝐻
𝐸
𝐸
𝐸
𝐻
Lightening Rods at Work
 Complex Field Representation
2 2
∂E ∂E 1D wave equation
= μ0ε0
∂z 2 ∂t 2 in free space

The wave equation and the Poynting vector indicate that we often have to differentiate or
combine field expressions. What solutions are convenient?
E(z, t) = a ̂ | E | cos(ωt − kz + φ),
where ω is the angular frequency, k is the wavenumber, φ is a fixed phase, and | E | is the
amplitude of the vector field oriented along a.̂
Substitution into the wave equation shows that this is a possible solution. However, sin(x) and cos(x)
functions are cumbersome to differentiate, integrate, and combine. In contrast, complex exponentials are
iφ i(ωt−kz)
much more amenable. We introduce the complex field E(z, t) = a ̂ | E | e e as a convenience. We
should bear in mind that EM fields must be real, so we will always take the real part as the actual solution:
E(z, t) = ℜ[a ̂ | E | e iφe i(ωt−kz)]
1 iφ i(ωt−kz)
= a[̂ | E | e e + c . c.]
2
= a ̂ | E | cos(ωt − kz + φ)
 Transversality of Electromagnetic Fields
From the divergence equations we can show that in homogeneous, charge-free media our earlier plane wave solutions for E, H are
perpendicular to the direction of wave propagation:

∇⋅E=0

∇ ⋅ E0e i(ωt−k⋅r) = − ike i(ωt−k⋅r) ⋅ E0

Thus, k ⋅ E0 = 0

Hence, E0 is perpendicular to k. Similarly, for the magnetic eld,

∇⋅H=0

∇ ⋅ H0e i(ωt−k⋅r) = − ike i(ωt−k⋅r) ⋅ H0

k ⋅ H0 = 0

Thus, H0 is also perpendicular to k.

fi
 Orthogonality of Electromagnetic Fields
From Ampere’s curl equation we can also show that in homogeneous, charge-free media our earlier plane wave solutions for E, H are
orthogonal to one another:

∂E
∇ × H = ε0
∂t
LHS = ∇ × H0e i(ωt−k⋅r) = − ik × H

∂E0e i(ωt−k⋅r)
RHS = ε0 = iε0ωE
∂t

(k × H) = − ε0ωE

( ε0ω )
k |E| μ0
(k × H) = − η0(k × H), where η0 =
Thus, E = − ̂ ̂ = = 377 Ω
|H| ε0

Hence E is perpendicular to H, and the ratio | E/H | = η0 is determined by the impedance of vacuum, which is 377 Ω. Note that this is a
non-dissipative impedance, i.e., no energy loss for electromagnetic elds propagating in vacuum.
fi
Energy Stored in an Electromagnetic Field
N

∑
Total work: WE = Wi
Energy stored = Work needed to assemble i=1
One can show that
the charges and form the eld 1
2 ∫V
WE = dVE ⋅ D

q2 Likewise, the work needed to form a magnetic eld is

1
2 ∫V
Spent W2 WM = dVB ⋅ H
q1
q4 As such the energy stored in an electromagnetic eld is
Spent W1 1
2 ∫V
Spent W4 W = WE + WM = dV (E ⋅ D + B ⋅ H)
q3
Spent W3 The energy stored per volume (energy density) is de ned as
1
U = (E ⋅ D + B ⋅ H)
2
fi
fi
fi
fi
 Power Dissipation
L

J
Voltage: V
Current: I A
R E
Resistor

Now consider a volume with area A and length L with

Energy dissipated per unit time
input current density J and electric eld E
P = IV
P = IV = AJ ⋅ LE
Thus, J ⋅ E is the energy dissipated per unit time in a
unit volume
fi
 Poynting’s Theorem
Energy dissipated per unit time in a unit volume: J ⋅ E
∂D
J=∇×H−
∂t
∂D
J ⋅ E = E ⋅ ( ∇ × H) − E ⋅
∂t
Now let us use a vector identity:
∂B
E ⋅ ( ∇ × H) = H ⋅ ( ∇ × E) − ∇ ⋅ (E × H) = − H ⋅ − ∇ ⋅ S, where S ≡ E × H.
∂t
∂B ∂D
Thus, J ⋅ E = − H ⋅ −E⋅ − ∇ ⋅ S.
∂t ∂t
1
Recall U = (E ⋅ D + B ⋅ H), we arrive at
2
∂U
J⋅E=− −∇⋅S
∂t
Power loss equals the loss rate of stored energy plus the ux of power out.
fl
 Quantities’s Poynting’s Theorem
∂U
J⋅E=− −∇⋅S
∂t
1
U = (E ⋅ D + B ⋅ H)
2
Energy ux
S≡E×H

U: energy density of electromagnetic elds (J/m3)

S: energy ux (J/m2/s = W/m2)
| S | : power per unit area (W/m2) carried by electromagnetic wave in the direction of S
∇ ⋅ S: the net power change through electromagnetic ux
fl
fl
fi
fl
 Time Averaging
Most optical frequencies are in the order on 1014 Hz (several hundred of THz). No detectors can respond this fast. So time averaged values
are useful.

⟨2 ⟩
1 1
⟨S⟩ = ⟨E × H⟩ = x̂ [ | E | e e
iφ i(ωt−kz)
+ c . c . ] × ŷ [ | H | e iφe i(ωt−kz) + c . c . ]
2

⟨ ⟩ 2
1 2i(ωt−kz+φ) −2i(ωt−kz+φ) 1
= | E | | H | ẑ 2 + e +e = | E | | H | ẑ
4
High frequency oscillations

Alternatively, we can start with the complex elds E(z, t) = | E | e iφe i(ωt−kt), H(z, t) = | H | e iφe i(ωt−kt) and calculate ⟨S⟩ as

1 1
⟨S⟩ = ℜ{E × H*} = | E | | H | z,̂
2 2
where the asterisk * represents the complex conjugate. With this formula, one can also calculate ⟨U⟩ based on the complex elds:

[2 ⟨2 ⟩]
1 1 1
U=ℜ E ⋅ D* + B ⋅ H*
2
fi
fi
 Classical Electron Model
Atoms are very small compared to the wavelength of light. Hence we assume the electric eld is uniform over each atom. The nucleus is much
more massive than the electron, so we also assume that it does not move.

Newton’s second law of motion: x(t)

me a = Ftot
∂2 x ∂x
me 2 = − kx − meγ − qE
∂t ∂t
For linear response, x must have the same temporal behavior as E.

E(t) = E0e iωt and x(t) = x0e iωt

Hence steady-state solution for x0 is found by substitution into the equation of motion:

me(iω)2 x0e iωt = − kx0e iωt − meγ(iω)x0e iωt − qE0e iωt

( ω02 − ω 2 + iγω )
−qE0 /me
x0 =

me: electron mass, ω0 = (k/me) : resonant frequency

1/2

q: electron charge, γ: damping coef cient

fi
fi
 Classical Electron Model
The steady state solution for displacement x(t) is

( ω02 − ω 2 + iγω )
iωt
−q/me iωt
x(t) = x0e = E0 e

Hence the induced dipole moment on each atom p(t) is

q 2 /me
( ω0 − ω + iγω )
iωt iωt
p(t) = − qx(t) = 2 2
E0 e = αE0 e ,

q 2 /me
( ω0 − ω + iγω )
where α = is the atomic polarizability.
2 2

The macroscopic electric polarization P due to N atoms is

Nq 2 /me
( ω0 − ω + iγω )
iωt
P(t) = Np(t) = 2 2
E0 e

The electric polarization P captures many aspects of optical response, such as dispersion, absorption, etc.!
 Classical Electron Model
Electric polarization

Nq 2 /me
( ω0 − ω + iγω )
iωt
P(t) = Np(t) = 2 2
E0e

The optical response of the medium is described by the constitutive equation:

D = ε0E + P = ε0(1 + χe)E

One can thus derive the susceptibility χe as

Nq 2 /ε0me
χe = 2
ω0 − ω 2 + iγω

Nq 2 /ε0me
The relative permittivity is then εr = 1 + χe = 1 + , and the refractive index is derived as n 2 = εr
ω02 − ω 2 + iγω

The classical electron model, albeit simple, captures many critical properties for the optical response of the medium!
 Classical Electron Model
In general, χe, εr, and n are complex. They can be written in terms of their real and imaginary parts. For example the susceptibility is expressed
as

Nq 2 /ε0me
χe = χ′ − iχ′′ = 2 ,
ω0 − ω + iγω
2

with real and imaginary parts being

ω02 − ω 2
ε0me [ (ω02 − ω 2)2 + (γω)2 ]
Nq 2
χ′ =

ε0me [ (ω02 − ω 2)2 + (γω)2 ]

Nq 2 γω
χ′′ =

Likewise, the complex refractive index can be expressed as n = n′ − in′′. The solution for the propagating electric eld is thus

E(z, t) = E0e i(ωt−kz) = E0e i(ωt−nk0z) = E0 e −n′′k0z e i(ωt−n′k0z)



attenuation propagation

The real part of n sets propagation speed, while the imaginary part of n determines the decay rate (absorption).










fi
 Classical Electron Model
For a small susceptibility (χe ≪ 1), the refractive index is approximated by

1 Nq 2 /2ε0me
n= 1 + χe ≈ 1 + χe = 1 + 2
2 ω0 − ω 2 + iγω

with the real and imaginary parts being

ω02 − ω 2
2ε0me [ (ω02 − ω 2)2 + (γω)2 ]
Nq 2
n′ = 1 +

2ε0me [ (ω02 − ω 2)2 + (γω)2 ]

Nq 2 γω
n′′ = 1 +

For a dielectric material, ω0 is in the UV region, thus ω < ω0 in the visible. We have n′red < n′blue, so (vp)red > (vp)blue. At the off-resonance limit of
Nq 2
ω ≪ ω0, n′ → 1 + and n′′ ∼ 0, i.e., the material is non-absorptive (useful for optical communication).
2ε0meω02

Nq 2
In contrast, on resonance at ω ∼ ω0, n′ → 1 and n′′ → 1 + ≫ 1, i.e., the material becomes highly absorptive (useful for photodetection).
2ε0meγω











 Classical Electron Model
The refractive index is dependent on the density of the medium. For air at 1 atmosphere of pressure, assuming the nearest resonance is in the
ultraviolet (UV), the index is approximately unity, derived as follows:

ρNA
N= , where ρ = 1.2047 × 103 g/m3 is the density of the air, NA = 6.022 × 1023 is the the Avogadro constant de ning a mole, GMW (gram
GMW
molecular weight) = 28.00 g is the mass (in grams) of 1 mole of nitrogen (the dominant compound in atmosphere). One obtain

N = 2.591 × 1025 m-3.

ω02 − ω 2
2ε0me [ (ω0 − ω ) + (γω) ]
Nq 2
We next derive the refractive index using the formula n′ = 1 + with the constants
2 2 2 2

q = 1.6 × 10−19 C, ε0 = 8.85 × 10−12 J/V2·m, me = 9.1091 × 10−31 kg. Also, the resonance frequency of the air is in the UV region, so
ω0 = 2π × 12 × 1014 rad·Hz. We obtain

n′ = 1.000727.

Note that in solids and liquids, the particle density is ~1000 times higher than in air. Consequently, N is larger, leading to refractive indices
considerably higher (nsolid ∼ 1.4 − 2.4)


fi
 Classical Electron Model
Frequency-dependent refractive index with multiple resonances

n′
2.0

n
Vibrational/
1.0 rotational Electronic
resonances resonances
Visible spectral
n′′ region
ω

ω02 − ω 2
2ε0me [ (ω02 − ω 2)2 + (γω)2 ] 2ε0me [ (ω02 − ω 2)2 + (γω)2 ]
Nq 2 Nq 2 γω
n′ = 1 + n′′ = 1 +






 Spectral Components of Pulses
We have learned that monochromatic plane waves were solutions to Maxwell’s equations. However, plane waves are overly
idealized with rare practical relevance. In reality, short pulses radiation are more common in, e.g., optical communication to
transmit information. A short pulse entails a spread of spectral components or wavelengths, as mathematically described by
the Fourier transform:

2π
E(t) T=
ω0
τ0 τ0
−
2 2
t

Pulse Duration: τ0
Spectral Components of Pulses
2π
E(t) T=
ω0
τ0 τ0
−
2 2
t

Pulse Duration: τ0

{ 0,
e iω0t, −τ0 /2 < t < τ0 /2
E(t) =
elsewhere
τ0 /2
∞ τ0 /2

2π [ i(ω0 − ω) ]
i(ω0−ω)t
1 1 1 e
2π ∫−∞ 2π ∫−τ0/2
−iωt i(ω0−ω)t
g(ω) = E(t)e dt = e dt =
−τ0 /2
τ0 sin[(ω0 − ω)τ0 /2] τ0
= = sinc[(ω0 − ω)τ0 /2]
2π (ω0 − ω)τ0 /2 2π
 Spectral Components of Pulses
2π
E(t) T=
ω0
τ0 τ0
−
2 2 t

Pulse Duration: τ0
τ0
g(ω) = sinc[(ω0 − ω)τ0 /2]
2π

( 2π )
2
τ0
I(ω) ≡ | g(ω) |2 = sinc2[(ω0 − ω)τ0 /2]

I(ω) Time-frequency uncertainty

FWHM: full-width at half maximum The time-bandwidth
2π product, (Δf )FWHM × (Δt)FWHM = 1,
(Δω)FWHM =
τ0 is a constant.
1
(Δf )FWHM =
ω0 τ0

( τ0 ) ( τ0 )
π π ω
ω0 − ω0 +
 Group Velocity
A pulse = Sum of monochromatic plane waves with different frequencies
A perfectly monochromatic wave truncated in time consists of a distribution of frequencies, given by the Fourier transform of the pulse in the time domain.
Each Fourier component is a plane wave solution of Maxwell’s equations. Since the Maxwell’s equations are linear, the sum of components is also a solution.
Now let’s write a pulse as the sum of monochromatic plane waves:
∞

∫−∞
E(z, t) = A(k)e i(ωt−kz)dk

ω and k are connected by the dispersion relation ω(k). Expanding ω(k) around ω0 = ω(k0) using Taylor series:
dω
ω(k) = ω0 + (k − k0) + . . .
dk k=k0
we obtain
∞

( dk )
dω
∫−∞
E(z, t) = e i(ω0t−k0 z)
A(k)exp i t − z (k − k0) dk = e i(ω0t−k0z)V(z, t),
k=k0

where e i(ω0t−k0z) describes the propagation of phase while V(z, t) describes the propagation of the pulse envelope. Assuming the pulse envelope takes from t0 to t1
dω dω
to propagate from z0 to z1. We have t0 − z0 = t1 − z1. The speed for pulse propagation is de ned as the group velocity:
dk k=k0 dk k=k0
z1 − z0 dω
vg = =
t1 − t0 dk k=k0

fi
 Group Velocity vs Phase Velocity
Group Velocity (·) Phase Velocity (·)

dω ω
vg = vp =
dk k=k0 k k=k0

The propagating electric eld is described as E(z, t) = e i(ω0t−k0t)V(z − vgt), indicating that the pulse moves along undistorted (to the rst
order of the Taylor expansion) at the group velocity vg, which differs from phase velocity vp = c/n = ω/k = ωλ/2π (λ is the wavelength in
the medium, λ0 is the central wavelength in the medium):

dλ ( k 2 )
dω d dvp dvp dλ dvp −2π
dk ( )
vg = = vp k = vp + k = vp + k = vp + k
dk k=k0 k=k0 dk k=k0 dλ dk k=k0 k=k0

( λ=λ0 )
λ dn c
vg = vp 1 + =
n dλ n + ω(dn/dω)
ω=ω0

dω ω c
In general, vg = ≠ vp = =
dk k=k0 k k=k0 n(ω0)
fi
fi
 Fast & Slow Light
Normal
dispersion Anomalous
n′
2.0 region: dispersion
Group velocity dn/dλ < 0 region:

( n dλ )
λ dn n dn/dλ > 0
vg = vp 1 +
λ=λ0 Vibrational/
1.0 rotational Electronic
resonances resonances
Visible spectral
n′′ region
ω

Near resonances the slope of the dispersion, dn/dλ, can have either sign and be large, resulting in very different vg’s

dn/dλ > 0 → Fast light in the anomalous dispersion region

dn/dλ < 0 → Slow light in the normal dispersion region (with applications in quantum storage)



 Chromatic Dispersion
∞

∫−∞
We now come back to the expansion of the propagating electric eld E(z, t) = A(k)e i(ωt−kz)dk. Let us now consider expanding the dispersion
relation to the second order:

dω 1 d 2ω
ω(k) = ω0 + (k − k0) + (k − k0)2 + . . .
dk k=k0 2 dk 2
k=k0

With the quadratic term accounted for, the electric eld is written as

{ ( dk 2 dk 2 ) }
2
dω 1 d ω
∫−∞ ( )
i(ω0t−k0 z)
E(z, t) = e A(k)exp i + (k − k0) t − z k − k0 dk

( )
2
i(ω0t−k0 z) 1 d ω
=e V z − vg + (k − k0) t
2 dk 2

1 d 2ω
The term (k − k0) introduces a frequency-dependent group-velocity variation, i.e., different spectral components travel at different group
2 dk 2
velocities, leading to distortion on the pulse during propagation.
fi
fi
 Group Velocity Dispersion
With the second order correction in the transform of the eld, there is no longer only one velocity for all wavelengths. The
pulse distorts as it propagates!

Transmitted pulses Received pulses

fi
 Group Velocity Delay
It is important to characterize the frequency-dependence of group velocity, referred to as the group velocity dispersion (GVD), because GVD is
detrimental to communication performance. The traditional coef cient that serves this purpose is related to 2nd-order terms in a Taylor series of
propagation constant k:

dk 1 d 2k
k(ω) = k0 + (ω − ω0) + (ω − ω0)2 + . . .
dω ω=ω0 2 dω 2
ω=ω0

We now introduce the GVD coef cient D, de ned as the delay time dT incurred between components separated by dλv in vacuum per unit
propagation distance:

L ( dλv )
1 dT
D≡ (ps/nm/km)

Since T = L/vg, we obtain the expression for the GVD coef cient:

L [ dλv ( vg )] c ( dλv2 ) λv2 ( dω 2 )

1 d L λv d 2n 2πc d 2k
D= =− or equivalently, D = −
λv=λv0 λv=λv0 ω=ω0
fi
fi
fi
fi
 Group Velocity Dispersion in Silica Fiber
In silica bers the GVD has a wavelength at which D = 0, called the zero-dispersion wavelength λD.

Anomalous GVD
(Short λ travel faster)

λD

Normal GVD
λD (Short λ travel slower)

dvg 2πc d 2k λv d 2n
=0 D=− 2 =− =0
dλv λv=λD
λv dω 2 ω=ωD c dλv2 λv=λD
fi
 Group Velocity Dispersion Example
A student tries to use a dispersion-shifted silica ber (n = 1.5; dn/dλ = 0.1μm-1) at λ = 1.5 μm with GVD of D=15 ps/nm/km
for communication at 40 GHz over a distance of L=10 km. Will it work?

fi
 Group Velocity Dispersion Example
A student tries to use a dispersion-shifted silica ber (n = 1.5; dn/dλ = 0.1μm-1) at λ = 1.5 μm with GVD of D=15 ps/nm/km for
communication at 40 GHz over a distance of L=10 km. Will it work?

Solution

1 dT ΔT (ps)
D≡ →D=
L dλv Δλv (nm) ⋅ L (km)

Now let us nd the bandwidth Δλ in wavelength:

dλv λv λv2 λ2
fλv = c, so dfλv + dλv f = 0 → = − = − . Thus, Δλ = − Δf. Plugging in the numbers we nd Δλv ≈ 0.3 nm.
df f c c

We next estimate the pulse stretching due to GVD:

ΔT = D ⋅ L ⋅ Δλv = 15 ps/nm/km × 0.3 nm × 10 km = 45 ps.

In a 40 GHz communication system, each bit is transmitted with a pulse of δT = 1/40 GHz = 25 ps. After transmission, the pulse is
stretched by 45 ps, thus causing signi cant overlap with the adjacent bits. Thus, the communication system won’t work!
fi
fi
fi
fi
 Optical Absorption Coefficient
We have shown that the imaginary part of the refractive index caused attenuation of the electric eld, according to
E(z, t) = E0e i(ωt−kz) = E0e i(ωt−nk0z) = E0e −n′′k0ze i(ωt−n′k0z)
As a convention, the absorption (or “attenuation”) coef cient α describes the spatial decay of intensity rather than eld amplitude. So to
determine α we must compare
I(z) = I(0)exp(−αz) = | E(z, t) |2 = | E0 |2 exp(−2n′′k0z)
4πn′′
Thus, α = 2n′′k0 =
λ0
Note that from here on we may use a different notation to represent the real and imaginary parts of the refractive index, following the text by
writing
4πκ
κ ≡ n′′ ⇒ α =
λ0
In optical communication, α is expressed in units of dB/km:

( I(0) )
10 I(L)
α=− log10 . A typical value is 0.2 dB/km.
L











fi
fi
fi
 n of Electron Gas (Metals)
In metals, the electrons are not bound to individual nuclei but move around freely. That is, there is no restoring force in the Lorentz model (ω0 = 0). The
equation of motion now becomes
∂2 x ∂x
me 2 + meγ = − qE
∂t ∂t
Solving the equation yields
2
2
Nq /ε0me
n =1− 2
ω − iγω
If the damping rate is much lower than the frequency, i.e., γ ≪ ω, we have
2
2
Nq /ε0me
n ≈1−
ω2
ωp2
=1− 2
ω
Nq 2
where ωp ≡ is referred to as the plasma frequency. Note that for ω ≪ ωp, n is imaginary, indicating exponential decay in propagation. The wave
ε0me
is caused to re ect from the metal. For ω ≫ ωp, n is real, leading to propagation. For metals, N = 1023/cm3, ωp ∼ 2 × 1016 rad/s, corresponding to 100 nm.
fl
Metals Reflect Light
Propagation in Upper Atmosphere
The ionosphere is so named because atoms and molecules in this layer have
been ionized by radiation from space. The free electrons of the ionosphere
constitute a free electron gas with metal-like properties. In particular they
reflect all electromagnetic waves at frequencies ω ≪ ωp. This phenomenon
supports over-the-horizon radio communication.
 Discussions
1. Why is the electric eld in a metal always zero?

2. Is the orthogonality of electromagnetic eld, (k × H) = − ε0ωE, universally valid? How about wave propagation in a
medium?

3. Does Poynting’s theorem S ≡ E × H also hold in a medium?

fi
fi