Nonlinear Optics

Instructor:

Mark G. Kuzyk

Students:

Benjamin R. Anderson
Nathan J. Dawson
Sheng-Ting Hung
Wei Lu
Shiva Ramini
Jennifer L. Schei
Shoresh Shafei
Julian H. Smith
Afsoon Soudi
Szymon Steplewski
Xianjun Ye

August 4, 2010
2
Dedication

We dedicate this book to everyone who is trying to learn nonlinear optics.

i
ii
Preface

This book grew out of lecture notes from the graduate course in Nonlinear
Optics taught in spring 2010 at Washington State University. In its present
form, this book is a first cut that is still in need of editing. The homework
assignments and solutions are incomplete and will be added at a later time.

Mark G. Kuzyk
July, 2010
Pullman, WA

iii
iv
Contents

Dedication i

Preface iii

1 Introduction to Nonlinear Optics 1

1.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Kerr Effect . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Two-Photon Absorption . . . . . . . . . . . . . . . . . 3
1.1.3 Second Harmonic Generation . . . . . . . . . . . . . . 4
1.1.4 Optical Kerr Effect . . . . . . . . . . . . . . . . . . . . 5
1.2 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Example: Second Order Susceptibility . . . . . . . . . . . . . 8
1.4 Maxwell’s Equations . . . . . . . . . . . . . . . . . . . . . . . 10
1.4.1 Electric displacement . . . . . . . . . . . . . . . . . . . 11
1.4.2 The Polarization . . . . . . . . . . . . . . . . . . . . . 13
1.5 Interaction of Light with Matter . . . . . . . . . . . . . . . . 14
1.6 Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2 Models of the NLO Response 21

2.1 Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.1 Linear Harmonic Oscillator . . . . . . . . . . . . . . . 21
2.1.2 Nonlinear Harmonic Oscillator . . . . . . . . . . . . . 23
2.1.3 Non-Static Harmonic Oscillator . . . . . . . . . . . . . 26
2.2 Macroscopic Propagation . . . . . . . . . . . . . . . . . . . . 30
2.3 Response Functions . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.1 Time Invariance . . . . . . . . . . . . . . . . . . . . . 39
2.3.2 Fourier Transforms of Response Functions: Electric
Susceptibilities . . . . . . . . . . . . . . . . . . . . . . 40
2.3.3 A Note on Notation and Energy Conservation . . . . . 41

v
vi

2.3.4 Second Order Polarization and Susceptibility . . . . . 42

2.3.5 nth Order Polarization and Susceptibility . . . . . . . 43
2.3.6 Properties of Response Functions . . . . . . . . . . . . 43
2.4 Properties of the Response Function . . . . . . . . . . . . . . 46
2.4.1 Kramers-Kronig . . . . . . . . . . . . . . . . . . . . . 47
2.4.2 Permutation Symmetry . . . . . . . . . . . . . . . . . 51

3 Nonlinear Wave Equation 55

3.1 General Technique . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2 Sum Frequency Generation - Non-Depletion Regime . . . . . 58
3.3 Sum Frequency Generation - Small Depletion Regime . . . . 62
3.4 Aside - Physical Interpretation of the Manley-Rowe Equation 66
3.5 Sum Frequency Generation with Depletion of One Input Beam 68
3.6 Difference Frequency Generation . . . . . . . . . . . . . . . . 72
3.7 Second Harmonic Generation . . . . . . . . . . . . . . . . . . 75

4 Quantum Theory of Nonlinear Optics 87

4.1 A Hand-Waving Introduction to Quantum Field Theory . . . 87
4.1.1 Continuous Theory . . . . . . . . . . . . . . . . . . . . 87
4.1.2 Second Quantization . . . . . . . . . . . . . . . . . . . 91
4.1.3 Photon-Molecule Interactions . . . . . . . . . . . . . . 93
4.1.4 Transition amplitude in terms of E ⃗ . . . . . . . . . . . 96
4.1.5 Stimulated Emission . . . . . . . . . . . . . . . . . . . 97
4.2 Time-Dependent Perturbation Theory . . . . . . . . . . . . . 98
4.2.1 Time-Dependent Perturbation Theory . . . . . . . . . 99
4.2.2 First-Order Susceptibility . . . . . . . . . . . . . . . . 102
4.2.3 Nonlinear Susceptibilities and Permutation Symmetry 104
4.3 Using Feynman-Like Diagrams . . . . . . . . . . . . . . . . . 105
4.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . 105
4.3.2 Elements of Feynman Diagram . . . . . . . . . . . . . 108
4.3.3 Rules of Feynman Diagram . . . . . . . . . . . . . . . 109
4.3.4 Example: Using Feynman Diagrams to Evaluate Sum
Frequency Generation . . . . . . . . . . . . . . . . . . 111
4.4 Broadening Mechanisms . . . . . . . . . . . . . . . . . . . . . 114
4.5 Introduction to Density Matrices . . . . . . . . . . . . . . . . 117
4.5.1 Phenomenological Model of Damping . . . . . . . . . 121
4.6 Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.7 Sum Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.8 Local Field Model . . . . . . . . . . . . . . . . . . . . . . . . 130
Downloaded from NLOsource.com vii

5 Using the OKE to Determine Mechanisms 135

5.1 Intensity Dependent Refractive Index . . . . . . . . . . . . . 135
5.1.1 Mechanism of χ(3) . . . . . . . . . . . . . . . . . . . . 137
(3)
5.2 Tensor Nature of χijkl . . . . . . . . . . . . . . . . . . . . . . 140
5.3 Molecular Reorientation . . . . . . . . . . . . . . . . . . . . . 143
5.3.1 Zero Electric field . . . . . . . . . . . . . . . . . . . . 147
5.3.2 Non-Zero Electric field . . . . . . . . . . . . . . . . . . 148
5.3.3 General Case . . . . . . . . . . . . . . . . . . . . . . . 149
5.4 Measurements of the Intensity-Dependent Refractive Index . 150
5.5 General Polarization . . . . . . . . . . . . . . . . . . . . . . . 153
5.5.1 Plane Wave . . . . . . . . . . . . . . . . . . . . . . . . 153
5.6 Special Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
5.6.1 Linear Polarization . . . . . . . . . . . . . . . . . . . . 154
5.6.2 Circular Polarization . . . . . . . . . . . . . . . . . . . 154
5.6.3 Elliptical Polarization . . . . . . . . . . . . . . . . . . 155
5.7 Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.7.1 Electronic Response . . . . . . . . . . . . . . . . . . . 157
5.7.2 Divergence Issue . . . . . . . . . . . . . . . . . . . . . 158
5.7.3 One- and Two-Photon States . . . . . . . . . . . . . . 158

6 Applications 163
6.1 Optical Phase Conjugation . . . . . . . . . . . . . . . . . . . 163
6.2 Phase Conjugate Mirror . . . . . . . . . . . . . . . . . . . . . 164

7 Appendix - Homework Solutions 169

7.1 Problem 1.1-1(a to c) . . . . . . . . . . . . . . . . . . . . . . . 169
7.2 Problem: nlosource.com/Polarizability.html . . . . . . . . . . 172
7.3 Problem 4.1-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
7.4 Problem 4.2-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
7.5 Problem 4.3-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
viii
Chapter 1

Introduction to Nonlinear
Optics

The traditional core physics classes include electrodynamics, classical me-

chanics, quantum mechanics, and thermal physics. Add nonlinearity to
these subjects, and the richness and complexity of all phenomena grows
exponentially.
Nonlinear optics is concerned with understanding the behavior of light-
matter interactions when the material’s response is a nonlinear function of
the applied electromagnetic field. In this book, we focus on building a fun-
damental understating of wave propagation in a nonlinear medium, and the
phenomena that result. Such un understanding requires both an under-
standing of the Nonlinear Maxwell Equations as well as the mechanisms of
the nonlinear response of the material at the quantum level.
We begin this chapter with a brief history of nonlinear optics, examples
of some of the more common phenomena, and a non-rigorous but physically
intuitive treatment of the nonlinear response.

1.1 History
1.1.1 Kerr Effect
The birth of nonlinear optics is often associated with J. Kerr, who observed
the change in the refractive index of organic liquids and glasses in the pres-
ence of an electric field. [1, 2, 3] Figure 1.1 shows a diagram of the experi-
ment.
Kerr collimated sunlight and passed it through a prism, essentially mak-

1
2

Spectrometer
Prism

Sun
Collimating Lenses
Aperture E
Hole in
V
the Wall Polarizer 1
Polarizer 2
Sample Between Electrodes
E
Screen
Applied
Field
Polarizer 2 Polarizer 1

Figure 1.1: Kerr observed the change in transmittance through a sample

between crossed polarizers due to an applied voltage. Inset at the bottom
left shows the orientations of the polarizers and the applied electric field due
to the static voltage.

ing a spectrometer that could be used to vary the color of the light incident
on the sample. He placed an isotropic sample between crossed polarizers
(i.e. the polarizers’ axes are perpendicular to each other), so that no light
makes it to the screen. A static electric field is applied 45o to the axis of
each polarizer as shown in the bottom left portion of Figure 1.1. He found
that the transmitted intensity is a quadratic function of the applied voltage.
This phenomena is called the Kerr Effect or the quadratic electrooptic
effect; and, originates in a birefringence that is induced by the electric field.
The invention of the laser in 1960[4] provided light sources with high-enough
electric field strengths to induce the Kerr effect with a second laser in that
replaces the applied voltage. This latter phenomena is called the Optical
Kerr Effect (OKE). Since the intensity is proportional to the square of the
electric field, the OKE is sometimes called the intensity dependent refractive
index,
n = n0 + n2 I, (1.1)

where n0 is the linear refractive index, I is the intensity and n2 the material-
dependent Kerr coefficient.
Downloaded from NLOsource.com 3

E2 E2

Blue Red
Photon Photons

E1 E1

Figure 1.2: (left)A system is excited by a photon if its energy matches the
difference in energies between two states. (right) Two-photon absorption
results when two photons, each of energy (E2 − E1 ) /2, are sequentially ab-
sorbed.

1.1.2 Two-Photon Absorption

Two Photon Absorption (TPA) was first predicted by Maria Goeppert-
Meyer in 1931,[5] and is characterized by an intensity dependent absorption,
of the form,
α = α0 + α2 I, (1.2)

where α0 is the linear absorption coefficient and α2 the two-photon ab-

sorption coefficient. Since the absorption coefficient is proportional to the
imaginary part of the refractive index, a comparison of Equations 1.1 and
1.2 implies that α2 is proportional to imaginary part of n2 . Thus, both the
OKE and TPA are the real and imaginary parts of the same phenomena.
Figure 1.2 can be used as an aid to understand the quantum origin of
TPA. A transition can be excited by a photon when the photon energy
matches the energy difference between two states (sometimes called the
transition energy), h̄ω = E2 − E1 , where ω is the frequency of the light.
Two-photon absorption results when h̄ω = (E2 − E1 ) /2, and two photons
are sequentially absorbed even when there are no two states that meet the
condition h̄ω = E2 − E1 .
We can understand TPA as follows. Imagine that the first photon per-
turbs the electron cloud of a molecule (or any quantum system) so that the
molecule is in a superposition of states |ψ⟩ such that the expectation value
of the energy is h̄ω = ⟨ψ| H |ψ⟩. This is called a virtual state, which will be
more clearly understood when treated rigourously using perturbation the-
ory. If a second photon interacts with this perturbed state, it can lead to
a transition into an excited energy eigenstate. Since the process depends
on one photon perturbing the system, and the second photon interacting
with the perturbed system, the probability of a double absorption is pro-
4

Quartz
Prism
Glass

Ruby
Laser Collimating Lenses

Fundamental Second
Harmonic Photographic
Plate

Exposed
Photographic
Plate

Figure 1.3: The experiment used by Franken and coworkers to demonstrate

second harmonic generation. The inset shows an artistic rendition of the
photograph that recorded the two beams.

portional to the square of the number of photons (the probability of the

first absorption times the probability of the second absorption), and thus
the intensity.

1.1.3 Second Harmonic Generation

Second harmonic generation was the first nonlinear-optical phenomena whose

discovery relied on the invention of the laser. P. A. Franken and coworkers
focused a pulsed ruby laser into a quartz crystal and showed that light at
twice the laser frequency was generated.[6] Figure 1.3 shows the experiment.

A prism was used to separate the the fundamental from the second har-
monic, and the two beams were recorded on photographic film. The film’s
appearance is shown in the inset of Figure 1.3 . Since the second harmonic
signal is weak, the spot on the film appeared as a tiny spec. The story is
that someone in the publications office thought this spec was a piece of dirt,
and air-brushed it away before the journal went to press. So, in the original
paper, an arrow points to the location of the second harmonic spot, but the
tiny spot is not visible.
Downloaded from NLOsource.com 5

Dichroic Reflector
(pass ruby and reflect
Polarizer 1
Xe Flashlamp flash lamp) E Polarizer 2
E
(probe)

Filter Ruby
Sample Laser
Pin Hole (reject ruby)
(pump)
Polarizer 3 E
Filter
(reject ruby)
PMT
Detector

Figure 1.4: In the Optical Kerr Effect, a strong pump laser is polarized 45o
to the weak probe beam, causing a rotation of its polarization.

1.1.4 Optical Kerr Effect

Three years later, the Optical Kerr Effect was observed by Mayer and
Gires.[7] In their experiment, shown in Figure 1.4, a strong light pulse from
a ruby laser and a weak probe beam from a xenon flashlamp counterprop-
agate in the sample. The electric field of the pump and probe beams are
polarized 45o to each other. The pump beam induces a birefringence in the
material that causes a rotation of the polarization of the probe beam. The
probe beam is reflected from a dichroic mirror to the detector while allowing
the pump beam to pass unaltered. The reflected probe beam passes through
polarizer 3, which rejects the original polarization of the probe. Thus, the
amount of light reaching the detector is related to the degree of depolariza-
tion of the probe beam, which in turn, is proportional to the pump intensity.

While the OKE is similar to the quadratic electrooptic effect, in that a

strong electric field rotates the polarization of a weaker probe beam, OKE
has many more diverse applications. Interestingly, two-photon absorption
also contributes to polarization rotation, as you will show in the exercises
below.
6

Problem 1.1-1(a): A monochromatic plane wave passes through a

quartz plate of thickness d with its Poynting Vector perpendicular
to the plate. The k-vector is given by ⃗k = kẑ and the refractive
indices along x̂ and ŷ are nx and ny . If the polarization vector of the
beam when it enters the sample makes an angle θ with respect to
ŷ, calculate the fraction of the beam intensity that passes through a
polarizer that is perpendicular to the light’s original polarization.
(b): Now assume that the refractive index is complex so that
nx → nRx + inx and nx → ny + iny . Recalculate the transmittance
I R I

(transmitted/insident intensity) through the crossed polarizers.

(c): Next assume that the material is isotropic but that a strong
pump beam of intensity Ip is incident at normal incidence and po-
larized along ŷ. If nR R R I I I
y = n0 + n2y I and ny = n0 + n2y I, where the
refractive index changes only along the direction of the pump beam’s
polarization, calculate the transmittance of probe beam through the
crossed polarizer as a function of the pump intensity.

Homework Hint

2nd Polarizer
Perpendicular
to the 1st

Polarizer d

Figure 1.5: Polarizer set perpendicular to the original wave propagation

direction (the first defines the polarization of the incident wave.)

Consider a polarizer that is set perpendicular to the polarization of the

⃗ 0 as is shown in Figure 1.5. The transmittance is
incident electric field E
defined as
It
T = , (1.3)
Ii
Downloaded from NLOsource.com 7

where the initial intensity is calculated from the electric field,

Ii ∝ |E(z
⃗ = 0)|2 . (1.4)

The intensity transmitted through the polarizer is proportional to the square

of the transmitted electric field, or

If ∝ |E(z
⃗ = d) · (cos θx̂ − sin θŷ)|2 , (1.5)

where the constant of proportionality is the same for each intensity calcu-
lation. The electric field of the wave at any point in the material is given
by ( )
E⃗ = E0 sin θx̂ei(knx z−ωt) + cos θŷei(kny z−ωt) , (1.6)

where k = ω/c = 2π/λ.

If the refractive index has both a real and an imaginary component, it
can be expressed as
n = nR + inI , (1.7)
where nR and nI are the real and imaginary parts of the refractive index.
For a pump along ŷ, assume that

nx = n0x (1.8)
ny = n0y + n2y I, (1.9)

where Equations 1.8 and 1.9 are substituted into Equation 1.6 with the
understanding that all quantities are complex.

1.2 Units
Most of the scientific literature in nonlinear optics has accepted the use of
SI units (i.e. meter, kilogram, second, coulomb). However, they lead to ugly
equations and mask the beauty of the unification of electric and magnetic
fields.
For example, consider Faraday’s Law in Gaussian units,

⃗
⃗ = − 1 ∂B .
∇×E (1.10)
c ∂t
8

The units of the electric and magnetic fields are the same, placing them on
an equal footing.
Coulomb’s Law, on the other hand, is given by,

∇·E
⃗ = 4πρ, (1.11)

or for a point charge leads to

⃗ = q.
E (1.12)
r2
Note that ϵ0 and 4π are absent. Also, ϵ0 does not appear in the induced
polarization, which is simply given by
(1)
Pi = χij Ej , (1.13)

where summation convention is implied (i.e. double indices are summed).

There are many conventions for defining the various quantities of non-
linear optics, and this is often a source of much confusion. For example.
Boyd’s book, “Nonlinear Optics,” uses SI units and defines a sinusoidal lin-
early polarized electric field to be of the form,

⃗
E(t) = E x̂ exp (−iωt) + c.c.. (1.14)

I prefer the fields to be of the form cos ωt rather than 2 cos ωt, so I will use
the convention,
⃗ 1
E(t) = E x̂ exp (−iωt) + c.c. (1.15)
2
Other issues of convention will be explained when they are needed.

1.3 Example: Second Order Susceptibility

It is instructive to give a simple example of the ramifications of a nonlin-
ear polarization before using the full rigor of tensor analysis and nonlocal
response. We begin by illustrating the source of second harmonic generation.
Charges within a material are displaced under the influence of an electro-
magnetic field. The displacement of charges can be represented as a series of
electric moments (dipole, quadrupole, etc). Currents can be similarly rep-
resented as a series of magnetic moments. The electric dipole field of bound
charges is usually (but not always) the largest contribution to the electro-
magnetic fields. Next in order of importance is the electric quadrupole and
Downloaded from NLOsource.com 9

magnetic dipole. We use the electric dipole approximation for most of this
book and ignore these higher order terms.
In the linear dipole approximation, the induced dipole moment per unit
volume of material – called the polarization – is proportional to the electric
field. The constant of proportionality that relates the applied electric field
to the polarization is called the linear susceptibility. If the induced polar-
ization, P , is in the same direction as the applied electric field, E, the linear
susceptibility, χ, is a scalar, and we have,

P = χ(1) E. (1.16)

It should be stressed that Equation 1.16 is not a fundamental relationship,

but rather a model of how charges react to an applied electric field. In the
case of Equation 1.16, the model makes an intuitively reasonable assumption
that the charges are displaced along the direction of the applied electric field.
As we will see later, such models can become more complex when applied
to real systems.
If the applied field is sinusoidal, as is the case for an optical field, the
induced dipole will oscillate and reradiate the light. In the linear case, the
dipole moment oscillates at the same frequency as the incident light.
Now consider a nonlinear material. The polarization will be given by,

P = χ(1) E + χ(2) E 2 + . . . , (1.17)

where χ(2) is called the second-order susceptibility. The source of the nonlin-
ear terms ordinates in charges that are bound by an anharmonic potential,
which we will study later. If we substitute the scalar form of Equation 1.14
into the second term of Equation 1.16, we get,

1 2
E 2 (t) = E (exp [−iωt] + exp [+iωt])2
4 ( )
1 2 exp [−i2ωt] + exp [+i2ωt]
= E +1 (1.18)
2 2

Equation 1.18 when substituted into Equation 1.17 clearly shows that the
polarization oscillates at frequency 2ω, which acts as a source that radiates
light at the second harmonic frequency. Also note that the constant offset
(the “+1” term) is the source of the phenomena called optical rectification,
which results in a static polarization, and therefore a static electric field.
The origin of all the nonlinear-optical processes described in Section 1.1
can be explained using the nonlinear polarization.
10

1.4 Maxwell’s Equations

In this section, we introduce the origin of interactions between light and
matter. This requires that we incorporate into Maxwell’s equation the effect
of light on matter. Motivated by fact that electric fields separate charges in
a neutral material, we develop the standard model of describing the charge
distribution as a series of moments. These various moments are effected
by light depending on how the electric field field varies in from point to
point in the material: A uniform electric field induces a dipole moment, a
field gradient the quadrupole, etc. This interaction is formulated using the
concept of a response function.
We begin by introducing conventions and motivate the use of Gaussian
units when expressing Maxwell’s Equations due to the symmetry between
the electric and magnetic fields that result. We assume non-magnetic mate-
rials with no free charges and no free currents. In particular no free charges
and free currents implies the absence of sources such as static charge or
currents due to batteries.
When light interacts with a material it separates the charges from their
equilibrium positions within the constituent molecules or within the lattice
of a crystal. Since these charges are intimately associated with the material,
they are called bound charges, and are “swept under the rug” by the intro-
⃗ which has only free electrons
duction of the electric displacement vector, D,
as its source..
In Gaussian units, Maxwell’s Equations in the absence of free charges
and currents are,

⃗
⃗ = − 1 ∂B ,
∇×E (1.19)
c ∂t
⃗
⃗ = 1 ∂D ,
∇×B (1.20)
c ∂t
∇·D
⃗ = 0, (1.21)
∇·B
⃗ = 0. (1.22)

Note that in Gaussian units E ⃗ and B⃗ have the same units, and are therefore
considered on an equal footing.
To gain familiarity with Gaussian units, conisder the divergence of the
electric field for a point charge q in vacuum. Gauss’ law in Gaussian units
is,
∇·E ⃗ = 4πρ, (1.23)
Downloaded from NLOsource.com 11

where ρ is the charge density, which when integrated over the volume yields,
q
E= . (1.24)
r2
The potential and then the work are given by

q q2
V = and U= . (1.25)
r r
In SI units, ϵ0 , the permittivity of free space, and µ0 , the permeability,
√
yield the speed of light, c ϵ0 µ0 = 1. Indeed, this relationship suggests
that light is an electromagnetic wave. However, ϵ0 and µ are absent in in
Gaussian units. Instead, the speed of light appears explicitly.
It is instructive to compare Coulomb’s law in SI and Gaussian units. For
two electrons separated by an Ångstrom, in SI units the electron charge is
e = 1.6 × 10−19 C, the radius of orbit in a typical atom is about r = 1Å =
·m2
10−10 m, and k = 4πϵ 1
0
= 9 × 109 N C , the resulting force is,

|F | = 2.304 × 10−8 N

expressed in Newtons.
In Gaussian units the charge is expressed in statcoulombs (usually called
“stat coul” or “esu”), as e = 4.8 × 10−18 esu, and the orbital radius is
r = 10−8 cm. In Gaussian units, Equation 1.24 with F⃗ = q E⃗ yields,

|F | = 2.304 × 10−3 dyn,

where the force is expressed in dynes. Applying the conversion factor 1 dyn =
10−5 N , we confirm that the two forces are indeed the same, as expected.

1.4.1 Electric displacement

The electric field in a material is the superposition of the applied electric
field, the field to free charge, and the fields due to all of the displaced charges.
The electric field due to the bound charges can be expressed as a series in
the moments. Ignoring terms higher order than the dipole term, the electric
displaced is given by,
D⃗ =E ⃗ + 4π P⃗ + . . . (1.26)
The higher-order terms will be discussed later. In Equation 1.26, E ⃗ is the
⃗ is the electric displacement due
total field of all charges(bound and free), D
12

E
++ ++ ++ ++ ++ ++ ++
p
- - - - - - - - - - - - - -

Figure 1.6: Material polarized by an external electric field.

to free charges and P⃗ is the electric dipole moment per unit volume, and
accounts for the bound charge.
Figure 1.6 shows a material with an applied electric field and the induced
charge. The electric field polarizes the material, causing the positive and
negative charges to move in opposite directions until they encounter the
surface. The net result is a dipole moment. Note that the polarization has
the same units as the electric field. The electric field due to the polarization
opposes the applied electric field, thus reducing the field strength inside the
dielectric.
An electric field gradient results in a different magnitude of force on two
neighboring electric charges of the same sign. Thus, two identical neighbor-
ing charges will be displaced by differing amounts, resulting in a quadruple
movement. Similarly, higher-order field gradients will couple to higher-order
moments. To take this into account, the electric displacement can be ex-
pressed as a series of moments of the form
∂ ∂2
Di = Ei + 4πPi + 4π Qij + 4π Oijk + . . . , (1.27)
∂xj ∂xj ∂xk

where Qij is the quadrupole moment and Oijk is the octupole moment, etc.1
1
Here we use Summation Convention where,
∑
3
Aij Vj = Aij Vj .
j=1

An index appearing once represents a vector component; an index appearing twice

indicates the component that is to be summed over the three cartesian components.
Downloaded from NLOsource.com 13

Z
E
++ ++ ++ ++ ++ ++ ++

p
- - - - - - - - - - - - - -

E
Figure 1.7: Plot of the electric field across the interface of a polarized ma-
terial.

A vast number of nonlinear-optical processes can be approximated by

the dipole term in Equation 1.27. However, we must be keep in mind that
the dipole approximation does not always hold, especially when there are
large field gradients, as can be found at interfaces between two materials.
Figure 1.7 shoes a plot of the electric field as a function of the surface-normal
coordinate z. The abrupt drop at the interface due to the surface charge
leads to a large surface gradient, so higher order terms need to be taken into
account.

1.4.2 The Polarization

The electric polarization, P⃗ , is the dipole moment per unit volume, and is
a function of the electric field. When the applied electric field strength is
much smaller than the electric fields that hold an atom or molecule together,
the polarization P⃗ (E)
⃗ can be expanded as a series in the electric field,

P⃗ (E)
⃗ = P⃗ (0) (E)
⃗ + P⃗ (1) (E)
⃗ + P⃗ (2) (E)
⃗ + ..., (1.28)

where the first two terms constitute linear optics and the subsequent terms
constitute nonlinear optics.
It is instructive to consider the units of the polarization to confirm that
they are the same as the units of the electric field. Doing a dimensional
14

analysis, we get,
[ ] [ ] [ ]
dipole charge · length charge
[P ] = = = .
volume length3 length2

Comparing Equation smith:Punits with the electric field of a point charge

given by Equation 1.24 confirms that the polarization is of the same units as
the electric and magnetic field. Doing a similar analysis of Equation (1.27)
of the higher-order terms, we can determine the units of the quadrupole,
octupole, and higher-order moments.

1.5 Interaction of Light with Matter

Material
Volume
E

Center Amplitude
larger
Wavelength= E

Figure 1.8: Light inducing a polarization inside a material. The light source
is a monochromatic plane wave.

We will consider the most general case of light interacting with matter
in the dipole approximation. Figure 1.8 shows the material and the beam of
light. We then select a volume, V , that is large enough to contain so many
molecules that the material appears homogeneous, but much smaller than
a wavelength, so that the electric field is approximated uniform over the
Downloaded from NLOsource.com 15
√
volume. The long wavelength approximation that λ ≫ 3 V will generally
hold in the visible part of the electromagnetic spectrum since light has a
wavelength on the order of 1 µm and an atom’s size is on the order of
10−4 µm.
Now we are ready to formulate the most general theory. Figure 1.8
shows an expanded view of the small volume element in Figure 1.9. Over
⃗ is uniform but varying with time. In this
this volume the electric field, E,
case the most general response is ahs the following properties.

Figure 1.9: A small volume element of material in an electric field

1. P⃗ is not parallel to E:
⃗ We assume that the polarization, P⃗ , can point
in any direction, depending on the direction of the electric field and
on the material’s anisotropy. To account for this, the susceptibility
(which relates P⃗ and E)⃗ must be a tensor.

2. P⃗ (E)
⃗ = Series: When the electric field is small, we approximate P⃗ as
a series of the electric field as given by Equation 1.28.

3. The polarization, P⃗ , at point“b” can depend on E ⃗ at point “a”: The

response may be nonlocal if information about the electric field is
transmitted to another point. For example, light absorbed at point“a”
can excite an acoustical wave that propagates to “b” and leads to a
polarization there. Thus, the polarizability at one point in the material
to the electric field everywhere in the material.

4. The polarization at a given time can depend on the electric field in

the past: The polarization response can be delayed relative to the
electric field. For example, an impulse can start the oscillation of a
spring that persists long after the impulse has subsided. Similarly, the
charges in a material may come to rest long after the excitation pulse
16

has left the material. Thus, the polarization at any time depends on
the electromagnetic history of the material.

After the polarization is expanded as a series in the electric field,

∞
∑
P⃗ = P⃗ (n) (r, t) (1.29)
n=0

the most general for of the first-order polarization, P⃗ (1) (r, t), is given
by, ∫ ∫ ∞
(1) 3
Pi (r, t) = d R dτ Tij (r, t; R, τ )Ej (R, τ ), (1.30)
all −∞
space

where Tij (r, t; R, τ ) is the response function, which carries all of the infor-
mation that relates the applied electric field to the polarization. The tensor
form of the response function accounts for anisotropy of the response, and
relates the induced polarization along i to the field polarized along j. The
non-locality of the response resides in the position dependence; that is, the
polarization at coordinate r is related to the electric field at coordinate R.
The integral over all space thus accounts for the influence of every part of
the material on the polarization. The polarization at time t in related by
the response function to the field at time τ . The integral over time thus
takes into account the influence of the the fields in the past.
Clearly, causality must be built into the temporal part of the response
function as does the delay to the retarded potentials. Causality will be
described later, while the small size of the sample will allow us to ignore
retarded potentials.
The seconf0order polarization is of the form,
∫ ∫ ∫ ∞ ∫ ∞
(2) 3 3
Pi (r, t) = d R1 d R2 dτ1 dτ2
all all −∞ −∞
space space (1.31)
Tijk (r, t; (R1 , τ1 ), (R2 , τ2 ))Ej (R1 , τ1 )Ek (R2 , τ2 ).

As in the linear case, the polarization at one point in the material is deter-
mined by the fields at other points in the material, thus the integration over
al space. Again, the polarization at a given time depends on the electric
field at previous times, requiring the integral over time. However, the po-
larization depends on the square of the electric field, so the induced dipole
Downloaded from NLOsource.com 17

vector depends on the directions of the two electric fields, which is taken
into account by the sum over all possible polarizations of both fields. The
response function is therefore a third-rank tensor. Boyd described symme-
try arguments that can be used to reduce the number of independent tensor
components of the response function,[8] thus reducing significantly the com-
plexity of the problem. Details of such symmetry arguments will not be
discussed here.
Finally the most general form of the nth -order polarization is expressed
as,
∫ ∫ ∫ ∞ ∫ ∞
(n) 3 3
Pi (r, t) = d R1 . . . d Rn dτ1 . . . dτn
all all −∞ −∞
space space (1.32)
Tijklm...z Ej (R1 , τ1 ) . . . Ez (Rz , τz ).

Bulk Material
It is instructive to consider the static limit of a spatially uniform electric
field. The polarization is then simply given by,
(0) (1) (2)
Pi = Pi + χij Ej + χijk Ej Ek + . . . , (1.33)
where Pi is the electric dipole moment unit per volume, the first two terms
are focus of linear optics, and χ(n) = nth −order susceptibility.

Molecular
For a molecule, the quantum description of volume is not a well defined
concept. Thus, it is more appropriate to characterize a molecule in a static
electric field by its dipole moment, given by
pi = µi + αij Ej + βijk Ej Ek + γijkl Ej Ek El , (1.34)
where pi is the dipole moment of the molecule; µ is static dipole moment;
α(units of volume) is the polarizability, or otherwise known as the first-
order molecular susceptibility; β is the hyperpolarizability, also known as
the second-order molecular susceptibility; and γ is the second hyperpolariz-
ability, or the third order molecular susceptibility.

Calculating the Nonlinear Susceptibilities

(n) ⃗ by
In light of Equation 1.33, χijk... can be calculated from a model of p⃗(E)
differentiation, 
N ∂ n pi 
(n)
χijk... =  (1.35)
nx !ny !nz ! ∂Ej ∂Ek . . . E=0
⃗
18

where ni represents the number of fields polarized along the ith cartesian
direction and N is the number density of molecules. A similar differentiation
can be applied to the molecular dipole moment given by Equation 1.34 to
get the molecular susceptibilities.

Example:

(a) A model of the polarization gives,

Px = aEx2 . (1.36)
(1) (2)
Determine χxx and χxxx .

(b) Given the polarization

Px = aEx + bEy Ez2 , (1.37)

(1) (3)
find χxy and χxyzz .

Solution:

(a) Taking the derivative as with respect to the field and evaluating it at
zero field 
1 ∂Px 
(1)
χxx = = 2aEx |0 = 0. (1.38)
1! ∂Ex 0

1 ∂ 2 Px 
χ(2) = = a. (1.39)
xxx
2 ∂Ex2 0

(b) Here we apply the same method as in part (a)


(1) 1 ∂Px 
χxy = (1.40)
1! ∂Ey E=0
⃗

and taking the derivative


2
χ(1)
xy = bEz  , (1.41)
⃗
E=0

We get,
χ(1)
xy = 0. (1.42)
Downloaded from NLOsource.com 19

(3)
For χxyzz we can apply the same method, getting

1 1 ∂ 3 Px 
χ(3) =  = b. (1.43)
xyzz
1! 2! ∂Ey ∂Ez ∂Ez E=0
⃗

1.6 Goals
Nonlinear optics encompasses a broad range of phenomena. The goal of
this course is to build a fundamental understanding that can be applied to
both the large body of what is known as well as to provide the tools for
approaching the unknown. In your studies, keep the following questions in
in mind:

• Are the nonlinear optical process related to each each other, and if so,
how?

• What are the underlying microscopic mechanisms?

• How are the microscopic mechanisms related to macroscopic observa-

tions?

• What governs wave propagation in the nonlinear regime?

• How can nonlinear optics be applied?

20
Chapter 2

Models of the NLO Response

In this chapter, we develop the classical nonlinear harmonic oscillator model

of the nonlinear-optical response of a material, show how the nonlinear-
optical susceptibility in the frequency domain is related to the response
function, and show how symmetry arguments can be used to reduce the
number of independent tensor components.

2.1 Harmonic Oscillator

In this section, we begin by developing a model of the static response of a
linear harmonic oscillator and generalize it to the nonlinear case to show
how the nonlinear susceptibility is related to the linear and nonlinear spring
constant. The model is subsequently applied to a nonlinear spring in a
harmonic electric field to develop an understanding of the dependence of
the nonlinear-optical susceptibility on the frequency of the electric field.

2.1.1 Linear Harmonic Oscillator

Consider as an example a charge on a spring as shown in Figure 2.1. For the
static, one dimensional harmonic oscillator model, we apply a static electric
⃗ which causes the spring to stretch.
field, E,
The applied electric field can be so large that it becomes unraveled and
leads to anharmonic terms. But, we assume the field is small enough to
keep the spring intact. This is analogous to a molecule where the applied
field is large enough to strongly displace the electrons from their equilibrium
positions, but not strong enough to ionize the molecule.
Figure 2.1.1 shows a free body diagram illustrating the forces acting on

21
22

m, q

x
E

Figure 2.1: A charge on a spring with mass, m, and charge, q, at a distance,

x, from the origin. There is an applied external electric field,E.

q
Fs = kx FE= qE

Figure 2.2: Free body diagram for a charge attached to a spring with an
applied electric field.

the charge. The force on the charge due to the electric field is balanced by
an opposing force due to the tension of the spring.
At equilibrium, the forces are equal to each other, that is,

Fs = FE . (2.1)

Substituting the forces into Equation 2.1 yields,

kx = qE. (2.2)

Solving for x,
q
x=E, (2.3)
k
and using the definition of the dipole moment,

p = xq, (2.4)

substitution of Equation 2.3 into Equation 2.4 yields

q2
p= E. (2.5)
k
The static linear susceptibility is defined as,

∂p 
α= , (2.6)
∂E E=0
Downloaded from NLOsource.com 23

which applied to Equation 2.5 yields the linear susceptibility,

q2
α= . (2.7)
k
When k is small, the charge is loosely bound and large displacements re-
sult from small applied fields. When q is large, the linear susceptibility, or
polarizability, is large because the applied force scales in proportion to q.
For N springs per unit volume, as shown in Figure 2.3, we get

χ(1) = N α. (2.8)

Substituting Equation 2.7 into Equation 2.8,we get

q2
χ(1) = N . (2.9)
k
Note that Equation 2.9 explicitly assumes that the springs are non-interacting,
and this approximation is often called the dilute gas approximation.

Figure 2.3: A system of N charges on springs with an applied electric field.

The susceptibility is linear with respect to N for low concentrations.

However, as N becomes sufficiently large, the susceptibility diverges from
linearity (as depicted in Figure 2.4). In systems such as dye-doped polymers,
deviations from linearity is a sign of molecular aggregation.

2.1.2 Nonlinear Harmonic Oscillator

Now we will consider the nonlinear harmonic oscillator model. In this sys-
tem, the potential has a quadratic dependence on the displacement, x. Real
systems deviate from the quadratic potential of a simple harmonic oscilla-
tor when the amplitude of x gets large (Figure 2.5). In this limit, we can
24

χ(1)

N
Figure 2.4: Susceptibility as a function of N, the number of molecules. With
large N, the susceptibility diverges from linearity because the molecules will
begin to aggregate.

V
molecule
ionization

small
E
x
0
Figure 2.5: The electric potential as a function displacement. As x gets
large, the potential no longer follows a quadratic equation and the molecule
will ionize beyond some critical displacement. We will consider small dis-
placements when E is small.

approximate x is small when the electric field, E, is small. The spring force
is,
Fs = k (1) x + k (2) x2 . (2.10)
In equilibrium, the spring force is equal to the electric force so that,
k (1) x + k (2) x2 = qE. (2.11)
Rearranging Equation 2.11 yields,
k (1) q
x2 + x − (2) E = 0. (2.12)
k (2) k
Solving for x, we get
√
( )2
−k (1) 1 k (1) 4q
x= ± + E, (2.13)
2k (2) 2 k (2) k (2)
Downloaded from NLOsource.com 25

( )2
k(1)
and factoring out k(2)
,
 √ 
( )2
−k (1) 1 ∓ 4q k (2)
x= 1+ E . (2.14)
2k (2) k (2) k (1)

For a small electric field, we can expand the argument and get
√ ( )2
( )2
4q k (2) 1 4q k (2)
x = 1 + (2) E =1+ E + .... (2.15)
k k (1) 2 k (2) k (1)

Using the series expansion of the square root, and keeping only the negative
root because it yields the correct linear result, we get,
( )2
k (1) 1 4q k (2)
x = E + ...
2k (2) 2 k (2) k (1)
q
x = (1)
E + .... (2.16)
k
2
Note that this is the same result as the linear case where α = qk .
Expanding the position given by Equation 2.16 to include higher-order
terms, and using Equation 1.35, the susceptibilities are,

χ(1) = N α (2.17)
q2
= N (1) (2.18)
k
and (2.19)

(2) 1 ∂ 2 P 
χ = (2.20)
2 ∂E 2  E=0
−N q k (2)
3
= , (2.21)
(k (1) )2 k (1)

where N is the number density. Note that as k (2) → 0 then χ(2) → 0,

in agreement with the fact that the nonlinearity originates in k (2) . The
third-order susceptibility depends quadratically on k (2) and is of the form,

(3) 1 ∂ 3 P 
χ =
6 ∂E 3 E=0
( )2
2N q 4 k (2)
= . (2.22)
(k (1) )3 k (1)
26

It is useful to review the units for the susceptibilities in Gaussian units.

They are:

[ ]
α = cm3 χ(1) = [dimensionless] ,
[ 5] [ cm ] [ cm2 ] [ cm3 ] 12
cm (2)
β= χ = = = ,
esu stat volt esu erg
[ 6] [ 5]
cm (3) cm
γ= χ = .
erg erg

Orders of magnitude ranges for typical molecular susceptibilities are:

α 10−21 − 10−22 cm3 ,

cm5
β 10−27 − 10−30 ,
esu
cm6
γ 10−36 .
erg
In order to observe these nonlinear effects, the field strengths must be
large enough to induce a displacement of charge but small enough so that
the molecules do not ionize. For example, using typical field strengths for
hydrogen atoms, the field strength must be on the order
V
E << 1011 ,
m
stat volt
E << 107 .
cm

2.1.3 Non-Static Harmonic Oscillator

For the non-static 1D linear harmonic oscillator model, the equation of mo-
tion of an electron derived from Newton’s Laws, is,

F = mẍ
= −mω02 x − mΓẋ − eE0 cos(ωt), (2.23)

where −mω02 x is the restoring force, mΓẋ is the damping force, and eE0 cos(ωt)
is the driving force on the electron. For the nonlinear case, the force includes
higher-order terms in the potential,
( )
F = −mω02 x − mΓẋ − eE0 cos(ωt) − ξ (2) x2 + ξ (3) x3 + . . . . (2.24)
Downloaded from NLOsource.com 27

In the linear case ξ (n) = 0 for n > 2.

A harmonic field of the form,

E = E0 cos(ωt), (2.25)

can be expressed in complex form,

E0 −iωt
E= e + c.c., (2.26)
2
where the electric field is the real part
[ ]
E = Re E0 eiωt . (2.27)

Once transients have decayed, it is commonly assumed that the electron

oscillates at the driving frequency so that a trial solution of the form,

A −iωt A∗ iωt
x= e + e , (2.28)
2 2
is used. The phase of A takes into account the phase shift between the
driving field and the displacement of the electron. Alternatively, we can
express the position of the electron by

x = x0 cos(ωt + ϕ) (2.29)

or using a trigonometric identity,

x = x0 (cos(ωt) cos(ϕ) − sin(ωt) sin(ϕ)) . (2.30)

For the linear case ξ (n) = 0 for n > 2, we show that these two representations
are equivalent with
A = AR + iAI , (2.31)
as follows.
The first order solution is of the form,
[ ]
(1) A −iωt
x = Re e (2.32)
2
[ ]
AR + AI i
= Re (cos(ωt) − i sin(ωt)) (2.33)
2
= Re [AR cos(ωt) + AI sin(ωt) + i ∗ stuff] (2.34)
= AR cos(ωt) + AI sin(ωt), (2.35)
28

π
where AR cos(ωt) is in phase with the field and AI sin(ωt) is 2 out of phase
with the field. Solving for the components,

AR = x0 cos(ϕ), (2.36)
AI = −x0 sin(ϕ), (2.37)
( )
AI
ϕ = − tan−1 , (2.38)
AR

and √
x0 = (A2R + A2I ). (2.39)
Equations 2.36 through 2.39 show that one can transform between (Ar , AI )
to (x0 , ϕ).
Substituting the ansatz given by Equation 2.28 into the linear equation
of motion given by Equation 2.23 and solving for A, we get
−eE0 1
A= . (2.40)
m ω02 − 2iγω − ω 2
(( ) )
Multiplying the numerator and denominator by ω02 − ω 2 + 2iγω , A be-
comes,
−eE0 (ω02 − ω 2 ) + 2iγω
A= . (2.41)
m (ω02 − ω 2 )2 + 4γ 2 ω 2
The dipole moment can be written as,

P (1) = −eN x(1) . (2.42)

Substituting Equation 2.28 into Equation 2.23, with the help of Equation
2.41 we get,
e2 E0 N 1 −iωt
P (1) = e + c.c. (2.43)
2m D(ω)
But,
(1) (1)
Pω −iωt P−ω iωt
P (1) = e + e , (2.44)
2 2
where Pω∗ = P−ω . These are the Fourier components at ω and −ω. Then
the dipole moment can be written as
[ ]
P (1) = Re χ(1) E
χ(1)
= E0 e−iωt + c.c. (2.45)
2
Downloaded from NLOsource.com 29

Setting Equations 2.43 and 2.45 equal to each other and taking the Fourier
component at ω, we get

e2 N (ω02 − ω 2 ) + 2iΓω
χ(1) =
m (ω02 − ω 2 )2 + 4Γ2 ω 2
e2 N 1
= . (2.46)
m D(ω)

Figure 2.6 illustrates the real and imaginary components of χ(1) . The real
part represents the in-phase component while the imaginary part is the π2
out of phase component of the induced dipole moment with the driving field.

χ (1)
R
Imaginary
Real
χ(1)
I

ω
ω0

Figure 2.6: The real and imaginary parts of the susceptibility of a harmonic
oscillator.

In the second-order non-linear case, ξ (2) ̸= 0 but ξ (n) = 0 for n > 2.

Using classical perturbation theory, we assume that the solutions are of the
form x = λx(1) + λ2 x(2) . For small fields, E = λE0 cos(ωt). We already
solved this equation to first order where we calculated x(1) . For order x(2)
(calculated by considering terms of order λ) the force equation gives
[ ]
m ẍ(2) + 2Γẋ(2) + ω02 x(2) − ξ (2) (x(1) )2 = 0, (2.47)

where ξ (2) (x(1) )2 acts as the driving term. Thus, we must first solve for x(1)
before we can solve for x(2) . For the general case of 2 fields,
E1 −iω1 t E2 −iω2 t
E= e + e + c.c, (2.48)
2 2
and
x(1) = x(1) (1)
ω1 + xω2 . (2.49)
30

Then,
A1 −iω1 t A∗1 iω1 t A2 −iω2 t A∗2 iω2 t
x(1) = e + e + e + e (2.50)
2 2 2 2
Substituting Equation 2.50 into Equation 2.47, we get terms of the form
e−i2ω1 t , e−i2ω2 t , e−i(ω1 ±ω2 )t , and a constant in time. Thus, we solve the
problem by projecting out each of these Fourier components. To illustrate,
we project out Fourier component at frequency 2ω1 ,
∫ inf ( [ ] )
dω m ẍ(2) + 2Γẋ(2) + ω02 x(2) − ξ (2) (x(1) )2 = 0 cos(2ω1 t), (2.51)
− inf

which yields,
[ ] [ ]
(2) (2) (2) (2) m A2 −2iω1 t
m ẍ2ω1 + 2Γẋ2ω1 + ω02 x2ω1 =ξ e + c.c , (2.52)
2 2
(2) (2) (2)
where we have used the fact that x(2) = x2ω1 + x2ω2 + xω1 ±ω2 + . . .. Note
that Equation 2.52 is the same form as Equation 2.23. Solving Equation
2.52 using the same approach that is used to solve Equation 2.23 yields the
displacement,
(2) ( e )2 E12
x2ω1 = ξ (2) 2 m , (2.53)
D (ω1 )D(2ω1 )
where E1 is the amplitude of the field at ω1 . To solve for the response at the
sum frequency, we project out the ω1 + ω2 Fourier component of Equation
2.47. This yields,
e 2
(2) (m ) E1 E2
xω1 +ω2 = −2ξ (2) . (2.54)
D(ω1 + ω2 )D(ω1 )D(ω2 )
We can use a similar approach to get the response at other frequencies. The
nonlinear susceptibilities can be calculated from Equations 2.53 and 2.54 by
recognizing that the dipole moment is given by p = −ex with a polarization
given by P = N p, and then applying Equation 1.35.

2.2 Macroscopic Propagation

In the previous sections, we developed intuition about the bulk nonlinear
processes and then studied the nonlinear oscillator to learn about the micro-
scopic models at the molecular level. Next we considered the macroscopic
behavior of the non-interacting gas model, where we related χ(n) to α, β
and γ. Here we study the bulk response independent of the the source of
Downloaded from NLOsource.com 31

the response. First we consider macroscopic propagation crudely to gain

physical insight, then we study it in more detail using the full mathematical
formalism.
Understanding χ(2) (ω1 , ω2 ) enables us to describe a number of processes
like second harmonic generation(SHG), sum frequency generation(SFG) and
optical rectification(OR). On the other hand, χ(2) is a function of frequencies
ω1 and ω2 , which provides significant information about the characteristics
of the system.
Two frequencies are sufficient to understand all the characteristics of
χ because P (2) is proportional to the product of two fields like E ω1 E ω2 ,
(2)

E ω1 E ω3 and so forth. Thus the fields in χ(2) always appear in pairs and any
pair of fields are represented by χ(2) (ω1 , ω2 ). Similarly, n frequencies are
required to describe the nth order nonlinear response.
As mentioned above, studying the resonance frequencies of a system is
a general approach to explore its characteristics. The resonance frequency
is defined as a frequency for which the real part of the responses (e.g. χ(1) )
changes sign and the imaginary part is characterized by a peak. χ(2) and
the higher order processes show more complete resonance structures. To
describe the nonlinearities of a system, one needs to characterize all the
frequencies. Quantities such as χ(1) and its resonance frequencies provide
some information about the system. χ(2) can give more information because
it is sensitive to different symmetries than is χ(1) . Therefore resonances in
χ(2) may indicate states that can’t be seen as features in the linear absorption
spectrum. As an example, nonlinear microscopy is used in biological systems
to bring out certain structures that are associated with orientation order and
two-photon states.
To begin, let’s look at the resonance frequency for the simple case of χ(1) .
In Figure 2.7, we plot the imaginary and real parts of χ(1) versus frequency
ω. The natural (resonant) frequency of the oscillation of the system is ω0 .
For χ(2) we have

2 (qN x ) 
1 ∂ i 
χ(2) (ω1 , ω2 ) = ω2  , (2.55)
D ∂Ej ∂Ek 
ω1
E=0

where N is the number density of the molecules, qx is dipole moment, qN x

is the polarization and D is the degeneracy factor, which should not be
confused with the energy denominator D(ω1 , ω2 ). In previous cases, D de-
pended on the tensor components of the frequencies of the system, but for
χ(2) , it also depends on ω1 and ω2 . The surface plot of χ(2) as a function
of frequencies becomes complicated since the energy denominators in χ(2)
32

Absorption
Imaginary

Real

ωo ω

Figure 2.7: Plot of the real and imaginary parts of χ(1) . ω0 is the resonant
frequency of the system.

include different combination of frequencies. Hence it is more interesting to

visual the resonances with a two-dimensional plot of the second-order sus-
ceptibility as a function of the frequencies ω1 and ω2 . In the ω2 -ω1 plane, one
can determine where the resonances occur. Since the energy denominators
of χ(2) are of the form D(ω1 ), D(ω2 ), D(2ω1 ) and D(2ω2 ) the system has
resonances when ω1 = ω0 , ω2 = ω0 , ω1 = ω0 /2, ω2 = ω0 /2, and ω2 ±ω0 = ω0 .
These resonances are shown in Figure 2.8, where the various lines shows the
position of the resonances.
A simple macroscopic model can be used to explain the concept of re-
fractive index. In Figure 2.9, plane waves are incident from the left on the
sample. This external electric field, E ω , induces dipoles in the molecule,
which consequently generate dipole fields. The superposition of the incident
and induced electric fields propagates through the material then exits to
the right. The induced dipoles oscillate at the frequency of the plane wave.
Assuming the plane waves and the sample, which is made of dipoles, are
infinite in the transverse direction, the superposition of the induced dipole
fields leads to planes of constant phase that are parallel to the incoming
wave’s phase fronts. The wavelength of the reflected dipole fields propagat-
ing in the opposite direction of the incident wave to the left (Figure 2.9) is
also the same as of the plane wave.
The superposition of the incident and dipole fields inside the material
leads to a phase shift relative to the incident wave and of lower phase velocity
Downloaded from NLOsource.com 33

- =

- =

+ =
/2
0

/2
0

Figure 2.8: Map of resonances of χ(2)

in the sample. The refractive index quantifies the reduced phase velocity
inside the material, hence, the wave propagates more slowly, and the peaks
are closer together. Outside, the net effect is a phase shift of the transmitted
wave.
When a nonlinear material is placed in the path of a beam of light of
frequency of ω (for now, consider only a second order nonlinearity, χ(2) ̸= 0),
the dipoles will oscillate at frequency 2ω and if the field is of the form
E ω1 + E ω2 , then the frequencies of oscillation are 0, 2ω1 , 2ω2 ,ω1 + ω2
and |ω1 − ω2 |. Therefore, outside the sample we observe fields with other
frequencies. For χ(1) , radiation with frequencies ω1 and ω2 are observed.
Next, we use Maxwell’s equations to rigorously solve for nonlinear wave
propagation. First, we find a relationship for ϵ0 when the nonlinear response
vanished and then, we explore how the refractive indices are generalized for
systems in which χ(2) and χ(3) are dominant.
In Gaussian units,
↔
D = E + 4πP = ε E, (2.56)
↔
where ε is a rank two tensor. Note that in Gaussian units, The permittivity
of free space, ϵ0 , is absent. When the nonlinear response vanishes,

P = χ(1) E. (2.57)

Then ( )
D = E + 4πχ(1) E = 1 + 4πχ(1) E, (2.58)
34

ω
E

Figure 2.9: Illustration of the macroscopic model that defines refractive in-
dex. Longer (red) lines represent phase fronts of the external electric field
in and outside the material and the shorter (dark blue) lines represent the
electric field generated by the induced dipoles fields (arrows) inside the sam-
ple. The superposition of the induced electric dipole fields and plane waves
inside the sample leads to a plane wave with a decreased phase velocity,
which defines the concept of the refractive index.

whence
ϵ0 = 1 + 4πχ(1) , (2.59)
or in tensor form,
(1)
ϵij = δij + 4πχij . (2.60)
ϵ0 is the linear dielectric constant of the material. The method used to
find ϵ0 , through Equation 2.59 provides a simple approach to quantifying
nonlinear wave propagation without solving Maxwell’s equations.
To obtain the wave equation for the regions inside the material, we apply
Faraday’s and Ampere’s laws. Based on Faraday’s law

1 ∂B
∇×E=− , (2.61)
c ∂t
and taking the curl of Equation 2.61, leads to

1∂
∇×∇×E=− ∇ × B. (2.62)
c ∂t
Downloaded from NLOsource.com 35

Inside the sample, where there are no free currents – as, for example,
one finds when an external source is applied – Ampere’s law gives,

1 ∂D
∇×B= , (2.63)
c ∂t
where H is equivalent to B when the material is nonmagnetic and the mag-
netic susceptibility is equal to unity. Inserting Equation 2.63 in Equation
2.62 yields
ϵ0 ∂ 2 E 4π ∂ 2 PN L
∇×∇×E+ 2 2 =− 2 , (2.64)
c ∂t c ∂t2
where the linear polarization, 4πχ(1) E, is absorbed into the second term of
Equation 2.64. The nonlinear polarization, PN L , is given by

PN L (E) = χ(2) E2 + χ(3) E3 + · · · = P(2) + P(3) + · · · . (2.65)

In this section, we will only consider terms up to third order in the electric
field.
In the following treatment, we assume that the material is local in space
and time. This implies that the polarization at one point of the material is
not affected by the field at any other point. These restrictions can later be
relaxed, but the phenomena that result will be the same.
If the sample is exposed to external fields with frequencies of ω1 and ω2 ,
the total electric field in scalar form (for simplicity) is given by,

E ω1 −iω1 t E ω2 −iω2 t
E= e + e + c.c., (2.66)
2 2
and the polarization P , is given by

P 2ω1 −2iω1 t P 2ω2 −2iω2 t P ω1 +ω2 −i(ω1 +ω2 )t

P = e + e + e + · · · + c.c. (2.67)
2 2 2
We take
PN L = χ(2) E 2 , (2.68)
and in Equation 2.64 project out the Fourier component of interest. First
let’s calculate E 2 using Equation 2.66

(E ω1 )2 −2iω1 t (E ω2 )2 −2iω2 t E ω1 E −ω1

E2 = e + e + +
4 4 2
E ω1 E −ω2 −i(ω1 −ω2 )t
e + · · · + c.c. (2.69)
4
36

To project out the 2ω1 component we calculate the integral

∫ ∞ ( )
e−2iω1 t P = χ(2) E 2 dt, (2.70)
−∞

which yields
1
P 2ω1 = χ(2) (E ω1 )2 . (2.71)
2
The coefficient of 1/2 in Equation 2.71 arises from squaring the field.
This coefficient for other process is calculated in the same way – that is, the
desired fourier component of the polarization is calculated after taking the
superposition of input fields to the nth power for an nth -order response.
For the next case, P 0 ,

P 0 = χ(2) E −ω1 E ω1 . (2.72)

Despite the fact that there was no applied static field, the driving field
produces this field through nonlinear interaction with the sample. The same
procedure applied to output frequency ω1 + ω2 gives

P ω1 +ω2 = χ(2) E ω1 E ω2 . (2.73)

Now, let’s assume there are two fields E ω and E 0 . The field with zero
frequency is treated separately because it is usually expressed as a real quan-
tity. In this case, the total field is
( ω )
E −iωt
E= e + c.c. + E 0 . (2.74)
2

Then, squaring Equation 2.74 leads to

(E ω )2 −2iωt
E 2 = E ω E 0 e−iωt + e + ··· , (2.75)
4
whence
P (2) = 2χ(2) E ω E 0 . (2.76)
χ(3) , in general, is written in terms of ω1 , ω2 and ω3 as the frequencies
of the three incident fields. However, this general case is too complicated to
serve as an example. So, for the sake of simplicity, let’s assume there is only
one field of frequency ω,
E ω −iωt
E= e + c.c. (2.77)
2
Downloaded from NLOsource.com 37

From Equation 2.77,

(E ω )3 −3iωt 3 ω 2 −ω
E3 = e + (E ) E + · · · + c.c., (2.78)
8 8
whence,
1 ω 3 (3)
P 3ω =
(E ) χ . (2.79)
4
Following the same procedure for P ω yields
3
P ω = χ(3) E ω E −ω E ω , (2.80)
4
where E ω E −ω = |E ω |2 . ( Recall that E ω∗ = E −ω .)
To understand how nonlinear interactions lead to changes in the effective
refractive index, we solve a specific problem in which a field with frequency
ω interacts with a static field. Consider a beam that travels through a
material between capacitor plates, which are used to apply a static electric
field. Ignoring second-order nonlinearities (and higher than third order), we
get

Dω = Eω + 4πχ(1) Eω + 8πχ(2) E 0 Eω
( )
= 1 + 4πχ(1) + 8πχ(2) E 0 Eω
= ϵ Eω , (2.81)

where ϵ0 = 1 + 4πχ(1) and

ϵ = ϵ0 + 8πχ(2) E 0 . (2.82)

This indicates that the resulting field, generated by the interaction of

incident field and the sample oscillates at frequency ω. The relationship
between refractive index n and permittivity ϵ is given by
√
n = ϵ
√
1/2 8πχ(2) E 0
= ϵ0 1+ , (2.83)
ϵ0

which for small fields yields

( )
4πχ(2) 0
n = n0 1+ E . (2.84)
ϵ0
38

Finally,
4πχ(2) 0
n = n0 + E , (2.85)
n0
where n0 = ϵ20 . Equation 2.85 shows how the refractive index depends on the
static field. The coefficient of E 0 in Equation 2.85 is called the electro-optic
coefficient, n1 ,
4πχ(2)
n1 = . (2.86)
n0
Considering now only the third order susceptibility, χ(3) , for only one
applied field,
3
P ω = χ(1) E ω + χ(3) |E ω |2 E ω . (2.87)
4
Once again we use Equation 2.56 in scalar form. ϵ is then given by
3
ϵ = 1 + 4πχ(1) + (4π)χ(3) |E ω |2 , (2.88)
4

where we have neglected χ(2) , a good approximation for a centrosymmetric

material such as a liquid. In analogy to Equation 2.83, the refractive index
for the third-order response is

3πχ(3) ω 2
n = n0 + |E | . (2.89)
2n0
The optical kerr effect coefficient, n2 , is given by

3πχ(3)
n2 = . (2.90)
2n0
In general, depending on the symmetry properties of the material, either
Equations 2.85 or 2.89 dominate. For example, in a centrosymmetric sample,
where χ(2) is zero, the refractive index is calculated via Equation 2.89.

2.3 Response Functions

In general, the polarization of a material depends on a response function
which characterizes the material. For the first order case the polarization is
∫∞
(1) (1)
Pi (t) = Tij (t; τ )Ej (τ )dτ , (2.91)
−∞
Downloaded from NLOsource.com 39

where we have assumed that the response is spatially local, and the response
function is Tij .1 This form for the polarization takes into account how fields
at different points in time,τ , will contribute to the material’s response at
time t. To understand some of the properties of response functions, we will
begin with the first order case, and then we will generalize the approach to
higher orders.

2.3.1 Time Invariance

One of the most important axioms of physics is that the laws of physics are
(1)
invariant under time translation. So if we let t → t + t0 in, Pi and Ej the
laws of physics should remain unchanged. Advancing the polarization by
t → t + t0 in equation 2.91 yields
∫
(1) (1)
Pi (t + t0 ) = Tij (t + t0 ; τ )Ej (τ )dτ . (2.92)

Advancing the polarization and fields by t → t + t0 yields:

∫
(1) (1)
Pi (t + t0 ) = Tij (t; τ )Ej (τ + t0 )dτ . (2.93)

Making the substitution τ → τ ′ in equation 2.92 and τ ′ = τ + t0 in equation

2.93, equations 2.92 and 2.93

∫ ∫
Tij (t + t0 ; τ ′ )Ej (τ ′ )dτ ′ = Tij (t; τ ′ − t0 )Ej (τ ′ )dτ ′ .
(1) (1)
(2.94)

If equation 2.94 is to hold for all Ej (τ ′ ), then it follows that

Tij (t + t0 ; τ ′ ) = Tij (t; τ ′ − t0 ).

(1) (1)
(2.95)
with t = 0 and t0 → t′ we get,

Tij (t′ ; τ ′ ) = Tij (0; τ ′ − t′ ),

(1) (1)
(2.96)

we see that the response depends only on τ ′ −t′ . Now changing from T (t, τ )(a
function of two variables) to R(τ − t) (a function of one variable) we can
rewrite the polarization as given by equation 2.91
1
In this section limits of integrals are assumed to be from -∞ to ∞ unless otherwise
stated.
40

∫
(1) (1)
Pi (t) = Rij (t − τ )Ej (τ )dτ . (2.97)

(1)
If we enforce causality by setting Rij (t − τ ) = 0 for τ > t (i.e. demanding
that a polarization is possible only after the application of a field and defining
τ ′ = t − τ , we get

∫∞
Rij (τ ′ )Ej (t − τ ′ )dτ ′ .
(1) (1)
Pi (t) = (2.98)
0

The result is a description of the polarization of a material at time t due to

fields acting locally over all times in the past.

2.3.2 Fourier Transforms of Response Functions: Electric

Susceptibilities
In the previous section we derived the polarization of a material as a function
of time due to an electric field. Often it is more convenient and appropriate
to consider the polarization in frequency space. The Fourier Transforms of
the electric field are defined by
∫
Ej (t) = dω Ẽj (ω)e−iωt (2.99)

and ∫
1
Ẽj (ω) = dtEj (t)eiωt . (2.100)
2π
Substituting Ej (t) into equation 2.98 yields
∫ ∫
′
Pi (t) = Rij (τ ) Ẽj (ω)e−iω(t−τ ) dωdτ ′ ,
(1) (1) ′
(2.101)

which we can regroup into the form,

∫ (∫ )
′
Rij (τ ′ )e−iωτ dτ ′ Ẽj (ω)e−iωt dω.
(1) (1)
Pi (t) = (2.102)

The term in parentheses in equation 2.102 is the inverse Fourier trans-

(1)
form of Rij , which we define as
∫
′
χij (ω) = Rij (τ ′ )e−iωτ dτ ′ ,
(1) (1)
(2.103)
Downloaded from NLOsource.com 41

the first order electric susceptibility. Substituting equation 2.103 into equa-
tion 2.102, we find
∫
χij (ω)Ẽj (ω)e−iωt dω.
(1) (1)
Pi (t) = (2.104)

Next, we transform the polarization into frequency space according to,

∫
(1) 1 (1)
Pi (ωσ ) = Pi (t)eiωσ t dt (2.105)
2π
substituting equation 2.104 into equation 2.105 yields
∫ ∫
1
χij (ω)Ẽj (ω)e−iωt eiωσ t dωdt.
(1) (1)
Pi (ωσ ) = (2.106)
2π
Recall that the dirac delta function can be expressed as,
∫
1
δ(ω − ωσ ) = e−it(ω−ωσ ) dt, (2.107)
2π
we can rewrite equation 2.106 as
∫
(1) (1)
Pi (ωσ ) = χij (ω)Ẽj (ω)δ(ω − ωσ )dω. (2.108)

Performing the integration yields

(1) (1)
Pi (ωσ ) = χij (ωσ )Ẽj (ωσ ). (2.109)
Thus we have found that in the frequency representation, the polarization is
a simple relationship between the electric field and the electric susceptibility,
where the susceptibility is defined as the Fourier transform of the response
function.

2.3.3 A Note on Notation and Energy Conservation

Throughout our book we will be focusing on electric susceptibilities, which in
general can depend on multiple frequencies, and the order in which they ap-
pear. In anticipation of future needs we introduce the notationχ(n) (−ωout ; ωin ).
The angular frequency to the left of the semi colon represents the outgo-
ing wave, or the frequency we are measuring in the experiment, the angular
frequencies on the right represent all the incoming fields. The negative sign
on the outgoing field represents energy leaving the material, where as the
42

positive frequencies to the right of the semicolon represent energy enter-

ing
∑ the system.
∑ This notation explicitly denotes energy conservation where
ωin + ωout = 0.
For the first order susceptibility the polarization is expressed as

(1) (1)
Pi (ω) = χij (−ω; ω)Ẽj (ω), (2.110)
where the sum of input frequencies and output frequencies is simply −ω +
ω = 0.

2.3.4 Second Order Polarization and Susceptibility

For the second order case, we begin with the relationship between the po-
(2)
larization P (2) (t) and response function Rijk ,

∫ ∫
(2) (2)
Pi (t) = Rijk (τ1 , τ2 )Ej (t − τ1 )Ek (t − τ2 )dτ1 dτ2 , (2.111)

(2)
where Rijk (τ1 , τ2 ) = 0 if τ1 or τ2 = 0. Substituting the Fourier transform
again of the electric field using equation ??, we get

∫ ∫ (∫ ) (∫ )
(2) (2) −iω1 (t−τ1 ) ˜ −iω2 (t−τ2 )
Pi (t) = Rijk (τ1 , τ2 ) Ẽj (ω1 )e dω1 Ek (ω2 )e dω2 dτ1 dτ2
∫ ∫ ∫ ∫
Rijk (τ1 , τ2 )Ẽj (ω1 )E˜k (ω2 )e−iω1 (t−τ1 ) e−iω2 (t−τ2 ) dω1 dω2 dτ1 dτ2
(2)
=
∫ ∫ ∫ ∫
dω1 dω2 dτ1 dτ2 Rijk (τ1 , τ2 )e−iω1 τ1 e−iω2 τ2 Ẽj (ω1 )E˜k (ω2 )e−iω1 t e−iω2 t .
(2)
=
(2.112)

In a manner similar to the derivation of equation 2.109 we can simplify

the expression by introducing the second order susceptibility,
∫ ∫
dτ1 dτ2 Rijk (τ1 , τ2 )e−iω1 τ1 e−iω2 τ2 .
(2) (2)
χijk (ω1 , ω2 ) = (2.113)

The second-order polarization, then, is related to the second-order suscep-

tibility,

∫ ∫
dω1 dω2 χijk (ω1 , ω2 )Ẽj (ω1 )E˜k (ω2 )e−iω1 t e−iω2 t .
(2) (2)
Pi (t) = (2.114)
Downloaded from NLOsource.com 43

(2)
We once again take the inverse Fourier transform of Pi (t), yielding
∫
(2) 1 (2)
Pi (ω) = Pi (t)eiωt dt. (2.115)
2π
substituting equation 2.114 into equation 2.115 yields,

∫ ∫ ∫
dω1 dω2 dtχijk (ω1 , ω2 )Ẽj (ω1 )E˜k (ω2 )ei(ω1 +ω2 −ω)t (2.116)
(2) (2)
Pi (ω) =
∫ ∫
(2)
= dω1 dω2 χijk (ω1 , ω2 )Ẽj (ω1 )E˜k (ω2 )δ(ω1 + ω2 − ω). (2.117)

Performing the integration we get,

(2) (2)
Pi (ω) = χijk (−ω; ω1 , ω2 )Ẽj (ω1 )E˜k (ω2 ), (2.118)
where again, we use the notation where all frequency arguments sum to zero.

2.3.5 nth Order Polarization and Susceptibility

We can easily generalize the method from the previous section to find higher-
order polarizations and susceptibilities. The general form of the nth -order
polarization is

(n) (n)
Pi (ω) = χijkl... (−ω; ω1 , ω2 , ....ωn )Ej (ω1 )Ek (ω2 )... (2.119)

where χ(n) is given by

∫ ∫ ∫
(n) (n)
χijk... (−ω; ω1 , ω2 , ....ωn ) = dτ1 dτ2 .... dτn Rijk... (τ1 , τ2 .....τn )eiω1 τ1 eiω2 τ2 ....eiωn τn .
(2.120)

2.3.6 Properties of Response Functions

Now that we have found a general relationship between the polarization
and the applied fields, we now consider the properties of response functions.
Response functions have full permutation symmetry, meaning that if we
have a polarization
∫ ∫
(n) (n)
Pi (t) = dτ1 dτ2 ..Rijk... (t; τ1 , τ2 , ....τn )Ej (τ1 )Ek (τ2 )..... (2.121)
44

we can exchange two spatial and time components simultaneously and not
affect the polarization as follows, Exchanging τ1 and τ2 , and also exchanging
j and k gives
∫ ∫
(n) (n)
Pi (t) = dτ1 dτ2 ..Rikj... (t; τ2 , τ1 , ....τn )Ek (τ2 )Ej (τ1 )..... (2.122)

(n)
Note that Pi (t) in both equations 2.121 and 2.122 must be the same be-
cause τ1 , τ2 , j and k are dummy indicies and so can be renamed in this
war without affecting the integrals and sums. But, since Ej (τ1 )Ek (τ2 ) =
(n) (n)
Ek (τ2 )Ej (τ1 ), then Rijk... (t; τ1 , τ2 ...) = Rikj... (t; τ2 , τ1 ...). Thus, the response
function is unchanged when k ↔ j and τ1 ↔ τ2 . This is called permutation
symmetry. Exchange of any two indicies(excluding the first one) and the
corresponding two time arguements leaves the response function unchanged.
As an example, consider the 2rd order response function,

(2)
Rxyx (2)
= Rxxy ̸= Ryxx
(2)
, (2.123)
where the third permutation is not equivalent because the first index is
involved.
As we saw in section 2.2 the definitions of the fields leads to numerical
(n)
prefactors in the relationship between Pµ (ω) and the fields. In the most
general case for the notation used here, the polarization is given by
∑ ∑
Pµ(n) (ω) = K(−ω; ω1 , ω2 , ...ωn )χ(n)
µ,α1 ,α2 .... (−ω; ω1 , ω2 , ....ωn )
n α1 ,α2 ...
(2.124)
Eα1 (ω1 )Eα2 (ω2 )...Eαn (ωn )

where K(−ω; ω1 , ω2 ....ωn ) is defined in our convention to be

K(−ω; ω1 , ω2 ....ωn ) = 2l+m−n Π. (2.125)

Π is the distinct number of permutations of ω1 , ω2 ...ωn , n is the order of the
nonlinearity, m is the number of times ωi = 0, and l = 1 if ω ̸= 0 and l = 0
if ω = 0.

Example: Optical Kerr Effect

In the Optical Kerr Effect we have two photons of frequency ω incident on
a sample and two emitted photons as diagramed by Fig 2.10. This diagram
corresponds to a third order susceptibility χ(3) (−ω; ω, −ω, ω). There are
Downloaded from NLOsource.com 45

Detector

Crystal
w

Figure 2.10: In the optical Kerr effect two beams of ω enter the sample and
one causes a change in the index of refraction of the other, which we measure
on the other side.

no zero frequencies so m = 0. Since the output is nonzero, l = 1. The

order of the nonlinearity is 3 and the number of distinct permutations of the
frequencies (ω, −ω, ω) is Π = 3. Therefore

3
K = 21+0−3 (3) = , (2.126)
4

which gives

3
P (3) (ω) = χ(3) (−ω; ω, −ω, ω)E(ω)E(−ω)E(ω), (2.127)
4

the result we got in equation 2.80

46

Problem 2.3-1(a): An electric field pulse of the form E(t) =

t0 E0 δ(t) is applied to a material. δ is the Dirac Delta function,
t0 has units of time and E0 has units of electric field strength. The
polarization responds as a damped oscillator of the form P (1) (t) =
P0 e−t/τ sin ω0 t, where ω0 is the frequency of oscillation and τ the
damping time. Use this result to determine the response function
R(1) (t) and from this determine the linear susceptibility χ(1) (−ω; ω).
Compare the result you get with χ(1) (−ω; ω) for the nonlinear spring.
(b): For a response function of the form

R(2) (t1 , t2 ) = R0 θ(t1 )θ(t2 )e−t1 /τ sin (ω0 t1 ) e−t2 /τ sin (ω0 t2 ) , (2.128)

calculate χ(2) (ω1 , ω2 ) AND P (2) (t). To calculate P (2) (t), assume
that the electric field is a step function of the form E(t) = E0 θ(t).
(c): Consider a nonlinear spring with only k (1) ̸= 0 and k (2) ̸= 0. If
Γ = 0 and there is no driving field, show that

x = Ae−iω0 t + δe−i2ω0 t

is a solution of the equations of motion when δ ≪ A. Show

how δ depends on A, k (1) and k (2) . Do the results make
physical sense in light of what you got in parts (a) and (b)?

2.4 Properties of the Response Function

The nonlinear polarization P (ω) can be described in the frequency domain,
as in the case when combinations of monochromatic fields are applied to
an optical material. The optical nonlinear polarization can also be written
in the time domain as P (t). This is especially effective when considering
problems where the applied fields appear as short pulses. To get a better
understanding of the time domain polarization let’s look at the case of a
material with a linear response.
In the case of linear polarization we had,
∫ ∞
(1)
P (t) = R(1) (τ )E(t − τ )dτ. (2.129)
0

Here P (1) (t) and E(t − τ ) are observed quantities and are therefore real,
which implies that R(1) (τ ) is also real.
Downloaded from NLOsource.com 47

The linear susceptibility, χ(1) (ω), is a complex quantity with real and
imaginary parts, and is given by,
∫ ∞
(1)
χ (ω) = R(1) (τ )eiωt dτ. (2.130)
0

Here R(1) (τ ) is the linear response function that relates the polarization at
a time t to an applied electric field at some earlier time t − τ . The lower
limit of integration is set to zero rather than −∞ because we require that
P (1) (t) does not depend on the future values of E(t); only the past values of
E(t) can affect P (1) (t). This is equivalent to setting R(1) (τ ) = 0 for τ < 0
to impose the “causality condition.” From equation 2.130 and the fact that
R(1) (τ ) is real we get the following relationship

χ(1) (ωi )∗ = χ(1) (−ωi ). (2.131)

2.4.1 Kramers-Kronig
To develop the Kramers-Kronig equations we will evaluate the integral
∫ ∞ (1) ′
χ (ω ) ′
′
dω . (2.132)
−∞ ω − ω

We will treat ω as complex for the purpose of this integration, with ω →

ω + iωI . To evaluate this integral, we consider the contour integral shown
in Figure 2.11. The contribution to the contour integral from the large
semicircular path goes to zero as R ≡ |ω| → ∞ since the denominator goes
to ∞ as R → ∞.
The principal value of an integral whose path passes through a singularity
is defined as the part of the integral that excludes the singular point ω = ω ′ ,
and takes the form
(∫ ′ ∫ ∞ (1) ′ )
ω −δ (1) ′
χ (ω ) ′ χ (ω ) ′
lim dω + dω . (2.133)
δ→0 −∞ ω′ − ω ′
ω ′ +δ ω − ω

The path of integration, shown in Figure 2.11, is drawn in a way that

all poles are excluded. Thus, Cauchy’s Integral Formula yields
I ∞ (1) ′
χ (ω ) ′
′
dω = 0. (2.134)
−∞ ω − ω

This contour integral can be separated in the three paths as shown in Figure
2.11: (1) the integral along the real axis that excludes the singularity (i.e.
48

Avoid this problem point,which is

treated by using the Cauchy integral
formula

Figure 2.11: The value of the contour integral is composed of contributions

from the three major pieces of the path: (1) The semicircle of increasing
radius for which ωI > 0. (2) The real axis, which avoids the singular point
ω = ω ′ by detouring around on a semicircular arc. (3) The pole ω = ω ′ is
integrated using Cauchy’s integral formula.
Downloaded from NLOsource.com 49

the principle value integral); (2) the small semicircular path that circles the
pole; and (3) the large semicircular path at infinity. According to Equation
2.134 These three terms give,
∫ ∞ ∫
χ(1) (ω ′ ) ′ χ(1) (ω ′ ) ′
dω + iπχ(1) (ω) + dω = 0. (2.135)
−∞ ω′ − ω R ω′ − ω

The last term in Equation 2.135 vanishes as R tends to ∞. Therefore,

Equation 2.135 yeilds,
∫ ∞
(1) −i χ(1) (ω ′ ) ′
χ (ω) = dω . (2.136)
π −∞ ω′ − ω

(1) (1) (1) (1)

Since we can write χ(1) = χR +χI ; then, the fact that −i(χR +χI ) =
(1) (1)
−iχR + χI leads to the Kramers-Kronig relationships,
∫ [ ]
∞
1 χ(1) (ω ′ )
Re[χ(1) (ω)] = Im dω ′ , (2.137)
π −∞ ω′ − ω

and [ ]
∫ ∞
−1 χ(1) (ω ′ )
Im[χ(1) (ω)] = Re dω ′ . (2.138)
π −∞ ω′ − ω

The Kramers-Kronig equations relate the real part of the nonlinear sus-
ceptibility to the imaginary part provided that the imaginary part is known
over a broad enough range of angular frequencies, ω, to perform the required
integration. Similarly, the imaginary part of χ(1) depends on integral of the
real part. The imaginary part of χ is dispersive and since it is more common
to measure absorption spectra, Equation 2.137 can be used to predict the
frequency dependence of the refractive index.
The Kramers-Kronig equations can be re-expressed in a more useful
form, as follows. From equation 2.131, we find that

Re[χ(1) (−ω)] = Re[χ(1) (ω)], (2.139)

Im[χ (1)
(−ω)] = −Im[χ (1)
(ω)], (2.140)

so, we can rewrite Equation 2.138 as

∫ ∫ ∞
1 0
Re[χ(1) (ω ′ )] ′ 1 Re[χ(1) (ω ′ )] ′
Im[χ (1)
(ω)] = − dω − dω , (2.141)
π −∞ ω′ − ω π 0 ω′ − ω
50

which is equivalent to
∫ ∫
1 ∞ Re[χ(1) (ω ′ )] ′ 1 ∞ Re[χ(1) (ω ′ )] ′
(1)
Im[χ (ω)] = dω − dω . (2.142)
π 0 ω′ + ω π 0 ω′ − ω
This gives the final form
∫ ∞
2ω Re[χ(1) (ω ′ )] ′
Im[χ(1) (ω)] = − dω . (2.143)
π 0 ω ′2 − ω 2
The other complex component can be found by the following method,
∫ [ ] ∫ [ ]
1 0 χ(1) (ω ′ ) ′ 1 ∞ χ(1) (ω ′ )
(1)
Re[χ (ω)] = Im dω + Im dω ′ (2.144)
π −∞ ω′ − ω π 0 ω′ − ω
∫ [ ] ∫ [ ]
1 0 χ(1) (−ω ′ ) ′ 1 ∞ χ(1) (ω ′ )
(1)
Re[χ (ω)] = Im (−dω ) + Im dω ′
π −∞ −ω ′ − ω π 0 ω′ − ω
(2.145)
∫ [ ] ∫ [ ]
∞ ∞
1 χ(1) (ω ′ ) 1 χ(1) (ω ′ )
Re[χ(1) (ω)] = − Im ′
(dω ′ ) + Im dω ′
π 0 ω +ω π 0 ω′ − ω
(2.146)
∫ [ ]
∞
2ω χ(1) (ω ′ )
Re[χ(1) (ω)] = Im dω ′ . (2.147)
π 0 ω ′2 − ω 2

If we look at the Kramers-Kronig relations using a viewpoint of Hilbert

transform filters of the kind,
∫ ∞ ∫ ∞
−iωt
H(ω) = e h(t)dt = (h(t) cos(ωt) − ih(t) sin(ωt))dt, (2.148)
−∞ −∞

the concept of causality is captured in the statement of the Kramers-Kronig

relationships. Applying a frequency domain convolution amounts to sliding
a kernel function, in this case projecting out the even and odd parts of the
causal response of χ(1) (ω) which correspond to the real and imaginary pieces
of the frequency response. For example,

∫ ∞ ∫ ∞
χ(1) (ω−ω ′ )H(ω)dω = χ(1) (ω−ω ′ )[cos(ωt) + i sin(ωt)]dω. (2.149)
−∞ −∞

The Kramers Kroning relationships can help determine when a response

is causal or not realistic in time. We can tell whether the response is causal
by testing whether the real and imaginary parts of the Hilbert transforms
are equal to each other.
Downloaded from NLOsource.com 51

2.4.2 Permutation Symmetry

In general for different orders of nonlinearity, χ will be a tensor of many
indices (3 in the case of 2nd order nonlinear susceptibility) and will depend
on combinations of the input and output frequencies. The negatives of these
frequencies (representing outgoing fields) add additional members to the
set possible of possible frequency combinations contributing to each tensor
element. Fortunately, we do not need to keep track of all these separate
members by using symmetry properties.
From the defining relationship expressing the polarization as a function
of space and time, the time dependence of the sum-frequency polarization
for two incident monochromatic waves of frequencies ω1 and ω2 is given by

P (r, t) = Pi (ω1 + ω2 )e−i(ω1 +ω2 )t + Pi (−ω1 − ω2 )ei(ω1 +ω2 )t . (2.150)

The function P (r, t) is real since it is a physical and therefore (theoretically)

measurable quantity and by equation 2.150 we get

Pi (−ω1 − ω2 ) = Pi (ω1 + ω2 )∗ . (2.151)

The electric fields are also real, leading to

Ej (−ω1 ) = Ej (ω1 )∗ ,
Ek (−ω2 ) = Ek (ω2 )∗ . (2.152)

From Equations 2.151 and 2.152, we get

(2)
Pi (−ω1 − ω2 ) = χijk (−ω1 − ω2 , −ω1 , −ω2 )Ej (−ω1 )Ek (−ω2 ), (2.153)

and

Pi (ω1 + ω2 )∗ = χijk (ω1 + ω2 , ω1 , ω2 )∗ Ej (ω1 )∗ Ek (ω2 )∗ ,

(2)
(2.154)

with an implied sum over i and j. We can conclude that the positive and
negative components of the nonlinear susceptibility tensor have the symme-
try
χijk (−ω1 − ω2 , −ω1 , −ω2 ) = χijk (ω1 + ω2 , ω1 , ω2 )∗ ,
(2) (2)
(2.155)

which reduces the number of independent components needed to describe

the susceptibility tensor χ.
52

Full Permutation Symmetry

Some of the indices of the susceptibility tensor can be exchanged without
affecting the polarizability. For example, the quantity χijk Ej Ek will lead
to the same polarization response when the two fields are interchanged,
i.e. when χijk Ej Ek → χijk Ek Ej = χikj Ej Ek because j and k are dummy
indices and can be renamed. Thus, χijk = χikj . When the electric fields
are at different frequencies ωn and ωm , the product of the fields is of the
form Ek (ωn )Ej (ωm ), so the polarization remains unchanged when i and j
are interchanged along with ωn and ωm , thus,
(2) (2)
χijk (ωn + ωm , ωn , ωm ) = χikj (ωn + ωm , ωm , ωn ). (2.156)

Equation 2.156 is called intrinsic permutation symmetry.

For lossless media, the components of χ are all guaranteed to be real and
the interchange of any two indices (including the first one) along with the
corresponding frequencies leaves the tensor unchanged. This is called full
permutation symmetry and holds for non-dispersive materials.

Kleinman’s Symmetry
When the Bohr frequencies in a material are much larger that the frequen-
cies of all the photons in a nonlinear process, called the off-resonance regime,
all of the photon frequencies can be approximated to be small enough to set
them to zero frequency. In this case, the interchange of any two frequen-
(2)
cies leaves χijk unchanged, so equivalently, by the full intrinsic permutation
symmetry condition, any two indices can be interchanged. Therefore, the
full intrinsic permutation symmetry condition,
(2) (2) (2)
χijk (ω3 = ω1 + ω2 ) = χjki (ω1 = −ω2 + ω3 ) = χkij (ω2 = ω3 − ω1 ) =
(2) (2) (2)
χikj (ω3 = ω2 + ω1 ) = χkji (ω2 = −ω1 + ω3 ) = χjik (ω1 = ω3 − ω2 )
(2.157)

becomes
(2) (2) (2) (2) (2) (2)
χijk = χjki = χkij = χikj = χkji = χjik . (2.158)

Contracted Notation
In order to simplify the notation when discussing systems with Kleinman
Symmetry or second-harmonic generation (which is fully symmetric in its
Downloaded from NLOsource.com 53

indices since ωn and ωm are equal), we can introduce the second harmonic
tensor dijk , which is defined as

1 (2)
dijk = χijk . (2.159)
2

Given permutation symmetry, and the fact that the two incident photons are
of the same frequency, the second two indices can be interchanged without
changing the polarization. This allows us to define a 3×6 contracted matrix,
where:

• For index (j,k)=(1,1) the second index is 1

• For index (j,k)=(2,2) the second index is 2

• For index (j,k)=(3,3) the second index is 3

• For the index (j,k)=(1,2) or if (j,k)=(2,1) the second index is 4

• For index (j,k)=(1,3) or (j,k)=(3,1) the second index is 5

• for index (j,k)=(1,2) or (j,k)=(2,1) the second index is 6

This gives the contracted tensor dil , where i goes from 1 to 3, and l is
between 1 and 6. Under Kleinman symmetry, the second harmonic tensor
is given
 by, 
d11 d12 d13 d14 d15 d16
dil =  d16 d22 d23 d24 d14 d12 .
d15 d24 d33 d23 d13 d14
54

Problem 2.4-1(a): Two light beams of frequency ω1 and ω2 im-

pinge on a nonlinear material with a scalar and spatially local re-
sponse (χ(n) will not depend on ⃗k). Calculate the Fourier amplitude
(4)
P2ω1 −2ω2 in terms of χ(4) , E ω1 and E ω2 . In other words, write an
(4)
expression of the form P2ω1 −2ω2 = Kχ(4) (− (2ω1 − 2ω2 ) ; . . . ) E ω1 . . .
(b): Two high-intensity light beams of frequency ω1 and ω2 are
launched into a centrosymmetric material. A spectrometer finds that
light at frequency ω1 − 2ω2 is generated for an arbitrary set of input
frequency ω1 and ω2 only when a static electric field is applied to
the sample. Write an expression in a form as you did in (a) for the
polarization at ω1 − 2ω2 for the lowest-order nonlinearity that could
be responsible for the effect. Be sure to evaluate K. How must the
intensity of the light at frequency ω1 − 2ω2 depend on the applied
static voltage?
Problem 2.4-2(a): A physicist and an engineer set out to measure
the χ(2) and χ(3) tensor of a material using a laser that is far off-
resonance. The engineer runs to the lab and measures all 27 χ(2)
tensor components and all 81 χ(3) tensor components. The physi-
cist uses full permutation symmetry to eliminate redundant mea-
surements. How many measurements are required to determined all
tensor components? Make sure to list all of the tensor components
and how they are related.
(b): If the material is isotropic, how many measurements are re-
quired in each case? Hint: apply the inversion operation.
Problem 2.4-3: The figures below show the ground state charge
distributions of several molecules. In each case, use symmetry con-
(2) (2)
siderations to determine when either χxxx and/or χyyy are disallowed.

-q -q -2q +q
+q +q
+4q
-5q
-q -q +q +q +q +q
a) b) c)

+q y
+q +q -q
x
+q +2q
+q
-7q +q -q
+q +q -q
d) e)
Chapter 3

Nonlinear Wave Equation

3.1 General Technique

In this section we develop a general technique to obtain the nonlinear wave

equation. Subsequently, we will apply this technique to different nonlinear-
optical processes. First we need to make a series of physically-reasonable
assumptions to simplify the calculation, as follows.

We assume that there are no free charges present in space, ρf = 0;

however, there are bound charges in any material. These bound charges will
be the source of nonlinear polarization. Likewise, we assume that there are
no free or induced currents, J⃗ = 0. In fact, we do not consider any magnetic
properties of the material, which implies B ⃗ = H.
⃗ This assumption will be
valid in a broad range of materials; however, we must exercise care in cases
where this assumption fails.

Secondly, recall from electrostatics, that the electric displacement field,

⃗
D, can be written as D ⃗ =E ⃗ + 4π P⃗ . Since every material consists of atoms
or molecules, applying an electric field results in an induced polarization
in the material, which leads to generation of an internal field. To get the
wave equation inside the material, we substitute the fields into the Maxwell’s
equations. The Maxwell’s equations are,

55
56

∇·D
⃗ = 4πρf ,
∇·B
⃗ = 0, (3.1)
⃗
∇×E⃗ = − 1 ∂B ,
c ∂t
4π ⃗
1 ∂D
∇×H⃗ = J⃗ + .
c c ∂t
Using the above approximations and assumptions, the last two Maxwell’s
equations yield,
2⃗
∇×∇×E ⃗ + 1 ∂ D = 0. (3.2)
c2 ∂t2
The polarization of the material is of the form
(1)
Pi = χij Ej + PiN L . (3.3)

We substitute the polarization into the electric displacement field and

then into the Maxwell’s wave equation (Equation 3.2). The linear term of
the polarization yields the linear wave equation,
2⃗
⃗ + ϵ ∂ E = 0,
∇×∇×E (3.4)
c2 ∂t2
(1)
where ϵ = 1 + 4πχij , and the nonlinearity comes from the nonlinear terms
⃗ It
in the polarization, PiN L , which depends on E 2 and higher orders of E.
is worth mentioning that all the quantities depend on position and time,
⃗ t), p⃗(z, t), etc...
E(z,
The general strategy for solving the wave equation is

1. We assume E ⃗ =∑ E ⃗
n n , which includes incident and outgoing fields.
Each index n represent a field of one fourier component.

2. We make the fields approximately plane waves,

[ ]
⃗ 1 i(kn z−ωn t) 1 ∗ −i(kn z−ωn t)
En = An (z, ρ)e + An (z, ρ)e ρ̂. (3.5)
2 2

The plane wave propagates along z and is cylindrically symmetric, and

ρ is the only transverse coordinate. The direction of the polarization
of the field is also ρ̂, which implies that it is a transverse field, ρ̂⊥ẑ. In
Downloaded from NLOsource.com 57

λ
A(z)

E(z)

Figure 3.1: Slowly varying envelope approximation: the variation of A(z)

with respect to z is much smaller than the wavelength of the field’s oscilla-
tion.

addition, we assume that the slowly varying amplitude approximation

holds, which means the amplitude may vary but not quickly with z
and ρ. The amplitude An (z, ρ) depends on the propagation distance
z and the transverse direction ρ.
For example, in the case of second harmonic generation, the second
harmonic field can grow with z, which is included in the z-dependence
of An . Another example is self focusing, where in this case the beam
width will change as it propagates. An gives us the shape of the beam
as a function of both ρ and z.
The slowly varying condition is applied to Maxwell’s equations when
differentiating the field with respect to the position. The derivative
of An with respect to z should be much smaller than the derivative of
exp[i(kn z − ωn t)], or
∂An
≪ An kn . (3.6)
∂z
As it is shown in Figure 3.1 the electric field oscillates with position
and the envelope of the oscillation is the amplitude An , which changes
very slowly on the scale of the wavelength of the oscillation.
Similarly,
∂An
≪ An kn , (3.7)
∂ρ
where in this case the beam’s amplitude changes negligibly in the
transverse direction on the scale of the light’s wavelength. Basically,
58

when the amplitude is approximately constant over oscillations, the

plane wave approximation holds. In terms of ρ, when the width of the
wave is much larger than the wavelength, the plane wave approxima-
tion can be used.

3. Project out the fourier component of Maxwell’s wave equation and

sequentially keep terms of the same order in the electric field strength.
We select only one frequency by fourier transformation and make the
terms of the same order equal to each other.
∫ [ ]
1 ∂ ⃗
2D
dte iωn t
∇×∇×E⃗+ = 0. (3.8)
c2 ∂t2

4. Simplify the expression

∇×∇×E
⃗ = −∇2 E
⃗ + ∇(∇ · E).
⃗ (3.9)

If An is constant along the direction of polarization, ρ̂, then ∇ · E

⃗ = 0,
and we can ignore the second term of the right hand side of Equation
3.9.

3.2 Sum Frequency Generation - Non-Depletion

Regime
Now we apply the same technique that we established for sum frequency
generation in the non-depletion regime. In this regime, the sum frequency
generation process can be described as two driving waves with frequencies ω1
and ω2 , and a generated wave of frequency ω3 = ω1 +ω2 . However, the inten-
sity of the generated beam is so small that the initial beam is not depleted.
Therefore, we can assume that the incident beams are unchanged, so they
propagate as they would in a linear material. A schematic representation of
the process is shown in Figure 3.2.
The electric field amplitudes are assumed to be independent of z, and
are of the form,
[ ]
⃗ n = 1 An (z)ei(kn z−ωn t) + 1 A∗ (z)e−i(kn z−ωn t) ρ̂,
E (3.10)
2 2 n

where the amplitudes are independent of ρ, and therefore represent infinitely-

wide gaussian beams. To further simplify this case, we assume that all
Downloaded from NLOsource.com 59

ω1
ω3
ω2

Figure 3.2: Sum frequency generation: the nonlinear interaction of two

incident waves of frequencies ω1 and ω2 with a sample of with thickness l
results in the generation of a wave of frequency ω3 = ω1 + ω2 .

quantities, aside from wave-propagation along z, depend only on ρ; there-

fore, they are scalars, and the wave equation becomes,

ϵ ∂2E 4π ∂ 2 P N L
−∇2 E + = − . (3.11)
c2 ∂t2 c2 ∂t2
In the non-depletion regime, the only nonlinear polarization that leads to
sum frequency generation is,
( )ω1 +ω2
P NL = χ(2) E1ω1 E2ω2 . (3.12)

The ω1 and ω2 components of P N L vanish. The wave equation of the

fundamental beams, therefore, is linear, and the only nonlinear component
is at frequency ω3 = ω1 + ω2 . The solution to the linear wave equation
(Equation 3.4) is a plane wave with k1 = ωc1 n1 and k2 = ωc2 n2 . The zeroth
order solution for ω3 is also a plane wave of the same form, since χ(2) is small
for small P N L ,
Aj −i(kj z−ωj t)
Ej = e , j = 1, 2, 3. (3.13)
2
To get the first order correction at frequency ω3 , we substitute the sum
of the incoming and outgoing fields into the wave equation (Equation 3.11)
60

and project by Fourier trnasform the ω3 components, which yields,

ik3 ∂A3 −i(k3 z−ω3 t) k32 ϵ3 ω32
e + A3 (z)e−i(k3 z−ω3 t) − A3 e−i(k3 z−ω3 t)
2 ∂z 2 2c2
πω 2
= 23 χ(2) (−ω3 ; ω1 , ω2 )A1 e−i(k1 z−ω1 t) A2 e−i(k2 z−ω2 t) .
c
(3.14)

Since we have used the slowly varying envelope approximation, the terms
with second derivative or higher with respect to z are negligible.
To solve for the zeroth order equations, when χ(2) = 0, the right hand
side of Equation 3.14 is zero. Additionally the first term on the left hand
side is also zero because, to zeroth order, there will be no conversion of
the intensity of the ω1 + ω2 wave to other colors due to nonlinearty. The
zeroth-order terms in Equation 3.14 yield,
ϵ3 ω32 ( ) 2
(1) ω3
k32 = = 1 + 4πχ . (3.15)
c2 c2
Using the fact that n2 = ϵ, Equation 3.15 becomes,
n3 ω3
k3 = . (3.16)
c
Equation 3.16 indicates that all the zeroth-order solutions of the wave
equation propagate with ki = nicωi , where i = 1, 2, 3. As previously dis-
cussed, in the non-depletion regime there are no nonlinear effects acting on
the fields at frequencies ω1 and ω2 . Thus, the exact solution for those two
frequencies is simply a linear wave as given by Equation 3.13. The solution
of the wave equation for the ω3 wave is the superposition of the linearly-
propagating wave and the next order correction. The first-order correction
can be calculated from the two first-order correction terms, which we previ-
ously ignored in the Equation 3.14, which yields
ik3 ∂A3 −i(k3 z−ω3 t) πω32 (2)
e = 2 χ (−ω3 ; ω1 , ω2 )
2 ∂z c
−i(k1 z−ω1 t)
A1 e A2 e−i(k2 z−ω2 t) . (3.17)

The fourier transform of Equation 3.17 is obtained by multiplying it by

e−iω3 t
and integrating over time (one should be careful that k3 ̸= k1 + k2 ).
This yields
ik3 ∂A3 −ik3 z πk32 (2)
e = χ (−ω3 ; ω1 , ω2 )A1 e−ik1 z A2 e−ik2 z , (3.18)
2 ∂z ϵ3
Downloaded from NLOsource.com 61

therefore,
∂A3 2πk3 (2)
= χ (−ω3 ; ω1 , ω2 )A1 A2 e−i∆kz , (3.19)
∂z iϵ3
where ∆k = k1 + k2 − k3 is called the phase mismatch.
In the non-depletion regime the coefficients of the exponentials, A1 and
A2 are constant. Thus, it is straightforward to integrate Equation 3.19.
Using the geometry shown in Figure 3.2, Equation 3.19 yields

2πk3 χ(2) (−ω3 ; ω1 , ω2 ) [ ]

A3 = A1 A2 e−i∆kl − 1 . (3.20)
ϵ3 ∆k
Equation 3.20 represents that field amplitude of the ω3 wave at the
output plane of the material located at z = l. Light at frequency ω3 is
generated by the nonlinear interaction of light at frequencies ω1 and ω2 . But
there are lots of photons with frequencies ω1 and ω2 that do not interact
within the material and they pass right through the sample unchanged.
To simplify Equation 3.20 we write it as follows,
( ) ( )
4iπk3 χ(2) A1 A2 ∆kl −i∆kl
A3 = − sin exp . (3.21)
ϵ3 ∆k 2 2
Since we are interested in measurable quantities, it is useful to determine
the intensity from the field. The intensity is calculated using Poynting’s
theorem, which in gaussian units is of the form,
c
Ii = ni |Ai |2 . i = 1, 2, 3 (3.22)
8π
Given this definition, we can determine I3 by substituting Equation 3.21
into Equation 3.22,
( )2 ( )
c cn 4πk χ (2) A A
3 3 1 2 2 ∆kl
I3 = n3 |A3 | =
2
sin . (3.23)
8π 8π ϵ3 ∆k 2

Replacing all values of |Ai |2 with 8πIi /cni yields

( )
32π 3 ω32 ( (2) )2 I1 I2 2 2 l
I3 = 3
χ Lc sin , (3.24)
c n1 n2 n3 Lc

where Lc = 2/∆k is called the coherence length. Equation 3.24 shows that
the output intensity at the sum frequency oscillates as a function of the
sample thickness over a coherence length. The oscillation in the intensity is
62

I3(ℓ)

Figure 3.3: The intensity of generated light as a function of sample length,

l, is proportional to A1 A2 χ(2) and a small-amplitude oscillating function
in the non-depletion regime (solid red curve) and in the phase-matching
condition (the dashed green curve), it is proportional to l2 .

produced by the phase mismatch between the three waves that are traveling
at different speeds, vi = c/ni and causing interference. In order to get the
most efficient from sum frequency generation, the waves ω1 and ω2 should
travel with a phase velocity such that the wave produced at frequency ω3
at one point in the material is always in phase with the ω3 wave, which has
already been generated up stream.
There are two conditions to be considered in Equation 3.24, which are
shown in Figure 3.3. First, the intensity of the generated wave depends on
the intensity of the incident waves, and on χ(2) ; so, the higher the intensity
of the initial waves or the larger the χ(2) , the higher the intensity of the gen-
erated wave (solid red line). Secondly, phase matching in the non-depletion
limit with ∆k → 0 leads to sin2 (l/Lc ) → (l/Lc )2 ; so, I3 ∝ l2 , which is shown
as the dashed green curve in Figure 3.3. For the non-depletion limit to hold,
the amount of light being generated should be very small compared to the
intensity of the incident waves. Thus, in the case of no phase matching,
the intensity of the light that is incident on the sample should be small;
and, in the second case the sample should be thin enough to not create an
appreciable amount of second harmonic light.

3.3 Sum Frequency Generation - Small Depletion

Regime
Now we consider sum frequency generation in the small depletion regime,
relaxing the condition that the incident beams are not depleted. As shown in
Figure 3.4, two incident photons of frequencies ω1 and ω2 , and amplitudes
Downloaded from NLOsource.com 63

A1 and A2 interact to produce a photon at frequency ω3 . Downstream,

the photon at frequency ω3 interacts with another photon of frequency ω2
to produce a photon of frequency ω1 . Similarly, a photon at ω2 could be
produced. In the no depletion approximation, not enough light is generated
at frequency ω3 to to lead to back conversion; but, in the small depletion
regime, ω3 can convert back to ω1 by mixing with ω2 .
Before proceeding, it is interesting to point out that the back-conversion
process appears to violate energy conservation. This issue can be be resolved
by recognizing the actual process shown in the inset of the Figure 3.4. Here,
the photon at frequency ω2 stimulates the emission of another photon with
frequency ω2 through the annihilation of a photon of frequency ω3 . The orig-
inal photon of frequency ω2 passes through the sample without any changes.
In this process energy is conserved. So, energy conservation appears to be
violated because not all the photons are shown in the diagram. However,
using such diagrams that appear to violate energy conservation yield the
correct mixing efficiency.
We begin with by rewriting Equation 3.19 and its complex conjugate,
∂A3 2πik3 (2)
=− χ (−ω3 ; ω1 , ω2 )A1 A2 e−i(k1 +k2 −k3 )z , (3.25)
∂z ϵ3
and,
∂A∗3 2πik3 (2)
= χ (ω3 ; −ω1 , −ω2 )A∗1 A∗2 e−i(−k1 −k2 +k3 )z . (3.26)
∂z ϵ3
Here we used the fact that complex conjugate of the response function yields
a change of sign in the arguments. With this notation it is easier to under-
stand how ω1 and ω2 combine to give ω3 . The outgoing wave is always
accompanied with a minus sign in the argument, and the positive frequency
means the wave is incident on the material. This is consistent with the
processes shown in Figure 3.4.
We can equivalently write χ(2) (ω3 ; −ω1 , −ω2 ) as χ(2) (−ω1 ; ω3 , −ω2 ) to
rewrite Equation 3.26 as
∂A∗3 2πik3 (2)
= χ (−ω1 ; ω3 , −ω2 )A∗1 A∗2 e−i(−k1 −k2 +k3 )z . (3.27)
∂z ϵ3
The signs of the frequencies in the argument of χ(2) clearly match the signs
of ki in the exponential. This conjugate equation represents the process
where ω3 is being depleted. We can write this equation for A1 and A2 as
well,
∂A1 2πik1 (2)
=− χ (−ω1 ; ω3 , −ω2 )A∗2 A3 e−i(−k1 −k2 +k3 )z , (3.28)
∂z ϵ1
64

ω1 ω3

ω2

ω1
ω3

ω2

ω2

Figure 3.4: Sum frequency generation in the small depletion regime is rep-
resented by the nonlinear interaction of two incident waves with frequencies
ω1 and ω2 within the material, to generate a photon of frequency ω3 . Down-
stream, this generated photon and a photon of frequency ω2 interact and
create a photon of frequency ω1 . The inset shows how the energy is con-
served in the process of the destruction of a photon of frequency ω3 and
creation of two photons of frequencies ω1 and ω2 .

and
∂A2 2πik2 (2)
=− χ (−ω2 ; ω3 , −ω1 )A∗1 A3 e−i(−k1 −k2 +k3 )z . (3.29)
∂z ϵ2
As one may notice, the susceptibilities are the same for Equations 3.27,
3.28, and 3.29. This allows us to derive a relationship between the derivatives
of the intensities by using the derivative of amplitudes in the small depletion
regime. Differentiating Equation 3.22 with respect to z yields,
( )
∂Ii cni ∂|Ai |2 cni ∂Ai ∗ ∂A∗i
= = A + Ai . i = 1, 2, 3. (3.30)
∂z 8π ∂z 8π ∂z i ∂z

Next, we substitute Ai and ∂Ai /∂z and their complex conjugates into
Equation 3.30 for each of the intensities, using Equations 3.27, 3.28, and
3.29. Before proceeding we must impose additional assumptions. First, we
assume that all light frequencies are far from the resonance frequencies of
the material. Therefore, χ(2) is real. Secondly, we assume full permutation
Downloaded from NLOsource.com 65

symmetry, which implies that χ(2) is independent of the wavelength. Then

χ(2) is independent of frequency, allowing it to be removed as a common
factor. Finally, we assume the refractive indices are real. Taking all these
considerations into account results in,
( )
∂I1 cn1 2πik1 χ(2) ∗ ∗ i∆kz 2πik1 χ(2) ∗ −i∆kz
= − A2 A3 A1 e + A2 A3 A1 e ,(3.31)
∂z 8π ϵ1 ϵ1

∂I2 cn2 ik2 χ(2) ( )

= −A∗1 A3 A∗2 ei∆kz + A1 A∗3 A2 e−i∆kz , (3.32)
∂z 4 ϵ2
and
∂I3 cn3 ik3 χ(2) ( )
= −A1 A2 A∗3 e−i∆kz + A∗1 A∗2 A3 ei∆kz . (3.33)
∂z 4 ϵ3
In Equations 3.31, 3.32, and 3.33, the first term represents the deple-
tion of the photon with specified frequency into other frequencies; and, the
second term shows generation of that frequency. To simplify the intensity
derivatives, we can take advantage of the dispersion relations, k = nω/c and
ϵ = n2 , and rewrite Equations 3.31 - 3.33 as follow,

∂I1 iω1 χ(2) ( )

= −A∗2 A3 A∗1 ei∆kz + A2 A∗3 A1 e−i∆kz , (3.34)
∂z 4

∂I2 iω2 χ(2) ( )

= −A∗1 A3 A∗2 ei∆kz + A1 A∗3 A2 e−i∆kz , (3.35)
∂z 4
and
∂I3 iω3 χ(2) ( )
= −A1 A2 A∗3 e−i∆kz + A∗1 A∗2 A3 ei∆kz . (3.36)
∂z 4
Adding these equations together, gives

∂I1 ∂I2 ∂I3 iχ(2) [ −i∆kz

+ + = e (ω1 A2 A∗3 A1 + ω2 A1 A∗3 A2 − ω3 A1 A2 A∗3 )
∂z ∂z ∂z 4 ]
+ei∆kz (−ω1 A∗2 A3 A∗1 − ω2 A∗1 A3 A∗2 + ω3 A∗1 A∗2 A3 )
iχ(2) ( )
= ei∆kz − e−i∆kz (ω3 − ω1 − ω2 ) [A∗1 A∗2 A3 + A1 A2 A∗3 ] .
4
(3.37)
66

Using the fact that ω3 = ω1 + ω2 ,

∂I1 ∂I2 ∂I3

+ + = 0, (3.38)
∂z ∂z ∂z
which implies

I1 + I2 + I3 = constant. (3.39)

This is a simple restatement of the conservation of the energy. In the case

of sum frequency generation, the total intensities remains unchanged with a
value of the initial intensity of the two input beams, I1 (z = 0) + I2 (z = 0).
Superposition holds for the linear Maxwell’s equations, so we can express
the total electric field as a sum over all fields that are present. In a nonlinear
equation, superposition doesn’t hold; but, it approximately holds and so
the energy should be approximately conserved. However, it may not come
out exactly that way because of all the approximations used in solving the
problem. Thus we should directly check for conservation of energy. The
energy might not be a constant and it may flow to higher order terms that
we are not taking into account. The other issue is that we only considered
the energy of the fields. If we are on resonance we also have to include
the energy of the material. The energy of the fields might not be constant
separately and the energy can flow in or out of the molecules. The total
energy of the fields and the molecules therefore should be constant. Thus
we should be careful when applying these results to situations where the
approximations we made may not hold.

3.4 Aside - Physical Interpretation of the Manley-

Rowe Equation
Let’s revisit Equations 3.34 through 3.36. If we divide the derivative of each
intensity by its frequency, the result will be the same for all frequencies,
but only differs in the sign. It will be positive for the incoming waves and
negative for the outgoing waves, or,

1 ∂I1 1 ∂I2 1 ∂I3

= =− . (3.40)
ω1 ∂z ω2 ∂z ω3 ∂z

Equation 3.40 is called the Manley-Rowe expressions and they are related
to energy conservation.
Downloaded from NLOsource.com 67

S
ω3

ω1

ω2

Figure 3.5: The Manley-Rowe equation expresses the fact that the absolute
change in the number of photons at every frequency is the same in each
interaction. For example, the destruction of one photon at ω1 and one
photon at ω2 yields one photon created at frequency ω3 .

To understand this relationship more deeply, let’s consider the process

shown in Figure 3.5. We integrate the Manley-Rowe equations over a surface
that is perpendicular to the beam propagation direction, and goes to infinity
in all directions is a region where all three beams, I1 , I2 , and I3 are present,
[ ∫ ∫ ∫ ]
∂ 1 1 1
I1 ds = I2 ds = − I3 ds . (3.41)
∂z ω1 ω2 ω3

It is useful to remember the units of the intensity, [ Power

Area ]. The integration
Energy
of the intensity over area gives power, [ Time ]. On the other hand if we
divide 1/ω by h̄, h̄ω in the denominator also has dimensions of energy. Thus,
the energy passing through the surface divided by the energy per photon h̄ω
is simply the number of photons passing through the surface. Equation 3.41
thus indicates that the change in the number of photons in the first beam is
equal to the change in the number of photons in the second beam and the
change in the number of photons in the third beam with a minus sign, or

∆N1 = ∆N2 = −∆N3 . (3.42)

This is exactly the picture we have been using: two photons are destroyed
at frequencies ω1 and ω2 , and a photon is created at frequency ω3 . These
results in the small depletion regime are illustrated in Figure 3.6. I1 and
I2 have high intensity and can be depleted, and wherever these two are
minimum, I3 is peaked.
68

I (z) I
3 3

I2(z) I
2

I1(z) I1

Figure 3.6: Comparison of the intensities of the three beams of frequency

ω1 , ω2 , and ω3 . in the sum frequency generation process. I1 and I2 have
their minima when I3 peaks.

3.5 Sum Frequency Generation with Depletion of

One Input Beam
Previously, we considered the process of sum frequency generation for two
undepleted beams. We now turn to the case where one of the two beams
is allowed to be depleted. In this particular case the undepleted beam, ω1 ,
will be referred to as the pump and the depleted beam, ω2 , will be referred
to as the idler. The process is shown in figure 3.7.
There are four characteristics of the assumptions underlying the calcu-
lation of sum frequency generation with depleted idler:

1. Strong undepleted pump at frequency ω2 .

2. Weak idler at frequency ω1 which can be fully depleted.

3. The sum frequency and probe intensity are much smaller than the
pump, I1 + I3 ≪ I2 where subscripts denote the frequencies ω1 , ω2 ,
and ω3 .

4. The phase matching condition holds, therefore, ∆k = 0.

Downloaded from NLOsource.com 69

Figure 3.7: The idler, ω1 , and the pump, ω2 , undergo sum frequency genera-
tion in a nonlinear crystal and create an output with frequency ω3 = ω1 +ω2 .

These conditions are easily met experimentally. The previous sections

have already defined the amplitudes and coordinates that we use here. The
intensity of the ith beam, Ii , is related to its amplitude, A, according to,

cni
Ii = |Ai |2 . (3.43)
8π

Also, as previously shown in the derivation of Equation 3.19,

dAj 4π ωj (2)
= χ A3 A∗l ei∆kz . (3.44)
dz inj c

The processes we are considering are the cases defined by j = 1 and l = 2;

and, j = 2 and l = 1. Also, note that the phase matching condition implies
exp[i∆kz] = 1.
Because there is no depletion of the pump, Equation 3.44 implies that
∂A2 /∂z = 0. Therefore, A2 is a constant. So, considering only A1 and A3 ,
we get
4π ω1 (2) ∗
k1 = χ A2 , (3.45)
in1 c

where k1 is a constant. Likewise,

4π ω3 (2)
k3 = χ A2 , (3.46)
in3 c
70

where k3 is also a constant. Therefore, Equation 2 can be expressed as,

dA1
= k1 A3 . (3.47)
dz
dA3
= k3 A1 . (3.48)
dz
Equations 3.47 and 3.48 can now be combined to write the second spatial
derivative of A3 with respect to the ẑ direction. This gives,
( )
d 1 dA3
= k1 A3 , (3.49)
dz k3 dz
and therefore,
d2 A3
= k1 k3 A3 . (3.50)
dz 2
If we assume that we are off resonance, then χ(2) is real. Therefore,
k1 k2 = −K 2 , and so the resulting one-dimensional Helmholtz equation can
be easily solved. So, with the resulting equation,
d2 A3
= −K 2 A3 , (3.51)
dz 2
the solution is in the form of sines and cosines with the two constants to
be evaluated using the proper boundary conditions of the experiment. The
solution of A3 can then be written as

A3 (z) = B cos (Kz) + C sin (Kz) . (3.52)

The solution to A1 is found by substituting Equation 3.52 into Equation

3.48, which yields,
−BK CK
A1 (z) = sin (Kz) + cos (Kz) . (3.53)
k3 k3
Since, A2 is a constant, we express it in complex form,

A2 = A20 eiϕ2 , (3.54)

where A20 is the amplitude and ϕ2 the phase.

When z = 0, A1 (z = 0) is at the input interface of the nonlinear crystal.
The intensity of the beam at frequency ω1 is maximum at the input interface.
Therefore, this boundary condition yields,

A1 (z) = A1 (0) cos (Kz) (3.55)

Downloaded from NLOsource.com 71

I
1

I
3
Intensity

Depth

Figure 3.8: The intensities at frequencies ω1 and ω3 plotted as a function of

propagation distance.
√
To evaluate the integration constant for A3 , we recognize that kK3 = √kk3 ,
1
which using Equations 3.45 and 3.46 gives,
√
k3 n1 ω3 A2
= . (3.56)
K n3 ω1 A∗2
√
Using Equation 3.54, we recognize that A2 /A∗2 is the phase, so Equation
3.56 becomes, √
k3 n1 ω3
= −ieiϕ2 . (3.57)
K n3 ω1
Using the boundary condition that A3 (0) = 0 will lead to B = 0 in Equa-
tion 3.52. Substituting Equation 3.52 into Equation 3.51 with the help of
Equation 3.57 yields,
√
n1 ω3
A3 (z) = −ieiϕ2 A1 (0) sin (Kz) . (3.58)
n3 ω1
The intensity of each field is calculated using Equation 3.43, and is pro-
portional to the square of the amplitude of the waves as they propagate
through the nonlinear crystal. Thus the intensities are given by,
cn1
I1 = |A1 (0) |2 cos2 (Kz) . (3.59)
8π
[ ]
cn3 2 n1 ω3
I3 = |A1 (0) | 2
sin (Kz) . (3.60)
8π n3 ω1
72

The intensities as a function of z are shown in Figure 3.8.

3.6 Difference Frequency Generation

Next we consider difference frequency generation. The difference beam at
frequency ω3 = |ω1 − ω2 | is generated by a strong pump beam at frequency
ω1 and the amplified beam (described later) at frequency ω2 . The process
is shown by Figure 3.9.

Figure 3.9: The pump beam at frequency ω1 and the amplified beam at
frequency ω2 undergo difference frequency generation in a nonlinear crystal.
This creates the difference beam at frequency ω3 = ω1 − ω2 .

The simplifying assumptions used in the following derivation are as fol-

lows:

1. Strong undepleted pump at frequency ω1 .

2. The beam frequencies are far from material resonances so there is full
permutation symmetry.

3. χ(2) is independent of frequency.

4. The phase matching condition holds, therefore, ∆k = 0.

As in the sum frequency generation calculations in Sections 3.1 and 3.2,

Maxwell’s equations are reduced to,

dA2 ω 2 1 (2)
= 4π 22 χ A1 A∗3 , (3.61)
dz c ik2
Downloaded from NLOsource.com 73

since A is a constant in the undepleted pump approximation. Defining the

constant K2 as,
ω 2 1 (2)
K2 = 4π 22 χ A1 , (3.62)
c ik2
Equation 3.61 can then be rewritten in the form

dA2
= K2 A∗3 . (3.63)
dz
In parallel to Section 3.5, the spatial derivative of A3 that results from
Maxwell’s equations yields,

dA3 ω 2 1 (2)
= 4π 23 χ A1 A∗2 . (3.64)
dz c ik3

Since A is a constant, we define the constant K3 to be

ω32 1 (2)
K3 = 4π χ A1 . (3.65)
c2 ik3

Therefore, Equation 3.64 can be rewritten in the form

dA3
= K3 A∗2 . (3.66)
dz
Equations 3.63 and 3.66 yield

d2 A2 dA∗3
= K 2 , (3.67)
dz 2 dz
which can be rewritten as

d2 A2
= K2 K3∗ A2 . (3.68)
dz 2
Equation 3.68 is in the form of Equation 3.50 except that K2 K3∗ > 0.
Setting K 2 = K2 K3∗ , the solution becomes,

A2 (z) = C sinh (Kz) + D cosh (Kz) . (3.69)

Using Equation 3.66 and Equation 3.69, A3 is found to be of the form,

−K
A3 (z) = [C cosh (Kz) + D sinh (Kz)] . (3.70)
K2∗
74

I
2

Intensity
I
3

Depth

Figure 3.10: The intensities at frequencies ω2 and ω3 as a function of depth

into the material.

z = 0, corresponds to the input interface of the sample, so A3 (z = 0) = 0

and A2 (z = 0) ̸= 0. These boundary conditions imply that C = 0 and
D = A2 (0), so
A2 (z) = A2 (0) cosh (Kz) . (3.71)

By inserting Equation 3.71 into Equation 3.63 we get,

√
A1 k3 n2
A3 (z) = −i A2 (0) sinh (Kz) . (3.72)
|A1 | k2 n3

It should be noted that A1 is assumed undepleted and is considered to be

constant, which was used in the solution of A2 and A3 .
The intensities can then be written as

cn2 [ ]
I2 = |A2 (0) |2 cosh2 (Kz) , (3.73)
8π
[ ]
cn3 2 k3 n2
I3 = |A2 (0) | 2
sinh (Kz) , (3.74)
8π k2 n3

and are plotted in Figure 3.10. Note that hyperbolic functions diverge as
z → ∞, so Equations 3.73 and 3.73 must only be used in the low-intensity
regime. Note that the wave at frequency ω2 is amplified in the process that
generates the difference-frequency wave at frequency ω3 .
Downloaded from NLOsource.com 75

ω 2ω
Figure 3.11: In second harmonic generation, two incident wave of frequency
ω interact with a nonlinear medium, generating an output wave of frequency
2ω.

2ω
ω ω

Figure 3.12: This process is difference frequency generation as we have seen

in the previous section. The fundamental wave of frequency ω and the gener-
ated second harmonic wave of frequency 2ω interacts with the material and
propagates through the material resulting in an output wave of frequency
ω.

3.7 Second Harmonic Generation

We now consider second harmonic generation, a special case of sum fre-
quency generation. Second harmonic generation can be described as a pro-
cess that two incident light of the same frequency ω ≡ ω1 interact with the
material, generating an output light of frequency 2ω ≡ ω2 , as shown in Fig-
ure 3.11. Under certain conditions, the intensity of the fundamental wave
(which is the incident light) decreases and that of the second harmonic wave
increases as two waves propagate in a lossless dielectric material (which is the
case we are considering throughout our discursion), but the total intensity
remains invariant.
As we did previously in sum frequency generation, starting from the
76

Non-linear Wave Equation 3.11, with the nonlinear polarization associated

with second harmonic generation (similar to Equation 3.12),
( )2ω 1
P NL = χ(2) E ω2 , (3.75)
2
and ( )ω+2ω
P NL = χ(2) E ω E 2ω , (3.76)
where E ω and E 2ω are complex amplitudes of electric fields of the funda-
mental wave and the second harmonic wave, respectively. We obtain the
coupled equations
dA2 2π ω2 (2) 2 i∆kz
= χ A1 e , (3.77)
dz in2 c
and
dA1 4π ω1 (2)
= χ A2 A∗1 e−i∆kz , (3.78)
dz in1 c
where ∆k = 2k1 − k2 . Equation 3.77 represents the process of two fun-
damental waves interacting with a nonlinear medium and generating the
second harmonic wave, as graphically shown in Figure 3.11. Unlike the
non-depletion case, A1 is no longer a constant but depends on z, the prop-
agating distance. Once the second harmonic wave is generated, it may un-
dergo the process of difference frequency generation by interacting with the
material together with the fundamental wave, as mathematically presented
in Equation 3.78 and graphically illustrated in Figure 3.12. If we rewrite
these coupled equations in terms of intensities, the resulting solutions can
be expressed by Jacobi elliptic functions. We will not solve these coupled
equations here, but we will discuss the results under various conditions.
We first consider phase-matching condition, ∆k = 0. The behavior of
waves propagating through a material depends on initial conditions of the
coupled equations. If two waves of frequency ω and 2ω are incident on
a nonlinear medium, hence given the initial conditions that the intensity
of the fundamental wave I1 (z = 0) and that of the second harmonic wave
I2 (z = 0) both have nonzero values on the incident surface (z = 0), I1 and
I2 will oscillate while two waves propagate through the medium, but the
total intensity is conserved according to Manley-Rowe relations. In other
words, the fundamental wave and the second harmonic wave exchange energy
when traveling in the medium as illustrated in Figure 3.13. The next case,
which is the most often used one, is to generate the second harmonic wave
with only the fundamental wave incident on the medium; hence the initial
conditions are now I1 (z = 0) ̸= 0 and I2 (z = 0) = 0. The initial conditions
Downloaded from NLOsource.com 77

Intensity

Figure 3.13: Intensities oscilate while two waves of of frequencies ω and 2ω

propagate in the medium under phase-matching condition. This behavior is
due to the initial non-zero incident intensities of both waves.

imply A1 ((z = )0) ̸= 0 and A2 (z

( = 0)) = 0 in the coupled equations, therefore
we have dz dA2
̸= 0 and dzdA1
= 0. When the fundamental wave
z=0 z=0
is incident on the material, the second harmonic wave is generated and
becoming stronger as it propagates in the material. The energy gained by
the second harmonic wave comes from the fundamental wave incident on
the material, as shown in Figure 3.14.
In the case of phase-mismatching which occurs when ∆k ̸= 0 as solved
by Armstrong et al., I1 and I2 oscillate as the waves propagate through
the material. We are interested in second harmonic generation, therefore we
examine the amplitude of the second harmonic wave as it travels through the
material. With varying values of ∆k as illustrated in Figure 3.15, if ∆k = 0,
the amplitude of the second harmonic wave gradually increases; if ∆k ̸= 0,
the second harmonic wave can not be generated efficiently. Consequently
phase-mismatching reduces the efficiency of converting energy to the second
harmonic wave from the fundamental wave.
The solutions of coupled equations at the undepleted limit will give the
same results as we saw previously in sum frequency generation under the
same limit, with the incident waves of frequency ω and output wave of
frequency 2ω. Therefore we have the intensity of the second harmonic wave
acording to Equation 3.24
( )
32π 3 ω22 ( (2) )2 I12 2 2 ∆kl
I2 = χ l sinc , (3.79)
c3 n21 n2 2
78

Normalized Amplitude

Figure 3.14: The energy is converted to the second harmonic wave from the
fundamental wave. Notice that initially the slope of the fundamental wave
is zero but that of the second harmonic wave begins with a non-zero value.

Δk = 0
Second Harmonic Amplitude

Δk = 0

Δk = 0

Figure 3.15: Phase-mismatching reduces the efficiency of generating second

harmonic wave significantly.
Downloaded from NLOsource.com 79

Second Harmonic Signal (I )

2

kl / 2

Figure 3.16: The undepleted limit solution of coupled equations.

where l is the propagation length through the material. Substitute

( )
2 ∆kl
sinc = sin2 (∆kl/2) / (∆kl/2)2 (3.80)
2

into Equation 3.79, we see that I2 is proportional to sin2 (∆kl/2) as shown

in Figure 3.16.
We are interested in measuring χ2 under the undepleted approximation.
Experimentally, we measure the varying intensity of the second harmonic
wave as varying l by rotating a quartz sample with thickness d, as illustrated
in Figure 3.17. The intensity variation with respect to the refractive angle
(of the fundamental wave)ϕ can be obtained by substituting l = d/cosϕ into
Equation 3.79 as graphically shown in Figure 3.18. The reflection coeffi-
cient also depends on ϕ in such a way that the transmittance is maximum
at normal incident and decreases as |ϕ| increases. As a consequence, the
intensity of second harmonic signals will depend on the orientation of the
quartz sample as shown in Figure 3.19, which was experimentally observed
by Maker et al. in 1962.
As we saw from the results of the coupled equations, in order to obtain
a stable output of second harmonic generation, phase-matching condition

∆k = 2k1 − k2 = 0 (3.81)

is required. Substituting k1 = ωn (ω) /c and k2 = 2ωn (2ω) /c into Equation

3.81, phase-matching condition for second harmonic generation becomes

n (ω) = n (2ω) . (3.82)

80

Figure 3.17: Rotating a quartz sample changes the effective thickness of

the sample. Consequently, the intensity of the second harmonic generation
varies with respect to the orientation of the quartz sample.

I2

Figure 3.18: The intensity of second harmonic wave depends on the refrac-
tive angle ϕ, of the fundamental wave. Here we neglect the ϕ dependence of
the transmittance.
Downloaded from NLOsource.com 81

Second Harmonic Signal

Φ
0

Figure 3.19: The intensity of second harmonic generation varies with respect
to the orientation of a quartz sample.

Equation 3.82 suggests a possible way of achieving phase-matching by

considering the dispersion of the refractive index. In the region of normal
dispersion, where the refractive index monotonically increases with increas-
ing frequency (or, decreasing wavelength), as shown in Figure 3.20. In this
region, we see clearly that phase-matching condition can not be satisfied.
Nevertheless, in the region of anomalous dispersion, which is in the vicinity
of a linear absorption peak, the refractive index does not have a monotonic
relation with frequency (or wavelength) as illustrated in Figure 3.21. Thus,
it is possible to have a specific pair of frequencies ω and 2ω (or wavelengths
λ and λ/2) satisfying the phase-matching condition. However, the phase-
matched case is very limited. For instance, if we find an arbitrary frequency
ω ′ , the corresponding 2ω ′ may not have the same refractive index. Exper-
imentally, the use of anomalous dispersion is not a good way to look for
phase-matching because if we change the frequency in the neighborhood of
ω, then the second harmonic wave may no longer be phase-matched.
The best way to achieve phase-matching condition is by using birefrin-
gent materials. Birefringence is characterized by an index ellipsoid. Here our
focus would be on uniaxial birefringence characterized by the index ellipsoid
in Figure 3.22. The components of the refractive index in xy plane is inde-
pendent of the polarization in that plane nx = ny = no , and nz = ne ̸= no ,
where no and ne are called ordinary and extraordinary components of the
refractive index, respectively.
A plane wave propagating in a cross section of the index ellipsoid, for
example yz plane, is shown in Figure 3.23. Notice that choosing yz plane
is just for the purpose of convenience. In other words, we did not lose any
82

Refractive index n( )

Wavelength

Figure 3.20: Phase-matching can not be found in the normal dispersion

region.
Refractive index n( )

Wavelength

Figure 3.21: A specific phase-matched pair of wavelengths may be found

in the anomalous region, but not arbitrary wavelengths satisfy the phase-
matching condition.
Downloaded from NLOsource.com 83

z
nz ne

nx no ny no y
x

Figure 3.22: Uniaxial birefringence is characterized by an ellipsoid showing

the ordinary refractive index no and the extraordinary refractive index ne .

ne
θ
no
E
E

Figure 3.23: A plane wave travels along k̂ direction. no is the component of

the refractive index along the perpendicular polarization E⊥ , and ne is the
component along the parallel polarization E∥ .
84

generality since the refractive index does not change if we rotate the index
ellipsoid along z axis. Define the perpendicular polarization of the plane
wave, which corresponds to no , to be the component of the polarization
perpendicular to both z axis and the wave vector. The other component of
the polarization that is perpendicular to both the perpendicular polarization
and the wave vector is defined as the parallel polarization which corresponds
to ne . The refractive index will depend on the angle θ between z axis and
the wave vector. Since no does not depend on the perpendicular polariza-
tion, which implies independent of θ, the ordinary refractive index is the
component along the direction of the perpendicular polarization and can be
defined as
no (θ) = n̄o . (3.83)
The associated light along the direction of the perpendicular polarization
is called ordinary ray. The extraordinary refractive index is the component
along the direction of the parallel polarization and is defined as

1 cos2 θ sin2 θ
= + . (3.84)
n2e (θ) n̄2o n̄e2

The associated light along the direction of the parallel polarization is called
extraordinary ray. Here n̄o and n̄e are the principle values measured as
the plane wave traveling along the semi-major axis and semi-minor axis,
respectively. Evidently we have ne (0) = n̄o = no (θ) and ne (π/2) = n̄e .
As we have seen earlier in this section, the refractive index depends on
wavelength, therefore this dependence should be included in the calculations.
Equation 3.83 becomes
no (θ) = n̄o (λ) , (3.85)
and Equation 3.84 becomes

1 cos2 θ sin2 θ
= + . (3.86)
n2e (θ) n̄2o (λ) n̄2e (λ)

It should be noted that to obtain phase-matching condition, the incident

light must remain linearly polarized. In order to do so, the plane wave must
travel in such a way that the polarization is along a semi-major axis or on
a plane as shown in Figure 3.23. In contrast, elliptically polarized light can
not generate second harmonic light continuously because it doesn’t always
satisfy phase-matching condition.
If we plot the dispersion of refractive index with respect to wavelength,
since no does not depend on θ, there is only one dispersion-curve; but ne
Downloaded from NLOsource.com 85

θ = π/2

θ3
θ2

Refractive index
θ1
θ =0

ne

no

λ/2 λ
Wavelength

Figure 3.24: The dispersion of the refractive index under different orien-
tations of a positive uniaxial crystal. Second harmonic generation can be
achieved by varying the orientation of the crystal.

depends on θ so varying θ will give us a series of curves of ne . For a positive

uniaxial crystal (no < ne ), ne increases as θ increases (Figure 3.24), while
for a negative uniaxial crystal (no > ne ), ne decreases as θ increases (Figure
3.25). Experimentally, we continuously rotate the crystal until the orienta-
tion of the index ellipsoid gives the same refractive index for ordinary ray
and extraordinary ray. Moreover, using birefringent materials also allows us
to obtain different frequencies of second harmonic generation by varying the
frequency (or wavelength) of the fundamental beam. In other words, the
second harmonic generation is not limited at a specific wavelength.
An extraordinary ray of frequency 2ω can be generated from two inci-
dent ordinary rays of frequency ω, and this process is called type I phase-
(2)
matching, which corresponds to χijj with i and j referring to the extraor-
dinary and two ordinary rays, respectively. Type II phase-matching is to
generate an extraordinary ray of frequency 2ω from one ordinary ray and one
extraordinary ray both with frequency ω. Type II phase-matching is corre-
(2)
sponding to χiij , where the first i refers to the outgoing extraordinary ray
and the second i and j refer to the incident ordinary ray and extraordinary
ray, respectively.
86

θ =0
θ1
θ2
Refractive index

θ3

no

ne

λ/2 λ
Wavelength

Figure 3.25: The dispersion of the refractive index under different orienta-
tions of a negative uniaxial crystal.
Chapter 4

Quantum Theory of
Nonlinear Optics

4.1 A Hand-Waving Introduction to Quantum Field

Theory
In this section, we motivate second quantization of the electromagnetic fields
using historic arguments. Since we can calculate all the fields in free space
⃗
from the vector potential, we start this section by considering A.

4.1.1 Continuous Theory

Starting with Maxwell’s equations one can write the wave equation in terms
⃗ To solve the wave equation for A
of A. ⃗ we choose the Coulomb gauge which
is given by ∇ · A = 0. This condition holds only in free space where there
⃗
are no free charges or currents. Given the wave equation, we get plane wave
solution to A⃗ in terms of spatial and temporal coordinates given by,

∑
2 ∑
⃗ x, t) =
A(⃗ ⃗ 0r (⃗k)ei(⃗k·⃗x−ωk t) .
A (4.1)
r=1 ⃗k

The expression is summed over all the modes of different wavevectors ⃗k and
the polarizations, r, with the following transverse condition satisfied

⃗k · A
⃗ 0r (k) = 0, (4.2)

for r=1,2.

87
88

⃗ according to,
The electric and magnetic fields can be calculated from A

⃗ = ∇ × A,
B ⃗ (4.3)

and
⃗
⃗ = − 1 ∂A .
E (4.4)
c ∂t
⃗ from Equation 4.1 into Equations 4.3 and 4.4 we get
By substituting A
2 ∑[
∑ ] ∑
⃗ =
B ⃗ 0r (⃗k) ⃗kei(⃗k·⃗x−ωk t) ≡
ι⃗k × A ⃗ r (⃗k),
B (4.5)
r=1 ⃗k r,⃗k

∑
2 ∑
ιωk ∑
⃗ =
E ⃗ 0r (⃗k)ei(⃗k·⃗x−ωk t) ≡
A ⃗ r (⃗k).
E (4.6)
c
r=1 ⃗k r,⃗k

As we can see, using k = ω/c the magnitude of the fields can be written as,

⃗ r (⃗k)| = |E
|B ⃗ r (⃗k)| = k|A
⃗ 0r (⃗k)| ≡ k|A0r (⃗k)|, (4.7)

Since space is infinite, to avoid divergence problems in the plane wave

equations, we divide the space into unit cells of dimensions L3 as shown in
Figure 4.1 and apply periodic boundary conditions to the vector potential.
That is,
⃗ y, z, t) = A(x
A(x, ⃗ + L, y, z, t), (4.8)
where L is an arbitrary unit length. We apply the same condition in y and
z directions.
By applying the boundary conditions in Equation 4.1, we get

1 = eikx L , (4.9)
⇒ kx L = 2πnx , (4.10)

where nx is an integer. By also applying the boundary conditions along y

and z directions, we get

⃗k = 2π (nx x̂ + ny ŷ + nz ẑ) . (4.11)

L

So, ⃗k can only take on discrete values satisfying the boundary conditions.
Eventually we can set L → ∞,which will allow the choice of an arbitrary ⃗k.
Downloaded from NLOsource.com 89

L
L
X
L
Figure 4.1: Space is equally divided into unit cells of volume L3
90

We can calculate the energy of a single unit cell using the corresponding
Hamiltonian(from Poynting’s theorem) of the radiation in a single cube.
∫
1 ⃗ ·E
⃗ +B ⃗ · B).
⃗
Hrad = d3 ⃗x(E (4.12)
8π L
Here the integration is over one of the cells. The periodic boundary con-
ditions assure that the result is independent of which cell is selected for
integration.
Substituting the expressions for E⃗ and B ⃗ from Equations 4.5 and 4.6
into Equation 4.12 leads to a summation over different modes. The cross
terms vanish by orthonormality of the modes. For example, consider mode
⃗k, which contributes an energy
∫ L
Hmode = E02 cos2 (kx − ωt)dx (4.13)
0
∫ L [ ]
2 1 + cos(2kx − 2ωt)
= E0 dx (4.14)
0 2
[ ]
2 1 sin(2kx − 2ωt) L
= E0 + (4.15)
2 2k 0
L
= E02 . (4.16)
2
So including the x, y and z directions, and, both the E and B fields, the
Hamiltonian for a single mode can be written as,
L3 ( 2 )
Hrad (⃗k) = B0 + E02 , (4.17)
16π
where E0 and B0 are the magnitudes of E ⃗ and B,⃗ respectively. Or, using
Equation 4.7, we can write
{ 3 }
⃗ L 2 ⃗ ⃗ 2
Hrad (k) = k |A0r (k)| , (4.18)
8π
which is also equal to the energy per mode.
We can write A ⃗ 0r = ⃗εr A0r where A0r is the magnitude of the vector
potential polarized along r. To find a qualitative expression for A0r , we
consider the following. Lets assume there is one photon per unit cell of
volume L3 . Then, according to the observation that a photon has energy
Ek = h̄ωk , we assign one photon per mode, so,

[Energy/mode] = h̄ωk . (4.19)

Downloaded from NLOsource.com 91

By combining Equations 4.18 and 4.19, we get

L3 2 ⃗ ⃗ 2
h̄ωk = k |A0r (k)| (4.20)
8π
√
h̄ω8π
⇒ A0r (⃗k) = . (4.21)
L3 k 2
Since we only measure the real part of the quantities, we can consider
⃗ is given by,
only the real part of the fields or potentials. The real part of A

⃗+A
A ⃗∗
⃗ x, t)] =
Re[A(⃗ . (4.22)
2
From Equations 4.1, 4.20 and 4.22 we can have,
√
∑ 2πh̄c2 { }
⃗ x, t)] =
Re[A(⃗ ⃗
εr ar (⃗k)ei(⃗k·⃗x−ωk t) + a∗ (⃗k)e−i(⃗k·⃗x−ωk t) . (4.23)
r
L3 ω
⃗k,r

Here, ar (⃗k) and a∗r (⃗k) are dimensionless complex quantities. Eventually we
will relate them to the number of photons with wave vector ⃗k.
From the Equation 4.23 we can find E ⃗ using the Equation 4.6, which
yields
√
∑ 2πh̄ωk { }
⃗ i(⃗k·⃗
x−ωk t) ∗ ⃗ −i(⃗k·⃗
x−ωk t)
⃗ x, t) = i
E(⃗ ⃗
εr ar (k)e − ar (k)e . (4.24)
L3
⃗k,r

⃗ x, t).
We can write a similar expression for B(⃗

4.1.2 Second Quantization

Recalling the harmonic oscillator, we have,
√
2h̄mω
x̂ = (a + a† ), (4.25)
2mω
√
−i 2h̄mω
p̂ = (a − a† ), (4.26)
2 ( )
p2 1 2 2 † 1
H = + mω x = h̄ω a a + , (4.27)
2m 2 2

with [a, a† ] = 1 and [x, p] = ih̄.

92

Using the analogy with a simple harmonic oscillator(E and B behaves

like x and p), we will make the assumption that ar (⃗k) is annihilation operator
and a†r (⃗k) is the creation operator, annihilating and creating one quantum
of energy with wave vector ⃗k and polarization r.
So, the corresponding commutator is given by,
[ ]
ar (⃗k), a†r′ (k⃗′ ) = δk,k′ δr,r′ . (4.28)

With this assumption made, we calculate the expression for the Hamiltonian
as follows.(Note that the validity of the assumption is based on the success
of the resulting theory in its predictions.)
By using the orthogonality condition,
∫ L
′
eikx eik x dx = δkk′ L, (4.29)
0

one can prove that in calculating Hrad , only cross terms (⃗k, −⃗k) from the
same mode survive. From Equations 4.18 and 4.23, Hrad is given by,
∑ 1 { [ ]}
Hrad = 2πh̄ωk 2 ar (⃗k)ar (⃗k)† + ar (⃗k)† ar (⃗k) . (4.30)
8π
⃗k,r

Here |⃗εr | is unity. With the commuter [ar (⃗k), a†r (⃗k)] = 1 we get
∑ [ ]
† ⃗ ⃗ 1
Hrad = h̄ωk ar (k)ar (k) + . (4.31)
2
⃗k,r

As we know, ar (⃗k)a†r (⃗k) is the number operator Nr (⃗k). Thus

∑ ( )
⃗ 1
Hrad = h̄ωk Nr (k) + . (4.32)
2
⃗k,r

Here an interesting point to note is that the second term on the right hand
side goes to infinity independent of the number of photons as there are an
infinite number of possible modes. That is,
∑ 1
h̄ω = ∞. (4.33)
2
⃗k,r

Since we always deal with differences of energies, we ignore this term in

further calculations. The Hamiltonian is thus given by,
∑
Hrad = h̄ωk Nr (⃗k). (4.34)
r,⃗k
Downloaded from NLOsource.com 93

4.1.3 Photon-Molecule Interactions

We now consider the Hamiltonian of an electron in an atom or molecule, or

p⃗2
He = + V. (4.35)
2m
Here V includes all internal forces on the electron due to the nucleus and
other electrons. In an external electrical field, the Hamiltonian becomes
( )2
1 qA⃗
′
H = p⃗ − + V + ϕm (4.36)
2m c
( )
1 2 e2 |A| ⃗ · p⃗
⃗ 2 2eA
= p⃗ + + + V, (4.37)
2m c c

where the charge of electron is q = −e. ϕm = 0, [A, ⃗ p⃗] = 0 in the Coulomb

gauge. Here we also neglect higher order terms such as |A| ⃗ 2 . The total
Hamiltonian of the electron is thus given by,
( )
1 2eA⃗ · p⃗
H′ = p⃗2 + + V, (4.38)
2m c

or,
H ′ = He + HI (4.39)
where the interaction Hamiltonian HI is given by,

⃗ · p⃗
eA
HI = . (4.40)
mc
The Hamiltonian of the whole system including the fields, is given by,

H = He + Hrad + HI = H0 + HI , (4.41)

where eigen functions for He and Hrad are known and we treat the mixing
term HI using perturbation theory.
⃗ is given by,
The potential energy of a dipole in an electric field E

U = −⃗
µ·E
⃗ = e⃗x · E,
⃗ (4.42)

where e is the magnitude of the electron charge. Here we are ignoring higher
order moments and higher-order terms in E.
94

Substituting expression for E from 4.24,

U = e⃗x · E
⃗ (4.43)
√ [ ]
∑ 2πh̄ωk
⃗ i(⃗k·⃗
x−ωk t)
= ie ⃗
x · ⃗
εr ar (k)e + H.A. , (4.44)
L3
⃗k,r

where H.A. is the hermitian adjoint due to the operators for electron position
⃗x and annihilation of a photon, ar . One important point to note is that ⃗x
is an operator in electron space whereas ar is an operator in photon space.
For a molecule that is small compared with the lights wavelength,
⃗
ei(k·⃗x) = 1 + i(⃗k · ⃗x) + ... (4.45)

For an atom or molecules, ⃗x is the position coordinate of an electron and its

in the order of few angstroms. Then, ⃗k · ⃗x is very small for visible light. For
light of wavelength 1 × 10−4 cm and atom of 1Å radius,

⃗k · ⃗x ∼ 2π ∼ 2π × 10−4 . (4.46)
λ
Thus we assume that,
⃗
ei(k·⃗x) ∼ 1 = constant. (4.47)

Equation 4.47 is often called as dipole approximation. Some materials

may not have dipole moment µ ⃗ but have an octuple moment or, in some
cases, the size of the system may be comparable to the wavelength of the
light. In those cases, the dipole approximation is not valid.
Since He and Hrad belong to different spaces, we can write,

[He , Hrad ] = 0. (4.48)

Now, construct a wave function as a direct-product of state vectors | ψe ⟩

and | ψrad ⟩, which are eigenfunctions of He and Hrad , respectively,

| ψ⟩ =| ψe ⟩ | ψrad ⟩. (4.49)

Then,

H0 | ψe ⟩ | ψrad ⟩ = (He | ψe ⟩) | ψrad ⟩+ | ψe ⟩Hrad | ψrad ⟩ (4.50)

= (Erad + Ee ) | ψ⟩, (4.51)

where Erad , Ee are eigen values of the corresponding state vectors.

Downloaded from NLOsource.com 95

Figure 4.2: Spontaneous emission of a photon due to de-excitation

We define the state vectors as follows,

| ψ⟩ =| n; N⃗k1 ,r1 , N⃗k2 ,r2 , N⃗k3 ,r3 ...⟩, (4.52)

where n represents quantum number(s) of electrons(k,l,m etc.,) and N⃗k1 ,r1

represents the number of photons with wavevector ⃗k = ⃗k1 and polarization
r1 , and so on.
Consider a two level system with ground state(q) and excited state(l).
If we assume the electron is initially in the excited state space(Figure 4.2)
then the probability of deexcitation is proportional to the transition moment.
That is,

℘f i ∝ |⟨ψf | HI | ψi ⟩|2 (4.53)

⃗ · p⃗
eA
= κ|⟨ψf | | ψi ⟩|2 , (4.54)
mc

where κ is a proportionality constant. Here, by energy conservation, El −

Eq = h̄ω. The wave functions for the states of the system are given by,

| ψi ⟩ = | l; 0, 0, 0...⟩, (4.55)
| ψf ⟩ = | g; 0, 0, 1k= ωc ,r , 0, 0...⟩, (4.56)

where the one photon in | ψf ⟩ corresponds to the photon emitted due to

de-excitation.
96

4.1.4 ⃗
Transition amplitude in terms of E
We saw that the probability for making a transition between states due to
the absorption or emission of light is related to matrix elements of A ⃗ · p⃗.
⃗ · E.
Here we show that this is equivalent to using matrix elements of µ ⃗
By considering the dipole approximation discussed earlier, we can write
Equations 4.53 and 4.54 as follows,
e2
℘f i = κ |⟨1 ω | A⃗ | 0⟩ · ⟨g | p⃗ | l⟩|2 . (4.57)
m2 c2 k= c ,r
Now to calculate ⟨g | p⃗ | l⟩, consider the commutator
p⃗
[⃗x, He ] = ih̄ . (4.58)
m
Matrix elements of the above commutator in electron space are given by,
ih̄
⟨g | ⃗xHe − He ⃗x | l⟩ = ⟨g | p⃗ | l⟩, (4.59)
m
ih̄
(El − Eg )⟨g | ⃗x | l⟩ = ⟨g | p⃗ | l⟩, (4.60)
m
mω
⇒ ⟨g | p⃗ | l⟩ = ⟨g | ⃗x | l⟩. (4.61)
i
When we have used the fact that El − Eg = h̄ω, that is emitted photon
energy is the energy difference between the two states.
Substituting Equation 4.61 into Equation 4.57 we get,
e2 ( )
℘f i = κ 2 2 |⟨1k= ωc ,r | A⃗ | 0⟩ · mω ⟨g | ⃗x | l⟩|2 (4.62)
m c i
−ω 2
= κ 2 |⟨1k= ωc ,r | A⃗ | 0⟩ · ⟨g | e⃗x | l⟩|2 . (4.63)
c
Since there is only one photon created, we can write the vector potential as
⃗ = a† A0 ⃗εr ,
A (4.64)
so, the transition probability is given by,
−ω 2
℘f i = κA20 |⟨1k= ωc ,r | ⃗a† | 0⟩⟨g | e⃗x · ⃗εr | l⟩|2 . (4.65)
c2
ω2 2
Since A
c2 0
has same units as of E 2 , we can write

℘f i = κ|⟨g | e⃗x · E
⃗ | l⟩|2 (4.66)
= κ|⟨g | µ · E⃗ | l⟩|2 . (4.67)
Downloaded from NLOsource.com 97

Figure 4.3: Stimulated emission from deexcited system in the presence of

identical photons

⃗ in the dipole
Thus, the interaction hamiltonian can equally expressed as µ· E
approximation and Equation 4.67 is equivalent to Equation 4.54. We can
use whatever form be simpler for the particular application.

4.1.5 Stimulated Emission

Consider a two level system in its excited state. Initially there are N photons
with wave vector ⃗k and polarization direction r as shown in Figure 4.3. When
the system de-excites, a photon will be emitted, so if the de-excitation is
driven by stimulated emission, the final state consists of N + 1 photons with
the same wave vector and polarization. The probability of a transition is
proportional to the matrix element of the interaction hamiltonian. Initially,

N⃗kr = N. (4.68)

The final and initial wavefunctions are,

| ψi ⟩ = | i; N ⟩, (4.69)
| ψf ⟩ = | f ; N + 1⟩, (4.70)
98

Here we assume all other occupation numbers are 0. Then the transition
strength is given by

℘f i ∝ |⟨ψf | HI | ψi ⟩|2 (4.71)

= |⟨f | e⃗x | i⟩ · ⟨N + 1 | E
⃗ | N ⟩| 2
(4.72)
= |⟨f | µ⃗ | i⟩| |⟨N + 1 | a†r (⃗k) | N ⟩|2
2
(4.73)
(√ )2
= |⃗
µf i |2 N +1 (4.74)
= |⃗
µf i |2 (N + 1). (4.75)

Thus, if N = 0, the result is spontaneous emission. When N is large, the

probability of de-excitation increases with the number of photons already
present, which is the process of stimulated emission.

Problem 4.1-1: A physicist discovers a process in which light

at frequency ω1 and ω2 is converted by a sample to light at fre-
quency 2ω1 + ω2 only when all the light is polarized along x̂. The
intensity
( of the output peaks )at frequency ω0 and has the form
exp − [{(2ω1 + ω2 ) − ω0 } /ω0 ]2 . Express the simplest interaction
Hamiltonian HI in terms of photon creation and annihilation oper-
ators that describes this process. The constant of proportionality is
unimportant.

4.2 Time-Dependent Perturbation Theory

Our goal is to calculate the nonlinear susceptibility of a quantum system.
As is always the case, the first step requires a calculation of the induced
dipole moment as a function of the electric field, which is then differentiated
according to Equation 1.35 to get the nonlinear susceptibilities.
All of the information about a quantum system is contained in the wave-
function and any particular observable is simply the expectation value of an
operator that represents the observable. Thus, we need to calculate the
wavefunction of a molecule in an electric field, from which we calculate the
expectation value of the dipole moment, which we differentiate with respect
to the electric field to get the⟩susceptibility.
 ⃗ as the nth energy eigenstate of an atom/molecule
We therefor define ψn (E)
Downloaded from NLOsource.com 99

⃗
in the presence of electric field E(t), where

# of∑fields
⃗
E(t) = ⃗ ωi (t).
E (4.76)
i=1

At zero temperature the atom/molecule is in its ground state, so the polar-

ization is given by ⟨   ⟩
P⃗ (E)
⃗ = ψ0 (E)⃗  P⃗ ψ0 (E)
⃗ , (4.77)
 ⟩
 ⃗
where ψ0 (E) is the perturbed ground state. The susceptibility is then
given by,

∂n ⟨   ⟩ 1 
(n)  
⃗  Pi ψ0 (E)
⃗ 
χijk... (−ωσ ; ω1 , ω2 , ...) = ψ0 (E)  (4.78)
ω1 ω2
∂Ej ∂Ek ... D ⃗
′
E=0

where D′ is the frequency dependent degeneracy denominator introduced

earlier in Chapter 2.

4.2.1 Time-Dependent Perturbation Theory

Our strategy is to determine the wavefunction of the molecule in the presence
of the electric field using the zero-field wavefunctions as a basis. The field
dependence will be calculated using dipole coupling between the field and
the molecule as a time-dependent perturbation.
Let H0 be the molecular Hamiltonian without perturbation, that is, when
the electric field is not present, the time evolution of a state |ψ⟩ is given by

∂
ih̄ |ψ⟩ = H0 |ψ⟩ . (4.79)
∂t
We consider as a perturbation

V (t) = −⃗
µ · E(t),
⃗ (4.80)

which is due to the interaction energy of the electric dipole moment of the
system and the applied electric field, where
1∑⃗
⃗
E(t) = E(ωp )e−iωp t (ω−p = −ωp ), (4.81)
2 p

where µ
⃗ is the dipole moment of the atom/molecule.
100

We define the unperturbed eigenstates as

 ⟩  ⟩
 (0) 
ψn (t) = u(0)
n e−iωn t , (4.82)
 ⟩
 (0)
where ωn = En /h̄ and un are the eigenstates of the original Hamiltonian.
We can then write the total Hamiltonian, H, as the sum of the molecular
Hamiltonian and the perturbation multiplied by a small parameter λ,

H = H0 + λV. (4.83)

The perturbed states can be expressed as a sum of different orders of cor-

rection that are labeled by λm ,
∞
∑  ⟩

|ψn (t)⟩ = λm ψn(m) (t) , (4.84)
m=0

where m is the order of correction to |ψ⟩.

Substituting Equation 4.84 into the Schrodinger equation, and keeping
all terms of order m, we get
∂  ⟩ 

⟩ 

⟩
ih̄ ψn(m) (t) = H0 ψn(m) (t) + V (t) ψn(m−1) (t) . (4.85)
∂t
 ⟩
 (m)
Using the fact that the eigenvectors, ψn (t) , can be expanded in terms
 ⟩
 (0) (m)
of unperturbed states ψl (t) with coefficients aln (t),
 ⟩ ∑  ⟩
 (m) (m)  (0)
ψn (t) = aln (t) ψl (t) (4.86)
l

the ground state wavefunction in the presence of an electric field can be

expressed as  ⟩ ∑  ⟩
 (m) (m)  (0)
ψ0 (t) = al (t) ψl (t) , (4.87)
l
(m) (m)
where al (t) ≡ al0 (t). Substituting Equation 4.87 into Equation 4.85 we
get
( )
∑  ⟩ ∑  ⟩
(m)  (0) (m)  (0)
ih̄ ȧl ψl (t) + al (−iωl ) ψl (t)
l l
∑  ⟩ ∑  ⟩
(m)  (0) (m−1)  (0)
= El al ψl (t) + V (t) al ψl (t) .
l l
(4.88)
Downloaded from NLOsource.com 101
⟨ 
(0) 
Operating on Equation 4.88 from left with up  , we get

−iωp t −iωp t −iωp t

ih̄ȧ(m)
p e + h̄ωp a(m)
p e = Ep a(m)
p e
∑⟨   ⟩
  (0) (m−1) −iωl t
+ u(0)
p  V (t) ul al e . (4.89)
l

By defining ωlp and Vpl (t) as

ωlp = ωl − ωp
⟨   ⟩
  (0)
Vpl (t) = u(0)
p  V (t) ul

(4.90)
(m)
we can write ȧp as
1 ∑
(t)e−iωlp t .
(m−1)
ȧ(m)
p = Vpl (t)al (4.91)
ih̄
l

Integration of Equation 4.91 gives

∫
1 ∑ t
(t)e−iωlp t dt.
(m) (m−1)
ap (t) = Vpl (t)al (4.92)
ih̄ −∞ l

(0)
Since initially the system is in its ground state, al = δl,0 , Equation 4.92
with the help of Equation 4.80 and Equation 4.81 yields
∫
1 ∑ t 1∑⃗
(1)
ap (t) = ⃗ pl ·
µ E(ωq )e−iωq t δl,0 e−iωlp t dt. (4.93)
ih̄ −∞ 2 q
l

The integral at negative infinity vanishes if we make Ωp0 slightly complex,

so Equation 4.93 yields

1 ∑µ ⃗ p0 · E(ω
⃗ q)
a(1)
p (t) = exp [i (Ωp0 − ωq ) t] . (4.94)
2h̄ q Ωp0 − ωq

(2)
The coefficient ap is derived by substituting Equation 4.94 into Equa-
tion 4.92, and gives
[ ][ ]
1 ∑ ∑ rv µ
⃗ · ⃗
E(ω s ) µ
⃗ v0 · ⃗
E(ω q )
a(2)
r (t) = exp [i (Ωr0 − ωq − ωs ) t] ,
4h̄ q,s v (Ωr0 − ωq − ωs ) (Ωv0 − ωq )
2

(4.95)
102

(3)
and ap is calculated using Equation 4.95 in Equation 4.92,
[ ][ ][ ]
1 ∑ ∑ µ⃗ pm · ⃗ s) µ
E(ω ⃗ mn · ⃗ q) µ
E(ω ⃗ n0 · ⃗ r)
E(ω
a(3)
p (t) =
8h̄3 q,r,s m,n (Ωp0 − ωq − ωr − ωs ) (Ωm0 − ωq − ωr ) (Ωn0 − ωq )
× exp [i (Ωp0 − ωq − ωr − ωs ) t] .
(4.96)

Recall that Ωnm = Ωn − Ωm is the Bohr frequency between states n and m.

We assume that the system remains in the ground state in the pres-
ence of the field; but, because the field is time-dependent, the ground state
wavefunction will evolve with time and is of the form,
∑ ∑
−iΩp0 t −iΩp0 t
|ψ0 (t)⟩ = |0⟩+λ p (t) |p⟩ e
a(1) +λ2 p (t) |p⟩ e
a(2) +. . . , (4.97)
p p

where we have used the fact that

 ⟩  ⟩
 (0) 
ψp (t) = u(0)
p e−iΩp0 t . (4.98)

Note that since only energy differences are important, Ωp and Ωp0 are equiv-
alent.
Using Equation 4.97, the expectation value of the dipole moment will be
of the form
( )
∑
+iΩ∗p0 t
⟨⃗
µ⟩ (t) = ⟨0| + λ ap (t) ⟨p| e
(1)∗
+ ... µ⃗
p
( )
∑ ∑
−iΩp0 t −iΩp0 t
× |0⟩ + λ p (t) |p⟩ e
a(1) + λ2 p (t) |p⟩ e
a(2) + ... .
p p
(4.99)

Here, we allow Ωp0 to be complex. The meaning of this will be clarified

later.
Given Equation 4.99, the nth -order nonlinear susceptibilities will be re-
lated to dipole moment to order λn .

4.2.2 First-Order Susceptibility

Using Equation 4.99, the dipole moment to first order in λ is given by
∑ ∑
+iΩ∗p0 t −iΩp0 t
µ⟩(1) (t) =
⟨⃗ a(1)∗
p (t) ⟨p| e µ
⃗ |0⟩ + ⟨0| µ
⃗ p (t) |p⟩ e
a(1) . (4.100)
p p
Downloaded from NLOsource.com 103

We evaluate Equation 4.100 using Equation 4.94. However, the calculation

(1)∗
is made simpler if we first recast ap , as follows.
Taking the complex conjugate of Equation 4.94, we get,

1 ∑µ ⃗ 0p · E(−ω
⃗ q) [ ( ) ]
a(1)∗
p (t) = ∗ exp −i Ω∗p0 − ωq t , (4.101)
2h̄ q Ωp0 − ωq

where we have used µ ⃗ ∗p0 = µ

⃗ 0p . But, recall that q sums over positive and
negative frequencies with ω−q = −ωq . Thus, if we invert the sum by taking
q → −q, the sum remains unchanged since we are still summing over all q.
This re-indexing of Equation 4.101 leads to,

1 ∑µ ⃗ 0p · E(ω
⃗ q) [ ( ) ]
a(1)∗
p (t) = ∗ exp −i Ω∗p0 + ωq t . (4.102)
2h̄ q Ωp0 + ωq

Substituting Equations 4.102 and 4.94 into Equation 4.100 gives,

( )
1 ∑ ∑ µ
⃗ · ⃗
E(ω ) µ
⃗ · ⃗
E(ω )
e−iωq t , (4.103)
0p q p0 q
µ⟩(1) (t) =
⟨⃗ µ
⃗ p0 + µ
⃗ 0p
2h̄ p q Ω∗p0 + ωq Ωp0 − ωq

where we have used the shorthand notation µ ⃗ nm = ⟨n| µ

⃗ |m⟩. But the po-
larization is given by,
( )
∑ P⃗ (ω) ⃗ (−ω)
P
−iωt
P⃗ (t) = e + eiωt
µ⟩(1) (t),
= N ⟨⃗ (4.104)
ω
2 2

so projecting out the ω fourier component yields,

( )
N ∑ µ ⃗ 0p · ⃗ σ)
E(ω µ
⃗ p0 · ⃗ σ)
E(ω
P⃗ (ωσ ) = µ
⃗ p0 + µ
⃗ 0p . (4.105)
h̄ p Ω∗p0 + ωq Ωp0 − ωσ

The first-order susceptibility is calculated using Equation 4.78 and yields,

( j )
(1)
∑ µ ⃗ ip0
⃗ 0p µ µ ⃗ jp0
⃗ i0p µ
χij (−ωσ ; ωσ ) = −N ∗ + h̄ω + E − h̄ω , (4.106)
p
Ep0 σ p0 σ

⃗ 0p , Ep0 = h̄Ωp0 , and where we

⃗ k0p is the k th cartesian component of µ
where µ
(1)
have used the fact that the definition of χij assigns i to the direction of the
polarization P⃗ (ωσ ), and j to the direction of the applied field E(ω
⃗ σ ).
104

4.2.3 Nonlinear Susceptibilities and Permutation Symmetry

The second- and third-order (and higher order) susceptibilities can be calcu-
µ⟩(n) (t)
lated using a similar approach as the first-order one by first finding ⟨⃗
from ⟨⃗µ⟩ (t) using n -order λ terms and differentiating the result according
th

to Equation 4.78 after selecting the ωσ Fourier component. The second-order

susceptibility is then given by,

[
(2) N e3 ∑ i rj rk
r0n nm m0
χijk (−ωσ ; ω1 , ω2 ) = PI
h̄2 m,n
(Ω n0 − ω 1 − ω2 )(Ωm0 − ω2 )
]
j i k j k i
r0n rnm rm0 r0n rnm rm0
+ + ,
(Ωn0 + ω1 )(Ωm0 − ω2 ) (Ωn0 + ω1 )(Ωm0 + ω2 + ω1 )
(4.107)
where we have expressed µ ⃗ as e⃗r, and PI is the ‘intrinsic permutation oper-
ator’. It dictates that we take an average over all permutations of ω1 and ω2
with simultaneous permutations of the Cartesian components. For example,
the first term in brackets in Equation 4.107 under permutation yields
[ ]
1 i rj rk
r0n i rk rj
r0n
nm m0 nm m0
+ . (4.108)
2 (Ωn0 − ω1 − ω2 )(Ωm0 − ω2 ) (Ωn0 − ω2 − ω1 )(Ωm0 − ω1 )
Equation 4.107 can be rewritten using the full permutation operator PF .
When ω is far from resonance ωσ can be permuted with ωn with simultaneous
permutation of the Cartesian components, so the first order susceptibility
can be rewritten as,
( )
(1) N e2 ∑ i rj
r0m
χij (−ωσ ; ωσ ) = PF m0
. (4.109)
h̄ m
(Ω m0 − ωσ )
Next we consider the second-order susceptibility. After some manipu-
lations, the ground state is found to be excluded from the sum and the
matrix elements of r have subtracted from them r00 , thus the second order
susceptibility is given by,
′
( )
(2) N e3 ∑ i r̄ j r k
r0n nm m0
χijk (−ωσ ; ω1 , ω2 ) = PF , (4.110)
2h̄2 m,n
(Ω m0 − ω σ )(Ω n0 − ω 2 )

where the prime indicates that the sum excludes the ground state, and
j (2)
r̄j = rj − r00 . Equation 4.107 for χijk indeed reduces to Equation 4.110
upon the full permutation operation.
Downloaded from NLOsource.com 105

And similarly the third order susceptibility is given by,

∑ ′
(3) N e4
χijkh (−ωσ ; ω1 , ω2 , ω3 ) = PF
6h̄3 m,n,l
i r̄ j r k r h
r̄0l ln nm m0
.
(Ωl0 − ωσ )(Ωn0 − ω2 − ω3 )(Ωm0 − ω3 )
(4.111)

Problem 4.2-1: Using time dependent perturbation theory, we

found that
1 ∑ ⟨ (0)   ⟩
 (0) (m−1)
ȧp(m) (t) = up  V (t) ul al (t)e−iωlp t , (4.112)
ih̄
l

where V (t) = e⃗r · E.

⃗ Express the electric field in terms of photon
(m)
creation and annihilation operators and solve for ar (t) using the
approximation ⃗k · r = 0.

4.3 Using Feynman-Like Diagrams

4.3.1 Introduction
A Feynman diagram is a pictorial technique to evaluate the nonlinear
susceptibilities, and it is a simple way to picture the corresponding physical
process when photons are being absorbed or emitted. A Feynman diagram
can be used to evaluate a nonlinear process to various orders of perturba-
tion, and allows one to write down immediately the mathematical expression
associated with that particular process. Thus it is a convenient tool for eval-
uating complicated higher order nonlinear optical responses.
Consider the two collision processes shown in Figure 4.4.Figure 4.4 shows
the physical processes in space, and Figure 4.5 shows the same events on a
space-time diagram that can be diagrammatically evaluated.
Figure 4.6 shows a Feynman diagram for the electromagnetic interaction
between two charged particles such as electrons. Since we understand that
the vertical axis is time, and horizontal axis is space, we will omit the axes
in the diagrams:
A incident electron emits a photon for no apparent reason and then
deflects. This is a virtual process because energy is not conserved. A second
106

AB A
B
B

A A B

Figure 4.4: Particle trajectories of collision. (left): Inelastic collision; (right):

Elastic collision.

T T

AB

t
t
A
A
B B

O X1 X O X1 X

Figure 4.5: Space-time diagram for above.

Downloaded from NLOsource.com 107

E
E t h
e e

Figure 4.6: The Feynman diagram that represents the interaction between
two electrons.

incident electron absorbs the photon emitted by the first electron, and also
deflects. In this diagram, the electron moves more slowly than the photon;
the electron on the left gives off energy, so its slope decreases; the electron
on the right which absorbs energy goes the opposite way. At each vertex,
energy is not conserved, but if we consider the diagram as a whole, energy
is conserved. If we observe the process from a distance, we don’t see the
virtual photon that is emitted and absorbed by the other electron because
it violates energy conservation and can not be measured. So it is not a real
process.
This diagram also contains other physics. For example, there is a time
interval ∆t, and an energy difference ∆E associated with the virtual photon.
The product of ∆E and ∆t is approximately h̄, which is demanded by the
uncertainty principle. It either indicates that there is a big energy violation
within a very small time period or small energy violations over a long time
period.
Another example is light absorption by a molecule as show in Figure 4.7.

n Excited state

Ground state 0

Figure 4.7: Molecular absorption of light

The molecule is initially in the ground state, and absorbs a photon caus-
ing a transition to an excited state. At some later time, the photon will be
emitted and the system will return to the ground state. This process is not
shown in the diagram.
108

We will no longer be concerned with the slopes of the molecule and

photon lines in the diagrams because they will not affect the computational
results. These diagrams will later be shown to be associated with specific
terms in perturbation theory.

4.3.2 Elements of Feynman Diagram

The elements of the Feynman diagrams are:

• a line represents a molecule

• a wavy line represents a photon

• a vertex represents absorption (or emission) of a photon

• a diagonal line is a specific state

• a vertical line represents many possible states which do not necessarily

conserve energy through a vertex

At the vertex, there is a virtual process, so energy does not have to be

conserved.But energy is conserved within the whole process. These Feynman
diagrams are just symbolic; they do not represent real particle trajectories.
The vertical direction is time, and the horizontal direction is space but does
not represent physical separations.
Next let’s consider an example of linear susceptibility, draw all the pos-
sible diagrams as below:

n
n

0
0
Photon is emitted first

Figure 4.8: The two Feynman diagrams for the linear susceptibility

These two diagrams are equally likely, and they correspond to the two
(2)
terms in the linear susceptibility χij (−ωσ ; ωσ ). The fact that the photon
is emitted before a photon is absorbed, as shown in the second figure, may
seem strange because it is not a real process that is not really observed. For
Downloaded from NLOsource.com 109

each case, energy is conserved for the whole process outside the dashed red
regions.
Recall that
( j )
(2)
∑ µ ⃗ ip0
⃗ 0p µ µ ⃗ jp0
⃗ i0p µ
χij (−ωσ ; ωσ ) = −N ∗ + h̄ω + E − h̄ω , (4.113)
p
Ep0 σ p0 σ

⃗ 0p , Ep0 = h̄Ωp0 , and where we

⃗ k0p is the k th cartesian component of µ
where µ
(2)
have used the fact that the definition of χij assigns i to the direction of
the polarization P⃗ (ωσ ), and j to the direction of the applied field E(ω
⃗ σ ).
We will show in the next section how the two terms in Equation 4.113 are
related to the two diagrams shown in Figure 4.8.

4.3.3 Rules of Feynman Diagram

Question:
How does one use Feynman diagrams to evaluate nonlinear susceptibili-
ties?
We will use:

• Vertex rules

• Propagator rules

Vertex Rule: A vertex represents absorption or emission of a photon.

In the dipole approximation the vertex represents a dipole transition between
states of the system. If the initial state is |0⟩ with one incident photon,
then the dipole energy operator µ ⃗ ·E⃗ couples the field to the final state
|n⟩ of the molecule with zero photons. Recall that µ ⃗ is an operator that
⃗
acts on molecular state and E is an operator that acts on the state of the
electromagnetic field. So, the expression for the vertex (Figure 4.9) gives a

Figure 4.9: Example for vertex rule when photon is absorbed

transition amplitude ⟨n; 0| µ

⃗ ·E
⃗ |0; 1⟩.
110

Recall that E ⃗ ∝ a+a+ . So ⟨0| E⃗ |1⟩ ∝ ⟨0| a |1⟩ = 1. Then from the dipole
moment of the molecule µ ⃗ , we have ⟨n| exi |0⟩, which is simply expressed as
exin0 , where i is the field polarization direction.
Figure 4.10 shows a photon that is emitted first,

Figure 4.10: Example for vertex rule when photon is emitted

which yields the vertex contribution ⟨n; 1| µ

⃗ ·E
⃗ |0; 0⟩.
Each vertex contributes to the Feynman diagram with terms of the form
i
exn0 .
Propagator Rule:
Time-dependent perturbation theory gives integrals of the form:
∫
dte−iΩn0 t ⟨f | µ
⃗ ·E
⃗ |i⟩ . (4.114)

⃗ of light is proportional to ae−iωt + a+ eiωt , the

Since the electric field E
integrand includes a term of the form e−i(Ωn0 ±ω)t . Thus integration yields a
factor proportional to Ωn01±ω ,
∫ ∫
1
dte−iΩn0 t ⟨f | µ
⃗ ·E
⃗ |i⟩ = e−i(Ωn0 ±ω)t dt ∝ (4.115)
Ωn0 ± ω

Clearly, the product of the two vertices and propagator in Figure 4.8 yields
the two terms in Equation 4.113.
Next, we consider the second-order case shown in Figure 4.11:
The Feynman diagram in Figure 4.11 gives the propagator

1
, (4.116)
Ωm0 − ω1 − ω2

which is evaluated between the two crosses.

In summary, the Propagator Rule tells us about time evolution as
given by the time integral.
Downloaded from NLOsource.com 111

n
0

Figure 4.11: Example for propagator rule to second-order in the field.

4.3.4 Example: Using Feynman Diagrams to Evaluate Sum

Frequency Generation
In the previous sections, we derived the mathematical expressions of the
nonlinear susceptibilities using a time-consuming approach. In this section,
we will see how we can write down all those mathematical expressions im-
mediately just simply evaluating the Feynman Diagrams.
First, let’s draw all possible diagrams for sum frequency generation (Fig-
ure 4.12). If we apply full permutation symmetry, we will get a total of 6
diagrams from this case (recall that 3! = 6).
Next, let’s examine the first diagram which is drawn in more detail in
Figure 4.13:

1. |0⟩ to |n⟩:

• from the vertex rule, we have exjn0

• from the propagator rule, we have 1
h̄(Ωn0 −ω1 )

2. |n⟩ to |m⟩:

• from the vertex rule, we have exkmn

• from the propagator rule, we have h̄(Ωm0 −ω1
1 −ω2 )
, whre Ωm0 =
Ωmn + Ωn0 . Note that in the denominator the energy difference
is always from ground state to excited state.

3. |m⟩ to |0⟩:

• from the vertex rule, we have exi0m

• from the propagator rule, since the transition is from the ground
state back to the ground state, there is no energy denominator
here.
112

+ +
0 0 0
+
m m m

n n n

0 0 0
(a) (b) (c)

0 0 0
+
m m
m
+ +
n n n

0 0 0

(d) (e) (f)

Figure 4.12: Feynman diagrams for sum frequency generation

Downloaded from NLOsource.com 113

0 +
Ei

Time n
Ek
0
Ej
Figure 4.13: Sum frequency generation case a

+
n
Ei

Figure 4.14: Outgoing photon in the propagator

We multiply all those terms together, which give us the contribution to

β for that diagram

e3 xi0m xkmn xjn0

β=− (4.117)
h̄2 (Ωm0 − ω1 − ω2 )(Ωn0 − ω1 )
But Equation 4.117 is only for the first diagram. In order to get the
total β, we need to add all the other five diagrams, which gives the result:
β:

e3 xi0m xkmn xjn0

β=− +the other five diagrams. (4.118)
h̄2 (Ωm0 − ω1 − ω2 )(Ωn0 − ω1 )
It is important to mention that if the outgoing photon is in the propa-
gator, Ωm0 is changed to Ω∗m0 . For example,
In Figure 4.14 the propagator rule gives Ω∗ +ω1 1 +ω2 . And all the other
n0
propagators above the one with the first complex conjugate are of the form
Ω∗ .
114

Imaginary α
Γn0

Figure 4.15: A plot of the imaginary part of α as a function of energy is

used to demonstrate the decay width.

Problem 4.3-1: Using the Feynamn-like diagram below, calculate

(−)
its contribution to χ−−−− (−; − − . . . ), where you are to fill in the
blanks. Note that the field at frequency ω2 − ω1 is the field that is
being observed.
w2-w1
Ei
0
2w1
l Ej
m
n w2
Ep
w1
0 Ek

4.4 Broadening Mechanisms

Decay from an excited state to a lower state in the two-state system can
be approximated by making it’s Hamiltonian non-Hermitian with a small
imaginary part of the energy,
iΓn0
En0 → En0 − , (4.119)
2
where Γn0 is real and related to the inverse decay time constant through
the uncertainty principle. Γn0 can be represented as the energy width at
half (or 1/e) maximum of the imaginary part of α, as shown in Figure
Downloaded from NLOsource.com 115

4.15. The decay width can be directly measured with linear absorption
spectroscopy. This is an example of homogeneous broadening mechanisms,
where transition energy and lifetime of each molecule in the ensemble are
related to each other with the uncertainty principle.
In nature, broadening is mainly inhomogeneous since molecules are found
in different environments. As an example, consider a gas in which the veloc-
ities of the molecules obey the Boltzman distribution, i.e. they are moving
with different velocities in different directions. Figure 4.16a shows the imag-
inary part of α versus energy for three particles, one that moves away from
the observer, v < 0, one with zero velocity, v = 0, and one that is moving
toward the observer, v > 0. The dashed curve represents the velocity distri-
bution, which is used as a weighting factor in summing the individual peaks
to determine the inhomogeneously-broadened spectrum.
As another example of inhomogeneous broadening, consider two molecules
that collide and scatter so that the electric field, originating from Coulomb
interactions, grows as a function of time when they get closer and decreases
when they move away, leading to a shift in the eigenenergy due to the Stark
effect. The position of the peak shifts continuously as the molecules ap-
proach each other. Figure 4.16b shows one such shift. The superposition of
all the shifts, weighted according to the distribution of shifts from a scat-
tering calculation, yields the inhomogeneous width.
The Doppler shift leads to the same kind of behavior. In the non-
relativistic case, the shift in energy due to the Doppler effect is proportional
to v/c. This implies that particles at rest with respect to the observer,
v = 0, experience no Doppler shift. The effect of all the different processes
should be taken into account when studying a quantum system. Figure
4.16c shows the superposition over all homogeneous-broadened molecules in
the ensemble, which are weighted in amplitude to yield the dashed curve
that represents the inhomogeneous spectrum.
In general, a quantum system is never isolated because it will interact
with the environment. For example, when a molecule is in contact with a
temperature bath, the energy has a Boltzman distribution. In the quantum
picture, the molecules in vacuum are all in their ground states, but because
of the temperature bath, other states may be populated as well. When
the quantum system is not isolated and interacts with the environment, the
system is characterized by both a quantum probability and an ensemble
average weighted according to a statistical mechanical probability. In such
cases, the density matrix is a convenient tool for understanding a system’s
properties.
116

(a) v<0 v=0 v>0 (b)

Imaginary α
Imaginary α

Vacuum

E E
(c)
Imaginary α

Figure 4.16: a) The absorption spectrum of three different molecules with

different velocities (red), and the Boltzman distribution of peak positions
(dashed curve). b) Shift in the spectrum of molecules that are interacting
with each other. c) A schematic picture of the superposition of molecules in
an ensemble weighted by the Boltzman factor, leading to a inhomogeneous
broadening.
Downloaded from NLOsource.com 117

4.5 Introduction to Density Matrices

The density matrix, ρ, in Dirac notation is given by,
∑
ρ= p(s) |s(t)⟩ ⟨s(t)| , (4.120)
s

where summation is over states of the system, |s(t)⟩, which are assumed
to be normalized and p(s) is the probability of finding the system to be
in a state |s⟩. The states |s⟩ are not necessarily orthogonal and may or
may not span the full Hilbert space. The consequence of this property
will be described later. We stress that p(s) is not necessarily a quantum
probability. For example, when the system is in thermal equilibrium with a
bath of temperature T , p(s) would be proportional to e−Es /kT where Es is
the energy of state s.
Density matrices describe a quantum system that interacts with the
environment, where the interaction is taken into account via the probability
term p(s). The equilibrium condition determines which states are populated.
If |n⟩ is the eigenvector of the Hamiltonian,

H |n⟩ = En |n⟩ , (4.121)

then any given state, |s⟩, can be expanded as a superposition of energy

eigenvectors, |n⟩,
∑
|s⟩ = an |n⟩ , (4.122)
n

where an is the quantum probability of finding the particle in state n and

should not be confused with p(s). an is the quantum amplitude and p(s)
is due to the interaction of the system with the outside world. We can
calculate the quantum time evolution of state |s⟩ via,
∑
|s(t)⟩ = e−iHt/h̄ |s⟩ = an e−iEn t/h̄ |n⟩ . (4.123)
n

Density matrices can be used to calculate the expectation value of any

observable, A, by taking
⟨A⟩ = T r [ρA] . (4.124)

Equation 4.124 takes into account both thermal and quantum effects. The
118

meaning of Equation 4.124 becomes apparent by expressing it as

[ ]
∑
T r [ρA] = T r p(u) |u ⟩⟨ u| A
u
∑∑
= ⟨v |p(u)| u⟩ ⟨u |A| v⟩
u v
∑
= p(u) ⟨u |A| v⟩ ⟨v|u⟩
u,v
∑
= p(u) ⟨A⟩uu
u
= ⟨A⟩ , (4.125)

where we have used closure,

∑
|u ⟩⟨ u| = 1. (4.126)
u

Thus, the expected value of an observable is the sum over the quantum
expectation values of that observable weighted by the probabilities of finding
the system in those states.
The upper limit of the summation index in Equation 4.125 is the di-
mension of the Hilbert space, which for continuous variable is infinite. The
dimensions of Hilbert space for a discrete systems such as spin 1/2 particles
is 2: spin up, |↑⟩, and spin down, |↓⟩. The important point of the density
matrix is that the sum in Equation 4.120 extends over the states of the
system as it is initially prepared. When preparing the system, all the eigen-
states may not be included or the system may include a state that is the
superposition of many eigenstates, therefore the summation may span the
Hilbert space, or may have more or less terms than the dimension of the
Hilbert space. Examples will follow.
As a first example we consider a system of a spin half particle in a uniform
magnetic field, B = B0 ẑ, in equilibrium with a bath at temperature T . The
particle can be spin up or spin down. The energy of a particle with spin Sz
is,

H = −µ · B = −aSz B0 , (4.127)

where a is a constant and Sz is the spin component in z direction. The

Hamiltonian in matrix form is
( )
−ah̄B0 /2 0
H= , (4.128)
0 ah̄B0 /2
Downloaded from NLOsource.com 119

and the probabilities of finding finding a particle with spin ±h̄ is,

e±ah̄B0 /2kT
p± = . (4.129)
eah̄B0 /2kT + e−ah̄B0 /2kT
Inserting Equation 4.129 in Equation 4.120 yields the density matrix,

ρ = p+ |+⟩ ⟨+| + p− |−⟩ ⟨−|

( )
p+ 0
= . (4.130)
0 p−

Now, using Equations 4.128 and 4.130 in Equation 4.124, the expectation
value of the Hamiltonian can be calculated,

ah̄B0 ah̄B0 ah̄B0

⟨H⟩ = T r[Hρ] = − (p+ − p− ) = − tanh . (4.131)
2 2 2kT
When the system is at zero temperature, the population is in its ground
state, so Equation 4.131 yields, ⟨H⟩ = −ah̄B0 /2. Similarly, when the tem-
perature becomes infinite, ⟨H⟩ = 0. This makes sense because at infinite
temperature both states are equally populated so that the average of the
energies is zero.
As a second example, we describe a situation in which the number of
states exceeds the dimension of the Hilbert space. Consider,
( )
1 |+⟩ + |−⟩ 1 1
p (|+⟩) = , p √ = , p (|−⟩) = . (4.132)
4 2 4 2

So, there are 3 states in the population, which is larger than the dimen-
sionality of the Hilbert space, which is 2. The density matrix is then given
by,
( )( )
1 1 | +⟩ +| −⟩ ⟨ +| +⟨ −| 1
ρ = |+⟩ ⟨+| + √ √ + |−⟩ ⟨−|
4 4 2 2 2
( )
3/8 1/8
= . (4.133)
1/8 5/8

The off-diagonal terms in the density matrix come from the quantum su-
perposition of states. They are called coherence terms because they are
non-zero only when the wave function is composed of a superposition of
states. The trace of the density matrix given by Equation 4.133 is equal to
unity. This is a general characteristic of density matrices.
120

For pure states, represented by a single vector in Hilbert space, the

density matrix can be written as

ρ = |s(t)⟩ ⟨s(t)| . (4.134)

In this case,
ρ = ρ† (4.135)
so a pure state density matrix must be Hermitian. Additionally, since the
wavefunction is normalized, ρ2 = |s(t)⟩ ⟨s(t)| · |s(t)⟩ ⟨s(t)| = |s(t)⟩ ⟨s(t)| = ρ.
These two properties makes the density matrix for a pure state a projector.1
As an example, for a spin 1/2 particle, the most general real state vector
is
|s⟩ = cos(θ) |+⟩ + sin(θ) |−⟩ , (4.136)
whence ( )
cos2 θ sin θ cos θ
ρ= . (4.137)
sin θ cos θ sin2 θ
It is straightforward to show that this density matrix is a projector.
The time evolution of the density matrix operator can be studied in the
Heisenberg representation, which for any operator A, yields,

dA ∂A i
= − [H, A]. (4.138)
dt ∂t h̄
For density matrix operator,

dρ i
= − [H, ρ], (4.139)
dt h̄
where we assume that the density matrix does not depend explicitly on time.
The evolution of each component of ρ is given by

dρnm i i
= − ⟨n| Hρ − ρH |m⟩ = − (En − Em )ρnm , (4.140)
dt h̄ h̄
so, solving Equation 4.140, we see that the time evolution of the density
matrix is given then by
−iEnm t/h̄
ρnm = ρ(0)
nm e , (4.141)
(0)
where ρnm is the density matrix at t = 0.
1
see any textbook on quantum mechanics.
Downloaded from NLOsource.com 121

In matrix form, the expectation value of an operator A, ⟨A⟩, is calculated

from the density matrix according to
∑
⟨A⟩ = ρnm Amn . (4.142)
n,m

Using Equation 4.141, the time-dependent expectation value of the off-

diagonal components, n ̸= m, are found to oscillate at the Bohr frequency.
For diagonal terms, n = m, the expectation value is time independent.

4.5.1 Phenomenological Model of Damping

In this section, we obtain an expression for the density matrix of a molecule
that is not in equilibrium with its surroundings. When there is no interaction
with the environment, all the particles are in their ground states, possessing
the lowest energy. When particles interact with the environment, the system
may get excited. So, the equilibrium state of the system will no longer be the
ground state. The density matrix describes how the system is distributed,
i.e. how many particles are in state |a⟩, how many are in state |b⟩ and so
on. Consider a system interacting with light, where after being knocked out
of the equilibrium, it decays back to equilibrium with a rate γ. The time
evolution of the density matrix for this system can be modeled by
dρnm i
= − [H, ρ]nm − γnm (ρnm − ρeq
nm ) , (4.143)
dt h̄
where ρeqnm is the density matrix at equilibrium and the second term (corre-
sponding to ∂ρ/∂t), represents the relaxation rate when the system is out
of equilibrium. Since γ represents the decay rate, γnm is a real quantity,
which implies γnm = γmn . This means that a system that is underpopu-
lated (depleted) will return back to the equilibrium at the same rate as if it
is overpopulated. This implies a coherence between states n and m.
In general, a system can be in a superposition of multiple states. How-
ever, if we consider an ensemble, the particles must be in eigenstates and
not in a superposition of states. For example, for a population of molecules
in thermal equilibrium the energy is a parameter that determines the pop-
ulation, ρ(En ) ∝ e−βEn . The energy is in turn determined by the energy
eigenvalues. Thermal effects will wash out coherence, i.e. for off-diagonal
elements, ρeqnm = 0. The population of each state is determined by the
Boltzman factor, e−En /kT , which are just the diagonal elements of the den-
sity matrix. Based on the discussion above, one can define γnm as a rate at
which coherence between states n and m is lost.
122

The time evolution of ρnn is given by

i ∑ ∑
ρ̇nn = − [H, ρ]nn + Γnm ρmm − Γmn ρnn , (4.144)
h̄ m>n m<n

where ρnn is the population of state n. When m > n, Γmn represents the
decay rate from the higher states to the lower state n and when n > m, the
decay rate represents transitions from n to the lower states. Thus, the sum
over m includes all states from which n is populated (second term on the
right-hand side of Equation 4.144), and to all the states to which state n
can decay (the last term in Equation 4.144).
γ in Equation 4.143 is related to Γ in Equation 4.144. Consider a pure
state, |ψ(t)⟩, for which ρ = |ψ(t) ⟩⟨ ψ(t)|. Its time dependent is given by,
∑ ∑
|ψ(t)⟩ = e−iHt/h̄ an |n⟩ = e−iEn t/h̄ an |n⟩ . (4.145)
n n

To include the decay of population, we can add a small imaginary part to the
energy eigenstates given by Equation 4.119, so the density matrix becomes
∑
ρ= am a∗n e−iωmn t e−(Γm +Γn )t/h̄ |m⟩ ⟨n| , (4.146)
nm

whence
ρnn = |an |2 e−Γn t , (4.147)
which indicates that ρnn decays with rate Γn . Taking the time derivative of
Equation 4.147 yields,
ρ̇nn = −Γn ρnn . (4.148)
Recall that Γn is the decay from state n. Comparing Equation 4.148 with
Equation 4.144 leads us to the conclusion that
∑
Γn = Γmn . (4.149)
m<n

On the other hand, associating the coefficient

( of |m⟩ ⟨n|) in Equation
∑
4.146 with ρmn , both m<n Γnm ρnn and γnm ρmm − ρequil. nn constribute
ρnm . Thus,
1 col
γnm = (Γn + Γm ) + γnm , (4.150)
2
where the last term is due to collisions dephasing, where molecules collide
without a gange in population; but, the coherence between the states n and
m is lost.
Downloaded from NLOsource.com 123

To show the usefulness of the density matrix approach, consider a two-

level atom where its dipole operator, µ, has the matrix form,
( )
0 µab
µ= , (4.151)
µba 0
where a and b are the ground and excited states. Since the dipole moment
operator is Hermitian, µab = µ∗ba . The expectation value of the dipole
operator is given by
[( )( )]
ρaa ρab 0 µab
⟨µ⟩ = T r[ρµ] = T r = ρab µ∗ab + ρ∗ab µab .
ρba ρbb µ∗ab 0
(4.152)
Equation 4.152 indicates that in the case of atoms, which have no dipole
moment, coherence leads to a dipole moment: The dipole moment is induced
in an atom that initially has no dipole moment when it is in contact with a
thermal bath.
Furthermore, we can use perturbation theory to calculate the first order
polarizability using the density matrix. The procedure is the same for the
higher order nonlinearities. Given the Hamiltonian,

H = H0 + λV (t), (4.153)

where V (t) is the perturbation potential, we can use Equation 4.143 to study
the evolution of the density matrix,
i i
ρ̇nm = − (En − Em ) ρnm − [V (t), ρ]nm − γnm (ρnm − ρeq
nm ) . (4.154)
h̄ h̄
The density matrix can be expanded in a series of the order of the per-
turbation,
ρ = ρ(0) + λρ(1) + · · · + λ(n) ρ(n) + · · · . (4.155)
Keeping the zeroth order terms of Equation 4.154 after substitution of Equa-
tion 4.155 leads to
( )
nm = −iωnm ρnm − γnm ρnm − ρnm ,
ρ̇(0) (0) (0) eq
(4.156)

and to first order, λ(1) ,

i[ ]
nm = −iωnm ρnm −
ρ̇(1) − γnm ρ(1)
(1)
V (t), ρ(0) nm
h̄ [ nm
]
i ∑( )
= − (iωnm + γnm ) ρ(1)
nm − pm − ρnp Vpm
Vnp ρ(0) (0)
(, 4.157)
h̄ p
124

where we have used closure, which is given by Equation 4.126.

Because of thermal fluctuations, zeroth-order off-diagonal elements of
the density matrix are zero in equilibrium, so only the diagonal terms in the
summation in Equation 4.157 remain. Thus

i ( )
nm = − (iωnm + γnm ) ρnm −
ρ̇(1) mm − ρnn ,
(1)
Vnm ρ(0) (0)
(4.158)
h̄
where the last term is zero for n = m. When the perturbation term is zero,
V (t) = 0, by definition ρnm = ρeq
nm , and it is zero when n ̸= m. This implies
that

ρ̇(0)
nm = 0 f or n ̸= m, (4.159)

and

ρ̇(0)
nn = 0 f or n = m, (4.160)

which indicates that both diagonal and off-diagonal components are consis-
tent with the zeroth order result.
Rearranging equation 4.157, we get

i ( )
ρ̇(1)
nm + (iω nm + γnm ) ρ (1)
nm = − V nm ρ (0)
mm − ρ (0)
nn . (4.161)
h̄
Equation 4.161 can be solved using Green’s function as follows. Defining
the operator, Ôt ,
∂
Ôt = + iωnm + γnm , (4.162)
∂t
Equation 4.157 can be written in the form,

Ôt ρ(1)
nm = f (t) (4.163)

where f (t) is the right hand side of Equation 4.157. The general solution of
(1)
Equation 4.163 is the homogeneous solution of ρnm,hom , for which f (t) = 0,
plus the inhomogeneous solution, where f (t) ̸= 0,
(1) (1)
ρ(1)
nm = ρnm,hom + ρnm,inhom . (4.164)

We can easily find a solution for the homogeneous equation, yielding

(1)
ρnm,hom = Ae(−iωnm +γnm )t , (4.165)
Downloaded from NLOsource.com 125

where A is the integration constant. To solve the inhomogeneous equation,

we define a Green’s function, G(t − t′ ), so that,

Ôt G(t − t′ ) = δ(t − t′ ). (4.166)

Once we have the Green’s function, then we can find the inhomogeneous
solution for the density matrix using the following relationship,
∫ ∞
G(t − t′ )f (t′ )dt′ .
(1)
ρnm,inhom (t) = (4.167)
−∞

Clearly operating with Ôt on Equation 4.167 with the use of Equation 4.170
yields f (t),
∫ ∞
f (t′ )Ôt G(t − t′ )dt′ = f (t),
(1)
Ôt ρnm,inhom (t) = (4.168)
−∞

where we have used the fact that

∫
f (t′ )δ(t′ − t)dt′ = f (t). (4.169)

To solve a particular differential equation riquires a Green’s function

that can be constructed from a superposition of the homogeneous solutions.
For a first order differential equation, the Green’s function is simply a step
function. We construct the Green’s function by setting A = 0 for t′ < t
and A = 1 in Equation 4.165 when t′ > t. These two regions are connected
through a step function where its derivative is a delta function. Thus
{
′ 0 t′ − t < 0
G(t − t) = (4.170)
exp [−(iωnm + γnm )t] t′ − t > 0.
Equivalently,
G(t′ − t) = Θ(t′ − t)e−(iωnm +γnm )t , (4.171)
where Θ(t′ − t) is a step function, with Θ(t′ − t) = 0 for t′ − t < 0 and
Θ(t′ − t) = 1 when t′ − t ≥ 1. Using Equation 4.171, the solution for ρnm (t)
(1)

in Equation 4.161 is obtained,

[∫ ∞ ( )]
−i ′ −(iωnm +γnm )t
(1)
ρnm (t) = dt e Vnm ρmm − ρnn + Ae−(iωnm +γnm )t .
(0) (0)
h̄ t
(4.172)
The expectation value of the dipole moment with
E0 −iωp t
V = −µ · E = −µ · e + c.c., (4.173)
2
126

and the first order density matrix,

ρnm = ρ(0) (1)

nm + λρnm . (4.174)

eventually leads to a susceptibility,

N∑ µimn µjnm ( )
(1)
mm − ρnn .
ρ(0) (0)
χij (−ωp ; ωp ) = (4.175)
h̄ n,m ωnm − ωp − iγnm

Manipulating the index in Equation 4.175 gives an expression for χ(1) , which
is very similar to what we had before,
[ ]
N ∑ µ i µj µ i µj
(1) mn nm nm mn
χij (−ωp ; ωp ) = ρ(0) + .
h̄ n,m mm ωnm − ωp − iγnm ωnm + ωp + iγnm
(4.176)
If we start from the ground state, m = 0, where we get ρ00 = 1, the
expression for χ(1) in Equation 4.176 is exactly as same as the results for
vacuum where the particles are in the lowest energy state. However, in
(0)
reality, the system can start from any other state. Therefore, ρmm is a
weight factor that describes the distribution of molecules in different states.
Hence, the summation is over all possible initial states.

4.6 Symmetry
In this section we are going to use the symmetry of the Hamiltonian of a
quantum system to investigate the properties of the nonlinear response. For
a Hamiltonian given by
p2
H= + V (r), (4.177)
2m
the energy eignestates, |n⟩, are solutions of

H |n⟩ = En |n⟩ . (4.178)

When the potential is symmetric,

V (⃗r) = V (−⃗r), (4.179)

then the parity operator, π, commutes with the Hamiltonian [H, π] = 0,

therefore

π |n⟩ = ± |n⟩ . (4.180)

Downloaded from NLOsource.com 127

As a result, the wave function can be characterized as either odd |oi ⟩ or even
|ei ⟩ wave functions,

π |oi ⟩ = − |oi ⟩ ,
π |ei ⟩ = + |ei ⟩ . (4.181)

In position space,
⟨⃗x| π |oi ⟩ = ⟨−⃗x|oi ⟩ = − ⟨⃗x|oi ⟩ , (4.182)
which implies that the wave function is spatially asymmetric, ψoi (−⃗x) =
−ψoi (⃗x).
We can apply these wave functions to determine the first order suscep-
tibility,
∑ [ ]
1 1
(1)
χ (ωp ) = N e 2
|x0n |2
+ ∗ + h̄ω ) . (4.183)
n
(En0 − h̄ωp ) (En0 p

Since x is an odd function under parity, ⟨ei |x| ej ⟩ = ⟨oi |x| oj ⟩ = 0, so no elec-
tric dipole moment exists for a centrosymmetric potential. Therefore, only
transitions of the form ⟨ei |x| oi ⟩ are allowed. According to Equation 4.183
there is a resonance in the susceptibility when the photon energy matches the
transition energy, which results in a peak in the linear absorption spectrum
at frequency Ei0 /h̄.
Usually, the ground state of a system with a symmetric potential is sym-
metric, which we will call |e0 ⟩. In this case only transitions to odd-parity
excited states are allowed. These states are called one-photon states. On
the other hand the even-parity states are called two-photon states because
the transition moment from the ground state to a two photon state vanishes
and so such states do not contribute to the first order susceptibility. There-
fore, two-photon states are not observed in the linear absorption spectrum.
However, as we will see later, transitions to two-photon states are observed
in χ(3) processes.
The contribution to the second order susceptibility from states 0 and 1,
(2)
χ2L , is given by
(2)
χ2L ∝ N e2 |x01 |2 ∆x10 , (4.184)

in which case both |x01 |2 and ∆x10 must be nonzero if state 1 is to contribute.
For a centrosymmetric potential, when the first excited state is odd, |x01 |2 =
⟨e0 |x| o1 ⟩ ̸= 0. However, ∆x10 = xoo − xee = ⟨e |x| e⟩ − ⟨o |x| o⟩ = 0, and this
(2)
results in χ2L = 0.
128

In the more general case that includes 3-level terms,

∑ ∑
χ(2) ∝ N e2 |x0n |2 ∆xn0 + N e2 x0n xnm xm0 , (4.185)
n n,m

where the first sum vanishes term-by-term according to the same argument
as above. And x0n xnm xmo in the second term is always zero due to the fact
that two of the three excited states are either both even or odd, which forces
one of the matrix elements to vanish. Therefore, χ(2) is always zero for a
centrosymmetric potential.
In a two-level model, the third order susceptibility is of the form
(3)
χ2L ∝ −|x01 |2 (|x01 |2 − |∆x10 |2 ). (4.186)
(3)
Since the dipole moment is zero for a centrosymmetric potential, then χ2L ∝
|x01 |4 . As we saw above, this term is nonzero. Thus, one strategy to make
a good χ(3) material is to use a centrosymmetric material that has a large
χ(1) value.
For potentials with no symmetry, we can always express the potential as
a superposition of even- and odd-symmetry wave functions. In general we
can write the nth state as

|n⟩ = sin αn |ne ⟩ + cos αn |no ⟩ , (4.187)

where sin αn and cos αn are normalization coefficients. The susceptibilities

for assymetric potentials can be expressed in terms of the the phase factors
αn . We leave the first order susceptibility derivation as an exercise. To
write χ(2) for a two-level system, we have to calculate |x01 |2 and ∆x10 .
Using Equation 4.187, the wave functions are

|1⟩ = sin α1 |1e ⟩ + cos α1 |1o ⟩ ,

|0⟩ = sin α0 |0e ⟩ + cos α0 |0o ⟩ . (4.188)

Whence,

|x01 | = ⟨0|x|1⟩ = (sin α0 ⟨0e | + cos α0 ⟨0o |) x (sin α1 |1e ⟩ + cos α1 |1o ⟩)
= sin α0 ⟨0e |x|1o ⟩ cos α1 + cos α0 ⟨0o | x |1e ⟩ sin α1 ,
(4.189)

and

∆x10 = x11 − x00 = sin α1 cos α1 ⟨1e |x|1o ⟩ − sin α0 cos α0 ⟨0e |x|0o ⟩ .
(4.190)
Downloaded from NLOsource.com 129

The only nonzero transitions are between states with opposite parities.
The contribution to χ(2) from these two states is determined by multiplying
together Equation 4.189 and Equation 4.190. The result depends on the
phases α1 and α0 . If any of these angles are zero or π, then χ(2) becomes
zero. The optimum value of χ(2) is between these two extremes.

4.7 Sum Rules

Now we demonstrate a simple approach to understand the magnitude of
the nonlinear response using broad fundamental principles. Basically, to
optimize a nonlinear response, one has to optimize infinite sums - a difficult
if not impossible task. If a quantum system can be approximated by a
two-level model, then a deeply-colored molecule, which necessarily has large
|x01 |, will have a large linear susceptibility.
However, we can calculate an upper limit of the first order susceptibility
using the sum rules. The sum rules, which are derived directly from the
Schrodinger equation without approximation, state that
∑ h̄2
|xn0 |2 En0 = Nel , (4.191)
n
2m

where Nel is the number of electrons in the system and m is the mass of
the electron. First we simplify Equation 4.183 by assuming that the light is
in the off-resonant regime, so that we can approximate ωp ≈ 0. Assuming
En0 ∼ En0∗ , yields,

∑ 2|x0n |2
χ(1) = N e2 , (4.192)
n
En0

where N is the number density of molecules.

All terms in the sum in Equation 4.192 are positive and the energy
denominator insures that each term, on average, gets smaller with increasing
n. Multiplying each term in Equation 4.192 by En0 /En0 we get
∑ 2|x0n |2 En0
χ(1) = 2N e2 2 . (4.193)
n
En0

Using the fact that 1/En0 ≤ 1/E10 , we can replace En0 by E10 in the
denominator, leading to
∑ 2|x0n |2 En0
χ(1) ≤ 2N e2 2 . (4.194)
n
E10
130

a) b)

Figure 4.17: a) Discrete model: the material consists of atoms or molecules.

b) Continuous model: The material is assumed a uniform dielectric and a
molecule inside the material is treated as a discrete entity inside a cavity.

2 out from the sum yields

Factoring 1/E01

N e2 ∑
χ(1) ≤ 2 2 |x0n |2 En0 , (4.195)
E10 n

and applying the sum rules gives

N e2 Nel h̄2
χ(1) ≤ 2 2 . (4.196)
E10 m

Equation 4.196 gives the fundamental limit for χ(1) . A similar derivation
can be carried out for χ(2) and χ(3) . Thus, for a given number of electrons
and energy scale, there is a fundamental limit to all susceptibilities. These
limits can be used to assess molecules and aid in the design of better mate-
rials.

4.8 Local Field Model

Figure 4.17a shows a material that is made of atoms or molecules. While the
molecules are discrete, we are going to approximate the material surrounding
each molecule as being continuous, as shown in Figure 4.17b. Under this
assumption we can describe the material by a dielectric constant ϵ, and
assume that the molecule can be described as a discrete entity in a cavity.
⃗ Charges are
Figure 4.18 illustrates the material in an electric field, E.
induced on the surface of the cavity, which creates a polarization field E⃗P .
Downloaded from NLOsource.com 131

EP

Figure 4.18: An electric field applied to a material induces charges on its

surfaces. A molecule in the material, modeled as being inside a dielectric
cavity (small arrow), is affected by both the applied field and the polarization
field. The polarization field induced in the cavity is shown in the inset.

In the discrete model, however, this polarization field that acts on a dipole
is due to the neighboring dipoles.
The local field, F⃗ , is the superposition of the applied field and polariza-
tion field,

F⃗ = E
⃗ +E
⃗P . (4.197)

In experiments, is is most convenient to measure the material response to the

applied electric field rather than in terms of the local field. So, we need to
develop a model that relates these two fields.[9] The Lorentz-Lorenz model
is the simplest one, and assumes a uniform electric field, yielding

ϵ+2⃗
F⃗ = E. (4.198)
3

For a plane wave of wave length λ, Equation 4.198 can be used when the
cavity radius, a, is much smaller than the wavelength of the field (a ≪ λ),
as shown in Figure 4.19. The field is spatially uniform in the cavity but
depends on time. The local field can then be expressed as a function of
frequency

F⃗ (ω) = f (ω)E(ω),
⃗ (4.199)
132

z
E Cavity

Figure 4.19: If the cavity radius is much smaller than the plane wave’s
wavelength, the electric field will be spatially uniform inside the cavity,
which is necessary for the Lorentz-Lorenz local field model. The arrows are
the electric fields.

and

ϵ(ω) + 2
f (ω) = , (4.200)
3

where f (ω) is the local field factor, and in general it is a tensor when the
dielectric is anisotropic.
More complicated local field models consider the interaction between
molecules, such as Onsager local field model, which includes the effects of
the molecule on the dielectric. We are not going to discuss them here.
We can evaluate the local field factor using the relationship between the
dielectric function and the refractive index, ϵ = n2 . For visible light in glass,
for example, n ∼ 1.5, thus f (ω) ∼ 1.42. The local field in the glass is thus
bigger than the applied field, F⃗ (ω) > E(ω).
⃗
For an anisotropic material the generalized ith component of the local
field is,
∑
Fi (ω) = fij (ω)Ej (ω), (4.201)
j

in which f (ω) is a tensor. To apply this model to nonlinear optics, we can

express the polarization in terms of the local fields

(n) (n)
pi′ (ωσ ) = ζi′ jk··· (−ωσ ; ω1 , ω2 , · · · )Fj (ω1 )Fk (ω2 ) · · · , (4.202)
Downloaded from NLOsource.com 133

(n)
where ζi′ jk... is the nth order molecular susceptibility in vacuum. Equation
4.202 indicates that the induced dipole moment is due to the total field at
the molecule. Substituting Equation 4.201 into Equation 4.202 yields
(n) (n)
pi′ (ωσ ) = ζi′ jk··· (−ωσ ; ω1 , ω2 , · · · )fjj ′ (ω1 )fkk′ (ω2 ) · · · Ej ′ (ω1 )Ek′ (ω2 ) · · · ,
(4.203)

where summation convention holds. Multiplying both sides by fii′ (ωσ ), and
summing over i′
(n) (n)
fii′ (ωσ )pi′ (ωσ ) = ζi′ jk··· (−ωσ ; ω1 , ω2 , · · · )fii′ (ωσ )
×fjj ′ (ω1 )fkk′ (ω2 ) · · · Ej ′ (ω1 )Ek′ (ω2 ) · · · . (4.204)
(n)∗ (n)∗
We define pi′ (ωσ ) as the dressed dipole moment, and ζi′ jk··· (−ωσ ; ω1 , ω2 , · · · )
as the dressed nth order nonlinear molecular susceptibility, which are given
by,
(n)∗ (n)
pi′ (ωσ ) = fii′ (ωσ )pi′ (ωσ ),
(n)∗ (n)
ζi′ jk··· (−ωσ ; ω1 , ω2 , · · · ) = ζi′ jk··· (−ωσ ; ω1 , ω2 , · · · )fii′ (ωσ )fjj ′ (ω1 )fkk′ (ω2 ) · · · .
(4.205)

The dressed polarization can thus be expressed in terms of the dressed sus-
ceptibility and the applied fields,
(n)∗ (n)∗
pi′ (ωσ ) = ζi′ jk··· (−ωσ ; ω1 , ω2 , · · · )Ej ′ (ω1 )Ek′ (ω2 ) · · · . (4.206)

These are the quantities that are measured by experiments.

The polarization is related to the dipole moment per unit volume, which
can be calculated according to
(n) 1 ∑ (n)∗
Pi (ωσ ) = pi (ωσ ). (4.207)
V
In Equation 4.207, the sum is over the dipoles in the given volume, V, shown
in Figure 4.17a. On the other hand,
(n) (n)
Pi (ωσ ) = χijk··· (−ωσ ; ω1 , ω2 , · · · )Ej (ω1 )Ek (ω2 ) · · · . (4.208)

Substituting Equation 4.206 into Equation 4.207 and subsequently into Equa-
tion 4.208 yields
(n) N (n)∗
χijk··· (−ωσ ; ω1 , ω2 , · · · ) = ζ (−ωσ ; ω1 , ω2 , · · · ), (4.209)
V ijk···
134

where N is the number of dipoles per volume.

We have assumed that all the dipoles are aligned in the same direction
so that the sum over the dipole moments is simply given by the number of
dipoles in the volume, N times the dipole moment of one molecule. If this
assumption is not valid, then on the right hand side of Equation 4.209, we
(n)∗
must replace ζijk··· with the orientational average of the molecular suscep-
⟨ ⟩
(n)∗
tibility ζijk··· over the molecules in the unit volume, which is calculated
using Equation 4.207. Thus, Equation 4.209 becomes

(n) N ⟨ (n)∗ ⟩
χijk··· (−ωσ ; ω1 , ω2 , · · · ) = ζ (−ωσ ; ω1 , ω2 , · · · ) , (4.210)
V ijk···

It is worth mentioning that sometimes L is used to express the local field

factor tensor in the form
(n)
Lii′ jj ′ kk′ ··· (−ωσ ; ω1 , ω2 , · · · ) = fii′ (ωσ )fjj ′ (ω1 )fkk′ (ω2 ) · · · . (4.211)

To illustrate the effect of the local field on the susceptibilities, we con-

sider χ(3) , where χ(3) ∼ f 4 . The local field factor, for example, for visible
light in glass is f ≃ 1.42, leading to a correction to γ by a factor of 4. There-
fore, the local field factor results in an enhancement of the bulk nonlinear
susceptibility relative to the sum over the microscopic units. Local fields
can therefore be used to enhance the nonlinear optical response. For exam-
ple, metal nanoparticles can act as strong electric field intensifiers that can
theoretically lead to an enhancement of 8 orders of magnitude when using
silver spheres at the surface plasmon resonance.
Chapter 5

Using the OKE to Determine

Mechanisms

This chapter focuses on how the various mechanisms of the Optical Kerr
Effect in a liquid affects the third-order susceptibility tensor; and, how mea-
surements of the χ(3) tensor can be used to study the underlying mecha-
nisms. We begin with a general description of the mechanisms and end with
a detailed discussion of how molecular reorientation and electronic mecha-
nisms can be separated using polarized nonlinear refractive index measure-
ments.

5.1 Intensity Dependent Refractive Index

Assume |E(ω)|2 ≫ |E(ω ′ )|2 , where E(ω) and E(ω ′ ) are the pump and probe
field, respectively. Since we are interested in measuring the pump beam, we
focus on the polarization at frequency ω ′ ,

Figure 5.1: A pump beam of light influences the propagation of a probe

beam.

135
136

3 ( )
Pω′ = χ(3) −ω ′ ; ω, −ω, ω ′ (Eω′ Eω′ ∗ + Eω Eω ∗ ) Eω′ .
(3)
(5.1)
2
We can use a color filter or beam block that only allows the beam at ω ′
to pass through to a detector. For weak pump fields, Equation 5.1 can be
expressed to first order in Eω′ ,
3 ( )
Pω′ = χ(3) −ω ′ ; ω, −ω, ω ′ |Eω |2 Eω′ ,
(3)
(5.2)
2
which is the lowest order term that describes two-beam mixing. Note that
the polarization in Equation 5.2 is proportional to the intensity of the strong
beam.

To observe ω ′ , we project out the ω ′ fourier component from Maxwell’s

wave equation. We obtain essentially the same result as when we did using
a similar approach in Section 2.2. This leads to,

ω ′2 ω ′2 3 (3)
kω2 ′ = ε(ω ′ ) + 4π χ |Eω |2 , (5.3)
c2 c2 2
or

ω ′2
kω2 ′ ≡ εeff 2 , (5.4)
c
where εeff = ε(ω ′ ) + 6πχ(3) |Eω |2 .
Then the effective refractive index is
√
√ 6πχ(3)
neff = εeff = n(ω ′ ) 1 + 2 ′ |Eω |2 , (5.5)
n (ω )
where the refractive index is evaluated at ω ′ . For a small nonlinear phase
shift χ(3) |Eω | ≪ 1, the square root in Equation 5.5 can be expanded in a
series,

3πχ(3)
neff ≈ n(ω ′ ) + |Eω |2 . (5.6)
n(ω ′ )
This is similar to the one-beam result given by Equation 2.89. Recalling
that
c
Iω = n(ω) |Eω |2 , (5.7)
8π
and substituting Equation 5.7 into Equation 5.6, the effective intensity-
dependent refractive index becomes
Downloaded from NLOsource.com 137

Figure 5.2: Electronic mechanism of χ(3) . A molecule with a nuclear frame-

work represented by a blue arrow, and electrons represented by the sur-
rounding green dots. The electron cloud is deformed by the electric field, as
shown above. Note that the degree of deformation is greatly exaggerated.

24π 2 χ(3) (−ω ′ ; ω, −ω, ω ′ )

neff ≈ n(ω ′ ) + Iω . (5.8)
cn(ω)n(ω ′ )
The most general form of the intensity-dependent refractive index can be
written as,
n = n0 + n2 I + n4 I 2 + ..., (5.9)
where n4 ∝ χ(5) , and nl ∝ χ(l+1) .

5.1.1 Mechanism of χ(3)

Electronic
Figure 5.2 shows a molecule and its electron cloud at an arbitrary initial ori-
entation. As an electric field is applied, the shape of the molecule changes
as the electrons move in response to the electric field. The refractive indices
then change as the electron cloud changes. In practice the change in refrac-
tive index is measured by using the strong electric field of a very powerful
laser pulse, and then probing the refractive index change with a weak laser.

Molecular Reorientation
A set of randomly oriented molecules is shown in Figure 5.3. The applied
electric field is in the optical range, so oscillates too quickly for the molecules
138

Figure 5.3: Reorientational mechanism of χ(3) . The nuclei of each molecule

are shown as arrows.

Intensity

Force

Force

Figure 5.4: Electrostriction mechanism of χ(3) . The effective width of the

light beam is shown by the dotted lines. The green dots represent molecules
in solution.

to keep up. The molecules will align axially along the polarization axis of
the electric field as shown in Figure 5.3.

Electrostriction(In liquid solution or gas phase)

A Gaussian beam, as shown in Figure 5.4, with full width half maximum
shown by dashed lines, travels through a solution of molecules and creates
an inward force on the molecules, which then move into the region of the
beam. More molecules in a region results in a density change and thus a
change in refractive index. In a solid material the force is on the material
itself. For instance, rhythmic mechanical contraction and relaxation can be
Downloaded from NLOsource.com 139

Figure 5.5: Saturated absorbtion mechanism of χ(3) .

Material

Laser
Hot

Liquid

Figure 5.6: Thermal mechanism of χ(3) .

created by successively pulsing light into the material such that acoustical
waves are produced.

Saturated Absorbtion
In the electronic mechanism there is deformation of the electron cloud but
there are no real excitations to excited states because the energy of the
photon is much smaller than the transition energy in the molecule. However
in saturated absorbtion, as shown in Figure 5.5, a molecule originally in an
initial state is excited to a state of higher energy when the photon energy
matches the transition energy, thus changing the refractive index.

Thermal Effect
When a laser impinges on a material as shown in Figure 5.6, it heats the ma-
terial even in the off-resonance regime. In the region of the laser beam, this
leads to a density change and a resulting refractive index change. Generally,
these thermal effects have a negative refractive index.
An interesting example of a device that converts a continuous light beam
to a pulsed source is a liquid placed in a thin prism cell. The prism bends the
light, but through the thermal effect the liquid gets hot and the refractive
index decreases. As the beam deflects due to heating, it moves away from
the heated region, allowing the liquid to cool. Subsequently, the cooled
140
( )
cm3
Mechanism χ(3) erg Response time (sec)
Electronic 10−14 10−15
Reorientation 10−12 10−12
Electrostriction 10−12 10−9
Saturated Absorbtion 10−8 10−8
Thermal 10−4 10−3

Table 5.1: χ(3) and the response times of several mechanisms.

liquid is back to its original state and the effect repeats. This leads to
time-dependent periodic deflection of a continuous laser beam.
In Table 5.1, the strength of χ(3) and the response time for each of the
mechanisms are shown. There is a large change in the magnitude of χ(3)
from the electronic mechanism to the thermal mechanism; but, the ratio of
the strength to the response time remains approximately constant. A de-
vice using the aforementioned mechanisms ideally has a large nonlinearity
and also operates quickly. However, according to Table 5.1, if the device
is fast χ(3) is small so the device is less efficient and therefore requires a
higher light intensity to operate. Thus a device operating under the differ-
ent mechanisms, over any particular time scale, is essentially equivalently
effective.

(3)
5.2 Tensor Nature of χijkl
(3)
The third-order susceptibility tensor, χijkl , is described by 81 separate ele-
ments. Without symmetry, many of these components can be independent
and nonzero. However, when a material has a high degree of symmetry, the
number of tensor components is significantly reduced. Liquids have centro-
symmetric potentials and have vanishing χ(2) so are thus ideal for studying
third order susceptibility processes.
To understand the centro-symmetric behavior of liquids, it is convenient
to define the inversion operator, I, ˆ which changes the signs of all vector
ˆ
components. We also define Ii , such that any vector becomes its inverse
along only the ith component. In a liquid,

ˆ (3) = χ(3)
Iχ and Iˆi χ(3) = χ(3) , (5.10)
since if we rotate the liquid about any arbitrary axis it would look the same
everywhere. The liquid is then said to be isotropic, i.e. the same in all
Downloaded from NLOsource.com 141

directions; and homogeneous, i.e. the same in all places. For our purposes
the isotropic behavior is particularly useful.

Consider
(3) (3)
P1 = χ1123 E1 E2 E3 . (5.11)
Applying Iˆ2 to Equation 5.11 yields,
[ ] [ ]
(3) (3)
Iˆ2 P1 = Iˆ2 χ1123 E1 E2 E3 , (5.12)
or
(3) (3) (3)
P1 = χ1123 E1 (−E2 )E3 = −χ1123 E1 E2 E3 . (5.13)
(3) (3)
Equations 5.11 and 5.13 reveal that χ1123 = −χ1123 , establishing that
(3) (3) (3)
χ1123 must be zero. By applying similar operations, χ1122 , χ1111 , and other
susceptibilities with even numbers of repeated components are nonzero. In
total, we have 21 nonzero components of χ(3) out of the original 81.

Rotating the system 900 about any particular axis does not change the
physical properties of the material, thus

χ1111 = χ2222 = χ3333 . (5.14)

(3)
Rotating any other component, say χ1212 , 900 about any axis, should leave
the material unchanged, so

χ1212 = χ1313 . (5.15)

(3) (3)
Following this same logic, all χiijj are the same for i ̸= j, all χijji are
(3)
the same for i ̸= j, and all χijij are the same for i ̸= j. Thus,

χ1111 = χ2222 = χ3333 , (5.16)

χ1122 = χ1133 = χ2211 = χ2233 = χ3311 = χ3322 , (5.17)
χ1212 = χ1313 = χ2323 = χ2121 = χ3131 = χ3232 , (5.18)
χ1221 = χ1331 = χ2112 = χ2332 = χ3113 = χ3223 . (5.19)

From the 21 nonzero terms, only four are independent components, three
of which are off diagonal. We can write all the off-diagonal components in
terms of 1’s and 2’s,
142

(3) (3) (3) (3)

χijkl = χ1122 δij δkl + χ1212 δik δjl + χ1221 δil δjk , (5.20)
since any other component can be determined from Equations 5.16-5.19.
(3)
Applying the Euler’s Rotation Matrix to χ1111 by 450 should leave the
liquid unchanged, because it is isotropic, and the results of the measurement
should not change, thus
(3) (3) (3) (3)
χ1111 = χ1122 + χ1221 + χ1212 . (5.21)
Equation 5.20 was derived under the assumption i ̸= j, but, by virtue of
Equation 5.21, when i = j, Equation 5.20 still produces the same sum.
The third order susceptibility tensor thus has only three independent tensor
components in an isotropic substance.

Now we use permutation symmetry to reduce our independent compo-

nents even further. In general the intensity dependent refractive index is of
the form
(3)
χijkl (−ω2 ; ω1 , ω2 , −ω1 ) , (5.22)
and for the special case of one beam, we have,
(3)
χijkl (−ω; ω, ω, −ω) . (5.23)
If we exchange the j and k tensor components and the associated two
degenerate frequencies under permutation symmetry, then,
(3) (3)
χijkl (−ω; ω, ω, −ω) = χikjl (−ω; ω, ω, −ω) , (5.24)
(3) (3)
and χ1212 = χ1122 . Therefore Equation 5.20 gives
(3) (3) (3)
χijkl = χ1122 (δij δkl + δik δjl ) + χ1221 δil δjk . (5.25)
Thus, the third order polarization with the appropriate degeneracy fac-
tor, is given by
3 (3)
P (3) = χijkl Ej Ek El∗ . (5.26)
4
Substituting the susceptibility from Equation 5.25 into Equation 5.26 yields

3 [ (3) ( ( ⃗ ⃗ ∗ ) ( )
⃗ ∗ Ek δik
)
P (3) = χ1122 Ej E · E δij + E ⃗ ·E
4 ( ) ]
+χ
(3)
E ⃗ E ∗ δie .
⃗ ·E (5.27)
1221 l
Downloaded from NLOsource.com 143

(3) (3)
Defining A = 32 χ1122 and B = 32 χ1221 , Equation 5.27 becomes
( ) ( )
P⃗ (3) = A E⃗ ·E
⃗ E ⃗ + 1B E ⃗ ·E⃗ E⃗ ∗. (5.28)
2
It is useful to write Pi in terms of the effective nonlinear susceptibility,

Pi = χef f
ij Ej , (5.29)

where
( ) 1 ( )
′ ⃗ ⃗∗
δij + B ′ Ei Ej∗ + Ei∗ Ej .
(1)
χef
ij
f
= χ ij + A E · E (5.30)
2
Here A′ = A − 12 B and B ′ = B. We can find the sum of the diagonal
components using the trace
( ) ( )
= 3 A′ + B ′ E · E ∗ .
(1)
T r χef f
ij − χij (5.31)

In the next section we show how A′ and B ′ are related to the mecha-
nisms of χ(3) . In particular by measuring them we will be able to determine
which mechanisms are playing an important role in any given process and
on various time scales.

5.3 Molecular Reorientation

We begin by considering the process of molecular reorientation, which is
slower than the electronic mechanism. We will assume the material to be
composed of a collection of freely rotating anisotropic molecules. Com-
mon examples of such materials are liquids and gases made of anisotropic
molecules with anisotropic polarization. Figure 5.7 is a diagram of a “cigar”-
shaped molecule where the electric field is applied along the z-axis. Here, α
⃗ is the electric field, P⃗ is the polarization, and
is the polarizability tensor, E
θ is the angle between the long axis of the molecule and the direction of the
electric field. Note that the induced dipole moment, p⃗, is not necessarily
along E⃗ or the principle axes of α.
An anisotropic molecule will tend to align with an applied electric field.
Thus, the refractive index of a medium made of anisotropic molecules in-
creases along the direction of the field, and the refractive index decreases
perpendicular to the direction of the electric field. Molecular reorientation
is due to an electric field induced torque, ⃗τ , which is given by

⃗τ = p⃗ × E.
⃗ (5.32)
144

E p
α3
θ

α1

Figure 5.7: An anisotropic molecule in an electric field with anisotropic

polarizability. Associated with the semi-major and semi-minor axes are the
polarizabilities, α3 and α1 , respectively.

It follows that the internal energy per molecule is,

U = −⃗
p · E.
⃗ (5.33)

Equation 5.33 shows that the energy is lowest when the molecule’s long
axis is aligned with the direction of the electric field. For a molecule that
lacks a permanent dipole moment, there is no preferred polar orientation;
that is, up and down orientation is equivalent. Figure 5.8 shows these lowest
and highest energy configurations.
The polarization of an anisotropic molecule is given by

Pi = αij Ej . (5.34)

When the polarizability is anisotropic, the internal energy is found by inte-

grating the right hand side of Equation 5.33. This gives,
∫ ⃗
E
U =− p⃗ · dE.
⃗ (5.35)
0

When a molecule is uniaxial as shown in Figure 5.7, the refractive

index ellipsoid is described by two refractive indices. The refractive index
component along the semi-minor axis is related to α1 and refractive index
component along the semi-major axis is related to α3 . Equation 5.35 for
Downloaded from NLOsource.com 145

(a) E (b) E

Figure 5.8: (a) The lowest orientational energy state; and, (b) the highest
orientational energy state of an ellipsoidal anisotropic molecule in a uniform
electric field.

a uniaxial molecule can be evaluated with the help of Equation 5.34 and
yields,
∫ E⃗
[ ]
U =− α3 E3′ dE3′ + α1 E1′ dE1′ . (5.36)
0
E3 and E1 are the electric field components along the direction of the semi-
major and semi-minor axis of the molecule, where E3 = E cos θ and E1 =
E sin θ. Upon integration, Equation 5.36 becomes,
1 ( )
U = − E 2 α3 cos2 θ + α1 sin2 θ . (5.37)
2
Figure 5.9 shows U (θ) given by Equation 5.37. We can rewrite this in a
simplified form in terms of cos θ,
1 [ ]
U = − E 2 α1 + (α3 − α1 ) cos2 θ . (5.38)
2
The electric field in a plane wave of light is given by
E = E0 cos (ωt) , (5.39)
where E0 is the amplitude and ω is the angular frequency. Then the internal
energy is proportional to cos2 (ωt). By using a trigonometric identity, we
can rewrite this as,
E0 [ ][ ]
U= 1 + cos2 (2ωt) α1 + (α3 − α1 ) cos2 θ . (5.40)
4
146

⃗
Figure 5.9: The energy per anisotropic molecule in an electric field, E.

Since the reorientational response time is on the order of 10−12 seconds,

any electric field of frequency greater than the reciprocal of the response
time will not lead to significant reorientation. Since an optical cycle is on
the order of 10−15 seconds and is therefore much shorter than reorientational
time scales, we can take a time average of all quantities over an optical cycle.
The time-averaged energy per molecule, U , is given by

1 [ ]
U = − E 2 α1 + (α3 − α1 ) cos2 θ . (5.41)
2
(1)
The linear refractive index is given by n2ii = 1 + 4πχii . Note that there
is no summation convention from double indices when the left hand side
of the equation has those same indices. The refractive index is then given
by the orientational ensemble average of the linear molecular susceptibility
according to,
n2ii = 1 + 4πN ⟨←
→
α ⟩ii (5.42)

The linear molecular susceptibility in the lab frame, αzz , can be found
by rotating the molecule from the molecular frame. Thus,
∑
αzz = αIJ aIz (Ω) aJz (Ω) , (5.43)
IJ

where the a’s are the rotation matrices and Ω represents the Euler angles θ,
ϕ, and ψ.
Downloaded from NLOsource.com 147

We define P (Ω) as the probability density of finding a molecule at the

angle Ω from Boltzmann statistics. Therefore,
e−βU
P (Ω) = ∫ (5.44)
dΩ e−βU
where β = kT with k denoting the Boltzmann constant, and T denoting
the temperature. Since no intermolecular forces are being considered in the
energy, the orientational ensemble average of the linear molecular suscepti-
bility is ∫
←→
⟨ α ⟩ii =
1
dΩαii (Ω) e−βU . (5.45)
Z
where dΩ = d (cos θ) dϕdψ. Z is the partition function defined as,
∫
Z = dΩe−βU . (5.46)

5.3.1 Zero Electric field

We will start by examining the susceptibility when no electric field is present.
αxx and αzz are then given by Equation 5.43;
αxx = α1 + (α3 − α1 ) sin2 θ cos2 ϕ, (5.47)
αzz = α1 + (α3 − α1 ) cos θ. 2
(5.48)
From Equation 5.41, when E ⃗ = 0, then U = 0, then P (Ω) = 1. Substi-
tuting Equations 5.47 and 5.48 into Equation 5.45, we get
∫ 2π ∫ 1 ∫ 2π
[ ]
dψ d (cos θ) dϕ α1 + (α3 − α1 ) sin2 θ cos2 ϕ
⟨←
→
α ⟩xx = 0 −1 0
∫ 2π ∫ 1 ∫ 2π ,
dψ d (cos θ) dϕ
0 −1 0
(5.49)
∫ ∫ ∫
2π 1 2π [ ]
dψ d (cos θ) dϕ α1 + (α3 − α1 ) cos2 θ
⟨←
→
α ⟩zz = 0 −1 0
∫ 2π ∫ 1 ∫ 2π . (5.50)
dψ d (cos θ) dϕ
0 −1 0
After integration, the two non-zero linear molecular susceptibilities are,

⟨←
→ 1 2
α ⟩xx = α3 + α1 (5.51)
3 3

⟨←
→ 1 2
α ⟩zz = α3 + α1 (5.52)
3 3
148

Equations 5.51 and 5.52 are equivalent to one another as would be expected
in an isotropic media consisting of anisotropic molecules which are randomly
oriented. Therefore, the refractive index given by Equation 5.42, can be
written as
4
n20 ≡ n2ii = 1 + πN (α3 + 2α1 ) . (5.53)
3

5.3.2 Non-Zero Electric field

⃗ ̸= 0, the
Next we consider the case when an electric field is present. When E
time-averaged energy per molecule due to an external field is greater than
zero. This means that the probability of finding a molecule at an angle Ω is
no longer uniform. We define
1
J = βE 2 (α3 − α1 ) , (5.54)
2
which is independent of the Euler angles and will thus remain after an
integration is performed to calculate the ensemble average.
The orientational ensemble average of the polarizability will be used to
determine the reorientational response. By, using Equations 5.48, 5.45, and
5.54, we get
∫ 1
[ ] 2
d (cos θ) α1 + (α3 − α1 ) cos2 θ eJ cos θ
⟨←
→
α ⟩zz = −1 ∫ 1 . (5.55)
J cos2 θ
d (cos θ) e
−1

When the magnitude of the field is small enough so that the alignment energy
is much smaller than thermal energies, we can expand the exponential term,
2
eJ cos θ
≈ 1 + J cos2 θ + · · · (5.56)
and
⟨←→ 1 2 2
α ⟩zz = α3 + α1 + βE 2 (α3 − α1 )2 . (5.57)
3 3 45
Substituting Equation 5.57 into Equation 5.42, the refractive index in the z
direction is
8πN
n2zz = n20 + βE 2 (α3 − α1 )2 . (5.58)
45
We define ∆n = nzz − n0 . But n2zz − n20 = (nzz − n0 ) (nzz + n0 ) ≡
(∆n) (2n). For a very small reorientationally-induced refractive index change,
∆n = δnzz , where n ≈ n0 , we can write
8πN
2n0 ∆n = βE 2 (α3 − α1 )2 . (5.59)
45
Downloaded from NLOsource.com 149

Therefore,
4πN
δnzz = βE 2 (α3 − α1 )2 . (5.60)
45n0
Defining δnzz = n2 E 2 , and including the Lorentz local field correction L (ω),
we can find the fourth rank tensor nzzzz
2 to be of the form,

4πN
nzzzz
2 = L4 (ω) β (α3 − α1 )2 . (5.61)
45n0
Then we can calculate nxxzz
2 using the relationship

1
nxxzz
2 = − nzzzz . (5.62)
2 2
Thus, the refractive index increases along the field and decreases perpendic-
ular to it, as expected.

5.3.3 General Case

In the most general case, there are three independent linear susceptibilities
that can be represented by the diagonal elements of ←→
α,
 
a 0 0
←→
α = 0 b 0 . (5.63)
0 0 c

The orientational ensemble average of ← →

α ij is the non-zero field result added
to the zero field result. Thus, using the Kronecker delta function, δij ,
( )
⟨αij ⟩ = Qδij + γij E 2 , (5.64)

where the first term on the righthand side of Equation 5.64 is from the
generalized zero field solution and the second term is attributed to the third
order susceptibility. Here, we state without proof that
1
Q= (a + b + c) , (5.65)
3
where a, b, and c are the diagonal elements in Equation 5.63.
The second term on the righthand side of Equation 5.64 is calculated
analogously to the derivative leading to Equation 5.61, yielding

γij = C (3δik δjl − δij δkl ) Ekloc Elloc . (5.66)

150

Here, the superscript “loc” refers to the local electric field. The constant C
is then of the form
β [ ]
C= (a − b)2 + (b − c)2 + (a − c)2 . (5.67)
90
We can represent the electric field as,

1
Ek (t) = Ek e−iωt + c.c. (5.68)
2
The polarization will be given by
(3)
∑
Pi =N γij Ej . (5.69)
j

Using Equations 5.66 - 5.68, the polarization can be found from Equation
5.69. In vector form, the polarization derived from Equation 5.69 is given
by,
( ) ( )
P⃗ (3) = A E ⃗∗ E
⃗ ·E ⃗ + 1B E ⃗ ·E
⃗ E ⃗ ∗, (5.70)
2
where A = N C/4 and B = 3N C/2.
Recalling that
δnlinear B
=1+ , (5.71)
δncircular 2A
δn
then the fraction δn linear = 4. This is larger than the 3/2 of the electronic
circular
response, and therefore the two mechanisms are easily separable. Figure
δn
5.10 shows a schematic diagram of the fraction δn linear as a function of
circular
laser pulse width.
This completes the full general description of molecular reorientation of
ellipsoidal molecules.

5.4 Measurements of the Intensity-Dependent Re-

fractive Index
In this section we will discuss the use of circularly or elliptically polarized
light to measure the intensity-dependent refractive indices of a liquid to
determine χ(3) . Using circular or elliptically polarized light, we can measure
the tensor components without having to keep track of phase factors. Recall
that we are considering the process governed by χ(3) (−ω; ω, ω, −ω).
Downloaded from NLOsource.com 151

δnlinear
Figure 5.10: δncircular as a function of the laser pulse width τ .

We begin by defining new unit vectors,

x̂ ± iŷ
σ̂± = √ , (5.72)
2
that contain real and an imaginary components. Using the same notation
as Boyd, σˆ+ describes left-hand circular polarization and σˆ− describe right-
hand circular polarization. Some properties of the unit vectors include:

σ̂± · σ̂± = 0 (5.73)

∗
σ̂± = σ̂∓ (5.74)
σ̂± · σ̂∓ = 1. (5.75)

The dot product of the unit vector with itself gives zero. This seems curious
since unit vectors are orthonormal, but since these unit vectors contain a
real and imaginary part, the dot product of the unit vector σ̂± with itself
gives zero. The complex conjugate of the unit vector σ̂± gives the opposite
unit vector, σ̂∓ . From these two properties, we can see that the dot product
of a unit vector σ̂± with its complex conjugate gives unity.
In electrodynamics, classical mechanics, and statistical mechanics, all
quantities are real. In quantum mechanics, wavefunctions are complex, al-
though measured values are always real. The complex quantities introduced
here are used as a computational tool while the actual measurements will
always yield real values.
The electric field can be written in terms of the unit vectors as a linear
152

combination of right circular and left circular polarized light,

⃗ = E+ σ̂+ + E− σ̂− .
E (5.76)

Some other properties of the electric field are

⃗ ·E
E ⃗ = 2E+ E− (5.77)
E ⃗ ∗ = |E+ |2 + |E− |2 ,
⃗ ·E (5.78)

where |E+ |2 and |E− |2 are the intensities of each of the two polarizations.
The dot product of the electric field vector with itself thus gives a cross term.
The dot product of the electric field vector times its complex conjugate gives
the total intensity.
Last time we considered the nonlinear polarization and derived an equa-
tion for the nonlinear polarization in terms of right-circular and left-circular
polarized light,

P⃗ N L = A(|E+ |2 + |E− |2 )E ⃗ ∗,
⃗ + B(E+ E− )E (5.79)

where A and B are constants that can be calculated from the χ(3) tensor. We
can rewrite the nonlinear polarization in terms of the circular polarization
unit vectors such that

P⃗ N L = P+N L σ̂+ + P−N L σ̂− , (5.80)

⃗ and P−N L contains the E−
where P+N L contains the E+ components of E
⃗
components of E,
∗
P+N L = A|E+ |2 E+ + A|E− |2 E+ + BE+ E− E− (5.81)

and
∗
P−N L = A|E+ |2 E− + A|E− |2 E− + BE+ E− E+ . (5.82)
Substituting Equations 5.81 and 5.82 into Equation 5.80 yields an Equa-
tion for P⃗ N L in terms of σ̂+ and σ̂− ,
∗
P⃗ N L = A|E+ |2 E+ σ̂+ + A|E− |2 E+ σ̂+ + BE+ E− E− σ̂+
∗
+A|E+ |2 E− σ̂− + A|E− |2 E− σ̂− + BE+ E− E+ σ̂− . (5.83)
∗ = |E |2 , we can simplify Equation 5.83 by combining these
Since E± E± ±
terms,

P⃗ N L = A|E+ |2 E+ σ̂+ + (A + B)|E− |2 E+ σ̂+

+A|E− |2 E− σ̂− + (A + B)|E+ |2 E− σ̂− . (5.84)
Downloaded from NLOsource.com 153

Recall that P±N L = χN L

± E± . In previous sections we have used χ
ef f , but

this is the same as χN L . We can solve for the nonlinear susceptibility term
in terms of the squared electric field such that

± = A|E± | + (A + B)|E∓ | .
χN L 2 2
(5.85)

This equation depicts the asymmetry of the refractive index for positive and
negative circular polarizations.
From the nonlinear susceptibility, we can write the refractive index as a
function of intensity,

2π
n± ≈ n0 + [A|E± |2 E+ + (A + B)|E∓ |2 E− ], (5.86)
n0

where n0 is the isotropic index of refraction. As previously shown, we have 2

independent tensors: χ1122 and χ1221 . Circularly polarized light probes both
components and the fields induce a refractive index change. In other words,
the field induces birefringence in the material. The left and right circularly
polarized light probe different tensor components since there are different
phase velocities for each polarization, and therefore both are required to
measure all of the susceptibility tensor components. Linear polarized light
is unaffected, because, if the sample is rotated, it remains unchanged by
virtue of the fact that it is isotropic.

5.5 General Polarization

5.5.1 Plane Wave
The most general form of a plane wave written in terms of right and left
circular polarization unit vectors is
[ ] [ ]
⃗ in+ ωz in− ωz
E(z) = A+ exp σ̂+ + A− exp σ̂− , (5.87)
c c

where A+ and A− are the amplitudes of the left and right circular polarized
components. The components of the wave propagate at nc+ and nc− . This
expression can be simplified by defining a phase factor and a k-vector,
( )
⃗
E(z) = A+ eiθ σ̂+ + A− e−iθ σ̂− eikm z , (5.88)
154

where
1 ω
θ = ∆n z (5.89)
2
( c )
1 ω
km = n− + ∆n (5.90)
2 c
2πB ( )
∆n = n+ − n− = |E− |2 − |E+ |2 . (5.91)
n0

5.6 Special Cases

5.6.1 Linear Polarization
For the case of linear polarization, the left polarized electric field component
is equal to the right polarized electric field component, E− = E+ , and the
refractive index difference is zero, that is ∆n = 0. The amplitudes of the
left and right polarization electric fields are equivilant, so A− = A+ . The
left and right circular components of the refractive indices are defined as,
2π [ ]
n± = A|E+ |2 + (A + B)|E+ |2 + n0 . (5.92)
n0
Since the right and left polarized electric field components are equal, the
intensity of light can be written as twice the intensity of the left polarization,
|E|2 = 2|E+ |2 . Substituting 21 |E|2 for |E+ |2 in Equation 5.92, we get
[( ) ]
2π B
n± = A+ |E| + n0 .
2
(5.93)
n0 2
We can write the nonlinear term as the difference between the refractive
indices of the left and right linear polarizations as
[( ) ]
2π B
δnlinear = A+ |E|2
(5.94)
n0 2
= n± − n0 . (5.95)

5.6.2 Circular Polarization

For the case of left circular polarized light, the electric field for left circular
polarized light is nonzero and the the electric field for right circular polarized
light is zero, E+ ̸= 0 and E− = 0. The general form for the electric field is
in+ ωz
⃗
E(z) = A+ e c σ̂+ . (5.96)
Downloaded from NLOsource.com 155

Using Equation 5.86, the nonlinear term for the change in refractive index
for left and right polarization is
2π [ ]
δncircular = A|E± |2 (5.97)
n0
= n+ − n0 . (5.98)

In order to construct the susceptibility tensor for a material, one needs

two independent experiments. The first experiment uses linear polarized
light to measure n± and we get (A + B2 ). The second experiment uses
circular polarization to measure n± from which we get A. Thus, we can use
these two experiments to find the coefficients A and B.
Different mechanisms have different tensor properties. When measuring
the susceptibility tensor for a material, one can take the ratio of different
components to find the mechanisms responsible for the nonlinearities.

5.6.3 Elliptical Polarization

Consider the case of elliptically polarized light. The electric field vector can
be written as Equation 5.88 but with A+ ̸= A− , A+ ̸= 0, and A− ̸= 0. In
the first term, we can write the unit vector σ̂+ in terms of x and y unit
vectors as a function of phase angle,

x̂ + iŷ
σ̂+ eiϕ = √ (cos θ + i sin θ) . (5.99)
2
Expanding the x̂ and ŷ terms yields
1
σ̂+ eiϕ = √ [x̂ cos θ − ŷ sin θ + i (x̂ sin θ + ŷ cos θ)] . (5.100)
2
We can define the real terms and the imaginary terms as

x̂′ = x̂ cos(θ) − ŷ sin(θ) (5.101)

and
ŷ ′ = x̂ sin(θ) + ŷ cos(θ). (5.102)
Then we can relate the set of coordinates x̂, ŷ to the transformed set of
coordinates x̂′ , ŷ ′ using a matrix rotation,
[ ′ ] [ ][ ]
x̂ cos θ − sin θ x̂
′ = . (5.103)
ŷ sin θ cos θ ŷ
156

y y’

x’

Figure 5.11: Rotation of axes x̂ and ŷ by an angle θ to give new axes x̂′ and
ŷ ′ .

The axes x̂, ŷ represent the axes of the polarization ellipse of the beam.
The same rotation can be applied to the unit vector σ̂− . The matrix trans-
formation rotates the axes in a clockwise direction, as shown in Figure 5.11.
In general, the net effect is that the axes of the ellipse rotate as the light
propagates through the material. The light remains left and right polarized,
but there is a phase shift. Physically, this represents rotation of the axes
of the elliptical light as it propagates due to a phase shift between polar-
izations. Recall the refractive index difference from Equation 5.91 and the
change in the angle of axes from Equation 5.89 and notice that the rate of
rotation depends on the difference in the refractive index. The degree of
rotation depends on the amount of material the light propagates through.
If we measure the change in angle of the axes, θ, we can find the difference
in refractive index ∆n, and obtain the constant, B.

5.7 Mechanisms

In order to calculate the tensor properties of the different mechanisms of χ(3) ,

we will consider how different mechanisms contribute to A and B. Different
mechanisms will give rise to different ratios of A and B. So by measuring
A and B and taking the relative ratios, we can determine the mechanisms
causing changes in the refractive index.
Downloaded from NLOsource.com 157

5.7.1 Electronic Response

An electronic response comes from moving charges. Recall the classical
anharmonic oscillator, for example, where the general three-dimensional po-
tential energy can be written as

1 1
U (⃗r) = mω02 |⃗r|2 − mb |⃗r|4 . (5.104)
2 4

In this case, the system is centrosymmetric, so χ(2) = 0. The solution for

the general χ(3) term in three dimensions is

(3) N be4 [δij δkl + δik δjl + δil δjk ]

χijkl (−ω; ω, ω, −ω) = , (5.105)
3m3 D3 (ω)D(−ω)

where D(ω) = ω02 − ω 2 − 2iωΓ.

Recall that A and B in terms of the χ(3) components are

(3) (3)
A = 3χ1122 + 3χ1212 (5.106)
(3)
B = 6χ1221 . (5.107)

(3)
For χ1122 , only the first term in Equation 5.105 is nonzero. Similarly, for
(3) (3)
χ1212 , only the second term is nonzero. Also, for χ1221 , the last term is
nonzero. But each of the terms in brackets in Equation 5.105 are unity, so
this nonlinear oscillator model gives A = B. The ratio of the measurements
δnlinear and δncircular yields

δnlinear B
= 1+ (5.108)
δncircular 2A
3
= . (5.109)
2

For the electronic response mechanism, the ratio of δnlinear and δncircular
is 32 . Notice that this ratio is independent of ω because the denominator is
the same for all of the tensor components.
If a combination of χ(3) mechanisms is acting, there will be a convolution
of the ratios, so the mechanisms may not be simple to isolate. In this case
we can use measurements of varying temporal pulse widths that probe the
time scales characteristic of each mechanism.
158

5.7.2 Divergence Issue

Recall the general form of χ(3) is written as

(3)
χkjih (−ωσ ; ωr , ωq , −ωp ) =
[ ]
∑ µk0n µjnm µiml µh
N
P
h̄3 F lmn
l0
(ωn0 −ωσ )(ωm0 −ωq −ωp )(ωl0 −ωp ) , (5.110)

where PF is the full permutation symmetry operator over the frequencies

ωσ , ωq , and ωp and the sum is over all states, including the ground state.
The outgoing frequency is a sum over the incoming frequencies,

ω σ = ωr + ωq + ωp . (5.111)

There is a divergence problem in the case where there is one beam such
that ωr → ω, ωq → ω, ωp → −ω. In the ground state, ω00 = 0. Adding
two of the beams together yields zero, −ωq − ωp = 0, which diverges at all
wavelengths. In order to solve the divergence problem, we would need to
write out all 24 terms for χ(3) which would yield two types of denominators
of the general form,
1 1
+ . (5.112)
(X + Y )Y (X + Y )X

When X = −Y , both terms diverge; however if we sum both terms, we get

1 X 1 Y X +Y
+ = (5.113)
(X + Y )Y X (X + Y )X Y XY (X + Y )
1
= (5.114)
XY
−1
= , (5.115)
X2

so the sum of two diverging terms yields a well-behaved one.

5.7.3 One- and Two-Photon States

When we add all of the 24 terms and eliminate the divergence problem by
combining divergent terms as described above, we get two different types of
Downloaded from NLOsource.com 159

energy denominators,
(3)
χkjih (−ωσ ; ωr , ωq , −ωp ) =
[ ′
N ∑ µk0n µjnm µiml µhl0
P F
h̄3 (ωn0 − ωσ )(ωm0 − ωq − ωp )(ωl0 − ωp )
lmn
′
]
∑ µk0n µjn0 µi0l µhl0
− . (5.116)
(ωn0 − ωσ )(ωl0 − ωq )(ωl0 − ωp )
ln
∑′
The indicates a summation excluding the ground state. The first term
is non-vanishing only when a two-photon state is included while the second
term requires only one-photon states. If we assume that the incident photons
are on or near resonance, then ωm0 − ωq − ωp ≈ 0 and ωl0 − ωp ≈ 0. Then,
an incoming photon will match state n and will resonate.
even m

odd n
odd

even 0

Figure 5.12: State n is a one-photon state where only one photon is required
to induce a transition to this state. State m is a two-photon state where
direct transitions from state 0 to state m are disallowed but transitions from
states 0 to n and n to m are allowed.

The energy levels are illustrated in Figure 5.12 where the ground state,
0, is assumed to be of even parity; the state, n, is odd; and the state, m,
is even. Additionally, n is a one-photon state such that transitions from 0
to n or n to m require one photon. The state m is a two-photon state such
that one-photon cannot induce a transition from 0 to m.
From the quantum mechanical viewpoint, wave functions for centrosym-
metric potentials have either even or odd parity. If the ground state is even,
then the first transition state must be odd and the next transition state
must be even. Therefore a transition from an even state to an even state
is disallowed. In order to excite a transition to a state m, two photons are
required where one photon induces a transition from the ground state to the
odd-parity state and a second photon induces a transitions from the odd
state to the even state. Both states n and l are one-photon states.
160

l n

Figure 5.13: In the second term in Equation 5.116 only one-photon state
transitions from the ground state, 0, to the first excited state, n or l, are
allowed, but there are no two-photon states.

(3)
χ1111 will have both terms in Equation 5.116 where both one- and two-
(3)
photon states are present. For χ1221 , the first term in Equation 5.116 does
not have to be zero but the second term must be zero. This can be illustrated
y

Figure 5.14: The ground state and first excited state wave functions in the
x and y directions are depicted, where the ground state is even and the first
excited state is odd. |ex0 ⟩ is red, |ox1 ⟩ is green, |ey0 ⟩ is blue, and |oy1 ⟩ is
orange.

in terms of wavefunctions inside a two-dimensional box as shown in Figure

5.14. In the most general case, the ground state will be an even state and
then the first excited state is odd. Consider wavefunctions along x and y,
ψn (x) and ψm (y). Then the centrosymmetric ground state can be written
as
|ψ1 ⟩ = |ex0 ⟩ |ey0 ⟩ , (5.117)
Downloaded from NLOsource.com 161

and the excited states can be written as

|ψ2 ⟩ = |oxn ⟩ |oym ⟩ , (5.118)

where e and o refer to the even and odd parity state vectors, respectively.
In the case where the first excited state is odd in the x-direction and even
in the y-direction, the wave function,

|ψ2 ⟩ = |oxn ⟩ |ey0 ⟩ . (5.119)

has a nonzero x-transition moment between the ground state and the first
excited state,
⟨ψ2 |x| ψ1 ⟩ ̸= 0, (5.120)
which we can see as follows.
We can use the wave functions defined in Equations 5.117 and 5.119 to
calculate the transition moment,

⟨oxn |⟨ey0 |x| ex0 ⟩| ey0 ⟩ . (5.121)

This can be simplified since the y states do not act on x, so we can move
them outside of the bras and kets and then using ⟨ey0 | ey0 ⟩ = 1, because
the states are orthonormal, we get,

⟨oxn |x| ex0 ⟩ ̸= 0. (5.122)

This represents a transition from an even state to an odd state.

If we use these state vectors that give nonzero transitions in the x-
direction to calculate the transition moment of y, we get

⟨ψ2 |y| ψ1 ⟩ = 0, (5.123)

because
⟨ψ2 |y| ψ1 ⟩ = ⟨0xn |⟨ey0 |y| ex0 ⟩| ey0 ⟩ . (5.124)
Since the x operators do not act on y, we can move them outside the other
bras and kets and are left with

⟨ey0 |y| ey0 ⟩ = 0. (5.125)

This expectation value vanishes because the bra, ket, and y operator are all
odd in the y-direction.
(3)
Equation 5.125 shows that χ1221 cannot have two-photon character, but
(3)
only have one-photon character. Thus, χ1221 has only two-photon character
162

(3)
while χ1122 has both one- and two-photon character. The ratio of the χ(3)
(3)
χ1221
components, (3) is frequency-dependent so we cannot uniquely determine
χ1122
A and B, and therefore, we cannot learn about the mechanisms using any
combination of polarized measurements.
In the off-resonance case, ω → 0 so that all frequencies vansish. Then,
the two-photon terms are greater than or equal to zero and the one-photon
terms are less than or equal to zero. Using Klienman symmetry, where we
permute only the two interior wavelengths, then A = B. In this case, the
ratio of the change in the linear refractive index to the change in the circular
refractive index is equal to 32 ,

δnlinear B
= 1+ (5.126)
δncircular 2A
3
= . (5.127)
2
Thus, polarization-dependent measurements that yield A and B can be used
to determine if the electronic or reorientational mechanisms are responsible.
Chapter 6

Applications

6.1 Optical Phase Conjugation

Let us consider a plane wave incident on a mirror. The incident plane wave
field is given by
E⃗ in = ϵ̂ A(⃗r) ei(kz−ωt) + c.c. (6.1)
2
The polarization ϵ̂ can be either circular, linear, or elliptical. From our
boundary conditions the field at the surface must be zero
⃗r + E
(E ⃗ in )|z=0 = 0, (6.2)
where E⃗ r and E
⃗ in ) are the reflected and incoming fields respectively. Because
the reflected wave moves opposite to the incident wave we find
⃗ r = −ϵ̂ A(⃗r) ei(−kz−ωt) + c.c.,
E (6.3)
2
which is similar to the incoming wave where the polarization has been re-
versed and the wave vector is in the negative direction. We can incorporate
the spatial portion of the exponential into A(⃗r) and we find

⃗ = ϵ̂ A(⃗r) e−iωt + c.c.

E (6.4)
2
The phase conjugate wave is defined as

∗
⃗ c (⃗r, t) = ϵ̂∗ A (⃗r) e−iωt + c.c,
E (6.5)
2
A(⃗r) iωt
= ϵ̂ e + c.c, (6.6)
2
⃗ r, −t).
= E(⃗ (6.7)

163
164

Reflected

Incoming

Figure 6.1: In a phase conjugate mirror the beam retraces the original path
in reverse.

Where equation 6.6 is obtained by exchaning the two terms in equation 6.5.
Phase conjugation is thus equivalent to time reversal, where t → −t. The
difference between reflection from a regular mirror and a phase conjugate
mirror is shown in Figure 6.1 and Figure 6.2

6.2 Phase Conjugate Mirror

A phase conjugate mirror can be made using degenerate four wave mixing, a
process that is mediated by χ(3) . We begin with two pump beams (E1 , E2 ),
a probe beam E3 and the reflected beam E4 (see figure 6.3). The source of
phase conjugation can be understood by recognizing that the probe beam
and one of the pump beams interfere and form a “grating” from which the
other pump scatters. The scattered pump beam then travels in the direction
of the phase conjugate beam, as shown in figure 6.3
For our derivation we assume the pump beams are much more intense
than the probe and reflected beams, therefore we can assume that pump
beam depletion is negligible. We will also assume that the fields are scalars,
or,

⃗ i (⃗r, t) ⇒ A(⃗r) ei(⃗ki ·⃗r−ωt) + c.c..

E (6.8)
2
Note that the polarization of the fields determine which components of the
χ(3) tensor are responsible, so long as χ(3) ̸= 0 for the components of interest,
Downloaded from NLOsource.com 165

Phase Front

PCM Mirror

Figure 6.2: When a wavefront is incident on a mirror, the reflected wave’s

phase front is reversed. In a phase conjugate mirror the phase front is the
same as that of the incident wave.

Figure 6.3: In a phase conjugate mirror, two counter propogating pump

beams(1,2) interact with a probe beam(3) to create a phase conjugate(4).
166

the scalar approximation does not limit the generalility of the calculation.
From the wave equation we know

3
E4 ∝ (P N L(3) ) = χ(3) E1 E2 E3∗ . (6.9)
2
Substituting the fields from equation 6.8 into equation 6.9,

3 ⃗ ⃗ ⃗
E4 ∝ χ(3) A1 A2 A∗3 ei(k1 +k2 −k3 )·⃗r . (6.10)
2
Where the two pump beams are counter propogating ⃗k1 = −⃗k2 , so equation
6.10 becomes

3 ⃗
E4 ∝ χ(3) A1 A2 A∗3 e−ik3 ·⃗r , (6.11)
2
which shows that E4 ∝ E3∗ making it the phase conjugate wave. Note that
if A1 and A2 are sufficiently large, E4 can be an amplified version of E3 .
To solve the problem in detail we consider the polarizations to first order
in E3 and E4 as they are much smaller than E1 and E2 . The third order
polarizations are

3 (3) [ 2 ∗ ]
P1 = χ E1 E1 + 2E1 E2 E2∗ , (6.12)
2
3 (3) [ 2 ∗ ]
P2 = χ E2 E2 + 2E2 E1 E1∗ , (6.13)
2
3 (3)
P3 = χ [2E3 E1 E1∗ + 2E3 E2 E2∗ + 2E1 E2 E4∗ ] , (6.14)
2
3 (3)
P4 = χ [2E4 E1 E1∗ + 2E4 E2 E2∗ + 2E1 E2 E3∗ ] . (6.15)
2
To solve for P4 , we must substitute the polarizations into the third or-
der nonlinear wave equation and solve using the slowly varying envelope
approximation. For A1 we find that

( )
d εω 2 −4π 1 (3) ( 2 2
)
−k12 + 2ik1 ′ + 2 A1 = χ |A1 | + 2 |A 2 | A1 , (6.16)
dz c c2 2

where z ′ is along k1 . Remembering that k1 = ϵω

2
c2
, and that A1 and A2 are
counter propagating, we find two similar differential equations for A1 and
A2
Downloaded from NLOsource.com 167

dA1 iπω (3) ( 2 2

)
= χ |A 1 | + 2 |A 2 | A1 = iκ1 A1 , (6.17)
dz ′ nc
dA2 −iπω (3) ( 2 2
)
= χ |A 2 | + 2 |A 1 | A2 = −iκ2 A2 , (6.18)
dz ′ nc
where κ1 and κ2 are defined as

πω (3) ( )
κ1 = χ |A1 |2 + 2 |A2 |2 , (6.19)
nc
πω (3) ( )
κ2 = χ |A2 |2 + 2 |A1 |2 . (6.20)
nc
Under the plane wave assumption the phase changes as the wave propa-
gates, but its amplitude|A|2 remains constant. Thus, κ1 and κ2 are constants
and the solutions for A1 and A2 are trivially complex exponentials

′
A1 (z ′ ) = A1 (0)eiκ1 z , (6.21)
′ iκ2 z ′
A2 (z ) = A2 (0)e . (6.22)

We have now determined the fields E1 and E2 . Since we are dealing

with plane waves, |E1 |2 and |E2 |2 are constants, so the only term that isn’t
constant is E1 E2 . From our solution above
′
E1 E2 = A1 (0)A2 (0)ei(κ1 −κ2 )z . (6.23)
Substituting E1 E2 into the nonlinear wave equation for A3 and A4 yields,

dA3 2iπω (3) [( ) ]

= χ |A1 |2 + |A2 |2 A3 + A1 A2 A∗4 (6.24)
dz ′ nc
dA4 −2iπω (3) [( ) ]
= χ |A1 |2 + |A2 |2 A4 + A1 A2 A∗3 . (6.25)
dz ′ nc
Now if we assume |A1 | = |A2 |, easily attainable using one laser and
a beam splitter, we find that κ1 = κ2 from Equation 6.21 and Equation
6.22, and that A1 A2 is constant. This implies automatic phase matching as
κ1 − κ2 → 0. In this case equation 6.24 and equation 6.25 become

dA3
= iκ3 A3 + iκA∗4 , (6.26)
dz
dA4
= −iκ3 A3 − iκA∗3 . (6.27)
dz
168

Solving equation 6.26 and equation 6.27

A′∗
3 (0)
A′∗
3 (L) = , (6.28)
cos (|κ|L)
iκ
A′∗
4 (0) = tan(|κ|L)A′∗
3 (0). (6.29)
|κ|

where A′4 is an amplified version of A3 and L is the length the beam has
traveled into the material.
List of Tables

5.1 χ(3) and the response times of several mechanisms. . . . . . . 140

175
176
List of Figures

1.1 Kerr observed the change in transmittance through a sample

between crossed polarizers due to an applied voltage. Inset at
the bottom left shows the orientations of the polarizers and
the applied electric field due to the static voltage. . . . . . . . 2
1.2 (left)A system is excited by a photon if its energy matches
the difference in energies between two states. (right) Two-
photon absorption results when two photons, each of energy
(E2 − E1 ) /2, are sequentially absorbed. . . . . . . . . . . . . 3
1.3 The experiment used by Franken and coworkers to demon-
strate second harmonic generation. The inset shows an artis-
tic rendition of the photograph that recorded the two beams. 4
1.4 In the Optical Kerr Effect, a strong pump laser is polarized
45o to the weak probe beam, causing a rotation of its polar-
ization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Polarizer set perpendicular to the original wave propagation
direction (the first defines the polarization of the incident wave.) 6
1.6 Material polarized by an external electric field. . . . . . . . . 12
1.7 Plot of the electric field across the interface of a polarized
material. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.8 Light inducing a polarization inside a material. The light
source is a monochromatic plane wave. . . . . . . . . . . . . . 14
1.9 A small volume element of material in an electric field . . . . 15

2.1 Charge On A Spring . . . . . . . . . . . . . . . . . . . . . . . 22

2.2 Harmonic Oscillator Free Body Diagram . . . . . . . . . . . . 22
2.3 N Number of Springs . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 Susceptibility Divergence . . . . . . . . . . . . . . . . . . . . 24
2.5 Harmonic Oscillator Potential . . . . . . . . . . . . . . . . . . 24
2.6 Real and Imaginary Susceptibility . . . . . . . . . . . . . . . 29

177
178

2.7 Plot of the real and imaginary parts of χ(1) . ω0 is the resonant
frequency of the system. . . . . . . . . . . . . . . . . . . . . . 32
2.8 Map of resonances of χ(2) . . . . . . . . . . . . . . . . . . . . 33
2.9 Illustration of the macroscopic model that defines refractive
index. Longer (red) lines represent phase fronts of the exter-
nal electric field in and outside the material and the shorter
(dark blue) lines represent the electric field generated by the
induced dipoles fields (arrows) inside the sample. The super-
position of the induced electric dipole fields and plane waves
inside the sample leads to a plane wave with a decreased phase
velocity, which defines the concept of the refractive index. . . 34
2.10 Optical Kerr Effect . . . . . . . . . . . . . . . . . . . . . . . . 45
2.11 The value of the contour integral is composed of contributions
from the three major pieces of the path: (1) The semicircle of
increasing radius for which ωI > 0. (2) The real axis, which
avoids the singular point ω = ω ′ by detouring around on a
semicircular arc. (3) The pole ω = ω ′ is integrated using
Cauchy’s integral formula. . . . . . . . . . . . . . . . . . . . . 48

3.1 Slowly varying envelope approximation: the variation of A(z)

with respect to z is much smaller than the wavelength of the
field’s oscillation. . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2 Sum frequency generation: the nonlinear interaction of two
incident waves of frequencies ω1 and ω2 with a sample of with
thickness l results in the generation of a wave of frequency
ω3 = ω1 + ω2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3 The intensity of generated light as a function of sample length,
l, is proportional to A1 A2 χ(2) and a small-amplitude oscillat-
ing function in the non-depletion regime (solid red curve) and
in the phase-matching condition (the dashed green curve), it
is proportional to l2 . . . . . . . . . . . . . . . . . . . . . . . 62
3.4 Sum frequency generation in the small depletion regime is
represented by the nonlinear interaction of two incident waves
with frequencies ω1 and ω2 within the material, to generate a
photon of frequency ω3 . Downstream, this generated photon
and a photon of frequency ω2 interact and create a photon of
frequency ω1 . The inset shows how the energy is conserved
in the process of the destruction of a photon of frequency ω3
and creation of two photons of frequencies ω1 and ω2 . . . . . 64
Downloaded from NLOsource.com 179

3.5 The Manley-Rowe equation expresses the fact that the abso-
lute change in the number of photons at every frequency is
the same in each interaction. For example, the destruction
of one photon at ω1 and one photon at ω2 yields one photon
created at frequency ω3 . . . . . . . . . . . . . . . . . . . . . . 67
3.6 Comparison of the intensities of the three beams of frequency
ω1 , ω2 , and ω3 . in the sum frequency generation process. I1
and I2 have their minima when I3 peaks. . . . . . . . . . . . 68
3.7 The idler, ω1 , and the pump, ω2 , undergo sum frequency
generation in a nonlinear crystal and create an output with
frequency ω3 = ω1 + ω2 . . . . . . . . . . . . . . . . . . . . . . 69
3.8 The intensities at frequencies ω1 and ω3 plotted as a function
of propagation distance. . . . . . . . . . . . . . . . . . . . . . 71
3.9 The pump beam at frequency ω1 and the amplified beam
at frequency ω2 undergo difference frequency generation in
a nonlinear crystal. This creates the difference beam at fre-
quency ω3 = ω1 − ω2 . . . . . . . . . . . . . . . . . . . . . . . . 72
3.10 The intensities at frequencies ω2 and ω3 as a function of depth
into the material. . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.11 In second harmonic generation, two incident wave of fre-
quency ω interact with a nonlinear medium, generating an
output wave of frequency 2ω. . . . . . . . . . . . . . . . . . . 75
3.12 This process is difference frequency generation as we have seen
in the previous section. The fundamental wave of frequency
ω and the generated second harmonic wave of frequency 2ω
interacts with the material and propagates through the ma-
terial resulting in an output wave of frequency ω. . . . . . . . 75
3.13 Intensities oscilate while two waves of of frequencies ω and
2ω propagate in the medium under phase-matching condition.
This behavior is due to the initial non-zero incident intensities
of both waves. . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.14 The energy is converted to the second harmonic wave from
the fundamental wave. Notice that initially the slope of the
fundamental wave is zero but that of the second harmonic
wave begins with a non-zero value. . . . . . . . . . . . . . . . 78
3.15 Phase-mismatching reduces the efficiency of generating sec-
ond harmonic wave significantly. . . . . . . . . . . . . . . . . 78
3.16 The undepleted limit solution of coupled equations. . . . . . . 79
180

3.17 Rotating a quartz sample changes the effective thickness of

the sample. Consequently, the intensity of the second har-
monic generation varies with respect to the orientation of the
quartz sample. . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.18 The intensity of second harmonic wave depends on the refrac-
tive angle ϕ, of the fundamental wave. Here we neglect the ϕ
dependence of the transmittance. . . . . . . . . . . . . . . . . 80
3.19 The intensity of second harmonic generation varies with re-
spect to the orientation of a quartz sample. . . . . . . . . . . 81
3.20 Phase-matching can not be found in the normal dispersion
region. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.21 A specific phase-matched pair of wavelengths may be found in
the anomalous region, but not arbitrary wavelengths satisfy
the phase-matching condition. . . . . . . . . . . . . . . . . . . 82
3.22 Uniaxial birefringence is characterized by an ellipsoid show-
ing the ordinary refractive index no and the extraordinary
refractive index ne . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.23 A plane wave travels along k̂ direction. no is the component
of the refractive index along the perpendicular polarization
E⊥ , and ne is the component along the parallel polarization
E∥ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.24 The dispersion of the refractive index under different orienta-
tions of a positive uniaxial crystal. Second harmonic genera-
tion can be achieved by varying the orientation of the crystal. 85
3.25 The dispersion of the refractive index under different orien-
tations of a negative uniaxial crystal. . . . . . . . . . . . . . . 86

4.1 Space is equally divided into unit cells of volume L3 . . . . . 89

4.2 Spontaneous emission of a photon due to de-excitation . . . . 95
4.3 Stimulated emission from deexcited system in the presence of
identical photons . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.4 Particle trajectories of collision. (left): Inelastic collision;
(right): Elastic collision. . . . . . . . . . . . . . . . . . . . . . 106
4.5 Space-time diagram for above. . . . . . . . . . . . . . . . . . 106
4.6 The Feynman diagram that represents the interaction be-
tween two electrons. . . . . . . . . . . . . . . . . . . . . . . . 107
4.7 Molecular absorption of light . . . . . . . . . . . . . . . . . . 107
4.8 The two Feynman diagrams for the linear susceptibility . . . 108
4.9 Example for vertex rule when photon is absorbed . . . . . . . 109
4.10 Example for vertex rule when photon is emitted . . . . . . . . 110
Downloaded from NLOsource.com 181

4.11 Example for propagator rule to second-order in the field. . . . 111

4.12 Feynman diagrams for sum frequency generation . . . . . . . 112
4.13 Sum frequency generation case a . . . . . . . . . . . . . . . . 113
4.14 Outgoing photon in the propagator . . . . . . . . . . . . . . . 113
4.15 A plot of the imaginary part of α as a function of energy is
used to demonstrate the decay width. . . . . . . . . . . . . . 114
4.16 a) The absorption spectrum of three different molecules with
different velocities (red), and the Boltzman distribution of
peak positions (dashed curve). b) Shift in the spectrum of
molecules that are interacting with each other. c) A schematic
picture of the superposition of molecules in an ensemble weighted
by the Boltzman factor, leading to a inhomogeneous broad-
ening. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.17 a) Discrete model: the material consists of atoms or molecules.
b) Continuous model: The material is assumed a uniform di-
electric and a molecule inside the material is treated as a
discrete entity inside a cavity. . . . . . . . . . . . . . . . . . . 130
4.18 An electric field applied to a material induces charges on its
surfaces. A molecule in the material, modeled as being inside
a dielectric cavity (small arrow), is affected by both the ap-
plied field and the polarization field. The polarization field
induced in the cavity is shown in the inset. . . . . . . . . . . 131
4.19 If the cavity radius is much smaller than the plane wave’s
wavelength, the electric field will be spatially uniform inside
the cavity, which is necessary for the Lorentz-Lorenz local
field model. The arrows are the electric fields. . . . . . . . . . 132

5.1 A pump beam of light influences the propagation of a probe

beam. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.2 Electronic mechanism of χ(3) . A molecule with a nuclear
framework represented by a blue arrow, and electrons rep-
resented by the surrounding green dots. The electron cloud
is deformed by the electric field, as shown above. Note that
the degree of deformation is greatly exaggerated. . . . . . . . 137
5.3 Reorientational mechanism of χ(3) . The nuclei of each molecule
are shown as arrows. . . . . . . . . . . . . . . . . . . . . . . . 138
5.4 Electrostriction mechanism of χ(3) . The effective width of
the light beam is shown by the dotted lines. The green dots
represent molecules in solution. . . . . . . . . . . . . . . . . 138
5.5 Saturated absorbtion mechanism of χ(3) . . . . . . . . . . . . . 139
182

5.6 Thermal mechanism of χ(3) . . . . . . . . . . . . . . . . . . . . 139

5.7 An anisotropic molecule in an electric field with anisotropic
polarizability. Associated with the semi-major and semi-minor
axes are the polarizabilities, α3 and α1 , respectively. . . . . . 144
5.8 (a) The lowest orientational energy state; and, (b) the high-
est orientational energy state of an ellipsoidal anisotropic
molecule in a uniform electric field. . . . . . . . . . . . . . . . 145
5.9 The energy per anisotropic molecule in an electric field, E. ⃗ . 146
δnlinear
5.10 δncircular as a function of the laser pulse width τ . . . . . . . . . 151
5.11 Axes Transformation . . . . . . . . . . . . . . . . . . . . . . . 156
5.12 Transitions for One- and Two-Photon States . . . . . . . . . 159
5.13 Transitions for One-Photon States in a Liquid . . . . . . . . . 160
5.14 Wavefunctions in x and y . . . . . . . . . . . . . . . . . . . . 160

6.1 Phase Conjugate Mirror . . . . . . . . . . . . . . . . . . . . . 164

6.2 Phase Conjugate Mirror . . . . . . . . . . . . . . . . . . . . . 165
6.3 Phase Conjugate Mirror . . . . . . . . . . . . . . . . . . . . . 165
( )
7.1 For α = 0 the transmittance is given by T → sin2 2θ sin2 ∆nkd 171
−2
2 ( )
7.2 For ∆n = 0 the transmittance is given by T → sin 2θe
2 ᾱkd sinh ∆αkd2 171
Bibliography

[1] J. Kerr, “A new relation between electricity and light: dielectrified media
birefringent,” Phil. Mag. S. 50, 337–348 (1875).

[2] J. Kerr, “Electro-optic observations on variaou liquids,” Phil. Mag. 8,

85–102,229–245 (1875).

[3] J. Kerr, “Electro-optic observations on various liquids,” J. Phys. Theor.

Appl. 8, 414–418 (1879).

[4] T. H. Maiman, “Stimulated optical radiation in ruby,” nature 187, 493–

494 (1960).

[5] M. Goeppert-Mayer, “ber Elementarakte mit zwei Quantensprngen,”

Annalen der Physik 401, 273–294 (1931).

[6] P. A. Franken, A. E. Hill, C. W. Peters, and G. Weinreich, “Generation

of Optical Harmonics,” Phys. Rev. Lett. 7, 118–119 (1961).

[7] G. Mayer and F. Gires, “Action of an intense light beam on the refractive
index of liquids,” Comptes Rendus, Acad. Sci. Paris 258, 2039 (1964).

[8] R. W. Boyd, Nonlinear Optics (Academic Press, 2009), 3rd edn.

[9] J. D. Jackson, Classical Electrodynamics (Wiley, New York, 1996), 3rd

edn.

183