Wave and Oscillation PDF PDF

Waves and Oscillations
This page intentionally left blank

A Prelude to Quantum Mechanics
Walter Fox Smith
3
2010
3
Oxford University Press, Inc., publishes works that further
Oxford University’s objective of excellence
in research, scholarship, and education.
Oxford New York

Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Copyright © 2010 by Oxford University Press

Published by Oxford University Press, Inc.
198 Madison Avenue, New York, New York 10016
www.oup.com
Oxford is a registered trademark of Oxford University Press
All rights reserved. No part of this publication may be reproduced,

stored in a retrieval system, or transmitted, in any form or by any means,
electronic, mechanical, photocopying, recording, or otherwise,
without the prior permission of Oxford University Press.
Library of Congress Cataloging-in-Publication Data

Smith, Walter Fox.
Waves and oscillations : a prelude to quantum mechanics /
Walter Fox Smith.
p. cm.
Includes index.
ISBN 978-0-19-539349-1
1. Wave equation. 2. Mathematical physics. I. Title.
QC174.26.W28S55 2010
530.12’4–dc22 2009028586
9 8 7 6 5 4 3 2 1
Printed in the United States of America

on acid-free paper
This book is dedicated to my mother, Barbara Leavell Smith,
and to my wife, Marian McKenzie
Preface
To the student
I wrote this book because I was frustrated by the other textbooks on this subject.
Waves and oscillations are enormously important for current research, yet other books
don’t stress these connections. The ideas and techniques that you will learn from this
book are exactly what you need to be ready for a study of quantum mechanics. Every
physics professor understands this linkage, and yet other books fail to emphasize it,
and often use notations which are different from those used in quantum mechanics.
Other books make little effort to keep you engaged. I can’t teach you by myself, nor
can your professor; you have to learn, and to do this you must be active. In this book,
I’ve provided tools so that you can assess your learning as you go; these are described
immediately after the table of contents. Use them. Read with paper and pencil handy.
As a scientist, you know that only by understanding the assumptions made and the
details of the derivations can you have your own logical sense of how it all fits together
into a self-consistent whole. Visit this book’s website. There, you will find links to
current physics, chemistry, biology, and engineering research that is related to the
topics in each chapter, as well as lots of other stuff, some purely fun and some purely
educational (but most of it both). Hopefully, there will be a second edition of this book
in the future; if you have suggestions for it, please e-mail me: [email protected].
To the instructor
Please visit the website of this book. You’ll find materials in the website that will make
your life easier, including full solutions and important additional support materials
for the end-of-chapter problems, lecture notes which complement the text (including
additional conceptual questions, worked examples, applications to current research
and everyday life, animations, and figures), as well as custom-developed interactive
applets, video and audio recordings, and much more. The following sections can be
omitted without affecting comprehension of later material: 1.10, 1.12, 2.3–2.6, 3.5–3.6,
4.5, 4.7–4.8, 6.6–6.7, 8.6–8.7, 9.9, 9.11, 10.8–10.9, and Appendix A. If necessary, one
can skip all of chapter 6, except for the part of section 6.5 starting with the “Core
example” through the end of the section; however omitting the rest of chapter 6 means
viii Preface
that the students won’t be exposed to any matrix math or to the idea of an eigenvalue
equation. (They are exposed copiously to eigenvectors and eigenfunctions in other
chapters, but the word “eigenvalue” is used only in chapter 6.) If you have questions
or comments, please contact me: [email protected].
Acknowledgments
This book builds on the enormous efforts of my predecessors. Like any textbook author,
I have consulted many dozens of other works in developing my presentation. However,
three stand out as particularly helpful: Vibrations and Waves, by A. P. French (Norton
1971), The Physics of Vibrations and Waves, 6th Ed., by H. J. Pain (Wiley 2005), and
The Physics of Waves, by H. Georgi (Prentice-Hall, 1993).
I am deeply grateful to my physics colleague Peter J. Love, who cheerfully
answered endless questions from me, taught from draft versions of the book and
gave me essential feedback, and made key suggestions for several sections. I am also
most thankful to my other colleagues in physics who supported me in this effort and
answered my many questions: Jerry P. Gollub, SuzanneAmador-Kane, Lyle D. Roelofs,
and Stephon H. Alexander. I also received very valuable inputs from colleagues in
math, particularly Robert S. Manning, and chemistry, including Casey H. Londergan,
Alexander Norquist, and Joshua A. Schrier. I also thank Jeff Urbach of Georgetown
University and Juan R. Burciaga of Lafayette College who used draft versions of the
text in their courses, and provided helpful feedback.
I am profoundly thankful for the proof-reading efforts, and suggested edits and
end-of-chapter problems from Megan E. Bedell, Martin A. Blood-Forsythe, Alexander
D. Cahill, Wesley W. Chu, Donato R. Cianci, Eleanor M. Huber, Anna M. Klales, Anna
K. Pancoast, Daphne H. Paparis, Annie K. Preston, and Katherine L. Van Aken. Special
thanks are due to Andrew P. Sturner for his tireless efforts and suggestions, right up to
the last minute.
Finally, I am most deeply grateful to my family, for their support and encourage-
ment throughout the writing of this book. My children Grace, Charlie, and Tom checked
up on my progress every day, and suggested things in everyday life connected to
waves and oscillations. My good friend Michael K. McCutchan gave deep proofreading
and editing help, and support of all kinds throughout. Finally, words cannot express
my gratitude for the efforts of my wife, Marian McKenzie, who did almost all the
computerizing of figures, helped with editing, and provided the much-needed emotional
support. This book would never have been published without her encouragement.
Learning Tools Used in This Book
Throughout this text you will find a number of special tools which are designed to help
you understand the material more quickly and deeply. Please spend a few moments to
read about them now.
Concept test
This checks your understanding of the ideas in the preceding material.
Self-test
Similar to a concept test, but more quantitative. It will require a little work with pencil
and paper.
Core example
Unlike an ordinary example, these are not simply applications of the material just
presented, but rather are an integral part of the main presentation. There are some topics
that are much easier to understand when presented in terms of a specific example, rather
than in more abstract general terms.
Your turn
In these sections, you are asked to work through an important part of the main
presentation. Be sure to complete this work before reading further.
x Learning Tools Used in This Book
Concept and skill inventory
At the end of each chapter, you’ll find a list of the key ideas that you should understand
after reading the chapter, and also a list of the specific skills you should be ready to
practice.
Contents
Learning Tools Used in This Book ix
1. Simple Harmonic Motion 1

1. 1 Sinusoidal oscillations are everywhere 1
1. 2 The physics and mathematics behind simple sinusoidal motion 3
1. 3 Important parameters and adjustable constants of simple harmonic
motion 5
1. 4 Mass on a spring 8
1. 5 Electrical oscillators 10
1. 6 Review of Taylor series approximations 12
1. 7 Euler’s equation 13
1. 8 Review of complex numbers 14
1. 9 Complex exponential notation for oscillatory motion 16
1.10 The complex representation for AC circuits 18
1.11 Another important complex function: The quantum mechanical
wavefunction 24
1.12 Pure sinusoidal oscillations and uncertainty principles 26
Concept and skill inventory 29
Problems 31
2. Examples of Simple Harmonic Motion 39

2.1 Requirements for harmonic oscillation 39
2.2 Pendulums 40
2.3 Elastic deformations and Young’s modulus 42
2.4 Shear 47
2.5 Torsion and torsional oscillators 49
2.6 Bending and Cantilevers 52
Problems 58
3. Damped Oscillations 64
3.1 Damped mechanical oscillators 64
3.2 Damped electrical oscillators 68
xii Contents
3.3 Exponential decay of energy 69

3.4 The quality factor 70
3.5 Underdamped, overdamped, and critically damped behavior 72
3.6 Types of damping 74
Problems 77
4. Driven Oscillations and Resonance 84

4.1 Resonance 84
4.2 Effects of damping 91
4.3 Energy flow 95
4.4 Linear differential equations, the superposition principle for driven
systems, and the response to multiple drive forces 99
4.5 Transients 101
4.6 Electrical resonance 104
4.7 Other examples of resonance: MRI and other spectroscopies 107
4.8 Nonlinear oscillators and chaos 114
Problems 129
5. Symmetric Coupled Oscillators and Hilbert Space 137

5.1 Beats: An aside? 137
5.2 Two symmetric coupled oscillators: Equations of motion 139
5.3 Normal modes 142
5.4 Superposing normal modes 146
5.5 Normal mode analysis, and normal modes as an alternate
description of reality 149
5.6 Hilbert space and bra-ket notation 153
5.7 The analogy between coupled oscillators and molecular energy
levels 163
5.8 Nonzero initial velocities 165
5.9 Damped, driven coupled oscillators 166
Problems 170
6. Asymmetric Coupled Oscillators and the Eigenvalue Equation 179

6.1 Matrix math 179
6.2 Equations of motion and the eigenvalue equation 182
6.3 Procedure for solving the eigenvalue equation 186
6.4 Systems with more than two objects 191
6.5 Normal mode analysis for multi-object, asymmetrical systems 194
6.6 More matrix math 198
6.7 Orthogonality of normal modes, normal mode coordinates,
degeneracy, and scaling of Hilbert space for unequal masses 201
Contents xiii

Problems 210
7. String Theory 216

7.1 The beaded string 216
7.2 Standing wave guess: Boundary conditions quantize the
allowed frequencies 219
7.3 The highest possible frequency; connection to waves in a
crystalline solid 222
7.4 Normal mode analysis for the beaded string 226
7.5 Longitudinal oscillations 227
7.6 The continuous string 230
7.7 Normal mode analysis for continuous systems 231
7.8 k-space 234
Problems 236
8. Fourier Analysis 246

8.1 Introduction 246
8.2 The Fourier Expansion 247
8.3 Expansions using nonnormalized orthogonal basis functions 250
8.4 Finding the coefficients in the Fourier series expansion 251
8.5 Fourier Transforms and the meaning of negative frequency 254
8.6 The Discrete Fourier Transform (DFT) 258
8.7 Some applications of Fourier Analysis 265
Problems 268
9. Traveling Waves 280

9. 1 Introduction 280
9. 2 The wave equation 280
9. 3 Traveling sinusoidal waves 284
9. 4 The superposition principle for traveling waves 285
9. 5 Electromagnetic waves in vacuum 287
9. 6 Electromagnetic waves in matter 296
9. 7 Waves on transmission lines 301
9. 8 Sound waves 305
9. 9 Musical instruments based on tubes 314
9.10 Power carried by rope and electromagnetic waves; RMS
amplitudes 316
9.11 Intensity of sound waves; decibels 320
9.12 Dispersion relations and group velocity 323
Problems 334
xiv Contents
10. Waves at Interfaces 343

10.1 Reflections and the idea of boundary conditions 343
10.2 Transmitted waves 349
10.3 Characteristic impedances for mechanical systems 352
10.4 “Universal” expressions for transmission and reflection 356
10.5 Reflected and transmitted waves for transmission lines 359
10.6 Reflection and transmission for electromagnetic waves in
matter: Normal incidence 364
10.7 Reflection and transmission for sound waves, and summary of
isomorphisms 367
10.8 Snell’s Law 368
10.9 Total internal reflection and evanescent waves 371
Problems 379
Appendix A Group Velocity for an Arbitrary

Envelope Function 388
Index 393
1 Simple Harmonic Motion
All around us, sinusoidal waves astound us!

From “The Waves and Oscillations Syllabus Song,” by Walter F. Smith
1.1 Sinusoidal oscillations are everywhere
You are sitting on a chair, or a couch, or a bed, something that is more or less solid.
Therefore, every atom within it has a well-defined position. However, if you could look
very closely, you’d see that every one of those atoms right now is vibrating relative to
this assigned position. The hotter your chair the more violent the vibration, but even
if your chair were at absolute zero, every atom would still be vibrating! Of course, the
same is true for every atom in every solid object throughout the universe—right now,
each one of them is vibrating relative to its assigned or “equilibrium” position within
the solid.
The vibration of a particular one of these atoms might follow the pattern shown in
the top part of figure 1.1.1. The pattern appears complicated, but we will show in the
course of this book that it is really just a summation of simple sinusoids (as shown in
the lower part of the figure), each of which is associated with a “normal mode” of the
solid that contains the atoms. (Over the next several chapters, we’ll explore what the
term “normal mode” means.)
The complexity shown in the top part of the figure arises because the solid has
many “degrees of freedom”; every one of the atoms in the solid can move in three
dimensions, and each atom is affected by the motion of its neighbors. The approach of
physics, and it has been enormously successful in an astonishing variety of situations,
is to build up an understanding of complex systems through a thorough understanding
of simplified versions. For example, when studying trajectories, we begin with objects
falling straight down in a vacuum, and gradually build up to an understanding of three-
dimensional trajectories, including effects of air resistance and perhaps tumbling of
the object.
So, to understand the motion of the atom, we begin with systems that have only one
degree of freedom, that is, systems that can only move in one direction and moreover
don’t have neighbors that move. A good example is a tree branch. If you pull it straight
up and then let go, the resulting motion looks roughly as shown in figure 1.1.2. Again,
we see a sinusoidal motion, although in this case it is “damped,” meaning that over
1
2 Waves and Oscillations
Figure 1.1.1 Top: motion of an atom in a

solid. Bottom: Sine waves that, when
added together, create the waveform
shown in the top part.
time the motion decays away. Hold a pen or a pencil loosely at one end with your
thumb and forefinger, with the rest of the pencil hanging below. Push the bottom of the
pencil to one side, and then let go—the resulting motion looks similar to figure 1.1.2,
though this time the quantity being plotted is the angle of the pencil relative to
vertical.
In fact, if you take any object that is in an equilibrium position, displace it from
equilibrium, and then let go, you’ll get this same type of damped sinusoidal response,
as we will show quite easily in section 1.2. This type of oscillation is enormously
important, not only in the macroscopic motion of objects, machine parts, and so on but
also, perhaps surprisingly, in the performance of many electronic circuits, as well as in
Figure 1.1.2 Motion of a tree branch

when pulled up and then released.
Chapter 1 ■ Simple Harmonic Motion 3
the detailed understanding of the motions of atoms and molecules, and their interaction
with light.
So, sinusoidal motion really is all around us, and something which any scientist
must understand deeply. However, there is another perhaps even more important
reason to study oscillations and waves: the mathematical tools and intuition you
will develop during this study are exactly what you need for quantum mechanics!
This is not surprising, since much of quantum mechanics deals with the study of the
“wave function” which describes the wave nature of objects such as the electron.
However, the connection of the field of waves and oscillations to that of quantum
mechanics is much deeper, as you’ll appreciate later. For now, rest assured that
you are laying a very solid foundation for your later study of quantum mechanics,
which is the most important and exciting realm of current physics research and
application.
1.2 The physics and mathematics behind simple sinusoidal motion
To start our quantitative study, we follow the approach of physics and consider
the simplest possible system: one with no damping. This means that all the forces
acting on the object are conservative and so can be associated with a potential
energy.
A body in stable equilibrium is, by definition, at a local minimum of the potential
energy versus position curve, as shown in figure 1.2.1. For convenience, we choose
x = 0 at the equilibrium position. Except in pathological cases, the potential energy
function U(x) near x = 0 can be approximated by a parabola, as shown. We write this
parabolic or “harmonic” approximation in the form U(x) ≈ 21 kx 2 + const. for reasons
that will become apparent in the next sentence.
Figure 1.2.1 The Harmonic Approximation, valid for small vibrations around equilibrium.
dU
The force acting on the body can then be found using F = − = −kx. The
dx
relation
F = −kx (1.2.1)
is known as “Hooke’s Law,” after its discoverer Robert Hooke (1635–1703).1 The
quantity k is called the “spring constant.” To find the position of the body as a function
of time, x(t), we will follow a three-step procedure. We’ll use the same procedure
throughout the book, for progressively more complex systems. To save space, we
simply write x remembering that this is shorthand for the function x(t).
1. Write down Newton’s second law for each of the bodies involved.
In this case, there is only one body, so we have
⎫
d2 x⎬ 2
F = ma = m 2 ⇒ m d x = −kx . (1.2.2)
dt ⎭ dt 2
F = −kx
This is a “differential equation” or DEQ which simply means that it is an equation

that involves a derivative. (If you haven’t had a course in DEQs, don’t worry; we’ll
go through everything you need to know for this course and for a first course in
quantum mechanics.) This is called a “second order DEQ,” because it contains a second
derivative. The “solution” for this equation is a function x(t) for which the equation
holds true—in this case, a function for which, when you take two time derivatives
and multiply by m (as indicated on the left side of the equation), then you get back
the same function times −k (as indicated on the right side of the equation). This is
the solution that we are trying to find, since it tells us the position of the object at
all times. One important thing to know right away is that there is no general recipe
for finding the solution that works for all second-order DEQs. However, for many of
the most important such equations in physics, we can guess a solution based on our
intuition and then check to determine whether our guess is really right, as shown in the
following steps.
1. Some scholars feel that Robert Hooke is one of the most underappreciated figures in science. He
was the founder of microscopic biology (he coined the word “cell”), he discovered the red spot on
Jupiter and observed its rotation, he was the first to observe Brownian motion (150 years before
Brown), and discovered Uranus 108 years before the more-publicized discovery by Herschel.
Unfortunately, it seems that Hooke spread himself too thin, and never got around to publishing
many of his results. Hooke and Newton, though originally on friendly terms, later became fierce
rivals. It appears that Hooke conceptualized the inverse square law of gravity and the elliptical
motion of planets before Newton, and discussed this idea briefly with Newton. Newton (unlike
Hooke) was able to show quantitatively how the inverse square law predicts elliptical orbits,
and felt that Hooke was pushing for more recognition than he deserved in this very important
discovery. Some scholars feel that, when Newton became the president of the Royal Society (the
leading scientific organization of the time in England), he may intentionally have “buried” the
work of Hooke, but there is no hard evidence to support this.
d2 x
To save space, we write as ẍ. (Each dot represents a time derivative,2 so that
dt 2
dx
ẋ represents .) We rearrange equation (1.2.2) slightly to give
dt
k
ẍ = − x . (1.2.3)
m
This is called the “equation of motion.”
2. Using physical intuition, guess a possible solution.
Observation of a mass bouncing on a spring suggests that its motion may be sinusoidal.
The most general possible sinusoid can be expressed as
x = A cos (ωt + ϕ ) (1.2.4)
The values of the “adjustable constants” A and ϕ depend on the initial conditions, as
we will discuss later.
3. Plug the guess back into the system of DEQs to see if it is actually a solution,
and to determine whether there are any restrictions on the parameters that appear
in the guess.
In this case, the “system of DEQs” is the single equation (1.2.3). Before you look at the
next paragraph, plug the guess (1.2.4) into (1.2.3), verify that it is indeed a solution,
and find what the “parameter” ω must be in terms of k and m.
You should have found that

ω= k m (1.2.5)
So, we see that sinusoidal vibration, also known as “simple harmonic motion” or
SHM, is universally observed for vibrations that are small enough to use the Harmonic
Approximation shown in figure 1.2.1.
As described in section 1.3, ω equals 2π times the frequency of the motion and is
called the “angular frequency.”
1.3 Important parameters and adjustable constants of simple

harmonic motion
Figure 1.3.1 shows a graph of the SHM represented by equation (1.2.4). Any such
sinusoidal motion can be described with three quantities:
1. The amplitude A. As shown, the maximum value of x is A, and the minimum
value is −A.
2. The dot notation was invented by Isaac Newton. It is very convenient for us, because we have
to deal with time derivatives so frequently. However, it is generally felt that, because historical
English mathematicians continued to use this notation so long, they were held back relative
to their German counterparts, who used Gottfried Leibniz’s d/dt notation instead. (Leibniz’s
notation is more flexible, and we will use it where convenient.)
Figure 1.3.1 Simple harmonic motion

of period T and amplitude A.
2. The period T . This is the time between successive maxima, or equivalently

between successive minima. The period is the time needed for one complete
cycle, so that when the time t changes by T , the argument of the cosine in
x = A cos (ωt + ϕ ) must change by 2π . Therefore,
ω (t + T ) + ϕ = ωt + ϕ + 2π,
so that
T = 2π/ω (1.3.1)
(This equation is shown with a double border because we’ll be referring to it

so frequently. Equations shown this way are so very important that you will
find it helpful to begin memorizing them right away.) The frequency f is given
by 1/T , so that
ω = 2π f (1.3.2)
For this reason, ω is called the “angular frequency.” We will use it continually
for the rest of the text, so get accustomed to it now! We will encounter various
different angular frequencies later, so we give the special name ω0 to the angular
frequency of simple harmonic motion, that is,3

ω0 ≡ k /m (1.3.3)
(Note: the “0” subscript here does not indicate a connection to t = 0, but it is
universally used.)
3. The “initial phase” ϕ . The position at t = 0 is determined by a combination of
A and ϕ . It is easy to find the relation between these two “adjustable constants”
3. Physicists use the symbol “≡” to mean “is defined to be.”

on one hand and the initial position x0 and the initial velocity v0 on the other.
From equation (1.2.4): x = A cos (ωt + ϕ ) we obtain:

dx
x0 = A cos ϕ and v0 = = −ω0 A sin ϕ
dt t =0
Your turn: From these, you should now show that

2

v0 −v0
A = x02 + (1.3.4a) and ϕ = tan−1 . (1.3.4b)
ω0 ω0 x0
(We use the term “parameter” to refer to a quantity determined by the physical
properties of a system, such as mass, spring constant, or viscosity. Thus, ω0 is a
parameter. In contrast, we use “adjustable constant” to designate a quantity that is
determined by initial conditions. Thus, A and ϕ are adjustable constants.)
As mentioned earlier, the equation of motion (1.2.3) is a second-order DEQ,
because the highest derivative is of second order. It can be shown that the most general
solution to a second-order DEQ contains two (and no more than two) adjustable
constants.4 (We know that this must be true for our case, since we need to be able
to take into account (1) the initial position and (2) the initial velocity when writing
out a particular solution, therefore we need to be able to adjust two constants.)
So, we can be confident that equation (1.2.4): x = A cos (ωt + ϕ ) is the general
k
solution to equation (1.2.3): ẍ = − x. An example of a nongeneral solution would
m
be x = A sin ω0 t; you should verify that this satisfies equation (1.2.3). But this is the
same as equation (1.2.4), with the particular choice ϕ = −π/2.
√
Look again at equation (1.3.3): ω0 = k /m. There is something about it that is
absolutely astonishing. The angular frequency depends only on the spring constant and
the mass – it doesn’t depend on the amplitude! It would be very reasonable to expect
that, for a larger amplitude, it would take longer for the system to complete a cycle,
since the mass has to move through a larger distance. However, at larger amplitudes
the restoring force is larger and this provides exactly enough additional acceleration
to make the period (and so ω) constant. The fact that the frequency is independent of
amplitude is critical to many applications of oscillators, from grandfather clocks to
radios to microwave ovens to computers. Most of these do not actually have separate
masses and springs inside them, but instead have combinations of components which
are described by exactly analogous DEQs, and so exhibit exactly analogous behavior.
We’ll explore many of these in chapter 2, but we start now with the two most basic,
and most important, examples.
4. For the special case of a “linear” (meaning no terms such as x 2 or x ẋ), “homogeneous” (meaning
no constant term) DEQ, such as equation (1.2.3), this theorem is often phrased in the alternate
form, “The general solution of a linear, homogeneous second-order DEQ is the sum of two
independent solutions.” An example for our case would be x = A1 cos ω0 t + A2 sin ω0 t. However,

you can easily show (see problem 1.7) that this can be expressed in the form x = A cos ω0 t + ϕ ,

with A = A21 + A22 and ϕ = tan−1 −A2 /A1 .
Figure 1.3.2 Left: the tangent function. Right: Because the arctan function is multivalued, you
can add π to the result your calculator returns (shown by the curve which passes through the
origin), and sometimes you need to do this to get the physically correct answer.
Aside: The arctangent function
The arctan function, which appears in equation (1.3.4b), is a slippery devil, because it’s
multivalued, that is, tan−1 x is only defined up to an additive factor of π . For example,
tan−1 (−1) can equal either −π /4 or 3π /4, as shown in figure 1.3.2b.
Your calculator is programmed always to return the value between −π /2 and π /2,
but this is not always the correct answer for the particular situation. For example, consider
a case with A = 5 m and ω0 = 7 rad/s, with v0 = −24.75 m/s and x0 = −3.536 m. If you
use equation (1.3.4b) and plug in the numbers on your calculator, it will return ϕ = −π /4,
but this is wrong, because x = A cos(ω0 t − π /4) would mean x0 = A cos(−π /4) > 0 and
ẋ0 = −Aω0 sin(−π /4) > 0. To get the correct signs for x0 and ẋ0 you must add π to the
result from your calculator, giving ϕ = 3π /4. So, every time you use your calculator to
find tan−1 , you must think carefully about the result, and use other information from the
problem to determine whether you should add π to it to get the truly correct answer.
See problem 1.10.
1.4 Mass on a spring
Any system described by a DEQ of the form (1.2.2), mẍ = −kx, has a time evolution
of the form (1.2.4), x = A cos(ωt + ϕ ). The very simplest example is a mass that
feels only one force, from an attached ideal spring. It is difficult to eliminate the force
of gravity, so instead we often counteract it with a frictionless supporting surface, as
shown in figure 1.4.1a. The spring has an equilibrium length ℓ. However, if we measure
the position of the mass relative to its equilibrium position, as shown, then the force
exerted by the spring has a very simple form:
F = −kx . (1.4.1)
Figure 1.4.1 a: Mass on a frictionless surface. Important: The vertical line and horizontal
arrow marked “x” at the bottom of the figure show the definition of x: it is zero at the position
of the vertical line, and becomes positive in the direction of the arrow. In this example, this
means that when the mass moves to the right of the position shown, x is positive, whereas if
the mass moves to the left of the position shown then x is negative. We will use this
combination of line and arrow to define the displacement x throughout the book. b–d: Mass on
a vertical spring. The direction of positive x is downward.
As mentioned earlier, this is called Hooke’s law. It simply states that, when the mass
is to the right of its equilibrium position, so that x > 0, and the spring is stretched, the
spring pulls back to the left, that is, in the −x direction. If instead the mass is to the
left of its equilibrium position (x < 0) and the spring is compressed, then (as predicted
both by equation (1.4.1) and common sense), the spring pushes to the right, that is, in
the positive x direction.
Often, we happen not to have any frictionless surfaces handy, so it is more
convenient to suspend the mass vertically, as shown in figure 1.4.1 b–d. As a thought
experiment, we consider what would happen in the absence of gravity, as shown in
figure 1.4.1b. As before, we measure the position of the mass relative to its equilibrium
position (in the absence of gravity); we’ll call this x ′ , as a reminder that this is before
gravity is turned on. As shown, we define x ′ to be positive downward. The force of the
spring is just the same as before:
F = −kx ′ . (1.4.2)
Now, we turn on gravity, as shown in figure 1.4.1c. This causes the spring to stretch out
by an additional distance d, so that x ′ = d, and the spring force is F = −kx ′ = −kd.
(The spring force is negative, which means that it is upward.) At the new equilibrium
position, the net force on the mass must be zero, that is, the spring force must cancel
the force of gravity. Since we have defined the down direction to be positive, the force
of gravity is positive, so
mg
−kd + mg = 0 ⇔ d = . (1.4.3)
k
We now measure the position x of the mass relative to its new equilibrium position,
as shown in figure 1.4.1c. In figure 1.4.1d, an additional downward force is applied,
stretching the spring further and so creating a positive x. We see that
x′ = x + d
The total force on the mass is

FTot = −kx ′ + mg = −k(x + d) + mg = −kx − kd + mg.
Substituting for d from equation (1.4.3) gives
mg
FTot = −kx − k + mg = −kx .
k
Thus, as long as we measure relative to the new equilibrium position, the combined
effects of the spring and gravity give a total force which follows Hooke’s Law!
So, for either the situation of figure 1.4.1a or 1.4.1c and d, we have a total force
FTot = −kx, so we can use the result of equation (1.2.2), that is,

F = ma = mẍ
⇒ mẍ = −kx
F = −kx
with the solution we found in section 1.2, x = A cos(ω0 t + ϕ ).
1.5 Electrical oscillators
Consider the circuit shown in figure 1.5.1. The capacitor, designated C, stores electrical
charge and potential energy, in much the same way that a spring can store potential
energy. The capacitor always has equal and opposite charge q on its two plates. For
example, at some instant in time it might have a charge +1.2 nC on the top plate
and −1.2 nC on the bottom plate. At this instant, q = +1.2 nC. The capacitance C is
defined as the ratio of the charge to the voltage across the capacitor:
q q
C≡ ⇔ Vc = . (1.5.1)
Vc C
The inductor, designated L, consists of a number of loops of wire. As you’ll recall
from a previous course, when electrical current I flows through the loops, it creates
a magnetic field B, with associated magnetic flux φB linking through the loops. The
inductance is defined as
φ
L ≡ B. (1.5.2)
I
Faraday’s law tells us that there is an emf across the inductor given by
ε = −φ̇B = −L İ . (1.5.3)
Figure 1.5.1 Electrical oscillator.

Recall that the current is defined to be a time rate of change of charge. We define a
positive current to be one that flows clockwise in the circuit, as shown in figure 1.5.1.
We also define q to be positive when the upper plate is positive, as shown. Current is
the time derivative of charge, but, with our sign definitions, a positive I decreases the
charge on the capacitor. Therefore,
I = −q̇. (1.5.4)
Combining this with equation (1.5.3) gives
ε = +L q̈. (1.5.5)
This is the voltage across the inductor, so
VL = L q̈. (1.5.6)
Next, we will apply Kirchhoff’s loop rule, which says that when you go around the
loop, the voltage changes must add up to zero:
q
Vc + VL =0 ⇒ + L q̈ = 0 ⇔
C
1
L q̈ = − q (1.5.7)
C
This is isomorphic to equation (1.2.2), mẍ = −kx, meaning that it is exactly the same,
except with different symbols. Right away, then, we know that the solution, which
must be isomorphic to x = A cos(ω0 t + ϕ ), is q = A cos(ω0 t + ϕ ). The isomorphism
is summarized in table 1.5.1.
Your turn (answer below5 ): Using the isomorphism, deduce what the angular
frequency ω0 is for the electrical oscllator.
Electrical oscillators are tremendously important in electronic circuits, from radio

tuners to the clocks that regulate the speed of computers.
Table 1.5.1. Isomorphism between mechanical and electrical oscillators
Mass and spring Electrical oscillator
Position relative to equilibrium x Charge q on capacitor

Mass m Inductance L
Spring constant k Inverse capacitance 1/C
5. Answer to self-test: Comparing equations (1.2.2)

and (1.5.7), we seethat m gets replaced by L,
k 1
while k gets replaced by 1/C. Therefore, ω0 = becomes ω0 = .
m LC
1.6 Review of Taylor series approximations
To move forward efficiently, we must take a little time now to go over two important
mathematical techniques. Later in this chapter, we’ll show that oscillatory motion can
be expressed in a more elegant way by using complex exponential functions. However,
to develop those, we’ll need to use Taylor series, which we review in this section.
Much of the creative effort in physics is devoted to making reasonable approxima-
tions so that we can study the most important behaviors of complex systems without
getting bogged down in a morass of hundreds of complex equations. The most important
approximation tool is the Taylor series approximation.
The goal is to find the value of a function f (x) at the position x = x0 + a, if we
are given complete information about the function at the nearby point x0 . The simplest
approximation, shown by the dot labeled “zeroth order approximation” in figure (1.6.1),
is simply to say that f (x0 + a) ≈ f (x0 ). We can get a better approximation (shown in
gray) by using our knowledge of the slope of f (x) at the point x0 . We write this slope
df
as , which is read as “the derivative of f with respect to x, evaluated at x0 .” The
dx x0
slope equals the “rise” over the “run,” so by multiplying it by the run (i.e., by a), we get
the rise, and by adding this to the initial value f (x0 ), we get a closer approximation to
the true value f (x0 + a); in doing so, we approximate the function as a straight line. We
can do even better by approximating f (x) as a parabola, as shown by the dashed curve.
If we wanted to get even more accurate, we could use a third-order approximation:

2 d2 f 3 d3 f

df a a
f x0 + a ∼ (1.6.1)

= f x0 + a + + .
dx x0 2! dx 2 3! dx 3
x0 x0
You can see the pattern. Assuming a is small, each additional correction term
gets smaller and smaller, so that usually we don’t need to go beyond a second-
order approximation. (In fact, most often a first-order approximation will suffice.)
The complete version would be
∞ n n
df an dn f a d f
f x0 + a = f x0 + a + ··· + + ··· = . (1.6.2)
dx x0
n! dx x0
n n! dx n x0
n=1
Figure 1.6.1 Graphical illustration of the Taylor series approximation.

As an example, let’s find the Taylor series for the sine function:
df d2 f d3 f d4 f
f (θ ) ≡ sin θ ⇒ = cos θ, = − sin θ, = − cos θ, = sin θ, . . .
dθ dθ 2 dθ 3 dθ 4
Plugging this into equation (1.6.2), using θ as the variable instead of x, and expanding
around θ0 = 0 gives
θ2 θ3 θ4 θ5
sin θ = sin 0 + θ cos 0 + (− sin 0) + (− cos 0) + (sin 0) + (cos 0) + · · ·
2! 3! 4! 5!
θ3 θ5
⇒ sin θ = θ − + − ··· (1.6.3)
3! 5!
From this, we can see why the approximation
sin θ ∼
= θ (θ in radians) (1.6.4)
works so well for small θ : there is no second-order correction term – the next
correction term is third order. (In fact, as you can show yourself on your calculator,
this approximation works pretty well up to about θ = 0.4 radians = 23◦ .)
Your turn: Show that

θ2 θ4
cos θ = 1 − + − ··· (1.6.5)
2! 4!
1.7 Euler’s equation
We will see in section 1.9 that there is a different way of expressing the solution
for simple harmonic motion, x = A cos(ωt + ϕ ), one which will become much more
convenient when we begin treating more complicated systems. We will make use of
Euler’s equation:
eiθ = cos θ + i sin θ. (1.7.1)
√
Here, i ≡ −1. The proof of this statement, and also the understanding of what it
means to have a complex number as an exponent, comes through consideration of
series expansions. Using the Taylor expansions we just derived for cos and sin, we can
express the right side of this as
θ2 θ3 θ4
cos θ + i sin θ = 1 + iθ − −i + + ··· (1.7.2)
2! 3! 4!
Now, we express the left side of equation (1.7.1) using a Taylor series, again
expanding around θ0 = 0:
θ 2 2 i0 θ 3 3 i0 θ 4 4 i0
eiθ = ei0 + θ iei0 + i e + i e + i e + ···
2! 3! 4!
θ2 θ3 θ4
= 1 + iθ − −i + + ···
2! 3! 4!
This is just the same as equation (1.7.2), which proves equation (1.7.1). This was first
demonstrated by Leonhard Euler6 in 1748. You will use Euler’s equation every day
for the rest of your life- , so you are encouraged to commit it to memory.
1.8 Review of complex numbers
Let us briefly review complex numbers. It is helpful to use the “complex plane,” as
shown in figure 1.8.1a in which the vertical axis is used for the imaginary part of a
number and the horizontal axis for the real part. A complex number z can be represented
Figure 1.8.1 a: The complex plane.

b: Multiplication by eiα is equivalent to
rotating counterclockwise by α in the
complex plane.
6. Euler earned his Master’s degree from the University of Basel at the age of 16. During his life, he
published over 900 works. He is responsible for many of our mathematical notations, including
f (x) to denote a function,
x to denote a difference, and e for the base of the natural logarithm.
as a vector in this plane, and we can write the “Cartesian representation” for the number
as z = a + ib. We see √ from the diagram that the length or magnitude of the vector is
given by A = |z| = a2 + b2 . Simple trigonometry then provides that
a = A cos θ and b = A sin θ, so that
z = a + ib = A cos θ + i(A sin θ ) = Aeiθ
(using Euler’s equation) and also
θ = tan−1 (b/a).
Your turn: If you’ve not already done so, read the aside about the arctan function in
section 1.3. Then, explain why a more complete version of the above equation is

−1
0 if a > 0
θ = tan b a +
π if a < 0
Let us collect all these useful relations into a single box:
z = a + ib “Cartesian representation”
– or –
z = Aeiθ “Polar representation,”
where
0 if a > 0
A = |z | = a 2 + b2 and θ = tan−1 b a +
π if a < 0
What happens in the complex plane when we multiply z by eiα ?
eiα z = eiα Aeiθ = Aei(θ+α) .
This is a vector in the complex plane which still has length A, but has been rotated
counterclockwise by the angle α so that it now points in the direction given by θ + α ,
as shown in figure 1.8.1b. Thus,
Multiplying a number by eiα is equivalent to rotating its vector counterclockwise by α

in the complex plane.
Often we need to take the “complex conjugate” of a complex number:
To form the complex conjugate, simply replace every instance of i by −i.
For example, the complex conjugate of a+ib is just a−ib. We denote the complex
conjugate with a star: the complex conjugate of z is z∗ . As another example, if z = eiθ ,
then z∗ = e−iθ . The complex conjugate is often used to calculate the magnitude of a
complex number. This is perhaps the easiest to see with a number expressed in polar
form: if z = Aeiθ (with A real), then z∗ = Ae−iθ , so that z∗ z = A2 = |z|2 . So, in general,
we have
|z |2 = z ∗ z .
Finally, we introduce the notation for the real and imaginary parts of a complex number:
If z = a + ib, then Re z = a and Im z = b.
(Note that the i is not included in Im z.)

Self-test (answer below7 ): Show that, for any complex number z, Re z = Re z ∗ .
1.9 Complex exponential notation for oscillatory motion
Finally, we are ready to apply these ideas to the simple harmonic oscillator. We can
very easily rewrite the solution using complex exponential notation:

x = A cos(ω0 t + ϕ ) = Re Aei(ω0 t +ϕ ) = Re eiω0 t Aeiϕ . (1.9.1)
Written this way, we can see that the complex plane vector that represents the system
has length A and at t = 0 points in the direction given by the angle ϕ . This vector
is then multiplied by eiω0 t , that is, it is rotated counterclockwise by the angle ω0 t, as
shown in figure 1.9.1. Since this angle increases in time, the vector rotates around and
around the origin. The “angular velocity” is the time derivative of this angle, that is,
d
ω t + ϕ = ω0 . The actual position of the oscillator is given by the real part of the
dt 0
vector, that is, the projection onto the horizontal axis. As the vector rotates in uniform
circular motion, this projection changes sinusoidally. It is convenient to define
z ≡ Aei(ω0 t +ϕ ) (1.9.2)
so that x = Re z.
Figure 1.9.1 Complex plane representation

of SHM.
7. Answer to self-test: Express z in Cartesian form: z = a + ib. The real part of z is just a. The
complex conjugate is z∗ = a − ib, and the real part of this is also a.
This method of portraying the motion brings out the physical significance of A
and ϕ more clearly. From figure 1.9.1, we can see that
x
ϕ = cos−1 0 , (1.9.3)
A
where x0 is the initial position. It is also clearer, perhaps, that A is related to the total
energy of the system. From the discussion surrounding figure 1.2.1, we know that the
potential energy of a harmonic oscillator is given by
1 2
U (x) = kx + const.
2
Usually, it is convenient to choose the constant so that U(x) = 0 at the equilibrium
position x = 0, so that
1 2
U (x) = kx . (1.9.4)
2
When x = A, the oscillator is at its maximum displacement, and so is momentarily at
rest. Therefore, all the energy is in the form of potential energy, so that

1 2E
E = kA2 ⇔ A = , (1.9.5)
2 k
where E is the total energy.
It is also easy to show the relationships between position, velocity, and acceleration
using this complex plane picture. Take a moment now to convince yourself that the
operations of taking time derivatives and taking the real part of a quantity “commute,”
that is, the order doesn’t matter, that is,
dz d Re z
Re = .
dt dt
Therefore, we can write
ẋ = Re ż and ẍ = Re z̈.
Plugging in for z from equation (1.9.2) gives
ż = iω0 z and z̈ = −ω02 z. (1.9.6)
π
Using Euler’s equation, we see that ei 2 = i, so that multiplication by i rotates a
complex plane vector counterclockwise by π/2. Equation (1.9.6) thus says that the
complex plane vector representing the velocity, ż, is rotated by a constant angle π /2
“ahead” of the position vector z (and is scaled by the factor ω0 ). Similarly, since
−1 = i · i, multiplication by −1 rotates a complex plane vector through 2 · (π/2) = π .
Equation (1.9.6) thus says that z̈ is always an angle π ahead of z (and is scaled by
ω02 ). These relationships are shown in figure 1.9.2; bear in mind that the position,
velocity, and acceleration have different units, so the relative lengths of the vectors in
each picture are not meaningful. As shown in the upper left part of the figure, for the
important special case of zero initial velocity, A = x0 . We will use this result again
later.
This is a good time to point out that, although taking the real part does commute
with taking the derivative, addition, and multiplication by a real number, taking the
real part does not commute with multiplication by a complex number. For example,
Figure 1.9.2 Phase relationships for SHM. Top left: initial velocity = 0 shown for t = 0 . Top
right: initial velocity = 0 shown for t > 0 . Bottom: Similar pictures for initial velocity <0.

if z1 = A1 eiϕ1 and iϕ2 i(ϕ1 +ϕ2 ) = A A cos ϕ + ϕ ,
z2 = A2 e , then Re z1 z2 = Re A1 A2 e 1 2 1 2
whereas Re z1 Re z2 = A1 A2 cos ϕ1 cos ϕ2 . It is especially important to bear this
in
2
1 1 1

mind when calculating energies; for example, KE = 2 mẋ = 2 m (Re ż) = 2 mRe ż2 .
2
1.10 The complex representation for AC circuits
In section 1.5, we discussed one type of electrical oscillator. However, there are many
other circuit examples in which the voltage and current vary sinusoidally in time.
Such circuits are essential to the operation of virtually all analog (that is, nondigital)
electronics, and the concepts involved in analyzing them are critical to the detailed
understanding of all circuits. It is convenient to use a complex representation for
the currents and voltages in AC circuits, and this approach is used essentially by all
scientists who work with circuits and essentially by all electrical engineers. So, this
section will provide a good chance to exercise the skills involving complex numbers
that we have just reviewed.
The simplest possibleAC circuit is shown in figure 1.10.1a. The signal generator on
the left of the circuit is a device that produces a voltage difference V0 cos ωt between its
two terminals. In this circuit, it is connected to a resistor. Since only voltage differences
are physically important, we can define V ≡ 0 at any point in the circuit. For many
actual circuits, the V ≡ 0 point is ground (literally the voltage of the dirt under your
building). For the rest of this section, we’ll use this convention; in the circuit shown, we
Figure 1.10.1 a: A signal generator (denoted by a sine wave

inside a circle) connected to a resistor. b: Same circuit,
drawn in the more conventional way using the ground
symbol. c: Complex representation for an AC voltage.
have grounded the lower terminal of the signal generator. Again, this does not affect the
operation of the circuit at all; it merely fixes the reference point with respect to which
voltages are measured. The circuit can then be redrawn as shown in figure 1.10.1b; all
points in the circuit with the ground symbol are connected together.
V
For a resistor, V = IR ⇔ I = V /R, so for this circuit I = 0 cos ωt.
R
Now, let’s apply the complex representation. The voltage is
V = V0 cos ωt = Re Ṽ , where Ṽ = V0 eiωt
is the complex version of the voltage, as shown in figure 1.10.1c.8 Similarly, the
current is
V V
I = 0 cos ωt = Re Ĩ , where Ĩ = 0 eiωt .
R R
Comparing this with the definition of Ṽ , we can write the complex version of Ohm’s
Law,
Ṽ = ĨZR ,
where ZR = R is the “impedance” of the resistor. The impedance is a generalized
version of the resistance; we will see below that it can be complex, so that we could
write it as |Z | eiϕ . Then, the general complex version of Ohm’s law would be
Ṽ = Ĩ |Z | eiϕ .
Since multiplying a complex number by the factor eiϕ rotates the complex plane vector
by angle ϕ , we see that the complex phase ϕ of the impedance is the phase difference
between the current and the voltage. For a resistor, Z is real, so that the current and the
voltage are in phase, but we will see that, for inductors and capacitors, Z is imaginary,
8. We use the tilde (∼) above a symbol to indicate “complex version of,” so that Ṽ is the complex
version of V . Unfortunately, in many texts, Ṽ is simply written as V , and one must remember
that, to find the actual voltage, one must take the real part.
Figure 1.10.2 a: A signal generator connected to a

capacitor. b: An RC low-pass filter.
so that the current and voltage are not in phase, meaning that the sinusoidally varying
voltage across a capacitor or inductor reaches a peak at a different time than the
sinusoidally varying current flowing through it. (See problems 1.19, 1.22, and 1.23
for more about these phase differences.) Note that the use of Z as a symbol for the
impedance is meant to tip you off that it may be a complex quantity, so it is conventional
not to include the tilde over the Z.
So far, this is not terribly exciting. Things get more interesting when we introduce
a capacitor, as shown in figure 1.10.2a. We again use a signal generator to apply a
voltage V0 cos ωt across the capacitor. This creates a time-dependent charge Q(t) on
the capacitor. To find the current, we use
d/dt dQ dV
Q = CV −−→ =C .
dt dt
dQ
Since I = , we have
dt
dV d
I=C =C V cos ωt = −CV0 ω sin ωt .
dt dt 0
Now, let’s apply the complex representation. As before, we have
V = V0 cos ωt = Re Ṽ , where Ṽ = V0 eiωt .
The current is
I = −CV0 ω sin ωt = Re Ĩ , where Ĩ = iCV0 ωeiωt .
the complex version of Ohm’s Law in this case is
Ṽ = ĨZC ,
where ZC is the impedance of the capacitor.
Your turn (answer below9 ): What is ZC in terms of C and ω?
Ṽ V0 eiωt 1
9. ZC = = = .
Ĩ iCV0 ωeiωt iω C
To really see the power of this approach, we need to consider a more complicated
circuit, such as that shown in figure 1.10.2b. No current is allowed to flow out of the
circuit at the point labeled VOUT ; instead, this is a place at which we will later calculate
the voltage. Again, we apply a voltage V0 cos ωt, this time to a series combination of a
resistor and capacitor. What is the resulting current? It is related to the voltage across
the resistor by VR = IR, which is the real part of
ṼR = ĨZR . (1.10.1)
The total voltage across the series combination is the sum of the individual voltages:
VIN = VR + VC , which is the real part of

ṼIN = ṼR + ṼC ⇔ ṼR = ṼIN − ṼC ⇒ ṼR = ṼIN − ĨZC .
Substituting for ṼR from equation (1.10.1) gives

ĨZR = ṼIN − ĨZC ⇔ ṼIN = Ĩ ZR + ZC .
So, the total impedance is ZTOT = ZR + ZC , meaning that impedances in series add,
just as one would expect from the behavior of resistors, i.e.,
Zseries = Z1 + Z2 . (1.10.2)
In problem 1.20, you can show that impedances in parallel combine as one would
expect, that is,
1 1 1
= + . (1.10.3)
Zparallel Z1 Z2
Core example: The low-pass filter. The circuit in figure 1.10.2b is one of the most useful
and common elements in analog circuitry. To understand why, let’s calculate VOUT :
Ṽ ZC
ṼOUT = ṼC = ĨZC = IN ZC = ṼIN .
ZTOT ZR + ZC
You may recognize this as the equation for a voltage divider; the fraction of the total
voltage ṼIN that appears across the capacitor equals the fraction of the total impedance
that is due to the capacitor. Often, we are only interested in the amplitude of the output
voltage expressed as a fraction of the amplitude of the input voltage. Recall that the
amplitude of an oscillating quantity is the magnitude of the complex number that
represents it, as shown in figure 1.10.1c. Therefore,

amplitude of VOUT ṼOUT
= .
amplitude of VIN
ṼIN

A |A|
You can show in problem 1.15 that, for any two complex numbers A and B, = .
B |B|
So,

amplitude of VOUT VOUT ZC Z
C
= = = .
amplitude of VIN VIN ZR + ZC ZR + ZC
continued
Referring to section 1.8, we see that

1
= 1 and Z + Z = R + 1 = R2 + 1 .

Z =
C R C
iωC ωC iωC ω2 C 2
Therefore,
1
amplitude of VOUT 1 1
= ωC = = ,
amplitude of VIN 1 2 2 2
ω R C +1

ω 2
R2 + 2 2 1+
ω C ω LO
where
1
ωLO ≡ .
RC
The dependence of this ratio on the frequency of VIN is shown in figure 1.10.3. You can
see from these graphs, especially the log–log graph on the bottom, why this circuit is
called a “low-pass filter.” If the angular frequency of VIN is well below ωLO , then the
amplitude at the output is the same as the amplitude at the input, whereas if the angular
frequency of VIN is well above ωLO , then the output is dramatically smaller than the input.
Figure 1.10.3 Amplitude ratio for the low-pass filter shown in figure 1.10.2b. Top: linear
axes. Bottom: same data with log–log axes.
We can understand this behavior qualitatively, by remembering that the fraction

of the input voltage that appears across the capacitor equals the fraction of the total
impedance due to the capacitor. At low frequencies, the impedance of the capacitor is
large, so the fraction is nearly equal to 1, whereas at high frequencies the impedance of
the capacitor is low, so the fraction approaches zero.
22
Finally, let us
consider
an inductor. We use the signal generator to apply a voltage
V0 cos ωt = Re V0 eiωt across the inductor. The magnitude of the voltage across the
inductor is
dI dI V
VL = L ⇔ = L
dt dt L

1 1 V
⇒I= VLdt = V0 cos ωt dt = 0 sin ωt + constant.
L L Lω
If the voltage amplitude V0 is zero, the current must be zero, so the constant must be
zero. So,
V0 V
I= sin ωt = Re Ĩ , where Ĩ = −i 0 eiωt ,
Lω Lω

iω t
since sin ωt = Re −ie . Thus, the impedance of the inductor is
Ṽ V0 eiωt
ZL = = = iω L .
Ĩ V
−i 0 eiωt
Lω
Summarizing:
1
ZR = R ZC = ZL = iωL . (1.10.4)
iωC
Concept test (answer below10 ): Does the circuit shown in figure 1.10.4 function
amplitude of VOUT
as a low pass filter (for which = 1 for low frequencies and 0
amplitude of VIN
for high frequencies), or instead does it function as a high-pass filter (for which
amplitude of V OUT
= 1 for high frequencies and 0 for low frequencies)? You
amplitude of V IN
should be able to answer this question using qualitative reasoning, combined with
equation (1.10.4).
Figure 1.10.4 An RL filter.
10. The output voltage divided by the input voltage equals the impedance of the resistor divided by
the total impedance. The impedance of the inductor is low for low frequency, meaning that most
of the total impedance is in the resistor, so VOUT ≈ VIN . On the other hand, at high frequencies,
the impedance of the inductor is high, so the resistor represents a small fraction of the total
impedance, and VOUT ≈ 0. So, this is a low-pass filter.
1.11 Another important complex function: The quantum mechanical

wavefunction
Our topics of study are waves and oscillations. However, one of the important reasons
for mastering the concepts and techniques associated with these topics is that they
apply directly to quantum mechanics. Therefore, although we will not study quantum
mechanics per se in this book, we will point out some of the connections as we come
to them.
As you may know, small particles such as electrons display many wave-like
properties. The wave nature of such a particle is described by the “wavefunction”
(the Greek capital letter psi). The wavefunction depends on both position and time,
so we could write it as (x , t), though usually we will simply write . One of the
most remarkable things about quantum mechanics is that is inherently complex! All
the information that can be known about the particle is contained in , and in a later
course you will learn how to extract from quantities such as the momentum, angular
momentum, and energy. One aspect of is relatively easy to understand: | (x , t)|2
is called the “probability density,” and is proportional to the probability of finding
the particle near the position x. For example, for the probability density shown in
figure 1.11.1, the particle is likely to be found near x = −3 or 1 nm (1 nm = 10−9 m).
This is called a “delocalized” wavefunction, since the particle might be found in two
different places. You’ll learn more about this in a later course.
Another example is that of an electron traveling at constant speed through a
vacuum; this is called a “free electron.” The wavefunction in this case is (x , t) =
ψ0 e−iωt eikx , where ψ0 (Greek lower case psi, with a naught subscript) is a constant,
ω
k = is called the “wavenumber,” and vp is the speed of the wave.11 This wavefunction
vp
has an oscillatory dependence both on t and on x:
(x , t) = ψ0 e−iωt eikx = ψ0 (cos ωt − i sin ωt) eikx = ψ0 e−iωt (cos kx + i sin kx) .
Wavefunction for a free electron
This function is plotted in figure 1.11.2a and b.
Figure 1.11.1 A delocalized wavefunction.
11. The wavenumber k is also equal to 2π /λ, where λ is the “wavelength,” that is, the repeat
interval along the x-axis.
Figure 1.11.2 Real and imaginary parts of

the wavefunction for an electron moving
with a constant velocity in a vacuum (i.e., a
“free electron”). a: dependence on time of
the wavefunction at x = 0. b: dependence on
position of the wavefunction at t = 0. c: The
probability density is given by ||2 ,
showing that a free electron is completely
delocalized.
Self-test (answer below12 ): Show for (x , t) = ψ0 e−iωt eikx that the probability density
2
||2 is equal to ψ0 .
The result of this self-test is remarkable—there is no space or time dependence in

the probability density ||2 for this case! This means that the particle is completely
delocalized—it is just as likely to be found at x = −100 nm as at x = +100 km, or
at any other point, as shown in figure 1.11.2c. This waveform is an idealized version,
since no real particle could be so infinitely spread out. However, real particles can be
highly delocalized, even over macroscopic distances.
We will encounter several more times in this text.
2
12. Answer to self-test: ||2 = ∗ = ψ0∗ eiωt e−ikx ψ0 e−iωt eikx = ψ0∗ ψ0 = ψ0
1.12 Pure sinusoidal oscillations and uncertainty principles
In real life, oscillations usually only last for a finite period of time. For example, we
might strike a piano key, hold it down for a short length of time (allowing the string
inside the piano to vibrate), and then release the key (which immediately stops the
vibration). If we strike the key at t1 and release it at t2 , the resulting vibration of a
particular point on the string might look as shown in figure 1.12.1.
It is important to realize that the waveform of figure 1.12.1 is not a pure sinusoidal
oscillation, since it is not of the form (1.2.4): x = A cos(ωt + ϕ ). Equation (1.2.4)
describes an oscillation which goes on infinitely in time, stretching back in time to
t → −∞, and forward in time to t → +∞. This is not merely a semantic distinction.
We will see in our study of Fourier Analysis (chapter 8) that a function of the form
shown in figure 1.12.1 can be created by adding together a very large number of pure
sinusoids, only one of which is at the angular frequency ω.
This means that, for a function such as that shown in figure 1.12.1, the angular
frequency is not “well-defined,” that is, the function cannot be characterized by a
single angular frequency. (If it could, we could write it as x = A cos(ωt + ϕ ).) We
don’t have to wait for chapter 8 to see this – we can develop a qualitative argument
now that shows it. Imagine that we try to determine the frequency of the waveform
shown in figure 1.12.1 by counting the number of times it crosses zero. (This, in fact, is
how frequencies are determined in most experiments, which usually rely on electronic
“frequency counters.”) There are two zero-crossings per period. Therefore, if we define
N ≡ number of zero crossings and

t ≡ t2 − t1 ,
Then,

t 1 N
T= ⇒f = = .
(N /2) T 2
t
However, in any real signal, the beginning and the end are not defined with absolute
crispness – there is always a question about exactly where we should begin counting
the zero-crossings and where we should stop. We’re only making a rough argument
here, so let’s say that there’s an uncertainty of 1 in the value of N. In other words, we
N N +1
can’t really be sure whether we should write f = or f = . Thus, there is an
2
t 2
t
uncertainty in f :
N +1 N 1

f = − = .
2
t 2
t 2
t
Figure 1.12.1 Motion of one point on a piano string.

We have used simple arguments about a simple way of determining the frequency.
However, more sophisticated arguments give the same qualitative results: the shorter
the time interval
t, the greater the uncertainty in the frequency
f . The more
sophisticated arguments, which use a more careful definition of
t and
f , show
1
that
f is always at least as big as . This is for an ideal circumstance, where the
4π
t
profile of the wave has an ideal shape. Of course, it is always possible to have a wave
with a less ideal profile, or to be sloppy in doing the measurements, so that in general
1

f ≥ ⇔
4π
t
1

f
t ≥ . (1.12.1)
4π
Frequency–time uncertainty relationship
(valid for all types of waves and oscillations)
We can apply identical arguments to something (such as a wave on the surface of the
ocean) that oscillates as a function of x instead of as a function of t. A pure sinusoidal
oscillation as a function of t can be written
y (t) = A cos(ωt + ϕ ) ⇒

2π
y (t) = A cos t+ϕ , (1.12.2)
T
where T is the repeat interval in time (the period). (Here, we use y (t) for the oscillating
function, to avoid confusion with the variable x which is used for the position in the
next equation. However, y (t) can stand for exactly the same types of things as we’ve
previously discussed, such as the position of an oscillating tree branch.) Similarly, we
can write a pure sinusoidal oscillation as a function of x as

2π
y (x) = A cos x+ϕ , (1.12.3)
λ
where λ is the repeat interval in space (the “wavelength”). We see that equations (1.12.2)
and (1.12.3) are isomorphic. (Recall that this means that they are exactly the same,
except that different symbols appear.) In the isomorphism t becomes x and T becomes
λ. We can use exactly the same arguments that lead us to equation (1.12.1), but apply
them to oscillations as a function of x instead. Since f = 1/T , this gives

1 1

x ≥ . (1.12.4)
λ 4π
Uncertainty relationship between inverse wavelength and position

(valid for all types of waves and oscillations)
The uncertainty relations (1.12.1) and (1.12.4) apply to any type of oscillation or wave,
whether it is a simple harmonic oscillator or the quantum mechanical wavefunction.
One of the fundamental relations of quantum mechanics relates the energy E
of the particle to the frequency of oscillation of the wavefunction. (e.g., for a free
ω
electron, the wavefunction is (x , t) = ψ0 e−iωt eikx , and the frequency is f = .)
2π
This fundamental relation is
E = hf , (1.12.5)
where h = 6.6260690 × 10−34 J s is called “Planck’s constant.” As you will learn in

a later course, this relation comes from experimental results, and cannot be derived
from arguments based on classical physics. As you read through this text, bear in mind
that essentially everything we say about classical waves and oscillators applies to
as well. In particular, essentially everything we say about the frequency of a classical
oscillator applies to the frequency of as well, and that, by equation (1.12.5), this
frequency is proportional to the energy of the particle.
Combining equation (1.12.5) with (1.12.1) gives

E 1

t ≥ ⇒
h 4π
h

E
t ≥ . (1.12.6)
4π
This is called the “energy–time uncertainty principle.” It states that it is not possible to
exactly determine the energy of a system if one can only observe it for a short time. If
the energy of a system cannot be precisely determined on short time scales, then this
“energy–time uncertainty principle” requires that we modify our understanding of the
conservation of energy to allow for “quantum fluctuations.” For example, it is possible
for pairs of particles (one normal matter and one antimatter) to be spontaneously created
out of the vacuum, which requires a tremendous amount of energy (on a particle scale).
These particles can exist only for a fleeting moment, and then annihilate with each
other, releasing the energy they had “borrowed” before anyone could notice it was
missing! Perhaps surprisingly, the effects of these “virtual particles” can be observed
experimentally, for example through the Casimir effect.13
The other fundamental relation of quantum mechanics is
h
p= , (1.12.7)
λ
where p is the momentum of the particle. As with equation (1.12.5), this relation
comes from experimental results, and cannot be derived from classical principles.
13. In the region between two metal plates, the density of virtual particles is lower than in the region
outside the plates, resulting in a force that pushes the plates together. For more information, see
Precision Measurement of the Casimir Force from 0.1 to 0.9 μm, by U. Mohideen and Anushree
Roy, Phys. Rev. Lett. 81, 4549 (1998). A summary is available at http://focus.aps.org/v2/st28.
html
Combining equation (1.12.7) with (1.12.4) gives

p 1

x ≥ ⇔
h 4π
h

p
x ≥ . (1.12.8)
4π
This is the more famous “Heisenberg uncertainty principle,” which states that it is
not possible to simultaneously determine a particle’s position and its momentum with
absolute precision. Both of these important quantum mechanical uncertainty relations
(1.12.6) and (1.12.8) are direct consequences of attributing a wave nature to particles
and not the result of any other “quantum mechanical weirdness.”
Despite all the above arguments, we are very often in the situation where the time
interval
t is much, much longer than the period T . In such a case, the frequency is
fairly well defined (i.e.,
f is small), and we need not worry much about the concerns
raised in this section.
Concept and skill inventory for chapter 1
After reading this chapter, you should fully understand the following
terms:
Stable equilibrium (1.2)
Hooke’s Law (1.2)
DEQ (1.2)
Simple Harmonic Motion (SHM) (1.2)
Harmonic approximation (1.2)
Amplitude, phase, frequency, angular frequency, period (1.2–1.3)
Adjustable constants (1.2)
Capacitor, inductor (1.5)
Kirchhoff’s loop rule (1.5)
Isomorphism (1.5)
Taylor series (1.6)
Euler’s equation (1.7)
Complex plane (1.8)
Magnitude of a complex number (1.8)
Complex conjugate (1.8)
Signal generator (1.10)
Ground for electrical circuits (1.10)
Complex version of Ohm’s Law (1.10)
Impedance, as applied to electrical components (1.10)
Low-pass filter (1.10)
Log–log axes (1.10)
Quantum mechanical wavefunction (1.11)
Probability density (1.11)
Free electron (1.11)

Wavenumber (1.11)
Frequency–time uncertainty relation (1.12)
Inverse wavelength – position uncertainty relation (1.12)
Energy–time uncertainty relation (1.12)
Heisenberg (momentum–position) uncertainty relation (1.12)
You should know what happens when:

A system described by the harmonic approximation is displaced from equilibrium and
then released (1.2–1.3)
A complex number is multiplied by eiα (1.8)
An input voltage is applied to a low-pass filter (1.10)
You should understand the following connections:

Frequency, angular frequency, & period (1.2–1.3)
Mass on a spring & an LC oscillator (1.5)
Current in a circuit & charge on a capacitor; be able to tell whether I = q̇ or instead
I = −q̇ (1.5)
Cosine representation & complex exponential representation for SHM (1.9)
x & z (1.9)
One-dimensional oscillation & rotation in the complex plane (1.9)
V & Ṽ (1.10)
You should be familiar with the following additional concepts:

Dot notation for time derivatives (1.2)
Angular frequency for a mass/spring (1.3)
The frequency for SHM doesn’t depend on amplitude (1.3)
Angular frequency for an LC oscillator (1.5)
First-order Taylor series approximation for sin (1.6)
Second-order Taylor series approximation for cos (1.6)
Taking the real part doesn’t commute with multiplication by a complex number (1.9)
Impedance of resistors, capacitors, and inductors (1.10)
You should be able to:

Check a proposed solution to a DEQ to determine if it’s correct and if there are
restrictions on the parameters (1.2)
Find the amplitude and phase given the initial position and velocity (1.3)
Deal with the multi-valued aspect of the arctan function (1.3)
Use the isomorphism between the mass/spring and the LC oscillator to quickly adapt
results from one system to the other (1.5)
Given a function, create the Taylor series for it (1.6)
Explain what it means to take the exponential of a complex number (1.7)
Go back and forth between Cartesian and polar representations for complex num-
bers (1.8)
Find the magnitude of a complex number (1.8)
Find the energy of an oscillator given either the spring constant and the amplitude or
the mass and the
maximum velocity (1.9)
VOUT
Calculate
for any simple combination of resistors, inductors, and capacitors
VIN
(1.10)
In addition to all of the above, you should be able to combine the concepts
you’ve learned to address new situations.
Problems
Note: Additional problems are available on the website for this text.
Instructor: Ratings of problem difficulty, full solutions, and additional support

materials are available on the website.
1.1 What is the difference between a stable and unstable equilibrium? Give
examples of each type of equilibrium from everyday life.
1.2 Plasma oscillations. To turn a gas into a “plasma,” one or more electrons
must be completely separated from each nucleus. This can be accomplished
by applying a very strong electric field (as in a spark or a lightning bolt), by
the absorption of ultraviolet light (as happens in the “ionosphere,” one of the
upper layers of our atmosphere), or by heating to a very high temperature
(as occurs in the sun). We will model a plasma as a “gas” of electrons
with number density n. (In other words, there are n electrons per cubic
meter.) Each electron has charge -e and mass me . Occupying the same
volume is a gas of positively charged ions, each with charge +e. Because
the ions were created by removing the electrons from neutral atoms, the
number density of ions is also n. (In other words, there are n ions per cubic
meter.)
Consider a cube of plasma, with sidelength ℓ. Somehow (e.g., by applying
an electric field), the electrons are all displaced a small distance x to the right.
This creates a layer of negative charge on the right, and leaves behind a layer
of positive charge on the left, as shown in figure 1.P.1. For simplicity, treat
each of these layers as an infinite thin sheet of charge. The layers create an
electric field throughout the plasma, pointing to the right. This field exerts a
force to the left on the electrons filling the central region of the figure, and
to the right on the ions. However, we’ll focus on the electrons because they
are so much lighter, and assume that the ions remain stationary.
When we turn off the original force that caused the displacement x of
the cube of electrons, the electric field from the surface charge layers
causes the electrons to spring back toward the left. However, because of
their finite mass, they overshoot the equilibrium position (corresponding to
Figure 1.P.1 Model for the ionosphere. The figure

shows a cross-sectional view of a cube of sidelength ℓ.
The electrons are represented by the region that is
shaded with down-slanting lines, while the positive
charges are represented by the region shaded with
up-slanting lines. Elayers is the field due to the thin
layers of charge on the left and right surfaces.
displacement x = 0), move to the left of the ions, then are pulled back to the
right, and so on, in an oscillatory motion. In this problem, you’ll calculate
the frequency of this oscillation.
(a) Explain why the field produced by the combination of the two charge
nex
layers in the region between them is E = . (You may need
ε0
to refer back to your intro. E&M textbook. Remember that we’re
treating the layers of charge as infinite sheets.)
(b) Explain why this leads to a restoring force on the electrons F =
n2 e2 ℓ3
− x. (Remember that x ≪ ℓ, so that the total charge of
ε0
electrons that experiences the electric force is not significantly
changed by the small number in the displaced region x.)
(c) Explain
why this means that the oscillation frequency is ω =
ne 2
. This is called the plasma frequency. Only radio waves with
m e ε0
frequencies significantly higher than this can propagate through
the plasma. Lower frequency waves are instead reflected. Thus,
AM radio waves (which are low frequency) can bounce off the
ionosphere, leading to very long transmission range under good
conditions (at night), whereas FM radio waves (higher in frequency)
don’t bounce off the ionosphere, and so the transmission range is
much more limited.
1.3 Derive equations (1.3.4a) and (1.3.4b) from the equations immediately
preceding them.
1.4 Consider the potential energy described in problem 1.14. For low amplitudes,
the motion of the object is well described by simple harmonic motion, so
that the period is independent of amplitude. However, once the amplitude
gets high enough this is no longer true. As the amplitude increases, does the
period increase or decrease? Explain your reasoning thoroughly, and assume

that the amplitude is always less than π/β .
1.5 A particle of mass m moves in the potential energy U = α x 2 + β x 4 , where
α and β are both positive. (a) What is the angular frequency of oscillation
for small amplitudes? (b) If the amplitude is increased far enough, one finds
that the angular frequency starts to depend on amplitude. Does the angular
frequency increase or decrease as the amplitude is increased? Explain your
reasoning.
v
1.6 We can rewrite equation (1.3.4a) to get ω0 = 0 . This makes it seem
A2 − x02
that ω0 depends on the amplitude and initial conditions. Explain this seeming
contradiction with equation (1.3.3).
1.7 Before doing this problem, read footnote 4 in section 1.3 about linear
homogeneous DEQs. Show that, as claimed in that footnote, A1 cos ω0 t +

A2 sin ω0 t = A cos(ω0 t + ϕ ), with A = A1 + A22 and ϕ = tan−1 −A2 /A1 .
2
1.8 A mass sits on a platform which oscillates vertically in simple harmonic

motion at a frequency of 5 Hz. Show that the mass loses contact with the
platform when the amplitude exceeds 1 cm. (Assume g = 9.8 m/s2 .) Hints:
The mass loses contact with the platform if the downward acceleration of
the platform exceeds g. To keep the math simple, choose a zero phase factor,
i.e., choose x = A cos ω0 t .
1.9 Home experiment (work with only one other student on this, and include
his/her name with your problem): The light emitted by a nonflatscreen TV
makes a good stroboscope. A given point on the screen is actually dark most
of the time; it is lit a small fraction of the time at a regular repetition rate,
which we’ll call fTV . The object of this experiment is to measure fTV ; I’ll
tell you that it’s either 30 or 60 Hz. As a very crude measurement, wave
your finger in steady oscillation in front of the screen at a rate of about 4
Hz, for example. Your finger will block the light from the screen wherever it
happens to be when the screen flashes on. Roughly measure (as best you and
your partner can) the amplitude of your finger’s oscillation. Then measure
the separation between successive finger shadows at the point of maximum
velocity. Assume the motion is sinusoidal. Calculate the maximum velocity
of the finger, given the amplitude and finger frequency. Put all of this together
to find fTV . (Recall that you’re told that it’s either 30 or 60 Hz, so your
measurements need only be precise enough to get a rough answer which
allows you to determine which of these two possibilities is correct.)
1.10 Read the aside about the arctan function in section 1.3. Explain
why a
−1
−v0
more complete version of equation (1.3.4b) would be ϕ = tan +
ω x0

0 if x0 > 0
.
π if x0 < 0
1.11 Before doing this problem, read footnote 4 in section 1.3 about linear
homogeneous DEQs. A particle far from the Earth feels the pull of gravity
GMm
F = − 2 , where G is the gravitational constant, M the mass of the Earth,
x
m the mass of the particle, and x is the distance from the center of the Earth to
the particle. Assume that the particle moves along a straight line that passes
through the center of the Earth, that gravity is the only force acting on it,
and that for the time period we are considering the particle doesn’t touch the
surface of the Earth (i.e., that it’s out in space).
(a) Write a DEQ of motion for the particle. (As an example of what
I mean by a DEQ of motion, for the simple harmonic oscillator it
would be ẍ + ω02 x = 0.)
(b) If x1 (t) is one solution of the DEQ and x2 (t) is another, is the
combination x1 + x2 necessarily a solution? Why or why not?
1.12 Analogy between a capacitor and a spring. (a) How is adding charge to
a capacitor like compressing a spring? (b) Why is it that the inverse of C
isomorphic to k, rather than just C itself ?
1.13 In research-level theoretical physics, it is almost never possible to get an
exact solution because of the complexity of the problems being considered.
Therefore, it is essential to make appropriate approximations, so as to get
physical insight. The Taylor series is central to many of these approximations.
You have already seen two of the three most common applications of the
2
∼ θ (for small θ ) and cos θ =
Taylor series: sin θ = ∼ 1 − θ (for small θ ).
2
In this problem, you will demonstrate the third of the three most common
applications. Show that (1 + x)n ∼ = 1 + nx for x ≪ 1. (Note that this works
whether n is positive or negative, integer or fractional.)
1.14 The potential energy for a particular object is U (x) = −L cos β x, where L
and β are both > 0. (This potential energy function is important in the study
of superconductivity.)
2π 2π
(a) Make a sketch of this potential energy from x = − to x = + .
β β
Indicate the scale on the vertical axis.
(b) The object has mass m and total energy ETOT = −L + G, where
0 < G ≪ L. (The symbol “≪” means “much less than.”) Add a
dashed line to your sketch indicating this total energy.
(c) At t = 0, the object is at x = 0. Show that its motion can
be approximated by a simple harmonic oscillation, and find the
approximate frequency of oscillation. Hint: recall that the Taylor
θ2 θ4
series expansion for cos θ is cos θ = 1 − + − · · · , so that
2! 4!
θ 2
for θ ≪1, cos θ ∼=1− .
2!

C1 C
1.15 Let C1 and C2 be complex numbers. Show that = 1 . Reminder:

C2 C
2
C means “magnitude of C ”
1 1
1.16 For each of the following, express the quantity shown either in the
form a+ib (i.e., Cartesian representation) or in the form Aeiα (i.e., polar
representation), whichever you find easier for each part of the problem.
In case you might be confused by the way I’ve written things: “i 4.5” means
“i times 4.5.”
(3.2 + i 6.7) + (5.6 – i 4.5)
(a)
6.1 ei 1.2 + 1.2 ei 1.7
(b)
(3.2 + i 6.7)(5.6 – i 4.5)
(c)
(6.1 ei 1.2 )(1.2 ei 1.7 )
(d)
What point is this problem trying to get across?
(e)
√
1.17 Let z1 = 8eiπ/6 and z2 = 2ei3π/4 .
(a) Represent z1 and z2 in the complex plane.
(b) Find the real and imaginary parts of z1 and z2 .
Express each of the following in the form Aeiϕ , where A and ϕ are
real:
(c) z1 + z2
(d) z13 z22
1.18 A mass m = 10 kg is oscillating on a spring with k = 10 N/m with little
damping. The
displacement
of the mass can be described by
x(t) = Re Ceiωt , where C =(1 – i) cm.
(a) What is the value of ω?
(b) What is the amplitude of the motion?
(c) The solution can also be expressed in the form x(t) = A cos(ωt + ϕ ).
What is the value of ϕ ?
(d) Describe the initial conditions of the motion, that is, specify the
position and velocity at t = 0. As for all numbers in physics (except
dimensionless quantities) be sure to include units!
(e) Sketch two graphs, one of position versus time and the other of
velocity versus time. Be sure to label both axes of each graph
quantitatively.
(f) What is the energy of the oscillator?

V
1.19 For an RC low-pass filter (figure 1.10.2b), show that OUT drops by a
VIN
factor of 10 for each factor of 10 increase in ω, for ω ≫ ωLO .
1.20 Using methods similar to those leading to equation (1.10.2), show that the
complex impedances of two circuit elements in parallel combine in the way
specified by equation (1.10.3).
1.21 A voltage V (t) = V0 cos ωt is applied across a capacitor with capacitance
C. (a) Without using a symbolic algebra program or graphing calculator,
make a sketch with curves for both V (t) and I (t), showing two full
periods of oscillation. Label your sketch quantitatively. (b) For a capacitor,
does the current “lead” the voltage (meaning that the current reaches
a peak before the voltage does), or does the voltage lead the current?
(c) Using simple ideas about charging of the capacitor and the connection
between voltage and charge, explain why your answer to part (b) makes
sense.
1.22 A voltage V (t) = V0 cos ωt is applied across an inductor with inductance L.

(a) Without using a symbolic algebra program or graphing calculator, make
a sketch with curves for both V (t) and I (t), showing two full periods of
oscillation. Label your sketch quantitatively. (b) For an inductor, does the
current “lead” the voltage (meaning that the current reaches a peak before
the voltage does), or does the voltage lead the current? (c) Using elements
from the isomorphism between an LC oscillator and a mass/spring system,
explain why your answer to part (b) makes sense.
1.23 A voltage V (t) = V0 cos ωt is applied to the input of an RC low-pass filter
(figure 1.10.2b). At the output, one observes a voltage VOUT cos(ωt + ϕ ).
(a) Find ϕ as a function of ω, R, and C. (b) Using whatever software you
like, make two graphs of -ϕ as a function of ω/ωLO , ranging from ω/ωLO =
0.01 to 100, one plot with linear axes and one with logarithmic axes.
1.24 The RC high-pass filter. For the circuit shown in figure 1.P.2, a sinusoidal
input voltage is applied relative to ground, and the resulting output voltage is
measured relative to ground. If a sinusoidal voltage Vin = Vi cos ωt is applied
to the input, one observes a sinusoidal voltage Vout = Vo cos(ωt + ϕ ) at the
amplitude of output voltage
output (relative to ground). (a) Show that =
amplitude of input voltage
Vo 1 1
= , where ωHI ≡ . (b) Show that the phase shift
Vi ωHI 2
RC
1+
ω ω
HI
of the output relative to the input is ϕ = tan−1 . (c) Given that
ω
ϕ is positive, does the output “lag” the input (i.e., do the peaks in the
output voltage occur after the peaks in the input), or does the output “lead”
the input (i.e., do the peaks in the output occur before the peaks in the
input)?
1.25 More on the RC high-pass filter. Read all of problem 1.24 before beginning
this problem. (a) Using a suitable computer program, make two plots of
amplitude of output voltage V ω
= o versus . In your first plot, use linear
amplitude of input voltage Vi ωHI
scales for both axes. In your second plot, use logarithmic scales for both
ω
axes. For both plots, let vary from 0.01 to 100. (b) Make two plots of ϕ
ωHI
ω
versus . In your first plot, use linear scales for both axes. In your second
ωHI
ω
plot, use a linear scale for the ϕ axis and a logarithmic scale for the axis.
ωHI
Figure 1.P.2 The RC high-pass filter.

ω
For both plots, let vary from 0.01 to 100. (c) For ω ≪ ωHI , show that
ωHI
amplitude of output voltage V
= o drops by a factor of 10 for each factor of
amplitude of input voltage Vi
10 decrease in ω.
1.26 More on the RC low-pass filter. (a) Consider the RC low-pass filter shown
in figure 1.10.2b. If a voltage (relative to ground) Vin = Vi cos ωt is applied
to the input, one observes a sinusoidal voltage Vout = Vo cos (ωt + ϕ ) at the
output (relative to ground). Show
that the phase shift of the output voltage
−1 ω 1
relative to the input is ϕ = tan − , where ωLO ≡ . (b) Given
ωLO RC
that ϕ is negative, does the output “lag” the input (i.e., do the peaks in
the output voltage occur after the peaks in the input), or does the output
“lead” the input (i.e., do the peaks in the output occur before the peaks in the
input)?
1.27 The RL low-pass filter. One can use resistors and capacitors to build either
low-pass filters (see section 1.10) or high-pass filters (see problems 1.24
and 1.25). It is also possible instead to use resistors and inductors for
these tasks; we’ll explore the low-pass version in this problem. In practical
circuits, RC filters are much more common than RL filters, partly because
it is easier to build an essentially ideal capacitor than to build an ideal
inductor. (There is always resistance in the wires used to wind an actual
inductor, and there is always capacitance between the windings.) Also,
inductors are generally bulkier and more expensive than the capacitors
that could be used to build comparable filters. However, one does see RL
filters in some high frequency applications. Consider the circuit shown in
figure 1.10.4. If a voltage (relative to ground) Vin = Vi cos ωt is applied to
the input, one observes a sinusoidal voltage Vout = Vo cos(ωt + ϕ ) at the
amplitude of output voltage
output (relative to ground). (a) Show that =
amplitude of input voltage
Vo 1 R
= , where ωLO ≡ . (b) Show that the phase shift of
Vi ω 2 L

1+
ωLO

ω
the output relative to the input is ϕ = tan−1 − . (Note that, except
ωLO
for the definition of ωLO , these expressions are the same as for the RC
low-pass filter.)
1.28 An electron in an atom can be excited from its original “ground state” to
a well-defined and reproducible higher energy metastable “excited state.”
The electron can then “fall” back down to the ground state, emitting a
photon in the process which has energy equal to the difference between
the ground state and the excited state. (As it turns out, this energy is
characteristic of the type of atom, and so analysis of such photons can be
used for determining the elements which are present in a sample.) Many
of these excited states have only a very short lifetime. For such a state,
if the experiment is performed carefully, one can determine that the photons
emitted from a large sample of material do not all have exactly the same
energy, but instead there is a small spread. If a particular excited state has
a lifetime of 10−7 s, about how big a spread in photon energy would you
expect? Hint: Use the energy-time uncertainty relation; this is meant to be
a very easy problem.
2 Examples of Simple Harmonic Motion
And I saw the cantilever jutting through the mist, resplendent in the light of dawn,
oscillating jauntily like a promise of joy.
–Marian McKenzie
2.1 Requirements for harmonic oscillation
In this chapter, we will explore several examples of the remarkable variety of systems
that show the harmonic oscillations described in chapter 1. There are two basic
requirements for a system to oscillate: (1) If the system is disturbed from equilibrium,
there must be something (such as a force) that tends to bring it back toward equilibrium.
For the oscillations to be of the sinusoidal form described in chapter 1, this restoring
drive must be proportional to the displacement from equilibrium, for example, the
spring force F is proportional to x: F = −kx. (2) As the system moves toward
equilibrium, there must be something (such as inertia) which tends to make the system
overshoot the equilibrium point.
Saying the same thing mathematically, if the system is described by a differential
equation of the form
d2 x
F = −kx ⇒ m 2 = −kx! , (2.1.1)
dt !

restoring term:
inertial term: satisfies
satisfies condition 1
condition 2
then it will exhibit harmonic oscillations.

Sometimes, it is easier to consider the energy of a system. If the system can be
described by an equation of the form

2
1 dx 1 2
2m dt
+ 2 kx = constant, (2.1.2)
!
!
potential
kinetic energy
energy
then it will exhibit harmonic oscillation.
39
In other words, if we can show that a system obeys an equation either of the form
(2.1.1) or of the form (2.1.2), then we can immediately conclude that

k
x = A cos ω0 t + ϕ , where ω0 = . (2.1.3)
m
2.2 Pendulums
Professor Roger Newton, author of the book Galileo’s Pendulum, recounts this
wonderful legend about Galileo, the world’s first experimental physicist, and a
revelation that occurred during a church service in 1581:
He was seventeen and bored listening to the Mass being celebrated in the cathedral of
Pisa. Looking for some object to arrest his attention, the young medical student began
to focus on a chandelier high above his head, hanging from a long, thin chain, swinging
gently to and fro in the spring breeze. How long does it take for the oscillations to repeat
themselves, he wondered, timing them with his pulse. To his astonishment, he found
that the lamp took as many pulse beats to complete a swing when hardly moving at all
as when the wind made it sway more widely.1
This description of one of the first quantitative observations of experimental physics
shows the historic importance of pendulums in physics. We will see in chapter 4 that
pendulums provide an excellent illustration of chaos theory. Pendulums are common
in everyday life, from a baby’s swing to a grandfather clock, from a fair ride to a
wrecking ball.
We recognize Galileo’s observation that the period is independent of the amplitude,
as a characteristic of simple harmonic motion. In the case of the pendulum, although
there is an obvious mass, there is no obvious spring. Yet, since it does have a position of
stable equilibrium, we should be able to model the potential energy near this position
as a parabola, and so we should be able to find an “effective spring constant” that
arises from the combination of gravity and the tension in the string.
Let’s consider an arbitrary rigid object of mass m that can rotate in the x–y plane
about a pivot P, as shown in figure 2.2.1. We’ll show that, for small amplitudes of
swing, the energy takes the form (2.1.2). In the figure, we displace the pendulum by
an angle θ from equilibrium. The potential energy is determined by the position of the
center of mass (marked CM in the figure): U = mgy, where y is the height of the CM
above its equilibrium position.
Your turn: Show that, if θ is small, then
U∼
= 1
2 mg ℓCM θ 2 . (2.2.1)
1. Roger G. Newton, Galileo’s Pendulum: From the Rhythm of Time to the Making of Matter,
Harvard University Press, Cambridge, MA, 2004, p. 1.
Chapter 2 ■ Examples of Simple Harmonic Motion 41
Figure 2.2.1 A pendulum of arbitrary form.
Hint: Use the Taylor expansion for cos θ , which you derived in section 1.6:
θ2 θ4
cos θ = 1 − + − ··· (1.6.5)
2! 4!
θ2
( If θ is small, this means that cos θ ∼
= 1 − .)
2!
The motion is a pure rotation about P, so the kinetic energy is K = 21 I ω2 , where

I is the moment of inertia for rotations about P and ω = θ̇ is the angular velocity.
Combining this with your result (2.2.1) gives the total energy:
E = U + K = 12 mg ℓCM θ 2 + 21 I θ̇ 2 . (2.2.2)
Since the total energy is constant, this has exactly the same form as
(2.1.2) : 12 kx 2 + 12 mẋ 2 = constant,
so that we immediately know that the pendulum displays simple harmonic motion,
that is, that

mg ℓCM
θ = θ0 cos ω0 t + ϕ , where ω0 = . (2.2.3)
I
Motion of a pendulum (mass need not be concentrated at a point).
I = moment of inertia about pivot, ℓCM = length from pivot to CM
Core example: the simple pendulum. The simplest example of a pendulum is a compact
mass (the “pendulum bob”) at the end of a thin rod of length ℓ; we assume the mass
of the rod is negligible compared to that of the bob. In this case, the moment of inertia
about the pivot point is I = mℓ2 . Plugging this into equation (2.2.3) gives

g
ω = . (2.2.4)
simple ℓ
pendulum
For more complicated objects, one often uses the parallel axis theorem, which you
may have seen proved in an introductory physics book:
I = ICM + mh2 , (2.2.5)
The parallel axis theorem
where ICM is the moment of inertia for rotations about the center of mass and h is the
distance from the pivot point P to the center of mass. By breaking a complicated object
up into smaller symmetrical objects and applying the parallel axis theorem, one can
compute I of the complicated object.
Although the harmonic motion of the pendulum is most easily seen by considering
the time dependence of θ (as we have done earlier), we can also show that there is a
harmonic variation in the horizontal position x. For a simple pendulum I = mℓ2 , so
that equation (2.2.2) becomes
" #2
mg 2 2 1 2 2 1 mg d
E = 12 mgℓθ 2 + 21 mℓ2 θ̇ 2 = 21 ℓ θ + 2 mℓ θ̇ = 2 (ℓθ )2 + 21 m (ℓθ ) .
ℓ ℓ dt
As we can see from the figure, in the limit of small displacements, the arc length ℓθ ∼= x,
so that
mg
E∼
= 1
2 x 2 + 12 mẋ 2 .
ℓ
Since this has the same form as the energy of a mass/spring system, E = 12 kx 2 + 21 mẋ 2 ,
mg
we see that the effective spring constant for the pendulum is kpendulum = . This
ℓ
means that the net restoring force, which is created by a combination of gravity and
the string tension, is
mg
Fpendulum = −kpendulum x = − x. (2.2.6)
ℓ
“Pendulum force” resulting from the combination
of gravity and tension for a simple pendulum.
We will use this result again in chapter 5.
2.3 Elastic deformations and Young’s modulus
All materials are at least a little stretchy, although the amount an object can be stretched
before breaking is often too small to see with the naked eye. This stretchiness, and
the vibrations that occur when an object is stretched or twisted and then released,
determine the engineering limits of building materials, the performance limits of
automotive components, and the behavior of a new class of devices known as Micro
ElectroMechanical Systems (MEMS).2 In the remainder of this chapter, we’ll explore
2. These devices, which combine mechanical motion with electrical actuation or sensing, are
fabricated using techniques of photolithography, electron beam lithography, and various types of
Figure 2.3.1 Top: a relaxed spring at

its equilibrium length. Bottom: When
the spring is stretched, the left and
right halves stretch by equal amounts.
various types of elastic (i.e., reversible) deformations, and their importance in science
and everyday life.
The simplest way to deform an object is to stretch or compress it. Consider a long
object of uniform cross-section, such as a beam, which is anchored at the left end. If
a force is applied to the right end, one always observes that the beam stretches by an
amount proportional to the force. This comes as no surprise – before the force is applied,
the beam is in equilibrium, and by the arguments in chapter 1 any displacement from
equilibrium is countered by a force Fspring = −kx. Therefore, to produce a displacement
x, we must apply Fapplied = −Fspring = kx.
The spring constant k depends on the material from which the beam is made;
diamond is stiffer than rubber. However, k also depends on the cross-sectional area
and length of the beam. We wish to divide out these geometric dependencies to get
a parameter that describes the springiness or stiffness of the material itself. First, we
consider the dependence on length. How does k change if the beam is cut to half its
length? Since we’re modeling the beam as a spring, this is equivalent to asking what
happens to the spring constant of a spring when it is cut in half.
Imagine a spring of equilibrium length ℓ which is attached to a wall on the left side.
A dog pulls on the right end, stretching it by an amount
ℓ, as shown in figure 2.3.1.
The force exerted by the spring on the dog is
Fby spring = −k
ℓ
Fby spring Fby spring
⇔k=− =− . (2.3.1)

ℓ extension
anisotropic (meaning directionally dependent) etching. The sizes of the devices range from the
diameter of a human hair down to the molecular regime, allowing extremely fast response times
and, for devices designed to detect trace chemicals, extraordinary sensitivity. We will discuss
MEMS devices in sections 2.6 and 3.4. You can learn more about MEMS in problem 2.14, and
in the website for this text, under the entry for this section.
The dog must be exerting an equal and opposite force k

ℓ on the right end of the
spring. Since the right half of the spring doesn’t accelerate, we know there must be
another force to balance this; this force is exerted by the left half on the right half
Fby left = −k
ℓ.
half

ℓ
The extension of the left half is . Therefore, by analogy with equation (2.3.1), the
2
spring constant of the left half is
Fby left
half (−k
ℓ)
kleft = − =− = 2k .
half
extension (
ℓ/2)
Therefore, when a spring is cut in half, the spring constant gets doubled. (Another way
to see this: for the same extension, the coils of a short spring are distorted more than
the coils of a long spring, therefore the shorter spring exerts a bigger force.)
Of course, this also means that, if the length of the spring (or in our case the length
1
of the beam) is doubled, the spring constant is halved. Thus, k ∝ . (The symbol ∝
ℓ
means “proportional to.”)
Your turn (answer3 at bottom of page): Explain why k is proportional to the cross-
sectional area A of the beam.
A
Putting these results together, we can write k ∝ or
ℓ
A
k=E , (2.3.2)
ℓ
where the proportionality constant E is called “Young’s Modulus,”4 and ℓ is the
equilibrium length of the beam.
We can also write
EA Fspring x
Fspring = −kx = − x ⇔ = −E .
ℓ A ℓ
The force applied to the beam, Fapplied , is equal and opposite to the force Fspring applied
by the beam, so that
Fapplied x
=E . (2.3.3)
A ℓ
3. Answer: We can divide the beam into N smaller beams running in parallel along the length. The
force from a single one of these would be Fsmall = −ksmall x, where x is the amount by which
the beam is stretched. The force
from the entire beam is the total of the forces from the small
beams: FTOT = NFsmall = − Nksmall x, so that the total spring constant is k = Nksmall . If the
cross-sectional area A is doubled, then N doubles, so k ∝ A.
4. This is named after Thomas Young, who is most famous for his 1801 two-slit experiment, which
established the wave nature of light. Young was also a physician, and figured out how the eye
focuses on objects at different distances. He was familiar with twelve languages by the age of
fourteen, and later was involved in the translation of the Rosetta stone.
The quantity
F
σ ≡ (2.3.4)
A
is called the stress. It is usually best to think of the F in this relation as being Fapplied .
The stress has units of pressure; in SI units, 1 Pa (“Pascal”) = 1 N/m2 . The quantity
x
ε≡ (2.3.5)
ℓ
(the amount by which the beam stretches divided by its equilibrium length) is called
the strain. Combining equations (2.3.3)–(2.3.5),
σ = E ε. (2.3.6)
In idiomatic English, the terms “stress” and “strain” are used in very similar ways,
for example, “The stress of this job is killing me.” or, “I can’t stand the strain of this
responsibility.” We can see above that, as used in physics and engineering, these are
quite different quantities, with different units. The stress is usually best thought of
as something applied to the sample, while the strain is usually best thought of as the
response of the sample to the stress. It might help you to remember that “stress” sounds
like the first part of “pressure.”
Figure 2.3.2 shows a graph of strain versus stress for a typical metal beam.
Rearranging equation (2.3.6) gives ε = σ/E, so that the slope of this graph is 1/E.
For low stress, the graph is linear, as predicted by equation (2.3.6). However, once
the stress becomes large enough, the beam no longer follows this relation. As shown,
when the strain reaches the “elastic limit” (typically about 0.2% for metals), the spring
constant decreases, then the beam “yields” and stretches a great deal with no additional
increase in the applied stress. Then (again for metal beams), the microscopic structure
changes in a process called “strain hardening,” and finally the beam breaks. Table 2.3.1
lists typical values of three important material parameters.
Figure 2.3.2 Strain vs. stress for a typical metal beam.

Table 2.3.1. Typical values of material parameters:
Material Young’s Modulus E Yield stress Shear Modulus G

(109 N/m2 ) (106 N/m2 ) (109 N/m2 )
Aluminum 75 300 28
Brass 100 250 40
Steel 200 400 75
Concrete 20 40a
(compression only)
Rock 50 100a
(compression only)
Plastic 2 20 0.1
Rubber 0.002 3 0.0005
Wood 10 40
Carbon nanotube 1000 60000
a
These materials do not exhibit yielding. The number quoted is the ultimate stress. i.e.
the stress at breakage.
Self-test (answer below5 ): From table 2.3.1, we see that the values of E are a few hundred
times larger than the values of the yield stress. How are E and the yield stress related?
Example: Scientists often need to isolate sensitive scientific equipment, such as atomic
force microscopes (AFMs), from building vibrations. An economical and effective way to
do this is to place the equipment on a platform that is suspended on soft springs. As we’ll
see in chapter 4, the effectiveness
of this vibration isolation is best when the vibration
k
frequency of the system, ω0 = m is as low as possible. A scientist wants to suspend
an AFM of mass 5.4 kg (including the mass of the platform) using a rubber bungee cord
of equilibrium length 1.2 m. If she wants ω0 = 10.0 rad/s, what is the required diameter
for the rubber cord? (Assume the mass of the cord is negligible compared to that of
the AFM.)

k
Solution: ω0 = m ⇒ k = ω02 m. Plugging in the numbers gives k = 540 N/m. From
equation (2.3.2), we have k = E Aℓ ⇔ A = kEℓ . From table 2.3.1, the Young’s Modulus
for rubber is E = 0.002 × 109 N/m2 . Plugging in ℓ = 1.2 m gives A = 3.2 cm2 . Since
2
A = π d /2 , where d is the diameter, this gives d = 2.0 cm.
The cord stretches a distance given by |F | = kx = mg ⇔ x = mg
k = 9.8 cm when the
load is applied to it. The stress under load is approximately σ = mg 5
A = 1.6× 10 N/m ,
2
well below the yield stress.6
5. Answer to self-test: The yield stress is approximately equal to the Young’s modulus multiplied
by the elastic limit, which is about 0.2%.
6. As the cord stretches, its cross-sectional area decreases somewhat, so this value for the stress
is less than the true value. We can get a different estimate by assuming that the volume of the
Aℓ
cord remains constant, so that Aℓ = Astress (ℓ + x) ⇔ Astress = ℓ+ x , where Astress is the cross-
sectional area with the load applied. Plugging in the numbers gives Astress = 3.0 cm2 , so that our
new estimate of the stress under load is σ = Amg stress
= 1.8× 105 N/m2 . This is greater than the
actual stress, because the volume actually increases when a beam (or in this case a bungee cord)
2.4 Shear
If we apply a force perpendicular to the face of a solid object, then it is stretched or

compressed, and the force over the area is the stress, as discussed in section 2.3. The
force can be spread uniformly over the surface, resulting in a uniform pressure or stress
σ = F /A. A force that is instead applied parallel to the face is called a shear force. If
this force is applied uniformly over the top surface of the cube shown in figure 2.4.1a,
and this face has area A, then the shear stress is
Fapplied
σshear ≡ . (2.4.1)
A
Figure 2.4.1 a: A force applied parallel to a face is called a shear force. For static equilibrium,
forces must be applied to three other faces of the cube, as shown. This results in shear strain, as
shown on the right. By imagining that we divide the cube into small parallel beams, such as
that shown shaded on the right, we can see that the force with which the cube resists the shear
deformation is proportional to the area of the top surface. b: Shear oscillations (side view).
c: Front and back surfaces of a crystal from a quartz crystal microbalance, with gold electrodes
applied. Image from Wikimedia Commons.
is put under stress. For most materials, the volume increases by a fraction in the range ε/3 to ε/2.
So, the correct value for σ is between this overestimate of 1.8 × 105 N/m2 and the underestimate
of 1.6 × 105 N/m2 given in the main text above. For metal beams, the changes in volume and in
cross-sectional area are much smaller, and can often be ignored.
For static equilibrium to be achieved, this force to the right on the top surface must
be balanced by an equal force to the left on the bottom surface. (Again, we assume
this force is applied uniformly over the surface.) However, this pair of forces would
tend to rotate the cube clockwise. So, to provide zero net torque about the center
of the cube, we need forces on the left and right sides, as shown in figure 2.4.1a,
and again we assume these are spread uniformly over the surface. This combination
of forces produces a condition of pure shear, resulting in the deformation shown in
the right side of figure 2.4.1a. If we imagine slicing the cube into thin horizontal
slices, each slice is moved a little to the right of the one below it; this is called
“shear strain.”
By now, it will not surprise you that the solid resists an applied shear force with
a force F = −kshear x. We can see from the figure that the displacement x is
spring
shear
Fspring
Fapplied
proportional to the length ℓ. Since kshear = − shear = , we see that kshear ∝
x x
1
. By the same arguments as in section 2.3, we could imagine dividing the solid
ℓ
into smaller parallel beams, each of which resists the shear force, so that kshear ∝ A.
A
Combining these results, we get kshear ∝ , or
ℓ
A
kshear = G , (2.4.2)
ℓ
where the proportionality constant G is called the Shear Modulus.7
As in section 2.3, we can also write
A Fapplied x
kshear x = Fapplied ⇒ G x = Fapplied ⇔ =G .
ℓ A ℓ
We define the shear strain to be
x
γ ≡ . (2.4.3)
ℓ
Therefore,
σshear = Gγ . (2.4.4)
In addition to the shear modulus, we can also define the shear yield stress and the
shear ultimate stress, which play roles analogous to those discussed in section 2.3. The
values of all three numbers are typically about half those of the corresponding numbers
for stretching/compression. (Some values are listed in table 2.3.1.)
Shear is particularly important in the design of earthquake-resistant buildings. In
many earthquakes, there is a strong lateral oscillation of the ground, resulting in large
shear stress applied to the building. Buildings which are not specifically designed to
withstand this type of stress are severely damaged by it.
7. This is named after William Shear (1701–1785), who also invented scissors. He received a
doctorate from Oxford at the age of 14, and … OK, just kidding.
Since we can define a spring constant for shear, we see that an object can undergo
shear oscillations, as shown in figure 2.4.1b. You can easily produce such oscillations
by placing a cube of jello on a plate and moving the plate sideways (simulating an
earthquake).
Connection to current research: Quartz Crystal Microbalances
A piezoelectric material responds to an applied voltage by changing its shape. The most
commercially important examples are quartz crystals (which exhibit a small but very
reproducible shape change) and lead zirconium titanate ceramics (which produce a
larger but less reproducible change). In a quartz crystal microbalance (QCM), a quartz
crystal in the shape of a thin disk is prepared in such a way that it can be excited into
shear oscillations by the application of appropriate voltages to gold electrodes on the
crystal surface, as shown in figure 2.4.1c.
Because the mass is distributed through the thickness of the crystal, the quantitative
analysis of these oscillations is beyond the level of this book. However, the frequency of
keff
oscillation is given by ω0 = , where keff is an effective spring constant (proportional
meff
to the shear modulus), and meff is an effective mass.
The most common use for QCMs is for monitoring vacuum depositions of thin films.
For example, one might wish to coat a silicon wafer with a layer of copper, to form
connecting wires between micro-fabricated circuits. One can apply such a coating in
several ways, including “thermal evaporation.” In this technique, the wafer is mounted
face down in a vacuum chamber, and almost all the air is pumped out. Below the
wafer, a pellet of copper is heated until it melts. Once the copper is liquid, it starts to
evaporate. The thermal energy of the copper atoms that evaporate off the liquid is quite
high so that they travel in straight lines until they strike the wafer. The thickness of
the copper film thus applied to the wafer is measured using a QCM, which is mounted
just to the side of the wafer. The copper atoms coat the surface of the quartz crystal
(as well as the silicon wafer), increasing its mass and thus changing ω0 . This change
can be measured so accurately that it is routine to measure thicknesses down to one
atomic layer!
QCMs can also be used in aqueous solutions. Typically, the surface of the QCM sensor
crystal is first coated with a receptor chemical, such as the antibody for a particular virus.
When the virus is introduced into the solution, it binds to the antibody resulting in
an increase in the effective mass of the oscillating crystal. This technique can be used
to make sensors, to study the progress of chemical reactions, to sort through possible
anti-cancer drugs, and for many other applications.
2.5 Torsion and torsional oscillators
As a child, you probably played with a yo-yo, and had trouble with it when the string
got twisted up. You may have held the top of the string, with the yo-yo resting at the
bottom of the string, and watched it slowly untwist, picking up rotational velocity as
Figure 2.5.1 a: A yo-yo twisting on

the end of its string. b: Twisting a
thin-walled tube.
it went, as shown in figure 2.5.1a. You may have noticed that the yo-yo didn’t stop
rotating when the string was untwisted, but instead kept rotating, twisting the string
in the other direction. Eventually, the yo-yo came to rest briefly before starting to
rotate in the opposite direction because of this new twist. Your yo-yo was acting as
a torsional oscillator. As we’ll explore, more sophisticated versions have been very
important in the history of physics and continue to be important in current research on
superfluids, supersolids, and other topics.
We begin by considering a thin-walled tube, as shown in figure 2.5.1b. Think of
it as a collection of thin strips, one of which is highlighted. We wish to calculate the
potential energy stored in this system when we twist it, and show that this has the form
of the energy stored in a spring, U = 12 kx 2 . From there, we will be able to find the
angular frequency of oscillation.
If we twist the top of the tube by a small angle θ , the strip experiences a shear
lateral displacement x rθ
strain given by = = . Therefore, it produces a force
length ℓ ℓ
Fspring = −kshear x = −kshear r θ.

shear
Using equation (2.4.2): kshear = GA/ℓ we then have
Astrip
Fspring = −G r θ,
ℓ
where Astrip is the cross-sectional area of the strip. The resulting torque is
Astrip
τ = rF = −G r 2 θ.
spring spring ℓ
strip shear
To find the total torque from all the strips, we simply replace Astrip by the total cross-
sectional area of the tube, A = 2π rt (valid for t ≪ r), where t is the wall thickness of
the tube:
r3t
τ = −2π G θ. (2.5.1)
spring ℓ
tube
To create the twist, we must apply a torque of equal magnitude and opposite sign:
r3t
τapplied = 2π G
θ.
ℓ
The work that we do in creating the twist is stored as potential energy.$ Recall from
your earlier study of mechanics that the work done in$ a rotation is W = τ dθ , which
is analogous to the work done in a translation, W = F dx. Therefore,
θ θ

r3t r3t
U = Won system = τapplied dθ = 2π G θ dθ = 1
2 2π G θ 2, (2.5.2)
ℓ ℓ
0 0
which, as hoped, has the same form as the potential energy of a conventional spring.
To complete our argument and find the angular frequency of oscillation, we must
consider the rotational kinetic energy. In most cases of interest, the object being twisted
(the string of the yo-yo, or the tube) is used to support a much more massive object,
such as the body of the yo-yo, which is called the rotor. For simplicity, we assume that
the rotational kinetic energy of the rotor is much larger than that of the twisted element
that supports it, so that

2
dθ
K = 12 Irotor ω2 = 12 Irotor .
dt
Therefore, the total energy is

2

dθ r3t
E =K +U = 1
2 Irotor + 1
2 2π G θ 2. (2.5.3)
dt ℓ
This is isomorphic to (i.e., is identical to except for a change of symbols)
equation (2.1.2):

2
dx
1
E = 2m + 21 kx 2 ,
dt

so that we can rewrite ω0 = mk for this case as

2π Gr 3 t
ω = . (2.5.4)
tube Irotor ℓ
torsion
In problem 2.12, you can show that, if the rotor is supported by a solid cylinder (e.g.,
a wire or a string) instead of a tube, the torque and angular frequency of oscillation are
instead given by

r4 π Gr 4
τ = −π G θ ω = . (2.5.5)
spring 2ℓ wire
torsion
2Irotor ℓ
wire
Perhaps the most famous use of a torsional oscillator occurred in 1798, when Henry
Cavendish8 used one to measure the gravitational attraction between two pairs of lead
8. Henry Cavendish made important contributions to chemistry as well as physics. He was the first to
isolate hydrogen gas, and showed that water is made from hydrogen and oxygen. He performed
Figure 2.5.2 Schematic diagram of the Cavendish

experiment. (Figure by Chris Burks.)
spheres, as shown in figure 2.5.2. The larger sphere in each pair was 1 ft (30.5 cm) in
diameter, and the smaller was 2 in. (51 mm). To isolate the experiment from air currents,
the instrument was housed inside a wooden box inside a specially built brick shed. The
operation and observations could all be performed by Cavendish from outside the shed,
using ropes, pulleys, and telescopes. To measure the gravitational interaction, he used
the apparatus in “torsion balance” mode: he brought the large spheres close to the small
ones, and measured how much the gravitational attraction caused the wire to twist; this
was a static, not oscillating, experiment. However, to determine the torsional spring
π Gr 4
constant of the wire (the quantity in equation (2.5.5)), he moved the large balls
2ℓ
far away, and set the rotor (composed of the two small balls and the horizontal crossbar
that connects them) into oscillation. By measuring the resulting oscillation frequency,
he could then deduce the spring constant.
Connection to current research: Torsional oscillators are used to investigate the phase
transitions of liquid helium to a superfluid and to a supersolid. See problem 2.13 to
learn more.
2.6 Bending and Cantilevers
When you jump on the end of a diving board, setting up a beautiful dive or an
enthusiastic cannonball, you apply a force straight down. Although this is parallel
many important experiments on electricity, including a precise demonstration of the inverse

square law for the force between two charges, but never published most of his work. He was very
shy, especially around women; he communicated with his women servants via notes. His main
social outlet was attendance at scientific meetings. However, even once his scientific reputation
was well-established, he could sometimes be observed standing outside the door of a meeting,
trying to work up enough courage to go in. He was fabulously wealthy; upon his death his estate
amounted to nearly a million pounds sterling, a tremendous sum. (At that time, a gentleman
could live comfortably on 500 pounds a year.)
to the (vertical) end face of the board, it is neither spread uniformly over this face nor
are there forces applied to the sides of the board to create the pure shear condition
shown in figure 2.4.1a. Therefore, the resulting deformation of the board is not pure
shear, although shear is involved along with other types of distortion. We consider the
situation in figure 2.6.1a, in which the left end of the board is anchored to a wall; this
configuration is called a cantilever.
The board, of course, tends to spring back to its original position. We want to find
the spring constant k, defined by
Fapplied = −Fspring = kd , (2.6.1)
where d is the displacement of the end of the board from equilibrium.

The full analysis is beyond the level of this text, however we can understand the
way that the spring constant depends on the dimensions of the board with a simplified
analysis. When the board is bent, the top half is stretched while the bottom half is
compressed; these distortions produce part of the restoring spring force. The board is
also subject to shear, which affects the displacement and the restoring force. In our
analysis later, we will neglect the effects of shear, so that the final proportionality
constant will not be correct, but the functional dependence on the dimensions of the
board will be.
In static equilibrium, a small part of the top half of the board (the shaded section
in the figure) must feel equal forces from the left and from the right. This is true for
any similar section along the length of the board. Thus, the amount of stretching (i.e.,
the “tensile strain”) in the top half must be uniform along the length. A constant stretch
along the length is only possible if the board is in the shape of a circular arc. (The
actual shape is more complicated, because of the effects of shear.) The radius of the arc
(called the “radius of curvature”) for a line along the center of the board is defined as r.
The board has a length L, a thickness t, and a width w (the dimension perpendicular
to the page in figure 2.6.1). Figure 2.6.1b shows an enlarged view of a section of the
board. Along the centerline, the length of this section is r θ , while a line a distance
z above the center is stretched to a length (r + z)θ . Thus, the tensile strain at the
position z is
amount of stretch [(r + z) θ − r θ ] z

ε= = = .
equilibrium length rθ r
Using equation (2.3.6), σ = E ε we see that the stress at the position z is

Ez
σ = .
r
This stress is the result of forces applied to the left end of the board by the wall,
as shown in figure 2.6.1a. The force applied to an infinitesimally thick horizontal
slice of the board (figure 2.6.1c) is the stress times the cross-sectional area of the
slice:
Ez
dF = σav w dz = w dz.
r
Figure 2.6.1 a: A cantilever is bent. The top half is stretched, while the bottom half is
compressed. The highlighted segment must feel equal forces from the left and right.
b: Enlarged view of a section of the cantilever. c: Enlarged view of the end of the cantilever
that is attached to the wall. d: Relation between displacement and radius of curvature.
This creates a torque z dF around the point P at the center of the board:
Ez2
dτ = z dF = w dz.
r
The total torque applied to the top half of the board is then
t /2 2
Et 3 w

Ez
τtop = dτ = w dz = .
r 24r
top 0
A torque of the same magnitude arises from the forces applied to the bottom half, so
that the total torque around point P arising from the forces exerted by the wall is
Et 3 w
τwall = 2τtop = .
12r
For static equilibrium, this must be equal in magnitude to the torque produced by the
force applied at the end of the board, Fapplied = kd. For small deflections, this torque is
τapplied = Fapplied L = kd L .
Equating the magnitudes of the torques gives

Et 3 w
kd L = ⇔
12r
Et 3 w
k= . (2.6.2)
12rd L
Now, we must find the radius of curvature r in terms of the displacement d. Recall that
we have approximated the shape of the bent board as a circle. We define the origin to
be at the center of the circle. The end of the board is at the position (x, y), where for
small displacement x ∼ = L. As we can see from figure 2.6.1d, the displacement is

L2
d =r−y ∼ = r − r 2 − L2 = r − r 1 − 2 .
r
For small displacements, the curvature is slight, that is, √
r ≫ L. So, we can approximate
the square root using a Taylor series approximation: 1 + x ∼ = 1 + x /2 for x ≪ 1.
Therefore,

L2 L2
d∼=r−r 1− 2 = . (2.6.3)
2r 2r
Substituting this into equation (2.6.2) gives
Et 3 w Et 3 w
k= 2
= . (2.6.4)
L 6L 3
12r L
2r
The full analysis, including effects of shear, gives the exact result
Et 3 w
kcantilever = . (2.6.5)
4L 3
This is proportional to 1/L 3 ; a factor of 1/L 2 comes from the fact that the board is
curved, so that the displacement is geometrically proportional to L 2 , equation (2.6.3).
The third factor of L comes from the moment arm for the torque exerted by Fapplied .
The spring constant is also proportional to t 3 w = t 2 (tw). The first two factors of t come
from the fact that the strain is proportional to the thickness, and from the moment arm
for the force from the wall which provides the stress leading to this strain. The factor
(tw) is the cross-sectional area of the board—a board with larger area of course has a
larger spring constant.
The angular frequency of vibration is

kcantilever
ω0 = . (2.6.6)
meffective
If a compact mass m is attached to the end of the cantilever, and m is much greater
than the mass of the cantilever itself, then the effective mass meffective is simply m.
(This is the same approximation we use for a conventional mass on a spring when we
neglect the mass of the spring itself.) However, in many applications there is no such
extra mass, and the effective mass is due to the cantilever itself. In this case, one can
show9 that the effective mass is
meffective = 0.243 m, (2.6.7)
where m is the mass of the cantilever.
Connection to current research: Cantilevers are widely used as sensors in Nano

ElectroMechanical Systems (NEMS). One approach is to coat the cantilever with a
“receptor” molecule, which can bind the target molecule that is to be sensed. When
this binding occurs, the effective mass of the cantilever increases, so the frequency of
vibration goes down. In an extreme example of this approach, Professor Alex Zettl and
his research group have used a carbon nanotube cantilever to detect tiny amounts of
gold which are deposited onto the cantilever. Because the mass of the cantilever itself is
so small, this system has a detection threshold close to a single gold atom! As you can see
from figure 2.6.2, the addition of 51 gold atoms produces a change in frequency that is
well above the noise level.
For further information about the topics in this chapter, see Mechanics of Materials,
4th Ed., by James M. Gere and Stephen P. Timoshenko, PWS Publishing, Boston,
1997.
terms:
Pendulum force (2.2)
Parallel axis theorem (2.2)
Young’s Modulus (2.3)
Stress, strain (2.3)
Yield stress, elastic limit, ultimate stress (2.3)
Shear stress, shear strain (2.4)
Shear modulus (2.4)
9. Dynamics of Vibrations, by Enrico Volterra and E. C. Zachmanoglou, Charles E. Merrill Books,

Columbus, OH, 1965, p. 319.
Figure 2.6.2 Deposition of gold atoms onto a cantilever made from a carbon nanotube (shown
schematically at the top) increases the mass of the cantilever, thus decreasing the vibration
frequency (left vertical scale of the graph). During the shaded time periods (e.g. 60 s to 76 s),
no atoms are deposited, so the frequency is constant. During the unshaded periods, atoms are
added, and the frequency decreases. Images Courtesy Zettl Research Group, Lawrence
Berkeley National Laboratory and University of California at Berkeley.
Rotor (2.5)
Cantilever (2.6)

A spring is cut in half (2.3)

Recognize by considerations of either force or energy when a system will oscillate,
and be able to identify the effective mass and spring constant (2.1).
Calculate the oscillation frequency for any pendulum made up from symmetrical
objects (2.2)
Calculate the oscillation frequency for extension/compression (2.3), shear (2.4),
torsional (2.5), and cantilever (2.6) oscillators
Problems
Instructors: Difficulty ratings for the problems, full solutions, and important
additional support materials are available on the website.
2.1 True or False? If true, explain. If false, give a corrected version.
If the net force on an object is zero at some position, and the object is
moved a short distance away and then released, it will then oscillate in
harmonic motion.
2.2 The Sears Tower in Chicago was the tallest building in the world for 22 years
and still holds the record for the highest antennas on top of a building. The
building itself is 442 m high. The building sways considerably in the famous
winds of Chicago; on a typical day, the top floors sway laterally by up to
15 cm, causing the toilets to slosh and occasionally giving people motion
sickness. The total mass of the tower is 2.02 × 108 kg. The average cross-
sectional area is equivalent to a square 63 m on a side. If the tower is hit by
a sudden gust of wind (which then suddenly stops), the tower is observed to
sway back and forth with a period of 8 s. Model the building as a cantilever
with square cross-section (63 m on a side) and length of 442 m. (a) If we
pretend the building is made from a uniform slab of material, what is the
Young’s modulus of this material? (b) You should have found a rather low
value, which is not surprising given that the volume of the Sears Tower is
mostly air. To get a reasonable comparison, multiply your result by the ratio
of the density of structural steel (7,850 kg/m3 ) to the average density of the
Sears Tower. You should still get a Young’s modulus which is considerably
less than that of steel, but this is reasonable since much of the weight of the
tower does not contribute to its rigidity.
2.3 Radioactive materials emit three kinds of particles. When radioactivity was
first discovered, the identity of these particles wasn’t known, so they were
named α , β , and γ particles. We now know that α -particles are helium nuclei,
that is, two neutrons and two protons combined into a nucleus, β -particles
are electrons, and γ -particles are high-energy photons. In this problem, treat
each of these as a point particle. An α -particle is fixed at the origin. An
electron is fixed at x0 = 2.00 nm. (a) A negative fluorine ion is written F− .
This is just a fluorine atom with an extra electron, and has essentially the
same mass as a fluorine atom. If the F− is placed at a position xeq , which
is to the right of x0 , the total force on it is zero. What is the value of xeq ?
Treat all three particles as point charges. (b) Make a qualitative argument
for why this is a position of stable equilibrium for the F− . (c) If the F− is
moved slightly from this position and then released, it oscillates. What is the
frequency of oscillation? (Hint: it is of order 1010 Hz. Begin by finding the
effective spring constant kspring . You are expected to find the mass of an F
atom by using the web or another reference source.)
2.4 You must do exercise 2.3 before doing this exercise. We allow the F− to
move relative to its equilibrium position xeq . (a) What is the potential energy
function U(x) for the F− , assuming it is to the right of x = 2.00 nm? (Set
U ≡ 0 at x = xeq .) (b) Using a suitable computer program, graph the potential
energy for the F− over the x-range from xeq − 1 nm to xeq + 1 nm, and
superimpose on this the graph of the harmonic (i.e., parabolic) approximation
to this potential energy using your effective spring constant from exercise 2.3.
2.5 The Lennard–Jones Potential. Two neutral atoms have an attractive
interaction due to a dynamic rearrangement of the electron clouds, called the
Van der Waals attraction. It is frequently modeled using the Lennard–Jones
potential (proposed in 1924 by Sir John Lennard–Jones):
" #
σ 12 σ 6
U (r) = 4ε − ,
r r
where r is the distance between the two atoms and ε and σ are constants. This
function is shown in figure 2.P.1. The positive term represents the repulsion
experienced at very short distances, which increases extremely rapidly as
the atoms are brought close together, while the negative term represents the
attraction, which is more important at longer distances. (a) For what value
of r is the potential energy a minimum? (Express your answer in terms
of σ .) (b) For the interaction of two argon atoms, ε = 1.654 × 10−21 J and
σ = 3.405 Å, where 1 Å (pronounced “Angstrom”) is 10−10 m, and is named
after Anders Ångström, one of the founders of spectroscopy. Assume these
numerical values are exact. What is the value of U at the minimum in this
case? (This is called the “bond energy” since it is the difference in potential
energy between the bonded atoms and the situation with the atoms very far
apart.) Quote your answer to six significant digits, and express your answer
in electronvolts, where 1 eV = 1.60217653 × 10−19 J is the amount of energy
an electron acquires as it moves through a voltage difference of 1 V. A typical
Figure 2.P.1 The Lennard-Jones potential.

covalent bond has a bond energy of 2–4 eV, so we can see that Van der Waals
bonds are much weaker. (c) What is the value of U for an r which is 0.1 Å
larger than the minimum? (Again, express your answer in electronvolts, and
quote to six significant digits.) (d) Using your numbers from parts (b) and
(c), estimate the effective spring constant k for this system? (e) If the system
is displaced from equilibrium, by pulling the atoms further apart than the
equilibrium distance you found in part (b) and then releasing them, both
atoms oscillate relative to the center of mass. This means that the center of
the “spring” that connects them doesn’t move. So, you can picture this as
each atom oscillating on a spring half as long connected to a brick wall.
Recalling that a spring half as long has twice the spring constant, what is the
frequency of oscillation? (You are expected to look up the mass of an Argon
atom on the web or other reference source.)
2.6 What is wrong with the following reasoning: For a pendulum made from
a thin rod, we can consider all the mass to be concentrated at the center
the effective length is ℓ/2, so the angular frequency is
of mass. Therefore,

g 2g
ω0 = = .
ℓ/2 ℓ
2.7 (a) If one wishes to make a pendulum with the longest possible period, how
should the mass be distributed along the length of the pendulum? Explain
your answer. (b) Your friend thinks it would be keen to make a pendulum
with a period of 1 h. Is this a practical idea? (Explain your reasoning.)
2.8 A pendulum is made from a light string that is 0.750-m long, with a small
bob of mass 0.500 kg at the end. The pendulum swings with an x-amplitude
of 5 cm. What is the energy present in this oscillation?
2.9 The tallest skyscraper (defined as having the highest occupied floor) in the
world is Taipei 101, in Taipei, Taiwan. The highest occupied floor is 439.2 m
above ground. A pendulum is suspended from this height (using a cantilever
to get the support point out laterally away from the building). The pendulum
is made from a steel rod that is 1 cm in diameter, with a 30-cm radius steel
sphere attached at the bottom. The bottom of the sphere is 10 cm above the
ground. Assuming wind forces can be ignored, what is the period of this
pendulum?
2.10 Assuming that wind forces can be neglected, what is the highest that one can
make a solid concrete tower of uniform cross-section (Take the density of
concrete to be 2,400 kg/m3 .)?
2.11 You are part of the design team for a manufacturer of air conditioning units. A
particular large unit is to be mounted on four rubber feet, one at each corner.
You are charged to decide on the size of these feet. The mass of the unit is
1,000 kg. Its weight is equally distributed on the four feet. (a) What is the
minimum area for each of the four feet? (Assume the yield stress for rubber
in compression is the same as that in extension.) To play it safe, you decide
on an area that is three times the minimum. (b) Your boss tells you, “We have
to be careful to avoid resonance with the frequency of the motors inside the
unit.” (You’ll learn more about resonance in chapter 4.) She continues, “So,
make sure the frequency of vertical oscillations is less than 5 Hz.” Does this
impose a minimum or a maximum limit on the thickness of the feet? (Assume

the mass of the rubber is negligible compared to that of the air conditioning
unit.) Explain. (c) Based on what your boss has told you, choose a thickness
for the feet. (d) If the air conditioning unit moves laterally in sinusoidal
motion, this imposes forces on the rubber that we will approximate as pure
shear. What would the frequency of such oscillations be?
2.12 For a torsional oscillator, show that, if the rotor is supported by a solid
cylinder (e.g., a wire or a string), the torque and angular frequency of
oscillation are given by

r4 π Gr 4
τ = −π G θ ω = .
spring 2ℓ wire
torsion
2Irotor ℓ
wire
(The rotor is assumed to have a much greater moment of inertia than the
wire.)
2.13 Supersolid helium At very low temperatures, helium exhibits a variety of
astounding behaviors. At atmospheric pressure, helium gas liquefies when
the temperature is reduced below 4.2 K. When the temperature is reduced
below about 2 K, the liquid becomes a “superfluid” meaning that it has zero
viscosity, so that objects can move through it with zero drag. This effect
was demonstrated experimentally using torsional oscillators. Unlike all other
materials, helium remains liquid all the way down to zero Kelvin, unless it is
placed under a pressure of at least 25 atmospheres. Above 25 atmospheres,
it does become solid. In 2004, E. Kim and M. H. W. Chan (Science 305,
1941–1944) used a torsional oscillator to demonstrate that, below about
0.23 K the solid becomes a “supersolid,” which can flow without resistance
in much the same way that the superfluid can. (It is difficult to conceive of
how a solid could behave this way, but the experimental evidence is fairly
clear.) Figure 2.P.2 shows a schematic diagram of the torsional oscillator they
used. The rotor is suspended on a hollow tube made of beryllium–copper
(BeCu), a very springy alloy. The tube has an inner diameter of 0.400 mm and
an outer diameter of 2.20 mm. The rotor itself consists of a solid cylinder of
BeCu 12.8 mm in diameter and 5.00 mm high.Around this is a hollow annular
region that is filled with the helium, and around this is a hollow aluminum
cylinder that has an inner diameter of 15.0 mm, an outer diameter of 17.0 mm,
and a height that is also 5.00 mm. A disk of BeCu that is 1-mm thick and has a
diameter of 17.0 mm is attached to the top, and a second such disc is attached
to the bottom, sealing off the annular region between the BeCu cylinder
and the aluminum cylinder. The annular region is filled with solid helium.
Take the density of BeCu to be 8,350 kg/m3 , that of aluminum to be 2,700
kg/m3 , and that of solid helium to be 172 kg/m3 . Take the shear modulus of
BeCu to be 4.83 × 1010 N/m2 . Assume all the preceding values are exact.
Assume that the moment of inertia for the rotor assembly (including top and
bottom caps, both cylinders, and the annular region filled with helium) is
much larger than that of the BeCu tube from which it is suspended. (a) When
the helium is in the normal solid state, what is the oscillation frequency f for
Figure 2.P.2 Torsional oscillator used to detect the transition to supersolid helium.
this system, calculated to five significant figures? (Recall that the moment
of inertia for a disk or solid cylinder is 21 MR2 ). (b) When the helium makes
the transition to the supersolid state, it no longer rotates along with the
rest of the rotor, so it doesn’t contribute to the moment of inertia. What is
the oscillation frequency under this condition, calculated to five significant
figures? (It was by detecting this change in oscillation frequency that Kim
and Chan demonstrated the transition to supersolid helium.)
2.14 Micro ElectroMechanical Systems (MEMS). You are designing a sensor
to detect biological pathogens in drinking water. Your sensor is based on a
cantilever which is micromachined from silicon, with a length of 300 μm, a
width of 100 μm, and a thickness of 1 μm. (Assume these dimensions are
exact.) Because silicon is a crystal, the relevant value of Young’s modulus
depends on the direction relative to the crystal axes; assume it is exactly 165
GPa for this problem. Assume the density of silicon is exactly 2,330 kg/m3 .
(a) What is the oscillation frequency f for your cantilever? (Give a result
with eight significant figures, assuming all the numbers used as inputs are
exact.) (b) You now coat your cantilever with a layer of receptor molecules,
which can bind the pathogen. Each receptor molecule has a molecular weight
of 278.1 u, where 1 u = 1.66053886 × 10−27 kg. The receptor molecules
coat all the exposed surfaces of your cantilever, each occupying an average
area of exactly (5 nm)2 . Assume that this distributed mass counts toward
the effective mass in the same way that the mass of the silicon itself counts.
What is the new oscillation frequency? (c) Each of the pathogens you must
detect has a mass of 20,000 u. Assuming that 1% of the receptor molecules
bind a pathogen, what is the new oscillation frequency? (Again, assume that
the distributed mass of the bound pathogens counts the same way that the
mass of the silicon itself counts.) (d) What is the minimum time for which
you must make a frequency measurement to detect the difference between
the frequencies in parts (b) and (c)?
3 Damped Oscillations
Decay is inherent in all component things.

Work out your salvation with diligence.
— The next-to-last thing said by the Buddha
3.1 Damped mechanical oscillators
Any real macroscopic oscillator that is displaced from equilibrium and released does not
actually show the pure sinusoidal oscillations described in chapter 1; instead, the oscil-
lations decay over time, as shown in figure 3.1.1. Fundamentally, this damping occurs
because a macroscopic oscillator is always coupled to its surroundings, even if weakly,
and the energy initially present in the oscillation leaks away through these couplings.
For example, the string of a violin is coupled to the body of the violin. If the string
is set into vibration, it causes the body to vibrate, which in turn sets the air vibrating
and broadcasts sound waves, which carry away energy. The violin body also transmits
vibrations into the violinist, providing another channel for energy to leak away.
Following the approach of physics, we start by studying the simplest possible form
of energy leakage: a drag force, also known as a “viscous damping force.” If you stick
your hand out of the window of a moving car, you feel the force of the air pushing
against you. The force is opposite to the velocity of the car and increases the faster you
go. Thus, we might reasonably expect that the force of the air is
Fdrag = −bv, (3.1.1)
where b is a constant, and the minus sign shows that the force is in the opposite direction
to the velocity.
Quantitative experiments with various objects show that equation (3.1.1) is correct,
but only if the object is moving slowly enough that the airflow around it isn’t turbulent.
The transition from nonturbulent or “laminar” flow at low speeds to turbulent flow at
higher speeds can be visualized in a wind tunnel. In wind tunnel tests, it is easier to hold
the object fixed and blow the air past it, as shown in figure 3.1.2a. At low to moderate
wind speeds, the air simply moves around the object, as shown in figure 3.1.2b. At
higher speeds, turbulent flow sets in, as shown in figure 3.1.2c, and the drag force is
no longer described by equation (3.1.1). We will discuss the turbulent regime a little
more in section 3.6.
64
Chapter 3 ■ Damped Oscillations 65
Figure 3.1.1 Decaying oscillations.
Figure 3.1.2 a: A stream of smoke is

used to test the aerodynamics of a car in a
wind tunnel. The car is stationary, and
large fans (not shown) force a flow of air
past it. The smoke flows smoothly over
the car without turbulence, indicating
laminar flow. This is desirable because
it reduces air drag, improving fuel
efficiency. (Image courtesy of NASA.)
b: A sphere in a wind tunnel. At low
speeds, the airflow is laminar. c: At
higher speeds, the flow is turbulent.
For now, we focus on the laminar flow regime. It is important in its own right,
and mathematically convenient. Further, many of the conclusions we will draw apply
qualitatively to other energy leakage mechanisms. In this regime, the drag force is
given by equation (3.1.1).
Now, we apply the same procedure as in chapter 1 to find the motion of a mass
experiencing both a restoring force from a spring and a viscous damping force.
1. Write Newton’s second law for each object in the system

There is only one object in this system, so we have
mẍ = FTotal ⇒ mẍ =

−kx!
−bv!
spring viscous
force damping
that is,
mẍ + bẋ + kx = 0 (3.1.2)
or
ẍ + γ ẋ + ω02 x = 0, (3.1.3)
Figure 3.1.3 (decaying envelope) × (oscillating function) = decaying oscillation.
where
b
γ ≡ (3.1.4)
m

characterizes the damping, and ω0 ≡ mk is the angular frequency of oscillations in
the absence of damping (i.e., in the absence of a drag force).
Following the lead of chapter 1, we now cast the problem into complex form.
Equation (3.1.3) is the real part of
z̈ + γ ż + ω02 z = 0, (3.1.5)
where z may be complex. If we can find a solution to equation (3.1.5), we automatically
get a solution to equation (3.1.3) by setting x = Re z (since the operation of taking the
real part commutes with taking derivatives).
2. Use physical and mathematical intuition to guess a solution

As suggested in figure 3.1.3, the solution might be of the form
x = [decaying envelope function] × A0 cos(ωv t + ϕ ),
where ωv is the angular frequency in the presence of viscous damping (to be
determined). It is not clear what the mathematical form of the decaying envelope
function should be. However, we can start by trying a decaying exponential, since this
is easy to handle mathematically.1 So, we’ll try
?
[decaying envelope function] = e−σ t , (3.1.6)
1. We could instead try to guess the envelope function from energy considerations. Since Work =
Force × distance, Power = Force × velocity. Therefore, the power dissipated by the damping
force is P = Fdamp v = −bv2 . The average power over a cycle is then given by Paverage = P =
% & ' 2(
dE
dt = −b v , where the angle brackets indicate the average over a cycle. Let us assume for
now that the damping is relatively light. Then, the velocity will vary approximately sinusoidally
over a single cycle. The average value of the square of a sinusoid is exactly
' half
( the maximum
of the square, as you can show by integration in problem 3.4. Therefore, v2 = 12 vmax 2 = E /m
where the constant σ is to be determined. So, our complete guess is

?
x = e−σ t A0 cos ωv t + ϕ = Re (z) ,
where z = A0 e−σ t ei(ωv t +ϕ ) (3.1.7)
3. Plug the guess back into the differential equation to see if it is indeed a
solution, and whether there are restrictions on the parameters.
Let’s prepare to plug equation (3.1.7) into (3.1.5) by computing the derivatives:
z = A0 eiϕ e(−σ +iωv ) t ,

ż = −σ + iωv z,
2
z̈ = −σ + iωv z.
Plugging into equation (3.1.5), z̈ + γ ż + ω02 z = 0, gives

2
−σ + iωv z + γ −σ + iωv z + ω02 z = 0.
Cancelling the z’s gives

2
−σ + iωv + γ −σ + iωv + ω02 = 0,
⇒ σ 2 − i2σ ωv − ωv2 − γ σ + iγ ωv + ω02 = 0.
This is actually two equations; the real part of the left side must equal the real part of
the right side, and also the imaginary part of the left side must equal the imaginary part
of the right side. So,
γ
imaginary part: − 2σ ωv + γ ωv = 0 ⇒ σ =
2
(since E = 12 mvmax
2 ). Plugging this in yields
) *
dE E
= −b ⇒
dt m
) *
dE
= −γ E . (3.1.8)
dt
If we restrict ourselves to timescales much longer than one period, we can write this as
dE dE
= −γ E ⇒ = −γ dt
dt E
Integrating both sides gives ln E = −γ t + const. ⇒ E = e-γ t +const. = econst. e−γ t . Setting
E (t = 0) ≡ E0 , we can identify the constant, so that
E = E0 e−γ t . (3.1.9)
The damping force does not affect the potential energy, so we still have U = 21 kx 2 ⇒ E = 21 kxmax
2 .
γ
Combining this with equation (3.1.9) gives xmax = 2Ek o e− 2 t . This is of the same form as
equation (3.1.6).
(in agreement with footnote 1 about guessing the envelope function from energy
considerations), and
real part: σ 2 − ωv2 − γ σ + ω02 = 0.
Substituting for σ gives

γ2 γ2
− ωv2 − + ω02 = 0 ⇒
4 2

γ2
ωv = ω02 − . (3.1.10)
4
Thus, our guess
γ
z = A0 e− 2 t ei(ωv t +ϕ ) . (3.1.11)
is indeed a solution, but only if equation (3.1.10) is satisfied. This means that x is
given by
γ
x = Re (z) = A0 e− 2 t cos ωv t + ϕ . (3.1.12)
Note that ωv is smaller than ω0 , as shown by equation (3.1.10). This is perhaps

intuitively reasonable, since the viscous damping slows down the oscillating mass.
However, for most cases of practical interest, the difference between ωv and ω0 is
small, as we’ll explore further in section 3.4. The adjustable constants A0 and ϕ are
determined by the initial position and velocity, as you can explore in problem 3.7.
3.2 Damped electrical oscillators
In section 1.5, we analyzed an LC oscillator. Now, we add a damping element to the

circuit: a resistor, as shown in figure 3.2.1. Air drag converts mechanical energy into
thermal energy; a resistor converts electrical energy into thermal energy. The voltages
across the capacitor and inductor are
q
VC = (1.5.1) and VL = L q̈. (1.5.6).
C

The magnitude of the voltage across the resistor is given by Ohm’s Law: VR = IR.
We go clockwise around the loop when evaluating voltage changes, and we also define
Figure 3.2.1 Damped electrical

oscillator.
positive current to be clockwise. Since the voltage drops across the resistor, VR = −IR.
As in section 1.5, positive Idecreases the charge on the capacitor, so I = −q̇. Therefore,
VR = −IR = q̇R.
Kirchhoff’s loop rule (which says that the sum of the voltage changes around a loop
must be zero) then gives
VL + VR + VC = 0 ⇒
1
q = 0.
L q̈ + Rq̇ + (3.2.1)
C
This is isomorphic to (i.e., identical to except for a change of symbols) (3.1.2):
mẍ + bẋ + kx = 0.
Therefore, we simply change symbols in the solution (3.1.12):
γ
q = A0 e− 2 t cos ωv t + ϕ , (3.2.2)

2
where, as before, ωv = ω02 − γ4 .
R
Your turn (answer2 below): Explain why, for the electrical oscillator, γ ≡ and
L
1
ω0 ≡ .
LC
3.3 Exponential decay of energy
The solution for the damped mass/spring is

γ
(3.1.12) : x = A0 e− 2 t cos ωv t + ϕ .
If the damping is relatively light, then γ is small, and the amplitude of oscillations
changes only slowly. In this limit, we can say that the amplitude of oscillation is
γ
A = A0 e− 2 t , (3.3.1)
so that, within each cycle the potential energy reaches a maximum value of
Umax = E tot = 12 kA2 = 12 kA20 e−γ t
⇒ E tot = E0 e−γ t , (3.3.2)
where the initial energy is E0 = 12 kA20 . This gives us a more direct interpretation of γ :
in a time γ −1 , the energy is reduced by a factor 1/e.
2. Comparing equation (3.1.2) with (3.2.1), we see that m becomes L, b becomes R, and k becomes
√
1
1/C. Making these substitutions into γ ≡ b/m and ω0 ≡ k /m gives γ ≡ RL and ω0 ≡ LC .
Concept test (answer3 below): Although the energy decays by a factor 1/e in a time
γ −1 , according to equation (3.3.1) it takes a time 2γ −1 for the amplitude to decay by
a factor 1/e. How can the energy decay more quickly than the amplitude?
Self-test (answer4 below): How long does it take for the energy to decay to 1% of
its initial value? Express your answer in terms of γ −1 .
3.4 The quality factor
In virtually all circumstances, whether in science or engineering, people don’t specify

the degree of damping by providing a value for the b that appears in Fdamp = −bv, or
for γ ≡ b/m. Instead, they quote the quality factor Q:
ω0
Q≡ . (3.4.1)
γ
Since ω0 and γ both have units of s−1 , this is a dimensionless number. It is a ratio of
the rate of oscillations to the rate of energy loss through damping. A system with low
damping has a low value of γ , and therefore a high Q.
High Q ⇔ low damping
For many applications, a large Q is desirable, because one often wants the oscillation
to persist for a long time. Also, a large Q means that the decay of oscillations is slow, so
that the actual waveform is close to a perfect sinusoid, allowing for the most accurate
timing. In a good watch, the time is determined by counting the vibrations of a quartz
crystal having a Q of about 105 . (The crystal is often in the shape of a tuning fork,
as shown in figure 3.4.1. The usual oscillation frequency is 32,768 Hz = 215 Hz. It is
easy in digital circuitry to divide a frequency into half. Doing this 15 times results in
a frequency of 1 Hz, which can be used to drive the second hand of the watch.)
However, in other applications, a low Q is preferable. For example, when the
suspension of a car is set into oscillation by driving over a pothole, the occupants
would prefer that it not keep oscillating for a long time! The suspension of a typical
car has a Q of about 1.
√
3. Since E = 12 kA2 , the amplitude need only decrease by a factor of 2 in order for the energy to
decrease by a factor of 2.
= 4.6γ −1 .
4. ln (100) γ −1 ∼
Figure 3.4.1 The quartz crystal

oscillator from a watch must
have a high Q, so that its
oscillation frequency is very
well-defined. (Image courtesy
of and © Dr. Erhard Schreck.)
We can re-express the angular frequency in the presence of damping using Q.

From equation (3.1.10), we have

γ2 γ2
ωv = ω02 − = ω0 1 −
4 4ω02

1
⇒ ωv = ω0 1 − . (3.4.2)
4Q2
Thus, we can see that if Q is large, then ωv ∼
= ω0 . To make a more quantitative statement
about this, you must first show the following:
Your turn: For a system with light damping (for which E = E0 e−γ t and ωv ∼
= ω0 ),
show that
E = E0 e−n2π /Q , (3.4.3)
where n = t /T is the number of oscillation cycles in time t, and the period is T = 2π /ω0 .
So, if Q = 2π , then the energy decays by about a factor of e per cycle of oscillation.
This is fairly heavy damping by most people’s standards, and yet equation (3.4.2) tells
us that ωv is only about 0.3% smaller than ω0 for this case! So, it many circumstances,
one can ignore the difference between ωv and ω0 .
Connection to current research: There are some applications that depend on detecting
the small difference between ωv and ω0 . For example, the micromachined cantilever
shown in figure 3.4.2 can be used to detect the pressure of gas or the molecular weight of
the gas in its surroundings. Higher gas pressure, and gases with higher molecular weight,
produce more damping and so a greater shift in the oscillation frequency. (A maximum
frequency shift of 7% was observed for one atmospheric pressure of Argon gas, as
compared to the frequency in vacuum.)
Figure 3.4.1 A micromachined silicon

cantilever used to detect the pressure or
molecular weight of gas via the change in
the oscillation frequency due to damping.
The cantilever is produced by a
combination of photolithography and
etching (“micromachining”) and is
suspended a few μm above the substrate,
so that it can oscillate in a plane parallel
to the substrate, as shown in the
schematic diagram. Top image reprinted
with permission from Y. Xu, J. –T. Lin,
B. W. Alphenaar, and R. S. Keynton,
Appl. Phys. Lett. 88, 143513 (2006).
Copyright 2006, American Institute of
Physics.
3.5 Underdamped, overdamped, and critically damped behavior
Concept test: (Please cover the bottom part of the page, below this box, so thatyou don’t
1
see the answer right away.) What is disturbing about equation (3.4.2): ωv = ω0 1 −
4Q2
if you consider the behavior at large damping?
What’s disturbing is that, if Q is less than 1/2 , then ωv is imaginary. Since low Q
corresponds to heavy damping, we can definitely make Q less than 1/2. This limit is
called “overdamping.” (Damping with Q > 1/2, which is of the most interest to us, and
corresponds to the damped oscillations discussed previously in this chapter, is called
“underdamping.”) The math that we went through in section 3.1 works fine even if ωv
is imaginary, so the solution for the overdamped case is still x = Re z, where
γ
(3.1.11): z = A0 e− 2 t ei(ωv t +ϕ ) .
It is revealing to rewrite things so that the imaginary nature of ωv (for Q < 1/2) is
shown explicitly:

1 1 1
ωv = ω0 1 − 2
= ω0 (−1) 2
− 1 = ±iω0 − 1.
4Q 4Q 4Q2
We define

1
β ≡ ω0 −1, (3.5.1)
4Q2
which is real if Q < 1/2. So, ωv = ±iβ . Because of the ±, plugging this into
equation (3.1.11) above gives two possible solutions:
γ γ γ
z1 = A1 e− 2 t ei (+iβ t +ϕ1 ) = A1 eiϕ1 e− 2 t e−β t and z2 = A2 eiϕ2 e− 2 t e+β t .
As discussed in footnote 3 of section 1.3, the general solution of a linear, homogeneous

second order differential equation, such as the differential equation (3.1.5): z̈ + γ ż +
ω02 z = 0 that describes the damped oscillator, is the sum of two independent solutions,
such as z1 and z2 above. Therefore, the most general solution is:
γ

z = z1 + z2 = e− 2 t A1 eiϕ1 e−β t + A2 eiϕ2 e+β t .
The actual behavior is

γ
x = Re z = e− 2 t A1 cos ϕ1 e−β t + A2 cos ϕ2 e+β t .
We define B1 ≡ A1 cos ϕ1 and B2 ≡ A2 cos ϕ2 , so that finally

γ
x = e− 2 t B1 e−β t + B2 e+β t . (3.5.2)
Overdamped behavior (i.e., behavior for Q < 1/2)
The constants B1 and B2 are determined by the values of x (t = 0) and ẋ (t = 0). Note
that there is no oscillation involved. Therefore, the system can cross the equilibrium
point at most once, as shown in figure 3.5.1a. (You can show this rigorously in
problem 3.14.)
The case Q = 1/2 is called “critical damping.” It is only of mathematical interest,
since in any real system Q would never exactly equal 1/2, but would always be at least
slightly greater or slightly less. However, for thoroughness, and because the name
“critical damping” makes it sound as though it ought to be important, we discuss it
briefly.
Again, the math from section 3.1 is unchanged, so that

x = Re A0 e− 2 t ei (ωv t +ϕ ) .
γ

1
However, now ωv = ω0 1 − = 0, so that
4Q2
γ
γ
x = Re A0 e− 2 t ei ϕ = A0 cos ϕ e− 2 t .
Figure 3.5.1 a: In the overdamped case, there is at most one crossing of the equilibrium point;
even this can only occur if the initial velocity is high and the initial position is close to x = 0.
b and c: Comparison of underdamped and critically damped systems. In part b, the initial
velocity is zero but the initial position is nonzero. In part c, the initial position is zero but the
initial velocity is non-zero. In both cases, the underdamped system (Q = 0.6) returns to
equilibrium a little more quickly than the critically damped system.
We define B = A0 cos ϕ , so that

γ
x = B e− 2 t , (3.5.3)
Where B is determined by the values of x(t = 0) and ẋ (t = 0).
Concept test: (Please cover the bottom part of the page, below this box, so that you
don’t see the answer right away.) Something about that last phrase should trouble you.
What is troubling about it?
As you’ve probably realized, we can’t get the desired values of both x(t = 0) and
ẋ(t = 0) with just the one adjustable constant B. Therefore, equation (3.5.3) can’t be
the complete solution for the critically damped case. In problem 3.15, you can show
that the complete solution is
γ γ
x = B e− 2 t + Dt e− 2 t . (3.5.4)
Critically damped behavior (i.e., behavior for Q = 1/2)
It is sometimes claimed that, if an engineer wants to design a system that moves

from a displaced value to equilibrium as quickly as possible, then s/he should make
it critically damped. This is incorrect, although it is not far from the truth, as shown
in figure 3.5.1b and c—an underdamped system with a Q a little above the critically
damped limit returns more quickly to equilibrium.
3.6 Types of damping
As we discussed in section 3.1, for low speeds the drag force is given by
(3.1.1) : Fdrag = −bv,
where v is the speed. For a sphere of radius r,
bsphere = 6π r μ, (3.6.1)
Stokes’ Law
where μ is the viscosity of the surrounding fluid. This relation was derived by George
Stokes in 1851.5
Recall, as discussed briefly in section 3.1, that the flow of the fluid around a moving
object is laminar at low speeds, and turbulent at higher speeds, as shown in figure 3.6.1.
5. Stokes made important contributions in mathematics and physics, including central results in
fluid dynamics, the first description of fluorescence, and early ideas that were close to the mark
in correctly explaining spectroscopic lines. Stokes’ theorem, a central result in vector calculus, is
named for him, although it was derived by Lord Kelvin. (It was associated with Stokes because
during oral exams he would ask students to derive it.) However, Stokes was himself generous in
sharing credit with Kelvin on other fronts, so perhaps this misnomer reflects his good karma.
Table 3.6.1. Viscosities of common fluids
Fluid μ (kg m−1 s−1 )
Air 1.784 .× 10−5

Water 1.004 .× 10−3
Light motor oil ∼ 0. 1
Honey ∼ 10
Peanut butter ∼ 250
Figure 3.6.1 Smoke from an incense stick

initially shows laminar flow, but when it
has accelerated to a high enough velocity as
it rises, it transitions to turbulent flow.
(Image © Matt Trommer/Dreamstime.com)
The conditions for this transition are determined by the Reynolds number:
ρvL
Re ≡ , (3.6.2)
μ
where L is the characteristic length and ρ is the density of the fluid. (For flow in a pipe,
L is the pipe diameter. For a sphere moving through a liquid, L is the diameter of the
sphere.) Re is a dimensionless number.6 For Re < 2,000, the flow is laminar, while for
6. The number is named for Osborne Reynolds (1842–1912). There is a wonderful anecdote told by
Sir J. J. Thomson (who discovered the electron), a student of Reynolds at Manchester University:
“Occasionally in the higher classes he would forget all about having to lecture and, after waiting
for ten minutes or so, we sent the janitor to tell him that the class was waiting. He would come
rushing into the room pulling on his gown as he came through the door, take a volume of Rankine
[a standard textbook of the time] from the table, open it apparently at random, see some formula
or other and say it was wrong. He then went up to the blackboard to prove this. He wrote on
Re > 4,000 the flow is turbulent. For a baseball moving through air, the transition thus
occurs at about 0.6 m/s. Be careful not to confuse the notation Re for Reynolds number
with Re meaning “real part of.” Usually the correct interpretation will be obvious from
context.
In turbulent flow, Fdrag ∝ v2 , as you can show in problem 3.21. The fact that the
force depends on the square of the velocity makes the differential equation describing
the motion of an oscillator experiencing such a force impossible to solve analytically,
although the motion can be solved by numerical methods. To be specific, in that
problem, you can show that in turbulent flow
Cd 2
Fdrag = Av ρ, (3.6.3)
2
where A is the cross-sectional area. The “drag coefficient” Cd varies from about 0.03
for a high-performance jet airplane to about 0.3 for an automobile to about 1.1 for a
person standing up. (In fact, the drag coefficient is not really constant, and depends
somewhat on the velocity.)
There are a number of other mechanisms which can suck energy out of an
oscillation, and so provide damping. A good example is friction, which provides a
force essentially independent of velocity. This can be addressed analytically, but it is
surprisingly messy, and is beyond the scope of this text. Another type of damping comes
from atomic-scale motions within the oscillator, the details of which are a subject of
current research. Because of these motions, if you put a mass/spring system in a vacuum
and set it oscillating, the motion eventually damps out, although all fluid damping has
been eliminated. Although important aspects of the behavior of damped oscillators
vary from one type of damping to another, we can still use the results obtained for
viscous damping in this chapter and the next as a qualitative guide.
terms:
Viscous drag (3.1, 3.6)
Laminar flow (3.1, 3.6)
Turbulence (3.1, 3.6)
Envelope function (3.1)
the board with his back to us, talking to himself, and every now and then rubbed it all out and
said it was wrong. He would then start afresh on a new line, and so on. Generally, towards the
end of the lecture, he would finish one which he did not rub out, and say this proved Rankine
was right after all. This, though it did not increase our knowledge of facts, was interesting, for it
showed the workings of a very acute mind grappling with a new problem.” (From Recollections
and Reflections, by Sir J. J. Thomson, Macmillan, New York, 1937, p. 15, cited in Eurekas and
Euphorias: The Oxford Book of Scientific Anecdotes, by Walter Gratzer, Oxford University Press,
Oxford, England, 2004, p. 333.)
Exponential decay (3.1)

Quality factor (3.4)
Underdamped, overdamped, critically damped (3.5)
Reynolds number (3.6)
Drag coefficient (3.6)

Damping and coupling to other parts of the universe (3.1)
Resistors and air resistance (3.2)
Damped electrical oscillators and damped mechanical oscillators (3.2)
Q and damping (3.4)
Laminar versus turbulent flow and v versus v2 damping (3.1, 3.6)

A single complex equation contains two equations, one for the real part and one for
the imaginary part (3.1)

Find the amplitude or energy of a damped oscillator as a function of time (3.1–3.2)
Calculate the speed at which the drag force changes from being proportional to v to
being proportional to v2
Problems
3.1 Consider two identical oscillators of the type shown in figure 1.4.1a. For one
of the oscillators, the surface is frictionless, but there is viscous damping
Fdrag = −bv. For the other, the surface has coefficient of kinetic friction
μk , but there is no viscous damping. The masses are both pulled away from
equilibrium the same distance to the right and then released. When they first
reach the equilibrium position, the magnitude of the viscous damping force
for the first oscillator is equal to the magnitude of the frictional force for
the second oscillator. Which oscillator damps down to one tenth the initial
amplitude more quickly? Explain your reasoning.
3.2 Eddy current damping. A changing magnetic field B creates an electro-
dφ $
motive force ε = − B , where φB ≡ B · n̂ dA is the magnetic flux. If a
dt
changing field is applied to a bulk piece of metal, the resulting current pattern
is complicated, since the current isn’t confined to simple geometries by
wires. Such currents are called “eddy currents,” and are used to provide a
damping force for vibration isolation in sensitive scientific equipment. They
are also used to provide a braking force for large trains, resulting in less wear
on the main brakes. When the train engineer throws the appropriate lever,
permanent magnets are brought close to the rails, without touching them.
dφ
The motion of the train then causes a B through any loop within the rail.
dt
dφB
Thus, is proportional to the train speed; we will take the proportionality
dt
dφ
constant to be α , so that B = α v. The current that flows within the rail is
dt
dφB
proportional to ε = − ; we will take the proportionality constant for the
dt
average current I to be R: ε = IR. What is the damping constant b in terms
of α and R?
3.3 Exponentials versus Power law. Using a suitable computer program,
superpose plots of the functions e−t /τ and t −n . Choose whatever value you
like for the constants τ and n. (Suggestion: the plots are easier to make if
you choose n to be no greater than 20.) Try to make the power law (t −n )
decay faster than the exponential. Make two plots: in the first, let the vertical
axis range from 0 to 2. In the second plot, choose the scales for the axes to
show that, even though the power law may initially decay faster than the
exponential, the exponential eventually always catches up and falls below
the power law.
3.4 Show that the average value of the square of a sinusoid (averaged
over one cycle) is exactly half the maximum value (of the squared
sinusoid).
3.5 Time dependence of the dissipated power. For a damped harmonic
oscillator, use a suitable computer program to superpose plots of x (t) and
the power dissipated as a function of time. (Choose whatever values you
like for the relevant parameters, and make sure your plot covers at least
one full period.) Comment qualitatively on why your plot looks the way
it does.
dB
3.6 A bank account earns interest according to = gB, where B is the account
dt
balance and g is a constant.
(a) Write an equation for B(t) (i.e., B as a function of time) in terms of
B0 (the value of B at t = 0) and g.
(b) If B0 = $100.00, and after 1 year B has increased by 5.00%, what is
the value of g? Hint: quote g in units of year −1 . In financial terms,
g is the “continually compounded interest rate,” while 5% is the
“Annual Percentage Rate” or APR.
3.7 We found in section 3.1 that the position of a damped oscillator is given
γ
by x = A0 e− 2 t cos(ωv t + ϕ ). Without using a symbolic algebra program
or calculator, find A0 and ϕ in terms of the initial position x0 , the initial
Figure 3.P.1 An RLC circuit suddenly has a nonzero voltage applied to it.
B
velocity ẋ0 , ωv , and γ . Hint: You should be able to show that tan ϕ = ,
D
where B and D are constants involving x 0 , ẋ0 , ωv , and γ . To find A0 , you
will need to find cos ϕ . To do this, draw a right triangle with B and D as the
two legs.
3.8 For the circuit shown in figure 3.P.1, at times t < 0 the output of the voltage
supply is set to 0 V, there is no current flowing, and there is no charge on the
capacitor. At t = 0, the voltage supply is suddenly changed to a voltage V0 ,
as shown, and remains constant thereafter.
(a) Show that, for t > 0, the charge q on the capacitor is described by
the following differential equation:
R 1 V
q̈ + q̇ + q = 0.
L LC L
(b) Show that
q = A + Be−α t cos(ωt + ϕ )
is the solution to the above differential equation, and determine the

values of A, α , and ω in terms of V0 , R, L, and C.
Hints:
1. Cast the problem into complex form.
2. If f(t) is a function of time, and f (t) + G = H, where G and
H are constants, then we must have G = H and f (t) = 0.
(Otherwise the equation f (t) + G = H could not be satisfied
for all times.)
3. You may, if desired, use exact analogies if they are appro-
priate in solving this problem. It is not necessary for you to
use analogies to get the answers, and the analogy approach

might not give answers for all three quantities A, α , and ω. If
you choose to use the analogy approach, you should explain
very clearly the analogy you’re using.
(c) Because the inductor will not allow the current to change discon-
tinuously, we must

have q̇0 = 0. Use this initial condition to show
− R
that ϕ = tan−1 .
2ωL
(d) It is clear from the above that ϕ → 0 or π when the damping is
light. Let’s choose ϕ → 0. In this light damping limit, use the fact
that q0 = 0 to show that B → −CV0 .
(e) Sketch q(t) for fairly light, but nonzero damping. Your sketch can
be qualitative.
3.9 Show that, perhaps surprisingly, the time between displacement maxima for
2π
a damped oscillator is exactly , independent of the degree of damping
ωv
(so long as the underdamped limit applies, that is, so long as equation (3.1.12)
describes the solution).
3.10 At t = 0, a particle of mass m attached to a spring of spring constant
k is at rest a distance x = A0 away from its equilibrium position. It is
released, and begins oscillating. The system is immersed in fluid which
leads to a damping force of the form Fdamping = −bẋ. You may assume the
damping is light. Find the time for the envelope of the oscillations to drop to
A0 /10.
3.11 Thermal vibrations.
The “Equipartition Theorem” of statistical mechanics tells us that thermal
fluctuations impart an average potential energy of 1/2kB T to a harmonic
oscillator and on top of this also impart an average kinetic energy of 1/2kB T ,
giving a total thermal energy of kB T , where kB = 1.38 × 10−23 J/K is
Boltzmann’s constant, and T is the absolute temperature of the oscillator.
A damped harmonic oscillator has mass m, spring constant k, and quality
factor Q, which is greater than 1/2 . (a) Explain how, by measuring the
amplitude of these thermal vibrations and the temperature, one can determine
the spring constant of a mass/spring system for which both the mass and
the spring constant are unknown. Make your explanation as quantitative
as possible, remembering the word “average” which appears above. This
method is used in atomic force microscopy to determine the spring
constant of microfabricated cantilevers. These cantilevers can then be
used to quantitatively measure the forces of interaction between individual
molecules. (b) A damped harmonic oscillator has mass m, spring constant k,
and quality factor Q, which is greater than 1/2. It is set into motion with an
initial amplitude A0 . In principle, the mass crosses the equilibrium point an
infinite number of times. However, the damping eventually causes the motion
to become smaller than random thermal motions. In terms of m, k , Q, A0 ,

kB , and T about how long does this take?
3.12 For a damped harmonic oscillator, show that if Fdamp = −bẋ β , then the
energy as measured on timescales longer than an oscillation period only
decays exponentially if β = 1. Assume light damping. You may also assume
that Paverage = GPmax , where P is the dissipated power and G is a constant.
Hint: If you assume that E decays exponentially, what differential equation
involving Pav = Ė must be obeyed? Then, in this equation, write E in terms
of the maximum kinetic energy. Finally, develop a second equation for Ė
using Fdamp = −bẋ β .
3.13 Damping by radiation. From classical electromagnetism, one can show
that the instantaneous power of electromagnetic radiation that is emitted
q 2 a2
by a point charge q undergoing acceleration a is P = , where c =
6π ε0 c3
2.9979 × 108 m/s is the speed of light and ε0 = 8.8542 × 10−12 C2 /Nm2 is the
permittivity of free space. Consider an electron that oscillates in harmonic
motion about a stable position with frequency f and amplitude A. It is held
close to this position by forces from other charges. As usual, we assume that
the net restoring force has a magnitude proportional to displacement away
from equilibrium. In this problem, you’ll show that the energy radiated away
provides an effective viscous damping force.Assume that the energy radiated
by the electron in one cycle is much less than the energy in the oscillator, so
that over one cycle the motion is approximately sinusoidal. (a) Show that the
4π 3 e2 A2 f 3
energy radiated in one cycle is Ecycle = . (b) Show that therefore
3ε0 c3

E
∝ E, where
E is the energy lost in one cycle and
t equals one period.

t
(c) By comparison with equation (3.1.8), which appears in footnote 1, show
that the energy decays exponentially, just as with a viscous damping force,
and find the time for the energy to decrease by a factor of e. (d) Show that
3ε mc3
Q∼ = 02 , where m is the mass of the electron.
e f
3.14 Show that an overdamped oscillator can cross the equilibrium point one
time at most.
γ
3.15 Show that x = Dte− 2 t is a solution to equation (3.1.3) for the case of critical
γ
damping. (In section 3. 5, we showed that x = Be− 2 t is also a solution, so
γ γ
the general solution is x = Be− 2 t + Dte− 2 t .)
3.16 For a given object moving through air or water, is the speed needed to
produce turbulence higher in air or water? Explain.
3.17 Consider a round rock of radius r in a shallow stream. For about what velocity
of the stream should the flow around the rock become turbulent, according
to the ideas discussed in this chapter? Is this velocity consistent with your
experience?
3.18 You are designing a chemical processing plant, in which water from a pipe
is used to fill a spherical reaction vessel that is 2 m in diameter. If there is
turbulence inside the pipe, then the required pressure to force the water
through is significantly higher, so this is to be avoided. If the reaction

vessel must be filled in less than 100 min, what is the minimum pipe
diameter?
3.19 Calibration of laser tweezers. As you may recall from an electricity and
magnetism course, matter is attracted to regions of high electric field. (The
atoms in the matter are polarized by the field, so that each atom forms a
dipole. The two ends of the dipole experience fields of different strength,
leading to a net force.) You may also recall that light consists of rapidly
oscillating electric and magnetic fields. Combining these two ideas, we see
that there is an attractive force which draws matter into a region with the
most intense light. For example, if you shine a laser beam into a drop of water
in which there are small plastic spheres suspended, the spheres are attracted
into the beam. This is called a “laser tweezers apparatus.” By moving the
beam around, you can drag one of these spheres wherever you like. You
can also grab a living cell that is floating around in the water, and drag it
around without harming it. By combining two laser tweezers with a more
powerful laser that can cut through the cell wall (“laser scissors”), you can
fuse two cells together. Going back to the plastic beads, you can attach one
sort of molecule to the surface of a bead, allow it to bind to another sort of
molecule on a different bead or on a solid surface, and measure the strength
of the binding by tugging with the laser tweezers. In order to perform this
experiment, you must calibrate the strength of the force applied to the bead
by the laser tweezers. Since the tweezers hold the bead in a position of stable
equilibrium, the force applied to the bead by the laser can be approximated
by Hooke’s law, F = −kx, where x is the displacement of the center of the
bead away from the center of the laser beam. One way to find the spring
constant x is to flow water (or another fluid) past the bead at a known speed
v0 and measure the resulting displacement x0 of the bead. Find an expression
for k in terms of x0 , v0 , the viscosity μ of the surrounding fluid, the density
ρ of the fluid, and the radius r of the bead, assuming the flow remains
laminar.
3.20 (a) Explain how our guess for the complex version of the solution of the
damped oscillator, (3.1.7): z = A0 e−σ t ei(ωv t +ϕ ) could be written in the form
z = Ceαt , where C and α are both complex. (b) Show that z = Ceαt is not
a solution for the damped oscillator in the turbulent regime, where (3.6.3):
C
Fdrag = d Av2 ρ .
2
3.21 Drag force for turbulent flow. (a) It is easier to think about drag forces
in a reference frame in which the object is stationary and the fluid medium
(air, water, etc.) is moving past it. In laminar flow, the fluid moves around
the object with relatively little net disruption to its velocity. However, in
turbulent flow, the velocity of the fluid is changed violently. Consider the
volume of fluid that will impinge on the object in a time
t, as shown in
figure 3.P.2. This has a length v
t, and a cross-sectional area A equal to that
of the object. Assume that a fraction Cd /2 of the momentum of this volume of
fluid is transferred to the object. (Cd is called the “drag coefficient.”) Show
Figure 3.P.2 In time

t, all the air in the cylinder will hit the front of the spherical object.
Cd 2
that the resulting force on the object has a magnitude Fdrag = Av ρ ,
2
where ρ is the density of the fluid. (b) For an airplane, Cd ≈ 0.03. What
power is required from the engines for an airplane with A = 15 m2 to fly
at a speed of 150 kph near sea level, where the density of air is 1.2 kg/m3 ?
(c) Repeat part (b) for a speed of 300 kph.
4 Driven Oscillations and Resonance
When you drive an oscillator at its resonant frequency

Then the amplitude of the oscillation will become huge.
In the equation above, it becomes infinite,
But in practice there will be some damping
That prevents that.
You have known this since your childhood,

This is how you swing on a swing.
If you live in a snowy climate, you know
(or at least should know) that a trick
To get your car out of a snow bank is
To rock it back and forth—
If you get the frequency right you will make the car oscillate
With a large amplitude
And dislodge it.
The electrical analogue is used to tune a radio.

From “The Driven Oscillator” (not originally intended to be a poem)
by Professor B. Paul Padley, Rice University
4.1 Resonance
As we saw in chapter 3, the oscillations of a macroscopic oscillator decay over time

because the energy leaks out into the surroundings. For an oscillation to be sustained,
this energy loss must be balanced by the energy added to the oscillator. We see examples
of this all the time. A child on a swing uses her legs to pump the amplitude higher and
higher. The seat in a bus vibrates, because it is driven by the engine which is shaking
in a regular way. A washing machine shakes, especially in the spin cycle; in this case,
the energy is provided by the motor which spins the tub with the clothes in it. The
current inside a radio receiver oscillates, driven by the radio waves broadcast by the
transmitter. The straps on your backpack, and your hair (if it’s long enough) swing
back and forth as you walk, driven by the energy you put into moving your legs (which
also moves your body up and down a little with each step).
84
Chapter 4 ■ Driven Oscillations and Resonance 85
The behavior of oscillators that are driven, including the above examples, is
perhaps the single most important idea in all of physics. Essentially every area of
physics has strong connections with driven oscillators, and many seemingly unrelated
phenomena can be understood qualitatively by analogy with driven oscillators.
You have probably noticed that if an oscillator is driven with a periodic force of
just the right frequency, the motion of the oscillator becomes very large, much larger
than for other frequencies of drive force. For example, if you have an old car that rattles
sometimes, you may have noticed that the rattle is worst when the engine is running at
a particular speed—for lower speeds or for higher speeds the rattle is less noticeable.
This phenomenon of a strong response at a certain frequency is called “resonance.” It
can be annoying and even destructive, leading to failure of mechanical components,
but it can also be used to almost magical effect in the design of ultrasensitive detectors,
and in medical imaging, as shown in figure 4.1.1.
Following the approach of physics, we begin with the simplest possible driven
oscillator: a mass on a spring. However, by now you recognize that this can represent
a vast array of physical systems, including electronic circuits (which we’ll explore
later in this chapter). We will also assume to start that the energy injected into the
oscillator comes from the simplest possible periodic source: a sinusoidal driving force.
However, this is actually no restriction at all, since we will show that any periodic
driving force can be represented as a sum of sinusoids, and that the response of the linear
Figure 4.1.1 Magnetic resonance images of a human brain. (Image © Katrina

Brown/Dreamstime.com)
Figure 4.1.2 An unbalanced washing machine (top view).
oscillators that are our main focus is simply the sum of the responses to the individual
sinusoids.
Furthermore, many real-world driving forces do have a simple sinusoidal form.
Many driving forces arise from a circular motion. For example, consider again the tub
of a washing machine, in which the clothes are unevenly distributed. The position of the
heaviest part of the clothes is shown by the dot in figure 4.1.2, and the angle of a line to
this point relative to horizontal is θ . If the tub rotates at constant angular velocity ω, then
θ = ω t. The horizontal position of the bunch of clothes is then x = r cos θ = r cos ωt,
so as the tub rotates, this creates a sinusoidal driving force in the x-direction.
For a mass hanging on a spring (figure 4.1.3), one way to apply the driving force
is to move the support point for the spring sinusoidally. The force from the spring is
clearly related to the motion of x relative to xc , i.e.,

Fspring = −k x − xc
(e.g., if x and xc are both shifted upward by the same amount, the spring force should
be zero.) We move xc sinusoidally with amplitude Ad and angular frequency ωd ,
where subscript “d” indicates “drive.” For example, if xc = Ad cos ωd t, then we get a
sinusoidal driving force:

Fspring = −k x − xc = −kx + kAd cos ωd t = −kx! + F0 cos ωd t
!
usual sinusoidal
spring driving
force force
with
F0 = kAd . (4.1.1)
Figure 4.1.3 A sinusoidal driving force can be applied by moving

the support point xc sinusoidally. The position x of the mass is
defined relative to the equilibrium position, as indicated by the short
horizontal line next to x. Similarly the position of the support point,
xc , is defined relative to its own equilibrium position, as indicated
by the short horizontal line next to xc .
Before starting our quantitative analysis, let’s qualitatively consider what to

expect. If the frequency of the drive is very low, then the mass should simply move
up and down together with the support point, that is, x = xc = Ad cos ωd t. On the
other hand, if ωd is very high, then the sign of the drive force keeps changing at a high
frequency. So, the drive force doesn’t have much time to accelerate the mass before the
sign of the drive force changes. So, for large ωd , we expect a small motion of the mass.
Somewhere between low ωd and high ωd we expect to find a resonant response,
that is, a special frequency of drive force for which the mass oscillates with large
amplitude.
The damped oscillator without a drive force is just a special case of the damped
driven oscillator, with zero drive amplitude, that is, F0 = 0. Thus, the full solution
for the oscillator with both driving and damping should include the possibility of
an oscillation at angular frequency ωv that decays away exponentially and has an
amplitude and phase that depend on the initial position and velocity. However, for
times well beyond t = 0, it is reasonable to expect that the effect of these initial
conditions will decay away, and that we will see a “steady-state” oscillation in which
the energy per cycle delivered to the system by the drive equals the energy per cycle
that leaks away from the system because of the damping. This “steady-state solution”
is actually the one of most interest, since it persists indefinitely.
We consider a general sinusoidal drive force
Fdrive = F0 cos ωd t , (4.1.2)
which might be applied by a motion of the support point (equation (4.1.1), or might
be applied in some other way.
To find the behavior of the system, x(t), we again follow our three-step procedure:
1. Write Newton’s second law for each object in the system:
mẍ = −kx − bẋ + F0 cos ωd t ⇔

mẍ + bẋ + kx = F0 cos ωd t ⇔ (4.1.3a)
F0
ẍ + γ ẋ + ω02 x = cos ωd t . (4.1.3b)
m
1b. Cast the DEQ into complex form
Following the lead given by previous chapters, we write the simplest complex
differential equation for which equation (4.1.3b) is the real part:
F0 iωd t
z̈ + γ ż + ω02 z = e (4.1.4)
m
with x = Re(z).
2. Guess a solution, based on physical and mathematical intuition
Driven oscillators appear to show a regular motion, at least once the steady-state
discussed earlier has been achieved. It is not obvious whether the angular frequency
of this motion would be that of oscillation without damping or driving (ω0 ) or that
of oscillation with damping but no driving (ωv ), or that of the drive (ωd ) , or some
combination of these. So, for now, we write the angular frequency of the motion in
steady-state as ωs . To be as general as possible, we include a phase factor, and of course

an amplitude. Thus, our guess is
?
z = Aei(ωs t −δ) .
Here, the phase factor is written as −δ , rather than, for example, +ϕ . We do this
because it will turn out that δ as defined this way is always positive, though this is not
yet obvious. The above guess can be rewritten
?
z = Ae−iδ eiωs t . (4.1.5)
3. Substitute the guess into the DEQ to see if it is a solution, and if there are
restrictions on the parameters
Your turn: Plug our guess (4.1.5) into (4.1.4) to show that
? F 0 i ωd t
(−ωs2 A + iγ ωs A + ω02 A)e−iδ eiωs t = e . (4.1.6)
m
We see that the left side oscillates at an angular frequency ωs , while the right side
oscillates at ωd . So, if they are to be equal, we must have ωs = ωd . In other words
(assuming other aspects of our guess turn out to be correct):
In the steady-state, the oscillator moves with the same angular frequency as
the drive.
This is perhaps an unexpected result. In the steady-state, the system does not move
with its “natural” angular frequency ω0 or at ωv , but rather it moves at ωd .
So, our guess now becomes
?
z = Ae−iδ eiωd t , (4.1.7)
and is shown graphically in figure 4.1.4. But we haven’t yet shown that our guess really
works. Since ωs = ωd , equation (4.1.6) becomes
F0 iωd t ?
(−ωd2 A + iγ ωd A + ω02 A)e−iδ eiωd t =e .
m
? F
⇒ (−ωd2 A + iγ ωd A + ω02 A)e−iδ = 0 .
m
? F
⇔ (ω02 − ωd2 )A + iγ ωd A = 0 eiδ
m
Figure 4.1.4 Our guess represented in the

complex plane. Note that, since the units of z and
F0 are different, the length of the two vectors
cannot be compared.
For this equation to hold, the real part of the left side must equal the real part of the
right side, and also the imaginary part of the left side must equal the imaginary part of
the right side:
F0
?
Real: (ω02 − ωd2 )A = cos δ. (4.1.8)
m
? F
Imaginary: γ ω d A = 0 sin δ. (4.1.9)
m
To make these equations true, we will need particular values of A and δ . To isolate A,
we square these two equations and add them, giving
F0 /m
A = A ωd = , (4.1.10)
(ω02 − ωd2 )2 + (γ ωd )2
in which we emphasize that A is a function of ωd . To isolate δ , we instead divide

equation (4.1.9) by (4.18), giving
γ ωd
tan δ ωd = , (4.1.11)
ω02 − ωd2
again emphasizing that δ is a function of ωd .1

So, our guess (4.1.7) is indeed a solution of the differential equation (4.1.4)! (But
only if A and δ are as given above.) Of course,
x = Re (z) = A cos (ωd t − δ ). (4.1.12)
We can see from equation (4.1.10) that the amplitude depends on the angular frequency
of the drive, ωd . It might appear from the equation that the maximum amplitude occurs
γ ωd
1. It is correct to write δ ωd = tan−1 , however recall that one must be careful with the
ω02 − ωd2
tan−1 function. For a negative argument, your calculator (or Mathematica or other equivalent
program) returns a negative value for the tan−1 . Since we want δ to be positive, we must add π
to such a result.
Figure 4.1.5 Amplitude and phase of a

damped driven oscillator as a function of the
angular frequency of the drive.
at ωd = ω0 ; this is almost right, but the actual maximum is at a slightly lower value2 of
ωd . However, the difference is quite small, except for heavy damping, and is usually
unimportant. (We shall discuss this in more detail in section 4.2.) This maximum
amplitude is the resonance discussed earlier in this section.
We also see from equation (4.1.11) that the phase δ by which the oscillator’s
response lags behind the drive force also depends on ωd . When ωd = ω0 , equation
(4.1.11) becomes tan δ → ∞, so that δ = π/2. The dependencies of A and δ on ωd are
shown in figure 4.1.5.
We see from equation (4.1.10) that the response amplitude at high frequencies
approaches zero, as anticipated in our initial qualitative discussion. In the opposite
limit, ωd → 0, equation (4.1.10) reduces to
F /m F
A ωd → 0 = 0 2 = 0 .
ω0 k
If the drive force is applied by moving the support point, then, using equation (4.1.1):
F0 = kAd , this becomes

A ωd → 0 = Ad , (4.1.13)
which is also as we anticipated.
2. To see this, we rewrite equation (4.1.10):
F 0 /m 1 F0 /m
A= = .
2 2 2 2 ω ( 2 − ω2 )2
( ω 0 − ω d ) + ( γ ωd ) d ω 0 d 2
+γ
ωd2
!
peaks at ωd =ω0
The second part of this has a peak at exactly ωd = ω0 . However, it is multiplied by the factor
1/ωd , which increases as ωd decreases, shifting the peak to a slightly lower value of ωd .
4.2 Effects of damping
The shapes of the curves in figure 4.1.5 are profoundly important for applications of
resonance, both those applications in which we want to maximize resonance effects
(such as in radio receivers) and applications in which we want to minimize them (such
as in building designs). These curves are strongly affected by the degree of damping,
as we’ll explore in this section.
Often, it is revealing to re-express functions in terms of their dependence on
dimensionless variables; this frequently reveals a universal behavior that was obscured
in the original form of the function. In our case, we will try re-expressing the steady-
state amplitude A and the phase shift δ of the response relative to the drive force.
In equations (4.1.10) and (4.1.11), these two functions are expressed in terms of the
angular frequency of the drive, ωd and the factor γ ≡ b/m which characterizes the
damping. We will now re-express them in terms of ωd /ω0 and Q ≡ ω0 /γ , both of
which are dimensionless.
F0 /m
Your turn: Starting from equation (4.1.10), A = , and using
(ω02 − ωd2 )2 + (γ ωd )2
F0 /m
γ = ω0 /Q, show that A =
2 .
ω0 ω 1
ω0 ωd − d + 2
ωd ω0 Q

F0 ω0 /ωd k
We can rewrite this result as A = . Since ω0 ≡ , this
mω02
ω ω
2
1 m
0
− d + 2
ωd ω0 Q
becomes
F0 ω0 /ωd
A= . (4.2.1)
k ω ωd
2
1
0
ωd − ω0 + Q2
We can now see the universal behavior that we hoped would arise; the particular values
of ωd and ω0 are not really central – what really matters is the ratio ωd /ω0 .
If the drive force is applied by moving the support point, then equation (4.1.1):
F0 = kAd , so that
ω0 /ωd
A = Ad 2 . (4.2.2)
1
ω0
ωd − ωωd0 + Q2
This relation is graphed in figure 4.2.1. Several different curves are shown for different
values of Q. The most important effect of increasing the Q (i.e., lowering the damping)
is to make the peak higher and sharper.
A more subtle effect is that the peak, which is always at an angular frequency close
to ω0 , moves even closer to ω0 as Q increases. You can show in problem 4.8 that the
Figure 4.2.1 Dependence of steady state amplitude on ωd /ω0 .
peak occurs at

1
ωd, peak = ω0 1 − . (4.2.3)
2Q2
Recall from chapter 3 that Q = 2π corresponds to reasonably heavy damping. (Without

a driving force, the energy decays by a factor of 1/e per cycle, and the amplitude decays
√
by a factor of 1/ e ∼ = 0.61 per cycle.) Substituting Q = 2π into equation (4.2.3) gives
ωd,peak = 0.994 ω0 . For light damping, ωd,peak is even closer to ω0 , so for most purposes
the difference can be ignored.
For moderate to light damping, for which the peak amplitude occurs at ωd ∼ = ω0 ,
we can obtain a useful result for the height of the peak. Substituting ωd = ω0 into
equation (4.2.2) gives

A ωd = ω0 = QAd . (4.2.4)
In other words,
For light to moderate damping:

(peak response amplitude) ∼
= Q × (drive amplitude).
Since the low-frequency response amplitude is equal to the drive amplitude, we could
also say that the peak response amplitude is Q times the low-frequency response
amplitude.
For ωd ≫ ω0 , the response becomes

2
ω0 /ωd lim ωd ≫ω0 ω0
A = A d 2 −−−−−−→ Ad , (4.2.5)
ω0 ωd 1
ωd
ωd − ω0 + Q2
so that, at high frequencies, the response is universal, and does not even depend on the
degree of damping.
Figure 4.2.2 Phase shift δ of the

steady-state response.
Now, we will consider how the graph of the phase shift δ is affected by damping.
Again, we begin by expressing δ in terms of the dimensionless quantities ωd /ω0 and Q.
γ ωd
Your turn: Starting from equation (4.1.11), tan δ = , and using γ = ω0 /Q,
ω02 − ωd2
show that
1/Q
tan δ = ω ω . (4.2.6)
0
− d
ωd ω0
This relation is graphed in figure 4.2.2.3 At low ωd , the phase shift is zero, meaning
(as we anticipated) that the oscillator moves in phase with the drive. At high ωd , the
phase shift is 180◦ ; the oscillator is exactly out of phase with the drive. At ωd = ω0 , the
phase shift is exactly 90◦ . The effect of increasing the Q (i.e., lowering the damping)
is to sharpen the transition from δ = 0 at low drive frequency to δ = π at high-drive
frequency.
Connection to current research: Tapping Mode Atomic Force Microscopy

The Atomic Force Microscope (AFM), invented in 1986 by Gerd Binnig, Calvin Quate,
and Christoph Gerber, images a sample by touching it very lightly with a sharp tip
mounted on the end of a cantilever (figure 4.2.3). The upward force exerted on the
tip by the sample causes a bend in the cantilever, so that this force can be measured
quantitatively. In the original mode of operation (“contact mode”), the tip is first lowered
into contact with the sample until a pre-set force is achieved (typically about 100
nN). Then, the tip is moved laterally across the sample. As it moves, the force is
monitored, and the base of the cantilever is moved up or down as needed to keep
the force constant at the pre-set level. In this way, the microscope “feels” the shape
continued
1/Q
3. We can rewrite equation (4.2.6) as δ = tan−1 ω ω . Recall that, in this case if the argument
0
− d
ωd ω0
of the tan−1 function is negative, that is, if ωd > ω0 , we must add π to the result given by a
calculator or a program like Mathematica in order to get the correct value of δ . (See footnote 1
after equation (4.1.11).)
of the surface, while only applying a light force. This works well for semiconductor and
metal samples, but the lateral motion of the tip is very destructive for the soft samples
which are of most interest in biology and in molecular electronics.
Figure 4.2.3 The cantilever for an atomic force microscope (top image) must have a
moderately high Q, so that a small amplitude vibration applied at the base of the
cantilever results in a vibration amplitude at the tip of about 100 nm. The length of the
cantilever is 125 μm, about the same as the diameter of a human hair. The
pyramid-shaped tip located at the end of the cantilever (bottom image) must be very
sharp to obtain high resolution images. (Image courtesy of Veeco Instruments Inc.)
In 1993, Zhong, Inniss, Kjoller, and Elings introduced the Tapping modeTM AFM. (The
trademark belongs to Digital Instruments (now part of Veeco Corporation), which was
founded by Virgil Elings and is the leading manufacturer of AFMs.) In this mode, the base
of the cantilever is vibrated vertically by means of a piezoelectric crystal. The frequency
of vibration is chosen to match the resonant frequency of the cantilever, so that the tip
vibrates with an amplitude of about 100 nm. The vibration amplitude is measured, and
the base of the cantilever is slowly lowered toward the sample. When the tip begins
to tap the sample, this contact reduces the oscillation amplitude. The base is lowered
further until the amplitude reaches a pre-set value. Then, as in contact mode, the base
of the cantilever is moved laterally across the sample. As it moves, the amplitude of the
tip’s oscillation is monitored, and the base of the cantilever is moved up or down as
needed to keep the amplitude constant. Because the tip only touches the sample briefly
during each cycle of oscillation, the lateral forces applied to soft features on the sample
are minimized, so that they can be imaged without damage.

The Q of the cantilever is typically about 100. Therefore, since A ωd = ω0 = QAd ,
the drive amplitude Ad applied by the piezoelectric crystal to the base of the cantilever
needs to be only about 1 nm in order to produce the desired tip vibration amplitude
A of 100 nm. This large ratio is essential, because if one had to vibrate the base by the full
100 nm, the entire AFM would vibrate, dramatically degrading image quality.
94
4.3 Energy flow
Energy is perhaps the most fundamental idea in physics. In many situations, including
many parts of quantum mechanics, finding the energy is the central problem. We will
find that, for damped driven oscillators, a careful consideration of the energy for the
steady-state behavior provides insights that can be very broadly applied.
The power (energy per time) supplied to the oscillator by the drive force is
Pdrive = Fdrive v.
In the steady-state, x = A cos (ωd t − δ ), and so

v = ẋ = −Aωd sin ωd t − δ = Aωd sin δ − ωd t = Aωd cos δ − π/2 − ωd t

= Aωd cos ωd t + π/2 − δ
Therefore,

Pdrive = F0 cos ωd t Aωd cos ωd t + π/2 − δ . (4.3.1)
! !
Fdrive v
Concept test (answer4 below): For what value of ωd /ω0 are the oscillations of Fdrive
in phase with the oscillations in v? Hint: review figure 4.2.2.
Since, for the value you just found, the oscillations are always in phase, Pdrive is
always positive, that is, the drive force always supplies power to the oscillator. For any
other value of ωd /ω0 , the oscillations of Fdrive are not in phase with the oscillations in v;
therefore, for some parts of the cycle Pdrive is positive (the drive force supplies power to
the oscillator), and for some parts of the cycle Pdrive is negative (the oscillator supplies
power to the entity providing the drive force). An example is shown in figure 4.3.1a.
We can define Pdrive, av to be the power supplied by the drive averaged over a cycle.
For the example shown in figure 4.3.1a, Pdrive is positive for most of the cycle, so
Pdrive, av > 0.
As ωd → 0, δ → 0, so that Fdrive and v are out of phase by π /2. For this condition,
the net energy flowing from the drive to the oscillator is zero (i.e., Pdrive, av = 0), as
suggested in figure 4.3.1b. (You can show rigorously that the net energy flow is zero
in problem 4.11.) As ωd → ∞, δ → π , so that Fdrive and v are out of phase by π /2 in
the other direction; again the net flow of energy from the drive over a cycle is zero.
Putting all this together, we expect that the graph of Pdrive, av as a function of ωd /ω0
must start at zero, reach a peak (probably at ωd /ω0 = 1), and then go back to zero.
Let’s examine this quantitatively. The average power, Pdrive, av , can be computed
using the average value theorem from calculus:
T
1
Pdrive,av = Pdrive dt ,
T
0
4. For the oscillations to be in phase, we need δ = π /2, which occurs when ωd /ω0 = 1.
Figure 4.3.1 a: Fdrive and v for the case δ = 45◦ . In the shaded regions, the two have opposite
sign, so Pdrive is negative, meaning that the oscillator supplies power to the drive. In the other
regions, Fdrive and v have the same sign, so Pdrive is positive, and the drive supplies power to
the oscillator. (Since Fdrive and v have different units, the vertical scales cannot be compared.)
b: Fdrive and v for ωd → 0; in this limit δ → 0, so Fdrive and v are out of phase by π /2. In the
shaded regions, the two have opposite sign, so Pdrive is negative, meaning that the oscillator
supplies power to the drive. In the other regions, Fdrive and v have the same sign, so Pdrive is
positive, and the drive supplies power to the oscillator. (The velocity scale for this plot is
greatly magnified compared to part a; for small ωd , the velocity is also small.)
where T is the period of the steady-state motion. Plugging in our expression from
equation (4.3.1), we get
T
1
Pdrive = F0 cos ωd t Aωd cos ωd t + π/2 − δ dt
T
0
T
F Aω
= 0 d cos ωd t cos ωd t + π/2 − δ dt .
T
0
This integral can be evaluated using Mathematica, or a similar symbolic algebra

program or calculator; it can also be done “by hand” as shown in the footnote.5 Bearing

5. First, we re-express the second term in the integral: cos ωd t + π/2 − δ = cos δ −
F0 Aωd $T
π/2 − ωd t = sin δ − ωd t . Therefore, Pdrive = cos ωd t sin δ − ωd t dt.
T 0
Next, we use sin (A + B) = sin A cos B + cos A sin B, so that Pdrive =
Figure 4.3.2 Geometric representation of equation (4.2.6).
in mind that T = 2π/ωd , the result is
F0 Aωd
Pdrive, av = sin δ. (4.3.2)
2
To evaluate sin δ , we can represent equation (4.2.6) geometrically, as shown in

figure 4.3.2. From this, we see that
1/Q
sin δ = 2 .
1
ω0
ωd − ωω0d + Q2
F0 ω0 /ωd
Your turn: Substitute the above, and also equation (4.2.1): A =
k
ω ω 2

1
0
− d + 2
ωd ω0 Q
into equation (4.3.2), and show that the result is
F02 ω0 1
Pdrive, av = . (4.3.3)
ωd 2

2kQ ω0 1
− + 2
ωd ω0 Q
+ ,
F0 Aωd $T $T
sin δ cos ωd t cos −ωd t dt + cos δ cos ωd t sin −ωd t dt . This simplifies
T 0 0
+ ,
F0 Aωd $T $T
to Pdrive = sin δ cos2 ωd t dt − cos δ cos ωd t sin ωd t dt . These are both
T 0 0
standard integrals which can be looked up in a table. The first one is especially important:
" #T
$T t 1
cos2 ωd t dt = + sin 2ωd t . Since T = 2π/ωd , the second term evaluates to
0 2 4ωd 0
zero at both limits, so that the integral is just T /2. Since we integrated over a whole
period, we can see that the average value of a squared sinusoid over one period is 1/2.
The second integral can also be looked up in a table (or you can integrate it by parts):
" #T
$T 1
cos ωd t sin ωd t dt = sin2 ωd t . Since T = 2π/ωd , this evaluates to zero at both limits.
0 2ωd 0
F Aω T F Aω
Putting this all together gives Pdrive = 0 d sin δ = 0 d sin δ.
T 2 2
This relation is plotted in figure 4.3.3a. Inspection of the equation shows that the peak
is at exactly ωd /ω0 = 1. As for the graph of A versus ωd (figure 4.2.1), larger Q (less
damping) leads to a higher and sharper peak.
The width of the peak is quite important for a variety of applications. One common
way to define the width is the “full width at half maximum,” or FWHM, which is the
width at half of the peak height, as shown in figure 4.3.3b. Let us calculate the FWHM
exactly. Referring to equation (4.3.3), we see that the values of ωd corresponding to
half the maximum height are determined by

2
ω0 ω 1
− ± = , (4.3.4)
ω± ω0 Q2
where ω± is either ω+ or ω− . First, we consider ω− . Since ω0 > ω− , equation (4.3.4)

gives

ω02
ω ω ω ω − ωQ0 ± Q2
+ 4ω02
ω02 − ω−
2
= 0 − ⇔ ω−
2
+ 0 − − ω02 = 0 ⇒ ω− = .
Q Q 2
Since ω− must be greater than 0, we choose the positive square root, giving

ω 1 ω02
ω− = − 0 + + 4ω02 .
2Q 2 Q2
Figure 4.3.3 a: Power delivered by drive force to oscillator (averaged over a cycle) as a
function of ωd . This is called the power resonance curve. b: Definition of FWHM.
Next, we consider ω+ . Since ω+ > ω0 , equation (4.3.4) gives

2 − ω2 = ω0 ω+ 2 ω0 ω+
− ω02 = 0
ω+ 0 Q ⇔ ω+ − Q
ω0 ω02
± + 4ω02
Q Q2
⇒ ω+ = .
2
Since ω+ must be greater than 0, we choose the positive square root, giving

ω0 1 ω02
ω+ = + + 4ω04 .
2Q 2 Q2
The FWHM is given by ω+ − ω− , so that
ω0
FWHM = = γ. (4.3.5)
Q
Full Width at Half Maximum for the graph of Pdrive, av versus ωd .
This important equation tells us that systems with more damping have a broader power
resonance peak. This is a very universal behavior, which is observed even for systems
with turbulence or frictional damping.
4.4 Linear differential equations, the superposition principle for

driven systems, and the response to multiple drive forces
What happens if we apply a nonsinusoidal drive force, or if we apply two or three

different sinusoids at different frequencies and amplitudes? We will show in chapter 8
that any function can be represented as a sum of sinusoids. Therefore, if we can figure
out how a damped driven oscillator responds to multiple sinusoidal drives, it will be
relatively easy, after understanding the contents of chapter 8, to determine the response
to any driving force.
We will make use of an important property of the differential equation governing
a damped driven oscillator:
F0
mẍ = −kx − bẋ + F0 cos ωd t ⇒ ẍ + γ ẋ + ω02 x = cos ωd t .
m
This differential equation is linear, meaning that it has no terms proportional to x 2 , or
to ẋ 2 , or to x ẋ, etc. That means, for example, that if we multiply x by two, then the
entire left side of the above equation becomes twice as big. (This would not be true,
for example, if the left side were x 2 .) It also means, as we’re about to show, that we can
simply add the steady-state solutions for each of the individual driving forces to get the
steady-state solution for the combined driving force. The term on the right side, which
does not depend on x, and which comes from the drive (also called the excitation) is
called the “source term.”
We’ll be able to see the pattern for the response to any number of sinusoidal driving
forces by considering the case of just two:
Fdrive = F1 cos ω1 t + F2 cos ω2 t .
To find the behavior of the system, x(t), we begin to follow our three-step procedure,
though this time it will be surprisingly easy so that we don’t have to go through the
whole process.
mẍ = −kx − bẋ + F1 cos ω1 t + F2 cos ω2 t ⇒

F1 F
ẍ + γ ẋ + ω02 x = cos ω1 t + 2 cos ω2 t . (4.4.1)
m m
If the drive force is simply Fdrive = F1 cos ω1 t, then we know that the steady-
state response is x1 (t) = A1 cos ω1 t − δ1 , where A1 and δ1 are as given by
equations (4.1.10) and (4.1.11), simply replacing F0 by F1 and ωd by ω1 :
F1 /m γ ω1
A1 = and tan δ1 = . (4.4.2)
(ω02 − ω12 )2 + (γ ω1 )2 ω02 − ω12
We know that, for this simple drive force, we have

F1
ẍ1 + γ ẋ1 + ω02 x1 = cos ω1 t . (4.4.3)
m
Similarly, if the drive
force is simply Fdrive = F2 cos ω2 t, then the steady-state response
is x2 (t) = A2 cos ω2 t − δ2 , where A2 and δ 2 are defined analogously to equation
(4.4.2). For this simple drive force, we have
F2
ẍ2 + γ ẋ2 + ω02 x2 = cos ω2 t . (4.4.4)
m
Adding equation (4.4.4) to (4.4.3) gives
F F
ẍ1 + γ ẋ1 + ω02 x1 + ẍ2 + γ ẋ2 + ω02 x2 = 1 cos ω1 t + 2 cos ω2 t
m m .
F F
⇔ ẍ1 + ẍ2 + γ ẋ1 + ẋ2 + ω02 x1 + x2 = 1 cos ω1 t + 2 cos ω2 t
m m
F1 F
⇒ ẍtot + γ ẋtot + ω02 xtot = cos ω1 t + 2 cos ω2 t , (4.4.5)
m m

where xtot (t) ≡ x1 (t) + x2 (t) = A1 cos ω1 t − δ1 + A2 cos ω2 t − δ2 . We see that
equation (4.4.5) is the same as equation (4.4.1). Therefore,
The steady-state response of a damped driven oscillator to two sinusoidal drive forces is
simply the sum of the steady state responses to each force by itself.
It is easy to see that if we have seven sinusoidal drive forces, or seventy times
seven sinusoidal drive forces, then the steady-state response is simply the sum of the
individual responses.
Figure 4.4.1 Waves on a water surface are governed by a linear differential equation (if they
are small in amplitude), and so different wave patterns simply add, as shown here. Image ©
Andrew Davidhazy, Rochester Institute of Technology.
This principle, that the total solution is simply the sum of the individual solutions,
is called the “superposition principle” and only works for linear differential equations.
Luckily, many important differential equations in physics are linear, so we can use this
principle in many contexts. (See, e.g., figure 4.4.1.)
The superposition principle for driven systems: For a system governed by a linear
differential equation, the total response to multiple excitations is the sum of the
response for each excitation applied individually.
4.5 Transients
Recall that the goal of section 4.1 was to find the steady-state behavior, that is, the
behavior after any effects of initial conditions have decayed away because of the
damping. Indeed, examination of the solution we found there shows that there are no
constants that can be adjusted to take into account variations of the initial position or
velocity:
(4.1.12): x = A cos(ωd t − δ ), where
F 0 /m
(4.1.10): A = and
(ω02 − ωd2 )2 + (γ ωd )2
γ ωd
(4.1.11): tan δ ωd = .
ω02 − ωd2
In almost every case, the steady-state behavior is the only thing of interest. But, there
are a few circumstances in which the initial behavior, which does depend on x0 and
ẋ0 , is important. (See, e.g., figure 4.5.1.)
The solution x = A cos (ωd t − δ ) is a solution to our differential equation,
F0
(4.1.3b): ẍ + γ ẋ + ω02 x = cos ωd t ,
m
Figure 4.5.1 The overall impression made by a musical instrument is strongly influenced by
the “attack,” that is, the way a note starts up from silence. This is an example of transient
behavior, and the transition to steady state. The graph shows a simulation of the attack for a
flute; the larger amplitude waveform shows air density variations inside the flute, while the
smaller waveform shows the variation outside near the mouthpiece. Top image © Galina
Barskaya | Dreamstime.com. Bottom image © and courtesy of Dr. Helmut Kuehnelt.
but it cannot be the general solution; we know the general solution must include two
adjustable constants, which can be changed to reflect the effects of x0 and ẋ0 . So, we
can anticipate that the general solution must be the sum of x = A cos (ωd t − δ ) with
something that decays away over time and that depends on x0 and ẋ0 . It is actually
quite easy to find the general solution, using the work we have already done.
Consider the special case F0 = 0; this is one of the cases that must be described
by the general solution. The differential equation for this special case is
ẍ + γ ẋ + ω02 x = 0, (4.5.1)
which describes the damped oscillator without driving. We already found the general
solution for this case in section 3.1:
γ
(3.1.12): x = A0 e− 2 t cos ωv t + ϕ ,
where A0 and ϕ are determined by the initial conditions x0 and ẋ0 . Since this is the
solution to equation (4.1.3b): ẍ + γ ẋ + ω02 x = Fm0 cos ωd t for the special case of
F0 → 0, it is reasonable to guess that the general solution to equation (4.1.3b) is the
sum of this and equation (4.1.12):
? γ
xG = A0 e− 2 t cos ωv t + ϕ + A cos ωd t − δ . (4.5.2)
! !
transient behavior steady-state behavior
We can see that this works for the special case F0 = 0 (since, according to equation
(4.1.10), A = 0 for this case), and also that it works for the steady-state, since the first
term decays away over time. It has two adjustable constants A0 and ϕ , as we know
the general solution must. Is it in fact a solution of our differential equation (4.1.3b)
above?
To facilitate our checking, we introduce some nomenclature: equation (4.5.1) is
called the “homogeneous” version of equation (4.1.3b), because it has zero on the right
side. So, equation (3.1.12) is the general solution to the homogeneous version of the
differential equation, and we’ll call it xH :
γ
xH = A0 e− 2 t cos ωv t + ϕ . (4.5.3)
The steady-state solution (4.1.12) is a “particular solution” to the full version (the
inhomogeneous version) of the differential equation (4.1.3b), so we call it xP :

xP = A cos ωd t − δ . (4.5.4)
?
Now, we are ready to check whether our guess xG = xH + xP is a solution to
equation (4.1.3b):
? F0
ẍG + γ ẋG + ω02 xG = cos ωd t ,
m
d2 d ? F
⇔ 2
xH + xP + γ xH + xP + ω02 xH + xP = 0 cos ωd t
dt dt m
? F
⇔ ẍH + γ ẋH + ω02 xH + ẍP + γ ẋP + ω02 xP = 0 cos ωd t .
m
We know that xH is a solution of equation (4.5.1): ẍ + γ ẋ + ω02 x = 0, so the first term

in parentheses equals zero. We know that xP is a solution of equation (4.1.3b), so the
second term in parentheses equals Fm0 cos ωd t. Thus, the equation is indeed satisfied,
and the general solution for the damped driven harmonic oscillator is
γ
xG = A0 e− 2 t cos ωv t + ϕ + A cos ωd t − δ , (4.5.5)
! !
transient behavior steady- state behavior
where A0 and ϕ are determined by the initial conditions x0 and ẋ0 . A typical example
is shown in figure 4.5.2.
Figure 4.5.2 The full solution (bottom trace)

for the damped driven harmonic oscillator is
the sum of the general solution for the
damped oscillator xH (top gray trace) and the
steady-state solution xP (top black trace).
4.6 Electrical resonance
In section 3.2, we discussed the RLC electrical oscillator, shown in the top part of
figure 4.6.1. We applied Kirchhoff’s loop rule to obtain
VL + VR + VC = 0 ⇒
q
L q̈ + Rq̇ + = 0.
C
To add driving, we open the loop and apply a drive voltage, as shown in the bottom
part of figure 4.6.1. Now, instead of the voltage changes around the loop summing to
zero, they must sum to the applied voltage, so that
q
L q̈ + Rq̇ + = V0 cos ωd t . (4.6.1)
C
This is isomorphic to the differential equation for the damped driven mass/spring
system, (4.1.3a):
mẍ + bẋ + kx = F0 cos ωd t .

So, the full isomorphism is as shown in table 4.6.1.
Concept test (answer below6 ): In terms of R, L, and C, what is Q for the circuit shown
in the bottom part of figure 4.6.1?
√
6. Q = ω0 /γ . For the mass/spring, ω0 = k /m, which translates for the electrical oscillator
√
into ω0 = 1/LC. For the mechanical oscillator, γ = b/m, which translates into γ = R/L.

1 L 1 L
Combining gives Q = = .
LC R R C
Mass and Spring Electrical Oscillator

Mass m Inductance L
Spring constant k Inverse of capacitance: 1/C
Damping constant b Resistance R
Drive force amplitude F0 Drive voltage amplitude V0
Figure 4.6.1 Top: RLC Oscillator. Bottom: Driven

RLC oscillator.
The series RLC oscillator can be used to make a simple radio receiver, as shown
in the top part of figure 4.6.2. The voltage from the antenna
VIN is used to drive a
series RLC circuit. The inductor is adjustable, so that the resonant angular frequency
√
of the circuit, ω0 = 1/LC can be tuned to match the angular frequency of the radio
station. (For AM radio, this is in the range of 2π times 520–1,610 kHz.) Typically, there
are many radio stations within range, but each broadcasts at a different frequency. By
tuning the resonant frequency to match the broadcast frequency of one of the stations,
that signal is amplified by the factor Q (which can be very high), while other stations
that do not match the resonant frequency are amplified by a much smaller factor. The
output could be taken across the capacitor (as shown), or across the resistor, or across
the inductor, since at resonance q, q̇, and q̈ all oscillate with large amplitude. Practical
radio receivers have more complex input circuits than shown here, but the circuit still
includes resonance in a circuit containing an inductor and a capacitor.
Example: Two adjacent radio stations. In the United States, the minimum frequency
separation between AM radio stations is 20.4 kHz. (Stations in a particular broadcast
area usually have a much greater separation than this.) In this example, station WINE
broadcasts at 1,210 kHz, while WART broadcasts at 1,230.4 kHz. A circuit such as that
continued
Figure 4.6.2 Top: The front end of a simple radio receiver. The arrow on the inductor indicates
that it has a variable inductance, allowing the resonance frequency of the LRC circuit to be
tuned to match the frequency of the desired radio station. Bottom: Radio broadcast towers are
often grouped together to take advantage of a good site. As we’ll see in chapter 9, the
differential equation governing radio waves is linear, so the different signals simply add. A
resonant RLC circuit inside a receiver, tuned to the broadcast frequency of one of the stations,
is used as part of the circuitry that selects one of the stations out of the many signals that are
received. Image © Tose | Dreamstime.com.
shown in figure 4.6.2 is tuned to a resonant angular frequency of 2π · 1, 210 kHz. What is
the required Q if the power dissipated in the resistor due to the WART signal is to be half
the power dissipated due to the WINE signal?
Solution: Because the circuit is tuned to the angular frequency of WINE, the center
angular frequency ω0 for the power resonance curve (see figure 4.3.3b) is that of WINE. In
steady state, the power dissipated is equal to the power provided by the drive. We need
the power dissipated by the WART signal to be half that dissipated by the WINE signal,
meaning that the angular frequency of the WART signal should correspond to the half
maximum point of the power resonance curve (marked ω+ in figure 4.3.3b). We know
from equation (4.3.5) that the full width at half maximum of the power resonance curve
ω0
is given by FWHM = ωQ0 ⇔ Q = . We need half the FWHM to equal the difference
FWHM
in angular frequency between WART and WINE:
FWHM
= 2π × 20.4 kHz ⇔ FWHM = 4π × 20.4 kHz.
2
Recall that the resonant angular frequency of the circuit is given as ω0 = 2π × 1, 210 kHz.
2π × 1, 210 kHz
So, we have Q = = 29.7.
4π × 20.4 kHz
106
4.7 Other examples of resonance: MRI and other spectroscopies
Of course, every oscillator described in chapter 2 can exhibit resonance when driven.
However, the idea of resonance, meaning a strong response of system at or near a
particular excitation frequency, appears in many other situations as well. In this section,
we briefly explore two of these.
Semiclassical description of Magnetic Resonance Imaging (MRI). MRI is
based on the phenomenon of Nuclear Magnetic Resonance (NMR). The nuclei in
the atoms of the body are comprised of protons and neutrons. We will treat these using
a “semiclassical” description, meaning a hybrid of classical and quantum mechanical
ideas. Although this description does not capture the full quantum mechanical truth,
it does allow us to get a feel of what is going on, and it also allows fully quantitative
predictions for the results of experiments. In this description, we visualize the protons
as tiny spinning spheres. Because the protons carry charge, the spin creates a circulating
electrical current. As you’ll recall from a course in electricity and magnetism, a
circulating current creates a magnetic field. Therefore, as shown in figure 4.7.1, each
proton acts as a tiny bar magnet, albeit one with some unusual properties. Because
the magnetic moment μ of the proton is due to the spin, it is proportional to the spin
angular momentum J:
μ = γ J, (4.7.1)
where the proportionality constant γ is called the gyromagnetic ratio, and depends
on the type of the nucleus. (In quantum mechanics, one often uses J to represent the
angular momentum, rather than L.)
Now, we apply an external magnetic field, Bapplied . (In an MRI instrument, this is
usually produced by current flowing through a large coil that surrounds the patient.)
From introductory electricity and magnetism, this field creates a torque on any magnetic
Figure 4.7.1 Semiclassical model for the magnetic moment of a proton. The proton is pictured
as a spinning sphere, which creates a magnetic field similar to that of a bar magnet. Thus, the
proton has a magnetic moment μ. An applied external magnetic field exerts a torque τ on this
magnetic moment; the torque points into the page.
moment (including that of the nucleus):

τ = μ × Bapplied . (4.7.2)
The torque tends to align the magnetic moment along the direction of Bapplied .
However, quantum mechanics tells us that, surprisingly, μ doesn’t actually fully
line up with Bapplied ; instead, there is a well-defined angle θ between these two vectors.
One of the most important types of nuclei used for MRI is hydrogen; this is abundant
in virtually all the molecules that make up the body. For hydrogen, θ = 54.7◦ .
Because μ is not fully lined up with Bapplied , we have a body with a significant
angular momentum (the nucleus) that experiences a torque. As you may recall from a
demonstration in an earlier course involving a weighted bicycle wheel or a gyroscope,
surprising things happen in this circumstance. Let’s take the example of a child’s top
that is tilted, as shown in figure 4.7.2a. In the side view, we can calculate the torque
due to gravity around the point at the bottom of the top. The weight can be taken as
applied at the center of mass, and therefore creates a torque
τ = r × F.
Using the right-hand rule for the cross-product, this torque points out of the page.
Another way to find the direction of the torque: If the top weren’t spinning, it would
fall down, rotating counterclockwise about the point at the bottom. Curling the fingers
of your right hand in the direction of rotation that the torque is “trying” to produce, your
thumb points out of the page, in the direction of τ . However, because the top already
has angular momentum, this torque does not cause it simply to fall over. Recall that
τ = dL/dt ⇔ dL = τ dt ,
where L is the angular momentum of the top. Therefore, the change in the angular
momentum is in the direction of τ , which is perpendicular to L. So, the length of L
doesn’t change, but its direction does: in time dt, the the tip of the L vector moves an
infinitesimal amount out of the page. Since L is along the axis of rotation of the top,
this means that the axis of rotation moves as well. This doesn’t fundamentally change
the situation, so the direction of the torque τ changes by an infinitesimal amount as
well, and is still perpendicular to L. Thus, the tip of L will trace out a circle with a
“radius” (in units of angular momentum) equal to the x-component of L, as shown in
figure 4.7.2c:
“radius” of circle = Lx = L sin θ.
This motion of the axis of rotation is called “precession.” If there were no friction to
slow down the spinning of the top, it would continue to precess in this way without
ever falling down.
The “speed” (in units of angular momentum per time) of the tip of L as it moves
along this circle is τ = dL/dt. Therefore, the time it takes for one full circle is
“distance” 2π “radius” 2π L sin θ
time = = = .
“speed” τ τ
The frequency of the precession is the inverse of this time:
τ
f = . (4.7.3)
2π L sin θ
Figure 4.7.2 a and b: A child’s top that is tilted experiences a torque which is perpendicular to
the angular momentum L. c: The torque causes L to precess around a circle. d: We define a
reference frame that rotates at the precession rate. In MRI, a pulse of RF radiation is applied,
with the magnetic field along the x-axis (in the nonrotating frame).
Applying these ideas to the nucleus with the applied external magnetic field, the tip of
μ precesses in the same way as does the axis of rotation of the top, moving in a circle
around the direction of Bapplied . As the tip moves in this circle, the vector μ sweeps
through a cone, as shown in figure 4.7.2d.7 Combining equations (4.7.2) and (4.7.3),
and using J to denote the magnitude of the spin angular momentum (rather than L),
we find that the precession frequency is
τ μBapplied sin θ μBapplied

f = = = .
2π J sin θ 2π J sin θ 2π J
7. Recall that this is a semi-classical treatment. In a fully quantum mechanical treatment, the
direction of μ cannot be fully determined, due to an uncertainty principle similar to those
discussed in section 1.12. We discussed there how the momentum and the position of a particle
cannot both be precisely defined simultaneously. Similarly, it turns out that the components of
μ along the x-, y-, and z-axes cannot all be precisely defined simultaneously. In our example, the
length of μ is precisely defined, and the component along the direction of Bapplied is precisely
defined, but the components in the other directions are completely undefined. We sometimes say
that μ is “delocalized” around the cone. Careful consideration of these effects produce the same
results as our semi-classical argument.
Substituting for μ from equation (4.7.1) gives
γ JBapplied
f = ⇒.
2π J
γ
fLarmor = B , (4.7.4)
2π applied
where this precession frequency is named the “Larmor frequency” after Joseph Larmor,
an Irish physicist who occupied the same professorship at Cambridge University as
did Isaac Newton. Because there is no dissipation in the system, the magnetic moment
μ of the nucleus precesses around the direction of Bapplied forever, unless something
else happens.
For MRI, the “something else” happens when we apply a pulse of radio frequency
(RF) electromagnetic radiation, with a frequency equal to fLarmor . Let us define the
z-axis to lie along Bapplied . The MRI machine is arranged so that this RF radiation
travels in the z-direction. As you may recall from a course in electricity and magnetism,
electromagnetic radiation consists of waves of perpendicular electric and magnetic
fields, both of which are perpendicular to the direction of travel. (We shall explore
these in more detail in chapter 9.) Therefore, the magnetic field of the RF radiation,
Brad could oscillate along any direction in the x–y plane. For convenience, we take it to
oscillate along the x-axis. This oscillation at frequency fLarmor and angular frequency
ωLarmor = 2π fLarmor can be considered to be the sum of two counter-rotating magnetic
field vectors, each of which has angular velocity ωLarmor , as shown in figure 4.7.3a.
We call the clockwise-rotating vector Brad, cl ; it rotates at the same rate and in the same
direction as does the precessing magnetic moment μ of the nucleus.
We now consider what things look like in a rotating reference frame x’, y’, z’ that
rotates clockwise along with μ, with z’ parallel to z. Since Brad, cl rotates at the same
rate as μ, it points in a constant direction in this frame. Let us define this direction to
be x’, as shown in figure 4.7.3b.
Figure 4.7.3 a: An oscillating magnetic field along

the x-axis equals the sum of two counter-rotating
magnetic fields. If the angular frequency of the RF
pulse equals the Larmor frequency, then Brad, cl
rotates at the precession rate, and so is stationary in
the rotating reference frame. b: Taking Brad, cl to lie
along x ′ , it causes a precession of the magnetization
around the x ′ axis. If the pulse length is chosen
suitably, the magnetization can be rotated by 90◦ .
At this point, we take a step back from thinking about the individual nuclei, and
instead consider the net magnetization M of a small region of the patient inside the
MRI machine. This magnetization is due to the average of the magnetic moments of
the nuclei, and so points along z. We have seen that a magnetic moment precesses
about the direction of an applied magnetic field. Because the magnetization is due
to the average of many magnetic moments, it also precesses about any applied field.
Before the application of Brad , M is parallel to Bapplied , and so is stationary. When
Brad is applied, Brad, cl points in a constant direction in the rotating reference frame,
and M precesses about it. (In the rotating reference frame, Brad, ccl rotates at angular
frequency 2ωLarmor , and so has no average effect.)
We can control how far M precesses by the duration of the RF radiation pulse.
Because M results from the average of all the individual magnetic moments of the
atoms, the precession of M around the magnetic field due to the RF pulse, as seen in
the rotating frame, is governed by the same physics leading to equation (4.7.4), so that
the angular rate of rotation is ω = γ Brad, cl . In the simplest version of MRI, the pulse
length is chosen so that M precesses by 90◦ in the rotating reference frame, so that
after this precession it lies along the y’ axis, as shown in figure 4.7.3b. A pulse of this
length is called a 90◦ or π /2 pulse.
In the stationary reference frame, the y’axis rotates clockwise at angular frequency
ωLarmor around the z-axis. Therefore, after a 90◦ pulse has been applied so that M lies
along the y’ axis, it also rotates clockwise at ωLarmor . This rotating magnetization
broadcasts electromagnetic radiation, which can be detected by the MRI machine.
This whole mechanism only works when the frequency of the radiation in the RF pulse
matches fLarmor . Any other frequency of radiation would have a magnetic field that is
not stationary in the rotating frame, and so produces no average effect.
Several different techniques are used to get contrast between different body tissues
in MRI. The simplest is associated with the effect of nearby electrons on the Bapplied
that is felt by a particular nucleus. The externally applied magnetic field affects the
motion of these electrons, and since moving electrons create a magnetic field, the total
magnetic field experienced by the nucleus depends on the local density of electrons.
Nuclei in different types of molecules, and in different parts of the same molecule, are
surrounded by different densities of electrons. From equation (4.7.4), fLarmor depends
on the total Bapplied . Therefore, these nuclei respond to different frequencies for the
RF radiation pulse, giving contrast.
The rotation of M away from Bapplied increases the potential energy of the system;
this energy comes from the RF radiation pulse. We have seen that this transfer of
energy from the radiation to the nucleus can only occur if the radiation has the correct
frequency, fLarmor .
There is a rather different way of understanding the transfer of energy from
the radiation to the nucleus, which generalizes to many other situations, but is less
connected to classical physics. Even in a rather strong applied magnetic field, the effect
of the field on the nuclei is small compared to that of random thermal fluctuations. This
fact does not affect the above arguments, which are based on the average magnetization.
Let us consider again the particular case of hydrogen nuclei. If one measures μz (the
component of μ in the direction of Bapplied ), then surprisingly one finds that the result
is always one of two possibilities. Slightly more than half the time, one gets the result
for μz that one would expect from figure 4.7.1, with θ = 54.7◦ ; this is called “spin up,”
because the direction of the nuclear spin axis is as close to the direction of Bapplied
as it ever gets. Slightly less than half the time one gets the result, one would expect
for θ = 180◦ − 54.7◦ = 125.3◦ ; this is called “spin down.” This surprising finding
that only these two results are measured and nothing in between, is a purely quantum
mechanical effect with no classical analog. You might ask, “Weren’t we just talking
about the magnetization precessing by 90◦ ? Isn’t that an in-between result?” However,
recall that that part of our discussion was centered on the magnetization of a small
region inside the patient, which is the combined result of the magnetic moments of
many individual spins. Each of the individual spins is either “spin up” (meaning that it
points somewhere along the upward-pointing cone, with an angle of 54.7◦ to Bapplied )
or it is “spin down.” The combined effect of all these spins is the magnetization of the
small region, which can change direction in an essentially continuous way.
The spin angular momentum of the hydrogen nucleus, which is simply a proton,
is a fixed quantity. It is a property of the particle, much as the charge is. The magnitude
of this angular momentum is
√
3
J= h̄,
2
where h̄ = 2hπ = 1.05457 × 10−34 Js and h = 6.6260693 × 10−34 Js are both called
“Planck’s constant.” To avoid confusion to the extent possible, h̄ is usually just called
“h-bar.” So, for the spin-up result, the z-component of the spin angular momentum is
√
3 h̄
jz↑ = h̄ cos 54.7◦ = ,
2 2
where the up arrow indicates “spin up.” Using equation (4.7.1), this means that
h̄
μz↑ = γ .
2
Similarly, for the “spin down” possibility, we have
h̄
μz↓ = −γ .
2
From introductory electricity and magnetism, the potential energy for a magnetic dipole
in an applied field is
U = − µ· Bapplied .
In our case, Bapplied is in the z-direction, so
U = −μz Bapplied .
Therefore, the potential energies for spin up and spin down are
h̄ h̄
U↑ = −γ Bapplied and U↓ = γ Bapplied .
2 2
The difference between these is

U = U↓ − U↑ = γ h̄Bapplied .
You may recall from section 1.12 that there is a connection between oscillation
frequency and energy for an electron:
(1.12.5): E = hf .
It turns out that the same relation applies to light. Experimentally, we find that,
whenever energy is absorbed from light, it is always absorbed in discrete packets
called “photons,” each of which has energy given by equation (1.12.5), with f equal
to the oscillation frequency of the light. To “promote” the hydrogen nucleus from the
lower energy spin up state to the higher energy spin down state, and to conserve energy,
the system must absorb a photon from the RF radiation pulse that has energy
E = hf =
U = γ h̄Bapplied ⇒
γ h̄Bapplied γ Bapplied
f = =.
h 2π
At this point, there may be chills running up and down your spine, because this
frequency, which we have arrived at using purely quantum mechanical methods,
is exactly the same as the Larmor frequency (4.7.4), which we derived using
semi-classical methods!8
Other types of spectroscopy. By either line of reasoning, only radiation of frequency

very close to fLarmor can transfer energy to the system of nuclear magnetic moments.
The quantum version of the argument generalizes perfectly to a large variety of
circumstances. For example, in ultraviolet/visible spectroscopy, light is passed through
a solution containing molecules of interest. Within the molecules, the electrons occupy
quantum mechanical states with well-defined energies. It is possible to promote an
electron from a low-energy state to a higher-energy state by the absorption of a photon
with a frequency f such that hf equals the energy difference between the two states.
This is essentially the same process that we described earlier, in which the absorption
of a photon from the RF radiation pulse promotes the hydrogen nucleus from the spin
up state to the spin down state.
In a more sophisticated version of the argument, we would see that the frequency
match need not be exact, although the energy transfer is greatest when there is an
exact match. This is closely analogous to the resonant response of a damped, driven
mass/spring system; the system absorbs the most energy from the drive force when
the frequency of the drive matches the resonant frequency of the system, but energy
can also be absorbed from drives with frequencies that do not match the resonance
exactly. Recall that the width of the power resonance curve for a mass/spring system
increases when the damping is increased. Also recall that the damping represents a
coupling of the energy in the oscillator to other parts of the universe, such as the air
surrounding the mass. In just the same way, the range of radiation frequencies that
can be absorbed by a molecule increases when the molecule is coupled to other things
8. For more detailed treatments, see Introduction to Physics in Modern Medicine, 2nd Ed., by
Suzanne Amador Kane, CRC Press, Boca Raton FL, 2009, and Physical Methods for Chemists,
2nd Ed., by Russell S. Drago, Surfside Scientific Publishers, Gainesville, FL, 1992.
in the universe. Thus, the frequency range is very small when the molecule is in a
dilute gas phase, becomes broader when the gas molecules collide more frequently
(“pressure broadening”), and becomes much broader when it sticks via multiple strong
bonds to the surface of a solid. (Note that there are additional mechanisms that broaden
the frequency range in actual spectroscopic experiments, such as local variations in
the environments felt by different copies of the same molecule.)
4.8 Nonlinear oscillators and chaos
In all real systems, Hooke’s Law F = −kx is only valid for a small enough displacement
away from equilibrium. Beyond this limit, the restoring force is no longer proportional
to displacement. This nonlinearity leads to a host of fascinating effects, many of which
are of great practical importance. Unfortunately, even the simplest modification to
F = −kx leads to a differential equation with no analytic solutions, meaning that no
combination of exponentials, trigonometric functions, polynomials, etc., gives an exact
solution. Therefore, scientists and engineers approximate actual systems with a linear
restoring force whenever possible. Nonlinear systems can be studied quantitatively,
using a combination of sophisticated approximation techniques and numerical (i.e.,
computer-based) methods, but these techniques are beyond the scope of this text.9 In
this section, we use rough and mostly qualitative arguments to explore a few of the
most important phenomena.
Harmonic generation10
The potential energy for Hooke’s Law is U = 12 kx 2 . It will make the math easier
when we consider a nonlinear oscillator to maintain the symmetry of the potential
energy around x = 0. (Later, we will consider nonsymmetrical potentials qualitatively.)
The simplest modification that preserves the symmetry is the addition of a term
proportional to x 4 :

U = 21 kx 2 1 + α x 2 , (4.8.1)
where α is a constant. We assume that the nonlinearity is small, meaning that

α A2 ≪ 1,
where A is the maximum magnitude of x. This potential energy is shown in figure 4.8.1a,
and compared to the harmonic potential energy U = 12 kx 2 . (You can see that, for small
x the harmonic potential is a good approximation.) Instead of only thinking about a
mass on a non-Hookian spring, we can extend our thinking and consider any particle
9. It is not too difficult to use a computer for numerical integration; you might inquire with your
instructor about doing an independent project in this area.
10. This discussion is adapted from that in Vibrations and Waves in Physics, 3rd Ed., by
Ian G. Main, Cambridge University Press, Cambridge, 1993, pp. 93–97.
which experiences this potential. For example, the particle might be an atom which
experiences a potential due to chemical bonds with its neighbors.
If the particle has total energy E, then it can move between points 1 and 2 marked
in figure 4.8.1a. When released from point 1, the particle oscillates
1 between
the two
2 1 2 2 2
points. The condition α A ≪ 1 is equivalent to 2 kA α A ≪ 2 kA . You can see
from figure 4.8.1a that this is not obeyed
for
the energy E chosen for our example,
although we do have 21 kA2 α A2 < 21 kA2 . (For our example, α A2 = 0.6.) We use
this example to highlight the changes that occur when the potential energy is not
harmonic. Because our example does not obey α A2 ≪ 1, the results obtained below
are only qualitatively correct for our example. However, they are quantitatively correct
for oscillations with lower E, corresponding to lower oscillation amplitude. (For an
energy corresponding to half the amplitude of our example, α A2 = 0.15, which is small
enough compared to 1 that the results below would be fairly good approximations.)
Because the potential energy is not harmonic, we expect that x (t) will not be a
simple cosine function, that is, we will not simply have x = A cos ωt as we did for
the simple harmonic oscillator. In fact, figure 4.8.1b shows the actual behavior that
we will derive below, for the example energy shown in figure 4.8.1a. Also shown in
figure 4.8.1b is a cosine (dashed line) with the same amplitude and angular frequency
as the actual x (t). You can see that the two curves are almost identical, but not quite.
Figure 4.8.1d shows the slopes (i.e., ẋ) for these two curves, which shows the difference
a little more clearly.
Concept test (answer below11 ): In the regions near x = 0, the shape of the ẋ curve
corresponding to x(t) is a little flatter than the ẋ curve corresponding to the cosine. Why
is this to be expected, based on the shape of U(x)?
However, even though x(t) is not sinusoidal, it still must be a periodic function
(if we ignore damping). We will show in chapters 7 and 8 that any periodic function
can be expressed as a weighted sum of the sinusoids with the same periodicity; this is
the process of Fourier synthesis. In our case, the period for oscillation from 1 to 2 and
back to 1 is T , as shown in figure 4.8.1b. We define ω ≡ 2π /T . Of course, the function
cos ωt has periodicity T , meaning that it has the same shape from t = 0 to t = T as it
does from t = T to t = 2T . However, the function cos 2ωt also has periodicity T ; it
completes two full cycles during the time T , and in the interval from t = T to t = 2T
it again completes two full cycles. In fact any function cos nωt, where n is an integer,
has periodicity T , as do the functions sin nωt. Even the integer n = 0 works; this just
gives a constant for either the sin or the cos. Therefore, we must be able to write
x(t) = const. + a1 cos ωt + b1 sin ωt + a2 cos 2ωt + b2 sin 2ωt

+ a3 cos 3ωt + b3 sin 3ωt + · · · (4.8.2)
where the a’s and b’s are constants.
11. Because the potential energy curve U(x) is flatter near x = 0 than a harmonic potential energy
1 2
2 kx , the velocity for the U(x) case does not change as much near x = 0.
Figure 4.8.1 a: A nonharmonic potential energy U (x) is compared with the harmonic potential
energy 21 kx 2 . A particle is released at rest from point 1, so that it has total energy E. Thereafter,
it oscillates between points 1 and 2. At point 1, x is only approximately equal to A, because of
the added cos 3ωt term in equation (4.8.7). b: The full solution x (t) from equation (4.8.7) is
plotted as a solid line, and compared with a cosine of the same amplitude and angular
frequency (dashed line). Although they are almost the same, x (t) is slightly below the cosine
from t = 0 to t = T /4, and slightly above the cosine from t = T /4 to t = T /2. c:Comparison
of three cosines and one sine. Sines have odd symmetry about t = T /2, and so cannot
contribute to x (t). Cosines of even multiples of ωt, such as cos 2ωt have even symmetry about
t = T /4, and so cannot contribute to x (t). Only cosines of odd multiples of ωt, such as the two
shown in black, have the correct symmetries. d: The black curves are the same as in part b –
x (t) is shown as a solid black line, and a cosine with the same amplitude and angular
frequency is shown as a dashed black line. The grey curves show the time derivatives, that is, ẋ
for each of the black curves, so that the solid gray curve is the time derivative of x (t) and the
dashed gray curve is the time derivative of the cosine.
However, because our potential energy function is symmetrical, we know that the
shape of x(t) from point 1 to 2 must be the mirror image of the shape from point 2 to 1,
that is, x(t) must have even symmetry about the time T /2, as shown in figure 4.8.1b.
The sines have odd symmetry about T /2 as shown in figure 4.8.1c, therefore all the b’s
must equal zero. Furthermore, again because of the symmetry of the potential energy,
the shape of x(t) as the particle moves from point 1 (at t = 0) to x = 0 (at t = T /4)
should be the time reversed version of the shape as the particle moves from x = 0
(at t = T /4) to point 2 (at t = T /2), that is, x(t) must have odd symmetry about the
time T /4, as shown in figure 4.8.1b. The odd cosines (cos ωt, cos 3ωt, etc.) have this
property, but the even cosines don’t, as shown in figure 4.8.1c. Therefore, the a’s with
even subscripts must all be zero. Finally, because of the symmetry of the potential
energy function, the average position of the particle must be zero, so that the first term
in equation (4.8.2) (“const.”) must be zero. Therefore, we have
x(t) = a1 cos ωt + a3 cos 3ωt + a5 cos 5ωt + · · · .
Because we have assumed that the nonlinearity is small, it is reasonable to expect that
the departure from harmonic behavior represented by the cos 3ωt and higher terms is
also small. Therefore, we can write

x(t) ∼
= A cos ωt + ε3 cos 3ωt + ε5 cos 5ωt + · · · , (4.8.3)
where A = a1 , ε3 = a3 /A, etc., and ε3 , ε5 , etc., are much less than 1.
The restoring force associated with the potential energy (4.8.1) is
dU d 1 2 1 4

Fr = − =− kx + k α x = −kx − 2k α x 3 .
dx dx 2 2
In the absence of damping, this is the only force that acts on the mass. Therefore,
we have
F
Fr = mẍ ⇒ ẍ − r = 0 ⇒
m
ẍ + ω02 x + 2ω02 α x 3 = 0, (4.8.4)
√
where ω0 ≡ k /m. To determine the coefficients in equation (4.8.3), we need to plug
it into equation (4.8.4). To prepare for this, we first evaluate the messiest term:
3
2ω02 α x 3 = 2ω02 α A2 A cos ωt + ε3 cos 3ωt + ε5 cos 5ωt + · · · .
Unlike the other terms in equation (4.8.4), this has the overall multiplicative factor of
α A2 , which is much smaller than 1. Therefore, since the ε’s are small, we ignore all
terms proportional to ε3 or ε5 , etc. (Such terms would be proportional, for example, to
ε3 α A2 , and so would be utterly negligible compared to the terms in equation (4.8.4)
that are not proportional to an ε or to α A2 .) This leaves us with
2ω02 α x 3 ∼
= 2ω02 α A2 A cos3 ωt .
1 3
You can show in problem 4.21 that cos3 θ = 4 cos 3θ + 4 cos θ . Applying this to the
above gives
ω 2 α A2
2ω02 α x 3 ∼
= 0 A(cos 3ωt + 3 cos ωt).
2
Substituting this and equation (4.8.3) into (4.8.4) and cancelling the common factor
of A gives

− ω2 cos ωt + 9ε3 cos 3ωt + 25ε5 cos 5ωt + · · ·

+ ω02 cos ωt + ε3 cos 3ωt + ε5 cos 5ωt + · · ·
ω02 α A2
+ (cos 3ωt + 3 cos ωt) ∼
= 0. (4.8.5)
2
For this to be valid at all times, the coefficients of cos ωt must sum to zero, as must
the coefficients of cos 3ωt and the coefficients of cos 5ωt, etc. Setting the sum of the
coefficients of cos ωt to zero we get
3ω02 α A2 ∼
− ω2 + ω02 + =0⇒
2

3α A2
ω∼
= ω0 1 + . (4.8.6)
2
Thus, depending on the sign of α , the angular frequency will be shifted up or down
relative to ω0 , by an amount that depends on the amplitude A. Thus, one of the hallmarks
of harmonic oscillation, that the frequency is independent of amplitude, is removed
when we consider a non-linear restoring force. In our example, ω = 1.38 ω0 , a quite
substantial shift.
Returning to equation (4.8.5), setting the sum of the coefficients of cos 3ωt to zero
gives
ω02 α A2 ∼
−9ε3 ω2 + ε3 ω02 + = 0.
2
Plugging in equation (4.8.6) yields
ω2 α A2 ∼

2 3α A2 27 1
−9ε3 ω0 1 + +ε3 ω02 + 0 = 0 ⇒ −8ε3 ω02 − ε3 ω02 α A2 + ω02 α A2 ∼
= 0.
2 2 2 2
The middle term is proportional to both ε3 and α A2 , and so is negligible compared to
the others. This leaves us with
1
ε3 ∼
= α A2 .
16
Thus, as we had anticipated, the departure from harmonic behavior in x(t) is small,
and proportional to the departure from linearity in the restoring force.
Finally, in equation (4.8.5) setting the sum of coefficients of cos 5ωt to zero gives
−25ε5 ω2 + ω02 ε5 ∼
= 0.
Plugging in equation (4.8.6) yields

2 3α A2 2 ∼ 2 3α A2 ∼
−25ε5 ω0 1 + + ω0 ε5 = 0 ⇒ −24ε5 ω0 1 + = 0.
2 2
Since α A2 ≪ 1, this gives
−24ε5 ω02 ≈ 0 ⇒ ε5 ≈ 0.
Thus, at the level of approximation we are using, there is a negligible response at

angular frequency 5ωt. (Similar arguments would hold for the higher harmonics, such
1
as 7ωt.) Plugging this result and ε3 ∼
= 16 α A2 into equation (4.8.3) gives

α A2
x∼ = A cos ωt + cos 3ωt . (4.8.7)
16
This relation is plotted for our example as the solid line in figure 4.8.1b. (If the departure
from linearity is large enough, the higher harmonics such as 5ωt become important.)
If we had considered an asymmetric potential energy, then we would not have been
able to discard the even harmonics in equation (4.8.2), that is, terms such as cos 2ωt. In
analogy with equation (4.8.7), we would find that the amplitude of the cos 2ωt term is
proportional to the departure from nonlinearity in the restoring force. Thus, we reach
the important conclusion:
If the restoring force is nonlinear but with symmetric magnitude about the equilibrium
point, then we observe oscillations with angular frequency ω ≈ ω0 superposed with
oscillations with angular frequency 3ω. If the force is nonlinear and asymmetric, we also
observe oscillations with angular frequency 2ω. Thus, harmonics of the fundamental
oscillation ω are generated. For highly nonlinear restoring forces, higher harmonics can
also be generated.
Harmonic generation is important in applications of lasers. The available laser

wavelengths are limited to certain discrete values depending on the material from
which the laser is made. However, if a laser of wavelength λ, corresponding to angular
frequency ωd , is focused at high intensity onto a suitable crystal, there is a response
of the electrons in the crystal at 2ωd , resulting in the production of light at angular
frequency 2ωd , corresponding to a wavelength λ/2, which might be exactly what is
needed for a particular application.
Sub-harmonic resonances
Now, we consider a damped, driven nonlinear oscillator. Above, we showed that, in
a circumstance where we would get a response at ω0 for a linear system, we did get
a response at an angular frequency near ω0 , but also a response near 3ω0 and (for an
asymmetric restoring force) near 2ω0 . For a damped driven linear oscillator, we get a
steady-state response at ωd . Therefore, given the above results, it may seem reasonable
that in the driven case we get a steady-state response at ωd , but also responses at 3ωd
and (for an asymmetric restoring force) at 2ωd . In other words, the response is

x (t) = A cos ωd t − δ + A2 cos 2ωd t − δ2 + A3 cos 3ωd t − δ3 .
For a highly nonlinear system, we can also get responses at higher harmonics. The
math needed to show this is more complicated than for the undriven case.
For the nondriven case, we saw that, for a small nonlinearity, the oscillation with
angular frequency near 3ω0 is much smaller than the oscillation with angular frequency
near ω0 . Similarly, for the damped driven case, one can show that the response at 2ωd
and 3ωd is ordinarily much smaller than the response at ωd . However, if ωd = ω0 /2,
then the response at 2ωd = ω0 is amplified by the resonance of the system. If Q is

large, the response at 2ωd = ω0 can thus be significantly larger than the response at
ωd . Similar comments apply for ωd = ω0 /3. Thus, we can excite the resonance of a
system with a drive force frequency below the resonant frequency.
Sub-harmonic resonances are often annoying. For example, parts of speakers in
an audio system can be modeled as mass/spring oscillators. (Since all the parts are in
equilibrium positions, we know that there are spring-like forces holding them there.)
If the volume is turned up high, the sound drives the mechanical components of the
speaker at high amplitude, so that the nonlinear region of the restoring force becomes
important. When the frequency of the sound is half or one third of the resonant
frequency of a part, it can be excited through the sub-harmonic resonance effect,
resulting in an unpleasant buzz.
Mixing
A resistor has a linear current–voltage relationship:
V = IR ⇒ I = GV ,
where G ≡ 1/R is the conductance. If we apply the voltage
V = V1 cos ω1 t + V2 cos ω2 t (4.8.8)
to a resistor, nothing very exciting happens. The resulting current is simply
I = V /R = GV1 cos ω1 t + GV2 cos ω2 t .
However, now let’s consider a circuit element with a nonlinear current–voltage

relationship, such as
I = GV (1 + α V ) = GV + α GV 2 ,
where α is a constant. (Note: the current as a function of voltage for any device can
be written as a Taylor series, so for anything other than a resistor there will be a
term proportional to V 2 and/or V 3 . Thus, we need not be talking about a very exotic
component; something like a diode would exhibit the effects we discuss here).Applying
the voltage (4.8.8) to this nonlinear device gives a current
2
I = GV1 cos ω1 t + GV2 cos ω2 t + α G V1 cos ω1 t + V2 cos ω2 t . (4.8.9)
It is the third term which is of most interest:

2
α G V1 cos ω1 t + V2 cos ω2 t

= α G V12 cos2 ω1 t + 2V1 V2 cos ω1 t cos ω2 t + V22 cos ω2 t . (4.8.10)
Now, we use two trigonometric identities:
cos2 θ = 1
2 (1 + cos 2θ ) and cos A cos B = 1
2 [cos (A + B) + cos (A − B)] .
Applying these to equation (4.8.10) gives

2 αG 2
α G V1 cos ω1 t + V2 cos ω2 t = V1 + V22 + V12 cos 2ω1 t + V22 cos 2ω2 t
2

+2V1 V2 cos ω1 + ω2 t + 2V1 V2 cos ω1 − ω2 t .
Thus, the current has components that oscillate at ω1 and ω2 (as we can see from
equation (4.8.9), but also components that oscillate at 2ω1 , 2ω2 , ω1 + ω2 , and ω1 − ω2 .
It is these last two that are perhaps the most interesting. We could call them the
sum angular frequency and the difference angular frequency. The difference angular
frequency plays the key role in the technique of “heterodyning,” which is used in
virtually every radio receiver. In AM radio, the information to be transmitted is used as
an envelope function for a “carrier wave.” The carrier wave must have a much higher
frequency (around 1 MHz) than the audio information (about 10 kHz). In FM radio,
the information to be transmitted is instead used to control the frequency of the carrier
wave; a variation in the audio wave causes a small change in the carrier frequency.
In either case, inside the radio receiver, the signal from the antenna is mixed with the
signal from a “local oscillator,” which is simply a source of AC voltage at a frequency
near that of the carrier. The mixing occurs by applying both signals to a nonlinear
element, as described earlier. Because the frequency of the local oscillator is close to
that of the carrier wave, the difference frequency generated by the mixing is a much
lower frequency, and so this signal can more easily be processed by subsequent circuits
in the receiver.
The same mixing effects occur in a nonlinear oscillator that is driven by two forces
at angular frequencies ω1 and ω2 ; this results in a response that is the sum of functions
that oscillate at 2ω1 , 2ω2 , ω1 + ω2 , and ω1 − ω2 , with amplitudes proportional to
the amplitudes of the drive forces. (The math required to show this for the case
of a nonlinear oscillator is more complicated than for the nonlinear circuit element
described earlier.)
Sensitivity to initial conditions

For some systems with nonlinear restoring forces, one observes (for appropriate choices
of parameters) very complicated motion, even though the system itself is simple,
and fully described by exact equations. In a system without a driving force, motion
eventually ceases due to dissipation, but the final state of the system can be sensitive
to the tiniest changes in the initial conditions.
One example is the “magnetic pendulum,” consisting of a rigid pendulum arm
ending in a small piece of iron, which hangs above a flat surface that has three identical
magnets on it, as shown in the top part figure 4.8.2. One of the magnets is colored white,
one black, and one gray. The piece of iron is attracted to each of the magnets, by a force
proportional to 1/r 3 . Thus, the end of the pendulum moves in response to a nonlinear
force. If the pendulum is started from rest near one of the magnets, it simply swings
toward the magnet, and oscillates around it until damping brings it to rest close to that
magnet. Each of these three possible final resting points is called an “attractor.”
If the pendulum is released at rest at a point far away from any of the magnets,
it follows a very complex trajectory. Eventually, it does come to rest at one of the
Figure 4.8.2 Top: In the magnetic pendulum, a rigid rod is free to swing from a top support.
The bottom of the rod is attached to a small piece of iron, which is attracted to the three
magnets on the surface below. Bottom: Basins of attraction for the three magnets, as simulated
on a computer using an approximate model for the forces. If the pendulum is released from rest
in any of the white regions, it eventually winds up pointing to the white magnet. If it is
released from one of the black regions, it winds up at the black magnet, and similarly for the
gray regions. The overall shading indicates the length of time required for the pendulum to
settle to a final resting place, with darker shading indicating a longer time. (Lower image
courtesy of and © Paul Nylander, www.bugman123.com)
three attractors. However, if we try to repeat the experiment by releasing the pendulum
a second time from the same initial point, we find that it may end up at a different
one of the three attractors. If we repeat the experiment over and over again, the final
resting place appears to be random, even though the system is governed by nonrandom,
deterministic equations.
This sensitivity can be demonstrated through a computer simulation, as shown
in the bottom part of figure 4.8.2. If the pendulum is released in one of the white
colored regions, it eventually winds up at the white magnet. Similarly, if released
in a black region it eventually winds up at the black magnet, and if released in the
gray regions it winds up at the gray magnet. Each colored region is called a “basin of
attraction”; it is conceptually similar to a watershed (the area of land from which rainfall
feeds into a particular river). The boundary between the three basins is extremely
complex. In fact, if you look at it in higher magnification, it looks just as complex as
it does at low magnification. This self-similarity at different scales is the hallmark of
a “fractal.” Many shapes in nature, such as coastlines and tree branches, have fractal
character.
Figure 4.8.3 a: Free body diagram for a pendulum. b: Phase space plot for the damped,
undriven pendulum.
Chaos
If a system with nonlinear restoring forces is also subjected to a periodic drive force,
this can result in “chaos,” in which the system is not only extremely sensitive to initial
conditions, but also is sensitive to minute perturbations at any later time. Even a tiny
perturbation, such as a gentle puff of air, leads to a huge change in the later behavior.12
Furthermore, a chaotic system exhibits very complicated, ongoing nonperiodic motion,
even though the drive force is periodic and the system is fully described by exact
equations.
The damped, driven pendulum provides a good example of chaotic behavior.
Let us find the exact version of the differential equation that governs this system.
We showed in section 2.2 that, for small displacements, the restoring force for the
pendulum is proportional to displacement. However, this is an approximation; as shown
in figure 4.8.3a, the restoring force is Fr = mg sin θ . Therefore, the restoring torque is
τ = −ℓmg sin θ.
12. This sensitivity to initial conditions and small perturbations is often referred to as the “butterfly
effect.” The name stems from a talk given in 1963 by one of the founding fathers of the field of
chaos. During the talk, Edward Lorenz said of chaos theory, “One meteorologist remarked that
if the theory were correct, one flap of a seagull’s wings would be enough to alter the course of
the weather forever.” In later talks, the seagull evolved into a butterfly.
We include a viscous drag force
Fdrag = −bv = −bℓθ̇ ,
which produces an associated torque
τdrag = −bℓ2 θ̇ .
Finally, we include a periodic driving torque τ0 cos ωd t. The angular version of

FTOT = mẍ is τTOT = I θ̈ , which gives us
−ℓmg sin θ − bℓ2 θ̇ + τ0 cos ωd t = I θ̈ .
For a simple pendulum, I = mℓ2 , so that
− ℓmg sin θ − bℓ2 θ̇ + τ0 cos ωd t = mℓ2 θ̈ ⇔

τ0
θ̈ + γ θ̇ + ω02 sin θ = cos ωd t , (4.8.11)
mℓ2
√
with γ ≡ b/m and ω0 ≡ g/ℓ. This is a differential equation in the variable θ . Because
of the sin θ term, it is nonlinear, and therefore there is the possibility that, for appropriate
choices of parameters, the system will show chaotic behavior.
To visualize the behavior of the system, we need a couple of new tools. A phase
space plot is simply a plot with one axis for each variable that is time-dependent and
that is needed to specify the state of the system. (Such quantities are called “dynamical
variables.”) For the case of zero driving torque, there are only two such variables: the
angular position θ and the angular velocity θ̇ . We can specify the state of the system
entirely13 by giving the values of θ and θ̇ . Thus, if we make a plot with θ on the
horizontal axis and θ̇ on the vertical axis, then each point in the plot specifies a state
of the system. If we release the pendulum from rest at θ0 , with zero driving torque, it
follows the phase space path shown in figure 4.8.3b, spiraling in toward rest at θ = 0.
One can show that, for a system to exhibit chaos, there must be at least three
dynamical variables. When we turn on the driving torque, we get the third variable:
ωd t, which is the argument of the cos ωd t term in equation (4.8.11), and is therefore
called the “drive phase.” Now, in order to completely specify the state of the system,
we need to specify θ , θ̇ , and ωd t. You might ask, “Why not just use t as the dynamical
variable instead of ωd t?” The answer is that we don’t actually need to know the time
in order to specify the state of the system, but we do need to know the argument of
the cos ωd t term; we do need to know whether the drive torque is at a maximum or
a minimum, or somewhere in between, and whether it is increasing or decreasing at
the moment in question. Note that ωd t is an angular variable, measured in radians; the
drive torque has the same value at ωd t = 0 as it does at ωd t = 2π .
Since we now have three-dynamical variables, we need a three-dimensional phase
space to represent the system. The axis for ωd t need only run from 0 to 2π . Let’s now
13. Note that we need not also specify θ̈ , since (for τ0 = 0) we can use equation (4.8.11) to calculate
θ̈ from θ and θ̇ . Also note that specifying θ by itself is not enough to determine the state of the
system. For example, if θ = 0, the pendulum could be swinging to the right or to the left.
Figure 4.8.4 a: Phase space plot for the damped, driven pendulum in steady state. The angle θ
is shown on the left-right axis, the angular velocity θ̇ is shown on the axis that is more-or-less
perpendicular to the page, and the drive phase ωd t is shown on the vertical axis. The system
starts at a maximum value of θ , as shown by point a and by the small picture labeled a, then
progresses through points b–e. b: Similar plot for the pendulum with arbitrary initial
conditions, leading to transient behavior that decays over time.
drive the pendulum at low amplitude (low enough that the nonlinearities are not
important). We know that, in steady state, it follows simple harmonic motion:

θ = A cos ωd t − δ ⇒ θ̇ = −ωd A sin ωd t − δ .
2
As an example, we choose ωd = 15 ω0 , which gives δ ∼
= 0. This means that θ = A
and θ̇ = 0 at t = 0. The phase space plot for this steady-state behavior is shown in
figure 4.8.4a; it is a helix that starts at the bottom. The phase space point representing
the system moves up the helix, and reaches the top of the helix at ωd t = 2π , which is
equivalent to ωd t = 0, so the phase space point jumps down to the bottom of the helix
and starts up again.
If we start the system with some arbitrary initial conditions, there will be some
transient behavior near t = 0, which damps away in time, until the system reaches
steady state, as we explored in section 4.5; this behavior is shown in figure 4.8.4b.
Once the transients damp away, the system returns to the helix phase space path that
represents the steady state. Therefore, this helix is the attractor for this system. It is a
Figure 4.8.5 a: Trajectory in the θ − θ̇ plane for pendulum driven at low amplitude. b: A
Poincaré section of this motion. Parts c and f of the figure represent a higher amplitude motion
of the pendulum, and so are at a larger scale than parts a and b. c: Trajectory in the θ − θ̇ plane
for higher drive torque, showing period doubling. d: A Poincaré section of this motion.
e:Trajectory in the θ − θ̇ plane for even higher drive torque, showing chaotic behavior. f: A
Poincaré section of this motion. g: A Poincaré section for a pendulum with less damping than
the one shown in parts a–f, showing chaotic behavior. (Note that, unlike the other figures in the
left column, this is a Poincaré section rather than a phase space plot.) h: Zoom-in on the upper
right part of the Poincaré section, showing that the complexity of the attractor does not
diminish as we zoom in. Parts c–h used with permission from Chaotic Dynamics: An
Introduction, 2nd Ed., by G. L. Baker and J. P. Gollub, Cambridge University Press,
Cambridge, 1996.
more complex attractor than the three points for the magnetic pendulum, and represents
a dynamic behavior, but the idea is the same; it represents the behavior that the system
eventually settles into.
Because it is complicated to look at the three-dimensional phase space plot, one
often instead just shows the projection onto the plane of the θ and θ̇ axes (the θ − θ̇
plane). For the driven pendulum at low-drive amplitude in steady state shown in
figure 4.8.4a, this projection is an ellipse, as shown in figure 4.8.5a.
Now, imagine that we illuminate the pendulum with a strobe lamp, which flashes
once per cycle of the drive torque. At each flash, we measure θ and θ̇ , and plot this
point in the θ − θ̇ plane. The resulting plot is called a “Poincaré section.” For the
steady-state motion shown in figure 4.8.5a, the Poincaré section is simply a point, as
shown in figure 4.8.5b. A single flash of the strobe is enough to create this particular
Poincaré section, because the pendulum is in simple periodic motion. It is called a
section because it is a slice through the three-dimensional phase space. We can choose
to have the strobe flashes occur at ωd t = 0, 2π , 4π , etc., in which case we get a slice at
the bottom of the cube shown in figure 4.8.4a. If instead we have the flashes occur at
ωd t = π , 3π , 5π , etc., then we get a slice halfway up the cube. By varying the timing
of the flashes, we can map out different slices through the attractor. (Again, for the
case of low-drive amplitude, the attractor is the helix shown in figure 4.8.4a).
Next, we gradually increase the amplitude of the drive, τ0 , while keeping ωd fixed.
For our example, we now choose ωd = 32 ω0 . We discussed earlier in this section that,
for small nonlinearity and a symmetric potential energy function, there is a response
not only at ωd , but there is also a third harmonic response at 3ωd , which in this case
equals 2ω0 . Therefore, there can be an interplay between this response and the natural
resonance angular frequency of the system, ω0 . The period of the drive torque is
T = 2ωπd = 23 2ωπ0 . In a time interval 2T , the drive torque goes through exactly two cycles,
the third harmonic response goes through exactly six cycles, while any oscillation at
ω0 goes through exactly three cycles. We can imagine therefore that the periodicity of
the system could change, due to this interplay, from T to 2T , and this “period doubling”
is indeed observed experimentally. The projection onto the θ − θ̇ plane of the phase-
space trajectory is shown in figure 4.8.5c, and a Poincaré section in figure 4.8.5d. In
this case, two successive flashes of the strobe are needed to complete the Poincaré
section; in subsequent flashes, the point representing the system alternates between
the first two points. As we increase the drive amplitude further, the system is driven to
higher amplitudes, and samples the nonlinearity of the restoring force more. Additional
complexities are added to the motion (such as the pendulum going “over the top”), and
the period gets longer, but the motion is still periodic.
Then, with a small additional increase in the drive amplitude, the motion suddenly
changes character. It is no longer periodic, as shown by the θ − θ̇ plane plot in
figure 4.8.5e, and the Poincaré section (figure 4.8.5f) becomes suddenly much more
complex. In this case, many strobe flashes (in principle an infinite number) are needed
to complete the Poincaré section, because the motion is not periodic. However, we see
that the motion is not completely random; even with a large number of flashes, many
parts of the Poincaré section remain blank. Again, the Poincaré section is a slice parallel
to the θ − θ̇ plane through the phase space attractor for the chaotic state (the equivalent
of the helix of figure 4.8.4a). Just from this slice, we can tell that the attractor has a
much more complicated shape, and so it is called a “strange attractor.” If we attempt to
examine the Poincaré section it in more detail, by expanding the plot in a small region,
it still looks just as complex, as suggested in figures 4.8.5g and h. Thus, the strange
attractor exhibits a fractal nature.
There are many plots, simulations, and animations relating to the chaotic pendulum
available on the web; for a few suggestions, go to the web page for this text, and consult
the part for this chapter section.
Chaos is important in a remarkably wide range of fields, including economics,

medicine, meteorology, physics, and chemistry. We have only scratched the surface
of the rich and active fields of chaos and nonlinear dynamics. For further study, you
might begin with the very approachable text by Baker and Gollub.14
terms:
Resonance (4.1)
Steady state (4.1, 4.5)
Amplitude resonance curve (4.1–4.2)
Power resonance curve (4.3)
FWHM (4.3)
Superposition principle for driven systems (4.4)
Transient (4.5)
MRI (4.7)
NMR (4.7)
Magnetic moment (4.7)
Precession (4.7)
RF (4.7)
Gyromagnetic ratio (4.7)
Larmor frequency (4.7)
Rotating reference frame (4.7)
Spin up and spin down (4.7)
Nonlinear differential equation (4.8)
Harmonic generation (4.8)
Frequency doubling (4.8)
Subharmonic resonance (4.8)
Mixing (4.8)
Sensitivity to initial conditions (4.8)
Dynamical variable (4.8)
Basin of attraction (4.8)
Phase space (4.8)
Attractor (4.8)
Strange attractor (4.8)
Chaos (4.8)
Fractal (4.8)
14. Chaotic Dynamics: An Introduction, 2nd Ed., by G. L. Baker and J. P. Gollub, Cambridge
University Press, Cambridge, 1996.

Motion of the support point and resulting drive force (4.1)
Frequency of drive and frequency of steady-state response (4.1)
Phase of drive and phase of response (4.1–4.2)
Height of amplitude resonance curve and Q (4.2)
Width of power resonance curve and γ (4.3)
Height and width of power resonance curve (4.3)
Average power supplied and average power dissipated in steady state (4.3)
General solution for damped driven harmonic oscillator and solution for damped
oscillator (4.5)
Damped driven mass/spring and driven RLC circuit (4.6)
Protons and a child’s top (4.7)
Torque and precession (4.7)
Angular momentum and magnetic moment (4.7)
Symmetry of potential energy curve and which harmonics are generated (4.8)

Advantages of expressing things in terms of dimensionless variables (4.2)

Find the steady-state response amplitude and phase given information about the
oscillator, the damping, and the drive (4.1–4.2)
Find the response of a damped oscillator to superposed drive forces (4.4)
Given the gyromagnetic ratio and Bapplied , calculate the RF frequency that produces
NMR (4.7)
Explain what a Poincaré section is (4.8)
Explain what an attractor is (4.8)
Problems
Instructor: Ratings of problem difficulty, full solutions, and important additional

support materials are available on the website.
4.1 The following statement is incorrect. Provide a corrected version. “If you
drive a damped oscillator at a frequency other than ω0 , the oscillator still
oscillates at its resonance frequency ω0 . The amplitude of the response
depends on how close the drive frequency is to the resonance frequency.”
4.2 Connection to current research: Frequency Modulation AFM. In Atomic
Force Microscopy, a very sharp tip is used to “feel” the topography of the
sample by moving the tip in an x–y pattern over the region to be imaged
while monitoring the force between the tip and sample. For the highest
resolution images, the tip is not allowed to “touch” the sample. Instead, it
is brought close enough to feel an attractive interaction (due to Van der
Waals and other forces). As the tip is moved laterally over the sample,
a feedback circuit adjusts the z-position of the tip to keep this attractive
force constant. The record of the adjustments needed in z then gives the
topography of the sample. The tip is mounted at the end of a cantilever,
which is set into oscillation, as shown schematically in figure 4.P.1a. (The
actual amplitude of vibration is much less than that shown; for the image
in figure 4.P.1b, the vibration amplitude was only 0.8 nm.) The attractive
force between tip and sample is measured through its effect on the resonant
frequency of the cantilever/tip system. Ordinarily, this technique is used in
ultrahigh vacuum, and can give sub-atomic imaging resolution, as shown in
figure 4.P.1b and c. For such an image, the measured forces are due to the
interactions of individual atomic orbitals!
The force of interaction between the tip and sample is shown schematically
in figure 4.P.1d. The position of the tip x is measured relative to the
equilibrium position in the absence of this interaction; note that downward
is defined as the positive direction for x. When the tip moves down (toward
the sample), the force is initially attractive (positive). As the tip starts to
“touch” the surface, the force becomes repulsive (negative). A typical point
near which the AFM might be operated is shown as x0 . (a) Briefly explain
why, near this point, we can model the force between tip and sample as

F∼ = F0 − kts x − x0 , where kts ≡ − dF dx . (b) Note that kts functions like

x0
a spring constant. It is negative because it acts oppositely to the spring of
the cantilever; when the tip is closest to the sample (the point marked “A” in
part a of the figure), the cantilever pulls it up whereas the tip–sample force is
attractive and so pulls it down. Let the spring constant of the cantilever
be k,
and the effective mass of the cantilever and tip be m. Explain why, if kts ≪ k,
then the ∼
oscillation angular frequency is approximately ω = ω0 +
ω, where
ω0 ≡ mk and
ω ≡ ω0 k2kts . (Therefore, by measuring the frequency shift,
one can measure the force of interaction between the tip and sample.)
4.3 It should be clear that, for a damped driven harmonic oscillator that has
reached steady state, the average power supplied to the system by the
driving forces is equal to the average power dissipated by the damping force.
However, for most drive frequencies there are parts of the cycle in which
power flows from the oscillator to the entity providing the driving force (the
“driver”), and other parts of the cycle during which power flows from the
driver into the oscillator. Thus, in general, the energy of the oscillator is not
constant during the cycle, even though its value averaged over the whole
cycle is constant. Assume that k, m, and b are known. For what finite, nonzero
value of ωd is the instantaneous power supplied by the driver exactly equal
to the instantaneous power dissipated by the damping force at every instant
in the cycle? Show your reasoning clearly. Hints: This means that the energy
Figure 4.P.1 a: Schematic representation of frequency–modulation AFM. Note that a

downward deflection of the tip is defined to be positive. b: Subatomic resolution image of an
atom on the tip of the AFM. c: Explanation for the image. As the tip passes over a protruding
atom on the sample (“adatom”), the adatom images the bottom-most atom of the tip, showing
the two lobes of the sp3 atomic orbitals. (The features shown in the image are actually
determined both by the atom on the surface and by the atom on the tip.) d: Schematic diagram
for the force of interaction between tip and sample. Parts a–c © and courtesy of Professors
Jochen Mannhart and Franz Giessibl.
of the oscillator is constant throughout the cycle, so you might wish to start
by writing an expression for the total energy as a function of time, and see
if you can determine what value of ωd would cause the total energy to be
constant.
4.4 A harmonic oscillator has an undamped angular frequency ω0 . It is then put in
a damping medium producing a damping characterized by Q. The oscillator
is driven at a frequency such that, in the steady state, the response x lags
behind Fdrive by 45◦ , meaning that the drive force reaches each peak a little
earlier than x does. The drive force amplitude is F0 , and k is the effective
spring constant for the oscillator.
√ 2
(a) Show that the response amplitude is A = √ 2 F0 Q , and that
k 1+4Q2 −1
ω QF 2
the dissipated power is Pdiss = 02k 0 sin2 ωd t − δ .
(b) Sketch qualitative graphs of the drive, response, potential energy,
velocity, and dissipated power over one complete steady-state cycle.
Align the graphs in a vertical column, or superpose the graphs, so
that the relationships between the five quantities are as clear as
possible.
4.5 A mass is subjected to a spring force Fspring = −kx, a damping force Fdamp =
−bẋ, an oscillating drive force Fdrive = F0 cos ωd t, and an additional
decaying force Fextra = De−β t , where D and β are positive constants.
(a) Write the differential equation for x that describes this system.
(b) Write the simplest possible complex version of your DEQ from part
a, that is, the simplest complex DEQ whose real part is your DEQ
from part a.
(c) Show that z = Aeiϕ eiωt is not a solution for your DEQ from part b,
no matter what the values of A, ϕ , and ω are.
4.6 Driving a pendulum with vertical motion. Make your own simple
pendulum. (One easy way is to squeeze the tea out of a used teabag that
has a string.) First, get your pendulum swinging by holding the end of
the string and moving your hand laterally. Find the resonant frequency,
and measure it roughly. With the pendulum still swinging, see if you can
keep it going by moving your hand vertically instead of horizontally. If you
experiment enough, you should be able to do this, though it only works if the
pendulum is already swinging when you start moving your hand vertically.
(a) What is the ratio of the frequency of your hand motion when you’re
moving your hand vertically divided by the frequency of your hand motion
when you move your hand horizontally, assuming you drive the pendulum at
resonance in both cases? (b) Explain your finding about the frequency ratio
qualitatively.
4.7 Go to the website for this text, and under chapter 4 open the “Damped Driven
Harmonic Oscillator” applet. Use it to answer the following questions:
(a) To use this animation, select a quality factor (Q) of 15 with the slider
and then select a drive frequency by clicking on either of the two
graphs. Try some different values of Q and drive frequency. At what
drive frequency do you get the greatest amplitude?
(b) The phase graph shows the phase relationship between the motion
of the drive and the motion of the oscillator. What is the phase
relationship between the motion of the drive and the motion of the
oscillator at the point that you just found of greatest amplitude?

How about at the two extremes of very low drive frequency and
very high drive frequencies?
(c) Now, click the check box labeled “Show Velocity” and observe the
small velocity indicator to the right of the oscillator. What is its
phase relationship to the drive force at resonance? Why does this
make sense in terms of the power being delivered by the drive force
to the oscillator?
(d) Generally speaking, how does varying Q affect the amplitude and
phase diagrams?
(e) This animation ignores “transients.” Because of this, what state of
a damped-driven harmonic oscillator system are we seeing?
(f) How does the steady-state response frequency compare to the drive
frequency at resonance? Well below resonance? Well above it?
4.8 Show that the peak ofthe amplitude resonance curve shown in figure 4.1.5
or 4.2.1 is at ωd = ω0 1 − 2Q1 2 . Do not use a symbolic algebra program or
calculator.
4.9 What, if anything, does the quality factor Q of a lightly damped mechanical
oscillator have to do with each of the following? Be brief, but as quantitative
as possible. (a) The rate at which the oscillator loses energy in the absence
of a driving force. (b) The “amplification factor” at resonance, that is, the
amplitude of the oscillator’s motion when a driving force at the resonance
frequency is applied, as compared to the amplitude of motion when the same
amplitude driving force is applied at very low frequency. (c) The phase shift
at resonance. (Assume the damping is light enough that amplitude resonance,
velocity resonance, acceleration resonance, and power resonance all occur
at essentially the same frequency.) (d) The full width at half-maximum of the
“power resonance curve,” that is, the graph of power absorbed as a function
of drive frequency.
4.10 A simple pendulum is made by attaching a steel sphere (ρ = 7,850 kg/m3 )
of radius 2.00 cm to the end of a thin string that is 1.00-m long. In this
√
problem, take the resonant frequency to be g/L, where L is the distance
from the support point to the center of the sphere; assume that the damping
is light enough that the differences between ω0 , ωv , and the peak angular
frequency for the amplitude response curve can be ignored. The support point
(at the top of the string) is moved laterally (in the x-direction) in a sinusoidal
pattern at the resonant frequency, with amplitude 1.00 μm = 1.00 × 10−6 m.
What is the steady-state amplitude of the sphere’s motion in the x-direction?
Assume that the air provides viscous damping, that is, that the airflow is not
turbulent.)
4.11 Using the results of section 4.1, show explicitly that, in steady state, the net
energy provided by the drive force to a damped driven oscillator over one
period equals the energy dissipated by the damping force.
4.12 Steady-state response to a square wave. In section 8.4, we will show that
the square wave (shown in black in figure 4.P.2) can be represented as an
Figure 4.P.2 Synthesizing a square wave

from sinusoids.
infinite sum of sinusoids:
cos 3ωt cos 5ωt

Square wave = cos ωt − + − ···
3 5
The sum of the first three terms is shown in gray in the figure—you can
see that it roughly approximates the square wave. The preciseness of the
approximation improves as more terms are added. If a force with a square
wave time dependence is used as the drive force for a damped driven
harmonic oscillator, is the steady-state response a square wave? Explain
your answer thoroughly.
4.13 The curve of power dissipated versus drive frequency for a damped driven
oscillator is shown in figure 4.P.3. (The power dissipated averaged over
a cycle in steady state is equal to the power absorbed from the drive
force.)
(a) What is the Q for this system?
(b) The oscillator is driven at resonance until steady state is achieved.
The drive force is then suddenly turned off. About how long does it
take for the oscillations to die away? (Give your answer in seconds;
all the numerical information you need is in the graph.)
(c) Now, instead, the oscillator is driven with fd = 850Hz. The drive
force is then turned off; thereafter, the oscillator continues to
oscillate for a short time. At what frequency (approximately) does
Figure 4.P.3 A power resonance curve.

it oscillate during this part of the experiment (after the drive force
is turned off)? Explain briefly.
4.14 For the damped driven harmonic oscillator, find expressions for the adjustable
constants A0 and ϕ that appear in equation (4.5.2) in terms of the initial
position x0 , the initial velocity ẋ0 , the steady-state amplitude A, the steady-
state phase shift δ , ωv , and ωd . Do not use a symbolic algebra program or
calculator. Hints: Don’t expect the final result to be “neat.” You should be
able to show that tan ϕ = DB , where B and D are constants involving x 0 , ẋ0 ,
ωv , and γ . To find A0 , you will need to find cos ϕ . To do this, draw a right
triangle with B and D as the two legs.
4.15 A weasel holds an object of mass m at its equilibrium position x = 0. The
mass hangs from a spring of constant k in a medium which provides Fdamp =
−bẋ. The support point for the spring is moving, with position given by
xC = Ad cos ωd t. The angular drive frequency ωd happens to exactly equal
k
m . (Don’t forget this, and its implications—otherwise, the math is a mess!)
At t = 0, the weasel releases the mass, so that it can begin moving. Given
these initial conditions, what is the complete solution x(t)? Everything in
your solution should be expressed in terms of the symbols above only.
However, you may make your life easier by defining symbols of your own
in terms of those above, then expressing your solution using these new
symbols.
4.16 An electrical engineer wishes to design an RLC oscillator, of the type shown
in figure 4.6.1. The engineer wants the resonant response to be very strong,
but also wants the system to respond strongly to a broad range of frequencies.
Explain briefly why both goals can’t be achieved.
4.17 Radio station WJJZ has hired you to help design radio receivers to go in
the waiting rooms of doctors’ offices. Each of these radios is to receive the
WJJZ broadcast only—there will be no “tuning” knob. One of your fellow
engineers hands you part of the schematic diagram for the radio, shown in the
top part of figure 4.6.2. The input voltage to the LCR circuit from the antenna
is
VIN = V0 cos ωd t, and the output voltage is taken across the capacitor
(and then sent to the rest of the radio). “I already chose L = 0.1 μH,” he
says. “You pick out the values for R and C.” (Note: “μ” means “×10−6 ”).
(a) The broadcast frequency of WJJZ is f = 101.9 MHz. (Note: “M”
means “ ×106 ”). What value of C should you pick so that the above
circuit will resonate at this frequency? (Assume the circuit is very
lightly damped.)
(b) Rival station WART broadcasts at f ′ = 101.5 MHz. The steady-
state oscillation amplitude of your circuit (i.e., the amplitude of

VOUT ) must be 100 times smaller at this frequency than at the
WJJZ frequency, assuming equal drive (input) voltages for the two
frequencies. What is the required value for R ? Again, assume very
light damping. Your answer need only be correct to within a few
percent. Hint: Don’t forget that q = CV ⇒
VOUT = Cq .
Figure 4.P.4 A parallel RLC oscillator.
(c) Of course, the WJJZ management would be happy if the amplitude

for WART is even less than 100 times below that for WJJZ. Bearing
this in mind, is the R you just calculated the maximum allowed R
or the minimum allowed R? Explain.
(d) What is the Q of your circuit?
4.18 The parallel RLC oscillator. Consider the circuit shown in figure 4.P.4.
This is driven by applying a current I0 cos ωt, as shown. Assume the system
is underdamped. (a) What is the resonance frequency ω0 ? (b) What is the
FWHM of the power resonance curve?
4.19 For this problem, the RLC oscillator shown in figure 4.6.1 is driven not
simply by the voltage V0 cos ωt, but instead by the more complicated voltage
V1 cos ω1 t + V2 sin ω2 t. What is the charge on the capacitor, q (t), once the
system reaches the steady state?
4.20 A patient is undergoing MRI. A particular proton in her brain experiences a
total applied magnetic field (including contributions from nearby electrons)
of 1.50 T. The gyromagnetic ratio for a proton is γ = 2π · 42.576 MHz/T. (a)
What frequency of RF radiation will this proton respond to most strongly?
(b) For simplicity, in this part of the problem assume that the net magnetic
moment and the net angular momentum of the protons in the local region of
the brain can be treated by considering only their z-components. The main
magnetic field in the MRI machine is applied in the z-direction. The magnetic
field of the RF radiation is described by Brad = B0 cos ωt î. If B0 = 30.0μT,
how long of an RF pulse should be applied to rotate the magnetization from
the z-direction into the x–y plane?
4.21 Show that cos3 θ = 41 cos 3θ + 34 cos θ . (You will probably want to use a
symbolic algebra program to do this.) This trigonometric identity is used in
section 4.8.
4.22 Online exploration of a chaotic pendulum. Go to the entry for this problem
on this book’s website, and follow the instructions there.
5 Symmetric Coupled Oscillators
and Hilbert Space
Gentlemen, I do not see that the sex of the candidate is an argument against her
admission as a privatdozent. After all, the senate is not a bathhouse.
– David Hilbert, arguing in favor of allowing Emmy Noether a position at the
University of Göttingen
5.1 Beats: An aside?
If you strike a tuning fork and listen, the resulting pressure variation at your ear is
sinusoidal. We might write the variation in pressure (relative to the average background
pressure) as
x1 = Re z1 , where z1 = Aeiω1 t .
Now, imagine that we strike two tuning forks of slightly different angular frequencies,
ω1 and ω2 . The resulting pressure variation at your ear is simply the sum of the
variations from the two forks. A/V: You can hear this right now by going to this
book’s web page and clicking on the “listen to beats” link under this chapter section.
This sound is remarkable – you perceive it as a single note (i.e., a single frequency)
with an oscillating loudness! We can see how the loudness variations come about
graphically, as shown in figure 5.1.1a. This effect turns out to be of tremendous
importance for our future studies, so let’s see how it comes about mathematically.
To simplify the math, we’ll look at the case where the two amplitudes are the
same, that is,
z = z1 + z2 = A(eiω1 t + eiω2 t ). (5.1.1)
It will help to define
ω1 − ω2 ω1 + ω2
ωe ≡ and ωav ≡ . (5.1.2)
2 2
(The subscript e is used for the first one because we will see that it is the angular
frequency of an envelope function.) You should verify right now that
ω1 = ωav + ωe and ω2 = ωav − ωe .
137
Figure 5.1.1 a: Two oscillations of slightly different frequencies (top) are added together
(bottom). At first, the two waves are in phase, and so the sum has large amplitude. As time
progresses, the two waves get out of phase, so the amplitude of the sum gets small. With
further passage of time, the waves start to get back into phase. b: A rapidly oscillating function
multiplied by a slowly varying envelope function of period 2π/ωe (top) produces beats
(bottom).
Therefore,

z = A ei(ωav +ωe )t + ei(ωav −ωe )t
Your turn: Use the above and Euler’s equation (eiθ = cos θ + i sin θ ) to show that
z = 2Aeiωav t cos ωe t
This means that
x = Re z = 2A!
cos ωav t cos ωe t
! !
constructive rapid slow oscillation
interference due to the transition
at maximum oscillation from constructive to
destructive interference
(and back):
an envelope function
We see that x is the product of three terms: an overall amplitude, a rapid oscillation at
the average angular frequency, and a slow oscillation (corresponding to the oscillating
loudness). The slow oscillation is called an “envelope function,” because you can think
of it as a time-dependent amplitude for the rapid oscillation, that is,
x = 2A cos ωe t cos ωav t (5.1.3)

! !
time-dependent rapid
amplitude oscillation
This multiplication is shown graphically in figure 5.1.1b. The slow variations in

amplitude are called “beats,” and you’ll often hear people talking about one frequency
“beating” against another, that is, the signals at two close frequencies combine to
Chapter 5 ■ Symmetric Coupled Oscillators and Hilbert Space 139
create beats. Following the usual convention, we define the beat period as the period
between maximum amplitudes, as shown, that is,
π 2π
Tbeat ≡ = (5.1.4)
ωe ω1 − ω2
Note that this is half the period of the envelope function. The beat frequency is then
simply
1 ω1 − ω2
fbeat = = ⇒
Tbeat 2π
fbeat = f1 − f2 (5.1.5)
Self-test (answer below1 ): Serious musicians sometimes use this phenomenon for
tuning their instruments. Two guitar players are trying to bring their guitars to the same
pitch. They both pluck their lowest string, and hear beats. One of the players begins
adjusting the pitch of her string, and the beat period gets shorter. Should she keep
adjusting in this direction?
Applet: Click on the “beats applets” link under this chapter section on this book’s web
page to see other ways of understanding beats.
5.2 Two symmetric coupled oscillators: Equations of motion
You probably have seen much of the material up to this point in previous courses, though
not at the same level of detail, nor with such a thrilling and masterful presentation.2
Now, we begin the truly new (and really neat!) stuff. We will begin by considering two
coupled oscillators, and this will lead directly to our treatment of waves (which we’ll
view as a large set of coupled oscillations), and to the most important ideas underlying
quantum mechanics.
The simplest and easiest-to-draw example of coupled oscillators is a set of two
pendula, connected by a spring, as shown in figure 5.2.1a. To start with, we use pendula
with the same length and the same mass. Note that we define the position of each mass
relative to its own equilibrium point, so that when the system is at equilibrium (both
pendula hanging straight down), we have x1 = 0 and x2 = 0. Also, for simplicity we
will ignore damping and driving forces for now.
If you displace mass 1 from equilibrium while holding mass 2 steady (at x2 = 0),
and then release both, you will observe a remarkable behavior. A/V: You can see a
1. If the beat period is getting shorter, the beat frequency is getting larger. According to equation
(5.1.5), this means the two frequencies are getting farther apart. Therefore, she should stop and
adjust in the other direction instead. Perfectly matched pitches will result in a very long beat
period (in principle infinitely long).
2. I asked my wife for ideas on how to punch up the humor of this sentence, but her suggestions
were all too funny to include in a physics text.
Figure 5.2.1 a: Coupled pendula. b:

Position of mass 1 plotted as a function
of time t. c: If the probability density
| (x , t)|2 of an electron is initially
localized on the left of this double-well
potential, it will oscillate to the right side,
then back to the left, etc.
video clip of this by clicking on the “coupled pendula” video under this chapter section
on this book’s website. At first, mass 1 oscillates back and forth, and mass 2 hardly
moves. Gradually the energy is transferred to mass 2; now mass 2 oscillates and mass
1 hardly moves. Gradually, the energy is transferred back to mass 1, and the process
starts over. If you watch just one of the two masses, its behavior should remind you
of something…something you’ve just been studying…what is it? Hmmm… Beats! If
you plot the position of either mass as a function of time, it looks as if it’s beating, as
shown in figure 5.2.1b. Recall that beats result when you superpose two oscillations of
slightly different frequencies. We will see that the best description of this motion of the
coupled pendula is in terms of a superposition of two different oscillations. However,
it is not at all obvious at this point why simply displacing one of the pendula from
equilibrium should somehow excite two different oscillations.
One observes similar behavior for any set of two symmetric coupled oscillators.
For the pendula, the coupling is obvious from just looking at the system, but there are
other coupled oscillator systems for which the coupling is more subtle. For example,
consider two cantilevers attached to the same support, with both cantilevers initially at
rest. If one is displaced from equilibrium and then released, its oscillation causes a tiny
amount of flexing in the support, which provides a coupling to the other cantilever. If the
damping is small enough for the oscillations to persist for a long time, the oscillation
energy starts to transfer to the other cantilever. (The transfer is slow, because the
coupling is weak.)
Another example comes from quantum mechanics. If an electron is initially
localized in the potential well on the left in figure 5.2.1c, its probability
density3 eventually moves entirely over to the right well, and then back to the left,
just as the energy of the two coupled pendula moves back and forth.
Getting back to the example of the coupled pendula, our task (similar to those in
the previous chapters) is to find x1 (t) and x2 (t) given the initial conditions. We begin our
three-step procedure, through instead of steps 2 and 3 we will use a different method,
in section 5.3.
Your turn: Convince yourself that the force due to the spring on mass 1 is given by

Fspring, 1 = −k x1 − x2 .
Hint: First find the force when mass 1 is held at x1 = 0 but mass 2 is allowed to move.
Then find the force when mass 2 is held at x2 = 0 but mass 1 is allowed to move.
The above equation shows mathematically that the force on mass 1 depends not only
on the position of mass 1, but also on the position of mass 2, so that the motion of
the two masses is “coupled.” The total force on mass 1 is the sum of the spring force
and the “pendulum force” discussed in chapter 3. Recall that, for small displacements
from equilibrium, the tension in the string and gravity combine to produce a spring-like
force given by
mg
Fpendulum = − x
ℓ
Therefore, the total force on mass 1 is
mg
F1 = − x1 − k x1 − x2 = mẍ1
ℓ
g k
⇔ ẍ1 + x1 + x1 − x2 = 0. (5.2.1a)
ℓ m
Similarly, the differential equation governing the motion of mass 2 is
g k
ẍ2 + x2 + x2 − x1 = 0. (5.2.1b)
ℓ m
These are a set of two DEQs, representing the motion of two objects. The equations
are “coupled” because x2 appears in the first equation and x1 appears in the second
equation. This mathematical coupling of the equations is a direct result of the physical
coupling of the masses. The coupling makes the equations difficult to solve, because
we must simultaneously find the solutions x1 (t) and x2 (t) for both equations. It would
be much easier to find solutions if we could decouple the equations, that is, if we could
find a way to write two other second-order differential equations that also completely
describe the system, but which aren’t coupled. There is a general recipe for doing this,
and we will study it, but for this simple case we will use physical insight instead.
3. The concept of probability density was discussed in section 1.11.

5.3 Normal modes
If we could succeed in describing the system with two uncoupled differential equations,
then (because they’re uncoupled), each would represent a completely independent type
of motion, without any energy transfer between the two. This would be analogous to the
two pendula shown in figure 5.2.1a, but with no spring between. These two uncoupled
oscillators are described by two uncoupled DEQs:
mg g ⎫
− x1 = mẍ1 ⇔ ẍ1 + x1 = 0⎪
⎬
ℓ ℓ
g Two uncoupled DEQs (5.3.1)
ẍ2 + x2 = 0 ⎭
⎪
ℓ
In this analogy, we can set mass 1 moving, and no energy is ever transferred to mass 2.
Mass 1 continues oscillating in a simple “steady-state” motion, that is, it oscillates with
constant amplitude.
For the coupled pendula, can we think of a way in which the system can move
in a steady state, in which each mass oscillates with constant amplitude? We will call
this way of moving a “normal mode.” In fact, we need to think of two normal modes,
since we know that we’ll need two second-order DEQs to describe the system. (After
all, we started with two DEQs, equations 5.2.1a and b.)
One of these normal modes is easy to guess, as shown in figure 5.3.1. This is
called the “pendulum mode”; the two masses swing in phase. Since x1 is always
equal to x2 , the spring never gets stretched or compressed, so it never exerts any
force. If there is no damping, the system would continue swinging in this mode
forever.
Concept test: See if you can figure out the other normal mode for this system. In other
words, figure out a different way in which the system can move forever in a steady state.
Don’t look any further until you’ve put an honest effort into thinking about this.
This other normal mode is called the “breathing mode,” and is shown in figure 5.3.2a.
This is a completely antisymmetrical motion. Therefore, if we excite this mode only,
there is no way to generate the symmetrical motion that would lead to the pendulum
mode. Another way to understand why this mode would continue in a steady state is
that the center point of the spring never moves. Therefore, you could attach the center
Figure 5.3.1 In the pendulum mode, the system oscillates between these two configurations.
Each mass oscillates with constant amplitude.
Figure 5.3.2 a: In the breathing

mode, the system oscillates
between these two configurations.
Each mass oscillates with
constant amplitude. b: In the
breathing mode, the center point
of the spring doesn’t move, so we
could pretend that it’s attached to
a wall.
point to an immovable anchor without changing anything, turning the two pendula into
two uncoupled oscillators moving in unison (though in opposite directions).
Now that we’ve found these two normal modes, let’s find the two differential
equations that describe them, and then see how they’re related to the original differential
equations 5.2.1a and b. We can characterize the pendulum mode by defining
1
sp ≡ √ x1 + x2 . (5.3.2)4
2
When the system moves in the pendulum mode, there is an oscillation of sp . If the
√ √
system is “purely” in the pendulum mode, then x1 = x2 , so sp = 2x1 = 2x2 ,
so this definition might seem to be of little use. However, we will soon consider
more complicated motions, involving a superposition of the pendulum mode and
the breathing mode, and then we will really need this definition. You can show in
problem 5.17 that sp always shows a simple harmonic oscillation, even when the
motions of x1 and x2 are more complicated (because of a superposition of the two
modes). Therefore, sp is called the “normal mode coordinate” for the pendulum
mode.
In the pure pendulum mode, the spring never stretches, so the motion of each
pendulum is that of a simple pendulum. Therefore, we have
mg g
F1 = mẍ1 = − x ⇔ ẍ1 + x1 = 0.
ℓ 1 ℓ
System in pendulum mode
√ √
In the pendulum mode, we have sp = 2x1 , so that s̈p = 2 ẍ1 . Therefore, we could
just as well write
g
s̈p + sp = 0. (5.3.3)
ℓ
This differential equation describes the motion of the pendulum mode. It has the form
k
of a simple harmonic oscillator DEQ, ẍ + x = 0. Therefore, the angular frequency
m
√
4. The factor of 1/ 2 in this definition might seem unneeded, but eventually it will make things
easier to think about.

k
of oscillation is given by the equivalent of ω = :
m

g
ωp = (5.3.4)
ℓ
There is a different way of deriving the DEQ (5.3.3), which is quite revealing. Inspired
by the definition of sp , equation (5.3.2), we try adding together equations (5.2.1), after
√
multiplying each by 1/ 2:
" #
1 g k
√ ẍ1 + x1 + x1 − x2 = 0 (5.2.1a)
2 ℓ m
" #
1 g k
+ √ ẍ2 + x2 + x2 − x1 = 0 (5.2.1b)
2 ℓ m
______________________________
1 g
√ ẍ1 + ẍ2 + x1 + x2 = 0
2 ℓ
1
Using the definition sp = √ (x1 + x2 ), this becomes
2
g
s̈p + sp = 0, (5.3.5)
ℓ
which is just the same as equation (5.3.3). Note that to do this second derivation, we
did not require that the system be in pendulum mode, so this equation must always
hold, whether the system is in the pendulum mode, the breathing mode, or some more
complicated motion.
Now, for the breathing mode. We can characterize the breathing mode by defining
1
sb ≡ √ x1 − x2 . (5.3.6)
2
When the system moves in the breathing mode, sb oscillates.

√ (If the system is “purely”
in the breathing mode, then x2 = −x1 , so sb = 2x1 .) Again, you can show in
problem 5.17 that, even when the system is in a superposition of the two modes,
sb still moves in simple harmonic fashion. Therefore, sb is called the normal mode
coordinate for the breathing mode.
Since the center of the spring never moves, we could imagine replacing the right
half of the apparatus by a brick wall, as shown in figure 5.3.2b; this would not affect the
motion of mass 1. Therefore, the motion of the left mass is that of a simple harmonic
oscillator; we just need to figure out the effective spring constant. (Of course, once
we have found the motion of the left mass, we can easily find the motion of the right
mass, since, in the breathing mode, x2 = −x1 .) There are two forces acting on the left
mg
mass: the “pendulum force,” with effective spring constant , and the force from the
ℓ
spring, which is effectively half its original length. Recall from section 2.3 that when
the length of a spring is cut in half, the spring constant is doubled. Applying this to our
mg
analysis of the breathing mode, the effective total spring constant is + 2k, so that
ℓ

mg g 2k
F1 = mẍ1 = − + 2k x1 ⇔ ẍ1 + + x1 = 0. (5.3.7)
ℓ ℓ m
System in breathing mode
√
Since (in the breathing mode), we have sb = 2x1 , we could just as well write

g 2k
s̈b + + sb = 0 (5.3.8)
ℓ m
This differential equation describes the motion of the breathing mode. It has the form
k
of a simple harmonic oscillator DEQ, ẍ + x = 0. Therefore, the angular frequency
m
k
of oscillation is given by the equivalent of ω = :
m

g 2k
ωb = + (5.3.9)
ℓ m
We see that the breathing mode has a higher frequency than the pendulum mode,
and that the difference in frequencies increases with the strength of the coupling,
represented by k. We’ll discuss this important feature at length in section 5.7.
As for the pendulum mode, there is a different, more mathematical way of deriving
√sb , equation (5.3.6), we try subtracting
the breathing mode. Inspired by the definition of
equations (5.2.1), after multiplying each by 1/ 2:
" #
1 g k
√ ẍ1 + x1 + x1 − x2 = 0 (5.2.1a)
2 ℓ m
" #
1 g k
− √ ẍ2 + x2 + x − x1 = 0 (5.2.1b)
2 ℓ m 2
_________________________________________
" #
1 g 2k
√ ẍ1 − ẍ2 + x − x2 + x − x2 = 0
2 ℓ 1 m 1
1
Using the definition sb = √ (x1 − x2 ), this becomes
2

g 2k
s̈b + + sb = 0, (5.3.10)
ℓ m
which is just the same as equation (5.3.8). Again, we did not need to require that
the system be in breathing mode, so this equation must always hold, whether the
system is in the pendulum mode, the breathing mode, or some more complicated
motion.
Recap: We began by describing the system of two coupled pendula in terms of the
motion of each mass, as described by the pair of differential equations
⎧
g k
⎨(5.2.1a): ẍ1 + x1 +
⎪ x1 − x2 = 0
DEQs describing the pendulum bobs: ℓ m
⎩(5.2.1b): ẍ + g x + k x − x = 0
⎪
2 ℓ 2 m 2 1
The motion of each pendulum bob can be complicated, with energy being transferred
back and forth between the two bobs, and it is difficult to solve these differential
equations because they are coupled.
Using physical insight, we discovered two simpler ways in which the system could
move, the “pendulum mode” and the “breathing mode.” When the system is in one of
these modes, the behavior is a simple oscillation, with no transfer of energy from one
mode to the other. We showed that the behavior of these two modes could be described
by the pair of differential equations
⎧ g
⎨(5.3.5) : s̈p + ℓ sp = 0
⎪
DEQs describing the normal modes:
g 2k

⎩(5.3.10) : s̈ +
⎪ + s =0
b ℓ m b
where
1 1
sp ≡ √ x1 + x2 and sb ≡ √ x1 − x2
2 2
are the “normal mode coordinates.” Since these DEQs have the form of simple harmonic
oscillators, we could easily see that

g
ωp = (5.3.4)
ℓ
and

g 2k
ωb = + (5.3.9)
ℓ m
We showed that the second pair of DEQs could be derived from the first simply by
adding or subtracting them.
Note that we don’t really need to continue with steps 2 and 3 of our three-step procedure
for solving the differential equations, since we’ve just shown that the motion of this
system can be understood in terms of two simple oscillators (the pendulum mode and
the breathing mode). However, we do need to get a better understanding of how these
modes combine to produce the complex behavior of the pendulum bobs that you saw
in the video.
5.4 Superposing normal modes
Because the pair of DEQs (5.3.5) and (5.3.10) that describes the two normal modes can
be derived from the pair of DEQs (5.2.1a) and (5.2.1b) that describes the motion of the
two pendulum bobs, each pair represents just as good a way of describing the complete
Figure 5.4.1 a: Any initial condition can be expressed as an appropriately weighted

superposition of the normal modes. b: Express this initial condition as a weighted
superposition of the normal modes.
behavior of the system as does the other. However, because each normal mode acts as
a simple oscillator (completely independent of the other mode), it is easy to predict its
behavior in time.
By choosing the phases and amplitudes of the two modes correctly, we can create
any desired initial condition for the two pendulum bobs. For example, as shown in
figure 5.4.1a, we can create the initial condition shown in the video clip (the one
you watched at the beginning of section 5.2) by adding together equal amounts of
pendulum and breathing modes.
After we let go, the pendulum and breathing modes oscillate independently.
For such a mixture of modes, we can’t directly observe the oscillations of each
mode. Instead, we see the effect of the combination of both modes on each of the
two masses:
1 1
sp ≡ √ x1 + x2 and sb ≡ √ x1 − x2
2 2
1 1
⇔ x1 = √ sp + sb and x2 = √ sp − sb
2 2
(5.4.1)
A remarkable insight: Thus, the “beating” behavior that we observe, say for mass
1,
g
is due to the superposition of the pendulum mode ( at angular frequency ωp =
ℓ

g 2k
and the breathing mode (at a different angular frequency, ωb = + ! (Recall
ℓ m
from section 5.1 that superposing two oscillations of equal amplitude but different
frequencies results in beating.)
In fact, we now understand that any behavior of the system can be understood in
terms of a superposition of the two normal modes.
Your turn: Describe the initial condition shown in figure 5.4.1b (both pendula are
initially at rest) in terms of a superposition of the two normal modes, in a way similar to
figure 5.4.1a.
Example: A set of coupled pendula has m = 0.10 kg for both masses, but k and ℓ are
unknown. At t = 0, the left mass is held at x1 = 2.0 cm and the right mass is held at
x2 = 0 cm. The masses are then released. At first, the left mass oscillates with a period
of about 1.1 s, and the right mass is nearly motionless. At t = 10 s, the two masses are
oscillating with approximately equal amplitudes. At t = 20 s, the right mass is oscillating
strongly, and the left mass is nearly motionless. At t = 30 s, the two masses are oscillating
with approximately equal amplitude. At t = 40 s, the left mass is oscillating strongly, and
the right mass is nearly motionless. Find approximate values for k and ℓ.
Solution: This beating behavior is caused by mixing together the two normal modes,
with their two characteristic frequencies. For example, the motion of the left mass is
a superposition of two different sinusoids of angular frequencies ωb and ωp , and the
variations in the amplitude for the left pendulum are due to the beating that results from
this superposition. The time between maximum amplitudes (e.g., for the left pendulum)
1
is 40 s, so that the beat frequency is fbeat = Hz .
40
We have that
(5.1.5) : fbeat = f1 − f2 ,
ωb ωp
where in this case f1 = fb = and f2 = f p = . So,
2π 2π
1 π
fb − fp = Hz ⇒ ωb − ωp = rad/s . (5.4.2)
40 20
2π 2π 4π
The period of the fast oscillations is given by Tfast = = ω +ω = ,
ωav b p ωb + ω p
2
4π
and we’re told that this equals 1.1 s for this example. Therefore, (1.1 s) = ⇔
ωb + ωp

4π
ωb + ω p = rad/s . Subtracting this from equation (5.4.2) gives
1.1
π
ωb − ω p = rad/s
20
"
#
4π
− ωb + ω p = rad/s
1.1
______________________
π
4π
−2ωp = rad/s − rad/s ⇔
20 1.1

g
ωp = 5.6 rad/s =
ℓ
⇒ ℓ = 32 cm
Finally,
π
ωb − ω p = rad/s ⇒ ωb = 5.7 rad/s
20

g 2k m 2 g
ωb = + ⇒k= ωb −
ℓ m 2 ℓ
⇒ k = 0.088 N/m
5.5 Normal mode analysis, and normal modes as an alternate

description of reality
The breathing mode and the pendulum mode are the two normal modes of the coupled
pendulum system. We will see that we can find similar normal modes even for very
complicated systems of many masses interacting through many different forces. All
these normal modes share the characteristics of the breathing and pendulum modes
shown above, leading us to the definition:
Definition of a normal mode: A way in which the system can move in a steady state,
in which all parts of the system move with the same frequency. The parts may have
different (perhaps zero or negative) amplitudes.
Given any set of initial conditions, we can determine the fractions of pendulum-mode
and breathing-mode which are needed to produce those initial conditions. This process
is called “normal mode analysis.” We can then use these normal mode amplitudes to
determine the future behavior of the system as a function of time:
Normal mode analysis:

1. Express the initial state of the system in terms of a superposition of the normal
modes, with an appropriate amplitude and phase for each mode.
2. Write the time dependent expressions for each normal mode coordinate,
e.g. sp = Re Ap ei(ωp t +ϕp )
3. If desired, form the appropriate combination of the normal mode coordinates to
find the positions of the masses (x1 , x2 , etc.)
For now, we will only consider situations in which the initial velocities are zero, since
this makes the math easier and shows the important ideas. However, we will treat the
more general case later in this chapter. It is easiest to explain normal mode analysis
with an example:
Core example: A set of coupled pendula has m = 0.10 kg, ℓ = 0.15 m, and k = 5.0 N/m.
At t = 0, the left mass is held at x1 = 1.0 cm and the right mass is held at x2 = 3.0 cm.
The masses are then released. What is the position of mass 1 at t = 5.0 s?
Solution: First we make the normal modes description of the system. As we saw in
chapter 1, for a single oscillator that begins at rest, the initial position is equal to the
amplitude of the motion. Similarly in this case, because the initial velocities are zero,
the amplitude for each normal mode is equal to the initial value of the normal mode
1 1
coordinate,5 as defined by sp ≡ √ x1 + x2 and sb ≡ √ x1 − x2 .
2 2
continued
5. You might be concerned that the two modes might have nonzero initial velocities ṡp0
and ṡb0 which, when added together, could produce zero initial velocity for both pendula.
4.0 √ 2.0
Plugging in the initial values of x1 and x2 , we get sp = √ cm = 2 2 cm and sb = − √
√ 2 2
cm = − 2 cm. Each mode acts as an independent oscillator. Since the initial velocities are
i(ωt +ϕ )
zero, the phase factor ϕ in the solution s = Re Ae is zero (as we saw in chapter 1),
so we can simply use s = A cos ωt. So, we have:
√ √
sp = 2 2 cm cos ωp t and sb = − 2 cm cos ωb t .
1
From equation (5.4.1), we have x1 = √ sp + sb , so
2
x1 = (2 cm) cos ωp t − (1 cm) cos ωb t .

g
The angular frequencies of the normal modes are ωp = = 8.1 rad/s and ωb =
ℓ
g 2k
+ = 13 rad/s. Plugging these and t = 5.0 s into the above expression for x1
ℓ m
gives x1 (t = 5.0 s) = −1.9 cm.
Why do we care so much about describing things in terms of normal modes? There
are several reasons. First, as shown in the above “core by example” section, and as we’ll
see as we progress through the book, the normal modes description provides by far the
easiest way of describing the behavior of a complicated system as a function of time,
given the initial conditions. Secondly, our two most important senses, sight and hearing,
perceive the world in a way that is closely related to normal modes. For example, when
you hear a musical note being played, you don’t perceive a rapid oscillation of air
pressure, but rather you hear a single pitch. When you see something that emits blue
light, you don’t perceive a rapidly oscillating electric field; instead, you see “blue.”
Normal modes are also very important in understanding the spectra of chemicals; the
infrared spectrum reveals the frequencies of the vibrational normal modes.
Finally, the normal mode picture presents an alternate, and a very powerful,
way of looking at the world. We will see that, for systems in which the motion is
one-dimensional, the number of normal modes is equal to the number of particles.
For example, in the two-pendulum system there were two normal modes. In a three
pendulum system, there would be three normal modes, etc. For each system, we can
either choose to describe the motion of each particle, or we can instead choose to
describe the motion of each normal mode. In a somewhat fantastic analogy, imagine
trying to completely describe a person. One way of doing it (analogous to describing the
positions of the two masses in our coupled pendula) would be to specify the positions
and velocities of all the electrons and nuclei that make up the person. But, instead,
we might say, “This person is a mixture of 40% Hillary Clinton, 20% Frank Sinatra,
20% Snoop Dogg, and 20% Julia Roberts.” Such a description would be similar to
However, for the pendulum mode the velocities of the two pendula are always equal, while
for the breathing mode they are always opposite. Therefore, while it is possible to add cancelling
velocities for one of the two pendula, the velocities for the other would not cancel. The only way
to get zero initial velocity for both pendula is to have zero initial velocity for both modes.
the normal modes way of describing the coupled pendula—the system is described in
terms of its archetypical behaviors. Such a description places no limits on the system,
and does not require any less information (to describe an arbitrary person in terms
of personality archetypes might require 1025 different archetypes), but is often more
revealing.
Energy of a superposition state

1
By equation (5.4.1), x1 = √ sp + sb . So, when the system is purely in the pendulum
2
Ap
mode, the amplitude of the left pendulum bob is √ , where Ap is the amplitude of the
2
pendulum mode, that is, the amplitude of sp . Therefore, the energy of the left bob is
Ap 2

1
keff √ , where the effective spring constant keff is entirely due to the “pendulum
2 2
mg g
force,” so that keff = . Since ωp = , we could also write that the energy of the
ℓ ℓ
1
left bob is 4 mωp2 A2p . The right bob has the same amount of energy, so that the total
energy is 12 mωp2 A2p . For convenience, we define Ep ≡ 21 mωp2 (1 m)2 ; this is the energy
the pendulum mode would have for an amplitude of 1 m. Using this, the total energy
A2p
of the system in the pure pendulum mode can be expressed as Ep .
(1 m)2
If instead the system is in a pure breathing mode, then identical arguments show
A2b
that the total energy is Eb , where Ab is the amplitude of motion of the left bob
(1 m)2
in this breathing mode (which of course is the same as that of the right bob), and
Eb ≡ 12 mωb2 (1 m)2 is the energy the mode would have for an amplitude of 1 m.
Now, let’s consider the energy when the system is in a state in which the two
modes are superposed. Since each mode can be considered as a completely independent
oscillator, the total energy is simply the sum of the energies of the two modes, that is,
A2p A2b
ETOT = E +
2 p
Eb (5.5.1)
(1 m) (1 m)2
We see that this is a sort of weighted sum, with the weights given by the squares of the
amplitudes of each mode. We’ll see in the next few pages that this is closely related to
the energy of a quantum mechanical system that is in a superposition of two quantum
states.
Analogy between quantum superpositions

and normal mode superpositions
In quantum mechanics, normal modes play an absolutely central role. For example,
each electron in an atom can be in one of several “quantum states,” labeled 1s, 2s, 2p,
etc. Each of these states is a different normal mode for the electron. For a quantum
mechanical particle, there is a simple relationship between the total energy of the
particle and the angular frequency of oscillation:
E = h̄ω,
h
where h̄ ≡ . (Both h̄ ≈ 1.055 × 10−34 J s and h ≈ 6.63 × 10−34 J s are called
2π
“Planck’s constant.”) Therefore, each normal mode in quantum mechanics not only has
a characteristic angular frequency of oscillation, but also a corresponding energy. This
explains why the normal modes of a quantum system are called “energy eigenstates.”
(“Eigen” means “characteristic” in German.) When the system is in one of these energy
eigenstates, the quantum mechanical wavefunction oscillates at the same frequency
at all points, just as the two pendula oscillate at the same frequency when the system
is in the pendulum mode (or the breathing mode). Just as in the coupled pendulum
system, it is also possible to add together or superpose energy eigenstates to create
more complicated states. For example, say that a represents a state with energy Ea ,
and b represents a state with energy Eb . We could create a state with a superposition
of these two:

mix = Aa eiϕa a + Ab eiϕb b
Here, we have explicitly indicated the amplitudes Aa and Ab being used in the
superposition, as well as the phases ϕa and ϕb . The quantum mechanical wavefunction
is inherently complex (!), so we don’t take the real part. This mixed state is exactly
analogous to mixing together a pendulum mode of amplitude Ap and a breathing mode
of amplitude Ab , for the coupled pendulum system.
Just as the energy in the coupled pendulum mixed state “sloshes” back and forth
between the two pendula, the probability distribution for an electron in such a quantum
state is complicated, and the peak of probability “sloshes around” from one place to
another.
However, there is one area in which this otherwise very good analogy doesn’t work.
If you measure the energy of the electron in this mixed quantum state, you always get
Ea or Eb , and never anything in between. This indeed is a “quantum mystery,” one that
you will learn more about elsewhere, and one that has no analogy in the classical world.
However, the probability of measuring energy Ea turns out to be A2a , and the probability
of measuring Eb turns out to be A2b . Therefore, if you could prepare a million identical
electrons, all initially in this same superposed state, and measure their energies, then
the average energy of all these measurements would be
Eaverage = A2a Ea + A2b Eb
The average energy, defined in this way, is called the “expectation value” of the
energy. (This is somewhat of a misnomer, since we don’t actually expect any single
measurement to match the expectation value. Rather, each single measurement will
yield either Ea or Eb .) Comparing this with equation 5.5.1 shows that there is a
close analogy between the classical energy of the coupled pendulum system and the
expectation value of the energy for the quantum mechanical system.
5.6 Hilbert space and bra-ket notation
There is a powerful way of thinking about superposing normal modes to create more
complicated behavior, or the inverse operation of analyzing complicated behavior into
its simpler normal mode components. Again, for now we restrict ourselves to situations
in which both pendula have zero initial velocity. We could then visualize all the possible
initial conditions as points or vectors on a plane, with the initial position of mass 1,
x10 , as the horizontal axis, and the initial position of mass 2, x20 , on the vertical axis.
Choosing a point in this plane, that is, choosing a set of initial conditions, completely
specifies the behavior of the system for all subsequent times, as shown in the example
of figure 5.6.1.
This plane is the simplest example of a Hilbert space6 ; so far there is not much
remarkable about it, but if we define things carefully we can generalize to much more
complicated Hilbert spaces, which can be used to describe much more complicated
systems, and we will gain enormous power! Mwah hah hah!!
Ahem. For example, if we had three masses in the system, we would need a
three-dimensional Hilbert space to represent all the possible initial states of the system
(again, sticking for now with the requirement that all the initial velocities be zero).
If we had four masses, we would need a four-dimensional Hilbert space. If we had
100 masses, we would need a 100-dimensional Hilbert space. If we want to describe
a continuous system, such as a rope, we could divide it up into an infinite number of
infinitesimally small masses; we would need an infinite-dimensional Hilbert space to
describe such a system. Choosing one point/vector in such a space would correspond
to specifying one coordinate along each of the infinite number of axes which define the
space. This infinite list of numbers is basically the same as a function; to completely
define a function, you must define the value of the function at an infinite number of
points along the x-axis. So, each different point in an infinite-dimensional Hilbert space
signifies a different function.
Figure 5.6.1 If we require the

initial velocities of both pendula to
be zero, then any initial condition
can be depicted as a vector on the
plane shown in the right part of the
figure. The vector can be specified
in column matrix format, with x10
in the top line and x20 in the
bottom line.
6. The general definition of Hilbert space is a vector space with a defined “inner product”. The
inner product is a rule for combining two vectors to create a new quantity; for ordinary vectors in
three-dimensional space, the dot product is the inner product. Formally, a Hilbert space must be
“complete” in a sense that is carefully defined by mathematicians, but this restriction is seldom
important in physics.
But, for now, let’s stick to our two-dimensional Hilbert space that describes our
system of two coupled pendula. Each vector in this space can be represented as a
column matrix, with the top entry showing the component along the x10 axis and the
bottom entry showing the component along the x20 axis, as shown in the figure. This
way of representing vectors may be new to you; it works well when you need to
multiply a vector by a matrix, which we’ll do in chapter 6.
Let’s see how vector multiplication works in this notation. Here’s the way you’re
used to taking dot products:
A · B = Ax Bx + Ay By . (5.6.1)
However, eventually (once we allow the initial velocities to be nonzero) we will need to
deal with vectors that have complex components. For such vectors, the above definition
must be modified. We want the dot product of any vector with itself to equal the square
of the length of the vector. This should be real, but
A · A = Ax A x + Ay A y
isn’t real if Ax or Ay is complex. Therefore, we instead define an extended version of

the dot product:
A · B ≡ A∗x Bx + A∗y By , (5.6.2)
where the ∗ indicates “complex conjugate.” Recall from section 1.8 that, to take the
complex conjugate of any number, you simply replace any occurrence of i by −i. For
example, if C = a + ib = Aeiϕ , then C ∗ = a − ib = Ae−iϕ . Thus, equation (5.6.2) is
the same as equation (5.6.1) if the components are real. The complex conjugate makes
it easy to find the magnitude of a complex number; if C = a + ib = Aeiϕ , then
Cartesian version: CC ∗ = (a + ib) (a − ib) = a2 + b2 = |C |2

Polar version: CC ∗ = Aeiϕ Ae−iϕ = A2
Using the extended definition of the dot product, equation (5.6.2), we have
2 2
A · A = A∗x Ax + A∗y Ay = Ax + Ay ,
which is thus guaranteed to be real.

In the language of Hilbert space, we refer to dot products as “inner products.”
To take the inner product of two vectors using matrix notation, we must first take the
adjoint (also called the “Hermitian transpose” or “Hermitian conjugate”) of the one on
the left. Taking the adjoint means: write the matrix as a row instead of a column, and
take the complex conjugate of the entries. Here’s how this works:

A1
A≡ ⇒ adjoint of A is A† ≡ A∗1 A∗2 ,
A2
where the † superscript means “adjoint.” The next step in taking the inner product of
A and B is to multiply A† with B using the rules of matrix multiplication. We’ll review
the full rules later. For now, we simply multiply the left entry of A† by the top entry of
B and add this to the product of the right entry of A† and the bottom entry of B:

B1 ∗ ∗ B1
B≡ Inner product of A and B = A1 A2 = A∗1 B1 + A∗2 B2 . (5.6.3)
B2 B2
This gives the same result as the extended definition of the dot product, equation (5.6.2).
(Recall that, for now, all the vectors in our Hilbert space are real, but we want to make
fully general definitions.)
Bra-ket notation In 1930, Paul Dirac introduced a wonderful notation that simplfies
the writing of inner products; the advantages are not immediately obvious, but this
notation is almost universally used for quantum mechanics, so you may as well begin
we instead
getting used to it. Instead of writing a vector in Hilbert space as A or A,
write it as |A. The “|” that surrounds the “A” plays the same role as the arrow in
– it simply tells you that the quantity is a vector. (In fact you can see that the
A
“|” sort of looks like an arrow.) A vector written with this new notation is called a
“ket” – part of a little physics joke, as you’ll see. Thus, we write

A1
|A =
A2
The adjoint of |A is written A| – this is called a “bra”:

A| ≡ A∗1 A∗2
Finally, the inner product of the vectors |A and |B is written A | B:

B
|B = 1
B2

B1
A | B = (A∗1 A∗2 ) = A∗1 B1 + A∗2 B2 .
B2
Because this brings together the “bra” A| with the “ket”|B, it forms a “bra-ket,”
or bracket. Get it? Heh, heh.
Now, that you’re done rolling in the aisles, let’s see if you really did get it.

2
Concept test: If the ket |A is given by |A = , what is the corresponding bra?
−7 + 3i
(Answer below.7 )
7. A| = (2 − 7 − 3i).

4 − 2i 2
Self-test: If |B = , and |A = , what is the value of their inner
5 −7 + 3i
product,A | B? (Answer below.8 )
We will use the terms “vector in Hilbert space” (or simply “vector”) and “ket”
interchangeably.9
Aside: Representing kets with column matrices. Most physicists and mathematicians
x
are comfortable writing a vector in ordinary space as r = , where the top line of
y
the column matrix means the position along the x-axis and the bottom line means the
position along the y-axis. Perhaps r indicates the position of a flea on the surface of a
table. If we choose to use a different coordinate system, perhaps one that is rotated by
45◦ clockwise relative to the original one, this does not affect the position of the flea, so
that the meaning of r is unchanged. However, we would have to change the way r is
represented by the column matrix. For example, if in the original coordinate system S we

1 0
have r = , then in the rotated coordinate system S’ we would have r = √ . In
1 2
order to make sense of either of these equations, we have to know what “basis” is being
used, in other words whether the top line of the column matrix means the value of x
(in the original coordinate system) or instead the value of x’ (in the rotated coordinate
system), and similarly for the bottom line. To avoid this possible confusion, we could

1
write r → , where the arrow with the S under it means “in the S coordinate system,
S 1
is represented by.” However, this notation, although explicit, can be cumbersome. So, as

1
long as it is clear which coordinate system is in use, we simply write r = .
1
Exactly the same arguments apply for vectors in Hilbert space. In our case, when we

A1 A1
write |A = , this is shorthand for |A → , where the arrow means “in a
A2 positions A2
of masses
coordinate system where the axes indicate the positions of the masses, is represented
by.” This is the coordinate system shown in the right part of figure 5.6.1. However, we
could describe the same Hilbert space using a coordinate system with rotated axes. The
meaning of |A would be unchanged, but we would have to change the entries in the
column matrix accordingly for this new basis, just as we did for the case of the flea.
continued
8.

4 − 2i
A | B = 2 − 7 − 3i = 2 (4 − 2i) + (−7 − 3i) 5
5
= (8 − 35) + (−4 − 15) i = −27 − 19i.
9. Note that mathematicians typically write the inner product as A, B, instead of A | B, but the
meaning is exactly the same.
In this book, we will consistently use the coordinate system where each line of the
column vector corresponds to the position of one of the masses, so we will simply write

A
|A = 1 with this understanding. As David Griffiths writes,10 “Technically, the ‘equals’
A2

A1
signs here [in equations such as |A = ] mean ‘is represented by’, but I don’t think
A2
any confusion will arise if we adopt the customary informal notation.”
Now, back to the Hilbert space for our coupled pendulum system. Recall that the
horizontal axis represents x10 (the initial position of the left bob), while the vertical
axis represents x20 . Also, recall that, to start with, we are restricting ourselves to the
important special case of zero initial velocities.
Some vectors in this Hilbert space represent the normal modes; two examples are
a
shown in figure 5.6.2. Note that any vector of the form , that is any vector with the
a
same entry in the top and bottom, would represent a “pure” pendulum mode. However,
the one shown in the diagram (the one labeled “pendulum”) is special: it has length 1.
A vector of length 1 is called a “normalized” vector. The components of the vectors
shown are dimensionless. That is, the length of each vector shown here is simply 1,
not (1 m). This is exactly like the unit vectors î and ĵ that you’re used to for regular
x–y space; they are vectors of length 1.
Figure 5.6.2 The eigenvectors for the

symmetric coupled pendulum system.
10. Introduction to Quantum Mechanics, 2nd Ed., Pearson Prentice-Hall, Upper Saddle River,
NJ, 2005, p. 120.
The normalized vectors in the Hilbert space that represent the normal modes are
called “eigenvectors.” As you can show in problem W5.1 (on the website for this text),
the eigenvectors lie along the axes that give the initial values of the pendulum mode
1 1
coordinate sp ≡ √ x1 + x2 and the breathing mode coordinate sb ≡ √ x1 − x2 .
2 2
Thus, we could label these axes sp0 and sb0 , as shown in figure 5.6.2. However, it is
the eigenvectors, rather than the axes, that are more important conceptually.
Note that the two eigenvectors are perpendicular or “orthogonal” to each other.
Although this is obvious to the eye, we can check it by taking their inner product,
which should equal zero for orthogonal vectors:

√
√ √ 1 2 1 1
1 2 1 2 √ = − = 0 (5.6.4)
−1 2 2 2
Let us define symbols for the two eigenvectors:

√

e = 1/√2 = √1 1 and e = √1 1
( (
p b ,
1/ 2 2 1 2 − 1
where the “e” stands

' for ( eigenvector. In terms of these symbols, we could write
equation (5.6.4) as ep e = 0.
b
' ( ' ( ' (

Your turn: Verify that ep eb = 0, ep ep = 1, and eb eb = 1.
We could describe any vector in the plane by suitable combinations of the two
eigenvectors. For example, the black vector shown in figure 5.6.3 has x10 = (1 cm)
and x20 = (0.5 cm). Therefore, we can write it in terms of its components (shown by

1 cm
the black dashed lines) along the x10 and x20 axes: . However, we could also
0.5 cm
write it in terms of its components (as shown by the gray dashed lines) along the axes
Figure 5.6.3 The black vector can be resolved into

components either along the x10 and x20 axes (as
shown by the black dashed lines) or along the
pendulum mode and breathing mode axes (as
shown by the gray dashed lines).
defined by the two eigenvectors. In other words, we could write it as

√
√
1 cm (1.5 cm) 1/ 2 (0.5 cm) 1/ 2
= √ √ + √ √
0.5 cm 2 1/ 2 2 −1 / 2
(1.5 cm) ( (0.5 cm) (
= √ ep + √ eb
2 2
Your turn: Verify that the sum of vectors on the right actually adds up to the vector on
the left.
This shows graphically that we can use the normal modes representation as a different
and equally good way to describe the system. Either we can describe the initial condition
shown in figure 5.6.3 as “the left pendulum is displaced 1 cm to the right and the right
pendulum is displaced 0.5 cm to the right,” or we could instead say “the pendulum mode
(1.5 cm) (0.5 cm)
has an amplitude of √ and the breathing mode has an amplitude of √ .”
2 2
Note that either description requires two pieces of information (plus the information
that the initial velocities are zero). The normal mode description can be seen as a way
of describing the same plane of points, but with axes that are rotated by 45◦ .
Since we can describe any vector in the Hilbert space using suitable combinations
of the eigenvectors, the set of two eigenvectors is called a “complete basis.” We might
also say that this set of vectors “spans” the space. Because the eigenvectors are normal
and orthogonal, they form a “complete orthonormal basis.” Often, the word “complete”
is dropped, and we simply say that the set of two eigenvectors form an “orthonormal
basis.”
Terminology review:
Hilbert space11 : A vector space with a defined inner product. As used in this book: a
space in which each point represents a particular configuration of the system.
In most applications of Hilbert space, the space has infinite dimensions, so each
point represents a function. In fact, it might be helpful for you to start thinking
of the vectors in our plane as functions that are evaluated only at two points
(at mass 1 and at mass 2). However, if that idea confuses you now, don’t start
thinking that way yet.

A
Column matrix (also called column vector): a matrix of the form 1 that represents a
A2
vector: the top entry represents the component along the horizontal axis, and
the bottom entry represents the component along the vertical axis.
continued
11. There are a number of amusing anecdotes about David Hilbert. When Hilbert spaces were just
starting to be used in a variety of mathematical fields, as well as in physics, he attended a
conference with the mathematician Richard Courant. Listening to a series of presentations about
different uses of Hilbert space, Hilbert leaned over to Courant and asked, “Richard, exactly
what is a Hilbert space?”

A
Adjoint (or Hermitian transpose or Hermitian conjugate): The adjoint of 1 is A∗1 A∗2 .
A

2
A1
Ket: Another name for a vector in Hilbert space. For example, |A = .
A2
∗ ∗
Bra: The adjoint of a ket. For example, A| = A1 A2

' ( B1
Inner product: Generalized version of the dot product: A B = A∗1 A∗2 = A∗1 B1 +
B2
A∗2 B2
Normalized vector: A vector in the Hilbert space that has length 1; the length is
dimensionless. (The unit vectors î and ĵ are examples of normalized vectors
in x–y space.) The inner product of a normalized vector with itself equals 1.
Orthogonal vectors: Vectors that are perpendicular to each other. They have an inner
product of zero.
Complete basis: A set of vectors, linear combinations of which can be used to create
any vector in the Hilbert space. For a two-dimensional Hilbert space, any two
vectors that are nonparallel would form a complete basis.
Complete orthonormal basis (or simply “orthonormal basis”): A set of orthogonal
normalized vectors that form a complete basis. (The unit vectors î and ĵ form a
complete orthonormal basis for x–y space.)
(
The component of an arbitrary vector x0 along the sp0 axis gives the amplitude of
(
the pendulum mode
( needed to create the initial conditions represented by x0 , and the

component of x0 along the sb0 axis gives the amplitude of the breathing mode that is
needed. To find these components, we simply take inner products (just as you can use
dot products to find the components of a vector in x–y space: Dx = î · D):

' ( ( 1 1
amplitude of pendulum mode = Ap = ep x0 where ep ≡ √

2
1
' ( ( 1 1 (5.6.5)
amplitude of breathing mode = Ab = eb x0 where eb ≡ √
( ' ( ( ' ( ( 2 − 1
⇒ x0 = ep x0 ep + eb x0 eb
(zero initial velocities)
Again, this is exactly analogous to the process of expressing ordinary vectors in

components:
(
î ⇔ ep
(
ĵ ⇔ eb
' (
Dx = î · D ⇔ amplitude of pendulum mode = Ap = ep x0
' (
Dy = ĵ · D ⇔ amplitude of breathing mode = Ab = eb x0
( ' ( ( ' ( (
D = î · D î + ĵ · D ĵ ⇔ x0 = ep x0 ep + eb x0 eb
How do we use Hilbert space to help with thinking about normal mode analysis?
Let’s use the symbol |x(t) to represent a vector with the positions of both pendula as
a function of time, that is,

x1 (t)
|x (t) ≡
x2 (t)
(
Consider the special case of a pure breathing mode. Then, x0 must have the form

( ( Ab 1
x = A e = √
0 b b , where Ab is the amplitude of the breathing mode. (Recall
2 −1

( A 1 √ √
that x0 = √b means x10 = Ap / 2 and x20 = −Ap / 2.) For this simple case
2 −1
of a pure breathing mode, we know that both pendulum bobs just oscillate

with angular
Ab 1
frequency ωb , so that the time dependence is |x (t) = √ cos ωb t .
2 −1
Similarly, if we take the special case of a pure pendulum mode, then

A Ap
x = A e = √p 1 1
( ( (
0 p p and x (t) = √ cos ωp t .
2 1 2 1
Example: Let’s rework the example from section 5.5 using these new ideas: A set of
coupled pendula has m = 0.10 kg, ℓ = 0.15 m, and k = 5.0 N/m. At t = 0, the left mass
is held at x1 = 1.0 cm and the right mass is held at x2 = 3.0 cm. The masses are then
released. What is the position of mass 1 at t = 5.0 s?
Solution (using Hilbert space to represent things graphically): The initial position
vector in Hilbert space is shown in figure 5.6.4. The components along the eigenvectors
represent the initial amplitude of each normal mode.
(
Figure 5.6.4 The (position vector x0 is resolved into components along the axes
( initial

defined by ep and eb .
continued
Concept test (answer below12 ): We’re about to find the components quantitatively,
but first estimate what they are from the figure (which is drawn to scale).
To find these components quantitatively, we take the inner product of each
(
eigenvector with the vector representing the initial condition of the system, x0 =

1.0 cm
:
3.0 cm

' ( 1 1.0 cm
component of pendulum mode = ep x0 = √ 1 1

2 3.0 cm
(1 cm) √
√ (1.0 + 3.0) = 2 2 cm
=
2

' ( 1 1.0 cm
component of breathing mode = eb x0 = √ 1 −1

2 3.0 cm
(1 cm) √
= √ (1.0 − 3.0) = − 2 cm
2
Using equation (5.6.6), we then have
+
√ ( ( √ 1 1
|x (t) = 2 cm 2 cos ωp t ep − cos ωb t eb = 2 cm 2 cos ωp t √
2 1
,
1 1
− cos ωb t √
2 − 1

2 cos ωp t − cos ωb t
= cm
2 cos ωp t + cos ωb t
Concept test: From the above, read off the expression for x1 (t) (the position of the left
mass as a function of time), and verify that it matches what we got when we worked this
same example in section 5.5.
For a superposition of breathing and pendulum modes, we then have

( ( ( Ab 1 Ap 1
x = A e + A e = √ +√
0 b b p p
2 −1 2 1

Ab 1 Ap 1
and |x (t) = √ cos ωb t + √ cos ωp t .
2 −1 2 1
12. The component of pendulum mode is about 2.8 cm, and the component of breathing mode is
about −1.4 cm.
' ( ' (
Since Ab = eb x0 and Ap = ep x0 , we can rewrite this as
' ( ( ' ( (
|x (t) = ep x0 cos ωp t ep + eb x0 cos ωb t eb (5.6.6)
(zero initial velocities)

This way of writing |x (t) is called the “normal mode expansion.”
This way of solving the problem may seem harder than the way we used in
section 5.5. However, the ideas you are using, of taking inner products in Hilbert space,
are much more powerful, and can be readily adapted to more complicated situations.
Once you get used to these ideas, you will find this approach quite intuitive. (OK, I
admit that that’s pretty much like saying, “Once you get used to eating ketchup on your
cereal, you’ll be used to eating ketchup on your cereal.” But Hilbert space really is a
superior and clearer way of thinking about the world.)
5.7 The analogy between coupled oscillators

and molecular energy levels
Our discussion of coupled oscillators is completely general, and applies to any two
identical coupled oscillators. Here, again, are the expressions we found for the angular
frequencies of the normal modes:

g g 2k
ωp = and ωb = + .
ℓ ℓ m
A larger k corresponds to a stronger spring connecting the two pendula, that is, to
stronger coupling. From the above equations, we can draw a very important conclusion,
which turns out to be true for all systems: the difference in frequency between the
two normal modes (called the “frequency splitting”) gets bigger as the coupling gets
stronger.
Stronger coupling produces larger splitting of normal mode frequencies.
As discussed in section 5.5, for quantum mechanical systems, the energy of

a normal mode (i.e., the energy of an “energy eigenstate”) is proportional to the
frequency. Therefore, we expect that when two quantum oscillators are coupled
together, the resulting system will have two energy eigenstates, with a splitting in
energy which increases with the strength of the coupling. This is precisely what is
observed! For example, let’s make a hydrogen molecule (H2 ) by bringing two hydrogen
atoms gradually closer together. Each hydrogen atom consists of a proton with an
electron near it. The electron is bound to the nucleus by electrostatic forces, and
cannot go far from the nucleus. This is analogous to the situation for one of our
pendula; the pendulum bob cannot go far from the equilibrium position, because the
“pendulum force” tends to push it back. The analogy is not perfect, because the potential
Figure 5.7.1 The analogy between molecular energy levels and coupled oscillators. Like the
protons discussed in section 4.7, the electrons have spin angular momentum. Like the protons,
each electron can be in a “spin up” or “spin down” state. Therefore, the electrons are shown as
arrows, with a spin up electron shown as an up arrow and a spin down electron shown as a
down arrow.
energy function experienced by the electron is not the same as that experienced by the
pendulum, but the qualitative conclusions are the same.
The energy eigenstates of the electrons in the isolated atoms (when they are far
apart) are called “atomic orbitals,” and the associated energies are called “atomic
energy levels.” If we bring the two atoms gradually closer together, they interact
more strongly. This is analogous to starting with two pendula which are completely
uncoupled, then gradually increasing the strength of the coupling by increasing k. In
the coupled pendula case, the system can assume one of two normal modes. In the
hydrogen case, the electrons in the molecule can assume one of two molecular energy
eigenstates; these are often called “molecular orbitals,” and each has an associated
“molecular energy level.” Figure 5.7.1 shows the way this process is usually shown,
with energy on the vertical axis. For stable molecules, such as H2 , one of the two
molecular energy levels lies below the original atomic energy levels. If both electrons
move into this one level, then the total energy of the system is lowered relative to the
separate atoms, so the molecule is stable. Therefore, the associated eigenstate is called
a “bonding orbital.”13
13. Note that we are using a different analogy between quantum and classical systems here than
at the end of section 5.5. There, we drew an analogy between a single atom and the system of
two coupled pendula, with the different atomic orbitals (1s, 2s, etc.) analogous to the different
normal modes of the coupled pendula. Here, instead, we think of the two pendula as analogous
to two H-atoms, but we only consider a single atomic orbital for each atom. Both analogies are
appropriate, but they are distinct.
5.8 Nonzero initial velocities
So far, we have concentrated on situations for which all parts of the system (i.e., both
pendula) are at rest at t = 0. But what if they aren’t? How do we determine the future
behavior of the system, and the weighting of the different normal modes, if the initial
velocities are nonzero?
It turns out that this question is not very important for quantum mechanics. The
time derivative of the real part of the wavefunction evaluated at a particular point
is analogous to the velocity of one of our pendula, and usually is chosen so that
this time derivative is zero at t = 0 for all points. (It turns out that this corresponds to
choosing to be real everywhere at t = 0.)
However, for mechanical systems, this is a reasonable question to ask, and it’s not
hard to answer. Since the two differential equations that describe the normal modes can
be derived from the original coupled differential equations that describe the motion
of the pendulum bobs (and vice-versa), any state of the system can be described as a
superposition of the normal modes. Therefore, the most general description is
(
|x (t) = Re Ap ei(ωp t +ϕp ) ep + Ab ei(ωb t +ϕb ) eb ,
(
where Ap is the amplitude of the pendulum mode, Ab is the amplitude of the breathing
mode, ϕp is the phase of the pendulum mode, and ϕb is the phase of the breathing mode.
To write things more compactly, we can set Cp ≡ Ap eiϕp and Cb ≡ Ab eiϕb , so that
( (
|x (t) = Re Cp eiωp t ep + Cb eiωb t eb , (5.8.1)
where Cp and Cb are the “complex amplitudes.” (The restriction of zero initial velocities
which we have used up to this point is equivalent to requiring Cp and Cb to be real, as
we will soon see.) Again, the above is called

the“normal mode expansion.” Remember
x (t)
that this is a vector equation: |x (t) ≡ 1 . Our goal is to find Cp and Cb , given
x2 (t)
the initial conditions.
The velocity is obtained by taking the time derivative:

ẋ1 (t) & (
|ẋ (t) = = Re iωp Cp eiωp t ep + iωb Cb eiωb t eb .

ẋ2 (t)
Plugging t = 0 into these expressions gives

x ≡ x10 = x1 (t = 0) = Re C e + C e
( & (
0 x20 x2 (t = 0) p p b b

ẋ (t = 0)
( & (
and ẋ0 ≡ 1 = Re iωp Cp ep + iωb Cb eb

ẋ2 (t = 0)
We write the complex amplitudes in Cartesian form:
Cp = Re Cp + i Im Cp and Cb = Re Cb + i Im Cb ,
so that
( ( ( ( ( (
x = Re C e + Re C e and ẋ = −ω Im C e − ω Im C e .
0 p p b b 0 p p p b b b
( (
Note that Re Cp ep is shorthand for Re Cp ep . Following what we did in section
5.6, let’s see what happens if we take inner products with these:
' ( ' 1 ( (2 ' ( ' (
ep x0 = ep Re Cp ep + Re Cb eb = ep Re Cp ep + ep Re Cb eb
' ( ' (
= Re Cp ep ep +Re Cb ep eb
! !
1 0
' (
⇒ ep x0 = Re Cp (5.8.2a)
and
' (
ep ẋ0 = −ωp ImCp . (5.8.2b)
Similarly,
' (
eb x0 = Re Cb (5.8.2c)
and
' (
eb ẋ0 = −ωb Im Cb . (5.8.2d)
(As promised, you can see that if the initial velocities are zero, then Cp and Cb are
(
real.) So, we’re done; given the initial positions x0 (which, recall
( has x10 in the top
row and x20 in the bottom row), and given the initial velocities ẋ0 , we can determine
the complex coefficients Cp and Cb in the normal mode expansion (5.8.1).
Self-test (answer below14 ): At t = 0, mass 1 (the pendulum bob on the left) has position
1.3 cm and velocity −1.9 cm/s, while mass 2 (the bob on the right) has position −0.3 cm
and velocity −0.2 cm/s. For this system, ωp = 5.0 rad/s and ωb = 7.0 rad/s. Find the
coefficients Cp and Cb that appear in the normal mode expansion of |x (t).
5.9 Damped, driven coupled oscillators
Let’s add damping forces to both pendula, and a drive force which is exerted only on
the left pendulum (this reflects a common real-world situation in which a driving force
is applied only at one point of a complicated system):
Fdamp, 1 = −bẋ1 , Fdamp, 2 = −bẋ2 , Fdrive, 1 = F0 cos ωd t , Fdrive, 2 = 0
14. Answer to self-test: Cp = (0.71 + i 0.30) cm , Cb = (1.1 + i 0.17) cm.

Your turn: Recall that γ ≡ b/m. Show that the differential equations that describe
this system are

k g k F
ẍ1 + γ ẋ1 + + x1 − x2 = 0 cos ωd t (5.9.1)
m ℓ m m
and

k g k
ẍ2 + γ ẋ2 + + x − x =0 (5.9.2)
m ℓ 2 m 1
Inspired by what we did for

√ the undamped, undriven case, we try adding these equations
(after multiplying by 1/ 2):

(5.9.1) (5.9.2) ẍ1 + ẍ2 ẋ1 + ẋ2 k g x1 + x2 k x1 + x2
√ − √ ⇒ √ +γ √ + + √ − √
2 2 2 2 m ℓ 2 m 2
F
= √ 0 cos ωd t
2m
√
g F0 2 1
⇒ s̈p + γ ṡp + sp = cos ωd t , where sp ≡ √ x1 + x2
ℓ m 2
This is just the equation for

an ordinary damped driven harmonic oscillator, with
g
resonant frequency ω0 = = ωp . The quantity that oscillates is sp , the normal
ℓ
mode coordinate for √the pendulum mode. Note that the effective amplitude of the
driving force is F0 2, reflecting the fact that the energy from the driving force is
responsible for the motion of two masses instead of just one.
If instead we√take the difference between equation (5.9.1) and (5.9.2) (after
multiplying by 1/ 2), we find

(5.9.1) (5.9.2) ẍ1 − ẍ2 ẋ1 − ẋ2 k g x1 − x2 k x1 − x2
√ + √ ⇒ √ +γ √ + + √ − √
2 2 2 2 m ℓ 2 m 2
F
= √ 0 cos ωd t
2m

√
g 2k F0 2
⇒ s̈b + γ ṡb + + s = cos ωd t ,
ℓ m b m
1
where sb ≡ √ x1 − x2 s
2
Again, this is just the equationfor an ordinary damped driven harmonic oscillator,
g 2k
with resonant frequency ω0 = + = ωb . The quantity that oscillates is sb , the
ℓ m
normal mode coordinate for the breathing mode.
The conclusion is that this system behaves like two completely independent
damped, driven oscillators, one corresponding to the pendulum mode and one to
Figure 5.9.1 For a damped driven coupled

oscillator system, each part of the system
shows a resonant response at the frequencies
corresponding to each of the normal modes.
This shows the amplitude of response of
mass 2 of a two-pendulum system, with the
drive force applied to mass 1. The curve for
the amplitude of response for mass 1 is
similar.
1
the breathing mode! Since x1 = √ sp + sb , mass 1 shows a large amplitude of
2
oscillation near both ωp and ωb . The steady-state amplitude of oscillation for mass 2
is shown in figure 5.9.1.
This behavior also occurs in more complicated coupled oscillator systems – one
observes a resonant response at each normal mode frequency. This explains, for
example, why the absorption spectrum for a chemical shows an absorption peak at
each normal mode frequency.
Self-test (answer below)15 : In figure 5.9.1, the peak at ωb is higher than the one at ωp .
Explain why. Hint: Recall that Q = ω0 /γ .
terms:
Beats (5.1)
Coupled differential equations (5.2)
Normal mode (5.5)
Breathing mode (5.3)
Pendulum mode (5.3)
Normal mode analysis (5.5)
Hilbert space, as applied to a system of two identical coupled oscillators (5.6)
Column matrix (5.6)
Adjoint (5.6)
ket (5.6)
Inner product (5.6)
bra-ket notation (5.6)
ω0
15. For light damping, the peak amplitude is about Q times the drive amplitude. Since Q ≡ , and
γ
γ is the same for both modes, the Q for the breathing mode (which has a higher characteristic
frequency) is higher, so the peak is higher.
Orthogonal vectors (5.6)

Normalized vector (5.6)
Orthonormal basis (5.6)
Eigenvector (5.6)

Two sinusoids of equal amplitude but different frequency are added together (5.1)
A system is excited into a pure normal mode (5.5)
A system is excited into a superposition of normal modes (5.4)
Two hydrogen atoms are brought close to each other (5.7)
A set of two identical coupled oscillators, with damping, is driven by a periodic force
applied to one of them. (5.9)

Beats & Superpositions of normal modes (5.1, 5.4)
Resolving an ordinary vector into its x- and y-components & Normal mode analysis
for a symmetric coupled oscillator system with zero initial velocity (5.6)
Vectors & Column matrices (5.6)
Kets & Vectors (5.6)
Bras & Row matrices (5.6)
Coupled oscillators & Molecular energy levels (5.7)
Superpositions of normal modes & Superpositions of quantum energy eigenstates (5.7)
Normal modes & Eigenvectors (5.6)
Normal mode coordinates (sp and sb ) & the coordinates of the two oscillators (5.3)

The effect of the strength of the coupling on the separation of normal mode
frequencies (5.7)

Go back and forth between beat frequency and the frequencies of the underlying
sinusoids (5.1)
Given the properties of a system of two identical coupled oscillators, describe the
normal modes and find their frequencies. (5.3)
Compute inner products (5.6)
Given the initial positions for a system of two identical coupled oscillators, and given zero
initial velocities, find the amplitudes of the normal modes. (5.6)
Similarly, for nonzero initial velocities, find the amplitudes of the normal modes. (5.8)
For either of the above, write down the positions of each oscillator as a function of
time. (5.6)
For a damped, driven pair of identical coupled oscillators, calculate the resulting steady
state motion (5.9)
Problems
5.1 Musical scales. The frequencies of notes in standard musical notation are
defined in terms of ratios. For example, an octave is defined as a factor of
two in frequency. The standard “concert A” is 440 Hz, so that one octave
below concert A is 220 Hz. Each octave is divided into twelve half steps, with
the same frequency ratio between any two notes separated by a half step, as
shown in figure 5.P.1 on a piano keyboard. Each of the black keys on the
piano has two names. For example, the black key just above concert A can be
called A# (pronounced “A sharp”), meaning that it is a half-step above A, or it
can be called B♭ (pronounced “B flat”), meaning that it is a half step below B.
(a) Show that the frequency ratio between half-steps is 1.05946.16
(b) The lowest note on a standard piano is the A that is four octaves
below concert A. If this note is played simultaneously with the A#
just above it, and both notes are played at the same volume, what
beat frequency is heard?
Figure 5.P.1 Musical notes on a piano keyboard.
16. This is called the “well-tempered” scale. It gives the exact desired frequency ratio for notes
separated by an octave (i.e., a ratio of 2), but only gives approximations for other harmonic
combinations. For example, the two notes in a “major fifth” chord should have a frequency ratio
Figure 5.P.2 A pair of masses connected by springs to immovable walls.
5.2 Two tuning forks of frequencies 440 and 446 Hz are struck at the same
time with the same intensity. The resulting beat pattern is recorded with a
microphone (placed an equal distance from both forks) and amplified. At the
output of the amplifier, the maximum amplitude of the rapid oscillations is
2 Volts (abbreviated 2 V).
(a) What is the time Tbeat between successive amplitude maxima?
(b) What is the voltage of this wave as a function of time?
5.3 Two identical pendula are coupled by a spring, and are oscillating. One has
"
# "
#
3 9
a position x(t) = (2 cm) cos rad/s t cos rad/s t . What are the
20 2
angular frequencies of the normal modes, ωp and ωb ?
5.4 Figure 5.P.2 shows a pair of masses, connected to immovable walls by
springs. Gravity is negligible. Describe the normal modes of this system,
and find their angular frequencies.
5.5 Two pendula, each with mass 0.30 kg and length 0.50 m, are coupled by
a spring. One of the masses is clamped while the other is pulled aside and
then released, resulting in an oscillation with ω = 5.00 rad/s. Find ωp , ωb ,
and Tbeat .
5.6 Careful observations of a pair of coupled pendula produce the plot shown
in figure 5.P.3 of the x-position of the left pendulum bob. (a) Given that the
time between maxima is 1.0 s (as shown), what is the approximate value of
the pendulum length ℓ? (Assume g = 9.80 m/s2 .) (b) Assuming your value
of ℓ from part a is exact, what is the value of k /m?
5.7 Quantum beats. As discussed in section 5.5, one can put a quantum system
into a superposition state, made up from two fundamental states called
“energy eigenstates.” This is analogous to putting a coupled pendulum
system into a superposition of breathing and pendulum modes. In either the
quantum or the classical system, one can then observe beating phenomena,
because the frequencies of the normal modes (for the classical system) or
energy eigenstates (for the quantum system) are not quite the same. Just as
of exactly 1.5; we can get close to this using the well-tempered scale by using two notes separated
by seven half steps: 1.059467 = 1.49831. However, for some musicians and unusually discerning
listeners, even this small departure from the ideal ratio creates a jarring effect and changes the
emotional quality of the music. Therefore, instruments which do not have fixed intervals, such as
violins, are sometimes played using scales with slightly different ratios between the half-steps,
so that chords such as the major fifth sound more perfect. In such a scale, A# and B♭ are slightly
different.
Figure 5.P.3 Left: A pair of coupled pendula. Right: The position of the left mass as a function
of time.
Figure 5.P.4 a: A signal proportional to the probability of ionization plotted as a function of

delay time between arrival of two different photons. The oscillations shown are due to
quantum beats. b: Schematic representation of the effect of the first photon, which excites an
electron from the ground state into a superposition of two excited states. Part (a) reprinted
from Optics Communications, 31, G. Leuchs et al., Quantum Beats Observed in
Photoionization, 313–316 (1979), © 1979, with Permission from Elsevier.
a guitarist can use beats to tune a guitar, an experimental physicist can use
quantum beats to measure the frequency difference between two quantum
states. Then, using E = h̄ω, she can deduce the energy difference between
the states.
Figure 5.P.4a shows data from such an experiment. First, the outermost
electron in an atom is excited from its ground state by the absorption of
a photon, that is, a particle of light, as shown in figure 5.P.4b. The energy
difference between the two excited states of interest (labeled A and B) is so
small that the photon excites the electron into a superposition of both states.
This is analogous to simultaneously exciting the breathing and pendulum
modes for the coupled pendula. In that case, we observed that the system
slowly evolved, first with almost all the swinging in the left pendulum bob
with the right bob motionless, then with almost all the swinging in the right
bob, and back again. Similarly for the electron, once it has been excited into
the superposition of states A and B, it “slowly” evolves from one shape of the
“electron cloud” (call it shape 1) to another (call it shape 2) and then back.
In this context, “slowly” means on the order of tens of ns (nanoseconds),
which is indeed slow on the timescale of many atomic events.
It turns out that, when the electron cloud is in shape 1, it is easier to
remove the electron entirely from the atom than it is when the electron cloud
is in shape 2. In other words, the probability for ionization of the atom (by
absorption of a second photon) is higher for shape 1 than shape 2. The authors
of this paper used a pair of photon pulses to exploit this fact. A photon from
the first pulse excites the electron into the superposition of states A and B,
and a photon from the second pulse might be able to ionize the atom, if the
photon arrives when the electron cloud is in shape 1. Figure 5.P.4a shows
a signal called “β4 ” measured by the experimenters, which is proportional
to the probability of ionization. The horizontal axis shows the time delay
between the two photon pulses.
Using this data, deduce the energy difference between states A and B;
express your answer in electron Volts (abbreviated eV), where 1 eV = 1.602
× 10−19 J. (For comparison, the energy difference between the ground state
and state B in this atom is 3.68 eV. You should find that the energy difference
between A and B is much smaller than this.) Note that the actual energy level
diagram is more complex than the simplified version shown here, which
explains why the graph above is only roughly sinusoidal.
5.8 State what is wrong with the following statement: “There is no difference
between a normal mode and an eigenvector – the two terms are inter-
changeable.” (As part of your response, provide a corrected version of the
statement.)
5.9 State what is wrong with the following paragraph: “Consider an undamped
mass on a spring. When the mass is given an initial ‘kick’ (by imparting
√
some initial velocity), thereafter it always oscillates at ω0 = k /m, no matter
whether the initial kick is to the left or to the right, and no matter how hard the
kick is. Now, consider an undamped symmetric coupled pendulum system.
In the same way, if the left mass is given an initial kick (while the right
mass is left initially motionless, but released immediately after the kick), the
system always oscillates at one of the normal mode frequencies, either ωp
or ωb , depending on whether the initial kick is to the right or to the left.”
(As part of your response, provide a corrected version of the statement.)
5.10 Two coupled identical pendula of mass 3 kg are oscillating, having started at
rest with positions x1 (0) = 0.22 m and x2 (0) = −0.15 m. Mass 1 has Tfast =
0.5 s (where Tfast is the period of the rapid oscillations) and Tbeat = 10 s.
(a) Find the amplitudes of the breathing and pendulum modes.
(b) Find the position of mass 1 as a function of time.

1
5.11 Two coupled pendula are oscillating with |x(t) = √ m cos [(5 rad/s) t]

7 2
1 1 1
+ √ m cos [(6.1 rad/s) t] . They started from rest. What
1 3 2 −1
were their initial positions?

1.2 − 3i 6.8
5.12 Practice with inner products Let |A = and |B = .
−2 7.1
(a) Evaluate A | B and B | A.
(b) Show that, for any two vectors |C and |D (which may have
complex components),C | D = D | C ∗ . In other words, show
that, to reverse the order of an inner product, you need only take
the complex conjugate of the value. (The vectors are assumed to be
two-dimensional, such as |A and |B are.)
(c) Find a vector that is orthogonal to |B, and verify the orthogonality
using an inner product.
(d) Find a vector that points in the same direction as |B but has length 1,
and verify its length using an inner product.
5.13 Symmetric coupled pendula applet. Please open a browser and go to the
listing for this problem on this book’s web page. Click on the link labeled
“Hilbert Space for Coupled Pendula,” and wait for the applet to load. (It may
take a couple of minutes.)
Exercises (please work through all these; written responses are needed only
as indicated by boldface):
(a) Try to set up the pendula in a pure pendulum mode (by clicking on
the Hilbert Space plot in the applet above) and then click on the
‘Go!’ button. Now try a few different amplitudes.
(b) Now, set up the pendula in a pure breathing mode. Again, try a few
different amplitudes.
(c) For positive breathing amplitude (downward and to the right), is
the initial position of the left pendulum bob positive or negative?
How about the right one?
(d) Now set up a beat pattern so that energy is transferred back and forth
between the two pendulum bobs periodically. First, make a perfect
beat pattern where each bob periodically comes to a complete stop,
and then set up some more complicated behavior.
(e) For the beating state that you just created, look carefully at the
graphs at the bottom of the applet. Are these graphs showing the
correct behavior of the pendulum and breathing modes? Is the
behavior of the system simply an addition of these two modes?
(f) Try pressing the button labeled “Drop Lines,” and describe what
this is showing.
(g) Describe how the colors used in the animation help to make
connections between the three different pictures of theapplet.
g
(h) The angular frequency of the pendulum mode is ωp = . The
ℓ

g 2k
angular frequency of the breathing mode is ωb = + .
ℓ m
Recall what we learned about beats: A cos ω1 t + A cos ω2 t =
ω + ω2 ω − ω2
2A cos ωe t cos ωav t where ωav = 1 and ωe = 1 .
2 2
Given a linear combination of these two normal modes, what

is the period and frequency of the envelope? Qualitatively, how
does changing the spring constant (k) affect these results? Test
this hypothesis by changing k in the animation.
(i) The normal modes are mutually orthogonal – they do not influence
each other. Why is this an illuminating way to look at a system?
(j) In order to predict the behavior of a system of coupled oscillators
as a function of time, why is it more useful to know the
amplitudes of the normal modes than the initial positions of
the pendulum bobs?
(k) What vectors in the Hilbert space are associated with unique
frequencies?
(l) Using the language of taking projections of vectors onto differ-
ent axes, describe the process of normal mode analysis.
5.14 For a pair of symmetric pendula (or any other symmetric coupled oscillator
system) that is restricted to zero initial velocities, verify that the following
peculiar-looking equation is correct:
|ep ep | + |eb eb | = 1.
Hint: Consider what happens when you apply each side of the equation to
an arbitrary Hilbert space vector |A, that is, when you write ( |ep ep | +
|eb eb | ) |A = 1|A . If you can explain why this equation must be true, then
(because |A is arbitrary), the original version of the equation must also be
correct.
5.15 Each pendulum bob in a pair of symmetric coupled pendula has mass 0.25 kg,
and pendulum length 2.5 m. The bobs are connected by a spring of constant
0.25 N/m. Initially, both are held immobile, with the left bob 0.25 m to the
left of its equilibrium position and the right bob at its equilibrium position.
At time t = 0.25 s, they are released. Find the position of each bob as a
function of t.
5.16 A set of identical coupled pendula has ωp = 5.0 rad/s and ωb = 7.0 rad/s.
The coefficients in the normal mode expansion are Cp = (0.10 m)ei(0.35) and
Cb = (0.15 m)ei(0.15) . What is the initial position and velocity of mass 2?
5.17 For the symmetric coupled pendulum system, consider an arbitrary superpo-
sition of pendulum and breathing modes, one for which the initial velocities
are not necessarily zero. Show that, even in this superposition, the normal
mode coordinates sp and sb oscillate in simple harmonic motion.
5.18 A symmetric coupled pendulum system has m = 1.5 kg, ℓ = 0.8 m, and
k = 30 N/m. At t = 0, the left bob is at a position of 5 cm, with a speed of
−2 cm/s, while the right bob is at -2 cm with a speed of 3 cm/s.
(a) Show that ωp = 3.50 rad/s and that ωb = 7.23 rad/s.
(b) Find |x (t). Hint: after finding the real and imaginary parts of the
complex amplitudes, express the complex amplitudes in polar form,
then do your taking of the real part.
5.19 A pair of symmetric coupled pendula has the following initial conditions:
the left bob is at position −3.1 cm with velocity −10 cm/s, and the right bob
is at position −0.2 cm with velocity 5 cm/s. Each bob has m = 1.2 kg and
string length ℓ = 0.45 m; the spring constant is k = 2.5 N/m. At t = 5.3 s,
what is the position of each bob? Show all your work clearly.
5.20 State what is wrong with the following paragraph: “Consider a symmetric
coupled pendulum system, with damping, and with a periodic drive force
applied to the left pendulum bob, as discussed in section 5.9. When a drive
force of angular frequency ωd is applied, the steady-state motion x1 (t) of
the left bob is always the sum of an oscillation at ωb and an oscillation
at ωp . As ωd is varied, the relative amplitude of the responses at ωb and
ωp changes.” (As part of your response, provide a corrected version of the
paragraph.)
5.21 Two lightly damped identical pendula are coupled together with a spring with
spring constant 15.0 N/m. The masses of the pendula are 3.00 kg, their lengths
are 1.20 m, and the damping coefficient is b = 0.500 kg/s. The support point
SL for the left pendulum (where the string is attached to the ceiling) can
be moved left and right in a sinusoidal pattern, while the support point for
the right pendulum is held motionless. (a) At what angular frequency should
the support point SL be moved to produce the largest steady state response
in the coupled pendulum system? (b) Make a semi-quantitative argument
showing that, if the system is driven at this angular frequency (by moving
SL), the response of the pendulum mode is negligible, bearing in mind that the
system is lightly damped. Hint: Calculate the FWHM of the power resonance
curve. If the drive frequency is further away from the resonance frequency
than four times the FWHM, the response will be very small indeed. (c) For
the angular frequency you calculated in part (a), what is the ratio of the
amplitude of the right pendulum to the amplitude of motion of SL?
5.22 (Please do problems 5.4 and 5.20 before beginning this exercise.) You are part
of a team designing an experiment for the International Space Station. Part
of the experiment involves a large box, labeled B in figure 5.P.5a. During
the experiment, box B will float inside the space station, in an air-filled
chamber. It will be tethered by a spring to a vibrating panel, labeled P, which
is attached to the side of the space station. Panel P will oscillate left and right
at a very low frequency (0.010 Hz); these vibrations should be transmitted
via the tether to box B. However, other team members have told you that
unwanted effects will cause panel P also to oscillate with the same amplitude
at frequencies of 75.0 Hz. In other words, the motion of panel P will be

xP = Ad cos ωw t + cos ωu t ,
where ωw = 2π (0.010 Hz) is the angular frequency of the wanted oscil-

lations, ωu = 2π (75.0 Hz) is the angular frequency of the unwanted
oscillations, and A = 0.05 m is the amplitude of oscillation.
Your job is to finalize the design of the tether, which must be 1.23 m long,
so as to minimize the transmission of the unwanted oscillations. You are given
Figure 5.P.5 Three configurations for

tethering box B to panel P, which is part of
the International Space Station.
a spring which is 1.23 m long, and has a spring constant of 154 N/m. Box B
has a mass of 54.7 kg. Tests of the air resistance of box B show that, when
it is moving at 2.00 m/s, it experiences a force from the air of 1.53 N.
(a) If you simply connect box B to panel P using the spring, as shown
in figure 5.P.5a, explain why the resulting steady-state behavior of
box B, xB (t), is simply the steady-state behavior for xP = Ad cos ωw t
added to the steady state behavior for xP = Ad cos ωu t.
(b) If you now wait for a steady state to be achieved, what is the ratio
of the amplitude of unwanted oscillations of box B (at 75.0 Hz) to
the amplitude of wanted oscillations of box B (at 0.010 Hz)?
You take the results of your calculations to your team leader.
“Well,” she says, “that’s a pretty impressive reduction of those 75 Hz
oscillations. Unfortunately, this is not good enough—the undesired
oscillations must be reduced in amplitude even further. Even more
unfortunately, it’s too late to order a different spring to use as the
tether.”
Then, inspiration strikes you! You could cut the spring into two
pieces, and insert an extra mass in the middle, as suggested in figure
5.P.5b. This would change the resonant characteristics of the system,
and so might improve rejection of the undesired oscillations. “Good
idea!” says your leader, “Crunch the numbers for me.”
You have a mass of 5.00 kg available; to simplify your calculation,

assume it has a diameter which is negligible on the scale of the
length of the spring. Air resistance tests on the 5.00 kg mass show
that, when it is moving at 2.00 m/s, it experiences a force from the
air of 0.33 N.
As you showed in part (b), the amplitude of oscillation of box B
at 75.0 Hz is very small in any case. Therefore, to make your life
simpler in determining the best configuration, you should make the
assumption that the amplitude of motion of box B at the unwanted
frequency (75 Hz) is essentially zero, although we will still be
interested in the forces exerted on it at this frequency.
(c) Explain qualitatively why the steady-state amplitude of oscillations
of box B at the wanted frequency of 0.010 Hz is essentially the same
for configurations a and b.
(d) Compare the force exerted on box B at 75 Hz for configuration
b to that for configuration a. (You should find that the force for
configuration b is almost 1,000 times lower!)
(e) “That’s great!” says your team leader. “But what if we take your
idea even further? How about if we cut the original spring into three
pieces, and add another intermediate mass?” She sketches another
possibility (figure 5.P.5c). At first you think, “If one mass made
the situation better, then two masses will work even better, and
would reduce the 75 Hz force exerted on box B even more!” But
then, you begin to realize that may not be right. Making the same
assumption that box B is essentially immobile, calculate the normal
mode frequencies for configuration c. Calculate the Q factor for
each of the normal modes. Then, use rough approximations to show
that, in fact the force transmitted at the unwanted frequency is a
little larger than for configuration b.
6 Asymmetric Coupled Oscillators
and the Eigenvalue Equation
Asymmetric bobs
Coupled in uneven dance—
Now I am enthralled.
—Marian McKenzie
6.1 Matrix math
What if we have a coupled oscillator system, similar to the coupled pendula of chapter 5,
in which the two oscillators are not identical? For example, what if one of the masses
is larger, or one of the pendulum strings is longer? This is a very important question for
mechanical systems. It also has important quantum mechanical analogs, such as the
formation of molecules from two different types of atoms (e.g., the CO2 molecule).
Before we dive into the physics, it will be helpful to set up the fundamentals
of matrix math. As you have seen in chapter 5, matrices provide a powerful way
of describing coupled oscillators. For asymmetric systems, we will need more
sophisticated matrix operations.
Matrix multiplication
It is easiest to explain matrix multiplication with an example:
⎛ ⎞⎛ ⎞
A B C j k l
⎝D E F ⎠⎝m n o⎠
G H I p q r
⎛ ⎞
Aj + Bm + Cp Ak + Bn + Cq Al + Bo + Cr
= ⎝ Dj + Em + Fp Dk + En + Fq Dl + Eo + Fr ⎠ .
Gj + Hm + Ip Gk + Hn + Iq Gl + Ho + Ir
We see that multiplying two 3 × 3 matrices creates a new 3 × 3 matrix, and that to
form each entry, we multiply each of the three terms in the corresponding row of the
179
left matrix with each of the three terms in the corresponding column of the right matrix:
⎛ ⎞
A B C ⎛ j k l⎞
−−−−→
⎜D E F ⎟⎝ m n o⎠
⎝ ⎠
−−−−−→
G H I p q r
⎛ ⎞
Aj + Bm + Cp Ak + Bn + Cq Al + Bo + Cr
= ⎝ Dj + Em + Fp ☛ Dk + En + Fq ✟Dl + Eo + Fr ⎟
⎜
⎠.
Gj + Hm + Ip Gk + Hn + Iq Gl + Ho + Ir
✡ ✠
This means that the width (number of columns) of the left matrix must match the height
(number of rows) of the right matrix. Therefore, we can multiply a three-entry column
vector by a 3 × 3 matrix:
⎛ ⎞⎛ ⎞ ⎛ ⎞
A B C j Aj + Bk + Cl
⎝ D E F ⎠ ⎝ k ⎠ = ⎝ Dj + Ek + Fl ⎠ .
G H I l Gj + Hk + Il
We see that multiplying a vector by a matrix creates a new vector, which will ordinarily
have a different length and direction. You can think of this as the matrix acting on a
vector to change it into a different vector. For example, the matrix

0 −1
1 0
◦
acts on a vector (one with two entries, of course) to rotate it by 90 counterclockwise,
without changing its length.

2
Self-test: Verify that the above matrix, when applied to the vector , works as
3
described.
Of course, we can also create a matrix that changes the length of any vector without
rotating it: the matrix

A 0
0 A
changes the length of any vector by a factor A:

A 0 a Aa a
= =A .
0 A b Ab b
However, most matrices, when they act on a vector, change both the length and the
direction.
The way that a matrix acts on a vector to change it into a different vector is
analogous to the way a function behaves: a function acts on a number to change it into
a different number. For example, the function y (x) = x 2 can act on the number 2 to
change it into the number 22 = 4. Taking this one step further, we will soon encounter
“operators.” An operator acts on a function to change it into a different function. For
d
example, the operator acts on the function x 3 to change it into the function 3x 2 .
dx
Chapter 6 ■ Asymmetric Coupled Oscillators and the Eigenvalue Equation 181
Figure 6.1.1 A view of someone (as seen from the back),

with right arm extended, palm facing down. The matrix Â
represents a counterclockwise (as viewed from above)
rotation about the vertical axis, while the matrix B̂
represents a counterclockwise (as viewed from the left)
rotation about the horizontal axis.
You should be aware that matrix multiplication doesn’t commute, that is, the order
of multiplication matters. For example,

A B e f Ae + Bg Af + Bh
=
C D g h Ce + Dg Cf + Dh

e f A B Ae + Cf Be + Df
while = .
g h C D Ag + Ch Bg + Dh
Since matrices can be used to represent rotations, we can see how this lack of
commutation works for one particular case. Stand up now (really!). Hold your right arm
up, so that it’s parallel to the floor, extending to the right away from your body, with your
palm facing down, as shown in figure 6.1.1. We denote matrices using capital bold-face
◦
letters that have hats. Let the matrix Â represent a 90 rotation counterclockwise (as
viewed from above) about a vertical axis passing through your right shoulder (ouch!).
◦
Let the matrix B̂ represent a 90 rotation counterclockwise (as viewed from someone
standing to your left) about a horizontal axis which passes through both your shoulders.
Now, let the combination ÂB̂ act on your arm:
ÂB̂ (arm pointing right, palm down) = Â [B̂ (arm pointing to right, palm down)]
= Â (arm pointing right, palm pointing back)
= arm pointing forward, palm pointing right
If instead you do the operations in the other order, the result is very different:
B̂Â (arm pointing right, palm down) = B̂ [Â (arm pointing to right, palm down)]
= B̂ (arm pointing forward, palm down)
= arm pointing down, palm pointing back
Determinants
The study of determinants dates back to the third century BC in China, actually
preceding the study of matrices. The word “determinant” was coined by Gauss in
1801, because they determine whether a system of equations (represented by a matrix)
has a unique solution. We will use this property to find the frequencies of the normal
modes.
The determinant of a 2 × 2 matrix is defined as follows:

a b
det ≡ ad − bc. (6.1.1)
c d
For larger matrices, you must use a multi-step procedure. First, you put alternating +
and − signs above the columns. Then, multiply the top left matrix element by the +
sign you just wrote above it and by the determinant of the smaller matrix formed by
ignoring the top row and the left column, thus forming the first term in the determinant:
⎛ ⎞
+ − +

⎜a b c⎟ e f
det ⎝
⎜ ⎟ = +a det + ...
d e f ⎠ h i
g h i
Now, move to the next entry in the top row. Multiply it by the − sign you wrote above
it, and by the determinant of the smaller matrix formed by ignoring the top row and
the second column:
⎛ ⎞
+ − +

⎜ a b c⎟
det ⎜ ⎟ = +a det e f − b det d f + . . .
⎝d e f ⎠ h i g i
g h i
Repeat this procedure, adding terms to the determinant, until you get to the rightmost
entry of the top row; for our example, there is only one more term:
⎛ ⎞
+ − +

⎜ a b c⎟ e f d f d e
det ⎝
⎜ ⎟ = +a det − b det + c det
d e f ⎠ h i g i g h
g h i
Using this procedure, you can, for example, reduce the calculation of the determinant
of a 5 × 5 matrix to the calculation of the determinants of five 4 × 4 matrices. Each of
these can in turn be reduced to the calculation of four 3 × 3 matrices. Each of these can
be reduced to the calculation of three 2 × 2 matrices, for which you can use equation
(6.1.1).
In practice, for the determinants of anything larger than a 3 × 3 matrix, you may
prefer to use a symbolic algebra program, such as Mathematica. Instructions are given
on the website for this book, under the listing for this section.
6.2 Equations of motion and the eigenvalue equation
We’ll start by considering just two nonidentical oscillators, but the method we develop
will be easily generalizable to a larger number. Our model system is one with two
coupled pendula, where the lengths and masses might be different, as shown in
figure 6.2.1. We expect that there are normal mode solutions for this system, and
Figure 6.2.1 An asymmetric coupled pendulum system.
once we have found them, we can use the techniques of chapter 5 to perform normal
mode analysis, and find the behavior of the system as a function of time, given the
initial conditions. However, it is no longer obvious exactly what the normal modes
are; it appears likely that there would be some type of breathing mode, and some type
of pendulum mode, but perhaps the amplitudes for the two bobs would be different.
In Hilbert space terms, we are trying to find the two eigenvectors, that is, the
two vectors that describe the initial positions of the bobs for the normal modes (for
zero initial velocities). It will turn out that these eigenvectors are always mutually
perpendicular, reflecting the fact that the different normal modes act as completely
independent oscillators.1
To find the normal modes, we fall back on the procedure used in chapters 1–4:
Step 1: Apply FTotal = mẍ to each object in the system to get one differential equaiton
for each object.
Your turn: Show that the application of FTotal = mẍ to the left bob gives
k
ẍ1 + ωA2 x1 − x = 0, (6.2.1)
m1 2
g k
where ωA2 ≡ + (6.2.2)
ℓ1 m1
Concept test: Explain why ωA represents the angular frequency at which bob 1 would
oscillate if bob 2 were held fixed at x2 = 0. (Answer below.2 )
1. However, for systems of coupled oscillators with unequal masses, the axes of Hilbert space (i.e.,
the x10 and x20 axes for the case of a two-oscillator system) must be scaled according to the
square root of the corresponding mass for this orthogonality to work out correctly. There is no
quantum mechanical analogy for this, so we will not spend much time on it. (The analogy would
be a particle whose mass depends on position.) The procedure is detailed in section 6.7.
Similarly, applying FTotal = mẍ to the right bob gives

k
ẍ2 + ωB2 x2 − x = 0, (6.2.3)
m2 1
g k
where ωB2 ≡ + . (6.2.4)
ℓ2 m2
Step 2: Guess a solution. We believe there will be normal mode solutions, that is,
solutions in which all parts of the system move with the same frequency, but may have
different (perhaps negative or zero) amplitudes. Also, based on our experience with
the symmetric coupled pendula, we expect that the phases of all the objects in a normal
mode will be the same.3 The most general way to represent such a motion would be
?
x1 = Re z1 z1 = X1 eiϕ eiωt .
?
x2 = Re z2 z2 = X2 eiϕ eiωt .
Note that the phases and frequencies are the same for both bobs; only the amplitudes
X1 and X2 are different. By the correct choice of when t = 0, we can arrange to have
ϕ = 0, so that our guess simplifies to
? ?
z1 = X1 eiωt z2 = X2 eiωt . (6.2.5)
With this choice of when t = 0, we see that the amplitudes have a simple interpretation:
X1 = x10 and X2 = x20 .
Step 3: Plug the guess into the DEQs, and see whether it works, and whether there are
restrictions on the parameters in the guess. By analogy with the symmetric coupled
pendulum system, we expect that there will be restrictions on ω and on the relative
X
amplitudes, that is, on the ratio 2 . Plugging equation (6.2.5) into (6.2.1) gives
X1
k
− ω2 X1 eiωt + ωA2 X1 eiωt − X eiωt = 0
m1 2
k
⇔ ωA2 X1 − X = ω2 X1 . (6.2.6a)
m1 2
m1 g
2. The total force on bob 1 is the pendulum force − x plus the spring force, which depends
ℓ1 1

on the relative position of the two bobs: Fspring = −k x1 − x2 . Thus, if x2 = 0, the total
m g
force reduces to Ftotal = −keff x1 , where keff = 1 + k. Therefore, bob 1 will oscillate with
ℓ1

keff g k
ω= = + .
m1 ℓ1 m1
◦
3. “But wait!” you say, “In the breathing mode, aren’t the bobs 180 out of phase?” This is indeed one
correct way of describing the breathing mode of symmetric coupled pendula: positive amplitudes
◦
for both bobs, and 180 phase difference. However, it is completely equivalent to say that one
bob has a negative amplitude, and that the phase factors are the same. That’s the option we’ll
choose.
Similarly, plugging into equation (6.2.3) gives
k
ωB2 X2 − X = ω2 X2 . (6.2.6b)
m2 1
We saw in chapter 5 that matrix methods were helpful. So, let us cast equations (6.2.6)
into matrix form, with the top line of the matrix equation representing (6.2.6a), and
the bottom line representing (6.2.6b):
⎛ ⎞
k
ωA2 −

⎟ X 1 = ω 2 X1 .
m1 ⎟
⎜
⎜
k ⎠ X (6.2.7)
⎝
− ωB2 2 X2
m2
Your turn: Verify that the top line of equation (6.2.7) really does represent
equation (6.2.6a), and that the bottom line really does represent equation (6.2.6b).
Note that this equation looks similar in some ways to one encountered in section 6.1,

A 0 a a
=A .
0 A b b

a
However, the important difference is that when any vector is multiplied by the
b

A 0
matrix , only the length of the vector is changed (and not its direction).
0 A
In contrast, equation (6.2.7) only works for a small number of special vectors. For
example,
if we apply the matrix in equation (6.2.7) to an “arbitrary” vector such as

1
, we get
0
⎛ ⎞
k ⎛ ⎞
2 − ωA2
⎜ ωA

m1 ⎟ 1
⎟ 1
⎠ 0 = ⎝ − k ⎠ = const. 0 .
⎜ ⎜ ⎟
⎝ k 2
− ωB m
m2 2
Each special vector for which equation (6.2.7) does hold represents the initial positions
X1 and X2 of the bobs in one of the normal modes. Therefore, these special vectors
are the eigenvectors, multiplied by an amplitude.
Example: in the symmetric coupled pendula, we found that the breathing mode was
( 1
represented by the eigenvector eb = √1 . We could excite this mode with any
2 −1
overall amplitude. For example, we could set x10 = 5 cm and x20 = −5 cm. Then, (making
use of the facts that, for the symmetric case ωA = ωB and m1 = m2 = m) the left side of
continued
equation (6.2.7) would read

⎛ ⎞
2 k ⎛ ⎞
k
⎜ ωA − 2 −

m 1
⎟ X1 ⎜ ωA m⎟ 5 cm
⎜
⎝ k
⎟ =⎝ ⎠ .
⎠ X2 k −5 cm
− ωB2 − ωA2
m2 m
Let’s verify that equation (6.2.7) really holds for this case, bearing in mind that for this
g k
symmetric case ωA2 ≡ + :
ℓ m
⎛ ⎞ ⎛ ⎞
2 k 2 k
ωA − 5 cm ωA (5 cm) − ( −5 cm)
k 5 cm
⎜ m ⎟ ⎜ m ⎟ 2
⎝ k ⎠ =⎝ k ⎠ = ωA +
− ωA2 −5 cm − (5 cm) +ωA2 (−5 cm) m −5 cm
m m
"
#

g k k 5 cm g 2k 5 cm 2 5 cm
= + + = + = ωb .
ℓ m m −5 cm ℓ m −5 cm −5 cm
As we had arranged, ω which appears in equation (6.2.7) is the normal mode frequency.
Each different eigenvector (or simple multiple thereof) which solves equation
(6.2.7) is associated with a different squared normal mode angular frequency ω2 . Our
job now becomes finding the eigenvectors and corresponding “eigenvalues” ω2 for
which equation (6.2.7) holds. This form of equation is called an “eigenvalue equation”:
⎛ ⎞
2 k
⎜ A ω −

(6.2.7): ⎜
m1 ⎟⎟ X1 = ω 2 X1
.
⎝ k 2 ⎠ X ! X2
− ωB 2
eigenvalue !
m2 !
! eigenvector eigenvector
matrix Â

( X1
To write this in bra-ket notation, we define eu ≡
, where the subscript “u”
( X2
indicates unnormalized, that is, the vector eu does not necessarily have length 1. Later,
we will discuss the (simple) process for normalizing eigenvectors. With this notation,
the equation becomes:
( (
Â eu = ω2 eu .
6.3 Procedure for solving the eigenvalue equation
Matrix eigenvalue equations such as equation (6.2.7) have been studied for a long time,
and there is a well-developed recipe to find the eigenvalues and eigenvectors.
Step A: Write the characteristic equation

The “characteristic equation” is the same as the eigenvalue equation, just
rearranged so that the right side is zero.
Your turn: Show that the eigenvalue equation

⎛ ⎞
2 k
⎜ Aω −
m1 ⎟ X1 2 X1
(6.2.7): ⎝ k
⎜ ⎟
⎠ X =ω
− ωB2 2 X2
m2
eigenvalue equation for two coupled oscillators
is equivalent to
⎛ ⎞
2 − ω2 k
ω
⎜ A −
⎜ m1 ⎟ ⎟ X1 = 0. (6.3.1)
⎝ k ⎠ X
− ωB2 − ω2 2
m2 !
! | eu
B̂
characteristic equation for two coupled oscillators

0
(Here, the “0” on the right side is shorthand for .) Hint: Remember that each matrix
0
equation is shorthand for two regular equations. If you can show that this pair of regular
equations for (6.3.1) is equivalent to the pair for (6.2.7), then you’ve shown that the matrix
equation (6.3.1) is equivalent to the matrix equation (6.2.7).
Step B: Use theorem from linear algebra

There is a general theorem that states that any equation of the form of
equation (6.3.1), that is, of the form B̂ r = 0, can hold if and only if det B̂ = 0.
Proof for 2 × 2 case:

a b e
=0⇒
c d f
!
B̂
⎫
bf ⎬

Top line : ae + bf = 0 ⇔ e = − bf
a ⇒ c − + df = 0 ⇔
Bottom line : ce + df = 0⎭ a
−cb + ad = 0 ⇔ det B̂ = 0
Applying this theorem to equation (6.3.1) gives us a quadratic equation for ω2 :
B̂ r = 0 ⇒ det B̂ = 0, that is,

k2
. ωA2 − ω2 ωB2 − ω2 − =0⇔
m1 m 2

k2

ω4 − ω2 ωA2 + ωB2 + ωA2 ωB2 − = 0.
m 1 m2
Using the quadratic equation for the variable ω2 , we obtain

2 2 k2
ωA2 + ωB2 ± ωA + ωB2 − 4 ωA2 ωB2 −
m1 m 2
ω2 =
2

2 2 2 k2
ωA2 + ωB2 ± ωA + 2ωA2 ωB2 + ωB2 − 4ωA2 ωB2 + 4
m 1 m2
=
2

2 2 2 k2
ωA2 + ωB2 ± ωA − 2ωA2 ωB2 + ωB2 + 4
m1 m2
=
2

2 2 k2
ωA2 + ωB2 ± ωA − ωB2 + 4
m1 m2
⇒ ω2 = (6.3.2)
2
eigenvalues for two coupled oscillators
Note that this equation tells us both values of ω (i.e., the angular frequencies of both
normal modes) because of the “±.”
Often, this is as far as we need to go, since often we only care about the frequencies
of the normal modes. However, sometimes we also wish to know the way in which the
masses are moving, which means we need to find the eigenvectors. (Again, for a pure
normal mode state, the eigenvectors tell us the relative initial positions of each mass,
and each mass oscillates in simple harmonic motion, with amplitude equal to its initial
position.)
Step C: To find the eigenvectors, take the eigenvalues, one at a time, and substitute
them into the characteristic equation. Solve for X 2 in terms of X 1 .
Note that we cannot find the actual values of both X1 and X2 , but only their
relative values, since we are only making use of the underlying interactions of the
system, and not of the initial conditions; each normal mode can be excited with an
arbitrary amplitude, as determined by the initial conditions. Implementing step C for
the general case, even for a system of only two coupled oscillators, is very messy and
not very instructive. Instead, we will illustrate with a special case.
Core example: Consider the special case of coupled pendula with equal masses√ m1 =
m2 = m = 1 kg , but unequal lengths ℓ1 = 4.33 m and ℓ2 = 2.30 m, and let k = 3 N/m.
This is shown in figure 6.3.1. Then,
k2
= 3 rad4 /s4 ,
m2
g k √
and ωA2 ≡ + = (2.26 + 3) rad2 /s2 = 4 rad2 /s2 ,
ℓ1 m1
Figure 6.3.1 An example of an asymmetric coupled pendulum system with equal

masses.
g k √
while ωB2 ≡ + = (4.26 + 3) rad2 /s2 = 6 rad2 /s2 .
ℓ2 m2
Plugging these into equation (6.3.2) gives

4 + 6 ± (4 − 6)2 + 4 · 3
ω2 = rad2 /s2
2
√ √
10 ± 4 + 12 2 2 7 2 2 7
= rad /s = rad /s ⇒ ω = √ rad/s.
2 3 3
We anticipate that the higher value of ω corresponds to a breathing-type mode, and the
lower value to a pendulum type mode. Now, to find the eigenvectors. For the breathing
mode, substitute ω2 = 7 rad2 /s2 into the characteristic equation (6.3.1):
⎛ ⎞
k
ω 2 − ω2 − √
⎜ A
⎜ m1 ⎟ ⎟ X1 = 0 ⇒ 4 − √ 7 − 3 rad2 /s2 X1 = 0.
⎝ k ⎠ X − 3 6−7 X2
− ωB2 − ω2 2
m2
We now divide both sides by 1 rad2 /s2 , giving

√
4− √ √
√7 − 3 X1
= 0 ⇒ Top line: − 3X1 − 3 X2 = 0 ⇔ X2 = − 3 X1
− 3 6−7 X2
Note: the bottom line of the above matrix equation tells us the same thing:
√ √
Bottom line: − 3 X1 − X2 = 0 ⇔ X2 = − 3X1 .
This redundancy arises because we cannot specify the actual values of X1 and X2 , but
only their relative values, since we are only making use of the underlying interactions
of the system, and not of the initial conditions. As expected for a breathing-type mode,
the amplitude for right bob, X2 is opposite in sign to that for the left bob. We also
continued
see that the amplitude for the right bob (the one with the shorter string) is larger in this
mode. Expressing our result in column matrix form, we have
found that the eigenvector
( 1
√ . This is illustrated in
for the breathing mode (not yet normalized) is eub =
− 3
figure 6.3.2a.
Figure 6.3.2 Normal modes for the system shown in figure 6.3.1. a: breathing mode.
b: pendulum mode.
Your turn: Find the eigenvector for the pendulum mode by substituting ω2 = 3 rad2 /s2
into the characteristic equation, (6.3.1), and then using the top line of the resulting
1
matrix equation to show that X2 = √ X1 . Now, show that the bottom line of the matrix
3
equation says the same thing.
As expected for the pendulum mode, the amplitudes have the same sign for both
bobs, but now the right bob has a smaller amplitude, as shown in figure 6.3.2b. Expressing
our result in column matrix form, we have
found that the eigenvector for the pendulum
&
1√
mode (not yet normalized) is eup = .
1/ 3
6.4 Systems with more than two objects
To generalize these ideas to a larger system of coupled oscillators, we use exactly the
same methods.4 Assuming there are N objects in the system, here is the procedure:
1. Write FTOT = mẍ for each object. This creates a system of N coupled
differential equations.
2. Guess a normal mode solution, that is, guess that the position of object j is
given by
?

xj = Re Xj eiωt ,
where ω is the angular frequency of the normal mode, and Xj is the amplitude of
oscillation of object j . As before, we have chosen the time when t ≡ 0 so that the
overall phase factor ϕ = 0. Therefore, the amplitude Xj of object j is equal to the initial
position xj0 of object j.
3. Plug the guess back into the system of differential equations. All the terms will
contain the eiωt factor, so this can be cancelled out, leaving a system of N coupled
linear equations involving the variables Xj .
4. Write the system of equations as an eigenvalue equation, that is, a matrix
equation of the form
( (
Â eu = ω2 eu ,
⎛ ⎞
X1
( ⎜X ⎟
where Â is an N × N matrix, and eu is the column vector ⎝ 2 ⎠.

..
.
5. Rearrange the eigenvalue equation to form the characteristic equation, that is,
the equivalent equation which has the form
(
B̂ eu = 0.
6. To find the eigenvalues ω2 , set
det B̂ = 0
This will give an Nth-order equation for ω2 , with N solutions.
4. The ideas described here work just as well for any system with more than two “degrees of
freedom,” whether there are one, two, or more objects. Here, a “degree of freedom” is a way
in which the system can move that is completely independent of other degrees of freedom, at
least before the coupling between degrees of freedom is added. For example, a rigid object in
three-dimensional space has six degrees of freedom, since it can move in three directions, and
rotate about three axes. As another example, if we had two particles, each of which was kept
near a particular point in three-dimensional space by restoring forces, and these particles were
coupled together by a spring, that would be a system with six degrees of freedom. In our main
discussion, however, we’ll stick to particles that are constrained to move in only one dimension,
since this is easier to illustrate and conceptually simpler.
7. To find the eigenvectors, take each value of ω2 , one at a time, and substitute it
back into the characteristic equation. Solve for X2 , X3 , etc., in terms of X1 .
This procedure is guaranteed to work. (For systems of five or more objects, it will
not usually be possible to solve for the eigenvalues analytically, but they can always
be found by numerical approximation.) This is a tremendously important realization,
since we have just proved that, for a system of N objects, we will always be able to find
N normal modes. In principle, we could write a system of N uncoupled differential
equations to represent these modes. (This is seldom necessary, but we certainly could
do it.) This system of DEQs would represent the behavior of the system just as well as
the system of N DEQs which describes the positions of the N objects. Therefore, we
can quite generally conclude:
For a one-dimensional system, the number of normal modes is equal to the

number of masses. Therefore, giving the amplitudes and phases of all the
normal modes provides just as complete a description of the system as giving
x(t) for each object in the system.
We will do a fairly generic example of a one-dimensional system with three objects,

from which it will be obvious how to generalize to a one-dimensional system with any
number of objects. Consider the system shown in figure 6.4.1. Masses 1 and 3 are
connected by the long spring k13 .
1. Applying FTOT = mẍ to m1 gives
m1 g
−k1 x1 − x1 − k12 x1 − x2 − k13 x1 − x3 = m1 ẍ1 . (6.4.1)
ℓ
This is the real part of
m1 g
−k1 z1 − z1 − k12 z1 − z2 − k13 z1 − z3 = m1 z̈1 . (6.4.2)
ℓ
(We won’t bother to apply FTOT = mẍ to the other masses, since the pattern will
become obvious just from m1 .)
Figure 6.4.1 system of three coupled

pendula with all possible couplings.
?

2. We guess the normal mode solution xj = Re Xj eiωt , that is, z1 = X1 eiωt ,
z2 = X2 eiωt , and z3 = X3 eiωt .
3. Plugging this guess into equation (6.4.2) gives
m g
− k1 X1 eiωt − 1 X1 eiωt − k12 X1 eiωt − X2 eiωt
ℓ

− k13 X1 eiωt − X3 eiωt = −m1 ω2 X1 eiωt ⇒

k1 g k12 k13 k k
+ + + X1 − 12 X2 − 13 X3 = ω2 X1 ⇒
m1 ℓ m1 m1 m1 m1
k12 k
ωA2 X1 − X − 13 X = ω2 X1 , (6.4.3)
m1 2 m1 3
where
k1 g k k
ωA2 = + + 12 + 13
m1 ℓ m1 m1
is the square of the angular frequency at which m1 would oscillate if all the other
masses were held fixed.
We can see that the result of applying FTOT = mẍ to m2 and plugging in our normal
mode guess would be
k12 k
− X1 + ωB2 X2 − 23 X3 = ω2 X2 , (6.4.4)
m2 m2
where ωB2 is the square of the angular frequency at which m2 would oscillate if the
other masses were held fixed.
Concept test (answer below5 ): What is the value of ωB2 in terms of ℓ, g, m2 , and the k’s?
Similarly, for m3 we would obtain
k13 k
− X1 − 23 X2 + ωC2 X3 = ω2 X3 , (6.4.5)
m3 m3
where ωC2 is the square of the angular frequency at which m3 would oscillate if the
k k k g
other masses were held fixed: ωC2 = 13 + 23 + 3 + .
m3 m3 m3 ℓ
k12 g k
5. ωB2 = + + 23
m2 ℓ m2
4. We can write equations (6.4.3), (6.4.4), and (6.4.5) in matrix form:
⎛ ⎞
2 k12 k13
⎜ ωA − −
⎜ m1 m1 ⎟⎛ ⎞
⎟ X1
⎛ ⎞
X1
⎜ k12 k23 ⎟
⎜−
⎜ m ωB2 − ⎟ ⎝ X ⎠ = ω2 ⎝ X ⎠ .
2 2 (6.4.6)
2 m2 ⎟
⎟ X
⎜
⎝ k13 k23 3 X3
ωC2
⎠
− −
m3 m3
In fact, now that we’ve gone through this procedure, you can probably construct the
equivalent of equation (6.4.6) for whatever system is given just by inspection.
From here, we could in principle go on to steps 5, 6, and 7 to find the eigenvalues
and eigenvectors. However, it is much easier to do this using a symbolic algebra
program (such as Mathematica), especially if there are more than three objects. Explicit
instructions for how to do this, including an example, are given on the website for this
text under the listing for this section.
6.5 Normal mode analysis for multi-object, asymmetrical systems
Our treatment of normal mode analysis in section 5.8 depended only on the
orthonormality of the normalized eigenvectors, so we can use exactly the same
methods for asymmetric systems with many objects, as long as we use the normalized
eigenvectors. We will go over things again below, to make sure this is clear. However,
first we need to understand the process for normalizing eigenvectors.
How to normalize eigenvectors

( ( (
If we find a vector e which solves the eigenvalue equation Â e = ω2 e , then the
( u u u
vector Deu (where D is a constant, perhaps a complex constant) will also solve the
equation, since
( ( ( ( ( (
Â eu = ω2 eu ⇔ DÂ eu = D ω2 eu ⇔ Â D eu = ω2 D eu .
Thus, there are an infinite number of vectors, all pointing along the same direction in
Hilbert space but with different lengths, which solve the eigenvector equation for a
particular eigenvalue ω2 . However, there is one vector that is particularly helpful: the
normalized eigenvector, that is, the vector that has length 1 in this direction. (Again,
just 1, not 1 m.)
Core example:
For the example from the end of section
6.3, we found that the eigenvector for
( 1
√
the breathing mode is of the form eub = . To normalize it, we just divide by the
− 3
√ 2
length: Length = 12 + − 3 = 2. So, the normalized eigenvector is
( ⎛ 1 ⎞
( e √
2
e = ub
b =⎝ 3 ⎠.
Length −
2
' (
As a check, if we’ve correctly normalized it we should have eb |eb = 1:
⎛ 1 ⎞

√
1 3 ⎝ √ 2 1 3
− 3 ⎠ = + = 1
2 2 − 4 4
2
The above normalization procedure is only valid when the masses are equal; the
more general procedure is treated in section 6.7.
Normal mode analysis

Here, I convince you that we only need a complete orthonormal basis in orderto (use
the same formulas for normal mode analysis as we( derived in section 5.8. Let e1 be
the normalized eigenvector for mode n = 1, and e2 be the normalized eigenvector for
mode n = 2, and so on.⎛Be careful⎞ not to confuse the mode index n with the mass index j.
X1
( ⎜ ⎟
For example, e1 = ⎝ X2 ⎠ is a vector of length 1 specifying the initial positions of
..
.
the
( masses (with zero initial velocities) for normal mode n = 1. The subscript “1” in
e refers to mode number 1, whereas the subscript “1” in X refers to mass number
1 1
1. It is easy to tell which
( type of index is being used from the context: any subscript
on a vector (e.g., e1 ) or a frequency (e.g., ω1 ) is a mode index. We will consistently

use the symbol n or m for the mode index and the symbol j for the mass index.
As detailed earlier, giving the amplitudes and phases of all the normal modes
provides just as complete a description of the system as giving x(t) for each object in
⎛ ⎞
x1 (t)
the system. This means that, for any state of the system, the vector |x (t) = ⎝ x2 (t) ⎠
⎜ ⎟
..
.
can be written as a superposition of the normal modes, that is,
+ N
,
(
|x (t) = Re Cn eiωn t en (6.5.1)
n=1
( ( (
(Again, the en ’s (i.e., e1 , e2 , etc.) are the normalized eigenvectors representing the
different normal modes.) This is the more general version
of the&normal mode expansion
i
(
for two coupled oscillators, (5.8.1): |x (t) = Re Cp e t ep + Cb eiωb t eb .
ωp

To find the complex amplitudes Cn in the normal mode expansion (6.5.1) from
the initial positions and velocities of the masses, we follow the example of section 5.8.
According to equation (6.5.1), the initial positions of the objects are given by6
⎛ ⎞
x10 + N ,
( ⎜ (
x ≡ ⎝ x20 ⎠ = Re
⎟
Cn en , (6.5.2)
0
.. n =1
.
and the initial velocities are given by

⎛ ⎞
ẋ10 + + N ,,
( ⎜ d d (
ẋ ≡ ⎝ ẋ20 ⎠ = [|x (t)]t =0 = Re Cn eiωn t en
⎟
0
.. dt dt
. n =1 t =0
+ N ,
(
= Re iωn Cn en .
(6.5.3)
n=1
Next, we write the complex amplitudes in Cartesian form:
Cn = Re Cn + i Im Cn .
This allows us to rewrite equations (6.5.2) and (6.5.3):

+ N , N
( ( (
x = Re C e = Re Cn en , (6.5.4)
0 n n
n=1 n=1
(
(since the en ’s are real) and
+ N , + N ,
( ( (
ẋ = Re iωn Cn en = Re
iωn Re Cn + i Im Cn en

0
n=1 n=1
+ N
, N
( (
Re ωn iRe Cn − Im Cn en = −
ωn Im Cn en . (6.5.5)
n =1 n =1
For the remainder of this section, we assume the masses of coupled oscillators
are equal. This is the case of most interest, because it is most closely analogous with
quantum mechanics. The general case of nonequal masses is treated in section 6.7.
Next, we need a couple of results regarding the normalized eigenvectors. The
normal modes do not interact with each other, therefore the eigenvectors are mutually
orthogonal, that is,
' (
em |en = 0 if n = m
(
(We prove this more rigorously in section 6.7.) The eigenvectors en are normalized,
that is,
' (
en |en = 1
6. Note that here, we use the subscript “0” to indicate t = 0. It does not indicate n = 0. (In fact,
there is no mode n = 0, since we start numbering the modes at n = 1.)
We can summarize the above two equations by writing

' (
em |en = δmn , (6.5.6)
where the “Kronecker delta function” (pronounced “Crow-neck-er”) is defined by

1 if n = m
δmn ≡ (6.5.7)
0 if n = m
(
Now, following the lead of section 5.8, we try taking the inner product of x0 with
each of the normalized eigenvectors:
N N
' ( ' ( ' (
em |x0 = em Re Cn en = Re Cn em |en .
n =1 n =1
' '
(Note that we must use em in the left side of the inner product above, rather than en ,
because the subscript n is already being used in the summation on the right side of the
inner product.) Because of orthonormality (as expressed in equation (6.5.6)), all the
'terms in (the sum are zero, except the one for which m = n, and for that term we have
em=n |en = 1. Therefore,
' (
em |x0 = Re Cm .
' (
Since we have eliminated the sum over n, we could just as well write this as en |x0 =
Re Cn
' (
⇒ Re Cn = en |x0 , (6.5.8a)
% &
which is the generalized version of equation (5.8.2a): Re Cp = ep |x0 .
( Next, again following the lead of section 5.8, we try taking the inner product of
ẋ with each of the normalized eigenvectors:
0
N N
' ( ' ( ' (
e |ẋ = e −
m 0 m ω Im C e = −
n n n ω Im C e |e = −ω Im C .
n n m n m m
n=1 n=1
' (
Again,
since the sum over n has been eliminated, we could write this as en |ẋ0 = −ωn
Im Cn
1 ' (
⇒ Im Cn = − en |ẋ0 , (6.5.8b)
ωn
1 % &
which is the generalized version of equation (5.8.2b): Im Cp = − ep |ẋ0 .
ω
( p
Note that for the important special case of ẋ0 = 0, Im Cn = 0, so that the
amplitudes
' ( that appear in the normal mode expansion (6.5.1) reduce to Cn =Re Cn =
en |x0 .
Stated in graphical
( terms, this means that, in the normal mode expansion for the
special case ẋ0 = 0, the coefficient along the Hilbert space axis corresponding
( to
mode n is given by the inner product of the initial state of the system x0 with the

(
normalized eigenvector for that mode, en . Just as for the coupled pendulum, this idea
of taking inner products to find the “projections” of the state of the system along the
normal mode “directions” in Hilbert space (by taking these inner products) is exactly
analogous to the process of finding the projections of an ordinary vector along the
x- or y-axis (by taking dot products with î or ĵ).
6.6 More matrix math
In this section, we go over a few additional aspects of matrix math that are needed for
section 6.7, and that are also quite important for the study of quantum mechanics.
Inverse matrices and the identity matrix

One can show that, if and only if the determinant of a matrix Â doesn’t equal zero,
then one can find the inverse matrix Â−1 , such that
ÂÂ−1 = Â−1 Â = Î,
where Î is the identity matrix:

⎛ ⎞
1 0 ···
Î = ⎝ 0 1 0 ⎠ .
⎜ ⎟
.. ..
. 0 .
Note that multiplying by Î has no effect on anything, just as multiplying by 1. For

example, in multiplication of 2 × 2 matrices, we have

1 0 a b a b
ÎÂ = = .
0 1 c d c d
This means that the matrix Â−1 “undoes” the effect of the matrix Â. For instance, if
◦
the matrix Â rotates vectors by 31 clockwise about the z-axis, then the matrix Â−1
◦
rotates vectors by 31 counterclockwise about the z-axis.
Another feature of bra-ket notation

We’ve discussed how, unlike multiplication of numbers, the order matters for matrix
multiplication. However, matrix multiplication is associative, meaning, for example,
that

Â B̂Ĉ = ÂB̂ Ĉ.
We can take advantage of this in

bra-ket
notation. Let’s define an arbitrary

bra
g| =
x1 a b
g1 g2 , an arbitrary ket |x = , and an arbitrary matrix Â = . Then,
x2 c d
we can write
g| Â |x .

One might think that this is ambiguous. Should we interpret this as g| Â |x , meaning
first do the matrix multiplication g| Â, then matrix multiply the result by |x ? Or,
should we instead interpret it as g| Â |x ? However, because matrix multiplication is
associative, these two interpretations give the same result. (You can show this explicitly
in problem 6.16.) So, when interpreting g| Â |x , we can think of Â “operating” to the
left on g| or operating to right on |x .
Hermitian matrices
In the next section, and frequently in the study of quantum mechanics, you’ll need to
take the adjoint of Â |x :
†
Â |x =?
By considering a system of two masses, we’ll be able to see the general pattern. Let

x1 a b
|x = and Â = .
x2 c d
Then,

a b x1 ax1 + bx2
Â |x = = .
c d x2 cx1 + dx2
To take the adjoint, we change it from a column matrix to a row matrix, and take the
complex conjugate of the entries:
†
Â |x = a∗ x1∗ + b∗ x2∗ c∗ x1∗ + d ∗ x2∗ .
Claim: We can write this as

†
Â |x = x | Â† ,
where Â† is the adjoint of Â, that is, the matrix formed by interchanging columns (the
first row becomes the first column, etc.) and then taking the complex conjugate of the
entries.
Check:

† a∗ c∗ †
a∗ c∗
Â = ⇒ x | Â = x1∗ x2∗
b∗ d∗ b∗ d∗

= a∗ x1∗ + b∗ x2∗ c∗ x1∗ + d ∗ x2∗
Most of the matrices in the next section, and virtually all the matrices of interest
in quantum mechanics, have the special property that they are “self adjoint,” meaning
that
Â† = Â.
We will use the more common term “Hermitian” for such matrices, instead of “self
adjoint.”
Concept test (answer below7 ): Which of the following matrices are Hermitian?
⎛ ⎞
k
ω 2 − √
⎜ A m1 ⎟ Ai 0 −√3 − 3
(a) ⎝ k
⎜
⎠ (b) 0 A (c) − 3
⎟ rad2 /s2
− 2
ωB − 1
m2
In fact, we can see that any 2 × 2 Hermitian matrix can be written as

a b
Â = . (6.6.1)
b∗ c
Form of a 2 × 2 Hermitian matrix, with a and c real.
For Hermitian matrices, we have

†
Â |x = x | Â† = x | Â. (6.6.2)
Adjoint of products of Hermitian matrices.

In the next chapter section (and in quantum mechanics), we will need to take the adjoint
†
of ÂB̂ |x , where Â and B̂ are both Hermitian. We will show that ÂB̂ |x = x | B̂Â.
To begin, we define
|β ≡ B̂ |x .
so that
† †
ÂB̂ |x = Â |β .
Using equation (6.6.2), this means that

†
ÂB̂ |x = β| Â. (6.6.3)
Since B̂ is Hermitian, we have that

†
β| ≡ B̂ |x = x | B̂† = x | B̂.
Plugging this into equation (6.6.3) gives

†
ÂB̂ |x = x | B̂Â. (6.6.4)
True if Â and B̂ are Hermitian.
We see that, when taking the adjoint, we must reverse the order of the two matrices.
†
Therefore, ÂB̂ = B̂Â. Since this is not necessarily equal to ÂB̂, we see that the
product of two Hermitian matrices need not be Hermitian.
7. Only (c) is Hermitian. However, (a) would be Hermitian if m1 = m2 .

6.7 Orthogonality of normal modes, normal mode coordinates,

degeneracy, and scaling of Hilbert space for unequal masses
For the symmetric coupled pendulum system, we saw in chapter 5 that the quantity sp ≡
1
√ x1 + x2 , which characterizes the motion of the pendulum mode, always oscillates
2
at ωp , even when the system is in a superposition of pendulum and breathing modes,
1
and that the quantity sb ≡ √ x1 − x2 , which characterizes the breathing mode,
2
always oscillates at ωb . The quantities sp and sb are called “normal mode coordinates”;
each is the special linear combination of the coordinates of the masses that oscillates
at the normal mode angular frequency.

& 1 1
Later in the chapter, we found the eigenvectors for these modes, ep = √

2 1
( 1 1 x1 (t)
and eb = √ . Therefore, defining |x (t) ≡ , we can write
2 −1 x2 (t)
' ( ' (
sp = ep |x (t) and sb = eb |x (t) .
We will show in this section8 that, for the case of equal masses, the analogous
expressions for normal mode coordinates hold even for asymmetrical systems with
many masses, that is, that the coordinate
' (
sn ≡ en |x (t) , (6.7.1)
(
where en is one of the eigenvectors of the system, oscillates at the associated angular
frequency ωn . (As for the symmetric coupled pendulum system discussed in chapter
5, the coordinate sn is associated
( with an sn0 axis in Hilbert space, and the direction of
this axis is defined by en .)
When the masses are unequal, we must modify the recipe for finding normal mode
coordinates. We will show that the normal mode coordinate for unequal masses is
'
sn ≡ en M̂ |x (t) , (6.7.2)
Normal mode coordinate for the case of unequal masses.

⎛ ⎞ ⎛ ⎞
m1 0 ··· x1 (t)
⎠ and |x (t) ≡ ⎝ x2 (t) ⎠. Note
⎜ 0 m2 0 ⎟
where the “mass matrix” is M̂ = ⎝
⎜ ⎟
.. .. ..
. 0 . .
that M̂ is
' Hermitian.
( Also, when all the masses equal m, equation (6.7.2) becomes
xn ≡ m en |x (t) ; this is the same as equation (6.7.1) except for the constant m, which
only affects the overall scaling of the normal mode coordinate.
Now, we must show that equation (6.7.2) is indeed the correct definition for the
normal mode coordinate in the general case, meaning that sn oscillates at frequency
8. Most of this presentation is based on that in The Physics of Waves, by Howard Georgi
(Prentice-Hall, Englewood Cliffs, NJ, 1993), pp. 81–2.
Figure 6.7.1 The same system of three

coupled pendula as shown in figure 6.4.1.
ωn , so that

sn ∝ cos ωn t + ϕn . (6.7.3)
To show this, we need only prove that
s̈n = −ωn2 sn . (6.7.4)
We consider the three-mass system shown in figure 6.4.1, reproduced here for
convenience as figure 6.7.1. (It will be clear how to extend the arguments to a system
with any number of masses.) Recall that the result of applying FTOT = mẍ to mass 1
was (6.4.1):
m g
−k1 x1 − 1 x1 − k12 x1 − x2 − k13 x1 − x3 = m1 ẍ1 ⇒
ℓ

− k11 + k12 + k13 x1 + k12 x2 + k13 x3 = m1 ẍ1 ,
where k11 is the total spring constant associated with restoring forces that act on m1
but are not associated with coupling to m2 or m3 . In the example of figure 6.4.1,
k11 = k1 + m1 g/ℓ. We could also write the above as
−kAx1 + k12 x2 + k13 x3 = m1 ẍ1 , (6.7.5a)
where kA = m1 ωA2 = k11 + k12 + k13 . Applying FTOT = mẍ to m2 and to m3 would
similarly result in
k12 x1 − kB x2 + k23 x3 = m2 ẍ2 (6.7.5b)
and
k13 x1 + k23 x2 − kC x3 = m3 ẍ3 , (6.7.5c)
where kB = m1 ωB2 = k12 + k22 + k23 , and k22 = m2 g/ℓ is the total spring constant
associated with restoring forces that act on m2 , but are not associated with coupling to
m1 or m2. (The constant kC is defined analogously to kA.)
Your turn: Show that we can then rewrite equations (6.7.5) as

d2
−K̂ |x (t) = M̂ |x (t) , (6.7.6)
dt 2
where
⎛ ⎞
k −k12 −k13
⎜ A ⎟
K̂ = ⎝ −k12 kB −k23 ⎠ ,
−k13 −k23 kC
and is Hermitian.
(
Taking the inner product of both sides of equation (6.7.6) with en gives
' ( ' d 2 (
− en K̂x t = en M̂ 2 x t .
dt
'
Since neither en nor M̂ has any time dependence, we can rearrange the right side,
giving
' ( d2 ' (
− en K̂x t = 2 en M̂x t ,
dt
' (
Recall that our proposed sn is given by equation (6.7.2): sn ≡ en M̂x t . Therefore,
the above gives
' (
s̈n = − en K̂x t .
Comparing this with what we need to prove, (6.7.4): s̈n = −ωn2 sn , we see that we
must show
' (
en K̂x t = ωn2 sn . (6.7.7)
The eigenvalue equation for this system is (6.4.6):

⎛ ⎞
2 k12 k13
⎜ Aω − −
⎜ m1 m1 ⎟ ⎛ ⎞
⎟ X1
⎛ ⎞
X1
⎜ k12 2 k23
⎟
⎜− ω − ⎟ ⎝ X ⎠ = ω2 ⎝ X ⎠ .
⎜ m B 2 2
2 m2 ⎟⎟ X
⎜
⎝ k13 k 3 X 3
− 23 ωC2
⎠
−
m3 m3
We could rewrite this as
( (
Â en = ωn2 en , (6.7.8)
where
⎛ ⎞
k k
ωA2 − 12 − 13
⎜
⎜ m1 m1 ⎟
⎟
⎜ k12 k23 ⎟
⎜−m
Â = ⎜ ωB2 − ⎟.
⎜ 2 m2 ⎟
⎟
⎝ k13 k23
ωC2
⎠
− −
m3 m3
It will be helpful to separate out the part involving the masses:
Â = M̂−1 K̂, (6.7.9)

where
⎛ ⎞
1
⎜m 0 0 ⎟
⎜ 1 ⎟
⎜ 1 ⎟
M̂−1 ⎜ 0
=⎜ 0 ⎟⎟,
⎜ m2 ⎟
⎝ 1 ⎠
0 0
m3
and is Hermitian.
Your turn: Do the matrix multiplication to verify that Â = M̂−1 K̂.
Therefore, we can rewrite equation (6.7.8) as

( (
M̂−1 K̂en = ωn2 en .
The adjoint of this equation is

−1 († '
M̂ K̂en = ωn2 en .
(†
Since both M̂−1 and K̂ are Hermitian, we can apply equation (6.6.4): ÂB̂x =
'
x B̂Â, giving
' '
en K̂M̂−1 = ωn2 en .
(
Taking the inner product with M̂x(t) gives
' ( ' (
en K̂M̂−1 M̂x(t) = ωn2 en M̂x(t) .
' (
The right side of this includes sn ≡ en M̂x(t) , so we can rewrite the above as
' (
en K̂M̂−1 M̂x(t) = ωn2 sn
Note that M̂M̂−1 = M̂−1 M̂ = Î. Therefore,

' ( ' ( ' (
en K̂M̂−1 M̂x(t) = en K̂x(t) ⇒ en K̂x(t) = ωn2 sn ,
which
' is (6.7.7),
( completing the proof that sn , as defined by equation (6.7.2): sn ≡
en M̂x(t) , is indeed the normal mode coordinate.
' (
Let us be clear about the interpretation of s ≡ e M̂x(t) . This does not mean
n n
that, in order to excite normal mode number
( 1, we must position the masses in an initial
pattern of positions proportional
( to M̂ e . Rather, as usual, we should position them in
1
a pattern proportional to e1 ; when they are then released, the system will be in a pure
mode 1 oscillation. However, if we instead excite the system into a superposition of
modes, then (assuming unequal masses)' the coordinate
( that displays a simple sinusoidal
' (
motion at angular frequency ω1 is s1 ≡ e1 M̂x(t) . Other coordinates, such as e1 |x(t) ,
do not show simple sinusoidal motion when the system is in a superposition. (They
do show simple sinusoidal motion when the system is in a pure mode, as do all
coordinates.)
Figure 6.7.2 Top: Coupled pendula with

unequal masses. For the example discussed in the
text, m1 = 1 kg, m2 = 2 kg, ℓ = 1 m, k = 1 N/m,
and g = 10 m/s2 . Bottom: Position versus. time for
each of the two masses (x1 and x2 ) and for each of
the normal mode coordinates (s1 and s2 ).
As an example, consider the two-pendulum system shown in the top of figure 6.7.2,
in which m1 = 1 kg, m2 = 2 kg, ℓ = 1 m, k = 1 N/m, and g = 10 m/s2 . You can show
in problem 6.17 that the (unnormalized) eigenvectors for this system are

e = 1 and e = −2 .
( (
1 1 2 1
Therefore, to excite mode 2 (which is a breathing-like mode), we could displace m1
2 cm to the left of equilibrium, displace m2 1 cm to the right of equilibrium, and release
them. The system would then oscillate in mode 2, in which m1 moves with an amplitude
of 2 cm and m2 moves with an amplitude of 1 cm, both with an angular frequency ω2 .
In this pure mode, any linear combination of x1 and x2 oscillates at ω2 , since both x1
and x2 oscillate at ω2 . If instead we excite the system by holding m1 at equilibrium,
displacing m2 to the right by 1 cm, and releasing, then both modes are excited. (Unlike
the case for equal masses, this does not excite both modes equally; we’ll see below
how to calculate the amplitudes for the two modes.) The resulting behavior is therefore
complicated, as shown in figure 6.7.2. However, s1 shows simple harmonic oscillation
at angular frequency ω1 and s2 shows simple harmonic oscillation at the slightly higher
angular frequency ω2 , as shown in the figure.
Next, we wish to show that, for the case of equal masses, the normal modes are
mutually orthogonal. We’ll keep our arguments general, so that we’ll easily be able to
see how to adapt things for unequal masses. Again, we’ll consider a system of three
masses, but it will be obvious how to generalize the argument to any number of masses.
We consider an arbitrary state of the system, which is formed by a superposition of the
normal modes:
( ( ( (
x t = A cos ω t + ϕ e + A cos ω t + ϕ e + A cos ω t + ϕ e
1 1 1 1 2 2 2 2 3 3 3 3
(
= Ap cos ωp t + ϕp ep ,
p
where p is the mode index, and ranges from 1 to 3. Taking the inner product with M̂|en
gives
' ( ' (
en M̂x t = en M̂ Ap cos ωp t + ϕp ep
p
' (
= Ap cos ωp t + ϕp en M̂ep .
p
' (
We know from equations (6.7.2) and (6.7.3) that sn = en Mx t ∝ cos ωn t + ϕn .
Inserting this into the above gives
' (
cos ωn t +ϕn ∝ Ap cos ωp t +ϕp en M̂ep .
p
Since both sides of this equation must oscillate at the same angular frequency, we must
have
' (
en M̂ep = 0 for ωn = ωp . (6.7.10)
' (
We cannot quite yet write en M̂ep = 0 for n = p, since it is possible that two or
more of the normal modes have the same angular frequency. Such modes are called
“degenerate.” (The quantum mechanical analog is a situation in which two or more
of the stable states have the same energy.) For example, with appropriate choices of
spring constants and masses,
( ( might be able to arrange ω1 = ω2 , so that using
one
2
equation (6.7.8): Â en = ωn en we would have

( (
Â e1 = ω12 e1 (6.7.11)
and
( ( (
Â e2 = ω22 e2 = ω12 e2 . (6.7.12)
Forming the combination β1 (6.7.11) + β2 (6.7.12), where β1 and β2 are constants
gives
( ( ( (
β1 Â e1 + β2 Â e2 = β1 ω12 e1 + β2 ω12 e2 ⇒
( ( ( (
Â β1 e1 + β2 e2 = ω12 β1 e1 + β2 e2 .
( (
Thus, the vector β1 e1 + β2 e2 is also an eigenvector with eigenvalue ω12 . In other
words,any linear combination of degenerate eigenvectors is also an eigenvector with
the same eigenvalue. In particular, the vector
' (
′( ( e M̂ e (
e ≡ e − ' 1 2 ( e
2 2 1
e1 M̂ e1
is an eigenvector with eigenvalue ω12 .
' (
Your turn: Show that e1 M̂ e2′ = 0.
Thus, we can always choose a set of eigenvectors that fulfill

' (
en M̂ep = 0 for n = p. (6.7.13)
Equivalent of orthogonality for unequal masses
For equal masses, we have

' ( ' (
en M̂ep = m en |ep ⇒
' (
en |ep = 0 for n = p, (6.7.14)
Orthogonality of normal modes for equal masses
which is the general orthogonality condition we set out to prove.

When the masses are not equal, we can still use all the results from section 6.5
simply by inserting the M̂ matrix inside each inner product. The normalization
condition becomes
' (
en M̂ en = 1 (6.7.15)
Equivalent of normalization for the case of unequal masses
If
' the (eigenvectors are scaled so that the above holds, then the equivalent of (6.5.6),
em |en = δmn becomes
' (
en M̂ en = δmn . (6.7.16)
Equivalent of orthonormality for the case of unequal masses
All the subsequent arguments of section 6.5 follow just the same way, simply inserting
M̂ inside each inner product. Thus, assuming the eigenvectors have been scaled to
satisfy equation (6.7.15), the coefficients in the superposition (6.5.1),
+ N ,
(
iωn t
|x (t) = Re C e n e ,n
n=1
' (
are given by the equivalent of (6.5.8a), Re Cn = en |x0 ,
' (
Re Cn = en M̂ x0 , (6.7.17a)
1 ' (
and by the equivalent of (6.5.8b), Im Cn = − e |ẋ ,
ωn n 0
1 ' (
Im Cn = − e M̂ ẋ0 . (6.7.17b)
ωn n
One way of interpreting this insertion of M̂ inside each inner product is to consider
Hilbert space to be rescaled. We can define
⎛√ ⎞
m1 0 0
√
M̂1/2 = ⎝ 0 m2 0 ⎠ , so that M̂ = M̂1/2 M̂1/2 .
√
0 0 m3
Therefore, for example,
' ( ' 1/2 1/2 (
en M̂ x0 = en M̂ M̂ x0 .
Since M̂1( /2 is Hermitian, we can

think
( of the above as the inner product of the vector
M̂1/2 x0 with the vector M̂1/2 en . Thus, we could consider each of the axes of Hilbert
√
space, corresponding to the motion of x1 , x2 , and x3 , to be scaled by the factor m1 ,
√ √ √
m2 , or m3 respectively, so
( that the x1 axis becomes the (x1 m1 axis, and so on. In
this scaled Hilbert space e1 would be perpendicular to e2 .
Example: Returning
tothe example
shown in figure 6.7.2, the two (unnormalized)
( 1 ( − 2
eigenvectors e1 = and e2 = are not perpendicular to each other in an
1 1
' (
unscaled Hilbert space space, as shown in figure 6.7.3a. However, e1 M̂ e2 = 0; we
can show this graphically using Hilbert space axes that are scaled by the square root
of the mass, as shown in figure 6.7.3b. Also shown in figure 6.7.3b are versions of the
eigenvectors after the equivalent of normalization has been applied, for example:
(
( e
“normalized” version of e1 = ' 1 ( .
e1 M̂ e1
terms:
Determinant (6.1)
Eigenvalue, eigenvector, eigenvalue equation (6.2)
Characteristic equation (6.3)
Mode index (6.5)
Kronecker delta function (6.5)
Inverse of a matrix (6.6)
Identity matrix (6.6)
Hermitian matrix (6.6)
Mass matrix (6.7)
Degenerate modes (6.7)

One takes the adjoint of a product of Hermitian matrices (6.6)
Figure 6.7.3 a: When the masses for a coupled oscillator system are unequal, the eigenvectors
are not orthogonal in an unscaled Hilbert space. b: If the axes are scaled by the square root of
the mass, then the eigenvectors are orthogonal. Also shown are the “normalized” versions of
the eigenvectors, that is, the versions that have length 1 in the scaled Hilbert space.

The number of modes & the number of masses for a 1D system (6.4)
You should understand the difference between:

The eigenvalue & the angular frequency for a mode (6.2)
Mass index & mode index (6.5)

Matrix multiplication doesn’t commute. (6.1)
Matrix multiplication is associative. (6.6)
Scaling of Hilbert space for unequal masses (6.7)

Multiply matrices. (6.1)
Find determinants by hand for 2 × 2 and 3 × 3 matrices. (6.1)
Explain what an eigenvalue equation is. (6.2)

For a system with several objects and couplings, derive the set of DEQs that describes
the system, and go from that set of DEQs to the eigenvalue equation. (6.4)
Given the eigenvalue equation, check a proposed vector to see if it’s an eigenvector
and (if it is) to find the associated eigenvalue. (6.4)
Given the eigenvalue equation, find the eigenvalues and eigenvectors for systems of
two or three coupled oscillators. (6.3–6.4)
Normalize an eigenvector. (6.5)
Given the eigenvectors and eigenvalues for a system of coupled equal masses, and given
the initial positions and velocities, express the state of the system as a superposition
of normal modes. Then be able to find the position as a function of time for each
mass. (6.5)
Given the eigenvectors and eigenvalues for a system of coupled unequal masses,
and given the initial positions and velocities, express the state of the system as
a superposition of normal modes. Then, be able to find the position as a function of
time for each mass. (6.7)
6.8 Problems


Find
6.1 (a) the matrix that reflects a conventional two-dimensional vector r =
x
across the y-axis, as shown in figure 6.P.1. Show your reasoning clearly.
y
(b) Find the matrix that reflects a vector across the x-axis. (c) Find the matrix
that reflects a vector across the line y = x.
6.2 2D rotation matrix. Derive the matrix that rotates a conventional two-

vector counterclockwise through an angle ϕ . In other words,
dimensional
x
if r = , then find the matrix Â such that Âr is a vector with the same
y
length as r, but rotated counterclockwise

by ϕ . Show your reasoning clearly.
x /r
Hint: start by writing r = r , where r is the length of r.
y/r
Figure 6.P.1 The vector r is reflected across the

y-axis.
6.3 Inverse of a matrix. As we have discussed, a matrix Â can operate on a

vector, either in conventional space or Hilbert Space, to transform it into
another vector through a combination of rotation, scaling, and reflections.
In many cases, one can find an inverse matrix Â−1 , which undoes these
changes. In other words, if Âr = r′ , then Â−1 r′ = r. We could also write
this as Â−1 Âr = r. This means that the combination Â−1 Â has no net effect,
so that we could write Â−1 Â = Î, where Î is the “identity matrix,” a matrix
with
1’s on the diagonal and 0 elsewhere. For a two-dimensional space,
1 0
Î = . (a) It is not always possible to find an inverse of a matrix. If
0 1
the operation of Â destroys information about the original vector r, then it
will be impossible to reconstruct r from the vector Âr = r′ . Give an example,
for a two-dimensional system, of a matrix Â that has no inverse, and explain
how your matrix destroys information. (b) One can show (see any text on
linear algebra) that Â has an inverse if and only if det Â = 0. Given

this,

2 1
which of the following matrices does not have an inverse: (i)
1 −2

1 2 1 2
(ii) (iii) . (c) Show that the determinant for your matrix
1 2 −1 2

1 2 −2 1
from part a is zero. (d) Verify that the inverse of is 3 .
3 4 2 − 12
6.4 (You should complete problems 6.1 and 6.2 before doing this problem.)
(a) Create a single matrix that rotates a conventional two-dimensional vector
counterclockwise through an angle ϕ and then reflects it across
the y-axis.
x0
Verify that your matrix has the expected effect on the vector for the
0
◦
case ϕ = 45 . (b) Create a single matrix that reflects a conventional two-
dimensional vector across the y-axis, and then rotates it counterclockwise
through
an angle ϕ . Verify that your matrix has the expected effect on the
x0 ◦
vector for the case ϕ = 45 .
0
⎛ ⎞
1 0 0
6.5 3D rotation matrices. (a) The matrix R̂x = ⎝ 0 cos θ − sin θ ⎠ rotates
0 sin θ cos θ
a conventional three-dimensional vector r through an angle θ counter-
clockwise around the x-axis (as viewed looking from the positive x-axis
toward the origin). Demonstrate that this matrix works as expected on
⎛ ⎞
x0
◦
the vector r = y0 ⎠ for the case θ = 45 . (Recall that î × ĵ = k̂.)
⎝
0
⎛ ⎞
cos θ 0 sin θ
(b) The matrix R̂y = ⎝ 0 1 0 ⎠ rotates a conventional three-
− sin θ 0 cos θ
dimensional vector r through an angle θ counterclockwise around the y-axis
(as viewed looking from the positive y-axis toward the origin). Create a single
◦
matrix that rotates a vector by 90 counterclockwise about the y-axis and
◦
then 90 counterclockwise about the x-axis. Show your reasoning clearly,
⎛ ⎞
0
and then show that your matrix has the expected effect on the vector ⎝ 0 ⎠.
z0
◦
(c) Create a single matrix that rotates a vector by 90 counterclockwise
◦
about the x-axis and then 90 counterclockwise about the y-axis. Show your
reasoning clearly, and then show that your matrix has the expected effect on
⎛ ⎞
0
the vector ⎝ 0 ⎠.
z0
6.6 For a particular coupled oscillator system, of the form shown in figure 6.2.1,
m1 = 0.200 kg, m2 = 0.400 kg, and k = 3.00 N/m. If m2 is fixed and m1 is
displaced from equilibrium and then released, one observes that m1 oscillates
with a period of 1.068 s. If instead m1 is fixed and m2 is displaced from
equilibrium and then released, one observes that m2 oscillates with a period
of 0.919 s. What are the periods for the normal modes of this system (with
neither mass held fixed)?
6.7 The coupled oscillator system shown in figure 6.P.2 has m1 = 0.300 kg,
m2 = 0.500 kg, and kR = 4.00 N/m. The masses slide on a frictionless
surface. If m2 is fixed and m1 is displaced from equilibrium and then released,
one observes that m1 oscillates with a period of 1.257 s. If instead m1 is fixed
and m2 is displaced from equilibrium and then released, one observes that
m2 oscillates with a period of 2.221 s. What are the periods for the normal
modes of this system (with neither mass held fixed)?
6.8 (a) Derive the set of coupled differential equations that describes the system
of coupled pendula in figure 6.P.3. The position of the left mass is
x1 , and is measured relative to its equilibrium position. Similarly, the
position of the right mass is x2 , and is measured relative to its equilibrium
position.
(b) Derive the eigenvalue equation for this system. Show all your steps
explicitly.
Hint: Answer is

ωA2 −ω02 X1 2 X1
=ω ,
−ω02 ωA2 X2 X2
g 2k k
where ωA2 ≡ + and ω02 ≡
ℓ m m
Figure 6.P.2 Two masses coupled by springs slide

on a frictionless surface.
Figure 6.P.3 A set of two coupled pendula.
(c) Show that the frequencies of the normal modes for this system are given
by
⎧
g k
⎨ +
⎪
ω2 = gℓ 3k m
⎪
⎩ +
ℓ m
(d) Show that the corresponding normalized eigenvectors are:

g k 1 1
for ω2 = + : √ and
ℓ m 2 1

g 3k 1 1
for ω2 = + :√
ℓ m 2 −1
(e) Describe the motion of each normal mode using words and pictures.
(f) Using our usual definition of the inner product for discrete systems, show
that the normal modes of the above system are orthogonal to each other.
Note: the math for this is very simple.
6.9 Consider the double-pendulum system shown in figure 6.P.4. As usual, we’ll
only consider small displacements. Recall that in this limit, the combined
forces of gravity and string tension for a generic pendulum give a spring-like
mg
restoring force for lateral displacements, with effective spring constant .
L
mg
Therefore, the restoring force for a generic pendulum is Fpendulum = − x,
L
where x is the lateral displacement of the pendulum bob relative to the lateral
Figure 6.P.4 A double pendulum system.

position of the support point. (a) Derive the eigenvalue equation (in matrix
form) for this system. (b) Now write the characteristic equation, and use it
to find the normal mode frequencies. (c) Finally, find the eigenvectors.
6.10 The CO2 molecule. Carbon dioxide is a linear molecule with a carbon atom
at the center which is double-bonded to two oxygen atoms, as suggested
in figure 6.P.5. The two springs are identical, and have spring constant
k = 3628 N/m. We’ll assume that the carbon atom is a 12 C isotope (the
most common kind, having six protons and six neutrons), meaning that it
has a mass of exactly 12 u = 1.993 × 10−26 kg, and that the oxygen atoms
are both 16 O, meaning that the mass of each is 2.66 × 10−26 kg. Find the
eigenvectors and normal mode frequencies for this system. For each normal
mode, describe the oscillation in words as well as by giving the eigenvector.
(You are encouraged to use a symbolic algebra program such as Mathematica
or Maple for this problem, though it is not necessary.)
6.11 (You should do problem 6.6 before this problem.) Find the normalized
eigenvectors for the system described in problem 6.6.
6.12 (You should do problem 6.7 before this problem.) Find the normalized
eigenvectors for the system described in problem 6.7.
6.13 Consider the system shown in figure 6.P.6. (a) Find the matrix Â that appears
in the eigenvalue equation Â |e = ω2 |e, where |e is an eigenvector. (b) For
the special case m1 = m2 = 1 kg, m3 = m4 = 2 kg, ℓα = 1 m, ℓβ = 2 m,
⎛ ⎞
0.653
( ⎜ −0.271 ⎟
k13 = 2 N/m, k24 = 4 N/m, and g = 4 m/s2 , show that ea = ⎜ ⎝ 0.653 ⎠ is
⎟
−0.271
an eigenvector for this system, without using a symbolic algebra program.
Figure 6.P.5 Model for a CO2 molecule.
Figure 6.P.6 A system of four coupled pendula.

(c) Determine
( the angular frequency of oscillation ωa for the mode described
by ea . (d) Make approximate sketches of the positions
( of the masses for
this mode at t = 0 and at t = π/ωa . (e) Show that ea is normalized, without
using a symbolic algebra program.
6.14 For the system shown in figure 6.P.6, consider the special case m1 = 1 kg,
m2 = 2 kg, m3 = 3 kg, m4 = 4 kg, ℓα = 1 m, ℓβ = 2 m, k13 = 3 N/m,
k24 = 6 N/m, and g = 10 m/s2 . Use a symbolic algebra program to find the
normalized eigenvectors and corresponding angular frequencies. Make clear
which eigenvector goes with which angular frequency. For each eigenvector,
make a rough sketch of the positions of the masses at t = 0 and at t = π/ω,
where ω is the angular frequency for that mode.
6.15 Do not use a symbolic algebra program or calculator for this problem. The
asymmetric coupled pendulum system shown in fig. 6.2.1 has l1 = 1.00 m,
l2 = 0.500 m, k = 1.50 N/m, m1 = m2 = 0.500 kg. The masses are released
from rest at t = 0, with initial positions x1 = −0.025 m and x2 = 0.045 m.
What are the positions of the two masses as a function of time?

x1
6.16 Let’s define an arbitrary bra g| = g1 g2 , an arbitrary ket |x = ,
x2

a b
and an arbitrary matrix Â = . Show explicitly that g| Â |x is
c d

unambigous, that is, show that g| Â |x is the same as g| Â |x .
6.17 Consider the two-pendulum system shown in figure 6.7.2, in which m1
= 1 kg, m2 = 2 kg, ℓ = 1 m, k = 1 N/m, and g = 10 m/s2 . Verify

( 1
that the (unnormalized) eigenvectors for this system are e1 =
1

( −2
and e2 = .
1
7 String Theory
Midnight. No waves,
no wind, the empty boat
is flooded with moonlight.
—Dogen (1200–1253)
Each string I touch

Sends out its love in waves
And now the boat is full.
—Marian McKenzie
In some of the most complicated theories of modern physics, elementary particles are
represented in terms of the normal modes of extremely short (10−35 m) strings. Usually,
these theories are hyperdimensional, that is, they use more than the normal number
(four) of spacetime dimensions; several string theories have ten dimensions or more.
Often, string theorists investigate a slice through this multidimensional space. A slice
through three-dimensional space has two dimensions, and is called a “membrane.”
A slice through a five-dimensional space might have four dimensions, and would be
called a “4-brane.” In general, a slice with p dimensions is called, get this, a “p-brane.”
This proves that even the most advanced theoretical physicists have a sense of humor!
So far, it has been very difficult to test string theories experimentally.
In this chapter, we will study ordinary macroscopic strings. This will lead us
to an understanding of the normal modes of continuous systems, and of how the
normal modes are modified when a system is not truly continuous (e.g., when it
consists of atoms). Surprisingly, our study of strings will also lead us to a fundamental
understanding of Fourier analysis, a completely mathematical idea in which any
arbitrary function can be constructed by adding together sine waves. Finally, as in
previous chapters, we will point out the deep connections between these macroscopic
oscillating systems and quantum mechanics.
7.1 The beaded string
We begin by considering a massless string of length L stretched between two walls,

so that it is under tension T . The string has N beads (each of mass m) spaced at even
216
Chapter 7 ■ String Theory 217
Figure 7.1.1 a: A massless string under tension T with small beads equally spaced along its
length. b: Geometry needed for calculating the force on bead j.
intervals a, as shown in figure 7.1.1a. We assume gravity is unimportant. If the string

is plucked, it vibrates. Each bead feels a force from the string coupling it to the bead on
its left, and a second force from the string coupling it to the bead on its right. Therefore,
this is a system of N coupled oscillators. As in chapter 6, we expect that such a system
will display normal modes, and we will use a similar procedure to find out what they
are. For mathematical convenience, we introduce two fictitious beads at the ends of
the string, as shown.
We wish to find the ways in which this system can move. We follow our multi-step
procedure.
Step 1: Write FTOT = mÿ for each mass. We will take the approximation of small
displacements from equilibrium. In this limit, when the system oscillates the change
in the magnitude of the string tension T is negligible. So, from figure 7.1.1b, we see
that the y-component of the force acting on mass j is
FTOT, y = −T sin θL − T sin θR , (7.1.1)
where
yj − yj−1
sin θL = 2
a2 + yj − yj−1

For small displacements from equilibrium, yj − yj−1 ≪ a, so we can write
yj − yj−1 yj − yj+1
sin θL ∼
= and similarly sin θR ∼
=
a a
Substituting these into equation (7.1.1) gives
T
FTOT, y = − yj − yj−1 + yj − yj+1 = mÿj
a
T
⇔− −yj−1 + 2yj − yj+1 = ÿj (7.1.2)
ma
Let us define

2T
ωA ≡ . (7.1.3)
ma
This is the angular frequency at which one mass would oscillate if its neighbors were
held fixed, as you can see by setting yj−1 = yj+1 = 0 in equation (7.1.2). The simplest
complex equation whose real part is the above would be
ωA2
− −zj−1 + 2zj − zj+1 = z̈j , (7.1.4)
2
where yj = Re zj . Since j can range from 1 to N, this represents a system of N
coupled DEQs.
Step 2: Use physical and mathematical intuition to guess a solution. Based on our
experience with the system of two coupled oscillators, we hope that this system may
display normal modes, in which each bead moves with the same frequency. The most
general possible such guess would be
?
zj = Yj eiωn t . (7.1.5)
Where ωn is the angular frequency of the normal mode, n is the mode index, and Yj is
the amplitude of bead j. (As before, we have chosen the definition of the moment when
t = 0 so that the phase of the normal mode ϕ = 0.) We expect for a system of N beads
that there will be N normal modes, each of which has its own characteristic frequency.
Step 3: Plug the guess into the system of DEQs. Plugging our normal modes
guess (7.1.5) into (7.1.4) gives
ωA2
− −Y( j−1) eiωn t + 2Yj eiωn t − Y( j+1) eiωn t = −ωn2 Yj eiωn t
2
ω2
⇒ A −Y( j−1) + 2Yj − Y( j+1) = ωn2 Yj (7.1.6)
2
This is a system of N coupled linear equations. In principle, we could solve it
using the methods of section 6.3, that is, we could write it as a matrix eigenvalue
equation, rearrange it to a characteristic equation, set the determinant equal to zero to
find the eigenvalues, and plug these back into the characteristic equation to find the
eigenvectors.
7.2 Standing wave guess: Boundary conditions quantize the allowed

frequencies
Obviously, this procedure would be tedious for a system of more than a few beads.
However, we can get to the answer much more quickly by using additional physical
insight to guess what the eigenvectors are, that is, to guess what the initial positions of
the beads are for a normal mode. We have all played with ropes, telephone cords, and
so on, and have observed “standing waves,” in which the rope oscillates between the
positions shown by the solid lines and the dashed lines in figure 7.2.1. Each part of the
rope oscillates up and down with the same frequency, so these represent the normal
modes of the system. It is reasonable to expect that the normal modes of the beaded
string would be similar.
In these standing waves, each part of the string oscillates with a different amplitude.
From figure 7.2.1c we see that the amplitude of oscillation for a point at position x is
given by,

x
amplitude = An sin 2π ,
λn
where λn is the wavelength of the standing wave and An is the amplitude of the standing
wave.
Therefore, we will make the following guess for the positions of the beads on the
beaded string:

? xj
amplitude = Yj = An sin 2π , (7.2.1)
λn
where xj is the x-position of bead j. This is our “standing wave guess” for the form of
the eigenvectors; each different value of j in the above equation (corresponding to each
different bead) gives a different line of the column matrix representing the eigenvector.
Figure 7.2.1 a: In the lowest-frequency standing wave (n = 1), the rope oscillates between the
shape shown by the solid curve and that shown by the dashed curve. b: Similar curves for the
n = 2 standing wave. c: The n = 3 standing wave.
Because the string must go to y = 0 at the boundaries, we must fit an integer

number of half-wavelengths between the walls, that is,
λn
n =L⇔
2
2L
λn = . (7.2.2)
n
Since the wavelength cannot take on a continuous range of values, but only those
allowed by the above equation, we see that the imposition of boundary conditions has
“quantized” the allowed values of λ. To save writing (and to follow well-established
convention), we define the “wavenumber”
2π π
kn ≡ =n . (7.2.3)
λn L
Thus, the wavenumber is 2π divided by the periodicity in space (λ) , just as the angular
frequency ω is 2π divided by the periodicity in time (T ). Using this, equation (7.2.1)
can be written
?

Yj = An sin kn xj (7.2.4)
(Again, n is the mode index and j is the mass index.)

Figure 7.2.1 shows that the n = 2 mode has one “node,” that is, one spot in the
middle where the rope doesn’t move. The n = 3 mode has two nodes, and so on. Quite
generally:
Number of nodes + 1 = n.
To see if our standing wave guess for the eigenvectors works, we plug equation (7.2.4)
into (7.1.6). (Recall that equation (7.1.6) was obtained by combining FTOT = mÿ with
our normal mode guess.) After canceling the common factor of An , this gives
ωA2 ω2 ?
− sin kn xj−1 + ωA2 sin kn xj − A sin kn xj+1 = ωn2 sin kn xj .
2 2
Now, xj−1 = xj − a, and xj+1 = xj + a, so this becomes
ωA2 ?
− sin kn xj − kn a + sin kn xj + kn a + ωA2 sin kn xj = ωn2 sin kn xj .
2
Next we use the standard formula sin (A + B) = sin A cos B + cos A sin B, and
recall that cos (−θ ) = cos θ , while sin (−θ ) = − sin θ , to obtain
ωA2
− sin kn xj cos kn a − cos kn xj sin kn a + sin kn xj cos kn a + cos kn xj sin kn a
2
?
+ ωA2 sin kn xj = ωn2 sin kn xj
?
⇔ −ωA2 sin kn xj cos kn a + ωA2 sin kn xj = ωn2 sin kn xj
?
⇔ −ωA2 cos kn a + ωA2 = ωn2
So, our standing wave guess does work, but only if1

ωn2 = ωA2 1 − cos kn a . (7.2.5)
Using the trigonometric identity2

θ
1 − cos θ = 2 sin2 ,
2
Equation (7.2.5) becomes
kn a πa
ωn2 = 2ωA2 sin2 = 2ωA2 sin2 n . (7.2.6)
2 2L
Recall that the imposition of the boundary conditions required that λn can only take on
discrete values, as given by equation (7.2.2), or equivalently that the wavenumber kn
can only take on discrete values as given by equation (7.2.3) . Now, we can see that
the boundary conditions also force ωn to take on discrete values. This is a very general
and very important observation:
Imposition of boundary conditions quantizes the normal mode frequencies.
Exactly the same thing happens in quantum mechanics: imposition of boundary

conditions (i.e., restricting the electron to a finite region of space, such as the region
near an atomic nucleus) quantizes the allowed frequencies of the wavefunction ,
and since E = h̄ω, this is equivalent to quantizing the allowed energies. Thus, we see
that one of the most famous results of quantum mechanics, the idea that an electron
in an atom can only take on certain allowed levels of energy, is directly the result of
attributing a wave nature to the electron, and restricting this wave to a small region
in space.
We can see from equation (7.2.6) that the smaller the region is (i.e., the smaller L
is), the greater the interval between allowed frequencies. Similarly in quantum
mechanics, the more tightly we confine the electron, the larger the interval between
allowed energies. This effect is important in determining the energy levels of atoms.
1. This is an example of a “dispersion relation,” that is, a relation between ω and k. It’s likely that
all the waves you have studied in previous courses, including standing waves in organ pipes or on
violin strings, had linear dispersion relations, that is, relations in which ω is directly proportional
λ 2π /k ω
to k. For example, for electromagnetic waves in vacuum, we have c = = = ⇔
T 2π /ω k
ω = ck. Clearly, the dispersion relation for standing waves on a beaded string, equation (7.2.5)
is nonlinear. We’ll discuss dispersion relations in more depth in chapter 9, and explain why they
are called “dispersion relations.”
2. It is not hard to prove this from more fundamental identities:

θ θ θ θ θ θ θ
cos θ = cos + = cos2 − sin2 ⇒ 1 − cos θ = 1 − cos2 + sin2 = 2 sin2
2 2 2 2 2! 2 2
θ
sin2
2
It also dictates the behavior of electrons that are not tightly attached to any single
atom, such as electrons in metals. Such electrons can roam throughout the metal.
Using electron beam lithography, experimentalists can make metal samples that are
very small, down to only about 100 atoms on a side! As the sample is made smaller,
they can observe the transition from having a “continuum” of energy levels (because
the energy interval between levels is too small to observe) to having clearly quantized
levels. Such “mesoscopic” samples (between the truly microscopic atomic scale and
the macroscopic scale) display many exciting properties, and may form the basis of
new types of extremely small transistors.
Let us collect all our results. Plugging the standing wave eigenvectors (7.2.4)
into our “normal mode guess” equation (7.1.5): zj = Yj eiωn t gives the complete
description of the normal modes:

zj = An sin kn xj eiωn t .
The actual displacement of bead j would be given by the real part of this, so that

yj = An sin kn xj cos ωn t
! !
spatial time
variation variation
(7.2.7)
2L π
xj = ja λn = n = 1, 2, 3, · · · kn = n
n L
√ πa 2T
ωn = 2 ωA sin n ωA ≡
2L ma
System in pure normal mode n.
7.3 The highest possible frequency; connection to waves in a

crystalline solid
We know that the number of normal modes should match the number of objects3
in the system (for a system such as this in which the objects can only move in one
dimension), that is, that the maximum value of the mode index n should be N. However,
from the above, there is no apparent limit on n. Let’s look more carefully. From
equation (7.2.7), we have that the frequencies of the normal modes are given by
√ π a
ωn = 2 ωA sin n . (7.3.1)
2L
√
We see right away that the normal mode frequencies can never exceed 2 ωA, so that
we already have a hint that the number of normal modes might actually be limited.
3. More generally, the number of normal modes should match the number of degrees of freedom.
Figure 7.3.1 A string with three beads.
Figure 7.3.2 ωn versus n for N = 3.
From figure 7.3.1 we see that
L = (N + 1) a. (7.3.2)

√ π n
ωn = 2 ωA sin . (7.3.3)
2 N +1
If we plot ωn as a function of n, it clearly reaches a maximum at n = N + 1; this is

shown in figure 7.3.2 for the case N = 3. We can already see graphically4 that ω5 = ω3 ,
or more generally that ωN +2 = ωN . This means that at least the time dependence of
the n = N + 2 “mode” is the same as that of the n = N mode, suggesting that the
n = N + 2 is not a new mode at all, but rather is really the same as the n = N mode.
(Similarly, we can see that the time dependence of the n = N + 3 “mode” is the same
as that of the n = N − 1 mode, etc.)
√
4. This is also easy to show symbolically: from equation (7.3.3) we have ωN +2 = 2 ωA

π N +2 √ π · 2 (N + 1) π ·N √ π ·N
sin = 2 ωA sin − = 2 ωA sin π − .
2 N +1 2 (N + 1) 2 (N + 1)
2 (N + 1)
√ π N
Now, sin (π − x) = − sin (−x) = sin x. So, ωN +2 = 2 ωA sin = ωN .
2 N +1
Does the spatial dependence of the n = N + 2 mode bear out this hunch? From
equation (7.2.7), we have that for mode n = N + 2
yj = AN +2 sin kN +2 xj cos ωN +2 t .
Where AN +2 is the overall amplitude of the mode. The spatial dependence of this is

N +2 2(N + 1) − N
yj (t = 0) = AN +2 sinkN +2 xj = AN +2 sin π x = AN +2 sin 2π xj
L j 2L
2(N + 1)xj N xj
" #9 " #9
2(N + 1)ja N xj
= AN +2 sin 2π − = AN +2 sin 2π −
2(N + 1)a 2L 2(N + 1)a 2L
9 9
N N
= AN +2 sin 2π j −π xj = AN +2 sin −π xj
L L
where the last step works because j is an integer. Finally, since sin (−x) = − sin x and
N
kN = π , we have
L

yj (t = 0) = −AN +2 sin kN xj .
This is the same as the spatial dependence of the n = N mode, differing only by a
minus sign that can be absorbed into the amplitude.
This is shown graphically in figure 7.3.3; note that the function An sin kn xj only
takes on physical reality at the positions xj of the beads; we can draw the continuous
mathematical function An sin kn x as shown by the dashed lines, but this does not
represent the shape of the string. Since the string is massless, it simply forms straight
line segments between the beads, as shown by the solid line. The same bead positions
Figure 7.3.3 As shown here for the case N = 3, the mode n = N + 2 gives the same bead
locations (i.e., has the same spatial dependence) as the mode n = N. The mathematical
functions A3 sin k3 x (black dashed line) and A5 sin k5 x (gray dashed line) only take on
physical reality at the positions of the beads. Because the string is massless, it follows straight
lines (solid lines) between the beads.
can be characterized using the black dashed curve (with kN ) or the gray dashed curve
(with kN +2 ).
Thus, since the spatial and time dependence of the n = N + 2 are the same as for
n = N, we see that the n = N + 2 mode is not actually a new, independent mode, but
is really the same as the n = N mode. We could make similar arguments to show that
the spatial dependence of the n = N + 3 mode is the same as the n = N − 1 mode,
and so on. So, all the modes with n = N + 2 or greater simply reproduce the modes
with n = N or smaller.
But what about the n = N + 1 “mode”? We see from figure 7.3.2 that the frequency
for this “mode” doesn’t match the frequency of any other mode, and yet we know that
there should be only N modes. The solution to this dilemma comes from considering
the spatial dependence. For n = N + 1, we have
π π π
kN +1 = (N + 1) = (N + 1) = .
L (N + 1) a a
2π
This means that λN +1 = = 2a. Therefore, the spatial dependence for this mode
kN +1
is as shown in the top part of figure 7.3.4. Each bead is at a node, and so the beads
never move. Therefore, this is not really a “mode,” but rather a very complicated way
of saying that the beads are allowed to remain motionless!
Here’s another way of understanding why there is a highest possible mode, that
is, a shortest possible wavelength. Note that, in the highest actual mode (n = N), the
positions of the masses alternate up–down–up, and so on, as shown in figures 7.3.3
and 7.3.4 bottom. Thus, this is the “most wiggliness” that can be represented by the
system, that is, the shortest wavelength.
Figure 7.3.4 Top: For n = N + 1, each bead is at a node. Bottom: For n = N, there is an
alternating up-down pattern, so that the string is displaying the most “wiggliness” possible.
These ideas are illustrated further in the applet for this section on the website for
this text.
A crystal is a regular array of atoms. Examples include all metals (usually made of
many tiny “crystallites” stuck together) and the silicon wafers used to make integrated
circuits. Each atom in a crystal is in stable equilibrium, and therefore (using the
arguments of chapter 1) can be modeled as a harmonic oscillator. So, one line of atoms
in a crystal can be modeled as a beaded string. Since there is a well-defined highest
frequency of normal mode for a beaded string, there is a well-defined highest frequency
of vibration for a crystal. Because the masses are small, and the restoring forces are
relatively strong, this frequency is fairly high – on the order of about 1013 Hz. This
highest frequency can be observed experimentally – both using spectroscopic methods
and using heat capacity measurements.
7.4 Normal mode analysis for the beaded string
Our analysis in section 6.5 was fully general, and applies perfectly well to the beaded
string. This means that any behavior of the system can be represented as a superposition
of the normal modes. Here’s the way we stated that back in section 6.5:
+ N ,
(
iωn t
(6.5.1): |x (t) = Re Cn e e .
n
n=1
For the beaded string, the beads move in the y-direction, so it makes more sense to
write this as
⎛ ⎞
y1 (t) + N ,
⎜ y (t) ⎟ (
iωn t
|y (t) ≡ ⎝ 2 ⎠ = Re Cn e en , (7.4.1)
.. n =1
.
Normal mode expansion for beaded string
(
where y1 (t) is the position of bead 1, and en is the normalized eigenvector for mode n.
We already know the form of the eigenvectors, though they aren’t yet normalized. In
section 7.2, we found that the entries in each line of the eigenvector are

(7.2.4): Yj = An sin kn xj ,
so that the eigenvectors are given by
⎛ ⎞
sin kn x1
sin kn x2 ⎟
(
e = A ⎜ ⎠.
n n⎝
..
.
To find the normalized eigenvectors, we set

' (
en en = 1
⎛ ⎞
sin kn x1
2 1
⇒ A∗n sin kn x1 sin kn x2 · · · An ⎝sin kn x2 ⎠ = 1 ⇒ An =
⎜ ⎟
.
.. N
: 2
. sin kn xj
j =1
You can show in problem 7.6 that

N
N +1
sin2 kn xj = . (7.4.2)
2
j=1
For convenience, we may as well choose An to be real, so that

2
An =
N +1
and
⎛ ⎞
sin kn x1
( 2 ⎜ sin k x ⎟
e =
n ⎝ n 2 ⎠ (7.4.3)
N +1 ..
.
Normalized eigenvectors for beaded string.
Self-test: Verify that the above is indeed correctly normalized for the case N = 3 and
n = 2.
Because the analysis of section 6.5 was fully general, we can find the coefficients Cn
in the normal mode expansion (7.4.1) by using the formulas
' ( 1 ' (
(6.5.8a): Re Cn = en | y0 and (6.5.8b): Im Cn = − en | ẏ0 .
ωn
(Here, we have replaced the x’s by y’s to fit the current situation.)
7.5 Longitudinal oscillations
As discussed in section 7.4, the beaded string is a good model for vibrations in a
solid. However, the atoms in a solid can vibrate in two distinct ways. So far, we have
discussed transverse vibrations, in which the motion is the y-direction, perpendicular
to the direction in which the wavelength is defined (the x-direction). For a solid, this
would correspond to planes of atoms sliding laterally without changing the distance
between planes, as shown in the left part of figure 7.5.1. However, the planes of atoms
can also vibrate along the x-direction, so that the distance between planes oscillates,
as shown in the right part of figure 7.5.1. The one-dimensional model for this type of
oscillation is a chain of beads connected by springs. We measure the x-position of each
Figure 7.5.1 Transverse and

longitudinal oscillations in a
solid.
Figure 7.5.2 Quantities defined for longitudinal oscillations.
bead relative to its equilibrium position, however to avoid confusion with the variable
x, we describe this displacement using the symbol δ , as shown in figure 7.5.2.

Your turn: Explain why the total force on bead j is FTOT = −k δj − δj −1 − k δj − δj +1 .
Since FTOT = mδ̈j , this gives
k
− −δj−1 + 2δj − δj+1 = δ̈j . (7.5.1)
m
This equation is isomorphic with equation (7.1.2), the differential equation describing
transverse oscillations on the beaded string:
T
− −yj−1 + 2yj − yj+1 = ÿj .
ma
Table 7.5.1. Isomorphism between transverse and longitudinal waves
Quantity Transverse Longitudinal
Displacement yj δj

2T 2k
Angular frequency of vibration ωA = am ωA = m
with neighbors fixed
Thus, the solutions are exactly the same as for transverse vibrations, with the
substitutions indicated in table 7.5.1 above, giving

δj = An sin kn xj cos ωn t
! !
spatial time
variation variation
(7.5.2)
2L π
xj = ja λn = n = 1, 2, 3, · · · kn = n
n L
√ πa 2k
ωn = 2 ωA sin n ωA ≡
2L m
System in pure normal mode n
Figure 7.5.3 shows the vibrations of the mode n = 1. At t = 0, all the beads are
displaced to the right, resulting in a bunching near the right side. A half period later,
the beads are all displaced to the left, resulting in bunching near the left side.
Let’s apply this model to a thin solid rod of cross-sectional area A, which is wedged
between two massive walls, with one end of the rod pushing against each wall. We
treat each plane of atoms perpendicular to the axis of the rod as a bead, with mass
m = ρ Aa, where a is the spacing between atomic planes. From equation (2.3.2), the
A
spring constant of a solid with cross-sectional area A and length a is k = E , where
a
Figure 7.5.3 Displacements at t = 0 and half a period later for the mode n = 1 for longitudinal
oscillations.

2k 2EA 1 2E
E is the Young’s modulus. Thus, ωA = = = . If we consider, for
m ρ Aa2 a ρ
example, the low n modes of a rod with at least several mm of length, we have na ≪ L,
so that
√ πa √ πa
ωn = 2 ωA sin n ∼
= 2 ωAn ,
2L 2L
where we have used the approximation (1.6.4): sin θ = ∼ θ , valid for small θ . Therefore,

√ 1 2E π a E π
ωn ∼= 2 n =n . (7.5.3)
a ρ 2L ρ L
7.6 The continuous string
It is easy to extend the results of section 7.4 to a continuous string with finite mass.
We simply imagine shrinking the distance a between the beads (by adding more beads
onto the string), while also making the beads less massive, so as to keep the mass per
unit length constant. Then, in the limit a → 0 (and N → ∞), we have a continuous,
massive string.
We can readily find the frequencies of the normal modes in this limit. From
equation (7.2.7), we have for the beaded string
πa

T T π n
ωn = 2 sin n =2 sin .
ma 2L ma 2 N +1
In the limit N → ∞, the argument of the sin becomes infinitesimal, so we can use
sin x −−→ x .
x →0
Therefore, as we allow N → ∞, we obtain

T πa √ a π
ωn = 2 n = T n . (7.6.1)
ma 2L m L
We define the mass per unit length:
m
μ≡ ,
a
so that equation (7.6.1) becomes

π T
ωn = n = n ω1 , (7.6.2)
L μ
Normal mode frequencies for continuous string

π T
where ω1 = . Note that, unlike for the beaded string, these frequencies are all
L μ
multiples of the fundamental frequency ω1 , and that they increase linearly with the
mode index n. One way to understand this is by considering the graph of ωn for the
beaded string, as shown in figure 7.6.1. The peak of this graph is at n = N + 1, so
Figure 7.6.1 Angular frequency vs. mode

index for a beaded string.
that in the limit N → ∞ we never approach the peak, and we are always in the linear
regime close to the origin.
As we take the limit N → ∞, there is nothing fundamental changing in the physics
of the situation. Therefore, the eigenvectors have the same form as for the beaded string.
In section 7.2, we found that the entries in each line of the eigenvector are

(7.2.4): Yj = An sin kn xj ,
so that the eigenvectors are given by

⎛ ⎞ ⎛ ⎞
Y sin kn x1
( ⎜ 1 ⎟
e = ⎝ Y2 ⎠ = A ⎜ sin kn x2 ⎟
⎠. (7.6.3)
n n⎝
.. ..
. .
For the limit N → ∞, this notation becomes awkward, since there are an infinite
number of lines in the column vector. So, instead we simply write the eigenvectors as
yn (x) = An sin kn x . (7.6.4)
Note that we no longer write these explicitly as vectors; instead we simply write
them as functions. Therefore, we will often call them “eigenfunctions” instead of
“eigenvectors.” However, it is still appropriate to think of them as vectors in an
infinite-dimensional Hilbert space. (The Hilbert space has infinite dimensions, because
there are an infinite number of infinitesimally small pieces making up the continuous
string.)
7.7 Normal mode analysis for continuous systems
To handle continuous systems, we’ll need to extend the definition of the inner product.
Let’s recall the general definition of the inner product for systems with
( a finite number
of objects, such as the beaded string. For example, let’s say that yA represents one
state of the beaded string (perhaps a normal mode, or perhaps a mixture of normal
(
modes), and that yB represents some different state. We can write
⎛ ⎞ ⎛ ⎞
yA1 yB1
( ⎜ ( ⎜
y = ⎝ yA2 ⎟
⎠ and yB = ⎝ yB2 ⎠ ,
⎟
A
.. ..
. .
(
where, for the state yA , yA1 is the position of bead 1, yA2 is the position of bead 2,
and so on.
The inner product is defined as
⎛ ⎞
y
' ( ∗ ⎜ B1 ⎟ N
∗
yA | yB ≡ yA1 ∗
yA2 · · · ⎝yB2 ⎠ = yAj yBj . (7.7.1)
.. j = 1
.
For a continuous system, there would be an infinite number of objects, so this would
become an infinite sum, which we can write using an integral. Therefore, we define
the inner product of two continuous functions yA (x) and yB (x) to be
' ( $L
yA (x) | yB (x) ≡ yA∗ (x) yB (x) dx, (7.7.2)
0
where the range x = 0 to x = L is the size of the continuous system. For a system
with a differently defined size, the limits of integration would be changed, so that the
integral is still over the entire system.
Although equation (7.7.2) is the well-established convention for the definition
of the inner product, it is not quite equivalent to the definition (7.7.1) for discrete
systems, since each term in the sum which the integral represents is multiplied by the
extra factor dx. It should be clear that, even with this extra factor, the eigenfunctions
are still orthogonal (i.e., the inner product of two different eigenfunctions is still zero),
since zero times dx is still zero.
5. From equation (7.6.3), we have yn (x) = An sin kn x and ym (x) = Am sin km x. According to
' ( $L
equation (7.7.2), their inner product is yn (x) ym (x) = A∗n sin kn x Am sin km x dx = A∗n Am ×
0
+ ,L
sin kn − km x sin kn + km x π π
− . Recall that kn = n , so that kn − km = (n − m)
2 kn − km 2 kn + km L L
0 +
π ' ( sin [(n − m) π ]
and kn + km = (n + m) . Therefore, yn (x) ym (x) = A∗n Am −
L 2 kn − km
,
sin [(n + m) π ]
− 0 + 0 . Since n and m are integers, this equals zero, Q.E.D.
2 kn + km
Self-test (answer below5 ): Verify that the eigenfunctions yn (x) and ym (x) for the
continuous string are indeed orthogonal if n = m. You may need this integral, given
in the form you would find it in an integral table:

sin p − q x sin p + q x
sin px sin qx dx = − p = ±q
2 p−q 2 p+q
(Although this is the common way to write this, it is somewhat ambiguous; it should
be read

sin p − q x sin p + q x
2 p−q 2 p+q
Your turn: Show that the normalized eigenfunctions for the continuous string are
given by

2
yn (x) = sin kn x (7.7.3)
L
Ignore the following paragraph if it confuses you, but you might find it
interesting. Compare equation (7.7.3) with the normalized eigenvectors for the beaded
string:
⎛ ⎞
sin kn x1
( 2 ⎜ sin k x ⎟
(7.4.3): en = ⎝ n 2 ⎠
N +1 ..
.
Normalized eigenvectors for beaded string
By recalling that L = (N + 1) a, we see that the result (7.7.3) for the continuous string
is the same as that for the beaded string, except that the continuous string result is
√
divided by a; for the continuous string, the “spacing” between
√ “beads” is a = dx, so
that the normalized eigenfunctions have been divided by √dx . When we take the inner
product of two such functions, we effectively divide by dx twice; this compensates
for the “extra” factor of dx in the definition of the inner product that was discussed at
the top of the previous page.
You should recall that the normalized eigenvectors for discrete systems were
dimensionless. From equation (7.7.3), we can see that the normalized eigenfunctions
for the continuous string have units of meters −1/2 . This might seem strange, but it
is indeed correct, and necessary to make the normal mode analysis formulas work
out correctly. It might make you feel better to know that the wave function for one-
dimensional quantum mechanics (analogous to our one-dimensional continuous string)
also has units of meters −1/2 .
As for the beaded string (and the coupled pendula), any state of the system can be
written as a superposition of the normal modes, that is,
"∞ #
:
y (x , t) = Re Cn eiωn t yn (x) . (7.7.4)
n=1
Normal mode expansion for continuous string
(If the system is truly continuous, then there are an infinite number of normal modes.)
Because the analysis of section 6.5 was fully general, we can find the coefficients Cn
in the normal mode expansion (7.7.4) by using the formulas
(6.5.8a): Re Cn = yn (x)|y0 (x) and

1
(6.5.8b): Im Cn = − y (x)|ẏ0 (x) .
ωn n
(Here, we have replaced the x’s by y’s, and vectors by functions to fit the current
situation.) Again, for the important special case ẏ0 (x) = 0, the process of normal
mode analysis is exactly analogous to the process of taking projections of ordinary
vectors onto the x- and y-axes (although now there are an infinite number of axes onto
which we must take projections!).
7.8 k-Space
As for the other coupled oscillator systems that we’ve studied, we can fully specify
the behavior of the continuous string either by specifying y (x , t) or by specifying
all the amplitudes Cn in the normal mode expansion. Let’s consider the important
special case ẏ0 (x) = 0, for which Im Cn = 0. Then, we can write the normal mode
expansion (7.7.4) as
N

y(x , t) = Cn cos ωn t yn (x) ,
n =1
' (
where Cn = Re Cn = yn (x) | y0 (x)
Normal mode expansion for ẏ0 (x) = 0.
At t = 0, this reduces to
N
2
y(x , t = 0) = Cn sin kn x . (7.8.1)
L
n=1
Each Cn is associated with a kn . We could thus make a plot of Cn versus k, as shown

in figure 7.8.1 for a particular y (x , t = 0) which involves mixing several normal
modes. The initial shape of the string can be specified either by plotting y versus
x, or by plotting Cn versus k. The latter view is called “k-space.” We can make a
Figure 7.8.1 k-space for the continuous string.
Figure 7.8.2 Three-dimensional view of normal mode analysis at t = 0. (Note that the picture
is somewhat schematic. For example, the actual perspective from viewpoint B would not allow
one to tell whether each Cn was positive or negative.)
three-dimensional plot, as shown in figure 7.8.2, which symbolizes the process of

normal mode analysis at t = 0. Looking from viewpoint A, we can imagine adding up
the sinusoids to produce y (x , t = 0); this is the real space view. However, we could just
as well look from viewpoint B, which gives the k-space view. This idea of looking at
things in k-space turns out to be tremendously important in crystallography, solid-state
physics, and many other areas.
terms:
Standing wave (7.2)
Wavenumber (7.2)
Dispersion relation (7.2)
Longitudinal waves (7.5)
k-space (7.8)
Real space (7.8)

Boundary conditions & allowed frequencies (7.2)
Mode index & number of nodes (7.2)
Bead spacing & maximum wiggliness (7.3)
Transverse & longitudinal oscillations (7.5)
Normal modes for beaded & continuous strings (7.6)
Hilbert space vectors & continuous functions (7.6)
Representations of a function in real space & in k-space (7.8)

Highest frequency/smallest wavelength standing wave for a beaded string (7.3)
Functions only taking on physical reality at certain positions (7.3)
Find the frequency for a given normal mode of a beaded string (7.2)
Explain why there is a minimum wavelength for standing waves on a beaded
string (7.3)
Analyze a given pattern of initial velocities and positions for a beaded string into a
superposition of normal modes (7.4)
Find the frequency for a given normal mode of a continuous string (7.6)
Analyze a given pattern of initial velocities and positions for a continuous string into
a superposition of normal modes (7.7)
Take inner products of continuous functions (7.7)
Problems

7.1 State what is wrong with the following: “When a beaded string is in pure
normal mode 2, the displacement of bead 1 is given by A sin k1 x2 cos ω1 t,
where A is the amplitude of motion of the bead.” (There might be more than
one thing wrong.)
7.2 Orthogonality of normal modes for the beaded string
(a) Find the normalized
( eigenvectors
( ( for the beaded string with N = 3.
Let’s call them e1 , e2 , and e3 . (The subscripts here refer to the

mode index' n, not( to the mass index j.)
(b) Show that en | en = 1 for all possible values of n, thus verifying
that you have
' correctly
( normalized the eigenvectors.
(c) Show that en | em = 0 for all possible combinations of n and m
where n = m, thus verifying that the eigenvectors are mutually
orthogonal.
7.3 Beaded string applet
Go to the website for this text, and under chapter 7 open the “Beaded String
Applet.”
Write out answers to all questions in boldface.
(a) This applet demonstrates some properties of standing waves on a
beaded string. You may select the number of beads (masses) and the
desired normal mode to display with the two sliders at the bottom of
the applet. Additionally, you may at any time choose to slow down
the animation with the “Slow” check box.
(b) First, look at a string with 4 beads on it. Set that up using the
“masses” slider. Now look at the normal modes for this configuration
by varying the “mode” slider. How many normal modes are there?
Try a few other configurations consisting of different numbers of
beads. In general, for this one dimensional case (beads can only
move up and down), how does the number of normal modes
relate to the number of beads?
(c) The main goal of this applet is to visually demonstrate the fact that
there is a maximum meaningful normal mode index n for waves on
a beaded string. As we have just seen, the maximum index relates
to the number of beads on the string. What happens if you try to
generate normal modes with indices higher than the maximum
possible mode for a situation? Try it. What do the blue and
red waves represent? Explain why these two waveforms do not
represent the actual string. What would the actual string look
like? Take a moment to try the animation with a number of different
normal mode frequencies and numbers of beads. Think about what
you have seen.
(d) Explain what the graph in the lower left represents. Why is the
right part shown as a dashed line?
(e) If there is a maximum normal mode index for a beaded string,

what does this imply about the possible frequencies of waves that
can exist on the string? Explain. What does this imply about
waves in a crystal lattice (which can be thought of as a three
dimensional array of beads on springs)?
(f) Given the normal modes for a beaded string, could one construct
a Hilbert space to represent this system? What would it look
like? How many dimensions would it have?
7.4 Driven beaded string. A string with N beads, each of mass m and spacing
a, is under tension T . At the left, the string is attached to a fixed wall, as
usual. However, on the right, it is attached to a wall which oscillates up
and down with amplitude h and angular frequency ωd . This means that the
“fictitious bead” on the left has y0 ≡ 0 as usual, but the fictitious bead on
the right has yN +1 ≡ h cos ωd t. The differential equations of motion for this
system are the same as for the undriven case (because the way the beads
interact with each other is unchanged); only the boundary conditions are
different.
(a) Explain why it’s reasonable to guess that a solution for this problem
?
would be yj = Yj cos ωd t − δ .
(b) Given the form of the above guess, explain why we expect that Yj
need not equal yj0 .
(c) Since the differential equations of motion are the same as for the
undriven case, it’s reasonable to guess that the amplitudes Yj vary
with position in the same way, that is, it’s reasonable to guess
?
Yj = A sin α xj , where the values of A and α are to be determined.
Use yN +1 ≡ h cos ωd t to find δ and A in terms of α , h, a, and N.
? ?
(d) Plug our guesses yj = Yj cos ωd t − δ and Yj = A sin α xj into the
equation (7.1.2) (which was derived simply by applying F = mẍ to
mass j):
T
− −yj−1 + 2yj − yj+1 = ÿj
ma
Verify that the guesses work, and show that cos (α a) = 1 −

ω2

d .
T
2
ma
T
(e) Explain why the above implies that if ωd > 2 then α is
ma
complex. Hint: Set α = αr + iαi , use cos (α a) = Re eiα a , and

T
show that for ωd > 2 , α must be nonzero.
ma i
(f ) Explain why a complex value of α means that the wave damps
ix
exponentially in space. Hint: recall that sin x = Im e .
(g) What does that phrase “damps exponentially in space” mean?

T
(h) Why is it reasonable that the frequency 2 should be the highest
ma
frequency which does not give exponential damping in space?
7.5 Normal mode analysis for the beaded string. (You are encouraged to use
Mathematica, Maple, or other symbolic algebra program for this problem,
though it is not necessary.) A beaded string has three beads, each of mass
1.00 kg, spaced at 1.00 m intervals. The string has tension 100 N. The initial
positions and velocities of the beads are:
y1 = −2.55033 cm ẏ1 = −35.8043 cm/s

y2 = 3.51028 cm ẏ2 = 1.38233 cm/s
y3 = −0.885746 cm ẏ3 = 15.6333 cm/s
(a) Write explicit equations for y1 (t) , y2 (t) , and y3 (t) that are valid
for all times.
Your equations should be in terms of cosines, rather
than Re eiωt . Note: Remember that the arctan returns a result
that is only defined up to an additive factor of π . Therefore, think
carefully about the result your calculator or program returns
to you – is it in the correct quadrant of the complex plane? If
not, you need to add π to it.
(b) Plug t = 0 into your expressions from part (a), and verify that they
give the correct values for the initial positions.
N N +1 nπ
sin2 kn xj =
:
7.6 Show that , where kn = , x = ja, L = (N + 1) a,
j=1 2 L j
and n and N are integers. (We used this result in section 7.4.) You should
probably use a symbolic algebra program for this; please see the additional
instructions on the web page for this book.
7.7 Fret spacing on guitars. (Read problem 5.1 before doing this problem; you
need not actually do that problem first, though.) A standard guitar has six
strings. Each is held under tension, stretched between the bridge (near the
middle of the main body of the guitar) and the nut (near the end of the neck),
as shown in figure 7.P.1. A series of frets is positioned just below the strings.
When the guitarist uses a finger just behind one of the frets to push one of
the strings against the fretboard, the effective length L of the string is the
distance from the bridge to the fret, as shown. The frets are spaced so that the
guitarist can create notes in half-step intervals. For example, if the highest
string is plucked without using any fret, then the full length of the string
(from the bridge to the nut) vibrates, producing a note of E above middle C.
(The main pitch produced is from the fundamental, i.e., the n = 1 mode.)
If the guitarist pushes down behind the first fret (the one closest to the nut),
and then plucks the string, it instead produces an F#. If the guitarist pushes
down behind the second fret, the string produces an F#. If the length from
bridge to nut is L0 , what is the equation that determines the position for
fret number j, where j = 1 for the fret closest to the nut, and positions are
Figure 7.P.1 Top: The frets on a guitar allow the guitarist to vary the effective length of each
string in a controlled way. Bottom: side view of a finger pressing a string down behind a fret.
(Top image © Milinz | Dreamstime.com)
measured relative to the bridge? (Recall from problem 5.1 that the ratio of
frequencies for notes that are a half-step apart is 1.05946.)
7.8 Harmonics on guitars. Rather than using the frets (see problem 7.7),
guitarists sometimes use an alternate technique. Instead of using a finger
to press the string against the fretboard, the finger is placed lightly on the
string, at a carefully selected point. The string is plucked, and then, as quickly
as possible, the lightly pressing finger is released. The string emits a note that
is bell-like in tone. If the lightly pressed finger is placed at a point halfway
along the length of the string, the frequency of the note sounded is twice
that of the fundamental frequency of the string. Explain what is happening.
(Various such “harmonics” can be produced by different placements of the
lightly pressed finger. These can be used to produce notes much higher
than the guitar could ordinarily make. For an example, go to the website
for this text, and under this chapter find the listing for this problem. This
technique is also very important for rock guitarists; you can see lots of
examples by doing an internet video search using the words rock guitar
harmonics.)
7.9 You can make a crude musical instrument by stretching a rubber band
and plucking it. Try it. This will work best with a relatively thick band,
held close to your ear. For the most reproducible results, hook your fingers
through the band, rather than pinching it between your fingers. Important:
you should notice that the band stretches fairly easily to a certain length, and
then becomes much stiffer. We will concentrate on the range of fairly easy
stretching. Start with the band stretched most of the way, and slowly let it
relax as you pluck it again and again, holding it very close to your ear so you
can hear clearly. Let it relax all the way to the point where it is slack. Then,
stretch it out most of the way, and listen again as you let it relax. Using the
ideas you’ve learned in this book, make a simple quantitative model for how
the pitch should depend on length, and compare your model qualitatively
with your observations. (You will probably find that the behavior is more
complicated if you start with the band slack and stretch it out than it is
if, as directed, you start with it stretched and allow it to relax. This shows
that the model you have developed doesn’t capture all the physics of this
situation.)
7.10 A reasonably accurate model for the dispersion relation of an actual stretched
piano string is
ω = ak + bk 3 ,
where a and b are constants. Find an expression for all of the possible
oscillatory frequencies of a piano string of length L.
7.11 A string of mass M and length L is attached to walls at either end, and is
under tension T . The string is held at rest in the following shape: y = B for
L 3L
<x< , and y = 0 for the rest of the string. The string is released
2 4
at t = 0. (a) Find y(x , t). You may express your answer as an infinite
series, so long as you have defined all the symbols in your series. (b) Using
words, diagrams, and equations, describe what you did in part a by making
analogies with a system of conventional two-dimensional vectors in ordinary
x − y space.
7.12 (You may wish to complete problem 7.11 first.) A string of mass M and
length L is attached to walls at either end, and is under tension T . At t = 0,
L 3L
the string has the following shape: y = B for < x < , and y = 0 for the
2 4
L L
rest of the string. At t = 0 the velocity distribution is ẏ = E for < x < ,
4 2
and ẏ = 0 for the rest of the string. E has units of m/s. Given these initial
conditions, find y(x , t). (Note that this velocity pattern is not the same as the
initial position pattern – one is nonzero on the right side of the string, and
the other is nonzero on the left side of the string.) You may express your
answer as an infinite series, so long as you have defined all the symbols in
your series.
7.13 A continuous string with mass per unit length μ is stretched with tension T
between two walls, one at x = 0 and the other at x = L. At t = 0, its shape
L x2
is given by y (x , t = 0) = , and it is at rest. What is the complex
100 L 2
coefficient for the n = 2 mode in the normal mode expansion for these
initial conditions? You may need the following integral (given in the form
you would find it in an integral table):

x2

2 2x 2
x sin ax dx = 2 sin ax + − cos ax .
a a3 a
7.14 A string of mass M and length L is attached to walls at either end, and is
under tension T . Prior to t = 0, the string is held fixed with the following
shape:
⎧ ⎫
4 2
L ⎪
⎪
⎨ 2 Gx
⎪ 0≤x≤ ⎪
2
⎬
L
y0 (x) = , where G is a constant.
⎩ 4 G(x − L)2 L
⎪
⎪ ⎪
⎪
≤x≤L ⎭
L2 2
(a) Sketch this shape.
(b) At t = 0, the string is released. For the normal mode expansion,
briefly explain why it is unnecessary to calculate the imaginary
parts of the expansion coefficients.
(c) Calculate the coefficients in the normal mode expansion. It might
be helpful to note the solutions to the following integrals:
(2 − A2 x 2 )cos(Ax) 2xsin(Ax)

x 2 sinAx dx = +
A3 A2
(2 − A2 (B − x)2 )cos(Ax) + 2A(x − B)sin(Ax)

(x − B)2 sin(Ax)dx = .
A3
(d) You should have found that coefficients for the even n terms in the
expansion are all zero. Explain why, based on symmetry.
(e) Write an expression for y(x , t). You may express your answer as an
infinite series, so long as you have defined all the symbols in your
series.
7.15 At t = 0, the shape of a string under tension T which has mass/length μ is
a complicated shape, as shown in figure 7.P.2, and the string is motionless
at this instant. (The vertical scale is greatly exaggerated; displacement from
equilibrium is actually small, so that our usual approximations work well.)
The distance between the walls is L. Assume damping is negligible. It is
Figure 7.P.2 Top: Initial shape of string. Bottom:

inverse of the original shape.
observed that at t = τ , the string returns to exactly the same shape. The
shape also recurs at t = 2τ , 3τ , and so on.
(a) Using ideas of normal mode analysis, explain this surprising result,
and also find the value of τ in terms of the other parameters above.
(b) Explain why the inverse of the original shape, that is, the one shown
in the lower part of the figure, is never observed.
7.16 The particle in a box. In section 1.11, we discussed the quantum mechanical
wavefunction (x , t). As you will learn in a later course on quantum
mechanics, it is governed by Schrödinger’s equation:
h̄2 ∂ 2 ∂
− + U (x) = ih̄ ,
2m ∂ x 2 ∂t
where h̄ is Plank’s constant, m is the mass of the particle (usually an electron),
U(x) is the potential energy the particle would have at position x, and =
(x , t) is the “wave function” which describes the particle. Note that this
is a linear differential equation, so that we can superpose solutions for it
∂
just as we did for the string stretched between two walls. The notation
∂t
means “partial derivative of with respect to x.” This simply means “take
the derivative of with respect to t, treating x as a constant.” Similarly,
∂ 2
the notation means “second partial derivative of with respect to x”,
∂ x2
which means “take the second derivative of with respect to x, treating t as
a constant.” You should not get too worried about this notation; it is needed
because y is a function of both x and t, but there is really nothing mysterious
about it.
You can see right away, from the presence of the “i” in the above equation
that is intrinsically complex. Therefore, in what follows there is no need
to think about taking real parts of anything.
0 0<x<L
Consider a particle in the potential U = . If the
∞ x < 0 or x > L
particle has a finite total energy then it cannot exist outside the region x = 0
to x = L, since outside this region its potential energy would exceed its total
energy by an infinite amount. Therefore, the boundary conditions are that
must go to zero at x = 0 and at x = L. This is very analogous to what
happens to y for a continuous string stretched between two walls.
(a) We can guess that the normal modes of this “particle in a box”
system are given by

Cn ψn (x) e−iωn t inside the well
n = ,
0 outside the well
where the functions ψn (x) = A sin kn x are the normalized eigen-
functions, and A is a constant that you’ll determine in part (c) of
this problem. Verify that this guess is correct by plugging it into
the Schrodinger equation and showing that it works. As part of
doing this checking, you should find the dispersion relation that
is required to make the guess work. (Note that this guess has
the factor e−iωn t , rather than the factor eiωn t that you might have
expected from our discussion of standing waves on a rope. For the
rope waves, since we take the real part anyway, it wouldn’t have
made a difference to include the minus sign in the exponential,
although we followed well-established convention by not including
it. As you can see from this exercise, the minus sign really is needed
in the quantum mechanical version.)
(b) What is the condition on kn in order for the boundary conditions to
be satisfied?
(c) Find the value of A that correctly normalizes the eigenfunctions for
this situation. You may do this mathematically, or by referring to
results we have previously obtained.
(d) Make an argument for why the eigenfunction for the lowest-
frequency normal mode is orthogonal to the eigenfunction for the
second lowest-frequency normal mode. Note that a response such
as, “All eigenfunctions are orthogonal.” is not acceptable; you must
demonstrate that the eigenfunctions are orthogonal by mathematical
or logical argument.
(e) Now
⎧ consider this particular initial condition: At t = 0, =
⎨ L L
B <x<
4 2 . Explain why this means that all the coefficients
⎩0 elsewhere
Cn that appear in the normal mode expansion of must be real.
Hint: This is a consequence of the fact that is real everywhere
at t = 0. Also, recall that there is no Re[…] in the normal mode
expansion for this situation, as indicated by the boldface text in the
introduction to this problem.
(f) Given the initial conditions of part (e), find (x , t). You may express
your answer in terms of an infinite series, so long as all quantities in
the series are explicitly defined, including evaluation of all integrals.
Hint: As you showed in part e, all the coefficients Cn are real.
Recall that, for a string stretched between two walls we get that
all real coefficents if all the initial velocities are zero. Therefore,
having a quantum mechanical wavefunction that is initially real
everywhere is fully analogous to a stretched string with zero initial
velocity everywhere.
(g) Now, consider a different initial condition, which is formed by
superposing equal amplitudes of the second-lowest-frequency nor-
mal mode and the third-lowest-frequency normal mode. The result-
ing superposition shows a “sloshing” back and forth of (x , t).
What is the period of this sloshing, in terms of h̄, m, and L ? Explain.
7.17 k-space picture for plucked guitar string. The tonal quality of a note
sounded by a guitar string depends on where along its length the string is
plucked. (For this problem, assume the string is played at full length, without
using any frets.) The graphs in figure 7.P.3 show the k-space picture for two
Figure 7.P.3 Two possible k-space graphs.
different plucking positions, with amplitude on the vertical axis and k on

the horizontal axis. The scales are the same for both graphs. For one of the
graphs, the string was plucked at the halfway point, that is, a finger halfway
along the length of the string pulled it slightly away from equilibrium and
then released it. For the other picture, the string was plucked one quarter
of the way along its length. Which graph is which? Explain your reasoning
thoroughly, including an explanation for why some of the peaks are missing
from one of the graphs.
8 Fourier Analysis
You need to move past Fourier transfers, and start thinking quantum mechanics.
—Maggie Madsen, signal analyst character, the Transformers movie (2007).
8.1 Introduction
Although the above quote is a bit mangled (presumably the writer meant “Fourier
transforms”), it accurately conveys that the first thing any scientist does with a
complicated data set is to subject it to Fourier analysis, that is, the scientist finds
how the data can be expressed as a sum of sinusoids. As with normal mode analysis,
this gives a powerful and drastically different view of the data, one that is often very
revealing. Fourier analysis is absolutely omnipresent in modern technology. The jpg
image compression algorithm is based on Fourier analysis. The performance of fiber
optic cables is evaluated using Fourier analysis. Diffraction methods used to determine
the structure of proteins are based on Fourier analysis.1
As an example of how illuminating this method can be, consider the function y(x)
shown in the left part of figure 8.1.1. It appears quite irregular, and would be difficult to
describe in any simple way. However, this is just the sum of the three sinusoids shown
in the middle part of the figure. The right part shows a graph of the amplitude of the
sinusoids (in other words the factor A in A cos (kx + ϕ )) as a function of wavenumber
2π
k = ), and a graph of the phase ϕ of the sinusoids. This is just as complete a
λ
description as the graph of y(x), yet it is much more revealing. In this chapter, we will
1. Joseph Fourier (1768–1830) lived a varied and tempestuous life. He was a strong supporter of
the French revolution, but later became aghast at the excesses of the Terror, and tried to withdraw
from the committee. This almost led to his beheading. He was sent to Egypt by Napoleon, along
with 164 other scholars, to “civilize” the country. Fourier spent three years there cataloging
antiquities and other discoveries. This exposure to warm climates may be responsible for his
habit of keeping his rooms uncomfortably warm, while wearing a heavy coat. He made important
contributions to the study of heat propagation, and it was in this connection that he developed
the idea of summing sinusoids to represent other functions. However, this notion met with a
great deal of resistance from the leading French mathematicians of the time, including Laplace,
Legendre, and Poisson.
246
Chapter 8 ■ Fourier Analysis 247
Figure 8.1.1 A seemingly complicated y(x) (on the left) is actually just the sum of the three
sinusoids shown in the center. The amplitude and phase of each sinusoid A cos(kx + ϕ ) are
shown on the right. (The amplitude is defined to be positive. For the three sinusoids here, all
happen to have a negative phase.)
describe how to go from the plot as a function of x to the plot as a function of k.

For a function of time, we can use the same procedure to go from the plot as a function
2π
of t to the plot as a function of ω = .
T
We begin by considering functions y(x) that are periodic (but complicated within
the period), and then go on in section 8.5 to consider how similar procedures can be
applied to functions that are not periodic.
8.2 The Fourier Expansion
In chapter 7, we saw how any behavior of a continuous string can be expressed as a

superposition of the normal modes. As an important special case, we explored how
any initial shape of the string with ẏ0 (x) = 0 can be expressed as a superposition of
normal modes with real amplitudes. Of course, the string goes to zero at x = 0 and at
x = L, as do all the functions for the modes.
Let’s look at this from a purely mathematical viewpoint. The string can take on
any shape between x = 0 and x = L. Therefore, we could think of the string shape
as a mathematical function y (x) which is only defined from x = 0 to L. Thus, over
this range of x, any function that goes to zero at x = 0 and L can be expressed as
a superposition (i.e., a weighted sum) of the sinusoidsthat go to zero at x = 0 and
2 π
L. Equivalently, we could say that the set of functions sin kn x, where kn = n ,
L L
forms a complete basis for describing functions that go to zero at x = 0 and x = L, but
this description is only valid over the range x = 0 to x = L.
Since these basis functions are periodic, it is reasonable to ask whether we can
extend this idea; instead of only describing functions in the range x = 0 to L, perhaps
we can describe any periodic function over the entire range x = −∞ to x = ∞.
For example, let’s try to describe
π the periodic function shown as a solid gray line in
figure 8.2.1, y (x) = A sin x . From x = 0 to L this is perfectly described by

L
L 2 π
a single one of our basis functions: y (x) = A sin k1 x , where k1 = .
2 L L
Figure 8.2.1 Solid gray line: A

periodic function y(x). Dashed
line: An effort to represent y(x)
using a sinusoid works only over
the range x = 0 to x = L.
π
Figure 8.2.2 Functions of the form sin n x with odd n (top graphs) don’t have periodicity
L
L, while those with even n (bottom graphs) do.
However, outside the range 0 to L, this description doesn’t work at all, as shown
2
in the figure. The problem is obvious: the basis function sin k x doesn’t match
L 1
2
the periodicity of y (x). In fact, of the set of basis functions sin kn x, all those
L
with odd n don’t have periodicity L, as shown in the top graphs of figure 8.2.2.
However, all those with even n do have periodicity L, as shown in the bottom
graphs.
So, it is tempting
simply to discard all the basis functions with odd n. That would
2 π
leave the functions sin n x , with even n. We could write this set more simply
L
L
2 2π
by writing them as sin n x , where now n can take on any value from 1 to ∞.
L L
We can simplify the notation even further by redefining the wavenumbers for Fourier
analysis:
2π
kn ≡ n for Fourier analysis, (8.2.1)
λ
where λ is the periodicity of the function being analyzed. (In the example above, λ = L.)
Each of the wavenumbers in equation (8.2.1) corresponds to fitting an integer number
n of wavelengths between 0 and λ. Compare this to our expression from normal mode
π
analysis, kn = n , which corresponds to fitting a half-integer number of wavelengths
L
between 0 and L. Using the Fourier analysis definition of kn , we can write the remaining
functions (the ones we didn’t throw out) as sin kn x. (Note that we have omitted
basis
2
the prefactor; this is the well-established convention for Fourier analysis. Because
λ
we are not including this in the basis functions, they are no longer normalized; we’ll
see soon that it is easy to compensate for this.)
However, since we started with a basis complete enough to describe functions
that go to zero at x = 0 and L, over the range x = 0−L, it’s clear that the basis will
no longer be complete once we discard half its members. To restore completeness, we
need to add back in an equal number of basis functions, but this time we’ll make sure
they have periodicity λ. What additional functions could we add to the basis? Hmm…
2π
if only there were some function other than sin n x = sin kn x that had periodicity
λ
λ. If only… Aha! cos kn x! In fact, it can be shown (though we will not do so here)
that the set of functions sin kn x combined with the set of functions cos kn x forms a
complete basis for describing functions y (x) with periodicity λ, so long as y (x) has an
average value of zero. To describe functions with nonzero average value, we must add
one more basis function to our set: a constant.
So, it is reasonable to expect that any function y (x) with periodicity λ can be
expressed as a weighted superposition of the functions sin kn x, cos kn x, and a constant.
This statement expressed mathematically reads
∞
a
y (x) = o + an cos kn x + bn sin kn x . (8.2.2)
2
n=1
Fourier series expansion
In the above, the an ’s and bn ’s are the coefficients in the expansion; in section 8.4, we
will find how to determine their values, and show that they are real.
Self-test (answer below2 ): What is the basis function that is the constant part of the
Fourier expansion?
We have not rigorously shown that the set sin kn x, cos kn x, and a constant forms
a complete basis for all functions with periodicity λ. However, this is plausible based
on our experience with the normal mode expansion (the completeness of which we
showed with full rigor), and can be shown rigorously.
2. Each coefficient an or bn multiplies the corresponding basis function. We see that a0 multiplies
y(x) = ½, so the constant basis function in the Fourier expansion is ½.
8.3 Expansions using nonnormalized orthogonal basis functions

2
As mentioned earlier, by a well-established convention, the normalization factor
λ
is not included in the basis functions
for Fourier analysis; for example, we use sin kn x
2
as basis functions, rather than sin kn x. It’s pretty obvious how to compensate in
λ
2
this particular case: we simply absorb the into the coefficients an and bn . However,
λ
compensating for nonnormalized basis functions in the more general case is easy.
Say we have a set of functions yn (x) that form a complete orthogonal basis. The
“complete” part of this means that we can write

y (x) = Cn yn (x), (8.3.1)
n
where y (x) is any function that can be described by this basis, and the Cn ’s are
the
' expansion
( coefficients (which may be complex). The “orthogonal” part means
ym (x) yn (x) = 0, if m = n.

' (
If the basis functions yn (x) are normalized, that is, yn (x) yn (x) = 1, then it’s
easy to find the Cn ’s:

' ( ' ( ' (
y (x) y (x) = y (x)
m m C y (x) =
n n C y (x) y (x) = C ,
n m n m
n n
' (
where in the last step we used ym (x) yn (x) = δmn . Therefore,
' (
Cm = ym (x) y (x) .
At this point, we can replace the index m by n, giving

' (
Cn = yn (x) y (x) . (8.3.2)
(Note that this is just what we got before for the case of the normal mode expansion
with ẏ0 (x) = 0.)
However, what if the basis functions aren’t normalized? In other words, what if
' (
yn (x) yn (x) = Fn , where Fn = 1?
Claim: The generalized version of equation (8.3.2) is

' (
yn (x) y (x)
Cn = ' (. (8.3.3)
yn (x) yn (x)
Obviously, this reduces to equation (8.3.2) when yn (x) is normalized.

Proof :

' ( ' ( ' (
ym (x) y (x) = ym (x)
Cn yn (x) =
Cn ym (x) yn (x)
n n
' (
= Cm ym (x) ym (x) ,
' (
where in the last step we used orthogonality, that is, ym (x) yn (x) = 0 if m = n.
' (
ym (x) y (x)
⇒ Cm = ' ( , Q.E.D.
ym (x) ym (x)
Here’s one way to see intuitively why equation (8.3.3) works. If yn (x) is not
normalized, then it has a “length” in Hilbert space that is not equal to 1. One factor of
this
' length appears
( in the normal mode expansion (8.3.1), ' and another
(factor appears in
yn (x) y (x) , so we must divide by (length)2 , that is, by yn (x) yn (x) , to compensate.
8.4 Finding the coefficients in the Fourier series expansion
Here, again is the Fourier series expansion from section 8.2:

∞
a0
(8.2.2) : y (x) = + an cos kn x + bn sin kn x .
2
n=1
As we can see, the basis functions are 1/2, the set of cos kn x, and the set of sin kn x.
To find the expansion coefficients an and bn , we use equation (8.3.3):
'1 ( λ
2 y (x) 4 ' 1
'1 1( 1 2 λ ( 2
a0 = ' 1 1 ( , 2 2 =

2 dx = ⇒ a0 = 2 y (x) = 1 | y (x)
2

2
4 λ λ
0
' ( λ
cos kn x y (x) ' ( λ
an = ' (, cos kn x cos kn x = cos2 kn x dx = ,
cos k x cos k x
n n 2
0
where in the last step we made use of the fact that the average value of cos 2 (or sin 2 )
over one wavelength is ½. So,
2' (
an = cos kn x y (x) .
λ
2π
Since kn = n , we have k0 = 0, so that cos k0 x = 1. Therefore, the above works
λ
for a0 as well as the other an ’s . Plugging in the definition of the inner product from
section 7.7, we get
λ
2
an = cos kn x y (x) dx .
λ
0
We can find the coefficients of the sines in the Fourier expansion in the same way:
' ( λ
sin kn x y (x) ' ( λ
bn = ' ( , sin kn x sin kn x = sin2 kn x dx = ,
sin kn x sin kn x 2
0
λ
2' ( 2
⇒ bn = sin kn x y (x) = sin kn x y (x) dx .
λ λ
0
Let’s collect these results:
∞
a0
y(x) = + an cos kn x + bn sin kn x .
2
n=1
λ
2
an = cos kn x y (x) dx .
λ
0 (8.4.1)
λ
2
bn = sin kn x y (x) dx .
λ
0
2π
kn = n .
λ
Fourier analysis for a function of x that has periodicity λ
Example: What is the Fourier series expansion for the square wave, shown as the left
graph in figure 8.4.1?
Solution: We can represent the square wave y(x) using the Fourier series expansion
shown in the top line of equation (8.4.1). Let’s begin by calculating the an ’s:
⎡ ⎤
λ λ/2 λ
2 2⎢ ⎥
an = cos kn x y (x) dx = ⎣ cos kn x (1) dx + cos kn x (−1) dx ⎦
λ λ
0 0 λ/2
+ λ/2 λ ,
2 1 1
= sin kn x
− sin kn x
λ kn 0 k n λ/2
"

#
2 2π λ 2π 2π λ
= sin n − sin 0 − sin n λ + sin n = 0.
λkn λ 2 λ λ 2
Thus, all of the cosine coefficients in the Fourier expansion are zero. We could have
anticipated this based on symmetry: the square wave function is antisymmetric about
the point x = λ/2, whereas the cosine functions are all symmetrical about this point.
Figure 8.4.1 Left: Square wave. Middle: The square wave is antisymmetrical about x = λ/2,
whereas the cosine is symmetrical. Right: Gray trace shows the sum of the first three non-zero
terms in the Fourier series expansion of the square wave.
This is shown for the case n = 1 in the middle part of the figure. However, the sine
functions do have the required symmetry, so we can anticipate that the bn ’s will be
nonzero:
⎡ ⎤
λ λ/2 λ
2 2⎢ ⎥
bn = sin kn x y (x) dx = ⎣ sin kn x (1) dx + sin kn x (−1) dx ⎦
λ λ
0 0 λ/2
+ λ/2 λ ,
2 1 1
= − cos kn x
+ cos kn x
λ kn 0 k n λ/2
"

#
2 2π λ 2π 2π λ
= − cos n + cos 0 + cos n λ − cos n
λkn λ 2 λ λ 2
2 2
= [− cos nπ + 1 + cos n2π − cos nπ ] = [1 − cos nπ ] .
2π nπ
λn
λ
4
If n is odd, then bn = , while if n is even then bn = 0. Plugging these results into the
nπ
top line of equation (8.4.1), we obtain

4 2π 1 3 · 2π
y (x) = sin + sin + ··· (8.4.2)
π λ 3 λ
Fourier expansion for a square wave.
As you can see, each succeeding term is smaller (because of the factor 1/n). The right
graph in figure 8.4.1 shows the sum of the first three terms. You can perhaps see how
the series begins to approximate the square wave. The approximation becomes better
as more terms are added.
Self-test (answer below3 ): Use symmetry arguments to explain why all the sines with
even n have zero coefficients.
You can show in problem 8.2 that an alternate version to equation (8.4.1) for the
Fourier series expansion is
∞
a
y (x) = 0 + An cos kn x + ϕn , (8.4.3)
2
n =1

b
where An = an2 + bn2 and ϕn = tan−1 − n . This is the version used for
an
figure (8.1.1).
We can also use Fourier analysis for a function of t. In fact, this is much more
common than using it for functions of x. Everything works in exactly the same way.
The periodicity in x is replaced by the periodicity in t: λ → T . The wave number is
3. Although these terms are antisymmetrical about x = λ/2, they are also antisymmetrical about
x = λ/4, whereas the square wave is symmetrical about this point.
2π 2π
replaced by the angular frequency: kn = n → ωn = n . With these substitutions,
λ T
equation (8.4.1) becomes
∞
a0
y(t) = + an cos ωn t + bn sin ωn t .
2
n=1
2 $λ
an = cos ωn t y (t) dt .
T 0
(8.4.4)
λ
2
bn = sin ωn t y (t) dt .
T
0
2π
ωn = n .
T
Fourier analysis for a function of t that has periodicity T
The version of equation (8.4.3) for functions of time is

∞
a0
y (t) = + An cos ωn t + ϕn . (8.4.5)
2
n=1
8.5 Fourier Transforms and the meaning of negative frequency
Complex exponential version of Fourier analysis

In problem 8.3, you will show that the Fourier expansion can be written in terms
of complex exponentials, rather than sines and cosines. You will show that we
can write
∞ λ

ikn x 1
y(x) = Cn e , where Cn = y(x) e−ikn x dx . (8.5.1)
n=−∞
λ
0
You will also show that, although the Cn ’s are complex, there is always cancellation
of the imaginary part of the term n = m with the term n = −m, so that there is no need
to take the real part of the summation to get y (x). One can show that the Cn ’s in this
expansion are related to the an ’s and bn ’s in the sine/cosine expansion:
1 1
Cn = an − ibn and C−n = an + ibn . (8.5.2)
2 2
This means for example, that for a case where C−n = Cn , we have bn = 0, that is, a
pure cosine term, while if C−n = −Cn , then we have an = 0, that is,
a pure sine term.
From equation (8.5.2), we can see that we must always have C−n = Cn .

Fourier transforms
If we take the limit λ → ∞, then the function y (x) is no longer truly periodic. However,
2π
we can still express it as an infinite sum of sinusoids. Since kn = n , as we increase λ
λ
the kn ’s get closer together. In the limit λ → ∞, the spacing between the kn ’s becomes
infinitesimal. Now, instead of associating each Cn with a kn , it is more helpful to think of
a continuous function C(k). By convention, this is instead called Y (k). In problem 8.4,
you can show that in the limit λ → ∞, equation (8.5.1) becomes
∞ ∞
1 ikx 1
y(x) = √ Y (k)e dk , where Y (k) = √ y(x)e−ikx dx (8.5.3)
2π 2π
−∞ −∞
Fourier Transforms for a function of x
Y (k), which represents the Fourier amplitude as a function of k, is called the “Fourier
Transform” of y (x). At first, this might appear intimidating, but remember that an
integral is just an infinite sum, so that we are still expressing y (x) as a sum of the
sinusoids eikx , each weighted by the factor Y (k).
2π 2π
It is easy to adapt this for functions of time: ω = plays the role of k = , so
T λ
that we can express any function of time y (t)as a weighted sum of the sinusoids eiωt :
∞ ∞
1 iω t 1
y(t) = √ Y (ω)e dω, where Y (ω) = √ y(t)e−iω t dt . (8.5.4)
2π 2π
−∞ −∞
Fourier Transforms for a function of t
Example: Fourier transform of a Gaussian. The Gaussian function is quite important

in several different contexts. It is defined as
2
/2 σ 2
y(t) = Ae−(t −t0 ) , (8.5.5)
a bell-shaped curved centered on t0 , as shown in figure 8.5.1a. For simplicity, let’s

examine the case t0 = 0, so that
2
/2 σ 2
y(t) = Ae−t. (8.5.6)

Your turn: Verify that y(t) falls to A / 2 at t = ±σ −2 ln 12 = ±1.18 σ , so that the Full
Width at Half Maximum (FWHM) is 2.35 σ .
Using equation (8.5.4), we see that the Fourier transform of equation (8.5.6) is
∞
1 2
/2σ 2 −iω t
Y (ω ) = √ Ae−t e dt . (8.5.7)
2π
−∞
continued
Figure 8.5.1 a: The Gaussian. b: A Gaussian centered on t = 0 and its Fourier transform.
The width (in time) of the y(t) Gaussian is inversely proportional to the width (in ω) of
the Y(ω) Gaussian. Therefore a broad range of frequencies is needed to synthesize a
narrow pulse.
This is a standard definite integral, which you can look up in a table (or evaluate with a
symbolic calculus program):
∞

ax 2 +bx +c
π b2 −4ac /4a
e− dx = e . (8.5.8)
a
−∞
1
In our case, x → t, a → , b → iω, and c → 0, so
2σ 2
A 2
2
2
2

Y (ω ) = √ 2π σ 2 e−ω / 2/σ = Aσ e−ω / 2/σ .
2π
We see that this has the same form as equation (8.5.6), that is, that the Fourier transform
of a Gaussian function of time y (t) is a Gaussian function of angular frequency, Y (ω).
1
The FWHM of y (t) is 2.35 σ , so the FWHM of Y (ω) is 2.35 , as shown in figure 8.5.1b.
σ
Thus, we arrive at a quite important conclusion: to Fourier synthesize a very narrow
Gaussian (one with a very small FWHM), we need a very broad range of frequencies (i .e.,
a Gaussian Y (ω) with a very large FWHM).4 This conclusion holds in general: to Fourier
synthesize a function y (t) with variations on time scales as short as
t, we must use
1
Fourier components covering a range of angular frequencies that is of order .

t
4. As you can show in problem 8.18, this leads directly to the Heisenberg uncertainty principle.
Meaning of the negative frequencies in the Fourier transform

The integral in equation (8.5.4) which represents the Fourier synthesis of y (t), y(t) =
1 $∞
√ Y (ω)eiω t dω, includes both positive and negative angular frequencies. Let’s
2π −∞
consider a particular positive angular frequency ω0 . We can see that for each such a
frequency in the integral, there is a corresponding negative frequency -ω0 . Since the
integral is an infinite sum,
the contributions
at these frequencies are added together as
part of the integral: Y ω0 eiω0 t + Y −ω0 e−iω0 t . This sum mustbe real, since y
(t)
is real, and this imposes a restriction on the relationship between Y ω0 and Y −ω0 ,
as we explore below.
The function eiω0 t represents a vector of length 1 rotating counterclockwise in the
complex plane, while for e−iω0 t the vector rotates clockwise, as shown in the top part
of figure 8.5.2. If we add them together, then the imaginary part of e−iω0 t cancels the
imaginary part of eiω0 t , leaving only a real function: eiω0 t + e−iω0 t = 2 cos ω0 t. Of
course, we could multiply both by any real number, and still get cancellation of the
imaginary parts: Aeiω0 t + Ae−iω0 t = 2A cos ω0 t. Finally, we can add a positive phase
shift to the eiω0 t and a negative phase shift of equal magnitude to the e−iω0 t , and still
get cancellation of the imaginary parts:
Aei(ω0 t +ϕ ) + Ae−i(ω0 t +ϕ ) = 2A cos ω0 t + ϕ ,

(8.5.9)
as shown in the bottom part of figure 8.5.2.

This is the only
way to get such cancellation.
Applying these ideas to the sum Y ω0 eiω0 t +
Y −ω 0 e
−iω0 t which is part of the

Fourier integral, we see that if we express Y ω0 in the form Y ω0 = Aeiϕ , then we
Figure 8.5.2 Top: The imaginary

parts of two counter-rotating
complex plane vectors can cancel
only if the angular frequencies are
equal and the magnitudes are
equal. Bottom: The phase factors
ϕ must be of equal magnitude and
opposite sign for the imaginary
parts to cancel.

must have Y −ω0 = Ae−iϕ . Thus, using equation (8.5.9),

Y ω0 eiω0 t + Y −ω0 e−iω0 t = 2A cos ω0 t + ϕ . (8.5.10)
To summarize:

Y ω0 is always equal in magnitude to Y −ω0 . The complex phase difference between

them determines the phase of the real oscillation cos ω0 t + ϕ .
8.6 The Discrete Fourier Transform (DFT)
In most applications of Fourier analysis, one does not actually deal with continuous
functions, but rather with discrete samples of a continuous function taken at regular
intervals. The most common applications are for functions of time. For example,
perhaps one wishes to perform Fourier analysis on an audio waveform. (Some of the
reasons for doing so are detailed in section 8.7.) In this case, a microphone converts
the sound waves into a time-varying voltage. The voltage is then sampled at regular
time intervals
, and these samples are recorded on a computer; this process is shown
schematically in figure 8.6.1. (We use the symbol
, rather than for example
t, just
for simplicity.) As long as
is smaller than the timescale on which the continuous
function changes significantly, then the resulting discretely sampled waveform is a
good approximation of the original.5
The samples occur at the times
tj = j
, (8.6.1)
where
j = 0, 1, . . . , N − 1, (8.6.2)
Figure 8.6.1 Sampling a

continuous waveform, in
preparation for the DFT.
2π t
5. For example, if we are sampling a sinusoidal waveform A cos ωt = A cos , where Tw is the
Tw
period of the waveform, then we must have
≪ Tw in order for the graph of the sampled signal
versus time to look approximately the same as the graph of the original signal versus time.

and
N is the number of samples. The result is a set of N data points: y t0 ,
y t1 , . . . , y tN −1 . (In the example of the audio waveform, y would be the voltage.)
How can we represent this dataset in angular frequency space (the equivalent of k-space
for functions of t), using ideas of Fourier analysis? In other words, how can we find
the “Fourier spectrum” of the dataset?
The algorithm used in such a situation is called the Discrete Fourier Transform
(DFT). It assumes that the function has a periodicity T = N
. However, most often
the continuous function that is being sampled (such as an audio waveform) does not
actually have this periodicity, since N and
are ordinarily determined by the hardware
and software used for data acquisition, rather than being determined by the system
being examined.6 This assumption that the DFT algorithm makes about the periodicity
can cause serious inaccuracies in the Fourier transform that it calculates, particularly
because the data point at the end of the set is usually at a different y-value than the
data point at the beginning. Thus, the assumed repeating function has a sharp step at
the beginning of each period. This would lead to spurious high-frequency peaks in the
Fourier spectrum. To avoid this, the input function is usually multiplied by a “window-
ing function,” as shown in figure 8.6.2, forcing the start and end points to zero.Although
this windowing procedure does introduce some distortions into the Fourier spectrum,
the benefits of eliminating the sharp step at the beginning of the period outweigh these
drawbacks. You can explore the windowing operation more in problem 8.16.
Because the algorithm assumes a periodicity of T = N
(equal to the measure-
ment time), the lowest angular frequency in the Fourier spectrum is
2π
ω1 = , (8.6.2)
N
and the higher angular frequencies are integer multiples of this:
2π
ωn = nω1 = n , where T = N
is the total measurement time (8.6.3)
N
Figure 8.6.2 Because the DFT assumes the function has periodicity T , problems occur if (as is
usually the case) the beginning and ending values of the sampled waveform (left) are not
equal; this causes a sharp step in the assumed waveform. To avoid this problem, the sampled
input waveform (left) is multiplied by a windowing function (center) to produce a waveform
(right) that goes to zero at the beginning and end of the time interval T . This example shows
the “Hanning window”, one of the most common windowing functions; it is simply an offset
cosine.
6. For example, if one is recording a vocalist singing a note at 440.3 Hz, corresponding to a period
of 2.271 ms, it is unlikely that the equipment used to make the recording will record for a time
corresponding to an exact multiple of 2.271 ms.
The resolution along the angular frequency axis of the Fourier spectrum equals the
spacing between the ωn ’s, so that:
The frequency resolution of the DFT is inversely proportional to the measurement

time T .
Recall in our discussion of the beaded string that there was a “maximum
wiggliness” that could be represented by the system, in which the beads form an
up–down–up–down pattern. In exactly the same way, there is a maximum wiggliness
that can be represented by the set of N data points, corresponding to a minimum period
of 2
, since we need at least two data points per period to represent the up–down
pattern. Therefore, the maximum angular frequency in the Fourier spectrum is
2π
ωmax = . (8.6.4)
2
By comparison with equation (8.6.3), we see that this corresponds to a maximum

value of n:
N
nmax = .
2
(We assume that N is even; this is almost always the case for reasons of computational
efficiency, as we’ll discuss briefly near the end of the section.)
a :∞
Following the top line of equation (8.4.4): y (t) = o + an cos ωn t +
2 n =1
bn sin ωn t , we will represent y(t) as a sum of a constant term, a series of cosines, and
a series of sines. However, for the reasons discussed earlier, the sum only extends to
N a
n = , instead of to n → ∞. Furthermore, instead of using 0 , cos ωn t , and sin ωn t
2 2
as the basis functions, it is conventional for DFT to divide them by N. Recall that,
since time is measured in the discrete increments
, we write y tj , where tj = j
and
j = 0, 1, . . . , N − 1. Putting these ideas together, we write the Fourier expansion as
⎧ ⎫
N /2
1 ⎨ ao ⎬
y tj = + an cos ωn tj + bn sin ωn tj .
N⎩2 ⎭
n=1

2π
However, when n = N /2, we have sin ωn tj = sin n j
= sin (njπ ) = 0, since
2
n and j are integers. (This is the same idea as the n = N + 1 “mode” of the beaded
string, in which all the beads are at nodes of the standing wave.) Therefore, we should
omit the sin term for n = N /2, leaving us with
⎧ ⎫
N/2−1
1 a0
⎨ ⎬
y tj = + an cos ωn tj + bn sin ωn tj + aN /2 cos ωN /2 tj . (8.6.5)
N⎩2 ⎭
n=1
N
In the above, we write − 1 as N /2 − 1 to save space. Note that there are N independent
2

N
coefficients in the above equation: a0 , aN /2 , and a total of 2 · − 1 = N − 2 an ’s
2
and bn ’s in the summation. Given that we started with N independent values of y, it is

logical that we should have N independent coefficients in the Fourier expansion.
It is customary to write the DFT using complex exponentials instead of sines and
cosines. The expansions for cosine and sine in terms of complex exponentials are:
eiωn t + e−iωn t eiωn t − e−iωn t eiωn t − e−iωn t
cos ωn t = and sin ωn t = = −i .
2 2i 2
Therefore, we can write equation (8.6.5) as
⎧ ⎫
/2−1
N
+ ,
1 ⎨ a0 eiωn tj + e−iωn tj eiωn tj − e−iωn tj ⎬
y tj = + an − ibn + aN /2 cos ωN /2 tj .
N⎩2 2 2 ⎭
n=1
(Because the cos ωN /2 t term has no accompanying sin term, we choose not to express
it as a complex exponential.)
Your turn: Show that, by defining

Cn ≡ 21 an − ibn and C−n ≡ 1
2 an + ibn , (8.6.6)
we can rewrite the above equation as

⎧ ⎫
/ 2−1
N
1 ⎨ a0 ⎬
y tj = + Cn eiωn tj + C−n e−iωn tj + aN/2 cos ωN/2 tj .
N⎩2 ⎭
n=1
2π
Because of equation (8.6.3): ωn = n , we have that −ωn = ω−n . Using this, we
N
can compress the above equation to

⎧ ⎫
N /2−1
1 ⎨ ⎬
y tj = Cn eiωn tj + aN /2 cos ωN /2 tj , (8.6.7)
N⎩ ⎭
n=−(N /2−1)
where we have defined b0 ≡ 0, so that the entry in the sum for n = 0 is C0 eiω0 tj =
a0 0 a0
e = .
2 2
For all values of n other than N/2, the basis functions in this expansion are
& eiωn tj
. To find the coefficients Cn , we can use the version of equation (8.3.3):

yn tj ≡
' N (
yn (x) y (x)
Cn = ' ( that would be appropriate for a function of time, that is,
y (x) y (x)
n n
' (
yn tj y tj
Cn = ' ( .
yn tj yn tj

Because we have a discrete set of data y tj , the inner products are taken in the same
way as we would for a beaded string. For example,
−1
∗ N −1 N −1
% & N eiωn tj eiωn tj e−iωn tj eiωn tj 1 1

yn tj yn tj = = = 2
= ,
N N N N N N
j=0 j=0 j=0
where the last step follows because there are N identical terms in the sum. Therefore,
the coefficients Cn are given by
N −1 −iωn tj
' ( e
Cn = N yn tj y tj = N y tj
N
j=0
N
−1

⇒ Cn = e−iωn tj y tj . (8.6.8)
j=0
Similarly, to find aN /2 , we use

' (
yN /2 tj y tj
aN /2 =' ( , (8.6.9)
yN /2 tj yN /2 tj
( cos ωN /2 tj
where yN /2 tj = . The denominator is
N
−1
' ( N cos2 ωN /2 tj
yN /2 tj yN /2 tj =
.
N2
j=0
2π
Recall from equations (8.6.3) and (8.6.1) that ωn = n and tj = j
, so that ωN /2 tj =
N
π j. Plugging this in gives

−1 −1
( N (±1)2
N
' cos2 π j 1
yN /2 tj yN /2 tj = 2
= 2
= .
N N N
j =0 j=0
Therefore, equation (8.6.9) becomes

N −1
' ( cos ωN /2 tj
aN /2 = N yN /2 tj y tj = N
y tj ⇒
N
j=0
N −1
cos π j
aN /2 = N y tj . (8.6.10)
N
j=0
Since
e−iωN /2 tj = e−iπ j = cos (−π j) + i sin (−π j) = cos π j,
we can write equation (8.6.10) as

N
−1

aN /2 = e−iωN /2 tj y tj .
j=0
Since this has the same format as equation (8.6.8), we can write
aN /2 cos ωN /2 tj = CN /2 cos ωN /2 tj = CN /2 cos π j. (8.6.11)
Since
eiωN /2 tj = eiπ j = cos π j + i sin π j = cos π j,

we can write equation (8.6.11) as
aN /2 cos ωN /2 tj = CN /2 eiωN /2 tj .
Now, we insert this into equation (8.6.7) to give

⎧ ⎫
N/2−1
1 ⎨ ⎬
y tj = Cn eiωn tj + CN /2 eiωN /2 tj . (8.6.12)
N⎩ ⎭
n=−(N /2−1)

N N
In the above, n ranges from − − 1 to − 1 . However, it is conventional for
2 2
2π
DFT to change this range. Recall from equations (8.6.3) and (8.6.1) that ωn = n
N
nj
and tj = j
, so ωn tj = 2π . Therefore,
N
(n+N)j Nj
2π i nNj 2π nj nj
eiωn+N tj = ei N 2π = ei N e = eij2π ei N 2π = ei N 2π = eiωn tj .
This means that we can add N to the index n for any of the terms of the sum in equation
(8.6.12) without changing anything. We will do this for all the negative values of n, so
that, for example,

N +N N +N
n=− − 1 −−→ + 1 and n = −1 −−→ N − 1. (8.6.13)
2 2
Using this, equation (8.6.12) becomes
⎧ ⎫
N /2−1 N −1
1 ⎨ ⎬
y tj = Cn eiωn tj + Cn eiωn tj + CN /2 eiωN /2 tj
N⎩ ⎭
n =0 n=N /2+1
−1
N
⇒ y tj = Cn eiωn tj , (8.6.14)
n=0
where equation (8.6.8):

N
−1

Cn = y tj e−iωn tj .
j=0
Equation (8.6.8) for the coefficients is called the Discrete Fourier Transform
(DFT),
while equation (8.6.14), which describes the Fourier synthesis of y tj , is called the
Inverse Discrete Fourier Transform (IDFT).
In equation (8.6.14), the term for n = 0 is the constant term, and the terms for n = 1
N
to n = − 1 correspond to the positive frequencies ω1 to ωN /2 − 1 . However, because
2
N
of the transformation (8.6.13), the terms from n = + 1 to n = N − 1 correspond to
2
the negative frequencies ω−(N /2 − 1) to ω−1 . Finally, the term for n = N /2 is derived
from a sum of positive and negative frequency terms. Although this is the conventional
sequence for the DFT, it is not very intuitive, because the highest frequency components
are those near n = N /2, while the components near n = N − 1 correspond to low
Figure 8.6.3 a: A function y sampled at 1000 times tj . b: The magnitude of the DFT for this
function. Note the symmetry about n = N /2. If one is only interested in the magnitude of the
N
Fourier spectrum, then the points from n = + 1 to n = N − 1 can be ignored.
2
frequencies (as do the components near n = 0). As with the Fourier Transform (see
section 8.5), there is no need to take the real part of anything in any of the above
discussions. There is therefore the same condition that the magnitude of the Cn for a
positive frequency must equal the magnitude of the Cn for the corresponding negative
frequency, so that these terms can combine to cancel the imaginary parts. Because of
the transformation (8.6.13), this means that

C = C
n N −n .
(8.6.15)
Figure 8.6.3a shows a sequence of N = 1,000 data points as a function of time,

in which there is no obvious pattern. Figure 8.6.3b shows the magnitude of the DFT
of this data (i.e., Cn as a function of n) , with the conventional sequence of n values,
that is, the sequence of equation (8.6.14). The pattern is symmetrical about the point
n = N /2 = 500. In fact, if one is only interested in the magnitudes Cn , which is often
N
the case, then the points from n = + 1 to n = N − 1 can be ignored, and one can
2
N
focus just on the values for n = 0 to n = .
2
It would appear from equations (8.6.14) and (8.6.8) that computing the DFT would
take a number of steps proportional to N 2 , since there are N different coefficients Cn ,
and to compute each of these one must do the sum (8.6.8) which has N entries. However,
if N = 2m , where m is an integer, then one can use a computer algorithm called the
“Fast Fourier Transform” (FFT) to compute the DFT, and this takes a number of
steps that is only proportional to N log N, a tremendous improvement. You may have
heard of research on quantum computers, in which data is represented by quantum

superpositions. Such computers have not yet been fully realized, but one can show
that they could be used to compute DFTs with a number of steps proportional only to
(log N)2 , a tremendous further improvement.
Because the DFT is of enormous scientific and commercial importance, a great
deal has been written about it. To learn more, you might begin with the classic
text “Numerical Recipes”,7 although you should beware that it follows the unusual
e−iωn tj
convention of using as the basis functions for the Fourier synthesis of y, rather
N
eiωn t j
than using as do most authors and as we have done above. The text by Julius O.
N
Smith is also very approachable, and includes reviews of all the necessary math.
8
8.7 Some applications of Fourier Analysis
Fourier analysis is so widespread in science and engineering that there are applications
in virtually every subfield. In this section, we explore two of them briefly.
Sonograms and whale calls. For scientists studying birds and animals, it can
be very helpful to have a visual representation of the characteristic calls made by a
particular species. A microphone converts the sound into a time-varying voltage. The
voltage is sampled at regular time intervals
for a measurement period T . The DFT
for this dataset is calculated, and then the process is immediately repeated.
Self-test (answer below9 ): (This is a hard self-test the first time you encounter it. So,
don’t spend too long on it before looking at the answer.) A scientist wishes
to create
a DFT of the sound of a whale call. She wants to make a plot of Cn versus angular
frequency, with the angular frequency ranging from 0 to 2π · (1,000 Hz), and with a
resolution in angular frequency of 2π · (10 Hz). What values of the sampling interval
and the measurement period T should she use?

A “sonogram” is a series of DFT plots at successive times, with Cn indicated by
a gray scale, frequency f = ω/2π on the vertical scale, and time on the horizontal
scale. (The time steps on the horizontal axis are equal to T , the time needed to collect
enough data for a DFT.) Figure 8.7.1a shows the sonogram for the call of a right whale,
showing that the call has three simultaneous frequency components, at about 250, 550,
7. Numerical Recipes: The Art of Scientific Computing, 3rd Ed., by W.H. Press, S. A. Teukolsky,
W. T. Vetterling, and B. P. Flannery, Cambridge University Press, Cambridge, 2007.
8. Mathematics of the Discrete Fourier Transform with Audio Applications, 2nd Ed., by Julius O.
Smith III, BookSurge Publishing, 2007.
2π 1
9. From equation (8.6.4), we have ωmax = , so we need = 1, 000 Hz ⇔
= 0.5 ms.
2
2
2π 2π 2π
Equation (8.6.3) states ωn = n , so the angular frequency resolution is = . Setting
N
N
T
this equal to 2π · (10 Hz) gives T = 0.1 s.
Figure 8.7.1 a: Sonogram of a Right whale. b: Sonograms for four different species of whale.
Note
the different time and frequency scales. Darker shades of gray darker gray indicate larger
C . Images courtesy of and © Prof. Christopher W. Clark, Cornell University.
n
and 850 Hz. This would be almost impossible to tell from the graph of voltage versus
time from the microphone. Figure 8.7.1b compares the calls from four different species
of whales.
There are fewer than 400 right whales remaining in the world. Although these
whales were heavily hunted in the past, collisions with ships account for many current
fatalities. Scientists are working to protect these whales using underwater microphones
on buoys to detect their calls, then warning ships away from possible collisions. It is
essential to distinguish the calls of the right whales from other sounds in the ocean,
to avoid false positives. The current generation of detection buoys uses relatively
simple techniques in an effort to select out the most significant signals, but the results
leave much to be desired. Scientists are hopeful that a future generation of buoys,
with computerized analysis of sonograms, will lead to more accurate detection. You
can learn more about this effort, including viewing a live map of the most recent right
whale detections, and listening to the calls, by visiting the links listed under this section
on the web page for this text.
JPEG image compression. According to the old saying, “A picture is worth
a thousand words.” In fact, it takes 11 kB(kilobytes) to store 1,000 words (at least
in the format used by my word processor), but it takes 9 MB (megabytes) to store
a reasonably high resolution (three megapixel) image if it is not compressed first.
Although storage has become inexpensive, the transmission of data is still a bottleneck
in many circumstances, so it is important to reduce the amount of data needed to
represent photographs and other images. After compression by the jpeg algorithm, the
same photo can be reduced to about 1 MB in size with no loss of quality that is visible
to the eye. In fact, it can be compressed to about 100 kB with only a little loss of
quality, unless one examines a magnified version. So, a more accurate version of the
saying might be, “A picture is worth at least 10,000 words.”
There are several image compression algorithms in wide use. The jpeg (“joint
photographic experts group”) method works especially well on photos, but can
introduce significant artifacts into schematic diagrams and other figures with sharp
edges and high contrast. In the jpeg method, the image is divided into squares of
8 × 8 pixels. The colors are represented by three numbers for each pixel, with one
number indicating the overall brightness and the other two indicating hue. Because
these numbers are functions of x and y (rather than of time), the Fourier transform is
in terms of wavenumbers (rather than angular frequencies). Each of the three sets of
color numbers for each 8 × 8 square is run through a version of the DFT called the
“Discrete Cosine Transform” (DCT). This uses only cosines as basis functions, rather
than both sines and cosines. However, the wavenumber spacing between the cosines
of the DCT is half that of the wavenumber spacing for DFT, so that the total number of
basis functions is the same. (Recall that, for normal mode analysis of a beaded string
π
we use only sines, with kn = n , whereas for Fourier analysis of a function of x we
L
2π
use both sines and cosines, but with twice the wavenumber spacing: kn = n .) In
λ
addition to the difference in basis functions, the DCT used for jpeg is a two-dimensional
transform; this is essentially a product of transforms in the x- and y-directions.
The advantage for jpeg of expressing the information in the 8 × 8 square as a
sum of cosines is that the eye is less sensitive to the high wavenumber (i.e., short
wavelength) components, and usually these components are smaller than the low
wavelength components. Therefore, after using the DCT to compute the coefficients
in the sum of cosines, the higher wavenumber coefficients can be represented with
very low accuracy, or even set to zero, with little perceptible change in the image
after the inverse transformation (to go back from the cosines to the real-space
colors of the pixels) has been applied. This is how the jpeg algorithm achieves
much of its compression, although significant compression is also obtained by
other steps.
terms:
Basis function (8.2)
Complete basis (8.2)
Orthogonal functions (8.3)
Fourier series expansion (8.4)
Fourier transform (8.5)
Negative frequency (8.5)
Discrete Fourier Transform (DFT) (8.6)
Sonogram (8.7)

Fourier analysis & normal mode analysis (8.2)
The frequency resolution of the DFT & the measurement time (8.6)
The maximum angular frequency of the DFT & the sampling interval (8.6)

Fourier series expansions & Fourier transforms (8.5)

The Fourier transform of a Gaussian is a Gaussian; the FWHMs are inversely
related (8.5)

Express any periodic function as a weighted sum of sines and cosines (8.4)
Use symmetry arguments to deduce which Fourier expansion coefficients are zero (8.4)
Given a complete set of orthogonal basis functions, express any function as a weighted
sum of the basis functions, even if they aren’t normalized (8.3)
Find the Fourier transform of a function that isn’t periodic (8.5)
Calculate the DFT by hand for very small datasets (e.g., up to four points) (8.6)
N
Explain the meaning of the DFT terms from n = + 1 to n = N − 1 (8.6)
2
Problems
8.1 Show that all the terms in the Fourier expansion (including sines, cosines,
and the constant term) of y(x) are orthogonal to each other. There are five
combinations you must test: sines versus sines, sines versus cosines, sines
versus constant, cosines versus cosines, and cosines versus constant.
You may need the following integrals:
$ sin (p − q) x sin (p + q) x
2 (p − q) 2 (p + q)
$ sin (p − q) x sin (p + q) x
cos px cos qx dx = + p = ±q
2 (p − q) 2 (p + q)
$ cos (p − q) x cos (p + q) x
sin px cos qx dx = − −
2 (p − q) 2 (p + q)
8.2 Show that an alternate version to equation (8.4.1) for the Fourier series
expansion is
∞
a
y (x) = 0 + An cos kn x + ϕn ,
2
n=1

2 2 − 1
bn
where An = an + bn and ϕn = tan − . (This is the version used
an
for figure 8.1.1.)
8.3 The complex version of the Fourier expansion. Frequently, instead of
thinking of the Fourier expansion in terms of sines and cosines, it is more
convenient to think of it in terms of complex exponentials. (a) Assume that
2π
the set of complex exponentials eikn x , where kn = n and n ranges from
λ
−∞ to +∞, forms a complete basis for functions y(x) with periodicity λ,
so long as we allow the expansion coefficients to be complex. (It is reasonable
to assume that this is a complete basis, since each exponential contains some
cos character and some sin character, and since we can vary the balance
between the sin and cos by varying the balance between real and imaginary
in the expansion coefficient.) This means that we can write
∞

y(x) = Cn eikn x .
n=−∞
Note: There is no “Re” in the above expression; this is intentional and

correct.
Show that the basis functions in this expansion are mutually orthogonal.
Hint: Remember that for a Fourier expansion, the inner product is defined
as an integral from 0 to λ. (b) Show that the expansion coefficients
1 $λ −ik x
are given by Cn = e n y(x) dx. Hint: you may wish to refer to the
λ0
discussion of generic orthonormal function expansions with nonnormalized
basis functions in section 8.3. (c) If y(x) is real (the usual case), then all the
imaginary parts of the expansion above must somehow cancel out. Show
in detail how this happens. Hint: Cancellation occurs between the term at
n = m and the term at n = -m.
8.4 The Fourier transform. So far, we’ve discussed how any function with
periodicity λ can be expressed as a sum of sines and cosines with the
same periodicity (plus a constant term); the difference in the wavenumber
k between the longest wavelength sinusoidal component and the next-to-
2π
longest is given by k1 = . (In fact the difference between any two
λ
“adjacent” (in k-space) sinusoidal components is k1 .) It is also possible to
express even nonperiodic functions in this way, so long as we allow this
spacing to become infinitesimal. In this problem, you’ll derive the equations
which apply in this limit.
In problem 8.3, you can show that the complex exponential version of
Fourier analysis is
∞

y(x) = Cn eikn x , (1)
n=−∞
λ
2π 1
where kn = n and Cn = e−ikn x y (x) dx .
λ λ
0
It should be fairly clear that, to evaluate the Cn ’s, it is only necessary to

integrate over one period of y(x); as long as we use the same integration
interval for all the Cn ’s, we can choose whatever interval corresponding to
one period that we like. In particular, it will make things look prettier if,
instead of integrating from 0 to λ, we integrate from −λ/2 to λ/2, that is, we
could just as well write:
λ/2
1
Cn = e−ikn x y (x) dx . (2)
λ
−λ/2
We can think of a nonperiodic y(x) as a periodic function with λ → ∞.

According to the above equation, all the Cn ’s appear to go to zero in this
limit, so we must proceed carefully. Although the Cn ’s do get smaller, the
number of them in any interval from k to k + dk gets bigger, so the net result
winds up still being OK, as you’ll show. As shown in the above equation
(on the right), each Cn depends on the value of kn = nk 1 , so instead of
writing them as Cn , we might write them as C(kn ). In the limit λ → ∞, the
spacing (in k-space) between the C’s becomes infinitesimal. In this limit, we
can write them as C(k) instead of C(kn ). Unless something pathological is
happening, all the C’s in the interval k to k + dk will have the same value, if
we’re taking the limit dk → 0. A representative one of the C’s in this interval
would have the value
λ/2
1
C (k) = e−ik x y (x) dx . (3)
λ
−λ/2
To form y(x) from these C’s, we must form an integral over k instead of the
sum shown in (1). This integral (in the limit λ → ∞) would be given by
∞
y(x) = (# of C’s in the interval k to k + dk) (value of a typical C(k)
−∞
in this range)eikx ,
where the (value of a typical C(k) in this range) would be given by (3)
(a) Show that the (# of allowed C’s in the interval k to k + dk) is given
λ
by dk. (This is almost trivial.)
2π
(b) Use this result to show that
∞ ∞
1 ikx 1
y(x) = √ Y (k)e dk , where Y (k) = √ y(x)e−ikx dx .
2π 2π
−∞ −∞
Note: Y (k) and y(x) are referred to as a “Fourier transform pair.” They contain
the same information, but one version, y(x), is expressed in regular space,
while the other, Y (k), is expressed in k-space.
8.5 What is the Fourier series representation for the triangle wave, shown in
figure 8.P.1?
Figure 8.P.1 The triangle wave.
Figure 8.P.2 a: Thomas Young 1773–1829. b: Model for two-slit apparatus. c: The Fourier
transform of the model shown in part b gives the interference pattern produced when light goes
through the two-slit apparatus.
8.6 Interference as a Fourier Transform. As you will recall from a previous

course, Thomas Young (figure 8.P.2a) showed in 1800 that light has wave
character. He did this by sending a beam of light through a pair of slits,
and observing the resulting interference pattern. We can model his two-slit
apparatus with the following function:

1 − b − a < x < −b and b < x < b + a
y (x) =
0 elsewhere
as shown in figure 8.P.2b. Surprisingly, Fourier transforms turn out to be very

handy in understanding the interference pattern. Show that the Fourier trans-
2 sin [k (a + b)] − sin (kb)
form of the above y (x) is given by Y (k) = .
π k
You are now done with the actual problem. However, it turns out that the
Fourier transform you just calculated is equal to the amplitude of the electric
field as the interference pattern impinges on the screen. Since the observed
intensity is proportional to E 2 , if we plot Y 2 , we get the interference pattern.
The plot of your Y 2 is shown in figure 8.P.2c.
You can see the short wavelength wiggles, which are due to the interference
between the two slits, and also the more slowly varying structure which is
due to the single slit diffraction pattern.
In fact, one can show that the Fourier transform of the “aperture function”
always gives the amplitude of the electric field, so that the square of the
Fourier transform gives the interference pattern.
8.7. (a) Find the Fourier transform of y = Ae−a|x| , where a > 0.
(b) Find the FWHM (full width at half maximum) of y, and set this
FWHM equal to W .
(c) Express the FWHM of the Fourier transform of y in terms of W .
8.8. In problem 8.4, you showed that we can express any function y(x) in the
form
∞
1
y(x) = √ Y (k)eikx dk . (1)
2π
−∞
You also showed that, if we write y(x) this way, then its “Fourier transform”
Y (k) can be found by
∞
1
Y (k) = √ y(x)e−ikx dx . (2)
2π
−∞
The above equations are my favorite way of writing the Fourier transform,
since they emphasize the symmetry between y(x) and Y (k). Although
about half the world uses the above way of writing the Fourier transform,
unfortunately the other half uses a slightly different formulation. In this
version, we instead express y(x) as
∞
1
y(x) = Y (k)eikx dk . (3)
2π
−∞
For this way of expressing y(x) , the expression for Y (k) (equation 2) needs to
be adjusted slightly to be correct. What would be the correct expression for
Y (k) to go with equation 3? Explain your answer thoroughly. Hint: There
is almost no additional math required for this problem. It will probably help
you to think in terms of basis functions/vectors in Hilbert space, although
there are other good ways to do this problem.
8.9 Time scaling of the Fourier transform. Consider a function y(t) with
Fourier transform Y (ω). The function y′ (t) = y (at) is the same as y(t),
except that it is compressed by the factor a along the time axis. For example,
if there is a peak in y(t) at t = 1s, and a = 2, then the same peak appears in
y′ (t) = y (at) at t = 0.5 s. Assuming a > 0 show that the Fourier transform
1 ω
of y′ (t) is Y ′ (ω) = Y . (This means that Y ′ (ω) is expanded by the
a a
factor a along the ω axis.)
8.10 Fourier transform of the Dirac delta function
(a) Explain briefly why the Kronecker delta function has the property
that it “picks out” the “matching” term in a series, that is,
∞
:
f (n) δmn = f (m), where f is a function.
n=−∞
(b) The Kronecker delta function only works for discrete variables, such
as n in the example above or kn for a finite-length string stretched
between two walls. However, it is often handy to have a similar
function which works for continuous variables. This is called the
“Dirac delta function,” and is written δ (x − a). It has the property
that for x = a, δ (x − a) = 0 (this is analogous to the Kronecker
delta function). However, for x = a, δ (x − a) = "∞", where the
infinity is in quotes because the “extent” of the infinity is carefully
defined below. However, it is correct to picture the graph of δ (x − a)
versus x as being zero everywhere except for an infinitely tall and
infinitely narrow spike at x = a. Here’s the final part of the definition
of the Dirac delta function: the height of the spike is the “right
amount of infinity” so that
∞
f (x) δ (x − a) dx = f (a) ,
−∞
which is analogous to the equation in part a) of this problem, that

is, the Dirac delta function “picks out” the “matching” term in the
integral. Find the Fourier transform of the Dirac delta function for
the case a = 0.
(c) Now that you know the answer, explain how you might have
predicted it qualitatively based on the example “Fourier Transform
of a Gaussian” in section 8.5.
(d) What is the “inverse Fourier transform” of δ k − k0 for the case
k0 = 0? (The “inverse Fourier transform”
means the function y(x)
whose Fourier transform is δ k − k0 .)
(e) When a (or k0 ) is not equal to zero, things are slightly more
complicated, though in a philosophical sense they are still about
the same. What is the inverse Fourier transform of δ k − k0 for k0
not necessarily equal to zero?
(f) Now that you know the answer to the above, explain why it makes
qualitative sense.
8.11 More about the Dirac delta function. Read the first two parts of problem
8.10. (You need not actually do them before you do this problem).√ In part b of
that problem, you can show that the Fourier transform of δ (x) is√
1/ 2π . This,
of course, means that the Fourier transform of δ (t) is also 1/ 2π . (a) Use
$∞
this as a starting point for this problem, and show that eiωt dω = 2π δ (t).
−∞
$∞
(b) Explain why, therefore, e−iωt dω = 2π δ (t).
−∞
8.12 The left column of figure 8.P.3 shows several different functions of time.
The right column shows the magnitudes of the Fourier amplitudes for these
functions, though the order has been scrambled, for example plot g does
not show the Fourier amplitudes for the function y (t) in part a. “Fourier
amplitude” means the amplitude An in the Fourier expansion expressed
Figure 8.P.3 The left column shows various functions of time, with the same vertical and
horizontal scale for each plot. The right column shows the magnitude of the Fourier amplitude
as a function of ω, with the same horizontal scale for each plot. The insets for j and k show
magnified views of the low-ω region; the horizontal scale on the two insets is the same.
Note that parts h and i each show just a single dot, near the top left.
a0 ∞
:
as (8.4.5): y (t) = + An cos ωn x + ϕn . The Fourier amplitudes are
2 n=1
plotted as a function of ω. Which entry from the right column goes with
each of the entries from the left column? Explain each of your choices
briefly.
8.13 Parseval’s theorem. We have seen that energy is proportional to the square
of amplitude. For example, the potential energy of a simple harmonic
oscillator is 12 kA2 . Therefore, it is reasonable to expect that the sums of
squares of amplitudes in the time domain are proportional to the sums of
squares of amplitudes in the frequency domain, since both sums should be
proportional to the energy. (a) We begin with a simple illustration of this
2π
idea. Let y (t) = A1 cos ω1 t + A2 cos ω2 t, where ω1 = and ω2 = 2ω1 .
T
The sum of squares of amplitudes in the frequency domain would simply
be A21 + A22 . To find the “sum of squares of amplitudes in the time domain,”
$T
we must integrate: [y (t)]2 dt. Show that this is proportional to A21 + A22 .
0
(b) In part a, we considered Fourier synthesis of a periodic function, as
described in section 8.4. Now, we consider Fourier transforms. As you’ll
recall from section 8.5, these involve complex exponentials. Therefore,
instead of discussing the “sum of squares of amplitudes,” we discuss the
“sum of squares of magnitudes.” Prove Parseval’s theorem, which states
$∞ $∞
that |y (t)|2 dt = |Y (ω)|2 dω, where y (t)and Y (ω) are a Fourier
−∞ −∞
transform pair, as defined by (8.5.4). Hint: use the result of problem 8.11 ;
since this is a mathematical result, you can interchange the symbols ω and
t and it will still be correct.
8.14 State whether each of the following is true or false. If true, explain why
briefly. If false, explain why and provide a corrected version that is not
simply a negation of the original statement. Important: assume that y (t) is real.
(a) In the Fourier transform (8.5.4), the imaginary part of Y (ω) is just
a mathematical convenience. We could just as well write
∞
1
y(t) = √ Y (ω)eiω t dω,
2π
−∞
⎡ ⎤
∞
1
where Y (ω) = Re ⎣ √ y(t)e−iω t dt ⎦ .
2π
−∞
(b) In the Fourier transform (8.5.4), the negative frequencies are just a
mathematical convenience. We could just as well write
⎡ ⎤
∞
1
y(t) = 2Re ⎣ √ Y (ω)eiω t dω⎦ ,
2π
0
∞
1
where Y (ω) = √ y(t)e−iω t dt .
2π
−∞
8.15 Aliasing. We have discussed why the maximum angular frequency that
can be represented by samples taken at time intervals spaced by
is
2π π ω N /2
ωN /2 = = . (The corresponding frequency, fc = is called the
2

2π
“Nyquist frequency.”) What happens if we sample a continuous waveform

with a higher angular frequency than this? The resulting effect is called
“aliasing,” and can cause serious problems in digital signal processing.
(a) Sketch several periods of a continuous sinusoidal waveform. Put
dots on it indicated the times at which samples are taken, with
slightly less that the period of your waveform.

(b) Based on your sketch, explain why the apparent frequency based
just on the sampled points is much lower than the actual frequency
1/Tw of your waveform.

(c) Based on your explanation from part b), for what value of
Tw
does the apparent frequency equal zero?
(d) Now, we will explore the same ideas more quantitatively. A
2π
continuous waveform A cos ωα t is sampled, where ωα = α ,
N

is the time interval between samples, and N is the number of
samples. For simplicity, we’ll assume α is an integer, but similar
things happen if it isn’t. Explain why, if ωα > ωN /2 , then peaks
appear in the Discrete Fourier Transform (DFT) at ωα and at ωN −α ,
where ωα > ωN −α > 0.
(e) Explain how your result from part d) is consistent with your result
from part c).
(f) Explain why a waveform A cos ωN −α would produce peaks at
ωα and ωN −α . (This means that DFT of the waveform with angular
frequency ωα is indistinguishable from the DFT for the waveform
with angular frequency ωN −α .)
8.16 Windowing. (Mathematica or other symbolic algebra program is required
for this problem. Further instructions for how to implement this problem in
Mathematica are available on the website for this text, under the entry for
this problem.) In this problem, you will make sets of N = 1, 024 samples
taken at intervals of
= 1 ms, so that the total sampling time is T = 1.024 s.
Recall that the Discrete Fourier Transform (DFT) algorithm assumes that the
input waveform y (t)has the periodicity T . In this problem, you’ll explore
what happens when this isn’t true.
(a) Create a dataset of 1,024 samples of the continuous wave y =
2π t
sin , with a sample interval
= 1 ms, and Tw = 10.24 ms,
Tw
so that exactly 100 periods fit into the sampling interval. Plot the
magnitude of the DFT of this dataset as a function of ω using a
logarithmic scale for the vertical axis, and comment briefly on it.
(b) Now create a new dataset of 1,024 samples, with Tw adjusted so
that 100.5 periods fit into the sampling interval. Because the DFT
assumes periodicity T , this creates a “glitch” at the border between
one interval of length T and the next, for example, at the border
between the interval t = 0 to t = T and the assumed repeat of the
function y that occurs in the interval t = T to t = 2T . Sketch by

hand what the assumed y looks like near this border.
(c) Plot the magnitude DFT of this new dataset (again with a logarithmic
scale for the vertical axis), and comment briefly on it.
(d) Now create a “windowed” version of the dataset from part b, using
the Hanning window described in figure 8.6.2. This eliminates the
“glitch,” but of course distorts the input waveform in other ways.
Again, plot the magnitude of the DFT, and comment briefly on how
this plot compares to those from parts (a) and (c).
8.17 Samples are taken of a continuous waveform at 1 ms intervals. Only N = 4
samples are taken. (a) For this part, the samples have the following values: A,
−A, A, −A. Compute by hand the Discrete Fourier Transform of this dataset,
and explain how your results fit with expectations. (b) For this part, the
samples have the following values: A, 0, -A, 0. Compute by hand the Discrete
Fourier Transform of this dataset, and explain how your results fit with
expectations.
8.18 The Heisenberg Uncertainty Principle and expectation values. As we’ve
discussed at various points in this text, the “probability density” for
a quantum mechanical particle is given by ||2 , where (x , t) is the
wavefunction. So far, I’ve only said that ||2 is proportional to the probability
of finding the particle near a particular position. The exact definition is
that the probability of finding the particle somewhere between x = a and
x = b is
$b
||2 dx
a
.
$∞
||2 dx
−∞
You might recognize the term on the bottom as a normalizing factor; the
numerator is the integrated probability density over the range of interest,
whereas the denominator is the integrated probability density over all space.
If the wavefunction is properly normalized (the same idea as normalizing an
eigenfunction), then the denominator equals 1.
In this problem, we will focus on what happens at t = 0. Let us define
ψ (x) ≡ (x , t = 0), so that the above probability of finding the particle
somewhere between x = a and x = b can be written as
$b
|ψ|2 dx
a
.
$∞
|ψ|2 dx
−∞
Now, we are equipped to understand the concept of “expectation value.”

Let’s say we measure some function of the position of the particle, f (x).
This means that if the particle is at x = a, then a measurement of f yields
f (a). If we imagine preparing a large number of particles in identical initial
states, and measuring f for each of them, then sometimes we’ll get f (a),
sometimes f (b), sometimes something else. If |ψ|2 is large near x = a and
small near x = b, we are more likely to get the result f (a) than the result
f (b). The appropriately weighted average of many measurements of f on a
large number of initially identical particles is called the “expectation value
of f ,” and is computed like this:
$∞
|ψ|2 f (x) dx
−∞
f = .
$∞ 2
|ψ| dx
−∞
For example, if f (x) = x 2 , then
$∞
% & |ψ|2 x 2 dx
−∞
x2 = .
$∞
|ψ|2 dx
−∞
The “standard deviation” of x is a measure of the width (in x) of the of the

graph of |ψ|2 versus x, and is defined as
% &

x ≡ (x − x )2 .
The quantity (x − x ) is the difference between x and the average value of

x. By squaring this, we get a number that is positive whether or not x is
greater than x . In the above equation, we take the average of this square,
then take the square root of the average, so as to get back to something that
has the same units as x . Thus,
x is the Root of the Mean of the Square of
the difference between x and x . I capitalized the words “root,” “mean,”
and “square” because you will encounter this same idea in chapter 9; at
that point it will be referred to as the “rms” amplitude. The
x defined
in this way is the “more careful definition” of the width that was referred
to in our initial discussion of the Heisenberg uncertainty principle, back in
section 1.12.
2 2
(a) Assume that ψ = Ae−x /2σ , where A and σ are constants. Show
σ
that
x = √ . Hint: because ψ is symmetrical about x = 0, we
2
have right away that x = 0.
(b) We can write any wavefunction ψ as a Fourier sum: ψ (x) =
1 $∞
√ Y (k)eik x dx . The Fourier transform Y (k) plays the same
2π −∞
role in k-space that ψ (x) plays in real space. For example, the
probability of measuring a k for the particle that lies between
ka and kb is
$kb
|Y |2 dk
ka
.
$∞
|Y |2 dk
−∞
We can define the width (in k) of the graph of |Y |2 versus k:

% &

k ≡ (k − k )2 ,
where, for example,

$∞
% & |Y |2 k 2 dk
−∞
k2 = .
$∞
|Y |2 dk
−∞
1
For the ψ assumed in part (a), show that
k = √ .
σ 2
1
(c) Combine the results of parts (a) and (b) to show that
x
k = ,
2
as claimed in section 1.12.
(d) For a quantum mechanical particle, the momentum is given by p =
h̄k, where h̄ is Planck’s constant. Show that therefore, for the ψ
h̄
assumed in part (b),
x
p = . (In fact, the ψ assumed in part
2
(b) gives the minimum value for
x
p, so that in general we have
h̄

x
p ≥ , which is Heisenberg’s uncertainty relation.)
2
9 Traveling Waves
Only for one does my heart resonate

Like a column of air in the wind
I seek the one whose frequency suits me
To release the sound within.
— Anonymous
9.1 Introduction
So far, we’ve examined oscillations limited to a finite region of space, such as the
standing waves on a rope stretched between two walls. Such situations are analogous
to the quantum mechanical problem of an electron confined to a finite region of space.
For example, the electron might be confined by its attraction to a nucleus, as suggested
in figure 9.1.1. These “bound states” will make up at least 75% of your work in an
introductory quantum mechanics course, and will lead, for example, to the structure
of the periodic table. However, we can also make waves that move.
One of the joys of physics is in discovering hidden connections. In this chapter, we
will explore electromagnetic waves (such as light) in vacuum and in matter, waves on
ropes, sound waves, and waves on transmission lines. We will find that the mathematical
structure for all of these waves is identical, once we find the appropriate variables to
study. In each case, we will find that there are two components to the wave; in the
example of electromagnetic waves in vacuum, these are the oscillating electric and
magnetic fields.
9.2 The wave equation
Let’s begin by looking at traveling waves on a string, because they’re easy to visualize.
(Later, we’ll examine other types of traveling waves, such as electromagnetic waves.)
We begin with the beaded string. Recall from section 7.1 that bead j experiences tension
forces from the the string connecting to bead j − 1 (on the left) and from the string
280
Chapter 9 ■ Traveling Waves 281
Figure 9.1.1 Standing waves on a rope (top) are

analogous to states of an electron for which it is
confined, such as that shown in the bottom part of
the figure; these are called “bound states.”
connecting to bead j + 1 (on the right, leading to
T
(7.1.2) : ÿj = − −yj−1 + 2yj − yj+1 ,
ma
where yj is the y-position of bead j, T is the tension in the string, m is the mass of
a bead, and a is the spacing between beads. We are interested in a continuous string,
so we allow the spacing of the beads to become very small. We can then think of a
continuous function y(x , t) which describes the string. For notational convenience, we
will stop bothering to indicate explicitly that y is a function of t as well as x, so that
we simply write yj = y(xj ). Since a is small, we can approximate yj−1 and yj+1 using
a second-order Taylor series:

∼
∂ y a2 ∂ 2 y
yj−1 = y xj − a = y xj − a + , (9.2.1a)
∂ x xj 2 ∂ x 2 xj

∂ y a2 ∂ 2 y
yj+1 = y xj + a ∼= y xj + a + . (9.2.1b)
∂ x xj
2 ∂ x 2 xj
∂y
The notation means “partial derivative of y with respect to x .” This simply means
∂x
“take the derivative of y with respect to x, treating t as a constant.” For example,
∂y ∂ 2y
if y = ax 2 t 3 , then = 2axt 3 . Similarly, the notation means “second partial
∂x ∂ x2
derivative of y with respect to x,” which means “take the second derivative of y with
∂ 2y
respect to x, treating t as a constant.” For example, if y = ax 2 t 3 , then 2 = 2at 3 . You
∂x
should not get too worried about this notation; it is needed because y is a function of
both x and t, but there is really nothing mysterious about it.
Your turn: Substitute equation (9.2.1) into 7.1.2 above, and show that the result is
Ta ∂ 2 y
ÿ xj ∼ = . (9.2.2)
m ∂ x2 xj
In the limit that a → 0, the second-order Taylor series approximation becomes

exact, so that the equality above becomes exact. As we take this limit, we imagine
m
simultaneously shrinking the mass of the beads, so that the mass per unit μ ≡
" 2 # a
∂ y T ∂2 y
remains constant. We can then rewrite equation (9.2.2) as1 = , where
∂ t2 μ ∂ x 2 xj
the subscript in square brackets indicates that both sides are evaluated at x = xj . Since
we can do this same analysis for any bead j, and since the beads are infinitesimally
close together, we see that the equation holds for any x, so we may as well write it
just as
∂ 2y T ∂ 2y
= . (9.2.3)
∂ t2 μ ∂ x2
T m 2
Your turn: Show that the units of are .
μ s
Given the above, we can define something with the units of velocity:

T
vp ≡ (9.2.4)
μ
(We’ll soon see just what this is the velocity of.) With this, we can rewrite
equation (9.2.3) as
∂2 y ∂2 y
2
= vp2 . (9.2.5)
∂t ∂ x2
The Wave Equation
This is the “wave equation.” It says that the second derivative of y with respect to t
is the same as the second derivative with respect to x, except for the additional factor
of vp2 .

Claim: Any function of the form y x − vp t , that is, a function that has as its argument

the factor x − vp t is a solution to the wave equation.
∂ 2y
1. Note that ÿ = , meaning the second partial derivative of y with respect to t. Again, this
∂ t2
notation is nothing to worry about; it simply means, “take the second derivative of y with respect
∂ 2y
to t, treating x as a constant.” For example, if y = ax 2 t 3 , then 2 = 6ax 2 t.
∂t

Figure 9.2.1 The function y x − vp t represents a rigid shape moving to the right at speed vp .
b
Demonstration by example: Let’s try y = x − vp t . To test whether this works, we
need to evaluate the derivatives:
∂2 b ∂ " ∂ b # ∂ " b−1 # b−2

x − vp t = x − vp t = b x − vp t = b (b − 1) x − vp t
∂ x2 ∂x ∂x ∂x
and
" b # " #
∂2 b ∂ ∂ ∂ b−1
x − vp t = x − vp t = b x − v p t − vp
∂ t2 ∂t ∂t ∂t
b−2
= b (b − 1) x − vp t vp2 .
b
Comparing these, we see that y = x − vp t is indeed a solution to the wave equation.

In fact, we can now see why any function of the form y x − vp t is a solution: the
process of taking a derivative with respect to t is identical to taking one with respect
to x,
except
that each t derivative also brings out (because of the chain rule) a factor
of −vp , so that taking two time derivatives brings out the extra factor vp2 , which is
just what we need for a solution to the wave equation.

Your turn: Show that any function of the form y x + vp t is also a solution to the
wave equation.
You know from your study of functions in high school that the function y(x − b),
where b is a constant, looks just the same
as y(x),
but shifted to the right by the amount
b. Therefore, a function of the form y x − vp t looks just the same as y(x), but shifted
right by the amount vp t. Since this shift increases linearly in time, at the rate vp , we

see that y x − vp t represents a rigid shape moving to the right at speed vp , as shown

in figure 9.2.1. Similarly, a function of the form y x + vp t represents a rigid shape
moving left at speed vp .

Conclusion: The solutions to the wave equation are of the form y x − vp t (representing

a rigid shape moving right at speed vp ) or of the form y x + vp t (representing a rigid
shape moving left at speed vp ).
So, we see that the string can indeed sustain traveling waves, with speed vp =

T
. Note that each infinitesimal segment of the string moves straight up and down
μ
(in the + or −y direction) as the wave passes, even though the wave is moving right
or left (in the + or −x direction).
There is something quite remarkable about the conclusion in the box above: the
speed of the wave, vp , doesn’t depend on the shape. In particular, for sinusoidal waves
the speed doesn’t depend on the wavelength or the amplitude!
9.3 Traveling sinusoidal waves
The most important example of a traveling wave is a sinusoid; We know from Fourier
analysis that any function can be formed by summing up sinusoids. Recall from
2π
chapter 7 that a sine wave of wavenumber k = (where λ is the wavelength) and
λ
amplitude
A is written A sin kx. To make this travel to the right, we simply replace x
by x − vp t :

y = A sin k x − vp t = A sin kx − kvp t . (9.3.1)
2π
When t advances from 0 to , the argument of the sin increases by 2π . Therefore,
k vp
2π 2π
the period of the wave is T = . Since the angular frequency is given by ω ≡ ,
k vp T
we see that ω = kvp ⇔
ω
vp = . (9.3.2)
k
This is really nothing new. You’ve known since you were a toddler that the speed
λ
of a sinusoidal wave is given by v = . We can reexpress this in terms of k and ω:
T
λ 2π/k ω
v= = = .
T 2π/ω k
Using equation (9.3.2), we can rewrite equation (9.3.1), to express a right-traveling
sinusoidal wave as
y = A sin (k x − ω t) ,
which is the most common way of writing it. The speed of the wave is the speed at
◦
which the crest (corresponding to a phase of 90 ) or the trough (corresponding to a
◦
phase of 270 ) advances, so it is called the “phase velocity,” hence the use of “p” as a
subscript in vp .
9.4 The superposition principle for traveling waves
In section 4.4, we found that when an oscillator is driven by multiple drive forces,
the resulting response is simply the sum of the responses to each drive force on its
own. This is a consequence of the linearity of the differential equation governing the
oscillator. In this section, we explore a related consequence of linearity, but this time
for a system that is not driven. In particular, we will explore whether we can combine
left- and right-moving waves, and what happens when they “collide.”
Claim: For any differential equation that is linear (i.e., there are no terms proportional
to y2 or to y ẏ, etc.) and homogeneous (i.e., there are no terms that are constant), if yA
is a solution to the differential equation, and yB is a different solution, then the linear
combination AyA + ByB (where Aand B are constants) is also a solution.
Proof for the wave equation: To say that yA is a solution to the wave equation simply
means that
∂ 2 yA 2
2 ∂ yA
= vp . (9.4.1)
∂ t2 ∂ x2
Similarly, to say that yB is a solution simply means that
∂ 2 yB ∂ 2 yB
2
= vp2 . (9.4.2)
∂t ∂ x2
Now, we multiply equation (9.4.1) by A, multiply equation (9.4.2) by B and add them
together:

∂ 2 AyA + ByB

2
∂ 2 yA ∂ 2 yB 2 ∂ yA ∂ 2 yB 2 ∂
2
A 2 + B 2 = vp A 2 + B 2 ⇔ 2
= vp 2
AyA + ByB ,
∂t ∂t ∂x ∂x ∂t ∂x
which shows that AyA + ByB is a solution to the wave equation. (Perhaps you can see
how you could do this same type of proof for any linear, homogeneous differential
equation.)
This principle of superposing (i.e., adding together) solutions is a very powerful
one, and one that you will use a lot in quantum mechanics, which is governed by the
Schrodinger equation
h̄2 ∂ 2 ∂
− 2
+ U = ih̄ , (9.4.3)
2m ∂ x ∂t
where m is the mass of the electron, U is the potential energy of the electron, and
(x , t) is the electron wavefunction, which is analogous to y (x , t). You can see that
this differential equation is linear (i.e., it contains no terms proportional to 2 or to
˙ , etc.) and homogeneous (i.e., it contains no constant terms), so it must obey the
Principle of Superposition.
As applied to waves that obey the wave equation, the Principle means, for example,
that we can have left- and right-traveling waves propagating at the same time. As they
move into the same area, they simply add up, as shown in figure 9.4.1, but they don’t
alter each other in any profound way – each just keeps going, and eventually they pass
through each other with no change!
Figure 9.4.1 For waves governed by a linear

differential equation, left- and right-traveling
waves simply add together, and pass through each
other without other effect; this is the Principle of
Superposition.
Recall from chapter 7 that when a string is stretched between two walls, that is,
when we impose the boundary conditions that y = 0 at x = 0 and x = L, the solutions
are standing waves. The basic physics of the string that we used in chapter 7 are exactly
the same as the physics we used to derive the wave equation. We know that the left- and
right-traveling wave solutions are the only solutions to the wave equation, so it must be
possible to express standing waves as a superposition of traveling waves. For example,
a pure normal mode n of a string between two walls is
y = Cn sin kn x cos ωn t . (9.4.4)
(Note how, for a standing wave such as this, the time and space dependence appear
in the arguments of two different functions, whereas for a traveling wave they appear
together in the argument of one function.) Can we express this as a superposition of
traveling waves?
Your turn: Use the basic trig identity sin (A + B) = sin A cos B + cos A sin B to show that
Cn C
sin kn x + ωn t + n sin kn x − ωn t = Cn sin kn x cos ωn t ,
2 2
that is, that the sum of two equal-amplitude waves traveling in opposite directions
produces a standing wave!
This result is important in a number of contexts. For example, in solid-state physics

the “energy gaps” which are responsible for the electronic properties of semiconductors
arise as the result of a standing wave which forms when electron waves of certain
special wavelengths undergo diffraction from the lattice of atoms in the semiconductor
crystal, creating a “backscattered” wave which travels in the opposite direction from
the original, and has the same amplitude.
Note that standing waves are a special case, because of the imposition of the
boundary conditions. Therefore, although we can superpose the more general solutions
(traveling waves) to create the special case solution (standing waves), we can’t
go the other way, that is, we can’t superpose standing waves to create traveling
waves.
9.5 Electromagnetic waves in vacuum
The phenomena of electricity and magnetism are governed by Maxwell’s four

equations:
A
Qnet, enclosed
Gauss’s Law: E · n̂ dA = (9.5.1)
ε0
“Electric field lines can only begin on + charges and can only end on − charges.
1 Q
The electric field due to a point charge is given by E = ”
4π ε0 r 2
A
Gauss’s Law for magnetic fields: B · n̂ dA = 0 (9.5.2)
“There are no magnetic monopoles.”

A
⇀ d
Faraday’s Law: E · d ℓ = − B · n̂ dA (9.5.3)
dt
“A time-varying B creates a spatially varying E.”
A
⇀ d
Ampère’s Law: B · d ℓ = μ0 I +μ0 ε0 E · n̂ dA (9.5.4)
net dt
threading
“Magnetic fields are created by currents or by time-varying E.”
Maxwell’s equations restrict which combinations of electric fields, magnetic fields,

and charge are allowed. For example, the magnetic field pattern shown in figure 9.5.1
is not allowed by equation (9.5.2). Let us ask whether “plane waves” are allowed in
Figure 9.5.1 This magnetic field is not consistent with

equation (9.5.2).
vacuum, that is, do Maxwell’s equations allow the following combination of electric
and magnetic fields in vacuum:
E = E (x , t) ĵ and B = B (x , t) k̂, (9.5.5)
where ĵ is the unit vector in the y-direction and k̂ is the unit vector in the z-direction.
The configuration described by equation (9.5.5) is called a “plane wave” because there
is no dependence of the electric or magnetic field strength on y or z. Therefore, in any
plane perpendicular to the x-axis, the E and B are uniform (because x has the same
value throughout the plane). The electric field component of one particular type of
plane wave is illustrated in figure 9.5.2, using the convention that the length of the
vector indicates the strength of the field. (Note: the planes shown could be infinite in
extent, but to conserve paper we have only shown a portion of each plane.)
It’s pretty clear that the plane wave does satisfy both the electric and magnetic
versions of Gauss’s Law, since the flux of either E or B would be zero through any
closed surface (as required, since there are no enclosed charges in a vacuum). Next, let’s
check whether this combination of electric and magnetic fields satisfies Faraday’s Law.
We apply Faraday’s Law to the rectangular loop shown in figure 9.5.3. We’ll
⇀
choose to have d ℓ point counterclockwise around this loop when evaluating the left
B ⇀
side of Faraday’s law: E · d ℓ . The closed loop integral is equal to the sum of the four
integrals along the four segments that make up the path:
A
⇀
E · d ℓ = + + + = 0 + E x2
y + 0 − E x1
y, (9.5.6)
1 2 3 4
⇀
where the 0’s arise because E is perpendicular to d ℓ along segments 1 and 3, and the
⇀
minus sign arises because E is anti-parallel to d ℓ along segment 4. We can
simplify
this by taking the limit
x → 0, and Taylor expanding E (x) around E x1 :
∂E
E x2 ∼
= E x1 +
x .
∂x
Figure 9.5.2 A plane wave in the electric field; the field strength and direction depends on x,
but not on y or z.
Figure 9.5.3 Geometry for

applying Faraday’s Law to the
plane wave (9.5.5).

A
⇀ ∂E
E · dℓ ∼
=
x
y . (9.5.7)
∂x
The equality becomes exact in the limit

x → 0.
d $
Now we turn our attention to the right side of Faraday’s Law: − B · n̂ dA. The
dt
⇀
direction of n̂ is determined by our choice of the directionality of d ℓ : curl the fingers
⇀
of your right hand in the direction of d ℓ , and your thumb points in the direction of
n̂, along k̂ in this case. The form of the magnetic field we’re investigating is (9.5.5):
B = B (x , t) k̂, therefore B · n̂ = B (x , t). Since
x is infinitesimal, and since B doesn’t
depend on y, B is uniform over the area bounded by the loop. Therefore, B · n̂ =
B
x
y. Plugging this and equation (9.5.7) into Faraday’s Law (9.5.3) gives
∂E ∂B

x
y = −
x
y ⇒
∂x ∂t
∂E ∂B
=− . (9.5.8)
∂x ∂t
So, fields of the form E = E (x , t) ŷ and B = B (x , t) ẑ are allowed by Faraday’s law,

so long as E and B are related as shown by equation (9.5.8).
There is one more Maxwell equation that we must check, Ampère’s law:
A
⇀ d
B · d ℓ = μ0 Inet +μ0 ε0 E · n̂ dA.
threading
dt
Since we are considering fields in a vacuum, Inet = 0. The reasoning following

threading
from this is almost exactly the same as the reasoning for Faraday’s law. You can show
in problem 9.6 that the result is
∂B ∂E
= −μ0 ε0 . (9.5.9)
∂x ∂t
∂
Let’s combine equations (9.5.9) with (9.5.8). We begin by taking of both sides of
∂x
equation (9.5.8), giving
∂ 2E ∂ ∂B ∂ ∂B
2
=− =− .
∂ x ∂x ∂t ∂t ∂x
∂B
Now, we use equation (9.5.9) to substitute for , giving
∂x

∂ 2E ∂ ∂E
= − −μ ε
0 0 ⇒
∂ 2x ∂t ∂t
∂ 2E ∂ 2E
= μ ε
0 0 .
∂ 2x ∂ t2
∂2 y 2
∂2 y
This is the wave equation, (9.2.5): = vp , with E playing the role of y, and
∂ t2 ∂ x2
phase velocity
1
c= √ (9.5.10)
ε0 μ0
If you plug in the numbers on this, you find a speed of 2.998 × 108 m/s, exactly equal
to the speed of light!
What about the magnetic field? You can make any wave by adding up sinusoids.
So, for simplicity, let’s consider
E = E0 sin k (x − ct) ĵ.
Since the phase velocity c = ω/k, we can rewrite this as E = E ĵ with
E = E0 sin (kx − ωt) . (9.5.11)

∂E ∂B
Plugging this into equation (9.5.8): =− gives
∂x ∂t
integrate with
∂B respect to t k
= −E0 k cos(kx − ωt) −−−−−−−→ B = E0 sin (kx − ωt) + const.
∂t ω
Because we are interested in waves, we’ll set the constant equal to zero. (The presence
of the constant means that we’re allowed to have a uniform, constant magnetic field
in addition to the electromagnetic wave.) Since c = ω/k, this gives
1 E
B = E0 sin (kx − ωt) = ⇒
c c
E = cB. (9.5.12)
Relation between amplitudes of E and B for an electromagnetic wave in vacuum
Because of this direct proportionality, we see that the electric and magnetic components
of the wave are in phase. This means that the magnetic field also propagates as a wave,
moving at speed c.
We have just shown that plane waves are allowed by all four of Maxwell’s
equations, but only if they travel at speed c and have E = cB. These electromagnetic
waves (which include light, radio waves, X-rays, microwaves, gamma rays, etc.)
consist of a self-sustaining oscillation, in which the changing magnetic field creates
a changing electric field, which in turn creates a changing magnetic field, etc. The
whole thing can only keep going if it travels at exactly c. The realization that the
speed of light is a consequence of Maxwell’s equations led directly to Einstein’s
theory of special relativity: If we assume that the laws of physics are the same in
all reference frames traveling at constant velocity, then Maxwell’s equations must
apply equally well in all such frames, and the speed of light must be exactly the same
in all such frames. This is the basic postulate of special relativity, and immediately
leads to such counterintuitive results as the speed of a light beam emanating from the
flash light being exactly the same as perceived in the frame in which the flashlight
is at rest as it is in a frame moving at speed 0.999 c in the direction the beam is
traveling!
Your turn: On the previous page, we showed that E (x , t) = cB (x , t) for a wave moving
in the positive x-direction, meaning that the E and B waves are in phase. Now, show that
E (x , t) = −cB (x , t) for a wave moving along the negative x-direction, meaning that the
E and B waves are out of phase.
For any electromagnetic (em) wave, the propagation direction is given by the
“Poynting vector”:
1
S= E × B. (9.5.13)
μ0
(The fortuitously named John Henry Poynting was a student of Maxwell.) In
section 9.10, we’ll see that this vector not only points in the direction of prop-
agation, but also has a magnitude equal to the intensity (power per area) of
the wave.
At the risk of being repetitive, let us reinforce the true nature of a plane wave,
the type of wave we’ve been discussing. Recall that we began with the assumption
(9.5.5) that E and B depend on x, but not on y or z. Some students misinterpret
this to mean that the wave exists only as a ray along the x-axis. This incorrect
thinking is reinforced by the usual graphical way of portraying a plane wave, shown
in figure 9.5.4a. This shows the magnitudes of E and B for points along the x-axis,
using the convention that the length of the vector represents the strength of the field.
However, this picture must not be taken to mean that E and B are zero at points
that aren’t on the x-axis. Instead, since E and B depend only on x, and not on y
or z, the picture would be the same for any line parallel to the x-axis, as shown
in figure 9.5.4b. Figure 9.5.5 shows the correspondence between the two ways of
portraying the strength of the electric field. In the foreground, the strength is indicated
by the spacing between field lines; this emphasizes that the field extends throughout
space, and is has the same strength and direction throughout any plane perpendicular
to the x-axis. In the background, the strength of the field is indicated by the length of the
vectors.
Figure 9.5.4 a: The conventional way of
portraying an electromagnetic wave,
showing the magnitudes of the electric and
magnetic fields along the x-axis. b: Another
portrayal, emphasizing that the magnitudes
of the fields are equal at equivalent points
along any line parallel to the x-axis.
Figure 9.5.5 Two ways of portraying a plane wave of the electric field.
292
Key equations for isomorphisms. Over the rest of this chapter, we will develop
isomorphisms (exact analogies) between em waves in vacuum and various other types
of waves. The essential relations are:
∂E ∂B ∂B ∂E
(9.5.8) : =− and (9.5.9) : = −μ0 ε0 .
∂x ∂t ∂x ∂t
All the other results follow from these two relations, including the fact that the two
components of the wave (in this case E and B) each obey the wave equation, the
relation between the magnitude of the two components (E = cB), and the speed of
1
the wave c = √ . So, if we can find equations relating the two components of
ε0 μ0
a wave that are isomorphic to equations (9.5.8) and (9.5.9), then there is a complete
1
isomorphism. It will sometimes make things a bit easier to substitute c = √ into
ε0 μ 0
equation (9.5.9), so that the two essential relations become
∂E ∂B
(9.5.8) : =−
∂x ∂t
∂B 1 ∂E
and =− 2 . (9.5.14)
∂x c ∂t
Isomorphism with rope waves. As a guide to developing the isomorphism between
rope waves and em waves in vacuum, we refer back to the isomorphism between the
mechanical oscillator and the electrical oscillator. From section 1.5, we have:
Mass and spring Electrical oscillator

Mass m Inductance L
Spring constant k Inverse capacitance 1/C
For the mechanical oscillator, the oscillations are a result of the restoring force
which tends to bring the system back toward equilibrium and the momentum which
tends to make the system overshoot the equilibrium point. The restoring force is
associated with the spring constant k which is isomorphic to 1/C. The capacitor is
associated with electric fields. Therefore, we may expect some type of connection
between the restoring force in a rope and the E for the em wave. The momentum
is associated with the mass m which is isomorphic to L. The inductor is associated
with magnetic fields. Therefore, we may expect some type of connection between the
momentum of a rope wave and the B for the em wave.
Now, let’s get more quantitative. For a rope wave, we speak of the mass per length,
μ, rather than simply the mass (which would be infinite for an infinitely long rope).
∂y
Thus, the transverse momentum per unit length is given by μ . For simplicity, we
∂t
∂y
will refer to this as the “momentum wave,” and use the symbol py ≡ μ , bearing in
∂t
mind that this is really the transverse momentum per unit length. We will hope that
this is isomorphic to B, and search for the quantity (related to the restoring force) that
is isomorphic to E.
Figure 9.5.6 Geometry needed for calculating the force on bead j.
For a right-traveling em wave in vacuum, E and B are in phase, so the quantity

we are searching for in the rope case should be in phase with the momentum wave.
From equation (9.2.2), which describes waves on a beaded rope, we have that the
Ta ∂ 2 y
acceleration of bead j is given by ÿ xj ∼
= , where T is the tension, a is
m ∂ x2 xj
the spacing between beads, and m is the mass of a bead. Therefore, the total force in
∂ 2y
the y-direction on the bead, which equals mÿ, is proportional to 2 . If we consider a
∂x
∂ 2y 2
sinusoidal wave y = A cos(kx − ωt), we have that 2 = −Ak cos (kx − ωt), which
∂x
∂y
is not in phase with the momentum wave py = μ = μAω sin (kx − ωt).
∂t
However, from figure 7.1.1, reproduced here as figure 9.5.6, we see that the y-
component of the force exerted from the left side on bead j is given by
yj − yj−1 T
−T sin θL = −T ∼
= − y − y ,
j j −1
2 a
2
a + yj − yj−1
where the last step follows because we assume a ≫ yj − yj−1 . We denote this force as
T
FL = − yj − yj−1 .
a
FL ≡ the y-component of force exerted on a bead or piece of rope from the left
(We will be using this quantity quite a bit in the rest of this section and in chapter 10,
so please make sure you understand this definition.)
Since a =
x, we then have

y
FL = −T .

x
In the limit of a continuous rope (for which we decrease a and m toward zero while
m
keeping the mass per length, μ = , constant), this becomes
a
∂y
FL = −T , (9.5.15)
∂x
where now we interpret this as the y-component of the force exerted from the left side
on a tiny segment of the rope. To check whether this is in phase with the momentum
wave, we again consider the sinusoidal wave y = A cos(kx − ωt). For this, FL =
∂y ∂y
−T = TAk sin (kx − ωt), which is in phase with the momentum wave py = μ =
∂x ∂t
μAω sin (kx − ωt).
It may seem odd that the y-component of the force exerted from the left should be
so important. However, if you hold the left end of a long rope with your hand, you must
exert a y-component of force (from the left) on this end to start a wave propagating
down the rope. Saying the same thing in a different way: for right-traveling waves, it is
the y-component of force exerted from the left which is responsible for the propagation
of a wavefront to the right on a rope that is initially at equilibrium; the force from the
left is what pulls an initially quiescent piece of rope away from equilibrium. We will
see in chapter 10 that this idea of the y-component of force exerted from the left is quite
important for the behavior of right-traveling waves when they encounter an interface
between two ropes with different values of μ.
∂y
So, are the momentum wave py ≡ μ and the y-force-from-the-left wave
∂t
∂y
F L = −T related in the same ways as B and E? From equation (9.2.2), we have for
∂x
the beaded rope

∂ 2 y xj 2
Ta ∂ y
2
∂ y

m ∂ 2y x
j ∂
"
∂y
# "
∂ m ∂y
#
= ⇔T = ⇒− −T = ⇒
∂ t2 m ∂ x 2 xj ∂ x 2 xj a ∂ t 2 ∂x ∂x ∂t a ∂t
" #
∂ ∂ ∂ y isomorphicwith ∂E ∂B
FL = − μ ←−−−−−−−−→(9.5.8) : =− .
∂x ∂t ∂t ∂x ∂t
∂y
We see that, in the isomorphism for this equation, FL plays the role of E, and py ≡ μ
∂t
plays the role of B, as expected.
To complete the isomorphism between rope waves and em waves in a vacuum,
we need to find the equation for a rope that is isomorphic with equation (9.5.14):
∂B 1 ∂E
=− 2 . Because the order of derivatives doesn’t matter, we have
∂x c ∂t
∂ ∂y ∂ ∂y
= .
∂x ∂t ∂t ∂x

T T
From equation (9.2.4), vp ≡ ⇒ μ = 2 . Combining this with the above gives
μ vp
" # " #
∂ ∂y T ∂ ∂y ∂ ∂y 1 ∂ ∂y
μ = 2 ⇒ μ =− 2 −T ⇒
∂x ∂t vp ∂ t ∂ x ∂x ∂t vp ∂ t ∂x
∂ 1 ∂ isomorphic with ∂B 1 ∂E
p = − 2 FL←−−−−−−−−→(9.5.14) : =− 2 .
∂x y vp ∂ t ∂x c ∂t
So, there is a complete isomorphism between rope waves and em waves in a vacuum.
T
Therefore, there is a isomorphism between the phase velocity for rope waves, vp = ,
μ
Table 9.5.2. Isomorphism between rope waves and electromagnetic waves in a vacuum
Rope waves Electromagnetic waves in vacuum
Transverse velocity ẏ B/μ0

y-component of force from the left Electric field E
∂y
FL = −T
∂x
Mass per length μ Permeability of free space μ0
Tension T Inverse of permittivity of free space: 1/ε0
1
and the phase velocity for em waves in a vacuum, c = √ . However, it is not clear
μ0 ε0
whether we can push things further than this. For example, is T isomorphic to 1/μ0 and
μ isomorphic to ε0 , or instead is T isomorphic to 1/ε0 and μ isomorphic to μ0 ? Since
μ is associated with momentum (which is isomorphic to B) and μ0 is associated with
B, one might expect that μ might be isomorphic to μ0 . Further, since T is associated
with the force from the left (which is isomorphic to E ) and ε0 is associated with E,
one might expect that T is isomorphic to 1/ε0 . In problem 9.5, you can show that these
∂y
are indeed the correct isomorphisms. Since py ≡ μ is isomorphic to B and μ is
∂t
∂y
isomorphic to μ0 , we see that ẏ = is isomorphic to B μ0 , and this version of the
∂t
isomorphism turns out to be somewhat more useful.
The remaining isomorphisms with other types of waves will take a lot less effort
to develop!
Self-test: Use this isomorphism and E = cB first to show that the FL wave is in phase
∂y ∂y
with the ẏ wave, and then to show that = −vp for a rope wave. (You should find
∂t ∂x
this helpful for problem 9.2. You can derive this expression using a different method
in problem 9.1.)
Since E = cB for right-moving em waves in vacuum, the E and B waves are in

phase. By the isomorphism, this means that the FL and ẏ waves are in phase for right-
moving rope waves. For left-moving waves, E = −cB (as you showed earlier in this
section), so E and B are out of phase. By the isomorphism, this means that the FL and
ẏ waves are out of phase for left-moving rope waves.
9.6 Electromagnetic waves in matter
A full description of electric and magnetic fields in matter is well beyond this text.
For a deeper understanding, you should consult an introductory electrodynamics text.2
2. The best treatment at the undergraduate level is Introduction to Electrodynamics, 3rd Ed.,
by David J. Griffiths (Prentice-Hall, Upper Saddle River, NJ, 1999).
Figure 9.6.1 Part a: when electric dipoles align

with a field, the field they create opposes the
applied field. Part b: when magnetic dipoles align
with a field, the field they create enhances the
applied field.
What follows is just the basics for simple geometries and the most common
materials.
When the space between the plates of a parallel plate capacitor is filled with an
insulating material, the atoms and molecules within the material become polarized
by the electric field; each atom or molecule becomes an electric dipole, as shown
schematically in figure 9.6.1a. (Note that the polarization is caused not only by the
field due to the plates, but instead by the combined field of the plates and all the other
dipoles in the material.) If we consider a line of such dipoles within the material, we
see that the electric field due to the dipoles opposes the electric field due to the plates,
so that the total field between the plates is reduced. The factor of reduction is called
the dielectric constant:
Eplates
κ= . (9.6.1)
Etotal
The parallel plate capacitor provides a particularly simple geometry; in a more
complicated geometry, the relationship between the total field and the field due to
the “free charge” (the charge on the plates) is more complicated. For almost all
materials, the dielectric constant really is constant, meaning that it doesn’t depend
on the strength of the electric field (at least until the field gets quite strong). Another
way of saying this is that the degree of polarization is linearly proportional to the
total electric field. We will focus exclusively on these “linear materials,” though
you should be aware that there are some materials for which the behavior is more
complicated.
An insulating material is also called a “dielectric,” derived from the Greek “dia”
meaning “through,” therefore indicating that an electric field can penetrate through a
dielectric. This is a bit of a misnomer, since the dielectric constant for typical insulating
solids is from about 2–8, meaning that the field in a parallel plate capacitor is reduced
by a factor of 2–8 by the presence of the dielectric. We define the permittivity of the
dielectric material to be
ε = κε0 . (9.6.2)
Thus, ε is always greater than or equal to ε0 .

Similarly, when the space inside a solenoid is filled with matter, the magnetic
dipoles associated with the spin of the electrons and with the orbital motion of
the electrons are affected by the magnetic field. The effect on the electron spins
is shown schematically in figure 9.6.1b. Again, the dipoles associated with the
electron spin align “with” the field of the solenoid coil (meaning that the north pole
of each dipole is on the right, and the south pole on the left), but for magnetic
dipoles the field due to the dipole within the dipole points from south to north,
in the same direction as the field from the coil. (This is opposite to what happens
for an electric dipole; inside the dipole, the field due to the dipole points from
positive to negative.) Therefore, the field of the spin dipoles enhances the field due
to the coils somewhat. A material with this type of behavior is therefore called
“paramagnetic,” from the Greek “para” meaning “alongside.” However, in other
materials, the magnetic response is dominated by the interaction with the orbital motion
of the electron, which is more complicated. The net effect of this interaction is to make
the magnetic moment associated with the orbital motion align opposite the field of
the coils. Thus, if the solenoid is filled with a material dominated by this type of
response, the total field is smaller than without the material. Such materials are called
“diamagnetic.” Again, we restrict ourselves to linear materials for which the degree
of polarization is proportional to the total field. We define the permeability of the
material to be

μ = μ0 1 + χm , (9.6.3)
where χm (Greek letter “chi”-sub-m) is the “magnetic susceptibility.” For diamagnetic

materials, the susceptibility is negative, and typically ranges from 10−6 to 10−5 . For
paramagnetic materials, the susceptibility is positive, and typically ranges from 10−5
to 10−4 . In any case, the effect on the total field is quite small, in strong contrast to the
case for dielectrics. (Note that we are not discussing ferromagnetic materials such as
iron, which dramatically affect the magnetic field.) Thus, for linear materials, μ is close
to μ0 ; it is slightly less for diamagnetic materials, and slightly more for paramagnetic
materials.
It is convenient to define a new field called “H”:
1
H= B. (9.6.4)
μ
Even for linear materials there is no simple physical interpretation for H other than
what is evident from the above equation. Within the material, it is still the magnetic field
B that exerts the force F = qv × B on moving charges. However, H is calculationally
convenient. (It turns out that, for symmetrical geometries, you can calculate H just
from the currents in coils, without referring to the complexities of the diamagnetic or
paramagnetic materials.)
The standard form of Maxwell’s equations is still correct within any material,
whether linear or not. However, it is sometimes convenient to separate the effects of the
“free charges” (such as the charges on capacitor plates and the charges flowing through
a solenoid coil) from those of “bound charges” (the charges which rearrange slightly
when atoms and molecules are polarized by an electric field) and “bound currents” (the
currents associated with electron spin and orbital motion). To create the alternate set of
Maxwell’s equations which is valid in a linear material, we simply replace ε0 by ε, μ0

by μ, and all charges by just the free charges. The two equations we care about most
are Faraday’s law and Ampère’s Law. Applying the recipe to Ampère’s Law, we get
A
⇀ d
B · d ℓ = μ0 Inet + μ0 ε0 E · n̂ dA →
threading
dt
!
Ampere’s Law for any circumstance
A
⇀ d
B · dℓ = μ I +μ ε E · n̂ dA.
free
net
dt
threading
!
An alternate version of Ampere’s Law
for an isotropic linear material
We can rearrange this a bit to obtain

A
1 ⇀ d
B · dℓ = I +ε E · n̂ dA.
μ free
net
dt
threading
If we restrict ourselves to materials that are not electrically conducting (so that
1
Ifree = 0), and make use of H = B, we get
net
μ
threading
A
⇀ d
H · dℓ = ε E · n̂ dA . (9.6.5)
dt
dB ⇀
Recall that the original version of Ampère’s Law in vacuum, B · d ℓ = μ0 ε 0
dt
$ ∂B ∂E
E · n̂ dA , when applied to a plane wave led to equation (9.5.9), = −μ0 ε0 .
∂x ∂t
So, we can see that equation (9.6.5) leads to
∂H ∂E
= −ε ⇒
∂x ∂t
∂ (μH) ∂ E isomorphic with ∂B 1 ∂E
= −εμ ←−−−−−−−−→(9.5.14) : =− 2 .
∂x ∂t ∂x c ∂t
We see that μH plays the role of B, E plays the role of E, and εμ plays the role of 1/c2 .
To complete the isomorphism, we must find the equation for em fields in matter
∂E ∂B
that is isomorphic to equation (9.5.8): = − . If we apply the recipe of replacing
∂x ∂t
ε0 by ε, μ0 by μ, and all charges by just the free charges to Faraday’s Law, there is no
change:
A A
⇀ d ⇀ d
E · dℓ = − B · n̂ dA −→ E · dℓ = − B · n̂ dA.
dt dt
! !
Faraday’s Law for any circumstance An alternate version of Faraday’s Law
for linear materials
However, since B = μ H, we can rewrite this as

A
⇀ d
E · dℓ = − μ H · n̂ dA. (9.6.6)
dt
Recall that the original version of Faraday’s law when applied to a plane wave led to
∂E ∂B
equation (9.5.8), = − , so we can see that equation (9.6.6) leads to
∂x ∂t
∂E ∂ (μH) isomorphic with ∂E ∂B

=− ←−−−−−−−−→(9.5.8) : =− .
∂x ∂t ∂x ∂t
Thus, the isomorphism between em waves in a vacuum and em waves in linear
materials is complete. As for the case with rope waves, it will be convenient to write
the isomorphism in terms of B/μ0 , which is isomorphic to μH /μ = H:
Table 9.6.1. Isomorphism between electromagnetic waves in linear materials and in a vacuum
Electromagnetic waves in linear materials Electromagnetic waves in vacuum
H B/μ0
Electric Field E Electric Field E
Permeability μ Permeability of free space μ0
Permittivity ε Permittivity of free space ε0
Since B = μH, you may well ask, “Why bother to write H in the isomorphism
rather than just B/μ”? Indeed, we could well have done so. We will see that when we
consider reflections at a boundary between two different media (with different values
of μ), it is somewhat easier to phrase things in terms of H.
Concept test (answer below3 ): What is the speed of em waves in linear materials?
Concept test (answer below4 ): Recall that in section 9.5 I stated that the intensity
1
(power per area) for em waves in vacuum is S = E × B. Use this to explain why the
μ0
intensity of em waves in linear materials is given by the magnitude of
S = E × H. (9.6.7)
1 1
3. In vacuum, c = √ , so in linear media vp = √ .
μ0 ε0 με

1 B
4. From equation (9.5.13), in vacuum S = E×B=E× . Using the isomorphism,
μ0 μ0
this translates into S = E × H.
Concept test (answer below5 ): Explain why the intensity of em waves in linear materials
is given by

ε 2
S= E . (9.6.8)
μ
9.7 Waves on transmission lines
One of the two most important applications for em waves is the transmission of
information. For example, we use em waves for radio and television broadcasts.
However, we frequently want to have better control of where information goes, and for
these applications we use cables, such as those used for telephone, cable TV, computer
networks, speaker wires, and so on. Such cables are called “transmission lines” by
physicists. The information is always transmitted through a pair of conductors. For a
telephone line, there are two wires twisted together. For cable TV, there is a central
wire surrounded by a cylindrical outer conductor, with plastic insulation between.
(This is called a “coaxial cable.”) For transmission of information on circuit boards,
the transmission line often consists of a wire above a sheet of metal called a “ground
plane”; the ground plane acts as the second conductor. Sometimes, the metal chassis
of an apparatus, or the ground itself, is used as one of the two conductors needed for
a transmission line.
Depending on the system, the information is either represented by the time-varying
voltage difference between the two conductors, or by the time-varying current traveling
through them (into a target device on one wire, and back out on the other). No matter
whether the information is transmitted by applying a controlled voltage to one end of
the transmission line or by applying a controlled current, it propagates as a linked wave
of current and voltage travelling along the line. The current creates a magnetic field,
and the voltage difference between the two conductors is associated with an electric
field between them, so that the wave is actually a self-sustaining em phenomenon;
the mathematics are analogous to those for an em wave in a vacuum, with the current
analogous to B and the voltage analogous to E. We will see that, in a standard coaxial
cable such as you may have used in the laboratory to connect electrical signals to an
oscilloscope, the wave travels at 2/3 c.
Our reasoning about these waves begins with two points: (i) Any two conductors
that aren’t infinitely far apart have a capacitance between them. (ii) Any length of wire,
even if it’s straight, has some inductance. One way to see this is that when you run a
current through a straight wire, it sets up a magnetic field. It takes energy to create this

1 B μ0 B μ
5. From (9.5.12), E = cB = √ μ0 = . This translates into E = H⇔
μ0 ε0 μ0 ε0 μ0 ε

ε
H= E. Substituting this into equation (9.6.7) gives equation (9.6.8).
μ
Figure 9.7.1 Model for a transmission line.
field (recall from introductory electricity and magnetism that there is an energy density
associated with the magnetic field), so you know by a Lenz’s law type of argument
that this means it will be “harder” to start current flowing through the wire, which is
dI
exactly the characteristic of an inductor: V = L is the voltage produced across an
dt
inductor when the current through it is changed, in the same way that V = IR is the
voltage produced across a resistor when the current I flows through it.
Using (i) and (ii), you should now find it reasonable that any transmission line
can be modeled as a series of inductors along one wire (the top wire in figure 9.7.1)
and capacitors to the other wire, as shown. For a coaxial cable, the top wire would be
the inner conductor and the lower wire would be the cylindrical outer conductor. For
the single wire running above a ground plane, the top wire would be the wire running
above the ground and the lower wire would be the ground plane. We begin with the
“lumped circuit element” model shown; very soon, we will take the limit where the
size of each cell in the model becomes infinitesimally small, corresponding to the limit
of a continuous transmission line. For mathematical simplicity, we assume the lower
wire is grounded, that is, that it is at zero voltage. We define the current circulating
within each cell to be positive when it is moving clockwise, as shown, that is, moving
to the right on the top wire and to the left on the bottom wire.
We will find an isomorphism between this system and em waves in vacuum.
dqj dVj
Your turn: (a) Explain briefly why Ij −1 − Ij = , and why this equals C .
dt dt
dIj
(b) Explain briefly why L = Vj − Vj +1
dt
From the result of part (b), we have
dIj
L = Vj − Vj+1 = − Vj+1 − Vj ≡ −
V .
dt
Dividing both sides of this by the cell size a gives
L dIj
V
V L dIj
=− ⇔ =− .
a dt a a a dt
lim a→0
We define L0 to be the inductance per unit length. Therefore L /a −−−−−→ L0 . So, in
the limit that the cell size becomes infinitesimal, the above becomes
∂V ∂I
= −L0 ⇔
∂x ∂t

∂V ∂ L0 I isomorphic with ∂E ∂B
=− ←−−−−−−−−→(9.5.8) : =− .
∂x ∂t ∂x ∂t
In the isomorphism, V plays the role of E, and L0 I plays the role of B.
To show that the isomorphism is complete, we must find the equation that is
∂B ∂E
isomorphic to equation (9.5.9): = −μ0 ε0 . From part (a) of “Your turn” given
∂x ∂t
earlier, we have
dVj
C = Ij−1 − Ij = − Ij − Ij−1 ≡ −
I .
dt
Dividing both sides of this by the cell size a gives

I C dVj
=− .
a a dt
lim a→0
We define C0 to be the capacitance per unit length. Therefore C /a −−−−−→ C0 . So,
in the limit that the cell size becomes infinitesimal, the above becomes
∂I ∂V
= − C0 ⇔
∂x ∂t

∂ L0 I ∂ V isomorphic with ∂B ∂E
= −L0 C0 ←−−−−−−−−→(9.5.9) : = −μ0 ε0 .
∂x ∂t ∂x ∂t
Again, V plays the role of E and L0 I plays the role of B. We also see that the combination
L0 C0 plays the role of the combination μ0 ε0 . As for the isomorphism between rope
waves and em waves in vacuum, it is not clear whether L0 is isomorphic with μ0 and C0
with ε0 , or instead whether L0 is isomorphic with ε0 and C0 with μ0 . However, since
L0 is associated with the current (which is isomorphic to B), and μ0 is associated with
B, one might expect that L0 might be isomorphic to μ0 . Further, since C0 is associated
with the voltage (which is isomorphic to E), and ε0 is associated with E, one might
expect that C0 is isomorphic to ε0 . By considering the power transmitted in the wave,
you can show that this hunch is correct; see problem 9.14.
So, the isomorphism between waves on transmission lines and em waves in
vacuum is complete. For convenience in comparing with other isomorphisms, we
say that 1/C0 is isomorphic to 1/ε0 , instead of saying that C0 is isomorphic to ε0
(which is equivalent). Also, we write the isomorphism in terms of L0 I /L0 = I, which
is isomorphic to B/μ0 . See table 9.7.1.
We can use this isomorphism to quickly obtain some important results for waves
on transmission lines:
1. Em waves in vacuum are a self-sustaining combination of a wave in E and
a wave in B. By the isomorphism, waves on a transmission line are a self-
sustaining combination of a wave in V and a wave in I.
Table 9.7.1. Isomorphism between waves on a transmission line and electromagnetic

waves in a vacuum
Waves on a transmission line Electromagnetic waves in vacuum
I B/μ0
V Electric Field E
Inductance per length L0 Permeability of free space μ0
Inverse of capacitance per length: 1/C0 Inverse of permittivity of free space: 1/ε0
1
2. The speed of em waves in vacuum is c = √ , so the phase velocity of
μ 0 ε0
waves on transmission lines is
1
vp = . (9.7.1)
L0 C0
Standard RG58 coaxial cable (the type used in laboratories, and the type that has
a BNC coaxial connector at the end) has C0 = 100 pF/m and L0 = 250 nH/m;
1
plugging in these numbers gives vp = = 2.00 × 108 m/s, or almost
L0 C0
exactly 2/3 c. The coaxial cable typically used for cable television, RG6,
has C0 = 53.1pF/m and L0 = 348 nH/m; plugging in these numbers gives
1
vp = = 2.33 × 108 m/s, which is a bit more than 3/4 c.
L0 C0
3. For em waves in vacuum travelling in the positive x-direction, E (x , t) =
1 μ0 B (x , t)

c B (x , t) = √ B (x , t) = . Using the isomorphism, this means
μ0 ε0 ε0 μ0
that for waves on transmission lines,

L0
V (x , t) = I (x , t) . (9.7.2)
C0
(Wave traveling in + x -direction)
This means that the current flowing to the right on the top wire is in phase
with the voltage; the current on the bottom wire is always opposite that on the
top wire.
As for em waves in vacuum, for a transmission line wave traveling in the
−x-direction, there is a negative sign added to the relation between V and I,
so that

L0
V (x , t) = − I (x , t) . (9.7.3)
C0
(Wave traveling in − x -direction)

9.8 Sound waves
For most of us, sound is second only to light as the most important type of wave.
Sounds fill our lives, inform us about our surroundings, and are the main medium for
inter-personal relations. Sound is a traveling wave of fluctuations in the pressure and
density of the air. (All the arguments we will make will work just as well for sound
waves in water or any other medium.) As shown in figure 9.8.1, the fluctuations in
pressure P and density ρ are small compared to the background pressure and density.
In fact, the fluctuations shown in the picture are greatly exaggerated; in a typical sound
wave, the pressure and density only change by about 0.1% of the background value.
We define P′ to be the change in pressure relative to the background, and ρ ′ to be the
change in density relative to the background, so that
P (t) = P0 + P′ (t) (9.8.1)

′
and ρ (t) = ρ0 + ρ (t) . (9.8.2)
Figure 9.8.1 A sound wave

consists of small variations in
the pressure and density,
which are in phase, as shown
in parts (a) and (b). Parts
(c)–(e) shown the relationships
between the variation in
pressure ρ ′ , the displacement
δ , and velocity v = dδ/dt. The
displacement (δ ) and variation
◦
in pressure (ρ ′ ) waves are 90
out of phase, but the velocity
wave is in phase with the
density wave. The uniformly
spaced dots underneath parts b
and d represent the original
positions of packets of air
molecules. The arrows
represent the displacement
imparted to these packets by
the sound wave, with positive
displacement δ corresponding
to right-pointing arrows. The
bottom row of dots shows the
displaced positions of the air
packets. From this part of the
figure, you can see how the
displacement wave gives rise
to the variation in density.
Higher density means higher pressure, so the pressure and density waves are in
phase, as shown. As we did for em waves, we will focus on plane waves in our study
of sound, that is, waves in which the pressure and density only depend on x, not on y
or z. This means that all points in a plane perpendicular to the x-axis have the same
pressure and density.
The density fluctuations are caused by displacements of the air molecules from
their evenly spaced positions, as suggested by the dots in the figure which are closer
together at the peaks in the density and by the arrows below the dots which indicate the
displacements required to achieve this clumping. The displacements point toward the
peaks in the density and away from the valleys. As we did in our study of longitudinal
standing waves, we define the displacement of an air molecule (in the x-direction)
relative to its original position to be δ (x , t).
We will show below that the pressure variations are governed by the wave
equation, so that we get traveling waves. This means, for example, that we can
have a sinusoidal traveling wave in the displacement, δ = δ0 sin (kx − ωt). Since
the displacement is in the x-direction, there is a corresponding wave in the velocity:
dδ
v≡ = −δ0 ω cos(kx − ωt), so that the velocity wave is in phase with the density
dt
wave, as shown. Because this velocity is due to the relatively small variations in P and
ρ , the velocity itself is also small.
The air or other medium must obey two basic equations: the continuity equation
(which is really a statement of the conservation of mass) and Euler’s equation (a
different Euler’s equation from eiθ = cos θ + i sin θ , and one that is really a statement
of F = ma). Below, we derive these two equations, and then combine them to show
that the pressure variations are described by the wave equation.
The continuity equation

We consider a region of square cross-section, with length along the x-axis, as shown
in figure 9.8.2. Let’s think about the mass of air in the shaded section of infinitesimal
thickness that lies between x and x + dx. At a particular instant t, the mass flowing into
this section per unit time is
mass mass volume A · distance

= · = ρ (x , t) = ρ (x , t) A v (x , t) .
time volume time time
Similarly, the mass flowing out of the section per unit time is ρ (x + dx , t) A v (x + dx , t).
Therefore, the change in mass contained in the section per unit time is
Figure 9.8.2 Geometry for the derivation

of the Continuity Equation and Euler’s
equation.
dm
= {ρ (x) A v (x)} − {ρ (x + dx) A v (x + dx)} .
dt
We can use a first-order Taylor series for ρ (x + dx) and v (x + dx):
∂ρ ∂v
ρ (x + dx) = ρ (x) + dx and v (x + dx) = v (x) + dx .
∂x ∂x
Because dx is infinitesimal, these expressions are exact. Plugging them into the
expression above gives
" # " #9
dm ∂ρ ∂v
= {ρ (x) A v (x)} − ρ (x) + dx A v (x) + dx .
dt ∂x ∂x
" #
∂ρ ∂v ∂ρ ∂ v
= −A dx v (x) + ρ (x) dx + dx dx .
∂x ∂x ∂x ∂x
Since the last term is of order dx 2 , it is negligible compared to the others, so
" #
1 dm ∂ρ ∂v
=− dx v (x) + ρ (x) dx ⇔
A dt ∂x ∂x
" #
1 dm ∂ρ ∂v
=− v (x) + ρ (x) . (9.8.3)
A dx dt ∂x ∂x
m ∂ (vρ ) ∂ρ
For the shaded section in figure 9.8.2, ρ = . We also have that = v (x) +
A dx ∂x ∂x
∂v
ρ (x) . Substituting these into equation (9.8.3) gives the continuity equation:
∂x
∂ρ ∂ (vρ )
+ = 0. (9.8.4)
∂t ∂x
The continuity equation for a system with variation in the x -direction only.
Next, we will do some massaging of this which, together with a massaged

version of Euler’s equation, will lead to the wave equation for the pressure. Plugging
equation (9.8.2):ρ (t) = ρ0 + ρ ′ (t) into the above gives
∂ρ0 ∂ρ ′ ∂
+ + v ρ0 + ρ ′ = 0 ⇒
∂ t ! ∂ t ∂x
=0
∂ρ ′ ∂ v ∂ (vρ ′ )
+ ρ0 + = 0.
∂t ∂x ∂x
Since both v and ρ ′ are small, the last term is negligible compared to the other two,
giving
∂ρ ′ ∂v
+ ρ0 = 0. (9.8.5)
∂t ∂x
Pressure is a function of density. Expanding the pressure in a Taylor series gives
∂P
P (ρ ) = P ρ0 + ρ ′ = P ρo +ρ ′ + terms of order ρ ′2 and higher .
! ∂ρ
P0
Since ρ ′ is small, we ignore the higher order terms. Also, we define

∂P
vs2 ≡ . (9.8.6)
∂ρ
(Soon, you will understand the reason why we chose the symbol vs2 to represent this
derivative.) So,
∂P ∂ρ ′ ∂ρ ′ 1 ∂P
P (ρ, t) = P0 + ρ ′ vs2 ⇒ = vs2 ⇔ = 2 .
∂t ∂t ∂t vs ∂ t
∂P ∂ P′
Since P (t) = P0 + P′ (t), we have that = . Plugging this into the above gives
∂t ∂t
∂ρ ′ 1 ∂ P′
= 2 .
∂t vs ∂ t
Now, we plug this into equation (9.8.5), giving
1 ∂ P′ ∂v
2
+ ρ0 = 0. (9.8.7)
vs ∂ t ∂x
∂
Taking of both sides yields
∂t
1 ∂ 2 P′ ∂ ∂v
2 2
+ ρ0 = 0.
vs ∂ t ∂t ∂x
The order of partial derivatives doesn’t matter, so we could write this as
∂ ∂v 1 ∂ 2 P′
ρ0 =− 2 2 . (9.8.8)
∂x ∂t vs ∂ t
This is as far as we need to go with massaging the continuity equation. We’ll use
this result a bit later to get the wave equation for the pressure. You should examine
figure 9.8.3 at this point.
Figure 9.8.3 This surfer is probably

thinking about partial derivatives,
since they are so important for the
understanding of waves. Image ©
Quincy Dein/Dreamstime.com
Euler’s equation. The shaded section in figure 9.8.2 experiences a force AP (x) from
the left side (pushing to the right), and a force of magnitude AP (x + dx) from the right
side (pushing to the left). So, the net force is Fnet = A [P (x) − P (x + dx)]. Applying
F = ma to the section then gives
dv
Fnet = A [P (x) − P (x + dx)] = (ρ A dx) .
dt
∂P
Again, we use a first-order Taylor series: P (x + dx) = P(x) + dx, so that
∂x
∂P dv
− dx = ρ dx ⇒
∂x dt
∂P dv
− =ρ . (9.8.9)
∂x dt
dv
Recall that v is a function of both x and t. Therefore, the full derivative can be
dt
expressed in terms of the partial derivatives via
dv ∂ v ∂ v dx ∂v ∂v
= + = + v.
dt ∂t ∂ x dt ∂t ∂x

∂P ∂v ∂v
− =ρ +v . (9.8.10)
∂x ∂t ∂x
Euler’s equation for a system with variation in thex − direction only.
Again, we will now do a bit of massaging of this, in our pursuit of the wave equation
for the pressure.
Your turn: Use equations (9.8.10), (9.8.1): P (t) = P0 + P ′ (t), (9.8.2): ρ (t) = ρ0 + ρ ′ (t),
and the fact that both v and ρ ’ are small to show that
∂ P′ ∂v
− = ρ0 . (9.8.11)
∂x ∂t
∂
Taking of both sides gives
∂x
∂ 2 P′ ∂ ∂v
− = ρ0 .
∂ x2 ∂x ∂t
∂ ∂v 1 ∂ 2 P′
Finally, we use equation (9.8.8), ρ0 = − 2 2 to substitute for the right side,
∂x ∂t vs ∂ t
giving
∂ 2 P′ 1 ∂ 2 P′
− 2
=− 2 2 ⇔
∂x vs ∂ t
∂ 2 P′ ∂ 2 P′
2
= vs2 2 . (9.8.12)
∂t ∂x
∂ 2y 2
∂ 2y
This is the wave equation, equation (9.2.5): = vp . So, sound behaves as a linear
∂ t2 ∂ x2
∂P
wave, with speed given by equation (9.8.6): vs2 ≡ !
∂ρ
Concept test (answer below6 ): What is the difference between vs and v?

∂P
The speed of sound. To find the speed of sound vs ≡ , we must ask what should
∂ρ
be held constant when evaluating the partial derivative. Clearly, the pressure does
depend on the absolute temperature T , so perhaps it is T that should be held constant.
However, in fact the compressions and expansions of the air in a sound wave are
fast enough that the temperature is not constant; it is higher in regions where the gas
is compressed and lower in regions where it is expanded. So, instead of holding T
constant, it is a better approximation to hold the total energy of the system constant. In
other words, when we think of the air being compressed in part of the sound wave, it is
better to assume that in this compression all the work done on the gas goes to increase
its energy (so that its temperature goes up), rather than assuming that the compression
is done at constant temperature (which would require that we allow energy to leak
out as the gas is compressed). Note that the work done on the part of the gas being
compressed is done by other parts of the gas, so that the total energy is constant.
This type of compression, in which all the energy used to compress the gas stays
in the gas and none leaves, is called an “adiabatic” compression, from the Greek a
dia batos, “not through to pass,” meaning that no energy leaks out of (or into) the
system. So,

∂ P
vs = . (9.8.13)
∂ρ adiabatic
Nm Nm
The density is ρ = ⇔V = , where m is the mass of a gas molecule and
V ρ
N is the number of molecules in volume V . Therefore, we have that

∂P ∂ P dV ∂P Nm
= = − 2 . (9.8.14)
∂ρ ∂ V dρ ∂V ρ
We can model air using the ideal gas law:
PV = NkB T = nRT , (9.8.15)

where n = N NA is the number of moles of gas, NA = 6.022 × 1023 is Avogadro’s num-
ber, kB = 1.381 × 1023 J/K is Boltzmann’s constant, and R = NAkB = 8.314 J mol−1
K−1 is the Universal Gas Constant. So, we see that, if T were held constant, we
6. The quantity vs is the phase velocity of the wave, that is, the speed at which the crests of the sound
wave move through space. It is constant in time. The quantity v is defined as dδ/dt, that is, the
time derivative of the displacement of the air molecules relative to their equilibrium positions.
As shown in figure 9.8.1e, v varies sinusoidally in space. So, as the wave propagates, v also
varies sinusoidally in time.
constant
would have P = , where the constant would be NkB T . However, in fact as
V
the gas is compressed adiabatically, the temperature increases, so that the pressure
increases more quickly with a decrease in volume. One can show that, for an adiabatic
compression,
C
P= , (9.8.16)
Vγ
where C is a constant and the “adiabatic index” γ depends on the type of gas; for air,
γ = 1.40. Therefore, using equation (9.8.14),

∂P ∂P Nm γ C Nm
= − 2 = γ +1 2 .
∂ρ ∂V ρ V ρ
Since C = PV γ , we have
∂P γ P Nm γP
= = .
∂ρ V ρ2 ρ

∂P
So, vs ≡ ∂ρ ⇒

γ P0
vs = , (9.8.17)
ρ0
Speed of sound in a gas
where we have explicitly indicated that one uses the equilibrium values of the pressure
and density to calculate the speed of sound, since the speed of sound is an average
property of the gas, not something that varies on the scale of the wavelength of the
sound. For air under standard conditions, ρ0 = 1.2 kg/m3 and P0 = 1.01 × 105 Pa.
(Recall that 1 Pa, pronounced “one Pascal,” equals 1 N/m2 .) Plugging in these numbers
gives vs = 343 m/s, which matches very well with experimental values.
We are also interested in the speed of sound in liquids. Recall from section 2.3 that
Fapplied x
Young’s modulus E was defined by equation (2.3.3): = E , where Fapplied
A ℓ
is the force applied to one face of a solid with cross-sectional area A and length ℓ
and x = −
ℓ is the magnitude of the resulting change in length of the solid. Since

ℓ
Fapplied /A is pressure, we could rewrite this as P = −E , so that
ℓ
∂P E ∂P
= − ⇔ E = −ℓ .
∂ℓ ℓ ∂ℓ
If the pressure is instead applied “hydrostatically” (meaning that it is applied to all
faces), then we usually think about the change in the volume of the solid, instead of the
change in the length. We define the bulk modulus by analogy with Young’s modulus
to be
∂P
B ≡ −V . (9.8.18)
∂V
The same definition works equally well for liquids, which also get compressed when
pressure is applied hydrostatically. Rearranging equation (9.8.18) gives
∂P
= − B /V .
∂V
∂P B Nm B
= 2
= .
∂ρ V ρ ρ

∂P
Since vs ≡ , we then have
∂ρ

B
vs = , (9.8.19)
ρ0
Speed of sound in a liquid.
where again we have explicitly indicated that we use the equilibrium value of the
density to calculate the speed of sound. Now, as discussed earlier, for sound waves
the compressions are close to adiabatic. However, the bulk modulus is usually
measured under constant temperature (“isothermal”) conditions. Luckily, for a liquid
the difference between adiabatic compressions and isothermal compressions is much
smaller than for a gas, because the pressure increase when the volume is decreased is
determined more by changes in the interaction between molecules due to the reduction
in the average distance between them, so that the change in the temperature of the
molecules is not as important in determining the change in pressure. So, using typical
tabulated values of B for liquids works well with equation (9.8.19). For example,
◦
experimental values for the bulk modulus of fresh water at 20 C range from 2.19 to
3
2.22 GPa, and the density is 1,000 kg/m . Thus, equation (9.8.19) predicts a speed of
sound in water of 1,480–1,490 m/s, which is fairly close to the experimental value of
1,498 m/s.
Finally, we are also interested in the speed of sound in solids. When part of a solid
is compressed, it bulges out. Unlike in a liquid, the neighboring parts of the solid resist
this bulging, effectively increasing the bulk modulus for the part being compressed.
Furthermore, the parts of the solid which are outside the region of the sound wave
exert shear stress on the parts that are inside the region, further increasing the effective
stiffness. A full discussion of these effects is beyond the level of this book. However,
one can show7 that for typical solids,

1.5 B
vs ≈ . (9.8.20)
ρ0
Approximation for speed of sound in a typical solid.
7. Understanding the Properties of Matter, 2nd Ed., by Michael de Podesta (Taylor and Francis,
London, 2002), p. 287.
For example, for aluminum the bulk modulus is 75.5 GPa, and the density is
2,698 kg/m3 , so that equation (9.8.20) predicts vs = 6,480m/s, whereas the experi-
mental value is 6,374 m/s.
Isomorphism with em waves in a vacuum. As with the other types of waves we’ve
studied, we can form an isomorphism between sound waves and em waves in a vacuum.
We start with equation (9.8.11):
∂ P′ ∂v
− = ρ0 ⇔
∂x ∂t

∂ P′ ∂ ρ0 v isomorphic with ∂E ∂B
=− ←−−−−−−−−→(9.5.8) : =− .
∂x ∂t ∂x ∂t
In the isomorphism, the pressure variation P′ plays the role of E and the momentum-
per-volume ρ0 v plays the role of B.
To complete the isomorphism, we must find the equation for sound that is
∂B 1 ∂E
isomorphic with equation (9.5.14): =− 2 . We make use of equation (9.8.7):
∂x c ∂t
1 ∂ P′ ∂v
2
+ ρ0 =0⇒
vs ∂ t ∂x

∂ ρ0 v 1 ∂ P′ isomorphic with ∂B 1 ∂E
=− 2 ←−−−−−−−−→(9.5.14) : =− 2 .
∂x vs ∂ t ∂x c ∂t
Again the pressure variation P′ plays the role of E and the momentum-per-volume ρ0 v
plays the role of B.
γ P0
For sound waves in a gas, we have equation (9.8.17): vs = , which, by the
ρ0
1
above reasoning, is isomorphic to c = √ . As in previous isomorphisms, it is not
μ0 ε0
clear exactly how to form the isomorphisms between the quantities γ , P0 , and ρ0 and
the quantitites μ0 and ε0 . However, since ρ0 is associated with the momentum-per-
volume ρ0 v (which is isomorphic to B), and μ0 is associated with B, one might expect
that ρ0 might be isomorphic to μ0 . Further, since P0 is associated with the pressure
variation (which is isomorphic to E), and ε0 is associated with E, one might expect
that γ P0 is isomorphic to 1/ε0 . By considering the power transmitted in the wave, you
can show that this hunch is correct; see problem 9.16. Therefore, the isomorphism is
complete. Instead of writing that ρ0 v is isomorphic to B, it is more convenient to say
that v is isomorphic to B/μ0 .
Table 9.8.1. Isomorphism between sound waves in gases and electromagnetic

waves in a vacuum
Sound waves Electromagnetic waves in vacuum
Velocity v ≡ dδ/dt B/μ0

Pressure variation P′ Electric Field E
Equilibrium density ρ 0 Permeability of free space μ0
γ P0 Inverse of permittivity of free space: 1/ε0
Concept test (answer below8 ): What is the relationship between the magnitude of
the pressure variation wave P′ and the magnitude of the velocity wave v ≡ dδ/dt?
9.9 Musical instruments based on tubes
All wind instruments, from the oboe to the organ, whether woodwind or brass, are based
on resonance in a tube filled with air. We saw in section 9.4 that we can superpose two
waves of equal amplitude traveling in opposite directions to create a standing wave.
This is true for any system described by the wave equation. Since sound is described
by the wave equation, we know that there can be standing waves of sound. These are
similar to the standing waves we found for strings; they are the normal modes of the
air-filled tube. For the string fixed at both ends, the normal modes fit an integer number
of half wavelengths between the walls. As we’ll see, the boundary conditions for tubes
of air can be different, so that in some cases the condition instead is that we must fit
an odd integer number of quarter wavelengths between the ends.
There are two types of boundary conditions. In a flute, the player blows air across
the mouthpiece, which is near one end of the flute. The mouthpiece is an open hole,
so that the pressure at this end of the flute is kept essentially constant at atmospheric
pressure. The other end of the flute is open, so the pressure there is also essentially
constant. Thus, the boundary conditions for a flute are that the variation in pressure P′
must go to zero at the ends, just as for a string fixed between two walls the amplitude
must go to zero at the walls. We could also say that there must be a node of the standing
wave in pressure at each end of the flute. Therefore, the condition for determining
the frequencies of the normal modes is that we must fit an integer number of half
wavelengths into the length of the flute:
λn 2L
L=n ⇔ λn = . (9.9.1)
2 n
The phase velocity of the wave is the speed of sound, so that
ω 2π f
vs = = = λf ⇒
k 2π λ
v
f = s. (9.9.2)
λ
Substituting from equation (9.9.1) gives
vs
fn = n . (9.9.3)
2L
Resonant frequencies for an air tube open at both ends.
The lowest frequency mode, n = 1, is called the “fundamental.” There is a pressure

node at each end of the flute, and an “antinode” (a point of maximum amplitude) at
B
8. We use the isomorphism to translate (9.5.12), E = cB = cμ0 into P′ = vs ρ0 v.
μ0
the center. The next mode, n = 2, is called the “first harmonic”; this has pressure nodes
at both ends and also in the middle. The n = 3 mode is called the “second harmonic,”
and so on. We can see from equation (9.9.3) that the frequency progression of the
normal modes is quite simple for a tube open at both ends: f1 , 2f1 , 3f1 , and so on. When
the system is excited with a wide range of frequencies simultaneously (as when a flute
player blows across the mouthpiece), the response at the resonant frequencies is much
stronger than the response at other frequencies (because the quality factor Q is high),
resulting in a well-defined musical pitch. Most instruments are designed so that the
fundamental mode is excited with the highest amplitude, but the excitation of the other
modes in addition is critical to the musical timbre of the instrument.
It is also worthwhile to consider what the wave in displacement looks like. We
◦
saw in figure 9.8.1 that the displacement wave is 90 out of phase with the pressure
wave. Therefore, for a standing wave:
A node in the pressure wave corresponds to an antinode in the displacement wave,

and vice-versa.
So, the standing waves for a flute look as shown in figure 9.9.1a.
Concept test (answer below9 ): In figure 9.9.1a, how can you tell that the solid line for
the displacement curve occurs at the same time as the solid line for the pressure curve,
rather than at the same time as the dashed line for the pressure curve?
Figure 9.9.1 a: Pressure and displacement standing waves for an air tube with both ends open,
for the mode n = 1. b: Displacement standing waves for a tube that is closed on the left end
and open on the right. Top: fundamental (n = 1). Bottom (n = 2).
9. High pressure requires high density. For the solid line, the pressure is highest at the center, so
we need positive displacements for the left half and negative displacements for the right half.
There are some types of musical instruments in which one end of the tube is closed.
For example, some pipes on organs are like this. In all organ pipes, air is blown in at
the bottom end, providing the constant pressure boundary condition discussed above.
However, for some pipes the top end is open, and for others it is closed. (These two
types produce a different quality of musical note, as we’ll explore below.) There must
be a node in the displacement at a closed end, since the end prevents motion along the
axis of the tube.
Also, for any instrument which is excited at one end by vibrating lips (trumpet,
etc.) or by a vibrating reed (clarinet, etc.), it is appropriate to count that end as closed, as
we can easily see. Consider a mass on a spring that is driven at its resonant frequency by
moving the support point up and down. In steady state, the amplitude of the motion of
the mass is Q times the amplitude of the support point motion (assuming the damping
is not too heavy). Therefore, compared to the motion of the mass, the support point
is almost a fixed point. The analogy for a trumpet is that the motion of the lips at the
mouthpiece is very small compared to the motion of the air molecules at the antinodes
of the standing wave, so we can consider the mouthpiece end to essentially be a node
of the displacement.
Let’s consider a tube with the left end closed (corresponding to a node in the
displacement) and the right end open (corresponding to a node in the pressure, and
so an antinode in the displacement). For the longest wavelength normal mode (the
fundamental), we fit a quarter wavelength into the length, as shown in the top part
figure 9.9.1b. In the mode with next shorter wavelength, shown in the lower part of
the figure, we fit 3/4 of a wavelength into the length. In the next mode, we would fit
5/4 of a wavelength. We can see that the general pattern is
λn 4L
L = (2n − 1) ⇔ λn = .
4 2n − 1
vs
From equation (9.9.2), f = , so
λ
vs
fn = (2n − 1) . (9.9.4)
4L
Resonant frequencies for an air tube open one end and closed at the other.
Unlike the case when both ends are open, the frequency progression is more
complicated: f1 , 3f1 , 5f1 , and so on. Because the frequencies corresponding to even
multiples of f1 are missing, the musical timbre of an instrument with one end closed
is quite different from one with both ends open.
9.10 Power carried by rope and electromagnetic waves; RMS

amplitudes
Earlier, we noted that transmission of information is one of the two most important
applications for waves. The other is the transmission of energy. Almost all the energy
we use on our planet has been transmitted to us via em waves from the sun. (Much of it
was transmitted hundreds of millions of years ago, and stored up in the form of fossil
Figure 9.10.1 A wave train propagates to

the right. The part of the rope to the right
of the dashed vertical line is initially
stationary, but starts moving once the
wave impinges on it, showing that the
wave must carry energy.
fuels.) Microwave ovens use em waves with a wavelength of about 10 cm to transmit

energy into the water molecules in food.10 Infrared lasers are used for surgery and for
cutting metal. Dentists use ultraviolet light (with a wavelength of about 300 nm) to
harden dental adhesives.
It is easy to see that energy is carried by a rope wave. For the wave train shown
in figure 9.10.1, the pieces of the rope on the right are initially at rest, and so have no
kinetic energy. However, as the wave reaches them, they begin to move, so the wave
has moved energy from left to right. The further the wavefront moves to the right, the
more energy has been moved from the region to the left of the dashed vertical line
into the region on the right. Thus, if the wave is originally semi-infinite (i.e., it extends
infinitely far to the left), it must contain an infinite amount of energy. So, we see that,
instead of discussing the total energy in a wave, it is more useful to discuss the power
carried by a wave, that is, the energy per unit time. For a sinusoidal rope wave, you
can show in problem 9.15 that the power is given by
P = 21 μA2 ω2 vp , (9.10.1)
where μ is the mass per unit length, A is the amplitude, ω is the angular frequency,
and vp is the speed. The important thing to note about this relation is that the power is
proportional to the square of the amplitude; this is a universal feature of all waves.
Most applications of waves for power transmission rely on em waves. It is not
difficult to find the power transmitted by em waves in vacuum, if we make use of two
results that you may have encountered in introductory electricity and magnetism: there
is energy density (i.e., energy per unit volume) in the electric field and in the magnetic
field:
ε0 2
Energy density of E : ηE = E (9.10.2)
2
1 2
Energy density of B : ηB = B . (9.10.3)
2μ0
10. Scientists have investigated the use of microwaves for heating homes for decades. The basic idea
is that very low level microwaves would be broadcast throughout the home, heating the humans
within. Because the heat is mostly absorbed by the humans and not the furniture or the air, much
less energy is wasted than in conventional heating systems. However, so far these ideas have
not progressed beyond the experimental stage, because of concerns over whether such a system
could be successfully marketed.
Figure 9.10.2 A box of volume Adx .
For em waves in vacuum, E = cB. Plugging this into equation (9.10.3) gives
1 E2 1 E2 ε
ηB = 2
= = 0 E2.
2μ0 c 2μ0 1/ μ0 ε0 2
So, we see that, for em radiation in vacuum, the energy is carried equally in the magnetic
and electric fields. The total energy density is thus
ηRad = ε0 E 2 (9.10.4)
Energy density of em radiation in vacuum
To calculate the power delivered by these waves, we first calculate the energy contained
in a box of cross-sectional area A, as shown in figure 9.10.2. The box has an infinitesimal
length dx along the direction of wave travel, so that E is essentially constant throughout
the box. The volume of the box is A dx, so the energy contained within it is ε0 E 2 A dx .
We imagine that the box moves forward with the wave, and that we have placed a
perfect energy absorber right in front of the box. Then, the entire energy content of the
box is deposited into the absorber in a time dx /c. The power is given by
Energy ε E 2 A dx
P= = 0 = cε0 E 2 A.
time dx c
The wavefronts of a plane wave are infinite in the y- and z-directions, so that to calculate
the power of such a wave, we would need a box with infinite A, which according to
the above would be infinite. This is not a helpful notion – it is much more useful to
quote the power per unit area, which is the definition of intensity:
Power
Intensity ≡ ≡ S = c ε0 E 2 . (9.10.5)
area
To bring this to the standard form, we use E = cB, so that

1 1
S = cε0 E (cB) = c2 ε0 EB = ε EB = EB. (9.10.6)
μ0 ε0 0 μ0
Recall that we had earlier found that the Poynting vector points in the direction of
propagation. We defined it as
1
S= E × B, (9.10.7)
μ0
Instantaneous intensity of an em wave in vacuum.
and now we see that it has magnitude equal to the intensity of the wave.
Since E and B vary sinusoidally in time, this shows that the intensity of an em wave
varies in time. This is seldom of any importance – usually we are far more interested in
the average intensity. We revert temporarily to the form of the instantaneous intensity
given by equation (9.10.5):
S = c ε0 E 2 .
So, the average intensity would be given by

% &
S = cε0 E 2 , (9.10.8)
where the angle brackets indicate an average over one period.

We define the “root mean square” or “rms” amplitude in terms of this average:
' (
Erms ≡ E2 . (9.10.9)
So, Erms is the square root of the mean of the square of E. The concept of rms amplitude
is quite important for the study of almost all time-varying phenomena.
Thus,
2
S = cε0 Erms .
Since E = cB, we have Erms = cBrms , so that

1
S = c2 ε0 Erms Brms = E B .
μ0 rms rms
Thus, we see that equation (9.10.7),
1
S= E × B,
μ0
Instantaneous intensity (if E and B are amplitudes) or average intensity
(if E and B are rms amplitudes) of an em wave in vacuum
can be interpreted in two ways, both of which are correct. Either it can indicate a time-
varying vector with instantaneous magnitude equal to the instantaneous intensity of
the wave, or (and this is the more usual interpretation) we can use the rms amplitudes
of E and B to calculate the cross product, with a result that gives the average
intensity.
Most often, rms amplitudes are used for sinusoidal waves. In this case, the relation
between the amplitude (i.e., the peak value) and the rms amplitude is simple. We make
use of the handy fact (which we encountered in chapter 4) that the average of a sinusoid
squared over a full wavelength is half the peak value,11 that is,
% & E2
peak
For a sinusoid, E 2 = .
2
Plugging this into the definition of rms amplitude, (9.10.9), gives

2
Epeak
' ( Epeak
Erms ≡ E = 2 = √ .
2 2
√
So, for a sinusoidal waveform, the rms amplitude is just the (peak) amplitude over 2.
Epeak
Self-test (answer below12 ): Sketch a waveform for which Erms ≈ .
20
9.11 Intensity of sound waves; decibels
For a sound wave, the energy is carried partly by the kinetic energy associated with
the oscillating longitudinal velocity, and partly by the potential energy associated with
the compressions and expansions of the medium. We first consider the density of the
kinetic energy, which is given by
1
kinetic energy mv2
ηK ≡ = 2 = 12 ρ v2 .
volume volume
Your turn: Employing the same ideas used to develop (9.10.5), show that the intensity
of kinetic energy for sound waves is
SK = 12 vs ρ0 v 2 . (9.11.1)
The above is the instantaneous kinetic energy intensity, which is highest at the velocity
peaks and valleys of the sound wave. Usually, we are more interested in the intensity
averaged over a wavelength. For a sinusoidal right-moving wave, we have
v = v0 sin (kx − ωt), (9.11.2)
where v0 is the amplitude of the longitudinal velocity wave, and is not to be confused
ω
with the phase velocity of the sound wave, vs = . We can compute the the average
k
1 ' (
11. Recall, that cos2 ωd t = 1 + cos 2ωd t and that cos 2ωd t = 0 over a complete cycle.
2
12. Answer to self-test: One possibility is a train of pulses, with the waveform equal to zero between
pulses, and with the repeat period equal to 20 times the pulse width.
kinetic energy intensity by plugging equation (9.11.2) into (9.11.1):

' ( 1 % & % &
SK = 2 vs ρ0 v2 = 12 vs ρ0 v02 sin2 (kx − ωt) .
Again, we make use of the fact that the average of the square of a sinusoid over one
period is half. Therefore,
' ( 1
SK = 4 vs ρ0 v02 .
In problem 9.11, you can show that the potential energy averaged over a wavelength
is equal to the kinetic energy averaged over a wavelength. (This is analogous to the
situation for em waves in a vacuum, for which we saw that the energy is equally
divided between the electric and magnetic components.) Therefore, the total average
intensity is
S = 21 vs ρ0 v02 . (9.11.3)
It is more common to discuss sound intensity in terms of pressure. Plugging
∂ P′ ∂v
equation (9.11.2) into (9.8.11), − = ρ0 , gives
∂x ∂t
∂P ′
= ωρ0 v0 cos(kx − ωt). (9.11.4)
∂x
We saw (figure 9.8.1) that the pressure wave is in phase with the velocity wave.
Therefore, we must have
P′ = Pm sin (kx − ωt) ⇒
∂ P′
= Pm k cos(kx − ωt). (9.11.5)
∂x
(Note that Pm is the amplitude of the pressure wave, while P0 is the constant background
pressure.) Comparing equations (9.11.5) and (9.11.4), we see that
k 1
Pm k = ω ρ0 v0 ⇔ v0 = Pm ⇒
ω ρ0
1
v0 = Pm . (9.11.6)
vs ρ0
So, we can re-express equation (9.11.3) as
Pm2
S = .
2vs ρ0
It is more common to deal with the rms amplitude of the pressure, which, for a sinusoidal
wave is
P
Prms = √m ,
2
so that
2
Prms
S = . (9.11.7)
vs ρ0
Average intensity of a sound wave in terms of the rms amplitude Prms of the pressure variation.
Sound intensity is usually quoted in decibels, abbreviated dB:
S
dB ≡ 10 log10 ' ( , (9.11.8)
Sref
' (
where Sref is the intensity of a sound wave with Pref ,rms = 2 × 10−5 Pa; this
corresponds to the quietest sound that can be perceived by a human. We can re-express
this equation as
S
' ( = 10dB/10 , (9.11.9)
Sref
which shows that each 10 dB increase in the intensity corresponds to a factor of 10

increase in intensity.
Self-test (answer below13 ): Does a 5 dB increase in intensity correspond to a factor of

5 increase in intensity?
' (
Since vs and ρ0 are the same for S and Sref , we also have
2
Prms
dB ≡ 10 log10 2
⇒
Pref, rms
Prms
dB = 20 log10 . (9.11.10)
Pref, rms
Example: What is the rms amplitude of pressure in a sound wave of intensity 0 dB?
P
Answer: dB = 20 log10 rms ⇒
Pref, rms
Prms = Pref, rms 10dB/20 . (9.11.11)
So, for 0 dB, Prms = Pref, rms = 2 × 10−5 Pa.
Concept test (answer below14 ): How much larger is the rms amplitude of pressure in
a sound wave of 10 dB intensity than in a sound wave of −10 dB intensity?
Decibels are used in many other measurements as well. For example, sometimes
voltage is quoted in decibels, and the symbol “dBV” is used, meaning decibels relative
S √
13. ' ( = 10dB/10 = 105/10 = 10 = 3.162, so a 5 dB increase corresponds to a factor of 3.162
Sref
increase in intensity.
1
14. For the −10 dB wave, Prms = Pref, rms 10−10/20 = √ Pref, rms . For the 10 dB wave, Prms =
√ 10
Pref, rms 1010/20 = 10 Pref, rms , so it has ten times larger amplitude than the −10 dB wave. We
can see that, in general, each increase of 20 dB corresponds to a factor of 10 increase in the
amplitude (and a factor of 100 increase in the intensity).
to a reference level of 1 V:
V
dBV ≡ 20 log10 .
(1 V)
Again, each increase of 20 dB corresponds to a factor of 10 increase in Voltage. For
a resistor, power is proportional to V 2 , so that an increase of 20 dB corresponds to a
factor of 100 increase in power, and an increase of 10 dB corresponds to a factor of
10 increase in power.
9.12 Dispersion relations and group velocity
When I first encountered the definition below as an undergraduate, I did not fully
appreciate how frequently it would be important. So, be forewarned: you will hear
about dispersion relations at least once a month for the rest of your physics life!
Dispersion relation: The relation between angular velocity ω and wave-

number k.
Why is this called the “dispersion” relation? We have seen that the phase
velocity (i.e., the velocity of the crests or, equivalently, of the troughs) is given by
equation (9.3.2):
ω
vp = .
k
This is one way of presenting the dispersion relation. For most of the waves we have
studied, vp is a constant, independent of k and ω. In other words, for most of the
waves we have studied, including waves on a rope, em waves in vacuum, and waves
on transmission lines, the speed of the wave does not depend on its wavenumber or
frequency. This fact becomes extremely important for the propagation of pulses, such
as that shown here in figure 9.12.1. (The vertical axis would be y for a wave on a
rope, E for an em wave, or V for a wave on a transmission line.) As you learned in the
section 8.5 (on Fourier Transforms), such a wave, even though it is not periodic, can
be synthesized by adding together an infinite number of sinusoids. Now, consider what
happens as the wave propagates. As long as all these sinusoids propagate at the same
speed, the pulse maintains its shape, since all the sinusoids continue to add together in
Figure 9.12.1 A propagating pulse.

the same way, except for the overall motion. However, if vp depends on wavelength,
then the different Fourier components travel at different speeds, and so the pulse soon
gets dispersed. Hence the name “dispersion relation.”
As we’ll argue below, there are many examples of wave propagation for which
vp does depend on ω or k, so the dispersion relation is nonlinear The phenomenon
of dispersion has enormous practical consequences. In some cases, the dispersion is
desirable, so as to separate different frequency components of a signal. For example, a
prism takes advantage of the differences in speed within the glass for different colors
(wavelengths) of light to bend them into different paths. In other cases, the dispersion
is very undesirable, and must be minimized. For example, when data is transmitted as
a series of pulses through fiber optic cables, it is quite important that the shape of each
pulse not change too much as it propagates through the kilometers of glass between
relay stations. (It turns out that there is a very fortuitous window of wavelengths,
for which the glass used for fibers has a nearly linear dispersion relation, and very
low absorption. This particular window happens to match almost perfectly with the
wavelength produced by inexpensive solid-state lasers.)
The propagation of light through glass is one of the most important examples
of dispersion. The full details of this propagation are beyond the scope of this book,
so here we present a qualitative model. As the light wave moves into the material,
it exerts a force, which varies sinusoidally in time, on the electrons, causing them to
vibrate. (It also exerts a force on the nuclei, but they are so much more massive that
they hardly move at all.) The vibrating electrons, because they are accelerated charges,
emit em radiation. The total wave propagating through the glass is the superposition
of the original wave with this re-radiated wave coming from each electron. The speed
at which this total wave propagates depends on the details of how all these waves
interfere with each other.
Since the electrons are originally in equilibrium, we can qualitatively model them
as harmonic oscillators. Therefore, the phase of their response relative to the drive
(provided by the incoming wave) depends on the ratio of the drive frequency (i.e., the
frequency of the incoming wave) to the resonant frequency of the oscillators. Since this
ratio changes as the frequency of the incoming wave changes, the phase relationship
between the incoming wave and the re-radiated wave from the electrons changes, and
so the speed of propagation changes. Therefore, although we can write the dispersion
ω
relation in the form (9.3.2): vp = , vp is a function of ω, and so the dispersion relation
k
is nonlinear.
You have already encountered another nonlinear dispersion relation: the one for a
beaded string, which is the same as the relation for atoms in a crystalline solid. Recall
from equation (7.2.7):

π √ π a 2T
kn = n ωn = 2ωA sin n ωA ≡
L 2L am
Combining these expressions gives
√ a
ωn = 2ωA sin kn , (9.12.1)
2
which is an example of a nonlinear dispersion relation.
We’ll consider one more example: quantum mechanical waves. We have men-
tioned a few times that quantum mechanical particles, such as electrons, have a wave
nature, and are described by a wavefunction (x , t). As for other types of waves, this
wavefunction has a frequency and a wavelength. As we discussed briefly in section 5.5,
it turns out that the energy of any quantum mechanical particle (such as an electron, a
photon, or a quantized vibration of a crystal called a “phonon”) is given by
E = h̄ω. (9.12.2)
(You may have seen this equation in an equivalent form, such as E = hf or E = hν ,
where h = 2π h̄ and f or ν (“nu”) represents the frequency.) It turns out that the
momentum of the particle is given by
p = h̄k . (9.12.3)
These two equations form the basis of quantum mechanics. Therefore, when you
encounter them in a quantum mechanics course, be sure you understand the exper-
imental justification for them.
For a particle with mass, such as the electron, we can express the kinetic energy
in terms of the momentum:
2
1 2 mve p2
KE = 2 mve = = ,
2m 2m
where ve is the speed of the electron. Substituting from equation (9.12.3) gives
h̄2 k 2
KE = .
2m
For a “free electron”, that is, an electron which is simply traveling through space
without any forces on it, the energy is entirely kinetic. Therefore,
h̄2 k 2
E = h̄ω = KE = ⇔
2m
h̄k 2
ω= . (9.12.4)
2m
Dispersion relation for a free electron
This is a another example of a nonlinear dispersion relation. The fact that it’s nonlinear
immediately tells us that quantum waves with different wavelengths travel at different
speeds:

ω h̄k 2 2m h̄k
vp = = = , (9.12.5)
k k 2m
so that higher wavenumber (smaller wavelength) quantum waves go faster.
Let’s check our understanding. Plugging into the above from equation (9.12.3)
gives
h̄k p mve v
vp = = = = e.
2m 2m 2m 2
The speed of the electron wave, the wave that represents the electron, is half the speed
of the electron itself!
OK, let’s not panic. If the electron wave has a well-defined k (as we assumed in
the above discussion), that means it is mathematically represented by a pure sinusoid,
which means that the wave must extend from x → −∞ to x → +∞. We discussed
such a wave in section 1.11: = ψ0 e−iω0 t eik0 x . Because we want to emphasize that
they are fixed quantities, we write the angular frequency as ω0 (instead of simply
ω) and the wavenumber as k0 (instead of simply as k). The quantity ||2 (called the
“probability density”) tells us the probability for finding the electron at a particular
place. For this case,

||2 = ∗ = ψ0 e−iω0 t eik0 x ψ0 eiω0 t e−ik0 x = ψ02 .
Since this doesn’t depend on x, we say that such an electron is “completely delocalized”:
it is equally likely to be found anywhere between x → −∞ and x → +∞. So, perhaps
it shouldn’t worry us that the velocity of the electron is twice the velocity of the wave
that represents it – after all, if the electron is already everywhere, what does it mean
for it to have a velocity anyway?
To discuss the velocity of the electron in terms we understand better, we need to
consider an electron that’s in a more localized state. We can create such a state, called
a “wavepacket,” by multiplying the free-electron wavefunction = ψ0 e−iω0 t eik0 x by
an envelope function, as shown in figure 9.12.2a. The figure shows the wavepacket at
t = 0, with the peak of the envelope at position xm .
How will the wavepacket evolve in time? The pulse could be Fourier-synthesized
by adding up a large number of sinusoids of the form = ψ0 e−iωt eikx , with different
Figure 9.12.2 a: We can create a wavepacket

(bottom) by multiplying a rapidly oscillating
function (in this case a free-electron
wavefunction, top) by an envelope function
(middle). b: The wavepacket as a function of
time, as it passes the point x0 .
k’s (and corresponding ω’s) for each sinusoid. Each of these sinusoids propagates at
h̄k
the speed given by equation (9.12.5), vp = , so we can propagate each sinusoid
2m
forward in time, and then add them up to see how the wavepacket has propagated. We
will show that, if we use an envelope which varies slowly enough, then, over short to
moderate time intervals, the envelope propagates without changing its shape, at a speed
vg (called the “group velocity”) which matches that of the electron itself. (This means,
of course, that the group velocity is different from the phase velocity vp .) This idea of
group velocity is critically important for all types of waves governed by a nonlinear
dispersion relation.
Figure 9.12.2a shows the wavepacket as a function of x. If instead we consider
the behavior as a function of time as the pulse passes the point x0 , the plot of versus
time would look as shown in figure 9.12.2b. For the rest of this section, we will think
primarily in terms of the dependence on time, instead of the dependence on x.
We will set aside this electron wavepacket for now, but come back to it after we’ve
figured out how to calculate the group velocity. For now, let’s consider a different
example for which the ideas of dispersion and group velocity are important; we’ll
use this example as a basis for explicitly showing that the envelope can propagate
without changing its shape, even for a system with a nonlinear dispersion relation. Our
example is the important problem of sending information through a fiber optic cable.
For example, figure 9.12.3 shows the voltage output from a microphone as I speak
the word, “Hi!.” The horizontal axis is in seconds, so that the entire word lasts about
1/ second. The first part of this is the “H” sound; if you look carefully, you can see
4
a repeating pattern in this part, one copy of which is highlighted with a box. You can
see a different repeating pattern in the last part of the word, which is the “i” sound.
The highlight box is 10-ms wide, and the sharpest features in it are perhaps 0.5-ms
wide. We can see that, later on in the word, there are even sharper features that are
about 0.1-ms wide. So, to Fourier synthesize this waveform, we’d need sinusoids with
1 1
frequencies from about to about m, or about 4 Hz to 10 kHz. This is
0.25 s 0.1 ms
a typical range for audio signals. We define the function shown in figure 9.12.3 to
be f (t).
Figure 9.12.3 The word “Hi!” as recorded with a microphone. The vertical axis is proportional
to the variation in air pressure relative to background, while the horizontal axis is time in s.
Say that we’d like to send this word over a fiber optic cable. We will study fiber
optics in more detail in section 10.9, but just from the name you can tell that they’re
intended to transmit optical signals, that is, signals with frequencies from about 1014
to 1015 Hz. As we discussed earlier, the frequencies in our audio signal f (t) only go up
to about 10 kHz, so if we converted the signal to an em wave, it wouldn’t be able to
propagate through the fiber optic. One solution is to use the audio signal to “modulate”
a fixed-frequency “carrier wave,” that is, to use f (t) as an envelope function for a pure
sinusoid:
y (t) = f (t) cos ωc t (9.12.6)

! !
envelope carrier
function wave
where the angular frequency ωc of the carrier is in the frequency range that can
propagate through the fiber optic. This idea is illustrated in figure 9.12.4.
To begin with, instead of the complicated f (t) that represents the “Hi!” sound,
let’s use a pure sinusoid as the audio signal that is to be transmitted:
f (t) = B cos ωa t ,
where ωa is an audio frequency, perhaps 2π · 10 kHz, and B is the amplitude. Thus,
y (t) = B cos ωa t cos ωc t .
Recall from our discussion of beats equation (5.1.3):
A cos ω1 t + A cos ω2 t = 2A cos ωe t cos ωav t ,
where, from equation (5.1.2),
ω1 = ωav + ωe and ω2 = ωav − ωe .
Figure 9.12.4 A part of an audio waveform (black) is used as an envelope function for a
rapidly oscillating carrier wave.
Figure 9.12.5 A rapidly

oscillating carrier wave
multiplied by a simple
sinusoidal envelope function
is equal to the sum of two
sinuoids, each at an angular
frequency close to that
of the carrier.
To apply this to our case, we make the following subsitutions: 2A → B, ωe → ωa , and

ωav → ωc . Therefore,
ω1 = ωc + ωa and ω2 = ωc − ωa , (9.12.7)
ω1 + ω2 ω1 − ω2
⇒ ωc = and ωa = , (9.12.8)
2 2
B B
and y (t) = B cos ωa t cos ωc t = cos ω1 t + cos ω2 t . (9.12.9)
2 2
This relation, shown graphically in figure 9.12.5, says that by using the audio signal
as an envelope function for the carrier
wave, we shift the
frequency
up close to ωc .
(For a fiber optic, ωc ∼ 2π 1014 Hz , while ωa ∼ 2π 104 Hz , so ωc ± ωa is very
close to ωc in percentage terms.) So, we can now transmit the signal through the
fiber optic.15
The glass from which the fiber optic is made has a nonlinear dispersion relation,
so we must inquire whether the shape of the envelope function is preserved as the
wave propagates. So far, we’ve only been considering the wave as a function of time.
To change it to a propagating wave, we replace ω1 t by k1 x − ω1 t, where k1 is the
wavenumber corresponding to ω1 , and we replace ω2 t by k2 x − ω2 t. Applying this
15. This is essentially

how AM
(“amplitude modulation”) radio works, but the carrier is a radio wave
with ωc ∼ 2π 106 Hz .
recipe to equation (9.12.9), and using equation (9.12.8) gives

+ , + ,
k1 x − ω1 t − k2 x − ω2 t k1 x − ω1 t + k2 x − ω2 t
y (x , t) = B cos cos
2 2
B B
= cos k1 x − ω1 t + cos k2 x − ω2 t . (9.12.10)
2 2
We will assume that the slope of the function ω (k) is approximately constant over the
range ω1 to ω2 , so that the wavenumber of the carrier wave is equal to the average of
k1 and k2 :
k1 + k2
kc = . (9.12.11)
2
We also define
k1 − k2
kd ≡ . (9.12.12)
2
Using these, together with equation (9.12.8), we can simplify equation (9.12.10) to

y (x , t) = B cos kd x − ωa t cos kc x − ωc t
! !
envelope function carrier wave
B B
= cos k1 x − ω1 t + cos k2 x − ω2 t . (9.12.13)
2 2
ωa ω − ω2
We see that the envelope function travels at the speed vg = = 1 . Recall
kd k1 − k2
that we assume that the slope of the function ω (k) is approximately constant over the
range ω1 to ω2 , therefore

dω
vg = . (9.12.14)
dk kc
Group velocity (the velocity of the envelope for a wave packet)
In appendix B, it is shown that these ideas all work equally well for a more
complicated envelope function f (t), rather than the simple sinusoid we used. If f (t)
can be Fourier synthesized from sinusoids with angular frequencies from 0 up to ωm ,
then y (t) = f (t) cos ωc t is shown in the appendix to have Fourier components with
angular frequencies from ωc − ωm to ωc + ωm . The corresponding traveling wave is
shown to have an envelope whose shape is determined by f (t), which travels at the
group velocity (9.12.14):
y (x , t) = carrier
wave! · envelope
!
travels at shape determined
vp = ωkcc by f (t), travels
at
vg ≡ ddkω

kc
Figure 9.12.6 An example of an envelope modulating a carrier for a travelling wave.
dω
(To obtain the above, we again require that be constant over the range ωc − ωm to
dk
ωc + ωm .) An example is shown in figure 9.12.6.
To recap:
If we start with an envelope function f (t) having nonzero Fourier components from 0
to ωm , and multiply it by a carrier wave cos ωc t:
1. The resulting wavepacket y (t) = f (t) cos ωc t has Fourier components with
angular frequencies from ωc − ωm to ωc + ωm .
dω
2. If is constant over the range ωc − ωm to ωc + ωm , then the envelope of
dk
the wavepacket
doesn’t change shape, and propagates at the group velocity
dω
vg ≡ .
dk kc
3. The peaks and valleys of the carrier wave (which is modulated by the envelope)
ω
propagate at vp = c .
kc
Does the idea of group velocity solve the conundrum about electron waves? Let’s
check to see whether the group velocity for a localized electron wavepacket equals the
velocity of the electron itself:
h̄k 2 dω 2h̄k p mve √

ω= ⇒ vg ≡ = = = = ve (Phew!!)
2m dk 2m m m
In general, vg can be equal to, greater than, or less than vp . It even occurs in some
real systems that vg is negative, even though all the Fourier component waves have
positive vp !
Figure 9.12.7 A hypothetical dispersion

relation.
Concept test (answer below16 ): For the hypothetical dispersion relation shown in
figure 9.12.7 here, identify (a) a point where vp = vg , (b) a point where vp < vg , (c) a
point where vp > vg > 0, and (d) a point where vp > 0 > vg .
terms:
Traveling wave (9.1)
Partial derivative (9.2)
The wave equation (9.2)
Superposition principle for traveling waves (9.4)
Plane wave (9.5)

16. vp ≡ ω k is the slope of a straight line drawn from the origin to a point on the dispersion curve,

while vg ≡ dω dk is the slope of the curve itself. Therefore, as shown in figure 9.12.8, at point 1
they are equal. At point 2, vp > vg > 0. (The group velocity is only slightly positive at point 2.)
At point 3, vp < vg . At point 4, vp > 0 > vg .
Figure 9.12.8 Solution to concept test.

Poynting vector (9.5)

Momentum wave (9.5)
Dielectric constant (9.6)
Permittivity (9.6)
Magnetic susceptibility (9.6)
Permeability (9.6)
H (9.6)
Transmission line (9.7)
Displacement wave for sound (9.8)
Velocity wave for sound (9.8)
Bulk modulus (9.8)
Adiabatic (9.8)
Continuity equation (9.8)
Euler’s equation for fluids (9.8)
rms amplitude (9.10)
dB (9.11)
Dispersion relation (9.12)
Carrier wave (9.12)
Group velocity (9.12)

Two waves “collide” (9.4)
Two equal amplitude sinusoidal traveling waves going in opposite directions are
superposed (9.4)

Phase velocity, angular frequency, & wavenumber (9.3)
Traveling waves & standing waves (9.4)
Magnitudes of electric & magnetic fields for em waves in vacuum (9.5)
Directions of E, B, & S for em waves (9.5)
ẏ, B, H , I , & v (9.5–9.8)
FL, E , V , & P′ (9.5–9.8)
μ, μ0 , μ , L0 , & ρ0 (9.5–9.8)
T , 1/ε0 , 1/ε , 1/C0 , & γ P0 (9.5–9.8)
H & B (9.6)
Maxwell’s equations in vacuum & in linear materials (9.6)
Density, pressure, displacement, & velocity components of a sound wave (9.8)
Fundamental frequency for a wind instrument, length of the air column, speed of sound,
whether one or both ends are open (9.9)
Nodes in the pressure wave & antinodes in the displacement wave for a wind
instrument (9.9)
rms amplitude & average value (9.10)
Graph of the dispersion relation, group velocity, & phase velocity (9.12)
You should understand the differences between:

The functional forms of left- & right-traveling waves (9.2)
v & vs (9.8)
Power & intensity (9.10)

The two quantities isomorphic to E and B are in phase for right-traveling waves and
out of phase for left-traveling waves. (9.5)
The energy, power, or intensity is proportional to amplitude.2 (9.10)

Take a partial derivative. (9.2)
Test a proposed solution to a differential equation involving partial dervatives. (9.2)
Recognize an isomorphism based on (9.5.8) and either (9.5.9) or (9.5.14).
Use isomorphisms to quickly copy results for em waves in vacuum to other
systems (9.5–9.8)
Calculate the speed of sound for a gas, liquid, or solid (9.8)
Calculate the resonant frequencies for a wind instrument with one or both ends
open (9.9)
Convert between dB and power or amplitude ratios (9.11)
Calculate the group velocity from the dispersion relation (9.12)
Explain the difference between the group velocity and the phase velocity (9.12)
Problems

9.1 (a) Use the fact that any right-travelling wave can be expressed as y (x − vt)
∂y
to show that ẏ = −v for a right-traveling wave.
∂x
(b) Use the fact that any left-travelling wave can be expressed as y (x + vt)
∂y
to show that ẏ = v for a left-traveling wave.
∂x
9.2 The left end of a long rope with μ = 0.1 kg/m and T = 50 N is moved in
the pattern shown in figure 9.P.1, creating a pulse which then propagates to
the right along the rope. The left end of the rope is at x = 0. Make a sketch
of the rope from 0 to 20 m at t = 0.50 s. On your sketch, label the width
and height of the pulse quantitatively, as well as the position of the leftmost
point on the pulse.
Figure 9.P.1 The left end of a rope is

moved in the pattern shown here.
9.3 In a region of outer space, the electric and magnetic fields are described by
E (x , y, z, t) = E0 sin (kx − ωt) ĵ B (x , y, z, t) = B0 sin (kx − ωt) k̂
The electric field magnitude is greater than zero for 0 < x < 0.4 m, goes to
zero at x = 0.4 m, and is negative for 0.4 m < x < 0.8 m.
(a) Using the convention that the length of the vector indicates the
strength of the field, sketch the electric and magnetic fields for points
along the x-axis from the origin to x = 1.6 m, at t = 0. Label your
sketch quantitatively.
(b) In which direction is the wave propagating? Explain how you
can tell.
(c) What is the value of k?
(d) What is the value of ω?
(e) If E0 = 5 × 10−5 N/C, what is the value of B0 ?
(f) Consider the following four points, with coordinates (x, y, z)
measured in meters: Point 1: (1, 0, 0). Point 2: (1, 0, 1). Point 3:
(1.1, 0, 1). Point 4: (1.1, 0, 0). At t = 0, rank the electric field at
these points from largest to smallest. If the field is zero at any point,
state that explicitly. If the field is equal at two or more points, state
that explicitly.
(g) Now, make a similar ranking for the magnitude of the magnetic field
at points 1-4.
(h) Now consider the following other points (again with dimensions in
meters): Point 5: (1, 1, 1). Point 6: (1, −1, 1). Point 7: (1, −1, −1).
Point 8: (1, 1, −1). Specify the strength and direction of the electric
and magnetic fields at each of these points at t = 0.
(i) A completely different wave is also sinusoidal, and also has
magnetic and electric fields that depend only on x and t. At t = 0,
the magnetic field at the origin is B0 ĵ and the electric field at the
origin is E0 k̂, where B0 and E0 are both positive. In which direction
is the wave propagating? How can you tell?
9.4 A wave pulse is traveling to the right on a rope with μ = 0.100 kg/m and
T = 20.0 N. The pulse consists of two straight sloping sections, with a sharp
peak at the center. (To the left of the peak, the rope has a positive slope, and
to the right of the peak it has a negative slope.) Both sloping sections are at
◦
angles of 30 relative to the horizontal. The peak height is 3.00 cm. Make a
sketch showing the transverse velocity (i.e., the velocity in the y-direction)
along the length of the rope, and label your sketch quantitatively, including
the numerical value for the maximum velocity.
9.5 From the isomorphism between rope waves and em waves in a vacuum,
T 1
we have that vp = is isomorphic to c = √ . We cannot tell from
μ μ0 ε0
this what the individual isomorphisms are between the quantities T and μ
on one hand and the quantities μ0 and ε0 on the other. However, since μ
is associated with the momentum μẏ (which is isomorphic to B) and μ0 is
associated with B, one might expect that μ might be isomorphic to μ0 . This
would then require that T is isomorphic to 1/ε0 , which makes sense, since
T is associated with the force from the left (which is isomorphic to E ), and
ε0 is associated with E. Note that we have not proved this isomorphism.
However, check it by applying it to the Poynting vector for an em wave in
vacuum. What should this translate into for a rope wave? Does it translate
correctly?
9.6 For an em plane wave in vacuum, assume that E is parallel to ĵ and B is
B ⇀ d $
parallel to k̂.ApplyAmpère’s law B · d ℓ = μ0 Inet +μ0 ε0 E · n̂ dA
threading
dt
∂B ∂E
to a rectangular loop in the x − z plane to show that = −μ0 ε0 . This
∂x ∂t
is equation (9.5.9) that we used to derive the wave equation for em waves.
Hint: The derivation of the above equation is essentially the same as the
derivation of equation (9.5.8).
9.7 Circular polarization. (a) The “polarization” of an em wave refers to
the direction of the electric field. Thus, for the em waves we’ve explicitly
considered, the polarization is along the y-axis. However, this is not at all
required. For example, we could have the polarization along the z-axis and
still have a wave propagating in the x-direction. The electric field for such a
wave would be given by E = E0 cos(kx − ωt) k̂ , where E0 > 0. Explain why
the magnetic field for this wave would be given by B = −B0 cos(kx − ωt) ĵ ,
where B0 > 0. (Your explanation need not be very mathematical; you should
include a diagram or two.)
For the rest of the problem, consider the following electric field (in
vacuum):
E = E0 cos(kx − ωt) ĵ + E0 sin(kx − ωt) k̂
(b) Qualitatively describe the behavior of the above electric field, as seen
by an observer at the origin. (Hint: radiation with this type of electric field
is called “circularly polarized.”) (c) How must k be related to ω? (d) What
is the magnetic field that must accompany the above electric field? Be sure
to specify both the magnitude (in terms of E0 and known constants) as
a function of position and time, and the direction, using the unit vectors
î, ĵ, and k̂. The best way to accomplish both these tasks will be to use a
Cartesian representation, as I did for E. Explain your reasoning briefly.
9.8 Waves in three dimensions. We have seen that waves in a one-dimensional

system are described by the wave equation
∂ 2y ∂ 2y
vp2 = ,
∂ x2 ∂ t2
where y = y (x , t). The function y might describe the displacement of a rope,
but we have also seen that this equation describes other types of waves, such
as em waves. We have seen that the solutions of this equation
" that represent
1 #
right-moving waves can be written as y x − vp t = y kx − kvp t =
" # k
1
y (kx − ωt) . Written in this way, we can think of the multiplication by
k
1
as part of the action of the function; this means that we could either think
k
of the generic form of a right moving wave as any function y x − vp t , or
as any function f(kx − ωt). For example, a right-moving sinusoid can be
written as f(x , t) = A sin (kx − ωt). This gives a wave with wavenumber k,
2π ω
wavelength λ = , angular frequency ω, and phase velocity vp = .
k k
For a three-dimensional medium which is isotropic (i.e., in which all
directions are equivalent), the wave equation becomes

2
∂ f ∂ 2f ∂ 2f ∂ 2f
vp2 2
+ 2
+ 2
= 2,
∂x ∂y ∂z ∂t
where now the “thing that is waving” (perhaps the displacement of the rope
or the electric field) is f (x , y, z, t). We can write this in shorter form by
introducing the “Laplacian” operator (named in honor of P. S. de Laplace):
∂2 ∂2 ∂2
∇2 ≡ + + .
∂ x2 ∂ y2 ∂ z2
⇀ ⇀
This is also often called “del squared,” since it can be thought of as ∇ · ∇ ,
⇀ ⇀ ∂ ∂ ∂
where ∇ is del, also known as the gradient operator: ∇ ≡ î + ĵ + k̂ .
∂x ∂y ∂z
With this, we can write the three-dimensional wave equation for isotropic
media as
vp2 ∇ 2 f = f¨ .
(a) Show that f (k · r − ω t) is a solution to this equation, where k =

kx î + ky ĵ + kz k̂ is the three-dimensional version of the wavenumber,
and is called the “wave vector.” (Of course, r = x î + y ĵ + z k̂.) In the
process, find the relation between kx , ky , kz , ω, and vp .
(b) Describe qualitatively what the components of k represent. Hint:
Consider the limiting cases when one of the three is equal to 1 m−1 ,
and the other two equal zero.
9.9 Phase relationships for left-traveling sound waves. In section 9.8, we
showed that, for a right-traveling sound wave, the pressure, density, and
◦
velocity are all in phase, while the displacement is 90 out of phase with
them. What are the phase relationships between these four components of
the wave for a left-traveling wave?
9.10 Standing waves of sound. We know that we can create a standing wave by
superposing left- and right-traveling waves of equal amplitudes.
δ0
(a) Superpose two traveling displacement waves of amplitude to
2
show that the displacement and velocity components of a standing
wave of sound are given by
∂δ
δ = δ0 sin kx cos ωt and v ≡ = −δ0 ω sin kx sin ωt .
∂t
(b) For a standing wave that is one wavelength long, sketch δ at t = 0,
and indicate the displacement vectors in a way similar to the vectors
shown in figure 9.8.1. From your sketch, explain why the density
wave must be ρ ′ = −ρm cos kx cos ωt.
(c) Recalling that the potential energy is associated with the pressure
wave, which is always in phase with the density wave, explain why
your results mean that at the time when the kinetic energy of the
standing wave is maximized, the potential energy is zero, and vice
versa.
9.11 Potential energy in sound waves. The potential energy of a sound wave
depends on the pressure. At the time of maximum pressure amplitude for
a standing wave, the distribution of pressure over a wavelength is exactly
the same as it is over a wavelength of a traveling wave. Therefore, the
potential energy of one wavelength’s worth of a standing wave is the same
as the potential energy of one wavelength’s worth of a traveling wave. In
problem 9.10, you can show that the kinetic and potential energies for a
standing wave of sound are out of phase, that is, that at the time when
the kinetic energy of the standing wave is maximized, the potential energy
is zero, and vice versa. By conservation of energy, this means that the
maximum kinetic energy of one wavelength’s worth of a standing wave
equals the maximum potential energy, which by the above reasoning equals
one wavelength’s worth of a traveling wave’s potential energy. (a) For
an air tube of length L and cross-sectional area A that is closed at both
ends, the standing wave in the velocity would be (see problem 9.10)
v = −δ0 ω sin kx sin ωt = −v0 sin kx sin ωt, where v0 is the amplitude of the
velocity wave. Use this to show that, for the mode with λ = L, the maximum
kinetic energy of the air tube is Kmax = 41 ρ0 Av02 L . Hint: Start by dividing
the tube into slices of infinitesimal thickness dx along the length of the tube,
and finding the maximum kinetic energy dKmax of a slice.
(b) Explain why this means ' (that the average potential energy intensity of a
traveling sound wave is SU = 14 vs ρ0 v02 .
9.12 Make yourself a mug of hot cocoa. Holding the mug by the handle, stir the
cocoa vigorously, then use your spoon to tap–tap–tap on the rim of the mug.
You should hear a musical pitch that changes over time. You can stir the cocoa
again and repeat the experiment. (If you are unable to do this yourself, you
can watch a video of me doing it by going to the entry for this problem on the
website for this text. However, it’s a lot more fun to do it yourself.) Explain
qualitatively what’s going on.
9.13 Is each of the following statements true or false? If true, explain briefly. If
false, briefly explain why, and provide a corrected version that is not simply
a negation of the original.
(a) In a vacuum, em radiation is carried in both the magnetic and electric

fields. The amount of energy carried in the electric field is sometimes
more (and therefore sometimes less) than the amount carried in the
magnetic field.
(b) If the electric field points along the −y axis and the magnetic field
points along +z axis, then the Poynting vector (which shows the
direction of wave propagation) points along the −x axis.
(c) Since we can superpose a left-moving wave with a right-moving
wave to create an expression for a standing wave, we can always
express any traveling wave as the sum of two or more standing
waves.
(d) An em wave is traveling in a linear material. When it enters
a different material that has the same permeability but a larger
permittivity, its velocity increases.
9.14 Power in transmission line waves. From the isomorphism between trans-
mission line waves and em waves in vacuum, we have that L0 C0 is
isomorphic to μ0 ε0 . We cannot tell from this what the individual isomor-
phisms are between the quantities L0 and C0 on one hand and the quantities
μ0 and ε0 on the other. However, since L0 is associated with the current
(which is isomorphic to B/μ0 ), and μ0 is associated with B, one might
expect that L0 might be isomorphic to μ0 . This would then require C0 is
isomorphic to ε0 , which makes sense since C0 is associated with the voltage
(which is isomorphic to E ), and ε0 is associated with E. Check this proposed
isomorphism by applying it to the Poynting vector for an em wave in vacuum.
What should this translate into for a waves on a transmission line? Does it
translate correctly?
9.15 Power in a rope wave. A traveling rope wave carries both kinetic and
potential energy. In this problem, you will find the total power carried by the
wave. Consider a sinusoidal wave traveling to the right y = A cos(kx − ωt).
The rope has tension T and mass per unit length μ.
(a) Show that the kinetic energy in one wavelength’s worth of this wave
is 14 λμA2 ω2 . Hint: you may as well make the calculation at t = 0.
Also, recall that the average value of the square of a sinusoid is 1/2 .
(b) Calculating the potential energy for a traveling wave is more
difficult, so we will use a trick. The potential energy is determined
by the shape, and the shape of one wavelength’s worth of a
traveling wave is the same as one wavelength’s worth of a standing

wave (at a moment when the wave is at its maximum amplitude).
Therefore, the potential energy of one wavelength’s worth of a
traveling wave is the same as the maximum potential energy of
one wavelength’s worth of a standing wave. Explain why this is the
same as the maximum kinetic energy of one wavelength’s worth of a
standing wave.
(c) Show that the maximum kinetic energy in one wavelength’s worth
of a standing rope wave of amplitude A is 41 λμA2 ω2 .
(d) Show that the power in a traveling rope wave is P = 21 μA2 ω2 vp .
9.16 Completing the isomorphism between sound waves and em waves in
γ P0
vacuum. For sound waves in a gas, we have vs = , which is
ρ0
1
isomorphic to c = √ . We cannot tell from this what the individual
μ 0 ε0
isomorphisms are between the quantities γ , P0 , and C0 on one hand and the
quantities μ0 and ε0 on the other. However, since ρ0 is associated with the
momentum-per-volume ρ0 v (which is isomorphic to B), and μ0 is associated
with B, one might expect that ρ0 might be isomorphic to μ0 . This would
then require that γ P0 is isomorphic to 1/ε0 , which is reasonable since P0
is associated with the pressure variation (which is isomorphic to E ), and
ε0 is associated with E. By considering the power transmitted in the wave,
show that this hunch is correct. Hint: Note that in (9.11.3), S = 12 vs ρ0 v02 ,
the quantity given is the average intensity.
9.17 The 287th Annual Solar System Olympics is being held on Venus, where
conditions are rather different from here. On Earth, the atmospheric pressure
is 101.3 kPa. The 100-m dash is held in Sandworm’s Borough Stadium on
the surface of Venus, where the atmospheric pressure is 92 times that of Earth
and the atmospheric density is 65 kg/m3 . The atmosphere is mostly CO2 ,
◦
and is at a temperature of 400 C. Assume the adiabatic index (γ ) of this
atmosphere is 1.235. The starter, standing at the start line, fires the starting
pistol.
(a) How long will it take for the sound of the gun going off to reach a
timer standing at the finish line?
(b) What is the rms amplitude of pressure in the sound wave that comes
out of the barrel of the gun, if the sound intensity measured right
next to the muzzle is 140 dB?
(c) After a blistering 8.69 s, a human crosses the finish line ahead of
all of the other competitors. We are now at the medal ceremony,
and the Aphrodite Orchestra is playing Joe Diffie’s “Third Rock
from the Sun,” the planetary anthem of Earth. The flautist, who
happens to be from Pluto, has placed an extension onto the end of
her flute to take into account the atmospheric conditions on Venus.
Assuming the conditions from part a still apply, how long must the
flute become if the flute is tuned so that its first harmonic occurs
at 880 Hz? How long must a clarinet be if it were to fit the same
characteristics?
9.18 A wave y1 = A cos(k1 x − ω1 t) is superposed with a wave y2 = A
cos(k2 x − ω2 t), where k1 is fairly close to k2 . (a) What is the group velocity of
the resulting waveform, that is, the velocity of the envelope? (b) What is the
wavelength of the rapidly oscillating function contained within the envelope?
(c) What is the speed of the crests of the rapidly oscillating function?
9.19 A Gaussian wavepacket of light travels through a sheet of glass which
has a nonlinear dispersion. Explain what is wrong with the following
statement: “When the wavepacket emerges from the glass, the high-
frequency components are shifted to the front of the packet, and the
low-frequency components are shifted to the back.”
9.20 For waves in deep water (depth at least as large as the wavelength), with
amplitude much less than the wavelength, the dispersion relation is ω2 =
γ k3
gk + . The gk part of this dominates at small k (i.e., at large wavelength),
ρ
and describes the situation when the restoring force is provided by gravity
(g is the acceleration of gravity). Waves in this regime are called “gravity
γ k3
waves.” The dominates at large k, and describes the situation when
ρ
the restoring force is provided by surface tension; γ is the surface tension
(7.2 × 10−2 N/m for pure water), and ρ is the density (1,000 kg/m3 ). Waves
in this regime are called “capillary waves.” (a) At what wavelength do these
two terms make equal contributions? (b) Use a graphing program to make
a plot of ω as a function of k for wavelengths ranging from 5 mm to 20 cm.
On the same graph show separate curves for pure gravity waves and pure
capillary waves as dashed lines, and label each of the three curves. (c) For
gravity waves (waves with a wavelength long enough that surface tension
effects are negligible), what is the phase velocity in terms of g and k? (d) For
gravity waves, show that the group velocity for waves having wavenumbers
in a range centered on k is half the phase velocity. (e) For capillary waves
(which have a wavelength short enough that the gk term can be ignored),
what is the phase velocity in terms of k, γ , and ρ ? (f) For capillary waves,
show that the group velocity for waves having wavenumbers in a range
centered on k is 3/2 the phase velocity.
9.21 Em waves in the ionosphere. (This is an example of a problem that looks
much more intimidating than it really is.) So far, we have considered em
waves in vacuum. One of the layers of the Earth’s atmosphere is made of
mostly ionized gas, and is called the “ionosphere.” It is responsible for long-
range transmission of AM radio, since the radio waves bounce off this layer
and then back down to Earth. Suppose that em waves in the ionosphere are
governed by the following differential equation:
∂ 2E ∂ 2E
2
= c2 2 − ω02 E ,
∂t ∂x
where c and ω0 are known constants. You are told that all possible solutions
of this equation can be expressed in the form
∞
E (x , t) = ε (k) ei(kx−ωt) dk ,
−∞
where ε (k) is a function of k.

(a) A naïve student might think that the above could only represent
right-moving waves. Explain why this is incorrect.
(b) Given that the above is a solution to the differential equation, what
must be the dispersion relation? Hint: rearrange the differential
equation so that you have zero on one side. Then, bear in mind that
integrating with respect to k commutes with taking the derivative
with respect to x or t. Finally, if you can make the integrand zero,
then the integral will definitely be zero.
(c) A burst of sine waves with a small range of angular frequencies
centered on ω1 is emitted from the point x = 0. The burst has a
duration τ . Approximately how long will it take the burst to reach
the point x = L?
10 Waves at Interfaces
Come with me if you want to live!

—Arnold Schwarzenegger as the Terminator in Terminator 2: Judgment Day
So far, we have discussed waves traveling through an unchanging medium. However,

interesting and surprising things happen at the interface between two different
mediums, leading to a wealth of important science and applications. In this chapter,
we begin by studying the behavior of wave pulses on ropes when they encounter a
boundary at which the mass/length of the rope suddenly changes. Although this may
not sound very exciting, bear in mind that waves on a rope are exactly analogous to
waves in any elastic medium. Therefore, for example, by understanding how transverse
waves on a rope behave, we immediately discover the behavior for torsional waves
in solids. For learning purposes, waves on ropes have the distinct advantage that they
are easily visualized and relatively easy to understand intuitively. Furthermore, the
methods we use for determining the behavior of rope waves at interfaces are almost
exactly the same as those you will use in a future course to understand the behavior of
quantum waves at interfaces, a most important problem indeed. We will also explicitly
discuss the behavior at interfaces of sound waves, waves on transmission lines, and
electromagnetic waves in a vacuum. The work we do with rope waves will serve us
well when we reach these other, perhaps more interesting, types of waves, since we
will again find isomorphisms for the behavior at interfaces.
10.1 Reflections and the idea of boundary conditions
In section 1.11, we discussed the simplest type of quantum mechanical wave, (x , t) =

ψ0 e−iωt eikx . We showed that an electron described by such a wave function is
completely delocalized, that is, it is equally probable to be found at any point in space.
Often, we are interested in electrons which are more localized; a possible wavefunction
for such an electron is shown schematically in figure 10.1.1a. (Recall that the quantum
mechanical wavefunction is inherently complex; only the real part is shown in the
figure.) This “wave packet” is travelling in a region of constant potential energy U,
but is about to encounter a place where U suddenly increases. We will not treat this
343
Figure 10.1.1 A Quantum mechanical

wavepacket approaching a place where
the potential energy suddenly changes
(a) is analogous to a rope wavepacket
approaching a place where the
mass/length suddenly changes (b).
problem quantitatively in this book, but it is quite analogous to the problem of a wave
packet on a rope wave shown in figure 10.1.1b.
At the moment shown, the wave packet (or pulse) is traveling on a rope with
low mass/length μ1 , and is about to encounter an interface at which the mass/length
suddenly increases to a larger value μ2 . (The tension is the same in the two parts of the
rope.) For now, we assume that any conversion of the energy in the wave into thermal
energy is negligible. We then recognize that some of the energy in the pulse might be
transmitted into the heavier part of the rope as a pulse traveling further to the right, but
some might also be reflected as a pulse traveling to the left in the lighter part of the
rope. Our job is as follows: given the shape of the incident pulse, find the shape of the
transmitted and reflected pulses. We define the following:
Table 10.1.1. Functions for rope waves

R1 (x − v1 t) Function describing the incident pulse, which travels to the right in medium 1
(the left part of the rope).
L1 (x + v1 t) Function describing the reflected pulse, which travels to the left in medium 1.
R2 (x − v2 t) Function describing the transmitted pulse, which travels to the right in medium 2
(the right part of the rope).
We want to describe things for all times, including the time before the pulse
encounters the interface, the time during the encounter, and the time after the encounter.
During the encounter, the left part of the rope has both the tail end of the pulse (moving
right) and also the front end of the reflection (moving left). So, in general, the motion
of the rope to the left of the interface (medium 1) is

y1 (x , t) = R1 x − v1 t + L1 x + v1 t , (10.1.1)
! ! !
motion of incident, reflected,
medium 1 right-moving left-moving
wave wave
Chapter 10 ■ Waves at Interfaces 345
while the motion of the rope to the right of the interface (medium 2) is

y2 (x , t) = R2 x − v2 t . (10.1.2)
! !
motion of transmitted,
medium 2 right-moving
wave
So, given the function R1 , our task is to find L1 and R2 . In this section, we’ll find L1 .
The properties of medium 1 determine how fast pulses travel, but place no
constraint on the shape of the pulse. Similarly, the properties of medium 2 determine
only the speed of waves in that medium. So, the details of L1 and R2 must be determined
by the properties of the interface itself, combined, of course, with the details of R1 .
There are two important properties of the interface; these are referred to as “boundary
conditions”:
1. At the interface, the rope must be continuous. We set x = 0 at the interface, so
this condition is
y1 (0, t) = y2 (0, t) (10.1.3)

⇒ R1 + L1 x=0 = R2 x=0 .
Taking the time derivative of this gives

" #
∂ R1 ∂L ∂ R2
+ 1 = . (10.1.4)
∂t ∂ t x=0 ∂ t x=0
2. Figure 10.1.2 shows the interface at a time when the pulse is passing through it.
We consider an infinitesimal segment of the rope, at the interface. This segment
has infinitesimal mass. Therefore, to avoid infinite acceleration, the net force
applied to it must be zero. It experiences a force T2 from the right part of the
rope and a force T1 from the left part. For these forces to cancel, they must
point in opposite directions. Since the tension force in a rope points along the
rope, this means that

∂ y1 ∂ y2
= (10.1.5)1
∂ x x=0 ∂ x x=0
+ ,
∂ R1 x − v1 t ∂ L1 x + v1 t ∂ R2 x − v2 t
⇒ + = . (10.1.6)
∂x ∂x ∂x
x =0 x =0

Since R1 is a function of the combination x − v1 t , we have that

∂ R1 x − v1 t ∂ R1 x − v1 t
= . (10.1.7)
∂x ∂ −v1 t
1. The same combination of boundary conditions, that is, continuity of the wavefunction (10.1.3)
and continuity of its derivative (10.1.5) are used to solve the equivalent problem in quantum
mechanics.
Figure 10.1.2 Top: Forces on an infinitesimal segment of rope at the interface must cancel, to
avoid infinite acceleration. Bottom: As we’ll see later in this chapter, reflection of light off the
surface of water is closely analogous to reflection of rope waves off the interface between a
light rope and a heavy one. This shows the 3rd Avenue bridge in Minneapolis. Image ©
Geoffrey Kuchera | Dreamstime.com
Now, we use the chain rule:
−1
∂ R1 x − v1 t ∂ R1 x − v1 t dt ∂ R1 x − v1 t d −v1 t
= =
∂ −v1 t ∂t d −v1 t ∂t dt

1 ∂ R1 x − v1 t
=− .
v1 ∂t
Combining this with equation (10.1.7) gives

∂ R1 x − v1 t 1 ∂ R1 x − v1 t
=− .
∂x v1 ∂t
Applying this, and the analogous expressions for R2 and L1 to equation (10.1.6) yields
" #
1 ∂ R1 1 ∂ L1 1 ∂ R2
⇒ − + =− .
v1 ∂ t v1 ∂ t x=0 v2 ∂ t x=0

∂ R2
Using equation (10.1.4) to substitute for in the above, gives
∂ t x=0
" # " #
1 ∂ R1 1 ∂ L1 1 ∂ R1 ∂ L1
− + =− + .
v1 ∂ t v1 ∂ t x=0 v2 ∂ t ∂ t x=0
Integrating with respect to time, and multiplying by −v1 v2 , we obtain

v2 R1 − v2 L1 = v1 R1 + v1 L1 + const. x=0 ⇔

v2 − v1 R1 = v2 + v1 L1 + const x=0 .
This equation must hold true under all conditions. In particular, it must be true when
the rope is at equilibrium, that is, when R1 = L1 = 0. Therefore, the constant must
equal zero. So, finally, we get
" #
v2 − v1
L1 = R1 . (10.1.8)
v2 + v1 x=0
Core example: Applying the reflection " equation. #

1
We consider the case v1 = 3v2 , so that L1 = − R1 . Each part of the incident pulse
2 x =0
causes the corresponding part of the reflected pulse, so that the sequence before and
after the encounter with the interface is as shown in figure 10.1.3a. Note that only the
behavior for x < 0 is shown; in section 10.3, we’ll figure out the behavior for x > 0.
Concept test 1 (answer below2 ): For the situation shown in the core example above,
make a sketch of R1 , L1 , and y1 for a time during the encounter of the pulse with the
interface, so that the sketch of R1 looks as shown in figure 10.1.3b. At this time, only
about the first quarter of the pulse has encountered the interface.
Concept test 2 (answer below3 ): If we attach the end of a rope to a wall, and launch a
pulse at this interface (as shown in figure 10.1.5), what does the reflection look like well
after the encounter of the pulse with the interface? Hint: Think about the wall as if it were
a very heavy rope. What is the speed in such a rope?
The opposite limit, of v2 → ∞, is more difficult to contrive. We must attach the

left part of the rope to something that has the same tension, but zero mass. The only
way to do this is to use a massless ring, which slides without friction on a pole, as
shown in figure 10.1.7. This arrangement is obviously a bit ridiculous, but the analogs
for other types of waves are important, and we’ll discuss them later in this chapter.
2. The answer to Concept test 1 isshown in figure 10.1.4.

3. The speed of a rope wave is T μ. We think of the wall as a rope with infinite μ, so that

v2 = 0. Using equation (10.1.8) this gives L1 = −R1 x=0 , so the reflection is the same size as
the incident pulse, but inverted, as shown in figure 10.1.6.
Figure 10.1.3 a: The behavior of
the part of the rope to the left of
the interface for the case
L1 = −R1 /2. b: R1 for a time early
in the encounter of the pulse with
the interface. Don’t look yet at the
next figure.
Figure 10.1.4 The reflection of a pulse

during the encounter with the interface.
Figure 10.1.5 A pulse about to hit a brick

wall.
348
Figure 10.1.6 The reflected pulse well

after the encounter with the brick wall.
Figure 10.1.7 Reflection of a pulse off a

massless ring which slides without
friction on a pole.

In this limit, equation (10.1.8) gives L1 = R1 x=0 , so that the reflected pulse has the
same amplitude as the incident pulse, and is not inverted.
10.2 Transmitted waves
Imagine you are a dog holding the left end of a rope which is under tension, as shown in
figure 10.2.1. Suddenly, you move your head up, launching a right-moving wavefront,
as shown.
The same thing happens at the interface, for a rope wave; the motion of the point
at the interface (the left end of the heavy part of the rope) launches the transmitted
pulse into the right part of the rope.
Figure 10.2.1 If you suddenly move your head up, you will launch a right-moving wavefront.
In this case, we only need the boundary condition that the rope is continuous at
the interface:
y1 (0, t) = y2 (0, t)

⇒ R1 + L1 x=0 = R2 x=0 .
Your turn: Use this, and the results of section 10.1, to show that
" #
2v2
R2 = R1 . (10.2.1)
v2 + v1 x =0
Thus, the transmitted wave has an amplitude relative to the incident wave of
2v2
. Because it travels at speed v2 , it is compressed or expanded in the horizontal
v2 + v1
direction by the ratio v2 v1 .
Core example: Shape of the transmitted wave pulse.

Again, we consider the case v1 = 3v2 . From equation (10.2.1), we see that R2 = R1 2.
We launch the same pulse as before toward the interface. As shown in figure 10.2.2a,
2v2
the transmitted pulse is scaled by the factor in the in the y-direction, and by the
v2 + v1
factor v2 v1 in the x-direction.
Concept test (answer below4 ): Sketch R1 , R2 , L1 , y1 , and y2 at the point in time when
the first quarter of the pulse has encountered the interface, that is, at the time shown in
figure 10.2.2b. Make your sketch as quantitative as possible.
1
Self-test (answer below5 ): Now, let v1 = v , so that the left side of the rope is heavier.
3 2
Use the same incident pulse as is shown in figure 10.2.2. Sketch R1 , R2 , L1 , y1 , and y2 for
a point in time well after the pulse has encountered the interface. Make your sketch as
quantitative as possible.
Sinusoidal waves are particularly important, partly because any wave can be
expressed as a sum of sinusoids, as we saw in chapter 8. What happens to a sinusoidal
4. The answer to the Concept Test is shown in figure 10.2.3.

5. Answer to Self-test: Using equation (10.1.8), the amplitude of the reflected pulse is half
that of the incident pulse, and the reflection is noninverted (i.e., the amplitude is positive).
Using equation (10.2.1), the amplitude of the transmitted pulse is 3/2 that of the incident pulse.
Because v2 is three times v1 , the transmitted pulse is stretched in the horizontal direction by
a factor of three. Putting this all together gives figure 10.2.4. It might at first appear that this
violates conservation of energy, since the transmitted pulse is higher and wider than the incident
pulse. However, since medium 2 is lighter, it takes less energy to create pulses in it.
Figure 10.2.2 a: An incident pulse (top) is partly transmitted and partly reflected (bottom).
Note that the reflected pulse travels three times as fast as the transmitted pulse. b: One
quarter of the way through the encounter. Don’t look yet at the next figure.
Figure 10.2.3 Reflection and transmission of a pulse during the encounter with the interface.
wave when it encounters an interface? As for all other

waves, the horizontal scale
for the transmitted wave is changed by the factor v2 v1 , so that the wavelength in
medium 2 is
v
λ2 = 2 λ1 . (10.2.2)
v1
Figure 10.2.4 Reflection and transmission for a case where the speed is three times greater on
the right side of the interface.
Does the angular frequency change?

ω2 2π v2 2π v 2π v1
v2 = ⇔ ω2 = v2 k2 = = v 2 = = v1 k1 = ω1 .
k2 λ2 2 λ1
λ1
v1
So:
When a sinusoidal wave moves from one medium to another, the wavelength
changes, but the frequency doesn’t change.
We can also see this by considering that each crest of the transmitted wave is launched
by a crest in the incident wave.
10.3 Characteristic impedances for mechanical systems
Concept test (answer below6 ): We saw in section 10.1 that, when a rope wave
encounters a brick wall, it is reflected with full amplitude but inverted. If it instead
encounters a massless ring that slides frictionlessly on a pole, it is reflected with full
amplitude but noninverted. One might expect that there would be some middle ground
where there is no reflection at all, that is, a condition halfway between an inverted and
a noninverted reflection. Perhaps if we attach the rope to a ring that has some carefully
chosen mass, and that slides frictionlessly on a pole, we could suppress the reflection.
Explain, using fundamental physical principles, why this couldn’t work.
As we saw in chapter 9, transmission lines (such as coaxial cables) can support

electromagnetic waves, which behave in many ways like waves on ropes. Thus, we
6. The wave pulse carries energy. Since the ring slides frictionlessly, it cannot dissipate the energy,
so the only way for energy to be conserved is by generating a reflection.
Figure 10.3.1 Top: model for a cable TV

transmitter attached to a coaxial cable.
Bottom: schematic for a cable TV
network; the lines represent coaxial
cables.
can expect that, when a wave pulse travelling along a coaxial cable reaches the end
of the cable, reflections will be generated (just as reflections are generated when a
pulse reaches the end of a rope attached to a wall or a ring). This could create a very
serious problem for the operator of a cable television system. To broadcast on such a
system, waves are launched into the cable, as shown schematically in the top part of
figure 10.3.1. This signal must be sent to many houses, as shown in the bottom part;
if reflections are generated at each one, there will be terrible interference problems,
dramatically reducing picture quality. What is needed is some way to end the cable (as
we must do at each house) but suppress the reflections. The device attached to the end
of the cable that accomplishes this suppression is called a “terminator.”
This is analogous to the problem of suppressing reflections for a rope wave, and
the basic solution is the same in both cases. If instead of ending the cable, we attached
an infinite length of cable of the same type, clearly no reflections would be generated.
So, how can we make a compact device that “feels” like an infinite length of cable?
For a rope, how can we make a compact device that feels like an infinite length of rope
under tension?
For the rope, we will borrow an idea from our earlier study of ac circuits, in
section 1.10. There, we encountered the generalized version of the resistance, called
the impedance. This is defined as Z ≡ Ṽ /Ĩ, where Ṽ is the complex version of the
voltage and Ĩ is the complex version of the current. For a resistor, we have simply
Z = R; because this is real, the current and voltage are in phase for a resistor. However,
1
for a capacitor we found that Z = ; because this is purely imaginary, the current
iω C
and voltage are 90 out of phase for a capacitor. One might say that the impedance
◦
characterizes the way a circuit element “feels.” We will define the equivalent quantity
for a mechanical system, and then show that, if we construct a compact object with the
same impedance it does indeed suppress reflections.
What is the mechanical equivalent of the impedance Z ≡ Ṽ /Ĩ? Recall the
isomorphism between the damped driven mechanical and electrical oscillators:
(4.1.3a) : mẍ + bẋ + kx = F0 cos ωd t

q
(4.6.1) : L q̈ + Rq̇ + = V0 cos ωd t
C
So, the voltage is isomorphic with the applied force, and the current (which equals q̇) is
isomorphic with the velocity ẋ. By analogy with Z ≡ Ṽ /Ĩ, this suggests the following
definition of the mechanical impedance:
F̃applied
Z≡ , (10.3.1)
ż
Definition of mechanical impedance for a compact object
where F̃applied is the complex version of the applied force (e.g., for the damped driven
oscillator, F̃applied = F0 eiωd t ) and ż is the complex version of the velocity ẏ (e.g., for the
d i(ω t −δ)
damped driven oscillator in a steady-state, ż = Ae d ). If Fapplied is in phase
dt
with the velocity ẏ, then Z is real, just as the impedance is real for an electrical circuit
if the voltage is in phase with the current. However, if Z is complex, then Fapplied is
not in phase with ẏ. Since Z may generally be complex, we don’t bother to include the
tilde above it.
Self-test (answer below7 ): For a damped driven harmonic oscillator, what is the
impedance Z? Express your answer in terms of F0 , A, δ , and ωd .
For an extended object such as a rope, we must be a bit more careful with our
definition. Imagine you are holding the left end of a rope in your hand. You can launch
waves down the rope by moving your hand up and down, exerting a force on the tiny
piece of the rope you are actually touching. However, this is not the only force this
piece of rope feels; it also feels a force from the rest of the rope, off to its right. So,
because our definition involves the applied force, we could also write it (for the case
of right-moving waves) as the force coming from the left side of a piece of rope:
F̃L
Z≡ ,
ż
Definition of mechanical impedance for right-traveling waves
where F̃L is the complex version of FL (the y-component of the force exerted on a
piece of rope from its left end), and ż (x , t) is the complex version of the velocity in the
y-direction, that is, ẏ = Re ż. We already know from the arguments of section 9.5 that
FL is in phase with ẏ for a right-moving wave, so Z is real, and we can simply write
FL
Z= (10.3.2)
ẏ
Mechanical impedance for right-traveling waves.
F̃applied F0 eiωd t F F
7. Answer to self-test: Z = = = −ieiδ 0 = ei(δ−π/2) 0 . So, when
ż d i(ωd t −δ ) ωd A ωd A
Ae
dt
δ = π/2 (which occurs when ωd = ω0 ), the impedance is real, meaning that the drive force is
in phase with the velocity, as we already knew.
It is not difficult to calculate Z, since we already know from equation (9.5.15) that
∂y ∂y
FL = −T . So, we need to express ẏ in terms of , so that when we take the ratio we
∂x ∂x
will get a constant. For a right moving wave, the solution to the wave equation is of the
∂y ∂y
form y (x − vt). Therefore, by the chain rule, ẏ = = −v . (For example, if y =
∂t ∂x
∂y ∂y
A sin [k (x − vt)], then = −kvA cos [k (x − vt)], while = kA cos [k (x − vt)].)
∂t ∂x
Therefore,
∂y
−T
Z= ∂ x = T.
∂y v
−v
∂x

T
From equation (9.2.4), v = μ , so that

Zrope = Tμ (10.3.3)
For a left-traveling wave, the solution to the wave equation is of the form y (x + vt),
∂y ∂y
so that ẏ = = +v . Since the impedance should be a property of the rope, and
∂t ∂x
should not depend on the direction of wave travel, we therefore see that we must alter
the definition (10.3.2) for the case of left-moving waves:
FL
Z=− .
ẏ
Mechanical impedance for left-traveling waves.
∂y
−T
Just to check: Z = − ∂ x = T = √T μ, as desired. You may be wondering why, for
∂y v
v
∂x
left-traveling waves, we don’t instead define Z = +Fr /ẏ , where Fr is the y-component
of force applied to a segment of rope from the right. This would be correct, but it is not
as convenient. We wish to build on the isomorphisms developed in chapter 9, which
are all based on FL, so it is better to stick with FL in this chapter as well. Furthermore, it
is easier to discuss boundary conditions when we use the same quantity FL to describe
left- and right-traveling waves.
Let us return now to right-traveling waves. If you hold the left end of the rope,
the force you apply to it is FL. Because the velocity ẏ is in phase with FL, it is in phase
with the force you apply. Thus, the rope “feels” different from a small rock, for which
the acceleration is in phase with the force you apply. The difference comes because
the end of the rope feels not only the force you are exerting now, but also the force
from the part of the rope just to its right. That part of the rope is at a position that
is determined by the propagating wave, that is, that is determined by the force you
exerted on the end of the rope a short time ago.
So, how can we terminate the rope in such a way as to suppress reflections? In
other words, which compact object has the same impedance as an infinite length of
rope, so that it feels the same as an infinite length of rope? In other words, which
compact system has a velocity that is in phase with the applied force? The answer
Figure 10.3.2 By connecting a rope to a massless ring

immersed in a suitably chosen damping fluid, we can
suppress reflections.
is a massless ring, sliding on a frictionless pole, and immersed in a viscous damping

medium. Because the ring is massless, the net force applied to it must be zero (otherwise
its acceleration would be infinite). Therefore, the force applied to it (in this case by the
rope) must be opposite the damping force, that is,
Fapplied = − (−bẏ) = bẏ, (10.3.4)
so that, as we need, the y-velocity is in phase with the applied force. The impedance of
such a ring is the ratio of the force to the velocity, so that Z damped = b, and to suppress
massless
ring
reflections of waves traveling down the rope, we just match impedances (so that the
ring feels like an infinite rope), that is, we choose the size of the ring and the viscosity
of the damping medium so that

b = T μ.
One way to accomplish this would be to have the rope be vertical, and the ring be
immersed in a pool of liquid, as suggested in figure 10.3.2. Note that the damping
provides a way to absorb the energy of incident pulses.
Now, unless you plan to create a telephone company based on tin cans tied
together with strings, the previous discussion might seem rather contrived. However, in
section 10.4 we will develop the ideas needed to determine reflection and transmission
for all the types of waves we have studied, making use of the isomorphisms uncovered
in chapter 9. Then, in section 10.5, we apply these ideas to the more practical
considerations of how to suppress reflections in transmission lines (such as coaxial
cables).
10.4 “Universal” expressions for transmission and reflection
We saw in section 9.5 that the displacement wave y (x , t) is not isomorphic to any of the
quantities that characterize other wave types. Instead, the velocity wave ẏ is isomorphic
to, for example, B/μ0 for an electromagnetic wave in vacuum, and FL is isomorphic
to, for example, the electric field E of an electromagnetic wave in vacuum. We will see
that all these isomorphisms remain faithful when we consider the boundary conditions
for each of the various types of waves. Therefore, it is worthwhile for us to consider
the reflection and transmission for the ẏ wave and the FL wave. However, for the most
general case we must allow the tension T to be different in the two parts of the rope.
To accomplish this, we imagine that the two halves are connected by a massless ring
which slides frictionlessly on a pole, so that the pole can supply the force needed for
the tensions to be different. To the left of the pole we have rope 1, and to the right we
have rope 2.
What is the boundary condition on FL? Consider the forces exerted on the massless
ring. The force exerted on the ring by the section of rope to its left is, of course, FL1 ,
where “1” in the subscript indicates the force is exerted on the left side of the interface.
The ring exerts a force FL2 on the rest of the rope to its right, where “2” indicates
the force is exerted on the right side of the interface. By Newton’s third law, this is
opposite to the force exerted by the rope to its right on the ring, so that the net force
(in the y-direction) exerted on the ring is FL1 − FL2 . Since the ring is massless, the net
force must be zero (otherwise its acceleration would be infinite), so that

FL1 = FL2 x=0 . (10.4.1)
Boundary condition for FL
In other words, the boundary condition for FL is that it must be continuous at the
interface.
As we did when considering the displacement y of the rope, we can express FL
in rope 1 (to the left of the interface) as the sum of the right-traveling incident wave
FL, R1 and the left-traveling reflected wave FL, L1 :
FL1 = FL, R1 + FL, L1 .
On the right side of the interface, there is only the transmitted wave:
FL2 = FL, R2 .
Therefore, the boundary condition becomes

FL, R1 + FL, L1 = FL, R2 . (10.4.2)
x =0
What isthe boundary

condition for ẏ? We know that the rope itself must be continuous,
that is, y1 = y2 x=0 . Taking the time derivative of this gives

ẏ1 = ẏ2 x=0 . (10.4.3)
Boundary condition for ẏ
Expressing this in the R1, L1, R2 notation, we have

ẏR1 + ẏL1 = ẏR2 x=0 . (10.4.4)
We will begin by finding the reflected and transmitted ẏ waves. For notational
convenience, in what follows we omit the explicit reminders that we are considering
the behavior at the interface x = 0. Dividing equation (10.4.2) by (10.4.4) gives
FL, R1 + FL, L1 FL, R2
= .
ẏR1 + ẏL1 ẏ R2
FL, R2
From equation (10.3.2), Z2 = , so that above becomes
ẏR2

FL, R1 + FL, L1 = Z2 ẏR1 + ẏL1 . (10.4.5)
From equation (10.3.4), we have that, for the reflected wave (which travels left)
FL, L1
Z1 = − ⇔ FL, L1 = −ẏL1 Z1 .
ẏL1

FL, R1 − ẏL1 Z1 = Z2 ẏR1 + ẏL1 .
Dividing by ẏR1 gives

FL, R1 ẏL1 ẏ ẏ
− Z1 = Z2 1 + L1 ⇒ Z1 − Z2 = L1 Z1 + Z2 ⇒
ẏ ẏR1 ẏR1 ẏR1
R1 !
Z1
Z1 − Z2
ẏL1 = ẏR1 , or (10.4.6)
Z1 + Z2
" #
Zi − Zt
ẏreflected = ẏincident , (10.4.7)
Zi + Zt x=0
where Zi = Z1 is the impedance for the rope on the incident side and Zt = Z2 is the
impedance for the rope on the transmitted side.
Your turn: Use equation (10.4.6) to substitute for ẏL1 in equation (10.4.4) and show that
" #
2Zi
ẏtransmitted = ẏincident , (10.4.8)
Zi + Zt x =0
Concept test (answer below8 ): A right-traveling pulse gets to the end of a rope, which
is attached to a massless ring which slides frictionlessly on a pole. Is the reflected ẏ pulse
inverted or not with respect to the incident ẏ pulse? Use the results of this section to get
your answer.
For FL, the expressions for reflection and transmission are different. The reflected
wave moves left in rope 1, so
FL, L1 FL, R1
Z1 = − ⇔ FL, L1 = −ẏL1 Z1 = −ẏL1 .
ẏL1 ẏR1
8. The massless ring has zero impedance (since an infinitesimal force in the y-direction is all that’s
needed to produce a velocity and Z = FL ẏ). Plugging Z2 = 0 into equation (10.4.6) gives
[ẏreflected = ẏincident ]x=0 , so the reflected velocity pulse is uninverted.
Using equation (10.4.6) to substitute for ẏL1 , we then have that, at the interface,
Z1 − Z2
FL, L1 = −FL, R1 or
Z1 + Z2
" #
Z − Zi
FL, reflected = FL, incident t . (10.4.9)
Zi + Zt x=0
The transmitted wave moves right in rope 2, so

FL, R2
Z2 = ⇔ FL, R2 = ẏR2 Z2 .
ẏR2
Using (10.4.8) to substitute for ẏR2 , we get that, at the interface

2Z1
FL, R2 = ẏR1 Z2 .
Z1 + Z2
FL, R1 FL, R1
Using Z1 = ⇔ ẏR1 = , we obtain
ẏR1 Z1

FL, R1 2Z1 2Z2
FL, R2 = Z2 = FL, R1 , so that
Z1 Z1 + Z 2 Z1 + Z2
" #
2Zt
FL, transmitted = FL, incident . (10.4.10)
Zi + Zt x=0
Concept test (answer below9 ): As in the previous concept test, a right-traveling pulse
gets to the end of a rope, which is attached to a massless ring which slides frictionlessly
on a pole. Is the reflected FL pulse inverted or not with respect to the incident FL pulse?
Again, use the results of this section to get your answer.
10.5 Reflected and transmitted waves for transmission lines
Now, let’s consider what happens when a wave pulse on a transmission line encounters
an interface where the properties of the line (such as the inductance per length L0 or
the capacitance per length C0 ) change suddenly, as suggested in figure 10.5.1. As for
Figure 10.5.1 An interface between two

types of coaxial cable.

9. Plugging Zt = 0 into equation (10.4.9), we see that FL, reflected = −FL, incident , so the
x =0
reflected FL pulse is inverted, and thus is opposite in phase to the reflected ẏ pulse.
a rope, we can anticipate that there will be a reflected pulse and a transmitted pulse.
We will be able to use our isomorphisms to find the expressions for the reflected and
transmitted pulses with no additional work, after checking to make sure the boundary
conditions are the same. We have previously made isomorphisms between rope waves
and electromagnetic waves in vacuum, and between electromagnetic waves in vacuum
and waves on transmission lines:
Table 10.5.1. Isomorphism between rope waves, em waves in vacuum, and waves on a
transmission line
Rope waves Electromagnetic waves Waves on a transmission

in vacuum line
Transverse velocity ẏ B/μ0 I

y-component of force from Electric field E V
∂y
the left FL = −T
∂x
Mass per length μ Permeability of free space μ0 Inductance per length L0
Tension T Inverse of permittivity of Inverse of capacitance per
free space 1/ε0 length 1/C0
Therefore, ẏ for the rope wave is isomorphic to I for the transmission line wave, and
so on.
We have two boundary conditions:
1 The voltage at the left side of the interface must equal the voltage on the right
side, that is,

Vi = Vt x=0 . (10.5.1)
From equation (10.4.1), this is the same as the boundary condition for the
isomorphic quantity FL: FLi = FLt x=0 .
2 The current at the left side of the interface must equal the current at the right
side of the interface, that is,

Ii = It x=0 .
From equation (10.4.3),

this is the same as the boundary condition for the iso-
morphic quantity ẏ, ẏi = ẏt x=0 . Therefore, the isomorphism is complete even
for reflections and transmissions, and we can immediately make use of the iso-
morphic expressions for transmission and reflection for a rope wave. To do so,
we need the expression for the impedance of the transmission line. For the rope
√
Z = T μ, so, reading off the isomorphism, the impedance for the transmission
line is

Z= L 0 C0 . (10.5.2)
FL
By equation (10.3.2), for the rope wave Z = for a right-moving wave, so for waves
ẏ
on transmission lines
V
Z= , (10.5.3)
I
Impedance for right-moving waves on transmission lines
in accordance with our expectations for the impedance of electrical circuits.

The voltage wave on the transmission line is isomorphic to FL. So, translating
equations (10.4.9) and (10.4.10), we have
" #
Zt − Zi
Vreflected = Vincident . (10.5.4)
Zi + Zt x=0
Reflection for the V of a transmission line wave.
and
" #
2Zt
Vtransmitted = Vincident . (10.5.5)
Zi + Zt x =0
Transmission for the V of a transmission line wave.
The I wave on the transmission line is isomorphic to ẏ. So, translating equations (10.4.7)
and (10.4.8) we have
" #
Zi − Zt
Ireflected = Iincident , (10.5.6)
Zi + Zt x =0
Transmission for the I of a transmission line wave
" #
2Zi
Itransmitted = Iincident , (10.5.7)
Zi + Zt x=0
Reflection for the I of a transmission line wave
Concept test (answer below10 ): A right-travelling wave gets to the end of a coaxial
cable, at which point a wire has been attached connecting the inner conductor to the
outer conductor. Is the reflected current pulse inverted or not with respect to the incident
current pulse?
10. The wire connecting the inner to the outer conductor has zero impedance, so Zt = 0. Plugging

this into equation (10.5.6), we see that Ireflected = Iincident x=0 , so the reflected current pulse is
uninverted.
Figure 10.5.2 a: Schematic picture of a

terminator for a transmission line. A
resistance equal to the characteristic
impedance of the transmission line is
connected between the two conductors.
This “feels” like an infinite length of
cable, and so generates no reflections
b: Two types of terminators for coaxial
cables. On the left is an inline terminator,
and on the right is a stub terminator. Both
have 50 between inner and outer
conductors. c: Typical arrangements for
suppressing reflections at the inputs of an
oscilloscope, using an inline terminator
on the left, and a coax T and stub
terminator on the right.
Concept test (answer below11 ): As in the previous concept test, a right-traveling wave
gets to the end of a coaxial cable, at which point a wire has been attached connecting
the inner conductor to the outer conductor. Is the reflected voltage pulse inverted or not
with respect to the incident current pulse?
Finally, to suppress reflections, at the end of the cable we simply connect a resistor
with resistance equal to the characteristic impedance of the cable, as illustrated in
figure 10.5.2a. Standard RG58 coaxial cable (the type used in laboratories, and the type
that has a BNC coaxial connector at the end) has C0 = 100 pF/m and L0 = 250 nH/m;
L0
plugging in these numbers gives Z = = 50 . Thus, a 50 resistor connected
C0
from the inner to the outer conductor feels (electrically) the same as an infinite length
of cable. In practice, one either connects the end of the cable to a target device that
already has an input impedance of 50 , or instead one uses a “terminator.” There are

11. Plugging Zt = 0 into equation (10.5.4), we see that Vreflected = −Vincident x=0 , so the reflected
voltage pulse is inverted, and thus is opposite in phase to the reflected current pulse. This makes
sense, since we know that for left-traveling waves the voltage is opposite in phase to the current.
two common styles, as shown in figure 10.5.2b and c, either of which can be used
to make a nonreflecting connection to a device with high input impedance such as an
oscilloscope. Either of the styles of terminators has a 50 resistor connecting the inner
and outer conductors of the coax.
We are now equipped to really understand what the term “characteristic impedance
of a transmission line” means:
The characteristic impedance of a transmission line is (1) The resistance one could
connect between the two wires at the end of the line to suppress reflections and (2)
The resistance one would measure at the left end of an infinitely-long transmission
line.
Meaning 2 may be difficult to understand at first. One way to measure resistance

is to apply a known voltage and measure the resulting current. When you first
apply the voltage to the transmission line, you launch a voltage wavefront down the
line, accompanied of course by a current wavefront, as shown in figure 10.5.3. The
equipment you’re using is supplying current to charge up the capacitance between the
two conductors of the transmission line; as time goes on, and the voltage wavefront
propagates out, more and more length must be charged up. Since the capacitance per
unit length is constant, this means that charge must be delivered at a constant rate
to the inner conductor (and removed from the outer conductor). Thus, you would
measure a constant current going into the inner conductor, and coming out of the
outer conductor. While the wavefronts are propagating out along the infinite cable, the
current you measure would be exactly the same DC current as you’d get from a 50
resistor.
If you take a 1-m length of standard coaxial cable, with nothing connected to
either end, apply a fixed voltage between the inner and outer conductors at the left
end, and measure the resulting current, you’ll get zero current, instead of the DC
current mentioned above. This is because it takes a very short time for the propagating
Figure 10.5.3 When you first apply a

voltage to the left end of a transmission line,
you generating a wavefront in the voltage
which propagates down the line,
accompanied by a wavefront in the current.
wavefronts in figure 10.5.3 to reach the end of this cable. Because it is not terminated
with a 50 resistor, a reflection is generated, which comes back to the left end of the
cable, generating a new reflection which goes back to the right end, and so on. Very
quickly, the whole thing settles down into a state with no propagating waves, but rather
a steady DC voltage difference and no current flowing.
Self-test (answer below12 ): A voltage difference of 50 V is suddenly applied to the inner

conductor of a standard coax cable, relative to the outer conductor. The cable is 100-m
long. For a very short time, the observed current going into the inner conductor (and
coming out of the outer conductor) has a well-defined constant value. It then starts
changing, and settling toward zero current. What is the well-defined constant value, and
for how long is it observed?
10.6 Reflection and transmission for electromagnetic waves in matter:

Normal incidence
Next, we consider what happens when a light wave encounters the boundary between
two linear materials, assuming the wave propagates in a direction perpendicular to
the boundary. We will be able to use our isomorphisms to find the expressions for the
reflected and transmitted pulses with no additional work, after checking to make sure
the boundary conditions are the same. We have previously made isomorphisms between
rope waves and electromagnetic waves in vacuum, and between electromagnetic waves
in vacuum and in linear materials:
Table 10.6.1. Isomorphism between rope waves, em waves in vacuum, and em waves in
linear materials
Rope waves Electromagnetic waves Electromagnetic waves in

in vacuum linear materials
Transverse velocity ẏ B/μ0 H = B/μ

y-component of force from Electric field E Electric field E
∂y
the left FL = −T
∂x
Mass per length μ Permeability of free space μ0 Permeability μ
Tension T Inverse of permittivity of free Inverse of permittivity 1/ε
space 1/ε0
So, for example, H is isomorphic to ẏ.
12. Until the wavefront propagates to the end of the cable and back, the information that the
cable is not infinitely long is unknown. As we saw in section 9.7, the speed of waves on
2 × (100 m)
standard coax cable is 2/3 c, so the constant current condition lasts for a time 2
=
3c
1μ s. The amount of current is I = V /Z = 1A.
What is the boundary condition for E? At the interface between two linear
materials, there can be surface charge, which affects the component of E perpendicular
to the interface but not the component parallel to the interface. Therefore, for an
electromagnetic wave which propagates perpendicular to the interface (so that E is
parallel to the interface), E must be continuous at the interface. This is the same
boundary condition as for the isomorphic quantity for a rope wave, FL, which must
also be continuous at the interface as discussed in section 10.4.
What is the boundary condition for H? Recall from section 9.6 that in linear
material, we can create an alternate version of Maxwell’s equations by replacing μ0
with μ, ε0 with ε, and all references to charge with references to free charge. Therefore,
the alternate version of Ampère’s law in an isotropic linear medium is
A
⇀ d
B · d ℓ = μ Ifree +μ ε E · n̂ dA.
net dt
threading
If μ and ε vary in space, then we must bring them inside the integrals:
A
1 ⇀ d
B · d ℓ = Ifree + ε E · n̂ dA. (10.6.1)
μ net dt
threading
Consider the Ampèrian loop shown in figure 10.6.1, which straddles the interface. If
we make the height
$ of the loop infinitesimal, then terms that depend on the area of the
loop, such as E · n̂ dA, and any ordinary current flowing through the loop become
negligible. Surface currents can still be important, since their current density is infinite.
However, we will assume that the materials in question are nonconducting, so the
1
surface current is zero. Then, since H = B for a linear material, equation (10.6.1)
μ
B ⇀
simplifies to H · d ℓ = 0. The contributions to this integral from the vertical arms of
the loop are negligible, since we have made the height of the loop infinitesimal. For
a plane wave propagating perpendicular to the interface, we align the top and bottom
arms of the loop along the axis of H, so that we get H1 ℓ − H2 ℓ = 0. (The direction
⇀
of d ℓ for the bottom arm of the loop is opposite to that for the top arm, leading to
the minus sign for the H2 ℓ term.) Therefore, H1 = H2 , meaning that H is continuous
across the interface. This is the same boundary condition as for the isomorphic quantity
in a rope wave, ẏ, which is continuous across the interface.
Therefore, the isomorphism is complete even for reflections and transmissions,
and we can immediately make use of the isomorphic expressions for transmission and
reflection for a rope wave. To do so, we need the expression for the impedance of
√
the transmission line. For the rope Z = T μ, so, reading off the isomorphism, the
Figure 10.6.1 An Ampèrian loop

at the interface. The top and
bottom arms have length ℓ.
impedance for electromagnetic waves in linear materials is

Z= μ ε. (10.6.2)
FL
By equation (10.3.2), for the rope wave Z = for a right-moving wave, so for waves
ẏ
in linear materials
E
Z= . (10.6.3)
H
Impedance for right-moving electromagnetic waves in linear materials
The E in an electromagnetic wave is isomorphic to FL. So, translating equation (10.4.9)

we have
" #
Z − Zi
Ereflected = Eincident t . (10.6.4)
Zi + Zt x=0
For problems in optics, it is conventional to work in terms of the index of refraction

c με
n≡ = . (10.6.5)
vp μ0 ε0
For most materials, μ ∼ = μ0 . Making use of this as we plug equation (10.6.2) into
(10.6.4), we obtain at x = 0

μ0 μ0
− √ √
εt εi εi − εt
Ereflected ∼
= Eincident = Eincident √ √ ⇒
μ0 μ0 εt + εi
+
εi εt
" #
∼ ni − nt
Ereflected = Eincident . (10.6.6)
nt + ni x=0
Concept test (answer below13 ): If ni > nt , is the reflection inverted relative to the
incident wave or not? If ni < nt , is the reflection inverted or not?
13. From equation (10.6.6), we can see that if ni > nt (e.g., a wave travelling from glass to air), then
the coefficient for the reflected amplitude is positive, so that the reflected wave is uninverted
(albeit smaller than the incident wave). On the other hand, if ni < nt (e.g., a wave travelling
from air to glass), then the coefficient is negative, so the reflection is inverted. This distinction
is important for various interference phenomena (see, for example, problem 10.16), and so it is
worth remembering. Here’s a mnemonic that may be helpful: “Air to glass, switch head and ass.
Glass to air, the phase don’t care.”
Often in optics
we are more interested in the intensity than the amplitude. From equation
ε 2
(9.6.8), S = E , so
μ

2
ni − nt
Sreflected ∼
= Sincident . (10.6.7)
nt + ni
Your turn: Use the isomorphism to show that

" #
2Zt
Etransmitted = Eincident . (10.6.8)
Zi + Zt x =0
and (assuming μ ∼
= μ0 for both materials)
" #
2ni
Etransmitted = Eincident . (10.6.9)
nt + ni x =0
In problem 10.13, you can show that if μ ∼

= μ0 then
4ni nt
St = Si 2 . (10.6.10)
ni + nt
10.7 Reflection and transmission for sound waves, and summary of

isomorphisms
Finally, we consider what happens when a sound wave encounters the boundary
between two media (e.g., air and water), assuming the wave propagates in a direction
dδ
perpendicular to the boundary. In section 9.8, we showed that the velocity wave v ≡
dt
(where δ is the longitudinal displacement) is isomorphic to B/μ0 for an electromagnetic
wave in vacuum, which is in turn isomorphic to the transverse velocity ẏ for a rope wave.
We also showed that the pressure variation P′ is isomorphic to E an electromagnetic
wave in vacuum, which is in turn isomorphic to FL for a rope wave.
What is the boundary condition for the pressure variation P′ ? Consider a volume
element of infinitesimal thickness centered on the interface. The net force on this
element must be zero (to avoid infinite acceleration). Therefore, the pressure must be
the same on either side of the element, so that P′ must be continuous at the interface.
This is the same boundary condition as for the isomorphic quantity for a rope wave,
FL, which must also be continuous at the interface as discussed in section 10.4.
What is the boundary condition for v? The two materials at the interface must
remain in contact with each other, so that the displacement δ must be continuous at the
dδ
interface. Therefore, its derivative v ≡ is continuous across the interface. This is
dt
the same boundary condition as for the isomorphic quantity in a rope wave, ẏ, which
is continuous across the interface. Therefore, the isomorphism is again complete.
We are now ready to summarize all the isomorphisms we have discussed:
Table 10.7.1. Summary of Isomorphisms for waves
Electromagnetic Waves on
waves in transmission
Rope waves linear matter lines Sound waves
Transverse H = B/μ I Velocity

velocity ẏ (B/μ0 invacuum) v ≡ ∂δ ∂ t
y-component of Electric field E V Pressure
force from the left variation P′
∂y
FL = −T
∂x
Mass per length μ Permeability μ Inductance per Equilibrium
length L0 density ρ0
Tension T Inverse Inverse of γ P0

permittivity 1/ε capacitance per
length: 1/C0
√
Phase velocity T μ 1 εμ 1 C0 L0 γ P0 ρ0
(for a gas)
√
Impedance Z μT μ ε L0 C0 ρ0 γ P0
(for a gas)

Impedance Z FL ẏ E H V I P′ v
for right-
moving waves
Amplitude Zt − Zi Zt − Zi Zt − Zi Zt − Zi
reflection FLr = FLi Er = E i Vr = Vi Pr′ = Pi′
Zi + Zt Zi + Zt Zi + Zt Zi + Zt
coefficients
Zi − Zt Zi − Zt Zi − Zt Zi − Zt
ẏr = ẏi Hr = Hi Ir = Ii vr = vi
Amplitude 2Zt 2Zt 2Zt 2Zt

transmission FLt = FLi Et = Ei Vt = Vi Pt′ = Pi′
coefficients
2Zi 2Zi 2Zi 2Zi
ẏt = ẏi Ht = Hi It = Ii vt = vi
10.8 Snell’s Law
What happens when waves of any type in a two- or three-dimensional system encounter
an interface at an angle, rather than encountering it at normal incidence? We define
the “wavefront” to be a line drawn along the crest of the wave. (For waves in three
dimensions, the wavefront is a plane.) Usually, the wavefront is perpendicular to the
direction of propagation; this direction is shown as a “ray.” Experimentally, we find
that if a wave has straight wavefronts in medium 1, the transmitted wavefronts in
medium 2 are also straight, as shown in figure 10.8.1. We found in our study of
one-dimensional waves that the frequency in medium 2 is the same as in medium
Figure 10.8.1 Straight wavefronts incident from medium 1 lead to straight wavefronts in
medium 2, with the same frequency.
1; this is a consequence of causality, since each crest in the incident wave launches a
crest in the transmitted wave.
λ v λ f λ
Therefore, v = = λf ⇒ 1 = 1 = 1 .
T v2 λ2 f λ2
v1 sin θ1
From the figure, λ1 = a sin θ1 and λ2 = a sin θ2 . Therefore, = .
v2 sin θ2
c
Your turn: Using the definition of the index of refraction, n ≡ and the above equation,
v
show that
n1 sin θ1 = n2 sin θ2 , (10.8.1)
Snell’s Law
where n1 is the index of refraction in medium 1 and n2 is the index in medium 2.
Snell’s Law describes refraction. For example, if medium 1 is air (for which
n1 ∼
= 1) and medium 2 is glass (for which n2 is in the range 1.5–1.9), so that n2 > n1 ,
then θ2 must be smaller than θ1 , meaning that the ray “refracts toward the normal,”
as shown in figure 10.8.1. On the other hand, if n2 < n1 (as would be the case for
a glass-to-air interface), then θ2 > θ1 , meaning that the ray refracts away from the
normal.
Example: A plano-convex lens. All of geometric optics can be derived from the
simple idea that there must be some curved surface which refracts parallel incoming
rays so that they meet a distance f behind the lens, where f is called the focal
length. For manufacturing convenience, most lenses are made with spherical surfaces,
continued
Figure 10.8.2 Refraction for a plano-convex lens.
even though this only approximately produces the desired convergence to a point. (The
lack of ideal performance for such lenses is called “spherical aberration.”) We consider
a plano-convex lens, as shown in figure 10.8.2, made from glass of refractive index n1 .
The lens is immersed in air, which we take to have index of refraction n2 ∼ = 1. The “optic
axis” is the horizontal line drawn through the center of the lens. Consider a ray of light
parallel to the optic axis, a distance y above it. Because it strikes the planar surface of
the lens at normal incidence, it is not refracted there, however it is refracted at the point
where it encounters the curved surface of the lens, at point A. This surface has radius of
curvature R; because the surface is spherical, a radius line drawn from A to the center
of curvature is perpendicular to the lens surface. Therefore, the angle of incidence θ1 is
the angle between this radius and the incident ray, as shown. We will consider only rays
traveling close to the optic axis; this is called the “paraxial approximation.” Therefore, all
the angles involved are small; for example, the smaller y is, the smaller θ1 and θ2 are.
At point A, Snell’s Law gives n1 sin θ1 = n2 sin θ2 . Since the angles are small, we have
sin θ ∼
= θ , so that Snell’s Law becomes
n1 θ1 ∼
= n2 θ2 = θ2 .
From the figure, the “bending angle” is given by
ϕ = θ2 − θ1 ∼
= n1 θ1 − θ1 ⇒

ϕ∼= θ1 n1 − 1 (10.8.2)
and
y
tan ϕ = ⇒
f
y
ϕ∼
= , (10.8.3)
f
where, in the paraxial approximation ϕ is small, so sin ϕ ∼
= ϕ , cos ϕ ∼
= 1, and tan ϕ ∼
= ϕ.
Combining equations (10.8.2) and (10.8.3) gives
y ∼
= θ1 n1 − 1 .
f
370
From the figure, y is approximately equal to the arclength θ1 R, so that

θ1 R ∼

= θ1 n1 − 1 ⇒
f
1 ∼ 1
= n −1 . (10.8.4)
f R 1
So, a more strongly curved lens (one with small R) has a shorter focal length f , as is
reasonable.
10.9 Total internal reflection and evanescent waves
Consider what happens when the incident medium has a higher index of refraction
than the transmitted medium. Snell’s law, equation (10.8.1), gives,
n
ni sin θi = nt sin θt ⇔ sin θt = i sin θi .
nt
Since ni > nt , the refracted angle θt is greater than the incident angle θi , as shown in
figure 10.9.1a. As θi increases, θt also increases, reaching 90◦ at a critical value of θi
determined by
n
sin 90◦ = i sin θi ⇒
nt
n
θi ≡ θc = sin−1 t . (10.9.1)
ni
For values of θi greater than θc , there is no real angle θt which satisfies Snell’s law, so
there is no transmitted ray. Therefore, all the incident energy goes into the reflected
ray; this is called total internal reflection.
If you’ve ever gone under water and looked up, you have seen total internal
reflection. If you look straight up, you can see whatever is above the water (usually
the sky). But, if you look off to the side, you’ll notice that the underside of the water
acts like a mirror, as shown in figure 10.9.2a.
Perhaps the most important application of total internal reflection is for fiber
optics. As shown in figure 10.9.2b, if you shine a laser into the end of a long piece
of glass or plastic, at an angle reasonably close to parallel to the long axis (i.e., at an
angle such that θi > θc ), then it reflects back and forth between the sides, as shown
in the figure. By encoding information into pulses of a laser beam, this can be used
to transmit information (e.g., telephone and internet signals) over long distances. In
fact, essentially all long-range information transmission is now done with either fiber
optics or satellite transmission.
The plastic used for the figure scatters the laser beam quite a bit (which allows us to
visualize the beam), resulting in considerable energy loss, as you can see. Commercial
fiber optics are made using ultra high purity glass, so that a signal can be transmitted
for very long distances. There is still the need to regenerate the signal periodically,
using a “repeater” which includes a receiver, elements that sharpen the pulses,
Figure 10.9.1 a: Angles of reflection and
refraction for a ray incident from
medium 1. For the case shown, ni > nt ,
so that the refracted ray is bent away
from the normal. If θi is large enough,
then θt = 90◦ . For even larger values of
θi , there is no real value of θt that satisfies
Snell’s Law, so there is no transmitted
propagating wave. The incident,
transmitted, and reflected rays all lie in
the x–y plane. b: View looking along the
incident ray (the viewpoint shown by the
eye symbol in part (a)). The electric field
vector for the incident wave can oscillate
along any axis perpendicular to the
direction of propagation. For example, it
can oscillate in the ±z-direction (which
of course is perpendicular to the direction
of propagation); this polarization is
labeled EA. Alternatively, the electric
field can oscillate in the x–y plane (but
perpendicular to the direction of
propagation); this polarization is labeled
EB . Of course, the polarization could be
along any axis in between those of EA
and EB .
Figure 10.9.2 a: The hand’s image is

reflected by the underside of the water, due
to total internal reflection. (Image courtesy
of and © Jacob J. Loman) b: A laser beam
bounces inside a block of the plastic PMMA,
due to total internal reflection. (Image from
Wikimedia Commons.)
372
and a transmitter. However, because of the very low absorption, the repeater stations
can be 100 km apart, roughly 100 times further than for conventional copper wire
cable. In fact, the maximum distance between repeaters for fiber optics is not limited
by absorption, but rather by effects of dispersion (see section 9.12) which tends to smear
out the pulses as they propagate. To avoid such dispersion, engineers must design the
fiber optic system to use wavelengths of light over which the dispersion relation is close
to linear. There are several “windows” of such wavelengths for properly processed
glass, and fortuitously some of these coincide with wavelengths at which other critical
components, such as optical amplifiers, can be made to work efficiently. Most fiber
optic cables are operated at wavelengths near 1,500 nm, corresponding to the infrared.
The frequency of this infrared light is 2 × 1014 Hz, so that, in principle, the length
of each laser pulse can be extremely short, allowing a vast amount of information to
be transmitted quickly over a single fiber. There are practical limitations associated
with dispersion and with the equipment connected to the fibers that limit the data
rate to much less than the theoretical maximum, although recent tests have achieved
transmission rates of up to 1.4 × 1013 bits per second, over distances of up to 160 km
between repeaters.14 By contrast, a typical copper cable has a maximum transmission
rate of 4 × 109 bits per second, with a much shorter run between repeaters. Thus, a
coaxial cable assembly containing 144 fibers, which is only 1.2 cm in diameter, can
carry the same information as three copper cable assemblies, each 7.5 cm in diameter,
each of which contains hundreds of twisted copper wire pairs. The weight and cost
of the fiber is far less. Additionally, the different twisted wire pairs within a copper
cable assembly exhibit significant “crosstalk” (the signal from one pair leaks onto
another); this is completely eliminated for fiber optics. Copper wiring is usually used
for local connections, because it is much simpler to connect the wires together, and the
equipment costs are lower.
Fiber optics are made with a central core that has a high index of refraction,
surrounded by a cladding layer with a lower index. As we have discussed earlier, by
using total internal reflection, one can eliminate the refracted beam. However, as we
will now show, the oscillating electric and magnetic fields actually do extend beyond
the interface, in the form of an “evanescent wave.” As we’ll see, this has important
implications for the design of fiber optics, as well as for the operation of certain
scientific instruments.
In sections 9.5 and 9.6, we considered a wave propagating in the x-direction, with
the electric field oscillating in the ± y-direction, and the H field (or the B field for a
wave in vacuum) oscillating in the ± z-direction. However, we could also make a wave
propagating in the x-direction for which the electric field oscillates in the ± z-direction
(and the H field in the ∓ y-direction), or in fact any direction perpendicular to the
direction of propagation. The axis along which the electric field points determines the
“polarization” of the light. For the geometry of figure 10.9.1a, there are two distinct
polarizations: (A) The electric field can oscillate parallel to the z-axis; this possible
polarization is labeled EA in part b of the figure. For this polarization, the electric field
14. NTT press release, Sept. 29, 2006.

is parallel to the interface. (B) The electric field can oscillate in the x–y plane, but in a
direction perpendicular to the propagation; this possible polarization is labeled EB . The
x–y plane contains the incident, refracted, and reflected rays, and is sometimes called
the “plane of incidence.” Of course, the electric field could oscillate along any axis
between these two extremes, but these other angles can be formed as a superposition of
EA and EB . We will consider EA and EB separately, and apply boundary conditions to
find the electric and magnetic fields in the transmitted medium for θi > θc , that is, for
angles of incidence such that there is no transmitted ray. We will find that the results
are similar for both cases.
Case A: Electric field polarized parallel to interface. We begin by considering
the case θi < θc , so that there is a transmitted ray. A wave traveling in the x-direction
can be written f (kx − ωt) , where k ≡ 2π /λ is the wavenumber. In problem 9.8, you
can show that f (k · r − ω t) represents a wave traveling in the direction of k, with
|k| = k = 2π /λ, where k is called the wavevector. Thus, the incident electric field can
be written

Ei = k̂ Ei0 cos ki · r − ωt = Re Ẽi , where Ẽi = k̂ Ei0 ei(ki ·r−ωt) , (10.9.2)
where ki is the incident wavevector

and k̂ is the unit vector in the z-direction. From
figure 10.9.1a, we have ki = ki î cos θi + ĵ sin θi . So, we can rewrite the magnitude
of the incident electric field as
Ei = Re Ẽi , where Ẽi = Ei0 ei(xki cos θi +yki sin θi −ωt) .
At the interface at t = 0, this reduces to
Ẽi (x = 0, t = 0) = Ei0 ei yki sin θi . (10.9.3)
Now, for the reflected beam. Recall from the concept test at the end of section 10.6
that when light reflects off a glass-to-air interface there is no inversion of the phase
(“glass to air, the phase don’t care”). Therefore, the reflected electric field is simply
Er = Re Ẽr , where Ẽr = k̂ Er0 ei(kr ·r−ωt) .

From figure 10.9.1a, we have kr = kr −î cos θr + ĵ sin θr . Plugging this into the
above gives
Ẽr = Er0 ei(−xkr cos θr +ykr sin θr −ωt) ⇒
Ẽr (x = 0, t = 0) = Er0 ei ykr sin θr . (10.9.4)
Finally, the transmitted electric field is
Et = ReẼt , where Ẽt = k̂ Et0 ei(kt ·r−ωt) ⇒
Ẽt = Et0 ei(xkt cos θt +ykt sin θt −ωt) ⇒ (10.9.5)
Ẽt (x = 0, t = 0) = Et0 ei ykt sin θt . (10.9.6)

Recall from section 10.6 that one of the boundary conditions is that the component
of the electric field parallel to the interface is continuous across the interface. For the
polarization we are considering in this case, the entire electric field is parallel to the
interface. Therefore, we must have
Ẽi (x = 0, t = 0) + Ẽr (x = 0, t = 0) = Ẽt (x = 0, t = 0) .
Substituting from equations (10.9.3), (10.9.4), and (10.9.6), we get
Ei0 ei yki sin θi + Er0 ei ykr sin θr = Et0 ei ykt sin θt . (10.9.7)
This equation can only hold for all values of y if the factors that appear in the
exponentials are equal.15 Thus,
iyki sin θi = iykr sin θr = iykt sin θt , (10.9.8)
and therefore
Ei0 + Er0 = Et0 . (10.9.9)
From equation (10.9.8) we get two significant results. First, since the speed is thesame
for the incident and reflected waves (as is the frequency), and since v = ω k ⇔
k = ω/ v, we have that ki = kr . So, using the first equality in equation (10.9.8),
we have
θi = θr , (10.9.10)
or “the angle of incidence equals the angle of reflection,” as you probably already
knew.
From equation (10.9.8) we also have

ki ω vi
sin θt = sin θi = sin θi .
kt ω vt
Since n ≡ c/ v, this gives

ni
sin θt = sin θi , (10.9.11)
nt
which is Snell’s Law arrived at by a somewhat different route from that in section 10.8.
Even for total internal reflection, where θi > θc , the boundary conditions still
hold, and so the above arguments still hold. Therefore, there is still a transmitted
wave, although we will see that it has a different mathematical form. (You can see
15. This may not be immediately obvious; the following may help to clarify. Define a new variable
α ≡ eiy . Then, equation (10.9.7) becomes Ei0 α ki sin θi + Er0 α kr sin θr = Et0 α kt sin θt . You know that
in any equation, the coefficients of like powers of the variable must be equal. In this equation, we
have three different-looking exponents of α . If the two exponents on the left side of the equation
(ki sin θi and kr sin θr ) were different, then the exponent on the right side (kr sin θr ) could only
equal one of them, so the equation could not hold. Therefore, all three exponents must be equal
to each other. (Also, we must have Ei0 + Er0 = Et0 .)
from equation (10.9.9) that there must be a transmitted wave, since we have defined
n
the amplitudes Ei0 , Er0 , and Et0 to be positive.) We still have sin θt = i sin θi , but for
nt
n
θi > θc we have i sin θi > 1, so that θt is an imaginary angle.
nt
eiθt − e−iθt
Your turn: Using sin θt = , show that we can have sin θt > 1 if θt is imaginary.
2
Must θt be a positive imaginary number (of the form +iA) or a negative imaginary
number (of the form –iA)?
We can write

ni2

cos θt = 1 − sin2 θt = 1− sin2 θi ⇒
nt2

ni2
cos θt = i sin2 θi − 1 ⇒
nt2
i

cos θt = n2 sin2 θi − nt2 , (10.9.12)
nt i
where for θi > θc the quantity inside the square root of equation (10.9.12) is positive.
The transmitted electric field is still given by equation (10.9.5):
Ẽt = Et0 ei(xkt cos θt +ykt sin θt −ωt) .
Substituting for cos θt using equation (10.9.12), and for sin θt using equation (10.9.11)
gives

n
ni2 sin2 θi −nt2 i ykt nit sin θi −ωt
k
−x nt
Ẽt = Et0 e t e .
ω c ω nt
Your turn: Since k = and n = , we have that kt = . Use this to show that
v v c
Ẽt = Et0 e−κ x ei(ke y −ωt ) , (10.9.13)
where

ω
κ≡ n2i sin2 θi − n2t , (10.9.14)
c
and the wavenumber for the “evanescent” wave in the transmitted medium is
ω ni
ke ≡ sin θi . (10.9.15)
c
Equation (10.9.13) describes a wave that travels in the +y-direction, parallel to

the interface. (The polarization is still along the z-axis.) As described by the e−κ x
term, the wave dies off exponentially with distance into the transmitted medium. We
will explore the implications of this “evanescent wave” after we consider the other
polarization.
Case B: electric field polarized in the plane of incidence. If the electric field
lies in the x–y plane, then H is parallel to the interface, that is, parallel to the z-axis.
Recall from section 10.6 that the component of H that is parallel to the interface is
continuous across the interface. Since this is the same condition as we had for E for the
case A polarization, the entire rest of the argument is identical, with the substitution of
H for E. Thus, we obtain
H̃t = Ht0 e−κ x ei(ke y−ωt) , (10.9.16)
with the same values for κ and ke . Again, this describes a wave propagating in the
+y-direction, and exponentially decaying in the x-direction. The connections between
E and H are more complicated in thiscircumstance than in an infinite expanse of linear
μ
matter (we don’t simply have E = H), so the waves in the transmitted medium
ε
are different for the two polarizations (in ways beyond their polarization), but they are
qualitatively very similar.
Because the evanescent wave propagates parallel to the interface, there is
ordinarily no transmission of power into the transmitted medium, so that the power
of the reflected beam equals the power of the incident beam. However, the oscillating
electromagnetic field of the evanescent wave can cause local excitation of atoms which
absorb at that frequency. For this reason, the cladding which provides the low index
of refraction around the high-index core of a fiber optic must be very pure, to avoid
absorption by impurity atoms.
Connection to current research: Total Internal Reflection Fluorescence Microscopy.

This type of absorption is used to advantage in Total Internal Reflection Fluorescence
Microscopy (TIRFM). In fluorescence, an electron in an atom is excited to a high energy
state using ultraviolet light. As the electron returns to its original state through a series of
steps, it emits visible light of a characteristic wavelength. In TIRFM, a sample of interest
(usually biological) is applied to the surface of a quartz substrate. Ultraviolet light is
transmitted within the substrate, contained by total internal reflection. The associated
evanescent wave extends at most 100 nm into the biological sample, so only atoms
within this thickness can be excited by the wave and fluoresce. Thus, the image of the
fluorescent light comes from the very thin layer of sample in contact with the quartz;
this localization can be very important for various experiments. For example, Gunnarson
and co-authors used a sequence of tethers to attach individual DNA molecules to a
quartz substrate, as shown schematically in figure 10.9.3a. The arrows at the bottom
of the figure schematically represent the rays of UV light bouncing off the top surface
of the substrate because of total internal reflection. The bound DNA molecules then
bind to complementary DNA strands which are attached to fluorescent self-assembled
nanospheres in the solution above. Only the bound nanospheres fluoresce, due to the
small extent of the evanescent wave. Thus, each bright spot in the image (figure 10.9.3b)
represents the binding of a nanosphere to a single DNA molecule. The simplest way
to use this technique is to test for the presence of the complementary DNA strand
in a target solution. However, it can also be used in other ways. For example, by
measuring how much time each nanosphere stays bound, the scientists can detect
whether there are any mismatches in the base sequences of the two complementary
DNA strands.
Figure 10.9.3 a: Schematic diagram showing how DNA molecules that are 30 bases long
(“30-mers”) are attached via a series of tethering molecules to a quartz substrate.
Self-assembled fluorescent nanospheres in the solution then attach to the DNA. b: Sequence of
TIRFM images. Each spot (highlighted by a grey circle) is a single nanosphere attached to a
single DNA molecule. Reprinted with permission from A. Gunnarson et al., Nano Letters 8,
183–8 (2008). Copyright 2008 American Chemical Society.
terms:
Boundary conditions (10.1)
Characteristic impedance for mechanical systems (10.3)
FL (10.3)
Terminator (10.5)
Characteristic impedance of a transmission line (10.5)
Index of refraction (10.6)
Snell’s Law (10.8)
Refraction (10.8)
Total internal reflection (10.9)
Evanescent wave (10.9)

ẏ, B, H , I , & v (10.5–10.7)
FL, E , V , & P′ (10.5–10.7)
μ, μ0 , μ , L0 , & ρ0 (10.5–10.7)
T , 1/ε0 , 1/ε , 1/C0 , & γ P0 (10.5–10.7)

The phase of an em wave reflected off a glass-to-air interface & the phase of an em
wave reflected off an air-to-glass interface (10.6)

For a right-traveling wave, the motion of the left end of the rope determines the wave
shape. (10.2)

Calculate the characteristic impedance of a mechanical system (10.3)
Find the shape of the transmitted and reflected pulses on ropes given the shape of the
incident pulse and the speeds on each side of the interface (10.1–10.2)
Find the shape of the transmitted and reflected pulses for rope waves, transmission line
waves, em waves in vacuum or linear media, and sound waves, given the shape of
the incident pulse and the information needed to calculate the impedance (10.4–10.7)
Explain how to suppress reflections with a compact terminator for rope waves (10.3)
or transmission line waves (10.5)
Calculate the angle of refraction (10.8)
Calculate the critical angle for total internal reflection (10.8)
Calculate the amplitude of an evanescent wave in the transmitted medium (10.8)
Problems

10.1 “The Wave” Spectators in large, round stadiums often perform a maneuver
called “the wave,” in which a “pulse” of people standing and waving
their arms propagates around the stadium while the rest of the people
remain seated. In this problem, you’ll model this phenomenon with a “peer
pressure” model, in which the motion of a particular spectator is completely
determined by what the people to his left and right are doing. (In other
words, there is no free will in this model!) Here’s the mathematical form
for our model:
β β
ÿp = −s yp − yp−1 − s yp − yp+1 ,
where β and s are constants, yp is the position of the pth spectator, yp−1
is the position of the spectator to his left, and yp+1 is the position of the
spectator to his right. Assume that negative values of y are allowed (perhaps
by crouching down in the seat), as well as positive values. The normal sitting
position corresponds to y = 0. We will focus only on spectators in the front
row, and assume that this row extends without a break around the entire
circular stadium. Note that, for pulses shorter than the circumference of the
stadium, this is like an infinitely long chain of people. We’ll assume that
Figure 10.P.1 Waveform for “the wave.”
the big changes in y occur on length scales much longer than the spacing
between spectators, but still much shorter than the stadium circumference.
(a) A pulse with a triangular shape (as shown in figure 10.P.1)
propagates without changing shape, and an identical pulse travels
in the opposite direction. When the peaks of the two pulses reach
the same point, the instantaneous value of y is twice the peak
height of either pulse alone. The pulses continue on, without having
affected each other by their “collision.” What must be the value of
β ? Explain.
(b) Now there is a single triangular pulse, as shown in the figure. Queen
Elizabeth II is in the front row. It is an experimental fact that she
will not participate in the wave. Mathematically, we could say that
y = 0 for the queen, no matter what the people near her are doing.
Describe what happens when the pulse gets to the queen’s position.
(c) The wave has now died away, and everyone in the stadium is
sitting (i.e., y = 0 for everyone). Now, the queen decides to
stand up, and continues standing. Describe qualitatively what
happens.
10.2 Reflection and transmission of two pulses. Two ropes under tension are
joined together, with v1 = 1/2v2 , where v1 is the velocity of waves in the left
rope and v2 is the velocity in the right rope. As shown figure 10.P.2, pulses
approach the interface from both sides. The pulse on the right has half the
height of that on the left, but the same width.
(a) Sketch the shape of the rope for the instant when the halfway point
of the left pulse reaches the interface.
(b) Sketch the shape of the rope well after both pulses have reached
the interface.
10.3 For sinusoidal waves on a rope that encounter an interface with a different
kind of rope, show that energy is conserved, that is, that the power of the
Figure 10.P.2 Pulses incident from left and right on an interface.

Figure 10.P.3 a: Photo of an Australian bull whip. b: Model for a whip. c: A sinusoidal wave is
incident from the left. Image credit for part a: © 1998 by C. Goodwin (license at
artlibre.org/licence/lal/en/).
incident wave equals the sum of the powers for the reflected and transmitted
waves.
10.4 Modeling a whip. Instead of an abrupt interface from one material to
another, one can create a “gradual” interface, in which the speed of wave
propagation changes smoothly from v1 to v2 . If such a gradual interface is
made just right, reflections can be almost eliminated. This has important
practical consequences for coupling high-frequency circuits to antennas,
and also for designing a whip. As shown in figure 10.P.3, a real whip has a
gradually tapering diameter. We will use a simplified model, as shown in
part (b). At the left, we assume a uniform diameter, with wave speed v1 .
In the middle, there is a tapering region. At the right, there is a uniform,
smaller diameter, with wave speed v2 . The tension T is the same throughout.
Assume the tapering region is made “perfectly,” so that there is no reflection
for waves coming in from the left.
A sinusoidal wave train propagates to the right, along the thick portion,
as shown in part (c). To make the whip “crack,” the transverse velocity
ẏ must exceed the speed of sound in air (vs = 330m/s), once the wave
train propagates into the thinner part of the rope on the right. What
is the minimum value of C1 required for this to happen, given the other
information in the figure? You should give a numerical answer. Hint:
Because of the gradual interface, you cannot use the formula for an
abrupt interface to find the amplitude of the transmitted wave. Instead,
use conservation of energy to find this amplitude.
10.5 (a) Write an expression for the mechanical impedance of a damped, driven
harmonic oscillator as a function of frequency in terms of k, b, m, F0 , and
ωd . To keep things neat, you may write your expression in terms of other
quantities, so long as you define these other quantities in terms of k, b, m,
F0 , and ωd . Assume the oscillator is in steady state. (b) You should find
that your expression for Z is complex. What does that imply?
10.6 A weasel holds the left end of a long rope which is under tension and initially
at rest. The weasel then starts to move the end up and down sinusoidally.
Show that power delivered by the weasel to the rope, averaged over a cycle,
is equal to the power in the wave.
10.7 In section 10.5, I claimed that if you had an infinitely long coaxial cable
and applied a constant voltage V to the left end of it, you would need to
supply a constant current I = V /Z, where Z is the characteristic impedance
of the coax, in order to charge up the capacitance between the inner and
outer conductors, as the wavefront of V propagates to the right. Although
we have already made arguments that show this is correct in section 10.5,
sometimes it is helpful to see things from more than one point of view.
Starting from Q = CV , and taking into account the time dependence of the
effective capacitance (as the wavefront propagates to the right), show that,
indeed, I = V /Z.
10.8 The characteristic impedance of standard coaxial cable is 50 . (a) Explain
what this means. (b) Why do we not measure a resistance of 50 when
we connect a normal ohmmeter between the inner and outer conductors of
a coaxial cable found in any of our labs?
10.9 Standard coaxial cable has C0 = 100 pF/m and L0 = 250 nH/m. The right
end of a length of standard coax is connected to a length of a different kind
of coax which also has C0 = 100 pF/m but has L0 = 500 nH/m. A right
moving current pulse is sent down this cable. Describe quantitatively what
happens when the pulse encounters the junction. Include comments about
the relative height and width of the incident current pulse, reflected current
pulse, and transmitted current pulse. Also include comments about the
relative height and width of the incident voltage pulse, reflected voltage
pulse, and transmitted voltage pulse.
10.10 Disappearing beads. If colorless, transparent plastic beads are put into a
liquid that has the same refractive index as the beads, they become invisible.
Explain.
10.11 RFID tags. A radio frequency identification (RFID) tag is a small device
attached to an item that needs to be tracked. To locate or identify the item,
one uses a “reader” unit to broadcast a radio frequency (RF) electromagnetic
wave (which we’ll call the interrogation signal) into the area containing the
item with the attached RFID tag. The tag then sends back an RF response
signal indicating both its presence and a tracking number. The lowest cost
versions of these tags are powered by the energy in the interrogation signal.
There are several different methods for generating the response wave. We
can model one of these technologies by an antenna which can be connected
to ground by an electrically controlled switch. When the switch is closed,
current can flow from the antenna to ground. When the switch is open, no
current flows. When the circuit within the RFID tag detects the interrogation
signal, it opens and closes the switch in a specific pattern determined by the
tracking number associated with that particular tag. By controlling the flow
of current in the antenna, the circuit controls the strength of the reflected
electromagnetic wave (as you’ll explore in this problem); this reflected
wave is used as the response signal. (a) Explain qualitatively why the
reflected signal is stronger when the switch is closed than when it is open.
(b) Assume the electric field of the interrogation signal at the position of
the RFID tag is Ei cos ωt, where the incident amplitude Ei is positive. Is
the reflected wave given by Er cos ωt or instead by −Er cos ωt, where Er is
positive. Explain your reasoning. (Note: you are not expected to determine
the magnitude of Er .)
10.12 Impedance matching. Suppose you want to maximize the transmission of
a wave from medium 1 to medium 2, but they have different impedances
Z1 and Z2 . You might think that inserting a slab of a third medium, call it
medium A, in between would lower the transmission, because now there are
two interfaces (one between 1 and A, and another between A and 2), with
reflections occurring at both interfaces. However, if the impedance of A,
ZA , is chosen correctly, the net transmission from 1 to 2 can be significantly
enhanced. (a)Assuming that all the impedances are real (and positive), what
is the best choice for ZA to maximize net transmission? (Note: you should
ignore second-order reflection effects. For example, you should not worry
about the fact that the wave which is reflected to the left off of the A-2
interface will be partly reflected back to the right off of the 1-A interface.
Such effects would only increase the overall enhancement achieved by
adding the slab of A.) (b) For the case Z2 = 3Z1 , by what factor can the
overall amplitude transmission be enhanced?
10.13 For electromagnetic plane waves that propagate in a direction perpendicular
to the interface between two linear materials, both of which have μ ∼ = μ0 ,
show that
4ni nt
St ∼
= Si 2 .
ni + nt
10.14 Show that, for electromagnetic waves in linear materials with μ ∼
= μ0
" #
n − ni
= Hincident t
Hreflected ∼ and
ni + nt x=0
" #
2nt
Htransmitted ∼
= Hincident .
ni + nt x=0
10.15 A plane wave of light with intensity 1.25 kW/m2 travels through the air,
and strikes a thick pane of glass at normal incidence. (The intensity is
averaged over one cycle of the wave; intensity is always quoted in this
way.) The glass has μ ∼ = μ0 , and dielectric constant κ = 5.6. Assume that
the electric and magnetic fields are mutually perpendicular in the glass and
the air. (a) What are the rms amplitudes of the electric and magnetic fields
for the incident wave? (b) What are the rms amplitudes of the electric and
magnetic fields for the reflected wave? (c) What are the rms amplitudes of
the electric and magnetic fields for the transmitted wave?
10.16 Thin film interference. Extraordinarily beautiful patterns can be made by
reflection of light off thin films. You have probably seen the colorful patterns
of light reflecting from a thin film of oil or gasoline as it spreads on a water
surface. By suspending a soap film vertically, one can observe spectacular
patterns as the soapy fluid flows toward the bottom. (For examples, simply
do an internet image search for “soap film.”) In this problem, you will
investigate this phenomenon. The index of refraction for soapy liquid is
approximately 1.4. (Take this as exact for this problem.) Consider a film
in the y–z plane of soapy liquid with thickness d. The film was created
by dipping a hoop into a bucket of soapy liquid, and then holding the
hoop vertically. Over time, the soapy liquid flows from the top to the
bottom, so that the thickness of the film at the top gets smaller. We aim
a beam of light at this top part. (a) Take the incident light to be traveling
in the +x-direction, so that it is at normal incidence to left side of the film
(hereafter referred to as the “front side”). Show that only 2.78% of the
incident intensity is reflected. (b) The result of part (a) means that the beam
transmitted into the soapy liquid is essentially as strong as the incident
beam. Explain briefly why the intensity reflected off the back surface of
the film (i.e., the right surface, where there is a liquid-to-air interface) is
2.78% of the intensity of the beam that was transmitted through the front
surface.
Almost all the light that is reflected off the back surface will get through
the front surface. (Again, only 2.78% gets re-reflected off the front surface
back to the right.) Therefore, there are two beams reflected to the −x-
direction, one from the front surface and one from the back, and these
beams are of almost equal intensity. Experimentally, one observes that
as the soap film gets thinner, eventually the top part (which is thinnest)
“disappears,” meaning that it reflects essentially no light. At this point, the
thickness of the film at the top part is less than 10 nm. (c) Explain why this
reflects essentially no light. (Bear in mind that the wavelength of visible
light in air is 380–750 nm.) (d) At a point lower down in the film, where
it is thicker, the film appears blue (corresponding to a wavelength in air of
480 nm), meaning that the light reflected back to the left is primarily blue.
What is the approximate thickness of the soap film at this point?
10.17 Anti-reflection coatings. (You should do problems 10.12 and 10.16 before
this.) In problem 10.12, you showed how to maximize transmission of
a wave from one medium to another by inserting a slab of appropriate
impedance between them. This, of course, also minimizes reflection. This
method for minimizing reflection works for all wavelengths of incident
waves, and for all types of waves. In this problem, we consider light
waves in particular. Frequently, optical components such as lenses have
“anti-reflection” coatings applied, to minimize the reflection. (a) For a lens
made from glass with index of refraction 1.52, which is to be used in
air, what index of refraction should the coating have? (Assume the light
impinges on the lens at normal incidence.) (b) The impedance matching
condition tells you the index of refraction needed for the coating, but
says nothing about the optimum thickness. For the soap film treated in
problem 10.16, the reflection was minimized at all wavelengths for a film
thickness less than 10 nm. Why not use this same strategy, and just make
the anti-reflection coating less than 10-nm thick? (c) In fact, one must
know the wavelength of the light being used to choose the best thickness
for the coating. If the lens is to be used for light from an Argon ion laser
(wavelength of 488 nm in air), what is the best choice for the thickness of
the anti-reflection coating?
10.18 Prisms. Figure 10.P.4 shows a schematic picture of white light entering a
prism (from the left), and being split into its component colors. Based on
this, make a qualitative sketch of the dispersion relation for light in glass.
Your sketch should show whether the curve slopes up or down, and whether
it is concave up, concave down, straight, or some other shape. Hint: The
wavelength of red light is longer than that of blue light.
10.19 In a particular setup for Total Internal Reflection Fluorescence Microscopy
(TIRFM), the ultraviolet light is provided by a laser, so that the angle of
the light relative to the surface of the quartz is well-defined, but can be
varied by turning a knob on the apparatus. Explain how this would allow
the experimenter to control the thickness of the sample that is being probed.
10.20 Dispersion in a fiber optic cable. A fiber optic cable is used to transmit
a series of laser pulses that represent computer data. The range of
wavenumbers used to Fourier synthesize the pulses spans from k1 to k2 ,
with an average value of kav . Over this range, the dispersion relation for the
glass used for the fiber optic can be approximated by ω = α k + αβ k 2 /kav ,
where β is a small dimensionless number. For the pulses to remain readable,
the wave at k1 must stay in almost exactly the same alignment relative to
the wave at k2 as the pulse propagates; the maximum allowable relative
shift of these two waves is a distance of 0.01/kav . What is the approximate
maximum distance between repeater stations, assuming that this dispersion
is the limiting factor?
10.21 One-way viewing? You have probably seen “one-way mirrors” in televi-
sion shows. Typically, a criminal suspect is being interrogated in one room,
while several police officers watch through a one-way mirror from another
Figure 10.P.4 A prism is used to separate white light into beams of different wavelengths.
room. To the criminal, the mirror looks like a mirror, so that he only sees
his own reflection, but for the policemen it works like a window, so that
they can see the criminal. In fact, this one-way effect is completely due to
the difference in lighting between the two rooms. The “one-way mirror”
is coated with a thinner layer of reflective metal than a normal mirror, so
that perhaps 80% of the light is reflected and 20% is transmitted, instead
of having 100% transmission. The fraction transmitted and reflected is
the same in both directions. However, the room with the criminal is kept
brightly lit, whereas the other room is kept fairly dark. Thus, almost all of
the light the criminal sees comes from his own room, whereas most of the
light that the policemen see comes from the criminal’s room.
You can easily demonstrate this for yourself. Find a CD-R (i.e., a CD
that you might write data onto using your computer) that is silvery on both
sides. (Some CD-R’s are painted or have a label on one side; these won’t
work for our purpose.) If you hold the CD-R up to a bright light, you should
be able to see the light dimly through the CD-R, because the metal coating
is not very thick. Now stand in front of a mirror and hold the CD-R tight
against your right eye. Looking at yourself in the mirror with your left eye,
you can’t see your right eye. But, looking with your right eye, you can still
see the reflection of yourself in the mirror. (It’s a bit hazy, because of the
grooves in the CD.)
Having recently learned about total internal reflection, an inventor
proposes a scheme for a true one-way viewing system that would work
even if both rooms are equally illuminated. The inventor wishes to patent
his idea. As a patent office clerk, you are called on to decide whether
the idea has merit. In the drawing that accompanies the patent application
(figure 10.P.5), the two rooms are separated by a large piece of glass with
a triangular cross section. The application reads, “Light from the police
officer on the left undergoes total internal reflection at the glass-to-air
interface, as shown by the black ray, and so does not reach the criminal.
However, since the index of refraction of air is lower than that of glass, there
Figure 10.P.5 A proposed scheme for one-way viewing. According to the scheme, light from
the prisoner reaches the policeman, but light from the policeman undergoes total internal
reflection, and so does not reach the prisoner. What is wrong with this reasoning?
is no such total internal reflection for the light coming from the criminal (as
shown by the gray ray), so the police officer can see him.” What is wrong
with this argument?
10.22 Your head is 1-m below the surface of a swimming pool. Looking straight
up, you can see a circle of the sky, but beyond this circle (i.e., at larger
angles relative to straight up) the bottom surface of the water looks silvery.
The index of refraction for water is 1.33. What is the radius of the circle on
the surface of the water outside of which the surface is silvery?
A Group Velocity for an Arbitrary
Envelope Function

dω
In this appendix, we wish to show that the group velocity vg = is the velocity
dk kc
of the envelope of a wavepacket that is composed of Fourier components with
wavenumbers centered on kc . First, we will show that we can easily construct such
a wave packet by multiplying a sinusoidal oscillation with wavenumber kc by an
envelope function.
We consider an envelope function f (x) that is composed of only low-wavenumber
(long wavelength) Fourier components, that is, the Fourier components are only
nonzero for |k | < km , where km is the maximum wavenumber needed to Fourier
synthesize f (x). We can write f (x) as a sum of its Fourier components:
∞
1
f (x) = √ F (k) eikx dk , (A1)
2π
−∞
where we write the Fourier transform as F(k) rather than Y (k), and F(k) is nonzero
only from −km to km , that is, only for |k | < km . We place no other restrictions on F(k),
so that our arguments are valid for a very general envelope function f (x), restricted
only by the range of frequencies of the sinusoids summed to create it. We use f (x) as
an envelope function for a carrier wave:
y (x) = f (x) cos kc x = Re [z (x)] , (A2)
where
z (x) = f (x) eikc x .
An example of the function y (x) is shown in figure A1. Using (A1), we can write
⎡ ⎤
∞
1
z (x) = ⎣ √ F (k) eikx dk ⎦ eikc x .
2π
−∞
This presentation is based in part on that in The Physics of Waves, by Howard Georgi,
Prentice-Hall, Englewood Cliffs, NJ, 1993, pp. 235–6.
388
Appendix A ■ Group Velocity for an Arbitrary Envelope Function 389
Figure A1 Multiplying an envelope function by a

high-wavenumber “carrier wave” creates the
modulated function shown at the bottom.
Because the integral is over k, we can take the factor eikc x inside the integral, giving
∞
1
z (x) = √ F (k) ei (k +kc )x dk .
2π
−∞
We define k ′ = k + kc ⇔ k = k ′ − kc , so that
∞
1 ′
z (x) = √ F k ′ − kc eik x dk ′ .
2π
−∞
Since k ′ is merely a variable of integration, we can drop the prime, so that

∞
1
z (x) = √ F k − kc eik x dk . (A3)
2π
−∞
Note that this has the same form as the Fourier expansion of z (x):
∞
1
z (x) = √ Z (k) eik x dk .
2π
−∞

Thus, the Fourier transform of z (x) is Z (k) = F k − kc .
The only
contributions to the integral in (A3) are from wavenumbers for which
F k − kc is nonzero.Recallthat F(k) is nonzero only for |k | < km , so that F k − kc
is only non-zero for k − kc < km , as shown in figure A2a and b. Thus, by using
an envelope function to modulate a carrier wave of wavenumber kc , we have indeed
constructed a wavepacket that is composed of Fourier components with wavenumbers
centered on kc .
Now, we want to make this function move as a traveling wave. To make a simple
function such as Aeik x move to the right as a traveling wave, we replace the eikx by
the traveling wave ei(kx−ω t) , giving Aei(kx−ω t) . We apply this same operation to each
of the factors eikx (A3) giving
∞
1
z (x , t) = √ F k − kc ei(kx−ωt) dk , (A4)
2π
−∞
where ω is a function of k, as described by the dispersion relation for the particular

system.
Again, the only contributions
to
the integral in (A4) are from the wavenumbers
k − k < k for which F k − k is nonzero. We assume that, over this range of
c m c
Figure A2 a: Magnitude of F (k), the Fourier

transform of the f (x) shown in fig A1. F(k) is
nonzero only for the range −km to km . b:
Multiplying f (x) by a sinusoidal carrier wave with
wavenumber kc shifts the Fourier spectrum so that
it is now centered on kc . c: If the dispersion relation
is linear over the range of nonzero Z(k), then we
can show that the envelope of the wave moves at vg
without changing shape.
Appendix A ■ Group Velocity for an Arbitrary Envelope Function 391
wavenumbers we can write the dispersion relation as

dω
ω = ωc + k − kc , (A5)
dk kc

where ωc = ω kc is the angular frequency corresponding to the wavenumber kc of
the carrier wave. In words, our assumption (A5) says that ω depends linearly on k,
though it need not be directly proportional. An example of such a dispersion relation
is shown in figure A2c. (In fact, if km is small enough, then any dispersion relation
will satisfy this restriction,
since any function is linear over a small enough range.)
dω
Defining vg = , we can rewrite equation (A5) as
dk kc

ω = ωc + k − kc vg . (A6)
Substituting this into equation (A4) gives us

∞
1
z (x , t) = √ F k − kc ei kx e−iωc t e−i kvg t eikc vg t dk .
2π
−∞
Now, we define k ′′ ≡ k − kc ⇔ k = k ′′ + kc , so that
∞
1
F k ′′ ei (k +kc )x e−iωc t e−i (k +kc )vg t eikc vg t dk .
′′ ′′
z (x , t) = √
2π
−∞
∞
1
F k ′′ ei (kc x−ωc t) eik (x−vg t ) dk ′′ .
′′
=√
2π
−∞
Again, k ′′ is just a variable of integration, so we can rewrite it as k:

∞
1
z (x , t) = √ F (k) ei (kc x−ωc t) eik (x−vg t ) dk ⇒
2π
−∞
∞
1
z (x , t) = e i (kc x −ωc t)
√ F (k) eik (x−vg t ) dk . (A7)
2π
−∞
1 $∞
Since equation (A1): f (x) = √ F (k) eikx dk, we have that
2π −∞
∞
1

f x − vg t = √ F (k) eik (x−vg t ) dk ,
2π
−∞
so that equation (A7) becomes

z (x , t) = ei (kc x−ωc t) f x − vg t .
Figure A3 The traveling wave y (x , t).

Note that f x − vg t is just the shape f (x) moving to the right at speed vg . The actual
wave is found by taking the real part:

y (x , t) = Re [z (x , t)] = cos kc x − ωc t f x − vg t
! !
carrier wave envelope
travels at travels at
ωc
vp = vg
kc
This is what we set out to show, and is illustrated in figure A3.
Index
AC circuits 18 beaded string, eigenvector 227

adiabatic compression 310 beaded string, normal mode analysis 226
adiabatic index 311 beats 137
adjoint 154 Boltzmann’s Constant 310
adjustable constants 7 bonding orbital 164
AFM 93, 130 bound charge 298
aliasing 275 bound current 298
AM radio 32, 106, 121, 329, 341 boundary condition 220–221
Ampère’s Law 287 boundary conditions 345
amplitude 5 bra 155
amplitude, resonance 89–92 bra-ket notation 155
analogy: normal mode analysis & vector breathing mode 142
analysis 160 bulk modulus 311
analogy: particle in a box & string 243 butterfly effect 123
analogy: quantum bound states & standing
waves on string 280
analogy: rope waves & quantum waves 344 c 290
analysis, general, with orthonormal basis 250 cantilever 52–56
Angstrom 59 cantilever, effective mass of 56
angular frequency 6 cantilever, spring constant of 55
Annual Percentage Rate 78 capacitance per unit length 303
antinode 314 capacitor 10
anti-reflection coating 384 capacitor, impedance of 19
APR 78 capillary waves 341
arctangent 8 carbon nanotube 57
atomic energy levels 164 carrier wave 121, 328, 330, 331, 388–390
Atomic Force Microscopy (AFM) 93, 130 Cartesian representation 15
attractor 122, 125 Casimir effect 28
average value of squared sinusoid 97 causality 368
Avogadro’s number 310 Cavendish experiment 52
Cavendish, Henry 51
chaos 123
bank interest 78 characteristic equation 186–187
basin of attraction 123 characteristic impedance 360, 363, 382
basis 156 characteristic length 75
basis functions 248 charge 10
beaded string 216 circular polarization 336
beaded string, driven 238 clarinet 316
393
394 Index
CO2 molecule 214 differential equation 4

coaxial cable 301 dipole 297
cocoa 338 Dirac delta function 272
commuting operations 17, 181 Dirac notation 155
complete basis 159 Discrete Cosine Transform 267
complex conjugate 15 Discrete Fourier Transform (see DFT) 263
complex exponential 13 dispersion relation 221, 323
complex plane 14 displacement in a sound wave 305
complex version of current 19 diving board 52–56
complex version of voltage 19 DNA 377
Continuity Equation 307 dog pulling a spring 43
continuous functions, inner dot notation 5
product of 232 double pendulum 213
continuous string 230 drag coefficient 76
contrast for MRI 111 drag force 64
cosine, Taylor series for 13 driven beaded string 238
coupled DEQs 141 driven coupled oscillators 166
coupled oscillators, driven 166 driven oscillator in complex plane 89
coupled pendula 139 driven oscillator, energy of 95–99
critically damped behavior 73 driven oscillator, full solution for 103
crystals, waves in 226 driving force, sinusoidal 86
current 10 dynamical variables 124
current, complex version of 19
earthquakes 48
damped electrical oscillator 68 eddy currents 77
damping 64 effective mass of a cantilever 56
damping constant 64 effective spring constant 40
dB 322 eigenfunctions for string 233
DCT 267 eigenstates 152
decibels 322 eigenvalue 186
degenerate modes 206 eigenvalue equation 186
del 337 eigenvalues for two coupled oscillators 188
delocalized particle 24 eigenvector 157, 186
delta function, Dirac 272 eigenvector for beaded string 227
delta function, Kronecker 197 elastic limit 45
denegeracy 206 electrical oscillator 10
DEQ 4 electrical oscillator, damped 68
DEQ, linear 7 electrical resonance 104
DEQs, coupled 141 electromagnetic waves 289, 290, 292
derivative, partial 281 electron volt 173
determinant 181–182 electron, group velocity of 331
DFT 263 em waves 289, 290, 292
DFT, frequency resolution 259 energy density of B 317
DFT, maximum frequency for 260 energy density of E 317
DFT, order of frequencies 263–264 energy density of em radiation 318
diamagnetic 298 energy of a superposition state 151
dielectric 297 energy of driven oscillator 95–99
dielectric constant 297 energy, exponential decay of 69
difference frequency 121 energy, potential, of an oscillator 17
Index 395
energy, total, of an oscillator 17 functions, inner product of 232

energy-time uncertainty principle 28 fundamental 314
envelope function 66 FWHM 98–99, 256
equation of motion 5
equilibrium 3
equilibrium length 8, 44 Galileo 40
equipartition theorem 80 gas pressure sensor 71–72
Euler, Leonhard 14 Gauss’s Law 287
Euler’s equation 13 Gaussian function 255
Euler’s equation for fluids 309 general solution 7, 103
eV 173 gravity waves 341
evanescent wave 373, 376, 377 ground plane 301
evaporation, thermal 49 ground, electrical 18
expectation value 152, 277 group velocity 327, 329, 330, 388
exponential decay of energy 69 group velocity derivation 388
exponential, complex 13 group velocity of electron 331
guitar 239–240
gyromagnetic ratio 107–108
Faraday’s Law 10, 78, 287 gyroscope 108
Fast Fourier Transform 265
FFT 265
fiber optics 324, 328, 329, 371, 373 Hanning window 259
filter, low pass 21, 37 harmonic approximation 3
first harmonic 314 harmonic generation 114, 119
first-order approximation 12 harmonics on a guitar 240
flute 102, 314 Heisenberg uncertainty principle 29, 277
flux, magnetic 10, 78 helium, supersolid 61
FM radio 32, 121 helix 125
FMAFM 130 Hermitian conjugate 154
force from the left 294 Hermitian matrix 199
Fourier analysis 252–254, 269 Hermitian transpose 154
Fourier analysis, complex heterodyning 121
exponential 254, 269 H-field 298
Fourier analysis, wavenumber 248 hi 327
Fourier expansion 252–254, 269 highest possible frequency 223
Fourier transform 255, 269 highest possible wavenumber 225
Fourier, Joseph 246 high-pass filter 36
fractal 123, 128 Hilbert space 153
free charge 297 Hilbert space, rescaling for unequal
free electron 24 masses 208
free electron dispersion relation 325 Hilbert, David 159
frequency 6 homogeneous differential equation 7, 102
frequency modulation AFM 130 Hooke, Robert 4
frequency resolution 259 Hooke’s Law 4
frequency, highest possible 223 hot cocoa 338
frequency-time uncertainty relation 27 hydrostatic pressure 311
fret 240
full solution for driven oscillator 103
Full Width at Half Maximum 98–99, 256 ideal gas law 310
functions for rope waves 344 identity matrix 198
396 Index
IDFT 263 laminar flow 64

impedance 19 Laplacian 337
impedance matching 383 Larmor frequency 110
impedance of capacitor 19 laser tweezers 82
impedance of inductor 23 LC oscillator 10
impedance of linear materials 366 Lennard-Jones potential 59
impedance, characteristic 360, 363, 382 lens 370
impedance, mechanical 354–355 lensmakers equation 371
index of refraction 366 light 291
inductance per unit length 303 light, speed of 290
inductor 10 linear DEQ 7
inductor, impedance of 23 linear materials 297
initial position 7 linear materials, impedance of 366
initial velocity 7 linear materials, Maxwell’s equations in 299
inner product 154 local oscillator 121
inner product of functions 232 longitudinal oscillations 227
instrument, reed 316 low-pass filter 21, 37
instruments, musical 314 lumped circuit element model 302
intensity 318
intensity of em waves 300
interest, bank 78
magnetic flux 10, 78
Inverse Discrete Fourier Transform 263
magnetic moment 107
inverse matrix 198, 211
magnetic pendulum 122
ionosphere 31, 341
Magnetic Resonance Imaging 85, 107
isomorphism 11
magnetic susceptibility 298
isomorphism: em waves in linear
magnetization 110
materials & in vacuum 300
magnitude of a complex number 15
isomorphism: em waves in vacuum & rope
waves 296, 336 mass matrix 201
isomorphism: longitudinal & transverse material parameters table 46
waves 228 matrix multiplication 179–180
isomorphism: mechanical & electrical maximum frequency for DFT 260
oscillators 105 maximum wiggliness 225
isomorphism: sound & em waves 313, 340 Maxwell’s equations 287
isomorphism: transmission line waves & Maxwell’s equations in linear materials 299
em waves in vacuum 304, 339 mechanical impedance 354–355
isomorphisms, key equations for 293 MEMS 42, 62
isomorphisms, summary of 368 microphone 327
isothermal compression 312 mixing 120
mnemonic for phase upon reflection 366
mode index 195
jpeg image compression 266 molecular energy levels 164
molecular orbitals 164
moment of inertia 41
ket 155 momentum wave 293
key equations for isomorphisms 293 momentum, quantum mechanical 28
keyboard 170 momentum, transverse 293
Kirchoff’s Loop Rule 11 MRI 85, 107
Kronecker delta function 197 musical instruments 314
k-space 234–235 musical scales 170
Index 397
Nano ElectroMechanical Systems 56 Pascal 45

nanotube, carbon 57 p-brane 216
negative frequencies 257 peak frequency 91
NEMS 56 pendulum 40
Newton, Isaac 4 pendulum force 42
NMR 107 pendulum mode 142
node 220 pendulum, simple 41
nonlinear dispersion 324 period 6
non-linear oscillators 114 period doubling 126
non-normalized basis functions 250 permeability 298
normal mode analysis 149, 195 permittivity 297
normal mode analysis for beaded personality archetypes 151
string 226 phase 6
normal mode coordinates 143, 144, 201 phase factor for driven oscillator 88, 89
normal mode expansion 161 phase space 124
normal modes 142, 149 phase upon reflection, mnemonic
normalized vector 157 for 366
normalizing eigenvectors 194 phase velocity 284, 323
Nuclear Magnetic Resonance 107 phonon 325
photon 113
piezoelectric 49
Ohm’s Law 19 Planck’s constant 28
operator 180 plane wave 288, 292
optic axis 370 plano-convex lens 370
organ pipes 316 plasma 31
orthogonal 158 plasma frequency 32
orthogonality of eigenvectors 205–207 Poincaré section 125
orthonormal basis 159 polar representation 15
orthonormal basis, general analysis with 250 polarization 297, 374
orthonormality, equivalent of for unequal polarization, circular 336
masses 207 potential energy of an oscillator 17
oscillation, harmonic, requirements for 39 power carried by waves 317
oscillator, driven, energy of 95–99 power for driven oscillator 98
oscillator, driven, full solution for 103 power in transmission line waves 339
oscillator, driven, in complex plane 89 power resonance curve 98
oscillator, electrical 10 Poynting vector 291, 300, 319
overdamped behavior 72 precession 108
pressure 305
pressure broadening 114
Pa 45 prism 385
panic, let’s not 326 probability density 24
parallel axis theorem 42 proton 107
parallel impedance 21 pure sinusoid 26
parallel RLC oscillator 136
paramagnetic 298
parameter 7 Q 70, 92
Parseval’s theorem 275 QCM 47, 49
partial derivative 281 quality factor 70, 92
particle in a box 243 quantization of energies 221
particular solution 103 quantization of frequencies 221
398 Index
quantum beats 171 self-sustaining oscillation 291

quantum superpositions 151 semiclassical description 107
Quartz Crystal Microbalance 47, 49 sensitivity to initial conditions 121
quietest sound 322 sensor, gas pressure 71–72
series impedance 21
shear 47
radiation damping 81 shear modulus 48
radio 32, 106, 121, 329, 341 shear modulus table 46
radio frequency 110 Shear, William 48
radio receiver 105–106 SHM 6
radio, AM 32, 106, 121, 329, 341 signal generator 18
radio, FM 32, 121 simple harmonic motion 6
radioactive particles 58 simple harmonic motion, in complex
ray 368 plane 16, 18
RC filter 21, 36, 37 simple pendulum 41
real space 235 sine, Taylor series for 13
reed instrument 316 sinusoid, pure 26
reflected wave 344, 347, 358–359 slippery devil 8
reflection 375 smallest possible wavelength 225
refraction 369
Snell’s Law 369, 375
rescaling Hilbert space for unequal
solution for a DEQ 4
masses 208
sonogram 265
resistor 19
sound intensity 321
resonance 84
sound waves 305
resonance amplitude 89–92
sound, speed of (see speed of sound) 311
resonance curve, power 98
spectroscopy 113
Reynolds number 75
speed of light 290
Reynolds, Osborne 75
speed of sound 310
RF 110
speed of sound in a gas 311
RFID tags 383
speed of sound in liquid 312
RG58 cable 304, 362
speed of sound in solid 312
rigid shape moving 283
spherical aberration 370
RL filter 23, 37
RLC oscillator 105, 136 spin angular momentum 107
rms amplitude 319 spin down 112, 164
room mean square amplitude 319 spin up 112, 164
rope waves, functions for 344 spring 9
rotating reference frame 110 spring constant 4
rotation matrix 210, 211 spring cut in half 43
rotor 51 spring, vertical 9
rubber band 240 square wave 134, 252
squared sinusoid, average value of 97
stable equilibrium 3
sampling 258 standard deviation 278
scales, musical 170 standing waves 219
Schrödinger’s equation 285 steady state behavior 87
Sears Tower 58 Stokes, George 74
second-order approximation 12 Stokes’ Law 74
second-order DEQ 4 strain 45
self-adjoint 200 strain hardening 45
Index 399
strange attractor 128 turbulent flow 64

stress 45 two-slit experiment 271
string, beaded, driven 238
string, eigenfunctions 233
string, frequencies for 230 ultimate stress 45
string, normal mode analysis 234 uncertainty 26
sub-harmonic resonance 119 underdamped behavior 72
superposing normal modes 146 Universal Gas Constant 310
Superposition Principle for driven
systems 101
Superposition Principle for traveling Van der Waals attraction 59
waves 285 viscosities, table of 75
supersolid 61 viscosity 74
surface tension 341 viscous damping 64
susceptibility 298 voltage 10
voltage, complex version of 19
Taipei 101 60
tapping mode AFM 93 washing machine 86
Taylor series 12 watch crystal 70–71
Taylor series, three important water waves 341
uses of 34 wave equation 282
television 33 wave equation, solutions 283
terminator 343, 362 wave function, real and imaginary
thermal evaporation 49 parts 25
thermal vibrations 80 wavefront 368
thickness monitor 49 wavefunction 24
thin film interference 384 wavenumber 24, 220
three-dimensional waves 337 wavenumber for Fourier analysis 248
three-step procedure 4 wavepacket 326
tilde 19 waves in crystals 226
TIRFM 377 wavevector 337
top, child’s 108 weasel 135, 382
torque due to magnetic field 108 well-tempered scale 170
torsion balance 52 whale call 266
torsional oscillator 51 wind instruments 314
total energy of an oscillator 17 wind tunnel 64
total internal reflection 371 windowing 259, 276
transients 102 wire, torsion of 61
transmission line waves, power in 339
transmission line, impedance
of 360, 363, 382 yield stress 45
transmission lines 301 yield stress table 46
transmitted wave 344, 350, 358–359 yielding 45
transverse momentum 293 Young, Thomas 44, 271
traveling wave 283 Young’s modulus 44
traveling wave speed 284 Young’s modulus table 46
triangle wave 270
trumpet 316
tube, torsion of 50–51 zeroth-order approximation 12

Wave and Oscillation PDF PDF

Uploaded by

Copyright:

Available Formats

Wave and Oscillation PDF PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Wave and Oscillation PDF PDF

Uploaded by

Copyright:

Available Formats

Waves and Oscillations

This page intentionally left blank

Walter Fox Smith

Oxford New York

Copyright © 2010 by Oxford University Press

Oxford is a registered trademark of Oxford University Press

All rights reserved. No part of this publication may be reproduced,

Library of Congress Cataloging-in-Publication Data

Printed in the United States of America

This checks your understanding of the ideas in the preceding material.

Concept and skill inventory

Learning Tools Used in This Book ix

1. Simple Harmonic Motion 1

2. Examples of Simple Harmonic Motion 39

3.3 Exponential decay of energy 69

4. Driven Oscillations and Resonance 84

5. Symmetric Coupled Oscillators and Hilbert Space 137

6. Asymmetric Coupled Oscillators and the Eigenvalue Equation 179

Concept and skill inventory 208

7. String Theory 216

8. Fourier Analysis 246

9. Traveling Waves 280

10. Waves at Interfaces 343

Appendix A Group Velocity for an Arbitrary

All around us, sinusoidal waves astound us!

1.1 Sinusoidal oscillations are everywhere

Figure 1.1.1 Top: motion of an atom in a

Figure 1.1.2 Motion of a tree branch

1.2 The physics and mathematics behind simple sinusoidal motion

This is a “differential equation” or DEQ which simply means that it is an equation

1.3 Important parameters and adjustable constants of simple

Figure 1.3.1 Simple harmonic motion

2. The period T . This is the time between successive maxima, or equivalently

(This equation is shown with a double border because we’ll be referring to it

3. Physicists use the symbol “≡” to mean “is defined to be.”

Your turn: From these, you should now show that

Aside: The arctangent function

1.4 Mass on a spring

The total force on the mass is

1.5 Electrical oscillators

Figure 1.5.1 Electrical oscillator.

Combining this with equation (1.5.3) gives

This is the voltage across the inductor, so

Electrical oscillators are tremendously important in electronic circuits, from radio

Table 1.5.1. Isomorphism between mechanical and electrical oscillators

Mass and spring Electrical oscillator

Position relative to equilibrium x Charge q on capacitor

5. Answer to self-test: Comparing equations (1.2.2)

1.6 Review of Taylor series approximations

Figure 1.6.1 Graphical illustration of the Taylor series approximation.

From this, we can see why the approximation

Your turn: Show that

1.7 Euler’s equation

eiθ = cos θ + i sin θ. (1.7.1)

1.8 Review of complex numbers

Figure 1.8.1 a: The complex plane.

a = A cos θ and b = A sin θ, so that

z = a + ib = A cos θ + i(A sin θ ) = Aeiθ

(using Euler’s equation) and also