Acoustics-A Textbook For Engineers and Physicists Volume II Applications

Jerry H.
Ginsberg
Acoustics—
A Textbook for
Engineers
and Physicists
Volume II: Applications
Acoustics—A Textbook for Engineers
and Physicists
Jerry H. Ginsberg
Acoustics—A Textbook
for Engineers and Physicists
123
Jerry H. Ginsberg
G. W. Woodruff School of Mechanical
Engineering
Georgia Institute of Technology
Dunwoody, GA
USA
ISBN 978-3-319-56846-1 ISBN 978-3-319-56847-8 (eBook)

DOI 10.1007/978-3-319-56847-8
Library of Congress Control Number: 2017937706
© Springer International Publishing AG 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
Printed on acid-free paper
This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
The ASA Press
The ASA Press imprint represents a collaboration between the Acoustical Society
of America and Springer dedicated to encouraging the publication of important new
books in acoustics. Published titles are intended to reflect the full range of research
in acoustics. ASA Press books can include all types of books published by Springer
and may appear in any appropriate Springer book series.
Editorial Board
Mark F. Hamilton (Chair), University of Texas at Austin

James Cottingham, Coe College
Diana Deutsch, University of California, San Diego
Timothy F. Duda, Woods Hole Oceanographic Institution
Robin Glosemeyer Petrone, Threshold Acoustics
William M. Hartmann (Ex Officio), Michigan State University
James F. Lynch (Ex Officio), Woods Hole Oceanographic Institution
Philip L. Marston, Washington State University
Arthur N. Popper (Ex Officio), University of Maryland
Martin Siderius, Portland State University
Andrea M. Simmons, Brown University
Ning Xiang, Rensselaer Polytechnic Institute
William Yost, Arizona State University
The Acoustical Society of America
On December 27, 1928, a group of scientists and engineers met at Bell Telephone
Laboratories in New York City to discuss organizing a society dedicated to the field
of acoustics. Plans developed rapidly and the Acoustical Society of America
(ASA) held its first meeting on May 10–11, 1929, with a charter membership of
about 450. Today, ASA has a worldwide membership of 7000.
The scope of this new society incorporated a broad range of technical areas that
continues to be reflected in ASA’s present-day endeavors. Today, ASA serves the
interests of its members and the acoustics community in all branches of acoustics,
both theoretical and applied. To achieve this goal, ASA has established technical
committees charged with keeping abreast of the developments and needs of
membership in specialized fields as well as identifying new ones as they develop.
The technical committees include acoustical oceanography, animal bioacoustics,
architectural acoustics, biomedical acoustics, engineering acoustics, musical
acoustics, noise, physical acoustics, psychological and physiological acoustics,
signal processing in acoustics, speech communication, structural acoustics and
vibration, and underwater acoustics. This diversity is one of the Society’s unique
and strongest assets since it so strongly fosters and encourages cross-disciplinary
learning, collaboration, and interactions.
ASA publications and meetings incorporate the diversity of these technical
committees. In particular, publications play a major role in the Society. The Journal
of the Acoustical Society of America (JASA) includes contributed papers and patent
reviews. JASA Express Letters (JASA-EL) and Proceedings of Meetings on
Acoustics (POMA) are online, open-access publications, offering rapid publication.
Acoustics Today, published quarterly, is a popular open-access magazine. Other key
features of ASA’s publishing program include books, reprints of classic acoustics
texts, and videos.
ASA’s biannual meetings offer opportunities for attendees to share information,
with strong support throughout the career continuum, from students to retirees.
Meetings incorporate many opportunities for professional and social interactions,
and attendees find the personal contacts a rewarding experience. These experiences
result in building a robust network of fellow scientists and engineers, many of
whom became lifelong friends and colleagues.
From the Society’s inception, members recognized the importance of developing
acoustical standards with a focus on terminology, measurement procedures, and
criteria for determining the effects of noise and vibration. The ASA Standards
Program serves as the Secretariat for four American National Standards Institute
Committees and provides administrative support for several international standards
committees.
Throughout its history to present day, ASA’s strength resides in attracting the
interest and commitment of scholars devoted to promoting the knowledge and
practical applications of acoustics. The unselfish activity of these individuals in the
development of the Society is largely responsible for ASA’s growth and present
stature.
To Leah Morgan, Elizabeth Rachel,
and Abigail Rose, my grandchildren.
Each is talented, each is beautiful,
each is unique, each is amazing.
I love them.
Preface
The Basic Concept
Because you are reading this book, there is a strong likelihood that you are familiar
with its companion, Volume 1. If that is so, then you are well acquainted with my
core philosophy of instruction. It is evident that students should be prepared to
address technical issues in the future. An essential component in doing so is the
instructor, but even the most talented instructor needs support. Fulfilling that need
has been the motivation for all my prior books, but even more so it is my objective
for these books. To meet the present and future needs of students and instructors,
treatments of various principles and concepts feature extensive explanations of the
motivation and organization of the derivation, as well as thorough descriptions
of the steps that are implemented. Some of the derivations and explanations
I believe are unique to this book. In most cases, the derived principles are
accompanied by discussions of their physical meaning.
Examples are numerous. All are my own creation. Indeed, the thought of
creating a suitable set dissuaded me for a long time from beginning to write
Acoustics—A Textbook for Engineers and Physicists, even though I believed that
there was a strong need for a text like it. Most examples are more than simple
applications of derived formulas. I selected many of these examples not only to
illustrate the associated theory, but also to be simplified versions of issues the
student might encounter in practice. Equally important to their selection was
whether treating the results as the basis for small case studies would enlighten the
student about the phenomena associated with that theory. I also used the examples
as a vehicle to bring to the fore the fact that in many situations, alternative for-
mulations of an analysis might be viable. I endeavored to use the examples as a way
of assisting the student to recognize when these alternatives exist, and to assist them
to recognize in other situations which one is best. At the same time, I endeavored to
recognize the imperative that the examples be cognizant of the capabilities that can
be expected of a student who is being exposed to acoustics for the first time.
xi
xii Preface
Each solution explains why the example is important, why the solution proceeds
as it does, how to perform unfamiliar operations, what can be learned from the results
about fundamental behaviors, and why the qualitative aspects of the results are
consistent with the underlying fundamental principles. Some examples analyze
systems by more than one method. This serves to enhance the student’s fundamental
understanding of the underlying physical processes, as well as enhancing the ability
to make the appropriate line of attack when confronted with a new situation. The
advent and wide availability of computational software is exploited to lend greater
realism to some examples. When the usage of software entails any potentially
problematic aspects, especially concerning algorithms and their implementation,
those issues are addressed explicitly, sometimes with program fragments. In
recognition of the importance of computations, and to help students concentrate on
the acoustical aspects when they solve homework exercises, the MATLAB code
used to solve the examples is available for download from the Springer server.
At its inception, Acoustics—A Textbook for Engineers and Physicists was a
single volume. The sequence of topics for the early chapters was a slightly modified
version of the courses I taught at Georgia Tech. I wrote the later chapters to fill the
spectrum of subjects that I consider to constitute the core concepts and techniques a
student in physics or engineering is likely to encounter. Space limitations, as well as
my desire that this should be a textbook, rather than a monograph, dictated the
scope in the later chapters. For this reason, they were written with the notion that
they should expose the student to the fundamental phenomena and provide the
fundamental tools required to study and research these phenomena. No chapter
attempts to bring the student up to the current state of the art in that subject.
Some instructor’s might consider that some chapters delve too deeply into the
subject. I do not agree with this sentiment. Some parts address questions students
have asked me, and the development in other parts serves to motivate students by
demonstrating interesting and enlightening phenomena. Nevertheless, if one does
not desire to take advantage of the depth of treatment, the first few sections of each
chapter should be adequate to proceed to the next chapter.
After I completed this book, it was evident that it would be quite large. This
caused me to recall my days long ago when I was a student, and disliked carrying
very large textbooks. Concurrently, when I surveyed the Table of Contents, I real-
ized that the manuscript divided naturally into two parts. The first six chapters
constitute a foundation that I consider requisite knowledge for anyone active in the
physical aspects of acoustics. My examination also led me to the realization that the
nature of the last seven chapters was different. Few individuals are equally familiar
with all of them. This certainly was true in my case. Filling in the gaps in my
knowledge, particularly in Chap. 11 on geometrical acoustics, was quite enriching
and fulfilling. I wanted to provide that same sense to instructors who might not be
familiar with the subject of one or more of the later chapters. I also wanted to
prepare students for further study in each of those topics. I foresee using Volume 2
either in a survey course covering selected or all chapters, or else to begin spe-
cialized courses on the subject of a single chapter. Both usages would have been
followed when I taught acoustics at Georgia Tech. The Acoustics II course covered
Preface xiii
Chap. 7 on radiation from objects in free space, Chap. 8 on radiation from vibrating
regions in a baffle, the beginning portion of Chap. 9 on waveguides, and the
beginning portion of Chap. 10 on enclosures. There also were specialized courses
on propagation on the ocean and atmosphere, nonlinear acoustics, and structural
acoustics. The associated chapters would not have been adequate for those spe-
cialized courses, but I believe they would provide a strong foundation with which to
initiate such courses.
Technical Content
The chapters to be found in this volume are sequenced according to what I believe
is sensible in terms of the level of sophistication and analytical difficulty. The
highlights of each chapter were discussed in the Preface to Volume I, but it is
appropriate to summarize the scope of each chapter. Radiation from vibrating
objects is the subject of Chap. 7. Configurations that are addressed are spheres and
hemispheres, infinitely long cylinders, and three numerical techniques that exem-
plify the formulations in current use. Radiation from a piston in a baffle is
encountered in a diverse set of applications, so Chap. 8 is devoted to a thorough
exploration of farfield and nearfield properties. Two examples explore radiation
resulting from square wave motion. The difference between these examples is
whether the analysis is formulated in the frequency or time domains. The reader
might find the results to be quite interesting.
The field within a waveguide is the subject of Chap. 9. First to be studied is the
Webster horn equation for one-dimensional waveguides. The investigation of
two-dimensional waveguides introduces the concept of modal analysis. It concludes
with an investigation of coupling of the acoustic field with elastic walls that are
described by plate theory. The basic theory for three-dimensional waveguides is
developed. It is applied to waveguides whose cross section is rectangular and then
cylindrical. Chapter 10 is devoted to the sound field within an enclosed region. It
begins by using the waveguide representation to explain the alternative descriptions
of the field as a set of waves that propagate in multiple directions, or as a set of
cavity modes that are standing waves. Both analytical descriptions are developed,
with emphasis on the situations where each is best employed. An interesting
example uses an infinite series of cavity modes to describe the field within a
two-dimensional rectangular enclosure due to a point source. After the analysis, an
alternative using the method of images highlights the notion that the selection of an
optimal analytical approach sometimes requires consideration of what the objective
is. The closure of this chapter describes the Rayleigh-Ritz method and a formulation
according to Dowell’s approximation. The former is often included in standard
texts, but the latter is relatively recent.
Chapter 11 is devoted to geometrical acoustics. Mean flow effects are excluded
in order to emphasize how high-frequency rays and wavefronts may be determined.
A vertically stratified fluid, which is a fundamental model for ocean acoustics,
xiv Preface
is examined first. After that, the ray tracing equations for media whose properties
depend arbitrarily on location are developed. Algorithms for solving the governing
equations in each case are formulated. Of particular interest is the example of the
passage of a plane wave through fluid that is cylindrically heterogeneous.
Demonstrations of a caustic and folding of the wavefront beyond the caustic are
quite captivating. The chapter closes with a presentation of Fermat’s principle. The
calculus of variations is developed there.
Scattering from bodies surrounded by a fluid is the subject of Chap. 12. The
Born approximation for heterogenous media is derived. Rayleigh scattering and its
relation to the Born approximation are the next topic. The metrics commonly
sought from a scattering study, such as target strength, are discussed. After that,
Kirchhoff scattering theory and its relation to geometrical acoustics are explored.
The chapter closes with the application of spherical harmonics to analyze scattering
from a rigid sphere and a spherical shell. These studies shed light on the transition
from Rayleigh to Kirchhoff scattering, as well as the fundamental importance of
fluid loading relative to elasticity.
Chapter 13 closes the textbook with an exploration of nonlinear acoustic anal-
yses and phenomena. The bulk of the chapter is devoted to simple plane waves
described by the Riemann solution. Techniques for evaluating it are discussed and
used to examine harmonic generation and depletion, followed by the propagation of
shocks. The Rankine-Hugoniot relations for weak shock are derived and shown
with simple mathematics to lead to the phenomena of old age and acoustic satu-
ration. A nonlinear wave equation is derived as the basis for study of multidi-
mensional nonlinear waves. Its first usage leads to differential equations governing
the position dependence of Fourier series coefficients for a plane wave in a dissi-
pative fluid. Perturbation analysis techniques for the nonlinear wave equation are
developed for plane waves and then extended to radially symmetric spherical waves
and two-dimensional waves radiated by a vibrating plate. The latter leads to
demonstration of the phenomenon of self-refraction, in which the rays and wave-
fronts are modified by the associated pressure and particle velocity fields.
Although the overriding precept of the text is that all topics must be fully
explained as they arise, two appendices are provided for further assistance. One is
devoted to derivation of the coordinate transformations and vector differential
operators in spherical and cylindrical coordinates. The second describes Fourier
transforms and their application. Fourier transforms appear in the main body only in
the treatment of radiation from an infinite cylinder and from a transducer in a baffle.
Those analyses could be addressed by invoking FFT techniques, albeit without the
benefit of an algebraic solution. However, Fourier transforms are a ubiquitous
thread that runs through the technical literature, and the treatment of this mathe-
matical tool in a sense ties together the dual nature of the time and frequency
domains.
The chapters of Volume II of Acoustics—A Textbook for Engineers and
Physicists are sequenced in a manner that I believe to be sensible from the view-
point of ascending difficulty, as well as the technical sophistication. For instance, it
would not make sense to study piston radiation in Chap. 8 before the Kirchoff–
Preface xv
Helmholtz integral theorem is developed in Chap. 7. From the opening motivational

discussion of waveguides in Chap. 10 to the development of natural cavity modes,
there is much reliance on the nature of propagation modes of a waveguide, which is
the subject of Chap. 9. Chapter 12 on scattering relies on the KHIT and multipole
expansions, which are developed in Chap. 7. Chapters 11 and 13 are exceptions.
Both only require concepts developed in the first four chapters. Their placement is
based on my perception of what instructors expect to find. My own research in
acoustics began in the nonlinear regime, so if I followed my preference, the last
chapter might have appeared earlier. In the same vein, I think that exposure of
students to ray tracing for heterogeneous media would interest them greatly and
thereby serve as a strong motivational tool.
Writing this volume has helped me close gaps in my knowledge of acoustics.
I very much enjoyed writing it. If you are a student, I hope you learn much from it
and find it useful for your future endeavors. If you are an instructor, I hope that this
work captivates you and enhances your teaching efforts. If you are a reader on a
self-study path, then I think you will find this volume to be exceptionally helpful
and instructive, for I have gone to great effort to provide all you need to pursue your
studies.
Dunwoody, GA, USA Jerry H. Ginsberg

Acknowledgements
I am indebted to many individuals for providing motivation to write these books.

Above all, neither would exist if I had not met my good friend, Allan Pierce, when
I interviewed in 1980 for a professorship at the Georgia Institute of Technology.
Working with him convinced me to extend my knowledge of acoustics beyond the
specialized subject of nonlinear acoustics that was part of my early career. Over the
years, our discussions were quite revelatory regarding where there were gaps in
my knowledge of the subject. Furthermore, I hope he is not offended by this remark,
but learning from, and then teaching from, his book convinced me of the necessity
that I write Acoustics—A Textbook for Engineers and Physicists.
In the six-year interval, during which I wrote these books what I needed most
was assurance that the effort was worth pursuing. Some of my colleagues at Georgia
Tech were quite supportive. Karim Sabra convinced me on several occasions that
I would be filling an important need. Students solving the homework exercises
should thank him because he suggested that the MATLAB code I used should be
publically available. Pete Rogers provided extremely useful critical remarks for an
early draft of my treatment of geometrical acoustics. My former Ph.D. students,
especially J. Gregory McDaniel at Boston University and Kuangchung Wu at the
NSWC Carderock Division of the Naval Sea Systems Command, were especially
enthusiastic. I also am indebted to those attendees at many Acoustical Society of
America meetings who I waylaid to discuss my writing efforts. They are too
numerous to list, and I am sure that I have forgotten some names, but I greatly
appreciate their attention. I owe Mark Hamilton of the University of Texas at Austin
a special debt because he convinced me to participate in the ASA Book program
under the aegis of Springer Publishing. Sara Kate Heukerott, my Editor at Springer,
was quite understanding of my requests. Her expertise was a great aid as we
assembled this project. Some might be surprised at the inclusion of my grand-
daughter, Leah Morgan Ginsberg, in the list of folks deserving recognition. Early in
the writing stage, because she was a proficient clarinetist, I sought her assistance for
the discussion of music in Chap. 1. Then at the conclusion, as she approached
graduation from Georgia Tech in the G. W. Woodruff School of Mechanical
xvii
xviii Acknowledgements
Engineering, from which I had retired, she served as my sounding board and
spokeswoman for students when I deliberated how best to disseminate this work.
In addition to Leah’s role, my family was essential to the effort. The forbearance
of my wife, Rona, while I focused on writing, ignored other responsibilities, and
forgot many things that I still cannot remember, astonishes me, even now that my
efforts are over. She went through this experience before when I wrote my prior
books on statics, dynamics, and vibrations. However, none of those experiences
could have prepared her for the intensity and duration of the present effort. My sons,
Mitchell and Daniel, had similar experiences when they lived at home. Although
they and their wives, Tracie and Jessica, were not as strongly impacted now,
I greatly appreciate their forbearance when I was not as communicative as I should
have been. My granddaughters, Leah, Beth, and Abby, inspire me by their dedi-
cation to their own activities. I hope that recognition of the pleasure their Papa
derived from creating these books will inspire them.
Dunwoody, GA, USA Jerry H. Ginsberg

Contents
Volume I: Fundamentals
1 Descriptions of Sound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Harmonic Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.1 Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2 Vectorial Representation . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.3 Complex Exponential Representation . . . . . . . . . . . . . . 9
1.1.4 Operations Using Complex Exponentials . . . . . . . . . . . 12
1.2 Averages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3 Metrics of Sound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.3.1 Sound Pressure Level . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.3.2 Human Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.3.3 Frequency Bands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.4 Transfer Between Time and Frequency Domains . . . . . . . . . . . . 41
1.4.1 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.4.2 Discrete Fourier Transforms . . . . . . . . . . . . . . . . . . . . . 52
1.4.3 Nyquist Sampling Criterion . . . . . . . . . . . . . . . . . . . . . 55
1.4.4 Fast Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . 59
1.4.5 Evaluation of Time Responses . . . . . . . . . . . . . . . . . . . 64
1.5 Spectral Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
1.5.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
1.5.2 Noise Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
1.6 Closure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
1.7 Homework Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2 Plane Waves: Time Domain Solutions. . . . . . . . . . . . . . . . . . . . . . . . 91
2.1 Continuum Equations in One Dimension . . . . . . . . . . . . . . . . . . 92
2.1.1 Conservation of Mass . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.1.2 Momentum Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 94
2.2 Linearization and the One-Dimensional Wave Equation . . . . . . 96
2.3 Equation of State and the Speed of Sound . . . . . . . . . . . . . . . . . 102
xix
xx Contents
2.4 The d’Alembert Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

2.4.1 Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
2.4.2 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
2.4.3 Harmonic Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
2.5 The Method of Wave Images . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
2.5.1 Initial Value Problem in an Infinite Domain . . . . . . . . . 124
2.5.2 Plane Waves in a Semi-infinite Domain . . . . . . . . . . . . 131
2.5.3 Plane Waves in a Finite Waveguide . . . . . . . . . . . . . . . 145
2.6 Analogous Vibratory Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 170
2.6.1 Stretched Cable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
2.6.2 Extensional Waves in an Elastic Bar . . . . . . . . . . . . . . 179
2.7 Closure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
3 Plane Waves: Frequency-Domain Solutions . . . . . . . . . . . . . . . . . . . 191
3.1 General Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
3.2 Waveguides with Boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
3.2.1 Impedance and Reflection Coefficients . . . . . . . . . . . . . 200
3.2.2 Evaluation of the Signal . . . . . . . . . . . . . . . . . . . . . . . . 207
3.2.3 Modal Properties and Resonances . . . . . . . . . . . . . . . . . 212
3.2.4 Impedance Tubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
3.3 Effects of Dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
3.3.1 Viscosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
3.3.2 Thermal Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
3.3.3 Molecular Relaxation . . . . . . . . . . . . . . . . . . . . . . . . . . 234
3.3.4 Absorption in the Atmosphere and Ocean. . . . . . . . . . . 238
3.3.5 Wall Friction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
3.4 Acoustical Transmission Lines . . . . . . . . . . . . . . . . . . . . . . . . . . 248
3.4.1 Junction Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
3.4.2 Time Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
3.4.3 Frequency-Domain Formulation for Long Segments . . . 257
3.5 Lumped Parameter Models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
3.5.1 Approximations for Short Branches . . . . . . . . . . . . . . . 272
3.5.2 Helmholtz Resonator. . . . . . . . . . . . . . . . . . . . . . . . . . . 278
3.6 Closure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
4 Principles and Equations for Multidimensional Phenomena . . . . . . 295
4.1 Fundamental Equations for an Ideal Gas . . . . . . . . . . . . . . . . . . 296
4.1.1 Continuity Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
4.1.2 Momentum Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 298
4.2 Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Contents xxi
4.3 Plane Waves in Three Dimensions . . . . . . . . . . . . . . . . . . . . . . . 306

4.3.1 Simple Plane Wave in the Time Domain . . . . . . . . . . . 307
4.3.2 Trace Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
4.3.3 Simple Plane Wave in the Frequency Domain . . . . . . . 316
4.4 Velocity Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
4.5 Energy Concepts and Principles . . . . . . . . . . . . . . . . . . . . . . . . . 324
4.5.1 Energy and Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
4.5.2 Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
4.5.3 Power Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
4.6 Closure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
5 Interface Phenomena for Planar Waves . . . . . . . . . . . . . . . . . . . . . . 347
5.1 Radiation Due to Surface Waves . . . . . . . . . . . . . . . . . . . . . . . . 347
5.1.1 Basic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
5.1.2 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
5.2 Reflection from a Surface Having a Local Impedance . . . . . . . . 358
5.2.1 Reflection from a Time-Domain Perspective . . . . . . . . . 358
5.2.2 Reflection from a Frequency-Domain Perspective . . . . . 365
5.3 Transmission and Reflection at an Interface Between Fluids . . . 373
5.3.1 Time-Domain Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 373
5.3.2 Frequency-Domain Analysis . . . . . . . . . . . . . . . . . . . . . 380
5.4 Propagation Through Layered Media . . . . . . . . . . . . . . . . . . . . . 390
5.4.1 Basic Analysis of Three Fluids . . . . . . . . . . . . . . . . . . . 390
5.4.2 Multiple Layers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
5.5 Solid Barriers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
5.5.1 General Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
5.5.2 Specific Barrier Models . . . . . . . . . . . . . . . . . . . . . . . . 414
5.6 Closure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
6 Spherical Waves and Point Sources . . . . . . . . . . . . . . . . . . . . . . . . . 433
6.1 Spherical Coordinates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
6.2 Radially Vibrating Sphere–Time-Domain Analysis . . . . . . . . . . 437
6.2.1 General Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
6.2.2 Radiation from a Uniformly Vibrating Sphere . . . . . . . 439
6.2.3 Acoustic Field in a Spherical Cavity . . . . . . . . . . . . . . . 444
6.3 Radially Vibrating Sphere–Frequency-Domain Analysis . . . . . . 453
6.3.2 Radiation from a Radially Vibrating Sphere . . . . . . . . . 454
6.3.3 Standing Waves in a Spherical Cavity . . . . . . . . . . . . . 462
6.4 Point Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
6.4.1 Single Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
6.4.2 Green’s Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
xxii Contents
6.4.3 Point Source Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . 483

6.4.4 Method of Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
6.5 Dipoles, Quadrupoles, and Multipoles . . . . . . . . . . . . . . . . . . . . 505
6.5.1 The Dipole Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 506
6.5.2 Radiation from a Translating Rigid Sphere . . . . . . . . . . 513
6.5.3 The Quadrupole Field . . . . . . . . . . . . . . . . . . . . . . . . . . 521
6.5.4 Multipole Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . 532
6.6 Doppler Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
6.6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
6.6.2 Moving Fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
6.6.3 Subsonic Point Source . . . . . . . . . . . . . . . . . . . . . . . . . 541
6.6.4 Supersonic Point Source . . . . . . . . . . . . . . . . . . . . . . . . 548
6.7 Closure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569

7 Radiation from Vibrating Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
7.1 Spherical Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
7.1.1 Separation of Variables . . . . . . . . . . . . . . . . . . . . . . . . . 2
7.1.2 Description of the Pressure Field . . . . . . . . . . . . . . . . . 15
7.1.3 Arbitrary Spatial Dependence . . . . . . . . . . . . . . . . . . . . 16
7.2 Radiation from a Spherical Body . . . . . . . . . . . . . . . . . . . . . . . . 19
7.2.1 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7.2.2 Important Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
7.2.3 Symmetry Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
7.2.4 Interaction with an Elastic Spherical Shell . . . . . . . . . . 36
7.3 Radiation from an Infinite Cylinder . . . . . . . . . . . . . . . . . . . . . . 53
7.3.1 Separation of Variables . . . . . . . . . . . . . . . . . . . . . . . . . 56
7.3.2 Transverse Dependence—Cylindrical Bessel
Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.3.3 Radiation Due to a Helical Surface Wave . . . . . . . . . . . 60
7.3.4 Axially Periodic Surface Vibration . . . . . . . . . . . . . . . . 68
7.3.5 Finite Length Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.4 Kirchhoff–Helmholtz Integral Theorem . . . . . . . . . . . . . . . . . . . 84
7.4.1 Derivation for an Acoustic Cavity . . . . . . . . . . . . . . . . 85
7.4.2 Acoustic Radiation into an Exterior Domain . . . . . . . . . 91
7.5 Numerical Methods for Radiation from Arbitrary Objects . . . . . 104
7.5.1 Source Superposition . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.5.2 Boundary Element Method . . . . . . . . . . . . . . . . . . . . . . 108
7.5.3 Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . 115
Contents xxiii
8 Radiation from a Source in a Baffle . . . . . . . . . . . . . . . . . . . . . . . . . 133

8.1 The Rayleigh Integral. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
8.2 Farfield Directivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
8.2.1 Cartesian Coordinate Description . . . . . . . . . . . . . . . . . 137
8.2.2 Farfield of a Piston Transducer . . . . . . . . . . . . . . . . . . . 144
8.3 Axial Dependence for a Circular Transducer . . . . . . . . . . . . . . . 153
8.4 An Overall Picture of the Pressure Field . . . . . . . . . . . . . . . . . . 158
8.5 Radiation Impedance of a Circular Piston . . . . . . . . . . . . . . . . . 166
8.6 Time Domain Rayleigh Integral . . . . . . . . . . . . . . . . . . . . . . . . . 176
9 Modal Analysis of Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
9.1 Propagation in a Horn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
9.1.1 The Webster Horn Equation . . . . . . . . . . . . . . . . . . . . . 188
9.1.2 Exponential Horn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
9.1.3 Group Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
9.1.4 WKB Solution for an Arbitrary Horn . . . . . . . . . . . . . . 207
9.2 Two-Dimensional Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . 216
9.2.2 Rigid Walls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
9.2.3 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
9.2.4 Flexible Walls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
9.2.5 Orthogonality and Signal Generation . . . . . . . . . . . . . . 243
9.3 Three-Dimensional Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . 253
9.3.1 General Analytical Procedure . . . . . . . . . . . . . . . . . . . . 253
9.3.2 Rectangular Waveguide . . . . . . . . . . . . . . . . . . . . . . . . 257
9.3.3 Circular Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
10 Modal Analysis of Enclosures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
10.1 Fundamental Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
10.1.1 Wall-Induced Signals . . . . . . . . . . . . . . . . . . . . . . . . . . 291
10.1.2 Source Excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
10.2 Frequency-Domain Analysis Using Forced Cavity Modes . . . . . 297
10.2.1 Rectangular Enclosures . . . . . . . . . . . . . . . . . . . . . . . . . 297
10.2.2 Spherical Cavities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
10.2.3 Cylindrical Enclosures . . . . . . . . . . . . . . . . . . . . . . . . . 323
10.3 Analysis Using Natural Cavity Modes . . . . . . . . . . . . . . . . . . . . 336
10.3.1 Equations Governing Cavity Modes . . . . . . . . . . . . . . . 337
10.3.2 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
10.3.3 Analysis of the Pressure Field. . . . . . . . . . . . . . . . . . . . 340
10.3.4 Rectangular Cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
10.3.5 Cylindrical Cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
10.3.6 Spherical Cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
xxiv Contents
10.4 Approximate Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363

10.4.1 The Rayleigh Ratio and Its Uses . . . . . . . . . . . . . . . . . 364
10.4.2 Dowell’s Approximation . . . . . . . . . . . . . . . . . . . . . . . . 379
11 Geometrical Acoustics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
11.1 Basic Considerations: Wavefronts and Rays . . . . . . . . . . . . . . . 406
11.1.1 Field Equations for an Inhomogeneous Fluid . . . . . . . . 409
11.1.2 Reflection and Refraction of Rays . . . . . . . . . . . . . . . . 413
11.2 Propagation in a Vertically Stratified Medium . . . . . . . . . . . . . . 417
11.2.1 Snell’s Law for Vertical Heterogeneity . . . . . . . . . . . . . 418
11.2.2 Intensity and Focusing Factor . . . . . . . . . . . . . . . . . . . . 428
11.3 Arbitrary Heterogeneous Fluids . . . . . . . . . . . . . . . . . . . . . . . . . 438
11.3.1 Ray Tracing Equations . . . . . . . . . . . . . . . . . . . . . . . . . 438
11.3.2 Amplitude Dependence . . . . . . . . . . . . . . . . . . . . . . . . . 443
11.4 Fermat’s Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
12 Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
12.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
12.2 Scattering by Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
12.2.1 General Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
12.2.2 The Born Approximation . . . . . . . . . . . . . . . . . . . . . . . 485
12.3 Rayleigh Scattering Limit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
12.3.1 The Rayleigh Limit of the Born Approximation . . . . . . 490
12.3.2 Mismatched Heterogeneous Region . . . . . . . . . . . . . . . 492
12.3.3 Scattering from a Rigid Body . . . . . . . . . . . . . . . . . . . . 494
12.4 Measurements and Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
12.5 High-frequency Approximation . . . . . . . . . . . . . . . . . . . . . . . . . 510
12.6 Scattering from Spheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
12.6.1 Stationary Spherical Scatterer . . . . . . . . . . . . . . . . . . . . 515
12.6.2 Scattering by an Elastic Spherical Shell . . . . . . . . . . . . 522
13 Nonlinear Acoustic Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
13.1 Riemann’s Solution for Plane Waves . . . . . . . . . . . . . . . . . . . . . 540
13.1.1 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 540
13.1.2 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543
13.1.3 Boundary and Initial Conditions . . . . . . . . . . . . . . . . . . 545
13.1.4 Equations of State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
13.1.5 Quantitative Evaluations . . . . . . . . . . . . . . . . . . . . . . . . 553
13.2 Effects of Nonlinearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 568
13.2.1 Harmonic Generation . . . . . . . . . . . . . . . . . . . . . . . . . . 568
13.2.2 Shock Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 574
13.2.3 Propagation of Weak Shocks . . . . . . . . . . . . . . . . . . . . 577
Contents xxv
13.3 General Analytical Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . 597

13.3.1 A Nonlinear Wave Equation . . . . . . . . . . . . . . . . . . . . . 598
13.3.2 Frequency-Domain Formulation . . . . . . . . . . . . . . . . . . 602
13.3.3 Regular Perturbation Series Expansion . . . . . . . . . . . . . 611
13.3.4 Method of Strained Coordinates . . . . . . . . . . . . . . . . . . 620
13.4 Multidimensional Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 625
13.4.1 Finite Amplitude Spherical Wave . . . . . . . . . . . . . . . . . 625
13.4.2 Waves in Cartesian Coordinates . . . . . . . . . . . . . . . . . . 637
13.5 Further Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654
Appendix A: Curvilinear Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 665
Appendix B: Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 677
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 689
About the Author
Jerry H. Ginsberg began his technical education at

the Bronx High School of Science, from which he
graduated in 1961. This was followed by a B.S.C.E.
degree in 1965 from the Cooper Union and an E.Sc.D.
degree in engineering mechanics from Columbia
University in 1970, where he held Guggenheim and
NASA Fellowships. From 1969 to 1973, he was an
Assistant Professor in the School of Aeronautics,
Astronautics, and Engineering Science at Purdue
University. He then transferred to Purdue’s School of
Mechanical Engineering, where he was promoted to
Associate Professor in 1974. In the 1975–1976 aca-
demic year, he was a Fulbright-Hayes Advanced
Research Fellow at the École Nationale Supérieure
d’Électricité et de Mécanique in Nancy, France. He
came to Georgia Tech in 1980 as a Professor in the
School of Mechanical Engineering, which awarded
him the George W. Woodruff Chair in 1989. He
retired in June 2008. His prior publications include
five textbooks in statics, dynamics, and vibrations,
most in several editions, as well as more than one
hundred and twenty refereed papers covering these
subjects. Dr. Ginsberg became a Fellow of the
Acoustical Society of America in 1987 and a Fellow
of the American Society of Mechanical Engineers in
1989. The awards and recognitions he has received
include Georgia Tech Professor of the Year (1994),
ASEE Archie Higdon Distinguished Educator in
Mechanics (1998), ASA Trent-Crede Medal (2005),
xxvii
xxviii About the Author
ASME Per Bruel Gold Medal in Noise Control and

Acoustics (2007), and the ASA Rossing Prize in
Acoustics Education (2010). In addition to his tech-
nical activities, he is an exceptional photographer.
List of Examples
The name of the MATLAB programs used to solve the respective example are
enclosed in parentheses. They are available for download from the Springer
website.
Example 7.1 (Legendre_function_series.m) Evaluation

of a Legendre polynomial series to fit a given function,
and evaluation of the series length required to have an
error below a specified level. . . . . . . . . . . . . . . . . . . .. 7
Example 7.2 (none) Determination of the spherical harmonic series
for the farfield radiated by a combination of a
monopole, a dipole, and a longitudinal quadrupole . . . .. 17
Example 7.3 (Cap_in_sphere_radiation.m) Determination
of the spherical harmonic series representation of the
axisymmetric pressure field radiated by a sphere that
has a cap of arbitrary size that executes a translational
vibration, and is rigid elsewhere . . . . . . . . . . . . . . . . .. 24
Example 7.4 (Sphere_radiation_baffle_vs_free.m) Evaluation
of spherical harmonic series for three cases: a
hemisphere that executes a translational vibration
in a rigid baffle, the same motion for a pressure-release
baffle, and a rigid sphere whose forward half translates
while the back half is stationary . . . . . . . . . . . . . . . . .. 32
Example 7.5 (Submerged_spher_shell_response.m) Evaluation
of the surface vibration and pressure when a
submerged spherical shell is excited by a harmonic
force concentrated at the polar apex . . . . . . . . . . . . . .. 46
Example 7.6 (Z_rad_helical.m) Determination and evaluation
of the radiation impedance for a single helical wave
of radial displacement on a cylinder . . . . . . . . . . . . . .. 67
xxix
xxx List of Examples
Example 7.7 (Line_source_finite.m) Determination of the pressure

field radiated by a uniform line source of finite length,
and comparison of its field properties to those of an
infinite line source . . . . . . . . . . . . . . . . . . . . . . . . . .. 73
Example 7.8 (Cello_string_radiation.m) Evaluation of the pressure
field radiated by a cello’s string based on a model
of the string as a cylinder whose cross sections execute
translational vibration . . . . . . . . . . . . . . . . . . . . . . . .. 79
Example 7.9 (KHIT_spher_cavity.m) Formulation of the
Kirchhoff-Helmholtz integral theorem for a spherical
cavity by using the analytical solution to set the
pressure and particle velocity at the wall, with special
attention to the behavior as the field point approaches
the wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 88
Example 7.10 (KHIT_cyl_radiation.m) Usage of the radiation
impedance of an infinite cylinder as the basis for an
approximate formulation of the Kirchhoff-Helmholtz
Integral Theorem for a finite cylinder, with emphasis
on evaluation of the integral to identify the farfield
directivity at several frequencies . . . . . . . . . . . . . . . . .. 94
Example 8.1 (Directivity_rectangular_transducer.m) Usage of the
Rayleigh integral to evaluate the farfield radiated by a
rectangular strip that executes a transverse vibration in
a rigid baffle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 140
Example 8.2 (Example_piston_directivities.m) Comparison
of the farfield properties for radiation from an
oscillating piston or a vibrating membrane in a rigid
baffle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 149
Example 8.3 (Anti_piston_example.m) Determination of the axial
dependence of the pressure radiated by a so-called
anti-piston, which consists of transducer that has an
outer region that vibrates 180° out-of-phase from the
central circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 156
Example 8.4 (Rayleigh_integral_hemisphere.m) Evaluation of the
axial dependence and farfield radiated by a hemisphere
in a rigid baffle that translates as a rigid body, based on
phase-shifting the normal velocity to place it in the
plane of the baffle, followed by comparison of the
results to those obtained from the spherical harmonic
solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 163
Example 8.5 (Piston_square_wave.m) Evaluation of the time
dependence of the radiated power and the pressure
waveform at a field point on the axis of a circular
piston that executes a square wave oscillation. . . . . . . .. 172
List of Examples xxxi
Example 9.1 Usage of the general solution of the Webster horn

equation to evaluate the two-port mobility for an
exponential transition between waveguides having
different cross-sectional sizes . . . . . . . . . . . . . . . . . . .. 199
Example 9.2 (Group_velocity_from_data.m) Evaluation of the
waveform at two locations in an exponential horn that
is driven by a biharmonic input, followed by the usage
of those waveforms to determine the group velocity,
and comparison of the group velocity to the value
derived from the dispersion equation . . . . . . . . . . . . . .. 205
Example 9.3 (Webster_horn_numerical_vs_WKB.m) Evaluation
of the WKB solution for the pressure distribution in
a plane waveguide whose cross-sectional area varies
sinusoidally, with a small minimum, followed by
comparison of the results to those obtained from a
finite difference solution of the Webster horn
equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 211
Example 9.4 (Compliant_2d_waveguide.m) Determination of the
phase and group velocity and the transverse mode
functions of a two-dimensional waveguide for which
one wall is rigid and the other is purely resistive . . . . .. 230
Example 9.5 (Elastic_2D_waveguide.m) Determination of the
eigenvalues of the transverse mode functions and the
ratio of plate displacement to the pressure amplitude at
selected frequencies for a two-dimensional waveguide
whose walls are identical elastic plates . . . . . . . . . . . .. 235
Example 9.6 (Two_dim_waveguide_excitation.m) Determination
of the axial and transverse dependence of pressure
amplitude in a two-dimensional waveguide whose
floor is rigid and whose upper boundary is a free
surface when the excitation stems from a vibrating
ribbon transducer aligned horizontally at one end . . . . .. 248
Example 9.7 (Rect_waveguide_line_source.m) Determination of the
axial and transverse dependence of pressure amplitude
in a three-dimensional waveguide whose sides and
floor are rigid and whose upper boundary is a free
surface when the excitation stems from a vibrating
ribbon transducer aligned vertically . . . . . . . . . . . . . . .. 262
Example 9.8 (Annular_waveguide_modes.m) Determination
of the eigenvalues and transverse mode functions
for a waveguide that is the region bounded
by concentric rigid cylinders . . . . . . . . . . . . . . . . . . .. 275
xxxii List of Examples
Example 9.9 (Piston_circular_waveguide.m) Determination

of the axial dependence of the pressure amplitude
in a circular waveguide when a vibrating piston of
lesser diameter is situated concentrically at one end . . .. 280
Example 10.1 (Three_D_cavity_force_plate.m) Evaluation
of the frequency dependence of the force required
to move a massive plate that is part of one wall
in a hard-walled rectangular cavity . . . . . . . . . . . . . . .. 301
Example 10.2 (Piston_transducer_rectangular_cavity.m) Evaluation
of the frequency dependence of the pressure amplitude
at selected locations and the force required to drive
a vibrating circular piston at the end of a rectangular
tank of water . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 305
Example 10.3 (Spherical_cavity_external_pressure.m) Evaluation
of the pressure field in a spherical cavity due to
passage of an exterior plane wave. . . . . . . . . . . . . . . .. 315
Example 10.4 (Concentric_spheres.m) Evaluation of the radially
symmetric pressure field in the region between
concentric spheres when the inner sphere vibrates,
with examination of the limit as the inner sphere
shrinks to a point source . . . . . . . . . . . . . . . . . . . . . .. 320
Example 10.5 (Cylindrical_waveguide_higher_modes_
impedance_tube.m) Determination of the effect
of sidewall compliance on the performance a cylinder
that is used as an impedance tube . . . . . . . . . . . . . . . .. 326
Example 10.6 (Cylindrical_cavity_vibrating_side_wall) Evaluation
of the volume velocity and pressure at the open end
of a cylindrical waveguide when the cylindrical wall
executes a specified vibration . . . . . . . . . . . . . . . . . . .. 332
Example 10.7 (Rect_cavity_modal_density.m) Evaluation of the
natural frequencies of a rectangular tank of water with
a free surface, with special emphasis on repeated
natural frequencies and increasing modal density . . . . .. 344
Example 10.8 (Cavity_impulse_response.m) Analysis using a series
of cavity modes of the time-dependent pressure
radiated by an impulsive line source in a hard-walled
rigid cavity, and discussion of an alternative
formulation using the method of images . . . . . . . . . . .. 347
Example 10.9 (Compare_cyl_and_rect_cavity_modes.m)
Determination of the sizes of a cylindrical tank
and a rectangular tank such that their fundamental
nonplanar mode has a specified natural frequency . . . . .. 355
List of Examples xxxiii
Example 10.10 (none) Analysis of the modal properties of a system

in which an inner sphere is a membrane filled
with a fluid, and surrounded by a concentric sphere
filled with a different fluid . . . . . . . . . . . . . . . . . . . . .. 360
Example 10.11 (none) Usage of the Rayleigh ratio in conjunction
with different trial functions to estimate the
fundamental natural frequency of a two-dimensional
cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 368
Example 10.12 (Rayleigh_Ritz_Ellipse.m) Evaluation of the natural
frequencies of a two-dimensional elliptical cavity
according to the Rayleigh-Ritz method, with emphasis
on convergence properties and agreement with the
separation theorem . . . . . . . . . . . . . . . . . . . . . . . . . .. 374
Example 10.13 (Dowells_approx_2D_cavity.m) Formulation and
solution of differential equations derived from
Dowell’s approximation for a harmonically driven
rectangular two-dimensional cavity whose walls have
properties that change discontinuously. . . . . . . . . . . . .. 388
Example 11.1 (Ray_reflection_spherical_cap.m) Evaluation
of the rays resulting from incidence of a plane wave
on the inner surface of a spherical cap. . . . . . . . . . . . .. 414
Example 11.2 (Ray_single_stratified_fluid.m) Evaluation of a ray
path and propagation time along that ray in a water
channel whose sound speed depends parabolically
on depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 424
Example 11.3 (Ray_reflections_stratified_fluid.m) Computation
of the full set of rays, including those reflected from
the surface, for a water channel whose sound speed
depends parabolically on depth, and usage of the result
to evaluate a focusing factor. . . . . . . . . . . . . . . . . . . .. 434
Example 11.4 (none) Proof that the ray tracing equations for
a channel whose sound speed varies linearly with
depth requires that the rays be circles whose center
is situated at the extrapolated depth where the sound
speed vanishes . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 441
Example 11.5 (Rays_cylindrical_heterog.m) Solution of the ray
tracing equations for the case of a plane wave that
passes transversely through a cylindrically symmetric
heterogeneous region, with special attention to the
caustic that is created, followed by application
of the transport equation to determine the variation
of sound along a wavefront . . . . . . . . . . . . . . . . . . . .. 451
xxxiv List of Examples
Example 11.6 (none) Usage of geometrical acoustics to evaluate

the eigenrays for a point source below the free
surface in a homogeneous channel, and comparison
of the result to that obtained from the method
of images. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 459
Example 11.7 (none) Application of Fermat’s principle to determine
the eigenrays for propagation from a point source in
water channel of constant depth bounded above by air
in a half-space, and below by a liquid sediment
in a half-space . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 469
Example 12.1 (Born_approximation_cylinder.m) Evaluation of the
angular dependence of the pressure scattered by a
cylindrical region of heterogeneity according to the
Born approximation . . . . . . . . . . . . . . . . . . . . . . . . .. 486
Example 12.2 (Scattering_rigid_body_Rayleigh.m) Application
of the Rayleigh approximation of scattering to evaluate
bistatic scattering of a rigid body . . . . . . . . . . . . . . . .. 497
Example 12.3 (Scattering_movable_rigid_disk.m) Evaluation
of low-frequency scattering at arbitrary incidence
on a rigid circular disk that is free to move . . . . . . . . .. 502
Example 12.4 (Submerged_spher_shell_scattering.m) Evaluation
of the backscatter cross section and total scattering
cross section for a spherical shell in water
and in the atmosphere . . . . . . . . . . . . . . . . . . . . . . . .. 527
Example 13.1 (none) Application of a graphical construction method
to evaluate the nonlinear propagation of two plane
wave pulses in the case where the initial waveform
of the second is inverted from that of the first . . . . . . .. 556
Example 13.2 (Nonlin_harmonic_plane_wave.m) Computation of the
nonlinear propagation of the nonlinear plane wave
generated by a transient sinusoidal excitation using
a simple algorithm . . . . . . . . . . . . . . . . . . . . . . . . . .. 560
Example 13.3 (Nonlinear_wave_Newtons_method.m) Application of
Newton’s method to evaluate the Riemann solution for
a nonlinear plane wave when the excitation is the sum
of a fundamental and second harmonic, followed by
FFT analysis to determine the frequency spectrum . . . .. 565
Example 13.4 (Nonlin_harmonic_wave_shock_fitting.m)
Implementation of the equal-area rule to fit shocks
into the Riemann solution for an initially sinusoidal
nonlinear plane wave . . . . . . . . . . . . . . . . . . . . . . . .. 587
List of Examples xxxv
Example 13.5 (Frequency_domain_nonlin_ODEs.m) Solution of the

coupled differential equations governing the nonlinear
propagation of the Fourier series coefficients of an
initially sinusoidal plane wave in the case where the
attenuation coefficient increases as the square of
frequency. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 607
Example 13.6 (none) Formulation of a regular perturbation series
to analyze the nonlinear wave generated by a periodic
sequence of beats in order to identify the growth of
harmonic and intermodulation distortion . . . . . . . . . . .. 607
Example 13.7 (Nonlin_spher_wave_2_harmonics.m) Evaluation
of the waveform of a nonlinear spherical wave
generated by boundary vibration that is the sum of a
fundamental and a third harmonic, and comparison
of the behavior of a nonlinear plane wave generated by
the same excitation . . . . . . . . . . . . . . . . . . . . . . . . . .. 633
Example 13.8 (Nonlin_2D_waveforms.m) Application of Newton’s
method to solve the transcendental equations resulting
from singular perturbation analysis of the
two-dimensional nonlinear plane wave radiated
by a periodically supported plate . . . . . . . . . . . . . . . .. 651
Chapter 7
Radiation from Vibrating Bodies
The previous chapters were devoted to the analysis and interpretation of specific
types of acoustic waves. In this chapter, the central theme is phenomenological. We
will consider a variety of situations in which the task is to determine the acoustic
field generated by a body that executes a specified vibration. These are the problems
in acoustic radiation.
This chapter opens with an investigation of spherical waves that are not radially
symmetric. A description of waves in terms of spherical harmonics, which may
be considered to be the extension of Fourier series to a spherical geometry, will be
applied to treat radiation from a sphere whose surface velocity pattern is arbitrary. At
the same time, it is reasonable to ask how a velocity pattern can be imposed in reality.
An object moves because forces are applied to it. The analysis of the vibration of a
body excited by forces is a fundamental subject unto itself. In an acoustic system, part
of the force system acting on a body is the surface pressure. It follows that to study the
vibration problem, we must determine the radiated field, and that determination of
the radiated field requires that we solve the vibration problem. In other words, the two
problems are coupled. The joint formulation and solution of this coupled problem
are the realm of structural acoustics. The behavior of a spherical shell immersed
in a fluid will be examined in order to shed light on the analytical approach and
phenomena associated with this subject.
To explore the field radiated by an infinitely long vibrating cylinder, we will
develop the general solution of the Helmholtz equation in cylindrical coordinates.
Spheres and infinitely long cylinders are just about the only geometrical configura-
tions that are amenable to analysis. Their solutions shed much light on fundamental
phenomena, but numerical simulation tools are required to address realistic systems.
One such formulation is boundary elements, which is based on a fundamental con-
cept called the Kirchhoff–Helmholtz integral theorem. This theorem, which we shall
abbreviate to KHIT, gives the pressure at a field point in terms of an integral of
the pressure and velocity on the surface of the radiating body. After the theorem is
derived, its implementation as a boundary element formulation will be described.
Two other numerical techniques for analyzing acoustic radiation are source superpo-
sition, which uses a set of sources to replace the vibrating body, and finite elements,
© Springer International Publishing AG 2018 1
J.H. Ginsberg, Acoustics—A Textbook for Engineers and Physicists, Volume II,
DOI 10.1007/978-3-319-56847-8_7
2 7 Radiation from Vibrating Bodies
which requires adaptation to fit the nature of acoustic radiation. These developments
close this chapter.
7.1 Spherical Harmonics
When the surface of a sphere vibrates in an irregular pattern, the radial velocity is
an arbitrary function of the polar and azimuthal angles. Given the difficulties that
arose for a radially symmetric wave in the time domain, it should not be surprising
that we will only pursue a frequency domain analysis. Furthermore, in order to avoid
excessive entanglement in the techniques of mathematical analysis, we will restrict
our attention to axisymmetric situations. The consequence is that the pressure is
p = Re[P(r, ψ) exp(iωt)]. Although this will limit the generality of the situations
we can address, it will not restrict the nature of the physical phenomena that arise.
7.1.1 Separation of Variables
Our objective is to determine the complex pressure amplitude P (r, ψ) that is the
general solution of the Helmholtz equation. The analysis will use the method of
separation of variables. It begins with an ansatz that represents the dependent variable
as a product of functions of each of the independent variables. In other words,
P = F (r ) G (ψ) (7.1.1)
It is helpful to rewrite the Helmholtz equation in axisymmetric spherical coordinates,

Eq. (A.1.16), as

1 ∂2 1 ∂ ∂P
(r P) + sin ψ + k2 P = 0 (7.1.2)
r ∂r 2 r 2 sin ψ ∂ψ ∂ψ
When we substitute the factorized representation of P into this equation, and divide
by FG/r 2 , the result is an equation in which each side is a function of a different
position variable, specifically,

r d2 1 d dG
(r F) + k 2 2
r = − sin ψ (7.1.3)
F dr 2 G sin ψ dψ dψ
The argument now is that the only way a function of r can equal a function of
ψ is if they both equal some constant, which we denote as C. Thus, we obtain two
ordinary differential equations,
7.1 Spherical Harmonics 3

d 2F 2 dF C
+ + k 2
− F =0
dr 2 r dr
r
2
(7.1.4)
1 d dG
sin ψ + CG = 0
sin ψ dψ dψ
These equations are solved independently. We begin with G (ψ), because doing so
will establish the separation constant C.
Actually, we will find that there are an infinite number of C values. The set of
corresponding G functions are referred to as spherical harmonics. The reason for
this terminology stems in part from the oscillatory nature of the functions, and also
from the fact that they are the basis functions for a Fourier-like series describing an
arbitrary function of the spherical angles. Because our interest is in axisymmetric
situations, the spherical harmonics derived here are a subset of the general set of
functions, which depend on θ, as well as ψ.
Dependence on the Polar Angle
The presence of sinusoidal terms in the coefficients of the second of Eq. (7.1.4)
complicates the task of finding a solution. Fortunately, a transformation from ψ to a
nondimensional variable η converts the differential equation to a form that has been
widely studied. The transformation is
1/2
η = cos ψ =⇒ sin ψ = 1 − η 2 (7.1.5)
The derivatives become

dG dG
= − sin ψ
dψ dη
1 d dG d 2G dG (7.1.6)
sin ψ = (sin ψ)2 − 2 cos ψ
sin ψ dψ dψ dη 2 dη
We use the transformation to eliminate ψ from the separated equation, which leads
to Legendre’s equation,
d 2G dG
1 − η2 2
− 2η + CG = 0 (7.1.7)
dη dη
One of the most important aspects of this equation is that the coefficient of its
highest derivative vanishes. Differential equations having this property contain a
solution that is singular. Dealing with this possibility will lead to an expression for
C. The fact that the coefficients in this equation are powers of η suggests that G may
be represented as a power series. Thus, we try
∞

G= ajηj (7.1.8)
j=0
The next step is to substitute this form into Legendre’s equation, and require that it
be satisfied independently for each power of η. The details of these operations may
be found in most textbooks on ordinary differential equations. The result is that the
coefficients are obtained as a recurrence relation that gives a higher order coefficient
in terms of a lower order value. The specific relation is
j ( j + 1) − C
a j+2 = a j , j = 0, 1, 2, ... (7.1.9)
( j + 2) ( j + 1)
The implication of this relation is that a0 and a1 are arbitrary and independent, from
which successive coefficients are found as factors multiplying the a0 and a1 values.
If C is arbitrarily selected, it can be proven that the infinite series for G does not
converge at n = ±1. This is how the singular nature of Eq. (7.1.7) is manifested. The
locations at which |η| = 1 are the poles of the spherical coordinate system, but there
is no physical aspect of the poles that would cause a singular response. The only way
to avoid this behavior and thereby obtain a solution that is valid everywhere, is to
select C such that the series becomes a finite sum. This happens if C makes one of
the a j coefficients to vanish, because the recurrence relation will lead to the higher
coefficients being zero.
If we wish that the mth coefficient be the last one that is nonzero, then we would
select
Cm = m (m + 1) (7.1.10)
Each value of m leads to a different solution of Legendre’s equation because m is the

degree of the polynomial. The solutions are written as Pm (η), which denotes that
they are Legendre polynomials of degree m. Their definition is

floor(m/2)
(−1) j (2m − 2 j)!
Pm (η) = ηm − 2 j (7.1.11)
j=0
2m j! (m − j)! (m − 2 j)!
One should exercise caution whether this formula is used for computations when
m is large, because the factorial terms might be larger than the largest computable
floating point number for the particular computer architecture. The lowest degree
polynomials are
1 2
P0 (η) = 1, P1 (η) = η, P2 (η) = 3η − 1
2 (7.1.12)
1 3
P3 (η) = 5η − 3η
2
The substitution of η = cos ψ leads to Pm (cos ψ), which is a Legendre function of
ψ. The first few functions are
1
P0 (cos ψ) = 1, P1 (cos ψ) = cos ψ, P2 (cos ψ) = (3 cos 2ψ + 1)
4 (7.1.13)
1
P3 (cos ψ) = (5 cos 3ψ + cos ψ)
8
To obtain these forms directly from the Legendre polynomial with η = cos ψ, one
must apply various trigonometric identities for powers of cos ψ. By virtue of their
dependence on cos ψ, the Legendre functions are periodic in ψ = 2π. Conse-
quently, Pm (cos ψ) may be represented by a Fourier series. The handbook edited
by Abramowitz and Stegun1 is an excellent resource for the properties of Legendre
functions.
Legendre polynomials are plotted in the upper set of graphs in Fig. 7.1. The even
degrees are symmetric functions with respect to the origin, while odd degrees corre-
spond to antisymmetric functions. As m increases, Pm (η) oscillates over a smaller
scale, but these oscillations are not harmonic. Recall that the coefficients a0 and a1 in
Eq. (7.1.8) may be selected arbitrarily. The plotted functions have been defined such
that |Pm (±1)| = 1. The lower set of graphs in Fig. 7.1 shows that replacing η with
cos (ψ) to form a Legendre function does not fundamentally alter the appearance of
the function at any degree. However, it does have the effect of compressing the region
near the midpoint, which now is ψ = π/2, and expanding the region near the limits
at η = 0 and π. The transformation from η to ψ leads to (d/dψ) Pm (cos (ψ)) being
identically zero at ψ = 0 and π. It follows that setting G to a Legendre function in the
separation of variables form, Eq. (7.1.1), will lead to d P/dψ being zero identically
1 m=0
1 m=1
m=2
0.5 0.5 m=3
m=6 m=5
Pm(η)
0 0
−0.5 −0.5 m=7

m=4
−1 −1
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
η η
1 m=0
1
m=2 m=1
Pm(cos(ψ))
0.5 m=6 0.5 m=5
0 0
−0.5 m=4 −0.5 m=7 m=3
−1 −1
0 0.5 1 0 0.5 1
ψ/π ψ/π
Fig. 7.1 Low-degree Legendre polynomials as a function of η, and Legendre functions as a function
of ψ
1 M.I. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, Chap. 8 (1965).
at the poles. The continuity equation requires that an axisymmetric field meets this
condition.
The Legendre polynomials have some useful properties. The first are a set of
recurrence relations for functions and derivatives at different degrees,
(m + 1) Pm+1 (η) = (2m + 1) η Pm (η) − m Pm−1 (η)

d d
η Pm (η) − Pm−1 (η) = m Pm (η)
dη dη
d d (7.1.14)
Pm+1 (η) − Pm−1 (η) = (2m + 1) Pm (η)
dη dη
d
1 − η2 Pm (η) = −mη Pm (η) + m Pm−1 (η)
dη
A different definition of the polynomials is offered by Rodrigue’s formula,
1 d m 2 m
Pm (η) = η − 1 (7.1.15)
2m m! dη m
An important property of Legendre polynomials is that they are mutually orthog-

onal. The polynomials are defined over the range −1 ≤ η ≤ 1, so orthogonality
refers to an inner product defined as an integral over that interval. The orthogonality
property states that if m and j are any two indices, then
1 π
2
Pm (η) P j (η) dη = Pm (cos ψ) P j (cos ψ) sin ψdψ = δ jn
−1 0 2m + 1
(7.1.16)
where δ jm is the Kronecker delta, which equals one if j = m and zero otherwise.
Recall that the orthogonality of harmonic functions is one of the primary reasons
that a Fourier series has great utility. Orthogonality has a similar usefulness when
a variable is represented by a Legendre series. In a Fourier series, the harmonic
functions are said to be the basis functions and the series coefficients are “distances,”
in a linear algebra sense, in the direction of each harmonic function. A Legendre
series is defined similarly, with Pm (η) as the basis functions, specifically,
∞
∞

F (η) = Fm Pm (η) or F (ψ) = Fm Pm (cos ψ) (7.1.17)
m=0 m=0
The similarity of this representation to a Fourier series causes some individuals to

say that the preceding is a generalized Fourier series.
Suppose we wish to determine the Fm coefficients associated with a specified F
function. For an acoustics problem, ψ is the independent variable, so we shall work
with the second form in Eq. (7.1.17). Both sides of the equation are multiplied by
P j (cos ψ) sin ψ, where j is a designated index. Then, both sides are integrated over
0 < ψ < π. We bring the integration inside the summation, with the result that
π ∞
π
F (ψ) P j (cos ψ) sin ψdψ = Fm Pm (cos ψ) P j (cos ψ) sin ψdψ
0 m=0 0
(7.1.18)
The orthogonality property states that all integrals vanish, except for the one where
the summation index m equals the selected value of j. In that case, the integral is
2/ (2 j + 1). Thus, the orthogonality property filters from the sum all except the jth
term. The result is that
π
2j + 1
Fj = F (ψ) P j (cos ψ) sin ψdψ (7.1.19)
2 0
One may attempt to evaluate the integral analytically, possibly by replacing P j (cos ψ)
with its definition as a polynomial, or else with the aid of a table of integrals. An
alternative is to use numerical methods to evaluate the integral.
EXAMPLE 7.1 Legendre polynomials have been used to describe crystal

vibrations.2 That analysis used a series of Legendre polynomials to represent
the dependence of displacement u on distance y from the midplane of a rectan-
gular plate. The graph depicts u = 0.2 tanh−1 (1.98y/ h), where h is the plate’s
thickness. Determine the Legendre series coefficients that represent this func-
tion. Identify at what degree the series may be truncated such that its evaluation
will differ everywhere from the actual by less than 2% of the maximum value.
0.5
0.25
y/h
-0.25
-0.5
−0.4 −0.2 0 0.2 0.4
Displacement u
Figure 1.
2 R.D.
Mindlin, “Thickness-Shear and Flexural Vibrations of Crystal Plates,”J. Appl. Phys. 22,
316–323 (1951).
Significance
In addition to showing how we may determine a Legendre series numerically, this
example will give some insight to the convergence properties of this type of series.
Solution
To use Eq. (7.1.19), the range of integration must be −1 ≤ η ≤ 1, but the range for the
plate is −h/2 ≤ y ≤ h/2. Thus, we set 2y/ h = η, so that u (η) = 0.2 tanh−1 (0.99η).
Because η is the dependent variable, we use the first form in Eq. (7.1.16). This entails
multiplying the series description of u by P j (η), then integrating over −1 < η < 1.
The result is 1
2j + 1
Fj = 0.2 tanh−1 (0.99η) P j (η) dη
2 −1
There is no simple way to evaluate this integral analytically, so we turn to numer-

ical methods. The details depend on the software to be used, but the basic elements
are the same. A routine to evaluate the Legendre functions at a set of values of η
is required, as well as a routine that performs numerical integration. If one wishes
to write their own procedures, then Eq. (7.1.11) may be coded, because the degree
of the Legendre polynomial will not be extraordinarily high. Numerical integration
may be carried out by Simpson’s rule, provided one verifies that the sampling interval
is sufficiently small. The present results were obtained with MATLAB, which con-
tains a function routine legendre.m that evaluates Pm (x) for a vector of x values.
However, it cannot be used directly because it also evaluates associated Legendre
functions. They are described in Sect. 7.1.3 as functions that would arise if we were to
investigate spherical waves that are not axisymmetric. The nature of legendre.m
is that it returns an array having m + 1 rows and n columns, where m is the degree
number and n is the number of elements of x. The first row of this array holds
the Pm (x) values, while the other rows hold the corresponding associated func-
tions. Thus, we evaluate the Legendre polynomials by creating a MATLAB function
leg.m that retains only the first row, as follows:
function P_m = leg(x, m); full_set = legendre(m, x); P_m = full_set(1,:)
The integral was evaluated with one of MATLAB’s numerical integration routines,
quadl.m. Implementation of this scheme requires that we define another function to
evaluate the integrand. It is a good idea to verify that everything has been implemented
correctly by following the same procedure to evaluate the integral in the orthogonality
condition.
The first few coefficients obtained from the numerical integration are F0 = 0,
F1 = 0.2869, F2 = 0, F3 = 0.0972, F4 = 0. All coefficients having even subscripts
are zero because u (η) is an odd function with respect to η = 0, and that is the
symmetry of the odd-degree Legendre polynomials. To examine the convergence
of the series, the odd-subscripted Fm are plotted in Fig. 2. It can be seen that the
coefficients decrease rapidly as m increases.
Coefficient Fm
0.1
0.01
0 5 10 15 20
Legendre polynomial order m
Figure 2.
It is requested that truncation of the Legendre series not lead to an error greater
than 2% of the maximum value of u, which is 0.529 at y = 0.5. This requires that
the series be synthesized at the full range of y. An estimate for the series length
is obtained from the observation that the maximum magnitude of Pm (x) is one, so
inclusion of a term FN Pm (x) in the Legendre series can change the series sum by no
more that |Fn |. It is specified that the error should be no more that 2% of 0.529, so
we search for the smallest Legendre degree N at which |FN | < 0.0106. This leads
to N = 13. However, a certain error level in the coefficients might not be manifested
by the same error when the series is synthesized.
In general, a efficient way in which a series may be evaluated at many points is to
write it as a matrix product. Toward that end, we let {y} denote a vector of y values, so
that {η} = (2/ h) {y}. Then,
we form a rectangular array [L] whose element at row j
and column m is Pm η j . The product of row j of [L] and a vector {F} is the same as

{u}
the series evaluation of u y j . Thus, the full set of values is found

by evaluating
[L] {F}. Figure 3 displays the error, defined as
u y j − u series y j
/u (y = h/2)
for N = 13. The convergence criterion is met away from y = ±h/2, but it is 4% at
the limits. A larger N is required to satisfy it everywhere. Successive evaluations at
increasing N revealed that N = 19 is the smallest degree at which truncation of the
series will not lead to an error exceeding 2% at any y.
0.5
0.25
y/h
-0.25
-0.5
10-5 10-4 10-3 10-2 10-1
Error, |u - useries|/u
Figure 3.
Radial Dependence
The separation of variables constant C depends on the degree m of the Legendre
polynomial, as given by Eq. (7.1.10). Thus, each m leads to a different radial function.
Each is governed by the first of Eq. (7.1.4), which now is

d 2F 2 dF m (m + 1)
+ + k −
2
F =0 (7.1.20)
dr 2 r dr r2
At this juncture, there are a number of ways by which F may be determined.

One approach is to implement the method of Frobenius, which is a standard tool
for solving ordinary differential equations whose coefficients are not constant. The
ansatz for this method is a power series multiplied by a factor that accounts for
singular behavior. Another approach is suggested by the fact that far from a radiating
body, spherical spreading requires that the solution decays reciprocally with r . We
can anticipate this effect by changing the dependent variable according to
F = (r ) /r (7.1.21)
The differential equation becomes

d 2 m (m + 1)
+ k −
2
=0 (7.1.22)
dr 2 r2
This is promising because it tells us that tends to a harmonic function at larger r.

Nevertheless, this equation does not appear to be easier to solve than the equation for
F. Our knowledge of acoustics principles suggests how to proceed. When m = 0,
Eq. (7.1.20) reduces to the Helmholtz equation in the case of a radially symmetric
field. Thus, for m = 0, we know that F = Be±ikr /r . Furthermore, regardless of m,
we know that the farfield pressure is a spherically spreading wave with an angular
directivity. The observations suggest that we change the dependent variable such that
1
F= Fm (r ) e−ikr (7.1.23)
r
Determination of the function Fm (r ) is expedited by writing the differential equation
as
1 d2 m (m + 1)
(r F) + k − 2
(r F) = 0 (7.1.24)
r dr 2 r2
When the representation of F is substituted into this equation, the complex expo-
nential cancels. What remains is an ordinary differential equation for Fm ,
d 2 Fm d Fm m (m + 1)
− 2i − Fm = 0 (7.1.25)
d (kr )2 d (kr ) (kr )2
This might not seem to be any more conducive to an analytical solution than the
original equation governing F, but it is. In the farfield, Fm must approach a constant
value, which suggests a series representation in powers of 1/ (kr ), that is,
∞
aj
Fm = (7.1.26)
j=0
(kr ) j
The coefficients a j are determined by substituting the series into Eq. (7.1.25), fol-
lowed by equating the collected coefficients of each power of 1/ (kr ) to zero. These
operations lead to a recurrence relation for a j+1 as a proportionality to a j , starting
with j = 0. Examination of this relation shows that it gives a j = 0 if j > m. In other
words, although we allowed for an infinite length series, the number of terms must
be one greater than the index m for the separation constant C. The solution for F
corresponding to each Fm is the second kind of spherical Hankel function of order
m. A detailed analysis would lead to
j
e−ikr (m + j)!
m
1
F = h (2)
m (kr ) = i
m+1
(7.1.27)
kr j=0 j! (m − j)! 2ikr
In this form, the starting coefficient a0 has been set to one. The lowest order functions
are
(2) ie−ikr (2) e−ikr i
h 0 (kr ) = , h 1 (kr ) = − 1−
kr kr kr
−ikr (7.1.28)
(2) ie 3i 3
h 2 (kr ) = − 1− −
kr kr (kr )2
At this juncture, the logical question is: Why is this function called the “second
kind”? We found it first, so why is it not the “first kind”? The answer lies in the fact
that we have used eiωt to represent a time function. The function associated with the
negative exponential temporal representation by convention is denoted as h (1) m (kr ),
which is the first kind of spherical Hankel function of order m. Because the proper
representation of a harmonic function adds e−iωt and eiωt terms, the coefficients of
the exponential must be complex conjugates in order that the time dependence be
real. Hence, it must be that
∗
h (1) (2)
m (kr ) = h m (kr ) (7.1.29)
This definition of another solution could have been anticipated from the fact that the
differential equation whose solution we sought, Eq. (7.1.20), is real. If a complex
function satisfies a linear differential equation whose coefficients are real, then its
complex conjugate must do so as well because the real and imaginary parts of the
differential equation must be satisfied independently. The same reasoning leads to the
recognition that the real and imaginary parts of the spherical Hankel functions must
satisfy the equation independently. The real part is the spherical Bessel function of
order m, denoted jm (kr ), and the imaginary part is the spherical Neumann function
of order m, denoted n m (kr ),
(2)
jm (kr ) = Re h (2)
m (kr ) , n m (kr ) = − Im h m (kr ) (7.1.30)
A different way of viewing these definitions is that
h (2) (1)
m (kr ) = jm (kr ) − in m (kr ) , h m (kr ) = jm (kr ) + in m (kr )
1 (1) 1 (1)
jm (kr ) = h (kr ) + h (2)
m (kr ) , n m (kr ) = h (kr ) − h (2)m (kr )
2 m 2i m
(7.1.31)
The lowest order functions are

sin (kr ) cos (kr )
j0 (kr ) = n 0 (kr ) = −
kr kr
sin (kr ) cos (kr ) cos (kr ) sin (kr )
j1 (kr ) = − n 1 (kr ) = − −
(kr )2 kr (kr )2 kr

3 1 3 1
j2 (kr ) = − sin (kr ) n 2 (kr ) = − − cos (kr )
(kr )3 kr (kr )3 kr
3 3
− cos (kr ) − sin (kr ) (7.1.32)
(kr )2 (kr )2
When we wish to refer to these solutions generically, we will say that they are
spherical Bessel functions.
Most mathematical software provides routines to evaluate Bessel functions Jν (x)
and Neumann functions Nν (x).3 Bessel’s equation arises in problems featuring a
planar circular geometry. For this reason, the functions Jν (x) and Nν (x) are said to
be a set of cylindrical Bessel functions. The cylindrical and spherical functions are
related by
π 1/2 π 1/2
jm (x) = Jm+1/2 (x) , n m (x) = Nm+1/2 (x) (7.1.33)
2x 2x
In the event that the available software only evaluates cylindrical Bessel function,
we would evaluate the cylindrical functions for order m + 1/2 and argument kr and
then multiply the result by π 1/2 / (2kr )1/2 . The spherical Hankel functions may be
found from Eq. (7.1.31).
The low-order spherical Bessel and Neumann functions are graphed in Fig. 7.2.
The abscissa for the Neumann functions is split because that function is singular at
is standard practice to denote a Bessel function as Jν (x), but a Neumann function often is
3 It
denoted as Yν (x).
1
m=0
0.5 m=1
jm(x) m=2
0 m=3
−0.5
0 2 4 6 8 10
1
0 m=0 1
m=1
−1 0.5
nm(x)
−2 m=2 m=2
−3 m=3
m=1 0
m=3 m=0
−4
−0.5
−5
0 2 4 6 8 10
Fig. 7.2 Dependence of the spherical Bessel and Neumann functions on the value of its order m
and argument x
the origin. The behavior for small x of n m (x) is quite different from that of jm (x),
but both become oscillatory with decreasing amplitude as x increases.
These observations are borne out by the asymptotic approximations for small and
large arguments. These behaviors for all functions may be extracted by the analysis
of Eq. (7.1.27), combined with application of Eqs. (7.1.29) and (7.1.30). When the
argument is very small, x 1, the results are
⎫
∗ (2m)! 1 m+1 ⎪
⎪
h (2) (1)
m (x) = h m (x) → i ⎪
⎪
m!2m x ⎪
⎪
m! (2x) m ⎬
jm (x) → as x → 0 (7.1.34)
(2m + 1)!
m+1 ⎪
⎪
⎪
⎪
(2m)! 1 ⎪
⎪
n m (x) → − ⎭
m!2m x
The trends for large arguments are
⎫
∗ 1 −i(x−(m+1)π/2) ⎪
h (2) (1)
m (x) = h m (x) → e ⎪
⎪
x ⎪
⎪
1 m+1 ⎬
jm (x) → cos x − π as x → ∞ (7.1.35)
x 2 ⎪
⎪
1 m+1 ⎪
⎪
n m (x) → sin x − π ⎪
⎭
x 2
An important aspect of the preceding is that they require that the argument x be larger
than the order m. Another asymptotic representation covers the situation where the
order is large,
1/2 ⎫
∗ 2 2m + 1 (m+1/2) ⎪
⎪
h (2) (x) =h (1) (x) → i ⎪
⎪
m m
(2m + 1) x ex ⎪
⎪
(m+1/2) ⎪
⎪
π 1/2 ⎬
1 ex as m → ∞
jm (x) →
2x (π (2m + 1)) 1/2 2m + 1 ⎪ with x fixed
⎪
1/2 (m+1/2) ⎪
⎪
⎪
⎪
2 2m + 1 ⎪
⎪
n m (x) → − ⎭
(2m + 1) x ex
(7.1.36)
The qualitative description of these relations is that only the spherical Bessel
function jm (x) is finite at zero argument, with all except the zero order approaching
zero. Each of the other functions is singular at x = 0, with an increasing growth rate
as the order is increased. Also, replacing x with kr shows that the farfield behav-
ior of h (2) (1)
m (kr ) matches that of a wave that propagates outward, whereas h m (kr )
is associated with inward propagation. Furthermore, jm (kr ) and n m (kr ) represent
standing wave patterns. The behavior at large orders is important because it affects
the convergence of series that sums over all orders.
There are many formulas that relate the various spherical Bessel functions and
their derivatives. A group that are particularly useful are the recurrence relations.
The following uses f m (x) to denote jm (x), n m (x), h (1) (2)
m (x), or h m (x),
2m + 1
f m−1 (x) + f m−1 (x) = f m (x) , m > 0
x
d m
f m (x) = f m (x) − f m+1 (x)
dx x
m+1
= f m−1 (x) − f m (x) , m > 0
x
1 (7.1.37)
= m f m−1 (x) − (m + 1) f m+1 (x) , m > 0
2m + 1

1 d n m+1
x f m (x) = x m−n+1 f m−n (x)
x dx

1 d n −m
x f m (x) = (−1)n x −m−n f m+n (x)
x dx
The first equation may be used to evaluate higher order functions in terms of lower
order values. We will have frequent need for derivatives of Bessel functions, so it is
good idea to create a subprogram that implements the first equation for d f m (x) /d x.
The last two relations seldom come into play in the analysis of a specific system. As
is true for Legendre functions, the handbook compiled by Abramowitz and Stegun4
provides a compact resource for many other formulas. The book by Sneddon5 is a
reference for those seeking detailed analyses of Bessel functions.
4 M.I. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, Chaps. 9–11
(1965).
5 I.A. Sneddon, Special Functions of Mathematical Physics and Chemistry Oliver & Boyd (1961).
7.1.2 Description of the Pressure Field
The separation of variables analysis began by seeking the conditions for which P =
F (r ) G (ψ) is a solution of the Helmholtz equation. This led to the recognition that
G (ψ) must be a Legendre function of integer order m and F (r ) must be one of the
types of spherical Bessel functions. Because the Helmholtz equation is homogeneous,
the trial solution may be multiplied by a constant. Furthermore, the equation is linear,
so a sum of solutions at various orders also is a solution.
Which type of spherical Bessel function we should use depends on the extent
of the fluid’s domain. We began with the stated interest of describing radiation into
an infinite domain. The Sommerfeld radiation condition requires an outgoing wave,
which means that we should use whichever function behaves as e−ikr /r when r
is large. Reference to Eq. (7.1.35) shows that this function is h (2) (kr ). Thus, the
axisymmetric pressure exterior to a sphere is described by
∞

P= Bm h (2)
m (kr ) Pm (cos ψ), r > a (7.1.38)
m=0
To specialize this expression to the farfield, we use the asymptotic representation of

h (2)
m (kr ) contained in Eq. (7.1.35), which leads to
∞
e−ikr i(m+1)π/2
Pff = Bm Pm (cos ψ) e (7.1.39)
m=0
kr
The pressure in a spherical cavity is a topic in Chap. 10. As a preliminary to

that study, we recall that the analysis of the radially symmetric field in Sect. 6.3.3
began with a solution that contains inward and outwardly propagating waves. The
corresponding form here would use both kinds of spherical Hankel function. The
finiteness condition in the previous analysis led to the recognition that these waves
must have complex amplitudes that sum to zero. A similar analysis would lead to the
same conclusion here. According to Eq. (7.1.31), the sum of h (2) (1)
m (kr ) and h m (kr )
gives a spherical Bessel function, while the difference gives a spherical Neumann
function. The latter is singular at r = 0, so the general solution for the axisymmetric
pressure within a spherical cavity is
∞

P= Bm jm (kr ) Pff (cos ψ), 0 ≤ r < a (7.1.40)
m=0
7.1.3 Arbitrary Spatial Dependence
The most general situation is a pressure field that is neither radially nor axially
symmetric. For example, a single dipole and a longitudinal quadrupole generate
fields that are axisymmetric, but a lateral quadrupole is not. An investigation of the
most general spherical wave would pursue a separation of variables analysis of the
Helmholtz equation by starting with
P = F (r ) G (ψ) (θ) (7.1.41)
Such an analysis would lead to the recognition that the azimuthal function (θ)
must be harmonic. This fact is apparent on a physical basis, because circling the
polar axis at a fixed radial distance and polar angle must return us to the point where
we began. Thus, it must be that P (r, ψ, θ) = P (r, ψ, θ + 2π). This is the condition
for representing the azimuthal dependence in a Fourier series. At our option, we can
take to be a complex exponential, einθ , where n is any positive or negative integer,
or we may consider it to be a real harmonic by letting it be sin (nθ) or cos (nθ).
Regardless of which representation is selected, the fact that (θ) is a harmonic
function whose argument is nθ reduces the task to a determination of the product
F (r ) G (ψ). Application of the separation of variables technique to the Helmholtz
equation leads to the recognition that the azimuthal harmonic number n affects the
equation for G, but not the one for F. The solution for G is found to be an associated
Legendre function Pmn (cos ψ), which is obtained from n differentiations of an mth
degree Legendre polynomial, followed by the substitution of η = cos ψ. The specific
definition is
m d n
Pm (cos ψ) = − 1 − η
n 2 1/2
n
Pm (η)
(7.1.42)
dη η=cos ψ
Because Pm (η) is an mth degree polynomial, any derivative above the mth degree
will be identically zero, from which it follows that
Pmn (cos ψ) = 0 if n > m (7.1.43)
The index n is said to be the order of the associated Legendre function and m is its
degree.
Because the separation equation for F (r ) does not depend on the azimuthal
number n, it is the same alternative set of Bessel functions as in the axisymmetric
case. Thus, for each polar number m, there are azimuthal harmonics from 0 to m.
The pressure for each m, n pair consists of a product of a spherical Bessel or Hankel
function of order n for the radial dependence, a Pmn (cos ψ) function, and a harmonic
function of nθ. The range of m is infinite, so the general solution for the pressure is
a double sum. For example, in the case of radiation, with a complex Fourier series
used to represent the azimuthal dependence, the solution would have the form
∞
1
m
P= Bm,n h m (kr ) Pmn (cos ψ) einθ (7.1.44)
2 m=0n=−m
The factors Pmn (cos ψ) einθ describe an arbitrary angular dependence; they constitute
the full set of spherical harmonics.
Like the axisymmetric set, the general set of spherical harmonics are orthogo-
nal when integrated over a sphere. Indeed, procedures that are used to analyze an
axisymmetric field in terms of spherical harmonics are readily modified to handle
arbitrary fields. Several books6 treat the mathematical aspects of a general spherical
harmonic series expansion and its application to acoustical systems. If we were to
study situations where there is no symmetry, we would not encounter much that is
new from a phenomenological viewpoint. The primary reason for this is the afore-
mentioned nature of the radial dependence, which is independent of the azimuthal
harmonic number.
EXAMPLE 7.2 Consider an axisymmetric field consisting of a monopole,

a dipole, and a longitudinal quadrupole. This combination of sources is col-
located, and the axes of the dipole and quadrupole are aligned. What is the
spherical harmonic representation of the farfield?
Significance
Experience in working with spherical harmonics is the primary emphasis. The result
will highlight the different pictures of the farfield that results from a spherical har-
monic series, rather than a multipole expansion.
Solution
The multipole representation of the farfield pressure is provided by Eq. (6.5.142).
To use it, we will let z ≡ x3 be the axis of symmetry, and set the dipole such that
its moment vector is aligned along z. A longitudinal quadrupole that is aligned in
this direction has Q 3,3 as the only nonzero component. The spherical coordinate
transformation gives z = r cos ψ, so the given pressure is
e−ikr
P = A + ik D3 cos ψ − k 2 Q 3,3 (cos ψ)2 (1)
r
where the strength factors are unspecified.
The spherical harmonic description of an axisymmetric field is Eq. (7.1.38), and
Eq. (7.1.39) is the corresponding farfield approximation. The Bm coefficients are
found by applying the orthogonality property. It is somewhat easier to formulate the
6 J.O.Hirschfelder, C.F. Curtiss, and R.B. Bird, Molecular Theory of Gases and Liquids, Wiley-
Interscience (1964).
analysis in terms of η = cos ψ. The actual and series descriptions of P are equated,
and both sides are multiplied by P j (η). Integration over −1 ≤ η ≤ 1 leads to
∞

1
e−ikr i(m+1)π/2
Bm Pm (η) e P j (η) dη
−1 kr
m=0 1
e−ikr
= A + ik D3 η − k 2 Q 3,3 η 2 P j (η) dη
−1 r
All terms in the summation integrate to zero, except for the term for which m = j.
The result is
1
2j + 1
Bj = ke−i( j+1)π/2 A + ik D3 η − k 2 Q 3,3 η 2 P j (η) dη (2)
2 −1
We could evaluate the integral numerically as we did in the previous example.

However, the facts that P j (η) is a polynomial in η, and that it is multiplied by pow-
ers of η, suggest we try analytical integration. One approach would be to work
with the series representation of Legendre functions, Eq. (7.1.11), but a simpler
analysis is available. Equation (7.1.12) indicates that P0 (η) = 1, P1 (η) = η, and
[2P2 (η) + P0 (η)] /3 = η 2 , so Eq. (2) may be written as
1
2j + 1 −i( j+1)π/2
Bj = ke A P0 (η) + ik D3 P1 (η)
2 −1

2P2 (η) + P0 (η)
− k 2 Q 3,3 P j (η)dη
3
The orthogonality property describes the integral of each term. Specifically, Pn (η)
P j (η) integrates to zero unless j = n, in which case the integral is 2/ (2n + 1). Thus,
the result is that
B0 = −ik A + ik 3 Q 33 , B1 = −ik 2 D3 , B2 = − (2/3) i Q 3,3
The most notable attribute is that the spherical harmonics are not in one-to-one
correspondence to the point source distributions. As the source distribution goes
up in order, more harmonics are required to represent it. In general, the field of an
axisymmetric n-pole will be proportional to cos ψ raised to powers up to n. The
resulting B j coefficients for j ≤ n will contain a contribution from that pole if j and
n have the same parity.
Given that several spherical harmonics are required to describe the higher mul-
tipoles, the utility of spherical harmonics might seem to be questionable. This is a
limited view. The multipole construction requires that the sources be a compact set,
whereas spherical harmonics may be used to describe any spherical field. Further-
more, the orthogonality property enables a complicated field to be mapped into a
simpler representation as a series.
7.2 Radiation from a Spherical Body 19
7.2 Radiation from a Spherical Body
7.2.1 Analysis
Determination of the field radiated by an axisymmetrically vibrating sphere is our

first application of a spherical harmonic series. The axis of symmetry is designated
as the polar axis z. The radial vibration of the sphere is an arbitrary function of the
polar angle, V (ψ). The sphere, whose radius is a, is surrounded by an ideal fluid.
Cases in which the sphere executes a breathing mode vibration, or oscillates as a rigid
body, which we studied in Chap. 6, are special cases. Generalization to an arbitrary
vibration pattern will lead to insights about what aspects of the vibration pattern are
important to the pressure field and radiated power.
In addition to the spherical harmonic series for the pressure in a radiation problem,
Eq. (7.1.38), we will require the associated radial velocity. It is obtained by applying
Euler’s equation to the pressure field. These expressions are
∞

P= Bm h m (kr ) Pm (cos ψ)
m=0
∞ (7.2.1)
i
Vr = Bm h m (kr ) Pm (cos ψ)
ρ0 c m=0
Explicit designation of the Hankel function as being the second kind has been aban-
doned because that is the only kind that is relevant to the analysis. (If we had adopted
the exp(−iωt) convention for a harmonic function, it would use the first kind of Han-
kel function.) Also, the prime designates differentiation of the function with respect
to its argument, that is,
d (2)
h m (kr ) ≡ h (ξ)
(7.2.2)
dξ m ξ=kr
The preceding is descriptive of any axisymmetric field. To fit it to the present

radiation problem, we match the radial particle velocity at the sphere’s surface to
that of the sphere. Toward that end, the surface velocity is represented as a spherical
harmonic series,
∞
V (ψ) = v0 Vm Pm (cos ψ) (7.2.3)
m=0
The parameter v0 is a measure of the magnitude of the surface velocity, such as its
maximum. It has been factored out of the series in order that the Vm coefficients
be dimensionless. An expression for these coefficients is found by invoking the
orthogonality property of the spherical harmonics, which gives
π
1 V (ψ)
Vm = m + Pm (cos ψ) sin ψdψ
2 0 v0
1 −1 (7.2.4)
1 V cos (η)
= m+ Pm (η) dη
2 −1 v0
Which of these forms is more conducive to the evaluation of the integral depends on
the nature of the V (ψ) function.
The continuity condition requires that Vr = V (ψ) at r = a. We perform this
matching by using Eq. (7.2.3) to represent the sphere’s motion and Eq. (7.2.1) to
represent the response of the fluid. Thus, we have
∞ ∞
i

Bm h m (ka) Pm (cos ψ) = v0 Vm Pm (cos ψ) (7.2.5)
ρ0 c m=0 m=0
The Legendre functions constitute an orthogonal set, so the preceding relation must
be satisfied separately by each term. This leads to
Vm
Bm = ρ0 cv0 (7.2.6)
i h m (ka)
The resulting expressions for particle velocity and pressure are
∞
h m (kr )
P = ρ0 cv0 Vm Pm (cos ψ)
m=0
i h m (ka)
∞
(7.2.7)
h (kr )
Vr = v0 Vm m Pm (cos ψ)
m=0
h m (ka)
Although these expressions call for infinite summations, in most cases, adequate
accuracy will be obtained when the summations are truncated at a relatively small
number of terms.
It is useful to verify that these expressions are consistent with prior developments.
In the case of a sphere that executes a breathing mode vibration, the surface velocity
V (ψ) = v0 . Because P0 (cos ψ) = 1, the only nonzero velocity coefficient is V0 = 1.
From Eq. (7.1.28), we have
ie−i x x − i −i x
h 0 (x) = =⇒ h 0 (x) = e (7.2.8)
x x2
These expressions, evaluated at x = kr for the former and x = ka for the latter, are
substituted into the m = 0 term in the summation, which yields
ie−ikr /r
P = ρ0 cv0 (7.2.9)
ka − i −ika
i e
(ka)2
When this expression is simplified, the result is the same as the first of Eq. (6.3.7).
Proof that the spherical harmonic description reduces to the previous result for
a translating sphere is a little more complicated. In this case, V (ψ) = v0 cos ψ,
so V1 = 1 is the sole nonzero coefficient. According to Eq. (7.1.28), the first-order
Hankel function and its derivative are
2
−x + i −i x i x + 2x − 2i
h 1 (x) = e =⇒ h 1 (x) = e−i x (7.2.10)
x2 x3
Substitution of these terms into the m = 1 term in the summation gives

−kr + i −ikr
e
(kr )2
P = ρ0 cv0 (7.2.11)
i (ka)2 + 2ka − 2i
i
(ka)3
Simplification of this expression leads to Eq. (6.5.102).
7.2.2 Important Limits
Not much about the nature of the field is apparent in the general solution, so we
turn to asymptotic trends, beginning with the farfield. When kr is large, Eq. (7.1.35)
indicates that
e−ikr i(m+1)π/2
h m (kr ) → e , h m (kr ) → −i h m (kr ) as kr → ∞ (7.2.12)
kr
Application of these trends to Eq. (7.2.7) leads to
∞
e−ikr eimπ/2
Pff = ρ0 c (Vr )ff = ρ0 cv0 Vm Pm (cos ψ) (7.2.13)
kr m=0 h m (ka)
Typically, the farfield property of interest is the directivity. Previously, we defined

D (ψ) such that its maximum value is one. It is more convenient here to use as the
reference pressure ρ0 cv0 (a/r ). Only the pressure magnitude is considered. Thus,
we have

Pff
r 1
Vm eimπ/2
D (ψ) ≡

a =
Pm (cos ψ)
(7.2.14)
ρ cv 0 0 ka
m h m (ka)
Another use of the farfield pressure is to evaluate the radiated power. Toward that
end, we form the time-averaged intensity at large kr , which is (Ir )av = Pff Pff∗ / (2ρ0 c).
To evaluate the product, we change the summation index for one factor, which leads
to
2 ∞
∞
1 1 ∗ ei( j−m)π/2
(Ir )av = ρ0 cv02 V j Vm ∗ P j (cos ψ) Pm (cos ψ)
2 kr j=0 m=0
h j (ka) h m (ka)
(7.2.15)
The radiated power is found by integrating this expression over the surface of a very
large sphere whose radius is r . Because of the axisymmetry of the field, a suitable
differential surface element is dS = 2πr sin ψ (r dψ). When we bring the surface
integral inside the double sum, we find that
∞ ∞
πρ0 c ∗ ei( j−m)π/2 π
P= V j Vm ∗ P j (cos ψ) Pm (cos ψ) sin ψdψ
k 2 j=0 m=0 h j (ka) h m (ka) 0
(7.2.16)
The integral is merely the orthogonality condition for Legendre functions, which
means that all terms in the double sum for which j
= m vanish. This simplifies the
result to
∞
1 2 |Vm |2
P = πρ0 cv02 a 2

(7.2.17)
(ka)2 m=0 2m + 1
h (ka)
2 m
This expression exhibits an important property: The spherical harmonics of the

surface velocity are uncoupled in the radiated power. In other words, each harmonic
gives a fixed contribution to the radiated power, independently of the magnitude of
the other harmonics.
Another important feature is related to convergence properties. When the har-
monic number m becomes increasingly large, h m (ka) will increase rapidly, see
Eq. (7.1.36). Consequently, the high-order terms in the series for the radiated power
have much less importance than they do for the surface velocity. Recall that increas-
ing spherical harmonic number represents more “wiggles” in the spatial fluctuation.
It follows that this behavior is another manifestation of the general notion that waves
on a surface that have a short wavelength tend to not radiate to the farfield.
Another limit of interest is a small sphere. The expressions for pressure and radial
velocity depend on the radius nondimensionally as ka, so the small radius limit also

is the low-frequency limit. The asymptotic behavior of h m (ka) may be found by
differentiating h (2)
m (x) in Eq. (7.1.34), which gives
(m + 1) (2m)!
h m (ka) → −4i as ka → 0 (7.2.18)
(2ka)m+2
With these, the pressure in Eq. (7.2.7) becomes

∞
(2ka)m+2
P = ρ0 cv0 Vm h m (kr ) Pm (cos ψ) , ka 1 (7.2.19)
m=0
4 (m + 1) (2m)!
Before we interpret this expression, let us consider the special case of a radially
symmetric vibration. The only nonzero velocity coefficient is V0 = V (ψ) /v0 = 1.
The definitions are h 0 (kr ) ≡ ie−ikr / (kr ) and P0 (cos ψ) = 1, and we also have

d ie−iξ
ka − i
h 0 (ka) ≡
≡ e−ika , (7.2.20)
dξ ξ ξ=ka k 2a2
When V0 is the only nonzero coefficient, Eq. (7.2.9) for arbitrary values of kr and
ka gives
h 0 (kr ) ka a −ik(r −a)
P = ρ0 cv0 = ρ0 cv0 e (7.2.21)
i h 0 (ka) ka − i r
Now consider Eq. (7.2.19) when ka 1. Each term in the summation raises ka
to a higher power, so it follows that the dominant term corresponds to the lowest
harmonic m for which Vm is nonzero. Usually, this corresponds to m = 0, in which
case, this description reduces to
a
P = ρ0 cv0 (ka)2 h 0 (kr ) P0 (cos ψ) ≡ ρ0 cv0 (ika) e−ikr (7.2.22)
r
Because eika ≈ 1 when ka 1, Eq. (7.2.21) reduces to this expression. Hence, we

conclude that in the small sphere/low-frequency limit, the only aspect of the surface
motion that is important is the volume velocity, Q̂ = 4πa 2 v0 . Only if the volume
velocity is zero does the pattern of the surface vibration affect the acoustic field.
The high-frequency/large sphere limit also leads to an interesting observation.
Let M denote the lowest spherical harmonic at which the summation in Eq. (7.2.3)
may be truncated without reducing the accuracy below the desired value. The def-

inition of high frequency is that ka M. In that case, an expression for h m (ka)
at all significant orders m may be obtained by differentiation of the large argument
approximation of h (2)
m (ka), Eq. (7.1.35), which gives
−i −i(ka−(m+1)π/2)
h m (ka) → e as ka → ∞ (7.2.23)
ka
This approximation converts Eq. (7.2.13) to
a M
Pff → ρ0 cv0 e−ik(r −a) Vm Pm (cos ψ) (7.2.24)
r m=0
The sum is the truncated representation of V (ψ) /v0 , so the corresponding directivity
is D = |V (ψ)| /v0 . In other words, the directivity replicates the pattern of the sphere’s
radial velocity pattern.
This limiting behavior is readily explained. Because Pm (η) is an mth degree
polynomial, it follows that Pm (cos ψ) may be represented as a Fourier series in
ψ whose highest harmonic is cos (mψ). The wavelength of a surface vibration on
the sphere at this harmonic is the circumference divided by m, λm = 2πa/m. The
shortest wavelength that is significant to the vibration is λ M . The acoustic wavelength
is 2π/k. Thus, defining high frequency to be ka M is equivalent to stating that the
high-frequency regime is such that all significant λm are much greater than 2π/k. In
other words, spherical harmonics whose surface wavelength is much larger than the
acoustic wavelength are responsible for most of the radiation to the farfield. This is
reminiscent of the behavior that was observed for surface waves on an infinite planar
surface, see Sect. 5.1.
EXAMPLE 7.3 A spherical cap is the region subtending an angle ψ0 relative

to a radial line that is designated as the z-axis. This cap executes a translational
vibration Re v0 eiωt parallel to z-axis, whereas the remainder of the sphere is
rigid and stationary. The corresponding complex amplitude of the radial veloc-
ity on the surface r = a is Vr = vcap ēz · ēr = v0 cos ψ if 0 ≤ ψ ≤ ψ0 , Vr = 0
if ψ0 < ψ ≤ π. Derive expressions for the coefficients Vm of the spherical har-
◦
monic series. Then, evaluate these coefficients for the case where ψ0 = 30 ,
and identify the harmonic number N for which a reconstruction of Vr deviates
by no more than 0.2v0 from the actual Vr . Use this value of N to evaluate
the pressure along the z-axis, ψ = 0, for ka = 12. Use that result to estimate
the value of kr at which the farfield begins. Then, use the same value of N to
evaluate the farfield directivity, and compare that result to the high-frequency
asymptotic trend.
Significance
A simplified version of this problem has appeared in many books. The simplification
entails considering the velocity of the spherical cap to be radial, rather than being
proportional to cos ψ. The availability of mathematical software, coupled with a
little ingenuity regarding the application of the recurrence relations for Legendre
polynomials, makes it possible to carry out a consistent analysis. The outcome will
show the utility of some of the asymptotic properties of the pressure field.
Solution
We begin by evaluating the Vm coefficients, which are given by Eq. (7.2.4). The
function of interest here is zero beyond ψ = ψ0 , so we have
ψ0
1
Vm = m + cos ψ Pm (cos ψ) sin ψdψ
2 0
1 (1)
1
= m+ η Pm (η) dη
2 cos ψ0
The question now is how to evaluate the integral? We could use a numerical inte-
gration algorithm. If we were to follow that route, we would encounter difficulty
because a rather large number of terms are required to meet the imposed conver-
gence criterion, and Pm (η) oscillates rapidly when m is large. An alternative is to
replace Pm (η) with its series representation. Here too, large m would lead to diffi-
culty because the factorials in the numerator and denominator of the series for Pm (η)
would be very large. Our approach relies on the recurrence relations. We use the first
of Eq. (7.1.14) to replace η Pm (η) with term that are purely Legendre polynomials.
To that result, we will apply the third of Eq. (7.1.14) in order to replace those polyno-
mials with derivatives of Legendre polynomials. These derivatives may be integrated
directly. The operations are
1
1 (m + 1) Pm+1 (η) + m Pm−1 (η)
Vm = m + dη
2 cos ψ0 2m + 1

1 1 m+1 d d
= Pm+2 (η) − Pm (η)
2 cos ψ0 2m + 3 dη dη

m d d
+ Pm (η) − Pm−2 (η) dη
2m − 1 dη dη
All Legendre polynomials are unity at η = 1, so we find that

1 m+1 (2m + 1)
Vm = − Pm+2 (cos ψ0 ) + Pm (cos ψ0 )
2 2m + 3 (2m + 3) (2m − 1) (2)
m
− Pm−2 (cos ψ0 )
2m − 1
These relations are valid for m ≥ 2. The first two coefficients may be found directly
by substituting P0 (η) ≡ 1 and P1 (η) ≡ η. The result is
1 1
V0 = 1 − (cos ψ0 )2 , V1 = 1 − (cos ψ0 )3 (3)
4 2
For comparison, the simplified version, in which Vr = v0 , would not have the η factor
in the integrand. The third of the recurrence relations in Eq. (7.1.14) would suffice
in that case, which would lead to
1
1
(Vm )simp = m+ Pm (η) dη
2 cos ψ0
1 (4)
= − Pm+1 (cos ψ0 ) − Pm−1 (cos ψ0 ) , m ≥ 1
2
1
(V0 )simp = [1 − (cos ψ0 )]
2
The consistent and simplified Vm coefficients are described by Fig. 1. The two
sets of values are quite close. This is a consequence of the fact that for 0 ≤ ψ ≤ 30◦ ,
the value of cos ψ is close to unity. The discrepancy would grow if ψ0 were larger.
Consistent
0.2 Simplified
Vm
0
−0.2
0 20 40 60 80 100
Polynomial degree m
Figure 1.
The Vm values decrease with increasing m, but slowly and in an oscillatory manner.
As we add more terms to the series for V (ψ), sometimes the additional terms will
add, and sometimes they will subtract. Consequently, examination of the magnitude
of the coefficients gives an uncertain picture of the significance of each term in the
series. We must assess convergence by constructing V (ψ) according to Eq. (7.2.3).
Let V (ψ, N ) denote the value obtained when the series is truncated at m = N . The
error metric of interest is |V (ψ, N ) − cos (ψ) [h (ψ) − h (ψ − ψ0 )]|. The following
table lists the maximum error for 0 ≤ ψ ≤ π for several N .
N 20 40 60 80 100
Error (%) 37 32 25 21 16
The error at N = 80 is very close to the requested value 0.2v0 , so that is what we
will use for the remainder of this investigation.
Figure 2 depicts the reconstruction of V (ψ) /v0 obtained for two values of N .
The synthesis of Eq. (7.2.3) can be done with the matrix algorithm described in
Example 7.1. Let {V } be the column vector of the N + 1 velocity coefficients Vm ,
and let {ψ} be a set of J polar angles at which it is desired to evaluate the series,
so that {η} = cos ({ψ}). A rectangular J × (N + 1) array [L] is defined such that
column m is Pm ({η}) for m = 0, 1, ..., N . Then, the radial velocity on the surface
is given by {Vsurf } = [L] {V }. The result for N = 20 exhibits a good resemblance of
the actual function, while N = 80 is very close, with a minor Gibb’s phenomenon
at ψ = ψ0 = 30◦ .
1 N = 20
V(ψ)
N = 80
0.5 Actual
0
0 20 40 60 80 100 120 140 160 180
ψ (deg)
Figure 2.
The implication thus far is that a good representation of the surface velocity is
necessary to describe the pressure field adequately, but this is not so. Let us rewrite
the solution for pressure in Eq. (7.2.7) as
∞
Vm
Pff = ρ0 cv0 m h m (kr ) Pm (cos ψ) , m =
m=0
i h m (ka)
Correspondingly, Eq. (7.2.14) for the directivity and Eq. (7.2.17) for the radiated
power become
∞
D (ψ) =
m e imπ/2
Pm (cos ψ)
ka
m=0
∞
2 |m | 2
P = πρ0 a 2 cv02
m=0
2m + 1 (ka)2
The m values indicate which spherical harmonics are important to the pressure
field. Figure 3 displays the m values and the magnitude of each term in the series
for radiated power. These graphs would be unaltered if we truncated the series for
V (ψ) at N = 20, so it is evident that the small scale details of the surface motion
are unimportant for the radiated field.
4
2
Γm
02
contribution
Power
0
0 10 20 30 40 50 60 70 80
Polynomial degree m
Figure 3.
The next feature to examine is the farfield and where it begins. The primary aspect
of the farfield in any case is that the pressure decays as 1/r along any radial line.
Thus, a plot of kr |P/(ρ0 cv0 )| as a function of kr will approach a constant value as
the kr reaches into the farfield. The only radial line we were requested to consider
is ψ = 0, but one should carry out such computations along several lines in order to
assure that what is observed on a specific line is not anomalous. An evaluation of
|P| as a function of kr also may be carried out by the matrix algorithm. Specifically,
{kr } is a column vector of kr j values, and column m of [h] holds the values of
h m ({kr }) for m = 0, ..., N , and [L] is a diagonal N + 1 array whose elements are
the Pm (cos {ψ}) values. Then, the values of pressure at locations kr j along the radial
line at fixed ψ are [h] [L] {V }.
The center line is ψ = 0, so all of the Legendre function values are one. The
result of this computation is graphed in Fig. 4. The onset of a constant value of kr P
is gradual, but it seems that a good estimate would be that the product approaches a
constant value in the vicinity of kr = 100. The value of ka is 12, so the observation
that the farfield begins at kr = 100 suggests that the farfield begins at a kr value whose
order of magnitude is (ka)2 . This turns out to be a good empirical rule in general.
Specifically, if a is a measure of the size of an object that radiates an acoustic signal,
then the farfield will begin at a nondimensional distance krff = β (ka)2 , where β is
a value having unit order of magnitude. The identified value of β for the present
system is approximately 0.75.
40
krP(r,ψ=0)
20
0
0 ka 50 100 150 200 250
kr
Figure 4.
The last property to consider is the farfield directivity. From Eq. (7.2.14), we have

∞
D (ψ) =
m ei(m+1)π/2 Pm (cos ψ)
ka
m=0

The matrix-based algorithm for evaluating a series may also be implemented here.
The directivity obtained from the consistent and simplified analyses appear in Fig. 5.
To avoid obscuring details, the data has been displayed as a rectangular plot, rather
than the standard polar plot. It is not surprising, given the closeness of the two sets of
coefficients, that the directivity obtained from the simplified representation of V (ψ)
is quite good. We can see a hint of the high-frequency limit, in which the directivity
is proportional to the surface velocity. However, the value at ψ = 0 is 40% greater
than the limiting value, and D does not drop abruptly from cos ψ0 to zero at ψ0 .
2
(r/a) Pff
Consistent
1 Simplified
0
0 20 40 60 80 100 120 140 160 180
ψ (deg)
Figure 5.
7.2.3 Symmetry Plane
Anyone who is interested in sound reproduction systems is aware that speakers

usually are mounted on a baffle. A baffle is a boundary, usually flat, whose purpose
is to direct the radiated sound to one side. A system that features a sphere in a baffle is
a hemispherical dome tweeter, which is commonly used in speaker systems to create
and radiate the higher frequencies in the audible spectrum. In this arrangement, a
circular cap is mounted on the baffle, but is free to vibrate. In practice, the cap might
not be a full hemisphere, but we will employ that idealization because it has an
analytical solution. The example will compare the radiated field obtained with and
without the baffle.
A hemispherical dome tweeter is not the only configuration in which a baffle
guides spherical waves. A great simplification of a surface ship represents it as a
sphere. Suppose that the sphere’s buoyancy is such that its equator coincides with
the free surface. The characteristic impedance of water is much greater than that
of air. To a wave in the water, the surface appears to be pressure-release, whereas
it appears to be rigid to an acoustic wave in the air. Thus, the model for acoustic
radiation from the (idealized) ship into the air would consider a hemisphere in an
infinite, planar, rigid baffle. In contrast, the model for acoustic radiation from the ship
into the water would consider a hemisphere in an infinite, planar, pressure-release
baffle.
Figure 7.3 describes the geometry of a hemisphere in an infinite baffle. The portion
of the boundary that is the hemisphere is defined by r = a, 0 ≤ ψ < π/2, and the
baffle is defined by r > a, ψ = π/2. Velocity continuity requires that the particle
velocity matches the hemisphere’s radial velocity. In addition, if the baffle is rigid,
the particle velocity normal to the baffle must vanish. Because ēr is parallel to the
baffle and ēψ is perpendicular to it, this requirement is that V̄ · ēψ = 0 at ψ = π/2 for
r > a. The case of a pressure-release baffle requires that P = 0 at the same locations.
Thus, the problem to solve is
Fig. 7.3 A hemisphere z

mounted on an infinite planar
baffle ψ
a r
x y
⎧
⎪ ∇ 2 P + k 2 P = 0, r > a, 0 ≤ ψ < π/2
⎪
⎪
⎪
⎪ ∂P
⎨ = −iωρ0 Vh , r = a, 0 ≤ ψ < π/2
∂r ⎫ (7.2.25)
⎪ ∂P ⎬
⎪
⎪ Rigid baffle: =0
⎪
⎪ ∂ψ , r > a, ψ = π/2
⎩
Pressure-release baffle: P = 0 ⎭
where Vh is the complex amplitude of the radial velocity of the hemisphere.

The response with either type of baffle may be obtained as a special case of a
general axisymmetric wave. To see how to do so, we recall the method of images. If
the boundary of an infinite half-space is rigid, the field may be obtained by placing a
mirror image(s) behind the boundary of the actual source(s), with all images vibrating
in-phase with their matching actual source. Here, the image of a hemisphere is also
hemisphere, so the source and its image form a full sphere. If Vh (ψ) is the radial
velocity at location ψ on the hemisphere, then the mirror image point is at π − ψ.
Hence, the radial velocity distribution for the image sphere is

Vh (ψ) , 0 ≤ ψ < π/2
Rigid baffle: V (ψ) = (7.2.26)
Vh (π − ψ) , π/2 ≤ ψ < π
In the case where the boundary of a half-space is pressure-release, the mirror image
hemisphere vibrates 180◦ out-of-phase relative to the source hemisphere. Thus, the
radial velocity of the full sphere in this case is

Vh (ψ) , 0 ≤ ψ < π/2
Pressure-release baffle: V (ψ) = (7.2.27)
−Vh (π − ψ) , π/2 ≤ ψ < π
The acoustic field that is radiated must share the same symmetry properties as
the excitation. Thus, if the baffle is rigid, the pressure must be an even function of
ψ relative to ψ = π/2, in other words, P (r, π − ψ) = P (r, ψ). This requires that
the only nonzero spherical harmonics in the general solution, Eq. (7.2.1), be those
whose degree is even, P0 (cos ψ), P2 (cos ψ), .... Thus, the method of images tells us
that the field radiated by a hemisphere mounted on an infinite rigid baffle must be
describable as
∞
h 2m (kr )
P = ρ0 cv0 V2m P2m (cos ψ)
m=0
i h 2m (ka)
∞ (7.2.28)
h 2m (kr )
Vr = v0 V2m P2m (cos ψ)
m=0
h 2m (ka)
To verify the correctness of this ansatz, we evaluate Vψ at ψ = π/2. This should be

zero for any set of V2m coefficients. By Euler’s equation, this condition is obtained if
∂ P/∂ψ = 0 at this surface. The only quantity that depends on ψ in the preceding is
P2m (cos ψ). By virtue of being an even function relative to ψ = π/2, it follows that
(∂/∂ψ) P2m (cos ψ) ≡ 0 at ψ = π/2, which verifies that the boundary condition is
satisfied identically.
Similar reasoning applies in the case where the hemisphere is mounted on a

pressure-release baffle. The pressure in this case is an odd function of ψ relative
to ψ = π/2, so that P (r, π − ψ) = −P (r, ψ). The Legendre functions that fit this
symmetry specification are those whose degree is odd. Thus, the field radiated by a
hemisphere in an infinite pressure-release baffle is given by
∞
h 2m+1 (kr )
P = ρ0 c2 V2m+1 P2m+1 (cos ψ)
m=0
i h 2m+1 (ka)
∞
(7.2.29)
h (kr )
Vr = c V2m+1 2m+1 P2m+1 (cos ψ)
m=0
h 2m+1 (ka)
The odd-degree Legendre functions vanish at the midpoint of their range, P2m+1 (0) ≡
0, so the above representation identically satisfies the boundary condition that P = 0
on the baffle.
The velocity coefficients in Eqs. (7.2.28) and (7.2.29) describe the composite
V (ψ) function in Eq. (7.2.26) or (7.2.27). An alternative to integrating over this
sphere is to use orthogonality over the hemisphere. The even-degree Legendre poly-
nomials are a mutually orthogonal set over 0 ≤ ψ ≤ π/2, as are the odd-degree
polynomials. (The even and odd sets are not orthogonal with respect to each other
in this range.) Thus, we have
1 π/2
P2m (η) P2 j (η) dη = P2m (cos ψ) P2 j+1 (cos ψ) sin ψdψ
0 0
1
= δ jn
2 (2m) + 1
1 π/2
P2m+1 (η) P2 j+1 (η) dη = P2m+1 (cos ψ) P2 j+1 (cos ψ) sin ψdψ
0 0
1
= δ jn (7.2.30)
2 (2m + 1) + 1
When we apply these orthogonality properties to the respective series for Vr , we find
that the velocity coefficients are given by
π/2
V (ψ)
Rigid baffle: V2m = (4m + 1) P2m (cos ψ) sin ψdψ
0 c
π/2
V (ψ)
Pressure-release baffle: V2m+1 = (4m + 3) P2m+1 (cos ψ) sin ψdψ
0 c
(7.2.31)
The representations in Eqs. (7.2.28) and (7.2.29) are like those for a sphere in
an unbounded space, except that only the even or odd spherical harmonics are rele-
vant. The fact that the domain is a half-space affects the radiated power. The power is
obtained by integrating the time-averaged intensity over a large hemisphere. Because
(Ir )av is formed by multiplying P and Vr∗ product, and both are either even or odd
functions relative to ψ = π/2, (Ir )av is an even function. Thus, the integrals over
the hemisphere are half the value of what they would be over the full sphere. Cor-
respondingly, the time-averaged radiated power is obtained by halving Eq. (7.2.17)
and using only the appropriate set of V j coefficient. In other words
⎧ ∞
⎪
⎪ |V2m |2
⎪ 2 2 1 1
πρ
⎪ 0
⎨ a cv

: rigid
(ka) m=0 4m + 1
h 2m (ka)
2
0 2
P= ∞ (7.2.32)
⎪ πρ a 2 cv 2 1 1
⎪
⎪
|V2m+1 |2
⎪
⎩ 0

2 : pressure-release
(ka)2 m=0 4m + 3
h (ka)
0
2m+1
Somewhere in the midst of this discussion, you might have wondered about the
relevance of the concept of an infinite baffle, given that nothing has infinite extent.
An important general aspect is that the pressure on a baffle surrounding a radiator
tends to fall off rapidly with increasing distance from the center. Thus, the baffle only
needs to extend sufficiently far to attain this decrease. In most cases, especially at
high frequencies, this distance has the order of magnitude of an acoustic wavelength,
which corresponds to an outer radius of the baffle being Rbaffle = a (1 + 2π/ka). In
some designs, the apparatus used to mount the transducer is sufficiently large that it
is a good approximation of an infinite baffle.
EXAMPLE 7.4 A tweeter consists of a hemisphere that translates as a rigid

body at v0 sin (ωt) in the direction of the axis of symmetry. It is desired to
compare the field that this transducer radiates when it is mounted on an infinite
rigid or pressure-release baffle to a reference configuration. The latter places the
sphere in an unbounded fluid with translational motion only over the forward
half, 0 ≤ ψ ≤ π/2, with the back half, π/2 < ψ ≤ π, stationary. Compare
the farfield distribution (r/a) | p| / (ρ0 cv0 ) for both types of baffles to that
obtained from the reference sphere. Also, evaluate the radiated power and the
pressure distribution acting on the surface of the hemisphere for each system.
Frequencies of interest are ka = 1, 4 and 20.
Significance
In addition to gaining experience working with spherical harmonics, this example
will exhibit some important general features of acoustic radiation and provide insight
to the behavior of a common device.
Solution
In each configuration, the complex amplitude of the radial velocity for 0 ≤ ψ ≤ π/2
is Vr = v0 cos ψ. Both baffle cases give rise to an image of the hemisphere that covers
π/2 < ψ ≤ π, thereby creating a virtual complete sphere. For a rigid baffle, the radial
velocity must be the same on the image, so that Vr = −v0 cos ψ. The union of these
for the virtual sphere is Vr = v0 |cos ψ|. For a pressure-release baffle, the radial
velocity for ψ = π/2 + > π/2 must be equal in magnitude but in the opposite
sense from the value at ψ = π/2 − . Thus, the radial velocity on the virtual sphere
is Vr = v0 cos ψ. The radial velocity on the reference full sphere is Vr = v0 cos ψ for
0 ≤ ψ ≤ π/2 and Vr = 0 for π/2 < ψ ≤ π.
The radial velocity on the virtual sphere for the rigid baffle case is an even function
with respect to ψ = π/2. This means that only the only nonzero velocity coefficients
are V2m , m = 0, 1, 2, .... According to the first of Eq. (7.2.31), these coefficients are
π/2
V2m = (4m + 1) (cos ψ) P2m (cos ψ) sin ψdψ
0
This is like Eq. (1) in Example 7.3 with ψ0 = π/2, except that 2m replaces m and
the integral is doubled. When we make these alterations, the result is

2m + 1 (4m + 1)
(V2m )rigid = − P2m+2 (0) + P2m (0)
4m + 3 (4m + 3) (4m − 1)
2m
− P2m−2 (0) , m > 0
4m − 1
1
(V0 )rigid =
2
In the case of a pressure-release baffle, the radial velocity function for the equiv-
alent full sphere is an odd function relative to ψ = π/2. In other words, it is cos ψ.
Thus, the surface velocity of the image sphere is that of a sphere that translates as
a rigid body. Because P1 (cos ψ) ≡ cos ψ, orthogonality of the Legendre functions
informs us that V1 = 1 is the only nonzero coefficient.
The radial velocity on the reference full sphere has no special symmetry properties
with respect to ψ = π/2. The velocity coefficients are given directly by Eq. (7.2.4)
for m = 0, 1, 2, .... Setting Vr = 0 for ψ > π/2 leads to
1
1
Vm = m + η Pm (η) dη
2 0
This equation is the same as Eq. (1) in Example 7.3 with ψ0 = π/2, so we quote the
results directly as

1 v0 m + 1 (2m + 1)
(Vm )ref =− Pm+2 (0) + Pm (0)
2 c 2m + 3 (2m + 3) (2m − 1)
m
− Pm−2 (0) , m > 1
2m − 1
1 v0 1 v0
(V0 )ref = , V1 =
4 c 2 c
Despite the similarity of the surface velocity in the three configurations, the V j
coefficients are significantly different, as is shown in Fig. 1. This figure also shows
that truncation of the spherical harmonic series at N = 12 should be adequate for
the rigid and reference cases.
1 Rigid
Pressure−release
Reference
0.5
Vj
0
0 5 10 15 20
Spherical harmonic number j
Figure 1.
The computation of the directivity proceeds as though we had a full sphere, except
that ψ only extends to π/2 when there is a baffle, and the velocity coefficients are
those of the relevant configuration.
∞ Equation (7.2.14) for the three cases
gives
1
(V2m )rigid i(2m+1)π/2
Drigid (ψ) =
e P2m (cos ψ)
ka
m=0 i h 2m (ka)
(V1 )p.r.
v0 (ka)2
Dp.r. (ψ) = cos ψ =

cos ψ
ka
i h 1 (ka)
c
2 + 2ika − (ka)2
(Vm )ref i(m+1)π/2
Dref (ψ) =
e P m (cos ψ)
ka
i h m (ka) m=0

The directivities for each specified frequency are displayed in Fig. 2.

Rigid baffle Pressure-release baffle Reference sphere
0.5 1.2 1.2
ka = 1 1 ka = 4 ka = 20
(r/a)Pff/(ρ0cv0)
0.4 1
0.8 0.8
0.3
0.6 0.6
0.2 0.4 0.4
0.1 0.2 0.2
0 0 0
0 30 60 90 0 30 60 90 0 30 60 90
ψ (deg)
Figure 2.
Because the pressure-release baffle corresponds to an image sphere that oscillates

as a rigid body in the z direction, the directivity is always proportional to cos ψ, but its
maximum value is frequency dependent. At the lowest frequency, ka = 1, the rigid
baffle and the reference sphere radiate approximately uniform. This is consistent
with the general observation that a body that executes a low-frequency vibration acts
like a monopole whose strength is proportional to the body’s volume velocity. In
contrast, at the highest frequency, ka = 20, each directivity fits the asymptotic trend
that it will be proportional to the surface velocity distribution.
The radiated power in each case may be found from Eq. (7.2.17) using the appro-
priate set of V j coefficients. For the configurations where there is a baffle, the
formula is halvedbecause a baffle reduces the domain to a half-space. The results

for P/ πρ0 a 2 cv02 are tabulated below.
P (nondim.)
ka 1 4 20
Rigid baffle 0.12588 0.32242 0.33331
Pressure-release baffle 0.066667 0.32821 0.33333
Reference sphere 0.096272 0.32531 0.33332
At low frequencies, the rigid baffle configuration gives twice as much power radiation
as the pressure-release case, while the reference sphere is midway between the two
baffle configurations. In contrast, at high frequencies, all configurations give the same
radiated power. This is to be expected because the directivity of all configurations at
large ka is essentially the same for ψ < 90◦ , and it is close to zero for ψ > 90◦ in
the case of the reference sphere.
The evaluation of surface pressure follows a direct implementation of Eq. (7.2.7),
with the V j coefficients for the rigid baffle case being those for even j, while the only
nonzero coefficient for the pressure-release baffle is V1 . Of course, only 0 ≤ ψ ≤ π/2
is relevant. Here too, the matrix algorithm for the evaluation of a series representation
may be applied.
The radial velocity on the surface is real and positive. Thus, Re (P) is the resistive
part, and Im (P) is the reactive part, with Im (P) > 0 for an inertance. The patterns
in Fig. 3 indicate that when the frequency is low, the surface pressure for both types
of baffles has a large inertance. The resistive part for the rigid baffle at low frequency
is approximately uniform, whereas Re (P) is proportional to cos ψ for the pressure-
release baffle. High frequencies lead to a surface pressure that is ρ0 cv0 cos ψ for both
configurations. This occurs because the surface, r = a, effectively is in the farfield
when ka is very large.
R e( P) Im ( P )
1 Rigid
0.5 Pressure-release
0
ka =1
−0.5
1
0.5
0
ka =4
−0.5
1
0.5
0
ka=20
−0.5
0 30 60 90 0 30 60 90
Figure 3.
In systems for sound reproduction, it usually is desirable to emit the sound omni-
directionally. This is the role of a hemispherical dome tweeter. Because a is relatively
small (typically less than 40 mm), the value of ka will be small even at the upper
limits of the audible spectrum. Thus, their use leads to radiation of high frequen-
cies over a broad angular range. The rigid baffle enhances the radiated power at low
frequencies. A designer would be interested in the surface pressure for design pur-
poses. For example, the fact there is a large reactive part means that the electronic
amplifier that drives the tweeter must be able to output an instantaneous power that
is substantially greater than the time-averaged power.
7.2.4 Interaction with an Elastic Spherical Shell
A situation we have not yet addressed arises when the surface vibration is produced
by a force applied to an elastic body. The vibration in that case must be determined
by solving laws of structural dynamics. Such an analysis must account for the pres-
sure applied to the surface by the acoustic signal that is generated. But that pressure
depends on the surface vibration. This is not circular logic. Rather, it requires a simul-
taneous approach that concurrently formulates the governing acoustic and structural
equations. Problems such as this fall into the category of fluid–structure interaction.
If the shape of the body and the nature of the structure are complicated, as in the
case of a submarine, then accurate determination of the radiated field at all frequen-
cies would require the most powerful computational tools available, and even such
resources might not be adequate. The decoupling properties of spherical harmonics
make the analysis of radiation from elastic spheres quite tractable.
We previously encountered an elastic plate, which refers to a planar sheet whose
thickness is much less than the dimensions of the sheet. A shell consists of a thin
sheet whose surface is curved. Doing so alters the nature of the internal stresses
in comparison to a plate, which thereby complicates the equations of motion. We
will begin with a membrane shell model, in which the internal stresses come from
stretching the sheet.
To some extent, the stresses resemble how a balloon carries a load, but there is a
fundamental difference. In the case of a balloon, the sheet is inflated before it is subject
to loading. Inflation stretches the sheet, thereby inducing a prestress. Displacement
of the sheet produces additional stress. The work that is done in this displacement,
which is stored as potential energy, comes from displacing the system in opposition
to the total stress. In the case of a membrane shell, there is no prestress, so the internal
work stems from displacing the shell in opposition to the stretching stress induced
by that displacement. A simpler system that has the same alternatives is a cable. The
inflated balloon is analogous to a tensioned cable, whereas a membrane shell is like
a taut, but untensioned cable. Interestingly, the taut cable has no load carrying ability
according to linear elasticity theory. This is not true for the membrane shell because
of its curvature.
The membrane shell theory we will employ becomes progressively less valid as the
thickness/curvature ratio increases. It also is not valid if the frequency is so high that
the wavelength of the surface displacement is comparable to the thickness. Improved
shell theories that address both situations are available. (The author once joked to a
colleague that there might be more improved shell theories than the number of people
who have been involved in the development of those theories.) We shall begin with
membrane shell theory because it is the simplest model. However, a breakdown of
this theory will cause an anomaly of the series, which we will address by introducing
a correction that accounts for flexural effects.
Equations of Motion
Figure 7.4 shows the midsurface of a spherical shell, which is the surface that is
midway between the outer and inner surfaces. We will limit our consideration to
situations in which the shell response is axisymmetric. This means that the displace-
ment of a point at (a, ψ) on the midsurface will consist of a radial component w
and meridional displacement u tangent to the surface. A positive displacement com-
ponent is one that moves the midsurface in the sense of ēr or ēψ . The restriction to
axisymmetric motions limits u and w to be functions of ψ and t only. The shell’s
thickness h is much less than a. Thus, we may also use a as the radius of the outer
surface when we formulate the equations governing the acoustic field exterior to the
sphere.
Fig. 7.4 Definition of the z

radial displacement w and
polar displacement u of a
spherical shell  w
r u
y

x
Junger and Feit7 employed equations of motion that incorporate the effects of flex-
ural rigidity. Omitting those effects leads to the following simplified set of equations
of membrane shell theory,
7 M.C. Junger and D. Feit, Sound, Structures, and Their Interaction, Acoustical Society; Second
edition, pp. 229–230 (1993 reprint).

∂2u ∂u ∂w 2∂ u
2
ce2 + cot ψ − ν + (cot ψ) 2
u + (1 + ν) − a =0
∂ψ 2 ∂ψ ∂ψ ∂t 2
2
∂u ∂ w
2
a
−ce2 (1 + ν) + (cot ψ) u + 2w − a 2 2 = pacoustic |r =a − qapplied
∂ψ ∂t ρe h
(7.2.33)
In this expression, qapplied represents the outward radial force per unit surface area
that is the excitation, ρe is the density of the shell’s material, ν is Poisson’s ratio, and
ce is the phase speed of an extensional wave of plane strain in a plate of the same
thickness, 1/2
E
ce = (7.2.34)
ρe 1 − ν 2
We know that the pressure exterior to a sphere may be represented by a spherical

harmonic series. Also, the acceleration term in the second of Eq. (7.2.33) is propor-
tional to w in the case of a harmonic response. Because the w term in that equation
is not differentiated with respect to ψ, it seems reasonable to try a Legendre function
series for w. In both equations of motion, most of the u terms differ from the w terms
by one differentiation with respect to ψ. This feature suggests that we should try a
series representation of u, with the derivatives of the Legendre functions as the basis
functions. Thus, the ansatz we begin with is
∞

p = Re Bm h m (kr ) Pm (cos ψ) eiωt
m=0
∞

w = Re Wm Pm (cos ψ) eiωt (7.2.35)
m=0
∞
d
u = Re Um Pm (cos ψ) eiωt
m=0
dψ
Many mathematical manipulations are required to prove that the basis functions
for w and u are consistent with the equations of motion, and to identify the equations
governing the coefficients Wm and Um . Both Baker8 and Junger and Feit9 apply the
transformation η = cos ψ prior to identifying the basis functions. In both analyses,
the series for w uses Pm (η), which is equivalent to the above. The u basis function
1/2
used by Junger and Feit is 1 − η 2 Pm (η), whereas the basis function for u in
Baker’s analysis is the associated Legendre polynomial Pm1 (η). Equation (7.1.42)
1/2
states this function is Pm1 (η) = (d/dψ) Pm (cos ψ) = − 1 − η 2 Pm (η). Thus,
Baker’s series matches the one used here, whereas Junger and Feit’s work corresponds
to Um coefficients that are the negative of the present definition.
8 W.E. Baker, “Axisymmetric Modes of Vibration of Thin Spherical Shells,” J. Acoust. Soc. Am. 33,
1749–1758 (1961).
9 M.C.
Junger and D. Feit, ibid.
The equations that result from substitution of Eq. (7.2.35) into the equations of
motion, followed by application of the derivative identities in Eq. (7.1.14), are
∞

−K UU + (ke a)2 Um + K U W Wm Pm1 (η) = 0
m=0
∞

K W U Um + −K W W + (ke a)2 Wm Pm (η)
m=0 ∞
a2
= Bm h m (ka) Pm (η) − q̂applied
ρs hce2 m=0
(7.2.36)
where the K coefficients are functions of m given by
K UU = m (m + 1) − (1 − ν) , K U W = (1 + ν)
(7.2.37)
K W U = (1 + ν) m (m + 1) , K W W = 2 (1 + ν)
The wavenumber ke is defined relative to ce , rather than c, in order to avoid confusion

when we consider the shell in a vacuum. Thus,
ω ce
ke = , k = ke (7.2.38)
ce c
The first of Eq. (7.2.36) yields uncoupled equations because the Pm1 (η) are a lin-
early independent set. To uncouple the second equation, we employ the orthogonality
property of Legendre functions, Eq. (7.1.16). The result of multiplying the equation
by a specific Pn (η), then integrating over −1 ≤ η ≤ 1, is

K UU − (ke a)2 Um − K U W Wm = 0
a2 a2 (7.2.39)
−K W U Um + K W W − (ke a)2 Wm = − Bm h m (ka) + Fm
ρe hce2 ρs hce2
The Fm coefficients are those of a Legendre series for the applied excitation,

2m + 1 1
Fm = q̂applied Pm (η) dη (7.2.40)
2 −1
These values are considered to be known, so there are three unknowns in Eq. (7.2.39).
The additional equation comes from enforcing continuity at the sphere’s surface.
The relation between the spherical harmonic coefficients of pressure and radial veloc-
ity on a sphere’s surface is provided by Eq. (7.2.6). The radial velocity of the surface
is ẇ, so the velocity coefficients are iωWm , from which it follows that
1
Bm = ρ0 cω Wm (7.2.41)
h m (ka)
We now have a solvable triad of equations. The most direct solution uses the
preceding to eliminate Bm and the first of Eq. (7.2.36) to eliminate Um . This leads to
KU W
Um = Wm
K UU − (ke a)2
2
KU W K W U ρ0 a c kah m (ka) Wm a Fm
KW W − (ke a) −
2
+
=
K UU − (ke a) 2 ρs h ce h m (ka) a h ρe ce2
(7.2.42)
Terms have been grouped to emphasize the manner in which the system parameters
affect the response. The pressure excitation is referenced to ρe ce2 , which is a bulk
modulus for the shell, and displacement is referenced to the radius. This nondimen-
sional displacement depends parametrically on the density and sound speed ratios,
Poisson’s ratio ν, and the nondimensional frequency.
Evaluation of the response at any frequency entails solving the second of the
above equations for the radial displacement coefficient Wm for a range of harmonic
numbers m, then using those coefficients to evaluate the corresponding pressure
coefficients Bm according to Eq. (7.2.41). The meridional displacement typically is
not of interest, but if it were, we would use the first equation above to evaluate the
Um values. The displacement and pressure fields would be synthesized according to
Eq. (7.2.35). The farfield description of the pressure is obtained by replacing h m (kr )
by its asymptotic behavior for large kr , which yields
∞

e−ikr
p = Re Bm ei(m+1)π/2 Pm (cos ψ) eiωt (7.2.43)
kr m=0
In-vacuo Vibration
An important property of the shell is its natural frequencies, at which the system may
vibrate without application of an external excitation. No free system can execute a
steady-state harmonic response if energy is removed from it. As we will soon see,
that is what the fluid always does. Thus, the natural frequencies correspond to the
harmonic motion of the in-vacuo shell when there is no excitation. The effect of fluid
loading is described by the last term in the bracket in the second of Eq. (7.2.42).
Dropping this term and setting Fm = 0 leads to the characteristic equation, which is
a quadratic equation whose roots are the natural frequencies m of the mth spherical
harmonic,
10
Ω(1)
m a/ce
5 Ω(2)
m a/ce
0
1 Um /Wm @ Ω(1)
m
Um /Wm @ Ω(2)
m
0
−1
0 2 4 6 8 10
Spherical harmonic m
Fig. 7.5 Natural frequencies and modal displacement ratios of a spherical shell according to mem-
brane shell theory
2 2
m a m a m a
χ = KW W − K UU − − KU W K W U = 0
ce ce ce
(7.2.44)
Thus, there are two natural frequencies (1) (2)

m < m at a specific m. At either nat-
ural frequency, Eq. (7.2.39) with Bm = Fm = 0 ceases to be linearly independent.
Consequently, we may only determine the ratio Um /Wm from either of those equa-
tions. When the natural frequencies are plotted as functions of m, they form two
curves called branches. They are depicted in Fig. 7.5 along with the displacement
ratio Um /Wm for each branch. The common parlance is to say that the displacement
pattern in a free vibration at a system’s natural frequency is a mode function or mode
shape. The radial displacement in a mode is the product of Wm and Pm (cos ψ), while
the meridional displacement is the product of Um and (d/dψ) Pm (cos ψ).
The lower branch has no natural frequency at m = 0. (The mathematical solu-
tion is not physical, because it corresponds to an imaginary natural frequency.)
At m = 1, the lower natural frequency is zero, which means that it corresponds
to a rigid body mode in which the shell moves without deformation. This is
(1) (1)
verified by the modal displacement
ratio U1 /W1 =1, which corresponds to
w = Re W1(1) cos ψei t , u = Re −W1(1) sin ψei t . This pattern is transla-
(1) (1)
tional oscillation in the axial direction.

If the natural frequencies are known, the second of Eq. (7.2.42) in the case of
forced response may be written as
⎡ 2
2
κ(1) − (ke a)2 κ(2) − (ke a)2
⎢ m m
⎣ 2
κ(0)
m − (ke a)2 (7.2.45)
2
ρ0 a c kah m (ka) Wm a Fm
+ =
ρe h ce h m (ka) a h ρe ce2
where
(1) a (2) a
κ(1)
m ≡ m , κ(2)
m ≡ m , κ(0)
m = (K UU )
1/2
(7.2.46)
c c
The special case of a shell in a vacuum is obtained by setting ρ0 /ρe = 0. It is
evident that removal of fluid loading will cause Wm to be singular when ke a equals
κ(1) (2)
m or κm . This condition marks a resonance, which we will examine more closely
in the following section. In contrast, this equation indicates that Wm will be zero
when ke a = κ(0) (1) (2)
m . In control system theory, ω = m or ω = m are “poles” of the
in-vacuo system, and ω = (ce /a) (K UU ) is a “zero”.
1/2
A troubling aspect of the displacement in any case becomes evident if we consider

the value of Wm as m is increased with ke a fixed. According to Fig. 7.5, κ(1)m tends to
a constant value as m increases. Also, examination of Eq. (7.2.37) indicates that K UU
approaches m 2 , so κ(0)
m approaches m. Consequently, for very large m at fixed ka, the
first term in Eq. (7.2.45) approaches a constant value. The asymptotic expansion of
h m (ka) and its derivative for large m, see Eq. (7.1.36), indicate that the second term
also tends to a constant value. If the Fm coefficients do not decrease with increasing
m, the Wm values will not decrease. In that event, none of the spherical harmonic
series will converge.
The source of this difficulty is the usage of membrane shell theory. In particular,
the fault lies in its prediction that the natural frequencies on the lower branch κ(1) m
approach a constant value as m is increased. Any shell theory that incorporates flex-
ural deformation effects will yield natural frequencies that increase with increasing
harmonic number. For this reason, we shall henceforth use the full theory presented
by Junger and Feit. The parameter that scales flexural effects relative to membrane
effects is
h2
β2 = (7.2.47)
12a 2
Fortunately, adoption of this improved shell theory only entails altering the definitions
of the K coefficients, which now are

K UU = 1 + β 2 [m (m + 1) − (1 − ν)]
K U W = (1 + ν) + β 2 [m (m + 1) − (1 − ν)] (7.2.48)
K W U = m (m + 1) (1 + ν) + β 2 m (m + 1) − β 2 (1 − ν)
K W W = 2 (1 + ν) + β 2 m 2 (m + 1)2 − β 2 (1 − ν) m (m + 1)
101 Ω(2) a/ce

Flexural correction
Membrane theory
Ω(1) a/ce
100
0 5 10 15 20
Fig. 7.6 Comparison of natural frequencies of a spherical shell according to membrane theory to
those obtained when flexural effects are included, a/ h = 50
The natural frequencies that result from inclusion of flexural effects are depicted
in Fig. 7.6. There is no perceptible effect on the natural frequencies of the upper
branch. The frequencies of the lower branch increase as expected, and the rate of
increase is more rapid than a simple proportionality to m. This increase is sufficient
to assure that the Wm values obtained from Eq. (7.2.45) decrease for sufficiently large
m, regardless of the nature of the excitation, qapplied . Decreasing a/ h increases the
values of (1) (2)
m , but has little effect on the values of m . The modal displacement
( j) ( j)
ratios Um /Wm are not displayed because the values associated with either branch
are indistinguishable from those in Fig. 7.5.
Even with the inclusion of flexural effects, an important limitation must be recog-
nized. Simple shell theory is not valid if the wavelength along the surface is compara-
ble to the thickness, because other deformation effects must be included. A qualitative
guideline for validity may be identified from the property that Pm (cos ψ) has m zeros
in the range 0 < ψ < π. If we ignore the fact that these zeros are not equidistant,
we arrive at a surface half-wavelength of 2πa/m. A reasonable criterion is that this
distance should be no less than 10h. This leads to a guideline that the shell theory
employed here can be expected to lose accuracy if m > (π/5) a/ h. For example, if
a/ h = 50, it is desirable that the spherical harmonic series converge prior to m = 32.
If they do not, we might obtain a mathematically convergent solution that does not
accurately describe the actual physical response.
Interpretation
In the present context, the in-vacuo vibration properties provide a lens for interpreting
the fluid-loaded response. To see why, let us compare Eq. (7.2.42) to the expression for
the displacement of a one-degree-of-freedom oscillator subjected to a harmonically
varying force. Let K , M, and D, respectively, denote stiffness, mass, and dashpot
constants, nondimensionalized based on the usage of ka to represent frequency. In
terms of a nondimensional velocity V = ika (W/a), the steady-state amplitude of
the oscillator is governed by

K
Z V = F, Z = i ka M − +D (7.2.49)
ka
The imaginary part of the impedance is the reactance. It is negative if ka is below

the natural frequency, which is (K /M)1/2 , and it is positive if ka exceeds the natural
frequency. The dashpot term is a positive real value at any frequency. If the dashpot
was not present, a frequency sweep would reveal a true resonance at the natural
frequency, in which the value of V is infinite.10 The dashpot removes energy at any
frequency, so the response never shows a true resonance if the dashpot is present.
To apply this perspective to the spherical shell, we convert Eq. (7.2.45) to an
impedance form by factoring ike a out of the coefficient. The coefficients in the first
line of this expression represent the effects of the structure, whereas the term in the
second line is due to the fluid. Let us denote these contributions as the structural
impedance (Z e )m and the fluid surface impedance (Z f )m , respectively. Thus, we
rewrite the equation as

Wm a Fm
(Z e )m + (Z f )m ike a =
a h ρe ce2

(1) 2 (2) 2
ke a κ m / (k e a) 2
− 1 κ m / (k e a) 2
− 1
(Z e )m = 2 (7.2.50)
i
κ(0)
m / (ke a)2 − 1

ρ0 a c h m (ka)
(Z f )m = −i
ρe h ce h m (ka)
The frequency dependence of (Z e )m + (Z f )m does not match that of a simple

oscillator, but we can see that (Z e )m is imaginary, so it is a reactance. If ke a < κ(1)
m ,
then it is negative imaginary, corresponding to a compliance. Also, (Z e )m is infinite
when ke a = κ(0)m , and it grows without bound to a large positive imaginary value
(inertance) with increasing ke a beyond κ(2)m . A large impedance leads to small velocity
and displacement amplitudes.
The fluid impedance, (Z f )m , is complex and nonsingular. The frequency depen-
dence of both impedances for the lowest spherical harmonics is described by Fig. 7.7
for the case of a steel spherical shell submerged in water. Several features are note-
worthy. The most important are that Im (Z f ) > 0 and Re (Z f ) > 0 for any frequency.
In other words the fluid loading adds an inertance and a resistance.
We have seen that if the shell vibrates in a vacuum, it will resonate when ke a = κ(1)
m
or ke a = κ(2)
m . Both conditions correspond to (Z e )m = 0. (In an actual system, internal
dissipation would limit Wm to a finite value.) Now consider what happens when the
impedance of the fluid is added to that of the shell. Because Im (Z f ) is always greater
than zero, a plot of Im ((Z e )m + (Z f )m ) as a function of ke a would yield a curve at
each m that is above the corresponding plot of Im(Z e )m in Fig. 7.7. Consequently,
the total inertance crosses the zero axis at a lower frequency. It is convenient to refer
10 An infinite value of Wm is an artifact. If any linear system is excited at a natural frequency, the
response grows in time without bound. This behavior was observed in Sect. 2.5.3. The mathematical
explanation may be found in Mechanical and Structural Vibration, J.H. Ginsberg, John Wiley and
Sons, Inc., Sect. 3.3 (2001).
10
Im(Z e )
0
-10
2 m= 0
m= 1
Im(Z f )
m= 2
1 m= 3
m= 4
0
2
Re(Z f )
0
0 1 2 3 4
Frequency ke = ωa/ce
Fig. 7.7 Structural and fluid impedances for several spherical harmonics as a function of frequency.
The media are steel and water, and a/ h = 50
to any condition in which the total reactance vanishes as a fluid-loaded resonance.

These resonances occur at a lower frequency than the corresponding (in-vacuo)
natural frequencies. Despite the terminology, it is essential that one not forget that
fluid resistance is always present. This resistance, being in-phase with the surface
velocity, plays the same role as a dashpot. The difference is that a dashpot dissipates
energy internally, whereas the power required to overcome the fluid resistance is
transferred into the fluid as the radiated pressure field. The name given to this effect
is radiation damping.
Another important observation stems from the fact that the magnitude of Z f is
proportional (ρ0 c)/(ρe ce ), that is, the ratio of characteristic impedances. For
water

and steel, this ratio is approximately 0.03, whereas it is approximately 0.9 10−5 in
the case of air and steel. The usual terminology is to say that the former constitutes
heavy fluid loading, whereas the latter is light fluid loading. In a lightly fluid-loaded
system, the fluid impedance will be much smaller than the structural impedance,
except at the in-vacuo natural frequencies, where Im (Z e ) = 0.
This attribute leads to a simplification that often is invoked in the cases of light
fluid loading. In it, the structural response is determined with the in-vacuo model. The
effect of fluid resistance when the frequency is close to in-vacuo natural frequencies
is estimated, typically on the basis of prior experience with similar systems. The
vibratory displacement is then used to predict the pressure field as a standard radiation
problem. The important aspect of this simplified approach is that increasing ρ0 c/ρe ce
will lead to a decrease in the peak amplitudes at resonances, but it will not account
for the downshift of the frequency at which these resonances occur.
The discussion thus far has focused on the behavior of individual spherical har-
monics. We have examined the factors influencing the displacement and pressure
coefficients, and have seen that the structure has an infinite number of in-vacuo
modes that have some relevance to resonant-type phenomena. We have not seen how
these behaviors combine to affect the displacement and acoustic responses to an
actual excitation. This we shall do by investigating a prototypical system.
EXAMPLE 7.5 A spherical shell (a/ h = 40) composed of aluminum is sub-

merged in water. It is excited by a time-harmonic point force that acts in the
radial direction. (a) Determine the amplitude of the displacement coefficients
Wm as a function of frequency for m ≤ 4 and 0 < ke a < 8. (b) Determine the
amplitude of the radial displacement at the point where the force is applied
and at the diametrically opposite point for 0 < ke a < 8. (c) Determine the
farfield pressure amplitude r |P| for 0 < ke a < 8 in the direction ψ = 0 and
ψ = 180◦ , where ψ is measured relative to the radial line through the point
where the force is applied. (d) Determine the directivity of the farfield pres-
sure amplitude. Carry out the analysis for three frequencies: the frequency in
the range ke a < 1 for which r |P| at ψ = 0 is maximized, ke a = 2.5, and the
frequency in the range ke a < 5 for which r |P| at ψ = 0 is maximized.
Significance
An important part of any analysis is interpretation and explanation of the results. This
example will explore how to do so when elastic and acoustic effects interact. Along
the way, issues regarding efficient computational approaches and series convergence
will come to the fore.
Solution
The only deviation of this system from radial symmetry is the point force. If its
location is defined as the ψ = 0 pole for a spherical coordinate system, the system is
axisymmetric with respect to the z-axis. The value of Wm as a function of frequency
is given by Eq. (7.2.42). Solution of those equations requires specification of the
Fm coefficients, which are described by Eq. (7.2.40). That expression is based on
a distributed force, that is,
a force
per unit area, whereas the present excitation is
a concentrated force Re F̂e iωt
applied at η = 0. We could use a Dirac delta to
describe this force, but a less mathematical approach is available. Let us consider
the force to be distributed over a small spherical cap defined by 0 ≤ ψ ≤ , where
1. The point force coefficients will emerge by recognizing the smallness of .
The distributed force is qapplied = F̂/A for ψ ≤ and qapplied = 0 for ψ > ,
where A is the area of the cap,

A= (2πa sin ψ) (adψ) = 2πa 2 (1 − cos ) ≈ πa 2 2
0
It is convenient to use η = cos ψ to change the integration variable in Eq. (7.2.40),

so that
2m + 1 F̂
Fm = P (cos ψ) sin ψdψ
2 2 m
2 0 πa
Because is extremely small, the integration variable is always small. Therefore,

the Legendre functions are well approximated as Pm (cos ψ) ≈ 1, and we may set
sin ψ ≈ ψ. The result is
2m + 1 F̂
Fm = (1)
4 πa 2
Before we proceed to any computations, it is useful to review the equations to be
solved. First, the nondimensional natural frequencies κ(1) (2)
m and κm , which are inde-
pendent of ka, will be evaluated by finding the roots of Eq. (7.2.44). With those
quantities known for each azimuthal harmonic m,we may solve Eq. (7.2.45) for all
Wm /a. The right side of that equation is (a/ h)Fm / ρe ce2 . In view of Eq. (1), we shall
nondimensionalize this term by defining a variable χm that replaces Wm /a, such that
Wm F̂
= χm (2)
a ρe ce2 ah
The result of changing variables in this manner is to convert Eq. (7.2.45) to

⎡ 2
2
κ(1) − (ke a)2 κ(2) − (ke a)2
⎢ m m
⎣ 2
κ(0)
m − (ke a)2 (3)
2
ρ0 a c kah m (ka) 2m + 1
+ χm =
ρe h ce h m (ka) 4π
The values of χm may be used to evaluate the coefficients of the pressure series.
Introducing Eq. (2) into (7.2.41) leads to
# $
ka Wm F̂ ρ0 c2 ka
Bm = ρ0 c 2
= χm (4)
h m (ka) a ah ρe ce2 h m (ka)
Thus, the χm values found from Eq. (3) are the (dimensional) displacement coeffi-
cients ratioed to F̂/ ρe ce2 h . Using those values to evaluate, Eq. (4) yields the pressure
coefficients ratioed to F̂/ (ah). The displacement and pressure at any location are
found by synthesizing the series in Eq. (7.2.35).
Part (a) requested the Wm coefficients, which now are χm , for m ≤ 4. However,
evaluation of the displacement and pressure series requires computations for increas-
ing m until the series converge. Thus, our strategy is to solve Eq. (3) at the full range
of frequencies, which we will designate (ke a) j . The result will be a matrix χ̃ whose
element at row m + 1 and column j is χ̃m, j = χm at (ke a) j . It is logical to follow
the evaluation of the χm coefficients with an evaluation of the corresponding Bm

coefficients. Doing so will yield a matrix [B] that matches [χ].
Standard properties for aluminum are ρe = 2100 kg/m3 , ce = 6420 m/s, and
ν = 0.33. The in-vacuo natural frequencies are tabulated below.
0 1 2 3 4
(1)
m – 0 0.7014 0.8327 0.8895
(2)
m 1.6125 1.9749 2.7221 3.6351 4.5967
Figure 1a shows the frequency dependence of |χm | for m = 0 to 4. As expected,

each amplitude has a zero at ke a = κ(0) m , and it falls off with increasing frequency.
Although the theory indicates that each spherical harmonic should show two fluid-
loaded resonances because there are two natural frequencies at each m other than
m = 0, only one resonance appears in the figure. Figure 1b provides a zoomed view
of this range. It indicates that |W0 | does not feature a strong resonance. The fluid-
loaded resonance of W1 occurs at ke a = 0. Increasing m beyond one shifts the peak
of |Wm | to a higher frequency, with a narrower width and greater height. (In the
study of mechanical and electrical oscillations, this feature would be said to be that
the Q factor of the resonance increases.) These fluid-loaded resonances occur at
substantially lower frequencies than the corresponding in-vacuo natural frequency.
102
|Wm|(ρe ce2h/F)
m =0 m =1 m=2 m=3 m=4
100
10-2
0 1 2 3 4 5 6 7 8
Nondimensional frequency kea
Figure 1a.
10-6
|Wm|/a
10-8
10-10
0 0.2 0.4 0.6 0.8 1
Figure 1b.
Reasonable questions to ask at this juncture are: Why don’t any of the curves show
two peaks if there are two natural frequencies for each m > 0? Why do peaks occur at
frequencies that are much less than either natural frequency for that m? Why doesn’t
the m = 0 curve have a sharp peak? Why does the m = 1 curve rise without bound
as the frequency approaches zero? The answer to each may be found in the general
analysis, but it is helpful to examine them from a slightly different perspective. Rather
than examining impedances that relate the applied force to velocity coefficients, let
us write Eq. (3) for the displacement coefficients as
2m + 1
(K e + K f ) χm =
4π
The stiffness of an elastic spring is the ratio of the internal force to the elongation.
The preceding describes a similar relation, so K e and K f are, respectively, called
the structural and fluid dynamic stiffnesses. A comparison of the above definition to
Eq. (7.2.50) shows that these quantities are related to the respective impedances by
K e = ike a Z e , K f = ike a Z f
We begin with an examination of the structural reactance. The above definitions show
that the real part of a dynamic stiffness is −ke a times the reactance. The imaginary
part of a dynamic stiffness is ke a times the resistance.
Dissipation
in the shell material
has been neglected, so the sole contributor to Im K e + K f is the fluid resistance,
which means that Im (K e + K f ) > 0 at any frequency.
Let us examine how the behavior of the dynamic stiffnesses affects each χm ,
beginning with m = 2. Fig. 2 depicts the nonzero parts of K e and K f . If fluid loading
was not present, a zero of Re (K e ) would lead to a singular value of χm . Close
inspection shows that this curve does indeed cross the zero axis at ke a = (1) 2 and
(2)
2 . Now suppose that fluid loading was such that Im (K f ) were zero. Then, a singular
value of Wm would occur at any frequency at which Re (K e + K f ) = 0. The plotted
data indicates that this condition occurs at ke a = 0.276, which is much less than
(1) (2)
2 , and at ke a = 2.668, which is slightly less than 2 . Of course, Im(K f ) is not
zero. However, in the vicinity of ke a = 0.276, the value of Im(K f ) is comparable to
the small value of Re (K e + K f ). Consequently, division of the F2 force by K e + K f
yields a large value of |χ2 |. In fact, the maximum is |χ2 | /a = 10.704 at ke a = 0.274.
10
m=2 Re(Ke)
Re(Ke+Kf)
0 Im(Kf)
−10
1 2 3 4 5
Figure 2.
Now consider the situation for χ2 in the vicinity of ke a = 2.668, at which

Re (K e + K f ) = 0. The value of Im (K f ) in this range is much greater than Re(K e +
K f ). This means that radiation damping is high, so there is no frequency in this range
at which |K e + K f | is noticeably diminished. It is for this reason that |χ2 | does not
feature a peak in this range.
The situation for m > 2 is similar to what we have seen for m = 2. The behaviors
for m = 0 and m = 1 are different because of the special nature of the natural fre-
quencies for each case. Figure 3 displays the dynamic stiffnesses for m = 0. There
is only one natural frequency, so there is only one frequency at which Re (K e ) = 0.
The addition of fluid loading leads to Re (K e + K f ) being zero at a slightly lower
value of ke a, but Im (K f ) is quite large in this range. This means that there is no
frequency at which |K e + K f | is greatly reduced, so is there no frequency at which
|χ0 | shows a sharp peak.
10 Re(Ke)
m=0 Re(Ke+Kf)
Im(Kf)
0
−10
0 1 2 3 4 5
Figure 3.
Figure 4 explains the m = 1 behavior. The lower natural frequency is (1) 1 = 0,

and K f = 0 at that frequency because low frequencies correspond to low velocities.
Hence, K e + K f = 0 at ke a = 0, which leads to a singular value of |W1 | /a. The
behavior in the vicinity of the higher natural frequency is the same as it was for m = 2.
Adding K f to K e adds a large imaginary part. Thus, even though there is a frequency
at which Re (K e + K f ) is zero, nowhere in that frequency range is |K e + K f | small.
The consequence is that large values of |W1 | /a only occur at low frequencies.
10 Re(Ke)
Re(Ke+Kf)
Im(Kf)
0
m=1
−10
0 1 2 3 4 5
Figure 4.
For Part (b), we observe that the displacement at polar angles ψ1 = 0 and ψ2 =
π when the frequency is a specific value (ka)n may be computed by the matrix

algorithm. Let [L] be a 2 × (N + 1) array whose elements are L j,m = Pm cos ψ j ,
that is, row j has the Legendre functions at all orders for ψ j . Also, define a rectangular
array [χ] whose columns hold the χm values at each (ka)n . Then, the displacement
series in Eq. (7.2.35) reduces to
ρe ce2 h
[w] = [L] [χ]
F̂
It is at this juncture that the question of series convergence arises. Given that the
Fm coefficients in Eq. (2) increase with increasing m, it is reasonable to anticipate
that the spherical harmonic M at which the summation is halted will be large. If we
perform the first calculation with a large M, then results for smaller M may be found
by deleting columns from the right of [L], and deleting the same number of rows
from the bottom of [χ]. Results for M = 100, 67, and 33 are described by Fig. 5.
There is no perceptible difference between the results for the two longer series. Thus,
it is reasonable to conclude that M = 67 is adequate for the range ke a < 8. The series
for M = 33 agrees well up to ke a = 4, then diverges drastically for ke a > 6. There
is a simple explanation for this behavior. Each χm , other than χ0 , has a narrow peak
in the vicinity of a single frequency. Because the displacement w is formed from a
weighted series of the χm coefficients, w/a may be expected to have peaks at each
of these frequencies. The highest peak for M = 33 occurs at ke a = 5.908, which is
where the series begins to diverge from the correct solution. The conclusion is that
the series length must be sufficiently long to capture all fluid–loaded “resonances”
that occur within the frequency range of interest. Unfortunately, this criterion is not
much use a priori, because we find these frequencies at part of the analysis.
104
M = 100 M = 67 M = 33
|w|(ρe ce2h/F)
102
100
ψ=0o
10-2
104
|w|(ρe ce2h/F )
102
100
ψ=180o
10-2 0 1 2 3 4 5 6 7 8
Figure 5.
A different perspective for convergence results from considering the criterion for
accuracy of the shell theory. It was stated as requiring that m < (π/5) a/ h. For the
present system, this requires m ≤ 25. A graph like Fig. 5 would show that a truncation
of the series at m = 25 would yield a result that is convergent only for ke a < 3.3.
Given that, we should not assume that our results for higher ke a are what would be
observed in an experiment.
The farfield pressure is given by Eq. (7.2.43), which also may be computed as a
matrix product. A rectangular array [C] is computed by multiplying each row [Bm ]
by a phase delay for that m, specifically
[Cm ] = [Bm ] ei(m+1)π/2
The farfield locations at which the pressure is to be evaluated for Part (c) also are
ψ = 0 and ψ = π, so we may also use [L] for this evaluation. Therefore, we evaluate

r Pff,axis = [L] [C]
Because of the manner in which the χm values are defined, the quantities that are
evaluated in this manner are (r/a)Pff (r, ψ = 0 or π) eikr (ah/ F̂).
Figure 6 displays the frequency dependence of the farfield pressures on the axis
of symmetry. A zoomed view gives a clearer picture of the many low-frequency
resonances. The frequencies at which peaks occur are those that maximize one of the
χm coefficients. An interesting attribute is that the radiation in the forward direction,
ψ = 0, is much greater than the backward direction, ψ = 180◦ , for nonresonant ke a
values between 0.5 and 5.
102
(r/a) | pff | (ah/F)
100
10-2
ψ = 0o
ψ = 180o
10-4
0 1 2 3 4 5 6 7 8
102
(r/a) | pff | (ah/F )
100
10-2
10-4
0 0.2 0.4 0.6 0.8 1
Figure 6.
The directivity requested in Part (d) can be found similarly to the farfield
pressure
at a specific angle. Let (ke a) j denote the frequency of interest, and let C j denote
the jth column of [C].

The farfield pressure at many angles ψ j may be found by
defining a matrix
that differs from [L] only by the fact that its has many rows,
L
L j,m = Pm cos ψ j .
A scan of the data shows that the largest value of r |P| for ke a < 1 occurs at ke a =
0.404, and the largest value in the interval ke a < 5 occurs at ke a = 4.935. Polar plots
describing r |P| at these frequencies and ke a = 2.50 constitute Fig. 7. The response at
the lowest and highest frequencies corresponds to peaks of χn for m = 4 and m = 30,
respectively. Consequently, the farfield pressure at these frequencies is proportional to
the Legendre functions P4 (cos ψ) and P30 (cos ψ). In contrast, the middle frequency,
ke a = 2.50, is not close to any resonance. Hence, all spherical harmonics contribute
significantly to the pressure. The consequence is that the directivity plot at ke a = 2.50
does not resemble that of a single Legendre function. It will be noted that the overall
amplitude at ke a = 2.5 is comparable to that at the lower frequency, even though |χ4 |
at ke a = 0.404 is two orders of magnitude greater than any |χm | value at ke a = 2.5.

This result is a consequence of the fact that the h m (ka) appears in the denominator
of Eq. (7.2.43). These values are much larger at the lower frequency. This is another
manifestation of the inefficiency of radiation at low frequencies.
60
kae = 0.404 kae = 2.50
150 30 150 0.4 30
0.5
180 0.4 0 0.3
180 0
0.2
0.8
150 30 150 30
120 60 60
120 60 120
kae = 4.935
150 20 30
10 25
15
180 0
150 30
Figure 7.
7.3 Radiation from an Infinite Cylinder
It is known that the separation of variables approach may be used to solve the
Helmholtz equation in eleven orthogonal coordinate systems, such as the spheri-
cal coordinates of the previous section.11 However, it does not follow that we will
always be able to use the technique to analyze radiation from a body whose shape
11 Morse and Feshbach, Methods of Theoretical Physics, vol. 1 (1953) pp. 494–523, 655–666.
Fig. 7.8 An otherwise rigid y

box undergoing a uniform
vibration at velocity v0
amplitude v0 along the edge x
x = a/2 b
a
matches one of these coordinate systems. The apparent paradox results from the
nature of boundary conditions.
Consider a rectangular box, which is simplified to a two-dimensional system.
Suppose the right edge translates, and the other edges are stationary. This is the
situation in Fig. 7.8.
The origin of x y has been placed at the center. Thus, the boundary conditions are
∂P
= −iρ0 ωV0 on x = a/2, − b/2 < y < b/2
∂x
∂P
= 0 on x = −a/2, − b/2 < y < b/2 (7.3.1)
∂x
∂P
= 0 on y = ±b/2, − a/2 < x < a/2
∂y
The separation of variables ansatz is that P = F (x) G (y). It is possible to find

ordinary differential equations for F (x) and G (y) whose satisfaction assures that P
is a solution of the Helmholtz equation. The difficulty is with the boundary conditions.
Substitution of the ansatz into the second of the preceding set of boundary conditions
gives # $

d F
G (y) = 0, −b/2 < y < b/2 (7.3.2)

dx
x=−a/2
In cannot be that G(y) = 0, so it must be that d F/d x is zero at x = −a/2. However,

if it is zero along this edge, then ∂ P/∂x is zero at x = −a/2 for all y, not just the
edge of the box. In other words, the resulting solution would describe the box plus
a rigid sheet at x = −a/2 whose extent is infinite. Similar situations occur for the
rigid edges at y = ±b/2. Thus, the system that is implied by making the separation
of variables solution satisfying the rigid boundary conditions is the one depicted in
Fig. 7.9. Clearly, this system is not at all like the one we set out to analyze, and it has
no physical significance.
Fig. 7.9 System implied by

rigid
rigid
making a separation of
variables ansatz satisfying rigid rigid rigid
the boundary conditions for rigid rigid
the vibrating box in Fig. 7.8. v0
The rigid lines represent rigid rigid
rigid thin planes that extend rigid rigid rigid
rigid
rigid
to infinity
7.3 Radiation from an Infinite Cylinder 55
Given this observation, it is reasonable to wonder what body shapes can be

analyzed for their radiated field by the separation of variables method? If we add
the requirement that the analysis not be excessively complicated, and that the sys-
tem is reasonably common, the answer is: Very few! Spheres were covered in the
previous section. Variants featuring ellipses, such as spheroids, have been analyzed,
but those studies require great effort because of the nature of the associated separation
functions.
Radiation from cylindrical bodies is of great interest because it is an issue that
arises in several contexts, including HVAC and automotive exhaust system, and
piping and pressure vessel applications. Also, in underwater acoustics, the shape of
the pressure hull of a modern submarine usually is a cylinder closed by rounded
caps. Unfortunately, the separation of variables solution for a cylindrical geometry
does not describe any of these systems. This is so because it is limited to situations
where the cylindrical boundary has infinite length. Otherwise, we encounter the same
difficulty as that in Fig. 7.9, with a cylindrical boundary extended beyond the ends of
the cylinder. Despite this limitation, we will pursue the analysis of infinite cylinders.
In part, the justification for doing so is that the investigation will provide results that
are good approximations in some regions of the actual field. In addition, analyses
like those pursued here often are used to develop simpler models that capture basic
phenomena encountered in finite length cylinders.
The system we will explore is described in Fig. 7.10. The z-axis is defined to
coincide with the axis of a cylinder whose radius is a when the cylinder is at rest.
The fluid domain extends outward from the cylinder’s surface without limit. The
surface of the cylinder executes a small amplitude vibration such that its velocity
component normal to the surface is Re Vs eiωt , where Vs is an arbitrary specified
function of the circumferential angle θ and axial position z. The unit vector ē R at an
arbitrary field point is said to be oriented in the transverse direction relative to the
z-axis. The circumferential direction is ēθ , and the axial direction is ēz . Both ē R and
ēθ depend on θ, but are independent of R and z, whereas ēz has a fixed orientation.
Fig. 7.10 Coordinate -e

systems and unit vectors for R y
describing a surface -
eθ
vibration of a cylinder
-e
z
z
z R
θ
a
x
7.3.1 Separation of Variables
The pressure field radiated by a vibrating cylinder is a function of R, θ, and z.

Because ē R is perpendicular to the surface, the continuity condition requires that the
transverse velocity of the fluid equals Vs . The frequency domain equations governing
the field are the Helmholtz equation subject to a boundary condition derived from
Euler’s equation. The gradient and Laplacian in terms of cylindrical coordinates are
derived in Appendix A. The governing equations are
⎧
∂2 P 1 ∂P 1 ∂2 P ∂2 P ⎨R >a
+ + + + k 2
P = 0, −π ≤ θ < π
∂ R2 R ∂R R 2 ∂θ2 ∂z 2 ⎩

−∞ < z < ∞ (7.3.3)
∂ P
−π ≤ θ < π
= −iρ0 ωVs ,
∂R
r =a
−∞ < z < ∞
We could immediately proceed to apply the method of separation of variables to the

Helmholtz equation. However, the circular geometry allows us to simplify identifi-
cation of the dependence on θ.
If we were to start from any point and follow a circle defined by constant values
of R and z, an increment of θ by 2π would bring us back to the starting point. It
follows that P must be θ-periodic in 2π, that is, P (R, θ, z) = P (R, θ + 2π, z), so
we introduce a Fourier series. The coefficients of this series may be functions of R
and z. It is convenient to begin by using a complex series, so we know that
∞
1
P= Fn (R, z) einθ (7.3.4)
2 n=−∞
Because of the linearity of the Helmholtz equation, each term of the Fourier series
must constitute a solution. If we substitute a generic term, the harmonic function of
θ appears throughout and may be eliminated. What remains is

∂ 2 Fn 1 ∂ Fn ∂ 2 Fn n2
+ + + k 2
− Fn = 0 (7.3.5)
∂ R2 R ∂R ∂z 2 R2
At this juncture, we introduce the separation of variable ansatz,
Fn = f (R) g (z) (7.3.6)
We substitute this form into the differential equation, divide by f g, then group terms
that are functions of R and z on either side of the equality. The k 2 term could be
placed on either side, but it preferable to keep it with the 1/R 2 term. Thus, these
operations lead to

1 d2 f 1 df n2 1 d2g
+ + k − 2 =−
2
(7.3.7)
f d R2 R dR R g dz 2
The terms on the left are a function of R, while the one on the right depends only on z.
This can only be true if both sides equal the same constant. The value of this constant
is not known a priori, but we do know that g (z) cannot grow as |z| increases. This
requires that the separation constant be positive or zero, because a negative value
would lead to solution for g that is a growing exponential. Thus, we shall denote the
separation constant as μ2 . The corresponding separated differential equations are

d2 f 1 df n2
+ + κ 2
− f =0
d R2 R dR R2
d2g (7.3.8)
+ μ2 g = 0
dz 2
κ2 = k 2 − μ2
The axial function g (z) is harmonic. In keeping with the usage of a complex
Fourier series for the radial distance, this function is written in a similar form as
g = Ae−iμz + Be+iμz (7.3.9)
The values of the axial wavenumber μ and the coefficients A and B are arbitrary in
regard to satisfaction of the Helmholtz equation.
7.3.2 Transverse Dependence—Cylindrical Bessel Functions
The value of μ affects f (R) by altering the constant κ. To begin, we will consider μ
to be less than k, so that κ is a real number, which may be taken to be positive. The
case where μ > k is important and will be studied after the primary development.
The differential equation for f (R) is reminiscent of the equation governing the
radial function for spherical waves. However, the absence of a two factor in the
second term, (1/R) d f /d R, is important. It is possible to derive a transformation
that converts the present equation to the spherical form. An alternative would derive
the solution by following the method of Frobenius, which modifies a power series
in R with a term that is an unspecified fractional power. Some readers might have
encountered a differential equation like this in other subjects. In any event, the present
endeavor is well served by merely observing that the first of Eq. (7.3.8) is Bessel’s
equation. Its solutions are Bessel and Neumann functions. We encountered them in
the previous section, where it was noted there that standard algorithms compute these
functions as intermediate steps in the evaluation of spherical Bessel functions.
A rigorous terminology would refer to these functions as cylindrical Bessel func-
tions, but their occurrence is so common that the adjective “cylindrical” usually is
dropped. To distinguish them from the spherical functions, capital letters are used for
the cylindrical functions: J for the Bessel function, N for the Neumann (although
Y also is often used), and H for the Hankel functions that we soon will define. The
value of n, which is the order of the function, appears in the subscript. Furthermore,
if we were to divide the first of Eq. (7.3.8) by κ, we would see that the independent
variable for this form of Bessel’s equation is κR, so that is the argument of the func-
tions. Among the many ways to evaluate Bessel functions, the least useful is from
their definitions. Nevertheless, they are listed here for the sake of completeness,
(−1)k z 2k+n
∞

Jn (x) =
k=0
k! (n + k)! 2
2 z 1 (n − k − 1)! z 2k−n
n−1
Nn (x) = ln Jn (x) − (7.3.10)
π 2 π k=0 k! 2
1 ∞
(−1) k z 2k+n
− [ (k + 1) + (n + k + 1)]
π k=0 k! (n + k)! 2
In the second expression, ( ) is the psi or digamma function, which is defined in

standard references on special functions.12
Both Jn (x) and Nn (x) are real. Therefore, neither by itself can represent a wave
propagating in the transverse direction. For spherical waves, the spherical Hankel
functions fit this specification. The same is true here. Hankel functions of the first
and second kind and order n are defined to be
Hn(2) (x) = Hn(1) (x)∗ = Jn (x) − i Nn (x) (7.3.11)
As was true for spherical waves, the second Hankel function represents an outgo-
ingwave at
large k R corresponding to our representation of harmonic response as
Re Peiωt . This fact is seen when we consider the asymptotic behavior of each
function as its argument κR increases,
1/2 ⎫
∗ 2 ⎪
−i(κR−(2n+1)π/4) ⎪
Hn(2) (κR) = Hn(1)
(κR) → e ⎪ ⎪
πκR ⎪
⎪
1/2 ⎪
⎬
2 2n + 1
Jn (κR) → cos κR − π as κR → ∞
⎪
πκR 1/2 4 ⎪
⎪
⎪
2 2n + 1 ⎪
⎪
Nn (κR) → sin κR − π ⎪
⎭
πκR 4
(7.3.12)
The Bessel and Neumann functions will arise when we explore cylindrical cavities
and waveguides, but we will have no need for the first kind Hankel function. Thus,
12 M.I. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, 9th ed., Dover, Sect. 6.3.
(1965).
we will drop the superscript “two” notation for acoustical analyses, as we did for
spherical waves, so that Hn (k R) will always refer to the second Hankel functions.
The behavior of the cylindrical Bessel functions at small arguments also is impor-
tant. The Bessel function is finite at x = 0, but the Neumann and Hankel functions
are singular. The trends are
⎧ ⎫
⎪ 2i ⎪
⎨ − ln (x) if n = 0 ⎪
⎪
⎪
Hn(2) (x) = Hn(1) (x)∗ → i π n
2
⎪
⎪
⎪
⎪
⎩ (n − 1)! if n ≥ 1 ⎪
⎪
⎪
⎪
π x ⎪
⎪
⎪
⎬
1 x n
Jn (x) → as x → 0 (7.3.13)
n! 2 ⎪
⎪
⎧ ⎪
⎪
2 ⎪
⎪
⎪
⎨ ln (x) if n = 0 ⎪
⎪
⎪
⎪
Nn (x) → π 1 n
2
⎪
⎪
⎪
⎪
⎩ − (n − 1)! if n ≥ 1 ⎪
⎭
π x
The limiting behaviors for small and large arguments are evident in Fig. 7.11. The
Bessel functions are finite at x = 0, with all except the zero order being zero there.
In contrast, all Neumann functions are negatively infinite at the origin, with a growth
rate that increases with increasing order. For large arguments, a Neumann function
resembles the Bessel function at that order with a lag of π/2. Both functions decay
with increasing values of their argument at fixed order. However, the decrease is
inversely proportional to the square root of their argument, which is a slower decay
than the decay rate for the spherical Bessel functions.
1
J0(x) J1(x) J2(x) J3(x)
0.5
−0.5
N0(x) N1(x) N2(x) N3(x)

0.5
−0.5
−1
0 5 10 15 20
x
Fig. 7.11 Low-order Bessel and Neumann functions as functions of κR

The order n affects the magnitude of the argument x at which it is appropriate

to employ Eq. (7.3.12). Specifically, the large argument trends will be observed if
x > n + 2π. The opposite trend is the behavior when the order is much greater than
the argument. In this case, we have
⎫
1 1/2 eκR n ⎪⎪
Jn (κR) → ⎬
2πn
1/2 2n−n if n κR (7.3.14)
2 eκR ⎪
⎪
Nn (κR) → − ⎭
πn 2n
Most standard mathematics software contains routines for evaluating the Bessel
and Neumann functions. We also will need to evaluate their derivatives. As always, a
prime denotes differentiation of a function with respect to its argument. The following
recurrence relations resemble those for spherical Bessel functions. Those for the
derivatives will be useful when it is necessary to satisfy velocity boundary conditions,
whereas the relations featuring functions at different orders are a core component of
most numerical algorithms by which the functions are evaluated. In the following,
Fn (x) represents any of the cylindrical functions.
2n
Fn (x) = Fn−1 (x) + Fn+1 (x)
x
1
Fn (x) = Fn−1 (x) − Fn+1 (x)
2
n
= Fn−1 (x) − Fn (x)
x
n
= −Fn+1 (x) + Fn (x)
x (7.3.15)
F0 (x) = −F1 (x)
F−n (x) = (−1) Fn (x)
n
j
d 1 j!
Fn (x) = j Fn− j (x) − Fn− j+1 (x)
dx j 2 1! ( j − 1)!

j!
+ Fn− j+2 (x) − · · · + (−1) j Fn+ j (x)
2! ( j − 2)!
Many other relations featuring the various Bessel functions are available.13
7.3.3 Radiation Due to a Helical Surface Wave
The first case we shall consider is that in which the surface motion consists of a
single term in the complex Fourier series for the circumferential dependence, and
13 M.I. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, (1965) Chaps.
9–11.
Fig. 7.12 Depiction of a (a) s=aθ (b)

helical wave vector k̄s . a μ
helix
Unwrapped cylindrical
surface. b Tangent to the - -
helix n/a ks
e-θ
ks (n/a)e-θ
e-z μe-z
z
the z dependence is a sinusoidal wave that propagates in the positive direction. In

other words, the radial velocity on the surface is taken to be

vs = Re V0 ei(ωt−μz−nθ) (7.3.16)
Clearly, this represents a wave, but in what direction is it propagating? To answer

this, note that the arclength along the cylinder in the circumferential direction is
s = aθ. Let us unwrap the surface of the cylinder and lay it on a plane, as is shown in
Fig. 7.12a. Position along this flat surface is x̄ = z ēz + s ēθ , where the unit vectors are
those in Fig. 7.10. This suggests that we rewrite the phase variable of vs as ωt − k̄s · x̄,
where the surface wavenumber vector is
n
k̄s = μēz + ēθ (7.3.17)
a
This vector forms a constant angle with ēz , which means that on this plane the wave
follows a straight line. When the plane is wrapped back onto the cylinder in Fig. 7.12b,
this straight line becomes a helix. Correspondingly, Eq. (7.3.16) is said to be a helical
surface wave. In the case where n = 0, the surface wave is axisymmetric and the
wavenumber vector is parallel to the axis of the cylinder.
In addition to the boundary condition at the cylinder’s surface, the acoustic
response must be an outgoing wave having suitably decaying amplitude as R
increases. The only radial dependence fitting this specification with our use of the eiωt
convention is the Hankel function of the second kind. Furthermore, because V̄ · ē R
is obtained from a derivative of P with respect to R, the manner in which pressure
depends on z and θ must match the dependence of Vs . Thus, we shall try
1/2
P = B Hn (κR) e−i(μz+nθ) , κ = k 2 − μ2 (7.3.18)
The boundary condition is

∂ P
∇ P · ē R | R=a ≡ = −iωρ0 Vs = −iωρ0 V0 e−i(μz+nθ) (7.3.19)

∂ R
R=a
Our ansatz will satisfy this condition if
BκHn (κa) = −iωρ0 V0 ≡ −ik (ρ0 cV0 ) (7.3.20)

The corresponding pressure and particle velocity expressions are
k Hn (κR) −i(μz+nθ)
P = −iρ0 cV0 e
κHn (κa)

1 ∂P 1 ∂P ∂P
V̄ = − ē R + ēθ + ēz (7.3.21)
iωρ0 ∂ R R ∂θ ∂z
n H (κR)
Hn (κR) n
= V0
ē R − i ēθ + μēz e−i(μz+nθ)
Hn (κa) R κHn (κa)
An important aspect of this expression is that n may be positive, negative, or

zero. Because of the way the Hankel, Bessel, and Neumann functions are defined,
functions of negative order are related to the positive order functions by an alternating
sign, specifically
H−n (x) = (−1)n Hn (x) , J−n (x) = (−1)n Jn (x) , N−n (x) = (−1)n Nn (x)
(7.3.22)
The sign change has no effect on Eq. (7.3.21) because the terms are ratios of a Hankel
function and its derivative at the same order.
The farfield decay of a cylindrical wave is different from the spherical spreading
of the field that radiates from a finite body. To see why, suppose we select a field
point at a very large value of R. The line from the field point to points in the cylinder
cross section at the same z is perpendicular to the cylinder’s axis. In contrast, lines
from the field point to the distant ends of the cylinder are essentially parallel to the
z-axis. There is no radial distance at which all lines from the field point to the surface
are parallel, so the prior theorems regarding farfield behavior do not apply.
To identify the farfield limit, we employ the first of Eq. (7.3.12) to write
1/2
kei(2n+1)π/4 2
Pff = −iρ0 V0 e−i(κR+nθ+μz) (7.3.23)
κHn (κa) πκR
The particle velocity in the farfield is obtained by substituting Pff into Euler’s equa-
tion. The fact that R is large renders the circumferential velocity negligible compared
to V̄ · ē R and V̄ · ēz . The result is
Pff κ μ
V̄ff = ē R + ēz (7.3.24)
ρ0 c k k
The relation of pressure and particle velocity in a plane wave is V̄ = (P/ρ0 c) k̄/k.
The above expression suggests a similar form. The phase variable of the farfield pres-
sure, Eq. (7.3.23), is κR + nθ + μz. It may be expressed in terms of a wavenumber
vector k̄ if we let x̄ be the position off the cylinder after it is unwrapped, that is,
x̄ = R ēr + s ēθ + z ēz . Then, setting k̄ · x̄ = κR + nθ + μz leads to
n
k̄ = κēr + ēθ + μēz (7.3.25)
R
κe-R
Fig. 7.13 Wavenumber
(n/3a)e-θ
vectors for the acoustic wave -
μe
generated by a helical wave z
on the surface of a cylinder
- κe-R
μe (n/2a)e-θ
z
3a (n/a)e-θ
2a μe-z
θ a
Figure 7.13 shows k̄ at two locations along a transverse line, as well as k̄s for the
surface wave. (In a strict sense, we should not depict k̄ close to the surface unless
κa 1, because Eq. (7.3.23) is only valid if κR 1.)
At the surface, the portion of k̄ that is parallel to the surface, that is the ēθ
and ēz components, matches the surface wavenumber k̄s . In other words, the trace
wavenumbers match, and the surface wave is supersonic. At very large R, the ēθ
component of k̄ is negligible. Thus, Eq. (7.3.24) states that

Pff k̄
V̄ff = lim (7.3.26)
ρ0 c R→∞ k

Furthermore, it follows from the definition of κ that

k̄
→ k as R → ∞. In other
words, in the farfield, the wave propagates in the plane formed by the field point and
the cylinder’s axis. Over short propagation distances, it seems to be locally planar,
but cylindrical spreading is observable for large changes in R.
Whether a surface wave results in radiation depends only on the scale of the
axial dependence. The condition that μ < k for a wave to radiate to the farfield
is equivalent to saying that the axial wavelength 2π/μ is greater than the acoustic
wavelength 2π/k, or equivalently, that the axial propagation speed cs exceeds the
speed of sound, c ≡ ω/k. Each is a manifestation of trace matching. It is evident
that the circumferential component of k̄s is not relevant to the matching of trace
velocities. In essence, as waves propagate along the cylinder, the axial features do
not change, whereas the surface falls away in the circumferential direction. Thus, the
circumferential effect is local, whereas the axial effect is global. (This is the reason
why the phenomenon of matching trace velocity is not evident in radiation from
spheres.)
Subsonic Surface Waves—Modified Bessel Functions
Recognition that the axial trace velocity of the acoustic wave must equal the axial
speed of the surface wave leads to the same question as that for a surface wave on a
plane, specifically, what if the surface wave is subsonic? The axial component of the
surface wave’s phase velocity is ω/μ. The supersonic case is obtained if μ < k, in
which case, the value of κ2 is positive. Conversely, a subsonic wave corresponds to
μ > k, in which case, κ2 is negative. When this situation occurs on a planar boundary,
the result is an evanescent acoustic wave. The same is true here, but the fact that the
dependence on R is described by a Hankel function is a slight complication.
To make it clear that κ2 is negative, we replace it with −β 2 , with β > 0. Then,
the differential equation resulting from the separation of variables procedure is

d2 f 1 n2
+ − β + 2 f = 0, β 2 = μ2 − k 2
2
(7.3.27)
d R2 R R
This is a modified Bessel’s equation, and its solutions are the real functions K n (β R)
and In (β R), which, respectively, are the modified Bessel functions of the first and
second kind. The K n (β R) function is singular at the origin and decays to zero as
β R → ∞. In contrast, In (β R) is finite at β R = 0 and grows without bound as
β R → ∞. The dependence of the lowest order functions is depicted in Fig. 7.14.
10
8 K0 (x)
K1 (x)
6
K2 (x)
4
K3 (x)
2
0
8 I0(x)
I 1(x)
6
I2(x)
4
I3(x)
2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
x
Fig. 7.14 Modified Bessel functions for orders zero to four
Asymptotic approximations of the modified functions are

n ⎫
1 2 ⎪ ⎪
K 0 (x) → − ln (x) , K n (x) → (n + 1)! ⎬
2 x as x → 0
1 x n⎪⎪
In (x) → ⎭ (7.3.28)
(n+ 2)! 2
π 1/2 1 1/2 x
K n (x) → e−x , In (x) → e as x → ∞
2x 2πx
The recurrence relations for these functions are slightly different from those for the
regular functions. In the following, Fm (x) is either In (x) or (−1)n K n (x),
2n
Fn (x) = Fn−1 (x) − Fn+1 (x)
x
1
Fn (x) = Fn−1 (x) + Fn+1 (x)
2
n (7.3.29)
= Fn−1 (x) − Fn (x)
x
n
= Fn+1 (x) + Fn (x)
x
K 0 (x) = −K 1 (x) , I0 (x) = I1 (x)

Although both In (β R) and K n (β R) satisfy Eq. (7.3.27), the former grows expo-
nentially with increasing R. This behavior is counter to the Sommerfeld radiation
condition, so it is discarded. The modified function of the second kind decays expo-
nentially as R increases, so it represents an evanescent wave. Thus, the pressure
generated from a high wavenumber on the surface is
Pev = B K n (β R) e−i(μz+nθ) , μ > k (7.3.30)
The corresponding particle velocity is
B n
V̄ = − β K n (β R) ē R − i K n (β R) ēθ + μēz e−i(μz+nθ) (7.3.31)
iωρ0 R
The transverse component must match the surface velocity in Eq. (7.3.16), which
sets B. When that expression is substituted into the above general solution, the result
is
k K n (β R) −i(μz+nθ)
Pev = −iρ0 cV0 e (7.3.32)
β K n (βa)
The fact that subsonic surface waves do not radiate to the farfield is the same
property as subsonic waves on the planar boundary of a half-space. Another aspect
becomes evident when we recognize that the modified Bessel functions are real. This
means that the transverse component of velocity, V̄ · ē R , is everywhere 90◦ out-of
phase from the pressure. One consequence is that power in not radiated when μ > k.
It also has implications for fluid–structure interaction, which is touched on in the
next example.
In some applications, such as computer programs, it is inconvenient to monitor
whether μ > k, and then select the appropriate description of P. In the case of waves
on a planar surface, we found that we could use the equations for supersonic waves
to describe the subsonic case. With the objective of doing so here, we note that the
modified functions are related to the regular Bessel functions, which should not be
surprising in view of the fact that the equations they satisfy only differ by a sign. The
relations are
In (x) = (−i)n Jn (i x)
π (7.3.33)
K n (x) = (−i)n+1 Hn (−i x)
2
In the present context, the values of x of interest are βa and β R. By definition,
κ = ±iβ. If we use the positive sign, then Hn (κR) will become Hn (iβ R), in which
the above relation indicates to be proportional K n (−β R). We wish that Hn (κR)
convert to K n (+β R), so we set κ = −iβ for subsonic surface waves. Thus, we
may employ the supersonic solution, Eq. (7.3.21), to treat any helical surface wave,
provided that the transverse wavenumber is set according to
1/2
k 2 − μ2 if μ < k
κ= 2
2 1/2
(7.3.34)
−i μ − k if μ > k
Of course, this scheme is only useful if the computational routine that evaluates
Hankel functions allows the argument to be imaginary.
The alteration from a regular to a modified Bessel function has an important
influence on the radiation impedance, which is a measure of how effectively a surface
motion generates a pressure wave. The impedance of a locally reacting surface is the
ratio of the surface pressure to the inward normal surface velocity. In contrast, the
radiation impedance is the proportionality of the surface acoustic pressure resulting
from a known outward surface velocity. The relation is Z rad = P/V̄s · n̄. Despite the
apparent similarity in their definitions, the two types of impedances are fundamentally
different. A local impedance is a property of the surface, such as its viscoelastic
properties. The radiation impedance is an acoustical property that depends on the
shape of the radiating body and the nature of the surface motion. If we know it as
a function of the fundamental parameters of the system, such as the wavenumbers
and frequencies of the surface motion, it may be used to describe the behavior in a
variety of situations.
The radiation impedance of a helical wave is obtained by evaluating the pressure at
R = a, then dividing it by the surface velocity. The velocity is given by Eq. (7.3.16).
The surface pressure in the case of a supersonic propagation in the axial direction is
found by setting R = a in Eq. (7.3.21), whereas setting R = a in Eq. (7.3.32) gives
the surface pressure in the subsonic case. The radiation impedance that results is
given by
⎧
⎪
⎪ ka Hn (κa)
⎨ −i if μa < ka
Z rad κa Hn (κa)
= ka K n (βa) (7.3.35)
ρ0 c ⎪
⎪
⎩ −i if μa > ka
βa K n (βa)
The nondimensional form of this expression tends to facilitate numerical evaluations,

which are conducted in the next example.
A fundamental difference between the supersonic and subsonic cases is evident in
this expression for Z rad . The regular Hankel function is complex, which means that
Z rad has both a resistive and a reactive component. The resistive part is associated
with transfer of power from the surface into the fluid. In contrast, the modified
Bessel functions are real, so Z rad is imaginary for subsonic waves. In other words,
it is purely a reactance. Because the pressure is 90◦ out-of-phase from the velocity,
the time-averaged power flow into the fluid will be zero.
EXAMPLE 7.6 The task here is to determine the radiation impedance asso-
ciated with a helical surface wave. Evaluate and graph the result as a function
of μa, with ka = 2 and 10, and n = 0, 2, and 4.
Significance
If the radiation impedance is known at the outset of the analysis, it can be used to
describe the pressure effect on the structure without performing a parallel acoustical
analysis. The parameter dependencies that will be disclosed in the course of this
example have much significance.
Solution
How we proceed depends on whether our computational software can evaluate
Hankel function for an imaginary argument. If so, then the field for any set of
wavenumbers may be computed by employing Eq. (7.3.34) in conjunction with the
first of Eq. (7.3.35) to evaluate Z rad for any value of μ. If not, then both parts of
Eq. (7.3.35) must be constructed explicitly. A vector {μa} that covers the full range
of axial wavenumbers m consists of two partitions. The upper vector {κa} corre-
sponds to values of m for which μm a > κa, so its elements are defined according to
Eq. (7.3.18). The lower vector {βa} corresponds to μm a > κa, so Eq. (7.3.27) defines
its elements. A vector {Z rad } then is obtained by stacking the values obtained from
{βa} below those obtained from {κa}.
The Hankel function is complex, so Z rad for supersonic waves is complex. In
contrast, the modified Bessel functions are real and positive, and their derivative is real
and negative. Therefore, Z rad for subsonic waves is always positive imaginary, that is,
it is an inertance. Figure 1 shows the low-frequency behavior. There is a singularity
at μ = k for n = 0, but not for n > 0. This difference stems from the behavior of
the Bessel functions as their argument approaches zero. For n = 0, H0 (κa) has a
logarithmic singularity as κa → 0. Equation (1) indicates that Z rad in that case is
proportional to ln(κa). The behavior for βa → 0 is similar. In contrast, for n > 0,
the trend is that Hn (κa) is proportional to (κa)−n as κa → 0. The result is that Z rad
is finite in the limit. Another interesting aspect is that Im (Z rad ) > 0 for all values
of μ, which means that its reactive part always is an inertance. We also see that the
impedance at any value of μa decreases as n increases.
Im(Zrad)/(ρ0 c) Re(Zrad)/(r0c)
4 ka = 2
n=0
2 n=2
n=4
0
0
0 1 2 3 4
Axial wavenumber μa
Figure 1.
Increasing the frequency changes some of these trends. The singular behavior in
Fig. 2 around μ = k is like Fig. 1, but both parts of Z rad for n > 0 are maximized
slightly below μ = k. Also, the intertances above μ = k are greater than they were
at the lower frequency.
Im(Zrad)/(ρ0c) Re(Zrad)/(ρ0c)
4 ka = 10
n=0
2 n=2
n=4
0
2
0
0 5 10 15 20
Axial wavenumber μa
Figure 2.
From the viewpoint of the structure, a large value of |Z rad | means that any move-
ment will be strongly opposed by the fluid. Larger values of |Z rad | in the vicinity of
μ = k mean that it would be more difficult to coerce the structure to move in the
transition from supersonic to subsonic surface waves.
7.3.4 Axially Periodic Surface Vibration
A helical wave is a building block from which more complicated patterns of surface
vibration may be synthesized. How that synthesis is implemented depends on the
extent of the vibration along the axis of the cylinder. The simpler situation is that
of an axially periodic vibration. In this case, the surface velocity may be described
by combining a Fourier series for the z dependence with a Fourier series in θ. This
pattern is replicated by the pressure. The more difficult situation to analyze is that in
which the surface vibration covers a finite length of the cylinder, with the remainder
of the surface being quiescent. That case is taken up in the next section.
Equation (7.3.4) described the surface velocity as a Fourier series composed of cir-
cumferential harmonics. Our interest here is the situation where the surface velocity
also is periodic over an axial distance L. Such a condition requires that the coefficient
of each circumferential harmonic has that period in the axial direction. Thus, the sur-
face velocity may be described by a double Fourier series, with coefficients denoted
as Vm,n to indicate the associated axial harmonic m and circumferential harmonic n.
Hence, the surface velocity in the transverse direction is represented as
∞ ∞
1
Vs = Vm,n e−i(2mπx/L+nθ) (7.3.36)
4 n=−∞ m=−∞
This velocity distribution is a superposition of helical surface waves whose

wavenumbers are μ = 2mπ/L in the z direction and n in the θ direction. The corre-
sponding pressure is obtained by adding the contribution associated with each helical
wave, which is given by the first of Eq. (7.3.21) for supersonic waves, and Eq. (7.3.32)
for subsonic surface waves. To distinguish the two in the summation, we define a
cutoff axial harmonic that marks the highest axial harmonic for which μ < k. This
value is
kL
M = floor (7.3.37)
2π
The axial harmonics for m ≤ M are supersonic, so they form the radiated pressure
Prad , while the higher harmonics form the evanescent field Pev ,
P = Prad + Pev
∞ M
1 k Hn (κm R) −i2mπz/L −inθ
Prad = − iρ0 c Vm,n
e e
4 n=−∞m=−M
κ m Hn (κm a)
∞ ∞

1 k K n (βm R) −inθ
Pev = − iρ0 c Vm,n e−i2mπz/L + V(−m),n ei2mπz/L e
4 n=−∞m=M+1
β K n (βm a)
(7.3.38)
A subscript m has been assigned to the wavenumbers of the Bessel functions to

indicate that they depend on the axial harmonic number. Specifically,
1/2 2 2 1/2
4m 2 π 2 4m π
κm = k 2 − , βm = − k 2
(7.3.39)
L2 L2
The signal generated by a subsonic surface wave does not reach the farfield. The
farfield representation is obtained by dropping Pev and applying Eq. (7.3.12) to
approximate Hn (κn R). This leads to
∞ M 1/2
1 kVm,n 2 a 1/2 i(2n+1)π/4
Pff = − iρ0 c
e
4 κ H (κm a) πκm a
n=−∞m=−M m n
R
× e−i(κm R+nθ+2mπz/L)
(7.3.40)
Equation (7.3.36) describes Vs as a superposition of helical waves that propagate
in the positive z direction (m > 0) and negative z direction (m < 0), with each set
consisting of waves that twist in the direction of increasing θ (n > 0) and decreas-
ing θ (n < 0). Some individuals prefer to describe the dependencies on θ and z as
a superposition of standing waves, rather than propagating helical waves. Such a
form may be obtained by applying Euler’s identity to the complex exponentials. The
result would be four terms that are either cos (nθ) or sin (nθ) multiplied by either
cos (2mπx/L) or sin (2mπx/L).
It is sufficient to determine the time-averaged power radiated from a segment of
the cylinder whose length is L because the spatial field is replicated over this distance.
An evaluation of radiated power begins by surrounding the radiating cylinder with an
infinitely long concentric cylindrical surface. The radius of the surrounding cylinder
is taken to be sufficiently large that the farfield approximation applies. The outward
normal to the surrounding cylinder is ē R , so we must construct (I R )av . The farfield
pressure is described by Eq. (7.3.40). When the terms in that series are denoted as
(Pff )m,n , the particle velocity for each term is described by Eq. (7.3.24). The transverse
velocity component is associated with radiation, so we have
∞
1
M ∞ M κ
(Pff )m,n (Pff )∗, j
m
(I R )av = Re (7.3.41)
2ρ0 c n=−∞ m=−M j=−∞ =−M k
The time-averaged radiated power crossing the surrounding cylinder over an axial
distance L is given by
L/2 π
P= (I R )av (Rdθ) d x (7.3.42)
−L/2 −π
Because the θ and z exponentials in a (Pff )m,n term constitute an orthogonal set, the
only terms that give a nonzero contribution are those for which = m and n = j.
This reduces the description to a double sum, which is i.e. P = 1/8, etc.
∞ (k/κm )2

2
M
1
P = ρ0 cL Vm,n (7.3.43)
8 H (κm a)2
n=−∞m=−M n
There is no cross-coupling between individual terms having different axial or

circumferential wavenumbers. In other words, each helical surface wave contributes
independently to the radiated power. If we had chosen to describe the circumferential
variation of the surface velocity in terms of a real Fourier series, we would have found
that the individual sin (nθ) and cos (nθ) terms do not couple with the radiated power.
The expression for P is consistent with the earlier theorem regarding radiated
power. The region contained between the radiating and surrounding cylinders is not
dissipative. Hence, the time-averaged power that flows into the fluid at the surface
must flow across the surrounding cylinder. The pressure and radial velocity spread
as 1/R 1/2 , so the radial intensity is proportional to 1/R. The area of the surrounding
cylinder over a spatial period L is 2π R L, so the power that flows across the cylinder
over a period does not depend on R.
Axisymmetric Cylindrical Waves
The n = 0 circumferential harmonic is descriptive of the vibration of a tube that con-
tains a flowing fluid. If the tube is filled with a fluid, and a pressure wave propagates
through that fluid, the walls will be pushed outward, thereby inducing an axisym-
metric motion of the walls. All points in the cylinder’s surface at a specific axial
position z share the same radial velocity, so the pressure field is axisymmetric. The
following study of the n = 0 harmonic shows the close relation of the result to the
field of an infinite line source.
The surface normal velocity of an axisymmetric wave propagating in the z direc-
tion is Vs = V0 e−i2π Z /L . The corresponding pressure signal may be obtained from
Eq. (7.3.38) by letting m = 1 and setting V0,1 = 4V0 as the sole nonzero coefficient.
(In the case, where the wave propagates in the negative z direction, the m = −1
velocity coefficient V0,−1 would be the nonzero term.) The farfield pressure obtained
by dropping all terms other than n = 0 in Eq. (7.3.38), then setting V0,1 = 4V0 , is

ka H0 (κ1 R) −i2πz/L
P = −iρ0 cV0 e (7.3.44)
κ1 a H0 (κ1 a)
It is instructive to examine the farfield behavior as a function of frequency. There

are two independent parameters to consider: ka and L/a. Fig. 7.15 is the result of a
calculation for four values of L/a.
(R/a)1/2|Pff|/(ρ0cV0)
101
10 0
L/a = L/a = 40 L/a = 4 L/a = 1

10-1
0 2 4 6 8 10
Frequency ka
Fig. 7.15 Normalized farfield pressure radiated by a supersonic axisymmetric surface wave prop-
agating axially along an infinite cylinder
The quantity that is plotted is (R/a)1/2 |Pff | /(ρ0 cV0 ), which is analogous to
the manner in which the farfield of a spherical wave is described, except for the
difference in the spreading factor. The minimum plotted frequency for each L/a
value is ka = 2πa/L, because no signal radiates to the farfield below that cut-
off. At high frequencies, the plotted quantity approaches one. The singularity at
ka = 2πa/L stems from Eq. (7.3.40) having (κ1 a)3/2 H0 (κ1 a) in the denominator.
If ka → 2πa/L, then κ1 a → 0. Differentiation of the first of Eq. (7.3.13) shows that
H0 (κ1 a) → −2i/(πκ1 a) for κ1 a 1. This means that the denominator approaches
zero as the frequency is reduced to the cutoff value. This singularity means that if
we were to induce a vibration amplitude that does not vary with frequency, then a
very large pressure would be obtained. The opposite view is that a large radiation
impedance indicates that a very large force would be required to sustain the vibration
amplitude at a constant value.
The result for infinite L at low frequencies is not singular. The behavior for
L/a = 40 appears to be a transition between the extremes of infinite and small axial
wavelengths. An infinite value of L corresponds to a field that is the same at all z.
Because this field only depends on R and θ, it represents a two-dimensional model.
Such models often are used to test concepts and examine physical phenomena. Thus,
it is useful to examine this case. We set 2π/L = 0 and κ1 = k in Eq. (7.3.38), which
gives
H0 (k R)
P = −iρ0 cV0 (7.3.45)
H0 (ka)
There is no axial wave in this case, so there is no minimum frequency for radiation to
the farfield. When H0 (k R) is replaced in the above with its approximation for large
k R, and H0 (ka) is approximated as 2i/(πka) for small ka, we find that the farfield
pressure is proportional to (ka)1/2 , rather than the 1/ (κ1 a)1/2 proportionality when L
is finite and ka → 2π/L. Thus, in the two-dimensional case, (R/a)1/2 |P| / (ρ0 cV0 )
approaches zero as ka → 0, whereas the finite L case shows a singularity at the
lowest ka.
We obtained a point source by shrinking a radially vibrating sphere to zero radius.
Similarly, we obtain an infinite line source by shrinking the two-dimensional vibrat-
ing cylinder to zero radius with the volume velocity per unit length held fixed. The
complex amplitude of the radial velocity on the surface is V0 . If we consider a unit
length of the cylinder, the surface area is 2πa, so the volume velocity per unit length
is Q̂ s = 2πaV0 . We use this definition to replace V0 , and also invoke the small ka
approximation of H0 (ka), see Eq. (7.3.13). Doing so converts the two-dimensional
solution to the line source pressure,
1
P= ρ0 ω Q̂ s H0 (k R) (7.3.46)
4
Recall that the free-space Green’s function was defined in Eq. (6.4.29) to be the
solution of the Helmholtz equation with a Dirac delta inhomogeneous term, with the
additional specification that it has unit mass acceleration. The mass acceleration per
unit length of the line source is iωρ0 Q̂ s , so the two-dimensional free-space Green’s
function is
1
G (x̄, x̄s ) = H0 (k R)
4i 1/2 (7.3.47)
R = |x̄ − x̄s | = (x − xs )2 + (y − ys )2
It often is more convenient to use the monopole amplitude of a line source, rather
than the volume velocity. The monopole amplitude A for a point source was defined
to be the coefficient of the position-dependent terms. We adopt the same definition
here, so that the line source pressure may be written in either of two forms,
P = iωρ0 Q̂ s G (x̄, x̄s ) = AH0 (k R)

1 (7.3.48)
Q̂ s = 2πaV0 , A = ωρ0 Q̂ s
4
This development is important because concepts and procedures that feature

Green’s functions for three-dimensional situations are equally applicable for a two-
dimensional model. For example, we may employ the method of images to analyze
the field radiated by a line source above a rigid or pressure-release plane.
EXAMPLE 7.7 The surface of a cylinder whose length is L undergoes a uni-

form radial vibration, such that v R = Re v0 eiωt everywhere. The cylinder has
finite length L, and its radius is very small relative to the acoustic wavelength
2π/k. The origin of the x yz coordinate system is situated at the center of the
cylinder, with the z-axis coincident with the cylinder’s centerline. Derive an
expression for the pressure at an arbitrary field point. Use this expression to
evaluate |P| as a function of distance from the centerline along the x-axis. Con-
sider k L = 0.5, 5, and 50, and compare each result to the field of an infinite
line source.
Significance
This is one of the few problems featuring a finite length cylinder that may be solved
without recourse to advanced numerical techniques. In addition to providing a quan-
titative picture that suggests when it might be acceptable to take a cylinder’s length
to be infinite, it is a useful reminder of the technique for superposing point source
fields.
Solution
The cylinder reduces to a line source when its radius is much less than a wave-
length. The case of an infinite length is described by Eq. (7.3.48), which may be
applied directly. To evaluate the finite length cases, we return to the concept of a
source distribution. The cylinder at zero radius may be considered to be a continuous
distribution of point sources that are differential segments of the line. A sketch of
the system shows the radial distance r from the cylinder segment to the field point,
whereas R is the perpendicular distance from the field point to the cylinder. The
distance from the origin to a differential segment is denoted as s in order to distin-
guish this distance from the coordinate z of the field point. Correspondingly, ds is the
length of the segment. The field is axisymmetric, so the field point may be situated
in the x z plane, as it is shown in Fig. 1.
x
z
ψ r R
z
s ds
Figure 1.
The radial velocity of a differential segment is VR = v0 and its surface area

is 2πads, so the complex amplitude of its volume velocity is d Q̂ = Q̂ s ds =
v0 (2πads). We retain Q̂ s rather than av0 because v0 is not meaningful in the limit as
a → 0. The corresponding mass acceleration of the element is iωρ0 Q̂ s ds. The radial
distance from the differential element to the field point is r = [R 2 + (s − z)2 ]1/2 , and
the (three-dimensional) Green’s function is G (x̄, x̄s ) = exp (−ikr ) /4πr . The pres-
sure at the field point is the sum of the signal received from each differential segment.
In other words, it is an integral,

P (R, z) = G (x̄, x̄s ) iωρ0 Q̂ s ds
L/2 2 1/2 (1)
e−ik [ R +(s−z) ]
2
= iωρ0 Q̂ s 1/2 (iωρ0 ) (2πav0 ) ds

−L/2 4π R 2 + (s − z)2
The requested evaluation places the field point on the x-axis. Setting z = 0 in the
general expression gives
L/2 −ik ( R 2 +s 2 )
1/2
Q̂ s e
P (R, 0) = iωρ0 1/2 ds (2)
2π 0 R2 + s2
where the fact that the integrand only depends on s 2 allows us to double the integral
over positive s.
The integral in Eq. (2) is not one that can be found in a standard tabulation. Nev-
ertheless, let us consider an analytical integration. The square root is a complicating
feature, so it is reasonable to consider changing the integration variable. We can trans-
form from s to the angle ψ in the sketch. The trigonometric relations are s = R tan ψ,
(R 2 + s 2 )1/2 = R sec ψ, and ds = Rdψ (sec ψ)2 . This does not make the integrand
more amenable to analysis. Another transformation that eliminates the square root
is s = R sinh ζ, for which (R 2 + s 2 )1/2 = R cosh ζ and ds = R (cosh ζ) dζ. This is

promising because it converts the integral to

sinh−1
L
Q̂ s 2R
P (R, 0) = iωρ0 e−ik R cosh ζ dζ (3)
2π 0
The complex exponential may be decomposed into its real and imaginary parts.
A search through integral tables yields results for the case where L is infinite, but
not finite. It is worth the effort to consider that limit, because it should agree with
Eq. (7.3.48), and thereby confirm our analysis. The tabulated formulas14 are
∞
2
J0 (k R) = sin (k R cosh ζ) dζ
π 0
∞
2
N0 (k R) = − cos (k R cosh ζ) dζ
π 0
Thus, if L were infinite, Eq. (3) would be

Q0 ∞
P (R, 0) = iωρ0 [cos (k R cosh ζ) − i sin (k R cosh ζ)] dζ
2π 0
Q0 π
= iωρ0 [−N0 (k R) − i J0 (kr )]
2π 2
The term in the bracket is −i H0 (k R), so the pressure for the case of infinite L reduces
to
1
Pinf (R, 0) = iωρ0 Q̂ s H0 (k R) (4)
4i
This is the same as (7.3.48). In addition to confirming the analysis, this result tells
us that evaluating Eq. (3) for increasing values of L/R should yield results that are
close to those obtained for an infinite length.
The only recourse for a finite value of L is a numerical evaluation of the integral.
The numerical integration scheme we employ could be one of our own construction
using a technique like Simpson’s rule. Instead, we shall use a routine provided by
MATLAB, which is quadl. Although the integrand in Eq. (2) does not have a singu-
larity, there is a potential difficulty in evaluating it for large values of k R. The cause
of this difficulty is that k R is like the frequency of harmonic terms. A large value of
k R leads to an integrand that is a rapidly oscillating function of ζ. Some say that the
integrand has a singularity at infinite ζ. This property must be acknowledged in the
numerical integration, either by using an adaptive routine, or by explicitly increasing
the number of integration points as the value of k R increases.
The maximum value of k R has not been specified. We know that if R L,
then the three-dimensional farfield approximation should apply. Because R/L ≡
14 Equation (9.1.23) in M.I. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions,
Dover, p. 360 (1965).
k R/ (k L), this necessitates adjusting the range of k R commensurate with k L. Thus,

the computation first sets the k L value. A program loop evaluates Eq. (2) at each k R
value from zero to 60k L , with an increment of k L/2.
Figures 2–4 display the results for each k L. The data indicates that the finite
and infinite line sources for k L ≥ 4 agree well in the region from R = 0 out to
the distance at which k R = O (k L), in other words, 0 ≤ R ≤ O (L). Beyond that
range, the infinite line source model overpredicts the pressure. For the shortest length,
k L = 0.5, the pressure for an infinite line source everywhere exceeds the pressure
radiated by the finite cylinder.
100
|P|/(ωρ0Qs)
kL = 50
10-1
Finite length
10-2 Infinite length
10-3
10-2 10-1 100 101 102 103
Radial distance, kR
Figure 2.
100
|P|/(ωρ0Qs)
kL =5
^
10-1 Finite length

10-3 -2
10 10-1 100 101 102
Radial distance, kR
Figure 3.
100
|P|/(ωρ0Qs)
kL = 0.5
^
10-1
Finite length
10-3 -2
10 10-1 100 101
Radial distance, kR
Figure 4.
These trends have a simple explanation. If the length is comparable to, or larger
than, one wavelength, a field point that is not too distant from the cylinder will most
strongly be affected by the signals that radiate from the closest source points (z ≈ 0).
As the field point is moved farther outward, the contribution of the sources near the
ends becomes relatively more significant because all points on the cylinder approach
being equidistant from the field point. In the case where L is much smaller than a
wavelength, the cylinder is acoustically compact. The best simple model in that case
is a point source, rather than an infinite line source.
Numerical simulations and experimental measurements of a variety of systems

have found that these trends are general. This knowledge allows us to use infinite
length models to investigate the nearfield of a finite length body, provided that the
length of that body is not much less than a wavelength. A useful corollary is that the
radiation impedance of a finite length cylinder may be expected to be close to that
of an infinite cylinder if the length is comparable to, or larger than, 2π/k. We will
exploit this attribute in Exercise 7.10.
Field Generated by a Vibrating Cable or Beam
When a flexural wave propagates along the z-axis of a cable or beam, all points on
a specific cross section displace perpendicularly to the axis by the same amount. In
other words, each cross section translates without alteration of its shape. We shall
define the x-axis to be such that the displacement ū lies in the x z plane, so that

Vx −i2πz/L iωt
ū = Re e e ēx (7.3.49)
iω
The component of this displacement normal to a circular cross section is ū · ē R .

We measure the circumferential angle θ relative to the x-axis, so ēx · ē R = cos θ.
Correspondingly, the surface velocity is
1
Vs = Vx e−i2πz/L cos θ ≡ Vx e−i2πz/L e−iθ + eiθ (7.3.50)
2
This representation tells us that the surface wave consists of n = 1 and n = −1 cir-
cumferential harmonics. A comparison with Eq. (7.3.36) shows that the only nonzero
coefficients in this motion are V1,(−1) = V1,1 = 2Vx . Thus, the pressure generated by
a supersonic wave propagating along a cable or beam is
k H1 (κ1 R) −i2πz/L
P = −iρ0 cVx e cos θ (7.3.51)
κ1 H1 (κ1 a)
Figure 7.16 depicts the frequency dependence of the farfield pressure ampli-
tude normalized by ρ0 cV0 (R/a)1/2 . The minimum frequency for radiation is ka =
2πa/L. The high-frequency limit is seen to be a value of one, as it was for
n = 0. Rather than being singular, the transition from a subsonic to supersonic sur-
face wave is marked by a zero value for the pressure. To see why this is so, we
observe that H1 (κ1 R) is proportional to 1/ (κ1 R)1/2 at large κ1 R (farfield), while
H1 (κ1 a) → − (2i/π) / (κ1 a)2 as κ1 a → 0 (small radius). The pressure that results
from combining these limiting forms is
π 1/2 a 1/2
P = ρ0 cVx ka (κ1 a)1/2 e−i(κ1 R−3π/4−2πz/L) cos θ (7.3.52)
2 R
Thus rather than P being singular, it approaches a proportionality to (κ1 a)1/2 as

k → 2π/L. However, if L/a is not too large, the figure indicates that values of ka
slightly larger than 2πa/L give rise to large pressures.
(R/a)1/2|Pff|/(ρ0cV0)
101 L/a = L/a = 40 L/a = 4 L/a = 1
10 0
10-1
0 2 4 6 8 10
Frequency ka
Fig. 7.16 Normalized farfield pressure radiated by a supersonic flexural wave propagating axially
along an infinite cylinder
A two-dimensional model of a translating beam is obtained by letting L → ∞.

When we studied point sources, we found that a translating sphere generates a dipole
field, and that a dipole field may be obtained from a Taylor series expansion of the
field generated by two oppositely phased monopoles. Let us see whether the same
is true in the two-dimensional case. We consider two line sources at x = d/2 and
x = −d/2, where their placement in the x z plane gives symmetry relative to that
plane, which is the property of cos θ. The mass accelerations are ±iωρ0 Q̂ s , so that

d d
Pdipole = iωρ0 Q̂ s G x̄, x̄0 + ēx − iωρ0 Q̂ s G x̄, x̄0 − ēx (7.3.53)
2 2
where the Green’s function is as given in Eq. (7.3.47) and x̄0 is situated on the z-axis.
Furthermore, for a two-dimensional analysis, we take z = 0 for both the field and
source points. We apply a Taylor series to both terms, and switch the gradient at x̄0
for the gradient at x̄, which changes the sign associated with each gradient. Thus,
we have
P = iωρ0 Q̂ s [−d ēx · ∇G (x̄, x̄0 )] (7.3.54)
Now we observe that G (x̄, x̄0 ) depends only on the distance R between the points
in the x y plane, and that the gradient of R at the field point is a unit vector pointing
in the direction of increasing R, which is ē R . This leads to
∂
P = −iωρ0 Q̂ s d (ēx · ē R ) G (x̄, x̄0 )
∂R (7.3.55)
1
= − ωρ0 Q̂ s d (cos θ) H1 (k R)
4
The dipole moment D is the product of the monopole amplitude and the separation
distance. Introduction of A from Eq. (7.3.48) into the preceding leads to
1
P = D H1 (k R) cos θ, D = Ad = − ωρ0 Q̂ s d (7.3.56)
4
A comparison of this result with Eq. (7.3.51) for a/L = 0 shows that the dipole
moment associated with a translating rigid cylinder is
iρ0 cVx
Dcyl = − (7.3.57)
H1 (κ1 a)
In principle, we could mirror every development using three-dimensional point

sources with an analogous one for the two-dimensional field for a set of infinite line
source. For example, we could develop a multipole expansion in two dimensions. In
practice, this is seldom done because of the artificiality of the infinite length model.
EXAMPLE 7.8 When the string of a musical instrument is plucked, the free
vibration that ensues is a standing wave that consists of a series of mode func-
tions. The displacement of mode number m is proportional to sin (mπz/) cos θ
and its frequency is mω1 , where ω1 is the fundamental frequency and is the
string’s length. In the situation of interest, the displacement amplitude Um
of modes 1 to 10 is known to be inversely proportional to the square of the
mode number, and the contributions of higher modes are negligible. Thus,
Um = U1 /m 2 if m = 1, ..., 10, Um = 0 if m > 10, where U1 is the amplitude
of the fundamental. Consider a model of a cello string that ignores the presence
of the cello body and takes the string to be an infinitely cylinder that undergoes
an axially periodic vibration. The string’s length is 670 mm and its diameter is
1.20 mm. For the case where the sound pressure level at 2 m from the string is
measured to be 75 dB and the fundamental frequency is 440 Hz, estimate the
RMS displacement amplitude of the string.
Significance
Inverse radiation problems, in which a system’s vibration is deduced from measured
properties of the pressure field, help reinforce understanding of an analytical solution.
They also mimic, in a small way, the nature of some experimental investigations.
Solution
We know the pressure at a field point and must deduce properties of the vibration that
generated it. Our strategy is to reverse the problem. The only quantity not specified
for the displacement is the amplitude U1 of the m = 1 mode. It will be determined by
computing ( p 2 )av in terms of U1 , then matching that result to the value corresponding
to 75 dB//20 µPa, which is 0.01265 Pa2 .
The specified displacement of the cable, which we take to be in the x direction, is

10
U1 mπz
ū = Re sin eimω1 t ēx (1)
m=1
m2
We differentiate this expression with respect to time, then take the radial component
in order to determine the normal velocity on the surface. Because ēx · ēr = cos θ, the
resulting expression is

10
iω1 U1 mπz
vs = Re sin (cos θ) eimω1 t
m=1
m
We will use Eq. (7.3.51) to describe the pressure, so we employ Euler’s identity to
decompose vs into waves traveling in opposite directions, which gives
10
U1 ω1 imπz/
vs = Re e − e−imπz/ eiθ + e−iθ eimω1 t (2)
m=1
4m
This expression tells us that each time harmonic consists of a pair of helical waves
at n = ±1 that spiral about the z-axis in opposite senses. The velocity coefficients
are Vm,1 = Vm,−1 = −V−m,1 = −V−m,−1 = ωU1 /(2m).
The nature of the pressure generated by each wave depends on whether it is
supersonic. The frequency of a term in Eq. (2) is mω1 and its axial wavenumber is
mπ/, so that
mπ 2 1/2
mω 2
= mκ1 = 6.780m m−1
1
κm = − (3)
c
All values of κm are real, which means that all of the terms in Eq. (1) are supersonic
helical waves.
Each term in Eq. (2) occurs at a different frequency, so the time factor must be
retained when Eq. (7.3.51) is used to evaluate the pressure. In each term of that
expression, we replace k with mω1 /c, κ1 with mκ1 , and 2/L with m/. The result is

10
U1 ω1 imπz/ ω1 H1 (mκ1 R)
−imπz/
p = Re −iρ0 c e −e (cos θ) e imω1 t
2m cκ1 H1 (mκ1 a)
m=1
ω12 H1 (mκ1 R)
10 mπz
= Re −iρ0 U1 sin (cos θ) e imω1 t
κ1 m=1 m H1 (mκ1 a)
(4)
Each term in Eq. (4) represents an oscillation at a distinct frequency, so the mean-
squared pressure is a sum of squares of the individual amplitudes. Thus, the predicted
mean-square pressure is
2 10

1
∞
1 ρ0 ω12
H1 (mκ1 R)
2 mπz 2
p av =
2 ∗
Pm Pm = |U1 |

(cos θ)2
2 m=1 2 κ1
m H (mκ a)
sin
m=1 1 1
(5)
The transverse distance for the pressure measurement is specified to be R = 2 m,
but the values of z and θ at which the pressure was measured are not given. It is
reasonable to assume that the value of θ is that which leads to a maximum, which is
θ = 0 or π, both of which give (cos θ)2 = 1. In contrast, no value of z is evident as

leading to a maximum. For example, if we selectthe midpoint, z = /2, the terms
for even m will be zero. Thus, let use the value of p 2 av averaged over the length of
the string, which is

10
2
1
1 ρ0 ω12
H1 (mκ1 R)
2
p 2
= 2
p av,av
dz = |U1 |

(6)
av,av θ=0 4 κ1
m H (mκ a)
0 m=1 1 1

The result of substituting all quantities into Eq. (6) is p 2 av,av = 784.3 |U1 |2 .
Matching this to the measured mean-squared pressure of 0.01265 Pa2 leads to |U1 | =
4.016 mm. With this, we have fully characterized the displacement. Its mean-squared
value is obtained by averaging ū · ū in Eq. (1) over 0 < t < π/ω1 and 0 < x < .
The result is
2 |U1 |2 1 −6 2
10
u x av = = 4.363 10 m =⇒ (u x )rms = 2.08 mm (7)
4 m=1 m 4
The actual displacement amplitude of the strings of a cello is barely perceptible to

the eye. The value we have determined is too large. The discrepancy between our
analysis and reality is only partially the consequence of taking the string’s vibration
to be spatially periodic and using a model of an infinitely long string. The primary
fault is that the strings are an extremely minor contributor to the sound emitted by
a stringed musical instrument. The role of the strings is to induce vibration of the
instrument’s body. This vibration radiates sound directly, and it also excites the cavity
enclosed by the body. The sound created in the cavity is emitted through the open
ports of the body.
7.3.5 Finite Length Effects
Two aspects of the vibrating cylinder model we have created limit its practical utility.
The most obvious one is taking the cylinder to be infinitely long. Unfortunately, there
are no analytical solutions for radiation of finite length cylinders. If one requires an
accurate solution of such problems, all available tools use numerical techniques. We
will examine a few later in this chapter.
Even within the context of a model that takes a cylinder to have infinite length,
there is another unrealistic aspect. We have considered helical waves and axially
periodic vibrations, which leads to the question of how would one generate such
a disturbance? A more realistic model allows the surface velocity to depend in an
arbitrary manner on the axial distance, subject to the condition that it be negligible
at large distances in either direction. In order to not obscure the analysis, we will
restrict attention to the case of a single circumferential harmonic. The solutions may
be superposed to create the Fourier series for an arbitrary circumferential excitation.
The surface velocity for this investigation is

vs = Re f (z) e−inθ eiωt , lim | f (z)| → 0 (7.3.58)
|z|→∞
This distribution is not periodic in the axial direction, which is equivalent to saying
that its period is infinite. The conditions imposed on f (z) fit the requirements for
its Fourier transform to exist. The transform, which is denoted as F̃ (μ), is
∞
F̃ (μ) = f (z) e+iμz dz (7.3.59)
−∞
As explained in Appendix B, the Fourier transform of f (z) essentially describes a

Fourier series whose wavenumber is a continuous spectrum μ, rather than a set of
discrete values nμ1 . The series coefficient at each μ is F̃ (μ) dμ. This perspective is
especially evident when the inverse Fourier transform is used to synthesize f (z). It
tells us that
1 ∞
Vs = F̃ (μ) e−i(μz+nθ) dμ (7.3.60)
2π −∞
The elementary view of an integral is that it is an infinite sum of terms described

by its integrand. From that viewpoint, the above description states that the surface
velocity is an infinite sum of helical waves whose axial wavenumber is μ, whose
circumferential wavenumber is n, and whose complex amplitude is Ṽ (μ) dμ. The
fact that the number of terms in the sum is infinite does not affect the validity of
superposing the pressure field for each helical wave. Equation (7.3.21) gives the
pressure for a single helical wave, so the superposition gives
∞
k Hn (κ (μ) R) −i(μz+nθ)
P (R, θ, z) = −iρ0 c F̃ (μ) e dμ (7.3.61)
−∞ κ (μ) Hn (κ (μ) a)
where 1/2
κ (μ) = k 2 − μ2 (7.3.62)
The transverse wavenumber has been written as κ (μ) as a reminder that its depen-
dence must be recognized when the integral is evaluated. The portion of the integrand
in Eq. (7.3.61) that is a factor of e−iμz is a Fourier transform. Thus, we have obtained
a representation of P as an inverse Fourier transform.
The preceding is a heuristic derivation. A more rigorous analysis provides under-
standing as to how one may apply the Fourier transform in other situations. We
begin with the argument that the dependence of P on the axial position also may be
represented by a Fourier transform. In addition to depending on the axial wavenum-
ber μ, this transform may depend on the transverse distance R, whereas exp (−inθ)
must describe the circumferential dependence. If we knew the transformed function,
which is denoted as P̃ (R, μ), then the inverse Fourier transform would enable us to
recreate P (R, θ.z) according to
∞
1
P (R, θ, z) = P̃ (R, μ) e−iμz e−inθ dμ (7.3.63)
2π −∞
This ansatz must satisfy the Helmholtz equation in cylindrical coordinates. Spatial
derivatives of the above are

∂ 1 ∞ ∂
P (R, θ, z) = P̃ (R, μ) e−iμz e−inθ dμ
∂R 2π −∞ ∂ R

∂ 1 ∞
P (R, θ, z) = (−in) P̃ (R, μ) e−iμz e−inθ dμ (7.3.64)
∂θ 2π −∞

∂ 1 ∞
P (R, θ, z) = (−iμ) P̃ (R, μ) e−iμz e−inθ dμ
∂z 2π −∞
Second-order derivatives follow directly from these forms, so substitution of the

assumed solution into the Helmholtz equation leads to

∞
d 2 P̃ 1 d P̃ n2
+ + κ (μ) − 2 P̃ e−iμz e−inθ dμ = 0
2
(7.3.65)
−∞ d R2 R dR R
The preceding must be satisfied at all z, which will only be true if the bracketed term
in the integrand vanishes. Thus, P̃ (R, μ) must be a solution of Bessel’s equation.
As previous, an outgoing wave is obtained by taking it to be a Hankel function of
the second kind. It may be multiplied by a coefficient that depends on μ and n, so
we now have
∞
P (R, θ, z) = B (μ, n) Hn (κ (μ) R) e−iμz e−inθ dμ (7.3.66)
−∞
The coefficient B (μ, n) is set by the boundary condition, which matches Vs to the
complex particle velocity in the radial direction at R = a. A derivative with respect
to R for Euler’s equation operates on the Hankel function, so using Eq. (7.3.60) to
represent the surface velocity leads to
∞ ∞
κ (μ) B (μ, n) Hn (κ (μ) a) e −iμz −inθ
e dμ = −iωρ0 F̃ (μ) e−i(μz+nθ) dμ
−∞ −∞
(7.3.67)
Here too, the equality must hold for all z, so the integrands must be equal. From this,
we solve for B (μ, n). Backsubstituting that results into the integral representation
of P leads to Eq. (7.3.61), which verifies the heuristic approach.
The spectrum of surface waves consists of supersonic waves for |μ| < k and
subsonic waves for |μ| > k. For subsonic waves, let us set κ (μ) = −iβ (μ), where
1/2
β = μ2 − k 2 . It is helpful for evaluations to rewrite Eq. (7.3.61) in a form that
splits the two in order to explicitly display the contributions of the supersonic and
subsonic spectra. For this, we note that neither κ nor β depends on the sign of μ, so
we may reduce the integral to extend only over μ ≥ 0. These operations lead to

k
k Hn (κ (μ) R) −i(μz+nθ)
P (R, θ, z) = −iρ0 c F̃ (μ) + F̃ (−μ) e dμ
0 κ (μ) Hn (κ (μ) a)
∞ k K n (β (μ) R) −(μz+nθ)
+ ρ0 c F̃ (μ) + F̃ (−μ) e dμ
k β (μ) K n (β (μ) a)
(7.3.68)
An important aspect of this representation is that the contribution from μ > k decays
rapidly with increasing R and z. Therefore, the second integral may be ignored if we
seek a farfield description.
Equation (7.3.68) is an inverse Fourier transform. It is an implementation of
a general analytical approach known as an angular spectrum decomposition. We
will encounter in the next chapter another system that has been analyzed by this
technique. Many research papers have derived solutions in the form of inverse Fourier
transforms, and then used sophisticated tools of mathematical analysis to identify
important properties. Such analysis exceeds the scope of this book. An alternative
is to employ numerical methods. The last part of Appendix B describes how to use
an FFT numerical procedure to evaluate the Fourier transform of a spatial function,
and how to convert a Fourier transform back to a spatial function by application of
an inverse FFT procedure. Thus, it might seem viable to employ FFT techniques
to evaluate Eq. (7.3.61). Despite the apparent simplicity of this approach, there are
several issues regarding aliasing and wraparound error, which were discussed in
Chap. 1, that must be addressed. An additional complication arises as either R or z
is increased, because the integrands become rapidly oscillating functions of μ. This
raises the sampling requirements, and sometimes leads to numerical instabilities.
Ultimately, evaluation of the pressure field would have limited use. As was shown
in Example 7.7, the fact that the cylinder is not infinite becomes important at radial
distances that exceed the length. Furthermore, even if we are only interested in the
field close to a cylinder, end effects cannot be ignored at locations that are close to
an end.
7.4 Kirchhoff–Helmholtz Integral Theorem
It should be apparent at this juncture that we need tools other than separation of
variables solutions of the Helmholtz equation. The first step toward that objective is
application of the definition of a Green’s function to derive the Kirchhoff–Helmholtz
integral theorem, which we will refer to as the KHIT. It is the basis for some common
numerical modeling technique, but it is not a solution of any problem. Rather, it will
tell us how to evaluate the pressure at a field point when we know the pressure
7.4 Kirchhoff–Helmholtz Integral Theorem 85
and normal velocity on the surface from which this signal emanates. The laws of
mechanics, as well as our experience thus far, tell us that both quantities cannot be
specified at the same location. (This fact is merely an extension of Newton’s second
law for a particle, which tells us that we can determine the velocity as a function of
time if we know the force history, or we can determine the force required to attain a
specified velocity history, but we cannot specify both histories.) The KHIT version
we will derive pertains to the response in the frequency domain.
7.4.1 Derivation for an Acoustic Cavity
The development begins with the simultaneous consideration of two acoustic fields:
the frequency domain pressure radiated by vibration of the boundary and a Green’s
function. Usually, the free-space Green’s function is the one we would use, but all
steps will be equally valid of the Green’s function satisfies boundary conditions for
the system at hand. We let V denote the domain of interest, and S is its boundary.
For our initial effort, we consider V to be enclosed by S. In other words, it is a cavity
whose shape is arbitrary. The configuration is depicted in Fig. 7.17.
Fig. 7.17 Configuration of xs

an acoustic cavity
r^
x0
n(xs)
The pressure at a field point x̄ satisfies the Helmholtz equation. The Green’s func-
tion for this point due to a source at x̄0 , which is G (x̄0 , x̄), satisfies an inhomogeneous
version of that equation, specifically
∇ 2 P + k 2 P = 0, ∇ 2 G + k 2 G = −δ (x̄ − x̄0 ) (7.4.1)
We multiply the equation for P by G (x̄0 , x̄) and the equation for G by P (x̄). Taking
the difference yields
G (x̄0 , x̄) ∇ 2 P (x̄) − P (x̄) ∇ 2 G (x̄0 , x̄) = P (x̄) δ (x̄ − x̄0 ) (7.4.2)
This expression is multiplied by a differential element dV at x̄ and integrated over

the entire domain. Gradients of products are covered by the same rule as the scalar
derivative of a product, and ∇ 2 ≡ ∇ · ∇, so we have

G (x̄0 , x̄) ∇ 2 P (x̄) − P (x̄) ∇ 2 G (x̄0 , x̄) dV
V

≡ ∇ · [G (x̄0 , x̄) ∇ P (x̄) − P (x̄) ∇G (x̄0 , x̄)] dV
V

= P (x̄) δ (x̄ − x̄0 ) dV (7.4.3)
V
The filtering property of a Dirac delta function simplifies the right side. If x̄0 is
inside V, then the integral will give P at x̄0 . This is the case that usually interests us,
but there also is reason to consider x̄0 outside V. Let us denote χ (x̄0 ) as a coefficient
that captures both alternatives,

1 if x̄0 ∈ V
χ= (7.4.4)
0 if x̄0 ∈
/V
The left side of Eq. (7.4.3) is described by the divergence theorem. The differential
element is situated at x̄ within V. To assure that we recognize that the resulting
surface integral applies to field points that are on S, we designate those points as x̄s
and denote the gradient as ∇s . Also, the usual statement of the divergence theorem
uses the normal vector on the surface that is oriented outward from V, but our studies
have taken this direction to be oriented into the fluid. Thus, we shall use the divergence
theorem with −n̄ (x̄s ), which leads to

(−n̄ (x̄s )) · [G (x̄0 , x̄s ) ∇s P (x̄s ) − P (x̄s ) ∇s G (x̄0 , x̄s )] d S = χ (x̄0 ) P (x̄0 )
S
(7.4.5)
The gradient of P (x̄s ) is related to the complex particle velocity amplitude by Euler’s
equation, whose application leads to the final form of the Kirchhoff–Helmholtz inte-
gral theorem (KHIT),

χ (x̄0 ) P (x̄0 ) = iωρ0 n̄ (x̄s ) · V̄ (x̄s ) G (x̄0 , x̄s )
(7.4.6)
S
+P (x̄s ) n̄ (x̄s ) · ∇s G (x̄0 , x̄s )] dS
This relation has a simple interpretation in terms of point sources. For the first term
in the integrand, we observe that iωρ0 n̄ (x̄s ) · V (x̄s ) dS is a complex amplitude of the
mass acceleration across a differential patch of the boundary. The pressure radiated
by a point source having this mass acceleration is iωρ0 n̄ (x̄s ) · V (x̄s ) dSG (x̄0 , x̄s ).
The interpretation of the second term follows from the observation that a similar
term occurred in Eq. (6.5.90) for a dipole. A comparison with that term shows that the
−P (x̄s ) n̄ (x̄s ) dS/4π is a differential dipole moment that is oriented perpendicularly
to the surface. Thus, we may regard the KHIT as a statement that a vibrating surface
acts as though it was a continuous sheet of monopoles and dipoles.
The KHIT that is found in a text or a research paper might not exactly replicate
Eq. (7.4.6). One difference might arise from the use of Re (exp (−iωt)) to represent a
harmonic variation. Here, like everywhere else in this book, this alternate convention
leads to a complex conjugate representation. Another difference might lie in the
definition of the Green’s function, which might insert a minus sign and/or omit the
4π factor. Differences in the definitions of the Green’s function cease to be an issue
if the definition is explicitly substituted into the theorem. The free-space Green’s
function we have defined is
1 −ikr̂
G (x̄0 , x̄s ) = e , r̂ = |x̄s − x̄0 | (7.4.7)
4πr̂
(The use of r̂ to denote the source-field point distance is intended to avoid confusion
with r as the radial coordinate.) The meaning of ∇s G (x̄0 , x̄ S ) in the integrand is that
it is the gradient at x̄s with x̄0 held fixed. This gradient was evaluated in Sect. 6.5.1
when we used a Taylor series to derive the dipole field. The direction in which r̂
increases most rapidly when x̄0 is fixed is the direction of increasing r̂ . Therefore,
the procedure we followed earlier yields

∂ 1 −ikr̂
n̄ (x̄s ) · ∇s G (x̄0 , x̄) = n̄ (x̄s ) · ∇s r̂ e
∂ r̂ 4πr̂
(7.4.8)
x̄s − x̄0 ik r̂ + 1 −ikr̂
= −n̄ (x̄s ) · e
r̂ 4πr̂ 2
Thus, the KHIT version that explicitly displays the free-space Green’s function is

1
χ (x̄0 ) P (x̄0 ) = iωρ0 n̄ (x̄s ) · V̄ (x̄s )
4π
S −ikr̂
x̄s − x̄0 1 e
− n̄ (x̄s ) · ik + P (x̄s ) dS
r̂ r̂ r̂
(7.4.9)
The KHIT has not been used frequently to study two-dimensional systems. If one
wishes to do so, it would be incorrect to employ the preceding form. The proper for-
mulation would use Eq. (7.4.6) with the two-dimensional free-space Green’s function
in Eq. (7.3.47), which is G (x̄, x̄s ) = H0 (k R) /(4i).
As was noted before we began the derivation, the pressure and particle velocity
cannot both be specified on S. Thus, it is reasonable to question whether the KHIT
has practical use. Indeed, it does! It may be used to verify a solution that was obtained
by another method, but it seldom is. A common application uses the KHIT to evaluate
a farfield pressure field when the surface response has been obtained by a numerical
simulation technique, or by experimental measurements. An important version of the
KHIT lies in the case not covered by the definition of χ (x̄0 ), Eq. (7.4.4), specifically
x̄0 being situated on S. In that case, the KHIT gives an equation for the surface
pressure at a specific field point on the surface in terms of the surface pressure
and velocity distributions. Just as a differential equation relates a variable to its
derivatives, this version of the KHIT is an integral equation for the pressure. The
solution of this equation by numerical methods is known as a boundary element
formulation. This concept is explored in Sect. 7.5.2.
EXAMPLE 7.9 Consider a spherical cavity in the case where the container
executes a radially symmetric oscillation. In that case, the complex radial veloc-
ity on the boundary, r = a, is Vr = V0 . This field was analyzed in Sect. 6.3.3. It
is desired to use KHIT to verify the analytical solution for the radially symmet-
ric pressure in the interior. To do so, use the analytical solution to determine
the pressure at r = a, then use that distribution to formulate the KHIT. Evalu-
ate the pressure as a function of kr according to KHIT, and compare it to the
analytical solution for the pressure field. The frequency for the evaluation is
ka = π.
Significance
The formulation of KHIT for a system whose geometry is not too complicated allows
us to focus on the meaning of each term. The computed results will suggest an attribute
that is crucial to the application of the KHIT as the foundation of a boundary element
formulation.
Solution
Equation (6.3.27) states that the radially symmetric pressure and particle velocity
inside a vibrating sphere are
a ka sin (kr )
P = −iρ0 cV0
r ka cos (ka) − sin (ka)
a 2 kr cos (kr ) − sin (kr ) (1)
Vr = V0
r ka cos (ka) − sin (ka)
At r = a, these expressions give
ka sin (ka)
P = −iρ0 cV0 , Vr = V0 (2)
ka cos (ka) − sin (ka)
The task is to use Eq. (2) to form the integrand of the KHIT, then evaluate the
pressure within the cavity, and compare it to the pressure in Eq. (1). The field is
radially symmetric, so there is no difference in the dependence of P along any radial
line. We shall designate this line as the polar axis z because doing so simplifies the
formulation. Figure 1 shows a generic surface point x̄s at polar angle ψ, and field
point x̄0 at distance z 0 on the z-axis. (This picture is descriptive of all points at the
same value of ψ for any azimuthal angle θ.)
xs -x0
x
r^ eR
xs
β
a
r^
ψ
z
z0 x0
Figure 1.
We could obtain a description of the terms in the KHIT integrand from a trigonometric
analysis. However, one objective here is to get an idea of how to formulate the
theorem in a general situation where the geometrical configuration is less amenable
to a pictorial representation. Thus, all vectorial quantities will be represented in terms
of components relative to x yz. The normal direction points into the spherical cavity,
so the terms appearing in Eqs. (7.4.7) and (7.4.8) are
x̄s = a (sin ψ ēx + cos ψ ēz ) , x̄0 = z 0 ēz
1/2
r̄ = |x̄s − x̄0 | = a 2 + z 02 − 2az 0 cos ψ
(1)
n̄ (x̄s ) = −ēr = − sin ψ ēx − cos ψ ēz
x̄s − x̄0 a − z 0 cos ψ
−n̄ (x̄s ) · = cos β =
r̂ r̂
An axisymmetric differential area element is 2π (a sin ψ) (adψ), so we have char-
acterized the geometric variables required to form the KHIT. For P (x̄s ), we use the
first of Eq. (2). Because the normal direction is defined to point into the fluid domain,
the surface velocity is n̄ (x̄s ) · V̄ (x̄s ) = −Vr = −V0 . All parameters are specified in
nondimensional form, so we shall replace a, r̂ , and z 0 with ka, k r̂ , and kz 0 every-
where.
The descriptions are substituted into the KHIT, with the result that we must eval-
uate
χP (x̄) i ka sin (ka)
= − (ka)2 I1 + I2 (2)
ρ0 cv0 2 ka cos (ka) − sin (ka)
where π −ik r̂
e
I1 = sin ψdψ
0 k r̂
(3)
π −ik r̂
e
I2 = (cos β) (sin ψ) dψ
0 k r̂
Equation (1) give k r̂ and cos β as functions of ψ for specified values of ka and kz 0 ,
so we proceed to the evaluation. If we were to search through tables of integrals,
we might find expressions for I1 and I2 . However, the idea here is to use numerical
methods because we would have no expectation of an analytical result for more
complicated shapes.
The data in Fig. 2 was obtained with MATLAB. For each value of kz 0 , the anony-
mous function capability was used to define kr_hat as a function of psi, which
then was used to define cos_beta as a function of psi. Both were then used to
define the integrands of I1 and I2 , and the integrals were found by invoking the
quadl routine. The error tolerance was decreased to 10−8 .
10
Im(P)/(ρ0 cv0)
0 KHIT
Analytical solution
−10
0 1 2 3 4 5 6
Radial distance kz0
Figure 2.
According to Eq. (2), P/ (ρ0 cVa ) within the cavity is purely imaginary. The value
obtained
from the KHIT nowhere differs from the analytical value by more than
1.4 10−7 , and the real part is no greater than ±1.8(10−10 ). The discontinuity at
kz 0 = 4 results because the pressure obtained from KHIToutside the domain should
be zero. For kz 0 > ka, the computed value is less than 6 10−8 out to kz 0 = 1.5ka.
The evaluation skipped kz 0 = ka because both integrands in that case are singular
at ψ = 0. Nevertheless, it is interesting to wonder what would happen if kz 0 were
very close to ka. To test this, a sequence of evaluations with kz 0 = ka ∓ 10−n were
carried out. The results are tabulated below.

p/ ρ0 cv0
kz 0 − ka KHIT Analytical

− 10−5 −5.1281 10−9 − 1.6295i −1.6295i
−6 −9
− 10 −5.1282 10 − 0.8146i −1.6295i

− 10−7 −5.1282 10−9 − 0.8145i −1.6295i
−8
− 10 Singular −1.6295i
−8
10 Singular 0
−7
10 −5.1282 10−9 − 0.8144i 0
−6 −9
10 −5.1282 10 − 0.8143i 0
−5
10 −5.1283 10−9 − 0.8127i 0
−4
10 −5.1288 10−9 + 6.5779 10−7 i 0
The table indicates that as the field point approaches the surface from the inside,
the pressure drops from the analytical value to approximately half that value, but
it is singular at the closest distance. The trend for points outside the sphere begins
with a singularity at the closest distance, then approximately half the value on the
surface, then zero. This behavior will be examined in our development of boundary
elements, where it will be proven that χ (x̄0 ) = 0.5 if x̄0 actually is situated on a
smooth surface.
7.4.2 Acoustic Radiation into an Exterior Domain
This situation we consider here is one where a vibrating surface S radiates into
a surrounding fluid domain V, and the extent of V is infinite. We cannot apply
KHIT directly because its derivation required that the vibrating boundary encloses
the domain. However, the derivation of KHIT did not require that the boundary
be continuous. Consequently, it is valid if we consider the boundary of the fluid
domain to be the composite of the radiating surface S and a very large sphere S O
having radius r O that encloses S. The fluid domain V is the region between S and
S O . Figure 7.18 depicts the decomposition of the region. After we modify KHIT to
describe this configuration, the limit as r O → ∞ will provide the theorem we seek.
Fig. 7.18 Acoustic region

for derivation of the
Kirchhoff–Helmholtz
integral theorem describing n(xs)
radiation from vibrating n(xs)
body S . The surrounding
sphere is virtual
O
rO
The integral in the KHIT now becomes the sum of integrals over the vibrating
surface S and the virtual S O . The first has the same form as the left side of Eq. (7.4.6).
The only difference stems from the convention that n̄ (x̄s ) always is oriented into the
fluid. Hence, for the radiation problem, it is outward from S, whereas it is oriented
inward when S is the outer boundary of a cavity. The other part of the surface integral
is the contribution from S O . The normal direction for this portion of the boundary
is −ēr , so that n̄ (x̄s ) · ∇ = −∂/∂r and n̄ (x̄s ) · V̄ (x̄s ) = −Vr . Also, in terms of
spherical coordinates, we have dS O = r O2 sin ψdθdψ. Thus, the contribution of the
surrounding sphere is

[iωρ0 n̄ (x̄s ) · V (x̄s ) G (x̄0 , x̄s ) + P (x̄s ) n̄ (x̄s ) · ∇s G (x̄0 , x̄s )] d S
SO π π
∂
= −iωρ0 Vr (x̄s ) G (x̄0 , x̄s ) − P (x̄s ) G (x̄0 , x̄s ) r O2 sin ψdθdψ
0 −π ∂r
(7.4.10)
Our study of radiated power in Sect. 4.5.3, which led to the Sommerfeld radiation
condition for a three-dimensional domain, proved that if r O is very large, then the
pressure should exhibit spherical spreading. The specific description was

1 1
P= f (ψ, θ) e−ikr O + O (7.4.11)
rO r O2
Application of Euler’s equation shows that the radial velocity is

1 1
Vr = f (ψ, θ) e−ikr O + O (7.4.12)
ρ0 cr O r O2
where the two variables have been written separately, rather than as P = ρ0 cVr to
emphasize that this property is only satisfied to the leading order in r O . The Green’s
function G (x̄s , x̄0 ) also will satisfy the farfield approximation when x̄s is situated
on SO . This is so because the source location x̄0 is at a fixed finite distance from the
center of the sphere, and r O can be as large as necessary to make this statement be
true. Thus, it must be that for any x̄s ∈ S O ,

1 1
G (x̄0 , x̄s ) =g (ψ, θ) e−ikr O + O
rO r O2
(7.4.13)
∂ ik 1
G (x̄0 , x̄s ) = − g (ψ, θ) e−ikr O + O
∂r rO r O2
When these farfield representations are substituted into Eq. (7.4.10), the terms that
are products
of the f and g functions cancel. What remains in the integrand is
O 1/r O3 . This term is multiplied by r O2 in the differential area dS, so the quantity
that is integrated is O (1/r O ). Consequently, the integral vanishes in the limit as
r O → ∞. Therefore, the KHIT only requires integration over S.
The consequence is that Eq. (7.4.6) also applies when V is the infinite fluid regions
surrounding a vibrating surface S. It is helpful to restate the theorem. The gen-
eral form leaves the Green’s function unspecified, so it is valid for two- and three-
dimensional problems. This description is

χ (x̄0 ) P (x̄0 ) = iωρ0 n̄ (x̄s ) · V̄ (x̄s ) G (x̄0 , x̄s )
(7.4.14)
S
+ P (x̄s ) n̄ (x̄s ) · ∇s G (x̄0 , x̄s )] dS
The usual application is to three-dimensional systems. Explicitly, representing the

free-space Green’s function in the preceding leads to

1 x̄s − x̄0 1
χ (x̄0 ) P (x̄0 ) = − n̄ (x̄s ) · ik + P (x̄s )
4π r̂ r̂
S −ikr̂ (7.4.15)
e
− iωρ0 n̄ (x̄s ) · V̄ (x̄s ) dS, r̂ = |x̄s − x̄0 |
r̂
This is the form in which KHIT typically appears.

The fact that KHIT has the same appearance for the interior and exterior problem
can cause confusion. Compare the following situations: (1) Surface S is the exterior
boundary of a cavity, (2) the same surface S is the vibrating surface of a transducer
surrounded by fluid. The normal velocity distribution in both systems is the same. So
what is the difference in KHIT for these systems? The answer lies in the definition
of the normal direction on the surface, n̄ (x̄s ) as the normal that points into the fluid
domain. Thus, if S is the surface that encloses a cavity, n (x̄) is oriented onto the
region bounded by S. In contrast, if S is the surface of a body that is surrounded by
fluid, then n (x̄) points outward from the region bounded by S. In other words, n̄ (x̄s )
is reversed between the two cases. The only other change is that χ (x̄0 ) is one if x̄0
is situated within the cavity, or if it is exterior to the transducer, and χ (x̄0 ) is zero if
x̄0 is outside the cavity or inside the transducer.
As was the case for the interior cavity, the KHIT is not a solution to a problem
because we cannot specify both the pressure and normal velocity at a location. A
common use is to determine the pressure, especially in the farfield, when the surface
response has been determined by another method. Example 7.10, which follows,
describes such an application in conjunction with an approximation of the surface
pressure.
If only the farfield pressure is of interest, the KHIT may be written in a form that
is somewhat simpler. This is done by introducing the same approximations as those
employed to describe the farfield of a collection of point sources. The origin is inside
S. As part of the restriction to the farfield, the distance to x̄0 from the origin must be
much greater than the maximum distance from the origin to a point on the surface,
that is, |x̄0 | max (|x̄s |) In that case, the line from x̄ S to x̄0 , which is x̄0 − x̄s , will
approach parallelism to x̄0 . Thus, the distance r̂ from the surface point to the field
point may be approximated as the radial distance r = |x̄0 | from the origin less the
projection of x̄s onto this radial line. The unit vector for the radial line is x̄0 /r , so we
have
x̄0
r = |x̄0 | , r̂ ≡ |x̄s − x̄0 | ≈ r − x̄s · (7.4.16)
r
where the approximation becomes increasingly accurate with increasing |x̄0 |. We use
this approximation to factorize the phase of the free-space Green’s function and its
normal derivative in Eq. (7.4.8). The result is that

e−ikr
G (x̄s , x̄0 ) ≈ eik x̄s ·x̄0 /r
4πr (7.4.17)
x̄0 ike−ikr
n̄ (x̄s ) · ∇s G (x̄s , x̄) ≈ n̄ (x̄s ) · eik x̄s ·x̄0 /r
r 4πr
The corresponding form of the KHIT is

e−ikr x̄0 ik x̄s ·x̄0 /r
Pff (x̄0 ) = ik ρ0 c V̄ (x̄s ) · n̄ (x̄s ) + P (x̄s ) n̄ (x̄s ) · e dS
4πr r
S
(7.4.18)
The sole dependence of the integrand on the location of the field point is the value
of r and the unit vector x̄0 /r . When we use the polar angle ψ0 and azimuthal angle
θ0 to locate that point, the farfield version of KHIT reduces to
e−kr
Pff (x̄0 ) = ik f (ψ0 , θ0 ) (7.4.19)
4πr
This is the same form as that in Eq. (4.5.43). The difference is that the KHIT tells
us how to evaluate the directivity factor, f (ψ0 , θ0 ). Of course, this assumes that we
know the pressure and normal velocity on the radiating body’s surface.
As was true for the interior problem, G (x̄0 , x̄s ) need not be the free-space Green’s
function, but it usually is because any other is difficult to obtain. However, one notable
exception is the Green’s function used to describe radiation from an infinite planar
boundary. Chapter 8 is devoted to this topic. The foundation for that study is the
Rayleigh integral, which essentially is the KHIT with a Green’s function that is
derived from application of the method of images.
EXAMPLE 7.10 A cylindrical tank of length L is submerged. Pulsations in

the feed line result in internal pressure fluctuations that induce an axisymmetric
vibration of the cylinder. The velocity of the cylinder’s surface in the normal
direction is v R = Re(v0 cos (πz/L) eiωt ) for −L/2 ≤ z ≤ L/2, which is a sin-
gle lobe of a sine function that extends over the length of the cylinder. The
ends of the cylinder are immobile. It is desired to determine the farfield pres-
sure distribution. The radiation impedance of an infinite cylinder provides an
approximation of the surface pressure generated by the specified velocity. Use
this approximation to formulate the farfield pressure according to the KHIT.
Evaluate the result for the directivity (r/a) |Pff (x̄0 )| / (ρ0 cv0 ) as function of
the spherical angle locating field point x̄0 . Parameters for this evaluation are
ka = 1 and 6, and L/a = 4 and 20.
Significance
In addition to showing how the KHIT may be formulated for three-dimensional
bodies, this example has the purpose of showing that in some occasions, one may
use knowledge of fundamentals to construct approximate solutions to problems that
otherwise would require sophisticated simulation software.
Solution
The analysis is based on the suggestion in Example 7.7 that the field close to a cylinder
might be reasonably well predicted by a model in which the cylinder is infinite. It is
an approximation to ignore the fact that the helical surface waves do not propagate
to infinity. In addition, deformation of the cylinder would result in movement of the
ends, even if they are made very rigid. Furthermore, even if the ends did not move, the
pressure on the ends would not be zero. Nevertheless, the analysis we will pursue is
useful because it enables us to understand some phenomena and it provides an order
of magnitude check for more exact analyses. Some might refer to this as a “back of
the envelope” analysis, but that description is only appropriate in comparison with
the effort required for a faithful simulation.
The decomposition of the given radial velocity into its constituent helical surface
waves is v π
eiμz + e−iμz eiωt , μ =
0
v R = Re (1)
2 L
Both are axisymmetric, n = 0. In comparison with Eq. (7.3.16), this motion is a
superposition of two helical waves whose velocity coefficients are V1,0 = V−1,0 =
v0 /2. The radial wavenumber for both is
1/2
a2
κa = (ka)2 − π 2 2 (2)
L
All combinations of the specified values of ka and L/a lead to κa being real, so
the helical waves are supersonic. The radiation impedance in Eq. (7.3.35) does not
depend on the sign of the axial wavenumber, so it is the same for both helical waves.
Adding the contribution of each leads to the complex amplitude of surface pressure
being
ka H0 (κa)
P = ρ0 cZ rad v0 cos (μz) , Z rad = −i (3)
κa H0 (κa)
The next step is to describe the geometric variables in the farfield representation
of the KHIT, Eq. (7.4.18). We do so by using a set of x yz axes to describe vector
components. The variables to be represented appear in Fig. 1.
- -)
n(x y
s
z
-x
s
z θ
a
ψ0
-x
0
r
x
Figure 1.
The field point x̄0 has been placed on the x z plane at polar angle ψ0 . This placement
will yield a general result because an axisymmetric surface vibration generates a field
that also is axisymmetric. The surface point has cylindrical coordinates (a, θ, z), and
the transverse direction ē R is normal to the surface. It is stated that the ends of the
cylinder are immobile. We shall assume that the pressure on the ends also is zero.
(This assumption often is based on the conjecture that radiation from the ends is
unimportant if L/a is not small.) The quantities of interest are
x̄0 = r sin ψ0 ēx + r cos ψ0 ēz

x̄s = a cos θēx + a sin θē y + z ēz
n̄ (x̄s ) = ē R = cos θēx + sin θē y
These variables depend on θ, so we use (adθ) dz as the differential area element and
integrate over −π < θ ≤ π, −L/2 ≤ z ≤ L/2. Substitution of the preceding terms
and P = Z rad VR into Eq. (7.4.18) leads to
L/2 π
Pff (x̄0 ) ika −ikr
= e cos (μz) (1
ρ0 cv0 4πr −L/2 −π (4)
ik(a cos θ sin ψ0 +z cos ψ0 )
+ Z rad cos θ sin ψ0 ) e dθdz
The exponential may be split into factors that depend solely on θ and z, and Z rad
is independent of θ and z. Furthermore, the θ integral depends only on cos θ, so it is
the same as twice the integral over 0 < θ ≤ π. Recognition of these features reduces
Eq. (4) to
Pff (x̄0 ) ika −ikr
= e (I1 + I2 Z rad sin ψ0 ) Iz (5)
ρ0 cv0 4πr
Integration over z is described by Iz . It is

L/2
Iz = cos (μz) eikz cos ψ0 dz
−L/2
sin [(k L cos ψ0 + μL)/2] sin [(k L cos ψ0 − μL)/2]
= +
(k cos ψ0 + μ) (k cos ψ0 − μ)
The θ integration is described by I1 and I2 . They appear in Eq. (9.1.21) in Abramowitz

and Stegun as π
eika cos θ sin ψo dθ = π J0 (ka sin ψ0 )
π
0
eika cos θ sin ψo cos θdθ = iπ J1 (ka sin ψ0 )

0
Substitution of these integrals into the Eq. (4) leads to
Pff (x̄0 ) ik L a −ikr

= e [J0 (ka sin ψ0 ) + i Z rad sin ψ0 J1 (ka sin ψ0 )]
ρ0 cv0 2 r
(6)
sin [(k L cos ψ0 + μL)/2] sin [(k L cos ψ0 − μL)/2]
× +
(k L cos ψ0 + μL) (k L cos ψ0 − μL)
It should be noted that the farfield pressure decays reciprocally to the radial distance.
This must be so because no matter how long the cylinder is, the farfield specification
places the field point sufficiently far away that the cylinder seems to occupy a single
point.
For each combination of ka and L/a, we compute (r/a) |Pff | / (ρ0 cv0 ) as a func-
tion of ψ0 . (The parameter k L is written as ka (L/a) for this computation.) Figure 2
displays polar plots for the four parameter combinations. Proper interpretation of
these plots requires that one recognize that because of the axisymmetry of the sys-
tem, these plots describe the view in any plane that contains the cylinder’s axis. The
overall picture is that only the shorter cylinder at the lower frequency radiates strongly
in all directions. This is not surprising because our earlier studies have indicated that
at very low frequencies, bodies radiate as a monopole whose strength depends on
the complex amplitude of the volume velocity. For the longer cylinder in both cases,
as well as the shorter cylinder at high frequency, the radiation is confined to a small
range of angles around ψ0 = 90◦ , which is said to be “broadside” or “beam aspect,”
both of which are nautical in their origin. Another trend displayed by the graphs is
that increasing the frequency increases the maximum farfield pressure.
90 90
120 60 120 60
1.0 2.5
150 30 150 30
180 0 180 0
L/a=4, ka=1 L/a=4, ka=6
90 5 90
120 60 120 60
5 15
150 30 150 30
180 0 180 0
L/a=20, ka=1 L/a=20, ka=6
Figure 2.
A fault of polar plots is that they obscure data close to the origin. For this reason,
the same data is plotted in Fig. 3 in abscissa–ordinate form. These graphs vividly
show how narrowly the radiation is aimed in all cases except the short cylinder at
the low frequency. In the intermediate cases, there are many side lobes, but they are
much smaller than the main lobe. The side lobes for the long cylinder at the high
frequency are negligibly small.
L/a = 4, ka = 1 L/a = 4, ka =6
1 2.5
0.8 2
0.6 1.5
0.4 1
0.2 0.5
(r/a)|Pff |/(ρ0cv0)
0 0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
L/a = 20, ka = 1 L/a = 20, ka =6

5 15
4
10
3
2
5
1
0 0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Polar angle ψ0 (deg.)
Figure 3.
Multipole Expansion of the KHIT

The initial discussion noted that the KHIT essentially describes the pressure field as
the superposition of monopoles and dipoles on the vibrating surface. However, the
description of the directivity derived in the previous section shows no evidence of a
set of sources. The first effort in the following will resolve this apparent conflict. By
itself, the development will merely provide a different perspective. However, further
analysis based on a restriction to acoustically compact radiators (size substantially
less than a wavelength) will lead to a recipe for actually evaluating the pressure field.
General Description We begin with a Taylor series expansion of the Green’s function
terms. A central point C inside the surface S, such as the centroid of the vibrating
body, is defined to be a reference location. Then, the location of any point on S is
¯ which is
described by the relative position ξ,
ξ¯ = x̄s − x̄C (7.4.20)
Because x̄C is a specified fixed point, we may consider the surface response to be
¯ An explicit description of the Green’s function as a function of ξ¯
functions of ξ.
results from application of Eq. (6.5.82). Doing so gives
1 ¯ 2
G (x̄0 , x̄s ) = G (x̄0 , x̄C ) + ξ¯ · ∇C G (x̄0 , x̄C ) + ξ · ∇C G (x̄0 , x̄C ) + · · ·
2
(7.4.21)
where the notation ∇C serves to emphasize that the gradient is evaluated at x̄C . The
normal derivative of G (x̄0 , x̄s ) in the KHIT is taken at x̄s . To put it on the same basis
as the above representation of G (x̄0 , x̄s ), we invoke Eq. (6.5.86), which states that
∇s G (x̄0 , x̄s ) = −∇0 G (x̄0 , x̄C ). For the present purpose, it is adequate to truncate
Taylor series at second derivatives, so we use the preceding to represent the Green’s
function. Doing so yields

n̄ ξ¯ · ∇s G (x̄0 , x̄s ) = −n̄ ξ¯ · ∇0 G (x̄0 , x̄C ) + ξ¯ · ∇C G (x̄0 , x̄C ) + · · ·
= −n̄ ξ¯ · ∇0 G(x̄0 , x̄C )
+ n̄ ξ¯ · ∇0 ξ¯ · ∇0 G (x̄0 , x̄C ) + · · ·
(7.4.22)
The dipole and quadrupole strengths were described in Chap. 6 in terms of deriva-
tives at the field point, with the source point defined to be the origin of a Cartesian
coordinate system. Therefore, we convert the preceding expression to scalar form
by explicitly describing the gradients. In the following, x1 , x2 , x3 are the Cartesian
coordinates of x̄0 . (In most situations, we use the free-space Green’s function and
designate x̄C as the origin.) Substitution of the series expansion of G (x̄0 , x̄s ) and its
gradient into the KHIT, followed by collection of the coefficients of each order of
derivative, yields
⎡

3
∂
P (x̄0 ) = 4π ⎣ AG (x̄0 , x̄C ) + Dj G (x̄0 , x̄C )
∂x j
j=1
⎤ (7.4.23)
3
3
∂ 2
+ Q j,m G (x̄0 , x̄C ) + · · ·⎦
j=1 m=1
∂x j ∂xm
The derivatives of the Green’s function are described by Eq. (6.5.123). The source
strength, dipole moments, and quadrupole strengths appearing in this expression are

1
A= ikρ0 c V̄ ξ¯ · n̄ ξ¯ dS
4π
S
1
Dj = − ikρ0 c V̄ ξ¯ · n̄ ξ¯ ξ j + P ξ¯ n j ξ¯ dS
4π
S
1
Q̂ j,m = ikρ0 c V̄ ξ¯ · n̄ ξ¯ ξ j ξm + P ξ¯ n j ξ¯ ξm + n m ξ¯ ξ j dS
8π
S
(7.4.24)
The expression for Q̂ j,m has a symmetric form because a mixed derivative does not
depend on the order of differentiation. The total strength of the lateral quadrupoles
is Q j,m ≡ 2 Q̂ m, j .
By itself, this representation generally is not useful because convergence will
require many higher order poles. The exception is low frequencies, in which case,
the body is compact. That is, its largest dimension a is much smaller than the acoustic
wavelength, so ka 1. In such situations, the preceding should be adequate. Indeed,
the monopole contribution, whose strength is proportional to the volume velocity of
the source, often is sufficient.
A Vibrating Rigid Body The multipole expansion provides an interesting perspec-
tive when the object from which the signal emanates is a rigid body. We restrict our
attention to acoustically compact bodies, or equivalently, to very low frequencies.
The theorems of kinematics state that the movement of such an object is the combi-
nation of a translation that follows an arbitrary point in the body, and a rotation in
which the arbitrary point does not move. We shall use the centroid C of the body as
the arbitrary point, and define this point to be the origin of the coordinate system.
The translational velocity v̄C and angular velocity ¯ are taken to be harmonic at
the same frequency. The complex velocity amplitude of any point in the body is
¯ × x̄
V̄ = V̄C + (7.4.25)
The normal velocity on the surface is found by evaluating this expression at surface
point x̄s , then taking the dot product with the surface normal n̄ (x̄s ).
The result is used to evaluate the source strengths in Eq. (7.4.24). For the monopole
amplitude, we have

1
A= ikρ0 c n̄ ξ¯ · V̄C + ¯ × ξ¯ dS (7.4.26)
4π
S
We apply the divergence theorem to this expression, and recognize

that V̄C is inde-
¯ × x̄ is identically zero.
pendent of position, so that ∇ · V̄C = 0. In addition, ∇ ·
Consequently, A = 0. The monopole amplitude is zero because the nature of rigid
body motion is that outward movement of some regions of the surface is balanced
by inward motion of other regions, so the net volume velocity is zero.
The dipole moment obtained from Eq. (7.4.24) is

1
Dj = − ikρ0 c V̄C ξ j · n̄ ξ¯ +
¯ × ξ¯ ξ j · n̄ ξ¯ dS
4π
S

1
− P ξ¯ n j ξ¯ dS, j = 1, 2, 3 (7.4.27)
4π
S
We use the divergence theorem to convert the first integral to a volume integral,

1
D j = − ikρ0 c ∇ · V̄C ξ j + ∇ · ¯ × ξ¯ ξ j dV
4π
V

1
− P ξ¯ n j ξ¯ dS, j = 1, 2, 3 (7.4.28)
4π
S

Both terms in the first integrand have the form ∇ · Ū ξ j . The gradient is taken at the
surface point. Thus, ∇ = ē1 (∂/∂ξ1 ) + ē2 (∂/∂ξ2 ) + ē3 (∂/∂ξ3 ). Expanding Ū into
its components gives
∂ ∂ ∂
∇ · Ū ξ j = U1 ξ j + U2 ξ j + U3 ξ j = U j + ξ j ∇ · Ū (7.4.29)
∂ξ1 ∂ξ2 ∂ξ3
¯ so ∇ · V̄C = 0.
In the first term in Eq. (7.4.28), Ū = V̄C . This term is independent of ξ,
For the second term, Ū = ¯ × ξ.¯ The same vector identity that led to vanishing of

the monopole amplitude states that ∇· ¯ × ξ¯ is zero. Thus, we have reduced the
dipole moment components to

1 1
Dj = − ikρ0 cV V̄C · ē j − ikρ0 c ¯ × ξ¯ · ē j dV
4π 4π
V

1
− P ξ¯ n j ξ¯ dS (7.4.30)
4π
S
The fact that we have designated point C to be the centroid of the body means that
the first moment of position is zero, that is,

ξ j dV = 0 (7.4.31)
V
Thus, the middle integral vanishes. The dipole moment vector that results from adding
the individual components of the remaining terms is
⎡ ⎤

1 ⎣
D̄ = − ρ0 V iω V̄C + P ξ¯ n̄ ξ¯ dS ⎦ (7.4.32)
4π
S
To interpret the dipole moment, we observe that ρ0 V is the mass m disp of the
body of fluid displaced by the vibrating body, and the complex amplitude of the
acceleration of point C is iω V̄C . Hence, the first term in the bracket is the force that
would be required to impart to the displaced fluid the accelerationof the vibrating
body. The second term also has a familiar explanation. Recall that n̄ ξ¯ is the normal
to the vibrating body pointing into the fluid. Thenormal
force exerted by the fluid on
a surface patch dS of the vibrating body is −n̄ ζ̄ P ξ¯ dS. The resultant pressure
force F̄ p exerted by the fluid is the surface integral of this term,

F̄P = − P ξ¯ n̄ ξ¯ dS (7.4.33)
S
It follows that the dipole moment in Eq. (7.4.32) may be written as
1
D̄ = − m disp iω V̄C − F̄P (7.4.34)
4π
The force exerted on the fluid by the body is the reaction, that is, it is − F̄ p . Thus,
the dipole moment is −1/4π times the sum of the force required to accelerate the
displaced fluid mass and the resultant force exerted on the fluid by the vibrating body.
To make this relation useful, it is necessary to quantify F̄P . Doing so requires an
analysis specific to the vibrating body’s shape. Nevertheless, certain aspects can be
anticipated. For low frequencies, the pressure should be proportional to the body’s
acceleration, but opposed to it. Therefore, F̄ p must be proportional to −iω V̄C . Given
that the first term in the dipole moment is proportional to the displaced mass, it is
reasonable to anticipate that F̄P is proportional to the density of the fluid. An aspect
that might not be apparent is that F̄P is not necessarily parallel to the acceleration.
A vibrating thin disk exemplifies this behavior because the resultant of the pressure
distribution must be perpendicular to the flat surface of the disk regardless of how
the disk is oriented.
The corollary of the possible misalignment of F̄P and iω V̄C is that the propor-
tionality is tensorial in nature. To describe it, we shall switch to a matrix description.
Let {FP } and {VC } be column vectors that are populated by the components of the
respective quantities. The preceding general observations lead to definition of the
factor of proportionality as a square matrix [W ], such that
{FP } = −iωρ0 [W ] {VC } (7.4.35)
The meaning of relation becomes somewhat clearer if we represent [W ] in partitioned

form as a column vector, that is
⎧ ⎫ ⎡ T ⎤
⎨ Fx ⎬ W (x)
⎢ ⎥
Fy = −iωρ0 ⎣ W (y) T ⎦ {VC } (7.4.36)
⎩ ⎭ T
Fz W (z)
This representation shows that the components of F̄P in the x, y, and z directions are,
respectively, the projection of V̄C onto vectors W̄ (x) , W̄ (y) , and W̄ (z) . Only if [W ] is
proportional to the identity matrix will F̄P be parallel to V̄C . Dimensional consistency
requires that the elements
of [W ] be volume quantities. Newton’s second law states
that F̄e + F̄P = M iω V̄ , where F̄e is the external force causing the body to move
and M is the body’s mass. Substitution for F̄ p gives
{Fe } = iω [M [I ] + ρ0 [W ]] {VC } (7.4.37)
This expression leads to ρ0 [W ] being referred to as the virtual mass, or the added
mass.
Although determination of [W ] requires further analysis specific to the body’s
shape, certain properties are general. The same vector F̄ p must result from a specified
vector V̄C , regardless of the orientation of the coordinate system associated with the
respective components. This condition requires that [W ] be symmetric. A special
property applies if the body is axisymmetric. Let this axis be z. The resultant force
then must lie in the formed by V̄c and ēz because the differential surface pressure
plane
resultant −P ξ¯ n̄ ξ¯ dS on points on either side of this plane balance. In addition,
the force component transverse to z must be invariant as the body is rotated about that
axis with V̄C fixed. These conditions require that [W ] be diagonal, with W1,1, = W2,2 .
Two shapes of interest are a sphere and a thin disk,15 for which the nonzero elements
of the respective matrices are
2 3 8
Wsphere j, j
= πa , j = 1, 2, 3; (Wdisk )3,3 = a 3 (7.4.38)
3 3
Determination of [W ] allows us to describe the dipole moment in matrix form.
The relation is
iωρ0
{D} = − [V [I ] + [W ]] {Vc } (7.4.39)
4π
15 Pierce, ibid, p. 427.

Knowledge that the monopole part of the radiated field is zero, combined with the
ability to evaluate the dipole contribution, usually is adequate to determine the pres-
sure radiated by a rigid body vibrating at low frequencies. This is so because the
dipole component will dominate the higher order terms. However, in some cases, it
will be found that D̄ = 0̄. This situation will occur if the body oscillates in a pure
rotation about the centroid C, so that V̄C is zero. The pressure resultant is zero in such
a motion because some pairs of points on the body’s surface move in opposite sense
as a result of the rotation. Consequently, the surface pressure distribution shows a
180◦ change of phase, with the net result that different regions cancel. In other words,
the small ka field radiated by a body that executes a pure rotation will be dominated
by various quadrupoles.
7.5 Numerical Methods for Radiation from Arbitrary

Objects
The techniques we have developed are limited to specific configurations. Determi-

nation of the field radiated by a body that has an arbitrary shape requires a numerical
simulation. We will examine three approaches having fundamentally different foun-
dations. The first is the source superposition method, which is based on the notion that
the radiating body may be replaced by a set of point sources in free space. Another
approach converts the Kirchhoff–Helmholtz integral theorem to an integral equation
for the surface pressure. A numerical analysis of this equation leads to the boundary
element method. Both of these approaches are founded on acoustical analyses we
have already carried out. The third is the finite element method, which is a general
tool for solving field equations associated with the various media. We will see that
each method has positive aspects and liabilities. The objective is to familiarize the
reader with the available alternatives. Hence, our investigations will focus on the
fundamental issues, rather than the details of their implementation. Each approach
has several variants, but we will not delve into them.
7.5.1 Source Superposition
The concept here is quite simple—replace the radiating body with a collection of
point sources. The idea is quite natural, but it leads to questions of how many sources
are needed, where should they be placed, and how strong should each source be?
Answering these questions is the contribution of the work by Koopman, Song, and
Fahnline.16 They began by considering a continuous sheet of sources distributed
over a surface interior to the radiating body. This was done to prove that the method
is equivalent to the Kirchhoff–Helmholtz integral theorem. That is the junction at
which we begin.
16 Koopman, Song, and Fahnline, “A method for computing acoustic fields based on the principle
of wave superposition,” J. Acoust. Soc. Am., 86 (1989), 2433–2438.

7.5 Numerical Methods for Radiation from Arbitrary Objects 105
Fig. 7.19 Sources interior to

a body whose normal Vs(xs)n(xs)
component of surface
velocity is Vs (x̄s )
xs
S
ξn
Figure 7.19 depicts a body surrounded by an ideal fluid. Its surface S executes a
specified
vibration in which the velocity normal to the surface, positive into the fluid,
is Re Vs (x̄s ) eiωt . Distributed throughout the region surrounded by S is a set of N
point sources whose strengths will be determined. The objective is to determine the
pressure at field point x̄0 .
The source superposition method may be used to analyze two- and three-
dimensional situations. For the two-dimensional case, G (x̄, x̄n ) is described by
Eq. (7.3.47), and Eq. (6.4.31) describes the three-dimensional version. The pressure
is the superposition of the contribution of each source. The strength of that contribu-
tion is the complex amplitude of the mass acceleration, which is iωρ0 Q̂ n , with Q̂ n
being the volume velocity amplitude. Thus, the pressure at any point x̄ is given by

N
P (x̄) = iωρ0 Q̂ n G (x̄, x̄n ) (7.5.1)
n=1
An expression for the particle velocity is obtained by applying Euler’s equation

to the preceding. The gradient is evaluated at x̄, and the free-space Green’s function
only depends on the distance between x̄n and x̄, so that
N
x̄ − x̄n d
V̄ (x̄) = − Q̂ n G (x̄, x̄n ) , rn = |x̄ − x̄n | (7.5.2)
n=1
r n dr n
To determine the unknown volume velocities, we require that the particle velocity
matches the known normal velocity distribution Vs (x̄s ). We cannot do so at every
point on the surface, so we select a set of surface points, designated ξ¯m , at which this
condition will be imposed. Let M be the number of surface points selected for this
purpose. Then, we have
# $

N
ξ¯m − x̄n d
Q̂ n n̄ ξ¯m · G ξ¯m , x̄n = −Vs ξ¯m , m = 1, 2, ..., M (7.5.3)
n=1
rn drn
The only quantities that are not set are the volume velocities. Thus, we have developed
M equations for these parameters.
Solution of these equations is expedited by writing them in matrix form. The
volume velocities are arranged sequentially in a column vector. The coefficient of
Q̂ n in equation #m is an element of the M × N rectangular array [U ],
# $
ξ¯m − x̄n d
Um,n = n̄ ξ¯m · G ξ¯m , x̄n (7.5.4)
rn drn
Thus, the task is to solve ( )

[U ] Q̂ = − {Vs } (7.5.5)
At this juncture, selection of the x̄n locations for the sources and of the ξ¯m locations
for velocity matching enters the picture. If the number M of the latter is less than
N , then there are more sources than the number of available equations. Thus, it is
necessary that M at least equals N . Koopman, Song, and Fahnline set M = N , but
this choice is problematic. If we were to select a different set of points to perform
the matching, then it is likely that the Q̂ n values would be different. However, based
on the proof that the source superposition and KHIT formalisms are equivalent, we
expect the volume velocities to be unique. Thus, setting M = N and increasing that
number should eventually lead to a convergent result.
An alternative that has been followed matches the surface velocity at more points
than the number of sources, M > N . Suppose we have identified a tentative solu-
tion for { Q̂} using M > N . Because there are more equations than the number of
unknowns, it is unlikely that { Q̂} will satisfy all equations. A velocity matching error
may be formed by substituting the solution into Eq. (7.5.5), then comparing the two
sides of the equation. This gives an error residual {E} whose definition is
( )
{E} = [U ] Q̂ + {Vs } (7.5.6)
A single error metric E 2 is the summed magnitude squared of all elements of {E}.
This quantity is useful because it cannot be negative, so minimizing it assures that all
elements of {E} are a minimum. Another term for this metric is that it is the square
of the Euclidean norm of {E}, which is obtained as a Hermitian dot product. The
Hermitian of a matrix is the complex conjugate of its transpose, so the product is

M ( )H ( ) ( )H
E2 ≡ |E m |2 ≡ {E}H {E} = Q̂ [U ]H [U ] Q̂ + Q̂ [U ]H {Vs }
m=1 ( ) (7.5.7)
+ {Vs } [U ] Q̂ + {Vs } {Vs }
H H
We wish to find the set of source strengths that minimize E 2 , which means that
the value of E 2 must be stationary with respect to each Q̂ j . References usually only
discuss the case where all quantities are real. The case where these quantities are
complex is slightly more complicated, but the result is only different by replacing
transpose operations by Hermitians, that is,
( )
[U ]H [U ] Q̂ = − [U ]H {Vs } (7.5.8)
This formulation is said to be the method of linear least squares. The coefficient
matrix [U ] has M rows (the number of velocity equations) and N columns (the
number of sources). Thus, [U ]H [U ] is N × N . The inverse of this matrix is called
the Morse–Penrose inverse, but any method may be used to solve for Q̂. After those
values have been obtained, the pressure at any field point may be determined from
Eq. (7.5.1).
The formulation seems to be straightforward, but its implementation requires that
several issues be addressed. Most significant is the possibility that [U ]H [U ] is ill-
conditioned. As explained by Ochmann,17 this situation might occur if the radiating
body’s shape is drastically different from spherical, as would be the case with a long
cylinder. Another reason might be that the distribution of sources is inconsistent with
the symmetry of the body. For example, a rectangular box should have sources placed
equally on both sides of each of its midplanes. Ill-conditioned situations typically are
addressed by application of singular value decomposition,18 but doing so requires
greater computational effort.
The question still remains as to where the sources should be located. Koopman
et al. provide some guidelines, as does Ochmann.19 Because the equations are quite
easy to formulate, practitioners typically use a very large number of sources placed
consistently with the symmetry of the system. For example, consider radiation from
a cylinder whose surface vibration is axisymmetric. Many sources should be dis-
tributed along the cylinder’s axis in order that the field has no variation around any
circumferential circle. When this tactic is used, then the number of surface points ξ¯m
for matching the surface velocity would be chosen to be much greater than N . Some
users distribute the source and surface locations uniformly, and some use a random
number generator for that purpose.
Regardless of how one selects these locations, the result should be verified. One
way of doing so is to redo the analysis with another set of source and surface points.
Another way to verify the solution for Q̂ n values is to use Eq. (7.5.2) to evalu-
ate n̄ (x̄s ) · V̄ (x̄s ) at several points on the surface that were not used to formulate
Eq. (7.5.8). The degree to which this set of normal velocities differ from the known
values of Vs (x̄s ) provides a measure of the error in the method.
17 M. Ochmann, “The full-field equations for acoustic radiation and scattering,” J. Acoust. Soc. Am.
105 (1999) 2574–2584.

18 W.H. Press, S.A. Teukolsky, W.T. Vetterling, & B.P. Flannery, Numerical Recipes, 3rd Ed., Chap.
12, Cambridge University Press (2007).

19 M. Ochmann, “The source simulation technique for acoustic radiation problems,” Acustica 81,
512–527 (1995).
7.5.2 Boundary Element Method
A very different approach is founded on the KHIT. A preview of this method arose
in Example 7.9, where the KHIT was used to evaluate the pressure at field points
that are very close to the surface. We begin with the radiation version of KHIT, in
which the field point x̄0 is exterior to the vibrating body. The fundamental equation
we seek has the field point on the surface. This is not a trivial matter because the
integrand of the KHIT is singular when the surface point x̄s at which the integrand
is evaluated coincides with x̄0 .
The Surface Helmholtz Integral Equation
The singularity of the KHIT integrand is handled by bringing the field point to the
surface in a limiting process. In Fig. 7.20, the field point is at distance ε from the
surface measured in the direction of the surface normal that intersects this field point.

The surface point for this normal is designated x̄0 , so that the field point’s position

is x̄0 = x̄0 + εn̄ x̄0 .
Fig. 7.20 Placement of a

field point close to a
x0
vibrating surface n(x0‘ )
ε
S0
x‘0
S1
The idea is to decompose the surface into two parts: S0 is a small region that
surrounds x̄0 and S1 is the remainder of S. Because x̄0 is situated in the fluid domain,
we set χ = 1 in Eq. (7.4.6), so the KHIT may be written as
P (x̄0 ) = I0 + I1
Ij = [iωρ0 n̄ (x̄s ) · V (x̄s ) G (x̄0 , x̄s ) + P (x̄s ) n̄ (x̄s ) · ∇G (x̄0 , x̄s )] dS
Sj
(7.5.9)
In order for n̄ x̄0 to be a unique normal vector, it must be that S is smooth at that
location, that is, there must be a unique tangent plane there. (This excludes sharp
corners, which we will address after the basic derivation.) When the geometry fits
this specification, we may consider S0 to be a small circle of radius lying in the
tangent plane and centered on x̄0 . This is the configuration depicted in Fig. 7.21.
The procedure we will follow is to consider the limiting behavior of the integral
over S0 as ε → 0 with held at a small but finite value. After that limit is established,
Fig. 7.21 Definition of a

spherical coordinate system x0
for evaluating the principal
part of the KHIT when the ε ψ
field point is situated on the r
vibrating surface
n(x0‘ ) S0
x‘0
S1 Δ R xs
we will take the limit as → 0. The resulting expression for the integral over
S0 is called the Cauchy principal part of the integral over S. We will derive the
result for a three-dimensional system, so the free-space Green’s function will be
G (x̄0 , x̄s ) = e−ikr / (4πr ).
The geometric properties of the terms in the integral over S0 are described in
term of a spherical coordinate system centered on x̄0 . The radius is small, and it
eventually will be reduced to zero, which means that the surface pressure and normal
velocity may be taken to be constant over S0 at their values at the center point. Thus,
the portion of KHIT associated with this region is

I0 ≈ iρ0 ωVs x̄0 G (x̄0 , x̄s ) dS + P x̄0 n̄ (x̄s ) · ∇G (x̄0 , x̄s ) dS
S0 S0
−ikr
1 e 1 x̄s − x̄0 d e−ikr
= iρ0 ωVs x̄0 dS + P x̄0 ēz · dS
4π r 4π r dr r
S0 S0
(7.5.10)
Each of the terms in the integrand is readily described in terms of the spherical
coordinate system in Fig. 7.21. The axis of this coordinate system is normal to the
circle, so the geometry is asymmetric. Consequently, we may use a ring element
to describe the differential area, so that dS = 2π Rd R. The value of R is related to
the radial distance by R 2 = r 2 − ε2 . The differential of this relation is Rd R = r dr
because ε is the constant distance to the surface. The range of radial distances covered
1/2
by S0 is ε ≤ r ≤ ε2 + 2 . This change of variables converts the integral to
2 2 1/2
1 (ε + ) −ikr
I0 = iρ0 ωVs x̄0 e dr
2 ε
2 2 1/2
1 (ε + ) x̄s − x̄0 1 −ikr
+ P x̄0 ēz · −ik − e dr (7.5.11)
2 ε r r
The dot product in the second integral is expressible in terms of ε and the radial
distance, specifically,
(x̄s − x̄0 ) ε
ēz · = − cos ψ = − (7.5.12)
r r
In addition, the smallness of ε and leads to simplification of both integrals, because
we may set exp (−ikr ) = 1 and approximate the last factor in the second integrand
as −1/r . The result is that the integral over S0 reduces to
2 2 1/2 2 2 1/2
1 (ε + ) 1 (ε + ) ε
I0 = iρ0 ωVs x̄0 dr + P x̄0 dr
2 ε 2 ε r2
1 2 1/2
1 1 1
= iρ0 ωVs x̄0 ε + 2 − ε + P x̄0 ε − 1/2
2 2 ε ε2 + 2
(7.5.13)
The limit of I0 as ε → 0 is
1 1
I0 = iρ0 ωVs x̄0 + P x̄0 (7.5.14)
2 2
When we take the limit as → 0, the first term vanishes but the second is unaltered.
In regard to the integral over S1 , we recognize that it approaches the original integral
over the entire surface S, except that the singular point, x̄s = x̄0 , is excluded from
the domain. (How this is done in a computational scheme will be discussed.) Thus,
to evaluate Eq. (7.5.9), we carry out a regular integration over the entire surface S
with x̄0 excluded, then add P x̄0 /2. When we bring the latter term to the left side,
we find that P x̄0 /2 equals the regular integral.
Let us step back to take an overview of the KHIT before we write the final form.
The left side is χP (x̄0 ). Previously, we found that χ = 1 if x̄0 is outside the radiating
body, and χ = 0 if x̄0 is inside the radiating body. Here, we brought x̄0 to the surface.
Doing so led to χ = 1/2. This result was derived by assuming that the surface is
smooth. If x̄0 is situated at a corner, then the smoothness condition is not met. That
situation is addressed by the concept of a solid angle. Figure 7.22 shows the four
cases for x̄0 : (a) exterior, (b) interior, (c) smooth surface, and (d) corner. For each, a
small sphere of radius δ is centered on x̄0 .
The portion of the surface area of the sphere that lies in the fluid domain is defined
to be δ 2 , where is the solid angle. Thus, = 4π for case (a), = 0 for case (b),
x0 x0
x0
x0 S
S S S
(a) Γ=4π, χ=1 (b) Γ=0, χ=0 (c) Γ=2π, χ=0.5 (d) 2π< Γ< 4π, χ=Γ/(4π)
Fig. 7.22 Parameter χ for the KHIT is obtained by surrounding the field point by a small sphere of
radius δ. The solid angle δ 2 is the portion of the sphere’s surface area that is situated in the fluid
= 2π for case (c), and is between 2π and 4π for case (d) because more than half
the sphere is in the fluid at that corner. The value of at a corner depends on whether
the slope is discontinuous in one direction or two orthogonal directions. For example,
if x̄0 is at the edge of cylinder with flat ends, then = 3π, whereas = (7/8) 4π if
x̄0 is at the corner of a rectangular box. The parameter χ (x̄0 ) describes the fraction
of a full sphere that lies in the fluid. In general, it is

χ (x̄0 ) = (7.5.15)
4π
An extremely useful feature stems from the fact that KHIT applies to acoustic
cavities, as well as radiation problems. The sole modifications required to treat either
situation is that n̄ must point into the fluid and the surface normal velocity must be
defined to be positive if it is in the sense of n̄. It follows that a similar modification
of the SHIE will make it applicable to cavities. In that case, the solid angle would be
defined by the extent of the interior fluid surrounding x̄0 .
The KHIT for all cases is described by

1 x̄s − x̄0
χ (x̄0 ) P (x̄0 ) = iωρ0 Vs (x̄s ) − P (x̄s ) n̄ (x̄s ) ·
4π |x̄n − x̄s |
S −ik|x̄n −x̄s | (7.5.16)
ik |x̄n − x̄s | + 1 e
× dS
|x̄n − x̄s | |x̄n − x̄s |
It is implicit to this expression that x̄s = x̄0 is excluded from the integration domain if
x̄0 is on the surface. In that case, the equation is called the surface Helmholtz integral
equation. We will use the abbreviation SHIE to refer to it.
Discretized Implementation
In a radiation problem, we seek the pressure field generated by a specified velocity
distribution on the vibrating surface. The SHIE states that the pressure on the sur-
face depends on the properties of the pressure on the surface. Just as an integral is
sometimes called an antiderivative, such a relation could be considered to be an “an-
tidifferential equation.” (Its proper description is that is Fredholm integral equation
of the second kind.) It often is necessary to use numerical methods to solve a differ-
ential equation, and that is the situation for the SHIE. This process begins by dividing
the surface into a set of small patches, which we will denote as S j . Each patch is one
element situated on the surface, from which the description as a boundary element
formulation follows.
The integral over a patch depends on the geometrical properties of the surface and
the spatial distribution of P (x̄s ) and Vs (x̄s ). Some formulations have used interpolat-
ing functions to describe these properties. We will adopt a simpler approach in which
both fields are considered to be constant over a patch. This is the formulation that
was implemented to develop CHIEF,20 which was one of the first effective numerical
20 H.A. Schenck, “Improved Integral Formulation for Acoustic Radiation Problems,” J. Acoust. Soc.
Am. 44 (1968) pp. 41–58.

codes given wide distribution. The meaning of the abbreviation will emerge in the
course of the development.
the location of the center of element S j , and define P j ≡ P x̄ j
Let x̄ j denote
and V j ≡ Vs x̄ j . Because x̄0 in the SHIE is situated on the surface, x̄0 is taken to be
one of these center points. The pressure and velocity are factored out of the integral
over a patch because the requisite smallness of all S j makes negligible their variation
over that region, Thus, Eq. (7.5.16) becomes
N −ik|x̄n −x̄s |
e
4πχn Pn = iωρ0 V j dS
j=1
|x̄n − x̄s |
Sj
N
ik |x̄n − x̄s | + 1 −ik|x̄n −x̄s |
− Pj n̄ (x̄s ) · (x̄s − x̄n ) e dS
j=1
|x̄n − x̄s |3
Sj
(7.5.17)
Both integrands depend only on the geometrical properties of S and the selection of
the points on S. Thus, the integrals constitute a set of coefficients that are independent
of the pressure and surface velocity,

ik |x̄n − x̄s | + 1 −ik|x̄n −x̄s |
Bn, j = n̄ (x̄s ) · (x̄s − x̄n ) e dS
|x̄n − x̄s |3
Sj
−ik|x̄n −x̄s | (7.5.18)
e
Cn, j = dS
|x̄n − x̄s |
Sj
These coefficients are evaluated by two different schemes. No singularity arises

in the off-diagonal terms, j
= n, because x̄n is not situated in S j . These terms are
evaluated by a two-dimensional Gaussian integration scheme, which uses the value
of the integrand at points interior to S j . A low-order scheme using four points is
illustrated in Fig. 7.23, but some investigations21 have suggested that more points are
required for accurate results.
y
a 3
4
4 Sj 1 b 3
f(x,y) dx dy = 4ab Σ 0.25f(xm, ym) 2b x
m=1 xj
Sj 3 2
2a
Fig. 7.23 A four-point Gaussian integration rule for integrating over surface patch S j
21 A.L. van Buren, “A Test of the Capabilities of CHIEF in the Numerical Evaluation of Acoustic
Radiation from Arbitrary Surfaces,” NRL Report 7160 (1970).

Evaluation of the diagonal coefficients Bn,n and Cn,n recognizes that the Cauchy
principal part gives the contribution of the first integral at x̄s = x̄n . The numerical
integration scheme excludes this point by further dividing the patch in Fig. 7.23 into
four quadrants indicated there by dashed lines. A sixteen-point Gaussian integration
rule is used for each subpatch. Because the Gaussian points are internal to each
subregion, evaluation of the integrands at x̄n is excluded.
The discretized form of the SHIE, Eq. (7.5.17), may be written in matrix form as

4π [χ] + [B] {P} = [C] {V } (7.5.19)
where [χ] is a diagonal array of the value of χ at each center point x̄n ; so χn,n = 1/2
except for points at corners and edges. Both [B] and [C] are N × N complex arrays,
which is the number of unknown P (xn ). It might seem that solving the equations for
the {P} given {V } is a straightforward task. However, this is only true if the domain
is an interior cavity. For the exterior radiation problem, there are certain frequencies
at which difficulties arise.
All cavities are like closed one-dimensional waveguides and spherical cavities, in
the sense that an infinite number of modes exist if the boundary is rigid. A mode is
a pressure field that can exists without excitation; the frequency of that field is the
natural frequency. Why is this important to the radiation problem? The KHIT for
interior and exterior domains have similar appearance. It was proven by Schenck22
that if the frequency matches a natural frequency of the domain inside the vibrat-
ing surface, then Eq. (7.5.19) has no unique solution. Some individuals refer to this
condition as a “forbidden internal cavity resonance.” If the frequency for a specific
computation was exactly one of these natural frequencies, the matrix [4π [χ] + [B]]
would be rank-deficient, so its inverse would not exist. The more common occur-
rence is that the overall computation entails a frequency sweep, and one or more of
the frequencies in the swept range is close to a natural frequency for the enclosed
region. Then, the matrix will be ill-conditioned. If one proceeds to solve for {P}
corresponding to this matrix, numerical errors will be greatly magnified, making the
solution unusable.
How to proceed in this situation depends on what one knows about the modes
of the domain contained inside S. In some special cases, like a sphere or a flat-
ended circular cylinder, the modal solution can be obtained analytically. In that case,
one could simply omit any results obtained from Eq. (7.5.19) for frequencies close to
natural frequencies. However, for arbitrary shapes, we seldom know what the interior
natural frequencies are.
Loss of uniqueness at the forbidden frequencies has been addressed in several
ways, each resulting in a different boundary element computer code. CHIEF appar-
ently was the first general procedure. It is based on a very simple idea. The approxi-
mation of the SHIE in Eq. (7.5.19) has reduced rank at a forbidden frequency, which
means that the number of independent equations is less than the size of {P}. Extra
equations are obtained by using the original Helmholtz integral equation, Eq. (7.4.6),
22 H.A. Schenck, ibid.

for the case where the fluid is exterior to the vibrating surface and the field point x̄0
is interior to it. The field point is not within the fluids, which means that χ = 0, so
the integral should evaluate to zero. As is done for the surface points, the values of
P and Vs are factored out of the integral over each S j , and Gaussian integration is
used to evaluate the contribution of the geometric terms in each integrand. Because
x̄0 is situated inside S, the integrand is never singular, so the Gaussian integration
scheme used to evaluate the off-diagonal terms of [B] and [C] is used here also.
Each interior field point to which this analysis we applied leads to an extra equation
to supplement what has been lost. The computer code name “CHIEF”, which is an
acronym for Combined Helmholtz Integral Equation Formulation, was the first to
implement this concept.
It is known that the rank of [B] is less than N at a forbidden frequency, but
the actual rank is not known. Without that knowledge, it is not known how many
equations for interior points are required to make the system of equations solvable.
Furthermore, as was noted, the interior field is a mode at a forbidden frequency. Such
modes have nodal surfaces, along which the pressure is zero. The Helmholtz integral
equation will give zero identically if x̄0 is situated on a nodal surface, so it will not
be an additional equation to supplement the equations for points on the surface. The
strategy that addresses both issues uses far more interior points than the number that
is required. Let ξ¯n be the positions of these interior points, whose location may be
selected in any convenient manner, including randomly. Let M be the number of
such points. Then, because χ = 0 for interior points, the numerical approximation
of the Helmholtz integral equation at such points gives

B {P} = C {V } (7.5.20)
where the coefficient matrices are

#

$
ik
ξ¯n − x̄s
+ 1 −ik |ξ̄ −x̄ |

Bn, j = n̄ (x̄s ) · x̄s − ξ¯n

e n s
dS

ξ¯n − x̄s
3
Sj
−ik |ξ̄n −x̄s | (7.5.21)
e
Cn, j =

dS

ξ¯n − x̄s
Sj
The equations for the interior points are stacked below Eq. (7.5.19) for the surface
points. There are M of the former and N of the latter, so the equations to be solved
are
+
4π [χ] [B]
{P} = [C] {V } (7.5.22)
B
This represents a total of N + M equations for the M values contained in {P}. The
equations are overdetermined, as they were for the source superposition method,
Eq. (7.5.5). This condition is handled in the same way, specifically by invoking the
method of linear least squares. The result is that the coefficient matrix is defined to be
an array [U ] whose size is (N + M) × N , and the equation to be solved is multiplied

by [U ]T . This leads to

[U ] = +
[4π [χ] [B]]
B (7.5.23)
[U ] [U ] {P} = [U ] [C] {V }
T T
After the solution for {P} has been obtained, the pressure at any field point exterior
to S is found by using the same formulation as that used for the interior points. If
the field point is in the farfield, this evaluation would be based on the farfield form
in Eq. (7.4.18).
CHIEF is just one of many boundary element codes. They differ in the way
that the variation of P (x̄s ) and Vs (x̄s ) over a patch is described, as well as how the
integral over the surface patches is obtained. In addition, some use a different integral
equation to handle the forbidden frequency issue. One that has had some popularity
is obtained by differentiating the Helmholtz integral equation to obtain an integral
equation for Vs on the surface. The combination of that equation and SHIE at the
forbidden frequencies is called the Burton–Miller formulation.23
Boundary element formulations have some common attributes. One is that they
tend to have a large number of variables. This is a consequence of a general guideline
that the surface patches should be approximately square, with each side being no
bigger than one-sixth of the acoustic wavelength at the frequency for the evaluation.
The number of equations is increased by the supplemental equations required to
handle the forbidden frequency issue. The requisite computation resources increase
substantially as the frequency is increased in a sweep. This is so because one has the
choice of setting the surface mesh to fit the one-sixth wavelength requirement at the
highest frequency, or else rezoning the mesh as the frequency increases. Furthermore,
all coefficient matrices are functions of frequency, so they must be recomputed at each
step in the sweep. Another attribute that causes a boundary element model to require
substantial computational resources is the fact that [B] and B are full matrices,
so only standard solution algorithms may be employed. This is contrasted by the
finite element method, whose matrices are diagonally dominant. Many algorithms
that greatly reduce core memory requirements and the number of operations have
been developed to solve such equations.
7.5.3 Finite Element Method
A variety of finite element formulations have been developed. We will consider a

simple one in order to expose the broad concepts, and then discuss ways in which
23 A.J.Burton and G.F. Miller, “The application of integral equation methods to the numerical
solution of some exterior boundary-value problems”, Proc. R. Soc. A 323 (1971) 201–210.
it may be used to study radiation into an infinite domain. As with the preceding
developments, we will only consider the frequency domain.
One can find finite element formulations for acoustics that employ displacement
as a primary variable, either individually, or in combination with pressure. We will
examine a simpler approach that only uses pressure. The derivation will apply the
Galerkin method to the Helmholtz equation. The first step is to decompose the domain
V of the fluid into small pieces, which are the finite elements. We shall denote a
generic element as Vn . Each element may have any shape; common ones are a box
or a tetrahedron. In general, a set of mesh points is defined in each Vn . Some of these
points may be interior to the element. However, placing some on the surface of the
element will assure continuity of pressure because those points will also lie on one
or more adjacent elements. Interpolation functions, typically polynomials, are used
to represent the pressure within an element.
To understand the basic aspects of the formulation, a one-dimensional waveguide
is described in Fig. 7.24. The waveguide is segmented into elements that are numbered
sequentially by index n = 1, 2, ...N . The length of element n is L n , which need not
be the same for all elements. A mesh point is located at the end of each element,
and no interior points are defined. A local coordinate xn is used to measure the axial
position within element n.
γ=1 γ=2 γ=3 γ=4 γ=5 γ=N γ =N + 1

n =1 n=2 n= 3 n =4 n= N
α =1 α= 2 α = 1 α = 2 α = 1 α= 2 α= 1 α =2 α= 1 α=2
L1 L2 L3 L4 LN
x1 x2 x3 x4 xN
Fig. 7.24 Numbering scheme for a finite element model of a one-dimensional waveguide
The mesh points within an element are numbered with the index α = 1 for the left
node and α = 2 for the right node. A superscript will be used to indicate which ele-
ment is under consideration if there is ambiguity. Therefore, the pressure at a node is
denoted as Pα(n) . The lowest order interpolating function is linear. The corresponding
description of the pressure distribution within an arbitrary element is
⎧
⎨
(n) xn xn
(n) P 1 − + P2(n) , 0 ≤ xn ≤ L n
P (xn ) = 1
Ln Ln (7.5.24)
⎩
0 otherwise
The interpolating functions are denoted as Nα(n) (xn ), where n is the element number
and α is the mesh point index. Thus, the pressure within an element may be written
as
*
2
P (n) (xn ) = Nα(n) (xn ) Pα(n) (7.5.25)
α=1
If one were to use higher order interpolating functions, the sole alteration in this
expression would be adjustment of the summation range to match the commensu-
rate increase in the number of mesh points per element. Because the interpolating
functions for an element are defined to be zero outside their element, the pressure
at any position may be considered to be a sum over all elements of the preceding
expression, that is,
*
N * 2
P= Nα(n) (xn ) Pα(n) (7.5.26)
n=1α=1
This representation does not account for the fact that some mesh points are shared
by elements, that is, P2(1) = P1(2) , P2(2) = P1(3) , ..., P2(N −1) = P1(N ) . This overlap is
described by assigning a unique index γ to each mesh point, and defining P̂γ to be
the pressure at that point. These values are the global pressures, whereas the Pα(n)
(n)
are the local values. The latter may be obtained by(using ) a filtering matrix S to
pick out the appropriate elements of the global set P̂ ,

P1(n) ( )
= S (n) P̂ (7.5.27)
P2(n)
For example, the filtering matrices for the first two elements in Fig. 7.24 are

1 0 0 0 ··· 0 1 0 0 ···
S (1) = , S (2) = (7.5.28)
0 1 0 0 ··· 0 0 1 0 ···
The [Sn ] matrices are more commonly referred to as the connectivity matrix.
Although these relations have been derived for planar waves in a waveguide,
they are applicable with a few modifications to a three-dimensional field. Position
within an element is described by a local coordinate system xn yn z n . Mesh points
are distributed over an element, and more than two points per element are required.
Other than the higher dimensionality, the basic relations are as above. The pressure
at any position x̄n within an element is a sum of terms consisting of an interpolating
function multiplied by the associated mesh pressure. The matrix representation of
this ansatz is

P (x̄n ) = N (n) (x̄n ) P (n) (7.5.29)
This is the fundamental description of the pressure field. Only the mesh pressures
are unknown. As before, the number of columns for the interpolating matrix equals
the number of mesh points in an element,
and it has one row.
The connectivity matrix S (n) still filters the pressures P (n) at an element’s
mesh points from the global set of pressures. The matrix representation of these
relations is
(n) (n) ( )
P = S P̂ (7.5.30)

Of course, the number of rows for S (n) will be the number of mesh points in
an element, which is greater than two. Equation (7.5.29) gives the pressure in a
specific element, and the interpolating functions are defined to give zero if the position
is outside that element. Thus, substitution of Eq. (7.5.30) into (7.5.29) gives the
contribution of the mesh points within an element to the entire field. Adding these
relations for all elements gives a representation of the pressure throughout the fluid,
N
* ( )
P (x̄) = N (n) (xn , yn , z n ) S (n) P̂ (7.5.31)
n=1
Various rules for finite elements correspond to using different interpolating functions,
and/or different shapes for the elements. The preceding will accommodate any rule.
The next task is to derive the equations governing the global pressures. Equation
(7.5.31) will not satisfy the Helmholtz equation. The remainder after its substitution
represents an error whose value will depend on the location within V. The Galerkin
method requires that this error be orthogonal to an arbitrary test function that lies
in the same functional space as the series for P. This means that the test function is
like Eq. (7.5.31), except that its coefficients are an arbitrary set of values ηn . In other
words
*N
= N (n) (xn , yn , z n ) S (n) {η} (7.5.32)
n=1
The orthogonality condition is defined relative to an inner product of functions

over the entire domain V. Thus, the condition required in the Galerkin method is

2
∇ P + k 2 P dV = 0 (7.5.33)
V
We introduce the identity that ∇ 2 P ≡ ∇ · (∇ P) − ∇ · ∇ P. Green’s theorem

converts the first term to a surface integral, so we now have

∇ P · ∇ − k 2 P dV = − [(n̄ · ∇ P) ] dS (7.5.34)
V S
Note that the signs in this expression correspond to definition of n̄ as the surface
normal that points into the fluid domain.
The series in Eqs. (7.5.31) and (7.5.32) are used to describe P and . To use the
matrix form, the components of the gradient operator are placed in a column vector
{∇},

∂ ∂ ∂ T
{∇} ≡ (7.5.35)
∂xn ∂ yn ∂z n
It will be noted that the gradient is expressed in terms of the local xn yn z n coordinate
system, because that is the coordinate system used for the interpolating functions.
The pressure gradient is obtained by operating on Eq. (7.5.31), which leads to

⎧ ⎫
⎨ ∂ P/∂xn ⎬ * N ( )
{∇} P = ∂ P/∂ yn = {∇} N (n) (xn , yn , z n ) S (n) P̂ (7.5.36)
⎩ ⎭ n=1
∂ P/∂z n
In this expression, only the interpolating

functions depend
on position. If J is the total
number of mesh points, then {∇} N (n) (xn , yn , z n ) is a 3 × J position-dependent
array given by
⎡ ⎤
∂N1(n) /∂xn ∂N2(n) /∂xn ∂N3(n) /∂xn · · ·
{∇} N (n) (xn , yn , z n ) = ⎣ ∂N1(n) /∂ yn ∂N2(n) /∂ yn ∂N3(n) /∂ yn · · · ⎦ (7.5.37)
∂N1(n) /∂z n ∂N2(n) /∂z n ∂N3(n) /∂z n · · ·
Changing P to in the preceding gives the matrix representation of ∇. A dot

product is obtained as a product of row and column vectors, so the first term in
Eq. (7.5.34) may be written as
∇ P · ∇ = ({∇})T ({∇}P)
*
N *N T T
= {η}T S (m) {∇} N (m) (xn , yn , z n ) (7.5.38)
m=1n=1
( )
× {∇} N (n) (xn , yn , z n ) S (n) P̂
Both {P} and {η} are global parameters, so they may be brought outside the sum.
Furthermore, the interpolating functions are zero outside their element. This means
that all terms in the double sum for which the elements are different are identically
zero. This reduces the double sum to a single sum over all elements, so that
N
* T T
∇ P · ∇ = {η} T
S (n)
{∇} N (n) (xn , yn , z n ) (7.5.39)
n=1
( )
× {∇} N (n) (xn , yn , z n ) S (n) P̂
The scalar product P is

N
* T (n) T ( )
P = {η}T S (n) N (xn , yn , z n ) N (n) (xn , yn , z n ) S (n) P̂
n=1
(7.5.40)
Now let us consider the right side of Eq. (7.5.34). This term is evaluated on the
boundary of the acoustic domain, so that n̄ · ∇ P = −iρ0 ωVs . Thus this integrand
may be written as
N
* T T
(n̄ · ∇ P) = −iωρ0 {η}T S (n) N (n) (xn , yn , z n ) Vs (xn , yn , z n ) (7.5.41)
n=1
The result of substituting these expressions into the Galerkin principle,

Eq. (7.5.34), is
( ) ( ) ( )
{η}T [A] P̂ − k 2 {η}T [B] P̂ = iωρ0 {η}T V̂ (7.5.42)
The matrices [A] and [B] are square and symmetric. Their definitions are

*
N T T
[A] = S (n) {∇} N (n) (xn , yn , z n ) {∇} N (n) (xn , yn , z n )
n=1

Vn
× S (n) dV

*
N (n) T (n) T
[B] = S N (xn , yn , z n ) N (n) (xn , yn , z n ) S (n) dV
n=1
Vn
(7.5.43)
The column vector {V̂ } contains the effect of the surface velocity that generates the
pressure field. It is given by
( ) *
N (n) T (n) T
V̂ = S N (xn , yn , z n ) Vs (xn , yn , z n ) dS (7.5.44)
n=1
S
The evaluation of {V̂ } entails an integration over the surface, but most finite elements
are on the interior. Those elements may be omitted from the sum. Also, it seldom
is convenient to evaluate the integral for the actual Vs function, especially in the
context of a finite element computer code. Rather, the usual practice is to use a set
of interpolating functions to represent the dependence of Vs on the surface position
in terms its value at those mesh points that lie on the surface.
Now we come to the crucial step. The test function is arbitrary, which means
that the η j coefficients are arbitrary. It is necessary that Eq. (7.5.42) be satisfied for
any {η}, which will only be true if the coefficients of {η}T in both sides of the equality
are equal. The result is the standard form of the finite element equations of motion
( ) ( )
[A] − k 2 [B] P̂ = iωρ0 V̂ (7.5.45)
Because k 2 is proportional to ω 2 , an individual who is conversant with the princi-

ples of structural dynamics might be tempted to call [B] the inertia matrix, and [A]
the stiffness matrix. In fact, this is the opposite of the effect each represents. The
kinetic energy density is (ρ0 /2) v̄ · v̄, and v̄ = − Re ((1/iρ0 ω) ∇ P). The integrand
for [A] contains the dot product of gradients, so it is the inertial effect. Similarly, the
acoustic potential energy per unit volume is p 2 / (2ρ0 c), and the integrand for [B]
contains a product of pressure functions. Hence, it is the stiffness term. The differ-
ence from the structural dynamics equations stems from the fact that we have used
Fig. 7.25 Finite element

domain used to analyze
radiation from the vibrating
body at the center S1
S0
pressure, which is a force-like quantity, as the basic variable, whereas displacement

is the basic variable in structural dynamics.
The formulation leading to Eq. (7.5.45) is quite general, but it does not address
the unique nature of a radiation problem in which energy is transported into an
unbounded space. The problem is that it is not possible to divide an infinite domain
into a finite number of elements. Individuals who have enormous computing power
at their disposal have represented that domain as a very large sphere. This is the finite
element model depicted in Fig. 7.25, where the radiating body is a hydrophone that
consists of a cylinder with vibrating pistons at both ends. The mesh depicted there
adheres strictly to a spherical coordinate grid, but it would likely be better to distort
the mesh such that the radial lines in the vicinity of the hydrophone are close to being
perpendicular to the surface.
Why is a spherical domain used? The answer lies in the farfield behavior of the field
radiated by any finite-sized object. We know that in terms of spherical coordinates
centered inside a radiating body, the pressure at a very large r is a spherical wave with
directivity. The complex pressure and particle velocity in that region are described by
f (ψ, θ) −ikr P
P= e , Vr = (7.5.46)
r ρ0 c
The surface S is a composite of the exterior of the hydrophone ( ) S0 and the outer
spherical surface S1 . Correspondingly, Equation (7.5.44) for V̂ is split into con-
tributions from each surface. The surface velocity on the hydrophone is known. On
S1 , we use the farfield approximation. Recall that Vs was defined to be positive if it is
into the fluid, so Vs = −Vr on the spherical boundary. Thus, the velocity coefficients
on S1 become

*
N T (n) T
{V̂ }1 = S (n) N (xn , yn , z n ) Vs (xn , yn , z n ) dS
n=1
S1
(7.5.47)
1 * N
(n) T
(n) T
= S N (xn , yn , z n ) P (xn , yn , z n ) dS
ρ0 c n=1
S1
The pressure in the integrand is described by evaluating Eq. (7.5.31) on the spherical
boundary. Doing so reduces the integrand to a product of interpolating functions,
with { P̂} brought outside the integral. The result is an equation for {V̂ }1 that depends
on { P̂}1 . Bringing these terms to the left side of Eq. (7.5.45) yields a set of equations
for the mesh pressures in which the velocity coefficients on the radiating body are
the inhomogeneous terms,
( ) ( )
[A] + k 2 [B] + ik [C] P̂ = iωρ0 V̂ (7.5.48)
0
where the additional contribution to the coefficient matrix is

*N (n) T (n) T
[C] = S N (xn , yn , z n ) N (n) (xn , yn , z n ) S (n) dS (7.5.49)
n=1
S1
An important feature is that the term containing [C] is 90◦ out-of-phase from
the contributions of [A] and [B]. As noted previously, [A] and [B] are associ-
ated, respectively, with the kinetic and potential energy of the fluid contained in
the finite element model. Both effects are conservative. In contrast, [C] represents a
loss of energy, specifically, radiation damping resulting from power flowing out of
the domain across the spherical boundary.
Although creating a model based on surrounding the radiating body with a very
large sphere seems to be a reasonable procedure, it has a serious flaw. We do not
know precisely where the farfield begins, but we do know that the sphere’s radius
r1 must be sufficiently large that adjacent lines from a point on the outer surface
to the vibrating body are essentially parallel. This means that the sphere’s radius r1
must be much greater than the largest dimension of the radiating body. The farfield
approximation also requires that kr1 1, that is, r1 must be much greater than an
acoustic wavelength. Furthermore, the general requirement that the element size’s
be one-sixth of an acoustic wavelength, which was identified for boundary elements,
also applies to finite elements. Meeting these criteria would require that the model
contains many mesh points. That is why this approach typically requires substantial
computational resources to analyze realistic radiation problems.
What should the rest of us do? Many concepts have been developed, all with
the idea of shrinking the size of the sphere, and possibly, adjusting the shape of the
outer boundary to conform more closely to the shape of the vibrating body. The
first modifies the spherical plane wave approximation to obtain a modified boundary
condition. If the radius of the sphere is sufficiently large, setting Vr = P/(ρ0 c) on the
outer body will result in the reflection coefficient being zero. If a boundary had a zero
reflection coefficient for any type of wave, we would say that is perfectly transmitting.
The reflection coefficient for a plane wave obliquely incident on a locally reacting
plane is
Z − Z1 ρ0 c
R= , Z1 = (7.5.50)
Z + Z1 cos ψI
One could think of selecting Z to minimize |R| over a range of ψI , and then using
that Z as the boundary condition for the exterior surface S of the finite element
mesh. That surface need not be spherical, so the approximate transmitting boundary
condition would be
P = −Z n̄ · V̄ (7.5.51)
where the negative sign results from defining n̄ to point into the domain represented
by the mesh.
This approximation does not work too well, so the next idea is to introduce dis-
sipation, in which case the result will be an absorbing boundary condition. This can
be done by allowing Z to be complex, or by introducing a dissipation operator into
the plane wave approximation. A simple absorbing boundary condition would be
ρ0 c n̄ · V̄ = − (P + αn̄ · ∇ P) (7.5.52)
where the value of α would be selected to minimize the signal that is reflected back
into the acoustic domain. The absorbing boundary conditions that are used are much
more complicated than these. An archival work is the paper by Bayliss and Turkel24 ;
a more accessible description is the paper by Assaad et al.25
The latest development is the concept of an infinite element. Such an element
uses the basic form of a radially diverging wave as the basis for its shape functions.
These elements are placed outside the fluid domain that is described by regular finite
elements. Most derivations have considered this domain to be spherical, which is the
configuration depicted in Fig. 7.26. The exception is the work of Burnett26 in which
the outer boundary of the fluid domain is a prolate spheroid, which can enclose a
long slender object without extending far in the broadside direction.
As shown in Fig. 7.26, some mesh points lie on the outer boundary S1 of the
finite element mesh. The field within the element is taken to be an outgoing spherical
wave with directivity. This dependence is expressed in terms of spherical coordinates
centered on the sphere, which leads to
24 A. Bayliss and E. Turkel, “Radiation BoundaryConditions for Wave-Like Equations,” Comm.

Pure Appl. Math., 33 (1980) pp.707–725.
25 J. Assaad, J.-N. Decarpigny, C. Bruneel, R. Bossut, and B. Hamonic, “Two-dimensional radiation
problems,” J. Acoust. Soc. Am., 94 (1993) 562-573.

26 D.S. Burnett, “A three-dimensional acoustic infinite element based on a prolate spheroidal mul-
tipole expansion,” J. Acoust. Soc. Am., 96, (1994) 2798–2816.

Fig. 7.26 Configuration of

an infinite element used to S1
approximate the free field in
a finite element model ψ θ 7
1 3 5
r
2 4 6 8
e−ikr Fm (ψ, θ)
M
P= (7.5.53)
r m=0 rm
Several methods have been used to set the Fm (ψ, θ) functions. One, described by
Burnett,27 selects the m = 0 function to match the regular finite element functions in
the two transverse directions, and then finds the others by fitting the expression to the
angular terms in the Laplacian operator. A different derivation based on a mapping
of the cylindrical coordinates into a hyperbolic grid was used by Astley et al.28 In
any event matching, the standard interpolation form in Eq. (7.5.31) to the diverging
spherical wave form leads to the interpolation functions for the infinite element. At
that juncture, the contribution of the infinite elements to [A] and [B] is folded into
those of the regular finite elements.
The advantages of the finite element formulation might not be apparent at this
juncture, especially because the one-sixth wavelength requirement also applies to a
finite element model. A boundary element model only entails a mesh on the surface,
whereas finite elements discretize the three-dimensional domain of the fluid. Hence,
a finite element model will have more mesh points, and therefore more equations to
solve, than a comparable boundary element model. The advantageous aspects of the
finite element model follow from its basic features. First, the coefficient matrices [A],
[B], and [C] are constants independent of frequency. Thus, in the likely event that it is
necessary to perform a frequency sweep, the basic terms for the equations of motion
are only evaluated once. In contrast, the coefficient matrices for a boundary element
model depend on frequency. Another difference is that the integrals for the evaluation
of coefficient matrices can be derived as standard formulas for a specified set of
interpolation functions. In contrast, the integrals for boundary element coefficients
contain singularities and therefore constitute a more formidable task to program and
evaluate. Some individuals find the absence of the “forbidden frequency” issue to be
a major asset of finite elements.
27 ibid.
28 R.J. Astley, G.J. Macaulay, J.-P. Coyette, and L. Cremers, “Three-dimensional wave-envelope
elements of variable order for acoustic radiation and scattering. Part I. Formulation in the frequency
domain,” J. Acoust. Soc. Am. 103 (1998), 49–63.
Perhaps the most significant difference between finite elements and boundary
elements may be found in the fundamental mathematical properties of their equations.
As was noted earlier, the coefficient matrix for boundary elements is full. In contrast,
the mesh points for a finite element model are shared by only a few elements, so
only a few elements are directly coupled. This means that the coefficient matrices
for a finite element model will be banded around the diagonal. Many highly efficient
methods have been developed to solve equations whose system matrices are banded.
Thus, although the finite element model has more mesh points, it tends to require
less computational resources.
An interesting feature of the finite element technique is that all developments,
including absorbing boundary conditions and infinite elements, have analogous for-
mulations in the time domain. Whether this is useful depends on what properties
of the response one wishes to obtain. Often, a complex frequency response is the
desired result. Extracting time domain data from a frequency domain simulation
would require FFT processing, thereby adding to the computational effort. On the
other hand, solving time domain equations of motion must be done with a numerical
differential equation solver, which require a great deal of computational resources if
it is necessary to a large number of coupled equations. In any event, the time domain
formulation lies more in the research arena because some features are still being
explored.
It is worth noting that the finite element formulation is equally valid for the analysis
of a cavity. The same is true for the boundary element formulation, if the direction of
the surface normal is reversed. Several commercial codes of both types are available
for such applications. Both are simplified by the fact that a cavity is a finite domain.
Hence, there is no need for absorbing boundary conditions or infinite elements, and
the forbidden frequency issue is irrelevant.
7.6 Homework Exercises
Exercise 7.1 The farfield radiation of a vibrating sphere is observed to match the
pattern of a longitudinal quadrupole. The fluid is air and the radius of the sphere is
100 mm. When the frequency is 1.2 kHz, the maximum pressure at 5 m from the
center of the sphere is 0.3 Pa. What vibration of the sphere’s surface would produce
a pressure field having these attributes?
Exercise 7.2 The surface velocity on a sphere is given by vr = Re [V4 cos (4φ)
exp (−iωt) . Derive an expression for the complex pressure amplitude in the radiated
field. Evaluate |P/ρ0 cV4 | along the polar axis, φ = 0. Plot the result as a function
of kr when ka = 4 and ka = 24. Based on these graphs, identify the value of kr at
which the farfield approximation is valid for each ka.
Exercise 7.3 A sphere

vibrates over a band surrounding its equator, specifically,
vr = Re V0 exp (iωt) if π/3 < ψ < 2π/3, vr = 0 otherwise. Derive an expression
for the acoustic pressure in the farfield. From this, evaluate the directivity at ka = 1,
10, and 50. Interpret these results relative to what one would expect qualitatively.
Exercise 7.4 Scanning the farfield of a vibrating sphere has led to identification of
the coefficients of a spherical harmonic series describing the polar angle dependence
of the pressure at a specific radial distance r1 = ka 2 . Specifically, it is known that
only spherical harmonics above m = 8 are insignificant, with the contribution of the
lower harmonics being

8
1
P (r1 , ψ) = P = ρ0 c2 Cm Pm (cos ψ) , Cm =
m=0
(m + 0.5)2
Determine the radial velocity on the surface of the sphere for ka = 2 and ka = 20.
Compare the polar pattern of the surface velocity to that of the measured pressure
field.
Exercise 7.5 Due to measurement error, the coefficients Cm in Exercise 7.4 have a
±20% error in amplitude and a ±40◦ error in phase relative to the nominal value
given there. The errors for each coefficient have a uniform probability distribution.
Create a new set of coefficients by adding the random error to each of the stated
values. Determine the radial velocity for ka = 0.5, 4, and 20 at the surface corre-
sponding contaminated coefficients. Compare that result to the surface velocity when
the coefficients have their nominal values. What conclusions can be drawn regarding
this inverse identification process?
Exercise 7.6 A vibrating sphere of radius a is surrounded by a polymeric material

that has been molded into a sphere of radius b that is concentric with the inner sphere.
Outside these spheres, the fluid is water. The shear strength of the polymeric material
is sufficiently small that the material may be considered to be a liquid whose density
is ρ1 and whose sound speed is c1 . The vibration of the inner sphere is radially
symmetric, so that vr = va sin (ωt) at r = a. Derive an expression for the pressure
in the water. Hint: The pressure and radial particle velocity must be continuous at
r = b.
Exercise 7.7 A hemispherical balloon is fastened on its edges to a rigid baffle.

A jet of air internally induces a vibration that has a bulbous distribution given by
vr = v0 cos (2ψ)sin(ωt) for r = a and 0 ≤ ψ ≤ π/2. (a) Derive a spherical harmonic
series describing the radiated pressure field. (b) Evaluate the farfield directivity for
ka = 0.5, 5, and 20.
Exercise 7.8 A hemispherical balloon is fastened on its edges to an infinite pressure-

release baffle. A jet of air internally induces a vibration that has a bulbous distribution
given by vr = v0 cos (2ψ)sin(ωt) for r = a and 0 ≤ ψ ≤ π/2. (a) Derive a spheri-
cal harmonic series describing the radiated pressure field. (b) Evaluate the farfield
directivity for ka = 0.5, 5, and 20.
7.6 Homework Exercises 127
Exercise 7.9 A hemispherical steel shell is fastened at its rim, ψ = 90◦ , to a rigid
baffle. The attachment is such that displacement parallel to the baffle is possible, but
displacement perpendicular to the baffle is not. In terms of the shell displacements,
these conditions require that u = 0 at ψ = π/2. The internal pressure is a uniform
distribution that fluctuates harmonically, so that ps (ψ, t) = P0 cos (ωt). The shell’s
diameter is a factor of 120 greater than its thickness, its density is 5ρ0 , and the sound
speed of the shell is ce = 3c. (a) Derive spherical harmonic series describing the
radiated pressure field and the shell displacement. (b) Evaluate the farfield directivity
for ka = 0.5, 5, and 20.
Exercise 7.10 A hemispherical steel shell is fastened at its rim, ψ = 90◦ , to a

pressure-release baffle. The attachment is such that displacement perpendicular
to the baffle is possible, but radial displacement parallel to the baffle is not. In
terms of the shell displacements, these conditions require that w = 0 at ψ = π/2.
The internal pressure is a uniform distribution that fluctuates harmonically, so that
ps (ψ, t) = P0 cos (ωt). The shell’s diameter is a factor of 120 greater than its thick-
ness, its density is 5ρ0 , and the sound speed of the shell is ce = 3c. (a) Derive spherical
harmonic series describing the radiated pressure field and the shell displacement. (b)
Evaluate the farfield directivity for ka = 0.5, 5, and 20.
Exercise 7.11 It is suspected that a long cylindrical section of an HVAC duct is

vibrating excessively, thereby radiating an unacceptable sound level. The diameter
of this section is 240 mm. Measurements at a distance of 2.5 m from the centerline
indicate that the sound pressure level is 95 dB//20 µPa, and that the wavenumber
is k̄ = 20ē R + 1.6ēθ + 15ēz m−1 . What properties of the duct’s vibration can be
deduced from these measurements?
Exercise 7.12 An extremely long pipeline is submerged in a deep region of the

ocean. The diameter of the pipe is 700 mm. A disturbance at a distant end generates a
flexural wave in the pipe at wavenumber. A reference location for the radiated pressure
is R = 600 mm, z = 0. The pressure measured at this location is 940 cos(3770t +
0.1160). Then, the position of the hydrophone is shifted gradually. Measurements
at various circumferential angles lead to the conclusion that field is axisymmetric.
The radial distance R is increased gradually. It is found that R2 = 2.48895 m is the
smallest distance greater than R1 = 0.6 m at which the pressure at the two locations
is 180◦ out-of-phase. After this measurement is made, the hydrophone is shifted
axially from the reference location, with R held constant. This new measurement
indicates that the pressure at R3 = 0.6, z 3 = 1.6017 also is 180◦ out-of-phase from
the reference signal. (a) Determine the axial and radial wavenumbers. (b) Determine
the complex amplitude of the surface vibration.
Exercise 7.13 A pressure field is the superposition of a plane wave that propa-
gates in the axial direction of a cylindrical pipe and an axisymmetric cylindri-
cal standing wave. Specifically, the pressure at an arbitrary field point is p =
B H0 (κR) cos (πz/L) cos (ωt) + C cos (ωt − kz). Derive an expression for the time-
averaged power per unit axial length radiated by the cylinder.
Exercise 7.14 The surface normal velocity on a very long cylinder consists of a
bulging motion on one side of the circumference, with a sinusoidal variation in the
axial direction. The specific form is v̄ · ē R = V0 cos (θ) cos (πz/L) if |θ| ≤ π/2, v̄ ·
ē R = 0 if π/2 < |θ| ≤ π. The wavelength is twice the cross section’s radius, L = 2a,
and the frequency is ka = 0.75π. (a) Derive an expression for the radiated pressure as
a function of R, z, and θ. (b) Evaluate the directivity function (R/a)1/2 |Pff | / (ρ0 cV0 )
as a function of θ in the plane z = 0. (c) Evaluate (R/a)1/2 |P| / (ρ0 cV0 ) as a function
of R along the line defined by θ = 0, z = 0. From that result, estimate where the
farfield begins.
Exercise 7.15 A cylindrical pipeline is supported at intervals spaced at distance L.

It transports a liquid whose flow velocity is high, with the consequence that the cross
sections undergo a translational displacement in a fixed transverse direction, which
is defined to be x. In addition, pressure fluctuations within the flowing liquid induce
a uniform expansion and contraction of a cross section. The result is that the normal
displacement on the cylinder’s surface is w = [W0 + W1 cos (πz/L) cos θ] sin (ωt),
where z is the axial distance from a reference point. The cylinder’s diameter is 2a and
the half-wavelength is L = a. Derive an expression for the dependence of the pressure
on the radial distance R, axial distance z, and circumferential angle θ measured from
the x-axis. It may be assumed for this analysis that end effects are negligible. (b) Post-
processing of data acquired in an experiment indicates that the pressure amplitude
when W0 and W1 are nonzero is the twice the amplitude when W0 = 0. The location
for this observation is situated on the transverse line for which θ = 0 and z = 0.
Determine the ratio W0 /W1 for each of the following combinations of frequency and
location: (1) ka = 1 and R = a, (2) ka = 1 and R = 10a, (3) ka = 10 and R = a,
(4) ka = 10 and R = 10a.
Exercise 7.16 The surface of a very long cylinder whose radius is a consists of
a sequence of circular bands of length L. These bands are piezoceramic elements,
each of which vibrates axisymmetrically at frequency ω with no axial variation.
Each band’s velocity is 180◦ out-of-phase from those to which it is adjacent. The
result is that the surface velocity has a square wave pattern in the axial direction,
given by v R = Re (V (z) exp (iωt)) , where the complex amplitude is V (z) = V0 if
0 < z < L, V (z) = −V0 if L < z < 2L, V (z) = V (z ± 2L). Derive an expression
for the farfield pressure. Plot (R/a)1/2 | p/ (ρ0 cV0 )| as a function of z for the case
where ka = 1 and ka = 10. Does either result resemble the square wave pattern on
the surface?
Exercise 7.17 The field radiated by a sphere that executes an oscillatory rigid
body translation has a known solution as a dipole, see Sect. 6.5.2. Use the results
derived there to evaluate the surface pressure when a sphere’s surface velocity is
Re (V1 exp (iωt)) cos ψ. Then, use this surface response to formulate the KHIT for
the pressure at an arbitrary field point. In the case where ka = 5, evaluate the result
of that formulation along the polar axis, ψ = 0, for 0 < r < 4a. Compare the result
to the analytical description, Eq. (6.5.102).
Exercise 7.18 A cylindrical rod whose length is L and diameter is 2a vibrates like
a simply supported beam. This means that all points on a cross section displace
by the same amount in a fixed transverse direction. In terms of cylindrical coordi-
nates whose origin is at the midpoint, the resulting velocity of a surface point is
v̄ = Re (V1 exp (iωt)) sin (πz/L) (ēr cos θ + ēθ sin θ). It is desired to ascertain the
farfield directivity, so it is not acceptable to approximate the rod as an infinite length
cylinder. Use the formulation in Example 7.10, in which the radiation impedance for
an infinite cylinder is used to estimate the surface pressure corresponding to the stated
surface velocity. An integral expression for the far field pressure can be obtained by
substituting the surface velocity and approximate surface pressure into the KHIT.
(a) Derive this integral expression. (b) Use analytical or numerical methods to eval-
uate and plot |R| Pff / (ρ0 cV1 ) in the plane θ = 0 as a function of the polar angle ψ.
Parameters are ka = 1 and 10, L/a = 6.
Exercise 7.19 The rigid disk in the sketch executes an oscillatory transverse vibra-
tion as a rigid body. The normal velocity on the front surface, whose normal is
ēz , is v̄front · n̄ = Re (V1 exp (iωt)), whereas on the back surface, it is v̄back · n̄ =
Re (−V1 exp (iωt)). It is very thin, so the surface response on the edge has negligi-
ble effect. Because the front and back motions are 180◦ out-of-phase, but otherwise
alike, the surface pressures on opposite sides also must be 180◦ out-of-phase. Fur-
thermore, the pressure distributions must be axisymmetric relative to the z-axis, so it
must be that pfront = − pback = Re (P (R) exp (iωt)). Another consideration comes
from the theory of diffraction. According to it, the zero thickness approximation
leads to a discontinuity at R = a. This requires that the pressure at the edge be zero
with an infinite slope, dpfront /d R = 0. A trial function fitting these requirements is
1/2
Pfront = PC 1 − R 2 /a 2 . (a) Show that using this guess to formulate the farfield
version of the KHIT, Eq. (7.4.18), leads to
1/2
e−ikr a
R2
Pff (x̄0 ) = ik (cos ψ) PC 1 − 2 J0 (k R sin ψ)Rd R
r 0 a
(b) In the limit as ka → 0, this expression should yield the same result as Eqs. (7.4.38)
and (7.4.39), which is the (low-frequency) multipole expansion for a transversely
oscillating disk. Use this fact to derive an expression for PC .
v1
front R
y
x
Exercise 7.19
Exercise 7.20 The sketch shows a rigid box mounted on a vertical shaft. Its rotation
rate is = 0 sin (ωt). Identify the restrictions that must be imposed on 0 and ω in
order that a linearized analysis be valid. Given that these restrictions are met, describe
the multipole expansion that would fit the field radiated by the box. In what directions
can the pressure amplitude be expected to have a maximum value? In what directions
can the pressure be expected to be zero? Estimate the largest pressure amplitude that
would be observed at a fixed distance r from the origin.
y
a
a
x
Exercise 7.20
Exercise 7.21 The surface Helmholtz integral equation was derived in Sect. 7.5.2
for the case of radiation by letting x̄0 be an exterior field point that approaches the
surface S. Show that the same result is obtained if x̄0 is a field point inside the
radiating body.
Exercise 7.22 The field radiated by a sphere that executes an oscillatory rigid body
translation has a known solution as a dipole, see Sect. 6.5.2. Thus, it offers the oppor-
tunity to test an implementation of the source superposition formulation. The arrange-
ment of sources and field points is described in the sketch. There are four sources
that are situated along the z-axis, which is the direction of the translational velocity
v̄ = V1 sin (ωt) ēz . These points are spaced at 0.4a centered on the sphere’s center.
There are nine surface points spaced equally along a meridian. (a) Explain why it is
preferable that all of sources be situated on the z-axis. (b) Explain why it would be
incorrect to use source or surface points at the same polar angle along more than one
meridian. (c) Without solving the superposition equations, anticipate how the source
strengths left of the center should be related to those to the right of the center. (d)
Formulate and solve the source superposition equations for the case where ka = 8.
Compare the pressure on the surface at ψ = 30◦ to the dipole solution.
vz vz
a
vz vz
0.2a 0.4a vz
vz
Source Surface field point

Exercise 7.22
Exercise 7.23 The sketch shows a model for an underwater projector. It is a cylin-
der whose surfaces and one end are stationary. The end that is visible in the sketch
executes a translational oscillation, such that all points on that surface vibrate in the
axial direction according to vz = V cos (ωt). The task is to use the source superpo-
sition method to evaluate the radiation. The eventual objective would be to evaluate
the radiated pressure field, but here it is sufficient to evaluate the source strengths.
The arrangement of sources and surface points is such that both are spaced at one-
fifth of the cylinder’s length centered on the midpoint in the axial direction, with an
additional six field points, three on each end. (a) Explain why it is preferable that
all sources be situated on the z-axis. (b) Explain why it would be incorrect to use
source points at the axial location along other circumferential angles. (c) Explain
why it would be unwise to locate a surface point at the junction of the cylinder and
its end. (d) For the case where L = 3a and ka = 4, determine the source strengths
as nondimensional quantities, Q̂ n / a 2 V .
0.2 L
0.2 L
0.2 L
0.2 L
vz vz
0.4 a
vz
0.4 a
vz
vz
vz L
a
Source
Surface field point
Exercise 7.23
Chapter 8
Radiation from a Source in a Baffle
Section 3.5.1 introduced a correction for the pressure-release condition at the open
end of a one-dimensional waveguide. The basis for that development is the radiation
properties of a circular piston flush mounted in a wall whose extent is large. The wall
is referred to as a baffle. A piston serves well as a model of real transducers that might
be piezoceramics in an underwater projector. It may be employed as an approximate
representation of the movable cone of a loudspeaker. The term “piston” is intended
to convey the notion that all points in the surface have the same velocity. Our study
will concentrate on such situations, but allowance will be made in the development
for more general velocity distributions.
Because the concern here is with the signal that is generated by a type of sur-
face vibration, it could logically be included in the previous chapter. However, the
phenomena that arise are extremely relevant to a number of every day systems. Con-
sequently, many aspects of radiation from a transducer in a planar baffle have been
thoroughly analyzed. The subject appears in its own chapter as a way of emphasizing
its importance.
The KHIT is the fundamental principle that is the foundation for the developments
that follow. Rayleigh identified a Green’s function that removes the need to know
the surface pressure, with the result that the pressure at a field point is described as
a surface integral that requires specification of the normal velocity. Given that the
KHIT we derived governs the frequency domain, that is the description with which
we begin, but we will also examine a time-domain solution.
8.1 The Rayleigh Integral
Figure
8.1 depicts the system to be analyzed. We wish to determine the pressure
Re P (x̄0 ) eiωt at an arbitrary field point x̄0 that results from a specified normal

velocity distribution Re n̄ (x̄s ) · V̄ (x̄s ) eiωt on S. This plane is the boundary of a
DOI 10.1007/978-3-319-56847-8_8
134 8 Radiation from a Source in a Baffle
Fig. 8.1 An aribitrary z

velocity distribution over _ _ _
points x̄s on a flat plane that n(xs)=ex
is the boundary S of a
half-space
y
x _ _ _ _
Re(n(xs) . V (xs) e it )
half space. The origin of the xyz coordinate system is situated on S, with the z-axis
pointing into the fluid domain. The only requirement imposed on V̄ (x̄s ) is that it
approach zero with increasing distance in all directions from the origin.
The KHIT entails integration over the surface that bounds the domain V in which
the signal occurs. In the present situation that domain is a half space. The plane is
only a part of the boundary, with the other part extending to infinity. The concept
of an enclosing virtual surface allows us to apply the KHIT to this configuration.
In Sect. 7.4.2, a sphere that surrounds the vibrating body was used to extend the
KHIT to radiation problems. Here, we use a hemisphere S0 centered on the origin,
whose radius r0 is allowed to grow without limit. The reasoning that previously led
to recognition that the contribution of S0 to the KHIT is negligible if r0 is very large
also applies here. The consequence is that the KHIT only requires integration over
the planar surface S.
The discussion in Sect. 6.4.2 showed that in addition to the free-space Green’s
function, other functions that satisfy boundary conditions can be defined. If we can
find a Green’s function for which n̄ (x̄s ) · ∇G (x̄0 , x̄s ) is identically zero, then the
integrand in the KHIT will not contain the surface pressure. Such an arrangement
would make it possible to determine the pressure field from a direct evaluation of
the integral.
The first step in identifying the requisite Green’s function is to remove any ambigu-
ity in what we seek. The position of the field point x̄0 is represented by its coordinates
x0 , y0 , and z0 and a source point x̄, not necessarily on the surface, is situated at coor-
dinates x, y, and z. If we reduce z to zero, the source point will be the surface point
x̄s . In other words, x̄ = x̄s + zēz . The gradient in the KHIT is evaluated at the source
point, and we want the source point to be on the surface. Thus, we seek a Green’s
function that satisfies
∂
n̄ (x̄s ) · ∇s G (x̄0 , x̄s ) ≡ lim G (x̄0 , x̄) = 0 (8.1.1)
z→0 ∂z
8.1 The Rayleigh Integral 135
Our experience with a point source in a half space suggests where to search for
the desired function. According to the method of images, the field associated with a
point source situated above a rigid plane is obtained by adding an equal point source
at the same distance below the plane. Figure 8.2 depicts a point source at distance z
above the surface and its image at distance z below the surface. The projection of x̄
onto the surface is also x̄s .
Fig. 8.2 Green’s function x

for a half space bounded by a z r
rigid planar boundary. The x0
source is at x̄ and the image z
is at x̄
xs
x
z r'
x'
The mass acceleration of the point source is defined to be one. The image location
is x̄ = x̄s − zēz . The sum of this source and its image is a Green’s function for the half
space. This is so because the field radiated by the pair satisfies the inhomogeneous
Helmholtz equation, ∇ 2 G + k 2 G = −δ (x̄0 − x̄) for any x̄ and x̄0 within the half
space, as well as the radiation condition, and the boundary condition of zero normal
velocity on the surface. To distinguish the half-space function G from the free-space
Green’s function let us use a subscript “fs” to identify the latter. Thus, the function
we have constructed is
G (x̄, x̄0 ) = G fs (x̄s + zēz , x̄0 ) + G fs (x̄s − zēz , x̄0 ) (8.1.2)
We wish to reduce z to zero. Before we do so, let us evaluate ∂G/∂z when z is

not zero. The radial distances to the field point from the source and image are
1/2
r = |x̄s + zēz − x̄0 | = (x − x0 )2 + (y − y0 )2 + (z − z0 )2
1/2 (8.1.3)
r = |x̄s − zēz − x̄0 | = (x − x0 )2 + (y − y0 )2 + (−z − z0 )2
Because the free-space Green’s function depends only on the radial distance between
the pair of points, the z derivative may be evaluated according to
∂ ∂r d ∂r d
G (x̄0 , x̄) = G fs (x̄s + zēz , x̄0 ) + G fs (x̄s − zēz , x̄0 )
∂z ∂z dr ∂z dr (8.1.4)
z − z0 d z + z0 d
= G fs (x̄s + zēz , x̄0 ) + G fs (x̄s − zēz , x̄0 )
r dr r dr
When we reduce z to zero, both x̄ and x̄ become x̄s . Consequently,

G fs (x̄s + zēz , x̄0 ) and G fs (x̄s − zēz , x̄0 ) both become G fs (x̄s , x̄0 ) , so that dG fs /dr
approaches dG fs /dr. Furthermore, in the limit, the fractions preceding each G fs term
in the above expression become −z0 /r and z0 /r. Thus, we conclude that the Green’s
function defined in Eq. (8.1.2) with z = 0 satisfies Eq. (8.1.1). In other words, the
Green’s function we seek is
G (x̄0 , x̄s ) = 2G fs (x̄0 , x̄s ) (8.1.5)
The result of using this Green’s function to form KHIT is

P (x̄0 ) = 2iωρ0 n̄ (x̄s ) · V (x̄s ) G fs (x̄0 , x̄s ) dS (8.1.6)
S
In retrospect, Eq. (8.1.5) could have been identified heuristically. A free-space

Green’s function represents the limit as its radius shrinks to zero of a radially sym-
metric spherical source, whose mass acceleration is unity. When this sphere is placed
on a rigid surface at x̄s , its mass flow is confined to the hemisphere above the bound-
ary. If the mass flow is held at a unit value, the particle velocity on the hemisphere
is double that for the full sphere in free space. This results in doubling the radiated
pressure relative to what would have been obtained if the surface were not present.
In regard to the gradient property, consider a spherical source in free space at loca-
tion x̄s . Changing the z coordinate of the source will change the distance to a field
point, and therefore the pressure at that point. In contrast, if a hemispherical source
is situated on the boundary, moving it to a small distance above the boundary moves
its image below the boundary by the same distance. The change in the distance from
the image to the field point is the negative of the change in the distance from the
source to the field point. The pressure changes for each cancel.
Equation (8.1.6) is valid for two-dimensional, as well as three-dimensional
models. This is so because the only place where dimensionality was considered
was in the representation of r1 and r2 , and setting ys and y0 to zero for a two-
dimensional geometry would not alter any result. However, our primary interest
is in three-dimensional configurations. When G fs (x̄s , x̄0 ) is specified as the three-
dimensional version, the form of Eq. (8.1.6) that results is known as the Rayleigh
integral.1 It is

iωρ0 e−ik|x̄0 −x̄s |
P (x̄0 ) = Vs (x̄s ) dS (8.1.7)
2π |x̄0 − x̄s |
S
where Vs (x̄s ) is the velocity normal to a surface at location x̄s ,
Vs (x̄s ) = n̄ (x̄s ) · V̄ (x̄s ) (8.1.8)
1 J.W. Strutt Lord Rayleigh, Theory of Sound, vol. 2, 2nd ed., Dover (1945 reprint) Sect. 78.
8.1 The Rayleigh Integral 137
The Rayleigh integral describes situations where S is the planar boundary of an

infinite half space. We will see in Sect. 8.4 that it may be used for a finite baffle,
provided that the baffle extends at least one acoustic wavelength beyond the portion
of S that moves. However, it has been used as an approximation when S is not
planar. Doing so requires a thorough verification through computational modeling
or experiment because it is not justified by the derivation. This issue will be examined
in Example 8.4.
A detailed mapping of the pressure field requires that the Rayleigh integral be
evaluated numerically. How to perform such an evaluation is addressed in Sect. 8.4.
However, our initial explorations will examine analytical solutions that are descrip-
tive of the field in specific regions.
8.2 Farfield Directivity
Suppose the surface vibration is confined to a finite portion of S, which will be

designated as Sv . If the distance from the origin to the field point x̄0 is much greater
than the maximum distance between two points in Sv , then we may invoke the farfield
approximation of the radial distance from a surface point to x̄0 .
8.2.1 Cartesian Coordinate Description
We consider first the case where the outline of the vibrating region is best described
in terms of the Cartesian coordinates x and y that lie on the surface. A point in Sv
is x̄s , and x̄0 is a field point. The farfield approximation requires that |x̄0 | be much
greater than the largest |x̄s | . When this condition applies, the radial line from x̄0 to
any x̄s is essentially parallel to the radial line from x̄0 to the origin of xyz, so that the
length of radial lines differs by the projection of x̄s onto x̄0 . The magnitude of x̄0 is
the radial distance r, so the approximation is
x̄0
|x̄s − x̄0 | ≈ |x̄0 | − x̄s · = r − x̄s · ēr (8.2.9)
|x̄0 |
The location of the field point is described in Fig. 8.3 in terms of spherical coor-
dinates. The radial unit vector is
ēr = sin ψ cos θēx + sin ψ sin θēy + cos ψēz (8.2.10)
The location of the surface point is x̄s = xs ēx + ys ēy , so we have
|x̄s − x̄0 | ≈ r − xs sin ψ cos θ − ys sin ψ sin θ (8.2.11)

Fig. 8.3 Locations of a z

surface point x̄s and a field
point x̄0 for implementation
of the farfield approximation
of the Rayleigh integral
_
x0

v

xs y
ys
x
Re(n(xs) . Ve it )
The condition that x̄0 be a farfield point requires that r xs and ys for any point
in the surface. In that case |x̄s − x̄0 | in the denominator is well approximated as 1/r.
The normal velocity is Vs (xs , ys ) , and an area element is dxs dys . The result is that
Eq. (8.1.7) becomes

iωρ0 −ikr
Pff (x̄0 ) = e Vs (xs , ys ) eik(xs sin ψ cos θ+ys sin ψ sin θ) dxs dys (8.2.12)
2πr
Sv
Suppose that Sv is a rectangle defined by −a ≤ xs ≤ a, −b ≤ ys ≤ b, and that the

velocity distribution has the special form of a product of functions of xs and ys ,
that is, Vs (x̄s ) = X (xs ) Y (ys ) . Because dS = dxs dys and the integration limits are
constants, the integrals over xs and ys may be evaluated independently. In this case,
the farfield pressure is given by
a b
iωρ0 −ikr ikxs sin ψ cos θ
Pff (x̄0 ) = e X (xs ) e dxs Y (ys ) eikys sin ψ sin θ dys (8.2.13)
2πr −a −b
Each integral fits the definition of a Fourier transform, which is defined in Eq. (B.1.5).
The factors appearing in the exponent in each integral are the wavenumbers
μx =
k sin ψ cos θ and μy = k sin ψ sin θ. The angular spectrum F̃ μx , μy is defined to
be the product of the individual transforms, according to

a b
F̃ μx , μy = X (xs ) e iμx xs
dxs Y (ys ) eiμy ys dys (8.2.14)
−a −b
In terms of this function, the farfield pressure is
e−ikr
Pff (x̄0 ) = iωρ0 F̃ (k sin ψ cos θ, k sin ψ sin θ) (8.2.15)
2πr
8.2 Farfield Directivity 139
If Sv is not rectangular, or if Vs is not a product of xs and ys functions, the area

integral does not factorize. Nevertheless, integration over the full range of xs and
ys leaves only the wavenumbers as variables. The angular spectrum in this case
is a surface integral that constitutes a two-dimensional Fourier transform. Thus,
Eq. (8.2.15) is descriptive of the farfield pressure in any situation, with the angular
spectrum defined to be
∞ ∞

F̃ μx , μy = Vs (xs , ys ) ei(μx xs +μy ys ) dxs dys (8.2.16)
−∞ −∞
In the case where the angular spectrum factorizes as in Eq. (8.2.14), tabulations
of functions and their Fourier transforms are available, see Tables B.1 and B.2 in
Appendix B. FFT technology may be employed to handle truly complicated velocity
functions. However, the normal velocity typically is characterized by an elementary
function, which makes it feasible to evaluate the integral analytically. Situations
in which the vibrating portion of Sv is circular may be analyzed by using polar
coordinates. Such a formulation is the topic of the next section.
Identification that the farfield pressure is directly related to the two-dimensional
Fourier transform of the surface velocity leads to an interesting interpretation. Sup-
pose we wish to determine the pressure at a specific field point in the farfield. The
radial unit vector to that point is ēr , which is defined in Eq. (8.2.10). The wavenumber
of the farfield signal propagating in this direction is k ēr . The trace of this vector onto
the xy plane is its projection, k̄s = k sin ψ cos θēx + k sin ψ sin θēy in Fig. 8.4. These
components are the wavenumbers μx and μy for the angular spectrum in Eq. (8.2.15).
This construction makes it evident that k̄s ≤ k, which leads to an interpretation of

the farfield radiation. We construct a surface F̃ μx , μy in a Cartesian wavenumber
z
ker

y
F(x,y)
ks

F(k sin  cos , ksin  sin )
x
Fig. 8.4 Trace wavenumber vector k̄s for the farfield spherical wave in a designated direction ēr
space such that the height of the surface is the value of F̃, as is done in Fig. 8.4.
In the μx μy plane, we construct a circle of radius k. Any point within this circle
corresponds to a signal that propagates to the farfield at polar angle ψ and azimuthal
angle θ. These angles are related to the wavenumbers μx and μy of the selected point
by cos θ = μx /(μ2x + μ2y )1/2 , sin θ = μy /(μ2x + μ2y )1/2 , and cos ψ = (μ2x + μ2y )1/2 /k.

Selection of a point in the μx μy plane for which k̄s > k corresponds to subsonic
surface waves that do not radiate to the farfield.
This interpretation of the relationship between the surface wave spectrum and
the signal that propagates in a specified direction can be used as a design tool.
Suppose we wish that the radiated pressure in a certain direction designated by
spherical angle ψ0 and θ0 be the maximum. For a specified frequency, we can form
the wavenumber vector in this direction. The components of this vector in the x
and y directions are the corresponding wavenumbers, μx = k sin ψ0 cos θ0 , μy =
k sin ψ0 cos θ0 . The maximum condition is attained if the two-dimensional Fourier
transform of n̄ (x̄s ) · V (x̄s ) has a maximum magnitude at this pair of wavenumbers.
Finding a surface vibration pattern whose Fourier transform has this attribute would
not be a trivial task, but tabulations of Fourier transform pairs like those in Appendix
B would be useful.
EXAMPLE 8.1 An electrostatic speaker essentially is a pair of oppositely

charged plates, one of which is fixed. The free plate vibrates in response to
oscillations of the voltage difference. This vibrating plate may be taken to be
rigid, so its normal velocity is v0 cos (ωt) everywhere on its surface Sv . The
plate is surrounded on all sides by a large rigid baffle. Two alternative designs
are contemplated. Both have equal surface area, but one has a square face
whose sides are 200 mm, and the other is 100 mm by 400 mm. The frequency
√ as a function of the polar angles ψ
is 2 kHz. Determine the farfield pressure
and θ for each design, and graph r/ Sv |Pff | / (ρ0 cv0 ) as a function of the
spherical angles.
Significance
The Fourier transform of the vibration pattern will be found with little effort, which
will allow us to focus on the correlation between the properties of the vibration and
those of the pressure field.
Solution
The shape of Sv is rectangular and the velocity function is a constant, so we shall
evaluate the angular spectrum directly. Let 2a and 2b be the lengths of the edges
parallel to the x and y axes, respectively. Then the complex vibration amplitude is

v0 if |xs | ≤ a and |ys | ≤ b
Vs (xs , ys ) =
0 otherwise
The two-dimensional Fourier transform of this function is

a
b
F̃ μx , μy = v0 ei(μx x+μy y) dys dxs
−a −b

sin (μx a) sin μy b
= 4v0
μx μy
A compact way to write this expression is to use the sinc function, which is defined
as sinc(ξ) ≡ sin (πξ) / (πξ) . Figure 1 is a plot of the sinc function.
1
0.5 sinc()
0
−0.5
−8 −6 −4 −2 0 2 4 6 8


At μx = 0 or μy = 0, the value of F̃ μx , μy is the finite value obtained as a limit,
which gives sinc(0) = 1. The transform of the surface velocity is thereby found to
be μ a
x μy b
F̃ μx , μy = 4v0 ab sinc sinc
π π
To evaluate the farfield radiation pattern, we define a hemispherical mesh formed

from meridional semicircles through the z-axis along which θ is a constant value
between −π and π, and circles centered on the z-axis along which ψ is a constant
between zero and π/2. The radial direction ēr at each mesh point is described
by
Eq. (8.2.10). The trace of k ēr on the xy plane is k sin ψ cos θēx + sin θēy , so the sur-
face wavenumbers are μx = k sin ψ cos θ and μy = k sin ψ sin θ. With an eye toward
computations,
we collect variables as nondimensional groups by multiplying and
dividing F̃ μx , μy by ab. Correspondingly, Eq. (8.2.15) gives

2kab e−ikr ka sin ψ cos θ kb sin ψ sin θ

Pff (x̄0 ) = iρ0 cv0 sinc sinc
π r π π
The value of ab for the alternative designs is the same, and the maximum value of
the sinc function is one, so the alternative
design gives similar pressure amplitudes.
When b > a, the oscillation of F̃ μx , μy in the μy direction will be more rapid

than that in the μx direction. Maxima of F̃ μx , μy correspond to the directions in

which the farfield pressure is a maximum, and zeros of F̃ μx , μy correspond to
zero farfield pressure in the corresponding ēr direction. It follows that the slender
rectangle configuration will exhibit more side lobes along θ = 90◦ and 270◦ than it
does along θ = 0 and 180◦ . To see the details, we need to evaluate the pattern. The
area on which the vibration occurs is 4ab, so the specified quantity to plot is given
by

|Pff (x̄0 )|
r
=
k (ab)1/2 sinc ka sin ψ cos θ sinc kb sin ψ sin θ
2 (ab)1/2 ρ0 cv0 π π π
Figure 2 describes the square configuration at 2 kHz. The data is displayed as

a spherical plot in which the radial distance is the quantity defined above. This is
accompanied by polar plots at θ = 0, which is the xz plane, and θ = 90◦ , which is
the yz plane. (The range of ψ was set as −π/2 < ψ < π/2 to depict the data on both
sides of the z-axis.) Because a = b, the polar plots are identical. The spherical plot is
close to being axisymmetric relative to the z-axis, and the pressure falls off rapidly as
ψ increases from zero. This type of pattern, in which the signal is essentially aimed
in a certain direction, is said to be a sound beam. There are several side lobes, but
the pressure amplitude in them is quite low relative to the main lobe.
x
=0
2
1  = 90o
z 2
1
z
 y
z

Figure 3 describes the case where the plate is a narrow rectangle at 2 kHz, as in
the previous case. This field also is a sound beam. The polar plots indicate that the
signal in the yz plane is much more tightly confined to the z-axis than it is in the xz
plane. This might seem to be counterintuitive because the width of the plate in the
x direction is large. To explain this feature, consider the limit in which the plate’s
dimensions a and b become infinite. In that case, the radiated field would consist of
a simple plane wave that radiates in the z direction. Such a wave would be zero in
all directions other than ψ = 0.
x
θ=0
2
 = 90o
1
2
z
1
z
y

z

We have encountered a number of situations in which decreasing the size of the

vibrating object (or the frequency) produces a field that varies more gradually in
space. Figure 4 exhibits this feature. The values of a and b are like those of Fig. 2, but
the frequency now is 200 Hz. Even though the rectangular plate is long and narrow
(we say that its aspect ratio b/a is high), the field is almost axisymmetric, and there
is little variation from ψ = 0 to ψ = 90◦ . The nondimensional parameters setting
the behavior are ka and kb, which are 0.37 and 1.48, respectively, in this last case.
Thus, both sides are much smaller than the acoustic wavelength, 2π/k. As a general
rule, a fairly uniform farfield will be obtained if the size of the radiating body is
substantially less than the acoustic wavelength.
x
θ=0
 = 90o
0.2
z 0.2
x
0.1 z
y z

8.2.2 Farfield of a Piston Transducer
In the previous section, the surface vibration was considered to be best described in
terms of a Cartesian coordinate system. Here, we consider the situation in which the
outline of the vibrating region is circular. A set of polar coordinates (Rs , θs ) centered
on this region are most suitable to formulate the Rayleigh integral. The development
will begin by allowing the velocity distribution to depend arbitrarily on Rs and θs .
However, in most cases of interest the vibration pattern is axisymmetric, which
means that the vibration amplitude depends solely on Rs . A special case is a piston
transducer, for which the vibration amplitude is uniform across the circular area.
This model often is used to represent a common loudspeaker. Some very interesting
phenomena are encountered in this case, so it will receive the most attention.
Figure 8.5 depicts a vibrating circular region Sv surrounded by a rigid baffle. The
z-axis for a cylindrical coordinate system is normal to the surface. The field point x̄0
is located by its spherical coordinates (r, ψ, θ), and a generic point on Sv has polar
coordinates (Rs , θs ) , so its position is x̄s = Rs cos θs ēx + Rs sin θs ēy .
The farfield approximation of the distance between a surface point and a field
point is |x̄ − x̄s | = r − x̄s · ēr . In view of Eqs. (8.2.9) and (8.2.10), this reduces to
|x̄ − x̄s | = r − Rs sin ψ cos (θs − θ) (8.2.17)
In combination with a polar coordinate description of dS, this expression converts
the farfield approximation of the Rayleigh integral, Eq. (8.2.12), to
a π
iωρ0 −ikr
Pff (x̄0 ) = e Vs (Rs , θs ) eikRs sin ψ cos(θs −θ) Rs dθs dRs (8.2.18)
2πr 0 −π
Fig. 8.5 Coordinate z

descriptions of a surface
point and a field point
suitable for the analysis of a Rs 
circular transducer in an x0
infinite baffle r
v
xs
s
y

x
where a is the radius of the area over which the vibration occurs.
Usually the velocity distribution is axisymmetric, so Vz is independent of θs . This
greatly simplifies the analysis because it allows the integration over θs to be factored
out. An integration over θs with Rs held fixed represents the contribution to the farfield
pressure of a circular ring whose radius is Rs and whose radial width is dRs . The
normal velocity along this ring is Vs (Rs ) , so that
π
iωρ0 −ikr
dPring = e Vs (Rs ) Rs dRs eikRs sin ψ cos(θs −θ) dθs (8.2.19)
2πr −π
The value of θ is fixed by selection of the field point. This allows us to change the
integration variable to θ = θs − θ. Furthermore, the integrand is periodic in any 2π
interval, and it is an even function of θ . This allows us to integrate over 0 ≤ θ ≤ π,
then double the result. Hence, the pressure radiated by a ring whose radius is Rs and
width is dRs is given
π
iωρ0 −ikr
dPring = e Vs (Rs ) Rs dRs eikRs sin ψ cos θs dθs (8.2.20)
πr 0
The integral may be found in most references on Bessel functions,2 which leads to
e−ikr
dPring = iρ0 cVs (Rs )kRs dRs J0 (kRs sin ψ) (8.2.21)
r
There is no dependence on the azimuthal angle θ for the field point because any
point-symmetric vibration pattern will generate an axisymmetric pressure field.
Equation (8.2.21) may be used for an actual ring by replacing the differential
width dRs with the actual width Rs and taking Rs to be the mean radius. However,
doing so requires that Rs /Rs be sufficiently small to that any fluctuation of the
Vz value from the inner to outer radius is negligible. The more important use is to
2 M.I. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover (1965) p. 360,
Eq. (9.1.21).
construct the signal for a full circular region. The Rayleigh integral in that case is
essentially a sum of contributions of all differential elements, so that

e−ikr ˜ a
Pff (x̄0 ) = iρ0 ck f (k sin ψ) , f˜ (μ) = Vs (Rs )J0 (μRs ) Rs dRs
r 0
(8.2.22)
This expression has the appearance of Eq. (8.2.15). The function f˜ (μ) is the angular
spectrum for a decomposition of the surface velocity into waves that propagate along
the surface radially outward from the center. From a mathematical perspective, it
is the Hankel transform of the axisymmetric Vs (Rs ). The Hankel transform may be
derived by transforming the two-dimensional Fourier transform to polar coordinates.3
Depending on the nature of Vs (Rs ), it might be possible to evaluate f˜ (μ) analytically.
If not, it can be evaluated numerically.
A particularly important case is that of a piston or translating rigid disk, for which
Vs = V0 for Rs ≤ a. Introduction of u = kRs sin ψ into the integrand of the angular
spectrum leads to a standard integral,4 with the eventual result that
J1 (ka sin ψ) a −ikr

Pff (x̄0 ) = iρ0 cV0 e (8.2.23)
sin ψ r
The Rayleigh distance R0 is a reference length often used to describe acoustic

radiation. We will see in the next section that it indicates the radial distance at which
the signal is well represented by its farfield approximation. The definition of R0 is
that it is the ratio of the area over which the vibration occurs to the wavelength. For
a circular piston, it is
Sv πa2 1
R0 = = = ka2 (8.2.24)
λ 2π/k 2
A simple expression for the maximum pressure at a specified radial distance, which
occurs at ψ = 0, results from using the Rayleigh distance in Eq. (8.2.23). As ψ → 0,
the first order Bessel function approaches (1/2) ka sin ψ, so that
R0
max (|Pff (x̄0 )|) = ρ0 cV0 @ψ=0 (8.2.25)
r
The corresponding farfield pressure is

2J1 (ka sin ψ) R0

Pff (x̄0 ) = iρ0 cV0 e−ikr (8.2.26)
ka sin ψ r
3 I.A. Sneddon, Fourier Transforms, Dover (1995 reprint) Chap. 2.

4 Abramowitz and Stegun, ibid., #11.3.20.
The directivity is the ratio of the pressure magnitude as a function of ψ to the value
on-axis, with r fixed in the farfield. Thus,

|Pff | 2 |J1 (ka sin ψ)|
D (ψ) = = (8.2.27)
max (|Pff |) r>R0 ka sin ψ
An interesting comparison is with the pressure radiated by a ring having the

same radius a and vibrational velocity V0 . We find from Eq. (8.2.21) that the ratio
(Pff )ring / (Pff )piston = 2Rs /a on the z-axis, where ψ = 0. The expression for a ring
is based on the width Rs being much less than the radius, from which it follows
that the ring radiates much more weakly as a consequence of the smaller face area.
The directivity factor for a ring is J0 (ka sin ψ) . The nulls of the directivity in each
case correspond to parameters for which ka sin ψ is a zero of the respective Bessel
functions.
The number of nulls that occurs is limited by the requirement that sin ψ < 1, from
which it follows that this number increases in a stepwise manner approximately in
proportion to the value of ka. Thus, below ka = 3.832 for a piston and ka = 2.405
for a ring, the directivity exhibits no nulls. Above these frequencies, the interval from
the axis to the first null is the main lobe, and the other intervals are the side lobes.
The maxima in the side lobes decrease rapidly with increasing angle from the z-axis.
Figure 8.6 displays some typical patterns. The field in each case is axisymmetric, so
a polar plot over −π/2 ≤ ψ ≤ π/2 is sufficient.
The rapid decrease in the magnitude of the side lobes obscures them in a polar
plot. A rectangular plot, such as that in Fig. 8.7, conveys a clearer quantitative picture.
Piston Ring
ka=2
ka=6
ka=18
Fig. 8.6 Directivity of a piston and a ring having radius a embedded in an infinite baffle
1 1
Piston Ring
0.5 0.5
0 0
−90 −45 0 45 90 −90 −45 0 45 90
Polar angle  Polar angle 
Fig. 8.7 Cartesian plot of the directivity of a piston and a ring embedded in an infinite baffle,
ka = 18
We can discern some general trends from Fig. 8.7 and Table 8.1. For a given value
of ka, the main lobe is narrower with a ring than it is with a piston. The ring has
more side lobes, and the side lobes are stronger. These trends, in combination with
the stronger signal radiated by the piston, indicate that if one’s objective is to aim
a signal in a certain direction at a specified frequency, they should use the piston
configuration. The piston’s radius should be as large as feasible, because a large ka
will yield a small value of ψ for the first null.
Table 8.1 Polar angles at which the directivity of a piston and a ring have a null
ka sin ψ
Null # 1 2 3 4
Piston 3.832 7.016 10.173 13.324
Ring 2.405 5.520 8.654 11.792
A quantity of interest is the beamwidth bw , which is the cone angle at which the
intensity is half the maximum on-axis value. In the farfield, the radial particle velocity
is p/ (ρ0 c) and the other velocity components are negligible, so the time-averaged
intensity is
2
P ρ0 c |V0 |2 R0
(Ir )av = = D (ψ)2 (8.2.28)
2ρ0 c 2 r
√
The half-power point corresponds to D (ψ) = 1/ 2. For the piston directivity in
Eq. (8.2.27), this condition occurs at ka sin ψ = 1.62, whereas Table 8.1 indicates that
the first null occurs at ka sin ψ = 3.82. Hence, if large ka is large, the beamwidth
approximately subtends the center half of the main lobe.
A question that sometimes is posed is why is the piston model used to describe
the radiation properties of transducers that do not look like a piston? One such device
is a speaker cone embedded in a planar baffle, which is a common driver for sound
reproduction systems. This arrangement is shown in Fig. 8.8.
The field is symmetric about the z-axis. The depth of the transducer below
the xy plane of the baffle is some function Z (R) . A well-designed speaker cone
translates in the z direction, so that all points on the cone have the same velocity
Fig. 8.8 Cross-sectional z

view of a conical speaker
embedded in a rigid baffle
a R
Baffle Z
Speaker x
cone

Re Vz eiωt . Because the angle of the cone is close to 90◦ , the normal to the cone’s
surface is nearly parallel to the z-axis, so we may say that the (time domain) surface
velocity is

Re Vz eiωt @ z = −Z (R) if |R| < a
n (x̄s ) · v̄ (x̄s ) = (8.2.29)
0 if |R| > a
An approximation of the field close to the surface is that each point emits a
plane wave that propagates in the axial direction. (This is a crude representation of a
complicated process, but it is adequate for the present purpose.) In this approximation,
a signal that is emitted from a point on the transducer’s face at radial distance R
requires an interval of Z (R) /c to arrive at the xy plane. This is a lag, so the phase
delay for a harmonic signal at frequency ω is kZ (R) . The complex surface velocity
on the xy plane in this approximation is

Vz e−ikZ(R) if |R| < a
Vs = (8.2.30)
0 if |R| > a
In most cases, the intent is to radiate sound over a broad range of polar angles,
in which case ka will be small. The preceding description of Vs states that there is
a position-dependent phase lag, which may be written as ka (Z (R) /a) . If ka is not
much larger than one and the cone angle is close to 90◦ , so that Z (R) /a is small
everywhere, then it is reasonable to ignore this phase lag. A different way of viewing
this is to write the phase lag as 2πZ (R) /λ, where λ is the wavelength of a plane
wave at frequency ω. In this view, a nonplanar piston can be considered to behave
like a piston if its maximum depth is much less than the plane wavelength. Con-
versely, in many cases, especially ultrasonic applications like acoustical microscopy,
a transducer is driven at a very large ka in order to obtain a narrow main lobe. In
such cases, it would be wrong to use a piston model to analyze the radiated field.
Equation (8.2.30) will be used in Example 8.4 to develop an approximate nonpiston
model for the pressure radiated by a nonplanar transducer at high ka.
EXAMPLE 8.2 Two designs for a circular transducer embedded in a rigid

baffle are under consideration. One is a conventional piston, whose surface
velocity is v0 cos (ωt) everywhere on its face. The face of the other is a mem-
brane that is stretched across the circumference. Its surface velocity is well
approximated be a parabolic distribution, that is, Vs = v1 1 − R2 /a2 . The
medium in which these devices will operate is water. It is desired that the
pressure amplitude at a distance of 100 m for either transducer be 0.8 atm, and
that the first null of the farfield directivity be at ψ = 2◦ in either case. The
frequency at which the transducers will operate is 30 kHz in order to test the
hearing of sea mammals. For each design, determine the required radius and
the maximum velocity on its face. Then, determine the polar angle of its second
null and the maximum pressure in the side lobe between the first and second
nulls.
Significance
Design problems like this improve our understanding of the relationship between
system parameters and performance.
Solution
The first task is to determine the directivity associated with the parabolic velocity
profile. The farfield description in Eq. (8.2.22) for the given velocity distribution is
simplified slightly by changing the integration variable to ξ = Rs /a, which gives
−ikr 1
2e
Pff (x̄0 ) = iρ0 cv1 ka 1 − ξ 2 J0 (kaξ sin ψ) ξdξ (1)
r 0
We could evaluate the integral numerically for each value of kaξ sin ψ, but an ana-
lytical evaluation is possible. A tabulated integral5 states that
π/2
Jμ (z sin t) (sin t)μ+1 (cos t)2ν+1 dt (2)
0
2ν (ν + 1)
= Jμ+ν+1 (z) , Re (μ) > −1 and Re (ν) > −1
zν+1
where denotes the gamma function. Our interest is with integer values, in which
case (ν + 1) ≡ ν! If this formula is to apply to Eq. (1), then the order of the Bessel
functions and the argument should match. This suggests that we should set μ = 0,
sin t = ξ, and z = ka sin ψ. Furthermore, setting sin t = ξ leads to matching integra-
ν
tion limits. This change of variables gives (cos t) dt = dξ and (cos t)2ν = 1 − ξ 2 .
Thus, setting ξ = sin t, ka sin ψ = z, μ = 0, and ν = 1 transforms Eq. (1) to the inte-
gral in Eq. (2). The result is that the farfield for the membrane transducer is

2J2 (ka sin ψ) e−ikr
Pff (x̄0 ) = iρ0 cv1 ka 2
(3)
(ka sin ψ)2 r
5 Abramowitz and Stegun #11.4.10.

As we did for the piston in Eq. (8.2.27), we define the directivity relative to the
maximum farfield pressure. At ψ = 0, the denominator is singular, but the asymptotic
expansion of the Bessel function gives J2 (ka sin ψ) ≈ (ka sin ψ)2 /8, so we find that

a ka 1 R0
max (|Pff (x̄0 )|) = ρ0 cv1 ≡ ρ0 cv1 @ψ=0
r 4 2 r
(4)
|Pff (x̄0 )| 8 |J2 (ka sin ψ)|
D (ψ) = =
max (|Pff (x̄0 )|) (ka sin ψ)2
Before we address the design criteria, it is worth noting that if v1 and ka for
the membrane equal v0 and ka for the piston, then the maximum farfield pressure
in Eq. (4) is half the value for a piston in Eq. (8.2.26). Separately, we know that at
very low frequencies the radiated pressure is omnidirectional, with an amplitude
that is proportional to the complex volume velocity amplitude. This value for each
design is

Q̂s = Vs dS = v0 πa2
piston
S 2

Rs 1 2
Q̂s = v1 1 − 2 2πRs dRs = v1 πa
parabolic a 2
S
Thus, the ratio of maximum farfield pressures for the two designs is the same as the
ratio of volume velocities.
One design criterion is the value of ψ for the first null, which corresponds to
the first positive zero of the respective Bessel function. For J1 (ka sin ψ) , we have
Table 8.1. We may find the zeros of J2 (ka sin ψ) by inspection of a graph of J2 (x) ,
and a high precision value may be obtained by using numerical methods to find the
roots of J2 (x) = 0. The first few values are
J2 (x) = 0 @ x = 5.136, 8.417, 11.620, 14.796
The frequency is ω = 60, 000π rad/s, and we take c = 1480 m/s. For a first null at
ψ = 2◦ , matching ka sin ψ to the first zero of the respective functions gives
3.832
Piston: ka = = 109.80 =⇒ a = 0.862 m
sin (2◦ )
5.136
Membrane: ka = = 147.15 =⇒ a = 1.155 m
sin (2◦ )
Both are enormous designs, but such transducers can be found on some ocean-going
research vessels.
The other design criterion sets the maximum pressure at 100 m to 0.8 atm =
81.06 kPa. The Rayleigh distance for the membrane transducer, which is the larger
value, is R0 = 85 m, so r = 100 for the specified pressure is in the farfield. Matching
the maximum farfield pressure to the specified value in each case gives

ka a 0.862 24.70
Piston: ρ0 cv1 = 1000 (1480) v0 = 81 103 =⇒
4 r 100 2
v0 = 0.1156 m/s

a ka 1.155 33.11
Membrane: ρ0 cv1 = 1000 (1480) v1 = 81 103 =⇒
r 4 100 4
v1 = 0.1287 m/s (8.2.31)
The required maximum surface velocities for the transducers are close, because the
membrane transducer has a larger radius, which compensates for its smaller volume
velocity per unit surface area.
The last factor to consider pertains to the first side lobe, which is bounded by
the first and second nulls. The second null is located at ka sin ψ = 7.016 for the
piston, and ka sin ψ = 8.417 for the membrane, which corresponds to ψ = 3.66◦ and
ψ = 3.28◦ , respectively. We could perform a mathematical search for the maximum
in each side lobe, but it is easier to scan the data for a plot of each directivity. The
angular dependence for both configurations are depicted in Fig. 1. The first side lobe
maxima are D = 0.1323 at ψ = 2.69◦ for the piston and D = 0.0586 at ψ = 2.49◦ for
the membrane transducer. The graphs show that although the main lobes are closely
matched, the side lobes of the membrane transducer are narrower and much lower
in amplitude. These differences are consequences of the fact that vs is a continuous
function of Rs for the membrane transducer, whereas it is discontinuous at the baffle
for the piston. Discontinuities generally lead to the occurrence of diffraction effects.
100
Piston
Membrane
10-2
10-4
0 30 60 90
100
10-2
10-4
0 1 2 3 4 5
Polar angle  (deg)
8.3 Axial Dependence for a Circular Transducer 153
8.3 Axial Dependence for a Circular Transducer
The farfield approximation yields valuable information, but it does not provide any
insight regarding the range at which it is applicable. The analysis in this section eval-
uates the Rayleigh integral for any distance from the surface. However, the analysis
that follows is not general. It requires that the vibrating region of the surface be circu-
lar, that the pattern of the surface vibration be axisymmetric, and that the field point
be situated on the axis of symmetry. One of its benefits is that the result will shed
light on the nearfield properties, and how those properties transition to the farfield.
The description of the location of a surface point and a field point in Fig. 8.5
are suitable here. The field point is on the z-axis, so ψ = 0. Then, the distance
1/2
from a surface point to a field point is |x̄0 − x̄s | = z2 + Rs2 . As a consequence
of taking the surface vibration to be axisymmetric, the surface velocity is Vs (Rs ) .
Furthermore, because |x̄ − x̄s | does not depend on the polar angle, we may use a ring
of radius Rs and width dRs as the differential element of area, so dS = 2πRs sRs .
According to Eq. (8.1.7), the contribution of this ring to the pressure at the field
point is
e−ik (z +Rs )
2 2 1/2
dPaxial (z) = iωρ0 Vs (Rs ) 1/2 Rs dRs (8.3.1)

z2 + Rs2
This differential contribution is integrated over the surface area, where the vibration
occurs. Thus, we find that

e−ik (z +Rs )
a 2 2 1/2
Paxial (z) = iωρ0 Vs (Rs ) 1/2 Rs dRs (8.3.2)

0 z2 + Rs2
In most situations, evaluation of this integral will require numerical methods, but
an analytical integration is possible in the special case of a translating piston, for
which Vs (Rs ) = V0 . Toward that end, we change the integration variable from Rs to
the radial distance r from a point on a ring to the field point. Because r 2 = Rs2 + z2
and z is constant in the integral, differentiation of this relation gives rdr = Rs dRs .
Thus, the change of variables leads to
(z2 +a2 )1/2
Paxial (z) = iωρ0 V0 e−ikr dr (8.3.3)
z
This change of variables gives the integrand the appearance of a plane wave, in that
there is no 1/r term to account for spreading loss. This perception seems to be borne
out by the integrated form, which is

Paxial (z) = ρ0 cV0 e−ikz − e−ik (z +a )
2 2 1/2
(8.3.4)
The first term is the center wave because its phase is that of a wave that propagates
distance z from the center of the piston to the field point. The distance from the edge
1/2
of the piston to the field point is z2 + a2 , so the second term is the edge wave. In
this view, Eq. (8.3.3) represents a superposition of waves from rings that begin at the
center and end at the edge. The integrated form says that the contributions of all the
ring waves between the center and edge annihilate each other, but there is nothing
coming from beyond Rs = a, which leaves the edge wave as the cancelling effect.
A clearer picture is obtained by using the identities that
1 1/2 1 1/2
z= z + z 2 + a2 + z − z 2 + a2
2 2
2
2 1/2 1 2 1
2 1/2
1/2 (8.3.5)
z +a = z+ z +a − z − z 2 + a2
2 2
Substitution of these expressions into Eq. (8.3.4) leads to

−i(k/2) z+(z2 +a2 )
1/2
k 2
2 1/2
Paxial (z) = 2iρ0 cV0 e sin z +a −z (8.3.6)
2
The complex exponential factor gives the phase lag relative to the surface velocity.
This lag does not increase linearly with increasing z. The sine term is more important
because it tells us that the magnitude of the pressure on the z-axis is
1/2
k a2 + z 2 − kz

|Paxial (z)| = 2ρ0 c |V0 | sin (8.3.7)
2
The maximum pressure that is observed on the axis of symmetry is 2ρ0 cV0 ,
which is twice the plane wave value. Maxima occur at discrete locations called
antinodes that are separated by nodes at which the pressure is zero. Although the
function describing this oscillation is sinusoidal, the argument of the function is not
proportional to z, so the nodes and antinodes are not spaced evenly along the z-axis.
Nodes occur at locations where the argument is nπ, with n > 0 because z ≥ 0. This
condition corresponds to

(ka)2 − (2nπ)2 ka
Nodes: kz = , n = 1, 2, ..., floor (8.3.8)
4nπ 2π
For antinodes, the argument of the sine function must be (2n − 1) π/2, also with
n > 0. This leads to

(ka)2 − (2n − 1)2 π 2 ka + π

Antinodes: kz = , n = 1, 2, ..., floor (8.3.9)
2 (2n − 1) π 2π
The corollary of these relations is that nodes do not exist if ka < 2π, and antin-
odes do not exist if ka < π. Below ka = π, the pressure decreases monotonically
with increasing z. The pressure amplitude at z = 0, which marks the center of the
piston, is found from Eq. (8.3.7) to be 2ρ0 cV0 sin (ka/2)). These observations lead
us to recognize that decreasing the frequency below ka = π will have the effect of
decreasing the pressure everywhere along the axis of symmetry. Above ka = 2π, the
farthest antinode (n = 1) is always more distant than the farthest node.
An important aspect is the behavior at large z. A binomial expansion of Eq. (8.3.6)
based on a/z being small gives

1/2
kz
−ikz a2 kz R0
Paxial (z) = 2iρ0 cV0 e sin 1+ 2 − ≈ 2ρ0 cV0 e−ikz sin
2 z 2 2z
(8.3.10)
where R0 is the Rayleigh distance defined in Eq. (8.2.24). This expression is valid
whenever z a. If z is sufficiently greater than R0 , then the sine function may be
approximated by its argument. As it must to be consistent with the general property of
spherical spreading in the farfield, the pressure amplitude tends to a 1/z dependence,
R0 −ikz
Paxial (z)ff = ρ0 cV0 e (8.3.11)
z
When we set ψ = 0 in Eq. (8.2.27), so that r = z, the result is identical to this

expression. This equivalence suggests that r > R0 marks the beginning of the farfield.
Although this suggestion is based on the axial behavior, it is confirmed by mappings
of the field obtained from numerical evaluations of the Rayleigh integral, as we will
see in the next section.
Figure 8.9, for ka = 30, is a typical axial pattern. There are four nodes and five
antinodes. The spacing between these locations increases with increasing axial dis-
tance, with the increment to the last antinode being much greater than the others.
The antinodes occur sharply, but peaks also occur over distances that are a small
fraction of the piston radius. This is an important feature if one wishes to measure
the nearfield at high frequencies. It means that the microphone/hydrophone must
have a very small diameter in order to probe these small regions, and the position
must be precisely controlled. Figure 8.9 also shows that the farfield approximation
is quite good from the Rayleigh distance outward.
2 Farfield approximation
|P/(0cV0)|
1
R0
0
0 5 10 15 20 25 30
2
|P/(0cV0)|
0
0 1 2 3 4 5 6
Axial distance, z/a
Fig. 8.9 Dependence on distance along the axis of symmetry of the pressure radiated by a piston
in an infinite baffle, ka = 30
EXAMPLE 8.3 The chief engineer of Weird Acoustical Devices, Inc. has
the idea to make an “anti-piston” transducer to be embedded in a baffle. The
idea is to use piezoceramic elements that form an inner circle that covers
0 ≤ Rs < a/2, and a ring covering a/2 ≤ Rs < a. Each set of elements will
be driven by a separate circuit in order that the surface velocity on the outer
ring will have the same magnitude but 180◦ out-of-phase relative to the inner
ring’s surface velocity. Derive an expression for the axial pressure radiated by
this device. Compare it to the axial field of a conventional piston transducer at
ka = 10 and 40. In particular, compare the pattern of nodes and antinodes and
the distance at which the farfield approximation becomes accurate.
Significance
This example will greatly enhance our understanding of the use of the Rayleigh
integral to evaluate an axial field, and the results will highlight the significance of
center and edge waves.
Solution
The given surface velocity is

V0 if 0 ≤ Rs < a/2
Vs =
−V0 if a/2 ≤ Rs < a
The on-axis version of the Rayleigh integral given by Eq. (8.3.1) is valid for any
axisymmetric surface velocity. The discontinuous velocity distribution is handled by
decomposing the integral into two parts,
a −ik (z2 +Rs2 )1/2
a/2 −ik (z2 +Rs2 )
1/2
e e
Paxial (z) = iωρ0 V0 1/2 Rs dRs − 1/2 Rs dRs
0 z2 + Rs2 a/2 z 2 + Rs2
The change of variables r 2 = Rs2 + z2 is appropriate here, so we find that

2 2 1/2 (z2 +a2 )1/2
(z +a /4)
Paxial (z) = iωρ0 V0 e−ikr dr − e−ikr dr
z ( z2 +a2 /4)
1/2

= ρ0 cV0 e−ikz − 2e−ik (z +a /4) + e−ik (z +a )
2 2 1/2 2 2 1/2
This expression resembles Eq. (8.3.6), but here there are three edge waves. The first
term is a signal that departs from the center, the second arrives from the middle radius,
and the third comes from the perimeter. The strength of each signal is proportional
to the velocity discontinuity at the location where it departs. That is, at the center the
surface velocity is +V0 , then it decreases by −2V0 at Rs = a/2, and then it increases
by +Vs to return to the zero value on the baffle. This expression is not amenable to
the type of manipulation that led to an explicit expression for the magnitude of the
pressure, so we will merely evaluate it as a complex function.
Each complex exponential in the preceding has a different argument. It follows that
the maxima and zeros of each term occur independently. Thus, at some locations they
might all have maxima, leading to a pressure amplitude, that is, close to 4ρ0 c |V0 |,
which is twice the maximum for a piston. It might also happen that at some locations
the first and last term are maxima, while the second is a minimum, or vice versa, in
which case the pressure would be close to zero. Figure 1, for ka = 10, confirms this
expectation. The peak pressure occurs close to the node for the circular piston, and
no antinode is evident. However, the farfield pressure is approximately half that for
a piston.
4
Discontinuous Vs
2 Uniform Vs
0
0 2 4 6 8 10
4
0
0 0.5 1 1.5 2
Axial distance, z/a
High values of ka lead to the pressure being very small at some z, but not zero,
as shown in Fig. 2 for ka = 40. The pressure amplitude is at or close to 4ρ0 cV0 at
several locations, but the number of such locations is fewer than for a piston. Only
one minimum is close to zero, and the pressure at another minimum is close to the
maximum for the piston. Overall, the “antipiston” generates a larger pressure in the
nearfield in comparison to the piston, but the pressure is lower in the farfield.
4
Discontinuous Vs
2 Uniform Vs
0
0 5 10 15 20 25 30 35 40
4
0
0 1 2 3 4 5 6 7 8
Axial distance, z/a
Whether this is a useful conceptual design depends on the application. If the intent
is to generate a higher pressure on-axis in the nearfield that varies less and is larger
overall than a piston, then the antipiston might be useful. In contrast, if one requires
a large pressure in the farfield, then the antipiston is not a good idea.
8.4 An Overall Picture of the Pressure Field
At this juncture, we can evaluate the pressure at any point in the farfield. In contrast,
the only aspect of the nearfield that we can evaluate readily is the pressure for a field
point on-axis in the case where the surface vibration occurs over a circular region.
An accurate determination of the pressure at an arbitrary location in the nearfield
requires numerical methods. We shall do so here for the case of a circular piston, but
the modifications to treat other systems are straightforward.
To integrate over Sv as required for the Rayleigh integral, Eq. (8.1.7), we will
approximate the region as a set of patches. We will use the radial distance from
the center of a patch to the field point to describe the contribution of that patch to
the Rayleigh integral. There are more sophisticated numerical algorithms that use
multiple points on a patch, analogously to Simpson’s and higher order rules for
integration over a single variable. The scheme we shall implement is not too difficult
to program, but it will require a very fine grid. The same procedure was used by
Zemanek6 to perform a detailed study that considered a wide range of parameters.
The procedure begins by dividing the circular area Sv into NR rings and a center
circle, as shown in Fig. 8.10. The circle’s radius is set to half the width R of a ring,
which leads to R = 2a/ (2NR + 1) .
Fig. 8.10 Decomposition of y

a circular region into patches

_ n
xj,n
Rj x
R
Each ring is divided into Nθ sectors, which leads to the sectorial increment being
θ = 2π/Nθ . Each patch is associated with
a pair of indices j and n, such that its cen-
tral point has polar coordinates Rj , θn , where Rj = nR and θn = (n − 1/2) θ.
The center circle is assigned to j = n = 0, while 1 ≤ j ≤ NR and 1 ≤ n ≤ Nθ for
the other patches. The inner and outer radii of ring j are rin = (n − 1/2) R and
rout = (n + 1/2) R, and the angle subtended by a patch is θ. Thus, the area of a
patch is Sj,n = (rout2
− rin2 )θ/2 = nθ (R)2 , with S0,0 = π (R)2 /4.
A map of the pressure in the nearfield is best obtained by depicting it in a cylindrical
coordinate grid. The distance from a field point x̄0 , whose cylindrical coordinates are
(R, θ, z) , to the central point x̄j,n on a surface patch is
6 J. Zemanek, “Beam behavior within the nearfield of a vibrating piston,” J. Acoust. Soc. Am. 49
(1971) pp. 181–191.

8.4 An Overall Picture of the Pressure Field 159
1/2
x̄0 − x̄j,n = R cos θ − Rj cos (θn ) 2 + R sin θ − Rj sin (θn ) 2 + z2
1/2
≡ R2 + Rj2 − 2RRj cos (θ − θn ) + z2 (8.4.1)
The integral in Eq. (8.1.7) is approximated as a sum of the contribution of each

surface patch, with that contribution based on using x̄ − x̄j,n for the entire patch.
The resulting discretized form of the Rayleigh integral is

iωρ0 e−ik|x̄0 | π 2
P (x̄0 ) = Vs (0, 0) R +
2π |x̄0 | 4 ⎫
Nθ NR
e−ik |x̄0 −x̄j,n | ⎬ (8.4.2)
+ Vs Rj , θn nθ (R)2
x̄0 − x̄j,n ⎭
n=1 j=1
Equation (8.4.2) may be used to evaluate the field resulting from anysurface veloc-
ity distribution. In the special case of a translating piston, we set Vs Rj , θn = V0 .
The field in that case is axisymmetric, so it is sufficient to compute P only on the
xz plane, where θ = 0. Another saving is that the xz plane divides the surface into
two parts whose contributions are equal, so only noncentral surface patches cover-
ing 0 ≤ θ ≤ π need to be evaluated, with the result doubled. The field is computed
on a grid of R and z values, and depicted as a three-dimensional surface plot, in
which pressure is the third coordinate. The results of such computations are shown
in Figs. 8.11 and 8.12. These computations used NR = 400 and Nθ = 180, that is, the
sectors subtended 2◦ . These values were selected by progressively increasing each
number until the plot of |P| as a function of z along R = 0 matched the on-axis solu-
tion, Eq. (8.3.6). Equally important, the grid of field points must be sufficiently fine
to capture the spatial fluctuations. The plotted results depict a grid whose points are
separated by z = a/50 in both directions. Another consideration is that placement
of the field point on the vibrating surface will lead to a singularity in the integrand if
the field point should happen to fall on one of the central points for a patch. This singu-
larity can be addressed mathematically, as was done to derive the Surface Helmholtz
Integral Equation. Here, the issue was avoided by setting min(z) = z. All of these
considerations mean that the second plot, for ka = 24, for which the maximum value
of z is R0 /2, constitutes a grid of 6 104 field points, and the
evaluation
of the pres-
sure at each field point required evaluation of a sum of 3.6 104 terms. Remarkably,
this computation required less than four minutes on a desktop computer with dual
quad 2.7 GHz Intel processors using MATLAB as the programming language.
The first case to be considered is ka = 6, which is chosen because it is slightly
lower than the frequency at which the on-axis pressure has a null. The maximum
axial distance in Fig. 8.11 is twice the Rayleigh distance, and the width of the plotted
region is twice the piston radius. The most evident feature is the smoothness of
the spatial variation away from the piston’s face. The farfield directivity, which
is shown in Fig. 8.6, is discernible here too. For example, the pressure amplitude
along R = 2a is close to zero at z ≈ 2.4a, which corresponds to a polar angle of
tan−1 (2/2.4) = 39.8◦ , whereas the zero of D (ψ) at this frequency is 39.7◦ . Another
feature worth noting is that the pressure very close to the baffle falls off quickly for
transverse distances that are greater than a.
1.5
0.5
0
6
5
4
Axial 3 0
distan 2
ce, z/a 1 R / a
1 e,
tanc
0 2 se dis
sver
Tran
Fig. 8.11 Pressure amplitude in the nearfield of a piston in an infinite baffle, ka = 6
The nearfield at high frequencies exhibits significant small-scale fluctuations, as

may be seen in Fig. 8.12. This attribute is a consequence of cancelation between
the signals that arrive at a field point from each location on the piston. As the fre-
quency increases, the wavelength of these signals decreases, thereby increasing the
0
0
Tr
an
1
sv
2
ers
1
3
ed
4 nce, z/a
ist
ista
Axial d
an
2 5
ce
6
,R
/a
Fig. 8.12 Pressure amplitude in the nearfield of a piston in an infinite baffle, ka = 24

possibility of constructive and destructive interference. On-axis, such interference is

perfect at the nodes and antinodes and its transitions to the farfield directivity with
increasing radial distance.
The spatial profiles in Fig. 8.13 provide a supplementary view of the data. The
pressure out to z = a fluctuates in the transverse direction, but it is reminiscent of
what happens if a square wave is synthesized from a Fourier series that has an
insufficient number of terms. Furthermore, the phase along these lines changes little
for R < a, which is the characteristic of a simple plane wave. As we saw in the case
when ka = 6, the pressure close to the baffle falls off beyond R > a, but the decay
is much more rapid at ka = 24. At the larger distances, the farfield behavior can be
seen to take over beyond the farthest maximum of the on-axis pressure. This distance
is approximately half the Rayleigh distance.
2
|P/0cV0|
z/a = 0.02 z/a = 1 z/a = R0 /4

1
0
arg(P) (deg)
0
−180
−360
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Transverse distance, R/a
2
R/a = 0 R/a = 1 R/a = 2
|P/0cV0|
1
0
0 2 4 6 8 10 12
Axial distance, z/a
Fig. 8.13 Spatial profiles of the pressure radiated by a piston in a baffle. The plotted data describe
variations parallel and transverse to the axis of symmetry, ka = 24
These trends are magnified in the case where ka = 72, whose profiles are depicted
in Fig. 8.14. The pressure within R < a along the transverse lines close to the piston
and at z = a varies little and has a nearly constant phase. The general features are
like those for ka = 24.
The fall-off of pressure on the baffle beyond the piston radius explains why we may
use analytical results for an infinite baffle model to describe a finite baffle. When
a is used to define dimensionless distances, an acoustic wavelength is 2π/ (ka) .
Maps such as Figs. 8.11 and 8.12 indicate that the pressure field close to the baffle,
z/a 1, is very small for R/a > 1 + π/ (ka) . Furthermore, in this region, the pres-
sure changes quite gradually with increasing R and z. As a consequence, both particle
velocity components are small in this region. Because the surface velocity is inher-
ently small at sufficiently large R/a, there is no need for an infinite baffle to attain that
condition; surrounding the piston with a rigid circular ring whose width equals the
acoustic wavelength should be adequate to mimic the infinite baffle condition. The
qualitative picture of a sound beam emerges from these computations. It is especially
|P/0cV0|
z/a = 0.02 z/a = 1 z/a = R0 /4
1
0
arg(P) (deg)
180
0
−180
−360
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Transverse distance, R/a
2
|P/0cV0|
1 R/a = 0 R/a = 1 R/a = 2
0
0 2 4 6 8 10 12 14 16 18
Axial distance, z/a
Fig. 8.14 Spatial profiles of the pressure radiated by a piston in a baffle. The plotted data describes
variations parallel and transverse to the axis of symmetry, ka = 72
suitable as a description of the radiated sound field for a nonacoustician, although

the fact that it is only correct in a macroscopic sense often is forgotten by practicing
individuals. Figure 8.15 conveys the salient aspects.
Fig. 8.15 Conceptual model

of the sound beam radiated
by a circular piston in a baffle

2a

R
The nearfield is considered to exist in the cylindrical region extending out from the
piston’s face to the Rayleigh distance. Within this region, the field is represented as a
planar wave. The signal beyond the Rayleigh distance is considered to be a spherical
wave within a cone whose semi-vertex angle is ψ0 , with the origin of this wave being
the center of the piston. The angle is found from Fig. 8.15 to be tan−1 (a/R0 ) . The
model is intended for situations, where ka 1, in which case a/R0 = 2/ (ka) is
small. This reduces the arctangent to its argument, which leads to
a 2
ψ0 ≈ = (8.4.3)
R0 ka
For comparison, the maximum polar angle at which the intensity is at least half the
on-axis maximum is half of the beamwidth, which was discussed in in Sect. 8.2.2.
For large ka, this angle was found to be bw /2 ≈ 1.62/ka. The conceptual angle ψ0
for the radiated field is slightly greater than this angle.
EXAMPLE 8.4 Equation (8.2.30) gives a velocity distribution on the plane

of a rigid baffle that supposedly describes the behavior of a transducer whose
moving face translates but is not flush with the baffle. The function Z (R)
is the depth of the transducer’s face at transverse distance R, with positive
Z (R) corresponding to the point being below the xy plane. The discussion of
Eq. (8.2.30) suggested that the phase delay associated with propagation from
the transducer’s face to the xy plane may be ignored if ka is sufficiently small.
On the other hand, if ka is very large, then the nearfield should be reason-
ably close to a plane wave, so the phase-shifted velocity distribution should
be accurate. If so, then using it as the input to the Rayleigh integral should
yield a reasonably accurate description of the pressure field. This hypothesis
may be tested by considering the field of a hemispherical transducer mounted
on a rigid baffle that executes a vibration as a rigid body. The radiated field
was analyzed in Example 7.4 by formulating a spherical harmonic solution.
1/2
The hemisphere’s shape is Z = − a2 − R2 , where the minus sign applies
because the hemisphere is above the xy plane. Correspondingly, the approxi-
mate surface velocity on the baffle is

ik (a2 −R2 )
1/2
Vs = Vz e if R < a
0 if R > a
This function could be used to map the entire field numerically in conjunc-
tion with Eq. (8.4.2). It is much simpler, although less definitive, if only the
farfield and on-axis fields are evaluated according to the Rayleigh integral.
Thus, the task is to evaluate the farfield and on-axis pressure by using the pre-
ceding expression for Vs as the input to the Rayleigh integral, then to compare
the results to those obtained from the spherical harmonic solution. Cases to
consider are ka = 3 and ka = 24.
Significance
Many applications use a transducer whose vibrating face does not lie in the plane of
the baffle. Accurate evaluations of their pressure field requires numerical modeling
techniques like those in Sect. 7.5. A simpler description like the one examined here
would be especially useful for design studies.
Solution
The farfield approximation of the Rayleigh integral for an arbitrary circular transducer
is Eq. (8.2.22). The angular spectrum corresponding to the phase-shifted surface
velocity is
a
f˜ (k sin ψ) = eik (a −Rs2 )
2 1/2
J0 (kRs sin ψ) Rs dRs
0
This expression cannot be integrated analytically, so we will perform a numerical

integration based on the nondimensional distance ξ = Rs /a, which leads to
1
f˜ (k sin ψ) = a2 eika(1−ξ ) J0 (kaξ sin ψ) ξdξ
2 1/2
How to evaluate the integral depends on the software package to be used. The
procedure implemented in MATLAB began by setting ka and ψ. Then an anony-
mous function that evaluates the integrand was used as the input to another function
that performs the integration. The anonymous function specification is F=@(xi)
exp(1i*ka*sqrt(1-xi.ˆ 2)).*besselj(0,ka*xi*sin(psi)).*xi;.
(The period before the exponent and the multiplication signs causes the operations
to apply element by element to data vectors.) Then the integral is found according
to quadl(F,0,1). Both steps must be repeated for each value of ψ from zero to
π/2 in order to construct a directivity pattern.
The on-axis version of the Rayleigh integral is Eq. (8.3.1). For the present Vs
function, this becomes

e−ik (z +Rs )
a 2 2 1/2
ik (a2 −Rs2 )
1/2
Paxial (z) = iωρ0 Vz e 1/2 Rs dRs
0 z2 + Rs2
Changing the integration variable to the radial distance for a field point slightly
1/2
simplifies the integral. Thus, we define aξ = z2 + Rs2 . This expression is solved
for Rs2 in order to transform the integral to
(z2 /a2 +1)1/2

−ika ξ−(z2 /a2 +1−ξ 2 )
1/2
Paxial (z) = iωρ0 Vz e dξ
z/a
This too is not integrable analytically. A MATLAB anonymous function of ξ that eval-
uates the integrand would be G=@(xi) exp(-1i*ka*(xi-sqrt(z_a.ˆ2+1-
xi.ˆ2);. The integral correspondingly is evaluated according to quadl
(G,z_a,sqrt(z_a.ˆ2+1)), where z_a is the value of z/a at the field point.
The spherical harmonic analysis of the pressure field radiated by a translating
hemisphere in a rigid baffle was described in Example 7.4. The first operation eval-
uated the velocity coefficients (V2m )rigid . Division of these values by the respective

values of H2m (ka) gave a set of pressure coefficients,
(V2m )rigid
Bm =
iH2m (ka)
The resulting farfield directivity was

M/2
P r
1
Drigid (ψ) = = B ei(2m+1)π/2
P (cos ψ)
ρ0 cVz a
m 2m
ka m=0
where M is the minimum spherical harmonic index required for convergence to the
desired precision. This series must be recomputed for each value of ψ.
The spherical harmonic series for the on-axis pressure is obtained by evaluating
the general solution in Eq. (7.2.28) at ψ = 0. The Legendre polynomials appearing
in the series at this angle are P2m (0) ≡ 1, and r = z on the axis of symmetry. Thus,
the spherical harmonic series reduces to

M/s z
Paxial (z) = ρ0 cvz Bm h2m ka
m=0
a
The results of this computation for a range of z values beginning at the hemisphere,
z/a = 1, are plotted as (z/a) |P/ (ρ0 cVz )| . Doing so compensates for spherical
spreading, which makes it easier to compare data over a large range of z values.
Figure 1 compares the farfield and axial dependencies obtained from the approxi-
mate Rayleigh integral and the spherical harmonic series for ka = 3. The directivity
patterns are similar for all ψ, with the approximate on-axis value being less than 2
dB higher than the correct one. (This error increases rapidly as ka is decreased below
three.) The on-axis data tends to a constant value at z/a ≈ 3, which suggests that this
is the distance at which the farfield begins. For comparison, the Rayleigh distance is
R0 = ka/2 = 1.5.
(z/a)|P(0cVz )|
Spherical harmonics
1
Rayleigh integral
0.5
0
0 30 60 90
(z/a)|P(0cVz )|
1.2
0.8
0.4
0 1 2 3 4 5 6
Axial distance z/a
Setting ka = 24 leads to very close agreement in the farfield directivity out to

ψ = 60◦ , as can be seen in Fig. 2. The Rayleigh distance for a piston at this frequency
is R0 a = ka/2 = 12, which is consistent with the spherical harmonic prediction of
the value of z/a at which r |P| on axis is a constant value. This is contrasted by
the approximate model, which indicates that a constant value of r |P| is attained at
z/a ≈ 20.
(z/a)|P(0cVz )|
Spherical harmonics
1 Rayleigh integral
0.5
0
0 30 60 90
(z/a)|P(0cVz )|
1.2
0.8
0.4
0 5 10 15 20 25 30 35 40 45
Axial distance z/a
The data for the axial pressure suggests that the approximate analysis is quite
good for the farfield, but exhibits significant errors in the nearfield. It is reasonable to
generalize the preceding observations. If one has a transducer whose vibrating face is
not flush to the plane of the baffle, it might be acceptable to use the Rayleigh integral
with a phase-shifted velocity distribution to evaluate the farfield at high frequencies.
Furthermore, although it might not be very accurate for the farfield at low frequencies
and for the nearfield at any frequency, predictions obtained from this model give a
qualitative picture that resembles the actual field. Such simulations might represent
a useful alternative to numerical modeling techniques, and they can also be used to
verify results obtained by such techniques.
8.5 Radiation Impedance of a Circular Piston
An important aspect of a transducer is the power required to drive it at a specified

vibration amplitude. Aside from dissipation within the transducer, this is the same as
the radiated power. Up to now, we have evaluated the time-averaged radiated power
by using the farfield properties. We could perform a similar analysis here. Instead, we
will determine the properties by analyzing the power that flows out of the piston. The
virtue of such analysis is that it will provide a description of the instantaneous power
output of the piston, which is important for such tasks as designing the electronics to
drive the piston. We begin with a determination of the resultant force exerted by the
fluid on the piston. This will lead to the radiation impedance, which we previously
used in Sect. 3.5.1 to form the end correction at the opening of a waveguide. We will
now see how that correction is derived.
The radiation impedance Zrad for a vibrating piston is defined as the complex
amplitude of the average pressure on the piston’s face divided by the complex
amplitude of the velocity, V0 . The average pressure is the total force divided by
the area, and the total force is obtained by integrating PdS over the face. Thus,
8.5 Radiation Impedance of a Circular Piston 167

Fs
Zrad = , Fs = P (x̄s ) dS (8.5.1)
πa2 V0
Sv
A slightly different interpretation of this definition, which is useful as an extension

to all transducers, is that the radiation impedance is the resultant force divided by
the volume velocity. If we use the Rayleigh integral to evaluate P at a point x̄ on the
surface, then that evaluation itself is an integral over the surface,

e−ik |x̄s −x̄s |

iωρ0
P (x̄s ) = V0 dS (8.5.2)
2π x̄s − x̄
s
Sv
Because x̄s is situated somewhere on Sv , the integrand contains a singularity at

x̄s = x̄s .
Rayleigh7 figured out an ingenious way to circumvent
this singularity. Consider a
pair of surface points x̄s and x̄s . Let us define τ x̄s , x̄s as the contribution to pressure
at x̄s due to a unit surface velocity for a differential patch at x̄s , that is,

e−ik |x̄s −x̄s |

iωρ0
τ x̄s , x̄s = (8.5.3)
2π x̄s − x̄
s

In terms of this quantity, p (x̄s ) is an integral τ x̄s , x̄s V0 dS over Sv , and the resultant
force is an integral of p (x̄s ) dS, also over Sv . Thus, Fs may be found by evaluating

Fs = τ x̄s , x̄s V0 dS dS (8.5.4)
Sv Sv

Rayleigh’s idea was to exploit the reciprocity property that τ x̄s , x̄s = τ x̄s , x̄s ,
which is obvious from its definition, as well as the principle of reciprocity. Rather
than performing the double integral in a single operation, he decomposed it into two
parts. To assist the discussion, we shall refer to xs as the source location and x̄s as
the pressure location. The first integral only considers sources that are closer to the
center than the pressure location, that is, x̄s < |x̄s | . This is the attribute of the lightly
shaded circle Sin in Fig. 8.16.
The second integral accounts for sources that are more distant than the pressure
location. For every pair of points in the first integral, the same pair of points arise in
the second integral. To recognize this, let x̄a and x̄b denote a specific pair of points,
with x̄b being the point that is closer to the center. Then, the first integral contains
τ (x̄a , x̄b ) V0 dSb dSa , while the second integral contains τ (x̄b , x̄a ) V0 dSa dSb . Reci-
procity of τ (x̄b , x̄a ) tells us that these contributions are equal. A similar equality
applies for every pair of points. Therefore, the evaluation of Fs becomes one of the
integrating over the circle Sin whose radius is the radial distance to an arbitrary point
7 J.W. Strutt Lord Rayleigh, Theory of Sound, vol. 2, 2nd ed., Dover (1945 reprint) Sect. 78.
Fig. 8.16 Decomposition of

the face area of a vibrating
B ‘
piston for the purpose of d
evaluating the resultant force A ‘
xs
v
Rs
a
r 
in
xs
O
d
x̄s on its perimeter. The location of this perimeter point is held fixed in the inte-
gral over Sin . It follows that x̄s is the only position variable on which this integral
depends. An integral of this functional dependence over Sin accounts for all sources
that are closer than the pressure location. Doubling that result accounts for all source
locations outside the pressure location, so the evaluation becomes

Fs = 2 τ x̄s , x̄s dS dS (8.5.5)
Sv Sin
The second part of Rayleigh’s contribution

is the way in which the integral over
Sin is carried out. The definition of τ x̄, x̄s depends on the distance between the
points. Recall that x̄s is held fixed in the integral over Sin . Designating it as the
origin for a set of polar coordinates (Rs , φ) is very useful because Rs = x̄s − x̄s ,
as depicted in Fig. 8.16. The integration is performed first over Rs , then over φ. The
length of diameter OA is 2r ≡ 2 |x̄s |. This diameter is the hypotenuse of right triangle
OAB, so the range of radial distances is 0 ≤ Rs ≤ 2a cos φ, and the range of polar

angles is −π/2 ≤ φ ≤ π/2. In this polar coordinate system, dS = Rs dRs dφ, which
cancels the occurrence of Rs in the denominator of τ x̄, x̄s , thereby removing the
singularity. Thus, we find that
π/2 2r cos φ
iωρ0
τ x̄s , x̄s
dS = V0
e−ikRs dRs dφ
Sin 2π −π/2
π/2 0
1
= ρ0 cV0 [1 − cos (2kr cos φ) + i sin (2kr cos φ)] dφ
2π −π/2
(8.5.6)
Because cos φ is an even function relative to φ = 0, the integral is twice the integral
over 0 ≤ φ ≤ π/2, so that
π/2

1
τ x̄s , x̄s dS = ρ0 cV0 [1 − cos (2kr cos φ) + i sin (2kr cos φ)] dφ
Sin π
1 0
π
= ρ0 cV0 − Ic + i Is
π/2 π 2 π/2
Ic = cos (2kr cos φ) dφ, Is = sin (2kr cos φ) dφ (8.5.7)
0 0
An integral like Ic arose in the derivation of the farfield directivity of a ring. The
tabulated integral8 used there extended over 0 ≤ φ ≤ π. The integrand of Ic is an
even function with respect to φ = π/2, so Ic is half the value of the tabulated entry,
that is,
π
Ic = J0 (2kr) (8.5.8)
2
The integration range for Is does not match any entry for Bessel functions in a
basic resource for special functions. However, it does appear in a chapter on Struve
functions, which are denoted as Hν (x) , where ν is the order. Alternative definitions
of this function describe it as a power series or as an integral,9
x ν+1
∞ 2 k
−x /4
Hν (x) =
2 (k + 3/2) (k + ν + 3/2)
k=0 π/2 (8.5.9)
ν
2 (x/2)
= 1/2 sin (x cos θ) (sin θ)2ν dθ
π (ν + 1/2) 0
These alternative definitions feature the Gamma function, denoted as (z) . It is

another special function described in our basic reference.10 It follows the recurrence
relation (z + 1) = z (z) . Thus, for integer arguments, it is related to a factorial
according to (n + 1) = n! For a noninteger positive argument z, it may be computed
by evaluating the function at the fractional part of the argument, zf = z− floor(z) ,
followed by application of the recurrence relation. The fractional part for the preced-
ing formulas is zf = 1/2 and (1/2) = π 1/2 . If the chosen mathematical software
does not provide a routine that evaluates the Struve function, it may be evaluated
numerically according to either representation in Eq. (8.5.9). However, both forms
will encounter difficulty with round-off error if the argument is large, e.g. x > 40.
An asymptotic approximation is quite effective in such situations.
A comparison of the integral representation in Eq. (8.5.9) and the definition of Is
show that
π 1/2 (1/2) π
Is = H0 (2kr) ≡= H0 (2kr) (8.5.10)
2 2
8 M.I. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, 9th ed., Dover (1965),
Eq. (9.1.21).
9 M.I. Abramowitz and I.A. Stegun, ibid, Eqs. (12.1.3) and (12.1.7).
10 M.I. Abramowitz and I.A. Stegun, ibid, Chap. 5.
Correspondingly, we find that

1
τ x̄s , x̄s V0 dS = ρ0 cV0 [1 − J0 (2kr) + iH0 (2kr)] (8.5.11)
Sin 2
The total resultant force is found by integrating this term over Sv . Because r is
the distance from the center to x̄s , the integration is done by changing to a polar
coordinate system centered on Sv . Thus, we must evaluate
π
a
Fs = 2 τ x̄s , x̄s dS dS = ρ0 cV0 [1 − J0 (2kr) + iH0 (2kr)]rdθdr
Sv Sin 0 −π
(8.5.12)
The θ integration gives a 2π factor. To integrate the Bessel and Struve function terms,
we invoke a recurrence relation that has a similar form for both, specifically,
d n d n
(y Jn (y)) = yn Jn−1 (y) , (y Hn (y)) = yn Hn−1 (y) (8.5.13)
dy dy
The result is
Fs = πa2 V0 Zrad , Zrad = ρ0 c χ (ka) (8.5.14)
where the specific impedance is

J1 (2ka) H1 (2ka)
χ (ka) = 1 − +i (8.5.15)
ka ka
Figure 8.17 repeats from Chap. 3 the graph of χ as a function of ka.
1
Re()
Im()
0.5
0
0 5 10 15 20
Nondimensional frequency ka
Fig. 8.17 Plot of the frequency dependence of the specific acoustic impedance χ for a vibrating
circular piston of radius a
For small ka, the first few terms of a Taylor series expansion are adequate,

1 1 4 2 8
χ (ka) = (ka)2 − (ka)4 + · · · + i (ka) − (ka)3 + · · ·
2 12 π 3 45
(8.5.16)
The specific impedance tends to χ (ka) = 1 as ka → ∞, which is merely the plane

wave limit associated with high frequencies.
One reason that the specific impedance is important is that it is needed to predict
the voltage and current required to drive the piston transducer at a specified velocity.
Knowledge of the specific impedance provides a simple way to evaluate the power
required to drive a piston. Aside from losses within the transducer, this is the same as
the power radiated by the piston. Let us consider this quantity as a function of time.
All points on the piston’s face translate at the same velocity, so the instantaneous
power is the product of the resultant force and that velocity. The instantaneous power
input to the fluid at the piston face, therefore, is
1
P (t) = Re Fs eiωt + c.c. V0 eiωt + c.c.
4
1
= Re Fs (V0 )∗ + Fs V0 e2iωt (8.5.17)
2
πa2
= (Re Zrad ) |V0 |2 + Re (V0 )2 Zrad e2iωt
2
The invariant part of this expression is the time-averaged power,
πa2
Pav = Re (Zrad ) |V0 |2 (8.5.18)
2
The fluctuating part is the power that flows into and out of the fluid
across
the piston’s
face. It oscillates at a frequency 2ω with an amplitude that is πa2 /2 |Zrad | |V0 |2 , so
|Zrad |
|Pfluct | = Pav (8.5.19)
Re (Zrad )
It follows that the average power flow across the piston face never exceeds the
fluctuating part, so there is a time interval during which the power flows from the
fluid into the piston. In this condition, the pressure resultant acting on the piston is
in the same sense as the piston’s velocity.
It is enlightening to compare the power flow across the piston face to the power
that flows across a very large surrounding hemisphere. We know that the average
portions should be the same, but what about the fluctuating parts? To answer this,
we us a large hemisphere whose radius is sufficiently large that we may use the
farfield approximation that P/Vr = ρ0 c at any polar angle. The time-domain version
of Eq. (8.2.23) is

J1 (ka sin ψ) a 1 −ikr iωt

pff (x̄0 , t) = ρ0 c V0 e e + c.c. (8.5.20)
sin ψ r 2
The instantaneous radial intensity is p2ff / (ρ0 c) , and the field is axisymmetric, so
dS0 = 2πr 2 dψ. Hence, the instantaneous power flowing across a hemisphere of
radius r is given by
π
1 2
Pff (t) = pff (x̄0 ) 2πr 2 dψ
0 ρ0 c
π
J1 (ka sin ψ) 2
= πρ0 ca 2
dψ |V0 |2 + Re (V0 )2 e2i(kr−ωt)
0 sin ψ
(8.5.21)
There is no need to evaluate the integral because the time-averaged power is the
constant part of this expression. The net power flowing into any region of an ideal
fluid must have a zero average value, so this portion of Pff (t) must equal Pav in
Eq. (8.5.18). Hence, writing V0 in polar form leads to
πa2
Pff (t) = Re (Zrad ) |V0 |2 1 + e2i(kr−ωt+arg(V0 )) (8.5.22)
2
This description of Pff (t) highlights the fact that the fluctuating part has an amplitude
that equals the mean value. This is different from the behavior at the transducer face,
where the magnitude of the fluctuating part of the power flow exceeds the average.
At any instant, the difference between the fluctuating parts of the power radiated by
the piston and the power that flows into the farfield represents the rate of change of
the fluid’s mechanical energy in the region close to the piston.
EXAMPLE 8.5 A circular 450 mm diameter piston is mounted in a baffle on

the floor in an anechoic chamber. In one test, the electrical signal creates a
square wave alternating between +V0 and −V0 with a fundamental frequency
of 1 kHz. The sound pressure on-axis 4 meters from the piston is measured to
be 125 dB/20 μPa. This measurement is cutoff at 18 kHz in order to account
for audible tones only. Determine the power output as a function of time. Also
determine the waveform measured at the 4 m distance on-axis. Graph both
time-dependent quantities.
Significance
Beyond the obvious objective of performing an evaluation of radiated power, we will
see that power requirements are significantly affected by the harmonic content. In
addition, the pressure waveform provides a preview of an interesting phenomenon
that arises in a time-domain analysis.
Solution
The value of V0 is not specified. We will determine it by evaluating the pressure at
z = 4 m on-axis in terms of V0 , which will yield a value of the mean-squared pressure
at that location. The square wave is specified to be a periodic signal defined by
vs (t) = V0 [h (t) − 2h (t − T /2)] if 0 ≤ t ≤ T , vs (t ± T ) = vs (t) (1)

The value of ω1 is 2000π rad/s. The Fourier series representation of this signal is

∞ 2π
vs = Re Vn eiωn t , ωn = nω1 , ω1 =
n=1 T (2)
2
Vn = 1 − (−1)n V0
iπn
Only the Vn values for odd n are not zero. Each harmonic contained in the surface
vibration is associated with a different wavenumber,
kn = nω1 /c (3)
Each value of kn leads to a contribution to the on-axis pressure according to

Eq. (8.3.6). The resulting Fourier time series is
∞
p (z, t) = Re Pn (z) einω1 t

n odd (4)
4V0 −i(kn /2) z+(z2 +a2 )1/2 kn 2 1/2
Pn (z) = 2iρ0 c e sin z + a2 −z
iπn 2
We take c = 340 m/s and ρ0 = 1.2 kg/m3 as representative values for air. The pressure
coefficients at z = 4 m, R = 0 of the odd harmonics are tabulated below. They are
cutoff at 18 kHz, which for a 1 kHz fundamental corresponds to n = 18.
n 1 3 5 7 9

Pn
Re 9.1155 −26.4029 40.9712 −51.3675 56.6292
V0 z=4
Pn
Im 59.9788 −54.3143 43.6173 −29.0723 12.2696
V 0 z=4
n 11 13 15 17

Pn
Re −56.3883 50.9007 −40.9956 27.9556
V0 z=4
Pn
Im 4.9911 −20.9190 33.9445 −42.8971
V0 z=4
The given sound pressure of 125 dB at z = 4 m corresponds to a mean-

level
squared pressure of 1.2649 103 Pa2 . At the same time, the mean-squared pressure is
the incoherent sum of the mean-squared pressure of each harmonic in the tabulation,
that is,
18
2 1 Pn 2
p av = |V0 |2 = 14719 |V0 |2 = 1.2649 103
2
V0 z=4
n odd
From this, we find that the piston’s velocity amplitude is

V0 = 0.2932 m/s
With this value determined, the actual Vn and Pn coefficients can be computed. The
primary factor deciding the time increment for synthesis of the Fourier series is how
well we wish to capture the rise and fall of the square wave. Let us select t = T /100
for this purpose. Presumably, the full harmonic content of the pressure is measured
and then filtered for the content below 18 kHz. It is possible that despite such filtering,
the existence of higher harmonics is important, especially for considerations related
to the power output of the transducer. To ascertain whether such speculation is correct,
all series shall be synthesized for a series length of N = 100, as well for N = 18.
The first evaluation tests the adequacy of either truncation by evaluating the input
velocity waveform according to Eq. (2).
We could use an inverse FFT routine to create the waveforms, but we shall imple-
ment the matrix algorithm in Eq. (1.4.39). Thus, vs at instant tj is indicated by Eq. (1)
to be

N
vs (tj ) = Re Vn eiωn tj =⇒ {vs } = [E]H {V } (5)
n odd
where
⎡ ⎤
e−iω1 t1 e−iω1 t2 ...
⎢ −iω2 t1 e−iω2 t2 ···⎥
{V } = [V1 V3 · ··]T , [E] = ⎣ e ⎦
.. .. ..
. . .
The superscript in Eq. (5) specifies the Hermitian of [E] , which is the complex
conjugate transpose of the array.
Figure 1 shows that either truncation gives an adequate reproduction of the square
wave, although the ringing for the smaller N is noticeable. Note that this ringing
stems from the fact that N = 18 does not meet the Nyquist criterion.
0.4
0.2 N = 100
vs (m/s)
N = 18
−0.2
−0.4
0 0.5 1 1.5 2
Time (ms)
Figure 2 depicts the pressure waveform at z = 4 m. It was computed by adapting

the matrix algorithm in Eq. (5) to Eq. (4). The result is somewhat surprising because
it indicates that the waveform is a sequence of spikes. One clue why this is so is
found in the Rayleigh distance (R0 )n = kn a2 /2 for each harmonic. This distance for
the first harmonic is 0.468 m, and it is not until n = 9 that z = 4 m is less than the
Rayleigh distance. Thus, for the most significant harmonics, the field point is in the
farfield. According to Eq. (8.3.11), the contribution of each harmonic at fixed z in
the farfield is proportional to (R0 )n Vn . The Vn values are inversely proportional to
n, and the Rayleigh distance of each harmonic is proportional to ωn , which means
that the lower Pn (z) values should have nearly equal magnitude. This expectation is
borne out by the tabulation of Pn (z) /V0 . A Fourier series whose coefficients have
equal magnitude, with suitable relative phase delays, represents a periodic train of
impulses. This is manifested as the pressure being a series of alternating spikes. But
why does this happen? That explanation will be found in the next section, which
explores the time-domain Rayleigh integral.
200 N = 100
p@z=4m
N = 18
−200
0 0.5 1 1.5 2
Time (ms)
It is not necessary to synthesize waveforms to evaluate the time-averaged power

because the intensity is the incoherent sum of contributions from each harmonic.
Therefore, Pav is the sum of terms described by Eq. (8.5.18), with Zrad for each n
evaluated at the respective kn value. That is,
πa2 ρ0 c N
Pav = Re χ (kn a) |Vn |2 (5)
2 n odd
The result for N = 100 is Pav = 5.27 W. Cutting off the series at N = 18 yields
Pav = 5.17 W, so the shorter series length seems to be quite adequate.
Unlike the evaluation of the average power, interference between harmonics is an
important feature of the instantaneous power. We begin this evaluation by describ-
ing the force resultant as a Fourier series. The complex amplitude for a harmonic
is described by Eq. (8.5.14), which we adapt by evaluating Zrad at the respective
harmonic frequency and multiplying it by the velocity coefficient Vn in Eq. (2). The
series is
N
Fs = πa ρ0 c Re
2
χ (kn a) Vn einω1 t (6)
n odd
The matrix algorithm in Eq. (5) for this calculation is

{Fs } = [E]H {{χ} · {V }} (8.5.23)
where {{χ} · {V }} denotes an element by element multiplication.

The instantaneous power may be computed by multiplying the Fs value at any t

by the vs value plotted in Fig. 1. The result of both computations is shown in Fig. 3.
The force waveform resembles the piston’s velocity, except that the plateaus are not
flat. The radiated power is positive if vs and Fs have the same sign, so the plot of
Prad is a periodic sequence of positive plateaus, with T /2 as the period. The instants
at which Prad is zero are the instants at which vs = 0. There is no interval in which
Prad is negative. This is so because all kn a > 3, so all χ (kn a) are close to unity. The
maximum radiated power, which occurs soon after the transitions of vs , is 6.18 W.
The power falls off after the peak, then drops off rapidly at the velocity transitions.
The consequence is that the average power does not differ much from the peak value.
An interesting aspect is that the inadequacies of using the smaller N are evident for the
force, and especially the power. The shorter series length leads to max (Prad ) = 8.84
W, which is more than 40% high.
20 N = 100
N = 18
Fs (N)
−20
10
Prad (W)
0 0.5 1 1.5 2
Time (ms)
8.6 Time Domain Rayleigh Integral
A time-domain description of the signal generated by a vibrating piston has advan-

tages in some circumstances. One approach describes the field as a superposition
of plane and evanescent waves of the type studied in Sect. 5.1. A two-dimensional
Fourier transform of the pressure leads to a representation that matches the angular
spectrum of the surface velocity. The requirement that the pressure at each wavenum-
ber matches the normal surface velocity yields a representation of the pressure as
an inverse Fourier transform. From a philosophical viewpoint, this formulation is
attractive because it provides a similar perspective to that provided by the prior
investigations of radiation from spheres to cylinders. The formulation is primarily
intended for numerical evaluations, for which FFT technology substantially reduces
8.6 Time Domain Rayleigh Integral 177
the computational effort. The paper by Wu, Kazys, and Stepinski11 explains the
details. We shall not delve into this line of investigation because it does not lead to
new insights as to the nature of the pressure field.
In contrast, the time-domain version of the Rayleigh integral provides a different
perspective. For example, it provides a simple explanation of the impulse-like wave-
form observed in the previous example. However, evaluation of the pressure field at
an arbitrary field point is more difficult than it is in the frequency-domain approach.
We have regarded the Rayleigh integral as a prescription for evaluating the
frequency-domain response when the surface vibration is a set of point sources asso-
ciated with Re (Vs (x̄s ) exp (iωt)). A different perspective views the Rayleigh integral
as the Fourier transform of a time-domain representation of the radiated field. Our
task is to determine what that time representation is.
We shall not call on the inverse Fourier transform to carry out this determination.
Rather, we will use the notion of function/transform pairs for operations in the time
domain, such as those in Table B1 of Appendix B. The first step is to recognize
that the factor iωVs (x̄s ) in the Rayleigh integral is the Fourier transform of the rate
of change of vs (x̄s , t) , that is, F ((∂vs /∂t) , ω) = iωVs (x̄s ). Hence, the Rayleigh
integral may be written as

−ik|x̄0 −x̄s |
ρ0 ∂ e
P (x̄0 ) = F vs (x̄s , t) , ω dS (8.6.1)
2π ∂t |x̄0 − x̄s |
S
The other occurrence of ω is in the complex exponential. The coefficients of a

Fourier series that are delayed by time τ relative to another function are described in
Eq. (1.4.11). It is shown there that if p (t) and q (t) are two functions that have the same
period T , with q (t) = p (t − τ ) , then their Fourier series coefficients are related by
Qj = Pj e−iωj τ , where ωj = j (2π/T ). A Fourier transform corresponds to the coef-
ficients of the Fourier series of a function whose period is infinite, which replaces
the discrete frequencies ωj with a continuous variable ω. Thus, if g (t) = g (t − τ ) ,
their Fourier transforms are related by F (g (t) , ω) = F (f (t) , ω) e−iωτ . (This prop-
erty also may be identified from the coordinate shifting property in Table B1 of
Appendix B, except that the sign of the exponent must be reversed because the tab-
ulation is based on our definition of a spatial Fourier transform.) Because k = ω/c,
the exponential in the Rayleigh integral corresponds to delaying a function by
|x̄0 − x̄s | /c. Correspondingly, the preceding form of the Rayleigh integral may be
written as

ρ0 ∂ |x̄0 − x̄s | 1
P (x̄0 ) = F vs x̄s , t − ,ω dS (8.6.2)
2π ∂t c |x̄0 − x̄s |
S
11 Ping Wu, Rymantas Kazys, and Tadeusz Stepinski, “Analysis of the numerically implemented
angular spectrum approach based on the evaluation of two-dimensional acoustic fields. Part I. Errors
due to the discrete Fourier transform and discretization,” J. Acoust. Soc. Am. vol. 99, Issue 3, pp.
1339–1348 (1996); (10 p).
By definition, the inverse Fourier transform of a Fourier transform is the original

time-domain function. Because the source-field point distance is independent of ω,
it is constant in the transformation. We thereby obtain the time domain Rayleigh
integral,

ρ0 1 ∂ |x̄0 − x̄s |
p (x̄0 ) = vs x̄s , t − dS (8.6.3)
2π |x̄0 − x̄s | ∂t c
S
This form is readily understood in terms of source distributions. Consider a differ-

ential element of the surface. The mass acceleration of this element is ρ0 dS dvs /dt,
and the distance from it to the field point is r = |x̄0 − x̄s | . The time-domain pressure
radiated by this element is double the value in Eq. (6.4.3) because the sound radiates
into a half space, rather than a free space. Thus, it is
ρ0 ∂ r
dp (x̄0 , t) = dQs t − (8.6.4)
2πr ∂t c
The time-domain Rayleigh integral is the superposition of the signals radiated by
a sheet of these sources distributed over the surface. The signals from different
sources that arrive contemporaneously at a field point departed from the surface at
different instants. Specifically, the signal from a differential source at x̄s that arrives
at the field point at time t must have left the surface at time t − |x̄0 − x̄s | /c. Hence,
the signal observed at a field point at time t consists of contributions that departed
from the surface in a time interval beginning at t − max (|x̄0 − x̄s |) /c and ending at
t − min (|x̄0 − x̄s |) /c.
An evaluation of the Rayleigh integral for an arbitrary location and surface velocity
requires numerical methods. The technique described by Fig. 8.10 also is suitable for
this application. The radial distance from the central point on a surface patch to the
selected field point is used to evaluate t − r/c for that patch. Then, Eq. (8.6.4) is
added for each patch.
If we limit our consideration to the on-axis pressure radiated by a circular piston,
we may evaluate the integral analytically. The analysis follows similar steps to that for
the frequency domain. Thus, a point on the surface is located by its polar coordinates
(Rs , θs ). The distance from such a point to the field point at distance z from the
1/2
surface is r = Rs2 + z2 . The value of r is independent of θs , so an integration
over this variable leads to a 2π factor. In addition, vs for a piston is the same for all
Rs . Thus, Eq. (8.6.3) becomes
2 1/2
a
1 ∂ Rs + z2
p (x̄0 , t) = ρ0 1/2 vs t − Rs dRs (8.6.5)
0 R2 + z 2 ∂t c
s
At this stage, we change variables from Rs to the radial distance r. Because z is con-
stant, differentiation of r 2 = Rs2 + z2 leads to rdr = Rs dRs . Furthermore, the limits
8.6 Time Domain Rayleigh Integral 179
1/2
Rs = 0 and Rs = a correspond to r = z and r = a2 = z2 , which leads to
(a2 +z2 )1/2

∂ r
p (x̄0 , t) = ρ0 vs t − dr (8.6.6)
z ∂t c
The last step is to recognize the meaning of the derivative of vs . Let denote the
argument, = t − r/c. Then, the chain rule for differentiation leads to
∂ r ∂ d dvs
vs t − ≡ vs () =
∂t c ∂t d d
(8.6.7)
∂ r ∂ d 1 dvs 1 ∂ r
vs t − ≡ vs () = − =− vs t −
∂r c ∂r d c d c ∂t c
Usage of this property to replace the time derivative with a derivative with respect
to r leads to
(a2 +z2 )1/2
∂ r
p (z, t) = −ρ0 c vs t − dr (8.6.8)
z ∂r c
The integrand is a perfect differential, so we find that

2 1/2
z a + z2
p (z, t) = ρ0 c vs t− − vs t − (8.6.9)
c c
This is the time-domain equivalent of Eq. (8.3.4). The first term is the center wave,
which leaves the center of the piston at time z/c prior to its arrival at the field point.
The second term is the edge wave, whose propagation time prior to arrival at the field
1/2
point is a2 + z2 /c.
Consider the case where the surface vibration is a pulse, such that vs (t) = 0 for
t < 0 or t > T . Then, the signal received at an axial point will be the same shape
pulse in the edge wave, followed by the negative of the first pulse
emanating from

1/2
the edge of piston. The interval between the arrival times is a + z2 2
− z /c.
If this interval is greater than T , the pulses will appear as separate events. This
inequality
may besolved for the range of z in which the pulses occur individually. It
is z < a2 − c2 T 2 / (2cT ).
Some insight into the farfield behavior may be gained by considering axial loca-
1/2
tions at which z/a 1. Then, a2 + z2 ≈ z + a2 / (2z) . The second term is much
less than z, so application of a Taylor series to Eq. (8.6.9) leads to
ρ0 a2 z
p (z, t) → v̇s t − (8.6.10)
2z c
Thus, the pressure signal received at an axial point far from the piston will mirror
the surface acceleration.
V0
vs
0
−V0
0 0.5 1 1.5 2
0.5V0 δ(t+11.0T-z/c)
pz/ (ρ 0 a 2 )
0.5V0 δ(t+10.0T-z/c)
0
−0.5V0 δ(t+11.5T-z/c) −0.5V0 δ(t+10.5T-z/c)
0 0.5 1 1.5 2
Time (ms)
Fig. 8.18 The pressure signal received at a point on the axis of a circular piston whose velocity vs
is a square wave
An interesting consequence of this property is what is heard when a piston moves

as a square wave, which was the situation in Example 8.5. No pressure is generated
when the velocity is constant, so the only sound that reaches the farfield is a series
of spikes at the discontinuities. This is illustrated in Fig. 8.18, which depicts the
time-domain construction of vs and the pressure waveform at z = 4 m, as opposed to
the representation of these signals as truncated Fourier series in the aforementioned
example. The value of z/c = 11.76 ms, so the first impulse to be observed for t > 0
is a negative one corresponding to the discontinuous decrease of vs at t = −11.5T .
Of course, it is not possible to change the velocity discontinuously, because doing
so would require an infinite force. Thus, the graph is an approximation of a rapid
transition, just as an impulsive force is an approximation of a large force that acts
over a very short time interval.
Additional analyses have been performed for various off-axis locations. For exam-
ple, Blackstock12 describes an approximate time-domain farfield analysis for a cir-
cular piston. Pierce13 shows how to determine the pressure radiated by a circular
piston at an arbitrary field point as a single integral, rather than an integral over an
area. These analyses are somewhat complicated, so we shall not explore them.
Exercise 8.1 A rectangular plate is flush mounted in an infinite rigid baffle that lies
in a vertical plane xy. The plate is very rigid, so it undergoes a uniform translational
vibration vz in the normal z direction. The frequency of this vibration might be any
value up to 1.2 kHz. Within this interval, the farfield mean-squared pressure in the
(horizontal) xz plane at 30◦ from the z-axis should be no more than 6 dB less than the
on-axis value. Also, the farfield mean-squared pressure at 5◦ from the z-axis in the
12 D.T. Blackstock, Fundamentals of Physical Acoustics, John Wiley & Sons (2000) pp. 461–463.
13 A.D. Pierce, Acoustics, McGraw-Hill (1981), reprinted by ASA Press (1989) pp. 227–231.
vertical yz plane should be no more than 6 dB less than the on-axis value. Determine
the horizontal and vertical edge lengths that meet these specifications.
rigid
baffle
b/2
x b/2 a/2 z
a/2
Exercise 8.2 The sketch depicts a square plate that is embedded in a very large rigid
baffle. The plate is flexible, with motion actuated 180◦ out-of-phase from one corner
to the next, with the result that the surface normal velocity is vz = V0 (xy/b2 ) cos (ωt) ,
where b is the length of each edge. Derive an expression for the farfield directivity.
For the case where kb = 10, evaluate and graph this function for points in the xz
plane and yz plane, as well as a plane that contains the z-axis at 45◦ from the x- and
y-axes.
z v
z
vz vz
x
vz
y
Exercise 8.3 The sketch depicts a submerged square transducer mounted on a large
rigid baffle that lies in the vertical plane. The edges of the transducer are horizontal
and vertical, with the upper edge at depth h below the free surface. This depth is
sufficiently large that the baffle may be considered to extend infinitely in all directions
for the purpose of formulating the Rayleigh integral. The transducer executes a
uniform vibration v0 sin (ωt) in the z direction, which is perpendicular to the plane
of the baffle. It is permissible to approximate the free surface as being pressure-
release for waves in the water. (a) Derive an expression for the complex amplitude
of the pressure in the farfield as a function of spherical coordinates for which z is

the polar axis and r measures the distance from the xyz origin. (b) Consider the case
where a = 0.4 m and the frequency is 5 kHz. Determine and plot the farfield radiation
(r/a) |pff | / (ρ0cv0 ) as a function of angle from the z-axis for points in the yz plane.
free surface
h
x
a
a/2
a/2
rigid
baffle
y
Exercise 8.4 It is desired to build a baffled circular piston projector that generates a
signal in air whose beamwidth is less than 1.5o for frequencies greater than 16 kHz.
What is the smallest radius a of the piston that fits this specification? For this value
of a, what is the beamwidth at 16 kHz, 1.6 kHz, and 160 Hz? Graph the farfield
directivity at each of these frequencies.
Exercise 8.5 A Gaussian velocity distribution on a flat surface is one for which the
velocity distribution
depends on the polar distance Rs from the center according to
vz = V0 exp −βRs2 sin (ωt) . This distribution is desirable because it limits diffrac-
tion effects that cause nulls due to a discontinuous surface velocity distribution. In
reality, this concept is an approximation because Rs cannot exceed the active radius
a. Thus, the Gaussian surface velocity distribution applies for Rs < a, and vz = 0
for Rs > a. Use numerical methods to evaluate the farfield directivity when ka = 12.
Perform this evaluation for β = 0, β = 0.5, β = 2, and β = 8. What conclusions
can be drawn from these evaluations?
Exercise 8.6 Derive an expression for the farfield directivity of the transducer in
Example 8.3. Evaluate the result for ka = 5 and ka = 20. Compare each pattern to
the directivity of a baffled piston.
Exercise 8.7 A circular plate flush mounted in an infinite baffle executes a harmonic
motion in which it rotates as a rigid body about a diametral line. The consequence
is that the surface velocity is v1 (y/a) sin (ωt) . Derive an expression for the farfield
pressure. Evaluate and graph this function for points in the xz plane and yz plane for
the case where ka = 10.
vz vz
vz
y
vz
vz
vz
vz x
vz
Exercise 8.8 Drawing I shows a projector embedded in the surface of a rigid infinite
baffle. The projector face is a semicircular piston, with the velocity being constant
along the face. If Rs , θs are polar coordinates for a point on the face, with x =
Rs cos (θs ) , y = Rs sin (θs ) , then a harmonic oscillation of the piston leads to a
surface velocity described by

v0 cos (ωt) , 0 < Rs < a, −π/2 < θs < π/2
vz =
0 otherwise
(a) Consider a field point x̄0 on the z-axis (ψ = 0) . Starting with the general Rayleigh
integral, derive an expression for the complex amplitude of the pressure at such a
location. How does that result compare to the result for the case of a full circular
piston? (b) Starting with the general Rayleigh integral, derive an expression for the
complex amplitude of the pressure at location x̄0 in the farfield, r a. The final
result may be left in integral form. (c) In Drawing II, an infinite baffle is situated in
the vertical plane. A vibrating piston is mounted on the baffle such that its center is
at the free surface of the water. For the purpose of the analysis, the free surface of
the water may be taken to be pressure-release for waves in the water and rigid for
waves in the air. Let Pb (r, φ, θ) denote the pressure derived in Part (b). Describe in
terms of Pb (r, φ, θ) the pressure in the water and in the air.
z y
 _
x0
a free
v0 surface
r
r z
y 
_
a x0
a 
x
(I) x (II)
Exercise 8.9 A circular piston in a rigid baffle executes a transverse vibration

v0 sin (ωt) everywhere along its surface. The following facts are known: (1) The
fluid is water. (2) The frequency is 50 kHz. (3) The first null in the farfield pressure
occurs at 3◦ from the z-axis. (4) The pressure at 3 m along the z-axis from the plate
is 14 kPa. Determine the radius a and velocity amplitude v0 that correspond to the
stated conditions.
Exercise 8.10 An ultrasonic piston transducer in a large rigid baffle transmits a
signal into water. A scan of the axial field indicates that the largest pressure is 195
dB//1 μPa, and that farthest distance at which it occurs is 28 mm from the face of the
piston. It also is observed that the farthest antinode occurs at 12 mm from the face.
Determine the radius a of the piston, the frequency, and the amplitude of the piston’s
velocity. Also estimate the minimum radial distance at which farfield properties can
be expected to be observed.
Exercise 8.11 A transducer is composed of multiple piezoeceramic elements that
are driven in-phase at frequency ω, with the individual elements approximately filling
a circle of radius a. This circle is surrounded by a rigid baffle, that is, sufficiently large
to consider it to be infinite. Because this transducer is not exactly a circular piston
the effective radius aeff for acoustical properties might not be exactly the geometrical
radius a. The value of aeff , as well as the local speed of sound, can be determined from
the axial distances of the farthest two minima of the on-axis pressure amplitude. Let
these distances be zA and zB < zA . (a) Derive formulas for aeff and c corresponding
to a known frequency ω and measured values of zA and zB . (b) Evaluate the results
in Part (a) for the case of a 10 kHz signal with zA = 0.395 m, and zB = 0.105 m.
(c) How many pressure antinodes will be observed on the symmetry axis when the
parameters are those in Part (b)? (d) For the parameters in Part (b), what distance
is a good estimate for the onset of farfield behavior? (e) What is the beamwidth
corresponding to the parameters in Part (b)?
Exercise 8.12 An underwater baffled piston projector must have a maximum sound
pressure level of 240 dB//1 μPa on-axis when the frequency is 8 kHz. The farthest
distance at which this maximum occurs should be 3 meters. (a) What is the required
radius of the piston? (b) At what distance is it reasonable to consider farfield spreading
to begin? (c) How many nulls are there on-axis? (d) What is the sound pressure level
at 30 m from the projector on-axis? (e) What is the beamwidth of this projector? (f)
How many nulls occur in the directivity?
Exercise 8.13 An infinite plane is the boundary of a half space filled with an ideal
fluid. The entire plane outside a circle of radius a executes an in-phase harmonic
vibration normal to the plane. With z being the axial direction for a set of cylindrical
coordinates, and z = 0 designated as the plane, the surface velocity is vz = v0 sin (ωt)
if R > a and vz = 0 if R < a. Derive expressions for the pressure along the z-axis
and the pressure at an arbitrary location in the farfield. For the case where ka = 20,
distribution |P| /(ρ0 cv0 ) as a function of z/a. Also, graph the farfield
graph theaxial
pressure Pff /(ρ0 cv0 ) as a function of distance R/a from the symmetry axis. The
axial distance for this evaluation should be z = R0 .
Exercise 8.14 A projector in an infinite baffle consists of a ring whose inner radius is
a/2 and whose outer radius is a. The complex normal velocity amplitude at frequency
ω is v0 over the surface of the ring, while the region inside the inner radius is
motionless. Derive expressions for the farfield pressure and the pressure on-axis.
Given that ka = 25, evaluate (r/a) |pff / (ρ0 cv0 )| as a function of the polar angle, and
the axial pressure distribution |p (ψ = 0) / (ρ0 cv0 )| as a function of distance from
the piston. Do nulls occur in either pattern?
Exercise 8.15 Example 8.2 analyzed the farfield radiation of a membrane trans-
ducer. The membrane’s
surface
velocity is well approximated be a parabolic distrib-
ution, Vs = v1 1 − R2 /a2 sin (ωt) if R < a, Vs = 0 if R ≥ a. Evaluate |p/ (ρ0 cv0 )|
as a function of distance along the axial center line from the plane of the baffle.
Perform this evaluation for ka = 3 and 20.
Exercise 8.16 The sketch depicts a transducer that radiates sound into water. It
consists of a piston that is flush mounted in an infinite baffle. The piston, whose
mass is M = 0.6 kg, is restrained by spring K. The displacement at the other end
of the spring is controlled by an actuator that can exert whatever force is required
to impose a specified harmonic motion vact cos (ωt) . Design specifications are as
follows: (1) The piston radius is 150 mm. (2) The operational frequency is 2 kHz. (3)
The natural frequency of the mass-spring system in a vacuum equals the operational
frequency. (4) The maximum pressure on-axis at the operational frequency is 200 kPa.
Given these requirements, determine vact and the force exerted by the fluid on the
piston.
vpiston vact
water
2a M
K
Exercise 8.17 The sketch depicts a piston transducer that is flush mounted in an
infinite baffle. The piston, whose mass is M, is restrained by spring K. The displace-
ment at the other end of the spring is controlled by an actuator that can exert whatever
force is required to impose a specified harmonic motion Re (Vact exp (iωt)) . The task
is to determine radiation properties as a function of the frequency. Such an analysis is
initiated by deriving a differential equation of motion for vpiston that accounts for the
fluid loading acting
on the piston.
Solution of this equation for the specified actuator
motion will yield Vpiston /Vact as a function of ω. From that result evaluate the time-

averaged radiated power Pav / ρ0 c |Vact |2 . Graph both quantities as functions of ω
in the range from zero to 20% greater than the natural frequency of the mass-spring
system in a vacuum. Carry out the analysis for the case where the fluid is water, then
repeat the computation for the case where the
fluid
is air. The transducer’s properties
are a = 150 mm, M = 0.2 kg, and K = 45 106 N/m.
Exercise 8.18 A piston, whose radius is 250 mm, is embedded in an infinite rigid
baffle. The fluid is air. The piston was at rest until t = 0, at which
instant it begins
a parabolic pulse, in which the surface velocity is vs = 0.3 t/T − t 2 /T 2 m/s if
0 < t < T . Evaluate the waveform at the on-axis positions z = 0, 4a, and 32a in
the case where T = 0.25a/c. Compare each to the limiting form in Eq. (8.6.10). For
each location, determine the time required for the signal to attain a steady state.
Exercise 8.19 The waveform plotted below is the normal velocity of a circular
piston in an infinite baffle. It consists of a single cycle of a sawtooth. Evaluate and
plot p/ (ρ0 cv0 ) as a function of the nondimensional retarded time (c/a) (t − z/c)
for on-axis locations. Perform the calculation for z = a and z = 4a, with the pulse
interval set at T = 0.3a/c and T = 3a/c.
vz
0
t
T
-v0
Chapter 9
Modal Analysis of Waveguides
We return to situations where boundaries guide the direction in which sound may
propagate. The difference from the one-dimensional waveguides considered in
Chaps. 2 and 3 is that some feature will vary transversely to the propagation direction.
It might be that the excitation varies in the transverse direction, which is analogous to
the effect of a spherical source that is not radially symmetric. This situation commonly
is encountered in turbine engines and ocean acoustics. Another cause for transverse
variation of a propagating signal is a nonconstant spacing between the confining
boundaries. Examples of this situation are musical horns and some speaker systems.
Both features often are encountered in HVAC systems. Fully realistic models that
account for all effects are difficult to analyze, so they are typically addressed with
computer methods. However, if we make some idealizations, there are a number of
configurations that are amenable to analysis. The studies that follow provide much
understanding of basic phenomena, and all are descriptive of situations that one might
encounter.
9.1 Propagation in a Horn
Unlike the one-dimensional waveguides considered previously, a horn has a cross

section that is not constant along the axis of propagation. For example, conical
horns, such as those commonly used by the coxswain in shell rowboat racing, have a
cross-sectional radius that is proportional to the axial distance x. A configuration of
particular interest is an exponential horn, in which the cross-sectional area depends
exponentially on x. In principle, the shape of the cross section could change, but
such configurations are rare. In any event, the actual shape of the cross section will
be irrelevant to the analysis.
The idea of attaching a horn to a small source is to attain some features of an
unadorned large source. One such feature is directivity. We know that a source whose
DOI 10.1007/978-3-319-56847-8_9
188 9 Modal Analysis of Waveguides
size is small relative to a wavelength will radiate almost omnidirectionally. The large
area of a horn at its open end has the effect of increasing the size of the radiator. This
aspect may be recognized by considering a conical horn. In Fig. 9.1a, a spherical
cap in a baffle executes a purely radial vibration. A rigid cone is mounted on the
perimeter of the cap, with the apex of the cone coincident with the center of the
sphere. In Fig. 9.1b, the cone is positioned on a freely standing sphere of the same
radius that executes the same radially symmetric vibration everywhere on its surface.
The particle velocity in the signal radiated by the sphere in case (b) is purely radial.
This is compatible with the rigidity of the cone, which requires that the particle
velocity normal to its surface be zero. It follows that the field inside the cone is the
same as what it would be if the cone were not present. (This assertion assumes that
there is no reflection from the cone’s open end.) Furthermore, what occurs outside
the cone is irrelevant to the field inside the cone. This leads to recognition that the
field in case (a) is a radially symmetric spherical wave within the cone. Thus, the
presence of the conical horn in case (a) changes the effective size of the radiator
from the diameter 2a0 of the cap to the diameter 2a1 of the opening. It also makes it
unnecessary to mount the cap in the baffle.
(a) (b)
2a0 2a1
Fig. 9.1 A conical horn mounted on a spherical transducer. In case a, the transducer is a cap in a
baffle, whereas in case b, it is in free space
Another reason to mount a horn on a source pertains to applications in which it

is desired to have a constant output over a broad frequency range. Without the horn,
attainment of this objective would require either a small transducer whose velocity
amplitude increases inversely to the frequency, or else a very large transducer. Neither
feature is desirable for reasons that will be disclosed when we see why a horn avoids
these issues.
9.1.1 The Webster Horn Equation
The model to be constructed is based on a fundamental assumption: The pressure

perturbation depends only on the axial position x and t. A corollary that follows
from Euler’s equation is that the particle velocity is solely in the axial direction,
with vx also being a function of x and t. Clearly, the latter cannot be exactly correct,
because if it were there would be no signal in the region into which the horn expands.
9.1 Propagation in a Horn 189
Intuitively, it is reasonable to expect that this assumption is valid if the cross-sectional

area is a function A (x) that varies slowly relative to the scale of a wavelength. The
correctness of this assumption is an item we will explore.
The field equation is obtained by enforcing conservation of mass and the
momentum-impulse principle. These considerations parallel those that led to the
one-dimensional wave equation in Chap. 2. However, rather than using finite-sized
control volumes and finite increments in time, here we shall take a less rigorous, but
more expedient, approach that considers the rates of change in the state of differen-
tial control volumes. We begin with conservation of mass for the stationary control
volume in Fig. 9.2.
Fig. 9.2 A differential

stationary control volume A(x) A(x+ dx)
used to describe conservation
of mass in a horn
(x, t)vx(x, t) (x+ dx, t)vx(x+ dx, t)
x
x+ dx
The wall of the horn forms the sides of this region. It is impermeable, so there
is no mass flow across the sides of the control volume. Thus, the net rate at which
mass flows into the control volume over the two cross sections must equal the rate at
which the mass of this control volume increases. For the cross section at x, positive
vx transports mass into the control volume, and ρvx evaluated at x is the mass flow
rate per unit area. Positive vx on the face at x + d x transports mass out of the control
volume, and ρvx evaluated at x + d x is the rate per unit area there. According to the
central limit theorem, ρ (x, t) A(x)d x is the mass of the control volume at time t.
Thus, it must be that
∂
ρ(x, t)vx (x, t)A(x) − ρ(x + d x, t)vx (x + d x, t)A(x + d x) = [ρ(x, t)A(x)d x]
∂t
(9.1.1)
Division by d x converts the left side of the preceding to a partial derivative of ρvx A,
so the expression becomes
∂ρ ∂
A + (ρvx A) = 0 (9.1.2)
∂t ∂x
The second term is nonlinear because it is the product of two state variables.
A linearized analysis restricts the particle velocity to being a small quantity (relative
to the speed of sound), and the density is restricted to being little changed from the
ambient value ρ0 . Correspondingly, we linearize the continuity equation by replacing
ρvx with ρ0 vx . For a homogeneous medium, ρ0 is the same at all locations, so the
linearized continuity equation for a horn is
∂ρ ∂
A + ρ0 (Avx ) = 0 (9.1.3)
∂t ∂x
If the cross section is constant, this relation reduces to the continuity equation for
truly one-dimensional plane waves.
The momentum equation is derived by considering a control volume that moves
with the particles. Because it tracks a specific set of particles, its selection simplifies
application of the momentum-impulse principle for a system of particles. The control
volume in Fig. 9.3a consists of the fluid contained between cross sections at x and
x + d x at time t. The forces depicted there consist of the pressure resultant acting
on each cross section and the distributed force exerted by the wall of the horn in
reaction to the pressure the fluid exerts in the wall. It would be awkward to account
for the axial component of the wall force.
(a) p(x, t)
(b) p(x, t)
A(x) A(x+ dx) A(x)
A(x)p(x, t) A(x+ dx)p(x+ dx, t) A(x)p(x, t) A(x)p(x+ dx, t)
p(x, t) p(x, t)
x x
x+ dx x+ dx
Fig. 9.3 Control volume used to form the derivative form of the momentum-impulse principle.
The domain moves in unison with the particles contained between two adjacent cross sections.
a The domain consists of all particles contained between the adjacent cross sections. b The domain
is a cylinder whose cross section is constant
The control volume in Fig. 9.3b is more amenable to the analysis. It is a general
cylinder whose cross section is A (x). This shape exploits the fact that shear stresses
are negligible for an ideal fluid. Thus, the force system exerted on the sides of this
cylindrical domain is the stress resultant of the pressure exerted by the adjacent fluid.
This resultant acts perpendicularly to the cylindrical surface, so it has no component
in the axial direction. (The transverse resultant actually must be zero for the assumed
motion, because there is no acceleration in that direction.)
There is no evidence of a variable cross section in the second force diagram, as in
Fig. 9.3b. Thus, the time-domain Euler equation, which is obtained from conservation
of momentum, may be applied to the motion in the x direction
d ∂p
ρ v (x, t) = − (9.1.4)
dx ∂x
A linearized equation of motion results from dropping any terms that are products of
small variables. Thus, the acceleration is approximated as a partial time derivative,
and the current density is approximated by its ambient value. Doing so reduces the
equation of motion for the control volume to
∂vx ∂p
ρ0 =− (9.1.5)
∂t ∂x
This is the same as the x component of Euler’s equation.

Equations (9.1.3) and (9.1.5) constitute two equations for three unknowns: p, ρ,
and vx . The third relation is the equation of state, whose linearized form is p =
c2 (ρ − ρ0 ). We use this equation to eliminate ρ from the continuity equation. To
eliminate vx , we multiply Eq. (9.1.5) by A and then differentiate it with respect to x.
Because A is independent of t, the result may be written as

∂2 ∂ ∂p
ρ0 (Avx ) = − A (9.1.6)
∂x∂t ∂x ∂x
Differentiation of Eq. (9.1.3) with respect to t yields the same vx term,
A ∂2 p ∂2
+ ρ0 (Avx ) = 0 (9.1.7)
c2 ∂t 2 ∂x∂t
Elimination of the mixed derivative term in these two equations yields the Webster
horn equation

1 ∂ ∂p 1 ∂2 p
A − 2 2 =0 (9.1.8)
A ∂x ∂x c ∂t
This partial differential equation has variable coefficients except in the case of a
uniform bar. In the frequency domain, it becomes an ordinary differential equation,

d2 P 1 dA dP
p = Re P (x) eiωt =⇒ + + k2 P = 0 (9.1.9)
dx2 A dx dx
This too is referred to as the Webster horn equation. There are horn shapes A (x)
for which it is possible to obtain an analytical solution in the frequency domain. We
also will develop an approximate solution for arbitrary A (x). However, before we
pursue these tasks, let us return to the conical horn because doing so will allow us to
identify when the Webster horn equation may be used.
Figure 9.4 depicts a conical waveguide whose semi-vertex angle is β. An arbitrary
point is located by its cylindrical coordinates (x, R) and its spherical coordinates
Fig. 9.4 Cylindrical and

spherical coordinates of a
field point within a conical
horn
r
 R
(r, ψ), with both coordinate systems defined relative to the apex of the cone. The
spherical cap at the left executes a uniform radial vibration at frequency ω.
From the discussion of Fig. 9.1, we know that the field is a radially symmetric
wave, so an outgoing wave is

D −ikr iωτ
ptrue = Re e e (9.1.10)
r
An analytical solution of the Webster horn equation for this case is available. Let us
compare it to the spherical wave solution in order to grasp the effect of the approx-
imations embedded in that equation. The cross-sectional radius of the waveguide is
x tan β, which leads to A (x) = π (tan β)2 x 2 . The Webster horn equation in this case
is
∂2 p 2 ∂ p ∂2 p
+ − 2 =0 (9.1.11)
∂x 2 x ∂x ∂t
If, instead of x, the independent variable were r , this would be the wave equation for
a radially symmetric spherical wave. Thus, if we ignore reflections from the far end
of the horn, the solution of the Webster horn equation for an outgoing wave is

D −ikx iωt
phorn = Re e e (9.1.12)
x
The radial distance to a point on a specific cross section is r = x/ cos ψ. Setting

D = D makes this solution and the spherical coordinate solution, Eq. (9.1.10), the
same on the axis, ψ = 0.
With increasing distance from the axis, the solutions diverge. The relative error in
the amplitude at a specific ψ is obtained by replacing r in Eq. (9.1.10) with x/ cos ψ,
which leads to
|Phorn | − |Ptrue |
εamplitude ≡ = cos ψ − 1 (9.1.13)
|Ptrue |
Another aspect is the phase error, which must be small if the waveform is to be
positioned correctly. This quantity is

1
εphase = kx − kr = kx 1 − (9.1.14)
cos ψ
Both errors are underestimates that increase monotonically with increasing ψ. The
largest errors at fixed x occur at the wall, where ψ = β. A trigonometric identity is
1 − cos β ≡ (sin β) tan(β/2). The resulting error estimates are
β β
εamplitude = − sin β tan , εphase = −kx tan β tan (9.1.15)
2 2
The amplitude error will be small if β 1, that is, if the cross section expands
slowly. However, even if the amplitude error is acceptable, the fact that εphase =
(kx/ cos β) εamplitude means that the amplitude and phase error will have comparable
magnitude only if kx is not large compared to unity. In other words, β should be
much less than 1 rad and x should not be much larger than a wavelength. Let us
assess these criteria for the audible range of a horn in air. Consider β = 10◦ . The
amplitude error at ψ = β is 1.5%. At the low end of the audible spectrum, 20 Hz,
the phase error will be less than 10% if x < 17.5 m. In contrast, at 10 kHz, the phase
error will be less than 10% if x < 35 mm. The low-frequency case suggests that an
English horn or euphonium might be well described by the Webster horn equation,
but this model might not be adequate for a flute or piccolo.
These criteria may be generalized to any shape horn. Let Rwall (x) be the trans-
verse distance to the wall at a specified axial position. For a cone, we have
d Rwall /d x = tan β, so small β corresponds to small values of d Rwall /d x and
tan (β/2) ≈ (1/2) d Rwall /d x. Thus, the amplitude and phase errors will not be
excessive if d Rwall /d x 1 and kx is not much larger than a wavelength.
9.1.2 Exponential Horn
Let us examine whether it is possible for a simple plane wave to exist within a horn.
Let κ be the constant wavenumber. Then, substitution of
P = Be−iκx (9.1.16)
into the frequency-domain Webster horn equation requires that
2 dA
k − κ2 A − iκ =0 (9.1.17)
dx
The wavenumber of a simple plane wave is a real quantity. Because A is real, it is

evident that in no circumstance, other than constant A, is it possible for a simple
wave to exist in a horn.
In the special case where dA/d x is proportional to A, that is, dA/d x = bA, the
area A factors out of Eq. (9.1.17). What remains is a complex quadratic equation
for κ. The cross-sectional variation fitting this description constitutes an exponential
horn
A (x) = A0 ebx (9.1.18)
The growth factor b may be negative, corresponding to a cross section whose radius
decreases with increasing x. In most cases, the cross section is either rectangular or
circular, with the shape held constant at all x, but these details are irrelevant to the
analysis.
When the cross-sectional area varies exponentially, Eq. (9.1.17) leads to a
quadratic equation for the wavenumber, specifically
κ2 + ibκ − k 2 = 0 (9.1.19)
If k > |b| /2, there are two roots complex roots. The one whose real part is positive
corresponds to propagation in the positive x direction. It is
1/2
b2 b |b|
κ = k2 − − i if k > (9.1.20)
4 2 2
The other root corresponds to waves that propagate in the direction of decreasing x.
It is 1/2
b2 b |b|
κ =− k − 2
− i if k > (9.1.21)
4 2 2
Both waves will exist if there are reflections at one end. The imaginary part of
both κ and κ correspond to waves that decrease with increasing x if b is positive.
As was noted above, a horn whose cross section decreases with increasing x may
be represented by a negative b. The alternative is to reverse the x-axis, so that it is
oriented in the direction of increasing cross section. In either view, the amplitude
decreases for a wave that propagates in the sense of increasing area, and it increases
for a wave that propagates in the sense of decreasing area. The general solution when
k > |b| /2 is

P = e−bx/2 B1 e−i (k −b /4)x + B2 ei (k −b /4)x if k > |b| /2
2 2 2 2
(9.1.22)
It is possible that k is smaller than b/2, in which case the square root yields an
imaginary value. Both wave numbers are negative imaginary values
1/2
b2 b |b|
κ = −i ± −k 2
+ if k < (9.1.23)
4 2 2
In this situation the waves that travel back and forth are evanescent waves. Neither
solution corresponds to a wave that propagates. Rather, both are standing waves in
which |P| decreases on the sense of increasing values of bx. The evanescent field
may be written as

P = e−bx/2 B1 e+(b /4−k )x + B2 e−(b /4−k x )
2 2 2 2
(9.1.24)
The case where k = |b| /2 marks a transition from a propagating wave to evanes-
cence. This condition marks the cutoff frequency,
ωcutoff = bc/2 (9.1.25)
Below the cutoff frequency, a waveguide is seldom of interest because the time-
averaged power transported through the horn is essentially zero. (Power flow will be
examined shortly.) The parameter κ is the complex wavenumber. Its real part gives
the wavelength. It also defines the phase speed, according to
ω c
cphase ≡ = 1/2 (9.1.26)
|Re (κ)| 1 − b /4k 2
2
Because the phase speed is a function of frequency, a signal consisting of more

than one frequency will spread out, or disperse, as it propagates. For this reason, a
relation like Eq. (9.1.26) is said to be a dispersion equation. However, the nature of
other systems is such that it might not be possible to obtain a closed form solution
for Re (κ) as a function of frequency. Thus, any equation, such as Eq. (9.1.20), that
relates the complex wavenumber to the frequency will be said to be a dispersion
equation.
In terms of dimensionless quantities, cphase /c for an exponential horn depends
only on k/b, which is the relation depicted in Fig. 9.5. At high frequencies, the phase
speed is close to the plane wave speed. Decreasing the frequency causes cphase to
increase, until it becomes infinite at the cutoff frequency. Below k = b/2, the phase
speed is undefined because κ is imaginary.
3
c phase / c
2
1
0
0 0.5 1 1.5 2 2.5 3
Nondimensional frequency  /bc
Fig. 9.5 Dispersion curve for an exponential horn

The cutoff effect results from the fact that |κ| = k for any |b| < 2k. Consider
a variety of waveguides at a specific frequency. As b increases from a small value
with k fixed, it is necessary that Im (κ) increase in order to match b. This requires
that Re (κ) decrease. At the cutoff condition, b = 2k, the imaginary part equals k,
so Re (κ) = 0.
The simpler alternative when the larger end is excited is to let b be negative, rather
than reversing the sense of x. Doing so allows us to set x = 0 as the driven end in
any case. In the absence of reflections at the undriven end, setting b < 0 leads to
|P (x = L) /P (x = 0)| = e−bL > 1. The signal at the exit would be substantially
enhanced if −bL 1. Such a configuration was used for antique hearing aids, as
well as in the apparatus that recorded sound on a phonograph platter. A horn amplified
the signal that actuated the encoding stylus.
The usual situation is that the particle velocity at the end x = 0 is specified. Let us
assume that the far end is terminated in a manner that prevents reflection. Accordingly,
we set B2 = 0 in Eq. (9.1.22). A general expression for the velocity may be obtained
by substituting the pressure ansatz, Eq. (9.1.16), into Euler’s equation. Doing so
gives
κ κ P
Vx = Be−iκx ≡ (9.1.27)
ωρ0 k ρ0 c
The expressions for P(x) and V (x) are general, but the behavior is fundamentally
different above and below the cutoff frequency. Above cutoff, κ is complex. This
means that there is a frequency-dependent phase difference between the velocity
and the pressure. Below the cutoff frequency, κ is imaginary. This corresponds to
frequency-independent 90◦ phase difference between the particle velocity and pres-
sure.
Satisfying the boundary condition will set the coefficient B. We equate Vx at
x = 0 from the preceding to V0 . The value of B is set by matching the particle
velocity according to this relation to the complex amplitude V0 of the velocity at that
end. The result is
k
P (x) = ρ0 cV0 e−iκx (9.1.28)
κ
The development began with two reasons horns are used presently. If the horn is
sufficiently long, the open aperture will be much larger than the size of the opening
where the horn is excited, thereby enhancing directivity of the signal that radiates
from that end. However, the signal at the aperture is much weaker due to spreading
loss, so the advantages of a horn regarding radiated power are not evident. An analysis
of power flow begins with the observation that there is no variation over a cross
section. Consequently, the power flowing across any cross section is the product of
the intensity and the area
1 iωt
P= A Pe + P ∗ e−iωt Vx eiωt + c.c.
4
(9.1.29)
1
= A Re P Vx e2iωt + P ∗ Vx
2
This expression is applicable to any system that fits the assumptions embedded
in the Webster horn equation. In the special case of an exponential horn, substitution
of P (x) and A (x) into this expression gives

1 2 k 2iωt (−2iκ+b)x 2 k −i(κ+κ∗ −b)x
P = A0 ρ0 c Re (V0 ) e e + |V0 | e (9.1.30)
2 κ κ
The complex wavenumber is given in Eq. (9.1.20). If ω < ωcut , then κ is negative
imaginary, which leads to

1 k
A0 ρ0 c Re (V0 )2 e2iωt e−(b −4k ) x if k < b/2
2 2 1/2
P= (9.1.31)
2 κ
Thus, for frequencies below cutoff, the time-averaged power flow is zero, and the
oscillatory part of the power flow decays exponentially with increasing distance.
Above the cutoff frequency, κ is complex. A few manipulations show that the power
flow in this case is

1 k k
P = A0 ρ0 c Re (V0 )2 e2iω(t−x/cphase ) + |V0 |2 if k > b/2 (9.1.32)
2 κ κ
This relation indicates that above the cutoff frequency, the time-averaged power
flow is independent of the cross section’s location. This property is predicted by the
general theorem regarding power, which tells us that the average power that flows
through the waveguide must be the same at all cross sections because energy does not
flow out of the walls. The fluctuating part of the power flow propagates downstream
as a second harmonic at the phase speed of the pressure wave.
The invariance of the time-averaged power means that we only need know the
amount that is input at x = 0. This is true regardless of the shape of the horn. For
this reason, a quantity of interest for any type of horn is the ratio of P at x = 0 to
the particle velocity V0 at that end. This ratio is the throat impedance,
P|x=0
Z throat ≡ (9.1.33)
V0
If we know this quantity, we may evaluate the time-averaged power input to the horn
by the transducer according Eq. (9.1.29), which gives
1 1
P|z=0 = A0 Re Z rad (V0 )2 e2iωt + A0 Re (Z rad ) |Z |2 (9.1.34)
2 2
For the case of an exponential horn, Eq. (9.1.27) gives

⎧
⎪ 2 1/2
⎪
⎪ b b
⎪
⎪ ρ c 1 − 22
+i , ω > ωcut
k ⎨ 0 4k 2
Z throat = ρ0 c = 1/2 (9.1.35)
κ ⎪⎪ 2
⎪
⎪ ρ0 ci b b
⎪
⎩ −k 2
+ , ω < ωcut
4 2
To gain insight to the design issues pertaining to a horn, let us compare an expo-
nential horn to a hemispherical source in an infinite rigid baffle in the case of a
radially symmetric vibration. To make the comparison, we assign the same active
surface area to both devices, so A0 = 2πa 2 . The time-averaged power radiated by
the hemisphere is half the value in Eq. (6.3.16),
2 (ka)2
(Pav )hemi = πρ0 ca 2 V0 (9.1.36)
(ka)2 + 1
Both devices should have a relatively flat response. That is, we wish that Pav be
nearly independent of frequency over as wide a band as possible. The power radiated
by an exponential horn is described by Eq. (9.1.32). According to it, if k/κ is close to
one, then Pav is independent of k. The value of κ approaches k as k increases beyond
cutoff. Thus, we require that the frequency be substantially greater than the cutoff
value, which leads to
ω ωcutoff =⇒ (Pav )horn = 2πρ0 ca 2 |V0 |2 (9.1.37)
Reference to Fig. 9.5 shows that this criterion is met if ω is as little as twice the cutoff
value.
It is not immediately apparent whether the equivalent hemisphere should be large
or small relative to a wavelength, so we shall consider both possibilities. Equation
2
(9.1.36) gives Pav ≈ πρ0 ca 2 V0 (ka)2 if ka 1. Thus, producing a constant
power output over a frequency band with a small hemisphere requires that V0 ka
be constant. The amplitude of the surface displacement is V0 /ω, so sustaining
a constant power output with a small diameter requires that the displacement be
inversely proportional to the square of the frequency. It is quite challenging to create
a small transducer that is capable of increasingly large displacement as the frequency
decreases. Hence, let us consider the alternative in which the hemisphere is large
relative to a wavelength. If ka 1 at the lowest frequency in the band, then the
2
power output would be Pav ≈ πρ0 ca 2 V0 . This is a similar dependence to that of
a horn, so an invariant |V0 | will yield a constant power output. The difficulty with
this design is that it entails a large transducer. Suppose the minimum frequency of
interest is 100 Hz. The criterion ka 1 in air leads to a 3.4 m! Even if we raise
the low end tenfold to 1 kHz, the transducer would be enormous. Large transducers
are expensive to manufacture, but there is another fundamental difficulty. The surface
acceleration is proportional to ω |V0 |, so the upper end of the frequency range will
be required to undergo relatively high accelerations. Imparting a large acceleration
to a large device requires very strong mechanical drivers to induce the vibration, and
it also introduces large stresses. In contrast, the cutoff frequency of an exponential
horn only sets the exponential constant b. We are free to select the horn’s length and
throat area to fit other specifications, such as a desired sound pressure level at the
opening.
The analysis of an exponential horn has been based on the assumption that a back-
ward propagating wave is not present. This is only true if the reflection coefficient at
that end is zero. In most cases, the undriven end of the horn is open. The end correc-
tion for the open end of a constant cross-sectional waveguide, which approximated
the impedance at an open end by the radiation impedance of a piston, may be used
for a horn. The impedance at that end is given by Eq. (3.5.53) as ρ0 cχ (ka). This
value is very close to the fluid’s characteristic impedance, ρ0 c, if ka > 3, which
corresponds to the radius of the opening exceeding half the acoustic wavelength.
In this condition, the reflection coefficient is close to zero. The alternative case in
which ka < 3 requires inclusion of backward waves resulting from reflection at the
open end. (This observation may be applied to waveguides whose cross section is not
circular by taking the mean radius to be [A (L) /π]1/2 .) The next example describes
how a nonzero reflection coefficient at the opening affects the field within the horn.
EXAMPLE 9.1 A horn may be used to obtain a gradual transition between

constant cross-sectional waveguides having different sizes. This transitional
segment typically is conical to join circular cross sections or a truncated pyra-
mid for square ones. However, we will consider an exponential section. It is
desired to determine the relationship between the pressure and particle veloc-
ity at the ends x = 0 and x = L. This property is a key part of the algorithm
developed in Sect. 3.4.3 to analyze waveguide networks. The requisite quantity
is the two-port mobility matrix [D], which was defined such that

Vx (x = 0) P (x = 0)
= [D]
Vx (x = L) P (x = L)
Derive an expression for [D] in terms of the values of k, L, and b.
Significance
The analysis will explore how reflection phenomena are affected by a nonuniform
cross section. It has practical application for waveguide networks.
Solution
We cannot assume that waves propagate in one direction, because that property only
occurs if the open end is large and exposed to an open space. Thus, we must retain
both waves in Eq. (9.1.22). Manipulations are simpler if we use symbols to represent
the characteristic exponents, so we set
1/2 1/2
b2 b b2 b
κ1 = k −
2
− i, κ2 = k −
2
+ i
4 2 4 2
The general solution reduces to
P = B1 e−iκ1 x + B2 eiκ2 x
Note that this form is valid for all k values. The particle velocity is obtained from
Euler’s equation
κ1 κ2
ρ0 cVx = B1 e−iκ1 x − B2 eiκ2 x
k k
The pressure and velocity at the ends resulting from these expressions are
B1 + B2 = P0
κ1 κ2
B1 − B2 = ρ0 cV0
k −iκ L k
B1 e 1
+ B2 eiκ2 L = PL
κ1 κ2
B1 e−iκ1 L − B2 eiκ2 L = ρ0 cVL
k k
The process of eliminating B1 and B2 is facilitated by writing these equations in
matrix form
P0 B1 ρ0 cV0 B1
= Tp , = [Tv ]
PL B2 ρ0 cVL B2
where the transfer matrices are

1 1
Tp =

e−iκ1 L eiκ2 L
κ1 /k −κ2 /k
[Tv ] =
(κ1 /k) e−iκ1 L − (κ2 /k) eiκ2 L
The pressure equations are solved for the wave amplitudes, and the result is substi-
tuted into the velocity equations. These operations yield

B1 −1 P0
= Tp
B2 PL
ρ0 cV0 −1 P0
= [Tv ] T p
ρ0 cVL PL
Hence, the two-port impedance matrix is
1 −1
[D] = [Tv ] T p
ρ0 c
The network algorithm iscomputational,

so it would be adequate in most situations
to use numerical values of T p and [Tv ] corresponding to specified values of k, b,
and L. However, an algebraic evaluation will
enable us to examine the effect of the
backward-traveling wave. The inverse of T p is

−1 1 eiκ2 L −1
Tp =
eiκ2 L − e−iκ1 L −e−iκ1 L 1
In turn, this leads to

iκ L
1 κ1 e 2 + κ2 e−iκ1 L − (κ1 + κ2 )
[D] = −iκ L iκ L −iκ L
ρ0 ck eiκ2 L − e−iκ1 L (κ1 + κ2 ) e 1 e 2 − κ1 e 1 + κ2 e 2
iκ L
This expression is general. To specialize it to the case where k > b/2, we define
1/2
b b b2
κ1 = K − i, κ2 = K + i, K = k 2 −
2 2 4
Application of Euler’s identity then reduces the transfer matrix to
1
[D] =
ρ0 ck i sin (K L)

[K cos (K L) + (b/2) sin (K L)] −2K ebL/2
2K e−bL/2 [−K cos (K L) + (b/2) sin (K L)]
A constant cross section corresponds to b = 0, which gives K = k. In that

case, the above expression is identical to Eq. (3.4.14), which is the two-port mobility
for a waveguide whose cross section is uniform. An interesting possibility is that it
[D] is singular if K L is a multiple of π. This corresponds to a resonance, like that
encountered in an impedance tube.
The acoustic mobility matrix has other uses in addition to the stated purpose of
analyzing waveguide networks. It describes how the state variables at the ends must
be related for any set of end conditions. In the case of an isolated exponential horn,
either the complex pressure amplitude or the axial velocity amplitude will be known
at the end where the signal is generated. At the far end, a rigid termination requires
that V be zero. Another possibility is that the far end is open. In that case, we might
set P = 0, or it might be appropriate to use the open end impedance described by
Eq. (3.5.53). Another alternative is that the far end is terminated by a material whose
impedance is Z L . In any case, one state variable is unknown at each end. The equation
{V } = [D] {P} gives two equations for two pressures and two particle velocities.
We may solve those equations for the unknown variables in terms of the excitation
amplitude, P0 or V0 at x = 0 and P (x = L) . Evaluation of such solutions over a
frequency band gives a complex frequency response. Depending on the impedance
at the far end, there might be frequencies at which the pressure is greatly enhanced.
These are cavity resonances, which will be studied from a general framework in the
next chapter.
9.1.3 Group Velocity
The term “dispersive” refers to waves whose waveform changes shape with increasing
propagation distance. In such a wave, the phase speed of a sinusoidal wave will
depend on the frequency. The waves at the higher phase speed advance farther than
the slower waves, so the arrival times at a specific location differ. The fact that a
wave is dispersive has another important effect, which is that the speed at which it
transports energy also depends on frequency. The phase speed is merely the speed
at which the fact that there is a disturbance is passed to downstream locations. For
example, a specific value of the pressure at a certain phase passes from point to point
at the phase speed. A simple plane wave is nondispersive, so all features of such a
wave, including its energy content, propagate at the speed of sound. The situation is
different if a wave is dispersive. The group velocity cg is the speed at which energy
is transported. (The group velocity is a scalar, so it would be more appropriate to say
that it is the group speed, but that term is not used.)
The group velocity is important for a variety of waveguide systems, so we shall
begin by considering it in general terms. The analysis will lead to a general formula by
which the group velocity may be extracted from a dispersion equation. The analysis
will close by applying that formula to an exponential horn. As implied by the adjective
“group”, identification of cg requires that we consider waves at different frequencies.
These frequencies will be taken to be very close, so the associated wave numbers
also will be very close. Close frequencies lead to beating. The packet of waves in
one beat contains a specific amount of energy that can be tracked.
Two waves of equal amplitude at frequencies ω1 and ω2 propagate in the x direc-
tion. We assume that the dispersion equation is known; the function describing the
frequency dependence of the wavenumber is κ (ω) . Our only interest is how the
waves propagate, so it is irrelevant if the field properties are such that the amplitude
depends on position transverse to x. Hence, the signal we shall consider is

p (x.t) = Re B e−iκ(ω1 )x+iω1 t + e−iκ(ω2 )x+iω2 t (9.1.38)
The location at which the waves originated also is irrelevant to this investigation, so
B may be considered to be real.
Because there is little difference between the frequencies of the individual har-
monics, a waveform at any location exhibits a beating pattern as a function of t
at any instant. Furthermore, because κ (ω) is taken to be an analytical function, a
small difference in frequency leads to wavenumbers that are close. Therefore, spatial
profiles also show a beating pattern. The upper graph in Fig. 9.6 shows the spatial
profiles at two adjacent instants t1 and t1 + t, while the lower graph are waveforms
at two adjacent positions, x1 and x1 and x1 + x. The dotted lines are the envelope
function for each beat pattern. In a time interval t, features of a profile, such as
maxima, minima, and zeros, advance at the phase speed. Thus, these features in the
upper graph shift to the right by cphase t. The group velocity is the rate at which the
envelope of the beat moves, so the envelope in the upper graph shifts to the right
p(x,t1) p(x,t1+t)
−10 −5 0 5 10
x (meters)
p(x1,t) p(x1+x,t)
−40 −30 −20 −10 0 10 20 30 40

t (ms)
Fig. 9.6 Addition of two harmonic waves at slightly different frequencies creates a beating pattern
in the waveform and spatial profile
by cg t. If cphase > cg , the extrema will seem to run past the envelope’s leading
zero, to be replaced by extrema that enter the packet at its trailing zero. In the lower
graph, features of the waveform are delayed by the time required for the phase to
travel distance x, so this delay is x/cphase . The delay of the envelope is x/cg .
In either view, the total energy density (kinetic and potential) contained between the
envelope’s zeros is transported at the group velocity.
The first step in identifying the group velocity entails writing the pressure in a
form that makes the beating pattern explicit. To that end, the center and difference
frequencies are defined to be
1
ω0 = (ω2 + ω1 ) , ω = ω2 − ω1 (9.1.39)
2
This corresponds to ω1 = ω0 − ω/2 and ω2 = ω0 + ω/2. It is assumed that
ω ωav , so we may use a two-term Taylor series to approximate the wavenumber
at each frequency
ω ω
κ (ω1 ) ≈ κ0 − κ0 , κ (ω2 ) ≈ κ0 + κ0 (9.1.40)
2 2
where
dκ
κ0 ≡ κ (ω0 ) , κ0 = (9.1.41)
dω ω=ω0
Substitution of these representations into Eq. (9.1.38) eventually leads to

p = E (x, t) Re 2B e−iκ0 x+iω0 t (9.1.42)
The function E (x, t), which is an amplitude modulation function, is given by

ω
E (x, t) = cos t − κ0 x (9.1.43)
2
The complex factor in Eq. (9.1.42) represents a nominal constant amplitude wave
at the center frequency. This term propagates at the phase speed cphase = ω0 /κ0 .
The envelope of the wave is ±E(x, t). The frequency of E (x, t) is ω/2, which is
much slower than ω0 , and its wavenumber is κ0 ω/2, which is much smaller than
κ0 . Equation (9.1.43) indicates that all features of the envelope, including its zeros
and maxima, propagate in the direction of increasing x at speed 1/κ0 . Therefore, the
interval between adjacent zeros of E (x, t) constitutes a wave packet that contains
a fixed amount of energy. Because κ0 is the derivative of the κ (ω) dispersion curve
at the average frequency, it follows that the phase speed and group velocity at any
frequency are
ω 1
cphase = , cg = (9.1.44)
κ dκ/dω
In principle, cg could be negative, which would indicate that energy propagates

oppositely to the direction of propagation. Such an occurrence is exceptional.
The analysis has taken κ (ω) to be real. However, the function described by
Eq. (9.1.20) is complex. This often is the case, as is evidenced by the wavenum-
bers for plane waves in a dissipative fluid, Sect. 3.3. The real part of the wavenumber
is associated with the propagation, and the group velocity also is extracted from
Re (κ). Thus, the general relations are
ω
1 dω
cphase = Re , cg = Re ≡ Re (9.1.45)
κ dκ/dω dκ
The second form in the equation for cg is useful if it is easier to solve the dispersion
equation for the frequency corresponding to a specified wavenumber.
The preceding is generally applicable. In the case of an exponential horn, the
dispersion equation is Eq. (9.1.20). It describes κ as a function of the plane wave
number k, whereas Eq. (9.1.45) takes κ to be a function of ω. The chain rule for
differentiation leads to
⎧ 1/2
⎪ ⎪
⎨ c 1− 2
b2
if k >
b
1
cg = c Re = 4k 2 (9.1.46)
dκ/dk ⎪
⎪ b
⎩ 0 if k <
2
This expression tells us that the group velocity in an exponential horn decreases with
decreasing frequency up to cutoff. Below the cutoff frequency, the group velocity
is zero. The latter aspect could have been anticipated by the earlier analysis, which
showed that the average power flow is zero below cutoff.
A comparison of the preceding relation for group velocity and Eq. (9.1.26) for
the phase speed shows that
cg cphase = c2 (9.1.47)
It is not unusual in acoustics to encounter this relation, but it is not a general property.
The relation in a sense helps us to resolve any uneasiness regarding the phase speed
increasing without bound as ω decreases the cutoff value. We now see that increasing
cphase is accompanied by decreasing cg , until energy ceases to propagate at ωcutoff .
EXAMPLE 9.2 The velocity of a source at the throat of an exponential horn in

air (ρ0 = 1.2 kg/m3 and c = 340 m/s) is v0 = 0.004 [sin (500t) + cos (501t)]
m/s. The area growth constant is b = 0.6 m−1 , and reflections at the open end
are negligible. Determine the pressure waveforms at x = 0.60 m and 1.20 m.
Use them to determine the group velocity, and compare the result to the value
obtained from the dispersion relation for an exponential horn.
Significance
The notion here is that seeing how features of a propagating wave are correlated to
the group velocity will enhance intuitive understanding of these concepts.
Solution
The specified analysis calls for extraction of the group velocity directly from com-
puted waveforms, followed by a comparison of the result to the one obtained from the
dispersion equation. Thus, our first task is to construct the waveforms. The pressure
in an exponential horn is described by Eq. (9.1.22) for a single harmonic. To combine
two harmonics, we write the sum in time-domain form. Before we do so, we deter-
mine the amplitude coefficient for each harmonic. The pressure in each harmonic is
related to the source’s velocity by Eq. (9.1.35) for the throat impedance. The values
for frequencies of 500 and 501 rad/s are
Z throat (ω1 = 500) = 399.42 + 83.23i, Z throat (ω2 = 501) = 399.45 + 83.07i Rayl
The corresponding pressure amplitudes are Pn (x = 0) = Z throat Vn , where V1 =

0.004/i and V2 = 0.004 m/s. The exact occurrence of beats requires that |P2 | = |P1 |.
The amplitudes obtained here are sufficiently close to treat the data as a true beat.
The waveform must be sampled at a very high rate in order to determine accurately
the instants at which p = 0. The data set we shall use corresponds to a sampling
interval of 0.001 (2π/ωav ). The data set that is generated is

0.004 iω1 tn 0.002πn
p = Re Z throat (ω1 ) e + Z throat (ω1 ) (0.004) eiω1 tn , tn =
i ωav
Figure 1 shows the pressure waveform at x1 = 0.6 m. The existence of a beat is

evident, but the oscillation at the average frequency is too rapid to see the actual
waveform.
4
2
p (x1,t)
−2
−4
0 2 4 6 8 10 12
Time (seconds)
Figure 1.
Figure 2 zooms the waveforms at x1 = 0.6 m and x2 = 1.2 m in the interval

surrounding the first zero of the envelope. We can evaluate the group velocity by
dividing the distance x2 − x1 by the time required for a zero to travel between these
locations. Hence, the feature we seek is a zero of the pressure, but which one? The
answer emerges when we consider the first zero after t = 1.57 s at which p transitions
from positive to negative. Both waveforms are such that the smallest peak is to the
left and the smallest valley is to the right. This is what we would expect at the location
where the amplitude modulation function vanishes. The other zeros correspond to
oscillations at the mean frequency, so the amplitude of the oscillation will be larger
than it is in the interval where the modulation function is small.
0.01
0.005
−0.005
p(x1,t)
p(x2,t)
1.55 1.56 1.57 1.58 1.59 1.6

Time (seconds)
Figure 2.
A less intuitive way in which to identify the instant at which the envelope function
is zero is to examine the zeros of the signal. In the interval depicted in Fig. 2, we
see that p (x1 , t) = 0 at tn = 1.5501, 1.5564, 1.5627, 1.5690, 1.5730, 1.5753,
1.5815, 1.5878, 1.5941, and 1.6004 seconds. The intervals between adjacent zeros
are tn+1 − tn = 0.0063, 0.0063, 0.0063, 0.0040, 0.0023, 0.0062, 0.0063, 0.0063, and
0.0063. A regular interval, which is 0.0063 s for the present data, corresponds to the
zeros of the high-frequency oscillation at the average frequency. An extra zero due
to the envelope function zero breaks up such regularity. This condition is manifested
here by the subintervals of 0.0040 s followed by 0.0023 s. The instant they mark is
t1 = 1.5730 s. For greater precision, we interpolate between the instants at which
the pressure changes sign. The value extracted in this manner is t1 = 1.573015 s.
A similar analysis of the x2 data in the zoomed interval leads to identification of
t2 = 1.574817 s as the instant at which the envelope is zero. The group velocity is
thereby estimated as
x2 − x1
cg data ≈ = 330.0 m/s
t2 − t1
The analytical expression for the group velocity is Eq. (9.1.46). The value of k is
set by the average frequency, so we have
1/2
b2
cg = c 1 − 2 = 332.9 m/s
4k
k=ωav /c
To further calibrate the accuracy of this procedure, let us evaluate the phase veloc-
ity by a similar calculation. The envelope modulates the amplitude of a wave at the
center frequency that propagates at the phase speed. The best feature to track is a
zero near an instant when the envelope is largest because its location can be identi-
fied with little ambiguity. The one we shall use occurs slightly later than 4.5 s. The
values identified by scanning the data and interpolating across the zero crossing are
t1 = 4.500292 s and t2 = 4.502019 s. This gives cphase = (x2 − x1 ) /(t2 −t1 ) = 347.4
m/s, whereas Eq. (9.1.20) gives cphase = 347.3 m/s.
Our identification scheme performed well. It should be evident from this exercise
that experiments to extract phase and group velocity from waveform data measured
over a small spatial range require highly accurate and precise equipment. Accuracy
would be improved by using measurements at points that are farther apart. However,
if that interval exceeds a wavelength at the average frequency, we will not be able to
correlate zeros at two locations because of the almost periodic nature of the waveform.
9.1.4 WKB Solution for an Arbitrary Horn
There are several reasons to use a horn whose profile is neither conical or exponential.
For example, to smoothly join two cylindrical tubes, it would be necessary that the
transition segment be circular with a radius function Rwall (x) for which d Rwall /d x =
0 at the ends. Except for a few cases, A (x) will be such that the Webster horn equation
has no simple solution. A technique that has been developed is the WKB method.1 It is
an approximate analysis in the frequency domain based on reasonable assumptions.
1 Theinitials recognize G. Wentzel, H. Kramers, and L.N. Brillouin, who employed the method to
analyze the Shroedinger wave equation. A detailed development is provided by C.M. Bender and
S.A. Orszag, Advanced Mathematical Methods for Scientists and Engineers (1978) Chap. 10.
Our starting point is consideration of the propagation of a plane wave in the

direction of increasing x. In an ideal unbounded fluid, propagation introduces a
phase lag −kx to the pressure. Therefore, it is reasonable to anticipate that the phase
lag of the arbitrary plane wave will be negative and increase with increasing x. In
view of these considerations, it must be that the general time-domain representation
of the wave may be written as

p (x, t) = |P (x)| Re ei(ωt−(x) (9.1.48)
The phase lag for this representation is the real variable (x).
The phase speed, cphase , is obtained by equating the phase of p at (x, t) and
x + d x, t + d x/cphase . Therefore,

dx
ωt − (x) = ω t + − (x + d x)
cphase

(9.1.49)
ωd x d
= ωt + − (x) + (x) d x
cphase dx
Solution of this relation for the phase speed yields
ω ω
cphase = ≡ (9.1.50)
d d
[ (x)] arg (P (x))
dx dx
The local wavenumber therefore is
ω d
klocal = = arg (P (x)) (9.1.51)
cphase dx
These expressions are suitable for any plane wave propagating in a fixed direction.
One of the outcomes of the WKB method will be a description of the phase speed
and local wavenumber in terms of the frequency and the properties of the waveguide.
The WKB analysis of the Webster horn equation assumes that there is no reflection
at the far end, so only waves that propagate in the direction of increasing x are present.
The starting point is an ansatz suggested by Eq. (9.1.48), but lacking the distinction
between the phase and the magnitude of P. Specifically, we begin with

p (x, t) = Re P (x) eiωt , P (x) = Be−i(x) (9.1.52)
where B is constant. The function (x) is allowed to be complex. If we write it as

Re () + i Im (), then we have
|P (x)| = |B| eIm() , (x) = Re () − arg(B) (9.1.53)

Substitution of Eq. (9.1.52) into the Webster horn equation gives

2
d 2 d d
−i 2 − − i + k Be−i = 0
2
(9.1.54)
dx dx dx
where
1 dA
(x) = (9.1.55)
A dx
The exponential factor is not identically zero, so a nontrivial solution corresponds to
vanishing of the bracketed term
2
d 2 d d
i 2 + + i (x) − k2 = 0 (9.1.56)
dx dx dx
This is a nonlinear second-order ordinary differential equation. The frequency-

domain Webster horn equation is linear, so it might seem that we have made the
analysis more difficult. However, such thinking ignores the difference in the behavior
of the complex pressure P and phase variable . In the case of an exponential
waveguide, P varies rapidly over the scale of a wavelength, whereas = κx, which
is a much more gradual dependence on x. In the case of arbitrary A (x), we can find an
exponential area function Bebx that closely fits the actual A (x) over any small range
of x values. Within this range, the exponential horn solution for this approximating
b should be close to the solution for the true A (x). This observation suggests that
in the short x interval where the area functions match, it is reasonable to consider
= κx. A different exponential factor b will fit A(x) in an adjacent x interval. Thus,
the value of κ for this interval will be different. However, unless the cross-sectional
area undulates greatly over a very small scale, the adjacent value of κ should not
differ greatly from the previous value; that is, dκ/d x may be expected to be small.
In other words, we can anticipate that in general ≈ κ (x) x, with κ (x) being a
slowly varying function. Because d 2 /d x 2 = d 2 κ/d x 2 x + 2dκ/d x, and dκ/d x
is taken to be small in some sense, perhaps it is reasonable to consider d 2 /d x 2 to
be even smaller. This is the WKB approximation.
The approximation is implemented as an iterative procedure. The first iteration
results from setting d 2 /d x 2 to zero in Eq. (9.1.56). We solve that equation for
d/d x and integrate to find the first approximation (1) (x). A second approxima-
tion, (2) (x), may be obtained by using the first approximation to estimate d 2 /d x 2 .
The specific equations to be solved are
2
d(1) d(1)
+ i (x) − k2 = 0
dx dx
(2) 2 (9.1.57)
d d(2) d 2 (1)
+ i (x) − k 2 = −i
dx dx dx2
It is not difficult to prove that the equation governing the nth iterant resembles the
one for (2) . It is
2
d(n) d(n) d 2 (n−1)
+ i (x) − k −i
2
= 0; n = 1, 2, ... (9.1.58)
dx dx dx2
The result of solving this quadratic equation for d(n) /d x is a recurrence relation

1/2
(0) d(n) i 1 d 2 (n−1)
= 0; = − (x) ± k − (x) −
2 2
, n = 1, 2, ...
dx 2 4 dx2
(9.1.59)
We seek the wave that propagates in the positive x direction, which corresponds
to Re () > 0. Thus, the root for the positive sign is selected. A direct integration,
with (n) (0) = 0, leads to

1/2
i x x
1 d 2 (n−1)
(n) = − (η) dη + k 2 − (η)2 − dη (9.1.60)
2 0 0 4 dη 2
The definition in Eq. (9.1.55) may be written as = (d/d x) ln (A), which leads to

1/2
(n) A (x)1/2 x
1 d 2 (n−1)
= −i ln + k − (η)2 −
2
dη (9.1.61)
A (0)1/2 0 4 dη 2
Let N be the last iteration step. Because e− ln(u) ≡ 1/u, the complex pressure ampli-
tude resulting from the nth iteration is given by

2 (N −1) 1/2
A (0)1/2 x
1 d
P (x) = B exp −i k 2 − (η)2 − dη (9.1.62)
A (x)1/2 0 4 dη 2
An important aspect of the preceding expression is that the integrand generally

will be complex for N > 1, which is evidenced by Eq. (9.1.61). Consequently, |P|
and (x) in Eq. (9.1.53) can only be extracted after the integral has been evaluated.
The exception is N = 1. In addition, it is evident that any iteration beyond the first
will be quite complicated. It is for this reason that a WKB analysis seldom goes
beyond the first iteration. For N = 1, the integrand in Eq. (9.1.62) is real. The
corresponding description of the pressure amplitude and phase angle is

1/2
A (0)1/2 x
1
|P| = |B| , (x) = k 2 − (η)2 dη − arg (B) (9.1.63)
A (x)1/2 0 4
The fact that the pressure depends inversely on the square root of the area is consistent
with the requirement that the time-averaged power flowing through a cross section
be independent of x. Substitution of (x) into Eqs. (9.1.50) and (9.1.51) yields

1/2
1
klocal = k − (x)
2 2
4
ω (9.1.64)
cphase =
1/2
1
k − (x)
2 2
4
In practice, the integral in Eq. (9.1.62) will be quite challenging to evaluate ana-
lytically at any iteration, including the first. One exception is the exponential horn,
for which (x) = b. The first iteration gives
1/2
(0) (1) i b2
= 0 =⇒ = − bx + k −
2
x (9.1.65)
2 4
This gives d 2 (1) /d x 2 = 0, which means that (2) will be the same as (1), , and
so on. From this, we conclude that (1) = , but the preceding expression states
that (1) = κx, where κ is the complex wavenumber in Eq. (9.1.20). Hence, the
WKB solution is identically correct for an exponential horn. This is as it should be,
because the fact that = κx for an exponential horn means that d 2 /d x 2 = 0 is
not an approximation.
When the A function is such that it is not possible to carry out the integral ana-
lytically, it could be evaluated numerically. However, it is reasonable to ask why we
should employ the WKB method in that case, because an alternative is to use numer-
ical methods to solve the Webster horn equation. Doing so does not entail assuming
that d 2 /d x 2 is very small. The next example will examine the various facets of
both numerical approaches.
EXAMPLE 9.3 In order to explore the effect of a constriction in a

waveguide,
consider the case where the cross-sectional area is A (x) =
A0 ε + cos (πx/L)2 , which varies periodically over length L/2 between a
minimum of εA0 and a maximum of (1 + ε) A0 . At x = 0, the particle velocity
is vx = Re V0 eiωt , and the far end, which is x = 2L, is terminated with an
absorptive material that has a zero reflection coefficient. The derivation of the
Webster horn equation requires that the cross section changes gradually. This
condition is satisfied if L is much greater than the nominal wavelength 2π/k,
so set k L = 20π. Use the WKB method to determine the complex pressure
amplitude P (x), and the corresponding local wavenumber and phase speed
for the case where ε = 0.1. Restrict the analysis to the first iteration. Then,
assess the quality of the approximate solution by using numerical methods to
solve the Webster horn equation.
Significance
The WKB method is interesting for the insights derived from its solution. However,
it is very difficult to perform higher iterations, so its accuracy is uncertain. A direct
numerical solution of the Webster does not entail fundamental approximations, so

we will have the opportunity to decide whether the WKB method is worth pursuing.
Solution
The first iteration of the WKB solution is indicated by Eq. (9.1.62) to be

1/2
A (0)1/2 x
1
P (x) = B exp −i k 2 − (η)2 dη (1)
A (x)1/2 0 4
The value of B is set by matching Vx at x = 0 to V0 . This requires application

of Euler’s equation. Differentiation of Eq. (1) and application of Leibnitz’ rule for
differentiation of an integral leads to

d P 1 1 dA
= − BA (0) 1/2
d x x=0 2 A (x)3/2 d x x=0

A (0)1/2 2 1
−iB k − (0) 2
= −iωρ0 V0
A (x)1/2 4
By definition, (0) = (dA/d x) /A evaluated at x = 0, so the preceding yields
ωρ0 V0
B=
1/2
1 i
k 2 − (0)2 − (0)
4 2
The function (x) for the given area function is

1 dA −2 (π/L) ε sin (2πx/L)
(x) = = (2)
A dx ε + cos (πx/L)2
At x = 0, we have (0) = 0, so we have established that
B = ρ0 cV0 (3)
Substitution of Eq. (2) into (1) leads to an integral that is in the class of ellip-
tic integrals, which are tabulated functions. However, that recognition still requires
considerable sophistication to manipulate the integral to a standard form. Since the
primary purpose is to determine the accuracy of the WKB method, we shall abandon
the effort to evaluate the integral analytically. Instead, we will evaluate the integral
numerically.
It is best to use nondimensional variables for numerical studies, so let x = x̃ L
and P = ρ0 cV0 P̃. Then, the first WKB iteration, Eq. (1), becomes
Ã (0)1/2 ˜
P̃wkb = e−i (x̃) (4)
Ã (x̃)1/2
where the phase angle is

x̃
1/2
1
˜ (x̃) =
(k L)2 − ˜ (η)2 dη (5)
0 4
The nondimensional area properties are
A (x̃ L)
Ã (x̃)1/2 ≡ = ε + cos (π x̃)2
A0
−πε sin (2π x̃)
˜ (x̃) = L (x̃ L) =
1 + cos (π x̃)2
Evaluation of Eq. (5) is straightforward for a numerical routine such as MATLAB’s

quadl because the integrand has no singularities. Evaluation at a large number
of locations x̃ j may be done efficiently by exploiting the incremental nature of an
integral, which gives

1/2
x̃ j
1
˜ x̃ j−1 +
˜ x̃ j = ˜
(k L) − (η̃)
2 2
d η̂
x̃ j−1 4
Now let us turn to the direct application of numerical methods. In terms of the
nondimensional variables defined above, the Webster horn equation in the frequency
domain is
d 2 P̃ d P̃
2
+ ˜ (x̃) + (k L)2 P̃ = 0
d x̃ d x̃
Standard numerical software offers routines for solving first-order differential equa-
tions. To convert this second-order differential equation to first-order form, we define
two variables Y1 and Y2 , such that
P d P̃ L dP
Y1 = P̃ = , Y2 = = (6)
ρ0 cv0 d x̃ ρ0 cV0 d x
The derivative identity is dY1 /d x̃ = Y2 , and the derivative of Y2 is the second

derivative of P̃, whose value is given by the differential equation. Thus, the first-
order equations to solve are

d Y1 Y2
= (7)
d x̃ Y2 − (k L)2 Y1 − ˜ (x̃) Y2
Solution of this pair of order differential equations requires values of Y1 and Y2 at

x̃ = 0. Euler’s equation gives d P/d x = −iωρ0 V0 at x = 0. It follows from Eq. (6)
that
Y2 |x=0 = −ik L (8)
What should we use for the boundary value of Y1 ? If we do not set it correctly, we
will not initiate a wave that propagates solely in the direction of increasing x. To
resolve this dilemma, we recall the earlier observation that we may fit an exponential
shape to A(x) in a small interval. Let us do so for the region around x = 0. Then
we can use the equivalent b factor to set the relation between P and V at x = 0
as though the horn were exponential. For an exponential horn, = b everywhere,
whereas (0) = 0 in the present system. Therefore, the exponential constant should
be beq = 0. According to Eq. (9.1.27) for an exponential horn, the relation between
P and V at x = 0 is
⎧ ⎫
⎨ 2 1/2 ⎬
κeq P|x=0 beq L 1 beq L
V0 = = 1− − V0 P̃|x̃=0 (9)
k ρ0 c ⎩ kL 2 kL ⎭
Setting beq = 0 leads to

Y1 |x̃=0 = 1
Many techniques have been developed to solve a set of first-order differential

equations. A fourth-to-fifth-order Runge-Kutta routine, which is implemented as
ode45 in MATLAB, works well for the present configuration. It requires definition
of a function that evaluates the right side of the first-order equation. A program loop
increments x and calls ode45 with the function name as the first argument. The
value of Y1 at each output location xn is saved.
Errors sometimes accumulate in numerical analyses, so it is useful to have ways of
checking our work. One way of doing so is to replace the given ˜ (x̃) with a constant
value for an exponential horn. Comparison of the computed pressure to the analytical
solution will serve to verify that the program steps are correct. It also is good practice
to verify that reducing the x increment in the program loop and decreasing the error
tolerances do not significantly alter the computed result. Another check is to compute
the time-averaged power flowing through any cross section. From Eq. (6), we have
P = ρ0 cV0 Y1 and d P/d x = (ρ0 cV0 /L) Y2 . Then, Euler’s equation gives
1 dP V0
Vx = − =i Y2
iωρ0 d x kL
Correspondingly, the power flowing across any cross section is

∗
1 1 i
Pav = A (x̃) Re P Vx∗ = A (x̃) Re (ρ0 cV0 Y1 ) V0 Y2 (10)
2 2 kL
∗
1 ρ0 c |V0 | 2
Y1 Y2
= A (x̃) Re
2 kL i

At all locations,
the numerical solution gave Pav / ρ0 c |V0 |2 A0 = 0.5500 ±
4 10−13 .
Figure 1 displays the numerical results for Y1 below the A and functions. The
pressure rises where A is relatively small, but its features otherwise are unremarkable.
10
0 1 ~

P/(0cV0) A/A
0

~
A
0 -10
4
0
Re(P) Im(P)
−4
0 0.5 1 1.5 2
Axial distance x/L
Figure 1.
˜
To compare the WKB and direct numerical solutions, we examine |P| and .
Equation (4) gives
A (0)1/2

P̃WKB =
A (x)1/2
and Eq. (5) gives WKB (x). These quantities are obtained in the direct numerical
solution by post-processing of the Y1 data. By definition Y1 = P̃ (x). Conversion of
the Y1 data to polar form places in a 2π range. The consequence is that can change
discontinuously. This data may be converted to a continuous variable by invoking an
unwrap routine that is available in many software packages. Alternatively, jumps of
by 2π may be canceled manually by comparing the phase angle at each x location
to its predecessor at an adjacent location.
Figure 2 compares the amplitude and phase for the alternative methods. The latter
is plotted as φ/x, because that quantity would be the constant wavenumber of a
simple plane wave. Both analyses are in extraordinarily good agreement, except for
the phase near the source.
4
|P| Numerical
|P| WKB
2
0
0 0.5 1 1.5 2
70
60
/x Numerical
50 /x WKB
40
0 0.5 1 1.5 2
x/L
Figure 2.
It also is useful to compare the local wavenumber and phase speed. These quan-
tities for the first iteration of the WKB method are given by Eq. (9.1.64). To find
klocal for the differential equation solution, we use the general relation, Eq. (9.1.51),
which requires values of (d/d x) arg (P (x)) ≡ d/d x. We shall use a central finite
difference to approximate Eq. (9.1.51), so that

d (xn−1 ) − (xn+1 )
klocal (xn ) = − ≈
d x xn (xn+1 − xn−1 )
The phase speed in either case is cphase (xn ) = ω/klocal (xn ).

Comparisons of the wavenumber and phase speed are provided by Fig. 3. Here, the
agreement is less good. In particular, the differential equation solution indicates that
the highest phase speed occurs where the area is smallest. The first WKB iteration
misses this feature. However, the numerical solution also has a troubling aspect,
specifically, the small-scale fluctuation in klocal and cphase . It is not clear whether this
is an artifact or a true property. Examination of the data in the interval 0.5 < x̃ < 1.5
indicates that there are 20 ridges. The value of k L is 20π, which means that there
are 20 half-wavelengths in this interval. Furthermore, the curves were found to be
unaltered when the error tolerance and time step for the Runge-Kutta solver were
decreased, nor were they altered when a backward difference was used to evaluate
klocal .
63
klocalL/kL
klocal (Horn eq.)

62 klocal (WKB)
1.01 cphase (Horn eq.)

cphase/c
cphase (WKB)
1
0 0.5 1 1.5 2
Figure 3.
Which method is preferable is discretionary in most cases. Once one has pro-
grammed either solution, adaptation of that program to analyze other configurations
merely requires altering the definition of (x). The WKB first iteration is easier to
program and requires very little CPU time. However, the differential equation solu-
tion also is quite fast because it only requires solving two coupled equations. The
main drawback of the WKB method is that its numerical implementation is limited
to the first iteration unless we wish to evaluate the second derivative of an iteration’s
results. Thus, if extreme accuracy is required, especially for the phase speed, then
numerical solution of the Webster horn equation is preferable. However, in most
situations, we are only interested in the amplitude and phase angle of P, and we
can accept a fractional dB error. In such situations, the WKB method might be quite
acceptable.
9.2 Two-Dimensional Waveguides
We now turn our attention to situations where the pressure within a waveguide varies
transversely to the propagation. Our starting point is the simplest system, in which
the fluid is situated in the region bounded by two infinite planes. It is assumed that
9.2 Two-Dimensional Waveguides 217
the pressure is invariant in one direction parallel to these planes. Thus, the signal
propagates parallel to the boundary, and it varies perpendicularly to the boundaries.
Despite the apparent simplicity, the various configurations we shall examine feature
most of the phenomena encountered in realistic systems.
9.2.1 General Solution
The model system is the fluid contained between two parallel infinite planes. There
is no variation in one of the directions parallel to these planes. This is the two-
dimensional situation depicted in Fig. 9.7, where x measures distance in the direc-
tion of propagation and y measures distance from one wall. A harmonic excitation
applied along the plane at x = 0 induces a signal that propagates in the direction
of increasing x. Our first task is to ascertain the general nature of its dependence on
both x and y.
Fig. 9.7 A two-dimensional y

waveguide consisting of the
region between parallel x
infinite planes
H
We may shortcut the separation of variables formalism by using a complex expo-

nential e−iκx to represent the propagation, where the constant axial wavenumber κ
must be determined. Hence, we begin with
P = (y) e−iκx (9.2.1)
Note that the above expression is equally valid if κ is replaced to −κ, which would
correspond to waves traveling in the negative x direction. This ansatz for P will
satisfy the Helmholtz equation if
d 2
+ μ2 = 0, μ2 = k 2 − κ2 (9.2.2)
dy 2
The y-dependence is generally described by
= Be−iμy + Ceiμy (9.2.3)

This form is useful because it will lead us to a fundamental interpretation. However,

most individuals prefer trigonometric functions for analysis, so the form we shall use
= B cos (μy) + C sin (μy) (9.2.4)
The coefficients B and C are not the same as they were in Eq. (9.2.3), but this is of no
concern because there will be no occasion where both forms are used simultaneously.
Equation (9.2.4) constitutes the general solution. The unknown parameters in it
are B, C, and μ. These will be set by satisfaction of boundary conditions at y = 0
and y = H . Determination of μ will then lead to the axial wavenumber, which is
given by
1/2
κ = ± k 2 − μ2 (9.2.5)
The plus/minus sign allows us to select the proper branch cut in situations where
μ > k or μ is complex. In any case the sign of the square root must be taken
to satisfy the Sommerfeld radiation condition for propagation in the direction of
increasing x, which requires that Re (κ) ≥ 0 and Im (κ) ≤ 0.
9.2.2 Rigid Walls
For our initial study, we consider the walls to be rigid. Acoustic parlance is to say
that this is a hard-walled waveguide. The general solution, Eq. (9.2.4), will satisfy
the rigid wall boundary condition at y = 0 if

∂ P
= μC = 0 (9.2.6)
∂ y y=0
It is possible that μ = 0, but in that case, P is independent of y. This is the case of

a plane wave. Thus, we set C = 0. Then, the rigid condition at y = H requires that

∂ P
= −μB sin (μH ) = 0 (9.2.7)
∂ y y=H
It cannot be that B = 0, because that would set P to zero. Thus, it must be that
nπ
sin (μH ) = 0 =⇒ μn = , n = 0, 1, 2, ... (9.2.8)
H
The description of P in Eq. (9.2.1) at any n is a waveguide mode, and n is the
mode number. The modes that arise in structural and mechanical vibration typically
are standing waves in all directions, whereas the modes of a waveguide propagate in
one direction and form a stationary pattern transversely to the propagation direction.
The values of μn are the eigenvalues, and the equation whose roots give the eigenval-
ues is the characteristic equation. Somewhat confusingly, , which describes the

transverse variation of P corresponding to μn , is called a transverse mode function,
or more concisely, a transverse mode.
The only aspect that we have not found is the coefficient B. The solution for
P satisfies the Helmholtz equation and rigid wall conditions independently of the
value of B. Furthermore, any value assigned to it would be compensated when we
determine the actual signal generated by a specified excitation. We will soon define a
standard definition of B, but a common practice for graphing a mode is to set B such
that the maximum value of n is one. Accordingly, we shall temporarily leave this
coefficient in algebraic form, so the transverse modes of a hard-walled waveguide
are nπ y
n = Bn cos (9.2.9)
H
The axial wavenumber in the nth mode is

nπ 2 1/2
κn = k −
2
(9.2.10)
H
The pressure and particle velocity corresponding to a specific n are

nπ y
P = Bn cos e−iκn x , n = 0, 1, 2, ...
H
Bn κn nπ y nπ nπ y (9.2.11)
V̄ = cos ēx − i sin ē y e−iκn x
ρ0 c k H kH H
An important feature is the n = 0 mode, 0 = B0 , for which κ0 = k. Both P and V̄

are independent of y in this mode, and V̄ is in the x direction; it is a one-dimensional
plane wave.
As the mode number increases, the value of κn decreases, until n is sufficiently
large that the value of κn becomes imaginary. The largest n for which κn is real is
the cutoff mode number. For the case of rigid walls, it is

kH
Ncut = floor (9.2.12)
π
Although Eq. (9.2.10) for n > Ncut is satisfied if κn is either a positive or negative
imaginary value, only the negative value leads to decay. Specifically, we set
1/2
κn = −i μ2n − k 2 if n > Ncut (9.2.13)
In this case, the x dependence of P and V̄ is
e−iκn x = e−(μn −k ) x
2 2 1/2
(9.2.14)
These modes evanesce. The decay constant |κn | for n > Ncut increases with increas-
ing n when k is fixed. This means that the cutoff modes become less significant at a
specific x as the number of “wiggles” of the mode in the y direction increases. The
terminology we will use is to say that a propagating mode has an axial wavenumber
that is positive real (for propagation in the sense of increasing x). An evanescent
mode has an axial wavenumber that is negative imaginary. (These definitions will be
refined when we examine situations where the axial wavenumber is complex.) The
value of κn may be computed in one operation regardless of the frequency according
1/2 1/2
to κn = conj( k 2 − μ2n ) or κn = − conj( k 2 − μ2n ). Which alternative should
be selected depends on whether the software gives Im (κ) < 0 if κ2 < 0.
If we consider a specific mode, that is, hold n fixed, |κn | decreases as k decreases
from a large value. The cutoff frequency of a mode is the value below which that
mode becomes evanescent,
nπc
ωcut = (9.2.15)
H
Rather than viewing the behavior of a specific mode as the frequency decreases, we
may consider Ncut = floor(ω H/(πc)) as the highest mode number at which a mode
at a specific frequency will propagate. The n = 0 (plane wave) mode never is cutoff.
The phase speed of a propagating mode is
ω c
cphase = =
nπ 2 1/2 (9.2.16)
κn
1−
kH
The group velocity obtained from Eq. (9.2.10) is

nπ 2 1/2
1
cg n = c =c 1− (9.2.17)
dκn /dk kH
For frequencies below the cutoff, the axial wavenumber is imaginary, so the phase and
group velocities are zero. An interesting aspect is the similarity of these properties
to those of an exponential horn. In both systems
cg cphase = c2 (9.2.18)
The next section will explain this relation in terms of the trace velocity.
9.2.3 Interpretation
The description of a mode in Eq. (9.2.11) indicates that a constant value of P,

whose amplitude is proportional to cos (nπ y/H ), propagates in the x direction. This
amplitude is sustained as the signal moves downstream. Thus, it is appropriate to

say that a waveguide mode is a nonuniform plane wave. This means that the rays of
constant phase are parallel and that the amplitude is not constant along the wavefronts,
which are perpendicular to the rays.
An alternate picture explains several aspects, including the transition to evanescent
behavior. To obtain it, we employ Euler’s identity to convert the transverse mode
functions to complex exponentials
Bn (−inπ y/H )
n = e + e(inπ y/H ) (9.2.19)
2
This is the incarnation of the general solution, Eq. (9.2.3), for the rigid wall case.
When we use this representation to form the nth waveguide mode, we find that
Bn −i k̄1 ·x̄
Pn = e + e−i k̄2 ·x̄ (9.2.20)
2
where the position is x̄ = x ēx + y ē y . The wavenumber vectors are
nπ
k̄1 = κn ēx + ē y
H
(9.2.21)
nπ
k̄2 = κn ēx − ē y
H
aredirected at angle ψ above

These vectors are depicted in Fig. 9.8. The vectors
and below the x-axis. In view of Eq. (9.2.10), k̄1 = k̄2 = k. Consequently, each
wavenumber vector is the product of k and a unit vector, which is the way each is
depicted in the figure.
The construction in Fig. 9.8 leads to an alternate description of the wavenumber
vectors as
Fig. 9.8 Wavenumber y

vectors for the plane waves n
that are equivalent to mode n
in a two-dimensional
hard-walled waveguide ke-
1
n/H

x

ke-2 n/H
k̄1 = k ē1 , k̄2 = k ē2

ē1 = (cos ψn ) ēx +
(sin ψn ) ē y , ē2 = (cos ψn ) ēx − (sin ψn ) ē y
nπ κ nπ (9.2.22)
= sin−1
−1 n
ψn = tan = cos−1
κn H k kH
Substitution of this representation of the wave numbers into Eq. (9.2.20) leads to
Bn −ik ē j ·x̄
Pn = Pn(1) + Pn(2) , Pn( j) = e (9.2.23)
2
This representation describes a mode Pn (x, y) as a superposition of simple plane
waves whose phase speed is c that propagate obliquely in opposite sense relative
to the x-axis. The particle velocity is the superposition of the contribution of the
individual waves.
Bn −i k̄1 ·x̄
ρ0 c V̄ = ē1 e + ē2 e−i k̄2 ·x̄ (9.2.24)
2
The picture of the signal as superposed plane waves explains many features. The
transition to an evanescent mode is a consequence of the monotonic increase of ψn as
n increases at fixed k. Ultimately, if n > Ncut , the last of Eq. (9.2.22) states that ψn is
an angle whose sine is greater than one. This leads to a complex ψn , with cos ψn being
a negative imaginary value. The corresponding x components of k̄1 and k̄2 become
negative imaginary values, which converts the mode from one that propagates to one
that evanesces.
Decomposition of a propagating mode into simple plane waves gives a new per-
spective for the phase speed in Eq. (9.2.16) and the group velocity in Eq. (9.2.17).
For a propagating mode, cos ψn = κn /k = [1 − (nπ)2 / (k H )2 ]1/2 . Thus, these
propagation speeds may be written as
c
cphase = , cg = c cos ψn (9.2.25)
cos ψn
The phase speed is recognizable as the trace in the x direction of the phase velocity
of the oblique plane waves. In contrast, the group velocity is the component in the x
direction of the phase velocity of each oblique plane wave. Increasing ψn , either by
raising n or decreasing k, increases the trace velocity along the x-axis. Ultimately,
at the cutoff frequency, the phase speed is infinite because the wavefronts of the
simple waves are parallel to the x-axis, so all points on the x-axis simultaneously see
a specific wavefront. The explanation of the group velocity lies in the fact that the
oblique plane waves are nondispersive, so both their group and phase velocities are
c. Thus, a packet of oblique waves formed from two slightly different frequencies
would propagate obliquely at speed c. The projection onto the x-axis of the packet’s
location would move in the x direction at speed c cos ψn .
Representation of a mode as two plane waves makes is possible to view the
propagation in terms of rays. Figure 9.9a follows the ray of each plane wave that
passes through an arbitrary point C. The ē1 ray is incident at the upper wall, where it
Fig. 9.9 Multiple reflections C2

in a hard-walled
two-dimensional waveguide
observed by decomposing a
mode into its constituent
plane waves
e-2
e-1 D
C 
 e-1
e-2
C1
is reflected, thereby generating an ē2 ray. Because the boundary is rigid, the reflection
coefficient is R = 1, so the pressure in the reflected ē2 ray matches the pressure in
the incident ray. In other words, the reflected wave seems to have originated from the
image C2 of point C. A similar process applies to the ē2 ray that intersects point C,
so the pressure on the ē1 ray at point D seems to have originated from image point
C1 . In other words, the signal within a two-dimensional waveguide with rigid walls
is a process of repeated reflections.
The standing wave pattern in the transverse y direction results from the require-
ment that the field be consistent for reflection from either wall. From the viewpoint
of the lower wall ē2 is the propagation direction for the incident wave, and ē1 is
the propagation direction for the reflected wave. The viewpoint of the upper wall is
opposite, with the incident direction being ē1 and the reflected direction being ē2 .
The reflection coefficient at each wall is R = 1 because the walls are rigid. Suppose
the angle of incidence (and reflection) is an arbitrary value χ. Then, Eq. (5.2.16)
states that the pressure field due to reflection at the lower wall is

Bn i(ωt−kx sin ψ1 ) iky cos χ
p (x, y, t) = Re e e + e−iky cos χ
2

= Re Bn ei(ωt−kx sin χ) cos (ky cos χ) (9.2.26)
Reflection from the upper wall may be described similarly if we replace y with
y = H − y, so that y = 0 at the upper wall and y = H at the lower wall. Then,
the pressure field from this viewpoint is

p (x, y, t) = Re Bn ei(ωt−kx sin χ) cos (k (H − y) cos χ)

≡ Re Bn ei(ωt−kx sin χ) [cos (k H cos χ) cos (ky cos χ)
+ sin (k H cos χ) sin (ky cos χ)]} (9.2.27)
These alternate descriptions must represent the same pressure field, so it must be
that sin (k H cos χ) = 0. In other words, k H cos χ = nπ. The angle χ was defined to
be the angle of incidence, so the angle ψ for the decomposition into oblique waves
is π/2 − χ. It follows that
nπ
sin ψ = (9.2.28)
kH
This is identical to the third relation for ψn in Eq. (9.2.22). Thus, we have shown that
the waveguide mode is the superposition of a pair of waves that propagate obliquely to
the x-axis. The propagation angle for each oblique wave is such that the interference
patterns resulting from reflection at the lower and upper walls are compatible.
9.2.4 Flexible Walls
The model of a waveguide with rigid walls is a useful idealization. However, the
walls in some applications may be quite flexible, and they might be composed of an
absorptive material, as is the case for turbine engine exhaust systems. Some coastal
regions of the ocean may be considered to be a constant-depth waveguide with a
pressure-release upper wall (the surface) and a compliant lower wall (the ocean
bottom). In the situation to be addressed here, the planar boundary at y = 0 has
local impedance ρ0 cζ0 and the boundary at y = H has local impedance ρ0 cζ H ;
both parameters are considered to be the same at all locations on the respective
surfaces. The normal directions into the fluid are n̄ = ē y and n̄ = −ē y at y = 0 and
y = H , respectively. The local impedance condition is P = ρ0 cζ −n̄ · V̄ . We use
Euler’s equation to describe V̄ , which leads to the boundary conditions at the walls
being
ζ0 ∂ P
P = −ρ0 cζ0 Vy = at y = 0
ik ∂ y
(9.2.29)
ζH ∂ P
P = ρ0 cζ H Vy = − at y = H
ik ∂ y
The general solution for the pressure in any two-dimensional waveguide is given
by Eq. (9.2.4). We substitute this representation into the boundary conditions. The
requirement that the result be satisfied for all x allows cancelation of the e−iκx factor.
The result is a pair of linear algebraic equations for B and C,
ζ0 μC − ik B = 0
(9.2.30)
ζ H μ [B sin (μH ) − C cos (μH )] − ik [B cos (μH ) + C sin (μH )] = 0
The matrix form of these equations is

[D] [B C]T = [0 0]T

−ik ζ0 μ
[D] =
[ζ H μ sin (μH ) − ik cos (μH )] − [ζ H μ cos (μH ) + ik sin (μH )]
(9.2.31)
The coefficient equations are homogeneous, so their solution would give B =

C = 0. We seek a nontrivial solution, which can only exist of the equations are
rank-deficient. This condition is marked by the coefficient matrix having a zero
determinant. The condition that |[D]| = 0 constitutes the characteristic equation

|[D]| = − k 2 + ζ0 ζ H μ2 sin (μH ) + i (ζ0 + ζ H ) kμ cos (μH ) = 0 (9.2.32)
Multiplication of this equation by H 2 reveals that it contains only nondimensional

parameters: μH , k H , ζ0 , and ζ H . The frequency is specified, so the only unknown in
the characteristic equation is μH . The presence of sinusoidal terms whose argument
is μH tells us that there are an infinite number of nondimensional roots, whose values
we designate as ηn ; the corresponding eigenvalues are μn = ηn /H and n is the mode
number.
The preceding characteristic equation has a root at μ = 0, but that root is extra-
neous and must be disregarded if either ζ0 or ζ H is finite. This is so because μ = 0
corresponds to a simple plane wave. The particle velocity in a plane wave is solely
in the x direction. The pressure in such a wave would cause the wall to move in
the y direction, which is incompatible with the velocity field. The only exception is
the case where both walls are rigid. It is noteworthy that the plane wave mode does
not have a cutoff condition. The consequence is that only if both walls are rigid can
sound propagate at any frequency, including values that are nearly zero.
The B and C coefficients corresponding to an eigenvalue must satisfy Eq. (9.2.31).
Because this pair of equations are rank-deficient when μ equals any μn , only one
of the scalar equations is independent. This means that we only can solve those
equations for one coefficient as a proportionality to the other. We take Bn to be the
arbitrary coefficient, where “n” subscript allows for a value that varies from mode to
mode. When we replace μ with ηn /H in the first of Eq. (9.2.30) and solve for C/B,
we find that the waveguide mode is
P = n (y) e−iκn x
y y
n = Bn k H sin ηn − iζ0 ηn cos ηn
H H (9.2.33)

− k 2 H 2 + ζ0 ζ H ηn2 sin (ηn ) + i (ζ0 + ζ H ) k H ηn cos (ηn )
1/2
κn = ± k 2 − ηn2 /H 2
As was true for the case where both walls are rigid, the alternative sign for κn
corresponds to waves that propagate in the x direction in either sense. If either wall’s
impedance has a resistive part, the root ηn will be complex. In such case, care must be
taken to select the alternative sign. Stated differently, we must select the appropriate
branch cut for the square root. Let us focus on the wave that propagates in the sense
of increasing x, which means that we only are interested in ηn values that lead to
Re (κn ) ≥ 0. Furthermore, the radiation condition requires that the magnitude of the
pressure not grow in the direction of propagation, so it must be that Im (κn ) ≤ 0.
To identify the restriction this requirement imposes on the axial wavenumber,
let us write κn in polar form as κn = |κn | eiφn . Thus, we seek the conditions for
which a root μn will lead to −π/2 ≤ φn ≤ 0. From the definition of μn , we have
μ2n = k 2 − |κn |2 exp(2iφn ) ≡ k 2 − |κn |2 cos (2φ n2)− i |κn | sin (2φn ). If φn is in
2
the proper range, then −1 ≤ sin 2φn ≤ 0, so Im μn ≥ 0. There is no limitation to

|κn |, so the preceding tells us that μ2n is in either the first or second quadrant, that
is, μ2n = u exp(iθ) with u > 0 and 0 ≤ θ ≤ π. DeMoivre’s theorem gives two
values: μn = u 1/2 exp(iθ/2) and μn = −u 1/2 exp(iθ/2). The second root, being
the negative of the first, is accounted for as the second complex exponential in Eq.
(9.2.3). Therefore, only the first value is retained. Because θ must lie in the first or
second quadrant of the complex plane, it follows that the roots μn of the characteristic
equation that are consistent with the radiation condition lie in the first quadrant of
the complex plane. Roots that lie in other quadrants are extraneous.
A complex wavenumber κn leads to an oscillating function that decays as it
propagates in the x direction. To understand the nature of the transverse dependence
when κn is complex, we recall identities that convert sinusoidal functions having a
complex argument to hyperbolic functions whose argument is real. They give
sin (α + ιβ) ≡ sin (α) cosh (β) + i cos (α) sinh β

(9.2.34)
cos (α + ιβ) ≡ cos (α) cosh (β) − i sin (α) sinh β
If neither α nor β is negative, then both hyperbolic functions increase with increasing
β. Thus, setting α + iβ = μn y shows that n varies in an oscillatory manner with a
growing amplitude.
Cases where the eigenvalues are real are exceptional. One consists of the com-
bination of a rigid and a pressure-release wall, which we take respectively to be
y = 0 and y = H . The general characteristic equation, Eq. (9.2.32 ), is divided by
ζ0 to handle the infinite value of that impedance. Because ζ H = 0, what remains is
kμ cos (μH ) = 0. The cosine function must vanish because μ = 0 is extraneous.
Thus, the roots ηn are odd multiples of π/2. The modal properties are

π 2 1/2
2n − 1 π
μn = , n = 1, 2, ..., κn = k 2 − n 2
2 H H

(2n − 1) π y
n = Bn cos (9.2.35)
2 H

kH 1 2n − 1 πc
Ncut = floor + , ωcut =
π 2 2 H
As was true in the case where both walls are rigid, the variation in the y direction is
a real harmonic function, which means that the pressure at all points situated at the
same x oscillates in-phase.
Whereas a water channel might be approximated as having a lower rigid wall
and an upper pressure-release wall, a waveguide with two pressure-release walls
has no common incarnation. Such a system could be approximated by using lightly
tensioned thin mylar sheets for the walls. However, this model is primarily useful as
a limiting behavior. If ζ0 = ζ H = 0, then the general characteristic equation reduces
to sin (μH ) = 0. This is the same equation as that for a rigid-rigid waveguide, except
that μn cannot be zero here, so

2 1/2
π 2 π
μn = n , κn = k − n 2
, n = 1, 2, ...
H H
nπ y
n = Bn sin (9.2.36)
H

kH πc
Ncut = floor , ωcut = n
π H
The case where one or both walls are purely reactive is representative of some
common systems. For simplicity, let us take the wall at y = 0 to be rigid. The specific
impedance of the wall at y = H is taken to be ζ H = iσ, where σ is a real value
that may be positive or negative. The characteristic equation in this case is found by
dividing Eq. (9.2.32) by ζ0 and then substituting for ζ H . It is
− σμH sin (μH ) + k H cos (μH ) = 0 (9.2.37)
It is more difficult to extract the roots μn H = ηn of this equation, but once we do the
other features follow. The transverse mode function, axial wavenumber, and cutoff
parameters are described by
η y
n
n = Bb cos
H
1/2
ηn2
κn = k − 2
2 (9.2.38)
H
c
ηn < k H if n ≤ Ncut , ωcut = ηn
H
Due to the transcendental nature of the characteristic equation, numerical methods
are required to find the eigenvalues. Those methods work best when they are initiated
with a good estimate. It also is useful to have a general understanding of the nature
of the root relative to the stiffness and the frequency. Both objectives are met by
rewriting the characteristic equation as

kH 1
tan (μH ) = (9.2.39)
σ μH
The sole nondimensional parameter affecting the roots is the ratio of k H to σ.

The case where σ k H corresponds to either a very small σ, meaning a wall that
is very flexible, or else a very high frequency. The smaller roots μH = ηn will be
such that μH k H/σ, so the characteristic equation in this case is well approx-
imated as cot (μH ) = 0. The roots are the same as those of cos (μH ) = 0 for a
pressure-release wall at y = H . The opposite case is that in which σ k H , which
requires that the wall be quite stiff, as well as that the frequency not be very high.
In that case, the characteristic equation is approximately tan (μH ) = 0, and the
approximation improves with increasing mode number because of the presence of
μH in the denominator. The roots in this case are the same as those of sin (μH ) = 0
for a hard-walled waveguide.
These qualitative trends may be identified by a graphical construction of the
functions of μH on either side of Eq. (9.2.39). Figure 9.10 shows the plots for a
positive and a negative value of σ. The value of k H/ |σ| is 2.5π, which is not very
large. Nevertheless, the first root is close to the first singularity of tan (μH ), that
is, cos(μ1 H ) ≈ 0, which is the first root for a rigid-pressure-release configuration.
Increasing the value of μH brings (k H/σ) / (μH ) closer to the zero axis, so the
roots progressively become closer to the zeros of tan (μH ). Thus, they approach the
zeros of sin (μH ), which is the characteristic equation for a hard-walled waveguide.
Decreasing σ moves both hyperbolic curves farther from the zero axis. Nevertheless,
the higher roots always tend to approach the zeros of tan (μH ). In other words,
regardless of how flexible a wall is, the higher eigenvalues always tend to the values
for a hard-walled waveguide.
Fig. 9.10 Identification of the real eigenvalues as intersections of the functions forming the char-
acteristic equation for a two-dimensional waveguide bounded by a rigid wall and a locally reactive
wall, k H = π
An unexpected possibility is that an eigenvalue is purely imaginary, because an

imaginary ζ H can lead to a real characteristic equation. To explore this possibility,
we set μ = iβ. Identities give cos (μH ) ≡ cosh (β H ) and sin (μH ) ≡ i sinh (β H ).
A rearrangement of the terms in the characteristic equation gives

kH 1
tanh (β H ) = − (9.2.40)
σ βH
Only positive values of β are meaningful because of the earlier analysis, which
showed that μ must be in the first quadrant of the complex plane. Consequently,
tanh (β H ) > 0, which means that a real root for β H can exist only if the function
on the right side of the above equation is positive. In turn, this requires that σ < 0,
which is the case of a wall that is a pure compliance. The plot of both sides of this
equation in Fig. 9.11 confirms this interpretation. It also shows that if σ < 0. there
is only one root. We will designate this root as β H = ηim , so that μim = iηim /H . In
general, we will use a subscript “im” to denote this type of mode. As a generality,
this type of mode is not possible if a wall is an inertance.
2
kH/

H
1
tan(H), 
0

−1
−2
0 5 10 15
H
Fig. 9.11 Identification of the imaginary eigenvalue for a two-dimensional waveguide bounded by
a rigid wall and a locally reactive wall, k H = π
The transverse mode function corresponding to a real eigenvalue is given Eq.

(9.2.38). Substitution of ηn = iβim yields
y
im = B cosh ηim (9.2.41)
H
This function, like those for the real eigenvalues, is real. This means the pressure
in any mode oscillates either in-phase or 180◦ out-of-phase along any cross section
at fixed x. Fig. 9.12 shows a few transverse mode functions. Figure 9.11 indicates
that the value of the root ηim will be very large, unless k H/σ is very small. The
consequence of a large value of ηim is that im increases greatly in the vicinity of
the compliant wall. Another interesting aspect is that the axial wavenumber for this
1/2 2 1/2
mode is greater than k because κim = k 2 − μ2im = k + ηim 2
/H 2 . Thus, this
type of mode propagates much more slowly than the speed of sound.
Let us compare the results for cases where the wall impedance is infinite (rigid),
zero (pressure-release), or imaginary (reactive), to the general situation when the real
part of the impedance at either wall is nonzero. The former are idealized cases in
which dissipation is absent, whereas energy is absorbed in the wall when Re (ζ) > 0.
The presence of dissipation complicates the analysis by requiring the solution of a
complex characteristic equation. The eigenvalues μn are complex, which means that
0.4i  0.4i
1
0.5 n n
im
n
0 n
n
−0.5 n
n
−1
0 0.5 1 0 0.5 1
y/L y/L
Fig. 9.12 Mode functions for a two-dimensional waveguide bounded by a rigid wall and a compliant
wall, k H = π
the pressure as a function of y at fixed x shows phase differences. The axial wavenum-
ber κn also is complex, which means that a wave is attenuated as it propagates. We
would recognize the attenuation effect as a consequence of dissipation, even if we
did not perform an analysis. Phase differences are a more subtle effect of dissipation.
EXAMPLE 9.4 The lower wall of a two-dimensional waveguide is rigid, and

the upper wall is purely resistive with Z = 0.1ρ0 c. Determine the phase and
group velocities for the first three modes in the frequency range 0 < k H < 10.
Also, plot the transverse mode functions at k H = 0.1, 1.0, and 10.
Significance
A highlight is determination of the eigenvalues when the characteristic equation is
complex. Another new feature is a general approach for extracting group velocities
from data. The results will display an interesting phenomenon exhibited by the group
velocity.
Solution
Before we can solve the characteristic equation we must consider what is meant
by the specification “first three modes”. If the eigenvalues were real, we would
identify the mode number by sequencing μn in increasing order. A logical extension
for complex eigenvalues is to sequence them in ascending order of their real part,
Re (μ1 ) < Re (μ2 ) < .... The characteristic equation we shall solve is the limiting
form that results from multiplying Eq. (9.2.32) by H/ζ0 and then taking the limit as
ζ0 → ∞. The result is
f (μH ) = −ζ H (μH ) sin (μH ) + i (k H ) cos (μH ) = 0
where ζ H = 0.1 in the current case.

The standard methods for solving a single nonlinear equation f (x) = 0 when
x and f are scalars also are valid for complex variables. Many software packages
contain the necessary modifications. MATLAB (as of the 2009) does not. The results
presented here were obtained by programming a standard Newton-Raphson proce-
dure, which states that the jth estimate for a root is given by

( j) ( j−1) f (μH )
μn H = μn H−
d
f (μH )
dμ μH =μ( j−1) H
where
d
f (μH ) = − (ζ H + ikh) sin (μH ) − ζh (μH ) cos (μH )
d (μH )
Nonlinear equation solvers generally require initial estimates of the roots. A robust
approach that is suitable if it is only necessary to find the eigenvalues at a few values
of k H entails inspection of data contours. The characteristic equation is a complex-
valued function of the complex variable μH . When it and μH are decomposed into
real and imaginary parts, we may graph Re ( f (μH )) and Im ( f (μH )) as surfaces
above the complex plane μH whose orthogonal axes are Re (μH ) and Im (μH ). If
both parts of f are zero at the same point in the complex μH -plane, then that point
marks a root of the characteristic equation. The difficulty is that depiction of these
surfaces and their identification of the zeros would be challenging.
An alternative approach recognizes that if both parts of f (μH ) are zero at a root,
then it must be that | f (μH )| = 0 at such locations. A plot of | f (μH )| consti-
tutes a single surface above the μH complex plane. Its value may be depicted by a
contour plot. Because | f (μH )| cannot be negative, a root must occur at a location
where | f (μH )| is a minimum. Minima in a contour plot appear as a point inte-
rior to a contour that forms a closed curve. However, closed curves also surround
maxima, and a minimum value might not be zero. Thus a reasonable procedure is
to draw the contours, evaluate | f (μH )| at a point inside a closed loop, and test if
that value is close to zero. If so, that value of μH is a candidate as a starting value
for the root search. Recall that only the contours lying in the first quadrant should
be examined because the meaningful roots are those that give Re (μn ) ≥ 0 and
Im (μn ) ≥ 0.
This procedure is illustrated in Fig. 1, for which k H = π and ζ H = 0.1. Because
only the first quadrant is depicted there, a loop that actually is closed might appear
to be open if it is situated near the real or imaginary axis. In view of this possibility,
Fig. 1 suggests four roots, μn H ≈ (2n − 1) π/2 + 0.1ni. The value of | f (μH )| at
each of these points is less than 0.1, whereas | f (π + i)| = 6.4, which suggests that
the minima might be zeros. The actual roots were found to be μ1 H/π = 0.49949 +
0.0159i, μ2 H = 1.4984 + 0.0481i, μ3 H = 2.4972 + 0.0812i.
1.5
1
Im (H)
0.5
0
0 1 2 3 4
Re (H )/
Figure 1.
A different approach followed by some individuals finds estimates by considering

an ideal system, that is, one that is purely reactive. For example, in the present
situation the value of ζ H is much less than one. This suggests that the μn values are
close to the values for ζ H = 0, which is the pressure-release condition described by
Eq. (9.2.35). The real part of the estimates identified from the contour plot are these
values. The potential difficulty with this approach is that it does not address the case
where there is no comparable ideal system, which would be the case if Re (ζ H ) is
not small.
The present task calls for evaluation of the eigenvalues for a range of frequencies.
This suggests yet another approach for generating the initial guesses. The eigen-
values are continuous analytical functions of k H . Thus, if we increment k H by a
small amount, we can use the values identified for the previous frequency as starting
estimates for the new frequency. A benefit of following this approach is that it only
requires a contour map to identify the eigenvalues at the lowest k H . We begin with
a contour plot of | f (μH )| for k H = 0.01π, which is Fig. 2. It suggests that the
smallest eigenvalue is close to the origin, while the higher eigenvalues are multi-
ples of π plus a small imaginary part. Thus, the starting estimates used to initialize
the equation solver at k H = 0.01π are μ1 H ≈ 0.1π + 0.2i, μ2 H ≈ π + 0.1i,
μ3 H = 2π + 0.05i.
1.5
1
Im (H)
0.5
0
0 1 2 3 4
Re (H)/
Figure 2.
The procedure that was followed set a loop in which k H was increased gradually.
At each frequency, a loop over the mode number n initialized the μn H value with the
value found at the prior frequency. This value was input to the routine that solves the
characteristic equation. The output was the value of μn H at the current frequency.
After the eigenvalues were determined, the κn H were found from the definition,
Eq. (9.2.33), after which the frequency was incremented and the procedure repeated.
The result is described in Fig. 3. It will be noted that μn H is in the first quadrant,
and kn H is in the fourth quadrant, as required. The fact that all plots are continuous
curves is an indication that the procedure was carried out correctly.
10
Re (n)
0
1.5
Im (n)
1
n=1 n=2 n=3
0.5
0
15
Re (n)
10
5
0
0
Im (n)
−4
−8
0 5 10 15
Frequency kH
Figure 3.
Each mode shows a transition in its κn value as the frequency increases. Below a
certain frequency, Re (κn ) is small but nonzero, and Im (κn ) is relatively large nega-
tively. Above that frequency, Re (κn ) begins to grow and Im (κn ) is small negatively.
This behavior is analogous to cutoff for situations in which the wall does not dissipate
energy.
Division of a κn H value by the corresponding value of k H gives cphase /c. Evalu-
ation of the group velocity requires a little more effort. The nondimensional version
of Eq. (9.1.45) is

cg d (k H )
= Re
c d (κH )
Although we do not have a functional dispersion equation, we can use finite dif-
ferences to differentiate the data. A central difference formula based on a uniform
frequency increment = (k H ) j − (k H ) j−1 is
c
g
≈
c j Re (κH ) j+1 − Re (κH ) j−1
The results of these evaluations appear in Fig. 4.

6
4 cphase/c cg /c
2
0
n=1
−2
20
10 n=2
30
20
n=3
10
0
0 5 10 15
Frequency kH
Figure 4.
As the frequency increases, cphase approaches c from above, and cg approaches c

from below. The large values of cphase occur in the low-frequency range where κn
is essentially a negative imaginary value. This means that the waveguide mode is
evanescent. The consequence is that neither the high phase speed nor the negative
group velocity is meaningful, because the wave decays over a very short distance.
The transverse mode functions are computed by substituting into Eq. (9.2.33) the
values of μ1 H , μ2 H , and μ3 H at k H = 0.01, 1, and 10. The amplitude factor Bn
for each is arbitrary in regard to a pictorial representation. Because the eigenvalues
are complex, setting all Bn = 1 would lead to mode functions that have substan-
tially different magnitudes. To facilitate comparisons, the transverse mode functions
in Fig. 5 have been scaled by evaluating each as a function of y/H for Bn = 1
and then dividing each by its maximum magnitude. (The ordinate axis in Fig. 5 is
y/H , because the transverse direction runs vertically in the configuration depicted
in Fig. 9.7.)
Real part Imaginary part

1
kH = 10
kH = 1.0
Transverse distance y/H 0.5 kH = 0.1 Mode 1
0
1
Mode 2
0.5
0
1
Mode 3
0.5
0
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
Mode function n(y)
Figure 5.
A few features are worth noting. The highest frequency, k H = 10, exceeds the
cutoff value for each of the three modes. The consequence is that the imaginary
part of each mode function is much smaller than its real part. Indeed, the real parts
are nearly proportional to cos (πx/ (2H )), cos (3πx/ (2H )), and cos (5πx/ (2H )),
which are the first three modes for a rigid/pressure-release configuration. This behav-
ior indicates that the impedance is sufficiently low that the surface effectively is soft
at high frequencies. If k H is below the cutoff frequency for a mode, the imaginary
part of the mode is comparable to the real part, and it is oscillatory in a pattern that
resembles the real part
If we were to compare the present results to those for ζ H = 0, we would see that the
overall effect of dissipation associated with power transfer across the locally reacting
wall is quite small. This is especially true for modes that are not cutoff. For those
modes, the wave is gradually attenuated as it propagates. The modes whose cutoff
frequency exceeds the oscillation rate seem to be much affected by dissipation, as
suggested by a negative group velocity. However, these modes are rapidly attenuated,
so it would be difficult to observe such features.
EXAMPLE 9.5 The walls of a two-dimensional waveguide are identical elas-

tic plates whose properties are Young’s modulus E, density ρ p and thickness
h. The fluid is water, and the material properties of the plate are E = 100ρ0 c2 ,
ρ p /ρ0 = 5, and h/H = 0.02. Frequencies of interest are k H = 0.1, 1, and 10.
For each frequency, determine the eigenvalues of the transverse mode function
in the range μn H < 10. Also determine the corresponding values of κn H and
the ratio of the plate displacement amplitude to the pressure amplitude in each
mode.
Significance
The interaction between different media usually leads to interesting phenomena. In
a sense, this is an extension of the analysis of coincidence frequency for an elastic
plate in Sect. 5.5.2. The application of symmetry properties to simplify an analysis,
which is a general tool, plays a prominent role here.
Solution
The walls are flexible, but they are not locally reacting. Rather, application of a force
results in a displacement wave. The consequence it that it is necessary to formulate
the equation of motion for the plates. The pressure exerted by the fluid acts as a
load. In addition, enforcement of continuity of particle velocity at the walls must be
satisfied.
All features are the same above and below the central plane. A general property
of a system that has a plane of symmetry is that the response is either symmetric or
antisymmetric with respect to that plane. To exploit that feature the origin is placed
midway between the plates so that x z is the plane of symmetry. Figure 1 shows the
set up. The normal displacement of the plate w is defined to be positive into the fluid.
y
h
H/2 w(x,t)
x
Figure 1.
The frequency-domain representation of the pressure and displacement in a mode

are
p (x, y, t) = Re P (x, y) eiωt , w (x, t) = Re W (x) eiωt (1)
Every point (x, y) above the bisecting x z-plane is matched by a point at (x, −y).
A symmetric response has the property that P (x, −y) = P (x, y). In other words
P is an even function of y. An antisymmetric response is such that P (x, −y) =
−P (x, y), which means that it is an odd function of y. Because P is an analytic
function, it must be that ∂ P/∂ y = 0 at y = 0 for a symmetric response and P = 0
at y = 0 for an antisymmetric response. These, respectively, are the boundary condi-
tions for a rigid and pressure-release wall. Thus, we will consider two configurations
of a two-dimensional waveguide whose width is H/2: rigid at y = 0 and an elastic
plate at y = H/2 for symmetric modes, and pressure-release at y = 0 and an elastic
plate at y = H/2 for antisymmetric modes. We shall denote which type is under
consideration with a superscript “s” or “a ”, so the effective wall conditions are
∂ P (s) ∂ P (s)
= 0 at y = 0, n̄ · ∇ P (s) ≡ − = −iρ0 ω (−iωW ) at y = H/2
∂y ∂y (2)
(a)
(a) (a) ∂P
P = 0 at y = 0, n̄ · ∇ P ≡ − = −iρ0 ω (−iωW ) at y = H/2
∂y
The pressure must be a solution of the Helmholtz equation. Propagation in the

axial direction is represented by exp (−iκx). For the y direction we can immediately
take care of the boundary condition at y = 0 in Eq. (2) by recognizing that a cosine
function gives a zero gradient, whereas a sine function gives a zero value. We leave
the argument of each as μy, so the pressure may be written as
(s)
P (s) = B (s) cos μ(s) y e−iκ x
(a) (3)
P (a) = B (a) sin b μ(a) y e−iκ x
These forms satisfy the Helmholtz equation if

(s) 2 (s) 2 2 2
κ + μ = k 2 , κ(a) + μ(a) = k 2, (4)
The time-domain equation of motion for the plate is Eq. (5.5.27). The pressures
p1 and p2 appearing there are applied on either side of the plate. In the present
situation we shall ignore the fluid loading outside the waveguide. (This is acceptable
if there is a liquid within the waveguide and a gas outside, but it might not be a valid
assumption if the fluids are similar.) The displacement is defined to be positive if it
is into the waveguide, and the pressure pushes the plate away from the fluid. Hence
a negative p should cause a positive value of ∂ 2 w/∂t 2 , which means the pressure
term in the equation of motion that is preceded by the negative sign is the one to use.
In the two-dimensional problem, there is no variation in the z direction. Thus, the
equation of motion is
∂2w ∂4w
ρph + D = − p| y=H/2
∂t 2 ∂x 4
where p may be either the symmetric or antisymmetric variable. The parameter D is
the flexural rigidity. If the plate is composed of a homogeneous material, rather than
a composite, then
Eh 3
D=
12 1 − ν 2
Equation (1) gives the frequency-domain representation of w. To use it we must

recognize that the amplitude function W (x) must match the x dependence of the
pressure, because it is necessary to match W and P at the plate’s surface. Thus, W
must be representable as
W (x) = W̃ e−iκx (5)

where κ and W̃ may be the value for symmetric or antisymmetric modes. Substitution
of the representations of W and P into the plate equation of motion leads to

(s) (s) (s) H
Dκ − ρ p hω
4 2
W̃ = −B cos μ
2 (6)
H
Dκ4 − ρ p hω 2 W̃ (s) = −B (a) sin μ(a)
2
The last set of equations result from enforcing the continuity conditions in Eq. (2).
These equations are

H
B (s) μ(s) sin μ(s) = ρ0 ω 2 W̃ (s)
2 (7)
H
−B (a) μ(a) cos μ(a) = ρ0 ω 2 W̃ (s)
2
Equations (6) and (7) describe the coupling of the acoustic and structural responses.
Let us begin with the analysis of the symmetric modes. The matrix form of Eqs.
(7) and (6) for B (s) and W̃ is

(s)
B (s) 0
D = (8)
W̃ (s) 0
where ⎡ ⎤
(s) (s) H
(s) ⎢ μ sin μ −ρ0 ω 2
⎥
D =⎢
⎣

H
2

⎥
⎦
(s)
cos μ Dκ − ρ p ω
4 2
2
Equation (8) leads to B(s) = W̃ (s) = 0 unless the pair of scalar equations is rank-
deficient. Thus, we set D (s) = 0, which gives

(s) (s) H (s) H
Dκ − ρ p hω
4 2
μ sin μ + ρ0 ω cos μ
2
=0
2 2
Because the frequency is specified, the value of κ2 is defined in terms of μ(s) . This
means that the characteristic equation for fixed system parameters and frequency is
a function of μ(s) . In a nondimensional sense it is a function of η = μ(s) H , such that
% &
(k H )2
− η 2 2 η H ρ η
F (s) (η) = S
0
− 1 η sin + cos =0 (9)
(k H )2 2 h ρp 2
The parameter S is a measure of the dynamic stiffness of the plate. Based on an

assumption that Poisson’s ratio is ν = 0.3, it is
2
D 1 E ρ0 h
S≡ = = 0.080
ρph H c
2 2 12 1 − ν ρ0 c2
2 ρp H
The derivation of the equations for the antisymmetric modes follows a similar
path. When the pressure is P (a) in Eq. (3), the coefficient equations are like Eq. (8),
% &
B (a) 0
(a)
D =
W̃ (a) 0
⎡ ⎤
(a) H
(a) (10)
μ cos μ ρ0 ω 2
(a) ⎢
⎢ 2 ⎥
⎥
D =⎢ ⎥
⎣ (s) H ⎦
sin μ Dκ − ρ p ω
4 2
2
The characteristic equation for the antisymmetric modes is

% &
(k H ) 2
− η 2 2 η H ρ η
F (a) (η) = S
0
− 1 η cos − sin =0 (11)
(k H )2 2 h ρp 2
The rank reduction of the Eqs. (8) and (10) at an eigenvalue means that only one
equation is independent. We solve the first for the plate displacement in terms of the
pressure amplitude, which gives

(s) H
μ(s)
sin μ
n H (s)
Wn(s) 2 B
=
H (k H )2 ρ0 c2
(12)
(s) (a) H (a)
μn H cos μ
Wn(a) 2 B
=−
H (k H ) 2 ρ0 c2
When the characteristic equations, Eqs. (9) and (11), are evaluated at a fixed
k H , they become functions of a single variable, η. These equations may be solved
numerically, but doing so requires initial guesses. We shall obtain these by plotting
F (a) (η) and F (s) (η) as functions of η. The eigenvalues are the values of η at which
a function crosses the zero axis. Figures 2, 3, and 4 show such plots for k H = 0.1,
k H = 1, and k H = 10. The eigenvalues are marked as circles. The curves marked
as Fac(s) and Fac(a) denote the second term in each characteristic equation. They are the
terms that remain as ρ p h/ (ρ0 H ) → 0. Hence, they represent the purely acoustic
effect if the plate did not offer any opposition to an applied pressure, that is, if the walls
were pressure-release. The eigenvalues in that case would be μn H = (2n − 1) π for
the symmetric modes and μn H = 2nπ for the antisymmetric modes. The contrary
case is ρ0 H/ ρ p h → 0. This corresponds to a plate that is extremely massive, and
therefore immobile. In such situations the first part of each characteristic equation
is dominant. The same behavior is obtained in the limit as k H → 0, that is, low
frequencies. The corresponding eigenvalues are μn H = 2nπ for the symmetric
modes and μn H = (2n − 1) π for the antisymmetric modes. For the most part, the
roots in Fig. 2 match those for the rigid wall limit. The exception are the first two
antisymmetric modes. We will see that the axial wavenumber for these modes match
the wavenumber at which a flexural wave at k H = 0.1 can propagate along the
plate in a vacuum. The plate cannot sustain an external load (the pressure) in such
conditions.
kH = 0.1
20
F (a)()
F (s)( )
10
0
(s)
F ac ( ) F (a)
ac (  )
−10
−20
0 5 10 15 20 0 5 10 15 20
H H
Figure 2.
kH = 1.0
20
10
(s)
F ac ( )
0 F (a)
ac (  )
−10 F (s)()
F (a)( )
−20
0 5 10 15 20 0 5 10 15 20
H H
Figure 3.
At the higher frequency, k H = 10, described in Fig. 4, neither the pressure-

release nor rigid limits for the wall match the computed eigenvalues. Furthermore,
the increment between adjacent eigenvalues is irregular. Both features indicate that
the structural dynamics behavior of the wall has an important influence on the acoustic
response. The modal properties obtained from numerical solution of both characteris-
tic equations are tabulated below. Perhaps the most obvious feature is the irregularity
of the sequence in which symmetric and antisymmetric modes occur at low frequen-
cies. Another manifestation of the structural dynamic effect is the fact that the wall
displacement to pressure ratio is much lower at the higher frequency. An important
property is that all κn H values are imaginary for k H = 0.1 and 1.0. This means that
all of the low-frequency modes are evanescent, so a signal will not propagate in the
low-frequency range.
kH = 10
20
10
F (s)(  ) F (a)( )
F (a)
ac (  )
0
(s)
F ac ( )
−10
−20
0 5 10 15 20 0 5 10 15 20
H H
Figure 4.
Wn /H
k H Mode #n μn H κn H
Bn / ρ0 c2
0.10 Antisymmetric 1 1.7770 −1.7742i −112.054
Antisymmetric 2 3.0445 −3.0428i −14.780
Symmetric 1 6.2806 −6.2798i 0.8044
Antisymmetric 3 9.4244 −9.4239i 0.1585
1.0 Symmetric 1 2.7933 −2.6082i 2.7511
Symmetric 2 4.3647 −4.2486i 3.5736
Symmetric 3 5.8555 −5.7695i 1.2426
Antisymmetric 1 9.3891 −9.3357i 0.1673
10.0 Symmetric 1 2.9461 9.5562 0.02932
Antisymmetric 1 5.6095 8.2785 0.05294
Symmetric 2 8.1489 5.7961 −0.06546
A graph showing how the eigenvalues depend on k H would be helpful for a

closer examination. Such a picture could be obtained by following the preceding
analysis for a sequence of closely space k H values. Fortunately, a more efficient
method is available. It is based on considering the characteristic equations, Eqs.
(9) and (11), to be functions of k H , as well as μH . The value of each function
may be considered to form a surface above the κH, μH -plane. The eigenvalues
are the κH, μH where the elevation of this surface is zero. Thus, the numerical
procedure entails defining a grid of k H and μH values, evaluating F (s) and F (a)
at each grid point, and then finding the zero contour of each set of function data.
The contours of equal values of F (s) and F (a) are computed. The only contours of
interest are those for which F (s) (η) = 0 or F (a) (η) = 0. In MATLAB, the oper-
ations would be hold on; kH=0:d_k:kH_max; eta=0:d_e:eta_max;
C_sym = contour(kH, eta/pi, F_val_sym.’, [0 0]); C_anti
= contour(kH, eta/pi, F_val_anti.’, [0 0]);. In these opera-
tions, the contour command graphs the data, the last argument being [0 0] spec-
ifies that only the zero contour should be plotted, and hold on at the beginning
of the procedure retains the first set of contours when the second set are plotted.
A useful aspect is that the k H, μH coordinates of the points on each contour are
embedded in C_sym and C_anti. This data may be used for further study, for
example, to determine the group velocity.
Figure 5 shows the resulting contours. When viewed as plots of μn H as a function
of k H , each curve is multivalued over a small range of k H . The irregular spacing of
the tabulated data stems from this property. The shading marks the region in which
μH > k H , or equivalently, where κH is imaginary. The associated modes evanesce.
Inspection of the contour data shows that all eigenvalues lie in this shaded region if
k H < 2.64. This means that k H = 2.64 is the minimum frequency at which a signal
can propagate.
9
8 Antisymmetric
7 Symmetric
6 Antisymmetric
5
nH
4
Fpl = 0 Symmetric
3
Antisymmetric
2 Symmetric
Fpl = 0
 Antisymmetric
Symmetric
0
0 2 4 6 8 10 12 14 16 18
kH
Figure 5.
The dashed curves in Fig. 5 marked Fpl = 0 are the locus of points at which the
term within braces in both characteristic equations vanishes. This term is
2
D (k H )2 − η 2
Fpl = −1
ρ p h H 2 c2 (k H )2
This factor is the nondimensional version of Dκ4 −ρ p ω 2 after substitution of c2 κ2 =

ω 2 − c2 μ2 . In other words it is the dispersion equation for waves in an elastic plate
that is not loaded by the fluid. The intersection of an Fpl curve with one of the
eigenvalue contours corresponds to the coincidence frequency for a plate bounding
a half-space. When Fpl = 0, the inertia and stiffness effects of the plate cancel. In
the vicinity of the lower curve, the value of Fpl is relatively small, so the eigenvalues
are relatively insensitive to this term. The upper curve is the contrary condition. The
intersection of this Fpl curve with any eigenvalue curve is centered in the region
where each curve is multivalued, which suggests that these values are very sensitive
to the plate’s properties. Note that the axial phase speed is ω/κ, so the lower Fpl
curve describes waves in the plate/wall that are supersonic.
A lesson that can be gleaned from this example is that systems in which there
is a strong interaction between an acoustic medium and a structure might exhibit
unexpected effects. We also have seen that the cause of these effects can be identified if
the properties of the system’s response are examined from an appropriate perspective.
It is worth noting that the steps leading to the governing equations are typical of all
systems in which elastic and acoustic subsystems strongly interact. Continuity of the
normal displacement at the interface leads to a boundary condition for the fluid, and
the pressure applied to the structure is accounted for as a dynamic load.
9.2.5 Orthogonality and Signal Generation
The primary motivation for identifying the modes of a waveguide is that they simplify
an analysis of the signal that is generated by an excitation. This excitation is taken
to be a specification of the axial particle velocity on the cross section at x = 0. The
velocity distribution is Vs (y), so the boundary condition is
1 ∂P
Vx = − = Vs (y) at x = 0 (9.2.42)
iωρ0 ∂x
The other condition is that waves are assumed to be absorbed or perfectly transmitted
at a long range, so we are solely concerned with waves that propagate in the positive
x direction.
How to satisfy the source condition is the question addressed here. The situation is
actually close to that which arose for a spherical source. There the signal propagated
to an infinite radial distance, and we used a spherical harmonic series to represent
the transverse variation. Here the propagation is in the axial direction. We know how
the signal in a mode varies in the transverse y direction, so it is logical to consider a
modal series,
'∞
P= Pn n (y) e−iκn x (9.2.43)
n=0
In this expression the Pn coefficients are called the modal amplitudes of the mode
functions. The sum is indicated to begin at n = 0, but that is reserved for the plane
wave mode in the case of a hard-walled waveguide. Also, if the waveguide has modes
whose eigenvalues are purely imaginary, they must be inserted into this series.
The path leading to an expression for the modal amplitudes is like that by which
the coefficients of a Fourier series are determined. Indeed, some individuals refer to
a modal series, Eq. (9.2.43), as a generalized Fourier series. Both series are a sum
of a basis function, modal or trigonometric, multiplied by an unknown coefficient.
Each function represents a direction in a generalized linear space and the coefficient
represents a vector component in that direction. The similarity extends further, in
that the basis functions form an orthogonal set.
The procedure for deriving the orthogonality property is part of Sturm–Liouville

analysis in a course on partial differential equations. We begin by observing that the
transverse modes satisfy the mixed boundary conditions at y = 0 and y = H stated
in Eq. (9.2.29),
ζ0 dn
n = at y = 0
ik dy
(9.2.44)
ζ H dn
n = − at y = H
ik dy
Within the domain of the fluid the mode functions satisfy a one-dimensional
Helmholtz equation,
d2
n + μ2n n = 0 (9.2.45)
dy 2
The key step is selection of two arbitrary mode indices n and j. The Helmholtz
equation for each equation is multiplied by the other mode function. The difference
of the two products is
2
d2 d
j n + μ2

n n − n j + μ2

j j =0 (9.2.46)
dy 2 dy 2
The next operation entails integrating this expression over 0 < y < H ,

H
d2 d2 H
j 2 n − n 2 j dy + μ2n − μ2j j n dy = 0 (9.2.47)
0 dy dy 0
Integration by parts is applied to the integral containing derivatives. Let us examine

one of those terms,
H
H
d2 d d d
j 2 n dy ≡ − j n dy + j n
0 dy 0 dy dy
dy y=H
(9.2.48)
d
− j n
dy y=0
If a boundary is rigid, then the normal derivative is zero, whereas the function is
zero if a boundary is pressure-release. In either case the terms that are evaluated at
the boundaries vanish. Otherwise the impedance boundary conditions in Eq. (9.2.44)
relate the y derivative to the function at each wall. Their substitution into the preceding
gives
H
H
d2 d d ik
j 2 n dy = − j n dy + j n
0 dy 0 dy dy ζH y=H
(9.2.49)
ik
− j n
ζ0 y=0
When the same set of operations is performed on the other derivative term in
Eq. (9.2.47), the result has the form of this relation with n and j swapped. However,
such an interchange does not alter the result, which means that all terms cancel.
Thus, regardless of the wall impedances, the integral in Eq. (9.2.47) that contains
derivatives vanishes. What remains is the orthogonality property,

H
μ2n − μ2j j n dy = 0 (9.2.50)
0
If j and n are different, then the integral must vanish. If the indices are equal,
then the integral will not evaluate to zero. We can use this fact to set the arbitrary
coefficient Bn in Eq. (9.2.33). In particular we will select it such that the integral
equals H , which will make the transverse mode functions dimensionless. Let ñ
denote a transverse mode function when Bn = 1, so that
n = Bn ñ

H 2 −1/2
1 (9.2.51)
Bn = ˜ n dy

H 0
The integral property for any pair of n and j values is captured by using a Kronecker
delta, δn, j = 0 if j = n and δn, j = 1 if j = n. Thus,
H
j n dy = H δn, j (9.2.52)
0
From a linear algebra perspective the integral is the scalar product of n and j in
the functional space of a single Cartesian coordinate. Correspondingly, we say that
the mode functions form an orthonormal set. Without the above definition of Bn ,
the mode functions would be the analog of orthogonal vectors that have arbitrary
magnitude. With it, the mode functions are analogous to a set of orthogonal unit
vectors. A rule that scales the arbitrary coefficient of a mode according to a specified
rule is said to be normalization. To some extent it would be less confusing to say that
the rule “standardizes” a mode because “normal” is synonymous with “orthogonal”,
which is a property the modes always have. Nevertheless, the term is generally
accepted, so we shall use it.
To determine the modal coefficients. we express the given boundary excitation as
a series whose basis functions are the transverse modes n (y), specifically
∞
'
Vn n (y) = Vs (y) (9.2.53)
n=0
An explicit description of the velocity coefficients is obtained from the modal orthog-
onality property. Thus, the preceding is multiplied by a specific function j (y), and
both sides are integrated over the cross section. This operation filters the jth term
from the sum, with the result that
H
1
Vj = Vs j dy (9.2.54)
H 0
At the same time, Euler’s equation states that the axial particle velocity at x = 0
corresponding to the modal pressure series is
∞
i d P 1 '
Vs = = κn Pn n (y) (9.2.55)
ωρ0 d x x=0 ωρ0 n=0
This description of the axial velocity must be the same as Eq. (9.2.53). The alternatives
are modal series with the same orthogonal set of basis functions. Equality requires
that the coefficients match, which leads to
k
P j = ρ0 c Vj (9.2.56)
κj
The corresponding expression for the pressure field is
'∞
k
P (x, y) = ρ0 c V j j (y) e−iκ j x (9.2.57)
j=0
κ j
It is implicit to this expression that evaluation of κn is based on a branch cut for the
square root that places κn in the fourth quadrant of the complex plane.
Equation (9.2.56) tells us that the source might generate an infinite number of
waveguide modes. The degree to which the source velocity matches a mode function
dictates the magnitude of the velocity coefficient. The other factor affecting the
pressure coefficients is κn , whose minimum magnitude at a specific frequency occurs
in the vicinity of n cutoff . For n n cutoff , κn will be much larger than k, which is one
reason that the pressure series converges. This property leads to convergence of the
modal series at any x. Convergence improves as x increases because the evanescent
modes decay increasingly.
The case of a point source situated somewhere on the cross section at x = 0
leads to an explicit description. Recall that the volume velocity is the limit as the
source’s surface area shrinks to zero and the normal velocity of the source increases
commensurately. Accordingly, the source is described by a Dirac delta function. For
a two-dimensional waveguide with the source situated at y = y0 , the volume velocity
is per unit width in the invariant direction. In other words, we are considering a line
source. Hence, we have
% &
Q̂/ε if y0 < y < y0 + ε
Vs = lim = Q̂δ (y − y0 ) (9.2.58)
ε→0 0 otherwise
The velocity coefficients obtained by application of Eq. (9.2.54) are V j = Q̂ j

(y0 ) /H . The corresponding pressure field is
∞
Q̂ ' k
Psource = ρ0 c j (y0 ) j (y) e−iκ j x (9.2.59)
H j=0 κ j
It is evident that if the source is situated at a location where mode # j is largest,

that mode will be excited to the greatest extent possible. Conversely, if the source
is situated at a node of the jth mode, that mode will not be excited. Furthermore, if
the pressure is observed at a location where the jth mode has a node, there will no
evidence of that mode at that location, even though it might be an important mode
for the pressure at other points.
Orthogonality of transverse mode functions plays an important role in an analysis
of axial power flow. To evaluate the axial component of intensity we require an
expression for Vx , which we obtain by applying Euler’s equation to the modal series
for pressure. This operation yields
∞
'
Vx = V j j (y) e−iκ j x (9.2.60)
j=0
The time-averaged power flow past any cross section is the average intensity at fixed
x integrated over y,
⎧ ⎡ ⎤ ⎫
1 ⎨ H '∞
k '∞ ⎬
Pav (x) = Re ⎣ρ0 c V j j (y) e−iκ j x ⎦
∗
Vn∗ n (y) eiκn x dy
2 ⎩ 0 κ ⎭
j=0 j
⎧ n=0
⎫
⎨'∞ '
∞ H ⎬
ρ0 c k
V j Vn∗ e−i (κ j −κn )x
∗
= Re j (y) n (y) dy
2 ⎩ κj 0 ⎭
j=0 n=0
(9.2.61)
Because the transverse mode functions form an orthogonal set that is normalized to
H , the terms in the double sum for which n = j integrate to zero. This reduces the
expression to
⎡ ⎤
'∞
ρ0 cH k 2 2 Im κ x
Pav (x) = Re ⎣ V j e ( j ) ⎦ (9.2.62)
2 j=0
κ j
The first notable aspect of this expression is that the modes do not couple, so the
power flow is the sum of the contribution of each mode. If both walls are purely
reactive, then the κ j values below cutoff ( j ≤ Ncutoff ) will be real. The exponential
decay factor in that case is unity, which means that these modal contributions to the
power flow are independent of x. The axial wavenumber is negative imaginary for
j > Ncutoff . The exponential factor in the summation is real for these modes, but
the presence of κ j in the denominator means that each of these terms is imaginary.
The consequence is that evanescent modes do not contribute to the power flow. (The
fundamental reason for this property is that the axial velocity in an evanescent mode
is 90◦ out of phase from the pressure.) This means that if there is no dissipation the
power input to a waveguide at one end is transported downstream by the propagating
modes without loss. The situation is different if the impedance of either wall has
a resistive part. In that case all axial wavenumbers are complex. We established
previously that the κj values must lie in the fourth quadrant of the complex plane,
which means Im κ j < 0. Thus the power input to each mode at one end of the
waveguide is progressively dissipated through the walls(s) as the mode propagates
downstream. The decay rate is twice as fast as the decay of the pressure amplitude. If
the waveguide is sufficiently long, the pressure and therefore the power, that reaches
the far end will be negligible.
EXAMPLE 9.6 The sketch shows a rectangular tank of water that is open to
the atmosphere. On the left wall a ribbon transducer is mounted at mid-depth.
The transducer executes a uniform translational oscillation in the direction
normal to the wall. Its width is sufficiently narrow that itmay be approximated
as a line whose a volume velocity per unit length is Re Q̂ 0 eiωt . The bottom
and side walls are composed of a thick glass sheet that may be considered to be
rigid. The far wall consists of a mosaic of piezoceramic tiles that are actively
controlled to achieve a nonreflecting boundary condition. The tank dimensions

are H = 600 mm and L = 3 m, the frequency is 20 kHz, and Q̂ 0 = 8 10−4
m2 /s. The wall containing the ribbon transducer is the plane x = 0, and y
is measured from the bottom. Determine the dependence of | p| on the axial
distance x at mid-depth, y = H/2, and also determine the depth dependence
of | p| at x = 0, L/2, and L. At each location, identify the portion that is
attributable to the modes that are below the cutoff wavenumber. In addition,
determine the spatially averaged impedance required of the piezoceramic tiles.
Ribbon transducer
id Rigid
Rig
H H/2 Rigid
L
Piezoceramic tiles
Figure 1.
Significance
The best way to realize the effect of the cutoff phenomenon and the role of the
evanescent spectrum, is to carry out a computation. The results will illustrate the
general nature of multimodal signals in a waveguide.
Solution
It might not be obvious that this system constitutes a two-dimensional waveguide
because the tank is not infinitely wide in any direction parallel to the plane of the
excitation. However, the vertical side walls are rigid, which means that ∂ P/∂z must
be zero along them. A pressure field that does not depend on z will satisfy this
condition. The ribbon transducer executes the same vibration at all z, so a pressure
field fitting the description Re P (x, y) eiωt is fully consistent. Thus, the pressure
is described by the modal series in Eq. (9.2.43). The bottom of the tank is rigid, and
the top is the free surface, which is well represented as pressure-release. The modes
for such a configuration are described by Eq. (9.2.35),

1 π
n = Bn cos μ y y , μn = n − , n = 1, 2, ... (1)
2 H
The Bn coefficients are normalized according to Eq. (9.2.51), which gives

2 −1/2
H
(2n − 1) π y
Bn = H 1/2
cos dy = 21/2 (2)
0 2 H
The excitation is a line source having a specified velocity. Aside from the line at
y = H/2, the wall is rigid, so the velocity boundary condition at x = 0 is

H
Vx |x=0 = Q̂ 0 δ y − , 0<y<H (3)
2
According to Eq. (9.2.54), the coefficients of a modal series for this velocity distri-
bution are

1 H
Q̂ 0 Q̂ 0 1/2 (2n − 1)π
Vn = Q̂ 0 δ(y − H/2)n dy = n (y = H/2) = 2 cos
H 0 H H 4
(4)
The resulting modal series obtained from Eq. (9.2.57) is
'∞
k
P (x, y) = ρ0 c Vn n (y) e−iκn x (5)
n=0
κ n
The last consideration before we may evaluate the pressure is selection of the
number of terms N to include in the series. The transverse wavenumbers are given
by Eq. (1). The cutoff condition is marked by μn = k = 84.91 based on c = 1480 m/s.
This leads to

kH 1
Ncut = floor + = 16
π 2
As was noted, convergence of the modal series, Eq. (9.2.43), can result from reduc-
tion of |Pn | to a negligible value above some n. This possibility constitutes the
convergence criterion for small x, because |exp (−iκn x)| will be O (1) regardless
of whether κn is real or negative imaginary. An important aspect is the fact that κn
becomes very small near n cutoff . Because κn appears in the denominator of Eq. (3),
achieving convergence in this manner requires that N > Ncut . The other terms in
Eq. (3) do not grow with increasing n, so setting N substantially larger than n cutoff
should be adequate. We shall use N = 3Ncut , which leads to |Pn | < 0.07 max |Pn |
for n > N .
The other attribute that leads to convergence of a modal series is the decay of
the evanescent modes with increasing x. This decay increases with increasing mode
number. Thus, fewer and fewer terms are required as x increases. We could use this
attribute to reduce N at each x based on the value of exp (−iμn x) for n > n cutoff .
We shall not do so to avoid complicating the computational program. Furthermore,
the processing time required for computations at fixed N is not prohibitive. The
only remaining consideration is a caution against a common error. The erroneous
procedure evaluates the magnitude of each term in the modal series. The correct
procedure evaluates the sum first.
Before we examine data, it is useful to recall the earlier matrix based algorithm for
evaluation of a series. Our interest is in the values of P at a set of x and y locations.
In Eq. (5) the mode function depends on y, and the exponential function depends
on x. Let these locations be x j , j = 1, ..., J and ym , m = 1, ..., M. The summation
over the mode number n can be expressed as a matrix sum. To do so, let [F] be a
matrix of transverse mode functions and [E] be a matrix of exponentials,
such that
Fn,m ≡ n (ym ) and E j,n ≡ exp −iμn x j . Then Pm, j ≡ P x j , ym is the element
of a rectangular array [P] at all locations that is computed from the matrix version
of Eq. (4), which is
[P] J ×M = ρ0 c [E] J ×N [K ] N ×N [F] N ×M (6)
where [K ] is a diagonal matrix whose elements are K n,n = (k/κn ) Vn . A row of [P]
will be the pressure values at all ym for fixed x j , whereas a column will be the values
at all x j for fixed ym . A bonus of this formulation is that changing the modal index
to extend from n = 1 to n = Ncutoff is the only modification required to exclude the
evanescent modes.
The first set of results we shall examine are the pressure amplitudes. According
to Eq. (5), they are
k
P̂n = ρ0 c Vn
κn
These values are described by Fig. 2. As was anticipated, the largest coefficients
occur in the vicinity of Ncut .
^
Re(Pn)
5000
^ )
Im(Pn
−5000
0 5 10 15 20 25 30 35 40 45
Mode number
Figure 2.
The axial dependence of the pressure may be computed according to Eq. (6)
with x j being a set of axial locations and y1 = H/2. Figure 3 shows the result.
Many individuals anticipate a smooth curve, whereas the actual result shows rounded
maxima separated by narrow near-null regions. Capturing these features requires a
fine scan; the data for the Fig. 3, as well as for the computations along a cross section
used a point spacing set at = (2π/k) /25. The cause of this behavior is interference,
as it is for a sound beam. The interference occurs here because the multiple modes
have different wavelengths. At some locations many are nearly in-phase creating
maxima, whereas at other locations the modal contributions nearly annihilate each
other. Another noteworthy feature is that the pressure very close to the transducer is
10 dB higher than it is at any downstream maximum.
102 All modes

Propagating modes
| p| (kPa)
100
0 0.5 1 1.5 2 2.5 3

Axial coordinate x (meter)
Figure 3.
Figure 3 shows that only very close to x = 0 is there significant difference between
the pressure obtained from the full series and the series that consists of only propa-
gating modes. This observation is confirmed by the transverse profiles in Fig. 4. To
obtain the data for Eq. (6), x j was set to the axial position and ym was a sequence of
values spaced at . Interference is evident at each cross section. The cross section at
the boundary, x = 0, shows a sharp pressure peak in the vicinity of y = 0.3 m, which
is the location of the transducer. With increasing distance from the excitation, that
tendency is overcome by the interference effect. Note that the profiles at x = 1.5 m
and x = 3 m are similar in magnitude, as well as spatial dependence, but they are
not the same. Further only at x = 0 is the contribution of the evanescent modes
significant.
All modes Propagating modes

0.6
Transverse coordinate y (meter)

0.5
0.4 x=0
0.3
x=3 m
0.2 x= 1.5 m
0.1
0
0 50 0 20 0 20
|p| (kPa)
Figure 4.
The last quantity to evaluate is the average impedance at x = L. This quantity

is merely the average complex pressure amplitude divided by the average particle
velocity. There is no variation in the z direction, so averages per unit width are
obtained by integrating over y. We use the modal pressure series in Eq. (5) to describe
P as a function of y, so that
∞ H

ρ0 c ' k 1 H
(2n − 1)π y −iκn L
Pav |x=L = Vn 21/2 cos P(L , y)dy = e dy
0 H n=1 κn H 0 2H
'∞ 3/2
k 2 Vn
= ρ0 c (−1)n−1 e−iκn L
n=0
κ n (2n − 1) π
(7)
The average axial velocity is obtained by integrating Eq. (9.2.60) over y,
∞
1 H
1' H
(2n − 1)π y
(Vx )av |x=L = Vx (L , y)dy = Vn e−iκn L 2 1/2
cos dy
H 0 H n=1 0 2H
'∞
23/2 Vn
= (−1)n−1 e−iκn L
n=1
(2n − 1) π
(8)
Despite the similarities of the sums in Eqs. (7) and (8), each sum must be evaluated
individually. The average specific impedance that results from these expressions is
Z av Pav |x=L
ζav = = = 1.086 − 0.0043136i
ρ0 c ρ0 c (Vx )av |x=L
The average impedance is very close to ζ = 1, which is the value for which the
reflection coefficient of a plane wave is zero.
Before we leave this system, it is interesting to contemplate an alternative solution
based on the method of images. Example 6.8 developed such a solution for the
case of a point source in a hard-walled waveguide. The ribbon transducer in the

present system effectively is a line source, so all images also are line sources. Images
above the free surface, y = H , invert images below that surface, while images
below the bottom are in-phase replicas of the images above that surface. This creates
an infinite sequence of line sources lying in the x = 0 cross section centered at
y = H/2 and separated by distance H . The pressure amplitude radiated by a line
source decays as 1/R 1/2 when the distance R transverse to that line is large. If we
were to impose a comparable convergence criterion to that employed for the modal
series, we would cease a summation of all images when the nth line source above
or below has an amplitude that is 1% of the largest contribution. For a field on the
axis, the closest image is the actual transducer, which is at y = H/2, x = 0. The
corresponding source-image distance is R0 = x. The distance to the same field point
1/2
from the nth image is Rn = x 2 + (n H )2 . The aforementioned convergence
condition suggest that we may omit from the image summation any image for which
1/2 1/2 1/2
1/Rn < 0.01/R0 . This simplifies to x 2 + (n H )2 > 104 x, which is well
approximated as n > 10 x/H . In other words, the number of images required to
4
obtain a suitably convergent representation of the on-axis pressure must increase

drastically in proportion to the distance. The signal arriving at a field point from each
image corresponds to a ray. In general, the number of rays required to represent a
signal increases greatly with increasing range. In contrast, a fixed, relatively small
number of modes usually suffices.
9.3 Three-Dimensional Waveguides
The two-dimensional model of a waveguide encapsulates the conceptual and phe-

nomenological features of other waveguides whose cross section is invariant. This
will be evident in the following general formulation and its application to cases of
rectangular and circular cross sections.
9.3.1 General Analytical Procedure
As we did previously, we designate the direction in which signals propagate as x.

The cross section is A, whose shape and size are constant. The other restriction
imposed on this general exploration is that the side wall, which forms the perimeter
C, is locally reacting. The specific impedance ζ may vary along the perimeter, but it
is independent of x (e.g., a rectangular cross section could have walls different types
of sides). Let ξ2 and ξ3 denote the transverse coordinates. These would be Cartesian
coordinates if the shape is rectangular, or polar coordinates if the shape were circular,
but their definition is unimportant at this juncture. The transverse mode functions
therefore are n (ξ2 , ξ3 ), and a waveguide mode is obtained by multiplying these
functions by a complex exponential representing the propagation. In other words,

we anticipate that the waveguide modes will be described as
P = n (ξ2 , ξ3 ) e−iκn x (9.3.1)
As always, the wavenumber must lie in the fourth quadrant of the complex plane in
order to satisfy the Sommerfeld radiation condition. The transverse mode functions
n are solutions of a Helmholtz equation whose general form is
∇˜ 2 (n ) + μ2n n = 0, μ2n = k 2 − κ2n (9.3.2)
where ∇˜ 2 denotes the portion of the Laplacian that contains derivatives with respect
to ξ2 and ξ3 .
As was the case for a two-dimensional waveguide, values of μn will be found
from the condition that there be a nontrivial solution consistent with homogeneous
boundary conditions at the walls. Application of Euler’s equation to describe the par-
ticle velocity in the local impedance model, with n̄ defined to be into the waveguide,
leads to
ζ
P = ρ0 cζ −n̄ · V̄ = n̄ · ∇˜ P if x̄ ∈ C (9.3.3)
ik
where C is the perimeter of the cross section A. Substitution of the representation
of P as a waveguide mode leads to a boundary condition for the transverse mode
functions,
ζ ˜ n
n = n̄ · ∇ (9.3.4)
ik
In addition to similarities of the equations that must be solved to determine the

transverse mode functions, the general problem is like the two-dimensional one
because its transverse mode functions are orthogonal. The general orthogonality
condition is obtained similarly to the derivation of Eq. (9.2.52). Thus, we consider
two eigensolutions and form

j ∇˜ 2 (n ) + μ2n n − n ∇˜ 2 j + μ2j j
(9.3.5)
˜ n − ∇˜ · n ∇
≡ ∇˜ · j ∇ ˜ j + μ2n − μ2j j n = 0
This relation is integrated over a cross section. Application of the divergence theorem,
with n̄ defined as the normal on the perimeter C pointing into A, leads to

2
˜ ˜
n̄ · n ∇ j − j ∇n dC + μn − μ j
2
j n h 2 (ξ2 ) h 3 (ξ3 ) dA = 0
C
A
(9.3.6)
9.3 Three-Dimensional Waveguides 255
The terms evaluated on C will vanish for the rigid (n̄ · ∇ ˜ n = 0) and pressure-
release cases. If the wall is locally reacting, then Eq. (9.3.4) gives

˜ ik ik ˜ n j
n n̄ · ∇ j = n j = n j = n̄ · ∇ (9.3.7)
ζ ζ
Thus the integral over C reduces to zero for any local impedance, even if ζ varies
along C.
What remains in Eq. (9.3.6) is the integral over A. If the two mode functions
correspond to different eigenvalues, μ j = μn , then the integral must vanish. In the
case where μ j = μn , the integral may have any value. We shall use this arbitrariness
to normalize the mode functions. Specifically, the n are scaled such that the integral
equals the cross-sectional area. Thus, we will have

j n dA = Aδ j,n (9.3.8)
A
Fulfillment of this condition when j = n sets the Bn coefficient of n , which is

arbitrary in regard to satisfying the eigenvalue problem.
Given the similarity of the general relations governing three-dimensional trans-
verse mode functions and those for a two-dimensional waveguide, it is consistent
that a series of these functions is used to represent the pressure in the general case.
The coefficients of the modes are again designated as Pn , so that
∞
'
P= Pn n (ξ2 , ξ3 ) e−iκn x (9.3.9)
n=1
The axial velocity distribution corresponding to this description of the pressure is
∞
' κn Pn
Vx = n (ξ2 , ξ3 ) e−iκn x (9.3.10)
n=1
κ ρ0 c
The procedure for determining the modal amplitudes is the same as it was for the
two-dimensional waveguide. Typically, the excitation consists of a specified axial
velocity Vs (ξ2 , ξ3 ) at x = 0. This distribution is described by a modal series, specif-
ically
∞
'
Vs (ξ2 , ξ3 ) = Vn n (ξ2 , ξ3 ) (9.3.11)
n=1
To obtain an expression for the velocity coefficients the preceding is multiplied by

an arbitrarily selected j and integrated over A. Orthogonality, Eq. (9.3.8), filters
the jth term out of the summation, which leads to

1
Vj = j (ξ2 , ξ3 ) Vs (ξ2 , ξ3 ) dA (9.3.12)
A
A
Equation (9.3.10) evaluated at x = 0 must be the same as the modal series of Vs .

This requires that the coefficients of n match, which leads to
k
Pn = ρ0 c Vn (9.3.13)
κn
There is no difference between this expression and Eq. (9.2.56) for the two-
dimensional system.
These expressions lead to a general description of the power flow. The time-
averaged power flowing across any cross section is the integral over that surface of
the time-averaged intensity,
' ∞

1 1
P (x) = Re P Vx∗ dA = Re Pn n (ξ2 , ξ3 ) e−iκn x
2 2 n=1
A ⎡ A ⎤
'∞ ∗
κ j P j∗ ∗
×⎣ j (ξ2 , ξ3 ) e+iκ j x ⎦ dA
j=1
κ ρ0 c
(9.3.14)
Different indices are used for the summations to assure that all products are included.
The only quantities that depends on position are the n functions. This allows us to
sum the integrals of the products of these functions. The result is that
'∞ ' ∞ ∗
1 κ j Pn P j∗ −i κn −κ∗j x
P (x) = Re e n (ξ2 , ξ3 ) j (ξ2 , ξ3 ) dA
2 n=1 j=1
κ ρ0 c
A
(9.3.15)
The last step is to apply the orthogonality of the transverse mode functions, which
eliminates all coupling between modes. Thus, we find the power flow to be a sum of
modal contributions,
∞
' κ∗
A
P (x) = Re n
|Pn |2 e+2 Im(κn )x (9.3.16)
2ρ0 c n=1
k
This is a useful relation because it allows for identification of modes that transport
the greatest amount of energy. In the case of a purely reactive wall, the axial wave
numbers are real for modes that propagate, and imaginary for
evanescent modes.
The exponential factor for the latter modes is real, but Re κ∗n = 0, so the evanes-
cent modes do not contribute the power transfer. Furthermore, Im (κn ) = 0 for
the propagating modes, so the contribution of these modes to the power transfer is
independent of x. Thus, P (x) is independent of x if the walls are purely reactive,
regardless of the shape of the cross section. The modal contributions to the power
flow depend on x only if the wall has a nonzero resistance. In that case the eigenval-
ues and, therefore the axial wavenumbers, are complex. We selected the branch cut
for the square root such that Im (κn ) < 0 in order to satisfy the radiation condition.
The consequence is that power flow is attenuated in the downstream direction. The
energy that is dissipated is absorbed in the wall.
The relations we have derived for the pressure, particle velocity, and power flow are
generally applicable. However. their implementation assumes that we have identified
transverse mode functions that satisfy the Helmholtz equation, Eq. (9.3.2), subject
to the wall condition in Eq. (9.3.4). The ease of that determination depends on the
shape of the cross section. The following studies explore rectangular and circular
cross sections, which are the most amenable to formal analysis. Even within that
limitation, an analysis using standard methods is not always possible. The derivation
allowed the wall impedance to vary along the perimeter. However, the analyses we
shall perform require that the boundary condition be constant along any segment of
a wall that corresponds to a constant coordinate. In other words, the impedance of
each side of a rectangular cross section is required to be constant, and the impedance
of a circular wall cannot depend on the circumferential angle. If these conditions are
not met, or if the subject is a waveguide whose cross section has an irregular shape,
numerical techniques will be required to determine the transverse mode functions.
Nevertheless, they do exist for such systems, and they do constitute an orthogonal set.
9.3.2 Rectangular Waveguide
A waveguide whose cross section is rectangular is the least challenging to analyze

because we can define a Cartesian coordinate system whose coordinate planes are
parallel to the walls. The walls in Fig. 9.13 correspond to y = 0, y = H , z = 0,
z = W , and the excitation is applied to the end at which x = 0. Waves propagate in
the positive x direction, and it is assumed that the far end is terminated in a manner
that suppresses reflections.
Fig. 9.13 Definition of the y

Cartesian coordinate system
used to analyze propagation
in a rectangular waveguide W
z x
The waveguide mode is
P (x, y, z) = n (y, z) exp (−iκn x) (9.3.17)
The differential equation governing a transverse mode function (y, z) is found by

substituting this form into the Helmholtz equation, which leads to

∂2 ∂2
+ 2 + k 2 − κ 2 n = 0 (9.3.18)
∂y 2 ∂z
Each wall may have a local impedance ζ j . The normal directions into the fluid are
ē2 for the wall at y = 0, −ē2 for the wall at y = H , ē3 for the wall at z = 0, and −ē3
for the wall at z = W , so the specific forms of the general local impedance boundary
condition, Eq. (9.3.4), are
ζ1 ∂ P
P = −ρ0 cζ1 Vy = at y = 0
ik ∂ y
ζ2 ∂ P
P = ρ0 cζ2 Vy = − at y = H
ik ∂ y
(9.3.19)
ζ3 ∂ P
P = −ρ0 cζ3 Vz = at z = 0
ik ∂z
ζ4 ∂ P
P = ρ0 cζ4 Vz = − at z = W
ik ∂z
We could identify by separation of variables, but it is more direct to guess the

form of the general solution and then seek the condition that such a form must
satisfy. The analysis of a two-dimensional waveguide began by expressing as a
sum of complex exponentials, so it is logical to try the same here. Because either
y or z is constant on a boundary, while the other coordinate has a range of values,
a solution that is capable of simultaneously solving the boundary conditions on all
walls must be a product of functions of y and z. Thus, the appropriate ansatz is
= φ y (y) φz (z)
(9.3.20)
φ y = B1y e−iμy + B2y eiμy , φz = B1z e−iνz + B2z eiνz
The trial solution will satisfy the Helmholtz equation if
μ2 + ν 2 + κ2 = k 2 (9.3.21)
We substitute the trial form for in Eq. (9.3.20) into the boundary conditions. The
φz function is a common factor in the conditions at constant y, and φ y is a common
factor in the conditions on the walls at constant z. Cancelation of these common
factors leads to two pairs of boundary conditions, which may be written as
T
[D (μH, k H, ζ1 , ζ2 )] B1y B2y = [0 0]T
T (9.3.22)
[D (νW, kW, ζ3 , ζ4 )] B1z B2z = [0 0]T
Let α and β denote the first two variables in the coefficient matrices for each direction,
and ζa and ζb be the corresponding specific impedances. Then these matrices are
described by

(ζa α + β) (−ζa α + β)
[D (α, β, ζa , ζb )] ≡ (9.3.23)
(−ζb α + β) e−iη (ζb α + β) eiα
In this way the analysis reduces to finding the transverse modes for two two-
dimensional waveguides. The wall conditions at constant y set the μn eigenvalues,
and the conditions at constant z set the νn values. The characteristic equations are
|[D (μH, k H, ζ1 , ζ2 )]| = 0, |[D (νW, kW, ζ3 , ζ4 )]| = 0 (9.3.24)
The representations in Eq. (9.3.20) account for values of μ and ν that have opposite
sign, so the only roots for which Re (μ) ≥ 0 and Re (ν) ≥ 0 should be sought.
The transverse mode function is denoted as n, j and referred to as the (n, j)
mode. Only the ratios B2y /B1y and B2z /B1z can be obtained from Eq. (9.3.22) because
the respective [D] matrix is rank-deficient when μH is an eigenvalue. We use the
first row of [D] to determine these ratios. Furthermore, when the y and z functions
are multiplied, only one arbitrary coefficient, which we denote as Bn , remains. The
result is that the transverse mode functions may be written as a product of functions
of y and z, which we denote respectively as φ y,n and φz, j . Thus, we have established
that
n, j (y, z) = Bn, j φ y,n (y) φz, j (z) (9.3.25)
where the factors are
(ζ1 μn H − k H ) e−iμn y + (ζ1 μn H + k H ) eiμn y

φ y,n =
(ζ1 μn H − k H ) (9.3.26)
ζ3 ν j W − kW e−iγ j y + ζ3 ν j W + kW eiν j y
φz, j =
ζ3 ν j W − kW
After the eigenvalues for each direction have been determined, the set of axial
wavenumbers are 1/2
κn, j = ± k 2 − μ2n − ν 2j (9.3.27)

where
the alternative sign should be selected such that Re κn, j > 0 and
Im κn, j ≤ 0. If all walls are reactive, then the μn and ν j values are either real
or purely imaginary. In such cases cutoff occurs if μ2n + ν 2j > k 2 . A corollary is that
there is no single cutoff. Rather, for a fixed mode number in one direction there is a
mode number for the other direction beyond which the signal evanesces. For exam-
ple, if μ5 = 0.8k, then modes are cutoff in the z direction if ν j > 0.6k. It follows
that if μn > k, then all modes at that n and higher are evanescent, regardless of the
value of ν j . Similarly, if ν j > k, then all modes at that j and higher are evanescent.
The alternative situation occurs when one or more wall has a resistance, so that its
impedance has a real part. In such circumstances all modes decay axially, so that
the demarcation between propagating and evanescent modes does not exist. It is rea-
sonable to consider
modes to be cutoff if the scale over which they decay, which
is

−1/ Im κn, j , is comparable to or smaller than an axial wavelength, 2π/ Re κn. j .
Such a transition will be a noticeable feature of dispersion curves if the resistive part
of the wall impedance is small, as it was in Example 9.4. On the other hand, it might
be meaningless to characterize a mode in this manner if the resistance is large.
Like a two-dimensional waveguide, a waveguide mode in a rectangular waveguide
may be viewed either as a nonuniform plane wave that travels in the x direction with
the transverse dependence described by n, j , or as a superposition of a set of plane
waves that travel obliquely to the x-axis. Merging Eqs. (9.3.17) and (9.3.20) for a
specific (n, j) pair shows that a waveguide mode is a sum of four simple plane waves
described by
'
4
n, j e−iκn, j x = Cm e−i k̄m ·x̄
(9.3.28)
m=1
k̄m = κn, j ēx ± μn, j ē y ± νn, j ēz
where the Cm are set by satisfying the boundary conditions at the walls. If κn, j is
real, the four plane waves that combine to form a waveguide mode undergo multiple
reflections at the side walls. Their interference pattern on a cross section produces
the transverse mode function. This is the same phenomenon as the one illustrated
in Fig. 9.9, except that each k̄m has nonzero direction angles in both the x y- and
x z-planes.
The general orthogonality property, Eq. (9.3.8), in the case of a rectangular cross
section states that H W
n, j m, dzdy = H W δn.m δ j. (9.3.29)
0 0
The case where m = n and = j sets the coefficient Bn, j in Eq. (9.3.25). Substitution
of that expression into the orthogonality condition leads to
−1/2 −1/2
H 2 W 2
Bn, j = (H W ) 1/2
φ y,n dy φz, j dz (9.3.30)
0 0
We have seen that the factors φ y,n and φz, j in Eq. (9.3.25) are transverse mode
functions for a two-dimensional waveguide. This can be a very useful attribute,
because it might be quite simple to form the transverse mode functions for a three-
dimensional system. For example, consider the three-dimensional transverse mode
functions of the tank in Example 9.6. In the y direction, it is a rigid/pressure-release
configuration, whereas it is a hard-walled waveguide in the z direction. Multiplying

a mode function for each gives

(2n − 1) π y jπz
n, j = Bn, j cos cos , n = 1, 2, ..., j = 0, 1, ... (9.3.31)
2H W
A system of particular interest is a hard-walled waveguide, that is, one whose four
walls are rigid. The combination of cosine functions in each direction gives
nπz
jπ y
n, j (y, z) = Bn, j cos cos , j, n = 0, 1, 2, ... (9.3.32)
H W
The corresponding eigenvalues and axial wavenumber are

nπ jπ nπ 2 jπ 2 1/2
μn = , νj = , κ j,n = ± k −
2
− (9.3.33)
H W H W
The alternative sign in this expression should be selected such that Re (κn ) > 0 and
Im (κn ) ≤ 0.
The case where j = n = 0 is the plane wave mode, in which 0,0 is independent
of y and z. The hard-walled waveguide is the only one in which a plane wave mode
can exist. This is so for the same reason as a two-dimensional waveguide: In the
plane wave mode n̄ · ∇ P is identically zero at the walls, which conflicts with the
proportionality of pressure and velocity imposed by a finite local impedance.
Examination of Eq. (9.3.33) leads to recognition of an unusual feature. Suppose
that the cross section is square, W = H . Then any (n, j) mode for which n = j has
the same axial wavenumber as the ( j, n) mode. The same situation can arise even
if W = H . To explore this possibility, let W = r H , and let ( j1 , n 1 ) and ( j2 , n 2 ) be
the numbers for a pair of modes that have the same κ value. Repeated values of κ j,n
occur if 2 2
j2 j1
(n 2 )2 + = (n 1 )2 + (9.3.34)
r r
This condition is met if n 2 = j1 /r and j2 = r n 1 . Because the mode indices are

integers, repeated kn, j values can arise only if r is a rational number, that is, if
W/H is a ratio of integers. Furthermore, unless this ratio consists of small numbers,
the overlapping modes will be very high, and possibly evanescent. For example
if r = 3/2, then κ3,2 = κ2,3 , κ6,4 = κ4,6 , etc. In contrast of r = 20/21, then
κ20,21 = κ21,20 , etc. Similar situations might arise for other combinations of rigid
and pressure-release boundary conditions. Equality of the axial wavenumbers for
different modes does not present a complication for the analysis. However, it can
cause confusion for an examination of measured data.
After we have evaluated the transverse mode functions, the response to an exci-
tation may be analyzed. The modal summation in Eq. (9.3.9) now must extend over
the index for each direction, that is,
∞ '
' ∞
P= Pn, j n, j e−iκn, j x (9.3.35)
n=0 j=0
As always, a zero subscript applies only if a plane wave mode exists, in which case
κn, j = k. Similarly, if there is an eigenvalue that is purely imaginary, its contribution
must be inserted. The velocity excitation Vs (x, y) extends over the cross section at
x = 0. Its modal series also is a double sum,
∞ '
' ∞
Vs = Vn, j n, j (9.3.36)
n=0 j=0
The modes form an orthogonal set according to Eq. (9.3.29). This leads to the specific
form of Eq. (9.3.12),
W H
Bn, j
Vn, j = Vs φ y,n φz, j dydz (9.3.37)
HW 0 0
These coefficients are matched to the modal series for Vx at x = 0 obtained from
Euler’s equation, as was done to obtain Eq. (9.3.13), with the result that
k
Pn, j = ρ0 c Vn, j (9.3.38)
κn, j
The considerations regarding convergence for two-dimensional systems are

equally applicable here. The largest y mode number Ncut for a propagating mode
is such that μ Ncut < k and μ Ncut +1 > k. Similarly, the z mode number for cutoff is
such that ν Jcut < k and ν Jcut +1 > k. In regions that are not close to the excitation at
x = 0, truncation of the modal series at n = Ncut and j = Jcut will be adequate.
This is so because the modes that are omitted correspond to κn, j = −i Im κn , j ,
so that exp(−iκn, j x) is negligible. In regions very close to the excitation at x = 0,
the evanescent modes have not decayed much. Convergence of the pressure series
requires inclusion of the evanescent modes. The values of κn, j increase rapidly if
n > Ncut + 1 or j > Jcut + 1. A truncation at N = 2Ncut and j = 2Jcut should
be adequate, unless x is smaller than a wavelength. For such cases, truncation at
3Ncut and 3Jcut will suffice.
EXAMPLE 9.7 Consider the tank in Example 9.6 in the situation where the
ribbon transducer is mounted on the vertical center line at x = 0, rather than
the horizontal center line. All parameters are as stated there, and W = 0.5
m. Graph the dependence of |P| on the axial distance along the center line
y = H/2, z = W/2. Also, graph the profile of |P| for cross sections at x = 0
and x = L along the horizontal line y = H/2 and the vertical line z = W/2.
Significance
The number of modes is the product N J of the series lengths, which might be quite
large. Thus, it is desirable that the computational algorithm be efficient. The need to
use two indices to specify a mode complicates development of such an algorithm.
Both issues are addressed.
Solution
We shall use the same coordinate system as that in Example 9.6. Unlike the previous
arrangement, the ribbon transducer’s velocity distribution varies in the z direction.
The consequence is that more than one mode in that direction is excited. Furthermore,
although the transducer velocity now is independent of y, there is no plane wave mode
for this direction. Consequently, all modes in that direction participate, as they did
in the previous example. The y dependence of the transverse mode function is the
same rigid/pressure-release combination as it was in the previous Example, while the
z dependence consists of the modes of a two-dimensional hard-walled waveguide.
Thus, the transverse mode functions are

(2n − 1) π y jπz
n, j = Bn, j cos cos , n = 1, 2, ..., j = 0, 1, 2, ...
2H W
The corresponding eigenvalues are
(2n − 1) π jπ
μn = , νj = (1)
2H W
The axial wavenumber is either a positive real value or a negative imaginary value,
⎧ 1/2
⎨ k 2 − μ2 − ν 2 if μ2n + ν 2j < k 2
n j
κn, j = 1/2 (2)
⎩ −i μ2 + ν 2 − k 2 if μ2n + ν 2j > k 2
n j
The value of Bn is set by the normalization in Eq. (9.3.30), which leads to

√
Bn, j = 2 if j = 0, otherwise Bn, j = 2 (3)
Alignment of the transducer in the vertical direction means that the axial velocity
everywhere on the cross section at x = 0 is zero, except along the line z = W/2.
Thus, the normal velocity at x = 0 is Vs = Q̃ 0 δ (z − W/2). The corresponding
velocity coefficients are indicated by Eq. (9.3.37) to be
Bn, j Q̃ 0 (H
Vn, j = φz, j (z = W/2) 0 Vs φ y,n dy
H
(4)
2Bn, j Q̃ 0 (2n − 1) π jπ
= sin cos
(2n − 1) πW 2 2
The series length is set by the largest cutoff mode numbers for both directions,
which are Ncut = floor((k H/π+1/2) and Jcut = floor(kW/π). For the present system
parameters ( f = 20 kHz, c = 1480 m/s, H = 0.6 m, W = 0.5 m), these values are
Nev = 16, and Jev = 13. We wish to construct a scan along locations that are close
to the boundary, so evanescent modes must be included. Truncation of the series at
n = 3Nev and j = 3Jev should be adequate. In principle, we could reduce these
numbers as x increases. However, the algorithm that follows is quite efficient, so
there is no need to reduce the computational effort.
We have seen in a number of contexts that it is desirable to evaluate a modal-
type series by the matrix algorithm employed in Example 9.6. Doing so enables the
computation to be carried out in a vectorized manner. To use that algorithm here we
must arrange the modes in a sequence described by a single subscript. Let us define
an index m = 1, 2, ... that is in a one-to-one correspondence to the n, j indices of
a transverse mode function. This can be done by incrementing m as n is increased
with j fixed, then repeating the process with j increased by one, until the range of n
is exhausted. Because 1 ≤ n ≤ N and 0 ≤ j ≤ J , the rule is
m = n + jN (5)
Integer arithmetic allows us to identify the n and j values associated with a specific
m. If we divide the preceding by N we find that m/N ≥ j, which leads to

m−ε
j = floor , n = m − jN (6)
N
where ε is a very small number required to handle round-off error.

This arrangement allows us to implement the algorithm described by Eq. (4) of
Example 9.6. The first step is to rewrite the modal series in a single subscript form.
The number of modes is N (J + 1), so the series becomes
(J +1)
N'
k
P ≈ ρ0 c Vm m (y, z) e−iκm x
m=1
κm
To convert this to matrix form let the diagonal array [K ] hold the factors in the sum
that do not depend on position,
k
K m,m = Vm (7)
κm
The exponential terms are placed in [E], with a row holding the terms for all m at
fixed x. The number of rows is the number of x locations at which the pressure will
be computed. Thus,
E q,m = exp(−iκm xq ) (8)
The [F] array holds the values of the transverse locations at selected points on a cross
section. In any case a column of [F] holds the m values for all m at a designated
y and z. If we are interested in the pressure distribution over a range of y locations
at a fixed z, then [F] is formed by adjoining columns for a sequence of y values.
Similarly, if our interest is in a range of z locations at fixed y, then the adjoined
columns correspond to the successive z values. Thus,
Fm,r = m (yr , z) or Fm,r = m (y, zr ) (9)
In either case, the pressure evaluation is
[P] = ρ0 c [E] [K ] [F] (10)
The computational reduction achieved by this algorithm compared to one that actually
accumulates sums is quite substantial. Indeed, the computations here, for which the
total number of modes is 1920 and the number of points along the axial centerline
was 1000, required 0.60 s to execute on a laptop computer with an Intel I7 processor
and 8 GB of RAM. Truncating the modal series at 2Ncut and 2Jcut reduced this time
to 0.37 s.
A few details of the MATLAB program are worth reviewing. The vector {m} =
{1 2 · · · N (J + 1)} consists of the single mode indices. Vectors {n} and { j}, which
hold the respective mode numbers corresponding to each element of {m}, were set
according to Eq. (6). (The length of {n} and { j} is the same as the length of {m} .)
The eigenvalues corresponding to {n} and { j} are placed in arrays {μ} and {ν}. The
κm values also are computed in a vectorized manner. In MATLAB, this operation
is kappa=conj(sqrt(kˆ2+mu.ˆ2-nu.ˆ2)), where “.ˆ” indicates that the
square should be done element by element. (The use of the conjugate here is dictated
by MATLAB’s implementation of the branch cut for a square root.) The vector {κ}
is used to compute the successive rows of [E], with x held constant in each row. The
evaluation of [F] is done in a row-wise manner using {μ} and {ν}, with the y and z
values fixed in a column.
Figure 1 shows |P| as a function of x along the geometric centerline, y = H/2,
z = W/2. The overall pressure amplitude is comparable to the result in Example
9.6. That is, the pattern of relatively broad peaks separated by narrow antinodes is
like the preceding result. This is to be expected because Eq. (9.3.28) shows that the
signal is a superposition of plane waves, just as it is in the two-dimensional case.
30
25
|p| (kPa)
20
15
10
5
0 0.2 0.4 0.6 0.8 1

x/L
Figure 1.
Profiles of |P| along a cross section appear in Fig. 2. The upper pair describe
x = 0 and the lower pair are the distributions at x = 3 m. The vertical axis of the
graphs on the right depicts the y dependence because that is the way the y-axis is
physically aligned in Example 9.6. Here too we see interference patterns
40 1
y/H, x = 0, z = W/2
30
|p| (kPa)
20 0.5
10
0 0
0 0.5 1 0 5 10 15 20
z/W, x = 0, y = H/2 |p| (kPa)
20 1
y/k, x = L, z = W/2
15
|p| (kPa)
10 0.5
0 0
0 0.5 1 0 5 10 15 20
z/W, x = L, y = H/2 |p| (kPa)
Figure 2.
resulting from superposing the contributions of numerous modes. At x = 0 the

scan along z = W/2 shows the pressure on the transducer. Interference leads to
fluctuations of |P| on the face of the transducer, even though Vx is constant. The
scan along y = H/2 shows a spike of |P| in the vicinity of the transducer. This is
like the behavior on the face of a translating piston, where a uniform particle velocity
generates a fluctuating face pressure. Neither pattern is evident at the farther cross
section because the location is many wavelengths from the transducer.
The symmetry properties of these profiles is important. The waveguide’s shape is
symmetric with respect to the vertical and horizontal mid-planes, that is, the planes
that contain the axial centerline. The pressure does not share this symmetry, because
the upper boundary is pressure-release, whereas the lower boundary is rigid. How-
ever, it is symmetric with respect to the vertical mid-plane because the walls to the
left and right are rigid, and the transducer properties are the same on either side of
that plane. This symmetry is manifested in Eq. (4), which indicates that the Vn, j
velocity coefficients are identically zero if j is odd. As a result, all modes in the
vertical direction are excited, but only the even modes for the horizontal dependence
are excited. The consequence is that a scan of |P| along a horizontal line is an even
function with respect to the middle, whereas a scan along a vertical line shows no
pattern repetition. We could have exploited this symmetry by limiting the evaluation
of |P| to the range 0 ≤ z ≤ W/2 and only including the even functions of z in the
modal series.
9.3.3 Circular Waveguide
Circular waveguides are commonly used in such diverse applications as HVAC and
nuclear reactor cooling systems. Cylindrical coordinates, with the axial direction des-
ignated as x, are appropriate to this configuration. Propagation in a circular waveguide
is the complement of radiation from a vibrating cylinder. Both are governed by the
Helmholtz equation in cylindrical coordinates. The difference is that the waves prop-
agate in the transverse direction in the radiation problem, whereas the propagation is
axial for a waveguide. Cylindrical Bessel functions are part of the general solution
of the Helmholtz equation in cylindrical coordinates. Consequently, much reference
will be made to Sect. 7.3.2, where the fundamental properties of these functions are
described.
Fundamental Solution for a Waveguide Mode
The pressure is taken to be a wave that propagates in the axial direction with
wavenumber κ. The complex amplitude is allowed to depend on the radial distance R
and circumferential angle θ. The θ dependence is set by the fact that (R, θ ± 2π, z)
and (R, θ, z) designate the same point, so that any signal must be expressible as a
Fourier series in θ with a period of 2π. We shall use a complex exponential to express
the θ-dependence of a term in this series. Thus, the ansatz we have identified is

p = Re f (R) e−inθ e−iκx eiωt (9.3.39)
The choice of κ as the axial wavenumber is consistent with the general analysis in
Sect. 9.2.1. However, the variable for the transverse dependence will be μR. This
notation is opposite the usage in Sect. 7.3, where the transverse wavenumber of a
cylindrical wave was κ and the axial wavenumber was μ. One should bear this in
mind when reference is made to prior developments. Another difference from the
analysis of radiation is that the axial wavenumber was taken to be known as part of
the specification of surface motion, whereas here it is a parameter to be determined.
Substitution of the assumed form of p into the Helmholtz equation in cylindrical
coordinates leads to

d2 f 1 df n2
+ + μ2
− f =0 (9.3.40)
d R2 R dR R2
where
μ2 = k 2 − κ2 (9.3.41)
This is Bessel’s equation, which we previously encountered as Eq. (7.3.8). Its fun-
damental solutions are the Bessel function Jn (μR) and the Neumann function
Nn (μR) , so the general solution is
f = Bn Jn (μR) + Cn Nn (μR) (9.3.42)

The Neumann function is singular at R = 0, so it is discarded by setting Cn = 0.

Satisfaction of the boundary condition on the cylindrical wall will be seen to
require specific values of μ; these are the eigenvalues. These values will depend on
which azimuthal harmonic is being considered. Each μ value will lead to a different
κ, so the modes are dispersive. With n restricted to be positive or zero, the waveguide
modes and associate transverse mode functions are given by
P = ⎧(R, θ) e−iκx
⎨ B Jn (μR) e−inθ
(9.3.43)
(R, θ) = or
⎩
B Jn (μR) einθ
Adding and subtracting these fundamental solutions yields sine and cosine functions
for the azimuthal dependence. Thus, an alternative form of the waveguide modes is
(R, θ) e−iκx
P =⎧
⎨ B Jn (μR) cos (nθ)
(9.3.44)
(R, θ) = or
⎩
B Jn (μR) sin (nθ)
The difference between Eqs. (9.3.43) and (9.3.44) is that the former describes a pair
of helical waves that spiral about the x-axis in opposite senses, whereas the latter
describes standing wave patterns around the x-axis. The sine and cosine functions
are 90◦ out-of-phase and therefore constitute separate solutions. In the special case
where n = 0, the two helical wave coalesce to a plane wave that propagates in the
axial direction, and the sine mode vanishes.
Just as the pattern in the circumferential direction may be viewed as either oppo-
sitely traveling waves or standing waves, the same is true for the transverse direction.
According to Eq. (7.3.11), the Bessel function is equivalent to the sum of both kinds
of Hankel functions,
1 (1)
Jn (μR) = H (μR) + Hn(2) (μR) (9.3.45)
2 n
Thus, any of the descriptions of the modes may equivalently be considered to be
waves that propagate inward and outward from the center. Each wave individually
would be singular at R = 0, but their singularities cancel.
In addition to finiteness, any pressure field must be continuous at R = 0. This
means that the same pressure must be obtained as a field point approaches R = 0
along any θ, which is stated as
lim P (R, θ, x) = g (x) (9.3.46)

R→0
where g (x) is used to indicate that the limit may depend on x. Continuity of ∇ P
also is a fundamental requirement. Otherwise, ∇ 2 P will be infinite, which according
to the Helmholtz equation corresponds to infinite P. As R → 0, the ēθ direction
becomes meaningless, so we may focus on the behavior of ∂ P/∂ R at R = 0.
Because ē R (θ + π) = −ē R (θ), the transverse derivative at the origin must change
sign when θ is incremented by 180◦ ,

∂ P ∂ P
=− (9.3.47)
∂ R R=0,θ±π ∂ R R=0,θ
Let us assess whether these conditions are satisfied by the waveguide modes.
All modes have the form n (R, θ) exp(−iκx) with n (R, θ) = f R (R) f θ (θ).
The transverse dependence is f R (R) = Jn (μR) ≈ (μR/2)n /n! when R is small,
and f θ (θ) = exp (−inθ). For n > 0, f R = 0 at R = 0, so the g function is
zero independent of x. Thus, the modes for n ≥ 1 are continuous. For n = 0, we
have f R = 1 at R = 0 and f θ = 1, so the n = 0 modes also are continuous. In
regard to continuity of the gradient, for n ≥ 1, the radial derivative at the center
is ∂n /∂ R = (μR/2)n−1 exp (−inθ) / (n − 1)! evaluated at R = 0. This quantity
is zero for n ≥ 2, which satisfies Eq. (9.3.47). For the n = 1 modes, we have
∂n /∂ R = exp (−iθ) at R = 0, which also satisfies Eq. (9.3.47). The n = 0 modes
also are such that ∂0 /∂ R = 0 at R = 0. Thus, the Bessel function modes satisfy
the continuity requirements at the origin.
Figure 9.14a and b, respectively, shows the behavior of the n = 0 and n = 1
waveguide modes in the vicinity of R = 0. These are contrasted by Fig.
9.14c,
which shows a discontinuous pressure field described by J0 (μR) Re eiθ , and by
Fig. 9.14d, where J1 (μR) has a discontinuous gradient at R = 0.
(a) (b)
(c) (d)
Fig. 9.14 Behavior of pressure fields in the vicinity of R = 0: a The n = 0 propagation mode,
P = B J0 (μR). b The n = 1 propagation mode, P = B J1 (μR) e−iθ . c A discontinuous pressure
field, B J0 (μR) e−iθ . d A field having a discontinuous gradient, P = B J1 (μR)
The radial wavenumber is set by the boundary condition at the wall, where R = a.
A positive value of VR corresponds to movement into the wall, so an impedance
boundary condition corresponds to
ζ ∂P
P = ρ0 cζ VR = − at R = a (9.3.48)
ik ∂ R
When we substitute the general solution for a waveguide mode into this condition,
the exponential factors cancel, as does the amplitude coefficient B. What remains is
the characteristic equation,
iζ
Jn (μa) − (μa) Jn (μa) = 0 (9.3.49)
ka
Regardless of the nature of ζ, the characteristic equation for a specified value of

the circumferential number n will have an infinite number of roots. These nondi-
mensional roots of the characteristic equation are denoted as ηn, j . The corre-
sponding eigenvalues are μn, j = ηn, j /a. The eigenvalues might be imaginary or
complex,
so we shall
arrange
them sequentially in increasing order of their real part,
Re μn,1 < Re μn,2 < ... The manner in which the characteristic equation has
been written makes it evident that the ηn, j values depend only on ζ and ka.
Numerical methods will be required to obtain accurate roots of the characteristic
equation. Few software packages offer routines that evaluate derivatives of Bessel
functions. A recurrence relation in Eq. (7.3.15) gives

n
Jn (μa) = −Jn+1 (μa) + Jn (μa) (9.3.50)
μa
This expression may be incorporated as an auxiliary routine, but it is just as easy to

incorporate it into the characteristic equation, which becomes

iζ n
Jn (μa) − (μa) −Jn+1 (μa) + Jn (μa) = 0 (9.3.51)
ka μa
The methods that are commonly used to solve transcendental equations require
initial guesses of the roots ηn, j = μn, j a. If ζ is infinite, zero, or purely imaginary, the
equation is real. Guesses for the lower roots in that case may be obtained by graphing
the characteristic equation as a function of μa. For large μa (μa > 8 is reasonable),
we may employ the asymptotic representation of Jn (μa) for large arguments
1/2
2 2n + 1
Jn (μa) → cos μa − π (9.3.52)
πμa 4
When μa is large, increasing μa changes the square root factor much less than the
cosine term. It follows that for large μa, the characteristic equation is approximately
periodic in (μa) = 2π. The consequence of this property is that after we have
found a large eigenvalue μn a 1, a good starting guess for the next value is
μn+1 a = μn a + π.
Two indices are required to uniquely specify which eigenvalue is under consider-
ation: the circumferential harmonic number n and the root number j, which gives the
sequential placement of the root within the set of eigenvalues at that n. Specification
of a transverse mode function requires a third index, denoted as α. This index is used
to specify the dependence on θ. We will adopt the convention that if this subscript is
α = 1 or α = −1, we are using helical waves, such that

n, j,α = Bn, j,α Jn μn, j R eiαnθ (9.3.53)
Alternatively, if we wish to use real sinusoidal terms to describe the θ dependence,

we would set α = c or α = s, such that

n, j,c = Bm,n,c Jn μn, j R cos (nθ)
(9.3.54)
m,n,s = Bn, j,s Jn μn, j R sin (nθ)
The discussion thus far covers the generalities. However, examination of specific
impedance cases discloses issues that are not yet evident, as well as some significant
behaviors.
Rigid Walls
The model of a rigid wall serves as a reference system for discussing other configu-
rations, as well as being useful for its own sake. Satisfaction of Eq. (9.3.49) when ζ
is infinite requires that Jn (μa) = 0. Thus, the eigenvalues correspond to locations
at which Jn (μa) has an extreme value. Figure 9.15 plots some low-order Bessel
functions and shows the smallest roots ηn, j for each n.
0,1 0,2 0,3

1,1 1,2 1,3
2,1 2,2
1
n
Jn(a)
0.5
n
n
0
−0.5
0 2 4 6 8 10
a
Fig. 9.15 Characteristic equation for a circular waveguide with rigid walls
The first few eigenvalues for a rigid-walled cylinder are listed in Table 9.1.
The axial wavenumbers are defined by Eq. (9.3.41). As was true for the previ-
ous situations, a large eigenvalue will lead to cutoff of the axial propagation. The
possibilities are
Table 9.1 Roots of the characteristic equation for a rigid-walled cylindrical waveguide
ηn, j ≡ μn, j a j =1 j =2 j =3 j =4
n =0 0 3.8317 7.0156 10.1735
n =1 1.8412 5.3314 8.5363 11.7060
n =2 3.0542 6.7061 9.9695 13.1704
n =3 4.2012 8.0152 11.3459 14.5858
⎧
⎨ k 2 − ηn, j /a 1/2 if ηn, j < ka
κn, j = 1/2 (9.3.55)
⎩ −i ηn, j /a 2 − k 2 if ηn, j > ka
Because the eigenvalues are roots of a transcendental equation, it is not possible to

identify a priori how many eigenvalues are below the cutoff value. Nevertheless,
the tabulation indicates that the eigenvalues at a fixed root number increase with
increasing n. Furthermore, by definition, the eigenvalues at fixed n increase with
increasing root number, Thus, it is sufficient to identify the root number J for n = 0
at which η0,J > ka, and the circumferential harmonic N at which η1,N > ka.
Using these as the cutoff numbers assures that all propagating modes are included.
Figure 9.16 shows contours of some typical transverse mode functions. Bold contours
η0,1 = 0 η0,2 = 3.832 η0,3 = 7.016
η1,0 = 1.841 η1,2 = 5.331 η1,3 = 8.536
η2,1 = 3.054 η2,2 = 6.706 η1,0 = 9.969
Fig. 9.16 Contours of the transverse mode functions for a rigid-walled circular waveguide. An
ηn, j value is the jth root of the characteristic equation corresponding to circumferential harmonic
number n
are zeros, which separate regions of positive and negative pressures. The patterns
describe the cosine functions, n, j,c ; the sine functions would be rotated by π/ (2n).
Compliant Walls
Many features exhibited by the modes in the rigid-walled case are typical of those
for other wall conditions. Equation (9.3.43) or the alternative form in Eq. (9.3.44)
describes any case. A corollary of the occurrence of ka in the characteristic equation,
Eq. (9.3.49), is that it will be necessary to recompute its roots if the frequency of
interest changes, for example, to perform a frequency sweep in order to determine a
dispersion curve. The only exceptions are rigid or pressure-release walls.
If ζ is finite and nonzero, then neither term in the characteristic equation is zero. It
follows that unless ζ is purely imaginary, the equation is complex. This means that the
eigenvalues must be complex, which is a possibility we will address. The exceptional
case is that where ζ is purely imaginary, corresponding to a purely reactive wall. To
analyze that case, we set ζ = σi, with σ > 0 for an inertance and σ < 0 for a
compliance. The corresponding form of the characteristic equation is
μa
Jn (μa) + σ J (μa) = 0 (9.3.56)
ka n
The Bessel function is oscillatory, from which it follows that this equation has an
infinite number of roots ηn, j = μn, j a, j = 1, 2, ...
Let us consider the behavior of the eigenvalue for a fixed circumferential harmonic
n and root number j as the frequency increases. The nondimensional parameter
affecting the eigenvalues is σ/ (ka). The case of a pressure-release wall, which is the
limit as |ζ| → 0, corresponds to Jn (μa) = 0. In contrast, ζ = ∞ gives Jn (μa) = 0,
which is the characteristic equation for a rigid wall. It follows that this eigenvalue
will be close to that of a rigid waveguide at very low frequencies, while it will be
close to the eigenvalue of a pressure-release configuration for high frequencies. A
different trend is that in which n and ka are held fixed. By definition, increasing j
corresponds to increasing eigenvalue. Because μa multiplies σ in the characteristic
equation, the higher eigenvalues will approach those for a rigid wall regardless of the
size of σ. We previously encountered this trend in the analysis of a two-dimensional
waveguide whose walls are reactive.
Like a two-dimensional waveguide, a circular waveguide whose wall is purely
reactive might have a purely imaginary eigenvalue. To explore this possibility, we
set μ = iβ, β > 0. Rather than dealing with Bessel functions whose argument is
imaginary, it is preferable to convert Jn (μa) to the first kind of modified Bessel
function, which is given by Eq. (7.3.33) to be
Jn (iβa) = i n In (βa) (9.3.57)

Substitution of this expression into the third recurrence relation for a derivative of a
Bessel function in Eq. (7.3.15) gives

n
Jn (iβa) = i n−1
In+1 (βa) + In (βa) (9.3.58)
βa
Substitution of these relations into the characteristic equation converts it to

βa n
In (βa) + Im (ζ) In+1 (βa) + In (βa) = 0 (9.3.59)
ka βa
One feature is immediately apparent. Each of the modified Bessel functions appearing
above is positive for βa > 0. It follows that a real root of this characteristic equation
can exist only if Im (ζ) < 0. In other words, a mode whose eigenvalue is purely
imaginary exists only if the wall impedance is a pure compliance. This condition is
consistent with the previous result for a two-dimensional waveguide. We will say
that this is the (n, im) mode, so μn,im = iβn,im a.
To get a more detailed picture, let us consider the modal properties for σ = ±5
(ζ = ±5i), n = 0, and ka = π. The characteristic equation for real eigenvalues,
Eq. (9.3.56), and an imaginary eigenvalue, Eq. (9.3.59), are depicted in Fig. 9.17. The
function that is plotted for imaginary eigenvalues is the above divided by I0 (βa).
(This is done because In (βa) increases exponentially with increasing βa, which
would make it difficult to see the location where the plotted function is zero.) The
intersection of each curve with the zero axis marks a root.
10 8
 = 5i  = 5i
 = 5i = 5i
5
4
0
0
−5
−10 −4
0 5 10 15 20 0 0.5 1 1.5 2
a a/i=
Fig. 9.17 Characteristic equation for a circular waveguide whose wall is compliant, ka = π, n = 0
Each eigenvalue in the range μa > 3 for σ = −5 has a close companion for
positive σ. The exception is the first eigenvalue for σ = 5, μ1 a = 1.215, whereas
no real eigenvalue for σ = −5 that is that small. However, the imaginary eigenvalue
for σ = −5 also has a small magnitude, which suggests that they might have some
similarities. The first few eigenvalues and corresponding axial wavenumbers are
listed in Table 9.2. In both reactance cases, only one mode is propagative. As ka is
increased, more and more modes enter the spectrum that propagates.
Table 9.2 Eigenvalues of a cylindrical waveguide with a reactive wall, ka = pi

j = im j =1 j =2 j =3
ζ = −5i η0, j ≡ μ0, j a 1.2146i 3.6657 6.9257 10.1116
κ0, j a 3.368 −1.8888i −6.1722i −9.6112i
ζ = 5i η0, j ≡ μ0, j a − 1.0387 3.9910 7.1044
κ0, j a − 2.9649 −2.4614i −6.3720i
The transverse mode functions for an inertance and a compliance are depicted
in Fig. 9.18. The functions associated with real eigenvalues μ0, j = η0, j /a are like
those in Fig. 9.16. This is to be expected because ζ = ±5i is effectively quite close
to the rigid condition. However, a finite impedance requires that dn, j /d R = 0 at
R = a, so the maxima, minima, and zeros of the R-dependence are shifted. The
single imaginary eigenvalue, μn,im = iηn,im /a, corresponds to a mode function that
is reminiscent of the plane wave mode. However, the first mode for ζ = 5i also
shows relatively little variation along a transverse lines, so it is too reminiscent of
the plane wave mode.
1
0, im
0.5
0, 1
0  = 5i 0, 2
−0.5
1
0, 1
0.5 0, 2
0  = 5i 0, 3
−0.5
0 0.2 0.4 0.6 0.8 1
R/a
Fig. 9.18 Transverse mode functions for a circular waveguide whose wall is reactive, ka = π,
n = 0, ζ = −0.2i
EXAMPLE 9.8 An annular waveguide consists of the region between two

concentric cylinders. The outer radius is a, and the inner radius is b. Consider
the case where both cylinders are rigid. Derive the characteristic equation and
an expression for the transverse mode functions. Evaluate the n = 0, n = 1,
and n = 2 modes for the case where b/a = 0.4. Also, determine the highest
frequency for which all nonplanar modes evanesce.
Significance
Coaxial pipes are used in some systems to convey fluid back and forth, especially if the
surroundings are narrow. The analysis suggests how the formulation for a complete
cylindrical cross section may be generalized. To some extent, it is a precursor of the
analyses of a cavity, which is the topic of the next chapter.
Solution
The transverse mode function must satisfy the rigidity condition at both walls, which
are
∂
= 0 at R = a and R = b
∂R
Because R = 0 is not a point in the fluid, both the Bessel and Neumann functions
in the general solution for R dependence, Eq. (9.3.42), are appropriate. A trans-
verse mode function is the product of that function and the nth harmonic for the θ
dependence, so the form is
= [Bn Jn (μR) + Cn Nn (μR)] e−inθ (1)
The coefficients are found by making this general solution fit the boundary con-
ditions. The matrix form of the result of these operations is

Jn (μa) Nn (μa) Bn 0
= (2)
Jn (μb) Nn (μb) Cn 0
The characteristic equation, which results from setting the determinant of these equa-
tions to zero, is
Jn (μa) Nn (μb) − Nn (μa) Jn (μb) = 0 (3)
The frequency does not occur in this characteristic equation, which means that the
eigenvalues are independent of the frequency. Figure 1 graphs the characteristic equa-
tion as a function of μa with μb ≡ (b/a) μa.
(a)2[Jn(b)Nn(a)Nn(b)Jn(a)]
1
b/a=0.4
0.5
0
−0.5 n=0
n=1
−1 n=2
0 5 10 15 20
a
Figure 1.
The zero crossing of the curve for each n is used to initiate a numerical search
for the respective eigenvalues. The walls are rigid, so a plane wave mode may exist.
Thus, n = 0 is a plane wave, for which μ0,1 = 0 and 0,1 = 1. The result appears
in the following tabulation.
μn, j a
n j =1 j =2 j =3 j =4
0 0 5.3912 10.5577 15.7665
1 1.46178 5.6591 10.6833 15.8481
2 2.8424 6.4160 11.0560 16.0916
When μa is one of the eigenvalues, only one of the scalar equations described by
Eq. (2) is independent. Thus, only the ratio of the coefficients is described uniquely.
When we use the second coefficient equation, which is the boundary condition at
R = a, to evaluate C/B, we find that

Jn μn, j a −inθ
n, j,1 = Bn, j Jn μn, j R − Nn μn, j R e
Nn μn, j a
(Recall that e+inθ , cos (nθ), and sin (nθ) are equally valid descriptions of the
θ-dependence.) Figure 2 shows some transverse mode functions for b/a = 0.4. The
value of B0, j for each was selected such that the maximum of the respective function
is one. If we were to let s be the distance from the inner cylinder, it would not be a
terrible approximation to say that these modes fit cos ( jπs/ (a − b)), which are the
transverse modes of a hard-walled two-dimensional waveguide.
n=0
1 j=1
j=2
n,j
0 j=4
j=3
−1
n=1
1 j=1
j=2
n,j
0
j=4
j=3
−1
n=2
1 j=1
j=2
n,j
0
j=4
j=3
−1
0.4 0.5 0.6 0.7 0.8 0.9 1
R/a
Figure 2.
Signal Generation
Our interest here is the signal generated by the imposition of an axial particle veloc-
ity at x = 0. If the spatial pattern of the velocity imposed at x = 0 does not match
a specific mode, all modes are excited. Thus, we consider an arbitrary distribu-
tion, Vx = Re Vs (R, θ) eiωt at x = 0. The procedure for identifying the mix of
waveguide modes that is generated by this excitation follows the general outline laid
out in Sect. 9.3.1. The first task is to decide how to represent the θ-dependence of
the modes. Complex exponentials are suitable for an excitation that rotates about the
x-axis, which is typical of noise generation mechanisms in rotating machinery. How-
ever, most excitations have a fixed distribution in the θ direction, and trigonometric
functions share that property. Hence, we shall use the transverse mode functions in
Eq. (9.3.44),

n, j,c = Bn, j Jn μn, j R cos (nθ)
(9.3.60)
n, j,s = Bn, j Jn μn, j R sin (nθ)
In the usual situation the range of indices is n ≥ 0, j ≥ 1. This scheme is modified

if the wall is rigid by designating the first mode for n = 0 as j = 0 to represent
the plane wave mode. In addition, if the wall is purely compliant, the imaginary
eigenvalue at any n is designated with the label j = “im”. Also, for n = 0, B0, j,s are
set to zero.
According to the general orthogonality property in Eq. (9.3.8), it must be that
a π
n, j,α m,,β Rdθd R = πa 2 δn,m δ j, δα,β (9.3.61)
0 −π
The Kronecker delta term δα,β tells us that the sine and cosine modes are orthogonal,
even if their eigenvalues are the same. When the subscripts are the same, the orthog-
onality condition defines the normalization of the Bn, j,α coefficients. Substitution of
the mode function into Eq. (9.3.60) with μn, j = ηn, j /a, followed by integration over
θ, reduces the normalization condition to
a π 2
2 2 a
R
n, j,α Rdθd R = 1 + δn,0 π Bn, j Jn ηn. j Rd R = πa 2
0 −π 0 a
(9.3.62)
There is no need to evaluate this integral numerically. Changing the integration
variable to χ = R/a converts the equation to
2
1 + δn,0 Bn, j In, j = 1 (9.3.63)
where
1 2
In, j = Jn ηn, j χ χdχ (9.3.64)
0
Because ηn, j is a root of the characteristic equation, Eq. (9.3.56), it fits the condition
for which a standard formula2 is valid. The result depends on the nature of the
impedance according to
⎧
⎪ 1 2
⎪
⎪ Jn ηn, j if ζ = 0
⎨2
In, j = (9.3.65)
⎪
⎪ 1 ζ 2 n 2 + (ka)2 2
⎪
⎩ 1− Jn ηn, j if ζ = 0
2 ζ ηn, j
2 2
Thus, we have established that the normalizing coefficient is
1
Bn, j = 1/2 1/2 (9.3.66)
1 + δn,0 In, j
The incarnation of the general modal series, Eq. (9.3.9), is

∞ '
' ∞ '
P (x, R, θ) = Pn, j,α n, j,α e−iκn, j x (9.3.67)
n=0 j=1 α=c,s
Equation (9.3.13) describes the pressure coefficients in terms of the coefficients of a

modal series for the velocity input, Vs (R, θ). For the case of a cylinder, Eq. (9.3.12),
becomes
a π
1
Vn, j,c = Vs (R, θ) Jn μn, j R cos (nθ) Rdθd R
πa 0 −π
2
a π (9.3.68)
1
Vn, j,s = Vs (R, θ) Jn μn, j R sin (nθ) Rdθd R
πa 2 0 −π
The corresponding pressure coefficients are
k
Pn = ρ0 c Vn, j,α (9.3.69)
κn, j
Thus, the modal pressure series is
'∞ ' ∞
k
P (x, R, θ) = ρ0 c Vn, j,c cos (nθ) + Vn, j,s sin (nθ)
κn, j (9.3.70)
n=0 j=1
×Bn, j J0 μn, j R e−iκn, j x
2 M.I.Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, (1965) Eq. 11,
pp. 485.
The associated axial particle velocity is
∞ '
' ∞

Vx (x, R, θ) = Vn, j,c cos (nθ) + Vn, j,s sin (nθ) Bn, j J0 μn, j R e−iκn, j x
n=0 j=1
(9.3.71)
The overall procedure for determining the response begins with solution of the
characteristic equation for the eigenvalues. The normalizing coefficients Bn, j are
evaluated with the aid of Eqs. (9.3.65) and (9.3.66). Evaluation of the integrals for
the velocity coefficients in Eq. (9.3.68) might require a numerical algorithm. These
items are used to form the modal series, Eq. (9.3.70). The contribution from each
circumferential number n may be carried out sequentially. Truncation of the modal
series may be done based on the highest n for which κn,1 < k and the highest j for
which μ0, j < k, as was explained after Eq. (9.3.55).
EXAMPLE 9.9 A circular piston is mounted concentrically at one end of a

very long cylindrical tube filled with air. The other end is terminated with a
sand barrier that eliminates reflections. The tube is rigid, with a diameter of
100 mm, and the piston diameter is 50 mm. The piston oscillates at 40 kHz.
The vibration amplitude is such that the maximum on-axis pressure would be
110 dB//20μPa if the piston were mounted in an infinite baffle in free space.
Determine and graph the sound pressure on the axis of the waveguide in a
range from 20 mm to 1 m.
Significance
From an operational viewpoint, this example will illustrate some of the facets of
constructing a modal series. It also will demonstrate the drastic effect of confining a
sound beam.
Solution
Let V0 be the velocity amplitude of the piston. It is given that the piston is situated
concentrically, which means that the axial velocity it imposes is the same at all θ.
The specific function is

V0 , 0 ≤ R < b
Vx (x = 0) =
0, b < R ≤ a
where b = 0.025 m is the radius of the piston and a = 0.05 m.

Only the n = 0 modes are excited because Vx is an axisymmetric distribution.
The transverse mode functions of a hard-walled circular waveguide were found in
Sect. 9.3.3 to be

R
0, j = B0, j J0 η0, j
a
J0 η0, j ≡ −J1 η0, j = 0
The first four roots η0, j for the n = 0 modesare listed there in Table 9.1. For the
present parameters, c = 340 m/s, ω = 2.513 105 rad/s, a = 0.05 m, the highest
mode that is not cutoff is the twelfth, for which η0,12 = 35.33. The value of κ for this
mode and the first cutoff mode are κ0,12 = 216.95 m−1 and κ0,13 = −213.73i m−1 .
At x = 20 mm, the exponential decay factor for the latter is approximately 1.4%,
which clearly demonstrates that we need not be concerned here with evanescent
modes.
The scaling coefficients B0, j are given by Eq. (9.3.66). The specific impedance

and we are only interested in n = 0. Therefore, we may set J0 η0, j =
is infinite,

−J1 η0, j , which is squared in the aforementioned equation. Hence, the coefficients
are
21/2
B0,1 = 1, B0, j = if j > 1
J1 η0, j
The velocity coefficients are described in general by Eq. (9.3.68). For the above mode
functions, and zero Vx for R > b, that expression becomes
b π b π
V0 V0 R
V0, j = 0, j Rdθd R = B0, j J0 η0, j Rdθd R
πa 2 0 −π πa 2 0 −π a
Changing the integration variable to u = η0, j R/a leads to a form that is another
standard integral.3 The result is that
⎧
⎪
⎪ b2
⎨ V0 2 B0, j , j = 1
V0, j = a
⎪
⎪
b
η
b
, j >1
⎩ 0
2V B J
0, j 1 0, j
η0, j a a
The first few terms are V0, j /V0 = 0.25, −0.3763, 0.063695, 0.13228, ...
The value of V0 is set by the given pressure in a free space. The maximum on-axis
pressure in a sound beam is twice the plane wave value, max (|P|) = 2ρ0 cV0 . The
specified value of 110 dB corresponds to max (|P|) = 8.9443 Pa, from which we
find that V0 = 10.961 mm/s. The Rayleigh distance is kb2 /2 = 0.231 m. Hence, a
scan out to x = 1 m extends into the farfield for the infinite baffle case.
The on-axis pressure in the waveguide is obtained by evaluating the modal series,
Eq. (9.3.70) at R = 0. Because J0 (0) ≡ 1, the modal series reduces to
'
12
k
P (x, 0) = P0, j e−iκ0, j x , P0, j = ρ0 c V0, j
j=1
κn, j
3 M.I. Abramowitz and I.A. Stegun, ibid, Eq. (11) pp. 484.
Our standard technique

for computing a summation like this with x fixed is to define
column vector P j that holds the coefficients and row vector [E (x)] whose elements
are the complex exponentials for each j. To evaluate it at many x, we stack [E (x)]
for each location and compute
⎡ ⎤
[E (x1 )]
⎢ ⎥
[P] = ⎣ [E (x2 )] ⎦ P j
..
.
The result is shown in Fig. 1. The on-axis pressure for the case where the piston
is mounted on an infinite baffle also is shown there. The maximum pressure in the
nearfield is much less than it would be in the infinite baffle case, but the pressure in
the waveguide does not show spherical decay. Both the large fluctuations over short
distance and the weaker signal in the closer ranges have the same explanation. The
axial signal for a baffled piston results from constructive and destructive interference
between equal strength center and edge waves. In contrast, we found that waveguide
modes may be represented as waves that propagate in and out in the transverse
direction as they propagate downstream. There are many modes, which means that
there are many waves having different amplitudes. There is no location at which all
of these waves are in-phase to create a maximum.
10
6
|P| (Pa)
0
0 0.2 0.4 0.6 0.8 1.0
Axial distance x (mm)
Figure 1.
Exercise 9.1 The sketch shows a conical waveguide that is driven harmonically at its
left end by a piston. The right end is terminated by a damping material that eliminates
reflections. It is desired to compare the pressure at point B at the wall according to
the Webster horn equation to the result obtained by taking the signal to be a radially
symmetric wave. For a waveguide that is filled with air at frequencies ranging from
20 Hz to 5 kHz, compare the predictions of both formulations for the magnitude and
phase angle of the pressure at point B.
200 mm
20 mm
o
10
Exercises 9.1 and 9.2
Exercise 9.2 The sketch shows a conical waveguide that is driven harmonically at
its left end by a piston. The right end is terminated by a damping material that elim-
inates reflections. One model for propagation considers the signal to be a radially
symmetric wave. Another approximates the radius variation as an exponential expan-
sion, according to which the radius at point B, where x = 0.2 m, is related to the end
radius by a B = a0 exp (0.2b), where b has units of m−1 . Determine the magnitude
and phase of the pressure of the pressure at point B according to each model. The
frequency varies from 10 Hz to 5 kHz and the fluid is air.
Exercise 9.3 An exponential horn is filled with air. It is 1.2 m long, with end diam-
eters of 10 mm and 100 mm. The pressure at the large end varies harmonically at
800 Hz with an amplitude of 50 Pa. The small end is open to the atmosphere and
is surrounded by a large rigid baffle that is flush with the end. (a) Determine the
complex amplitudes of the pressure and particle velocity at the left end. Use the end
correction on Eq. (3.5.54) for this evaluation. (b) Determine the time-averaged power
radiated into the fluid domain to the left of the baffle.
10 mm 100 mm
1.2 m
Exercise 9.3
Exercise 9.4 An exponential horn is driven at x = 0 by a transducer that imposes a

constant velocity amplitude V0 . The specific impedance at x = L is a fixed value ζ
independent of the frequency. Derive expressions for the amplitudes of the forward
and backward waves as a function of k L when k > b/2.
Exercise 9.5 It is desired to design an exponential horn in air to connect to a piston
driver. The design requirements are as follows (1): The cross-sectional area at x = 0
should be 0.8 m2 . (2) At 200 Hz, the axial particle velocity at x = 0 should be 20 dB
greater than it is at x = L. Equation (3) The cutoff frequency should be 100 Hz. It
may be assumed that there are no reflections at the open end. Determine the growth
constant b and the length L.
Exercise 9.6 Solve Exercise 9.5 for the case where the undriven end’s specific
impedance is ζ = 0.2 + 0.1i.
Exercise 9.7 The small end of an exponential horn is driven by a velocity source.
The other end is joined to a very long cylindrical waveguide that is terminated
by a nonreflecting material. Consequently, only waves that propagate to the right
exist within that section. Derive a set of equations for the pressure amplitude in the
cylindrical waveguide. Assume that the waves in the cylinder are planar.
2m
100 mm 200 mm
Exercise 9.7
Exercise 9.8 Use the WKB approximation, rather than the exact solution of the
Webster horn equation, to carry out the analysis requested in Exercise 9.1.
Exercise 9.9 A horn’s radius expands quadratically with axial distance according to
a = α+βx 2 . The frequency is ω, and the complex pressure amplitude at x = 0 is P0 .
(a) Derive expressions describing the WKB approximation for P (x). (b) Evaluate and
plot |P/P0 | as a function of x corresponding to the following parameters: α = 0.25
m, β = 0.1 m−1 , f = 1000 Hz, ρ0 = 1.2 kg/m3 , c = 340 m/s.
Exercise 9.10 A seismic wave in the bottom of an ocean channel induces a wave that
travels horizontally at 3200 m/s. The dominant harmonic component of this wave
occurs at 12 Hz, and the amplitude of the vertical displacement in this component
is 2 mm. The channel’s depth is 150 m. What is the maximum pressure amplitude
within the channel, and at what depth does it occur?
Exercise 9.11 A scan across the width of a two-dimensional waveguide has yielded
the transverse mode function plotted below. Use the measured data to identify the
specific local impedance of each wall and the eigenvalue of this mode. The fluid is
air, the frequency is 400 Hz, and the width of the waveguide is 5 m.
Transverse mode 
0.8
Real
0.4
Imag
0
−0.4
−0.8
0 0.2 0.4 0.6 0.8 1

Transverse distance y/H
Exercise 9.11
Exercise 9.12 A porous tile is used as the bottom of a two-dimensional water chan-
nel whose height is H . A locally reacting model of the bottom sets the specific
impedance as ζ = i (βω − γ/ω) + δ. The surface of the channel may be consid-
ered to be pressure-release. (a) Derive the characteristic equation whose roots are
the eigenvalues of the transverse mode functions at a specified value of ω. (b) In the
expression for ζ, the parameter β is proportional to a mass per unit surface area, γ
is proportional to a stiffness per unit surface area, and δ is a dissipation factor. What
is the significance of the frequency being (γ/β)1/2 in the situation where δ = 0? (c)
For the ideal case in which δ = 0, derive an equation whose roots are the frequencies
at which the various modes are cutoff.
Exercise 9.13 A waveguide is formed by two parallel walls having infinite extent
and separated by distance H . One wall is well approximated as pressure-release,
while the other is locally reacting with specific impedance ζ. For the case where
k H = 4π and ζ = −2i, determine the eigenvalues and axial wavenumbers of the
first four transverse modes. Plot each mode as a function of the transverse distance
y/H , with each mode scaled such that max (|n |) = 1.
Exercise 9.14 Solve Exercise 9.13 for the case where the specific impedance of the
locally reacting wall is ζ = 0.9 − 2i, and all other parameters are as specified there.
Exercise 9.15 The rigid walls of a two-dimensional waveguide are separated by

distance H . The far end inhibits reflections. The waveguide is excited by a line
source whose volume velocity per unit length is Q̂. The source is perpendicular to
the x y plane at y = H/2 on the cross section x = 0. Except for the source, vx = 0
along the cross section at x = 0. (a) Derive an expression for the pressure at an
arbitrary location for any frequency k H . Which transverse modes contribute to the
response? (b) Suppose the value of k H is very close to 2π. What is the nature of the
pressure distribution along any cross section at large kx?
Exercise 9.16 A two-dimensional horizontal waveguide has a rigid upper and lower
walls. The velocity normal to the vertical wall at x = 0 is a translational oscillation
over the lower half, while the upper half of the wall is stationary. Thus, the boundary
condition at x = 0 is vx = v0 cos (ωt) [h (y) − h (y − H/2)]. The fluid is oil,
ρ0 = 930 kg/m3 , c = 1320 m/s, the height H = 5 m, and ω = 25000 rad/s. (a)
Determine the value of |v0 | if it is observed that the sound pressure level is 195 dB//1
μPa at a point at y = H/2 on the cross section at x = 8 m. (b) Repeat the analysis in
Part (a) for the case where the sound pressure level is 195 dB at y = H/4, x = 8 m.
Exercise 9.17 The walls at y = 0 and y = H of a rectangular waveguide are rigid,

while the walls at z = 0 and z = W are pressure-release. The end x = 0 is rigid,
except that a point source is situated at the corner where x = y = z = 0. The
termination at the far end of the waveguide prevents reflection of waves. Derive an
expression for the pressure at an arbitrary location.
Exercise 9.18 A signal propagates in the horizontal x direction within a long

waveguide whose walls at y = 0, y = H, z = 0, and z = W are rigid. The
termination at the far end inhibits reflections. The signal is generated by a harmonic
vibration at the end x = 0 whose spatial pattern is a mix of the plane wave mode and
the (2,2) mode, that is, vx (x = 0) = [v0 + v2 cos (2π y/H ) cos (2πz/W )] cos (ωt).
The fluid is air, with ω = 500π rad/s, H = 3 m, W = 4 m, and v0 = v2 = 400
mm/s. (a) Evaluate |P| as a function of x along the center line, y = H/2, z = W/2
for 0 ≤ x ≤ 10 m. (b) Evaluate |P| as a function of y along the transverse line
z = W/2 at x = 1 m and x = 10 m.
Exercise 9.19 The walls at y = 0, y = H , and z = W of a rectangular waveguide
are rigid, and the wall at z = 0 is pressure-release. The excitation is a uniform
velocity distribution over one-quarter of the end x = 0, specifically vx = v0 cos (ωt)
for 0 < y < H/2 and 0 < z < W/2. The aspect ratio is W/H = 2.5, and
k H = 10. Derive an expression for the pressure at an arbitrary location knowing that
the termination at the far end of the waveguide prevents reflection of waves. Then,
evaluate this result for the pressure along the lines y = H/2 and z = W/2 at the
cross section x = 20π/k.
Exercise 9.20 The sketch shows a two-dimensional model of a rectangular wave-
guide in which a layer of liquid having density ρ2 and sound speed c2 overlays
a denser liquid whose properties are ρ1 and c1 . The bottom is rigid, whereas air
is above the upper liquid. Consequently, that interface may be considered to be
pressure-release from the viewpoint of the liquid. Vibration of the vertical wall at
x = 0 generates waves that propagate to the right. The far end attenuates these
waves, so that there are no return reflections. (a) Derive the characteristic equation
governing horizontal propagation modes. (b) Derive an expression for the transverse
mode function corresponding to an eigenvalue that would be obtained by solving
the characteristic equation. (c) Prove that if ρ2 = ρ1 and c2 = c1 , then the modal
properties reduce to those of two-dimensional waveguide of height H whose walls are
rigid and pressure-release. Hint: Let μ and η denote the transverse wavenumber of
for 0 < y < H/2 and H/2 < y < H , respectively. The pressure and vertical particle
velocity derived from a mode function must be continuous. In addition to leading to
the conditions that and d/dy be continuous at y = H/2, this requirement tells
us that the axial wavenumber κ must be the same for both layers. Thus, it must be
that κ = (ω 2 /c12 − μ2 )1/2 = (ω 2 /c22 − η 2 )1/2 .
Liquid 2 H/2
vx
Liquid 1 H/2
x
Exercise 9.20
Exercise 9.21 The wall of a cylindrical waveguide is locally reacting, with specific
impedance ζ = 4i. Determine the eigenvalues of the first four axisymmetric trans-
verse modes. Plot dependence of these mode functions on the transverse distance R.
Carry out the analysis for ka = 1 and ka = 10.
Exercise 9.22 The wall of a cylindrical waveguide is locally reacting, with specific
impedance ζ = 4i. Determine and graph the phase and group veocities of the first
four modes from ka = 15 down to the cutoff frequency of each mode.
Exercise 9.23 A circular waveguide whose wall is rigid is drive by a vibrating plate
at x = 0. The spatial pattern of the vibration matches the fundamental nonplanar
mode of the waveguide, that is, vx = v0 0,1 cos (ωt). Determine and graph the
nondimensional mean-squared pressure p 2 av / (ρ0 cv0 )2 and nondimensional mean-

squared particle velocity vx2 av /v02 , averaged over any cross section. Perform the
analysis for frequencies ranging from cutoff to ka = 10. Compare these dependencies
to that of the power transported in that mode.
Exercise 9.24 A cylindrical waveguide whose wall is rigid is excited by a point

source situated on the centerline at x = 0. Determine the pressure amplitude on-axis
at kx = 40 as a function of frequency in the range 0.1 ≤ ka ≤ 16.
Exercise 9.25 A cylindrical waveguide whose wall is rigid is excited by a point

source situated at x = 0. The location of the point source is R = a/2 from the
centerline. Derive an expression for the pressure field generated by the source.
Exercise 9.26 The region between the inner and outer cylinders is an annular
waveguide. A disturbance interior to the inner cylinder induces a harmonic dis-
placement wave that propagates along the inner wall. This wave propagates down-
stream at sound speed cw > c. In other words, the transverse velocity at R = b
is v R = V sin (ωt − ωx/cw ). The outer cylinder is rigid. (a) Derive an expression
for the pressure wave that is generated in the annular region between the cylindrical
walls. Does the analysis indicate that there are values of ω that cause a resonance?
(b) For the case where cw = 3c, evaluate the pressure at the outer wall as a function
of ka < 20.
a
b
Exercise 9.26
Exercise 9.27 A circular disk is the closure of a very long cylindrical waveguide.
The disk is mounted on a shaft that rotates at angular speed . The shaft’s centerline is
collinear with the centerline of the waveguide. Because of a manufacturing error, the
disk is not correctly aligned. Rather, the normal to its surface forms a small constant
angle relative to the shaft. The sketch is a side view of the configuration when
the plane containing both centerlines is vertical. The effect of the misalignment is to
cause a fluctuation in the x coordinate of the boundary. Consider two points at fixed
R and θ. One is on the nominal end, x = 0, and the other is on the disk. This distance
oscillates as a consequence of the rotation, and there is an axial velocity at (R, θ) at
the end. This velocity is vx = R sin (t − θ). The cylindrical wall is rigid, and
the far end is terminated in a manner that inhibits reflections. Derive an expression
for the pressure within the waveguide. Which transverse modes are excited?
 a
x
 a
Exercise 9.27
Exercise 9.28 A cylindrical waveguide, whose wall is rigid, is excited at the end x =
0. Waves that reach the far end are fully absorbed. The excitation at x = 0 consists of a
harmonic axial velocity that is uniformly distributed over two nonadjacent quadrants.
The velocity in one quadrant is 180◦ out-of-phase from the other, and the velocity
in the intervening quadrants is zero. What transverse modes are excited? Derive an
expression for the pressure at an arbitrary field point.
Exercise 9.29 A different type of waveguide that features cylindrical waves occurs
in an ocean channel. The cylinder in the sketch extends from the ocean floor up to
the free surface at the atmosphere. The acoustic domain here is R > a, 0 < z < H ,
−π < θ < π. The surface of the cylinder undergoes a vibration in the transverse
direction, ē R . (One context in which this configuration might arise is a pipe that brings
oil to the surface from a drill hole.) Waves propagate away from the cylinder, as they
do for an infinitely long cylinder whose surface vibrates. The fundamental difference
is the finite extent of the axial direction. To avoid unnecessary complications, con-
sider an axisymmetric pressure field p (R, z, t). (a) Determine the transverse mode
functions for this system. (Note that propagation is in the sense of increasing R, so
the transverse modes are functions of z.) Do these modes have a cutoff frequency?
(b) Describe the orthogonality and normalization conditions for the mode functions.
(c) Describe the modal series for the pressure corresponding to the general situation
where the vibration at R = a is a harmonic, with an arbitrary dependence on z,
v R = Re ( f (z) exp (iωt)). Derive an expression for the pressure field generated by
this vibration.
Pressure-release
a
x y
x
R
z
Rigid
Exercise 9.29
Chapter 10
Modal Analysis of Enclosures
To say that a region containing an ideal fluid is a “cavity” is to imply that it is a

void in some solid region. The synonymous term “enclosure” is more descriptive,
in the sense that it conveys the notion that the domain is surrounded by a solid
boundary. A tube terminated at both ends is the simplest configuration to analyze.
Sections 2.5.3 and 3.2.2 addressed planar waves in such a system, but our concern here
will extend to higher order waveguide modes. As is true for planar waves, closure
results in reflection of the incident waves, thereby setting up a field that features
standing waves in all directions. The developments that follow will primarily deal
with regular geometries. The chapter will close by developing an approximate method
for irregular cavities. The closure will be an approximate method for cavities and
elastic structures.
10.1 Fundamental Issues
10.1.1 Wall-Induced Signals
In our study of waveguides in the previous chapter, the fundamental premise was that
the far end is terminated in a manner that prevents reflections. We now remove that
restriction. Figure 10.1 depicts a rectangular cavity. The length in the x direction is
L, and the height in the y direction is H . To avoid unnecessary complications, we
begin with a situation in which there is no spatial variation in the z direction.
Thus,
the frequency-domain response will have the form p = Re P (x, y) eiωt .
Let us consider the situation where a known normal velocity distribution Vs (y)
exists on the face at x = 0, and the walls at y = 0, y = H , and x = L are rigid.
We know that if L were infinite, the imposed velocity at x = 0 would generate a set
of waveguide modes that propagate in the direction of increasing x. The presence

DOI 10.1007/978-3-319-56847-8_10
292 10 Modal Analysis of Enclosures
Fig. 10.1 Coordinate y

systems for a
two-dimensional cavity. An
x
excitation acts at x = 0 and
the other walls are rigid H
L
z
of a wall at finite L, does not alter this fact, but it does result in reflections that
are manifested as modes that propagate in the direction of decreasing x. The y
dependence of both sets of modes must be the same set of transverse mode functions
n (y). The transverse mode functions for a hard-walled two-dimensional waveguide
are nπ y
n (y) = Bn cos (10.1.1)
H
√
where normalization gives Bn = 1 if n = 0 and 2 otherwise. The axial wavenumber
for both the forward and backward propagating modes must be κn , but we do not
know a priori the modal amplitudes for either set of modes. Hence, the appropriate
modal series is
∞
nπ y
P= Dn,1 e−iκn x + Dn,2 e+iκn x Bn cos
n=0
H
∞ nπ y (10.1.2)

= Dn,c cos (κn x) + Dn,s sin (κn x) Bn cos
n=0
H
where the representation of the x dependence as sinusoidal functions makes it easier

to satisfy boundary conditions.
As was noted, this representation satisfies the rigidity conditions at y = 0 and
y = H , and each term in the series individually satisfies the Helmholtz equation
because the κn wavenumbers are those in Eq. (9.2.10). It remains to satisfy the con-
ditions at x = 0 and x = L, which are

∂ P
= −iωρ0 Vs
∂x x=0
(10.1.3)
∂ P
=0
∂x x=L
The second condition is satisfied by each term in the modal series if
Dn,s cos (κn L) = Dn,c sin (κn L) (10.1.4)

10.1 Fundamental Issues 293
We use this relation to eliminate Dn,s in Eq. (10.1.2) and then define a new coefficient
Dn ≡ Dn,c / cos (κn L). This converts the series to
∞
nπ y
P= Dn Bn cos (κn (L − x)) cos (10.1.5)
n=0
H
The Dn coefficients are found by satisfying the boundary condition at x = 0. To

that end, Vs (y) is expanded in a series of the transverse mode functions. Orthogo-
nality of those functions yields
∞
nπ y
H nπ y
1
Vs (y) = Vn cos , Vn = Vs Bn cos dy (10.1.6)
n=0
H H 0 H
This series and Eq. (10.1.5) are substituted into the first of Eq. (10.1.3). Matching the
coefficients of each mode leads to
∞
Bn nπ y
P = −iωρ0 Vn cos (κn (L − x)) cos (10.1.7)
κ sin (κn L)
n=0 j
H
This representation of the response to excitation on a boundary is quite similar to

the result of the analysis in Sect. 9.2.5. One difference is that the dependence on x here
is a superposition of standing waves, rather than propagating waves. Each series term
describing the spatial dependence is a forced cavity mode. A new feature contained
in Eq. (10.1.7) is the possibility of a singularity when any of the axial wavenumbers
are such that sin (κn L) = 0. Because κn depends on k, these singularities correspond
to resonant frequencies. We shall return to this feature after we consider a different
type of excitation.
10.1.2 Source Excitation
Problems pertaining to the response within an enclosure that is induced by motion

of a wall are particularly relevant to moving vehicles. However, if the enclosure is a
room, the excitation of interest is likely to be a source, such as a person or a musical
instrument. Suppose the cavity in question is that in Fig. 10.1, with all walls now
taken to be rigid. Because there is no wall that generates propagating modes, there is
no conceptual difference between the x and y directions. If we consider the former to
be the axial direction, the transverse mode functions are Bn cos (nπ y/H ), whereas
taking the y direction to be axial leads to transverse functions that are B j cos ( jπx/L).
It is not reasonable to favor either direction, so we use one of each function to represent
P. There is no limit to the values of n and j, so it seems reasonable to try an ansatz
for the pressure that is the sum of all possible combinations of the transverse mode
functions for each direction. That is, let us try
∞
∞ nπ y
jπx
P= B j Bn D j,n cos cos (10.1.8)
j=0 n=0
L H
This expression has the appearance of a Fourier cosine series in each direction, but
we soon will find a more profound interpretation.
The above representation clearly satisfies all rigid wall conditions, but there is
a fundamental difficulty—none of the terms in the double summation individually
satisfy the Helmholtz equation at an arbitrary frequency. Furthermore, there is no
indication of the role of a source. Suppose a line source is situated at an arbitrary loca-
tion (x0 , y0 ). It was shown as part of the definition of Green’s functions, Sect. 6.4.2,
that the field of a harmonic point source is governed by an inhomogeneous Helmholtz
equation, specifically Eq. (6.4.22). Thus, the pressure must satisfy
∂2 P ∂2 P
+ + k 2 P = −iωρ0 Q̂δ (x − x0 ) δ (y − y0 ) (10.1.9)
∂x 2 ∂ y2
The terms in Eq. (10.1.8) will not satisfy this equation individually, but they might
do so collectively. Conceptually, we want to find the set of coefficients D j,n for which
the discrete, doubly infinite summation satisfies the field equation at a continuum of
points within the cavity. Contemplating matching these different types of infinities is
bewildering, so let us look at the matching process from a linear algebra perspective.
Suppose we were to substitute Eq. (10.1.8) into the inhomogeneous wave equation,
with arbitrary values assigned to the D j,n coefficients. The result would be a differ-
ence (x, y) between the left and right side of the equations. Rather than setting
to zero at each location, we invoke the interpretation of the series as a sum of unit
vectors in a linear functional space (the product of cosines) multiplied by component
lengths (the D j,n values). Because all possible solutions of the cavity with rigid walls
lie in this linear space, we can assert that the correct set of coefficients is that which
causes to be orthogonal to all directions in the linear space. The pressure functions
are defined in the domain 0 < x < L, 0 < y < H , so orthogonality constitutes an
inner product over this space. The functions in the pressure series terms are sinu-
soidal, which means that the orthogonality property is that of a Fourier series. Hence,
a function F (x, y) is orthogonal to the ( j, n) term if

H
L nπ y
jπx
F (x, y) B j cos Bn cos d xd y = 0 (10.1.10)
0 0 L H
To implement this concept, we substitute the pressure series into the inhomoge-
neous wave equation, multiply both sides by a selected function, and then integrate
over the domain of the cavity. It is convenient to change the summation indices to
r and s, so that we may multiply the series by the ( j, n) function. These operations
lead to
∞
H
L r πx sπ y r π 2 sπ 2
Br Bs Dr,s cos cos k −2
−
r =0 s=0 0 0 L H L H
nπ y
jπx
× B j Bn cos cos d xd y
L H

H
L
= −iωρ0 Q̂ δ (x − x0 ) δ (y − y0 )
0 0
nπ y
jπx
B j bn cos cos d xd y (10.1.11)
L H
On the left side, we invoke the orthogonality of the functions in each direction. Thus,
only the term for which r = j and s = n is nonzero. In that case, inclusion of the
Bm in the integrals over x and y, respectively, gives L and H factors. On the right
side, we have the integral property of a Dirac delta function, which evaluates the
analytical part of the integrand at the location where the delta occurs. The result is

2 jπx0 nπ y
0
L H k 2 − χ j,n D j,n = −iωρ0 Q̂ B j Bn cos cos (10.1.12)
L H
where 1/2
2 nπ 2
jπ
χ j,n = + (10.1.13)
L H
Substitution of the resulting expression for D j,n into Eq. (10.1.8) leads to
∞ ∞ 2
Q̂ B j Bn jπx0
P = −iωρ0 2 cos
L H j=0 n=0 k 2 − χ j,n L
nπ y nπ y
0 jπx
cos cos cos (10.1.14)
H L H
This is a remarkable result because it is the response of the cavity to a point source at
an arbitrary location. In other words, it is Green’s function for that domain. Observe
in this regard that the expression is reciprocal, as is evident by the fact that swapping
x and x0 , or y and y0 does not alter the result.
Thus, we have identified alternative sets of functions that describe the field within
the cavity. The n functions in the previous section were derived from the propagation
modes of a waveguide. Changing the excitation changes these functions. We shall
refer to them as forced cavity modes. These attributes are contrasted by the functions
from which Eq. (10.1.14) is constructed. We shall denote them as j,n and refer to
them as natural cavity modes, although we often will omit “natural.” These functions
are independent on the nature of the excitation and therefore represent intrinsic
properties of the system. A specific set of wall conditions leads to a specific set of
j,h . The functions appearing above are the two-dimensional functions for a hard-
walled cavity,
nπ y
jπx
j,n = cos cos (10.1.15)
L H
Let us substitute this function into the (homogeneous) Helmholtz equation. Doing
so leads to recognition that any j,n is a solution when the acoustic wavenumber is
k = χ j,n . These observations are equivalent to saying that χ j,n are the eigenvalues and
j,n are the eigenfunctions of the cavity. The natural frequencies are ω j,n = cχ j,n .
It now is clear why Eq. (10.1.14) is singular if k equals any χ j,n , or equivalently,
when ω = (ωnat ) j,n . Excitation of any linear system at its natural frequency leads to
resonance. This phenomenon occurs because all internal actions balance to sustain
a free response, so there is no remaining ability to resist an excitation. The only
exception to this statement occurs if the excitation is such that the mode is not excited.
For example, suppose that k = χ2,3 ( j = 2, n = 3) in Eq. (10.1.14), but x0 = L/4.
Then, cos ( jπx0 /L) = 0. Mathematically, this gives a term in the pressure series in
which zero is divided by zero. The reality is that a true singularity is unlikely because
it requires that k exactly equals χ2,3 . Furthermore, the analysis excluded dissipation
effects that ameliorate resonances. Another aspect to consider is a property identified
in Sect. 2.5.3, specifically, that resonance in a linearized analysis results in progressive
growth of the transient response. As the response grows, nonlinear effects become
increasingly important. Thus, the singularities at resonance merely flag exceptional
situations.
An interesting aspect results from rewriting the pressure series, Eq. (10.1.14), in
terms of the mode function notation, rather than explicit harmonic functions. Doing
so gives
∞ ∞
Q̂ j,n (x0 , y0 ) j,n (x.y)
P = −iωρ0 (10.1.16)
L H j=0 n=0 k 2 − χ2j,n
This form is descriptive of Green’s function for any two-dimensional cavity, with
the mode functions appropriate to that domain. Furthermore, three-dimensional sit-
uations are described by altering the summation to extend over three indices for the
three directions. We will see how this result is obtained in Sect. 10.2.1.
The coexistence of forced and natural cavity modes, which result in different
modal series, might cause some degree of confusion, so let us contrast their essential
features. The n in Eq. (10.1.7) represent the transverse dependence of waveguide
modes that propagate in opposite senses at any frequency. The j,n represent the
pressure fields that can exist within a closed region without excitation only at the
natural frequencies. Some individuals refer to natural cavity modes as standing wave
modes. Perhaps the most important distinction is that natural modes exist for any
enclosed domain, whereas a series of propagation modes obviously assumes that
the picture of a waveguide closed at one end is appropriate. For example, although
the analysis would not be simple, it is possible to find the cavity modes for an L-
shaped room. However, the L-shape does not suggest a conceptual view in which
waves propagate back and forth in one direction. Indeed, the direction that might be
considered to be axial in one leg of the L-shape would correspondingly be viewed
as the transverse direction for the other leg.
Resonance of any dissipationless system is manifested as a singularity at a natural

frequency. Regardless of whether the field within a cavity is described as a series of
forced or natural modes, resonances will be manifested as an exceptionally large value
of a modal coefficient. Resonance occurs in Eq. (10.1.7), which is the forced mode
series, when sin (κn L) in the denominator is zero for a specific n. This occurs when
κn L = jπ, j = 0, 1, ... The axial wavenumber κn for a waveguide mode is related
to the eigenvalue μn of the transverse mode function by κ2n = k 2 − μ2n , and μn =
nπ/H for a hard-walled waveguide. Hence, the resonances occur when n 2 π 2 /H 2 +
j 2 π 2 /L 2 = k 2 . This condition is identical to the resonance condition k = χ j,n for
the natural cavity mode series in Eq. (10.1.14).
Each type of cavity mode has situations for which it is most useful. A series
of forced modes is well suited to situations where the excitation originates on the
boundary. However, a forced mode series cannot be implemented directly if the cavity
is excited by one or more sources within the fluid. This is so because these modes
satisfy the homogeneous Helmholtz equation. We will see in Example 10.4 that in
some cases it is possible to use a forced mode analysis to determine the effect of a
source. However, the situation addressed there is rather special, and its extension to
more general systems would be quite difficult.
The utility of a natural cavity mode formulation is quite opposite. These functions
are solutions of the Helmholtz equation that satisfy passive (that is, homogeneous)
boundary conditions, so they cannot be used directly to satisfy an inhomogeneous
boundary condition associated with an excitation. On the other hand, such a formu-
lation readily handles cases of source excitation.1
10.2 Frequency-Domain Analysis Using Forced

Cavity Modes
The developments in this section assume familiarity with the properties of propaga-
tion modes for waveguides. The various derivations and results in Chap. 9 are not
repeated here.
10.2.1 Rectangular Enclosures
A cavity in the shape of a rectangular box, with one wall that undergoes a specified
motion, is relatively uncomplicated to analyze, yet it exhibits the same phenomena as
1 It is possible to use the forced mode formulation to determine the field of a source within an enclosed
regions. The method follows the development in Sect. 6.4.2, in which a solution is constructed
by adding a function F (x̄) to the free-space source solution G (x̄). Similarly, a natural mode
formulation can be used to address boundary excitation by adding to the series a term that satisfies
all boundary conditions. Equations for the modal coefficients would be obtained by requiring that the
sum of the added term and the natural mode series satisfy the inhomogeneous Helmholtz equation.
any other cavity. The direction normal to the vibrating wall is designated as the axial
direction x, with the axial length designated as L. The vibrating wall is defined to be
situated at x = 0. The velocity distribution at this boundary may be arbitrary, so it
is required that Vx = Vs (y, z) at x = 0. The sidewalls are the planes y = 0, y = H ,
z = 0, and z = W . Conceptually, the signal within this cavity consists of modes that
are radiated from the vibrating wall and propagate in the direction of increasing x.
When they arrive at x = L, they are reflected, which results in waves that travel back
to x = 0, where they are reflected, ad infinitum. The waves that propagate forward
and back in the axial direction are the propagation modes evaluated in Sect. 9.3.2.
The amplitudes of the forward and backward modes are quantities to be determined.
All walls other than the vibrating one are taken to be locally reacting, which
includes the rigid and pressure-release conditions. The boundary conditions on the
sidewalls are the same as those posed in Eq. (9.3.19) for a rectangular waveguide.
Thus, the transverse mode functions are as described in Eqs. (9.3.25) and (9.3.26).
For a rectangular waveguide, these functions are the product of the transverse modes
for two-dimensional waveguides in each direction, specifically
j,n = B j,n φ y, j (y) φz,n (z) (10.2.1)
Eigenvalues associated with j,n are μ j for the y functions, and νn for the z functions.
It is essential that the modal series account for the effect of all modes. A plane wave
mode can exist in the y direction if the walls at y = 0 and y = H are rigid. In that
case, the sum over the modes for the y direction would begin with j = 0, with
the first eigenvalue μ0 = 0. Another possibility is that y = 0 and y = H are purely
compliant. In that case, there would be a purely imaginary eigenvalue. This mode
can be assigned to j = 0, so that μ0 = Im (μim ) i. In any other case, the modal sum
would begin with j = 1. The same considerations apply to the φz,n (z) functions,
whose eigenvalues are νn . A further consideration is that the scaling factors B j,n are
required to have been set according to the normalization rule stated in Eq. (9.3.29).
A modal series consists of sums in which the transverse functions multiply com-
plex exponentials representing forward and backward propagation. The amplitudes
of these exponentials are denoted as α j,n and β j,n , respectively, so the ansatz is
∞
∞

P= α j,n e−iκ j,n x + β j,n e+iκ j,n x j,n (y, z) (10.2.2)
n=0 j=0
The axial wavenumbers κ j,n must be such that each term in the series individually
satisfies the Helmholtz equation, which means that
1/2
κ j,n = k 2 − μ2j − νn2 (10.2.3)
The modal series accounts for forward and backward waves in the x directions, so
we may continue the previous practice of placing κ j,n in the fourth quadrant of the
complex plane.
10.2 Frequency-Domain Analysis Using Forced Cavity Modes 299
Because the transverse mode functions satisfy the boundary conditions at the
sidewalls, it only remains to make the particle velocity Vx at x = 0 match the imposed
distribution Vs (y, z) and to make P/Vx be consistent with the locally reacting model
at x = L. These conditions are

∂ P
= −iωρ0 Vs (y, z)
∂x x=0
ζ L ∂ P (10.2.4)
P|x=L = −
ik ∂x x=L
Substitution of the modal series into these boundary conditions gives

∞
∞

iκ j,n −α j,n + β j,n B j,n j,n (y, z) = −iρ0 ω V̂s (y)
n=0 j=0
∞ κ j,n −iκ j,n L κ j,n +iκ j,n L
∞
α j,n 1 − ζ L e + β j,n 1 + ζ L e B j,n j,n (y, z) = 0
n=0 j=0
k k
(10.2.5)
It might appear that the preceding constitutes two equations for many unknown α j,n
and β j,n values, but such thinking ignores the orthogonality of the transverse mode
functions. Each equation is multiplied by an arbitrarily selected m, ≡ Bm, φ y,m φz,
and integrated over the cross section, 0 < y < H , 0 < z < W . These operations filter
out all term in the summations for which n = m and j = remain. What remains is
a pair of equations,
k
αm, − βm, = ρ0 c Vm,
κm,
κm, −iκm, L κm, +iκm, L (10.2.6)
αm, 1 − ζ L e + βm, 1 + ζ L e =0
k k
The Vm, coefficients are those of a modal series for Vs defined in Eq. (9.3.37),
specifically

W
H
1
Vm, = m, (y, z) Vs (y, z) dydz (10.2.7)
HW 0 0
Orthogonality has reduced the problem to that of solving two simultaneous equa-
tions for αm, and βm, , which leads to
κm, +iκm, L
1 + ζL e k
αm, = k ρ0 c Vm,
(k, j, n) κm,
κm, −iκm, L (10.2.8)
−1 + ζ L e k
βm, = k ρ0 c Vm,
(k, j, n) κm,
where (k, j, n) is the determinant of Eq. (10.2.6),

κ j,n iκ j,n L κ j,n −iκ j,n L

κ j,n = 1 + ζ L e + 1 − ζL e (10.2.9)
k k
Substitution of the solution into the modal series, Eq. (10.2.2), leads to
∞
∞
kV j,n κ j,n iκ j,n (L−x)
P = ρ0 c 1 + ζL e
κ κ j,n
n=0 j=0 j,n
k
(10.2.10)
κ j,n −iκ j,n (L−x)
− 1 − ζL e j,n (y, z)
k
The fact that the variable for the axial dependence in Eq. (10.2.10) is L − x highlights
the reflection process at x = L. Another manifestation of that attribute comes from
factoring 1 + ζ L κ j,n /k out of the bracketed term. Doing so displays the role of a
modal reflection coefficient,
κ j,n
ζL −1
(R L ) j,n = κkj,n (10.2.11)
ζL +1
k
so that
∞
∞

k eiκ j,n (L−x) + (R L ) j,n e−iκ j,n (L−x)
P = ρ0 c V j,n j,n (y, z)
n=0 j=0
κ j,n eiκ j,n L − (R L ) j,n e−iκ j,n L
(10.2.12)
Setting j,n (y, z) = 1 and κ j,n = k would make each term in the summation be like
Eq. (3.2.21) for plane waves.
Some individuals prefer a form of Eqs. (10.2.9) and (10.2.10) that results when
Euler’s identity is used to remove the complex exponentials in favor of sine and cosine
functions. The resulting expressions are not displayed here because such forms offer
no merit if ζ L or κ j,n is complex.
Resonances correspond to values of k for which κ j,n = 0. It can proven that
this condition occurs only if the impedance ζ L at x = L and the impedances of
the sidewalls are infinite, zero, or purely imaginary, that is, purely reactive. As was
shown earlier, any frequency at which = 0 is a natural frequency of the cavity.
If any impedance has a nonzero real part, energy is absorbed by that wall, thereby
inhibiting unlimited growth of the pressure amplitude. However, if the resistive part
of each impedance is substantially smaller than ρ0 c, a plot of pressure as a function
of frequency will exhibit prominent peaks in the vicinity of the natural frequencies
associated with zero resistive parts.
When any wall impedance is a finite compliance or inertance, determination of
the natural frequencies by solving the eigenvalue problem is not a simple task. The
difficult lies in the fact that a transverse eigenvalue, μ j or νn , will depend on the fre-
quency because the associated characteristic equation contains k, see, for example,
Eq. (9.2.32). Concurrently, the axial wavenumber κ j,n must satisfy its own character-
istic equation, specifically Eq. (10.2.8). In addition, the natural frequency is related
to the eigenvalues through Eq. (10.2.3), so that
1/2
(knat ) j,n = κ2j,n + μ2j + νn2 (10.2.13)
Thus, solution of the eigenvalue problem in this situation requires simultaneous solu-
tion of three characteristic equations and one algebraic equation. The unknowns are
the three wavenumbers and the knat . Such a solution could be obtained by implement-
ing a four variable Newton’s method. A simpler, though less efficient, procedure is
to sweep through a range of k. For each value, each characteristic equation is solved
for the associated wavenumber. If these values are consistent with Eq. (10.2.13), then
that value of k marks a natural frequency. Regardless of the procedure that is imple-
mented, care must be exercised to obtain all natural frequencies in the interval of
interest.
One of the useful aspects of a series representation of the pressure field is the
uncoupled nature of the equations to be solved. That is, we can find the amplitude
coefficients α j and β j for a mode without finding the coefficients of the other modes.
In some situations, it might be that more than one wall moves. In that case, the
response may be found by superposing solutions. In each subproblem, one wall
moves and the other vibrating walls are held fixed; that is, they become rigid. The
axial direction for each subproblem would be normal to the wall that is vibrating.
Addition of the pressure from each subproblem yields the pressure field.
EXAMPLE 10.1 All walls of a rectangular cavity are rigid. The dimensions
are L in the x direction, H in the y direction, and W in the z direction. The upper
half of the wall at x = 0 is stationary. The lower half is a rigid plate having
mass M that is forced to undergo an oscillation at frequency ω. Thus, the
end condition may be written as vx = Re (V0 exp (−iωt)) [h(y) − h(y H/2)],
where h denotes the step function. Derive an expression for the amplitude of the
force that must be applied to the plate. Evaluate this force as a function of the
nondimensional frequency k L, for the case where H/L = 0.2 and W/L = 0.1.
The mass of the plate equals the mass of the fluid contained in the cavity.
Significance
This basic operations for the analysis are fairly straightforward. What makes the
problem interesting is the interpretation of the nature of the forces acting on the
plate, which will afford a different perspective to the meaning of resonances.
Solution
The force acting on the plate must overcome the inertia of the plate and the resultant
force exerted by the pressure. This resultant, which acts in the negative x direction if
the pressure is positive, is the integral over the plate of the pressure at x = 0. Thus,
the complex amplitude of the pressure resultant is

W
H/2
Fac = P|x=0 dydz (1)
0 0
The acceleration of the plate is Re (iωV0 exp (iωt)). Thus, the force amplitude
required to translate the plate is
F = iω M V0 + Fac (2)
Now, we turn our attention to the fluid. The walls that coincide with planes of
constant y and z are not forced, so we take them to be the sidewalls of the waveguide.
The corresponding transverse mode functions are
nπz
jπ y
j,n = B j,n cos cos , j, n = 0, 1, 2, ... (3)
H W
Normalization of these functions leads to

4
B j,n =
1 + δ j,0 1 + δn,0
The axial wavenumber is

2 1/2
jπ nπ 2
κ j,n = k − 2
− (4)
H W
with the usual proviso that κ j,n is negative imaginary if the quantity within the
brackets is negative.
In Eq. (10.2.4), the axial velocity imposed at x = 0 is Vs (y, z). Here, this function
is specified to be
Vs = V0 [h(y) − h(y − H/2)], 0 < y < H, 0 < z < W
The corresponding velocity coefficients obtained from Eq. (10.2.7) are
V j,n = V0 B j,n I j,n

W
H/2 nπz
1 jπ y
I j,n = cos cos V0 dydz
HW 0 0 H W
⎧
⎪
⎪ 0 if n > 0 (5)
⎪
⎪
⎨1
if j = n = 0
= 2
⎪
⎪
⎪
⎪ 1 jπ
⎩ sin if j > 0 and n = 0
jπ 2
The V j,n coefficients are zero if n > 0 because the imposed velocity is constant in
the z direction, and a constant is orthogonal to the z dependence of all transverse
mode functions for n > 0. The velocity distribution is the excitation, so any velocity
coefficient that is zero means that the corresponding mode is not excited. In addition,
the factor sin ( jπ/2) is zero if j is even, so V j,n ≡ 0 for even j. Thus, rather than
requiring evaluation of a double sum, Eq. (10.2.10) reduces to a single sum over
j = 0, 1, 3, ... with n = 0. The only other adaptation required to use that equation
is to let ζ L be infinite. The resulting series for the pressure field is
∞

2 kL eiκ j,n (L−x) + e−iκ j,n (L−x) jπ y
P = ρ0 cV0 B j,0 I j,0 cos
j=0,1,3,...
κ j,0 L eiκ j,n L − e−iκ j,n L H
(6)
The last step is to form the resultant force by substitution of this expression,
evaluated at x = 0, into Eq. (1). The integral for this evaluation is the same as the
one for Im, in Eq. (5). When the complex exponential sum and difference are replaced
by sinusoidal functions, the pressure resultant reduces to
Fac = H W Z ac V0
∞

2 2 kL (7)
Z ac = iρ0 c B j,0 I j,0 cot κ j,0 L
j=0,1,3,...
κ j,0 L
An expression for the force that must be applied to the plate is obtained by sub-
stituting Fac into Eq. (2), which results in
F = H W Z F V0 (8)
The factor Z F is the total impedance seen by F. It consists of a contributions from

the acoustic pressure and the inertia of the plate,

iω M M
Z F = Z ac + Z M , Z M = ≡ iρ0 ck L (9)
HW ρ0 L W H
Because of the simplicity of this system, the value of Z F / (ρ0 c) depends on only
three parameters: the nondimensional frequency k L, the ratio of the plate’s mass to
that of the fluid contained within the cavity M/ (ρ0 L W H ), and H/L. If j is less
than the cutoff value k H/π, then κ j,0 is real, whereas κ j,0 is negative imaginary if j
exceeds the cutoff value. In either case, the terms inside the summation are real. It
follows that both Z ac and Z M are reactive at any frequency. The corollary is that the
time-averaged power input to the cavity by F is zero. This is as it must be, because
the walls are rigid and therefore cannot dissipate energy.
The value of j at which the series in Eq. (7) may be truncated is readily identified
from Eq. (4), which indicates that κ j,0 L ≈ −i jπ (H/L) when jπ/H k. For very
large j, the terms in the summation are proportional to coth ( jπ (H/L)), which
approaches unity. The coefficients B j,0 = 21/2 and I j,0 are inversely proportional to
jπ. It follows that for large j the terms in the summation are inversely proportional
to j 3 . Thus, a truncation criterion that j > k L (H/L) /π and j > 20 should be

adequate. Figure 1 shows the result of computing Z F for the given parameters with
the series truncated at j = 21. In order to understand the role of the resultant acoustic
force, the frequency dependence of Z ac also is shown in Fig. 1. The second graph
gives a finer picture of the behavior in the vicinity of the first resonances.
ZF Zac
20
Im(Z/(0c))
10
−10
−20
0 5 10 15 20
Frequency, kL
10
5
Im(Z/(0c))
−5
−10
0 1 2 3 4 5 6 7 8
Frequency, kL
Figure 1.
Resonances occur if the denominator in Eq. (6) vanishes for any j. The denomi-
nator may be written as 2i sin(κ j,0 L). Equating it to zero yields κ j,0 L = mπ. Then,
Eq. (4) yields the natural frequencies,
2 1/2
H
(k L)nat = (mπ) + ( jπ)
2 2
, m = 0, 1, 2, ..., j = 0, 1, 3, ... (10)
L
These values are marked with an asterisk in Fig. 1. The values in ascending order are
tabulated below.
j 0 0 0 0 0 0 1 1 1 1 0
m 0 1 2 3 4 5 0 1 2 3 6
(k L)nat 0.00 3.14 6.28 9.42 12.57 15.71 15.71 16.02 16.92 18.32 18.85
Two different j, m pairs give rise to the same natural frequency at k L = 15.71, and
the spacing between natural frequencies decreases with increasing frequency. Both
of these features will be explored in detail when we analyze natural cavity modes.
As expected, |Z F | is very large in the vicinity of a natural frequency, whereas the
value of Z M is an inertance that is proportional to the frequency. The plots in Fig. 1
indicate that Z M is the dominant effect away from a resonance. As the frequency
is increased beyond a resonance, Z ac transitions to negative imaginary; that is, it
is a compliance. Further increase toward resonance results in a sharp increase of

this compliance, eventually becoming singular as a resonance is approached from
below. The consequence is that at some frequency between resonances the value of
Z F ≡ Z M + Z ac is zero. This has an important implication.
Let us change our perspective to ask what the plate’s displacement would be if
we were to drive it with a specified F. The displacement is U = V0 / (iω), so Eq. (8)
tells us that
F 1
U= (11)
iω H W Z F
From this viewpoint, a large value of Z F at a resonance leads to a very small dis-
placement. In contrast, a very small value of Z F leads to a large displacement. This
is a resonance of the coupled system formed by the plate and the cavity. We say that
it is a fluid-structure resonance.
This might seem to be somewhat confusing, but the two situations are readily
distinguished if we think of an experiment. The situation implied in the problem
statement is that a machine exerts a force amplitude F that can be as large as necessary
to maintain a specified value of V0 regardless of the pressure within the cavity. In this
scenario, the pressure will grow greatly when the frequency is close to any (k L)nat ,
so the value of F required to sustain V0 will also be very large. In another experiment,
the machine applies the constant amplitude force F. In that case, Eq. (11) gives the
displacement U . Setting V0 = iωU in Eq. (7) gives the corresponding acoustic force
Fac applied to the plate,
Z ac
Fac = F (12)
Z M + Z ac
The value of Fac / (H W ) serves as a metric for the pressure within the cavity. In the
second experiment, the displacement and pressure within the cavity are very large at
frequencies for which Z M + Z ac = 0; this is the fluid-structure resonance condition.
The contrasting situation places the frequency at one of the natural frequencies of
the cavity, as given in Eq. (10). At these frequencies, Z ac is very large. Equation (11)
gives a small value for U in this case, and Eq. (12) gives Fac = F. It follows that when
we talk about a resonance, we must be unambiguous in specifying how the system is
driven. To some extent, we encountered this same issue when we considered closure
of one-dimensional waveguides, specifically in Example 3.4.
EXAMPLE 10.2 A rectangular tank of water has width and height 4a. The
bottom and side walls are rigid, and the top is a free surface. A circular piston
transducer whose radius is a is flush mounted on the rigid wall at x = L,
so that Vx = V0 exp (iωt) on the piston face and zero otherwise. Derive an
expression for the pressure at an arbitrary location as a function of ka. Then
evaluate | p| / (ρ0 cV0 ) at x = 0, L/2, and L on the centerline, y = z = 2a, for
the frequency range 0 < ka < 8. Also evaluate the face-averaged impedance
seen by the transducer. The aspect ratio for the evaluation is L/a = 5.
4a
x
2a
2a L y
Figure 1.
Significance
From a conceptual viewpoint, the system in this example is very much like the one
analyzed in the previous example. The primary difference is the fact that forced cavity
modes covering a range of wavenumbers in both the y and z directions are excited
here. The other significant difference from the previous example is that numerical
methods will be required to evaluate the modal velocity coefficients. A different view
of the behavior at resonances will emerge.
Solution
Identification of symmetry properties is a useful starting point. The vertical plane that
contains the centerline cuts the tank in half, and the walls have the same properties
on either side of this plane. Furthermore, the input velocity field is symmetric with
respect to this plane. Consequently, the pressure field must share this symmetry.
The normal to the symmetry plane is the y direction, so only forced cavity modes
that are even functions with respect to y = 2a will participate in the response. The
walls at y = 0 and y = 4a are rigid. The full set of two-dimensional modes for rigid
walls separated by distance H is cos (sπ/H ), s = 0, 1, 2, ... The symmetric ones
correspond to even s. In contrast, there is no symmetry with respect to any horizontal
plane, so the full set of rigid/pressure-release two-dimensional modes are required
to describe the field’s dependence on z. The length of both sides of the cross section
is 4a, so the normalized transverse modes for the analysis are

2 jπ y (2n − 1) πz
j,n = Bn, j cos cos ; j = 0, 1, 2, ..., n = 1, 2, ...
4a 2 (4a)

4
B j,n =
1 + δ jn,0
(1)
The associated axial wavenumbers are
2 2 1/2
jπ (2n − 1) π
κ j,n = k − 2
− (2)
2a 8a
As always, the square root should be such that κ j,n is either positive real or negative
imaginary.
For the sake of variety, the boundary excitation has been situated at x = L. The
piston’s axial velocity is Vx = V0 over its face and Vx = 0 for points that are not on
the piston. Consequently, the integrand for the modal velocity coefficients is zero
outside the piston, so that

4a
4a
1 V0
V j,n = Vx j,n dydz = j,n dydz (3)
16a 2 0 0 16a 2 Apiston
In view of the circular shape of the piston’s face, it is logical to use polar coordinates
centered on the piston to formulate the integral. The coordinate transformation is
y = 2a + R cos θ, z = 2a + R sin θ
The face of the piston corresponds to 0 < R < a, −π < θ < π, which converts the
velocity coefficients to

a
π
1 jπ R
V j,n = Bn, j cos 2 + cos θ
16a 2 2 a
0 −π (4)
π R
× cos (2n − 1) 2 + sin θ Rdθd R
8 a
No analytical result for these integrals is readily available, so they are evaluated
numerically. Some software packages contain a double integration routine. (In
MATLAB, it is dblquad.) Otherwise, the integration may be performed by dividing
a circle into a set of small patches subtending R and θ. The double integral is
approximately the sum of the value of n, j at the center point of each patch, multi-
plied by the area of the patch. Figure 2 plots the values Vn,1 and V0, j . It can be seen
that the coefficients decrease, but slowly and in an oscillatory manner.
10
5 V0,j /V0
−5
Vn,1 /V0
−10
0 5 10 15 20
Mode number
Figure 2.
The values of V j,n are the last set of parameters required to form the modal series,
Eq. (10.2.10). If we wish to employ the derived formulas, we must recognize that
the x direction used for the derivation is reversed from the one we have used for the
present system, with the excitation occurring at x = L rather than x = 0. We may

adjust for this difference by replacing x in the formulas with L − x. The undriven
end is rigid, so we let ζ L be infinite in Eq. (10.2.10). This is handled by dropping

terms that do not contain ζ L in the numerator and denominator, which is κ j,n in
Eq. (10.2.9). Thus, the modal pressure series is transformed into
∞ ∞ −iκ j,n x
k e + e+iκ j,n x
P= ρ0 cVn, j n, j (y, z)
κ
j=0 n=1 j,n
e+iκ j,n L − e−iκ j,n L
(5)
∞ ∞
k cos κ j,n x
≡ ρ0 cVn, j n, j (y, z)
κ
j=0 n=1 j,n
i sin κ j,n L
Resonances correspond to the zeros of the denominator. This is examined in detail

in the discussion.
The face-averaged impedance, which is the average pressure on the piston face
divided by the average velocity, may be computed directly from the other parameters.
The average velocity amplitude on the piston is V0 . The average pressure is found as
an integral over the face of the piston,

1
Pav = P|x=L dydz
πa 2 Apiston
The pressure is described by the modal series, Eq. (5), which we evaluate at x = L. In
order to better understand the magnitude of the terms, we shall replace the complex
exponentials with equivalent forms derived from Euler’s identity. Thus, we form
∞ ∞
1 ka
Pav = ρ0 cV j,n cot κ j,n L j,n (y, z) dydz (6)
πa j=0 n=1 iκn, j a
2
Apiston
The integral appeared previously in Eq. (3). It follows that

V j,n
j,n (y, z) dydz = 16a 2
Apiston V0
The ratio of pressure to velocity is an impedance. Thus, the average acoustic

impedance seen by the piston may be determined by evaluating
∞ ∞
Pav ka V j,n 2
Z ac = = 16 ρ0 c cot κn, j L (7)
V0 j=0 n=1
iκ j,n a V 0
Evaluation of the pressure and average impedance requires computation of dou-

ble sums over the n and j mode numbers. Let J = max ( j) and N = max (n) be the
numbers at which the summations are halted. One criterion is that J and N may be
set such that V j,n is negligible for j > J and n > N . The slow oscillatory reduction
of V j,n indicated in Fig. 2 suggests that this criterion would require that J and N
be quite large. Another possibility is that the series lengths may be set according
to the behavior of κ j,n a. Examination of Eqs. (5) and (7) shows that convergence
of the pressure and impedance will be attained if the κn, j values for omitted modes
are large in magnitude. This condition will be attained if J is sufficiently large that
κ J,n is negative imaginary regardless of n, and N is sufficiently large that κ j, N a
is negative imaginary regardless of j. Equation (2) for κ j,n shows that this condi-
tion can be met if J > 2ka/π and N > 4ka/π + 1/2. The highest frequency for the
computations is ka = 8, which gives J > 6 and N > 11. Doubling these values is
more than adequate
because κ J, 1 a ≈ −17i and κ0, N a ≈ −15i, whereas the smallest
values of κn, j are close to zero for modes that are closest to the cutoff condition. A
by-product of selecting J and N in this manner is certainty that all modes that might
be resonant are included. This is so because κn, j of all omitted modes are negative
imaginary, so the exponential terms in the denominators are real.
The computational algorithm developed for a rectangular waveguide in
Example 9.7 is equally applicable here. Preliminary to the main frequency loop,
all quantities that are independent of frequency are evaluated. This entails setting up
an outer loop over j = 0, ..., J and an inner loop over n = 1, ..., N . The element
index is set at m = j N + n. Within these loops, the values of B̃m ≡ B j,n are com-
puted according to Eq. (1), and the double integral for Ṽm ≡ V j,n /V0 is carried out.
We are working nondimensionally, so the latter coefficients are given by

1
π
1
Ṽm = j,n (2 + cos θ, 2 + sin θ) R dθd R
16 0 −π
Another item that may be computed independently of the frequency is the value of
each transverse mode function on the centerline, y = z = 2a, at which the pressure
will be computed. These values form ˜ m ≡ j,n (2a, 2a), which are stored in a
diagonal matrix [m ].
The frequency loop finely increments the ka value between 0 and 8. At each
frequency, the loops over j = 0, ..., J and n = 1, ..., N are repeated. The index
m = j N + n is set as the subscript for storing arrays as column vectors. The nondi-
mensional axial wavenumber κ̃m a is obtained from Eq. (2). The pressure at the des-
ignated points x1 = 0, x2 = L/2,̃ and x3 = L and the impedance value are computed
by placing the pressure in Eq. (5) and the impedance coefficient in Eq. (7) into matrix
form as
P
= [E] [D]diag ˜m Ṽ
ρ0 cV0 diag
T (8)
Z av
= Ṽ [F]diag Ṽ
ρ0 c
where ⎛ ⎞
ka ⎜ ⎜ 1 ⎟
Dm,m = ⎟
iκm a ⎝ L ⎠
sin (κ̃m a)
a

L xs
E s,m = cos (κ̃m a)
a L
ka L
Fm,m = cot (κ̃m a)
iκm a a
The pressure amplitude at the designated locations is plotted as a function of

frequency in Fig. 3. A casual viewing might lead one to conclude that each plot
describes random noise. That this is not so is evidenced by the behavior in the lower
frequency range, where distinct resonance peaks can be identified. Each of the high-
frequency peaks also is caused by a resonance.
Thedenominators in Eqs. (5) and (7)
may be written equivalently as 2i sin κ j,n a (L/a) . Resonance corresponds to this
quantity being zero, which occurs if κ j,n a = (a/L) sπ, s = 0, 1, 2, .... In view of
the definition of κ j,n in Eq. (2), the resonance condition is marked by
nπ 2 2 a 2
(2 j − 1) π
+ + π s 2 = (ka)2 (9)
2 8 L
As the frequency increases, it becomes increasingly likely that there will be one
or more n, j pairs that satisfy Eq. (9). For example, the first resonance occurs at
ka = π/8 = 0.393 corresponding to n = 0, j = 1, and s = 0. This is contrasted by
the situation for ka = 7.54, for which sin κ j,n a (L/a) < 0.05 for n = 2, j = 4
(s = 10), n = 1, j = 9 (s = 5), and n = j = 4 (s = 5). The fact that the resonances
become very evidenced by the fact that a slight shift to ka = 7.55
close is further
leads to sin κ j,n a (L/a) < 0.05 for n = 0, j = 1 (s = 12), as well as n = 2,
j = 4 (s = 10).
102
x/L = 0
10-2
|P|/(0 cV0)
102
x/L = 0.5
10-2
102
x/L = 1
10-2
0 1 2 3 4 5 6 7 8
Frequency ka
Figure 3.
Figure 4 zooms in on the low-frequency range. With one exception, a resonance

frequency is manifested by the pressure at each location showing a peak. The excep-
tion occurs at ka = 0.74, where the pressure at x = L/2 does not show a peak.
This is a consequence of the fact that the resonant (0, 1) mode corresponds to
cos (κ̃m (x/L) (L/a)) = 0 at x = L/2. Hence, its does not contribute to the pressure
at this location even though the amplitude of this mode is large. Unlike resonances,
sharp minima of |P| are not correlated between different locations. These occur
because the many terms in the modal sums accumulate to a very small value, and the
terms in that sum depend on position.
104
|P|/(0 cV0)
102
100 x/L = 0
x/L = 0.5
x/L = 1
10-2
0 0.5 1 1.5 2
Frequency ka
Figure 4.
The Vn, j coefficients given in Eq. (4) are real, so Z ac is imaginary. This means
that Z ac is reactive, as it was in the previous example. The average impedance on the
piston face is depicted in Fig. 5. (The dashed lines describe the discontinuous drop
from an infinite inertance to zero compliance as a natural frequency is passed.) The
overall appearance of Fig. 5 is like Fig. 1 of the previous example, except that the
spacing between natural frequencies is irregular here.
1000
Im(Zac)/(0c)
500
0
−500
−1000
0 0.5 1 1.5 2
Frequency ka
Figure 5.
10.2.2 Spherical Cavities
One seldom encounters a spherical room, although the author does recall visiting a
combat flight simulator at an aerospace company in which images were projected
on a nearly complete spherical wall. Nevertheless, being able to analyze waves in
an enclosed spherical space affords us the opportunity to examine the differences
between different types of cavities and thereby understand how curved boundaries
affect basic phenomena. We first encountered a spherical cavity in Chap. 6. The earlier
treatment was limited to radially symmetric waves. The development of spherical
harmonics in Chap. 7 makes it possible to generalize the analysis.
It might seem that a spherical domain is unlike the interior of a box, but there are
many similarities. Although a sphere has one physical boundary, whereas a box has
six walls, there are additional boundary conditions for the former that play the roles
of walls. One we encountered in the study of radially symmetric waves, wherein

a primary consideration is that the general solution of the wave equation must not
be singular at the center. Other conditions are continuity. If we know the complex
amplitude P at a set of spherical coordinates (r, ψ, θ), an increase or decrease of θ by
2π must yield the same values for the pressure and particle velocity. (This condition
is identically met by an axisymmetric field, which is the situation we will consider.)
Continuity also is an issue at the poles, because the pressure and particle velocity at
ψ = 0 and ψ = π must be finite.
Fortunately, these issues are addressed by a spherical harmonic series representa-
tion. Thus, the starting point for our exploration is the general solution for axisym-
metric waves in a spherical geometry. This was the subject of Sect. 7.1.
Axisymmetric Excitation
We wish to determine the pressure in the region enclosed by a vibrating spherical wall.
The wall’s motion is harmonic. A further restriction is that the motion is limited to
being axisymmetric. In accord with our standard practice we designate the symmetry
axis as z. Axisymmetry of the excitation
leads to axisymmetry
of the response, so
the pressure field is p (x̄, t) = Re P (r, ψ) exp (iωt) . The general solution of the
Helmholtz equation in axisymmetric spherical coordinates is a series of spherical
harmonics. Each harmonic is a product of a radial function and a polar function.
The polar angle is ψ, but the mathematical variable that was used in Chap. 7
was the transformed variable η = cos ψ. Some individuals view η as the projection
onto the z axis of a unit radial line at angle ψ. The angular functions are Legendre
polynomials, Pm (η). When η is replaced by its definition, the result is that the angular
functions are Pm (cos ψ). These are said to be Legendre functions. An analysis of
the Helmholtz equation usually will proceed equally well using either ψ or η.
Radiation exterior to a sphere causes waves to propagate outward. A spherical
Hankel function h (2)
m (kr ) is appropriate to this situation because at long distances it
approaches a proportionality to exp (−ikr ). This function is not appropriate by itself
for the interior problem because it is singular at the origin, r = 0. The singularity of
this solution could be canceled by adding the Hankel function h (1) m (kr ) for an inward
propagating spherical wave. Such a representation would lead to an analysis that
parallels the analysis of a rectangular cavity. Finiteness at r = 0 would be the analog
of the boundary condition for the wall opposite the driven one in the rectangular
geometry.
The merits of seeing this analogy are counterbalanced by the complications asso-
ciated with dealing with (complex) Hankel functions. The Helmholtz equation is
linear and real, so the real and imaginary parts of each Hankel function satisfy that
equation individually. The singularity at the origin occurs in the imaginary part, so
we discard it. The result is that the radial part of a spherical harmonic for an enclosed
region is a spherical Bessel function jm (kr ). The order m is the same as the degree
of one of the Legendre functions.
A general solution for the field within a sphere is obtained by multiplying the
product of each Legendre function and the corresponding spherical Bessel function
by a coefficient whose value must be determined. Thus, the basic ansatz for the
complex pressure amplitude is
∞

P= Cm jm (kr ) Pm (cos ψ) (10.2.14)
m=0
Before we determine the series coefficients, let us consider the finiteness and conti-
nuity conditions that were noted at the beginning of the development. Finiteness was
addressed by using only the spherical Bessel function for the radial dependence. This
condition also led to truncation of the Legendre polynomial, without which the series
would have diverged at ψ = 0 and ψ = π. The fact that the preceding series repre-
sents a continuous pressure field results identically from the axisymmetry property.
However, there is another finiteness condition that applies to the particle velocity.
The velocity components are found from Euler’s equation to be
∞
i
Vr = Cm jm (kr ) Pm (cos ψ)
ρ0 c m=0
∞ (10.2.15)
i
Vψ = − sin ψ Cm jm (kr ) (sin ψ) Pm (cos ψ)
ρ0 ωr m=0
There is nothing irregular about these expressions, but that is only because Pm
(cos ψ) ≡ 0 at ψ = 0 and ψ = π. Suppose this was not true. Points on opposite
sides of the polar axis correspond to ψ = ε at azimuthal positions θ and at θ + π.
In the limit as ε → 0, the value of Vψ would be the same for both points, but the
direction of ēψ is opposite. This would correspond to an infinite velocity gradient.
According to the continuity equation, such a condition corresponds to an infinite
pressure.
Now that we know that the spherical harmonic series satisfies all finiteness and
continuity conditions, we may proceed to evaluate the coefficients. They are deter-
mined by matching the particle velocity to the vibration of the spherical wall. The
normal to the wall is −ēr . Our convention is that the surface normal points into the
fluid, so we consider positive values of the wall’s normal velocity to be inward. Thus,
we require that
vr = − Re V̂S (ψ) eiωt at r = a (10.2.16)
The amplitude VS may be complex, in which case it represents a motion of the surface
in which not all points vibrate in-phase.
Our task now is to determine the coefficients Cm of the pressure series by satisfying
the velocity boundary condition. This is done by using a Legendre function series
to describe Vs (ψ). Doing so yields a representation that is consistent with the above
expression for Vr of the fluid. Thus, we write
∞

V̂S = Vm Pm (cos ψ) (10.2.17)
m=0
An expression for the coefficients of this series results from application of the orthog-
onality property of Legendre functions, Eq. (7.1.16). To that end, we multiply the
series by P j (cos ψ) sin ψ, where j is a specific index, and then integrate over the
full range of ψ. Changing the index j back to m gives

π
2m + 1
Vm = V̂S Pm (cos ψ) (sin ψ) dψ (10.2.18)
2 0
We equate the radial velocity at r = a, which is obtained from the first of

Eq. (10.2.15) to the imposed velocity in Eq. (10.2.17). We could apply the orthogo-
nality property, but a briefer approach argues that the Legendre functions constitute
a complete set of linearly independent basis functions. Therefore, it must be that the
series coefficients match, which leads to
i
Cm jm (ka) = Vm (10.2.19)
ρ0 c
The resulting description of the pressure is
∞
jm (kr )
P = ρ0 c Vm Pm (cos ψ) (10.2.20)
m=0
i jm (ka)
Earlier we identified resonances as singularities of a frequency-domain description.

The singularities of the preceding expression correspond to zeros of the denominator,
so we have identified that the mth resonance occurs when
m
jm (x) x=ka ≡ jm (ka) − jm+1 (ka) = 0 (10.2.21)
ka
This condition can occur for any spherical harmonic. Furthermore, the spherical
functions are oscillatory, so there are multiple roots for a specific m. Thus, there is a
double infinity of ka values at which the cavity will resonate.
Proper evaluation of the series requires that an adequate number of terms be
included. Resonances are an important factor. For a selected value of ka, there are
values of m for which the resonance condition is close to being satisfied. However,
jm (ka (r/a)) /jm (ka) in Eq. (10.2.20) decreases monotonically with increasing m,
provided that m is substantially larger than ka and r/a ≤ 1. Thus, series truncation
at a value of m that is a multiple of ka, such as max (m) = 4ka, should be adequate.
Another aspect of series truncation is the behavior of the Vm coefficients. At some
value of m, these values will decrease because the corresponding Legendre function
series represents a nonsingular Vs function. Usually, truncation based on the behavior
of the Bessel functions will assure inclusion of a sufficient number of terms.
In addition to the acoustic pressure, power transfer is an important physical aspect.
The instantaneous radial intensity at a point on the wall is

Ir = P Vs e2iωt + P Vs∗ r =a + c.c.
∞ ∞
V j Vm e2iωt + V j Vm∗ jm (ka)
= ρ0 c + c.c. P j (cos ψ) Pm (cos ψ)
j=0 m=0
i jm (ka)
(10.2.22)
According to this expression, at any point on the wall power might be flowing out
of, or into, the wall. The power transfer across a patch surrounding this point might
have a nonzero mean value. The total power flow is obtained by integrating this
expression over the entire wall. An element of area for this axisymmetric situation is
dS = 2πa 2 (sin ψ) dψ. Orthogonality of the Legendre functions filters out the cross
terms. In addition, Vm Vm∗ is a real quantity, so the complex conjugate part cancels
this term. The result is that the instantaneous power flow is
∞

Vm2 jm (ka)
P = 4πa ρ0 c
2
e2iωt + c.c. (10.2.23)
m=0
i (m + 1) jm (ka)
From this, we conclude that the average total power transfer is zero. Each term
fluctuates about zero at twice the excitation frequency, but all terms are not necessarily
in-phase, so there might be no instant at which P = 0.
EXAMPLE 10.3 Consider a plane wave that is external to a spherical con-

tainer. Suppose it induces a radial velocity of the wall that is proportional to the
external pressure. Evaluate the pressure field within the container at ka = 2
and ka = 5.
Significance
The notion that the wall velocity is proportional to the external pressure is a vast
simplification. For one, it ignores the fact that the presence of the sphere induces an
exterior scattered pressure field. Nevertheless, the stated conditions serve to create a
situation from which we may learn much. The analysis will also help us sharpen our
computational skills.
Solution
The direction in which the plane wave propagates defines the symmetry of the field,
so we let this direction be parallel to the z axis. When the origin of x yz is placed
at the center of the cavity, the complex pressure amplitude of the incident wave is
P0 exp (−ik x̄ · ēz ) = P0 exp (−ikz). For a point on the surface, z = a cos ψ, so we
begin by setting Vs = V0 exp (−ika cos ψ).
The velocity coefficients for the Legendre series representation of Vs are

π
2m + 1
Vm = V0 e−ika cos ψ Pm (cos ψ) sin (ψ) dψ
2 0
In general, we cannot expect that a formula for the integral is available, but it is for
this case.2 The result is
Vm = (2m + 1) e−mπi/2 jm (ka)
If this formula were not available, we would evaluate the coefficients numerically.
The MATLAB procedure would entail setting up a loop ranging from m = 0 to m =
max (m). For each m, the integrand would be evaluated with an anonymous function
that depends on η, such as Coeff_int=@(eta) exp(-1i*ka*eta).*leg
(m,eta), where leg.m was described in Example 7.1. The quadl function
would yield the coefficient according to V(m + 1) = 0.5*(2*m+1)*quadl
(Coeff_int,-1,1,1e-6);
The Vm values for ka = 2 and ka = 5 are plotted in Fig. 1. The largest values
are centered on a harmonic number that is slightly larger than ka. Beyond that
maximum, the coefficients fall off rapidly. According to the above formula, Vm is a
real quantity if m is even, and it is imaginary if m is odd. To understand why this
is so, consider the integrand in conjunction with writing the complex exponential as
cos (ka cos ψ) − i sin (ka cos ψ). With respect to ψ = π/2, the first term is an even
function, whereas the second is an odd function. At the same time, Pm (cos ψ) sin ψ
is even with respect to π/2 for even m and odd for odd m. The integral from 0 to π of
the product of even or odd functions is nonzero, whereas the integral of the product
of an even and an odd function is zero. This parity behavior will be seen to have an
interesting consequence in regard to the pressure distribution.
2
1.5 ka = 5
|Vm|
0.5
ka = 2
0
0 5 10 15 20
Figure 1.
Figure 1 demonstrates that the Vm values are negligible well before M = max
(m) = 4ka, which is the suggested rule to truncate the series. The manner in which
the pressure field is to be displayed affects the algorithm that is used to evaluate
the series. We will construct contours of equal Re (P (x̄)) and Im (P (x̄)). Doing
so entails evaluating the pressure at a set of grid points in axisymmetric spherical
coordinates and then calling on the graphics capabilities of our chosen mathematical
software. Typically, a contouring graphics routine expects the position coordinates
to be Cartesian. Thus, we define a set of r and ψ values at equal increments, rn /a =
nr and ψ = ψ . The corresponding Cartesian coordinates of the grid points are
2 M.I. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, p. 440,
Eq. 10.1.47 (1965).

z n, /a = nr cos ψ and xn, /a = nr sin ψ . (Because of axisymmetry, it
is adequate to view the field in the x z plane.)
Equation (10.2.20) describes P at each grid point. The matrix algorithm for eval-
uating a modal series at any points is suitable for the present situation. Thus, we
define a diagonal matrix [K ] whose elements are K m,m = Vm /jm (ka), a rectangu-
lar matrix [E] whose elements are E n,m = jm (krn ), and a rectangular matrix [F]
whose elements are Fm, = Pm (cos ψ ). The pressure at all grid points is found by
computing
ρ0 c
[P] = [E] [K ] [F]
i
Figure 2 shows the contours of Re (P) and Im (P) for ka = 2. The distribution
of Re (P) is antisymmetric with respect to the plane ψ = π/2, and the distribution
of Im (P) is symmetric with respect to that plane. This behavior is a consequence
of the real/imaginary property of the Vm coefficients. Both [E] and [F] are real, and
[K ] is complex solely because its elements are proportional to the Vm coefficients.
For even m, Vm is real and Pm (cos ψ) is even with respect to ψ = π/2. Division by i
results in imaginary quantities that are even. The converse occurs for odd m, because
Vm is purely imaginary and Pm (cos ψ) is odd with respect to ψ = π/2. Thus, the
real pressure contours stem from the odd m terms, whereas the imaginary contours
are the contributions of the even m terms.
Re(P) Im(P)
ka = 2
o o o o
= 180 = 0 = 180 = 0
−20 −10 0 10 20 0 0.5 1 1.5

Figure 2
Figure 3 shows the contours for ka = 5. The even/odd properties displayed here
are the same as those for Fig. 2. There is more fluctuation in the values from point to
point, which is to be expected because the frequency is higher. What is unexpected
is that the overall pressure level is lower. Also notable is the fact that the real and
imaginary parts of P are comparable, whereas Re (P) is much greater than Im (P)
at ka = 2.
Re(P) Im(P)
ka = 5
= 180o = 0o = 180o = 0o
−5 0 5 −6 −4 −2 0 2
Figure 3.
The magnitude differences are a manifestation of resonance. To explore this

aspect, let us return to the pressure series, Eq. (10.2.19). To compensate for differ-
ences of the overall Bessel function magnitudes for different order m, let us rewrite
that equation in terms of a relative coefficient (Bm )rel , such that
jm (ka)
(Bm )rel = Vm
i jm (ka)
∞
jm (ka (r/a))
P = ρ0 c (Bm )rel Pm (cos ψ)
m=0
jm (ka)
The ratio of Bessel functions in the summation is always O(1), so (Bm )rel is a measure
of the amount each term contributes to the summation. Figure 4 indicates that the
dominant contribution at ka = 2 comes from the m = 1 harmonic. The large value of
(B1 )rel stems from j1 (ka) being very small; ka = 2.0816 is a zero of j1 (ka). Because
this is a term for which m is odd, the complex pressure amplitude is essentially a real
value everywhere. In other words, it is essentially in-phase. An in-phase pressure
distribution, as well as large pressure values, is hallmarks of a cavity resonance.
At ka = 5, the primary contributors are the m = 3 and m = 4 spherical harmonics.
The values of (B3 )rel and (B4 )rel are comparable in magnitude, which is why Fig. 3
indicates that the real and imaginary parts of P (x̄) have similar magnitudes overall.
The ratio of these parts is not constant over the field, which means that the pressure
field does not consist of an in-phase oscillation. We conclude that ka = 5 is not very
close to a resonance.
30
|(Bm)rel|
20
ka = 2
10
ka = 5
0
0 5 10 15 20
Figure 4.
To close this example, it is useful to take an overview. The excitation of the

spherical cavity is a wavelike particle velocity at the wall. Nevertheless, there is no
evidence of a comparable wave process in the pressure field. The field at any ka
will consist of a complex pressure amplitude that combines real and imaginary parts.
The former is antisymmetric with respect to the plane ψ = π/2, whereas the latter
is symmetric with respect to that plane. If ka is close to a resonance, one part will be
dominant and the pressure will show an overall maximum magnitude. Off-resonance,
the mix of real and imaginary parts results in an instantaneous pressure distribution
that fluctuates throughout a period because of the different parities of the real and
imaginary parts of the complex amplitude.
Concentric Spheres
Systems in which the acoustic field is the region between concentric spheres are not
common. Nevertheless, their analysis serves a number of purposes. The radius of
the inner sphere is b, so the acoustic domain is b < r < a. The excitation might be
applied to either the inner or outer wall as a specified harmonic radial velocity or
pressure oscillation. The undriven surface might be rigid or pressure-release. In order
to avoid unnecessary complications, we will not consider the alternate possibility that
the undriven wall is locally reacting.
It is necessary to satisfy a boundary condition at both walls, but the center, r = 0, is
not contained in the region occupied by the fluid. This means that we cannot exclude
a solution of the spherical Bessel equation that is singular at the center. We could
use both types of spherical Hankel functions, h (1) (2)
m (kr ) and h m (kr ), to form a spher-
ical harmonic series. These functions, respectively, represent inward and outward
propagation. However, as we saw in the previous example, the field in a cavity tends
to have a standing wave attribute. Adding and subtracting the two types of Hankel
functions lead to the spherical Bessel and Neumann functions, jm (kr ) and n m (kr ),
see Eq. (7.1.31). A spherical harmonic series for the complex pressure amplitude
results from multiplying each function at a specific m by a different coefficient and
then multiplying the sum of these terms by a Legendre function. In other words,
∞

P= [Cm jm (kr ) + Dm n m (kr )] Pm (cos ψ) (10.2.25)
m=0
The corresponding radial particle velocity is

∞
i
Vr = Cm jm (kr ) + Dm n m (kr ) Pm (cos ψ) (10.2.26)
ρ0 c m=0
The restriction to the passive boundary being rigid or pressure-release, accompa-

nied by allowance that the driven sphere is either the inner or outer boundary, admits
four possible sets of boundary conditions. We will address the case where the inner
sphere is rigid and the outer sphere is driven. The modifications required to address
the other cases will be evident. We let Va (ψ) be the complex amplitude of the radial
velocity at r = a. Our convention takes a wall velocity to be positive if it points
into the fluid, so Va > 0 represents a velocity that is in the direction of −ēr . Thus,
velocity continuity at the boundaries requires that
∞

Cm jm (ka) + Dm n m (ka) Pm (cos ψ) = iρ0 cVa (ψ)
m=0
∞
(10.2.27)

Cm jm (kb) + Dm n m (kb) Pm (cos ψ) = 0
m=0
As we did for the case of a full spherical cavity, we expand Va in a Legendre

function series,
∞

Va = Vm Pm (cos ψ) (10.2.28)
m=0
The series coefficients are

π
2m + 1
Vm = Va Pm (cos ψ) (sin ψ) dψ (10.2.29)
2 0
The Legendre function series for Va is substituted into the velocity boundary
conditions, Eq. (10.2.27). Application of the orthogonality properties of the Legendre
functions leads to
Cm jm (ka) + Dm n m (ka) = iρ0 cVm
(10.2.30)
Cm jm (kb) + Dm n m (kb) = 0
Solution of these equations yields
n m (kb)
Cm = iρ0 c Vm
(ka, kb) (10.2.31)
j (kb)
Dm = −iρ0 c m Vm
(ka, kb)
where
m (ka, kb) = jm (ka) n m (kb) − jm (kb) n m (ka) (10.2.32)
Resonances correspond to singular values of a Cm and Dm . They occur when

m (ka, kb) = 0. This condition occurs for each harmonic. As was true for the com-
plete spherical cavity, the oscillatory nature of the Bessel and Neumann functions
causes m to have zeroes at an infinite number of k values. Because there are an
infinite number of harmonics, there are a doubly infinite set of resonances.
EXAMPLE 10.4 A concentric inner sphere in a spherical cavity executes a

radially symmetric vibration, so that vr = v0 cos (ωt) over its entire surface.
The outer sphere is rigid. (a) Evaluate the radial dependence of the pressure
for ka = 2, 8, and 32 in the cases where b = 0.5 and b = 0.1. (b) Evaluate
the dependence of |P| on ka at r = a, r = (a + b) /2, and r = b for the case
where b = 0.1a. (c) Consider the limit as b/a → 0 with the source’s volume
velocity, Q̂ = 4πb2 v0 , held constant. Derive an expression for the pressure
within the sphere.
Significance
The analysis will connect the present development with our investigations of radi-
ally symmetric waves and sources. The outcome will be enhanced understanding of
acoustic behavior of cavities.
Solution
Radially symmetric situations correspond to the m = 0 spherical harmonic. Although
the formulas derived in this section are quite general, the simplicity of the m = 0
relations allow us to formulate the solution in terms of elementary functions. For
m = 0, we have j0 (x) = sin (x) /x and n 0 (x) = − cos (x) /x. Thus, the m = 0
spherical harmonic for the region between concentric spheres is
1
P= [C0 sin (kr ) − D0 cos (kr )] (1)
kr
The complex radial velocity on the inner sphere is v0 , and the outer sphere is
stationary, so the boundary conditions are

d P d P
= −iρ0 ωv0 , =0
dr r =b dr r =a
Substitution of Eq. (1) into these conditions gives
1
[kb cos (kb) − sin (kb)] C0 + [kb sin (kb) + cos (kb)] D0 = −iρ0 ωv0
kb2
1
[ka cos (ka) − sin (ka)] C0 + [ka sin (ka) + cos (ka)] D0 = 0
ka 2
Introduction of the identities for the sine and cosine of the difference of two angles
reduces the solution of these equations to
ka sin (ka) + cos (ka)

C0 = −iωρ0 kb2 v0
0 (ka, b/a)
(2)
ka cos (ka) − sin (ka)
D0 = iωρ0 kb2 v0
0 (ka, b/a)
where

b b b b
0 (ka, b/a) = (ka) + 1 sin ka 1 −
2
− ka 1 − cos ka 1 −
a a a a
(3)
An expression for P as a function of r results from substitution into Eq. (1) of the
coefficients in Eq. (2). Introduction of ω ≡ kc and b ≡ a (b/a) yields a nondimen-
sional form that exhibits the fundamental parametric dependencies,
(b/a)2 ka
P = −iρ0 cv0 [ka cos (ka (r/a − 1)) + sin (ka (r/a − 1))]
r/a 0 (ka, b/a)
(4)
According to this expression, the complex pressure amplitude everywhere is purely

imaginary. This means that the pressure is proportional to sin (ωt). The radial velocity
of the inner sphere is a cosine, so the pressure is in-phase with the surface acceleration.
To obtain the radial pressure distributions called for in Part (a), we evaluate Eq. (4)
for fixed b/a and ka over b/a < r/a < 1. The radial profiles of pressure in Fig. 1
are typical. The overall magnitude of each profile is set by the factors preceding the
bracketed term in Eq. (4). The value of 0 is especially important in this regard.
In most cases, the increase of pressure with decreasing radial distance is evident.
However, there is little dependence on r/a at low ka, regardless of the value of
b/a. This behavior is consistent with our general expectation for low-frequency
oscillations in a cavity. At fixed b/a, the radial interval over which the pressure
fluctuates decreases with increasing frequency. Note that d P/dr is zero at r = a, as
required by that boundary being rigid.
20
Im(P/(0 cv0 ))
10 b/a = 0.5
0
−10
−20
0.5 0.6 0.7 0.8 0.9 1
Radial distance, r/a
200
Im(P/(0 cv0 ))
0
ka = 2
−200 ka = 8
b/a = 0.1 ka = 32
−400
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Radial distance, r/a
Figure 1.
Frequency response functions offer a different view. Figure 2 shows the result at
three locations for the case b = a/10. The peaks are resonances at which 0 = 0.
Resonance is a system feature, so they are manifested at each of the radial dis-
tances. The nulls are frequencies at which the numerator in Eq. (4) vanishes for
each r . For a given value of b/a, the occurrence of this condition depends on both
ka and r/a, so the nulls are not aligned. The frequencies at which nulls occur
at r = (a + b) /2 become increasingly close to the resonances as the frequency
increases. This behavior is a consequence of b/a being quite small in the present
example. When ka 1, b/a 1, and r = (a + b) /2, the numerator in Eq. (4) for
r = (a + b) /2 is approximately proportional to ka cos (ka/2) and 0 is approxi-
mately proportional to (ka)2 sin (ka) ≡ 2(ka)2 sin (ka/2) cos (ka/2). Resonances,
at which 0 = 0, occur approximately at sin (ka/2) = 0 and cos (ka/2) = 0. The
latter are approximately the same as the zeros of the numerator. When all terms are
included, the values of ka at which the numerator for r = (a + b) /2 is zero are close
to some frequencies at which 0 = 0.
r=a r=(a+b)/ 2 r=b
102
|P/(0 cVa )|
100
0 5 10 15 20 25 30 35 40
Nondimensional frequency (ka)
Figure 2.

To evaluate the limit of a very small sphere, we use v0 = Q̂/ 4πb2 in order to
remove v0 from Eq. (4). Doing so gives
Q̂ a ka
P = −iρ0 c [ka cos (ka (r/a − 1)) + sin (ka (r/a − 1))]
4πa r 0 (ka, b/a)
2
(5)
This expression does not contain b, so it is valid in the limit as b → 0. Note that
the limiting form requires that the inner sphere be present, so r = 0 is excluded. In
effect, Eq. (5) is the solution for a spherical cavity with a point source at its center.
The pressure is singular at the source.
10.2.3 Cylindrical Enclosures
The mathematical representation of the pressure field in a rectangular cavity does not
depend on which wall vibrates. For a spherical cavity, there is only one mathematical
form. In contrast, the mathematical nature of the field in a closed cylinder depends
on whether the excitation is an imposed vibration of an end cap or the cylindrical
surface. We will explore both possibilities. For the sake of simplicity and brevity, only
axisymmetric situations are addressed. More general situations are addressed by a
simple adaptation of the present results. This is so because a Fourier series description
of an arbitrary θ dependence uncouples due to orthogonality of the circumferential
harmonics.
Vibrating End Wall
We consider the end x = 0 to execute a specified vibration in an axisymmetric pat-
tern, so that Vx = Vs (R). The end x = L is a locally reacting surface with specific
impedance ζ L . The transverse mode functions within a cylinder were derived in
Sect. 9.3.3. The axisymmetric modes are

R η0, j
0, j (R) = B0, j J0 η0, j , μ0, j = (10.2.33)
a a
The B0, j values in Eq. (10.2.33) are selected to normalize the mode functions accord-
ing to Eq. (9.3.67).
The η0, j quantities are eigenvalues. The characteristic equation from which they
are obtained results from satisfying velocity continuity at the cylindrical wall, where
the specific impedance is ζa . Equation (9.3.57) is the result that was derived in
Chap. 9. Because only the n = 0 harmonic is being considered,
that
expression may
be simplified by exploiting the identity that J0 η0, j ≡ −J1 η0, j . The characteristic
equation correspondingly may be written as
iζa
J0 η0, j + η0, j J1 η0, j = 0 (10.2.34)
ka
Evaluation of the modes was discussed in Sect. 9.3.3. The eigenvalues are
sequenced in ascending order of their real part. In the case where the wall is rigid,
j = 0 is the smallest eigenvalue, η0,0 = 0. This is the plane wave mode. It cannot
exist if the cylindrical wall is not rigid, because a pressure in the wall would induce
a particle velocity in the ē R direction, whereas the velocity in a plane wave is solely
in the x direction.
If the cylindrical wall impedance is a compliance, that is, if ζa is a negative
imaginary value, then there is one imaginary eigenvalue. That mode is designated
as j = 0. The modal series that we will develop will assume that a mode for j = 0
exists. If it does not, then the associated modal series coefficient should be set to
zero.
In the axial direction, we are presented with the same situation as that for a
rectangular enclosure, in that each transverse mode is associated with a pair of waves
that propagate forward and back in the x direction. Hence, a modal series will have
the same form as Eq. (10.2.2), except that a single summation is used because only
the n = 0 azimuthal harmonic is excited. The representation of the pressure field is
∞

P= α0, j e−iκ0, j x + β0, j e+iκ0, j x 0, j (R) (10.2.35)
j=0
The axial wavenumbers are the values that are set by the requirement that each term
independently satisfies the Helmholtz equation, which gives
1/2
κ0, j = k 2 − η0, j /a (10.2.36)
The series accounts for waves that propagate in both directions, so we may retain
the
requirement
that the square root be selected such that Re κ0, j ≥ 0 and Im
κ0, j ≤ 0. In turn, this requires that the η0, j values be in the first or second quadrant
of the complex plane, which includes values on the positive real axis.
The steps leading to the determination of the modal coefficients parallel those
for analysis of a rectangular cavity. The boundary conditions at the driven end and
the opposite passive end are like those in Eq. (10.2.4). Consequently, the results
may be derived from those for the rectangular case by modifying any variables
associated with the transverse behavior to match the circular geometry. The analog
of Eq. (10.2.7) is an inner product over the circular cross section. A differential
element of area for an axisymmetric integrand may be taken to be dA = 2π Rd R,
and the cross-sectional area is πa 2 , so the modal velocity coefficients are

a
1
V0, j = 0, j (R) Vs (R) 2π Rd R (10.2.37)
πa 2 0
Aside from the fact that the transverse modes are those for a cylindrical waveguide,
the pressure field is the same as Eq. (10.2.10),
∞
kV0, j κ0, j iκ0, j (L−x)
P = ρ0 c 1 + ζL e
κ0, j κ0, j k (10.2.38)
j=1
κ0, j −iκ0, j (L−x)
− 1 − ζL e 0, j (R)
k

The quantity κ0, j in the denominator is the same as the determinant in
Eq. (10.2.9), except that the axial wavenumbers are those for the cylinder,
κ0, j iκ0, j L κ0, j −iκ0, j L

κ0, j = 1 + ζ L e + 1 − ζL e (10.2.39)
k k
As was true for a rectangular cavity, resonances correspond to = 0. This con-

dition can occur only if Re(ζ L ) = 0 and κ0, j is real. This feature is evident in the
alternate form of obtained by application of Euler’s identity. It is
κ0,n
κ0, j = 2 cos κ0, j L + iζ L sin κ0, j L (10.2.40)
k
The procedure by which the natural frequencies may be analyzed when ζa or ζ L

is reactive is essentially the same as that discussed in regard to Eq. (10.2.13). The
primary difference is that the restriction to axisymmetric modes means that the trans-
verse mode function depends on a single eigenvalue η0, j , and the characteristic equa-
tion for that value features Bessel functions, rather than harmonic functions. The
natural frequencies are found by a frequency sweep ranging over k at fixed j. The
value η0, j is combined with the set of roots κ(m)
0, j for which κ0, j = 0 to evaluate
tentative natural frequencies according to
2 η 1/2
κ(m
0, j
(knat )0, j = 0, j + (10.2.41)
a
Any of these values that is very close to the value of k for the computations marks a
(m)
natural frequency, ω0, j = (knat )0, j /c.
EXAMPLE 10.5 The concept of an impedance tube developed in Sect. 3.2.4

is that measured properties of the field within a closed waveguide can be used to
identify the impedance of a material placed at one end of the tube. The concept
was derived on the basis of the field being planar in the transverse direction, but
that is not possible in reality. This is so because a planar waveguide mode exists
only if the sidewalls are rigid, which cannot be for actual materials. Consider the
impedance tube in Example 3.6. As described there, the tube is a cylinder whose
diameter is 120 mm, and L = 2 m, The pressure field is generated at x = 0 by a
vibrating piston whose diameter equals that of the tube. The specific impedance
of the material that covers the end at x = L was ζ L = 2.763 + 1.168i. (This
is the value used to generate the original data.) The fluid within the impedance
tube is water, ρ0 = 1000 kg/m3 , c = 1480 m/s. Determine and graph the mean-
squared pressure distribution along the centerline when the cylindrical wall is
compliant, with ζa = −5i. Compare the result to the axial distribution obtained
from an idealized rigid wall model. Frequencies to consider are 600 Hz, as
stated in the original problem, and 6 kHz.
Significance
Flexibility of the sidewall is a primary issue for the design of impedance tubes. From
an instructional standpoint, this example serves to connect the multidimensional
phenomena to our earlier studies of plane waves.
Solution
The first task is evaluation of the eigenvalues of the transverse mode functions. The
cylindrical wall’s impedance is a pure compliance. Consequently, the real eigenvalues
η0, j are the roots of Eq. (10.2.34). There also is one imaginary eigenvalue, η0,im =
iβa, where β is the root of Eq. (9.3.60) with n = 0. This is the type of mode for which
j = 0 is reserved. The first few eigenvalues for each frequency and corresponding
axial wavenumbers are tabulated below. The oscillatory nature of Bessel functions is
such that eigenvalues for the higher modes are well approximated by incrementing
η0,4 by multiples of π.
j
f (Hz) im 1 2 3 4
600 η0, j 0.5022i 3.7997 6.9981 10.1614 13.3145
κ0, j 0.7911 −3.7502i −6.9714i −10.1430i −13.3005i
6000 η0, j 1.8306i 3.5107 6.8409 10.0532 13.2319
κ0, j 6.3816 5.0048 −3.0700i −7.9808i −11.7349i
At 600 Hz, for which ka = 0.611, all modes except the one associated with η0,im are
cut off. Raising the frequency to 6 kHz only adds one mode to the set of real axial
wavenumbers.
The B0, j coefficients that normalize the transverse mode functions are described
in Eqs. (9.3.67) and (9.3.66). The result for n = 0 is
2 −1/2
η0, j ka 2
B0, j = + η0, j
J0 η0, j ζa
To evaluate the coefficients of the modal series for the velocity excitation, we use
Eq. (10.2.37) with Vs set to a constant value Vp for a vibrating piston. The resulting
integral is a standard one,3

V0, j 1 a 2B0, j
= B0, j J0 η0, j R 2π Rd R = J1 η0, j
Vp πa 2 0 η0, j
This expression indicates that the V0, j values decrease with increasing j, both as a
consequence of the overall decrease of Bessel functions as their argument increases
and as a consequence of the presence of η0, j in the denominator. Figure 1 describes
these values, with V0,im assigned to j = 0. Beyond j = 8, the coefficients are less than
1% of the largest at both frequencies. Thus, the modal series will omit terms for which
j > 8. (This actually is many more terms than necessary for series convergence,
because all modes for j > 2 at both frequencies are cut off. The corresponding
eigenvalues are negative imaginary, which will lead to (k, j) being very large.)
100 6 kHz
|V0, j|/Vp
600 Hz
10-2
10-4
0 1 2 3 4 5 6 7 8
Mode number j
Figure 1.
It is desired to evaluate the pressure at R = 0, where the zero-order Bessel func-

tion is one. Therefore, the on-axis pressure is obtained by setting 0, j = B0, j in
Eq. (10.2.38). Computation of the field at a set of points xm along this axis may be
performed by the matrix algorithm used previously. The V0, j /Vp values form column
vector {V }. A rectangular array [] holds the terms in Eq. (10.2.38) that depend on
axial position, that is,
κ0,n iκ (L−xm ) κ0,n −iκ0,n (L−xm )
0,n = 1 + ζ L e 0,n − 1 − ζL e
k k
Then, the complex pressure amplitudes at the successive locations are found by
evaluating

P k
= [] {V }
ρ0 cVp κ0,n κ0, j
diag
3 M.I. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, (1965)
Eq. 11.3.20, p. 484.

where κ0, j is described by Eq. (10.2.39). This computation is performed for both
frequencies.
The pressure for planar waves in a closed waveguide (with rigid walls) is given
in Eq. (3.2.21). It will be easier to interpret the results if this expression is written in
terms of ζ L rather than the reflection coefficient at z = L. Such a form is obtained
by multiplying that expression by ζ L + 1, which leads to
Prigid (x) (1 + ζ L ) eik(L−x) − (1 − ζ L ) e−ik(L−x)

=
ρ0 cVp rigid
rigid = (1 + ζ L ) eik L + (1 − ζ L ) e−ik L
The quantity used for an impedance

tube calculation is the mean-squared pres-
sure. The axial dependence of p 2 av is depicted in Fig. 2. Clearly, there is a drastic
difference between the behavior at the two frequencies, as well as between the rigid
waveguide model and the model that account for the compliance of the wall. To
understand why this is so, let us first focus on the results at 600 Hz.
3
2
All modes
(P )av /(0cVp)
Rigid wall
2
1 600 Hz
2
2
6 kHz
2
(P )av /(0cVp)
1.5
0.5
2
0
0 0.5 1 1.5 2
Axial distance x (m)
Figure 2.
We found that the zero mode, corresponding to η0,im , is the only one that is not
cut off at 600 Hz. For all other modes, κ0, j is a negative imaginary value. The value
of (k, j) for these modes is very large, which means thatthese modes contribute
little to the modal pressure series. Indeed, the graph of p 2 av would be unchanged

if only the “im” mode was used. This explains why the plot of p 2 av at 600 Hz has
a sinusoidal appearance.
To explain why this plot differs for that for a rigid wall, we need to compare
the n = 0 term in Eq. (10.2.38)
to the expression for Prigid . The contribution of the
“im” mode to P/ ρ0 cVp is proportional to k/κ0,im , which is 0.77, and V0,im /Vp ,
which is 0.9998. Thus, the different magnitude of the modal description of P and
Prigidmust stem from differences of the respective values. For the “im” mode, this
is κ0,0 = 0.9838 + 2.1789i, whereas rigid = 2.9133 − 5.1275i. A resonance
corresponds to being zero, so we conclude that the “im” mode is closer at 600 Hz
to a resonance than is the rigid model.
The fact that the pressure amplitudes are different does not necessarily invalidate
application of the impedance
tube concept, which relies on the maximum and min-
imum values of p 2 av . The procedure derived in Sect. 3.2.4 for extracting ζ L from
measured data should not be applied directly because the axial wavenumber is κ0,im ,
rather than k. A comparison of the “im” term in the modal series, Eq. (10.2.38), to
the above expression for Prigid suggests that the procedure may be used if we set
the wavenumber to κ0,im and replace the specific impedance with an effective value
given by
κ0,im
ζeff = ζ L
k
identify κ0,im by measuring the distance between

Note that in an actual setup we could
adjacent maxima or minima of p 2 av .

The ratio of the maximum and minimum values of p 2 av is defined as S 2 in
Sect. 3.2.4. The present data for 600 Hz gives S = 4.2572. From this, we find the
magnitude of the modal reflection coefficient, see Eq. (10.2.11),

(R L )0,im = S − 1 = 0.620
S+1
To find δ, which is the argument of (R L )0,im , we use Eq. (3.2.57) with k replaced
by κ0,im . The value of δ we shall use is the average of the value obtained from
each maximum or minimum. The result is δ = 0.230, so the reflection coefficient is
(R L )0,im = 0.620 exp(0.230i). In turn, the effective impedance is found to be
1 + (R L )0,im
ζeff = = 3.4727 + 1.5937i
1 − (R L )0,im
The corresponding true impedance is

k
ζL = ζeff = 2.6836 + 1.2316i
κ0,im
The latter value is quite close to the impedance used to generate the data.
The evaluation of ζ L at 600 Hz yielded a meaningful result because the concept
is based on the axial pressure distribution being sinusoidal. This is not the case at
6 kHz. We previously established that the j = 1 mode, as well as the “im” mode, is
not cut off at this frequency. Furthermore, V0,1 is 20% of V0,im . The respective axial
wavelengths are 2π/κ0,im = 0.236 m and 2π/κ0,1 = 0.301 m. The combination of
the contributions of these modes is the irregular dependence of p 2 av that appears
in Fig. 2. This negates identification of ζ L at 6 kHz by the scheme developed in
Sect. 3.2.4.
Our analysis leads to the conclusion that the given waveguide design is useful at
any frequency where the “im” mode is the only one whose eigenvalue is real. As the
frequency is increased from 600 Hz, the j = 1 mode “cuts on” when ka = η0, j . This
frequency is 3572 Hz. For comparison, if the wall were rigid, the axial wavenumber
of the j = 1 mode, which is the lowest nonplanar mode, is real above 3760 Hz.
Increasing the stiffness corresponds to making Im (ζa ) a larger negative number.

This causes η0,im to decrease toward zero and each η0, j for j ≥ 1 to approach the
respective value for a rigid wall. From this, we may deduce some design objectives
for an impedance tube. The first nonplanar mode is cut off below ωcut = η0,1 c/a.
Thus, a should be minimized. Then, the walls should be made as rigid as possible, so
that the cutoff frequencies of the nonplanar modes are close to the ideal values for a
rigid wall. Although a locally reacting model is a useful convenience, any wall is part
of a dynamic structure that has its own set of natural frequencies. The impedance
of the boundary as seen by the fluid within the cavity is greatly reduced at these
frequencies. Nevertheless, these characteristics do not alter the basic conclusions,
most important of which is that usage of an impedance tube is limited to frequencies
below the cutoff frequency of the second mode.
Vibrating Cylindrical Wall
Regardless of how a cylindrical cavity is excited, a series of forced modes must
feature a sum of products of a Bessel function of μR and a complex exponential
exp (±iκx), because such terms constitute a fundamental solution of the Helmholtz
equation for a cylindrical domain. The excitation in the previous section was an
imposed vibration at an end, so that waves propagated in the x direction. When the
excitation of a cylindrical cavity is an imposed vibration of the cylindrical boundary, a
cylindrical wave propagates inward. Its arrival at the centerline generates an outward
wave that cancels the singularity of the inward wave. The axial direction x now is
transverse to the propagation direction. Correspondingly, the x dependence is set by
the need to fit axial waves between the ends. In this view, the significance of R and
x is swapped relative to that for vibration at an end.
The wall vibration Va is a radial velocity, positive into the fluid, that may depend
arbitrarily on the axial position. An axisymmetric field occurs when Va is independent
of θ. The ends at x = 0 and x = L may be locally reacting, with specific impedances
ζ0 and ζ L . Thus, we seek a solution of the Helmholtz equation that is finite at R = 0
and satisfies impedance boundary conditions at x = 0 and x = L,
ζ0 ∂ P
P= at x = 0
ιk ∂x (10.2.42)
ζL ∂ P
P=− at x = L
ιk ∂x
In addition, the (inward) particle velocity at the cylindrical wall must match the input,
so it must be that
∂P
= + iωρ0 Vs (x) at R = a (10.2.43)
∂R
A comparison of the boundary conditions at the ends, Eq. (10.2.42), to those at the
sidewalls of a two-dimensional waveguide, Eq. (9.2.29), shows that they only differ
in the labels of parameters and coordinates. Thus, the axial mode functions here will
be the same as the transverse functions in Eq. (9.2.33), with y, H , and ηn changed to
x, L, and η0, j = κ0, j L, respectively. These functions, therefore, are
x x
0, j (x) = B0, j k L sin η0, j − iζ0 η0, j cos η0, j (10.2.44)
L L
Like the notation for excitation at an end, using a zero first subscript reminds us that
this is an axisymmetric function. The second subscript is the number of the root of
the characteristic equation. As always, j = 0 is reserved for a zero or an imaginary
eigenvalue. Adaptation of Eq. (9.2.32) to the current notation gives the characteristic
equation for the eigenvalues η0, j ,
2
− (k L)2 + ζ0 ζ L η0, j sin η0, j + i (ζ0 + ζ L ) (k L) η0, j cos η0, j = 0
(10.2.45)
The functions ˜ 0, j in Eq. (10.2.44) have a magnitude that varies substantially as
the eigenvalue changes. Normalization ensures that the mode functions we use do not
share that attribute. The coefficient B0, j is normalized consistently with Eq. (9.2.51).
An inner product of the mode function is defined relative to the x direction, so the
B0, j values are evaluated according to
2 L
B0, j =
(10.2.46)
L 2
0, j d x
0
The other part of the forced modes for this type of excitation describes how waves
propagate in the R direction. This dependence must be Bessel functions because the
Neumann functions are singular at R = 0. A modal series will feature a product of
a transverse mode and a Bessel function that depends on R. Specifically, the modal
series is taken to be
∞

P= P0, j J0 μ0, j R 0, j (x) (10.2.47)
j=0
The transverse wavenumber is denoted μ0, j , just as it was for the case of end exci-
tation. The difference is that previously this wavenumber was obtained by solving
a characteristic equation, whereas here it is set by the requirement that each term
in the series satisfies the Helmholtz equation. Hence, it must be that the transverse
wavenumbers are
η 2 1/2
0, j
μ0, j = k −
2
(10.2.48)
L
It remains to match the particle velocity to the imposed velocity of the cylinder
wall. Substitution of the modal series into Eq. (10.2.43) and application of the fact
that J0 (β) ≡ −J1 (β) give
∞

− μ0, j P0, j J1 μ0, j a 0, j (x) = iωρ0 Va (x) (10.2.49)
j=1
The 0, j functions are an orthogonal set, as described by Eq. (9.2.52). We exploit
this property by multiplying the preceding relation by an arbitrarily selected 0,n (x)
and then integrating the result over 0 < x < L. The result is that

L
ik 1
P0, j = − ρ0 c V0, j , V0, j = Va (x) 0, j (x) d x (10.2.50)
μ0, j J1 μ0, j a L 0
The pressure field corresponding to these coefficients is
∞
ikV0, j
P = ρ0 c J0 μ0, j R 0, j (x) (10.2.51)
μ J μ0, j a
j=0 0, j 1
Resonances correspond to situations where the denominator of any term in the

preceding summation is zero, that is, whenever the transverse wavenumber of the
jth mode is a zero of the first-order Bessel function. There are many such zeros; we
will use to denote the root number. Thus, resonances are marked by
μ0, j a = μ̃ where J1 (μ̃ ) = 0, = 1, 2, ... (10.2.52)
The first few roots are
μ̃0 = 0, μ̃1 = 3.8317, μ̃2 = 7.0156, μ̃3 = 10.1735 (10.2.53)
Higher roots are essentially integer multiples of π greater than the last listed value.
To find the natural frequencies we recall Eq. (10.2.48),
2 1/2
η μ̃
0, j
(knat )0, j, = + (10.2.54)
L a
An η0, j value is one of the infinite sets of roots of the characteristic equation for
the axial direction, Eq. (10.2.45), but k appears in that equation. Thus, unless both
ends are either rigid or pressure-release, the value of η0, j depends on k. However,
the preceding defines the value of k at a natural frequency as a function of η0, j .
Identification of the natural frequencies in the presence of this dilemma was addressed
by the discussion of Eq. (10.2.13).
EXAMPLE 10.6 A tentative design of a woofer, which is a loudspeaker

that radiates low-frequency sound, consists of a cavity whose cylindri-
cal wall is forced to undergo a uniform expansion at frequency ω. One
end of the cavity is rigid, and the other is closed by a loosely tensioned
membrane that well approximates a pressure-release condition. The sound
generated by this speaker radiates from the vibrating membrane, so the
velocity of the membrane is of interest. In particular, at low frequencies,
ka < 1, the radiated power is proportional to the volume velocity of a source.
Therefore, it is desired to derive an expression for the average velocity across
the membrane. Evaluate this expression for the range from 20 Hz to 1 kHz for
cases where the fluid within the cavity is air and water. The diameter of the
cavity is 300 mm and the length is 500 mm.
Significance
The analysis will provide an in-depth exploration of the manner in which a vibrating
cylindrical wall influences the interior field. It also will introduce some rudimentary
issues for transducer design.
Solution
Formation of the modal series begins with determination of the axial mode functions.
It is convenient to let x = 0 be the pressure-release end, so that ζ0 = 0 and ζ L is
infinite. We divide the characteristic equation, Eq. (10.2.45), by ζ L in order to handle
its infinite value, which reduces the equation to
(κL) cos (κL) = 0
The root κL = 0 leads to 0, j = 0, so it is discarded. Thus, κL must be an odd

multiple of π/2,
2j − 1
η0, j = κ0, j L = π, j = 1, 2, ... (1)
2
The axial mode functions obtained from Eq. (10.2.44) are

(2 j − 1) πx
0, j = B0, j sin
2L
To normalize the modes we set

2
L
(2 j − 1) πx 2 L
B0, j sin d x ≡ B0, j =L
0 2L 2
so the normalized modes are

√ (2 j − 1) πx
0, j = 2 sin (2)
2L
The transverse wavenumbers are given, in Eq. (10.2.48),

2 1/2
1 2 j − 1
μ0, j = (k L)2 − π2 (3a)
L 2
The negative imaginary root is taken if k L < ( j − 1/2) π.

It is stated that the radial vibration of the cylindrical wall is uniform, which means
that Va is a constant value. Substitution of Eq. (2) into Eq. (10.2.50) shows that the
corresponding velocity coefficients are

L
√
Va 2 2
Vj = 0, j d x = Vs (4)
L 0 (2 j − 1) π
Substitution of these quantities into the modal series, Eq. (10.2.51), gives
∞
4 ik J0 μ0, j R (2 j − 1) πx
P = − ρ0 cVa sin (5)
j=1
(2 j − 1) π μ0, j J1 μ0, j a 2L
If (2 j − 1) π > 2k L, the mode number is above the cutoff value. Then, μ0, j is neg-
ative imaginary, that is, μ0, j = −i μ̂0, j . We may either directly input this imaginary
value to the computer routine for Bessel functions, or alternatively invoke the identity
that Jn (−iβ) = (−i)n In (β), see Sect. 9.3.3.
The quantity of interest here is the axial particle velocity at x = 0. From Euler’s
equation, we have
∞
1 ∂P 2J0 μ0, j R (2 j − 1) πx
Vx = − = Va cos (6)
iρ0 ω ∂x j=1
μ0, j L J1 μ0, j a 2L
The volume velocity is the area integral of the particle velocity at x = 0. A reference
volume velocity is the cylindrical wall’s velocity Va multiplied by the cross-sectional
area. Thus,

a
Q̃ 1 Vx |x=0
= (2π R) d R
πa Va
2 πa 0
2 V2
An equivalent interpretation of this expression is that it is the ratio of the average axial
velocity at x = 0 to the
wall’s velocity. Substitution of Eq. (6) reduces this equation
to an integral of R J0 μ0, j R , for which a standard formula4 is available. The result
is that

∞
Vav Q̃ 1 a
4 a/L
≡ = Vx |x=0 (2π R) d R = Va (7)
Va πa Va
2 πa 2 0 π j=1 μ0, j a 2
Interpretation of the frequency response graphs will be easier if we first compute

the natural frequencies. The procedure for this analysis discussed in connection
4 M.I. Abramowitz and I.A. Stegun, ibid.

with Eq. (10.2.54) is not needed because the ends are rigid and pressure-release.
Consequently, the axial wavenumbers are η0, j /L are given by Eq. (1), regardless
of the frequency. The resonance condition is J1 μ0, j a = 0, whose lower roots
are given in Eq. (10.2.53). Substitution of these values into Eq. (10.2.54) yields the
natural frequencies. The following tabulation lists the values that occur below 1 kHz.
Air Water
Axial # j μ̃ (ωnat ) j, (Hz) Axial # j μ̃0, j (ωnat ) j, (Hz)
1 0 0 170 1 0 0 740
2 0 0 510
1 1 3.8317 711.7
3 0 0 850
2 1 3.8317 858.9
Figure 1 shows the result of evaluating Eq. (7) for the average velocity. When the
cylinder is filled with water, the single resonance at 740 Hz matches the tabulated
natural frequency. In contrast, for the case where air is the fluid, only the natural
frequencies at 170, 510, and 850 Hz are manifested as resonances. (The data was
probed to ascertain that the latter is not the adjacent natural frequency of 858.9 Hz.)
102
Water
Air
Vav /Vs
100
10-2
200 400 600 800 1000
Frequency (Hz)
Figure 1.
Do the missing resonances represent an anomaly? Let us look at the pressure

on-axis at the midpoint, which is described by Fig. 2. Every natural frequency is
manifested by a peak pressure, as expected.
P(x=L/2)/(0cVs )
10 2 Water
Air
10 0
200 400 600 800 1000

Frequency (Hz)
Figure 2.
To explain why the average velocity does not show a resonance at some natural
frequencies, we must examine Eqs. (5) and (7) more carefully. The former
indicates
that the jth pressure term will be singular if at some frequency, J1 μ0. j a = 0, which
encompasses the alternative that the factor μ0. j = 0. This is the attribute of any nat-
ural frequency. The situation for Vav in Eq. (7) is different, because a singularity only
occurs if the frequency is such that one of the μ0. j terms in the denominator is zero.
The preceding tabulation indicates μ0. j = 0 is associated with the natural frequencies
at 170, 510, and 850 Hz. This merely confirms that the graphs are consistent with
the analysis, but not why the pressure and average velocity behave differently. The
answer lies in Eq. (6) for theaxial particle
velocity at any location. The denominator
of that quantity contains J1 μ0, j a , so it is singular at any natural frequency. How-
ever, the
numerator
indicates that the velocity distribution at x = 0 is proportional
to J0 μ0,j R . The average value of this function over the area is proportional to
J1 μ0. j a , which cancels the singular denominator term. Recall that μ0, j = 0 is the
property of a plane wave mode, in which there is no dependence on R. Hence, we
have established that if any mode is not constant over a cross section, its contribu-
tion to the average velocity will be finite at any frequency, even resonant ones. The
average velocity will show a resonance only if the resonant mode is one in which
there is no variation in the R direction.
To close this example, let us consider whether this device is useful as a low-
frequency loudspeaker. Flatness of the frequency response is a desirable attribute.
Filling the cavity with air leads to many resonances in the frequency band, but the
average velocity in the water-filled design is essentially flat out to 500 Hz. This might
be a desirable attribute, but another feature of the response is not. In the low-frequency
range, where the response is flat, the value of Vav is nearly equal to Va . Thus, no
advantage is gained by driving the cylindrical surface relative to using a piston that
matches the cylinder’s cross section. Resonances and amplification factors are two
aspects that must be considered in the design of any transducer.
10.3 Analysis Using Natural Cavity Modes
The forced cavity modes depend on the frequency of excitation and what surface
is forced to vibrate. In contrast, the natural cavity modes that are the present topic
depend on the size and shape of the enclosed region, as well as the nature of the
boundaries, but they do not depend on the way in which the field is generated. These
modes are properties of the system.
They must be determined prior to analysis of the response to an excitation. If
the region has a shape that fits one of the standard coordinate systems, there will
be wavenumbers describing the spatial dependence in each coordinate direction. In
that case, the mode will be characterized by a subscript whose value marks the set
of associated wavenumbers. For example, the modes of a rectangular room would
be denoted as j,m,n (x, y, z). However, many situations feature an enclosed region
whose shape is irregular They do not lend themselves to decomposition into identifi-
able wavenumbers. Such systems are better described by a single subscript denoting
the hierarchy of modes. We will begin with an overview, which is more easily pre-
sented if we employ the latter representation. Hence, the cavity modes in the general
development will be denoted as n (x̄).
Before we embark, it is necessary to limit the scope of our investigations to situ-
ations where the walls fit some combination of the rigid or pressure-release models.
10.3 Analysis Using Natural Cavity Modes 337
This restriction is imposed because the formulation required to treat any other wall
condition is quite complicated, as is evident from a fairly recent analysis.5 Even under
this constraint, a variety of configurations are possible. The development that follows
begins with the general aspects of cavity modes that do not depend on the shape and
properties of the enclosure. These are the governing equations, orthogonality, and
the usage of a modal series to determine the pressure. The general development will
be followed by specific analyses of standard shapes. Another limitation is that only
the response generated by sources situated within the fluid will be considered. Recall
that this is the situation for which a formulation using natural cavity modes is best
suited, whereas the field generated by an imposed wall vibration is best described
in terms of forced cavity modes. The common practice is to omit “natural” from the
terminology. We shall do so because each formulation is used for a different class
of excitation. Consequently, there is no context where both natural and forced cavity
modes will be encountered.
10.3.1 Equations Governing Cavity Modes
By definition, a cavity mode n is a pressure field that can exist within the fluid
domain V, even though there is no external excitation. Such a function must be a
solution of the Helmholtz equation. In accord with the restrictions stated above, we
subdivide the boundary into a rigid portion SV on which the normal velocity is zero
and a pressure-release portion S P on which the pressure is zero. The pressure field in a
cavity mode is p (x̄, t) = Re (n exp (iωt)), and Euler’s equation describes the parti-
cle velocity. Hence, the equations governing a cavity mode for this set of surfaces are
∇ 2 n + (kn )2 n = 0, x̄ ∈ V
n̄ (x̄s ) · ∇n = 0 if x̄s ∈ SV (10.3.1)
n = 0 if x̄s ∈ S P
If we assign an arbitrary value to kn , the only possible solution of this homogeneous

set of equations is n = 0 everywhere. Thus, we must identify the values of kn for
which there is a nontrivial solution. This is a statement of an eigenvalue problem.
The eigenvalues kn give the natural frequencies as ωn = kn c, and the eigenfunction
corresponding to each kn is a cavity mode.
It should be evident that it will be quite difficult, or even impossible, to solve
for the eigenvalues kn unless V has a shape that fits one of the standard coordinate
systems. Thus, we shall pursue analytical solutions only for rectangular, spherical,
and cylindrical cavities. Any combination of rigid and pressure-release subregions
of S is admissible from the viewpoint of general properties. However, we will not
be able to solve analytically for the mode functions if the condition on a constant
coordinate surface changes. For example, we can use analytical methods if entire
5 J.H.
Ginsberg, “Derivation of a Ritz series modeling technique for acoustic cavity-structural sys-
tems based on a constrained Hamilton’s principle,” J. Acoust. Soc. Am. 127, 2749–2758 (2010).
walls of a box are pressure-release or rigid, but we cannot use analytical methods if
any wall of that box is partially rigid and partially pressure-release. Such situations
require computational or approximate methods.
The existence and nature of a zero frequency mode has general importance.
If the frequency is zero, the Helmholtz equation reduces to Laplace’s equation,
∇ 2 P = 0. Because ∇ 2 P ≡ ∇ · ∇ P, it must be that ∇ P is a constant vector. If the
entire boundary is rigid, then n̄ · ∇ P must be zero at the boundary, which is satis-
fied only if the constant value of ∇ P is zero. In other words, a constant value of
P constitutes a nontrivial solution at zero frequency if a fluid is fully enclosed by
a rigid boundary. The contrary case is that in which any portion of the boundary is
pressure-release. A field whose gradient is constant with a zero value at any location
must be zero everywhere. In other words,
If a region of fluid is fully enclosed by a rigid boundary, then there exists a zero
frequency cavity mode in which the pressure is constant. If any portion of the boundary
is pressure-release, a zero frequency mode does not exist.
10.3.2 Orthogonality
Orthogonality of the cavity modes greatly simplifies determination of the response to

source excitation. To derive this property, we consider a pair of modes j and n, whose
natural frequencies are ck j and ckn . The first step entails multiplying the Helmholtz
equation for mode n by mode j and then integrating over V, which gives

j ∇ n dV +
2
kn2 j n dV = 0 (10.3.2)
V V
In order to obtain a form in which j and n appear in the same manner, we exploit
the definition that ∇ 2 ≡ ∇ · ∇ to convert the first integrand,

∇ · j ∇n − ∇ j · ∇n dV + kn2 j n dV = 0 (10.3.3)
V V
Application of the divergence theorem to the first term leads to

− j n̄ (x̄s ) · ∇n dS − ∇ j · ∇n dV + kn2 j n dV = 0

S V V
(10.3.4)
where the negative sign preceding the surface integral results from the definition of
n̄ as the normal oriented into the cavity.
We have limited our consideration to systems in which the wall consists of a rigid
portion SV and a pressure-release portion S P . On SV , we have n̄ (x̄s ) · ∇n = 0 and
n (x̄s ) = 0 on S P . In either case, one of the factors in the surface integral is zero,
so it must be that

kn2 j n dV − ∇ j · ∇n dV = 0 (10.3.5)

V V
If we had begun by multiplying the Helmholtz equation for mode j by mode n, we

would have obtained

2
kj j n dV = ∇n · ∇ j dV (10.3.6)
V V
The difference of these two relations is

2
kn − k 2j j n dV = 0 (10.3.7)
V
This relation is reminiscent of the one for the transverse mode function of a
forced cavity mode. If modes j and n are different, then the factor multiplying
the integral is not zero, so the integral must vanish. In other words, two different
modes are orthogonal over the space V. Equation (10.3.7) also applies if n and
j are the same function. In that case, the factor multiplying the integral is zero,
so the condition is met regardless of the value of the integral. Indeed, we know that
the integration will yield a positive value because the mode functions are real, so
(n )2 > 0 everywhere. This positive value is used to normalize the cavity modes.
Specifically, a cavity mode n may be multiplied by any constant Bn without altering
the fact that they are solutions of the basic governing equations. The arbitrariness
of Bn is addressed by setting it such that the integral equals the volume V of the
domain. Doing so yields mode functions that are dimensionless. Thus, we have the
first orthogonality condition,

j n dV = Vδ j,m (10.3.8)
V
(For two-dimensional domains, V would be the area in the plane on which the pressure
varies.) When we substitute the first orthogonality condition into Eq. (10.3.5), we
obtain the second orthogonality condition,

∇ j · ∇n dV = V (kn )2 δ j,m (10.3.9)

V
An exception to this development arises if two modes have the same natural fre-
quency. This situation is common for rectangular-shaped enclosures, as we will see
in the upcoming Example 10.7. Furthermore, the repeated natural frequencies always
arise for the nonaxisymmetric modes of a cylindrical cavity because of the arbitrari-
ness of how θ is defined. However, in both geometries, the derived cavity modes will
be orthogonal, even though k j = kn . This property stems from the differing spatial
patterns, with one being obtainable from the other by rotating the coordinate system
by 90◦ from the other.
10.3.3 Analysis of the Pressure Field
We may now proceed to formulate a modal analysis of the pressure induced by

sources. One of the assets of the cavity mode approach is that it can be used to derive
directly a time-domain response. We shall develop this type of solution and then use
the result to obtain a description of the frequency-domain response.
We begin with a series of cavity modes in which the modal coefficients are allowed
to be arbitrary functions of time,
∞

p (x̄, t) = n (x̄) pn (t) (10.3.10)
n=1
An arbitrary collection of sources are situated within the cavity. The volume velocity
of each source is Q m (t), and the locations are x̄m . The field in the time domain for
a single source is governed by the inhomogeneous wave equation, Eq. (6.4.23), with
C = −ρ0 Q̇ m . Thus, the field equation for the pressure is
1 ∂2 p
∇2 p − = −ρ0 Q̇ m (t) δ (x̄ − x̄m ) (10.3.11)
c ∂t
2 2
m
Differentiation of the modal series with respect to t operates only on the modal
coefficients, and spatial derivatives operate only on the cavity modes. Thus, substi-
tution of the series into the preceding wave equation leads to
∞

2 2
c ∇ n pn − n p̈n = −ρ0 c2 Q̇ m (t) δ (x̄ − x̄m ) (10.3.12)
n=1 m
The procedure mirrors that which led to Eq. (9.2.52). The preceding
is multiplied by a
specific j and integrated over V. Because j ∇ 2 n ≡ ∇ · j ∇n − ∇ j · ∇n ,
the divergence theorem gives

j ∇ nn dV =
2
j n̄ · ∇n dS− ∇ j · ∇n dV (10.3.13)
V S V
The mode functions satisfy rigid or pressure-release boundary conditions, as pre-

scribed in Eq. (10.3.1), so the surface integral vanishes. Hence, we have found that
∞

∞

c 2 pn ∇ j · ∇n dV + p̈n j n dV
n=1 n=1
V

V (10.3.14)

= j ρ0 c 2
Q̇ m (t) δ (x̄ − x̄m ) dV
m
V
The last step is to invoke both orthogonality properties, Eqs. (10.3.8) and (10.3.9),
which serve to filter out of both summations all terms for which n = j. The remaining
terms constitute an ordinary differential equation for the jth modal coefficient,
2
p̈ j + ω j p j = S j (t), j = 1, 2, .. (10.3.15)
The S j excitations are time-dependent modal source strengths. They map the spatial
distribution of sources into the modal space, according to

ρ0 c2 ρ0 c2
Sj = j Q̇ m (t) δ (x̄ − x̄m ) dV = Q̇ m (t) j (x̄m )
V m
V m
V
(10.3.16)
Equation (10.3.15) has the same form as the differential equation of motion for a
simple spring-mass system. Its solution is quite accessible. If there is a single source
whose time signature is a unit impulse function, that is, Q̇ 1 = δ (t − t0 ), then the
modal coefficients are the impulse response,
ρ0 c2 j (x̄1 )
pj = sin ω j (t − t0 ) h (t − t0 ) (10.3.17)
V ωj
By definition, the time-domain Green’s function is the pressure due to an impulsive

point source. Thus, we have established that for a cavity it is
∞
ρ0 c2 j (x̄1 ) j (x̄)
G (x̄, t) = sin ω j (t − t0 ) h (t − t0 ) (10.3.18)
V j=1 ωj
Depending on the nature of the sources’ time signatures, the modal differential
equations might be solvable by standard methods. If not, they can be solved numer-
ically. One case where an analytical solution is possible is that of steady-state field
generated by time harmonic sources. Such an analysis yieldsthe frequency-domain
solution. The sources are set to Q̇ m (x, t) = δ (x̄ − x̄m ) Re iω Q̂ m exp(iωt) . The

frequency-domain representation of the pressure coefficients is p j = Re X j eiωt .
The modal source strengths in Eq. (10.3.16) become

iωρ0 c2 Q̂ m iωt
Sn = n (x̄m ) Re e (10.3.19)
m
V
Synthesis of the modal series, Eq. (10.3.10), using the particular solution for the P j
coefficients yields
∞

n (x̄m ) n (x̄) iω Q̂ m iωt
p (x̄, t) = ρ0 c 2
Re e (10.3.20)
n=1 m
ωn2 − ω 2 V
In the special case of a single source whose strength is a unit value, Q̂ 1 = 1, the
preceding gives the frequency-domain Green’s function of the cavity. The similarity
of this result and Eq. (10.1.16), which was derived heuristically, is obvious.
The general development is applicable to the cavity modes of any configuration,
regardless of the shape of the enclosed region and the nature of the walls. However,
Eq. (10.3.20) presumes that the cavity modes have been determined. The reality is that
only rectangular and circular shapes may be analyzed without considerable effort.
10.3.4 Rectangular Cavity
To study the field within a closed box, we place the origin of x yz at one corner,
with the axes aligned parallel to the edges. Then, the sides of the cavity coincide
with the planes x = 0, x = L, y = 0, y = H , z = 0, and z = W . The pressure in the
region between two parallel walls was the subject of Sect. 9.2. It was shown there
that any combination of rigid and pressure-release surfaces is represented by either
a sine or cosine function of the coordinate that is normal to the surfaces. That is the
present situation, but there are now three sets of parallel walls. Consequently, there
are three sets of harmonic functions: x (x) for the walls at x = 0 and x = L, y (y)
for the walls at y = 0 and y = H , and z (z) for the walls at z = 0 and z = W . The
wavenumbers are denoted as κ, μ, and ν in the x, y, and z directions, respectively.
A product of these harmonic functions will satisfy the Helmholtz equation if the
wavenumbers are set properly. Furthermore, because the differential equation and
boundary conditions are homogeneous, any solution may be multiplied by a constant
B. It follows from these considerations that a mode function is described by
(x, y, z) = Bx (x) y (y) z (z) (10.3.21a)
where the factors are

x = cx cos (κx) + sx sin (κx)
y = c y cos (μy) + s y sin (μy)
z = cz cos (νz) + sz sin (νz) (10.3.21b)
Either a c or s coefficient will be zero, depending on the nature of the side that
contains the origin. The nonzero coefficient may be set to one because B serves to
collect all arbitrary factors. For example, if the surface y = 0 is rigid, then it must be
that ∂ P/∂ y = 0 at y = 0, which is met by setting s y = 0 and c y = 1. Alternatively,
if y = 0 is a pressure-release side, so that P = 0 at y = 0, then c y = 0 and s y = 1.
The wavenumber in each direction is found by satisfying the boundary conditions
at the far walls, x = L, y = H , and z = W . In particular, the condition that there be a
nontrivial solution for each direction will require that the argument of the respective
sinusoidal function be an integer multiple of π/2. For example, suppose P = 0
on z = 0, so that z = sin (νz). Then, the rigid condition, ∂ P/∂z = 0 on z = W ,
requires that cos (νW ) = 0, which is satisfied if νW is an odd multiple of π/2. The
other possibility, P = 0 at z = W , requires sin (νW ) = 0. This is met if νW is an
even multiple of π, but ν cannot be zero. This is so because it would lead to the trivial
solution = 0.
Determination of the possible values of the wavenumbers allows us to evaluate
the natural frequencies. By definition, Eq. (10.3.21) is a solution of the Helmholtz
equation at a natural frequency. Let j, m, n be the sequence number for k, μ, and
v, respectively, so that the corresponding natural frequency is denoted as ω j,m,m =
ck j,m,n . Then, it must be that
1/2
ω j,m,m = c κ2j + μ2m + νn2 (10.3.22)
This leaves only the scaling coefficient B j,m,n to determine. These values are
obtained by normalizing the mode functions to satisfy Eq. (10.3.8) when the two
modes are the same. The mode function described in Eq. (10.3.21) is products of the
functions of each coordinate, and the boundaries consist of constant values of each
coordinate. Hence, Eq. (10.3.8) leads to
⎛ ⎞1/2
⎜ LHW ⎟
B j,m,n = ⎜
⎝
W
H
L
⎟
⎠ (10.3.23)
x (x) d x
2
y (y) dy
2
z (z) dz
2
0 0 0
A closed form expression for the coefficient may be found, but it is easier to do so
after the functions are specified.
EXAMPLE 10.7 A column of water is used to test vertical propagation in a

channel. The height is 4 m, and the dimensions of the horizontal cross section
are 0.6 m × 1.2 m. The bottom and sides of the tank are rigid. Determine the
natural frequencies of this tank up to 4 kHz.
Significance
The evaluation of the frequencies is straightforward. What makes this example inter-
esting is what the results tell us about the general properties of natural frequencies.
Solution
We shall align the y-axis vertically and set the x-axis horizontally along the short
side. This corresponds to L = 0.6 m, H = 4 m, and W = 1.2 m. The surface y = H
is the interface between the water and air, so it is well approximated as pressure-
release for waves in the water. The vertical sides and bottom are rigid, so it is required
that the normal derivatives on them be zero. Satisfaction of this condition at x = 0,
y = 0, and z = 0 requires that
x = cos (κx) , y = cos (μy) , z = cos (νz) (1)
The other vertical sides are x = L and z = W . Setting the normal derivative on
each to zero yields

∂x ∂z
= sin (κL) = 0 and = sin (νW ) = 0
∂x x=L ∂z z=W
These conditions are met if

jπ nπ
κj = , j = 0, 1, ...; νn = , n = 0, 1, ... (2a)
L W
Note that κ0 = 0 and ν0 = 0. Both correspond to the plane wave modes of two-
dimensional waveguide whose walls are rigid. The horizontal surface at y = H is
pressure-release, so setting y = 0 there leads to
(2m − 1) π
μm = , m = 1, 2, ... (2b)
2H
The natural frequency corresponding to these wavenumbers is
1/2
j 2 2m − 1 2 n 2
ω j,m,n = πc + + (3)
L 2H W
The mode functions, which are not needed in the present context, are
nπz
jπx (2m − 1) π y
j,m,n = B j,m,n cos cos cos (4)
L 2H W
It is specified that we should limit our attention to the range up to 4 kHz. To

determine the range of indices to consider, we observe that the natural frequency
increases monotonically as either j, m, or n is increased with the other two held fixed.
For a fixed frequency, the largest value of one index corresponds to the smallest value
of the other two. The range of indices for the computation is 0 ≤ j ≤ J , 1 ≤ m ≤ M,
0 ≤ n ≤ N . Thus, we set the range such that J , M, and N are the smallest values
for which ω J,1,0 > ωmax , ω0,M,0 > ωmax , and ω0,1,N > ωmax . Setting ωmax = 8000π
rad/s leads to J = 4, M = 23, and N = 7.
A four column list of j, m, n, and ω j,m,n for j = 0, .., 3, m = 1, .., 22, and
n = 0, .., 6 could be sorted by saving them to a spreadsheet. A graphical display
is a better way to see the qualitative behavior. Each curve in Fig. 1 describes ω j,m,n
for a fixed j and n as m is increased from one. The lowest frequency, which cor-
responds to j = 0, m = 1, and n = 0, is 92.5 Hz. This is the cavity’s fundamental
natural frequency.
(2,m,5) (3,m,1)
4000 (1,m,6) & (3,m,2)
(0,m,6) & (3,m,0)
(2,m,4)
(1,m,5)
(0,m,5) & (2,m,3)
3000
(1,m,4) & (2,m,2)
(2,m,1)
j,k,nHz
(0,m,4) & (2,m,0)

(1,m,3)
2000 (0,m,3)
(1,m,2)
(1,m,1)
(0,m,2) & (1,m,0)
1000
(0,m,1)
(0,m,0)
0
1 5 10 15 20
Mode index m
Figure 1.
A few combinations of j and n correspond to overlapping curves, which means

the natural frequencies are equal regardless of m. The lowest curve for which this
condition occurs is (0, m, 2) and (1, m, 0). Other combinations are (0, m, 4) and
(2, m, 0), (1, m, 4, ), (2, m, 2, ), and so on. This overlap is a consequence of the fact
that W = 2L. Thus, if a natural frequency corresponds to indices j1 , m, and n 1 , the
same value will occur for j2 = n 1 /2 and n 2 = 2 j1 , with the same m. This is not to
suggest that natural frequencies overlap only if j2 n 2 = j1 n 1 . Other combinations are
possible. This is the case for the (0, m, 5) and (2, m, 3) modes. Furthermore, even if a
frequency curve does do not overlap others, it is possible that a value on that curve will
equal a value on a different curve at a different m. An interesting manifestation of that
possibility is a triple value, ω1,8,3 / (2π) = ω1,13,0 / (2π) = ω0,13,2 / (2π) = 2621 Hz.
In general, different modes will have the same natural frequency in the low-
frequency range if a ratio of cavity dimensions is a small rational fraction. For
example, in the present system the ratio of lengths in the x and z directions is
W/L = 2, so the occurrence of equal frequencies with different ( j, n) pairs is rather
common. In contrast, H/L = 20/3, so equal natural frequencies with different ( j, m)
are less common. The same is true for natural frequencies associated with different
(m, n) pairs because H/W = 10/3.
It was stated in the general discussion that two cavity modes will be orthogonal
even if they have the same natural frequency. Let us test that assertion. The orthogo-
nality condition is the volume integral of the product of two functions. In view of the
fact that the modes in Eq. (4) are products of x, y, and z functions, the orthogonality
integral factorizes to integrals over the range of x, y and z. Each has the same general
form,

1 sin ((βr − βs ) ) sin ((βr + βs ) )
cos (βr η) cos (βs η) dη ≡ +
0 2 βr − βs βr + βs
For the x functions, βr represents κ j L = jπ, while βr represents νn W = nπ for

the z functions. In both cases, the argument of the above sine functions is a multiple
of π, and therefore zero. For the y functions, βr represents μm H = (m − 1/2) π,
so that (βr ± βs ) also is a multiple of π. It follows that the orthogonality integral
vanishes for any combination of two different mode functions, regardless of what
their natural frequencies are.
The fact that two or more modes share a natural frequency does have some inter-
esting implications. Suppose a field is established at the natural frequency belonging
to two modes. A linear combination of these modes also satisfies the governing
equations. For example, in the case of a pair identified in the preceding paragraph,
= b1 0,1,2 + b2 1,1,0 satisfies the Helmholtz at k = ω0,1,2 /c = ω1,1,0 /c for any
constants b1 and b2 . We could consider the combined function to be a mode, but
shall not do so because it is not independent of the ones we derived. The values of
b1 and b2 that occur depend on how the field is induced. This phenomenon was first
encountered in the vibration of rectangular membranes and plates. Chaladni6 placed
sand on a vibrating square plate. This made it possible to visualize the vibration
pattern because the sand migrated toward the nodal lines, at which the plate is quiet.
The excitation consisted of a violin bow that was stroked across an edge. Chaladni
showed that nodal patterns were not unique. He also showed that they were very
difficult to reproduce. This is a consequence of the fact that the factors b1 and b2 for
the vibration were highly sensitive to small changes in the location where the plate
was stroked. In recognition of his contributions, nodal lines are commonly known
as Chaladni lines.
Even if natural frequencies are not exactly equal, the likelihood that two or more
natural frequencies are very close increases drastically as the frequency increases.
This attribute may be recognized by drawing a horizontal line through any natural
frequency in the graph. Above ω0,1,1 , this line will intersect two or more curves. As
the frequency increases, the number of curves such as line intersects will increase.
Some of those intersections will be close to integer values of m, which mark the
natural frequencies. We say that the modal density increases with increasing natural
6 E.
Chaladni, Entdeckungen über die Theorie des Klanges (“Discoveries in the Theory of Sound”)
1787.
frequency. This phenomenon is a general attribute of two- and three-dimensional

systems that conserve energy. It can be quite problematic for experimental methods
that attempt to identify modal properties from measured data.
EXAMPLE 10.8 Consider a hard-walled chamber whose dimensions are

4 × 3 × 6 m. A line source is situated along one of the 6 m edges. The source
is impulsive, with a volume velocity per unit length that changes discontinu-
ously at t = 0 from zero to a constant value q0 . This corresponds to a mass
acceleration per unit length that is ρ0 q0 δ (t). Determine the signal heard in the
middle of the chamber. The fluid is air.
Significance
The solution is a direct application of the modal series formalism. The results will
lead to some useful insights regarding our prior study of point sources, and it will
highlight some of the difficulties one might encounter in the application of modal
series representations.
Solution
The x yz coordinate system in Fig. 1 has the z-axis coincident with the line source
and the x-axis coincident with a 4 m edge. The line source has the same properties
along its length, and the hard-walls at z = 0 and z = 6 m allow for the existence
of a pressure field that does not vary in that direction. Thus, the pressure field is
p = Re (P (x, y) exp (iωt)).
y
3m
Line
sourc
e
6m 4m
z
Figure 1.
Despite independence of the z coordinate, it is best to treat the field as three-

dimensional, rather than considering it to be two-dimensional. The boundaries in the
x and y directions also are rigid, so the modes that contribute to the pressure field
are nπ y
jπx
j,n,0 = B j,n,0 cos cos , j, n = 0, 1, ... (1)
L H
where L = 4 m and H = 3 m. The corresponding natural frequencies are

2 1/2
jπ nπ 2
ω j,n,0 = c + (2)
L H
Normalizing the mode functions to the volume V = L H W , in accord with

Eq. (10.3.8), leads to
2
B j,n,0 = 1/2 1/2 (3)
1 + δ0, j 1 + δ0,n
Because the source is situated in the corner, its mass flows into one quadrant, rather
than into free space. To account for this effect, we multiply the source strength by a
factor of four. Another consideration is that Eq. (10.3.16) describes the modal source
strength S j for a discrete set of sources. The line source is a continuous distribution of
differential mass accelerations 4ρ0 q0 δ (t) dz, so a summation of sources is replaced
by an integral over z. Hence, the first form of Eq. (10.3.16) in this case gives

ρ0 c2 L H W W
S j,n,0 = 4ρ0 q0 δ (t) dξ δ (x) δ (y) j,n,0 dzdyd x
LHW 0 0 0 0
Equation (1) indicates that all modes at the origin are j,n,0 = B j,n,0 . Thus, the modal
source strengths are
4ρ0 c2 q0
S j,n,0 = B j,n,0 δ (t) (4)
LH
The modal differential equations, Eq. (10.3.15), corresponding to these coeffi-
cients are
2 4ρ0 c2 q0
p̈ j,n,0 + ω j,n,0 p j,n,0 = B j,n,0 δ (t) (5)
LH
Quiescent initial conditions are p j,n,0 = 0 and ṗ j,n,0 = 0 for any t preceding activa-
tion of the impulse. After the impulse, the excitation is gone, so only the homoge-
neous solution remains. The initial conditions for the ensuing response are obtained
by considering the infinitesimal interval from t = 0− before the impulse to t = 0+
afterward. The pressure cannot change suddenly in that interval, because doing
so
+
would lead to an infinite
particle
acceleration. Thus, it must be that p j,n t = 0 = 0.
To determine ṗ j,n t = 0+ , we integrate Eq. (5) over 0− < t < 0+ . The inte-
gral of pn,n over this
infinitesimal
interval is zero, and the integral of p̈ j,n is
ṗ j,n t = 0+ − ṗ j,n t = 0− . Thus, we find that
4ρ0 c2 q0
ṗ j,n,0 t = 0+ = B j,n,0
LH
The solution of Eq. (5) fitting the initial conditions is the one-degree-of-freedom
impulse response,
4ρ0 c2 q0
P j,n,0 = B j,n,0 g j,n,0 (t)
LH
sin ω j,n,0 t (6)
g j,n,0 = h (t) if j > 0 or n > 0
ω j,n,0
g0,0,0 = t h (t)
where the step function serves to remind us that the response is zero for t < 0. The
special form of the impulse response for the (0, 0, 0) mode is a consequence of the
associated natural frequency being zero. The corresponding representation of the
pressure at an arbitrary field point is found from Eq. (10.3.10) to be
nπ y
4ρ0 c2 q0 2
J N
jπx
p (x̄, t) = B j,n,0 cos cos g j,n,0 (t) (7)
L H j=0 n=0 L H
The dimensionality of q0 is (volume/time)/length = length2 /time. Therefore, dividing

p by ρ0 cq0 /L is a convenient way of eliminating dependence on the unspecified value
of q0 .
An important aspect is truncation of the series. The impulse response function
g j,n,0 (t) is the only factor in Eq. (7) whose magnitude varies significantly from term
to term. Its magnitude is proportional to 1/ω j,n , so it decreases monotonically as j
or n increases. Thus, truncation will be based in assuring that any frequencies that
are omitted exceed some maximum value. Let us describe this frequency limit as
σ min (ωnat ), where σ is a factor to be determined. Because L > H , the fundamental
nonzero frequency is ω1,0,0 = πc/L. (The zero frequency is not relevant.) Thus, the
task is to identify values of J and N such that ω J,n,0 , for any n, and ω j,N ,0 , for any j,
exceed σ (πc/L). The largest J corresponds to n = 0, so we want ω J,0,0 ≥ σ (πc/L),
that is J ≥ ceil(σ). A similar analysis sets N ≥ ceil(σ H/L).
A matrix algorithm may be implemented to evaluate Eq. (7) at specified loca-
tions over any time interval. The values of J and N will ultimately be found
to be quite large, which means that there are many modes to compute. There
are several ways in which the algorithm in Example 10.2 may be adapted to the
present formulation. The following was found to be quite efficient relative to oth-
ers that were tried, yet able to run on a laptop whose memory workspace is lim-
ited. Variants that work on more powerful computers might be more rapid, but
their workspace requirements could be excessive. The first step creates a row vec-
tor of mode numbers [m] = [1 2 3 ...M], where M = (J + 1) (N + 1). The pair
of wavenumbers in the x and y direction are placed in row vectors [ j] and [n],
whose length is the same length as [m]. Operations to create these quantities are
[ j] = floor(([m] − 1)/(N + 1)) followed by [n] = [m] − 1 − [ j] ∗ (N + 1). These
row vectors are used to vectorize computation of the natural frequencies in Eq. (2)
and normalizing coefficients in Eq. (3). The MATLAB code for these operations
is omega=c*sqrt((j*pi/L).ˆ2+(n*pi/H).ˆ2); B=2./sqrt((1+
(j==0)).*(1+(n==0)));. The result is row vectors [ω] and [B] whose

size matches [m]. To evaluate the modal values, let x̄ f , f = 1, ... F, denote the
field points at which we wish to evaluate the pressure, and let t be a sequence
of instants at which the pressure is to computed. The values of all mode func-
tions at fixed x are stored in row f of a rectangular array [χ]. Pressure val-
ues at several locations may be computed by incrementing f , thereby stacking
rows of [χ]. In MATLAB, the vectorized implementation of this operation defines
two vectors x_f and y_f that hold the coordinates of the set of x̄ f locations.
Then, Chi(f,:)=B.ˆ 2.*cos(j*pi*x_f/L).*cos(n*pi*y_f/H);.
The pressure in Eq. (7) at the various locations at a specific instant t may be com-
puted as the product of [χ] and a column vector of gm ≡ g j,m,0 values, that is,
ρ0 cq0 4c
{ p ({x̄} , ts )} = [χ] {g (ts )} (8)
L H
A time history may be constructed by adjoining column vectors of { p} values at each
instant, that is,
ρ0 cq0 4c
[{ p ({x̄} , t1 )} { p ({x̄} , t2 )} · ··] = [χ] [{g (t1 )} {g (t2 )} · ··]
L H
The highest frequency sets the time step according to the Nyquist criterion. It states
that the sampling interval should be less than half the period of the highest frequency
contained in a signal. The modal series has been truncated J and N . This sets the
highest natural frequency at ω J,N ,0 . The Nyquist criterion is t ≤ 0.5 2π/ω J,N ,03 .
The system parameters, c = 340 m/s, L = 4 m, H = 3 m, give ω J,M,0 = 377 10
rad/s, which leads to t ≤ 8.32 μs.
For the sake of completeness, the signals received at three points along a diagonal
were computed: (x, y) = (L/4, 3H/4), (L/2, H/2), and (3L/4, H/4). The factor
for modal truncation was set at σ = 100, which leads to J = 100, N = 75. The result
appears in Fig. 2.
60
40 x=L/4, y=3L/4
pL/(0 cq0 )
x=L/2, y=3L/2
x=3L/4, y=L/4
20
−20
5 6 7 8 9 10
Time (ms)
Figure 2.
The plots have a jagged appearance. Calculations with t set at one-tenth of

the Nyquist value primarily smoothed out the fluctuations, whereas increasing the
value of σ led to curves whose shape changed from Fig. 2. In other words, a modal
truncation based on σ = 100 is not adequate. All computations were repeated with
σ = 1000. This leads to J = 1000, N = 750, which corresponds to 751751 modes.
The sampling interval under the Nyquist sampling guideline is t = 0.832 μs. Rais-
ing the number of modes and number of instants by a factor of ten relative to Fig. 2
leads to a factor of 100 increase in the computational time. Nevertheless, the effort
is still manageable, with an elapsed time of 1.5 min on a laptop computer with an
Intel I7 CPU.
The result of the refined computation is the data plotted in Fig. 3. Each plot is
much smoother. Some high-frequency wiggles are present. For reasons elucidated
below, it is not possible to eliminate them. At each location described in Fig. 3, the
pressure shows a sharp peak, followed by a rapid decay. The shape of each signal
after this arrival time is reminiscent of the solution to Example 6.5, which explored
the field radiated by an impulsive line source in free space. Indeed, that solution can
be used in conjunction with the method of images to construct the response of the
present system.
150
100 x=L/4, y=3L/4

pL/(0cq0)
x=L/2, y=3L/2
50 x=3L/4, y=L/4
−50
5 6 7 8 9 10
Time (ms)
Figure 3.
According to the method of images the walls at constants x and y serve as mirrors
for the line source at x = y = 0. Figure 4 shows that an infinite number of image
sources is required to satisfy the condition that n̄ · ∇ p = 0 at all walls. A subtle aspect
of the image construction is that the walls at z = 0 and z = W also serve as mirrors.
This leads to a continuous line extending to infinity in the positive and negative z
direction. In other words, each image is an impulsive infinite line source, each of
whose mass acceleration per unit length is 4ρ0 q0 δ (t). Confirmation of the correctness
of this model is the fact that the time at which each peak occurs in Fig. 3 is the distance
from a field point to one of the images in Fig. 4, divided by c. This is the time required
for a signal leaving the image line source at t = 0 to arrive at the field point.
y
2H
2H
2H
H
x
2H L
2L 2L 2L 2L 2L
Figure 4.
A corollary is that the waveform received at any point within the cavity begins
with arrival of the signal that travels directly from the source, followed by the
arrival of pulses resulting from one or more reflections of the source signal. As
the distance along the reflected path increases, the arriving signal is weakened by
cylindrical spreading, as explained in Example 6.5. This conceptual picture explains
the waveforms in Fig. 5, which were obtained by halting computation of Eq. (8) at
tmax = 3L/c. Multiple pulses are seen at each field point. The arrival time for some
is sufficiently close that they overlap.
x=L/4, y=3L/4
100
x=L/2, y=3L/2
pL/(0 cq0 )
100
x=3L/4, y=L/4
100
0
0 5 10 15 20 25 30 35
Time (ms)
Figure 5.
The availability of an alternate solution makes it possible to recognize some faults

with the cavity mode series solution. An exceptionally large number of modes must
be included in the series in order to obtain a high-quality waveform. In contrast, the
method of images merely requires inclusion of only those images whose signal arrives
at the selected field point within the designated time interval. The field of an impulsive
line source is singular at the first arrival time. The modal solution cannot replicate
this singularity. Furthermore, the modal solution shows high-frequency wiggles prior
to the arrival of the signal from each image. This is an impossibility, because early
arrival corresponds to a signal that travels supersonically from an image to the field
point. Both artifices stem from the fact that the impulsive line source excites all modes
equally, as is evident in Eq. (4) for the S j,n values. A very large number of modes is
required to reproduce the discontinuity. Increasing the number of modes as we did
decreases the scale of the precursor signal in duration and amplitude. However, the
precursor cannot be eliminated, and the singularity cannot be replicated, with a finite
length modal series.
10.3.5 Cylindrical Cavity
The analysis of cylindrical cavity modes proceeds much like the development for
rectangular cavities. We align the x-axis with the centerline, with the origin placed
at an end. The mode functions we seek are solutions of the homogeneous Helmholtz
equation subject to rigid or pressure-release conditions at the ends x = 0 and x = L,
and on the interior wall of the cylinder, R = a. In addition, the mode function must
be finite at R = 0. This excludes the Neumann function as a possible description of
the transverse dependence, so that dependence will be a Bessel function. (If there
is a concentric inner cylinder, the finiteness condition is replaced by a boundary
condition on the inner cylinder. In that case, a mode function is formed from both the
Bessel and Neumann functions.) We designate the transverse wavenumber as μ. The
behavior in the axial direction x is like that between parallel walls in a rectangular
waveguide. If the field is not axisymmetric, then the dependence on the azimuthal
angle θ must be harmonic. Hence, any cavity mode may be described as
= B Jn (μR) F (nθ) [cx cos (κx) + sx sin (κx)] (10.3.24)
where
F (nθ) = cos (nθ) , sin (nθ) , or exp (±inθ) (10.3.25)
Setting P = 0 or ∂ P/∂x = 0 at x = 0 leads to either cx = 0 or sx = 0. Then,

satisfying P = 0 or ∂ P/∂x = 0 at x = L yields an infinite number of κL values.
The possible values of μa are determined by satisfying the rigid or pressure-release
conditions at R = a. This leads to the μa values being either a zero of a Bessel
function, or a value at which a Bessel function has an extreme value. The specific
conditions are

∂ ηn,m
Rigid: = 0 =⇒ Jn ηn,m = 0, μ = , m = 1, 2, ...
∂ R R=a a
ηn,m
Pressure-release: | R=a = 0 =⇒ Jn ηn,m = 0, μ = , m = 1, 2, ...
a
(10.3.26)
Thus, three subscripts are required to denote a cylindrical cavity mode. The first is the
order n of the Bessel function, which is the same as the azimuthal harmonic n. The
second is the root number m for the radial wavenumber. The third is the root number
j for the axial wavenumber κ. Furthermore, as indicated by Eq. (10.3.25), two inde-
pendent functions describe the dependence on θ for a given triad of n, m, j values.
Cavity modes are standing waves, so the sinusoidal functions are more appropriate in
the present context. We shall use a subscript α = c or α = s to indicate which sinu-
soidal function is being considered, with this subscript omitted for the axisymmetric
mode, n = 0. Correspondingly, the modes n,m, j,α of a cylindrical cavity are

R
0,m, j = Bn,m,0 J0 η0,m cx cos κ j x + sx sin κ j x
a
R
n,m, j,c = Bm,n, j Jn ηn,m cx cos κ j x + sx sin κ j x cos(nθ), n > 0
a
R
n,m, j,s = Bm,n, j Jn ηn,m cx cos κ j x + sx sin κ j x sin(nθ), n > 0
a
(10.3.27)
It still remains to determine the natural frequency. This is done by ascertaining

the frequency at which a mode function satisfies the Helmholtz equation,
∂2 P 1 ∂P 1 ∂2 P ∂2 P
+ + 2 + + k2 P = 0 (10.3.28)
∂R 2 R ∂R R ∂θ 2 ∂z 2
We also have ∂ 2 n,m, j,α /∂θ2 = −n 2 n,m, j,α and ∂ 2 n,m, j,α /∂x 2 = −κ2j n,m, j,α .

In the transverse direction, we know that Jn ηn,m R/a is a solution of Bessel’s
equation, which means that

∂ 2 n,m, j,α 1 ∂n,m, j,α ηn,m
2
n2
+ =− − 2 n,m, j,α (10.3.29)
∂ R2 R ∂R a2 R
It follows that n,m, j is a solution of the Helmholtz equation if

1/2
ηn,m
2
ωn,m, j,α ≡ ck j,m,n = c κ2j + 2 (10.3.30)
a
The azimuthal harmonic number n does not appear explicitly in this expression.
Nevertheless, inclusion of n in the list of subscripts for the natural frequency is
warranted by the fact that the transverse wavenumbers ηn,m depend on which Bessel
function order is under consideration.
An important aspect of the nonaxisymmetric (n > 0) modes is that they occur as a
pair α = c and α = s with equal natural frequencies. Despite the equality of natural
frequencies, each function is orthogonal to all others. The specific forms of the
general modal orthogonality properties and normalization definition, Eqs. (10.3.8)
and (10.3.9), for a cylindrical cavity are

L
π
a
n,m, j,α r,s,,β Rd Rdθd x = δn,r δm,s δ j, δα,β πa 2 L
0 −π 0
L
π
a
∂n,m, j,α ∂r,s,i,β ∂n,m, j,α ∂r,s,i,β
+
0 −π 0 ∂x ∂x ∂R ∂R

1 ∂n,m, j,α ∂r,s,i,β 2
+ 2 Rd Rdθdz = δn,r δm,s δ j, δα,β ω j,m,n,α πa 2 L
R ∂θ ∂θ
(10.3.31)
There is no need to prove these properties because they are covered by the general
proof.
The case where all mode numbers match, that is, n = r , m = s, j = , and α = β,
is used to set the normalization coefficients Bn,m, j . Note that the same Bn,m, j value
applies to the cosine and sine modes because one is merely rotated about the x-axis
relative to the other. An actual determination of the Bn,m, j values will entail evaluation
of the integral of a product of Bessel functions. Because ηn,m as an eigenvalue, a
formula7 is available for this integral. The result depends on whether the surface is
pressure-release or rigid, specifically
7 M.I. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, p. 485, Eq. 11.4.5
(1965).
⎧ 2
⎪a 2
a 2 ⎪
⎪ Jn ηn,m if Jn ηn,m = 0
R ⎨ 2
Jn ηn,m Rd R =
a ⎪
⎪a2 n 2 2
0
⎪ 1−
⎩ Jn ηn,m if Jn ηn,m = 0 and ηn,m = 0
2 ηn,m
(10.3.32)
Upon determination of the mode functions and natural frequencies, an analysis of
response in the time domain or frequency domain may proceed by following the
steps in the general development provided in Sect. 10.3.1.
EXAMPLE 10.9 A box and a cylinder contain air at standard conditions. The
cross section of the box is square, with a side length of w, and the radius of
the cylinder is a. Both have equal length L in the axial direction, which is
designated as x. The ends at x = 0 and x = L are open to the atmosphere
and each cross section is sufficiently large to set P = 0 at both ends. The other
sides are rigid. The parameter that is of interest is the smallest natural frequency
corresponding to a mode that is not planar transverse to the x axis. In particular,
it is desired that this frequency be a specified value for both configurations.
Derive expressions for a and w for which this condition occurs.
Significance
This example will provide quantitative insight into the manner in which dimensions
and shape affect natural frequencies.
Solution
Let us begin with a circular cross section. The ends are pressure-release, and the same
frequency results from using cos (nθ) and sin (nθ), so a mode function is described by

jπx
n,m, j,c = B Jn μn,m R cos (nθ) sin
L
where j is the root number for the axial eigenvalue, n is the azimuthal harmonic, and
m is the root number for the transverse eigenvalue. The latter comes from satisfying
the condition VR = 0 at R = a, for which the characteristic equation is
ηn,m
Jn ηn,m = 0, μn,m =
a
Equation (10.3.30) tells us that the lowest frequency corresponds to minimum
values of μn,m and κ j . To identify the appropriate value of the former, we examine
Fig. 1, which is a graph of the Bessel function at various orders. The roots ηn,m are the
argument x for which Jn (x) has a maximum or minimum. The first maximum occurs
for n = 0, but this root gives J0 (0) = 1, which is a planar mode. The smallest nonzero
value of x leading to an extreme value of Jn (x) occurs at x ≈ 1.8 for n = 1. A more
accurate value may be found by invoking the identity J1 (ξ) = J0 (x) − J1 (x) /x to
implement a numerical solution. Doing so yields η1,1 = 1.8412.
1
J0(x) J1(x) J2(x) J3(x)
0.5
−0.5
0 5 10 15 20
x
Figure 1.
The axial wavenumber is κ j = jπ/L, with j > 0. The minimum occurs for j = 1.
Thus, the fundamental frequency for nonplanar modes is found from Eq. (10.3.30)
to be
π 2 η1,1 2 1/2
min ω j,m,n = ω1,1,1 = c +
L a
The x dependence of the modes of the box is the same as it is for the cylinder.
The dependence on y and z must be harmonic, with zero slope at the walls. The
corresponding modes are
mπ y nπz
jπx
j,m,n = B cos sin , m = 0, 1, ..., n = 0, 1, ...
w w L
The natural frequencies are

2 1/2
jπ mπ 2 nπz 2
ω j,m,n = c + +
L w w
The lowest natural frequency corresponds to the minimum values of j, n, and m.

The minimum value of j is one. Although the smallest values of m and n are zero,
both being zero lead to a planar mode. However, setting either m or n to zero and the
other to one is acceptable. Thus, we find that

π 2 π 2 1/2
min ω j,m,n = ω1,1,0 = ω1,0,1 = c +
L w
The minimum for each configuration should be , so

2
π 2 η1,1 2 π 2 π 2
= + = + 2
c L a L w
This yields
1.8412 π
a = 2 1/2 , w = 2 1/2
π 2 π 2
− −
c L c L
Although both dimensions depend on L and /c, their ratio is a constant,
a 1.8412
=
w π
The corresponding ratio of the cross-sectional areas is
Acyl πa 2
= 2 = 1.079 1
Arect w
The areas are quite close. The associated mode functions also have some attributes
in common. Figure 2 shows equal value contours of the modes at x = L/2. The
thickened contours are nodal lines. The (1, 1, 1) mode with a cos θ dependence and
the (1, 1, 0) mode of the box are symmetric with respect to the horizontal centerline
and antisymmetric with respect to the vertical centerline. Their only nodal line is
along the vertical centerline. The patterns are rotated 90◦ for the (1, 1, 1, cos θ)
mode of a cylinder and the (1, 0, 1) mode of a box.
z z
1,1,0 1,1,1,cos 
z
 z
sin 
y
Figure 2.
10.3.6 Spherical Cavity

The objective here is to identify the axisymmetric natural modes of a spherical cavity
whose wall is either rigid or pressure-release. In general, the individual terms in a
series of forced modes satisfy the Helmholtz equation. Equation (10.2.14) is such a
series for an axisymmetric field within a sphere. The individual terms in that series are
spherical harmonics. The terms are summed in order to satisfy the boundary condition
of an arbitrary excitation on the spherical wall. An analysis of natural cavity modes
instead requires that each spherical harmonic individually be consistent with either a
zero radial velocity or zero pressure at the boundary. Thus, we begin by considering
a natural mode whose form is
= B jm (kr ) Pm (cos ψ) (10.3.33)
Although the spherical Neumann function also satisfies the Helmholtz equation, it
is excluded by the condition that the pressure at the center must be finite.
As in the previous analyses, the B coefficient will be set by the orthogonality
property. The only other parameter not yet set is k, which is the ratio of the natural
frequency to the speed of sound. This quantity is found by satisfying the condition at
the wall. Rigidity there requires that ∂/∂r = 0 at r = a, whereas a pressure-release
condition requires that = 0 at r = a. The characteristic equations that result from
satisfying these alternative conditions are

Rigid: jm ηm,n = 0 a
, ηm,n ≡ km,n a ≡ ωn,m (10.3.34)
Pressure-release: jm ηm,n = 0 c
The η quantities, which are the eigenvalues, have two subscripts. The first, n, is the
polar harmonic number. Because the spherical Bessel function is oscillatory about a
zero value, there are multiple roots at a specified n. The second subscript, m, is the
root number. Usually, the roots and therefore the natural frequencies are sequenced
in ascending order, so that ηm,n < ηm,n+1 .
Each eigenvalue might lead to a different normalizing coefficient, so those values
are denoted as Bn,m . Thus, we have established that the axisymmetric natural modes
of any spherical cavity are
r c
n,m = Bn,m jm ηn,m Pn (cos ψ) , ωn,m = ηn,m (10.3.35)
a a
Despite the common mathematical form of the mode functions, their radial derivative
will vanish at r = a for a rigid wall, whereas they will be zero at that location for a
pressure-release wall. Because j0 (0) = 1, the first root for m = 0 in the rigid wall
case is η0,0 = 0. Such a mode corresponds to a uniform pressure distribution in the
transverse direction. All other eigenvalues are positive.
The first few roots for each case are given in Table 10.1. As is usual for Bessel
functions, the higher roots are well approximated by adding multiples of π to these
values.
Table 10.1 Nondimensional natural frequencies for the axisymmetric modes of a spherical cavity
whose wall is either rigid or pressure-release
Wall condition Polar harmonic (m) km,1 a km,2 a km,3 a km,4 a
Rigid 0 0 4.4934 7.7253 10.9041
1 2.0816 5.9404 9.2058 12.4044
2 3.3421 7.2899 10.6139 13.8461
Pressure-release 0 3.1416 6.2832 9.4248 12.5664
1 4.4934 7.7253 10.9041 14.0662
2 5.7635 9.0950 12.3229 15.5146
Example 10.7 showed that the modal density of a rectangular cavity increases
as the range of natural frequencies is increased. The spherical Bessel functions are
not exactly periodic functions, so there will not be cases where the eigenvalues for
different modes are identical. However, they are nearly periodic for large arguments,
so modes having different harmonic numbers n might have very close natural fre-
quencies.
The orthogonality conditions for the spherical cavity modes are obtained by spe-
cializing the general conditions, Eqs. (10.3.8) and (10.3.9), to an axisymmetric spher-
ical geometry. Thus, we use a differential volume element in the shape of a ring whose
radius is r sin ψ, that is, d V = (2πr sin ψ) (r dψ) dr . The result is
$a $π 4
2π m,n r,s r 2 sin ψdψdr = πa 3 δm,r δn,s
0 0
3
$ a $ π ∂m,n ∂r,s 1 ∂m,n ∂r,s 2 4π 3
2π + 2 r sin ψdψdr = a δm,r δn,s ωm,n
2
0 0
∂r ∂r r ∂ψ ∂ψ 3
(10.3.36)
As was done for other geometries, the case where the mode numbers match,
m = r and n = s are used to set the scaling coefficients Bm,n . The first of the above
equations is used for this purpose. The actual evaluation will entail evaluation of
an integral containing products of Bessel functions. We may obtain a formula by
introducing into a tabulated integral,8 the relation of spherical and cylindrical Bessel
functions, specifically,

r πa 1/2 r
jn ηn,m = Jn+1/2 ηn,m (10.3.37)
a 2ηn,m a
This relation is useful for computations, as well as for the normalization integrals.
For the latter, it leads to
8 M.I. Abramowitz and I.A. Stegun, ibid, p. 485, Eq. 11.4.5.

⎧
⎪
⎪ 1 a 3 jn+1 ηn,m 2 if jn ηn,m = 0
⎪

a ⎪
⎪
r 2 2 ⎨ 2
jn ηn,m r dr = 1 3 n + 1/2 2 2
a ⎪
⎪ a 1− jn ηn,m if jn ηn,m = 0
0
⎪2
⎪ η
⎪
⎩ n,m
and ηn,m > 0
(10.3.38)
The features unique to the spherical geometry are the specific form of the mode
functions, the natural frequencies, and evaluation of the normalizing coefficients.
After they have been determined, analysis of the pressure field may follow the general
path described in Sect. 10.3.1.
EXAMPLE 10.10 A spherical tank used for chemical processing consists of

concentric spheres that separate different fluids. The outer sphere, whose radius
is a, is rigid. The inner sphere, whose radius is b, is a permeable membrane
that allows the fluids to mix slowly by diffusing through it. This sphere is well
approximated as having zero thickness with neither inertia nor stiffness. The
density and sound speed within the inner sphere are ρ1 and c1 , while ρ2 and c2
are the properties of the fluid contained between the inner and outer spheres.
Derive the characteristic equation and an expression for the cavity modes of
this system. Also identify the orthogonality conditions for these modes.
Significance
Our study of propagation through planar layers entailed making solutions of the
Helmholtz equation satisfies continuity conditions at interfaces. Those concepts are
extended here to spherical waves.
Solution
It is stated that the permeable membrane physically separates the inner and outer
regions, but cannot resist any movement. Thus, it must be that the pressure and
particle velocity on either side of the inner sphere are equal. In addition, the pressure
in both regions must satisfy the Helmholtz equation. The definition of a mode requires
that all locations in a system exhibit an oscillation at the same frequency. Thus, the
acoustic wavenumbers for the Helmholtz equation are k1 = ωc1 and k2 = ωc2 . The
other boundary conditions to be met are zero radial velocity at r = a and finiteness
of P at r = 0. Hence, we seek pressure fields P (1) and P (2) that satisfy
∇ 2 P (1) + (k1 )2 P (1) = 0, 0 < r < b

∇ 2 P (2) + (k2 )2 P (2) = 0, b < r < a
∂ P (2)
= 0 at r = b, P (1) is finite at r = 0
∂r
P (1) = P (2) at r = a
1 ∂ P (1) 1 ∂ P (2)
= at r = b
iωρ1 ∂r iωρ2 ∂r
The continuity conditions at r = b require that the same polar harmonic Pn (cos )
be associated with the mode in the inner and outer regions. Finiteness at r = 0 requires
that the radial dependence of P (1) be a spherical Bessel function. The outer region
does not contain r = 0, so both the spherical Bessel and Neumann functions are
required to describe the radial dependence there. It follows that a suitable ansatz for
the pressure in each region is
(1) (1)
(2)
P(2) = B jn (k1(2)
r ) Pn (cos )
(1)
P = B jn (k2 r ) + C n n (k2 r ) Pn (cos )
When these expressions are substituted into the boundary and continuity condi-
tions, the Legendre function is a common factor for each, so it may be canceled. The
resulting equations are

k2 B (2) jn (k2 a) + C (2) n n (k2 a) = 0

B (1) jn k (1) b = B (2) jn (k2 b) + C (2) n n (k2 b)

1 1 (2)
B (1) jn (k1 b) = B jn (k2 b) + C (2) n n (k2 b)
ρ1 c1 ρ2 c2
The first equation is satisfied by k2 = 0, which leads to ω = 0. Because the outer

boundary is rigid, this is the zero frequency mode. The general discussion suggests
that such a mode exists for this system. This will be proven after we find the modes
whose natural frequency is greater than zero. Thus, k2 may be factored out of the
first equation.
It is convenient to express the coefficient equations in matrix form. In recognition
that the frequency sets the acoustic wavenumbers, we replace k1 and k2 with ω/c1 and
k2 = ω/c2 , respectively. Hence, the matrix representation of the coefficient equations
is T
[D (ω)] B (1) B (2) C (2) = {0}
⎡ ⎤
0 jn (ωa/c2 ) n n (ωa/c2 )
⎢ (ωb/c1 ) − jn (ωb/c2 ) −n n (ωb/c2 ) ⎥ (2)
[D (ω)] = ⎢ jn ⎥
⎣ ρ2 c2 ⎦
j (ωb/c1 ) − jn (ωb/c2 ) −n n (ωb/c2 )
ρ1 c1 n
At an arbitrary ω, the solution of these equations is B (1) = B (2) = C (2) = 0. A non-

trivial solution requires that ω be a value that causes the determinant of [D (ω)] to
be zero, thereby making it impossible to invert the matrix. Thus, the characteristic
equation is
|[D (ω)]| = 0 (3)
The roots of this (scalar) equation constitute the eigenvalues. Stated differently,
they are the natural frequencies corresponding to the axisymmetric cavity modes.
We could reduce the determinant to an algebraic expression, but there is little reason
to do so because evaluation of its roots would require numerical methods. As was
the case for a homogeneous cavity, the oscillatory nature of the Bessel and Neumann
functions leads to an infinite number of natural frequencies ωn,m for a specified polar
harmonic n.
When the determinant of [D] vanishes, one of the coefficient equations described
in Eq. (2) is not independent of the other two. If we drop the third equation, we may
solve the first two equations for B (2) and C (2) as ratios to B (1) , that is
B (2) = βn,m B (1) , C (2) = γn,m B (1)

jn ωn,m b/c1 n n ωn,m a/c2
βn,m = B (1)
jn ωn,m b/c2 n n ωn,m q/c2 − n n ωn,m b/c2 jn ωn,m a/c1

jn ωn,m b/c1 jn ωn,m a/c2
γn,m = − B (1)
jn ωn,m b/c2 n n ωn,m q/c2 − n n ωn,m b/c2 jn ωn,m a/c1
The coefficient B (1) is set by the normalization part of the orthogonality condition,
so it also will depend on n and m. Therefore, we have established that a mode is
described by
(1)
n,m = Bn,m Fn,m (r ) Pn (cos ) (4)
where the radial function is

+
jn ωn,m r/c1 , 0 < r < b
Fn,m (r ) = (5)
βn,m jn ωn,m r/c2 + γn,m n n ωn,m r/c2 , b < r < a
Nowhere in the derivation of the general orthogonality condition in Eqs. (10.3.8)

and (10.3.9) was it required that the fluid be homogeneous. It was implicit to
the derivation that the mode functions are continuous and that their gradient is
continuous. These requirements in the present system are assured by the conti-
nuity conditions at r = b. The general statements call for an integration over the
entire domain, but the modes have different functional dependence in each fluid.
Therefore, we decompose the volume integral into integrals over 0 < r < b and
b < r < a. Each integral accounts for the different definitions of the mode function
in the inner and outer regions. Because the functions are axisymmetric, we may
use dV = 2π (r sin ) (r dψ) dr . The volume of the sphere is (4/3) πa 3 . Thus, the
condition tells us that

π
b
a
n,m j, dV = 2π n,m j, r 2 dr + n,m j, r 2 dr sin dψ
0 0 b
V

π
b
(1) (1)
= 2π Bn,m B j, Pn (cos ψ) P j (cos ψ) (sin ψ) dψ Fn,m (r ) F j, (r ) r 2 dr
0 0

a
4 3
+ Fn,m (r ) F j, (r ) r 2 dr = δn, j δm, πa
b 3
If n and j are different, the left side vanishes regardless of m and because the
Legendre functions are orthogonal according to Eq. (7.1.16). The condition that is
satisfied by the F function is obtained by setting j = n. This property reduces the
orthogonality integral to

(1) (1)
b a
2n + 1 3
Bn,m Bn, Fn,m (r ) F j, (r ) r dr +
2
Fn,m (r ) F j, (r ) r dr = δm,
2
a
0 b 3
(1)
Normalization of Bn,m is obtained by setting = m.
To close the analysis, let us verify that there is a zero frequency mode. If ω = 0,
then k1 = k2 = 0. In that case, jn (k1r ), jn (k2 r ), and n n (k2 r ) are identically zero
unless n = 0. Thus, a zero frequency mode can only occur for n = 0. Furthermore,
because n 0 (0) is infinite, the zero frequency mode requires that C (2) = 0. Substitu-
tion of these values, combined with the property that j0 (0) = 0 and P0 (cos ) ≡ 1,
leads to recognition that only the second of Eq. (2) is not trivial when ω = 0. The
second equation in that case leads to B (2) = B (1) . Thus, the zero frequency mode is
a constant value that is the same in both regions.
(1) (2)
0,1 = 0,1 = B (1) , 0<r <b
This fits the general property of zero frequency cavity modes.
10.4 Approximate Methods
Several factors limit the types of enclosed domains that can be analyzed mathe-
matically. The primary issues pertain to the shape of the cavity and the nature of the
surface bounding the enclosed region. There is an obvious need for a formulation that
is more general. Discretizations using either boundary elements or finite elements
fit this specification. Their implementations for radiation from a vibrating surfaces,
which were described in Chap. 7, are readily modified to address an enclosed region.
The fundamental alteration of a boundary element formulation entails reversing the
surface normal n̄ to point into the domain enclosed by the boundary, rather into
the region exterior to the boundary. A merit of this formulation is that concern for
the issue of “forbidden frequencies” is not relevant. Finite element implementations
also benefit when applied to a cavity because there is no need for infinite elements
or any of the other techniques for modeling an infinite domain.
In any event, a model created with either discretization method is likely to feature
a large number of variables that are the pressures at points an a spatial grid. The
fineness of such a grid must increase as the frequency increases. Consequently,
the computational resources required to implement numerical simulation techniques
makes them prohibitive for the early stages of the design process, as well as for
gaining a preliminary insight when an existing system exhibits anomalous behavior.
Two approximate approaches we shall consider generate relatively small models.

The first formulation, which originated with Rayleigh,9 has two objectives: estimate
the lower natural frequencies and find a series solution that best fits the actual cavity
modes. The development retains the restriction that the boundary be rigid and/or
pressure-release. The second method is known as Dowell’s approximation.10 It is
used to model systems in which the walls are flexible, usually because they are part
of an elastic structure.
10.4.1 The Rayleigh Ratio and Its Uses
We consider here a limited task: Determine the natural frequencies and modes of an
enclosed region whose shape is not amenable to formal mathematical analysis. The
concepts we shall explore are applicable to a variety of physical systems. Indeed,
they usually are encountered in a course on mechanical and structural vibration. Our
implementation is based on the expressions for the acoustic kinetic and potential
energy.
Estimation of the Fundamental Frequency
The foundation for the development is the statement of the basic work-energy prin-
ciple, Eq. (4.5.30). Because we have restricted our attention to cavities whose walls
are rigid (no movement) or pressure-release (no traction force), no power is trans-
ferred out of the fluid domain. Hence, the acoustical energy K E + P E must be
constant. We know that
the cavity modes n are real functions, and Euler’s equation
gives v̄ (x̄, t) = Re i∇n eiωt / (ρ0 ω). This means that the pressure everywhere is
in-phase, as is the particle velocity, with the particle velocity leading the pressure by
90◦ . Consequently, at instants separated by T /2 = π/ω, the kinetic energy is a max-
imum and the potential energy is zero. Midway between these instants, the potential
energy is maximum and the kinetic energy is zero. It follows from Eq. (4.5.30) that
max (K E) = max (P E). When we use the field properties of the nth mode to form
the energy expressions, we find that

1 1
max (K E) = ∇n · ∇n dV
2 ρ0 (ωn )2
V
1 1 (10.4.41)
max (P E) = (n )2 dV
2 ρ0 c2
V
9 J.W.Strutt Lord Rayleigh, Theory of Sound, Vol. 2, Dover (1945 reprint) Sect. 88.
10 Dowell, E.H., and Voss, H.M. (1962). “The effect of a cavity on panel vibration,” AIAA J., 1,
476–477.
Dowell, E.H., Gorman, G.F., and Smith, D.A. (1977). “Acoustoelasticity: General theory,
acoustic natural modes and forced response to sinusoidal excitation, including comparisons with
experiment” J. Sound Vib., 52, 519–542.
10.4 Approximate Methods 365
Equating these expressions yields

1
∇n · ∇n dV
ρ0
V
(ωn )2 =

(10.4.42)
1
(n )2 dV
ρ0 c2
V
This relation is of no direct use, because the analysis leading to a mode function
also gives the natural frequency. Its value lies in how Rayleigh generalized it. The
numerator and denominator are said to be functionals of n . They shall be denoted
as N [n (x̄)] and D [n (x̄)]. The Rayleigh ratio R is the same ratio of functionals
as above, except that n is replaced by a function (x̄) that is not necessarily a
cavity mode,

c 2
∇ · ∇dV
N [ (x̄)] V
R= ≡
(10.4.43)
D [ (x̄)]
2 dV
V
The form of this relation is based on the cavity fluid being homogeneous, which
is typical for an enclosed region. This restriction allows us to factor ρ0 and c out
of the integrals. Nevertheless, the concepts that will emerge are equally valid for a
heterogeneous medium.
The function (x̄) is said to be a trial function. It should be selected to satisfy
two requirements. The first stems from the fact that the pressure within the fluid must
be continuous because the particle velocity must be finite. Therefore, (x̄) must be
continuous everywhere. It also is necessary that any function we select be zero on
any portion of the surface that is pressure-release. The converse condition is that
should not evaluate to zero on the rigid portion of the boundary. (Satisfaction of the
latter condition is not mandatory, but it is advisable.) Why do we need to worry about
these alternatives? The mathematical explanation lies in the calculus of variations.11
A heuristic explanation comes from the observation that the derivation of the Rayleigh
ratio required that there be no power transfer at the boundary. We also know that the
pressure and normal velocity cannot be specified concurrently anywhere. If = 0 on
the pressure-release portion of the boundary, the trial function inherently is consistent
with zero power flow across that region, regardless of the value of n̄ · ∇. On the
other hand, if is not zero on the rigid portion of the boundary, then the fact that
the principle is based on zero power flow will lead to a result that corresponds to
taking n̄ · ∇ to be zero there. This argument does not explain why = 0, rather
than n̄ · ∇ = 0 is the condition that must be set. Nevertheless, it is an inviolable
requirement that the trial function be zero at any point on the surface that is
11 Weinstock, Calculus of Variations, Dover (1974).

pressure-release. (In the calculus of variations, this is said to be a geometric or

imposed boundary condition.)
To identify some important properties of R, we represent the trial function in
terms of the actual mode functions of the cavity. We do not know these functions,
but if we did, such a modal series would feature a set of coefficients an ,
∞

= an n (10.4.44)
n=1
It will be noted that no allowance has been made in this series for the n = 0 mode,
which is the zero frequency mode for an enclosed region whose entire boundary is
rigid. We will examine this case after we have completed the derivations.
Although we do not know the mode functions, we know that they satisfy the
orthogonality conditions in Eqs. (10.3.8) and (10.3.9). In the latter, kn is the wavenum-
ber at the nth natural frequency, kn ≡ ωn /c. We consider the mode function to be
normalized in accord with the first of those conditions. (Normalization merely expe-
dites the derivation.) Thus, substitution of the modal series description of into the
energy functionals leads to
∞
∞
∞

N [ (x̄)] = c2 an a j ∇n · ∇ j dV = c2 (kn )2 an2 V
n=1 j=1 V n=1
∞
∞
∞
(10.4.45)
D [ (x̄)] = an a j n j dV = an2 V
n=1 j=1 V n=1
The modes have been ordered in ascending order of their natural frequency, so that
kn /k1 > 1 if n > 1. Hence, using the preceding to form R, followed by factoring
(k1 )2 ≡ (ω1 /c)2 out of the numerator leads to
∞

kn 2
an2
N [ (x̄)] n=1
k1
R= = (ω1 )2 ∞ (10.4.46)
D [ (x̄)]
an2
n=1
Except for the n = 1 term, each term in the numerator’s sum is greater the corre-
sponding term in the denominator’s sum. It follows that
R ≥ (ω1 )2 (10.4.47)
with the equality sign being the case only if all an other than a1 are zero, which
means that = 1 . In other words, we have shown that the Rayleigh ratio gives
an upper bound to the square of the fundamental frequency of the cavity.
This does not tell us the quality of the upper bound—is it too high by 10%,
100%, or much more? To examine this, let us suppose that the function is close
to the first mode multiplied by a scaling factor. To indicate this proximity, let a1 in
Eq. (10.4.44) be the scaling factor and let all of the other coefficients
be small values
by setting an = εan , n = 2, 3, ..., where |ε| 1 and all an ≤ 1. Equation (10.4.46)
correspondingly indicates that
∞
kn 2 2
a12 + ε2 an
n=2
k1
R = (ω1 )2 ∞ = (ω1 )2 1 + O ε2 (10.4.48)
2
a12 + ε2 an
n=1

The notation O ε2 refers to a number whose order of magnitude is ε2 . The preceding
tells us that the relative error of the estimate for (ω1 )2 will be less than the deviation
of the trial function from the true fundamental mode, provided that the trial function
is a reasonably good estimate, ε < 1.
This leads to a common use of the Rayleigh ratio. To obtain an estimate of the
fundamental frequency of a cavity, we may fabricate a trial function that is zero at all
locations where the surface is pressure-release. Substitution of that function into the
definition of R can be expected to yield an estimated fundamental natural frequency
that is better than the degree to which the trial function approximates the first mode
function.
There is one condition that must be satisfied if this expectation is to be met. As
was stated, the error in the trial function relative to the true fundamental mode must
be O (ε), but we do not know the true mode. If we blindly select trial functions,
we could make a terrible choice. For example, suppose we were to select a trial
function for which a1 = 0.1, a2 = 1, and all other an = 0. This would lead to R =
0.9901[(10−4 ) (ω1 )2 + (ω2 )2 ], which obviously is not close to (ω1 )2 .
This condition usually can be avoided if one is cognizant of a fundamental aspect:
the fundamental mode of a system usually exhibits the minimum number of nodal
lines consistent with any pressure-release conditions. If we select a trial function
having this property, then the estimate for ω1 can be expected to be less than 40%
above the true fundamental frequency, corresponding to R < 2ω1 . We will take up
this issue in the next example.
When all of the boundaries of the cavity are rigid, the fundamental mode is a
uniform pressure and the fundamental frequency is zero. We do not need the Rayleigh
ratio to tell us this, but that is what Eq. (10.4.48) indicates will be approximated.
Can we use our knowledge of the nature of the zero frequency mode to obtain
an approximation of the lowest nonzero natural frequency? The answer lies in the
observation that if the trial function is orthogonal to the first mode, then the coefficient
a1 will be zero. In that case, Eq. (10.4.46) tells us that R ≥ (ω2 )2 , and Eq. (10.4.48)
tells us that the error in the estimate of (ω2 )2 obtained from R will be much less
than the amount by which the trial function differs from the second mode. Thus, we
must modify the trial function in a manner that leads to a1 = 0. The zero frequency
mode is a constant value everywhere. We may scale a mode by any factor, so we may
set 0 (x̄) = 1. (This is what would be obtained from the standard normalization of
modes.) A trial function (x̄) is orthogonal to 0 if

0 dV = (1) dV = 0 (10.4.49)

V V
This integral is the mean value of multiplied by the size of the domain.
Thus, if we wish to determine the fundamental nonzero frequency of a rigid-
walled cavity, we should employ a trial function whose mean value is zero. If our
initial choice for does not meet this specification, we can obtain a corrected function
ˆ (x̄) by subtracting the mean value of (x̄), that is,
1
ˆ (x̄) = (x̄) − mean , mean =
Rigid − walledcavit y : (1) dV
V
V
(10.4.50)
ˆ to construct the Rayleigh ratio will lead to R ≥ (ω2 ) .
Using 2
EXAMPLE 10.11 Consider a two-dimensional cavity whose sides at x = 0,

x = L, and y = 0 are rigid and whose side at y = H is pressure-release. Iden-
tify three alternative trial functions that may be used to estimate the funda-
mental natural frequency of the system. Then, determine which gives the best
estimate when H = L/4, H = L/2, and H = 2L.
Significance
Because the analytical solution for the modal properties is readily obtained, this
example will demonstrate how to interpret the performance of the Rayleigh ratio
approximation.
Solution
The best way to select a trial function usually is to consider a similar system for
which the modes are known. However, we will proceed as though we had no prior
experience with a two-dimensional cavity in order to understand how an unfamiliar
system might be approached. The boundaries correspond to constant values of x
or y, so we consider trial functions whose form is = f (x) g (y). If g (y) were
chosen to be a constant, satisfaction of the pressure-release condition at y = H
would require that the constant be zero. The next simplest function is g (y) being
linear. A linear function whose value is zero at y = H is g (y) = H − y. A constant
nonzero value of f (x) would not violate any requirements because x = 0 and x = L
are rigid, which requires that not be zero at those locations. The overall magnitude
of f (x), as well as g (y), is irrelevant because any scaling factor will cancel when
the Rayleigh ratio is formed. We therefore may set f (x) = 1, which leads to our
first choice for a trial function, 1 = H − y. In general, monomials, that is, x or y to

an integer power, are the easiest to work with because they lead to simple integrals.
Because 1 may be written as x 0 (H − y), it seems reasonable to allow for variation
in the x direction by raising the power to one, which leads to 2 = x (H − y). This
will test what happens if the trial function is zero on a surface that is not pressure-
release. For the third function, let us increase the power of the y dependence, so
that 3 = x (H − y)2 . A portion of the boundary is pressure-release, so the issue of
eliminating the zero frequency mode is irrelevant.
The analytical steps leading to R for each trial function, based on integration over
0 < x < L, 0 < y < H , are tabulated below.
H−y x (H − y) x (H − y)2
∇ −ē y (H − y) ēx − x ē y (H − y)2 ēx − 2x (H − y) ē y

H L H 2 + L2 H5L 4H 3 L 3
N [ (x̄)] c2 H L c2 c2 +
3 5 9
H3L H 3 L3 H 5 L3
D [ (x̄)]
3 9 15

3 3 3 20 3
R c2 c2 + 2 c2 + 2
H2 H2 L 3H 2 L
2
L
R 48 51 109.7
c
H =L/4
2
L
R 12 15 29.67
c
H =L/2
2
L
R 0.75 3.75 4.67
c
H =2L
The value of R cannot be less than the square of the fundamental natural frequency.
The consequence is the general fact that the smallest value of R is closest to the true
value of ω12 . The lowest value for each H/L is listed in bold type. In each case,
the best estimate is obtained with 1 . The actual fundamental frequency is either
ω0,1 = 0.5πc/H or ω1,0 = πc/L, depending on the size of H relative to L. These
values and the lowest estimate obtained from the Rayleigh ratio are listed below.

L L L
H/L ω0,1 ω1,0 R1/2
c c c
0.25 6.283 3.142 6.928
0.5 3.142 3.142 3.464
2 0.785 3.142 0.8660
The true fundamental frequency is shown in bold. The error of the Rayleigh ratio
estimate for H/L = 0.5 and 2 is 10%, but it is greater than 100% for H/L = 0.25.
To understand why the estimate for H/L = 0.25 is poor, we need to consider
the two possible fundamental mode functions, 0,1 = cos (0.5π y/H ) and 1,0 =
cos (πx/L). If H > L/2, the former is the fundamental mode. In it, there is no
variation in the x direction and its value decreases monotonically with increasing y.
The first trial function also does not depend on x, and the fact that it increases with
increasing y, rather than decreasing, is irrelevant. In contrast, when H < L/2, the
fundamental mode is 1,0 , which is independent of y and decreases monotonically
with increasing x. None of the trial functions matches this independence on y. Indeed,
the results suggest that they are approximating 0,1 .
There is no simple rule for selecting a trial function. Picking it to match the
fundamental mode of a similar system is a good idea. Another aspect is recognizing
that the fundamental mode will entail the minimum fluctuation in all directions
consistent with the requirement that it be zero at boundaries that are pressure-release.
If it is crucial to obtain a good estimate, multiple trial functions may be considered,
as was done here. Ultimately, a better approach is to use several trial functions, which
is the procedure developed in the next section.
Rayleigh–Ritz Method
The previous example closed with a suggestion is that the Rayleigh ratio could be
evaluated for several trial functions. Whichever function led to the smallest value of
R would mark the best estimate of (ω1 )2 . This is a flawed approach because it does
not provide a definitive path to an improved answer. Much better results, including
estimates of mode functions, may be obtained from another property of R.
Suppose that the trial function is close to the jth cavity mode, where j may be any
integer. Proximity of to j does not preclude them differing by a scaling factor.
Therefore, this proximity would be manifested in the modal series representation
of the trial function, Eq. (10.4.46), by the coefficient a j being much larger than any
other coefficient. We indicate this by setting an = εan , n = j, where |ε| 1. In
other words, the series representation is
∞

= ajj + εan n (10.4.51)
n=1
n= j
When an in Eq. (10.4.46) is replaced by εan , the result is
∞
kn 2 2
a 2j + ε2 an
kj
2 n=1
n= j
R = ωj ∞ (10.4.52)
2
a 2j + ε2 an
n=1
n= j
The case where the trial function is exactly proportional to the jth normalized mode
corresponds to ε = 0. This is important because differentiation of the above expres-
sion with respect to ε leads to

dR
=0 (10.4.53)
dε ε=0
To understand this property, let us pretend that we know the cavity modes. Then,
we actually could construct the modal series by multiplying any j by a coefficient
a j and then adding to that term small contributions of the other modes, with each
contribution scaled by a small parameter ε. This construction would allow us to
compute the numerator and denominator of the Rayleigh ratio. Altering the value
of ε would yield a different value of R. In a plot of R as a function of ε, the slope
would be horizontal, that is, d R/dε would be zero, at ε = 0. This is the property of
stationarity.
Of course, we do not know the actual mode function. Instead, the concept is to
select a trial function that has adjustable parameters. We can use the stationarity
property to find the conditions those parameters must satisfy in order that the trial
function matches a mode function. It would be problematic to attempt to pursue this
concept by selecting a single trial function of position that contains adjustable para-
meters. It is far easier to use two or more fully prescribed trial functions of position
and to let the adjustable parameters be weighting coefficients for their combination.
To see how this concept may be implemented, suppose we have two identified
trial functions, 1 and 2 . Rather than evaluating R for each function, as we did in
the previous example, let us form a new trial function that is a linear combination of
1 and 2 . That is, let
= b1 1 + b2 2 (10.4.54)
Substitution of this expression into the functionals in Eq. (10.4.43) converts N and
D and therefore R, to algebraic functions of b1 and b2 . The values of b1 and b2 that
lead to a function that most closely fits a mode are marked by R having an extreme
value.
To understand how we may use this property, suppose we were to create a three-
dimensional Cartesian plot in which b1 and b2 are plotted along two axes and the
value of R as a function of a b1 and b2 pair is plotted on the third axis. We wish to
identify points in the b1 b2 plane that lead to extreme values of R. We could do so by
searching for a high or low point or any point where it levels out locally. However,
there is no reason to actually construct such a plot, because the point we seek is
characterized by the gradient with respect to b1 and b2 being zero. The components
of this gradient are the respective partial derivatives, so we seek the condition for
which
∂R ∂R
= 0, =0 (10.4.55)
∂b1 ∂b2
This does not mean that satisfaction of these conditions will yield the actual mode
function. Rather, it states that if we start off with a two-term series representation
of a mode, the coefficients b1 and b2 that give the closest fit to a mode function are
those for which the above derivatives vanish.
We may generalize this notion by considering an N -term series, which requires
that we identify N suitable trial functions. They must be a linearly independent set,
which means that we cannot obtain one of them as a linear combination of the others.
The series is
N
= bn n (x̄) (10.4.56)
n=1
Substitution of this expression into the numerator and denominator functionals con-
verts them to quadratic sums,

⎛ N ⎞
N
N
N
N = c2 bn ∇n · ⎝ b j ∇ j ⎠ dV = N j,n b j b N
n=1 j=1 n=1 j=1
V

⎛ N ⎞ (10.4.57)
N
N M
D= bn n ⎝ b j j ⎠ dV = D j,n b j bn
V n=1 j=1 n=1 j=1
where the coefficients are

N j,n = c2 ∇ j · ∇n dV

V
(10.4.58)
D j,n = j n dV
V
This reduces R to a function of the bm coefficients, so the stationarity property

requires that
∂R
= 0, m = 1, 2, ..., M (10.4.59)
∂bm
According to Eq. (10.4.43), R = N /D, so we have

∂R 1 ∂N ∂D 1 ∂N ∂D
= 2 D −N ≡ −R =0 (10.4.60)
∂bm D ∂bm ∂bm D ∂bm ∂bm
Because D is finite, it must be that
∂N ∂D
−R = 0, m = 1, 2, ..., M (10.4.61)
∂bm ∂bm
These equations are known as the Rayleigh–Ritz method. Walther Ritz derived the
formulation close to the time of Rayleigh’s derivation, but he did so by following a
very different route.12
12 Rayleigh is widely recognized for this formulation because it appears in his monumental texts on
The Theory of Sound. Walther Ritz was a pioneering nuclear physicist who independently developed
The Rayleigh–Ritz equation, Eq. (10.4.61), is quite general. To derive its specific
form, we differentiate Eq. (10.4.57) with respect to a specific bm . This operation is
∂N N N
∂ N N

= N j,n b j bn = N j,n δ j,m bn + b j δn,m
∂bm n=1 j=1
∂bm n=1 j=1
(10.4.62)

N
N
N
= Nm,n bn + N j,m b j = 2 Nm,n bn
n=1 j=1 n=1
The last form results from the symmetry of the coefficients, N j,m ≡ Nm, j , which
follows from their definition. A similar result applies to the derivative of D. Thus,
we obtain the Rayleigh–Ritz equations,

N
N
Nm,n bn − R Dm,n bn = 0, m = 1, 2, ..., N (10.4.63)
n=1 n=1
The nature of the equations to solve becomes evident from their matrix form,
[[N ] − R [D]] {b} = {0} (10.4.64)
This constitutes a general (matrix) eigenvalue problem, whose solution will yield M
eigenvalues Rm /c and corresponding eigenvectors {b}m .
The eigensolutions have several remarkable properties. Identification of those
properties requires considerable analysis, primarily in linear algebra,13 so we will
only discuss the highlights here. The first is an extension of the upper bound theorem.
It asserts that the r th eigenvalue, Rr , is an upper bound for the square of the r th natural
frequency,
Rr ≥ (ωr )2 (10.4.65)
This provides no guidelines as to how much larger is the value of Rm . However,

unless one makes special effort to find unusual functions, all eigenvalues except the
largest can be expected to be within 10% or less of the respective true values, with
the error increasing with increasing mode number.
(Footnote 12 continued)
the formulation. He came to it by applying variational calculus concepts he first developed for
quantum mechanics. The Rayleigh–Ritz formulation actually is a special application of Ritz’ general
approach, which is known as the Ritz series method, see J.H. Ginsberg, Mechanical and Structural
Vibrations, John Wiley and Sons (2000), Chap. 6. This approach has been used for many analyses
in physics and engineering. It also is the foundation for the finite element method. Thus, it is
remarkable that Ritz wrote only two papers that discussed acoustic-type systems: “On the new
method for solving some variational problems of mathematical physics” (1908) and “Theory of
the transverse oscillations of a square plate with free boundaries” (1909). These were the first to
explain mathematically the occurrence of Chaladni lines in plate vibrations.
13 L. Meirovitch, Principles and Techniques of Vibrations, Prentice-Hall (1997) Chap. 8.
The upper bound property is applicable for any set of trial functions. A much
stronger property holds if the series of trial functions is formed in a certain way. Let
us denote the basis functions of the N -term series in Eq. (10.4.56) as n(N ) . To this,
we add another term consisting of a different trial function F, so that the enlarged
set of functions is
+1)
n(N +1) = n(N ) , n = 1, ..., N , N(N+1 =F (10.4.66)
Correspondingly, the eigenvalues obtained from the two series are denoted as Rr(N ) ,
r = 1, 2, ..., N , and Rr(N +1) , r = 1, 2, ..., N + 1. The result will satisfy the sepa-
ration theorem, which states that the eigenvalues for the shorter length series fall
between the eigenvalues for the longer series, according to
R1(N +1) ≤ R1(N ) ≤ R2(N +1) ≤ R2(N ) ≤ ... ≤ R (N ) (N +1)

N ≤ R N +1 (10.4.67)
In words, this property tells us that as we lengthen the series, the estimate for any
natural frequency will decrease or be unchanged. This is a remarkable result in light
of the upper bound theorem, because the estimated value must be greater than the
true value. Therefore, it must be that lengthening the series will cause the eigenvalues
to converge from above to the true natural frequencies.
A corollary of convergence of the eigenvalues is convergence to the true mode
functions of the trial function series. This attribute follows from Eq. (10.4.52), which
2
states that if a value of R is close to ω j , then it must be that the corresponding a j
coefficient is much larger than any of the other coefficients. To construct a mode, let
(bn )r be the nth element of the eigenvector corresponding to Rr . The trial function
series corresponding to the r th eigenvector is

N
r(N ) = (bn )r n (x̄) ≡ [1 (x̄) 2 (x̄) · · · N (x̄)] {b}r (10.4.68)
n=1
The superscript “N ” has been inserted to remind us that we have constructed an

N -term series approximation of the actual cavity mode r .
EXAMPLE 10.12 It is desired to identify the natural frequencies and mode

functions of a two-dimensional cavity in the shape of an ellipse defined as
(x/a)2 + (y/b)2 = 1, where a and b are the semimajor and semiminor axis
lengths. The elliptical surface of the cavity is rigid. The mode functions of
interest are those that are symmetric with respect to the x z and yz bisecting
planes, with a nonzero natural frequency. For the case where a = 2b, use
series containing two to six terms to determine the natural frequencies and
mode functions by the Rayleigh–Ritz method. Assess the frequencies in light
of the separation theorem, and compare the results for the first three mode
functions.
Significance
Analysis of the mode functions for an elliptical shape is an imposing problem if it
is done by seeking eigensolutions of the Helmholtz equation. Thus, it is systems
like this for which the Rayleigh–Ritz method is especially useful. The solution will
demonstrate how exploiting a system’s symmetry properties can improve the analysis
without requiring greater computational effort. The results will exemplify the general
performance characteristics of the Rayleigh–Ritz method.
Solution
The elliptical wall is rigid, so the trial functions should not vanish on the boundary.
Using a power series in each direction is acceptable, and those functions are easy
to use. One consideration is the zero frequency mode, which this system possesses.
When the Rayleigh ratio is used to estimate the fundamental frequency for such
systems, the mean value of the basis function is subtracted from the function in
order to estimate the fundamental nonzero frequency. We could do the same here by
subtracting from each trial function its mean value over the elliptical cross section.
In the Rayleigh–Ritz method, many modes are obtained, so a simpler alternative is to
include a trial function that is uniform over the cross section. Suppose we use power
series for both the x and y dependencies. We form as a product of the individual
series, and designate the independent coefficients as cm ,

= α0 + α1 x + α2 x 2 + · · · β0 + β1 y + β2 y 2 + · · ·
(1)
= c1 + c2 x + c3 y + c4 x 2 + c5 y 2 + c6 x y + · · ·
It makes sense to halt the series at a stage where all combinations of x and y up to the
same degree have been included. Equation (1) corresponds to truncation at quadratic
terms.
Equation (1) would be a valid starting point for a general analysis. However, it
has been stipulated that we should find the symmetric mode functions. A function
is symmetric with respect to the x z plane if f (x, y) = f (x, −y), and symmetry
with respect to the yz plane requires f (x, y) = f (−x, y). These properties will be
obtained if the x and y polynomials in Eq. (1) are even functions of x and y, which
means that they should be polynomials that contain only even powers. Thus, the
series we shall use is formed as

= α0 + α2 x 2 + α4 x 4 + · · · β0 + β2 y 2 + β4 y 4 + · · ·
(2)
= c1 + c2 x 2 + c3 y 2 + c4 x 4 + c5 y 4 + c6 x 2 y 2 + · · ·
Note that this series has the same number of terms as Eq. (1), but it is quartic and
therefore can be expected to better fit the true modes. The trial functions are the
coefficients of the respective cm in Eq. (2). It is useful to work with their dimensionless
equivalents, which are
x2 y2 x4 y4 x 2 y2
1 = 1, 2 = , 3 = , 4 = , 5 = , 6 = (3)
a2 b2 a4 b4 a 2 b2
Because the functions are even with respect to x and y, we may implement
Eq. (10.4.58) by integrating over one quadrant and then multiplying the result by four.
1/2
Hence, the domain of integration will be 0 < y < b 1 − (x/a)2 , 0 < x < a.
These functions have the generic form j = (x/a)r j (y/b)s j . Thus, the coefficients
for the numerator of R are described by

a
b(1−(x/a)2 ) 1/2 r −1 s
rj x j yj j s j x r j y s j −1
N j,m = 4c 2
ēx + ē y ·
0 0 a a b b a b

rm x rm −1 y sm
sm x rm y sm −1
ēx + ē y d yd x
a a
b b a b

1
1−(x ) 1/2
2
r j rm r j +rm −2 s j +sm
= 4abc2 x y
a2
s j sm
0 0
r +r s +s −2
+ 2 x j m y j m dy d x
b +

1
b r j rm r +r −2 2 (s j +sm +1)/2
= 4c 2 x j m 1 − x
0 a s j + sm + 1
,
a s j sm r j +rm 2(s j +sm −1)/2
+ x 1− x d x , s j > 0, sm > 0
b s j + sm − 1
(4)
Cases where either s j = 0 or sm = 0 are readily handled separately. The denominator
coefficients are

a
b(1−(x/a)2 )1/2 r s r s
x j y j x m y m
D j,m =4 d yd x
0 0 a b a b

1 (5)
1 r +r 2 (s j +sm +1)/2
= 4ab x j m 1 − x dx
0 s j + sm + 1
The most expedient approach is to use numerical methods to evaluate the integrals.
A small saving is that [D] and [N ] are symmetric, so it is sufficient to compute only
D j,m and N j,m for m ≤ j.
To address the fact that only the value of b/a is specified, let us place the
eigenvalue problem into nondimensional form. To that end, let [N ] = c2 N̂ and

[D] = a 2 D̂ . Doing so converts Eq. (10.4.64) to

a2
N̂ − 2 R D̂ {b} = {0} (6)
c

The eigenvalues are a 2 /c2 Rn , which are approximations of the square of nondi-
mensional natural frequencies kn a.
The computer algorithm that was implemented began by defining row vectors for
the exponents in Eq. (3) sequenced in the order of the trial functions, such that
[r ] = [0 2 0 4 0 2] , [s] = [0 0 2 0 4 2]
The coefficients in Eqs. (4) and (5) were determined numerically. The MATLAB
numerical integration routine that was used is quadl. It requires that the function that
is integrated depends only on the integration variable. This is achieved by defining [r ]
and [s] prior to an outer loop over j = 1, 2, .., 6 and an inner loop over m = 1, .., j.
Inside these loop, anonymous functions defining each integrand are defined with the
r and s values referred to explicitly, for example
N_integrand=@(x) 4*b_a*r(j)*r(m)*x.ˆ(r(j)+r(m)- 2).*(sqrt(1 - x.ˆ2))...
.ˆ(s(j) + s(m) + 1)/(s(j) + s(m) + 1);
The name of each function is thefirst argument
for the quadl routine. Completion
of the loops over m and j fills in N̂ and D̂ . The next set of operations is solution
of the eigenvalue problem for series lengths ranging from two tosix. This
is done by
creating a loop for N from to six. Within it, submatrices of N̂ and D̂ are input
to the eigenvalue solver. In MATLAB, the operation is [b_vals, eigvals] =
eig(N_coeff(1 : N, 1 : N), D_coeff(1 : N, 1 : N));. Each
column of b_vals is an eigenvector,
and the diagonal elements of eigvals are
the corresponding values of a 2 /c2 Rn . The diagonal elements of eigvals for
each N are saved in a matrix for examination upon completion of all operations.
The eigenvectors are needed to generate the mode functions. A useful way in
which they may be visualized is with a surface plot, but doing so entails a change in
how Eq. (10.4.68) is evaluated. Suppose [X ] and [Y ] hold the values of x/a and y/b
at a grid of points in the ellipse. (These matrices may be created by the meshgrid
command in MATLAB.) Let us expand the summation form of the aforementioned
equation, such that
r(N ) = 1 (x̄) (b1 )r + 2 (x̄) (b2 )r + · · · (10.4.69)
where r(N) is the r th mode function, the j (x̄) are trial functions evaluated at
x̄, and b j r is the jth element in column r of the eigenvector matrix. (In the
present case, this matrix is b_vals.) If we define j (x̄) to be a matrix evalu-
ated at each point in the grid, then the result will be the value of r(N ) at all grid
points. The sum may be evaluated cumulatively. MATLAB code for these operations
is Psi_mode_r=zeros(size(X)); for j=1:N; Psi_j= X.ˆr(j)*
(Y/b_a).ˆs(j);. Psi_mode_r=Psi_mode_r+Psi_j*b_vals(j,r);
end;.
Because 1 = 1, so that ∇1 = 0, the first row and column of [N ] are zero,
which leads to R1 = 0 for any series length. The corresponding mode function is a
constant. The results for the values of Rn (a/c)2 are described in the tabulation. The
separation theorem states that any value in a column for N > 2 should be less than
or equal to the value to the left of it, and greater than or equal to the value above
it. The tabulated values are consistent with this theorem. We also see that the last
value of R at any series length is the one that changes most when a term is added
to the series. This means that the last mode, whose natural frequency is highest, will
be the most in error. The typical overall trend is that at any series length, the error
in the eigenvalue, which may be identified by comparison to a solution of the field

equations or by examining the trend in a tabulation, increases with increasing mode
number. All except the highest may be expected to be reasonably close to the true
value.
N =2 N =3 N =4 N =5 N =6
R1 (a/c)2 0 0 0 0 0
R2 (a/c)2 16.0000 15.4534 11.9353 11.8629 11.7446
R3 (a/c)2 74.5466 70.7570 53.9301 46.5883
R4 (a/c)2 87.3552 86.8430 69.9076
R5 (a/c)2 354.1028 177.7429
R6 (a/c)2 434.1451
The mode functions converge more slowly than the natural frequencies. The
changes as the series is lengthened are less readily quantified, so comparisons usu-
ally are visual. Figure 1 compares the result of using Eq. (10.4.68) to evaluate 2 for
several series lengths. The differences are not great. Indeed, the high quality of the
N = 2 result, for which the corresponding value of R2 is 36% too large, is somewhat
surprising.
N = 2, Mode #2 N = 4, Mode #2
2
1
1.5
1 0.5
0.5 0
0 −0.5
−0.5 1 −1 1
0.5 0.75 0.5 0.75
0.25 0.5 0.5
0.25 x/a 0.25 0.25 x/a
y/a 0 0 y/a 0 0
N = 6, Mode #2
1.5
1
0.5
0
−0.5
1
0.5 0.75
0.5
0.25 0.25 x/a
y/a 0 0
Figure 1.
Length limitations prevent a full comparison of the mode functions for various
series lengths. Figure 2 shows some modes for N = 6. There seems to be little evi-
dence of the modal distributions for a circular cross section, but 2 does resemble
the (1, 0) mode for a rectangular shape whose aspect ratio is 2 : 1.
N = 6, Mode #2 N = 6, Mode #3
1.5 1.5
1 1
0.5
0.5
0
0 −0.5
−0.5 −1
1 −1.5
1
0.5 0.75 0.5 0.75
0.25 0.5 0.5
0.25 0.25
0 0
x/a 0 0.25
x/a
y/a y/a 0
N = 6, Mode #4
N = 6, Mode #5
1 1
0
0
−1
−1
−2
−3 −2
0.
−4 1 1
0.5 0.75 5 0.75
0.5 0.25 0.5
0.25 0.25 0.25
y/a 0 0
x/a y/a 0 0 x/a
Figure 2.
10.4.2 Dowell’s Approximation
Evaluation of the time-domain response within a cavity whose walls are compliant
is a core problem for the subject of structural acoustics. Usually, such walls are the
surface of an elastic structure. The two media interact in two ways. The pressure
exerted by the fluid constitutes a force system that is applied to the structure. Its
effect must be incorporated into the structure’s equations of motion. The excitation
applied to the fluid stems from the requirement that the normal component of its
particle velocity and that of the structure be equal everywhere along the surface at
which they meet. Analytical solutions of such problems can be found only for very
simple circumstances. Numerical modeling techniques often are used, but as noted
previously, they might be unwieldy for some applications.
The fundamental formulation of the Rayleigh ratio is valid if the structure is elastic
because all that was assumed is that the system is time-invariant and that the sum of
kinetic energy and potential energy is conserved in a free vibration. The difficulty
with implementation of either the Rayleigh ratio or Rayleigh–Ritz method lies in
the selection of trial functions. In addition to selection of one or more functions
for the fluid, it would be necessary to formulate a series to describe the structure’s
displacement. This is where a dilemma arises, because there is no convenient way to
assure that the pressure and displacement series satisfy continuity conditions at the
fluid-structure interface.
The modeling formulation derived from Dowell’s approximation addresses this
issue. The analytical method is general. It has the beneficial feature of leading to
model equations that feature a relatively small number of variables. However, these
benefits are obtained by approximating one of the fundamental fluid-structure inter-
action mechanisms.
The portion of the fluid’s boundary that is the interface with the structure is denoted
as Se . The displacement of the structure normal to Se is an unknown function w (x̄s , t)
that must be determined as part of the analysis. The remainder of the bounding surface
is denoted as Sv on which the normal velocity is taken to be a known function v̂ (x̄s , t).
Thus, Sv includes any regions that are rigid. Both w and v̂ are taken to be positive if
they correspond to movement into V, which is the sense in which the surface normal
n̄ (x̄s ) is defined. Hence, continuity at the surface requires that
n̄ · v̄ = ẇ, x̄s ∈ Se
(10.4.70)
n̄ · v̄ = v̂, x̄s ∈ Sv
The core concept of Dowell’s approximation is a series representation of pressure

in which the basis functions are the cavity modes for a domain having the same shape,
but whose walls are rigid. Analysis of such modes has already been discussed, but
their actual determination might represent a significant task if the cavity is irregular
in shape. We shall proceed on the assumption that we have found those modes, which
we denote as R,n , where n is the mode number and “R” serves to remind us that
these are rigid-cavity mode functions. The objective is to derive a set of equations
whose solution will describe the pressure field within the cavity and the response
of the structure as functions of time. Correspondingly, the coefficients of the cavity
mode series are taken to be functions of time to be determined. Thus, the ansatz is

N
p (x̄, t) = R,n (x̄) pn (t) (10.4.71)
n=0
where the summation index n begins from zero as a reminder to include the zero
frequency mode in which the pressure is uniform throughout the cavity.
The fundamental nature of Dowell’s approximation is embedded in the above
series. In the time domain, Euler’s equation states that the particle acceleration normal
to the surface is proportional to the normal component of ∇ p. By definition, n̄ ·
∇ R,n = 0 on the boundary. Hence, the series construction gives a zero acceleration
on the surface, whereas the time derivative of Eq. (10.4.70) requires that it be ẅ. The
development proceeds by ignoring this paradox, but the surface continuity conditions
will enter the analysis at a later juncture.
In addition to excitation by movement of the boundary, there might be a set of
sources at several locations x̄m . Thus, the pressure p is governed by the inhomoge-
neous wave equation, Eq. (10.3.11), which is
1 ∂2 p
∇2 p − = −ρ0 Q̇ m (t) δ (x̄ − x̄m ) (10.4.72)
c2 ∂t 2 m
Equations (10.4.70) are the boundary conditions governing p (x̄, t). The equations
satisfied by the rigid-cavity modes are
2
∇ 2 R,n + kr,n R,n = 0, x̄ ∈ V
(10.4.73)
n̄ · ∇ R,n = 0, x̄ ∈ Se ∪ Sv
With the aim of employing Green’s theorem, we multiply the field equation for a
specific rigid body R,n mode by p, and the field equation for p by R,n . Then, we
subtract the latter from the former and integrate over the fluid domain V. The result
of these operations is

2 1 ∂2 p
p ∇ R,n + kr,n r,n − r,n ∇ p − 2 2
2 2
dV
c ∂t
V

= ρ0 Q̇ m (t) R,n (x̄m ) (10.4.74)
m
Green’s theorem converts the terms containing the Laplacian to a surface integral
according to

p∇ R,n − R,n ∇ p dV ≡
2 2
∇ · p∇ R,n − R,n ∇ p dV
V V

=− p n̄ · ∇ R,n − R,n (n̄ · ∇ p) dS
S
(10.4.75)
The minus sign preceding the surface integral stems from the normal direction n̄
being defined to point into V.
The surface integral extends over the entire boundary S. The normal derivative
of R,n is zero everywhere on this surface, whereas n̄ · ∇ p is related to the surface
normal acceleration by Euler’s equation and the time derivative of Eq. (10.4.70).
Thus, we have established that

∂ v̂
p∇ 2 R,n − R,n ∇ 2 p dV = − R,n ρ0 ẅdS − R,n ρ0 dS
∂t
V Se Sv
(10.4.76)
It is important to recognized that although the rigid-cavity series cannot give the
correct particle velocity at the boundary, the implementation of Green’s theorem
uses the actual boundary velocity. Thus, the formulation accounts in an average
sense for the way movement at the boundary influences the pressure.
Substitution of the preceding relation into Eq. (10.4.74) leads to

1 ∂ 2 p 2 ∂ v̂
R,n + k r,n p dV − R,n ρ0 ẅd S − R,n ρ0 d S
c ∂t
2 2 ∂t
V Se Sv
= ρ0 Q̇ m (t) R,n (x̄m )
m
(10.4.77)
The integrands of the surface integrals consist of known quantities, so they belong
on the right side of the equation. Furthermore, we have already represented p as
a series of rigid-cavity modes according to Eq. (10.4.71). When we substitute that
description of p into the preceding, R,n multiplies every R, j in the summation.
The first orthogonality condition for cavity modes is

R,n R, j dV = V (10.4.78)
V
Thus, orthogonality filters out of the summation all modes except number n. We
thereby obtain a set of uncoupled differential equations for the modal coefficients,
2 ∂ v̂
V p̈n + ω R,n pn = ρ0 c2 R,n ẅdS + ρ0 c2 R,n dS
∂t
Se Sv (10.4.79)
+ ρ0 Q̇ m (t) R,n (x̄m ) , n = 0, 1, ..., N
m
where ω R,n ≡ ck R,n are the natural frequencies of the rigid cavity. Each of the terms in
this expression represents a different physical process: p̈n represents compressibility,
(ωn )2 pn represents inertia, Q̇ m is an excitation that results from injecting mass into the
domain, ∂ v̂/∂ v̂ is a surface source distribution resulting from causing the boundary
to move, and ẅ is a surface source distribution resulting from movement of the elastic
structure at its interface with the fluid.
Equation (10.4.79) constitutes a set of N + 1 ordinary differential equations for
the N + 1 pressure coefficients. If there were no structure, these equations would suf-
fice. Our interest here is the interaction of the acoustic fluid and the structure. In such
situations, the surface displacement w is a quantity that must be determined. It is here
that the equations of motion for the structure enter the analysis. The displacement
of the structure may be described in terms of a set of variables called generalized

coordinates and denoted as q j . These variables might be the coefficients of a series,
or they might be the mesh displacements in a finite element model. Our interest is
in systems that are linear and time-invariant, which means that a reference configu-
ration exists in which the system may remain in a state of static equilibrium. In such
cases, it is possible to define the generalized coordinates such that zero values for
all place the structure at its reference position. For a given system, there are several
methods for formulating the differential equations of motion governing the general-
ized coordinates. Usually, the derivation of those equations is founded on Lagrange’s
equations, Hamilton’s principle, or the principle of virtual work, which are closely
related.14 The result will be the structure’s equations of motion in a canonical form,
whose matrix description is
- .
[M] {q̈} + [K ] {q} = {F} + Fpressure (10.4.80)
The matrices [M] and [K ], which are constant, square, and symmetric, are the inertia
and stiffness. Another structural effect is dissipation, but the actual mechanisms
by which energy is lost usually are uncertain. Consequently, dissipation is usually
introduced at a later stage. - .
The elements of the column vectors {F} and Fpressure are generalized forces.
The former represents the excitation forces acting on the structure, whereas the
latter contain the effects of the surface pressure. Determination of {F} is part of
the
- standard . process by which equations of motion are derived. Determination of
Fpressure is an extension of that process. Identification of the force terms begins
with a general description of the surface displacement w (x̄s , t). A linear structural
model leads to a description of the displacement of any point as a series in which each
generalized coordinate is multiplied by a spatial function. These functions depend
on how the structure is modeled, but the details are not important here. The surface
normal displacement is obtained by evaluating that series at an arbitrary surface
point x̄s and then taking a dot product of that vector with the outward normal n̄ (x̄s ).
The consequence is that w also is represented as a linear series in which the spatial
functions depend on x̄s . The general form of this series is

M
w= χm (x̄s ) qm (10.4.81)
m=1
Clearly, the specific form of the χm functions depends on how the structure is mod-
eled.
The surface displacement enters the description of the pressure loading through the
concept of virtual work. This quantity is unlike the actual work done by forces. Rather,
it is the work that would result if each generalized coordinate were incremented by
an arbitrary differential amount. These increments are denoted as δqm to distinguish
14 J.H. Ginsberg, Engineering Dynamics, Cambridge University Press (2008) Chap. 7

them from actual increments dqm that occur in the course of a response. Increments
of the generalized coordinates lead to a virtual displacement. At the surface, this
displacement is δw. The functions χm are set by the manner in which the structure
has been modeled and therefore are unchanged. Thus, the virtual displacement is
given by
M
δw = χm (x̄s ) δqm (10.4.82)
m=1
The resultant force exerted by the pressure acting on a patch of the surface is
p (xs , t) dS in the direction of −n̄ (x̄s ) and δw at that location is positive if it is in
the direction of +n̄ (x̄s ). Therefore, virtual work done by the pressure acting on Se
is
δWpressure = − p (x̄s , t) δwd S (10.4.83)

Se
Substitution of the cavity mode series for p and the preceding expression for δw into
the virtual work gives
M
N

δWpressure = − pn R,n (x̄s ) [χm δqm ] d S (10.4.84)
n=0 m=1 S
e
- .
On the other hand, the fundamental definition of the elements of Fpressure is that
they are the coefficients in a canonical description of δWpressure , specifically

M

δWpressure = Fpressure m
δqm (10.4.85)
m=1
These alternative descriptions of δWpressure must be the same for any set of δq j values.
This can only be true if corresponding coefficients of δqm are the same, which leads
to
N
Fpressure m = −Se pn m,n

n=0 (10.4.86)
1
m,n = χm (x̄s ) R,n (x̄s ) dS
Se
Se
It follows that the pressure force vector is

- .
Fpressure = −Se [] { p} (10.4.87)
In turn, the structure’s equations of motion become
[M] {q̈} + [K ] {q} + Se [] { p} = {F} (10.4.88)

The series description of surface displacement, Eq. (10.4.81), is also used to

describe the surface acceleration term in the pressure equations. Substitution of the
displacement series into Eq. (10.4.79) shows that
⎡ ⎤

M

M
R,n ẅdS = ⎣ R,n (x̄s ) χm (x̄s ) dS ⎦ q̈m = Se m,n q̈m (10.4.89)
Se m=1 Se m=1
When the preceding expression for the surface displacement is substituted into
Eq. (10.4.79), the result is
2
M
V p̈n + ω R,n pn − ρ0 c2 Se m,n q̈m = n (t) , n = 0, 1, ..., N
m=1
(10.4.90)
where the source coefficients are

∂
n = ρ0 c2 R,n (x̄s ) v̂ (x̄s , t) dS + ⊂0 R,n (x̄m ) Q̇ m (t) (10.4.91)
∂t m
Sv
In keeping with the matrix representation of the structural equations of motion,

the equations for the acoustical response may be written in that form. Stacking the
pressure equations above the structural equations leads to a set of ordinary differential
equations whose form is

d2 { p} { p}
[M] 2 + [K] = {F} (10.4.92)
dt {q} {q}
The system matrices are

V [I ] −ρ0 c2 Se []T
[M] =
[0] [M]

V (ω R )2 [0]
[K] = (10.4.93)
Se [] [K ]

{}
{F} =
{F}

where (ω R )2 is a diagonal matrix whose elements are the rigid-cavity natural fre-
quencies and [I ] is an identity matrix. Thus, we have derived a set of coupled equa-
tions whose number equals the number of unknown pressure coefficients, N + 1,
plus the number of unknown displacement coefficients, M. The column vector {}
represents source excitations stemming from injecting mass into the fluid, as well as
from imposing a movement of the boundary. The displacement and pressure variables
are coupled by the matrix [].
These equations cover most situations, but the case where some region in the wall
is pressure-release is not addressed. Just as rigid mode series, Eq. (10.4.71), cannot
give a nonzero normal velocity on the boundary, it cannot give a zero pressure on the
boundary because the rigid-cavity modes are not zero there. An approximation of that
condition may be obtained by replacing the pressure-release region with an extremely
compliant structure, such as an extremely thin membrane whose surface tension is
very low. However, doing so might cause the final equations to be ill-conditioned if
the thickness and surface tension are too small.
How to solve the equations derived from Dowell’s approximation depends on the
type of response we seek. If the excitation is time harmonic, we may obtain the
steady-

state response from a complex variable
representation.
We set p n = Re Pn eiωt ,

qn = Re ξn eiωt , n = Re ˆ n eiωt , and Fn = Re F̂n eiωt , which leads to
⎧ ⎫
⎨ ˆ ⎬
{P}
[K] − ω 2 [M] = (10.4.94)
{ξ} ⎩ F̂ ⎭

Resonances occur at frequencies for which the determinant [K] − ω 2 [M] vanishes.
Solution of the differential equations in the time domain is a little more difficult.
Coupled linear ordinary differential equations like Eq. (10.4.92) arise in mechanical
and structural vibratory systems. In that area, the equations are commonly solved by
modal analysis, in which the modes constitute the eigensolutions of the homogeneous
equations. That formulation is not directly applicable here because the coefficient
matrices are not symmetric. Consequently, a modal analysis would require consid-
eration of left and right eigenvectors, which is the way in which the equations of
motion for rotordynamic systems may be solved.15 The formulation of such solu-
tions exceeds the scope of the present work.
The differential equations also may be solved by implementing a numerical
algorithm, such as one of the Runge-Kutta versions. Most algorithms require that
Eq. (10.4.92) be placed in first-order form. This is done by defining a state vector
{Z } that consists of all unknown variables and their first derivatives,
T
{Z } = { p}T {q}T { ṗ}T {q̇}T (10.4.95)
The variables and their first derivatives may be extracted from {Z } according to

{ p} d { p}
= [[I ] [0]] {Z } , = [[0] [I ]] {Z } (10.4.96)
{q} dt {q}
15 Ginsberg, Mechanical and Structural Vibrations (2001), Ch. 11.

where the size of the partitions equals the number of unknowns contained in { p}
and {w}, which is N + M + 1. The derivative identity describes the fact that the
derivative of the first expression equals the second, from which it follows that
d
[[I ] [0]] {Z } = [[0] [I ]] {Z } (10.4.97)
dt
We also use Eq. (10.4.96) to describe the basic differential equations, Eq. (10.4.92),
as
d
[[0] [M]]T {Z } + [[K] [0]] {Z } = {F} (10.4.98)
dt
Stacking the derivative identity above this form yields

[I ] [0] d [0] [I ] {0}
{Z } = {Z } + (10.4.99)
[0] [M] dt − [K] − [0] {F}
To solve these equations with mathematical software, one merely needs to provide
the equation solver with the coefficient matrix multiplying d {z} /dt and a function
that evaluates the right side of the equality sign at an arbitrary instant based on the
current value of {z}. Initial values required to begin the integration process are the
values of { p}, {w}, { ṗ}, and {ẇ} at t = 0.
We have seen that Dowell’s approximation leads to a set of differential equa-
tions that are readily solved and are quite comprehensive in the variety of systems
to which the may be applied. However, they do have two flaws. The structural and
fluid velocities are not inherently continuous in the normal direction at the interface
where the media meet. This was observed at the outset, but another fault was encoun-
tered when the differential equations were assembled. The coupled equations are not
symmetric. It was shown in Chap. 6 that the laws of acoustics are reciprocal. That
is suppose a source is situated at one point and the response is measured at another.
The same response would be measured at the original source location if the source
were applied at the measurement point. It can be proven that this property requires
that the matrix equations for the response be symmetric. This is not the nature of
Eq. (10.4.92). Another consequence of the asymmetry of [M] and [K] arises in the
computation of natural frequencies. These correspond to nonzero responses in the
absence of excitation, so they are solutions of an eigenvalue problem,
T
[K] − ω 2 [M] { p}T {q}T = 0 (10.4.100)
It is possible that some eigenvalues ω 2 will be negative, or they might occur as pairs
of complex conjugate values, even though both matrices are real. Neither condition
corresponds to a periodic motion. Such a prediction is erroneous, because periodic
free vibration is the only possibility in any system that does not dissipate energy, as
we saw in the development of the Rayleigh ratio.
Despite these flaws, results obtained from Dowell’s method have been found to
be quite usable, especially when the fluid is air. Indeed, the formulation has been
widely employed to predict cabin noise for aircraft and automobiles. The reason for
this success is that the air presents a low impedance to movement of a solid structure.
Consequently, the fluid’s particle velocity at the interface will be much lower than
it is somewhere in the interior of the cavity. (Such an assertion is not likely to be
true for a heavy fluid like water.) In addition, issues regarding reciprocity seldom
are at the forefront of the questions that the model equations must answer. In any
event, the availability of a rather accessible simulation technique makes Dowell’s
approximation a very useful predictive tool for design and diagnostic tasks.
EXAMPLE 10.13 The sketch depicts a rectangular cavity in which both walls
parallel to the plane of the sketch are rigid. At the top, a portion of the wall
consists of a plate whose mass is M1 . The wall at the left side is another plate
whose mass is M2 . Both plates are extremely stiff, so their movement is essen-
tially a rigid body translation, vertically for the upper plate and horizontally
for the side plate. The width of both plates perpendicular to the plane of the
diagram is the same as the width of the waveguide, which is 300 mm. The upper
plate is supported by spring K and dashpot D, so this subsystem constitutes a
one-degree-of-freedom oscillator. A horizontal force applied to the plate on the
left is adjusted such that at any frequency ω, it induces a translational velocity
of V0 sin (ωt), where V0 = 50 mm/s. All other regions of the wall are rigid.
The value of K /M1 has been selected such that the natural frequency of the
oscillator in the absence of fluid loading would equal the lowest nonzero nat-
ural frequency of the cavity if all walls were rigid. The value of D is set to give
a 0.001 ratio of critical damping for the isolated oscillator. Both M1 and M2
are fifty times greater than the mass of the fluid within the cavity. Determine
the pressure at three points: x = 90, y = 37.5 mm; x = 180, y = 75 mm; and
x = 360/21/2 , y = 150/21/2 mm, as a function of frequency up to 25% greater
than the fourth natural frequency of the system. Also, determine the amplitude
of the plate displacement within this range.
y
K D
M1
M2 150 mm
F 120 mm
c
x
360 mm
Figure 1.
Significance
The analysis will demonstrate all of the operations required to implement Dow-
ell’s approximation without requiring that we formulate a complicated model of the
structure. Features of special interest are the way in which the discontinuous proper-
ties of the surface are handle and the influence of the structure on the overall response
characteristics.
Solution
The coordinate system we shall use is shown in the sketch. Each plate is stated
to only translate in the direction normal to its surface. “Translation” means that
all locations execute the same displacement in the z direction. The motion of the
plate at x = 0 is the excitation, and a pressure field that is independent of z may
be sustained because the sidewalls are rigid. This observation permits formulation
of a two-dimensional model, in which the pressure field depends only on the x and
y coordinates. Correspondingly, integrals over a surface will feature a width factor,
W = 0.3 m.
The interface of the horizontal plate and the fluid, which constitutes Se , occupies
0.24 < x < 0.36, y = 0.15 m. The total area of this portion of the surface is Se =
W (0.12) = 0.036 m2 . All points on this surface displace by the same amount w (t),
positive for downward movement (into the fluid). This corresponds to Eq. (10.4.81)
with a single generalized coordinate and a surface displacement function that is one,
that is,
w = χ1 q, χ1 = 1 for 0.24 < x < 0.36 (10.4.101)
The plate that is situated at x = 0 undergoes a specified motion, and the remainder
of the boundary is rigid. This means that allportions of the boundary excluding the
oscillator’s face constitute Sv , with v̂ = Re V0 eiωt on x = 0 and v̂ = 0 elsewhere.
Consequently, integrals containing v̂ will have a nonzero contribution from x = 0
only.
A rigid-cavity modal series was formulated in Example 10.8. A variant of the
single index sequencing scheme used there is somewhat simpler to formulate. Index
m is associated with cavity mode ( j, n), with n being the number for x dependence.
The terms associated with each cavity mode may be defined in a double loop over
j = 0, ..., J and n = 0, ..., N . The associated value of m is
m = n + 1 + j (N + 1) , j = 0, 1, ..., J, n = 0, 1, ..., N (1)
where the highest mode numbers in each direction are J and N . Correspondingly,
the rigid-cavity modal properties are
nπ y
jπx
R,m = Bm cos cos , m = 1, 2, ..., (J + 1) (N + 1)
L H 1/2
jπ 2 nπ 2
(2)
ω R,m =c +
L H
where H = 0.15 m and L = 0.36 m. The modes are normalized according to

Eq. (10.3.8), which gives
2
Bm = 1/2 1/2 (3)
1 + δ0, j 1 + δ0,n
There are no sources within the fluid, so the source coefficients only contain a
contribution from the imposed motion of the plate on the left. Therefore, the surface
integral in Eq. (10.4.91) for these coefficients extends only over that plate,

∂
m,1 = ρ0 c2 R,m (x̄s ) v̂(x̄s , t) dS
∂t

Sv H

= ρ0 c2 R,m (x̄s ) x=0 (ωVc cos (ωt))(W dy)
0

H nπ y
= ρ0 c2 W Bm cos (ωV0 cos (ωt)) dy = ρ0 c2 W H Bm ωV0 cos (ωt) δn,0
0 H
(4)
This tells us that only the modes that are constant in the y direction are directly
excited.
The matrix [] mathematically couples the modes and the displacement of the
upper plate. The coefficients obtained from Eq. (10.4.86) are

1 L
1,m = χ1 (x̄s ) R,m (x̄s ) y=H (W d x)
Se 2L/3

W L jπx
= Bm cos cos (nπ) d x
Se 2L/3 L
(5)
⎧
⎪ W L 2 jπ
⎪
⎨S Bm sin (−1)n+1 if j = 0
e jπ 3
=
⎪
⎪
⎩ W L B (−1)n if j = 0
3Se
Note that each element of [] and [] is computed within the loops over j and n.
The equations for the acoustical field are supplemented by those for the structure.
In the present case, a single differential equation of motion governs the displacement
variable q of the oscillator. The only external force acting on this system is the resul-
tant force exerted by the pressure field, which is obtained by integrating the pressure
over the face of the piston. Because a positive q has been defined to be positive for
downward displacement, a positive pressure represents a negative resultant force.
Thus, this force is

+1)
(J +1)(N L
Fpressure = − p (x̄s , t) dS = R,m (x̄s ) y=H pm (W dy) ,
m=1 2L/3
Se (6)
(J +1)(N
+1)
= −Se 1,m pm
m=1
The equation of motion for the oscillator is M1 q̈ + C q̇ + K q = Fpressure . Substitution

of Eq. (6) gives
(J +1)(N
+1)
M1 q̈ + C q̇ + K q + Se 1,m pm = 0 (7)
m=1
The mass is specified to be M1 = 50ρ0 V, and K is set by the requirement that

the isolated natural frequency (K /M1 )1/2 equals the fundamental nonzero natural
frequency of the rigid-cavity. This frequency belongs to either the (1, 0) or (0, 1)
mode. Because L > H , the former is the choice, so we set
K π 2
= c2
M1 L
The fundamental nonzero frequency of the cavity is 472 Hz. The other parameter is
D, which sets the ratio of critical damping to 0.001. This ratio is such that division
of Eq. (7) by M1 converts the coefficient of q̇ to D/M1 = 2 (ratio) (K /M1 )1/2 , so
D = 2 (0.001) (K M1 )1/2 . The values that result are

M1 = M2 = 0.972 kg, K = 8.57 106 N/m, D = 57.7 N-s/m
The general frequency-domain equations are described in Eq. (10.4.94). To adapt

them to the present analysis, we populate {P} with the Pm coefficients and replace the
structural displacement vector {W } with a single coefficient q̂, which is the complex
amplitude of q. Correspondingly, [] is a row vector formed from the coefficients
1, , [K ] is the spring stiffness K , and [M] is the piston mass M2 . The structural
damping term gives rise to a term iωC that adds to K − ω 2 M in the oscillator’s
equation of motion. The frequency-domain equations that result are
+ ,
V (ω R )2 − ω 2 [I ] −ρ0 c2 Se ω 2 []T {P} ˆ
= (8)
Se [] K + iωC − ω 2 M2 ξ {0}
The computational procedure begins by computing all quantities that do not

depend on the frequency. This is done with a pair of nested loops in which j ranges
from 0 to J , within which n ranges from 0 to N . Inside these loops, the vector index is

m = n + 1 + j (N + 1). The mth element of (ω R )2 , [], and Q̃ are evaluated.
Then, a loop over frequency formulates and solves Eq. (8) over a range of frequen-
cies. The column vector{P} of complex pressure amplitudes at each ω is saved as a
column of an rectangular matrix [Pall ], and the corresponding complex displacement
amplitude ζ is saved as an element of a vector.
One response that is requested is the pressure
at the midpoint of the cavity. Equa-
tion (10.4.71) indicates that at any point x f , y f the pressure is
(J +1)(N
+1) nπ y
jπx f f
P= Pm Bm cos cos (9)
m=1
L H
The upper frequency limit is specified to be 25% greater than the system’s fourth
nonzero natural frequency. We do not know what that frequency is, so let us assume
that the natural frequencies are close to those of the rigid cavity. The lower natural
frequencies correspond to n = 0, 1, or 2 and j = 0, 1, or 2. The values are tabulated
below.
n=0 n=1 n=2

j =0 0 7121 14 242
j =1 2967 7714 14 548
j =2 5934 9269 15 429
The fourth nonzero value belongs to the (1, 1) mode. Setting the cutoff to 25% above
that value gives 9643 rad/s. We will round it up to 1.6 kHz.
We must decide how many modes to include. One guideline comes from making
sure that the highest natural frequency of the model is well above the maximum
excitation frequency. This will not be a difficult criterion to meet. A more stringent
requirement is that the number of modes should be sufficient to recreate in a modal
series the impedance discontinuity of the upper wall, but that provides no definitive
guideline. It is reasonable to set the highest mode numbers J and N such that the
(0, N ) and (J, 0) modes have comparable natural frequencies and that these frequen-
cies are at least twice the highest excitation frequency. We shall try J = 10, which
leads to N = ceil(J (H/L)) = 5. Because of the vagueness of these guidelines, the
computed results were verified by reducing J to 5 and N to 3. Doing so resulted in
barely perceptible alterations of the graphed data.
The upper graph in Fig. 2 describes the pressure responses at the prescribed loca-
tions. The lower graph shows the corresponding plate displacement. The vertical
lines mark the rigid-cavity natural frequencies. The peaks that occur are very close
to these values. This attribute is a consequence of the oscillator having a large mass. It
appears to the fluid to be a nearly rigid obstacle. Consequently, the resonances in the
frequency responses occur close to the natural frequencies of the rigid-cavity modes.
However, the pressure at the midpoint is very small when ω is close to the natural
frequency of the (1,0) rigid body mode. The pressure at the other locations seems to
be nullified at that frequency. The explanation for this feature lies in the concept of
a tuned vibration absorber, in which a low-mass oscillator is mounted on a system.
The natural frequency of the attachment is set equal to that of the unadorned system.
The result is that the previously unadorned part of the system does not respond at that
frequency. The piston-spring oscillator on the upper wall serves the same purpose,
except that its mass is much greater than the fluid.
x = L/ 2 , y = H/2
x = L/ 4 , y = H/4
x = L/2½, y = H/ 2½ 1130 1135
104
102
|P| (Pa)
100 (1,1) (3,0)
(0,0) (2,0) (0,1) (2,1)
(1,0)
10-2
10-5
|w| (m)
10-9
0 200 400 600 800 1000 1200 1400 1600

Frequency (Hz)
Figure 2.
It still remains to explain why some locations do not show evidence of resonances
that are seen at other locations. This behavior is a corollary of the proximity of
resonances to the rigid-walled behavior. Consequently, at each resonant frequency
ωm ≈ (ω R )m , the pressure coefficient Pm for the resonant mode is much larger than the
other coefficients. When the rigid-cavity modal series is evaluated at this frequency,
the mth mode will make the dominant contribution. However, if the corresponding
mode function R,m is small at a specific location, then the pressure there will be
much smaller than it is elsewhere. In the case of the midpoint, where x = L/2 and y =
H/2, we have R,m = B cos ( jπ/2) cos (nπ/2). Thus, if a resonance corresponds
to a cavity mode for which j or n is odd, then that resonance will not be observed at
the midpoint. Because x = 360/21/2 and y = 150/21/2 are irrational numbers, none
of the rigid-cavity mode functions are zero at that point. Consequently, every natural
frequency is manifested in a peak of the pressure at that point.
Another interesting feature is the narrow peak and adjacent null that occurs at some
frequencies. These all correspond to modes for which n = 1. The explanation lies in
the source coefficients Q̂ , which represent the excitation. According to Eq. (4), these
values are nonzero only if n = 0. Thus, only the modes that correspond to n = 0 are
excited directly. If the resonance is of a cavity mode for which n = 0, the excitation
is much weaker than it is for n = 0 modes. The drastic change from a maximum to
near-null results from the fact that crossing a resonance results in 180◦ phase shift
for the P coefficients that are resonant, so instead of adding to the contribution of
the nonresonant coefficients, they subtract.
The last effect of note is the fact that the oscillator displacement shows no evidence
of the (3, 0) resonance, even though the pressure does. This feature is a consequence
of the fact that the length of the piston in the x direction is L/3. Equation (5) indicates
that the 1, values are zero if corresponds to a mode for which j is an integer
multiple of 3. This is the mathematical consequence of the fact that 1, represents
the interaction between the piston and a mode and that interaction averages out over
the length of the piston if the pressure along it varies as cos (3πx/L), cos (6πx/L),
etc
Many of the features discussed in the preceding arise because the structural mass
M2 is large. A system in said to be a case of “light fluid loading” if the structure’s
mass is large compared to the fluid’s. Decreasing the mass of the upper plate leads
to system with “heavy fluid loading,” although it would be more accurate to say
that it is a system with low structural impedance. Dowell’s approximation might not
be applicable in this parameter range. Nevertheless, carrying out the computation
with the plate mass decreased by a factor of 50, so that it equals the mass of the
fluid within the cavity and enables us to see how the mass ratio affects the response.
Although the plate mass is decreased, the value of K /M2 is maintained. This has the
effect of increasing the importance of the term containing [] in the lower part of
Eq. (8). As a result, coupling of the Pm coefficients is stronger. Figure 3 shows the
pressure responses for this case. The oscillator now is a less substantial obstacle to
movement of the fluid, so the cavity’s behavior differs considerably from a cavity with
rigid walls. One consequence is that the resonant frequencies are noticeably different
from the natural frequencies of the rigid-cavity. A by-product is that resonances are
more likely to be seen at all points. Stronger coefficient coupling also accounts for
the increased occurrence of displacement nulls at various frequencies.
x = L/2 , y = H/2 x = L/ 4 , y = H/ 4 x = L/2½, y = H/2½
104
102
|P| (Pa)
100
(1,1) (3,0)
(0,0) (2,0) (0,1) (2,1)
(1,0)
10-2
10-5
|w| (m)
10-9
0 200 400 600 800 1000 1200 1400 1600

Frequency (Hz)
Figure 3.
An overview indicates that Dowell’s approximation is readily implemented to

describe fairly complicated systems. The most difficult operation in such cases is
likely to be determination of the rigid-cavity modes, which are a prerequisite to the
formulation. That was not aan issue for the box cavity, nor would it be for any of the
rectangular, cylindrical, or spherical cavities we have studied.
Exercise 10.1 An exponential horn whose length is L is closed at its narrow end and
open to the atmosphere at its wide end. The cross section is sufficiently small that
only plane waves can propagate within it. Derive the characteristic equation whose
roots are the horn’s natural frequencies. Then, evaluate these frequencies for a horn
whose length is L = π/k with b = 0.08/L. Compare these values to the natural
frequencies for plane waves in a uniform waveguide whose length is L with one end
open and the other closed.
Exercise 10.2 The sketch depicts a two-dimensional model of a box filled with a
liquid. The sides are rigid, and the top is open to the atmosphere. The bottom is bonded
to the ground, so any vertical motion of the ground is replicated by the bottom. In the
situation of interest, vibration of a nearby machine induces a vertical displacement
w = W cos (2πx/L) sin (ωt). Derive an expression for the pressure at any location
within the tank. Use that expression to evaluate the vertical velocity v y / (W ω) at the
midpoint of the free surface for frequencies in the range 0 < k L ≤ 4π. The aspect
ratio is H/L = 0.6.
H
w
Exercises 10.2, 10.3, and 10.4.
Exercise 10.3 Consider the situation in which a polymer sheet is placed on the top
of the liquid of the box in Exercise 10.2. The result is that the specific impedance at
this surface ζ = 0.8 − 3i. Evaluate the vertical velocity |vz | / (W ω) at the midpoint
of the free surface for frequencies in the range 0 < k L ≤ 4π. Then, compare this
response to that which would be obtained with a pressure-release condition, ζ = 0
The aspect ratio is H/L = 0.6.
Exercise 10.4 The system of interest is the one in Exercise 10.2, with the sole dif-
ference that the vertical wall at x = L is an intertance, with ζ = 0.25i. The aspect
ratio is H/L = 0.6. The excitation frequency is the lowest value at which the system
would resonate if x = L were rigid. Evaluate the vertical velocity |vz | / (W ω) at the
midpoint of the free surface.
Exercise 10.5 A container whose sides are otherwise rigid is excited by an oscil-
lation in which the vertical wall at x = 0 pivots about its horizontal centerline. The
result is that its normal velocity is vx = v0 (y/H − 0.5) cos (ωt). The dimensions of
the container are L, H , and W in the x, y, and z directions, respectively. Derive an
expression that describes the pressure field within the container.
Exercise 10.6 A rectangular water tank has sides that are composed of very thick
glass panels that may be taken to be rigid. The top, y = H , is open to the atmosphere,
so it is pressure-release. The bottom, y = 0, is a thin steel plate that flexes because
of ground vibration. The consequence is that the plate undergoes a vibration normal
to its surface, specifically, vz = v0 sin (πx/L). The dimensions of the bottom plate
are L × W . Derive an expression for the pressure p/ (ρ0 cv0 ) at an arbitrary location
within the tank.
Exercise 10.7 The normal velocity at the left side of a water tank, x = 0, is har-
monic with a parabolic dependence on y given by vz = v0 (y/b − y 2 /b2 ) sin(ωt) for
0 < y < b. The opposite side, x = a, is locally reacting, with local impedance
ζa = 0.6 − 0.3i. The sides z = 0 and z = d are rigid, as is the bottom, y = 0, whereas
y = b is pressure-release. (a) Derive an expression for the pressure at any location.
(b) The dimensions are a = 10 mm, b = 5 mm, and d = 3 mm. Evaluate the pressure
at the midpoint x = a/2, y = b/2, and z = d/2. The frequency range is ka ≤ 16.
Are resonances evident in this frequency response?
Exercise 10.8 An engineered material has been installed in the ceiling of a closed
room. It has the property of behaving like a continuous sheet of bricks, so that it
is locally reacting, with an impedance of 0.1k L. The wall at x = 0 undergoes a
translational oscillation in which vx = v0 sin (ωt). The other walls and the floor are
rigid. (a) Derive an expression for the pressure at an arbitrary location. (b) Identify the
natural frequencies by examining the expression for the pressure. (c) The proportions
of the room are L : H : W = 5 : 3 : 4. Evaluate the pressure | p| / (ρ0 cv0 ) at the
middle of the room as a function of frequency for ka < 12.
Exercise 10.9 The walls, floor, and ceiling of a reverberant room are essentially
rigid. Its height is H , and its horizontal dimensions are L × W . It is excited by
a loudspeaker that emits a constant amplitude at frequency ω. The loudspeaker is
situated at the middle of the ceiling. It is sufficiently
small thatit may be considered to
be a point source having volume velocity is Re Q̂ exp (iωt) . Derive an expression
for the pressure at the middle of the room.
Exercise 10.10 The sound pressure at the middle of the reverberant room in
Exercise 10.9 has been measured. At 1 kHz, it is 92 dB//20 µPa. Determine the vol-
ume velocity Q̂. The room dimensions are 3 m high and 5 m × 4 m horizontally.
Exercise 10.11 A small spherical object has been placed at the center of a spherical
cavity filled with water. The outer wall of the cavity executes a radially symmetric
vibration whose amplitude is fixed at 0.2 m/s, but the frequency is variable in the
range from 1 to 10 kHz. The surface of the spherical object is locally reacting with
ζ = 0.6 − 0.4i. The radii are a = 800 mm, b = 10 mm. The region between the
surfaces is filled with water. Determine the frequency at which the pressure on the
surface of the object is maximum. What are the pressure amplitudes on the inner and
outer surfaces at this frequency?
Exercise 10.12 A transducer is flush mounted on the inner surface of a spherical

cavity whose walls otherwise are rigid. The transducer is very small, so it may be
approximated as a point source whose volume velocity is Re( Q̂ exp (iωt)). Derive
an expression for the pressure at the center of the cavity.
Exercise 10.13 A cylindrical tank filled with oil has a piston transducer situated
concentrically at x = 0. The radius of the tank is a, and the radius of the piston is
a/2. Consequently, the resulting velocity at x = 0 is vx = v0 cos (ωt) if R < a/2,
vx = 0 if R > a/2. The other surfaces are rigid. Derive an expression for p/ (ρ0 cv0 )
at an arbitrary location. Then, use that expression to obtain an expression for the
acoustic acting on the wall at x = L. Evaluate | p/ (ρ0 cv0 )| at x = L, R = 0 and
force
|F| / πa 2 as functions of ka < 15. Parameters are ρ0 = 910 kg/m3 , c = 1460 m/s,
L = 4 m, and a = 0.25 m.
Exercise 10.14 Exercise 9.27 described the axial velocity at the near end of a cylin-
drical waveguide when the termination is a misaligned disk that rotates about the
centerline at angular speed . The velocity at a point whose cylindrical polar coordi-
nates are (R, θ, x = 0) is well described as vx = R sin (t − θ), where is the
angle between the normal to the disk and the centerline. Rather than taking the far
end to be nonreflecting, as was done in Exercise 9.27, suppose that the end at x = L
is open to the atmosphere, which corresponds to a pressure-release termination. The
cylindrical wall is rigid. Derive an expression for the pressure at an arbitrary location.
Exercise 10.15 The fluid contained in a cylinder whose length is L and diameter is
2a is excited by harmonic transverse vibration of the cylindrical wall. The velocity
amplitude is axisymmetric with a variation in the axial direction as a single lobe of
a sine, so that v R = v0 sin (πx/L) sin (ωt). The ends of the cavity are rigid. Derive
an expression for the pressure amplitude |P| / (ρ0 cv0 ) at an arbitrary location. Then,
evaluate this expression at the midpoint, x = L/2, R = 0, as a function of frequency
in the range ka < 12. The aspect ratio is L/a = 5.
Exercise 10.16 A cylindrical tank was filled with methyl alcohol (c = 1103 m/s,
ρ0 = 792 kg/m3 ). A small transducer was mounted concentrically at one end. The
far end was closed with mylar, which well fits a pressure-release model, whereas the
cylindrical wall was lined with a special insulating coating. The radius of the tank
was 3 m. The general analysis of the pressure field in a cavity indicates that in a nearly
resonant situation the spatial distribution is essentially proportional to the resonant
mode. With this in mind, the transducer was swept through a range of frequencies,
with the result that 154.3 Hz was identified as the lowest frequency at which the
interior field attains a peak amplitude. At this frequency, the pressure amplitude
along a transverse line was measured. The data that was collected is described by
the following table. The pressure waveforms at these locations were found to be
in-phase. (a) Determine the transverse wavenumber μ by fitting the measured pressure
to the analytical expression for the pressure field. (b) Determine the dependence of
the pressure on the axial distance measured from the driven end. (c) Determine the
specific impedance of the liner material. (d) Determine the length of the cylinder.
R/a 0.2 0.4 0.6 0.8

P (kPa) 310 268 203 125
Exercise 10.17 The sketch depicts a rectangular enclosure in which a small source
is situated at its midpoint. The source’s volume velocity is Q(t) = Re( Q̂ exp (iωt)),
where Q̂ is independent of the frequency ω. The enclosed fluid is oil (ρ0 = 910 kg/m3 ,
c = 1460 m/s). All sides of the enclosure are rigid, except that the closure at the left
end is a lightly tensioned membrane. The fluid on the other side of the membrane is air,
so the membrane effectively constitutes a pressure-release termination. Furthermore,
the left end is flush mounted with a very large rigid baffle, so the vibration of the
membrane causes sound to be radiated into the air. In other words, this system acts
like an unconventional type of loudspeaker. The primary property of a low-frequency

loudspeaker is its volume velocity. Determine the magnification factor Q̂ mem / Q̂
for the vibrating membrane’s volume velocity. The frequency range of interest is
zero to 2 kHz, and L = 0.5 m, W = 0.25 m.
fle
baf
membrane
Q W
L W
Exercises 10.17
Exercise 10.18 Solve Exercise 10.17 for the case where the cavity is a cylinder. As
described there, the closure at the left end is a pressure-release membrane and the
other surfaces are rigid. The length is L = 0.5 m, and the radius is a = 0.15 m.
Exercise 10.19 The sketch shows two oppositely phased line sources arranged hor-
izontally relative to the middle of the water tank. The sides and bottom are essentially
rigid. Consider the situation where the line sources are harmonic, with volume veloc-
ities per unit length ± Re( Q̂ exp (iωt)). Derive an expression for the pressure field
when the horizontal separation s is finite. (b) Take the limit as s → 0 with Q̂s held
constant at D and thereby obtain the pressure field for a dipole line source at the
center. (c) Evaluate the pressure distribution (amplitude and phase) along the bot-
tom when ω = 0.95ω1 and ω = (ω1 + ω2 )/2, where ω1 and ω2 are the two smallest
natural frequencies of modes that are excited by the dipole line source. The depth to
length ratio is H/L = 0.4.
H -Q +Q
H/2
L/2
L
Exercises 10.19
Exercise 10.20 A harmonic point source whose volume velocity is Re(Q 0 exp (iωt))
is situated at the midpoint of a cylindrical cavity. The ends at x = 0 and x = L are
pressure-release, whereas the cylindrical wall, R = a, is rigid. Derive an expression
for the pressure amplitude
at an arbitrary
point on the cylinder’s axis. Evaluate the
pressure amplitude Pa 2 / (ρ0 cQ 0 ) as a function of distance along this axis in the
case where L = 2a. Perform this evaluation for cases where the frequency is half
the cavity’s fundamental natural frequency and 10% greater than that frequency.
Exercise 10.21 The sketch shows a cylinder whose cross section is one-third of a
full circle. All surfaces are rigid. Derive expressions for the natural frequencies and
mode shapes of the fluid contained within this cavity.
120
L
a
Exercises 10.21.
Exercise 10.22 The wall of a spherical cavity is rigid. The volume of the fluid region
is the same volume as the rectangular cavity in Example 10.7. Evaluate all natural
frequencies of axisymmetric modes that are below 3 kHz. Graph the data by plotting
the frequencies at a fixed polar harmonic against the root number.
Exercise 10.23 The sketch shows a bowl that is filled with water. The hemispherical
wall is rigid, whereas the surface is free. Determine the axisymmetric mode functions
of the water contained in the bowl.
a
x y
z
Exercises 10.23
Exercise 10.24 The system of interest is a rigid-walled sphere whose radius is a. A

point source whose volume velocity is Re( Q̂ exp (iωt)) is situated at the center. The
pressure field may be described in terms of a series of the cavity’s natural modes.
An alternative analysis is suggested in Sect. 6.4.2. According to it, the field may be
considered to be the superposition of the field generated by the source in free space
and a field whose properties are selected to satisfy the boundary conditions, which
are violated by the free-space field. In the present case this condition is that the
radial particle velocity should be zero at r = a. The task here is to formulate both
solutions and then compare them by evaluating the dependence of the pressure on the
radial distance from the center when ka = 12. Part of the comparison should be an
assessment of convergence of the modal series as the number of modes is increased.
Exercise 10.25 A harmonic point source whose volume velocity is Re(Q 0 exp(iωt))
is situated at radial distance r = b from the center of a spherical cavity whose radius
is a. In order to exploit axisymmetry properties, the radial line through the source
is defined to be ψ = π. The wall of the cavity is rigid. Derive an expression for the
pressure amplitude due the point source. Evaluate
that
expression
to determine the
2
radial dependence of the pressure amplitude Pa / ρ0 c Q̂ along the radial lines
ψ = 0 and ψ = π/2. Parameters for these evaluations are b = 2a/3 and ka = 5.
Exercise 10.26 An exponential horn is closed at the narrow end and open to the
atmosphere at the wide end. Use a series of one, two, and three Ritz basis functions
to estimate the natural frequencies kn L in the case where the exponential growth
factor is β = 0.7. Compare the results to those obtained by solving the Webster horn
equation. It may be assumed that the complex pressure amplitude depends only on
the axial position x.
Exercise 10.27 The apex angle (side to side) of a cone is 10◦ . The cone is open at
its a large end, and its wall is rigid. Use a series of one to four basis functions Ritz
basis to the natural frequencies kn L to determine the natural frequencies of modes
whose amplitude depends only on the radial distance from the apex.
its a large end, and its wall is rigid. The task here is to employ spherical coordinates
r, ψ, θ centered at the apex to formulate the Rayleigh–Ritz equations. (a) Identify
a set of basis functions that are suitable for identifying the natural frequencies of
modes that are axisymmetric relative to the cone’s axis, but vary in both the radial
and polar direction. (b) Use a series of six of the Ritz basis in Part (a) to identify the
nondimensional natural frequencies kn L.
its large end, and its wall is rigid. The task here is to employ cylindrical coordinates
R, θ, z to formulate the Rayleigh–Ritz equations. The z-axis should coincide with
the cone’s axis, and the origin should be at the cone’s apex. (a) Identify a set of basis
functions that are suitable for identifying the natural frequencies of modes that are
axisymmetric relative to the cone’s axis, but vary in both the radial and polar direction.
(b) Use a series of six of the Ritz basis in Part (a) to identify the nondimensional
natural frequencies kn L.
Exercise 10.30 The cavity in the sketch is a sector of the region between concentric
cylinders. It is filled with water. All walls are rigid. Use the Rayleigh–Ritz method
to estimate the cavity’s natural frequencies. Carry out the analysis using from one to
eight basis functions.
1 mm
35o
2 mm
3 mm
Exercises 10.30
Exercise 10.31 Suppose that rather than being the interface with an elastic struc-
ture, Se in the development of Dowell’s approximation is the interface with a locally
reacting material whose impedance is position dependent, Z (x̄s ). What are the cor-
responding differential equations governing the coefficients of the pressure series?
Exercise 10.32 A piston of mass M, which closes the waveguide in the sketch,
is restrained by a spring whose stiffness is K . The termination at the other end
is rigid. A harmonically varying force F is applied to the piston. Use Dowell’s

approximation to determine the pressure amplitude on the face of the piston as a
function of the nondimensional frequency k L < 20. Parameters for the evaluation
are M/(ρ0 AL) = 5 and (K /M)1/2 = 2πc/L. Also determine the lowest five natural
frequencies of the system.
F
M 0, c A
L
Exercises 10.32
Exercise 10.33 The sketch depicts a closed tank filled with nitrogen, ρ0 =
1.165 kg/m3 , c = 354 m/s. A plate that fully closes the right end is supported by
a spring. The mass per unit surface area of the plate is 25 kg/m2 and the spring stiff-
ness is 500 MN/m. The left end is driven by a plate that extends over half the height
of the tank. Both plates translate as rigid bodies, and both extend horizontally across
the full width of the waveguide. All other surfaces are rigid. Consequently, the sys-
tem is two-dimensional in the plane of the sketch. The cavity is excited by harmonic
vibration of the left plate at 1.6 kHz, such that v = 20 sin (ωt) mm/s. Determine the
pressure distribution along the bottom and the amplitude of the displacement of the
plate at the right end.
K
200 mm 0, c
v
M
100 mm 400 mm
Exercises 10.33.
Exercise 10.34 The region enclosed by the cylinder is filled with an ideal fluid. The
left end is terminated by a concentric piston whose radius is a/2 and whose mass
is M. The remainder of the left end is rigid. Translational displacement w of the
piston is resisted by spring K . The cavity is excited by a piston at the right end
whose radius is a. As a result, the axial velocity at this end is a specified function
v L (t) for 0 ≤ R ≤ a. (a) Use Dowell’s approximation to derive a set of differential
equations whose solution would be the pressure coefficients for the acoustic field
and the displacement w of the left piston. (b) The parameters are M/(ρ0 AL) =
20, L/a = 3, and (K /M)1/2 = 2πc/L. The imposed vibration at the right end is
v L = v0 cos (ωt). Determine and graph the amplitude of the displacement w and the
pressure at the center of the right end as a function of frequency for 0 < ω < 6πc/L.
(c) For the parameters in Part (b), determine the first eight natural frequencies of the
axisymmetric modes.
w (t)
a
M
K L
Exercises 10.34
Chapter 11
Geometrical Acoustics
Our explorations thus far have analyzed sound fields from a global perspective, in
which field equations and boundary conditions for the entire system were satisfied
concurrently. The concepts in this chapter take a localized view to the question of
how a signal is modified as it propagates. It constructs rays, which are the paths
along which a specific phase propagates through space, and wavefronts, which are
surfaces of constant phase at a specific instant. Some aspects of the formulation are
an application of differential geometry, so the method is referred to as geometrical
acoustics. Because ray paths are the primary property that is derived, an alternative
name for the formulation is ray acoustics. The concept is limited to high-frequency
signals, but it is quite general otherwise.
A common application of geometrical acoustics is analysis of propagation through
a heterogeneous fluid, in which the ambient density ρ0 and pressure p0 are functions
of position. However, the ambient properties shall be considered to be independent
of time, on the basis that they vary much more slowly than the time scale for the
acoustic signal. Analysis of propagation through such a heterogeneous medium is a
core task for environmental acoustic studies of the atmosphere and the ocean. The
literature on geometrical acoustics and its application to propagation in the ocean
and atmosphere is rich. The objective here is to introduce the fundamental concepts.
The texts written by Brekhovskikh and Lysanov,1 Pierce,2 and Jensen et al.3 are
recommended starting points for further studies.
In order to emphasize development of the basic concepts, we shall add another
restriction to the systems we shall consider. Situations where the fluid is not quiescent
1 L.M. Brekohovskikh and Yu. P. Lysanov, Fundamentals of Ocean Acoustics, Springer-Verlag,

1982.
2 A.D. Pierce, Acoustics: An Introduction to Its Physical Principles and Applications, Acoustical
Society of Amer (1989).

3 F.B. Jensen, W.A. Kuperman, M.B. Porter, and H. Schmidt, Computational Ocean Acoustics, 2nd
ed (2011).
DOI 10.1007/978-3-319-56847-8_11
406 11 Geometrical Acoustics
in its ambient state are excluded. Winds in the atmosphere might be a reasonably
large fraction of the speed of sound. Thus, propagation of sound in the atmosphere is
not covered here. In a strict sense, the ocean also is not at rest, but the speed of ocean
currents is much less than the speed of sound. Although flow effects are omitted from
the present development, the principles and techniques that are derived provide the
foundation for study of sound propagation in the atmosphere.
11.1 Basic Considerations: Wavefronts and Rays
We begin with the general representation of a harmonic signal that depends arbi-
trarily in position, which is p (x̄, t) = Re F (x̄) eiωt . The polar representation of
the complex amplitude may be written as F (x̄) = P (x̄) exp(−i (x̄)), where P (x̄)
and (x̄) are real. The geometrical acoustics approximation defines the nature of
(x̄). It considers any signal to be locally planar, with ē (x̄) being the propaga-
tion direction and c (x̄) being the local sound speed. Such a signal corresponds to
(x̄) = x̄ · ē (x̄) /c (x̄). (No reference time lag is incorporated because we can define
t = 0, however, we wish to obtain a pure sine or pure cosine, or any other phase.)
Correspondingly, the geometrical acoustics ansatz for the pressure field is

p (x̄, t) = P (x̄) Re eiω(t−x̄·ē(x̄)/c(x̄)) (11.1.1)
The fact that the amplitude function P (x̄) is real will be crucial to later developments.
To identify how the preceding description is related to wavefronts, we begin with
the fundamental definition that a wavefront is a surface along which the signal at
the same instant has the same phase. A surface that is fixed in space is the locus of
points at which a function of the position coordinates has a constant value. A moving
surface may be obtained by replacing the constant value by a function of time. Thus,
an equation whose form is f (x̄) = t defines a surface that moves through space.
The phase θ of the signal in Eq. (11.1.1) is the coefficient of i in the complex
exponential, so we have
θ x̄ · ē (x̄)
=t− (11.1.2)
ω c (x̄)
If we fix t, then the locus of points on a wavefront of constant specified θ may be

found by solving the preceding for all possible x̄. Thus, this relation is the equation
defining the wavefront of constant phase θ. It is useful for further developments to
collect the position-dependent terms in a single function τ (x̄), whose dimensionality
is time,
x̄ · ē (x̄)
τ (x̄) ≡ (11.1.3)
c (x̄)
The description of a wavefront provided by Eq. (11.1.2) actually has two interpre-
tations. The values of x̄ on a wavefront, and therefore wavefront’s shape and overall
11.1 Basic Considerations: Wavefronts and Rays 407
position, depend of the value of t − θ/ω. If we hold the phase at a constant θ and
vary t, we will see the wavefront flow through space as time evolves. Alternatively,
we can fix t and construct a family of wavefronts at that instant corresponding to
different values of the θ. In other words, wavefronts may be considered to be a set
of surfaces at a specific instant or a single surface that moves through space. Both
views are equivalent, because only the value of t − θ/ω is important. This duality
tells us that everything we need to know may be found by following the wavefront
corresponding to zero phase. Therefore, we take the wavefront equation to be
τ (x̄) = t (11.1.4)
After we have described the zero-phase wavefront at an instant t, the wavefront along
which the phase is θ = 0 will occupy the same location at the instant t = t + θ/ω.4
The description on Eq. (11.1.1) is quite general. The geometrical acoustic approx-
imation is based on a fundamental assumption regarding spatial dependence of the
amplitude and phase. For a simple plane wave, the phase is θ = ω x̄ · ē/c = kx so
that ∇θ = k ē = (2π/λ) ē. In view of the definition of θ in Eq. (11.1.2), we may
consider ∇θ to be proportional to the reciprocal of a local wavelength. Thus, varia-
tion of the phase over a small spatial scale implies that ∇θ is relatively large. With
this observation in mind, let us differentiate Eq. (11.1.1) to determine the pressure
gradient,
∇p ≡ [∇P − iP∇θ] Re ei(ωt−θ(x̄)) (11.1.5)
An overview of the various signals we have considered indicates that in many cases,
the wavelength scale is much smaller than the spatial scale over which the pressure
amplitude varies. A geometric acoustics analysis is based on the a priori assumption
of this behavior. More definitively, it is required that |∇P| |P∇θ|. When this
condition is met, we say that the amplitude P (x̄) is a slowly varying function of
position. A further assumption comes from differentiation of Eq. (11.1.2), which
gives
x̄ · ē (x̄) ωē ∇c
∇θ = −ω∇ =− +θ (11.1.6)
c (x̄) c (x̄) c (x̄)
Unless the fluid is homogeneous, ∇c is not zero. A geometrical acoustic analysis

assumes that the sound speed varies slowly relative to the local wavelength, in which
case ∇c ≈ 0̄. The combination of both assumptions leads to the geometrical acoustics
assumption of the pressure gradient as
4 This assertion that the set of wavefronts at a specific instant may be identified by following the
evolution of a specific wavefront is only valid in the context of linear acoustics. For example, it
assumes that the signal does not alter the properties of the fluid, even though the speed of sound
depends on the pressure. If nonlinear effects are included, wavefronts and rays might also depend on
the pressure and particle velocity associated with a specific phase. This condition will be encountered
in Sect. 13.4.2.

ωē i(ωt−θ(x̄))
∇p ≈ − Re i Pe (11.1.7)
c (x̄)
Another requirement, which will be relaxed later, is that the radius of curvature of
the wavefront should be large compared to the wavelength. When both conditions are
met, the signal may be considered to be locally planar. Conversely, ray theory might
require correction, or it might not be valid at all, if either property is not obtained in
some region.
The extension of Eq. (11.1.1) to an arbitrary signal is

p (x̄, t) = F x̄, t , t = t − τ (x̄) (11.1.8)
The essence of the geometrical acoustics approximation is that the explicit depen-
dence of F on x̄ is taken to be very gradual relative to the manner in which τ (x̄)
varies spatially. This leads to the pressure gradient being represented by
∂F
∇p ≈ − ∇τ (11.1.9)
∂t
The nature of the wavefronts is defined by Eq. (11.1.4). An examination of
Eqs. (11.1.1) and (11.1.8) indicates that knowledge of the wavefronts is fundamental
to describing the pressure field. This observation should cause the reader to wonder
how rays enter the picture? That question is addressed by Fig. 11.1, which follows a
¯ The location
zero-phase wavefront. At instant t, a specific signal is located at x = ξ.
ξ¯ shifts in time by the amount necessary to keep it on the wavefront. The locus of
positions at which this signal is located is a curved line whose equation is x̄ = ξ¯ (t).
This is a parametric description of a ray, in which time is the parameter. Of course,
at this stage we do not know the functional dependence.
Fig. 11.1 Relationship of

rays and wavefronts in the
_ Ray
geometrical acoustics
Wavefront:(x) t + dt _ _ _
approximation. The normal ē
to the wavefront at any
x d
location is the direction in
which the signal at that _
location propagates Wavefront:(x)=t _
c()dt
_
x  _ _
e()
If we know the wavefront at any instant t and the sound speed at any location, there
is a graphical construction that will lead to wavefronts and rays at later times. We
consider a single step from t to t +dt. The planar nature of the propagation means that

the phase velocity at ξ¯ is c ξ¯ in the direction that is normal to the wavefront at ξ.
¯ The

¯
unit vector for this normal direction is ē ξ in Fig. 11.1. Given the current wavefront,
we can draw this normal, either manually or with a CAD program. In a time interval

dt, the signal propagates through distance c ξ¯ dt, which we lay out along ē ξ¯ . This

gives the next point on the ray as x̄ = ξ¯ (t + dt) = ξ¯ (t) + ē ξ¯ c ξ¯ dt. If we apply
this construction to many points on the wavefront at t, we will establish where the
wavefront is at t + dt. Thus, we have updated the location of the wavefront. The
process may be repeated as long as necessary. If we track the position of a particular
point on the initial waveform as time evolves, we will obtain the ray path for the
signal that was at that initial point at that initial instant.
There are several reasons why this graphical procedure is only used conceptu-
ally. Most significantly, it is approximate because it can only be implemented for
a finite time step. In addition, any method we use to graphically identify ē ξ¯ will
inherently be approximate. There is no way to quantify either type of error. Beyond
considerations of accuracy, the procedure is tedious, because many time steps would
be required to identify the rays and wavefronts over a reasonably large interval.
As we progress through this chapter, we will encounter two approaches by which
geometrical acoustics analyses are implemented. The first, which is limited to fluids
whose properties depend on a single spatial coordinate, directly determines the rays.
The second approach is suitable for an arbitrary heterogeneity. It simultaneously
determines wavefronts and rays. Both formulations are used primarily as the basis
for computational algorithms that we will explore. However, before we can address
either formulation, it is necessary that we identify the governing field equation.
11.1.1 Field Equations for an Inhomogeneous Fluid
The ambient density ρ0 and pressure p0 are known function of position, and the fluid
is taken to be at rest in this state. We seek to determine the acoustic perturbations,
ρ (x̄, t) , p (x̄, t), and v̄ (x̄, t). (It is convenient to temporarily revert to the use of
a prime to denote increments of state variables from their ambient value.) A body

force acting on the fluid may depend on the density, so it is denoted as b̄ x̄, ρ0 + ρ .
The body force adds a term to the momentum equation, Eq. (4.1.7), but the general
continuity equation, Eq. (4.1.5), is unmodified.
A linearized set of equations is obtained by introducing the small signal approx-
imations,
ρ = ρ0 (x̄) + ρ (x̄, t) , p = p0 (x̄) + p (x̄, t) , v̄ = v̄ (x̄, t) (11.1.10)

in which ρ ρ0 , p p0 , and |v̄| is much less than the speed of sound at ambient
conditions. When we introduce this decomposition into the continuity equation and
drop all terms that are products of primed quantities, we find that
∂ρ
+ ∇ · ρ0 v̄ = 0 (11.1.11)
∂t
Substitution of the small-signal representation into the momentum equation gives

∂ v̄ ∇p0 + ∇p 1 ∂ b̄
=− + b̄ (x̄, ρ0 ) + ρ (11.1.12)
∂t ρ0 ρ0 ∂ρ
0
Note that we have used the smallness of ρ to replace b̄ by its first two terms in a
Taylor series relative to the ambient state. All primed variables are zero in the ambient
state, so the ambient pressure must satisfy
∇p0 = b̄ (x̄, ρ0 ) (11.1.13)
Hence, the linearized momentum equation is

∂ v̄ ∇p ∂ b̄ ρ
=− + (11.1.14)
∂t ρ0 ∂ρ ρ0
0
Three state variables appear in the continuity and momentum equations. The third
relation is the equation of state. The assumption that heat flow is negligible remains
valid. However, because the entropy now may be position-dependent, the pressure
must be taken to depend on both the density and entropy, p = p (ρ, S). Negligible
heat flow between particles corresponds to taking the entropy of a specific particle
to be constant as the particle moves through space. The material derivative describes
how a property of a moving particle changes as it moves. Thus, we seek a relation
between p and ρ based on dS/dt = 0. The total pressure is p = p0 + p , but
dp0 /dt = 0 because the ambient pressure is taken to vary only with position. Thus,
the material derivative of the pressure equation of state is

dp ∂p dρ ∂p dS ∂p ∂ρ
= + = + v̄ · ∇ρ (11.1.15)
dt ∂ρ dt ∂S dt ∂ρ ∂t
This expression does not invoke the small-signal approximation. To do so we intro-

duce ρ = ρ0 (x̄) + ρ and drop higher order terms. Furthermore, the small-signal
approximation indicates that dp /dt and ∂p /∂t differ by second-order terms, which
are dropped. Thus, the equation of state for a heterogeneous fluid is a linear differ-
ential equation,

∂p ∂ρ ∂p
= c2 + v̄ · ∇ρ0 , c =
2
(11.1.16)
∂t ∂t ∂ρ 0
Although this definition of c is the same as it is for a homogeneous fluid, c now is a

function of position.
We have derived three differential equations that govern p , ρ , and v̄ , but we
seek a single differential equation for p . Toward that end, we substitute ∂ρ /∂t from
Eq. (11.1.11) into the equation of state, which leads to
∂p
= −ρ0 c2 ∇ · v̄ (11.1.17)
∂t
Next, we take the time derivative of this relation. Substitution of ∂ v̄/∂t from
Eq. (11.1.14) removes the particle velocity. The result is

1 ∂ 2 p ∂ v̄ ∇p ∂ b̄ ρ
= −∇ · =∇· −
ρ0 c2 ∂t 2 ∂t ρ0 ∂ρ ρ0 (11.1.18)
0
A rearrangement of terms yields

∇p 1 ∂ 2 p ∂ b̄ ρ
ρ0 ∇ · − 2 2 = ρ0 ∇ · (11.1.19)
ρ0 c ∂t ∂ρ ρ0
0
If the term in the right side were not present, this would be a field equation for p .
Indeed, most references do not list this term because they ignore the body force at
the outset. To see why it is allowable to do so, consider a plane wave propagating in
direction ē in theocean. The body force per unit volume is gravitational attraction,
so b̄ = ρgēz and ∂ b̄/∂ρ 0 = gēz . Let us use the relations for a homogeneous fluid
to compare the two terms whose divergence is taken in the above. Thus, we set
p = Re P̂ exp (iωt − ik ē · x̄) and ρ ≈ p /c2 . This leads to estimates of the two

terms as
∇p k p ∂ b̄ ρ
g p

ρ ≈ ρ , ∂ρ ≈
ρ0 ρ0 c2
(11.1.20)
0 0
0
The ratio of the second term to the first is g/ (ωc), which is extremely small even
at ω = 1 rad/s, unless the fluid is remarkably slow. For this reason, we shall proceed
without considering the body force effect. (It might not be allowable to do in some
situations, such as a strong magnetic field confining a plasma).
In the case of a gas, notably the atmosphere, dropping the body force term in
Eq. (11.1.19) is the sole fundamental simplification of the field equation. The equation
for a heterogeneous gas therefore is
1 1 ∂ 2 p
∇ 2 p − ∇ρ0 · ∇p − =0 (11.1.21)
ρ0 (x̄) c(x̄)2 ∂t 2
If the density changes gradually, the preceding field equation may be simplified
further by setting ∇ρ0 to zero. To determine the conditions for which this approx-
imation is reasonable, let us again make use of the description of a plane wave in
a homogeneous fluid to estimate the effect. The density gradient is approximated
as the difference ρ between the minimum and maximum values divided by the
distance over which these extreme values are observed. For ρ0 , we use the average
value. For a harmonic plane wave propagating in direction ē, we have ∇p = −ik ēp .
The maximum magnitude of ∇ρ0 · ∇p occurs if ē is parallel to the density gradient.
Thus, we have the estimates that
∇ρ0 ρ
∇ 2 p ≈ −k 2 p , · ∇p ≈ −ik p (11.1.22)
ρ0 ρav
With k replaced by 2π/λ, the order of magnitude of the ratio of these terms is
O ((ρ/ρav ) /(2π/λ)) .
Density variations in the atmosphere can be quite large, but that is not true for
water. To demonstrate this assertion consider standard formulas for the density of
water provided by Pierce.5 The value of ρ0 depends on the ambient pressure and
temperature, and salinity in the case of the ocean. The reference value for seawater is
ρ0 = 1027 kg/m3 at 10◦ C. Raising the ambient pressure to 2 MPa, which corresponds
to a 200 m depth, raises the density by less than 0.1%. A temperature increase to 30◦ C
decreases the density by less than 2%. Distilled water exhibits similar tendencies.
The density of seawater also depends on salinity, whose reference value is 35 parts
per thousand (ppt). It is very rare that the salinity varies by more than 2 ppt from
the reference, and such a fluctuation changes the density by 0.15%. Furthermore,
it is unusual that these extreme changes of ambient conditions occur over less than
a kilometer. In contrast, the wavelength at 10 Hz is approximately 150 m, and the
wavelength decreases with increasing frequency. In other words, 2π/λ will be large.
Thus, neglecting the contribution of the density gradient to the field equation is well
warranted if the medium is a large body of water or most other liquids. One exception
occurs at the interface between distinct layers, which was studied in Chap. 5. The
geometrical acoustic formulation addresses discontinuous changes of density and
sound speed as separate considerations.
When the simplifications that ignore the effects of a body force and of a spatially
varying density are invoked, what remains of the field equation is
1 ∂ 2 p
∇ 2 p − =0 (11.1.23)
c (x̄)2 ∂t 2
This equation resembles the linear wave equation for a homogeneous fluid. However,
the spatial dependence of c has a fundamental effect.
It should be apparent at this juncture that any analysis we pursued in previ-
ous chapters probably will not be valid for a heterogeneous fluid because the field
equation is a partial differential equation with variable coefficients. Indeed, any direct
mathematical analysis would be challenging. Nevertheless, it is important that we
have derived this equation because it validates considering the pressure to behave
5 A.D. Pierce, Acoustics, McGraw-Hill (1981), reprinted by ASA Press (1989) pp. 33–34.
locally like a plane wave. In addition, Eq. (11.1.21) will be crucial for the analysis
of the pressure amplitude in an fluid whose heterogeneity is an arbitrary function of
position.
11.1.2 Reflection and Refraction of Rays
The geometrical acoustic approximation that a ray behaves locally as though it were
planar has some important corollaries. Suppose a wave is incident on a locally react-
ing solid surface. The reflected wave may be identified by adapting the laws for
reflection of a plane wave at oblique incidence to a planar boundary. In keeping with
the restrictions of geometrical acoustics, the surface’s principal radii of curvature
are required to be much greater than a typical wavelength. When this restriction is
fulfilled, the surface may be replaced by the tangent plane at the point of incidence.
Other than the dependence of the normal n̄ on the point of incidence, the reflection
process is the same as that of planar waves from a flat surface. Thus, the normal n̄ and
the incident ray form a plane. The reflected ray lies in this plane at the same angle as
the angle of incidence. A consequence of this behavior is that rays that were parallel
prior to reflection will not be parallel after reflection, as is illustrated by Fig. 11.2.
Fig. 11.2 Reflection of rays

from a curved surface
A A B
B
B
A
At very high frequencies, surfaces that fit a local impedance model tend to be
very stiff. For that reason, solid surfaces are typically considered to be rigid in the
geometrical acoustics approximation. Then, the amplitude of the reflected wave at
the point of incidence equals that of the incident wave. If the impedance Z is finite,
the reflection coefficient in Eq. (5.2.15) may be used,
Z − ρ0 c/ cos θ
PR = RPI , R = (11.1.24)
Z + ρ0 c/ cos θ
As is true for plane waves, a description using a reflection coefficient should be

limited to frequency-domain analyses, unless Z is real.
It is unusual that two fluids meet at a curved interface. A possibility is that two
fluids are separated by a highly compliant solid sheet. Like the case of reflection from
a solid surface, the reflection and transmission of an incident ray is defined relative
to the normal at the point of incidence. The process is illustrated in Fig. 11.3. The
reflected and refracted rays lie in the plane formed by the incident ray and the normal
at the point of incidence. The angle of reflection equals the angle of incidence. The
angle of transmission is governed by Snell’s law, which states that
sin θ2 sin θ1
= (11.1.25)
c2 c1
Fig. 11.3 Transmission and

reflection of a ray incident on c2 c1 PR
the curved interface of two
fluids 1
PT
1
2
PI
The amplitudes of the signal on the reflected and transmitted rays are taken to be
those derived for a plane wave. The reflection and transmission coefficients are as
given in Eq. (5.3.8),
Z2 − Z1 2Z2 ρj cj
R= , T= , Zj = (11.1.26)
Z2 + Z1 Z2 + Z1 cos θj
The angle of transmission is measured relative to the local normal. Consequently,

even if incident rays are parallel, the amplitudes of transmitted and reflected rays
will vary with position on the interface.
We saw in Sect. 5.3 that supercritical incidence occurs if cos θ1 > c1 /c2 , which
obviously can only occur if c2 > c1 . Supercritical incidence leads to a wave in the
receiving fluid that evanesces with increasing distance perpendicularly the interface.
In the case of a truly planar interface, such a wave propagates parallel to the surface.
In a strict sense such a wave does not fit the geometrical acoustic model.
EXAMPLE 11.1 A plane wave propagates parallel to the axis of symmetry

of the cap of a sphere whose radius is a. The wave is reflected from the inner
concave surface of the cap, which is rigid. Determine and graph the paths of
the reflected rays.
Significance
This system, which is the acoustical analog of the mirror in a reflecting optical
telescope, has an acoustic application as part of an extremely sensitive microphone.
The only mathematical tools required for the analysis are those of trigonometry, so
the basic aspects of ray paths will be easy to identify. The results will exhibit the
phenomenon of a caustic, at which rays are concentrated.
Solution
The system is axisymmetric, so the picture is the same for any slicing plane that
contains the axis of the spherical cap. The range of polar angles ψ ≥ 0 for one slice
is depicted in Fig. 1. The spherical radius of the cap is a. The sketch shows a ray at
an arbitrary distance R from the axis of symmetry. This defines the polar angle ψ
to the point where the ray is incident on the mirror according to ψ = sin−1 (R/a).
The normal direction n̄ coincides with the radial line. Because the angle of reflection
is ψ, the reflected ray, the radial line, and the axis of symmetry form an isosceles
triangle. The distance where the reflected ray intersects the axis of symmetry is
s = (a/2) / cos ψ inward from the center.
 _
n
a 
R
s
Figure 1.
A full picture requires consideration of many rays, so we shall consider a range

of R values. If the maximum ψ is too large, a reflected ray will go on to be incident
at a second point on the cap. This situation may be avoided if the cap is such that
the maximum polar angle is less than 45◦ . We will consider a cap whose maximum
polar angle is 30◦ .
Construction of the rays is assisted by defining a coordinate system whose x-axis
coincides with the axis of symmetry, with the origin at the center. We construct a set
of incident rays, spaced at equal increments of the offset distance. Because of the
axisymmetric nature of the system, it is sufficient to consider positive polar angles.
For each R, we draw an incident ray from (0, R) to (a cos ψ, R), and a reflected ray
from (a cos ψ, R) to (s, 0), where s is described above. In Fig. 2, the reflected ray are
extended beyond the axis of symmetry.
a
Figure 2.
Every cutting plane features the same picture. Furthermore, Fig. 2 only shows half
the rays in a cutting plane because there are additional rays for ψ < 0. The full set for
a cross section are depicted in Fig. 3. The reflected rays extend beyond the window
that is plotted. The collection of rays corresponding to all cutting planes represents
an axisymmetric distribution. Thus, the locus of straight rays at offset distance R is
a cylinder, and the locus of reflected rays corresponding to a specified R is a cone
whose apex is situated at distance s.
Figure 3.
There is a region near the z-axis where the rays are closely spaced. Our later
investigations will show that the amplitude grows in regions where the spacing of
rays decreases. Another interesting aspect is this region is delimited by a curve beyond
which there are no neighboring reflected rays. To better understand this feature Fig. 4
zooms in on the reflected rays in this region of concentration. The decreased spacing
between rays and the absence of rays beyond a bounding curve are produced by
nearly parallel rays intersecting. The minimum s from such an intersection to the
center of the spherical cap is the ray associated with R → 0. This corresponds to
ψ → 0, for which s → a/2. The bounding curve generated by the intersecting rays
is called a caustic, and s = a/2 marks the arrête, which is the term given to the
location where a caustic terminates.
a/2
Figure 4.
In some applications, it is desirable that all rays intersect in a very compact region.
Convergence at a point, which is the definition of a focus, can be attained with a mirror
that is the interior of a paraboloid of revolution. (This is the subject of Homework
Problem 11.1.) All rays for the case of a spherical cap nearly converge at a point if
the largest polar angle is relatively small. However, there is no size that leads to a
perfect focus. In optics, this is known as spherical aberration.
11.2 Propagation in a Vertically Stratified Medium
An important model in environmental acoustics considers the fluid to have properties

that vary with vertical elevation, which is designated as z. The ambient density,
pressure, and temperature are taken to depend solely on z, and the fluid is quiescent
in the ambient state. The analysis that follows parallels the manner in which a classic
textbook on underwater acoustics6 begins. The development has the attractive feature
of emphasizing the physical properties of rays. Modifications to account for flow
6 L. Brekohovskikh and Yu. Lysanov, Fundamentals of Ocean Acoustics, Springer-Verlag (1982).

in the ambient state, such as winds in the atmosphere, are best deferred until the
basic concepts of geometrical acoustics are understood. The text by Pierce7 offers
an excellent portal to this subject. For underwater acoustics flow is less of an issue
because the flow rates are lower than in the atmosphere, and the sound speed is much
higher.
11.2.1 Snell’s Law for Vertical Heterogeneity
We begin with the case where finite thickness layers are stacked vertically. Propa-
gation in the ocean typically is analyzed by measuring z downward with z = 0 at
the free surface. That is the convention used here. A ray starting in layer n behaves
like a plane wave, so it is transmitted into layer n + 1. The direction of the transmit-
ted ray is governed by Snell’s law. Consequently, the ray will remain in a constant
vertical plane. Whereas the treatment of transmission in Sect. 5.3 used the angle of
incidence, the developments that follow use the grazing angle χ of the ray, which is
measured from the horizontal to the ray. The arrangement is depicted in Fig. 11.4.
It will be necessary to distinguish between propagation that is downward (increas-
ing z) and upward (decreasing z). This can be done by letting χ be in the range
−π/2 ≤ χ ≤ π/2, with negative values corresponding to upward propagation.
Fig. 11.4 Definition of the x

grazing angle for
propagation in a vertically cn
n
stratified fluid n
n+1
n+1
cn+1
z
Let “I” denote the conditions at the location where the ray was generated. Then,
Snell’s law requires that
cn+1 cn cI
= =···= (11.2.1)
cos χn+1 cos χn cos χI
The situation where the sound speed c (z) is a continuous function corresponds to
layers having differentially small thickness. This condition is described by replacing
cn with c (z), and χn with χ (z) . Thus, the grazing angle at any depth is defined by
1 cI
cos χ = cos χI , μ = (11.2.2)
μ (z) c (z)
The quantity μ is the index of refraction. It captures the manner in which the depth
dependence affects any ray, whereas the cos χI factor is specific to one ray. It is
7 Pierce, ibid.
11.2 Propagation in a Vertically Stratified Medium 419
evident that the grazing angle changes continuously if c is a continuous function of

depth. This means that each ray path will be curved, so that χ measures the angle
between the tangent to this curve and the horizontal.
It is worth noting that no consideration has been given to a reflected ray. The
reflection coefficient depends on the difference of the ρc/ sin χ values on either side
of the interface. A differential difference of these value associated with a continuous
variation of ρ and c leads to a reflection coefficient that is infinitesimal.
The observation that χ defines the local tangent to the ray path is the basis for
identification of these paths. Consider Fig. 11.5, which depicts a differential arc d.
Its slope is dz/dx, so we have
dz cos χ dz 1
dx = = dz, d = = 1/2 dz
tan χ 1 − (cos χ) 2 1/2 sin χ 1 − (cos χ)2
(11.2.3)
Snell’s law and trigonometry lead to descriptions of both differential distances in

terms of cos χI ,
cos χI
dx = 1/2 dz
μ − (cos χI )2
2
μ (11.2.4)
d = 1/2
dz
μ2 − (cos χI )2
Fig. 11.5 Pythagorean x

theorem for the arclength
along a ray dx

dz
d
z
The first of Eq. (11.2.4) defines x as a function of z. Because μ is a function of z,

the relation x (z) may be found as a definite integral starting from the first location
(xI , zI ). At the location where the ray is generated, the sense of the ray might be
downward. In that case z > zI . Upward propagation leads to z < zI . In either case, x
increases monotonically. Both cases are captured by taking the absolute value of the
integral, so that

z cos χI

x = xI + dξ (11.2.5)
zI μ (ξ)2 − (cos χI )2 1/2
The second of Eq. (11.2.4) is useful because the local propagation speed in geo-
metrical acoustics is c (z). Thus, the time required for a signal to propagate distance
d along a ray is dt = d/c (z) ≡ μd/cI . The arclength is described by the second
of Eq. (11.2.4), so
μ (z) μ (z)2
dt = d = 1/2 dz (11.2.6)
cI cI μ2 − (cos χI )2
If a signal departs from the first location at tI , the time at which it arrives at another
point on a specified ray is

z μ (ξ)2

t = tI + dξ (11.2.7)
zI cI μ (ξ)2 − (cos χI )2 1/2
The absolute value of the integrand is used here because the elapsed time increases
monotonically, regardless of whether the propagation is upward or downward.
Evaluation of Eq. (11.2.5) for a range of z gives the ray path, and evaluation of
Eq. (11.2.7) gives the time at which the signal will arrive at a point on that path.
Propagation into a region where the sound speed decreases in the direction of propa-
gation leads to increasing values of |χ|. This is the situation depicted Fig. 11.6a and
d. There is no limit to the vertical propagation in these cases, other than incidence
at the bottom. The behavior of rays in cases where the sound speed increases in the
direction of propagation are depicted in Fig. 11.6b and c. There the grazing angle
progressively decreases.
Fig. 11.6 Influence of the x

gradient of the sound speed
on the curvature of a ray

dc 
dz
(a) (b)
z
 dc
dz

(c) (d)
A simple explanation of the tendency for a ray to bend may be gained by consid-
ering a wave that initially is planar, with a wavefront that lies in the vertical plane. If
the sound speed increases with depth, so that the signal propagates faster as the depth
increases, the wavefront will advance progressively farther as the depth increases.
The direction of propagation is perpendicular to the wavefront, therefore the ray
begins to point upward. The case where dc/dz < 0 leads to the opposite effect,
wherein the wavefront advances more as the depth decreases.
It might ultimately happen that the grazing angle is zero. The sound speed at
which this condition occurs will be denoted as c0 . Setting χ = 0 in Snell’s law gives
cI
c0 = (11.2.8)
cos χI
This value is dependent on the initial grazing angle. If a ray reaches a depth at which
c (z) = c0 for its associated χI , then it will attain a horizontal tangency, which is
the limit of its vertical propagation. This condition occurs only if there is a depth at
which c (z) equals the value of c0 for the ray. For instance, rays that are launched
vertically remain straight. After a ray attains this horizontal tangency, the vertical
sense of the propagation reverses, so the horizontal tangency is said to be a turning
point.
It might happen that there are turning points above and below the launch point.
In Fig. 11.7a c (zB ) and c (zD ) equal c (zA ) / cos χI . The ray is launched downward,
but it curves upward because dc/dz > 0. Beyond turning point B it continues to
curve upward until it reaches a depth at which dc/dz < 0. This results in downward
curvature. Upward propagation ceases at turning point D, and beyond that point the
propagation direction is downward. It follows that that this ray is confined to the
range zD ≤ z ≤ zB . Depending on the sound speed profile c (z) and the value of χI ,
other rays will be confined differently or not at all.
Fig. 11.7 Turning points (a) x

and the surface limit the D
zD
vertical propagation of rays
I I I
A
C E
zB
B
z
(b) x
D’
A C’ I I
I E’
zB’
z B’
An exceptional situation occurs if a turning point occurs at the depth where the
sound speed is a maximum. In the region above that depth, a ray curves upward,
whereas a ray in the region below that depth curves downward. At the turning point
the propagation direction is horizontal. If c is a maximum at that depth the tendency
for the horizontal ray to curve downward is equal to the tendency for it to curve
upward. The consequence is that the horizontal ray splits in two, with one part going
up and the other going down.
Another possible condition is that a ray arrives at the free surface or the horizontal
bottom of a channel. Such a ray is reflected from the boundary. Because of the locally
planar nature of the signal, the reflected ray obeys the law for reflection of plane waves
at a free surface. Thus, χ for the reflected ray is −χ for the incident one. This is the
situation in Fig. 11.7b. The launch angle χI is larger, so the lower turning point occurs
at a greater depth zB . Reflection at the surface will occur if there is no turning point
at a lesser depth than zI , or if the depth at which c (z) = cI / cos χI is negative.
An interesting feature of a ray path results from the observation that |χ| for a
specific ray depends only on z. Thus, the grazing angle at a given depth beyond a
turning point is the negative of the value at that depth prior to the turning point.
This means that a ray is symmetric with respect to a vertical line through a turning
point. This feature is exhibited in Fig. 11.7 which shows that the grazing angle at
the original depth zA is either χI or −χI , depending on whether the propagation is
downward or upward. This attribute also applies relative to a surface reflection point.
Furthermore, the situation at points E and E in Fig. 11.7 is the same as point A,
except that the horizontal range is different. Thus, the pattern beyond these points is
a periodic repetition of the pattern from point A.
It is useful to consider how to handle the integrals in Eqs. (11.2.5) and (11.2.7)
in view of the nature of the ray paths in Fig. 11.7. From Snell’s law, we know that
0 ≤ |cos χI | ≤ μ (z). From the starting location to the lower turning point, where
z = zB and x = xB , the integral’s upper limit progressively increases. Consequently,
the integral leads to a value of x that increases monotonically with increasing z.
Beyond the turning point, the upper limit decreases. The consequence is that both
integrals decrease from their values at the turning point. This cannot be correct
because it leads to x values beyond the turning point that are less than xB , and
elapsed times required for the signal to propagate beyond the turning point that are
less than the time to reach the turning point. This seeming paradox is actually a
consequence of ignoring the existence of a singularity. The condition for a turning
point is μ (z) = cos χI , so the integrands are singular at these points. To go beyond
this point, we must perform piecewise evaluation of the integrals. That is, if xtp , ztp
are the coordinates of a turning point, then we must perform the evaluations as

ztp cos χI z cos χI

x = xI + +
zI μ (ξ)2 − (cos χI )2 1/2 ztp μ (ξ)2 − (cos χI )2 1/2
dξ dξ

ztp μ (ξ)2 z μ (ξ)2

t = tI + dξ + dξ
zI cI μ (ξ)2 − (cos χI )2 1/2 ztp cI μ (ξ)2 − (cos χI )2 1/2
(11.2.9)
A better alternative is to recognize that the first term in the right side of these
expressions gives xtp and ttp . This allows us to write these equations as

z cos χI

x = xtp +
ztp μ (ξ)2 − (cos χI )2 1/2
dξ
(11.2.10)
z μ (ξ)2

t = ttp + dξ
ztp cI μ (ξ)2 − (cos χI )2 1/2
The same modifications are required to evaluate the ray path and elapsed time
beyond a reflection point. The integrand is not singular at that point. Thus, the
integration may be carried without special consideration up to the reflection point,

then using the values of x and t at the reflection point to proceed onwards. For exam-
ple, in Fig. 11.7b, we may use the above procedure to find x and t for any point on
segments A B and B C D . To go beyond the reflection point, and thereby find seg-
ment D E , we would start with xD and tD and integrate over 0 < z ≤ zE . Beyond
point E , periodic replication applies. The symmetry property of the waveform rel-
ative to turning points and waveforms could be used to reduce the integrations. In
any event, this piecewise method of handling slope reversals at turning points and
reflections is troublesome to incorporate into an autonomous computer program. An
alternative approach is developed in the next example. It evaluates the integrals in a
stepwise manner.
The curvature of rays is related to the gradient of the sound speed. To identify this
relation, we again consider a differential arc d, but now account for the change of
slope along it. Thus, at the upper end of the element in Fig. 11.8 the grazing angle
is χ, while at the lower end it is χ + dχ. The center of curvature C is the point at
which lines perpendicular to the tangents intersect. The radius of curvature R is the
distance from the center of curvature to a point on the arc at the specified z.
Fig. 11.8 Geometric

construction of the radius of
curvature R using the
properties of a ray
The angle between these perpendiculars is the difference of the grazing angles,
so the differential arclength is d = Rdχ. The grazing angle on a specific ray
may be considered to be a function of the arclength, so dχ = (∂χ/∂) d. Implicit
differentiation of Snell’s law, Eq. (11.2.2), gives

∂χ ∂ cos χI cos χI dc dz
− (sin χ) = c (z) = (11.2.11)
∂ ∂ cI cI dz d
We found from Fig. 11.5 that dz = (sin χ) d. Assembling these relations leads to
1 ∂χ cos χI dc
= =− (11.2.12)
R ∂ cI dz
Most treatments regard the radius of curvature as a positive number, so they take
the absolute value of the preceding relation. This is not done here because the sign
of R describes the direction in which the arc is curving. As shown in Fig. 11.6a
and b, increasing z leads to decreasing c. Therefore, the slope increases with depth,
dχ/dz > 0 and R > 0. In Fig. 11.6c and d, the value of c increases with increasing
z. In that case, dχ/dz > 0 and R > 0. A different view is that the center of curvature
lies below the ray path if R > 0, whereas it lies above the ray path if R < 0.
Fig. 11.9 Construction of circular ray paths for a fluid whose sound speed varies linearly with
depth. The reference depth z = 0 need not be situated in the fluid
Knowledge of the radius of curvature is most useful when the sound speed depends
linearly on z, so that dc/dz is constant. It is evident that R is constant in this case,
but more information is embedded in Eq. (11.2.12). Let us define z = 0 to be the
depth at which c = 0, so that c = βz. (It is not important that this reference depth
probably is outside the fluid.) Correspondingly, we have
cI c z
R=− =− =− (11.2.13)
(cos χI ) β (cos χ) β cos χ
The center of curvature is at distance R in the direction of the inward normal to

the ray. This distance is constructed in Fig. 11.9 for cases where dc/dz is positive,
β > 0, and negative, β < 0. In either case, the radius of curvature is the hypotenuse
of a right triangle whose vertical side is z. The means that the center of curvature is
situated at the virtual depth where the sound speed vanishes. This knowledge may be
used for a graphical construction. Given the launch location and the initial grazing
angle, a line perpendicular to the initial ray tangent is constructed. The intersection
of this line with the z = 0 reference elevation marks the fixed center of curvature.
The ray path is a circle of radius R centered on this point.
It has been suggested that a case of arbitrary sound speed dependence may be
analyzed by approximating it as a sequence of layers in which c (z) depends lin-
early on z. This is equivalent to replacing c (z) with a pointwise linear interpolation.
Mathematical software enables direct evaluation of the integrals in Eqs. (11.2.5) and
(11.2.7), so there is little reason to pursue an approximate procedure.
EXAMPLE 11.2 The sound speed profile in a certain region of the

ocean varies parabolically with depth according to c (z) = c (zmin ) +
[c (0) − c (zmin )] (z/zmin − 1)2 , with zmin = 200 m, c (zmin ) = 1460 m/s, and
c (0) = 1520 m/s. Consider two rays emitted by a source at a depth of 150 m,
both of which are launched toward the surface. One departs at a grazing angle
of 12◦ , whereas the other is such that it has a turning point at the surface.
Determine and plot the path and the travel time along each ray as a function
of the horizontal range x.
Significance
The numerical algorithm developed herein handles turning points without the
requirement of user intervention. The results will shed some light on the overall
significance of sound speed variation on propagation in the ocean.
Solution
The first task is to determine the range of z covered by each ray. For the first,
the initial grazing angle is χI = −12◦ and the sound speed at zI = 150 m is
cI = 1463.8 m/s. The corresponding sound speed for a turning point is c0 ≡
cI / cos χI = 1496.5 m/s. There are two depths at which c (z) = c0 : zB = 44.11 m and
zD = 355.87 m. The initial grazing angle is upward, opposite the case in Fig. 11.7a.
Thus, the ray first travels upward from zI to zB , then downward from zB to zD , then
upward from zD to zB , and so on. We only need to follow one cycle, after which we
may invoke the periodicity property. To determine the initial angle for the second
ray, we compute c at z = 0, which is 1520 m/s. The grazing angle at this depth is
zero in order to have a turning point. Snell’s law states that the initial angle should
be χI = cos−1 (c (0) /cI ) = −15.636◦ . The depth of the lower turning point corre-
sponding to these values is 400 m.
Determination of a path requires evaluation of Eqs. (11.2.5) and (11.2.7) at a
sequence of depths zn . The computational effort will be reduced greatly if the integrals
in Eqs. (11.2.5) and (11.2.7) are carried out in an incremental manner from one depth
to the next. If the values of xn−1 and tn−1 at a specific xn−1 have been determined,
and the values at the next point are given by

zn cos χI

xn = xn−1 + dξ
zn−1 μ (ξ)2 − (cos χI )2 1/2
(1)
zn μ (ξ)2

tn = tn−1 +
zn−1 cI μ (ξ)2 − (cos χI )2 1/2
dξ
In most cases, the integrals may be evaluated with standard software. Indeed, if
zn − zn−1 is extremely small, a crude strip rule would be adequate.
However, turning
points are such that at some depth xj it happens that μ xj = cos (χI ), so both
integrands are singular at such points. This condition may be addressed by using
Eq. (1) only if the value of z at a turning point does not fall in the interval between
zn−1 and zn . If it does, an alternative evaluation based on a series expansion of the
denominator will allow us to pass the turning point. To see how to proceed, consider
an upper turning point. Let J be the index of this point, so that zJ = zB . Because this
is an upper point, adjacent points are at a greater depth. Thus, zJ−1 = zJ+1 = zJ + ,
with 1. The integration variable is expressed as ξ = zB + ε. Series expansions

based on the smallness of ε in this interval give

dc
c (z) = c (zJ + ε) ≈ c (zJ ) + ε
dz J
−1/2
−1/2 2
cI
μ (ξ)2 − (cos χI )2 = − (cos χ I ) 2
c (zJ + ε)2
2 −1/2
cI 2ε dc
≈ 1 − − (cos χ I ) 2
c (zJ )2 c (zJ ) dz J
Because zJ is a turning point, it must be cI /c (zJ ) = cos χI . This observation leads to

⎡ ⎤1/2
−1/2 ⎢ c (zJ )3 ⎥
μ (ξ)2 − (cos χI )2 ≈⎢
⎣− ⎥ ⎦ ε−1/2 (11.2.14)
2 dc
2cI
dz J
Note that dc/dz is negative at the upper turning point, so the square root yields a real
value. Substitution of this approximation and dξ = dε into Eq. (1) leads to
⎡ ⎤1/2
⎢ c (zJ ) ⎥
xJ − xJ−1 = xJ+1 − xJ = (2)1/2 ⎢ ⎥
⎣ dc ⎦

dz
⎡ J ⎤
1/2 (2)
(2) ⎢ ⎢ c (zJ) ⎥
1/2 ⎥
tJ − tJ−1 = tJ+1 − tJ =
cI μ (zJ ) dc ⎦
⎣
dz
J
The values of xJ−1 and tJ−1 presumably have been determined by invoking Eq. (1).
The above expressions yield the corresponding values of xJ and tJ , then xJ+1 and tJ+1 .
These expressions also apply to the case of a lower turning point, for which (dc/dz)J
> 0. In that case, the points neighboring the turning point will be at a slightly smaller
depth, so that zj−1 = zJ+1 = zj − .
The actual ray path has two turning points. Figure 1 describes the scheme by which
the depth z is discretized. It is based on the recognition that a selected sampling
interval I between the launch point and the first turning point might not fit the
interval between the zI and the lower turning point. Therefore, two intervals are
used: 1 = (zI − zB ) /N1 for points at z ≤ zI and 2 = (zD − zI ) /N2 for points at
z ≥ zI . This leads to assignment of the depths according to zn = zI − (n − 1) 1
for n = 1 to N1 + 1; zn = zB + (n − N1 − 1) 1 for n = N1 + 2 to 2N1 + 1;
zn = zI + (n − 2N1 − 1) 2 for n = 2N1 + 2 to 2N1 + N2 + 1; and zn = zC −
(n − 2N1 − N2 − 1) 2 for n = 2N1 + N2 + 2 to 2N1 + 2N2 + 1.
B:N1+1
zB
N11
zI A:1 C:2N1+1 E:2N1+1+2N2
N22
zD
D:2N1+1+N2
Figure 1.
The initial values are x1 = 0 and t1 = 0 at x1 = 0. The values of xj and tj at regular

points are found by using a numerical integration routine to evaluate Eq. (1). A typical
step in MATLAB is x(j)=x(j-1)+abs(quadl(F_x, z(j-1), z(j))).
This scheme is used to evaluate points 2 to N1 , then the upper turning point is
passed by using Eq. (2) to evaluate the z and t values at points N1 + 1 and N1 + 2.
The next phase is evaluation of the regular points N1 + 3 to 2N1 + 1. From there,
the procedure is replicated, with Eq. (2) used to evaluate points 2N1 + N2 + 1 and
2N1 + N2 + 2. For points beyond 2N1 + 2N2 + 1, the data is replicated according to
xn = x2N1 +2N2 +1 + xn−2N1 −2N2 −1 and tn = t2N1 +2N2 +1 + tn−2N1 −2N2 −1 .
The ray path in Fig. 2 is constructed by plotting (zn , xn ) pairs such that z is dis-
played as a function of x. The first graph shows that the path is moderately curved,
with a depth variation of approximately 310 m over a range of more than 4 km. The
result for travel time along the ray appears to depend linearly on the range. This
property stems from the facts that the grazing angle is small along the entire path
and the phase velocity c (z) varies little. Consequently, the horizontal component of
the phase velocity which is c (z) cos(χ) is nearly constant.
0
Depth (m)
200
400
4
Time (s)
0
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
Horizontal range (km)
Figure 2.
The path appears to be sinusoidal. Whether this is the true nature of the path is
tested by plotting a sine curve relative to the average value of z, which is zav =
[max(z) + min(z)] /2. The fitted curves are z = zav + A sin [2π (x − x0 ) /L] and
t = max (t) b, where A = (zD − zB ) /2, L/2 is the horizontal distance between
adjacent locations at which z = zav , and x0 is the minimum range at which z = zav
with dz/dx > 0. This sine curve is indistinguishable from the computed path.
The ray corresponding to χI = −15.636◦ is described by Fig. 3. Launching a ray

at an upward or downward initial angle that exceeds this value will generate a ray
that reflects from the free surface.
0
Depth (m)
200
400
4
Time (s)
0
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
Horizontal range (km)
Figure 3.
11.2.2 Intensity and Focusing Factor
Thus far, we have not examined how the pressure magnitude is altered as a sig-
nal propagates. Although we applied Snell’s law for an interface between different
fluids to derive equations describing the ray paths, we cannot extend the analogy
by applying the reflection and transmission coefficients. The reason for this is that
the fluid properties vary continuously. Because ρc/ sin χ changes by an infinitesi-
mal amount across a layer of depth dz, the reflection coefficient is essentially zero
and the transmission coefficient is essentially one. Our need is to determine how
much the transmission differs from one, so we shall pursue a different approach. The
analysis examines the manner in which power flows along rays in order to obtain an
expression for intensity.
A common measurement in the ocean is mapping of the field from a small source.
Typically the source radiates axisymmetrically relative to the vertical, which means
that the field is independent of the circumferential angle of a cylindrical coordinate
system whose axial direction is vertical. This situation is depicted in Fig. 11.10. The
ray path in this picture is described by giving the transverse distance R as a function of
z. The fact that the properties are the same for any vertical plane containing the z-axis,
combined with the locally planar nature of the geometrical acoustics approximation,
requires that R depends on z in the same manner as x depends on z for two-dimensional
Cartesian coordinates. In addition, a ray path is altered if the depth z0 of the source
or the grazing angle at the source are altered. Thus, we may consider the ray path to
be defined functionally as R = R (z, z0 , χI ) .
The analysis of intensity is founded on the concept of a ray tube, which is the region
enclosed by a group of rays that leave a source at orientations that are infinitesimally
different. A typical tube appears in Fig. 11.10. The sides of this ray tube lie in vertical
Fig. 11.10 A ray tube

consisting of all rays 
R
emanating from a point at x
grazing angles between χI z0 dI
and χI + dχI , and between d
azimuthal angles between θ
and θ + dθ
z
dA Rd
dh
planes whose circumferential angle differ by dθ. The upper and lower bounding
surfaces correspond, respectively, to initial grazing angles χI and χI + dχI . The
transverse distance to a cross section dA of the tube is R. The cross section is a
rectangle whose sides are Rdθ horizontally and dh in the vertical plane.
The significance of a ray tube stems from the fact that the particle velocity, and
therefore the intensity, is parallel to a ray. Consequently, there is no power flow
across the sides, so the time-averaged power that flows across any cross section of
the ray tube must be constant. The particle velocity is parallel to each ray, so the
local intensity is perpendicular to the cross section at which it occurs. It follows that
the constant power in a ray tube is such that
|P|2
dP = dA is constant along a ray (11.2.15)
2ρ0 c
An equivalent way of stating this principle is that the time-averaged power that
flows across any two cross sections of the same ray tube must be the same. A corollary
is that the power radiated by a source into the tube must flow across the cross section
at the far end of the tube. Our objective is to describe this principle in terms of the
properties of the rays in order to obtain an expression for the intensity.
The cross section of a ray tube in the vicinity of a specified depth is depicted in
Fig. 11.11. Both ray paths in this Figure have been terminated at the same depth,
so the transverse distance to the end of the upper ray is R (z, z0 , χI ), while it is
R (z, z0 , χI + dχI ) for the lower ray.
Because power is the area integral of intensity, we seek an expression for the
cross-sectional area dA = (Rdθ) (dh). The side dh forms one side of a right triangle
whose hypotenuse is the line connecting the points in the bounding rays at depth z.
The horizontal distance between the ends is solely a consequence of the dependence
of R on the initial grazing angle. These angles are χI and χI + dχ for the rays in the
sketch,
from which it follows that the horizontal distance between points on the rays
is R z, z0, χI + dχ − R z, z0, χI . Therefore, we have
Fig. 11.11 Side view of a

ray tube generated by a
source at depth z0
z
dh

R (z,z  ,  d)
R (z,z  ,)

∂R
dh = R z, z0, χI + dχI − R z, z0, χI |sin χ| = dχI |sin χ| (11.2.16)
∂χI
Thus, the cross-sectional area of the ray tube at the output end is

∂R

dA = (Rdθ) dh = sin χ RdχI dθ (11.2.17)
∂χI
The normal to dA is parallel to the rays within the tube, and therefore parallel to
the particle velocity on that cross section. Hence, the time-averaged power flowing
across dA is
∂R
dP = Iav dA = Iav sin χ RdχI dθ (11.2.18)
∂χI
The preceding is the time-averaged power that passes the cross section of the ray
tube at depth z. This must be the same as the power that is input to the ray tube
by the source. To describe this quantity, we observe that heterogeneous effects are
significant only over distances that are many wavelengths. Thus, if we surround the
source with a small sphere of radius ε, the pressure on this sphere will be the same
as it would be if the fluid was homogeneous. If a sphere radiates omnidirectionally,
its power P is distributed evenly
2
over the surface area 4πε , so the intensity would
be a constant value, P/ 4πε . The radial intensity for a source that is axisymmetric
2
relative to the vertical axis is defined by multiplying this constant reference value by a
directivity factor DI . This factor depends on the polar angle, which is the complement
of the grazing angle, ψ = π/2 − χI . Thus, the intensity at radial distance ε from the
source is given by
P
I0 (ε, χI ) = DI (π/2 − χI ) (11.2.19)
4πε2
The power input dP0 to the ray tube is the product of I0 and the area of a patch of
the surrounding sphere, whose sides are εdχI in the vertical plane and (ε cos χI ) dθ
horizontally. Thus, we have
P
dP0 = I0 (ε, χI ) ε2 (cos χI ) dχI dθ = DI (π/2 − χI ) (cos χI ) dχI dθ
4π
(11.2.20)
Equating this quantity to the power flowing across an arbitrary cross section in
Eq. (11.2.18) yields

P DI (π/2 − χI ) cos χI
Iav = (11.2.21)
4π ∂R
R
sin χ
∂χ I
The same concept may be employed to examine the intensity when a field is
generated by some agent other than a small source. If we know the intensity I1 along
a cross section at one location, and we wish to determine the value at another cross
section, equality of the power flows requires that
dP = (Iav )2 dA2 = (Iav )1 dA1 (11.2.22)
When we use Eq. (11.2.18) to describe each term, we obtain

∂R

R1 sin χ1
∂χ
(Iav )2 = (Iav )1 I 1 (11.2.23)
∂R

R2 sin χ2
∂χI 2
If we know the ray path, we can evaluate all terms in this expression other than the
value at the input end. Thus, we can use this relation to monitor the intensity as the
signal propagates along the ray path provided that we have a starting value for the
intensity.
The focusing factor is used to compare the intensity to the hypothetical value that
would be obtained at a specified location if the same source radiated into a homo-
1/2
geneous fluid. The radial distance would be r = R2 + x 2 . Equation (11.2.19)
describes the intensity for a homogeneous fluid, so we may find the reference inten-
sity by replacing ε in that relation by r. Thus, the focusing factor is
Iav r 2 cos χI
= (11.2.24)
I0 (r, χI ) ∂R
R sin χ
∂χI
The condition where ∂R/∂χI = 0 marks a location at which the cross-sectional

area of a ray tube narrows to zero. The actual pressure must be finite, so this condition
marks a condition where ray theory requires some adjustment. Depending on the
specific nature of the signal, it might be that the singularity may be corrected by
using a higher order approximation within the context of a linear acoustic theory,
or it might be appropriate to incorporate a nonlinear correction. In any event the
pressure will be greatly enhanced in the vicinity of a location where ∂R/∂χI = 0.
Extreme narrowing of ray tubes is manifested in either of two ways. One is a
focus, like that encountered in an optical lens. A focus is marked by convergence
(a) (b)
Focal point Caustic
Fig. 11.12 Two situations in which geometric acoustics predicts a singularity of intensity:
a a focus, b a caustic
of a group of rays on a single point. According to geometrical acoustics, the rays

emerge from the focus continuously, unless the focus is at an interface between
media. Figure 11.12a displays a focus at which a cone of rays intersect. The quantity
∂R/∂χI is zero because there is no change in the transverse distance R for rays
associated with initial grazing angles that differ by an infinitesimal amount. This
is different from the situation where two rays associated with finitely different χI
values intersect. In that case, the pressure at the intersection is the (finite) sum of the
pressure associated with each ray.
Figure 11.12b illustrates a different singularity. In it, there is a curve in two dimen-
sions, or a surface in three dimensions, on which rays for infinitesimally different χI
arrive tangentially. We previously encountered this phenomenon in Example 11.1.
The line or surface that is the locus of points on which R does not change when χI
is changed infinitesimally is a caustic . When the fluid properties and the starting
location of a ray are specified, the transverse distance R to a point on a ray path
depends only on the depth and the initial grazing angle, that is, R = f (z, χI ). Hence,
the occurrence of a caustic in the axisymmetric field in a vertically stratified fluid is
marked by
∂f
R = f (z, χI ) and =0 (11.2.25)
∂χI
In principle, we could identify this locus by solving the second equation for z over
a range of χI values, then using the first equation to find the corresponding R values.
In practice, the ray path usually is determined numerically, so the function f will
not be available. A caustic is marked visually by a greatly reduced spacing between
adjacent rays, with all arriving rays being on one side of the caustic. The absence
of rays on the other side of the caustic is an artifice of the geometrical acoustics
approximation. The actual behavior is that the field evanesces very rapidly in the
direction normal to the caustic. Nevertheless, geometrical acoustics is quite useful
even here because it informs us that there is a region where the pressure is greatly
enhanced. Furthermore, Eq. (11.2.23) may be used to evaluate the intensity beyond
the caustic. (This extension is a topic in Sect. 11.3.2.)
1
Depth z (km)
1.48 1.54 0 20 40 60 80
c (km/s) Horizontal distance R (km)
Fig. 11.13 Ray paths for a point source in a channel modeled by Munk in the Journal of the
Acoustical Society
of America, vol.55, pp. 220–226 (1974). The dependence of sound speed on
depth is c = c1 1 + ε ξ + e−ζ − 1 where ξ = 2 (z − z1 ) /z1 . The coefficients are c1 = 1492 m/s,
ε = 0.0074, and z1 = 1300 m
A classic demonstration of the occurrence of caustics may be found in a paper

by Munk.8 The sound speed profile described in Fig. 11.13 was suggested there to
be a good prototype for studies of phenomena in the ocean at moderate latitudes.
The sound speed profile has a minimum at z1 = 1300 m. An omnidirectional source
is located at this depth, so that zI = z1 . Setting dc/dz = 0 at zI results in the ray
paths having an inflection point at the depth of the source. However, the sound speed
profile is not symmetric relative to this depth. Consequently, the appearance of a ray
path below zI does not mirror that appearance above.
The rays appearing in Fig. 11.13 correspond to initial grazing angles ranging from
14.40◦ upward to 14.40◦ downward. This is the range of angles for which rays are
not reflected from the surface. The rays described here extend to a greater range than
those computed by Munk, whose results were not shown in their entirety beyond a
range of 25 km.
The most evident feature of the ray paths is that there are regions where there are
none. These are shadow zones. Usually, one thinks of a shadow as the blocking effect
of a solid object. Here, a caustic obstructs the sound field. Regions of enhanced inten-
sity correspond to decreased spacing between adjacent rays. They appear as darkened
regions in the figure. The source is at R = 0, z = 1300 m, so the enhancement near
that location is not surprising. The next enhanced region occurs in the range from
10 to 20 km at depths between 400 m–1.3 km. This appears to be a caustic. Other
regions of interest are close to the source depth at ranges in the vicinity of 50 and
70 km. These seem to be somewhat like a focus, but they also seems to resemble a
caustic. Figure 11.14 zooms in on the regions where rays are concentrated.
8 W.Munk, “Sound channel in an exponentially stratified ocean, with application to SOFAR,”

Acoustical Society of America, Vol. 55 (1974) pp. 220–226.
0
0
0.2 0.5
Depth z (km)
0.4
1
0.6
0.8 1.5
1 2
1.2
2.5
10 15 20 25 40 50 60 70 80
Horizontal distance R (km)
Fig. 11.14 Zoomed views of regions of enhanced intensity in the sound channel studied by Munk
in the Journal of the Acoustical Society of America, vol. 55, pp. 220–226 (1974)
The caustic in the left figure is a boundary beyond which the pressure is zero (or
very small if the geometrical acoustics approximation is corrected). The low point
of this feature occurs at the depth of the source. The concentration of rays in the
vicinity of R = 47 km and R = 70 km in the right figure also occurs in the vicinity of
the source depth zI . The zoomed views show that these concentrations are not foci,
because only rays at nearly equal grazing angles intersect. This is a fundamental
aspect of caustics. Each has a distinct arrete,which is the minimum distance for the
caustic. The rays become more diffuse at the other end of each caustic. Additional
computations would show that caustics like these will occur even if zI is not the same
as depth z1 for minimum c (z). However, in that case no caustics would be observed
at zI or z1 .
EXAMPLE 11.3 A radially symmetric point source is situated at zI = 150

m. The sound speed dependence on depth is as described in Example 11.2.
Identify the regions where the pressure is enhanced. Then explain how this
data may be used to evaluate the focusing factor at a designated location.
Significance
In addition to showing how to map rays, some interesting phenomena, including
the role of reflections, will be explored.
Solution
We begin with an evaluation of the ray paths that remain below the surface. For
this, we follow the procedure developed in the previous example. Stated briefly, the
initial grazing angle is set in a range whose limits are the values extracted from
Eq. (11.2.8), specifically χI = cos−1 (cI /c (z = 0)) = ±15.363◦ . The turning point
depths min (z) and max (z) for each angle are computed, and the depth is discretized
into intervals of 1 for min (z) ≤ z ≤ zI and 2 for zI ≤ z ≤ max (z). Starting from
the source location, the transverse value of R at each depth up to the first turning
point is computed by applying the first of Eq. (1) in Example 11.2. The turning point
and the next are evaluated by applying Eq. (2). Then, Eq. (1) is used to evaluate
points up to the second turning point, followed by Eq. (2) for that point and the point
beyond. The final part of evaluation uses Eq. (1) to return to zI , after which the path
is periodically replicated. Figure 1 is the result of this computation.
0
100
Depth z (m)
200
300
400
1.45 1.55 0 1 2 3 4 5
c (km/s) Horizontal distance R (km)
Figure 1. Rays radiated by the source that do not reflect from the surface.
It can be seen that all rays have the same periodicity horizontally, and that they
all intersect at a common point. These intersections are foci. Both properties are a
consequence of c (z) being an even function relative to the depth zmin at which it is a
minimum. Another consequence of the symmetric nature of the sound speed profile
is that the depth of foci are either zI or max (z) − zI .
In addition to the occurrence of foci, there also appears to be large shadow zones.
However, Fig. 1 does not give a complete picture, because it does not account for rays
that reflect from the free surface. Such a condition is marked by one of the turning
points being above the surface, that is, z0 < 0. Thus, reflected rays in the present
case correspond to 15.363◦ < χI < 90◦ and −90◦ < χI < −15.363◦ . Rays that
reflect are computed by using Eq. (1) in Example 11.2 to evaluate the values of R
and x at the reflection point. If we let J equal to the index of the reflection point, then
the properties for the next point are set to place its properties symmetrically, so that
RJ+1 = RJ + Rj − RJ−1 .
The result of including rays that reflect is Fig. 2. It appears that only one shadow
zone remains. (Even that zone would disappear if there were a bottom at a finite
depth, in which case there would also be reflections at that boundary.) The ray paths
are far apart in the regions that were quiet in the previous figure. This tells us that
the intensity is low in those regions. There appears to be a caustic that touches the
surface at a range of 2.5 km formed from a group of rays for which χI > 0, and
another at a range of 2.8 km formed from a group for which χI < 0.
Depth (m) 500
1000
1500
2000
0 1 2 3 4 5
Horizontal distance (km)
Figure 2.
To examine these features more closely, Fig. 3 displays only paths of rays that
reflect from the surface. The details remain masked in the overview, but they can be
seen clearly in the zoomed views. The broad darkened region in the vicinity of the
source is due to nearly vertical rays that reflect from the surface. They return to the
depth of the source at a comparatively small horizontal distance, so they enhance
the already large pressure in the nearfield of a spherical source. In the zoomed view
of the horizontal range from 2 to 3.2 km, we see that there are indeed two caustics
formed from rays that are incident at the surface. We also see that reflection of the
rays associated with each caustic seems to result in each forming another caustic.
0
500
1000
1500
Depth (m)
2000
0 1 2 3 4 5
0 0
100
200
200
400
300
400 600
0 0.05 0.1 0.15 0.2 2 2.5 3
Horizontal distance (km)
Figure 3. Rays that reflect from the free surface when
the sound speed depends parabolically on
depth.
Graphs such as these give a qualitative picture, but the data also may be used to
calculate the focusing factor. The quantity in Eq. (11.2.24) that is not readily available
is ∂R/∂χI . Because z is held constant in this derivative, we could estimate it with a
finite difference if we had values of R corresponding to a discretization of z that is
common to the ray paths. The problematic aspect is that our algorithm for evaluating
ray paths leads to the type of data depicted in Fig. 4. The two ray paths correspond
to slightly different initial grazing angles χ1 and χ2 . Points marked a and b on each
curve are the result of computations, so the R and z coordinates of each point are
known. The algorithm that we use sets an equal depth increment for the computation
of a ray, but that increment is not constant from ray to ray. Thus, za (χ2 ) = za (χ1 )
and zb (χ2 ) = zb (χ1 ).
R
a Ra (2), za (2)
z
a
Rb (2), zb (2)
b
Ra (1), za (1)
b
Rb (1), zb (1)
z
Figure 4.
We wish to determine the intensity at the indicated depth z midway between these
ray paths. To that end we first evaluate the transverse distance to the point on each
ray path at depth z. This is done by interpolating between the coordinates of the
computed point. A linear interpolation gives

zb − z z − za
R z, χj = Ra + Rb
zb − za zb − za χj
Then a first-order finite difference approximation gives

∂R R (z, χ2 ) − R (z, χ1 )
=
∂χI z,(Ra +Rb )/2 χ2 − χ2
Application of this procedure to create an intensity map for the entire field would
require effort. One approach would establish a spatial grid at which the algorithm
would be implemented. A search algorithm would identify the two rays that are
closest to a selected grid point, and the data points on each ray that are directly above
and below the selected point. Section 11.3.2 will develop a formulation that is readily
programed and more accurate.
11.3 Arbitrary Heterogeneous Fluids
The model of a vertically stratified fluid is useful for environmental acoustic studies.
However, it might be that the sound speed depends on the horizontal range, as well as
depth, especially at very long ranges. The analytical approach in the previous section
has the attractive feature of being pictorial, and therefore readily understood. The
method that follows is mathematical and more abstract, but it is general.
11.3.1 Ray Tracing Equations
The formulation begins with the general equation for the zero-phase wavefront, which
was stated previously in Eq. (11.1.4),
x̄ · ē (x̄)
τ (x̄) ≡ =t (11.3.1)
c (x̄)
The corresponding pressure is given by Eq. (11.1.1) for a harmonic signal or

Eq. (11.1.8) for an arbitrary time signature. In either case, a constant value of τ
locates the wavefront on which the phase is zero when t is that value of τ .
The geometrical acoustics approximation of propagation was depicted in Fig. 11.1,
where a specific signal is observed at position x̄ = ξ¯ at time t. At a slightly later
instant t + dt, it is situated on the same ray at x̄ = ξ¯ + d ξ.
¯ Because ξ¯ marks the
position of a specific phase, it must be that d ¯
ξ/dt is the phase velocity. A different

view is that the local sound speed is c ξ¯ , and the propagation is in the direction

ē ξ¯ tangent to the ray path. Both views are correct, so it must be that
d ξ¯
= c ξ¯ ē ξ¯ (11.3.2)
dt
Another view of the propagation is that the phase moves along a ray such that it
remains on the same wavefront. The wavefront at time tis described by Eq. (11.3.1),
and its shape at time t + dt is described by τ ξ¯ + d ξ¯ = t + dt. A Taylor series
expansion this expression gives

τ ξ¯ + ∇τ · d ξ¯ = t + dt (11.3.3)

The fact that τ ξ¯ = t reduces this relation to
d ξ¯
∇τ · =1 (11.3.4)
dt
A fundamental property of the gradient of a function is that it is normal to the
surface

on which the function is constant. Along a wavefront τ is constant, and ē ξ¯ is the
11.3 Arbitrary Heterogeneous Fluids 439

normal, so we have ∇τ = |∇τ | ē ξ¯ . Substitution of this description and Eq. (11.3.2)

into the preceding leads to |∇τ | = 1/c ξ¯ . An equivalent description of this property
is
1
∇τ · ∇τ = 2 (11.3.5)
c ξ¯
This is the eikonal equation. Because the magnitude of ∇τ is inversely propor-

tional to the local sound speed, and it is in the sense of
the propagation, it is called
the wave slowness. We shall denote this quantity as s̄ ξ¯ , so we have

ē ξ¯
s̄ ξ¯ ≡ ∇τ = (11.3.6)
c ξ¯
Manipulation of the relations developed thus far will lead to a set of first-order
differential equations in which the dependent variables are ξ¯ and s̄. The derivation
begins by evaluating the rate at which s̄ changes. One way in which that derivative
may be found is by the chain rule based on considering s̄ to be a function of position
along a ray, which leads to
ds̄ d ξ¯
= ∇ s̄ · = ∇ s̄ · (cē) (11.3.7)
dt dt
(The quantity ∇ s̄ is the gradient of a vector, which we usually do not encounter.
In Cartesian coordinates, each component of s̄ may depend on x, y, and z, and the
unit vector parallel to each axis is constant. Correspondingly, we would find that
∇ s̄ = ∇sx ēx + ∇sy ēy + ∇sz ēz .) We use the definition of s̄, Eq. (11.3.6), which
indicates that ē = cs̄, to eliminate the unit vector,
ds̄ 1
= c2 ∇ s̄ · s̄ ≡ c2 ∇ (s̄ · s̄) (11.3.8)
dt 2
The eikonal
equation gives s̄ · s̄ = 1/c2 . The gradient of this quantity is ∇ (s̄ · s̄) =
−2/c3 ∇c. Substitution of this relation converts the preceding equation to
ds̄ ∇c
=− (11.3.9)
dt c ξ¯
In addition, substituting ē = cs̄ into Eq. (11.3.2), which describes the phase velocity,
leads to
d ξ¯ 2
= c ξ¯ s̄ ξ¯ (11.3.10)
dt
Equations (11.3.9) and (11.3.10) are coupled ordinary differential equations that
govern ξ¯ and s̄ as functions of time. They are known as the ray tracing equations.
Except for a few special cases, their solution requires computational methods. The
numerical algorithms typically implemented in software solve first-order equations
that have been placed into a matrix form,
d
{X} = {F} (11.3.11)
dt
The ray tracing equations are expressed in this form by placing the components of s̄
and ξ¯ in the column vector {X}. The components of the right side of each equation are
placed in the same sequence in {F}. The result is a set of six differential equations,
T
{X} = sx sy sz ξx ξy ξz

1 ∂c ∂c ∂c T (11.3.12)
{F} = − c 2 sx sy sz
c ∂ξx ∂ξy ∂ξz
The sound speed is a function of position. Hence, it is implicit that embedded in the
computation of {F} are routines that evaluate c and the components of its gradient
as functions of the position coordinates ξx , ξy , and ξz .
To start the numerical solution, the initial conditions must be stated. The initial
instant may be set as t = 0 because time does not appear explicitly in the ray
tracing equations. The analysis of rays in a vertically stratified fluid began by setting
the starting position and the grazing angle. The analogous information here is a
specification of where a ray is situated, ξ¯ = ξ¯0 , and the direction in which it is
aimed, ē = ē0 , at t = 0. The corresponding initial wave slowness is s̄0 = ē0 /c ξ¯0 .
The result of the numerical solution is a set of ξ¯ and s̄ values at a set of t value for a
single ray. Depending on how the signal is generated, we might find families of rays
by changing the starting point or the initial tangent direction.
It generally is a good idea to have an external check to monitor the error when
differential equations are solved numerically. For example, if energy is conserved in
a dynamic system, the mechanical energy contained in the response at each instant
can be verified to be constant. A conservation-like principle is the eikonal equation,
s̄ · s̄ = 1/c2 . Solution of the differential equations gives the values of ξ¯ and s̄ at some
2
instant t. Knowledge of these values makes it possible to compute c ξ¯ s̄ · s̄ − 1.
The degree to which the computed value differs from zero is a fractional measure of
the error.
One situation where the ray tracing equations may be solved analytically is that
of a homogeneous fluid, for which ∇c ≡ 0. Equation (11.3.9) states that s is constant
in this case, so it never changes from its initial value, s̄0 . In turn, this means that
the right side of Eq. (11.3.10) is constant at the initial value s̄0 . The solution of that
equation is
ξ¯ = ξ¯0 + c2 s̄0 t = ξ¯0 + ē ξ¯0 ct (11.3.13)
This is the equation of a straight line. Thus, the ray tracing equations for a homo-
geneous fluid yield rays that are straight, which is a property we already knew.
Furthermore, the signal at instant t is at distance ct from the starting point, as it must
for plane wave propagation.
EXAMPLE 11.4 Consider a fluid whose sound speed depends linearly on

the depth. Start with an ansatz for a ray path that is the general equation for a
circle. Show that the ray tracing equations are satisfied by this representation if
the center of the circle is situated at the extrapolated elevation where the value
of c is zero.
Significance
The process of working backward to fit the trial solution to the ray tracing equa-
tions will clarify what the forward solution process actually entails. It also serves to
tie the graphical approach for vertically stratified media to the mathematical formu-
lation for arbitrary heterogeneity.
Solution
The mathematical operations are somewhat simplified by defining z = 0 to be the
depth at which c = 0. (This location might be outside the region in which the fluid
resides.) This definition of z leads to c = βz, where β is an arbitrary constant. There
are several ways in which a circular path may be represented mathematically. We
will use a parametric form in which the grazing angle χ is the parameter. Figure 1
shows that χ also is the angle of the radial line to a point on the ray. The coordinates
of the center are (x0 , z0 ). Neither these coordinates nor the radius R is presumed to
be known.
Figure 1.
The diagram shows that the coordinates of a point on the circular arc are
ξx = x0 + R sin , ξz = z0 − R cos (1)

Thus, the sound speed at x̄ = ξ¯ is

c ξ¯ = β (z0 − R cos ) (2)
The slowness vector is tangent to the ray, and its magnitude is 1/c, so we find from
the sketch that
1
s̄ = (ēx cos + ēz sin χ) (3)
c ξ¯
The last aspect of the trial solution is the introduction of time. A signal travels along
a ray at the local speed of sound. This is the rate at which point ξ¯ moves along the
arc. The path is circular, so the corresponding angular speed is

dχ c ξ¯
= (4)
dt R
Our approach is to substitute these representations into the ray tracing equations.
Because all quantities have been defined in terms of χ, derivatives with respect to
time may be evaluated by the chain rule. We begin with Eq. (11.3.10), for which we
have
c
d ξ¯ dξx dξz dχ
= ēx + ēz = R (ēx cos + ēz sin ) ≡ c2 s̄
dt dχ dχ dt R
This shows that Eq. (1) identically satisfies Eq. (11.3.10) independently of the radius
and coordinates of the center. In other words, any circular path is consistent with that
equation.
Consideration of Eq. (11.3.9) proceeds similarly. An expression for ds/dt may
be found by differentiating Eq. (3). The key observation is that Eq. (2) gives c as a
function of χ, which means that Eq. (3) actually is a description of s as a function of
χ. This calls for the chain rule for differentiation, which leads to

ds̄ dχ ds̄ c d 1
= = (ēx cos χ + ēz sin χ)
dt dt dχ R dχ c
1 d
+ (ēx cos χ + ēz sin χ)
c dχ
c 1 dc 1

= − (ēx cos χ + ēz sin χ) + (−ēx sin χ + ēz cos χ)
R c2 dχ c
1 βR sin χ
= − (ēx cos χ + ēz sin χ) + (−ēx sin χ + ēz cos χ)
R βz0 − R cos χ
Collection of like components reduces this expression to
ds̄ β
= z0 (−ēx sin χ + ēz cos χ) − Rēz (5)
dt β (z0 − R cos χ) R
This expression constitutes the left side of Eq. (11.3.9). The right side is found by
using the original description c = βz to evaluate the gradient. Thus,

∇c 1 ∂c 1
− =− ēz =− (β ēz ) (6)
c c ∂z z=ξx β (z 0 − R cos χ)
When we equate Eqs. (5) and (6), we see that the ēx and ēz components match
only if z0 = 0. Because z0 is the depth of the center of curvature, the finding that
z0 = 0 places the center of the circular ray path at the depth where c vanishes. As is
true for the first ray tracing equation, Eq. (11.3.9) is satisfied for any R. This quantity
and the actual location of the center are set by the initial position and angle at which
the ray is launched. The circular nature of the ray path in a linearly varying fluid
was identified previously. Solution of the ray tracing equations gives us a concrete
illustration of what the ray tracing equations represent.
11.3.2 Amplitude Dependence
Transport Equation
Suppose we know the waveform of a signal at the starting location ξ¯0 of a ray. We
wish to know the waveform at a downstream point ξ¯A on that ray. Part of such a
determination requires that we find the travel time for the signal. The discussion of
Eq. (11.1.4) informs us that we can extract this property directly from the solution of
the ray tracing equations. The solution algorithm sets the time base such that t = 0 for
the starting point ξ¯0 . The wavefronts that are computed correspond to zero phase. This
means that the signal that departed from ξ¯0 at t = 0 will be observed at the computed
location ξ¯A at time tA . According to the aforementioned discussion, a signal that
departs from ξ¯0 at some instant t > 0 will arrive at ξ¯A at time tA + t . Consequently,
if we retain the time value associated with each ξ¯ value obtained in a numerical
solution of the ray tracing equations, then we may use that time to determine the
time delay of the waveform function at each location. A different route leading to
the same conclusion comes from the observation that the independent variable for
the ray tracing equations is t, but it does not appear explicitly anywhere in those
equations other than as the rate variable for derivatives. Thus, shifting the time at ξ¯0
from t = 0 to t = t merely increments the time value at any point ξ¯A from tA to
tA + t .
Knowledge of the time delay allows us to determine when a specific phase arrives
at any location along a computed ray. However, this knowledge sheds no light on the
amplitude dependence of the signal. Here, we will develop a procedure by which this
dependence may be identified if the signal is harmonic. The basic notion is to seek
conditions that must be satisfied in order that the geometrical acoustics description
of pressure be a solution of the position-dependent wave equation. For this analysis,
we use the field equation that accounts for density variation, which is Eq. (11.1.21).
Doing so yields a result that is valid for gases, as well as liquids.
We begin with the basic ansatz in Eq. (11.1.1). In term of the wavefront function
τ (x̄), it is
p (x, t) = P (x̄) Re eiω(t−τ (x̄)) (11.3.14)
From this, we find the pressure gradient to be

∇p = Re (∇P − iωP∇τ ) eiω(t−τ (x̄)) (11.3.15)
The divergence of this expression is

∇ 2 p = Re ∇ 2 P − 2iω∇P · ∇τ − iωP∇ 2 τ − ω 2 P∇τ · ∇τ eiω(t−τ (x̄))
The dependence of p on t is harmonic. Thus, substitution of the preceding into

Eq. (11.1.21) leads to

1 1 ∂2p
∇ p − ∇ρ0 · ∇p − 2 2 = Re
2
∇ 2 P − 2iω∇P · ∇τ − iωP∇ 2 τ
ρ0 c ∂t
1
−ω 2 P∇τ · ∇τ − ∇ρ0 · (∇P − iωP∇τ )
ρ0
ω
+ 2 P eiω(t−τ (x̄)) = 0
c
(11.3.16)
This equation must be satisfied for any combination of x̄ and t values, so the
complex exponential may be canceled. The fact that both P and τ are real assists
decomposition of the factor of this exponential into its real and imaginary parts. The
real part is

1 1
∇ P + ∇ρ0 · ∇P − ω ∇τ · ∇τ − 2 P = 0
2 2
(11.3.17)
ρ0 c
Geometrical acoustics considers ω to be very large, which means that terms that
are proportional to ω 2 are dominant. This attribute makes it permissible to drop the
terms that not contain ω. What remains is the factor in parentheses multiplying P.
The relation must be true for any P. Therefore, the term inside the parentheses must
be zero. The result is a restatement of the eikonal equation, so it provides no new
information.
The imaginary part of Eq. (11.3.16) is the one that is useful. All of these terms are
proportional to ω, so none may be dropped as being insignificant relative to others.
Factoring out −ω converts the imaginary part to the transport equation,
1
2∇P · ∇τ + P∇ 2 τ − P∇ρ0 · ∇τ = 0 (11.3.18)
ρ0
This expression features three terms, each of which is a derivative of one variable in a
triple product. The term − (∇ρ0 ) /ρ0 is identically ρ0 ∇ (1/ρ0 ) and the two factor for
the first term suggests that the underlying term is P2 . Consequently, let us multiply
the transport equation by P/ρ0 . Doing so gives

P P2 2 1 P2
2 ∇P · ∇τ + ∇ τ + P2 ∇ · ∇τ ≡ ∇ · ∇τ =0 (11.3.19)
ρ0 ρ0 ρ0 ρ0
This relation is further simplified by the definition that ∇τ ≡ s ≡ ē/c. Substitution

of this relation reduces the transport equation to

P2
∇· ē = 0 (11.3.20)
ρ0 c
The particle velocity in the geometrical acoustic approximation fits the relation
for a planar wave, according to which V̄ = ēP/ (ρ0 c). Therefore, the quantity inside
the parentheses is twice the time-averaged intensity tangent to the ray. If we already
have solved the ray tracing equations, then the intensity is the sole unknown in
this version of the transport equation. In principle, we could employ a numerical
differential equation solver to determine how the intensity varies along the ray path.
However, there is a simpler way of extracting P along a ray. It does not require
solution of a differential equation.
Relation to Ray Tube Area
The occurrence of intensity as the variable governed by the transport equation sug-
gests that this relation is fundamentally like the result for intensity in a depth-
dependent heterogeneity in Sect. 11.2.2. To examine this observation, we consider a
ray tube, such as the one in Fig. 11.15. The starting point for the ray tube is x̄ = ξ¯0 ,
and the cross-sectional area at any location ξ¯ is the infinitesimal value dA ξ¯ .
Fig. 11.15 Ray tube in an __

arbitrary heterogeneous e()
fluid. The edges are rays
whose initial tangent
corresponds to four sets of
direction angles that differ
by differential amounts __
d e()
Green’s (divergence) theorem is applied to the transport equation, with the domain
V being the region contained within the ray tube. The intensity is in the direction
ē tangent to the rays, and the normal to the side walls is perpendicular to ē. Con-
sequently, there is no contribution to the integral from the side walls. It follows
that
2 2 2
P P P
∇· ē dV = ē · ē dA + (−ē) · ē dA = 0
ρ0 c ρ0 c ρ0 c
V dA(ξ̄ ) dA(ξ̄0 )
(11.3.21)
The ray tube cross section is infinitesimal in extent, so the pressure may be considered
to be constant along it. (The behavior at caustics will be addressed separately.) The
direction −ē in dAI corresponds to an inflow. Hence, the above relation states that
the time-averaged power that flows into the ray tube at its beginning must equal the
power that flows out at the end, that is
2 2
P ξ¯ P ξ¯0
dA ξ = dA ξ¯0
¯ (11.3.22)
¯
ρ0 ξ c ξ ¯ ¯
ρ0 ξ0 c ξ0 ¯
The explicit solution for the pressure amplitude at position ξ¯ on a ray is

1/2
ρ0 ξ¯ c ξ¯ dA ξ¯0
P ξ¯ = P ξ¯0 (11.3.23)
ρ0 ξ¯0 c ξ¯0 dA ξ¯
Equations (11.3.23) and (11.2.23) are equivalent. Indeed, Eq. (11.3.22) could have
been obtained from the principle that the time-averaged power that flows into the
ray tube must equal the power that emerges. This was the basis for the analysis of
vertically stratified media. In view of this observation, it is reasonable to ponder why
the present derivation derived a transport equation, and then used it to determine the
amplitude dependence. The answer is that the energy conservation argument does
not hold if there is a flow in the ambient state, whereas a modified transport equation
would be available.
The preceding developments indicate that we can determine the amplitude of the
signal anywhere along a ray if we know the amplitude on that ray at some starting
point. But to do so we must evaluate the ray tube area. With that determination
as the objective, let us return to the situation addressed in Sect. 11.2, specifically,
an axisymmetric field in a depth-dependent fluid. The solution of the ray tracing
equations gives position along a ray as a function of time. Different initial values of
the grazing and azimuthal angles generate different rays, so position on a ray is a
function of t, χI , and θI . However, by virtue of axisymmetry, the range R and depth
z of a point depend only on t and χI because the azimuth angle is invariant along a
ray. Rather than considering t to be the independent variable, it is advantageous for
this discussion to use the arclength measured along a ray from the starting position.
This quantity is in a one-to-onecorrespondence
to t. Indeed, the differential position
increment in a time t is d = c ξ¯ dt, so we could determine by a direct integration
along the ray path. With this change of variables, the transverse distance and depth
are described functionally as R (, χI ) and z (, χI ). This is the situation depicted in
Fig. 11.16. (The difference between this Figure and Fig. 11.11 is that R previously
was considered to depend explicitly on z.)
A ray tube cross section consists of all points at the same t. Because all rays within
the tube are very close, the rays intersect the cross section at the same arclength .
Thus, the coordinates of the end point on all upper rays in Fig. 11.16 are R (, χI )
and z (, χI ), while the coordinates at the end of all lower rays are R (, χI + dχI )
and z (, χI + dχI ) .
R(l,I )
R(l,I )

dI
dI z(l,I )
l z(l,I ) z(l,I +dI )
l l dh
z R(l,I +dI )
dA RdI
dh
Fig. 11.16 Evaluation of ray tube area for the axisymmetric field radiated by a point source. The
initial grazing angle is χI , the azimuthal angle is θ, and is the arclength along a ray measured
from the source
The Pythagorean theorem gives the distance in the vertical plane between these
points,
1/2
dh = [R (, χI + dχI ) − R (, χI )]2 + [z (, χI + dχI ) − z (, χI )]2

1/2
∂R 2 ∂z 2
= + dχI (11.3.24)
∂χI ∂χI
The distance in the horizontal plane between the rays bounding the tube is RdθI . The
corresponding cross-sectional area is
2 2
1/2
∂R ∂z
dA ξ¯ = R + dχI dθI (11.3.25)
∂χI ∂χI
This expression would be useful if we knew R and z as analytical functions, but

numerical solution of the ray tracing equations gives a digitized representation. In
that case if two adjacent rays n and n + 1 are launched at grazing angles that differ
by a very small value χI , then the derivatives may be approximated as first-order
finite differences,
∂R Rn+1 − Rn ∂z zn+1 − zn
≈ , ≈ (11.3.26)
∂χI χI ∂χI χI
where Rn and zn are the coordinates on ray n at the instant of interest, and Rn+1 and
zn+1 are the coordinates for ray n + 1 at that instant.
A hint as to how to proceed when the field is not axisymmetric comes from
the observation that the bracketed term in Eq. (11.3.25) describes dh. This distance
is measured in the vertical plane between points on adjacent rays that are at the
same arclength. Thus, dh represents the magnitude of the change of position when
and θI are held constant. An equivalent statement is that dh is derived from the
Jacobian of the transformation from the (R, z, θ) position variables to the (, χI , θI )
variables of a ray path when θI and are held constant. A description of ray tube
area in terms of a Jacobian is the basic approach described by Jensen et al.9 for
handling an arbitrary situation. Ultimately, evaluation of this quantity would be done
via numerical methods because the ray paths are known as discrete data sets. Rather
than following that approach, we will develop a formulation that relies solely on
vector algebra.
In Fig. 11.17, a ray tube is formed by a selected ray, designated as number 1, and
two adjacent rays, numbered 2 and 3. The adjacent rays must be independent of ray
1. For example, if ray 1 corresponds to launch angles χI and θI , then ray 2 could
correspond to χI +χI and θI , while ray 3 could correspond to χI and θI +θI , where
the angle increments are extremely small. This analysis of ray tube area requires that
the evaluation of points on each ray use a common time base. (This would be the case
if we prescribe the same output instants when the ray tracing equations are solved
numerically.) This scheme allows us to identify points ξ¯1 , ξ¯2 , and ξ¯3 on the respective
rays that correspond to any selected instant tn . Therefore, these three points lie on
the same wavefront τ = tn .
Fig. 11.17 Vectors used to

evaluate the cross-sectional
area of a ray tube _
_
r 
_ _ _ _
 r _ e()
_

h
_
h
From ξ¯1 we construct relative position vectors to the other points,
h̄2/1 = ξ¯2 − ξ¯1 , h̄3/1 = ξ¯3 − ξ¯1 (11.3.27)

The cross section area of this ray tube is A = h̄2/1 h̄3/1 sin angle h̄2/1 , h̄3/1 .
This is the magnitude of a scalar triple product, with the tangent vector as the third
vector. Furthermore, although this area is finite, the fact that the rays correspond to
initial conditions that are only slightly different allows us to consider the area to be
an approximation of dA, so we have found that

dA ≈ h̄2/1 × h̄3/1 · ē1 ξ¯1 ≡ c ξ¯1 h̄2/1 × h̄3/1 · s̄1 ξ¯1 (11.3.28)
9 F.B. Jensen et al., W.A. Kuperman, M B. Porter, and H. Schmidt, Computational Ocean Acoustics,
2nd ed., Chap. 3, Springer (2011).

If we wish to ascertain how the amplitude varies along a specific ray, which
has been designated as ray 1, we would employ the preceding description of dA
to formulate Eq. (11.3.23) for each computed point on that ray. Alternatively, we
can apply the computation of dA to the point on each ray at a common instant.
The outcome of such an evaluation would be the distribution of pressure along a
wavefront. Repetition of either calculation for all rays and all wavefronts would
yield a map of the pressure field.
Reflections and Caustics
The procedure we have developed entails computing a set of rays by solving the ray
tracing equations. Each ray corresponds to a different initial condition at tI = 0, which
may be a different initial position ξ¯I and/or a different initial tangent direction ēI .
The differential equation solution marches forward incrementally. Upon reaching the
farthest location, the solver is halted. After all rays have been computed, the properties
of the rays may be used to determine how the pressure varies along any ray, provided
that the computation has described a sufficient number of rays to approximate ray
tubes.
The evaluation of each ray may be carried without interruption, unless a ray
encounters a boundary. In that case, the ray is reflected. In view of the locally planar
nature of a signal in the geometrical acoustics approximation, the rules for reflection
of a plane wave that is incident on an infinite plane may be employed here. In
¯
at ξB. The
Fig. 11.18, a ray is incident on a solid boundary normal to the surface,
pointing into the fluid, at this location is n̄ ξB , and ē ξ¯B− is the tangent to the ray
¯
path immediately before it touches the boundary. The tangent to the reflected ray
immediately after it leaves the surface is ē ξ¯B+ . The angle of incidence ψB equals
the angle of reflection.
Fig. 11.18 Unit vectors n(B)

B
describing the incidence and e(B -) B
reflection of a ray at position e(B +)
ξ̄B in a solid surface
et (B)
B

We wish to describe ē ξ¯B+ in a manner that does not require defining a coordinate

system. To that end, let ēt ξ¯B denotes the tangent to the surface at the point of

incidence in the plane formed by ē ξ¯B− and n̄ ξ¯B . It follows that the component
representation of the tangent directions is

ē ξ¯B− = ēt ξ¯B sin ψB − n̄ ξ¯B cos ψB , ē ξ¯B+ = ēt ξ¯B sin ψB + n̄ ξ¯B cos ψB
(11.3.29)
Subtracting the first equation from the second yields the desired relation. The angle
of incidence may be found from a dot product, so we have

ē ξ¯B+ = ē ξ¯B− + 2n̄ ξ¯B cos ψB , ψB = cos−1 −ē ξ¯B− · n̄ ξ¯B (11.3.30)
Subsequent to reflection, solution of the ray tracing equations mayresume.

The initial
conditions for that computation would be the values of ξ¯B and ē ξ¯B+ at tB .
The computation of the amplitude variation along a ray also must account for
a reflection. Because the angles of incidence and reflection are equal, the ray tube
area immediately prior to incidence is the same as the area after reflection. Hence,
the only reason the amplitude might be altered by the reflection is that the reflection
coefficient is not unity. Suppose the surface is locally reacting with impedance Z.
Then, the reflection coefficient is

Z ξ¯B − ρ0 c/ cos ψB
R= (11.3.31)
Z ξ¯B + ρ0 c/ cos ψB
Neither the local impedance nor the angle of incidence is required to be constant along
the surface. If that is the situation, some rays will be attenuated more than others as
a result of reflection. However, this effect often is ignored by taking the surface to be
rigid, R = 1, based on the fact that at very high frequencies most materials have a
large impedance. In any case, the amplitude immediately after reflection is R times
the amplitude before. Using the properties at the reflection point as the input for the
reflected ray tube leads to a description of the reflected signal as
⎛
1/2 ⎞
¯ ¯
ρ0 ξ c ξ dA ξB ¯
¯ t = Re ⎝
p ξ, RP ξ¯B− eiω(t−τ (x̄)) ⎠ (11.3.32)
ρ0 ξ¯B c ξ¯B dA ξ¯
The occurrence of a caustic is the consequence of the global behavior of a family

of rays, rather than a special feature of a single ray. Thus, there is no need to interrupt
solution of the ray tracing equations upon arrival at a caustic. (Indeed, one would
be unlikely to recognize the existence of a caustic until the full set of rays has been
computed.) The occurrence of a caustic does alter the amplitude dependence. We
shall merely describe how to account for that effect because the analysis is beyond
the present scope. The interested reader is referred to Pierce,10 which also contains
references to the original analyses.
Mathematical analysis would reveal that a harmonic signal undergoes a π/2 phase
delay as it propagates along its ray through a caustic. Specifically, if A and B ,
respectively, are the arclengths to points on a ray prior to, and following, the caustic
at caustic , then we have
⎛
1/2 ⎞
c ξ¯B dA ξ¯A iω(t−τ (x̄))
p ξ¯B , t = Re ⎝ P ξ¯A e ⎠ if B < caustic
c ξ¯A dA ξ¯B
⎛
1/2 ⎞
c ξ¯B dA ξ¯A −iπ/2 iω(t−τ (x̄))
p ξ¯B , t = Re ⎝ P ξ¯A e e ⎠ if B > caustic
c ξ¯A dA ξ¯B
(11.3.33)
10 A.D. Pierce, Acoustics, Sect. 9-4, ASA Books reprint of 1981 edition.
If it should be that two caustics are situated on a ray between ξ¯A and ξ¯B , then the phase
delay would be 2 (π/2), and so on. Note that the ray tube area decreases to zero at a
caustic, which means that the linear acoustics approximation requires correction. In
the vicinity of the caustic, the amplitude will be greatly enhanced.
A heuristic explanation of this phase shift may be obtained by considering the ray
tube area beyond a caustic to be negative. This view stems from the description of
ray tube area as a proportionality to the Jacobian. The rays cross over as they pass
the caustic. The Jacobian is positive ahead of the caustic, zero at the caustic, and
negative beyond it. The amplitude is proportional to the square root of the area, and
the square root of a negative number gives a ±i ≡ exp (±iπ/2). The root that applies
is −i for a delay of the signal. This aspect may be viewed as a manifestation of the
principle of causality. A factor +i as the signal passes through a caustic corresponds
to a phase lead. If that were the case, it would mean that a feature is observed at a
location infinitesimally beyond the caustic prior to its observation at a location that
infinitesimally precedes the caustic.
EXAMPLE 11.5 Air is contained within a rectangular waveguide whose side

walls are composed of tempered glass, which is well approximated as being
rigid. As sketched below, circular plates of radius a are attached to the hor-
izontal walls. The plates are cooled, thereby altering the sound speed in the
region between them. Because the depth H is small and the walls are rigid, it
is reasonable to consider the sound speed to vary only in the horizontal direc-
tions. Hence, the cooled region may be taken to constitute a vertical
cylinder of

lower sound speed. The specific dependence is c = c0 +(c1 − c0 ) 1 − R2 /a2 ,
where R is the transverse distance from the centerline connecting the plates.
Outside this region, the sound speed is c0 . The wall at x = 0 generates a 200
kHz harmonic plane wave, and the wall at x = L is actively controlled to elimi-
nate reflections. Draw the rays and wavefronts. Then determine the dependence
of pressure amplitude on distance along the wavefront that touches the front
of the cylindrical region, where x = 2.5a. Parameters for the evaluation are
c0 = 330 m/s, c1 = 280 m/s, and a = 250 mm.
y
1.5a
a
x
L
z Figure 1.
Significance
Application of numerical methods to solve the ray tracing and transport equations
is a primary thrust, and the results will greatly enhance our understanding of the
behavior of rays and wavefronts, including a much deeper insight to the behavior
near caustics.
Solution
The sound speed depends on the transverse distance R, but the waveguide is best
described in terms of Cartesian coordinates. Because the sound speed is independent
of the vertical position y and the top and bottom walls are rigid, the pressure field
will be independent of y, so s̄ and ξ¯ may be described solely in terms of x and z
components. The column vector of variables for a numerical solution is
T
{X} = sx sz ξx ξz
The corresponding vector driving d{X}/dt is

T
1 ∂c ∂c
{F} = − 2
c sx sz
c ∂ξx ∂ξz
The transverse
distance
from the vertical centerline of the cooling plates to ξ¯ is
¯
R = ξ − 1.5aēx , which is used to evaluate the given expression for c(ξ). ¯ The
corresponding gradient of c is ∇c = (dc/dR) ∇R. We know that ∇R = ēR , where ēR
¯ Thus,
is the unit vector from the center to point ξ.
! 2R
dc − (c1 − c0 ) ēR if R < a
∇c = ēR = a2
dR 0 if R ≥ a
This gradient must be described in term of x and z components, with ξx and ξz as

the variables. The required expressions are obtained by a transformation to Cartesian
coordinates, which sets
ξ¯ − 1.5aēx
ξ¯ = ξx ēx + ξz ēz , R = ξ¯ − 1.5aēx , ēr =
R
The numerical solver routine must be selected. Matlab® offers several. The one
that was used to generate the data is ODE45, which implements a fourth/fifth-order
Runge–Kutta algorithm. The input arguments for this routine are the name of a
function F_rhs that will evaluate {F}, followed by a column or row vector t_span
holding t values at which the solution will be output. The next input argument is the
initial value of {X}, which will be addressed below. The last input argument for
ODE45 is an optional set of parameters setting error limits. The results here were
obtained by setting the relative error tolerance to 10−6 .
There are two alternative procedures for stepping through time with ODE45. If
t_span holds all of the time instants, the output from a single call to ODE45 would
be a rectangular array, which we will denote as [Xout ]. The nth row of [Xout ] would
be the solution {X} at time tn . The alternative procedure explicitly steps through time
by setting t_span as a two-element vector [tn−1 tn ]. The value of {X} at tn is the
last row of [Xout ] in that case. (The other rows are {X} values at intermediate instants
listed in {tout }.) The time history for all variables in this procedure may be constructed
by adjoining the transpose of the column vectors {X}. In either treatment, the time
values at which data is output are stored in a column vector {t}. Clearly, the single-
call procedure is easier to implement, and it was found to require much less time
to execute. However, the explicit stepping procedure was found to be significantly
more accurate.
The numerical integration must be performed for each ray. This was done within
a loop over the ray number. The solution for all rays should use a common base of
t values. Doing so aligns the data such that the computed ξ¯ values for all rays lie
on a common set of wavefronts. The time increment for the evaluation was taken
to be the time required to travel a distance a/25 at speed c0 , and the maximum
time was set as the time to travel 6a at speed c0 . Thus, the number of tn values is
N = (6a/c0 ) / (a/25c0 ) = 150. Each pass through the ray number loop corresponds
to a ray that departs from the plane x = 0 at an incremented distance zj in the z
direction. Each ray initially is a straight line, so ē = ēx at the start, which is defined
to be t = 0. Let J be the number of rays. The diameter of the cooled region is 2a,
so the rays were initialized at a uniform spacing to cover a width of 3a, which gives
a separation of = 3a/ (J − 1) between rays. The corresponding initial conditions
for the solution for ray j were sx = 1/c0 , sz = 0, ξx = 0, ξz = −1.5a + (j − 1).
A single pass through the outer loop over ray numbers gives [Xout ] for ray j. The
four rows of [Xout ] consist, respectively, of sx , sz , ξx , ξz at each tj . The row of ξx data
was stored as column j of a rectangular array [ξx ], and the ξz row was likewise stored
in column j of [ξx ]. The eikonal error metric for each ray, numbered j, was obtained
by evaluating c2 [(sx )2 + (sz )2 ] − 1 for each instant tn , and then storing the error
values at all instants as column j of the rectangular array [εeik ]. The largest eikonal
errors were found to occur along rays 21 and 25. Figure 2 displays the variation of
this quantity along the respective rays. The largest error for any ray was found to be
less than 0.08%.
0.001
Error eik
Ray 21: h=16.67 mm

0
Ray 25: h=10 mm
-0.001
0 1 2 3 4
Time t (ms)
Figure 2.
When the data is stored as described above, a ray may be drawn by plotting
a column of [ξx ] against the corresponding column of ξy , while wavefronts
are
obtained by plotting a row of [ξx ] against the corresponding row of ξy . The rays and
wavefronts appear in Fig. 3. For the sake of clarity, only every tenth of the evaluated
wavefronts is shown there. There are many interesting facets to this picture. The
rays that pass through the cooled region are bent inward toward the x-axis. This
is consistent with the property observed for vertical stratification. It results from a
wavefront being retarded where the sound speed is lower, thereby having the effect
of turning the wavefront. There appears to be regions where there is no ray, but these
are not shadow zones. Rather the rays are sparse there. If we were to greatly increase
the number of computed rays, we would see that rays do pass through those regions.
Regions where the rays are sparse correspond to low, but not zero, pressure.
0.4
0.2
Transverse position z (m)
−0.2
−0.4
0 0.2 0.4 0.6 0.8 1 1.2
Figure 3.
Figure 3 shows that there is a region in which many rays converge. Figure 4 zooms
in on this region. The concentration is not a focus because the rays do not intersect at
a single point. Another feature leading to the same conclusion is that the wavefronts
beyond the convergence are not circles, as they would be if the rays were radiated from
a focus. Rather, the concentration of rays is a caustic. Inspection of a region beyond
the arrete shows that at any point there are three rays. Two arrive at comparatively
large angles from either side of the x-axis, whereas the third comes at a small angle
from the opposite side of the x-axis. This results in folding of a wavefront. The part
closer to the arrete is a curved arc. It is the locus of constant τ values for the shallow
rays. The portions of a folded wavefront that intersect at the x-axis correspond to the
large angle rays.
0.1
Transverse position y (m) 0.05
−0.05
−0.1
0.8 0.9 1
Figure 4.
The location of the caustic can be shifted by adjusting the value of c1 with c0
fixed. Decreasing c1 increases the heterogeneity, thereby increasing the curvature of
the rays within the cooled region. This brings the caustic closer to the origin. Indeed,
if c1 is sufficiently large, the caustic will occur inside that region.
It still remains to evaluate the pressure along a wavefront. For this analysis, there
are several items we must determine. First, we must determine where the computed
ray paths intersect the wavefront that was stipulated. The point where this wavefront
tangentially touches the cooled region is ξx = 2.5a = 0.625 m, ξz = 0. The ray
passing through this point is the straight one along the center line. Its ray number
is j = (J + 1) /2 = 31. None of the computed ξx values for this ray exactly equals
0.625 m. We can use interpolation to reach the specified wavefront. First, we scan
the ξx values for ray 31 to identify the points that bracket ξx = 0.625 m. These are
(ξx )69,31 = 0.6225 and (ξx )70,31 = 0.6325. A linear interpolation between these
points should give (ξx )wf,31 = 0.625 m, that is,
0.625 = 0.6225 + β (0.6325 − 0.6225) =⇒ β = 0.25
The interpolation fraction β is used to find the value of τ for the wavefront, according
to
τwf = τ69 + β (τ70 − τ69 ) = 2.0681 ms
This interpolation also is employed to determine the x and y coordinates at which

any ray j intersects the wavefront, according to
! " ! " ! "
(ξx )wf,j (ξx )69,j (ξx )70,j − (ξx )69,j
= +β , j = 1, ..., J
ξy wf,j ξy 70,j ξy 70,j − ξy 70,j
Now that the ray-wavefront intersections have been located, the next task is to
determine the ray tube areas. Reference to Fig. 5 will assist following this develop-
ment. The depth perpendicular to ray diagrams is constant, and we are only interested
in the ratio of areas. Therefore, an area ratio reduces to the ratio of distances between
intersection points. Let hwf, j be the distance between the points at which rays j and
j + 1 intersect the wavefront. These distances are

hwf,j = ξ¯wf, j+1 − ξ¯wf, j , j = 1, ..., J − 1
The rays are equally spaced at distance when they depart from x = 0, so h1,j =
for all j.
wf, j+1 av, j+1
hwf, j
Ray j+1
Ray j wf, j wf, j

wf, j-1
Ray j-1
Ray 1
wf,1
Figure 5.
The last quantity required to evaluate the pressure amplitude according to

Eq. (11.3.23) is the characteristic impedance. The stipulated wavefront does not lie
in the cooled region, so the ratio of ρ0 c along it to the ρ0 c value at the initial point
is unity. Nevertheless, we shall incorporate this effect into the computation because
doing so would allow us to probe inside the cooled region. The value of ρ0 is not
specified for any location, but we can deduce it from the given ambient properties.
In the ambient state, the fluid everywhere is quiescent, so there can be no pressure
gradient. It follows that the pressure within the cooled region must be the same as it
is outside this region. We shall assume that the ambient pressure is atmospheric, so
P0 = 1 atm = 101.32 kPa. We know that c = (γP0 /ρ0 ) for an ideal fluid, with
1/2
¯
γ = 1.4 for air, and the value of c ξ at any location is given. Solving this relation
2
for the density gives ρ0 ξ¯ = γP0 /c ξ¯ .
The ray tube numbered j covers the distance between ξ¯wf,j and ξ¯wf,j+1 . It is reason-
able to use the impedance midway between
those points to formulate Eq. (11.3.23).
These values are (ρ0 c)av,j ≡ ρ0 ξ¯av,j c ξ¯av,j , where ξ¯av,j is the midpoint position,
1 ¯
ξ¯av,j = ξwf,j + ξ¯wf,j+1
2
The pressure at the midpoint corresponding to these parameters is indicated by
Eq. (11.3.23) to be

Pwf,j (ρ0 c)av,j 1/2 1/2
=
PI (ρ0 c)x=0 hwf, j
where PI is the pressure amplitude in the plane wave at x = 0.

These pressure ratios are associated with the midpoint between rays. To plot them,
we will use as the abscissa the distance along the wavefront measured from the point
where the wavefront intersects the first ray. This distance, which we shall denote as
ηwf,j , is shown in Fig. 5 to be approximately the accumulation of the ray tube widths
preceding it, according to.
1 1 # j−1
ηwf,1 = hwf,1 , ηj = hwf,j + hwf,n
2 2 n=1
Figure 6 plots these the pressure ratio against distance along the wavefront, with
the abscissa shifted such that η = 0 at the centerline. The number of rays for this graph
was raised to 121 to give the graphs a less jagged appearance. The wavefront that
was specified contains the point x = 0.625 m, y = 0. Its equation is τ = 2.0681 ms.
The pressure distribution along two other wavefronts also appear in this graph. The
point x = 0.5 m, y = 0 lies on the wavefront τ = 1.6633 ms. This wavefront passes
through the cooled region. The third wavefront is τ = 2.4469 ms. It contains the point
x = 0.75 m, y = 0, which places it slightly to the left of the arrete. Reference to Fig. 3
shows that the intervals where |P| /PI is less than one correspond to regions where the
rays are sparse, while the rays are closely spaced in regions where the pressure ratio
is enhanced. This enhancement becomes more prominent as the wavefront comes
closer to the caustic.
4
3  = 1.6633 ms
|P()|/PI
 = 2.0681 ms
2  = 2.4469 ms

−0.5 0 0.5
Distance  along the wavefront
Figure 6.
Evaluation of the Signal at a Field Point – Eigenrays

The operations described thus far entail evaluating ray paths by solving the ray tracing
equations. The pressure along a ray is then determined by solving the transport
equation. In this procedure, we cannot specify a priori the points where the ray falls.
This leads to the question of how can we find the pressure at a specific point? Stated
differently, the question is how can we determine the ray path that connects a source
and a specific field point?
The situation is depicted in Fig. 11.19 where the source is at ξ¯I , and the receiver’s
location is the designated field point ξ¯f . Only direct and surface-reflected rays are
considered. It is permissible to ignore bottom-reflected rays if the water is very deep
compared to the depths of the source and field point, because signals that propagate
along bottom-reflected rays to ξ¯f will have traveled over very large arclengths. Unless
a caustic occurs, the ray tube area of this set of rays will be greatly enlarged, so the
amplitude will be greatly reduced. (Bottom reflections are an important aspect of
propagation in a shallow channel.)
Rf
zf
I
f Ray n
Ray n+1
Direct
Surface
reflected
Eigenray
Fig. 11.19 Identification of eigenray connecting a point source and a selected field point
Contributors to the signal at ξ¯f are a ray that arrives directly from the source, and
a ray that arrives after it is reflected from the free surface. Both are referred to as
eigenrays, which is the term given generally to a ray that connects the source and a
selected field point. Solution of the ray tracing equations presumably has given the
two families of rays in Fig. 11.19, where each ray corresponds to a different grazing
angle at the source. How can we identify the eigenrays given that none of the rays in
the Figure actually intersects ξ¯f ? This is not a trivial questions, and it is much more
difficult if some rays undergo many surface and bottom reflections.
The simplest approach is to use interpolation. To do so, we first identify the pair
of direct rays n and n + 1 that bracket ξ¯f , that is, the rays that come closest to ξ¯f above
and below. Let Rf be the horizontal range to ξ¯f , and let zn and zn+1 be the depths of
the points on the respective rays that are at Rf . A linear interpolation of the initial
grazing angles gives
(χI )f − (χI )n (χI )n+1 − (χI )n
= (11.3.35)
zf − zn zn+1 − zn
The value of (χI )f is used to find the eigenray. To do so, the ray tracing equa-
tions are solved with (χI )f as the initial grazing angle. The eigenray for surface
reflection is found by repeating these steps using the rays in the reflected set that
bracket ξ¯f .
The total pressure at ξ¯f is the sum of the contributions from all eigenrays connect-
ing the source and field points. In the case of Fig. 11.19, let τd and τr be the values
of the wavefront functions for the direct and reflected eigenrays, respectively. Also,
let Pd and Pr be the corresponding amplitudes. Both amplitudes are obtained by
evaluating the cross-sectional areas of the respective ray tubes, with the appropriate
corrections for reflection coefficients and phase delays at caustics. Then the total
pressure at ξ¯f is
p ξ¯f , t = Re Pd eiω(t−τd ) + Pr eiω(t−τr ) (11.3.36)
If more than two eigenrays intersect the selected field point, than the contribution of
each is additive.
EXAMPLE 11.6 A radially symmetric point source is situated at depth H

below the free surface of a deep body of water. Density and sound speed
variations are negligible. Use geometrical acoustics to derive an expression for
the pressure at range RF and depth zF . Sound speed variations over this range
are negligible. Compare that result to the result of an analysis that uses the
method of images.
Significance
The main feature of the analysis is that it explains how the Jacobian is used to
evaluate analytically the ray tube area. The juxtaposition of geometrical acoustics
and the method of images illustrates their different perspectives.
Solution
The fluid is homogeneous, so all rays are straight lines. There are only two eigen-
rays because reflections from the bottom are deemed to be negligible. The direct
eigenray is the line from the source to the field point, and the reflected eigenray
intersects the surface somewhere between the source and the receiver. The arrange-
ment is described in Fig. 1.
RF
s
A B C
 
r l1 r
Reflected
H
 eigenray zF
r
l2
I ld
d Direct eigenray
F
Figure 1.
A geometrical acoustics analysis begins with determination of the ray paths.

Because they are formed from straight lines, we may use geometry for the determina-
tion, rather than solving the ray tracing equations. The angle of incidence equals the
angle of reflection. Consequently, right triangles IAB and BCF are similar. Equating
the ratio of sides yields an expression for the horizontal distance s to the reflection
point,
s RF − s HRF
= =⇒ s =
H zF zF + H
We must evaluate ray tube areas for each eigenray. We could do so by considering
each eigenray to be one limit of its ray tube with rays at initial grazing angles
χd + dχ and χr + dχ as the other limit. However, doing so would be tedious. An
analytical approach employs the Jacobian in Eq. (11.3.25). This formulation requires
description of the (R, z) coordinates of a field in terms of the initial grazing angle
for the ray path through that point and the distance along that ray path to the field
point. The source depth H is fixed in that evaluation
For the direct path, the arclength is d , and the initial grazing angle is χd . In terms
of these variables, the coordinates of the field point F are
RF = d cos χd , zF = H + d sin χd
We differentiate these expressions with respect to χd , and substitute the results into
Eq. (11.3.25). This gives

1/2
∂RF 2 ∂zF 2
dAd = RF + dχd dθd = (d )2 (cos χd ) dχd dθ (1)
∂χd ∂χd
The presence of a reflection complicates the evaluation of ray tube area for reflected
rays. We seek a description of RF and zF in terms of χr and the total arclength along
the reflected ray. Figure 1 indicates that this arclength is r = 1 + 2 . Because H is
fixed, the distance 1 depends solely on χr . Using this relation to form 2 = r − 1
gives 2 as a function of χr and r . We employ this relation to describe RF and zF
for the reflected ray,
H H
1 = , 2 = r − 1 = r −
sin χr sin χr
RF = 1 cos χr + 2 cos χr = r cos χr
zF = 2 sin χr = r sin χr − H
The ray tube area obtained from Eq. (11.3.25) is

1/2
∂RF 2 ∂zF 2
dAr = RF + dχI dθI
∂χr ∂χr (2)
1/2
= RF (r sin χr )2 + (r cos χr )2 dχr dθ = (r )2 (cos χr ) dχr dθ
Equation (11.3.22) equates twice the intensity at an arbitrary location to the value
at the start of the ray tube. In the present situation, the input to both ray tubes is a
radially symmetric point source. The time-averaged intensity is the source’s radiated
power divided by the surface area of a surrounding sphere. For both ray tubes,
this sphere has a very small radius ε. Therefore, for both the direct and reflected
eigenvalues the starting value is

P2 2P
= (3)
ρ0 c I 4πε2
The cross-sectional area for both ray tubes at the input end consists of a small patch
of the surrounding sphere at polar angle π/2 − χ covering a polar angle dχ and
azimuthal angle dθ. Therefore, we have
(dAI )d = (εdχd ) (ε cos χd ) dθ, (dAI )r = (εdχr ) (ε cos χr ) dθ (4)
Application of Eq. (11.3.22) in the case where the input to a ray tube is a radially
symmetric point source leads to

Pd2 P2 P2 P2
dAd = (dAI )d and r dAr = (dAI )r
ρ0 c ρ0 c I ρ0 c ρ0 c I
Substitution of Eqs. (1)–(4) into these relations gives

Pd2 2 2P
(cos χd ) dχd dθ = ε2 (cos χd ) dχd dθ
ρ0 c d 4πε2

Pr2 2P
(r )2 (cos χr ) dχr dθ = ε2 (cos χr ) dχr dθ
ρ0 c 4πε
2
1 ρ0 cP 1/2
Pd =
d 2π
From this we find that the pressure amplitudes are

1/2 1/2
1 ρ0 cP 1 ρ0 cP
Pd = , Pr = (5)
d 2π r 2π
For a wave that departs from the source at zero phase and propagates at constant
phase speed c along a ray, the phase variable is τ (x̄) = /c. The reflection coefficient
at the free surface may be taken to be −1. Adding the signal received at the field
point from each eigenray therefore gives
1/2
p ξ¯f , t = Re Pd eiω(t−d /c) + Pr (−1) eiω(t−r /c)
1/2 −iωd /c
cP e e−iωr /c iωt
= Re − e
2π d r
Evaluation of the pressure by the method of images was addressed in Chap. 6.

The source 1 and its image 2 are depicted in Fig. 2. Because the free surface is
pressure-release, the strength of the image is the negative of the source’s strength. It
is evident that the radial distance r1 from the source to the field point equals d and
that r2 = 1 + 2 = r .
2
r2
H l1
r zF
r1=ld l2
1

d
RF F
Figure 2.
The complex pressure amplitude radiated by a point source in a homogenous fluid

is [ρ0 cP/ (2π)]1/2 exp (−ikr) /r, where P is the time-averaged power. The pressure
received at the field point from the image is inverted, so the combination of the source
and image pressures is
1/2
ρ0 cP e−ikr1 e−ikr2
¯ t =
p ξ, Re − iωt
e
4π r1 r2
Because r1 = d and r2 = r , this is the same result as that obtained from ray theory.
Although the analysis using the method of images is much easier, seeing how ray
theory is implemented, particularly the analytical evaluation of ray tube area, makes
some of the abstract concepts more understandable.
11.4 Fermat’s Principle 463
11.4 Fermat’s Principle
The developments thus far rest on Snell’s law, which ultimately is founded on the
basic principles that underlie other acoustic phenomena. A different formulation
for analyzing ray paths is available. Fermat stated his principle in 1662 without a
mathematical derivation. It postulates that the ray connecting two points is such that
the path minimizes the travel time between the points. We will see that it leads to
Snell’s law, as well as a differential equation that is equivalent to the ray tracing
equations. However, that is not the reason we study it. Rather it offers a different
perspective regarding the nature of rays.
It is possible to justify Fermat’s principle with a heuristic argument based on an
analogy. This might be the thought process that led Fermat to state the principle.
Suppose a large number of people have gathered at the base of a mountain. Each
individual is made to follow a different path as they travel to a specified point at
the other side of the mountain. Some paths might be easier to walk, so individuals
following such paths can travel more quickly. However, these fast paths might be
long. Other paths might be much shorter, thereby counterbalancing the fact that those
paths are difficult to traverse. If we consider when individuals arrive at the destination,
the first arrival will be the individual who followed the optimum path. Individuals
who follow paths that are very close to the optimum will arrive slightly later, whereas
individuals who follow nonoptimum paths will arrive at a wide range of later times.
In the acoustical analogy, signals that follow the optimum and near optimum rays
will arrive nearly synchronously, thereby reinforcing each other. Signals following
nonoptimal paths will be unsynchronized, and therefore destructively interfere with
each other.
Any derivation of basic principles must start from a set of postulates. We could
begin with the ray tracing equations as the foundation for a derivation of Fermat’s
principle. Instead, we will derive the ray tracing equations from it. One benefit of
proceeding in this manner is that it will introduce a general mathematical tool called
the calculus of variations, which is very useful for a wide range of subjects, especially
classical and relativistic mechanics.
In Fig. 11.20 several paths connect selected points A and B. One of these paths
is optimal in the sense that the time for a signal to travel along it is minimal. Some
other paths are far from optimal, but one path in the Figure is very close. This is
a variational path. The general approach is to compare the travel time along the
optimal path to the time along a variational path. The variational path is defined to be
separated from the optimal path by an infinitesimal difference δ ξ¯ of their positions.
More specifically, if x̄ = ξ¯ is a point on the optimal path, then the corresponding
point on the variational path is placed at x̄ = ξ + δξ. The reason for taking δ ξ¯ to be
infinitesimal is that doing so allows us to truncate a Taylor series at the second term.
Although we do not know the optimal path, that is, the eigenray, let us pretend
that we do. The minimum travel time along this ray between points A and B may be
found by evaluating
Fig. 11.20 The optimal ray B

path corresponds to the
minimum time between two
designated points A and B Nonoptimal path
_ _
_ _
x 
_ _ _
Optimal path x   
_
Variational path 
A
Nonoptimal path
B
d
Tmin =C (11.4.1)
A c ξ¯
The differential arclength d is the distance between two adjacent points on the opti-
mal path. We shall use the symbol “d ” to indicate differential increments pertaining
to the same path. Thus, the positions of the adjacent points are ξ¯ and ξ¯ +d ξ,
¯ as shown
in Fig. 11.21. It follows that
1/2
d = d ξ¯ · d ξ¯ (11.4.2)
The variational path is obtained by shifting from position x̄ = ξ¯ to position x̄ =

ξ¯ + δ ξ.
¯ This corresponds to infinitesimal amount δ ξ. ¯ The symbol δ is used to indicate
that the increment is infinitesimal, but it pertains to a difference between a quantity on
the optimal and variational paths. Thus, any quantity that contains the symbol δ is not
the result of a physical process. For this reason, it is said to be a virtual increment.
Thus, δ ξ¯ is a virtual displacement, whereas d ξ¯ is an actual displacement. Other
than being required to be infinitesimal, the direction and magnitude of the virtual
displacement are arbitrary. As shown in Fig. 11.21, d ξ¯ is tangent to the optimal path,
whereas δ ξ¯ may be in any direction. Because δ ξ¯ is arbitrary, there is no implication
that the displaced point, whose position is x̄ = ξ¯ + δ ξ, ¯ is on the same wavefront as
x̄ = ξ.¯
Fig. 11.21 Relation between _ _

a displacement d ξ̄ along _ _ _  ( d)
x d
minimum-time path and a _ _ _
variational displacement δ ξ̄ x  d _ _
to a neighboring sub-optimal d ()
path _
 _ _ _
x   
We need to compare the propagation time in Eq. (11.4.1) to the time for the
variational path. This requires a description of the differential arclength along the
variational path, which in turn requires that we describe the position of the displaced
point on the variational path. Figure 11.21 shows that this position may be regarded
as the result of a virtual displacement from the point ξ¯ + d ξ¯ on the optimal path,
or the result of an actual displacement from point ξ¯ + δ ξ¯ on the variational path.
Both constructions must lead to the same point. An equivalent statement is that the
same new position must be obtained, regardless of whether the virtual displacement
precedes or follows the actual displacement. Thus, it must be that

ξ¯ + d ξ¯ + δ ξ¯ + d ξ¯ = ξ¯ + δ ξ¯ + d ξ¯ + δ ξ¯ =⇒ δ d ξ¯ = d δ ξ¯
(11.4.3)
This property will soon be found to be quite useful.

The arclength along the variational path is the magnitude of the differential

increase in the position of a point on the variational path, that is, d = d ξ¯ + δ ξ¯ .
A dot product may be used to construct this distance. Higher order differentials are
negligible, so we have
1/2
d = d ξ¯ + δ d ξ¯ · d ξ¯ + δ d ξ¯
1/2 1/2
= d ξ¯ · d ξ¯ + 2d ξ¯ · δ d ξ¯ ≡ (d)2 + 2d ξ¯ · δ d ξ¯ (11.4.4)
d ξ¯ ¯
= d + · δ dξ
d
We use this expression to construct the propagation
time along the variational
path. The sound speed along the variational path is c ξ¯ + δ ξ¯ , so the time required
for a signal to travel along this path is
B
1 d ξ¯ ¯
Tvar =C d + · δ dξ (11.4.5)
A c ξ¯ + δ ξ¯ d
It is easier to proceed if we replace the sound speed with its reciprocal
1
U ξ¯ ≡ (11.4.6)
c ξ¯
(Some treatments use the index of refractionμ in Eq. (11.2.2), but that would require
the unnecessary introduction of a reference sound speed.)
At this juncture, we call on Fermat’s principle. According to it, the travel time
along the variational path can differ at the most from the minimal time by second-
order differentials. Therefore, we express Tvar in terms of the lowest order differentials
that distinguish it from Tmin in Eq. (11.4.1). A Taylor series is used to treat the
integrand, so that
B

¯ ¯
d ξ¯ ¯
Tvar = C U ξ + δ ξ d + · δ dξ
d
A
B
d ¯
ξ
= C U ξ¯ + δ ξ¯ · ∇U ξ¯ d + · δ d ξ¯ (11.4.7)
d
B
A
B
d ξ¯
¯
= C U ξ d + C U ξ ¯ ¯ ¯ ¯
· δ d ξ + δ ξ · ∇U ξ d
A A d
The first integral is Tmin . Thus, the statement that the lowest order representation of
Tvar must be the same as Tmin leads to the condition that
B
d ξ¯
C U ξ ¯ ¯ ¯ ¯
· δ d ξ + δ ξ · ∇U ξ d = 0 (11.4.8)
A d
The ray tracing differential equations consider the position of a point on a ray
to be a function of time, but none of the terms in the integral contains time. This
suggests an alternative description of position on a ray path, in which it is defined
by the arclength to a point. In other words, we consider to be the independent
variable, so that the position of a point on the optimal path is described functionally
as x̄ = ξ¯ (). This is where Eq. (11.4.3) enters, because we can assert that

d ¯
δ d ξ¯ = d δ ξ¯ = δ ξ d (11.4.9)
d
This step converts the statement of Fermat’s principle in Eq. (11.4.8) to an integral
over , specifically,
B
d ξ¯ d
C U ξ¯ · δ ξ¯ + δ ξ¯ · ∇U d = 0 (11.4.10)
A d d
The last operation is based on the fact that although the dependence
of δ ξ¯ on is
arbitrary, it is a function that we select. Its selection sets d δ ξ¯ /d, so this derivative
is not an independent quantity. It may be eliminated with an integration by parts,
which gives
B
d ξ¯ B
d d ξ¯
¯
U ξ ¯
· δξ − C ¯
U ξ − ∇U · δ ξ¯ d = 0 (11.4.11)
d A d d
A
At this juncture, we take an overview. It was stated that δ ξ¯ is an arbitrary function

of , but that is not exactly true. The variational path is an alternative between
point A and B, so the variational path must intersect these points. This means that
δ ξ¯ (A ) = δ ξ¯ (B ) = 0. The consequence is that the terms in the preceding relation
that are evaluated at the ends of the path are zero. In turn, this requires that the integral
must evaluate to zero for any δ ξ¯ function we select. If we wish that the integral of
ā · b̄ be zero for any function b̄, it must be that ā is identically zero. Thus, we find

d d ξ¯
U ξ¯ − ∇U = 0 (11.4.12)
d d
Minimization of integral quantities arise in a variety of areas, notably analyti-

cal dynamics. The mathematical process we have followed is known as the calcu-
lus of variations. Equation (11.4.12) in a general analysis is known as the Euler-
Lagrange equation. It is a second-order differential equation governing ξ, ¯ which is
a fundamental difference with the (first order) ray tracing equations. Another funda-
mental difference is the absence of the wave slowness s̄.
Our intent at the outset was to derive the ray tracing equations from Fermat’s prin-
ciple, but the appearance of Eq. (11.4.12) does not suggest how to meet that objective.
Let us try to work backward, by eliminating s̄ from the ray tracing equations. The
second ray tracing equation, Eq. (11.3.10), states that
d ξ¯
= c2 s̄ (11.4.13)
dt
A different view is that ξ¯ depends on the arclength

, which
leads to a description
¯
of the preceding derivative according to d ξ/dt ¯
= d ξ/d (d/dt). The local sound
speed is the rate at which the arclength changes, so that
d
c= (11.4.14)
dt
Matching the two descriptions of the rate of change of the position on a ray leads to
d ξ¯ d ξ¯ d d ξ¯
= ≡ c = c2 s̄ (11.4.15)
dt d dt d
Replacing c with 1/U thereby converts the second ray tracing equation to
d ξ¯
s̄ = U (11.4.16)
d
Now we turn to the first ray tracing equation, Eq. (11.3.9), which is
ds̄ ∇c
=− (11.4.17)
dt c
On the right side, we set c = 1/U , so that

∇c 1 1
=U∇ =− ∇U (11.4.18)
c U U
We introduce the chain rule to handle ds̄/dt on the left side of Eq. (11.4.17) because
our view now is that s̄ is a function of . The result is that the first ray tracing equation
becomes
ds̄ d ds̄ 1
≡ c = ∇U (11.4.19)
d dt d U
Replacement of c with 1/U in this equation leads to
ds̄
= ∇U (11.4.20)
d
The last step is to substitute s̄ from Eq. (11.4.16) into this expression. The result is
the Euler-Lagrange equation, Eq. (11.4.12). In other words, the ray tracing equations
and the Euler-Lagrange equation are equivalent specifications of a ray path. The
differences are that the ray tracing equations give position as a function of time, with
additional information about the tangent to the path in the form of s̄. If we wish to
also determine the arclength to a point on a ray, we can do so by adding to the set
of unknowns {X} in Eq. (11.3.12)
and inserting
into the scalar differential equations
another equation: d/dt = c ξx , ξy , ξz . The Euler-Lagrange equation describes the
position in terms of . If we wish to know a local tangent or the propagation time,
we must extract that information after the equation has been solved.
Some mathematical analyses are best performed with a single differential equa-
tion. However, realistic situations usually require numerical methods. In that case,
the ray tracing equations are preferable. The reason for this assertion is that most
numerical algorithms require that the equations be first order, which is how the ray
tracing equations are posed. Of course, the Euler-Lagrange equation could be trans-
formed into first-order form, but as we have seen, such equations would be the ray
tracing equations or a variant of them.
We have demonstrated that Fermat’s principle correctly describes the rays in a
fluid whose sound speed is a continuous function of position. It also is valid at
discontinuities. To demonstrate this, consider Fig. 11.22. We wish to identify the
horizontal distance xO where the ray from A to B crosses the interface between two
fluids. Points A and B are fixed. A coordinate system whose z-axis measures depth
into the second fluid is situated at the horizontal position of A.
Fig. 11.22 Construction of a

A
ray path for transmission I
across an interface between c1
1 O
two fluids x
c2
T 2
B
z
The travel times along each segment of the ray are 1 /c1 and 2 /c2 . These distances
are
1/2 1/2
1 = zA2 + (xO )2 , 2 = zB2 + (xB − xO )2 (11.4.21)
We seek the value of xO that minimizes the sum of the travel time along each segment.
This requires that
1/2 1/2
d 1 2 d zA2 + (xO )2 d zB2 + (xB − xO )2
+ = +
dxO c1 c2 dxO c1 dxO c2 (11.4.22)
xO (xB − xO )
= − =0
c1 1 c2 2
According to Fig. 11.22, sin ψ1 = xO /1 and sin ψ2 = (xB − x0 ) /2 , so the preceding
is a restatement of Snell’s law.
There is a degree of elegance in having a general principle that leads to the
equations for ray paths in inhomogeneous fluids as well as at interfaces. Indeed,
the generality of Fermat’s principle probably is its most important attribute. For
example, it can been used to derive equations describing the rays in the presence of
an ambient flow.11 It also can link geometrical acoustics to optics and the Hamiltonian
formulation of analytical mechanics.
EXAMPLE 11.7 A point source is located at position A in an ocean channel

and a hydrophone is situated at point B. The sound speeds are c1 for the water,
c2 < c1 for the air, and c3 > c1 for the sediment. Describe the eigenrays for
point B.
Air
dA dB
H A Water B
*
Sediment
Figure 1.
Significance
This application of Fermat’s principle to the determination of eigenrays is quite
straightforward, but the results have a surprising feature. Thus, this example affirms
that the perspective offered by Fermat’s principle might cause us to re-examine our
initial assumptions.
11 O. Bühler, Waves and Mean Flows, Cambridge University Press (2009) pp. 81–83.
Solution
We begin with a sketch of possible paths between points A and B. Paths 1, 2, and 3
in Fig. 2 are familiar, but path 4 might be considered to be unexpected. In addition to
these, others exist, but we will consider them later. Path 1 is the direct path. Because
the sound speed along it is c1 , the minimum travel time is obtained for the shortest
path. Thus, Fermat’s principle confirms that this path is a straight line.
c2 RC R-R C
dA A2 2 C B2
c1 dB
1
H A
4
A4
3
D B4
*B
c3 E F
Figure 2.
Paths 2 and 3 feature a single reflection. Let us determine whether Fermat’s prin-
ciple yields ray paths for which the angle of reflection equals the angle of incidence.
In the case of the surface-reflected wave, points A and B are fixed and we know that
the elevation of point C at which it is incident. This ray path is set if the horizontal
location of point C is specified. Therefore, we select the horizontal range RC to this
point as the unknown to be determined by Fermat’s principle. Trigonometric equa-
tions describe the length of the two straight segments that form path 2. Division of
those lengths by the sound speed gives the travel time along each. Fermat’s principle
applies to the sum of these contributions, which is
2 1/2 2 1/2
dA + RC2 dB + (R − RC )2
T2 (RC ) = +
c1 c1
Fermat’s principle states that the correct value of RC is that which gives dT2 /dRC = 0.
Evaluation of the derivative gives
dT2 1 Rc 1 (R − RC )
= − 2 =0 (1)
dRC c1 d 2 + R2 1/2 c1 d + (R − RC )2 1/2
A C B
The length ratios in Eq. (1) are the cosines of the respective grazing angle depicted
in the sketch. Hence, Eq. (1) reduces to
cos χ2,A cos χ2,B
− =0
c1 c1
Cancelation of c1 shows that the grazing angles are equal. Thus, Fermat’s principle
is consistent with the familiar reflection law. There is no need to repeat this analysis
for path 3, because the arrangement is the same as path 2.
Path 4 involves two fluids. It consists of a straight segment from the source to
point E on the interface, then a horizontal path in the sediment to point F, from
which a straight path brings the signal to point B. The location of points E and F
each depend on the horizontal range, so this path is a function of two parameters. For
the sake of variety, let us use the grazing angles χA,4 and χB,4 . The corresponding
expression for the travel time is

1 H − dA 1 H − dA H − dB 1 H − dB
T4 χA,4 , χB,4 = + R− − +
c1 sin χA,4 c3 tan χA,4 tan χB,4 c1 sin χB,4
The value of T4 must be stationary with respect to both grazing angles, so we set

∂T4 H − dA − cos χA,4 H − dA (−1)
= 2 − 2 = 0
∂χA,4 c1 sin χA,4 c 3 sin χA,4
(2)
∂T4 H − dB (−1) H − dB − cos χB,4
=− 2 + 2 = 0
∂χB,4 c2 sin χB,4 c3 sin χB,4
These expressions reduce to

c1
cos χA,4 = = cos χB,4
c3
The angles of the line segments relative to the normal to the interface are ψA,4 =
π/2 − χA,4 and ψB,4 = π/2 − χB,4 , so Eq. (2) leads to
c1
sin ψA,4 = sin ψB,4 = (3)
c3
This is the expression for the critical angle of incidence. It is a real value only if
the sound speed in the receiving fluid is greater than the speed in the transmitting
fluid. Thus, Fermat’s principle tells us that the alternative path exists if c3 for the
sediment is greater than c1 for the water. If that is the case, then the ray arrives and
departs from the interface at the critical angle. One way to view this is that the source
at point A emits many rays. Rays that arrive at the sediment at an angle of incidence
less than the critical value are reflected upward and transmitted downward. However,
the ray that arrives at the critical angle of incidence generates a transmitted ray that
is parallel to the interface. This ray runs along the interface on the sediment side. The
water now becomes the receiving fluid. The ray emerges from the sediment when the
angle from the interface to the receiving point is the critical angle for transmission
from the fluid into the sediment.
Paths 2 and 3 entail a single reflection. Multiple reflection paths also are possible.
A few are shown in Fig. 3. In general, there the two paths that feature N reflec-
tions, with the their difference being whether the first reflection is at the top or
bottom. We will analyze ray 5, from which the procedure for other cases should be
apparent.
c2 RC R-R C
dA c1 C
5 dB
H A
6
D
*B
c3 R-R D
RD
Figure 3.
The reflection points are defined by the ranges RC and RD , so the travel time is a
function of two parameters,
1/2 2 1/2
(dA )2 + (RC )2 H + (RD − RC )2
T5 (RC , RD ) = +
c1 c1
1/2
(H − dB ) + (R − RD )2
2
+
c1
For an extreme value, the derivative of T5 with respect to both parameters must be
zero,
∂T5 1 RC 1 (RD − RC )
= − =0
∂RC c1 (dA )2 + (RC )2 1/2 c1 H 2 + (RD − RC )2 1/2
(4)
∂T5 1 (RD − RC ) 1 (R − RD )
= 1/2 − =0
∂RD c1 H 2 + (RD − RC ) 2 c1 (H − dB ) + (R − RD )2 1/2
These equations may be solved for RC and RD . To interpret these equations we

observe that the grazing angle of the segments of ray 5 in the Figure are

−1 RC
χAC = cos 1/2
(dA )2 + (RC )2

−1 RD − RC
χCD = cos 1/2 (11.4.23)
H 2 + (RD − RC )2

R − RD
χAC = cos−1 1/2
(H − dB )2 + (R − RD )2
Thus, the values of RC and RD that extremize T5 according to Eq. (4) correspond to
equal grazing angles for the arriving and departing rays at points C and D. In other
words, the angle of reflection equals the angle of incidence at points C and D. There
is nothing new in this.
The fact that Fermat’s principle identifies the critical angle path through the sed-
iment is another demonstration of its generality. One of the paradoxes of this path is
that the concept of a ray tube having an identifiable cross-sectional area breaks down.
This is so because grazing angles less than the critical value in Eq. (3) do not lead to
real rays, while grazing angles greater than critical lead to rays that are transmitted
into the sediment, and therefore cannot reach the vicinity of point B. This is one of
many phenomena that do not fit the geometric acoustic model. A few others are the
fact that a sound is heard in a shadow zone, and that sound is heard behind a rigid
barrier. Phenomena that do not fit into the geometrical acoustic model are referred
to as diffraction. Analyses of diffraction phenomena generally require a level of
mathematics that is beyond the scope of this book.
Exercise 11.1 The mirror in the sketch is a surface of revolution z = f (R), with
f (R) increasing monotonically. A plane wave propagating in the negative z direction
is reflected. Consider a ray that is incident at arbitrary R. The angle of reflection
equals the angle of incidence, so tan ψ = df /dR. Derive an expression in terms of
R, f (r), and df /dR for the coordinate zf where the reflection of this ray intersects
the z-axis. Then consider the specific case of a paraboloid of revolution, for which
f (R) = hz (R/a)2 . Show that zf is a constant in this case, which means that all
reflected rays intersect at a true focus.
f(R)
 _
n

R
z
zf
Exercise 11.1
Exercise 11.2 The cross section of the prism in the sketch is an isosceles triangle
with a vertical axis of symmetry. Assume that the prism is a fluid with sound speed
cb , and set ca as the sound speed in the surrounding fluid. (a) Derive an expression for
the angle of transmission ψ for rays that emerge from the prism. (b) If ca = 200 m/s
and cb = 400 m/s, what is the largest value of φ for which a plane wave emerges
from the prism? (b) If ca = 400 m/s and cb = 200 m/s, what is the largest value
of φ for which a plane wave emerges from the prism? (c) It is given that ca = cb .
Qualitatively explain why there is no value of the apex angle 2φ > 0 for which the
plane wave that emerges propagates horizontally, regardless of whether ca > cb or
ca < cb .
ca
ca 2
cb
Exercise 11.2
Exercise 11.3 The reflection from a flat surface of the signal radiated by a point
source is describable by the method of images if the surface is rigid or pressure-
release. In the case of a locally reacting surface, geometrical acoustics also leads to
the conclusion that the reflected rays seem to emanate from a point source below the
surface. It also leads to the conclusion that the image source has a directivity factor.
Consider an omnidirectional point source situated at height H above a surface whose
local impedance is Z. The source is harmonic at frequency ω. Determine the location
of the image and the directivity of that image. Does the field of the image source
actually satisfy the Helmholtz equation in the actual fluid?
Exercise 11.4 The sound speed profile in a very deep ocean channel consists of a
lower layer of depth H in which the sound speed is constant at c1 , and an upper layer in
which the sound speed decreases linearly with depth. The sound speed is continuous
across the interface of these layers. It is convenient to describe c in terms
of the
height z above the bottom. Thus, the sound speed is c = c1 + m z − H h z − H ,
with m > 0 and z = 0 at the bottom. Any ray path within the upper layer must be
a downward curving circular arc. Consequently, a ray departing from the bottom at
any upward grazing angle must eventually return to the bottom. Derive an expression
for the distance between the points of departure and return.
Exercise 11.5 At the surface of an ocean channel, the sound speed is 1470 m/s. It
decreases linearly with depth, down to 500 m, where the sound speed is 1420 m/s.
Downward from that depth the sound speed increases linearly, with a value of
1470 m/s occurring again at 1000 m. Determine the path of a ray launched hori-
zontally at a depth of 200 m. Base the analysis on the fact the path is circular if the
sound speed is linearly dependent on the depth.
Exercise 11.6 External heating of a waveguide has resulted in a sound speed that
increases linearly from axial position x = 0 to position x = L, but the sound speed
is constant transversely. The result is that c = c0 (1 − x/L) + c1 x/L. A point source
is situated in the middle of the waveguide, at x = 0, y = H/2, where H = 2 m
is the waveguide’s width. Consider rays launched at grazing angle of 15◦ , 45◦ , and
75◦ . (Note that the heterogeneity varies in the x direction, so the grazing angle are
measured relative to the y direction.) Do any of these rays intersect a wall? If so,
what is the location of this intersection, and what is the propagation time from the
instant the ray is launched to the instant when it arrives at that wall? Parameters are
c0 = 300 m/s, c1 = 400 m/s, H = 3 m.
Exercise 11.7 The sound speed in a hard-walled two-dimensional HVAC duct varies
parabolically according to c (z) = c0 + c1 (1 − 2z/H)2 , where z is the transverse
distance. The walls are at y = 0 and y = H. Determine the path of two rays that are
launched downward from z = H/2 : χIa is such that a wall is a turning point, and
χIb = 0.5χIa . Parameters are c0 = 300 m/s, c1 = 50 m/s, H = 2 km.
Exercise 11.8 The sound speed in a hard-walled two-dimensional HVAC duct varies
parabolically according to c (z) = c0 + c1 (1 − 2z/H)2 , where z is the transverse dis-
tance. The walls are at z = 0 and z = H. Determine the paths of rays launched down-
ward from z = H/2 resulting from an initial grazing angle that is 1.25χIw , where
χIw is such that a wall is a turning point. Parameters are c0 = 300 m/s, c1 = 50 m/s,
H = 2 km.
Exercise 11.9 The sound speed in a large body of water varies sinusoidally with
depth relative to a mean value, c (z) = c0 + ε sin (2πz/H). A radially symmetric
point source is situated at z0 = 10H. This depth is sufficiently large that there is
no need to consider reflections from the surface or bottom. (a) Consider a ray that
is launched horizontally. Does it have a turning point? If so, at what depth? (b)
Determine the path followed by this ray. (c) Determine the travel time along this ray.
Plot this property as a function of the horizontal distance from the source, and also
as a function of the depth. Parameters are c0 = 1500 m/s, ε = 25 m/s, H = 100 m.
Exercise 11.10 The sound speed in a large body of water varies sinusoidally with
depth relative to a mean value, c (z) = c0 + ε sin (2πz/H). A radially symmetric
point source is situated at z0 = 10H. This depth is sufficiently large that there is no
need to consider reflections from the surface or bottom. Determine the paths of rays
launched at initial grazing angles ranging from −45◦ to 45◦ in a 1◦ increment.
Exercise 11.11 Consider the waveguide in Exercise 11.10. (a) Evaluate the paths of
rays that are launched from the source at grazing angles of 2.99◦ and 3.01◦ . (b) Use
the result of Part (a) to evaluate the ray tube area corresponding to rays that depart
from in the source range 2.99◦ ≤ χ ≤ 3.01◦ in an interval of azimuthal angles
subtending 0.02◦ . (c) Determine the intensity in this ray tube at a horizontal distance
of 500 m from the source.
Exercise 11.12 Evaluate the ray tube area A for two incident rays in Exercise 11.2
that are infinitesimally separated by elevation dh. From that expression determine
the amplitude |PT | of the wave that emerges from the prism if the incident wave was
harmonic with amplitude |PI |. Graph dA/dh and |PT | / |PI | as functions of θ for two
cases: cb /ca = 2, ρb /ρa = 1.2, and cb /ca = 0.5, ρb /ρa = 0.8.
Exercise 11.13 A radially symmetric source is situated at depth z0 in a vertically

stratified fluid. The sound speed is an arbitrary function c (z), where z is the depth.
It is desired to determine the intensity within a ray tube that surrounds the vertical
ray path. The first step in the analysis considers a ray that departs from the source
slightly off vertical, so the initial grazing angle is χI = π/2 − , with 1.
(a) Show that the slope of this ray at an arbitrary depth is given, to leading order in
powers of , by
dR c (z)
≈
dz cI
where R is the transverse distance from the z-axis. (b) This approximation is valid at
any depth for which c (z) /cI is O (1). Subject to this limitation derive an expression
for the cross section area of the ray tube that emanates from the source as a cone
whose apex angle is 2. Then use that area to determine the depth dependence of
the time-averaged intensity along the downward ray.
Exercise 11.14 Consider the waveguide in Exercise 11.6. The task here is to deter-
mine the pressure at the wall when the radially symmetric source, which is situated
at x = 0, y = H/2, is time-harmonic, with time-averaged power output P. Derive an
expression for the distance xf at which a ray that departs from the source at grazing
angle χI intersects a wall. Then use that expression to determine the pressure ampli-
tude at the location where rays intersect the wall. Consider initial grazing angles of
χI = 0◦ , 30◦ , and 60◦ . (Note that the heterogeneity varies in the x direction, so the
grazing angle are measured relative to the y direction.)
Exercise 11.15 Suppose that the reference sound speeds in Example 11.5 are
reversed, so that c0 = 280 m/s, c1 = 330 m/s. All other parameters are as stated
there. Compute the ray paths and wavefronts. Do caustics form? Are there shadow
zones?
Exercise 11.16 Due to extraordinary circumstances the sound speed in the atmo-
sphere has been found to depend on the horizontal distance x and altitude y measured
from a point that is 1000 m above the ground. The dependence c = 350+xy/200 m/s,
where the units of x and y are meters. A weather balloon transmits an omnidirectional
point source from that location. (a) Compute the paths of rays that are launched in the
xy plane in all directions covering the full 360◦ range in 10◦ increments. (b) Determine
the wavefronts corresponding to the rays in Part (a).
Exercise 11.17 In the region x < 0, a fluid is homogeneous, with sound speed c0 .
To the right, x > 0, the sound speed varies sinusoidally in both the x and y directions,
according to c = c0 − sin (πx/L) cos (πy/H) for 0 < x < L. For x > L the sound
speed returns to c = c0 . The parameters are c0 = 1500 m/s, = 60 m/s, L = 3 m,
H = 1.5 m. An initially plane wave whose propagation direction is ē = ēx is incident
at x = 0 on the heterpgeneous region. Determine the ray paths and wavefronts in the
region 0 > x > 6L.
Exercise 11.18 Consider the fluid in Exercise 11.17. A harmonic plane wave having
amplitude B propagates at 8◦ above the x-axis in the region x < 0. Consider the ray
that is incident on the heterogeneous region at x = 0, y = H/4. (a) Determine the
path of this ray. (b) What is the location of this ray and the direct of the tangent to
that ray when it emerges from the heterogeneous region at x = L. (c) Determine the
amplitude of the signal described by this ray at the location where the ray emerges
from the heterogeneous regions
Exercise 11.19 A source at point A emits rays in all directions. Use Fermat’s prin-
ciple to prove that the ray that arrives at field point B after reflection from the corner
is parallel to the ray that reflects from the vertical wall.
xB
A
xA
B
yA
yB
Exercise 11.19
Exercise 11.20 The sketch shows two points at opposite sides of a rigid cylinder.
The fluid in which it is immersed is homogeneous, with constant sound speed c.
Suppose there is a ray connecting these points. Because path segments within the
fluid must be straight, the only way a connecting ray might exist is if it wraps around
the surface of the cylinder, as shown in the sketch. Given distances L1 and L2 use
Fermat’s principle to determine the angles φ1 and φ2 that mark the limits of this
contact. Draw a sketch of the result. Also determine the corresponding propagation
time for a signal that follows this path.
A   B
L L
Exercise 11.20
Chapter 12
Scattering
Have you ever wondered why you can hear sound even though there is a building
between you and the source, or why the sky is blue, or how dolphins can iden-
tify objects underwater? These phenomena are manifestations of scattering. At a
fundamental level, scattering is the reflection of a signal from a finite-sized object.
Knowledge of the scattering properties of a body probably is of the greatest concern
for underwater acoustics, not only for its relevance to naval applications like sonar,
but also for animal bioacoustics. Scattering phenomena often also are important for
studies of atmospheric propagation.
We will see that sometimes the analysis of scattering is closely related to the
description of acoustic radiation. Like that problem, scattering analyses are usually
performed in the frequency domain. The key parameter is the wavelength relative to
the size of the scattering region. The primary tools for low frequencies are derived
from the Kirchhoff–Helmholtz integral theorem, whereas geometrical acoustics is the
approach we will employ for high frequencies. Both formulations entail the judicious
application of well-justified approximations.
The problematic situation is the mid-frequency range, in which the object’s dimen-
sions are comparable to the wavelength. There are few objects that are conducive
to an analytical study of the scattered field at an arbitrary frequency. The sphere is
one that we shall take up. Spherical harmonics will play an important role in that
study. As is true for many simplified models, the availability of this solution will help
us understand the validity of the low- and high-frequency approximations. Never-
theless, we seldom encounter spheres as actual acoustical systems. Realistic shapes
require computational methods.

DOI 10.1007/978-3-319-56847-8_12
480 12 Scattering
12.1 Background
The general situation has two finite-sized bodies that are surrounded by a fluid whose
extent is infinite. One generates a signal that encounters the other body as an obsta-
cle to its propagation. The modification of the incident signal is the scattered signal.
Reflection is one aspect of the scattering process, but other phenomena are involved.
In the most accurate model, the pressure distribution on the surface of the scattering
body causes that body to move. Such motion modifies the scattered signal in compar-
ison with what it would be if the body were immobile. Incorporation of the effect of
that motion might require consideration of the laws of structural dynamics. Another
complication is that the scattered signal returns to the generating body. Consequently,
that body also is a scatterer, so its surface response is different from that associated
with generation of the incident signal. For example, if the generating body is an
electromechanical transducer, the pressure and particle velocity corresponding to a
specified voltage are modified. This alters the signal that is incident on the second
body that was considered at the outset to be the scatterer, and so on. This is the phe-
nomenon of mutual scattering. The extended version features many bodies whose
scattered fields interact.
An analysis of mutual scattering is quite complicated and requires prior under-
standing of single-body scattering. The latter is a very good model if the distance
between the two bodies is much greater than the acoustic wavelength and the largest
dimension of either body, because the farfield of either scattered signal will eventu-
ally decay inversely with the distance from that body. The consequence is that the
signal radiated by the generating body will be little affected by the signal returning
from the scatterer.
Thus,
the general
problem we shall address is that body S0 generates the signal
Re PI (x̄) eiωt that is incident on the scattering body S. The scattered signal is
Re Ps (x̄) eiωt . The total pressure is the sum of these contributions,
P (x̄) = PI (x̄) + Ps (x̄) (12.1.1)
As noted, because S and S0 are widely separated, the field generated by S0 is essen-
tially the same as it would be if S were not present. Thus, PI is a solution of the
Helmholtz equation that matches the normal velocity on S0 , and obeys the Sommer-
feld radiation condition at large radial distances r0 from S0 ,
∇ 2 PI + k 2 PI = 0
n̄ · ∇ PI = −iωρ0 V0 , x̄ ∈ S0
∂ PI (12.1.2)
lim r0 + ik PI = 0
r0 →∞ ∂r0
In principle, the incident field can be any field consistent with these equations.
However, we have imposed the restriction that S is very distant from S0 , which
means that it is reasonable in most cases to consider the field that is incident on S to
be locally planar.
12.1 Background 481
If the scattering body is capable of motion, either as a rigid body, or as a con-

sequence of its deformability, then the surface velocity on S is V̄ (x̄). The normal
velocity in the acoustic signal must match the normal component of V (x̄). Also, the
total pressure must satisfy the Helmholtz equation. Because P is the superposition
in Eq. (12.1.1) and PI individually satisfies the Helmholtz equation, the scattered
pressure must satisfy
∇ 2 Ps + k 2 Ps = 0
(12.1.3)
n̄ · ∇ Ps = −n̄ · ∇ PI − iωρ0 n̄ · V̄ , x̄ ∈ S
In addition, the scattered field at large radial distances r from S must satisfy the
Sommerfeld radiation condition relative to that body,

∂ Ps
lim r + ik Ps =0 (12.1.4)
r →∞ ∂r
These equations may be interpreted as a statement that the scattered field is the
pressure radiated by S when the normal velocity of its surface is n̄ · V̄ + (1/iωρ0 ) n̄ ·
∇ PI .
By definition, the scattered pressure is the field created by the insertion of S
into the fluid domain. The pressure on that surface is the total quantity P (x̄). By
analogy with a radiation problem, the Kirchhoff–Helmholtz integral theorem (KHIT)
describes Ps at any location x̄0 exterior to S. Equation (7.4.14) states that

Ps (x̄0 ) = [P (x̄s ) n̄ (x̄s ) · ∇G (x̄0 , x̄s ) − n̄ (x̄s ) · ∇ P (x̄s ) G (x̄0 , x̄s )] dS
S
(12.1.5)
Although KHIT is valid for any Green’s function, the one we will use is the free-space
version,
e−ikr̂
G (x̄0 , x̄s ) = , r̂ = |x̄0 − x̄s | (12.1.6)
4πr̂
The gradient operator is applied at x̄s , so that

x̄s − x̄0 d 1 1 −ikr̂ x̄0 − x̄s
∇G (x̄0 , x̄s ) = G (x̄0 , x̄s ) = ik + e (12.1.7)
r̂ d r̂ 4πr̂ r̂ r̂
The viewpoint in Eq. (12.1.5) is that the scattered pressure is the field radiated
from S by the total pressure field. An alternative form stems from the perspective
of Eqs. (12.1.3) and (12.1.4), which holds that Ps is a solution of the Helmholtz
equation that satisfies the Sommerfeld radiation condition. Thus, Ps by itself must
be consistent with the KHIT,
482 12 Scattering

Ps (x̄0 ) = [Ps (x̄s ) n̄ (x̄s ) · ∇G (x̄0 , x̄s ) − n̄ (x̄s ) · ∇ Ps (x̄s ) G (x̄0 , x̄s )] dS
S
(12.1.8)
In this view, the scattered pressure at a field point is the field radiated from S due to
the scattered pressure and particle velocity on that surface.
Some individuals are puzzled by the availability of two forms of the KHIT for Ps ,
so let us examine this issue. By definition, the total pressure is P = Ps + PI . Hence,
if we subtract the above version from Eq. (12.1.5), we find that

[PI (x̄s ) n̄ (x̄s ) · ∇G (x̄0 , x̄s ) − n̄ (x̄s ) · ∇ PI (x̄s ) G (x̄0 , x̄s )] dS = 0 (12.1.9)
S
This identity may be derived in a different way. The incident field satisfies the KHIT
for the surface S0 from which it emanates, that is,

PI (x̄0 ) = [PI (x̄s ) n̄ (x̄s ) · ∇G (x̄0 , x̄s ) − n̄ (x̄s ) · ∇ PI (x̄s ) G (x̄0 , x̄s )] dS
S0
(12.1.10)
At the same time, we could form the KHIT for PI (x̄0 ) by including the scattering
body in the domain. Then, the surface integral would consist of contributions from
S and S0 . According to the preceding, the contribution from S0 gives PI (x̄0 ), so
the contribution from S must be zero. This is the condition stated in Eq. (12.1.9).
Whether it is preferable to use Eq. (12.1.5) or Eq. (12.1.8) sometimes depends on
what is known about the field.
12.2 Scattering by Heterogeneity
In Example 11.5 of the previous chapter, we used ray theory to examine how a signal
passes through a region in which the fluid properties vary spatially. In the situation
considered there, the density and sound speed were continuous functions of position,
so there was no impedance mismatch. Consequently, sound was transmitted into and
out of the heterogenous region without reflections. In the situation, we consider here
S is the surface that bounds a region in which the density ρ and sound speed c
are different from the ambient properties exterior to S. Thus, the incident signal is
scattered by S, as well as being transmitted into the interior, then out again. Both ρ
and c are taken to be constant, so the scattering region is inhomogeneous from the
perspective of the overall domain, although it is internally homogeneous.
12.2 Scattering by Heterogeneity 483
12.2.1 General Equations
We cannot determine the scattered pressure directly from the KHIT integral because
we do not know the surface pressure and particle velocity. Perhaps we can gain some
insight if we examine the field in the homogeneous region V interior to S. A prime
will denote that a quantity is associated with the interior, so the position of a point
in V is labeled in Fig. 12.1 as x̄ . Point C is central to V, such as the centroid, and
x̄0 is an arbitrary field point exterior to V. The z-axis has any convenient reference
orientation. It serves as the axial direction for a set of spherical coordinates, r, ψ0 , θ0
that locates the field point.
Fig. 12.1 Definition of S x0

positions and coordinates n(xs)
relative to fluid region V xs
whose properties differ from
those of the surrounding fluid
x’ C 

 z
The pressure and normal particle velocity must be continuous along the interface
S between the inner and outer fluids, so it must be that
1 1
P (x̄s ) = P (x̄s ) and n̄ (x̄s ) · ∇ P (x̄s ) = n̄ (x̄s ) · ∇ P (x̄s ) (12.2.1)
iωρ iωρ0
Upon substitution of these relations, Eq. (12.1.5) becomes

ρ0
Ps (x̄0 ) = n̄ (x̄s ) · P (x̄s ) ∇G (x̄0 , x̄s ) − ∇ P (x̄s ) G (x̄0 , x̄s ) dS
ρ
S
(12.2.2)
The divergence theorem converts the integral to one that extends over the interior
domain. Note that n̄ (x̄s ) points out of the interior domain, so no sign change is
entailed in this operation. Thus, we have

ρ0
Ps (x̄0 ) = ∇ · P x̄ ∇G x̄0 , x̄ − ∇ P (x̄) G x̄0 , x̄ dV (12.2.3)
ρ
V
where ∇ is the gradient at interior point x̄ .

The fact that x̄0 is exterior to V simplifies this integral because
nowhere does
an interior point x̄ coincide with x̄0 . Consequently, G x̄0 , x̄ is a solution
of

the Helmholtz
equation with no
additional source term, that is, ∇ 2
G x̄ 0 , x̄ =
−k 2 G x̄0 , x̄ . In addition, P x̄ is a solution of the Helmholtz equation for
the interior domain, where the sound speed is c . It follows that ∇ 2 P x̄ =
2
− ω/c P x̄ . These relations convert the KHIT to
484 12 Scattering

ρ0
Ps (x̄0 ) = 1 − ∇ P x̄ · ∇G x̄0 , x̄
ρ
V

ρ0 c2
−k 2 1 − 2 P x̄ G x̄0 , x̄ dV (12.2.4)
ρ (c )
Further simplifications result if we limit consideration to cases where x̄0 is in

the farfield. The farfield approximations apply if the smallest distance from a point
within V to x̄0 is much greater than the largest dimension of the region. The farfield
approximations use some central point C as the reference point, so that x̄C ≡ 0̄. The
length of the line from point C to x̄0 is the radial distance r , and the direction is ēr .
The line from any point x̄ to x̄0 is considered to be parallel to ēr , so the distance
between x̄ and x̄0 differs from r by the projection of the vector connecting those
points. The relations are
x̄0
r = |x̄0 | , ēr = , r̂ ≡ x̄0 − x̄ ≈ r − x̄ · ēr (12.2.5)
r
This leads to a simplified form of the Green’s function in Eq. (12.1.6),
e−ikr ik x̄ ·ēr
G x̄0 , x̄ ≈ e (12.2.6)
4πr

Furthermore, ∇G x̄0 , x̄ is the gradient at x̄ , with x̄0 held fixed, so that
d ike−ikr ik x̄ ·ēr
∇G x̄0 , x̄ = ∇ r̂ G x̄0 , x̄ ≈ ēr e (12.2.7)
d r̂ 4πr
The orientation of x̄0 is described by the polar angle ψ0 and azimuthal angle θ0 .
Hence, substitution of these expressions into Eq. (12.2.4) leads to recognition that
farfield scattering from a heterogeneity is described by two directivity factors. The
general representation is

e−ikr ρ0 ρ0 c2
Ps (x̄0 ) = 1 − F1 (ψ0 , θ0 ) + 1 − 2 F2 (ψ0 , θ0 ) (12.2.8)
4πr ρ ρ (c )
where the directivities are

F1 (ψ0 , θ0 ) = ik ēr · eik x̄ ·ēr ∇ P x̄ dV
V (12.2.9)

F2 (ψ0 , θ0 ) = −k 2 eik x̄ ·ēr P x̄ dV
V
Without further development, these relations merely tell us that the strength of
the scattered field depends on the mismatch of densities and of the bulk moduli, with
the former being attributable to the particle velocity within the scattering volume
and the latter due to the pressure in that region. Evaluation of the directivity factors
requires judicious approximations of the interior field. Two that have been used
consider different situations. The Born approximation is based on an assumption
that the properties of the scattering region do not drastically differ from those of
the surrounding homogeneous fluid. An approximation due to Rayleigh addresses
objects whose size is much smaller than an acoustic wavelength.
12.2.2 The Born Approximation
If the region inside S was the same as the fluid in the outer region, the pressure inside
S would be the incident value at that location. The Born approximation addresses the
situation in which
the difference between the fluids is not substantial, that is, when
2
ρ ≈ ρ0 and ρ c ≈ ρ0 c2 . In that case,it is reasonable
to assume that the inside
pressure differs little from PI . Setting P x̄ = PI x̄ in Eq. (12.2.4) converts the
directivity factors in Eq. (12.2.9) to computable functions.
The typical situation is a plane wave that propagates in direction ēI . Let P̂I be the
amplitude at the origin, so that

PI x̄ = P̂I exp −ik x̄ · ēI , ∇ P x̄ = −ik ēI P̂I exp −ik x̄ · ēI (12.2.10)
Substitution of these representations into Eq. (12.2.9) gives
F1 (ψ0 , θ0 , ψI , θI ) = k 2 P̂I ēr · ēI L (ψ0 , θ0 , ψI , θI ) ,

F2 (ψ0 , θ0 , ψI , θI ) = −k P̂I L (ψ0 , θ0 , ψI , θI )
2
(12.2.11)
eik (x̄ ·ēr −x̄ ·ēI ) dV

L (ψ0 , θ0 , ψI , θI ) =
V
Because the integrals for the directivity factors F1 and F2 extend over the domain
of the scattering region, it makes sense to align the coordinate system consistently
with the shape of the body. Then, ēI and ēr to the field point have arbitrary orientation
relative to the coordinate system. We use spherical angles ψI and θI defined relative
to the z-axis to describe ēI , while angles ψ0 and θ0 describe ēr .
ēI = sin ψI cos θI ēx + sin ψI sin θI ē y + cos ψI ēz

(12.2.12)
ēr = sin ψ0 cos θ0 ēx + sin ψ0 sin θ0 ē y + cos ψ0 ēz
The presence of ψI and θI as variables affecting the directivity factors emphasizes

that the scattered pressure depends on the propagation direction of the incident wave
486 12 Scattering
relative to the body, as well as the direction to the field point, which is defined by
ψ0 and θ0 . The position x̄ and differential volume element dV may be described
in terms of Cartesian coordinates, cylindrical coordinates, or spherical coordinates,
whichever is most convenient.
This description of Born scattering is not limited to a certain frequency range.
However, its utility is limited by the condition that the scattering region is homoge-
neous with properties that are very much like those of the surrounding fluid.
EXAMPLE 12.1 A plane wave is incident on a region in the shape of a

cylinder whose radius is a and length is L. The sound speed and density in
this region are very close to the ambient properties, so the Born approximation
may be applied. Derive
an expression for the scattered pressure in the form

(r/a) Ps (x̄0 ) / P̂I for arbitrary ēI and ēr . Then, specialize this expression to
the case of backscatter, for which ēr = −ēI . Use that expression to evaluate
the dependence of the backscattered pressure on the orientation of the incident
wave for ka = 0.1, 1, and 10. The aspect ratio is L/a = 20, and the fluid
properties are ρ0 /ρ = c/c = 0.95.
Significance
The solution demonstrates the way in which scattering integrals are formulated in
general. The results that are obtained are a specific case of backscatter, which later
will be addressed in greater detail. Most importantly, the results are illustrative of
general scattering properties.
Solution
The starting point is Fig. 1, which illustrates the scattering region, the unit vectors,
and the x yz coordinate system for the description of vectors. The origin is situated
at the centroid of the cylinder, and the z-axis is aligned with the centerline. The
axisymmetry of the system makes it permissible to situate the x-axis in the plane
containing ēI and ēz . This means that we can set θI = 0 without losing generality of
the analysis.
x
L/2
ψΙ
e-r
θ0
e-Ι ψ0 a z
Figure 1
The shape is best described in terms of cylindrical coordinates defined relative to

x yz, so that dV = Rdθd Rdz. The relevant vectors are
ēI = sin ψI ēx + cos ψI ēz

ēr = sin ψ0 cos θ0 ēx + sin ψ0 sin θ0 ē y + cos ψ0 ēz (1)
x̄ = R cos θēx + R sin θē y + z ēz
When these representations are substituted into the integral in Eq. (12.2.11), the
result is
L/2 a π
L(ψ0 , θ0 , ψI , θI ) = exp[ik R(sin ψ0 cos θ0 − sin ψI ) cos θ
−L/2 0 −π
+ik R(sin ψ0 sin θ0 ) sin θ + ikz(cos ψ0 − cos ψI )]Rdθd Rdz
Several simplifications are available. The first two terms in the integrand are
independent of z, whereas the last term is independent of R and θ. The integral
may be factorized accordingly. Furthermore, the integral over z may be evaluated
analytically. The integral over the range of θ may be split into integrals over positive
and negative θ. Replacing θ by −θ̃ in the latter interval allows the integral to extend
only over positive θ. The result is that
L (ψ0 , θ0 , ψI , θI ) = a 2 LG (ψ0 , θ0 , ψI ) H (ψ0 , ψI ) (2)
The H factor is the result of integration over z,

1/2

H (ψ0 , ψI ) = exp ik L ẑ (cos ψ0 − cos ψI ) d ẑ
−1/2
It may be integrated analytically, which leads to

⎧
⎨ sin [(k L/2) (cos ψ0 − cos ψI )]
if ψ0 = ψI
H (ψ0 , ψI ) = (k L/2) (cos ψ0 − cos ψI ) (3)
⎩ 1 if ψ = ψ
0 I
The G factor comes from integration over θ and R. It is

1 π
G (ψ0 , θ0 , ψI ) = 2 exp ika R̂ (sin ψ0 cos θ0 − sin ψI ) cos θ
0 0 (4)
× cos ka R̂ (sin ψ0 sin θ0 ) sin θ R̂dθd R̂
Numerical methods are required to evaluate this function for arbitrary orientation
angles. However, a closed form result is available for the backscatter case.
488 12 Scattering
After the value of L(ψ0 , θ0 , ψI , θI ) is determined for a specific set of angles, the
directivity factors in Eq. (12.2.11) are known. Their substitution, and ēr · ēI from
Eq. (1), into Eq. (12.2.8) gives the scattered pressure. Collecting the parameters into
nondimensional groups leads to

P̂I a −ikr ρ0
Ps (x̄0 ) = e (ka) (k L) 1 − (sin ψ0 cos θ0 sin ψI + cos ψ0 cos ψI )
4π r ρ

ρ0 c2
− 1− 2 G(ψ0 , θ0 , ψI )H(ψ0 , ψI ) (5)
ρ (c )
Evaluation of the G function for arbitrary ψI , ψ0 , and θ0 requires numerical integra-
tion, which is easier if the available software has a routine for double integration. In
MATLAB, it is dblquad, whose input arguments are the name of a function that
evaluates the integrand, as well as the lower and upper integration limits for both
variables.
Equation (5) offers little insight to the nature of the scattered field. The results for
an arbitrary set of angles could be displayed as a spherical
plot for fixed ψI , in which

ψ0 and θ0 are the spherical angles and (r/a) Ps / P̂I is the radial distance. Many
such plots would describe the dependence on ψI . Fortunately, the information that is
requested sets ēr = −ēI . This corresponds to setting θ0 = π and ψ0 = π − ψI . For
these angles, the preceding expression reduces to
r P (x̄ )
s 0 e−ikr ρ0 c2
= (ka) (k L) 1 + 2 − 2
a P̂I 4π ρ (c ) (6)
G (π − ψI , π, ψI ) H (π − ψI , ψI )
where
1 π
G (π − ψI , π, ψI ) = 2 exp −2ika R̂ sin ψI cos θ R̂dθd R̂
0 0
sin (k L cos ψI )
H (π − ψI , ψI ) =
k L cos ψI
Analytical evaluation is possible by invoking general formulas provided in

Abramowitz and Stegun’s compendium.1 The eventual result is
J1 (2ka sin ψI )
G (π − ψI , −π, ψI ) = π (7)
ka sin ψI
There are different ways in which parametric trends may be viewed. We can
consider the value of Ps that would be measured at a fixed location, which sets r .
1 M.I.Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, (1965), p. 360
for Eq. (9.1.210), and p. 484 for Eq. (11.3.20).
Because a 2 L = V/π, that view considers Ps to be proportional to the volume of the

heterogenous medium, as well as the square of the frequency. However, changing the
dimensions or the frequency also has a significant effect on the angular dependence.
The first of Eq. (6) indicates that G is a function of ka, as well as ψI . The value of G
for ψI = 0 is π for any ka, as well as for any ψI if ka
1. Furthermore, G = π is the
maximum value for any ka and ψI . These properties mean that the G function leads
to enhanced directivity in the axial direction as ka increases. In the case of H, its
maximum is H = 1. This value occurs at any ψI if k L
1, and at ψI = π/2 for any
k L . Thus, H leads to greater directivity in the transverse direction as the frequency
increases. The low-frequency limit is examined in detail in the next section.
Figure 2 displays polar plots of (r/a) |Ps /PI | at the designated frequencies.
Backscatter at broadside incidence is somewhat larger at the lowest frequency, and
it is strongly directional in the broadside direction at the middle frequency. At the
highest frequency, backscatter is comparable for end-on and broadside incidence,
and both are confined to a smaller range of angles.
ka = 0.1 90
ka = 1 90
1
0.01
0.5
180 0 180 0
90
ka = 10 1
0.5
180 0
Figure 2
To gain insight to this shifting trend, let us consider individually G (π − ψI ,

−π, ψI ) and H (π − ψI , ψI ), whose product describes the backscatter directivity.
Examination of Eq. (7) reveals that the maximum value of the former is G = π at
end-on incidence and that G = π J1 (2ka) / (ka) at broadside incidence. This value
decreases in an oscillatory manner with increasing ka. In contrast, the maximum
value of H broadside is H = 1, and H = sin (k L) / (k L) for end-on incidence. This
too decreases in an oscillatory manner with increasing frequency. With these trends
in mind, consider the graphs in Fig. 3. Each function is large in a narrow interval.
The directivity of Ps is the product of the functions, which explains why it is very
small away from ψI = 0◦ , 90◦ , and 180◦ . Increasing ka increases k L , so J1 (2ka)
and sin (k L) will grow and decrease. If either term is near zero, the value of ps will be
very small in the respective direction. Consequently, whether backscatter is greater
at broadside or end-on incidence strongly depends on the frequency.
490 12 Scattering
4 1
H at ka = 10
G at ka = 10
0.5
2
1
0
0
−1 −0.5
0 45 90 135 180 0 45 90 135 180
Figure 3
12.3 Rayleigh Scattering Limit
The Rayleigh scattering limit is a general term that refers to the field scattered by any
object whose size is much less than the acoustic wavelength. Specifically, if a is the
largest dimension, then the Rayleigh scattering limit is the behavior when ka
1.
This condition leads to several analytical simplifications. It also will disclose some
qualitative aspects of the field that are not evident from the general equations.
The Rayleigh limit is sometimes referred to as a quasistatic approximation. This
terminology stems from the fact that because of the small size of the body, the phase
of any long-wavelength signal will be essentially constant over the extent of the
body. This means that any complex exponential representing a phase shift within the
scattering body or on its surface may be approximated as having unit value.
12.3.1 The Rayleigh Limit of the Born Approximation
The Born approximation is not limited by the size of the heterogeneous region. Its
associated directivity factors must be evaluated for a specific body, and a specific
frequency. The Rayleigh limit of the Born approximation yields a formula for the
scattered field that does not require evaluation of integrals. At the same time, it is
less general because it considers the scattering region to be much smaller than a
wavelength.
The previous example indicates how we should proceed to obtain the low fre-
quency version. As before, we shall only consider the case of an incident plane
wave, which is described by Eq. (12.2.10). Let a be the largest dimension of the
scattering region. If the origin
is interior to the region, then any position within
this region is such that x̄ ≤ a. Furthermore,
we are interested in the situation
when 2π/k a. It follows that k x̄
1, so that the integrand in Eq. (12.2.11)
is essentially one, which gives L = V. The corresponding directivity factors are
12.3 Rayleigh Scattering Limit 491
F1 = k 2 P̂I ēr · ēI V and F2 = −k 2 P̂I V. The only directional factor that arises is ēr · ēI .
Therefore, we may define the z-axis to be parallel to ēI . This reduces Eq. (12.2.8) to

e−ikr ρ0 c2 ρ0
Ps (x̄0 ) = k P̂I
2
V −1 − − 1 cos ψ0 (12.3.1)
4πr ρ (c )2 ρ
where ψ0 is the angle between ēI and the radial direction to the field point. This
expression tells us that the strength of the scattered field at a fixed location is pro-
portional to the volume of the scatterer. This attribute surprises some individuals
who intuitively think that the cross-sectional area represents the amount the sound
is blocked. Another important observation is that the field consists of a monopole
whose strength is proportional to the difference of the bulk moduli, combined with a
dipole component that is proportional to the difference of densities. The polar angle
to the field point is measured from the incident direction ēI . Thus, the dipole is aligned
in the direction of the incident wave. Both the monopole and dipole strengths are
proportional to the square of the frequency.
It is instructive to compare this result to the solution of Example 12.1, which is
specific to a cylindrical region. That result is valid for any frequency, but it is only
descriptive of a specific shape. The Rayleigh limit described Eq. (12.3.1) is valid only
for low frequencies, but it is descriptive of scattering from any region, regardless of
its shape. Another difference is that the example was limited to situations where
the scattered signal is observed on the same line as that from the planar source to
the region. This corresponds to ψI = 180◦ in the above expression. The result for
a cylindrical region when ka
1 is identical to Eq. (12.3.1) when V = πa 2 L and
cos ψI = −1. Despite the generality of the present result, the assumptions embedded
in the model lead to a description of the scattered pressure that depends on the location
of the field point, but is independent of the orientation of the body. In contrast, such
dependence is displayed in the Born approximation.
Viewing the dependence of the scattered pressure in Eq. (12.3.1) in terms of
nondimensional variables offers a different perspective. The largest dimension has
so the volume can be described as βa , where β is a shape factor.
3
been denoted as a,

Then r/a Ps / P̂I depends solely on β, (ka) , and ψI .
2
This trend was used by Rayleigh2 to explain why the sky is blue. At midday, the
light rays that are incident at our location from the sun essentially are white, which
means that they are composed of all colors in the visual spectrum in nearly equal
amounts. These rays constitute the image of the sun. If there were no scattering, these
are the only rays that would travel from the sun to an observer on the ground. The rest
of the sky would appear to be black, as it does on the Moon. What actually happens
is that the light rays that propagate through the atmosphere in other directions are
scattered by dust and water particles in the atmosphere. Some of these rays reach
an observer on the ground. The Rayleigh description of scattering indicates that the
2 J.W.
Strutt, Lord Rayleigh, “On the light from the sky, its polarization and colour,” Philosophical
Magazine, series 4, vol. 41 (1871) pp. 107–120.
492 12 Scattering
blue spectrum, which comprises the short-wavelength/high-frequency portion of the

visual spectrum, scatters more strongly than the red spectrum. Thus, the contribution
of the blue spectrum to what people see when they view the sky is stronger than the
contribution of the red spectrum.
12.3.2 Mismatched Heterogeneous Region
If the properties of a small region of fluid differ significantly from those of the sur-
rounding fluid, the Born approximation is not valid. A different analysis, also based
on ka being small, is available in that case. Let us multiply the Helmholtz equation by
a 2 and define a nondimensional gradient ∇ˆ = a∇. Smallness of ka means that ∇ˆ P
is the dominant term in the Helmholtz equation within the heterogeneity. Hence, the
Helmholtz equation for the interior field reduces to the Laplace equation. Smallness
approximations do not apply to the exterior scattered field, which must satisfy the
Helmholtz equation. The boundary conditions on the surface S, which is the inter-
face between the media, are continuity of pressure and normal particle velocity. Let
P (x) be the pressure within the heterogeneity, and Ps (x̄) be the scattered field. The
governing equations are
∇ 2 P = 0, x̄ ∈ V
∇ Ps + k 2 Ps = 0, x̄ ∈
2
/V

P = Ps + PI (12.3.2)
ρ0
n̄ · ∇ P = n̄ · (∇ Ps + ∇ PI ) x̄ ∈ S
ρ
In addition the scattered field must satisfy the radiation condition in Eq. (12.1.4).
To see how the limitation to small ka simplifies the analysis, we shall consider
a specific situation. Suppose V is a sphere and the incident signal is a plane wave
propagating in the axial direction, so that ēI = ēz . Let r̂ = r/a be the nondimensional
radial position, and set n̄ = ēr . Then, the incident pressure and its normal derivative
on the surface of the sphere are
−ika cos ψ
PI |r̂ =1 = P̂I e

∂ PI (12.3.3)
= −ika cos ψ P̂I e−ika cos ψ
∂ r̂
r̂ =1
Smallness of ka allows us to expand the complex exponential in a series that is

truncated at the lowest order frequency-dependent term, so that
PI |r̂ =1 = (1 − ika cos ψ) P̂I

∂ PI (12.3.4)
= −ika cos ψ P̂I
∂ r̂ r̂ =1
We may use a spherical harmonic expansions to represent Ps and P̃. Both series
may be truncated at the m = 1 harmonic, because that is the highest harmonic appear-
ing in the incident pressure and pressure gradient on the surface, both of which
constitute the excitation. The series for the scattered pressure is

1
(2)
Ps = B̂m Pm (cos θ) h (2)
m ka r̂ = B̂0 h 0 kar̂ + B̂1 h (2)
1 kar̂ cos ψ (12.3.5)
m=0
This description is valid everywhere, but it may be simplified in the vicinity of the
sphere based on ka
1 and r̂ = O (1) . This form is
B0 B1
Ps ≈ + 2 cos ψ, kar̂
1
r̂ r̂ (12.3.6)
i i
B0 = B̂0 , B1 = B̂0
ka 2 (ka)2
The value of kar̂ is much less than one everywhere within V, so the spherical har-
monic representation of the pressure in the scattering region may be represented by
a similar form,
∞

P = Am Pm (cos θ) jm (kr ) ≈ A0 + A1r̂ cos ψ (12.3.7)
m=0
These series are substituted into the boundary conditions in Eq. (12.3.2), and like
dependencies on ψ are matched. This operation yields
A0 = P̂I + B0
A1 = −ika P̂I + B1
0 = −B0 (12.3.8)
ρ0
A1 = −ika P̂I − 2B1
ρ
Solution of these equations leads to an explicit description of the pressure and its
gradient within V,
3ik
P = P̂I 1 − r cos ψ
2 + ρ0 /ρ (12.3.9)
3ik
∇ P = − P̂I r cos ψ
2 + ρ0 /ρ
These expressions are used to evaluate the volume integrals in Eq. (12.2.9). The
result is

e−ikr 4πa 3 ρ0 c2 3 ρ0 /ρ − 1
Ps (x̄0 ) = P̂I k 2 −1 − cos ψ0 (12.3.10)
4πr 3 ρ (c )2 2 + ρ0 /ρ
494 12 Scattering
The above has several features that are like the behavior predicted by the Born
approximation. The magnitude of the scattered pressure increases as the square
of the frequency. It consists of a monopole whose strength is proportional to the
difference of the bulk moduli, and a dipole whose strength is proportional to the
difference of densities. The amplitude of the monopole term matches the result of
the Born approximation, Eq. (12.3.1), but the dipole moment is different. In the case
where ρ0 /ρ ≈ 1, the dipole moment derived here is 50% greater than the previous
result. The Born approximation was introduced empirically, whereas the present
result stems from an analysis that only assumes that ka is sufficiently small.
Although the above formula was derived for a sphere, we know that at very
low frequencies, geometric details that are much smaller than the wavelength are
unimportant. Thus, it is reasonable to use Eq. (12.3.10) with (4/3) πa 3 replaced by
V, rather than Eq. (12.3.1), to describe scattering from a small embedded region of
fluid, regardless of its shape.
12.3.3 Scattering from a Rigid Body
Analysis of the field scattered by an immovable body is quite complicated if the

wavelength is comparable to the size of the body. Restricting our consideration to
situations where the body is small enables us characterize the general nature of the
scattered field. The development is founded on the Kirchhoff–Helmholtz integral
theorem in Eq. (12.1.8), which states that Ps is the field radiated from S due to the
scattered pressure and particle velocity on that surface. We begin by considering the
situation where the rigid body is immobilized by some means. After that, we will
formulate a description that accounts for the effect of removing restraints against
movement.
Stationary Bodies
When the surface S is stationary, the normal component of the particle velocity
must vanish everywhere on S. The total signal is the sum of the incident and scattered
parts, so it must be that
n̄ · V̄s = −n̄ · V̄I , x̄ ∈ S (12.3.11)
Specific results may be obtained for low-frequency scattering of an incident plane

wave,
P̂I −ik ēI ·x̄
PI = P̂I e−ik ēI ·x̄ =⇒ V̄I = ēI e (12.3.12)
ρ0 c
where ēI is the propagation direction. It will be useful for the analysis if we situate the
origin at the geometric centroid C of the body, and set the phase lag of the wavefront
through point C to zero. This is the arrangement depicted in Fig. 12.2.
Fig. 12.2 Plane wave is z

incident on a rigid body
n-
e-I
x-s
C
x y
The definition of the low-frequency range is that the wavelength is much larger
than the size of the scattering body. In that case, it must be true that k |x̄|
1 for
any point in the body. Consequently, V̄I will be adequately described by the first few
terms in a Taylor series,
P̂I
V̄I = ēI [1 − ik ēI · x̄ + · · ·] (12.3.13)
ρ0 c
The surface normal velocity in the scattered signal must cancel the incident contri-
bution, so we have
P̂I
n̄ · V̄s = n̄ · ēI [−1 + ik ēI · x̄ + · · ·] ; x̄ ∈ S (12.3.14)
ρ0 c
A useful way to interpret this expression is that the scattered pressure is equivalent
to the pressure radiated by the body when the normal velocity on the surface is the
quantity on the right side. This viewpoint allows us to invoke the multipole expansion
of radiation from a rigid body, which was the topic in Sect. 7.4.2. The first contribution
to n̄ · V̄s is the component of the particle velocity V̄I in the direction of the inward
normal. In a rigid body translation, all points have the same velocity. Therefore, the
first term represents a rigid body translational oscillation whose complex velocity
amplitude is −V̄I . Such a motion leads to dipole radiation, but does not contribute
to the monopole amplitude. The second term in Eq. (12.3.14) may be written as
i n̄ · V̄I (ēI · k x̄) . This term contributes to both the monopole amplitude and dipole
moment. However, |x̄| can be no larger than the largest dimension a, and we have
restricted our consideration to situations where ka
1. This observation makes it
permissible to neglect the contribution of this term to the dipole moment. Hence, by
these considerations, we have recognized that the monopole amplitude A accounts for
the volume velocity associated with a normal velocity that is i n̄ · V̄I (ēI · k x̄) , whereas
D̄ is the dipole moment for a rigid body oscillation whose complex amplitude is −V̄I .
496 12 Scattering
The multipole expansion is given in Eq. (7.4.23). We let G (x̄0 , x̄C ) be the free-
space Green’s function and ignore higher order terms on the basis that they are weak
if 2π/k is much larger than the body’s size. The corresponding expression for the
scattered pressure in the farfield is
e−ikr
P (x̄0 ) = A + ik D̄ · ēr (12.3.15)
r
The monopole amplitude is stated by the first of Eq. (7.4.24) to be

iωρ0 P̂I k 2 P̂I
A= n̄ · ēI ik ēI · ξ¯ dS = − (n̄ · ēI ) ēI · ξ¯ dS
4π ρ0 c 4π
S S

k 2 P̂I

=− ∇ · ēI ēI · ξ¯ dV
4π
V
(12.3.16)

where the divergence theorem is applied with no sign change because n̄ ξ¯ is oriented
outward from V. To evaluate the gradient, let j denote the direction cosines of
ēI , so ¯
ēI · ξ = 1 x + 2 y + 3 z. Because ēI is a constant, the integrand reduces to
¯
ēI ∇ ēI · ξ ≡ ēI · ēI = 1. Correspondingly, the monopole amplitude reduces to
k2V
A=− P̂I (12.3.17)
4π
Determination of the dipole moment does not require evaluation of a surface

integral. The requisite analysis has been performed in Sect. 7.4.2, which treated the
field radiated by an oscillating rigid body. Thus, we may employ Eq. (7.4.39), with
the translational velocity sets at −V̄I ≡ −ēI P1 / (ρ0 c) . This equation is written in
matrix form to facilitate describing the possibility that the dipole moment might not
be aligned with the translational velocity. Thus, we find that
ik
{D} = PI [V [I ] + [W ] ] {eI } (12.3.18)
4π
The quantity ρ0 [W ] is the virtual mass, defined in Eq. (7.4.35).

The scattered pressure is obtained by adding the monopole and dipole contribu-
tions according to Eq. (12.3.15). In keeping with the matrix form of the relation for
the dipole moment, this pressure field is
−ikr
k 2 P̂I i e
Ps (x̄0 ) = − V − {er } [ V [I ] + [W ] ] {eI } 1 −
T
(12.3.19)
4π kr r
where r is the distance from the centroid of the scattering body to the field point. The
farfield version is obtained by dropping the i/kr term. The corresponding description
of the force exerted by the scattered pressure is stated in Eq. (7.4.35). Replacing V̄C
in that expression with ēI PI / (ρ0 c) gives
{F} = ik P̂I [W ] {eI } (12.3.20)
Both V and [W ] are shape-dependent quantities that scale as the cube of the size
of the body, which is measured by dimension a. Therefore, the scattered pressure
observed at a fixed field point depends on the cube of the size, the shape of the
body, the square of the frequency, and the orientation
of ēI and ēr relative to the

body. Nondimensionally, it is proportional to P̂I , a/r, and (ka)2 , and the direction
angles for both directions. These attributes, as well as the fact that the field is a
superposition of a monopole and dipole, are the same as those found for Rayleigh
scattering from a region of heterogeneity.
Evaluation of the scattered pressure when [] is known requires a description of
{ei } and {er }. For this, we may use spherical angles defined relative to the z-axis,
with ψI and ψr being the respective polar angles. The representation is
{eI } = [sin ψI cos θI sin ψI sin θI cos ψI ]T

(12.3.21)
{er } = [sin ψr cos θr sin ψr sin θr cos ψr ]T
The expression for the scattered pressure simplifies considerably when the body
is axisymmetric . We shall designate z as this axis. A consequence of axisymmetry
is that we may take ēI to lie in the zx, so that θI = 0. Setting the off-diagonal terms
of [] to zero reduces the farfield scattered pressure in Eq. (12.3.19) to
k 2 P̂I

Ps (x̄0 ) = − V − V + W1,1 sin ψI sin ψr cos θr
4π
e−ikr
− V + W3,3 cos ψI cos ψr
r
(12.3.22)
For a sphere, W1,1 = W3,3 = (2/3)πa 3 , which corresponds to the added mass being
half the mass of the displaced fluid. For a thin disk, W1,1 = 0 and W3,3 = (8/3) a 3 .
In the thin disk approximation, the displaced volume is essentially zero, so the added
mass is not related to the displaced mass. It is important to recognize that although
V is very small, the monopole term should not be taken to be zero because the dipole
contribution might also be very small. If the scattering body is homogeneous with
mass density ρb , then we can set V = M/ρb .
498 12 Scattering
EXAMPLE 12.2 The sketch shows an arrangement in which projector A and

hydrophone B are situated along orthogonal lines relative to the center of a
scattering body that is rigid and stationary. The distance from the source to
the body is sufficiently large to consider the incident wave to be planar, and
the wavelength of the incident wave is much greater than the body’s largest
dimension. The body is axisymmetric relative to the z-axis, which is coplanar
with ēI and ēr . The virtual mass matrix is described relative to the volume of the
body, according to W1,1 = W2,2 = 5V, 3,3 = V. Determine the dependence
on the angle of incidence ψ of the scattered
pressure at point B, normalized
according to 4πr |Ps (x̄ A ) /PI | / k 2 V .
B
z
_
er
_ 
eI
A
Figure 1.
Significance
Considering a specific case of scattering from a rigid body gives a clearer picture of
the angular dependence, as well as some understanding to the influence of the virtual
added mass.
Solution
To employ Eq. (12.3.22), we must define the direction angles. We define the x-axis to
lie in the plane formed by ēI and ēr , which also contains the z-axis. This is allowable
because it does not matter how much the body is rotated about its axis of symmetry.
The formula was derived on the basis that ēI has a positive component in the x
direction, which means that the x-axis is down and to the right in the sketch. Thus,
the unit vectors are
ēI = cos ψ ēz + sin ψ ēx , ēr = sin ψ ēz − cos ψ ēx (1)
A comparison of these expressions to the general description in Eq. (12.3.21) leads

to recognition that ψI = ψ, θI = 0, ψr = π/2 − ψ, and θr = π. The result is that

4πr 1 W1,1 − W3,3
|Ps (x̄0 )| = 1 + W1,1 − W3,3 cos ψ sin ψ = 1 + sin (2ψ)
k2V V 2V
(2)
This function is depicted in Fig. 2. Note that this is not a directivity. Rather, a radial
line at a specific ψ indicates the magnitude of |Ps | when the axis of symmetry, which
is x, is aligned at that angle.
90o
o 3 o
120 60
o
2 o
150 30
z
1
ψ
o
180 0o
210o 330o
240o 300o
o
270
Figure 2.
The plot has interesting features, some of which might be considered to be counter-
intuitive. At ψ = 0, 90◦ ,180◦ , and 270◦ , the normalized scattered pressure is one.
This is the monopole effect. But it is one also at ψ = 135◦ and ψ = 315◦ . At these
angles, the dipole contribution associated with the amplitude of the dipole contribu-
tion is twice the monopole value, but it is 180◦ out-of-phase from the monopole. In
contrast, the scattered pressure arriving at hydrophone B has a maximum at ψ = 45◦
and ψ = 225◦ because the two contributions are in-phase. The nulls which occur at
ψ = 105◦ , 165◦ , 285◦ , and 245◦ result from the two effects having equal magnitude
by opposite phase.
Many individuals think of scattering as a mirror-like effect, which is the viewpoint
of geometrical acoustics. In such thinking, ψ = 45◦ and ψ = 225◦ are like placing a
mirror in the x y-plane. The angle of incidence of ēI on this plane equals the angle of
reflection of ēr , which should lead to a maximum pressure at hydrophone B. This
expectation is confirmed by Fig. 1, but that does not mean that the underlying rea-
soning is correct. If it were, then ψ = 135◦ and ψ = 270◦ would be nulls because the
corresponding angle of reflection is downward. In general, low-frequency scattering
is not a process of reflection.
Effect of Rigid Body Motion
We now turn to the fact that the scattering body often is immersed in the fluid
without support, as in the case of a rain drop. Such movement generates a radiating
pressure field that adds to the pressure field obtained when the body is stationary.
Section 7.4.2 developed a multipole expansion that describes low-frequency radiation
from an oscillating rigid body. That development led to the observation that rotation
of a body at low frequencies generates a very weak quadrupole that is negligible if
500 12 Scattering
the dipole contribution is nonzero. This is the present situation, so we only need to
consider translational motion, which is governed by Newton’s second law.
The most expedient way to proceed is to return to Eq. (12.1.3), which is the
velocity continuity condition for any scattering problem,
n̄ · ∇ Ps = −n̄ · ∇ PI − iωρ0 n̄ · V̄ , x̄ ∈ S (12.3.23)
This relation allows us to regard the scattered field as the superposition from two
incident plane waves: the actual one, and one whose particle velocity is −V̄ . The
pressure amplitude of the latter is ρ0 cV. In other words, the effective incident pressure
is
(PI )effective = P̂I e−ik ēI .x̄ − ρ0 cV e−ik ēv ·x̄ (12.3.24)
The first place we introduce this replacement is the analysis of the monopole part
of the scattered pressure, Eq. (12.3.17). This effect does not depend on the direction in
which the incident wave propagates, but the phase lag between the incident velocity
and V̄ is 180◦ . Therefore, the monopole amplitude is
k2V k2V
A=− A=− P̂I − ρ0 c |V | (12.3.25)
4π 4π
The modification of the dipole moment entails replacement of PI ēI with PI ēI − ρ0 c V̄ .
Doing so converts the dipole moment in Eq. (12.3.18) to
ik
{D} = [V [I ] + [W ] ] P̂I {eI } − ρ0 c {V } (12.3.26)
4π
The last step in the analysis is determination of V̄ . With that as our objective, we
formulate Newton’s Second Law for the object. It equates the inertia of the body
to the resultants of the incident and scattered pressure distributions. Let M be the
body’s mass. The complex amplitude of its acceleration is iω V̄ , so it must be that

M iω V̄ = F̄I + F̄ (12.3.27)
The scattered pressure resultant is described by (12.3.20), with the incident pres-
sure set to PI ēI − ρ0 c V̄ ,

{F} = ik [W ] P̂I {eI } − ρ0 c {V } (12.3.28)
The resultant of the incident pressure is

F̄I = (−PI n̄) dS (12.3.29)
S
The sense of n̄ is outward from the region V enclosed by S, so the divergence theorem
converts this to

F̄I = − (∇ PI ) dV (12.3.30)
V
The scale over which PI varies is an acoustic wavelength, which is restricted here to
being much larger than the size of the body. Therefore, we may consider ∇ PI to be
a constant in this expression. For a plane wave, ∇ PI = −ik P̂I ēI exp (−ik ēI · x̄) . We
evaluate this gradient at the origin and invoke the smallness of k x̄, so that
F̄I = ikV P̂I ēI ⇐⇒ {FI } = ikV P̂I {ēI } (12.3.31)
Substitution of {F} and {FI } into Newton’s law gives

Miω {V } = ik [W ] P̂I {eI } − ρ0 c {V } + ikV P̂I {ēI } (12.3.32)
A rearrangement of terms yields an expression that may be solved for {V } ,

specifically

M P̂I
[I ] + [W ] {V } = [V [I ] + [W ]] {ēI } (12.3.33)
ρ0 ρ0 c
This expression confirms that it was correct to say in Chap. 7 that ρ0 [W ] is an added
mass.
The closure of the analysis uses the value of {V } found from the preceding to form
the dipole moment in Eq. (12.3.26). Then, the expressions for the monopole ampli-
tude and dipole moment are substituted into the mulipole expansion, Eq. (7.4.23).
Letting G (x̄0 , x̄C ) be the free-space Greens function ultimately leads to

k2
Ps (x̄0 ) = − V P̂I − ρ0 c |V | − {er }T [V [I ]
4π
i
−ikr
e
(12.3.34)
+ [W ] ] P̂I {eI } − ρ0 c {V } 1 −
kr r
As is true for scattering from a stationary rigid body, a quantitative evaluation of

this expression requires that we know [W ]. The values for a sphere and thin circular
disk are provided in Eq. (7.4.38) and at the concluding paragraph of the preceding
section.
The special case of a neutrally buoyant object, such as a submarine, is interesting.
The mass is the same as that of the displaced fluid, so M/ρ0 = V. According to Eq.
(12.3.33), this situation leads to ρ0 c V̄ = P̂I ēI , from which it follows that Ps = 0.
This situation occurs because the neutral buoyant body moves in unison with the
fluid, so it does not impede the propagation of the incident wave. This phenomenon
assumes that the acoustic wavelength is much larger than any dimension of the object.
502 12 Scattering
It also ignores the fact that no object is truly rigid. The role of structural flexibility
is examined in the last section of this chapter.
EXAMPLE 12.3 A thin aluminum disk is suspended in a large vat of water

in a manner that allows it to be rotated about its vertical diameter without
otherwise restraining the disk. A plane wave traveling horizontally is incident
on the disk, with the angle between −ēI and the normal to the disk being ψI . The
mass of the disk is 1.7 g, and its radius is 20 mm. The frequency of the incident
wave is 100 Hz, and its amplitude is 10 kPa. (a) Determine the magnitude and
direction of the disk’s complex velocity relative to the z direction as a function
of ψI . (b) Determine the amplitude of the backscattered pressure as a function
of ψI and compare the result to backscatter for the case of a stationary disk.
Significance
The focus is on the tensorial nature of the added mass effect and how it affects
the backscattered pressure. The quantitative results illuminate the general effect of
allowing an object to move.
Solution
The coordinate system and geometry for this system are shown in Fig. 1. The given
parameters correspond to ka = 8.491 10−3 , which certainly is sufficiently small to
warrant application of Rayleigh scattering theory. We begin by assembling the basic
quantities. According to Eq. (7.4.38), the virtual mass tensor is a single element,
3,3 = (8/3) a 3 . It is not surprising that only the 3,3 element is nonzero, because
incidence of a wave traveling parallel to the surface of the disk will not lead to
significant scattering. Another consideration is that V is not given. We could take it
be zero, based on thinness of the disk, but the monopole contribution is dominant at
orientations for which the dipole contribution is very small. Therefore, we determine
it from the mass, which gives V = M/ρal .
x
z
a
ψI
_
eI
a
Figure 1
Axisymmetry allows us to situate ēI in the x z-plane, so
{eI } = [sin ψI 0 − cos ψI ]T

All terms in Eq. (12.3.33) have been characterized, so we may solve that equation
for the complex velocity amplitude. Doing so leads to
⎡⎡ ⎤⎤−1 ⎡ ⎤
M M
⎥⎧ ⎫
0 0 0 0
⎢⎢ ρ0 ⎥⎥ ⎢ ρal
P̂I ⎢ ⎢ ⎥⎥ ⎢ ⎥ ⎨ sin ψI ⎬
⎢⎢ M ⎥⎥ ⎢ M ⎥
{V } = ⎢⎢ 0 0 ⎥⎥ ⎢ 0 0 ⎥ 0 (1)
ρ0 c ⎢ ⎢ ρ ⎥⎥ ⎢ ρal ⎥⎩ ⎭
⎣⎣ 0
M 8 3 ⎦⎦ ⎣ M 8 ⎦ − cos ψI
0 0 + a 0 0 + a3
ρ0 3 ρal 3
The matrix being inverted is diagonal, so we find that

ρ0 P̂I
Vx = sin ψI , Vy = 0
ρal ρ 0c (2)
M + (8/3) ρal a 3 ρ0 P̂I
Vz = − cos ψI
M + (8/3) ρ0 a 3 ρal ρ0 c
Both complex velocity components are real, so their relative phase is zero. This
corresponds to a temporal velocity vector that is oriented in a constant direction. It
lies in the x z-plane, so we may describe
V̄ in terms its magnitude |V | and angle
of
θ relative to the z-axis as Vx = V̄ sin θ, Vz = V̄ cos θ. (Both components are
required to assure that θ is assigned to the proper quadrant.) The velocity components,
as well as |V | and θ, are graphed in Fig. 2. Perhaps the most unexpected feature is
the fact that V̄ is parallel to ēz when ψI = 90◦ . Although the disk is thin, it moves at
this incidence because its volume is finite.
|V| Vx Vz
0.01
0.005
−0.005
−0.01
0 90 180 270 360
180
90
θ (deg)
−90
−180
0 90 180 270 360
ψI (deg)
Figure 2
504 12 Scattering
The complex velocity amplitude is used to compute (r/a) |PI | /(ρ0 c) according
to Eq. (12.3.34). We are interested in the farfield backscatter, so we set ēr = −ēI and
drop the term that is the reciprocal of kr. Thus, we have
r |P (x̄ )|
s 0 (ka)2 M
=− P̂I − ρ0 c |V |
a ρ0 c 4π ρal a 2
⎡ ⎤
M
⎧ ⎫T ⎢ ρ 0 0
⎥
sin ψI ⎬ ⎢ al M
(ka)2 ⎨ ⎢
⎥
⎥
+ 0 ⎢ 0 0 ⎥
4π ⎩ − cos ψ ⎭ ⎢ ρal ⎥ (3)
I ⎣ M 8 3⎦
0 0 + a
⎧ ⎧ ⎫ ⎧ ⎫ρ⎫al 3
⎨ ⎨ − sin ψI ⎬ ⎨ Vx ⎬⎬
× P̂I 0 − ρ0 c 0
⎩ ⎩ ⎭ ⎩ ⎭⎭
cos ψI Vz
Figure 3 is constructed by substituting {V } at each ψI into Eq. (3). The value of |Ps |
for the case of a stationary disk is obtained by setting {V } = {0} .
0.25
|Ps| |Ps| (stationary)
0.2
0.15
0.1
0.05
0
0 90 180 270 360
ψI (deg)
Figure 3
It can be seen that the backscatter is much greater if the disk is immobilized. An
explanation of this behavior may be found in an evaluation of the ratio of Vx and Vz
to the respective components of V̄I , which are VI sin ψI and −VI cos ψI . Equation (2)
indicate that

Vx ρ0 Vz M + (8/3) ρal a 3 ρ0
= = 0.370, = = 0.954 (12.3.35)
V̄I · ēx ρal V̄I · ēz M + (8/3) ρ0 a 3 ρal
Movement of the disk in the ēx direction leads to little scattering, so the mismatch
between Vx and V̄I · ēx has little effect on the scattering. In contrast, movement of the
disk in the ēz direction is important. This velocity component is very nearly equal to
V̄I · ēz . If they truly were equal, there would be no scattering. This phenomenon is
comparable to ocean waves. If they are incident on a stationary object, there is a large
splash. In contrast, a floating object induces little ripple. Another aspect to note is
that the scattered pressure is essentially zero at ψI = 90◦ and 270◦ . At these angles,
the contribution associated with [W ] is zero. The remaining terms are negligible
because M/ρal = V is much less than a 2 .
12.4 Measurements and Metrics 505
12.4 Measurements and Metrics
Acoustic scattering is the core phenomenon for a number of applications, ranging

from biomedical ultrasonics to fishing to naval sonar. Thus, it is undoubtedly true
that scattered signals are measured much more frequently than they are evaluated
analytically. This bias is even stronger if we include marine mammals and bats, who
have remarkable capabilities to “see” with sound.
Sonar has two implementations. Passive sonar is a process of listening, in that
it detects sound radiated by a body. If the object that is to be located is stationary
and does not vibrate, it will emit no sound and therefore cannot be located. Active
sonar does not suffer from this shortcoming. It is used recreationally and commer-
cially to locate fish. Typically, this entails broadcasting a signal from the boat, then
measuring the signal returned to the boat. Suitable processing of the returned signal
can determine whether it is scattered by fish, and if so, where the fish are located.
This application is a monostatic scattering measurement. The “mono” prefix is used
because signal generation and measurement are at a single location. Many of the
sonar concepts follow those for radar, wherein “static” refers to the apparatus being
stationary. Monostatic scattering is the primary concept for biomedical ultrasound,
wherein the device that generates the probing signal also detects the signal scattered
by a heterogeneous region, such as a tumor.
Naval sonar systems applications rely on both monostatic and bistatic scattering
measurements. In a bistatic measurement, the incident signal is generated at one
location and the scattered signal is detected at another. This can be done from an
airplane or helicopter by dropping sonobuoys, or it can be done from ships and
submarines by using a towed array.
Ultrasonic imaging for medical diagnostics is the acoustical analog of our visual
sense, in that they rely on contrast. The field scattered by a heterogeneity is scanned
spatially. Strong returns correspond to greater impedance mismatch, so they are
displayed as lighter regions. Weaker returns correspond to little or no heterogeneity,
and therefore are displayed as dark regions. The field properties of the scattered field,
such as its multipole resolution, typically is not an issue in biomedical applications.
The underlying properties of the scattered field are of vital importance to sonar
applications. The general objective there is to deduce the nature of the scattering
body from measurements. We shall not delve into how the measured signal is used
to identify the fundamental nature of the scattering body, which is a process of target
identification. However, if we consider the result for the field scattered by a rigid
body, we can gain a hint at the possibility of using scattering data to characterize the
scatterer. The virtual mass matrix ρo [W ] in Eq. (12.3.19) is symmetric, so it contains
six unknown elements. Another unknown parameter is the volume of the scattering
body. Thus, if we have several measurements, corresponding to angles of incidence
and/or different measurement locations, it is conceivable that we could infer these
unknown parameters.
Obviously, this notion relies on several assumptions. Several sources and/or
receivers are required to obtain bistatic measurements. It is necessary to determine
506 12 Scattering
the location of the body. Range information can be determined by using a pulse,
rather than a continuous wave. The simplest measurement uses the time τ for a sig-
nal to return to the source location in a monostatic measurement. Because the travel
time to the body is τ /2, the range will be cτ /2. Orientation of the line from the
source/receiver location to the body can be determined by rotating a highly direc-
tive source until the scattered signal is largest, or by using a phased array. Another
assumption is that the orientation of the scattering body does not change between
measurements. This can be addressed by taking the measurements over a short inter-
val. The most fundamental assumption buried in the notion of identifying V and []
is that the analysis is valid. If the body is flexible, then the scattered field will be much
more complicated than a simple superposition of monopole and dipole fields. Indeed,
in that case, the nature of the field is likely to be highly dependent on the frequency.
An efficient way of performing scattering measurements over a frequency range is
to alter the pulse from a pure tone burst. The shape of the envelope can be adjusted
(amplitude modulation), and the frequency can be varied (frequency modulation)
within the pulse’s interval. Then, FFT processing of the scattered signals, which also
will be pulses, will yield frequency spectra that can be converted to transfer functions.
Unless the duration of the pulse is very long relative to the range divided by the
sound speed, the scattered signal will be measured distinctly from the incident. The
logical question is what to do with it? Other than travel time and location, the most
fundamental metric is based on a concept developed for radar. It originates from
a hypothetical situation. Suppose a planar harmonic wave is normally incident on
a rigid stationary disk. Let us further suppose that the planar wave is a confined
beam whose cross section exactly matches the disk. This is the situation depicted in
Fig. 12.3. In the high-frequency limit of ray acoustics, the scattered signal Ps is the
regular reflection of the incident signal, corresponding to a reflection coefficient of
one. In that case, the power transported in the reflected signal is the same as that of the
incident signal. The magnitude of the time-averaged intensity in the incident wave is
II = |PI |2 / (2ρ0 c), so the power in the scattered signal is (Ps )av = II A, where A is
the cross-sectional area of the side of the disk that is irradiated by the incident signal.
Division of the time-averaged power in the reflected wave by the intensity of the
incident wave yields a measure of the cross-sectional area that blocks the incident
wave.
Fig. 12.3 Model of PI

scattering from a disk used to
define the scattering cross Ps Ps
PI
section. The picture
generally is not descriptive of PI Ps
the actual scattering process
PI Ps
PI
Ps
To extend this idea to the actual field scattered by an arbitrary body, the time-
averaged power transported in the scattered signal is computed by integrating over
a sphere S0 that surrounds the scatterer. The spherical coordinates of a point on the
sphere’s surface are r0 , ψ0 , θ0 . The radius r0 of this sphere is taken to be much greater
than 2π/k, as well as the largest dimension of the body. Then, the radial velocity
on the sphere’s surface is Vr = Ps (r0 , ψ0 , θ0 ) / (ρ0 c) . The general definition of the
scattering cross section is that it is the time-averaged power transported by the
scattered signal divided by the intensity of the incident wave,

(Ps )av Is (r0 , ψ0 , θ0 ) |Ps |2
σs = = dS0 , Is = (12.4.1)
II II 2ρ0 c
S0
The intensity in the farfield is proportional to 1/r02 , so setting dS0 = r02 (sin ψ0 )
dθ0 dψ0 will yield a value for σs that is independent of r0 .
In the hypothetical model of scattering from a disk, σs = A. This leads to inter-
pretation of σs as the apparent area of the scattering object that blocks the incident
wave. Because the field scattered by an arbitrary body depends on the orientation of
ēI relative to the body, the scattering cross section depends on that orientation. The
value of σs might exceed the area of the body’s projection on to the plane normal to
the ēI , or it might be much smaller.
Consider the case of scattering from a stationary rigid body. Equation (12.3.19)
describes the scattered field as the sum of a monopole whose strength is propor-
tional to the displaced mass ρ0 V, and a dipole whose moment is proportional to
[V [I ] + [W ]] {eI } . It is useful to identify the direction of the dipole moment cor-
responding to a known [W ] because scattering associated with the dipole will be
strongest in that and the opposite direction. Toward that objective, we write the vec-
torial part of the dipole moment in terms of its magnitude and a unit vector {e D }
that is the direction of the vector. In other words, we set {e D } = [V [I ] + [W ]] {eI } .
It follows that
[ V [I ] + [W ] ] {eI }
= |[ V [I ] + [W ] ] {eI }| , {e D } = (12.4.2)

This transformation yields a description of {e D } in terms of components relative
to the x yz coordinate system associated with the known components of [W ] . We
also represent the direction {er } from the origin to a field point on the surface of
the surrounding sphere S0 in terms of components relative to the same coordinate
system. Doing so allows us to evaluate product {er }T [[V [I ] + [W ]]] {eI }, which
now is {er }T {e D } . In other words, the directivity of the dipole is the dot product
of ēr and ē D , and the strength of the dipole is proportional to . Let ψ0 be the angle
between the vectors, that is,

ψ0 = cos−1 {er }T {e D } (12.4.3)
508 12 Scattering
This corresponding representation of the farfield pressure scattered by a stationary

rigid body is
k 2 P̂I e−ikr0
Ps (x̄0 ) = − [V − cos ψ0 ] (12.4.4)
4π r0
Integration of this expression over the surface of the sphere yields the radiated
power. We define the polar direction for a set of spherical coordinates to be ē D . The
field is axisymmetric with respect to an axis aligned with {e D } , so the differential
2

area element is set to 2πr02 sin ψ0 dψ0 . Division of Ps by |II | = P̂I / (2ρ0 c) yields
the scattering cross section. The result is

k4 1
σs = V + |[] {eI }|
2 2
(12.4.5)
4π 3
It should be noted that the integration over S0 has removed any dependence on the
orientation of the dipole moment, so there is no need to actually evaluate {e D } .
Specific results are
7 2
Sphere: σs = πa (ka)4
9
(12.4.6)
16 2
Thin circular disk: σs = πa (ka)4 (cos ψI )2
27π 2
where ψI is the angle between the normal to the disk and the incident wave’s direction.
In the case of a sphere, the factor πa 2 is the physical area that blocks the inci-
dent wave at any angle. For a thin disk, the maximum blocking area is πa 2 , which
corresponds to normal incidence. At arbitrary incidence, the physical blocking area
is πa 2 cos ψI , so the above expression indicates that the decrease of σ when ψI is
increased is due to more than the decrease of the blocking area. The Rayleigh scat-
tering limit describes cases where ka
1, so the dependence of σs for each shape
on (ka)4 tells us that the bodies appear to be much smaller than their cross-sectional
area.
The scattering cross section represents a spatial average of what is seen at various
field points. A different metric is used to convey what is seen at a specific location.
This metric is defined in terms of a solid angle, which is a geometric parameter we
encountered in Sect. 7.4.1. A solid angle will arise in the next section, so it is helpful
to review the concept. Consider the normal n̄ (x̄s ) to the surface. The intersection of
the surface and any plane that contains this normal is a curve. We are interested in the
differential arc of this curve that is centered at x̄s . The center of curvature of this arc
is the intersection of n̄ (x̄s ) and any other normal to the arc, such as that at either end.
The distance from the center of curvature to the surface is the radius of curvature of
the arc. The plane containing n̄ (x̄s ) is defined by its rotation about n̄ (x̄s ) . At some
rotation angle, the arc has the largest radius of curvature. A further rotation by 90◦
forms an arc with the smallest radius. These planes and arcs are depicted in Fig. 12.4.
The radii r1 and r2 are the principal radii of curvature. (For a sphere r1 = r2 = a,
whereas r1 = a and r2 is infinite for a cylinder.) The arclengths are d1 = r1 dθ1 and
d2 = r1 dθ2 where dθn is the angle between the normals to the surface at the ends of
dn . The area of the patch that is formed is dS = d1 d2 . The definition of a solid
angle is
dS
d = (12.4.7)
r1 r2
where is the symbol we shall employ for the solid angle. Thus, the solid angle
subtended by the patch is d = dθ1 dθ2 .
Fig. 12.4 Surface patch r1

formed from arcs whose dl 2
radii of curvature are the
principal values at a point
dS
--)
n(xs r2
dl1
The scattering cross section is defined in Eq. (12.4.1) as an integral over a sphere.
Both principal radii of that surface are r0 , so the differential element of area may
be written as dS0 = r02 d. Concurrently, any integral may be represented as the
antiderivative of its integrand. Thus, the definition of σs may be written in two ways,
as

Is (r0 , ψ0 , θ0 ) 2 dσs
σs = r0 d = d (12.4.8)
II d
S0 S0
The quantity dσs /d is the differential scattering cross section. Matching the above
integrands gives
dσs Is (r0 , ψ0 , θ0 ) 2 r02 |Ps (r0 , ψ0 , θ0 )|2

= r0 = (12.4.9)
d II |PI |2
In other words, dσs /d is proportional to the directivity of the farfield intensity. It
is implicit to this expression that the scattering cross section is defined for a specific
orientation of ēI relative to the body, so that ψ0 and θ0 are defined relative to a body-
fixed direction. If we define that direction to be ēr , then backscatter corresponds to
ēr = −ēI (ψ0 = π). A measurement at an arbitrary field point gives the differential
scattering cross section in a bistatic measurement. The special case where ēr = ēI
(ψ0 = π) is said to be forward scatter.
510 12 Scattering
A quantity that causes confusion for some is the backscatter cross section, which
is denoted as σback . Despite the name, it is not derived from the actual scattering cross
section. Rather, it is the value of the scattering cross section that would be obtained
if the scattering was isotropic at the backscatter value. Accordingly, if dσs /d were
constant at its value for ēr = −ēI , integrating it over S0 would give a 4π factor, so
that

dσ Is (r0 , π, 0) Ps (r0 , π, 0) 2
σback = 4π = 4πr02 = 4πr02
(12.4.10)
d ēr =−ēk II PI
Like the other metrics, σback depends on the orientation of the incident signal relative
to the body. Because |Ps | is proportional to 1/r0 , the value of σback is independent
of r0 .
Typically, the backscatter cross section is described logarithmically as the target
strength. The definition is

σback
T S = 10 log 2
(12.4.11)
4πrref
where rref is a reference distance; the standard value is rref = 1 m. An expression for
the target strength in terms of measured pressure values results from substitution of
the definition of σback and Eq. (12.4.9) into the above. Let Lback be the sound pressure
level at the source/receiver location in a monostatic measurement, and let LI be the
sound pressure level in the incident wave at that location. Then

r0
T S = Lback − LI + 20 log (12.4.12)
rref
Measurement of backscatter calls for a monostatic setup, in which the source and
receiver are collocated, and possibly the same device. The usage of the above to
evaluate target strength requires that the incident signal not be included in Lback .
This condition will be attained if the incident signal is a pulse that terminates before
the scattered signal returns to the receiver.
12.5 High-frequency Approximation
Low-frequency approximations of scattering phenomena are associated with a great

acoustician, Rayleigh. It is appropriate that another pioneer, Kirchhoff, should be
associated with high-frequency scattering. Kirchhoff approximations are essentially
applications of ray acoustics to the scattering problem. We will restrict our attention
to the usual case wherein the incident wave is planar.
12.5 High-frequency Approximation 511
We begin with an analysis of backscatter from a rigid, stationary body. The first
task is to locate the point on the surface at which the normal is opposite ēI . This
location depends on the shape of the body and the orientation of ēI relative to the
body. This point is labeled as C in Fig. 12.5, which is a side view in the first plane of
principal curvature. The intensity of the backscattered wave is found by analyzing
the reflection of the incident ray tube that arrives at a differential element dS of
surface area that is centered at point C. The edges of dS are the arcs d1 = r1 dθ1
depicted in Fig. 12.5 and d2 = r2 dθ2 in the second plane of principal curvature.
The angles dθ j are the angles subtended by each relative to the respective centers of
curvature.
Fig. 12.5 Description of _

backscatter as a process of vs
specular reflection
d1/2
_
vI
d l1 d1/2
d1
d1/2 O
_ _ _
O’
eI n(xC) C d1/2
d1
d1/2
_
vI r1
d1/2
_
vs
The intensity of the incident plane wave is I¯I = ēI |PI |2 / (2ρ0 c) everywhere, so
the power transported by this ray tube is
dPI = II dS = IIr1r2 dθ1 dθ2 (12.5.1)
In the geometrical acoustics approximation , the reflection of each ray is like it

would be if the surface was an infinite plane with normal n̄. The angle of incidence
at the extremities of the ray tube in Fig. 12.5 is dθ1 /2. The angle of reflection equals
the angle of incidence, so the reflected rays are confined to a tube that is bounded
by rays that are at angle dθ1 relative to −ēI . The consequence is that the reflected
rays in this bundle represent a spherical wave that radiates from a virtual source
at point O . (There is a single center for all rays because the angle of incidence
for all is infinitesimal. If one were to consider incident rays at large distances from
the backscatter centerline, the reflected rays would appear to radiate from centers
distributed along the centerline behind point C. This is a manifestation of spherical
aberration, which we first encountered in Example 11.1.)
512 12 Scattering
The reflected rays in the plane of Fig. 12.5 are centered at point O . At a radial
distance r0 , the arc subtended between the extremities of d1 is r0 (2dθ1 ) . A similar
description applies to the other principal plane of curvature, so the arc subtended
by the rays in that plane is r0 (2dθ2 ) . Thus, the reflected rays at distance r0 cross a
differential surface patch whose area is dS0 = 4r02 dθ1 dθ2 . All rays contained within
the scattered ray tube have the same amplitude, so the intensity along dS0 is constant
at Is . The power transported across this patch in the scattered wave is

dPs = Is dS0 = Is 4r02 dθ1 dθ2 (12.5.2)
The surface is rigid, so the reflection coefficient for each incident ray is one. It
follows that the power flow in the tube of scattered rays must equal the power flow
in the tube of incident rays. Equating dPs to dPI yields
r1 r2
Is = II (12.5.3)
4r02
The backscatter cross section is found by using this intensity to form Eq. (12.4.9),
which leads to
σback = πr1r2 (12.5.4)
A review of the derivation will reveal that it assumes that there is a single point
C on the surface at which the normal is exactly opposite to ēI . If there are multiple
points that meet this criterion, the conservation of energy argument may still be
applied, provided that it accounts for the contribution from each scattering site. Each
contribution is a tube centered on direction −ēI . Because the scattering sites are
separated by a finite distance, the tubes would not intersect. Hence, the intensity of
each tube may be added as scalars to find the backscatter cross section. It also is
possible that there is no point at which there is normal incidence. An example of this
condition is a thin disk whose normal is not aligned with ēI . Another assumption
is implicit, specifically, that the surface is convex. The situation where the surface
is concave at the normal incidence point can be analyzed by merging the ray tube
approach here with the analysis of focusing carried out in Example 11.1.
The preceding analysis of backscatter may be generalized to predict bistatic scat-
tering. The first step in doing so is to locate point C at which the normal to the
surface is such that the angle of reflection to a designated field point equals the angle
of incidence. From there, the derivation invokes concepts from differential geometry.
The interested reader is referred to Pierce’s presentation.3
It is much easier to evaluate the scattering cross section σs for a rigid body in the
high-frequency limit. Figure 12.6 depicts all rays that are incident on the illuminated
side of the body. They are confined to a (general) cylinder. The projection of the
body’s shape onto a plane perpendicular to ēI defines the cross section SI of this
bundle.
3 A.D. Pierce, Acoustics, Mc-Graw-Hill (1981), ASA reprint, (1989), Sect. 8-8.
Fig. 12.6 Bundle of rays of

a plane wave that is incident
on body S from a cylinder
whose cross section is SI
S
SI
_
eI
The power flow in this bundle is II SI . Although the incident rays are reflected
over a wide range of directions, their reflection is lossless. Thus, the power scattered
on the illuminated side of the body equals the power contained in the incident ray
tube. On the shadow side, the total pressure is essentially zero, but the total pressure
is the sum of incident and scattered contributions. Thus, the power in the scattered
field on that side equals the power flow in the bundle of incident rays that appear to
have passed through the body. This is the same ray tube as that which is incident on
the illuminated side, so the power in the scattered field on the shadow side equals the
power flow in the incident side, which is II SI . The total power flow in the scattered
field, therefore, is twice this value. Hence, we find from Eq. (12.4.9) that
σs = 2SI (12.5.5)
It is interesting to consider the geometrical acoustics results for a sphere and

a thin circular disk. Both principal radii for a sphere equal the radius a, and the
perpendicular area is SI = πa 2 . If the angle between the normal to the disk’s surface
and the incident direction is χ, the cross section blocked by the disk is an ellipse
whose semi-major radii are a and a cos χ. Therefore, the incident ray tube area is
SI = πa 2 cos χ. The principal radii of curvature for the disk are infinite because the
surface is flat. It follows that
Sphere: σs = 2πa 2 , σback = πa 2

(12.5.6)
Disk: σs = 2πa 2 cos χ, σback = ∞
The infinite value might seem surprising, but its meaning becomes obvious when we
recall the general interpretation of a scattering cross section as the area that blocks
the incident wave. Consider the reflection from an infinite plane surface. An incident
wave at normal incidence is reflected backward, so that I¯s = − I¯I everywhere. To
evaluate the scattered power in this case, we could define S0 to be a hemisphere of
radius r0 centered on surface, then let r0 → ∞. The result would be an infinite
the
value of Ps because I¯s is independent of r0 . Thus, the scattering cross section σ of an
infinite wall is infinite. The backscatter cross section takes the backscatter differential
cross section to be representative of the scatterer’s properties in all directions. In other
514 12 Scattering
words, it is infinite because it does not recognize the fact that the surface has finite
extent.
The preceding result may be extended to treat incidence on a surface that is locally
reacting, with impedance ρ0 cζ. Backscatter corresponds to normal incidence, so the
coefficient of reflection is
ζ −1
R= (12.5.7)
ζ +1
The geometrical features of the reflection process are unaltered. Thus, the tube of
scattered rays has the same extent as it does for the case where the surface is rigid.
From the analysis in Sect. 5.2.2, we know that the power flow in the reflected ray
tube as it leaves the surface is dPs = |R|2 dPI . The time-averaged intensity in the
tube when it crosses the surrounding sphere still is Is = dPs /dS0 . The result is that
the previous expression for the farfield scattered intensity becomes
r1 r2
Is = II |R|2 (12.5.8)
4r02
The corresponding expression for the backscatter cross section is
σback = πr1r2 |R|2 (12.5.9)
We know that |R| < 1 for any passive nonideal material, which means that rigid and
pressure-release conditions lead to the largest backscatter at high frequencies.
The geometrical acoustics approximation may also be invoked to analyze bista-
tic scattering. The general approach entails formulating the KHIT with the surface
response described as though the surface is locally planar. Thus, the reflected signal
constitutes the scattered field on the surface. The reflection coefficient for the ray
that is incident at surface point x̄s is
ζ ēI · n̄ (x̄s ) − 1
R (x̄s ) = (12.5.10)
ζ ēI · n̄ (x̄s ) + 1
In the geometrical acoustics limit, rays are incident on the illuminated side only.
Correspondingly, the pressure and the particle velocity are taken to be zero on the
shadowed side. This region is identifiable by ēI being outward from the surfaces.
Thus, the total pressure and normal velocity on the surface are
⎧
⎨ P (x̄s ) = [1 + R (x̄s )] PI (x̄s )
ēI · n̄ (x̄s ) < 0 =⇒ 1
⎩ V̄s (x̄s ) · n̄ (x̄s ) = [−1 + R (x̄s )] PI (x̄s ) ēI · n̄ (x̄s )
ρ0 c
ēI · n̄ (x̄s ) > 0 =⇒ P (x̄s ) = 0 & V̄s (x̄s ) · n̄ (x̄s ) = 0
(12.5.11)
To implement Eq. (12.1.5), we could define ēz to align with ēI , and place the origin
at the centroid of the body. Then, if we set the phase of the incident plane wave to
zero at the origin, we would have
PI (x̄s ) = P̂I e−ik x̄s ·ēz (12.5.12)
Substitution of these representations into the KHIT would define all of the terms
in the integrand, but that does not mean we can directly proceed to evaluating the
integral numerically. One difficulty that awaits is the necessity to formulate a way of
describing x̄s and n̄ (x̄s ) in a way that fits the selected integration scheme. This is a
standard issue for any formulation of KHIT. The more difficult feature stems from
the fact that this is a high-frequency theory. Thus, k x̄s · ē3 will be much greater than
unity over most of the surface. This means that the integrand of Eq. (12.1.5) would be
a rapidly oscillating function of position. It is difficult to obtain a convergent result in
such situations without expending a great deal of computational effort. Asymptotic
methods, such as the method of stationary phase,4 might be useful for such efforts.
12.6 Scattering from Spheres
We have analyzed scattering from a variety of bodies in the low-frequency limit of

the Rayleigh approximation and the high-frequency limit of the Kirchhoff approx-
imation. However, we have no idea of the actual frequency range for which either
approximation is valid, nor of how accurate either is in its range of validity. Answer-
ing these questions for an arbitrary scattering object would require considerable
computational or experimental work. Fortunately, the spherical shape is amenable to
analytical investigations. As was noted at the outset, a basic concept for an analysis
of scattering is to convert the problem to an analogous radiation problem. Radiation
from spheres was the subject of Sect. 7.2.
12.6.1 Stationary Spherical Scatterer
For scattering from a sphere, there is no special orientation of the polar axis. There-
fore, we align the z-axis with the propagation direction of the incident plane wave,
ēz = ēI . The origin of the coordinate system is situated at the center of the sphere, and
the phase of the incident wave is defined to be zero at that location. Thus, the incident
signal is P̂I exp (−ikz) . In spherical coordinates, the axial distance is z = r cos ψ.
4 C.M. Bender & S.A. Orszag, Advanced Mathematical Methods for Scientists and Engineers (1978).
516 12 Scattering
An identity known as an addition theorem5 converts the exponential to a spherical

harmonic series,
∞

PI = P̂I e−ikz = P̂I (2m + 1) (−i)m jm (kr ) Pm (cos ψ) (12.6.1)
m=0
Rigidity requires that the surface velocity be zero. The surface normal is the radial
direction. Correspondingly, continuity of the normal velocity on the surface requires
that
∞
∂ Ps ∂ PI
=− = −k P̂I (2m + 1) (−i)m jm (ka) Pm (cos ψ) (12.6.2)
∂r r =a ∂r r =a m=0
A spherical harmonic series for Ps satisfies the Helmholtz equation and the
Sommerfeld radiation condition. The incident field is independent of the azimuthal
angle θ, which means that the scattered field will be axisymmetric relative to the z-
axis. Therefore, the scattered pressure may be represented as a series of axisymmetric
spherical harmonics, see Eq. (7.2.1),
∞

Ps = Bm h m (kr ) Pm (cos ψ) (12.6.3)
m=0
Recall that the spherical Hankel functions appearing here are the second kind, which
correspond to waves that propagate outward when the convention is to use exp (iωt).
The series coefficients are determined by using this form to satisfy the surface bound-
ary condition. Doing so leads to
jm (ka)
Bm = − (2m + 1) (−i)m P̂I (12.6.4)
h m (ka)
The corresponding expression for the scattered pressure at any location is

∞
jm (ka)
Ps = − P̂I (2m + 1) (−i)m h m (kr ) Pm (cos ψ) (12.6.5)
m=0
h m (ka)
The series expansion may be specialized to the farfield by applying the asymptotic
expansion in Eq. (7.1.35), which is
i m+1 −ikr
h m (kr ) → e (12.6.6)
kr
5 M.I.Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, (1965), p. 440,
Eq. 10.1.47.
12.6 Scattering from Spheres 517
This limiting form is valid if kr m. This attribute might seem to be problematic

because the spherical harmonic series sums over all m. Fortunately, the Bm coeffi-
cients decrease rapidly as m increases. (To verify this behavior, consider Eq. (7.1.36),
which describes the behavior of the spherical functions as m increases.) Thus, for
any kr, there always is a value m = M for which B M is sufficiently small to permit
truncation of the series. We may use Eq. (12.6.6) to describe terms for m ≤ M. We
thereby obtain
a M
j (ka)
Ps = PI e−ikr χ (ψ, ka) , χ (ψ, ka) = −i (2m + 1) m Pm (cos ψ)
r m=0
kah m (ka)
(12.6.7)
One approach to evaluate the series is to accumulate the contributions of increas-

ing m until there is negligible change in the value of the sum. However, modern
software is most efficient when the operations are vectorized, which requires that the
range of the summation be set prior to the actual computation. A viable approach
sets the maximum m as an integer multiple of ka. If that multiple is too large, the
value of jm (ka) / h m (ka) will cause an underflow; the results reported here were
obtained using max(m) = ceil(5ka) . Vectorization is achieved by defining acol-
umn vector {F} that consists of the values of (2m + 1) jm (ka) / kah m (ka) for
0 ≤ m ≤ max (m) . If the Legendre function values Pm (cos ψ) are stored in a col-
umn vector {L} , then the summation reduces to {F}T {L} .
The function χ (ψ, ka) is analogous to the directivity function for radiation. From
it, the differential scattering cross section Eq. (12.4.9) may be evaluated as
dσ
= a 2 |χ (ψ, ka)|2 (12.6.8)
d
The backscatter cross section is obtained from the time-averaged intensity at ψ =

180◦ , for which Pm (−1) = (−1)m , so that
∞
jm (ka)
χ (π, ka) = (−1)m (2m + 1) (12.6.9)
m=0
kah m (ka)
The backscatter cross section is
σback = 4πa 2 |χ (π, ka)|2 (12.6.10)
For comparison, the value of σback in the Rayleigh scattering limit may be found
from Eq. (12.3.22) by setting W3,3 = 2πa 3 , ψI = 0, and ψ0 = −π, whereas the
geometrical acoustics limit is independent of ka,
518 12 Scattering
25 2
σback = πa (ka)4 , ka
1
9 (12.6.11)
σback = 2πa 2 , ka 1
In Fig. 12.7, the computed backscatter cross section is compared to the low- and
high-frequency approximations. The lower graph shows that the Rayleigh limit is
indistinguishable from the spherical harmonic series up to ka = 0.6, while the geo-
metrical acoustic approximation essentially is close to the mean about which the com-
puted value fluctuates when ka > 10. The amplitude of this fluctuation decreases as
ka increases, with an excursion that is less than 5% of the mean value above ka = 20.
1.5
back/(a2)
1
0.5
Spherical Rayleigh Geometrical
harmonics limit acoustics
0
0 10 20 30 40 50
100
back/(a2)
10-2
10-4
10-6
0 0.2 0.4 0.6 0.8 1
Frequency ka
Fig. 12.7 Frequency dependence of the backscatter cross section of a rigid, stationary sphere
Evaluation of the scattering cross section requires integration of the time-averaged

bistatic intensity. This quantity in the farfield is ēr |Ps |2 / (ρ0 c) . It contains all cross
products of spherical harmonics. However, an evaluation of radiated power entails
integration over the surrounding sphere, and the Legendre functions are orthogonal
over that domain, see Eq. (7.1.16). Orthogonality was shown in Eq. (7.1.17) to
decouple the spherical harmonics in regard to evaluation of radiated power. In the
present context, the radiated power in the scattered field is
π
Ps Ps∗
Ps = 2πr 2 sin ψdψ
0 2ρ0 c 2
∞ (12.6.12)
|PI |2 jm (ka)
= 4πa (2m + 1)
2
2ρ0 c m=0 ka h m (ka)
The factor preceding the summation is the intensity of the incident wave, so the
scattering cross section of a rigid sphere is
∞
2
Ps jm (ka)
σs = = 4πa 2
2 (2m + 1) (12.6.13)
II m=0
ka h m (ka)
Figure 12.8 shows that unlike the backscatter cross section, σs monotonically
increases toward the geometrical acoustics limit, σs = 2, without any fluctuation.
The value of σs in the Rayleigh limit, see Eq. (12.4.6), closely matches the computed
function for ka < 0.45.
2
1.5
/(a2)
1 Spherical Rayleigh Geometrical

0.5
harmonics limit acoustics
0
0 10 20 30 40 50
Frequency ka
Fig. 12.8 Frequency dependence of the scattering cross section of a rigid, stationary sphere
Before we examine why σback and σs behave differently, let us examine the scat-
tering response from the perspective of the scattering directivity function χ (ψ, ka).
Graphs like those in Fig. 12.9 were published by Stenzel in 1938,6 which is quite
impressive given the nonexistence of electronic computing. At low frequencies, the
angular dependence of χ fits the Rayleigh limit in Eq. (12.3.22), with the mono-
pole and dipole contributions being in-phase for ψ > 90◦ and oppositely phased
for ψ < 90◦ . As the frequency increases, side lobes become increasingly numerous,
ka = 0.2 ka = 2
_ _
eI eI
0.02 0.25
0.04 0.5
2 2.5 5
_ _
eI 1 eI
ka = 5 ka = 10
Fig. 12.9 Scattering directivity function χ (ψ, ka) as a function of the observation angle relative
to the propagation direction ēI of the incident wave
6 H. Stenzel, “On the perturbation of a sound field caused by a rigid sphere,” Electr. Nachrichtentech.,
vol. 15, (1938) pp. 71–28.

520 12 Scattering
which is a trend for acoustic radiation problems. An unexpected trend that grows as
the frequency increases is strong forward scattering, in which the maximum value of
χ increases and the beamwidth decreases as ka increases. This behavior is not pre-
dicted by ray theory, according to which the incident signal should be blocked in the
shadow of the sphere. Because P = Ps + PI , such reasoning leads to the expectation
that forward scattering should be such that |Ps | = |PI | .
The inability of geometrical acoustics to predict forward scattering is not a failure
of the theory. This phenomenon is another diffraction effect. Such effects arose in our
study of geometrical acoustics, which does not predict that sound will be heard on the
quiet side of a caustic. It also arose there as the fact that a ray at critical incidence to
the interface of two fluid might propagate along the interface, then re-emerge into the
transmitting fluid. The term “diffraction” refers to any high-frequency phenomenon
that cannot be described by ray acoustics. The process that ray acoustics missed in
scattering is creeping waves. Their existence would be difficult to detect from direct
evaluations of the spherical harmonic description of the pressure field. At the same
time, availability of a series solution is crucial, because it is the beginning point for
a mathematical proof of the existence of creeping rays.
Watson in 19187 addressed the fact that series such as those derived here require an
increased number of terms to converge as the frequency increases. Only rudimentary
computational tools were available, so he needed to be “smart”. He developed a
technique that transforms the series to a contour integral. Evaluation by the methods
of complex analysis produces a solution in the form of a series of residues. Each
residue corresponds to a pair of waves that pass a point on the surface. Figure 12.10
provides a qualitative picture of the first pair, which are the dominant contributors to
the signal.
The incident rays depicted in Fig. 12.10 touch the sphere at ψ = 90◦ , where they
are tangent to the sphere. In other words, they are at grazing incidence. They launch
rays that propagate along meridian circles (constant θ) in the sense of decreasing ψ,
which is the reason they are said to creep over the surface. Ultimately, they focus at
the shadowed pole (ψ = 0). They emerge from there as waves that travel in the sense
of increasing ψ along the diametrically opposite meridian, for which the azimuthal
Fig. 12.10 Incident rays PI PI

that are at grazing incidence
PI (Ps )creep
curve around the sphere as
creeping waves (the dotted (Ps )creep
lines). They circumnavigate (Ps )creep
PI
the sphere along meridians (Ps )creep 
PI
(Ps )creep 
PI
7 G.N. Watson, “The Diffraction of Electric Waves by the Earth; the Transmission of Electric Waves
Around the Earth,” Proc. Royal Soc. London A95 (1918), pp. 83–88, 546–563.
angle is −θ. In other words, an individual ray of the creeping wave follows a great
circle that contains the polar axis. Each ray circumnavigates the sphere along its great
circle. Eventually, they return to the location where they were launched. From there,
the rays execute another circumnavigation, then another, and so on. Each passage is
described in Watson’s solution by another residue in the complex integration.
The reason it is sufficient to only account for the first circumnavigation is that the
wavenumber for these rays is complex, so they attenuate as they propagate. Because
this is a high-frequency phenomenon, the behavior in the vicinity of any surface patch
is like what it would be if that patch laid in a planar boundary. Thus, the surface ray
launches a tangential ray. This signal is said to be shed from the surface. The shedding
at a few surface locations is depicted in Fig. 12.11, but it occurs continuously along
the surface. These shed waves are the agents that illuminate the shadow zone.
Fig. 12.11 Rays shed from Ray of a shed wave

creeping wave on the surface
of a sphere
Ray of a creeping wave
The wave that is shed in the backscattering direction combines with the reflected
signal predicted by geometrical acoustics. The latter is referred to as the specularly
reflected wave. The fluctuations of σback displayed in Fig. 12.7 stem from constructive
and interference between the contribution of the creeping waves and the specularly
reflected wave. As the frequency increases, the decay rate for the creeping waves
increases. Above ka = 20, this attenuation is sufficiently strong to make shedding
in the backscatter direction negligible. This leaves only the contribution of specular
reflection, which explains why the computed σback differs little from the geometrical
acoustic limit at very high frequencies.
Diffraction is responsible for acoustical phenomena in other systems. Sometimes,
we can identify the mechanism directly from a mathematical analysis. For example,
the side lobes of a sound beam are a diffraction effect associated with the discontinuity
of the velocity at the edge of a piston in a baffle, which leads to center and edge waves
for the on-axis signal. In other situations, specialized analyses based on ka being
very large are required. This is the situation regarding the Watson transformation.
The mathematical sophistication required to carry out such analyses exceeds the
level set for this text. An excellent introduction is available in the text by Junger and
Feit.8 Pierce9 explores the same phenomena, as well as diffraction around wedges
8 M.C. Junger and D. Feit, “Sound, Structures, and Their Interaction,” 2nd ed., MIT Press, ASA
reprint, Chap. 12 (1986).
9 A.D. Pierce, Acoustics, Mc-Graw-Hill (1981), ASA reprint, (1989), Chap. 9.
522 12 Scattering
and in ocean acoustics, and Keller’s geometrical theory of diffraction10 offers much
in terms of comprehensiveness and generality.
12.6.2 Scattering by an Elastic Spherical Shell
Acoustic radiation from a spherical shell was shown in Sect. 7.2.4 to be profoundly
dependent on frequency. At the resonance frequencies, the sole agent sustaining
any excitation applied to the shell is the resistive part of the acoustic loading. Now,
consider the implications for scattering from a spherical shell. The surface pressure
in the incident signal may be regarded conceptually as the excitation, so the pressure
in the scattered signal is analogous to the radiated field. Thus, if the frequency is
close to one of these fluid-loaded resonances, it is reasonable to expect that the field
will be quite different from what it would be away from these frequencies. It also is
reasonable to expect that if the shell is composed of a material whose density and
sound speed are much greater than the fluid’s properties, the scattered signal away
from resonances would be like that for a rigid target. Our objective is to explore the
correctness of these expectations.
We will analyze scattering from an elastic spherical shell by merging the analyses
of scattering from a rigid sphere and radiation from a spherical shell. Toward that
end, let us decompose the scattered field in two parts. The first is that which would be
found if the shell was rigid and stationary. It is called the blocked pressure, denoted
as pb = Re(Pb exp (iωt)). The second contribution is induced by the motion of the
shell resulting from its elasticity. Correspondingly, we will denote this part of the
scattered field as pe = Re(Pe exp (iωt)). Thus, we have
Ps = Pb + Pe (12.6.14)
This decomposition is substituted into Eq. (12.1.3), which govern the scattered field.
When each term is associated with its respective effect, we obtain
∂ Pb ∂ PI
∇ 2 Pb + k 2 Pb = 0, =− atr = a (12.6.15a)
∂r ∂r
∂ Pe
∇ 2 Pe + k 2 Pe = 0, = ω 2 ρ0 W at r = a (12.6.15b)
∂r
In the boundary condition for Pe , the parameter W is the complex amplitude of the
radial displacement of the shell, that is, w = Re(W exp (iωt)). The corresponding
radial velocity is vr = Re(iωW exp (iωt)). In addition to satisfying the Helmholtz
equation and a surface boundary condition, both Pb and Pe must satisfy the Som-
merfeld radiation condition, Eq. (12.1.4).
10 J.B. Keller, “Geometric Theory of Diffraction,” J. Opt. Soc. Am.52, (1962) pp. 116–130.
The conditions that the blocked pressure must satisfy are identical to those for
the pressure scattered by a rigid sphere. Our interest is the case of an incident plane
wave, which was analyzed in the previous section. Consequently, we may employ
Eqs. (12.6.3) and (12.6.4) to describe the blocked pressure, so that
∞
jm (ka)
Pb = − P̂I (2m + 1) (−i)m h m (kr ) Pm (cos ψ) (12.6.16)
m=0
h m (ka)
The corresponding farfield description is provided by Eq. (12.6.7).

Now, let us analyze Pe . The radial displacement w represents the response of
the elastic shell to the total acoustic pressure applied at its surface. This is the fully
coupled problem solved in Sect. 7.2.4. That analysis was limited to excitations that are
axisymmetric, which is true for an incident plane wave. It was shown that the spherical
harmonic expansions of the radial displacement w and meridional displacement u
in Eq. (7.2.35) will satisfy the equations of motion if the coefficients are selected
properly. The specific expressions are
∞

w = Re W eiωt , W = Wm Pm (cos ψ)
m=0
∞ (12.6.17)
d
u = Re U eiωt , U = Um Pm (cos ψ)
m=0
dψ
Furthermore, the pressure induced by the shell’s motion constitutes a radiated field,
so it too must be expressible as a spherical harmonic series,
∞

Pe = Cm Pm (cos ψ) h m (kr ) (12.6.18)
m=0
Satisfaction of the velocity boundary condition, Eq. (12.6.15b), requires that
kCm h m (ka) = ω 2 ρ0 Wm (12.6.19)
Thus, Pe will be determined when we find an expression for Wm .

The total pressure acting on the surface of the shell is the sum of the incident,
blocked, and structural contributions. In the basic differential equations of motion
for a spherical shell, the surface pressure is decomposed into two parts: an applied
loading that does not depend on the shell’s motion, and an acoustic part that consists
of the pressure induced by the motion. In the present context, the part that is generated
by the shell’s motion is Pe , and the effective loading is the sum of PI and Pb . The
series for Pe is the same as the description of the acoustic pressure obtained from
Eq. (7.2.41), so we may determine Pe directly from the radiation analysis. To do
so, we equate the applied pressure qapplied to the negative of the sum of the incident
524 12 Scattering
and blocked pressures, where the negative sign applies because qapplied was taken
to be positive if it is outward. Equation (12.6.1) is a spherical harmonic series for a
plane wave propagating in the direction of increasing z. When it is combined with
Eq. (12.6.16) for Pb , the total pressure loading is
qapplied = − (PI + Pb )|r =a

∞

j (ka)
= − P̂I (2m + 1) (−i)m jm (ka) − m h m (ka) Pm (cos ψ)
m=0
h m (ka)
(12.6.20)
When the fraction in the bracketed term is cleared, the numerator is a quantity called
a Wronskian. Application of the definition that h m (ka) = jm (ka) − in m (ka) gives
a form that is described by a simplifying identity,11

jm (ka) h m (ka) − jm (ka) h m (ka) ≡ −i jm (ka) n m (ka)

−i
− jm (ka) n m (ka) = (12.6.21)
(ka)2
The result is that the applied load reduces to

∞
(−i)m+1 (2m + 1)
qapplied = − P̂I Pm (cos ψ) (12.6.22)
m=0
(ka)2 h m (ka)
The radiation analysis derived algebraic expressions for the displacement coef-
ficients Wm and Um in terms of the coefficient Fm of a series representation of the
applied pressure. The preceding equation is such a description, so the force coeffi-
cients are
(−i)m+1 (2m + 1)
Fm = − P̂I (12.6.23)
(ka)2 h m (ka)
The relation between Fm and Wm is described by Eq. (7.2.50) in terms of a

structural stiffness (Z e )m and an acoustic impedance (Z f )m , such that

Wm a Fm
(Z e )m + (Z f )m (ike a) = (12.6.24)
a h ρe ce2
where ρe and ce are the density and a certain wave speed for the shell’s material,
and ke = ω/ce . The structural impedance describes the combination of inertial and
elastic effects,
11 M.I. Abramowitz and I.A. Stegun, ibid., p. 439. Use Eq. (10) to eliminate n n−1 (z) and jn−1 (z)
in Eq. 10.1.31.
2
2
ke a κ(1)
m / (ke a)2 − 1 κ(2)
m / (ke a)2 − 1
(Z e )m = 2 (12.6.25)
i
κ(0)
m / (ke a)2 − 1
( j) ( j)
In this expression κm = m a/ce , where (1) (2)
m and m are the natural frequencies
for axisymmetric-free vibration of the shell in the mth spherical harmonic. Also,
κ(0) (0)
m is a nondimensionalization of the frequency m at which there is no radial
displacement, regardless of the excitation. The determination of these parameters
is described in Sect. 7.2.4. The parameter ke = ω/ce , where ce is the phase speed
of an extensional wave of plane strain. The value (Z e )m is purely imaginary, so it
is a reactance. Inclusion of the effects of internal dissipation would add a resistive
part. However, this resistance would be much less than the reactance, except in the
vicinity of either natural frequency, where the reactive part is zero.
The fluid impedance is
ρ0 c a h m (ka)
(Z f )m = −i (12.6.26)
ρe ce h h m (ka)
The real part is positive, representing resistance associated with radiation damping.
The imaginary part also is positive, and therefore an inertance, corresponding to an
added mass effect. Figure 7.7 indicates that the reactive part has a broad peak that
occurs at a frequency that increases monotonically with increasing m. In contrast,
the resistive part of (Z f )m approaches a constant value with increasing frequency.
Equation (7.2.41) describes the coefficients of a spherical harmonic series for the
pressure. Substitution of the expressions for Wm and Fm , followed by synthesis of
the series, leads to
∞
ρ0 c a (−i)m (2m + 1) h m (kr )
Pe = P̂I
Pm (cos ψ)
ρe ce h m=0 (Z e )m + (Z f )m (ka)2 h m (ka)2
The farfield version is obtained by applying Eq. (12.6.6), which is the asymptotic
property of the Hankel function. The result is
ρ0 c a a −ikr
M
(2m + 1)
(Pe )ff = i P̂I e
Pm (cos ψ)
ρe ce h r m=0
(Z e )m + (Z f )m (ka)3 h m (ka)2
(12.6.27)
where M is the largest index for which the factor of the Legendre function is signif-
icant.
526 12 Scattering
The amplitude of the mth spherical harmonic is maximized when the magnitude
of (Z e )m + (Z f )m is a minimum. Here, an analogy with standard vibration theory
comes into play. The portion (Z e )m represents the effect of inertia and elasticity.
If internal dissipation is ignored, (Z e )m is imaginary. The fluid effect is contained
in (Z f )m . It is complex, with the imaginary part acting as a virtual mass that adds
to (Z e )m . The real part represents radiation damping, in which power is radiated
into the fluid. In a standard spring-mass-dashpot system, a resonance is associated
with the magnitude of the combined impedance being a minimum. However, if the
damping is not too large, the resonance condition will be close to that for which
the imaginary part is zero. Similarly, although the true maximum of a spherical
harmonic corresponds to the frequency that minimizes |(Z e )m + (Z f )m | , it is simpler
to approximate the frequency at which a fluid-loaded resonance occurs as the value
for which Im ((Z e )m + (Z f )m ) = 0.
These resonances affect the selection of the harmonic M at which the series may be
truncated. In the analysis of scattering from a rigid sphere, the criterion for selection
of M was found to be M ka, whose satisfaction assures that h m (ka)2 is very large.
That criterion applies here also, but there is another consideration. It is imperative that
any harmonic m̃ that is close to resonance be included in the summation. That is, we
should set M > m̃. Figure 7.7 describes the impedances
as a function of frequency
for fixed m, but the issue here is the dependence of Im (Z e )m + (Z f )m on m at fixed
ka.
The natural frequencies as functions of m are described in Fig. 7.5, and (0) m falls
between these frequencies. If M is set well above the intersection of the line at the
current frequency ka with the lower branch, then Im (Z e )m will be large negatively.
This accelerates the convergence rate of the series for Pe , beyond what is attributable
to the behavior of the Hankel function. The next example uses M = ceil(10ka) ,
which is much larger than necessary.
Elastic effects occur
in a number of ways in Eq. (12.6.27), so let us evaluate Pe
in the case where Im (Z e )m̃ + (Z f )m̃ actually is zero. Doing so will lead to the
pressure at a fluid-loaded resonance. The frequency for this resonance is denoted as
ω̃ ≡ k̃c. The total impedance reduces to Re (Z f )m̃ at k̃a. Equation (12.6.26) describes
the fluid impedance, so we have
⎡ ⎤

ρ ca h m̃ k̃a
(Z e )m̃ + (Z f )m̃ k̃a = Re (Z f )m̃ |k̃a = Re ⎣−i ⎦
0
(12.6.28)
ρe ce h h k̃a
m̃
The Hankel function of the second kind is defined

in
terms
of the
real
spherical
(2)
Bessel and Neumann functions to be h m̃ k̃a = jm̃ k̃a − in m̃ k̃a . With this,
the impedance becomes
⎡ ∗ ⎤

ρ0 ca h m̃ k̃a h m̃ k̃a
⎢ ⎥
(Z e )m̃ + (Z f )m̃ k̃a = Re ⎣−i ∗ ⎦
ρe ce h
h m̃ k̃a h m̃ k̃a
⎛ ⎞
ρ0 ca −n m̃ k̃a jm̃ k̃a + jm̃ k̃a ym̃ k̃a
⎜ ⎟
= ⎝ 2 ⎠
ρe ce h
h m̃ k̃a
(12.6.29)
The numerator is the Wronskian in Eq. (12.6.21), so the total impedance at a fluid-
loaded resonance reduces to

ρ0 ca 1
(Z e )m + (Z f )m̃ k̃a = 2 2 (12.6.30)
ρe ce h
k̃a h m k̃a
Presumably, the m̃ term in (Pe )ff is dominant. Substitution of the impedance at

resonance into Eq. (12.6.27) yields
e−i k̃r
(2m̃ + 1) e−2i arg(h m̃ (k̃a )) Pm̃ (cos ψ)

(Pe )ff ≈ P̂I (12.6.31)
k̃r
The surprising aspect of this result is that the peak magnitude of (Pe )ff /PI appears
to depend only on the harmonic number and the acoustic wavenumber at which the
resonance occurs. However, this is a deceptive view, because both m̃ and k̃ depend
strongly on all system parameters.
EXAMPLE 12.4 A spherical shell (h/a = 40) composed of aluminum (ρe =

2100 kg/m3 , ce = 6410 m/s) is submerged in water. Determine the backscatter
cross section σback and total scattering cross section σ in the range ka < 10.
Compare this to the frequency dependence of each parameter if the sphere was
rigid and stationary. Then, repeat the analysis for the case where the sphere is
suspended in the air.
Significance
The dependence on a number of system parameters, as well as the interplay between
spherical harmonic number and frequency, can best be understood by performing
quantitative analyses like the one to be performed here. The evaluation itself will
provide a useful review of some fundamental concepts.
Solution
The total scattered field is the sum of the blocked and structural contributions.
Because Pb is the same as the scattered field for the rigid case, we shall compute it, and
528 12 Scattering
Pe as separate entities. Our strategy is rather straightforward. The in-vacuo natural

frequencies (1) (2)
m and m , as well as the zero-radial-displacement frequency m ,
(0)
are independent of the excitation frequency. The truncation criterion is M = 10ka,

and the maximum frequency to be considered is ka = 10. Therefore, we begin by
computing the modal properties for azimuthal harmonics from zero to 100. This com-
putation entails solving the characteristic equation, Eq. (7.2.44). The K coefficients
to use are those in Eq. (7.2.48), which describe membrane and flexural deformation
( j) ( j)
effects. The roots of this quadratic equation are the values κm = m a/ce , j = 1, 2,
(0)
and κm in Eq. (7.2.46).
The remaining operations are performed for a succession of ka value. The values of
h m (ka) for 0 ≤ m ≤ M + 1 are computed and used in conjunction with a recurrence
relation to evaluate h m (ka) for 0 ≤ m ≤ M. These values are required to evaluate
both Pb and Pe , and the latter also requires jm (ka) , which is the real part of h m (ka) .
Let us denote the coefficients of the spherical harmonic series for Pb /PI in the farfield
as Cb,m and those for Pe /PI as Ce,m . Thus, from Eqs. (12.6.7) and (12.6.27), we have
Pb a M
Pe a M
= e−ikr Cb,m Pm (cos ψ) , = e−ikr Ce,m Pm (cos ψ)
PI r m=0
PI r m=0
jm (ka) (1)

Cb,m = −i (2m + 1)
kah m (ka)
ρ0 ca (2m + 1)
Ce,m = i

ρe ce h (Z e )m + (Z f )m (ka)3 h m (ka)2
All coefficients for a specific ka may be computed in a single vectorized step.

The backscatter cross section was defined in Eq. (12.4.10). The backscatter direc-
tion is ψ = π, for which Pm (−1) = (−1)m . Thus,
M 2

(σback )e = 4πa 2
(−1) Cb,m + Ce,m
m

m=0
M 2 (2)

(σback )b = 4πa 2 (−1)m Cb,m

m=0
Evaluation of the scattering cross section would be more difficult were it not for the
orthogonality property of Legendre functions. Because of it, there is no coupling of
harmonics in the scattered power. The steps for this analysis are
π
∗
Ps Pb Pw Pb Pw∗
σs = = + ∗ + ∗ 2πr 2 sin ψdψ
II 0 PI PI PI PI

M π
M
∗ ∗

= 2πa 2 Cb, j + Ce, j Cb,m + Ce,m P j (cos ψ) Pm (cos ψ) sin ψdψ
j=0 m=0 0

M
4
= πa 2 Cb,m 2 + C ∗ 2 + 2 Re Cb,m C ∗
2m + 1 e,m e,m (3)
m=0
The scattering cross section for the rigid sphere is obtained by setting Ce,m = 0.
Figure 1 shows the backscatter and total scattering cross sections. The spikes cor-
respond to fluid-loaded resonances, whose occurrence was anticipated in the earlier
discussion. The puzzling aspect is their occurrence only in the low-frequency band.
102
σback
100
Pe + Pb
10-2 Pb
102
100
σs
10-2
0 2 4 6 8 10
Frequency ka
Figure 1
An examination of this behavior begins by considering the resonance condition

in the absence of fluid loading. In that case, (Z f )m̃ is identically zero, so a resonance
corresponds to (Z e )m , which is imaginary, being zero. At a randomly selected ka,
a plot of Im (Z e )m as a function of m would show that it crosses the zero axis at
a value of m that is not an integer, so it is not a resonant frequency. Increasing ka
eventually will bring on a condition where Im (Z e )m = 0 at an integer value of m.
Because we are ignoring the role of fluid loading, the corresponding value of (Ce )m̃
would be infinite. Further increase of ka increases the m value at which (Z e )m = 0,
until it reaches another integer value of m, and so on.
A crucial aspect of the effect of fluid loading is that (Z f )m is zero only at ka = 0.
two quantities that affect the magnitude of (Ce )m are | (Z e )m + (Z f )m | and
The
h (ka) , both of which are in the denominator of the above expression. A fluid-
m
loaded resonance was defined as a frequency k̃a at which Im((Z e )m̃ + (Z f )m̃ ) = 0.
However, Re (Z f )m̃ is not zero, so it is possible that | (Z e )m + (Z f )m | is not close
to a minimum. Furthermore, even if it is, the denominator of (Ce )m̃ might not be
530 12 Scattering

minimized because the magnitude of its denominator is | (Z e )m + (Z f )m | h m (ka) ;
a small impedance might be multiplied by a large Hankel function.
To see how these factors work together, Fig. 2 examines them at ka = 1.188, at
which the first peak in σback and σs occur. This value of ka is very close to the first
natural frequency of the m = 2 spherical harmonic, ka ≈ κ(1) 2 . The uppermost graph
indicates that the reactance at m = 2 is zero, and the resistance is small. The actual
values are (Z e )2 + (Z f )2 = 0.13341 − 0.02512i. The second graph shows the total
impedance is minimized at m = 2. The general trend for h m (ka) is that it rises rapidly
as m increases beyond ka. However, the third graph indicates that this increase is
not drastic for m = 2, for which h 2 (1.188) = 0.12829 + 4.8275i. The logarithmic
scale in the fourth graph obscures relative magnitudes, but it is evident that (Ce )2 is
the largest coefficient.
ka = 1.188
4
0
−4 Im(Z e + Z f ) Re(Z e + Z f )
−8
10
5 | Z e + Z f|
0
106
103 |h'm (k a)|
1
1
|(C e ) m |
10-3
10-6
0 1 2 3 4 5 6 7 8 9 10
Figure 2
As ka is increased, the plots of (Z e )m + (Z f )m and h m (ka) as functions of m

maintain their general shape, except that Im[(Z e )m + (Z f )m ] crosses the zero axis
at an increasing m. Figure 3 describes the behavior at ka = 4, which does not
correspond to a peak in either scattering cross section. The plot of Im[(Z e )m +
(Z f )m ] crosses the zero axis between m = 12 and m = 13, and the minimum value

of | (Z e)m + (Z f )m | occurs m = 11, where it is 0.03865. However, h 12 (4) =
4at
−6
4.4811 10 + 2.0608 10 i, which is quite large, and this quantity is squared in
the denominator of (Ce )m . The consequence is that the largest value of (Ce )m occurs
at m = 3, but it is not especially large relative to the values for adjacent m. Hence,
this frequency does not lead to enhanced scattering.
ka = 4
10
0
−10 Im(Z e + Z f ) Re(Z e + Z f )
50
25 | Z e + Z f|
0
108
104 |h'm (k a)|
1
1
10-3 |(C e ) m |
10-6
0 5 10 15 20 25 30
Figure 3
These observations suggest that the increase of |h m (ka)| for m > ka is respon-
sible for the suppression of high-frequency resonances. An example of this
−10 behav-
ior is
−5 ka = 8.710 at m̃ = 19, for which (Z e ) m̃ + (Z f ) m̃ = 8.8890 10 −
6.7896

10 i. For comparison, at m = 18 the value is (Z e ) 18 + (Zf ) 18 = 1.7893 10−8 +
0.85717i. However, at m = 19, h 19 (8.71) = 3.2851 104 , so the relative small-
ness of the impedance term is overwhelmed by the largeness of h 19 (ka). The largest
coefficient
value at ka = 8.710 is (Ce )m = 1.4761 at m = 7, whereas (Ce )19 =
6.5216 10−5 . Thus, the absence of high-frequency resonances in the scattering
cross sections may be attributed to the fact that the spherical harmonic excitation
becomes much weaker with increasing m, as well as the low radiation efficiency
of the higher spherical harmonics of the surface displacement. (A spherical har-
monic whose number is large corresponds to a short wavelength/subsonic surface
motion.)
Although resonances are not present in scattering cross sections at high frequen-
cies, the shell’s flexibility does influence the scattering properties at all frequencies.
As shown in Fig. 1, the value of the total scattering cross section is considerably
increased by the flexibility of the shell, and the backscatter cross section shows
considerable variability.
There is no need to display the cross sections for the case where the surrounding
fluid is air, because they are identical to those in Figs. 12.7 and 12.8 for a rigid,
stationary sphere. The flexibility of the shell has little effect on the scattering because
there is an enormous difference in the properties of the shell relative to air and water.
The parameter
that governs the overall magnitude of Pe relative to Pb is ρ0 c/ (ρe ce ) ,
which is 3 10−5 for air and aluminum, and 0.11 for water and aluminum. Thus, Pe
is very a small contributor to the scattered field. Even resonances are insignificant
relative to the blocked pressure.
532 12 Scattering
Exercise 12.1 A solid plastic sphere immersed in an ideal liquid has density and
sound speed that are slightly larger than the liquid. A plane wave is incident on this

object. Specialize the Born approximation of (r/a) Ps (x̄0 ) / P̂I to fit this spherical
configuration. Doing so will yield an integral representation of the scattered pressure
at a field point in the farfield as a function of ka, ρ/ρ0 , and c/c0 , and the direction
angles to the field point. Evaluate this integral for the case of backscatter correspond-
ing to ρ /ρ0 = 1.02 and c /c0 = 1.05. Plot the backscatter amplitude in the interval
ka < 2, and compare it to the Rayleigh limit in Eq. (12.3.1).
Exercise 12.2 A submerged object is surrounded by eight equally spaced

hydrophones, as shown in the sketch. Projector A emits an omnidirectional signal at
20 Hz. This signal is measured to be 210 dB//1 μPa at one meter from the source. The
table provides the amplitude of the scattered signal measured at each hydrophone.
The ambient properties for the water are ρ0 = 1004 kg/m3 and c = 1470 m/s. A
visual inspection of the object indicates that its volume is 0.12 m3 . (a) Assuming that
the object’s physical properties are comparable to water, use the tabulated data to
deduce the density and sound speed within the object. Use the Rayleigh limit of the
Born approximation to analyze the data. (b) To explore the effect of measurement
error, contaminate the tabulated data by adding to each value a random error that
ranges over ±0.4 Pa with a uniform probability distribution. Repeat the calculation
in Part (a) using the error-contaminated data. What does the result say about the
significance of measurement error?
Hydrophone # 1 2 3 4 5 6 7 8
|Ps | (mPa) 1.5296 1.3526 1.1022 0.92514 0.92514 1.1022 1.3526 1.5296
2 3
45o
1 45o 45o 4
A 22.5o
8 5m 5
7 6
30 m
Exercise 12.2
Exercise 12.3 A transducer capable of transmitting and receiving is situated at point

A, and a receiving hydrophone is situated 400 m away at point B. It is desired to use
this arrangement to measure the physical properties of a bottle-nosed dolphin, based
on the notion that it does not differ much from water. The scheme uses transducer A
to generate a tone burst consisting of three cycles of a 50 Hz harmonic. The scattered
signal is measured by both transducers. The received signals are plotted below. (The
pulse received at position B directly from the source at point A has been removed.)
The timescale is based on t = 0 being the instant when the tone burst is initiated at
source A. The measured signals allow for determination of the geometrical properties
required to evaluate the Rayleigh limit of the Born approximation. It is known that
the transducer A generates a radially symmetric signal that is 49 kPa at 1 m from
the center, and a measurement of the displacement of the dolphin indicates that its
volume is 0.987 m3 . The density and sound speed of water are the nominal values,
c = 1480 m/s and ρ0 = 1000 kg/m3 . Based on the assumption that the dolphin is
approximately homogeneous, determine its average density and sound speed.
rA rB
A B
400 m
120
Pressure (μPa)
80 PA
40 PB
0
−40
−80
−120
320 340 360 380 400 420 440 460 480 500
Time (ms)
Exercise 12.3
Exercise 12.4 Consider the cylinder in Example 12.1 when the incident wave arrives
broadside, ψI = 90◦ . Determine the directivity of the scattered pressure in the plane
formed by the incident wave’s propagation direction and the axis of the cylinder.
Frequencies of interest are ka = 0.1, 1, and 10.
Exercise 12.5 A plane wave is incident at direction ēI on the stationary rigid rectan-
gular box in the sketch. The box, whose walls have negligible stiffness and mass, is
filled with a liquid for which ρ = 0.92 ρ0 and c = 0.94. The box sides are a, b, and
d. (a) Use the Born approximation to derive an expression for the low-frequency scat-

tered pressure in the form (r/a) Ps (x̄0 ) / P̂I for arbitrary ēI and scattering direction
ēr . For the development, describe ēI and ēr in terms of spherical angles described
relative to the z-axis. (b) Specialize the result in Part (a) to the backscatter case,
ēr = −ēI when a plane wave is incident in the x z-plane. (c) Let = V 1/3 , where V is
the volume, be a representative length scale. Compare the backscatter directivity for
two shapes having the same volume: (1) a = b = d = , (2) a = b = 0.5, d = 4.
534 12 Scattering
b d
a
C
z
y
Exercise 12.5
Exercise 12.6 The cylinder in the sketch is rigid and stationary. It is insonified by a
plane wave at frequency ω whose propagation direction is parallel to the centerline of
the cylinder. The radius a is extremely small compared to the length L, and ka
1.
Derive an expression for the scattered pressure at an arbitrary field point x̄ in the
farfield.
x
r
ψ
a z PI
L
Exercise 12.6
Exercise 12.7 The sketch shows a mass-spring oscillator consisting of a sphere

whose mass is M that is supported by a spring whose stiffness is K . A harmonic
plane wave, which is traveling parallel to the direction in which the spring is oriented,
causes the sphere to oscillate. The sphere is rigid, so the only motion that it executes
is a rigid body translation u = Re (U exp (iωt)). It may be assumed that ka
1.
(a) Derive an expression for U as a function of ka. (b) Derive an expression for the
scattered pressure Ps at an arbitrary field point in the farfield. (c) Determine the value
of ka that will maximize |Ps |.
M eI
Exercise 12.7.
Exercise 12.8 A 20 mm diameter ball bearing is insonified by a 1 kHz plane wave.

Evaluate its scattering cross section in situations where it is restricted from moving,
and where it is suspended in a manner that permits it to translate. Compare the

results for a ball compose of lead (ρ = 11340 kg/m3 , c = 1322 m/s), to those for an
aluminum ball (ρ = 2100 kg/m3 , c = 6410 m/s).
Exercise 12.9 A plane wave at 60 Hz in air is obliquely incident on a stationary

rigid disk whose radius is 200 mm. Its direction of propagation is 20◦ off the normal
to the disk. Determine the scattering cross section, the backscatter cross section, and
the target strength corresponding to this angle of incidence.
Exercise 12.10 A land mine is buried in sand, ρ0 = 1140 kg/m3 , c = 105 m/s,
which may be regarded as a liquid. It is insonified by a 15 Hz plane wave whose
amplitude is 500 Pa. The mine is buried sufficiently deeply that reflections of the
scattered wave from the surface may be ignored. It is a 200-mm-long cylinder whose
radius is 150 mm, and both ends are closed with hemispherical caps. The mass of
the cylinder is 120 kg and the virtual mass coefficients are 1,1 = 2,2 = 10V,
3,3 = 2.5V. Determine its scattering cross section and target strength in the situation
where ēI = −ēx . Do these values depend strongly on whether the mine is stationary
or movable?
Exercise 12.11 A plane harmonic wave in the atmosphere is incident on an object

that is symmetric about the z-axis, with ēI = −ēz . The frequency is 500 Hz. The
scattered pressure at a point in the farfield is described in terms of spherical harmonics
as
∞
e−ikr0
Ps = PI Bm Pm (cos ψ0 ) ei(m+1)π/2
kr0 m=0
The coefficient values are B0 = 0.5, B1 = 0.2 − 0.4i, B2 = 1.5 + 0.3i, B3 = −

0.2 + 0.2i Pa, while all higher values are zero. The forward scattered pressure is
2 Pa at r = 10 m. (a) What is the backscattered pressure at r = 10 m. (b) What are
the differential scattering cross section values at ψ = 0, 90◦ , and 180◦ ? (c) What is
the scattering cross section of this object? (d) What is the target strength?
Exercise 12.12 The axisymmetric object in the sketch is a body of revolution. It

is generated by rotating about the centerline a circular arc of radius 4a whose cen-
ter is situated at 5a from the centerline. A plane harmonic wave traveling in the
direction −ēx is incident on the object. According to geometrical acoustics, what is
the backscatter cross section of this object? Would this value change if the incident
wave’s direction deviated slightly from broadside incidence?
536 12 Scattering
4a
a
z
x
Exercise 12.12
Exercise 12.13 A crude model for scattering from a rough surface considers normal
incidence of a plane wave on a surface that is nominally flat, but actually has a
sinusoidal variation in its elevation. Such a situation is depicted in the sketch, where
the elevation is h = h 0 cos (2πx/L). If the acoustic wavelength 2π/k is much smaller
than the spatial period L of the surface, it is permissible to invoke geometrical
acoustics. Construct a sketch of the manner in which the incident rays scatter. What
general conclusions can be drawn from this construction?
Exercise 12.13
Exercise 12.14 A plane harmonic wave propagating horizontally in water is incident

on a solid sphere. The sphere is suspended by a cable that permits it to move in the
horizontal plane. Derive expressions for the backscattered pressure and the velocity
of the center of the sphere as a function of ka < 10 for the case where the sphere’s
mass is 5πa 3 ρ0 .
Exercise 12.15 Exercise 12.7 entailed an analysis of scattering from a spring-

supported rigid sphere in the low-frequency regime, ka
1. (a) Perform the investi-
gation requested there, but do not assume that ka is small. (b) Evaluate ρ0 c |VC | / |PI |
as a function of ka < 5 for four parametric combinations: M = σρ0 πa 3 and K =
χM (c/a)2 with σ = 2 and 8, χ = 1 and 2. Identify the frequency at which |VC | is
maximized.
Exercise 12.16 A 3-m diameter steel shell is neutrally buoyant in a very large vat of
ethyl alcohol (ρ0 = 785 kg/m3 , c = 1144 m/s). It is insonified by a plane harmonic
wave. Determine the scattering directivity, that is, (r/a) |Ps /PI | as a function of ψ.
Identify the portion of this function that is attributable to the elasticity of the shell.
Carry out this evaluation for cases where the frequency of this wave is 99% of the
upper natural frequency of the m = 0, 1, and 2 in-vacuo modes.
Exercise 12.17 The sketch shows a plane wave that is incident on a rigid hemisphere
mounted on an infinite rigid baffle. The sum of the incident and scattered fields
must have zero particle velocity normal to the baffle, as well as the normal to the
hemisphere’s surface, that is, (v̄I + v̄s ) · ēz = 0 for r > a, ψ = π/2 and (v̄I + v̄s ) ·
ēr = 0 for r = a, 0 ≤ ψ < π/2. A suitable representation draws on the development
in Sect. 7.2.3, which showed that the field radiated by a vibrating hemisphere on a rigid
baffle can be represented as a series of the even-numbered spherical harmonics. At the
same time, if the hemisphere was not present, the incident wave would be reflected
with the reflection coefficient being one. Thus, an ansatz for the field scattered by
the hemisphere and baffle is
P = PI + PR + Ph
∞

PI = P̂I e+ikz , PR = P̂I e−ikz , Ph = Bm h 2m (kr ) P2m (cos ψ)
m=0
Derive an expression that describes the Bm coefficients.
Exercise 12.17
Exercise 12.18 Consider the hemisphere-baffle configuration in Exercise 12.17.

The scattered pressure is Ps = PR + Ph . The frequency of interest is ka = 10. (a)
Evaluate the scattered pressure as a function of distance along the z -axis. Compare
the result to the geometrical acoustics approximation for the backscatter. (b) Evaluate
the dependence of the scattered pressure along a line parallel to the baffle at z = 4a.
(c) Evaluate the dependence of the scattered pressure on the polar angle ψ at r = 4a.
Chapter 13
Nonlinear Acoustic Waves
When told that a system’s response is quite large, a knowledgeable individual

will anticipate that nonlinear effects are important. This certainly is the case in
acoustics, as exemplified by an explosion, wherein peak pressures might be well
above atmospheric. However, even if the pressure is not large, there is another para-
meter whose largeness might require consideration of nonlinear effects. That para-
meter is the propagation distance. The signal radiated by a transducer might not
seem to be especially large, but nonlinear effects grow with increasing distance. We
might observe harmonic and intermodulation distortion of the waveform, decreased
signal strength, or the formation of shocks (discontinuous changes of the waveform).
Although these effects would seem to be deleterious, there are some applications that
rely on them.
Analysis of nonlinear effects encounters two fundamental difficulties relative to
linear theory. The most obvious is the inapplicability of the principle of superposition.
But there is another. In our studies thus far, conversion to the frequency domain made
it possible to analyze many systems. The concept of using a complex exponential
to remove the explicit dependence on time ceases to be valid in the presence of
nonlinearity. The reason becomes evident when we consider a nonlinear term such
as p2 . If p oscillates at frequency ω, then p2 contains a term whose frequency is 2ω.
The occurrence of such a term in the field equation for p means that p cannot be a
pure harmonic. The consequence is that we will need to work in the time domain, so
far fewer systems are amenable to analysis.
In many treatises, the term finite amplitude wave is used to refer to a signal whose
amplitude is sufficiently large that nonlinear effects cannot be ignored. The notion
behind this terminology is that ignoring terms like p2 is equivalent to considering
p to be infinitesimal. Whether it is allowable to do so depends on several factors,
most notable of which are the amplitude of the signal and the propagation distance.
If either is large, one might need to consider nonlinear effects.
We will not pursue a historical survey of the numerous approaches that have
been used. Rather, the methods we shall explore have been selected either for their
DOI 10.1007/978-3-319-56847-8_13
540 13 Nonlinear Acoustic Waves
generality or for their directness. We will follow Riemann’s seminal analysis,1 which
is based on the fundamental properties of differential equations. One of its beneficial
attributes is that it does not require specification of the excitation. Another approach
we will develop applies Fourier series to analyze systems in which the excitation is
temporally periodic. The idea is not to convert the problem to the frequency domain.
Rather, the formulation yields differential equations for the Fourier coefficients that
are easier to solve. A third approach for our consideration is perturbation analysis,
which decomposes the nonlinear problem to a sequence of linearized problems. This
method cannot proceed if the analogous linear problem cannot be solved.
13.1 Riemann’s Solution for Plane Waves
It is remarkable that Riemann’s solution was published in 1859, and that it was
contemporaneous with Earnshaw’s analysis2 of the same problem. The two papers
followed very different paths. Riemann used the continuity and momentum princi-
ples to obtain equations governing the fundamental state variables: pressure, particle
velocity, and density. Earnshaw formulated the analysis in terms of displacement
of a specific particle in what is called a Lagrangian kinematical description. Most
importantly for us, Riemann’s analysis proceeds in a logical manner, whereas Earn-
shaw’s requires a degree of intuition. The results of the analyses are equivalent. Both
may be found in treatises by Rayleigh3 and Lamb.4 Poisson5 actually was the first to
solve this problem, so some refer to the final result as the Poisson solution. However,
that treatment assumed isothermal conditions. We will follow Riemann’s analysis
because of its directness and generality.
13.1.1 Analysis
The fluctuating acoustic part of the pressure is p, so the absolute pressure in the
current state is p0 + p. The corresponding density is ρ. The direction of propagation
is designated as x. The particle velocity, positive in the sense of increasing x, is v.
The basic equations of continuity and momentum for a one-dimensional wave are
1 B. Riemann, “Über die Fortpflanzung ebener Luftwellen von endlicher Schwingungsweite,”

Abhandlungen der Königlichen Gesellschaft der Wissenschaften zu Göttingen, Vol. 8 (1859) pp.
245–264.
2 S. Earnshaw. “On the mathematical theory of sound,” Trans. Royal Soc. London (1860) pp. 133–
148.
3 J.W. Strutt, Lord Rayleigh, “Theory of Sound,” Vol 2, (1877), Dover reprint (1945) Articles 252
and 253.
4 Lamb, Hydrodynamics, (1916), Dover reprint, (1945). Articles 260 and 261.
5 S.D. Poisson, “Memoire sr lat théorie de son,” J. l’École Polytechnique Paris, Vol 7 (1808) pp.
319–392.
13.1 Riemann’s Solution for Plane Waves 541
Eqs. (2.1.6) and (2.1.10). They are
∂ρ ∂
+ (ρv) = 0
∂t ∂x (13.1.1)
∂v ∂v ∂p
ρ +v =−
∂t ∂x ∂x
In addition, we retain the assumption of adiabatic compression and expansion, based

on the expectation that the pressure varies at a rate that is too high to permit significant
heat transfer.
The notion motivating the Riemann solution is that there is a family of curves
in space-time along which p and v are constant. If this is true, then as long as we
stay on one of these curves, we can consider p to be a function of t and x, and v to
be a function of p. A curve t as a function of x (or x as a function of t) having this
property is a characteristic. Thus, along a characteristic, the functional dependence
is v = v (p). As a consequence of the adiabatic assumption, the equation of state
may be taken to be ρ = ρ (p).
The task is to solve the first-order conservation equations for the relation between
x and t consistent with the assumed functional dependence on p. The derivatives of
ρ and v appearing in the basic equations are converted by the chain rule to
∂ρ 1 ∂p ∂ d ∂p
= , (ρv) = (ρv)
∂t dp/dρ ∂t ∂x dp ∂x
(13.1.2)
∂v dv ∂p ∂v dv ∂p
= , =
∂t dp ∂t ∂x dp ∂x
In linear theory, the square of the sound speed is defined to be dp/dρ in the ambient
state. In the nonlinear formulation, the relevant value is this derivative at the current
state. Many references define c2 to be this value and denote c0 as the linear speed of
sound. It is less confusing if c is retained as the linear speed of sound, so c̃2 is dp/dρ
at the current state,
1/2 1/2
dp 1
c̃ = ≡ (13.1.3)
dρ dρ/dp
When Eq. (13.1.2) is substituted into the basic continuity and momentum equa-
tions, the result is

1 ∂p 1 dv ∂p
+ 2v +ρ =0
c̃2 ∂t c̃ dp ∂x
(13.1.4)
dv ∂p dv ∂p
ρ + ρv +1 =0
dp ∂t dp ∂x
Because of the assumed interdependence of variables, the only quantities that depend
explicitly on x and t are the derivatives of p. Thus, the preceding may be considered
to be a pair of first-order equations for ∂p/∂t and ∂p/∂x in which the coefficients
are functions of p. The matrix form of the equations is
⎡ ⎤
1 1 dv ⎧ ⎫ ⎧ ⎫
⎢ c̃2 2
v + ρ ⎥ ⎨ ∂p/∂t ⎬ ⎨0⎬
⎢ c̃ dp ⎥ = (13.1.5)
⎣ dv dv ⎦⎩ ⎭ ⎩ ⎭
ρ ρv +1 ∂p/∂x 0
dp dp
These equations are homogeneous. If the relation between p and v were arbitrary,
their solution would be that both ∂p/∂t and ∂p/∂x are zero. These conditions lead
to the trivial solution that p is constant. A nontrivial solution is obtained only if
the pair of equations cannot be solved. In other words, a nontrivial solution along a
characteristic curve exists only if the determinant of the coefficient matrix vanishes.
This condition reduces to
1 dv 2
− ρ =0 (13.1.6)
c̃2 dp
Thus, we find that

dv 1
=± (13.1.7)
dp ρc̃
The equation of state relates ρ and p, and c̃ is obtained by differentiating that rela-
tion. Hence, the preceding constitutes a first-order ordinary differential equation that
relates p and v on a characteristic. It may be integrated, subject to the condition
that the fluid is quiescent in the ambient state, v = 0 if p = 0. Thus, the integrated
relation is
p
dp
v= (13.1.8)
0 ρ (p ) c̃ (p )
We shall proceed on the assumption that we have evaluated this integral.

It is useful at this juncture to recall the relation between the particle velocity and
the pressure fluctuation in a linear plane wave, v̄ = ēp/ (ρ0 c). The propagation here
may be in either the positive or negative x direction, so ē = ±ēx , and v̄ = vēx . Thus,
if we were to differentiate the relation for a linear wave, we would obtain dv/dp =
±1/ (ρ0 c). The present nonlinear analysis tells us that the effective characteristic
impedance is the value at the current state. An analogous relationship occurs when
the force in a spring depends nonlinearly on the strain, which is equivalent to the strain
being a function of the force. Then, the incremental force dF required to increase the
strain by dε depends on the value of the current force. In other words, dF = K (F) dε,
where K = 1(dε/dF). For a linear spring, K is the (constant) stiffness, so K (F) is
said to be the tangential stiffness. If we were to use similar terminology for a nonlinear
plane wave, we would call ρc̃ the tangential characteristic impedance of the fluid.
A corollary of the coefficient matrix in Eq. (13.1.5) having a zero determinant is
that the two differential equations are equivalent. When Eq. (13.1.7) applies, both
require that
∂p ∂p
+ (v ± c̃) =0 (13.1.9)
∂t ∂x
Note that the correspondence of alternative signs must be maintained, so a negative

sign corresponds to a wave that propagates in the direction of decreasing x.
The preceding tells us that anywhere along a characteristic, ∂p/∂t is proportional
to ∂p/∂x, where the proportionality factor is − (v ± c̃), whose constancy follows
from the fact that both v and c̃ depend on the p value for that characteristic. Constancy
of p along a characteristic may be expressed alternatively by equating p at any two
(x, t) pairs on the same characteristic. The points we select are differentially apart,
so we require that p (x + dx, t + dt) = p (x, t). Application of a Taylor series leads
to
∂p ∂p
dx + dt =0 (13.1.10)
∂x ∂t constant p(x,t)
When we substitute Eq. (13.1.7), the result is

dx
− (v ± c̃) =0 (13.1.11)
dt constant p(x,t)
Because v and c̃ are constant in this condition, it follows that dx/dt is constant.
In other words, the slope dx/dt defining where a specific p occurs in space-time is
constant at ±c̃ + v. Stated differently, the characteristics are straight lines! If p̃ is
known at position x0 at time t0 , then the same pressure will occur at position x at
time t, with
x − x0 = (v ± c̃) (t − t0 ) (13.1.12)
This simple property has some of profound implications, which we shall now
examine.
13.1.2 Interpretation
From a viewpoint of phenomenology, the most important aspect is that the phase
velocity of a plane wave that propagates in the direction of increasing x is (c̃ + v)ēx .
The difference between c̃ and the linear speed of sound, c, is a consequence of the
variability of the equation of state. Consider an ambient state in which the fluid is at
rest and the pressure is the sum of the actual ambient pressure p0 plus the acoustic
perturbation p. Then, c̃ would be the linear sound speed c. The phase speed also
differs from c because the particle velocity is nonzero. Suppose that in this ambient
state, the fluid flows uniformly at vēx . In this condition, the fluid is at rest relative to a
reference frame that translates in unison with the ambient flow. That reference frame
is inertial, so sound propagates relative to the moving reference frame at speed c̃ēx .
The propagation velocity as seen by a fixed observer is the velocity relative to the
moving reference frame plus the velocity of the reference frame, that is, c̃ēx + vēx .
In the terminology of nonlinear mechanics, the difference between c̃ and c is said to
be a consequence of material nonlinearity. The fact that v adds to the speed of sound
is said to be a convective nonlinearity. Although the two sources of nonlinearity are
unrelated in regard to their cause, they affect the propagation speed similarly.
The (linear) d’Alembert solution for plane waves actually is quite similar to the
present result. It states that constant values of p propagate along lines for which either
t − x/c or t + x/c is constant. These lines in space-time are the characteristics of
the linear wave equation. Figure 2.3 describes the characteristics as graphs of t as a
function of x. Similar graphs are extremely useful for nonlinear waves. Let us begin
with a boundary value problem, in which v at x = 0 is a specified function of t.
The propagation is taken to be in the sense of increasing x, so the positive sign
preceding c̃ applies in Eq. (13.1.11) and everywhere else. The abscissa is x, so the
slope of a characteristic line is dt/dx = 1/ (c̃ + v). This slope is less than 1/c if
p > 0. To see why, note that Eq. (13.1.7) in this case gives dv/dp > 0, which means
that v > 0 if p > 0. Furthermore, the equation of state for ideal gases and liquids is
such that c̃ > c if p > 0. Because the slope is less than 1/c, a characteristic along
which v > 0 describes propagation that is faster than the linear speed of sound. The
converse is true for characteristics associated with v < 0. They describe propagation
at a speed that is less than c. (It is best to avoid using terms like subsonic or supersonic
here.)
To visualize these properties, let us suppose we know the particle velocity at x = 0
as a function of time. (This is a simplification of the actual boundary condition for
a source, such as a vibrating wall. The proper description of this condition may be
found in the next section.) The case of a single period of a sine pulse starting at t = 0
is plotted along the t-axis in Fig. 13.1.
Fig. 13.1 Characteristics t

lines for propagation of the
nonlinear wave generated by v=0
a sinusoid velocity pulse at
x=0 v<0
v(t) @ x= 0
v=0
v>0
v=0
x
The characteristics through the instants when v = 0 at x = 0 represent propagation

at the linear speed of sound because v and p are zero along them. Because v at the
boundary is zero for t < 0 and t > 2π/ω, the pressure is zero in the region below the
lowest characteristic and above the uppermost characteristic. Signals for which v > 0
cover the same distance in a shorter elapsed time, whereas signals for which v < 0
take longer to travel a given distance. It follows that the shape of a waveform will
change as the wave propagates. A linear wave, which is described by the d’Alembert
solution, is said to be dispersionless, because all waveforms maintain their shape. In
a waveguide, nonplanar modes have a phase speed that depends on the frequency,
which means that the waveforms undergo frequency dispersion. The changing shape
of the waveform for a nonlinear plane wave is said to be amplitude dispersion.
Because the characteristics for different v values are not parallel, it is evident that
these lines will eventually intersect. At such intersections, the analysis indicates that
two signals having different values of v will exist at the same x and t. This is an
impossible condition. What actually happens is that discontinuities, which are called
shocks, form. We will explore this phenomenon later. For now, we will consider the
intersection of characteristics as a limit beyond which the Riemann solution is not
valid.
A consequence of p being constant on a characteristic is that lines that originate at
instants when p is maximum will retain this property. Similarly, lines that originate
from minimum or zero p remain so. We could use this attribute to construct a sketch
of a waveform at fixed x or a profile at fixed t. This is one approach described in the
next section. However, before we address the quantitative evaluation of a signal, let
us consolidate what we have learned.
The pressure and particle velocity are constant along a characteristic, and a char-
acteristic is the locus of points in space-time at which t − x/ (c̃ + v) is a specified
constant. This constant value is a phase variable for the wave. We shall use τ to
denote it, so that
x
p = F (τ ) , τ = t − (13.1.13)
(c̃ + v)
The logical question at this juncture is what is F (τ )? The answer lies in matching
the preceding general form to the manner in which the signal is generated.
13.1.3 Boundary and Initial Conditions
Consider the situation where the plane wave is generated by movement of a large
piston or a wall that translates as a rigid body. Let x = X (t) denote the (known)
position of the wall at any instant. Then, continuity of particle velocity requires that
the velocity of the fluid at the current location of the piston face equals the piston’s
velocity, that is,
v = Ẋ (t) at x = X (t) (13.1.14)
This is referred to as a moving boundary condition. Because of the intricate manner

in which the state variables are related, we shall take a somewhat reverse approach,
in which relations that we seek are taken to be known.
Because the boundary condition imposes a particle velocity, the first step is to
remove pressure from the functional dependencies. Toward that end, we invert the
expression of v in terms of p obtained from Eq. (13.1.8). This step allows us to con-
sider the relationship to be p = p (v). The implication of this switch of dependencies
is that c̃ is known if we know v. This feature is made explicit by letting C̃ be the
value of c̃ at a specified v, so that
C̃ (v) = c̃ (p (v)) (13.1.15)
It follows that the slope of a characteristic is dx/dt = C̃ (v) + v.

Given this, let us track a signal that departs from the boundary at time τ . The
particle velocity
everywhere along the characteristic for this signal is Ẋ (τ ), so its
slope is C̃ Ẋ (τ ) + Ẋ (τ ). Also, the starting point of the characteristic is (X (τ ) , τ ).
The time required for this signal to arrive at an arbitrary point (x, t) is t − τ , and
the propagation distance to this point is x − X (τ ). Equating [x − X (τ )]/ (t − τ ) to
dx/dt yields
x − X (τ )
= C̃ Ẋ (τ ) + Ẋ (τ ) (13.1.16)
t−τ
A minor rearrangement of terms yields a solution whose form is like the one derived
by Earnshaw, specifically
x − X (τ )
τ =t− , v = Ẋ (τ ) (13.1.17)
C̃ Ẋ (τ ) + Ẋ (τ )
These relations fully describe the signal. To realize this, suppose we select a value
of τ . Given a function
X (t), we may evaluate X and Ẋ at t = τ , which sets the
value of C̃ Ẋ (τ ) . Then, the first of the above relations gives the locus of x, t
values for passage of this signal, that is, it defines the characteristic along which
the particle velocity is Ẋ (τ ). The value of p along this characteristic may be found
from the relation p = p (v) that we would determine from Eq. (13.1.8). Note that the
derived solution is consistent with Eq. (13.1.13), because it merely entails measuring
the propagation distance from the location of the piston, rather than the reference
location at which x = 0.
Despite the availability of the general solution in Eq. (13.1.17), it seldom is
important to satisfy the moving boundary condition. To see why this is so, con-
sider Fig. 13.2, which depicts the position of the wall in space-time as the curve
x = X (τ ) that oscillates about the t-axis. The specific function is one cycle of
X = − (εc/ω) cos ωτ ), which leads to Ẋ = εc sin (ωτ ). The parameter ε is the
acoustic Mach number. Later examination will show that ε is a very small num-
ber, with values exceeding 0.001 being extraordinarily uncommon.
t
v=0
- 
v=
v=0
X v=
 v=0
x
Fig. 13.2 Characteristics originating from the instantaneous position X (τ ) of a moving boundary
whose nominal position is x = 0. The solid lines are the result of satisfying velocity continuity at
the instantaneous position of the boundary, whereas the dashed lines are the characteristics if the
boundary’s velocity is approximated as occurring at x = 0
Each point on the x = X (τ ) curve is the starting point for a characteristic, where
τ is the instant at which the
associated
signal was generated. The slope of any char-
acteristic is x/t = C̃ Ẋ (τ ) + Ẋ (τ ). The characteristics described in Fig. 13.2
are those for values of τ at which Ẋ is either zero, or a maximum, or a minimum.
These characteristics are compared in the figure to the nominal characteristic that
would result if the value of Ẋ (τ ) occurred at x = 0. For the selected X function, the
maximum and minimum values of Ẋ occur when X = 0. These characteristics are
the same as the corresponding nominal ones. This arrangement is contrasted with
those associated with instants when X is a maximum or minimum. The value of Ẋ is
zero at these instants, which means that the phase speed is c, but these characteristics
are shifted by the maximum amount relative to the nominal value.
The key aspect is that satisfying the moving boundary condition shifts a charac-
teristic parallel to the x-axis by X (τ ). This effect is quite different from the effect of
the dependence of phase speed on particle velocity, which changes the slope of the
characteristics. Consequently, the associated time shift increases proportionally to x.
In the vicinity of the boundary, these time shifts might be comparable. However, with
increasing x, the velocity dependence of the phase speed overwhelms the moving
boundary condition effect.
It follows that if a signal’s amplitude is within the range of common experience,
the error resulting from applying velocity continuity at the undisplaced position of a
boundary is insignificant. This allows us to replace the actual boundary condition in
Eq. (13.1.14) with an approximate stationary boundary condition that sets v = Ẋ (t)
at x = 0. Correspondingly, the particle velocity anywhere is described by removing
X (τ ) from the numerator of Eq. (13.1.17). Thus, the solution reduces to
x
v = Ẋ (τ ) , τ = t − (13.1.18)
C̃ (v) + v
As in the case of the moving boundary formulation, the corresponding value of p

may be found by inverting Eq. (13.1.18). This relation requires specification of the
equation of state, which is the topic for the next section.
Figure 13.1 describes the signal that propagates in the positive x direction when
the excitation at x = 0 is a known function of time. How would the analysis be altered
if the domain was x < 0? In that case, the negative sign in Eqs. (13.1.7) and (13.1.11)
would apply. Hence, the slope of a characteristic would be dt/dx = −1/ (c̃ − v). As
before p > 0 corresponds to c̃ > c. However, now dv/dρ < 0, so p > 0 corresponds
to v < 0. Thus, as was true for the previous case, the phase speed of a positive
pressure is c̃ + |v| > c, and the phase speed of a negative pressure is c̃ − |v| < c.
When we studied linear wave propagation, we satisfied arbitrary initial conditions
by combining the d’Alembert solution for waves that propagate in either direction.
This process cannot be implemented for nonlinear waves. A general reason is that
solutions cannot be superposed. If we add solutions for forward and backward prop-
agation, the result will generate product terms that do not fit the general scheme.
A more specific reason is that each characteristic line describes a constant pressure
value. If we contemplate combining waves that propagate in both directions, charac-
teristic curves for propagation in each direction would intersect. The pressure at any
(x, t) would be the sum of the value for the forward and backward characteristics that
intersect there. Thus, any attempt to combine forward and backward waves would
contradict the condition that the pressure is constant along a characteristic.
The consequence is that we may only use the Riemann solution to satisfy a limited
initial value problem. Suppose we know that a forward propagating wave has been
established somehow. (The argument is equally applicable to a backward propagating
wave.) Further suppose that at some instant, which we designate as t = 0, we have
measured the profile v as a function of x. This situation is depicted in Fig. 13.3.
Other than where they originate from, this diagram shows that there is no difference
between the characteristics for this initial value problem and those in the previous
figure for a boundary value problem.
Fig. 13.3 Characteristics for t

a nonlinear wave that
propagates in the sense of
increasing x. The properties
of this wave are taken to be v= 0
known at t = 0
v> 0 v= 0
v< 0
v= 0
x
v(x) @ t= 0
The task of actually satisfying this type of initial value problem is expedited by a
minor modification. The phase variable for a boundary value problem, Eq. (13.1.13),
is τ = t − x/ (c̃ + v). This definition facilitates matching the general solution to

whatever function is specified at the boundary. For the initial value problem, we
wish to match the general solution to whatever x function is specified at t = 0. This
process is assisted by redefining the phase variable to be θ = x − (c̃ + v) t, so that
θ = x at t = 0. The corresponding form of the general solution is
p = f (θ) , θ = x − (c̃ + v) t (13.1.19)
This is the form obtained by Poisson.6
13.1.4 Equations of State
If we wish to construct waveforms (p as a function of t at specified x) and profiles (p

as a function of x at specified t), it is necessary to evaluate the integral in Eq. (13.1.8),
which relates the values of p and v along a characteristic. Doing so requires specifi-
cation of the equation of state, which relates p and ρ. We begin with the case of an
ideal gas that is compressed and expanded adiabatically,
γ 1/γ
p ρ ρ p
+1= ⇐⇒ = +1 (13.1.20)
p0 ρ0 ρ0 p0
The linear speed of sound in an ideal gas is c2 = γp0 /ρ0 . We use this relation to
eliminate the ambient pressure, which converts the preceding relations to
γ 1/γ
p 1 ρ ρ γp
= − 1 ⇐⇒ = + 1 (13.1.21)
ρ0 c 2 γ ρ0 ρ0 ρ0 c2
From this, we find that
γ−1
c̃ γp 2γ
= +1 (13.1.22)
c ρ0 c 2
To determine the relation between p and v, we substitute ρ from Eq. (13.1.21) and
c̃ from the preceding into Eq. (13.1.8), which becomes

− γ+1
1 p
γp 2γ ρ0 c2 γp
v= + 1 d (13.1.23)
ρ0 c 0 ρ0 c2 γ ρ0 c2
6 S.D.
Poisson, “ Memoire sr la théorie de son,” J. l’École Polytechnique Paris, Vol. 7 (1808) pp.
319–392.
The result is
⎡ ⎤

γ−1
2 ⎣ γp 2γ
v=c +1 − 1⎦ (13.1.24)
γ−1 ρ0 c2
This equation may be solved for p as a function of v, which is

2γ/(γ−1)
p 1 γ−1v
= 1+ −1 (13.1.25)
ρ0 c2 γ 2 c
We can gain some insight into the significance of nonlinearity by examining graphs
of c̃ and v as functions of p; Fig. 13.4 describes both functions nondimensionally for
the case where the gas is air, γ = 1.4. The dotted lines are the relations for linear
acoustics, ρ = ρ0 + p/c2 , c̃ = c, v = p/ (ρ0 c).
Fig. 13.4 Nonlinear 2

dependence of ρ, c̃, and v on ~
c/c
the pressure in an ideal gas 1
for γ = 1.4. The analogous /0
0
relation for a linearized
equation of state is shown as −1 v/c
a dotted line
−2
-1 -0.5 0 0.5 1
Pressure, p/p0
The linear acoustic approximations are very close to the corresponding exact
curve in the range |p| /p0 < 0.2. This range corresponds to an acoustic amplitude
that is 20% of the ambient pressure. In air, for which p0 is 1 atm, this amplitude is
approximately 20 kPa, which corresponds to a harmonic signal at 217 dB//20 µPa.
Such a signal is beyond anything that would be encountered in a conventional acoustic
application.
The fact that these relations are close to those of linear theory does not mean that
nonlinearity is unimportant, because the phase variable τ is not simply t − x/c. One
could say that the linearized analysis gives the correct solution, except that it places
that solution in the wrong space-time location.
The logical question at this juncture is: What modifications are required if the
fluid is a liquid? The equation of state for an adiabatic process in a liquid is given by
Eq. (2.3.18),
2
p ρ 1B ρ
= −1 + −1 +··· (13.1.26)
ρ0 c2 ρ0 2 A ρ0
This series relation is derived experimentally. The value of A is the bulk modulus,
A = K = ρ0 c2 . It is convenient to consider B/A to be a single coefficient, because
that is the manner in which B arises in nonlinear waves. The value of B/A for
water ranges from 4.2 for distilled water at atmospheric pressure and 0◦ C to 6.2 for
distilled water between 200 and 4000 atm at 30◦ C. A nominal value B/A = 5 may
be used if environmental conditions are unspecified. It also is the value for seawater
at atmospheric pressure, 20◦ C, and 35 parts per thousand salinity.7
The equation of state for a liquid contains only two terms because the bulk modulus
is very large. For example, ρ0 c2 = 2.2 GPa for water. Hence, large values of ρ/ρ0 − 1
lead to very large values of p that are outside the usual realm of acoustics. Indeed, it is
challenging to generate pressures that are sufficiently large to measure the coefficients
of the cubic and higher terms in the equation of state. To illustrate this fact, a harmonic
signal at 260 dB//1 µPa in water corresponds to |ρ/ρ0 − 1| = 0.0065.
A corollary of halting the equation of state at quadratic terms is that any con-
stitutive parameters that are derived from it may be truncated at a consistent level.
To formulate Riemann’s solution, we need the density as a function of pressure. A
Taylor series expansion of the quadratic equation gives
2
ρ p 1B p
=1+ − +··· (13.1.27)
ρ0 ρ0 c 2 2A ρ0 c2
Differentiation of Eq. (13.1.26) gives

1/2
c̃ B ρ B p
= 1+ −1 =1+ +··· (13.1.28)
c A ρ0 2A ρ0 c2
In turn, the differential equation relating v to p becomes
dv 1 1
= = (13.1.29)
dp ρc̃ p B p
ρ0 c 1 + +··· 1+ +···
ρ0 c2 2A ρ0 c2
A consistent series truncation of a binomial series simplifies the differential equation

to
dv 1 B p
= 1− +1 +··· (13.1.30)
dp ρ0 c 2A ρ0 c2
Separating variables with the lower limits of the integral set to give v = 0 if p = 0
yields

v p 1 B p 2
= − + 1 +··· (13.1.31)
c ρ0 c2 2 2A ρ0 c2
7 R.T.
Beyer, “The Parameter B/A,” M.F. Hamilton & D.T. Blackstock, Nonlinear Acoustics, eds.,
Acoustical Society of America (2008) p. 34.
The inverse relation is

p v 1 B v 2
= + +1 +··· (13.1.32)
ρ0 c 2 c 2 2A c
The expressions for all quantities derived from

the equation of state for a liq-
uid have been truncated on the basis that |p| / ρ0 c2 is very small. Although the
bulk modulus
of any gas is much less than that of water, it nevertheless is large,
so |p| / ρ0 c2 for a signal in an ideal gas typically is much less than one. Series
expansions of the basic quantities based on taking |p| / ρ0 c2 to be much less than
one are said to constitute small-signal approximations. Waves that have this feature
are said to be weakly nonlinear. Acoustical studies seldom need to consider strong
nonlinearity.
In such circumstances, the relations for ideal gases and liquid may be combined.

To see how this is achieved, consider a Taylor series in powers of p/ ρ0 c2 of the
equation of state for an ideal gas, Eq. (13.1.21). When this series is truncated at the
first term beyond the expression for linear theory, the result is
2
ρ p γ−1 p
=1+ − +··· (13.1.33)
ρ0 ρ0 c 2 2 ρ0 c2
This relation and Eq. (13.1.27) are power series. Changing B/A in the latter to
γ − 1 yields the expression for a gas. A description that covers both media is obtained
by coefficient of nonlinearity defining a β such that
⎧
⎪ 1
⎨ (γ + 1) : Ideal gas
β= 2 (13.1.34)
⎪
⎩ B + 1 : Liquid
2A
Because the equations of state now match, the expressions for c̃ and v in terms of p
derived from them also will match. The unified constitutive relations that result are

ρ p p 2
=1+ − (β − 1) +···
ρ0 ρ0 c2 ρ0 c2

c̃ p (13.1.35)
= 1 + (β − 1) +···
c ρ0 c2
p v
= +···
ρ0 c2 c
The coefficient of nonlinearity is defined as it is in Eq. (13.1.34) because the difference

between the nonlinear and linear phase speeds is proportional to β, specifically
p
c̃ + v = c + βv = c + β (13.1.36)
ρ0 c

Smallness of p/ ρ0 c2 has the further implication that it usually is acceptable to
use v = p/ (ρ0 c), which is the same as the relation according to linear theory. In that
case, we may shorten the analysis slightly. Instead of solving Eq. (13.1.18) for v and
then evaluating the corresponding p, we may convert that relation to one that governs
p directly,
x
p = ρ0 cẊ (τ ) , v = Ẋ (τ ) , τ = t − (13.1.37)
c 1 + βp/ ρ0 c2
where the boundary condition is taken as v = Ẋ (t) at x = 0.

In summary, four fundamental aspects collectively constitute the small-signal
approximation:
• Continuity of particle velocity at a vibrating surface is satisfied at the static refer-

ence position of that boundary.
• Liquids and gases are governed by the same equations.
• The pressure and particle velocity for a plane wave propagating in the direction of
increasing x are related by p = ρ0 cv.
• The phase speed along a characteristic is c + βp/ (ρ0 c).
Each of these simplifies some aspect of an analysis, but none is essential.
13.1.5 Quantitative Evaluations
Now that we know c̃ and v as functions of p, the next task it to extract profiles and
waveforms from the Riemann solution. We will address the boundary value solution
in Eq. (13.1.13), but the developments are readily modified to treat the initial value
solution in Eq. (13.1.19). It is convenient to nondimensionalize the velocity excitation
at x = 0 by setting
Ẋ (τ ) = εcV (t) (13.1.38)
The parameter ε is the acoustic Mach number we encountered earlier. If we define

V (t) such that max (|V |) = 1, then ε is the ratio of the maximum particle velocity
to the linear speed of sound. In the small-signal approximation, ε is very small.
Graphical Construction
A useful approximation in some situations is a representation of V (t) as a piecewise
linear function, that is, a sequence of straight lines with a few locations at which the
slope changes. As the signal propagates, waveforms and spatial profiles of pressure
retain that attribute if the small-signal approximations apply. Consequently, it is only
necessary to follow the propagation of the phases at which the slope of the initial
waveform changes. After those phases are located, the intervals between them remain
straight lines.
Suppose τ is an instant at which the slope of V (τ ) changes. The particle velocity
for this phase is εcV (τ ), and the small-signal approximation of the pressure is
p (τ ) = ερ0 c2 V (τ ). To construct a waveform, the value of x is fixed. According
to Eq. (13.1.36), the phase speed is c [1 + βεV (τ )]. Hence, the time required for
this signal to propagate from the origin to the specified x is x/[c + βεcV (τ )]. By
definition, this signal was generated at x = 0 when t = τ , so the corresponding arrival
time of p (τ ) is τ + x/ [c + βεcV (τ )]. A plot of the waveform observed at x would
show this time as the abscissa and p (τ ) as the ordinate. This process is repeated for
each value of τ at which V (τ ) changes slope, after which the waveform is obtained by
joining the plotted points with straight lines. The procedure is illustrated in Fig. 13.5
for a triangular pulse whose duration is T .
p
2=0.75T Waveform at x = 0
1 2 0c2
3=T t
3
p
2 + xA / (c + c) Waveform at xA
1 2 0c2
t
xA /c 3
3+xA /c
p 2 + xB / (c + c) Waveform at xB
0c2 2
1
t
xB /c 3
3+xB /c
Fig. 13.5 Graphical construction of the waveforms at two spatial locations xA < xB . The phase
speed of the signal at each instant of slope discontinuity is constant
There are three instants along which the slope changes discontinuously, τ1 = 0,
τ2 = 0.75T , and τ3 = T . The particle velocity at τ2 is the maximum value εc. The first
plot is the waveform at x = 0, which is obtained by setting p = ρ0 cεV (t). Because
v (τ1 ) = v (τ3 ) = 0, the phase speed of these signals is the linear speed of sound,
c. The time required for p (τ1 ) and p (τ3 ) to propagate to a position xA > 0 is xA /c.
The pressure along the characteristic line associated with τ2 is p = ρ0 c2 ε. The phase
speed along this characteristic is constant at c + βεcV (τ ). The propagation time
required for p (τ2 ) to arrive at xA is xA / (c + βεcV (τ )), which is less than the travel
time for p (τ1 ) and p (τ3 ). Thus, the waveform seems to begin to lean in the sense of
earlier time. (If the waveform has an interval in which p is negative, that portion of
the waveform will seem to lean in the sense of later time.)
To obtain the waveform at a farther location xB distance, we can evaluate the
arrival time for each phase by adding the departure time τ to xB divided by the phase
speed for that pressure, as is done in Fig. 13.5. An alternative is to use the waveform
at xA as the reference. The propagation time from xA to xB is xB − xA divided by the
phase speed corresponding to the pressure on that characteristic.
At a certain x, the arrival time of p (τ2 ) will equal that of p (τ1 ). Such a
condition marks the formation of a shock. Equating the two arrival times gives
τ2 + x/ (c + βεc) = τ1 + x/c. This value of x is the shock formation distance,
c + βεc
xshock = c (τ2 − τ1 ) (13.1.39)
βεc
(This expression could be simplified by approximating the numerator as c based

on ε 1.) The value of xshock depends on the magnitude and shape of the initial
waveform. For now, we regard this as the limiting range for our evaluations.
Evaluation of a spatial profile follows a similar procedure. We select τ as the phase
of a significant feature we wish to follow. The phase speed of this feature is c + βεc.
It departed the boundary at time τ . At a designated instant t, the propagation time is
t − τ , so the feature will be situated at x = (c + βεc)t.
Figure 13.6 depicts this construction for the initial waveform in Fig. 13.5. The
first profile describes tA = τ3 . This is the instant when p (τ3 ) leaves the boundary,
so x3 = 0. Because p (τ1 ) = 0, its phase speed is c and its location is x1 = ctA . The
propogation time for p (τ2 ) is tA − τ2 . According to the Riemann solution, its location
Fig. 13.6 Construction of p  =0.75T

spatial profiles at various 2 Waveform at x = 0
instants
1 2 0c2
t
3=T 3
p (c + c)(A - 2)
Profile at tA = 3
2 0c2
x
3 cA 1
p Profile at tB >> 3
(c + c)(B - 2)
0c2 2
x
c(tB -3) 3 1
ctB
is x2 = c (tA − τ2 ) + βεc (tA − τ2 ). However, the pulse is very brief. Consequently,

c (tA − τ2 ) is small, and βεc (tA − τ2 ) is a negligible addition. Thus, at this early
instant the spatial profile differs little from the prediction of linear theory. The second
profile is for a much later instant tB . The phase speed of p (τ1 = 0) and p (τ3 ) is c,
so the spatial extent of the pulse remains at cτ3 . However, the profile changes shape
because the phase speed of p (τ2 ) is greater than c. Because p (τ2 ) propagates farther
than the beginning and end of the pulse, the profile seems to lean in the sense of
increasing x. (Regions where the pressure is negative would seem to lean in the
sense of decreasing x.)
With increasing time, the profile propagates farther, and the p (τ2 ) signal increas-
ingly advances relative to the p (τ1 ) and p (τ3 ) signals. Eventually, it catches up to
p (τ1 ), which marks the advent of a shock. The instant tshock when this condition
occurs may be found by equating the locations where p (τ1 ) and p (τ2 ) occur. The
shock formation distance xshock = ctshock is the same as Eq. (13.1.39).
EXAMPLE 13.1 The waveforms described in the sketch are alternative

velocity inputs at x = 0. Their only difference is phase-reversal. For each case,
determine the shock formation distance, xshock . Then, evaluate each waveform
at x = xshock /2 and x = xshock . The value of v0 is sufficiently small to permit
considering the propagation to be weakly nonlinear.
v Case A v Case B
T
T/4 3T/ 4
v0 v0
2 4 3
1 t 1 t
3 2 4
T/2 T/4
3T/ 4
T T/2
Figure 1.
Significance
Graphical constructions enhance intuitive understanding, which is further enhanced
by examining the similarities and differences between the propagation properties of
the alternative waveforms.
Solution
The input functions are the velocity at x = 0, where τ = t. There are four instants at
which the slope of each input changes. For case A, v1(A) = 0 at τ = 0, v2(A) = v0 at
τ = T /4, v3(A) = −v0 at τ = 3T /4, and v4(A) = 0 at τ = T . For case B, the velocities
at τ2 and τ3 are the negative

of the value for case A. According to the weakly nonlinear
approximations, p/ ρ0 c2 = v/c, so the equations of the relevant characteristics are
Case A Case B
x x
t1 = t1 =
c c
T x T x
t2 = + t2 = +
4 c + βv0 4 c − βv0
3T x 3T x
t3 = + t3 = +
4 c − βv0 4 c + βv0
x x
t4 = T + t4 = T +
c c
A shock forms at the location where characteristics for different values of τ
intersect. The denominator of the x term for t2 in case A is bigger than the denominator
for t1 , so there is a value of x at which t2 and t1 are equal. Similar reasoning suggests
that it is possible that t4 = t3 in case A, whereas in case B the only possibility is that
t3 = t2 . These conditions occur at
x x T v0 T
Case A: t2 = t1 =⇒ − = =⇒ β x = (c + βv0 )
c c + βv0 4 c 4
x x T v0 T
Case A: t4 = t3 =⇒ − = =⇒ β x = (c − βv0 )
c − βv0 c 4 c 4
x x T v0 c2 − (βv0 )2 T
Case B: t3 = t2 =⇒ − = =⇒ 2β x =
c − βv0 c + βv0 2 c c 2
The value of v0 /c is stated to be sufficiently small to justify using weak shock

approximation, so the expression for x in each case reduces to
cT
xshock =
4β (v0 /c)
To plot the waveforms at x = xshock /2, we substitute x shock /2 into the expres-
sionsfor each tj and then plot each time value against the corresponding pressure
ρ0 cv τj . The time values are simplified by using the smallness of v0 /c to approxi-
mate 1/ (c ± βv0 ) by the first two terms of a binomial series. For example,
v0
xshock T 1 − β cT T xshock
At x = , t2 ≈ + c = +
2 4 c 8β (v0 /c) 8 2c
The full set of values is
Case A (x = xshock /2) Case B (x = xshock /2)

xshock xshock
t1 = t1 =
2c 2c
T xshock 3T xshock
t2 = + t2 = +
8 2c 8 2c
7T xshock 5T xshock
t3 = + t3 = +
8 2c 8 2c
xshock xshock
t4 = T + t4 = T +
2c 2c
The waveform corresponding to these instants is shown in Fig. 2.
p Case A p Case B
T/8 T
5T/8
0cv0 0cv0
t t
T/ 2
xshock 7T/ 8 xshock 3T/8
2c T 2c T/2
Figure 2.
The same procedure with x set to xshock leads to
Case A (x = xshock ) Case B (x = xshock )

xshock xshock
t1 = t1 =
c c
xshock T xshock
t2 = t2 = +
c 2 c
xshock T x shock
t3 = T + t3 = +
c 2 c
xshock xshock
t4 = T + t4 = T +
c c
Figure 3 shows that both waveforms have indeed developed a shock at xshock .
p Case A p Case B
T
0cv0 0cv0
t t
T/ 2
xshock T xshock T/2
c c
Figure 3.
Although it might seem as though there is a fundamental difference between the

signals generated by the alternative inputs, they actually are quite similar. Rather
than considering case B to be the negative of case A at x = 0, the similarities become
evident if we compare the behavior of the positive and negative phases. In comparison
with case A, the positive phase in case B is retarded in time by T /2, and the negative
phase is advanced by T /2. From this perspective, the corresponding phases in each
case are altered in the same way as the signal propagates. Both waveforms have the
shape of teeth of a sawblade. We will see that this is a tendency that is shared by all
oscillatory waveforms.
Computation of Characteristics Graphical construction of waveforms and profiles
is impractical if the input is not piecewise linear. It also is inappropriate if we believe
that the acoustic Mach number is too large to consider the signal to be weakly
nonlinear. The method developed here is a computational version of the graphical
algorithm. It allows for evaluation of as many points as necessary to resolve the input
waveform, so it can be used in any situation.
We begin with the weakly nonlinear signal. In the graphical method, a value of the
phase variable τ is selected. The initial velocity v (τ ) is used to evaluate the associated
pressure p (τ ), from which either x or t is determined by solving the equation for the
nonlinear phase speed of that signal. The concept is to use a computer algorithm to
perform these operations for many values of τ in situations where the input function
is complicated.
The first step is to discretize the interval during which the input function is
nonzero. At x = 0, τ and t are equal, so we let τn denote the sampled instants. (Usu-
ally we select a uniform increment, so that τn = n, but doing so is not essential.)
The corresponding set of particle velocities is vn = εcV (τn ), and the pressures are
pn = ερ0 c2 V (τn ).
If we wish to construct the waveform at some point xA , we solve Eq. (13.1.36) for
the time tn corresponding to the selected values of xA and τn , which gives
xA
tn = τn + (13.1.40)
c (1 + βεV (τn ))
Plotting pn as the ordinate and tn as the abscissa for all n yields the waveform. To
obtain the profile at a specific instant tA , we solve the characteristic’s equation for
the location xn ,
xn = (1 + βεV (τn )) c (tA − τn ) (13.1.41)
The desired profile results from plotting pn versus xn .

Unlike the graphical method, the procedure is readily modified to handle a sit-
uation where ε ≡ max(v)/c is sufficiently large that the weakly nonlinear approx-
imations might not be valid. It is seldom necessary to consider this possibility for
a liquid, so Eq. (13.1.25) for an ideal gas would be the governing equation of state.
Rather than using pn = ρc cvn to evaluate the pressure associated with characteristic
τn , we would use Eq. (13.1.25). The corresponding value c̃n would be found from
Eq. (13.1.22). Characteristic lines in this case are described by the original expres-
sion, Eq. (13.1.13), so the relation between x and t for the signal that departed the
boundary at time τ is
x = (t − τn ) (c̃n + vn ) (13.1.42)
In all other respects, the evaluation would proceed as it does when the signal is
weakly nonlinear.
EXAMPLE 13.2 A plane wave in air is induced by a harmonically varying

velocity at x = 0. The motion begins at t = 0, and the velocity amplitude is
0.002c. Evaluate the spatial distribution of the pressure signal at t = 50T ,
where T is the period of the boundary velocity. Also determine the waveform
at x = 50λ, where λ is the wavelength according to linear theory.
Significance
Application of the general procedure for tracing signals along characteristic lines
serves to enhance our ability to implement the method for other problems. Further-
more, harmonic signals are the fundamental building blocks of linear acoustics, so
using such a signal as the exemplar will bring to the fore some important phenomena.
Solution
The excitationis v = εc sin (ωτ ) h (t) at x = 0 with ε = 0.002 as specified. For ref-
erence, ε = 2 10−3 corresponds to 140 dB//20 µPa. The fluid is specified to be air,
so γ = 1.4. The sound pressure level is quite high. For this reason, we should assess
whether it is appropriate to use the small-signal approximations. To do so, let us
consider the signal at the instant when the velocity is εc. According to Eq. (13.1.25),
v/c = 0.002 corresponds to p = 0.0020024ρ0 c2 , whereas the small-signal approxi-
mations give p = 0.002ρ0 c2 . The error for the latter value is 0.1%, so application of
the small-signal approximations is justified.
There is no need to specify the frequency if we work nondimensionally. To that
end, we write the equation for the characteristic as
kx
ωτn = ωt −
(1 + βvn /c)
In other words, we take the characteristic value to be ωτn and consider the coordinates
in space-time to be kx and ωt.
Before we can select the τn values, we must decide the interval to be discretized.
The velocity input at x = 0 begins at t = 0, at which instant v = 0. This corresponds
to p = 0, so the characteristic line through the origin is t − x/c = 0. Thus, at any
instant t, the signal will extend in space from x = 0 to x = ct. Standard terminology
is to say that x = ct is the leading edge. We seek the profile for t = 50 (2π/ω),
so the range to be sampled is 0 ≤ ωτ ≤ 100π. To assure that the sampling leads to
smooth curves, the sampling rate is selected to be 20 per period, so ωτn = n (2π//20),
n = 0, 1, ...1000. In regard to evaluation of waveforms, the acoustic wavelength is

λ = 2π/k, so we must determine the waveform at kx = 100π. It will be proven later
that a periodic excitation at the boundary leads to waveforms having the same period.
Consequently, we only need to see one period of the waveform. The sampling of ωτn
for the evaluation of profiles is adequate for this purpose.
The algorithm is implemented by defining a column vector of τn values. The
column vector of velocity values at the sampled instants is vn /c = ε sin (ωτn ). The
pressure corresponding to each vn is pn / ρ0 c2 = vn /c. To find a spatial profile at
time ωt, we solve the characteristic definition for the kxn value associated with pn
at this ωt, so that kxn = (1 + βεvn /c) (ωt − ωτn ). Waveforms result from solving
Eq. (1) for ωtn at location kxn corresponding to pressure pn , which gives ωtn =
ωτn + kx/ (1 + βvn /c).
Figure 1 describes the profile at ωt = 100π. (Division of kxn by 2π gives distance
as a fraction of a linear wavelength. Division of ωtn by 2π gives time as a fraction of
the period.) The first plot displays the full extent of the signal. The compressed nature
of the x scale makes it difficult to see specific features, so the second graph shows
the profile from the boundary outward to five wavelengths. Nothing remarkable is
evident—the distribution seems to be the sine curve predicted by linear theory. The
third plot shows the region five wavelengths back from the wavefront, x = ct. The
wave in this region seems to be leaning forward in space. This is a consequence
of the positive pressure phase propagating faster than c, while the negative phase
propagates slower than c.

p/( 0 c 2 )
−
0 10 20 30 40 50

p/( 0 c 2 )
−
0 1 2 3 4 5

p/( 0 c 2 )
−
45 46 47 48 49 50 51
Distance, x/
Figure 1.
What would happen if we were to perform the same construction for a much larger
time? The leaning effect will be so exaggerated that the computation would yield a
multivalued profile. This behavior is exhibited in Fig. 2, which describes the forward
part of the spatial profile at ωt = 300π. Only one
p/( 0 c 2 )
0
−
146 147 148 149 150 151
Distance, x/
Figure 2.
pressure can exist at a specific location and instant, so it is evident that there is
something wrong. What is missing is the formation of a shock, at which the pressure
changes discontinuously. The instant at which the shock forms is that at which a
multivalued condition first occurs. Because the positive lobe at the leading edge,
x = ct, leans forward the most, occurrence of a shock is first manifested at the
leading edge as infinite ∂p/∂x. We will investigate shocks in a later section, where
we will find that an infinite gradient is a general criterion marking shock formation.
The time required for a positive pressure to propagate to a specific x is less than it
is for a negative pressure. Therefore, the leaning appearance of waveforms is in the
opposite sense of a spatial profile. This feature is evident in Fig. 3. With increasing
x, the leaning tendency increases, until a shock forms at a certain distance. The first
occurrence of a shock in a waveform is manifested by a vertical tangent at some
instant, at which ∂p/∂t is infinite.

p/( 0 c 2 )
−
49 50 51 52 53 54 55
Time, t/T
Figure 3.
The waveform has other interesting properties. It is zero until t = x/c and then,
it is periodic at the period T of the boundary excitation. Periodicity follows from the
fact that the characteristic lines in x, t space corresponding to a periodic v (τ ) form
a repeated pattern. This attribute will be proven to be general. A periodic waveform
generated by a periodic excitation may be represented as a Fourier series. Thus,
another effect of nonlinearity is to create harmonics whose amplitude is dependent
on the propagation distance. It might happen that the some harmonics are depleted,
while others are generated or enhanced. In any event, propagation leads to an ever
increasing level of distortion. Distortion is another effect that will be explored in
greater detail.
Solution of a Nonlinear Algebraic Equation
The method of constructing profiles and waveforms by computing x or t along a
characteristic is reliable and readily programmed, so why seek another method? One
reason is that analyses that include dissipation or nonplanar effects might be posed in
a different form. A more common reason lies in the fact that the interval between the
instants at which a waveform is computed is not uniform, that is, a constant increment
τn+1 − τn does not lead to a constant increment tn+1 − tn . The consequence of the
waveform being known discretely at instants that are not uniformly spaced is that
standard FFT routines cannot be used to decompose a waveform. This is a serious
shortcoming, because the harmonic content of the distorted wave often is of primary
interest.
The concept underlying the method we shall now develop is that the Riemann
solution may be assembled as a nonlinear equation for p as a function of x and t.
Such an equation may be solved with the aid of numerical methods. We will derive
the method for the boundary value problem, but the procedure may be modified with
little effort to address the initial value problem. As before, the boundary condition is
v = εcV (t) at x = 0.
Let us begin by considering the procedure when ε is sufficiently small to
warrant considering the signal to be weakly nonlinear. Correspondingly, we set
p = ρ0 c2 εV (t) at x = 0. The same approximation is used to replace v/c in the equa-
tion for the characteristics. Thus, the nonlinear plane wave matching the boundary
velocity is defined by

x/c
p = ερ0 c V2
t− (13.1.43)
1 + βp/ ρ0 c2
It is evident that Eq. (13.1.43) is a function relating x, t, and p. Cases where p

and either x or t are specified are equivalent to the previous method, which used
constancy of the phase speed along a characteristic. Our interest here is in finding
the value of p at specified x and t. The equation cannot be solved analytically, so we
use numerical methods. It is useful to nondimensionalize the pressure relative to the
bulk modulus, such that
p
q= (13.1.44)
ρ0 c2
Correspondingly, the equation to be solved is

x/c
G (q, x, t) ≡ q − εV t − =0 (13.1.45)
1 + βq
Many methods have been developed to find the roots of a nonlinear algebraic
equation. One that works well for the function described above is Newton’s method,
which is sometimes referred to as the method of tangents. Let q(j) denote the jth
iteration. The next iteration is described by
⎛ ⎞

⎜ G (q, x, t) ⎟
q(j+1) (j)
=q −⎝ ⎜ ⎟ (13.1.46)
∂ ⎠
G (q, x, t)
∂q q=q(j)
The derivative of the G function in Eq. (13.1.45) is

∂ x/c βx/c
G (q, x, t) = 1 − εV̇ t − (13.1.47)
∂q 1 + βq (1 + βq)2
where V̇ ( ) denotes the time derivative of the excitation function.

A straightforward modification allows for evaluation of the signal in situations
where the signal is strongly nonlinear. In that case, the phase speed as a function
of pressure is c̃ (p) + v (p), which are the functions given by Eqs. (13.1.22) and
(13.1.24), respectively. The corresponding function representing the Riemann solu-
tion is

x
G (q, x, t) ≡ q − εV t − 2 =0 (13.1.48)
c̃ ρ0 c q + v ρ0 c2 q
Carrying out a numerical solution by Newton’s method requires the value of ∂G/∂q
corresponding to the latest iterative value of q(j) . The expression for that derivative
is complicated by the intricate nature of the expressions for c̃ (p) and v (p). Fortu-
nately, it is adequate to use Eq. (13.1.47) for a weakly nonlinear wave to evaluate
∂G/∂q, because a small error in this quantity merely alters the increment of q(j) in
Eq. (13.1.46).
The Newton solver algorithm requires an initial guess, q(0) . The fact that |q| 1
suggests that a good guess would be the value according to linear theory, which is
x
q(0) = εV t − (13.1.49)
c
This usually will lead to a root in a few iterations, but this assertion only holds if
x is not too close to the shock formation distance. A more robust approach is to
evaluate G (q, x, t) for a set of values qn . Then, the best starting guess is that for
which G (qn , x, t) is closest to zero. A computational operation would implement
min (|G (qn , x, t)|) =⇒ q(0) = qn (13.1.50)
In MATLAB, if G is a vector array of these values, one can find n by writing

[n,min_G]=min(abs(G)). The appropriate range of qn values is that which
matches the range of εV (t) for all possible t. For example, if V (t) = sin (ωt), then
the range of qn would be −ε to +ε.
To illustrate this procedure, consider the signal in Example 13.2. Let us determine
the value of q at kx = 100π at the instant ωt = 101.98π, which is slightly less than
one period after the wavefront arrives at that location. The range of qn values is −ε
to ε, but the range is extended in Fig. 13.7 in order to view the behavior of the G
function with greater clarity. The function is sampled at qn = ε/50. A scan of the
computed values of G shows that it crosses the zero axis between q91 = −4 10−4 ,

where G = −2.747 10−6 , and q92 = −3.6 10−4 , where G = 7.623 10−6 . The
first qn gives the smaller magnitude
of G, so the solution algorithm would begin with
the initial value q(0) = −4 10−4 . Theroot is found after one iteration
based
on a
convergence criterion that q(j) − q(j−1) < ε/105 ; it is q = −3.8944 10−4 .
Figure 13.7 also shows a situation where the concept of finding q by solving
a nonlinear equation becomes problematic. When ωt = kx + 1.98π at kx = 200π,
which is twice as far from the boundary, there are three values of q for which G = 0
at the selected x and t. This means that three values of q are consistent with the
Riemann solution. Hence, the occurrence of multiple roots for G at a selected x and
t is another way that shocks are manifested. The Riemann solution is not valid in the
presence of shocks. Nevertheless, the method by which shocks are incorporated into
an analysis will require the multi-valued portion of the Riemann waveform. Such an
evaluation could implement the root search procedure, with the scan initiated near
each q value at which the G function changes sign.
Fig. 13.7 Graph of the 2

function G (q, x, t) whose
G(q,x,t) x 103
1
root gives the
kx = 200
nondimensional
pressure 0
q = p/ ρ0 c2 at kx = 100
−1
ωt = kx + 1.98π, ε = 0.002
−2
−1.5 −1 −0.5 0 0.5 1 1.5
q/
EXAMPLE 13.3 The particle velocity at x = 0 is εV (t) =

εc [sin (ωt) + 0.5 sin (2ωt)] for t > 0, and it was zero for t < 0. The
fluid is air. Consider the case where ε = 9 10−5 . Compare the pressure
waveforms at kx = 800π and kx = 1600π to the waveform predicted by linear
theory. Further, because the excitation is periodic at T = 2π/ω, it may be
represented as a Fourier series. Construct a graph showing the dependence of
the Fourier pressure coefficients on the propagation distance.
Significance
The evaluation of the signal is unremarkable, but the growth or decay of harmonics
is a primary issue for nonlinear acoustics. This evaluation is a useful reminder of
spectral techniques that were developed in Chap. 1.
Solution
We will implement the Newton solver algorithm in nondimensional form because
the value of ω is not given. The value of ε is sufficiently small to justify considering
the wave to be weakly nonlinear. The function to be solved is

kx 2kx
G (q, kx, ωt) ≡ q − ε sin ωt − + 0.5 sin 2ωt − h (t)
1 + βq 1 + βq
The derivative of this function with respect to q is

∂ kx 2kx (kx) β
G(q, x, t) = 1 − ε cos ωt − + cos 2ωt − h (t)
∂q 1 + βq 1 + βq (1 + βq)2
The value of β for air is 1.2.

The value of q predicted by linear theory is the most efficient way to initiate
Newton’s method. If doing so fails to lead to a convergence, then the search algo-
rithm described by Eq. (13.1.50) may be invoked. (This alternative procedure was
not required to obtain the present results.) Thus, the root search is initiated with
q(0) = εc [sin (ωt − kx) + 0.5 sin (2ωt − 2kx)] h (t)
We select an even increment for the time values, ωtn = 2πn/N, n = 0, ..., N − 1,
in order to exploit FFT technology. The nondimensional fundamental frequency is
ωτ = 2π, and the sampling interval is ωt = 2n/N. Thus, the fundamental fre-
quency of the data is one, and its highest frequency is N/2. Although we will only
examine harmonics m ≤ 5, it is crucial that the highest frequency be greater than
the significant frequencies contained in the data. The waveform at x = 0 contains
only two harmonics, so the sampling criterion would be met with a small value
of N. However, as the signal propagates its steepens in a portion of a period. This
raises the Nyquist requirement. Consequently, the evaluations will be carried out
with N = 128.
The procedure followed for each kx is to perform a program loop in which tn is set,
and the corresponding value of qn is found by Newton’s method. After completion
of that loop, the set of qn values is input to an FFT routine. The output of that routine
is a set of DFT coefficients that usually must be scaled to obtain Fourier series
coefficients Pm . (In MATLAB, this factor is 2/N.) The Pm values obtained at each kx
are preserved for plotting. A simple check of the computations uses the fact that the
mean value of v (t) at x = 0 is zero. This attribute should be preserved as the signal
propagates. (The first value returned by most FFT routines is the mean value, P0 .)
The waveforms at the requested locations appear in Fig. 1. The abscissa is the
retarded time, that is, the elapsed time after the wavefront arrives at ωt = kx. Two
periods appear there, but the data for the second period merely periodically replicates
that of the first, with ωtn+N = ωtn + 2π and pn+N = pn for n = 0, ...N − 1.
v = [sin(t) + 0.5 sin(2t)] @ x = 0

2(10-4)
p/(0c2)
0 kx = 0
kx = 800
kx = 1600
−2(10-4)
0 0.5 1 1.5 2
Retarded time, (t - kx)/2
Figure 1.
The waveform at kx = 0 is the same as the linear solution. Although it has the
appearance of being advanced for positive p and retarded for negative p, this quality
is a consequence of the manner in which the harmonics combine, not the effect of
nonlinearity. The distortion of the waveform enhances this initial appearance. At
kx = 1600π, the effect is sufficiently large that the slope of the waveform seems to
be nearly vertical at the start of each period.
Waveforms at two values of kx were requested, but such a sample is too sparse
for a description of the position dependence of the harmonic amplitudes. Therefore,
the waveforms at kxm = 400π, 800π, 1200π, and 1600π were processed. The result
is Fig. 2. Only the imaginary parts of the amplitudes are shown because the real parts
remain zero. None of the amplitudes are constant. The first and second harmonics are
the ones contained in the input, whereas the third and higher harmonics are generated
by nonlinearity. There are no dissipation mechanisms, so the total energy in a period
is conserved. Typical discussions of the generation of harmonics state that energy
is transferred from lower harmonics into higher harmonics as the wave propagates.
This graph shows this description to be only partially true. The harmonics that are
not directly generated are n ≥ 3. They do indeed grow as the wave propagates. The
directly generated harmonics are n = 1 and n = 2, but only the former is depleted,
whereas m = 2 grows slightly.
v = [sin(t) + 0.5 sin(2t)] @ x = 0
0
Im(Pn)/(0c2)
−5(10-5)
−10(10-5) n=1 n=2 n=3 n=4 n=5
0 200 400 600 800

Propagation distance kx/
Figure 2.
The interchange of energy between harmonics and the resulting distortion of

the waveform are sensitive to the content of the excitation. To demonstrate this,
Fig. 3 shows the waveforms that result from shifting the second harmonic of the
input by 180◦ , so that v = ε [sin (ωt) − 0.5 sin (2ωt)] at x = 0. The interference of
the harmonics of p (t) at x = 0 is such that the starting waveform has less of the
leaning appearance than it did in the previous case. The nonlinear advancement
and retardation are the same as in the previous case. However, the overall leaning
appearance of the initial waveform is less noticeable than it was in the previous
case. The consequence is that the maximum slope of the waveform at kx = 1600π
is reduced, which suggests that shocks will form at a larger distance.
v = [sin(t) - 0.5 sin(2t)] @ x = 0
2(10-4)
p/(0c2)
0 kx = 0
kx = 800
kx = 1600
−2(10-4)
0 0.5 1 1.5 2
Retarded time, (t - kx)/2
Figure 3.
The harmonic content of these signals is described in Fig. 4. Reversal of the

second harmonic at x = 0 has a rather drastic effect on the growth and decay of
harmonic amplitudes. Increasing propagation distance leads to depletion of the m = 2
harmonic. The energy lost from that harmonic is transferred into the other harmonics,
including the first. The growth of the higher harmonics is similar to the previous case,
but now the fourth harmonic grows more rapidly than the third.
v = [sin(t) - 0.5 sin(2t)] @ x = 0
5(10-5)
Im(Pn)/(0c 2)
−5(10-5) n=1 n=2 n=3 n=4 n=5
−10(10-5)
0 200 400 600 800
Propagation distance kx/
Figure 4.
13.2 Effects of Nonlinearity
The examples disclosed that some interesting phenomena arise in nonlinear plane
waves. Two of particular interest are harmonic distortion and the occurrence of
shocks. Both features are encountered in other types of waves. Both are sufficiently
important to warrant a more detailed analysis, which is what we shall now do.
13.2.1 Harmonic Generation
A periodic excitation at the boundary leads to a periodic waveform. This is so because

the characteristics form a periodic pattern parallel to the t-axis. Periodicity is readily
13.2 Effects of Nonlinearity 569
proven. Two phases separated by a period are τ and τ + T . The arrival time of these
phases at a specified x is t and t + t, where
x x
t=τ+ , t + t = (τ + T ) + (13.2.1)
c + βεV (τ ) c + βεV (τ + T )
Because V (τ ) is periodic, V (τ + T ) = V (τ ), so it must be that t = T . A corollary

of periodicity is that the signal may be expressed as a Fourier series. Rather than using
an FFT to process discretized data, as we did in Example 13.3, let us consider a formal
analysis of the Fourier coefficients.
It is convenient to measure time relative to the signal’s arrival at position x. Because
the pressure at the wavefront is zero, the arrival time is x/c. Thus, the time variable
is the retarded time t , where
x
t = t − (13.2.2)
c
We consider the weakly nonlinear wave generated by a periodic boundary vibration
v = εcV (t) at x = 0. We further require that V (t) = 0 for t < 0, in order that the
velocity be continuous at the initial instant. This condition is added because a dis-
continuity in the waveform means the existence of a shock, which invalidates using
the Riemann solution throughout a period.
A periodic function that is zero at t = 0 may be represented as a Fourier sine
series. The period of the excitation is T , so the fundamental frequency is ω = 2π/T .
A Fourier sine series whose independent variable is t is used,
∞
#
p= Pn sin nωt , t > 0 (13.2.3)
n=1
Concurrently, the Riemann solution in Eq. (13.1.37) also describes p, so it must be

that
#∞
x
Pn sin nωt = ρ0 c2 εV (τ ) , τ = t + βεV (τ ) (13.2.4)
n=1
c
The relation for τ differs from the earlier form by application of a binomial series
expansion to clear the denominator, based on the fact that |ε| 1.
The Fourier pressure coefficients are found by applying the orthogonality property
of sine functions to Eq. (13.2.4). The first period following t = 0 covers the interval
0 < t < T . Therefore, the coefficients are given by

2 T
Pn = ρ0 c2 ε V (τ ) sin nωt dt (13.2.5)
T 0
Because V (τ ) is an arbitrary function, we use Eq. (13.2.4) to change the integration

variable from t to τ . Doing so gives
T $ % $ %
2 x x
Pn = ρ0 c ε2
V (τ ) sin nω τ − βεV (τ ) 1 − βεV̇ (τ ) dτ (13.2.6)
T 0 c c
The factor of V (τ ) is a perfect differential, which suggests an integration by parts.

Because V (0) = V (T ) = 0, this gives
T $ %
2 x
Pn = ρ0 c2 ε V̇ (τ ) cos nω τ − βεV (τ ) dτ (13.2.7)
nωT 0 c
The periodic nature of the integrand allows the integration range to be shifted to
−T /2 < τ < T /2. Furthermore, because V (τ ) is an odd function relative to τ = 0,
V̇ (τ ) is even, so that the product of V̇ (τ ) and the cosine term also is even. Therefore,
the integration interval may be reduced to 0 < τ < T /2 with the result of the integral
doubled. The result is

2 T /2
2nπ x
Pn = ρ0 c2 ε V̇ (τ ) cos τ − βεV (τ ) dτ (13.2.8)
nπ 0 T c
It might be desirable to compare these Pn values to those obtained from FFT analysis
of a waveform, which is the procedure in the previous example. To do so, it is
necessary to be cognizant of the fact that the FFT gives the coefficients of a complex
Fourier series, whereas Eq. (13.2.8) gives the coefficients of a Fourier sine series.
Equation (13.2.8) may be used to describe the propagation of any initial waveform
for which p = 0 at the beginning of a period. The expression seems to be unpromising
for an analytical integration, but it can be evaluated numerically. Harmonic excitation
is one case where an analytical result can be obtained. If V (τ ) = sin (ωτ ), then an
identity for the product of cosines, accompanied by the change of variables ξ = ωτ ,
leads to
π
2
Pn = ρ0 c2 ε cos (ξ) cos [n (ξ − kxβε sin (ξ))] dξ
nπ 0
π
1
≡ ρ0 c2 ε {cos [(n − 1) ξ − nkxβε sin (ξ)]
nπ 0
+ cos [(n + 1) ξ − nkxβε sin (ξ)]} dξ (13.2.9)
Both parts of the integrand match a fundamental formula for Bessel functions.8 From
it, we find that
Pn ε& ' 2
= Jn−1 (nβεkx) + Jn+1 (nβεkx) ≡ Jn (nβεkx) (13.2.10)
ρ0 c 2 n nβkx
8 M.I. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover (1965)
Eq. (9.1.21).
where the last form stems from a recurrence relation for the Bessel functions. A more
meaningful form results when distance is described relative to the shock formation
distance of an initially harmonic plane wave, (xshock )sine . Shock formation is the
subject of the next subsection. It will be found that (xshock )sine = 1/(εβk), so that
Pn 2ε
= Jn (nσ) (13.2.11)
ρ0 c 2 nσ
The symbol σ is the distance ratio,
x
σ= = βεkx (13.2.12)
(xshock )sine
The corresponding series representation is
#∞
2
p = ρ0 c2 ε Jn (nσ) sin nωt (13.2.13)
n=1
nσ
Because this analysis began with the Riemann solution, the basic limitation to regions
where x < (xshock )sine , or σ < 1, applies equally to the Fourier series.
The Fourier series analysis was first worked out by Fubini-Ghiron in 1935.9 A
remarkable aspect is the simplicity of the result, for it tells us that the harmonic ampli-
tudes depend on only two parameters: the harmonic number and distance relative to
the position where a shock first forms. However, this property is specific to a sinu-
soidal excitation function; other excitations will lead to dependence on additional
parameters.
Figure 13.8 plots the amplitudes. We see that the fundamental harmonic amplitude
lessens with increasing distance, whereas the higher harmonic amplitudes grow.
Another important property of a sinusoidal excitation is that the amplitudes at any x
decrease as the harmonic number increases. This brings up an ambiguity regarding
the description of harmonics. The terminology used here is to refer to n = 1 as the
fundamental or first harmonic, n = 2 as the second harmonic, and so on. However,
one can find many technical works wherein n = 1 is referred to as the fundamental,
n = 2 is the first harmonic, and n = 3 is the second harmonic. This usage is consistent
with terminology for music, but we shall avoid it.
Simpler expressions for the range dependence of the amplitudes may be obtained
if we restrict our attention to the region close to the source at x = 0. Application to
Eq. (13.2.13) of the series approximation of Bessel functions gives
9 I.E. Fubini-Ghiron. “Anomalies in acoustic wave propagation of large amplitudes,” Alta Freq. Vol.
4 (1935) pp. 530–581.

1
n=1
0.75
Pn /( 0 c 2 )
0.50 n=4
n=3 n=2
0.25
0
0 0.2 0.4 0.6 0.8 1
x/x shock
Fig. 13.8 Plot of the Fourier series coefficients for the signal generated by a sinusoidal excitation at
x = 0. The distance xshock is the location where a shock forms, and the analysis is not valid beyond
that distance
2
P1 1 x
= 1 − + ...
ρ0 c ε
2 8 xshock 2 (13.2.14)
P2 1 x P3 3 x
= + ..., = + ...
ρ0 c2 ε 2 xshock ρ0 c2 ε 8 xshock
This confirms quantitatively the trend that is evident in Fig. 13.8 that the primary
effect near the source, x = 0, is growth of the second harmonic linearly with prop-
agation distance. The depletion of the fundamental, as well as the growth of third
harmonic, depends quadratically on the propagation distance. All of these effects are
weak until x is a substantial fraction of xshock .
In view of the fact that we could have obtained the data in Fig. 13.8 from FFT
analysis, as we did in Example 13.3, one might think that Fubini-Ghiron’s devel-
opment is important only from a historical perspective. This is far from the truth.
For one, many evaluations would be required before we realized that the harmonic
amplitudes depend on only two parameters. In addition, the availability of a formula
for the amplitudes allows us to delve further into basic properties. One of particular
interest is energy conservation.
Consider a temporally periodic wave in the region between two planes at x1 and
x2 , both of which are less than xsh . As a consequence of periodicity of the pressure and
particle velocity, the sum of the kinetic energy and potential energy per unit volume
at any location x is the same at the end of any period as it was at the beginning. Thus,
the total energy in the region between the two planes is conserved over any period.
It follows that the average energy transported into this region at x1 in an interval of
a period must equal the average energy that flows out of this region at x2 . The cross
sectional areas at x1 and x2 are identical and the locations are arbitrary, so it must be
that the time-averaged intensity in the propagation direction is constant.
The time-averaged intensity is the same basic expression as it is in linear theory,
t0 +T /2
1
(Ix )av = pvdt (13.2.15)
T t0 −T /2
t0 is arbitrary. The order of magnitude of p and v is ε, so the intensity is

where
O ε2 . However, as we proceed some higher-order terms might arise. To assure that
we consistently account for all such terms, we shall employ the small-signal approx-
imations, but retain the quadratic terms in the relation between v and p. Equation
(13.1.35) describes this relation, so we have
2
t0 +T /2
c p 1 p
(Ix )av = p − β dt (13.2.16)
T t0 −T /2 ρ0 c 2 2 ρ0 c2
The series is defined in terms of the delayed time t , Correspondingly, the intensity
may be computed by integrating over T /2 < t < 3T /2. (This is the first full interval
centered on an instant when p = 0.) This changes the preceding expression to
2 3
3T /2
ρ0 c3 p 1 p
(Ix )av = − β dt (13.2.17)
T T /2 ρ0 c2 2 ρ0 c2

The pressure is an odd function with respect to t = T , which means that p t =
3
−p 2T − t . Therefore p t also is an odd function. This means that its contri-
bution to the integral from T /2 < t < T will be the negative of its contribution
from T < t < 3T /2. The consequence is that the contribution of p3 vanishes. Thus
the time-averaged intensity is given by the same expression as it is in linear theory.
Periodicity permits adjusting the interval to 0 < t < T , so that
T
1
(Ix )av = p2 dt (13.2.18)
T ρ0 c 0
We wish to prove that the Riemann solution for any excitation leads to a time-
averaged intensity that is independent of x. Doing so is much easier if we limit
consideration to weakly nonlinear signals. In that case the pressure is related to the
boundary velocity by p (τ ) = ρ0 c2 εV (τ ), and τ is related to t by Eq. (13.2.4). The
latter relation is used to convert the integration variable from t to τ , which leads to
T $ %
1 x
(Ix )av = p2 d τ − βεV (τ )
T ρ0 c 0 c
$ %
ρ0 c3 2 T x
= ε V (τ )2 1 − βεV̇ (τ ) dτ
T 0 c
T T
ρ0 c 2
3
x
= ε V (τ )2 dτ − βεV (τ )3 (13.2.19)
T 0 3c 0
Because V (τ ) is periodic, the cubic term vanishes. Furthermore, the integrand

depends only on the nature of V (τ ), so it is independent of x. This proves that
(Ix )av is independent of x, being

ρ0 c3 2 T
(Ix )av = ε V (τ )2 dτ (13.2.20)
T 0
An alternative form of this expression results from Parseval’s theorem, Eq. (1.4.12),
which describes the mean-squared pressure corresponding to a Fourier series. The
fact that the integration is performed over a period of the τ variable, not time t, is irrel-
evant. Thus, the time-averaged intensity is related to the Fourier series coefficients
of V (τ ) by
#∞
1
(Ix )av = ρ0 c3 ε2 |Vn |2 (13.2.21)
2 n=1
Because the derivation wasrestricted

to weakly nonlinear waves, this expression
actually only is accurate to O ε2 . An analysis that does not invoke the weak non-
linearity would confirm that the Riemann solution describes a signal in which power
flows constantly to the location where a shock forms, regardless of the magnitude of
the signal.
13.2.2 Shock Formation
We have seen three manifestations of the occurrence of a shock: intersection of

characteristic lines, multivaluedness of the function whose roots give the pressure,
and vertical tangency of a waveform or spatial profile. Any of these three features may
be used as the basis for a quantitative analysis of the distance where a shock forms.
The result from each is the same, but the differences in the analysis demonstrate
different techniques for analyzing other systems.
We begin with an excitation at x = 0 that is a piecewise linear waveform like the
one in Example 13.1. Whereas we previously tracked the propagation of the phases
at which p is maximum and zero, we now consider signals that depart the boundary at
two arbitrary instants τ1 and τ2 > τ1 ; the corresponding pressures are p1 and p2 . The
phase speeds associated with these signals are c̃1 + v1 and c̃2 + v2 . The equations
for the two characteristics are
x x
τ1 = t − , τ2 = t − (13.2.22)
c̃1 + v1 c̃2 + v2
At the intersection of these lines, the values of x and t are the same. Because τ2 > τ1 ,
such an intersection can occur only if the slope dx/dt = 1/(c̃2 + βv2 ) of the τ2
characteristic is less than the slope 1/(c̃1 + βv1 ) of the τ1 characteristic. We have
seen that both c̃ and v increase monotonically with increasing p. It follows that the
characteristic lines will intersect only if p2 > p1 . This is another way of saying that
a shock forms because a later high-pressure phase catches up to an earlier lower
pressure phase. To solve for x, we take the difference of the two equations, which
leads to
(c̃1 + v1 ) (c̃2 + v2 ) (τ2 − τ2 )
x= (13.2.23)
(c̃2 + v2 ) − (c̃1 + v1 )
The shock formation distance, xshock , in general is the minimum value of x at

which a shock occurs at any instant. Rather than solving this relation for a variety
of τ1 , τ2 pairs, we can identify directly which pair of characteristics intersect at the
smallest x. Specifically, the τ2 characteristic should be the one for which c̃2 + v2 is
the largest value, and the τ1 characteristic should be the one for which c̃1 + v1 is the
smallest. Equation (13.1.39) is the result for a simple triangle waveform because in
that case, p2 is the maximum and p1 = 0.
Equation (13.2.23) applies if p (τ ) at x = 0 is piecewise linear. If it is a contin-
uously differentiable function, it is more useful to consider two adjacent instants
by letting τ1 = τ and τ2 = τ + dτ . Then, (c̃2 + v2 ) − (c̃1 + v1 ) becomes d (c̃ + v).
Furthermore, c̃ + v is a function of p, and p is a function of x and τ . These observa-
tions lead to identification of the condition for formation of a shock,
(c̃ + v)2
x= (13.2.24)
d ∂p
(c̃ + v)
dp ∂τ
We know that c̃ + v increases monotonically with increasing p, so that d (c̃ + v) /

dp > 0. Therefore, we conclude that a shock occurs in the portion of the waveform
for which p increases with increasing t.
Equation (13.2.24) is valid for a signal at any amplitude (within the limits of the
assumptions of adiabatic propagation in an inviscid fluid). In acoustic applications,
the acoustic Mach number ε is small, so Eq. (13.1.36) may be used to describe the
phase speed. Doing so gives

p
c 1+β
ρ c2
x= 0 (13.2.25)
d p
β
dτ ρ0 c2

In the small-signal approximation p/ ρ0 c2 = v/c = εV (τ ). Thus, when terms
whose order of magnitude is ε or smaller are dropped, the preceding becomes a
simple expression for x. This value depends on the value of τ . The shock formation
distance is defined to be the smallest x at which characteristics intersect because
we must assume that the basic formulation ceases to be valid in a region where a
shock exists at any instant. The minimum distance, xshock , corresponds to the positive
maximum of dV /dτ . Thus,
c
xshock = (13.2.26)
dV
βε max
dt
This expression confirms that the shock forms at the phase of the waveform that
changes most rapidly. It also tells us that the distance at which a shock first forms is
inversely proportional to the magnitude of the excitation.
Our second encounter with shock formation occurred when we developed a numer-
ical procedure to evaluate the pressure. This entailed finding the value q ≡ p/ ρ0 c2
that satisfies G (q, x, t) = 0. Figure 13.7 describes this function at x < xshock , in
which case there is one root, and at x > xshock , in which case there are three roots.
As x is increased from the first value, the slope of G vs. q at the root decreases. The
transition from one root to three is marked by a horizontal tangent at the root. This
suggests that the condition leading to x shock is marked by
∂
G (q, x, t) = 0 (13.2.27)
∂q
This derivative is given by Eq. (13.1.47), so we set

dV x β
1−ε =0 (13.2.28)
dτ τ =t− x/c c (1 + βq)2
1+βq
Because the order of magnitude of q is ε, it is acceptable to approximate the denomi-

nator terms as one. The smallest value of x at which this condition occurs corresponds
to the largest value of dV /dτ , regardless of when it occurs. The resulting expression
for xshock is the same as Eq. (13.2.26).
Another way in which we may locate a shock is by explicitly searching for a
vertical tangency of the waveform or spatial profile. This property is recognizable
by the fact that the interval in space for a profile, or in time for a waveform, between
the characteristics associated with τ and τ + dτ decreases as they come closer to
intersecting. We shall pursue this analysis for a waveform and leave the corresponding
analysis of a profile for Exercise 13.13. The first step is to derive an expression for
∂p/∂t. The substitution p = ρ0 c2 q somewhat simplifies differentiation of the small-
signal approximation in Eq. (13.1.43), which gives

∂q dV x/c ∂q
≡ε 1+ β (13.2.29)
∂t dτ τ =t− x/c (1 + βq)2 ∂t
1+βq
from which we find that

⎡ ⎤

∂q dV
⎣ x/c βε dV − 1⎦ = −ε (13.2.30)
(1 + βq) 2 dτ τ =t− x/c ∂τ dτ τ =t− x/c
1+βq 1+βq
For a vertical tangency, we seek the condition that leads to ∂q/∂t being infinite. This
occurs if the bracketed term is zero. The minimum x at which this condition occurs
corresponds to the maximum value of dV /dt, so a vertical tangent first occurs at
x (1 + βq)2
= (13.2.31)
c dV
βε max
dτ
Dropping βq in comparison with 1 yields Eq. (13.2.26).

Identification of the location where a shock forms is possible by three alternative
methods because of the manner in which the Riemann solution is posed. For other
systems, these approaches might not be equally viable. Indeed, if the solution is
the result of a numerical analysis, the only suitable approach might be a numerical
evaluation of the slope of a waveform or profile.
The value of xshock depends on the nature of the input function V (τ ). The typical
concern is a sine function, V (τ ) = sin (ωτ − φ). In that case, max (dV /dτ ) = ω,
so that
1
(xshock )sine = (13.2.32)
εβk
For reference values, a harmonic excitation at 1 kHz in air at 140 dB//20µPa cor-
responds to ε = 0.0020 and xshock = 23 m, while a 200 dB//1 µPa signal at 1 kHz
underwater corresponds to ε = 6.5 10−5 and xshock = 1.04 km.10 These values may
be scaled for
other
signal amplitudes and frequencies, so that a 1 MHzsignal in water
at ε = 6.5 10−5 shocks at 1.04 m, while the same signal at ε = 6.5 10−6 forms a
shock at 10.4 m.
13.2.3 Propagation of Weak Shocks
Propagation of a wave beyond the location where a shock first forms causes its
spatial profile or waveform to develop an extremely steep slope over a very small
interval. This condition is depicted in Fig. 13.9. Within this interval, shear stresses
grow, which means the some accounting must be given to the effects of viscosity
and heat conduction. Model field equations that account for the role of viscosity and
other nonideal effects would be much more difficult to derive and solve. Further-
more, the specific details of the small-scale region within which the signal changes
drastically might be of little interest if we can find a simpler way to describe how this
phenomenon affects the overall propagation features. The formulation that meets this
objective replaces the actual velocity dependence by a model in which the velocity
10 A 200 dB signal in water has an amplitude that is 1.4 atm. Hence, generating such a signal without
causing cavitation requires that the depth be sufficiently large that the hydrostatic pressure exceeds
the amplitude.
Actual Model
x dis (t) t>t 
x<x  tdis (x)
p p
x t
p p
x>x  t< t 
Profile at t Waveform at x
Fig. 13.9 Actual pressure profile and waveform at a shock and the idealized model. The latter
replaces extremely steep gradients and high accelerations with discontinuous changes in space and
time
changes discontinuously. Figure 13.9 shows that the actual pressure profile changes
over a distance that is much less than a wavelength in space. It also shows that the
actual waveform changes over an interval that is much shorter than a period. If we
do not closely inspect the signal, the model of a discontinuous change is a good
approximation. A discontinuity of the pressure in either space or time is an artifice,
but one implies the other. This is so because Euler’s equation and the equation of
state require that ∂p/∂t is infinite if ∇p is infinite.
The term “shock” is synonymous with discontinuity. We have already used xshock
to denote the distance at which the wave first develops a vertical tangency, so xdis (t)
will denote the location of the discontinuity in a spatial profile at an arbitrary instant.
For a wave propagating in the sense of increasing x, it must be that xdis ≥ xshock .
Similarly, tdis (x) is the instant at which a shock is observed in the waveform at an
arbitrary location. Other parameters are described in Fig. 13.9. The position infini-
−
tesimally behind the discontinuity is denoted as xdis and the corresponding pressure
− +
is p , whereas the position infinitesimally ahead of the discontinuity is xdis and the
+
corresponding pressure is p . A minor confusion results from the usage of + and
− to denote when and where a feature occurs, rather than relative size. Specifically,
we have p− > p+ , t − > t + , and x − < x + . In terms of the characteristic variable,
because p+ is ahead of p− in space, it must be that p+ departed from the boundary
at an earlier time, so that τ − > τ + . Multiple discontinuities may exist, but each is
treated individually.
Rankine–Hugoniot Jump Conditions
If the model of a discontinuous pressure change at a shock is to be useful, we must
determine how the discontinuity moves in space. This analysis modifies the derivation
of the conservation principles to include a discontinuous change of the state variables.
The domain we consider is a control volume contained between two fixed locations,
x1 < xdis < x2 . Figure 13.10 shows this control volume at two instants t and t + dt.
The position of the discontinuity is xdis at each instant. The field is independent of
xdis(t) xdis(t+dt)
dsin dsout
x1 x1
x2 x2
(a) Time t (b) Time t + dt
Fig. 13.10 A stationary control volume surrounding a moving shock
position transverse to the propagation distance, so the cross section may be considered
to be a unit area.
In Fig. 13.10(a), a chunk of fluid is situated to the left of the control volume.
This chunk contains the fluid that will enter the control volume during the interval
from t to t + dt. The width of this chunk is dsin . Because this width is infinitesimal,
the velocity of all particles contained in it is essentially v (x1 , t). By definition, the
left face of this chunk arrives at x1 in an elapsed time of dt, so dsin = v (x1 , t) dt. In
Fig. 13.10(b), the chunk of fluid to the right of the control consists of the particles that
exited in the interval from t to t + dt. As a consequence of dsout being infinitesimal,
the velocity of these particles is essentially v (x2 , t). Hence, dsout = v (x2 , t) dt.
The principles we shall consider are conservation of mass, momentum and
impulse, and energy and work. The formulation of each balance principle is essen-
tially the same. Let H denote the quantity that is the subject of the principle. Thus, the
mass per unit volume is H = ρ, the momentum per unit volume is H = ρv, and the
energy per unit volume is H = (1/2) ρv 2 + U , where U is the potential energy per
unit volume stored within the fluid when it is compressed and expanded. Equation
(4.5.7) describes U in terms of the pressure and the equation of state.
There is an external agent F that changes H. For momentum, it is the impulse of
the pressure acting on the control volume, which is positive for p& (x1 ) and negative '
for p (x2 ). The impulse is the net force times dt, so we define F = p (x1 ) − p (x2 ) dt
corresponding to a unit cross section. The external agent changing the energy is the
net work done by the pressure at the two ends. This& work may be evaluated as'the
instantaneous power multiplied by dt, so we set F = p (x1 ) v (x1 ) − p (x2 ) v (x2 ) dt.
Mass is conserved, so F = 0 for conservation of mass.
The description of the contributions from the chunks of fluid that enter and leave
the control volume is based on their infinitesimal extent. According to the central limit
theorem, H (x, t) within each chunk is the same as the value at the adjacent end of the
control volume. Hence, H (x1 , t) dsin enters the control volume and H (x1 , t) dsout
exits. We also must account for the amount of the H quantity contained within the
control volume at instant t, which is obtained by integrating H (x, t) over x1 < x <
x2 . Let I (t) denote this amount.
The existence of a shock means that H is not a continuous function of x, so the
integration to describe I must be done in a piecewise manner from x1 to xdis and then
from x dis to x2 . The positions to the left and right of the discontinuity are x − and x + ,
so the integration over the control volume at time t is described by
x2 x− x2
I (t) ≡ H (x, t) dx = H (x, t) dx + H (x, t) dx (13.2.33)
x1 x1 x+
We could follow similar arguments to describe the integral at time t + dt, with the
shock situated at xdis (t + dt). A simpler procedure employs Leibnitz’ rule for dif-
ferentiating an integral with a variable limit. The difference between x − and x + is
infinitesimal, so dx − /dt = dx + /dt = dxdis /dt ≡ vdis . The result is that
d
I (t + dt) = I (t) + dt I (t)
( dt
) x− ∂ )x ∂
= I (t) + x1 H (x, t) dx + x+2 H (x, t) dx (13.2.34)
∂t ∂t
& '*
+ vdis (t) H x − , t − H x + , t dt
Each balance principle states that I (t + dt) must equal I (t) plus the net amount
of H that enters the control volume plus the increment due to the F quantity, that is,
I (t + dt) = I (t) + H (x1 , t) v (x1 , t) dt − H (x2 , t) v (x2 , t) dt + Fdt (13.2.35)
This relation applies in any situation. Substitution of the representations of I at t and

t + dt into the conservation equation reduces it to
) x− ∂ )x ∂ & '
H (x, t) dx + x+2 H (x, t) dx + vdis (t) H x − , t − H x + , t
x1
∂t ∂t
= H (x1 , t) v (x1 , t) − H (x2 , t) v (x2 , t) + F
(13.2.36)
The last step is to shrink the control volume by letting x1 approach xdis from below,
x1 → x − , and letting x2 approach xdis from above, x2 → x + . This operation reduces
the integrals to zero. In addition, because x1 and x2 now are at the shock, we have
−
v (x1 , t) → vdis
and v (x2 , t) → vdis . To simplify the notation, let H ≡ H x− , t
and H+ ≡ H x + , t . A minor rearrangement of terms lead to a general description
of each balance principle as

H+ v + − vdis = H− v − − vdis + F (13.2.37)
Replacing the symbols H and F with their definition for each balance principle
leads to the Rankine–Hugoniot (jump) conditions. The mass equation is

ρ+ v + − vdis = ρ− v − − vdis (13.2.38)
The momentum equation is

ρ+ v + v + − vdis + p+ = ρ− v − v − − vdis + p− (13.2.39)
The energy condition is

1 + + 2 +
+ + + 1 − − 2 −
ρ v +U v − vdis + p v = ρ v +U
2 − 2 (13.2.40)
− −
v − vdis + p v
These relations are general and therefore apply to nonacoustic applications such as
blast waves and combustion in ramjet engines, as well as shocks formed by supersonic
aircraft.
Our objective is to determine where the discontinuity is located and how fast it is
moving. The analysis is simplified if we limit consideration to weak nonlinearity, in
which case the pressure on either side of the shock is much less than the bulk modulus.
Nevertheless, the operations are intricate, see for example Pierce’s treatment,11 We
shall only take an overview of the analysis. A key step is to form a linear combination
of the Rankine–Hugoniot equations in order to obtain an expression for the enthalpy
per unit mass, h = (U + pv) /ρ. This quantity is important because the differential
increment of entropy per unit mass in a process is relatedto thechange of h according
to TdS = dh − (dp) /ρ. The order of magnitude of p/ ρ0 c2 is taken to be O (ε).
Simplification of the recast Rankine-Hugoniot equations based on keeping thelargest

terms eventually leads to recognition that the order of magnitude of dS is O ε3 . In
other words, entropy is essentially constant across a weak shock.
The implication of constancy of entropy is that the nonlinear acoustic relations
derived from the Riemann solution remain valid everywhere other than xdis . This
property leads to a relation between the velocity of the shock and state of the fluid
behind and ahead of the shock. Let vav = v − + v + /2 and c̃av = c̃ (vav ). The weakly
nonlinear properties lead to
p− p+ p− + p+ 1 & − '
v− = , v+ = , vav = , c̃av = c̃ v + c̃ v +
ρ0 c ρ0 c 2ρ0 c 2
(13.2.41)
The velocity of the discontinuity is found to be
vdis = c̃av + vav (13.2.42)
11 Pierce, ibid., pp. 576–577.

These relations describe a situation in which the shock is propagating in the positive
x direction. Because a shock is characterized by a forward-leaning of the spatial
profile, p− will be greater than p+ . Thus, the shock’s location will move slower than
the phase speed behind the shock and faster than the phase speed ahead. If it should
happen that p+ = −p− , then c̃av will be c. In other words, the shock will move at the
linear speed of sound.
Equal-Area Rule
It still remains to determine how xdis depends on time. One approach solves a differen-
tial equation based on setting vdis = dxdis /dt in Eq. (13.2.42). Figure 13.11 describes
the parameters required to carry out this analysis. In it, the multivalued pressure pro-
file is the Riemann solution beyond the shock formation distance. Suppose the value
of xdis is known at instant t. We can use the method in Sect. 13.1.5 to determine the
characteristics τ1 and τ3 , and the associated pressures corresponding to the values
of xdis and t. (The pressure at τ2 is irrelevant to this discussion.) The unshocked
portion of the waveform adjacent to τ3 is farther advanced than xdis , so p+ = p (τ3 ).
Similarly, the unshocked portion of the waveform adjacent to τ1 is behind the shock,
so p− = p (τ1 ). Because τ1 and τ3 are functions of xdis and t, so too are p− and p+ .
This means that Eqs. (13.2.41) and (13.2.42) combine to give vav as a function of
xdis and t. Therefore, Eq. (13.2.42) is a statement that dxdis /dt = f (x dis , t)—it is a
differential equation. The initial condition is the value of xdis at the instant when the
profile first develops a vertical tangency.
xdis(t)
p= p(  )
p(  )
x
p+=p(  )
Fig. 13.11 Relationship of the parameters for a shock to the pressure waveform obtained from the
Reimann solution
Although this procedure is feasible, numerical methods would be required to

solve the differential equation, and the need to find multiple roots of the Riemann
solution at each integration step would require considerable effort. Furthermore, the
method provides little understanding of the underlying process. Fortunately, there
is an alternative that provides a graphical interpretation and determines xdis directly.
Figure 13.12 shows a multivalued profile obtained from the Riemann solution. The
difference between it and the previous figure is that the portions of this waveform that
are advanced beyond xdis and retarded behind xdis are shaded. The ultimate result will
be that xdis is such that the total shaded area, positive for the region where x > xdis
x( ,t)
x dis (t) p()
dp=d

p = p(  ) >0 p( )

p(  )= 0
x
p =p( )
<0
Fig. 13.12 Spatial pressure profile obtained by evaluation of the Riemann solution beyond the
location where a shock begins to form. This profile is a multivalued function of x for characteristic
values in the range τ − < τ < τ + . The position of the area of the profile at x > xdis is positive,
while the area for x < xdis is negative
and negative for x < xdis , is zero. We shall prove this property, because doing so
leads to a method by which xdis can be determined at any instant.
If t is specified, then p and x are unique functions of the characteristic variable τ .
The values of τ corresponding to p− and p+ are τ − and τ + , respectively. The lower
portion of the profile after the discontinuity is farther advanced than the discontinuity,
so p+ is the smallest value of p obtained from the Riemann solution when x = xdis
and t is the time for the profile. Similarly, the upper portion of the profile preceding
the discontinuity is less advanced than the discontinuity, so p− is the largest value
of p obtained from the Riemann solution at x = xdis when t is the designated time.
If follows that the range of characteristic variables associated with the shaded area
is τ + < τ < τ − . The small-signal approximation of the equation for a characteristic
tells us that the location of a specific p (τ ) at instant t is

β
x (τ , t) = (t − τ ) (c + βv) = (t − τ ) c + p (τ ) (13.2.43)
ρ0 c
As suggested by Fig. 13.12, the total area A is best evaluated by accumulating

(x − xdis ) dp, so that
p−
A (t) = [x (τ , t) − xdis (t)] dp (13.2.44)
p+
If A (t) is constant, then it should be that dA/dt = 0. Leibnitz’ rule is used to

differentiate the integral,
p−
dA d & ' ∂p−
= x (τ , t) − vdis dp + x τ − , t − xdis (13.2.45)
dt p+ dt ∂t
& ' ∂p+
− x τ + , t − xdis
∂t

The definitions of τ − and τ + are that x τ − , t = x τ + , t = xdis (t), so both terms
obtained from the integration limits vanish. We know from the boundary condition
that p (τ ) = ρ0 cV (τ ), so p is a single-valued function of τ . Accordingly, we change
the integration variable from p to τ , so that dp = (∂p/∂τ ) dτ . Also, differentiation
of Eq. (13.2.43) with respect to t, with τ held fixed, gives
dx β
=c+ p (τ ) (13.2.46)
dt ρ0 c
The result is that

τ−
dA βp (τ ) ∂
= c+ − vdis p (τ ) dτ
dt τ+ ρ0 c ∂τ
τ =t −
β
2
= (c − vdis ) p (τ ) + p (τ ) (13.2.47)
2ρ0 c τ =t +
A rearrangement of terms lead to

dA − p− − p+
= p − p+ c − vsh + (13.2.48)
dt 2ρ0 c
The term in the bracket is zero according to Eq. (13.2.42), which was derived from
the Rankine-Hugoniot relations. Thus, we have proven that dA/dt = 0. This means
that the total area A is constant at its value at the instant when the shock first formed,
which is A = 0. Because the total area bounded by a graph of p (x.t) versus x and
the line x = xdis is zero, then the area to the left of the discontinuity must equal the
area to the right. This is the equal-area rule.
Evaluation of Spatial Profiles
Enforcement of the equal-area rule leads to a method for determining the location of
the discontinuity at any instant. There is a common situation in which little analysis is
required. Let τ0 be a characteristic along which the pressure is zero. Suppose that p (τ )
is a continuous odd function relative to τ0 , that is, p (τ0 + δ) = −p (τ0 − δ). It follows
that p+ = −p− regardless of t, so that vdis = c and xdis (t) = x (τ0 , t) = c (t − τ0 ).
The application of the equal-area rule when p (τ ) is an arbitrary function begins
by selecting a trial value for the discontinuity’s location; let us denote this choice as
xtr . Obviously, x tr must be somewhere in the interval where the profile is multivalued.
The area A created in this manner will depend on xtr , as well as the value of t for the
profile. The basic notion is to identify which xtr in a range of values yields A = 0.
To do so, we need a procedure for evaluating A. To that end, the integration variable
in Eq. (13.2.44) is changed from p to τ , which leads to
τ−
∂
A (xtr , t) = [x (τ , t) − xtr ] p (τ ) dτ (13.2.49)
τ+ ∂τ
This expression can be specialized to the situation where the velocity at x = 0 is

εcV (t). Then, the weakly nonlinear pressure is ρ0 c2 εV (τ ). In view of the position
dependence in Eq. (13.2.43), the area function is
τ−
A (xtr , t) = ρ0 c2 ε {c (t − τ ) [1 + βεV (τ )] − xtr } V̇ (τ ) dτ (13.2.50)
τ+
Several terms are exact differentials. Integration by parts is used to handle the terms
that contain τ as a factor. The result is
( +τ −
1
A (xtr , t) = ρ0 c2 ε c (t − τ − xtr ) V (τ ) + βεc (t − τ ) V (τ )2
2 τ+ (13.2.51)
) τ − 1
+ρ0 c2 ε τ + c V (τ ) + βεV (τ )2 dτ
2
A determination of xdis for the spatial profile at a specified t may be performed

by following a set procedure. It begins by letting xtr be the smallest x in the multi-
valued range, solving for the smallest and largest roots, τ + and τ − , respectively, of
Eq. (13.2.43) when x = xtr and t is the instant of interest, then evaluating A for that
choice of xtr . Progressively incrementing xtr will reduce the value of A. The value of
xdis lies between the values of xtr for which A changes sign.
Evaluation of Waveforms
We have established that the equal-area rule can be used to determine where a discon-
tinuity occurs in a pressure profile at an arbitrary instant. It also can be used to deter-
mine the instant at which the discontinuity occurs in the waveform at a designated
x. The multivalued waveform in Fig. 13.13 is the result obtained directly from the
Riemann solution. The instant at which the discontinuity is observed at this location
is tdis (x), which leads to a total area A (x, tdis ). The interval in which the waveform
is multivalued corresponds to p+ < p < p− . According to the discussion of profiles,
the value of τ is a unique function of p, so it is best to us dA = (tdis − t) dp.
Therefore,
p−
A (x, tdis ) = [tdis (x) − t (τ , x)] dp (13.2.52)
p+
To verify that the equal-area rule applies to waveforms, we evaluate dA /dx.
Application of Leibnitz’ rule to the preceding gives
Fig. 13.13 Construction of tdis(x)

the area bounded by a
multivalued waveform and a t( ,x)
the location of the
discontinuity at an arbitrary dp=d p()
location x  ‘
p( )
>0 p =p(  )
‘
p =p( ) <0
p+
d d ∂ & ' ∂ −
A (x, tdis ) = tdis (x) − t (τ , x) dp + tdis (x) − t τ − , x p
dx p− dx ∂x ∂x
& ' ∂ +
− tdis (x) − t τ + , x p (13.2.53)
∂x

By definition, t τ − , x = t τ + , x = tdis (x), so the terms from the integration limits
vanish. Also, (d/dx) tdis (x) ≡ 1/vdis (x). To continue the analysis, it is convenient to
rewrite Eq. (13.2.43), which is the small-signal approximation of the equation for a
characteristic, as
x
t (τ , x) = τ + (13.2.54)
c + βv (τ )
Differentiation of this relation with respect to x, with τ held fixed, yields
∂ 1
t (τ , x) = (13.2.55)
∂x c + βv (τ )
The result is that the spatial rate of change of A becomes

p−
d 1 1
A (x) = − dp (13.2.56)
dx p+ vdis (x) c + βv (τ )
The analysis is restricted to weakly nonlinear signals, so v = p/ (ρ0 c) and c v (τ ).

These approximations lead to
p−
d c + βp/ (ρ0 c) − vdis (x)
A (x) = + · · · dp
dx p+ cvdis (x)
p−
1 1 p
= c+ β − vdis (x) p (13.2.57)
cvdis (x) 2 ρ0 c p+
The bracketed term is identically zero according to Eq. (13.2.42), so we have proven
that the gradient of A (x) is identically zero. Consequently, A (x) is constant at the
value when the discontinuity began to form, so A (x) = 0. Thus, we have proven
that the equal-area rule applies to waveforms, as well as spatial profiles.
We may use the equal-area rule to determine when the discontinuity occurs in
the waveform at a designated x. If the excitation function V (τ ) is odd relative to a
phase τ0 at which V (τ ) = 0, then symmetry places the shock at tdis = τ0 + x/c. The
procedure when V (τ ) is arbitrary is a modification of the procedure for profiles.
To carry out the analysis, we need an expression for A in terms of the trial
instant ttr for the discontinuity. To that end, we use the equation for a charac-
teristic, Eq. (13.2.54), to replace t in Eq. (13.2.52) and then set v = εcV (τ ) and
p = ερ0 c2 V (τ ), and change the integration variable to τ . These operations give
τ−
x dV (τ )
A (x, ttr ) = ερ0 c 2
ttr − τ − dτ (13.2.58)
τ+ c + βεcV (τ ) dτ
A binomial series simplifies the fraction, so that

τ− $ % dV (τ )
x
A (x, ttr ) ≈ ερ0 c2 ttr − τ − (1 − βεV ) dτ (13.2.59)
τ+ c dτ
We apply integration by parts to the term for which τ is a coefficient. All other terms
are perfect differentials, so that
, τ − τ − -
x 1 x

A (x, ttr ) = ρ0 c ε
2
ttr − τ − V (τ ) + βε V (τ )2 + V (τ ) dτ
c 2 c τ+ τ+
(13.2.60)
The value of tdis is the value of ttr that gives A = 0 when x is set. Carrying
out the search for this value of ttr requires determination of the values of τ + and
τ − corresponding to specified values of x and ttr . These values respectively are the
smallest and largest roots extracted from Eq. (13.2.54). The procedure for carrying
out the analysis is described in the next example.
EXAMPLE 13.4 In Example 13.2, the excitation at x = 0 is a harmonically

varying velocity starting at t = 0. Determine the waveforms at x/ (xshock )sine =
1.2, 1.5, 2, and 4.
Significance
A primary focus is development of a general computational algorithm for fitting
shocks into the Riemann solution. The resulting waveforms will shed light on the
trend for the signal to approach a sawtooth waveform.
Solution
The starting point for the analysis is the evaluation of the Riemann solution. The
direct computational method used to solve Example 13.2 may be used for this purpose
because there is no need to have the pressure data on a base of equally spaced t values.
The nondimensional shock formation distance is kxshock = 1/ (εβ), so we seek the
waveform at locations where kx = σkxsh = σ/ (βε). The acoustic Mach number is
ε = 0.002. The Riemann solution at a set of equally spaced ωτn values starting at
ωτ1 = 0 yields a set of qn ≡ p/ ρ0 c2 values corresponding to instants ωtn , where
qn = ε sin (ωτn ) h (ωτn )

σ 1 (1)
ωtn = ωτn +
βε 1 + βε sin (ωτn ) h (ωτn )
The evaluation of the Riemann solution yields a correlated set of ωτn , ωtn , and qn
data at fixed σ in which the constant interval for the ωτ values is 2π/N.
The Riemann solution beyond xshock is typified by the waveform at σ = 4 in Fig. 1.
Two different conditions are seen to exist. At the leading edge, which corresponds to
τ = 0, there is no prior signal. The multivalued waveform has no special symmetry
properties in this interval, so fitting a discontinuity will require an analysis of the
waveform area. Subsequent to arrival of the shock at the leading edge, the multivalued
intervals are centered on ωτ = 2jπ, which is the phase at which p = 0 and dp/dτ
is the maximum positive value. With respect to these instants, p is an odd function.
Thus, the shock will occur at the time values for which ω (tdis − x/c) = 2jπ.
tM 
1
N+
p/(0c2)
ttr  N/+

-1
kx kx+  kx+2 
t
Figure 1.
We begin by modifying the computed waveform in the periods after the first. Let
τJ be any instant for which p (τJ ) = 0. In view of the periodic nature of p (τ ), and
the fact that p (0) = 0, the second and subsequent discontinuities correspond to J =
N + 1, 2N + 1, .... Therefore, ωtdis = ωτJ + kx for these discontinuities. Now that
we know the instant at which a discontinuity occurs, we can correct the multivalued
pressure. Smaller τ values that occur later than ωtJ , and subsequent τ values that
occur earlier are physically impossible. These phases belong to the discontinuity.
This condition can occur for no more than N/2 points on either side of ωτJ . Thus,
the procedure is to set ωτJ−n = ωτJ , ωtJ−n = ωtdis , and pJ−n = 0 if ωτJ−n > ωτJ ,
and ωτJ+n = ωτJ , ωtJ−n = ωtdis , and pJ+n = 0 if ωτJ+n < ωτJ . These operations are
implemented for n = 1, ..., N/2.
Fitting the shock at the leading edge requires explicit examination of the waveform
area according to Eq. (13.2.60) for each of the Riemann data values. The excitation
function is V (τ ) = sin (ωτ ) h (ωt). Furthermore, the area of the waveform is situated
in p ≥ 0, so the lower limit in that expression corresponds to τ + = 0 and V (τ + ) = 0.
Thus, the waveform area corresponding to a trial instant ωttr for the discontinuity is
(
1 2
ωA (kx, ωttr ) = ρ0 c2 ε ωttr − ωτ − − kx sin ωτ − + βεkx sin ωτ −
& '. 2
+ 1 − cos ωτ −
(1)
The true value of ωtdis is obtained by setting ωttr = ωtn for a range of n and
then searching for the value of n that best fits the criterion that ωA (kx, ωtn ) = 0.
Explanation of the procedure is facilitated by considering Fig. 2, which zooms in on
the multivalued interval at the leading edge. The dots mark points on the waveform
that are the computed data. The first point, ωτ1 = 0, corresponds to the largest t in
the multivalued interval. The earliest time in the data set is ωtM . The value of M is
determined by seeking the index corresponding the minimum of the ωtn data.
 -
1
tM
p/(0c2)
n
0
ttr
1=0
-1
kx kx+ kx+2
t
Figure 2.
The search procedure entails setting ωttr equal to a computed ωtn value in the range
1 ≤ n ≤ M. By definition, ωτ − is the characteristic value on the upper part of the
multivalued segment when the time is ωttr . In Fig. 2, this point is the intersection of
the vertical line through point n and the upper branch. Because a constant increment
of the ωτ values leads to uneven spacing of the ωt values, this intersection will not
fall on a point that was computed along the upper branch. Therefore, the value of
ωτ − is not yet known, but evaluation of the area in Eq. (1) requires that we locate
this intersection. There are several options for this determination. The most accurate
one uses the method in Sect. 13.1.5 to determine the largest of the three values of p
given by the Riemann solution at kx and this ωttr . A simpler alternative interpolates
between data points on the upper branch on either side of ωttr . The simplest to
implement selects the computed point on the upper branch that is closest to ωtn . The
latter procedure will be adequate if the value of N is sufficiently large. Regardless
of which method is used, the concept is to use the value of ωτ − to compute ωA .
This is done beginning at n = 1, and ending at the most at n = M. The search may
be halted when the sign of A changes.
The resulting values of ωA in the full range of possible ωttr are shown in Fig. 3
for x = 1.2xsh . The value of A is zero at ωttr = 199.9464, so this is the value of ωtdis .
The computations used N = 512, which corresponds to (ωτ ) = 2π/512. This is a
reasonably fine resolution, but the nearly linear nature of the plotted curve suggests
that it would have been adequate to use a smaller N.
0.04
0.02
−0.02
199.94 199.96 199.98 200
ttr
Figure 3.
After ωtdis has been identified, the waveform may be corrected. The first step is to
identify the instant that is closest to ωtdis on the middle branch. We denote this instant
as ωtK . The index of points that lie on the middle branch is 1 < K < M. The points
in the multivalued interval that are delayed too much are adjusted by assigning them
to the bottom of the discontinuity, so that ωtn = ωtdis and pn = 0 if ωtn > ωtdis and
n < K. Data values that occur too early are shifted to the top of the discontinuity by
setting ωtn = ωtdis and pn = pK if ωtn < ωtdis and n > K.
These procedures yield a set of (ωt, p) data that is single-valued function of time.
Figure 4 is the result at the four specified locations. At 20% farther than the location
where the shock begins to occur, the discontinuity is barely noticeable. Nevertheless,
there is a visible similarity of the waveform to the teeth of a saw. Propagation to
farther distances enhances this appearance. Indeed, at σ = 4, there is no evidence of
a sinusoidal function. Another important aspect is the effect of the discontinuity on
the peak pressure. At σ = 1.2 and σ = 1.5, the maximum pressure in the Riemann
solution occurs slightly after each of the multivalued intervals. In contrast, at σ = 2
and at σ = 4 the maximum and minimum pressures in the Riemann solution occur
in the multivalued interval. Thus, fitting a discontinuity to those waveforms results
in reduction of the peak pressure. From this, we may deduce that the shock wave
dissipates energy.
1
0
 = 1.2
−1
1
0
p/(0c2) −1
 = 1.5
1
0
=2
−1
1
0
=4
−1
−0.5 0 0.5 1 1.5 2 2.5 3 3.5
(t-x/c)/(2)
Figure 4.
An interesting question is how the discontinuities propagate relative to the wave-

form. The waveform is an odd function relative to the second and subsequent dis-
continuities. These discontinuities propagate at the linear speed of sound, so their
relative spacing is maintained at cT , and the interval between their occurrence in a
waveform will be T . This is not true for the leading discontinuity because for it, p−
is positive and p+ = 0. The average pressure across this shock is pav = p− /2. The
Rankine-Hugoniot
relation,
Eq. (13.2.42), states that this discontinuity will propa-
gate at c̃ p− / (2ρ0 c) + p− / (2ρ0 c) > c. Therefore, the distance from the shock at
the leading edge to the next shock will decrease with increasing t. For the same
reason, the leading edge shock will precede arrival of the next shock by more than T .
Eventually, the value of p− progressively decreases as the discontinuity cuts off an
increasing portion of the multivalued waveform. In terms of nondimensional vari-
ables, the discontinuity’s arrival at a specified x is advanced by ωtdis − kx. Figure 5
is the result of computing this quantity over a broad range of σ values. Plotting the
time advancement as a ratio to 2π allows us to see it as a fraction of a period. The
rate of this advancement decreases with increasing distance due to the decrease of
p− . These features are examined more closely in the next section.
0
tdis− kx)/(2)
−0.1
−0.2
−0.3
1 2 3 4 5 6 7 8
  x/xshock
Figure 5.
Old-Age Behavior
The previous example demonstrates the drastic ways that a signal is affected by
the creation of shocks. These effects grow as a signal propagates well beyond the
location where a shock first forms. A square wave provides a convenient means
for exploring these phenomena because the equal-area rule can be enforced without
requiring integration to evaluate the waveform area. Although this is the only type
of waveform we will consider, the observations are widely applicable because of the
tendency of all waves to evolve to a sawtooth appearance.
The excitation is a particle velocity at x = 0 that is a periodic square wave begin-
ning at t = 0. The peak particle velocity is set at εc, with ε 1. The small-signal
approximation tells us that p = ρ0 cv, so the pressure also is a square wave, as shown
in Fig. 13.14, where the maximum value is p̂ = ρ0 c2 ε. The presence of discontinu-
ities in which the pressure increases tells us that the signal at x = 0 already contains
shocks.
p @ x=0
T/2 T/2
p^
t
-p^
Fig. 13.14 A periodic square wave generated at x = 0. The pressure is zero for t < 0
The positive pressures propagate faster and the negative pressures propagate
slower than the linear speed of sound. Because |ε| 1, the Riemann solution has the
property that straight segments of the waveform remain straight as the wave propa-
gates. Thus, according to the Riemann solution, the square lobes become trapezoids
that lean forward in time as the signal propagates. Figure 13.15 shows this waveform
at a propagation distance that is not too large. The phase speed of the maximum
positive pressures is c (1 + βε), whereas the phase speed at zero pressure is c. Thus,
Reimann solution Linear theory

p @ x<xold T/2 T/2
t t
p^
t
-p^ x/c t
Fig. 13.15 Comparison of the Riemann solution and linear theory for propagation of a periodic
square wave. The location is not too distant from the source
the time advancement t of the positive pressure plateaus relative to the arrival time
of the preceding p = 0 phases is
x x x
t = − ≈ βε (13.2.61)
c c (1 + βε) c
The phase speed of the negative pressure plateaus is c (1 − βε). To O(ε), these signals
are retarded in time relative to p = 0 by the same t. As part of the analysis, we will
identify the largest distance, labeled xold , for which this picture is appropriate.
According to the Riemann solution, p = 0 and ∂p/∂τ > 0 when t equals x/c plus
an integer multiple of the period. The multivalued portion of the Riemann solution
contains these instants. They are corrected by application of the equal-area rule.
The piecewise linear nature of the Riemann waveform simplifies this adjust-
ment. After the first cycle, the waveform in the multivalued intervals are odd func-
tions relative to the instant when p = 0. Figure 13.16 shows a vertical line through
p = 0 within these intervals. The waveform area for negative p is the negative of the
area for positive p. It follows that these lines define the time tdis of discontinuities.
The discontinuity values are p− = p̂ and p− = −p̂. The Rankine-Hugoniot relation
states that shocks propagate at the linear speed of sound plus the average of the par-
ticle velocities ahead of and behind the discontinuity. This construction in Fig. 13.16
is consistent with this principle.
p @ x<xold Reimann solution Equal-area rule
p^
t
t/2
-p^
x/c T/2 T/2
Fig. 13.16 Application of the equal-area rule to a periodic square wave. The propagation distance
is relatively small
The situation at the leading edge is different, because no signal precedes the
p = 0 phase. A vertical line that splits t leads to a waveform area that consists of
identical triangles to the left and right, so the discontinuity at the leading edge is
advanced by t/2. It follows that the leading edge arrives at position x when t =
x/c − t/2 = (x/c) (1 − βε/2). The corresponding propagation speed of this shock
is x/t = c/ (1 − βε/2) ≈ c (1 + βε/2). This too is consistent with the Rankine–
Hugoniot relations because it is the mean of the sound speeds for p− = p̂ and p+ = 0.
Increasing the propagation distance leads to an increase of t. At some location, the
Riemann solution predicts that the last pressure on a positive plateau occurs at the
instant of a discontinuity, and that the first pressure on a negative plateau also occurs
at that instant. This condition, which is characterized by t = T /2, is described by
Fig. 13.17. Only the leading edge is exempt from this behavior.
Beyond the location described by Fig. 13.17, the waveform is altered in a fun-
damentally different manner that is ongoing. This position marks the advent of the
old-age stage, so it is labeled as xold . Setting t = T /2 gives
Reimann solution Equal-area rule

p @ x=xold t=T/2
p^
t
t/2
-p^
x/c T/2
Fig. 13.17 Distortion of an initial square wave at the onset of old age. The time advancement of
the positive pressure equals half a period

cT cT ρ0 c2
xold = = (13.2.62)
2βε 2β p̂
In the present case of a periodic square wave, the old-age stage begins when
a plateau has advanced sufficiently that its entire occurrence falls in the interval
where the Riemann solution is multivalued. Beyond that occurrence, the discontinuity
intersects the Riemann solution on a sloping portion. A negative plateau also is cut off
because it is delayed into the multivalued interval. This is the situation in Fig. 13.18.
The equal-area rule maintains all discontinuities, other than the leading edge, at the
location where the Riemann solution gives p = 0. The consequence is that the peak
maximum and minimum pressure amplitudes are reduced. Any wave that propagates
beyond its shock formation distance will ultimately exhibit this phenomenon. It is
the hallmark of the old-age stage.

t
p @ x>xold pmax
T/2
p^
t
-p^
x/c T/2
pmax
Fig. 13.18 Waveform of an initially square wave at old-age. Distortion results in a decrease of the
peak pressure
An expression for the peak pressure in the old-age stage is obtained from similar
triangles formed by a discontinuity and the Riemann solution, specifically
pmax p̂
= (13.2.63)
T /2 t
Substitution of t from Eq. (13.2.61) and p̂ = ρ0 c2 ε leads to

ρ0 c2 cT xold
pmax = = p̂ , x > xold (13.2.64)
2β x x
The first form is remarkable because it indicates that the peak pressure is independent
of p̂. In other words, if all other parameters are held fixed, increasing the amplitude
of the input will not alter the pressure observed at a specified location in the old-age
region. Why is this so? If we increase the amplitude, the value of xold decreases.
Consequently, the signal enters old-age at a closer location, so x old /x decreases.
Thus, the increase in the input amplitude is offset by the greater old-age reduction.
This phenomenon is called saturation. It limits the loudness of the sound that can
be heard. To see why, suppose we select a location x. At very small values of p̂, this
location is much smaller than the value of xold given by Eq. (13.2.62). Increasing the
amplitude of the input proportionally raises the amplitude of the Riemann solution.
However, the increase also reduces the value of xold . Further increase eventually
reduces xold to x at the field point, which occurs when p̂ = ρ0 c2 (cT ) / (2βx). This
is the saturation pressure. Further increase of p̂ leads to x > xold , which means that
Eq. (13.2.64) now describes the peak pressure. As an example, consider a signal in
air for which 1/T = 1 kHz. The saturation amplitude at x = 100 m is 197 Pa. The
RMS pressure in one cycle of the sawtooth waveform is pmax /31/2 , so the sound
pressure level at saturation is 135 dB. The only way in which a louder sound at that
location can be obtained with a plane wave is to raise T , that is, lower the fundamental
frequency.
The reduction of peak pressure in the old-age stage brings to the fore the issue of
mechanical energy. Because we are dealing with simple plane waves, and the particle
velocity is p/ (ρ0 c) in the small-signal approximation, the time-averaged intensity
has the same form as it does for a linear signal whose waveform is arbitrary. For the
present purpose, we may use any period subsequent to the first, so that

1 x/c+(n+1)T
p (x, t)2
Iav (x) = dt (13.2.65)
T x/c+nT ρ0 c
For the initial waveform in Fig. 13.14, we have Iav (0) = p̂2 / (ρ0 c). The area under the
square of trapezoidal waveform in Fig. 13.15 may be found by adding the contribution
of the plateau and the linear decrease, which gives

2 T 1 p̂2 2 x
Iav (x) = p̂2 − t + p̂2 t = 1− , x < xold
ρ0 cT 2 3 ρ0 c 3 xold
(13.2.66)
In the region where old-age behavior has taken over, integration of the square of the
triangular waveform in Fig. 13.18 leads to

2 1 2 T 1 p̂2 xold 2
Iav (x) = p = , x > xold (13.2.67)
ρ0 cT 3 max 2 3 ρ0 c x
2
At xold , both expressions give Iav = p̂ / (3ρ0 c). We see that the time-averaged
intensity decreases with increasing x at all locations. This means that some energy is
lost, rather than flowing downstream. Before the onset of old-age, the loss is linear.
In old-age, the power flow decreases as the inverse square of distance, so the energy
loss is much more drastic.
The loss of energy is a consequence of the thermodynamic mechanisms captured

by the Rankine–Hugoniot relations. This effect arises even if nonideal effects are
negligible. However, dissipative effects like viscosity and molecular relaxation are
always present. Attenuation of harmonics composing a periodic signal typically
increases with increasing harmonic number, as will be seen in Sect. 13.3.2. This
effect is not necessarily large, depending on the properties of the fluid. However, it
combines with the old-age behavior, which reduces all harmonics. Ultimately, what
remains is a fundamental harmonic that has been reduced sufficiently to permit the
usage of linear acoustic theory.
One of the surprising attributes of shocks is that they affect a pulse differently
from a steady-state signal. We encountered this difference in the case of continuous
square wave excitation, where the discontinuity at the leading edge was unlike those
embedded in the subsequent response. To explore this further, we consider a pulse at
x = 0 that is the first cycle of the square wave, after which it is cut off. In Fig. 13.19,
the pulse has been propagating for a while, but it has not yet reached old-age.

p @ x<xold
t
p^ T/2
t/2
t
t/2
-p^
x/c T/2
t
Fig. 13.19 Early stage in the propagation of single period of a square wave
The Riemann solution for the pulse is the same as its prediction for the first period
of the continuous wave. As a consequence, the shock at the leading edge behaves in
the same way as the shock for the continuous wave. Termination of the waveform
after one cycle leads to p+ = −p̂ and p− = 0. The multivalued waveform at the
trailing edge is not what it was for a continuous wave. Rather, it is an image of the
waveform at the leading edge. Therefore, the retardation of the discontinuity at the
trailing edge is the same as the advancement at the leading edge. The consequence is
that the pulse duration is extended to x/c − t/2 < t < x/c + T + t/2, whereas
it is x/c < t < x/c + T in linear theory.
Further propagation eventually brings the pulse to the transition where old-age
behavior sets in. This occurs when the last pressure on the positive plateau catches up
to the discontinuity, which is the same instant as that at which the first pressure on the
negative plateau is delayed to the trailing discontinuity. This occurs when t = T ,
which leads to xold being twice as distant as the value in Eq. (13.2.62). Figure 13.20
shows the waveform beyond the old-age location.
p @ x=xold Reimann solution Equal-area rule

T/2 pmax
p^ t=T/2 pmax
t
-p^ t
x/c T/2 t T/2
Fig. 13.20 Distortion of a square wave pulse beyond the distance where old-age behavior begins
Application of the equal-area to locate the discontinuity in the old-age signal

would be a little more difficult than it was for the continuous wave because the
waveform area has a different shape on either side of the discontinuity. Nevertheless,
it is evident that the waveform preserves its sawtooth shape with an ever decreasing
amplitude, and that its duration increases progressively as the propagation distance
increases.
It is reasonable to ponder whether the old-age behavior of a square wave pulse
is representative of general trends. Evidence that it may be found in the results of
Example 13.4, where the waveform strongly resembles a sawtooth at a distance as
small as x = 2xshock . Indeed, if we set T = 2π/ω for the fundamental frequency, the
expression in Eq. (13.2.62) gives xold = π (xshock )sine . A comparison of the waveforms
at x = 2xshock and x = 4xshock in Fig. 3 of the aforementioned example shows that old-
age amplitude reduction has occurred at the larger distance. The old-age behavior
of any signal will be like that of the square wave, but the details of where it sets
in, where the discontinuities are situated, and how the amplitude decreases with
increasing distance will depend on the nature of the input function.
13.3 General Analytical Techniques
We began our study of linear acoustics with the d’Alembert solution. Like the Rie-
mann solution, it is obtained by analyzing the characteristics. To extend the scope of
linear systems that could be investigated, we invoked other methods to solve the wave
equation. In the same way, if we are to expand our exploration of nonlinear acoustic
waves, we will need an appropriate field equation. The nonlinear wave equation
we will derive contains no approximations beyond the assumption of an isentropic
process in an inviscid fluid. To the author’s knowledge, the first appearance of this
equation was in a text on aeroelasticity.12 The derivation parallels that in a classic
fluid mechanics treatise,13 and an alternative derivation appears in an acoustics text.14
After the field equation is derived, two general methods for solving it will be derived
for the case of plane waves.
12 R.L. Bisplinghoff, H. Ashley, and R.L. Halfman,Aeroelasticity, Addison Wesley (1955) Chap. 3
13 S.Goldstein, Lectures in Fluid Mechanics, Wiley-Interscience (1960) Chap. 4
14 D.T. Blackstock, Physical Acoustics, Wiley (2000) pp. 84–86.
13.3.1 A Nonlinear Wave Equation
The basic state variables are the acoustic pressure p, the particle velocity v̄, and the
current density ρ. The ambient state corresponds to p = 0, v̄ = 0̄, and ρ = ρ0 . The
field equation we seek governs the velocity potential φ (x̄, t), whose gradient is the
particle velocity,
v̄ = ∇φ (13.3.1)
In terms of this function, the continuity equation, Eq. (4.1.5), is
1 ∂ρ ∇ρ
+ · ∇φ + ∇ 2 φ = 0 (13.3.2)
ρ ∂t ρ
and the momentum equation, Eq. (4.1.7), is
∂ 1 ∇p
∇φ + ∇ (∇φ · ∇φ) = − (13.3.3)
∂t 2 ρ
The overall task now is to replace terms that contain p or ρ with terms that
solely contain φ. Our strategy entails manipulating the momentum equation and
the equation of state to obtain suitable representations of the first two terms in the
continuity equation. Removal of the second term is the more direct. The corollary
of the assumption of constant entropy is that p is a known function of ρ. Taking the
gradient of this function shows that ∇p is proportional to ∇ρ, specifically,
1/2
dp
∇p = c̃2 ∇ρ, c̃ = (13.3.4)
dρ
The quantity c̃ is the tangential sound speed we first encountered in the derivation of
the Riemann solution. The result of substituting this definition into the momentum
equation is an expression we will use to replace the second term in the continuity
equation, Eq. (13.3.2), specifically,

∇ρ 1 ∂ 1
=− 2 ∇φ + ∇ (∇φ · ∇φ) (13.3.5)
ρ c̃ ∂t 2
The task now is to find a suitable description, the first term in the continuity
equation using relations derived from the momentum equation and/or equation of
state. The means for doing so entails working with the pressure, rather than the
density, using an ingenious application of Leibnitz rule for differentiating an integral.
According to it

1 ∂p ∂ p dp
≡
ρ ∂t ∂t 0 ρ (p )
p
∇p dp
≡∇
(13.3.6)
ρ 0 ρ (p )
13.3 General Analytical Techniques 599
Both identities follow from the fact that only the upper limit of the integral depends
explicitly on position and time.
These relations merely replace one form of dependence on ρ with another. The
reason they are useful is that the right sides are respectively the time derivative or
gradient of a quantity. Substitution of the gradient form into the original momentum
equation, Eq. (13.3.3), gives

∂φ 1 p
dp
∇ + ∇φ · ∇φ + =0 (13.3.7)
∂t 2 0 ρ (p )
This equation requires that the term within the parentheses be independent of position,
but it could be a function of time. However, it was stipulated that the fluid is initially
at rest in the ambient state. The term in parentheses is zero at that initial state and
therefore always. Thus, we have

∂φ 1 p
dp
+ ∇φ · ∇φ + =0 (13.3.8)
∂t 2 0 ρ (p )
In later developments, this equation will lead to a relation for the pressure corre-
sponding to a known solution for the velocity potential. Here, we use it to eliminate
the density from the continuity equation. Toward that objective, we differentiate the
equation with respect to time and then recall the first of Eq. (13.3.6). The result is

∂2φ ∂ 1 1 ∂p
+ ∇φ · ∇φ + =0 (13.3.9)
∂t 2 ∂t 2 ρ ∂t
The pressure is eliminated by applying the chain rule for differentiation to the
equation of state, which gives ∂p/∂t = c̃2 ∂ρ/∂t. Thus, these manipulations of the
momentum equation have led us to a relation that matches the first density term in
Eq. (13.3.2),
1 ∂ρ 1 ∂2φ ∂ 1
=− 2 + ∇φ · ∇φ (13.3.10)
ρ ∂t c̃ ∂t 2 ∂t 2
The result of substituting this relation and Eq. (13.3.5) into Eq. (13.3.2), followed by
collection and rearrangement of some terms, is
∂2φ ∂ 1
c̃2 ∇ 2 φ − − (∇φ · ∇φ) − ∇ (∇φ · ∇φ) = 0 (13.3.11)
∂t 2 ∂t 2
All product terms in the preceding are the result of kinematical nonlinearity.
Constitutive nonlinearity is contained in the equation of state, but that effect is buried
in c̃2 . Differentiation of the equation of state will give c̃2 as a function of either p or
ρ, depending on how the equation of state is written. Unfortunately, neither form is
acceptable because we seek a field equation in which φ is the sole dependent variable.
Thus, we cannot proceed until we specify the equation of state. Let us begin by
considering an ideal gas. For an isentropic process, the relation is
1/γ
ρ p
= +1 (13.3.12)
ρ0 p0
The definition of c̃ gives

(γ−1)/γ
1 p
c̃2 ≡ = c2 +1 (13.3.13)
dρ/dp p0
We wish to eliminate p in favor of φ in this equation. The means for doing so is to

substitute the equation of state into Eq. (13.3.8). This operation makes it possible to
evaluate the integral, specifically
−1/γ
∂φ 1 1 ) p p
+ ∇φ · ∇φ + +1 dp
∂t 2 ρ0 0 p0
2 (γ−1)/γ
∂φ 1 c p
≡ + ∇φ · ∇φ + +1 −1 =0
∂t 2 γ−1 p0
(13.3.14)
where the relation for the linear speed of sound, c2 = γp0 /ρ0 , has been used. Solving
this equation for p in terms of φ yields
γ/(γ−1)
p γ − 1 ∂φ 1
= 1− + ∇φ · ∇φ −1 (13.3.15)
p0 c2 ∂t 2
We seek an expression for c̃2 in terms of p, so we substitute the above expression for
p into Eq. (13.3.13). The result is

∂φ 1
c̃ = c − (γ − 1)
2 2
+ ∇φ · ∇φ (13.3.16)
∂t 2
Substitution of c̃2 into Eq. (13.3.11) yields a nonlinear field equation,

∂2φ ∂φ ∂
c ∇ φ − 2 − (γ − 1) ∇ 2 φ
2 2
+ (∇φ · ∇φ)
∂t ∂t ∂t
(13.3.17)
(γ − 1) 1
+ (∇φ · ∇φ) ∇ 2 φ + ∇φ · ∇ (∇φ · ∇φ) = 0
2 2
This is the nonlinear wave equation. When the terms contained within the bracket
are dropped, what remains is the linear wave equation. The full equation governs any
wave in an ideal gas. Its appearance is quite daunting. We may simplify it somewhat
by invoking the small-signal approximations. The implication of |p| being much less
than the bulk modulus is that φ also is small. If that is so, then the cubic terms are
much less than the quadratic terms, which are much less than the linear terms. It
is necessary to retain the quadratic products as the first approximation of nonlinear
effects, but the cubic terms may be dropped.
In addition to simplification of the field equation, dropping the cubic terms has
another beneficial feature. When the equation of state for an ideal gas is truncated at
quadratic terms, it is the same as the equation for a liquid if we state both in terms
of the coefficient of nonlinearity in Eq. (13.1.34). Hence, setting γ + 1 = 2β in the
nonlinear wave equation and deleting the cubic terms yield a field equation that is
descriptive of weakly nonlinear waves in an ideal gas, as well as almost any wave in
a liquid. This version of the nonlinear wave equation is

∂2φ 2 ∂φ ∂
c ∇ φ − 2 − 2 (β − 1) ∇ φ
2 2
+ (∇φ · ∇φ) = 0 (13.3.18)
∂t ∂t ∂t
We will seek solutions of this equation. However, the state variable of interest
is the pressure. Equation (13.3.15) describes p in terms of φ. This relation also is
simplified by the small-signal approximations. The bracketed term is expanded in a
binomial series. Truncation of the expansion at quadratic terms gives
2
p γ ∂φ 1 1 ∂φ
=− 2 + ∇φ · ∇φ − 2 (13.3.19)
p0 c ∂t 2 2c ∂t
A minor simplification comes from the relation for the linear speed of sound in an
ideal gas, γpo /ρ0 = c2 . This converts the preceding to
2
∂φ 1 1 ∂φ
p = −ρ0 + ∇φ · ∇φ − 2 (13.3.20)
∂t 2 2c ∂t
The absence of γ or β from this expression tells us that the nonlinear effect contained
in this relation is kinematical in origin and may be used for both liquids and gases.
We also see that the relation reduces to the linear one when the quadratic terms are
dropped.
In summary, we have found that any acoustic signal in a liquid and all except
the loudest acoustic signals in an ideal gas are governed by the quadratic version of
the nonlinear wave equation, Eq. (13.3.18). Boundary conditions typically impose
conditions on the particle velocity, which is v̄ = ∇φ. The pressure corresponding to a
solution for φ is given by Eq. (13.3.20). In the exceptional case of a signal in an ideal
gas whose amplitude is deemed to be too large to use a small-signal approximation,
the appropriate nonlinear wave equation is Eq. (13.3.17) and Eq. (13.3.15) should be
used to determine the pressure.
The derivation of the nonlinear wave equation is based on the assumptions that
dissipation mechanisms are insignificant, and that the signal propagates isentropi-
cally. It is possible to modify this equation by adding terms that capture dissipative
effects. Linear terms are adequate for this purpose, because dissipation, relaxation,
and such are small effects that are comparable to the effect of nonlinearity. Suitable
dissipation terms may be found in the modified versions of the linear wave equation
that were developed in Sect. 3.3.
Even if we exclude nonideal effects, a direct attempt to solve the nonlinear wave
equation by standard techniques, such as separation of variables, would not be suc-
cessful. For this reason, several simplified field equations have been developed.15
They differ in their assumptions regarding scales over which state variables change
in various directions and the relative order of magnitude of terms representing var-
ious effects. Some model equations are particularly amenable to analysis by finite
difference methods. Our concentration will be on systems that are amenable to ana-
lytical techniques. The quadratic nonlinear wave equation will be the framework for
our investigations, but the same techniques may be applied to other model equations.
13.3.2 Frequency-Domain Formulation
Numerical techniques have been used to solve the nonlinear wave equation. The first
thought might be to use finite difference formulas to approximate the derivatives in
the nonlinear wave equation. However, that approach has inherent limitations, not
the least of which is the accumulation of error with increasing distance and time. This
is a crucial issue because the error might overwhelm the nonlinear effects, especially
for a weakly nonlinear signal. Another issue is the large number of mesh points
resulting from the requirements that the spatial mesh size be considerably smaller
than a wavelength and the temporal time increment be a fraction of the interval over
which the signal fluctuates.
The best computational approaches begin with an ansatz that captures as many
aspects as possible of the fundamental physics of a system. If we are solely interested
in situations where the excitation is periodic in time, a formulation that inherently
recognizes the periodicity of the steady-state signal can reasonably be expected to
be optimal. Periodicity allows the time signature to be represented by a Fourier
series. In linear theory, each harmonic in that series would be a solution of the
linear wave equation, and the series coefficients would be constants. We have seen
with the Fubini-Ghiron solution that the amplitude of each harmonic in a plane
wave is position dependent. This suggests that the periodic signal in any system
may be described as a Fourier series whose harmonic functions are solutions of the
linear wave equation for that system, and whose coefficients are position dependent.
Making this ansatz satisfy the nonlinear wave equation will lead to a coupled set
of first-order differential equations for the coefficients. Such equations are readily
solved by standard numerical methods.
15 M.F. Hamilton and C.L. Morfey, “Model Equations,” M.F. Hamilton and D.T. Blackstock, Non-
linear Acoustics, eds., Acoustical Society of America (2008) Chap. 3.

This approach is said to work in the frequency domain because the signal is
decomposed into its harmonics. However, the most useful property of linear equa-
tions, which is the principle of superposition, is not applicable to nonlinear equations.
A corollary is that one cannot simplify matters by using real or imaginary parts and
factoring out complex exponential factors. We shall develop the basic approach by
applying it to the analysis of a plane wave.
The period is T , so the fundamental frequency is ω1 = 2π/T . The corresponding
phase variable for a linear wave is
ω1
θ1 = ω1 t − kx, k = (13.3.21)
c
Thus, the Fourier series representation of a plane wave is taken to be

∞
1 #
φ= n (x) einθ1 , (−n) = ∗n (13.3.22)
2 n=−∞
Equations for the n functions are obtained by requiring that this expression satisfies
the quadratic nonlinear wave equation, Eq. (13.3.18). An important aspect of those
operations arises when we consider ∇ 2 φ, which is
∞
∂2φ 1 # & '
∇2φ = = n − 2inkn − n2 k 2 n einθ1 (13.3.23)
∂x 2 2 n=−∞
where a prime will denote an ordinary derivative with respect to x. Based on our
experience thus
far, as
well as the expectation that nonlinear effects grow gradually,
we consider n (x) k |n (x)|. In other words, we take each of the amplitude
factors to be a slowly varying function. Correspondingly, second derivatives of n
are ignored, so the linear terms in the wave equation become
#∞
∂2φ ∂2φ
c2 − = −ic 2
k nn einθ1 (13.3.24)
∂x 2 ∂t 2 n=−∞
The nonlinear terms in the wave equation, Eq. (13.3.18), require that derivatives
of the Fourier series be multiplied. The manipulations are simplified by the slowness
property, which allows derivatives of the n coefficients to be dropped when the
product is formed. This means that only variation of the complex exponential is
considered, so that
∂φ ∂φ ∂θ1 ∂φ ∂φ ∂φ ∂θ1 ∂φ
≈ = ω1 , ≈ = −k (13.3.25)
∂t ∂θ1 ∂t ∂θ1 ∂x ∂θ1 ∂x ∂θ1
In turn, these properties lead to
∂φ ∂ 2 φ ∂φ ∂ 2 φ ∂φ
∇2φ = ≈ k 2
ω 1
∂t ∂x 2 ∂t ∂θ12 ∂θ1
(13.3.26)
∂ ∂ φ ∂φ
2
∂ 2 φ ∂φ
(∇φ · ∇φ) = 2 ≈ 2k 2 ω1 2
∂t ∂x∂t ∂x ∂θ1 ∂θ1
The preceding relations allow us to combine the product terms in the wave equation.
Each factor in the product must be described by a different summation index in order
to assure that all individual products are described. Thus, the product terms become
∂φ ∂ ∂ 2 φ ∂φ
2 (β − 1) ∇ 2 φ + (∇φ · ∇φ) = 2βk 2 ω1 2
∂t ∂t ∂θ1 ∂θ1
#∞ #∞
i
=− β mn2 k 2 ω1 m n ei(n+m)θ1
2 n=−∞ m=−∞
(13.3.27)
Substitution of this expression and Eq. (13.3.24) into the nonlinear wave equation,
Eq. (13.3.18), gives
∞
# #∞ #∞
i
− ic2 k nn einθ1 = − βk 2 ω1 mn2 m n ei(n+m)θ1 (13.3.28)
n=−∞
2 n=−∞ m=−∞
The frequencies of the harmonic functions are multiples, so the functions are
orthogonal over their common period. This property makes it possible to obtain a
set of first-order differential equations for the amplitude functions. To that end, we
multiply the equation by exp (−ijθ), where j may be any integer. We then integrate
over a period, −π < θ1 < π. All terms in the single summation, other than the one
for which n = j, are orthogonal to this complex exponential. The only terms in
the double summation that are not orthogonal to exp (−ijθ1 ) are those for which
m = j − n. Thus, these operations filter a single term out of the single summation
and reduce the double summation to a single one. The result is a set of coupled
ordinary differential equations,
∞
1 k2ω #
jkj = β 2 (j − n) n2 (j−n) n (13.3.29)
2 c n=−∞
It would be more meaningful if we worked with the Fourier pressure coefficients.

We may describe p (x, t) as a Fourier series,
∞
1 #
p= Pn (x) einθ1 , P(−n) = Pn∗ (13.3.30)
2 n=−∞
The pressure is related to the velocity potential by p = −ρ0 ∂φ/∂t. (Nonlinearity

is uniformly a second-order effect here.) The result of replacing p and φ by their
respective Fourier series and matching like harmonics shows that, for any n,

i
Pn = −inωρ0 n =⇒ nkn = Pn (13.3.31)
ρ0 c
When this relation is substituted into Eq. (13.3.29), the result is

∞
Pj i ω # P(j−n) Pn
= β 2 n (13.3.32)
ρ0 c 2 c n=−∞ (ρ0 c)2
This form suggests nondimensionalization of the position and pressure coefficients

according to
Pn
x̃ = kx, P̃n = (13.3.33)
ρ0 c2
This replacement converts the differential equations to
∞
d P̃j i #
= β nP̃(j−n) P̃n (13.3.34)
d x̃ 2 n=−∞
These differential equations apply for all j, but it would be repetitive to evaluate
the coefficients for j < 0 because P̃(−j) is the complex conjugate of P̃j . If we are
to make use of this property to reduce the number of variables, we must explicitly
account for the complex conjugate property within the summation. The first step is
to break up the summation into two parts, n > 0 and n < 0. (The contribution from
n = 0 is zero.) For the second part, n is replaced by −n, which gives
∞ ∞

d P̃j i # #
= β nP̃(j−n) P̃n − nP̃(j+n) P̃(−n) (13.3.35)
d x̃ 2 n=1 n=1
In the equation for j = 0, the sums cancel, so that
d P̃0
=0 (13.3.36)
d x̃
In other words, the mean pressure P̃0 is independent of the position.

For j > 0, the second summation in Eq. (13.3.35) features a coefficient with a
negative subscript. The complex conjugate property gives P̃(−n) = P̃n∗ . The first sum-
mation also has a coefficient with a negative subscript when n > j. This is handled
∗
by splitting the summation into two ranges: n ≤ j and n > j, with P̃(j−n) = P̃(n−j) for
the latter range. These operations lead to
⎡ ⎤
∞ ∞
i ⎣# # #
j
d P̃j
= β nP̃(j−n) P̃n + ∗
nP̃(n−j) P̃n − nP̃(j+n) P̃n∗ ⎦ , j > 0 (13.3.37)
d x̃ 2 n=1 n=j+1 n=1
Replacing n with j + n in the middle summation reduces this expression to

j ∞

d P̃j i # #
∗
= β nP̃(j−n) P̃n + jP̃n P̃(j+n) , j > 0 (13.3.38)
d x̃ 2 n=1 n=1
This is the form we seek because it only features Pn values for n > 0.
The first summation in this equation describes the growth or decay of harmonic
j due to the interaction of harmonics below j. The second summation captures the
influence on harmonic j of a higher harmonic interacting with all harmonics. For
example, the first summation indicates that second harmonics ( j = 2) are generated
by the interaction of the first harmonic with itself, while the second summation indi-
cates that second harmonics are also generated by the interaction of many harmonic
pairs consisting of first and third, second and fourth, and so on.
The differential equations must be solved numerically, so it is necessary to truncate
the pressure series. Let N be the highest harmonic that is retained. Thus, we set P̃n = 0
for n > N. Neither subscript in the first summation will exceed N, so that summation
requires no adjustment. The lowest value of n in the second summation that leads to a
factor being zero is that for which j + n = N + 1. Thus, that summation is truncated
at n = N − j, which means that the equations to be solved numerically are
j
d P̃j i # #
N−j
∗
= β nP̃j−n P̃n + jP̃n P̃(j+n) , j = 1, 2, ..., N (13.3.39)
d x̃ 2 n=1 n=1
The second summation is not executed if j = N. The value of P̃0 is set at whatever
constant value it has at x = 0, so there are N pressure coefficients to determine and
N differential equations to solve.
An attractive feature of the frequency-domain analysis is its generality. Monitor-
ing the term on the right side of Eq. (13.3.39) enables one to delve into the interaction
of harmonics at a fundamental level, rather than merely as a computed result. Another
beneficial aspect of the formulation is that it is readily modified to incorporate the
effects of dissipation. Section 3.3 discusses how dissipation is manifested in the
frequency domain. It was shown that weak dissipation attenuates the complex ampli-
tude of a plane wave. Dissipation is important for nonlinear propagation because the
decay constant typically increases with increasing frequency. Thus, attenuation due
to dissipation counters the nonlinear tendency for harmonics to grow as the wave
propagates.
For propagation in the direction of increasing x, the attenuation factor is e−αx . The
absorption coefficient α is a real quantity that depends on frequency. Different dissi-
pation effects, such as bulk viscosity or wall friction, lead to different dependencies,
so we will write it as α (ω). If nonlinear effects were not present, this exponential
decay would be obtained if the coefficients were governed by
d P̃j
+ α ( jω1 ) P̃j = 0 (13.3.40)
d (x̃/k)
The assumption that dissipation is weak implies that it is not altered by nonlinear
effects, and vice versa. Consequently, we may insert the dissipation term into the
nonlinear amplitude equations, with the result that Eq. (13.3.39) becomes
j
d P̃j α (jω1 ) i # #
N−j
∗
=− P̃j + β nP̃j−n P̃n + jP̃n P̃(j+n) , j = 1, 2, ..., N
d x̃ k 2 n=1 n=1
(13.3.41)
Note that the presence of dissipation does not alter the fact that P̃0 is independent
of x.
Starting values for these differential equations are obtained from the boundary
condition. In the weakly nonlinear regime, ε 1, the particle velocity at x = 0
equals the motion of the boundary. Let εcV (t) denote the velocity there. The small-
signal approximation gives the corresponding pressure ρ0 c (εcV ). The phase variable
at x = 0 is θ = ω1 t, so matching the Fourier series to the motion that is imposed gives
∞
1 2 #
ρ0 c P̃n (x̃ = 0) einω1 t = ρ0 c [εcV (t)] (13.3.42)
2 n=−∞
Orthogonality of the harmonic functions in a Fourier series then leads to
π/ω1
ε
P̃j (x = 0) = V (t) e−ijω1 t d (ω1 t) (13.3.43)
2π −π/ω1
EXAMPLE 13.5 Determine the Fourier series pressure coefficients for the
steady-state plane wave generated by a harmonic source at x = 0. The parti-
cle velocity at that location is v = εc sin (ωt) with ω = 2π (5000) rad/s and

ε = 2 10−5 . The fluid is air at standard conditions. Let the decay con-
stant be α = 2.54 10−13 ω 2 neper/m. Carry out the evaluation to the dis-
tance where a shock would form if there were no dissipations, which is
(xshock )sine = 1/ (βεk). Compare the results to those obtained from the Fubini-
Ghiron solution for the dissipationless case. Also evaluate the waveform at
(xshock )sine .
Significance
Issues such as implementation of a numerical procedure for solving the coupled
differential equations and the number of harmonics that are required shall be covered.
Equally important is the demonstration of the interplay of dissipation and nonlinearity
effects.
Solution
The analysis entails solving Eq. (13.3.41) subject to the initial conditions
described

by Eq. (13.3.43). The velocity at x = 0 is specified to be v = εc eiωt − e−iωt / (2i).
Matching ρ0 cv to the Fourier series for p at x = 0 leads to
P̃1 (x = 0) = −iε, otherwise Pn (x = 0) = 0 (1)
The mean value, P̃0 , is zero. Nevertheless, we shall treat it as a variable because
doing so leads to the implementation of a numerical procedure that is valid for any
(periodic) boundary velocity.
The coupled differential equations may be solved with a Runge–Kutta numerical
integrator. Several versions are available; we shall use a forth-fifth order scheme.
The form that is required is
d P̃j
= Fj P̃n , x̃ , j = 0, ..., N (2)
d x̃
Equation (13.3.41) matches this general form, with

F0 P̃n , x = 0
j
α(jω1 ) i # #
N−j
∗
Fj P̃n , x = − P̃j + β nP̃j−n P̃n + jP̃n P̃(j+n) , j = 1, ..., N − 1
k 2 n=1 n=1
N
α(Nω1 ) i #
Fj P̃n , x = − P̃j + β nP̃j−n P̃n , j = N (3)
k 2 n=1
There is no simple way to vectorize computation of the Fj terms, so the operations

are implemented term by term. Within a loop that increments j from 1 to N, the
computation begins by equating Fj to the linear term. A loop over n from 1 to j adds
the terms in the first summation. Then, if j is less than N, a loop over n from one to
N − j adds the terms in the summations. The only caution is that if one’s software
requires that subscripts begin at one, as is the case with MATLAB, then the subscript
number must be one greater than the associated harmonic number.
The dissipation factor that appears in the differential equations is α (jω1 ) k. For
the given α dependence, this is
α (jω) j2 ω12
= 2.54 10−13 = 2.54 10−13 (ω1 c) j2 (4)
k k
The units of α are nepers/m, so α/k is a dimensionless quantity. This means that the
numerical coefficient has units of s2 /m.

The computation requires selection of N. We know from Fig. 13.8 that P̃n P̃1
if n > 5, and dissipation will affect the higher harmonics more than P̃1 because
the dissipation factor increases monotonically with increasing harmonic number.
We shall begin with N = 10, which would seem to be adequate to assure that all
significant nonlinear interactions are properly described. As one should always do
when making an a priori decision like this, we will verify the adequacy of this
truncation by comparing the results to those obtained from a larger N.
It is requested that the series coefficients be compared to those of the Fubini-
Ghiron solution of the dissipationless case. That solution is described in Eq. (13.2.13)
in terms of the retarded time t = t − x/c and x nondimensionalized relative to the
shock formation distance for harmonic excitation, (xshock )sine = 1/(βεk). In terms of
the present variables, the complex Fourier series for the signal in the case of an ideal
fluid is
# ∞
1 2Jn (nβεx̃) Pn 2ε
p = ρ0 c2 P̃n einθ1 , P̃n = = Jn (nσ) (5)
2 n=1
inβ x̃ ρ0 c 2 nσ
One computational check is to compare the solution of the differential equations for

α = 0 to this expression. Evaluation of P̃n − P̃n /ε showed an error

comp F-G
that rose from 0.02% for n = 1 to 0.14% for n = 9. The error for n = N = 10 was
found to be 4.8%, which is typical behavior when a Fourier series is truncated. For
reasons that will become apparent when we examine waveforms, the computation
was repeated with N = 50. The error for the first ten harmonics in that case was
found to be less than 0.005%.
The pressure coefficients when α(jω) is given by Eq. (4) are the solid curves in
Fig. 1. The dashed curves are the values for the dissipationless case. Each amplitude
is attenuated. For clarity, both linear and logarithmic scales are used. Viewed as a
ratio, the loss is greater as the harmonic number increases. Whether this is a drastic
effect depends on the appearance of the waveform at the shock formation distance.
Fourier coefficients,
2(10 -5) n=1
105 x Pn /(0c2)
1(10 -5)
n=2
n=3
0
Fourier coefficients,
n=1
10 -5
Pn /(0c2)
n=2
10 -6 n=3
n=4
0 0.5 1 1.5 2 2.5 3 3.5 4

Propagation distance, 10-4 x kx
Figure 1.
A waveform is obtained by evaluating Eq. (13.3.30). The result at the shock for-
mation distance, which is x̃ = k(xshock )sine = 1/(βε), appears in Fig. 2. The result for
the ideal case could be obtained by evaluating the Riemann solution; instead, it also
was obtained by computing the series. The cutoff number N was raised to obtain
these waveforms because a much higher resolution is required to capture the steep
slope of the α = 0 curve. This is unlike the behavior of the curve for α = 0, which
was found to differ negligibly from the one obtained from N = 10.
2
105x p/(0c2)
1
0
−1 =0
−2
0 0.5 1 1.5 2
Time, t / (2)
Figure 2.
The waveform in Fig. 2 shows that the presence of dissipation has two effects.
The first is to attenuate the overall amplitude, which would be p/ρ0 c2 = ε if α
were zero. The second effect is to reduce the slope in the phase where ∂p/∂t > 0.
Thus, it forestalls the formation of a shock. The differential equations lose validity
when a shock forms because they do not account for the Rankine–Hugoniot con-
ditions. Conversely, they can be used for any x̃ in the α = 0 case, provided ∂p/∂t
remains finite at lesser x̃. Figure 3 is the waveform at x̃ = 4/ (βε). It appears that the
attenuation is sufficiently great that a shock will not form at any location. Because
the attenuation increases with increasing harmonic number, at a sufficiently large
distance the signal will be reduced to a greatly attenuated fundamental. This is one
of the reasons that sounds that are generated in everyday life do not form shocks.
105x p/(0c2)
1
0
−1
−2
0 0.5 1 1.5 2
Time, t / (2)
Figure 3.
13.3.3 Regular Perturbation Series Expansion
There are a number of fundamental systems and phenomena that can be solved
analytical, rather than by recourse to numerical methods. One approach that has had
success is a perturbation series representation of the pressure or velocity potential.
The concept is limited by the requirement that the linearized version of the problem
be solvable. In actuality, this is not much of a limitation, because it is unreasonable
to expect to solve the nonlinear equations if the linearized version of those equations
cannot be solved. The starting point is identification of a basic parameter that can
be used as a metric of the relative importance of linear and nonlinear terms in the
field equation. We have already encountered such a parameter, the acoustic Mach
number ε.
In a linear problem, any constant that multiplies the excitation scales the response
in the same way. This is not true for a nonlinear signal, so it follows that the poten-
tial function and therefore, the acoustic pressure and particle velocity are unknown
functions of the Mach number ε, as well as position and time. This dependence on
ε can be expected to be analytical, at least in regions where shocks do no occur. An
analytical dependence on ε implies that the variables can be represented as a Taylor
series relative to ε = 0. The response variable for the nonlinear wave equation is φ,
so its Taylor series is
2
∂ 1 ∂
φ(x̄, t, ε) = φ(x̄, t, ε)|ε=0 + ε φ(x̄, t, ε) + ε2 φ(x̄, t, ε) +···

∂ε ε=0 2 ∂ε 2
ε=0
(13.3.44)
The functional dependence of φ is not known when the analysis is initiated, so the
derivatives may be replaced by functions to be determined. Furthermore, because ε
is the scaling factor of the excitation, setting ε = 0 means there is no excitation, in
which case φ will be zero. This means that the series should not contain a term that
is independent of ε. Thus, the series reduces to
φ (x̄, t, ε) = εφ1 (x̄, t) + ε2 φ2 (x̄, t) + ε3 φ3 (x̄, t) + · · · (13.3.45)

When this series is used as a Taylor series, the concept is to add terms until it
converges. Adding more terms will decrease the error of the truncated series relative
to the true φ. This is not the viewpoint if the representation is considered to be a
perturbation series. Rather, the number of terms is fixed. If |ε| 1, we know that
the difference between the series that is truncated beyond εN and the true φ will
be proportional to εN+1 . In other words, lengthening a perturbation solution would
reduce the error at a specified ε, rather than increasing the range of ε for which the
error is less than a specified amount. The change in viewpoint from a Taylor series to
a perturbation series is fortunate, because the analysis of each function φn becomes
progressively more difficult as n increases. It follows that there is no intent to increase
the length of the series, even though doing so might lead to a convergent answer.
Frequent reference will be made to the order of magnitude of a term, so it is
important to have a clear definition of that metric. A function f (ε) is said to be O (εn )
if, for two very small ε values, it is found that f (ε1 ) /f (ε2 ) is essentially (ε1 /ε2 )n .
We will use the weakly nonlinear version of the wave equation, Eq. (13.3.18). Cubic
terms have been omitted from this equation. The velocity potential is O (ε), so the
omitted terms are O ε3 . It would be inconsistent to seek terms at that order in the
response, so the perturbation series
for the velocity potential pressure and particle
velocity are truncated at O ε2 . The forms are
φ = εφ1 (x, t) + ε2 φ2 (x, t) + · · ·

p = εp1 (x, t) + ε2 p2 (x, t) + · · · (13.3.46)
v = εv1 (x, t) + ε2 v2 (x, t) + · · ·
We have no intrinsic interest in the density as a response variable, but if we did, its
perturbation series would begin with ρ0 , which is the O (1) term.
The next step is to assemble the equations to be satisfied. For a planar wave, they
are
∂φ
v=
∂x
2 2
∂φ 1 ∂φ 1 ∂φ
p = −ρ0 + − 2 (13.3.47)
∂t 2 ∂x 2c ∂t

∂2φ ∂2φ ∂ 2 φ ∂φ ∂φ ∂ 2 φ
c2 2 − 2 − 2 (β − 1) 2 +2 =0
∂x ∂t ∂x ∂t ∂x ∂x∂t
The early development of the method of characteristics

showed that the effect of a
moving boundary condition is always O ε2 , see Eq. (13.1.17). Nevertheless, it is
useful to see how such effects may be incorporated in a perturbation analysis. We
have defined ε to measure the magnitude of the boundary motion, so we take the
displacement of the boundary to be εX (t). Thus, the boundary condition is

∂φ
= εẊ (t) (13.3.48)
∂x x=εX(t)
Statement of the problem is completed by imposing the radiation condition, which

requires that p be an outgoing wave as x → ∞, and that the fluid was at the ambient
state for t < 0, so that p = v = 0 at t = 0.
The fundamental step in the analysis stems from the requirement that the pertur-
bation series constitute a solution for any ε. When any such series is substituted into
an equation, terms like order of ε may be collected. The result will have the form
ε × (term 1) + ε2 × (term 2) = 0. It is necessary that this equation be satisfied for
many different values of ε. This requirement can be met only if the coefficient of
each power of ε is zero. Thus, a single equation becomes an equation for each power
of ε: term 1 = 0, term 2 = 0, .... An equivalent statement is that the coefficients of
like powers of ε in each equation must match.
In this way, we obtain a sequence of problems in which ε does not appear. Each
set of equations is described as being the equations at the order of the associated
power of ε. The first-order set consists of term 1 from each governing equation, the
second-order set consists of term 2 for each equation, and so on. Thus, the first-order
equations derived from Eq. (13.3.47) are
∂φ1
v1 =
∂x
∂φ1
p1 = −ρ0 (13.3.49)
∂t
∂ 2
φ 1 ∂ 2
φ 1
c2 2 − =0
∂x ∂t 2
The second-order equations are
∂φ2
v2 =
∂x

∂φ2 1 ∂φ1 2 1 ∂φ1 2
p2 = −ρ0 − − 2 (13.3.50)
∂t 2 ∂x 2c ∂t
∂ 2 φ2 ∂ 2 φ2 ∂ 2 φ1 ∂φ1 ∂φ1 ∂ 2 φ1
c2 − = 2 (β − 1) + 2
∂x 2 ∂t 2 ∂x 2 ∂t ∂x ∂x∂t
A slightly simpler differential equation

for φ2 is obtained if the first-order equation
is used to replace ∂ 2 φ1 /∂x 2 with 1/c2 ∂ 2 φ1 /∂t 2 . Doing so leads to
2 2
∂ 2 φ2 ∂ 2 φ2 ∂ (β − 1) ∂φ1 ∂φ1
c2 2 − = + (13.3.51)
∂x ∂t 2 ∂t c2 ∂t ∂x
It will be noted that at each order the terms that are lower order have been placed to
the right of the equality sign. Doing so anticipates the way in which the equations are
solved. The first-order equations are linear in the dependent variables, which means
we should be able to solve them. The left side of the first- and second-order equations
is the same, but each second-order equation is inhomogeneous. Substitution of the
first-order solutions will convert them into terms that represent a spatial distribution
of sources.
The moving boundary condition also is transformed to a set of perturbation equa-
tions by means of a Taylor series relative to x = 0, specifically,

∂φ ∂φ ∂ 2 φ 1 2
2 ∂ φ
= + εX (t) + (εX (t)) + · · · = εẊ (t)
∂x x=εX(t) ∂x x=0 ∂x 2 x=0 2 ∂x 2 x=0
(13.3.52)
Only effects up to O ε2 are of interest.
Thus, we substitute the perturbation series
for φ and match the O (ε) and O ε2 terms on either side of the equality. The result
is a sequence of boundary conditions,

∂φ1
= Ẋ (t)
∂x x=0 (13.3.53)
∂φ2 ∂ 2 φ1
= −X (t)
∂x x=0 ∂x 2 x=0
In addition, the radiation and initial conditions must be satisfied for any ε, which
leads to the requirements that each order be an outgoing wave, and that each order
of p and v be zero for t < 0.
In most systems, solution of the linear problem requires specification of the exci-
tation. However, the d’Alembert solution, which is general, is available for plane
waves. An outgoing wave that satisfies the linear wave equation in Eq. (13.3.49) is
φ1 = f (t − x/c). The function f is determined by satisfying the first-order boundary
condition, Eq. (13.3.53),
1
− f (t) = Ẋ (13.3.54)
c
The notation f prime denotes differentiation with respect to the argument of f , which
is t − x/c. The chain rule for differentiation leads to f (t − x/c) ≡ (∂/∂t) f (t − x/c),
so the boundary condition may be integrated with respect to time to find that
x
φ1 = − cX t − (13.3.55)
c
The corresponding state variables are
x x
v1 = Ẋ t − , p1 = ρ0 cẊ t − (13.3.56)
c c
It is implicit to this solution that X (t) = 0 for t < 0, so these terms are consistent
with the initial condition.
Now that we have determined the first-order terms, we proceed to the next order.
Substitution of φ1 into the second-order wave equation, Eq. (13.3.51), gives

2∂
φ2 2
∂ 2 φ2 ∂ x 2
c − =β Ẋ t − (13.3.57)
∂x 2 ∂t 2 ∂t c
The corresponding second-order boundary condition is

∂φ2 1
= X (t) Ẍ (t) (13.3.58)
∂x x=0 c
The solution of the second-order wave equation consists of a complementary

equation solution (φ2 )c and a particular solution (φ2 )p . The definition of (φ2 )p is
that its substitution into the left side of the equation exactly matches the right side,
whereas substitution of (φ2 )c gives zero. The left side of the equation is the linear
wave equation, so any term that gives zero on the right side must be a function of either
characteristic variable. The requirement that such a term be a forward propagating
wave removes dependence on the backward characteristic. Therefore, we know that
x
(φ2 )c = Fc t − (13.3.59)
c
We now are presented with a small dilemma, because the source term on the right
side of Eq. (13.3.57) is a function of t − x/c. Any function with such dependence is
a homogeneous solution, so we cannot use that combination of variables to form a
particular solution. Let us try a particular solution that is modified by introducing a
factor x, specifically, x
(φ2 )p = xFp t − (13.3.60)
c
Substitution of this ansatz into the second-order wave equation yields

x ∂ x 2
− 2cFp t− =β Ẋ t − (13.3.61)
c ∂t c
According to the chain rule for differentiation, (∂/∂t) Fp (t − x/c) = Fp (t − x/c).
Thus, the preceding may be integrated directly. The resulting particular solution is
β x 2
(φ2 )p = − x Ẋ t − (13.3.62)
2c c
One might wonder at this juncture why the trial form of the particular solution was
not taken to be tFp (t − x/c), which could also be made to match the source term.
The term that was selected could be justified by noting that it matches the behavior
observed in the Riemann solution, specifically that distortion of the pressure profile
increases with increasing x. However, invocation of this argument would violate
the intention of solving the problem without having prior knowledge of the Riemann
solution. A justification fitting this criterion is that the form selected for the particular
solution must be valid for any X (t). If X (t) is periodic for t > 0, then regardless
of the fact that the system is nonlinear, the response must evolve to have that same
temporal periodicity. Such a property cannot be obtained if t is a factor in the trial
solution.
The function Fc for the complementary solution in Eq. (13.3.59) is arbitrary in
regard to satisfying the second-order wave equation. It is set by satisfying the second-
order boundary condition, Eq. (13.3.58). This condition must be satisfied by the sum
of the complementary and particular solutions, which is why the particular solution
for φ2 was found first. Thus, we seek a function Fc (t − x/c) for which

∂ & ' 1 β 1
(φ2 )c + (φ2 )p ≡ − Fc (t) − Ẋ (t)2 = X (t) Ẍ (t) (13.3.63)
∂x x=0 c 2c c
Because Fc (t) ≡ Ḟc (t) at x = 0, this equation may be integrated, although the inte-
gral can be evaluated only upon specification of X (t). Thus, we find that
t
β 2
Fc (t) = − Ẋ t dt + X t Ẍ t dt (13.3.64)
0 2
The second-order potential at any x is obtained by replacing t with t − x/c in this

expression and then combining it with the particular solution. Doing so yields
t−x/c
β x 2 β 2
φ2 = − x Ẋ t − − Ẋ t dt + X t Ẍ t dt (13.3.65)
2c c 0 2
The second-order pressure and particle velocity may be obtained from

Eq. (13.3.50). The result shall not be listed here because doing so would not advance
the dual objective of demonstrating how to implement a perturbation analysis, and
how to handle a moving boundary condition. However, the expression for φ2 that we
have derived serves several purposes.
It is evident that the full solution, which is obtained by combining first- and second-
order effects, is not at all like the Riemann solution. The combination is not consistent
the small-signal approximation that p = ρ0 cv. However, the discrepancy is
with
O ε2 everywhere, so the difference is not significant. It is far more important that
the perturbation solution has a limited range of validity. This is so because (φ2 )p is
proportional to x, so it grows in magnitude. In the vicinity of the shock formation
distance, x = O (1/ε). This makes ε2 φ2 have the same magnitude as εφ1 , but the
basic concept of a perturbation series is that each succeeding term is small compared
to the preceding terms. Therefore, the velocity potential we have derived is only valid
in regions where x = O (1). A perturbation series that is not equally accurate at all
locations is said to be nonuniformly valid. In contrast, the Riemann solution is valid
everywhere up to the shock formation distance. Some individuals say that φ2 has a
singularity at x = ∞ because it grows without bound as x increases.
Despite these faults, a nonuniformly valid series might be quite useful. Suppose we
were interested in mapping the field around the boundary and we had no knowledge
of the Riemann solution. The perturbation solution would explain certain nonlinear
effects, such as the occurrence of second harmonics if X (t) is sinusoidal, and the fact
that the difference between the linear and nonlinear solutions grows with increasing
propagation distance. The nonuniformly valid solution also illustrates a general fact
that nonlinear effects arising from moving boundary conditions do not grow with
increasing distance from the boundary. Ultimately, the most important justification
for the analysis is that its faults can be corrected. Methods for doing so are the subject
of the next section.
EXAMPLE 13.6 A plane wave is induced by a wall vibration that consti-

tutes a beat. The steady state motion is v = εc sin (σt) sin (ωt), where σ ω.
Identify the terms in the second-order velocity potential that are not uniformly
valid.
Solution
An analysis based on the d’Alembert solution, like the one for the preceding devel-
opment, will not be suitable for more general systems. Frequency-domain tools often
are likely to be employed in such circumstances. We will see how to do so here.
Solution
Although the basic perturbation equations have already been stated, it is good practice
to assemble them anew. We seek a solution for the velocity potential whose form is
φ = εφ1 (x, t) + ε2 φ2 (x, t) + · · · (1)
The first-order term must satisfy the linear wave equation in Eq. (13.3.49), and the
gradient of this term at x = 0 must match the wall vibration. Thus, the first-order
problem is
∂ 2 φ1 ∂ 2 φ1
c2 2 − =0
∂x ∂t 2
∂φ1
= c sin (σt) sin (ωt)
∂x x=0
The velocity of the wall is a beat. We describe it in terms of complex exponentials.

Doing so expedites solving the first-order equations, and it will also allow us to avoid
recalling trigonometric identities for products when we go on to the second order.
Thus, the boundary condition is converted to

∂φ1 1
= − εc eiσt − e−iσt eiωt − e−iωt
∂x x=0 4 (2)
1 i(ω−σ)t
= εc e − ei(ω+σ)t + c.c.
4
This is a familiar problem. Each harmonic in the excitation generates a forward

propagating wave at that frequency. Because there are two uncorrelated frequencies,
it is preferable to use the retarded time to represent each term,
x
τ =t−
c
Thus, a suitable trial solution is
φ1 = a1 ei1 τ + a2 ei2 τ + c.c. (3)
where
1 = ω − σ, 2 = ω + σ
Because ∂τ /∂x = −1/c, substitution of this form into Eq. (2) yields
ic2 ic2
a1 = , a2 = − (4)
41 42
We have seen that the moving boundary condition leads to second-order terms in
φ and p that are uniformly valid. Therefore, we may ignore that effect. Because the
gradient of the first-order term matches the specified wall velocity, it must be that
∂φ2 /∂x = 0 at x = 0. The resulting second-order problem is
∂ 2 φ2 ∂ 2 φ2 ∂ 2 φ1 ∂φ1 ∂φ1 ∂ 2 φ1
c2 − = 2 (β − 1) + 2
∂x 2 ∂t 2 ∂x 2 ∂t ∂x ∂x∂t (5)
∂φ2
=0
∂x x=0
The differential equation is simplified by the properties that
∂φ1 dφ1 ∂φ1 1 dφ1

= , =−
∂t dτ ∂x c dτ
Consequently, the differential equation in Eq. (5) may be written as
2
2∂φ2 ∂ 2 φ2 β dφ1 d 2 φ1 β d
2
dφ1
c − = 2 ≡ 2
∂x 2 ∂t 2 c2 dτ dτ 2 c dτ dτ
The term on the right side is formed by substituting for φ1 from Eq. (3). The use of
c.c. in that expression is a convenience; care must be taken to account for the unlisted
terms.
∂ 2 φ2 ∂ 2 φ2 β d /& i1 τ i2 τ

'2 *
c2 − = i1 a1 e + i2 a2 e + c.c.
∂x 2 ∂t 2 c2 dτ
β d &
= 2 −21 a12 e2i1 τ − 22 a22 e2i2 τ − 21 2 a1 a2 e2iωτ
c dτ
' (6)
+21 2 a1∗ a2 e2iστ + 21 a1 a1∗ + 22 a2 a2∗ + c.c.
2iβ & 3 2 2i1 τ
=− 1 a1 e + 32 a22 e2i2 τ
c2
'
+2ω1 2 a1 a2 e2iωτ − 2σ1 2 a1∗ a2 e2iστ + c.c.
The task now is to determine the portion of the particular solution for φ2 that grows
as x increases. Each term on the right side is solely a function of τ , so each identically
satisfies the homogeneous terms to the left of the equality side. Accordingly, we form
a trial solution by multiplying each term by x. (The alternative of multiplying each
term by t must be rejected because the excitation is a periodic function of time, so
the response must share that property.) Consequently, the trial form for (φ2 )p is
& '
(φ2 )p = x b1 e2i1 τ + b2 e2i2 τ + b3 e2iωτ + b4 e2iσt + c.c. (7)
Substitution of this ansatz into the left side of Eq. (6) generates terms that match the
right side if
β β
b1 = 3 21 a12 , b2 = 3 32 a22
2c 2c
β β
b3 = 3 1 2 a1 a2 , b4 = − 3 1 2 a1∗ a2
c c
The complementary solution of Eq. (6) is the d’Alembert solution, which does not
grow. Hence, we have completed the analysis. The velocity potential is constructed
by substituting Eqs. (3) and (7) into Eq. (1), which yields

c c
φ = ε(a1 ei1 τ + a2 ei2 τ ) + ε2 b1 x + e2i1 τ + b2 x + e2i2 τ
2i
1 % 2i2
c c
+ b3 x + e2iωτ + b4 x + e2iσt + c.c + O ε2
2iω 2iσ
(8)
This expression shows that the beating excitation at the boundary, which consists of
two harmonic signals at close frequencies 1 = ω − σ and 2 = ω + σ, generates
double-frequency harmonics at 2 (ω + σ) and 2 (ω − σ), as well as a sum frequency
harmonic at 2ω, and a difference frequency harmonic at 2σ. All frequencies except
2σ are close, so it seems that the waveform should display a beating pattern. These
frequencies appear in both the particular and complementary solutions for φ2 . The
second-order terms grow with increasing x, so the representation is valid only for
locations at which x is small relative to some reference length, like the shock forma-
tion distance.
The expression we have derived does not provide much insight regarding the
appearance of a waveform at substantial distances from the source. Although it was
not requested, Fig. 1 shows the waveform obtained from the Riemann solution at
xshock = c/(βεω). The frequencies are ω/ (2π) = 1000 Hz and σ/ (2π) = 40 Hz. The
envelope function that has been plotted is ±εc sin (σt). This is the true envelope for
linear theory, but a very close examination would show that the maxima and minima of
each oscillation of p sometimes do not reach the envelope, and sometimes it exceeds
the envelope. These discrepancies are minor. On the other hand, the oscillation within
the envelope shows the advancement and retardation that causes the waveform to lean
over.
1
p/(0c2)
−1
0 5 10 15 20 25
1
p/(0c2)
−1
0 2 4 6 8 10
t (ms)
Figure 1.
Before we see how a nonuniformly valid perturbation series can be corrected,

let us consider applying the Fourier series analysis in Sect. 13.3.2 to this problem.
The condition satisfied by a period is that it contain an integer number of cycles at
each frequency. In the case of the present excitation, it must be such that an interval
T contains n oscillations at ω − σ, and m oscillations at ω + σ. In other words,
T = n2π/ (ω − σ) = m2π/ (ω + σ). The smallest integers meeting this condition
for the frequencies in Fig. 1 are m = 13 and n = 12, which corresponds to a period of
T = 2π/80 s. In other words, in a Fourier series analysis the frequencies contained
in the excitation are the 12th and 13th harmonics of the fundamental frequency.
The highest harmonic for the frequency series must be many times the frequency
of the excitation if the propagation distance is comparable to xshock . Thus, a very
long series would be required if that formulation were to be applied to analyze the
pressure generated by a beating vibration. The formulation that follows provides an
alternative.
13.3.4 Method of Strained Coordinates
The occurrence of a nonuniformly valid description of a variable is not unique

to analyses of nonlinearity. Consider a harmonic waveform whose frequency is
slightly different from a nominal value ω0 , for example, y = sin ((ω0 + ε) t) ≡
sin (ω0 t) cos (εt) + cos (ω0 t) sin (εt). This expression may be expanded in a
Taylor series of increasing powers of ε. Truncation of that series at the first power
gives y ≈ sin (ω0 t) + εt cos (ω0 t). This is a nonuniformly valid perturbation series.
Rather than considering the nonuniformity to stem from a small difference of the
frequency from ω0 , we could consider the frequency to be ω0 but the actual time
variable to be slightly different from t. This is the essential idea underlying singu-
lar perturbation methods. Specifically, the nonuniformly valid perturbation series is
regarded as the correct solution in the wrong place in space and/or time.
A widely employed method for correcting a nonuniformly valid perturbation
series is Lighthill’s method of strained coordinates.16 It distorts the space-time grid
in a manner that is gradual when viewed over a small distance, but accumulates to a
significant level at long range. If the nonuniformly valid term grows with increasing
x, then the coordinate straining transformation replaces x with a variable ξ that differs
from x at a rate that matches the nonuniformly valid term(s), specifically,
x = ξ + εxF (ξ, t) (13.3.66)
The concept is to introduce this transformation into an expression that is not uniformly
valid and then select the function F that regularizes the expression.
We will use the regular perturbation series for plane waves derived in the previous
section as the framework for the development. The second-order velocity potential
in Eq. (13.3.65) is not required to be uniformly valid because it is an abstract quantity
that does not directly characterize the state of the fluid. Rather, the expressions for the
pressure and particle velocity derived by differentiating φ must behave properly. An
expression for φ is obtained by combining the first-order potential in Eq. (13.3.55)
and the second-order potential in Eq. (13.3.65). It is permissible to ignore terms in
φ that will lead to terms in p and v that are O ε2 everywhere, because those terms
will decrease in importance relative to the nonuniformly valid terms as the wave
propagates. Therefore, we may begin with
x β x 2
φ = − εcX t − − ε2 x Ẋ t − + O ε2 (13.3.67)
c 2c c

A corollary of being allowed to drop terms that uniformly are O ε2 or smaller is
that we may use p = −ρ0 (∂φ/∂t). No approximation is contained in v = ∂φ/∂x.
The state variables that result are
x β x x
p = ερ0 cẊ t − + ε2 ρ0 x Ẋ t − Ẍ t − + O ε2
c c c c

x β 2 x x x 2
v = εẊ t − +ε 2
x Ẋ t − Ẍ t − + Ẋ t − + O ε2
c 2c c c c c
(13.3.68)
16 M.J. Lighthill, “A technique for rendering approximate solutions to physical problems uniformly
valid,” Philosophical Mag., Vol. 40 (1949) 1179–1201.

The last term within the bracket for v does not grow as x increases. It is uniformly

valid, so it may be placed inthe group of terms that are uniformly O ε2 . Doing
so leads to p = ρ0 cv + O ε2 , which is like the relation according to linear theory.
This is not merely an interesting observation. Rather, it tells us that whatever we do
to make p uniformly valid will do the same for v.
We are now ready to introduce the coordinate straining transformation. We sub-
stitute the expression for x in Eq. (13.3.66) into the preceding equation for p. The
O (ε) term is expanded in a Taylor series in powers of ε, according to

x ξ x
Ẋ t − = Ẋ t − − ε F (ξ, t)
c c c

ξ x ξ
= Ẋ t − − ε F (ξ, t) Ẍ t − +··· (13.3.69)
c c c

Because O ε3 terms are not being considered, the Taylor series may be truncated
at the listed terms. For the same reason, we can simply replace x with ξ in the phase
of the O(ε2 ) part of φ. The result of these operations is

ξ x ξ
p = ερ0 c Ẋ t − − ε F (ξ, t) Ẍ t −
c c c (13.3.70)
β ξ ξ 2
+ ε ρ0 x Ẋ t −
2
Ẍ t − +O ε
c c c
The role of the F function is to make the perturbation series regular. Thus, we
select this function to cancel the growing term, that is

ξ ξ βξ ξ ξ
ρ0 cε −ε F (ξ, t) Ẍ t − + ε ρ0
2
Ẋ t − Ẍ t − =0
c c c c c
(13.3.71)
In other words
β ξ
F (ξ, t) = Ẋ t − (13.3.72)
c c
The assembly of the results is

p = ρ0 cv + O ε2 = ρ0 cεẊ (τ ) + O ε2
ξ x (13.3.73)
τ = t − , x = ξ + εβ Ẋ (τ )
c c
The second-order terms that remain in the expression for p are those in Eq. (13.3.68)
that do not grow with increasing x. They represent effects that always are small
relative to the overall signal and therefore seldom are of interest. Some individuals
refer to the process of removing the growing terms as a renormalization of the
perturbation series. (“Normal” here refers to regularity, as opposed to singularity.)
Is the solution consistent with the Riemann solution? To show this to be true, we
write the last of Eq. (13.3.73) as

Ẋ (τ )
ξ = x 1 − εβ (13.3.74)
c
Correspondingly, the first and second of Eq. (13.3.73) are

p = ρ0 cv = ρ0 cεẊ (τ ) + O ε2

x Ẋ (τ ) x p (13.3.75)
τ =t− 1 − εβ =t− 1−β
c c c ρ0 c2
For comparison recall Eq. (13.1.37), which is the small-signal approximation of the
Riemann solution. The term containing p in the denominator is small compared to
one, so the denominator may be expanded in a binomial series. Doing so leads to

x x p 2
p = ρ0 cẊ (τ ) , τ = t − =t− 1−β +O ε
c 1 + βp/ ρ0 c2 c ρ0 c2
(13.3.76)
For second order, this is the same as the expressions derived by the method of renor-
malization. It is useful to have this confirmation of the renormalization formulation,
but for more general systems any confirmation must come from experiments.
Before we consider such generalizations, it is appropriate that we consider another
singular perturbation technique that has been employed for nonlinear acoustics analy-
ses. The method of multiple scales had been used in other areas for many years prior
to Nayfeh and Kluwick’s application17 of the method to analyze a finite amplitude
plane wave. It recognizes that the spatial scale over which nonlinear effects become
important is much greater than the scale of a wavelength. Each scale is defined to be
an independent spatial variable ξn that changes more slowly as n increases. For the
nonlinear wave equation, these variables are
ξn = εn x, n = 0, 1, ... (13.3.77)
The velocity potential is considered to depend on each of these scaled variables, even
though the value of each coordinate is set when x and ε are set. Thus, a derivative
with respect to x becomes
∂φ ∂φ ∂φ ∂φ
= + ε2 + ε2 +··· (13.3.78)
∂x ∂ξ0 ∂ξ1 ∂ξ2
17 A.H. Nayfeh and A. Kluwick, “A comparison of three perturbation methods for nonlinear waves,”
J. Sound and Vibration (1976) pp. 293–299.

This transformation is introduced to the nonlinear wave equation and the boundary
condition, after which the velocity potential is expanded in the perturbation series in
Eq. (13.3.46). The first-order equations are
∂ 2 φ1 ∂ 2 φ1
c2 − =0
∂ξ02 ∂t 2
(13.3.79)
∂φ1
= Ẋ (t)
∂ξ
0 x=0
and the second-order equations are
∂ 2 φ2 ∂ 2 φ2 ∂ 2 φ1 ∂ 2 φ1 ∂φ1 ∂φ1 ∂ 2 φ1
c2 − = −2c2 + 2 (β − 1) +2
∂ξ0 2 ∂t 2 ∂ξ0 ∂ξ1 ∂ξ0 ∂t
2 ∂ξ0 ∂ξ0 ∂t

∂φ2 ∂φ2 ∂ 2 φ1
=− − X (t)
∂ξ0 x=0 ∂ξ0 x=0 ∂ξ02 x=0
(13.3.80)
Note that the second-order boundary condition accounts for the displaced location
of the boundary, even though we know that effect does not lead to growing nonlinear
effects.
Because we are not interested in φ terms that are O ε3 or smaller, only the
physical scale ξ0 and the slow scale ξ1 enter into the analysis. The first-order equations
have the same form as those for a regular perturbation, but they allow φ1 to have
arbitrary dependence on ξ1 . It is easier to satisfy boundary conditions after the general
solution has been identified. Hence, the first-order solution is written as
ξ0
φ1 = f t , ξ1 , t = t − (13.3.81)
c
This form is used to generate the second-order equations. Equation (13.3.57) may
be used to describe the quadratic terms, so the second-order wave equation becomes

∂ 2 φ2 ∂ 2 φ2 ∂ ∂ ∂ ∂ 2
c2 − = 2c f t , ξ1 + β f t , ξ1 (13.3.82)
∂ξ0 2 ∂t 2 ∂ξ1 ∂t ∂t ∂t
We know from the regular perturbation series analysis that the term whose coefficient
is β will lead to a particular solution for φ2 that is not uniformly valid. The role of
the first term on the right side is to prevent this condition from occurring. That is, we
set
∂ ∂ β ∂ ∂ 2
f t , ξ1 = − f t , ξ1 (13.3.83)
∂ξ1 ∂t c ∂t ∂t
Fulfillment of this condition represents the primary difficulty in using the method
of multiple scales. If considered in isolation, without prior knowledge of the nonlinear
behavior of a plane wave, solution of this equation would not be a simple task. The
result, after it is made to satisfy the first-order boundary condition, is
β
φ1 = −cX (τ ) + ξ1 Ẋ (τ )2
2c (13.3.84)
β
t = τ − 2 ξ1 Ẋ (τ )
c
From here, expressions for the pressure and particle velocity are obtained by
differentiation of this function. We shall not pursue the details because the result will
be equivalent to the Riemann solution. The primary motivation for this discussion
of the method of multiple scales is to demonstrate that it will be substantially more
difficult to implement than the method of strained coordinates. Such is especially
the case when we consider situations where a signal depends on more than one
spatial coordinate. Nevertheless, multiple scales method is a more general method
than strained coordinates, so it is conceivable that only it would be suitable in some
as yet unexplored situation.
13.4 Multidimensional Systems
Most linear acoustic systems of practical interest feature signals that are not truly pla-
nar waves. Higher-order modes in a rectangular waveguide are best characterized by
Cartesian coordinates, whereas the signal within a cylindrical waveguide or radiated
by a circular piston is best described in cylindrical coordinates. In the latter systems,
the axial direction of the coordinate system would be the direction of propagation,
whereas radiation from a cylinder features cylindrical coordinates with propagation
in the transverse direction. Spherical coordinates are appropriate for radiation from
a vibrating sphere. Nonlinear effects arise in each of these systems. Each has been
successfully researched, some by analytical techniques, some by computational tech-
niques, and some by both approaches. The scope of the investigations is too vast to
survey here. An excellent starting point is the monograph edited by Hamilton and
Blackstock,18 but even that work is merely an entry point to an enormous body of
research. Our objective here is not to delve deeply into specific topics. Rather, we
will consider a few systems that highlight methods and phenomena not seen in plane
waves.
13.4.1 Finite Amplitude Spherical Wave
The choice of coordinate system to use for the analysis of any wave usually is dictated
by the shape of the boundary. Here, we shall study how nonlinearity affects the prop-
agation of a radially symmetric spherical wave. The signal is generated by a spherical
body whose radius oscillates relative to a mean value a, such that r = a + εY (t). The
18 M.F. Hamilton & D.T. Blackstock, Nonlinear Acoustics, eds., Acoustical Society of America
(2008).
magnitude of the radial velocity εẎ is much less than the speed of sound c, but it is
sufficiently large to warrant consideration of nonlinear effects. In a sense, the signal
is not a multidimensional wave because it depends on a single spatial coordinate, the
radial distance r. Indeed, the basic line of investigation we shall pursue parallels the
perturbation analysis of plane waves. However, dependence on the spherical coor-
dinate r rather than the Cartesian coordinate x considerably complicates each stage
of the analysis, and these complications are typical of those encountered in formula-
tions using more general curvilinear coordinates. It is important to note that a much
simpler analysis of spherical waves is available.19 The analysis here will show how
the method of strained coordinates may be employed when simpler approaches are
not available.
The radiated signal is radially symmetric, so φ is a function of radial distance r
and time t. The perturbation series for the velocity potential is
φ = εφ1 (r, t) + ε2 φ2 (r, t) (13.4.1)
The first-order equation derived from the quadratic nonlinear wave, Eq. (13.3.18), is
∂ 2 φ1
c 2 ∇ 2 φ1 − =0 (13.4.2)
∂t 2
We use this relation to replace ∇ 2 φ1 in the second-order equation, which thereby is

simplified to
2
∂ 2 φ2 ∂ (β − 1) ∂φ1
c ∇ φ2 −
2 2
= + ∇φ1 · ∇φ1 (13.4.3)
∂t 2 ∂t c2 ∂t
[Equations (13.4.2) and (13.4.3) have been highlighted because they apply in any
situation.] The operators in radially symmetric spherical coordinates are
1 ∂2 ∂φ
∇2φ = (rφ) , ∇φ = ēr (13.4.4)
r ∂r 2 ∂r
It is convenient to address satisfaction of the boundary condition at the surface,
r = a, after the general solution has been obtained. The radiation condition requires
that the signal be an outgoing wave, so we write the first-order solution as
a
φ1 = f t (13.4.5)
r
19 D.T. Blackstock, “On plane, spherical, and cylindrical waves of finite amplitude in lossless fluids,”
J. Acoust. Soc. Am., Vol. 36 (1966) pp. 217–219.

13.4 Multidimensional Systems 627
where the nature of the function f is not specified at this juncture. The phase variable
t is the retarded time, which accounts for the time required for a signal to travel from
the sphere’s surface to the field point, that is,
r−a
t = t − (13.4.6)
c
Because ∂( )/∂t ≡ ∂( )/∂t , substitution of the expression for φ1 into the second-
order wave equation leads to

1 2 ∂2 ∂2 ∂ βa2 ˙ 2
c (rφ2 ) − (rφ2 ) = 2 2 f t
r ∂r 2 ∂ (t )2 ∂t c r

2a2 ˙ a2 2
+ 3f t f t + 4f t (13.4.7)
cr r
A consequence of the interchangeability of derivatives with respect to t and t is that

an overdot will be used to denote differentiation with respect to the argument of f .
The preceding is a linear partial differential equation, so we may obtain its par-
ticular solution by superposing the contribution
of each term on the right side. Each
of those terms has the general form r −n Q t . Let us examine the possibility that the
particular solution has a similar form by considering rφ2 = zn , where

zn = g (r) G t (13.4.8)
Then, the left side of Eq. (13.4.7) would be

1 2 ∂2 ∂2 g (r) g (r)
c z n − z n = −2 Ġ t + G t (13.4.9)
r ∂r 2 ∂ (t )2 r r

If we set g (r) = 1/r, then Ġ t can be selected to match the first term within the
bracket of Eq. (13.4.7). However, doing so generates a term that is like the second
term in the bracket because g (r) = −1/r 2 . Suppose we add to zn another term
for which g (r) = 1/r 2 in order to match the second term in the bracket plus the
residual from the first term. Doing so generates a new residual term that resembles
the third term in the bracket because it gives g (r) = −2/r 3 . Addition of another
term to zn for which g (r) = 1/r 3 would handle the 1/r 3 terms to the right, but also
add another residual at 1/r 4 . The process continues ad infinitum, with each added zn
term selected to match the residual from the previous term. Hence, we conclude that
we may construct a particular solution as an infinite series,
∞
#
(rφ2 )p = gn (r) G n t
n=1 (13.4.10)
gn (r) = r −n
If we were
to
carry out the full analysis, we would be
led
to a recurrence relation
wherein Ġ n t for n > 1 is defined in terms of G n−1 t . Such an analysis would
be required if it were necessary to identify all parts of φ2 . Fortunately, this seldom
is the case because of the nature of the gn (r) functions. Integration of the second of
Eq. (13.4.10) leads to
1 1−n
g1 (r) = ln (r) , gn (r) = − r if n > 1 (13.4.11)
n−1
Hence, each term in Eq. (13.4.10) other than n = 1 represents a contribution to rφ2
that decays at least as rapidly as 1/r. Because rφ1 does not decay with increasing r,
all of those terms will be such that ε2 φ2 is much less than εφ1 everywhere.
The same is not true for the n = 1 term. Its contribution to rφ2 is proportional
to ln (r). Although logarithmic growth is relatively slow, there is some large value
of r that would lead to ε2 φ2 having the same order of magnitude as εφ1 . This term
represents the part that is responsible for the perturbation series not being uniformly
valid.
Our interest is identification of terms that describe cumulative growth of nonlinear
effects, so we modify Eq. (13.4.10) to be
r
1
rφ2 = a ln G1 t + O (13.4.10)
a r
where the introduction of the factor a will help avoid confusion regarding the
dimensionality of quantities. Substitution of this expression converts the left side
of Eq. (13.4.7) to

1 2 ∂2 ∂2 2ac
c (rφ )
2 p −
(rφ 2 p = − 2 Ġ 1 t
) (13.4.12)
r ∂r 2 ∂(t ) 2 r
We equate this term to the first term on the right side of Eq. (13.4.7), which leads to
2ac βa2 2
− 2
G 1 t = 2 f˙ t (13.4.13)
r c r
The terms on therightside of Eq. (13.4.7) that have not been matched would lead
to O (1/r) and O 1/r 2 contributions to rφ2 . Neither grows relative to rφ1 , so these
terms may be ignored. Thus, we have found that

βa2 r ˙ 2 1
(rφ2 )p = − ln f t + O (13.4.14)
2c3 a r
A corollary of this development is that there is no need to consider the comple-

mentary solution for φ2 because it is a solution of the linear wave equation. Such a
term cannot lead to ε2 φ2 being comparable to εφ1 . In turn, because the role of the
complementary solution is solely to satisfy the second-order boundary condition,
only the first-order condition need be satisfied. This is fortunate, because satisfying
the second-order boundary condition, possibly accounting for the moving boundary
that the surface is at r = a + εY , would require that we employ the full particular
solution in Eq. (13.4.10).
The velocity potential obtained by substitution of the first- and second-order veloc-
ity potentials, Eqs. (13.4.5) and (13.4.14), into the perturbation series is
βa2 r ˙ 2
rφ = εaf t − ε2 3 ln f t + O ε2 (13.4.15)
2c a
be L /T. The O (ε) term has those

3
As a check, note that the dimensionality of rφmust
dimensions if f is L /T, in which case the O ε also is consistent. Identification of
2 2
the coordinate straining

transformation
requires expressions for p and vr . Terms
that are uniformly O ε2 /r or less may be ignored, so we use p = −ρ0 ∂φ/∂t . In
combination with vr = ∂φ/∂r, we obtain
βa2 r ˙ ¨
rp = −ρ0 εaf˙ t + ρ0 ε2 3 ln f t f t + O ε2
c a
$ a a % r (13.4.16)
βa 2
rvr = −ε f˙ t t̃ + f t + ε2 4 ln f˙ t f¨ t + O ε2
c r c a
The loss of uniform validity for rp and rvr is caused by a term that grows as
ln (r/a) relative to the O (ε) terms, so we shall strain the r coordinate with a growth
factor that is logarithmic. Thus, we try
r
r = ξ + εa ln R (13.4.17)
a
where R is an unknown function to be determined. The coordinate straining converts
the retarded time to
r−a a r
t ≡ t − = τ − ε ln R (13.4.18)
c c a
The variable τ is a nonlinear retarded time. Its definition is
ξ−a
τ =t− (13.4.19)
c
The task now is to determine the function R that makes the expressions for p and vr
uniformly valid. Toward that end, Eq. (13.4.18) is used to replace t̃ as the argument
of the F and Ḟ functions. Taylor series expansions give
a r
f t = f (τ ) − ε ln Rf˙ (τ ) + O ε2
c a
a r (13.4.20)
f˙ t = f˙ (τ ) − ε ln Rf¨ (τ ) + O ε2
c a
We substitute the transformation in Eq. (13.4.18) into Eq. (13.4.16). The result for p
is $ a r %
rp = − ρ0 εa f˙ (τ ) − ε ln Rf¨ (τ )
c a
2 (13.4.21)
βa 2
r ˙ ε a
+ ρ0 ε2 3 ln f (τ ) f¨ (τ ) + O
c a r
Similar operations for vr give

ε $ a r %
rvr = − a f˙ (τ ) − ε ln Rf¨ (τ )
c c a
a$ a r %
−ε f (τ ) − ε ln Rf˙ (τ ) (13.4.22)
r c a
βa 2
r ˙
+ ε2 4 ln f (τ ) f¨ (τ ) + O ε2
c a
The role of R is to cancel all terms in rp that have the factor ε2 ln (r/a). This
condition occurs if
β
R = − 2 f˙ (τ ) (13.4.23)
c
The pressure term that remains is
rp = − ρ0 εaf˙ (τ ) (13.4.24)
The expression for rvr corresponding to the selection of R is
εa $ ˙ c % a2 r ˙
rvr = − f (τ ) + f (τ ) − ε2 β 3 ln f (τ )2 + O ε2 (13.4.25)
c r c r a
The presence of ln (r/a) as a factor in the O(ε2 ) term might seem to mean that this
expression is not uniformly valid. However, that factor is divided by r and the limit
as r → ∞ of ln (r/a) /r is zero.
At this juncture, we have not addressed the determination of the f function. It is
somewhat less confusing to carry out that task if we work with a function that is
dimensionless. Accordingly, we let
f (t) = −c2 F (t) (13.4.26)
This substitution converts the coordinate straining transformation to

r
r = ξ + εβa ln Ḟ (τ ) (13.4.27)
a
The corresponding nonlinear retarded time is
r−a a r
t ≡ t − = τ − εβ ln Ḟ (τ ) (13.4.28)
c c a
The corresponding uniformly valid expressions for the state variables are
2
a ε a
p = ρ0 c ε Ḟ (τ ) + O
2
r r
a$ c % r a2 2
ε a
vr = εc Ḟ (τ ) + F (τ ) + ε cβ ln
2
2
Ḟ (τ ) 2
+ O
r r a r r
(13.4.29)
Aside from identification of τ dependence of F, the pressure has been determined. To

see that this is so, suppose we specify r and t, which sets t − r/c. Then, Eq. (13.4.28) is
an algebraic equation for τ . Substitution of the value of τ into the first of Eq. (13.4.29)
gives the value of p at that r and t.
Determination of the F function is done by satisfying the boundary condition on
the sphere. At r = a, the coordinate straining gives τ = t. To be consistent with the
notation for plane waves, we let the specified velocity on the boundary be εcV (τ ).
Velocity continuity at the surface requires that
vr |r=a = εcV (t) (13.4.30)
at r = a. We only need to substitute

Equation (13.4.29) describes the particle velocity
the O (ε) terms, because the terms that are O ε2 will merely lead to small corrections
of F (τ ). Thus, the substitution leads to a first-order differential equation for F (τ ),
c
Ḟ (τ ) + F (τ ) = V (τ ) (13.4.31)
a
This is the same differential equation as that which must be solved in linear theory
when the radial velocity is known at r = a. Indeed, the expressions for p is the same
as the description of the pressure for an outgoing wave according to linear theory,
see Eq. (6.2.5), except that the retarded time has been replaced by τ .
Before we proceed to an example, let us consider some qualitative aspects.
Because nonlinear effects become progressively more significant at large r, it is
reasonable to wonder whether they only occur in the farfield. A convenient ref-
erence distance marking where nonlinear effects are significant is that at which a
shock forms. This is the closest location at which ∂t/∂τ = 0 occurs at any instant.
Differentiation of Eq. (13.4.28) gives
∂t a rshock
= 1 − εβ ln F̈ (τ ) = 0 (13.4.32)
∂τ c a
from which we find that

r c/a
shock
ln = (13.4.33)
a εβ max F̈
The farfield of a spherical wave is characterized by p being essentially ρ0 cvr .

To identify where this condition occurs
let T be a timescale representing the rate
at which F changes, that is, O Ḟ = O (|F|) /T . We can assert from Eq. (13.4.29)
that p ≈ ρ0 cvr if O (|F| /T ) (c/r) O (|F|), which is equivalent to r cT . (If
a cT , then the farfield condition occurs everywhere.) Thus, the shock formation
distance is inversely proportional to the maximum rate at which the pressure changes,
whereas the farfield distance is independent of the pressure amplitude. This obser-
vation suggests that it might be possible to adjust the pressure to attain a shock at
any range.
This hypothesis can be tested by considering the steady-state harmonic pres-
sure corresponding to setting f˙ (t) = sin (ωτ ). The steady-state nonlinear signal is
described by a
p = ερ0 c2 sin (ωτ ) (13.4.34)
r
An appropriate reference time is 1/ω, so the farfield occurs at kr 1. The maximum
value of f¨ is ω, so the shock formation distance in this case is
rshock = ae1/(εβka) (13.4.35)
If we arbitrarily define the farfield to occur when kr > 10, we obtain
rshock ka 1/(εp βka)

= e (13.4.36)
rff 10
The value of ε is much less than one and β is O (1). Hence, it is evident that rshock
will always be much greater than rff . For example, a large value of ε in air is 0.001,
in which case the above expression leads to rshock > 170rff for ka < 100. A corollary
of rshock being much greater than rff is that wherever a spherical wave is significantly
affected by nonlinearity, it will be true that p ≈ ρ0 cvr .
Although the solution for a nonlinear spherical wave appears to be different from
the Riemann solution, they actually are quite similar. To demonstrate the resem-
blance, we note that the first of Eq. (13.4.29) gives Ḟ (τ ) = (r/a) p/ ρ0 c2 . Substi-
tution of this expression into Eq. (13.4.28) shows that the nonlinear retarded time is

r−a r r p
τ =t− + β ln (13.4.37)
c c a ρ0 c2
For comparison, the characteristic variable for a weakly nonlinear plane wave was
written in Eq. (13.3.76) as

x x p
τ =t− +β + O ε2 (13.4.38)
c c ρ0 c2
The quantity (r − a) /c in Eq. (13.4.37) is the time required for a linear signal to
propagate from the surface to the field point. This time is analogous to x/c, which
is the time required for a linear plane wave to travel from the boundary at which it
is generated. One beneficial aspect of this similarity of the descriptions is that it is
possible to adapt the earlier Fourier series analysis to handle situations where the
boundary excitation is time-periodic. The adaptation is based on the fact that time
occurs in the same manner in Eqs. (13.4.37) and (13.4.38). The implementation of
this notion is the topic of Exercise 13.24.
The difference between linearand nonlinear
retarded times for a spherical wave is
p/2 ρ0 c . In contrast, this difference for a planar wave
proportional to ln (r/a) (r/c) 2
is proportional to (x/c) p/ ρ0 c . To assess the relative strength of the two effects,

it is essential to bear in mind that spherical spreading leads to r |p| being constant.
Accordingly, we write the ratio of nonlinear effects as

Spherical nonlinear term ln (r/a) (r/c) pspher a ln (r/a) rpspher /a
= ≡
Planar nonlinear term (x/c) pplanar xpplanar
(13.4.39)

If pspher at r = a has the same order of magnitude as p planar , then rpspher /a will
have that magnitude at all r. This reduces the ratio of terms to ln (r/a) / (x/a). If
r and x both equal some value a, then this ratio is much less than one. This
observation leads to the conclusion that the nonlinear effect for a spherical wave is
much weaker than it is for a plane wave.
Either of the computational algorithms in Sect. 13.1.5 for evaluating profiles and
waveforms of a planar signal may be adapted with small modification to evaluate
spherical waves. For example, if we select a value of τ , we may find p from the
first of Eq. (13.4.29). Then, Eq. (13.4.37) may be solved algebraically for t if r is
specified. Solution for r at a specified t is more difficult because Eq. (13.4.37) is a
transcendental function of r. For the same reason, determination of p at specified r
and t requires more effort. Like the case of planar waves, a nonlinear equation solver
would be needed.
EXAMPLE 13.7 It is desired to compare the effect of nonlinearity in planar

and spherical waves. Toward that end, consider a boundary motion in which
the velocity normal to the surface is εc [sin (ωt) + (1/3) sin (3ωt)]. For the
spherical wave, this is vr at r = a, whereas it is vx at x = 0 for a plane wave
propagating in the x direction. Parameters are ε = 0.002, f = 12 kHz, and
a = 400 mm. The fluid is water, for which c = 1480 m/s and β = 3.5. Deter-
mine the steady-state planar waveform at x = xshock . Compare that result to the
waveform of a spherical wave at r = xshock and at r = rshock .
Significance
Evaluation of the equations describing a nonlinear spherical wave is the primary
emphasis. The results will vividly demonstrate the much slower advent of nonlinear
effects that results from spherical spreading, as well as some surprising similarities
to nonlinear distortion of plane waves.
Solution
The function F is obtained by solving the linear differential equation that is the
boundary condition, Eq. (13.4.31). The specified boundary velocity is

1 εc iωt 1 3iωt
vr = εcV = εc sin (ωt) + sin (3ωt) = e + e + c.c.
3 2i 3
The corresponding differential equation is

c 1 1 3iωτ
Ḟ (τ ) + F (τ ) = e + e
iωτ
+ c.c. (1)
a 2i 3
The form of the steady-state solution of Eq. (1) is
1 iωτ
F (τ ) = B1 e + B3 e3iωτ + c.c. (2)
2i
Substitution of this ansatz into Eq. (1) leads to expressions for the coefficients, which
are found to be
a/c a/c
B1 = , B3 =
1 + ika 3 + 9ika
where k = ω/c, as always. It is useful to work with a nondimensional version of Ḟ,

which is defined according to

a 1/i iωτ 1/i
F (τ ) = Re e + e3iωτ (3)
c 1 + ika 3 + 9ika
Determination of F allows us to finalize the expressions to evaluate. Equation

(13.4.28) becomes

r−a a r ka ka
t− = τ − εβ ln Re e +
iωτ
e3iωτ
(4)
c c a 1 + ika 1 + 3ika
The pressure resulting from substitution of Eq. (2) into the first of Eq. (13.4.29) is

a a ka ka
p = ρ0 c2 ε Ḟ (τ ) = ρ0 c2 ε Re eiωτ + e3iωτ (5)
r r 1 + ika 1 + 3ika
The radial distance at which an arbitrary wave forms a shock is given by

Eq. (13.4.33). In terms of the current functional representation, that expression is
r c3 1
shock
ln = ≡ (6)
a εβa max −F̈ (τ ) εβka max (D (τ ))
where
i 3i
D (τ ) = Re i eiωτ + e3iωτ
1 + ika 1 + 3ika
We could determine the maximum value of D (τ ) by seeking the value of ωτ at

which Ḋ = 0. Such an evaluation is complicated by the fact that there is more than
one local maximum. A simpler approach is to evaluate D (τ ) for closely spaced
values of ωτ ranging over a period. The specified parameters give ka = 20.38. A
data scan gives max (D (τ )) = 1.9978 at ωτ / (2π) = 0.9984. The shock formation
distance obtained from Eq. (6) is rshock = 13.37 m.
We are now ready to evaluate waveforms. There is no requirement to compute
the pressure values at a uniform time increment. Consequently, we may follow the
simpler procedure in which we select ωτn at equal increments in the range of a period
2π. The corresponding value of pn is obtained from Eq. (5). The instant tn at which
pn occurs at a specified r is found from Eq. (4), which gives
r
ωtn = kr − ka + ωτ − εβka ln Ḟ (τ ) (7)
a
The plane wave also must be evaluated. The velocity at x = 0 in the Riemann
solution was written as εcV (t). The function V (t) in the present situation is

vx 1 1 3iωt
V (τ ) = = e + e
iωt
+ c.c.
εc x=0 2i 3
The Riemann solution for the corresponding plane wave pressure is

1 iωτ 1 3iωτ kx
p = ερ0 c Re e + e
2
, ωt = ωτ +
i 3i 1 + εβV (τ )
The shock formation distance for a plane wave is given by Eq. (13.2.26). For the
present V (τ ), the maximum value of dV /dτ is 2ω at ωτ = 0. This leads to shock
formation of the plane wave at
c
xshock = = 1.402 m
2βεω
All computations may be carried out in a vectorial manner. Figure 1 compares

the boundary velocity V (τ ) to the pressure function Ḟ (τ ). Both are very close, and
suggestive of a square wave. Both functions differ little because ka is quite large.
This means that the spherical wave exhibits its farfield behavior everywhere, so that
p ≈ ρ0 cvr at the surface of the sphere.
1 .
F()
0.5 V()
−0.5
−1
0 0.5 1 1.5 2

Figure 1.
The spherical waveforms at r = x shock and r = rshock in Fig. 2 have been delayed by
the respective value of (r − a) /c, in order to place them in a common time base. It is
difficult to compare them because of their vastly different magnitude. But it appears
that the spherical wave is essentially undistorted at r = xshock . This is not surprising
because rshock xshock .
r=xshock r=rshock
0.2
p/(0c2)
0.1
0
−0.1
−0.2
0 0.5 1 1.5 2
Retarded time [t-(r-a)/c]
Figure 2.
Multiplying each pressure waveform in Fig. 2 by their respective value of r/a

makes it possible to compare them and the plane wave at xshock . This is the data in
Fig. 3. Other than the spherical spreading factor, there is no perceptible difference
between the spherical and plane waves at their respective shock formation distance.
This behavior might seem to be surprising. It is attributable to the fact that ln (r/a)
for a spherical wave and x/xshock for a plane wave are merely factors that scale the
difference between the nonlinear and linear retarded times.
p @ r=xshock p @ r=rshock Plane wave p @ x=xshock
1
scaled p/(0c2)
0.5
−0.5
−1
0 0.5 1 1.5 2
Retarded time [t-(r-a)/c]
Figure 3.
To close this discussion, it is instructive to consider what would happen if ka were

much smaller. Reducing the frequency
by a factor of ten with all other parameters
constant leads to rshock = 21.7 1015 m. On the other hand, lowering the frequency
by a factor of ten and raising the acoustic Mach number by a factor of ten change the
shock formation distance by a relatively small amount, to rshock = 18.86 m. These
trends show that nonlinear effects for a spherical wave are relatively small unless
either the frequency, as measured by ka, or the acoustic Mach number is very large.
13.4.2 Waves in Cartesian Coordinates
There is no variation of pressure along the wavefronts of plane waves and radi-
ally symmetric spherical waves. Variation of pressure transverse to the direction of
propagation gives rise to a phenomenon we have not yet encountered, specifically
deviation of rays due to the signal associated with those rays. The system we shall
investigate here exhibits this phenomenon, but it is somewhat academic. It concerns
radiation from an infinite planar boundary, which is unrealizable. Nevertheless, with
some modifications the solution can be used to describe the nonlinear propagation
of signals in a hard-walled waveguide.20
Figure 13.21 depicts the system we will investigate. A flat plate extends infinitely
in both directions. It is the boundary of the fluid in the infinite half-space above it.
Knife-edge supports running perpendicularly to the plane of the diagram are spaced
equidistantly at distance L. The plate vibrates in the direction normal to its surface,
with the displacement w defined to be positive into the fluid, which is the sense of
increasing z. The displacement must be zero at its supports. The longest possible
half-wavelength is L, so the displacement is set to
εc πx
w= sin (ωt) sin (13.4.40)
ω L
(Shorter wavelengths along the plate may be studied by replacing L with L/n, n =
1, 2, ... throughout the analysis.) We shall perform a steady-state analysis, so w is
presumed to have existed for a long time prior to t = 0.
L L w(x,t)
x
Fig. 13.21 A periodically supported vibrating plate, whose rest position is the plane z = 0, gener-
ates a two-dimensional acoustic field
20 H.C.Miao and J.H. Ginsberg, “Finite amplitude distortion and dispersion of a nonplanar mode
in a waveguide,” J. Acoust. Soc. Amer., Vol. 80 (1986), pp. (911–920).
The task before us is to find a solution of the nonlinear wave equation such that
the particle velocity and the velocity of the plate have equal components normal
to the plate’s surface. Another condition is that the acoustic signal must satisfy
the Sommerfeld radiation condition, which requires that the pressure and particle
velocity be outgoing or evanescent waves at large z. A further requirement follows
from the fact that Fig. 13.21 does not change if we shift the coordinate system by
any multiple of 2L in either sense of the x direction. Consequently, the solution must
have period 2L in the x direction.
The condition of continuity of particle velocity at the fluid/plate interface consti-
tutes a moving boundary condition. Eventually, the analysis will lead to a simplified
boundary condition, as it did for plane and spherical waves. However, such an approx-
imation might not be warranted in other systems, or in situations where nonlinear
effects near the boundary are of interest. Hence, let us consider the “exact” condition
for the plate. Figure 13.22 zooms in on a small portion of the surface at an arbitrary
instant.
z 
n(x,t) 
w(x,t)
x
Fig. 13.22 Formulation of the continuity condition of particle velocity for fluid at the interface
with a vibrating plate. The condition applies at the current position of the plate in the direction n̄
normal to the current tangent to the plate
Figure 13.22 represents the displaced plate as the surface defined by z = w (x, t).
The normal to this surface is n̄ = − sin χēx + cos χēz , where χ is the angle between
the normal to the plate and the z-axis. The particle velocity normal to the surface of
the plate at its current position must equal the normal velocity of the plate at that
position. In terms of the velocity potential, this condition is stated as
∂w
n̄ · ∇φ|z=w = n̄ · ēz (13.4.41)
∂t
Figure 13.22 shows that χ also is the angle between the tangent to the surface in the xz-
plane and the x-axis, so χ = tan−1 (∂w/∂x). The displacement is O (ε), so |χ| 1.
Therefore, cos χ ≈ 1 and sin χ ≈ tan χ ≈ ∂w/∂x, so n̄ = − (∂w/∂x) ēx + ēz . When
the gradient is replaced by the first two terms of a Taylor series relative to z = 0, the
boundary condition becomes

∂w ∂ ∂w
− ēx + ēz · ∇φ|z=0 + w ∇φ| = (13.4.42)
∂x ∂z ∂t
z=0 z=0

The result of expanding this expression and dropping terms that are O ε3 is
∂φ ∂w ∂φ ∂2φ ∂w
− +w 2 = at z = 0 (13.4.43)
∂z ∂x ∂x ∂z ∂t
The first term is the one that would be employed for a linear analysis, the second
compensates for the fact that vx has a normal component because n̄ is rotated from
the z-axis, and the third term compensates for the plate not being on the x-axis.
There is no variation of any feature of the system perpendicular to the plane of
Fig. 13.21, so the velocity potential is a function of x, z, and t. The perturbation series
is
φ = εφ1 (x, z, t) + ε2 φ2 (x, z, t) + · · · (13.4.44)
Equations (13.4.2) and (13.4.3) respectively govern the first and second-order
potentials.
The boundary conditions for the first and second-order potentials are found by
substituting Eqs. (13.4.40) and (13.4.44) into the moving boundary condition and then
matching like powers of ε. It is useful to use a complex exponential representation
to handle products of harmonic time functions. Thus, we write

∂φ1 ∂ w 1 iωt πx
= = ce sin + c.c.
∂z z=0 ∂t ε 2 L
∂φ2 ∂ w ∂φ1 w ∂ 2 φ (13.4.45)
1
= −
∂z z=0 ∂x ε ∂x z=0 ε ∂z2 z=0
The sinusoidal dependence on x on the boundary is like a higher-order mode in

a two-dimensional waveguide. This suggests that the general solution for φ1 has the
form πx
1
φ1 = Bei(ωt−κz) sin + c.c. (13.4.46)
2 L
This ansatz satisfies the first-order (linear) wave equation if the wavenumber is
1/2
π2
κ = k2 − 2 (13.4.47)
L
If kL < π, the pressure will decay exponentially with increasing z. Such a situation
is irrelevant to a study of nonlinearity at small ε because the pressure will decay to
insignificance before φ2 grows to a significant level. Thus, we shall only consider the
case where the wavelength in the z direction is less than 2L, that is, kL > π, which
means that κ is real. The coefficient B is found by satisfying the first of Eq. (13.4.45).
The resulting first-order potential is
πx
1 ic i(ωt−κz)
φ1 = e + c.c. sin + c.c. (13.4.48)
2 κ L
To determine the second-order potential, we substitute φ1 into the second-order

inhomogeneous wave equation, Eq. (13.4.3), which becomes

∂ 2 φ2 1 ∂ (β − 1) ωc i(ωt−κz) 2 πx 2
c 2 ∇ 2 φ2 − = − e + c.c. sin
∂t 2 4 ∂t c2 κ L
2
icπ i(ωt−κz) πx 2
+ e + c.c. cos
κL L

2 πx 2
+ cei(ωt−κz) + c.c. sin (13.4.49)
L
The product of the complex exponential and its complex conjugate is a constant.
There is no need to track such terms because each square is differentiated with
respect to time. For the remaining terms we use the formulas for the cosine of a
double angle to handle the x dependence. Doing so leads to
(
∂ 2 φ2 1 ∂ 2i(ωt−κz) (β − 1) ω
2
2πx
c ∇ φ2 −
2 2
= e +c 2
1 − cos
∂t 2 8 ∂t κ2 L
2 2 +
π c 2πx
− 2 2 1 + cos + c.c.
κ L L
(13.4.50)
The definition of κ in Eq. (13.4.47) reduces the second-order equation to
(
∂ 2 φ2 iω 2i(ωt−κz) π 2 c2
c 2 ∇ 2 φ2 − = e ω 2
β − 2
∂t 2 4κ2 L2
+ (13.4.51)
2πx
−ω 2 β cos + c.c.
L
The second term is a solution of the linear wave equation. We can generate a
particular solution by multiplying its form by x, z, and/or t. Other considerations
dictate which of these alternatives should be chosen. Specifically, we know that the
pressure must be periodic in x and t because shifting t by 2π/ω or x by 2π/L does
not alter the excitation. Factors of t or x multiplying sinusoidal terms would violate
these periodicity requirements. Although the presence of a z factor in φ2 might seem
to violate the Sommerfeld radiation condition, the intent is to identify terms that lead
to nonuniform validity. Thus, we shall try as a particular solution

2πx
(φ2 )part = B0 + B1 z cos e2i(ωt−κz) + c.c. (13.4.52)
L
The coefficients are found by substituting this expression into Eq. (13.4.51) and then
matching like terms. The result is

ick βk 2 − 2π 2 /L 2 c k3
B0 = , B1 = β (13.4.53)
16κ2 k −κ
2 2 16 κ3
The B1 term leads to growth of φ2 relative to φ1 as z increases, whereas the

B0 term remains bounded at all z. By definition, (φ2 )comp is a solution of the
(homogeneous) linear wave equation. As such, it too is bounded and therefore repre-
sents a contribution to the second-order potential that does not grow. For this reason,
we need not determine it. (This is fortunate because substitution of w and φ1 into the
second of Eq. (13.4.45) would generate several terms that would need to be matched.)
Consequently, the solution for the velocity potential reduces to
πx
ic cβ k3 2πx
φ = ε ei(ωt−κz) sin + ε2 2i(ωt−κz)
ze cos + c.c. + O ε2
2κ L 16 κ3 L
(13.4.54)
Although we could work with this representation, the presence of complex con-
jugate terms would complicate the process of identifying the coordinate straining
transformation. The equivalent real form is
c πx
φ = −ε sin (ωt − κz) sin
κ 3 L
cβ k 2πx 2 (13.4.55)
+ ε2 z cos (2ωt − 2κz) cos + O ε
8 κ3 L

As always, O ε2 should be understood to denote terms that everywhere have that
order of magnitude.
The nonuniform validity of this expression is not addressed directly, because the
velocity potential is not a state variable for the state of the fluid. The fact that φ is
not uniformly valid is passed on to the pressure and particle velocity, which are state
variables. However, the growth factor will be different. This is a two-dimensional
system, so there are two particle velocity components, given by
∂φ $ π πx
vx = = εc − sin (ωt − κz) cos
∂x κL L
β π k3 2πx
−ε z cos (2ωt − 2κz) sin + O ε2
4 L κ 3 L
∂φ $ πx
vz = = εc cos (ωt − κz) sin
∂z L
β k3 2πx
+ε z sin (2ωt − 2κz) cos + O ε2
4 κ 2 L
(13.4.56)
The pressure may be evaluated according to p = −ρ0 ∂φ/∂t because the quadratic
terms in Eq. (13.3.20), being formed from φ1 , would merely add O ε2 terms that
time only occurs in φ as ωt − κz,
do not grow relative to the O (ε) terms. Because
we have (∂φ/∂t) /ω = − (∂φ/∂x) /κ + O ε2 , from which it follows that

p = (ρ0 ω/κ) vz + O ε2 (13.4.57)
The descriptions of p, vx , and vz that have been derived are not uniformly valid.
However, because p is proportional to vz , it is only necessary to find a coordinate
straining transformation by considering vx and either vz or p. Both x and z should
be transformed because two state variables require correction, and the second-order
term for vx and vz is different. A different perspective is that the velocity components
are comparable in magnitude, so we should anticipate that the spatial grid should be
strained comparably in both directions. The growth factor is proportional to z, so let
us try
x = η + εzF (η, ξ, t)
(13.4.58)
z = ξ + εzG (η, ξ, t)
We substitute the coordinate transformation into the real forms of the state vari-
ables, which gives
vx π $π %
=− sin [ωt − κ (ξ + εzG (η, ξ, t))] cos (η + εzF (η, ξ, t))
εc κL L

β π k3 2πη
−ε z cos (2ωt − 2κξ) sin + O (ε)
4 L κ3 L
(13.4.59a)
$π %
vz κ p
= = cos [ωt − κ (ξ + εzG (η, ξ, t))] sin (η + εzF (η, ξ, t))
εc k ερ0 c2 L

βk 3
2πη
+ε z sin (2ωt − 2κξ) cos + O (ε)
4 κ2 L
(13.4.59b)
The terms containing F or G are expanded in Taylor series that truncated at O (ε).
are
In addition, the double angle formulas are applied to the O ε2 z terms,
vx π $ πη
=− sin (ωt − κξ) cos
εc κL Lπη
− (κεzG) cos (ωt − κξ) cos
π L %
πη
− εzF sin (ωt − κξ) sin
L L

β π k3 &
−ε z cos (ωt − κξ)2
2 L κ3
' πη πη
− sin (ωt − κξ)2 sin cos + O ε2
L L
vz κ p πη πη
= = cos (ωt − κξ) sin + (κεzG) sin (ωt − κξ) sin
εc k ερ0 c2 L L
π πη
+ εzF cos (ωt − κξ) cos
L L
β k3
+ε z sin (ωt − κξ) cos(ωt
2 κ2
πη 2 πη 2
− κξ) cos − sin + O ε2 (13.4.60)
L L
The task now is to determine F and G functions that remove all terms in both vx
and vz that have z as a factor. A clue suggesting the appropriate choice comes from
recognition that in both expressions, the sinusoidal factors of G are shifted by 90◦
relative to those for F. This suggests that G also should be 90◦ out-of-phase relative
to F in both its ξ and η dependencies. Each should match the nonuniformly valid
term derived from φ. Thus, we try
πη
F = CF sin (ωt − κξ) cos
L
πη (13.4.61)
G = CG cos (ωt − κξ) sin
L
These forms are substituted into Eq. (13.4.60), and terms having like dependence on
ξ and η are grouped. The values of CF and CG are those that result in annihilation of
all terms that have z as a factor. This leads to
2
βkL k β k 3
CF = − , CG = (13.4.62)
2π κ 2 κ
The corresponding coordinate straining transformation is
πη
1 kL k 2
x = η − εβz sin (ωt − κξ) cos
2 π κ L
3 (13.4.63)
1 k πη
z = ξ + εβz cos (ωt − κξ) sin
2 κ L
The terms that remain in the state variables are
vx π πη
=− sin (ωt − κξ) cos + O ε2
εc κL L
πη (13.4.64)
vz κ p
= = cos (ωt − κξ) sin + O ε2
εc k ερ0 c 2 L
Evaluation of the pressure corresponding to specified x, z, and t requires that

Eq. (13.4.63) be solved for η and ξ. Such a determination will require numerical
methods because of the transcendental nature of the transformation. These matters
will be addressed in the next example. Here, we shall focus on the meaning of the
coordinate straining. The general concept of a coordinate straining transformation
is that it shifts the location where the linear signal is observed. The linear solution
is recovered by setting ε = 0 in Eq. (13.4.63), which leads to ξ = z and η = x in

Eq. (13.4.64). Thus, one way to view the nonlinear effect is to view lines of constant
ξ and constant η as a grid. Such a view is afforded by Fig. 13.23, which shows the
transformation grid at two instants separated by a quarter-period. (The value of ε
used there is quite large because doing so assists visualizations.)
t = 0 t = /2
6 6
t = 0
6.5
5 5
6
4 4
5.5
z/(2)
z/(2)
z/(2)
3 5 3
2 4.5 2
4
1 −1 −0.5 0 0.5 1 1
z/L
0 0
−1 0 1 −1 0 1
z/L z/L
Fig. 13.23 Instantaneous visualizations of the coordinate straining transformation for the nonlinear
signal radiated by a vibrating flat plate, ε = 0.005, kL = 3.628. Constant ξ lines are solid and red;
constant η lines are dashed and blue
The phase variable for the signal is
θ = ωt − κξ (13.4.65)
The coordinate transformation gives ξ = 0 at z = 0, so all signals on the wavefront

at specific values of t and ξ left the plate at t0 = t − (κ/ω) ξ. Thus, lines of constant
ξ are wavefronts. The transformation also states that η = x at z = 0. The signal
corresponding to specific ξ and η at time t left the plate from location x = η at time
t − (κ/ω) ξ. Thus, a line of constant η is the locus of locations at instant t of signals
that departed from x = η prior to time t. (These lines were mistakenly referred to as
rays in prior publications describing this system.21 The rays are identified later in this
discussion). Figure 13.23 shows that the constant ξ and η lines near the plate differ
little from the linear grid, z = η and ξ = η. The differences grow in an oscillatory
manner with increasing z, but the lines ξ = ±L/2 are straight.
21 J.H. Ginsberg, “A new viewpoint for the two-dimensional nonlinear acoustic wave radiating
from a harmonically vibrating flat plate,” J. Sound and Vib., Vol. 63 (1978) pp. 151–154. The
analysis of the plate problem by J.H. Ginsberg, “Perturbation Methods,” M.F. Hamilton and D.T.
Blackstock, eds., Nonlinear Acoustics, Acoustical Society of America (2008) Chap. 10 repeated the
misinterpretation.
Each of these behaviors has a relatively simple physical explanation. Note that
the combination of sinusoidal terms in each of Eq. (13.4.63) matches a velocity
component in Eq. (13.4.64). When we use this fact to eliminate the sinusoidal terms
in the coordinate straining, we see that
2
1 kL k vx
x = η + βz
2 π κ c
3 (13.4.66)
1 k vz
z = ξ + βz
2 κ c
A signal corresponding to specific ξ and η would occur at x = ξ and x = η in the

absence of nonlinearity. The nonlinear signal occurs at a location that is shifted in
the x direction by an amount that is proportional to its vx value. That location also
is shifted in the z direction by an amount that is proportional to its vz value. In other
words, the manner in which a signal propagates is altered by the signal. The constant
ξ lines in Fig. 13.23 are oscillatory because vz depends sinusoidally on η. Similarly,
the oscillatory nature of the constant η lines is due to the sinusoidal dependence of
vx on η. The lines η = ±L/2 are straight because vx is zero for signals that may be
traced back to the plate at = ±L/2.
Advancement and retardation of a wavefront is the amplitude dispersion process
encountered in plane and spherical waves. Indeed, if we set κ = k, the difference
between z and ξ is half the value for a plane wave. This process is weaker here because
vz fluctuates along the wavefront. Shifting of the signal in the direction transverse to
the propagation due to the signal’s particle velocity in that direction is the process
of self-refraction. It does not occur in plane waves. We will see in the next example
that the presence of both processes sometimes leads to waveform distortion that is
unlike what we have observed thus far.
The zoomed view in Fig. 13.23 shows that at large z the grid lines for different ξ
and η have regions where they intersect or are out of synchronization. For example,
a wavefront for a lesser value of ξ might be at a larger value of z. This, of course,
corresponds to multivaluedness of the signal. In other words, a shock is previously
formed as some distance closer to the boundary. One way of identifying a shock is to
extend the approach for plane waves in which the transformation is analyzed. Suppose
we look at the signal at a point x, z that corresponds to strained coordinates η, ξ. If
we increment the latter by differential amounts dη and dξ, we will obtain the signal
at x + dx, z + dz. Let us write Eq. (13.4.63) generically as x = fx (η, ξ, t, z) , x =
fx (η, ξ, t, z). Then the differential increments are related to
∂fx ∂fx
dx = dη + dξ
∂η ∂ξ
(13.4.67)
∂fz ∂fz
dz = dη + dξ
∂η ∂ξ
The matrix representation of these relations is

⎡ ⎤
∂fx ∂fx
( + ( +
dx ⎢ ∂η ∂ξ ⎥
=⎢ ⎥ dη (13.4.68)
dz ⎣ ∂fz ∂fz ⎦ dξ
∂η ∂ξ
The coefficient matrix is the Jacobian of the transformation. If its determinant is

zero, the transformation is not well-defined. This is the condition for the advent of
a shock. Expansion of the determinant and the formula for the cosine of the sum of
two angles show that either of the two conditions leads to the Jacobian being zero,
2
1 k π
εβ (kz) cos ωt − κξ + η = 1
2 κ2
L (13.4.69)
1 k π
εβ (kz) cos ωt − κξ − η = −1
2 κ L
Either relation in combination with Eq. (13.4.63) constitutes three equations relating
the values of η, ξ, x, at a specified t. Their simultaneous solution for fixed t defines
the locus of points at which a shock forms at that instant.
The views in Fig. 13.23 provide an instantaneous picture of the location of signals
that were radiated from the plate in the prior interval. In some respects, a more
meaningful question is: What is the path followed by the signal that radiated from
the plate at a specific instant? By definition, this path is a ray. All properties of the
signal on a ray other than its amplitude are constant. Consequently, a ray in the
present system is the locus of points at which a constant value of the phase variable
θ = ωt − κξ and strained coordinate η is observed as time elapses.
In linear theory, η = x and ξ = z. Constant θ in that case is observed at
z = (ωt − θ) /κ on a line of constant x. Thus, the rays of a linearized description are
perpendicular to the plate, and the signal propagates along the ray at phase speed
ω/κ. The situation is different when nonlinear effects are included. Identification of
rays and wavefronts begins by noting that the strained coordinate corresponding to
designated values of θ and t is ξ = (ωt − θ) /κ. This relation is used to eliminate ξ
from Eq. (13.4.63). Then the equation is solved for x and z as functions of θ, t, and
η. The second equation gives
ωt − θ
z= 3 πη
1 k
κ 1 − εβ cos (θ) sin
2 κ3 L (13.4.70)
3 πη
ωt − θ 1 k
= 1 + εβ cos (θ) sin + O ε2
κ 2 κ 3 L
This expression is used to eliminate z in the first of Eq. (13.4.63), with the result that
πη
ωt − θ kL k 2
x=η− εβ sin (θ) cos + O ε2 (13.4.71)
κ 2π κ L
To obtain a ray, we fix θ and η, and consider t to be a variable parameter. Equations

(13.4.70) and (13.4.71) give the x and z coordinates of the point at which this phase
is situated at a specific t. Both equations are linear in t, so it follows that a ray is a
straight line. The slope of this line relative to the z-axis is
πη
dx ∂x/∂t kL k2
= = −εβ sin (θ) cos + O ε2 (13.4.72)
dz ray ∂z/∂t 2π κ2 L
Self-refraction is evidenced here by the fact that the sinusoidal terms match vx in
Eq. (13.4.64), which leads to
2 2
dx 1 kL k vx
= β (13.4.73)
dz ray 2 π κ c
The value of ε is small, so the slope essentially is the angle. From this we conclude
that a ray is rotated from perpendicularity to the z = 0 plane by an amount that is
proportional to the particle velocity vx in the signal associated with that ray.
To determine the phase speed at which a signal propagates along its ray as t
elapses, we observe that if θ and η are constant, then Eq. (13.4.70) is an explicit
expression for z. Differentiating it gives
ω 3 πη
dz 1 k
= 1 + εβ cos (θ) sin (13.4.74)
dt constant θ and η κ 2 κ3 L
The actual phase velocity is parallel to the ray; the preceding is the z component
of that velocity. However the angle between the ray and the z-axis is O (ε), so the z
component is essentially the magnitude of the vector. Furthermore, the trigonometric
terms in the preceding are the same as those for vz in Eq. (13.4.64). This leads to the
description of the phase speed as
ω 1
3
k vz
cphase = 1+ β (13.4.75)
κ 2 κ3 c
If κ were the same as k and the half factor were not present, this would be the same
of the phase speed of a nonlinear planar wave.
An important corollary of this analysis is the observation that rays leaving a
specific point on the boundary at different instants do not coincide. This behavior is
unlike the properties of rays according to the linear version of geometrical acoustics
in which the rays are time invariant. Let us consider the rays that depart from a
specific location x on the plate’s surface. Because the plate displacement is O (ε),
this surface is essentially z = 0. Thus, θ/ω is the instant when the signal departed
from the plate. According to Eq. (13.4.64), vx = 0 for θ = 0, π, 2π, .... It follows
from Eq. (13.4.72) that the rays for these values of θ are parallel to the z-axis, as are
the rays obtained from linear theory. The maximum positive value of vx corresponds
to θ = π/2, 5π/2, .... Equation (13.4.72) indicates that the rays that depart from
location x on the plate at these instants undergo the greatest rotation to the right of
the z-axis. Conversely, θ = 3π/2, 7π/2, ... correspond to the largest negative vx . The
rays departing at these instants undergo the greatest rotation to the left of the z-axis.
Thus, within a period the rays rotate from one side to the other as time elapses.
An analogy explains the difference between this description of a ray and the nature
of lines of constant η in Fig. 13.23. Consider the stream of water that emerges from a
hose. Suppose the nozzle is rotated about the vertical axis in an oscillatory manner.
Any particle of water follows a straight line in the horizontal plane because the only
force acting on that particle is gravity. (Obviously, this is not true if there is a sub-
stantial wind.) This straight line is the ray. Further suppose that we take a photograph
looking down on the horizontal plane. The stream in such a photograph would be
situated along a line that resembles the oscillatory constant η line in Fig. 13.23. Both
views: the path followed by a particle and the instantaneous location of signals, are
meaningful, but the ray picture might be more revealing.
Another corollary of the analysis is the observation that rays for a specific
phase that depart from the boundary at the same time, but different x, are not
parallel. The maximum plate displacement |w| at any instant occurs at locations
where |sin (πx/L)| = 1. These are the midpoints between the locations where the
plate is supported. Because η = x at the boundary, these locations correspond to
cos (πη/L) = 0. The deviation angle for rays leaving these locations is ψ = 0 regard-
less of θ. In contrast, at the locations where the plate is supported, w = 0, which
means that sin (πx/L) = sin (πη/L) = 0. Hence, |cos (πη/L)| = 1 for these rays, so
the fan of rays that radiate from these points is the widest.
A description of rays sometimes is accompanied by a depiction of wavefronts.
Such a picture allows us to monitor the movement of the signal along its ray. We have
already constructed a wavefront by fixing ξ and t and allowing η to be a parameter for
the coordinate transformation. Here, we wish to construct wavefronts for a specified
θ at many instants. A systematic method for doing so is to treat Eqs. (13.4.70) and
(13.4.71) as a parametric description of a coordinate transformation with θ and t set,
and η considered to be a free parameter. Figure 13.24 displays these properties at four
instants separated by a quarter period. The response is periodic in the x direction, so
only one interval needs to be described.)
If θ is a multiple of π, the particle velocity in the x direction is zero. Consequently,
the rays departing from z = 0 at any x are perpendicular to the boundary. These
values of θ also correspond to vz having its maximum magnitude at a specified η, so
the wavefronts for these phases undergo the most distortion as they propagate. The
situation for θ being an odd multiple of π/2 is opposite. These phases correspond to
the magnitude of vx being its maximum, so the rays departing from z = 0 at any x
is at the extreme angle relative to the z-axis. Concurrently, these phases give vz = 0,
so these wavefronts remain straight as the signal propagates. Intermediate phases
feature rays that are rotated less than the extreme, and wavefronts that undergo less
than maximum distortion.
A consequence of the rays being rotated by different amounts is that they will inter-
sect. This is evidenced by the region of concentration along x = −L/2 for θ = π/2
and along x = L/2 for θ = 3π/2. The z value at which this condition occurs seems
to be close to the value in Fig. 13.24 at which the coordinate straining transformation
becomes multivalued. To interrogate the behavior of the rays, Fig. 13.25 zooms in on
the region along x = −L/2 for θ = π/2. This evaluation extends to a greater distance
from the plate, and the density of rays close x = 0 has been increased. Aside from
shifting the x values, the same picture would be obtained around the line x = 0.5L
for θ = 3π/2.
=0  = 0.5 =  = 1.5
4
z/(2)
0
−1 0 1 −1 0 1 −1 0 1 −1 0 1
x/L x/L x/L x/L
Fig. 13.24 Rays of the nonlinear signal radiated by a vibrating flat plate. The wavefronts are
depicted at intervals of one-quarter period. ε = 0.005, kL = 3.628
Fig. 13.25 Zoomed view of  = 0.5

a caustic formed by the rays
of the nonlinear signal 7.2
radiated by a vibrating flat
plate, ε = 0.005, kL = 3.628 7
6.8
6.6
6.4
6.2
−0.55 −0.5 −0.45
The enlarged figure shows that multiple rays at large angles do not have a common
intersection, which means that the convergence of rays is not a focus. Rather, it is
a caustic that results from the close intersection of rays at small angles relative to
the line x = −0.5L. This feature marks shock formation. Let us consider a range of
θ values. For θ = 0, all rays are parallel, so the caustic is infinitely distant. As θ is
increased, the caustic occurs closer to the plate somewhere along x = −L/2. The
caustic is closest to the plate when θ = π/2. Beyond that value, the caustic occurs
at increasing distance, eventually ceasing to exist for θ = π. Further increase of θ
from π to 2π repeats this behavior along x = 0.5L
We can derive an expression for the distance to the caustic. The derivation is based
on Fig. 13.26, which depicts a ray that intersects the line x = −0.5L. The ray departs
Fig. 13.26 Intersection of a z

ray departing from the plate
close to the ray departing
from x = −L/2; ε = 0.005, dx
kL = 3.628, θ = π/2
tan( dz )
ray
zint
x
L/2

from the plate at x = η. A caustic on x = −L/2 is generated by rays for which η < 0,
so the horizontal distance between the launch point and x = −0.5L is −η − L/2. This
is one side of a right triangle, with the opposite angle being tan−1 (dx/dz)ray , which
is given by Eq. (13.4.72). The other side of the right triangle is the distance zint to the
intersection of the ray with x = −L/2. From trigonometry, we have
−η − L/2 η + L/2
zint = −1 = πη (13.4.76)
tan tan (dx/dz)ray kL k2
εβ sin (θ) cos
2π κ2 L
The caustic occurs because the rays at small angles converge, and small angles
correspond to η being close to −L/2. The arrete, which is the minimum distance to
the caustic, is the limit as the angle goes to zero, that is, as η → −L/2. Thus
1
zarrete = lim (zint ) = (13.4.77)
η→−π/2 k k2
εβ sin (θ)
2 κ2
The shock formation distance is the minimum distance to the caustic at any instant,
which corresponds to sin(θ) = 1,
2 κ 2
zshock = min (zarrete ) = (13.4.78)
εβk k
For comparison, the plane wave shock formation distance for a harmonic excitation
is 1/ (βεk). For the parameters used to construct Fig. 13.26, the above equation gives
zsh / (2π/κ) = 6.631, which is slightly less than the distance at which two rays in
Fig. 13.25 intersect.
The motivation for the analyses of the coordinate straining grid, as well as of rays
and wavefronts, is to understand the processes that affect how a signal propagates.
The main concepts are the existence of the self-refraction phenomenon, as well as
the altered behavior of rays as properties that depend on the phase of the signal
that is transported along them. These occur concurrently with amplitude dispersion,
which is the process that is responsible for distortion of plane waves. However, these
observations are significant, and in some respects fascinating, we also are interested
in the nature of waveforms. Evaluation of the signal at specified x, z, and t requires
a quantitative solution of the coordinate straining transformation. The next example
addresses the procedure for doing so, as well as the effects of the nonlinear processes.
EXAMPLE 13.8 A plate of large extent forms the surface of a large body
of water. The plate is periodically supported at intervals of L = 1.5 m.
It is mechanically driven such that its displacement amplitude is w =
W cos (ωt) sin (πx/L). The maximum displacement is 0.5 mm, the frequency
is 2500 Hz, and it may be assumed that the water remains in contact with the
plate throughout the vibration. Determine the steady-state waveforms of pres-
sure at x = 0, L/4, and L/2 at a depth that is 95% of the shock formation
distance. Also determine the particle velocity waveforms at those locations.
Significance
Thus far, we have only examined how the rays and waveforms are affected by self-
refraction. Here, we will see its effects on waveforms. In doing so, we will see how
to solve a coordinate transformation that features more than one position coordinate.
Solution
The displacement in Eq. (13.4.40) is w = (εc/ω) sin (ωt) sin (πx/L). A time shift
of π/ (2ω) will convert this expression to the given displacement. Accordingly, the
derived formulas for the coordinate straining, as well as for the pressure and velocity
components, are converted by setting ε = W k and replacing t with t + π/ (2ω). It is
convenient to work with a functional representation of the coordinate transformation.
Upon introduction of the time shift, Eq. (13.4.63) becomes
x = fx (η, ξ, t, z) , z = fz (η, ξ, t, z) (1)
where the functions are

2 πη
1 k
kL
fx (η, ξ, z, t) = η − (W k) βz cos (ωt − κξ) cos
2 π
κ L
3 (2)
1 k πη
fz (η, ξ, x, z, t) = ξ − (W k) βz sin (ωt − κξ) sin
2 κ L
Correspondingly, Eq. (13.4.64) becomes

vx π πη
= − (W k) cos (ωt − κξ) cos + O ε2
c κL L
πη (3)
vz κ p
= = − (W k) sin (ωt − κξ) sin + O ε2
c k ρ0 c 2 L
It is desired to determine waveforms at specified points. This requires that we solve

the coordinate transformation, Eq. (13.4.63), for the values of η and ξ corresponding
to specified values of x, z, and t. Repeating such a computation for a sequence of
t values will yield one period of a waveform, which can be periodically replicated
without requiring further solutions.
There are several numerical algorithms that may be used to solve two simultaneous
nonlinear equations. We shall use the extended version of Newton’s method. Suppose
η (n) and ξ (n) are estimates for the values of η and ξ that satisfy Eq. (1) at specified
x, z, and t. We seek to improve that estimate, such that η (n+1) = η (n) + η and
ξ (n+1) = ξ (n) + ξ satisfy both transformations, that is,

fx η (n) + η, ξ (n) + ξ, t, z = x, fz η (n) + η, ξ (n) + ξ, t, z = z
This constitutes two simultaneous nonlinear equations for η and ξ. Let us expand
each function in a first-order Taylor series relative to the values obtained with the
prior estimates. Then the equations to solve are linear in the increments,
(n)
∂fx (n) ∂fx
(fx )(n) + η + ξ = x
∂η (n) ∂ξ (n)
∂fz ∂fz
(fz )(n) + η + ξ = z
∂η ∂ξ
where the superscript for the functions and their derivatives means that the quantity
is evaluated for the variables at the nth estimate. The matrix representation of these
equations is ( + ( +
& (n)
' η x − (fx )(n)
J (fx , fz ) =
η z − (fz )(n)
& '
where J (Fx , Fz )(n) is the Jacobian of the transformation evaluated at the nth esti-
mate, ⎡ (n) (n) ⎤
∂fx ∂fx
& ' ⎢ ∂η ∂ξ ⎥
J (fx , fz )(n) = ⎢ ⎥
⎣ ∂fz (n) ∂fz (n) ⎦
∂η ∂ξ
Since η = η (n+1) − η (n) and ξ = ξ (n+1) − ξ (n) , the improved estimates of the
strained coordinates are given by
( + ( + ( +
η (n+1) η (n) & '
(n) −1 x − (fx )
(n)
= + J (fx , fz )
ξ (n+1) ξ (n) z − (fz )(n)
The iterative procedure requires an initial estimate. All of the results that follow
were obtained by setting η (0) = x and ξ (0) = z. The procedure terminates when a
convergence criterion has been met. The one used here is based on the observation
that |fx − x| should be much less than εx and |fz − z| should be much less than εz. Both
2 2
conditions
will be met if the iteration is halted when (fx )(n) − x + (fz )(n) − x <

ε4 x 2 + z2 . The iterative algorithm is used to compute the waveforms of pressure
and particle velocity at each of the specified locations. To do so the procedure is
carried out in a loop of t values covering one period. After the values of η and ξ
have been determined at each instant, the corresponding values of p, vx , and vz are
determined from Eq. (13.4.64). All response variables have a period of 2π/ω, so the
waveforms are replicated for another period in order to see the pattern more clearly.
The parameters for the evaluation are c = 1480 m/s, β = 3.5 corresponding
to B/A = 5, ω = 5000π rad/s, k = 10.613 m−1 , κ = 10.405 m−1 , and ε = W k =
5.3067 10−3 . The wavelength in the z direction is 2π/κ = 0.6039 m. The shock
formation distance is xshock = 9.705 m, so this is a fairly strong signal. The wave-
forms that are computed at x = 0, 0.375, and 0.75, with z = 9.263 m are shown in
Fig. 1. They are best understood if they are compared to their linear analog. Linear
theory recovered by setting ε = 0 in Eq. (2),
vx π πx
= − (W k) cos (ωt − κz) cos
c κL L
vz κ p πx (5)
= = − (W k) sin (ωt − κz) sin + O ε2
c k ρ0 c 2 L
Thus, in linear theory p should be zero along x = 0 and vx should exhibit its maximum
excursion along that line, whereas the amplitude of p should be maximum and vx
should be zero along x = L/2. The situation at x = L/4 should be intermediate to
these extremes.
= 103 x (k/)vz /c
103 x p/(0c2)
0
−5
x=0 x = L/4 x = L/2
1
103 x vx /c
0.5
0
−0.5
−1
0 0.5 1 1.5 2
Time, t/T
Figure 1.
Because vx = 0 along x = L/2, the ray leaving the plate at that location does not
rotate. The pressure signal evolves along a ray in a manner similar to planar wave.
Hence, we see the same distortion effect along this line, with positive pressures
arriving earlier, and negative pressures arriving later. Along x = 0, the pressure is
nonzero because different rays pass through a point at (0, z) at different instants as a
result of the self-refraction effect. Rays on which vx > 0 arrive from x < 0 for half
a period, and rays with vx < 0 arrive from x > 0 for the other period. The variation
of pressure along both sets of rays is similar, so the pressure undergoes a double
oscillation in one period. Hence, the pressure at that location has half of the overall
period.
The waveforms for vx are interesting in that they are symmetric, rather than leaning
over due to advancement and retardation. This symmetry occurs because vx in a ray
coming from the plate at x > 0 is merely 180◦ out-of-phase from vx on a ray arriving
from x < 0. If we were to examine vx as a function of t at x = L, we would see the
inverted version of vx at x = 0 because all variables are shifted there by 180◦ .
13.5 Further Studies
Nonlinear phenomena have been the subject of a vast body of research. The top-
ics covered here provide the primary tools and concepts required to go further. A
historical perspective is available in a comprehensive treatise by Hamilton and Black-
stock,22 which is an excellent starting point for further exploration. The purpose of
the discussion here is to make the reader aware of some interesting phenomena and
applications without delving into the technical details.
22 M.F. Hamilton and D.T. Blackstock, ibid.

13.5 Further Studies 655
• Radiation pressure and acoustic levitation
Our studies have concentrated on the growth of nonlinear effects as a wave prop-
agates. The consequence was that a temporally periodic excitation with zero mean
value was found to lead to a pressure field that has similar properties. This is not quite
true. In actuality, there is a small mean value for the pressure at a point. This field
may be traced back to the nonlinear terms that do not result in cumulative growth of
distortion. For example, consider the nonlinear momentum equation, Eq. (13.3.3). It
contains the term ∇φ· ∇φ. If ∇φ is harmonic, then ∇φ · ∇φ consists of a mean value
and a second harmonic of equal magnitude.
The mean value of a pressure field is referred to as the radiation pressure. This
leads to a radiation force, which is the mean value for the resultant force exerted
on a body immersed in the pressure field. Although this concept might appear to be
simple, the analysis of the concept raises some difficult issues. One concern is the
obvious fact that placing a body in the field modifies the field due to scattering. That
is, what is the radiation pressure distribution on the surface of a body that scatters
the incident signal? Another question that arises pertains to the nature of this body.
Is it fixed in space, or is it free to vibrate? Because the radiation pressure is small
compared to the pressure amplitude, small differences are important, which is why
the question of whether the body is restrained is an essential aspect.
One of the most interesting applications of radiation pressure is for acoustic
levitation. As was noted, the radiation force is small. Using it to support a body at
the earth’s surface would require that the radiation force equals the body’s weight,
which would require an incredibly high acoustic pressure. However, for a body in
orbit, for example, in the International Space Station, the weight is the centripetal
force required to maintain the body in orbit. From the reference frame of the Space
Station, the body is weightless and free to drift unless it is restrained. The radiation
force has been used to restrain a body without contacting it. This is achieved by
setting up a resonant field in a closed chamber. An unconstrained body then will drift
to a pressure node. This concept is useful for any process in which contact with a
container will lead to contamination or distortion of an object.
• Sound beams
Suppose you wish to send an acoustical message to listeners at a long range but in
a limited direction. We select a loudspeaker or hydrophone for this purpose, embed
it in a baffle, and set the frequency sufficiently high to obtain the desired directivity.
To overcome amplitude reductions due to spherical spreading and dissipation, we
drive the transducer harder. Several alternatives arise. As the input is increased, the
amplifier that drives the transducer might begin to distort or cut out entirely. The
speaker might be overdriven, which also might cause distortion or cause the speaker
to fail. The first two possibilities can be avoided with proper selection of the apparatus,
but the third possibility is not avoidable. Raising the amplitude of the signal radiated
by the transducer leads to the growth of nonlinear effects as the signal propagates.
If the drive amplitude is sufficiently large, shock waves might form. Raising the
amplitude further will merely bring the location of shocks closer to the transducer,
rather than further raising the amplitude that is received. The ultimate result might
be saturation. Another possibility for an underwater signal is that the amplitude will
be sufficiently large that the absolute pressure (ambient plus acoustic disturbance)
is negative. The consequence of such an occurrence is creation of bubbles, which is
called cavitation. This process also limits the power that can be transferred into the
radiated signal. (Negative pressure is not an issue for atmospheric sound. To see why

consider a plane wave for which the acoustic Mach number is ε = |p| / ρ0 c2 . A
negative absolute pressure corresponds to |p| = 1 atm, but ρ0 c2 is approximately 1
atm, and we know that ε = 1 is only attained with an explosion. In contrast ρ0 c2 ≈ 2
GPa for water and the ambient pressure near
the surface is approximately 1 atm.
Setting |p| = 1 atm leads to ε ≈ 0.5 10−4 , which is quite attainable.)
A primary application of sound beams is a method of long-range communication.
They also are at the core of some imaging concepts, such as an acoustical microscope
and some biomedical ultrasonic devices. In each case, it is imperative that nonlinear
effects be understood. Our study of sound beams according to linear theory required
numerical methods to evaluate the general field properties. Thus, it is obvious that
studying nonlinear effects is even more challenging. Many analyses, particularly
the early ones were based on a regular perturbation analysis, in which the linear
solution for a sound beam is used to describe the second order, that is, quadratic,
nonlinear terms. The Rayleigh integral is one way of describing the linear solution,
but there are others. Furthermore, to reduce the level of difficulty various simplified
field equations often provided the basis for the analysis. The simplifications were
based on properties of sound beams at high frequency, such as that the spatial scale
of diffraction is large compared to the axial wavelength, but small compared to the
Rayleigh distance.
A regular perturbation analysis is limited in its range of validity. A more complete
picture of nonlinear effects was obtained by using some of the methods developed
here, notably the frequency domain Fourier series approach and the singular pertur-
bation formulation.
• Parametric arrays
An application that relies on nonlinear effects is a concept that combats a basic aspect
of the linear field of a sound beam. Specifically, if one wishes the beamwidth to be very
small, then the frequency must be high, but high frequencies mean greater dissipation,
thereby reducing the effective range. The concept entails imparting to the piston on
the boundary a vibration that is the sum of two harmonic signals at closely spaced
frequencies. The linear field, which is the first-order solution for a regular perturbation
series, is the superposition of sound beams at each frequency. If both frequencies
are high, both beams will be narrow, and their directivities will be quite similar
because the frequencies are close. When the first-order solution is used to generate
the quadratic terms in a nonlinear wave equation, approximate or exact, various
combinations of harmonic signals are created. Second harmonics of each primary
frequency are like the behavior of each beam if it were the only one present. Mixing
of the two harmonics results in a sum frequency signal, which is close to the second
harmonics of the individual beams, and a difference frequency signal. The latter part
13.5 Further Studies 657
is the one that is of interest because proximity of the two drive frequencies means
that the difference frequency is very small. Furthermore, because these terms are the
product of two narrow directivities, the difference frequency terms are significant
only near the axis of the sound beam. The homogeneous terms of the second-order
wave equation are the same as those of the first-order equation, but the quadratic
terms occur on the right side of the equation as virtual sources. These difference
frequency sources are confined to the region close to the axis. In combination with
the fact that the propagation is predominantly axial, the consequence is that the virtual
sources act like a narrow end-fire array. The creation of this array is attributable to
the parameters of the primary frequency sound beams and hence the name of the
difference frequency sources as a parametric array. This concept has been used in
some applications, notably as a narrowly confined megaphone. It has not been used
widely because it relies on nonlinear interactions, which are relatively weak, with a
magnitude that is the order of the square of the acoustic Mach number.
• Biomedical applications
The use of sound beams for ultrasonic diagnostics was mentioned earlier. Increasing
the depth of penetration by raising the amplitude enhances nonlinear effects. One
aspect of this is beneficial in that higher harmonics have smaller wavelengths. This
serves to enhance the visual resolution. At the same time, larger amplitudes lead to
greater tissue heating. This is a harmful effect as a diagnostic tool. However, it is
useful as a method for destroying tissue that is harmful, such as cancer cells. One of
the key requirements for understanding these phenomena is a description of how large
amplitude sound waves propagate through living tissue, which is heterogeneous and
dissipative through heat transport and viscosity. This is a subject of ongoing research.
Lithotripsy is a different way in that finite amplitude effects are exploited. Stones
can form in several organs, such as kidneys. They can become extremely painful, in
which case highly invasive surgery is the usual treatment. A lithotripter provides a
nonsurgical alternative. The basic concept is to immerse the torso in a tub of water,
with the walls of the tub driven as a pulse. This generates an inwardly propagating
transient wave that converges on the stone. The consequence of convergence is that
the pressure field applied to the stone is greatly enhanced and changes even more
abruptly than the original pulse. (Recall the behavior in linear theory of an inwardly
propagating spherical wave at the focus, see Sect. 6.2.3.) Hence, the pressure acting
on the stone is large and the pressure gradient is very large. Furthermore, the stone
typically is somewhat irregular in shape and has some internal faults. The combina-
tion of these attributes is that large stresses induced within the stone might cause it
to fracture into smaller pieces that might cause it to pass through the duct. Transfer
from a concept to a functioning apparatus required a considerable body of research
concerning focusing of nonlinear waves and their interaction with solid objects.
Exercise 13.1 An explosion in air produces a pressure pulse in the form of a single
lobe of a sine function. The amplitude is 20 kPa and the duration is 0.1 s. Draw
the characteristics that track the peak pressure, and the zero pressure values at the
beginning and end of the pulse. Identify the smallest distance at which any two of
these lines intersect. Sketch the waveform predicted by the Riemann solution at this
location. What does that sketch indicate regarding the location at which a shock
forms?
Exercise 13.2 A transducer at the near end, x = 0, of a very long tube imparts a
velocity wave form in the shape of a triangular pulse of duration T , given by
, 2t − T
εc , 0<t<T
v= T
0 otherwise
Thus, t = T is the last instant at which the transducer is moving. (a) Draw the spatial
profile of this wave at t = T . (b) Evaluate the shock formation distance xs for this
wave. (c) Evaluate the waveform at x = xs .
Exercise 13.3 At t = 0 the signal in a waveguide is observed to be a plane wave

propagating in the direction of increasing x. Its spatial profile at that instant has the
shape of an equilateral triangle,
⎧
⎪
⎪
2x
ε , 0 < x < L/2
⎪
⎪
⎪
⎨ L
p
= 2 (L − x)
ρ0 c2 ⎪
⎪ ε , L/2 < x < L
⎪
⎪ L
⎪
⎩
0 otherwise
Determine the shock formation distance xshock according to weak shock theory. Then
draw the waveform at x = xshock and x = xshock /2.
Exercise 13.4 A projectile impacts

the back of a wall, thereby inducing a displace-
ment pulse u = (4/3)εc t 2 /T 2 (3T − 2t) that occurs over 0 ≤ t ≤ T . The parame-
ters are ε = 0.0001 and T = 10 ms, and the fluid is air at standard conditions. (a)
Determine the minimum distance at which a shock forms according to the weakly
nonlinear approximation. (b) Use the weakly nonlinear approximation of the Rie-
mann solution to evaluate the waveform at x = xshock . (c) Use Eq. (13.1.17), which
accounts for movement of the wall, to evaluate the waveform at x = xshock . Estimate
the overall error in the result in Part (b) that results from using a stationary, rather
than moving, boundary condition.
Exercise 13.5 The velocity of a vibrating wall consists of a pulse that occurs over
an interval T . The waveform of that pulse is a single cycle of a sine function, v =
εc sin (2πt/T ). The fluid is air at standard conditions. To test whether the limits
of the weakly nonlinear approximation set ε = 0.1, T = 4 ms. (a) Use the weakly
nonlinear approximation to evaluate the waveform at the shock formation distance.
(b) Use the exact ideal gas relations to evaluate the waveform at the same location.
Exercise 13.6 If the particle velocity in a weakly nonlinear plane wave at x = 0

is v = εc sin (ωt), then the shock formation distance is xshock = 1/(βεk). Consider
the case of a 1 kHz signal in water with ε = 0.0001. (a) Use numerical methods to
evaluate and graph the waveform at x = xshock . (b) Evaluate and graph the waveform
at x = 1/(βεk) when the input waveform (at x = 0) is the negative of the waveform
at x = xshock found in Part (a). (c) Based on the properties of the waveforms at x = 0
and x = 1/(βεk), identify the minimum value of x at which the inverted signal will
form a shock.
Exercise 13.7 An experiment in the Tunnel de Sainte-Marie-aux-Mines in east-

ern France measured the waveform of a pressure pulse at 3 Km from the western
entrance. With t = 0 defined as the instant when the pulse arrived, a function fit-
ting the measured data is p = 42(t/T )1/2 (1 − t/T ) [h(t) − h (t − 1.25T )] Pa, with
T = 12 ms. Because ∂p/∂t is infinite at t = 0 and t = 1.25T , with p increasing at
both instants, the location where this waveform was measured marks shock forma-
tion. Determine the waveform at the tunnel entrance, where the pulse was generated.
It may be assumed that waves only propagate from the source to the measurement
point, and that the propagation is planar with negligible dissipation.
Exercise 13.8 The pressure input at x = 0 consists of second and third harmonics,
such that
p = ερ0 c2 [a2 sin (2ωt) + a3 sin (3ωt)]
When a3 = 0, the shock forms at x = 1/(2a2 εβk), whereas this location is kx =

1/(3a3 εβ) when a2 = 0, where k = ω/c. Cases of interest are a2 = 1, a3 = 0.1,
and a2 = 1, a3 = 2/3. For both cases ε = 0.002 and β = 3.6. Evaluate and graph
p/ ρ0 c2 as a function of ωt − kx at kx = 0.25/(βεa2 ) for each case. Compare each
waveform to its linear counterpart. Identify the instants at which ∂p/∂t is greatest in
each case, and examine each waveform to determine whether a shock has formed.
such that
p = ερ0 c2 [sin (2ωt) + (2/3) sin (3ωt)]
The period of this function is ωT = 2π. The parameters are ε = 0.002 and β = 3.5.
(a) Use numerical methods to analyze the waveform at a sequence of position kxn =
0.04n/ (βε), n = 0, 1, ..., 6. (b) Use FFT techniques to evaluate the amplitude of
harmonics 1 to 5 at each position.
Exercise 13.10 Consider the situation in Exercise 13.9. Rather than using FFT
analysis of computed waveforms, use Fourier series analysis of the Riemann solution
to determine the amplitude of harmonics 1 to 5 at kxn = 0.04n/ (βε), n = 0, 1, ..., 6.
such that
p = ερ0 c2 [sin (2ωt) + (2/3) sin (3ωt)]
The fluid is air with the attenuation described by Eq. (3.3.33). The parameters are
ω = 800 rad/s, ε = 0.002, and β = 1.2. Solve the differential equations governing
the waveform’s Fourier series coefficients in order to evaluate the x dependence of the
lowest four harmonic amplitudes. Halt the evaluation at kx = 0.24/(βε). Compare
the waveform at the farthest location to the result that would be obtained if dissipation
were negligible.
Exercise 13.12 The pressure generated by a periodic source at x = 0 is a com-

bination of the ninth and tenth harmonics, specifically, p = ρ0 c2 ε (sin (9ωt) + sin
(10ωt)). At the origin, this signal constitutes a beat, but an important aspect of
the nonlinear process is harmonic and intermodulation distortion, in which higher
and lower harmonics are generated. Derive an expression for the shock formation
distance, xshock . Then, solve the differential equations in Sect. 13.3.2 for 0 ≤ x ≤
0.9xshock . Use Eq. (3.3.33) to describe the frequency dependence of the attenuation
coefficient. The system parameters are ω/(2π) = 500 Hz and ε = 0.002. The fluid is
water, for which ρ0 = 1000 kg/m3 , c = 1480 m/s, and β = 3.6. Plot the amplitudes
of pressure harmonics 1 to 15 as a function of x. Use a logarithmic scale for the
amplitude values.
Exercise 13.13 The spatial profile of a wave is described as p as a function of x

at fixed t. Prove that the shock formation distance is manifested in a profile as an
infinite value of ∂p/∂x with p decreasing.
Exercise 13.14 A piston at the end x = 0 of a semi-infinite waveguide vibrates

periodically. Its steady-state velocity is the sawtooth waveform in the sketch. The
waveform at any x preceding shock formation is a scalene triangle. (a) Use the
graphical construction method to establish the shape of the triangle at an arbitrary x
preceding shock formation. (b) Carry out a Fourier analysis of the waveform in
Part (a), and thereby, derive an expression for the dependence of the harmonic
amplitudes on x. (c) Prove that the shock formation distance for this signal is xshock
= cT / (2εβ). (d) Evaluate the waveform and harmonic content at xn = nxshock /4,
n = 0, 1, ..., 4. Which of these harmonics are enhanced and which are depleted as
the signal propagates?
v/(c)
0
-1
0 0.5 1 1.5 2
Time t/T
Exercise 13.14
Exercise 13.15 The periodic boundary velocity in Exercise 13.14 generates a steady-
state plane wave in an infinitely long waveguide. (a) Determine the distance xold at
which the wave enters the old-age stage. (b) Determine the waveform at x > xold .
From that result, derive an expression for the maximum pressure as a function of x.
Exercise 13.16 A pressure surge occurs at x = 0 in a waveguide filled with water.

The duration of the pulse
is 0 ≤ t ≤ T , and its signature in that interval is
p (x = 0, t) = 0.50 106 (t/T ) exp (3 (1 − t/T )) Pa where T = 50 ms. (a) Deter-
mine the shock formation distance xshock . (b) Evaluate the Riemann solution at evenly
spaced locations x = xold , 2xold , 3xold , ... Sequentially examine each waveform to esti-
mate the smallest distance at which old age begins. (c) Evaluate the waveform at the
location estimated in Part (b) to be the beginning of old age. Does the shock forma-
tion distance obtained from the analysis indicate that the estimated distance was too
large or too small?
Exercise 13.17 When the spring in a one-degree-of-freedom oscillator is weakly

nonlinear, the force it exerts deviates from proportionality to its elongation. A model
that is descriptive of some systems adds a force term that is proportional to the square
of the displacement. The resulting equation of motion is
M ÿ + μẏ + K1 y + K2 y2 = f (t)
where f (t) is the excitation force and μ is a viscosity coefficient that accounts for
dissipation. In the case where f (t) is a periodic function, the steady-state response
must share that period. This means that y, as well as f , may be represented by Fourier
series. Thus, let
1 # 1 #
N N
y= Yn einωt , Y(−n) = Yn∗ f = Fn einωt , F(−n) = Fn∗
2 n=−N 2 n=−N
where the harmonic N for truncation is based on convergence of the series. Derive
a set of algebraic equations whose solution would give the Yn values when the Fn
values are specified.
Exercise 13.18 If an elastic beam is fastened at its ends to support that are immobile,
any displacement increases the length of the beam. This elongation is an accumulation
of the effect of displacement w along the beams length, so it is represented by an

integral. The effect of the elongation is to make the beam stiffer. The field equations
and boundary conditions for the beam are
2
∂ 4 w 1 EA L
∂w ∂2w
EI + dx + σ =q
∂x 4 2 L2 0 ∂x ∂t 2
where E is Young’s modulus, A and I are the area and the area moment of inertia of
the cross section, σ is the mass per unit length, and q is the force loading per unit
length of the beam. If the beam is supported at its ends by stationary pins that allow
the beam to rotate freely, they are said to be simple supports. The associated boundary
conditions are w = ∂ 2 w/∂x 2 = 0 at x = 0 and x = L. If the displacement is small
relative to the span L, then it may be represented nondimensionally as w = εLξ (x, t),
where |ε| 1. Write the dimensionless displacement as a regular perturbation series,
ξ = ξ1 + εξ2 + · · ·. Derive the differential equation and boundary conditions gov-
erning ξ1 and ξ2 .
Exercise 13.19 When the spring in a one-degree-of-freedom oscillator is weakly

nonlinear, the force it exerts deviates from proportionality to its elongation. This
effect often is modeled by adding a force term that is proportional to the cube of the
displacement. The resulting equation of motion is
M ẍ + K1 x + K3 x 3 = F (t)
where F (t) is the excitation force. The nonlinear coefficient K3 is much less than
K1 , so it may be written as K3 = εK1 , |ε| 1. A regular perturbation series for the
oscillator displacement is x = x0 + εx1 + · · ·. (a) Derive the differential equations
governing x1 , x2 , and x3 . (b) When the force varies harmonically, F (t) = F̂ cos (ωt),
the steady-state response also must be periodic. Find the steady-state response of
the differential equations obtained in Part (a). Is there any value of ω at which the
solution fails to be uniformly valid?
Exercise 13.20 Free vibration of the oscillator in Exercise 13.19 corresponds to

f (t) = 0. Consider an initial state in which x = X0 and ẋ = 0 when t = 0. Division
of the equation of motion by M leads to

ẍ + ωnat
2
x + εx 3 = 0
where ωnat = (K1 /M)1/2 is the natural frequency of the linear system. Use the regular
perturbation series x = x0 + εx1 + · · · to obtain a solution of the equation of motion
that is consistent with the initial conditions. Then, apply the method of strained
coordinates to the time variable in order to obtain a uniformly valid description of
the response.
Exercise 13.21 The diagram is the pressure pulse measured on the surface of a
sphere, r = a. This radius is much greater than the spatial extent of the pulse, that is,
a cT . (a) Use the graphical construction method to sketch the pressure waveform
at an arbitrary radial distance. (b) Use the waveform obtained in Part (a) to derive an
expression for the radial distance rshock at which a shock forms. (c) Determine the
waveforms of the radial particle velocity at r = a, r = rshock /2, and r = rshock .
1
v/(c)
-1
0 0.25 0.5 0.75 1
Time t/T
Exercise 13.21.
Exercise 13.22 The pressure resulting from a huge explosion in the atmosphere
is measured at r1 = 2 km from the ignition point. A function that correlates well
with the measured data is p = p1 (t/T ) (1 − 5t/T ) exp (−μt), with p1 = 20 kPa and
μ = 4.54 s−1 . This function applies for t < 2 seconds, after which it may be set to
zero. It is believed that the propagation is well described by the weakly nonlinear
approximation beyond r0 = 5 m. (Inside this distance, the signal is highly nonlinear
and the water is in a nonideal state.) Plot p at r0 as a function of the retarded time t =
t − (r0 − r1 ) /c. Without performing calculations, conjecture what the waveform
would look like at a radial distance r that is significantly greater than r1 ?
Exercise 13.23 A 10 m diameter sphere will be used as the living chamber of a
deepwater submersible. One of its tests will apply an internal pressure fluctua-
tion when it is submerged. This pressure induces a radially symmetric vibration
of the outer surface in which the surface velocity is vr = 0 for t < 0 and t > T ,
and vr = εc(1 − 4/T )2 (5t/T − 1) for 0 < t < T . Parameters are ε = 0.0032 and
T = 0.94 ms. (a) Determine the pressure waveform at the radial distance at which a
shock develops. (b) Determine the radial velocity at r = a and r = rshock . Plot vr /c
at each location as a function of the retarded time t − (r − a)/c). Compare each
waveform to (r/a) p/(ρ0 c2 ) as a function of the retarded time. What can be deduced
about the distance at which the farfield begins?
Exercise 13.24 If −Ḟ (τ ) in Eq. (13.4.29) is replaced by c2 V (τ ), the resulting
expression for (r/a) p in a spherical wave has the same appearance as p = ρ0 c2 εV (τ )
for a plane wave. Use this similarity to modify the general Fourier series analysis
in Sect. 13.2.1 in order to determine the Fourier coefficients of a nonlinear spherical
wave when the function Ḟ (t) is periodic. Specialize the general result to the case
where the pressure at x = 0 is a single harmonic.
Exercise 13.25 At a reference radial distance r0 , the waveform of a radially sym-
metric spherical wave is observed to be the square wave depicted below. (a) Derive
an expression for the radial distance rold at which this wave enters old age. (b) Derive
an expression for the maximum positive pressure in this wave if the radial distance
exceeds rold .
p @ r=r0
T/2 T/2
p^
t
-p^
Exercise 13.25.
Appendix A
Curvilinear Coordinates
A.1 Spherical Coordinates
When p is known in terms of the spherical coordinates of the location at which is

observed, the most expedient evaluation of ∇ p uses a form in terms of the radial,
polar, and azimuthal components. This form is more complicated in appearance
than merely the derivative of a component with respect to a coordinate because
incrementing the angular variables alters the direction of the unit vectors. There are
several approaches by which these effects may be described in a derivation of the
gradient ∇ p and the Laplacian ∇ 2 p in spherical coordinates. We shall employ a direct
approach based on the coordinate transformation to derive ∇ p, and then switch to
a slightly more subtle approach based on the properties of the spherical coordinate
unit vectors to derive ∇ 2 p.
A.1.1 Transformations
The derivation of a gradient in spherical coordinates begins by considering the pres-

sure to be a function of those coordinates. In turn, the spherical coordinates are treated
as functions of the Cartesian coordinates. Doing so calls for the chain rule for partial
differentiation. It is helpful to have the transformations at hand. The coordinates are
defined in Fig. A.1.
The Cartesian coordinates corresponding to a given a set of spherical coordinates
are
x = r sin ψ cos θ
y = r sin ψ sin θ (A.1.1)
z = r cos ψ
The inverse transformation gives the spherical values for a given set of Cartesian
coordinates. It is

DOI 10.1007/978-3-319-56847-8
666 Appendix A: Curvilinear Coordinates
Fig. A.1 Spherical z

coordinates and associated Constant
unit vectors r and  e- r

Constant -
e
 and 
r e- 
y
Constant
Equator r and 

x
1/2
r = x2 + y 2 + z 2
−1 z
ψ = cos 1/2
x 2 + y2 + z2 (A.1.2)

−1 y −1 y −1 x
θ = tan = sin 1/2 = cos 1/2
x x 2 + y2 x 2 + y2
We also will have need for the definition of spherical unit vectors. Each vector is the
change of the position vector from the origin when the corresponding coordinate is
increased by a unit value, with the others held constant. The result of doing so is
ēr = sin ψ cos θēx + sin ψ sin θē y + cos ψ ēz

ēψ = cos ψ cos θēx + cos ψ sin θē y − sin ψ ēz (A.1.3)
ēθ = − sin θēx + cos θē y
A.1.2 Gradient
The Cartesian components of a gradient are the derivative with respect to the corre-
sponding coordinate. The chain rule is needed to evaluate these derivatives when p
is known in terms of r , ψ, and θ, so we have
∂p ∂ p ∂r ∂ p ∂ψ ∂ p ∂θ
= + +
∂x ∂r ∂x ∂ψ ∂x ∂θ ∂x
∂p ∂ p ∂r ∂ p ∂ψ ∂ p ∂θ
= + + (A.1.4)
∂y ∂r ∂ y ∂ψ ∂ y ∂θ ∂ y
∂p ∂ p ∂r ∂ p ∂ψ ∂ p ∂θ
= + +
∂z ∂r ∂z ∂ψ ∂z ∂θ ∂z
Appendix A: Curvilinear Coordinates 667
The transformation from Cartesian to spherical coordinates is used to fill in the

derivatives appearing in these equations. Implicit differentiation is useful for the
angles. The derivatives of Eq. (A.1.2) with respect to x proceed as follows:
∂r ∂ 2 1/2 x
= x + y2 + z2 = 1/2
∂x ∂x x 2 + y2 + z2

∂ ∂ψ ∂ z ∂ z
(cos ψ) ≡ − sin ψ = = 1/2
∂x ∂x ∂x r ∂x x 2 + y2 + z2
xz
= − 3/2
x + y2 + z2
2
∂ 1 ∂θ ∂ y y
(tan θ) ≡ = =− 2 (A.1.5)
∂x (cos θ) ∂x
2 ∂x x x
Derivatives with respect to y and z are obtained by similar operations.

We seek a description of ∇ p in which only variables associated with spherical
coordinates appear. The transformations from Cartesian to spherical coordinates,
along with x 2 + y 2 + z 2 = r 2 , are used to eliminate x, y, and z from the preceding.
Doing so leads to
∂r ∂r ∂r
= sin ψ cos θ, = sin ψ sin θ, = cos ψ
∂x ∂y ∂z
∂ψ cos ψ cos θ ∂ψ cos ψ sin θ ∂ψ sin ψ
= , = , =− (A.1.6)
∂x r ∂y r ∂z r
∂θ sin θ ∂θ cos θ ∂θ
=− , = , =0
∂x r sin ψ ∂y r sin ψ ∂z
These expressions are substituted into Eq. (A.1.4), which then are used to form
the gradient in Cartesian components. This step yields

∂p 1 ∂p 1 ∂ p sin θ
∇p = sin ψ cos θ + cos ψ cos θ − ēx
∂r r ∂ψ r ∂θ sin ψ

∂p 1 ∂p 1 ∂ p cos θ
+ sin ψ sin θ + cos ψ sin θ + ē y (A.1.7)
∂r r ∂ψ r ∂θ sin ψ

∂p 1 ∂p
+ cos ψ − sin ψ ēz
∂r r ∂ψ
The last step is to convert this expression into spherical coordinate components.
By definition, a component is the projection of a vector onto a coordinate axis, so
we may write

∇ p ≡ (∇ p · ēr ) ēr + ∇ p · ēψ ēψ + (∇ p · ēθ ) ēθ (A.1.8)
Substitution of the unit vectors in Eq. (A.1.3) and the gradient on Eq. (A.1.7) ulti-
mately reduces to
∂p 1 ∂p 1 ∂p
∇p = ēr + ēψ + ēθ (A.1.9)
∂r r ∂ψ r sin ψ ∂θ
One device for remembering this expression is that dr is the displacement when r is
incremented infinitesimally, r dψ is the meridional displacement for an infinitesimal
increment of ψ, and (r sin ψ) dθ is the azimuthal displacement when θ is incremented.
A.1.3 Laplacian
The final form of the gradient in spherical coordinates is understandable as changes

in the field when the position is shifted infinitesimally in each coordinate direction.
The Laplacian is less intuitive and somewhat more difficult to derive. One approach
for deriving ∇ 2 p parallels that used to describe ∇ p.Specifically, it applies the chain
rule to convert (∂/∂x)(∂ p/∂x), (∂/∂ y)(∂ p/∂ y), and (∂/∂z)(∂ p/∂z) to derivatives
with respect to r , ψ, and θ. The derivation that follows proceeds differently. It uses
Eq. (A.1.9) to describe both terms in ∇ 2 p ≡ ∇ · ∇ p, so that

∂ 1 ∂ 1 ∂ ∂p
∇2 p = ēr + ēψ + ēθ · ēr
∂r r ∂ψ r sin ψ ∂θ ∂r (A.1.10)
1 ∂p 1 ∂p
+ ēψ + ēθ
r ∂ψ r sin ψ ∂θ
Neither the components of ∇ p nor the unit vectors are constant. For this reason, the
derivatives must be evaluated prior to taking dot products. In other words,

∂ ∂p 1 ∂p 1 ∂p
∇ p = ēr ·
2
ēr + ēψ + ēθ
∂r ∂r r ∂ψ r sin ψ ∂θ
1 ∂ ∂p 1 ∂p 1 ∂p
+ ēψ · ēr + ēψ + ēθ (A.1.11)
r ∂ψ ∂r r ∂ψ r sin ψ ∂θ
1 ∂ ∂p 1 ∂p 1 ∂p
+ ēθ · ēr + ēψ + ēθ
r sin ψ ∂θ ∂r r ∂ψ r sin ψ ∂θ
Further progress entails evaluation of the derivatives of the unit vectors with
respect to each spherical coordinate. These operations are applied to the unit vectors
in Eq. (A.1.3). For example,
∂ ēψ
= − sin ψ cos θēx − sin ψ sin θē y − cos ψ ēz (A.1.12)
∂ψ
It will be easier to evaluate the dot products in Eq. (A.1.11) if the unit vector deriv-
atives are expressed in terms of ēr , ēψ , and ēθ . Toward that end we form the dot
product of each unit vector with the derivative of each unit vector. For the preceding
description of ∂ ēψ /∂ψ, these operations yield
∂ ēψ ∂ ēθ
ēr · = sin ψ cos θēx + sin ψ sin θē y + cos ψ ēz · = −1
∂ψ ∂ψ
∂ ēψ ∂ ēψ
ēψ · = − cos ψ cos θēx − cos ψ sin θē y − sin ψ ēz · =0 (A.1.13)
∂ψ ∂ψ
∂ ēψ ∂ ēψ
ēθ · = − sin θēx + cos θē y · =0
∂ψ ∂ψ
These components are used to construct the vector representation of ∂ ēθ /∂θ
according to

∂ ēψ ∂ ēψ ∂ ēψ ∂ ēψ
= ēr · ēr + ēψ · ēψ + ēθ · ēθ = −ēr (A.1.14)
∂ψ ∂ψ ∂ψ ∂ψ
The other derivatives are found by a similar process. The full set of results is
∂ ēr ∂ ēψ ∂ ēθ

= 0̄, = 0̄, = 0̄
∂r ∂r ∂r
∂ ēr ∂ ēψ ∂ ēθ
= ēψ , = −ēr , = 0̄ (A.1.15)
∂ψ ∂ψ ∂ψ
∂ ēr ∂ ēψ ∂ ēθ
= sin ψ ēθ , = cos ψ ēθ , = − sin ψ ēr − cos ψ ēψ
∂θ ∂θ ∂θ
Now that derivatives of the unit vectors have been characterized, we may proceed
to carry out the derivatives in Eq. (A.1.11). Some terms are zero because the unit
vectors form an orthonormal set, and others vanish because some partial derivatives
of a unit vector are zero. What remains leads to
∂2 p 2 ∂ p 1 ∂2 p cot ψ ∂ p 1 ∂2 p
∇2 p = + + 2 + 2 + (A.1.16)
∂r 2 r ∂r r ∂ψ 2 r ∂ψ r (sin ψ) ∂θ2
2 2
A.1.4 Velocity and Acceleration
To describe the velocity and acceleration of a particle, it is necessary to account for

the time dependence of the spherical coordinates. The position of a particle depends
explicitly on the value of r because x̄ = r ēr . However, the position also depends on
ψ and θ because the ēr depends on ψ and θ. Accordingly, differentiating position
requires application of the chain rule,
d x̄ ∂ ēr ∂ ēr ∂ ēr

v̄ ≡ = ṙ ēr + r ṙ + r ψ̇ + r θ̇ (A.1.17)
dt ∂r ∂ψ ∂θ
Equation (A.1.15) describe the derivatives of the unit vectors, so the velocity is
v̄ = ṙ ēr + r ψ̇ ēψ + r θ̇ sin ψ ēθ (A.1.18)
An expression for the acceleration is derived similarly. However, the motion vari-
able of interest in fluid motion is the particle velocity, so we consider the velocity
components to be functions of time that are known. For this reason, the preceding is
rewritten as
v̄ = vr ē + vψ ēψ + vθ ēθ
vψ vθ (A.1.19)
ṙ = vr , ψ̇ = , θ̇ =
r r sin ψ
Differentiation of this description of velocity leads to

∂ ēr ∂ ēr ∂ ēr
ā = v̇r ēr + v̇ψ ēψ + v̇z ēz + vr ṙ + ψ̇ + θ̇
∂r ∂ψ ∂θ
(A.1.20)
∂ ēψ ∂ ēψ ∂ ēψ ∂ ēθ ∂ ēθ ∂ ēθ
+ vψ ṙ + ψ̇ + θ̇ + vθ ṙ + ψ̇ + θ̇
∂r ∂ψ ∂θ ∂r ∂ψ ∂θ
The rates of change of the spherical coordinates are replaced by the velocity com-
ponents in Eq. (A.1.19), and Eq. (A.1.15) are used to describe how the unit vectors
change. Collection of like components yields
v vθ
ψ
ā = v̇r ēr + v̇ψ ēψ + v̇θ ēθ + vr ēψ + ēθ
r r
vψ vθ vθ
+ vψ − ēr + cos ψ ēθ + vθ − sin ψ ēr − cos ψ ēψ
r r sin ψ r sin ψ
(A.1.21)

vψ2
v2 v2 v2 vr vθ
ā = v̇r − − θ ēr + v̇ψ − r − θ cot ψ ēψ + v̇θ + ēθ
r r r r r
(A.1.22)
The primary use of Eq. (A.1.18) for acoustics is to formulate a velocity continuity
condition at an interface, and to evaluate the intensity. There are few contexts in
which the acceleration arises, so Eq. (A.1.22) seldom is needed.
A.2 Cylindrical Coordinates
Cylindrical coordinates are an extension of polar coordinates in which the distance

from the plane, which is the axial distance z, provides the third coordinate. Thus,
if a variable only occurs in the x y plane, or if a three-dimensional field variable is
constant in the axial direction, setting all derivatives with respect to z will convert
the formulas derived here to polar coordinates.
A.2.1 Transformations
The definition of cylindrical coordinates begins with the selection of a reference

direction to be the axis of the coordinate system. In most cases, we are free to label
directions as we wish, so we shall label this reference axis as z. The plane for the
polar coordinates is defined such that z is perpendicular to it. The polar coordinates
are R and θ, but here too, other labels might be used. In Fig. A.2, θ is defined relative
to the x-axis in the sense of the right hand rule. (Point the thumb of the right hand
parallel to the z-axis. The curled fingers of that hand define the sense in which θ
increases.) This rule is somewhat arbitrary, for example, one might wish to define θ
relative to the y-axis, but the definition in the figure will suffice for our purposes.
The coordinate transformations between (x, y, z) Cartesian coordinates and
(R.θ, z) cylindrical coordinates may be derived by projecting a point onto the coor-
dinate planes. Figure A.2 shows that the axial distance z locates a point above or
below the x y plane; it is the same as the Cartesian coordinate. In a three-dimensional
situation, we should think of R as the perpendicular distance from the point to the
z-axis. Thus, we say that it is the transverse distance. In addition to being descriptive,
using the adjective transverse aids us to avoid confusing it with the radial distance r in
spherical coordinates, which is distance to a reference point. The angle θ measures
how much the transverse line is rotated about the z-axis. It is the circumferential
angle because it locates a point along the circumference of the circle whose radius
is R.
Trigonometry shows that the forward transformation is
x = R cos θ
y = R sin θ (A.2.23)
z=z
We may determine the inverse transformation either by solving the above, or else
through another trigonometric analysis. The result is
Fig. A.2 Definition of z

cylindrical coordinates when
z defines the axial direction Constant e- z
R
 and z e- 
e- R
Constant
R and z z
Constant
R and 
y

x
1/2
R = x 2+ y 2
x −1 y
θ = cos−1 1/2 = sin 1/2 (A.2.24)
x 2 + y2 x 2 + y2
z=z
It should be noted that the circumferential angle has not been described as θ =
tan−1 (y/x) because doing so leads to ambiguity in the assignment of quadrant.
Every curvilinear coordinate system has an associated set of unit vectors. They
describe the direction in which a point would displace if one coordinate were
increased and the others held constant. The cylindrical coordinate directions are the
transverse direction ē R , the circumferential direction ēθ , and the axial direction ēz .
Increasing R displaces the point outward perpendicularly to the z-axis, and increas-
ing z displaces the point parallel to the z-axis. When θ is increased with R and z
held constant, the point moves along a circle of radius R parallel to the x y plane.
Thus, ē R , ēθ , and ēz form a mutually orthogonal set of unit vectors. A cylindrical
coordinate description of a vector entails finding the projections of the vector along
each of these unit vectors. In turn, these unit vectors may be defined as components
relative to the Cartesian coordinate directions. Reference to Fig. A.2 shows that
ē R = ēx cos θ + ē y sin θ

ēθ = −ēx sin θ + ē y cos θ (A.2.25)
ēz = ēz
The orthogonal nature of the cylindrical coordinate directions may be verified by

using the preceding to evaluate ē R · ēθ , ē R · ēz , and ēθ · ēz .
A.2.2 Gradient
The derivation of ∇ p in terms of cylindrical coordinates follows the general proce-

dure for the spherical coordinate derivation. Because z is the same for cylindrical
and Cartesian coordinates and z is independent of x and y, some terms in the chain
rule for differentiation with respect to x, y, and z are identically zero. The nonzero
terms that result when p is a function of R, θ, and z are
∂p ∂p ∂R ∂ p ∂θ
= +
∂x ∂ R ∂x ∂θ ∂x
∂p ∂p ∂R ∂ p ∂θ
= + (A.2.26)
∂y ∂R ∂y ∂θ ∂ y
∂p ∂p
=
∂z ∂z
We could evaluate the derivatives of the spherical coordinates by differentiating

Eq. (A.2.24). However, an indirect approach that differentiates Eq. (A.2.23) is more
efficient, because the resulting derivatives only depend on the spherical coordinate.
The procedure forms
∂R ∂θ ∂R ∂θ
1= cos θ − R sin θ, 0 = cos θ − R sin θ
∂x ∂x ∂y ∂y
(A.2.27)
∂R ∂θ ∂R ∂θ
0= sin θ + R cos θ, 1 = sin θ + R cos θ
∂x ∂x ∂y ∂y
The matrix form of these equations is

⎡ ∂R ∂R ⎤

cos θ −R sin θ ⎢ ∂x ∂ y ⎥ 10
⎣ ∂θ ∂θ ⎦ = (A.2.28)
sin θ R cos θ 01
∂x ∂ y
Inversion of the coefficient matrix gives

⎡ ∂R ∂R ⎤

⎢ ∂x ∂ y ⎥ cos θ sin θ
⎣ ∂θ ∂θ ⎦ = 1 1 (A.2.29)
− sin θ cos θ
∂x ∂ y R R
Substitution of these derivatives into Eq. (A.2.26), followed by assembly of the

individual terms to form ∇ p leads to

∂p ∂p 1
∇p = cos θ − sin θ ēx
∂R ∂θ R
∂p ∂p 1
+ sin θ + cos θ ē y (A.2.30)
∂R ∂θ R
∂p
+ ēz
∂z
When we collect the coefficients of each derivative of p, we obtain

∂p ∂p ēx ē y ∂p
∇p = ēx cos θ + ē y sin θ + − sin θ + cos θ + ēz (A.2.31)
∂R ∂θ R R ∂z
A comparison of these coefficients to Eq. (A.2.25) reveals that
∂p 1 ∂p ∂p
∇p = ē R + ēθ + ēz (A.2.32)
∂R R ∂θ ∂z
A.2.3 Laplacian
The procedure by which we will describe ∇ 2 P ≡ ∇ · ∇ P is to use Eq. (A.2.32).This

will entail describing the derivatives of the cylindrical coordinate unit vectors. The
axial direction ēz is constant, and Eq. (A.2.25) indicates that ē R and ēθ depend solely
on θ. Therefore, the derivatives are
∂ ē R ∂ ē R ∂ ē R
= 0, = −ēx sin θ + ē y cos θ = ēθ , =0
∂R ∂θ ∂z
∂ ēθ ∂ ēθ ∂ ēθ
= 0, = ēx cos θ + ē y sin θ = −ē R , =0 (A.2.33)
∂R ∂θ ∂z
∂ ēz ∂ ēz ∂ ēz
= = =0
∂R ∂θ ∂z
The next step is to use Eq. (A.2.32) to describe both the divergence operator and
∇ P. The operation requires that the derivatives be applied to each component of ∇ p
before the dot product is applied. Thus, the operations leading to the Laplacian are

∂ ∂p 1 ∂p ∂p
∇ p ≡ ∇ · ∇ p = ē R ·
2
ē R + ēθ + ēz
∂R ∂R R ∂θ ∂z

1 ∂ ∂p 1 ∂p ∂p
+ ēθ · ē R + ēθ + ēz (A.2.34)
R ∂θ ∂ R R ∂θ ∂z

∂ ∂p 1 ∂p ∂p
+ ēz · ē R + ēθ + ēz
∂z ∂ R R ∂θ ∂z
The unit vectors are mutually orthogonal, and they only depend on θ. Thus, the
nonzero terms arising from the preceding are

∂ ∂p 1 ∂p ∂ ē R 1 ∂ ∂p ∂2 p
∇2 p = ē R · ē R + ēθ · + 2 ēθ · ēθ + 2 ēz · ēz
∂R ∂R R ∂R ∂θ R ∂θ ∂θ ∂z
(A.2.35)
The result of substituting Eq. (A.2.33) is
∂2 p 1 ∂p 1 ∂2 p ∂2 p
∇2 p = + + + 2 (A.2.36)
∂2 R R ∂R R 2 ∂θ2 ∂z
Some individuals prefer to write the Laplacian in a manner that groups the derivatives
with respect to R,

1 ∂ ∂p 1 ∂2 p ∂2 p
∇2 p = R + + 2 (A.2.37)
R ∂R ∂R R 2 ∂θ2 ∂z
Other than brevity, there is little advantage to this form. For example, there is no
power n for which R n f (t − R/c) satisfies the wave equation in a situation where p
is independent of θ.
A.2.4 Velocity and Acceleration
The derivation of velocity and acceleration in terms of cylindrical coordinates follows

the analysis for spherical coordinate. The first step is to construct the position by
examining Fig. A.2. The field point is reached from the origin by traveling R units
in the transverse direction and z units in the axial direction. The angle θ implicitly
affects this construction because it defines the orientation of ē R . Thus
x̄ = R ē R + z ēz (A.2.38)
The axial unit vector points in a constant direction and ē R depends on θ, so

differentiation of the position gives
∂ ē R
v̄ = Ṙ ē R + ż ēz + R θ̇ (A.2.39)
∂θ
The derivative of ē R is described in Eq. (A.2.33). The result is that the velocity is
v̄ = Ṙ ē R + R θ̇ēθ + ż ēz (A.2.40)
In terms of the velocity components, the preceding is
v̄ = v R ē R + vθ ēθ + vz ēz
(A.2.41)
v R = Ṙ, vθ = R θ̇, vz = ż
To describe the acceleration, only the velocity components are considered to

depend explicitly on t, whereas the unit vectors depend on time implicitly through
their dependence on θ. Hence, differentiation of the velocity expression leads to
∂ ē R ∂ ēθ
ā = v̇ R ē R + v̇θ ēθ + v̇z ēz + v R θ̇ + vθ θ̇ (A.2.42)
∂θ ∂θ
Setting θ̇ = vθ /R and collecting like components lead to the acceleration being

described by

v2 v R vθ
ā = v̇ R − θ ē R + v̇θ + ēθ + v̇z ēz (A.2.43)
R R
Appendix B
Fourier Transforms
A Fourier transform may be used to represent a function whose independent variable

extends from minus to plus infinity. A corollary is that a Fourier transform is suitable
for acoustic models of an infinite plane or an infinitely long cylinder. In both cases, the
Fourier transform may be used to describe the dependence of the field on coordinates
that measure position parallel to the boundary. In the time domain, the range for use
of a Fourier transform is −∞ < t < ∞. However, we usually take t = 0 to be the
initial time. We may bypass the range requirement in this case by defining the function
to be zero for t < 0, but doing so requires that we not consider initial conditions.
Moreover, response in the time domain is readily handled by standard methods, as
well the Laplace transforms, which is derived from the Fourier transform. For this
reason, the focus here is application of Fourier transforms to functions of position.
B.1 Derivation
We begin with a function p (z) that is periodic in a distance . The Fourier transform
will emerge as → ∞. The complex amplitude of a plane wave propagating in the
direction of increasing z has been represented as e−ikz , so we use the same convention
for the complex exponential of the Fourier series,
∞
1
p (z) = Pm e−im2πz/ (B.1.1)
2 m=−∞
Because we are dealing with a spatial function, we say that is the wavelength and
2π/ is the fundamental wavenumber. The Fourier coefficients corresponding to a
specified function p (z) are found by evaluating
/2
2
Pm = p (z) eim2πz/ dz (B.1.2)
−/2

DOI 10.1007/978-3-319-56847-8
678 Appendix B: Fourier Transforms
To obtain the limiting behavior as → ∞ we denote the wavenumber of the mth

harmonic as μm = m2π/ and note that the separation between adjacent wavenum-
bers is μ = 2π/. In terms of these parameters, the coefficients are given by
/2
μ
Pm = p (z) eiμm z dz (B.1.3)
π −/2
For a given p (z) and , the quantity obtained from the integral is solely a function of
μm . Let us designate this integral as P̃ (μm ). Accordingly, we shall replace Pm with
(μ/π) P̃ (μm ). This changes the Fourier series to
∞
1
p (z) = P̃ (μm ) e−im2πz/ μ (B.1.4)
2π m=−∞
Increasing decreases μ, but μ is the wavenumber spacing between adjacent val-
ues of P̃ (μm ). Hence, a plot of P̃ (μm ) against its associated value of μm approaches
a curve describing P̃ (μ) as a function of μ. This function is the Fourier transform of
p (z). A variety of notations have been used to denote the operation of Fourier trans-
forming a function. Our choice is F ( p (z) , μ), which explicitly lists the function
that is transformed and the transform variable. Hence, we find that
∞
F ( p (z) , μ) ≡ P̃ (μ) = p (z) eiμz dz (B.1.5)
−∞
To determine p (z) when P̃ (μ) is known we return to Eq. (B.1.4). Because μ

becomes dμ as → ∞, the summation becomes an integral. The result
is the inverse
−1
Fourier transform. This operation is designated as F P̃ (μ) , z , so we have

1 ∞
F −1 P̃ (μ) , z ≡ p (z) = P̃ (μ) e−iμz dμ (B.1.6)
2π −∞
This is a heuristic derivation, but there is a formal mathematical foundation.1 Not

all functions have a Fourier transform. A sufficient condition for its evaluation is
that the integral of | p (z)| be finite. No periodic function, such as a sine or cosine,
meets this requirement, but we will see in the next section that it is possible to
use Fourier transforms to deal with such functions. The inverse Fourier transform
requires values of P̃ (μ) covering the full wavenumber spectrum −∞ < μ < ∞.
However, if the original function p (z) is real, then P̃ (−μ) = P̃ (μ)∗ , so only the
positive spectrum needs to be evaluated.
1 I.N.
Sneddon, Fourier Transforms, Dover reprint (1951).
H.F. Davis, Fourier Series and Orthogonal Functions, Dover reprint (1963).
Appendix B: Fourier Transforms 679
At the outset, we restricted our attention to the Fourier transformation of a spatial

function based on the usage of exp (−iμz) for a wave propagating in the positive
z direction. However, it is quite likely that you will eventually encounter a work in
which the analysis is based on using exp (iμz). Furthermore, it might be appropriate
to employ the Fourier transform to a describe function of time based, for which we
have adopted exp (iωt) for harmonic functions. Both situations require a modification
of the definition of a Fourier transform and its inverse. Let us denote this alternative
definition as ( p (ξ) , μ) where ξ is the independent space or time variable and μ is
the transform parameter. This alternative is
∞
( p (ξ) , μ) ≡ P̂ (μ) = p (ξ) e−iμξ dξ
−∞ (B.1.7)
1 ∞
−1 P̂ (μ) , ξ ≡ p (ξ) = P̂ (μ) eiμξ dμ
2π −∞
When it is necessary to distinguish between the alternatives, we will refer to

Eq. (B.1.5) as the spatial Fourier transform, and Eq. (B.1.7) as the temporal Fourier
transform.
There is much similarity of this alternative definition to the one with which we
began. Therefore, it should not be surprising that given one, the other is readily
recovered. The following relations are readily verified
( p (ξ) , −μ)

= F ( p (ξ) , μ) (B.1.8)
( p (ξ) , μ)∗ = F p (ξ)∗ , μ , F ( p (ξ) , μ)∗ = p (ξ)∗ , μ
Variants on the present Fourier transform definition multiply the integral by different
coefficients. One that is often used leads to a transform and its inverse that have
similar forms. This quality is obtained if the integral in Eq. (B.1.7) or (B.1.5) is
multiplied by 1/(2π)1/2 , which also becomes the factor multiplying the integral for
the inverse transform. Another version, which is not often used, has a 1/(2π) factor
for the transform and none for the inverse. Knowledge of the relationships between
the various definitions is needed when we wish to use a work that has a definition that
differs from our selection. Equation (B.1.8) can be quite useful in this regard because
it allows us to employ any set of tables of functions and their Fourier transforms.
B.2 Evaluation Techniques
Our concern here is evaluation of the Fourier transform of a spatial function p (z), as
well as evaluation of the spatial function when we know a transform P̃ (λ). If p (z)
is not too complicated, we can evaluate the integrals directly, possibly with the aid
of a table of integrals. However, there are alternative methods of evaluation that are
likely to require less effort than that entailed for a formal integration. One is based
on using tabulations of transform pairs rather than of integrals. The other makes use
of FFTs.
B.2.1 Transform Pairs
If a function p (z) is so familiar that its Fourier integral can be obtained with the aid
of a table of integrals, it is likely that the result appears in a table of Fourier transform
pairs. An entry in such a tabulation An entry in such a tabulation will list a function
p (z) and the adjacent entry will list the corresponding Fourier transform P̃ (μ).
Such a tabulation may be used in either direction: from the physical variable that is
a function of z to the transformed function of μ, or vice versa. (Here we are using
z as the independent variable and μ as the transform parameter. Obviously, other
symbols may be used interchangeably.) The first consideration in using a table of
Fourier transforms is to ascertain that it is based on a definition of a Fourier transform
that is consistent with one’s work. If not, then the tabulation will require conversion
as described in the previous section.
Extensive tables of Fourier transforms are available in several books,2 (1954), Ch.
III. as well as some web sites. A condensed table, whose entries tend to be relevant
to acoustic systems, are listed here. Table B.1 gives some basic properties that tell
us how to adapt a known transform pair p (z) and P̃ (μ) to handle related functions.
The last entry in Table B.1 is known as the convolution integral. It seldom is used
to find a transform, but it is quite relevant for performing the inverse. One reason
for this is that it fits a situation in which P̃ (μ) is the transform of an excitation and
Q̃ (μ) is a transfer function. However, we will see in the next section that an FFT
provides a much simpler alternative for evaluating a convolution integral.
Table B.2 lists some specific functions of z and their transform. Several entries
contain the Dirac delta function δ (μ). In each case, the associated function p (z)
does not meet the specification that the integral of | p (z)| be finite, which is why
the transform is singular. One way to verify these transforms is to evaluate the cor-
responding p (z) function as an inverse transform. For example, to verify the entry
for a complex exponential we observe that the integrand for the inverse transform is
2πδ (μ − μ0 ) e−iμz over an infinite range of μ yields 2πe−iμz evaluated at μ = μ0 ,
that is, 2πe−iμ0 z . The integral for the inverse transform is multiplied by 1/ (2π), so
we have shown that F −1 (2πδ (μ − μ0 ) , z) = exp (−iμ0 z).
The entries in Table B.2 may be extended by using the properties in Table B.1.
For example, suppose we wish to transform sin (kz). The Euler identity and linearity
leads to

1 π
F (sin (kz) , μ) ≡ F eikz − e−ikz , μ = [δ (μ + k) − δ (μ − k)] (B.2.9)
2i i
2 A. Erdelyi Bateman Project Manuscript, Vol. I - Table of Integral Transforms (1954) Ch. III.
Table B.1 Basic properties relating an arbitrary function and its Fourier transform
Name Function Fourier transform
∞ ∞
1
Definition p (z) = P̃ (μ) e−iμz dμ P̃ (μ) = p (z) eiμz dz
2π −∞ −∞
Linearity ap (z) + bq (z) a P̃ (z) + b Q̃ (z)
1 μ
Coordinate scaling p (αz) P̃
|α| α
Coordinate shifting p (z − z 0 ) P̃ (μ) eiμz 0
Modulation p (z) e−iμ0 z P̃ (μ − μ0 )
Complex conjugate p (z)∗ P̃ (−μ)∗
dn
nth derivative p (z) (−iμ)n P̃ (μ)
dz n
z 1
Definite integration p (η) dη − P̃ (μ) + π P̃ (0) δ (μ)
−∞ iμ
dn
Differentiate the (i z)n p (z) P̃ (μ)
transform dμn
∞
Convolution p (ξ) q (z − ξ) dξ = P̃ (μ) Q̃ (μ)
−∞
∞
p (ξ) q (z − ξ) dξ
−∞
A particularly useful function is sometimes referred to as the “box” function. It

is defined as box(z) = h (z − a) − h (z − b) with b > a. This function is one for
a < z < b, zero outside that band, and one half at the discontinuities. We could find
the corresponding transform by starting with the rect function, but application of the
linearity and coordinate shifting properties to the definition is more direct. Thus,
F (box (z) , μ) = F (h (z − a) ,
μ) − F (h (z −
b) , μ) = e F (h (z) , μ) − e F (h (z) , μ)
iμa iμb
1 1 iμb
= eiμa − eiμb πδ (μ) − = e − eiμa
iμ iμ
(B.2.10)
where the last form results from δ (μ) being zero except at μ = 0.
Table B.2 Common functions and their Fourier transform

Descriptive name Function Fourier transform
Constant 1 2πδ (μ)
1
Step h (z) πδ (μ) −
iμ
Impulse δ (z − z 0 ) eiμz 0
Complex exponential e−iμ0 z 2πδ (μ − μ0 )
One-sided exponential decay e−αz h (z) , Re (α) > 0
1
Attenuated ramp ze−αz h (z)
(α − iμ)2
e−(z/z0)
2 /2
(2π)1/2 z 0 e−(μz 0 )
2 /2
Gaussian
1 μ
Normalized sinc: sinc(αz) , α > 0 rect
sin (πz) α 2πα
sinc(z) ≡
πz
μz
z 0
Rectangle:
≡
rect(z) rect
z0
z 0 sinc
2π
1 1
h z+ −h z−
2 2
2α
Symmetric exponential e−α|t| , Re (α) > 0
α2 + μ2
B.2.2 Fast Fourier Transforms
For nonstandard functions analytical evaluation of a Fourier transform or inversion

of a transform will require significant analytical effort, possibly involving contour
integration in the complex space. If numerical results are sufficient, the Fast Fourier
Transform (FFT) technology is ideal for that purpose. The presumption here is that
the usage of Fast Fourier Transforms in conjunction with periodic time data, which
was addressed in Sect. 1.4, is a familiar concept. The first objective here is to ascertain
how an FFT computer routine may be used to evaluate the Fourier transform of an
excitation whose spatial dependence is arbitrary. Then, we will examine how FFTs
may be used in conjunction with transfer functions to evaluate an acoustical response.
Whether the Fourier transform of an arbitrary function p (z) exists is a complicated

issue. It is assured whether the function is zero outside some region centered around
the origin.This is the situation for our studies, because the objects we are interested
in have finite extent. More generally, the transform will be finite if the function is
absolutely integrable, that is,
∞
| f (x)| d x < ∞ (B.2.11)
−∞
If we avoid processing periodic functions, this property will be satisfied for finite
acoustical systems.
Usage of an FFT to treat nonperiodic data requires that we define a window whose
width L serves as the effective period. There are two considerations to selecting this
window. The selection begins by identifying the interval −/2 < x < /2 outside of
whichthe data is negligible. For example, if the data is a Gaussian function, p (z) =
a exp −βx 2 , then the criterion | p (z)| < max (| p|) /100 if |x| > /2 would be met
by setting /2 ≥ (ln (100) /β)1/2 . Data that suddenly drops off requires special care
in this regard because the interval should be sufficient to capture the nature of such
data. For example, if p (z) is a rectangle function, then should be substantially
larger than the width of the rectangle.
The value of that is identified should not be the period L. This prohibition stems
from our ultimate objective of using transfer functions to determine the response
to an excitation that has been processed with an FFT. This operation introduces
wraparound error that narrows the interval in which the data is valid. It is possible
to avoid contaminating the response with wraparound error. To do so, the length
of the window must be at least twice . The data is set to zero in the extended
range. This is a process of zero padding. In other words, if the data is such that
| p (z)| < ε max (| p|) /100 if |x| > /2, then the data window for the FFT would be
−L/2 ≤ x ≤ L/2, with L ≥ 2. In the intervals −L/2 ≤ x < −/2 and /2 < x ≤
L/2, we would set p (z) = 0. If we follow this protocol, then FFT results will be
valid for the original data interval, −/2 < x < /2.
Let us first consider the formal definition of a DFT. After we have done so, we
will examine how the definition is adjusted to a form that is suitable for an FFT
algorithm. The period is L, and the data set consists of N samples. Therefore, the
discrete wavenumbers for the Fourier series are

2π
μn = n , n = −N /2 + 1, ..., N /2 (B.2.12)
L
Suppose we knew the function p (z) underlying the sampled data. The Fourier trans-
form of that function would integrate over −L/2 < z < L/2, because we are consid-
ering p (z) to be zero outside that interval. Thus, the Fourier transform at the series
wavenumbers is
L/2 L/2
F ( p (z) , μn ) ≡ P̃ (μn ) = p (z) e dz ≡
iμn z
p (z) ei(2nπ/L)z dz (B.2.13)
−L/2 −L/2
Rather than attempting to carry out this integration analytically, let us consider a
numerical approximation using a strip rule whose width is z = L/N . There are N
strips and the sample locations are z j = jz, j = −N /2 + 1, ..., N /2. The resulting
approximation is
N /2
L N N
F ( p (z) , μn ) ≈ p j ei2π jn/N , n = − + 1, ..., (B.2.14)
N j=−N /2+1
2 2
The significant aspect is that although the definition of a DFT is based on this approxi-
mation, the operation itself exists independently of the concept of a Fourier transform.
Rather, it is a formal prescription of a way in which data may be processed. This
distinction is indicated by using a slightly different notation. A “caret” will denote
DFT data, whereas a “tilda” has been used for a Fourier transform. The definition of
the DFT is
N /2
N N
P̂n ≡ p j ei2π jn/N , n = − + 1, ..., (B.2.15)
j=−N /2+1
2 2

The inverse spatial DFT evaluates the p z j data given a set of DFT values. It is
N /2
1 N N
pj = P̂n e−i2πn j/N , j =− + 1, ..., (B.2.16)
N n=−N /2+1 2 2
One manifestation of the fact that a DFT is an independent operation, separate from a
Fourier
transform, is that the inverse operation is exact. That is, if we process a set of
p z j data to obtain a DFT
set P̂n , then take the inverse DFT of this data, the result
will exactly be the p z j values (aside from numerical errors). No approximation is
involved.
In order to implement an FFT algorithm and its inverse, the spatial data set and
the DFT values are manipulated slightly. One alteration is that the spatial data is
considered to be situated in the window 0 ≤ z ≤ L, which constitutes a forward shift
by half the fundamental wavelength. Shifting the spatial data leads to an algorithm
that evaluates
N −1

Pn
= p j ei2π jn/N , n = 0, 1, ..., N − 1 (B.2.17)
j=0
The notation Pn
is intended to distinguish the FFT output from the DFT values,
which are denoted as P̂n . The shift of the data window is unimportant for our current
purpose. In the event that it is necessary to compare the FFT output to the DFT values
obtained from the formal definition in Eq. (B.2.15), the coordinate shifting property
in Table B.1 for a shift of L/2 multiplies the transform by exp (iμn L/2) = (−1)n .
The other adjustment arises because the range of n appearing in the FFT definition
above differs from the range of n in Eq. (B.2.15). The FFT exploits the periodicity
properties of a DFT to place the data for negative wavenumbers after the data for zero
and positive wavenumbers. This is the complex conjugate mirror image property first
described in Eq. (1.4.40). The rearrangement is described by
Pn
= P̂n , n = 0, ..., N /2
(B.2.18)
Pn
= P̂n−N , n = N /2 + 1, ..., N − 1
The rearrangement of wavenumbers follows the same rule
2πn
μ
n = μn = , n = 0, ..., N /2
L (B.2.19)
2π (n − N )
μ
n = μn−N = , n = N /2 + 1, ..., N − 1
L
This shift in the wavenumber spectrum is a crucial issue for the objective of using
transfer functions to find a response.
Any routine that performs a Fast Fourier Transform will be accompanied by one
that performs the inverse transform.FFT definition in Eq. (B.2.15) is
N −1
1
pn = P̂ j e−i2πn j/N , n = 0, 1, ...N − 1 (B.2.20)
N j=0
We will refer to a computer routine that implements the definition in Eq. (B.2.15)
as a spatial FFT, whereas the FFT definition in Sect. 1.4.4 will be said to be a a
temporal FFT. (The only difference is the sign of the argument of the complex
exponential.) If the available software package only has a routine that implements a
temporal FFT, it will be necessary to adjust its output. One way to do so is to use the
first of Eq. (B.1.8), which calls for swapping the negative and positive index of the
DFT data. This can be somewhat problematic because of the arrangement of data for
FFT algorithms, as described by Eq. (B.2.18). An alternative is to follow the last of
Eq. (B.1.8) which calls for using the available FFT routine to transform the complex
conjugate of the spatial data, then taking the complex conjugate of that output.
The preceding addresses the fundamentals of FFT processing. Let us now turn our
attention to the reason that an FFT might be useful for acoustical system analysis.
Suppose we have used an FFT algorithm to transform the spatial dependence of
an excitation. Examples of this would be transformation of an arbitrary vibrational
pattern along one direction of an infinite plane, or transformation of the axial pattern
of vibration on the surface of a circular cylinder. The transfer function in the first
case would be the acoustic wave radiated into the fluid by a sinusoidal surface wave,
which was determined in Chap. 5. In the case of a cylinder, an analysis in Chap. 7
found the wave radiated into the fluid by a helical wave surface wave at a fixed
circumferential harmonic number. For both systems, the transfer function is the
amplitude of a wave that is radiated due to an excitation that propagates along the
surface in a certain direction, which we will designate as x. The wavenumber in that
direction is designated as μ. Let us denote the amplitude of the wave that propagates
in the x direction as G (μ). To be a transfer function, the excitation causing it must
be a wave in the same direction with unit amplitude. Our task is to determine the
signal when the excitation is an arbitrary function.
Let us analyze radiation from a plane to illustrate the process. Suppose the surface
vibration is harmonic at frequency ω. The distribution in the x direction is arbitrary,
such that the surface velocity normal to the plane is Re (v (x) exp (iωt)), where v (x)
may be complex. It is necessary that there is a DFT window L that accommodates
the significant portion of v (x). The transfer function is the y dependence of the wave
radiated from the surface when v (x) = exp (−iμx). This quantity is described by
Eq. (5.1.16), with the velocity amplitude iωW set to one. Thus, the transfer function
is such that
k
P = G (μ) e−iμx , G (μ) = ρ0 c e−iκy
1/2 κ
k 2 − μ2 if k > μ (B.2.21)
κ= 2 1/2
−i μ − k 2 if k < μ
If the amplitude of the surface vibration is V (μ) = 1, then the wave radiated from
the surface would be V (μ) G (μ) exp (−iμx). Addition of the contribution of the
full spectrum of V (μ) leads to

p (x, y, t) = Re P (x, y) eiωt (B.2.22)
where the complex pressure amplitude is given by

∞
1 ∞ −iμx k (μ) −iκ(μ)y
P (x, y) = V (μ) G (μ) e dμ = ρ0 c V (μ) e e−iμx dμ
2π −∞ −∞ κ (μ)
(B.2.23)
This expression is an inverse Fourier transform. It is readily adapted to fit the

definition of an inverse FFT in Eq. (B.2.20). First we evaluate the FFT of the excitation
distribution V (μ), which produces velocity coefficients Vn
. The FFT of the pressure
is analogous to the integrand, specifically,
k
−iκn y
Pn
= ρ0 c V e , κn = κ (μn ) (B.2.24)
κn n
The last step is to take the inverse FFT of the Pn

data. The output of the inverse FFT
will be values of P (x, y) at fixed y at the points xn used to sample v (x). The only
precaution for implementing this procedure is associated with the manner in which
the DFT data is arranged in an FFT algorithm. Specifically, the first N + 1 values
are the associated with λn ≥ 0. These values are followed by the data for λn < 0.
The wavenumbers corresponding to the data that is input to the inverse FFT are
⎧
⎪ 2nπ N
⎨ , n = 0, 1, ...,
μn = L 2 (B.2.25)
⎩ 2 (n − N ) π , n = N + 1, ..., N − 1
⎪
L 2
If the wavenumbers are arranged in this sequence as a vector array {μ} in a
computer program, all operations may carried out vectorially, for example,
kappa=conj(sqrt(kˆ2-mu.ˆ2)) in MATLAB.
Although this FFT procedure is fairly straightforward, there is a profound diffi-
culty that can inhibit its application for acoustics. The system we have set up consists
of an interval along the boundary within which there is a vibration. This is the two-
dimensional analog of a piston in an infinite baffle. In both cases, increasing distance
from the boundary leads to ever increasing spreading of the acoustic field parallel
to the surface. The inverse FFT that generates this distribution must have a window
width that accommodates the spreading effect. In other words, the window length
must be selected on the basis of the extent of the signal at the selected distance y
to the boundary, rather than the nature of the surface vibration. This is where the
difficulty arises, because the extent of the field is not known until the system has
been analyzed.
Index
A Axial wavenumber, 57, 217, 267

Absorbing boundary conditon, 123 complex, 229
Absorption coefficient, 607 Axisymmetric body, scattering by, 497
Absorptive liner, 224 Axisymmetric field, 19, 61, 71
Acceleration Azimuthal angle, 94, 140, 353
in cylindrical coordinates, 675 Azimuthal harmonic, 16, 268, 351
in spherical coordinates, 670
Acoustical imaging, 656
Acoustically compact, 99 B
Acoustic cavity, 85 Backscatter, 486, 502, 509, 511, 514
Acoustic levitation, 655 Backscatter cross section, 510, 512, 514, 517
Acoustic Mach number, 546, 553, 611 Backward propagating wave, 199, 292, 298,
Acoustic perturbation, 409 548
Acoustic-structure interaction, 238 B/A coefficient, 551
Added mass, 103, 501, see also Virtual mass Baffle, 29, 133, 160
Adiabatic assumption, 541 Beam aspect, 97
Algorithm Beamwidth, 148, 162
for a modal series, 252, 264, 282, 308 Beating signal
for a nonlinear wave, 559, 563, 633 for group velocity, 202
Ambient conditions, 409, 412 from a supersonic source, 242
Amplitude nonlinear, 607
dispersion, 545, 645 Bending of rays, 420
slowly varying, 407, 428, 443 Bessel function
Angular spectrum, 176 cylindrical, 12, 57, 145
Annular waveguide, 275 derivative of, 14, 270
Antinode, 154 for modes, 268, 271
Anti-piston transducer, 156 integral for, 570
Area function modified, 64, 273
for a multivalued waveform, 585 of negative order, 62
for the Webster horn equation, 191 spherical, 11, 15, 57, 312, 319
Arrête, 417, 434 Bessel’s equation, 57
Asymptotic approximation modified, 64, 273
of a spherical harmonic series, 21 Biomedical applications, 657
of Bessel functions, 58, 270, see also Blocked area, 491, 513, 520
Farfield approximation Blocked pressure, 522, 527
Atmosphere, propagation in, 406 Blue sky, 491
Attenuation, 230, 521, 596, 607 Body force, 409
DOI 10.1007/978-3-319-56847-8
690 Index
Born approximation, 485, 490 Conservation

Boundary condition of energy, 572, 579
for a cavity, 243, 300, 313 of mass, 189, 541, 579
for a perturbation series, 614, 639 Constitutive nonlinearity, see Equation of
for a waveguide, 188, 196, 214, 218 state
for farfield radiation, 125 Continuity condition
mixed, 244 at a cavity wall, 361, 380
nonlinear, 638 at a vibrating plate, 638
Boundary elements, 111, 363 on the surface of an elastic shell, 39
Box function, 681 Continuity equation, 409, 540, 598
Branch cut, 218, 226, 246 Control volume, 190, 578
Breathing mode, 20 Convective nonlinearity, 544
Bulk modulus, 40, 485, 491, 551 Convergence
Burton-Miller formulation, 115 of a Legendre series, 8, 26
of a modal series, 42, 47, 250, 262
of a spherical harmonic series, 22, 526
C of Rayleigh-Ritz method, 374
Calculus of variations, 365, 463 Coordinate straining, 621, 629, 642
Cauchy principal part, 109, 113 Corner, for Kirchhoff–Helmholtz integral
Causality, principle of, 451 theorem, 110
Caustic, 417, 432 Coupled equations
effect on amplitude, 450 for a spherical shell, 40
example, 454 for cavity-structure interaction, 385
formed by nonlinear rays, 651 for waveguide-structure interaction, 238
Cavitation, 577, 656 Creeping waves, 520
Cavity, 91, 93, 113, 291 Critical incidence, 471
Cavity mode, 293, 295, 297, 336, 365 Cross-sectional area
Center of curvature, of a ray path, 423 of a horn, 189
Center wave, 153, 179 of a ray tube, 445
Chaladni lines, 346 Cumulative growth, 628
Characteristic curve, 542, 544 Curvature of rays, 423
Characteristic equation, 219, 225 Cutoff, 259
for a cylinder, 325 for an exponential horn, 194, 204
for a spherical shell, 41 for radiation from a cylinder, 68
for a waveguide, 219, 224, 238, 259 of waveguide modes, 218, 260
CHIEF computer code, 114 Cylindrical Bessel function, see Bessel func-
Circular disk, scattered field, 508, 513 tion
Circular ray path, 424, 443 Cylindrical coordinates, 56, 144, 267
Circumferential angle, 671 definition of, 671
Circumferential harmonic, 69, 77, 271, 323 unit vectors, 672, 674
Coefficient of nonlinearity, 552 Cylindrical spreading, 63
Coincidence frequency, 236
Compact set of sources, 18
Complex eigenvalues, 225 D
Complex wavenumber, 226 D’Alembert solution, 544, 548, 614
Compliant boundary, 273, 298, 326, 379, Decay constant, 220, 246
386 Decomposition of a propagating mode, 222
Concave reflector, 512 Degree of a Legendre function, 4, 16
Concentric cylinders, 352 Derivative of a unit vector
Concentric spheres, 319, 360 for cylindrical coordinates, 675
Conical horn, 188, 191 for spherical coordinates, 668
Connectivity matrix, 117 Differential scattering cross-section, 509,
517
Index 691
Diffraction, 152, 473, 520 for a liquid, 550

Dipole, 495 for an ideal gas, 549
for scattering, 491 relation to sound speed, 593
two-dimensional, 78 Equations of motion
Dipole moment, 78, 100, 101 for a spherical shell, 36
Dirac delta function, 295 for a structure, 383
Directivity Error residual, 106
for a transducer in a baffle, 137, 144, 166 Euler’s equation, 190, 381, 541, 578
for radiation, 53, 97, 147 Euler-Lagrange equation, 468
for scattering, 484, 517 Evanescent mode, 222, 248, 260, 262
from a spherical harmonic series, 21 Evanescent wave, 195, 202
of a horn, 187, 196 Exponential horn, 194
of a ring, 169 group velocity for, 204
Disk Extensional elastic wave, 525
radiation from, 103
scattering from, 497, 501
Dispersion, 195, 202 F
Displacement Farfield
of a plate, 637 for a transducer, 127, 133, 144, 149, 154,
of a spherical shell, 37, 522 163
structural, 383 for radiation from a cylinder, 69
Dissipation, 229, 567, 607 for radiation from a spherical shell, 41,
Divergence theorem, 86, 338, 340 52
Dome tweeter, 29 of a nonlinear spherical wave, 632
Dowell’s approximation, 380 of a spherical harmonic series, 21
of Kirchhoff–Helmholtz integral theo-
rem, 94
E of the Rayleigh integral, 144
Earnshaw solution, 540, 546 Fast Fourier Transform, 682
Edge wave, 154, 179 for radiation, 84, 139
Eigenray, 458 Fermat’s principle, 463
Eigenvalue FFT, see Fast Fourier Transform
complex, 225 Finite amplitude wave, see Nonlinear wave
for a rigid-walled cylinder, 218, 271 Finite elements, 115, 363
for a waveguide, 218, 225 Fitting shocks, 582, 587
purely imaginary, 228, 274 Flexible wall, 42, 224, 236, 379
Eigenvalue problem, 337 Fluid impedance, 44, 525
Eikonal equation, 439, 440 Fluid-loaded resonance, 51, 526
Elastic plate, 230, 637 Fluid-loading effect of, 43
Elastic spherical shell, 36, 523 Fluid-structure interaction, 36, 243, 379
Elastic structure, 380 Focus, 431, 435
Elliptic cavity, 374 Focusing factor, 431
End-fire array, 657 Force
Energy absorption, 300 acting on a sphere, 38
Energy and work, 579 exerted on a piston, 166
Energy functionals, 366 for low frequency radiation, 102
Energy, in the old-age stage, 594 Forward and backward waves
Entropy, 581 in a cavity, 292, 298
Equal area rule, 582, 592 in a horn, 194, 199
Equal eigenvalues, 261 Forward characteristic, 544
Equal natural frequencies, 339, 345, 354 Forward scattering, 509, 520
Equation of state Fourier coefficient, 677
for a heterogeneous fluid, 410 Fourier series, 677
692 Index
for a cylindrical waveguide, 267 in spherical coordinates, 2

for a nonlinear wave, 569, 603 inhomogeneous, 135
for radiation from a cylinder, 56, 69 Hemisphere, in a baffle, 29, 165, 198
Fourier transform, 82, 138, 177, 677 Hermitian, 106
alternative definitions, 679 Heterogeneity, 409
conditions for, 678 arbitrary dependence, 438
of a surface velocity, 82, 139, 176 cylindrical region, 451
table of, 679, 680 scattering from, 484, 491
Fubini-Ghiron, 571 vertically stratified, 418
High frequency approximation
for a spherical wave, 24
G for cylindrical waves, 71
Galerkin method, 116, 118 for geometrical acoustics, 407
Gamma function, 169 Horizontal tangency of a ray path, 421
General eigenvalue problem, 373 Horn, 187
Generalized coordinates, 383 conical, 188, 191
Generalized force, 383 exponential, 193
Generalized Fourier series, 6 WKB method, 207
Geometric boundary condition, 366 Hyperbolic function, 226
Geometrical acoustics, 405, 511
Gibb’s phenomenon, 174
Gradient
I
for scattering, 492
Ideal gas law, 549
in cylindrical coordinates, 672
Images, method of, 30, 135
in spherical coordinates, 665
Impedance boundary condition
Graphical construction
of a characteristic equation, 228 in a cavity, 330
of a spatial profile, 553 in a waveguide, 254, 259, 273
of a waveform, 554 orthogonality condition with, 254
of wavefronts and rays, 408 Impedance tube, 326
Grazing angle, 418 Imposed boundary condition, 366
Green’s function, 85, 481 Impulsive source, 347
for a cavity, 295, 296 Incident wave, 485
for a rigid plane, 135 Index of refraction, 418, 465
free space, 87 Inertia matrix, 120, 383
two-dimensional, 72, 87 Infinite elements, 123
Green’s theorem, 118, 381 Inhomogeneous Helmholtz equation, 294
Group velocity, 202, 222, 233 Inhomogeneous wave equation, 340, 381
Initial conditions, for a nonlinear wave, 548
Integral equation, 111
H Intensity,
Half-power point, 148 for a spherical harmonic series, 22
Hamilton’s principle, 383 for radiation of a hemisphere, 31
Hankel function from a baffled piston, 148, 166
cylindrical, 58, 268 from a spherical harmonic series, 22
spherical, 11, 19 from a vibrating cylinder, 70
Hankel transform, 146 in a ray tube, 429, 437, 445
Harmonic amplitudes, nonlinear wave, 567, in a scattered wave, 512, 514
568, 603 in a waveguide, 196, 247, 256
Heavy fluid loading, 394 in the old-age stage, 595
Helical surface wave, 61, 66, 69 radiated by a source, 429
Helmholtz equation, 10, 85, 217 using a farfield approximation, 171
for scattering, 480, 492 Interaction of harmonics, 175, 606
in cylindrical coordinates, 56, 267 Interference
Index 693
in a sound beam, 154, 161, 173 Long range communication, 656

in a waveguide, 260, 266 Loudspeaker, 133, 144, 332
in scattering, 521
Interpolating function, 111, 116
Inverse Fourier transform, 82, 678
M
Mass acceleration, 72, 74, 86, 105
J Material derivative, 410
Jacobian, 448, 646 Material nonlinearity, 544
Jump conditions, 580 Matrix equation
for a cavity, 309, 350
for a waveguide, 250, 264, 266
K for finite elements, 120
KHIT, see Kirchhoff–Helmholtz integral for radiation, 106, 113, 120
theorem for ray tracing, 440
Kinematical nonlinearity, 599
for Rayleigh-Ritz method, 373
Kinetic energy, 364
Membrane shell model, 36
Kirchhoff approximation, 510
Modal amplitudes, see Modal series
Kirchhoff–Helmholtz integral theorem
Modal density, 346
evaluation at a surface point, 90
farfield approximation, 93, 121 Modal series
for a cavity, 85 for a cavity, 293, 298, 300, 313, 324, 331,
for a cylinder, 94 340
for a half-space, 134 for a circular waveguide, 243, 246, 255,
for a spherical cavity, 88 279
for radiation, 91 Modes, 336
for scattering, 481 of a cylindrical cavity, 352
multipole expansion of, 99 of a cylindrical waveguide, 272
of a rectangular cavity, 342
of a spherical cavity, 358
L of a spherical shell, 41
Lagrange’s equations, 383 of a spherical waveguide, 272
Laplace equation, 492 two-dimensional, 217, 224, 235
Laplacian Modified Bessel function, 64, 273
in cylindrical coordinates, 674 Momentum equation
in spherical coordinates, 668 for a heterogeneous fluid, 410
Legendre function, 4, 5, 8, 16, 22 for a horn, 190
Legendre polynomial, 4 for a shock, 578
Legendre series, 6, 8 in terms of a velocity potential, 597
Legendre’s equation, 3
nonlinear wave, 544, 598
Leibnitz’ rule, 212, 580, 583, 585
Monopole, see Green’s function, Point
Light fluid loading, 394
source
Lighthill’s method, see Strained coordinates
Monopole amplitude, 73, 78
Line source, 72
finite length, 73 for scattering from a rigid body, 450, 501
Linear least squares, 107, 114 Monostatic scattering measurement, 505
Lithotripter, 657 Morse–Penrose inverse, 107
Locally planar approximation, 408 Moving boundary condition, 612, 638
Locally reacting importance of, 547
cylindrical wall, 330 Mulitvalued waveform, 565, 577
effect on cavity resonance, 305 Multipath propagation, 458
end, 323 Multiple scales, method of, 623
for a waveguide, 253 Multipole expansion, 99, 130, 496
for geometrical acoustics, 511 Mutual scattering, 480
694 Index
N of axial mode functions, 330

Natural frequency of cavity modes, 338, 365
of a cylindrical cavity, 330, 353 of harmonic functions, 604
of a rectangular cavity, 291, 301, 342 of Legendre functions, 31
of a spherical cavity, 311, 358 of transverse mode functions, 245, 255,
of a spherical shell, 41, 523 260, 278
Neumann function Output from a horn, 188
cylindrical, 58, 268, 353
spherical, 12, 15, 319
Newton’s method, 563 P
Nondispersive wave, 202 Parametric arrays, 657
Nonlinear propagation Parseval’s theorem, 574
in two-dimensions, 637 Particle velocity
of a plane wave, 540 according to geometrical acoustics, 445
of a spherical wave, 625 for a rigid body, 101
Nonlinear wave equation, 598, 604, 611, 623 for a set of sources, 106
Nonlinearity, effects of, 568, 574, 577 in a cylindrical wave, 62
Nonplanar transducer in a baffle, 163 in a nonlinear wave, 544, 548, 551
Nonuniform plane wave, 221 in a ray tube, 429
Nonuniform validity, 621, 641 Perfectly transmitting boundary conditon,
Normal direction 123
to a surface, 86, 91, 413 Perturbation series, 612, 626, 639
to a vibrating plate, 638 Phase delay, due to a caustic, 450
to a wavefront, 408 Phase speed
Normalization for a plate, 38
of cavity modes, 339 for a waveguide, 223, 228, 243
of transverse mode functions, 245, 255, in a horn, 195, 202, 204, 207, 208
260, 278 in ray tracing equations, 438
Normal velocity, 133, 638 nonlinear, 543, 647
Numerical method Phase variable, 209, 603, 644
for a horn, 213 Phase-reversal, 556
for a nonlinear wave, 559, 563, 649, 652 Piecewise linear waveform, 553, 574
for characteristic equation, 227, 231 Piston
for nearfield of a sound beam, 158 at the end of a cylinder, 279
for ray tracing equations, 440, 452 in the time domain, 176
for ray tube area, 448 nearfield, 158
for transport equation, 452 on-axis pressure, 153, 159
Nyquist criterion, 174 radiated power, 171
for a nonlinear waveform, 567 radiation impedance, 166
Plane wave
for group velocity, 202
O in a horn, 193
Oblique plane waves, 123, 222 in a waveguide, 215, 221, 268
Ocean, propagation in, 405 nonlinear, 541
Old-age stage, 594 refraction by a cylindrical heterogeneity,
Omnidirectional source, 430 451
Order Plane wave approximation, 62, 123, 149
of a Bessel function, 58 Plane wave mode, 220, 225, 261
of an associated Legendre function, 16 Point source, set of, 104
of Neumann and Hankel functions, 11, Poisson solution, 540
59 Position
Order of magnitude, definition, 612 in cylindrical coordinates, 671
Orthogonality in spherical coordinates, 665
Index 695
Potential energy, 36, 120, 364 for a nonlinear wave, 644, 647
Potential function, see Velocity potential graphical construction of, 408
Power flow, 256 in vertically stratified fluid, 417
in a ray tube, 430, 446 reflected, 223, 413, 421
in a waveguide, 248, 257 transmitted, 414
Power input, 171, 197, 248, 303, 430 Ray tracing equations, 466
Power, Radiated derivation, 438
at an interface, 256 numerical solution of, 440, 452
by a hemisphere, 32 relation to Fermat’s principle, 467
for a spherical harmonic series, 24 Ray tube
from a baffled piston, 166, 171 definition, 428, 445
from a vibrating cylinder, 72 Jacobian for, 448, 459
in a scattered wave, 506, 512 Rayleigh distance, 146, 155, 162
using a farfield approximation, 172 Rayleigh integral
Pressure application for nonplanar surface, 146,
amplitude along a ray, 446 163
from the ray tracing equations, 459, 497 derivation, 134
in a cylindrical wave, 62 farfield approximation of, 138, 144
in a spherical cavity, 15 for a nonplanar transducer, 163
in terms of strained coordinates, 621, for a surface point, 167
630, 643 for an axial field point, 153
in terms of velocity potential, 598, 599 in Cartesian coordinates, 137
relation to particle velocity, see Euler’s in cylindrical coordinates, 144
equation numerical approximation of, 158
scattered, see Scattered pressure time domain version, 177
Pressure loading, of a sphere, 38, 497, 524 Rayleigh ratio, 364, 370
Pressure-release Rayleigh-Ritz method, 370
baffle, 29 Rayleigh scattering limit, 490, 508, 517, 519
end correction, 133 Reactive boundary, 227, 247
wall, 226, 353 Reciprocity, 167, 388
Principal curvature, 511 Rectangular cavity, 291, 301, 342
Product, of Fourier series, 604 Rectangular waveguide, 257, 298, 625
Pulse, 596 Recurrence relation
for a gamma function, 169
for Bessel functions, 270
R for Legendre polynomials, 6, 24
Radial symmetry, 23, 625 for modified Bessel functions, 65
Radiation for second order velocity potental, 629
from a piston, 157 for spherical Bessel functions, 14, 65
from a spherical shell, 522 Reflecting telescope, 415
from a vibrating plate, 218, 637 Reflection
from a vibrating sphere, 15, 19 effect on shadow zones, 435
from an infinite cylinder, 55 from a curved surface, 413, 511
Radiation damping, 45, 122, 525 from a free surface, 421, 459
Radiation impedance in a hard-walled waveguide, 223
of a piston, 166 of waveguide modes, 298
of an infinite cylinder, 63 Reflection coefficient
specific, 170 from a curved surface, 414, 450, 511
Radiation pressure, 655 modal, 300
Range determination, 506 Refraction
Rankine–Hugoniot conditions, 580 according to geometrical acoustics, 413
Ray Renormalization, see Strained coordinate
definition, 408 Resonance, 296, 300, 325
696 Index
effect of fluid-loading, 44, 522, 526 SHIE, 111, see Surface Helmholtz integral
in a cavity, 296, 300, 322, 325, 393 equation
of a mechanical oscillator, 44, 526 Shock
of a spherical shell, 43, 46 condition for, 575, 646
Retarded time, 569 for a sinusoidal input, 577
Riemann solution, 541, 623 formation, 539, 555, 562, 564, 574
compared to nonlinear spherical wave, in a spherical wave, 632
632 in a two-dimensional wave, 651
energy conservation, 572 location, 550, 577, 582, 631, 646
evaluation of, 553, 559, 564 old age, 594
fitting shocks to, 588 Shock formation, 545
for initial conditions, 548 Short wavelength behavior, 22
interpretation, 543 Side lobe, 98, 142, 147, 519
Rigid body, scattering from, 494, 499, 507 Sinc function, 141
Rigid surface, 337, 353 Singular perturbation, see Strained coordi-
Rigid walled waveguide, 218, 271 nate
Rigid-walled cavity, 336, 353 Slowly varying function, 603
Ring, vibrating, 145 Slowness, see Wave slowness
Rodrigue’s formula, 6 Small signal approximations
for linearization, 409
for nonlinear propagation, 554
S for shock formation, 575
Salinity, 412 Snell’s law
Saturation, 595 at an interface, 414
Sawtooth waveform, 588, 592, 595 for vertical heterogeneity, 418, 424
Scattering from Fermat’s principle, 463
Born approximation, 485 Solid angle, 110, 508
by a disk, 497, 501, 502 Sommerfeld radiation condition
by a movable body, 499 for a spherical wave, 15
by a sphere, 497, 515 for radiation, 92
due to strong heterogeneity, 497 for scattering, 480, 517, 522
from a rigid body, 494 for the Kirchhoff–Helmholtz integral, 91
governing equations, 492 from a plate, 638
Rayleigh limit, 490 in a waveguide, 218, 254
Scattering cross section, 507, 512, 517 Sonar, 505
Scattering directivity function, 519 Sound beam, 142
Scattering volume, 491 axial pressure, 156
Second orthogonality condition, 339 farfield pressure, 142, 146
Selection of trial functions, 367 intensity, 148, 171, 175
Self-refraction, 645 nearfield, 160
Separation of variables, 2, 9, 16, 53, 56, 217 nonlinear, 655
Separation theorem, 374 qualitative description, 162
Series Sound speed in a heterogenous fluid, 412,
for structural displacement, 383 418, 439, 492
of forced cavity modes, 297, 312, 323 Source
of natural cavity modes, 342 in a cavity, 291, 381, see also Point
of rigid cavity modes, 381 source
of spherical harmonics, 17, 19 Source distribution, 105, 341
of transverse mode functions, 245, 298 for surface vibration, 178
of trial functions, 370 Source strength, see Monopole amplitude
Shadow zone, 433, 435, 521 Source superposition method, 104
Shedding, of creeping waves, 521 Spatial Fourier transform, 679
Shell theory, 37 Speaker cone, 148
Index 697
Specular reflection, 521 Tensor, 102

Sphere Throat impedance, 197
radially vibrating, 19, 103, 625 Time delay, 177, 443
scattered field, 492, 496, 513, 516, 522 Time domain Rayleigh integral, 177
Spherical aberration, 417 Time, along a ray, 420, 440, 442
Spherical cap, 188 Trace velocity, 222
Spherical cavity, 311, 358 Trace wavenumber, 63, 139
Spherical coordinates, 137, 665 Transducer impedance, 248
for an elastic shell, 37, 515 Transfer function, 683
for scattering, 483 Transient response, 296
Spherical Hankel function, 11, 19, 319, 516 in a cavity, 296
Spherical harmonics, 36, 493 of a piston, 176
Spherical spreading, 10, 62, 92 Transmission coefficient, 414
Square wave Transport equation, 444
distortion of, 592 Transverse distance, 671
radiation from a piston, 172, 180 Transverse mode function, 219
Standing wave, 223, 268, 296 coupling with an elastic plate, 235
State vector, 386 relation to forced cavity modes, 254, 258,
Stationarity 259, see alsoMode function
of a variational path, 463 Transverse wavenumber, 82, 353
of the Rayleigh ratio, 371 Trial function, 365, 367
Stiffness matrix, 120, 383 Truncation
Strained coordinates, 621, 642 for Dowell’s approximation, 394
Stringed musical instrument, 79 of a Fourier series, 602
Structural acoustics, 1, 379 of a modal series, 280, 350
Struve functions, 169 of a perturbation series, 612
Sturm–Liouville analysis, 244 of a spherical harmonic series, 23, 517,
Subsonic surface wave, 63 525
Supercritical incidence, 414 of equation of state, 551
Superposition Turning point, 421
of plane waves, 222, 260
of rays, 459
of standing waves, 293 U
Supersonic surface wave, 63, 242 Ultrasonic imaging, 505, 656
Surface Helmholtz integral equation, 108, Underwater acoustics, 418
111 Unit vector
derivation, 108 in cylindrical coordinates, 672
discretization of, 113 in spherical coordinates, 666
Symmetry Upper bound, for natural frequency, 366
in a rectangular waveguide, 236, 266, 306
of a ray path, 422
of cavity-structure equations, 387 V
radial, 22, 625 Variational path, 463
Velocity
in cylindrical coordinates, 675
T in spherical coordinates, 669
Table of a shock, 581
of Fourier transform pairs, 681, 682 Velocity potential, 598, 605, 623, 626
Tangential stiffness, 543 Vertical stratification, 417
Target identification, 505 Vertical tangency, of a waveform, 562, 576,
Target strength, 510 578
Taylor series, 170, 410, 465, 543, 551, 611 Vibrating cylinder, 56
Temporal Fourier transform, 679 Vibrating pipe, 71
Vibrating plate, 637
698 Index
Virtual displacement, 384, 464 with compliant walls, 224

Virtual mass, 103, 502 Waveguide mode, see Transverse mode
Virtual sources, 657 function
Virtual work, 383 Wavenumber
Volume velocity, 340 axial, 217
for multiple sources, 100, 105 complex, 195
of a piston, 151, 166 dispersion equation for, 202
of a source, 23, 72, 74, 105 for a cylindrical wave, 62
for a spherical shell, 37
for a surface vibration, 137
W for an arbitrary surface vibration, 137
Water, standard properties of, 412, 551 vector, 61, 137, 221
Watson transformation, 521 Webster horn equation, 191, 209
Wave equation criteria for using, 193
for heterogeneous fluid, 412 derivation, 189
nonlinear, 600 numerical solution of, 212
Wave packet, 204 Window
Wave slowness, 439
for Fast Fourier Transform, 683
Waveform, nonlinear, 539, 545, 553, 559,
WKB method, 207
633, 651
Woofer, 332
Wavefront
Work-energy principle, 364
according to geometrical acoustics, 407,
Wraparound error, 683
438, 448
Wronskian, 524, 527
definition, 406
for a nonlinear wave, 644
Waveguide
circular, 267 Z
end correction, 199 Zero frequency mode, 367
three-dimensional, 253 Zero padding, 683
two-dimensional, 217 Zeros of a Bessel function, 151

Acoustics-A Textbook For Engineers and Physicists Volume II Applications

Uploaded by

Copyright:

Available Formats

Acoustics-A Textbook For Engineers and Physicists Volume II Applications

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

Acoustics-A Textbook For Engineers and Physicists Volume II Applications

Uploaded by

Copyright:

Available Formats

Jerry H.

ISBN 978-3-319-56846-1 ISBN 978-3-319-56847-8 (eBook)

© Springer International Publishing AG 2018

Printed on acid-free paper

This Springer imprint is published by Springer Nature

Mark F. Hamilton (Chair), University of Texas at Austin

The Basic Concept

Helmholtz integral theorem is developed in Chap. 7. From the opening motivational

Dunwoody, GA, USA Jerry H. Ginsberg

I am indebted to many individuals for providing motivation to write these books.

Dunwoody, GA, USA Jerry H. Ginsberg

2.4 The d’Alembert Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

4.3 Plane Waves in Three Dimensions . . . . . . . . . . . . . . . . . . . . . . . 306

6.4.3 Point Source Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . 483

Volume II: Applications

8 Radiation from a Source in a Baffle . . . . . . . . . . . . . . . . . . . . . . . . . 133

10.4 Approximate Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363

13.3 General Analytical Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . 597

Jerry H. Ginsberg began his technical education at

ASME Per Bruel Gold Medal in Noise Control and

Example 7.1 (Legendre_function_series.m) Evaluation

Example 7.7 (Line_source_ﬁnite.m) Determination of the pressure

Example 9.1 Usage of the general solution of the Webster horn

Example 9.9 (Piston_circular_waveguide.m) Determination

Example 10.10 (none) Analysis of the modal properties of a system

Example 11.6 (none) Usage of geometrical acoustics to evaluate

Example 13.5 (Frequency_domain_nonlin_ODEs.m) Solution of the

7.1 Spherical Harmonics

7.1.1 Separation of Variables

It is helpful to rewrite the Helmholtz equation in axisymmetric spherical coordinates,

The derivatives become

Each value of m leads to a different solution of Legendre’s equation because m is the

−0.5 −0.5 m=7

0.5 m=6 0.5 m=5

−0.5 m=4 −0.5 m=7 m=3

(m + 1) Pm+1 (η) = (2m + 1) η Pm (η) − m Pm−1 (η)

A different definition of the polynomials is offered by Rodrigue’s formula,

An important property of Legendre polynomials is that they are mutually orthog-

The similarity of this representation to a Fourier series causes some individuals to

EXAMPLE 7.1 Legendre polynomials have been used to describe crystal

There is no simple way to evaluate this integral analytically, so we turn to numer-

Each is governed by the first of Eq. (7.1.4), which now is

At this juncture, there are a number of ways by which F may be determined.

The differential equation becomes

This is promising because it tells us that tends to a harmonic function at larger r.

A different way of viewing these definitions is that

The lowest order functions are

The trends for large arguments are

7.1.2 Description of the Pressure Field

To specialize this expression to the farfield, we use the asymptotic representation of

The pressure in a spherical cavity is a topic in Chap. 10. As a preliminary to

7.1.3 Arbitrary Spatial Dependence

P = F (r ) G (ψ) (θ) (7.1.41)

Pmn (cos ψ) = 0 if n > m (7.1.43)

EXAMPLE 7.2 Consider an axisymmetric field consisting of a monopole,

We could evaluate the integral numerically as we did in the previous example.

B0 = −ik A + ik 3 Q 33 , B1 = −ik 2 D3 , B2 = − (2/3) i Q 3,3

7.2 Radiation from a Spherical Body

Determination of the field radiated by an axisymmetrically vibrating sphere is our

Because eika ≈ 1 when ka 1, Eq. (7.2.21) reduces to this expression. Hence, we

(V2m )rigid i(2m+1)π/2

(Vm )ref i(m+1)π/2

formula is halvedbecause a baffle reduces the domain to a half-space. The results

Thus, there are two natural frequencies (1) (2)

Figure 4 explains the m = 1 behavior. The lower natural frequency is (1) 1 = 0,

In the second expression, ( ) is the psi or digamma function, which is defined in