Elastic Waves at High Frequency

This page intentionally left blank
Elastic Waves at High Frequencies
Techniques for Radiation and Diffraction

of Elastic and Surface Waves
Elastic Waves at High Frequencies
Techniques for Radiation and Diffraction
of Elastic and Surface Waves
JOHN G. HARRIS
Late, University of Illinois at Urbana–Champaign
Edited and Prepared for Publication by
GARETH I. BLOCK
Flemington, NJ
RICHARD V. CRASTER
University of Alberta
ANTHONY M. J. DAVIS
University of California, San Diego
PAUL A. MARTIN
Colorado School of Mines
ANDREW N. NORRIS
Rutgers University
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore,
São Paulo, Delhi, Dubai, Tokyo
Cambridge University Press

The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521875301
© J. G. Harris 2010
This publication is in copyright. Subject to statutory exception and to the

provision of relevant collective licensing agreements, no reproduction of any part
may take place without the written permission of Cambridge University Press.
First published in print format 2010
ISBN-13 978-0-511-78982-3 eBook (NetLibrary)

ISBN-13 978-0-521-87530-1 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy

of urls for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Do not go gentle into that good night,
Old age should burn and rave at close of day;
Rage, rage against the dying of the light.
Though wise men at their end know dark is right,
Because their words had forked no lightning they
Do not go gentle into that good night.
Do not go gentle into that good night

By Dylan Thomas, from THE POEMS OF DYLAN
THOMAS, copyright 1952
c by Dylan Thomas.
Reprinted by permission of New Directions
Publishing Corp. and J. M. Dent & Sons Ltd.
Contents
Foreword page ix
1 Linear elastic waves 1
1.1 Model equations 1
1.2 Continuity and boundary conditions 6
1.3 Flux of energy 6
1.4 The Fourier and Laplace transforms 9
1.5 A wave is not a vibration 12
1.6 Dispersive propagation 15
1.7 General references 20
2 Canonical acoustic-wave problems 23
2.1 Radiation from a piston in an infinite baffle 23
2.2 Diffraction of an acoustic plane wave by an edge 40
2.3 Summary 50
3 Canonical elastic-wave problems 52
3.1 The scattering of a spherical wave from a fluid–solid
interface 52
3.2 Rayleigh–Lamb modes and Rayleigh surface waves 62
4 Radiation and impedance 71
4.1 Reciprocity 71
4.2 Green’s tensor 72
4.3 Reciprocity revisited 78
4.4 Force on a particle from an elastic wave 79
5 Integral equations for crack scattering 81
5.1 Formulation 81
5.2 A hypersingular integral equation 84
5.3 Low-frequency scattering 86
vii
viii Contents
5.4 Some strategies 88

5.5 Flat cracks as a special case 88
5.6 Flat cracks: direct approach 89
5.7 Flat cracks: how to compute [u] 91
5.8 Curved cracks 93
6 Scanned acoustic imaging 96
6.1 Scanned, reflection acoustic microscope 97
6.2 Fresnel and F number 99
6.3 Converging spherical wave 99
6.4 Focused acoustic beam 101
6.5 Scattered focused beam 103
6.6 An electromechanical reciprocity relation 109
6.7 Measurement model 113
6.8 Acoustic material signature 116
6.9 Summary 119
7 Acoustic diffraction in viscous fluids 121
7.1 Theory 121
7.2 Diffraction by a half-plane 123
7.3 Scattering of a spherical wave at a plane interface 127
7.4 Diffraction by an elastic sphere 130
8 Near-cut-off behavior in waveguides 133
8.1 Shear horizontal and acoustic waveguides 133
8.2 Elastic waveguides 136
8.3 Long waves 138
Appendix A Asymptotic expansions 147
Appendix B Some special functions 154
References 157
Index 163
Foreword
This volume is dedicated to the memory of John G. Harris, whose

life ended prematurely on the 6th of May, 2006. John’s friendship
and research impacted many people – he was a dedicated and loving
husband, an accomplished scientist and applied mathematician, a pas-
sionate teacher, and an important mentor to many young scientists.
This book was originally intended to be John’s second book on elas-
tic wave theory and diffraction. It grew from four lectures that were
given at the Department of Mathematics and Mechanics, within IIMAS,
at the National Autonomous University of Mexico, in January 2004.
After John’s passing, several of his colleagues, inspired by his wife
Beatriz, began to convert these unfinished notes into a form suitable
for publication. We have worked to combine the existing chapters with
additional, contributed chapters from experts in the field of elastic wave
theory.
Born and raised in Toronto, John entered McGill University as a
mature student and graduated with a Bachelors in Electrical Engineer-
ing (Honours). After receiving a Masters of Science in Applied Physics
from Stanford University, John traveled to Northwestern University to
work toward a doctorate in Applied Mathematics with Jan Achenbach,
which he completed in 1979. J. D. Achenbach had a lasting impact on
John’s work in elastic wave scattering, which formed the basis of much
of John’s research as a professor at the Department of Theoretical and
Applied Mechanics at the University of Illinois at Urbana–Champaign
between 1979 and 2005.
John’s initial research focused on the scattering of ultrasonic elastic
waves from cracks, in particular the use of surface waves to interro-
gate curved shells and surface-breaking cracks in cases of interest to
acoustic microscopy. He developed a great interest in WKBJ theory and
ix
x Foreword
asymptotic approximations of wavefields, which allowed him to share

his clear and elegant understandings of diffraction in an extensive list
of publications. John also wrote a graduate-level textbook, Linear Elas-
tic Waves, and a monograph on elastic wave theory during his visit to
Mexico. He began this volume in Quaid-i-Azam University in Islamabad,
Pakistan, where he spent a semester lecturing in 1994.
Two unifying themes are used throughout this book. The first is that
wave propagation and scattering are among the most fundamental pro-
cesses that we use to comprehend the world around us. The second is
that waves are best understood in an asymptotic approximation, where
they are free of the complications of their excitation and are governed
primarily by their propagation environment. This book is not intended
as a textbook, in the sense that it is not written to accompany a specific
course. However, the chapters do follow a course of increasing com-
plexity, beginning with plane-wave propagation and spectral analyses,
which allows for the development of advanced techniques for studying
diffraction. A short synopsis of the chapters is as follows.
John’s writing forms the basis of Chapters 1, 2, and 3. Chapter 1
introduces the topics of elastic wave theory, energy flux, and Fourier and
Laplace representations of time-harmonic wavefields. Chapters 2 and 3
solve canonical scattering problems using asymptotic approximations to
Fourier integrals.
Chapter 4, written by A. Norris, explores the use of reciprocity iden-
tities and mechanical impedance to describe radiation and scattering
problems. Radiation of elastic waves is examined using Gaussian beam
solutions as a model for transducers, while reciprocity is used to derive
the force on a particle caused by an incident elastic wave.
Chapter 5, written by P. A. Martin, concerns integral formulations
of scattering from cracks. In particular, a special case of scattering
from a screen is solved in the low-frequency limit by approximating a
hypersingular integral equation, and a general strategy for solving more
complex problems is proposed to analyze scattering from curved cracks
and non-planar geometries.
Chapter 6, written by J. G. Harris, develops techniques for scanned
acoustic imaging that utilize a converging spherical wave generated by
a transducer above a fluid–solid interface. The incident and scattered
fields are written in terms of an angular spectrum of plane waves, and
a reciprocity relation is used to express unknown material variability
(in a thin film attached to the interface, for example) in terms of the
measured transducer voltage.
Foreword xi
Chapter 7, written by A. Davis, explores the effects of viscosity on

acoustic diffraction. Compressibility and viscosity are seen to be inter-
twined, as dilatational waves become coupled to vorticity disturbances
near scattering interfaces. Scattering solutions are derived for plane
waves diffracted by a half-plane, and for spherical waves scattered by
a plane interface and by an elastic sphere.
Chapter 8, written by R. Craster, elucidates the phenomenon of
channeling of wave energy along guided structures. The chapter first
summarizes guided waves in acoustics, and then proceeds to describe
elastic waves in straight waveguides (including the Rayleigh–Lamb
modes). Asymptotic expressions for wave propagation in waveguides
that are inhomogeneous, possibly bent or of varying thickness, are the
ultimate focus of this chapter.
Appendix A, written by J. G. Harris, discusses asymptotic expansions
and asymptotic approximations of integrals, methods that are used rou-
tinely throughout the book. Appendix B, also by J. G. Harris, lists, with-
out derivation, properties of the special functions that arise in the book.
These nine chapters cover both the necessary introductory material
and a broad survey of applications in diffraction and scattering theory.
John’s dedication, creativity, and clear understanding of these subjects
have inspired us to take on the task of completing his final work. We
thank him dearly for his friendship and collaboration, and hope that
future readers will find the topics as compelling and captivating as he
did.
Sincerely,
Gareth I. Block
Richard V. Craster
Anthony M. J. Davis
Paul A. Martin
Andrew N. Norris
Bibliography of J. G. Harris
Achenbach, J. D. and Harris, J. G. (1978). Ray method for elastodynamic
radiation from a slip zone of arbitrary shape. J. Geophys. Res.,
83:2283–2291.
Achenbach, J. D. and Harris, J. G. (1979). Acoustic emission from a brief
crack propagation event. J. Appl. Mech., 46:107–112.
xii Foreword
Harris, J. G. (1980). Diffraction by a crack of a cylindrical longitudi-

nal pulse. Zeit. Angew. Math. Physik, 31:367–383, 1980. Correction:
34:253, 1983.
Harris, J. G. (1980). Uniform approximations to pulses diffracted by a
crack. Zeit. Angew. Math. Physik, 31:771–775.
Harris, J. G. and Achenbach, J. D. (1981). Near-field surface motions
excited by radiation from a slip zone of arbitrary shape. J. Geophys.
Res., 86:9352–9356.
Achenbach, J. D. and Harris, J. G. (1982). Focusing of ground motion
due to curved rupture fronts. In J. Boatwright, editor, Proceedings of
Workshop XVI: The Dynamic Characteristics of Faulting Inferred
from Recordings of Strong Ground Motion, volume 1 (Report
85-591), pages 50–70. U.S. Geological Survey, Menlo Park, CA.
Harris, J. G. and Achenbach, J. D. (1983). Love waves excited by discon-
tinuous propagation of a rupture front. Geophys. J. Roy. Astron.
Soc., 72:337–351. Correction: 74:647, 1983.
Harris, J. G., Achenbach, J. D. and Norris, A. N. (1983). Rayleigh waves
excited by the discontinuous advance of a rupture front. J. Geophys.
Res., 88:2233–2239.
Shield, T. W. and Harris, J. G. (1984). An acoustic lens design using
the geometrical theory of diffraction. J. Acoust. Soc. Am., 75:1634–
1635.
Pott, J. and Harris, J. G. (1984). Scattering of an acoustic Gaussian
beam from a fluid–solid interface. J. Acoust. Soc. Am., 76:1829–
1838.
Harris, J. G. and Pott, J. (1984). Surface motion excited by acoustic
emission from a buried crack. J. Appl. Mech., 51:77–83.
Harris, J. G. (1984). Wave-front approximations in a moving coordinate
system. J. Appl. Mech., 51:934–935.
Achenbach, J. D. and Harris, J. G. (1984). Excitation of near- and
far-field ground motions by sliding events on a fault plane. In
S. K. Datta, editor, Earthquake Source Modeling, Ground Motion,
and Structural Response, volume PVP-80, pages 13–27. ASME,
New York.
Harris, J. G. and Pott, J. (1985). Further studies of the scattering of a
Gaussian beam from a fluid–solid interface. J. Acoust. Soc. Am.,
78:1072–1080.
Harris, J. G. (1987). Edge diffraction of a compressional beam. J. Acoust.
Soc. Am., 82:635–646.
Foreword xiii
Achenbach, J. D. and Harris, J. G. (1987). Asymptotic modeling of

strong ground motion excited by subsurface sliding events. In
B. A. Bolt, editor, Seismic Strong Motion Synthetics, pages 1–54.
Academic Press, New York.
Harris, J. G. (1988). The wavefield radiated into an elastic halfspace by
a transducer of large aperture. J. Appl. Mech., 55:398–404.
Choi, H. C. and Harris, J. G. (1989). Scattering of an ultrasonic beam
from a curved interface. Wave Motion, 11:383–406.
Harris, J. G. (1989). An asymptotic description of an elastodynamic
beam. In M. F. McCarthy and M. A. Hayes, editors, Elastic Wave
Propagation, pages 505–510. Elsevier, Amsterdam.
Choi, H. C. and Harris, J. G. (1990). Focusing of an ultrasonic beam by
a curved interface. Wave Motion, 12:497–511.
Choi, H. C. and Harris, J. G. (1990). Focusing of an ultrasonic beam
by a concave interface. In S. K. Datta, J. D. Achenbach and
Y. S. Rajapakse, editors, Elastic Waves and Ultrasonic Nondestruc-
tive Evaluation, pages 177–182. Elsevier, Amsterdam.
Achenbach, J. D., Ahn, V. S. and Harris, J. G. (1991). Wave anal-
ysis of the acoustic material signature for the line focus
microscope. IEEE Trans. Ultrason. Ferroelectr. Freq. Control,
38:380–387.
Ahn, V. S., Harris, J. G. and Achenbach, J. D. (1991). Acoustic
material signature for a cracked surface. In Proceedings 1990
IEEE Ultrasonics Symposium, volume 2, pages 921–924. IEEE,
New York.
Achenbach, J. D., Ahn, V. S. and Harris, J. G. (1991). BEM analysis for
the line focus acoustic microscope. In D. O. Thompson and D. E.
Chimenti, editors, Review of Progress in Quantitative Nondestruc-
tive Evaluation, volume 10A, pages 225–232. Plenum Press, New
York.
Ahn, V. S., Harris, J. G. and Achenbach, J. D. (1992). Numerical analysis
of the acoustic signature of a surface-breaking crack. IEEE Trans.
Ultrason. Ferroelectr. Freq. Control, 39:112–118.
Rebinsky, D. A. and Harris, J. G. (1992). An asymptotic calculation of
the acoustic signature of a cracked surface for the line focus scanning
acoustic microscope. Proc. R. Soc. A, 436:251–265.
Rebinsky, D. A. and Harris, J. G. (1992). The acoustic signature for a
surface-breaking crack produced by a point focus microscope. Proc.
R. Soc. A, 438:47–65.
xiv Foreword
Yogeswaren, E. and Harris, J. G. (1994). A model of a conformal

ultrasonic inspection system for interfaces. J. Acoust. Soc. Am.,
96:3581–3592.
Qi, Q., O’Brien, Jr., W. D. and Harris, J. G. (1995). The propaga-
tion of ultrasonic waves through a bubbly liquid into tissue: a
linear analysis. IEEE Trans. Ultrason. Ferroelectr. Freq. Control,
42:28–36.
Qi, Q., Johnson, R. E. and Harris, J. G. (1995). Boundary layer attenua-
tion and acoustic streaming accompanying plane-wave propagation
in a tube. J. Acoust. Soc. Am., 97:1499–1509.
Harris, J. G. and Yogeswaren, E. (1995). A model of a conformal ultra-
sonic imaging system for solid-solid interfaces. In J. L. Wegener and
F. R. Norwood, editors, Nonlinear Waves in Solids, pages 196–200.
ASME, New York.
Harris, J. G., Rebinsky, D. A. and Wickham, G. (1996). An integrated
model of scattering from an imperfect interface. J. Acoust. Soc.
Am., 99:1315–1325.
Rebinsky, D. A., Harris, J. G. and Wickham, G. (1996). Interrogating a
thin layer of heterogeneity with confocal transducers. In Review of
Progress in Quantitative Nondestructive Evaluation, volume 15A,
pages 1027–1033. Plenum Press, New York.
Ti, B. W., O’Brien, Jr., W. D. and Harris, J. G. (1997). Measurements
of coupled Rayleigh wave propagation in an elastic plate. J. Acoust.
Soc. Am., 102:1528–1531.
Harris, J. G. (1997). Modeling scanned acoustic imaging of defects at
solid interfaces. In G. Chavent, G. Papanicolaou, P. Sacks, and
W. W. Symes, editors, Inverse Problems in Wave Propagation,
volume 90 of IMA Volumes in Mathematics and its Applications,
pages 237–257. Springer-Verlag, New York.
Folguera, A. and Harris, J. G. (1998). Propagation in a slowly vary-
ing elastic waveguide. In J. A. DeSanto, editor, Mathematical and
Numerical Aspects of Wave Propagation, pages 434–436. SIAM,
Philadelphia.
Folguera, A. and Harris, J. G. (1999). Coupled Rayleigh surface waves
in a slowly varying elastic waveguide. Proc. R. Soc. A, 455:
917–931.
Goueygou, M., Harris, J. G. and O’Brien, Jr., W. D. (1999). Time-
domain solution of the temperature increase induced by diagnostic
ultrasound. In Proceedings 1999 IEEE Ultrasonics Symposium,
volume 2, pages 1385–1388. IEEE, New York.
Foreword xv
Block, G., Harris, J. G. and Hayat, T. (2000). Measurement models

for ultrasonic nondestructive evaluation. IEEE Trans. Ultrason.
Ferroelectr. Freq. Control, 47:604–611.
Asghar, S., Hayat, T. and Harris, J. G. (2001). Diffraction by a slit in
an infinite porous barrier. Wave Motion, 33:25–40.
Harris, J. G. (2001). Linear Elastic Waves. Cambridge University Press,
Cambridge, U.K.
Harris, J. G. (2002). A biographical note on Jan D. Achenbach with a
foreword to the special issue. Wave Motion, 36:307–309.
Harris, J. G. (2002). Rayleigh wave propagation in curved waveguides.
Wave Motion, 36:425–441.
Harris, J. G. and Achenbach, J. D. (2002). Comment on “On the complex
conjugate roots of the Rayleigh equation: the leaky surface wave” [J.
Acoust. Soc. Am. 110, 2867 (2001)]. J. Acoust. Soc. Am., 112:1747–
1748.
Harris, J. G. (2002). Propagation in curved waveguides. In I. D. Abra-
hams, P. A. Martin, and M. J. Simon, editors, IUTAM Symposium
on Diffraction and Scattering in Fluid Mechanics and Elasticity,
pages 321–328. Kluwer, Dordrecht.
Harris, J. G. (2004). Elastic Waves at Microwave Frequencies: Math-
ematical Models using Asymptotic Methods. Universidad Nacional
Autónoma de México.
Harris, J. G. and Block, G. (2005). The coupling of elastic, surface-wave
modes by a slow, interfacial inclusion. Proc. R. Soc. A, 461:3765–
3783.
Block, G. I. and Harris, J. G. (2006). Conductivity dependence of seismo-
electric wave phenomena in fluid-saturated sediments. J. Geophys.
Res. – Solid Earth, 111:B01304 (12 pages).
1
Linear elastic waves
Chapter 1 provides the background, both the model equations and some
of the mathematical transformations, needed to understand linear elas-
tic waves. Only the basic equations are summarized, without derivation.
Both Fourier and Laplace transforms and their inverses are introduced
and important sign conventions settled. The Poisson summation formula
is also introduced and used to distinguish between a propagating wave
and vibration of a bounded body. A general survey of books and collec-
tions of papers that bear on the contents of the book are discussed at
the end of the chapter.
A linear wave carries information at a particular velocity, the group
velocity, which is characteristic of the propagation structure or envi-
ronment. It is this transmission of information that gives linear waves
their special importance. In order to introduce this aspect of wave prop-
agation, we discuss propagation in one-dimensional periodic structures.
Such structures are dispersive and therefore transmit information at a
speed different from the wavespeed of their individual components.
1.1 Model equations

The equations of linear elasticity consist of:
(1) the conservation of linear and angular momentum; and

(2) a constitutive relation relating force and deformation.
In the linear approximation the density ρ is constant. The conservation

of mechanical energy follows from (1) and (2). The most important fea-
ture of the model is that the force exerted across a surface, oriented
by the unit normal nj , by one part of a material on the other is given
1
2 Linear elastic waves
by the traction ti = τji nj , where τji is the stress tensor. The conserva-
tion of angular momentum makes the stress tensor symmetric: that is,
τij = τji . The conservation of linear momentum, in differential form, is
expressed as
∂k τki + ρfi = ρ∂t2 ui . (1.1.1)
The term fi is a force per unit mass. In general we use Cartesian tensors
such as τij , where the indices i, j = 1, 2, 3, or a bold-face notation τ .
The symbol ∂i is used to represent the partial derivative with respect to
the ith coordinate; it is the ith component of the gradient operator ∇.
Similarly, sometimes dx f is used to represent df /dx. Repeated indices
are summed over 1, 2, 3 unless otherwise indicated. For problems
engaging only two coordinates, subscripts using Greek letters such as
α, β = 1, 2 are used so that a vector component would be written as
uβ and a partial derivative as ∂α . When these subscripts are repeated
they are summed over 1, 2. At times we use symbols such as cL or cT
when there is a need to distinguish between parameters that relate to
compressional or shear disturbances, but when that distinction is not
important we drop the subscript. Constants such as A are used over and
over again and have no special meaning.
Deformation is described using a strain tensor
ij = (∂i uj + ∂j ui )/2, (1.1.2)
where ui is ith component of particle displacement. The underlying

dependence of the deformation is upon the symmetric part of the dis-
placement gradient ∇u, which ensures that no rigid body rotations are
included. For a homogeneous, isotropic, linearly elastic solid, stress and
strain are related by
τij = λkk δij + 2μij , (1.1.3)
where λ and μ are Lamé’s elastic constants and δij is the Kronecker
delta symbol. Substituting (1.1.2) in (1.1.3), followed by substituting the
outcome into (1.1.1), gives one form of the equation of motion, namely,
(λ + μ)∂i ∂k uk + μ∂j ∂j ui + ρfi = ρ∂t2 ui . (1.1.4)
Written in vector notation the equation becomes
(λ + μ)∇(∇ · u) + μ∇2 u + ρf = ρ∂t2 u . (1.1.5)

Using the identity ∇2 u = ∇(∇ · u) − ∇ ∧ ∇ ∧ u, the equation can also

be written in the form
(λ + 2μ)∇(∇ · u) − μ∇ ∧ ∇ ∧ u + ρf = ρ∂t2 u. (1.1.6)
This last equation indicates that elastic waves have both dilatational ∇·u
and rotational ∇ ∧ u deformations. If ∂R is the boundary of a region
R occupied by a solid then commonly the tractions t or displacements
u, or combinations of either, are prescribed on ∂R. When t is given
over part of ∂R and u over another part, the boundary conditions are
said to be mixed. One very common boundary condition is to ask that
t = 0 everywhere on ∂R. This models the case where a solid surface
is adjacent to a gas of such small density and compressibility that it is
almost a vacuum. When R is infinite in one or more dimensions, special
conditions are imposed such that a disturbance decays to zero at infinity
or radiates outward toward infinity from any sources contained within R.
Another common situation is that in which ∂R12 is the boundary
between two regions, 1 and 2, occupied by solids having different prop-
erties. Contact between solid bodies is quite complicated, but in many
cases it is usual to assume that the traction and displacement, t and
u, are continuous. This models welded contact. One other simple con-
tinuity condition that commonly arises is that between a solid and an
ideal fluid. Because the viscosity is ignored, the tangential component of
t is set to zero, while the normal component of traction and the normal
component of displacement are made continuous.
These are only models and are often inadequate. To briefly indicate
some of the possible complications, consider two solid bodies pressed
together. A (linear) wave incident on such a boundary would experience
continuity of traction and displacement when the solids press together,
but would experience a traction-free boundary condition when they pull
apart (Comninou and Dundurs, 1977). This produces a complex non-
linear interaction.
The reader may consult Hudson (1980) for a succinct discussion of lin-
ear elasticity or Atkin and Fox (1980) for a somewhat more general view.
1.1.1 One-dimensional models

We assume that the various wavefield quantities depend only on the
variables x1 and t. For longitudinal strain u1 is finite, while u2 and u3
are assumed to be zero, so that (1.1.2) combined with (1.1.3) becomes
τ11 = (λ + 2μ)∂1 u1 , τ22 = τ33 = λ∂1 u1 , (1.1.7)
and (1.1.1) becomes

(λ + 2μ)∂12 u1 + ρf1 = ρ∂t2 u1 . (1.1.8)
For longitudinal stress all the stress components except τ11 are assumed
to be zero. Equation (1.1.3) becomes
3λ + 2μ
τ11 = E∂1 u1 , E=μ , (1.1.9)
λ+μ
and
λ
∂2 u2 = ∂3 u3 = −ν∂1 u1 , ν= . (1.1.10)
2(λ + μ)
Equation (1.1.1) now becomes
E∂12 u1 + ρf1 = ρ∂t2 u1 . (1.1.11)
Equations (1.1.8) and (1.1.11) are essentially the same, though they have
somewhat different physical meanings. The longitudinal stress model is
useful for rods having a small cross-section and a traction-free surface.
Stress components that vanish at the surface are assumed to be negligible
in the interior.
1.1.2 Two-dimensional models

We assume that the various wavefield quantities are independent of x3 .
As a consequence, (1.1.1) breaks into two separate equations, namely,
∂β τβ3 + ρf3 = ρ∂t2 u3 (1.1.12)
and
∂β τβα + ρfα = ρ∂t2 uα . (1.1.13)
We use Greek subscripts α, β = 1, 2 to indicate that the independent
spatial variables are x1 and x2 . The case for which the only non-zero dis-
placement component is u3 (x1 , x2 , t), namely (1.1.12), is called antiplane
shear motion, or SH motion for shear horizontal. In this case
τ3β = μ∂β u3 , (1.1.14)
giving, from (1.1.12),
μ∂β ∂β u3 + ρf3 = ρ∂t2 u3 . (1.1.15)
Note that this is a two-dimensional scalar equation, similar to (1.1.8) or
(1.1.11).
The case for which u3 = 0, while the other two displacement com-
ponents are generally non-zero, (1.1.13), is called inplane motion. The
initials P and SV are used to describe the two types of inplane motion,
namely, compressional and shear vertical, respectively. For this case
(1.1.3) becomes
ταβ = λ∂γ uγ δαβ + μ(∂α uβ + ∂β uα ), (1.1.16)
and
τ33 = λ∂γ uγ . (1.1.17)
The equation of motion remains (1.1.4): that is,
(λ + μ)∂α ∂β uβ + μ∂β ∂β uα + ρfα = ρ∂t2 uα . (1.1.18)
This last equation is a vector equation and contains two wave types,
compressional and shear, whose character we explore shortly. It leads to
problems of some complexity.
These two-dimensional equations are the principal models used. The
scalar model (1.1.14) allows us to solve complicated problems in detail
without being overwhelmed by the size and length of the calculations
needed, while the vector model (1.1.18) allows us enough structure to
indicate the complexity found in elastic wave propagation.
1.1.3 Displacement potentials

When (1.1.4)–(1.1.6) are used, a boundary condition, such as t = 0, is
relatively easy to implement. However, in problems where there are few
boundary conditions, it is often easier to cast the equations of motion
into a simpler form and allow the boundary condition to become more
complicated. One way to do this is to use Helmholtz’ theorem (Phillips,
1933, pp. 182–196) to express the particle displacement u as the sum of
a scalar ϕ and a vector potential ψ: that is,
u = ∇ϕ + ∇ ∧ ψ, ∇ · ψ = 0. (1.1.19)
The second condition is needed because u has only three components
(the particular condition selected is not the only possibility). Assume
f = 0. Substituting these expressions into (1.1.6) gives

1 1
(λ + 2μ)∇ ∇2 ϕ − 2 ∂t2 ϕ + μ∇ ∧ ∇2 ψ − 2 ∂t2 ψ = 0. (1.1.20)
cL cT
The equation can be satisfied if
c2L ∇2 ϕ = ∂t2 ϕ, c2L = (λ + 2μ)/ρ, (1.1.21)
and
c2T ∇2 ψ = ∂t2 ψ, c2T = μ/ρ. (1.1.22)
The terms cL and cT are the compressional or longitudinal wavespeed,

and shear or transverse wavespeed, respectively. The scalar potential
describes a wave of compressional motion, which in the plane-wave case
is longitudinal, while the vector potential describes a wave of shear
motion, which in the plane-wave case is transverse. Knowing ϕ and ψ,
do we know u completely? Yes we do. Proofs of completeness, along with
related references, are given in Achenbach (2003).
1.2 Continuity and boundary conditions

Consider a plane fluid–solid interface oriented by means of a unit normal
vector n̂ pointing into the fluid. The traction acting on the surface of
the solid is ts = n̂ · τ . The continuity conditions at the interface are thus
expressed as
ts · n̂ = −pf , n̂ ∧ ts = 0, us · n̂ = uf · n̂. (1.2.1)
Because the fluid is ideal, no condition is placed on n̂ ∧ uf,s .

The only other boundary condition needed in this work is one at infin-
ity. The waves must in general be outgoing, though when the focused
beam is discussed an incoming wave is considered. The principle of lim-
iting absorption (Harris, 2001, pp. 62, 63) is used, in most cases, to
determine this. Either by Fourier transforming a signal or by considering
a time-harmonic one, in the far field, it will have the form
A(φ, θ) i(kr−ωt)
ϕ= e ,
kr
where (r, φ, θ) are spherical coordinates, and k = ω/c is the wavenumber,
with c being the wavespeed. The angular frequency is defined such that
ω = ω0 + i, ω0 > 0, ≥ 0. The wavenumber then becomes k = (ω0 /c) +
i(/c). Therefore
|ϕ| ∼ e−r/c et , (1.2.2)
as r → ∞ with t fixed; that is, the wave vanishes provided the combi-
nation [i(kr − ωt)] appears in some guise. The parameter can be sent
to zero at the end of the calculations.
1.3 Flux of energy

The remaining conservation law of importance is the conservation of
mechanical energy. Again assume f = 0. This law can be derived directly
1.3 Flux of energy 7
from (1.1.1)–(1.1.3) by taking the dot product of ∂t u with (1.1.1). This

gives, initially,
(∂j τji )∂t ui − ρ(∂t2 ui )∂t ui = 0. (1.3.1)
Forming the product τkl kl , using (1.1.3) and making use of the decom-
position ∂j ui = ji + ωji , where ωji = (∂j ui − ∂i uj )/2, allows us to write
(1.3.1) as
1
∂t {ρ(∂t ui )(∂t ui ) + τki ki } + ∂k (−τki ∂t ui ) = 0. (1.3.2)
2
The first two terms on the left-hand side become the time rates of
change of
1 1
K = ρ(∂t uk )(∂t uk ), U = τij ij . (1.3.3)
2 2
These are the kinetic and internal energy density, respectively. The
remaining term is the divergence of the energy flux, F , where Fj is
given by
Fj = −τji ∂t ui . (1.3.4)
Equation (1.3.2) can then be written as
∂E
+ ∇ · F = 0, (1.3.5)
∂t
where E = K + U and is the energy density. This is the differential
statement of the conservation of mechanical energy. To better under-
stand that the energy flux or power density is given by (1.3.4), consider
an arbitrary region R, with surface ∂R. Integrating (1.3.5) over R and
using Gauss’ theorem gives

d
E(x, t) dV = − F · n̂ dS. (1.3.6)
dt R ∂R
Therefore, as the mechanical energy decreases within R, it radiates

outward across the surface ∂R at a rate F · n̂.
Because time-harmonic problems are being considered (see the next
section), the time average of the flux of energy per unit area, or the
intensity, is of interest. The time average of a quantity a(x, t) is defined as

1 t+T
a(x, t) := a(x, τ )dτ, (1.3.7)
T t
where T = 2π/ω, and the time-dependence is e−iωt . Given two terms
a(x, t) = Re[a(x)e−iωt ], b(x, t) = Re[b(x)e−iωt ],
the time average of their product is

1
a(x, t) b(x, t) = Re[a(x) b∗ (x)]. (1.3.8)
2
This is derived by a direct substitution of the product into (1.3.7). The
superscript ∗ indicates the complex conjugate.
Equation (1.3.8) is especially useful when calculating the time-average
flux of energy per unit area, which is also called the intensity. Expressing
the stress tensor and particle displacement as
τ (x, t) = Re[τ (x)e−iωt ], u(x, t) = Re[u(x)e−iωt ],
the time average of (1.3.4) is

1
F = Im[ω τ (x) · u∗ (x)]. (1.3.9)
2
Two additional time-average quantities will be of interest: the time-

average flux of energy P across a surface ∂S oriented by the unit
normal n̂, and its complex counterpart P c . These are given by

P = ReP =
c
F · n̂ dS (1.3.10)
∂S
and

iω
P c = (τ · u∗ ) · n̂ dS. (1.3.11)
2 ∂S
Lastly, in all the cases treated in this book it can be shown (Auld,
1990a, pp. 201–207; Lighthill, 1965) that
F = CE , (1.3.12)
where C is the group velocity and the energy density E is given following
(1.3.5).
Cautionary note. There are waves, such as mode L3 of an elastic

plate (see §3.2 and §8.2), whose group velocity C and wave (or phase)
velocity c are in different directions. In such cases, the principle of lim-
iting absorption, (1.4.2), must be applied with care, and it is more
direct to ask that F , or equivalently C, be directed away from the
source.
1.4 The Fourier and Laplace transforms

All waves are transient in time. One useful representation of a transient
waveform is its Fourier one. This representation imagines the transient
signal decomposed into an infinite number of time-harmonic or frequency
components. One important reason for the usefulness of this representa-
tion is that the transmitter, receiver, and propagation structure usually
respond differently to the different frequency components. The linearity
of the problem ensures that we can work out the net propagation out-
comes for each frequency component and then combine the outcomes to
recreate the received signal.
The Fourier transform is defined as
∞
ū(x, ω) = eiωt u(x, t)dt. (1.4.1)
−∞
The variable ω is complex. Its domain is such as to make the above

integral convergent. For t > 0 this domain is Im(ω) > 0. ū is an analytic
function within the domain of convergence, though it can be analytically
continued beyond it.1 The inverse transform is defined as
∞
1
u(x, t) = e−iωt ū(x, ω)dω . (1.4.2)
2π −∞
Thus we have represented u as a sum of harmonic waves e−iωt ū(x, ω).
Note that there is a specific sign convention for the exponential term that
we shall adhere to throughout the book.
A closely related transform is the Laplace one. It is usually used for
initial-value problems so that we imagine that, for t < 0, u(x, t) is
zero. This is not essential and its definition can be extended to include
functions that extend through values of negative t. This transform is
defined as ∞
ũ(x, p) = e−pt u(x, t)dt. (1.4.3)
0
As with ω, p is a complex variable and its domain is such as to make

ũ(x, p) an analytic function of p. The domain is initially defined as
Re(p) > 0, but the function can be analytically continued beyond this.
Note that p = −iω so that Im(ω) > 0. The inverse transform is given by
+i∞
1
u(x, t) = ept ũ(x, p)dp, (1.4.4)
2πi −i∞
1 Analytic functions defined by contour integrals, including the case in which the
contour extends to infinity, are discussed in Titchmarsh (1939, pp. 85–86) in a
general way and in more detail by Noble (1988).
where ≥ 0. The expressions for the inverse transforms (1.4.2) and

(1.4.4) are misleading. In practice we define the inverse transforms on
contours that are designed to capture the physical features of the solu-
tion. A large part of this book will deal with just how those contours
are selected. But, for the present, we shall work with these definitions.
Consider the case of longitudinal strain. Imagine that at t = −∞ a
disturbance started with zero amplitude. Taking the Fourier transform
of (1.1.8) gives
d2 ū1
+ k 2 ū1 = 0, (1.4.5)
dx21
where k = ω/cL and cL is the compressional wavespeed defined in
(1.1.21). The parameter k is called the wavenumber. Equation (1.4.5)
has solutions of the form
ū1 (x1 , ω) = A(ω)e±ikx1 . (1.4.6)
If we had sought a time-harmonic solution of the form
u1 (x1 , t) = ū1 (x1 , ω)e−iωt , (1.4.7)
we should have gotten the same answer except that e−iωt would be
present. In other words, taking the Fourier transform of an equation
over time or seeking solutions that are time-harmonic are two slightly
different ways of doing the same operation.
For (1.4.7), it is understood that the real part can always be taken to
obtain a real disturbance. Much the same happens in using (1.4.2). In
writing (1.4.2) we implicitly assumed that u(x, t) was real. That being
the case, ū(x, ω) = ū∗ (x, −ω), where the superscript ∗ to the right of the
symbol indicates the complex conjugate. From this it follows that
∞
1
u(x, t) = Re e−iωt ū(x, ω)dω . (1.4.8)
π 0
The advantage of this formulation of the inverse transform is that we may

proceed with all our calculations using an implied e−iωt and assuming ω
is positive. The importance of this will become apparent in subsequent
chapters. Equation (1.4.8) can be regarded as a generalization of taking
the real part of a time-harmonic wave (1.4.7).
Equation (1.4.6), when the + sign is taken, is a time-harmonic, plane
wave propagating in the positive x1 direction. We assume that the
wavenumber k is positive, unless otherwise stated. The wave propagates
in the positive x1 direction because the term (kx1 −ωt) remains constant,
and hence u1 remains constant, only if x1 increases as t increases. The
speed with which the wave propagates is cL . The term ω is the angular
frequency or 2πf , where f is the frequency. That is, at a fixed position,
1/f is the length of a temporal oscillation. Similarly k, the wavenumber,
is 2π/λ where λ, the wavelength, is the length of a spatial oscillation.
Note that if we combine two of these waves, labeled u+ −
1 and u1 , each
going in opposite directions, namely,
u+
1 = Ae
i(kx1 −ωt)
, u−
1 = Ae
−i(kx1 +ωt)
, (1.4.9)
we get
u1 = 2Ae−iωt cos(kx1 ). (1.4.10)
Taking the real part gives
u1 = 2|A| cos(ωt + α) cos(kx1 ). (1.4.11)
We have taken A as complex so that α is its argument. This disturbance

does not propagate. At a fixed x1 the disturbance simply oscillates in
time and at a fixed t it oscillates in x1 . The wave is said to stand or is
called a standing wave.
Continuing with the case of longitudinal strain, consider the follow-
ing boundary, initial-value problem. Unlike the previous discussion in
which the disturbance began, with zero amplitude, at −∞, here we shall
introduce a disturbance that starts up at t = 0+ . Consider an elastic half-
space, occupying x1 ≥ 0, subjected to a non-zero traction at its surface.
The problem is one-dimensional, and invariant in the other two, so that
(1.1.8), the equation for longitudinal strain, is the equation of motion.
At x1 = 0 we impose the boundary condition τ11 = −P0 e−ηt H(t), where
H(t) is the Heaviside step function and P0 is a constant. As x1 → ∞ we
impose the condition that any wave propagates outward in the positive
x1 direction (why?). Moreover, we ask that, for t < 0, u1 (x1 , t) = 0
and ∂t u1 (x1 , t) = 0. Note that, using integration by parts, the Fourier
transform, indicated by F, of the second time derivative is

F ∂t2 u1 = −ω 2 ū1 (x1 , ω) + iωu1 (x1 , 0− ) − ∂t u1 (x1 , 0− ). (1.4.12)
In deriving this expression we have integrated from t = 0− to ∞ so

as to include any discontinuous behavior at t = 0. Taking the Fourier
transform of (1.1.8) gives (1.4.5). Then the inverse transform of the stress
component τ11 is given by

P0 ∞ ei(kx1 −ωt)
τ11 (x1 , t) = dω. (1.4.13)
2πi −∞ ω + iη
In the course of making this step we need to choose between the solutions
to the transformed equation, (1.4.5). Why is the solution leading to
(1.4.13) selected? The integral is readily evaluated by capturing the pole.
Note that, if the disturbance is to decay with time, η must be positive.
We can also show that
τ11 (x1 , t) = −P0 H(t − x1 /cL )e−η(t−x1 /cL ) . (1.4.14)
The conditions for convergence of the integral, as its contour is closed

at infinity, give rise to the Heaviside function. Note how the sign con-
ventions for the transform pair, by affecting where the inverse transform
converges, give a solution that is causal.
The preceding example indicated the natural association between
ikx1
e and e−iωt and thereby suggests how we shall define the Fourier
transform over the spatial variable x. Suppose we have taken the tem-
poral transform obtaining ū(x, ω). Then its Fourier transform over x is
defined as
∞
∗
ū(k, ω) = e−ikx ū(x, ω)dx, (1.4.15)
−∞
and its inverse transform is

∞
1
ū(x, ω) = eikx ∗ū(k, ω)dk . (1.4.16)
2π −∞
Again note the sign conventions for the transform pair. Moreover, note
that
∞ ∞
1
u(x, t) = ei(kx−ωt)∗ū(x, ω)dω dk . (1.4.17)
4π 2 −∞ −∞
This shows that quite arbitrary disturbances can be decomposed into a

sum of time-harmonic plane waves and thereby indicates that the study
of such waves is very central to understanding linear waves.
1.5 A wave is not a vibration

A continuous body vibrates when a system of standing waves is estab-
lished within it. Vibration and wave propagation can be explicitly linked
by means of the Poisson summation formula. This formula might better
be termed a transform and is quite useful, especially for asymptotically
approximating sums.
1.5 A wave is not a vibration 13
Theorem. Consider a function f (t) and its Fourier transform f¯(ω).

Restrictions on f (t) are given below. The Poisson summation formula
states that
∞ ∞
1 ¯
f (mλ) = f (2πk/λ), (1.5.1)
|λ|
m=−∞ k =−∞
where λ is a parameter.
This formula relates a series to one comprised of its transformed terms.

If we want to approximate the left-hand side of (1.5.1) for λ small,
then knowing the Fourier transform of each term enables us to use an
approximation based on the fact that λ−1 is large. The left-hand side of
(1.5.1) may converge only slowly for a small λ.
Proof.2 This proof follows that of de Bruijn (1970). Consider the function
ϕ(x) given by
∞

ϕ(x) = f [(m + x)λ], (1.5.2)
m =−∞
where the sum converges uniformly on x ∈ [0, 1]. The function ϕ(x) has
period 1. We assume that f (t) is such that ϕ(x) has a Fourier series,
∞
ϕ = k=−∞ ck eik2πx . Its kth Fourier coefficient, ck , equals
1 1 ∞

e−ik2πx ϕ(x)dx = e−ik2πx f [(m + x)λ]dx
0 0 m =−∞
∞
(m+1)
= e−ik2πx f (xλ)dx
m =−∞ m
∞
1 −ik2π(x/λ)
= e f (x)dx. (1.5.3)
|λ| −∞
(m+t) −ikx
Note that the integral m e f (xλ)dx → 0 as m → ± ∞, uniformly
in x ∈ [0, 1], as the sum (1.5.2) converges uniformly. This completes the
proof.
These conditions are more restrictive than needed. Lighthill (1978a,

pp. 67–71), among others, indicates that the Poisson summation formula
holds for a far more general class of functions than assumed here.
2 A minimum amount of analysis is used, both here and elsewhere, and no attempt
at rigorous proofs is made. The conclusions are usually valid under more general
conditions than those cited.
Consider the impulsive excitation of a rod of finite length 1. The gov-

erning equation is (1.1.11). Assume f1 = 0, set cb and ρ = 1 (c2b = E/ρ),
and assume that for t ≤ 0, u1 (x1 , t) = ∂t u1 (x1 , t) = 0. The boundary
conditions are
τ11 (0, t) = −T δ(t), τ11 (1, t) = 0. (1.5.4)
By using a Fourier transform over t and solving the boundary-value

problem in x1 , we get3
∞
sin [nπ(1 − x1 )]
τ11 = iT H(t) e−inπt . (1.5.5)
n =−∞
cos (nπ)
Thus the rod is filled with standing waves. One usually considers a solu-
tion in this form as a vibration. This is a very useful way to express the
answer, assuming the pulse has reverberated within the rod for a time
long with respect to that needed for one echo from the far end. But the
individual interactions with the ends have been obscured. To find these
interactions we apply the Poisson summation formula to (1.5.5). Break
up the sin [nπ(1 − x1 )] term into exponential ones and apply (1.5.1) to
each term. The crucial intermediate step is the following, where we have
taken one of these terms.
∞
1 −1
e−imπ[t−(1−x1 )] = (π|t + x1 − 2|)
m =−∞
cos mπ
∞
∞

2n
× exp − iω 1 − dω
n =−∞ −∞
|t + x1 − 2|
∞

=2 δ (|t + x1 − 2| − 2n) . (1.5.6)
n =−∞
The outcome to our calculation is

∞
∞

τ11 = T δ(t + x1 − 2k) − T δ(t − x1 − 2k). (1.5.7)
k=1 k=0
This is a wave representation. It is very useful for times on the order of

the echo time. For example, if t ∈ (1, 2), then τ11 = T δ[(t − 1) + (x1 − 1)].
We have thus isolated the pulse returning from its first reflection at
the end x1 = 1. However, the representation is not very useful for t
large because it becomes tedious to work out exactly at what reflection
3 The reader should check that this is the solution.
you are tracking the pulse. Moreover, the representation (1.5.7) would
have been awkward to work with if, instead of delta-function pulses, we
had had pulses of sufficient length that they overlapped one another.
Nevertheless, the representation captures quite accurately the physi-
cal phenomenon of a pulse bouncing back and forth in a rod struck
impulsively at one end.
A vibration therefore is defined and confined by its environment. It
is the outcome of waves reverberating in a bounded body. A period
of time, sometimes a long one, is needed for the environment to settle
into a steady oscillatory motion. In contrast, a wave is a disturbance
that propagates freely outward, returning to its source perhaps only
once, experiencing only a finite number of interactions. Understanding
how an individual wave interacts with its environment and tracking it
through each of its interactions constitute the principal problems of wave
propagation. Moreover, while one works frequently with time-harmonic
propagating waves, one usually assumes that at some stage a Fourier
synthesis will be carried out, mapping the unending oscillatory motion
into a disturbance confined both temporally and spatially.
1.6 Dispersive propagation

1.6.1 An isolated interaction
A basic interaction of a wave with its environment is scattering from a
discontinuity. Continue to consider waves in a rod using the longitudinal
stress approximation. Consider time-harmonic disturbances of the form
u1 = ū1 (x1 )e−iωt , τ11 = ρc2b ∂1 u1 . (1.6.1)
We shall not indicate the possible dependence on ω of ū1 unless this is

needed. The equation of motion (1.1.11) becomes
d2 ū1
+ k 2 ū1 = 0, k = ω/cb . (1.6.2)
dx21
Assume there is a region of inhomogeneity within x1 ∈ (−L, L).
Incident on this inhomogeneity is the wavefield

i A1 eikx1 , x1 < −L,
ū1 (x1 ) = (1.6.3)
A2 e−ikx1 , x1 > L,
where we have allowed waves to be incident from both directions. Upon

striking the inhomogeneity the scattered wavefield

s B1 e−ikx1 , x1 < −L,
ū1 (x1 ) = (1.6.4)
B2 eikx1 , x1 > L,
is excited. Note that the scattered waves have been constructed so that
they are outgoing from the scatterer. The linearity of the problem sug-
gests that we can write the scattered amplitudes in terms of the incident
ones as

B1 S11 S12 A1
= , (1.6.5)
B2 S21 S22 A2
or, more compactly, as

B = SA. (1.6.6)
The matrix S is called a scattering matrix and characterizes the

inhomogeneity.
We next consider a specific example. Imagine that the rod has a cross-
sectional area 1 and that the inhomogeneity is a point mass M , at x1 = 0.
The left-hand figure in Fig. 1.1 indicates the geometry. The conditions
across the discontinuity are
u1 (0− , t) = u1 (0+ , t), M ∂t2 u1 = −τ11 (0− , t) + τ11 (0+ , t); (1.6.7)
that is, the rod does not break, but the acceleration of the mass causes
the traction acting on the cross-section to be discontinuous. Setting
tan θ = kM/2ρ, with θ ∈ (0, π/2), the matrix S is calculated to be

sin θei(θ+π/2) cos θeiθ
S= . (1.6.8)
cos θeiθ sin θei(θ+π/2)
It is also of interest to relate the wave amplitudes on the right to

those on the left. This matrix T, called the transmission matrix, gives
0 1 2
x1 x1
0 L 2L
Fig. 1.1. One or more point masses M are embedded in a rod of cross-sectional
area 1. The left-hand figure shows a single, embedded point mass. The right-
hand figure shows an endless periodic arrangement of embedded point masses,
each separated by a distance L. The masses are labeled 0, 1, 2, . . . with the 0th
mass at x1 = 0.
T T
R = TL, where L = [A1 , B1 ] and R = [B2 , A2 ] . The matrix T is
readily calculated from S and is given by

1 + i tan θ i tan θ
T= . (1.6.9)
−i tan θ 1 − i tan θ
Note that the amplitudes A1 and B2 are those of right-propagating

waves, while B1 and A2 are those of left-propagating ones.
1.6.2 Periodic structures

One of the more interesting aspects of wave-bearing structures is that
they often contain several length scales. Propagation in such a structure
often can only take place if the angular frequency ω is linked to the
wavenumber – a term we must define a bit more carefully here – in a non-
linear way. To consider this possibility we use the matrix T, (1.6.9), to
analyze propagation in a periodic structure. We imagine an infinite rod,
of cross-sectional area 1, in which equal point masses, M , are periodically
embedded. The right-hand figure of Fig. 1.1 indicates the geometry and
how the masses are labeled. One such mass has the nominal position
x1 = 0 and is labeled n = 0. Each mass is separated from its neighbors
by a distance L. A cell of length L is thereby formed and is labeled n
if the nth mass occupies its left end. There are thus two length scales,
the wavelength λ = 2π/k and the cell length L. In this problem we do
not concern ourselves with how the waves are excited, but only with the
simpler question, what waves does this structure support?
Consider the 0th cell, where x1 ∈ (0, L). Within that cell the solution
to (1.6.2) is
ū1 (x1 ) = R0 eikx1 + L0 e−ikx1 . (1.6.10)

−ikL T
At x1 = L− the wavefield is R0 eikL , L0 e . This can be written as
T
L1 = PR0 , with R0 = [R0 , L0 ] . The matrix P is called the propagator
or the propagation matrix and is given by
ikL
e 0
P= . (1.6.11)
0 e−ikL
At x1 = L+ , within the first cell, the wavefield amplitudes R1 =

T
[R1 , L1 ] are
R1 = TPR0 . (1.6.12)
T
This relation is readily generalized. If Rn = [Rn , Ln ] then
Rn+1 = TPRn . (1.6.13)
The central feature of the propagation structure is that it has transla-
tional symmetry. The central feature of the disturbance we seek is that
its phase changes from cell to cell in a way that represents propagation.
Specifically consider propagation to the right. To capture these two fea-
tures, the wavefield at a point within the (n + 1)th cell can differ from
that at a point within the nth cell, where the two points are separated
by a distance L, by at most a multiplicative phase factor. This kinematic
constraint is expressed by the relation
Rn+1 = eiκL Rn , (1.6.14)
where κ is unknown: κ is positive, if real, and such as to cause decay,
if complex. Combining (1.6.13) and (1.6.14) gives a 2 × 2 system of
algebraic equations that has a non-trivial solution if and only if

det TP − eiκL I = 0. (1.6.15)
Recalling our previous definition of tan θ = kM/2ρ, this equation can
be written compactly as
cos κL = cos(kL + θ)/ cos θ. (1.6.16)
This is a non-linear relationship between the angular frequency ω = cb k
and the effective wavenumber κ, though it may not be apparent, as yet,
that κ (and not k) is the wavenumber of interest.
Note that if κL ∈ [−π, π] is a solution, then κL±2nπ, for n = 1, 2, . . . ,
is also a solution. Accordingly we need only consider κL ∈ [−π, π]. The
term κL is real provided | cos κL| ≤ 1. Therefore the boundaries between
real and complex κL are given by
cos(kL + θ)/ cos θ = ±1. (1.6.17)
Taking +1, the solutions are sin(kL/2) = 0 or tan(kL/2) = − tan(θ).
That is, kL = 2nπ or kL + 2θ = 2mπ, where n and m are integers. Tak-
ing −1, the solutions are cos(kL/2) = 0 or cot(kL/2) = tan(θ). That
is, kL = (2n − 1)π or kL + 2θ = (2m − 1)π, where, again, n and m
are integers. All these cases are covered by kL = nπ or kL + 2θ = mπ.
For kL ∈ [(n − 1)π, (nπ − 2θ)], κL is real. These intervals are called
passbands. Elsewhere κL is complex, causing the disturbance to decay
as it propagates, and the intervals are called stopbands. At the lower
boundary of a passband, L is an integer number of half wavelengths.
If T is real and such as to allow only weak transmission, then all the
reflected waves add constructively and little or nothing is transmitted.
The actual situation is complicated by the complex T, but the construc-
tive interference of the reflected waves is the basic physical mechanism
giving rise to the stopbands. This phenomenon is referred to as Bragg
scattering.
Consider the interval x1 ∈ (nL, (n + 1)L). Then
ū1 (x1 ) = Rn eik(x1 −nL) + Ln e−ik(x1 −nL)

= eiκL Rn−1 eik(x1 −nL) + Ln−1 e−ik(x1 −nL)
= eiκL ū1 (x1 − L). (1.6.18)
This equation is a restatement of (1.6.14). Further, it indicates that
ū1 (x1 ) must satisfy the functional equation ū1 (x1 + L) = eiκL ū1 (x1 )
if the kinematic constraint is to be enforced. Within each cell there
are nominally two waves, as indicated in (1.6.18), which we call partial
waves. However, we seek a solution for the wave globally propagating to
the right along the structure, as distinguished from the right and left
propagating partial waves in each cell. With this in mind, the solution
to the functional equation is
ū1 (x1 ) = eiκx1 ϕ(x1 ), (1.6.19)
where ϕ(x1 + L) = ϕ(x1 ).4 That is, ϕ(x1 ) is a periodic function and can
be represented by a Fourier series whose coefficients are cn . Therefore,
ū1 (x1 ) becomes
∞

ū1 (x1 ) = cn eix1 (κ−2πn/L) . (1.6.20)
−∞
The time-harmonic wavefield ū1 (x1 )e−iωt is thus a consequence of an

infinite number of space harmonics. Note that shifting κL by ±2mπ
would not change this expression.
More importantly, it is clear that it is κ, through the term ei(κx1 −ωt) ,
that is the wavenumber. Equation (1.6.16) indicates that ω is a function
of κ, or κ a function of ω. A relation such as this is called a dispersion
relation. Writing κ as ω/c(ω), we see that ū1 (x1 , ω) propagates at a
different speed for each ω. If we excited the structure with a pulse, then
the pulse would be comprised of an infinite number of such components,
as indicated by (1.4.8). Each component would then propagate at its
own speed and the pulse would become dispersed. A pulse is information,
4 This is a partial statement of Floquet’s theorem (Friedman, 1956, pp. 65–67).
whereas a sinusoid is not. Hence what we have inferred is that dispersion

can cause the distortion of or loss of information from a signal. We shall
explore this topic further in Chapter 6.
There are many fascinating aspects to propagation in periodic struc-
tures. The discussion here has followed parts of Levine (1978, pp.
273–308, 339–345 and elsewhere) and a reader seeking to learn more
may wish to read this work further.
1.7 General references

Specific references will be given as the book progresses; there are, how-
ever, several general books and reviews that are useful, though not
always referred to elsewhere. These are listed here.
Propagation of elastic and acoustic waves. There are now several books
on elastic waves, including that by Harris (2001). Among those con-
cerned with elastic waves are Achenbach (2003), Auld (1990a; 1990b),
Eringen and Şuhubi (1975), Graff (1991), Hudson (1980), Miklowitz
(1978), Poruchikov (1993), and Rose (1999). In particular, Achenbach
(2003), Miklowitz (1978), Poruchikov (1993) and Rose (1999) discuss
many of the advanced techniques used to describe elastic waves, includ-
ing the scalar Wiener–Hopf method. The two volumes by Auld (1990a;
1990b) are quite original in presenting the subject: elastic waves are
approached as an extension of the study of electromagnetic waves. Erin-
gen and Şuhubi (1975), Graff (1991), and Rose (1999) are workmanlike,
but do not give imaginative presentations of the mathematics of wave
propagation, and are more suited to an applied readership. Lastly, Aki
and Richards (2002) is a wide-ranging and very useful textbook, at an
advanced level, that describes many aspects of elastic-wave propagation
as it applies to seismology. Both Morse and Ingard (1968) and Pierce
(1981) describe the equations of linear acoustics and solve many of the
standard problems. Pierce (1981) is the more modern in its outlook.
Radiation and edge diffraction. Bouwkamp (1954) gives a comprehensive

overview of formulating and solving diffraction problems, with edges, to
about 1950, and forms the approximate point in time from which the
writer gives references to edge diffraction. Geometrical diffraction theory
(Keller, 1958) has had a defining impact on the study of diffraction
problems, as is evidenced by the collection of papers in Hansen (1981).
Apertures produce well-collimated beams and these beams, which form
at a Fresnel length from the aperture, have their own propagation
1.7 General references 21
properties; Felsen and Deschamps (1974) is a collection of papers that

describe work on the relation between ray and beam descriptions, up
to about 1970. Diffraction through an aperture can be described by
directing attention to the wavefield filling the aperture, or by directing
attention at scattering from the edges. This latter viewpoint is the one
taken in this book; one review of edge diffraction, based on physical
optics, is that by Ufimtsev (1989, pp. 463–474). The study of diffrac-
tion has a long history, primarily in optics; one helpful place to begin to
study this work is the collection of papers in Oughstun (1991). Diffrac-
tion from edges, and by extension wedges, still remains an important
mathematical challenge; Osipov and Norris (1999) review the important
work of Malyuzhinets, which echoes Sommerfeld’s original work on edge
diffraction. Sommerfeld’s original work has recently been translated and
combined with various historical and mathematical notes in Sommerfeld
(2004); briefer descriptions of his work are given in Baker and Copson
(1987, pp. 124–152) and Sommerfeld (1967, pp. 247–272).
There are now several books that deal with various aspects of radia-
tion and edge diffraction: Achenbach et al. (1982), Babic̆ and Buldyrev
(1991), Borovikov and Kinber (1994), James (1980), and Jull (1981).
Achenbach et al. (1982) and this book both discuss the edge diffrac-
tion of an incident compressional wave in three dimensions. Babic̆ and
Buldyrev (1991) and Borovikov and Kinber (1994) are comprehensive
monographs describing many aspects of the asymptotic solution to lin-
ear wave problems: these are not books for the faint-hearted. James
(1980) is directed at electromagnetic propagation, as is Jull (1981).
Elastic waveguides and surface waves. All the books on elastic-wave

propagation discuss guided waves and surface waves to some extent.
Miklowitz (1978) in particular discusses elastic waveguides with an exci-
tation at the end of a plate at length; however, his approach and the one
taken here are quite different. Biryukov et al. (1995) and this book both
discuss elastic surface waves on curved surfaces. Malischewsky (1987)
deals with guided waves, in somewhat the same spirit as does this book,
though he is concerned primarily with anti-plane shear problems.
Mathematical methods. Many of the standard techniques of applied

mathematics are used in this book; however, some emphasis is given
to asymptotic approximations to the Fourier-like integrals that arise in
radiation and diffraction problems. Borovikov (1994) has as its sole goal
to describe such asymptotic approximations. He discusses all the cases of
integral approximations that are described in this book. Also of use are
chapters 8 and 9 of Bleistein and Handelsman (1975) and most of Copson
(1971). The angular spectrum representation of a wavefield plays a signif-
icant role in the description of scanning acoustic microscopy. Clemmow
(1966) introduces this way of representing a wavefield using examples
from electromagnetic propagation, and this technique is given many
applications to physical optics by Nieto-Vesperinas (1991). A working
knowledge of the Wiener–Hopf technique can be learned from Noble
(1988, pp. 11–27) or Weinstein (1969), though the former may be easier
to read.
Acoustic microscopy. Briggs (1992) is a comprehensive study of many

aspects of acoustic microscopy, and its list of references is very com-
plete (up to its publication date). However, the model of the scanning
acoustic microscope described in this book differs substantially from that
described in Chapter 6, though the imaging equation that is derived, the
one describing how imaging works, is identical.
2
Canonical acoustic-wave problems
Four problems form a framework for understanding the excitation and

propagation of elastic waves at high frequencies. The purpose of this
chapter is to describe two acoustic problems; the next chapter will
describe two elastic-wave problems.1 The two acoustic problems are
described as follows:
(1) The radiation from an oscillating piston, in contact with an ideal

fluid, is calculated in three distinct ways, each indicating a different
interpretation of the source wavefield. The radiation impedance is
also calculated to indicate how the source and the radiated wavefield
interact.
(2) The cylindrical wave diffracted from the edge of a rigid barrier when
it is struck by an acoustic plane wave is calculated using the Wiener–
Hopf method. A uniform asymptotic description of the diffracted
wavefield is calculated.
In each case applied mathematical techniques are introduced, and

expanded upon in subsequent chapters.
2.1 Radiation from a piston in an infinite baffle

The centerpiece of an introductory course on physical acoustics is often
a discussion of the piston source. While it is not a very realistic model,
its solution informs how one subsequently treats radiation from loud-
speakers and ultrasonic transducers. Moreover, the study of this problem
allows one to make a number of far-reaching statements about modeling
1 The adjective ‘acoustic’ is used when propagation of a compressional wave in
an ideal fluid is of interest; the adjective ‘elastic-wave’ is used when shear-wave
propagation must be taken into account.
23
24 Canonical acoustic-wave problems
x1
φ
ê
ρ
ρ
r′
r
x = (x1, x3, x3)
x2
x3
ê3
Fig. 2.1. A piston of radius a undergoes small-amplitude, time-harmonic

vibrations in a rigid baffle. The region R, defined by x3 ≥ 0, is occupied
by an ideal fluid. The baffle and piston together occupy the boundary region
∂R, defined by x3 = 0, and the piston face occupies the region ∂Sa , defined
by ρ ∈ [0, a] and φ ∈ [0, 2π), so that ∂Sa ⊂ ∂R.
sources that extend over several wavelengths. For these reasons radiation
from an oscillating piston is taken as a canonical problem.
Figure 2.1 shows the geometric arrangement: a piston embedded in a
rigid baffle radiates sound into an ideal fluid, which occupies the region
R. The piston itself need not be rigid, though for some calculations this
assumption is made.
The problem of radiation from a piston is essentially equivalent to
diffraction through a circular aperture when the wavefield in the aperture
is approximated as being composed of only the incident wavefield. In an
exact diffraction problem the aperture also contains the edge-diffracted
wavefield, in addition to the incident one, so that the wavefield in the
aperture is not known. However, in the case of radiation by a piston,
the motion of the piston is known; it can be considered as an incident
wavefield filling an aperture. Accordingly, much of the very extensive
literature on approximate diffraction through an aperture is pertinent
to the description of radiation from the piston, and names the reader
may associate with optics are used here as well.2
2 For example, the approximation to aperture diffraction just referred to is a vari-

ation of the Kirchhoff boundary conditions. In associating names with equations,
approximations, and so on, the writer has tried to be careful, but not necessarily
historically accurate. The introduction and notes to Sommerfeld (2004) give some
of the history of diffraction.
There are at least three distinct ways to represent mathematically

the radiation from a piston. The most common way is to use Huygens’
principle: referring to Fig. 2.1, ∂Sa , the piston face, is considered an
initial wavefront, and each point in ∂Sa acts as a source of secondary
spherical waves, each of radius r = cΔt. The envelope of these sec-
ondary spherical surfaces gives the wavefront of the radiated wavefield
after it has propagated at wavespeed c through time interval Δt. This
idea was used by Fresnel to explain diffraction from an aperture, and
by Rayleigh and Helmholtz to describe radiation from a piston. The
approach is described in most books on acoustics, among them Morse
and Ingard (1968), Pierce (1981), and Skudrzyk (1971, pp. 489–499,
532–537, 603–605). This last book has a useful though discursive dis-
cussion of many aspects of the piston radiator, as well as of diffraction
through an aperture in a screen. Greenspan (1979) is also a useful review
article on aspects of radiation from a piston source.
A second way to describe the radiated wavefield is to imagine a cylin-
drical region with base equal to ∂Sa and extending normally into R.
Within the cylinder a plane wave radiated by the face of the piston
propagates, while outside the cylinder it is zero. Radiating from the
edge ρ = a, and compensating for this discontinuity, is a wave with
toroidal wavefront. In optical diffraction this latter wave is referred to
as the boundary diffraction wave (Born and Wolf, 1999, pp. 499–503;
Skudrzyk, 1971, pp. 519–531; Sommerfeld, 1967, pp. 311–318), though
here it is named an edge-diffracted wave. Its existence was first sug-
gested by Young to explain diffraction through an aperture, somewhat
before Fresnel proposed using Huygens’ principle, but a mathemati-
cal representation was achieved only much later by Rubinowicz (see
the citations in Gniadek and Petykiewicz, 1971). This picture could
be considered as a beginning for the geometrical theory of diffraction
(Keller, 1958).
A third way is to represent the radiated wavefield as a Hankel trans-
form in planes parallel to ∂R, the plane containing the piston and baffle,
a calculation first undertaken by King (1934) and subsequently extended
by Bouwkamp (1946) (see also Oughstun, 1991; pp. 41–67). The Hankel
transform can then be re-expressed as an integration over plane waves;
this gives a plane-wave or angular spectrum representation of the radi-
ated wavefield (Clemmow, 1966). This particular representation is very
useful because knowing how a single plane wave interacts with a bound-
ary or edge allows one to calculate how the beam radiated by a piston
interacts with them.
The problems are formulated throughout this section using ϕ, the

particle displacement potential. The end results can readily be expressed
in terms of the acoustic pressure p: using p = −ρf ∂t2 ϕ implies the relation
p = (ρf c)(ω k)ϕ for time-harmonic waves. Moreover, the amplitude of ϕ
is normalized by the constant A. If one imagines that V0 is the magnitude
of the particle velocity, then A = V0 /(ω k). ρf is the density of the fluid, c
the wavespeed in the fluid, ω the angular frequency, and the wavenumber
k = ω/c.
2.1.1 Huygens’ principle and the Fresnel length

Figure 2.1 indicates the geometry of the problem being studied. The
region R = {x | (x1 , x2 ) ∈ (−∞, ∞), 0 ≤ x3 < ∞}, where x =
(x1 , x2 , x3 ) locates the observation point, and is occupied by an ideal
fluid. The baffle and piston together occupy the region ∂R = {ρ, φ, x3 |
ρ ∈ [0, ∞), φ ∈ [0, 2π), x3 = 0}, while the piston face occupies the region
∂Sa = {ρ, φ, x3 | ρ ∈ [0, a), φ ∈ [0, 2π), x3 = 0}, so that ∂Sa ⊂ ∂R. In
the time-harmonic approximation, ϕ satisfies the reduced wave equation
in R, namely,
∇2 ϕ + k 2 ϕ = 0, x ∈ R. (2.1.1)
This follows from the wave equation ∇2 ϕ = c−2 ∂t2 ϕ. On the boundary
∂R, ϕ satisfies
∂3 ϕ = ikA v(ρ/a) H(1 − ρ/a), x ∈ ∂R (2.1.2)
(here H(ξ) is the Heaviside step function); that is, the function v(ρ/a)
is defined in ∂Sa . Equation (2.1.2) is equivalent to the third continuity
condition, (1.2.1). The function v(ξ) is normalized so that
1
v(ξ)v ∗ (ξ)ξdξ = 1/2.
0
Lastly, as r = |x| → ∞, then ϕ must satisfy a radiation condition

such as that imposed by the principle of limiting absorption, (1.2.2).
The surface ∂R also includes a hemisphere of radius r, centered at the
origin; however, sending r to infinity removes any contribution from this
surface so that it is not discussed further.
It is useful to consider an auxiliary problem before solving (2.1.1) and
(2.1.2), namely that of the Green’s function g(x, x ) for this problem
(Harris, 2001, pp. 56–76). It satisfies
∇2 g + k 2 g = −δ(x − x ), x, x ∈ R, (2.1.3)
subject to the boundary conditions that
∂3 g = 0, x ∈ ∂R (2.1.4)
and that a radiation condition as |x| → ∞ be satisfied. The solution to

this problem is

1 eik|x−x | eik|x−x |
g(x, x ) = + , (2.1.5)
4π |x − x | |x − x |
where x = (x1 , x2 , x3 ) and x = (x1 , x2 , −x3 ), and x3 > 0.
The bilinear form
∇ · (g∇ϕ − ϕ∇g) = ϕ(x)δ(x − x )
is constructed using (2.1.1) and (2.1.3); it is then integrated throughout

the half-space R and transformed to a surface integral. Then, using
(2.1.2) and (2.1.4), and noting that the Green’s function is symmetric
in its arguments, ϕ is represented as

ϕ(x) = − g(x, x ) ∂3 ϕ(x ) dS(x ), (2.1.6)
∂Sa
where x ∈ ∂Sa . Substituting (2.1.5) in (2.1.6), with x3 = 0, gives

k2 A eikr
ϕ(x) = −i v(ρ/a) dS(x ). (2.1.7)
2π ∂Sa kr
Here r = |x − ρ êρ |, and êρ is a unit vector in the ρ direction (Fig. 2.1).
The element of surface is dS(x ) = ρ dρ dφ. Note that the particular
cases solved here are axisymmetric so that the dφ integration can be
done immediately.
Equation (2.1.7) is a mathematical expression of Huygens’ principle:
each point on the initial wavefront ∂Sa is a source of spherical waves,
of radius r , weighted at each point by A v(ρ/a). The integral governs
the subsequent evolution of the radiated wavefield; in particular, the
envelope of the various secondary spherical wavefronts defines the overall
wavefront.
As has been indicated by the previous citations, this integral has
been extensively studied in numerous ways. There is little to add to
the techniques used to evaluate it. Nevertheless, there is one important
observation to be made: because the problem is axisymmetric, the obser-
vation point can be placed in the x2 = 0 plane, without loss of generality.
The distance r can then be approximated as

ρ ρ2
r ≈ r 1 − sin θ cos φ + 2 (1 − sin2 θ cos2 φ) ,
r 2r
where r = |x|. The quadratic term may be dropped, provided ka2 /2r π
(that is, less than half a cycle), where max(ρ) = a. This leads to the
definition of the Fresnel length, Fl , as
Fl := ka2 /(2π). (2.1.8)
When r > Fl the wavefield is in the far field. In this region,

eikr 1
ϕ(r, θ) = −i[(ka)2 A] J0 (ka ξ sin θ)v(ξ)ξdξ + O(Fl /r), (2.1.9)
kr 0
where J0 is the Bessel function of order zero.
The Hankel transform is defined by
∞
∗
ϕ(kρ , x3 ) = ρϕ(ρ, x3 )J0 (kρ ρ)dρ, (2.1.10)
0
∞
ϕ(ρ, x3 ) = kρ ∗ϕ(kρ , x3 ))J0 (kρ ρ)dkρ . (2.1.11)
0
The radiated wavefield of (2.1.9) appears as a spherical wave radiated by

an anisotropic point source; the anisotropy is described by the Hankel
transform of the distribution of particle velocity on the piston face. The
Hankel transform is useful in deriving solutions to the wave equation, as
we will see in §2.1.3.
In §2.1.2 the wavefield in the neighborhood of the Fresnel length will
be investigated, and there it will be found that it is well collimated
and can be described as a beam. When microwave frequency sound is
used to probe an environment, it is usual that one wants a beam rather
than a spherical wave because when a beam strikes something one can
more readily sort out where the echo is coming from. Thus at microwave
frequencies the wavefield at the Fresnel length is more important than
that in the far field, (2.1.9).
2.1.2 The edge-diffraction integral

To explore the wavefield at the Fresnel length, the edge-diffraction
interpretation of the radiation from the piston is a more useful start-
ing point. Consider the cylindrical region S = {ρ, φ, x3 | ρ ∈ [0, a], φ ∈
[0, 2π), x3 ∈ [0, ∞)} shown in Fig. 2.2. Its surface ∂S = ∂Sa ∪ ∂Ss ;
the surface ∂Ss = {ρ, φ, x3 | ρ = a, φ ∈ [0, 2π), x3 ∈ [0, ∞)}. Within the
cylinder a plane wave, the geometrical wavefield ϕg propagates outward,
x1
êρ φ
α
a
r′
r e^3
x2 ϑ β x
x3
Fig. 2.2. A cylindrical region S projects normally from the piston face ∂Sa
into R. The surface ∂S = ∂Sa ∪ ∂Ss . Also shown are the angles α, β, and
θ, and the length r = |x − a êρ |. Note that r does not equal that used in
Fig. 2.1, unless ρ = a.
while outside the cylinder it vanishes. Again there is a surface at the far
right in Fig. 2.2 that caps ∂S; however, sending x3 to infinity removes
any contribution from this surface, so that it is not discussed further.
The geometrical wavefield satisfies
d2 ϕ g
+ k 2 ϕg = 0, x ∈ S, (2.1.12)
dx23
subject to
dϕg
= ikA, x ∈ ∂Sa . (2.1.13)
dx3
Note that the boundary condition on ∂Sa is equivalent to setting
v(ρ/a) = H(1 − ρ/a) in (2.1.2); however, ϕg ≡ 0 outside of S. As
x3 → ∞, ϕg represents an outgoing wave, satisfying the principle of
limiting absorption, (1.2.2).
The solution to this problem is
ϕg (x) = A H(1 − ρ/a)eikx3 . (2.1.14)
Note that ϕg is allowed to jump to zero at ρ = a.

Letting ê3 be a unit vector in the x3 direction, represent the gradient
operator as
∇ = ∇t + ê3 ∂3 ,
where ∇t indicates the gradient in a plane parallel to ∂R. Using (2.1.3)

and (2.1.4) in combination with (2.1.12) and (2.1.13), the bilinear form
∇ · [ g (ê3 ∂3 ϕg ) − ϕg ∇g ] + ∇t g · ∇t ϕg = ϕg (x)δ(x − x )
is constructed, where x, x ∈ R. The term ∇t ϕg contains a delta function

because of the discontinuity at ρ = a. Accordingly the second term in
this expression can be written as
A
∇t g · ∇t ϕg = − ∇t g · êρ δ(1 − ρ/a) eikx3 .
a
Integrating the bilinear form throughout R, taking account of the
delta function, and noting that ϕg and g, and their derivatives, go to
zero as x3 → ∞ gives

ϕg (x ) = − g(x, x ) ∂3 ϕg (x) dS(x)
∂Sa
(2.1.15)
− [∇1 g(x, x ) · êρ ] ϕg (x) dS(x).
∂Ss
The subscript 1 indicates that the gradient is taken with respect to

the first argument in g(x, x ). The unit normal to ∂Ss is êρ . Because
ϕg (x ) ≡ 0 for x ∈ S c , the complement of S, there is a correspond-
ing discontinuity located in the surface integral when x lies in ∂Ss .
Interchanging the arguments x, x , and noting the symmetry of g(x, x ),
allows the previous expression to be written as

ϕg (x) = − g(x, x ) ∂3 ϕg (x ) dS(x )
∂S
a (2.1.16)
− [∇2 g(x, x ) · êρ ] ϕg (x ) dS(x ).
∂Ss
The subscript 2 indicates that the gradient is taken with respect to the
second argument in g(x, x ). From (2.1.6) one notes that the integral over
∂Sa equals ϕ, because ∂3 ϕg = ∂3 ϕ there: compare (2.1.2) with (2.1.13)
and note that v(ρ/a) = 1 in ∂Sa (for the problem of this section).
Therefore (2.1.16) can be written as

ϕ(x) = ϕg (x) + [∇2 g(x, x ) · êρ ] ϕg (x ) dS(x ), (2.1.17)
∂Ss
where x ∈ R.
2.1.3 The Maggi transformation

The next step is to reduce the integral over the surface ∂Ss to a line
integral over the boundary of ∂Sa . This transformation is done with
varying degrees of sophistication in the previously cited references, and
in Baker and Copson (1987, pp. 74–84) and Gniadek and Petykiewicz
(1971), though none of these references uses the particular representation
given by (2.1.17). The name ‘Maggi transformation’ is used here by
analogy with its use in Baker and Copson (1987, pp. 74–84).
The points x = (x1 , x2 , x3 ) and x = (x1 , x2 , −x3 ), where x3 > 0, in
(2.1.18) now lie in ∂Ss and in its reflection in the plane ∂R, respectively.
The distances from these points to the observation point x are given by
1/2
s∓ = (x1 − x1 )2 + (x2 − x2 )2 + (x3 ∓ x3 )2 .
Referring to (2.1.5), the gradient in (2.1.17) can be expressed as

eiks∓ d eiks∓
∇2 · êρ = cos χ∓ , (2.1.18)
s∓ ds∓ s∓
where
cos χ∓ = (∇2 s∓ ) · êρ .
The subscript 2 indicates that the derivative is taken with respect to x

(the −) or x (the +). Note that ∇2 s∓ = −∇1 s∓ , where the subscript
1 indicates that the derivative is taken with respect to x.
The distance r = |x − a êρ |, from a point on the boundary of ∂Sa to
the observation point x (see Fig. 2.2), is
1/2
r (x, φ) = (x1 − a cos φ)2 + (x2 − a sin φ)2 + (x3 )2 . (2.1.19)
Next, note that the length of a perpendicular dropped from x to a point

on ∂Ss is such that
s∓ cos χ∓ = r cos α, cos α = (−∇1 r ) · êρ .
The angle α is shown in Fig. 2.2. We identify a second angle β, also

shown in Fig. 2.2, using
cos β = ∇1 r · ê3 .
By direct computation one finds that

eik(s∓ +x3 ) 1 d eik(s∓ +x3 )
ik − = . (2.1.20)
(s∓ )2 s∓ dx3 s∓ (s∓ + x3 ∓ r cos β)
Equations (2.1.18) and (2.1.20) constitute the Maggi transformation as

used here.
Using (2.1.18) and (2.1.20) in (2.1.17), it follows that

A 2π a ∇1 r · êρ
ϕ(x) = ϕg (x) + 2
eikr dφ, (2.1.21)
2π 0 r 1 − (∇1 r · ê3 )
where x ∈ R. The terms cos α and cos β have been replaced by their
definitions. The distance r is given by (2.1.19). Note that, for an
observation point x ∈ ∂Ss , the integrand becomes singular. The inte-
gral is then defined so as to maintain the overall continuity of ϕ. Lastly,
the geometrical term ϕg is given by (2.1.14).
Though it has been complicated to arrive at, (2.1.21) is a remarkable
result because it expresses Young’s and Rubinowicz’ ideas very precisely:
the radiated wavefield consists of a geometrical term ϕg that vanishes
outside S, and an edge-diffracted wave, the integral over the bound-
ary of ∂Sa , that eliminates the discontinuity in ϕg across ∂Ss . It is
remarkable that two such seemingly different representations as (2.1.7)
and (2.1.21) are equal when v(ρ/a) = H(1 − ρ/a). Had something more
complicated than v(ρ/a) = H(1 − ρ/a) been prescribed, then the result
(2.1.21) would not be quite so simple; the reader is invited to consider
this case. Alternatively, one could also have begun directly with (2.1.7)
and approximated the double integral asymptotically; see Bleistein and
Handelsman (1975, pp. 321–366), Borovikov (1994, pp. 148–155), Chako
(1965), or Stamnes (1986, pp. 136–162).
To understand (2.1.21) more fully, the integral is approximated using
the method of stationary phase (Appendix A). The problem is axisym-
metric, so that the observation point can again be placed in the plane
x2 = 0 without losing any generality. The distance r , given by (2.1.19),
can now be expressed as
r (r, θ; φ) = (r2 − 2ar sin θ cos φ + a2 )1/2 , (2.1.22)
where (r, θ) is the observation point and (a, φ) is a point on the boundary
of ∂Sa ; both sets of coordinates are shown in Fig. 2.2. Differentiating
r with respect to φ indicates that the stationary points are φ = 0, π;
the first indicates the point on the piston edge closest to the observation
point, and the second, the point furthest away. Let

r− = r (r, θ; 0),
r+ = r (r, θ; π),
and identify

cos α∓ = (a ∓ r sin θ)/r∓ .
x1
α–
r−′
r
α+
x2 r+′ x = (x1, 0, x3)
x3
e^3
Fig. 2.3. The lengths and angles arising from the stationary-phase approxi-
mation to the line integral along the piston edge.

The lengths r∓ and the angles α∓ are shown in Fig. 2.3. From the edge
of the piston a diffracted wave with a toroidal wavefront is emitted; the
shape of its wavefront will be evident from the stationary-phase approx-
imation. The principal radii of curvature (Kreyszig, 1975, pp. 78–99),
ρ1(∓) , ρ2(∓) , of this torus, along the rays emitted from φ = 0, π, are
a − r∓ cos α∓
ρ1(∓) = − , ρ2(∓) = r∓ . (2.1.23)
cos α∓
The radii ρ1(∓) carry a sign that is determined by cos α∓ , so that it is
necessary to define the square root: (−ρ1(∓) )1/2 = eiπ/2 (ρ1(∓) )1/2 .
It is also helpful to define the diffraction coefficients3 D(α∓ ) as
ie−iπ/4 1
D(α∓ ) := . (2.1.24)
(2π)1/2 cos α∓
Note that the diffraction coefficient has no length scale in its structure,
so that it does not change even if a → ∞.
With these terms identified, the stationary-phase approximation to
(2.1.21) is given by
a/ cos α∓
1/2

ϕ(x) ∼ ϕg (x) − A D(α∓ ) eikr∓ ,
∓
(kρ2(∓) )(−ρ1(∓) )

kr∓ → ∞, (2.1.25)
3 Other names such as ‘excitation coefficient’ are sometimes given to these terms. In
this book only the label ‘diffraction coefficient’ will be used.
where ϕg is given by (2.1.14) and the radii of curvature ρ1(∓) and ρ2(∓)

by (2.1.23). Looking at Fig. 2.3, it is seen that the (r+ , α+ ) ray, for the
observation point shown there, crosses the center axis, making ρ+ change
sign; thus, as (2.1.23) indicates, it undergoes a phase shift of π/2.
Equation (2.1.25) is the representation of the radiated wavefield that
would be obtained using the geometrical theory of diffraction: it has a
purely geometrical ray, represented by ϕg , followed by edge-diffracted
ones. There are two such rays, one from the point nearest the observa-
tion point and one from the furthest point; this information is contained

in the propagation terms exp(ikr∓ ); thus both rays satisfy Fermat’s
principle. The amplitudes are determined, in part, by the radii of cur-
vature (2.1.23): the square-root term is a consequence of energy flux
being conserved and it thus describes the geometrical spreading of the
edge-diffracted wave. Lastly the amplitude is determined by a diffrac-
tion coefficient (2.1.24) that contains no length; it could be calculated
by solving a radiation problem formed by the boundary condition
∂3 ϕ = ikA H(1 − x1 /a), x3 = 0. (2.1.26)
The geometrical ray is expected from geometrical optics, but geomet-
rical optics does not include diffracted rays. The notions that diffracted
rays originating on an edge satisfy Fermat’s principle, and that the
diffracted amplitudes can be determined by a geometrical spreading fac-
tor and a diffraction coefficient, are the essence of the geometrical theory
of diffraction, at least as applied to problems such as the piston radiator.
This brief description of the geometrical theory of diffraction is
complete enough for the purposes of this section, but is by no means thor-
ough. There are now several books on the subject, namely Achenbach
et al. (1982), Babic̆ and Buldyrev (1991), Borovikov and Kinber (1994),
and James (1980), that provide comprehensive treatments. Moreover,
while there are numerous references to the geometrical theory of diffrac-
tion from apertures, Keller (1957) is arguably the best reference to begin
studying. This reference can be found in the collections by Hansen (1981)
and Oughstun (1991).
However, (2.1.25) is not an accurate representation in many regions of
R: the representation fails along the central axis x3 and along the shadow
boundaries at α∓ = π/2. The first difficulty arises because the rays
emitted from the piston edge for α± < π/2 focus along the central axis
and consequently undergo a second diffraction process. This is examined
next. The second difficulty arises at a shadow boundary where the edge-
diffraction integral undergoes a jump to accommodate the jump in ϕg .
In §2.2, it will be shown that a region forms around the shadow boundary
that steadily grows, eventually choking off the aperture at a distance
equal to the Fresnel length, (2.1.8). For observation points beyond the
Fresnel length, (2.1.25) ceases to be a useful representation and (2.1.9)
becomes the appropriate one.
By the word ‘beam’ is meant a well-collimated wavefield that forms
in a neighborhood where the edge-diffracted wavefield and the geomet-
rical one begin to merge to form a single wavefield, that is, at the
Fresnel length Fl , defined by (2.1.8). If the beam is to remain collimated
for a useful propagation length, Fl or, equivalently, ka must be larger
than 1.
Continuing to leave the observation point in the x2 = 0 plane, the
scaled coordinates
x̄1 = x1 /a, x̄3 = x3 /Fl
are introduced. The goal here is to approximate the line integral in

(2.1.21) in the neighborhood x̄1 < 1, x̄3 ≈ 1, in inverse powers of
(kFl )1/2 . The resulting wavefield will be a beam in the sense just defined.
Note that
kx1
x̄1 = ,
(2π)1/2 (k Fl )1/2
so that for kx1 ≤ 1 (equivalently ka > 1), x̄1 = O[(kFl )−1/2 ].

Referring back to Fig. 2.2, with a bit of work, it is readily shown that
a(1 − x̄1 cos φ) Fl x̄3

(−∇1 r ) · êρ = ; ∇1 r · ê3 = .
r r
The distance r , first given by (2.1.19), and subsequently by (2.1.22), is
rewritten as
r (x̄1 , x̄3 ; φ) = (a2 x¯1 2 + Fl2 x¯3 2 − 2a2 x̄1 cos φ + a2 )1/2

π 2 −2
= Fl x̄3 1 + (x̄ − 2x̄1 cos φ + 1) + O[(kF ) ] .
kFl x̄23 1
l
Gathering the various pieces, it is readily shown that
a ∇1 r · êρ
= 1 + x̄1 cos φ + O[(kFl )−1 ]
r 1 − (∇1 r · ê3 )2
and
2
eikr = eik Fl x̄3 eiπ(x̄1 −2x̄1 cos φ+1)/x̄3 + O[(kFl )−1 ].
Using the integral representation (B.3) for the Bessel function of zero
order,
2π
1
J0 (z) = eiz cos μ dμ, (2.1.27)
2π 0
where arg(z) ∈ (0, π), and substituting the previous approximations in
(2.1.21), gives, after removing the scaling,

ikx3 ik(x21 +a2 )/(2x3 ) kax1 x1 kax1
ϕ(x) = ϕ (x) − A e
g
e J0 − i J1
2x3 a 2x3
+ O[(kFl ]−1 ) . (2.1.28)
Noting that x1 = r sin θ, this expression can be written for an arbi-

trary observation point. This expression can also be derived by seeking
a parabolic approximation to the reduced wave equation (Naze Tjøtta
and Tjøtta, 1980; Harris, 1987).
Equation (2.1.28) is a mathematical description of a beam: it is a
wavefield collimated close to a central axis, propagating as a plane wave
on a scale set by the wavenumber k, but having an amplitude – the
expression in braces – that evolves on a scale set by the Fresnel length
Fl . This expression matches the one given by (2.1.9) for v(ξ) = H(1−ξ),
small θ (equivalently kx1 < 1), and x3 Fl .
Let the observation point x now be given by the cylindrical coordinates
(ρ, φ, x3 ). A solution to (2.1.1) and (2.1.2) is sought by using the Hankel
transform pair, (2.1.10) and (2.1.11). One readily finds that
∗
ϕ = Ceik3 x3 ,
where C is a constant determined from satisfying (2.1.2). The radical k3

is defined as
k3 := (k 2 − kρ2 )1/2 , Im(k3 ) ≥ 0 ∀ kρ . (2.1.29)
Figure 2.4(a) shows the Riemann sheet defined by (2.1.29). On this sheet
the radiation condition describing outgoing waves is fulfilled, and it is
referred to as the physical Riemann sheet. Defining the transform
1
∗
v(kρ a) := ξv(ξ)J0 (kρ aξ)dξ (2.1.30)
0
and using the definition of the Hankel transform in (2.1.11), ϕ is readily

found to be given by
∞ ∗
v(kρ a)
ϕ(ρ, x3 ) = A(ka2 ) kρ J0 (kρ ρ) eik3 x3 dkρ , (2.1.31)
0 k3
kρ plane Im (k3) ≥ 0 (b) ξ plane
∀kρ
Re (kρ) > 0
k 0 π/ 2
–k
Re (kρ) > 0
(a) kρ = k sin ξ
Fig. 2.4. (a) The kρ plane: the Riemann sheet is selected so that Im(kρ )
≥ 0, ∀ kρ . (b) The ξ plane: the integration contour runs from 0 to π/2 − i∞.
The transformation kρ = k sin ξ has removed the branch cuts present in (a).
where k3 is defined by (2.1.29) and ∗v(kρ a) by (2.1.30), and the

integration contour is sketched in Fig. 2.4(a).
In §3.1.2 it is shown that, using (2.1.27) and the transformation kρ = k
sin ξ, (2.1.31) can be expressed as
2π
A(ka)2 π/2−i∞
∗
ϕ(x) = v(ka sin ξ) eikp̂·x sin ξ dν dξ, (2.1.32)
2π 0 0
where
p̂ = sin ξ(cos ν ê1 + sin ν ê2 ) + cos ξ ê3 . (2.1.33)
Figure 2.4(b) sketches the ξ plane: the transformation kρ = k sin ξ, which

is sometimes named the Sommerfeld transformation, has eliminated the
branch cut present in Fig. 2.4(a). This transformation will be explored
in more detail in §3.1.2. For the present, note that the radiation from
the piston has been expressed as an integration over plane waves, each
propagating in the direction p̂, (2.1.33). Also note that some directions
of propagation are complex, implying that some of the plane waves
are inhomogeneous or evanescent (Harris, 2001, pp. 22, 23). Because
(2.1.32) is expressed as an integration over angles, it is named an angu-
lar spectrum representation. As previously indicated, this representation
is especially useful for exploring the scattering of a beam; it will be the
representation used to discuss the scattering of a spherical acoustic wave
from a fluid–solid interface in §3.1 and imaging in Chapter 6.
2.1.4 Radiation impedance

Figure 2.5 shows an idealized model of a rigid piston radiator and its
mounting: the piston is treated as point mass m, with particle velocity
va (t) = Re(V0 e−iωt ), and its mounting as a spring with constant ks
and a dashpot with damping constant cs . Contrary to the suggestion of
(2.1.2), it is not the particle velocity of a radiator that is set; rather there
is an applied force fa (t) = Re(Fa e−iωt ) imposed by an electromechanical
actuator. As the piston face moves, it is loaded by the fluid with a net
force fr (t) = Re(Fr e−iωt ). The force Fr is given by

Fr = p(x )dS(x ) = Zr (ω)V0 ,
∂Sa
where p = (ρf c)(ωk)ϕ, and ϕ is given by any of (2.1.7), (2.1.21), or

(2.1.32). This is the force that is excited by the radiated wavefield. The
term Zr (ω) is called the radiation impedance; it is a complex function
of ω, though the ω dependence will not usually be explicitly indicated.
Using (2.1.7), the multiple integral can be explicitly calculated for the
rigid piston; this is described in Morse and Ingard (1968, pp. 383–
387), Pierce (1981, pp. 221–225), and Skudrzyk (1971, pp. 663–665).
ds fr = Re(Fr e–iωt)
m
ks
fa = Re(Fa e–iωt)
Fig. 2.5. A mechanical model of the rigid piston radiator: m, ks , and ds are
the mass, spring constant, and damping constant, respectively, of the piston
and its mounting. A force fa = Re(Fa e−iωt ) is applied to the piston by an
electromechanical actuator; the fluid loads the piston face, ∂Sa , with a force
fr = Re(Fr e−iωt ).
Approximating p with a plane wave gives Zr = (ρf c)(πa2 ), the specific

acoustic impedance of a plane wave times the surface area of ∂Sa .
To understand the importance of Zr , the time-harmonic solution to
the problem indicated in Fig. 2.5 is sought; it is expressed as
V0 (Zs + Zr ) = Fa , where Zs = ds − i(ωm − ks /ω), Zr = Rr + iXr .
The time-average rate at which work is done on the piston P a by the

applied force fa is then given by
V0 V0∗ V0 V0∗
P a = ds + P r , where P r = Rr .
2 2
The result (1.3.8) was used here. The first term is energy lost to the
mounting, while the second, P r , is the energy radiated into the far
field and therefore lost to the piston radiator.
The radiation impedance Zr measures how the wavefield loads the
radiator: as the energy balance indicates, its real part Rr describes how
much power is lost by the radiator and transmitted into the far field,
while comparison of Zs with Zr suggests that the imaginary part Xr
characterizes the inertial loading by and compressibility of the fluid. For
the designer of a transducer, Zr is a very important quantity to estimate
because it is needed as part of an equivalent circuit model of the overall
radiating system.
This brief discussion of transducers is intended only to draw atten-
tion to Zr . Actual transducer models are much more complicated; Auld
(1990b, pp. 317–349) and Thurston (1974, pp. 257–275) give an idea of
how piezoelectric transducers are modeled.
To circumvent the need to consider a specific model of a transducer,
following Skudrzyk (1971, pp. 666–669), the complex time-average flux
of energy across the surface ∂Sa , given by (1.3.12), is used to define Zr :
2
Zr := P cr , (2.1.34)
V0 V0∗
where, in the case of a piston with a general particle velocity distribution
across its face, such as indicated by (2.1.2),

V0
P cr = p v(ρ/a) dS.
2 ∂Sa
Recall that V0 = (ωk)A. The radiator and its electromechanical actu-

ator may exhibit complicated dynamics, so the wavefield specified on the
face of the radiator must usually be estimated. The definition (2.1.34)
does not depend on an explicit transducer model and is such that the
time-average flux of energy across ∂Sa and transmitted into the far field
is given, quite generally, by
V0 V0∗
P r = Rr , (2.1.35)
2
irrespective of the dynamics of the transducer.
Equation (2.1.35) for the piston radiator with arbitrary v(ρ/a) can
be evaluated in terms of the Hankel transforms of p and v by using
Parseval’s relation, which is given by
∞ ∞
xf (x)g(x)dx = u ∗f (u) ∗g(u)du. (2.1.36)
0 0
After introducing the scaled transform variable ξ, by setting kρ = kξ,

the Hankel transform of p can be calculated directly from (2.1.30) as
∗
∗ v(kaξ)
p(kaξ) = (ρf c)V0 ,
(1 − ξ 2 )1/2
where ∗v is given by (2.1.30) and the square root is defined just as was k3
by (2.1.29). Then, from the definition (2.1.34), it is readily shown that
1 ∗ ∞ ∗
2 2 | v(kaξ)|2 | v(kaξ)|2
Zr = 2πa (ρf c)(ka) 2 1/2
ξdξ − i ξdξ .
0 (1 − ξ ) 1 (ξ 2 − 1)1/2
(2.1.37)
A very similar use of Parseval’s relation will be made in §6.7.
2.2 Diffraction of an acoustic plane wave by an edge

A recurrent topic of the book is diffraction from edges; see also
Chapters 5 and 7. The problem to be discussed here is the simplest
one that exhibits the structure of edge-diffracted waves; it is thus very
much a canonical problem. The presentation uses the Wiener–Hopf tech-
nique, and follows that in Harris (2001, pp. 101–103) and Noble (1988,
pp. 48–85). Its presentation is intended to introduce the reader to more
detailed information about edge diffraction, as well as the more com-
plex Wiener–Hopf method that arises in studying edge diffraction in
general. The original solution to this problem was given by Sommerfeld
in 1896 using what he called branched solutions (Baker and Copson,
1987, pp. 124–149; Sommerfeld, 1967, pp. 247–266).4 Several years later,
4 As indicated in §1.7, Sommerfeld’s original paper has been translated, with notes
and historical comments, and is available as Sommerfeld (2004).
Lamb (1906) gave a simpler derivation of Sommerfeld’s solution using

parabolic coordinates and a similarity solution. As with radiation from
a piston, the edge-diffraction problem has been intensively studied, as
the general references cited in §1.7 indicate.
As in the previous section, §2.1, the problem is formulated using ϕ,
the particle displacement potential. The end results can be expressed in
terms of the acoustic pressure p = (ρf c) (ωk) ϕ. Moreover, the amplitude
of ϕ is again normalized by A = V0 /(ω k).
2.2.1 Problem solution using the Wiener–Hopf technique

Figure 2.6 indicates the geometry of the problem being considered.
A rigid barrier lies along the x1 ≥ 0 axis. It is struck by a plane acoustic
wave described as
i
ϕi = A eiκp̂ ·x
, (2.2.1)
where the direction of incidence is given by
p̂i = cos φ0 ê1 + sin φ0 ê2 . (2.2.2)
The complex wavenumber κ = k + i, where k = ω/c and > 0, has

been introduced, in a manner that is consistent with the principle of
limiting absorption, (1.2.2), in order to facilitate the implementation
of the Wiener–Hopf technique. The parameter will be sent to 0 at
the end of the calculation. Moreover, initially it will be assumed that
φ0 ∈ (0, π/2].
The plane x2 = 0 is one of reflection symmetry so that the problem
can be split into two problems, one antisymmetric and one symmetric
î
p
x1
φ
ρ
x2
Fig. 2.6. A rigid barrier lies along x1 ≥ 0; a plane wave is incident at a

direction p̂i . (ρ, φ) are polar coordinates identifying a point in the wavefield.
φ ∈ {(−π, 0) ∪ (0, π]} and the barrier lies along φ = 0.
with respect to the reflection plane. However, the symmetric part of

(2.2.1) excites no scattered wavefield. As a consequence the scattered
wavefield must be antisymmetric with respect to the plane of reflection.
This permits one to consider the solution to the scattered wavefield in
the x2 > 0 half-space; the solution in x2 < 0 is just the antisymmetric
reflection of this one. In terms of the polar coordinates (ρ, φ) shown in
Fig. 2.6, φ ∈ (0, π] for x2 > 0, while φ ∈ (−π, 0) for x2 < 0, with the
barrier at φ = 0 .
The total wavefield ϕt is conveniently expressed as the sum of the
incident field, ϕi , and the scattered wavefield, ϕ, so ϕt = ϕi + ϕ.
The scattered wavefield satisfies the reduced wave equation, (2.1.1), in
the half-space x2 > 0, with κ2 replacing k 2 . Along x2 = 0 it satisfies the
mixed boundary conditions
ϕ = 0, x1 < 0; ∂2 ϕ = −iκ sin φ0 A eiκ cos φ0 x1 , x1 > 0. (2.2.3)
In addition ϕ must satisfy the edge condition
∂2 ϕ = O(ρ−1/2 ), ρ → 0, φ = 0. (2.2.4)
Because reflected plane waves are excited along the barrier, in addition
to the diffracted wave, it may be inferred that ϕ ∼ constant on φ = 0,
as x1 → 0+ (that is, as x1 goes to zero through positive values), where
the constant can be zero.5 Lastly, ϕ must satisfy the principle of limiting
absorption, (1.2.2), as kρ → ∞, which forces the scattered wavefield to
be outgoing from the barrier.
Introduce two further unknowns ϕ+ (x1 ) and τ − (x1 ) by writing the
boundary conditions (2.2.3), at x2 = 0, as
ϕ = 0, ∂2 ϕ = τ − , x1 < 0,
+ + iκ cos φ0 x1
ϕ=ϕ , ∂2 ϕ = τ = −iκ sin φ0 A e , x1 > 0. (2.2.5)
To seek a relationship between ϕ+ and τ − , the plane-wave spectral
representation of ϕ, namely,
∞
1 ∗
ϕ(x1 , x2 ) = ϕ(β)ei(βx1 +γx2 ) dβ, (2.2.6)
2π −∞
is introduced. Equation (2.2.6) is described as a plane-wave representa-
tion, as opposed to an angular one, because the integration is over the
Cartesian wavenumber β. The radical γ is defined as
5 The edge condition given by equation (5.89) of Harris (2001) is stronger than
necessary; only the condition on ∂2 ϕ is needed. The discussion in Noble (1988,
pp. 72–76) of edge conditions is particularly helpful.
β plane ξ plane
*τ–
κ
π
0 φ0 φ
ε cos φ 0
*ϕ + κ cos φ 0
(φ)
(a) (b)
φ <ξ1> π + φ
Fig. 2.7. (a) The β plane: both ∗ϕ+ and ∗τ − are analytic in the horizontal
strip between the dashed lines. (b) The ξ plane: C is the initial contour; Cs (φ)
is the steepest-descents contour passing through the saddle point φ.
γ := (κ2 − β 2 )1/2 , Im(γ) ≥ 0 ∀ β; (2.2.7)
the branch cuts in the β plane are shown in Fig. 2.7(a). This definition
is identical to that for k3 , (2.1.29). Using the inverse Fourier transform,
∞
∗ ∗ +
ϕ(β) = ϕ (β) = ϕ+ (x1 ) e−iβ x1 dx1 . (2.2.8)
0
+ − cos φ0 x1
Note that ϕ (x1 ) = O(e ) as kx1 → ∞, because it is dominated
∗ +
by reflected waves; therefore, ϕ (β) is analytic for Im(β) < cos φ0
(Noble, 1988, pp. 11–21). In a similar way,
∞
∗ +
τ (β) = τ + (x1 ) e−iβ x1 dx1
0
(2.2.9)
sin φ0 κA
=− ;
(β − κ cos φ0 )
and
0
∗ −
τ (β) = τ − (x1 ) e−iβ x1 dx1 . (2.2.10)
−∞
Note that τ − (x1 ) = O(e−|x1 | /k|x1 |1/2 ) as kx1 → −∞, because it is com-
posed of a cylindrical, edge-diffracted wave; therefore, ∗τ − (β) is analytic
for Im(β) > −. Note, in Fig. 2.7(a), the strip of overlapping analyticity
and the pole at κ cos φ0 .
Using (2.2.8)–(2.2.10), the Fourier transform of particle displacement

at x2 = 0 is calculated in terms of ∗ϕ+ and ∗τ ± , and the two expressions
equated; this gives
sin φ0 κA
iγ ∗ϕ+ = − + ∗τ − .
(β − κ cos φ0 )
Isolating the pole κ cos φ0 , this functional relation can be written as
∗ −

τ (κ + κ cos φ0 )1/2 − (κ + β)1/2
− sin φ 0 κA
(κ + β)1/2 (β − κ cos φ0 )(κ + κ cos φ0 )1/2 (κ + β)1/2
sin φ0 κA
= i(κ − β)1/2 ∗ϕ+ + . (2.2.11)
(β − κ cos φ0 )(κ + κ cos φ0 )1/2
The right-hand side is analytic in the half-space Im(β) < cos φ0 , while
the left-hand side is analytic in the half-space Im(β) > −; both sides
share the common region of analyticity − < Im(β) < cos φ0 . Each
side is therefore the analytic continuation of the other and together they
represent a function that is entire in the β plane.
To learn what this function is, the asymptotic behavior of ∗τ − , as
|β| → ∞, in the upper half of the β plane, is needed. Using an Abelian
theorem (see Appendix A.2; Abelian and Tauberian theorems as they
relate to integral transforms are discussed in Noble (1988)), the condition
(2.2.4) implies that
∗ −
τ = O(β −1/2 ), |β| → ∞, Im(β) > −.
Therefore the left-hand side of (2.2.11) goes to zero. Invoking Liouville’s

theorem (Titchmarsh, 1939), the entire function is zero, and it follows
then that
∗ i sin φ0 κA
ϕ(β) = .
(β − κ cos φ0 )(κ + κ cos φ0 )1/2 (κ − β)1/2
Removing the assumption that x2 > 0, the scattered wavefield is given by
i A sin(φ0 /2) κ 1/2
ϕ(x1 , x2 ) = sgn(x2 )
π 1/2 2π
∞
ei(βx1 +γ|x2 |)
× 1/2
dβ, (2.2.12)
−∞ (β − κ cos φ0 )(κ − β)
where sgn(x2 ) = ±1 as x2 > 0 or < 0, respectively. At this point, → 0,

so that κ becomes k; and the restriction φ0 ∈ (0, π/2] can be relaxed
and φ0 allowed to lie in (0, π).
Equation (2.2.12) added to ϕi , (2.2.1), is the total wavefield, ϕt . The

integral in (2.2.12) can be expressed in terms of a Fresnel integral, which
is defined in Appendix B.2; the steps in this reduction are given by
Born and Wolf (1999, pp. 645–648). This reduction is not always possi-
ble in more complicated problems, so it will not be undertaken here;
rather, asymptotic approximations will be sought because these can
almost always be calculated, no matter how complicated the integral.
However, before doing so, the problem of diffraction of an incident, three-
dimensional plane wave is considered because the scattered wavefield
entails only a slight modification of (2.2.12).
2.2.2 Diffraction of an incident, three-dimensional

plane wave
The incident, plane acoustic wave, propagating in three dimensions, is
described as
ϕi 3D = ϕi eik cos θ0 x3 , κ = k sin θ0 + i,
where ϕi is given by (2.2.1) and (2.2.2), provided k is replaced by k sin θ0 ,
in the expression for κ. The angle θ0 is measured from the positive x3
axis, and will be restricted to be such that θ0 ∈ (0, π/2); the angle φ0
continues to be defined as it is in Fig. 2.6. The boundary conditions for
the scattered, three-dimensional wavefield ϕ3D now become, for x2 = 0
and x3 ∈ (−∞, ∞),
ϕ3D = 0, x1 < 0; ∂2 ϕ3D = ∂2 ϕ eik cos θ0 x3 , x1 > 0;
where ∂2 ϕ is given in (2.2.3), with κ now given by k sin θ0 + i. The
boundary conditions do not engage the x3 coordinate, so that solv-
ing the three-dimensional problem is equivalent to solving the two-
dimensional one.
In (2.2.12), κ becomes k sin θ0 (after → 0) and the whole expression
is multiplied by exp(ik cos θ0 x3 ). Therefore, the scattered wavefield ϕ3D
is given by
1/2
3D i A sin(φ0 /2) k sin θ0
ϕ (x1 , x2 , x3 ) = sgn(x2 )
π 1/2 2π
∞ i(βx1 +γ|x2 |)
e
× eik cos θ0 x3 1/2
dβ. (2.2.13)
−∞ (β − k sin θ0 cos φ0 )(k sin θ0 − β)
In the corresponding elastic-wave problem it becomes evident that the

two-dimensional result cannot be trivially extended to three dimensions
because the boundary conditions engage all three traction components

on the barrier. The full solution involves a matrix Wiener–Hopf problem
as compared with the scalar Wiener–Hopf equation solved here. The
method is conceptually identical in form with the derivation of (2.2.13),
although the technical difficulty is necessarily greater; Achenbach et al.
(1982) provide a thorough discussion of the solution and its application
to three-dimensional edge diffraction. The related problem of diffraction
from a semi-infinite screen in a viscous fluid is considered and solved in
Chapter 7.
2.2.3 A uniform asymptotic approximation

It will be helpful first to explain the qualitative features of the total
wavefield; the asymptotic analysis will explicitly exhibit each of these
features. Figure 2.8 shows a sketch of the various wavefronts incident and
scattered by the barrier. In region 1 the scattered wavefield contains both
a plane wave and the cylindrical diffracted wave; the plane wave cancels
the incident plane wave so that the total wavefield is solely composed
of the diffracted wave. In region 2 the scattered wavefield is composed
of the cylindrical diffracted wave, but the total wavefield is composed of
both this wave and the incident one. In region 3 the scattered wavefield
contains both a plane wave, representing the plane wave reflected from
the barrier, and the cylindrical diffracted wave; and the total wavefield
has the incident plane wave added to this. The various plane waves do
not suddenly turn on or off: penumbrae form within the parabolic regions
e
av
w
ed
5 ct
fle
Re
3
Diffracted wave Incident wave
2 x1
1
φ0
Y 4
X
x2 Incident wave
Fig. 2.8. A sketch of the wavefronts and their directions of propagation. The
major disturbances merge into one another across the penumbra, or Fresnel
regions, 4 and 5. The local coordinates (X, Y ) describe region 4.
4 and 5, and are characterized by Fresnel integrals (see Appendix B.2)

that combine the plane geometrical waves with the diffracted one. In
this book the penumbrae are named Fresnel regions.
Again, take x2 > 0, allow φ0 ∈ (0, π), and transform (x1 , x2 ) to the
polar coordinates (ρ, φ). Moreover, introduce the transformation β = k
cos ξ so that γ = k sin ξ; this removes the branch cut. Equation (2.2.12),
for φ ∈ (0, π), becomes

eiπ/4 A
ϕ(ρ, φ) = − D(ξ, φ0 )eikρ cos(ξ−φ) dξ, (2.2.14)
(2π)1/2 C
where D(ξ, φ0 ), a diffraction coefficient, is defined as
i e−iπ/4 cos(ξ/2) sin(φ0 /2)

D(ξ, φ0 ) := . (2.2.15)
(2π) 1/2 sin[(ξ + φ0 )/2] sin[(ξ − φ0 )/2]
Figure 2.7(b) shows the ξ plane and the initial contour C; this con-
tour begins at (0 + i∞) and ends at (π − i∞). Note the similarities
between the transformations illustrated by Fig. 2.4(b) and Fig. 2.7(b).
Note also the pole at ξ = φ0 . The contour of integration C is next shifted
to the contour of steepest descents Cs (φ) (Felsen and Marcuvitz, 1994,
pp. 377–391, 399–406; Harris, 2001, pp. 91–94), which is defined by the
equation
cos(ξ1 − φ) cosh ξ2 = 1, where ξ = ξ1 + iξ2 .
The saddle point is ξ = φ. Figure 2.7(b) indicates the pole at φ0 , the

saddle point φ, and the contour Cs (φ). To ensure convergence, the con-
tour Cs (φ) must begin in the region φ−π < ξ1 < φ and end in the region
φ < ξ1 < π + φ; the ending region is indicated in Fig. 2.7(b). Note that
for φ ∈ (0, φ0 ), a pole contribution equal to −ϕi is picked up. Also note
that as |φ − φ0 | ranges over some domain, the saddle point and pole can
come close to one another, so that (2.2.14) cannot be approximated by
the usual saddle-point approximation for all values of this parameter.
Stated more precisely, a saddle-point (or steepest-descents) approxima-
tion to (2.2.14) is not uniform in the parameter |φ − φ0 |, and a uniform
approximation should be sought.
To isolate the pole term, (2.2.15) is written as
E(ξ, φ0 )
D(ξ, φ0 ) =
sin[(ξ − φ0 )/2]
(2.2.16)
E(ξ, φ0 ) − E(φ0 , φ0 ) E(φ0 , φ0 )
= + .
sin[(ξ − φ0 )/2] sin[(ξ − φ0 )/2]
This decomposition splits the integral (2.2.14) so that ϕ = ϕnp + ϕp : one

part ϕnp has a removable singularity at ξ = φ0 , while the second part
ϕp is defined by this singularity. The integral ϕnp can be approximated
by using the standard saddle-point approximation, described in Harris
(2001, pp. 91–95), to give
E(φ, φ0 ) − E(φ0 , φ0 ) eikρ
ϕnp ∼ −A , kρ → ∞. (2.2.17)
sin[(φ − φ0 )/2] (kρ)1/2
The function E(φ, φ0 ) is defined by (2.2.16). In the limit φ → φ0 ,
i e−iπ/4
E(φ0 , φ0 ) = .
23/2 π 1/2
The integral ϕp can be reduced to a Fresnel integral. This integral is
defined and its asymptotic properties are derived in Appendix B.2. The
reduction of ϕp to a Fresnel integral is also described in Born and Wolf
(1999, pp. 645–648); the outcome is
ϕp = −A eikρ cos(φ−φ0 ) H(φ0 − φ)

e−iπ/4
+ A eikρ cos(φ−φ0 ) sgn(φ0 − φ) F [21/2 sin(|φ − φ0 |/2)(kρ)1/2 ].
π 1/2
The first term is the pole contribution, and the second term is the contri-
bution from the contour integral. The Fresnel integral changes smoothly
as |φ − φ0 | becomes small. Using (B.7), the two terms can be combined
to give
e−iπ/4
ϕp = −A eikρ cos(φ−φ0 ) F {21/2 sin[(φ − φ0 )/2](kρ)1/2 }. (2.2.18)
π 1/2
In summary, the total scattered wavefield is approximated, uniformly
in |φ − φ0 |, by
ϕ = ϕnp + ϕp , φ ∈ (0, π], (2.2.19)
where ϕnp is approximated by (2.2.17), for kρ → ∞. For φ ∈ (−π, 0), φ

is replaced by (−φ), and ϕnp and ϕp are multiplied by (−1), in (2.2.17)
and (2.1.21) respectively. And to ϕ must be added ϕi to obtain the total
wavefield ϕt .
Equation (2.2.18) allows one to determine those regions in which the
Fresnel function can be replaced by its asymptotic approximation, as
given by (B.8), and hence to estimate the sizes of the Fresnel regions 4
and 5 shown in Fig. 2.8. Define
Fr (ρ, φ − φ0 ) := sin[(φ − φ0 )/2] (kρ)1/2 ,

making 21/2 Fr the argument of the Fresnel function in (2.2.18). For

21/2 |Fr | > π 1/2 the Fresnel integral can be replaced by its asymptotic
approximation (B.8); this right-hand side is a reasonable, though not
unique, choice. Thus 21/2 |Fr | = π 1/2 defines the boundaries of the
Fresnel regions shown in Fig. 2.8. Using the local coordinates (X, Y ),
also shown in Fig. 2.8, it is straightforward to show that the boundary
of region 4 is described by the parabola
(kY )2 = 4 (π/2)[kX + (π/2)]. (2.2.20)
Outside of the Fresnel regions, (2.2.18) can be asymptotically approx-
imated for kρ large. Combining this approximation with (2.2.17) gives
eikρ
ϕ ∼ −A eikρ cos(φ−φ0 ) H(φ0 − φ) − A D(φ, φ0 ) , kρ → ∞,
(kρ)1/2
(2.2.21)
for φ ∈ (0, π]. The term D(φ, φ0 ), defined by (2.2.15), is the diffrac-
tion coefficient. For φ ∈ (−π, 0), φ is replaced by (−φ), and (2.2.21)
is multiplied by (−1). And to ϕ must be added ϕi to obtain the total
wavefield.
Were one to solve the radiation problem suggested by (2.1.26) – a less
complicated but physically similar problem to the edge-diffraction one –
one would find that a Fresnel region, having the same geometrical shape
as described by (2.2.20), projects perpendicularly from the edge x1 = a.
a α− r−′ x
α+
r+′
ka2
2π
Fig. 2.9. The Fresnel region, shown in cross-section, for the piston radiator.
Note how the Fresnel region chokes off the projected aperture at approxi-
mately the Fresnel length Fl = ka2 /(2π). This figure should be considered in
conjunction with Fig. 2.3.
And approximating the radiated wavefield non-uniformly would lead to

an expression of the form (2.2.21). Returning to the piston radiator,
§2.1.2, and in particular noting the non-uniform asymptotic approxima-
tion (2.1.9), one can conclude that a Fresnel region, composed of the
parabola (2.2.20) rotated about the x3 axis, is present. As Fig. 2.9 sug-
gests, eventually the Fresnel region grows thick enough that the aperture
projected into S is choked off; this happens for x3 ≈ Fl . Therefore,
(2.1.25) is not a particularly useful representation because it is not
accurate for much of the region S, as is suggested in Fig. 2.9. One is
thus led to consider (2.1.28) as a better description.
2.3 Summary
As has been noted previously, the problems were formulated through-
out using ϕ, the particle displacement potential. The end results can be
readily expressed in terms of the acoustic pressure p = (ρf c) (ωk) ϕ.
Moreover, if one imagines that V0 is the magnitude of the particle
velocity, then A = V0 /(ω k). As an example, (2.1.7) is expressed in terms
of the acoustic pressure as

i(ρf c)(k 2 V0 ) eikr
p(x) = − v(ρ/a) dS(x ).
2π ∂Sa kr
The study of the piston radiator has shown that there are three ways
of representing the radiation from an extended source at a boundary:
(1) The extended source, or aperture distribution, can be viewed as a

distribution of point sources; thus the radiated wavefield (2.1.7) is
a consequence of the interference of many spherical waves emitted
from a dense set of point sources. This is an adaptation of Huygens’
principle. Moreover, as indicated in its definition, (2.1.8), the Fresnel
length Fl marks the distance at which the radiated wavefield begins
to evolve into the spherical wave, represented by (2.1.9).
(2) The extended source radiates a geometrical wave and an edge-
diffracted wave; the latter focuses rays along the central axis, as
indicated by (2.1.21), and generates a thickening Fresnel region as
indicated in Fig. 2.9. It is not hard to show that this region, cal-
culated using (2.2.20), chokes off the projection of the aperture at
approximately the Fresnel length, Fl = ka2 /(2π). This restates the
inference originally drawn from (2.1.8). As a consequence, an approx-
imation such as (2.1.25) is not particularly useful. Rather, one can
2.3 Summary 51
take advantage of the representation (2.1.21) to construct a beam

wavefield, (2.1.28), in the neighborhood x1 < a, x3 ≈ Fl . It is this
region that is of most interest at high frequencies.
(3) The extended source can be decomposed into an integration over
homogeneous and inhomogeneous plane waves, as indicated by
(2.1.7). This representation allows one to derive a general expression
for the radiation impedance (2.1.37). This is a quantity of impor-
tance in the construction of equivalent circuit models, which in turn
are needed to integrate the acoustic source of radiation with its elec-
tromechanical actuator. Moreover, as will be shown in subsequent
sections, the angular spectrum representation (2.1.32) is an appro-
priate representation when the scattering of a beam from a boundary
or edge is to be calculated.
The study of the edge-diffraction problem, in addition to introducing
the Wiener–Hopf technique for solving mixed boundary-value problems,
has suggested the following:
(4) The approximation to the diffraction integral (2.2.14) has high-
lighted the distinction between uniform and non-uniform asymptotic
approximations. Moreover, it has allowed a determination of the
size of the Fresnel region, a geometrical or kinematic property of an
edge-diffracted wavefield.
(5) In the case of diffraction from an edge, the Fresnel regions, while
important, do not dominate the character of the scattered wave-
field. However, when radiation from an extended source such as the
piston radiator is studied, then these regions determine much of the
character of the radiated wavefield.
In both cases these canonical problems show the interplay between
physical understanding, intuition, and the mathematical techniques
required to distill the formulae into their most insightful form. These
techniques are fundamental to the study of wave scattering and diffrac-
tion phenomena.
3
Canonical elastic-wave problems
This chapter continues the work of the previous one. Its purpose is to
describe the following two canonical elastic-wave problems:
(1) The scattering of a spherical wave from a fluid–solid interface is

calculated using angular spectrum representations of the incident
and scattered wavefields. The principal features described are:
(a) the integral representations of the incident and scattered wave-
fields;
(b) the inter-relation between the analytic structure of the reflec-
tion coefficient and the scattered waves, particularly the leaky
Rayleigh wave;
(c) the importance of the saddle-point method in interpreting the
scattered waves. An angular-spectrum representation of the
various wavefields is used.
(2) As an elastic plate becomes thicker and thicker, the two lowest
Rayleigh–Lamb modes combine to form a Rayleigh surface wave.
The excitation of a Rayleigh wave is explored using both the
eigenmodes of an elastic plate and a Green’s state construction.
Again various applied mathematical techniques are introduced.
3.1 The scattering of a spherical wave from

a fluid–solid interface
The mathematical description of the waves scattered at a fluid–solid
interface, when a spherical acoustic wave strikes it, forms the basis
for understanding the excitation of a surface wave, and for describing
scanned acoustic imaging. By examining this interaction, not only are
52
3.1 The scattering of a spherical wave 53
a number of useful mathematical techniques introduced, but also the

underlying physical structure of the scattering at a fluid–solid inter-
face is made manifest. Moreover, it will be shown in Chapter 6 that
its solution can be progressively reworked so as to describe the imaging
mechanism of a scanned acoustic microscope. This problem is therefore
taken as canonical.
3.1.1 Description of the problem

The plane x3 = 0 separates an ideal fluid, in x3 < 0, from a homoge-
neous, isotropic, elastic solid, in x3 > 0, as indicated in Fig. 3.1. A unit
normal vector n̂ (= −ê3 ) points from the solid into the fluid. A spherical
acoustic wave is emitted from a very localized source at (0, 0, −b). When
it strikes the fluid–solid interface, part of the spherical wave is reflected,
part is transmitted as compressional and shear waves, and part becomes
a leaky wave propagating along the interface. The solution to this prob-
lem is not new: a detailed one is described in the book by Brekhovskikh
and Godin (1999, pp. 1–40), and another reworking of it is given by Pott
and Harris (1984).
The equation of motion is
δ(r)
∇2 ϕ + k 2 ϕ = −C , (3.1.1)
4πr2
where r = |x + bê3 | and the wavenumber k = ω/c, with ω being the
angular frequency and c the fluid’s wavespeed. The particle displacement
(0,0, –b)
^ Ideal fluid
n
x1
Elastic solid
x3
Fig. 3.1. The arrangement of the coordinate system. x3 > 0 is occupied by an

elastic solid, x3 < 0 by an ideal fluid. The source is placed at (0, 0, −b), where
b > 0.
54 Canonical elastic-wave problems
in the fluid is u = ∇ϕ, and the acoustic pressure p = ρf ω 2 ϕ, where ρf is

the density of the fluid. The quantity C is a constant parameter giving
the strength of the source at (0, 0, −b).
The equation of motion describing similar disturbances in the solid is
κ2 ∂i (∂k uk ) − eijk ∂j (eklm ∂l um ) + kT2 ui = 0. (3.1.2)
This is a restatement of (1.1.18) where eijk is the permutation symbol.

The compressional and shear wavenumbers are kL = ω/cL and kT =
ω/cT , where cL and cT are the respective wavespeeds given by (1.1.21)
and (1.1.22). The parameter κ = cL /cT .
The continuity conditions to be satisfied at the interface are, restating
(1.2.1),
ts · n̂ = −pf , n̂ ∧ ts = 0, us · n̂ = uf · n̂, (3.1.3)
where the traction ts = n̂ · τ . The stress tensor τ is related to the

particle displacement by (1.1.16). The subscripts s and f indicate prop-
erties of the solid and fluid, respectively. Equations (3.1.1)–(3.1.3), in
combination with the requirement that the scattered waves propagate
outward from the interface, constitute the mathematical statement of
the problem to be addressed.
3.1.2 Representations for the spherical wave

Ignoring for the moment the presence of the elastic solid, the solution
to (3.1.1) is
eikr kC
ϕ=A , A= .
kr 4π
The constant A, as in the previous chapter, normalizes the particle dis-
placement potential, but can be related to the magnitude of the particle
velocity V0 by A = V0 /ωk. Following Harris (2001, pp. 24–28), the
spherical wave can first be written as
∞ ∞
iA dk1 dk2
ϕ= ei(k1 x1 +k2 x2 +k3 |x3 +b|) , (3.1.4)
2πk −∞ −∞ k3
where
k3 = (k 2 − k12 − k22 )1/2 , Re(k3 ) ≥ 0, Im(k3 ) ≥ 0. (3.1.5)
This choice of the branch of k3 is made so that the spherical wave is

outgoing or decays as |x3 + b| → ∞. Second, it can be written as

i i A 2π π/2−i∞ ik p̂ i ·x ikb cos ξ
ϕ = e e sin ξ dν dξ, (3.1.6)
2π 0 0
where
p̂ i = sin ξ (cos ν ê1 + sin ν ê2 ) + cos ξ ê3 . (3.1.7)
Note the similarity between (3.1.6) and (2.1.32), and between (3.1.7)
and (2.1.33). The superscript i has been added in (3.1.6) to indicate
that this is the incident spherical wave. Equation (3.1.6) assumes that
(x3 + b) > 0; it was derived using the transformation
k1 = k sin ξ cos ν, k2 = k sin ξ sin ν, k3 = k cos ξ,
where the accessible parts of the ξ plane are determined by (3.1.5);
Fig. 2.4(b) indicates this.
Equation (3.1.6) is an integration of the plane waves
i
eik p̂ ·x
not only over all real directions of propagation, but also over a range
of complex directions. The contour for ξ (see Fig. 2.4(b)) is taken to
be from 0 to π/2, and then to (π/2 − i∞); for ξ ∈ (π/2, π/2 − i∞),
ξ = π/2 + iξ2 , with ξ2 ≤ 0. In this case the direction given by (3.1.7)
becomes complex:
p̂ i = cosh ξ2 (cos ν ê1 + sin ν ê2 ) − i sinh ξ2 ê3 .
The imaginary part of p̂ i produces a decay in the ê3 direction. The
integration over complex angles is essential if the complete curvature of
the spherical wave is to be captured by (3.1.6).
It will be useful to have two additional representations for the spherical
wave. Using cylindrical coordinates (ρ, φ, x3 ), where x1 = ρ cos φ and
x2 = ρ sin φ, ϕi can be written as
2π
i i A π/2−i∞ ik cos ξ (x3 +b) ikρ sin ξ cos(ν−φ)
ϕ = e sin ξ e dν dξ.
2π 0 0
Using the representation (2.1.27) for the Bessel function J0 this integral
can be written, first, as
π/2−i∞
i
ϕ = iA eik cos ξ (x3 +b) J0 (kρ sin ξ) sin ξ dξ, (3.1.8)
0
and, second, as

i i A π/2−i∞ ik cos ξ (x3 +b) (1)
ϕ = e H0 (kρ sin ξ) sin ξ dξ. (3.1.9)
2 −π/2+i∞
The contour for (3.1.9) passes above the branch cut for the Hankel func-
tion; that cut proceeds from 0 to (−π/2 − i∞). Both representations
are integrals over cylindrical waves. Equation (3.1.9) is very useful when
calculating an asymptotic expansion to ϕi .
3.1.3 Scattered wavefields

Consider briefly the expressions for the plane waves reflected and trans-
mitted at a fluid–solid interface when a plane wave, incident from the
fluid, strikes it. Harris (2001, pp. 37–55) indicates one way to calculate
these waves. Reflection and transmission, and their respective coeffi-
cients, are calculated using the particle displacement u. Each plane wave
is characterized by a unit propagation vector p̂ and a unit polarization
vector d̂. The incident plane wave is written as
i
ui = d̂ i eik p̂ ·x
, d̂ i = p̂ i ,
where p̂ i is given by (3.1.7). The reflected and transmitted waves must

all phase-match to the incident wave; thus,
k sin ξ = kL sin ξL = kT sin ξT ,
where the angles ξL and ξT are indicated in Fig. 3.2. These waves are
given by
î
p ^r
p
ξ ξ
^
dT
x||
ξL ^L
p
ξT
^T
p
x3
Fig. 3.2. The geometry of the incident and scattered rays. The incident and
scattered propagation vectors, with the defining angles ξ, ξL , and ξT , are
shown, as is the polarization vector d̂ T .
r
ur = R(ξ) d̂ r eik p̂ ·x
, d̂ r = p̂ r , p̂ r = sin ξ ê − cos ξ ê3 ,
L
uL = TL (ξ) d̂ L eik p̂ ·x
, d̂ L = p̂ L , p̂ L = sin ξL ê + cos ξL ê3 ,
T
uT = TT (ξ) d̂ T eik p̂ ·x
, p̂ T ∧ d̂ T = êT , p̂ T = sin ξT ê + cos ξT ê3 .
(3.1.10)
Two new unit vectors,
ê = cos ν ê1 + sin ν ê2 , êT = ê3 ∧ ê , (3.1.11)
have been introduced; (ê , êT , ê3 ) form a right-handed triad. The
superscripts r, L, and T indicate the reflected, transmitted longitudi-
nal, and transmitted transverse waves, respectively. The reflection and
transmission coefficients, R(ξ) and TL,T (ξ) are given by
A− (ξ) 2κ κf (ρf /ρs ) cos ξ cos(2ξT )
R(ξ) = , TL (ξ) = ,
A+ (ξ) A+ (ξ)
2κf (ρf /ρs ) cos ξ sin(2ξL )

TT (ξ) = − , (3.1.12)
A+ (ξ)
where
A± = cos ξ[sin(2ξL ) sin(2ξT ) + κ2 cos2 (2ξT )] ± κ κf (ρf /ρs ) cos ξL .
(3.1.13)
The parameter κf = c/cT . Recall that κ = cL /cT .
Returning to the problem under discussion and using (3.1.6), the
incident wave becomes

k A 2π π/2−i∞ i i
ui = − d̂ (ξ, ν) eik p̂ ·x eikb cos ξ sin ξ dν dξ.
2π 0 0
Noting that the problem is linear and that the scattered plane waves
are given by (3.1.10), the waves scattered from the interface are directly
calculated to give the equations

k A 2π π/2−i∞ r r
u =−
r
d̂ (ξ, ν)R(ξ) eik p̂ ·x eikb cos ξ sin ξ dν dξ
2π 0 0
(3.1.14)
and

k A 2π π/2−i∞ L,T L,T
u L,T
=− d̂ (ξ, ν)TL,T (ξ) eikL,T p̂ ·x eikb cos ξ
2π 0 0
× sin ξ dν dξ. (3.1.15)
In (3.1.10) are listed the propagation and polarization vectors. It should
be clear to the reader that all further outcomes depend on the analytic
structure of the phases in the integrands and the analytic structure of

A± (ξ), (3.1.13).
3.1.4 Wavefield in the fluid

For the remainder of this chapter only the wavefield in the fluid, (3.1.14),
will be considered, because it is the wavefield used to form an image
by a scanning acoustic microscope, discussed in Chapter 6. Using the
transformation that yielded the representation (3.1.9), ur is written as
π/2−i∞
iA (1)
r
u = ∇ R(ξ) eik(b−x3 ) cos ξ H0 (kρ sin ξ) sin ξ dξ.
2 −π/2+i∞
(3.1.16)
Recall that cylindrical coordinates (ρ, φ, x3 ) are introduced to arrive at
(1)
this representation. Noting the Sommerfeld representation of H0 (see
Appendix B.1) it is seen that (3.1.16) is a double integral whose phase is
Φ(ξ, μ) = (b − x3 ) cos ξ + ρ sin ξ cos μ.
Note: the second, mixed derivative ∂ξ ∂μ Φ = 0 when evaluated at the
stationary point (ξ ∗ , μ∗ ), where (b − x3 ) sin ξ ∗ = ρ cos ξ ∗ and μ∗ = 0.
As a consequence, (3.1.14) can be treated as an iterated integral, to
leading order, when asymptotically approximating it; see Bleistein and
Handelsman (1975, pp. 321–366), Chako (1965), and Stamnes (1986,
(1)
pp. 136–162). Though an asymptotic approximation to H0 is used next,
one could work directly with (3.1.14).
(1)
Using the asymptotic approximation for H0 , and introducing the
spherical coordinates (r, θ, φ), where (b − x3 ) = r cos θ and ρ = r sin θ,
(3.1.14) can be approximated as
1/2
kA 2
ur ∼ − e−iπ/4 d̂ r (ξ)R(ξ) eikr cos(ξ−θ) (sin ξ)1/2 dξ,
2 πkρ C(θ)
kρ → ∞. (3.1.17)
The phase of the integrand of (3.1.15) has a saddle point at ξ = θ. The
contour C(θ) is the steepest-descents contour; Fig. 3.3(b) is a sketch of
a part of this contour as it passes through θ. The method of steepest
descents, which is introduced in Appendix A.3, is described more fully in
Felsen and Marcuvitz (1994, pp. 370–391), and the saddle-point method
(which gives the leading-order term of a steepest-descents approxima-
tion) in Harris (2001, pp. 91–94). Lastly, note that the dependence on
ξ plane ξ plane
ξbL ξR
ξbT
ξR π/ 2
θ
ξS
(a) (b)
Fig. 3.3. The structure of the complex ξ plane for Re(ξ) ≥ 0. (a) The con-
tour for (3.1.14), the branch cuts beginning at the branch points ξbL and
ξbT , the leaky Rayleigh pole ξR , and the Stoneley pole ξS are indicated.
(b) The steepest-descents contour C(θ) of (3.1.15) is sketched (the steepest-
descents method is described in Appendix A.3). The spatial relation between
the steepest-descents contour and the leaky Rayleigh pole ξR is shown.
ν in d̂ r has been dropped, because it is clear, at this point, that the

wavefield is axisymmetric.
The subsequent asymptotic approximation of this integral depends on
the proximity of θ to the poles and branch points of R(ξ). A+ (ξ), which
is given by (3.1.12), has four zeros, ±ξR and ±ξS , and four branch points,
±ξbL and ±ξbT . The branch points are defined by the equations
sin ξbL = c/cL , sin ξbT = c/cT .
Figure 3.3(a) shows the right half of the complex ξ plane, with these
poles and branch points indicated. The poles of R(ξ) correspond to
leaky Rayleigh waves (±ξR ) and Stoneley waves (±ξS ), and the branch
points to various lateral waves. The wiggly lines in Fig. 3.3 indicate the
branch cuts corresponding to these branch points. These cuts, which
define the Riemann sheet on which the problem is solved, are determined
by demanding that Im(cos ξI ) ≥ 0, I = L, T , everywhere on the Riemann
sheet of interest. Because R(ξ) is a reflection coefficient for an incident
plane wave, the branch cuts are chosen to ensure that when critical
refraction takes place in the solid, the refracted waves decay with depth.
In this chapter, only the effect of the leaky Rayleigh-wave pole is
considered; a non-uniform expansion is used that does not accurately
account for what happens when the saddle point lies in the neighborhood
of a pole, a branch point, or both a pole and a branch point. It was
shown in §2.2.3 how a uniform approximation can be calculated if a
pole is present. The case here requires a uniform approximation when a
pole and branch point are close to a saddle point, which is possible to
evaluate; unfortunately, this uniform expansion is itself complicated. For
the purposes of this section the non-uniform approximation is adequate.
A non-uniform asymptotic approximation to (3.1.17) is given as
ur ∼ ug d̂ r (θ) + χ(ξR , θ) uR d̂ r (ξR ), kr → ∞, (3.1.18)
where

1 if C(θ) encloses ξR ,
χ(ξR , θ) =
0 otherwise.
The polarization vector d̂ r (φ) is given by (3.1.10) with ξ replaced by θ;

similarly, the polarization vector d̂ r (ξR ) is given by (3.1.10) with ξ
replaced by ξR . The specularly reflected spherical wave ug is given as
eikr
ug = ikA R(φ) . (3.1.19)
kr
Note that the condition that kρ → ∞, used in (3.1.17), does not influence
the final outcome describing ug ; that is, (3.1.19) remains accurate even
as ρ → 0.
Describing uR requires somewhat more detail. The leaky Rayleigh
pole is
ξR = βR + iαR , βR , αR > 0. (3.1.20)
In all cases of interest here, αR /βR 1. Somekh et al. (1985) show that
A− (ξR )
≈ 2iαR . (3.1.21)
dA+ /dξ(ξR )
This approximation will be used frequently. Moreover, except in the
phase terms, the approximation ξR ≈ βR is used. Again noting the
various unit vectors introduced in (3.1.10), the unit vectors p̂ R and d̂ R ,
given by the expressions
p̂ R = sin βR ê − cos βR ê3 , p̂ R ∧ d̂ R = êT ,
are introduced; êT is defined by (3.1.11). Note that d̂ r (ξR ) ≈ p̂ R (βR ).

Incorporating these approximations and the newly defined unit vectors,
uR can be expressed as
Reflected spherical wave

(0,0,–b) βR
Leaky Rayleigh wave
x||
(0,0,b)
x3
Fig. 3.4. A drawing indicating the nature of the leaky Rayleigh wave excited
by a ray incident at the angle βR . Also shown is the ray describing the reflected
wave being emitted from the image point (0, 0, b).
1/2
2π R
R
u = kA 2αR e−iπ/4 eikp̂ ·x cosh αR
krR
R
× eikb cos βR cosh αR
ekd̂ ·x sinh αR
ekb sin βR sinh αR
, (3.1.22)
where rR = ρ/ sin βR .
Equation (3.1.22) describes an inhomogeneous wave propagating away
from the interface in the direction p̂ R , and decaying in the direction
−d̂ R . The possibility of unlimited growth in the direction d̂ R is removed
from ur , (3.1.16), by the indicator function χ(ξR , φ). Viewing it in the
coordinates indicated in Fig. 3.1, (3.1.22) describes a Rayleigh surface
wave that steadily radiates, or leaks, into the fluid so that, as it propa-
gates along the interface, it also decays. Further, it only appears outside
a right circular cone whose vertex is at (0, 0, b) and which opens upward,
cutting the surface at x3 = 0 in a circle of radius b cot βR . Figure 3.4 is
a drawing that attempts to indicate these relationships.
3.1.5 Summary of §3.1

(1) Equations (3.1.18), (3.1.19), and (3.1.22) are the principal results of
this section. They describe the most important waves scattered from
the fluid–solid interface, namely, the specularly reflected spherical
wave and the leaky Rayleigh wave.
(2) By using the angular spectrum representation of the incident wave,
(3.1.6), representations of the reflected wavefield, (3.1.14), and
the transmitted wavefields, (3.1.15), are readily calculated, and

expressed in a form that indicates the underlying role of the scat-
tering of a plane wave. Moreover, these representations indicate
that many of the physical outcomes of the problem can be inferred
directly from the analytic behavior of the reflection and transmis-
sion coefficients. The general similarity between the expressions of
§2.1.3 and §3.1.2 should be noted.
(3) Representations of the spherical wave as an integral over cylindrical
waves, (3.1.8) and (3.1.9), are also given. Equation (3.1.8) exhibits
the spherical wave as an inverse Hankel transform; (3.1.9) proves
quite useful in calculating the asymptotic approximation begun with
(3.1.14).
3.2 Rayleigh–Lamb modes and Rayleigh surface waves

3.2.1 Introduction
Rayleigh–Lamb modes are described in several books: Achenbach (1973,
pp. 220–236), Auld (1990b, pp. 76–94), Brekhovskikh and Goncharov
(1985, pp. 75–86), and Miklowitz (1978, pp. 178–209). They are inplane
elastic waves that are guided in a uniform, two-dimensional waveguide
with traction-free surfaces. The equations describing two-dimensional,
inplane, elastic waves are obtained from (1.1.18) by setting u3 and all
derivatives ∂3 to zero; that is, the only particle displacement occurs in
the (x1 , x2 ) plane and the independent spatial variables are (x1 , x2 ).
3.2.2 Notation
The wavenumber kT = ω/cT and is used, initially, to scale lengths;
ω is the angular frequency and cT is the shear wavespeed. The Rayleigh
wavespeed cR is approximately 0.9cT , so that at high frequencies the
shear-wave wavelength approximates the Rayleigh-wave wavelength.
Upper-case letters represent the lengths and lower-case letters the scaled
lengths, unless otherwise indicated. The coordinates are xα = kT Xα ,
the particle displacement components are uα = kT Uα , and the nominal
thickness of the waveguide is 2h0 = 2kT H0 .
The shear coefficient μ is used to scale the stress components: τ1 =
τ11 /μ, τ2 = τ12 /μ, and τ3 = τ22 /μ, where the ταβ are the components.
The subscripts α, β = 1, 2. Note: this is the only place in this section
where two subscripts are needed; elsewhere, τp α indicates the pth mode,
α component.
It is useful to introduce combinations of elastic parameters that will
arise frequently. These are:
a = λ/(λ + 2μ), b = μ/(λ + 2μ), c = 4(a + b). (3.2.1)
Also of use is the relation
τ3 = aτ1 + c ∂2 u2 . (3.2.2)
3.2.3 Rayleigh–Lamb modes

The two lowest Rayleigh–Lamb modes are given by the following
expressions, which are taken from Brekhovskikh and Goncharov (1985,
pp. 75–86). The lowest antisymmetric mode ua is given as
⎡ ⎤
sinh(γL x2 ) γL γT sinh(γT x2 )
⎡ ⎤ −
ua 1 (x2 ) ⎢ cosh(γL h0 ) pβa cosh(γT h0 ) ⎥
⎢ ⎥
⎣ ⎦ = Ca ⎢ ⎥
⎢ ⎥ , (3.2.3)
ua 2 (x2 ) ⎣ −iγL cosh(γL x2 ) βa cosh(γT x2 ) ⎦
−
βa cosh(γL h0 ) p cosh(γT h0 )
while the lowest symmetric mode us is given as
⎡ ⎤
cosh(γL x2 ) γL γT cosh(γT x2 )
⎡ ⎤ −
us 1 (x2 ) ⎢ sinh(γL h0 ) pβs sinh(γT h0 ) ⎥
⎢ ⎥
⎣ ⎦ = Cs ⎢
⎥
⎥ . (3.2.4)
⎢
us 2 (x2 ) ⎣ −iγ L sinh(γ x
L 2 ) βs sinh(γT 2 ⎦
x
−
βs sinh(γL h0 ) p sinh(γT h0 )
In (3.2.3), βa is substituted wherever needed; in (3.2.4), βs is substituted
wherever needed and Ca , Cs are arbitrary constants.
The equations whose solutions give the dispersion relation for each
mode are
γL γT tanh(γT,L h0 ) = p2 tanh(γL,T h0 ), (3.2.5)
where the first subscript gives the antisymmetric relation and the second
the symmetric one. The radicals γL and γT are defined as
γL = (κ2 β 2 − 1)1/2 /κ, γT = (β 2 − 1)1/2 , κ = cL /cT = b−1/2 ,
where cL is the compressional wavespeed; it is readily found that γL,T
are real and positive. The term p is defined as
p = (β 2 − 1/2)/β.
For these two lowest modes it is useful to express βa, s as
βa = βR + a , βs = βR − s , a + s = 2, (3.2.6)
where βR is the wavenumber of a Rayleigh surface wave, and 2 =

(βa − βs ). In general, is not small and must be found from a knowledge
of βa, s . However, as Fig. 3.5 suggests, as h̄0 (or more simply h0 ) becomes
large, a → s → , and is given by
df −1
2 −2γL h̄0 −2γT h̄0 R
= 2p e −e . (3.2.7)
dβ
This approximation is taken from Brekhovskikh and Goncharov (1985,
pp. 79–81). βR is used wherever β is needed. The approximation (3.2.7)
is accurate provided γT h̄0 > 1 or, more approximately stated, provided
h̄0 1. The Rayleigh function, given as
fR (β) = p2 − γL γT ,
has roots ±βR . The Rayleigh wavenumber βR is approximated (Auld,

1990b, p. 92) by
cL
cT
h0
cR
ε
β h0
Fig. 3.5. A sketch of the dispersion relation for the two lowest Rayleigh–Lamb
modes. Recall that h0 = ωH0 /cT . Solid curve: antisymmetric mode, eigenvalue
βa = βR + a . Short-dash, long-dash curve: symmetric mode, eigenvalue βs =
βR − s . a + s = 2 measures the horizontal distance between the two curves.
At large h0 , a = s = , where 1, as indicated in the figure. cL , cT , and
cR are the compressional, shear, and Rayleigh wavespeeds, respectively; the
long-dashed lines indicate the slopes cL /cT , 1, and cR /cT .
cR (0.87 + 1.14ν)
βR = ω/cR , = ,
cT 1+ν
where ν = λ/ [2(λ + μ)].
For a uniform waveguide, a and s are constant, x̄2 = x2 and
h̄0 = h0 . Therefore, the sum of these two modes gives the net particle
displacement:

u = eiβR x1 C eia x1 ua + e−is x1 us , (3.2.8)
where Ca = Cs = C. At large h0 , it is very difficult to excite just one mode

because a, s ≈ 1. Hence, even for a uniform waveguide, both modes
would be excited. Once excited, however, both modes propagate inde-
pendently. Equation (3.2.8) becomes a wave propagating at the Rayleigh
wavespeed, but with a modulated amplitude. Approximating (3.2.3) and
(3.2.4) for large h0 would indicate that (3.2.8) has most of its amplitude
near the upper surface and decays toward x2 = 0; its amplitude near the
lower surface is negligible. It is effectively a Rayleigh wave. When the
term ‘Rayleigh surface wave’ was used in §3.1.1, this was the wave being
referred to. If the propagation path is long enough, the two terms will
move out of phase so that the Rayleigh wave will shift from the upper
surface to the lower one; see Auld (1990b, pp. 93, 94) and Brekhovskikh
and Goncharov (1985, pp. 79–81). Experimental confirmation of this
shifting is described by Ti et al. (1997).
3.2.4 A framework for elastic waveguide problems

The equations of motion will be written in a form that is not found in
the books cited in §1.7, though it is used by Harris and Block (2005),
Kirrmann (1995), and Maupin (1988). This manner of writing the
equations of motion was also used for antiplane shear problems by
Malischewsky (1987, pp. 54–69). The purpose of finding an alternative
way of writing the equations of motion, (1.1.18), is to eliminate second-
order derivatives of the form ∂1 ∂2 so that, when the pth eigenmode is
sought, the eigenvalue βp will appear in an isolated way.
The vector
U := [u1 , u2 , τ1 , τ2 ]T
is introduced, where the superscript T indicates the transpose. The

equations of motion are then written as
(L − ∂1 ) U = F, (3.2.9)
where
⎡ ⎤
0 −a ∂2 b 0
⎢−∂2 0 0 1 ⎥
L := ⎢
⎣ −1
⎥. (3.2.10)
0 0 −∂2 ⎦
0 −c ∂22 − 1 −a ∂2 0
F is a vector containing body forces and is described by
F = [0, 0, f1 , f2 ]T ,
where the fα = Fα /(kT μ). The Fα are the body-force components per
unit mass. The constants a, b, and c are given by (3.2.1). This system
of equations is derived by using both the equation of motion, (1.1.18),
and the constitutive relation (1.1.16), and subsequently using (3.2.2) to
eliminate τ3 from the equations.
An inner product may be defined as
h0
∗
V, U := −i vβ τβ − σα∗ uα dx2 , (3.2.11)
−h0
where
V = [v1 , v2 , σ1 , σ2 ]T , U = [u1 , u2 , τ1 , τ2 ]T .
Using this inner product, the following integration by parts relation is

calculated:
LV, U + V, LU = i [(v1∗ τ2 − σ2∗ u1 ) + (v2∗ τ3 − σ3∗ u2 )] |−h
0 h
0
. (3.2.12)
τ3 is given by (3.2.2) and σ3 is defined similarly.

For the moment, imagine that the waveguide is uniform, with thick-
ness 2h̄0 . Consider a wave of the form
U = up (x2 )eiβp x1 .
Substituting this into (3.2.9), with F set to zero, gives the following
eigenvalue problem:
Lup = iβp up ,
(3.2.13)
τp 2 = τp 3 = 0, at x2 = −h0 , x2 = h0 ,
where the x1 is fixed. up is the eigenmode and βp the eigenvalue.
The eigenmodes of (3.2.13) are numbered as follows: they occur in
pairs and are labeled so that one member of each pair carries energy
or decays exponentially in the +x1 direction, while the other member
does so in the −x1 direction. Subscripts p are integers and take plus and
minus values, but not 0; a plus value is taken for propagation in the +x1
direction and a minus is taken for propagation in the −x1 direction.
Thus β−p = −βp . Equations (3.2.3) and (3.2.4) are two eigenmodes
that satisfy (3.2.13). However, they have different symmetries: β1 = βa
satisfies the antisymmetric dispersion equation, β2 = βs the symmetric
dispersion equation; both dispersion equations are given by (3.2.5). In
the development of this section, subscripts 1, 2, . . . will be used to identify
the eigenvalues βp , with no indication of the symmetry.
Consider next two eigenmodes up and uq . Substituting them into
(3.2.12) gives
− i(βq∗ − βp )Pqp = 0, Pqp := uq , up . (3.2.14)
This is the orthogonality condition. Note that the eigenvalues can be

complex, though in practice the two eigenvalues βa,s of interest are
always real.
A more precise description of the previous expressions is to state that
the eigenmodes form a biorthogonal system; that −L, with the same
boundary conditions as indicated in (3.2.13), is the adjoint operator; and
that (3.2.14) is an expression of biorthogonality. It is assumed through-
out this chapter that the up (x2 ) form a complete set, and thus a vector
U(x2 ) can be uniquely expressed as

U(x2 ) = cn un (x2 ).
n
These issues are discussed further by Besserer and Malischewsky (2004)

and Kirrmann (1995), and a general discussion of biorthogonal expan-
sions is given by Herrera and Spence (1981).
By taking the complex conjugate of (3.2.13), and giving close attention
to the mode labeling previously noted, it can be shown that
if βq∗ = βp , then u∗q = u−p .
Moreover, by asking that the modes of (3.2.13) be symmetric upon

reflection in x1 = constant, it can be shown that, if βq∗ = βp , then
u∗q 1 = −up 1 , u∗q 2 = up 2 , τq∗1 = τp 1 , τq∗2 = −τp 2 , τq∗3 = τp 3 .

(3.2.15)
The modes are normalized so that the conditions (3.2.15) are satisfied.
By setting Ca = Cs = i, the eigenmodes ua and us , (3.2.3) and (3.2.4)
respectively, are normalized so that these conditions are satisfied.
3.2.5 Excitation of a waveguide

In practical terms one is often interested in the excitation of a wave-
guide by a transducer that applies a known traction on one surface.
The discussion of §3.2.3 draws on the connection between the lowest
Rayleigh–Lamb modes and the Rayleigh wave; we now turn our attention
to exciting such modes within a waveguide and draw upon the treatment
given by Folguera and Harris (1998). By using a Rayleigh wave trans-
ducer one can strongly excite the lowest two Rayleigh-like modes. To see
this, consider a flat, straight waveguide and apply a localized traction
on the upper surface, x2 = h0 , of the form
τ3 = 0, c∂2 u2 + aτ1 = AeikT βR X1 [H(X1 + a) − H(X1 − a)],

(3.2.16)
where H(X1 ) is the Heaviside function and A is a constant.

In the spirit of the eigenvalue problem (3.2.13) and the framework of
§3.2.4, we solve
(L − ∂1 )U = 0 (3.2.17)
subject to (3.2.16).
We use the Green’s states, UG
(α) , which for the infinite waveguide
satisfy

(α) = [0, 0, δ1α , δ2α ] δ(x1 − x1 )δ(x2 − x2 ),
(L − ∂1 )UG T
(3.2.18)
where δαβ is the Kronecker delta symbol and α, β = 1, 2. The subscript

(α) delineates the two Green’s states that are used. The forcing is now
a source located at (x1 , x2 ) and the boundary conditions for the Green’s
state are homogeneous, i.e.,
G
τ3(α) = 0, c∂2 uG G
2(α) + aτ1(α) = 0. (3.2.19)
The framework (3.2.12) leads to

U, (L − ∂1 )UG∗
(α) + (L − ∂1 )U, U(α) = iu(x1 , x2 ) · δα δ(x1 − x1 ),
G∗
(3.2.20)
where δα = (δ1α , δ2α ). We are now in a position to use the boundary

conditions (3.2.16) and integrate (3.2.20) over x1 . Then, assuming that
the waves are damped at infinity, one arrives at
kT a
(iβR x1 )
uα (x1 , x2 ) = A uG
2(α) (x1 , kT h0 ; x1 , x2 )e dx1 . (3.2.21)
−kT a
As we are primarily exciting the two lowest Rayleigh–Lamb modes,

the Green’s state can be expanded to

(−a) iβa |x1 −x1 | (−a)
UG
(α) = [Bα e U + Bα(−s) eiβs |x1 −x1 | U(−s) ]H(x1 − x1 )

+ [Bα(a) eiβa |x1 −x1 | U(a) + Bα(s) eiβs |x1 −x1 | U(s) ]H(x1 − x1 ) + . . . ,
(3.2.22)
where here the superscripts a, s denote antisymmetric and symmetric
modes, and the associated + or − signs before these letters indicate
that they propagate to the right or left, respectively. The waves propa-
gate outwards from the source. These Green’s states involve constants,
the B, that are identified from another application of the inner product
framework:
(a,s)
(L − ∂1 )UG
(α) , U = [0, 0, δα1 , δα2 ]T δ(x2 − x2 ), U(a,s) δ(x1 − x1 ).
(3.2.23)
Integrating across x1 from x1 − 0 to x1 + 0, and using (3.2.14), gives
Bα(±a,±s) = −iu∗(±a,±s) · δα /Pnn , (3.2.24)
where n is either a or s, depending on whether the mode is anti-
symmetric or symmetric.
Putting all this together, using (3.2.21), the resulting displacement is
sin(kT a) (iβR x1 )
u(x1 , x2 ) = −i2kT aA e A(x1 , x2 ), (3.2.25)
kT a
with
u∗a2 (h0 ) u∗ (h0 )
A(x1 , x2 ) = ua (x2 )eix1 + s2 us (x2 )e−ix1 . (3.2.26)
Paa Pss
Here ua and us are the lowest antisymmetric and symmetric modes,
respectively, as given in §3.2.3. Notably the factor 1/kT a ensures
that, although higher-order mode contributions are present, and omit-
ted in the analysis, the contribution from the lowest two modes
dominates.
3.2.6 Summary of §3.2

(1) The modes that exist within a straight, uniform waveguide are the
Rayleigh–Lamb modes which satisfy a dispersion relation (3.2.5).
As the waveguide gets thicker the lowest two modes are perturba-
tions from the Rayleigh surface wave that exists near the surface
of an elastic halfspace. It is difficult to excite these modes indepen-

dently; the result is that a modulated Rayleigh wave will often be
observed.
(2) A useful framework for elastic waveguide problems in presented and
applied to the excitation of a waveguide by a Rayleigh transducer.
The resulting excitation is dominated by contributions from these
lowest modes.
4
Radiation and impedance
Radiation from transducers is defined by the near field, from which the
far field can be determined. The near field of a general source can be
represented as an integral over the source surface in terms of unknowns
on the surface. The integral can be derived using the powerful argument
that is known as reciprocity. The key ingredient in the integration is the
fundamental solution for a point force. The purpose of this chapter is to
introduce reciprocity and point-force solution, and to understand them
through example. The fundamental solution, known as the Green’s ten-
sor, is used to generate Gaussian beam solutions that are commonly used
in practical simulation of transducer fields. Reciprocity is used to con-
sider the elastic fields generated by the motion of a finite-sized particle
and, at the same time, to find the force on a particle due to plane-wave
incidence. A common feature of the point-force solution and the parti-
cle radiation is the notion of impedance or its inverse, admittance. This
is a direct measure of the energy loss through radiation, and can be
interpreted in terms of both near- and far-field effects.
4.1 Reciprocity
The time-harmonic equations of motion for two independent elastic
wavefields, indicated by the superscripts 1 and 2, are
1,2
∂j τji + ρfi1,2 + ρω 2 u1,2
i = 0. (4.1.1)
We consider solutions in a bounded region Rx with surface ∂Rx . The
tractions on ∂Rx are t1,2
i
1,2
= τji n̂j . The unit normal n̂ to ∂Rx is outward.
Take the scalar product of each equation in (4.1.1) with the particle
displacement of its reciprocating wavefield, and subtract. This gives

ρ fi2 u1i − fi1 u2i = −u1i ∂j τji2
+ u2i ∂j τji
1
. (4.1.2)
71
72 Radiation and impedance
Using the relation between stress and strain, (1.1.3), we note that
1 2 2 1
τji ∂j u i = τji ∂j ui , so that the right-hand side of (4.1.2) equals
1 2 2 1
∂j τij ui − τij ui . The fundamental statement of elastodynamic reci-
procity follows after integration:

2 1 2 1
f · u − f 1 · u2 ρdV = u · t − u1 · t2 dS. (4.1.3)
Rx ∂Rx
One special case of (4.1.3) is of particular interest: point forces at two

points and the associated displacements at each point produced by the
force at the other. Let us assume that the integral over ∂Rx vanishes and
that f 1,2 = a1.2 δ(x − x1,2 ), where the a1,2 are constant vectors. Then
the reciprocity statement becomes a2 · u1 (x2 ) = a1 · u2 (x1 ). This is a
pointwise version of the general integral identity (4.1.3).
4.2 Green’s tensor

We consider the time-harmonic Green’s tensor. The defining equation
follows from (1.1.4), with f = âδ(x), and the associated displacement u
satisfies
c2L ∇∇ · u − c2T ∇ ∧ ∇ ∧ u + ω 2 u + âδ(x) = 0. (4.2.1)
The solution is well known and is obtained, for instance in Harris (2001),
using transforms. Here we describe an alternative method based on the
Helmholtz decomposition (1.1.19) for u. Using the identity
â
âδ(x) = −∇2
4πr
â â
= −∇∇ · +∇∧∇∧ ,
4πr 4πr
where r = |x|, gives the pair of equations
1 1 1 1
∇2 ϕ + kL
2
ϕ= â · ∇ , ∇2 ψ + kT2 ψ = â ∧ ∇ .
4πc2L r 4πc2T r
These in turn suggest
ϕ = â · ∇ΦL , ψ = â ∧ ∇ΦT ,
where ΦL and ΦT satisfy similar equations

1
∇2 ΦI + kI2 ΦI = , I = L, T.
4πc2I r
The solutions must be regular at the origin, be radially symmetric

(depend only on r), and contain no incoming waves. Thus,
1 1 1 ikI r
ΦI = − G(kI r), G(kI r) = e , I = L, T. (4.2.2)
ω 2 4πr ω 2 4πr
The appearance of the non-radiating first term might appear unusual,
but recall that the individual potentials are not bound by the same
radiation conditions that govern the Green’s tensor. Substituting back
into the Helmholtz decomposition (1.1.19) for u, we find that the 1/r
terms exactly cancel a delta function singularity arising from the second
G
derivatives of the G functions. The end result is ui = Uik âk , where UG
is a second-order tensor, the Green’s tensor (Harris, 2001):

UG = ω −2 kT2 G(kT r)I + ∇∇ G(kT r) − G(kL r) . (4.2.3)
The identity a2 · u1 (x2 ) = a1 · u2 (x1 ), derived in §4.1 on the basis of

the general reciprocity identity (4.1.3), is reflected in the symmetry of
the Green’s tensor.
4.2.1 Point impedance

The Green’s tensor is singular at the point of application of the point
force. An expansion of (4.2.3) about r = 0 indicates that the leading
term is of order 1/r, and it is identical to the static Green’s tensor
UG0 = UG |ω=0 . Proceeding to the next term in the expansion of (4.2.3)
yields

iω 1 2
UG (x) = UG0 (x) + + I + O(r).
12π c3L c3T
The particle velocity near the point of application is also singular. How-
ever, the time average of the power dissipated by the force is finite. Let
us assume ρf = δ(x)ReF0 e−iωt ; then the average power radiated per
period is equal to the power expended by the force
2π/ω

ω
P = lim ReF0 e−iωt · v(x, t)dt,
x→0 2π
0

where v(x, t) = Re − iωu(x)e−iωt is the particle velocity. Hence,
F0 F0∗
P = , (4.2.4)
2Zr
where Zr is the radiation impedance,

−1 ω2 1 2
Zr = + 3 . (4.2.5)
12πρ c3L cT
The inverse, defined by Ar = Zr−1 , is called the drive point admittance.
These reflect the two-fold interpretation of Zr , or its inverse Ar , as defin-
ing the radiation loss at infinity, or the work done by the applied force
at the point of application. These measures of power must be equivalent
in the absence of material damping.
4.2.2 Complex source: Gaussian beams

Let us consider the case of a point force applied at a complex source
point: f = F0 δ(x−ibê3 ), b > 0. In order to understand the Green’s tensor
for this seemingly unusual forcing, we note that the length s = |x − ibê3 |
must be understood as a complex number:
1/2 2 1/2
s = R2 + (x3 − ib)2 = r − b2 − 2ibx3 ,
where R2 = x21 + x22 is the cylindrical radius about the axis in the x3
direction. We take the branch cut of the square-root function along the
negative real axis, and note that the physical coordinates corresponding
to this are the disc of radius b: R ≤ b, x3 = 0. We may associate
each face of the disc with the√ different branches of the square root,
thus: s(R ≤ b, x3 = ±0) = ∓i b2 − R2 . The Green’s tensor is therefore
discontinuous across the disc. The effect of the complexification of the
source point is to introduce a finite region of discontinuity, the size of
which grows as the source is moved further into complex space.
The far-field behavior of the beam follows from the asymptotic form
of the complex length s = r −ib cos θ+O(r−1 ), where cos θ = x3 /r. Thus,
F0 −2 ikL r kL b cos θ
u= · cL e e x̂ ⊗ x̂ + c−2
T e
ikT r kT b cos θ
e (I − x̂ ⊗ x̂)
4πr
+ O(r−2 ), (4.2.6)
where ⊗ signifies the tensor product. The amplitude in the forward direc-
tion θ = 0 is exponentially larger than in the opposite direction. Let us
examine the behavior of the beams near the forward symmetry axis.
Using cos θ = 1 − θ2 /2 + · · · with θ ≈ sin θ = R/r implies
R2 R2
F0 ikL r kL b 1− 2r2 −2 ikT r kT b 1− 2r2
u≈ · c−2 e e x̂ ⊗ x̂ + c e e (I − x̂ ⊗ x̂) .
4πr L T
The Gaussian decay away from the central axis makes this type of
solution useful in modeling piston-like sources with a suitable summa-
tion of beams, e.g. by taking a finite set of sources with different values
of F0 and b, say {Fi , bi }, i = 1, 2, . . . , n. The total power then follows
from (4.2.6) as

ω2
n
1 2
P = Fi · F∗i 3 I0 (kL bi ) + 3 I0 (kT bi ) ,
24πρ i=1 cL cT
where I0 is the modified Bessel function of order zero. This reduces to

the point-force result of (4.2.4) as b → 0, by virtue of I0 (0) = 1.
Wen and Breazeale (1988) describe a procedure for choosing a suitable
basis set of Gaussian beam parameters for a piston-like transducer in an
acoustic fluid. The same basis set was used by Spies for transducers in
elastically isotropic (Spies, 1994) and anisotropic (Spies, 1999) materials.
The analytic nature of the complex point-source Green’s tensor make it
suited for examining Gaussian beam-like interaction with surfaces and
interfaces, and scattering from objects such as spheres.
4.2.3 Impedance of a particle

The notions developed above, specifically reciprocity and Green’s ten-
sors, are most powerful when applied to radiation and scattering from
finite regions. We consider in detail the example of a spherical particle
with a rigid surface that is free to move. This does not imply that the
particle is itself rigid, only that its bounding surface moves as a rigid
body. In fact we will specifically consider later the case of a particle
of finite mass or, more generally, of given impedance. We first con-
sider the radiation caused by imposed motion of the particle, and at
the same time calculate the forcing required. This replaces the notion
of a point force with the realistic model of a forced particle, and its
radiation.
Let us assume the particle has radius a and moves rigidly so that
its center oscillates according to u = u0 with, as usual, the e−iωt term
omitted and understood. We are interested in finding the displacement
in the exterior region r > a in an infinitely extended solid. The force F
required to maintain the motion must be in the same direction. This is
the resultant of the tractions over the surface r = a.
The solution was originally found by Oestreicher (1951), and we
follow his general approach here. Let us assume the displacement is
ui (x) = Uij (x)u0j , where, based on the form of the Green’s tensor (4.2.3),
we take

U = −A∇∇G(kL r) + B kT2 I + ∇∇ G(kT r). (4.2.7)
This only depends on two constants, A and B, which are found by
requiring that the displacement equals u0 everywhere on the surface
r = a. Note that G(kI r) = (ikI /4π)h0 (kI r), where h0 is the spheri-
cal Hankel function of the first kind. The first three spherical Hankel
functions are

1 iz i+z 2 z + 3iz − 3
h0 (z) = e , h1 (z) = − eiz = −h0 (z), h2 (z) = i eiz .
iz z2 z3
Using these we may express the displacement as

3

ikL h1 (kL r) 0 1 0
u =A u − h2 (kL r) (u · x)x̂
4π kr r

ikT3 0 0 3 0
+B 2h0 (kT r)u − h2 (kT r) u − (u · x)x̂ . (4.2.8)
12π r
Two coupled equations are obtained by requiring u = u0 on r = a, then
equating the u0 component on either side of (4.2.8) and setting the x̂
component to zero. After some algebra, we can express the solutions as
(Norris, 2008)

1 Z + ZM 1 Z + ZM
A= 2 , B= 2 ,
kL G(kL a) ZL + ZM kT G(kT a) ZT + ZM
(4.2.9)
where
4
ZM = iω πa3 ρ,
3
ZL = (iω)−1 4πa(λ + 2μ) (1 − ikL a),
ZT = (iω)−1 4πaμ(1 − ikT a),
and Z satisfies
3 1 2
= + . (4.2.10)
Z + ZM ZL + ZM ZT + ZM
This relation between the impedances is similar to the effective
impedance of a lumped parameter system. The right member of (4.2.10)
is equivalent to (ZL in parallel with ZM ) in series with (ZT in parallel
with ZM ) twice.
The impedance parameter Z relates the displacement to the force
acting on the sphere, according to
F = −iωZu0 . (4.2.11)
This can be verified by first noting that the traction on a spherical

surface, ti = τij x̂j , can in general be written

t = x̂λ(∇ · u) + (μ/r) ∇(x · u) + r∂r u − u .
The resultant of the traction is obviously in the same direction as the

imposed oscillation. Therefore, integrating the inner product t · u0 over
r = a gives, after considerable simplification, −iωZ. The identity (4.2.11)
then follows.
The non-dimensional tensor or matrix U(x) relates the time-harmonic
displacement of the particle of radius a to the displacement in the elas-
tic material. It should reduce to the Green’s tensor if the particle is
small or, what is essentially the same limit, when the frequency is small.
The proper scaling is in fact kT a 1, meaning that the particle is
sub-wavelength and acts effectively like a point. Taking the limit of a
vanishingly small particle yields
U(x) → iωZρ−1 UG (x), a → 0.
0.8
0.6
A
0.4
0.2
0
0 1 2 3 4 5
kLa
Fig. 4.1. The relative admittance A is the ratio of the real part of −1/Z to
Ar of (4.2.5). The quantity Z is the impedance for a rigid spherical inclusion
of radius a. The figure illustrates that the radiation loss from a given time-
harmonic force applied to the rigid sphere is a decreasing function of a.
In the same limit, the impedance Z simplifies:

2
1 iω 1 2 ω2 1 2 iω 3 a 1 2
= + 2 − + 3 − + +O(a2 ).
Z 12πρa c2L cT 12πρ c3L cT 36πρ c2L c2T
In this form the impedance acts as three lumped elements in series:
a stiffness, a damper, and a mass, respectively. Recall that lumped
parameter elements in series combine as (Z1−1 + · · · + Zn−1 )−1 . The mid-
dle term in (4.2.12), associated with the damper, reproduces the point
admittance result of (4.2.4) and (4.2.5); see Fig. 4.1.
4.3 Reciprocity revisited

We may apply the general reciprocity formulation to a rigid particle,
i.e. one with a bounding surface S that moves as a rigid body. In this
particular case the fields u1 and u2 are constants for the surface inte-
grals in (4.1.3). We again take field 1 as the solution for the inclusion
undergoing arbitrary rigid-body displacement, with f 1 = 0. Let field 2 be
the solution for a point force at x in the exterior: ρf 2 (y) = F δ(y − x).
The solution u2 is in fact the Green’s function in the presence of the
movable inclusion. We can, however, avoid explicit calculation of the
Green’s function. The displacement u2S on the inclusion surface must
be another rigid-body oscillation, and therefore the reciprocity identity
becomes
u1S · τ 2 dS − F · u1 (x) = u2S · τ 1 dS. (4.3.1)
S S
The integral involving τ 1 gives the resultant force −iωZu1S . For field
2, let Fp denote the resultant caused by the point force at x,

τ 2 dS = Fp .
S
The displacement at s for field 1 follows from the definition of the tensor
U as u1 (x) = U(x)u1S . Using the fact that the rigid-body displacement
u1S is arbitrary, we deduce that
Fp = U(x)F − iωZu2S . (4.3.2)
This contains two unknowns: the force and the displacement on the
sphere. These are physically related by the nature of the material in
the sphere. A wide class of particles is covered by assuming a general
impedance ZP of the particle. This could be, at the simplest level, a
mass-like impedance ZP = iωm, where m is the particle mass. More
4.4 Force on a particle from an elastic wave 79
generally, the particle impedance could include or model internal struc-

ture in the particle. Whatever the precise form, the particle undergoes
rigid oscillation, and the equilibrium condition (Newton’s second law) is
Fp = iωZP u2S . (4.3.3)
Eliminating the linear displacement between (4.3.2) and (4.3.3) gives
Fp = ZP (Z + ZP )−1 U(x)F. (4.3.4)
This is the desired result: it provides an expression for the force acting
on the particle, caused by the point force in the elastic solid. Note that
it involves the fundamental matrix U that was previously interpreted
as a generalized Green’s tensor for a force applied to the particle of
size a. Equation (4.3.4) is a complementary result, giving the force on
the finite-sized particle, caused by a point force elsewhere.
4.4 Force on a particle from an elastic wave

The force on a particle due a remote point load is given directly by the
tensor ZP (Z + ZP )−1 U of (4.3.4). Taking the source point to infinity,
the effect on the particle is equivalent to a combination of two incident
longitudinal and transverse plane waves. The far-field form of the tensor
follows from (4.2.7) and (4.2.9) as

ZP Z + ZM a eikL (r−a)
U(x) = ZP x̂ ⊗ x̂
Z + ZP Z + ZP r ZL + ZM

eikT (r−a)
+ I − x̂ ⊗ x̂ + O(r−2 ). (4.4.1)
Z T + ZM
Consider, for instance, a unit point force in the far field at x in the
direction n = −x̂ . This produces a longitudinal plane wave at the origin
of the form u = u0 neikL n·x , where u0 = eikL r /(4πμκ2 r). The force
on the spherical particle due to an incident longitudinal plane wave
u(x) = eikL n·x u0 is therefore

4πaZP Z + ZM
Fp = (λ + 2μ) e−ikL a u0 . (4.4.2)
Z + ZP ZL + ZM
In the same manner, the force on the spherical particle due to an incident
transverse plane wave u(x) = eikT n·x u0 , with u0 · n = 0, is

4πaZP Z + ZM
Fp = μ e−ikT a u0 . (4.4.3)
Z + ZP ZT + ZM
A rigid, immovable particle is obtained in the limit as ZP → ∞. The

values of the plane-wave-induced forces for the rigid, immovable par-
ticle coincide as the frequency tends to zero, the static limit. Despite
the relatively simple formulae for the force on the particle from wave
incidence, these expressions would be difficult to obtain without the
use of arguments based on the reciprocity identity. Related results and
other applications of these ideas to particles in elastic solids are given in
Norris (2008).
5
Integral equations for crack scattering
This chapter is concerned with cracks. Real cracks in solids are

complicated: they are thin cavities, their two faces may touch, and the
faces may be rough. We consider ideal cracks. By definition, such a crack
is modeled by a smooth open surface Ω (such as a disc or a spherical
cap); the elastic displacement is discontinuous across Ω, and the traction
vanishes on both sides of Ω (so that the crack is seen as a cavity of zero
volume). We suppose that we have one crack with a smooth edge, ∂Ω,
embedded in an infinite, unbounded, three-dimensional solid. Extensions
to multiple cracks, to cracks in two dimensions, to cracks in halfspaces
or in bounded domains, or to cracks with less smoothness may be made,
with varying degrees of difficulty. For a variety of applications, see the
book by Zhang and Gross (1998).
After formulating our scattering problem, we give the governing
hypersingular integral equation in §5.2. This equation is solved approx-
imately for long waves (low-frequency scattering) in §5.3. The approach
used is elevated to a well-known ‘strategy’ in §5.4 prior to further
applications. For flat cracks and screens, we can simplify the govern-
ing hypersingular integral equation. This is done in §5.5. Alternatively,
we can use a direct approach, using Fourier transforms; see §5.6. Meth-
ods for solving the resulting equations are discussed in §5.7. The final
section is concerned with curved cracks and screens. Results for cracks
that are almost flat are described.
5.1 Formulation
To keep the analysis relatively simple, we shall focus on analogous scalar
problems coming from acoustics. Thus, we suppose that Ω is a thin screen
in a compressible fluid. The screen is hard (or rigid), which means that
we have a Neumann boundary condition.
81
82 Integral equations for crack scattering
We are interested in scattering time-harmonic waves by the screen.

Much is known about how to calculate scattering from objects of non-
zero volume (Martin, 2006). Except in a few special cases (such as
scattering by a sphere), it is usual to derive and solve (numerically)
a boundary integral equation over the boundary of the scatterer. How-
ever, special methods are needed for zero-volume obstacles such as cracks
and screens. In particular, the Neumann boundary condition means that
it is inevitable that we shall encounter hypersingular boundary integral
equations over the screen. These equations can be tackled directly (using
boundary elements, perhaps), or they may be recast into other equiva-
lent forms. For example, if the screen is flat, various simplifications can
be made. Integral equations can also be used as the basis for various
approximation schemes.
We consider acoustic scattering by a thin rigid screen Ω surrounded by
a compressible fluid; we model the screen as a smooth simply-connected
bounded surface with a smooth edge ∂Ω. We write the scattered field
as Re{usc e−iωt }, where ω is the circular frequency. Then, usc solves the
Helmholtz equation in three dimensions,
∇2 usc + k 2 usc = 0 in the fluid, (5.1.1)
the Sommerfeld radiation condition at infinity, and the boundary

condition
∂usc ∂uin
+ = 0 on Ω. (5.1.2)
∂n ∂n
In addition usc is required to be bounded everywhere: we do not permit
sources on ∂Ω. Here, k = ω/c, c is the constant speed of sound, uin is
the given incident field, and ∂/∂n denotes normal differentiation. The
total field is u = usc + uin , so that (5.1.2) gives ∂u/∂n = 0 on Ω.
Denote the two sides of Ω by Ω+ and Ω− , and define the unit nor-
mal vector on Ω, n, to point from Ω+ into the fluid. Then, define the
discontinuity in u across Ω by
[u(q)] = lim+ u(Q) − lim− u(Q),

Q→q Q→q
where q ∈ Ω, q ± ∈ Ω± , and Q is a point in the fluid. Notice that

[u] = [usc ] as [uin ] = 0.
The scattered field has the integral representation

1 ∂
usc (P ) = [u(q)] G(P, q) dSq , (5.1.3)
4π Ω ∂nq
5.1 Formulation 83
where
G(P, q) = R−1 exp (ikR) (5.1.4)
is the free-space Green’s function and R is the distance between P
and q ∈ Ω.
To be more explicit, we introduce Cartesian coordinates Oxyz and
suppose that the surface Ω is given by
Ω : z = S(x, y), (x, y) ∈ D,
where D is a region in the xy-plane with edge ∂D. We define a normal
vector to Ω by
N = (−∂S/∂x, −∂S/∂y, 1),
and then n = N/|N| is a unit normal vector.
Also, if uin represents an incident plane wave, then
uin (x, y, z) = eik(xα1 +yα2 +zα3 ) , (5.1.5)
where α12 + α22 + α32 = 1.
Suppose that P has position vector r = (x0 , y0 , z0 ) and q ∈ Ω has
position vector q = (x, y, S(x, y)). Let
[u(q)] = w(x, y).
Then, we find that (5.1.3) becomes, exactly,

1 dA
usc (x0 , y0 , z0 ) = w(x, y) (N(q) · R2 )(1 − ikR2 ) eikR2 3 ,
4π D R2
where dA = dx dy, R2 = r − q, and R2 = |R2 |.
In the far field, we have
usc (P ) ∼ r−1 eikr f (r̂) as r → ∞,
where r = |r|, r̂ = r/r and

k
f (r̂) = [u(q)] {r̂ · n(q)} exp (−ikq · r̂) dSq (5.1.6)
4πi Ω

k
= w(x, y) {r̂ · N(q)} exp (−ikq · r̂) dA; (5.1.7)
4πi D
f is the far-field pattern.
The formula (5.1.7) is exact. Although the integration is over a flat
region, the geometry of the screen enters through w, N, and q. Thus,
we can expect that reasonable approximations to w will generate good
approximations to f .
For example, in some applications, a static approximation to w (or [u])

could be used; this idea, leading to low-frequency approximations, will
be developed in §5.3.
For another example, we might be able to use high-frequency
approximations for [u]. The most popular of these is the Kirchhoff
approximation, [u] uK , where uK (q) is the total field at q when the inci-
dent field is reflected by an infinite flat plane perpendicular to n(q). This
approximation is used widely in models of ultrasonic non-destructive
evaluation; see, for example, Schmerr and Song (2007, chapter 10).
5.2 A hypersingular integral equation

Application of the boundary condition (5.1.2) to (5.1.3) gives

1 ∂2 ∂uin
× [u(q)] G(p, q) dSq = − , p ∈ Ω, (5.2.1)
4π Ω ∂np ∂nq ∂np
where the integral must be interpreted in the finite-part sense. Equa-
tion (5.2.1) is the governing hypersingular integral equation for [u]; it is
to be solved subject to the edge condition
[u(q)] = 0 for all q ∈ ∂Ω. (5.2.2)
Equation (5.2.1) could be solved numerically, using boundary elements

combined with regularization techniques. Indeed, this is probably the
only option if Ω has a complicated, three-dimensional shape. As a sam-
ple, we cite a paper by Tada et al. (2000), in which a formulation for
transient elastodynamics and non-planar cracks is developed. The lit-
erature on methods for solving hypersingular equations numerically is
extensive.
The right-hand side of (5.2.1) reduces to −ikuin α̂ · n(p) when uin is
given by (5.1.5); here,
α̂ = (α1 , α2 , α3 )
is a unit vector giving the direction in which the incident plane wave is
propagating.
What does ‘hypersingular’ mean? This question can be answered pre-
cisely, using the notion of pseudodifferential operators acting between
function spaces. However, for many purposes, it is enough to gain
intuition through simple examples. Suppose we have an expression Lf ,
where L is a linear operator and f is a function. If L is an integral
operator with a weakly singular kernel, Lf will be smoother than f : we
5.2 A hypersingular integral equation 85
usually think of integration as a smoothing process. If L is an integral

operator defined by a Cauchy principal-value integral (a singular inte-
gral operator), Lf will have the same smoothness as f : in some sense,
L is similar to the identity operator, I. Hypersingular operators coarsen.
If L is such an operator, Lf will have less smoothness than f : in some
sense, L is similar to a differential operator, and this operator will be of
first order in our applications.
We usually identify hypersingular operators in one of two ways. One
is as in (5.2.1): the integral operator is defined in terms of a finite-part
integral (and does not exist as an improper integral or as a Cauchy
principal-value integral). The second way involves Fourier transforms.
Roughly, we can write (locally) Lf = F −1 {σF{f }}, where σ is called
the symbol and F denotes Fourier transformation. Then, if σ is a lin-
ear function of the transform variable(s), L is a first-order differential
operator; if σ is linear for large values of the transform variable(s), then
L will be identified as one of our hypersingular operators – we will see
examples in §5.6.
Returning to (5.2.1), let us project onto D (as in §5.1), giving

1
× K(x0 , y0 ; x, y) w(x, y) dA = b(x0 , y0 ), (x0 , y0 ) ∈ D, (5.2.3)
4π D
where
K = R1−3 (1 − ikR1 ) eikR1 {N(p) · N(q)}

− R1−5 (3 − 3ikR1 − k 2 R12 ) eikR1 (N(p) · R1 )(N(q) · R1 ),
R1 = (x − x0 , y − y0 , S(x, y) − S(x0 , y0 )), R1 = |R1 |, and
b(x, y) = −∂uin /∂N = −ikN · α̂ exp (ikq · α̂) (5.2.4)
when uin is given by (5.1.5). Notice that K(x0 , y0 ; x, y) = K(x, y;

x0 , y0 ).
Equation (5.2.3) is to be solved subject to the edge condition
w(x, y) = 0 for all points (x, y) on ∂D. (5.2.5)
Let
S1 = ∂S/∂x and S2 = ∂S/∂y, evaluated at (x, y), (5.2.6)
with S10 and S20 being the corresponding quantities at (x0 , y0 ). Then
N(q) = (−S1 , −S2 , 1) and N(p) = (−S10 , −S20 , 1). Let
R = {(x − x0 )2 + (y − y0 )2 }1/2 (5.2.7)

and Λ = {S(x, y) − S(x0 , y0 )}/R. Also, define the angle Θ by

x − x0 = R cos Θ and y − y0 = R sin Θ,
whence R1 = R(cos Θ, sin Θ, Λ). Hence

eikRX 1 − ikRX
K= (1 + S1 S10 + S2 S20 )
R3 X3

Y 2
− 5 (3 − 3ikRX − (kRX) ) , (5.2.8)
X
√
where X = 1 + Λ2 and
Y = (S1 cos Θ + S2 sin Θ − Λ)(S10 cos Θ + S20 sin Θ − Λ).
This formula for K is exact. If we expand K for small R about p, we
find that
K ∼ R−3 σ(p; Θ), (5.2.9)
where
1 + (S10 )2 + (S20 )2
σ(p; Θ) = .
1 + (S10 cos Θ + S20 sin Θ)2
In particular, σ ≡ 1 when S is constant. Equation (5.2.9) exhibits
the strong singularity in the kernel K, and is typical of hypersingular
operators defined over surfaces.
5.3 Low-frequency scattering

Before the advent of computers, it was traditional to obtain approximate
solutions of scattering problems by assuming that the frequency is low,
so that the crack is assumed to be small compared to the wavelength of
the incident field. These approximations are still useful today. In three
dimensions, it is known that the scattered field is an analytic function
of k: it has a Maclaurin expansion with respect to k. Thus, associated
static problems (where k = 0) feature.
The basic static problem is as follows. Let φa denote the velocity
potential of a given ambient flow. Then, we seek another potential φ,
where φ is a bounded solution of ∇2 φ = 0 in the fluid, with
∂φ ∂φa
+ =0 on Ω
∂n ∂n
and φ = O(r−1 ) as r → ∞. For a uniform ambient flow, we have
φa (x, y, z) = xα1 + yα2 + zα3 . (5.3.1)
5.3 Low-frequency scattering 87
In general, we can find φ by solving a hypersingular integral equation

analogous to (5.2.1), namely

1 ∂2 ∂φa
× [φ(q)] G0 (p, q) dSq = − , p ∈ Ω, (5.3.2)
with [φ(q)] = 0 for all q ∈ ∂Ω. Here, G0 = R−1 is the static free-space
Green’s function; see (5.1.4). The right-hand side of (5.3.2) reduces to
−α̂ · n(p) when φa is given by (5.3.1).
Returning to (5.2.1), we seek solutions in powers of k. We can write
G = G0 + kG1 + · · · and uin = u0in + ku1in + · · · ,
and then (5.2.1) implies that [u] has a similar expansion,
[u] = u0 + ku1 + · · · .
Substituting and collecting up powers of k, we obtain a sequence of
equations from which uj can be determined. The first two are
L0 u0 = b0 and L0 u1 = b1 ,
where

1 ∂2 ∂u0
L0 u = × u(q) G0 (p, q) dSq , b0 (p) = − in ,
1 2
∂u 1 ∂
b1 (p) = − in − × u0 (q) G1 (p, q) dSq ,
∂np 4π Ω ∂np ∂nq
and uj = 0 on ∂Ω for j = 0, 1, 2, . . .. Thus, each uj is obtained by solving
a certain static problem. The problem itself could be solved by any
convenient method, not necessarily via the equation L0 uj = bj .
For an incident plane wave, (5.1.5) gives
uin (x, y, z) = 1 + ik(xα1 + yα2 + zα3 ) + · · · .
Thus, u0in ≡ 1 and b0 ≡ 0. As the equation L0 uj = bj is uniquely solvable
(subject to uj = 0 on ∂Ω), we obtain u0 ≡ 0. Then, the equation for u1
reduces to
L0 u1 = −iα̂ · n(p).
Hence, comparison with (5.3.2) gives u1 = i[φ] and
[u] ik[φ] for small k.
This approximation can then be inserted in (5.1.6) to give a low-
frequency approximation for the far-field pattern.
5.4 Some strategies

In general, we encounter equations, such as (5.2.3), that we can write in
operator form as
Hw = b, (5.4.1)
where H is a linear operator, b is known, and w is to be found. We write
(5.4.1) as
(L0 + L1 )w = b, with L1 = H − L0 , (5.4.2)
where L0 is another operator. If L0 is invertible, we obtain
(I + M)w = g, with M = L−1 −1
0 L1 and g = L0 b. (5.4.3)
Several strategies for solving (5.4.1) follow this general pattern. To be
effective, L−1
0 should be available explicitly or it should be easier to
compute than H−1 .
For example, if L0 is H evaluated at k = 0, we obtain a method
with two virtues. First, the operator M will not be hypersingular: the
operator H has been regularized. Second, we have access to analytical
approximations for low-frequency scattering, as discussed in §5.3. This
is because M is small in some norm, so that (I +M)w = g can be solved
by iteration.
For another example, if L0 is H evaluated for a simpler geometry, we
obtain a method for perturbed screens (such as wrinkled discs).
We shall see examples of these strategies later.
5.5 Flat cracks as a special case

Almost all the literature on scattering by three-dimensional screens (and
cracks) assumes that the screen is flat. Thus, we assume that S(x, y) ≡ 0,
whence D ≡ Ω. From (5.2.3) and (5.2.8), the governing hypersingular
integral equation reduces to

1 eikR
× (1 − ikR) 3 [u(x, y)] dA = b(x0 , y0 ), (x0 , y0 ) ∈ D, (5.5.1)
4π D R
where R is defined by (5.2.7) and
b(x, y) = −∂uin /∂z evaluated on z = 0. (5.5.2)
Noticing that

1 ∂2 ∂2 eikR ∂2 ∂2 eikR
(1 − ikR) eikR = + + k2 = + + k2 ,
R3 ∂x2 ∂y 2 R 2
∂x0 ∂y02 R
5.6 Flat cracks: direct approach 89
we can rewrite (5.5.1) as

2
∂ ∂2 2 eikR
2 + 2 + k [u(x, y)] dA = 4πb(x0 , y0 ), (x0 , y0 ) ∈ D.
∂x0 ∂y0 D R
(5.5.3)
This can be regarded as a kind of regularization, because the finite-
part integral has gone, although it has been replaced by a differential
operator.
For an incident plane wave, (5.5.2) gives
2
ik(xα1 +yα2 ) 1 ∂ ∂2
b(x, y) = −ikα3 e = + 2 + k eik(xα1 +yα2 )
2
ikα3 ∂x2 ∂y
(for α3 = 0). Then, (5.5.3) can be integrated to give

eikR 4π ik(x0 α1 +y0 α2 )
[u(x, y)] dA = e + Ψ0 (x0 , y0 ), (x0 , y0 ) ∈ D,
D R ikα3
(5.5.4)
where Ψ0 (x, y) solves
2
∂ ∂2 2
+ + k Ψ0 (x, y) = 0, (x, y) ∈ D.
∂x2 ∂y 2
Denote the left-hand side of (5.5.4) by S[u]; S is a single-layer opera-
tor. The equation Sw = g arises when the analogous scattering problem
for a sound-soft screen (Dirichlet condition) is solved. It is known that,
in general, the solution of Sw = g is infinite around ∂D, whereas we
want w to satisfy the edge condition (5.2.5). This condition can only be
satisfied by making an appropriate choice for Ψ0 ; it is not clear how to
make this choice in practice.
It is known that S smooths by one order. Thus, the operator on the
left-hand side of (5.5.3) coarsens by one order.
5.6 Flat cracks: direct approach

Here, we assume from the outset that the crack or screen is flat and lying
in the xy-plane. To proceed, we use two-dimensional Fourier transforms.
Thus, define
∞ ∞
U (ξ, η, z) = F{usc } = usc (x, y, z) ei(ξx+ηy) dx dy
−∞ −∞
with inverse
∞ ∞
1
usc (x, y, z) = F −1 {U } = U (ξ, η, z) e−i(ξx+ηy) dξ dη.
(2π)2 −∞ −∞
Transforming (5.1.1) gives (k 2 − ξ 2 − η 2 + ∂ 2 /∂z 2 )U = 0 with solution
U (ξ, η, z) = A± (ξ, η) e±γz for ±z > 0.
Here, A+ and A− are arbitrary functions, and γ is defined as follows:

√ 2 − k2 ,
s√ s > k,
γ= with s = ξ2 + η2 .
−i k 2 − s2 , 0 ≤ s < k,
This definition ensures that the radiation condition is satisfied.

As ∂usc /∂z is continuous across z = 0 for all (x, y), we infer that
A+ = − A− . This implies that usc (x, y, z) is an odd function of z,
usc (x, y, −z) = −usc (x, y, z), so we can now assume that z ≥ 0 and
write
∞ ∞
1
usc (x, y, z) = A(ξ, η) e−γz e−i(ξx+ηy) dξ dη, z > 0.
(2π)2 −∞ −∞
(5.6.1)
Let us identify A in terms of [u]. From (5.6.1), we have
∞ ∞
2
[u(x, y)] = A(ξ, η) e−i(ξx+ηy) dξ dη = 2F −1 {A},
(2π)2 −∞ −∞
so that 2A = F{[u]}. Explicitly, as [u(x, y)] = 0 for (x, y) ∈ D, we have

2A(ξ, η) = F{[u]} = [u(x, y)] ei(ξx+ηy) dx dy. (5.6.2)
D
Then, application of the boundary condition (5.1.2) on z = 0+ yields

∞ ∞
∂uin 1
= γA(ξ, η) e−i(ξx+ηy) dξ dη, (x, y) ∈ D,
∂z (2π)2 −∞ −∞
or, more concisely,
− 12 F −1 {γF{[u]}} = b(x, y), (x, y) ∈ D, (5.6.3)
where b is defined by (5.5.2). This is another equation for [u]; it should be

compared with (5.5.1). Once solved, usc is given by (5.6.1) with (5.6.2).
Equation (5.6.3) can be regarded as a hypersingular equation. This
can be seen by noticing that γ ∼ s as s → ∞, so that the right-hand
side of (5.6.3) is similar to a first derivative of [u]. For an extensive review
of the use of (5.6.3) for scattering computations, see Boström (2003).
5.7 Flat cracks: how to compute [u] 91
Write γ = s + β(s), where β = γ − s. Then, we can write (5.6.3) as

(5.4.2), with
L0 w = − 12 F −1 {sF{w}}, L1 w = − 12 F −1 {βF{w}},
and w = [u]. The operator L1 is similar to a single-layer operator: its
symbol β ∼ − 21 k 2 /s as s → ∞. The operator L0 is hypersingular but
it does not depend on k: it is the corresponding static operator. If D
has a simple shape (such as a circular disc), L0 has an explicit inverse
(subject to the edge condition on ∂D) and this can be used in order to
obtain the regularized equation (5.4.3).
5.7 Flat cracks: how to compute [u]

For a flat screen D, we found two equations for [u], the discontinuity in
u across D, namely (5.5.1) and (5.6.3); write these formally as H[u] = b.
A familiar way to solve such equations is to expand [u] with a set of
basis functions, writing

[u(x, y)] = an un (x, y);
n
evidently, the functions un have to be selected and then the coefficients

an have to be found. Substitution in H[u] = b gives

an (Hun )(x, y) = b(x, y), (x, y) ∈ D,
n
and then various methods (such as collocation) immediately suggest

themselves for the numerical determination of an .
For the expansion functions, un , one option is radial basis functions.
Thus, choose N points (xn , yn ) ∈ D, n = 1, 2, . . . , N , put

un (x, y) = χ (x − xn )2 + (y − yn )2 ,
2 2
and then choose the function χ. Examples are χ(r) = e−r /c (Visscher,
1983) and χ(r) = (1 − r2 /c2 )α H(c − r) (Glushkov and Glushkova, 1996),
where c and α are positive constants and H is the Heaviside unit func-
tion. Other choices could be made but, to be effective, one should be
able to compute Hun efficiently if not analytically.
One virtue of radial basis functions is that they provide flexibility:
the shape of D is relatively unimportant. On the other hand, we know
that [u] = 0 around the edge ∂D; in fact, we know that [u] must vanish
as the square root of the distance from ∂D. This knowledge could be
incorporated by using special approximations near ∂D. However, if D is

simple in shape, such as circular, elliptical, or rectangular, it is possible
to construct functions un with the correct square-root edge behaviour.
As an example, we consider a circular screen D of radius a. For this
geometry, we have functions un satisfying L0 un = bn , where L0 is H at
k = 0 and bn is known explicitly.
Introduce plane polar coordinates (r, θ) so that the crack occupies
0 ≤ r < a. For simplicity, suppose that un (x, y) is an even function of
x = r cos θ and put
un (x, y) = awn (r/a) H(a − r) cos nθ.
Then, a standard calculation (using the Jacobi expansion (Martin, 2006,
p. 37) twice) gives
∞ 1
−1 3
F {γF{un }} = cos nθ a γJn (rs) wn (ρ) Jn (asρ) ρ dρ s ds,
0 0
where Jn is a Bessel function. This formula simplifies if we expand wn (ρ)

n
in a series of functions wm (ρ), defined by
n+1/2

n
wm (r) = rn C2m+1 ( 1 − r2 ),
λ n
where
√ Cm is a Gegenbauer polynomial. Each function wm (r) is equal to
2
1√− r multiplied by a polynomial in r; in particular, w0n (r) = (2n + 1)
rn 1 − r2 . Hence (Krenk, 1979),
1
n 2 Γ(n + m + 3/2) jn+2m+1 (as)
wm (r) Jn (asr) r dr = ,
0 Γ(n + 1/2) m! as
where jn is a spherical Bessel function. Then, writing

[u(x, y)] = aH(a − r) Wm n n
wm (r/a) cos nθ, (5.7.1)
n=0 m=0
we find that

n n
(H[u])(r, θ) = Wm tm (r/a) cos nθ, r < a,
n=0 m=0
where
∞
Γ(n + m + 3/2)
tnm (r/a) =− a2 γJn (rs) jn+2m+1 (as) ds.
Γ(n + 1/2) m! 0
The remaining integral must be evaluated numerically. However, in the
static limit (replace γ by s), we have (Bell, 1979)
∞ n
Γ(m + 3/2) Γ(n + 1/2) wm (r/a)
a2 sJn (rs) jn+2m+1 (as) ds = ,
0 (n + m)! 1 − r2 /a2
a polynomial in r/a, and this gives the explicit evaluation of L0 [u].

Also, if
n
n wm (r/a)
b(r, θ) = Bm cos nθ,
n=0 m=0 1 − r2 /a2
then the solution of L0 [u] = b is given by (5.7.1) with
n
(n + m)! Bm
n
Wm =− ;
Γ(n + m + 3/2) Γ(m + 3/2)
n
the coefficients Bm can be found using the orthogonality of the
Gegenbauer polynomials, which gives
1
r dr
wln (r) wm
n
(r) √ = hnm δlm ,
0 1 − r2
where δij is the Kronecker delta and hnm is a known constant. Thus, in
effect, L−1
0 is available, and it can be used as in §5.4.
If D is flat but not circular, a possible strategy is the following. Find
a conformal mapping that maps the interior of D onto the interior of a
disc. Use this mapping in the integral equation (5.5.1); as it is a con-
formal mapping, it does not change the basic hypersingularity in the
kernel, and so the dominant operator is the static operator L0 for the
disc. The operator L1 will include some effects due to the mapping and
some due to the dynamics. The details have not been worked out, except
for static problems (Martin, 1996).
5.8 Curved cracks

If the crack or screen is not flat, one may have to resort to solving (numer-
ically) the hypersingular integral equation over the screen Ω, (5.2.1), or
the version of this equation obtained by projection onto the flat region D,
(5.2.3). Obviously, care must be taken in handling the hypersingularity
and with the edge condition, (5.2.2) or (5.2.5).
The surface Ω is defined by z = S(x, y) for (x, y) ∈ D. For non-
constant S, the singularity in the kernel of the integral equation (5.2.3)
is essentially different from that occurring in the integral equation for
constant S; this is revealed by the presence of σ in (5.2.9). This difference
means that the equation over D, (5.2.3), cannot be regularized using
known results for flat screens. However, we can make analytical progress
when Ω is almost flat.
Suppose that
S(x, y) = εf (x, y),
where ε is a small dimensionless parameter and f is independent of ε.

Setting
Λ = ελ, with λ = {f (x, y) − f (x0 , y0 )}/R, (5.8.1)
we find from (5.2.8) that
K = R−3 eikR {1 − ikR + ε2 K2 + O(ε4 )} as ε → 0,
where
K2 = (1 − ikR)(f1 f10 + f2 f20 − 32 λ2 ) + 12 λ2 (kR)2
− 3(1 − ikR − 13 (kR)2 )(f1 cos Θ + f2 sin Θ − λ)
(f10 cos Θ + f20 sin Θ − λ)
and fj , fj0 are defined similarly to Sj ; see (5.2.6).
We expand b similarly. For an incident plane wave, (5.2.4) gives
b(x, y) = ik{b0 (x, y) + εb1 (x, y) + · · · },
where, for example, b0 (x, y) = −α3 eik(xα1 +yα2 ) .
Then, if we expand w in (5.2.3) as
w(x, y) = ik(w0 + εw1 + ε2 w2 + · · · ),
we find that
Hk w0 = b0 , Hk w1 = b1 , and Hk w2 = b2 − K2 w0 ,
where

1 eikR
(Hk w)(x0 , y0 ) = × (1 − ikR) 3 w(x, y) dA
4π D R
is the basic hypersingular operator for acoustic scattering by a flat sound-
hard screen D (see (5.5.1)) and

1 eikR
(K2 w)(x0 , y0 ) = × K2 (x, y; x0 , y0 ) 3 w(x, y) dA.
4π D R
Thus, we have a sequence of hypersingular integral equations, Hk wn =
fn , to solve.
When k = 0, w0 , w1 , and w2 have been found explicitly for par-
ticular geometries, namely inclined elliptical discs and spherical caps
(Martin, 1998). The results obtained agree with known exact results.
Similar results for perturbed penny-shaped cracks have also been
obtained (Martin, 2001).
For k > 0, we can see that w0 is simply the solution for a flat screen;
however, the far-field pattern will be different, it being given by (5.1.7)
with w replaced by w0 . It should be possible to obtain w1 without too

much difficulty, as f1 = b1 is simple. For higher-order terms, one would
have to evaluate K2 w.
If it is assumed further that the incident waves are long compared to
the diameter of the scatterer, 2a, low-frequency approximations may be
made. Then, each wn can be expanded in powers of ka. This approach
has been pursued for a shallow crack in the shape of a spheroidal
cap (Mikhas’kiv and Butrak, 2006).
One difficulty with these approximation methods is that there are
few results to compare them with. For acoustic scattering by spherical
caps, see, for example, the papers by Miles (1971) and Thomas (1963).
Numerical results for elastic-wave scattering by cracks in the shape of
spherical and spheroidal caps have been given by Boström and Olsson
(1987).
6
Scanned acoustic imaging
An angular spectrum representation of a converging spherical wave is

derived, and then combined with a similar representation for a diverging
spherical wave, to construct a model of a focused acoustic beam. This
beam is directed at a fluid–solid interface and the scattered wavefield
calculated, much as was done in §3.1. Representations of the incident
and scattered wavefields are also expressed as Hankel transforms.
A model of the scanned acoustic microscope is next constructed. Using
an electromechanical reciprocity relation, the following imaging equation
is derived:
βa
δV (s, zs ) = 2iD G2 (ξ) R(ξ, s)ei2kzs cos ξ dξ.
0
This expression maps the mechanical reflection coefficient R(ξ, s) to the

measured change in voltage δV (s, zs ). This change depends on the lateral
position of the microscope s, and the relative distance zs between the
geometrical focal plane, and the interface that is being imaged. The
function G(ξ) and angle βa describe how the finite aperture of the lens
limits the acquisition of information. Varying s produces an image, while
varying zs allows the device to be used as an interferometer.
The chapter begins with a description of the acoustic microscope, and
its defining parameters, such as the F and Fresnel numbers. The model of
a converging/diverging focused beam is then presented in detail, followed
by the scattered version of the same in §6.5. Before dealing with the
acoustic signature, some reciprocity identities are necessary; these are
introduced in §6.6 with a sample application. The reciprocity is between
pairs of elastic–electromagnetic fields, which allows determination of the
received voltage response, culminating in the acoustic material signature,
derived in §6.8.
96
6.1 Scanned, reflection acoustic microscope 97
6.1 Scanned, reflection acoustic microscope

Figure 6.1 shows the acoustic microscope that will be modeled in this
chapter. A piezoelectric transducer is placed at one end of a buffer
rod and an acoustic lens is ground into the opposite end, as shown in
Fig. 6.1(a). The transducer, when excited by a tone-burst, converts the
incident electromagnetic wave into an elastic compressional wave that
propagates down the buffer rod to a lens. It is transmitted through the
lens into a coupling fluid. If the overall region being scanned is small, the
fluid is little more than a drop at the end of the buffer rod, as suggested
in the figure, while, if the region being scanned is large, then the whole
device is immersed in the fluid. The lens focuses the sound at or near
the fluid–solid interface. An echo is excited and subsequently collected
by the lens. It propagates back through the buffer rod to the transducer,
where it is converted into an electromagnetic wave. The buffer rod intro-
duces a time delay between the initial excitation and the echo, allowing
the two to be separated. The buffer rod, with its transducer and lens, is
δV
Lens
aperture b
Buffer rod
Reflected
focal plane f
x3 = 2zs
Lens
Interface
x3 = zs
Fluid
Focal
plane
Solid x1
Geometric
focal point βa
x1
x3
x3
(a) (b)
Fig. 6.1. (a) A drawing of the transducer, buffer rod, and lens that together
constitute the acoustic microscope. Note that the origin of the coordinate
system is placed at the geometrical focal point. (b) A drawing of the principal
geometrical features of the region between the plane of the lens aperture and
the geometric focal plane. Note that, in this arrangement, zs < 0.
98 Scanned acoustic imaging
mechanically scanned, in a raster pattern, across the interface. The echo

at each point, after being converted to an electrical signal, is displayed
as a dot on a screen, having a color or level of gray determined by its
strength. The total of all the dots acquired during the scan is the image.
As will become apparent, the microscope can also be used as an inter-
ferometer. As was shown in §2.3, a leaky Rayleigh surface wave is also
excited at the interface. It is collected along with the reflected wave,
and both waves are added together by the transducer. If the geometrical
focal point is placed below the interface, in the solid, at a distance zs ,
as shown in Fig. 6.1(b), then the signal received from the transducer
exhibits interference, between the reflected wave and the leaky Rayleigh
wave, as zs is varied.
If a discontinuity, perpendicular to the surface, is present near the
surface, the leaky Rayleigh wave is reflected and transmitted by it. The
reflected wave also radiates into the fluid and combines with all the other
signals collected by the lens. For simple geometrical configurations, fixing
zs and varying s in a direction perpendicular to the discontinuity causes
the microscope to behave as a standing-wave tube.
The work of this chapter is based on two papers by Rebinsky and
Harris (1992a; 1992b), and a Ph.D. dissertation by Rebinsky (1991).
Many of the figures shown in this chapter are taken from this dis-
sertation. These references also discuss the role of the leaky Rayleigh
wave in enhancing images of surface-breaking cracks, a topic discussed
briefly in §6.8. A somewhat broader overview of mathematically mod-
eling acoustic imaging in solids is provided in the review article by
Harris (1997). The monograph by Briggs (1992) gives a comprehen-
sive overview of acoustic microscopy, at higher frequencies, along with
many interesting photographs. Note that the derivation given in Briggs
(1992, pp. 109–123) of the imaging equation differs from that described
here. The approach of Rebinsky and the writer was initially taken from
Liang et al. (1985).
A note on the word ‘microscope’. Throughout this chapter the imag-

ing device is called an acoustic microscope. This is the common name;
however, it is not the most descriptive one. The device described here
is used in many settings. When used for non-destructive evaluation, the
operating frequency is on the order of 10 to 200 MHz. Useful images are
made, but their resolution is not so great as to merit calling the device a
microscope. It only really behaves as a microscope at frequencies on the
order of 500 MHz or higher. However, at these frequencies the material
6.3 Converging spherical wave 99
is either very attenuating, if it is isotropic, or it is anisotropic; isotropy

cannot be assumed. Except for the case of a surface-breaking crack, only
homogeneous isotropic materials are considered in this chapter; thus, the
frequency range of interest is 10 to 200 MHz.
6.2 Fresnel and F number

The work of Chapter 3 has shown that, at the Fresnel distance, a plane
wave transmitted through an aperture and the waves diffracted by its
edges begin to merge: it is here that a beam is formed. If b is the aperture
radius, and k the wavenumber, the Fresnel distance is kb2 /(2π). Placing a
lens in the aperture converts the transmitted plane wave into a spherical
wave converging to a point, a distance f away; in this book f is the
called the focal length.1 The waves diffracted by the edges continue to be
present. The resulting wavefield depends critically on whether focusing
is made to occur before or after the edge-diffracted waves merge with
the converging spherical one. The Fresnel number N = kb2 /(2πf ) is a
measure of this, and is an important number characterizing an imaging
system. A second one is the F number, where F = f /b. Note that
kb/(2π) = N F . The scanned acoustic microscope considered in this
chapter is such that N 1 and F ≈ 1. This makes the scaled radius of
the aperture, kb, large; this is the large parameter that is used for the
asymptotic calculations.
6.3 Converging spherical wave

A mathematical description of a focused acoustic beam is constructed in
this and the next section. To begin, the idealized case of a spherical wave
converging to a point and then diverging to infinity is first described.
Figure 6.2 indicates the geometry: the x3 axis is the centerline. A spher-
ical wave in x3 < 0 converges to the origin and then, in x3 > 0, diverges
to infinity. In crossing the plane x3 = 0 it is made to undergo a phase
jump of π. It is known that focused beams experience such a phase jump,
though the phase variation occurs more gradually within a finite focal
region (Born and Wolf, 1999, pp. 494–499).
1 The focal length can also be defined as the distance from the center of the face of
the acoustic lens to the focal point; this is the more common definition in acoustic
microscopy.
Converging spherical
wave
ϕ = A e–iπe–ikρ/kρ
Focal plane
x1
ϕ = A eikρ/kρ
Diverging spherical
wave
x3
Fig. 6.2. The arrangement of the coordinate system. The form taken by the
converging and diverging spherical waves on either side of the focal plane, at
x3 = 0, is indicated. Note that there is a phase jump of π across this plane.
Equations (3.1.6) and (3.1.7), with b = 0, describe an outgoing or

diverging spherical wave in x3 > 0. Equation (3.1.6) becomes, with b = 0,

i A 2π π/2−i∞ ik p̂ i ·x
ϕ= e sin ξ dν dξ, (6.3.1)
2π 0 0
where p̂ i is given by (3.1.7) but is repeated here:

p̂ i = sin ξ (cos ν ê1 + sin ν ê2 ) + cos ξ ê3 .
At x3 = 0+ , ϕ reduces to
eikρ
ϕ=A ,
kρ
where x1 = ρ cos θ and x2 = ρ sin θ. Equation (6.3.1) describes a diverg-
ing spherical wave in x3 > 0; the accessible regions of the ξ plane, as
x3 → ∞, are those for which Re(cos ξ) ≥ 0 and Im(cos ξ) ≥ 0. Recall that
k3 = k cos ξ, so that these conditions are equivalent to those in (3.1.5).
To construct the incoming or converging spherical wave, in x3 < 0, ϕ
is written as
∞ ∞
1 ∗ dk1 dk2
ϕ= ϕ(k1 , k2 )ei(k1 x1 +k2 x2 +k3 x3 ) , (6.3.2)
(2π)3 −∞ −∞ k3
where
k3 = (k 2 − k12 − k22 )1/2 , Re(k3 ) ≥ 0, Im(k3 ) ≤ 0. (6.3.3)
Note that the definition of k3 has changed from that given by (3.1.5).
The changed definition ensures that, as x3 → −∞, the exponential term
6.4 Focused acoustic beam 101
decays or acts as an incoming plane wave. The transform ∗ ϕ is deter-

mined by asking that, at x3 = 0− , ϕ be given by the expression (see
Fig. 6.2)
e−ikρ
ϕ = Ae−iπ .
kρ
It is found that
∗ i2πA
ϕ= .
kk3
Again introducing the transformation
k1 = k sin ξ cos ν, k2 = k sin ξ sin ν, k3 = k cos ξ,
the angular spectrum representation of the converging spherical wave is

given as

i A 2π π/2+i∞ ik p̂ i ·x
ϕ= e sin ξ dν dξ. (6.3.4)
2π 0 0
The accessible parts of the ξ plane, as x3 → −∞, are determined by the

definition of k3 , given by (6.3.3), which implies that Re(cos ξ) ≥ 0 and
Im(cos ξ) ≤ 0. In fact the only apparent difference between the diverging
spherical wave, (6.3.1), and the converging one, (6.3.4), arises from the
definition of k3 and the contour of integration for ξ.
6.4 Focused acoustic beam

A focused wave constructed from simultaneously converging and
diverging spherical waves is not physically realizable. First, a spheri-
cally converging wave could not be excited because it would demand
that all the plane waves in its spectrum, including those propagating at
complex angles, be available at some initial aperture. Second, the wave-
field in the focal plane should be continuous and not have a jump in
phase of π. Such a change is permissible in the sense that, between the
source of the converging wave and the far field of the diverging wave,
a net phase change of π may occur, but the change should not cause a
discontinuity at the focal plane. Using the knowledge gained from con-
structing the representations of the converging and diverging spherical
waves, the following is proposed as a model of a focused acoustic beam:

i A 2π π/2 i
ϕi = G(ξ)eik p̂ ·x sin ξ dν dξ. (6.4.1)
2π 0 0
By eliminating the complex legs of the contours, the sudden jump at

x3 = 0 has been eliminated, and by introducing a function G(ξ), the
focal region at the origin has been modulated to capture more accurately
the fact that radiation cannot be focused to a point. The function G(ξ)
is defined so that it is symmetric in its argument ξ and is non-zero
for ξ ∈ (−π/2, π/2). Lastly, a superscript i has been added because,
in subsequent sections, (6.4.1) will become the incident focused beam
emitted from the lens aperture.
Two additional representations of (6.4.1) are useful. Using the results
from Appendix B, the first is
π/2
ϕi = iA G(ξ) eikx3 cos ξ J0 (kρ sin ξ) sin ξ dξ, (6.4.2)
0
and the second is

iA π/2 (1),(2)
i
ϕ = G(ξ) eikx3 cos ξ H0 (kρ sin ξ) sin ξ dξ. (6.4.3)
2 −π/2
(1)
The Hankel function H0 is selected when x3 > 0. Its branch cut pro-
ceeds from 0 to (−π/2−i∞); the integration contour passes above it. The
(2)
Hankel function H0 is selected when x3 < 0. Its branch cut proceeds
from 0 to (−π/2 + i∞); the integration contour passes below it.
The function G(ξ) is specified by asking that, far from the focal plane,
in the halfspace x3 < 0, ϕ match a prescribed wavefield. This wavefield
is thought of as that which fills an aperture, of radius b, placed at a
distance |x3 | = f from the focal plane. It is specified as
2n e−ikr
ϕi = −H(b − ρ) A e(ρ/a) , (6.4.4)
kr
where H(x) is the Heaviside step function and r = (ρ2 + x32 )1/2 . Note
that the aperture is filled with a converging spherical wave. Recall that
x1 = ρ cos θ and x2 = ρ sin θ. The coordinate φ has been introduced
such that φ ∈ [0, π/2), ρ = r sin φ, and |x3 | = r cos φ. The parameters
a and n, where n is a positive integer, allow one to describe how fully the
incident wavefield fills the aperture. Figure 6.3 shows three profiles for
various values of a/b and n. The larger a/b, the more the wavefield fills
the aperture, and the larger n, the more the function resembles a top-hat.
Equation (6.4.3) is asymptotically approximated as kr → ∞, using
a procedure identical to that used in §2.4 to approximate the wavefield
scattered into the fluid. This approximation gives
2
n = 1, a/b = 0.50
n = 2, a/b = 0.75
|k2 ϕ i |
n = 4, a/b = 0.85
0
0 ρ /b 1
Fig. 6.3. The distribution of the wavefield over an imagined aperture at

|x3 | = f . A is selected arbitrarily to give the vertical scale shown. The
parameters n and a/b measure how fully the converging spherical wave fills
the aperture.
e∓ikr
ϕi ∼ ∓A G(φ), kr → ∞, x3 ≶ 0. (6.4.5)
kr
Matching (6.4.5) with (6.4.4) gives
2n
f tan φ
G(φ) = H(βa − φ) exp − , φ ∈ [0, π/2). (6.4.6)
a
The angle βa is defined by cot βa = F . Equations (6.4.1) and (6.4.6)

together define the model of a focused beam to be used here.
Note that the choice of the wavefield in the aperture is such that
edge-diffracted waves are made small. This choice was based on the
observation that the microscope’s acoustic lens did not have sharp edges;
accordingly, it was reasoned that only weak edge diffraction occurred.
6.5 Scattered focused beam

6.5.1 Representations
The reflection of the focused beam is needed next. It is calculated in
much the same way as was done in §3.1. Noting that the incident focused
beam (6.4.1) is an integral over plane waves, an integral representation
of the reflected focused beam can be written as

iA 2π π/2 r
ϕr = G(ξ)R(ξ)eik p̂ ·x ei2kzs cos ξ sin ξ dν dξ. (6.5.1)
2π 0 0
The propagation vector p̂ r is given by (3.1.10), but is repeated here:

p̂ r = sin ξ(cos ν ê1 + sin ν ê2 ) − cos ξ ê3 .
The plane-wave reflection coefficient R(ξ) is given by (3.1.12) and
(3.1.13). It is important to recall the branch cuts for R(ξ) in the ξ plane;
these are discussed in the paragraph following (3.1.15) and partially
indicated by Fig. 3.3. These cuts determine the Riemann sheet on which
the contour of integration of (6.5.1) is taken; the integration contour
passes below the branch cuts whose branch points are ξbL and ξbT .
To interpret (6.5.1) it is helpful to refer to Fig. 6.1(b). First recall
that the origin of the coordinate system for the focused beam is the
geometrical focal point, which in Fig. 6.1(b) is a distance |zs | from the
interface. If the incident beam strikes the interface before it focuses,
then its reflection focuses in a reflected focal plane at x3 = 2zs , the
mirror image of the focal plane. Combining the phase terms, a distance
|x3 − 2zs | emerges, indicating that, for 2zs < x3 < zs , (6.5.1) represents
a beam converging to (0, 0, 2zs ) while, for x3 < 2zs , it represents a beam
diverging from (0, 0, 2zs ).
Again two additional representations of (6.5.1) are useful. Using the
results from Appendix A, the first is similar to (6.4.2):
βa
ϕr = iA G(ξ)R(ξ) e−ik(x3 −2zs ) cos ξ J0 (kρ sin ξ) sin ξ dξ. (6.5.2)
0
The second is similar to (6.4.3). However, this representation will be

used to asymptotically approximate the reflected wavefield; thus, it is
important to scale the variables to identify what parameter is large. The
scaled variables ρ̄ = ρ/b, x̄3 = x3 /f , and z̄s = zs /f are introduced. This
gives the second representation as

iA βa (1),(2)
ϕr = G(ξ)R(ξ) e−ikbF (x̄3 −2z̄s ) cos ξ H0 (kbρ̄ sin ξ) sin ξ dξ.
2 −βa
(6.5.3)
(1)
The Hankel function H0 is selected when x3 < 2zs . In this case the
(1)
contour passes above the branch cut for H0 , whose cut proceeds from
0 to (−π/2 − i∞). It also passes above the branch cuts of R(ξ), whose
branch points are −ξbL and −ξbT , and below those whose branch points
are ξbL and ξbT . This is a beam diverging from the reflected focal plane.
(2)
The Hankel function H0 is selected when 2zs < x3 < zs . The contour
passes below the branch cut for the Hankel function, which proceeds
from 0 to (−π/2 + i∞). However, it passes above the branch cuts of
R(ξ), whose branch points are −ξbL and −ξbT , and below those whose
branch points are ξbL and ξbT . This is a beam converging toward the
reflected focal plane.
Recalling the discussion of §6.2, F ≈ 1 and kb 1. Therefore, the
integral (6.5.3) can be asymptotically approximated in the same way as
was ur in §3.1.4, though in this case kb has been explicitly identified
as the large parameter. However, proceeding in this way will lead to an
unexpected difficulty for the case of the reflected converging beam. It
will become apparent that, when the reflected focal plane lies above it,
leaky Rayleigh waves are excited at the interface that focus at ρ = 0.
Therefore, the asymptotic approximation in this latter case must be
modified somewhat from that described in §3.1.4.
6.5.2 Diverging scattered beam, x3 < 2zs

(1)
Where x3 < 2zs , the asymptotic approximation for H0 , taken from
Appendix A, is introduced into (6.5.3). As well, the spherical coordinates
(r, φ, θ) are introduced by setting ρ = r sin φ and |x3 − 2zs | = r cos φ
(θ never appears because the wavefields are axisymmetric about the x3
axis). Note that φ ∈ [0, π/2). Equation (6.5.3) is thus approximated as
1/2 βa
iA 2
ϕ ∼
r
e iπ/4
G(ξ)R(ξ) e±ikbF r̄ cos(ξ−φ) sin1/2 ξ dξ,
2 πkbρ̄ −βa
kb → ∞, (6.5.4)
where r̄ = r/f .
Equation (6.5.4) is very similar to (3.1.15). The steepest-descents
contour for this case is identical to that shown in Fig. 3.3(b). The contri-
butions from the end points at ±βa are not significant, primarily because
G(βa ) is small. Carrying out the approximation gives
ϕr ∼ ϕg + χd (ξR , φ) ϕR , kb → ∞, (6.5.5)
where

1 if C(φ) encloses ξR ,
χd (ξR , φ) =
0 otherwise.
This is a non-uniform approximation in the sense indicated in §3.1.4.
The specularly reflected beam ϕg is given as
eikr
ϕg = AG(φ)R(φ) . (6.5.6)
kr
Using Eqs. (3.1.20) and (3.1.21), and making approximations identical to

those used to calculate (3.1.22), the leaky Rayleigh wave ϕR is given as
1/2
2π
ϕ = −AG(βR )
R
2αR eiπ/4 eikr cos(ξR −φ) , (6.5.7)
krR
where rR = ρ/ sin βR . The leaky Rayleigh-wave pole is defined by

(3.1.20). The approximation (3.1.21) and the approximation ξR ≈ βR
have been used in the amplitude. The phase term can be expanded
into real and imaginary parts to exhibit the fact that (6.5.7) is a leaky
Rayleigh wave.
Equations (6.5.5)–(6.5.7) describe a scattered wavefield that consists
of a diverging spherical wave emerging from r = 0 and a leaky Rayleigh
wave radiating from the interface. Both these waves are collected by the
aperture and transmitted to the transducer. Figure 6.4 shows |k 2 ϕr | as
a function of the scaled radius ρ/b, at x3 = −f , the plane of the lens
aperture. The spikes in the graph arise from the fact that the asymptotic
approximation is not uniform. The oscillations to the right of the large
spike indicate the presence of the leaky Rayleigh wave in the aperture. It
is important to note that the transducer responds to this entire signal;
it does not sense or separate the individual parts.
Leaky Rayleigh waves

⏐k 2 ϕ r⏐
0
0 ρ /b 1
Fig. 6.4. A plot of the magnitude of |k2 ϕr | against ρ/b. Note the presence
of the leaky Rayleigh wave, starting at ρ/b ≈ 0.3, in the wavefield filling the
aperture. Compare this figure with the corresponding solid curve in Fig. 6.3.
The various parameter values used in Fig. 6.4 are the following:
3
ρs = 2200 kg/m , cL = 5960 m/s, cT = 3760 m/s,
3
ρf = 998 kg/m , c = 1480 m/s,
F = 0.75, N = 178, n = 1, a/b = 0.5. (6.5.8)
The first line lists the parameters describing the solid, and the second
those describing the coupling fluid; the solid is fused quartz and the
fluid water. The third line lists the parameters describing the lens and
aperture. The frequency of operation is taken as 225 MHz. The leaky
Rayleigh pole ξR is found numerically. The value of zs is chosen such
that kzs /2π = −10. The constant A is given the same value as that used
in Fig. 6.3.
6.5.3 Converging reflected beam, 2z s < x3 < z s

To approximate asymptotically the converging beam, where 2zs < x3 <
zs , (6.4.3) is again used. The spherical coordinates (r, φ, θ) are intro-
duced, with φ ∈ [0, π/2). The contour in ξ is distorted to the contour
(2)
Re[cos(ξ − φ)] = 1 with no approximation of H0 being made. This will
become the contour of steepest descents. Figure 6.5 shows the complete
distortion of the original contour into this contour. The branch cut for
(2)
the H0 is not shown. The dashed lines indicate that the contour must
be distorted onto a lower Riemann sheet in order for the contour to pass
through the stationary point ξ = φ. This means that branch cuts are
surrounded; however, these contributions are ignored. The pole terms
are now extracted, giving
ϕR = −i 2πA G(βR ) αR sin βR e−ikr cos φ cos ξR

!
(1) (2)
× H0 (kρ sin ξR ) − χc (ξR , φ)H0 (kρ sin ξR ) , (6.5.9)
where

1 if C(φ) encloses ξR ,
χc (ξR , φ) =
0 otherwise,
and ρ = r cos φ. The leaky Rayleigh-wave pole is given by (3.1.20). The
approximation (3.1.21) and the approximation ξR ≈ βR have been used
in the amplitude. Note that the indicator χc (ξR , φ) sets the second term
to zero when φ is greater than βR , whereas χd (ξR , φ) sets the correspond-
ing term in (6.5.5) to zero when φ is less than βR . When φ is less then
βR both Hankel functions are present, indicating that the leaky Rayleigh
ξ plane
–βa π /2
– π /2
φ βa
Fig. 6.5. A sketch of the steepest-descents contour for the case where 2zs >
(2)
x3 > zs . The branch cut for H0 is not shown. The somewhat complicated
path is necessary because the integration contour can be closed at infinity only
in the indicated quadrants.
waves are focusing on the interface and along the axis ρ = 0. More-
over, the first Hankel function is always present because the pole −ξR
is always captured. When φ is greater then βR this term represents an
outgoing, cylindrical, leaky Rayleigh wave. Writing the pole terms as
Hankel functions gives an expression that is accurate when ρ → 0.
Once the pole terms have been extracted, the remaining integral can
(2)
be approximated by introducing the asymptotic approximation to H0
and noting that the contour of integration is that of steepest descents.
The geometrical term is approximated as
e−ikr
ϕg ∼ −AG(φ)R(φ) . (6.5.10)
kr
This expression is accurate when 2zs < x3 < zs , provided zs = 0
or, equivalently, r = 0. Note the phase shift of π in passing from (6.5.10)
to (6.5.6).
The asymptotic approximation to ϕr is therefore given as

ϕr ∼ ϕg + ϕR , kb → ∞, (6.5.11)
Aperture
βa
βR
Outermost φ Leaky
ray Rayleigh
rays
Geometrical
ray x3 ⫽2zS
x3 ⫽zS
Fig. 6.6. A sketch of the rays describing the focusing scattered beam. The
solid lines indicate the outermost geometrical rays, and the dashed ones the
leaky Rayleigh waves.
where ϕR is given by (6.5.9), with no restriction on ρ, and ϕg by (6.5.10),

with the restriction that r = 0. Accordingly, the overall expression is
only accurate provided the reflected focal plane and the interface are
separated. Moreover, (6.5.11) is a non-uniform approximation in the
sense indicated in §3.1.4. While (6.5.11) is limited in its usefulness, it
does provide a good physical picture of what is happening. Figure 6.6 is a
sketch that attempts to illustrate this equation. The solid lines indicate
the outermost rays of the focused cone of rays: they reflect from the
interface and are focused in the reflected focal plane. The dashed lines
indicate the leaky Rayleigh waves: within a right circular cone, which
opens downward and ends at the interface, intersecting it to form a circle
with radius |zs | tan βR , they focus on the axis ρ = 0 whereas, outside
the cone, they form an outgoing cylindrical wave steadily leaking into
the coupling fluid.
6.6 An electromechanical reciprocity relation

6.6.1 Piezoelectric coupling and reciprocity
Having considered the wavefield in the fluid and in the solid, we now turn
to the question of how to combine this knowledge to determine the elec-
tric response of the transducer/microscope. This requires a model of the
measurement device which is described in the next section. Here the the-
ory is described that enables us to combine the mechanical and electrical
effects. The key is reciprocity, and goes beyond the concepts introduced
in Chapter 4 to include a fully coupled field theory, as compared with
the purely elastic reciprocity identities derived previously.
Transducers operating at microwave frequencies frequently use piezo-
electric coupling to convert electrical to mechanical signals and vice
versa. While the equations of piezoelectricity are not explicitly used
in this book, an electromechanical reciprocity identity, which assumes
the coupling is piezoelectric, is used; and its use is central to the
model of scanned acoustic imaging described in this chapter. To sketch
its derivation, a summary of the equations of linear piezoelectricity is
needed.
A description of piezoelectricity can be found in Auld (1990a, pp.
265–298). The electromechanical reciprocity identity itself is also derived
in Auld (1990b, pp. 153,154), and is further discussed in Achenbach
(2003, pp. 233–246).
Maxwell’s equations,
eijk ∂j Ek + ∂t Bi = 0, eijk ∂j Hk − ∂t Di = Ji ,
are coupled with (1.1.1) through the constitutive relations
Di = κij Ej + ρijk jk ,

τij = −ρijk Ek + cijkl kl .
In addition Bi = νHi . The terms E, B, H, and D are the electric

intensity, the magnetic induction, the magnetic intensity, and electric
displacement. The vector J is an imposed current; the piezoelectric
materials are assumed to be non-conducting. Piezoelectric materials
must be anisotropic; thus, the various coupling parameters are tensors
of the order indicated by their subscripts: κij are the components of the
dielectric permittivity tensor, ρijk the components of the piezoelectric
tensor, and cijkl the components of the elastic tensor. The scalar ν is
the magnetic permeability. All the terms are assumed to be constant.
Using energy arguments and the symmetry of τ , it can be shown that
cijkl = cklij , cijkl = cjikl ,

ρikl = ρkli , ρijk = ρjik ,
κij = κji .
Other symmetries can be derived from these.

Let (E1 , H1 , τ 1 , u1 ) and (E2 , H2 , τ 2 , u2 ) be two electromechanical

states that satisfy the previously listed equations. Moreover, in keeping
with the previously stated convention, assume that ∂t → −iω. In essence,
the equations satisfied by state 1 are multiplied by the variables of state
2, and the equations satisfied by state 2 are multiplied by those of state
1; the equations are then added or subtracted to remove common terms,
and combined in a way that gives a divergence on one side and forcing
terms on the other. The only point to note is that it is the particle veloc-
ity v = −iωu, rather than the particle displacement, that is needed. The
result of these operations is the following reciprocity identity:

∂i −iω(u1j τij 2
− u2j τij
1
) + eijk (Ej1 Hk2 − Ej2 Hk1 )
= −iωρs (u2j fj1 − u1j fj2 ) + (Ej2 Jj1 − Ej1 Jj2 ). (6.6.1)
We will return to this identity in §6.7 after a brief example showing

application to a simpler but similar configuration.
6.6.2 Example: reflection from a rigid surface

To demonstrate the use of (6.6.1), reflection, from a perfectly rigid sur-
face, of a normally incident sound beam radiated by a circular transducer
is considered. This example is taken with little change from Block et al.
(2000). Figure 6.7 schematically indicates the configuration: a cylindri-
cal beam is emitted by the transducer, propagates through a fluid used
to couple the transducer and rigid surface, and is reflected, the reflection
being received by the same transducer.
Equation (6.6.1) is converted to an integral over a surface that cuts
the coaxial cable of the transducer along ∂Re , goes outward to infinity,
where the wavefields are assumed to vanish, and is closed by the rigid
surface ∂Rr . The integral then becomes

1 2
eijk Ej1 Hk2 − eijk Ej2 Hk1 n̂i dS = −iω ui p − u2i p1 n̂i dS.
∂Re ∂Rr
(6.6.2)
The only electrically active part of ∂Re is the cross-sectional area

of the coaxial cable. The area of the rigid surface ∂Rr insonified is
essentially that directly below the transducer. Diffraction spreading from
the transducer is taken as negligible. Accordingly, the incident beam is
approximated as
ui3 = Aeik(x3 +d) H(1 − ρ/a), ui1 = ui2 = 0, (6.6.3)

∂Re
∂Rm
x3
Fig. 6.7. A schematic drawing of the measurement of the reflection coefficient

of a rigid surface ∂Rm . A cross-section is shown. The surface ∂Re cuts the
coaxial cable coming from the transducer. d is the distance to the surface from
the transducer aperture; a is the aperture radius.
where ρ = (x21 + x22 )1/2 and a is the radius of the transducer aperture.
The Heaviside step function H(ξ) introduces a jump that is not physi-
cally present, but for this example this affects nothing. In §2.1.2 this is
explored in detail.
The reflected beam is
ur3 = −Ae−ik(x3 −d) H(1 − ρ/a), ur1 = ur2 = 0, (6.6.4)
where we have used the fact that the reflection coefficient for the particle
displacement amplitudes at the rigid surface is (−1).
The reciprocating wavefield 1 is taken as the incident wavefield (that
when no rigid surface is present), and the reciprocating wavefield 2 as the
total wavefield, incident and reflected. Although the wavefield is time-
harmonic, it is assumed that it is but one component of a temporally
separated signal; the wavefields are assumed windowed in time such that
only the initially reflected beam (very often a tone burst in time) is
received, and subsequent reverberations are not measured. It is assumed
that the coaxial cable is cut sufficiently far from the transducer that the
overwhelmingly dominant contribution to the wavefield in the cable is
that of the lowest TEM mode (Auld, 1979; Collin, 1991, pp. 247–259).
At ∂Re we therefore take as electromagnetic wavefield 1
+ +
Ei1 = 1 + Ree
1
Ei , Hi1 = 1 − Ree1
Hi , (6.6.5)
and as electromagnetic field 2

+ +
Ei2 = 1 + Ree
2
Ei , Hi2 = 1 − Ree
2
Hi . (6.6.6)
1 2
The reflection coefficients measured in the coaxial line are Ree and Ree .
Equation (6.6.2) is evaluated using (6.6.3) through (6.6.6). This gives
ω 2 ρc
ΔRee = − |A|2 πa2 ei2(kd+α) , (6.6.7)
2P
2 1
where ΔRee = Ree − Ree and A = |A|eiα . The argument α is assumed to
be introduced by the transducer. The change in voltage, symbolized by
ΔV , is proportional to ΔRee , namely, ΔV = C(ω)ΔRee , where C(ω) is a
function of the transducer only and not the scatterer. The time-average
power P incident on the transducer is

−2P = eijk Ej+ Hk+ n̂i dS,
∂Re
where the minus sign arises because of the orientation of the unit normal.
We can arrange that the integrand be real.
1
If the transducer is perfectly matched to the fluid, then Ree = 0.
Moreover, if it is lossless (α real), then the time-averaged incident power
is simply P = (1/2) ω 2 ρc|A|2 πa2 , and therefore Ree2
= −ei2(kd+α) , the
displacement reflection coefficient for a perfectly rigid surface.
6.7 Measurement model

The previous section describes the reflected focused beam and leaky
surface wave, but does not indicate how that scattered wavefield is trans-
formed into a change in voltage. Looking back at Fig. 6.1(a), it is readily
appreciated that the basic function of the lens, buffer rod, and piezoelec-
tric transducer is to map the total echo collected by the aperture of the
lens to a change in voltage δV . To construct a mathematical model of this
mapping, the electromechanical reciprocity identity described in §6.6,
and given by (6.6.1), is used. An early use of this identity to construct
such models is given by Auld (1979); a more recent application that
significantly influenced the writer is Liang et al. (1985). The outcomes
from using it are often called measurement models. When no sources are
present this identity becomes

∂i −iω(u1j τij
2
− u2j τij
1
) + eijk (Ej1 Hk2 − Ej2 Hk1 ) = 0. (6.7.1)
To use (6.7.1), several assumptions must be made. The electromechan-
ical reciprocity identity is assumed to hold throughout R, whatever the
internal structure of the real device. The wavefields on the surface ∂R
are assumed to be such that all electromagnetic waves enter or leave R
through the electrical plane ∂Re , and all mechanical waves enter or leave
through the mechanical plane ∂Rm ; elsewhere on ∂R the wavefields van-
ish. The two planes are considered subsets of the overall surface ∂R and
are identified in Fig. 6.8.
The transducer is connected to the electrical source and receiver
through a coaxial cable. Only a transverse, electromagnetic plane wave
propagates in a coaxial cable (Collin, 1991, pp. 247–251); all higher
modes are cut off. By placing ∂Re far enough from the junction of the
coaxial cable with the transducer, it can be assumed that any higher
modes that might be excited at the junction have become negligible;
therefore, at ∂Re only a transverse plane wave is present.
Lastly, it is assumed that the multiple scattering that occurs between
the lens aperture and the interface is negligible. This is a reasonable
assumption because tone-bursts, not time-harmonic signals, are used,
and thus any multiply-scattered signals can often be separated, in time,
∂Re
R ∂R
^
n
∂Rm
Fig. 6.8. The volume R in which the reciprocity relation is applied. The surface
∂R includes the electrical plane ∂Re and the mechanical plane ∂Rm .
from the first echo. The δV calculated in this chapter, which uses the
time-harmonic assumption, is a Fourier transform of the change in volt-
age detected in time. At the mechanical plane ∂Rm , therefore, it is
assumed that the scattered disturbance has been scattered from the
interface only once.
Equation (6.7.1) is integrated over the volume R, shown in Fig. 6.8,
and transformed into a surface integral over ∂R. This gives

eijk (Ej1 Hk2 − Ej2 Hk1 )ni dS = iω (u1j τij
2
− u2j τij
1
)ni dS.
∂Re ∂Rm
(6.7.2)
Wavefield 1 is taken as
Ei1 = (1 + Re )Ei+ , Hi1 = (1 − Re )Hi+ , u1j = uij , τij
1
= −pi δij .
(6.7.3)
The first two terms are assumed to be evaluated at ∂Re , and the last two
at ∂Rm . This is the wavefield that is present when there is no interface.
The first two terms describe the transverse electromagnetic wave in the
coaxial cable; there is both an incident and a reflected term, the reflec-
tion arising from the junction between the cable and the transducer. In
practice the designer tries to make Re as small as possible. The second
two terms describe the focused sound beam that is radiated from the
lens aperture. Wavefield 2 is taken as
Ei2 = (1 + Re + δRe )Ei+ , Hi2 = (1 − Re − δRe )Hi+ ,
u2j = uij + urj , 2
τij = −(pi + pr )δij . (6.7.4)
As with (6.7.3), the first two terms are assumed to be evaluated at ∂Re ,
and the last two at ∂Rm . This is the wavefield that is present when sound
is scattered back to the lens aperture from the interface. The first two
terms again describe the transverse electromagnetic wave in the coaxial
cable; there is now an increment to the reflection, δRe , arising from
the sound collected at the aperture. The second two terms describe a
focused sound beam that is incident on and scattered from the interface.
Substituting (6.7.3) and (6.7.4) into (6.7.2) gives

iω
δRe = (ur pi − uij pr )nj dS, (6.7.5)
4 P ∂Rm j
where
1
P =− eijk Ej+ Hk+ ni dS.
2 ∂Re
The normalization P is the incident power injected into the junction of

the cable with the transducer, so that it is known.
Despite the assumptions leading to it, (6.7.5) is quite remarkable for
at least two reasons. First, it maps the mechanical wavefield at ∂Rm to
what is measured at the electrical plane ∂Re . Note that it clearly indi-
cates how the transducer responds only to the complete signal, returning
a single scalar quantity δRe . Second, the incident focused beam has a
limited angular spectrum because the aperture is finite; thus, when it
combines with the scattered wavefields, it explicitly limits the angular
spectrum of the scattered wavefield that can be collected.
6.8 Acoustic material signature

Because the incident and reflected wavefields in the coupling fluid have
been expressed in terms of potentials, it is convenient to express (6.7.5)
in terms of them as well. From §1.1,
pi,r = ρf ω 2 ϕi,r , ui,r i,r

j = ∂j ϕ .
It is also convenient to renormalize (6.7.5) so that its maximum absolute

value is approximately one. The plane ∂Rm is taken to be the interface
x3 = zs . The outcome is an expression for a term labeled δV which
is proportional to δRe . δV is named the change in voltage, though it
is dimensionless, because it is proportional to the change in measured
voltage caused by the presence of the interface and any features on it.
The δV is given as

δV = E (ϕs ∂n ϕi − ϕi ∂n ϕs ) dS, (6.8.1)
∂Rm
where

E −1 = 2 (∂n ϕi ϕi ∗ ) dS.
∂Rm
Equations (6.4.2) and (6.5.2) indicate that the incident and reflected
wavefields are inverse Hankel transforms. Examining (6.8.1), it becomes
clear that it is an integral of the product of Hankel transforms
over a semi-infinite domain. This permits Parseval’s relation, given in
Appendix A (see also (2.1.36)), to be used to collapse what might seem
to be a complicated triple integral into a single one. This is a key step
in the derivation, but not one that could be foreseen when the work was
first begun. Carrying out this step gives
6.8 Acoustic material signature 117
βa
δV (s, zs ) = 2iD G2 (ξ) R(ξ, s)ei2kzs cos ξ dξ, (6.8.2)
0
where D = k 5 E or
βa
D−1 = 2i G2 (ξ) sin ξdξ.
0
This is the central result of this chapter, and mentioned in the

introduction. A dependence on the position of the interface is indicated
by s, and a dependence on the position of the geometric focal plane by
zs . If zs ≈ 0 then (6.8.2) describes how the reflection acoustic micro-
scope makes an image when it is scanned across the interface. In this
case no leaky Rayleigh wave is excited. Assuming that R(ξ, s), as a func-
tion of s, changes slowly with respect to wavelength – this is equivalent
to assuming that no vertical discontinuities are present – then (6.8.2)
maps the reflection at s to a change in voltage at the same position. The
set of all these reflections, when converted to electrical signals, forms
the image. Note that δV is a weighted average over the lens aperture of
the reflection coefficient; the weight function is the square of the aperture
function G(ξ). In other words, perfect information about the interface
is not recovered.
When s is fixed and zs varied, the acoustic microscope acts as an
interferometer. A non-uniform asymptotic approximation of (6.8.2), for
|zs | = 0, gives
ei2kzs
δV (s, zs ) ∼ ±DG2 (0) R(0, s)
k|zs |
− H(−zs ) DG2 (βR ) 8πi αR sin βR e−2ik|zs | cos ξR ,
zs ≷ 0, kb → ∞; (6.8.3)
here, H(x) is the Heaviside step function. Note that no leaky Rayleigh
wave is collected by the lens aperture for zs > 0. The approxima-
tions (3.1.19) and ξR ≈ βR have been used in the amplitude, in the
second term.
For zs negative, (6.8.3) indicates that δV records an interference pat-
tern as zs is varied. Figure 6.9 shows a plot of this interference pattern
for fused quartz. The values are identical to those used for Fig. 6.4 and
are cited in (6.5.8). The solid line indicates the asymptotic result (6.8.3),
and the dashed line a numerical evaluation of (6.8.2). The interference
pattern is evident. The distance between the peaks, Δ, is given by the
equation
−1
Δ = 2(1 − cos βR ) .
From this, the variation in the Rayleigh wavespeed can be determined.

The graph shown in Fig. 6.9 is named the acoustic material signature.
A similar analysis for the line focus can be performed; see Rebinsky and
Harris (1992b) for details. Figure 6.10 illustrates the relative magnitude
of the acoustic material signature for the two-dimensional and three-
dimensional focusing microscopes.
Much more can be done with (6.8.2). If the geometrical focal plane is
placed just slightly below the interface, the leaky Rayleigh wave partic-
ipates in the image formation, causing vertical surface features, such as
the opening of a surface-breaking crack, to be detected acoustically, even
when they are difficult to detect with an optical microscope. The mono-
graph by Briggs (1992) gives many examples of this feature. Rebinsky
and Harris (1992b) show that the leaky surface wave can be used to set
up an interference pattern between the incident surface wave and one
reflected from a vertical discontinuity. Lastly, (6.8.2) is a more general
1.0
Spectral
0.8
Asymptotic
0.6
| δV |
Δ
0.4
0.2
0
–30 –20 –10 0 10
kzs /(2π)
Fig. 6.9. The acoustic material signature of fused quartz. The solid line indi-
cates the result of plotting the asymptotic approximation, and the dashed line
indicates a numerical evaluation of (6.8.2). The distance between the peaks is
denoted by Δ.
6.9 Summary 119
1.0
Point focus
Line focus
0.8
0.6
⏐δVo⏐
0.4
0.2
0.0
–30 –20 –10 0 10
Zs /λ w
Fig. 6.10. Comparison of the acoustic material signature for line vs. point
focus, from Rebinsky and Harris (1992b).
imaging expression than its derivation might suggest. Many scanning,

acoustic detection schemes used in industrial non-destructive evalua-
tion work more or less in the same manner as does a reflection acoustic
microscope, so that (6.8.2) or a minor modification of it describes a wide
variety of such schemes.
6.9 Summary
(1) Equations (6.4.1) and (6.4.6) describe a model of a focused beam. It
does not exactly describe focusing through an aperture, as a study
of focusing calculations that begin at the plane of the aperture
indicates (Born and Wolf, 1999, pp. 484–499), but its form allows
one to readily calculate the scattering of a focused beam and, just
as importantly, to invoke Parseval’s relation to calculate δV .
(2) Equation (6.5.1) and the analysis of §6.5 describe the reflected
focused beam. Of particular interest is the fact that the leaky
Rayleigh waves form a second focal point at the interface.
(3) The discussion of §6.6, leading to (6.7.5), indicates how one may
relate the measured electrical signal from a piezoelectric transducer
to the scattered mechanical wavefields, and as such this plays an

important role in many elastodynamic measurement models.
(4) Equation (6.8.2) is the principal result of this chapter. It is the
equation that describes how a reflection acoustic microscope makes
an image and acts as an interferometer. In one guise or another it
describes acoustic imaging in many other situations that arise in the
non-destructive evaluation of manufactured parts.
7
Acoustic diffraction in viscous fluids
When a time-harmonic acoustic plane wave in a viscous fluid is incident

on a rigid wall, the additional constraint of zero tangential velocity
excites a transverse wave, damped by viscosity, of the type described
by Batchelor (1967), for example, in a section on unidirectional flows.
A semi-infinite viscous fluid is bounded by a wall that oscillates in its
own plane, and the purely imaginary square of the wavenumber shows
that the induced flow is essentially confined near the wall in what is often
described as a Stokes layer. In this chapter, the effects of compressibility
and viscosity are seen to be intertwined; the dilatation propagates as a
damped acoustic wave and the vorticity propagates with the dilatation
along the wall, with its damping in the normal direction consequently
modified.
The dilatation satisfies a wave equation but, in contrast to the rota-
tion in an elastic solid, the vorticity (time derivative of the rotation) is
governed by a diffusion equation. However, for a disturbance of given
frequency, the governing equations have the same form as in the elastic
case and thus the worked examples below relate well to two of the prob-
lems solved in Chapters 2 and 3. A contrast is provided by the deduction
of the displacement fields directly from solution forms for the dilatation
and rotation, without introducing the potentials.
7.1 Theory
The constitutive relation for linearized flow in a homogeneous viscous
fluid is expressed by
1
τij = λf ∂k uk δij + 2ν∂t ij − ∂k uk δij , (7.1.1)
3
121
122 Acoustic diffraction in viscous fluids
where ν is the kinematic viscosity (Batchelor, 1967, equation (3.3.11)).

Thus, with ρ replaced by ρf and the acceleration f absent, the equation
of motion (1.1.1) can be written as
1
c2 + ν∂t ∂i ∂k uk + ν∂t ∂j2 ui = ∂t2 ui . (7.1.2)
3
The acoustic particle velocity v is expressed in terms of the displacement
by ∂t u. On taking the divergence (operator ∂i ) of (7.1.2), the dilatation
satisfies the damped wave equation
4
c2 + ν∂t ∂i2 − ∂t2 ∂k uk = 0. (7.1.3)
3
On taking the curl (operator elmi ∂m ) of (7.1.2), the vorticity satisfies
the diffusion equation
(ν∂j2 − ∂t )elmi ∂m (∂t ui ) = 0, (7.1.4)
as in unsteady creeping (Stokes) flow. Vorticity is created at√rigid

boundaries and, according to (7.1.4), diffuses on the length scale νt.
As in earlier chapters, the time dependence e−iωt is assumed and sup-
pressed below. Then (7.1.3) and (7.1.4) reduce to the pair of Helmholtz
equations
(∂i2 + ka2 )∂k uk = 0, (7.1.5)
iω
∂j2 + elmi ∂m (∂t ui ) = 0, (7.1.6)
ν
where
−1/2 √
ω 4i νω
ka = 1 − 2 , = 1. (7.1.7)
c 3 c
Since the viscous decay length (ν/ω)1/2 is assumed here to be much

smaller than the acoustic length c/ω, as in air or water at normal acoustic
frequencies, the viscous effects are essentially confined to a thin Stokes
layer adjacent to a boundary. Comparison with (1.1.21) and (1.1.22)
shows that c2L and c2T are replaced here by c2 (1 − 4i2 /3) and −i2 c2 ,
respectively.
The reflection alluded to in the introduction to this chapter has the
damped acoustic plane wave
ui = A(ê1 cos φ0 − ê2 sin φ0 )eika (x1 cos φ0 −x2 sin φ0 ) , 0 ≤ φ0 < π/2,
(7.1.8)
incident on a rigid wall at x2 = 0. The no-slip condition, namely, zero

total displacement in the oscillatory flow, requires the reflected acoustic
wave to have amplitude different from A so as to include a vorticity field
parallel to ê3 exp[ika x1 cos φ0 ] exp[−x2 (ka2 cos2 φ0 − iω/ν)1/2 ], according
to (7.1.6). The vorticity generates a unique solenoidal displacement and
the reflected displacement is found to be
ur = [−1 + D(φ0 )(ka2 cos2 φ0 − iω/ν)1/2 ]
× A(ê1 cos φ0 + ê2 sin φ0 )eika (x1 cos φ0 +x2 sin φ0 )
− D(φ0 ) cos φ0 A[ê1 (ka2 cos2 φ0 − iω/ν)1/2 + ê2 ika cos φ0 ]
2 2
φ0 −iω/ν)1/2
× eika x1 cos φ0 e−x2 (ka cos , (7.1.9)
where
2 sin φ0
D(φ0 ) = .
(ka2 cos2 φ0 − iω/ν)1/2 sin φ0 − ika cos2 φ0
But (7.1.7) implies that iω/ν dominates ka2 cos2 φ0 , whence the solenoidal
field is almost parallel to the rigid wall, with exponential decay factor
∼ (−iω/ν)1/2 . Thus it is indeed an acoustic modification of the unidirec-
tional transverse wave generated by an oscillating boundary. Note that
(7.1.9) represents ur as a ‘soft’ reflection of the incident wave plus a
wave combination whose tangential component vanishes at the wall.
7.2 Diffraction by a half-plane

Suppose that the damped acoustic plane wave (7.1.8) is incident on
the semi-infinite barrier x2 = 0, x1 > 0. The geometry is therefore that
depicted in Fig. 2.6 except that p̂i lies in the second quadrant, that is, φ0
is replaced by −φ0 in (2.2.2). Express the total wavefield ut as ui +u. The
scattered wavefield satisfies the pair of reduced wave equations (7.1.5)
and (7.1.6) and, for no-slip at the barrier, must satisfy
u(x1 , 0) = −ui (x1 , 0)
= −A(ê1 cos φ0 − ê2 sin φ0 )eika x1 cos φ0 , x1 > 0. (7.2.1)
At x2 = 0, x1 < 0, u and ∂2 u are continuous. It is assumed that, at
infinity, u has only plane and diffracted waves.
The plane x2 = 0 is one of reflection symmetry so that the problem
can be split into two problems, one antisymmetric and one symmetric
with respect to the reflection plane. Thus write u = ua + us , where ua1
and us2 are odd functions of x2 while us1 and ua2 are even functions of x2 .
The solution in the half-space x2 > 0 then suffices and is such that, at
x2 = 0,
ua1 = 0 = us2 , −∞ < x1 < ∞, (7.2.2)
(us1 , ua2 ) = −A(cos φ0 , − sin φ0 )eika x1 cos φ0 , x1 > 0, (7.2.3)
according to (7.2.1), and
(∂2 us1 , ∂2 ua2 ) = (τ1+ , τ2+ ), x1 > 0, (7.2.4)
∂2 us1 = 0 = ∂2 ua2 , x1 < 0, (7.2.5)
where τ1+ (x1 ), τ2+ (x1 ) are unknown functions.

With implementation of the Wiener–Hopf technique in mind, intro-
duce the Fourier transforms
∞
(∗ua , ∗us ) = (ua , us )e−iβx1 dx1 , (7.2.6)
−∞
which, according to (7.2.2)–(7.2.5), are such that

∗ a
u1 (β, 0) = 0 = ∗us2 (β, 0), (7.2.7)
0
[∗us1 (β, 0), ∗ua2 (β, 0)] = [us1 (x1 , 0), ua2 (x1 , 0)]e−iβx1 dx1
−∞
∞
−A [cos φ0 , − sin φ0 ]ei(ka cos φ0 −β)x1 dx1
0
[cos φ0 , − sin φ0 ]
= [∗u− ∗ −
1 (β), u2 (β)] −A ,
i(β − ka cos φ0 )
(7.2.8)
∞
[∂2 ∗us1 (β, 0), ∂2 ∗ua2 (β, 0)] = [τ1+ (x1 ), τ2+ (x1 )]e−iβx1 dx1
0
= [∗τ1+ (β), ∗τ2+ (β)]. (7.2.9)
Note that the principle of limiting absorption need not be invoked

because ka , defined by (7.1.7), has a positive imaginary part. The integral
in (7.2.8) converges provided φ0 is acute, as assumed, due to the damp-
ing along the barrier. If φ0 is obtuse, then the reflection (7.1.9) must be
included in the incident wavefield and a convergent forcing integral arises
from the need to achieve a smooth total wavefield along the negative x1
axis where damping occurs.
The reduced wave equations (7.1.5) and (7.1.6) and the Fourier
transforms (7.2.6) imply that
2 ∞ a
d 2 2 ∂u1 ∂ua2
0= − β + ka + e−iβx1 dx1
dx22 −∞ ∂x1 ∂x2
2
d 2 2 ∗ a d∗ua2
= − β + k iβ u 1 + ,
dx22 a
dx2
2 ∞ a
d 2 iω ∂u2 ∂ua1
0= − β + − e−iβx1 dx1
dx22 ν −∞ ∂x1 ∂x2
2
d 2 iω ∗ a d∗ua1
= − β + iβ u 2 − .
dx22 ν dx2
Seeking a solution that vanishes as x2 → ∞, it is noted that
d∗ua2 2 2 1/2
iβ ∗ua1 + = ka2 e−(β −ka ) x2
dx2
uniquely yields the irrotational dilatation determined by
2 2 1/2
∗ a
u = [−iβê1 + (β 2 − ka2 )1/2 ê2 ]e−(β −ka ) x2
,
while
d∗ua1 iω 2 1/2
iβ ∗ua2 − = e−(β −iω/ν) x2
dx2 ν
uniquely yields the solenoidal rotation determined by

∗ a iω 1/2 2 1/2
u = −(β − ) ê1 − iβê2 e−(β −iω/ν) x2 .
2
ν
A suitable combination that satisfies (7.2.7) is therefore

2 2 1/2
∗ a
u = B(β) [−iβê1 + (β 2 − ka2 )1/2 ê2 ]e−(β −ka ) x2

β2 −(β 2 −iω/ν)1/2 x2
+ iβê1 − 2 iω 1/2 ê2 e . (7.2.10)
(β − ν )
Similarly, the symmetric solution that satisfies (7.2.7) is

∗ s β2 2 2 1/2
u = D(β) − 2 2 1/2
ê1 − iβê2 e−(β −ka ) x2
(β − ka )

2 iω 1/2 −(β 2 −iω/ν)1/2 x2
+ [(β − ) ê1 + iβê2 ]e . (7.2.11)
ν
Evidently, in a modification of the plane wave representation used in
§2.2, the square roots must have positive real part for all β.
The functions B(β) and D(β) are determined by the mixed conditions
at x2 = 0, expressed in the β plane by (7.2.8) and (7.2.9). Substitution
of (7.2.10) and (7.2.11) shows that
2

β iω A cos φ0
D(β) − + (β 2 − )1/2 = ∗us1 (β, 0) = ∗u−
1 (β) − ,
(β 2 − ka2 )1/2 ν i(β − ka cos φ0 )

2 β2 A sin φ0
B(β) (β − ka2 )1/2 − 2 iω 1/2 = ∗ua2 (β, 0) = ∗u−
2 (β) + ,
(β − ν ) i(β − ka cos φ0 )
∗ + iω
τ1 (β) , ∗τ2+ (β) = ∂2 ∗ua2 (β, 0) = B(β)ka2 .
= ∂2 ∗us1 (β, 0) = D(β)
ν
Elimination of B(β) and D(β) now yields the disjoint pair of scalar
Wiener–Hopf equations
[cos φ0 , − sin φ0 ]
[∗u− ∗ −
1 (β), u2 (β)] − A
i(β − ka cos φ0 )
1/2
ν∗ + β 2 − iω
ν 1∗ +
= K(β) τ (β) , 2 τ2 (β) ,
iω 1 β 2 − ka2 ka
where K(β) is the Wiener–Hopf kernel, given by
iω −1/2
K(β) = (β 2 − ka2 )1/2 − β 2 (β 2 −
) . (7.2.12)
ν
The ratio of square roots in the first equation, necessitated by viscosity,
corresponds to interchanging the square roots in K(β). Note that K(β)
is defined so that it reduces, except for the i factor, to (2.2.7) in the
inviscid limit.
The
function K(β) is analytic except for branch
cuts from ka to
eiπ/4 ω/ν to ∞ eiπ/4 and from −ka to −eiπ/4 ω/ν to −∞ eiπ/4 . The
product
K(β)[2(β 2 − ka2 )1/2 − K(β)]
has zeros at ±βK , where, by use of (7.1.7), βK = (ω/c)(1 − 7i2 /3)−1/2 .
2
But the chosen branch cuts imply that, at β = ±βK , (βK − iω/ν)1/2
2 2 1/2 2
(βK − ka ) = −βK , i.e. K(k) has no zeros. Although general inte-
gral representations are available (Noble, 1988) for the decomposition,
K(β) = K + (β)K − (β), it is simpler and still highly accurate to derive the
functions K + (β) and K − (β) from a Padé approximation to the kernel
function K(β), as fully discussed by Abrahams (2000).
Davis and Nagem (2002) used the Padé approximant

c1 ω 2
K(β) ≈ (β 2 − ka2 )1/2 , (7.2.13)
c2 ω 2 − β 2 c2 2
where the coefficients c1 () and c2 () are, after substituting (7.1.7),
given by

1 2 3 2
c1 = i+ 4i
, c2 = c1 − .
2 1 − 3 2 2 1 − 7i
3
2
Although (7.2.13) is based on a simple Padé approximant with respect

to the point at infinity, excellent agreement with the kernel function is
obtained over the entire range −∞ < β < ∞, because the poles thus
introduced are sufficiently far from the real β axis. Approximations to
the factored kernel functions K + (β) and K − (β) can be obtained directly
from the immediate factorization of (7.2.13), and the Fourier integrals
for the displacement, u = ua + us , can then be evaluated by standard
numerical integration techniques.
The Helmholtz representation for u is inappropriate because it
becomes problematic at x2 = 0. Inverse Fourier transforms for
ϕa , ϕs , ê3 ψs , ê3 ψa can be readily deduced from (7.2.10) and (7.2.11),
but the reflections in x2 < 0 of the antisymmetric functions yield dis-
continuities. In the inviscid case discussed in §2.2, only ϕa is needed
and its reflective continuity is assured by condition (7.2.7). But in the
viscous case, (7.2.7) is satisfied by a balance of dilatational and rota-
tional waves whose break-up can be rectified only by adding conjugate
harmonic functions.
A similarity solution near the barrier edge shows that the velocity is
almost Stokesian, with ψ and ϕ ultimately biharmonic, thus verifying
the need for extra terms described above. Stokes flows become unique
when the edge singularity is minimized, giving uα = O(|x1 |1/2 ), ∂2 uα =
O(|x1 |−1/2 ) on x2 = 0. These estimates are needed for completion of
the Wiener–Hopf technique. Thus viscosity eliminates the physically
unrealistic inviscid displacement/velocity singularity.
7.3 Scattering of a spherical wave at a plane interface

Lighthill (1978b) discusses the sound field generated by a volume outflow
m(t) from the origin (r = 0), showing that the displacement poten-
tial is −m(t − r/c)/4πr. With m = m0 e−iωt and the time dependence
suppressed, this reduces to ϕ = −m0 eikr /4πr. This form is given in §3.1,
where the wavenumber k = ω/c in an ideal fluid is introduced.
The somewhat obvious modification due to viscosity can be readily
determined by inserting a source term in the continuity equation and
hence the term c2 m(t)∂i [δ(r)/4πr2 ] on the right-hand side of (7.1.2).
The time-periodic form of the resulting equation can be arranged as

4i δ(r)
∂i 1 − 2 ∂k uk − m0 + i2 (∂i ∂k uk − ∂j2 ui ) + k 2 ui = 0,
3 4πr2
with defined by (7.1.7). The middle term is solenoidal and hence

2 m0 δ(r)
∂k u k = k a 2 −ϕ , (7.3.1)
k 4πr2
where
ka2 m0 δ(r)
∂i2 ϕ + ka2 ϕ = . (7.3.2)
k 2 4πr2
Referring to (3.1.1), the solution is
ka2 m0 eika r
ϕ=− , (7.3.3)
k 2 4πr
and comparison of (7.3.1) and (7.3.2) shows that ui = ∂i ϕ.
Now consider the scattering problem depicted in Fig. 3.1 but with
a viscous fluid in the halfspace x3 < 0 and, in the first instance, a
rigid solid in x3 > 0. As in (7.1.9), the reflected field consists of a
‘soft’ reflection of the spherical wave, easily represented as an image
sink (out-of-phase source), and a wave combination that is algebraically
similar to the antisymmetric displacement whose Fourier transform is
given by (7.2.10). This latter calculation requires the cylindrical wave
representation (see (3.1.4) and (3.1.8))
∞ √ 2 2
eika r kρ dkρ
= J0 (kρ ρ)e− kρ −ka |x3 | " , (7.3.4)
r 0 kρ2 − ka2
"
in which kρ2 − ka2 lies in the fourth quadrant. With this decomposi-
tion in use for the dilatation, it is evident that the relevant solution of
(7.1.6) is
iω 2 1/2
∇ × u = J1 (kρ ρ)e−(kρ −iω/ν) |x3 | êφ ,
ν
whose solenoidal displacement is
iω 1/2 2 1/2
u = [ kρ2 − sgn(x3 )J1 (kρ ρ)êρ + kρ J0 (kρ ρ)ê3 ]e−(kρ −iω/ν) |x3 | .
ν
(7.3.5)
Thus, when the source defined by the potential (7.3.3) is placed at

(0, 0, −b) in a viscous fluid bounded by a rigid wall at x3 = 0, the total
displacement is, by use of (7.3.4) and (7.3.5),

√
k 2 m0 ∞ kρ J1 (kρ ρ) 2 2
u= a 2 " êρ + sgn(x3 + b)J0 (kρ ρ)ê3 e− kρ −ka |x3 +b|
4πk 0 2
kρ − ka 2
" √
kρ J1 (kρ ρ) 2 2
+ [B(kρ ) kρ2 − ka2 − 1] " êρ − J0 (kρ ρ)ê3 e− kρ −ka (b−x3 )
2
kρ − ka 2

kρ2 J0 (kρ ρ)
+ B(kρ ) − kρ J1 (kρ ρ)êρ + 2 ê3
(kρ − iω/ν)1/2
√ 2 2 2
1/2
× e−b kρ −ka e(kρ −iω/ν) x3 kρ dkρ , (7.3.6)
for x3 < 0, where

" −1
kρ2
B(kρ ) = 2 kρ2 − ka2 − . (7.3.7)
(kρ2 − iω/ν)1/2
Note that this denominator has the same form as the Wiener–Hopf kernel
in (7.2.12).
If, instead, there is an elastic solid in x3 > 0, with wavespeeds defined
by (1.1.21) and (1.1.22) and corresponding wavenumbers, kL , kT , the
displacement fields can be similarly written, in the fluid as
√
ka2 m0 ∞ kρ J1 (kρ ρ) 2 2
u= " êρ + sgn(x3 + b)J0 (kρ ρ)ê3 e− kρ −ka |x3 +b|
4πk 2 0 kρ2 − ka2
" √
kρ J1 (kρ ρ) 2 2
2 2
+ [B(kρ ) kρ − ka − 1] " êρ − J0 (kρ ρ)ê3 e− kρ −ka (b−x3 )
2
kρ − ka2

+ C(kρ )kρ − (kρ2 − iω/ν)1/2 J1 (kρ ρ)êρ + kρ J0 (kρ ρ)ê3
√ 2 2 2
1/2
× e−b kρ −ka e(kρ −iω/ν) x3 kρ dkρ , (7.3.8)
for x3 < 0, and in the elastic solid as

∞ !
E ka2 m0 − 2 −k2 x
kρ L 3
u = D(kρ ) kρ J1 (kρ ρ)êρ + kρ2 − kL
2
J0 (kρ ρ)ê3 e
4πk2 0

+ E(kρ )kρ (kρ2 − kT2 )1/2 J1 (kρ ρ)êρ + kρ J0 (kρ ρ)ê3
√
2 2 1/2 2 2
× e−(kρ −kT ) x3 e−b kρ −ka kρ dkρ , (7.3.9)
for x3 > 0. The functions B, C, D, E are determined by imposing conti-

nuity of displacements and stresses at x3 = 0. The matching of (7.3.8)
and (7.3.9) at x3 = 0 yields
2 − B(kρ )(kρ2 − ka2 )1/2 + C(kρ )kρ2 = D(kρ )(kρ2 − kL

2 1/2
) + E(kρ )kρ2
= F (kρ ), (7.3.10)
B(kρ ) − C(kρ )(kρ2 − iω/ν)1/2 = D(kρ ) + E(kρ )(kρ2 − kT2 )1/2

= G(kρ ), (7.3.11)
after writing the common interfacial displacement as

2
∞ √
ka m0 2 −k2
(u)x3 =0 = [G(kρ )kρ J1 (kρ ρ)êρ + F (kρ )J0 (kρ ρ)ê3 ] e−b kρ a
kρ dkρ .
4πk2 0
On noting that the parameter definitions and the assumed periodic-

ity allow the solid and fluid stress tensors, (1.1.3) and (7.1.1), to be
expressed similarly by
−2 −2
τij /ρs 2 kL kT
=ω ∂k uk δij − 2 (ij − ∂k uk δij ) ,
τij /ρf ka−2 ν/iω
the introduction of the functions F , G enables the two equations that

ensure stress continuity to be written concisely as

1 ρf ν ρf
D(kρ ) = 2 − kρ2 G(kρ ) + B(kρ ), (7.3.12)
kT2 ρs iω ρs

1 ρf ν ρf
E(kρ ) = 2 − F (kρ ) + C(kρ ). (7.3.13)
kT2 ρs iω ρs
7.4 Diffraction by an elastic sphere

Suppose that the damped acoustic plane wave with potential ϕ =
Aeika x3 /ika is incident on the elastic sphere, r < R. This alignment of
the axes yields axisymmetric displacement fields. In terms of spherical
Bessel functions and Legendre polynomials, the Rayleigh equation,
∞

eikr cos θ = (2n + 1)in jn (kr)Pn (cos θ), (7.4.1)
n=0
allows the incident wave to be written as

∞
jn (ka r)
ui = −A (2n+1)in+1 jn (ka r)Pn (cos θ)êr − Pn (cos θ) sin θêθ .
n=0
ka r
7.4 Diffraction by an elastic sphere 131
(1)
Note that jn , and similarly hn , is such that
#
π 2 n(n + 1)
jn (x) = Jn+1/2 (x), jn + jn + 1− jn = 0. (7.4.2)
2x x x2
Suitable axisymmetric solutions of (7.1.5) and (7.1.6) are
∞ (1)
Bn hn (ka r)
∇ · ur = ka A (2n + 1)in+1 −jn (ka R) + Pn (cos θ),
n=0
ka R h(1)
n (ka R)
(7.4.3)
∞
A
(1)
hn (r iω/ν)
∇ × ur = (2n + 1)in+1 Cn (1) Pn (cos θ) sin θeφ .
R n=1 hn (R iω/ν)
(7.4.4)
Following Davis and Nagem (2006), two distinct rearrangements of
the differential equation (7.4.2) enable the irrotational and solenoidal
reflected displacements to be identified from (7.4.3) and (7.4.4). Thus,
in r > R,
∞

ur = −A (2n + 1)in+1 − jn (ka R) + Bn ka R
n=0
(1) (1)
hn (ka r) hn (ka r)
× (1)
Pn (cos θ)êr − (1)
Pn (cos θ) sin θêθ
hn (ka R) ka rhn (ka R)
∞ (1)
n(n + 1)hn (r iω/ν)
+ AR (2n + 1)in+1 Cn (1) Pn (cos θ)êr
n=1 rhn (R iω/ν)

1 d $ (1) %
− (1) rhn (r iω/ν) Pn (cos θ) sin θêθ .
rhn (R iω/ν) dr
(7.4.5)
The displacement in the elastic solid, r < R, has a similar form,
expressed as
∞

uE = −AkL R (2n + 1)in+1 Dn
n=0

jn (kL r) jn (kL r)
× Pn (cos θ)êr − P (cos θ) sin θêθ
jn (kL R) kL rjn (kL R) n
∞
n(n + 1)jn (kT r)
+ AR (2n + 1)in+1 En Pn (cos θ)êr
n=1
rjn (kT R)

1 d
− {rjn (kT r)} Pn (cos θ) sin θêθ . (7.4.6)
rjn (kT R) dr
The coefficients Bn , Dn (n ≥ 0), Cn , En (n ≥ 1) are determined by

imposing continuity of displacements and stresses at r = R. Writing
the common interfacial displacement as

∞

u|r=R = AiF0 êr + A (2n + 1)in+1 Fn Pn (cos θ)êr + Gn Pn (cos θ) sin θêθ ,
n=1
the coefficients Fn and Gn follow, from matching of (7.4.5) and (7.4.6)

at r = R, as
(1)
i ka Rhn (ka R)
Fn = (1)
− Bn (1)
+ n(n + 1)Cn
(ka R)2 hn (ka R) hn (ka R)
kL Rjn (kL R)
= −Dn + n(n + 1)En , (7.4.7)
jn (kL R)
# (1)
iω hn (R iω/ν)
Gn = Bn − Cn 1 + R
ν h(1)
n (R iω/ν)

kT Rjn (kT R)
= Dn − En 1 + , (7.4.8)
jn (kT R)
for n ≥ 0, n ≥ 1 respectively. Mimicking the previous section, the intro-
duction of Fn , Gn enables the stress continuity conditions to be written
concisely as

1 ρf ν 1 ρf
Dn = 2 − [n(n + 1)Gn + 2Fn ] + Bn , (7.4.9)
kT2 ρs iω R2 ρs

1 ρf ν 1 ρf
En = 2 − (Gn + Fn ) + Cn , (7.4.10)
kT2 ρs iω R2 ρs
for n ≥ 0, n ≥ 1 respectively. B0 , D0 are determined by (7.4.7) and
(7.4.9), and the four equations are solved for Bn , Cn , Dn , En ; n ≥ 1.
This construction fully exploits the similar governing equations for
viscous acoustic and elastic disturbances, despite the diffusive nature of
vorticity. Both here and in §7.3, which complements §3.1, the solution
does not use the Helmholtz potentials. They are shown to be inappro-
priate to the half-plane problem discussed in §7.2 because, in contrast to
§2.2, the introduction of viscosity generates not only vorticity but also
a symmetric component of the gradient field.
8
Near-cut-off behavior in waveguides
Thus far we have seen many applications and examples of the diffraction
of waves by cracks and sharp edges, and the main interest is often in the
field far from the scatterer. We now turn our attention to another type
of wave interaction; here we are concerned with the channeling of wave
energy along guiding structures. Waveguides and the guiding of waves,
and thereby energy, through and along structures is a fundamental issue
in wave propagation, and arises often in applications. We begin with a
summary of guided waves in acoustics, as the equations and ideas are
simpler in that context, and then discuss guided elastic waves. We revisit
some of the topics discussed in Chapter 3, namely Rayleigh–Lamb modes
for straight waveguides. In practice many waveguides are not straight,
but are either bent or thicken, or are inhomogeneous, and so we turn our
attention to an asymptotic method that can be used in these situations.
8.1 Shear horizontal and acoustic waveguides

Many of the basic concepts of guided waves can be understood and seen
more readily in acoustics, and hence we consider this first. We begin with
a perfectly straight infinite waveguide of thickness 2h. It is convenient
to non-dimensionalize, so x̂ = hx, ŷ = hy and the hatted variables x̂, ŷ
are dimensional. Our field variable φ = φ(x, y) satisfies the Helmholtz
equation
φxx + φyy + (k0 h)2 φ = 0, −∞ < x < ∞, |y| < 1, (8.1.1)
with k0 = ω/c, ω being the frequency and c the wavespeed. The wave-
guide itself, of course, can have different boundary conditions upon the
walls; Dirichlet, φ = 0, or Neumann, φy = 0, are the usual choices. If φ is
the out-of-plane displacement and we are considering SH elasticity, then
133
134 Near-cut-off behavior in waveguides
these would correspond to clamped or traction-free conditions, respec-

tively. But it is important to realize that the Helmholtz equation arises
in a host of different applications, from acoustics and electromagnetism
to quantum mechanics, as well as in elasticity.
It is natural, given this waveguide, the Helmholtz equation, and the
boundary conditions, to ask what solutions exist. We treat both Dirichlet
and Neumann boundary conditions, and it is readily shown that modal
solutions can be found in the form
∞

sin nπ
ikn x 2 (y + 1)
φ= An e × , (8.1.2)
n=0 cos nπ
2 (y + 1)
(the upper/lower solutions being Dirichlet/Neumann) for waves propa-

gating to x = +∞, where each of the discrete modes in the summation
has a characteristic wavenumber, kn , satisfying a dispersion relation

kn = (k0 h)2 − (nπ/2)2 (8.1.3)
connecting it to the non-dimensional frequency k0 h = hω/c. The branch
cuts for the square root are chosen such that it is positive imaginary for
k0 h < nπ/2. For convenience we have written this as a single cosine or
sine, but expanding it reveals that the modes fall into two families: a
symmetric (in y) set for n even and an anti-symmetric set for n odd.
The dispersion curves from this dispersion relation, for the Neumann
case, are plotted in Fig. 8.1, from which we immediately notice sev-
eral things. First, all the modes except the lowest – it is conventional
to number the modes with increasing frequency – cut the frequency
axis at non-zero values: nπ/2. These frequencies are called the cut-off
20
15
kn
10
0
0 2 4 6 8 10 12 14 16 18 20
k0 h
Fig. 8.1. The dispersion curves relating frequency and wavenumber for the
straight waveguide with Neumann boundary conditions.
8.1 Shear horizontal and acoustic waveguides 135
frequencies for each mode; if the frequency is below that value, then kn
is complex and the solution decays exponentially in x, no longer propa-
gates, and thus cannot transport energy efficiently. As we approach the
cut-off frequencies from above, the wavenumbers kn tend to zero. Since
these are related to the wavelength Δ via kn = 2π/Δn the wavelength is
tending to infinity and thus these waves are ‘long.’ Next, we note that all
the dispersion curves are not straight lines, except for the lowest mode,
and this explains why the relation relating frequency to wavenumber
is called a dispersion curve: these modes are dispersive, i.e. the group
velocity, defined as
∂ω
cg = ,
∂k
is not equal to the phase velocity c = ω/k. Thus the energy and wave
crests move at different speeds. The slope of the curves, in this case,
is always positive, so cg > 0 and the group velocity is always posi-
tive. The lowest mode, for which n = 0, is different: it is absent in the
Dirichlet case, it simply corresponds to a plane wave in the x direc-
tion, it is non-dispersive, and it automatically satisfies the Neumann
boundary condition. Thus, even in this simple situation, the dispersion
curves have told us useful information encapsulated in a readily under-
stood graph; dispersion curves and relations are fundamental to guided
waves, as is the concept that the field can be broken down into discrete
modes.
All of these details are important when it comes to asymptotic
methods. If a waveguide is not perfectly straight, contains a defect, or
widens, we can no longer expect to separate variables and have a per-
fect modal structure. Nonetheless one would hope to use the knowledge
gained from studying modal solutions and dispersion curves to be of
value. For instance, if the waves are at very high frequencies and have
short wavelengths relative to geometric or material variations within the
waveguide, then, heuristically, one would hope that a local modal struc-
ture would persist and that its phase would gradually change as one
moved along the waveguide. This is the motivation behind WKBJ and
ray-like methods.
In the opposite regime, if the waves are very long and the local modes
are almost cut off, then one would hope that expanding around this local
situation would give valuable information, and an asymptotic theory
could be generated for long waves. In this chapter we shall consider the
latter situation, which is not usually covered in books.
8.2 Elastic waveguides

We consider an isotropic homogeneous elastic straight waveguide occu-
pying the region −h ≤ x2 ≤ h and |x1 | < ∞. The edges x2 = ±h of the
waveguide are taken to be stress-free. Before applying the stress-velocity
equations, dimensionless variables are introduced:
x̃i = xi /h, t̃ = cT t/h, ũi = ui /h, τ̃ij = τij /μ,

ω̃ = hω/cT ,
(8.2.1)
and in subsequent equations the tilde will be omitted for notational
convenience. There are two bulk wavespeeds in isotropic elasticity, of
shear/transverse √ waves cT and of compressional/longitudinal waves cL
(with cL /cT > 2); the parameter μ in (8.2.1) is the shear modulus. We
consider time-harmonic motion, so the multiplicative factor e−iωt , where
ω is the frequency, is considered understood and suppressed henceforth.
The non-dimensional governing equations of isotropic and homogeneous
elasticity (in two dimensions) are:
∂ 2 u1 ∂ 2 u1 ∂ 2 u2
γ −2 + + (γ −2
− 1) + ω 2 u1 = 0, (8.2.2)
∂x21 ∂x22 ∂x1 ∂x2
∂ 2 u2 2
−2 ∂ u2 −2 ∂ 2 u1
+ γ + (γ − 1) + ω 2 u2 = 0, (8.2.3)
∂x21 ∂x22 ∂x1 ∂x2
where u = (u1 , u2 ) are the displacements in the x1 , x2 directions, and

the ratio of the wavespeeds, γ, where γ = cT /cL , emerges as a key
parameter.
Solutions of (8.2.2), (8.2.3) are modal sums:

u1 ikn Bn ikn An
= cos αn x2 + sin αn x2
u2 n
αn An −αn Bn

βn Cn −βn Dn
+ cos βn x2 + sin βn x2 eikn x1 . (8.2.4)
−ikn Dn −ikn Cn

The parameters αn = ω 2 /γ −2 − kn2 , βn = ω 2 − kn2 , An , Bn , Cn , and
Dn are found by applying the boundary conditions at x2 = ±1, as are
the values of kn : the summations are carried out over all of the modes.
The following analysis is greatly simplified by considering the even (or
symmetric) modes, which have u1n ∼ cos ax2 , u2n ∼ sin ax2 , separately
from the odd (or antisymmetric) modes, which have u1n ∼ sin ax2 and
u2n ∼ cos ax2 , where a is constant and u1n , u2n are the first and second
components of displacement associated with the nth mode, respectively.
8.2 Elastic waveguides 137
Physically, excitations can be split into even and odd functions of x2

and considered separately, so there is no loss of generality in doing this.
8.2.1 Rayleigh–Lamb modes

We have already seen Rayleigh–Lamb modes in §3.2, with an emphasis
on the first couple of modes which approximate Rayleigh surface waves
at high frequencies. Here we are more interested in all modes, close to
cut-off, which is a different regime. For the even modes (8.2.4) becomes

u1 = (ikn Bn cos αn x2 + βn Cn cos βn x2 ) eikn x1 , (8.2.5)
n

u2 = (−αn Bn sin αn x2 − ikn Cn sin βn x2 ) eikn x1 . (8.2.6)
n
Expressions for the stresses in terms of the displacements are

∂u2 ∂u1 ∂u1 ∂u2
τ12 = + , τ22 = (γ −2 − 2) + γ −2 . (8.2.7)
∂x1 ∂x2 ∂x1 ∂x2
The expression for τ11 has been excluded here as it is not required.
Applying the boundary conditions, requiring τ12 and τ22 to be zero at
x2 = ±1, leads us to one of the Rayleigh–Lamb dispersion relations, as
shown in many textbooks on elasticity, for instance in Achenbach (1973):
tan αn (k 2 − β 2 )2
= − n2 n , (8.2.8)
tan βn 4kn αn βn
and
Bn = (kn2 − βn2 ) sin βn , Cn = 2ikn αn sin αn . (8.2.9)
This dispersion relation for even modes allows values of kn to be

calculated for any given value of ω (or vice-versa). It is a non-trivial
relationship, but is easily solved numerically using root-finding, general
purpose packages, or eigenvalue/spectral methods; either way, dispersion
curves (such as those of Fig. 8.2) readily emerge.
The even eigenmodes are calculated by substituting these expressions
into (8.2.5)–(8.2.6) and then using (8.2.7) for the stresses.
Similarly, for the odd modes the displacements are written as

u1 = (ikn An sin αn x2 − βn Dn sinn βn x2 ) eikn x1 , (8.2.10)
n

u2 = (αn An cos αn x2 − ikn Dn cos βn x2 ) eikn x1 , (8.2.11)
n
10
6
k
0
0 1 2 3 4 5 6 7 8 9 10
ω
Fig. 8.2. The Rayleigh–Lamb dispersion curves for aluminum (γ = 0.4593).

Even (odd) modes are indicated by solid (dashed) lines.
with
An = (kn2 − βn2 ) cos βn , Dn = −2ikn αn cos αn , (8.2.12)
and the dispersion relation is

tan αn 4k 2 αn βn
= − 2n 2 2 . (8.2.13)
tan βn (kn − βn )
The odd eigenmodes can be calculated by substituting these expressions
into (8.2.10)–(8.2.11) and then using (8.2.7).
It is worthwhile inspecting the dispersion curves before embarking
upon any analysis. The lowest two modes pass through the origin and
are asymptotic to a straight line; these are precisely the modes discussed
in Chapter 3, the straight line being associated with Rayleigh waves
which are non-dispersive. The other modes have distinct cut-offs, and
separate naturally into even and odd modes; near cut-off the modes are
dominated by either compressional or shear terms (this is not evident
from the curves alone). The second even mode turns back upon itself,
recalling that the group velocity is defined as the gradient of frequency
with respect to wavenumber, and thus has negative group velocity; the
picture is thus more complex than that for SH modes.
8.3 Long waves

We consider a two-dimensional waveguide of infinite extent in the x
direction which is bent in the vicinity of the point x = 0 and flattens
out at infinity. The geometry of the problem in Cartesian coordinates is
8.3 Long waves 139
η = +1
η
y σ
x α
0
η = −1
Fig. 8.3. The geometry of a bent waveguide, showing the new coordinates σ, η
along and normal to the centerline.
shown in Fig. 8.3. The waveguide thickness is constant and taken as 2,

in non-dimensional units. We employ a new coordinate system (σ, η)
which is more natural for this type of geometry. Here, 0 ≤ σ ≤ ∞ is the
arc length along the centerline, and −1 ≤ η ≤ 1 is the signed shortest
distance between the centerline and a point within the waveguide; it
takes negative values beneath the centerline. The shape of the centerline
is characterized by the angle α between the tangent to the centerline and
x axis. Therefore, we assume that the curvature function ασ , where the
subscript means the derivative with respect to σ, vanishes as σ → ∞, so
considering a bent waveguide that flattens out at infinity. The following
derivatives are handy when transforming to the new coordinates:
σx = κ cos α, σz = κ sin α, ηx = − sin α, ηz = cos α,
where
κ = (1 − ασ η)−1 .
Since the waveguide is assumed to be weakly bent, we introduce the

small parameter and a new variable ξ such that ξ = σ, which means
that the angle function α is a slowly varying function of the coordinate σ.
In terms of the new coordinate system (ξ, η) the Laplacian is
rewritten as
Δ = 2 κ2 ∂ξξ + ∂ηη + 3 κ3 αξξ η∂ξ − καξ ∂η , κ = (1 − αξ η)−1 .
Thus for a bent geometry within which the Helmholtz equation applies
(we treat SH waves first) the equation that one wishes to solve is (8.1.1),
which in the new coordinates is
Δφ + λ2 φ = 0, (8.3.1)
where λ2 = (ωh/c)2 , i.e. the squared non-dimensional frequency. This

Helmholtz equation then requires boundary conditions, as earlier we
assumed either Dirichlet or Neumann conditions; we take φ = 0 or

φη = 0 on η = ±1, respectively.
Long-wave theory explores the physical limit just as one approaches
the cut-off frequency from above, and thus it is natural to consider reg-
ular perturbation expansions of φ and the squared frequency λ2 . Posing
expansions
φ = φ0 + φ1 + 2 φ2 + · · · , λ2 = λ20 + λ21 + 2 λ22 + · · ·
leads to a hierarchy of equations:
φ0ηη + λ20 φ0 = 0, (8.3.2)
φ1ηη + λ20 φ1 = −λ21 φ0 + αξ φ0η , (8.3.3)
φ2ηη + λ20 φ2 = −φ0ξξ + αξ [φ1η + αξ ηφ0η ] − [λ22 φ0 + λ21 φ1 ], (8.3.4)
with the conditions on η = ±1 that φi = 0 or φiη = 0 for the Dirichlet
or the Neumann case, respectively (i = 1, 2, 3, . . .). The first of these,
(8.3.2), has the trivial solution

sin (λ0 (η + 1)) nπ
φ0 (ξ, η) = f0 (ξ) × , λ0 =
cos (λ0 (η + 1)) 2
(the upper/lower solutions being Dirichlet/Neumann, respectively). The

equations contain an as yet unknown function f0 (ξ), and the aim of
the expansion is to generate an equation that will determine this and
the frequency.
The second equation, (8.3.3), contains the unknown frequency λ21 ,
which is determined using a solvability condition: multiplying the right-
hand side of the equation by the homogeneous solution, i.e. φ0 , and
integrating from η = −1 to η = +1 with respect to η must yield zero –
this gives λ21 = 0. A little more algebra gives

αξ η sin((η + 1)λ0 ) αξ 0
φ1 (ξ, η) = f1 (ξ)+ f0 (ξ) × − f0 (ξ)×
2 cos((η + 1)λ0 ) 2λ0 sin((η + 1)λ0 ).
The final equation we consider, (8.3.4), can likewise be simplified using

a solvability condition in the same manner as earlier, and this gives an
equation to determine λ22 :

2
αξ2 1
f0ξξ + λ2 f0 + f0 × = 0. (8.3.5)
4 (−3)
This ordinary differential equation (ODE) in f0 (ξ) yields useful infor-
mation about the solutions near cut-off frequencies. If one knows the
8.3 Long waves 141
curvature function, αξ , then this ODE can be solved in either finite or

infinite domains to determine the correction, λ22 , to the cut-off frequency.
For instance, if the waveguide is straight, then αξ = 0 and the ODE
simply has solutions f0 (ξ) ∼ exp(iκξ); thus λ22 = κ2 and
nπ 2
λ2 = (k0 h)2 = + 2 κ2 .
2
With kn = κ one is led back to the straight-waveguide dispersion
relation (8.1.3).
If the waveguide has the shape of a circular annulus, i.e. with constant
curvature, then αξ = 1 and one may use a polar coordinate system (r, θ)
with a < r < b. The centerline, R, is given by R = (a+b)/2 and the half-
width, h, by 2h = (b−a), with r = R−hη and = h/R 1. The circular
annulus has, assuming an exp(ikθ) dependence, a dispersion relation
Jk (k0 a)Yk (k0 b) − Jk (k0 b)Yk (k0 a) = 0 (Dirichlet),
Jk (k0 a)Yk (k0 b) − Jk (k0 b)Yk (k0 a) = 0 (Neumann),
which relates the wavenumber k to frequency k0 h. These relations are
easily solved numerically, and long-wave theory gives
#
2 nπ 2
k = (k0 h)2 + −
4 2
in the Dirichlet case and
#
32 nπ 2
k = (k0 h)2 − −
4 2
for the Neumann case. These asymptotic approximations are compared
to the exact dispersion curves in Fig. 8.4. Also shown in Fig. 8.4 is
the straight-waveguide dispersion curve; notably, for a curved Dirichlet
waveguide the cut-off frequency is, relative to the straight-waveguide
curve, shifted to the left, with

k0 h = (nπ/2)2 − 2 /4,
and conversely for the Neumann waveguide it is shifted to the right, with

k0 h = (nπ/2)2 + 32 /4.
This has implications for modes in varying waveguides as there exists a
small window of frequencies for a locally curved waveguide where, in the
Dirichlet case, a mode is permitted but is promptly cut off if it enters
a straight region, the converse being true for the Neumann case. This
brings us to the topic of trapped modes. Before embarking on that, we
2.5
1.5
Dirichlet
k
1
Neumann
0.5
0
1.56 1.58 1.6 1.62 1.64 1.66
k0 h
Fig. 8.4. The n = 1 dispersion curve relating frequency and wavenumber for
an annular waveguide with Dirichlet or Neumann boundary conditions. The
parameter has been chosen rather large ( = 0.25) in order to allow some
difference to be seen. The dashed line is the curve for a straight waveguide.
The solid lines are calculated from the full dispersion relation and the solid
plus dots is from the long wave theory.
mention a generalization: the procedure has been illustrated here for a

waveguide of constant width with a localized change in curvature, but
similar ideas can be used to deal with localized thickness or material
variations.
The pedagogic SH/acoustic example we have treated in some detail is
quite simple, but nonetheless illustrates an important point: there is a
useful asymptotic regime that is readily accessed, limited not to low or
high frequencies but to near cut-off frequencies.
8.3.1 Trapped modes

The long-wave asymptotics have distilled all of the physics into a simple
ODE, and some quite general properties of the ODE can be deduced
using Sturm–Liouville theory and the properties of positive operators;
a useful reference, albeit from the perspective of integral equations, is
Porter and Stirling (1990). Let us consider an infinite waveguide that
has a bend localized at ξ = 0 and which straightens out at infinity, so
αξ → 0 as ξ → ±∞. We wish to investigate whether there are modes
trapped in the locality of the bend, i.e., mathematically, whether there
exist non-trivial eigensolutions that decay at infinity, i.e. f0 → 0, there.
If we consider the ODE in (8.3.5) for the Neumann case, we have
3
− f0ξξ + αξ2 f0 = λ22 f0 . (8.3.6)
4
8.3 Long waves 143
This has been rewritten slightly to emphasize the following point: if one
multiplies by f0 and then integrates with respect to ξ from −∞ to ∞,
then the left-hand side of (8.3.6) is positive (a positive operator) and
thus the eigenvalue λ22 is positive. At infinity we are then left with the
ODE
f0ξξ + λ22 f0 ∼ 0,
which has oscillatory solutions at infinity; thus there are no decaying
eigenfunctions. Thus, we immediately deduce that the Neumann bent
waveguide has no possibility of trapping.
Now we consider the Dirichlet ODE:
αξ2
− f0ξξ − f0 = λ22 f0 , (8.3.7)
4
which is very similar to the Neumann one aside from the curvature-
dependent term. This has now changed sign (and dropped a numerical
factor); this is important as we no longer have a positive operator, and
the eigenvalue could now be negative. Although this does not prove that
it is, it allows for the possibility that λ22 < 0, i.e. that an eigenvalue
could exist beneath the cut-off frequency for the straight waveguide and
thus that a mode (or modes) could be trapped. This quite general but
simple procedure is enlightening and immediately discounts the possibil-
ity of localized trapping at a bend in a Neumann waveguide. Numerical
investigation of the Dirichlet ODE demonstrates that there is indeed a
localized eigenfunction at the bend; see Fig. 8.5.
8.3.2 Inplane elasticity

Long-wave theory can also be applied to a bent waveguide for inplane
elasticity. The algebra is heavier, but the basic idea remains unchanged:
identify the cut-off frequencies, expand about them, and distill the essen-
tial physics into an ODE that is then easily studied. We will now derive
the analogous ODE to (8.3.5) for an elastic bent waveguide. The asymp-
totic scheme in elasticity becomes slightly more complicated owing to the
higher-order boundary conditions, and it is convenient to use displace-
ment potentials φ, ψ such that u = ∇φ + ∇ × (0, 0, ψ). The governing
equations are written as follows:
Δφ + γ 2 Λ2 φ = 0, Δψ + Λ2 ψ = 0, (8.3.8)

where γ = cT /cL = (1 − 2ν)/(2 − 2ν), Λ2 = (ω/cT )2 , and ν is the
Poisson ratio. We assume that the boundaries are traction-free and that
Frequency = 1.5702
10
0.8
8
0.6
f0
0.4
6 0.2
−50 0 50
4 ξ
0
−10 −5 0 5 10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Fig. 8.5. The solution in a Dirichlet waveguide for n = 1 and

αξ = (π/4)sech2 (ξ) (a locally bent bar) with = 0.25, excited by a line source
oscillating at λ = 1.5702, which is slightly less than π/2; the figure shows
the energy ω 2 |φ|2 /2 normalized to have a maximum of unity. The colorbar
shows the magnitude, in grayscale, of the function f0 shown in the inset. The
asymptotic scheme predicts the same λ0 , and in both computations strong
localization at the bend is observed.
τξη (ξ, ±1) = τηη (ξ, ±1) = 0,
where the τ are stresses. In terms of the potentials, these become
2κφξη + 22 κ2 αξ φξ − Δψ + 2ψηη = 0,

(γ −2 − 2)Δφ + 2φηη − 2κψξη − 22 κ2 αξ ψξ = 0,
evaluated on η = ±1. There are four sub-cases to consider, two shear-

dominated and two compressional-dominated cases, and these further
split into cases that to leading order are symmetric and anti-symmetric
about the centerline; we present one of the compressional cases and
just quote results for the others later. An inspection of the dispersion
relations in the limit as k → 0 shows that the cut-off frequencies occur at
π(2n − 1) πn π(2n − 1)
Λ= , , πn, .
2γ γ 2
These correspond to transverse resonances, where the waves do not prop-
agate but simply bounce back and forth across the waveguide. These also
follow from the leading-order asymptotics.
8.3 Long waves 145
For the compressional field, we assume the following asymptotic

expansion:
φ(ξ, η) = φ0 (ξ, η) + φ1 (ξ, η) + 2 φ2 (ξ, η) + · · · ,

ψ(ξ, η) = ψ1 (ξ, η) + 2 ψ2 (ξ, η) + · · · , (8.3.9)
2
Λ = Λ20 + Λ21 + 2
Λ22 + ··· ,
which leads to the hierarchy of equations for the components of the

expansions. Each level of the hierarchy represents an ODE and uses the
result obtained in the previous step.
The leading-order calculation leads to
π(2n − 1)
φ0 (ξ, η) = f0 (ξ) cos[γΛ0 η], Λ0 = .
2γ
The function f0 (ξ) is unknown for the moment, and the Λ0 are the cut-off
frequencies in a straight waveguide. At the next order the compressional
piece has the solution
αξ αξ f0
φ1 (ξ, η) = f0 η cos[γΛ0 η] − 2 sin[γΛ0 η] + f1 cos[γΛ0 η],
2 Λ0
where f1 is unknown and ultimately not required. The shear piece,
ψ1ηη + Λ20 ψ1 = 0,
is more interesting, with boundary conditions
Λ20 ψ1 + 2ψ1ηη + 2φ0ξη = 0, η = ±1.
Solving this problem, we obtain the equation for ψ1 :

f0ξ γ sin(γΛ0 )
ψ1 (ξ, η) = −2 sin(Λ0 η).
Λ0 sin Λ0
The equation of order 2 for φ2 is
φ2ηη + γ 2 Λ20 φ2 = −φ0ξξ + ηαξ2 φ0η + αξ φ1η − γ 2 Λ22 φ0 , (8.3.10)
with boundary condition
2φ2ηη − 2ψ1ξη − (1 − 2γ 2 )Λ20 φ2 = 0, η = ±1.
Solving the equation for φ2 and applying the boundary conditions

leads to

1
C (L,s) f0ξξ + 4 − 2 αξ2 f0 = Λ22 f0 , (8.3.11)
4γ
where

(L,s) 1 8 cot Λ0
C =− 2 + . (8.3.12)
γ Λ0
The notation (L,s) denotes that it is the coefficient associated with the
longitudinal (L, i.e. compression) piece and that it is symmetric (s).
This is clearly a similar ODE to (8.3.5), and again general properties of
it can be deduced. In particular, if γ > 1/4 (which is true for realistic
media) then one has a positive operator if C (L,s) is negative, and thus no
trapping. Conversely, if C (L,s) is negative then there is possibly trapping.
Examination of the symmetric Rayleigh–Lamb modes shows that the
group velocity cg is
C (L,s)
cg ∼ − k,
Λ0
and thus localized trapping near the bend is only possible for modes with
negative group velocity; this returns us to the dispersion curves shown
in Fig. 8.2 and the observation that the second even mode has negative
group velocity.
This theory is as accurate in its description of near-cut-off phenomena
as that shown for the SH wave/acoustic example. Several examples are
found in the literature (Kaplunov et al., 2005; Gridin et al., 2005). The
presentation here follows Gridin et al. (2005), where the other ODEs for
antisymmetry and shear dominance can be found.
Appendix A
Asymptotic expansions
Applications of asymptotic methods and asymptotic expansions in this

book are mainly to the evaluation of integrals, particularly Fourier-like
integrals. We begin with a brief review of the notation required.
A.1 Notation for asymptotics

Most, if not all, of the mathematical definitions and notations are stan-
dard. Nevertheless, the reader is reminded of the following, which are
taken, with very little change, from Bleistein and Handelsman (1975,
pp. 14–21). The definitions generally apply in an open angular sector in
the complex plane defined as follows.
The angular sector Sαβ : Consider an open, two-dimensional region

R in the complex plane; let z ∈ R and z0 ∈ R̄ (the overbar indicates
closure). Define the open sectoral region Sαβ as
Sαβ := {z | 0 < |z − z0 | < ρ, α < arg(z − z0 ) < β}.
It is sometimes sufficient to consider simply a neighborhood of z = z0

without needing to specify α or β.
The main tools of asymptotics are the order-of-magnitude nota-
tion, o and O, with their associated symbols for comparison, ∼, and
equality, =.
o[f (z)] An order of magnitude notation: if limz→z0 g(z)/f (z) = 0 as

z → z0 in Sαβ , then one writes g(z) = o[f (z)].
O[f (z)] An order of magnitude of f (z): one writes g(z) = O[f (z)] if
there exists a real non-negative constant C such that, as z → z0
in Sαβ , |g(z)| ≤ C|f (z)| everywhere in the sector.
147
148 Asymptotic expansions
∼ This is used to indicate an asymptotic expansion of a function, in

the angular sector Sαβ , when no explicit estimation of the error
is given. For example,
g(z) ∼ a0 f0 (z), z → z0 ,
means that f0 (z) is the first term of an asymptotic expansion of

g(z) for z as z → z0 in Sαβ . More generally, one writes

g(z) ∼ an fn (z), z → z0 ,
n≥0
where the {fn } form an asymptotic sequence (see below).

= In the context of using an asymptotic approximation, this is used
when an order of magnitude of the error is given. For example
g(z) = a0 f0 (z) + O(f1 ), z → z0 ,
or
g(z) = a0 f0 (z) + a1 f1 (z) + o(f1 ), z → z0 ,
where again the {fn } form an asymptotic sequence in the

angular sector Sαβ .
Types of asymptotic expansions

It is useful to distinguish several distinct varieties of asymptotic
expansions that are encountered in practice.
Asymptotic sequence: The sequence {fn (z)}n≥0 is an asymptotic

sequence as z → z0 in Sαβ , if each fn (z) is continuous in R and
fn+1 (z) = o[fn (z)], n ≥ 0,
uniformly as z → z0 .
Asymptotic expansion of Poincaré type: The formal series

n≥0 an fn (z) is an asymptotic expansion of g(z), as z → z0 in Sαβ ,
with respect to the asymptotic sequence {fn }, if the conditions

N
g(z) = an fn (z) + o[fN (z)]
n=0
A.2 Abelian theorems for Fourier transforms 149
hold uniformly as z → z0 in Sαβ . When the number N is unimportant,

one may also write

g(z) ∼ an fn (z).
n≥0
Asymptotic expansions not of the Poincaré type are briefly dis-

cussed next.
Asymptotic expansion based on an auxiliary asymptotic

sequence: The formal series n≥0 an Fn (z) is an asymptotic expan-
sion of g(z), with respect to the auxiliary asymptotic sequence {fn (z)},
if the conditions

N
g(z) = an Fn (z) + o[fN (z)]
n=0
hold uniformly as z → z0 in Sαβ . When the number N is unimportant,

one may also write

g(z) ∼ an Fn (z), z → z0 , {fn }.
n≥0
The auxiliary sequence {fn } may not be explicitly indicated. If the

sequence {fn } is to be useful,
Fn = O(fn ), Fn = o(fn ), z → z0 , z ∈ Sαβ .
Uniform asymptotic approximation: A uniform asymptotic

approximation to an integral I(x, κ, α) is one that remains accurate
for all values of the parameter α in its domain as κ → κ0 in Sαβ (in
practice κ is usually real and κ → ∞).
A.2 Abelian theorems for Fourier transforms

The Fourier transform pair used in this book is
∞
∗ 1 ∞∗
f (k) = f (x)e−ikx dx, f (x) = f (k)eikx dk. (A.1)
−∞ 2π −∞
This pair can be interpreted and used in a slightly more general form
(Titchmarsh, 1948, pp. 4–6):
∞, 0
∗ −ikx 1 ∞±ia ∗
f± (k) = f (x)e dx, f± (x) = f± (k)eikx dk, (A.2)
0, −∞ 2π −∞±ia
where a is a sufficiently large positive number. Thus

∗
f (k) = ∗f− (k) + ∗f+ (k), f (x) = f− (x) + f+ (x). (A.3)
The function f (x) is said to have the image ∗f (k); similarly, the functions
f± (x) are said to have images ∗f± (k).
Consider the pair {f+ (x),∗ f+ (k)}. Theorems which relate the known
asymptotic behavior of f+ (x) to the asymptotic behavior of its image
are called Abelian theorems; those that relate the known asymptotic
behavior of the image ∗f+ (k) to that of the function are called Tauberian
theorems, and are harder to establish. The following Abelian theorem
(Titchmarsh, 1948, pp. 173–174) is sufficient for most purposes.
Abelian theorem for the pair {f+ (x),∗ f+ (k)}: Let f+ (x) = xα g(x),
where α ∈ (0, 1) and g(x) is of bounded variation in (0, ∞). Then
∗
f+ (k) ∼ g(0+ ) (−α)! k α−1 e−iαπ/2 e−iπ/2 , k → ∞;
∗
f+ (k) ∼ g(∞+ ) (−α)! k α−1 e−iαπ/2 e−iπ/2 , k → 0+ . (A.4)
The function (−α)! := Γ(1 − α), where Γ(z) is the gamma function. An
extensive study of Abelian and Tauberian theorems as they relate to the
pair {f+ (x),∗ f+ (k)} is given in Doetsch (1974, chapters 32–37).
A.3 The method of steepest descents

Integrals of the following two forms frequently arise in the study of linear
waves. The first is
I(κ) = f (z)eκq(z) dz, (A.5)
C
where f (z) and q(z) are functions of the complex variable z, having such
properties that the integral is convergent, and C is a contour beginning
and ending at −∞ ei and ∞ eiζ , respectively, where , ζ ∈ (−π, π] with
|ζ| < π/2 < ||. The second is
β
I(κ) = f (x)eiκp(x) dx, (A.6)
α
where f (z) and p(z) are functions of the complex variable z, and p(x)
is real on the real axis. In both cases κ is real, positive, and large.
These integrals can be asymptotically approximated using one of three
interrelated methods: the method of steepest descents, the saddle-point
method, and the method of stationary phase.
A.3 The method of steepest descents 151
To advance the discussion, the terms ‘isolated, first-order saddle

point,’ ‘stationary-phase point,’ and ‘steepest-descents contour’ need to
be defined. In doing so, the notation dnz q = dn q/dz n is adopted.
Isolated, first-order saddle point: zs is an isolated, first-order saddle

point of q(z) if dz q = 0 at z = zs , but dnz q = 0, n ≥ 2 at z = zs . q(z) is
assumed to be an analytic function in a region Q containing zs .
Isolated, first-order stationary-phase point: xs is an isolated, first-

order stationary phase point of p(x) if dx p = 0 at x = xs , but dnx p =
0, n ≥ 2 at x = xs .
The steepest-descents contour: Assume that q(z) has an isolated

first-order saddle point at z = zs . The contour, passing through zs ,
along which Im(q) is constant and along which Re(q) decreases most
rapidly is the steepest-descents contour Cs . More explicitly stated, Cs is
the curve defined by
Re[q(z)] ≤ Re[q(zs )], Im[q(z)] = Im[q(zs )]. (A.7)
The definition is readily extended to the case of higher-order saddle

points.
The method of steepest descents is one in which the steepest-descents
contour Cs is determined throughout the complex plane, and is usually
used if more than one term of the asymptotic expansion is to be deter-
mined. Knowledge of the steepest-descents contour also permits one to
accurately take account of possible contributions from poles or branch
points. The saddle-point method is one in which the contour of integra-
tion is distorted to be near Cs in a neighborhood of zs , and is usually
used if only the leading-order, asymptotic contribution is to be deter-
mined. Possible contributions from poles or branch points are not always
taken account of accurately. This is more or less the distinction made by
Copson (1971, p. 64). Because knowing Cs is very useful when interpret-
ing the meaning of an integral, even when only the leading-order term
of its asymptotic expansion can be readily calculated, the method of
steepest descents will always be used. The method of stationary phase,
at least in this book, is used to approximate asymptotically integrals of
the form (A.6) and may include contributions both from the end points
and from the stationary point (Copson, 1971, pp. 27–35).
The method of steepest descents: Consider (A.5). Assume q(z) has

an isolated first-order saddle point zs in some region R where q(z) is
analytic. Moreover, assume f (z) is analytic in R containing zs and
f (zs ) = 0. And, let there be real constants K and b such that
|f (z)| < Keb|q(z)−q(zs )| , |z| → ∞,
in sectors of the complex plane where Cs begins and ends. Then1
(−2π)1/2
I(κ) = f (zs ) eκq(zs ) + O(κ−3/2 ), κ → ∞. (A.8)
[κ d2z q(zs )]1/2
The argument of the square-root term is identical to that of dz at zs .

The special case q = ir with r a real function of the complex variable z
simplifies (A.8), giving
(2π)1/2
I(κ) = f (zs )eiκr(zs ) e±iπ/4 + O(κ−3/2 ), κ → ∞. (A.9)
[κ |d2z r(zs )|]1/2
The ± are chosen as d2z r(zs ) is positive or negative.
Notes on the method of steepest descents.
(1) Harris (2001, pp. 91–95), following Felsen and Marcuvitz (1994,
pp. 382–386), constructs (A.8) by first deforming C to Cs , picking up
whatever pole and branch-point contributions may be present, and
then using the mapping
s2 = q(zs ) − q(z).
This has the effect of converting the integral along Cs into one which
can be asymptotically approximated by Watson’s lemma (Copson,
1971, pp. 49–50). The condition placed on f (z) as |z| → ∞ near
Cs has been chosen to allow a straightforward application of this
lemma.
(2) The parameter κ has been assumed real. With care, (A.8) and (A.9)
can be analytically continued to an angular sector of the complex κ
plane.
(3) Further, setting q = ir, where r is a real function of z, leads to an
integral very similar to (A.6), considered next.
1 Equation (5.60), p. 93 of Harris (2001) should not have an i before the κ.
A.3 The method of steepest descents 153
The method of stationary phase, including end-point contri-

butions: Consider (A.6). f (z) and p(z) are analytic functions of the
complex variable z in a region R containing the segment of the real axis
x ∈ [α, β]; p(x) is real on the real axis and has an isolated first-order
stationary point xs ∈ (α, β). Then
(2π)1/2
I(κ) = f (zs )eiκp(zs ) e±iπ/4
[κ |d2z p(xs )|]1/2

1 f (β) iκp(β) f (α) iκp(α)
+ e − e + O(κ−3/2 ), κ → ∞.
iκ dx f (β) dx f (α)
(A.10)
The ± are chosen as d2z p(xs ) is positive or negative.
Notes on the method of stationary phase.

(1) Copson (1971, pp. 29–34) gives a proof of this result without
reference to the method of steepest descents.
(2) The similarity between the first term of (A.10) and (A.9) suggests
the two methods are closely related. The method of stationary phase
is ideal when the interval of integration is finite and along the real
axis. However, as previously indicated, when the contour integra-
tion is infinite and may lie in the complex plane, then the method
of steepest descents is the more accurate, as one can be confident
that pole and branch point contributions have been accurately taken
account of.
(3) Note that if no stationary point appears in x ∈ [α, β] then the leading
contributions to the integral come from the end points.
Appendix B
Some special functions
Special functions of one sort or another arise frequently, often in the

course of seeking uniform asymptotic approximations to more compli-
cated functions. The purpose of this appendix is to collect the formulas
or representations used in the text in one place. Magnus et al. (1966)
is a well-organized handbook giving most properties of these functions.
Borovikov (1994) describes many of the special functions arising in the
uniform asymptotic approximation of integrals describing radiation and
diffraction. The internet site NIST (accessed August 2008) is also a very
useful source of information.
B.1 Hankel functions

The following facts are taken from Magnus et al. (1966); the section
numbers refer to this handbook.
(1)
(1) From §3.1.2: The principal branch for the Hankel functions H0 (z)
(2)
and H0 (z) is defined by arg(z) ∈ (−π, π). The connections among
the branches are
(1) (2) (2) (1)
H0 (zeiπ ) = −H0 (z), H0 (ze−iπ ) = −H0 (z).
Hankel functions are used in several places; however, their branch

cuts are not always explicitly shown in the diagrams of the complex
plane.
(2) From §3.6.4: Sommerfeld’s integral representations for the Hankel
functions are

(1) 1
H0 (z) = eiz cos μ dμ (B.1)
π C1
154
B.2 Fresnel integral 155
and

(2) 1
H0 (z) = e−iz cos μ dμ. (B.2)
π C2
The contour C1 starts at −η + i∞ and ends at η − i∞; the contour C2

starts at η − i∞ and ends at η + i∞, where η is any number between
0 and π; and arg(z) ∈ (−η, π−η). These two integral representations
can be combined to show that
2π
1
J0 (z) = eiz cos μ dμ, (B.3)
2π 0
where arg(z) ∈ (0, π), a result that can be arrived at in several
other ways.
(3) From §3.14.1: Asymptotic expansions for the Hankel functions are
(1),(2)
H0 (z) ∼ (πz/2)−1/2 e±i(z−π/4) , |z| → ∞, (B.4)
where, for the (1) function arg z ∈ (−π, 2π), and for the (2) function
arg z ∈ (−2π, π).
One additional fact is of use. There is a Parseval’s relation between

functions and their Hankel transforms (Sneddon, 1951, pp. 59, 60). The
Hankel transform pair are given as
∞
∗
f (u) = yf (y)J0 (uy)dy,
0
∞
f (x) = u ∗f (u)J0 (xu)du;
0
and the Parseval’s relation is

∞ ∞
xf (x)g(x)dx = u ∗f (u) ∗g(u)du. (B.5)
0 0
B.2 Fresnel integral

In this book the Fresnel integral F (z) is defined as
∞
2
F (z) := eiξ dξ, (B.6)
z
where | arg(z)| ∈ (0, π/2) as ξ → ∞. Note that this integral is, apart
from its normalization, usually named the complementary error function
(Magnus et al., 1966, §9.2.3). This definition is used in agreement with
156 Some special functions
its use in Born and Wolf (1999, p. 647). Much use is made of the following
property of the Fresnel integral:
F (z) + F (−z) = π 1/2 eiπ/4 . (B.7)
Defining the related function
2
G(z) := e−iz F (z),
the asymptotic behavior of F (z), as |z| → ∞, can be expressed as
i
G(z) ∼ + O(z −3 ), arg(z) ∈ (−π/2, π/2),
2z (B.8)
1/2 iπ/4 i −3
∼π e + + O(z ), arg(z) ∈ (π/2, 3π/2).
2z
The presence of different expansions in different sectors of the z plane is
an example of the Stokes phenomenon. Moreover, as z → 0,
π 1/2 eiπ/4
F (z) ∼ − z + O(z 3 ). (B.9)
2
References
Abrahams, I. D. (2000). The application of Padé approximants to Wiener–

Hopf factorization. IMA J. Appl. Math., 65: 257–281.
Achenbach, J. D. (1973). Wave Propagation in Elastic Solids. North-Holland,
Amsterdam.
Achenbach, J. D. (2003). Reciprocity in Elastodynamics. Cambridge University
Press, New York.
Achenbach, J. D., Gautesen, A. K. and McMaken, H. (1982). Ray Methods for
Waves in Elastic Solids. Pitman, Boston.
Aki, K. and Richards, P. G. (2002). Quantitative Seismology. University
Science Books, Sausalito, CA, second edition.
Atkin, R. J. and Fox, N. (1980). An Introduction to the Theory of Elasticity,
volume 1. Longman, London, second edition.
Auld, B. A. (1979). General electromechanical reciprocity relations applied
to the calculation of elastic wave scattering coefficients. Wave Motion,
1: 3–10.
Auld, B. A. (1990a). Acoustic Fields and Waves in Solids, volume 1. Krieger,
Malabar, FL, second edition.
Auld, B. A. (1990b). Acoustic Fields and Waves in Solids, volume 2. Krieger,
Malabar, FL, second edition.
Babic̆, V. M. and Buldyrev, V. S. (1991). Short-Wavelength Diffraction Theory.
Springer-Verlag, Berlin.
Baker, B. B. and Copson, E. T. (1987). The Mathematical Theory of Huygens’
Principle. Chelsea, New York, third edition.
Batchelor, G. K. (1967). An Introduction to Fluid Dynamics. Cambridge
University Press, Cambridge, U.K.
Bell, J. C. (1979). Stresses from arbitrary loads on a circular crack. Int.
J. Fracture, 15: 85–104.
Besserer, H. and Malischewsky, P. G. (2004). Mode series expansions at vertical
boundaries in elastic waveguides. Wave Motion, 39: 41–59.
Biryukov, S. V., Gulyaev, Yu. V., Krylov, V. V. and Plesky, V. P. (1995).
Surface Acoustic Waves in Inhomogeneous Media. Springer-Verlag,
Berlin.
Bleistein, N. and Handelsman, R. A. (1975). Asymptotic Expansions of Inte-
grals. Holt, Rinehart and Winston, New York.
Block, G., Harris, J. G. and Hayat, T. (2000). Measurement models for ultra-
sonic nondestructive evaluation. IEEE Trans. Ultrason. Ferroelectr. Freq.
Control, 47: 604–611.
157
158 References
Born, M. and Wolf, E. (1999). Principles of Optics. Cambridge University

Press, Cambridge, U.K., seventh (expanded) edition.
Borovikov, V. A. (1994). Uniform Stationary Phase Method. The Institution
of Electrical Engineers, London.
Borovikov, V. A. and Kinber, B. Ye. (1994). Geometrical Theory of Diffraction.
The Institution of Electrical Engineers, London.
Boström, A. (2003). Review of hypersingular integral equation method for
crack scattering and application to modeling of ultrasonic nondestructive
evaluation. Appl. Mech. Rev., 56: 383–405.
Boström, A. and Olsson, P. (1987). Scattering of elastic waves by non-planar
cracks. Wave Motion, 9: 61–76.
Bouwkamp, C. J. (1946). A contribution to the theory of acoustic radiation.
Philips Res. Rep., 1: 251–277.
Bouwkamp, C. J. (1954). Diffraction theory. Rep. Prog. Phys., 17: 35–100.
Brekhovskikh, L. M. and Godin, O. A. (1999). Acoustics of Layered Media,
volume 2. Springer-Verlag, Berlin, second edition.
Brekhovskikh, L. M. and Goncharov, V. (1985). Mechanics of Continua and
Wave Dynamics. Springer-Verlag, Berlin.
Briggs, A. (1992). Acoustic Microscopy. Oxford University Press, New York.
Chako, N. (1965). Asymptotic expansions of double and multiple integrals
occurring in diffraction theory. J. Inst. Math. Appl., 1: 372–422.
Clemmow, P. C. (1966). The Plane Wave Spectrum Representation of Electro-
magnetic Waves. Pergamon, Oxford.
Collin, R. E. (1991). Field Theory of Guided Waves. IEEE Press, New York,
second edition.
Comninou, M. and Dundurs, J. (1977). Reflexion and refraction of elastic
waves in the presence of separation. Proc. R. Soc. A, 356: 509–528.
Copson, E. T. (1971). Asymptotic Expansions. Cambridge University Press,
Cambridge, U.K.
Davis, A. M. J. and Nagem, R. J. (2002). Acoustic diffraction by a half-plane
in a viscous fluid medium. J. Acoust. Soc. Am., 112: 1288–1296.
Davis, A. M. J. and Nagem, R. J. (2006). Curle’s equation and acoustic
scattering by a sphere. J. Acoust. Soc. Am., 119: 2018–2026.
de Bruijn, N. G. (1970). Asymptotic Methods in Analysis. North-Holland,
Amsterdam.
Doetsch, G. (1974). Introduction to the Theory and Application of the Laplace
Transformation. Springer-Verlag, New York. Translated from the German
by W. Nader.
Eringen, A. C. and Şuhubi, E. S. (1975). Elastodynamics, volume 2. Academic
Press, New York.
L. B. Felsen, and G. A. Deschamps editors. (1974). Special Issue on Rays and
Beams, volume 62 of Proc. IEEE.
Felsen, L. B. and Marcuvitz, N. (1994). Radiation and Scattering of Waves.
IEEE and Oxford University Press, New York.
Folguera, A. and Harris, J. G. (1998). Propagation in a slowly varying wave-
guide. In J. A. DeSanto, editor, Mathematical and Numerical Aspects of
Wave Propagation, pages 434–436. SIAM, Philadelphia.
Friedman, B. (1956). Principles and Techniques of Applied Mathematics.
Wiley, New York.
References 159
Glushkov, Y. V. and Glushkova, N. V. (1996). Diffraction of elastic waves by

three-dimensional cracks of arbitrary shape in a plane. J. Appl. Math.
Mech. (PMM), 60: 277–283.
Gniadek, K. and Petykiewicz, J. (1971). Applications of optical methods in
diffraction theory of elastic waves. In E. Wolf, editor, Progress in Optics,
volume 9, pages 281–310. North Holland, Amsterdam.
Graff, K. F. (1991). Wave Motion in Elastic Solids. Dover, New York (reprint;
first published in 1975).
Greenspan, M. (1979). Piston radiator: some extensions of the theory.
J. Acoust. Soc. Am., 65: 608–621.
Gridin, D., Craster, R. V. and Adamou, A. T. I. (2005). Trapped modes in
curved elastic plates. Proc. R. Soc. A, 461: 1181–1197.
Hansen, R. C., editor. (1981). Geometrical Theory of Diffraction, IEEE Press
Selected Reprint Series. IEEE Press, New York.
Harris, J. G. (1987). Edge diffraction of a compressional beam. J. Acoust. Soc.
Am., 82: 635–646.
Harris, J. G. (1997). Modeling scanning acoustic imaging of defects at solid
interfaces. In G. Chavent, G. Papanicolaou, P. Sacks, and W. W. Symes,
editors, Inverse Problems in Wave Propagation, pages 237–257. Springer-
Verlag, New York.
Harris, J. G. (2001). Linear Elastic Waves. Cambridge University Press, New
York. A list of errors is maintained at http://www.diffractedwave.com.
Harris, J. G. and Block, G. (2005). The coupling of elastic, surface-wave modes
by a slow, interfacial inclusion. Proc. R. Soc. A, 461: 3765–3783.
Herrera, I. and Spence, D. A. (1981). Framework of biorthogonal series. Proc.
Nat. Acad. Sci. Am., 78: 7240–7244.
Hudson, J. A. (1980). The Excitation and Propagation of Elastic Waves.
Cambridge University Press, Cambridge, U.K.
James, G. L. (1980). Geometrical Theory of Diffraction for Electromagnetic
Waves. Peter Peregrinus, London, second edition.
Jull, E. V. (1981). Aperture Antennas and Diffraction Theory. IEE and Peter
Peregrinus, London.
Kaplunov, J. D., Rogerson, G. A. and Tovstik, P. E. (2005). Localized vibration
in elastic structures with slowly varying thickness. Quart. J. Mech. Appl.
Math., 58: 645–664.
Keller, J. B. (1957). Diffraction by an aperture. J. Appl. Phys., 28: 426–444.
Keller, J. B. (1958). A geometrical theory of diffraction. In L. M. Graves,
editor, Calculus of Variations and its Applications, pages 27–52. AMS,
Providence, RI.
King, L. V. (1934). On the acoustic radiation field of the piezoelectric oscillator
and the effect of viscosity on the transmission. Can. J. Res., 11: 135–155.
Kirrmann, P. (1995). On the completeness of Lamb modes. J. Elast.,
37: 39–69.
Krenk, S. (1979). A circular crack under asymmetric loads and some related
integral equations. J. Appl. Mech., 46: 821–826.
Kreyszig, E. (1975). Introduction to Differential Geometry and Riemannian
Geometry. University of Toronto Press, Toronto.
Lamb, H. (1906). On Sommerfeld’s diffraction problem; and on reflection by
a parabolic mirror. Proc. Lond. Math. Soc., 2: 190–203.
Levine, H. (1978). Unidirectional Wave Motions. North-Holland, Amsterdam.
160 References
Liang, K. K., Kino, G. S. and Khuri-Yakub, B. T. (1985). Material charac-

terization by the inversion of v(z). IEEE Trans. Sonics Ultrason., SU
32: 213–224.
Lighthill, M. J. (1965). Group velocity. J. Inst. Math. Appl., 1: 1–28.
Lighthill, M. J. (1978a). Fourier Analysis and Generalized Functions. Cam-
bridge University Press, Cambridge, U.K.
Lighthill, M. J. (1978b). Waves in Fluids. Cambridge University Press,
Cambridge, U.K.
Magnus, W., Oberhettinger, F. and Soni, R. P. (1966). Formulas and Theorems
for the Special Functions of Mathematical Physics. Springer-Verlag, New
York, third (enlarged) edition.
Malischewsky, P. (1987). Surface Waves and Discontinuities. Akademie-
Verlag, Berlin.
Martin, P. A. (1996). Mapping flat cracks onto penny-shaped cracks, with
applications to somewhat circular tensile cracks. Quart. Appl. Math.,
54: 663–675.
Martin, P. A. (1998). On potential flow past wrinkled discs. Proc. R. Soc. A,
454: 2209–2221.
Martin, P. A. (2001). On wrinkled penny-shaped cracks. J. Mech. Phys. Solids,
49: 1481–1495.
Martin, P. A. (2006). Multiple Scattering: Interaction of Time-Harmonic
Waves with N Scatterers. Cambridge University Press, Cambridge, U.K.
Maupin, V. (1988). Surface waves across 2-d structures: a method based on
coupled local modes. Geophys. J., 93: 173–185.
Mikhas’kiv, V. V. and Butrak, I. O. (2006). Stress concentration around a
spheroidal crack caused by a harmonic wave incident at an arbitrary angle.
Int. Appl. Mech., 42: 61–66.
Miklowitz, J. (1978). Elastic Waves and Waveguides. North-Holland, Amster-
dam.
Miles, J. W. (1971). Scattering by a spherical cap. J. Acoust. Soc. Am.,
50: 892–903.
Morse, P. M. and Ingard, K. U. (1968). Theoretical Acoustics. McGraw-Hill,
New York.
Naze Tjøtta, J. and Tjøtta, S. (1980). An analytical model for the nearfield
of a baffled piston transducer. J. Acoust. Soc. Am., 68: 334–339.
Nieto-Vesperinas, M. (1991). Scattering and Diffraction in Physical Optics.
Wiley-Interscience, New York.
NIST, (2008). Digital library of mathematical functions. http://dlmf.nist.gov,
accessed August. This is an ongoing project.
Noble, B. (1988). Methods Based on the Wiener-Hopf Technique. Chelsea, New
York, second edition.
Norris, A. N. (2008). Faxen relations in solids: a generalized approach to
particle motion in elasticity and viscoelasticity. J. Acoust. Soc. Am.,
123: 99–108.
Oestreicher, H. L. (1951). Field and impedance of an oscillating sphere in a
viscoelastic medium with an application to biophysics. J. Acoust. Soc.
Am., 23: 707–714.
Osipov, A. V. and Norris, A. N. (1999). The Malyuzhinets theory for scattering
from wedge boundaries: a review. Wave Motion, 29: 313–340.
Oughstun, K. E. editor (1991). Selected Papers on Scalar Wave Diffraction,
volume 51. SPIE, Bellingham, WA.
References 161
Phillips, H. B. (1933). Vector Analysis. Wiley, New York.

Pierce, A. D. (1981). Acoustics. McGraw-Hill, New York.
Porter, D. and Stirling, D. S. G. (1990). Integral Equations. Cambridge Uni-
versity Press, Cambridge, U.K.
Poruchikov, V. B. (1993). Methods of the Classical Theory of
Elastodynamics. Springer-Verlag, Berlin. Translated from the Russian by
V. A. Khokhryakov and G. P. Groshev.
Pott, J. and Harris, J. G. (1984). Scattering of an acoustic Gaussian beam
from a fluid-solid interface. J. Acoust. Soc. Am., 76: 1829–1838.
Rebinsky, D. A. (1991). Asymptotic description of the acoustic microscopy
of a surface-breaking crack. PhD thesis, University of Illinois, Urbana-
Champaign.
Rebinsky, D. A. and Harris, J. G. (1992a). The acoustic signature for a surface-
breaking crack produced by a point focus microscope. Proc. R. Soc.
A, 438: 47–65.
Rebinsky, D. A. and Harris, J. G. (1992b). An asymptotic calculation of the
acoustic signature of a cracked surface for the line focus scanning acoustic
microscope. Proc. R. Soc. A, 436: 251–265.
Rose, J. L. (1999). Ultrasonic Waves in Solids. Cambridge University Press,
Cambridge, U.K.
Schmerr Jr., L. W. and Song, S. J. (2007). Ultrasonic Nondestructive Evalua-
tion Systems: Models and Measurements. Springer-Verlag, New York.
Skudrzyk, E. (1971). The Foundations of Acoustics. Springer-Verlag, Vienna.
Sneddon, I. N. (1951). Fourier Transforms. McGraw-Hill, New York.
Somekh, M. G., Bertoni, H. L., Briggs, G. A. D. and Burton, N. J. (1985).
A two-dimensional imaging theory of surface discontinuities with the
scanning acoustic microscope. Proc. R. Soc. A, 401: 29–51.
Sommerfeld, A. (1967). Optics: Lectures on Theoretical Physics, Vol. 4. Aca-
demic Press, New York. Translated from the German by O. Laporte and
P. A. Moldauer.
Sommerfeld, A. (2004). Mathematical Theory of Diffraction. Birkhäuser,
Boston. Translated from the German by R. J. Nagem, M. Zampolli, and
G. Sandri. An introduction and translators’ notes are provided.
Spies, M. (1994). Elastic waves in homogeneous and layered transversely
isotropic media: plane waves and Gaussian wave packets. A general
approach. J. Acoust. Soc. Am., 95: 1748–1760.
Spies, M. (1999). Transducer field modeling in anisotropic media by superpo-
sition of Gaussian base functions. J. Acoust. Soc. Am., 105: 633–638.
Stamnes, J. J. (1986). Waves in Focal Regions. Adam Hilger, Bristol, U.K.
Tada, T., Fukuyama, E. and Madariaga, R. (2000). Non-hypersingular bound-
ary integral equations for 3-D non-planar crack dynamics. Comput. Mech.,
25: 613–626.
Thomas, D. P. (1963). Diffraction by a spherical cap. Proc. Camb. Phil. Soc.,
59: 197–209.
Thurston, R. N. (1974). Waves in solids. In C. Truesdell, editor, Mechanics of
Solids, volume 4, pages 109–308. Springer-Verlag, New York.
Ti, B. W., O’Brien, W. D. and Harris, J. G. (1997). Measurement of coupled
wave propagation in an elastic plate. J. Acoust. Soc. Am., 102: 1528–1531.
Titchmarsh, E. C. (1939). The Theory of Functions. Clarendon Press, Oxford,
second edition.
162 References
Titchmarsh, E. C. (1948). Introduction to the Theory of Fourier Integrals.

Clarendon Press, Oxford, second edition.
Ufimtsev, P. Ya. (1989). Theory of acoustical edge waves. J. Acoust. Soc. Am.,
86: 463–474.
Visscher, W. M. (1983). Theory of scattering of elastic waves from flat cracks
of arbitrary shape. Wave Motion, 5: 15–32.
Weinstein, L. A. (1969). The Theory of Diffraction and the Factorization
Method. The Golem Press, Boulder, CO. Translated from the Russian
by P. Beckmann.
Wen, J. J. and Breazeale, M. A. (1988). A diffraction beam field expressed as
the superposition of Gaussian beams. J. Acoust. Soc. Am., 83: 1752–1756.
Zhang, Ch. and Gross, D. (1998). On Wave Propagation in Elastic Solids with
Cracks. Computational Mechanics Publications, Southampton, U.K.
Index
Abelian theorems, 150 through a circular aperture, 24

acoustic material signature, 116, 118 diffraction by an elastic sphere, 130
acoustic microscope, 98 diffraction by a half-plane, 123
admittance, 74 diffraction coefficient, 33, 47
angular frequency dispersion, see velocity, group
defined, 11 periodic structure, 19
angular spectrum dispersion curve, 134
spherical wave, 55 dispersion relation, 134, 137, 138, 141
antiplane shear motion, 4 dispersion relation, Rayleigh–Lamb, 63
asymptotic approximation drive point admittance, 74
Abelian theorem, 44
double integral, 32n, 58
non-uniform, pole present, 49 edge diffraction
stationary phase, 34 acoustic, 40–49
steepest descents, 47–48 problem defined, 42
uniform, pole present, 46–48 acoustic in three dimensions,
asymptotic expansion, not Poincaré 45–46
type, 149 qualitative features, 47
asymptotic expansion, Poincaré type, Wiener–Hopf, 41–45
148 eigenvalue problem, 66
asymptotic sequence, 148 elastic waveguide, 136
electromechanical reciprocity,
110–113
beam reflection coefficient, 113
unfocused, acoustic, 35–36 reflection coefficient calculated, 111
electromechanical reciprocity identity,
circular annulus, 141 113
complex source, 74 energy flux, 8
compressional wave, defined, 6 complex, 8
continuity conditions, fluid–solid radiation resistance, 40
interface, 6–54 time average, 8
cracks, 81, 89 energy relations
curved waveguide, 141 conservation law, 7
equations of motion, 7
damped wave equation, 122 one-dimensional, 3
diffraction two-dimensional, 4
edge-diffracted wave, 25 antiplane motion, 4
edge-diffraction integral, inplane motion, 4
28, 32 error function, see Fresnel integral
163
164 Index
F number, 99 longitudinal strain, 3

focal length, 99 longitudinal stress, 4
focused beam, converging reflected, 107 long-wave theory, 140, 143
diverging reflected, 105 low-frequency scattering, 86
representations, 101
scattered, 103 Maggi transformation, 31, 32
Fourier transform Maxwell’s equations, 110
space, 12 measurement model, 113, 115
time, 9 microscope, acoustic reflection, imaging
frequency equation, 116
defined, 11 microscope, acoustic reflection,
Fresnel integral qualitative description, 97
asymptotic behavior, 155 mixed boundary condition, 3
Fresnel length
Fl , 28
beam, 35 Parseval’s relation, 40
Fresnel number, 99 partial waves, 19
Fresnel region, 47–49 passbands, 18
Fr , 49 periodic structure, 17–20
radiation from a piston, 51 effective wavenumber, 19
piezoelectricity, equations of, 109–110
plane-wave representation, 42
Gaussian beam, 75 Poisson summation formula, 12
geometrical (theory of) diffraction polarization vector, 57
radiation from a piston, 32–34 potentials, displacement, 5
Green’s function scalar potential, 5
piston problem, 27 vector potential, 5
Green’s tensor, 72, 73 propagation matrix, 17
group velocity, 8, 135, 146 propagation vector, 57
Hankel function, 154

radial basis function, 91
Hankel transform, 28, 36, 155
radiation
Helmholtz decomposition, 72
from a piston, 24–40
Helmholtz theorem, 5
problem defined, 26
Huygens’ principle, 25, 26–28
impedance, 38–40
hypersingular, 84
radiation impedance, 39
hypersingular boundary integral
Rayleigh function, 64
equations, 82
Rayleigh wave, focusing, 107
Rayleigh–Lamb dispersion relation, 137
imaging equation, 116 Rayleigh–Lamb modes, 62, 63, 137
impedance, 73, 76 reciprocity, 71, 78, 109
inplane motion, 4 elastodynamic, 72
isolated, first-order saddle point, 151 reflected plane wave, 57
isolated, first-order stationary-phase reflection coefficient, 57
point, 151 poles and branch points of, 59
reflection, measured at transducer, 115
Kirchhoff approximation, 84 Riemann sheet, physical, 36, 43, 59, 104
Lamé’s elastic constants, 2 scalar potential, see potentials,

Laplace transform displacement
time, 9 scattering
leaky Rayleigh pole, 59 Bragg, 19
leaky Rayleigh wave, 59, 98, 109, 118 from a lumped mass, 15
leaky Rayleigh wave, filling lens scattering matrix, 16
aperture, 106 shear horizontal motion, 4
leaky wave, 53 shear wave, defined, 6
limiting absorption, principle of, 6 Sommerfeld transformation, 37
Index 165
spectrum transmission coefficient, 57

angular, 25 transmission matrix, 16
spectrum, angular transmitted plane waves, 57
radiation from a piston, 36–37 trapped modes, 142
spherical wave
angular spectrum, 55 uniform asymptotic approximation, 149
at a plane interface, 127
converging, 100
vector potential, see potentials,
cylindrical-wave representation, 56
displacement
standing wave, defined, 11
velocity
steepest-descents contour, 47, 151
group
Stokes layer, 122
periodic structure, 20
Stoneley pole, 59
vibration, 12
Stoneley wave, 59
viscosity, 121
stopbands, 18
viscous fluid, 121
strain tensor, 2
vorticity, 122
stress tensor, 2
symmetries, waveguide eigenmodes, 67
waveguides, 133, 136
wavelength, defined, 11
Tauberian theorems, 150 wavenumber
time average, 7 defined, 10
product, 8 effective, periodic structure, 19
traction, 2 wavespeed
transforms, Fourier and Laplace, see compressional, 6
Fourier transform, Laplace shear, 6
transform Wiener–Hopf
defined, 9–12 acoustic edge diffraction, 42–45

Elastic Waves at High Frequency

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

Elastic Waves at High Frequency

Uploaded by

Copyright:

Available Formats

This page intentionally left blank

Elastic Waves at High Frequencies

Techniques for Radiation and Diﬀraction

Edited and Prepared for Publication by

Cambridge University Press

This publication is in copyright. Subject to statutory exception and to the

ISBN-13 978-0-511-78982-3 eBook (NetLibrary)

Cambridge University Press has no responsibility for the persistence or accuracy

Do not go gentle into that good night

5.4 Some strategies 88

This volume is dedicated to the memory of John G. Harris, whose

asymptotic approximations of waveﬁelds, which allowed him to share

Chapter 7, written by A. Davis, explores the eﬀects of viscosity on

Harris, J. G. (1980). Diﬀraction by a crack of a cylindrical longitudi-

Achenbach, J. D. and Harris, J. G. (1987). Asymptotic modeling of

Yogeswaren, E. and Harris, J. G. (1994). A model of a conformal

Block, G., Harris, J. G. and Hayat, T. (2000). Measurement models

1.1 Model equations

(1) the conservation of linear and angular momentum; and

In the linear approximation the density ρ is constant. The conservation

∂k τki + ρfi = ρ∂t2 ui . (1.1.1)

ij = (∂i uj + ∂j ui )/2, (1.1.2)

where ui is ith component of particle displacement. The underlying

τij = λkk δij + 2μij , (1.1.3)

(λ + μ)∂i ∂k uk + μ∂j ∂j ui + ρfi = ρ∂t2 ui . (1.1.4)

Written in vector notation the equation becomes

(λ + μ)∇(∇ · u) + μ∇2 u + ρf = ρ∂t2 u . (1.1.5)

Using the identity ∇2 u = ∇(∇ · u) − ∇ ∧ ∇ ∧ u, the equation can also

1.1.1 One-dimensional models

and (1.1.1) becomes

1.1.2 Two-dimensional models

1.1.3 Displacement potentials

The terms cL and cT are the compressional or longitudinal wavespeed,

1.2 Continuity and boundary conditions

ts · n̂ = −pf , n̂ ∧ ts = 0, us · n̂ = uf · n̂. (1.2.1)

Because the ﬂuid is ideal, no condition is placed on n̂ ∧ uf,s .

1.3 Flux of energy

from (1.1.1)–(1.1.3) by taking the dot product of ∂t u with (1.1.1). This

Therefore, as the mechanical energy decreases within R, it radiates

the time average of their product is

τ (x, t) = Re[τ (x)e−iωt ], u(x, t) = Re[u(x)e−iωt ],

the time average of (1.3.4) is

Two additional time-average quantities will be of interest: the time-

Cautionary note. There are waves, such as mode L3 of an elastic

1.4 The Fourier and Laplace transforms

The variable ω is complex. Its domain is such as to make the above

As with ω, p is a complex variable and its domain is such as to make

where  ≥ 0. The expressions for the inverse transforms (1.4.2) and

The advantage of this formulation of the inverse transform is that we may

Taking the real part gives

u1 = 2|A| cos(ωt + α) cos(kx1 ). (1.4.11)

We have taken A as complex so that α is its argument. This disturbance

In deriving this expression we have integrated from t = 0− to ∞ so

τ11 (x1 , t) = −P0 H(t − x1 /cL )e−η(t−x1 /cL ) . (1.4.14)

The conditions for convergence of the integral, as its contour is closed

and its inverse transform is

This shows that quite arbitrary disturbances can be decomposed into a

1.5 A wave is not a vibration

Theorem. Consider a function f (t) and its Fourier transform f¯(ω).

This formula relates a series to one comprised of its transformed terms.

These conditions are more restrictive than needed. Lighthill (1978a,

Consider the impulsive excitation of a rod of ﬁnite length 1. The gov-

τ11 (0, t) = −T δ(t), τ11 (1, t) = 0. (1.5.4)

ij = (∂i uj + ∂j ui )/2, (1.1.2)

τij = λkk δij + 2μij , (1.1.3)

where ≥ 0. The expressions for the inverse transforms (1.4.2) and

The time-average rate at which work is done on the piston P a by the