Professional Documents
Culture Documents
JOHN G. HARRIS
Late, University of Illinois at Urbana–Champaign
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
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
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
ix
x Foreword
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
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
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
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
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,
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.
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.
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).
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)
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
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 −∞ −∞
where λ is a parameter.
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.
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 =−∞
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.
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
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)
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.
1.6 Dispersive propagation 17
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
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)
−∞
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.
23
24 Canonical acoustic-wave problems
x1
φ
^e
ρ
ρ
r′
r
x = (x1, x3, x3)
x2
x3
^e3
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
∂3 g = 0, x ∈ ∂R (2.1.4)
where x = (x1 , x2 , x3 ) and x = (x1 , x2 , −x3 ), and x3 > 0.
The bilinear form
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
28 Canonical acoustic-wave problems
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
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
∇ · [ g (ê3 ∂3 ϕg ) − ϕg ∇g ] + ∇t g · ∇t ϕg = ϕg (x)δ(x − x )
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 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 Radiation from a piston in an infinite baffle 31
where
cos χ∓ = (∇2 s∓ ) · êρ .
cos β = ∇1 r · ê3 .
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.
34 Canonical acoustic-wave problems
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 .
2.1 Radiation from a piston in an infinite baffle 35
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
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 ].
36 Canonical acoustic-wave problems
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)
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
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
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 ).
2.1 Radiation from a piston in an infinite baffle 39
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
^i
p
x1
φ
ρ
x2
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.
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 .
44 Canonical acoustic-wave problems
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.
2.2 Diffraction of an acoustic plane wave by an edge 47
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.
50 Canonical acoustic-wave problems
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:
This chapter continues the work of the previous one. Its purpose is to
describe the following two canonical elastic-wave problems:
52
3.1 The scattering of a spherical wave 53
(0,0, –b)
^ Ideal fluid
n
x1
Elastic solid
x3
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∞
56 Canonical elastic-wave problems
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 .
where the angles ξL and ξT are indicated in Fig. 3.2. These waves are
given by
^i
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 .
3.1 The scattering of a spherical wave 57
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)
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
3.1 The scattering of a spherical wave 59
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.
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
60 Canonical elastic-wave problems
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
where
1 if C(θ) encloses ξR ,
χ(ξR , θ) =
0 otherwise.
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
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.
3.2.2 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
3.2 Rayleigh–Lamb modes and Rayleigh surface waves 63
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)
βa = βR + a , βs = βR − s , a + s = 2, (3.2.6)
fR (β) = p2 − γL γT ,
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 .
3.2 Rayleigh–Lamb modes and Rayleigh surface waves 65
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)
(L − ∂1 ) U = F, (3.2.9)
66 Canonical elastic-wave problems
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 .
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
3.2 Rayleigh–Lamb modes and Rayleigh surface waves 67
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
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.
68 Canonical elastic-wave problems
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)
(3.2.20)
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
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 ,
F0 F0∗
P = , (4.2.4)
2Zr
74 Radiation and impedance
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
4.2 Green’s tensor 75
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
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
F = −iωZu0 . (4.2.11)
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.
78 Radiation and impedance
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
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.
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
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 .
84 Integral equations for crack scattering
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
Let
with S10 and S20 being the corresponding quantities at (x0 , y0 ). Then
N(q) = (−S1 , −S2 , 1) and N(p) = (−S10 , −S20 , 1). Let
with inverse
∞ ∞
1
usc (x, y, z) = F −1 {U } = U (ξ, η, z) e−i(ξx+ηy) dξ dη.
(2π)2 −∞ −∞
90 Integral equations for crack scattering
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
5.8 Curved cracks 93
96
6.1 Scanned, reflection acoustic microscope 97
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
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.
100 Scanned acoustic imaging
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.
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
6.5 Scattered focused beam 103
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
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
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.
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
106 Scanned acoustic imaging
1/2
2π
ϕ = −AG(βR )
R
2αR eiπ/4 eikr cos(ξR −φ) , (6.5.7)
krR
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.
6.5 Scattered focused beam 107
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.
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
108 Scanned acoustic imaging
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).
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.
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 ,
∂Re
∂Rm
x3
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
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
6.7 Measurement model 113
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)
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.
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 .
6.7 Measurement model 115
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
116 Scanned acoustic imaging
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
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
118 Scanned acoustic imaging
−1
Δ = 2(1 − cos βR ) .
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).
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
120 Scanned acoustic imaging
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
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
ui = A(ê1 cos φ0 − ê2 sin φ0 )eika (x1 cos φ0 −x2 sin φ0 ) , 0 ≤ φ0 < π/2,
(7.1.8)
7.2 Diffraction by a half-plane 123
The solution in the half-space x2 > 0 then suffices and is such that, at
x2 = 0,
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.
126 Acoustic diffraction in viscous fluids
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
7.3 Scattering of a spherical wave at a plane interface 127
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
" √
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)
+ 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)
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.
133
134 Near-cut-off behavior in waveguides
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.
136 Near-cut-off behavior in waveguides
∂ 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
6
k
0
0 1 2 3 4 5 6 7 8 9 10
ω
with
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.
where
κ = (1 − ασ 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)
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.
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.
0.8
8
0.6
f0
0.4
6 0.2
−50 0 50
4 ξ
0
−10 −5 0 5 10
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
ψ1ηη + Λ20 ψ1 = 0,
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
147
148 Asymptotic expansions
g(z) ∼ a0 f0 (z), z → z0 ,
or
g(z) = a0 f0 (z) + a1 f1 (z) + o(f1 ), z → z0 ,
uniformly as z → z0 .
N
g(z) = an fn (z) + o[fN (z)]
n=0
A.2 Abelian theorems for Fourier transforms 149
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).
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
(−2π)1/2
I(κ) = f (zs ) eκq(zs ) + O(κ−3/2 ), κ → ∞. (A.8)
[κ d2z q(zs )]1/2
(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
(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
154
B.2 Fresnel integral 155
and
(2) 1
H0 (z) = e−iz cos μ dμ. (B.2)
π C2
where, for the (1) function arg z ∈ (−π, 2π), and for the (2) function
arg z ∈ (−2π, π).
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
157
158 References
163
164 Index