Ultrasonics 41 (2003) 229–236

www.elsevier.com/locate/ultras

A concise and efficient scattering matrix formalism for stable

analysis of elastic wave propagation in multilayered anisotropic solids
Eng Leong Tan *

Nanyang Technological University, Block S1, Nanyang Avenue, Singapore 639798, Singapore
Received 28 March 2002; accepted 30 October 2002

Abstract

This paper presents a concise and efficient scattering matrix formalism for stable analysis of elastic wave propagation in mul-
tilayered anisotropic solids. The formalism is capable of resolving completely the numerical instability problems associated with
transfer matrix method, thereby obviating the extensive reformulation in its modified versions based on delta operator technique. In
contrast to the earlier reflection matrix formalisms, all scattering matrices are obtained in a direct manner without invoking wave-
propagator or scatterer operator concepts. Both local and global reflection and transmission matrices corresponding to scatterings
in two and more layers are derived. The derivation of global scattering matrices in terms of the local ones is carried out concisely
based on physical arguments to provide better insights into scattering mechanism. Another formulation which is even more succinct
is also devised for obtaining the global scattering matrices directly from eigensolutions. The resultant expressions and algorithm are
terse, efficient and convenient for implementation.
Ó 2002 Elsevier Science B.V. All rights reserved.

PACS: 43.20.Gp; 43.20.Fn; 43.35.Cg; 43.35.Mr

1. Introduction frequency and/or thickness parameters become large.

For many decades, there has been extensive research
on elastic wave interaction with layered media due
to its many applications in geophysics, ocean acoustics,
composite materials, SAW devices, nondestructive eval-
uation, etc. To solve the problem of wave propagation
through planar stratified structures, there exists a very
systematic way based on the transfer matrix method
developed by Thomson [1] and Haskell [2], and well-
documented in the literature, e.g. [3–5]. This method
facilitates the transition of field variables from one layer
to the next while satisfying the interfacial condition
between them. By invoking the appropriate bound-
ary conditions at the top and bottom layers, a variety of
response solutions can be obtained conveniently using
computers.
Although the transfer matrix approach is applicable
in principle, numerical algorithms based on its direct
implementation have been found to suffer from a major
drawback concerning instabilities, particularly when the
*
Tel.: +65-67906190; fax: +65-67933318.
E-mail address: eeltan@ntu.edu.sg (E.L. Tan).
The cause of this problem can be alluded to the loss of
precision when performing calculations that combine
very large and small terms [6]. Many researchers have
proposed some modifications to the original transfer
matrix method by developing the delta operator tech-
nique which alleviates the precision problem [6–9]. For
propagation through general anisotropic media, the
technique however requires computation of large-order
(up to 20) delta matrices and derivation of many ana-
lytical expressions. Other simpler approach has been
suggested which performs certain numerical test to limit
the large values that ultimately cancel each other [10]. A
recent numerical truncation algorithm has also been
attempted to impose a maximum threshold for the ex-
ponential terms, thus limiting the error amplification
caused by them [11]. Still, the truncation approximation
depends much on the threshold value assigned and it
cannot be applied over wide range of cases.
Alternative to the transfer matrix method is the direct
global approach which involves a large single matrix,
assembled via simultaneous matching of boundary con-
ditions for all layers [12–15]. This approach is able to
circumvent the numerical difficulties above, but the

0041-624X/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved.
doi:10.1016/S0041-624X(02)00447-X
230 E.L. Tan / Ultrasonics 41 (2003) 229–236

matrix size will grow with the number of layers. For

z=Z N>
many layers, the direct global approach would lead to
considerable amount of memory storage and computa- layer N
tion time. On the other hand, seismologists have resorted > <
to the (constant-size) reflection matrix method in which z=Z N- 1 , Z N
the overall response of a stratified medium is built up layer N- 1
from the scattering properties of individual interfaces > <
one at a time [16–18]. This method leads to an efficient z=Z N- 2 , Z N- 1
recursive scheme which involves expressions not con- z^
taining any growing terms, and therefore completely
avoids the loss of precision problem. However, the
x^
derivation and implementation of the pertaining expres-
>
sions have been rather cumbersome due to substantial z=Z 2 , Z 3<
manipulations of wave-propagator partitions. To de-
layer 2
scribe the matrix algebra of wave propagators in a more
>
succinct fashion, some authors have introduced and dis- z=Z 1 , Z 2<
cussed on the concepts of scatterer operators and scat-
terer products [19,20]. Still, the combination of product layer 1
operations appears to be somewhat abstract and has not <
z=Z 1
provided much physical insight.
In view of the aforementioned instabilities and cum- Fig. 1. Geometry of planar multilayered anisotropic media. The top
bersomeness, the objective of this paper is to present an (N ) and bottom (1) layers can be semi-infinite or terminated at z ¼ ZN>
alternative formulation of scattering matrices which is and z ¼ Z1< respectively.
stable, efficient and illuminating. In Section 2, a first-
order 6
6 matrix differential system is first described to (Fig. 1). The top (labeled N ) and bottom (labeled 1)
prepare for simultaneous determination of both eigen- layers can be semi-infinite or terminated to represent
values and eigenvectors in a layered anisotropic medium. half-spaces or plates. For each layer f (f ¼ 1; 2; . . . ; N ),
The loss of precision in the transfer matrix method is its upper and lower bounding interfaces are denoted
delineated in terms of upward-bounded and downward- respectively by Zf> and Zf< such that Zf> P z and Zf< 6 z
bounded waves. In Section 3, the scattering matrix for- for z lies within the layer. Let the homogeneous medium
malism capable of resolving completely the numerical of each layer f be characterized by [21]
instability problems is presented. All scattering matrices
1
are to be obtained in a direct manner without appealing Tf ¼  C f : $ s vf ; ð1Þ
to wave-propagator or scatterer operator concepts. Both ix
local and global reflection and transmission matrices where Tf is the elastic stress dyadic, vf is the velocity
corresponding to scatterings in two and more layers are vector, and Cf is the elastic stiffness tensor which may be
derived. The derivation of global scattering matrices in complex to account for viscous damping in absorbing
terms of the local ones is carried out concisely based on media. To cater for arbitrary anisotropy, the stiffness
physical arguments to provide better insights into scat- tensors are assumed to have been transformed from the
tering mechanism. Another formulation which is even respective local coordinate representations to the prin-
more succinct is also devised for obtaining the global cipal coordinate system of Fig. 1. This can be performed
scattering matrices directly. The resultant expressions systematically by using BondÕs matrix and coordinate
and algorithm are kept as terse as possible so that their transformation matrix expressed in Euler rotation an-
implementation will be more convenient and efficient gles ð/; h; wÞ [21,22]. Without loss of generality, all fields
than the transfer matrix method. To illustrate the ap- are assumed to have common eikx x factor (no y depen-
plications of our formulas, Section 4 exemplifies the dence) with kx ¼ xsx denoting the wavenumber and sx
numerical computations for some cases that are sus- the slowness component along x ^ direction.
ceptible to numerical instabilities. Throughout the fol- To facilitate subsequent analysis, let us form a 6
1
lowing analysis, eixt time dependence is assumed and field vector f f for each layer f from the elements of
suppressed. velocity vector vf and normal stress vector sf ¼ ^z  Tf ,
i.e.


v
2. Problem formulation ff ¼ f : ð2Þ
sf
Consider the planar multilayered structure compris- Using (1) and the equation of motion (qf stands for
ing N rigidly-bonded solid layers stratified in ^z direction mass density)
 E.L. Tan / Ultrasonics 41 (2003) 229–236 231

$  Tf ¼ ixqf vf ; ð3Þ upward-bounded waves which remain (exponentially)

ð4Þ ð5Þ ð6Þ
bounded in the upper region. Likewise, szf , szf and szf
we obtain a first-order 6
6 matrix differential system are assumed to have negative imaginary parts and their
for f f as eigensolutions correspond to downward-bounded waves
d which stay bounded in the lower region. Note that
f f ¼ ixAf f f ; ð4Þ the descriptions of Ôupward-boundedÕ and Ôdownward-
dz
boundedÕ are used here, since they are more instructive
where the elements of Af are functions of transverse
than the usual ÔupgoingÕ and ÔdowngoingÕ in asserting
slowness, mass density and stiffness constants of layer f .
the radiation conditions for inhomogeneous waves, be-
The advantages of differential system (4) compared to
sides being more pertinent to the numerical issue of
the conventional ChristoffelÕs matrix equation have been
concern. For lossless materials which may result in real
described in [23]. In particular, both eigenvalues and
wavenumbers, the Poynting vector approach is usually
eigenvectors can be computed simultaneously for (j ¼
invoked to separate the types of waves associated [24,25].
1; 2; . . . ; 6)
Alternatively, one may introduce some slight loss to aid
ðjÞ ðjÞ ðjÞ
Af wf ¼ szf wf ; ð5Þ in the limiting analysis. Under such boundedness asso-
ðjÞ ciation, the phase matrix in (6) can be designated by 3

where denotes the jth vertical slowness compo-
szf
ðjÞ ðjÞ 3 submatrices as
nent related to the vertical wavenumber by kzf ¼ xszf ,
ðjÞ
> 
and wf is the corresponding eigenfield. Presently, there Pf 0
are many software packages available for solving Pf ¼ < ; ð11Þ
0 Pf
the eigensystem efficiently and reliably, even for degen-
erate (repeated) eigenvalues encountered along the axis where the superscripts Ô>Õ and Ô<Õ stand for Ôupward-
of rotational symmetry. Hence, this system is also ap- boundedÕ and Ô downward-boundedÕ respectively (0 is
plicable to materials of very high symmetry including the null matrix). With reference to (9), one can see that
isotropic media. the exponential matrix tends to mix the (exponentially)
Having determined the eigenvalues and eigenvectors, large and small terms through the matrix products. Such
they are to be collected into phase matrix Pf and ei- mixture may easily cause numerical loss of precision
genvector matrix Wf as thus giving rise to erroneous results [6].
h i
ð1Þ ð2Þ ð3Þ ð4Þ ð5Þ ð6Þ
Pf ðzÞ ¼ diag eikzf z ; eikzf z ; eikzf z ; eikzf z ; eikzf z ; eikzf z ;
ð6Þ 3. Scattering matrix formalism
h i
To eliminate completely the numerical instability
Wf ¼ wfð1Þ wf
ð2Þ ð3Þ
wf wf
ð4Þ
wf
ð5Þ
wf
ð6Þ
: ð7Þ
problems associated with transfer matrix method, one
The solution of (4) can then be inferred from the initial may resort to the unconditionally stable method in-
field vector at z0 via volving scattering matrices. However, the earlier for-
malisms of this method are rather cumbersome in their
f f ðzÞ ¼ eixAf ðzz0 Þ f f ðz0 Þ; ð8Þ implementation. In this section, we shall present a sim-
ple formulation of scattering matrices which is physi-
eixAf ðzz0 Þ ¼ Wf Pf ðz  z0 ÞW1
f ; ð9Þ cally illuminating as well as very convenient and efficient
where the exponential matrix constitutes the so-called for computation. To that end, let us first represent the
local transfer matrix of layer f . By imposing the field field as linear combination of waves through
continuity conditions across intervening interfaces, the f f ðzÞ ¼ Wf wf ðzÞ: ð12Þ
local transfer matrices can be cascaded accordingly to
form a global transfer matrix which relates the field Here, wf is a 6
1 wave vector which has lumped into
vector at z0 within layer f directly to that at z within its elements the various exponentials in (8) and (9). In
another layer g, e.g. for g > f , accordance with the boundedness association above, the
< > wave vector is to be decomposed into 3
1 partitions
f g ðzÞ ¼ eixAg ðzZg Þ eixAg1 ðhg1 Þ    eixAf ðZf z0 Þ f f ðz0 Þ which stay upward-bounded (Ô>Õ) and downward-
ð10Þ bounded (Ô<Õ) respectively, i.e.
(hf is the thickness of layer f ).
>
wf
Although the transfer matrix method is convenient to wf ¼ : ð13Þ
w<
f
implement, its practical computations may lead to nu-
merical instabilities or failures. For definiteness hence- Furthermore, in view of the composition of f f in (2),
ð1Þ ð2Þ ð3Þ
forth, let us assume that szf , szf and szf feature positive the 3
3 submatrices for Wf of (7) are well designated
imaginary parts and their eigensolutions correspond to by
232 E.L. Tan / Ultrasonics 41 (2003) 229–236


v> v< exemplified in [20,23], each reflection and transmission
Wf ¼ f> f
: ð14Þ
sf s<
f
matrix generally consists of nine reflection and trans-
mission coefficients. These coefficients represent the nine
Again, these notations are based on the superscripts Ô>Õ possible combinations obtainable from the couplings of
and Ô<Õ to signify the velocity and normal stress ei- different wave types in each layer, cf. one quasi-longi-
genmatrices of upward-bounded and downward-boun- tudinal and two quasi-shear. (Certain coefficient would
ded association. (In general, the convention herein be zero if there is no coupling or conversion takes place.)
attempts to use the same symbol but superscripted to In the similar way, the reflection and transmission ma-
denote the partition or constituent whose size is natu- trices for downward-bounded wave incident from layer
rally different from the original one, e.g. wf is of 6
1 f þ 1 downto f are defined as
while its partition w>f is of 3
1, vf is of 3
1 while its l
w> < < <
f þ1 ðZf þ1 Þ ¼ rf þ1;f wf þ1 ðZf þ1 Þ; layer f semi-infinite;
constituent v>f is of 3
3, etc.)
ð17Þ
3.1. Local reflection and transmission matrices l
w< > < <
f ðZf Þ ¼ tf þ1;f wf þ1 ðZf þ1 Þ; layer f semi-infinite:
At a plane interface between two anisotropic half- ð18Þ
spaces, the relationships between the upward-bounded All these matrices constitute the basic building blocks
and downward-bounded waves in either layer can be for subsequent ones associated with the structure having
described succinctly through reflection and transmission more layers.
matrices. Assuming that the interface is between the Upon imposing the field continuity condition to-
layers labeled as f and f þ 1 in Fig. 2(a), these matrices gether with the radiation condition for each layer in
are defined for upward-bounded wave incident from sequel, the local scattering matrices in (15)–(18) can be
layer f to f þ 1 as determined simultaneously from
l
w< > > >
f ðZf Þ ¼ rf ;f þ1 wf ðZf Þ; layer f þ 1 semi-infinite; " l # " #1 " #
rf ;f þ1 tlf þ1;f v<
f v>
f þ1 v>
f v<
f þ1
ð15Þ ¼ :
tlf ;f þ1 rlf þ1;f s<
f s>
f þ1 s>
f s<
f þ1
l
w> < > >
f þ1 ðZf þ1 Þ ¼ tf ;f þ1 wf ðZf Þ; layer f þ 1 semi-infinite: ð19Þ
ð16Þ If desired, these solutions can be cast into the expres-
The superscript ÔlÕ signifies that all wave interactions are sions involving merely matrices of size 3
3 by carry-
ÔlocalÕ to the interface of pertaining layers, and there is ing out the partitioned matrix inversion explicitly. This
no reflected wave contribution in the refracted layer. As would be useful particularly when only one or two
matrices are required but is at the expense of slightly
elaborated form of equations. Another equivalent form
of local scattering matrices has been derived in terms of
the partitions of upward or downward interfacial wave
propagator defined by
wf þ1 ðZf<þ1 Þ ¼ Qf ;f þ1 wf ðZf> Þ; Qf ;f þ1 ¼ W1
f þ1 Wf ; ð20Þ
or
wf ðZf> Þ ¼ Qf þ1;f wf þ1 ðZf<þ1 Þ; Qf þ1;f ¼ W1
f Wf þ1 ; ð21Þ
respectively, cf. [16,20,23]. In a more direct manner, Eq.
(19) demonstrates the simple swapping and grouping of
the columns of Wf þ1 and Wf in (14) to yield all reflection
and transmission matrices without involving wave-
propagator partitions.

3.2. Global reflection and transmission matrices: Physics-

based derivation

To deal with a stratified medium composed of more

Fig. 2. Geometry of (a) two and (b) more layers for the derivation
than two layers, one can build up the overall response by
of local and global reflection and transmission matrices for incident employing certain addition rules and recurrence rela-
upward-bounded wave in layer f . tions in terms of the local scattering matrices above. In
 E.L. Tan / Ultrasonics 41 (2003) 229–236 233

contrast to the earlier derivation of such pertaining ex- l l

w> < < < > >
f þ1 ðZf þ1 Þ ¼ rf þ1;f wf þ1 ðZf þ1 Þ þ tf ;f þ1 wf ðZf Þ: ð28Þ
pressions in [16–20], we will construct the recurrence
relations solely based on physical arguments in order Performing some substitutions and algebraic operations
to provide real insights into scattering mechanism. for (22)–(28), we arrive at
Moreover, the development will fully exploit the matrix rgf ;f þ1 ¼ rlf ;f þ1 þ tlf þ1;f Rgf þ1;f þ2 tgf ;f þ1 ; ð29Þ
solutions of (19) without calling for intermediate ma-
nipulation of wave-propagator partitions. h i1
Consider the structure which consists of layer f and tgf ;f þ1 ¼ I  rlf þ1;f Rgf þ1;f þ2 tlf ;f þ1 : ð30Þ
f þ 1 as before but with the latter now being bounded/
These relations are seen to bear the form of those de-
terminated by an interface with other layers (Fig. 2(b)).
rived in [16–18] through manipulation of wave-propa-
Assuming that the upward-bounded wave is incident
gator partitions. Therein, some interpretations have
from layer f , the reflection and transmission matrices
been given to the matrix inverse in (30) which can be
are defined as
recognized as the reverberation term. More perceptibly,
g
w< > > >
f ðZf Þ ¼ rf ;f þ1 wf ðZf Þ; ð22Þ (29) and (30) have separated the local (ÔlÕ) and global (ÔgÕ)
scattering terms distinctly. They have also identified
g
w> < > >
f þ1 ðZf þ1 Þ ¼ tf ;f þ1 wf ðZf Þ: ð23Þ clearly the matrices with and without incorporation of
exponential phase factors (e.g. as in Rg and rl , tl re-
These matrices are termed as ÔglobalÕ and superscripted spectively).
with ÔgÕ to indicate the incorporation of multiple-reflec-
tion effects due to the presence of more layers (unlike 3.3. Global reflection and transmission matrices: Direct
(15) and (16) having no reflection in layer f þ 1). Simi- derivation
larly, all reflections due to layer f þ 2 and beyond are to
be lumped into the global reflection matrix at z ¼ Zf>þ1 : The above local and global scattering matrices
g
w< > > >
f þ1 ðZf þ1 Þ ¼ rf þ1;f þ2 wf þ1 ðZf þ1 Þ: ð24Þ have been obtained in an alternative manner which is
more illuminating and concise without invoking wave-
This relation is in turn transferable to another interface propagator or scatterer operator concepts. Although the
at z ¼ Zf<þ1 by introducing derivation enables one to interpret the composition of
g global reflection and transmission in terms of local in-
w< < > <
f þ1 ðZf þ1 Þ ¼ Rf þ1;f þ2 wf þ1 ðZf þ1 Þ; ð25Þ
teractions (cf. (27) and (28)), it may not be of much
Rgf þ1;f þ2 ¼ P< < > g > > < concern especially to those who still prefer the transfer
f þ1 ðZf þ1  Zf þ1 Þrf þ1;f þ2 Pf þ1 ðZf þ1  Zf þ1 Þ
g
matrix method for its convenience of implementation.
¼ P< >
f þ1 ðhf þ1 Þrf þ1;f þ2 Pf þ1 ðhf þ1 Þ: ð26Þ To devise another formulation of global scattering ma-
Tracing from the right of (25) and (26), Rgf þ1;f þ2 is seen trices which is even more convenient and more succinct
to describe the global round-trip reflections which in- than the previous subsection, let us impose the field
clude the phase traversal in the uniform layer from Zf<þ1 continuity condition directly for the structure in Fig.
to Zf>þ1 followed by reflection back to Zf<þ1 . Note that 2(b). Taking the definitions of global scattering matrices
owing to the boundedness of P> in (22)–(26), we arrive at
f þ1 ðhf þ1 Þ as well as
P< ðh Þ, R g
will stay bounded even for very 2 n o 31
f þ1 f þ1 f þ1;f þ2
g  v <
v >
þ v <
R g
>
large thickness and high frequency. rf ;f þ1 4 f f þ1 f þ1 f þ1;f þ2
5 vf
¼ n o :
Eqs. (22)–(26) have merely set forth the definitions for tgf ;f þ1 s < >
s þs R < g s>
f
f f þ1 f þ1 f þ1;f þ2
various global scattering matrices yet undetermined. To
solve for these matrices, let us take a closer look at the ð31Þ
ÔlocalÕ interactions near the interface between layer f
Evidently, this equation permits a one-step determina-
and f þ 1. Considering the physical constraints, the
tion of global reflection and transmission matrices
downward-bounded wave at z ¼ Zf> must be a conse-
without involving any local matrices at all. In fact, once
quence of the local reflection of the upward-bounded
the eigensystem in (5) is solved, all matrix operations
wave in the same plane, plus the local transmission of
required for deducing the exponential matrices (cf. (9)),
downward-bounded wave from z ¼ Zf<þ1 , i.e.
the wave propagators (cf. (20) and (21)), or the local
l l
w< > > > < <
f ðZf Þ ¼ rf ;f þ1 wf ðZf Þ þ tf þ1;f wf þ1 ðZf þ1 Þ: ð27Þ scattering matrices (cf. (19)), have instead been fully
utilized for global scattering matrices directly (cf. (31)).
Furthermore, the upward-bounded wave at z ¼ Zf<þ1 In other words, given the eigensolutions, we now have a
should be attributed to the local reflection of the stable and efficient algorithm of
downward-bounded wave in the same plane, plus the
local transmission of upward-bounded wave from z ¼  n o
Wf ; Pf ! rgf ;f þ1 ; tgf ;f þ1 ;
Zf> :
234 E.L. Tan / Ultrasonics 41 (2003) 229–236

as compared to the potentially unstable tgf ;f þl ¼ tgf þl1;f þl P> g > g

f þl1 ðhf þl1 Þ    tf þ1;f þ2 Pf þ1 ðhf þ1 Þtf ;f þ1 :
  n o
Wf ; Pf ! eixAf z ! rgf ;f þ1 ; tgf ;f þ1 ; ð33Þ

Tracing from the right of (33), such cascading scheme

or the rather cumbersome [17]

 indeed complies with the structure geometry of Fig. 1
Wf ; Pf ! Qf ;f þ1 or Qf þ1;f
n o
! rlf ;f þ1 ; tlf ;f þ1 ; rlf þ1;f ; tlf þ1;f
n o
! rgf ;f þ1 ; tgf ;f þ1 :
Carrying out the matrix inversion explicitly, the system
(31) can be shown to be equivalent to that of (19), (29)
and (30). Furthermore, for large frequency–thickness
product whereby the effect of adjacent inhomogeneous
waves vanishes (i.e. Rgf þ1;f þ2 ! 0), the global scattering
matrices in (31) simply reduce to the local ones in (19),
thus providing immediate justification for their numer-
ical stability.
Unlike the iterative cascading of local transfer ma-
trices, (31) must of course be used in a recursive scheme
so that Rgf þ1;f þ2 or rgf þ1;f þ2 should be known before-
hand. These matrices are to be initiated from the first
one specified by the boundary condition for the top
rS refers to a stack of r superlayer-repetitions
layer, for instance
 in turn consists of few elementary layers oriented at
0; layer N semi-infinite;
rgN ;N þ1 ¼ < 1 >
½sN
sN ; layer N terminated with free surface:
Other types of boundary condition may be considered as
well to yield another form of global reflection matrix.
Meanwhile, the global transmission matrix resulted
from (31) can be used to generalize the relation in (23)
by connecting the upward-bounded wave at z ¼ Zf<þl
(with l > 1) directly to that at z ¼ Zf> :
1
0.9
upward from layer f to f þ l, cf. tgf ;f þ1 mimics an in-
terface while P> f mimics a layer. It also parallels to the
cascading sequence in the global transfer matrix formula
(10), cf. W1
f þ1 Wf mimics an interface while Pf mimics a
layer. Recall however that all 3
3 matrices in (33) will
stay bounded but this need not be the case for those
6
6 exponential matrices in (10).
4. Numerical examples
To illustrate some immediate applications for the
formulas presented in the previous sections, let us revisit
the case of viscoelastic fiber composites studied in [9,11]
where numerical instabilities have been found prevailing
when using the transfer matrix method. Following the
usual convention, the composite laminate denoted as
½superlayer
followed by their mirror symmetry (S). Each superlayer
various azimuthal angles. The viscoelastic constants
corresponding to the sample elementary layer have been
ð32Þ given in [11,26]. Due to our choice of coordinate system
as in Fig. 1 (^y is along the fiber direction), the material
data therein have been transformed through Euler an-
gles ð90°; 0°; 0°Þ and ð90°; 90°; 0°Þ. Both reflection and
transmission coefficients are computed versus incident
angles by making repeated use of (31) and (33). For the
sample immersed totally in fluid (water), the initial and
final coefficients for solid–fluid interfaces can be adapted
0.6

0.8
0.5
Reflection coefficient

Transmission coefficient

0.7

0.6 0.4

0.5
0.3
0.4

0.3
0.2

0.2

0.1
0.1

0
0 10 20 30 40 50 60 70 80 90
0
Incident angle (°) 0 10 20 30 40 50 60 70 80 90

Incident angle (°)

Fig. 3. Computed reflection coefficient for unidirectional graphite–
epoxy laminate at the plane of incidence along the fiber, f ¼ 2:242 Fig. 4. Computed transmission coefficient for unidirectional graphite-
MHz, d ¼ 3:434 mm, viscoelastic constants from [11]. (Squares rep- epoxy laminate at the plane of incidence along the fiber, f ¼ 2:242
resent data from [11, Fig. 1(b)].) MHz, d ¼ 3:434 mm, viscoelastic constants from [11].
 E.L. Tan / Ultrasonics 41 (2003) 229–236 235

to take into account the vanishing of shear stress com-
ponents.
To provide an initial validation, Fig. 3 reproduces the
plot of reflection coefficient versus incident angle for the
unidirectional graphite–epoxy laminate examined in
[11] at frequency f ¼ 2:242 MHz and with thickness
8S sample in
d ¼ 3:434 mm. The solid curve denotes the present
computation results which agree well with that of [11,
Fig. 1(b)] and can be obtained by treating the laminate
as a single layer or as an arbitrary repetition of identical
elementary layers. For completeness, the corresponding
plot of transmission coefficient versus incident angle is
also supplemented in Fig. 4 as a by-product. Using the
present formulations, these computations are found to
be stable without having to incorporate any numerical
truncation approximation as in [11]. We next consider
the composite samples made of different stacking se-
quences of unidirectional PEEK carbon layers in [9] and
[26]. Figs. 5 and 6 show the computed reflection and
transmission responses for ½45°=45°
/ ¼ 0° plane. The frequency and laminate thickness are
chosen as f ¼ 5 MHz and d ¼ 4 mm. No numerical
instabilities have been encountered like that occurred in
[9] despite the application of D2 operator technique.
Furthermore, one does not have to compute large-order
matrices or derive complicated expressions as required
by D3 operator technique for stable computations. To
further justify the robustness of our formulas, let us

0.7
1

0.9
0.6

0.8

Transmission coefficient 0.5

Reflection coefficient

0.7

0.6 0.4

0.5
0.3
0.4

0.3 0.2

0.2
0.1

0.1

0
0 0 10 20 30 40 50 60 70 80 90
0 10 20 30 40 50 60 70 80 90
Incident angle (°)
Incident angle (°)
Fig. 6. Computed transmission coefficient for ½45°=45°
8S sample with
Fig. 5. Computed reflection coefficient for ½45°=45°
8S sample with
/ ¼ 0°, f ¼ 5 MHz, d ¼ 4 mm, viscoelastic constants from [26].
/ ¼ 0°, f ¼ 5 MHz, d ¼ 4 mm, viscoelastic constants from [26].

1 0.1

0.9 0.09

0.8 0.08
Transmission coefficient
Reflection coefficient

0.7 0.07

0.6 0.06

0.5 0.05

0.4 0.04

0.3 0.03

0.2 0.02

0.1 0.01

0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90

Incident angle (°) Incident angle (°)

Fig. 7. Computed reflection coefficient for ½0°=5°=10°=15°= . . . =90°

6S Fig. 8. Computed transmission coefficient for ½0°=5°=10°=
sample with / ¼ 0°, f ¼ 5 MHz, d ¼ 50 mm, viscoelastic constants 15°= . . . =90°
6S sample with / ¼ 0°, f ¼ 5 MHz, d ¼ 50 mm, visco-
from [26]. elastic constants from [26].
236 E.L. Tan / Ultrasonics 41 (2003) 229–236
conceive a sample made up of many layers as ½0°=5°=
10°=15°= . . . =90°
6S . Taking the same set of parameters loaded multiaxial anisotropic media, J. Acoust. Soc. Am. 89 (2)
except for the thickness being increased to d ¼ 50 mm,
the results for reflection and transmission (small, as
expected) responses in / ¼ 0° plane are depicted in Figs.
7 and 8. For such extreme case, the frequency–thickness
product would be very high and there is a large number
(228) of layers needs to be dealt with. This will present
great difficulty (instability) for the transfer matrix
method, while the remedy using direct global approach
will be costly due to huge matrix of over one thousandth
order. Still, the present approach features its numerical
stability and reliability, apart from being efficient and
convenient for implementation.
5. Conclusion
This paper has presented a concise formulation of
scattering matrices for analysis of elastic wave propa-
gation in planar stratified anisotropic media. The use of
scattering matrices has eliminated completely the nu-
merical instability problems associated with transfer
matrix method, thereby obviating the extensive refor-
mulation in its modified versions based on delta opera-
tor technique. In contrast to the earlier reflection matrix
formalisms, all scattering matrices have been obtained in
a direct manner without appealing to wave-propagator
or scatterer operator concepts. Both local and global
reflection and transmission matrices corresponding to
scatterings in two and more layers have been derived.
The derivation of global scattering matrices in terms of
the local ones has been carried out concisely based on
physical arguments, thus providing better insights into
scattering mechanism. Another formulation which is
even more succinct has also been devised for obtaining
the global scattering matrices directly from eigensolu-
tions. The resultant expressions and algorithm have
been kept as terse as possible so that their implemen-
tation will be more convenient and efficient than the
transfer matrix method. Some numerical computations
have been illustrated to justify the robustness of our
formulas. Because of its numerical stability and effi-
ciency, the present method will be well-suited and useful
for wide range of frequency and thickness parameters.
References
[1] W.T. Thomson, Transmission of elastic waves through a stratified
solid medium, J. Appl. Phys. 21 (1950) 89–93.
[2] N.A. Haskell, The dispersion of surface waves on multilayered
media, Bull. Seismol. Soc. Am. 43 (1953) 17–34.
[3] A.H. Nayfeh, The general problem of elastic wave propagation in
multilayered anisotropic media, J. Acoust. Soc. Am. 89 (4) (1991)
1521–1531.
[4] A.H. Nayfeh, D.E. Chimenti, Elastic wave propagation in fluid-
(1991) 542–549.
[5] A.H. Nayfeh, Wave propagation in layered anisotropic media
with applications to composites, North-Holland, Amsterdam,
1995.
[6] J.W. Dunkin, Computation of modal solutions in layered elastic
media at high frequencies, Bull. Seismol. Soc. Am. 55 (1965) 335–
358.
[7] T. Kundu, A.K. Mal, Elastic waves in multilayered solid due to a
dislocation source, Wave Motion 7 (1985) 459–471.
[8] D. Levesque, L. Piche, A robust transfer matrix formulation
for the ultrasonic response of multilayered absorbing media,
J. Acoust. Soc. Am. 92 (1) (1992) 452–467.
[9] M. Castaings, B. Hosten, Delta operator technique to improve the
Thomson-Haskell method stability for propagation in multilay-
ered anisotropic absorbing plates, J. Acoust. Soc. Am. 95 (4)
(1994) 1931–1941.
[10] B. Hosten, Bulk heterogeneous plane waves propagation through
viscoelastic plates and stratified media with large values of
frequency domain, Ultrasonics 29 (1991) 445–450.
[11] K. Balasubramaniam, On a numerical truncation approximation
algorithm for transfer matrix method, J. Acoust. Soc. Am. 107 (2)
(2000) 1053–1056.
[12] L. Knopoff, A matrix method for elastic wave problems, Bull.
Seismol. Soc. Am. 54 (1964) 431–438.
[13] H. Schmidt, F.B. Jensen, A full wave solution for propagation in
multilayered viscoelastic media with application to Gaussian
beam reflection at fluid–solid interfaces, J. Acoust. Soc. Am. 77 (3)
(1985) 813–825.
[14] A.K. Mal, Wave propagation in layered composite laminates
under periodic surface loads, Wave Motion 10 (1988) 257–
266.
[15] M.J.S. Lowe, Matrix techniques for modeling ultrasonic waves in
multilayered media, IEEE Trans. Ultrason., Ferroelec., Freq.
Contr. 42 (4) (1995) 525–542.
[16] B.L.N. Kennett, N.J. Kerry, Seismic waves in a stratified half
space, Geophys. J. R. Astron. Soc. 57 (1979) 557–583.
[17] B.L.N. Kennett, Seismic Wave Propagation in Stratified Media,
Cambridge University Press, Cambridge, 1983.
[18] D.C. Booth, S. Crampin, The anisotropic reflectivity technique:
theory, Geophys. J. R. Astron. Soc. 72 (1983) 755–766.
[19] P.R. Saastamoinen, On propagators and scatterers in wave
problems of layered elastic media: A spectral approach, Bull.
Seismol. Soc. Am. 70 (1980) 1125–1135.
[20] G.J. Fryer, L.N. Frazer, Seismic waves in a stratified anisotropic
media, Geophys. J. R. Astron. Soc. 78 (1984) 691–710.
[21] B.A. Auld, Acoustic Fields and Waves in Solids, vol. I, R.E.
Krieger, Malabar, 1990.
[22] H. Goldstein, Classical Mechanics, Addison-Wesley, Reading,
1980.
[23] B. Mandal, Reflection and transmission properties of elastic waves
on a plane interface for general anisotropic media, J. Acoust. Soc.
Am. 70 (2) (1991) 1106–1118.
[24] S.I. Rokhlin, T.K. Bolland, L. Adler, Reflection and refraction of
elastic waves on a plane interface between two generally aniso-
tropic media, J. Acoust. Soc. Am. 79 (4) (1986) 906–918.
[25] P. Lanceleur, H. Ribeiro, J.F. De Belleval, The use of inhomo-
geneous waves in the reflection–transmission problem at a plane
interface between two anisotropic media, J. Acoust. Soc. Am. 93
(4) (1993) 1882–1891.
[26] B. Hosten, M. Castaings, Transfer matrix of multilayered
absorbing and anisotropic media. Measurements and simulations
of ultrasonic wave propagation through composite materials,
J. Acoust. Soc. Am. 93 (5) (1993) 1488–1495.