You are on page 1of 17

European Journal of Mechanics A/Solids 24 (2005) 820–836

Elastodynamic fundamental solutions for certain families


of 2d inhomogeneous anisotropic domains: basic derivations
T.V. Rangelov a , G.D. Manolis b,∗ , P.S. Dineva c
a Department of Mathematical Physics, Institute of Mathematics and Informatics, Bulgarian Academy of Sciences, 1113 Sofia, Bulgaria
b Department of Civil Engineering, Aristotle University, Thessaloniki, GR-54006, Greece
c Department of Continuum Mechanics, Institute of Mechanics, Bulgarian Academy of Sciences, 1113 Sofia, Bulgaria

Received 21 June 2004; accepted 11 May 2005


Available online 24 June 2005

Abstract
The existence of frequency-dependent fundamental solutions for anisotropic, inhomogeneous continua under plane strain
conditions is a necessary pre-requisite for studying wave motion, either in geological media or in composites with both depth
and direction-dependent material parameters. The path followed herein for recovering such types of solutions is (a) to use a
simple algebraic transformation for the displacement vector so as to bring about a governing partial differential equation of
motion with constant coefficients, albeit at the cost of introducing a series of constraints on the types of material profiles;
(b) to carefully examine these constraints, which reveal a rather rich range of possible variations of the elastic moduli in both
vertical and lateral directions; and (c) to use the Radon transformation for handling material anisotropy. Depending on the
type of constraints that have been introduced, two basic classes of materials are identified, namely ‘Case A’ where further
restrictions are placed on the elasticity tensor and ‘Case B’ where further restrictions are placed on the material profile. We note
at this point that for isotropic materials, the elasticity tensor constraints correspond to equal Lamé constants or, alternatively,
to a fixed Poisson’s ratio. The present methodology is quite general and the homogeneous anisotropic medium, as well as the
inhomogeneous isotropic one, can both be recovered as special cases from the results given herein.
 2005 Elsevier SAS. All rights reserved.

Keywords: Algebraic transformations; Anisotropy; Boundary integral equation methods; Elastic waves; Fundamental solutions;
Inhomogeneous media; Radon transformation

1. Introduction

The ability to model anisotropic, inhomogeneous materials is of paramount importance in many engineering fields such
as acoustics, fluid and solid mechanics, electromagnetism, geophysics and seismology (De Hoop, 1995; Pike and Sabatier,
2001). Information more specific to geological materials under seismically induced waves, which would be the basic field of
application of the present work, can be found in Keilis-Borok et al. (1989). In general, there has been a surge of work in the last
two decades regarding numerical work on the elastodynamics of various types of man-made composites and naturally occurring
media (Nayfeh, 1995). A review of the literature can be found in some of the earlier papers by the present authors (e.g., Manolis

* Corresponding author.
E-mail address: gdm@civil.auth.gr (G.D. Manolis).

0997-7538/$ – see front matter  2005 Elsevier SAS. All rights reserved.
doi:10.1016/j.euromechsol.2005.05.002
T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836 821

et al., 1999, 2002, 2004). In here, we will focus on the more recent work that is specific to anisotropic, inhomogeneous solids
(or both), with and without cracks, and to the derivation of fundamental solutions for the corresponding wave equation.

Anisotropic media. The boundary element method (BEM) has been used by Ahmad et al. (2001) to study steady-state vibra-
tions of 2D surface foundations in an orthotropic sub-base and by Chuhan et al. (2004) for transient vibrations of underground
structures in 2D orthotropic media. Also, free vibrations of anisotropic structural sheets were studied using the BEM by Albu-
querque et al. (2003), while Dravinski and his coworkers (Dravinski and Wilson, 2001; Niu and Dravinski, 2003) have looked
at time-harmonic response of 2D basins and 3D cavities in anisotropic media by an indirect version of the BEM. Next, integral
equation methods have been employed for reconstructing the elastostatic field (Dong et al., 2004) and the time-harmonic wave
field (Lee et al., 2004) in the elastic half-plane containing anisotropic inclusions. Furthermore, Rubio-Gonzalez and Manzon
(1999) presented an analysis of crack problems in orthotropic media under impact loads, while more recently Albuquerque et al.
(2004) developed the dual BEM for anisotropic dynamic fracture mechanics. Finally, Sharma (2002) computed group velocities
for 3D wave propagation in general anisotropic media, which is an inverse problem whose solution is based on information
along the directions of ray travel.

Inhomogeneous media. Functionally graded materials (FGM) are man-made composites that form a special class of inhomo-
geneous media, and their dynamic behavior has been the focus of substantial recent work. Reddy and Cheng (2003) employed
3D asymptotic theory in a transfer matrix setting to study harmonic vibrations of FGM plates, while Santare et al. (2003) per-
formed a finite elements analysis of elastic wave propagation through 2D continuously inhomogeneous structural components.
Transient wave propagation in strips of FGM was also studied by Berezovski et al. (2003) using an algorithm based on the nu-
merical solution of systems of hyperbolic equations, while a combination of analytical solutions and dual integral equations was
used by Aizikovich et al. (2002) for the spherical indentation problem of a half-plane with depth-dependent properties. Next,
the generalized BEM with internal collocation points was used for the static analysis of non-homogeneous solids under both 2D
and 3D conditions by Chen et al. (2001). A hybrid method, combining finite differences, finite elements and the discrete wave
number method, was introduced by Moczo et al. (1997) for seismic wave motion in inhomogeneous, viscoelastic topography
under time-harmonic conditions. Wang et al. (2002) considered an FGM strip with a crack perpendicular to the boundary by
dividing the strip into layers with homogeneous properties along the thickness direction, while in subsequent work Wang et al.
(2003) studied anti-plane fracture of such a strip. Finally, Auriault (2002) used multi-scale asymptotic expansions to represent
the evolution of physical processes in heterogeneous porous media by equivalent homogeneous ones.

Inhomogeneous and anisotropic media. There is a relatively small amount of work dealing with the combined inhomogeneous-
anisotropic material. At this point, we mention Ang et al. (2003), who employed the dual-reciprocity BEM for elastostatic
problems. Also, Ozturk and Erdogan (1997), as well as Chen et al. (2002), investigated mode I crack problems in an inhomoge-
neous orthotropic medium, with the former assuming static and the latter transient conditions. Finally, fracture of a plane FGM
orthotropic strip containing an edge crack as well as an internal crack perpendicular to the boundaries, was studied by Guo et
al. (2004). To the author’s best knowledge, in-plane elastic wave propagation problems involving inhomogeneous anisotropic
materials have not been treated so far.

Fundamental solutions. The mathematical background behind wave motion in non-homogeneous media involves solution of
partial differential equations with variable coefficients. In general, these equations do not possess explicit and easy to calculate
fundamental solutions (or Green’s functions), which prevents reduction of the physical boundary-value problem (BVP) to a
system of boundary-integral equations (BIE) that is processed by standard numerical quadrature techniques. The key role played
by the fundamental solution in a BEM formulation is to reduce a given BVP into a system of BIE through use of reciprocal
theorems (Beskos, 1987). It is for this reason that the recovery of fundamental solutions in analytical form, or at least in an
easy to calculate numerical form, is so important. In elastodynamics, we can identify the following ways to obtain fundamental
solutions for inhomogeneous continua:
(i) Solution of the original partial differential equation. Unfortunately, such solutions are generally not available. Exceptions
to this are Hook (1962), who succeeded in obtaining a Green’s function in closed-form for a vertically inhomogeneous medium
with constant velocity gradient and Watanabe and Takeuchi (2002), who derived a Green’s function for two-dimensional waves
in a radially inhomogeneous solid with elastic moduli and density that vary in the radial direction.
(ii) Use of available fundamental solutions for homogeneous materials. There are two possibilities here, namely: (a) Re-
duction of the partial differential equation with variable coefficients to one with constant coefficients. This can be done by
using algebraic transformations, as for instance by Azis and Clements (2001) who derived BIE for the static deformation
of inhomogeneous and anisotropic materials and by Manolis and Shaw (1996, 2000), who derived Green’s functions for
2D and 3D non-homogeneous continua. (b) Use of the dual-reciprocity BEM, whereby the resulting integral formulation
includes both surface and domain integrals. The latter integrals can be numerically processed, or alternative formulations
822 T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836

can be use to convert them to surface integrals. As examples, we mention various approaches for the mechanics of non-
homogeneous isotropic media (including heat conduction, poroelasticity, etc.) by Clements and co-workers (Clements, 1980;
Ang et al., 1997), Rangogni (1987), Kassab and Divo (1996), Park and Ang (2000), Ang et al. (2003), Tanaka et al. (2001),
Cheng (1987) and Gipson et al. (1995).
(iii) Approximate fundamental solutions. Here matrix formulations are introduced that correctly describe the main part of
the fundamental solution, but are not required to satisfy the original differential equations, apart from the singular point. This
approach does not reduce the problem to surface-only representations, but domain integrals are also introduced. Alternatively,
integro-differential equations are produced that require special numerical treatment (Beskos, 1997; Mikhailov, 2002; Bai et al.,
2002). In all cases, numerical implementation requires discretization of both boundary and interior domain of the problem in
question.
In this work, we derive fundamental solutions for a point force in a domain whose material properties are dependent on both
direction and position. Also, the density is assumed to be depth-dependent. The analytical methodology that has been developed
for this purpose combines an algebraic transformation with the Radon transform. Both techniques have been used separately
by the authors in previous work, but their combination allows closed-form solutions for media that are both anisotropic and
inhomogeneous. Briefly, the main points are as follows: (a) The algebraic transformation is applied to the displacement vector
and the governing equation of motion is transformed into an equation with constant coefficients; (b) the Radon transformation
reduces the system of coupled partial differential equations to a system of coupled ordinary differential equations. Specifically,
it is the ellipticity property of the anisotropic differential operator stemming from a positive-defined tensor of material constants
that allows this uncoupling to take place; (c) subsequent application of the inverse Radon transform yields the displacement
fundamental solution in terms of an integral over the unit circle; and finally (d) the asymptotic forms of the displacement field
and of the corresponding stress field are derived for small arguments. It should be mentioned here that the specific form of the
fundamental solutions turns out to be highly dependent on the interplay of numerical values for the various material parameters
of the medium in question.

2. Problem statement

Consider Cartesian coordinate system Ox1 x2 in R2 and let Ω be an inhomogeneous anisotropic domain, as shown in Fig. 1.
This domain has mass density ρ(x) and its mechanical behavior is defined through elastic tensor Cij kl (x), which is symmetric
and positive definite (Crouch, 1976):
Cij kl = Cj ikl = Cklij . (1)
Furthermore, Cij kl gij gkl > 0 for every non-zero, real and symmetric tensor gij . Assume now that all material parameters vary
in the same fashion with coordinate x = (x1 , x2 ) as
0 h(x),
Cij kl (x) = Cij ρ(x) = ρ0 h(x) (2)
kl

Fig. 1. Non-homogeneous material under plane-strain conditions.


T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836 823

where
h ∈ C 2 (Ω), h(x)  h0 > 0, x ∈ Ω. (3)
The governing equations of motion for this domain, written in terms of the displacement vector ui , in the absence of body
forces and under time-harmonic conditions, are as follows:
σij,j (x, ω) + ρ(x)ω2 ui (x, ω) = 0. (4)
The stress tensor is defined here as σij (x, ω) = Cij kl (x)uk,l (x, ω), while σij,j are its spatial derivatives and ω is the frequency
of vibration.
If Eq. (4) is to be treated by integral equation formulations, then a special solution for the point load needs to be recovered.
This fundamental solution is non-unique, but defined with respect to any function that satisfies Eq. (4). Furthermore, the con-
dition specified in Eq. (3) places a restriction on the domain of definition of the fundamental solution to a strip or a half-plane.
On the boundaries of such domains, h(x) = 0 and Eq. (2) indicates the material parameters to be zero; consequently, Eq. (4)
degenerates at these boundaries. These types of solutions, besides being useful on their own right, form the basic ingredient
(i.e., the kernel functions) of BEM formulations that have been most successful in dealing with problems involving semi-infinite
domains (Beskos, 1997). In general, if the coefficients of Eq. (4), which is of the elliptic type, are analytical functions, then
a fundamental solution exists (John, 1955). The corresponding BVP that yields the fundamental displacement tensor solution
denoted by the (*) superscript is
σj∗ki,i + ρω2 u∗kj = −ekj δ(x − ξ ) (5)
where σj∗ki = Cj kpq u∗ip,q , δ is Dirac’s generalized function and ekj is the unit tensor. Thus, our aim in this work is to identify
a class of functions h(x) and a range of values for constants Cij0 , ρ , ω for which fundamental solution u∗ = {u∗ } can be
kl 0 kj
derived in a closed form that is suitable for numerical implementation within the BEM.

3. Transformation of the governing equation

We will employ the algebraic transformation used for isotropic materials by Manolis et al. (1996, 1999) and more recently
by Azis and Clements (2001) for the anisotropic elastostatic case. Specifically, Eq. (4) can be transformed into an equation
with constant coefficients under some additional restrictions besides those given in Eq. (3) for function h(x). The key step is to
introduce a smooth transformation of ui in Ω as
ui = h−1/2 (x)Ui (6)
so that the homogeneous part of Eq. (4) can be written in terms of the transformed displacement Ui as follows:
 1/2 
Cij kl (x) Uk,j l + h−1/2 (h,j Uk,l − h,l Uk,j − Uk h,j l ) + ρ(x)ω2 Ui = 0.
1/2 1/2
(7)
By reducing common factor h(x) in both Cij kl (x) and ρ(x), and using Eqs. (2), (3), we obtain
0 U −1/2 (h1/2 U − h1/2 U ) − h−1/2 h1/2 U  + ρ ω2 U = 0.
Cij kl k,j l + h ,j k,l ,l k,j ,j l k 0 i (8)
0
We will now specify additional constraints on h and Cij kl under which Eq. (8) has constant coefficients. Suppose there exist
constants pik , qi such that
0 h 1/2 1/2 0 0 )h 1/2 1/2 .
Cij kl ,j l = pik h and (Ci1k2 − Ci2k1 ,k = qi h (9)
0 h−1/2 h 1/2
,j l Uk + ρ0 ω Ui ) has constant coefficients with respect to Ui if
Then, expression (−Cij 2
kl

0 h−1/2 h 1/2
(a) Cij kl ,j l = pik .

Also note that


0 h−1/2 (h 1/2 1/2 0 −1/2 (h 1/2 1/2
0 −1/2 (h 1/2 1/2
Cij kl ,j Uk,l − h,l Uk,j ) = Ci1k2 h ,1 Uk,2 − h,2 Uk,1 ) − Ci2k1 h ,2 Uk,1 − h,1 Uk,2 )
0 h−1/2 (∇h × ∇U ),
= Cik k
0 = C0 0
where Cik i1k2 − Ci2k1 and (×) denotes the vector product. This last expression has constant coefficients with respect to
the gradient of transformed displacement ∇Uk if either
824 T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836

0 = 0, without additional constraints on h(x), or


(b) Cik
(c) h−1/2 ∇h1/2 = d, where d is a constant vector, without additional constraints on Cij
0 .
kl

In sum, it is either under combinations (a) and (b) or (a) and (c) that Eq. (9) is fulfilled, in which case Eq. (8) becomes a partial
differential equation with constant coefficients.
We note here that combination (a) and (b) with pik = 0 was used in Azis and Clements (2001) for the elastostatic case.
For the isotropic case (Manolis et al., 2004), the first condition translates as λ0 = µ0 for the Lamé constants. Furthermore,
condition (b) simply imposes additional symmetry on the elastic tensor, while condition (c) is essentially a restriction on the
class of available functions h(x) and on the domain Ω. For the orthotropic case and with the principal elastic axes parallel to
the coordinate axes, C16 = C26 = 0, and the independent material constants are now four: C11 , C12 , C22 , C66 . We use here
compact notation (Su and Sun, 2003), whereby coefficients CI J are obtained from tensor Cij kl by using the following rule:
(11) ↔ 1, (22) ↔ 2, (12 = 21) ↔ 6. Finally, condition (b) implies C12 = C66 .

3.1. Material profiles

Combination (a) and (b), which from now on will be labeled as ‘Case A’, yields the following possibilities for h(x) that
represent inhomogeneous material profiles:

(1) h(x) = e2(a,x+b) , where a, x = a1 x1 + a2 x2 is the scalar product, in Ω = R2 ;


(2) h(x) = sinh2 (a, x + b), in the half-plane Ωδ = {x, a, x + b  δ > 0};
(3) h(x) = cosh2 (a, x + b), in Ω = R2 ;
(4) h(x) = sin2 (a, x + b), in the strip Ωδ1 ,δ2 = {x, 0 < δ1  a, x + b  δ2 < π };
(5) h(x) = cos2 (a, x + b), in the strip Ωδ1 ,δ2 = {x, − π2 < δ1  a, x + b  δ2 < π2 };
(6) h(x) = (a, x + b)2 , in the half-plane Ωδ = {x, a, x + b  δ > 0}.

For combination (a) and (c), labeled ‘Case B’, the only possible example is again

(1) h(x) = e2(a,x+b) in Ω = R2 .

We note that the above material profiles might correspond to special classes of composite structures or to certain categories of
geological deposits.

3.2. Transformed equations

By algebraically transforming Eq. (5) using Eq. (6) and conditions (9), we recover the following governing equation in lieu
of Eq. (8):
  ∗
h1/2 (x) Mik (∂ 2 ) + Nik (∂) + Γik Uks = −δis δ(x, ξ ), (x, ξ ), ∈ Ω × Ω, (10)
0 ∂ ∂ , N (∂) = C 0 (q ∂ − q ∂ ) are differential operators and Γ = δ ρ ω2 − p = ρ ω2 I − P (q).
where Mik (∂ 2 ) = Cij kl j l ik ik 1 2 2 1 ik ik 0 ik 0 2
Dividing Eq. (10) by h1/2 (x) and keeping in mind that h−1/2 (x)δ(x, ξ ) = h−1/2 (ξ )δ(x, ξ ), since the support of δ(x, ξ ) is on
(x = ξ ), we obtain
  ∗
Mik (∂ 2 ) + Nik (∂) + Γik Uks = −δis h−1/2 (ξ )δ(x, ξ ), (x, ξ ) ∈ Ω × Ω. (11)

The above is a system of two coupled linear partial differential equations of second order with constant coefficients, whose
solution will be recovered by use of the Radon transformation.

4. Radon transform

The Radon transform is a powered tool for deriving fundamental solutions of elastodynamics. As examples, see Franciosi
and Lormand (2004) for inclusion problems in 3D elastostatics, Georgiadis and Lycotrafitis (2001) for a steadily traveling point
load on the surface of the half-space, Wang and Achenbach (1994) and Rangelov (2003) for anisotropic homogeneous media,
and finally Manolis et al. (2002, 2004) for isotropic inhomogeneous media. For completeness, we give the definition plus some
T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836 825

basic properties of the Radon transform (Ludwig, 1966; Zayed, 1996), which is defined in R2 for f ∈ , the set of rapidly
decreasing C ∞ functions, as follows:
 
 
R(f ) = fˆ(s, m) = f (x) dS = f (x)δ s − x, m dx, s ∈ R1 , m ∈ S 1
x,m=s

is the direct transform and


 
    1 
f (x) = R−1 fˆ(s, m) = R∗ K(fˆ ) = R∗ (f˜ ) = K(f )(s, m)
ˆ dm
4π 2 s=m,x
|m|=1

is the inverse transform, where


∞
∂σ fˆ(σ, m)
K(fˆ ) = dσ.
s−σ
−∞

The transform is linear, which implies that R(af (y) + bg(y)) = aR(f (y)) + bR(g(y)). Furthermore, if L(∂) is a homogeneous
differential operator of degree k with constant coefficients, then
  ∂k
R L(∂)f (x) = L(m) k fˆ(s, m).
∂s
The Radon transform is defined on the space of distributions and R(δ(x, ξ )) = δ(s − m, ξ ).
Applying the Radon transform to both sides of Eq. (11) yields
  ∗  
M(m)∂s2 + N (m)∂s + Γ U  (s, m, ω) = −h−1/2 (ξ )δ s − m, ξ  I2 , (12)
0 m m , N = {N (m)}, N (m) = C 0 (d ×m), Γ = {Γ (ω)},
where I2 is the unit matrix in R2 , M = {Mik (m)}, Mik (m) = Cij kl j l ik ik ik ik
Γik (ω) = δik ρ0 ω2 − pik , and variable m ∈ R2 , |m| = 1 being the unit circle.
The above is a system of two second-order, ordinary differential equations with constant coefficients. Under the conditions
specified in Eq. (1), matrix M is symmetric and positive definite, matrix N is skew-symmetric and matrix Γ is symmetric. In
 by diagonalizing matrices M and Γ (see Rangelov, 1992). Specifically,
order to solve Eq. (12), we recover its canonical form
we start with the eigenvalues α1,2 (m) = 12 (Tr M ± (Tr M)2 − 4 det M) of M, where Tr is the trace and det is the determinant.
For an isotropic material, α1 = λ + 2µ, α2 = µ, while for the anisotropic material, α1,2 depend on m, |m| = 1. Next, let g1 , g2
be the two corresponding normalized eigenvectors of M so that orthogonal matrix
1

g1 g21
G=
g12 g22
transforms the basis to canonical form as

α1 0
G−1 MG = A = .
0 α2
∗ = GV
Now define U , in which case the transformed displacement vector V
 satisfies equation
 
[MG∂s2 + N G∂s + Γ G]V  = −h−1/2 (ξ )δ s − m, ξ  I2 . (13)

Left-hand side multiplication of Eq. (13) by G−1 gives


 
[A∂s2 + N ∂s + Γ1 ]V = −h−1/2 (ξ )δ s − m, ξ  G−1 , (14)
where the commutative property of skew-symmetric matrix N and of orthogonal matrix G has been used to give N G = GN ,
with Γ1 = G−1 Γ G. Matrix A is strictly positive for every m, |m| = 1, so A1/2 exists. A subsequent displacement vector
transformation, denoted by W , is introduced in Eq. (13) yielding
 = A1/2 V
 
[A1/2 ∂s2 + N A−1/2 ∂s + Γ1 A−1/2 ]W
 = −h−1/2 (ξ )δ s − m, ξ  G−1 . (15)

An additional left-hand side multiplication of Eq. (15) by A−1/2 results in


 
[∂s2 I2 + Q∂s + Γ2 ]W = −h−1/2 (ξ )δ s − m, ξ  A−1/2 G−1 . (16)
826 T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836

Observe Q = A−1/2 N A−1/2 is skew-symmetric, i.e. QT = (A−1/2 N A−1/2 )T = −A−1/2 N A−1/2 = −Q with superscript
(T ) denoting the transpose. Also, matrix Γ2 = A−1/2 Γ1 A−1/2 is symmetric, has two real eigenvalues η1,2 (m) = 12 (Tr Γ2 ±

(Tr Γ2 )2 − 4 det Γ2 ) and the corresponding normalized eigenvectors are e1 , e2 . Use of the orthogonal matrix
1

e e21
E= 1
e12 e22
transforms the basis to canonical form as

η1 0
E −1 Γ2 E = R = .
0 η2
 = E −1 W
A third (and final) displacement vector transformation, defined as Z  , is introduced in Eq. (16) resulting in
 
[∂s2 E + QE∂s + Γ2 E]Z  = −h−1/2 (ξ )δ s − m, ξ  A−1/2 G−1 . (17)
As before, pre-multiplication of the above equation by E −1 yields
 = δ(s − τ )F,
[∂s2 I2 + Q∂s + R]Z (18)
where F = −h−1/2 (ξ )E −1 A−1/2 G−1 , and τ = m, ξ . At this point, it is important to note that matrices Γ0 = M −1/2 Γ M −1/2
and Γ2 have the same eigenvalues, namely η1 , η2 , since the characteristic polynomials of Γ0 and Γ2 are the same.
In what follows, ‘Case A’ and ‘Case B’ shall treat separately in deriving fundamental solutions for Eq. (17), because matrix
Q = 0 in the former case, while Q = 0 in the latter case and we no longer have a simple second-order equation.

5. Fundamental solution for ‘Case A’

With matrix Q = 0, Eq. (18) assumes the form given below


 = F δ(s − τ )
[∂s2 I2 + R]Z (19)
and because matrix R is diagonal, Eq. (19) is uncoupled. As previously noted, since the elements ηj of matrix R are invariants,
the eigenvalues of matrix M −1/2 Γ M −1/2 depend on ρ0 , ω, pik and on m, |m| = 1. Using Silvester’s theorem (Rangelov,
1992) and compactness arguments with respect to m, we ascertain that the number of positive, negative and zero eigenvalues is
preserved. Therefore, the following five different sub-cases can be identified:
(i) η1  η2 > 0, (ii) η1 > 0, η2 = 0, (iii) η1 > 0, η2 < 0, (iv) η1 = 0, η2 < 0, (v) 0 > η1  η2 . (20)
The fundamental solutions corresponding to these five sub-cases behave differently. For instance, we note the possibility of
double eigenvalues for arbitrary values of m for sub-cases (ii) and (iv). Which sub-case is actually realized depends on the
elasticity constants (through matrix M), on the degree of inhomogeneity (through function h), on the density and frequency
(through matrix Γ ). For example, if matrix P = {pik } is negative, only sub-case (i) materializes; if P is positive, then all five
sub-cases can occur, with (i) corresponding to large values of frequency ω, (v) to small values of ω, and (ii)–(iv) to intermediate
values of ω. A rather careful investigation of this problem will be carried out in the Appendix using a simpler operator, namely
the Laplacian. In the quasi-homogeneous case, matrix P = 0 and only sub-case (i) can be realized (Manolis et al., 2004). In the
purely homogeneous case, it is also true that P = 0 and two wave numbers are recovered as

 ρ0
kj = ηj = ω.
αj (m)

Finally, wave numbers kj are independent of variable m in the isotropic case only (Rangelov, 2003).
Therefore, in view of the above, Eq. (19) can be re-written as
[∂s2 + η]v̂ = f δ(s − τ ), f = h−1/2 (ξ )f0 (m, ω), (21)
where parameter η can be positive, zero or negative. In order to find the fundamental solution of Eq. (11), which is one step away
from the final form, we start by solving Eq. (21) to recover v̂ for different values of η. Next, we compute the first component of
the inverse Radon transform as ũ = K(v̂), which is subsequently integrated with respect to m, |m| = 1. This step completes the
inverse Radon transformation and yields U ∗ in terms of the original basis functions ũ. It is also possible to compute the spatial
derivatives of U ∗ . Finally, the simple algebraic transformation specified in Eq. (6) produces u∗ , the fundamental solution of the
original equation of motion.
Following Vladimirov (1984), Gelfand et al. (1966) and using calculus of distributions we recover v̂ and ũ = K(v̂) for the
following three possibilities regarding parameter η:
T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836 827

(a) Possibility η is positive (superscript +). Let η > 0, or η = k 2 . Then, v̂ + = α eik|s−τ | , α = −if/(2k) and let γ + = −f/2.
We compute
   
ṽ + (s) = K(v̂ + ) = γ + cos kz − 2 ci(kz) cos(kz) + si(kz) sin(kz) z=|s−τ | , 


 (22)
2    
∂s ṽ + = γ + −k sin kz − + 2k ci(kz) sin(kz) − si(kz) cos(kz)  sgn(s − τ ), 
z z=|s−τ |
where
∞ ∞
cos t sin t
ci(p) = − dt, si(p) = − dt
t t
p p

are the cosine and sine integral functions, respectively (Bateman and Erdelyi, 1953).

(b) Possibility η is zero (superscript 0). Let η = 0, and then v̂ 0 = (f/2)|s − τ |, with
 
ṽ 0 = K(v̂ 0 ) = f ln zz=|s−τ | , 


 (23)
f 
∂s ṽ 0 =  sgn(s − τ ). 
z z=|s−τ |

(c) Possibility η is negative (superscript −). Let η < 0, η = −k 2 , k > 0. Then v̂ − = α e−k|s−τ | , α = −f/(2k) and γ − = f/2.
We compute
   
ṽ − (s) = K(v̂ − ) = γ − ch(kz) + 2 chi(kz) ch(kz) − shi(kz) sh(kz) z=|s−τ | , 


 (24)
2    
∂s ṽ − = γ − k sh(kz) + + 2k chi(kz) sh(kz) − shi(kz) ch(kz)  sgn(s − τ ), 

z z=|s−τ |
where
z z
ch t − 1 sh t
chi(z) = dt + ln z, z > 0, shi(z) = dt
t t
0 0
respectively are the hyperbolic cosine and sine integral functions (Bateman and Erdelyi, 1953).

5.1. Inverse Radon transform

∗ = GV
Keeping in mind that U  = GA−1/2 W  = GA−1/2 E Z,
 F = −h−1/2 (ξ )E −1 A−1/2 G−1 , τ = m, ξ , and U
∗ =
GA−1/2 EK(Z)
 gives the inverse Radon transform as


∗ ) = 1
U ∗ = R−1 (U ∗ (z)
U
4π 2 z=|m,x−ξ | dm. (25)
|m|=1

By denoting
1
1

f1 ũ1 f12 ũ1 t1 t12


 =
K(Z) and GA−1/2 E = ,
f21 ũ2 f22 ũ2 t21 t22
Eq. (25) for the fundamental solution and its spatial derivative assumes the form
 t 1 t 2
f 1 ũ f 2 ũ
 
∗ 1 1 1 1 1 1 1  

U (x, ξ ) =  dm, 

4π 2 1
t2 t22 f2 ũ2 f22 ũ2 z=|m,x−ξ |
1 

|m|=1 

(26)
 1
1
 

t1 t12 f1 ∂z ũ1 f12 ∂z ũ1    
∗ (x, ξ ) = 1  mk sgn m, x − ξ  dm, 

U,k  

4π 2 t21 t22 f21 ∂z ũ2 f22 ∂z ũ2 z=|m,x−ξ | 
|m|=1
828 T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836

where
 

 ũ+ , ηj > 0 
 ∂s ũ+
j , ηj > 0,

 j 

ũj (s, m, ξ ) = h−1/2 (ξ ) ũ0j , ηj = 0 and ∂s ũj (s, m, ξ ) = h−1/2 (ξ ) ∂s ũ0j , ηj = 0,

 


 ũ− , η < 0 
 ∂s ũ− , η < 0.
j j j j

The aforementioned five sub-cases are now


(i) ũj = ũ+
j , (ii) ũ1 = ũ+ 0
1 , ũ2 = ũ2 , (iii) ũ1 = ũ+ −
1 , ũ2 = ũ2 , (iv) ũ1 = ũ01 , ũ2 = ũ−
2, (v) ũj = ũ−
j .
Formulae (26) can be further simplified as
 g 1 g 1
g 1 ũ
 
1 1 1 g12 ũ1  
U ∗ (x, ξ ) = 1 2  dm, 

4π 2  

g12 g22 g21 ũ2 g22 ũ2 z=|m,x−ξ | 

|m|=1 
(27)
 1
1
 

g1 g21 g1 ∂z ũ1 g12 ∂z ũ1    
∗ (x, ξ ) = 1  mk sgn m, x − ξ  dm, 

U,k  

4π 2 g12 g22 g21 ∂z ũ2 g22 ∂z ũ2 z=|m,x−ξ | 
|m|=1

with gj = (gj1 , gj2 ) and ũj = ũ+


j for the anisotropic homogeneous case (Rangelov, 2003), while g1 = (m1 , −m2 ), g2 =
+
(m2 , m1 ) and ũj = ũj for the isotropic inhomogeneous case (Manolis et al., 2004).
Following Manolis et al. (2004) and combining Eqs. (6) and (26), we obtain the two fundamental solutions for the original
equation of motion, along with the necessary expressions for the spatial derivatives as follows:

u∗ij (x, ξ ) = h−1/2 (x)h−1/2 (ξ )Uij∗ (x, ξ ), 

(28)
0 u∗z (x, ξ ), z = x or ξ, 
σij∗zk (x, ξ ) = h(z)Cij 
ml mk,l
where
 −1/2  −1/2 
u∗x
mk,l (x, ξ ) = h (x) ,l h ∗ + h−1/2 (x)h−1/2 (ξ )U ∗x , 
(ξ )Umk mk,l 
(29)
∗ξ   ∗ + h−1/2 (x)h−1/2 (ξ )U ∗ξ . 

umk,l (x, ξ ) = h−1/2 (ξ ) ,l h−1/2 (x)Umk mk,l
Using the properties of the trigonometric and hyperbolic integral functions for small arguments we can now obtain the asymp-
totic form of the fundamental solutions. When field point x → ξ , continuity in h implies that h(x) = h(ξ ) + O(|x − ξ |). Thus,
from Eq. (28) it is possible to derive
∗ asym ∗ asym 
uij = h−1 (ξ )Uij = h−1 (ξ )bij ln |x − ξ |, 







∗ asym ∗ asym   1 
σij m = Cij kl umk,l = Cij kl (ξ ) h−1/2 (ξ ) h−1/2 (ξ ) ,l bmk ln |x − ξ | + h−1 (ξ )ηmkl
|x − ξ |  (30)




1 −1   1 

0
= Cij kl − h (ξ ) h(ξ ) ,l bmk ln |x − ξ | + ηmkl . 

2 |x − ξ |

Remark 1. The fundamental solutions for the inhomogeneous case depend not only on the relative distance (x − ξ ) between
field and source points as in the homogeneous case, but also on x and ξ separately. Also, the influence of inhomogeneity on
the displacement fundamental solution is felt through multiplication of the corresponding solution for the homogeneous case
by factor h−1/2 (x)h−1/2 (ξ ). In other cases that will be examined, inhomogeneity influences Uij∗ (x, ξ ) in more complex ways,
namely through the wave numbers and through the type of functions that comprise ũj .

Remark 2. For given values of elasticity tensor Cij 0


kl and for specific types of functions h(x), the conditions implied by
‘Case A’ are easy to check and if true, to proceed and find the corresponding fundamental solution in closed form. We note that
an interesting mathematical task is to start with a given Cij0 tensor satisfying Eqs. (1), (2) and condition (b) (see Eq. (9)) and
kl
then identify the set of all real functions h(x) with properties specified by Eq. (3) in domain Ω such that the over-determined
system of equations (a) is solvable. Although Section 3.1 showed a partial list of functions h(x), it is an open question if
additional functions can be found that are also acceptable.
T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836 829

Remark 3. Three basic mathematical tools were used in deriving a fundamental solution for the anisotropic inhomogeneous
domain, namely the Radon transform, linear algebra and theory of ordinary differential equations. In addition, Eq. (1) is a
condition expressing that internal energy density must remain positive; since this energy must be minimal in a state of equilib-
rium, it is useful in deriving the fundamental solution. Our approach would be complete for the homogeneous anisotropic case
where the elasticity tensor has constant coefficients, but as we move on to the inhomogeneous case, the fundamental solution is
defined in domain Ω with function h(x) specified by virtue of Eq. (3) only. Consider now a given h(x) such that parameter η
is either zero (superscript 0) or negative (superscript −). The equation of motion (4) has variable coefficients and is elliptic in
terms of frequency parameter ω, provided the zeroth order term is positive (which it is, since density is always a positive-valued
function). Following the Radon transformation of Eq. (4), we still obtain an elliptic equation but the sign of the zeroth order
term changes to negative when η < 0 or vanishes when η = 0. Now, although Eq. (4) is defined in the frequency domain, we
actually started from a hyperbolic equation and made the assumption that the solution can be represented in harmonic form as
u(x, t) = u(x, ω) e−iωt . Thus, the following inconsistency appears: either form u(x, ω) e−iωt is reasonable for large values of
frequency that lead to the η > 0 case and for small values of the frequency that are likely to lead to the η  0 cases, or repre-
sentation u(x, ω) e−iωt as a solution of the original hyperbolic equation is not correct. This last point implies that the equation
of motion in the time domain has changed type because of the presence of inhomogeneity. This inconsistency does not go away
if we change tactics and start with the equation of motion in the time domain, because following the Radon transformation
path there are still zero-order terms with a priori unknown signs that will influence the behavior of the transformed equation
with constant coefficients. In sum, it appears that the three different cases (η < 0, η = 0, η > 0) might correspond to different
mechanical models. These points are further explored in the Appendix by using the simpler Poisson’s equation as the vehicle.

Remark 4. The present method for finding a fundamental solution can be also applied to the elastostatic case by setting ω = 0
in Eq. (4). This case is much simpler and gives Γ = −P (q) in Eq. (10), while the type of solution obtained is dependent on
matrix Γ only. Now, if P (q) < 0 and h(x) is either material profile no. (4) or no. (5) from Section 3.1, then we have sub-case (i)
and both solutions come from the η > 0 possibility. If P (q) = 0 and h(x) is material profile no. (6), both solutions come from
using η = 0 (see (Azis and Clements, 2001)). If P (q) > 0 and h(x) corresponds to material profiles no. (1)–no. (3), we have
sub-case (v) and both solutions use η < 0. Finally, matrix P (q) is basically the same as matrix M(q), but vector q is purely
imaginary for P (q) < 0, real for P (q) > 0 and equal to zero for P (q) = 0.

6. Fundamental solution for ‘Case B’

All solutions for the spatial profile of the inhomogeneity for this case must be compatible with condition (c) for a given real
vector q (see Eq. (9)). Thus, we have h1/2 (x) = eq,x+q0 , with q0 arbitrary. Given this form, condition (a) (see Section 3) is
satisfied with pik = −Cij 0 q q , so that P (q) = −M(q). Then, Eq. (18) becomes
kl j l
  0 (q m − q m ) √ 1
Q = Qik (m) , Qik = (i − k)C12 1 2 2 1 , (31)
α1 (m)α2 (m)
where R = {Rik (m, ω)}, Rik = δik ηi . It is not possible to uncouple the above system of equations, so we will transform it into
a system of four, first-order equations.
Denote Yij = Zij , Y
i+2j = ∂s Zij (i, j = 1, 2) and define matrices

0 I2 0
B= and S = .
−R −Q F
Eq. (18) is then transform into the following 4 × 4 system:
 = BY
∂s Y  + Sδ(s − τ ). (32)
In order to solve the above system for every fixed m, |m| = 1, we follow Vladimirov (1984) and Fedoriuk (1980). We start
by transforming B into canonical form (i.e., diagonalization). The eigenvalues βj (m) of B are solutions of the characteristic
equation det |B − βI4 | = 0, which yields a bi-quadratic equation β 4 + (Tr R + det Q)β 2 + det R = 0. The four roots are
 
β1,2 = ± d1 , β3,4 = ± d2 ,

1 
d1,2 = − Tr R + det Q ± (Tr R + det Q)2 − 4 det R . (33)
2
For every fixed value of m, |m| = 1, there are four solutions βj as functions of m. Unfortunately, and in contrast to ‘Case A’
where Q = 0, we cannot control these roots using invariant conditions, because Q varies with m. Nevertheless, we know
830 T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836

that det Q  0 and four different kinds of roots βj are possible, namely real, zero, purely imaginary and complex. Therefore,
the types of solutions recovered for the homogeneous system ∂s Y  = BY may involve hyperbolic functions, polynomials or
trigonometric functions. Moreover, if multiple roots are manifested due to the presence of the Jordan cells, the solutions are
quasi-polynomials. In sum, the fundamental solution of Eq. (32) must be set-up for all possibilities, depending on the actual
values of m, and construction of the actual fundamental solution of Eq. (5) follows the inverse Radon transformation outlined
in Section 5.
There is a possibility that the type of the roots βj do not change with m, and for this case we can obtain the fundamental
solution in closed form. We will make an additional assumption regarding invariants ηj being strictly positive, i.e.,

η1  η2 > 0. (34)
0
This condition is always true for large frequencies ω. More specifically, for every combination of values of parameters qj , Cij kl
and ρ0 , there exists a value ω0 such that Eq. (34) holds for ω > ω0 . In this case, and in reference to Eq. (33), we have
 
d2 < d1 < 0 ∀m, |m| = 1 and β1,2 = ±i |d1 |, β3,4 = ±i |d2 |. (35)

Let lj = (lj1 , . . . , lj4 ) be normalized orthogonal eigenvectors corresponding to βj , j = 1, . . . , 4. Then,



1, j = k,
lj , lk  =
0, j = k

and 4 × 4 matrix L = {lji } transforms B to diagonal form as L−1 BL = Λ = {δij βj }. Define X


 = L−1 Y
 and pre-multiply
Eq. (32) by L−1 to obtain the system

∂s X  + L−1 Sδ(s − τ ).
 = ΛX (36)

The above system corresponds to four equations of the type ∂s w = β w + f δ(s − τ ). Following Vladimirov (1984), the solution
is w = H (s − τ )z(s), where H (s) is the Heaviside function and z(s) is a solution of the initial value problem

∂s z = βz, and z(s) = f. (37)

A unique solution of Eq. (37) is z = f eβ(s−τ ) , so that w = H (s − m, ξ )f eβ(s−m,ξ ) and the corresponding solution of
Eq. (36) is
 
Xkj = H s − m, ξ  fkj eβk (s−m,ξ ) (k = 1, . . . , 4, j = 1, 2). (38)

 = LX.
Finally, Y  Since Z ij (i, j = 1, 2) and U
ij = Y ∗ = GA−1/2 E Z,  then every U ∗ is a linear combination of functions of
ij
mn given in Eq. (38). The intermediate functions U
the type X ∗
 = GA −1/2 EK(Z)  required by the Radon transform are linear
combinations of the type

1   
kl = fkl
U − eβk (s−m,ξ ) Ei βk s − m, ξ   (k = 1, . . . , 4, l = 1, 2), (39)
s − m, ξ 
 x et
where Ei(x) = −∞ t dt is the exponential integral function defined for Re{x} ∈
/ R (Bateman and Erdelyi, 1953). By applying
the inverse Radon transform, we obtain the fundamental solution and its derivatives as follows:
 t1 t2

 
1 ũ11 ũ12  
U ∗ (x, ξ ) = 1 1  dm, 

4π 2 t21 t22 ũ21 ũ22 z=|m,x−ξ | 


|m|=1 

(40)
 t1 t2

 

1 ∂s ũ11 ∂s ũ12    


U,k (x, ξ ) = 1 1  mk sgn m, x − ξ  dm. 

4π 2 t21 t22 ∂ s ũ21 ∂s ũ22 z=|m,x−ξ | 

|m|=1

We note here that ũkl are of the type given in Eq. (39).
The asymptotic behavior for small arguments of the above fundamental solution and of its spatial derivatives is the same as
∗ . Finally, if the condition specified
before (see Section 5), with the leading terms being ln |x − ξ | for U ∗ and 1/|x − ξ | for U,k
in Eq. (34) is not fulfilled, then it is not possible to recover a unified form for the fundamental solution that covers sub-cases
(ii)–(v) of Eq. (20).
T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836 831

7. Inhomogeneous isotropic material

We will illustrate both ‘Case A’ and ‘Case B’ by considering a specific example involving the isotropic case, where inho-
mogeneity is of the form h(x) = e2q,x+q0 (see Section 3.1) with q1 > 0, q2 = q0 = 0.
All matrices involved in Eqs. (12)–(18) are given below for this specific type of material as

(λ0 + 2µ0 )m21 + µ0 m22 (λ0 + µ0 )m1 m2


M(m) = ,
(λ0 + µ0 )m1 m2 µ0 m21 + (λ0 + 2µ0 )m22

−(λ0 + 2µ0 )q12 0 0 (λ0 − µ0 )q1 m2


P (q) = = −M(q), N (m) = ,
0 −µ0 q12 −(λ0 − µ0 )q1 m2 0


ρ0 ω2 − (λ0 + 2µ0 )q12 0 λ0 + 2µ0 0 m1 m2


Γ (q, ω) = , A= , G= ,
0 ρ0 ω2 − µ0 q12 0 µ0 −m2 m1
 
0 √ (λ0 −µ0 ) q1 m2
µ0 (λ0 +2µ0 )
Q(q, m) =  ,
− √ (λ0 −µ0 ) q1 m2 0
µ0 (λ0 +2µ0 )

η1 0
R(m, ω) = and
0 η2
√ 1
 √ m1
λ0 +2µ0
0 m1 −m2 − √ m2
λ0 +2µ0 λ0 +2µ0
F = −e−q1 ξ1 = −e−q1 ξ1 .
0 √1 m2 m1 √m2 √m1
µ0 µ0 µ0

7.1. ‘Case A’ results


 
For this case, λ0 = µ0 , E = 10 01 and system of Eqs. (21) becomes
2

 m1
∂s + η1 0 v̂11 v̂12   √3µ0 − √m2
q ξ 3µ0
= −e 1 1 δ s − m, ξ  .
0 ∂s2 + η2 v̂21 v̂22 √m2 √m1
µ0 µ0

All sub-cases (i)–(v) listed in Eq. (20) can now be realized, depending on the actual values of material parameters µ0 , ρ0 , ω, q1 :

Sub-case (i): If ρ0 ω2 /(3µ0 ) − q12 > 0 and ρ0 ω2 /µ0 − q12 > 0, then all four solutions of Eq. (21) are as in case (+) (see
Eq. (20)).
Sub-case (ii): If ρ0 ω2 /(3µ0 ) − q12 = 0, i.e., q12 = ρ0 ω2 /(3µ0 ), then (ρ0 ω2 )/µ0 − (ρ0 ω2 )/(3µ0 ) = (2ρ0 ω2 )/3µ0 > 0 and
two solutions of Eq. (21) are as in case (+), and two more are as in case (0) .
Sub-case (iii): If ρ0 ω2 /(3µ0 ) − q12 < 0, but ρ0 ω2 /µ0 − q12 > 0, then two solutions of Eq. (21) are as in case (−), and two
more are as in case (+).
Sub-case (iv): If ρ0 ω2 /(3µ0 ) − q12 < 0, but ρ0 ω2 /µ0 − q12 = 0, then two solutions of Eq. (21) are as in case (−), and two
more are as in case (0).
Sub-case (v): If ρ0 ω2 /(3µ0 ) − q12 < 0 and ρ0 ω2 /µ0 − q12 < 0, then all four solutions of Eq. (21) are as in case (−).

It is seen that we can control the type of the fundamental solution we wish to recover by changing the value of the frequency ω.

7.2. ‘Case B’ results

In this case, λ0 = µ0 , and the system of Eqs. (18) becomes


 2 2 (λ0 −µ0 )q1 m2  
∂s + λ ρ+2µ

− q12 √
m1
− √ m2
µ0 (λ0 +2µ0 ) v̂ v̂12   √λ0 +2µ0 λ0 +2µ0
 0 0
 11 = −eq1 ξ1 δ s − m, ξ  .
(λ0 −µ0 )q1 m2 2
v̂21 v̂22 √m2 √m1
−√ ∂s2 + ρµ 0ω
− q12 µ0 µ0
µ0 (λ0 +2µ0 ) 0

The corresponding first-order system of Eqs. (32) comprises matrices of the following form:
832 T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836

   0 0 
v11 v12
 v v22   0 0 
= 
Y 
21 
,

S =  √ m1 √ −m2

,
 ∂s v11 ∂s v12   λ0 +2µ0 λ0 +2µ0 
∂s v21 ∂s v22 √m2 √m1
µ0 µ0
 
0 0 1 0
 0 0 0 1 
 
B =
 −η1 √ 0 −µ0 )q1 m2 
−(λ .
µ0 (λ0 +2µ0 ) 
0 0
 
(λ0 −µ0 )q1 m2
0 −η2 √ 0
µ0 (λ0 +2µ0 )
Having in mind that |m|  1 and after some calculations, it is observed that if ω2 > 2ρ0 (λ0 + 3µ0 )(λ0 + µ0 )2 q12 , then Eq. (34)
is fulfilled and Eq. (35) follows with d1 < 0, d2 < 0. For such a frequency ω value, sub-case (i) materializes and the fundamental
solution has the form given in Eq. (40).

8. Conclusions

In this work, the combined algebraic plus Radon transforms are applied to the governing time-harmonic equations of motion
for the general anisotropic, inhomogeneous material under plane strain conditions in an effort to recover certain classes of
fundamental solutions. The former transform address material inhomogeneity and yields a finite number of possible material
profiles. The latter transform is used to treat the equations of motion, which now posses constant coefficients, for a range
of direction-dependent elastic constants. Given the fact that the exact type of closed-form solutions depends on an interplay
of values of the material parameters (elastic constants, density, type of inhomogeneity) and of the wave frequency, numerical
implementation can only be done on a case-by-case situation and will be communicated in subsequent work. Furthermore, either
material model mentioned previously can be derived as a special case from the general formulation. Finally, the methodology
developed herein is equally applicable to the general three-dimensional case.

Acknowledgement

The authors wish to acknowledge the financial support provided through NATO Grant EST.CLG.980303.

Appendix A. Poisson’s equation

In order to check the methodology presented herein in more detail and to show what happens with the fundamental solution
as key parameter η sweeps through negative (superscript −), zero (superscript 0) or positive (superscript +) values, we focus
on Poisson’s equation
( + η)v ∗ = f δ(x − ξ ) (A.1)
where  = ∂x21 + ∂x22 is the Laplacian, while η, f are constants and v ∗ is the corresponding fundamental solution. By applying
the Radon transform to Eq. (A.1), where R(v) = ∂s2 R(v) = v̂, we obtain an equation similar to Eq. (19), i.e.,

(∂s2 + η)v̂ = f δ(s − τ ), τ = x, ξ . (A.2)


For all three cases (η < 0, η = 0, η > 0) we retrieve a solution of Eq. (A.1) and check it using calculus of generalized functions.
Of course, a given solution of Poisson’s equation is non-unique, since superposition of the homogeneous solution u (i.e.,
( + η)u = 0) yields a new function w∗ = v ∗ + u that again solves Eq. (A.1).

(a) Possibility η > 0 (superscript +). Let η = k 2 . Then, v̂ + = α eik|s−τ | with α = −if /(2k) satisfies Eq. (A.2). To prove this,
we have

∂s v̂ + = αik eik|s−τ | sgn(s − τ ) and


 2  
∂s2 v̂ + = −αk 2 eik|s−τ | sgn(s − τ ) + αi k eik|s−τ | sgn(s − τ ) = −αk 2 eik|s−τ | + αi k eik|s−τ | 2δ(s − τ )
= −k 2 v̂ + + f δ(s − τ )
T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836 833

since (sgn(s − τ )) = (H (s − τ ) − H (τ − s)) = δ(s − τ ) + δ(τ − s) = 2δ(s − τ ). The inverse Radon transform of v̂ + is
v ∗+ (x, ξ ) = R−1 (v̂ + ) = R∗ (K(v̂ + )) = 1/(4π 2 ) |m|=1 ṽ + (z)|z=|m,x−ξ | dm, where ṽ + is given in Eq. (22). To show that
function v ∗+ (x, ξ ) satisfies Eq. (A.1), evaluate

1   
∂xj v ∗+ = 2
∂z ṽ + (z)z=|m,x−ξ | mj sgn m, x − ξ  dm

|m|=1
 
1    1
= g1+ (z)z=|m,x−ξ | mj sgn m, x − ξ  dm + g2+ mj dm,
4π 2 4π 2
|m|=1 |m|=1

where

   2
g1+ = γ + −k sin kz − 2k ci(kz) sin(kz) − si(kz) cos(kz) , or g1+ = γ + − .
m, x − ξ 
Next,

1 
∂x2j v ∗+ = ∂z2 ṽ + (z)z=|m,x−ξ | m2j dm
4π 2
|m|=1
 
1  +  +    1
= ∂z g (z)
1
2 
z=|m,x−ξ | mj + g1 z=|m,x−ξ | mj 2δ m, x − ξ  dm + 4π 2 ∂xj g2+ m2j dm
4π 2
|m|=1 |m|=1
  m2j
k2 2γ +
=− ṽ + (m, x − ξ )m2j dm + dm.
4π 2 4π 2 m, x − ξ 2
|m|=1 |m|=1

Finally, we recover
 
k2 2γ + 1
v ∗+ = − ṽ + (m, x − ξ ) dm + dm = −k 2 v ∗+ + f δ(x, ξ ),
4π 2 4π 2 m, x − ξ 2
|m|=1 |m|=1

since δ(x, ξ ) = 1 2 |m|=1 1 dm in R2 (see John, 1955).
4π m,x−ξ 2

(b) Possibility η = 0 (superscript 0). If η = 0, function v̂ 0 = f2 |s − τ | satisfies Eq. (A.2) because ∂s v̂ 0 = f sgn(s − τ ) and
∂s2 v̂ 0 = f δ(s − τ ). The inverse Radon transform of v̂ 0 is

  1 
v ∗0 (x, ξ ) = R∗ K(v̂ + ) = ṽ 0 (z)z=|m,x−ξ | dm,
4π 2
|m|=1

where ṽ 0 was given in Eq. (23).


Now function v ∗0 (x, ξ ) satisfies Poisson’s equation because
 
f    f 1
∂xj v ∗0 = ∂z ṽ 0 (z)z=|m,x−ξ | mj sgn m, x − ξ  dm, ∂x2j v ∗0 = m2 dm.
4π 2 4π 2 m, x − ξ 2 j
|m|=1 |m|=1

Finally, v ∗0 = f 2 |m|=1 1 dm = f δ(x, ξ ).
4π m,x−ξ 2
In this case, fundamental solution v ∗0 (x, ξ ) is the same as the usual solution for Laplace’s equation in R2 (Vladimirov,
1984). From Eq. (23) we have ṽ 0 = f ln z|z=|s−τ | , and
  # $
f   f  x − ξ 
v ∗0 = lnm, x − ξ  dm = ln |x − ξ | m, dm
4π 2 4π 2 |x − ξ | 
|m|=1 |m|=1
  # $
f  x − ξ  f
= ln |x − ξ | dm + ln m, dm = ln |x − ξ |.
4π 2 |x − ξ |  2π
|m|=1 |m|=1
834 T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836

(c) Possibility η < 0 (superscript −). Let η = −k 2 and define v̂ − = α e−k|s−τ | to satisfy Eq. (A.2) with α = −f/(2k). Then,

∂s v̂ − = −α − k e−k|s−τ | sgn(s − τ ), and


∂s2 v̂ − = α − k 2 e−k|s−τ | − α − k e−k|s−τ | 2δ(s − τ ) = k 2 v̂ + + f δ(s − τ ).
The inverse Radon transform of v̂ − is

  1 
v ∗− (x, ξ ) = R∗ K(v̂ − ) = ṽ − (z)z=|m,x−ξ | dm,
4π 2
|m|=1

where ṽ − is given in Eq. (24). To show that function v ∗− (x, ξ ) satisfies Eq. (A.1), evaluate

1   
∂xj v ∗− = ∂z ṽ − (z)z=|m,x−ξ | mj sgn m, x − ξ  dm
4π 2
|m|=1
 
1    1
= g1− (z)z=|m,x−ξ | mj sgn m, x − ξ  dm + g2− mj dm,
4π 2 4π 2
|m|=1 |m|=1

where

   2
g1− = γ − k sh(kz) + 2k chi(kz) sh(kz) − shi(kz) ch(kz) , or g1− = γ − .
m, x − ξ 
Next,

1 
∂x2j v ∗− = ∂z2 ṽ − (z)z=|m,x−ξ | m2j dm
4π 2
|m|=1
 
1  −  −    1
= ∂z g (z)
1
2 
z=|m,x−ξ | mj + g1 z=|m,x−ξ | mj 2δ m, x − ξ  dm + 4π 2 ∂xj g2− m2j dm
4π 2
|m|=1 |m|=1
  m2j
k2 2γ −
= ṽ − (m, x − ξ )m2j dm + dm.
4π 2 4π 2 m, x − ξ 2
|m|=1 |m|=1

The final step is to show that


 − 
k2 − (m, x − ξ ) dm + 2γ 1
v ∗− = ṽ dm = k 2 v ∗− + f δ(x, ξ ).
4π 2 4π 2 m, x − ξ 2
|m|=1 |m|=1

References

Ahmad, S., Leyte, F., Rajapakse, R.K.N.D., 2001. BEM analysis of two-dimensional elastodynamic problems of anisotropic solids. J. Engrg.
Mech. ASCE 27 (2), 149–156.
Aizikovich, S.M., Alexandrov, V.M., Kalker, J.J., Krenev, L.I., Trubchik, I.S., 2002. Analytical solution of the spherical indentation problem
for a half-space with gradients with the depth elastic properties. Int. J. Solids Structures 39, 2745–2772.
Albuquerque, E.L., Sollero, P., Aliabadi, M.H., 2004. Dual BEM for anisotropic dynamic fracture mechanics. Int. J. Numer. Methods Engrg. 59,
1187–1205.
Albuquerque, E.L., Sollero, P., Fedelinski, P., 2003. Free vibration analysis of anisotropic material structures using the boundary element
method. Engrg. Anal. Boundary Elements 27, 977–985.
Ang, W.T., Clements, D.L., Vahdati, N., 2003. A dual-reciprocity boundary element method for a class of elliptic boundary value problems for
non-homogeneous anisotropic media. Engrg. Anal. Boundary Elements 27, 49–55.
Ang, W.T., Kusuma, J., Clements, D.S., 1997. A boundary element method for a second order elliptic partial differential equation with variable
coefficients. Engrg. Anal. Boundary Elements 18, 311–316.
Auriault, J.L., 2002. Upscaling heterogeneous media by asymptotic expansions. J. Engrg. Mech. ASCE 128 (8), 817–822.
Azis, M., Clements, D., 2001. A boundary element method for anisotropic inhomogeneous elasticity. Int. J. Solids Structures 38, 5747–5763.
Bai, H., Zhu, J., Shah, A.H., Popplewell, N., 2002. Three-dimensional steady-state Green’s function for a layered isotropic plate. J. Sound
Vib. 269, 251–271.
Bateman, H., Erdelyi, A., 1953. Higher Transcendental Functions. McGraw-Hill, New York.
T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836 835

Berezovski, A., Engelbrecht, J., Maugin, G.A., 2003. Numerical simulation of two-dimensional wave propagation in functionally graded mate-
rials. Eur. J. Mech. A Solids 22, 257–265.
Beskos, D.E. (Ed.), 1987. BEM in Mechanics. Elsevier Science, Amsterdam.
Beskos, D.E., 1997. Boundary element methods in dynamic analysis, Part II, 1986–1996. Appl. Mech. Rev. 50 (3), 149–197.
Chen, J., Liu, Z.X., Zou, Z.Z., 2002. Transient internal crack problem for a nonhomogeneous orthotropic strip (mode I). Int. J. Engrg. Sci. 40,
1761–1774.
Chen, L., Kassab, A.J., Nicholson, D.W., Chopra, M.B., 2001. Generalized boundary element method for solids exhibiting nonhomogeneities.
Engrg. Anal. Boundary Elements 25, 407–422.
Cheng, A.H.D., 1987. Heterogeneities in flows through porous media by boundary element method. Topics in Boundary Element Research,
Appl. Geomech. 4, 1291–1344.
Chuhan, Z., Yuntao, R., Pekau, O.A., Feng, J., 2004. Time-domain boundary element method for underground structures in orthotropic media.
J. Engrg. Mech. ASCE 130 (1), 105–116.
Clements, D.L., 1980. A boundary integral equation method for the numerical solution of a second-order elliptic partial differential equation
with variable coefficients. J. Australian Math. Soc. Ser. B 22, 218–228.
Crouch, S.L., 1976. Analysis of Stress and Displacements around Underground Excavations: An Application of the Displacement Discontinuity
Method, NSF RANN Program, University of Minnesota Geomechanics Report, Minneapolis.
De Hoop, A.T., 1995. Handbook of Radiation and Scattering of Waves, Acoustic Waves in Fluids, Elastic Waves in Solids, Electromagnetic
Waves. Academic Press, San Diego.
Dravinski, M., Wilson, M.S., 2001. Scattering of elastic waves by a general anisotropic basin. Part 1: 2D model. Earthquake Engrg. Structural
Dynamics 30, 675–689.
Dong, C.Y., Lo, S.H., Cheung, Y.K., 2004. Numerical solution for elastic half-plane inclusion problems by different integral equation ap-
proaches. Engrg. Anal. Boundary Elements 28, 123–130.
Fedoriuk, M., 1980. Ordinary Differential Equations. Nauka, Moscow.
Franciosi, P., Lormand, G., 2004. Using the Radon transform to solve inclusion problems in elasticity. Int. J. Solids Structures 41, 585–606.
Gelfand, I., Graef, I., Vilenkin, N., 1966. Generalized Functions, vol. V. Academic Press, New York.
Georgiadis, H.G., Lycotrafitis, G., 2001. A method based on the Radon transform for three-dimensional elastodynamic problems of moving
loads. J. Elasticity 65, 87–129.
Gipson, G.S., Ortiz, J.C., Shaw, R.P., 1995. Two-dimensional linearly layered potential flow by boundary elements. In: Brebbia, C.A., Power, H.,
Kim, S., Osswald, T.A. (Eds.), Boundary Elements XVII. Springer-Verlag, Berlin, p. 9.
Guo, L.C., Wu, L.Z., Zeng, T., Ma, L., 2004. Mode I crack problem for a functionally graded orthotropic strip. Eur. J. Mech. A Solids 23,
219–234.
John, F., 1955. Plane Waves and Spherical Means Applied to Partial Differential Equations. Wiley Interscience, New York.
Hook, J.F., 1962. Green’s function for axially symmetric elastic waves in unbounded inhomogeneous media having constant velocity gradients.
J. Appl. Mech. ASME E-29, 293–298.
Kassab, A.J., Divo, E.A., 1996. A generalized boundary integral equation for isotropic heat condition equation with spatially varying thermal
conductivity. Engrg. Anal. Boundary Elements 18, 273–286.
Keilis-Borok, V.I., Lewshin, A.L., Yanovskaya, T.B., Lander, A.V., Buckhin, B.G., Barmin, M.P., Ratnikova, L.I., 1989. Seismic Surface Waves
in a Laterally Inhomogeneous Earth. Kluwer Academic, Dordrecht.
Lee, J., Lee, H., Mal, A., 2004. A mixed volume and boundary integral equation technique for elastic wave field calculations in heterogeneous
materials. Wave Motion 39, 1–19.
Ludwig, D., 1966. The Radon transform in Euclidean space. Commun. Pure Appl. Math. 19, 49–81.
Manolis, G.D., Dineva, P.S., Rangelov, T.V., 2004. Wave scattering by cracks in inhomogeneous continua using BIEM. Int. J. Solids Struc-
tures 41 (14), 3905–3927.
Manolis, G.D., Rangelov, T.V., Shaw, R.P., 2002. Conformal mapping methods for variable parameter elastodynamics. Wave Motion 36 (2),
185–202.
Manolis, G.D., Shaw, R.P., 1996. Green’s function for a vector wave equation in a mildly heterogeneous continuum. Wave Motion 24, 59–83.
Manolis, G.D., Shaw, R.P., 2000. Fundamental solutions for variable density two-dimensional elastodynamic problems. Engrg. Anal. Boundary
Elements 24, 739–750.
Manolis, G.D., Shaw, R.P., Pavlou, S., 1999. Elastic waves in non-homogeneous media under 2D conditions: I. Fundamental solutions. Soil
Dynamics Earthquake Engrg. 18, 19–30.
Mikhailov, S.E., 2002. Localized boundary-domain integral formulations for problems with variable coefficients. Engrg. Anal. Boundary Ele-
ments 26, 681–690.
Moczo, P., Bystricky, E., Kristek, J., Carcione, J.M., Bouchon, M., 1997. Hybrid modeling of P-SV seismic motion at inhomogeneous vis-
coelastic topographic structures. Bull. Seismolog. Soc. Amer. 87 (5), 1305–1323.
Nayfeh, A.H., 1995. Wave Propagation in Layered Anisotropic Media. North-Holland, Amsterdam.
Niu, Y., Dravinski, M., 2003. Direct 3D BEM for scattering of elastic waves in a homogeneous anisotropic half-space. Wave Motion 38,
165–175.
Ozturk, M., Erdogan, F., 1997. Mode I crack problem in inhomogeneous orthotropic medium. Int. J. Engrg. Sci. 35, 406–413.
Park, S.H., Ang, W.T., 2000. A complex variable boundary element method for an elliptic partial differential equation with variable coefficients.
Commun. Numer. Methods Engrg. 16, 697–703.
Pike, R., Sabatier, P. (Eds.), 2001. Scattering and Inverse Scattering in Pure and Applied Sciences. Academic Press, San Diego.
Rangelov, T.V., 1992. Linear Algebra. Novisima-Verda Publishers, Bourgas, Bulgaria.
836 T.V. Rangelov et al. / European Journal of Mechanics A/Solids 24 (2005) 820–836

Rangelov, T.V., 2003. Scattering from cracks in an elasto-anisotropic plane. J. Theoret. Appl. Mech. 33 (2), 55–72.
Rangogni, R., 1987. A solution of Darcy’s flow with variable permeability by means of BEM and perturbation techniques. In: Brebbia, C.A.,
Wendland, W.L., Kuhn, G. (Eds.), Boundary Elements IX, vol. 3. Springer-Verlag, Berlin, pp. 359–368.
Reddy, J.N., Cheng, Z.Q., 2003. Frequency of functionally graded plates with three-dimensional asymptotic approach. J. Engrg. Mech.
ASCE 129 (8), 896–900.
Rubio-Gonzalez, C., Manzon, J.J., 1999. Response of finite cracks in orthotropic materials due to concentrated impact shear load. J. Appl.
Mech. ASME 66, 485–491.
Santare, M.H., Thamburaj, P., Gazonas, G.A., 2003. The use of graded finite elements in the study of elastic wave propagation in continuously
nonhomogeneous materials. Int. J. Solid Structures 40, 5621–5634.
Sharma, M.D., 2002. Group velocity along general direction in a general anisotropic medium. Int. J. Solids Structures 39, 3277–3288.
Su, R., Sun, H., 2003. Numerical solution of two-dimensional anisotropic crack problems. Int. J. Solids Structures 40, 4615–4635.
Tanaka, M., Matsumoto, T., Suda, Y., 2001. A dual-reciprocity boundary element method applied to the steady-state heat conduction problem
of functionally gradient materials. In: Kassab, A.J., Brebbia, C.A. (Eds.), Boundary Element Techniques XIV. WIT Press, Southampton.
Vladimirov, V., 1984. Equations of Mathematical Physics. Mir, Moscow.
Wang, B.L., Mai, Y.W., Noda, N., 2002. Fracture mechanics analysis model for FGM with arbitrary distributed properties. Int. J. Fracture 116,
161–177.
Wang, B.L., Mai, Y.W., Sun, Y.G., 2003. Anti-plane fracture of a functionally graded material strip. Eur. J. Mech. A Solids 22, 357–368.
Wang, C.Y., Achenbach, J.D., 1994. Elastodynamic fundamental solutions for anisotropic solids. Geophys. J. Int. 118, 384–392.
Watanabe, K., Takeuchi, T., 2002. Green’s function for two-dimensional waves in a radially inhomogeneous elastic solid. In: Proceedings of
the IUTAM Symposium in Japan.
Zayed, A., 1996. Handbook of Function and Generalized Function Transforms. CRC Press, Boca Raton, FL.

You might also like