Y. J.

Liu1
Mechanical Engineering, University of Cincinnati,
598 Rhodes Hall, 2600 Clifton Avenue,
Cincinnati, OH 45221-0072,
e-mail: [email protected]

S. Mukherjee
Sibley School of Mechanical and
Aerospace Engineering, Cornell University,
Ithaca, NY 14853
e-mail: [email protected]

N. Nishimura
Department of Applied Analysis and
Complex Dynamical Systems,
Graduate School of Informatics,
Kyoto University,
Kyoto, 606-8501 Japan,
e-mail: [email protected] Recent Advances and Emerging
M. Schanz
Institute of Applied Mechanics,
Graz University of Technology,
Applications of the Boundary
Technikerstr. 4, A-8010 Graz, Austria,
e-mail: [email protected]
Element Method
W. Ye Sponsored by the U.S. National Science Foundation, a workshop on the boundary ele-
Department of Mechanical Engineering, ment method (BEM) was held on the campus of the University of Akron during September
Hong Kong University of Science and 1–3, 2010 (NSF, 2010, “Workshop on the Emerging Applications and Future Directions
Technology, Clearwater Bay, Kowloon, of the Boundary Element Method,” University of Akron, Ohio, September 1–3). This pa-
Hong Kong, China, per was prepared after this workshop by the organizers and participants based on the
e-mail: [email protected] presentations and discussions at the workshop. The paper aims to review the major
research achievements in the last decade, the current status, and the future directions of
A. Sutradhar the BEM in the next decade. The review starts with a brief introduction to the BEM.
Department of Surgery and Then, new developments in Green’s functions, symmetric Galerkin formulations, bound-
Department of Mechanical Engineering, ary meshfree methods, and variationally based BEM formulations are reviewed. Next,
The Ohio State University, fast solution methods for efficiently solving the BEM systems of equations, namely, the
915 Olentangy River Road, Suite 2100, fast multipole method, the pre-corrected fast Fourier transformation method, and the
Columbus, OH 43212 adaptive cross approximation method are presented. Emerging applications of the BEM
e-mail: [email protected] in solving microelectromechanical systems, composites, functionally graded materials,
fracture mechanics, acoustic, elastic and electromagnetic waves, time-domain problems,
E. Pan and coupled methods are reviewed. Finally, future directions of the BEM as envisioned
Department of Civil Engineering, by the authors for the next five to ten years are discussed. This paper is intended for stu-
The University of Akron, ASEC Room 534, dents, researchers, and engineers who are new in BEM research and wish to have an
Akron, OH 44325-3905 overview of the field. Technical details of the BEM and related approaches discussed in
e-mail: [email protected] the review can be found in the Reference section with more than 400 papers cited in this
review. [DOI: 10.1115/1.4005491]
N. A. Dumont
Pontifical Catholic University of Rio de Janeiro,
22451-900 Brazil
e-mail: [email protected]

A. Frangi
Department of Structural Engineering,
Politecnico of Milano, P.za L. da Vinci 32,
20133 Milano, Italy,
e-mail: [email protected]

A. Saez
Department of Continuum Mechanics and
Structural Analysis, University of Seville,
Camino de los Descubrimientos s/n,
Seville, E-41092, Spain,
e-mail: [email protected]

1
Corresponding author.
Manuscript received May 25, 2011; final manuscript received December 12,
2011; published online March 30, 2012. Transmitted by Editor: J. N. Reddy.

Applied Mechanics Reviews Copyright V

C 2011 by ASME MAY 2011, Vol. 64 / 031001-1

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Table 1 A Comparison of the FEM and BEM

Finite element method Boundary element method

Unknown DOFs everywhere Unknown DOFs only on boundary

in domain for linear problems
Banded symmetric matrices Smaller full nonsymmetric matrices
Very efficient Less efficient in standard form (fast
with new forms)
Exterior problems somewhat Exterior problems in infinite domains
difficult easy
Nonhomogeneous problems Needs Green’s functions for maxi-
fine mum efficiency
Moving boundary problems Efficient for moving boundary
can be difficult problems Fig. 1 A 3D domain V with boundary S
Needs compatible elements Incompatible=nonconforming ele-
ments fine
Inter-element discontinuity of Continuous internal derivatives
derivatives 1.2 A Review of the BEM Formulation. We use the 3D
potential theory problem as an example to show the key steps and
main results in the BEM formulation. For a domain V with bound-
ary S (Fig. 1), we have the following boundary-value problem:
1 Introduction
The boundary element method (BEM) is a numerical method for r2 /ðxÞ ¼ 0; x2 V (1)
solving boundary-value or initial-value problems formulated in
boundary integral equations (BIEs). In some literature, it is also
where / is the potential field. The boundary conditions (BCs) are
called the boundary integral equation method because of this rela-
tionship. The most important advantage of the BEM as compared /ðxÞ ¼ /ðxÞ; x 2 S/ ðDirichlet BCÞ (2)
with other domain-based numerical methods is the reduction of the
dimensions of the problems to be solved, from 3D to 2D, or from 2D @/
qðxÞ  ðxÞ ¼ qðxÞ; x 2 Sq ðNeumann BCÞ (3)
to 1D, leading to much easier mesh generation as compared with @n
domain-based methods. The second advantage is the accuracy the
BEM offers, due to the nature of integrals used in the formulations. in which the overbar indicates the prescribed value, S/ [ Sq ¼ S,
The third advantage of the BEM is its capability of accurately model- and n is the outward normal of the boundary S (Fig. 1).
ing infinite domain problems without introducing additional, often The fundamental solution (also known as full-space Green’s
cumbersome, conditions at infinity. Table 1 is a comparison of the function) Gðx; yÞ for 3D potential problems is given by
BEM with the finite element method (FEM), which may change in 1
the future based on new developments in either one or both methods. Gðx; yÞ ¼ (4)
4pr
1.1 A Brief History. The direct BIE formulations (in which where r is the distance between source point x and field point y.
the unknown functions have clear physical meanings) and their The normal derivative of G is
modern numerical solutions with boundary elements originated
more than 40 years ago during the 1960s. The 2D potential prob- @Gðx; yÞ 1
Fðx; yÞ  ¼ r;k nk ðyÞ (5)
lem was first formulated in terms of a direct BIE and solved @nðyÞ 4pr 2
numerically by Jaswon, et al. [1–3]. This work was later extended
to the 2D elastostatic case by Rizzo in the early 1960s as his dis- with r;k ¼ @r=@yk ¼ ðyk  xk Þ=r. The fundamental solution
sertation work at the University of Illinois, which was later pub- Gðx; yÞ represents the response (potential) at y due to a unit source
lished as a journal article in 1967 [4]. Following these early at x (See Sec. 2.1 for the definition of a Green’s function).
works, extensive research efforts were carried out in the BIE for- If we apply the second Green’s identity with / and Gðx; yÞ in
mulations of many problems in applied mechanics and in the nu- domain V with boundary S, we obtain the following representation
merical solutions during the 1960s and 1970s [5–20]. The name integral:
boundary element method appeared in the middle of the 1970s in ð
an attempt to make an analogy with the FEM [21–23]. /ðxÞ ¼ ½Gðx; yÞqðyÞ  Fðx; yÞ/ðyÞdSðyÞ; x 2 V (6)
Some personal accounts of the historical developments of the S
BIE and BEM in various places of the U.S., Europe, and China Equation (6) is an expression of the solution / inside the domain
were given by Rizzo [24], Cruse [25], Watson [26], Shippy [27], V for Eq. (1). Once the boundary values of both / and q are
Mukherjee [28], Telles [29], and Yao and Du [30] in a special issue known on the entire boundary S, Eq. (6) can be applied to calcu-
of the Electronic Journal of Boundary Elements in 2003 [31]. A late / everywhere in V, if needed. To solve the unknown bound-
comprehensive review of the heritage and early history of the BIE ary values of / and q on S, we let x tend to the boundary S to
and BEM was given by Cheng and Cheng in 2005 [32]. This review obtain the following BIE:
provides vivid descriptions of the long history and major contribu- ð
tions to the mathematical foundations of the integral methods by cðxÞ/ðxÞ ¼ ½Gðx; yÞqðyÞ  Fðx; yÞ/ðyÞdSðyÞ; x 2 S (7)
the pioneers from the 18th to the first half of the 20th centuries and S
the early days of the BEM until the late of 1970s. Another compre-
hensive review of the BEM research up to the early of 1980s was in which cðxÞ is a coefficient and cðxÞ ¼ 1=2 if S is smooth around
given by Tanaka in 1983 [33]. A special review was given by x. Equation (7) is called a conventional or singular BIE for solving
Tanaka, Sladek and Sladek in 1994 [34] on the research efforts in potential problems.
the late 1980s and early 1990s regarding the various approaches to If we take the normal derivative of Eq. (6) and let the source point
dealing with the singular and hypersingular integrals in the BIEs. x go to boundary S, we obtain the so-called hypersingular BIE:
Ten representative books on the BEM, published during the last ð
three decades, can be found in Refs. [35–44] which can be con- cðxÞqðxÞ ¼ ½Kðx; yÞqðyÞ  Hðx; yÞ/ðyÞdSðyÞ; x 2 S (8)
sulted to further study the details of the BEM. S

031001-2 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 where the two new kernels for 3D are and (3) The solution of the system in Eq. (13) using direct solvers
such as Gauss elimination requires OðN 3 Þ operations. Thus, for
@Gðx; yÞ 1 several decades, the conventional BEM approach of solving Eq. (13)
Kðx; yÞ  ¼ r;k nk ðxÞ (9)
@nðxÞ 4pr 2 directly has been limited to small-scale models with, at most, a few
thousands of elements, for example, on a desktop PC running a
@Fðx; yÞ 1 32–bit operating system. In Secs. 2 and 3, new BIE formulations and
Hðx; yÞ  ¼ ½nk ðxÞnk ðyÞ  3r;k nk ðxÞr;l nl ðyÞ
@nðxÞ 4pr 3 new fast solution methods aimed to address these shortcomings of
(10) the conventional BEM approach are discussed.
The rest of this review paper is organized as follows: In Sec. 2,
Hypersingular BIEs are needed in the BEM when it is applied to new developments in the last decade in the theoretical formulations
model cracks, thin shapes and exterior acoustic wave problems, as related to the BIE and BEM are discussed. These developments are:
discussed in other sections. The Fðx; yÞ and Kðx; yÞ kernels are new development in the Green’s functions which are the key ingre-
strongly singular. Therefore, the second integral in BIE (Eq. (7)) and dients in the BEM, the symmetric Galerkin BEM formulation, the
the first integral in BIE (Eq. (8)) are Cauchy-principal value (CPV) BIE related boundary meshfree methods, and the variationally based
integrals in general. The Hðx; yÞ kernel is hypersingular, and the sec- BEM formulations. In Sec. 3, three fast solution methods that
ond integral in BIE (Eq. (8)) is a Hadamard-finite part (HFP) inte- emerged in the last decade for solving the BEM systems of equations
gral. There are different approaches [34,45–47] to evaluate or avoid are reviewed. These fast methods are the fast multipole method, the
the singular and hypersingular integrals in the BIE formulations pre-corrected fast Fourier transformation method, and the adaptive
which are no longer the barriers in the implementation of the BEM. cross approximation method. These fast solution methods are the key
The BIE as shown in Eq. (8) is called a direct BIE formulation factors in the last decade that have contributed to revitalizing the
in which the density functions / and q have clear physical mean- BEM. In Sec. 4, modern applications of the BEM in solving large-
ings. The so-called indirect BIE formulations are also available in scale, critical research and industrial problems using the advanced
the literature, in which only one integral is applied and the density BEM are presented. The areas of applications include: modeling
function has no clear physical meanings. In some cases, the indi- microelectromechanical systems, composite materials, functionally
rect BIE formulation is advantageous because of the reduction of graded materials, fracture mechanics, acoustic, elastic and electro-
the integrals with which to work. More details about the indirect magnetic waves, time-domain problems, and coupling of the BEM
BIE formulations can be found in many textbooks on the BEM. with other methods. In Sec. 5, we discuss the future directions of the
Complex variable BIE formulations in the BEM are also popu- BEM research, development and education for the next five to ten
lar for dealing with various 2D problems (see, e.g., Refs. [48–51]; years, based on our own opinions. More than 400 references that
[44] (Chapter 4 on 2D elasticity)), because of the efficiencies in highlight the history and advances of the BEM research in the last
using the complex functions. However, this approach has limited decade or so are listed at the end of this paper for readers to further
applicability as to solving 3D problems. study the related topics.
As an example to see how to discretize the BIEs, assume that
we use N constant boundary elements (surface patches for 3D, 2 New Formulations
line segments for 2D) on boundary S. We can obtain the following
discretized equation of BIE (7): 2.1 Green’s Functions for Anisotropic and Layered
2 38 9 2 38 9 Materials. Green’s function (GF) is named after George Green,
f11 f12    f1N > > /1 >> g11 g12    g1N > > q1 > > who wrote the famous essay on the application of mathematical
6 f21 f22    f2N 7< /2 = 6 g21 g22    g2N 7>
> > < > =
6 7 6 7 q2 analysis to the theories of electricity and magnetism [52]. In his
6 .. .. . . .. 7 .. ¼6 . . . . 7 . essay, Green coined the term “potential” to denote the results
4 . . . . 5> > . >
> 4 .. .. .. .. 5>
> .. >
>
>
: >
; >
: > ; obtained by adding the masses of all the particles of a system,
fN1 fN2    fNN /N gN1 gN2    gNN qN each divided by its distance from a given point. The general prop-
(11) erties of the potential function were subsequently developed and
applied to electricity and magnetism. The formula connecting the
with surface and volume integrals, now known as the Green’s theorem,
ð ð was also introduced in that work.
1 A Green’s function, also called singular function or fundamen-
gij ¼ Gi dS; fij ¼ dij þ Fi dS; for i; j ¼ 1; 2; :::; N
DSj 2 DSj
tal solution (a full-space GF), represents the solution in a given
system due to a point source. For example, the GF denoted by
(12) Gðx; yÞ for potential theory problems governed by Eq. (1) and in
an infinite domain satisfies
in which Gi and Fi are the kernels with the source point x placed at
node i, DSj represents element j, and dij is the Kronecker d symbol. r2 Gðx; yÞ þ dðx; yÞ ¼ 0; x; y 2 R2 =R3 (14)
In the conventional BEM approach, we form a standard linear system
of equations by applying the boundary condition at each node and
switching the columns in the two matrices in Eq. (11) to obtain: where r2 ¼ @ 2 ð Þ=@yi @yi , and R2 and R3 indicate the full 2D and
3D spaces, respectively. The Dirac d function dðx; yÞ in Eq. (14)
2 38 9 8 9 represents a unit source (e.g., heat source) at the source point x,
a11 a12  a1N > k1 > > b1 >
6 a21 7>
< > = > < b2 >
= and Gðx; yÞ represents the response (e.g., temperature) at the field
6 a22  a2N 7 k2 point y due to the source. The solution of Eq. (14) in 3D is given
6 .. .. .. .. 7 .. ¼ .. ; or A~
k¼b
4 . 5>
> > > > in Eq. (4). GFs play a key role in the BEM. Without the GFs, the
. . . : . >; > : . >;
aN1 aN2  aNN kN bN boundary nature of the BEM will be hard to obtain. On the other
hand, due to the use of the GF, the BEM is often regarded as a
(13) semi-analytical method and thus more accurate, because the BEM
solution is built upon the analytical result provided by the GF.
where A is the coefficient matrix, k~ is the unknown vector, and b More information about the GFs can be found in Refs. [53,54].
is the known vector. The GF concept is now extensively used in the solution of par-
The disadvantages of the conventional BEM formulations are: (1) tial differential equations. The recently re-emerged method of fun-
The construction of matrix A requires OðN 2 Þ operations using damental solutions also makes direct use of the GFs [55,56]. In
Eq. (12), (2) The size of the memory required for storing A is also the Eshelby-based micromechanics theory, the GFs further plays
OðN 2 Þ because A is in general a nonsymmetrical and dense matrix, an important role [57].

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-3

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 The GF also played an important role in the development of magnetoelectroelastic media under anti-plane deformation. Chen
various advanced material systems in the last decade. Anisotropy et al. [83] derived the explicit expressions for the dynamic poten-
and multiphase coupling are important features in advanced smart tials of an inclusion embedded in a “quasi-plane” magnetoelec-
material systems. The static GFs in 3D transversely isotropic pie- troelastic medium of transverse isotropy and for the dynamic GFs
zoelectric infinite, half, and bimaterial spaces were derived by of such a medium both in the time domain and in the frequency
Ding’s group at Zhejiang University [58,59] and Dunn’s group at domain. Rojas-Dı́az et al. [84] later derived the 2D and 3D time-
University of Colorado at Boulder [60] where the potential func- harmonic GFs by means of Radon transform.
tion method was employed. Furthermore, under the assumption of For the development of GFs in other fields, readers may consult
transverse isotropy, Ding’s group has further derived the corre- a special issue of the journal Engineering Analysis with Boundary
sponding magnetoelectroelastic GFs in the infinite space, half Elements published in 2005 [78] where some recent research on
space, and bimaterial space [61]. GFs and related BEM issues are discussed.
The corresponding 2D GFs in general anisotropy (including
also the piezoelectric and magnetoelectroelastic coupled systems) 2.2 Symmetric Galerkin BEM. There are two basic proce-
can be best represented by the Stroh formalism based on the com- dures that are generally used to reduce the continuous boundary in-
plex variable method for infinite, half, and bimaterial planes tegral Eqs. (7) and (8) to finite systems. The simpler procedure is
[62–64]. In general, GFs of the degenerated case where repeated collocation, wherein the BIEs are explicitly enforced at a finite set
eigenvalues appear cannot be directly reduced from these expres- of points. In its simplest form, these collocation points are chosen
sions. However, those of slightly perturbed cases (i.e., the quasi- to be the nodes used to discretize the boundary. If the boundary
isotropic) can give very accurate solutions [65]. On the other potential and flux are interpolated from their values at these N
hand, Yin [66] employed the Lekhnitskii’s formalism and derived points, then the pointwise enforcement of Eqs. (7) and (8) provides
the Green’s functions for infinite, half, and bimaterial planes the N equations needed to solve for the unknown values. Colloca-
where both general anisotropy (with distinguished eigenvalues) tion necessarily leads to non-symmetric matrices system.
and quasi-isotropic case (with repeated eigenvalues in different In contrast to collocation, the Galerkin approach does not require
ranks) were cast into unified expressions. that the integral equations be satisfied at any point. Instead, the equa-
Three-dimensional GFs in an anisotropic and elastic full space tions are enforced in a weighted average sense as follows:
were derived using various mathematical methods, including
ð
eigenvalues/vectors, Fourier transform, Radon transform, among
others. A comprehensive review on this topic can be found in Lav- PðxÞ  cðxÞ/ðxÞ ¼ ½Gðx; yÞqðyÞ  Fðx; yÞ/ðyÞdSðyÞ; 8x 2 S
S
agnino [67]. Very recently, Buroni et al. [68] derived the set of
solutions which contain any kind of mathematical degeneracy of (15)
ð
the eigenvalues. The corresponding half-space and bimaterial GFs
QðxÞ  cðxÞqðxÞ ¼ ½Kðx; yÞqðyÞ  Hðx; yÞ/ðyÞdSðyÞ; 8x 2 S
were obtained in Ref. [69]. Similar GFs were also derived for the S
half-space case with general surface conditions [70] and for the (16)
bimaterial case with general (or imperfect) interface conditions ð
[71]. Simple line-integral expressions for the interfacial GFs and wk PðxÞ dSðxÞ ¼ 0
trimaterial GFs can also be obtained.
Three-dimensional GFs in the corresponding anisotropic piezo- ðS (17)
electric and even magnetoelectroelastic full, half, and bimaterial wk QðxÞ dSðxÞ ¼ 0
S
space can be derived by following a similar procedure. Actually, to
derive these GFs, one only needs to extend the problem dimension
where wk are the chosen weight functions. The needed N equa-
from three to four for the piezoelectric case and to five for the mag-
tions can be generated by an appropriate choice of N weights. The
netoelectroelastic case. For instance, for the 3D magnetoelectroelas-
strict definition of Galerkin is that the weight function wk is com-
tic case, the GFs can be found in Refs. [72,73]. Besides the point
posed of the shape functions that are non-zero at the node x, the
“force” GFs, some point “dislocation” GFs were also derived, which
shape functions being the local basis functions used to interpolate
have been shown to be very efficient in dealing with fracture prob-
the boundary functions. For a linear interpolation, the Galerkin
lems in piezoelectric and magnetoelectroelastic solids [74,75].
weight functions are illustrated in Fig. 2.
Layered structures are very common in advanced composite
In mathematical terminology, collocation is a strong solution,
structures, and therefore, derivation of the GFs in these systems is
the equations being satisfied at the specified points, whereas
also important. Typical methods to find the GFs is to first apply
the double Fourier transform or Hankel transform in the horizontal
layer plane and solve the problem in the transformed domain via
the propagator matrix method. The physical domain solution is
obtained by applying the inverse transform, which usually
requires numerical integration. Besides the GFs for elastic iso-
tropic, transversely isotropic, thermoelastic, and poroelastic lay-
ered spaces, GFs for the piezoelectric layered half space and even
piezoelectric functionally graded half space were also derived
[76,77]. Similar approaches have also been developed to derive
the GFs in general anisotropic and layered elastic spaces [78].
Significant efforts have been conducted by different authors to
derive dynamic GFs in anisotropic solids and further extending
these ideas to the piezoelectric and magnetoelectroelastic cases.
Denda et al. [79] derived the 2D time-harmonic GFs by means of
the Radon transform, and implemented them for the eigenvalue
analysis of piezoelectric solids. Following the same approach,
Wang and Zhang [80] presented their time-domain, Laplace’s do-
main, and 3D counterparts of the GFs. Wu and Chen [81] later
derived explicit dynamic GFs for the 2D case. With respect to the
magnetoelectroelastic case, Ren and Liu [82] presented a time-
harmonic dynamic fundamental solution for transversely isotropic Fig. 2 Illustration of the Galerkin weight functions for 2D BEM

031001-4 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Galerkin is a weak solution. The requirement that the equations 2.2.2 Brief History. Almost all of the early BEM computa-
hold in an integrated sense has a geometric interpretation: The ap- tions employed a collocation approximation. In 1977, Bui [85]
proximate Galerkin solution is obtained by projecting the exact produced a Galerkin formulation for fracture analysis, and in
solution onto the subspace consisting of all functions which are a 1979, Sirtori [86] proposed the symmetric Galerkin formulation
linear combination of the shape functions. The Galerkin solution for linear elasticity. It was recognized almost after a decade that
is therefore the linear combination that is the closest to the exact collocation BEM failed to retain many of the nice properties of fi-
solution. In general, the Galerkin method is more accurate than nite elements: sign-definite symmetric matrix operators, a varia-
the collocation method, and also provides a more elegant treat- tional formulation, and proofs of convergence and stability. In
ment of boundary corners. However, the primary advantage of the addition, a collocation implementation for fracture analysis had to
Galerkin method is that the treatment of hypersingular integrals is resort to higher order C1 elements, e.g., Overhauser or Hermite
actually much simpler than with collocation. The Galerkin form elements to adequately deal with the hypersingular kernel, and
of the hypersingular integral exists if the interpolation of the these elements proved difficult to handle, especially in three
boundary potential is continuous C0. For collocation, a smoother dimensions. These negative aspects of collocation fueled the inter-
interpolation (differentiable C1) is required for the existence of est in the symmetric Galerkin boundary element formulation, and
the integral. Standard elements, e.g., linear or quadratic, are con- subsequent development was primarily carried out in Europe.
tinuous across element boundaries, and it is quite a bit more com- Hartmann et al. [87] in 1985 extended the symmetric Galerkin
plicated to implement a differentiable interpolation. formulation to beams and Kirchhoff plates, while Maier and
Polizzotto [88] in 1987 developed SGBEM for elastoplastic sol-
2.2.1 Symmetric Galerkin Formulation. With Galerkin, x and ids. Maier and Polizotto [88] extensively studied elastoplasticity,
y are treated equally, and thus it is possible to produce a symmet- gradient elasticity, cohesive crack simulation, elastodynamics and
ric coefficient matrix, the symmetric Galerkin boundary element plasticity problems using the SGBEM. Different computational
method (SGBEM). This algorithm is based upon the symmetry implementations appeared, such as those by Holzer [89], Bonnet
properties of the Green’s function, and Bui [90], and Kane and Balakrishna [91].
Treatment of hypersingular integrals in 3D is a hurdle [34].
Gðx; yÞ ¼ Gðy; xÞ
Two main approaches have been based upon Hadamard finite
@Gðx; yÞ @Gðx; yÞ @Gðy; xÞ parts, including Andra [92], Carini and Salvadori [93], Carini
¼  ¼ et al. [94], Haas and Kuhn [95], Frangi and Novati [96], Frangi
@nðyÞ @nðxÞ @nðxÞ (18)
and Guiggiani [97], and Salvadori [98,99], among others; and
2 2
@ Gðx; yÞ @ Gðy; xÞ direct limit methods by Gray et al. [100,101] for homogeneous
¼
@nðxÞ@nðyÞ @nðxÞ@nðyÞ materials, and Sutradhar et al. [102] for graded materials.
2.2.3 Recent Applications. One of the important application
It follows that if the potential is specified everywhere on the areas of symmetric Galerkin is fracture. An implementation for
boundary, then the Galerkin form of the potential equation will fracture analysis in plane orthotropic elasticity was reported by
yield a symmetric matrix; a similar statement holds for flux Gray and Paulino [103]. A weak form integral equation for 3D
boundary data and the flux equation. For a general mixed bound- fracture analysis was proposed by Li, Mear, and Xiao [104].
ary value problem, symmetry results if each equation is applied Frangi [105,106] developed a similar crack analysis based on inte-
on its appropriate surface. That is, the potential equation is applied gration by parts, and applied it to crack propagation and fatigue
on the Dirichlet portion of the boundary and the flux equation is crack growth. Salvadori [107] presented a crack formulation for
applied on the Neumann portion of the boundary. cohesive interface problems, and Sutradhar and Paulino [108] pre-
After discretization, the set of Eq. (17) can be written in block- sented a symmetric Galerkin formulation to evaluate T-stress and
matrix form as ½f / ¼ ½hq, and in block-matrix these equations stress intensity factors (SIFs) using the interaction integral
become method. Xu, Lie and Cen [109] presented a 2D crack propagation
      analysis using quasi-higher order elements. Phan et al. [110–114]
f 11 f 12 /bv h11 h12 qbv presented SIF analysis with frictional contact sliding at disconti-
¼ (19)
f 21 f 22 /un h21 h22 qun nuity and junctions, SIF calculations for crack-inclusion interac-
tion problems, and a SGBEM based quasi-static crack-growth
Symmetry of the coefficient matrix for a general mixed boundary prediction tool to investigate crack-particle interactions. Galerkin
value problem is achieved by the following simple arrangement. methods have found a broad range of engineering applications.
The BIE is employed on the Dirichlet surface, and the HBIE equa- For instance, an SGBEM formulation for 2D steady and incom-
tion is used on the Neumann surface. The first row represents the pressible flow was developed by Capuani et al. [115]. A Stokes
BIE written on the Dirichlet surface, and the second row repre- problem with general boundary condition with slip condition was
sents the HBIE written on the Neumann surface. Similarly, the reported by Reidinger and Steinbach [116]. Other works using
first and the second columns arise from integrating over Dirichlet SGBEM include; for example, steady state harmonic solution of
and Neumann surfaces, respectively. The subscripts in the matrix the Navier equation by Perez-Gavilan and Aliabadi [117], inter-
therefore denote known boundary values bv and unknown un face and multi-zone problems by Gray and Paulino [118], lower
quantities. Rearranging into the form Ak ¼ b, one obtains bound shakedown analysis by Zhang et al. [119], dynamic soil-
     structure interaction by Lehman and Antes [120], problems of fi-
h11 f 12 qun f 11 /bv þh12 qbv nite elasticity with hyperelastic and incompressible materials by
¼ (20)
h21 f 22 /un f 21 /bv  h22 qbv Polizotto [121], dynamic frequency domain viscoelastic problems
subjected to steady-state time-harmonic loads by Perez-Gavilan
The symmetry of the coefficient matrix, h11 ¼ hT11 , f 22 ¼ f T22 and and Aliabadi [122,123], analysis of Kirchhoff elastic plates by
f 12 ¼ hT21 , now follows from the properties of the kernel functions. Frangi and Bonnet [124].
The advantages of Galerkin, of course, come at a price: although,
as shown in Fig. 2, the weight functions have local support, the 2.3 Meshfree Methods Based on BIEs. Boundary-based
additional boundary integration with respect to x is nevertheless meshfree methods are first introduced in this section. This is fol-
computationally expensive. These costs can be reduced somewhat lowed by some details of the boundary node method (BNM) and
by exploiting symmetry or by fast solution methods, but Galerkin the extended boundary node method (EBNM).
BEM will nevertheless require more computation time than The primary idea in meshfree methods is to divorce the tradi-
collocation. tional coupling between spatial discretization (meshing) and

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-5

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 interpolation of the unknown solution, as is commonly practiced tages of the variable basis approach [138], together with allowing
in the FEM or the BEM. Instead, a “diffuse” approximation is possible discontinuities in the normal derivative of the potential
used to represent the unknown functions; cells, with a very flexi- function, across edges and corners, in an approach called the
ble structure (e.g., any cell can be arbitrarily subdivided without extended boundary node method (EBNM). A detailed formulation
affecting its neighbors), are used for integration. A domain-based for the EBNM for 2D potential theory, together with numerical
meshfree method was first proposed by Nayroles et al. [125]; this results for selected problems, appear in [148]; while the EBNM
idea was improved and expanded upon by Belytschko et al. [126] for 3D potential theory is the subject of [149]. A brief description
who proposed the element-free Galerkin method (EFG). The EFG of the EBNM for 3D potential theory is given below. Details are
couples the moving least squares (MLS) approximation scheme available in Ref. [149].
with the Galerkin weak form to obtain a domain-based meshfree It is assumed that the bounding surface S of a solid body occu-
method. The BNM, first proposed by Mukherjee and Mukherjee pying the region V is the union of piecewise smooth segments
[127], on the other hand, is a combination of the MLS approxima- called panels. The BNM employs a diffuse approximation in
tion scheme and the standard BIE, thus producing a boundary- which the value of a variable at a boundary point is defined in
based meshfree method. Thus, the BNM retains the meshless terms of its values at neighboring boundary points within its do-
attribute of the EFG and the dimensionality advantage of the main of dependence (DOD). Correspondingly, a boundary node
BEM. As a consequence, the BNM only requires the specification affects points within its range of influence (ROI). These regions
of points on the 2D bounding surface of a 3D body (including are shown in Fig. 3(a).
crack faces in fracture mechanics problems), together with Let /ðxÞ be the sought after harmonic function in 3D and qðxÞ
unstructured surface cells, thereby practically eliminating the be its outward normal derivative. The first step is to write:
meshing problem. In contrast, the FEM needs volume meshing,
the BEM needs surface meshing, and the EFG needs points /ðxÞ ¼ pT ðxÞa; qðxÞ ¼ qT ðxÞb (21)
throughout the domain of a body.
Some publications related to the BNM are [42,128–133]. Other Appropriate selection of the approximation functions pðxÞ and
examples of boundary-based meshfree methods are the local BIE qðxÞ (each is a vector of length m) is of crucial importance, and is
(LBIE) approach [134,135], the boundary only radial basis function discussed in detail in Ref. [149].
method (BRBFM) [136], the boundary cloud method (BCM) The coefficients ai and bi are obtained by minimizing weighted
[137,138], the boundary point interpolation method (BPIM) [139], discrete L2 norms. Finally, one gets
the hybrid boundary node method (HBNM) [140], the boundary
face method (BFM) [141], the boundary element-free method [142], X
n X
n
and the Galerkin boundary node method (GBNM) [143–147]. /ðxÞ ¼ MI ðxÞ/^I ; qðxÞ ¼ NI ðxÞ^
qI (22)
The BCM [137,138] is very similar to the BNM in that scat- I¼1 I¼1
tered boundary points are used for constructing approximating
functions and these approximants are used with the appropriate where
BIEs for the problem. However, a key attractive feature of these
papers is that, unlike the BNM where boundary curvilinear coor- MðxÞ ¼ pT ðxÞðA1 BÞðxÞ; NðxÞ ¼ qT ðxÞðC1 DÞðxÞ (23)
dinates must be employed, the usual Cartesian coordinates can be
used in the BCM. Use of Cartesian, rather than curvilinear bound- Here, Eq. (22) relates the nodal approximations of / and q to their
ary coordinates, is certainly preferable, especially for 3D prob- real nodal values. The matrices A, B, C, D are defined in Ref.
lems. One disadvantage of the variable basis approach [138] (as [149].
well as the standard BNM [127]), on the other hand, is that these A moving least squares (MLS) approach is adopted in the BNM
methods do not properly model possible discontinuities in the nor- and in the EBNM. Now, one has variable weight functions within
mal derivative of the potential function across edges and corners. each cloud, i.e., wI ðx; xI Þ. The basic idea behind the choice of a
Telukunta and Mukherjee [148] have tried to combine the advan- weight function is that its value should decrease with distance

Fig. 3 Domain of dependence and range of influence. (a) The nodes 1, 2 and 3 lie within the DOD of the evaluation point
E. The ROIs of nodes 1, 2, 3, 4 and 5 are shown as gray regions. In the standard BNM, the ROI of a node near an edge, e.g.,
node 4, is truncated at the edges of a panel. In the EBNM, the ROI can reach over to neighboring panels and contain edges
and=or corners - see, e.g., node 5 (b) Gaussian weight function defined on the ROI of a node (from Ref. [149]).

031001-6 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 from a node and that it should have compact support so that the ^ þ ½K2 f^
½K1 f/g qg ¼ 0 NB equations (27)
ROI of the node (see Fig. 3(b)) is of finite extent. A possible sim-
ple choice is the Gaussian weight function, used in Ref. [149]: Equation (27) is combined with Eq. (22) and the given boundary
( 2
values in order to solve for the unknown quantities on the
eðd=dI Þ for d  dI boundary S.
wI ðdÞ ¼ (24)
0 for d > dI It is well known that surface meshing (that suffices for BEM
applications in linear problems) is much easier than (3D) domain
meshing (as is typically required for the FEM). One might then
where d ¼ gðx; xI Þ is the minimum distance measured on the sur- wonder about the need for a (pseudo) mesh-free version of the
face S (i.e., the geodesic) between x and the collocation node I BEM. It is important to point out, however, that certain situations,
with co-ordinates xI ; and the quantities dI determine the extent of such as problems with moving boundaries, or those in which opti-
the ROI (the compact support) of node I. mal shape design and/or adaptive meshing is carried out, typically
The EBNM combines the advantages of the variable basis require multiple remeshings during the solution process. This pro-
approach presented in Ref. [138] together with allowing disconti- cess can be painful even for the BEM. The BNM has a very flexi-
nuities in q. A basis (i.e., the functions in p or q in Eq. (21)) for a ble cell structure; the cells, used just for integration, only need to
cloud must satisfy two competing requirements - it must be broad cover the surface of a body and not overlap. No other topological
enough to include all cases, yet it must be narrow enough such restrictions are imposed. Therefore, this method offers an attrac-
that the matrices A and C in Eq. (23) are nonsingular [149]. tive alternative to the standard BEM, especially for this class of
The next important step is to use the MLS approximations in problems. As an illustration of the power of the BNM, the initial
Eq. (22) for the functions / and q in the regularized BIE for 3D and final cell distribution on a cube, from an ONE-step adaptive
potential theory. This regularized BIE is: (standard) BNM calculation for 3D potential theory [131], is
ð reproduced as Fig. 4 in this paper. This is, in fact, a Dirichlet prob-
0 ¼ ½Gðx; yÞqðyÞ  Fðx; yÞf/ðyÞ  /ðxÞgdSðyÞ (25) lem with the prescribed potential on the cube surface given by
S
px  px
px

1 2 3
The boundary S is partitioned into Nc cells Sk and MLS approxi- / ¼ sinh sin pffiffiffi sin pffiffiffi (28)
2 2 2 2 2
mations (Eq. (22)) are used in Eq. (25). The result is:
" The advantage of a flexible cell structure is apparent in Fig.
XNc ð ny
X
0¼ Gðx; yÞ NI ðyÞ^
qI  Fðx; yÞ 4(b) in which the cube surface has 1764 cells of different sizes.
k¼1 Sk I¼1 Application examples of the BNM and other related methods
can be found in Refs. [42,127–133,148,149].
( ny
)#
X X
nx
 MI ðyÞ/^I  MI ðxÞ/^I dSðyÞ (26)
I¼1 I¼1 2.4 Variationally Based BEM Formulations. In this sec-
tion, we review the conventional, collocation BEM from a point
where MI ðxÞ and MI ðyÞ are the contributions from the Ith node to of view that is not common in the literature, but which helps to
the collocation (source) point x and field point y, respectively. understand some key concepts and also indicates how the BEM
Also, ny nodes lie in the DOI of the field point y and nx nodes lie relates to some variationally based developments [150]. This sec-
in the DOI of the source point x. When x and y belong to the same tion also shows the connection of the BEM with the FEM, that is
cell, the cell is treated as a singular cell and special techniques are not highlighted in the literature.
needed to numerically evaluate the weakly singular integrals that We start with the derivation of the BEM. For brevity, we
involve the kernel Gðx; yÞ. This method is described in Ref. [42] restrict to elastostatics, out of which the formulation of potential
for 3D problems. Regular Gaussian integration is used for numeri- problems may be obtained as a special case. Let an elastic body
cal evaluation of nonsingular integrals. be subjected to body forces bi in domain V and traction forces ti
The final assembled equations from the EBNM are of the form: on part Sr of the boundary S. Displacements ui are known on the

Fig. 4 ONE-step multilevel BNM cell refinement for the “sinsinh cube” problem: (a) initial configuration with 96 surface
cells; (b) final adapted configuration (obtained in one step) with 1764 surface cells (from Ref. [131])

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-7

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 complementary part Su of S. We look for an approximation of the when the collocation point is placed on the boundary. Using
stress field that best satisfies equilibrium in V: boundary elements and placing the collocation point at each node,
we arrive at the basic equation of the conventional, collocation
rji ;j þ bi ¼ 0 in V (29) boundary element method, given in matrix form as

and rji nj ¼ ti on Sr , with a corresponding displacement field that Hd ¼ Gt þ e (34)

best corresponds to ui ¼ ui on Su, where nj is the outward unit Ð
normal to S. Index notation is used here. where H ¼ ½Hmn   S rjim nj uÐin dS þ dmn is a kinematic transforma-
tion matrix and G ¼ ½Gm‘   S ti‘ uim dS is a flexibility-type matrix.
2.4.1 From Variational to Consistent Weighted-Residual The error term e in Eq. (34) corresponds to residuals whose
Formulations of the BEM. The present problem might be formu- magnitude depends on the amount of rigid-body displacements
lated in the following framework of the strong form of the princi- that are implicit in the fundamental solution, Eq. (32), as well as
ple of stationary (minimum) total potential energy [151], only on how refined is the boundary mesh. This vector of residuals is
assuming that rij is a symmetric tensor that satisfies a priori the usually disregarded in the implementations shown in the literature
constitutive equation rij ¼ Cijkl uk;l : [153]. A consistent numerical formulation must take this term ex-
ð ð   plicitly into account and end up with matrices that are independent
  from Csm instead of just disregarding it. Otherwise, uncontrolled
dP udi ¼  rdji;j dui dV þ rdji nj  ti dui dS ¼ 0 (30)
V S ill-conditioning related to the matrix G may appear in the equation
system of Eq. (13). This specific issue has already been the subject
for a variation dui of ui , already extending the boundary integral of a thorough theoretical investigation [150,154] and shall not be
from Sr to S, since dui ¼ 0 on Su. The weak form of Eq. (30) is addressed in the present review. For simplicity, it is henceforth
the basis of the displacement finite element method [152] and this assumed that, for a sufficiently fine mesh, e ¼ 0.
is explicitly assumed by the superscript ð Þd attached to the stress
2.4.2 The Hybrid BEM. A variational formulation was pro-
field in Eq. (30). For brevity, the body force bi is assumed to be
posed in the year 1987 as an attempt to preserve the mechanical
zero (see Ref. [150] for the general case).
consistency of Eq. (30) and the boundary-only features of the
For the BEM, the above problem may be formulated in terms of
BEM [151]. It turned out to be a generalization of the Hellinger-
weighted residuals, in a non-variational, less restrictive frame-
Reissner potential [152] and was called hybrid BEM in reference
work than Eq. (30), using as weighting functions a field of varia-
to Pian’s developments on finite elements [155]. We require that
tions of fundamental solutions:
the potential
ð ð ð h ð 
  i 
 rji;j dui dV þ rji nj  ti dui dS ¼ 0 (31)  PR ðrsij ; udi Þ ¼ U0C rsij þ rsji;j udi dV þ ti  rsji nj udi dS
V S
V S
(35)
The sequential presentation of the two latter equations helps to
clarify the main difference between the FEM and the BEM:
be stationary for boundary displacements udi interpolated along S
besides integrability issues that are inherent to any integral state-
and such that udi ¼ ui on Su , and for domain stresses given by Eq.
ment, the premise for Eq. (30) is that dui ¼ 0 on Su , whereas the
(32). The superscripts ð Þd and ð Þs characterize that two independent
premise for Eq. (31) is that dui be the fundamental solution of Eq.
fields – of displacements and stresses – are used as trial functions.
(29).
The displacement field corresponding to rsij is given by Eq. (32).
The BEM is derived from Eq. (31) for variations of the follow-
After substituting for udi , rsij and usi in Eq. (35) according to the
ing series of fundamental solutions rij and ui , given in terms of
element discretization, we arrive at the following sets of matrix
force parameters pm ,
equations, for arbitrary variations dd and dp :
rij ¼ rijm pm ; ui ¼ ðuim þ uris Csm Þpm (32)
HT p ¼ p; F p ¼ Hd (36)
where uris , r r
for s ¼ 1…n , are n rigid-body displacements that are
In these equations, H is the same matrix of the BEM, introduced in
multiplied by, in principle, arbitrary constants Csm . In the above
Eq. (34), which performs a kinematic transformation in the second
equations, m characterizes both location and direction of applica-
set of equations and, in its preceding transpose form, anÐ equilibrium
tion of pm . Then, rijm and uim are functions – with global support
transformation. The nodal force vector p ¼ ½pm   S ti uim dS is
– of the coordinates and directions of pm referred to by m (the
equivalent, in terms of virtual work, to the applied forces ti on S.
source point), as well as of the coordinates and directions referred
Any terms related to Csm are naturally void.
to by i (the field point), at which the effects of pm are measured.
In Eq. (36), F is a symmetric flexibility matrix that transforms
For simplicity of notation, the coordinates of source and field
point forces p into equivalent nodal displacements d , as referred
points are not explicitly represented.
to the sets of nodal quantities used to interpolate the domain stress
Substituting for drijm and duim in Eq. (31) according to
field (fundamental solutions). The definition of F is given in the
Eq. (32), we obtain, for arbitrary dpm , the modified expression of
following, also repeating the definition of H in order to clarify
the Somigliana’s identity,
some key concepts [150,151]:
ð ð ð
ð
um ¼ ti uim dS  rjim nj ui dS þ Csm ti uris dS (33)
S S S
½ Hmn Fmn   rjim nj h uin uin idS þ h dmn Umn 
i
S
ð ð
which is used to evaluate displacements um (and, subsequently,
stresses) at a domain point m for prescribed forces ti , and bound-  fp rjim nj h uin uin idS þ rjim nj h uin uin idS
S Sdisc
ary displacements ui . The last term vanishes only if ti are in equi- 
 Hfp Ffp þ Hdisc ½ I U 
librium, which is not necessarily true when we are dealing with
approximations. Then, the results are in principle influenced by (37)
some arbitrary constants Csm [153].
Equation (33) may also be used to evaluate displacements ui The matrix F shares the singularity features of the matrices H
and traction forces ti as the problem’s unknowns along Sr and Su (singularity of rjim ) and G (improper integral due to uim ), except

031001-8 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 when m and n refer to the same nodal point. The evaluation of the All vectors and matrices above have already Ð been introduced,
singular integral is carried out in terms of a finite part plus a dis- except for L, which is expressed as L‘m ¼ S uim ti‘ dS [150]. We
continuous term, as indicated in the second and third lines of Eq. may obtain from the above equations the stiffness relation
(37). The impossibility of obtaining the coefficients about the h i
main diagonal of F in Eq. (37) is evident, as the coefficients of LT Gð1Þ F GðTÞ L d ¼ p (43)
the matrix U  Umn 
are the displacement fundamental solution
evaluated at point n for a source point applied at m: If m ¼ n, this where Gð1Þ must be expressed in terms of generalized inverses,
would correspond to a meaningless value at a point that lies out- in a consistent formulation, as G is in general rectangular and pos-
side the domain of interest V. The response of U  Umn 
for sibly not full rank [150]. In the latter formulation, the undefined
m ¼ n is actually not infinite, but rather undefined in the frame of coefficients of F are obtained by means of a spectral property
an integral statement. The evaluation of such terms, for a finite do- related to G, which takes the place of Eq. (39), although equiva-
main, must succeed in a linear algebra framework, as briefly lent to it. The same topological concerns made with respect to the
explained in the following.
d r hybrid BEM affect the hybrid displacement BEM [150].
Let W ¼ ½Wns  2 Rn n be a matrix whose columns form an or- The aim of the present brief review in this subsection is to sit-
thogonal basis of the nodal displacements d of Eq. (36) related to uate the collocation BEM among several developments that are
the rigid-body displacements of a finite domain. Then, as by con- based on energy concepts, as in Eqs. (30), (31), (35), and (41).
struction W ¼ NðHÞ, we may define a matrix V ¼ NðHT Þ and Some applications of the variationally based BEM and many in-
check that, for linear algebra consistency of Eq. (36), depth conceptual details are given in Refs. [150,151,154,157,
158,160–169].
WT HT p ¼ WT p ¼ 0 and VT F p ¼ VT Hd ¼ 0 (38)
3 Fast Solution Methods
for any set of force parameters p . It results that, for a finite do-
In addition to the new mathematical formulations for the BEM,
main V,
as reviewed in the previous section, a few new fast solution meth-
ods for solving the BEM systems of equations also emerged in the
VT F ¼ 0 (39) last decade. With the developments of these fast solution methods,
the BEM can now be applied to solve large-scale and practical en-
which is a linear algebra means of evaluating the coefficients of gineering problems with the number of boundary elements on the
F for m and n referring to the same nodal point, which could not order of a few millions on desktop computers. In this section, we
be obtained from Eq. (37) [150,151,156]. will review three fast solution methods, namely, the fast multipole
Solving for p in Eq. (36), we obtain the stiffness relation method (FMM), the pre-corrected fast Fourier transformation
(pFFT) method, and the adaptive cross approximation (ACA)
h  1 i method. The first two methods were mainly developed in the U.S.,
HT F þ VVT H d ¼ p (40)
while the third one was originated in Europe. Application exam-
ples of the new formulations in the BEM and these fast solution
with an inverse matrix that comes from the theory of generalized methods will be presented in Sec. 4.
inverses [150,151]
The same developments outlined above apply to an unbounded 3.1 Fast Multipole Method. In the mid-1980s, Rokhlin and
domain that is complementary to V. The resulting equations help Greengard [170–172] pioneered the innovative fast multipole
to understand several topological issues related to Eq. (39), for a method that can be used to accelerate the solutions of the BEM by
general non-convex, multiply connected domain. Such relations several orders, promising to reduce the CPU time in the FMM-
have eventually led to the simplified hybrid BEM [157], which is accelerated BEM to O(N). We call the fast multipole accelerated
in general as accurate as the hybrid BEM or the CBEM, although BEM the fast multipole BEM (or FMBEM). Some of the early
computationally less expensive [158], since the time-consuming work on the fast multipole BEM in applied mechanics can be
evaluation of F in Eq. (37) can be circumvented. found in Refs. [173–177]. A comprehensive review paper on the
In either hybrid or simplified hybrid BEM, once d, p and p are fast multipole BIE and BEM up to 2002 can be found in Ref.
solved, for adequate boundary conditions, the results at internal [178], a tutorial paper in Ref. [179], and a textbook in Ref. [44].
points can be directly obtained via Eq. (32), thus avoiding The main idea of the fast multipole BEM is to apply iterative
the time-consuming evaluation of the Somigliana’s identity of solvers (such as GMRES [180]) to solve Eq. (13) and use the FMM
Eq. (33) [150,151]. to accelerate the matrix-vector multiplication (Ak) in each iteration,
without forming the entire matrix A explicitly. Direct integrations
2.4.3 The Hybrid Displacement BEM. Motivated by the are still needed when the elements are close to the source point,
hybrid BEM, Figueiredo and Brebbia proposed an alternative var- whereas fast multipole expansions are used for elements that are far
iational formulation, which was called hybrid displacement BEM away from the source point. A main reason for the reduction in
[159]. It consists of the application of the Hu potential [152], operations in the FMM is due to the fact that the Green’s functions
given in the following using the notation already introduced for in the BIEs can be expanded in the following form:
the BEM and hybrid BEM: X
ð ð ð Gðx; yÞ ¼ Gxi ðx; yc ÞGyi ðy; yc Þ (44)
PHD ðusi ; udi ; ti Þ ¼ U0 ðusi;j ÞdV  udi ti dS  ti ðusi  udi ÞdS i
V Sr S
where yc is an expansion point near y. This can be achieved by
(41)
use of various forms of expansions, including, but not limited to,
Taylor series expansions. By using an expansion as in Eq. (44),
also with fundamental solutions as domain interpolation functions, we can write the original integral, such as the one with the G ker-
but in terms of displacements. The compatibility between domain nel in BIE (Eq. (7)), as
and boundary displacements, usi and udi , is weakly enforced by
using the Lagrange multiplier ti . After numerical discretization, ð X ð
Eq. (41) leads to Gðx; yÞqðyÞdSðyÞ ¼ Gxi ðx; yc Þ Gyi ðy; yc ÞqðyÞdSðyÞ
Sc i Sc
F p ¼ Gt; p ¼ LT t; GT p ¼ Ld (42) (45)

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-9

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 where Sc is a subset of S away from x. In the conventional BEM, X
k
the integral is computed with the expression on the left-hand side Mk ðzc0 Þ ¼ Ikl ðzc  zc0 ÞMl ðzc Þ (53)
of Eq. (45) directly. Any changes in the location of the source l¼0
point x will require reevaluation of the entire integral. In the fast
multipole BEM, when the source point x is far away from Sc , the Next, we introduce the so-called local expansion about the source
original integral is computed with the expression on the right- point z0 . Suppose zL is a point close to the source point z0 , that is,
hand side of Eq. (45), in which the new integrals need to be eval- jz0  zL j jzL  zc j. From the multipole expansion in Eq. (51),
uated only once, independent of the locations of the source point we have the following local expansion:
x. That is, the direct relation between x and y is cut off by use of ð
the expansion and introduction of the new “middle” point yc . 1 X1

Additional expansions and translations, as well as a hierarchical Gðz0 ; zÞqðzÞdSðzÞ ¼ Ll ðzL ÞIl ðz0  zL Þ (54)
Sc 2p l¼0
tree structure of the elements, are introduced in the fast multipole
BEM to further reduce the computational costs.
As an example, we discuss the expansions in the FMM for 2D where the coefficients Ll ðzL Þ are given by the following moment-
potential problems and the main procedures and algorithms in the to-local (M2L) translation:
fast multipole BEM. Consider the following integral with the G X
1
kernel in BIE (Eq. (7)): Ll ðzL Þ ¼ ð1Þl Olþk ðzL  zc ÞMk ðzc Þ (55)
ð k¼0
Gðx; yÞqðyÞdSðyÞ (46)
Sc If the point is moved from zL to zL0 , we have the following local-
to-local (L2L) translation:
in which Sc is a subset of boundary S and away from the source
point x, and Gðx; yÞ ¼ ð1=2pÞ logð1=r Þ for 2D. Using complex X
p

notation, that is, replace the source point x ) z0 ¼ x1 þ ix2 Ll ðzL0 Þ ¼ Iml ðzL0  zL ÞLm ðzL Þ (56)
m¼l
and the field
pffiffiffiffiffiffipoint
ffi y ) z ¼ y1 þ iy2 in the complex plane,
where i ¼ 1, we can write
Multipole expansion moments for the integrals with the F kernel,
Gðx; yÞ ¼ RefGðz0 ; zÞg (47) as well as those in the hypersingular BIE, can be defined similarly.
The M2M, M2L, and L2L translations remain the same for these
where integrals.
The algorithms in the fast multipole BEM can be described as
1 follows:
Gðz0 ; zÞ ¼  logðz0  zÞ (48)
2p Step 1. Discretize the boundary S in the same way as in the
conventional BEM. For example, we apply constant ele-
is the fundamental solution in complex notation and Ref g indi- ments to discretize boundary S of a 2D domain as shown in
cates the real part of the function. Thus the integral in (46) is Fig. 5.
equivalent to the real part of the following integral: Step 2. Determine a tree structure of the elements. For a 2D do-
ð main, consider a square that covers the entire boundary S
Gðz0 ; zÞqðzÞdSðzÞ (49) and call this square the cell of level 0. Next, start dividing
Sc this parent cell into four equal child cells of level 1. Con-
tinue dividing in this way cells that contain elements. Stop
We first expand the kernel function to separate the source point z0 dividing a cell if the number of elements in that cell is fewer
and field point z. To do this, we introduce an expansion point zc than a pre-specified number (this number is 1 in the example
close to the field point z, that is, jz  zc j jz0  zc j. Applying the in Fig. 5). A cell having no child cells is called a leaf
Taylor series expansion, we obtain

1 X1
Gðz0 ; zÞ ¼ Ok ðz0  zc ÞIk ðz  zc Þ (50)
2p k¼0

where the two auxiliary functions Ik ðzÞ ¼ zk =k! ðfor k

0Þ,
Ok ðzÞ ¼ ðk  1Þ!=zk ðfor k
1Þ and O0 ðzÞ ¼  logðzÞ. The inte-
gral in (49) is now evaluated by the following multipole
expansion:
ð
1 X1
Gðz0 ; zÞqðzÞdSðzÞ ¼ Ok ðz0  zc ÞMk ðzc Þ (51)
Sc 2p k¼0

where
ð
Mk ðzc Þ ¼ Ik ðz  zc ÞqðzÞdSðzÞ; k ¼ 0; 1; 2; ::: (52)
Sc

are called moments about zc , which are independent of the collo-

cation point z0 and need to be computed only once. If the expan-
sion point zc is moved to a new location zc0 , we compute the
moment at the new location by the moment-to-moment (M2M) Fig. 5 A hierarchical cell structure covering all the boundary
translation elements

031001-10 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 6 A hierarchical quad-tree structure for the 2D boundary element mesh

(shaded cells in Fig. 5). A quad-tree structure of the cells lems. During the next decade, this technique has been improved,
covering all the elements is thus formed after this procedure extended and applied together with the BEM to obtain solutions
is completed (Fig. 6). of various large-scale engineering problems governed by Laplace
Step 3. Upward pass. Compute the moments on all cells at all equation [188,189], Stokes equation [190,191], linear elasticity
levels l
2 and trace the tree structure upward (Fig. 6). For [192], coupled electrostatic and elasticity [193], Poisson’s equa-
a leaf, Eq. (52) is applied directly (with Sc being the set of tion [194] and even quasi-linear equations [195]. In recent years,
the elements in the leaf and zc the centroid of the leaf). For a extensions to dynamic problems have also been conducted
parent cell, calculate the moment by summing the moments [196,197], resulting in a more powerful and versatile method. A
on its four child cells using the M2M translation (Eq. (53)), distinct advantage of the pre-corrected FFT technique, in addition
in which zc0 is the centroid of the parent cell and zc is the to its kernel independent nature, is its simplicity. Unlike the
centroid of a child cell. FMM, this method requires only one Cartesian grid and there is
Step 4. Downward pass. In the downward pass, we compute the no need to construct a hierarchical structure. As such, the imple-
local expansion coefficients on all cells starting from level 2 mentation is straightforward. The complexity of the method is
and tracing the tree structure downward to all the leaves OðNÞ þ OðNg log Ng Þ, where N is the total number of surface ele-
(Fig. 6). The local expansion associated with a cell C is the ments and Ng is the total number of grid points. For a discussion
sum of the contributions from the cells in the interaction list of the complexity, readers are referred to Ref. [187].
of cell C and from all the far cells (see Refs. [44,179] for Similar to most acceleration schemes, the pre-corrected-FFT tech-
definitions). The former is calculated by use of the M2L nique achieves acceleration also by computing the far-field interac-
translation, Eq. (55), with moments associated with cells in tion, that is, Ak, where the source point is located far away from the
the interaction list, and the latter is calculated by use of the field element, approximately via efficient means. Consider, for exam-
L2L translation, Eq. (56), for the parent cell of C with ple, the evaluation of potential at point x due to charges qi at points
the expansion point being shifted from the centroid of C’s yi distributed randomly inside a cell. The cell size is much smaller
parent cell to that of C. than the distances between x and yi . Instead of clustering all the dis-
Step 5. Evaluation of the integrals. Compute integrals from ele- crete charges qi into one lumped charge Q and evaluating the poten-
ments in leaf cell C and its adjacent cells directly as in the tial at x due to the lumped charge Q as is done in the FMM, the
conventional BEM. Compute the integrals from all other pre-corrected FFT scheme computes the potential at x by first replac-
cells (in the interaction list of C and far cells) using the local ing each charge qi with point charges laying on a uniform grid; for
expansion (Eq. (54)). example, the cell vertices. The grid representation allows the fast
Step 6. Iterations of the solution. The iterative solver updates Fourier transformation to be used to efficiently perform potential
the unknown solution vector k in the system Ak ¼ b and computation on the grid. This grid potential is then interpolated onto
continues at Step 3 to evaluate the next matrix and vector point x to obtain the potential at x. In the section that follows, we use
multiplication (Ak) until the solution k converges within a a 2D potential problem to illustrate the steps of the pFFT scheme and
given tolerance. discuss the algorithms.
As mentioned previously, in the pFFT scheme, the far-field inter-
More details of the fast multipole BEM can be found in Refs. action is calculated approximately with the aid of a regular Cartesian
[44,178,179]. The fast multipole algorithm discussed in this sec- grid. As such, the first step in this scheme is to construct the grid.
tion is the original algorithm [172], which is efficient for models This is done by first constructing a parallelepiped and superimposing
in which the elements are about the same size and distributed uni- it onto a 3D problem domain with its boundary being discretized into
formly in a bulky domain. For BEM models with non-uniform n surface elements. This parallelepiped is then subdivided into an
element distributions and especially with large elements adjacent array of small cubes so that each small cube contains only a few sur-
to smaller elements, the so-called adaptive FMMs are more effi- face elements. The vertices of the cubes form the grid which can be
cient, in which the definitions of the adjacent cells and cells in the further refined if needed. In addition, these cubes are also used to
interaction list are further refined. Discussions on the adaptive partition the near and the far fields. In most implementation schemes,
algorithms can be found in Refs. [181–185]. the near field of the elements within a cube includes only the nearest
neighbors of that cube. This, however, is not rigid. The optimum
3.2 Pre-Corrected Fast Fourier Transformation number of cubes should be determined based on the balance
Method. The pre-corrected-FFT accelerated technique was pro- between accuracy and efficiency. For a fixed surface discretization,
posed in 1997 by Phillips and White [186,187] in an effort to de- more cubes mean a larger far field and hence a higher efficiency, but
velop a kernel independent acceleration technique to rapidly solve a lower accuracy because the direct evaluation region would
integral equations associated with Laplace and Helmholtz prob- be smaller. Figure 7(a) shows a 2D view of a parallelepiped

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-11

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 (3) Projection of the element charge onto the surrounding grid
pointsÐ based on polynomial interpolation, i.e., to compute
q ¼ S P ðqðyÞdSðyÞÞ.
(4) Calculation of the grid-to-grid interactionP
using the fast Fou-
rier transformation, i.e., to compute ul ¼  Gðxl ; y Þ  q .
(5) Interpolation of the
P grid potentials back to elements, i.e., to
compute uðxÞ  l Wl  ul .
(6) Subtraction of the near-field interaction.
(7) Computation of the near-field interaction using direct
calculation.
(8) Summation of the near-field and far-field contributions.
For a detailed description of this technique, readers are referred
to Refs. [187,193].
Fig. 7 (a) 2D view of a parallelepiped superimposed onto a dis-
cretized sphere. The black dots indicate grid points and the
cubes are represented by the solid lines; (b) Illustration of 3.3 Adaptive Cross Approximation Method. The adaptive
the four steps in the pFFT scheme; the gray region denotes the cross approximation (ACA) follows a strategy different from the
near-field region of the yellow element. above discussed techniques to reduce the complexity of the BEM
with respect to operations as well as to storage. However, the prin-
cipal idea is the same and the same properties of the underlying
superimposed onto a discretized sphere. The parallelepiped has been kernel functions are used. The above discussed methodologies
divided into a 3  3  3 array of small cubes. Each cube has been allow for use of the matrix-vector multiplication with almost lin-
further discretized to form the Cartesian grid indicated by black dots. ear complexity. However, the only approach that allows for all
The grid is extended outside the parallelepiped in order to ensure that matrix operations (matrix-vector, matrix-matrix product, matrix-
the number of grid points is an even number. Note that the surface matrix addition, matrix inversion, LU-decomposition, etc.) with
elements and the pFFT cubes can intersect with each other. There is almost linear complexity are the so-called hierarchical matrices
no need to maintain any consistency between the surface elements (H-matrices) introduced by Hackbusch [198]. They can be under-
and the cubes. stood an as algebraic structure reflecting a geometrically moti-
To evaluate the far-field interaction, for example, to compute the vated partitioning into sub-blocks. Each sub-block is classified to
potential uðxÞ at the centroid of the purple element induced by the be either admissible or not (similar to the clustering in Figs. 5
charge qðyÞ distributed on the yellow element as shown in Fig. and 6). This block structure points out the fact that H-matrix arith-
7(b), three steps are involved. First, the element charge is projected metic is easily parallelizable [199].
onto the surrounding grid points q located on or within the cube After having concluded the setup of an H-matrix, admissible
that encloses the yellow element. The potential at each grid point blocks have to be approximated. The FMM deals with the analyti-
ul surrounding the evaluation element, that is, the purple element, cal decomposition of integral kernels; hence, the procedure
is then computed using the fast Fourier transformation technique. becomes problem dependent. The problem-independent classes
Finally, the potential uðxÞ is obtained by interpolating the grid are the so-called algebraic approximation methods. The singular
potentials ul back to the element. Mathematically
Ð this procedure is value decomposition (SVD) leads to the optimal approximation,
equivalent toPapproximating uðxÞ  ¼ ÐS Gðx; yÞqðyÞdSðyÞ by uðxÞ however, with OðN 3 Þ complexity. Less expensive algorithms are
P
 l Wl   Gðxl ; y Þ  q , q ¼ S P ðqðyÞdSðyÞÞ, where Wl
the Mosaic skeleton method developed in Ref. [200] and the suc-
is an interpolation operator that interpolates the grid potential ul to cessively developed ACA. It has been applied by Bebendorf [201]
the potential at x and P is the projection operator that projects the to the approximation of BEM matrices for the first time. The out-
charge at point y to charges at grid points y surrounding y. In the standing feature of ACA compared to SVD is that it requires only
implementation of this scheme, the projection is done for all ele- the evaluation of some original matrix entries and the approxima-
ments before the FFT is performed; hence the grid charges repre- tion is still almost optimal. The main idea is the approximation of
sent the distributed charges of the entire problem domain. The a matrix A 2 Cts
resulting grid potentials thus contain the contributions from the
neighboring grid charges. To ensure accuracy, this part of the A  Sk ¼ UV T with U 2 Ctk and V 2 Csk (57)
potential must be subtracted from the total potential and the contri-
butions from the neighboring elements, that is, the near-field inter- with a small rank k compared to t and s. This low rank representa-
actions, Ð which are computed by directly evaluating tion can be found whenever the generating kernel function G(x,y)
uðxÞ ¼ S Gðx; yÞqðyÞdSðyÞ, should then be added back. This step in the computational domain of A is asymptotically smooth. As
is the pre-correction step. shown in Refs. [202] and [203] all kernel functions G(x,y) of
Both projection and interpolation can be performed using poly- elliptic operators with constant coefficients and x 6¼ y have this
nomial interpolation techniques and the complexity of this process property. Only in the case x ¼ y they become singular and are not
is OðNÞ. The convolution step, which is conducted via FFT, smooth.
requires about OðNg log Ng Þ operations, where Ng is the total num- Due to the kernel independent idea, ACA can be used in a
ber of grid points. For most problems, Ng  N and hence the com- black-box like manner. Its coding and adaptation to existing codes
plexity of the pFFT scheme is OðN log NÞ. The direct calculation is straight forward. The algorithm is robust and it is based on a
also requires OðNÞ operations. stopping criterion depending on a prescribed approximation accu-
The main steps in the pre-corrected-FFT acceleration technique racy e. Applications in elasticity may be found in Refs. [204,205],
are summarized as follows: for crack problems in [206,207], and for electromagnetism in Ref.
[208], to cite a few.
(1) Construction and superposition of the 3D Cartesian grid
and the discretized problem domain (Fig. 7(a)). 3.3.1 Hierarchical Matrices. Due to the previously described
(2) Determination of the near and far fields for each element kernel properties of elliptic operators, the necessity to separate the
using the cubes of the parallelepiped; for example, for the near- from the far-field becomes evident. Low-rank approxima-
yellow element inside cube S, its near field includes all ele- tions of the type (Eq. (57)) can be obtained only for well-separated
ments inside cube S and its nearest neighboring cubes, as computational domains x 6¼ y. Thus, H-matrices [202,203] are
illustrated by the gray region in Fig. 7(b). used. Their setup is based on the following idea: The index sets I

031001-12 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 and J of row and column degrees of freedom are permuted in such
a way that those who are far away from each other do also obtain
indices with a large offset.
First, by means of a distance based hierarchical subdivision of I
and J cluster trees TI and TJ are created. In each step of this proce-
dure, a new level of son clusters is inserted into the cluster trees.
A son cluster is not further subdivided and is called to be a leaf if
its size reaches a prescribed minimal size bmin. Basically, two
approaches can be distinguished. First, the subdivision based on
bounding boxes splits the domain into axis-parallel boxes which
contain the son clusters. Second, the subdivision based on princi-
pal component analysis splits the domain into well-balanced son
clusters leading to a minimal cluster tree depth. Details for both
approaches can be found in Ref. [203].
Now, the H-matrix structure is defined by the block cluster tree
TIJ :¼ TI  TJ . Its setup is performed by means of the following
admissibility criterion:

minðdiamðtÞ; diamðsÞÞ  g distðt; sÞ (58)

with the clusters t  TI, s  TJ, and the admissibility parameter

0 < g < 1. The diameter of the clusters t and s and their distance is
computed as usual, that is,

diamðtÞ ¼ max jxi1  xi2 j Fig. 8 H-matrix with 3279 entries and rank numbers of ACA
i1 ;i2 2t

diamðsÞ ¼ max jyj1  yj2 j with the ith row vector and jth column vector defined as
j1 ;j2 2s

distðt; sÞ ¼ min jxi  yj j vþ1 ¼ cþ1 ðR Þi and uþ1 ¼ ðR Þj (62)
i2t;j2s

Each cluster t and s is associated with its computational domain xt The residuum R is minimized and the rank of the approximant S
and ys on C. The supports of the corresponding degrees of free- is increased step by step. The pivot cþ1 is chosen to be the largest
dom of row i and column j are denoted by xi and yj, i.e., entry in modulus of either the row ðR Þi or column ðR Þj . Finally,
[ [ the approximation stops if the following criterion holds:
xt :¼ xi and ys :¼ yj with xi ; yj 2 R3 (59)
i2I j2J kuþ1 kF kvþ1 kF < e kSþ1 kF (63)

If Eq. (58) is fulfilled, a block b ¼ t  s is admissible. If the condi- Note, the entire matrix A will never be generated. Therefore, spe-
tion in Eq. (58) is not fulfilled, the admissibility is recursively cial care has to be taken in order to find the pivot such that the
checked for their son clusters, until either Eq. (58) holds or both algorithm converges to the prescribed accuracy e [202,209]. The
clusters t and s become leafs. In the latter case, block b is not ad- numbers in Fig. 8 are the obtained ranks.
missible. Admissible blocks have well separated computational As already mentioned above, a crucial point is the right choice
domains xt and ys and the algorithm presented in the next section of the pivot value. This becomes even more important in case of
is used to approximate them. Not admissible blocks must be eval- vector valued problems like, e.g., elasticity. In this case each entry
uated in the standard fashion. An example for the hypersingular is of matrix type Aij 2 C33 and the pivot might be defined as
operator with 3279 entries is presented in Fig. 8.  
 
In this example, 1171 clusters have been used. The red colored c ¼ ðR Þ1
ij with  ðR  Þij  ¼ max (64)
p
blocks are not admissible, whereas the green ones are admissible.
The numbers represent the final low rank of the sub-blocks. They
are computed with ACA, which is discussed in the next section. In no norm k kp it is guaranteed that a proper pivot entry can be
found. For example, if ðR Þij contains one very small entry com-
3.3.2 Approximation of the Admissible Blocks. A remark to pared to the remaining entries of ðR Þi the pivot row U is scaled
the notation in this section: ðAÞij denotes the ij-th entry, whereas up by c and Eq. (63) does not hold. If ðR Þij contains a zero entry
ðAÞi and ðAÞj are the ith row vector and jth column vector, respec- the pivot c is not even defined. This becomes evident if, e.g., the
tively, for matrix A. The idea of ACA is to split up a matrix double layer operator, Eq. (5) is evaluated on some plane
A 2 Cts into A ¼ Sk þ Rk where Sk denotes the rank k approxi- Ck fx; y 2 R3 : yk  xk ¼ 0g that lies perpendicular to the coordi-
mation of A and Rk the residuum to be minimized. Starting from nate axis k. A work around is the partitioning of the system matrix
with respect to each degree of freedom (for details, see Ref. [210]).
S0 :¼ 0 and R0 :¼ A (60)

a first pivot c1 ¼ ðR0 Þ1

ij has to be found, where i and j are the row
4 Emerging Applications
and column indices of the actual (0–th in this case) approximation
step. Hints for the right choice of the initial pivot can be found in 4.1 Modeling of MEMS/NEMS
Ref. [202]. In each ongoing step , the scaled outer product of the 4.1.1 Electrostatically Driven MEMS/NEMS Devices. Numer-
pivot row and column is subtracted from R and added to S ical solutions of electrically actuated microelectromechanical sys-
tems (MEMS) have been carried out for nearly 20 years by using
Rþ1 :¼ R  uþ1 vTþ1 the BEM to model the exterior electric field and the finite element
(61) method or the BEM to model deformation of the structure. The
Sþ1 :¼ S þ uþ1 vTþ1 commercial software package MEMCAD [211], for example, uses

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-13

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 and its resonant frequency and the quality factor, defined as the ra-
tio of the total energy to the dissipated energy within one cycle,
can be extracted. In most applications, the sensing mechanism
relies on the fact that the change in the resonant frequency is
related to the physical quantities to be measured. The resolution
of these sensors thus depends strongly on air damping, a dominant
dissipation source for devices operating in open air and even in a
low vacuum.
Air damping is determined by gas transport and its interaction
with the structure. A key parameter that influences the transport is
Fig. 9 Two parallel conducting beams the Knudsen number (Kn), defined as the ratio of the mean free
path of gas molecules to the characteristic length of the fluid field.
the FEM package ABAQUS for mechanical analysis together with Based on the value of Kn, gas can be classified roughly into four
a BEM code FastCap [212] for electric field analysis. regimes, the continuum regime (Kn < 0.01), the slip regime
A generic MEMS structure, consisting of two parallel conduct- (0.01 < Kn < 0.1), the transition regime (0.1 < Kn < 10) and the
ing beams, is shown in Fig. 9. The electric potential /, in the free-molecule regime ((Kn > 10). In the free-molecule regime,
region V exterior to the two beams, is governed by Laplace’s gas transport is dominated by the collisions between molecules
equation (Eq. (1)) and is prescribed on the boundary S of the two and structures, while in the continuum and slip regimes it is domi-
beams (it equals /1 on the surface of beam 1 and /1 on the sur- nated by the collisions between gas molecules. This fact results in
face of beam 2). different governing equations and hence requires different model-
The corresponding BIE in 2D space is of the form (see, e.g., ing approaches for different gas regimes.
Ref. [213]) Continuum and slip regimes. For resonators oscillating at rela-
ð tively low frequencies, the incompressible Stokes equation
log r ðx; yÞrðyÞ describes accurately gas behavior in both continuum and slip
/ðxÞ ¼  dSðyÞ þ C (65)
2pe regimes. In the absence of body force, it reads

where r ¼ eð@/=@nÞ ¼ eq is the charge density at a point on a r  u ¼ 0 ixqu ¼ rP þ lr2 u (68)
beam surface and e is the dielectric constant of the exterior me-
dium. The total charge Q on the surface S is where u and P denote the amplitudes of the velocity and pressure
ð of gas, q is the density of gas, l is the coefficient of viscosity, and
Q ¼ rð yÞdSð yÞ (66) x is the oscillation frequency. The equivalent integral equation of
S Eq. (68) can be formulated as
ð
Given the potential /, and the total charge Q, Eqs. (65) and (66) cðxÞuðxÞ ¼ ½Uðx; y; xÞtðyÞ  Tðx; y; xÞuðyÞdSðyÞ (69)
can be solved for the charge density r on the total beam surface S. S
This charge density is then used to determine the traction h at any
point on the beam surface according to In Eq. (69), U and T are the free-space Green’s functions of oscil-
latory Stokes flows [231] and t are tractions exerted by the struc-
r2 ðxÞ ture on the fluid.
hðxÞ ¼  n (67)
2e The boundary conditions in the continuum regime follow the
classical no-slip conditions, that is, the tangential gas velocity
The mechanical problem of deformation of the beams, subjected equals the wall velocity at the boundary. In the slip regime, gas
to traction h, is generally solved by the FEM. One can employ an rarefaction causes slip between the tangential velocities of the
iterative procedure, as has been done by many authors, or a total flow and that of the solid surface (the wall). A phenomenological
Lagrangian approach, first proposed in Ref. [214]. slip model is shown in Eq. (70). In a non-dimensional form, the
Examples of MEMS simulation from around the mid-1990s are slip velocity at an isothermal and locally flat wall is given as
in Refs. [213,215–219]. The total Lagrangian approach has been
employed in Refs. [214,220–222]. (In addition, Refs. [217,222] 2  rv Kn @Ug
address fully-coupled dynamic analysis of MEMS). A coupled U g  Uw ¼ (70)
electrostatic structural analysis of MEMS devices is described in rv 1  bKn @n
Ref. [193], where a pre-corrected-FFT accelerated BEM approach
has been employed. Special BEM techniques to efficiently handle where Ug ; Uw are normalized tangential velocities of the flow and
thin structures, that are typically present in MEMS, have been the wall at the interface, respectively, @Ug=@n is the normal
employed in Refs. [223–228]. Of these papers, Refs. [226,227] derivative of the normalized tangential velocity at the interface,
address dynamic analysis while Ref. [228] addresses large-scale rv is the tangential momentum accommodation coefficient, and
problems by the fast multipole BEM. Examples of BEM simula- b is a high-order slip coefficient [232]. An integral equation for
tions of nanoelectromechanical systems (NEMS) can be found in the normal derivative of the velocity is thus required to solve
Refs. [229,230]. Stokes equation with slip boundary conditions.
The first implementations of Eq. (69) have been proposed in the
4.1.2 Inertial Sensors. Another successful application of the quasi static limit where inertia forces can be neglected
BEM in the modeling and design of MEMS/NEMS is air damping [190,233–235]. The effect due to the inertial force is addressed in
analysis of inertial sensors, like the currently widely used acceler- Ref. [190]. In the quasi-static case, Eq. (69) reduces to a classical
ometers and gyroscopes, and other types of micro resonators. Dirichelet problem of incompressible elasticity but turns out to be
Micro resonators are a class of MEMS devices with applications severely ill-conditioned. This issue has been resolved in Refs.
ranging from various physical, chemical and biological sensors to [233–235] with a Burton-Miller like approach coupling the veloc-
high-frequency filters, oscillators in wireless communication sys- ity and the traction BIEs.
tems. In a typical micro resonator, a mechanical structure is driven The extension to the moderate-frequency regime and to the slip
into its resonance by an integrated micro actuator. The vibration regime has been addressed in Refs. [191,236] and acceleration tech-
of the structure is then measured either electrically or optically niques such as the pre-corrected FFT technique [190,191,237,238],

031001-14 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 10 SEM pictures of (a) a laterally oscillating beam resonator, and (b) a biaxial
accelerometer

the fast multipole method [233–236,239] and wavelets [240] have is smooth around x), and Uij ðx; yÞ and Tij ðx; yÞ are the two kernel
been employed to accelerate the computation. functions (Kelvin’s solution). For each inclusion, the BIE can be
Air damping and quality factors of several MEMS resonators; written as
for example, a laterally oscillating beam resoantor (Fig. 10(a))
and a biaxial accelerometer (Fig. 10(b)) have been simulated and ð h i
1 ðaÞ ðaÞ ðaÞ ðaÞ ðaÞ
compared with experimental data [190,191,233–235]. Very good u ðxÞ ¼ Uij ðx;yÞtj ðyÞ  Tij ðx;yÞuj ðyÞ dSðyÞ; 8x 2 Sa
agreements have been achieved. 2 i Sa
(72)
Transition and free molecule regimes. The situation is, how-
ever, less developed and understood in the transition regime, based
ðaÞ ðaÞ
on the formal classification given above. Even though a boundary- for a ¼ 1; 2; :::; n, in which ui and ti are the displacement and
ðaÞ ðaÞ
volume integral equation approach has been formulated theoreti- traction, respectively, for inclusion a, and Uij ðx; yÞ and Tij ðx; yÞ
cally in Ref. [241], due to the complicated collision integral the are the two kernel functions of inclusion a. They are functions of
BEM is not suitable and other techniques of deterministic and sta- the shear modulus, Poisson’s ratio, and outward normal for inclu-
tistical nature are generally preferred. On the contrary, the collision- sion a.
less (or free-molecule) flow lends itself again to the development of Hypersingular or traction BIE can also be applied in the matrix
robust and competitive BIE approaches. The free-molecule flow as well as in the inclusion domains. In fact, the dual BIE formula-
represents the limiting case where the Knudsen number tends to in- tion (a linear combination of the conventional BIE and hypersin-
finity and collisions between molecules can be neglected. In inertial gular BIE) is preferred for modeling inclusion problems in which
MEMS, this applies typically at pressures in the range of a few thin shapes often exist and could present difficulties for the con-
mbars and below, as is often encountered in package sealed sensors ventional BIE formulation when it is applied alone.
(e.g., gyroscopes, magnetometers and resonators). Assume that the inclusions are perfectly bonded to the matrix,
Typical techniques for the simulation of gas dissipation in this re- that is, there are no gaps or cracks and no interphase regions, we
gime belong to the class of test particle Monte Carlo (TPMC) meth- have the following interface conditions:
ods [241] where the trajectories of large ensembles of molecules are
analyzed with suitable models for gas–surface interaction in order to ðaÞ ðaÞ
compute averages of the quantities of interest. Compared to deter- ui ¼ ui ; ti ¼ ti (73)
ministic techniques, the computational cost of TPMC is much higher
due to its statistical nature. In Refs. [242,243] a deterministic tech- for a ¼ 1; 2; :::; n. That is, the displacements are continuous and
nique based on integral equations has been put forward and has been the tractions are in equilibrium at the interfaces. Based on these
shown to be indeed very competitive with TPMC methods for the assumptions, we can write the discretized form of the multido-
typical working conditions of inertial MEMS. main BIEs as follows, which gives faster convergence with itera-
tive solvers [44]:
4.2 Modeling of Fiber-Reinforced Composites. In this sec-
tion, we review the BEM formulation for multi-domain elasticity
problems [44], that can be applied to model fiber-reinforced com-
posite materials, functionally graded materials, and other inclu-
sion problems in elasticity.
Consider a 2D or 3D elastic domain V0 with an outer boundary
S0 and embedded with n elastic inclusions Va with interface Sa ,
where a ¼ 1; 2; :::; n (Fig. 11). Note: In this section, n is used to
indicate the number of inclusions. For the matrix domain V0, we
have the following BIE:
ð
1 
ui ðxÞ ¼ Uij ðx; yÞtj ðyÞ  Tij ðx; yÞuj ðyÞ dSðyÞ; 8x 2 S
2 S
(71)

where ui and ti are the displacement and traction, respectively;

S ¼ [a¼0 n Sa is the total boundary of domain V0 (assuming that S Fig. 11 Matrix domain V0 and n inclusions

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-15

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 2 38 9 2 3
Matrix S0 A00 A01 B01 A02 B02  A0n B0n > u0 > B00
>
> >
>
6 7>
> > 6
> 7
Matrix S1 66 A10 A11 B11 A12 B12  A1n B1n 7>
> u1 >
7> >
>
6 B10 7
6 7
6 >
7> >
> 6 7
Inclusion S1 6 Af1 Bf1  0 7 >
> >
> 6 7
6 0 0 0 0 7>
>
>
t1 > >
> 6 0 7
6 > >
7> > 6 7
Matrix S2 6 B21 B22  B2n 7> > 6 7
6 A20 A21 A22 A2n 7< u2 = 6 B20 7
6 7 ¼ 6 7ft0 g (74)
Inclusion S2 6
6 0 0 0 Af2 Bf2  0 0 7 > t2 >
7> > 6 6 0 7
7
6 >
7> >
> 6 7
.. 6 .. .. .. .. .. .. .. >
.. 7> .. >> 6 .. 7
6 . >
> >
>
. 6 . . . . . . . 7>
> >
7> . >
> 6
6 . 7
7
6 7>
> > > 6 7
Matrix Sn 6 Bn1 Bn2  Bnn 7>
> u >
> 6 Bn0 7
4 An0 An1 An2 Ann 5>
>
>
n >
> 4 5
: > ;
Inclusion Sn 0 0 0 0 0  Afn Bfn tn 0

in which u0 and t0 are the displacement and traction vector on the More examples in modeling composites with the fast multipole
outer boundary S0 , ui and ti are the displacement and traction vec- BEM, with the total number of DOFs up to 28.8  106, can be
tor on the interface Si (viewed from the matrix domain), found in Refs. [44,175,246–250,252].
Aij and Bij are the coefficient submatrices from the matrix do-
main, and Afi and Bfi are the coefficient submatrices from inclu-
sion i. Studies of 2D composite models with periodic conditions 4.3 Modeling of Functionally Graded Materials. In gen-
can be found in Refs. [244,245]. eral, a functionally graded material (FGM) is a special type of
The fast multipole BEM and other fast solution methods can be composite in which the constituent volume fraction varies gradu-
applied to solve the BEM systems of equations shown in Eq. (74) ally leading to a non-uniform microstructure with continuously
for the inclusion problems. The multipole expansions and related graded macroproperties, e.g., thermal conductivity, density, and
translations for 2D and 3D elasticity BIEs can be found in Refs. specific heat. For instance, for problems governed by potential
[44,246–250]. Some other related work on the fast multipole theory, e.g., steady-state heat transfer, the thermal conductivity is
BEM for general 3D elasticity problems can be found in Ref. a function of position, i.e., k :k(x). Approaches to treat problems
[251], for 3D inclusion problems in Refs. [175,252], and for crack of potential theory in non-homogeneous media include the
problems in Refs. [176,253,254]. In the following, we present two Green’s function approach, domain integral evaluation, and vari-
examples using the BEM to show its potentials in modeling com- able transformation approach. In the context of the BEM, prob-
posite materials. lems of potential theory in non-homogeneous media have been
Figure 12 shows 2D BEM models of fiber-reinforced composites previously studied by Cheng [256], Ang et al. [257], Shaw and
used to extract the effective properties in the transverse directions. Makris [258], Shaw [259], Harrouni et al. [260], Divo and Kassab
The fibers are distributed randomly and with an decreasing density [261], and recently by Gray et al. [262] and Dumont et al. [165].
in one direction. Thus, the composites may be considered as func- Most of these works are focused on obtaining the Green’s func-
tionally graded materials. The plane-strain models are used to tion. In the Green’s function approach, the Green’s function has to
extract the moduli of the composites in the transverse directions. be derived and a boundary only formulation can be obtained.
Two cases are studies, one with fibers in a circular shape and the Several approaches have been developed to evaluate domain
other with fibers in a star shape. The computed effective Young’s integrals associated with boundary element formulations includ-
moduli for the two composite models are reported in Ref. [44], ing approximate particular solution methods, dual reciprocity
which agree with the estimates given in Ref. [255]. methods, and multiple reciprocity methods. The particular solu-
Figure 13 shows a 3D representative volume element (RVE) tion methods and dual reciprocity methods can be considered
model with 5832 fibers and 10,532,592 DOFs for a composite. The more or less to be equivalent in nature. These methods have been
volume fraction of the fibers is 3.85%, the fibers are distributed ran- widely used on the axiom that the domain integral in the boundary
domly. Theoretically, the moduli of the materials should be inde- integral formulation is somehow removed. In these methods the
pendent of the sizes of the RVE models. However, when the RVE inhomogeneous term of the governing differential equation is
sizes are small (with only a few fibers), there are significant approximated by a simple function such as (1 þ r) or radial basis
changes in the estimated moduli as we increase the model sizes, functions (RBFs). The mathematical properties and the conver-
which suggests that the models may not be representative of the gence rates of the RBF approximations have been studied exten-
composites [249]. The estimated moduli approach constant with the sively. In these techniques, the boundary-only nature of the BEM
increase of the fibers in the RVE as revealed in the results [249]. is compromised.
A transformation approach, called the “simple BEM,” for poten-
tial theory problems in non-homogeneous media where non-
homogeneous problems are transformed into known problems in
homogeneous media can be used [263,264]. The method leads to a
pure boundary-only formulation. This idea has been successfully
implemented in three dimensions for steady state, transient heat
conduction problems and crack problems where the material prop-
erty varies in one, two, and three dimensions. There is a class of
material variations, which can transform the problem equation to a
Laplace or standard/modified Helmholtz equation. The idea of the
“simple BEM” consists of transforming problems in non-
homogeneous media to known problems in homogeneous media
such that existing codes for homogeneous media can be reused
with simple modifications. By means of this approach, which con-
Fig. 12 2D models of fiber composites: (a) circular-shaped sists of simple changes in the boundary conditions of existing ho-
fibers, (b) star-shaped fibers (Ref. [44]) mogeneous heat conduction computer codes, the solutions for

031001-16 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 13 A 3D RVE with 5832 fibers and with the total DOFs 5 10,532,592 [249]

nonhomogeneous media with quadratic, exponential, and trigono- modified Helmholtz can be obtained, respectively. From these
metric material variations can be readily obtained. cases, a family of variations of k(x, y, z) can be generated, as given
For problems described by potential theory, the governing differ- in Refs. [263,264]. From an engineering point of view (for appli-
ential equation for a potential function / defined on a region X cations such as FGMs), material variation in one coordinate is of
bounded by a surface C, with an outward normal n, can be written as practical importance, which include thermal barrier coatings, bone
and dental implants, piezoelectric and thermoelectric devices,
r  ðkðx; y; zÞr/Þ ¼ 0 (75) graded cementitious materials, and optical application with graded
refractive indexes.
where k(x, y, z) is a position-dependent material function and the
Boundary conditions:
dot represents the inner product. Equation (75) is the field equa-
In order to solve the boundary value problem based on the
tion for a wide range of problems in physics and engineering such
modified variable v, the boundary conditions of the original prob-
as heat transfer, fluid flow motion, flow in porous media, electro-
lem have to be incorporated in the modified boundary value prob-
statics, and magnetostatics. The boundary conditions of the prob-
lem. Thus, for the modified problem, the Dirichlet and the
lem can be of the Dirichlet or Neumann type:
Neumann boundary conditions given by Eq. (76) change to the
following:
/ ¼ /; on C1
pffiffiffi
@/ (76) v ¼ k/ on C1
q ¼ kðx; y; zÞ ¼ q on C2 (81)
@n @v 1 @k q
¼ v  pffiffiffi on C2
@n 2k @n k
respectively, with C ¼ C1 þ C2 for a well-posed problem.
Notice that the Dirichlet boundary
pffiffiffi condition of the original prob-
Variable transformation approach: lem is affected by the factor k. Moreover, the Neumann bound-
By defining the variable ary condition of the original problem changes to a mixed
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi boundary condition (Robin boundary condition). This later modi-
vðx; y; zÞ ¼ kðx; y; zÞ /ðx; y; zÞ (77) fication is the only significant change on the transformed bound-
ary value problem.
one rewrites Eq. (75) as A numerical example of an FGM rotor with eight mounting

holes having an eight-fold symmetry is presented (Fig. 14). Owing
rk  rk r2 k to the symmetry, only one-eighth of the rotor is analyzed. The top
r2 v þ  v¼0 (78)
4k2 2k view of the rotor, the analysis region, and the geometry of the
region are illustrated in Fig. 14(a). The grading direction for the
or, alternatively rotor is parallel to its line of symmetry, which is taken as the z-
axis. The thermal conductivity for the rotor varies according to
r2 v þ k0 ðx; y; zÞ v ¼ 0 (79) the following expression:

where kðzÞ ¼ 20ð1 þ 420:7zÞ2 (82)

rk  rk r2 k The BEM mesh employs 1584 elements and 3492 nodes. A

k0 ðx; y; zÞ ¼  (80)
4k2 2k schematic for the thermal boundary conditions and the BEM mesh
employed is shown in Fig. 14(b). Here, the solution of the prob-
By setting k0 ðx; y; zÞ ¼ 0; þb2 ; and  b2 , three classical homoge- lem is verified using the commercial software ABAQUS. The
neous equations namely, the Laplace, standard Helmholtz and the FEM mesh consists of 7600 20–node (quadratic) brick elements

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-17

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 14 Analysis of an FGM rotor part using the BEM [263]

and 35,514 nodes. Figure 14(c) shows the comparison of the BEM medium containing a crack with the same geometry and boundary
and FEM results for the temperature around the hole. A contour conditions (generally traction-free) as in the problem to be solved.
plot of the temperature distribution using the BEM is shown in In this way, there is no need to mesh the crack, since the funda-
Fig. 14(d). All the results of the BEM and FEM solutions are in mental solution already accounts for it. The other approach
very good agreement. divides the domain into several regions via the definition of artifi-
cial interfaces that contain the cracks; thus it links the cracks to
4.4 Modeling of Fracture Mechanics Problems. Fracture the boundary of the resulting subdomains. Equilibrium and com-
mechanics is one of the areas where the BEM has been clearly patibility conditions are further enforced on the nodes of the
shown to be a powerful and effective numerical method when newly introduced interfaces to solve the problem.
compared to other computational techniques. Among its advan- These two approaches have limited applicability: ad-hoc funda-
tages one may cite that: (1) discretization of only the boundary is mental solutions are needed in the first case, while the artificial
required; thus simplifying preprocessing and remeshing in crack- boundaries required in the second approach are not uniquely-
growth analysis, (2) it shows improved accuracy in stress concen- defined and they also increase the number of unknowns. Therefore,
tration problems, since there are no approximations imposed on none of them is suitable for large-scale BEM models involving a
the stress solution at the interior domain points, (3) fracture pa- great number of cracks. Following the works of Hong and Chen
rameters (stress intensity factors (SIF), energy release rates, etc.) [270], Krishnasamy et al. [267–269], and Portela et al. [271], an
can be accurately determined from the computed nodal data in a alternative approach is further proposed to overcome these difficul-
straightforward manner, and (4) modeling of problems involving ties, that is, the dual or hypersingular BIE approach that results in
infinite and semi-infinite domains (like in wave propagation phe- a single region formulation by applying the displacement BIE on
nomena) is simple and accurate, because the radiation conditions one of the crack faces and the traction BIE (that follows from dif-
at infinity are automatically satisfied. ferentiation of the displacement BIE and substitution into the con-
However, the direct application of the displacement BIE to stitutive law) on the other.
cracks where the two surfaces coincide leads to a mathematically A review of the major contributions of the BEM in fracture
degenerate problem. This has been a sticking issue in the BEM for mechanics between 1972 and 1997 may be found in Ref. [272],
modeling fracture problems (see, e.g., Ref. [36]). Different with applications covering from linear elastic fracture mechanics
approaches have been adopted in the BEM to tackle fracture prob- to SIF computation, dynamics, composite materials and interface
lems, such as the displacement discontinuity approach [265,266] cracks, thermoelastic problems, crack identification techniques,
and the use of hypersingular BIEs [267–269]. and nonlinear problems.
Other alternative approaches also exist. The first approach is Later efforts have focused on applying the BEM to fracture of
based on the adoption of a Green’s function derived for an infinite multifield materials with coupled piezoelectric and/or elastomagnetic

031001-18 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 properties [273–278], crack growth analysis using the BEM alone or
together with FEM [279–281], or addressing fracture phenomena in
composites or at different scales [282–285].
However, when dealing with large-scale fracture problems, the
BEM demands great computer memory storage and significant
CPU time unless algorithms to accelerate the solution come into
play. The development of fast BEM approaches in the last 15 years
oriented towards applied mechanics problems has revived the
potential applications of the BEM to fracture mechanics. These
techniques include the fast multipole method [44,170,172,178],
panel clustering method [286], mosaic-skeleton approximation
[287] and the methods based on hierarchical matrices and adaptive
cross approximation [198,202]. All these techniques aim at reduc-
ing the computational complexity of the matrix–vector multiplica-
tion, which is the core operation in iterative solvers for linear
systems. FMM and panel clustering address the problem from an
analytical point of view and require the knowledge of some kernel
expansion prior to integration. Meanwhile, mosaic-skeleton approx-
imations and hierarchical matrices are based on purely algebraic
tools for the approximation of boundary element matrices which
are thus suitable for applications where analytical closed-form
Fig. 15 Arrays of 12 3 12 3 12 penny-shaped cracks in an elastic
expressions of the kernels are not available, as in anisotropic frac- domain solved by using the fast multipole BEM (total number of
ture applications. DOFs 5 1,285,632) [297]
Nishimura et al. [176] applied the original FMM by Rokhlin
[170] to solve crack problems for 3D Laplace’s equation. Yoshida
et al. [288] further improved the procedure by implementing ACA solver, which permitted the accurate analysis of crack
Greengard and Rokhlin’s diagonal form [289], leading to a CPU parameters.
time reduction of about 30% as compared to the original FMM Figure 15 shows an example of large-scale fracture analysis
when a dense array of cracks is distributed in the domain. In addi- using the fast multipole BEM. In this case, arrays of 12  12  12
tion, Yoshida et al. [253] developed a Galerkin formulation that penny-shaped cracks in an elastic domain were solved using the
took advantage of the improved accuracy of the Galerkin method. BEM. A total of 1,285,632 DOFs is used in this BEM model. The
Helsing [290] presented a 2D complex variable formulation crack opening displacements were computed and plotted in the
based on an integral equation of Fredholm second kind together figure [297].
with the FMM. The potential of the resulting approach is illus-
trated by accurately computing stress fields in a mechanically 4.5 Acoustic Wave Problems. The BEM has been applied to
loaded material containing 10,000 randomly oriented cracks. solve acoustic and other wave propagation problems for several
Wang et al. [291] improved the rate of convergence of fast multi- decades, due to its features of boundary discretizations and accu-
pole BEM by considering preconditioners based on sparse approx- rate handling of radiation conditions at the infinity for exterior
imate inverse type. Convergence within 10 steps is achieved for a problems. The developments of fast methodologies in the last dec-
2D elastic infinite body containing 16,384 cracks with 1,572,864 ade for solving large-scale acoustic problems are perhaps the most
DOF. Liu [247] developed a new FMM with the dual BEM for 2D important advances in the BEM that have made this method
multi-domain elasticity based on a complex variable representa- unmatched by other methods in modeling such problems.
tion of the kernels. The resulting formulation was more compact The fast multipole method developed by Rokhlin and Greengard
and efficient than previous FMM approaches, as shown in tests [170–172] has been extended to solving the Helmholtz equation
involving a large number of crack-like inclusions. governing the acoustic wave problems for quite some time (see
For the application of the ACA method in the modeling of papers in Refs. [183,184,298–309], a review in Ref. [178], and
crack problems, Kolk et al. [292] investigated 3D fatigue crack Chapter 6 of Ref. [44]). Large-scale acoustic BEM models with a
growth using a dual BEM. In order to accelerate the solution, the total number of DOFs around several millions have been attempted
resulting collocation matrix is compressed by a low-rank approxi- by the fast multipole BEM on desktop PCs. For example, acoustic
mation performed by the ACA method. The efficiency of the for- BEM models of entire airplanes [44,309,310], wind turbines [184],
mulation is illustrated with an industrial application in which a and a submarine [44,310] have been solved for noise predictions.
corner crack located at the dedendum of a gear propagates. This Coupled structural acoustic problems have also been solved suc-
approach was later revisited and improved by minimizing the cessfully using the fast multipole BEM with the FEM for analyzing
computation time of each incremental loop needed to simulate the ship structures [311], and, most interestingly, for computing physi-
crack growth process [293]. cally based and realistic sound in computer animations [312,313].
More recently, Benedetti et al. [294] developed a fast solver The adaptive cross approximation suited for acoustics has not
based on the use of hierarchical matrices for the representation of found so much attention like the FMM until recently. There are
the collocation matrix and implemented it for dual BEM analysis mathematical studies concerning ACA applied in acoustics which
of 3D anisotropic crack problems. The low-rank blocks were com- show that increasing the frequency decreases the efficiency [314].
puted by the ACA. Numerical tests demonstrated that the solver The counterpart in the FMM is that for higher frequencies, more
was rather insensitive to the degree of anisotropy of different terms in the series expansion have to be used. One application of
materials, which results in significant savings in memory storage ACA in acoustics in time domain can be found in [210]. An im-
and assembly and solution time. An order of magnitude improve- pressive example in determining the noise level in an aircraft can
ment has been observed for a system of order 104. Benedetti and be found in Ref. [315]. Preliminary comparison studies on the per-
Aliabadi [295] further extended the approach to the dynamic anal- formance of the fast multipole BEM and ACA BEM in modeling
ysis of 3D cracked isotropic bodies in the Laplace transform acoustic problems are provided in Refs. [315,316]. In the follow-
domain. Based on this approach, Benedetti et al. [296] developed ing, we first review the BIE formulation for solving acoustic wave
a model to analyze 3D damaged solids with adhesively bonded problems and then show a couple of applications.
piezoelectric patches used as strain sensors. The cracked structure We consider the linear time-harmonic acoustic wave problems in
was modeled using a fast dual BEM with a hierarchical domain E, which can be an interior domain, or an exterior domain

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-19

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 ð
outside a structure with boundary S. The acoustic pressure / is gov-
ceðxÞqðxÞ ¼ ½Kðx; y; xÞqðyÞ  Hðx; y; xÞ/ðyÞdSðyÞ þ qI ðxÞ
erned by the Helmholtz equation and boundary conditions given on S
S, such as pressure or velocity conditions. For exterior (infinite do- (87)
main) acoustic wave problems; in addition to the boundary condi-
tions on S, the field (radiated or scattered wave) at infinity must where ceðxÞ ¼ 1=2 if S is smooth around x. For 3D problems, the
also satisfy the Sommerfeld radiation condition. two new kernels are given by
Boundary integral equations have been applied in solving
acoustic wave problems for decades [317–320]. In the direct BIE @Gðx; y; xÞ 1
formulation, the representation integral of the solution / is (for Kðx; y; xÞ ¼ ¼ ½ikrðx; yÞ  1r;j nj ðxÞeikr
the derivation, see Refs. [320–322] and Chapter 6 of Ref. [44]) @nðxÞ 4pr 2
ð (88)
/ðxÞ ¼ ½Gðx; y; xÞqðyÞ  Fðx; y; xÞ/ðyÞdSðyÞ þ /I ðxÞ (83) @Fðx; y; xÞ 1 
Hðx; y; xÞ ¼ ¼ ð1  ikr Þnj ðyÞ
S @nðxÞ 4pr 3
2 2  
where x 2 E, q ¼ @/=@n, /I ðxÞ is a possible incident wave for an þ k r  3ð1  ikr Þ r;j r;l nl ðyÞ nj ðxÞeikr (89)
exterior problem, x the circular frequency, and the two kernels
for 3D are given by In HBIE (Eq. (87)), the integral with the kernel K is a strongly
singular integral, while the one with the H kernel is a hypersingular
1 ikr (HFP) integral. HBIE (Eq. (87)) can also be written in a regularized
Gðx; y; xÞ ¼ e (84) or weakly singular form [321,322]. For exterior acoustic wave
4pr
@Gðx; y; xÞ 1 problems, a dual BIE (or composite BIE [321]) formulation using a
Fðx; y; xÞ ¼ ¼ ðikr  1Þr;j nj ðyÞeikr (85) linear combination of CBIE (Eq. (86)) and HBIE (Eq. (87)) can be
@nðyÞ 4pr 2 written as

where k is the wavenumber. Let the source point x approach the CBIE þ b HBIE ¼ 0 (90)
boundary. We obtain the following conventional boundary inte-
gral equation (CBIE) [320–322]: where b is the coupling constant. This formulation is called
ð Burton-Miller formulation [318] for acoustic wave problems and
cðxÞ/ðxÞ ¼ ½Gðx; y; xÞqðyÞ  Fðx; y; xÞ/ðyÞdSðyÞ þ /I ðxÞ has been shown to yield unique solutions at all frequencies, if b is
S a complex number
pffiffiffiffiffiffi
ffi (which, for example, can be chosen as b ¼ i/k,
(86) with i ¼ 1 [319]).
As the first example, a submarine model (Fig. 16) is presented
where cðxÞ ¼ 1/2, if S is smooth around x. This CBIE can be which is solved by use of the fast multipole BEM (FastBEM
employed to solve for the unknown / and q on S. The integral AcousticsV R , V.2.2.0). This is an interesting example in solving

with the G kernel is a weakly singular integral, while the one with large-scale underwater acoustic problems, which has been a chal-
the F kernel is a strongly singular (CPV) integral, as in the poten- lenging task for other domain-based methods. The Skipjack sub-
tial case. A regularized or weakly-singular form of CBIE (Eq. marine is modeled, which has a length of 76.8 m and a total of
(86)) exists and can be derived by use of the identities for the 250,220 boundary elements are used in the discretization, with a
static kernels [44,46,47,323]. typical element size equal to 0.14 m. Velocity BCs are applied to
It is well known that the CBIE has a major defect for exterior the propeller and an incident plane wave in the direction (1, 0, 1)
problems, that is, it has non-unique solutions at a set of fictitious is specified. The model is solved at a non-dimensional wavenum-
eigenfrequencies associated with the resonate frequencies of the ber ka ¼ 384 (frequency f ¼ 1233 Hz). The computed sound pres-
corresponding interior problems [318]. This difficulty is referred sure on the surface of the model is shown in Fig. 16. The BEM
to as the fictitious eigenfrequency difficulty. A remedy to this model was solved in 85 min on a DellV R PC with IntelVR Duo Core

problem is to use the normal derivative BIE in conjunction with CPU and with the tolerance for convergence set at104.
the CBIE. Taking the derivative of integral representation (Eq. Next, we show an example using the ACA BEM for solving
(83)) with respect to the normal at a point x on S and letting x large-scale 3D acoustic problems. Figure 17 shows a BEM model
approach S, we obtain the following hypersingular boundary inte- of five wind turbines discretized with 557,470 boundary elements.
gral equation (HBIE): The wind turbines have the same height of 29 m and are spaced

Fig. 16 A BEM model of the Skipjack submarine impinged upon by an incident wave

031001-20 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 17 Sound field from five wind turbines evaluated using the ACA and fast multipole
BEM

50 m apart in the x- and y-directions. The ground is modeled as a domain. In frequency domain; for example, one notes that the fun-
rigid infinite plane where no elements are applied because of the damental solution Cij for elastodynamics is written as
use of the half-space Green’s function. The sound pressure on the
surfaces of the turbines and a field surface of the size 200  200  
1 @ @ @ @
m2 on the ground are computed at the non-dimensional wavenum- Cij ðx; yÞ ¼ G k ðx; yÞ þ e ipr e jqr G k ðx; yÞ
ber ka ¼ 5 (representing a rotating speed of 3 cycles/s for the kT2 l @xi @yi L @xp @yq T
blades). This large BEM model (with more than 500,000 (91)
complex-valued DOFs) was solved in 210 min using the ACA
BEM on a Dell PC with an IntelV R Duo Core CPU. For compari-
where Gk and kL;T are the fundamental solution of Helmholtz
son, the same large-scale acoustic model was solved on the same
equation and the wave numbers of P and S waves given by
PC in 108 min using the fast multipole BEM [184]. At present,
the fast multipole BEM is faster than the ACA BEM for models rffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffi
with more than about 100,000 DOFs and the fast multipole BEM 1 q q
Gk ðx; yÞ ¼ eikjxyj ; kL ¼ x; kT ¼ x
also uses smaller memory for storage. However, the ACA BEM 4pjx  yj k þ 2l l
has the kernel-independent advantage and is easier to implement.
and q, ðk; lÞ and x are the density, Lamé’s constants, and fre-
4.6 Elastic Wave Problems. Boundary element methods in quency, respectively. This shows that one may formulate fast
elastodynamics have a long history. Actually, they date back to methods for elastodynamics simply by differentiating those for
1930s or 1940s where Kupradze [324] started his theoretical work the Helmholtz equation.
on singular integral equations for elastodynamics. Kupradze’s The first fast method used with the BEM in elastodynamics
[324] contributions are not restricted to theoretical works. Indeed, seems to be the fast multipole method. Indeed, Chew’s group
his numerical method known as Kupradze’s functional equation [326] presented a 2D FMM formulation extending the high-
method (collocate direct BIEs in the exterior of the domain under frequency FMM (diagonal form) in Helmholtz proposed by Rokh-
consideration) is not just well-known, but also has been rediscov- lin [298]. They also showed an example of the scattering from a
ered by many authors! rough interface between two elastic materials. Fukui [327] pro-
It is noteworthy that Cruse and Rizzo, who are usually considered posed a low-frequency FMM in 2D with which they solved scat-
to be the founders of BEM (called boundary integral equation meth- tering problems for many holes. Fujiwara [328] also proposed a
ods by these authors), investigated the use of the BEM in elastody- low-frequency FMM in 2D and solved scattering problems for
namics [6] soon after their first work in elastostatics. It is probable many cavities and cracks.
that these pioneers were aware of the potential of the BEM in elasto- In 3D, Fujiwara [329] presented a diagonal form FMM, but he
dynamics since it can deal with exterior problems easily. Indeed, the solved low-frequency problems related to earthquakes. Yoshida
BEM in wave problems gives highly accurate results which are not et al. [330,331] considered low-frequency crack problems in 3D,
polluted by reflections from artificial boundaries, while many other which they solved with a low-frequency FMM utilizing a decompo-
numerical methods require special techniques to avoid such reflec- sition of the kernel in Eq. (91). Yoshida et al. [332] also proposed a
tions. Subsequent developments of the BEM in elastodynamics diagonal form version of the elastodynamic crack analysis.
were quite rich with many applications in areas such as non- Unfortunately, their error control scheme was immature. As a mat-
destructive testing or evaluation, earthquake engineering and others, ter of fact, the use of the diagonal form is effective only when the
as one can see in the review article by Kobayashi [325]. wavenumber times the cell (box) size used for FMM is above a cer-
As in other branches of the BEM, however, the O(N2) complexity tain threshold. In problems where the smaller cells do not satisfy
of the algorithm turned out to be a serious problem. With this prob- this criterion, one has to switch to the low-frequency FMM in order
lem, ordinary BEMs were found to be inefficient in solving large to maintain the accuracy. Namely, one may have to use both diago-
problems in spite of the boundary-only nature of the approaches. nal forms and low-frequency FMM in one problem. This type of
This problem, however, is now almost solved with the develop- FMM formulation is called a wideband FMM [309,332].
ment of the so called fast BEMs. Indeed, the fast methodologies Subsequent developments in elastodynamic FMMs were rather
are undoubtedly among the most important developments in slow until 2007 where Chaillat et al. [334,335] reconsidered diag-
BEMs in elastodynamics in the last decade, which will be the sub- onal forms in elastodynamics. Sanz et al. [336] utilized several
ject of the following subsection. This will be followed by another techniques including SPAI preconditioners. Tong and Chew [337]
subsection devoted to stability issues and inverse problems. applied techniques developed for Maxwell’s equations to the S
4.6.1 Fast BEMs in Elastodynamics wave part in their diagonal form formulation. Chaillat et al. [338]
considered half-space and multi-domain problems for seismologi-
Frequency domain. Fast methods in elastodynamics are usually cal applications. Isakari et al. [339] presented a low-frequency
obtained from certain fast methods for Helmholtz equation in fre- FMM for periodic inclusion problems in elastodynamics which
quency domain or from those for the scalar wave equation in time may be used in applications related to phononic crystals. These

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-21

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 [345]. See the section on time domain approaches for more
references.
Inverse problems. Applications to problems related to ultra-
sonic non-destructive evaluation have been among major motiva-
tions of developing BEMs for elastodynamics. The past decade
has seen many interesting applications of BEMs to inverse prob-
lems of determining defects in structures, of which we cite just a
few in the rest of this section.
Solution techniques of inverse problems of determining
unknown defects in an elastic structure by BEMs can be classified
into two categories, namely, those which require successive solu-
tions of direct problems and those which do not. Methods of the
first category typically determine the unknown defects by mini-
mizing cost functions which represent the difference between the
experimental data and the computed value of certain physical
quantities. Approaches of this type are considered suitable for
determining the detail of the defects, although they may not nec-
essarily be effective for obtaining global information such as the
Fig. 18 BEM model of 1152 spherical cavities in an infinite
elastic solid [343]. The spatial DOF is 1,105,290 and the number
location or the number of defects. For methods of this category,
of time steps is 200. The CPU time is 10 h 47 min and the one needs effective ways of evaluating the sensitivities of the cost
memory requirement is 152.8 GB. functions with respect to parameters which characterize the
defects. Adjoint variable (field) methods are known to be effective
for such a purpose particularly when one has many parameters to
authors also considered pre-conditioners based on Calderon’s for- identify the defects. For example, Bonnet et al. [351] considered
mulae [340]. shape sensitivities for cavities and cracks using the time-domain
FMMs are not the only acceleration methodology useful in elasto- BEM. Bonnet and Guzina [352] discussed similar problems for
dynamic BEMs. Indeed, one may mention the use of pre-corrected inclusions in the frequency domain.
FFT approach by Yan et al. [197], and the H-matrix (ACA) As examples of inverse techniques of the second category, we
approach in crack problems by Benedetti and Aliabadi [295]. mention topological derivatives and linear sampling methods.
4.6.1.2 Time domain. As we have seen, papers on fast meth- Topological derivatives are essentially the rates of changes of a
ods in elastodynamics in frequency domain are scarce in spite of certain cost function when an infinitesimal cavity is produced at a
the importance of the subject in applications. Still more scarce are given position in a structure. This concept turned out to be useful
the time-domain counterparts. Takahashi et al. [341] extended the as a “void indicator” which can be used in order to obtain a rough
so called PWTD (plane wave time domain) approach developed idea about the defects, as we can see in Bonnet and Guzina [353].
by Ergin et al. [342] for the wave equation to elastodynamics in In the near-field linear sampling method discussed in Fata and
3D. Roughly speaking, this approach is the time-domain counter- Guzina [354], these authors consider an elastic half space with
part of the diagonal forms in the frequency domain with the Han- unknown defects. To determine the defects, they excite a part of
kel functions in the translation operators replaced by Bessel the plane boundary and measure the response on another part
functions. This approach has been parallelized and extended to of the plane boundary. They can delineate the defects by examin-
the anisotropic case by Otani et al. [343]. Figure 18 shows a typi- ing the blow-up of solutions of certain integral equations whose
cal example from this investigation, i.e., the response of 1152 kernel functions are determined from the experimental data. An
spherical cavities in an infinite isotropic elastic solid subject to a adjoint of this formulation is easier to use when one has only a
plane P wave. Yoshikawa et al. [344] applied this approach to a small number of excitations (sources), as has been discussed in
simple inverse problem of crack determination. Fata and Guzina [355].
Another fast approach in time domain has been proposed by
Fukui’s group [345], in which these authors accelerated a convo-
lution quadrature formulation in 2D elastodynamics with a low- 4.7 Electromagnetic Wave and Periodic Problems. Elec-
frequency FMM. tromagnetics (EM) is undoubtedly one of academic/industrial
areas where the BEM is used widely. This is partly because there
are many good applications where the BEM is quite effective. For
4.6.2 Other Developments in Elastodynamic BEM example, propagation of electromagnetic waves often takes place
in an open air, which is a loss-less homogeneous exterior domain.
Stability issues. Stability in the time domain approach still The BEM is considered to be one of ideal tools for solving such
seems to be an issue, as can be seen from the continuing publica- problems and Maxwell’s equations describe the phenomenon per-
tions on this subject. See, for example, the work by Soares and fectly. It is therefore not surprising that the BEM and related
Mansur [346] and a very recent paper by Panagiotopoulos and methods are accepted enthusiastically in the EM community.
Manolis [347]. The former approach works as a filtering to the Indeed, BEM software for EM applications including those accel-
recent-in-time convolution operation and is related to similar efforts erated with FMM are now available commercially or even freely.
in electromagnetics (e.g., Rynne and Smith [348]). We also note There are many techniques common to the BEM in EM and in
that mathematical proofs of the stability of some variational mechanics, and the mechanical community has a lot to learn from
approaches are available (e.g., the work by Ha-Duong et al. [349]), developments in EM community. It is therefore deemed worth the
although collocation methods are of more interests to engineers. efforts for those from mechanics to have some knowledge on the
Convolution quadrature method (CQM) is a new approach for BEM for Maxwell’s equations. The first part of this section is thus
the time-domain BEM which improves the stability of the method. intended as a quick introduction to the BEM for Maxwell’s equa-
It also extends the applicability of the BEM in the time domain tions for those who are familiar with the BEM in mechanics. The
because CQM needs only the Laplace transforms of the funda- interested reader is referred to Chew et al. [356] for further details.
mental solutions with respect to time, which may be easier to We then discuss more recent topics on the periodic FMM for
obtain than their time-domain counterparts. We here mention Maxwell’s equations, which have many interesting applications
Kielhorn and Schanz [350] in addition to Fukui’s contribution such as photonic crystals and metamaterials. Finally we review

031001-22 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 periodic boundary value problems in mechanics and the BEM to BIEs. Boundary value problems for Maxwell’s equations are
solve them. solved as one solves either EFIE or MFIE for unknown boundary
densities j and/or m. These density functions have two independ-
4.7.1 BEM and MoM for Maxwell’s Equations. We consider ent components in 3D problems since they are tangential to the
an exterior domain D 2 R3 in which Maxwell’s equations are boundary, while their mechanical counterparts have three inde-
satisfied: pendent components. A typical boundary condition is m ¼ 0 on S
for perfect electric conducting scatterers. As in mechanics, solu-
r  E ¼ ixlH; r  H ¼ ixeE; in D tions of EFIE or MFIE may not be unique for particular values of
x called irregular frequencies. In that case, one may consider a
where E and H are the electric and magnetic fields, x is the fre- linear combination of EFIE and MFIE called CFIE to obtain
quency (with eixt time dependence), and e and l are the dielec- uniquely solvable integral equations as in Burton and Miller’s
tric constant and the magnetic permeability for the material method [318] in acoustics. Numerical approaches based on these
occupying D, respectively. We may combine these equations to integral equations and collocation method are called the BEM in
have the EM community.
One may alternatively solve these integral equations with the
pffiffiffiffiffi Galerkin method. In the special case of perfect electric conducting
r  ðr  EÞ ¼ k2 E; k¼ elx (92)
scatterers, one has
In addition, we have an incident wave denoted by (Einc , H inc ) in D ð ð ð 
and scattered waves Esca ¼ E  Einc and H sca ¼ H  H inc which j0 ðxÞ  Einc ðxÞdSðxÞ ¼ ixlj0 ðxÞ  Gðx; yÞjðyÞ
satisfy radiation conditions. S S S
The fundamental solutions Cip for Maxwell’s equations in Eq. 
(92) have the following form: i
 divS j0 ðxÞGðx; yÞdivS jðyÞ dSðyÞdSðxÞ
xe

1 @ @ eikjxyj
Cip ¼ þ dip G; Gðx; yÞ ¼ (97)
k2 @xi @xp 4pjx  yj

Note that Cij has stronger singularity than its counterparts in where j0 is the basis function used for the discretization for j.
mechanics. Also noteworthy is the fact that the curl of Cij is less Galerkin approaches based on Eq. (97) are usually called method
singular than Cij and that the double curl of Cij is proportional to of moments (MoM) in the EM community. We note that the basis
Cij modulo Dirac delta. We have the following integral represen- functions for j have to satisfy the continuity of normal compo-
tations for E and H: nents across the edges of elements, or “divergence conforming,”
for the use of the integration by parts to be justifiable. RWG func-
ð  
tions [357] are among the most popular choices of divergence
  i
EðxÞ ¼ Einc ðxÞ þ m  ry G  ixljG þ ry G divS j dSðyÞ; conforming basis functions used in MoM. These functions are
S xe
vector valued and have degrees of freedom on edges of elements.
x2D (93) Use of RWG functions for both of j and m, however, may not be a
ð   good choice for problems with dielectric scatterers since it may
  i
HðxÞ ¼ H inc ðxÞþ j ry G ixemGþ ry G divS m dSðyÞ; lead to ill-conditioned systems. Use of dual [358] (or Buffa-
S xl Christiansen [359]) basis function is a possible solution to this
x2D (94) problem.
4.7.2 Fast Methods for Maxwell’s Equations. Fast methods
In these formulae, j and m are the surface electric and magnetic for equations of the Helmholtz type seem to be investigated most
current vectors defined by vigorously in the EM community because of many important
applications such as radars, etc. Examples of very large problems
jðyÞ ¼ nðyÞ  HðyÞ; mðyÞ ¼ EðyÞ  nðyÞ
solved with fast solvers of integral equations are found in papers
such as Taboada et al. [360] (150  106 unknowns), Ergül and
where n stands for the outward unit normal vector on the surface Gürel [361] (more than 200  106 unknowns), etc. Not surpris-
of the domain D, and divS indicates the surface divergence. From ingly, there are overwhelming amount of publications devoted to
Eqs. (93) and (94), we have the subject of fast integral equation solvers which make it impos-
sible to cover even just the main developments in a review article
mðxÞ=2 ¼ minc ðxÞ  nðxÞ of this size. The reader may find it more informative to refer to a
ð  
  i book [356] or to introductory or review articles by specialists
 m  ry G  ixljG þ ry G divS j dSðyÞ; x 2 S [362–365].
S xe
(95) 4.7.3 Periodic Problems in Maxwell’s Equations. As a recent
inc
progress in computational electromagnetics, we can mention peri-
jðxÞ=2 ¼ j ðxÞþnðxÞ odic FMM. Periodic boundary value problems for Maxwell’s equa-
ð  
  i tions are increasingly important these days because there are new
 j ry G ixemGþ ry G divS m dSðyÞ; x 2 S micro and nano devices which utilize periodic structures, such as
S xl
photonic crystals and metamaterials. Photonic crystals are a class of
(96) periodic structures of dielectric materials having a periodicity com-
parable to the wavelength of light [366]. Photonic crystals are
where minc and jinc are defined in an obvious manner. Equations known for their peculiar properties such as frequency selectivity,
(95) and (96) are called EFIE (electric field integral equation) and etc., and are said to guide and store the light quite freely. Metama-
MFIE (magnetic field integral equation). Conceptually, EFIE terials are a class of composite materials with subwavelength peri-
(MFIE) corresponds to the displacement (traction) BIE in odic metallic structures. They exhibit curious behaviors such as
mechanics or vice versa. Note, however, that EFIE and MFIE negative apparent refractive indices for a certain frequency range.
have essentially the same structure, which is in contrast to BIEs in Applications of metamaterials are being sought in optical fibers, in
mechanics in which hypersingularity appears only in traction the so called “super lens” [367] with extremely high resolution, etc.

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-23

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 BEMs for periodic boundary value problems are obtained easily in Ref. [372] for doubly periodic problems and in Ref. [373] for
as one replaces fundamental solutions by periodic Green’s func- singly periodic problems. They also proposed a novel approach to
tions in the BIEs. Periodic Green’s functions for wave problems compute lattice sums accurately [372].
usually take the form of lattice sums of fundamental solutions.
This means that a solution of a periodic boundary value problem 4.7.4 Periodic Problems in Mechanics. Periodic problems
can be viewed as an infinite repetition of the solution in the unit are found in mechanics also. Actually, the seminal paper by
cell. This structure of the solution matches well with the tree Greengard and Rokhlin [171] already includes discussions on per-
structure of FMM. Indeed, one may identify the unit cell with the iodic boundary conditions in many-body Coulomb systems. Also,
level 0 box of FMM to formulate a periodic version of FMM. The periodic boundary value problems appear in microscopic analyses
effects from far replicas of the unit cell are then computed as an in the homogenization theory which is used in estimating macro-
infinite sum of M2L formulae between the unit cell and its repli- scopic moduli of composite materials. Houzaki et al. [374] and
cas. An attempt of this type for 3D doubly periodic problems in Otani and Nishimura [375] considered periodic boundary value
Maxwell’s equations is found in Otani and Nishimura [368] where problems for rigid inclusion problems in 2D antiplane problems
they considered the case of a cubic unit cell. This approach has and 3D problems of elastostatics. We note that static periodic
been extended to the case of a parallelepiped unit cell in [369] and problems are delicate mathematically because the periodic
to a tall unit cell case in [370]. Figure 19 shows an example from Green’s functions do not exist. However, their numerical treat-
Ref. [370] where layers of glass fibers in water are considered. ment is easier than the corresponding dynamic problems once the
This is supposed to be a model of the skin of worms, which is mathematical problem is resolved.
known to look blue if viewed from a certain angle: an example of In regard to dynamic problems, we mention Otani and Nishi-
the so-called structural color. mura [376] for singly periodic crack problems in 2D for Helm-
Figure 19(b) shows the energy transmittance when the wave holtz equation and Isakari et al. [339,377] for elastodynamic
length of the incident wave is 475 nm (blue). This result shows scattering problems in 3D by doubly periodic inclusions.
that one will observe a strong reflection of blue light when the
incident angle is 30, which is in agreement with experiments. Per-
iodic problems are characterized by anomalous behaviors of solu- 4.8 Time-Domain Problems. The BEM in the time domain
tions related to the existence of guided modes. The performance is based on the representation formula of the underlying time-
of the periodic FMM for Maxwell’s equations near anomalies are dependent partial differential equation. A BIE in time domain can
examined in Ref. [371]. be deduced either with partial integration or based on reciprocal
It is also possible to formulate FMMs for periodic boundary theorems. In the case of acoustic problems, the acoustic pressure
value problems without using periodic Green’s functions or lattice /ðx; tÞ can be written as
sums. Barnett and Greengard [372] presented well-conditioned in- ðt ð
tegral equations with layer potentials distributed not just on the cðxÞ/ðx; tÞ ¼ ½Gðx; y; t  sÞ qðy; sÞ  Fðx; y; t  sÞ /ðy; sÞ
surface of the scatterers but on the boundary of the unit cell as 0 S
well. Their efforts of this type in 2D Helmholtz equation are found  dSðyÞds 8ðx; tÞ 2 S  ½0; T  (98)

Different to the integral equation in Eq. (86) is the convolution in-

tegral in time and the fundamental solutions are those in Eqs. (84)
and (85) transformed to time domain. Fortunately, for most physi-
cal problems, the singular behavior of the time-dependent funda-
mental solutions with respect to space is the same as in the
respective static counterparts. A second hypersingular BIE as
counterpart to Eq. (87) can be as well deduced by applying the
gradient operator on the first integral equation. Details from a
mathematical point of view can be found in [378], where an engi-
neering representation may be found in [41,379].
To obtain a boundary element formulation, the domain as well
as the time (T ¼ MDt) is discretized and the shape functions in
space and time have to be chosen. For the acoustic example this is
XX p XX q
/ðy; sÞ ¼ ui ðyÞhpj ðsÞ/ji ; qðy; sÞ ¼ ui ðyÞhqj ðsÞqji
j i j i

(99)

with the nodal values in space and time /ji and qji . Note, the sum-
mation over i is the spatial counter, whereas the summation over j
is the counter in time. This approach is different from time-
dependent (standard) finite element approaches where the time
discretization is applied on the PDE. Such approaches applied in
BEM would result in domain integrals. The last step is the integra-
tion in space and time which yields the following matrix formula:

X
m1
C0 km ¼ bm þ ½Gmk qk  Fmk /k  m ¼ 0; 1; …; M (100)
k¼0

In Eq. (100), km and bm have the same meaning as in Eq. (13),

however, with changing values in each time step. The convolution
integral produces the summation over the past time steps. Hence,
Fig. 19 Skin of worm: (a) model; (b) energy transmittance a standard implementation has a complexity of O(N2M) for M

031001-24 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 time steps and N DOFs. An efficient implementation to reduce
this effort can be found in Ref. [380].
An alternative approach to the above sketched procedure is the
transformation to Laplace or Fourier domain and to solve the
elliptic problem. This avoids the high storage but requires a nu-
merical inverse transformation with all its difficulties (see, e.g.,
Refs. [381,382]). Furthermore, at each frequency the equation sys-
tem must be solved in contrast to Eq. (100) where only one equa-
tion solution is necessary. An overview on dynamic BEM
formulation until the year 1996 can be found in Refs. [383,384].

4.8.1 Analytical Integration in Time. The integration in time

is mostly performed analytically. This is due to the structure of
several time domain fundamental solutions used up to now. In
acoustics the fundamental solution is a Dirac distribution in time
and, hence, the time integral is reduced to the function at the re-
tarded time t  r=c. In elastodynamics, the integration can as well
be done analytically, but in this case first the shape functions (Eq.
(99)) have to be chosen (mostly linear for the displacements and
constant for the tractions) and then the integration is performed
within each time step. This procedure has been published initially
by Mansur [385]. In three dimensions, due to Huygens principle,
there exists a cut-off time at which the system matrices in Eq.
(100) becomes zero. However, this is not true in two dimensions
and for any problem with damping.

4.8.2 Convolution Quadrature Method. An alternative meth-

odology which also extends the time domain formulations to prob-
lems where the time domain fundamental solution is not known is
the so-called convolution quadrature method (CQM)-based BEM.
In this formulation, the convolution integral is approximated
numerically by the formula
ðt ð
Gðx; y; t  sÞ qj ðsÞ upj ðyÞ dSðyÞds
0 suppðupj Þ
X
n
 xnk ^ DtÞ qj ðkDtÞ
ðG; with
j
k¼0
ð
^ DtÞ ¼ Rn X
L1
^ y; s‘ Þ fn‘ up ðyÞ dSðyÞ
xnj ðG; Gðx; j
L ‘¼0 suppðupj Þ

cðf‘ RÞ 2pi
s‘ ¼ ; f¼eL (101)
Dt
Fig. 20 Sound pressure Level (dB) at the boundary of the
The CQM has been developed by Lubich [386] and is applied to amphitheatre
elastodynamics in Ref. [387].
5
In Eq. (101), R is a parameter to be
chosen (often R ¼ 102M ) and c(z) denotes an A- and L-stable level, it is observed that there are high negative values. Recalling
multistep method. The second-order backward differential for- the definition of the sound pressure level, those values denote
mula is a good choice [388] and A- and L-stable Runge-Kutta sound below the auditory threshold. At all areas which cannot be
methods can be used [389]. The essential advantage of this for- reached by the sound wave at the observation time such low val-
mula is that both the time dependent fundamental solution and ues are obvious. Also, after some time nothing more can be heard.
^ y; sÞ can be used. For many time-
its Laplace transform Gðx;
dependent problems, especially if damping is included, only the
transformed fundamental solutions are available. The applications 4.8.3 Numerical Instabilities and Weak Formulations. The
to visco- and poroelasticity can be found in Refs. [390,391]. classical space-collocation approach based on Eq. (98) suffers, in
Beside this advantage, the stability of the algorithm is improved some cases, from the numerical instability of the time-marching
compared to the analytical integration. Finally, the same recursion scheme which has been put into evidence, e.g., in Refs.
formula (Eq. (100)) is obtained where the matrix entries are now [392–395]. This instability is often called intermittent owing to
the integration weights xnj which present the time-integrated fun- the often unpredictable way in which it appears upon variation of
damental solutions. the time step adopted in the analysis.
As an example, the computed sound pressure level on the It has been shown that weak formulations can be very beneficial
boundary of an amphitheatre using the time-domain BEM are with respect to this issue. A first proposal in the time domain was
depicted in Fig. 20 at the times t ¼ 0.00625 s, t ¼ 0.046875 s, and put forward and implemented by Ha Duong [396] for acoustics; it
t ¼ 0.078125 s. The amphitheatre is discretized with 3474 linear couples the weak version - both in time and space - of the integral
triangles with 1761 nodes and a time step size of Dt ¼ 0.003125 s representation Eq. (98) with the weak gradient BIE which is
is used. Sound reflecting boundary conditions are considered and obtained from Eq. (98) by applying the gradient operator with
the sound propagates in the air with a speed of c ¼ 346m=s. The respect to x. Recently, Aimi et al. [397,398] have proposed an al-
sound field is excited by a prescribed sound flux of q ¼ 1 N=m2 in ternative direct space-time Galerkin BEM which leads to uncondi-
the middle of the stage. Looking on the lowest sound pressure tionally stable schemes. Instead of exploiting the gradient BIE,

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-25

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 this latter formulation makes use of the velocity BIE obtained by displacement continuity is enforced in strong form, the BEs at the
applying to Eq. (98) the time derivative with respect to t. interface and the sides of the FEs adjacent to the interface should
coincide, together with their nodes. On the other hand, if the kine-
4.8.4 Fast Methodologies. One of the first approaches for a matic continuity condition is enforced in weak form, the matching
fast evaluation of the retarded potentials related to scattering prob- of the two discretizations at the interface is no longer necessary,
lems was published in Ergin et al. [342]. The main idea is a plane with an evident flexibility gained in the meshing process. This
wave expansion of the respective integral kernel. This idea has permits the integration of independently discretized models possi-
been extended to elastodynamics in Refs. [341,343]. Another bly coming from different sources.
approach has been proposed by Hackbusch [399] for acoustics A non-matching node coupling of SGBEM and FEM is devel-
using the panel clustering technique to sparsify each matrix per oped by Ganguly, Layton and Balakrishna [413] in the context of
time step. An improvement of this approach can be found in Ref. a BE-based approach. In fracture mechanics, a coupling approach
[400]. For the heat equation, an approach based on a Chebyshev which preserves the independence of the FE and BE meshes is the
interpolation of the kernel has been proposed by Tausch [401]. An so-called hybrid surface-integral-finite-element technique, see the
FMM approach in viscoelasticity has been proposed in Zhu et al. works by Keat, Annigeri and Cleary [414], Han and Atluri [415]
[402]. Beside the above-mentioned approaches, the reformulation and Forth and Staroselsky [280]. Springhetti et al. [416] address a
of the CQM by Banjai and Sauter [403] allows the application of SGBEM-FEM coupling procedure in elastostatics characterized
fast techniques developed for elliptic problems. In Ref. [404] this by a weak enforcement of the both matching conditions at the
reformulated CQM is discussed as well for collocation interface.
approaches. In combination with ACA, a fast elastodynamic BEM Two different families of approaches seem to guarantee, at
has been proposed by Messner and Schanz [210]. The CQM in best, the complete independence of the various subdomains: itera-
combination with an FMM has been studied by Saitoh et al. [405]. tive (or alternating) techniques and Lagrange multiplier techni-
Finally, we mention related but different use of fast BEM in ques. The former type of approach could be in principle applied to
time-domain problems in fluid mechanics in which the governing any kind of coupling. The individual subdomains are treated inde-
equations are integro-differential equations. Solution methods for pendently by either method based on an initial guess of the inter-
such equations require evaluations of integrals, which are acceler- face unknowns. Then, the newly computed interface variables are
ated with fast methods. Examples of approaches of this type are synchronized and exploiting these updated values in another sub-
found in vortex methods (e.g., Ref. [406]) and in vesicular flow domain solution yields enhanced results. The procedure is
problems (e.g., Ref. [407]), to mention just a few. repeated until convergence is achieved. A comprehensive over-
view of such methods is given by von Estorff and Hagen [417].
4.9 Coupling of the BEM with the FEM. Even if the finite Within the iteration procedure, a relaxation operator is applied to
element method has a dominant status in the field of computa- the interface boundary conditions in order to enable and speed up
tional methods in engineering, mostly because of its great flexibil- convergence. In this sense, the iterative coupling approaches are
ity and wide range of applicability, approaches based on integral better called interface relaxation FEM–BEM. Although the itera-
equations are clearly superior for certain classes of problems. In tive coupling is very attractive from the point view of software
fact, the FEM and the BEM have relative benefits and limitations design, the convergence commonly depends on the relaxation pa-
depending on the problems to be solved. Hence, there are many rameters which are rather empirical (see Refs. [418–421].
applications where a coupling between the two approaches is in In the Lagrange multiplier approach matrix entries are never
principle very attractive. For example, different parts of a struc- mixed from the different subdomains and the interface conditions
ture can be modeled independently by the FEM and the BEM in are posed in a weighted sense such that nonconforming interfaces
order to exploit the advantages of both techniques. can be handled. The largest family in this category, represented
by the FETI-BETI approaches [422,423], can be formulated as
4.9.1 Methodologies. The first investigations on this idea follows. At the subdomain level, Dirichlet-to-Neumann maps are
dated back to 1977 with the pioneering work of Zienkiewicz, realized by means of each discretization method. In elastostatics,
Kelly and Bettes [408] based on the collocation non-symmetric for instance, this amounts to generating a Steklov Poincaré opera-
BEM. A symmetric coupling of the BEM and FEM was proposed tor expressing surface tractions in terms of surface displacements
later by Costabel and Stephan [409] followed by the variationally t ¼ Su. The realization of this operator for FEM and BEM is dis-
based coupling procedure between the FEM and an indirect ver- cussed at length, e.g., in Refs. [422–424]. Then, tractions along
sion of the Galerkin BEM developed by Polizzotto and Zito [410]. the interface are treated as Lagrange multipliers k and the equilib-
In many papers, the BEM subdomains, see for instance the work rium and continuity conditions are enforced in a weak manner
by Haas and Kuhn [411] and by Ganguly, Layton and Balakrishna along the interfaces in terms of u and k. Interface displacements
[412], or the FEM subdomains, e.g., the work by Ganguly, Layton and Lagrange multipliers are the unknowns of the reduced prob-
and Balakrishna [413], are handled as equivalent finite or bound- lem which can be solved by means, e.g., of a conjugate gradient
ary macro-elements, respectively. In recent years, however, the method [425]. The overall approach can be interpreted as a pre-
idea of adapting one of the two techniques to the other in a sort of conditioned conjugate gradient solver [422].
master-slave technique with macroelements has been gradually As far as evolving problems are concerned, one additional rea-
abandoned for a more balanced approach. son for avoiding the solution of the global coupled system of
The need to maintain the different subdomains as independent equations is that the use of different time steps for the subdomains
as possible has become a crucial point for different reasons. First, becomes possible. This is an important advantage, especially in
this aims at preserving the structure of the system matrices which the BEM, where the range of applicable time step lengths, result-
are sparse positive definite for the FEM and full, possibly non- ing in a stable and accurate solution, is limited. In the staggered
symmetric and not definite for the BEM. Second, the computa- solution approach, the equations for each subdomain are solved
tional cost and memory requirements of the BEM part must be once at each time step. However, the stability and accuracy of
imperatively reduced by means of algorithms such as fast multi- both the BEM itself and the staggered coupling approach impose
pole methods, panel clustering and ACA; this entails that the sys- requirements on the choice of the time step durations, which may
tem matrix is actually never available. Third, different be contradictory. Therefore, it is desirable to use a coupling proce-
subdomains might require different discretizations (both in time dure that is stable and accurate for a wide range of time steps.
and space) resulting in non-conforming meshes and interpolations. This can be obtained by introducing corrective iterations into the
This last point is closely related to the enforcement of the match- staggered algorithm [421]. Within the iteration procedure, a relax-
ing conditions (e.g., continuity of displacements and tractions in ation operator may be applied to the interface boundary conditions
solid mechanics) at the interface between the subdomains. When in order to speed up convergence. An overview of a variety of

031001-26 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 such techniques involving FEM–BEM as well as BEM–BEM cou- around the membrane are approximated as incompressible Newto-
pling, is given by Refs. [417–419]. nian flows with low Reynolds numbers. Seghir et al. [454] present
a numerical model coupling boundary and finite elements suitable
4.9.2 Applications. A rich literature on applications of for dynamic dam-reservoir interaction employing a standard finite
coupled FEM-BEM approaches has flourished in recent years element idealization of the dam structure and a symmetric bound-
combining and exploiting at best the potentials of both the FEM ary element formulation for the unbounded reservoir domain.
and BEM techniques. The FEM is the standard method of choice Other important applications include diffuse optical tomography
for the analysis of complex structural parts and, more in general, by Elisee et al. [455] and modeling of photonic crystals for semi-
of structural mechanics in the presence of shells, plates and conductor laser beams by Jerez-Hanckes et al. [456].
beams. It allows for an easy handling of general constitutive laws
(encompassing nonlinearities and anisotropy) and is widely used
in the industry. However, the BEM is clearly superior in the pres-
5 Future Directions
ence of domains of infinite extension featuring linear isotropic In this closing section, we discuss some future directions of the
constitutive behaviors, evolving domains, strong gradients and BEM and related research topics. The views expressed herein are
singularities. When both situations occur in a single analysis, then not meant to be exclusive. They are merely those of the authors of
a coupled approach turns out to be very promising. this review article, aiming to shed some light on the research
A typical application with these requirements is represented by efforts to be made in the coming years. All these efforts should
3D linear fracture mechanics, especially in the presence of propa- further improve the formulations, implementations, solutions, and
gating fractures in complex structures. The FEM is inefficient in applications of the BEM to make it an even more effective and ef-
the modeling of singular stress fields (or regions with strong gra- ficient numerical method in the fields of computational science
dients) and in the presence of evolving domains. Such difficulties and engineering.
have been elegantly circumvented by a coupling procedure (with
the symmetric Galerkin BEM in particular), e.g., by Frangi and 5.1 Multi-scale, Multi-physics, and Nonlinear Problems.
Novati [426], Forth and Staroselsky [280], Aour, et al. [427], and Multi-scale and multi-physics applications of the BEM should be
Lucht [428]. a productive area of research for the BEM in the coming years.
In elastoplastic analyses with a limited spread of plastic defor- Many of the multi-physics and multi-scale problems involve com-
mations, the FEM is utilized where the plastic material behavior is plicated geometries and large-scale numerical models, such as
expected to develop. The remaining bounded/unbounded linear modeling of nanomaterials [250], biomaterials, and MEMS, that
elastic regions are best approximated by the BEM [409]. A central can be solved efficiently now by the BEM. Any new development
aspect of coupling approaches is that they require the user to pre- of the BEM to solve realistic models in these areas will further
define and manually localize the FEM and BEM sub-domains demonstrate the unique capabilities of the BEM and benefit
prior to the analysis. Recently, Elleithy [429], Elleithy and Grzhi- industry.
bovskis [430] focused on the development of an adaptive FEM–- The authors believe that the BEM is ready to solve some non-
BEM coupling algorithm in which they provide fast and helpful linear problems on a larger scale as compared with the conven-
estimation of the FEM and BEM sub-domains. tional BEM used in the past. Because of the new formulations
Soil structure interaction in statics and dynamics has stimulated (symmetric Galerkin, various preconditioning techniques, etc.)
intensive research in coupled approaches where the BEM is which offer more stable solutions and fast solution methods
employed for the infinite (or semi-infinite) and possibly layered (FMM, pFFT, ACA, etc.) with iterative solvers for large-scale
soil, while the FEM is adopted for the finite structure and struc- problems, the BEM and the related methods should be able to han-
tural elements like piles [431–436]. Recently, applications in dle effectively nonlinear problems with elastic-plastic materials
vibration analysis of transport systems have been addressed by [35], large deformation [39,457,458], and moving-boundary prob-
similar techniques [437–439]. lems [38] such as in hydrodynamics [459].
In magnetostatics industrial applications, the BEM-FEM cou-
pling is often applied to the analysis of finite metallic parts of a
mechanism (like a relay) excited by coils. The surrounding air is 5.2 More Efficient Fast Solution Methods. The BEM
considered as an isotropic and linear medium addressed by the accelerated with the FMM, pFFT, or ACA can now solve large-
BEM with total or reduced scalar potential approaches. The nonlin- scale BEM models of several million DOFs very efficiently, if
ear constitutive behavior of the metallic parts and their high sur- they work. In some cases, the convergence of the FMM, pFFT, or
face-to-volume ratio call, on the contrary, for the application of the ACA with iterative solvers may not be achieved within reasonable
FEM. This approach also allows performing coupled analyses time limits. Therefore, developing better pre-conditioners for the
encompassing large displacements and rotations of the pivoting BEM systems of equations has been, and will continue to be, a
lamina. Investigations along these guidelines have been presented, focus for improving the fast multipole, pFFT, and ACA BEMs.
e.g., by Kuhn [440], Balac and Caloz [441], Kuhn and Steinbach Wavelet methods [460,461] can also be used to compress the
[442], Frangi et al. [443,444], Salgado and Selgas [445], Pusch and BEM matrices to yield better conditioned systems of equations for
Ostrowski [446]. The simulation of eddy currents can also be con- solving large-scale BEM models. To make kernel-independent
veniently addressed with similar approaches [447]. fast multipole BEM, the black box, or generalized fast multipole
BEM-FEM techniques have a dominant status in the simulation method (bbFMM) [462] can be further explored.
of elasto-acoustic coupling. While the FEM is the preferred tool in
the field of structural vibrations, using the BEM for the unbounded 5.3 Applications in Urgent Engineering Fields. The BEM
linear domain has the additional benefit that the Sommerfeld radia- is uniquely positioned to address some of the urgent problems in
tion condition for exterior domains is inherently fulfilled. Advanced biomedical engineering, alternative energy and environmental en-
applications have been presented by Chen et al. [448], Czygan and gineering. For example, the computer image-based BEM can be
von Estorff [449], Gaul and Wenzel [450], Langer and Antes [451], applied to study thermal and electrical responses of bones or
Fischer and Gaul [425], and Soares [452]. brains [44,463], which are difficult to model using other methods
Although the above categories have attracted most of the due to the complicated domains. The fluid-structure interaction
research effort, fluid structure interaction is also emerging as problem of red-blood cells in Stokes flow can also be attempted
promising. For example, Wang et al. [453] focused on the simula- by the BEM [44,464]. This is currently an active research topic.
tion of cell motion in a biological fluid flow. The membrane of a We believe that the BEM can play a significant role in the devel-
moving cell is represented by a thin shell composed of incompres- opment of the next generation, more efficient and environment-
sible neo-Hookean elastic materials and the movement of liquids friendly energy harvesting systems in the next decade to tackle

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-27

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 urgent energy-related problems. For example, the BEM can be the BIE theory, element formulations, programming, and real-
applied to predict thermal, electrical, fluid flow, and other mechani- world applications, is needed. Integration of the BEM with other
cal responses, fatigue life and noise levels of arrays of solar panels computational methods in a course on computational science and
and wind turbines, which are often in open spaces and in large engineering may be a better way to introduce the BEM to
scales. With full-scale and large-scale models of solar panels and students.
wind turbines, we will be able to study their interactions with the There have been several conference series dedicated to the
environment and evaluate their impacts on human beings and other BEM in the last decade, such as the International Conference on
species. Boundary Elements and other Mesh Reduction Methods (BEM/
In other areas of environmental engineering, the BEM can be MRM), International Conference on Boundary Element Techni-
applied to predict noise along a highway stretch caused by cars or ques (BETeq), International Association for Boundary Element
in a city block caused by high-speed trains or explosives, to track Methods (IABEM) Conference, and numerous minisymposia at
the distribution and propagation of pollutants over a city, or to the World Congress on Computational Mechanics and US
evaluate manmade noise in the ocean and the negative impacts on National Congress on Computational Mechanics. In addition, two
marine lives. workshops on the fast multipole BEM were conducted, one in Los
Angeles in 2006 and one in Kyoto in 2007. There were also two
5.4 Software Development. The transition of research codes GF related workshops earlier in the 1990s. The first workshop was
to commercial software for the BEM has been lagging for the last on “Green’s functions and boundary element analysis for model-
three decades as compared with those for other numerical meth- ing the mechanical behavior of advanced materials” which was
ods. This is not easy for researchers who are not familiar with the held in 1994 in Boulder, Colorado and sponsored by the National
marketplace. However, for the BEM to be useful and to be rele- Institute of Standards and Technology (NIST). In 1998, another
vant in computational science and engineering, researchers on the workshop titled “Library of Green’s Functions and Its Industrial
BEM have to be aware of this issue. BEM researchers should Applications” was held at Gaithersburg, Maryland. This work-
solve large-scale, realistic, industry relevant problems with their shop, also sponsored by NIST, was to demonstrate the idea and
developed BEM approaches and codes, moving beyond the aca- application of a library of discretized GFs to industrial problems.
demic problems they are used to solve. They should interact more There were two GF related educational projects supported by
often with engineers to be aware of the type of the problems that NSF: One was the GREEN Project (Green’s Functions Research
they can solve with the BEM and thus contribute to industry & Education Enhancement Network), the other was an extension
research and development efforts. of the GREEN project “NSF-NSDL GREEN Project: A Digital
In fact, there are many advantages of the BEM regarding the Library Partnership of Academia, Government, and Industry.”
development of a software package, as compared with the FEM. Nationwide and worldwide annual or biannual conferences,
The data structure for a BEM model is simpler and requires much workshops, and symposia on the BEM and related methods should
less storage space due to the use of boundary elements. The continue so that more students and young researchers can have a
required geometric (CAD) model of a problem can also be less chance to learn the BEM, develop interests in the BEM and other
stringent. For example, only the surface data are required in the computational methods, and eventually help to advance BEM
BEM, and that surface data need not to be perfect (or “air tight”). research and development in the future.
Meshing surfaces for the BEM are straightforward and there are
many codes for meshing available in the public domain. There-
fore, optimizing and integrating a BEM research code with other Acknowledgment
software components to develop a BEM software package is rela- The authors would like to acknowledge the funding for the
tively straightforward. BEM workshop from the US National Science Foundation (award
There have been some efforts by researchers in the BEM to de- number CMMI-1019262) with Dr. Glaucio H. Paulino as the pro-
velop commercial BEM software for industrial applications. gram director. The views expressed in this review paper, however,
Among the commercial BEM software available, we mention the do not necessarily represent those of the sponsoring agency.
packages GPBEST by BEST Corporation, BEASYTM by Computa-
tional Mechanics Inc., COMET BEAT by Comet Technology Cor-
poration, Virtual.Lab by LMS International, and FastBEM References
AcousticsV R by Advanced CAE Research. A list of free educational [1] Jaswon, M. A., 1963, “Integral Equation Methods in Potential Theory. I,”
computer codes developed at several universities for solving vari- Proc. R. Soc. London, Ser. A, 275, pp. 23–32.
[2] Symm, G. T., 1963, “Integral Equation Methods in Potential Theory. II,” Proc.
ous problems using the BEM can also be found at the Boundary R. Soc. London, Ser. A, 275, pp. 33–46.
Element Resources Network (BENET) [465]. BEM source codes [3] Jaswon, M. A., and Ponter, A. R., 1963, “An Integral Equation Solution of the
for education can also be found in textbooks, such as the code for Torsion Problem,” Proc. R. Soc. London, Ser. A 273, pp. 237–246.
Galerkin BEM in Ref. [43] and the code for fast multipole BEM [4] Rizzo, F. J., 1967, “An Integral Equation Approach to Boundary Value Prob-
lems of Classical Elastostatics,” Quart. Appl. Math., 25, pp. 83–95.
in Ref. [44]. [5] Rizzo, F. J., and Shippy, D. J., 1968, “A Formulation and Solution Procedure
More coordination among the BEM researchers and collabora- for the General Non-Homogeneous Elastic Inclusion Problem,” Int. J. Solids
tions with industries are urgently needed in order to develop the Struct., 4, pp. 1161–1179.
next generation BEM software that will be unique, efficient, and [6] Cruse, T. A., and Rizzo, F. J., 1968, “A Direct Formulation and Numerical So-
lution of the General Transient Elastodynamic Problem - I,” J. Math. Anal.
user-friendly, and to add more valuable computational tools in the Appl., 22, pp. 244–259.
computational toolbox for engineers. Establishing an open-source [7] Cruse, T. A., 1968, “A Direct Formulation and Numerical Solution of the Gen-
platform for the BEM, where researchers and developers can con- eral Transient Elastodynamic Problem - II,” J. Math. Anal. Appl., 22, pp.
tribute and share their BEM source codes and work together to- 341–355.
[8] Cruse, T. A., 1969, “Numerical Solutions in Three Dimensional Elastostatics,”
ward an integrated BEM software system, may be a first step Int. J. Solids Struct., 5, pp.1259–1274.
towards achieving these goals. [9] Rizzo, F. J., and Shippy, D. J., 1970, “A Method for Stress Determination in
Plane Anisotropic Elastic Bodies,” J. Compos. Mater., 4, pp. 36–61.
[10] Rizzo, F. J., and Shippy, D. J., 1970, “A Method of Solution for Certain Prob-
5.5 Education. Education of a new generation of students lems of Transient Heat Conduction,” AIAA J., 8, pp. 2004–2009.
who are interested in conducting research and development of the [11] Rizzo, F. J., and Shippy, D. J., 1971, “An Application of the Correspondence
BEM will be crucial for further advances in the BEM. Educators Principle of Linear Viscoelasticity Theory,” SIAM J. Appl. Math., 21, pp.
321–330.
need to consider more innovative ways to teach the subjects of the [12] Cruse, T. A., and Buren, W. V., 1971, “Three-Dimensional Elastic Stress
BEM that have sometimes been abstract and theoretical in the Analysis of a Fracture Specimen with an Edge Crack,” Int. J. Fract. Mech., 7,
past. Developing new course materials on the BEM, that include pp. 1–16.

031001-28 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 [13] Cruse, T. A., and. Swedlow, J. L., 1971, “Formulation of Boundary Integral [50] Mogilevskaya, S. G., 1996, “The Universal Algorithm based on Complex
Equations for Three-Dimensional Elasto-Plastic Flow,” Int. J. Solids Struct., Hypersingular Integral Equation to Solve Plane Elasticity Problems,” Comput.
7, pp. 1673–1683. Mech., 18, pp. 127–138.
[14] Cruse, T. A., 1973, “Application of the Boundary-Integral Equation Method to [51] Mogilevskaya, S. G., and Linkov, A. M., 1998, “Complex Fundamental Solu-
Three-Dimensional Stress Analysis,” Comput. Struct., 3, pp. 509–527. tions and Complex Variables Boundary Element Method in Elasticity,” Com-
[15] Cruse, T. A., 1974, “An Improved Boundary-Integral Equation Method for put. Mech., 22, pp. 88–92.
Three Dimensional Elastic Stress Analysis,” Comput. Struct., 4, pp. 741–754. [52] Green, G., 1828, An Essay on the Application of Mathematical Analysis to the
[16] Cruse, T. A., and Rizzo, F. J., eds.,1975, Boundary-Integral Equation Method: Theories of Electricity and Magnetism, T. Wheelhouse, Nottingham.
Computational Applications in Applied Mechanics, AMD-ASME, New York, [53] Seremet, V. D., 2002, Handbook of Green’s Functions and Matrices, WIT
Vol. 11. Press, Southampton.
[17] Lachat, J. C., and Watson, J. O., 1976, “Effective Numerical Treatment of [54] Stakgold, I., 1997, Green’s Functions and Boundary Value Problems, 2nd ed.,
Boundary Integral Equations: A Formulation for Three-Dimensional Wiley-Interscience Publications, New York.
Elastostatics,” Int. J. Numer. Methods Eng., 10, pp. 991–1005. [55] Fairweather, G., and Karageorghis, A., 1998, “The Method of Fundamental
[18] Rizzo, F. J., and Shippy, D. J., 1977, “An Advanced Boundary Integral Equa- Solutions for Elliptic Boundary Value Problems,” Adv. Comput. Math., 9, pp.
tion Method for Three-Dimensional Thermoelasticity,” Int. J. Numer. Methods 69–95.
Eng., 11, pp. 1753–1768. [56] Chen, C. S., Goldberg, M. A., and Hon, Y. C., 1998, “The Method of Funda-
[19] Stippes, M., and Rizzo, F. J., 1977, “A Note on the Body Force Integral of mental Solutions and Quasi-Monte-Carlo Method for Diffusion Equations,”
Classical Elastostatics,” Zeits Ang. Math. Physik (ZAMP), 28, pp. 339–341. Int. J. Numer. Methods Eng., 43, pp. 1421–1435.
[20] Wilson, R. B., and Cruse, T. A., 1978, “Efficient Implementation of Aniso- [57] Eshelby, J. D., 1957, “The Determination of the Elastic Field of an Ellipsoidal
tropic Three Dimensional Boundary-Integral Equation Stress Analysis,” Int. J. Inclusion, and Related Problems,” Proc. R. Soc. London, A241, pp. 376–396.
Numer. Methods Eng., 12, pp. 1383–1397. [58] Ding, H. J., Chen, W. Q., and Jiang, A. M., 2004, “Green’s Functions and
[21] Banerjee, P. K., and Butterfield, R., 1976, “Boundary Element Methods in Boundary Element Method for Transversely Isotropic Piezoelectric Materi-
Geomechanics,” Finite Elements in Geomechanics, G. Gudehus, ed., John als,” Eng. Anal. Boundary Elem., 28, pp. 975–987.
Wiley and Sons (U.K.), London, Chapter 16, pp. 529–570. [59] Ding, H. J., Wang, G. Q., and Chen, W. Q., 1997, “Fundamental Solutions for
[22] Brebbia, C. A., and Dominguez, J., 1977, “Boundary Element Methods for Plane Problem of Piezoelectric Materials,” Sci. China, E40, pp. 331–336.
Potential Problems,” Appl. Math. Model.,1, pp. 372–378. [60] Dunn, M. L., and Wienecke, H. A., 1999, “Half-Space Green’s Functions for
[23] Brebbia, C. A., 1978, The Boundary Element Method for Engineers, Pentech Transversely Isotropic Piezoelectric Solids,” J. Appl. Mech., 66, pp. 675–679.
Press, London. [61] Ding, H. J., Jiang, A. M., Hou, P. F., and Chen, W. Q., 2005, “Green’s Func-
[24] Rizzo, F. J., 2003, “Springs, Formulas and Flatland: A Path to Boundary Inte- tions for Two-Phase Transversely Isotropic Magneto-Electro-Elastic Media,”
gral Methods in Elasticity,” Electron. J. Boundary Elem., 1, pp. 1–7. Eng. Anal. Boundary Elem., 29, pp. 551–561.
[25] Cruse, T. A., 2003, “Boundary Integral Equations – A Personal View,” Elec- [62] Bacon, D. J., Barnett, D. M., and Scattergood, R. O., 1980, “Anisotropic Con-
tron. J. Boundary Elem., 1, pp. 19–25. tinuum Theory of Lattice Defects,” Prog. Mater. Sci., 23, pp. 51–262.
[26] Watson, J. O., 2003, “Boundary Elements from 1960 to the Present Day,” [63] Ting, T. C. T., 1996, Anisotropic Elasticity, Oxford University Press, Oxford.
Electron. J. Boundary Elem., 1, pp. 34–46. [64] Qin, Q. H., 2005, “2D Green’s Functions of Defective Magnetoelectroelastic
[27] Shippy, D. J., 2003, “Early Development of the BEM at the University of Solids under Thermal Loading,” Eng. Anal. Boundary Elem., 29, pp. 577–585.
Kentucky,” Electron. J. Boundary Elem., 1, pp. 26–33. [65] Pan, E. and Amadei, B., 1996, “Fracture Mechanics Analysis of 2–D Cracked
[28] Mukherjee, S., 2003, “Boundary Element Methods in Solid Mechanics - A Anisotropic Media with a New Formulation of the Boundary Element Meth-
Tribute to Frank Rizzo,” Electron. J. Boundary Elem., 1, pp. 47–55. od,” Int. J. Fract., 77, pp. 161–174.
[29] Telles, J. C. F., 2003, “A Report on some Boundary Element Adventures,” [66] Yin, W. L. 2005, “Green’s Function of Bimaterials Comprising all Cases of
Electron. J. Boundary Elem., 1, pp. 56–60. Material Degeneracy,” Int. J. Solids Struct., 42, pp. 1–19.
[30] Yao, Z., and Du, Q., 2003, “Some Aspects of the BEM Research in China,” [67] Lavagnino, A. M., 1995, Selected Static and Dynamic Problems in Anisotropic
Electron. J. Boundary Elem., 1, pp. 61–67. Linear Elasticity, Stanford University, Stanford, CA.
[31] Rudolphi, T. J., and Liu, Y. J., 2003, “BIE=BEM - The Past, Present and [68] Buroni, F. C., Ortiz, J. E., and Sáez, A., 2011, “Multiple Pole Residue
Future, a Special Issue in Honor of Professor Frank J. Rizzo,” Electron. J. Approach for 3D BEM Analysis of Mathematical Degenerate and Non-
Boundary Elem., 1, pp. 1-67. Degenerate Materials,” Int. J. Numer. Methods Eng, 86, pp. 1125–1143.
[32] Cheng, A. H.-D., and Cheng, D. T., 2005, “Heritage and Early History of the [69] Pan, E. and Yuan, F. G., 2000, “Three-Dimensional Green’s Functions in Ani-
Boundary Element Method,” Engineering Analysis with Eng. Anal. Boundary sotropic Bimaterials,” Int. J. Solids Struct., 37, pp. 5329–5351.
Elem., 29, pp. 268–302. [70] Pan, E., 2003, “Three-Dimensional Green’s Functions in an Anisotropic Half
[33] Tanaka, M., 1983, “Some Recent Advances in Boundary Element Methods,” Space with General Boundary Conditions,” J. Appl. Mech., 70, pp. 101–110.
Appl. Mech. Rev., 36, pp. 627–634. [71] Pan, E., 2003, “Three-Dimensional Green’s Functions in Anisotropic Elastic
[34] Tanaka, M., Sladek, V., and Sladek, J., 1994, “Regularization Techniques Bimaterials with Imperfect Interfaces,” J. Appl. Mech, 70, pp. 180–190.
Applied to Boundary Element Methods,” Appl. Mech. Rev., 47, pp. [72] Pan, E., 2002, “Three-Dimensional Green’s Functions in Anisotropic Mag-
457–499. neto-Electro-Elastic Bimaterials,” J. Appl. Math. Phys., 53, pp. 815–838.
[35] Mukherjee, S., 1982, Boundary Element Methods in Creep and Fracture, [73] Buroni, F. C., and Sáez, A., 2010, “Three-Dimensional Green’s Function and
Applied Science Publishers, New York. its Derivative for Materials with General Anisotropic Magneto-Electro-Elastic
[36] Cruse, T. A., 1988, Boundary Element Analysis in Computational Fracture Coupling,” Proc. R. Soc. A., 466, pp. 515–537.
Mechanics, Kluwer Academic Publishers, Dordrecht, The Netherlands. [74] Zhao, M. H., Fan, C. Y., Liu, T., and Yang, F., 2007, “Extended Discontinuity
[37] Brebbia, C. A. and Dominguez, J., 1989, Boundary Elements - An Introductory Green’s Functions for Three-Dimensional Transversely Isotropic Magneto-
Course, McGraw-Hill, New York. Electro-Elastic Media and Applications,” Eng. Anal. Boundary Elem., 31, pp.
[38] Banerjee, P. K., 1994, The Boundary Element Methods in Engineering, 2nd 547–558.
ed., McGraw-Hill, New York. [75] Zhao, M. H., Fan, C. Y., Yang, F., and Liu, T., 2007, “Analysis Method of Pla-
[39] Chandra, A. and Mukherjee, S., 1997, Boundary Element Methods in Manu- nar Cracks of Arbitrary Shape in the Isotropic Plane of a Three-Dimensional
facturing, Oxford University Press, New York. Transversely Isotropic Magnetoelectroelastic Medium,” Int. J. Solids Struct.,
[40] Paris, F. and Canas, J., 1997, Boundary Element Methods: Fundamentals and 44, pp. 4505–4523.
Applications, Oxford University Press, Oxford, UK. [76] Pan, E., and Han, F., 2005, “Green’s Functions for Transversely Isotropic Pie-
[41] Bonnet, M., 1995, Boundary Integral Equation Methods for Solids and Fluids, zoelectric Functionally Graded Multilayered Half Spaces,” Int. J. Solids
John Wiley & Sons, Chichester. Struct., 42, pp. 3207–3233.
[42] Mukherjee, S. and Mukherjee, Y. X., 2005, Boundary Methods: Elements, [77] Han, F., Pan, E., Roy, A. K., and Yue, Z. Q., 2006, “Responses of Piezoelec-
Contours, and Nodes, CRC Press, Boca Raton, FL. tric, Transversely Isotropic, Functionally Graded, and Multilayered Half
[43] Sutradhar, A., Paulino, G. H., and Gray, L. J., 2008, Symmetric Galerkin Spaces to Uniform Circular Surface Loadings,” Comput. Model. Eng. Sci., 14,
Boundary Element Method, Springer-Verlag, Berlin, Germany. pp. 15–30.
[44] Liu, Y. J., 2009, Fast Multipole Boundary Element Method - Theory and [78] Pan, E., 2005, “Anisotropic Green’s Functions and BEMs (Editor),” Eng.
Applications in Engineering, Cambridge University Press, Cambridge. Anal. Boundary Elem., 29, pp. 161.
[45] Mukherjee, S., 2000, “CPV and HFP Integrals and their Applications in the [79] Denda, M., Araki, Y., and Yong, Y. K., 2004, “Time-Harmonic BEM for 2–D
Boundary Element Method,” Int. J. Solids Struct., 37, pp. 6623–6634. Piezoelectricity Applied to Eigenvalue Problems,” Int. J. Solids Struct., 41,
[46] Liu, Y. J. and Rudolphi, T. J., 1991, “Some Identities for Fundamental Solu- pp. 7241–7265.
tions and their Applications to Weakly-Singular Boundary Element For- [80] Wang, C. Y., and Zhang, C., 2005, “3–D and 2–D Dynamic Green’s Functions
mulations,” Eng. Anal. Boundary Elem., 8, pp. 301–311. and Time-Domain BIEs for Piezoelectric Solids,” Eng. Anal. Boundary Elem.,
[47] Liu, Y. J., 2000, “On the Simple-Solution Method and Non-Singular Nature of 29, pp. 454–465.
the BIE=BEM - A Review and some New Results,” Eng. Anal. Boundary [81] Wu, K. C., and Chen, S. H., 2007, “Two Dimensional Dynamic Green’s Func-
Elem., 24, pp. 787–793. tions for Piezoelectric Materials,” Comput. Model. Eng. Sci., 20, pp. 181–196.
[48] Linkov, A. M., 2002, Boundary Integral Equations in Elasticity Theory, [82] Ren, D. L., and Liu, J. X., 2004, “Time-Harmonic Dynamic Fundamental Sol-
Kluwer Academic Publishers, Dordrecht. utions for Transversely Isotropic Magnetoelectroelastic Media under Anti-
[49] Linkov, A. M., and Mogilevskaya, S. G., 1998, “Complex Hypersingular Plane Deformation,” Int. J. Nonlinear Sci. Numer. Simul., 5, pp. 387–391.
BEM in Plane Elasticity Problems,” Singular Integrals in Boundary Element [83] Chen, P., Shen, Y., and Tian, X., 2006, “Dynamic Potentials and Green’s
Method, V. Sladek and J. Sladek,eds., Computational Mechanics Publications, Functions of a Quasi-Plane Magneto-Electro-Elastic Medium with Inclusion,”
Southampton, pp. 299–364. Int. J. Eng. Sci., 44, pp. 540–553.

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-29

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 [84] Rojas-Dı́az, R., Sáez, A., Garcı́a-Sánchez, F., and Zhang, C., 2008, “Time- [115] Capuani, D., Bigoni, D., and Brun, M., 2005, “Integral Representations at the
Harmonic Green’s Functions for Anisotropic Magnetoelectroelasticity,” Int. J. Boundary for Stokes Flow and Related Symmetric Galerkin Formulation,”
Solids Struct., 45, pp. 144–158. Arch. Mech. 57, pp. 363–385. Available at: http://dezoo.ippt.gov.pl/contents_pdf/
[85] Bui, H. D., 1977, “An Integral Equation Method for Solving the Problem of a 1210152977.pdf.
Plane Crack of Arbitrary Shape,” J. Mech. Phys. Solids, 25, pp. 29–39. [116] Reidinger, B., and Steinbach, O., 2003, “A Symmetric Boundary Element
[86] Sirtori, S., 1979, “General Stress Analysis Method by Means of Integral Equa- Method for the Stokes Problem in Multiple Connected Domains,” Math. Meth-
tions and Boundary Elements,” Meccanica, 14, pp. 210–218. ods Appl. Sci., 26, pp. 77–93.
[87] Hartmann, F., Katz, C., and Protopsaltis, B., 1985, “Boundary Elements and [117] Perez-Gavilan, J. J., and Aliabadi, M. H., 2001, “A Symmetric Galerkin BEM for
Symmetry,” Ingenieur-Archiv, 55, pp. 440–449. Harmonic Problems and Multiconnected Bodies,” Meccanica, 36, pp. 449–462.
[88] Maier, G., and Polizzotto, C., 1987, “A Galerkin Approach to Boundary Ele- [118] Gray, L. J., and Paulino, G. H., 1997, “Symmetric Galerkin Boundary Integral
ment Elastoplastic Analysis,” Comput. Methods Appl. Mech. Eng., 60, pp. Formulation for Interface and Multi-Zone Problems,” Int. J. Numer. Methods
175–194. Eng., 40, pp. 3085–3101.
[89] Holzer, S. M., 1992, “The Symmetric Galerkin BEM for Plane Elasticity: [119] Zhang, X. F., Liu, Y. H., and Cen, Z. Z., 2001, “A Solution Procedure for
Scope and Applications,” Numerical Methods in Engineering ’92, Elsevier, Lower Bound Limit and Shakedown Analysis by SGBEM,” Acta Mech. Sol-
Amsterdam. ida Sinica, 14, pp. 118–129.
[90] Bonnet, M., and Bui, H. D., 1988, “Regular BIE for three-dimensional cracks [120] Lehmann, L., and Antes, H., 2001, “Dynamic Structure-Soil-Structure Interac-
in elastodynamics,” Advanced Boundary Element Methods, Springer-Verlag, tion Applying the Symmetric Galerkin Boundary Element Method (SGBEM),”
Berlin and New York, pp. 41–47. Mech. Res. Commun., 28, pp. 297–304.
[91] Kane, J., and Balakrishna, C., 1993, “Symmetric Galerkin Boundary Formula- [121] Polizzotto, C., 2000, “A Symmetric Galerkin Boundary=Domain Element
tions Employing Curved Elements,” Int. J. Numer. Methods Eng., 36, pp. Method for Finite Elastic Deformations,” Comput. Meth. Appl. Mech., 189,
2157–2187. pp. 481–514.
[92] Andra, H., 1998, “Integration of Singular Integrals for the Galerkin-Type [122] Perez-Gavilan, J. J., and Aliabadi, M. H., 2003, “Symmetric Galerkin BEM
Boundary Element Method in 3D Elasticity,” Comp. Mech. Appl. Mech. Eng., for Shear Deformable Plates,” Int. J. Numer. Methods Eng., 57, pp.
157, pp. 239–249. 1661–1693.
[93] Carini, A., and Salvadori, A., 2002, “Analytical Integrations 3D BEM,” Com- [123] Perez-Gavilan, J. J., and Aliabadi, M. H., 2001, “A Symmetric Galerkin
put. Mech., 28, pp. 177–185. Boundary Element Method for Dynamic Frequency Domain Viscoelastic
[94] Carini, A., Diligenti, M., Maranesi, P., and Zanella, M., 1999, “Analytical Problems,” Comput. Struct., 79, pp. 2621–2633.
Integrations for Two Dimensional Elastic Analysis by the Symmetric Galerkin [124] Frangi, A., and Bonnet, M., 1998, “A Galerkin Symmetric and Direct BIE
Boundary Element Method,” Comput. Mech., 23, pp. 308–323. Method for Kirchoff Elastic Plates: Formulation and Implementation,” Int. J.
[95] Haas, M. and Kuhn, G., 2002, “A Symmetric Galerkin BEM Implementation Numer. Methods Eng., 41, pp. 337–369.
for 3D Elastostatic Problems with an Extension to Curved Elements,” Comput [125] Nayroles, B., Touzot, G., and Villon, P., 1992, “Generalizing the Finite Ele-
Mech., 28, pp. 250–259. ment Method: Diffuse Approximation and Diffuse Elements,” Comput.
[96] Frangi, A., and Novati, G., 1996, “Symmetric BE Method in two-Dimensional Mech., 10, pp. 307–318.
Elasticity: Evaluation of Double Integrals for Curved Elements,” Comput [126] Belytschko, T., Lu, Y. Y., and Gu, L., 1994, “Element-Free Galerkin Meth-
Mech., 19, pp. 58–68. ods,” Int. J. Numer. Methods Eng., 37, pp. 229–256.
[97] Frangi, A., and Guiggiani, M., 2000, “A Direct Approach for Boundary Inte- [127] Mukherjee, Y. X., and Mukherjee, S., 1997, “The Boundary Node Method for
gral Equations with High-Order Singularities,” Int. J. Numer. Methods Eng., Potential Problems,” Int. J. Numer. Methods Eng., 40, pp. 797–815.
49, pp. 871–898. [128] Kothnur, V., Mukherjee, S., and Mukherjee, Y. X., 1999, “Two-Dimensional
[98] Salvadori, A., 2001, “Analytical Integrations of Hypersingular Kernel in 3D Linear Elasticity by the Boundary Node Method,” Int. J. Solids Struct., 36, pp.
BEM Problems,” Comp. Meth. Appl. Mech. Eng., 190, pp. 3957–3975. 1129–1147.
[99] Salvadori, A., 2002, “Analytical Integrations in 2D BEM Elasticity,” Int. J. [129] Chati, M. K., Mukherjee, S., and Mukherjee, Y. X., 1999, “The Boundary
Numer. Methods Eng., 53, pp. 1695–1719. Node Method for Three-Dimensional Linear Elasticity,” Int. J. Numer. Meth-
[100] Gray, L. J., 1998, “Evaluation of Singular and Hypersingular Galerkin Bound- ods Eng., 46, pp. 1163–1184.
ary Integrals: Direct Limits and Symbolic Computation,” Singular Integrals in [130] Chati, M. K., and Mukherjee, S., 2000, “The Boundary Node Method for
the Boundary Element Method, Computational Mechanics Publishers, South- Three-Dimensional Problems in Potential Theory,” Int. J. Numer. Methods
ampton, pp. 33–84. Eng., 47, pp. 1523–1547.
[101] Gray, L. J., Glaeser, J., and Kaplan, T., 2004, “Direct Evaluation of Hypersin- [131] Chati, M. K., Paulino, G. H., and Mukherjee, S., 2001, “The Meshless Stand-
gular Galerkin Surface Integrals,” SIAM J. Sci. Comput., 25, pp. 1534–1556. ard and Hypersingular Boundary Node Methods- Applications to Error Esti-
[102] Sutradhar, A., Paulino, G. H., and Gray, L. J., 2005, “On Hypersingular Sur- mation and Adaptivity in Three-Dimensional Problems,” Int. J. Numer.
face Integrals in the Symmetric Galerkin Boundary Element Method: Applica- Methods Eng., 50, pp. 2233–2269.
tion to Heat Conduction in Exponentially Graded Materials,” Int. J. Numer. [132] Chati, M. K., Mukherjee, S., and Paulino, G. H., 2001, “The Meshless Hyper-
Methods Eng., 62, pp. 122–157. singular Boundary Node Method for Three-Dimensional Potential Theory and
[103] Gray, L. J., and Paulino, G. H., 1997, “Symmetric Galerkin Boundary Integral Linear Elasticity Problems,” Eng. Anal. Boundary Elem., 25, pp. 639–653.
Fracture Analysis for Plane Orthotropic Elasticity,” Comput. Mech., 20, pp. [133] Gowrishankar, R., and Mukherjee, S., 2002, “The ‘Pure’ Boundary Node Method
26–33. for Potential Theory,” Commun. Numer. Methods Eng., 18, pp. 411–427.
[104] Li, S., Mear, M. E., and Xiao, L., 1998, “Symmetric Weak Form Integral [134] Sladek, J., Sladek, V., and Atluri, S. N., 2000, “Local Boundary Integral Equa-
Equation Method for Three-Dimensional Fracture Analysis,” Comp. Meth. tion (LBIE) Method for Solving Problems of Elasticity with Non-
Appl. Mech. Eng., 151, pp. 435–459. Homogeneous Material Properties,” Comput. Mech., 24, pp. 456–462.
[105] Frangi, A., Novati, G., Springhetti, R., and Rovizzi, M., 2002, “3D Fracture [135] Zhu, T., Zhang, J. D., and Atluri, S. N., 1998, “A Local Boundary Integral
Analysis by the Symmetric Galerkin BEM,” Comput. Mech., 28, pp. 220–232. Equation (LBIE) Method in Computational Mechanics, and a Meshless Dis-
[106] Frangi, A., 2002, “Fracture Propagation in 3D by the Symmetric Galerkin cretization Approach,” Comput. Mech., 21, pp. 223–235.
Boundary Element Method,” Int. J. Fract., 116, pp. 313–330. [136] Chen, W., and Tanaka, M., 2002, “A Meshless, Integration-Free, and
[107] Salvadori, A., 1999, “Quasi Brittle Fracture Mechanics by Cohesive Crack Boundary-Only RBF Technique,” Comput. Math. Appl., 43, pp. 379–391.
Models and Symmetric Galerkin Boundary Element Method,” Ph. D. thesis, [137] Li, G., and Aluru, N. R., 2002, “Boundary Cloud Method: A Combined Scat-
Politecnico Milano, Italy. tered Point=Boundary Integral Approach for Boundary-Only Analysis,” Com-
[108] Sutradhar, A., and Paulino, G. H., 2004, “Symmetric Galerkin Boundary Ele- put. Methods Appl. Mech. Eng., 191, pp. 2337–2370.
ment Computation of T-Stress and Stress Intensity Factors for Mixed-Mode [138] Li, G., and Aluru, N. R., 2003, “A Boundary Cloud Method with a Cloud-By-
Cracks by the Interaction Integral Method,” Eng. Anal. Boundary Elem., 28, Cloud Polynomial Basis,” Eng. Anal. Boundary Elem., 27, pp. 57–71.
pp. 1335–1350. [139] Liu, G. R., 2002, Mesh-free Methods: Moving Beyond the Finite Element
[109] Xu, K., Lie, S. T., and Cen, Z., 2004, “Crack Propagation Analysis with Galer- Method,CRC Press, Boca Raton, FL.
kin Boundary Element Method,” Int. J. Numer. Analyt. Meth. Geomech., 28, [140] Zhang, J. M., Yao, Z. H., and Li, H., 2002, “A Hybrid Boundary Node Meth-
pp. 421–435. od,” Int. J. Numer. Methods Eng., 53, pp. 751–763.
[110] Kitey, R., Phan, A. V., Tippur, H. V., and Kaplan, T., 2006, “Modeling of [141] Zhang, J. M., Qin, X., Han, X., and Li, G., 2009, “A Boundary Face Method
Crack Growth Through Particulate Clusters in Brittle Matrix by Symmetric- for Potential Problems in Three Dimensions,” Int. J. Numer. Methods Eng.,
Galerkin Boundary Element Method,” Int. J. Fract., 141, pp. 11–25. 80, pp. 320–337.
[111] Phan, A. V., Napier, J. A. L., Gray, L. J., and Kaplan, T., 2003, “Stress Inten- [142] Liew, K. M., Cheng, Y., and Kitipornchai, S., 2006, “Boundary Element-Free
sity Factor Analysis of Friction Sliding at Discontinuity Interfaces and Method (BEFM) and its Application to Two-Dimensional Elasticity Prob-
Junctions,” Comput. Mech., 32, pp. 392–400. lems,” Int. J. Numer. Methods Eng., 65, pp. 1310–1332.
[112] Phan, A. V., Napier, J. A. L., Gray, L. J. and Kaplan, T., 2003, “Symmetric- [143] Li, X., and Zhu, J., 2009, “A Galerkin Boundary Node Method and its Conver-
Galerkin BEM Simulation of Fracture with Frictional Contact,” Int. J. Numer. gence Analysis,” J. Comput. Appl. Math., 230, pp. 314–328.
Methods Eng., 57, pp. 835–851. [144] Li, X., and Zhu, J., 2009, “A Galerkin Boundary Node Method for Biharmonic
[113] Phan, A. V., Gray, L. J., and Kaplan, T., 2007, “On Some Benchmark Results Problems,” Eng. Anal. Boundary Elem., 33, pp. 858–865.
for the Interaction of a Crack with a Circular Inclusion,” Trans. ASME, J. [145] Li, X., and Zhu, J., 2009, “A Galerkin Boundary Node Method for Two-
Appl. Mech., 74, pp. 1282–1284. Dimensional Linear Elasticity,” Comput. Model. Eng. Sci., 45, pp. 1–29.
[114] Williams, R. C., Phan, A. V., Tippur, H. V., Kaplan, T., and Gray, L. J., 2007, [146] Li, X., and Zhu, J., 2009, “A Meshless Galerkin Method for Stokes Problems
“SGBEM Analysis of Crack-Particle (s) Interactions due to Elastic Constants using Boundary Integral Equations,” Comput. Methods Appl. Mech., 198, pp.
Mismatch,” Eng. Fract. Mech., 74, pp. 314–331. 2874–2885.

031001-30 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 [147] Li, X., and Zhu, J., 2010, “Meshless Analysis of Two-Dimensional Stokes [179] Liu, Y. J., and Nishimura, N., 2006, “The Fast Multipole Boundary Element
Flows with the Galerkin Boundary Node Method,” Eng. Anal. Boundary Method for Potential Problems: A Tutorial,” Eng. Anal. Boundary Elem., 30,
Elem., 34, pp. 79–91. pp. 371–381.
[148] Telukunta, S., and Mukherjee, S., 2004, “An Extended Boundary Node [180] Saad, Y., and Schultz, M., 1986, “A Generalized Minimal Residual Algorithm
Method for Modeling Normal Derivative Discontinuities in Potential Theory for Solving Nonsymmetric Linear System,” SIAM (Soc. Ind. Appl. Math.) J.
across Edges and Corners,” Eng. Anal. Boundary Elem., 28, pp. 1099–1110. Sci. Stat. Comput., 7, pp. 856–869.
[149] Telukunta, S., and Mukherjee, S., 2005, “The Extended Boundary Node [181] Cheng, H., Greengard, L., and Rokhlin, V., 1999, “A Fast Adaptive Multipole
Method for Three-Dimensional Potential Theory,” Comput. Struct., 83, pp. Algorithm in Three Dimensions,” J. Comput. Phys., 155, pp. 468–498.
1503–1514. [182] Shen, L., and Liu, Y. J., 2007, “An Adaptive Fast Multipole Boundary Ele-
[150] Dumont, N. A., 2003, “Variationally-Based, Hybrid Boundary Element Meth- ment Method for Three-Dimensional Potential Problems,” Comput. Mech.,
ods,” Comp. Assist. Mech. Eng. Sc., 10, pp. 407–430. 39, pp. 681–691.
[151] Dumont, N. A., 1989, “The Hybrid Boundary Element Method: An Alliance [183] Shen, L., and Liu, Y. J., 2007, “An Adaptive Fast Multipole Boundary Ele-
between Mechanical Consistency and Simplicity,” Appl. Mech. Rev., 42, pp. ment Method for Three-Dimensional Acoustic Wave Problems based on the
S54–S63. Burton-Miller Formulation,” Comput. Mech., 40, pp. 461–472.
[152] Washizu, K., 1975, Variational Methods in Elasticity and Plasticity, 2nd ed., [184] Bapat, M. S., Shen, L., and Liu, Y. J., 2009, “Adaptive Fast Multipole Bound-
Pergamon, New York. ary Element Method for Three-Dimensional Half-Space Acoustic Wave Prob-
[153] Banerjee, P. K., and Butterfield, R., 1981, Boundary Element Methods in Engi- lems,” Eng. Anal. Boundary Elem., 33, pp. 1113–1123.
neering Science, McGraw Hill, London. [185] Bapat, M. S., and Liu, Y. J., 2010, “A New Adaptive Algorithm for the Fast
[154] Dumont, N. A., 1998, “An Assessment of the Spectral Properties of the Ma- Multipole Boundary Element Method,” Comput. Model. Eng. Sci., 58, pp.
trix G used in the Boundary Element Methods,” Comput. Mech., 22, pp. 161–184.
32–41. [186] Phillips, J. R., and White, J. K., 1997, “A Pre-Corrected-FFT Method for Elec-
[155] Pian, T. H. H., 1964, “Derivation of Element Stiffness Matrices by Assumed trostatic Analysis of Complicated 3–D Structures,” IEEE Trans. Comput.-
Stress Distribution,” AIAA J., 2, pp. 1333–1336. Aided Des., 16, pp. 1059–1072.
[156] Dumont, N. A., 2006, “Linear Algebra Aspects in the Equilibrium-Based [187] Phillips, J. R., 1997, “Rapid Solution of Potential Integral Equations in Com-
Implementation of Finite=Boundary Element Methods for FGMs,” FGM 2006 plicated 3–Dimensional Geometries,” Ph. D. thesis, MIT, MA.
– Multiscale and Functionally Graded Materials Conference 2006, Proceed- [188] Fata, S. N., 2008, “Fast Galerkin BEM for 3D-Potential Theory,” Comput.
ings of the International Conference FGM IX, American Institute of Physics Mech., 42, pp. 417–429.
Honolulu – O’ahu, Hawaii, pp. 658–663. [189] Fata, S. N., and Gray, L. J., 2009, “A Fast Spectral Galerkin Method for
[157] Chaves, R. A. P., 2003, “The Simplified Hybrid Boundary Element Method Hypersingular Boundary Integral Equations in Potential Theory,” Comput.
Applied to Time-Dependent Problems (in Portuguese),” Ph. D. thesis, PUC- Mech., 44, pp. 263–271.
Rio, Brazil. [190] Ye, W., Wang, X., Hemmert, W., Freeman, D., and White, J., 2003, “Air
[158] Oliveira, M. F. F., 2004, “Conventional, Hybrid and Simplified Boundary Ele- Damping in Laterally Oscillating Microresonators: A Numerical and Experi-
ment Methods (in Portuguese),” Master’s thesis, PUC-Rio, Brazil. mental Study,” J. Microelectromech. Syst., 12, pp. 557–566.
[159] Figueiredo, T. G. B., 1991, “A New Boundary Element Formulation in Engi- [191] Ding, J., and Ye, W., 2004, “A Fast Integral Approach for Drag Force Calcula-
neering,” Lecture Notes in Engineering, C. A. Brebbia and S. A. Orszag eds., tion due to Oscillatory Slip Stokes Flows,” Int. J. Numer. Methods Eng., 60,
Springer-Verlag, Berlin, Germany. pp. 1535–1567.
[160] Gaul, L., Wagner, M., Wenzel, W., and Dumont, N. A., 2001, “On the Treat- [192] Yan, Z. Y., Zhang, J., Ye, W., and Yu, T. X., 2010, “Numerical Characteriza-
ment of Acoustical Problems with the Hybrid Boundary Element Method,” tion of Porous Solids and Performance Evaluation of Theoretical Models via
iInt. J. Solids Struct., 38, pp. 1871–1888. the Pre-Corrected-FFT Accelerated BEM,” Comput. Model. Eng. Sci., 55, pp.
[161] Dumont, N. A., and Cossio, M. U. Q., 2001, “Sensitivity Analysis with the 33–60.
Hybrid Boundary Element Method,” Build. Res. J., 49, pp. 35–58. [193] Masters, N., and Ye, W., 2004, “Fast BEM Solution for Coupled 3D Electro-
[162] Dumont, N. A., and Prazeres, P. G. C., 2005, “Hybrid Dynamic Finite Element static and Linear Elastic Problems,” Eng. Anal. Boundary Elem., 28, pp.
Families for the General Analysis of Time-Dependent Problems,” Third Inter- 1175–1186.
national Conference on Structural Stability and Dynamics, Florida, USA, [194] Ding, J., Ye, W., and Gray, L. J., 2005, “An Accelerated Surface
Procs. ICSSD 2005, p. 10. Discretization-Based BEM Approach for Non-Homogeneous Linear Prob-
[163] Dumont, N. A., and Oliveira, R., 2001, “From Frequency-Dependent Mass lems in 3–D Complex Domains,” Int. J. Numer. Methods Eng., 63, pp.
and Stiffness Matrices to the Dynamic Response of Elastic Systems,” Int. J. 1775–1795.
Solids Struct., 38, pp. 1813–1830. [195] Ding, J., and Ye, W., 2006, “A Grid Based Integral Approach for Quasi-
[164] Dumont, N. A., and Lopes, A. A. O., 2003, “On the Explicit Evaluation of Linear Problems,” Comput. Mech., 38, pp. 113–118.
Stress Intensity Factors in the Hybrid Boundary Element Method,” Fatigue [196] Nie, X. C., Li, L. W., and Yuan, N., 2002, “Pre-Corrected-FFT Algorithm for
Fract. Eng. Mater. Struct., 26, pp. 151–165. Solving Combined Field Integral Equations in Electromagnetic Scattering,” J.
[165] Dumont, N. A., Chaves, R. A. P., and Paulino, G. H., 2004, “The Hybrid Electromagn. Waves Appl., 16, pp. 1171–1187.
Boundary Element Method Applied to Problems of Potential of Functionally [197] Yan, Z. Y., Zhang, J., and Ye, W., 2010, “Rapid Solution of 3–D Oscillatory
Graded Materials,” Int. J. Comput. Eng. Sci., 5, pp. 863–891. Elastodynamics using the pFFT Accelerated BEM,” Eng. Anal. Boundary
[166] Dumont, N. A., 2006, “On the Inverse of Generalized Lambda Matrices with Elem., 34, pp. 956–962.
Singular Leading Term,” Int. J. Numer. Methods Eng., 66, pp. 571–603. [198] Hackbusch, W., 1999, “A Sparse Matrix Arithmethic based on H-Matrices.
[167] Dumont, N. A., 2007, “On the Solution of Generalized Non-Linear Complex- Part I: Introduction to H-Matrices,” Composites, Part B, 62, pp. 89–108.
Symmetric Eigenvalue Problems,” Int. J. Numer. Methods Eng., 71, pp. [199] Bebendorf, M., and Kriemann, R., 2005, “Fast Parallel Solution of Boundary
1534–1568. Integral Equations and Related Problem,” Comput. Visualization Sci., 8, pp.
[168] Gaul, L., Koegl, M., and Wagner, M., 2003, Boundary Element Methods for 121–135.
Engineers and Scientists, Springer Verlag, Berlin, Germany. [200] Goreinov, S. A., Tyrtyshnikov, E. E., and Zamarashkin, N. L., 1997, “A
[169] Wagner, M., Gaul, L., and Dumont, N. A., 2004, “The Hybrid Boundary Ele- Theory of Pseudoskeleton Approximations,” Linear Algebra Appl., 261, pp.
ment Method in Structural Acoustics,” ZAMM, 84, pp. 780–796. 1–21.
[170] Rokhlin, V., 1985, “Rapid Solution of Integral Equations of Classical Potential [201] Bebendorf, M., 2000, “Approximation of Boundary Element Matrices,”
Theory,” J. Comp. Phys., 60, pp. 187–207. Numerische Mathematik, 86, pp. 565–589.
[171] Greengard, L. F., and Rokhlin, V., 1987, “A Fast Algorithm for Particle Simu- [202] Bebendorf, M., 2008, Hierarchical Matrices: A Means to Efficiently Solve
lations,” J. Comput. Phys., 73, pp. 325–348. Elliptic Boundary Value Problems, Springer-Verlag, Berlin, Germany.
[172] Greengard, L. F., 1988, The Rapid Evaluation of Potential Fields in Particle [203] Hackbusch, W., 2009, Hierarchische Matrizen: Algorithmen und Analysis,
Systems, The MIT Press, Cambridge, MA. Springer-Verlag, Berlin, Germany.
[173] Peirce, A. P., and Napier, J. A. L., 1995, “A Spectral Multipole Method for Ef- [204] Bebendorf, M., and Grzhibovskis, R., 2006, “Accelerating Galerkin BEM for
ficient Solution of Large-Scale Boundary Element Models in Elastostatics,” Linear Elasticity using Adaptive Cross Approximation,” Math. Models Meth.
Int. J. Numer. Methods Eng., 38, pp. 4009–4034. Appl. Sci., 29, pp. 1721–1747.
[174] Gomez, J. E., and Power, H., 1997, “A Multipole Direct and Indirect BEM for [205] Maerten, F., 2010, “Adaptive Cross-Approximation Applied to the Solution of
2D Cavity Flow at Low Reynolds Number,” Eng. Anal. Boundary Elem., 19, System of Equations and Post-Processing for 3D Elastostatic Problems using
pp. 17–31. the Boundary Element Method,” Eng. Anal. Boundary Elem., 34, pp.
[175] Fu, Y., Klimkowski, K. J., Rodin, G. J., Berger, E., Browne, J. C., Singer, J. 483–491.
K., Geijn, R. A. V. D., and Vemaganti, K. S., 1998, “A Fast Solution Method [206] Benedetti, I., Aliabadi, M. H., and Davı́, G., 2008, “A Fast 3D Dual Boundary
for Three-Dimensional Many-Particle Problems of Linear Elasticity,” Int. J. Element Method based on Hierarchical Matrices,” Int. J. Solids Struct., 45, pp.
Numer. Methods Eng., 42, pp. 1215–1229. 2355–2376.
[176] Nishimura, N., Yoshida, K., and Kobayashi, S., 1999, “A Fast Multipole [207] Kolk, K., Weber, W., and Kuhn, G., 2005, “Investigation of 3D Crack Propa-
Boundary Integral Equation Method for Crack Problems in 3D,” Eng. Anal. gation Problems via fast BEM Formulations,” Comput. Mech., 37, pp. 32–40.
Boundary Elem., 23, pp. 97–105. [208] Kurz, S., Rain, O., and Rjasanow, S., 2007, “Fast Boundary Element Methods
[177] Mammoli, A. A., and Ingber, M. S., 1999, “Stokes Flow around Cylinders in a in Computational Electromagnetism,” Boundary Element Analysis: Mathemat-
Bounded Two-Dimensional Domain using Multipole-Accelerated Boundary ical Aspects and Applications, M. Schanz and O. Steinbach, eds., Springer-
Element Methods,” Int. J. Numer. Methods Eng., 44, pp. 897–917. Verlag, Berlin Heidelberg, Germany, pp. 249–279.
[178] Nishimura, N., 2002, “Fast Multipole Accelerated Boundary Integral Equation [209] Rjasanow, S., and Steinbach, O., 2007, The Fast Solution of Boundary Integral
Methods,” Appl. Mech. Rev., 55, pp. 299–324. Equations, Springer-Verlag, Berlin, Germany.

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-31

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 [210] Messner, M., and Schanz, M., 2010, “An Accelerated Symmetric Time- [240] Tausch, J., 2003, “Sparse BEM for Potential Theory and Stokes Flow using
Domain Boundary Element Formulation for Elasticity,” Eng. Anal. Boundary Variable Order Wavelets,” Comput. Mech., 32, pp. 312–318.
Elem., 34, pp. 944–955. [241] Cercignani, C., 1988, The Boltzmann Equation and Its Applications, Springer,
[211] Senturia, S. D., Harris, R. M., Johnson, B. P., Kim, S., Nabors, K., Shulman, New York.
M. A., and White, J. K., 1992, “A Computer Aided Design System for Micro- [242] Frangi, A., Ghisi, A., and Coronato, L., 2009, “On a Deterministic Approach
electromechanical Systems (MEMCAD),” IEEE J. Microelectromech Syst., 1, for the Evaluation of Gas Damping in Inertial MEMS in the Free-Molecule
pp. 3–13. Regime,” Sens. Actuators, A, 149, pp. 21–28.
[212] Nabors, K., and White, J., 1991, “FastCap: A Multipole Accelerated 3–D Ca- [243] Frangi, A., 2009, “BEM Technique for Free-Molecule Flows in High Fre-
pacitance Extraction Program,” IEEE Trans. Comput.-Aided Des., 10, pp. quency MEMS Resonators,” Eng. Anal. Boundary Elem., 33, pp. 493–498.
1447–1459. [244] Helsing, J., 1995, “An Integral Equation Method for Elastostatics of Periodic
[213] Shi, F., Ramesh, P., and Mukherjee, S., 1995, “Simulation Methods for Micro- Composites,” J. Mech. Phys. Solids, 43, pp. 815–828.
Electro-Mechanical Structures (MEMS) with Application to a Microtweezer,” [245] Eischen, J. W., and Torquato, S., 1993, “Determining Elastic Behavior of
Comput. Struct., 56, pp. 769–783. Composites by the Boundary Element Method,” J. Appl. Phys., 74, pp.
[214] Li, G., and Aluru, N. R., 2002, “A Lagrangian Approach for Electrostatic 159–170.
Analysis of Deformable Conductors,” J. Microelectromech. Syst., 11, pp. [246] Liu, Y. J., 2005, “A New Fast Multipole Boundary Element Method for Solv-
245–254. ing Large-Scale Two-Dimensional Elastostatic Problems,” Int. J. Numer.
[215] Gilbert, J. R., Legtenberg, R., and Senturia, S. D., 1995, “3D Coupled Electro- Methods Eng., 65, pp. 863–881.
mechanics for MEMS: Applications Of Co-Solve-EM,” Proceedings of the [247] Liu, Y. J., 2008, “A Fast Multipole Boundary Element Method for 2–D Multi-
IEEE MEMS Conference, pp. 122–127. Domain Elastostatic Problems Based on a Dual BIE Formulation,” Comput.
[216] Aluru, N. R., and White, J., 1997, “An Efficient Numerical Technique for Mech., 42, pp. 761–773.
Electromechanical Simulation of Complicated Microelectromechanical [248] Liu, Y. J., Nishimura, N., and Otani, Y., 2005, “Large-Scale Modeling of
Structures,” Sens. Actuators, A, 58, pp. 1–11. Carbon-Nanotube Composites by the Boundary Element Method based on a
[217] Shi, F., Ramesh, P., and Mukherjee, S., 1996, “Dynamic Analysis of Micro- Rigid-Inclusion Model,” Comput. Mater. Sci., 34, pp. 173–187.
Electro-Mechanical Systems (MEMS),” Int. J. Numer. Methods Eng., 39, pp. [249] Liu, Y. J., Nishimura, N., Otani, Y., Takahashi, T., Chen, X. L., and Muna-
4119–4139. kata, H., 2005, “A Fast Boundary Element Method for the Analysis of Fiber-
[218] Ye, W., Mukherjee, S., and MacDonald, N. C., 1998, “Optimal Shape Design Reinforced Composites based on a Rigid-Inclusion Model,” J. Appl. Mech.,
of an Electrostatic Comb Drive in Microelectromechanical Systems,” ASME- 72, pp. 115–128.
IEEE J. Microelectromech. Syst., 7, pp.16–26. [250] Liu, Y. J., Nishimura, N., Qian, D., Adachi, N., Otani, Y., and Mokashi, V.,
[219] Ye, W., and Mukherjee, S., 1999, “Optimal Shape Design of Three- 2008, “A Boundary Element Method for the Analysis of CNT=Polymer Com-
Dimensional MEMS with Applications to Electrostatic Comb Drives,” Int. J. posites with a Cohesive Interface Model Based on Molecular Dynamics,” Eng.
Numer. Methods Eng., 45, pp. 175–194. Anal. Boundary Elem., 32, pp. 299–308.
[220] Li, G., and Aluru, N. R., 2003, “Efficient Mixed-Domain Analysis of Electro- [251] Popov, V., and Power, H., 2001, “An O(N) Taylor Series Multipole Boundary
static MEMS,” IEEE Trans. Comput.-Aided Des., 22, pp. 1228–1242. Element Method for Three-Dimensional Elasticity Problems,” Eng. Anal.
[221] Shrivastava, V., Aluru, N. R., and Mukherjee, S., 2004, “Numerical Analysis Boundary Elem., 25, pp. 7–18.
of 3D Electrostatics of Deformable Conductors using a Lagrangian [252] Wang, H. T. and Yao, Z. H., 2005, “A New Fast Multipole Boundary Element
Approach,” Eng. Anal. Boundary Elem., 28, pp. 583–591. Method for Large Scale Analysis of Mechanical Properties in 3D Particle-
[222] De, S. K., and Aluru, N. R., 2004, “Full-Lagrangian Schemes for Dynamic Reinforced Composites,” Comput. Model. Eng. Sci., 7, pp. 85–96.
Analysis of Electrostatic MEMS,” J. Microelectromech. Syst., 13, pp. 737. [253] Yoshida, K., Nishimura, N., and Kobayashi, S., 2001, “Application of Fast
[223] Bao, Z., and Mukherjee, S., 2004, “Electrostatic BEM for MEMS with Thin Multipole Galerkin Boundary Integral Equation Method to Crack Problems in
Conducting Plates and Shells,” Eng. Anal. Boundary Elem., 28, pp. 3D,” Int. J. Numer. Methods Eng., 50, pp. 525–547.
1427–1435. [254] Lai, Y.-S. and Rodin, G. J., 2003, “Fast Boundary Element Method for Three-
[224] Bao, Z., and Mukherjee, S., 2005, “Electrostatic BEM for MEMS with Thin Dimensional Solids Containing Many Cracks,” Eng. Anal. Boundary Elem.,
Beams,” Commun. Numer. Methods Eng., 21, pp. 297–312. 27, pp. 845–852.
[225] Telukunta, S., and Mukherjee, S., 2006, “Fully Lagrangian Modeling of [255] Gross, D., and Seelig, T., 2006, Fracture Mechanics with an Introduction to
MEMS with Thin Plates,” IEEE J. Microelectromech. Syst., 15, pp. 795–810. Micromechanics, Springer, The Netherlands.
[226] Ghosh, R., and Mukherjee, S., 2009, “Fully Lagrangian Modeling of Dynam- [256] Cheng, A. H. D., 1984, “Darcy Flow with Variable Permeability - A Boundary
ics of MEMS with Thin Beams - Part I: Undamped Vibrations,” ASME J. Integral Solution,” Water Resour. Res., 20, pp. 980–984.
Appl. Mech., 76, pp. 1–10. [257] Ang, W. T., Kusuma, J., and Clements, D. L., 1996, “A Boundary Element
[227] Ghosh, R., and Mukherjee, S., 2009, “Fully Lagrangian Modeling of Dynam- Method for a Second Order Elliptic Partial Differential Equation with Variable
ics of MEMS with Thin Beams - Part II: Damped Vibrations,” ASME. J. Coefficients,” Eng. Anal. Boundary Elem., 18, pp. 311–316.
Appl. Mech., 76, pp. 1–9. [258] Shaw, R. P. and Makris, N., 1992, “Green-Functions for Helmholtz and Lap-
[228] Liu, Y. J., and Shen, L., 2007, “A Dual BIE Approach for Large-Scale Model- lace Equations in Heterogeneous Media,” Eng. Anal. Boundary Elem., 10, pp.
ing of 3–D Electrostatic Problems with the Fast Multipole Boundary Element 179–183.
Method,” Int. J. Numer. Methods Eng., 71, pp. 837–855. [259] Shaw, R. P., 1994, “Green-Functions for Heterogeneous Media Potential Prob-
[229] Chen, H., and Mukherjee, S., 2006, “Charge Distribution on Thin Conducting lems,” Eng. Anal. Boundary Elem., 13, pp. 219–221.
Nanotubes - Reduced 3–D Model,” Int. J. Numer. Methods Eng., 68, pp. [260] Harrouni, K., Ouazar, D., Wrobel, L. C., and Cheng, A. H. D., 1995, “Global
503–524. Interpolation Function based DRBEM Applied to Darcy’s Flow in Heteroge-
[230] Chen, H., Mukherjee, S., and Aluru, N. R., 2008, “Charge Distribution on neous Media,” Eng. Anal. Boundary Elem., 16, pp. 281–285.
Thin Semiconducting Silicon Nanowires,” Comput. Methods Appl. Mech. [261] Divo, E., and Kassab, A. J., 1998, “Generalized Boundary Integral Equation
Eng., 197, pp. 3366–3377. for Heat Conduction in Non-Homogeneous Media: Recent Developments on
[231] Pozrikidis, C., 1992, Boundary Integral and Singularity Methods for Linear- the Sifting Property,” Eng. Anal Boundary Elem., 22, pp. 221–234.
ized Viscous Flow, Cambridge University Press, New York. [262] Gray, L. J., Kaplan, T., Richardson, J. D., and Paulino, G. H., 2003, “Green’s
[232] Karniadakis, G. E., and Beskok, A., 2002, Microflows: Fundamentals and Functions and Boundary Integral Analysis for Exponentially Graded Materi-
Simulation, Springer, New York. als: Heat Conduction,” Trans. ASME J. Appl. Mech., 70, pp. 543–549.
[233] Frangi, A., and Gioia, A. d., 2005, “Multipole BEM for the Evaluation of [263] Sutradhar, A., and Paulino, G. H., 2004, “A Simple Boundary Element Method
Damping Forces on MEMS,” Comput. Mech., 37, pp. 24–31. for Problems of Potential in Non-Homogeneous Media,” Int. J. Numer. Meth-
[234] Frangi, A., 2005, “A Fast Multipole Implementation of the Qualocation ods Eng., 60, pp. 2203–2230.
Mixed-Velocity-Traction Approach for Exterior Stokes Flows,” Eng. Anal. [264] Sutradhar, A., and Paulino, G. H., 2004, “The Simple Boundary Element
Boundary Elem., 29, pp. 1039–1046. Method for Transient Heat Conduction in Functionally Graded Materials,”
[235] Frangi, A., Spinola, G., and Vigna, B., 2006, “On the Evaluation of Damping Comput. Methods Appl. Mech. Eng., 193, pp. 4511–4539.
in MEMS in the Slip–Flow Regime,” Int. J. Numer. Methods Eng., 68, pp. [265] Crouch, S. L., 1976, “Solution of Plane Elasticity Problems by the Displace-
1031–1051. ment Discontinuity Method,” Int. J. Numer. Methods Eng., 10, pp. 301–343.
[236] Frangi, A., and Bonnet, M., 2010, “On the Fast Multipole Method for the [266] Crouch, S. L., and Starfield, A. M., 1983, Boundary Element Methods in Solid
Helmholtz Equation with Complex Frequency,” Comput. Model. Eng. Sci., Mechanics, George Allen & Unwin, London.
58, pp. 271–291. [267] Krishnasamy, G., Rizzo, F. J., and Liu, Y. J., 1994, “Boundary Integral Equa-
[237] Wang, X., White, J., Kanapka, J., Ye, W., and Aluru, N., 2006, “Algorithms in tions for Thin Bodies,” Int. J. Numer. Methods Eng., 37, pp. 107–121.
FastStokes and its Applications to Micromachined Device Simulation,,” [268] Krishnasamy, G, Rizzo, F. J., and Rudolphi, T. J., 1991, “Hypersingular
Design Automation Methods and Tools for Microfluidics-Based Biochips, Boundary Integral Equations: Their Occurrence, Interpretation, Regularization
K. Chakrabarty and J. Zeng, eds., Springer, New York, Chapter 4. and Computation,” Developments in Boundary Element Methods, P. K. Bane-
[238] Frangi, A., Ye, W., and White, J., 2008, “Evaluating Gas Damping in MEMS rjee and L. Morino, eds.,, Elsevier Applied Science Publishers, London, Chap-
using Fast Integral Equation Solvers,” Advances in Multiphysics Simulation ter 7.
and Experimental Testing of MEMS A. Frangi, N. Aluru, C. Cercignani, and [269] Krishnasamy, G., Rudolphi, T. J., Schmerr, L. W., and Rizzo, F. J., 1990,
S. Mukherjee, eds., Imperial College Press, London. “Hypersingular Boundary Integral Equations: Some Applications in Acoustic
[239] Liu, Y. J., 2008, “A New Fast Multipole Boundary Element Method for Solv- and Elastic Wave Scattering,” J. Appl. Mech., 57, pp. 404–414.
ing 2–D Stokes Flow Problems based on a Dual BIE Formulation,” Eng. Anal. [270] Hong H.-K., and Chen, J. T., 1988, “Derivations of Integral Equations of
Boundary Elem., 32, pp. 139–151. Elasticity,” J. Eng. Mech., 114, pp. 1028–1044.

031001-32 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 [271] Portela, A., Aliabadi, M. H., and Rooke, D. P., 1992, “The Dual Boundary [302] Greengard, L. F., Huang, J., Rokhlin, V., and Wandzura, S., “Accelerating
Element Method - Effective Implementation for Crack Problems,” Int. J. Fast Multipole Methods for the Helmholtz Equation at Low Frequencies,”
Numer. Methods Eng., 33, pp. 1269–1287. IEEE Comput. Sci. Eng., 5, pp. 32–38.
[272] Aliabadi, M. H., 1997, “Boundary Element Formulations in Fracture Mechan- [303] Tournour, M. A., and Atalla, N., 1999, “Efficient Evaluation of the Acoustic
ics,” Appl. Mech. Rev., 50, pp. 83–96. Radiation using Multipole Expansion,” Int. J. Numer. Methods Eng., 46, pp.
[273] Pan, E., 1999, “A BEM Analysis of Fracture Mechanics in 2D Anisotropic 825–837.
Piezoelectric Solids,” Eng. Anal. Boundary Elem., 23, pp. 67–76. [304] Gumerov, N. A., and Duraiswami, R., 2003, “Recursions for the Computation
[274] Liu, Y. J., and Fan, H., 2001, “On the Conventional Boundary Integral Equa- of Multipole Translation and Rotation Coefficients for the 3–D Helmholtz
tion Formulation for Piezoelectric Solids with Defects or of Thin Shapes,” Equation,” SIAM J. Sci. Comput., 25, pp. 1344–1381.
Eng. Anal. Boundary Elem., 25, pp. 77–91. [305] Darve, E., and Havé, P., 2004, “Efficient Fast Multipole Method for Low-
[275] Liu, Y. J., and Fan, H., 2002, “Analysis of Thin Piezoelectric Solids by the Frequency Scattering,” J. Comput. Phys., 197, pp. 341–363.
Boundary Element Method,” Comput. Methods Appl. Mech. Eng., 191, pp. [306] Fischer, M., Gauger, U., and Gaul, L., 2004, “A Multipole Galerkin Boundary
2297–2315. Element Method for Acoustics,” Eng. Anal. Boundary Elem., 28, pp. 155–162.
[276] Garcı́a-Sánchez, F., Sáez, A., and Domı́nguez, J., 2005, “Anisotropic and Pie- [307] Chen, J. T., and Chen, K. H., 2004, “Applications of the Dual Integral Formu-
zoelectric Materials Fracture Analysis by BEM,” Comput. Struct., 83, pp. lation in Conjunction with Fast Multipole Method in Large-Scale Problems for
804–820. 2D Exterior Acoustics,” Eng. Anal. Boundary Elem., 28, pp. 685–709.
[277] Garcı́a-Sánchez, F., Rojas-Dı́az, R., Sáez, A., and Zhang, C., 2007, “Fracture [308] Gumerov, N. A., and Duraiswami, R., 2004, Fast Multipole Methods for the
of Magnetoelectroelastic Composite Materials using Boundary Element Helmholtz Equation in Three Dimensions, Elsevier, Amsterdam.
Method (BEM),” Theor. Appl. Fract. Mech., 47, pp. 192–204. [309] Cheng, H., Crutchfield, W. Y., Gimbutas, Z., Greengard, L. F., Ethridge, J. F.,
[278] Qin, Q.-H., 2007, Green’s Function and Boundary Elements of Multifield Huang, J., Rokhlin, V., Yarvin, N., and Zhao, J., 2006, “A Wideband Fast
Materials, Elsevier, Amsterdam. Multipole Method for the Helmholtz Equation in Three Dimensions,” J. Com-
[279] Lauterbach, B., and Gross, D., 1998, “Crack Growth in Brittle Solids under put. Phys., 216, pp. 300–325.
Compression,” Mech. Mater., 29, pp. 81–92. [310] Liu, Y. J., and Bapat, M., 2009, “Fast Multipole Boundary Element Method
[280] Forth, S. C., and Staroselsky, A., 2005, “A Hybrid FEM=BEM Approach for for 3–D Full- and Half-Space Acoustic Wave Problems,” Proceedings of
Designing an Aircraft Engine Structural Health Monitoring,” Comput. Model. ASME 2009 IMECE, Lake Buena Vista, FL, IMECE2009–10165.
Eng. Sci., 9, pp. 287–298. [311] Brunner, D., Junge, M., and Gaul, L., 2009, “A Comparison of FE–BE Cou-
[281] Leonel, E. D., and Venturini, W. S., 2010, “Dual Boundary Element Formula- pling Schemes for Large-Scale Problems with Fluid–Structure Interaction,”
tion Applied to Analysis of Multi-Fractured Domains,” Eng. Anal. Boundary Int. J. Numer. Methods Eng., 77, pp. 664–688.
Elem., 34, pp. 1092–1099. [312] Chadwick, J., An, S., and James, D. L., 2009, “Harmonic Shells: A Practical
[282] Liu, Y. J., and Xu, N., 2000, “Modeling of Interface Cracks in Fiber- Nonlinear Sound Model for Near-Rigid Thin Shells,” ACM Transactions on
Reinforced Composites with the Presence of Interphases using the Boundary Graphics SIGGRAPH ASIA Conference Proceedings 2009, pp. 119:1–119:10.
Element Method,” Mech. Mater., 32, pp. 769–783. [313] Zheng, C., and James, D. L., 2010, “Rigid-Body Fracture Sound with Pre-
[283] Luo, J. F., Liu, Y. J., and Berger, E. J., 2000, “Interfacial Stress Analysis for Computed Soundbanks,” ACM Transactions on Graphics (SIGGRAPH, 2010,
Multi-Coating Systems using an Advanced Boundary Element Method,” Com- pp. 69:1–69:13.
put. Mech., 24, pp. 448–455. [314] Banjai, L., and Hackbusch, W., 2008, “Hierarchical Matrix Techniques for Low-
[284] Ohtsu, M., and Uddin, F. A. K. M., 2008, “Mechanisms of Corrosion-Induced and High-Frequency Helmholtz Problems,” IMA J. Numer. Anal., 28, pp. 46–79.
Cracks in Concrete at Meso- and Macro-Scales,” J. Adv. Concr. Technol., 6, [315] Brancati, A., Aliabadi, M. H., and Benedetti, I., 2009, “Hierarchical Adaptive Cross
pp. 419–429. Approximation GMRES Technique for Solution of Acoustic Problems using the
[285] Távara, L., Mantiĉ, V., Graciani, E., and Paris, F., 2011, “BEM Analysis of Boundary Element Method,” Comput. Model. Eng. Sci., 43, pp. 149–172.
Crack Onset and Propagation along Fibermatrix Interface under Transverse [316] Brunner, D., Junge, M., Rapp, P., Bebendorf, M., and Gaul, L., 2010,
Tension using a Linear Elastic Brittle Interface Model,” Eng. Anal. Boundary “Comparison of the Fast Multipole Method with Hierarchical Matrices for the
Elem., 35, pp. 207–222. Helmholtz-BEM,” Comput. Model. Eng. Sci., 58, pp. 131–158.
[286] Hackbusch, W., and Nowak, Z. P., 1989, “On the Fast Matrix Multiplication [317] Schenck, H. A., 1968, “Improved Integral Formulation for Acoustic Radiation
in the Boundary Element Method by Panel Clustering,” Numerische Mathe- Problems,” J. Acoust. Soc. Am., 44, pp. 41–58.
matik, 54, pp. 463–491. [318] Burton, A. J., and Miller, G. F., 1971, “The Application of Integral Equation
[287] Tyrtyshnikov, E., 1996, “Mosaic-Skeleton Approximations,” Calcolo, 33, pp. Methods to the Numerical Solution of Some Exterior Boundary-Value Prob-
46–57. lems,” Proc. R. Soc. London, Ser. A, 323, pp. 201–210.
[288] Yoshida, K., Nishimura, N., and Kobayashi, S., 2001, “Application of New [319] Kress, R., 1985, “Minimizing the Condition Number of Boundary Integral
Fast Multipole Boundary Integral Equation Method to Crack Problems in 3D,” Operators in Acoustic and Electromagnetic Scattering,” Quart. J. Mech. Appl.
Eng. Anal. Boundary Elem., 25, pp. 239–247. Math., 38, pp. 323–341.
[289] Greengard, L. F., and Rokhlin, V., 1997, “A New Version of the Fast Multi- [320] Seybert, A. F., Soenarko, B., Rizzo, F. J., and Shippy, D. J., 1985, “An
pole Method for the Laplace Equation in Three Dimensions,” Acta Numerica, Advanced Computational Method for Radiation and Scattering of Acoustic
6, pp. 229–269. Waves in Three Dimensions,” J. Acoust. Soc. Am., 77, pp. 362–368.
[290] Helsing, J., 1999, “Fast and Accurate Numerical Solution to an Elastostatic [321] Liu, Y. J., and Rizzo, F. J., 1992, “A Weakly-Singular Form of the Hypersin-
Problem Involving Ten Thousand Randomly Oriented Cracks,” Int. J. Fract., gular Boundary Integral Equation Applied to 3–D Acoustic Wave Problems,”
100, pp. 321–327. Comput. Methods Appl. Mech. Eng., 96, pp. 271–287.
[291] Wang, H., Yao, Z., and Wang, P., 2005, “On the Pre-Conditioners for Fast [322] Liu, Y. J., and Chen, S. H., 1999, “A New Form of the Hypersingular Bound-
Multipole Boundary Element Methods for 2D Multi-Domain Elastostatics,” ary Integral Equation for 3–D Acoustics and its Implementation with C0
Eng. Anal. Boundary Elem., 29, pp. 673–688. Boundary Elements,” Comput. Methods Appl. Mech. Eng., 173, pp. 375–386.
[292] Kolk, K., Weber, W., and Kuhn, G., 2005, “Investigation of 3D Crack [323] Liu, Y. J., and Rudolphi, T. J., 1999, “New Identities for Fundamental Solu-
Propagation Problems via Fast BEM Formulations,” Comput. Mech., 37, pp. tions and their Applications to Non-Singular Boundary Element For-
32–40. mulations,” Comput. Mech., 24, pp. 286–292.
[293] Weber, W., Kolk, K., and Kuhn, G., 2009, “Acceleration of 3D Crack Propa- [324] Kupradze, V. D., 1970, Three-Dimensional Problems of the Mathematical
gation Simulation by the Utilization of Fast BEM-Techniques,” Eng. Anal. Theory of Elasticity and Thermoelasticity, North-Holland, Amsterdam.
Boundary Elem., 33, pp. 1005–1015. [325] Kobayashi, S., 1987, “Elastodynamics,” Boundary Element Methods in
[294] Benedetti, I., Milazzo, A., and Aliabadi, M. H., 2009, “A Fast Dual Boundary Mechanics, North-Holland, Amsterdam, pp. 192–225.
Element Method for 3D Anisotropic Crack Problems,” Int. J. Numer. Methods [326] Chen, Y. H., Chew, W. C., and Zeroug, S., 1997, “Fast Multipole Method as
Eng., 80, pp. 1356–1378. an Efficient Solver for 2D Elastic Wave Surface Integral Equations,” Comput.
[295] Benedetti, I., and Aliabadi, M. H., 2010, “A Fast Hierarchical Dual Boundary Mech., 20, pp. 495–506.
Element Method for Three-Dimensional Elastodynamic Crack Problems,” Int. [327] Fukui, T., and Inoue, K., 1998, “Fast Multipole Boundary Element Method in
J. Numer. Methods Eng., 84, pp. 1038–1067. 2D Elastodynamics (in Japanese),” J. Appl. Mech. JSCE, 1, pp. 373–380.
[296] Benedetti, I., Aliabadi, M. H., and Milazzo, A., 2010, “A Fast BEM for the Available at: http://library.jsce.or.jp/jsce/open/00561/1998/01-0373.pdf.
Analysis of Damaged Structures with Bonded Piezoelectric Sensors,” Comput. [328] Fujiwara, H., 1998, “The Fast Multipole Method for Integral Equations of
Methods Appl. Mech. Eng., 199, pp. 490–501. Seismic Scattering Problems,” Geophys. J. Int., 133, pp. 773–782.
[297] Yoshida, K., Nishimura, N., and Kobayashi, S., 2001, “Application of New [329] Fujiwara, H., 2000, “The Fast Multipole Method for Solving Integral Equa-
Fast Multipole Boundary Integral Equation Method to Elastostatic Crack Prob- tions of Three-Dimensional Topography and Basin Problems,” Geophys. J.
lems in Three Dimensions,” J. Struct. Eng., 47A, pp. 169–179. Int., 140, pp. 198–210.
[298] Rokhlin, V., 1990, “Rapid Solution of Integral Equations Of Scattering Theory [330] Yoshida, K., 2001, “Applications of Fast Multipole Method to Boundary Inte-
in Two Dimensions,” J. Comput. Phys., 86, pp. 414–439. gral Equation Method,” Ph. D. thesis, Department of Global Environment En-
[299] Rokhlin, V., 1993, “Diagonal Forms of Translation Operators for the Helm- gineering, Kyoto University, Japan.
holtz Equation in Three Dimensions,” Appl. Comput. Harmon. Anal., 1, pp. [331] Yoshida, K., Nishimura, N., and Kobayashi, S., 2000, “Analysis of Three
82–93. Dimensional Scattering of Elastic Waves by a Crack with Fast Multipole
[300] Epton, M., and Dembart, B., 1995, “Multipole Translation Theory for the Boundary Integral Equation Method (in Japanese),” J. Appl. Mech. JSCE, 3,
Three Dimensional Laplace and Helmholtz Equations,” SIAM J. Sci. Comput. pp. 77–80. Available at: http://library.jsce.or.jp/jsce/open/00561/2000/03-0143.pdf.
(USA), 16, pp. 865–897. [332] Yoshida, K., Nishimura, N., and Kobayashi, S., 2001, “Applications of a Diag-
[301] Koc, S., and Chew, W. C., 1998, “Calculation of Acoustical Scattering from a onal Form Fast Multipole BIEM to the Analysis of Three Dimensional Scatter-
Cluster of Scatterers,” J. Acoust. Soc. Am., 103, pp. 721–734. ing of Elastic Waves by Cracks,” Trans. JASCOME, J. BEM, 18, pp. 77–80.

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-33

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 [333] Otani, Y., and Nishimura, N., 2003, “On the Improvement and Pre- [361] Ergül, Ö., and Gürel, L., 2009, “A Hierarchical Partitioning Strategy for an
Conditioning for FMM for Helmholtz’ Equation in 2D (in Japanese),” J. Appl. Efficient Parallelization of the Multilevel Fast Multipole Algorithm,” IEEE
Mech. JSCE, 6, pp. 283–292. Available at: http://library.jsce.or.jp/jsce/open/ Trans. Antennas Propag., 57, pp. 1740–1750.
00561/2003/06-0283.pdf. [362] Chew, W. C., 2004, “Computational Electromagnetics: The Physics of Smooth
[334] Chaillat, S., Bonnet, M., and Semblat, J. F., 2007, “A Fast Multipole Method Versus Oscillatory Fields,” Philos. Trans. R. Soc. London, Ser. A, 362, pp.
Formulation for 3D Elastodynamics in the Frequency Domain,” C. R. Mecani- 579–602.
que, 335, pp. 714–719. [363] Chew, W. C., Song, J. M., and Cui, T. J., 2004, “Review of Large Scale Com-
[335] Chaillat, S., Bonnet, M., and Semblat, J. F., 2008, “A Multi-Level Fast Multi- puting in Electromagnetics with Fast Integral Equation Solvers,” Comput.
pole BEM for 3–D Elastodynamics in the Frequency Domain,” Comput. Meth- Model. Eng. Sci., 5, pp. 361–372.
ods Appl. Mech. Eng., 49–50, pp. 4233–4249. [364] Chew, W. C., Chao, H. Y., Cui, T. J., Lu, C. C., Ohnuki, S., Pan, Y. C., Song,
[336] Sanz, J. A., Bonnet, M., and Dominguez, J., 2008, “Fast Multipole Method J. M., and Zhao, J. S., 2003, “Fast Integral Equation Solvers in Computational
Applied to 3–D Frequency Domain Elastodynamics,” Comput. Mech., 42, pp. Electromagnetics of Complex Structures,” Eng. Anal. Boundary Elem., 27, pp.
787–802. 803–823.
[337] Tong, M. S., and Chew, W. C., 2009, “Multilevel Fast Multipole Algorithm [365] Vikram, M., and Shanker, B., 2009, “An Incomplete Review of Fast Multipole
for Elastic Wave Scattering by Large Three-Dimensional Objects,” J. Comput. Methods-From Static to Wideband-As Applied to Problems in Computational
Phys., 228, pp. 921–932. Electromagnetics,” Appl. Comput. Electromagn. Soc. J., 24, pp. 79–108.
[338] Chaillat, S., Bonnet, M., and Semblat, J. F., 2009, “A New Fast Multi-Domain Available at: https://www.egr.msu.edu/ece/technical_papers/PublicViewPaper.
BEM to Model Seismic Wave Propagation and Amplification in 3–D Geologi- php?SubmissionNumber=MSU-ECE-2008-6.
cal Structures,” Geophys. J. Int., 177, pp. 509–531. [366] Joannopoulos, J. D., Meade, R. D., and Winn, J. N., 1995, Photonic Crystals,
[339] Isakari, H., Yoshikawa, H., and Nishimura, N., 2010, “A Periodic FMM for Princeton University Press, Princeton, NJ.
Elastodynamics in 3D and its Applications to Problems Related to Waves [367] Pendry, J. B., 2000, “Negative Refraction makes a Perfect Lens,” Phys. Rev.
Scattered by a Doubly Periodic Layer of Scatterers (in Japanese),” J. Appl. Lett., 85, pp. 3966–3969.
Mech. JSCE, 13, pp. 169–178. [368] Otani, Y., and Nishimura, N., 2008, “A Periodic FMM for Maxwell’s Equa-
[340] Isakari, H., Niino, K., Yoshikawa, H., and Nishimura, N., 2011, “Calderon tions in 3D and its Applications to Problems Related to Photonic Crystals,” J.
Preconditioning for Periodic FMM for Elastodynamics in 3D,” Int. J. Numer. Comput. Phys., 227, pp. 4630–4652.
Methods Eng., (in press). [369] Otani, Y., and Nishimura, N., 2009, “An FMM for Orthotropic Periodic
[341] Takahashi, T., Nishimura, N., and Kobayashi, S., 2003, “A Fast BIEM for Boundary Value Problems for Maxwell’s Equations,” Waves Random Com-
Three-Dimensional Elastodynamics in Time Domain,” Eng. Anal. Boundary plex Media, 19, pp. 80–104.
Elem., 27, pp. 491–506. [370] Otani, Y., and Nishimura, N., 2009, “Extension of the Periodic FMM in Max-
[342] Ergin, A. A., Shanker, B., and Michielssen, E., 1998, “Fast Evaluation of well’s Equations to Tall Cell Problems (in Japanese),” Trans. JASCOME, 9,
Three-Dimensional Transient Wave Fields using Diagonal Translation Oper- pp. 55–60. Available at: http://gspsun1.gee.kyoto-u.ac.jp/JASCOME/denshi-
ators,” J. Comput. Phys., 146, pp. 157–180. journal/9/JA0911.pdf.
[343] Otani, Y., Takahashi, T., and Nishimura, N., 2007, “A Fast Boundary Integral [371] Otani, Y., and Nishimura, N., 2009, “Behaviour of Periodic Fast Multipole
Equation Method for Elastodynamics in Time Domain and its Parallelisation,” Boundary Integral Equation Method for Maxwell’s Equations Near Wood’s
Boundary Element Analysis: Mathematical Aspects and Applications, Anomalies,” Imaging Microstructures, Contemporary Mathematics, H.
M. Schanz and O. Steinbach, eds., Springer-Verlag, Berlin-Heidelberg, Ger- Ammari and H. B. Kang, eds., AMS, Providence,pp. 43–59.
many, pp. 161–185 [372] Barnett, A., and Greengard, L. F., 2010, “A New Integral Representation for
[344] Yoshikawa, H., Otani, Y., and Nishimura, N., 2007, “A Large Scale Wave Quasi-Periodic Fields and its Application to Two-Dimensional Band Structure
Analysis Related to Laser-Ultrasonic NDE with a Fast Multipole BIEM in Calculations,” J. Comput. Phys., 229, pp. 6898–6914.
Time Domain (in Japanese),” Trans. JASCOME, 7, pp. 79–84. Available at: [373] Barnett, A., and Greengard, L. F., 2010, “A New Integral Representation for
http://gspsun1.gee.kyoto-u.ac.jp/JASCOME/denshi-journal/7-1/17-070629.pdf Quasi-Periodic Scattering Problems in Two Dimensions,” BIT Numer. Math.,
[345] Saitoh, T., Ishida, T., Fukui, T., and Hirose, S., 2008, “A New Time-Domain 51, pp. 67–90.
Boundary Element Method using the Operational Quadrature Method and the [374] Houzaki, K., Nishimura, N., and Otani, Y., 2006, “An FMM for Periodic
Fast Multiple Method in 2–D Elastodynamics (in Japanese),” J. Appl. Mech. Rigid-Inclusion Problems and its Application to Homogenisation,” Inverse
JSCE, 11, pp. 193–200. Available at: http://library.jsce.or.jp/jsce/open/00561/ Problems, Multi-Scale Analysis and Effective Medium Theory, Contemporary
2008/11-0193.pdf. Mathematics 408, AMS, Providence, pp. 81–98.
[346] Soares, D., Jr., and Mansur, W. J., 2007, “An Efficient Stabilized Boundary [375] Otani, Y., and Nishimura, N., 2006, “A Fast Multipole Boundary Integral
Element Formulation for 2D Time-Domain Acoustics and Elastodynamics,” Equation Method for Periodic Boundary Value Problems in Three Dimen-
Comput. Mech., 40, pp. 355–365. sional Elastostatics and its Application to Homogenisation,” Int. J. Comp.
[347] Panagiotopoulos, C. G., and Manolis, G. D., 2011, “Three-Dimensional BEM Eng., 4, pp. 487–500.
for Transient Elastodynamics based on the Velocity Reciprocal Theorem,” [376] Otani, Y., and Nishimura, N., 2008, “An FMM for Periodic Boundary Value
Eng. Anal. Boundary Elem., 35, pp. 507–516. Problems for Cracks for Helmholtz’ Equation in 2D,” Int. J. Numer. Methods
[348] Rynne, B. P., and Smith, P. D., 1990, “Stability of Time Marching Algorithms Eng., 73, pp. 381–406.
for the Electric Field Integral Equation,” J. Electromagn. Waves Appl., 4, pp. [377] Isakari, H., Niino, K., Yoshikawa, H., and Nishimura, N., 2010, “Preconditioning
1181–1205. Based on Calderon’s Formulae for the Periodic FMM for Elastodynamics in 3D
[349] Ha-Duong, T., Ludwig, B., and Terrasse, I., 2003, “A Galerkin BEM for Tran- (in Japanese),” Trans. JASCOME, 10, pp. 45–50. Available at: http://gspsun1.
sient Acoustic Scattering by an Absorbing Obstacle,” Int. J. Numer. Methods gee.kyoto-u.ac.jp/JASCOME/denshi-journal/10/JA109.pdf.
Eng., 57, pp. 1845–1882. [378] Costabel, M., 2005, “Time-Dependent Problems with the Boundary Integral
[350] Kielhorn, L., and Schanz, M., 2008, “Convolution Quadrature Method-Based Equation Method,” Encyclopedia of Computational Mechanics, E. Stein, R. de
Symmetric Galerkin Boundary Element Method for 3–D Elastodynamics,” Int. Borst, and T. J. R. Hughes, eds., John Wiley & Sons, New York; Chichester,
J. Numer. Methods Eng., 76, pp. 1724–1746. Weinheim.
[351] Bonnet, M., Burczynski, T., and Nowakowski, M., 2002, “Sensitivity Analysis [379] Dominguez, J., 1993, Boundary Elements in Dynamics, Computational
for Shape Perturbation of Cavity or Internal Crack using BIE and Adjoint Mechanics Publication, Southampton.
Variable Approach,” Int. J. Solids Struct., 39, pp. 2365–2385. [380] Yoshikawa, H., and Nishimura, N., 2003, “An Improved Implementation of
[352] Bonnet, M., and Guzina, B. B., 2009, “Elastic-Wave Identification of Penetra- Time Domain Elastodynamic BIEM in 3D for Large Scale Problems and its
ble Obstacles using Shape-Material Sensitivity Framework,” J. Comput. Phys., Application to Ultrasonic NDE,” Electron. J. Boundary Elem., 1, pp. 201–217.
228, pp. 294–311. [381] Narayanan, G. V., and Beskos, D. E., 1982, “Numerical Operational Methods
[353] Bonnet, M., and Guzina, B. B., 2004, “Sounding of Finite Solid Bodies by way of for Time-Dependent Linear Problems,” Int. J. Numer. Methods Eng., 18, pp.
Topological Derivative,” Int. J. Numer. Methods Eng., 61, pp. 2344–2373. 1829–1854.
[354] Fata, S. N., and Guzina, B. B., 2004, “A Linear Sampling Method for Near- [382] Gaul, L., and Schanz, M., 1999, “A Comparative Study of Three Boundary Ele-
Field Inverse Problems in Elastodynamics,” Inverse Probl., 20, pp. 713–736. ment Approaches to Calculate the Transient Response of Viscoelastic Solids with
[355] Fata, S. N., and Guzina, B. B., 2007, “Elastic Scatterer Reconstruction via the Unbounded Domains,” Comput. Methods Appl. Mech. Eng., 179, pp. 111–123.
Adjoint Sampling Method,” SIAM J. Appl. Math., 67, pp. 1330–1352. [383] Beskos, D. E., 1987, “Boundary Element Methods in Dynamic Analysis,”
[356] Chew, W. C., Jin, J.-M., Michielssen, E., and Song, J., 2001, Fast and Appl. Mech. Rev., 40, pp. 1–23.
Efficient Algorithms in Computational Electromagnetics, Artech House, [384] Beskos, D. E., 1997, “Boundary Element Methods in Dynamic Analysis: Part
Norwood, MA. II (1986–1996),” Appl. Mech. Rev., 50, pp. 149–197.
[357] Rao, S. M., Wilton, D. R., and Glisson, A. W., 1982, “Electromagnetic Scattering [385] Mansur, W. J., 1983, “A Time-Stepping Technique to Solve Wave Propaga-
by Surfaces of Arbitrary Shape,” IEEE Trans.Antennas Propag., 30, pp. 409–418. tion Problems Using the Boundary Element Method,” Ph. D. Thesis, Univer-
[358] Tong, M. S., Chew, W. C., Rubin, B. J., Morsey, J. D., and Jiang, L., 2009, sity of Southampton.
“On the Dual Basis for Solving Electromagnetic Surface Integral Equations,” [386] Lubich, C., 1988, “Convolution Quadrature and Discretized Operational Cal-
IEEE Trans.Antennas Propag., 57, pp. 3136–3146. culus. I=II.,” Numerische Mathematik, 52, pp. 129–145=413–425.
[359] Buffa, A., and Christiansen, S. H., 2007, “A Dual Finite Element Complex on [387] Schanz, M., and Antes, H., 1997, “Application of ‘Operational Quadrature Meth-
the Barycentric Refinement,” Math. Comput., 76, pp. 1743–1769. ods’ in Time Domain Boundary Element Methods,” Mechanica, 32, pp. 179–186.
[360] Taboada, J. M., Landesa, L., Obelleiro, F., Rodriguez, J. L., Bertolo, J. M., [388] Schanz, M., 2001, “Application of 3–D Boundary Element Formulation to Wave
Araujo, M. G., Mour1no, J. C., and Gomez, A., 2009, “High Scalability FMM- Propagation in Poroelastic Solids,” Eng. Anal. Boundary Elem., 25, pp. 363–376.
FFT Electromagnetic Solver for Supercomputer Systems,” IEEE Antennas [389] Banjai, L., and Schanz, M., 2011, “Wave Propagation Problems Treated with
Propag. Mag., 51, pp. 20–28. Convolution Quadrature and BEM,” Fast Boundary Element Methods in

031001-34 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Engineering and Industrial Applications, U. Langer, M. Schanz, O. Steinbach, [419] Elleithy, W. M., and Tanaka, M., 2003, “Interface Relaxation Algorithms for
and W. L. Wendland, eds.,Springer, New York. BEM-BEM Coupling and FEM-BEM Coupling,” Comput. Methods Appl.
[390] Schanz, M., and Antes, H., 1997, “A New Visco- and Elastodynamic Time Mech. Eng., 192, pp. 2977–2992.
Domain Boundary Element Formulation,” Comput. Mech., 20, pp. 452–459. [420] Elleithy, W. M., Tanaka, M., and Guzik, A., 2004, “Interface Relaxation
[391] Schanz, M., 2001, Wave Propagation in Viscoelastic and Poroelastic Con- FEM-BEM Coupling Method for Elasto-Plastic Analysis,” Eng. Anal. Bound-
tinua: A Boundary Element Approach, Springer-Verlag, Berlin, Heidelberg, ary Elem., 28, pp. 849–857.
Germany. [421] Soares, D., Jr., von Estorff, O., and Mansur, W. J., 2004, “Iterative Coupling
[392] Birgisson, B., Siebrits, E., and Peirce, A. P., 1999, “Elastodynamic Direct of BEM and FEM for Nonlinear Dynamic Analyses,” Comput. Mech., 34, pp.
Boundary Element Methods with Enhanced Numerical Stability Properties,” 67–73.
Int. J. Numer. Methods Eng., 46, pp. 871–888. [422] Langer, U., and Steinbach, O., 2003, “Boundary Element Tearing and Inter-
[393] Frangi, A., 1999, “Elastodynamics by BEM: a New Direct Formulation,” Int. connecting Methods,” Comput. Visualization Sci., 71, pp. 205–228.
J. Numer. Methods Eng., 45, pp. 721–740. [423] Rüberg, T., and Schanz, M., 2008, “Coupling Finite and Boundary Element
[394] Frangi, A., and Novati, G., 1999, “On the Numerical Stability of Time- Methods for Static and Dynamic Elastic Problems with Non-Conforming
Domain Elastodynamic Analyses by BEM,” Comput. Methods Appl. Mech. Interfaces,” Comput. Methods Appl. Mech. Eng., 198, pp. 449–458.
Eng., 173, pp. 403–417. [424] Langer, U., and Pechstein, C., 2006, “Coupled Finite and Boundary Element
[395] Frangi, A., 2000, “‘Causal’ Shape Functions in the Time Domain Boundary Tearing and Interconnecting Solvers for Nonlinear Problems,” ZAMM, 86, pp.
Element Method,” Comput. Mech., 25, pp. 533–541. 915–931.
[396] Ha Duong, T., 1990, “On the Transient Acoustic Scattering by a Flat Object,” [425] Fischer, M., and Gaul, L., 2005, “Fast BEM-FEM Mortar Coupling for
Jpn. J. Appl. Math., 7, pp. 489–513. Acoustic-Structure Interaction,” Int. J. Numer. Methods Eng., 62, pp.
[397] Aimi, A., Diligenti, M., and Guardasoni, C., 2011, “On the Energetic Galerkin 1677–1690.
Boundary Element Method Applied to Interior Wave Propagation Problems,” [426] Frangi, A., and Novati, G., 2003, “BEM-FEM Coupling for 3D Fracture
J. Comp. Appl. Math., 235, pp. 1746–1754. Mechanics Applications,” Comput. Mech., 32, pp. 415–422.
[398] Aimi, A., Diligenti, M., Guardasoni, C., Mazzieri, I., and Panizzi, S., 2009, [427] Aour, B., Rahmani, O., and Nait-Abdelaziz, M., 2007, “A Coupled
“An Energy Approach to Space-Time Galerkin BEM for Wave Propagation FEM=BEM Approach and its Accuracy for Solving Crack Problems in Frac-
Problems,” Int. J. Numer. Methods Eng., 80, pp. 1196–1240. ture Mechanics,” Int. J. Solids Struct., 44, pp. 2523–2539.
[399] Hackbusch, W., Kress, W., and Sauter, S. A., 2007, “Sparse Convolution [428] Lucht, T., 2009, “Finite Element Analysis of Three Dimensional Crack
Quadrature for Time Domain Boundary Integral Formulations of the Wave Growth by the use of a Boundary Element Sub Model,” Eng. Fract. Mech., 76,
Equation by Cutoff and Panel-Clustering,” Boundary Element Analysis: Math- pp. 2148–2162.
ematical Aspects and Applications, M. Schanz and O. Steinbach, eds.,, [429] Elleithy, W. M., 2008, “Analysis of Problems in Elasto-Plasticity via an
Springer-Verlag, Berlin-Heidelberg, Germany, pp. 113–134. Adaptive FEM-BEM Coupling Method,” Comput. Methods Appl. Mech. Eng.
[400] Kress, W., and Sauter, S., 2008, “Numerical Treatment of Retarded Boundary 197, pp. 3687–370.
Integral Equations by Sparse Panel Clustering,” IMA J. Numer. Anal., 28, pp. [430] Elleithy, W. M., and Grzhibovskis, R., 2009, “An Adaptive Domain Decom-
162–185. position Coupled Finite Element-Boundary Element Method for Solving Prob-
[401] Tausch, J., 2007, “A Fast Method for Solving the Heat Equation by Layer lems in Elasto-Plasticity,” Int. J. Numer. Methods Eng., 79, pp. 1019–1040.
Potentials,” J. Comput. Phys., 224, pp. 956–969. [431] von Estorff, O., and Firuziaan, M., 2000, “Coupled BEM=FEM Approach for
[402] Zhu, X. Y., Chen, W. Q., Huang, Z. Y., and Liu, Y. J., 2011, “A Fast Multi- Nonlinear Soil=Structure Interaction,” Eng. Anal. Boundary Elem., 24, pp.
pole Boundary Element Method for 2D Viscoelastic Problems,” Eng. Anal. 715–725.
Boundary Elem., 35, pp. 170–178. [432] Rizos, D. C., and Wang, Z., 2002, “Coupled BEM-FEM Solutions for Direct
[403] Banjai, L., and Sauter, S., 2009, “Rapid Solution of the Wave Equation in Time Domain Soil-Structure Interaction Analysis,” Eng. Anal. Boundary
Unbounded Domains,” SIAM J. Numer. Anal., 47, pp. 227–249. Elem., 26, pp. 877–888.
[404] Schanz, M., 2010, “On a Reformulated Convolution Quadrature based Bound- [433] Almeida, V. S., and De Paiva, J. B., 2004, “A Mixed BEM-FEM Formulation
ary Element Method,” Comput. Model. Eng. Sci., 58, pp. 109–128. for Layered Soil-Superstructure Interaction,” Eng. Anal. Boundary Elem., 28,
[405] Saitoh, T., Hirose, S., Fukui, T., and Ishida, T., 2007, “Development of a pp. 1111–1121.
Time-Domain Fast Multipole BEM based on the Operational Quadrature [434] Padrón, L. A., Aznárez, J. J., and Maeso, O., 2008, “Dynamic Analysis of
Method in a Wave Propagation Problem,” Advances in Boundary Element Piled Foundations in Stratified Soils by a BEM-FEM Model,” Soil Dyn. Earth-
Techniques VIII, V. Minutolo and M. H. Aliabadi, eds., EC Ltd. Eastleigh, pp. quake Eng., 28, pp. 333–346.
355–360. [435] Padrón, L. A., Aznárez, J. J., and Maeso, O., 2009, “Dynamic Structure-Soil-
[406] Feng, H., Kaganovskiy, L., and Krasny, R., 2009, “Azimuthal Instability of a Structure Interaction between Nearby Piled Buildings under Seismic Excita-
Vortex Ring Computed by a Vortex Sheet Panel Method,” Fluid Dyn. Res., tion by BEM-FEM Model,” Soil Dyn. Earthquake Eng., 29, pp. 1084–1096.
41, 051405. [436] Millán, M. A., and Domı́nguez, J., 2009, “Simplified BEM=FEM Model for
[407] Rahimian, A., Lashuk, I., Veerapaneni, S. K., Aparna, C., Malhotra, D., Dynamic Analysis of Structures on Piles and Pile Groups in Viscoelastic and
Moon, L., Sampath, R., Shringarpure, A., Vetter, J., Vuduc, R., Zorin, C., and Poroelastic Soils,” Eng. Anal. Boundary Elem., 33, pp. 25–34.
Biros, G., 2010, “Petascale Direct Numerical Simulation of Blood Flow on [437] O’Brien, J., and Rizos, D. C., 2005, “A 3D BEM-FEM Methodology for Simu-
200 K Cores and Heterogeneous Architectures,” ACM=IEEE SC Conference lation of High Speed Train Induced Vibrations,” Soil Dyn. Earthquake Eng.,
Series, pp. 1–11. 25, pp. 289–301.
[408] Zienkiewicz, O. C., Kelly, D. W., and Bettes, P., 1977, “The Coupling of the [438] Galvı́n, P., and Domı́nguez, J., 2009, “Experimental and Numerical Analyses
Finite Elements and Boundary Solution Procedures,” Int. J. Numer. Methods of Vibrations Induced by High-Speed Trains on the Córdoba-Málaga Line,”
Eng., 11, pp. 355–375. Soil Dyn. Earthquake Eng., 29, pp. 641–657.
[409] Costabel, M., and Stephan, E. P., 1990, “Coupling of Finite and Boundary Ele- [439] Galvı́n, P., François, S., Schevenels, M., Bongini, E., Degrande, G., and Lom-
ment Methods for an Elastoplastic Interface Problem,” SIAM J. Numer. Anal., baert, A., 2010, “A 2.5D Coupled FE-BE Model for the Prediction of Railway
27, pp. 1212–1226. Induced Vibrations,” Soil Dyn. Earthquake Eng., 30, pp. 1500–1512.
[410] Polizzotto, C., and Zito, M., 1994, “Variational Formulations for Coupled [440] Kuhn, M., 1998, “The Application of Coupled BE=FE Formulations in Tech-
BE=FE Methods in Elastostatics,” ZAMM, 74, pp. 533–543. nical Magnetic Field Computations,” Comput. Methods Appl. Mech. Eng.,
[411] Haas, M., and Kuhn, G., 2003, “Mixed-Dimensional, Symmetric Coupling of 157, pp. 193–204.
FEM and BEM,” Eng. Anal. Boundary Elem., 27, pp. 575–582. [441] Balac, S., and Caloz, G., 2002, “Magnetostatic Field Computations based on
[412] Ganguly, S., Layton, J. B., and Balakrishna, C., 2000, “Symmetric Coupling the Coupling of Finite Element and Integral Representation Methods,” IEEE
of Multi-Zone Curved Galerkin Boundary Elements with Finite Elements in Trans. Magn., 38, pp. 393–396.
Elasticity,” Int. J. Numer. Methods Eng., 48, pp. 633–654. [442] Kuhn, M., and Steinbach, O., 2002, “Symmetric Coupling of Finite and
[413] Ganguly, S., Layton, J. B., and Balakrishna, C., 2004, “A Coupling of Multi- Boundary Elements for Exterior Magnetic Field Problems,” Math. Models
Zone Curved Galerkin BEM with Finite Elements for Independently Modelled Meth. Appl. Sci., 25, pp. 357–371.
Sub-Domains with Non-Matching Nodes In Elasticity,” Int. J. Numer. Meth- [443] Frangi, A., Faure-Ragani, P., and Ghezzi, L., 2005, “Magneto-Mechanical
ods Eng., 59, pp. 1021–1038. Simulations by a Coupled Fast Multipole Method - Finite Element Method
[414] Keat, W. D., Annigeri, B. S., and Cleary, M. P., 1988, “Surface Integral and and Multigrid Solvers,” Comput. Struct., 83, pp. 718–726.
Finite Element Hybrid Method for Two and Three Dimensional Fracture [444] Frangi, A., Ghezzi, L., and Faure-Ragani, P., 2006, “Accurate Force Evalua-
Mechanics Analysis,” Int. J. Fract., 36, pp. 35–53. tion for Industrial Magnetostatics Applications with Fast BEM-FEM
[415] Han, Z. D., and Atluri, S. N., 2002, “SGBEM (for Cracked Local Subdomain) Approaches,” Comput. Model. Eng. Sci., 15, pp. 41–48.
- FEM (for Uncracked Global Structure) Alternating Method for Analyzing [445] Salgado, P., and Selgas, V., 2008, “A Symmetric BEM-FEM Coupling for the
3D Surface Cracks and their Fatigue-Growth,” Comput. Model. Eng. Sci., 3, Three-Dimensional Magnetostatic Problem using Scalar Potentials,” Eng.
pp. 699–716. Anal. Boundary Elem., 32, pp. 633–644.
[416] Springhetti, R., Novati, G., and Margonari, M., 2006, “Weak Coupling of the [446] Pusch, D., and Ostrowski, J., 2010, “Robust FEM=BEM Coupling for Magne-
Symmetric Galerkin BEM with FEM for Potential and Elastostatic Problems,” tostatics on Multiconnected Domains,” IEEE Trans. Magn., 46, pp. 3177.
Comput. Model. Eng. Sci., 13, pp. 67–80. [447] Kurz, S., Fetzer, J., and Lehner, G., 1998, “A Novel Formulation for 3D Eddy
[417] von Estorff, O., and Hagen, C., 2005, “Iterative Coupling of FEM and BEM in Current Problems with Moving Bodies Sing a Lagrangian Description and
3D Transient Elastodynamics,” Eng. Anal. Boundary Elem., 29, pp. 775–787. BEM-FEM Coupling,” IEEE Trans. Magn., 34, pp. 3068–3073.
[418] El-Gebeily, M., Elleithy, W. M., and Al-Gahtani, H. J., 2002, “Convergence [448] Chen, Z. S., Hofstetter, G., and Mang, H. A., 1998, “A Galerkin-Type BE-FE
of the Domain Decomposition Finite Element-Boundary Element Coupling Formulation for Elasto-Acoustic Coupling,” Comput. Methods Appl. Mech.
Methods,” Comput. Methods Appl. Mech. Eng., 191, pp. 4851–4868. Eng., 152, pp. 147–155.

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-35

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 [449] Czygan, O., and von Estorff, O., 2002, “Fluid-Structure Interaction by Coupling [457] Ye, T. Q., and Liu, Y. J., 1985, “Finite Deflection Analysis of Elastic Plate by
BEM and Nonlinear FEM,” Eng. Anal. Boundary Elem., 26, pp. 773–779. the Boundary Element Method,” Appl. Math. Model., 9, pp. 183–188.
[450] Gaul, L., and Wenzel, W., 2002, “A Coupled Symmetric BE-FE Method for Acous- [458] Wen, P. H., Aliabadi, M. H., and Young, A., 2005, “Large Deflection
tic Fluid-Structure Interaction,” Eng. Anal. Boundary Elem., 26, pp. 629–636. Analysis of Reissner Plate by Boundary Element Method,” Comput.
[451] Langer, S., and Antes, H., 2003, “Analyses of Sound Transmission through Struct., 83, pp. 870–879.
Windows by Coupled Finite and Boundary Element Methods,” Acta Acous- [459] Noblesse, F., Delhommeau, G., and Yang, C., 2010, “Ship Bow Wave and
tica, 89, pp. 78–85. Green Function Method,” 12es Journées de l’Hydrodynamique, Nantes,
[452] Soares, D., Jr., 2009, “Acoustic Modelling by BEM-FEM Coupling Proce- France, pp. 1–12.
dures: Taking into Account Explicit and Implicit Multi-Domain Decomposi- [460] Xiao, J. X., and Tausch, J., 2010, “A Fast Wavelet-Multipole Method for
tion Techniques,” Int. J. Numer. Methods Eng., 78, pp. 1076–1093. Direct BEM,” Eng. Anal. Boundary Elem., 34, pp. 673–679.
[453] Wang, S. Y., Chen, P. Q., Lim, K. M., and Khoo, B. C., 2010, “A BEM=FEM [461] Xiao, J. Y., Wen, L. H., and Tausch, J., 2009, “On Fast Matrix-Vector Multi-
Coupling Approach for Fluid-Structure Interaction Simulation of Cell plication in Wavelet Galerkin BEM,” Eng. Anal. Boundary Elem., 33, pp.
Motion,” Commun. Comput. Phys., 7, pp. 994–1026. 159–167.
[454] Seghir, A., Tahakourt, A., and Bonnet, G., 2009, “Coupling FEM and Sym- [462] Fong, W., and Darve, E., 2009, “The Black-Box Fast Multipole Method,” J.
metric BEM for Dynamic Interaction of Dam-Reservoir Systems,” Eng. Anal. Comput. Phys., 228, pp. 8712–8725.
Boundary Elem., 33, pp. 201–210. [463] Chen, X. L., Zhang, H., and Wu, M., 2010, “Modeling Brain Electrical Activ-
[455] Elisee, J. P., Gibson, A., and Arridge, S., 2010, “Combination of Boundary ity by an Image-Based Boundary Element Method,” Eng. Anal. Boundary
Element Method and Finite Element Method in Diffuse Optical Tomography,” Elem., 34, pp. 995–1002.
IEEE Trans. Biomed. Eng., 57, pp. 2737. [464] Peng, Z. L., Asaro, R. J., and Zhu, Q., 2010, “Multiscale Simulation of Eryth-
[456] Jerez-Hanckes, C., Durán, M., and Guarini, M., 2010, “Hybrid FEM=BEM rocyte Membranes,” Phys. Rev. E, 81, pp. 031904(1-11).
Modeling of Finite-Sized Photonic Crystals for Semiconductor Laser Beams,” [465] The University of Mississippi, 2003–2011,“Boundary Element Resources Net-
Int. J. Numer. Methods Eng., 82, pp. 1308–1340. work (BENET),” http://www.olemiss.edu/sciencenet/benet/

Yijun Liu is a Professor of Mechanical Engineering at the University of Cincinnati. He received his BS and
MS degrees in Aerospace Engineering from Northwestern Polytechnical University in China in 1982 and
1984, respectively, and his Ph. D. degree in Theoretical and Applied Mechanics from the University of Illi-
nois at Urbana-Champaign in 1992. He conducted post-doctoral research at Iowa State University for two
years and worked at Ford Motor Company as a CAE Analyst for about two years. He has been on the faculty
of the University of Cincinnati since 1996. His research interests are in computational mechanics, BEM,
modeling of composites, MEMS, fracture mechanics, and wave problems. He has authored two books and
over 90 papers in journals and conference proceedings, and edited four special journal issues. His book Fast
Multipole Boundary Element Method – Theory and Applications in Engineering was published by Cambridge
University Press in 2009.

Subrata Mukherjee obtained his BS degree in Mechanical Engineering from the Indian Institute of Technol-
ogy at Kharagpur, India, in 1967, and his Ph. D. in Applied Mechanics from Stanford University in 1972. He
joined the Cornell faculty in 1974, and has been a Professor since 1986. He was a visiting professor at the
University of Waterloo, the Technische Hochschule in Darmstadt, the University of Arizona in Tucson, the
Ecole Polytechnique in Palaiseau, France, and the Rensselaer Polytechnic in Troy. He is a Fellow of
the ASME and of the American Academy of Mechanics, and a member of several other professional societies.
His primary field of research is computational mechanics, with applications of the BEM and related meshfree
methods in solid mechanics, manufacturing processes, shape optimization and MEMS=NEMS. He has auth-
ored or co-authored three books, edited or co-edited four books, written 12 invited review articles and over
190 articles in archival journals. He was an Associate Editor of the ASME Journal of Applied Mechanics dur-
ing 2004–2010.

Naoshi Nishimura is a Professor of Informatics at Kyoto University. He received B Eng, M Eng, and Dr Eng
degrees in Civil Engineering from Kyoto University in 1977, 1979, and 1988, respectively. He joined the fac-
ulty of the Department of Civil Engineering at Kyoto University in 1979, moved to the Department of Global
Environment Engineering in 1991, and returned to the Department of Civil Engineering in 2002. He then
moved to Academic Center of Computing and Media Studies in 2002 and then to the Graduate School of
Informatics in 2006. His collaborations with other institutions include his stays in the Department of Civil En-
gineering at Northwestern University during 1983–1985 and in Centre de Mathématiques Appliquées at
Ecole Polytechnique in 1988–1989. He conducts research on fundamentals and applications of the BIEs in
continuum mechanics with special attention to wave propagation, fracture, inverse, and large-scale problems.
He has over 150 publications in archival journals and conference proceedings, and is an author of chapters
in 17 books.

031001-36 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Martin Schanz is a Professor of Mechanics at the Department of Civil Engineering at Graz University of
Technology. He studied mechanical engineering at the University of Karlsruhe (Germany) and earned his
Diploma degree in 1990. From 1990 he investigated in improving the BEM formulation for wave propagation
problems. His doctoral thesis, done in 1994 at the University of the Federal Armed Forces in Hamburg, is
about viscoelastic BEM formulations in time domain supervised by Professor L. Gaul. He received his habili-
tation in Mechanics in 2001 at the Technical University in Braunschweig with a thesis on a visco- and poro-
elastodynamic BEM formulation. In 1999 he had a research stay at the Department of Civil Engineering at
the University of Delaware (USA) and in 2002 he was visiting professor at the Department of Computational
Mechanics at the University of Campinas (Brazil). He has authored and co-authored about 36 journal papers,
three books, and several book chapters. His research interests cover visco- and poro-elasticity, homogeniza-
tion of materials with microstructure, and the BEM with focus on wave propagation. He is a member of
GAMM (International Association of Applied Mathematics and Mechanics), and several other professional
societies.

Wenjing Ye received her BS degree in Mechanical Engineering from the University of Science and Technol-
ogy of China in 1989, her MS degree in Mechanical Engineering from the University of California at San
Diego in 1993, and her Ph. D. degree in Theoretical and Applied Mechanics from Cornell University in
1998. She was a postdoctoral associate at the Research Lab of Electronics at Massachusetts Institute of Tech-
nology before joining the faculty of the Woodruff School of Mechanical Engineering at Georgia Institute of
Technology in 1999. From 2006–2007, She took a leave from Georgia Tech and was a Visiting Assistant Pro-
fessor at the Hong Kong University of Science and Technology (HKUST). She is now a regular faculty mem-
ber in the Department of Mechanical Engineering at HKUST. Her research interests are in numerical
methods and analysis including the BEM, molecular dynamics, computational mechanics, modeling and
design of MEMS=NEMS and submicron gas transport and its applications. She has over 50 publications in
archival journals and conference proceedings, and is an author of chapters in four books.

Alok Sutradhar is an Assistant Professor in the Department of Surgery and Department of Mechanical and
Aerospace Engineering at The Ohio State University. He obtained his BS and MS degrees in Civil Engineer-
ing from Bangladesh University of Engineering and Technology. Thereafter, he received his Ph. D. in Civil
Engineering with Computational Science and Engineering option from the University of Illinois at Urbana-
Champaign in 2005. He did a postdoctoral fellowship at the Digital Simulation Laboratory at UT MD Ander-
son Cancer Center for a year and worked as a postdoctoral researcher in the Department of Surgery at Ohio
State University Medical Center before joining as a faculty. He performs interdisciplinary research that employs
mechanics and mathematics to solve problems within the fields of engineering, biomechanics and biomedical
engineering. His research interests are in biomedical modeling, patient-specific simulations, breast biomechanics,
craniofacial reconstruction, cancer modeling, tissue engineering scaffolds, topology optimization, fracture, bound-
ary element and mesh-reduction methods. He has coauthored 14 peer-reviewed journal papers and the book The
Symmetric Galerkin Boundary Element Method published by Springer in 2008.

Ernie Pan is a Professor of Civil Engineering at The University of Akron and a Fellow of the ASME. His BS
degree is from Lanzhou University, MS degree from Peking University, and Ph. D. degree from the Univer-
sity of Colorado at Boulder. His research interests are in computational mechanics as applied to anisotropic
magnetoelectroelastic solids in general and to quantum nanostructures in particular. His research expertise
is in anisotropic and multilayered Green’s functions. He has co-authored more than 200 research papers in
journals and conference proceedings. He edited a special issue on “Anisotropic Green’s Functions and
BEMs” for Engineering Analysis of Boundary Elements in 2005. He has also pioneered various benchmark
solutions for multiphase and multilayered composites with functionally graded materials. He has played
various active roles in, and contributed to, the Green’s function research and education.

Ney Augusto Dumont is a Professor of Civil Engineering at the Pontifical Catholic University of Rio de
Janeiro (PUC-Rio), Brazil. He received his BS degree in Civil Engineering at the Federal University of
Bahia, Brazil, in 1972, and the MS and Dr.-Ing. degrees in Structural Engineering at PUC-Rio, in 1973, and
at the Technische Universität Darmstadt, Germany, in 1978, respectively. His research interests are in com-
putational mechanics, variational methods, hybrid methods, FEM, BEM, fracture mechanics, time-dependent
problems and, more recently, gradient elasticity. He has over 150 publications in international journals,
conference proceedings, and book chapters.

Applied Mechanics Reviews MAY 2011, Vol. 64 / 031001-37

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Attilio Frangi is an Associate Professor in the Department of Structural Engineering of the Politecnico di
Milano since 2004 and at the Department of Mechanics, Ecole Polytechnique, Palaiseau (Paris) since 2001.
After graduating cum laude in Aeronautical Engineering in 1994, he obtained a Ph. D. in Structural Engi-
neering in 1998. From 2000 to 2004 he was an Assistant Professor with the Department of Structural Engi-
neering of the Politecnico di Milano. He was an associate researcher at the Laboratoire de Mecanique des
Solides, Ecole Polytechnique, Paris (1994–1995) and at the Department of Civil Engineering, University of
Minnesota at Minneapolis (1997). He is a (co-) author of more than 120 publications on issues of computa-
tional mechanics and micromechanics. He has co-edited one scientific monograph on the multiphysics simu-
lation of MEMS and NEMS. He is the recipient of the Fondazione Confalonieri prize (1999), Young
Researcher Fellowship Award (MIT, 2001), and Bruno Finzi Prize for Rational Mechanics (Istituto Lom-
bardo Accademia di Scienze e Lettere, 2010). His research interests include the theoretical and numerical
analysis of singular integral equations and the simulation of multiphysics phenomena for industrial electro-
magnetic applications and micro=nano structures. He is a member of the International Advisory Board of the
European Journal of Computational Mechanics.

Andres Saez is a Professor at the Department of Continuum Mechanics and Structural Analysis of the Uni-
versity of Seville (Spain). He received his B Eng degree from the School of Engineering at the University of
Seville in 1992, his M Sc degree from the Civil Engineering Department at Northwestern University (USA) in
1994, and his Ph. D. degree from the University of Seville in 1997. His research mainly focuses on fundamen-
tals and applications of the BEM in continuum mechanics, with emphasis on wave propagation and fracture
of anisotropic and multifield (piezoelectric and magnetoelectroelastic) materials. He has co-authored more
than 100 research papers in journals and conference proceedings.

031001-38 / Vol. 64, MAY 2011 Transactions of the ASME

Downloaded 21 Apr 2012 to 131.175.24.250. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm