Download as pdf or txt
Download as pdf or txt
You are on page 1of 29

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING

Int. J. Numer. Meth. Engng 2008; 74:209–237


Published online 26 July 2007 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/nme.2147

A continued-fraction-based high-order transmitting boundary for


wave propagation in unbounded domains of arbitrary geometry

Mohammad Hossein Bazyar and Chongmin Song∗, †


School of Civil and Environmental Engineering, University of New South Wales, Sydney, NSW 2052, Australia

SUMMARY
A high-order local transmitting boundary is developed to model the propagation of elastic waves in
unbounded domains. This transmitting boundary is applicable to scalar and vector waves, to unbounded
domains of arbitrary geometry and to anisotropic materials. The formulation is based on a continued-
fraction solution of the dynamic-stiffness matrix of an unbounded domain. The coefficient matrices of
the continued fraction are determined recursively from the scaled boundary finite element equation in
dynamic stiffness. The solution converges rapidly over the whole frequency range as the order of the
continued fraction increases. Using the continued-fraction solution and introducing auxiliary variables,
a high-order local transmitting boundary is formulated as an equation of motion with symmetric and
frequency-independent coefficient matrices. It can be coupled seamlessly with finite elements. Standard
procedures in structural dynamics are directly applicable for evaluating the response in the frequency and
time domains. Analytical and numerical examples demonstrate the high rate of convergence and efficiency
of this high-order local transmitting boundary. Copyright q 2007 John Wiley & Sons, Ltd.

Received 20 December 2006; Revised 25 May 2007; Accepted 4 June 2007

KEY WORDS: scaled boundary finite element method; transmitting boundary; absorbing boundary
condition; continued fraction; unbounded domains

1. INTRODUCTION

The modeling of wave propagation in unbounded domains is required in many fields of engineering
and science, such as acoustics, electromagnetics, geophysics, meteorology and elastodynamics.
Over the last four decades, a large amount of numerical methods have been developed in this
area. Excellent literature reviews can be found in papers [1–6] and in books [7–10]. Most of the
existing techniques can be categorized as rigorous and approximate methods.

∗ Correspondence to: Chongmin Song, School of Civil and Environmental Engineering, University of New South
Wales, Sydney, NSW 2052, Australia.

E-mail: c.song@unsw.edu.au

Copyright q 2007 John Wiley & Sons, Ltd.


210 M. H. BAZYAR AND C. SONG

Rigorous methods are, in general, spatially and temporally global. They are computationally
expensive for large-scale problems and long-time calculations. One of the most popular rigorous
methods is the boundary element method [11, 12]. This method satisfies the governing equations
in the problem domain and the radiation condition at infinity automatically by using a fundamental
solution. Only the boundary needs to be discretized. However, it is very complicated to evaluate the
fundamental solution when the material is anisotropic. The thin-layer method (consistent boundary
condition) [13] has been developed for horizontally layered media. In this method, the boundary
in the vertical direction is discretized and displacement functions in the horizontal direction satisfy
the radiation condition at infinity. Exact non-reflecting boundary conditions, such as Dirichlet-to-
Neumann (DtN) map [14], can be constructed from analytical solutions for unbounded domains.
As analytical solutions are available only for simple geometries and material properties, their
application is limited [4].
The approximate methods are, in general, spatially and temporally local. They are computation-
ally efficient by themselves, but have to be applied at a so-called artificial boundary sufficiently far
away from the region of interest in order to obtain results of acceptable accuracy. This increases
the total computational effort. The viscous boundary [15] consists of dashpots that absorb plane
waves propagating perpendicularly to the artificial boundary. The Bayliss, Gunzburger and Turkel
(BGT) boundary [16] annihilates selected terms of cylindrical or spherical waves. Infinite elements
[5, 17] use decay functions representing the wave propagation toward infinity as shape functions
of the displacements. In the method of perfectly matched layer [18], the domain of interest is
surrounded by a finite absorbing layer that attenuates waves.
A high-order local transmitting boundary is an attractive alternative. It has the potential of leading
to accurate results as the order of approximation increases. At the same time, it is computationally
efficient owing to the local formulation. The paraxial boundary [19] constructs a differential
equation favoring the out-going waves by splitting the differential operator of the scalar wave
equation for plane waves. The BGT [16] boundary is derived from a series solution for spherical
waves. The multi-direction boundary [20] constructs a differential equation which absorbs plane
waves at preselected angles. However, the order of derivatives in these formulations increases with
the order of the transmitting boundary. Beyond the second order, the implementation in a finite
element computer program becomes complex and instability may occur [8].
The higher-order derivatives in the high-order transmitting boundary conditions can be eliminated
by introducing auxiliary variables. Grote and Keller [21] developed high-order local transmitting
boundary conditions based on spherical harmonic transformations for three-dimensional time-
dependent scalar wave equations, three-dimensional elastodynamics [22] and Maxwell’s equations
[23]. Hagstrom and Hariharan [24] derived high-order transmitting boundary conditions based on
the Bayliss and Turkel boundary [25] for two- and three-dimensional time-dependent scalar and
Maxwell’s wave equations. Huan and Thompson formulated local transmitting boundaries for scalar
waves on a sphere [26] and on a circle [27] by directly applying the Bayliss and Turkel boundary
[25]. Krenk [28] derived a high-order boundary condition from a rational function approxima-
tion of the plane wave representation for scalar waves. Guddati and Tassoulas [29] developed a
high-order local boundary condition based on recursive continued fractions for scalar wave equa-
tions in Cartesian coordinates. Guddati and Lim [30] extended the continued-fraction absorbing
boundary condition to polygonal computational domains. Givoli and Paltashenko [31] proposed
a frequency-domain high-order local boundary condition for a two-dimensional Helmholtz equa-
tion by localizing the non-local DtN map. It is applicable to cylindrical cavities and wave-guide
configurations. Givoli and Neta [32] developed high-order boundary conditions based on the

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 211

multi-direction boundary [20] for a plane boundary. Hagstrom and Warburton [33] generalized
the Givoli and Neta [32] boundary condition to a full-space configuration and enhanced its sta-
bility by deriving corner compatibility conditions for the auxiliary variable equations. van Joolen
et al. [34] extended the high-order boundary condition proposed in [32] to unbounded domains
with a rectangular boundary. Kechroud et al. [35] proposed a high-order Padé-type non-reflecting
boundary condition in the frequency domain for two-dimensional acoustic scattering problems.
This boundary condition can be applied on any convex fictitious boundary. Givoli et al. [36] in-
corporated the boundary condition developed in [33] in a finite element scheme and compared its
performance with the boundary condition developed in [32].
In many engineering applications, such as in dynamic soil–structure interaction problems, the
material in an unbounded domain may be anisotropic. The geometry of the unbounded domain is
usually arbitrary, with curved segments and corners. The extension of the above high-order local
transmitting boundaries to elastic waves in anisotropic unbounded domains of arbitrary geometry
is not a straightforward task. The progress made in developing a reliable procedure for such
unbounded domains is limited. When the dynamic-stiffness matrix of an unbounded domain is
available at discrete frequencies, for example from an analytical method, the boundary element
method or the scaled boundary finite element method (SBFEM), a temporally local procedure
can be constructed by transforming the Padé approximation of the dynamic-stiffness matrix into
a recursive formulation in the time domain [37, 38]. Similarly, when the unit-impulse response
matrix of an unbounded domain is available at discrete time steps, a recursive formulation can
be derived by applying the system theory [39, 40]. Although the subsequent transient analysis is
efficient, the calculation of the dynamic-stiffness matrix or the unit-impulse response matrix is
often computationally expensive for unbounded domains of arbitrary geometry.
The SBFEM [41] has emerged as a promising alternative for the dynamic analysis of unbounded
domains. Only the boundary is discretized as in the boundary element method, but no fundamental
solution is required. General anisotropic materials can be analyzed without additional efforts. This
method has been applied successfully in both frequency and time domains [10, 40, 42–44]. It is
extended to the analysis of non-homogeneous unbounded domains with the elasticity modulus
and mass density varying as power functions of spatial coordinates [45, 46]. The original solution
procedure of the scaled boundary finite element equation is global in space and time [10, 47], and
thus computationally expensive. To increase the computational efficiency for large-scale problems,
novel solution procedures have been developed recently. A reduced set of base functions leading
to spatially local formulations is developed [46, 48]. The technique of hierarchical matrices is
employed to speed up the matrix–vector products encountered, e.g. in convolution integrals [44].
A Padé series solution for the dynamic-stiffness matrix of an unbounded domain is developed
by Song and Bazyar [49] for frequency-domain analyses. The sparsity and the lumping of the
coefficient matrices of the scaled boundary finite element equation are exploited in [50]. The
objective of this paper is to develop a high-order local transmitting boundary condition directly
from the scaled boundary finite element equation. The computation of the dynamic-stiffness matrix
or the unit-impulse response matrix at discrete frequencies or time steps is not required.
The outline of the paper is as follows. In Section 2, the SBFEM is summarized. In Section 3, a
continued-fraction solution for the dynamic-stiffness matrix is constructed. The coefficient matrices
of the continued fraction are determined recursively and directly from the scaled boundary finite
element equation in dynamic stiffness. In Section 4, the high-order local transmitting boundary
condition is constructed by using the continued-fraction solution and introducing auxiliary variables.
The formulation is expressed as an equation of motion with symmetric and frequency-independent

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
212 M. H. BAZYAR AND C. SONG

coefficient matrices. It can be coupled seamlessly with finite elements. Standard procedures in
structural dynamics are directly applicable to evaluate the response in the frequency and time
domains. In Section 5, analytical and numerical examples are presented to evaluate the accuracy
and efficiency of this high-order transmitting boundary. In Section 6, concluding remarks are
stated.

2. SUMMARY OF THE SCALED BOUNDARY FINITE ELEMENT METHOD

The scaled boundary finite element formulation for elastodynamics is detailed in [41, 51]. For the
sake of completeness, a brief summary of the equations necessary for the development of the
high-order transmitting boundary condition is presented in this section.
In the SBFEM, a so-called scaling center O is chosen in a zone from which the total boundary,
other than the straight surfaces passing through the scaling center, must be visible (Figure 1(a)).
Only the boundary S visible from the scaling center O is discretized (see Figure 1(b) for a typical
line element to be used in two-dimensional problems and Figure 1(c) for a typical surface element
to be used in three-dimensional problems). The coordinates of the nodes of an element in a
three-dimensional Cartesian coordinate system are arranged in {x}, {y}, {z}. The geometry of the
isoparametric element is interpolated using the shape functions [N (, )] formulated in the local
coordinates ,  of an element on boundary as

x̂(, , ) = [N (, )]{x}, ŷ(, , ) = [N (, )]{y}, ẑ(, , ) = [N (, )]{z} (1)

where , ,  are called the scaled boundary coordinates.


The nodal displacement functions {u()} are introduced along the radial lines passing through
the scaling center O and a node on the boundary. (The dependency on time t or the excitation
frequency  in a frequency-domain analysis is omitted from the argument for simplicity when
it is not explicitly required.) The displacements at a point (, , ) are interpolated from the
displacement functions {u()} as

{u(, , )} = [N u (, )]{u()} = [N1 (, )[I ], N2 (, )[I ], . . .]{u()} (2)

The scaled boundary finite element equation is derived by applying the Galerkin’s weighted resid-
ual technique or the virtual work method in the circumferential directions ,  to the governing

y^
ζ
+1 7
ξ >1

4 3
ξ=1

x^
O
S -1 0 +1 η 8 6 η
-1 +1
V 1 3 2
1 2
(a) (b) (c) -1 5

Figure 1. (a) Representation of unbounded domain in the scaled boundary finite element method;
(b) three-node line element on boundary; and (c) eight-node surface element on boundary.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 213

differential equations. In the frequency domain, the scaled boundary finite element equation in
displacement is expressed as

[E 0 ]2 {u()}, + ((s − 1)[E 0 ] − [E 1 ] + [E 1 ]T ){u()},

+ ((s − 2)[E 1 ]T − [E 2 ]){u()} + 2 [M 0 ]2 {u()} = 0 (3)

where s (= 2 or 3) denotes the spatial dimension of the domain. [E 0 ], [E 1 ], [E 2 ] and [M 0 ] are


coefficient matrices obtained by assembling the element coefficient matrices as in the finite element
method. For three-dimensional problems, the element coefficient matrices are expressed as (for
the sake of simplicity, the same symbols are used for the element coefficient matrices)
 +1  +1
[E 0 ] = [B 1 (, )]T [D(, )][B 1 (, )]|J (, )| d d (4a)
−1 −1
 +1  +1
[E 1 ] = [B 2 (, )]T [D(, )][B 1 (, )]|J (, )| d d (4b)
−1 −1
 +1  +1
[E ] =
2
[B 2 (, )]T [D(, )][B 2 (, )]|J (, )| d d (4c)
−1 −1
 +1  +1
[M 0 ] = [N (, )]T [N (, )]|J (, )| d d (4d)
−1 −1

where [B 1 (, )] and [B 2 (, )] represent the strain–nodal displacement relationship and |J (, )|
is the determinant of the Jacobian matrix on the boundary. The element coefficient matrices for
two-dimensional and axisymmetric elastodynamics are available in [41]. The coefficient matri-
ces [E 0 ] and [M 0 ] are positive definite. [E 2 ] is symmetric. As for the mass matrices in finite
elements, [E 0 ] and [M 0 ] can be lumped to the nodes [50]. [E 0 ] will be a block-diagonal ma-
trix consisting blocks of size s × s (s = 2 or 3). [M 0 ] will be a diagonal matrix. The techniques
of using Gauss–Lobatto–Legendre shape functions and quadrature are investigated by Song and
Bazyar [50].
The internal nodal forces on a surface with a constant  are obtained by integrating the surface
traction over elements. They are expressed as
{q()} = s−2 ([E 0 ]{u()}, + [E 1 ]T {u()}) (5)
The internal nodal forces are related to the nodal forces {R} on the boundary by {R} = −{q( = 1)}
for an unbounded domain. The dynamic-stiffness matrix of an unbounded domain [S ∞ ()]
(superscript ∞ for unbounded) is defined by
{R()} = [S ∞ ()]{u()} (6)
where the notations {R()} = {R( = 1, )} and {u()} = {u( = 1, )} are introduced. For
simplicity, the word ‘amplitude’ will be omitted when referring to {R()} and {u()}. From
Equations (5) and (6), the relationship between the nodal displacements and the radial derivatives
of the displacements on the boundary is expressed as
[E 0 ]{u(, )}, |=1 = −([S ∞ ()] + [E 1 ]T ){u()} (7)

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
214 M. H. BAZYAR AND C. SONG

The scaled boundary finite element equation in dynamic stiffness is derived from Equations (3),
(5) and (6) (see [41]). It is written on the boundary ( = 1) as

([S ∞ ()] + [E 1 ])[E 0 ]−1 ([S ∞ ()] + [E 1 ]T ) − (s − 2)[S ∞ ()]


− [S ∞ ()], − [E 2 ] + 2 [M 0 ] = 0 (8)

It is a system of non-linear first-order ordinary differential equations in the independent variable .


An asymptotic power series solution is derived in [10, 45]. As it is only valid at high frequency, the
dynamic-stiffness matrix at low and intermediate frequencies has to be evaluated by integrating
Equation (8) numerically. To avoid this computationally expensive task, a Padé series solution
for the dynamic-stiffness matrix is developed in [49] directly from the scaled boundary finite
element equation. It converges throughout the whole frequency range from zero to infinity. The
dynamic-stiffness matrix can be calculated directly at a specified frequency.
It is proven in [52] that the scaled boundary finite element equation for statics can be decoupled
by weighted orthogonal base functions. Numerical examples demonstrate that high-order terms
have little contributions to the responses of unbounded domains. A reduced set of base functions
constructed with eigenvectors with small real parts of eigenvalues is developed to exclude the high-
order modes. This significant development led to a spatially local formulation for the SBFEM.
The technique of the reduced set of base functions is later extended to the dynamic analysis of
unbounded domains [46, 48, 49], resulting in a significant reduction in the computational effort.
As explained in [49], the technique of the reduced set of base functions introduces modifications
to the coefficient matrices of Equation (8) while maintaining the form of the equation. Therefore,
the development in this paper can be directly applied in combination with the reduced set of base
functions.

3. A CONTINUED-FRACTION SOLUTION OF A DYNAMIC-STIFFNESS MATRIX

A continued fraction is closely related to a Padé series [53, 54]. Both of them have much larger
convergence range and higher convergence rate than the corresponding power series. The Padé
series is widely employed in numerical analysis such as interpolation, integration of differential
equations [55], etc. In the analysis of unbounded domains, Ruge et al. [38] derived an equivalent
continued fraction from a Padé approximation of the dynamic-stiffness matrix, which is obtained by
fitting the values of the dynamic-stiffness matrix at discrete frequencies. A high-order transmitting
boundary is constructed from the continued-fraction approximation by using the technique of
mixed variables. In this section, a continued-fraction solution for the dynamic-stiffness matrix is
determined directly from the scaled boundary finite element equation in dynamic stiffness. Neither a
dynamic-stiffness matrix at discrete frequencies nor a Padé approximation of the dynamic stiffness
is required.
To start the continued-fraction solution, the dynamic-stiffness matrix [S ∞ ()] is decomposed as
[S ∞ ()] = (i)[C∞ ] + [K ∞ ] − [Y (1) ()]−1 (9)
The first two terms are the constant dashpot matrix [C∞ ] and the constant spring matrix [K ∞ ].
[Y (1) ()]−1 is the residual of the two-term expansion at high frequency. Substitution of Equation (9)
in Equation (8) leads to three terms in descending order of the power of (i), i.e. (i)2 term,

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 215

(i) term and the remaining lower-order term

(i)2 ([C∞ ][E 0 ]−1 [C∞ ] − [M 0 ])


+ (i)([C∞ ][E 0 ]−1 ([K ∞ ] + [E 1 ]T ) + ([K ∞ ] + [E 1 ])[E 0 ]−1 [C∞ ] − (s − 1)[C∞ ])
+ ([K ∞ ] + [E 1 ])[E 0 ]−1 ([K ∞ ] + [E 1 ]T ) − ((i)[C∞ ] + [K ∞ ] + [E 1 ])[E 0 ]−1 [Y (1) ()]−1
− [Y (1) ()]−1 [E 0 ]−1 ((i)[C∞ ] + [K ∞ ] + [E 1 ]T ) + [Y (1) ()]−1 [E 0 ]−1 [Y (1) ()]−1
− (s − 2)[K ∞ ] + (s − 2)[Y (1) ()]−1 − [E 2 ] + ([Y (1) ()]−1 ), = 0 (10)

Equation (10) is satisfied when all the three terms are equal to zero. Setting the coefficient of the
(i)2 term to zero results in
[C∞ ][E 0 ]−1 [C∞ ] − [M 0 ] = 0 (11)
The solution of Equation (11) is obtained from the general eigenvalue problem [10, Section 7.2]
[M 0 ][] = [E 0 ][]2  (12)
The eigenvalues   are positive as [E 0 ] and [M 0 ] are positive definite. The eigenvector matrix
2

[] is normalized as
[]T [E 0 ][] = [I ] (13)
yielding
[]T [M 0 ][] = 2  (14)
The solution for [C∞ ] satisfying the radiation condition is symmetric and positive definite
(assuming a time dependence of e+it ):
[C∞ ] = [E 0 ][][]T [E 0 ] (15)
When [E 0 ] and [M 0 ] are lumped, Equation (12) is a series of independent eigenvalue problems
of 2 × 2 (for two-dimensional problems) or 3 × 3 (for three-dimensional problems) matrices. []
and [C∞ ] are also block-diagonal with the same structure. For later use, pre-multiplying Equation
(15) with [E 0 ]−1 and using Equation (13) result in
[E 0 ]−1 [C∞ ] = [][]−1 (16)
Setting the (i) term in Equation (10) to zero leads to the following equation for [K ∞ ]
[C∞ ][E 0 ]−1 ([K ∞ ] + [E 1 ]T ) + ([K ∞ ] + [E 1 ])[E 0 ]−1 [C∞ ] − (s − 1)[C∞ ] = 0 (17)
Pre- and post-multiplying Equation (17) with []T and [], respectively, and using Equation (16)
yields
[k∞ ] + [k∞ ] = −[]T [E 1 ]T [] − []T [E 1 ][] + (s − 1) (18)
with (the superscript −T denotes the transpose of the inverse of a matrix)
[K ∞ ] = [E 0 ][][k∞ ][]T [E 0 ] (19)

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
216 M. H. BAZYAR AND C. SONG

Equation (18) can be solved directly by back substitution as the coefficient matrix at the left-hand
side is diagonal. [K ∞ ] follows from Equation (19). It is symmetric. The remaining part of Equation
(10) is an equation for [Y (1) ()]−1 :

([K ∞ ] + [E 1 ])[E 0 ]−1 ([K ∞ ] + [E 1 ]T ) − ((i)[C∞ ] + [K ∞ ] + [E 1 ])[E 0 ]−1 [Y (1) ()]−1


− [Y (1) ()]−1 [E 0 ]−1 ((i)[C∞ ] + [K ∞ ] + [E 1 ]T ) + [Y (1) ()]−1 [E 0 ]−1 [Y (1) ()]−1
− (s − 2)[K ∞ ] + (s − 2)[Y (1) ()]−1 − [E 2 ] + ([Y (1) ()]−1 ), = 0 (20)

The derivative ([Y (1) ()]−1 ), is determined starting from the definition of the matrix inversion
[Y (1) ()]−1 [Y (1) ()] = [I ] (21)
Differentiating Equation (21) results in
([Y (1) ()]−1 ), [Y (1) ()] + [Y (1) ()]−1 [Y (1) ()], = 0 (22)
Post-multiplying Equation (22) with [Y (1) ()]−1 leads to
([Y (1) ()]−1 ), = −[Y (1) ()]−1 [Y (1) ()], [Y (1) ()]−1 (23)
An equation for [Y (1) ()] is obtained after substituting Equation (23) into Equation (20), and pre-
and post-multiplying the resulting expression with [Y (1) ()]:

[E 0 ]−1 − [Y (1) ()](((i)[C∞ ] + [K ∞ ] + [E 1 ])[E 0 ]−1 − 0.5(s − 2)[I ])


− ([E 0 ]−1 ((i)[C∞ ] + [K ∞ ] + [E 1 ]T ) − 0.5(s − 2)[I ])[Y (1) ()]
+ [Y (1) ()](([K ∞ ] + [E 1 ])[E 0 ]−1 ([K ∞ ] + [E 1 ]T ) − [E 2 ] − (s − 2)[K ∞ ])[Y (1) ()]
− [Y (1) ()], = 0 (24)

Using Equation (16), Equation (24) is written as the case i = 1 of the following equation:

[a (i) ] − [Y (i) ()]((i)[b1(i) ]T + [b0(i) ]T ) − ((i)[b1(i) ] + [b0(i) ])[Y (i) ()]

+ [Y (i) ()][c(i) ][Y (i) ()] − [Y (i) ()], = 0 (25)

with the coefficient matrices defined as

[a (1) ] = [E 0 ]−1 (26a)


(1)
[b0 ] = [E 0 ]−1 ([K ∞ ] + [E 1 ]T ) − 0.5(s − 2)[I ] (26b)

[V (1) ] = [] (26c)


(1)
[b1 ] = [V (1) ]  [V (1) ]−1 (26d)

[c(1) ] = ([K ∞ ] + [E 1 ])[E 0 ]−1 ([K ∞ ] + [E 1 ]T )


− (s − 2)[K ∞ ] − [E 2 ] (26e)

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 217

(1)
Both [a (1) ] and [c(1) ] are symmetric. [b1 ] is automatically expressed as an eigenvalue decompo-
sition. Equation (25) is in the same form as Equation (8). The expansion in Equation (9) is the
starting term of a continued-fraction solution for [S ∞ ()] where the radiation condition at infinity
is satisfied. As the leading order of [Y (1) ()]−1 at high frequency is (i)−1 , the highest order
term of [Y (1) ()] is (i). [Y (i) ()] can be decomposed as a continued fraction:
(i) (i)
[Y (i) ()] = [Y0 ] + (i)[Y1 ] − [Y (i+1) ()]−1 (27)

Using Equation (27), Equation (25) is again expanded to (i)2 , (i) and remaining lower-order
terms
(i) (i) (i) (i) (i) (i)
(i)2 (−[Y1 ][b1 ]T − [b1 ][Y1 ] + [Y1 ][c(i) ][Y1 ])

+ (i)([Y0(i) ](−[b1(i) ]T + [c(i) ][Y1(i) ]) + (−[b1(i) ] + [Y1(i) ][c(i) ])[Y0(i) ]

− [b0(i) ][Y1(i) ] − [Y1(i) ][b0(i) ]T − [Y1(i) ]) + [a (i) ] − [b0(i) ][Y0(i) ]

− [Y0(i) ][b0(i) ]T + [Y0(i) ][c(i) ][Y0(i) ] + ([b0(i) ] + (i)[b1(i) ])[Y (i+1) ()]−1

+ [Y (i+1) ()]−1 ([b0(i) ]T + (i)[b1(i) ]T ) − ([Y0(i) ] + (i)[Y1(i) ])[c(i) ][Y (i+1) ()]−1

− [Y (i+1) ()]−1 [c(i) ]([Y0(i) ] + (i)[Y1(i) ]) + [Y (i+1) ()]−1 [c(i) ][Y (i+1) ()]−1

− [Y (i+1) ()]−1 [Y (i+1) ()], [Y (i+1) ()]−1 = 0 (28)

As for Equation (20), Equation (28) is satisfied when all the three terms in the power series are
(i)
equal to zero. Setting the (i)2 term to zero leads to an equation for [Y1 ]:
(i) (i) (i) (i) (i) (i)
−[Y1 ][b1 ]T − [b1 ][Y1 ] + [Y1 ][c(i) ][Y1 ] = 0 (29)
(i)
Pre- and post-multiplying the above equation with [Y1 ]−1 leads to a Lyapunov equation for
(i)
[Y1 ]−1 :
(i) (i) (i) (i)
[Y1 ]−1 [b1 ] + [b1 ]T [Y1 ]−1 = [c(i) ] (30)
(i)
[b1 ] is expressed as an eigenvalue decomposition (see Equation (26d) for i = 1 and Equation
(43c) for i>1)
(i)
[b1 ] = [V (i) ][V (i) ]−1 (31)
Substituting Equation (31) into Equation (30) and pre- and post-multiplying the resulting expression
with [V (i) ]T and [V (i) ], respectively, lead to
(i) (i)
[y1 ]−1  + [y1 ]−1 = [V (i) ]T [c(i) ][V (i) ] (32)
with
(i) (i)
[Y1 ]−1 = [V (i) ]−T [y1 ]−1 [V (i) ]−1 (33)

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
218 M. H. BAZYAR AND C. SONG

(i)
In Equation (32) as  is diagonal, the entries of [y1 ]−1 are evaluated by back substitution.
(i)
Again [Y1 ]−1 is symmetric as [c(i) ] is symmetric. The (i) term in Equation (28) leads to a
(i)
Lyapunov equation for [Y0 ] as (i1)
(i) (i) (i) (i) (i) (i)
(−[b1 ] + [Y1 ][c(i) ])[Y0 ] + [Y0 ](−[b1 ]T + [c(i) ][Y1 ])
(i) (i) (i) (i) (i)
− [b0 ][Y1 ] − [Y1 ][b0 ]T − [Y1 ] = 0 (34)

Using Equation (30), its coefficient matrix can be expressed as an eigenvalue decomposition
(i) (i) (i) (i) (i)
−[b1 ] + [Y1 ][c(i) ] = [Y1 ][b1 ]T [Y1 ]−1 = [V (i+1) ]−T [V (i+1) ]T (35)
with the eigenvectors
[V (i+1) ] = [Y1(i) ]−1 [V (i) ] (36)
Substituting Equation (35) into Equation (34) and pre- and post-multiplying the resulting expression
with [V (i+1) ]T and [V (i+1) ], respectively, yield
(i) (i) (i) (i) (i)
[y0 ] + [y0 ] = [V (i+1) ]T [b0 ][V (i) ] + [V (i) ]T [b0 ]T [V (i+1) ] + [y1 ]−1 (37)
with
(i) (i)
[Y0 ] = [V (i+1) ]−T [y0 ][V (i+1) ]−1 (38)
(i)
where [Y0 ] is symmetric. The remaining lower-order term in Equation (28) leads to an equation
for [Y (i+1) ()]−1 as
(i) (i) (i) (i) (i) (i)
[a (i) ] − [b0 ][Y0 ] − [Y0 ][b0 ]T + ([Y0 ][c(i) ][Y0 ])
(i) (i) (i) (i)
× ([b0 ] − [Y0 ][c(i) ] + (i)([b1 ] − [Y1 ][c(i) ]))[Y (i+1) ()]−1
(i) (i) (i) (i)
+ [Y (i+1) ()]−1 ([b0 ]T − [c(i) ][Y0 ] + (i)([b1 ]T − [c(i) ][Y1 ]))

+ [Y (i+1) ()]−1 [c(i) ][Y (i+1) ()]−1 − ([Y (i+1) ()]−1 ), = 0 (39)

Following Equations (21)–(23), the derivative of ([Y (i+1) ()]−1 ), is obtained as
([Y (i+1) ()]−1 ), = −[Y (i+1) ()]−1 [Y (i+1) ()], [Y (i+1) ()]−1 (40)
Substituting Equation (40) and pre- and post-multiplying Equation (39) with [Y (i+1) ()] results
in an equation for [Y (i+1) ()] as
(i) (i) (i) (i) (i) (i)
[Y (i+1) ()]([a (i) ] − [b0 ][Y0 ] − [Y0 ][b0 ]T + [Y0 ][c(i) ][Y0 ])[Y (i+1) ()]
(i) (i) (i) (i)
− [Y (i+1) ()](−[b0 ] + [Y0 ][c(i) ] + (i)(−[b1 ] + [Y1 ][c(i) ]))
(i) (i) (i) (i)
− (−[b0 ]T + [c(i) ][Y0 ] + (i)(−[b1 ]T + [c(i) ][Y1 ]))[Y (i+1) ()]

+ [c(i) ] − [Y (i+1) ()], = 0 (41)

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 219

The coefficient matrix of the linear term in (i) is nothing but the transpose of Equation (35):
(i) (i)
−[b1 ]T + [c(i) ][Y1 ] = [V (i+1) ][V (i+1) ]−1 (42)
Introducing the recursive formula for the coefficient matrices

[a (i+1) ] = [c(i) ] (43a)


(i+1) (i) (i)
[b0 ] = −[b0 ]T + [c(i) ][Y0 ] (43b)
(i+1) (i) (i)
[b1 ] = −[b1 ]T + [c(i) ][Y1 ] = [V (i+1) ]  [V (i+1) ]−1 (43c)
(i) (i) (i) (i) (i) (i)
[c(i+1) ] = [a (i) ] − [b0 ][Y0 ] − [Y0 ][b0 ]T + [Y0 ][c(i) ][Y0 ] (43d)

leads to Equation (25) for case i + 1:


(i+1) T (i+1) T (i+1) (i+1)
[a (i+1) ] − [Y (i+1) ()]([b0 ] + i[b1 ] ) − ([b0 ] + i[b1 ])[Y (i+1) ()]

+ [Y (i+1) ()][c(i+1) ][Y (i+1) ()] − [Y (i+1) ()], = 0 (44)

Equation (44) can be solved by following the same steps as for solving Equation (25). The
(i)
coefficient matrices [a (i) ], [b0 ] and [c(i) ] are evaluated recursively using Equations (36) and
(i)
(43), starting from those at i = 1. The coefficient matrix [b1 ] is used only for the derivation of
equations and does not need to be evaluated explicitly. An order M continued fraction terminates
with the approximation [Y (i+1) ()]−1 = 0. Unlike for the Padé series solution [49, Equations (19)
and (21)], increasing the order of continued fraction does not require the recalculation of the
coefficient matrices determined previously for a lower order.

4. CONSTRUCTING A HIGH-ORDER LOCAL TRANSMITTING


BOUNDARY CONDITION

In structural dynamics, an equation of motion is expressed by stiffness, damping and mass matri-
ces [56]. Reliable and efficient techniques have been developed to solve the equation of motion
for the structural responses in the time or frequency domain. Starting from the continued-fraction
solution of the dynamic-stiffness matrix, a high-order local transmitting boundary condition can
be constructed as such an equation of motion. The resulting coefficient matrices are frequency
independent and symmetric. This high-order transmitting boundary modeling the unbounded
domain can be coupled seamlessly and straightforwardly with finite elements modeling the
near field.
Substituting the first term of the continued-fraction solution in Equation (9) into Equation (6),
the interaction forces on the boundary ( = 1) {R()} are related to the displacements on the
boundary {u()} by
{R()} = [S ∞ ()]{u()} = ((i)[C∞ ] + [K ∞ ]){u()} − {u (1) ()} (45)
where the auxiliary variable {u (1) ()} is defined as
{u()} = [Y (1) ()]{u (1) ()} (46)

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
220 M. H. BAZYAR AND C. SONG

Equation (27), representing the remaining terms of the continued-fraction solution, is expressed as
(i) (i)
{u (i−1) ()} = ([Y0 ] + (i)[Y1 ]){u (i) ()} − {u (i+1) ()} for i1 (47)

after being post-multiplied with {u (i) ()} and defining the auxiliary variable {u (i+1) ()} as in
Equation (46):
{u (i) ()} = [Y (i+1) ()]{u (i+1) ()} (48)
In Equation (47), {u (0) ()} = {u()} is introduced to denote the displacement on the boundary.
An order M continued fraction terminates with the approximation {u (M+1) ()} = 0. The force–
displacement relationship on the boundary in Equation (45) and the relationships among the
auxiliary variables in Equation (47) can be combined into a matrix form for an order M continued
fraction as
([A] + (i)[B]){Z ()} = {F()} (49)
with the frequency-independent coefficient matrices [A], [B], the function {Z ()} and the external
excitation {F()} defined as
⎡ ⎤
[K ∞ ] −[I ] 0 ··· 0 0
⎢ ⎥
⎢ −[I ] [Y (1) ] −[I ] ··· 0 ⎥
⎢ 0 0 ⎥
⎢ ⎥
⎢ (2) ⎥
⎢ 0 −[I ] [Y0 ] · · · 0 0 ⎥
⎢ ⎥
[A] = ⎢ . .. .. .. ⎥ (50a)
⎢ . . ⎥
⎢ . . . −[I ] 0 ⎥
⎢ ⎥
⎢ (M−1) ⎥
⎢ 0 0 0 −[I ] [Y0 ] −[I ] ⎥
⎣ ⎦
(M)
0 0 0 0 −[I ] [Y0 ]
(1) (2) (M−1) (M)
[B] = diag([C∞ ], [Y1 ], [Y1 ], . . . , [Y1 ], [Y1 ]) (50b)
⎧ ⎫ ⎧ ⎫

⎪ {u()} ⎪ ⎪ ⎪ {R()} ⎪

⎪ ⎪
⎪ ⎪
⎪ ⎪


⎪ {u (1)
()} ⎪
⎪ ⎪
⎪ ⎪


⎪ ⎪
⎪ ⎪
⎪ 0 ⎪


⎪ ⎪
⎪ ⎪
⎪ ⎪


⎪ ⎪ ⎪ ⎪
⎨ {u ()} ⎪
(2)
⎬ ⎨ 0 ⎪
⎪ ⎬
{Z ()} = .. , {F()} = . ⎪ (50c)
⎪ ⎪ ⎪


⎪ . ⎪

⎪ ⎪ .. ⎪

⎪ ⎪


⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ (M−1)
⎪ ⎪
⎪ ⎪ 0 ⎪
⎪ ⎪


⎪ {u ()} ⎪
⎪ ⎪
⎪ ⎪


⎪ ⎪
⎪ ⎪
⎩ ⎪

⎩ (M) ⎭
{u ()} 0

Equation (49) is a standard equation of motion of a linear system in structural dynamics written
in the frequency domain. It is expressed in the time domain as a high-order temporally local
transmitting boundary condition
[A]{Z (t)} + [B]{ Ż (t)} = {F(t)} (51)

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 221

(i) (i)
As all the coefficient matrices of the continued fraction [K ∞ ], [C∞ ], [Y0 ] and [Y1 ] are symmetric,
both the tri-block-diagonal matrix [A] and the block-diagonal matrix [B] are symmetric and sparse.
Equations (49) and (51) can be assembled with finite elements straightforwardly when the same
shape functions are employed at the common boundary. The resulting equation of motion for the
global system can be solved by standard procedures in structural dynamics. In a time-domain
analysis, time-stepping techniques such as Newmark’s method can be used. The numerically
expensive task of evaluating convolution integrals is not required. The solution for Equation (51)
is stable when all the eigenvalues of the general eigenproblem for [A] and [B] have negative real
parts, which can be verified before computing the response.
Combining the reduced set of base functions [48, 49] with the high-order transmitting boundary
condition developed above, a temporally and spatially local high-order transmitting boundary
condition is obtained. The size of the system of equations is reduced to the number of base
functions retained in the reduced set. The order of the transmitting boundary (continued fraction)
can be chosen based on the largest amplitude of the eigenvalues for the reduced set.

5. NUMERICAL EXAMPLES

In this section, the accuracy of the proposed high-order local transmitting boundary in both
frequency and time domains is evaluated by numerical examples. A one-dimensional example, for
which the analytical solution is available, is chosen for a parametric study in Section 5.1. A rigid
footing embedded in a transversely isotropic half-plane is addressed in Section 5.2 to demonstrate
the capacity of the high-order transmitting boundary. The seamless coupling of this transmitting
boundary with finite elements is illustrated with a two-dimensional example in Section 5.3. The
technique of the reduced set of base functions is employed in modeling the two-dimensional
examples. Newmark’s method with  = 0.5 and  = 0.25 (average acceleration scheme) is employed
for the time integration of the equations of motion of the coupled system. The size of the time
step is selected in such a way that the dilatational wave travels less than the length of an element
in one time step.
To provide a reference solution for the two-dimensional examples, the rigorous, spatially and
temporally global solution procedure is applied. The full set of base functions is used in both
frequency and time domains. In a frequency-domain analysis, the dynamic-stiffness matrix at a
high but finite frequency is evaluated by using the first three terms of the asymptotic expansion
for high frequency as detailed in [45, 49]. It provides a starting value to numerically integrate the
scaled boundary finite element equation in dynamic stiffness (Equation (8)) in order to determine
the dynamic-stiffness matrix at intermediate and low frequencies. In a time-domain analysis, the
scaled boundary finite element equation in dynamic stiffness is transformed to one in unit-impulse
response [46, Equation (18)]. After time discretization, the unit-impulse response matrix at discrete
time stations is computed step by step. A transient response is then evaluated from the nodal force–
nodal displacement relationship expressed as a convolution integral.

5.1. Out-of-plane motion of a circular cavity embedded in a full plane


The out-of-plane (anti-plane) motion of a circular cavity embedded in a full-plane is shown in
Figure 2. This problem can be decomposed into a series of circumferential modes by the method
of separation of variables. The propagation of each mode is a one-dimensional problem and can be

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
222 M. H. BAZYAR AND C. SONG

RUCTU
R ST

E-
r0

M E DIU M
u0 u τzr

N r

I
TE
RFACE

Figure 2. Out-of-plane motion of a circular cavity embedded in a full-plane.

solved analytically. In [49, Section 4], the Padé approximation of the dynamic-stiffness coefficient
of a single mode is obtained and compared with the analytical solution. The accuracy is evaluated
for a large range of mode numbers and orders of Padé series [49, Table I]. A guideline is provided
on selecting the order of approximation for complex cases without available analytical solutions. In
this section, the coefficients of the continued-fraction solution for the dynamic-stiffness coefficient
are determined. It is shown that the order M = 1 solution is equivalent to Padé series solution
of the same order and to the second-order BGT boundary condition [16, 57]. The results from
frequency- and time-domain analyses are compared with the analytical solutions. The guideline
on selecting the order of Padé series is shown to be valid for the continued fraction and thus the
high-order transmitting boundary.
The radius of the circular cavity is r0 . The shear modulus of the full plane is G and the mass
density is . The out-of-plane displacement amplitude under excitation frequency  is denoted
as u = u(r, , ) in polar coordinates r , . Only the shear stresses zr and z  occurring in the
z-direction normal to the plane are present:

zr = Gzr = Gu,r (52a)


1
z = Gz  = G u, (52b)
r

The equation of motion is expressed in the frequency domain as a Helmholtz equation:

1 1 2
u,rr + u,r + 2 u, + 2 u = 0 (53)
r r cs

with the shear wave velocity



G
cs = (54)


The analytical solution of this circular cavity problem in the frequency domain is derived in [49]
by using the method of separation of variables. The dynamic-stiffness coefficient on the boundary

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 223

( = 1) is expressed as
 (2)


a0 Hn−1 (a0 )
Sex () = E 0
n− (2)
(55)
Hn (a0 )
(2)
where E 0 is a constant given below in Equation (59a). Hn (a0 ) denotes the second kind of Hankel
functions of order n. a0 is the dimensionless frequency
r0
a0 = (56)
cs
When constructing the high-order transmitting boundary, the displacement along the circumferential
direction of the cavity is decomposed into orthogonal circumferential modes obtained from the
method of separation of variables. The eigenfunction of the nth mode is chosen as the shape
function
N u () = cos(n) (57)
The displacement amplitude (Equation (2)) is expressed as
u(, ) = u() cos(n) (58)
The coefficient matrices of the scaled boundary finite element equation for this one-dimensional
problem (Equations (4)) are determined as (see [49] for derivation)

2 G, n = 0
E0 = (59a)
G, n1

E1 = 0 (59b)
E 2 = n2 E 0 (59c)
 2
r02 0 r0
M0 = E = E0 (59d)
G cs

Starting from the scaled boundary finite element equation in dynamic stiffness (Equation (8))
with the coefficients defined in Equations (59), a continued fraction of order M is constructed by
following the procedure in Section 3. Using Equations (13)–(15) and (18):

1
= (60a)
E0
r0
= (60b)
cs
r0
C∞ = E 0 (60c)
cs
K ∞ = 12 E 0 (60d)

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
224 M. H. BAZYAR AND C. SONG

are derived. Substituting Equations (59) and (60) into Equations (26) yields the coefficients
1
a (1) = (61a)
E0
(1)
b0 = 1
2 (61b)

(1) 1
V = (61c)
E0
(1) r0
b1 = (61d)
cs
c(1) = E 0 ( 14 − n 2 ) (61e)

The coefficients of the continued-fraction solution are determined from Equations (32) and (37) as
(i) r0 1
Y1 = 2 (62a)
cs c(i)
(i)
(i) 2b0 + 1
Y0 = (62b)
c(i)
using the coefficients defined recursively (Equations (43a))

a (i+1) = c(i) (63a)


(i+1) (i) (i)
b0 = −b0 + c(i) Y0 (63b)
(i) (i) (i)
c(i+1) = a (i) − Y0 (2b0 − Y0 c(i) ) (63c)

The coefficients of the continued-fraction solution are evaluated explicitly up to order M = 4:


2 r0 1
Y1(1) = (64a)
E cs (0.25 − n 2 )
0

(1) 2 1
Y0 = (64b)
E (0.25 − n 2 )
0

(2) r0 (0.25 − n 2 )
Y1 = 2E 0 (64c)
cs (2.25 − n 2 )

(2) (0.25 − n 2 )
Y0 = 4E 0 (64d)
(2.25 − n 2 )

(3) 2 r0 (2.25 − n 2 )
Y1 = (64e)
E 0 cs (6.25 − n 2 )(0.25 − n 2 )

(3) 6 (2.25 − n 2 )
Y0 = (64f)
E 0 (6.25 − n 2 )(0.25 − n 2 )

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 225

0.5 3
M=1
M=2
2.5 M=4
0.4

IMAGINARY PART
+ EXACT
2
REAL PART

0.3
1.5
0.2 M=1 1
M=2
0.1 M=4
EXACT 0.5
+

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
DIMENSIONLESS FREQUENCY a0=ω r0/cs DIMENSIONLESS FREQUENCY a0=ω r0/cs

Figure 3. Dynamic-stiffness coefficient of out-of-plane motion of a circular cavity; mode n = 0.

(4) r0 (6.25 − n 2 )(0.25 − n 2 )


Y1 = 2E 0 (64g)
cs (2.25 − n 2 )(12.25 − n 2 )

(4) (6.25 − n 2 )(0.25 − n 2 )


Y0 = 8E 0 (64h)
(2.25 − n 2 )(12.25 − n 2 )

For a one-dimensional problem, the continued-fraction solution of the dynamic-stiffness coefficient


is expressed as
1
S ∞ () = iC∞ + K ∞ − (65)
(1) (1) 1
Y0 + iY1 − (2) (2)
Y0 + iY1 − · · ·
Substituting Equations (64a), (64b), (60c) and (60d) into Equation (65), the order M = 1 continued-
fraction solution is written as
 
∞ 1 (1 − 4n 2 )
S () = E ia0 + −
0
(66)
2 8(1 + ia0 )
In [49], the same expression is derived using the Padé expansion and shown to be identical to the
second-order BGT boundary condition [57].
The dynamic-stiffness coefficients for modes n = 0, 1, 5 and 12 are calculated from the continued-
fraction solution in Equation (65) with the coefficients in Equations (62). The results are non-
dimensionalized with E 0 when n = 0 and with n E 0 when n > 0. The real and imaginary parts
of the non-dimensionalized dynamic-stiffness coefficients are compared with the exact solution
(Equation (55)) in Figures 3–6. Since the continued-fraction solution is based on the high-frequency
behavior, it is more accurate at high frequencies than at low frequencies. When the dimensionless
frequency a0 is higher than the mode number, i.e. a0 >n, an order M = 4 continued fraction leads
to highly accurate results. When the dimensionless frequency is lower than the mode number
(a0 <n), a higher-order continued fraction is required to obtain an accurate result over the whole
frequency range. A parametric study is performed in [49] on the order of Padé series required to

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
226 M. H. BAZYAR AND C. SONG

1 3
M=1
M=2
2.5 M=4
0.8

IMAGINARY PART
+ EXACT
2
REAL PART

0.6
1.5
0.4 M=1 1
M=2
0.2 M=4
+ EXACT 0.5

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
DIMENSIONLESS FREQUENCY a0=ω r0/cs DIMENSIONLESS FREQUENCY a0=ω r0/cs

Figure 4. Dynamic-stiffness coefficient of out-of-plane motion of a circular cavity; mode n = 1.

1.2 1.5
M=1 M=1
M=2 M=2
1 M=4
IMAGINARY PART

M=4 1
+ EXACT + EXACT
0.8
REAL PART

0.6 0.5

0.4
0
0.2

0 −0.5
0 1 2 3 4 5 6 0 1 2 3 4 5 6
DIMENSIONLESS FREQUENCY a0=ω r0/cs DIMENSIONLESS FREQUENCY a0=ω r0/cs

Figure 5. Dynamic-stiffness coefficient of out-of-plane motion of a circular cavity; mode n = 5.

achieve accurate results for a given mode number n. A guidance on selecting the order of Padé
series M based on the mode number n is provided. As a continued fraction can be formulated
as an equivalent Padé series [54], the same guideline should be valid for selecting the order of
continued fraction, which is also confirmed numerically by carrying out the same parametric study
with the continued-fraction solution. For easy reference, the result is reproduced in Table I.
The transient response of the circular cavity initially at rest to an impulse of shear force
shown in Figure 7(a) is evaluated. The duration of the impulse is 4r0 /cs . The peak value is P.
The Fourier transform P() of the impulse P(t) is portrayed in Figure 7(b). The exact solu-
tion for displacement response u(t) is evaluated by performing the inverse Fourier transformation
on u() = P()/Sex ∞ (), where S ∞ () is the exact solution of the dynamic-stiffness coeffi-
ex
cient in Equation (55). The numerical result is computed by applying the high-order transmitting
boundary condition (Equation (51)) directly on the cavity wall. The coefficient matrices [A] and
[B] in Equations (50) are formed by using the coefficients in Equations (62). [A] is diagonal
and [B] is tri-diagonal. Newmark’s time-stepping method with  = 0.5 and  = 0.25 is used to

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 227

1.2 1.5
M=2 M=2
M=4 M=4
1 M=7
M=7

IMAGINARY PART
+ EXACT 1 + EXACT
REAL PART

0.8

0.6 0.5

0.4
0
0.2

0 −0.5
0 5 10 15 0 5 10 15
DIMENSIONLESS FREQUENCY a0=ω r0/cs DIMENSIONLESS FREQUENCY a0=ω r0/cs

Figure 6. Dynamic-stiffness coefficient of out-of-plane motion of a circular cavity; mode n = 12.

Table I. Highest mode number n for which the error norm (see [49, Equation (45)]) of
order M continued fraction is smaller than 1%.
Order of continued fraction M 4 6 8 10 12 14 16 18 20
Mode number n 6 10 17 26 36 48 61 76 93

1.5
P (t)
FORCE AMPLITUDE P(ω)/P

P 0.5

1 3 −0.5
2 4 t = tcs /r0
−1
−P
−1.5
−5 0 5
DIMENSIONLESS FREQUENCY ω r0/cs
(a) (b)

Figure 7. Force impulse: (a) time history and (b) Fourier transform.

integrate Equation (51). The time step is chosen as 0.03r0 /cs . The displacement response is
non-dimensionalized by P/G. The results are shown in Figures 8–11 for modes n = 0, 1, 5 and
12, together with the exact solution. As the order of the transmitting boundary increases, the results
converge to the exact solutions for all modes.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
228 M. H. BAZYAR AND C. SONG

0.12
M=1
0.1
M=2

DISPLACEMENT UG/P
0.08 M=4
0.06 + EXACT
0.04
0.02
0
−0.02
−0.04
−0.06
0 5 10 15 20 25
DIMENSIONLESS TIME tcs/r0

Figure 8. Displacement response of a circular cavity to shear force impulse; mode n = 0.

0.2
M=1
M=2
DISPLACEMENT UG/P

0.1 M=4
+ EXACT

−0.1

−0.2
0 5 10 15
DIMENSIONLESS TIME tcs/r0

Figure 9. Displacement response of a circular cavity to shear force impulse; mode n = 1.

5.2. Rigid strip footing with prolated cross section embedded in a transversely isotropic
half-plane
A rigid mass-less strip footing with a prolated cross section is embedded in a transversely isotropic
half-plane of soil (Figure 12). The cross section of the footing consists of two semi-circles with
radius r0 and a straight line with length 2r0 . A plane-strain state is considered. The notation of the
elasticity matrix for a transversely isotropic material in [10, 45] is used. The material constants of
the half-plane of soil are E hh = 3.864G hv , E hv = 2.863G hv , hh = 0.301 and hv = 0.185 and mass
density . Both frequency- and time-domain analyses are performed.
Taking advantage of symmetry, only half of the half-plane is modeled. The scaling center is
selected at point O (Figure 12). A mesh consisting of nine three-node elements on the arc seg-
ment and six three-node elements on the straight segment (length r0 ) is generated. This mesh has
61 degrees of freedom. In the frequency-domain analysis, the dynamic-stiffness coefficients of
the rigid footing for the vertical motion (Sv∞ ()), the horizontal motion (Sh∞ ()) and the rock-
ing motion (Sr∞ ()) about the scaling center are calculated. The dynamic-stiffness coefficients are

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 229

0.08
M=1
0.06 M=2

DISPLACEMENT UG/P
0.04 M=4
+ EXACT
0.02

−0.02

−0.04

−0.06

−0.08
0 2 4 6 8 10 12
DIMENSIONLESS TIME tcs/r0

Figure 10. Displacement response of a circular cavity to shear force impulse; mode n = 5.

0.05
M=2
0.04
M=4
DISPLACEMENT UG/P

0.03 M=7
0.02 + EXACT
0.01
0
−0.01
−0.02
−0.03
−0.04
−0.05
0 2 4 6 8 10
DIMENSIONLESS TIME tcs/r0

Figure 11. Displacement response of a circular cavity to shear force impulse; mode n = 12.

Scaling
center
O
r0 r0
2 r0

Figure 12. Rigid strip footing with prolated cross section embedded in a transversely isotropic half-plane.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
230 M. H. BAZYAR AND C. SONG

non-dimensionalized with the shear modulus G hv for the two translational motions and with G hvr02
for the rocking motion. The real and imaginary parts of the non-dimensionalized
√ coefficients are
plotted as functions of the dimensionless frequency a0 = r0 /cs (cs = G hv /). In the rigorous
analysis using the full set of base functions and numerical integration, the starting value is deter-
mined at frequency h = 120cs /r0 . The results of the full-set analysis are plotted as dashed lines
in Figure 13. To construct the transmitting boundary, the first eight base functions are selected as
the reduced set. The largest amplitude of eigenvalues is equal to 7. A continued-fraction solution
of order M = 7 is determined from the scaled boundary finite element equation for the reduced
set analysis [49, Equation (53)]. The normalized dynamic-stiffness coefficients are illustrated in
Figure 13 as solid lines. The largest difference with the results of the full-set analysis occurs in
the real part of the horizontal coefficient at a0 = 4. As the imaginary part is much larger than
the real part, except at low dimensionless frequency, this difference of about 0.25 in the real part
represents an error of less than 1% in the amplitude.
In the time domain, displacement responses of the footing to the impulse force shown in
Figure 7(a) with a duration 4r0 /cs are obtained. The vertical displacement, horizontal displacement
and rocking angle are non-dimensionalized with P/G hv , H/G hv and M/(G hvr02 ), respectively,
where P, H and M are the peak values of the vertical force, horizontal force and rocking moment
at the scaling center of the footing. They are plotted as a function of the dimensionless time
tcs /r0 . For the rigorous procedure with the full set of base functions, the time step is chosen
as t = 0.025r0 /cs . The non-dimensionalized displacement responses are portrayed in Figure 14
as dashed lines. An order M = 7 transmitting boundary condition is constructed using the first
eight base functions. Applying the transmitting boundary directly on the footing–soil interface, the
displacement responses are obtained by solving the equation of motion with Newmark’s method.
The normalized displacement responses are illustrated in Figure 14 as solid lines. Results of an
extended finite element mesh are also plotted for comparison. An excellent agreement is observed
up to about t = 9r0 /cs . Afterwards, the extended mesh results are contaminated by the waves
reflected at the outer boundary of the extended mesh.

5.3. Rigid strip footing with rectangular cross section embedded in an isotropic half-plane
A rigid mass-less strip footing with a rectangular cross section of width 2b and height e = b
embedded in an isotropic half-plane (Figure 15) is addressed. A plane-strain state is considered.
The material properties of the half-plane are shear modulus G, Poisson ratio = 0.25 and mass
density . To demonstrate the coupling of the present high-order local transmitting boundary
condition with finite elements, a bounded domain (near field) is introduced adjacent to the footing.
The transmitting boundary is applied to represent the remaining part of the half-plane (far field).
A benefit of introducing the bounded domain is that fewer base functions can be used to model
the unbounded domain as the solution will be smoother at the interface.
Owing to symmetry, only half of the half-plane is modeled. As the authors’ computer program
does not include finite elements, the bounded domain is divided into 45 sub-domains modeled with
the scaled boundary finite elements. Each side of a sub-domain is discretized by one three-node
element so that a sub-domain resembles an eight-node finite element. The scaling center of a
sub-domain is located at its center of geometry. In the scaled boundary finite element model of the
unbounded domain, the scaling center is located at point O. Eighteen three-node elements with
74 degrees of freedom are present at the interface between the bounded and unbounded domains.
In the frequency-domain analysis, the dynamic-stiffness coefficients of the rigid footing Sv∞ ()

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 231

3 40

hv
FULL SET, INTEGRATION
hv

IMAGINARY PART OF S (ω)/G


REAL PART OF S∞(ω)/G 35
2.5 8 MODES, M=7


30

v
v

2
25

1.5 20

15
1
FULL SET, INTEGRATION 10
0.5 8 MODES, M=7
5

0 0
0 1 2 3 4 0 1 2 3 4
(a) DIMENSIONLESS FREQUENCY a0=ω r0/cs DIMENSIONLESS FREQUENCY a0=ωr0/cs

4 40

hv
FULL SET, INTEGRATION
hv

IMAGINARY PART OF S∞(ω)/G


3.5 35
REAL PART OF S∞(ω)/G

8 MODES, M=7
3 30

h
h

2.5 25

2 20

1.5 15

1 FULL SET, INTEGRATION 10


8 MODES, M=7
0.5 5

0 0
0 1 2 3 4 0 1 2 3 4
(b) DIMENSIONLESS FREQUENCY a0=ω r0/cs DIMENSIONLESS FREQUENCY a0=ωr0/cs

20 80
IMAGINARY PART OF Sr (ω)/(Ghv r0)
2
REAL PART OF Sr (ω)/(Ghv r2)
0

FULL SET, INTEGRATION


70
8 MODES, M=7
15 60

50

10 40

30

5 FULL SET, INTEGRATION 20


8 MODES, M=7
10

0 0
0 1 2 3 4 0 1 2 3 4
(c) DIMENSIONLESS FREQUENCY a0=ω r0/cs DIMENSIONLESS FREQUENCY a0=ω r0/cs

Figure 13. Dynamic-stiffness coefficients of rigid strip footing with prolated cross section: (a) vertical;
(b) horizontal; and (c) rocking.

for the vertical motion, Sh∞ () for the horizontal motion and Sr∞ () for the rocking motion about
the center of the base of the footing are computed. For the analyses using the full set of base
functions, the dynamic-stiffness matrices are determined at h = 100cs /r0 as the initial values for

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
232 M. H. BAZYAR AND C. SONG

(a) (b)

(c)

Figure 14. Displacement response to a force impulse applied to a rigid strip footing with prolated cross
section: (a) vertical; (b) horizontal; and (c) rocking.

0.5b
O
b
Scaling center of Bounded
the unbounded e sub-domains
domain

0.5e
Unbounded
domain

Figure 15. A rigid strip footing with rectangular cross section embedded in an isotropic half-plane.

the numerical integration. Dynamic-stiffness coefficients are non-dimensionalized with the shear
modulus G for the two translational motions and Gb2 for the rocking motion. Real and imaginary
parts of the dynamic-stiffness coefficients are plotted as dashed lines in Figure 16 as functions of

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 233

2 30
FULL SET, INTEGRATION

IMAGINARY PART OF S (ω)/G


REAL PART OF Sv (ω)/G
25 10 MODES, M=8


v
1.5 + BOUNDARY ELEMENT METHOD

20

1 15

10
0.5 FULL SET, INTEGRATION
10 MODES, M=8 5
+ BOUNDARY ELEMENT METHOD
0 0
0 1 2 3 4 0 1 2 3 4
(a) DIMENSIONLESS FREQUENCY a =ωb/c DIMENSIONLESS FREQUENCY a =ωb/c
0 s 0 s

3 30
FULL SET, INTEGRATION

IMAGINARY PART OF S∞(ω)/G


REAL PART OF S∞(ω)/G

2.5 25 10 MODES, M=8

h
+ BOUNDARY ELEMENT METHOD
h

2 20

1.5 15

1 10
FULL SET, INTEGRATION
0.5 10 MODES, M=8 5
+ BOUNDARY ELEMENT METHOD
0 0
0 1 2 3 4 0 1 2 3 4
(b) DIMENSIONLESS FREQUENCY a0=ωb/cs DIMENSIONLESS FREQUENCY a0=ωb/cs

7 20
IMAGINARY PART OF Sr∞(ω)/(Gb2)
REAL PART OF Sr (ω)/(Gb )

FULL SET, INTEGRATION


2

6 10 MODES, M=8
15 + BOUNDARY ELEMENT METHOD
5

4
10
3

2
FULL SET, INTEGRATION 5
1 10 MODES, M=8
+ BOUNDARY ELEMENT METHOD
0 0
0 1 2 3 4 0 1 2 3 4
(c) DIMENSIONLESS FREQUENCY a0=ωb/cs DIMENSIONLESS FREQUENCY a0=ωb/cs

Figure 16. Dynamic-stiffness coefficient of a rigid strip footing with rectangular cross section: (a) vertical;
(b) horizontal; and (c) rocking.


the dimensionless frequency a0 = b/cs (cs = G/). A reduced set of 10 base functions with
the largest amplitude of the eigenvalues equal to 9 is chosen for the unbounded domain. The
dynamic-stiffness matrix is approximated by an order M = 8 continued fraction. The normalized

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
234 M. H. BAZYAR AND C. SONG

0.2 0.2
FULL SET, CONVOLUTION FULL SET, CONVOLUTION
0.15 10 MODES, M=7 0.15 10 MODES, M=7
DISPLACEMENT U G/P

DISPLACEMENT U G/H
+ EXTENDED MESH METHOD + EXTENDED MESH METHOD
v

h
0.1 0.1

0.05 0.05

0 0

0 5 10 15 20 0 5 10 15 20
(a) DIMENSIONLESS TIME tc /b (b) DIMENSIONLESS TIME tc /b
s s

0.15
FULL SET, CONVOLUTION
θ Gb2/M

0.1 10 MODES, M=7


+ EXTENDED MESH METHOD
m

0.05
ROCKING ANGLE

0 5 10 15 20
(c) DIMENSIONLESS TIME tcs/b

Figure 17. Displacement response of a rigid strip footing with rectangular cross section to force impulse:
(a) vertical; (b) horizontal; and (c) rocking.

dynamic-stiffness coefficients are illustrated in Figure 16 as solid lines. The results of the boundary
element method [58] are also shown for comparison. Good agreement is observed.
In the time domain, for the foundation initially at rest, the force impulse shown in Figure
7(a) with a duration 4b/cs is prescribed. The vertical displacement, horizontal displacement and
rocking angle are non-dimensionalized with P/G, H/G and M/(Gb2 ), respectively, where P,
H and M are the peak values of the vertical force, horizontal force and rocking moment at the
center of the base of the footing. They are plotted as functions of the dimensionless time tcs /b.
In the rigorous analysis using the full set of base functions, the time step t = 0.025b/cs is
selected. The results of the full-set analysis using convolution integrals are plotted in Figure 17 as
dashed lines. Again, a reduced set of 10 base functions is constructed for the unbounded domain.
An order M = 7 transmitting boundary condition constructed from the scaled boundary finite
element equation for the reduced-set analysis is enforced on the interface between the bounded
and unbounded domains. The time step t = 0.08b/cs is chosen. The displacement responses are
shown in Figure 17 as solid lines. An extended mesh covering a rectangle region of 10b × 10b to
the right of the plane of symmetry is analyzed. It is discretized by 3600 eight-node elements with
a size of 0.1667b × 0.1667b. The total number of nodes is 11 041. The time step is selected as

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 235

t = 0.08r0 /cs . The agreement between the result of the extended mesh and that of the present
transmitting boundary is excellent before the waves reflected at the outer boundary of the extended
mesh reach the foundation at about t = 10b/cs .
To evaluate the efficiency of this continued-fraction-based high-order transmitting boundary, the
computer times spent on the above time-domain analyses are recorded on a laptop computer with
a Centrino 1.73 GHz CPU and 1 GB of RAM. The CPU times required to compute the responses
in the duration 0t10b/cs are recorded. The rigorous solution procedure, extended mesh and the
present transmitting boundary take 56, 58 and 0.75 s, respectively. In the analysis using the present
technique, most of the CPU time is spent on the construction of the continued-fraction solution. As
this high-order transmitting boundary is temporally local, i.e. no convolution integral is evaluated,
the increase of the CPU time with the duration of analysis is linear. For example, when the duration
of analysis is doubled, i.e. 0t20b/cs , the CPU time increases to only 0.88 s, which includes
the time spent on constructing the continued-fraction solution. This example demonstrates that
this technique is well suited for long-time analyses. Note that no extra computational efforts are
needed for anisotropic unbounded domains.

6. CONCLUSIONS

A high-order local transmitting boundary condition is developed for the modeling of wave propa-
gation in unbounded domains of arbitrary geometry. It is based on a continued-fraction solution of
the dynamic-stiffness matrix of the unbounded domain. The coefficient matrices of the continued
fraction are determined recursively and directly from the scaled boundary finite element equation.
After introducing auxiliary variables, the force–displacement relationship corresponding to the
continued fraction is formulated as an equation of motion with symmetric frequency-independent
coefficient matrices. Standard procedures in structural dynamics are applied to perform frequency-
and time-domain analyses. The computationally expensive task of evaluating convolution integrals
is circumvented. Seamless coupling with the finite elements is straightforward. The spatially local
formulation is achieved by applying the technique of the reduced set of base functions. This novel
high-order local transmitting boundary condition is applicable to anisotropic materials without
extra computational efforts. Analytical and numerical examples demonstrate its high accuracy and
efficiency.

ACKNOWLEDGEMENTS
The first author would like to thank the Iranian Ministry of Science, Research and Technology for awarding
him a postgraduate scholarship to study at the University of New South Wales.

REFERENCES
1. Luco JE. Linear soil–structure interaction: a review. In Earthquake Ground Motion and its Effects on Structures,
AMD, Datta SK (ed.), vol. 53. ASME: New York, 1982; 41–57.
2. Kausel E. Local transmitting boundaries. Journal of Engineering Mechanics (ASCE) 1988; 114:1011–1027.
3. Givoli D. Non-reflecting boundary conditions: a review. Journal of Computational Physics 1991; 8:1–29.
4. Tsynkov SV. Numerical solution of problems on unbounded domains: a review. Applied Numerical Mathematics
1998; 27:465–532.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
236 M. H. BAZYAR AND C. SONG

5. Astley RJ. Infinite elements for wave problems: a review of current formulations and an assessment of accuracy.
International Journal for Numerical Methods in Engineering 2000; 49:951–976.
6. Givoli D. High-order non-reflecting boundary conditions: a review. Wave Motion 2004; 39:319–326.
7. Wolf JP. Dynamic Soil–Structure Interaction. Prentice-Hall: Englewood Cliffs, NJ, 1985.
8. Wolf JP. Soil–Structure-Interaction Analysis in Time Domain. Prentice-Hall: Englewood Cliffs, NJ, 1988.
9. Givoli D. Numerical Methods for Problem in Infinite Domains. Elsevier: Amsterdam, 1992.
10. Wolf JP, Song Ch. Finite-element Modelling of Unbounded Media. Wiley: Chichester, 1996.
11. Hall WS, Oliveto G. Boundary Element Methods for Soil–Structure Interaction. Kluwer Academic Publishers:
Dordrecht, 2003.
12. Beskos DE. Boundary element methods in dynamic analysis. Applied Mechanics Reviews (ASME) 1987; 40:1–23.
13. Kausel E, Roesset JM, Wass G. Dynamic analysis of footings on layered media. Journal of Engineering Mechanics
(ASCE) 1975; 101:679–693.
14. Keller JB, Givoli D. Exact non-reflecting boundary conditions. Journal of Computational Physics 1989; 82:
172–192.
15. Lysmer J, Kuhlemeyer RL. Finite dynamic model for infinite media. Journal of Engineering Mechanics (ASCE)
1969; 95:859–877.
16. Bayliss A, Gunzburger M, Turkel E. Boundary conditions for the numerical solution of elliptic equations in
exterior regions. SIAM Journal on Applied Mathematics 1982; 42:430–451.
17. Bettess P. Infinite Elements. Penshaw Press: Sunderland, U.K., 1992.
18. Berenger JP. A perfectly matched layer for the absorption of electromagnetic waves. Journal of Computational
Physics 1994; 114:185–200.
19. Engquist B, Majda A. Radiation boundary conditions for acoustic and elastic wave calculations. Communications
on Pure and Applied Mathematics 1979; 32:314–358.
20. Higdon RL. Absorbing boundary conditions for difference approximations to the multi-dimensional wave equation.
Mathematics of Computation 1986; 176:437–459.
21. Grote M, Keller J. Exact non-reflecting boundary conditions for the time-dependent wave equation. SIAM Journal
of Applied Mathematics 1995; 55:280–297.
22. Grote MJ, Keller JB. Exact non-reflecting boundary condition for elastic waves. SIAM Journal on Applied
Mathematics 2000; 60:803–819.
23. Grote MJ, Keller JB. Non-reflecting boundary conditions for Maxwell’s equations. Journal of Computational
Physics 1998; 139:327–342.
24. Hagstrom T, Hariharan S. A formulation of asymptotic and exact boundary conditions using local operators.
Applied Numerical Mathematics 1998; 27:403–416.
25. Bayliss A, Turkel E. Radiation boundary conditions for wave-like equations. Communications on Pure and
Applied Mathematics 1980; 33:707–725.
26. Huan RN, Thompson LL. Accurate radiation boundary conditions for the time-dependent wave equation on
unbounded domains. International Journal for Numerical Methods in Engineering 2000; 47:1569–1603.
27. Thompson LL, Huan RN, He DT. Accurate radiation boundary conditions for the two-dimensional wave equation
on unbounded domains. Computer Methods in Applied Mechanics and Engineering 2001; 191:331–351.
28. Krenk S. Unified formulation of radiation conditions for the wave equations. International Journal for Numerical
Methods in Engineering 2002; 53:275–295.
29. Guddati M, Tassoulas J. Continued-fraction absorbing boundary conditions for the wave equations. Journal of
Computational Acoustics 2000; 8:139–156.
30. Guddati M, Lim K. Continued fraction absorbing boundary conditions for convex polygonal domains. International
Journal for Numerical Methods in Engineering 2006; 66:949–977.
31. Givoli D, Patlashenko J. An optimal high-order non-reflecting finite-element scheme for wave scattering problems.
International Journal for Numerical Methods in Engineering 2002; 53:2389–2411.
32. Givoli D, Neta B. High order non-reflecting boundary conditions for time-dependent waves. Journal of
Computational Physics 2003; 186:24–46.
33. Hagstrom T, Warburton T. A new auxiliary variable formulation of high-order local radiation boundary condition:
corner compatibility conditions and extensions to first-order systems. Wave Motion 2004; 39:327–338.
34. van Joolen V, Neta B, Givoli D. High-order Higden-like boundary conditions for exterior transient wave problems.
International Journal for Numerical Methods in Engineering 2005; 63:1041–1068.
35. Kechroud R, Antoine X, Soulaimani A. Numerical accuracy of a Padé-type non-reflecting boundary condition for
the finite element solution of acoustic scattering problems at high-frequency. International Journal for Numerical
Methods in Engineering 2005; 64:1275–1302.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme
A CONTINUED-FRACTION-BASED TRANSMITTING BOUNDARY 237

36. Givoli D, Hagstrom T, Patlashenko I. Finite-element formulation with high-order absorbing conditions for
time-dependent waves. Computer Methods in Applied Mechanics and Engineering 2006; 195:3666–3690.
37. Paronesso A, Wolf JP. Global lumped-parameter model with physical representation for unbounded medium.
Earthquake Engineering and Structural Dynamics 1995; 24:637–654.
38. Ruge P, Trinks C, Witte S. Time-domain analysis of unbounded media using mixed-variable. Earthquake
Engineering and Structural Dynamics 2001; 30:899–925.
39. Paronesso A, Wolf JP. Recursive evaluation of interaction forces and property matrices from unit-impulse
response functions of unbounded medium based on balancing approximation. Earthquake Engineering and
Structural Dynamics 1998; 27:609–618.
40. Yan JY, Zhang CH, Jin F. A coupling procedure of FE and SBFE for soil–structure interaction in the time
domain. International Journal for Numerical Methods in Engineering 2004; 59:1453–1471.
41. Song Ch, Wolf JP. The scaled boundary finite-element method–alias consistent infinitesimal finite-element cell
method—for elastodynamics. Computer Methods in Applied Mechanics and Engineering 1997; 147:329–355.
42. Zhang X, Wegner JL, Haddow JB. Three-dimensional dynamic soil–structure interaction analysis in the time
domain. Earthquake Engineering and Structural Dynamics 1999; 28:1501–1524.
43. Fan SC, Li SM, Yu GY. Dynamic fluid–structure interaction analysis using boundary finite element method-finite
element method. Journal of Applied Mechanics (ASME) 2005; 72:591–598.
44. Lehmann L, Ruberg T. Application of hierarchical matrices to the simulation of wave propagation in fluids.
Communications in Numerical Methods in Engineering 2006; 22:489–503.
45. Bazyar MH, Song Ch. Time-harmonic response of non-homogeneous elastic unbounded domains using the scaled
boundary finite-element method. Earthquake Engineering and Structural Dynamics 2006; 35:357–383.
46. Bazyar MH, Song Ch. Transient analysis of wave propagation in non-homogeneous elastic unbounded domains
by using the scaled boundary finite-element method. Earthquake Engineering and Structural Dynamics 2006;
35:1787–1806.
47. Song Ch, Wolf JP. The scaled boundary finite-element method—a primer: solution procedures. Computers and
Structures 2000; 78:211–225.
48. Song Ch. Dynamic analysis of unbounded domains by a reduced set of base functions. Computer Methods in
Applied Mechanics and Engineering 2006; 195:4075–4094.
49. Song Ch, Bazyar MH. A boundary condition in Padé series for frequency domain solution of wave propagation
in unbounded domains. International Journal for Numerical Methods in Engineering 2007; 69:2330–2358.
50. Song Ch, Bazyar MH. Development of a fundamental-solution-less boundary element method for exterior wave
problems. Communications in Numerical Methods in Engineering 2007; DOI: 10.1002/cnm.964.
51. Wolf JP, Song Ch. The scaled boundary finite-element method—a primer: derivation. Computers and Structures
2000; 78:191–210.
52. Song Ch. Weighted block-orthogonal base functions for static analysis of unbounded domains. Proceedings of
the 6th World Congress on Computational Mechanics, Beijing, China, 5–10 September 2004; 615–620.
53. Press WH, Teukolsky SA, Vetterling WT, Flannery BP. Numerical Recipes, Chapter 15. Cambridge University
Press: Cambridge, MA, 1992.
54. Baker G, Graves-Morris P. Padeé Approximants. Cambridge University Press: Cambridge, MA, 1996.
55. Moler CB, van Loan C. High-order hierarchical a-stable and l-stable integration methods. International Journal
for Numerical Methods in Engineering 1993; 36:2607–2624.
56. Chopra AK. Dynamics of Structures: Theory and Applications to Earthquake Engineering. Prentice-Hall:
Englewood Cliffs, NJ, 1995.
57. Harari I, Djellouli R. Analytical study of the effect of wave number on the performance of local absorbing
boundary conditions for acoustic scatting. Journal of Applied Numerical Mathematics 2004; 50:15–47.
58. Wang YP, Rajapakse RKND. Dynamic of rigid strip foundations embedded in orthotropic elastic soils. Earthquake
Engineering and Structural Dynamics 1991; 20:927–947.

Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2008; 74:209–237
DOI: 10.1002/nme

You might also like