Computers and Geotechnics 114 (2019) 103143

Contents lists available at ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Research Paper

An ABAQUS toolbox for soil-structure interaction analysis T

a b a,⁎
W. Zhang , E. Esmaeilzadeh Seylabi , E. Taciroglu
a
University of California, Los Angeles, CA 90095, USA
b
University of Nevada, Reno, NV 89557, USA

ARTICLE INFO ABSTRACT

Keywords: It is well established that the soil-structure interaction (SSI) effects can bear important consequences under
Domain Reduction Method strong earthquakes, and their accurate quantification can become a critical issue in designing earthquake-re-
Perfectly-Matched-Layer sistant structures. In general, SSI analyses are carried out by means of either direct or substructure methods. In
Soil-structure interaction either option, the numerical models feature truncated and/or reduced-order computational domains. For
Finite element method
truncation, boundary representations that perfectly absorb the outgoing waves and enable the consistent pre-
ABAQUS
scription of input motions are crucial. At the present time, the aforementioned capabilities are not broadly
User-defined element (UEL)
available to researchers and practicing engineers. To this end, we implement the so-called Domain Reduction
Method (DRM) and Perfectly-Matched-Layers (PMLs) in ABAQUS, by computing and prescribing the effective
nodal forces, and through a user-defined element (UEL) subroutine, respectively. We then verify the accuracy
and stability of these implementations for both homogeneous and heterogeneous soil domains, vertical and
inclined incident SV waves, and two- and three-dimensional problems. Finally, we present two useful application
examples of using the implemented features—namely, the extraction of impedance functions, the response
analysis of buried structures subjected to inclined plane waves. The implemented codes for both DRM and PML
will be disseminated for broader use.

1. Introduction surrounding soil media along with the structure. This approach is
known as the direct modeling [1,3] method. Apparently, it is not pos-
All civil structures have foundations and other support elements sible to discretize the semi-infinite soil domain with a finite number of
that either rest on, or are embedded in, soil. Because of complexities in elements; and thus, it is necessary to truncate it by introducing ap-
modeling the mechanical behavior of soils, and the high degree of propriate boundary conditions. For an exact representation of the
uncertainty and variability in their properties, it is not uncommon omitted domain—dubbed the far-field, the introduced boundaries on
among structural engineers to completely ignore their effects on the the computational domain (the near-field) must have the ability to
structural system. This simplistic approach, wherein the soil-structure transmit the energy of the outgoing and incoming waves perfectly. In
interaction (SSI) effects are unaccounted for, might yield acceptable problems where the source of excitation is inside the near-field, all
designs for certain cases—for example, for lightweight aboveground waves impinging upon the imposed boundaries are outgoing; and the
structures resting on, or stiff underground structures buried in, rock and inserted boundary condition must absorb the energy of these outgoing
stiff soils [1]. Nevertheless, the omission of SSI effects can also bear waves through the so-called, absorbing-boundary-conditions (ABCs).
perilous consequences under strong earthquakes—for example, for a Lysmer and Kuhlemeyer [4] proposed the first local ABC,1 which
massive structure resting on soft soil [2]. For buried structures, al- could only absorb waves traveling along a prescribed direction. Higdon
though the inertially induced tractions may become negligible, the [7] proposed the m-th order multi-directional boundary condition that
nominal contrast between the flexibilities of the foundation system and can absorb traveling waves with m different angles of incidence per-
its surrounding soil may significantly affect their responses. fectly. Although the accuracy of this boundary condition increases by
One approach to take the effects of SSI into account is to use the m, its usage in application is limited to m 2. This is because it is very
finite element method (FEM) to model a portion of the supporting/ complicated to define high-order derivatives in standard numerical

Corresponding author.
⁎

E-mail address: [email protected] (E. Taciroglu).

1
ABCs may be classified as nonlocal and local. Nonlocal ABCs are derived by using analytical solutions of waves propagating in the far-field, and involves a
boundary integral operator, meaning that all the nodes on the boundary interact with each other. However, local ABCs are represented as a differential operator, so
that after discretization, each node on the boundary interacts only with its neighbors, resulting less computational costs [5,6].

https://doi.org/10.1016/j.compgeo.2019.103143
Received 24 February 2019; Received in revised form 3 May 2019; Accepted 18 June 2019
Available online 27 June 2019
0266-352X/ © 2019 Elsevier Ltd. All rights reserved.
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

schemes, such as the finite element method. Since then, different high- domain solution with fixed boundaries. Excellent agreement is acquired
order ABCs have been proposed [8]. Almost all these boundary condi- in different cases in terms of loading conditions, material properties and
tions are limited to scalar problems such as electrodynamics and geometry dimensions.
acoustic problems. Although many infinite domain problems involve Besides prescribing proper ABCs, direct modeling of SSI problems
vectorial elastic waves, only a few high-order ABCs have been devel- also usually suffer from defining inappropriate input ground motions
oped for elastodynamics problems thus far [9,10]. However, all of them [31,32]. So far the Domain Reduction Method (DRM) [3,33] is re-
suffer from long-time instability2 issues [11]. Recently, Baffet et al. [11] cognized to be the best approach for modeling a semi-infinite domain
proposed the first long-time stable high-order local ABC. Thus far, this under remote excitations. In this study, we first implement the DRM by
high-order ABC is only available for the relatively simple case of 2D computing the effective nodal forces based on the formulations derived
elastodynamics with a single artificial boundary and two physical in [33], and then thoroughly verify it by comparing the DRM-generated
boundaries. Extensions to more involved cases, such as problems in SSI free-field responses with either analytical solutions (if exist) or results
analysis, which consist of a single physical boundary and three artificial computed from 1D site response analysis. Moreover, for both homo-
boundaries, do not yet exist. geneous and heterogeneous soil layers, vertical and inclined incident SV
The paraxial boundary is also an ABC that is based on the paraxial waves, 2D and 3D domains, we achieve excellent accuracy.
approximations of the one-way wave equations, and has been devel- In the end, by using the powerful coupled DRM-PML system, (1) we
oped for both scalar wave [12,13] and elastic wave equation [14–16]. compute the impedance functions for different problems and then
However, it was demonstrated that this approximation is very accurate compare our results with analytical or semi-analytical solutions. Given
only for high-frequency waves and for waves impinging the boundary the facts that impedance functions are indispensable for substructure
with small incident angles [15]. method, but they have been devised only for specific conditions, we can
Besides the high-order ABCs and the paraxial boundary, another use DRM-PML system to compute them regardless of foundation shapes
class of local ABCs is perfectly matched layers (PMLs), which were and soil profiles, with extremely low computational cost and desirable
originally introduced by Berenger [17] for absorbing electromagnetic accuracy. (2) A buried rectangular structure in 2D domain is modeled to
waves, and extended to elastodynamic problems by Basu and Chopra study the effects of the angle of incidence. Inclined SV waves with
[18], but using a rather complicated time integration scheme [19] to different angles of incidence are applied, and the resulting horizontal
compute the internal forces. In short, a PML is an absorbing layer ad- and vertical accelerations, maximum axial forces and bending mo-
jacent to the finite computational domain–i.e., near-field–with two ments, and maximum deformations of the structures are plotted.
main properties [10]:
2. Perfectly-Matched-Layer (PML)
• It results in no reflections at the truncated near-field boundary (i.e.,
“perfect matches” it) for all non-zero-frequency impinging waves, To render the semi-infinite extent of soils, we will truncate it by
irrespective of their angles of incidence. using a robust wave absorbing boundary, Perfectly-Matched-Layer
• It attenuates the wave energy within itself. (PML). The PML can eliminate reflections at the truncated near-field
boundary for all non-zero-frequency impinging waves, irrespective of
Due to the applicability for heterogeneous media [20], PMLs have their angles of incidence. And the wave energy will be attenuated ra-
become more popular than high-order ABCs in dealing with infinite pidly within its zones.
domain problems. Kucukcoban and Kallivokas [20] proposed a mixed
finite element implementation of the displacement-stress unsplit-field
2.1. Implementation
formulation in 2D elastic heterogeneous media, with superior stability
and efficacy. Later they derived the symmetric formulation of this ap-
In this study we implement both 2D and 3D PMLs developed by
proach [21] and also extended it to 3D elastic heterogeneous media
Kucukcoban and Kallivokas [20] and Fathi et al. [34], respectively, that
[22].
can be used for wave propagation related simulation in an arbitrarily
However, considering the relatively complex formulations of the
heterogeneous media. The PML is treated via an unsplit-field, but
PMLs, few of them have been implemented to commercial finite ele-
mixed-field, symmetric displacement-stress formulation, which then
ment packages for broader use, which eventually limits the consider-
can be directly coupled to a standard displacement-only formulation for
able advantages of combining the PMLs with other advanced tools, e.g.,
the interior domain, resulting to a relatively computationally efficient
contact elements, nonlinear constitutive models, etc. Plaxis [23] and
hybrid scheme.
FLAC [24] —two commonly used commercial software in geotechnical
In order to implement this PML-type element in ABAQUS, we write
engineering area—are equipped with viscous and free-field ABCs. Ob-
a user-defined element (UEL) subroutine in Fortran90. Adding the
viously, those two ABCs are not capable of efficiently absorbing the
convenient features of the solver in ABAQUS, we do not need to be
outgoing waves when the incident angle is neither 0 nor 90 degree.
concerned regarding the assembling process of the matrices, even
Basu [25] implemented an elastic PML in LS-DYNA [26], but it is only
though this mixed PML element consists of displacement and stress
available in 3D media and with explicit integration scheme. Poul and
components. The principal component required for constructing a PML
Zerva [27] implemented a viscoelastic PML in ABAQUS [28] based on
is the complex stretching function. And the idea is to “stretch” the
the work of Zheng and Huang [29]. However, it is 2D plane-strain only,
originally physical coordinates to the virtually infinite coordinates. The
and the recursive convolution technique used in the evaluation of in-
complex stretching function we employ in this study is defined as,
ternal forces incurs more computational efforts.
In this work, we employ the symmetric hybrid PML formulation 1
i (x i , )= (x i ) + (x i )
proposed by Kucukcoban and Kallivokas, and Fathi et al. [21,22], by j (1)
writing a user-defined element (aka UEL) subroutine with Hilber-
Hughes-Taylor implicit time integration scheme (HHT- method [30]), where and are the scaling functions that stretch the coordinate
for simulation of wave propagation in both 2D and 3D heterogeneous variable x, and the attenuation function that enforces the amplitude
half-spaces. This implementation has been verified by comparing our decay of outgoing waves, respectively, which are given in terms of
results obtained from PML-truncated domain against an enlarged polynomials, as
m
(x i x i0) ni
i (x i ) =1+ 0 , x i0 xi xit
2
The solution grows exponentially after a sufficiently long amount of time.
LPML, i (2a)

2
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 1. Illustration of a 2D PML domain attached to the interior domain.

m
(x i x i0) ni (m + 1) cp 1
i (x i ) = , xi0 xi x it 0 = log
0
LPML, i (2b) 2LPML, i R (3b)

where i = x , y and z , and 0 and 0 are two user-chosen parameters that where cp is the P-wave velocity, R is a user-tunable reflection coefficient
control the amplitude decay, is component of the outward unit normal and b is a characteristic length of the domain. In the following study we
at the interface in i direction, and m represents the polynomial degree. always set m = 2, R = 10 10 and b = ten times the average element size.
And here we use the suggested expressions [20] for them, as shown in Also, please be advised that the outer PML boundary should always be
Eq. (3). x i0 and x it denote the coordinates of the inner and outer PML fixed (in terms of the displacement field), and for heterogeneous soils,
boundaries. LPML, i is the thickness of the PML boundary in i direction. constant 0 is required for different soil layers. See Fig. 1 for an illus-
tration of those parameters in the 2D PML domain.
(m + 1) b 1 Based on the definition of the complex stretching functions, we can
0 = log
2LPML, i R (3a) derive the corresponding semi-discrete forms and submatrices for PMLs

Fig. 2. Configuration of (a) the original 2D semi-infinite rod problem and (b) the PML-truncated domain.

Fig. 3. Comparisons of horizontal displacements for two selected points.

3
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 4. Configuration of (a) the original 2D three-layered half-space problem and (b) the PML-truncated domain.

Table 1 FR = ¨ t+
MPML u t + (1 + )(Ftext
+ t
KPML ut + t CPML ut + t )
Material properties of the three-layered soil deposits. (Ftext KPMLut CPML u t ) (5b)
Shear wave velocity Vs (m/s) Poisson’s Ratio Density (kg/m3 )
For 3D domain,
Layer 1 200 0.3 2000
Layer 2 300 0.3 2000
1 1+ t
K eff = MPML + CPML + (1 + ) KPML + GPML
Layer 3 400 0.3 2000 t2 t (6a)

FR = ¨ t+
MPML u t + (1 + )(Ftext
+ t
GPML u¯ t + t KPML ut + t
in 2D and 3D domains [34], which are included in the Appendix.
In ABAQUS, the Hilber-Hughes-Taylor (HHT- method) implicit CPML ut+ t)
time integration scheme is the default approach for solving both linear (Ftext GPML u¯ t KPML ut CPML ut ) (6b)
and nonlinear problems. And in the UEL subroutine, the effective
where , and are time increment parameters for HHT- method that
stiffness matrix (AMATRX, K eff ) and the residual vector (RHS, FR ),
have the conditions as,
which have the relationship in incremental form, as shown in Eq. (4),
need to be computed and updated in every step. 1/3 0, = (1 + ) 2/4, = 1/2 + (7)
K eff u = FR (4) The default value is = 1/20 .
From Eq. (22) we can deduce that the governing equation for PML
where K eff and FR can be derived, as in 3D domain is a third-order ODE, however the solver in the ABAQUS
For 2D domain, is designed for second-order ODE. Therefore here we need to store the
1 1+ variable ū t (see Eq. (22b)) into SVARS (state variables in UEL). By
K eff = MPML + CPML + (1 + ) KPML using the extended Newmark-beta method [34], ūt + t can be expressed
t2 t (5a)
as,

Fig. 5. Comparisons of horizontal and vertical displacements for four selected points.

4
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 6. Contour plots of the total displacement field at different times obtained by using the PML boundary.

t2 t 1 2.2.1. 2D semi-infinite rod

u¯ t + t = u¯ t + t ut + u + t 3u
¨t + t 3 ut + t
2 6 (8) A 2D semi-infinite rod (see Fig. 2(a)) with uniform cross-section and
homogeneous material properties (i.e., Vs =
where = 1/12 is a parameter chosen to be consistent with the average- 200 lm/s, = 0.3, = 2000 kg/m3 ), excited by a horizontal load using
acceleration scheme. the function defined in Eq. (18) with an amplitude of 100 kN, a central
frequency of 5 Hz and a total period of t = 1.0 s, is analyzed by at-
taching (i) PML elements (see Fig. 2(b)), (ii) infinite elements (CINPE4)
2.2. Verification and placing (iii) dashpots on the right end (i.e., x = 30 m) in horizontal
direction with coefficient cx = Vp/2 . The rod has a thickness of 1 m,
In this section, we comprehensively verify the accuracy and stability and here we use the 4-node plane-strain element (CPE4) with a size of
of the implemented PML, and compare its performance with other ABCs 1 m by 1 m for the regular domain. The same element size applies to the
(i.e., Lysmer and Kuhlemeyer [4] and ABAQUS built-in infinite ele- PML boundary. Besides, the vertical DOF is fixed and hence only the
ment), in both 2D and 3D domains, homogeneous and heterogeneous horizontal DOF is free. To investigate the performance of the PML and
soil layers, under concentrated and distributed loads. Lysmer and other ABCs, we create an enlarged domain with a total size of
Kuhlemeyer [4] ABC simply places dashpots on the outside boundary. 1 m× 400 m , element size of 1 m× 1 m and fixed boundaries on the
The infinite elements in ABAQUS, per descriptions in the manual [28], outer surface, where the domain is large enough to avoid the reflected
provide “quiet” boundaries to the finite element model through the waves enter the regular domain, to be the reference solutions.
effect of a damping matrix and a suppressed stiffness matrix. And it Fig. 3 shows the comparisons of horizontal displacements for two
does not provide perfect transmission of energy out of the mesh except selected points obtained by using different ABCs, i.e., PML, dashpots
in the case of plane body waves impinging orthogonally on the and infinite element, against the enlarged domain solutions. All three
boundary in an isotropic medium. ABCs show excellent agreement, which is because both Lysmer and

5
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 7. Quarter model of a PML-truncated 3D homogeneous half-space model.

Fig. 8. Time history and its Fourier amplitude plots for the applied surface pressure.

Fig. 9. Comparisons of z displacements for center and corner points.

6
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 10. Configuration of the PML-truncated 3D three-layered half-space model.

Fig. 11. Time histories of displacements of six selected points.

7
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 12. Contour plots of the total displacement field at different times.

Fig. 13. (a) Configuration of modeling semi-infinite domain by using DRM and ABCs, and (b) equivalent 1D site response analysis model for evaluation of free-field
response for DRM interface.

Fig. 14. Schematic propagation of the inclined incident SV wave in a flat homogeneous half-space in 2D and 3D domains.

8
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 15. Displacement, velocity, acceleration and Fourier amplitude plots for the applied Ricker pulse.

Kuhlemeyer [4] and ABAQUS infinite plane-strain element (CINPE4) are presented in Fig. 6, which on the other hand, proves that the PML
are designed for 1D wave propagation analysis. boundary absorbs all the outgoing waves.

2.2.2. 2D three-layered half-space 2.2.3. 3D homogeneous half-space

A 2D three-layered half-space case (see Fig. 4(a)) with material A 3D homogeneous half-space model (see Fig. 7) with material
properties summarized in Table 1, excited by a vertical point load using properties as Vs = 1 m/s, = 0.25 and = 1 kg/m3 , is subjected to a
the function defined in Eq. (18) with an amplitude of 100 kN, a central uniformly distributed vertical pressure on the grey area. The function of
frequency of 5 Hz and a total period of t = 2.0 s, is analyzed by at- the applied pressure is defined in [19] with parameters as
taching (i) PML elements (see Fig. 4(b)), (ii) infinite elements (simply td = 10 s, f = 3 rad/s . The resulting time history and Fourier ampli-
replacing PML by CINPE4) and placing (iii) dashpots on both bottom tude plots are shown in Fig. 8. This problem is also analyzed by at-
( y = 110 m) and side boundaries ( x = ± 110 m) with coefficients de- taching (i) PML elements, (ii) 3D infinite elements (CIN3D8) and pla-
fined in Eq. (10). Similarly, here we use the 4-node plane-strain element cing (iii) dashpots on the outer surface with coefficients defined in Eq.
(CPE4) with a size of 1 m by 1 m for the interior domain. The same (10). 8-node tri-linear element (C3D8) with a size of
element size applies to the PML boundary. Again, an enlarged domain 0.1 m× 0.1 m× 0.1 m is used for the interior domain and t = 0.05 s is
with a total size of 750 m× 1500 m , element size of 1 m× 1 m and fixed used for the time step. The same element size applies to the PML
boundaries on the outer surface is constructed to be used for the re- boundary. Further, in order to investigate the performance of the PML
ference solutions. and other ABCs, an enlarged domain with more than 10 million ele-
cx = ·Vs· Ab ments are required for such problem [25]. Due to the limitation of the
For the bottom boundary
c y = ·Vp·Ab maximum number of elements that the software imposes, it is however
(9a) impossible to be done in ABAQUS. Therefore here we adopt the re-
cx = ·Vp·As ference solutions from [22].
For the side boundaries Fig. 9 shows the computed vertical displacements for the center and
c y = ·Vs· As (9b) corner points on the grey surface. Still, the PML displays obviously
where is the density of the soil; Ab and As are the areas of the bottom superb behavior compared with dashpots and infinite elements. This
and side soil elements that correspond to the dashpots used. example demonstrates that the PML can shrink the computational do-
Fig. 5 shows the comparisons of vertical displacements for CP1 and main as close to the loading area as possible, and maintains the same
CP2, both horizontal and vertical displacements for CP3 and CP4, ob- excellent performance.
tained by using different ABCs, i.e., PML, dashpots and infinite element,
against the enlarged domain solutions. As seen, the PML boundary 2.2.4. 3D three-layered half-space
behaves perfectly for all selected points, regardless of horizontal and A 3D three-layered half-space case (see Fig. 10) with material
vertical displacements. However, for dashpots and infinite elements, properties summarized in Table 1, excited by a vertical point load using
errors appear after the reflected waves enter the interior domain, which the function defined in Eq. (18) with an amplitude of 100 kN, a central
is because the outgoing waves are not perfectly absorbed. And due to frequency of 5 Hz a total simulation time of t = 10.0 s and t = 0.001 s,
the fact that these ABCs are sensitive to the angle of incidence, the is analyzed by using PML boundary. 8-node tri-linear element (C3D8)
discrepancies become more apparent for the corner point (i.e., CP4). with a size of 1 m× 1 m× 1 m is used for the interior domain and the
Besides, the contour plots for the total displacement field at different same element size for the PML elements.
times (i.e., t = 0.4, 0.6, 0.8, 1.0, 1.5 s ) obtained by using PML boundary, Because for such a problem there are no reference solutions in

9
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

m, fRicker = 5 Hz and t0 = 0.3 s , is used for the incident wave function.

literature and we are unable to conduct an enlarged domain analysis in
Angle of incidence
ABAQUS, we utilize this model to examine the long-time stability of the
implemented PML. As we know, some other PML formulations may

= 25o and
suffer long-time instability issue [35]. This simulation has a total of

= 45o
T / t = 10/0.001 = 10000 steps, which is considered adequate to ex-
amine the stability of the model. The time histories of displacements of
25o

0o

0o
s
six selected points are shown in Fig. 11. The locations of CP1, CP2, CP3
where b = (V0/VH )1/ n , V0 = 200 m/s, VH = 400 m/s, n = 0.5, H = 25 m .

where b = (V0/VH )1/ n , V0 = 200 m/s, VH = 400 m/s, n = 0.5, H = 25 m .

and CP4 are indicated in Fig. 10. CP5 and CP6 have the coordinates as
(25, 25, 0) and (25, 25, 25) , respectively. And the contour plots of the
total displacement field at different times are also presented in Fig. 12.
It can be observed that the whole model becomes silent after around
t = 0.7 s , and there is no visible reflection at the interface of the interior
domain and the PML boundary.

3. Domain Reduction Method (DRM)

DRM is a two-step finite element procedure proposed by Bielak et al.

[33] for modeling the seismic responses of heterogeneous subdomains.
The most appealing advantage of DRM is that it enables the conversion
of the half-space problem to an equivalent one in which the effects of
incoming waves due to remote excitations are translated into equiva-
4

lent nodal forces that are applied inside a domain that is truncated by
= 0.3 are selected. The Ricker pulse defined in Eq. (18), with ARicker = 1e

ABCs.

3.1. Implementation
,

In DRM, the equivalent nodal force vector is computed using Eq.

z n

z n
H

(10) per [33] and applied to the nodes located at a single layer of
b)

b)

elements that form the boundary between the ABCs and the near-field
Vs = VH b + (1

Vs = VH b + (1

domain as shown in Fig. 13(a),

200 m/s

200 m/s

Pieff 0
Peff = Peff = Mbe+u
¨ 0e Cbe+u 0e Kbe+u0e
Vs

b
Peff + Meb+u
¨ 0b + Ceb+u 0b + K eb+u0b
e (10)
Heterogeneous soil layer with vertical incident SV
Homogeneous soil layer with inclined incident SV

Homogeneous soil layer with inclined incident SV

where the subscripts i, b and e refer to the nodes inside the domain of
interest, along the inside and outside boundary of the one layer of
elements, respectively. The terms u0 and Peff respectively denote the
free-field displacements and forces along nodes of the one layer of
elements. M +, C + and K + are the mass, damping and stiffness ma-
trices assembled for only the single layer of elements that form the
interface between the exterior and interior domains. In this study, we
Heterogeneous soil layer.

(1) use analytical solutions (if exist) or perform 1D site response ana-
wave in 2D domain.

wave in 3D domain.

wave in 3D domain.

lysis to obtain the free-field response, (2) extract the nodes’ coordinates
and meshes’ connectivity information from ABAQUS input file, (3) use
= 2000 kg/m3 and Poisson’s ratio

the information and the computed free-field response in step (1) to

Definition

compute the M +, C +, K + matrices, and Peff vector, (4) insert the

computed Peff vectors, that are considered as concentrated force time-
history functions for all DRM nodes, into the ABAQUS input file. A
Five verification problems of the DRM implementation.

routine has been developed in MATLAB [36] to perform the afore-

mentioned tasks.
DRM-2D-Hetero-Ver (i. Ricker pulse, ii.

As for the inclined incident SV waves, it is extremely difficult to

calculate free-field response by conducting 1D site response analysis
and analytical solutions are only available for the homogeneous half-
space case in time domain. Here we adopt the time domain analytical
Note: for all 5 cases, the same density

solutions for the inclined SV-wave propagation in a homogeneous flat

DRM-3D-Hetero-Ver

half-plane derived by [37]. The displacement field solution is given in

DRM-2D-Homo-Inc

DRM-3D-Homo-Inc
earthquake motion)

Eq. (11)

u x (x , z )
u z (x , z )
= Usi
+ cos
+ sin
s
s
f ( x
cs
sin s +
z
cs
cos s )
+t +
Label

Usr
cos
+ sin
s
f ( x
cs
sin s
z
cs
cos s )
+t +
Dimension

s
Table 2

( )
+ sin p x z
Upr f sin p cos p +t
2D

3D

+ cos p
cp cp
(11)

10
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 16. Acceleration and Fourier amplitude plots for the applied earthquake motion.

Fig. 17. Numerical models constructed for DRM verification problems in (a) 2D, and (b) 3D domains.

where u x and uz are the displacement fields in horizontal and vertical sin(2 s )sin(2 p) k 2cos2 (2 s )
Usr = Usi
directions, respectively. Fig. 14 shows the schematic propagation of the sin(2 s )sin(2 p) + k 2cos2 (2 s ) (16a)
inclined incident SV wave in a flat homogeneous half-space in 2D and
3D domains. Eq. (11) is originally derived for plane-wave, but it can be 2k2sin(2 s )cos(2 s ) Cs i
Upr = Us
easily converted to 3D formulation, by using the knowledge of trans- sin(2 s )sin(2 p) + k 2cos2 (2 s ) Cp (16b)
formation of coordinate system (see Eq. (12) for details). Usi (usually
defined as 1), Usr and Upr are the amplitudes for the incident SV wave, Besides, s is the angle of SV incidence, which is also equivalent to
reflected SV wave and reflected P wave fronts, respectively, which are the angle of reflected SV-wave, and p is the angle of reflected P-wave
defined as, that can be obtained based on Snell’s law:

ux cos sin 0
T
ux cp
p = arcsin sin s
u = uy = Tu = sin cos 0 uy = 0 cs (17)
uz 0 0 1 uz (12)
Function f = f (t ) in Eq. (11) is any time-dependent function that
Usi = Asi ks, Usr = Asr ks, Upr = Apr kp determines the temporal variation of the incident wave.
(13)
where Asr and Apr have relationships with Asi as, 3.2. Verification
sin(2 s )sin(2 p) k 2cos2 (2 s)
Asr = Asi In the present study, the DRM method is implemented in ABAQUS
sin(2 s )sin(2 p) + k 2cos2 (2 s ) (14a)
[28], and the accuracy of this implementation is verified by comparing
2k 2sin(2 s )cos(2 s ) the numerical results obtained from DRM and analytical solutions (for
Apr = Asi homogeneous cases, vertical and inclined incident SV waves) and
sin(2 s )sin(2 p) + k2cos2 (2 s ) (14b)
single-soil-column simulations (for heterogeneous cases, vertically
and k = cp/cs , and cs and cp are the shear and compressional wave ve- propagated SV waves). In these verification problems, a single soil
locities. kp and ks in Eq. (13) are shear and compressional wave- column that has an identical height and material properties as the full
numbers, which are defined as, DRM model (see Fig. 13(b)) is used in 1D wave propagation analyses to
obtain free-field ground responses. Where b and Vb are the density and
ks = , kp = the shear wave velocity of the elastic bedrock, respectively;
cs cp (15)
ub (t ) = 2uI (0, t ) and uI is the incident wave function.
where is the natural frequency of the soil. By combining Eqs. (13)- For all the verification problems, we use Ricker pulse defined in Eq.
(15), we have (18), with the selected parameters as ARicker = 1e 4 m, fRicker = 5 Hz

11
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 18. Comparisons of horizontal and vertical displacement results for the case DRM-2D-Homo-Inc.

Fig. 19. Comparisons of horizontal displacement results for the case DRM-2D-Hetero-Ver (Ricker Pulse).

12
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 20. Comparisons of horizontal displacement results for the case DRM-2D-Hetero-Ver (earthquake motion).

and t 0 = 0.3 s , as the incident wave function. And a total step equals to 1 computing the impedance functions and investigating the effects of
s with t = 0.001 s is used. Corresponding displacement, velocity and angle of incidence.
acceleration time histories and Fourier amplitude are shown in Fig. 15.
Besides, A Northridge earthquake motion with duration equals to 28.6 s 4.1. Impedance functions
and t = 0.005 s, is adopted for the case DRM-2D-Hetero-Ver (see
Table 2 for definition) to verify the accuracy of the DRM when earth- It is well known that the substructure method is a computationally
quake motion is applied. Its displacement time histories and Fourier efficient alternative where a reduced-order model for the near-field is
amplitude are shown in Fig. 16. utilized [1,38,39]. However, the impedance function of the near-field
soil-foundation system is the ingredient of the substructure method.
fRicker )2,
u ( ) = ARicker [1 (2 fRicker ) 2] e ( =t t0 (18) The impedance function represents the complex-valued frequency-de-
pendent stiffness matrix, where its real part corresponds to the stiffness
In the numerical simulations, we verify DRM in a total of 5 different
and mass inertia effect of the soil and the imaginary part accounts for
cases. 3 for 2D and 2 for 3D domains, with (1) inclined incident SV
radiation damping.
waves in homogeneous media and (2) vertically propagated SV wave in
In this section, we follow the procedures mentioned by Seylabi et al.
heterogeneous media (one Ricker pulse and one earthquake motion).
[40] to extract the impedance functions with time-domain analysis in
For all 5 cases, uniform density = 2000 kg/m3 and Poisson’s ratio
ABAQUS for four different problems. And in the subsequent analyses,
= 0.3 are selected. And for homogeneous cases, the shear wave ve-
we always apply the Ricker pulse (see Eq. (18)) with parameters as
locity Vshomo = 200 m/s is used. For heterogeneous cases, we adopt a
z n ARicker = 1e 4 m, fRicker = 10 Hz to be the input displacements. A total
continuous function as Vshetero = VH b + (1 b) H , where
simulation time T = 0.5 s and stepsize t = 0.001 s are used. For 2D
b = (V0/VH )1/ n , V0 = 200 m/s, VH = 400 m/s, n = 0.5, H = 25 m, z is the problems, we use 4-node quadrilateral plane-strain element (CPE4)
downward vertical coordinate measured from the soil surface; and V0 with element size of 0.5 m× 0.5 m for the interior domain, and for 3D
and VH are shear wave velocities at z = 0 and z = H , respectively. All problems, we adopt 8-node tri-linear solid element (C3D8) with ele-
the necessary information for DRM verification problems is summarized ment size of 0.5 m× 0.5 m× 0.5 m . The PML elements always keep the
in Table 2 for a more transparent view. In the full DRM models (see same element size as the regular elements.
Fig. 17), with a uniform element size of 1 m, 98 4-node quadrilateral
plane strain elements (CPE4) and 7204 8-node linear isoparametric 4.1.1. 2D rigid strip surface foundation
elements (C3D8) are used for the single layer of elements for 2D and 3D The compliance function (i.e., inverse matrix of the impedance
models, respectively, and a thickness of 5 m PML are attached to the function) of the rigid strip foundation shown in Fig. 23(a), resting on
interior domains to absorb the noises coming from numerical impreci- the surface of an elastic homogeneous half-space, with material prop-
sion. erties as E = 1 GPa, = 0.25 and = 2000 kg/m3 , is computed. The
Figs. 18–22 displays the displacement results for the three selected reference analytical solution to this problem is provided by Luco and
points (see Fig. 17) obtained using DRM, analytical solutions and the Westmann [41]. As shown in Fig. 24, the numerically-evaluated com-
single soil column modeled as shown in Fig. 13(b). As seen, the DRM pliance functions show an excellent agreement with the reference so-
yields a near-perfect match compared to both analytical solutions and lutions for both diagonal terms (i.e., CHH , CVV , and CMM ) and the cou-
the single-soil-column results, for both homogeneous and hetero- pled term (i.e., CHM / CMH ). B denotes the half-width of the foundation.
geneous soil media, vertical and inclined incident SV waves, and in 2D
and 3D domains.
4.1.2. 2D embedded rigid foundation
The impedance function of a rigid foundation embedded in a
4. Applications homogenous soil shown in Fig. 23(b), with identical material properties
as the previous case, is computed. The reference solution to this pro-
After the successful implementation and verification of both DRM blem is due to Wang and Rajapakse [42], obtained by using the indirect
and PML in ABAQUS, in this section, we present some applications that boundary integral equation method. Fig. 25 shows the numerically-
can be conducted by using the PML or coupled DRM-PML system, e.g., computed impedance functions versus the reference solution, which

13
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 21. Comparisons of Ux , Uy and Uz for the case DRM-3D-Homo-Inc.

14
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 22. Comparisons of Ux for the case DRM-3D-Hetero-Ver.

Fig. 23. Configuration of (a) the 2D rigid strip surface foundation problem and (b) the 2D embedded rigid foundation problem.

Fig. 24. Compliance function of a rigid strip surface foundation computed using FEM-PML versus analytical solution.

15
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 25. Compliance function of an embedded rigid foundation computed using FEM-PML versus reference solution.

Fig. 26. Compliance function of a rigid circular plate surface foundation in homogeneous media computed using FEM-PML versus reference solution.

again are in good agreements. we also compute the impedance function of a 3D rigid circular plate on
the surface of a two-layered half-space shown in Fig. 27. The material
4.1.3. 3D rigid circular plate surface foundation in homogeneous media properties of two soil layers are included in Table 3. The analytical
The compliance function of the rigid circular plate foundation solution due to Luco [44] only provides the horizontal, vertical and
shown in Fig. 26, resting on the surface of an elastic homogeneous half- rocking terms (i.e., KHH , KVV and KMM ). For the torsional term (KTT ),
space, with material properties as E = 1 GPa, = 1/3 and we compare our result with a semi-analytical solution proposed by Lin
= 2000 kg/m3 , is computed. The reference analytical solution to this et al. [45]. And for the coupled term (KHM = KMH ), we just include our
problem is proposed by Luco and Westmann [43]. As shown in Fig. 26, result because no analytical/semi-analytical solution has been found in
by using the implemented 3D PML boundary, the numerically-calcu- the literature. In all cases, FEM-PML method could reproduce the re-
lated compliance functions match the analytical solutions very well, for ference solutions with high accuracy.
all horizontal, vertical, rocking and torsional results. Here B = 8 m re-
presents the radius of the rigid plate. 4.2. Effects of the angle of incidence

4.1.4. 3D rigid circular plate surface foundation in two-layered half-space As been reported in [46,47], the angle of incidence can significantly
media affect the response of the buried structures, especially for vertical ones.
To test the performance of the FEM-PML method in layered media, In this section, we analyze a 2D rectangular tunnel in homogeneous

16
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 27. Compliance function of a rigid circular plate surface foundation in two-layered half-space media computed using FEM-PML versus reference solution.

Table 3 condition is used for the soil-structure interface. Material properties for
Material properties of the two-layered soil deposits. the homogeneous soil and the rectangular structure are summarized in
Shear wave velocity Vs (m/s) Poisson’s Ratio Table 4. The effects of the distance between the DRM interface and the
Density (kg/m3 )
buried structure need to be further investigated [48].
Layer 1 400 1/3 1700 First, we compare the accelerations for two corner points on the
Layer 2 500 1/3 2000 structure (i.e., bottom left and top right). As we can see in Fig. 29, for
this specific case, by increasing the angle of incidence, both maximum
values of horizontal and vertical accelerations increases and the am-
media excited by SV waves with different angles of incidence. The
plification factors can reach 2.57 and 6.52, respectively. Then the
numerical model is shown in Fig. 28. Here we use the DRM interface to
racking displacements for different columns and its Fourier spectrums
prescribe the input motion, and the PML to truncate the domain and
are computed and shown in Fig. 30. As seen, the racking displacements
absorb the scattering waves. Eq. (18) with the selected parameters as
actually decrease when the angle of incidence increases, for all left,
ARicker = 1e 4 m, fRicker = 2 Hz and t 0 = 0.8 s , is used as the incident
middle and right columns.
wave function. A total simulation time is T = 3 s and time stepsize is
Finally, we explore the effects of the angles of incidence on the axial
t = 0.005 Section 4-node quadrilateral plane strain elements (CPE4)
force, bending moment, as well as the deformation mode. Fig. 31(a) and
with element size of 1 m× 1 m are used for the soil in the interior
(b) show the profiles of the maximum axial force and bending moment,
domain. 2-node beam elements (B21) with element size of 0.5 m and
respectively. As seen, with the increasing angles of incidence, the axial
thickness 1 m are used for the rectangular structure. No-slip interface

Fig. 28. Plan view of the buried rectangular tunnel for studying the effects of angle of incidence.

17
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Table 4 force increases, while the bending moment decreases. And Fig. 32
Material properties of homogeneous soil and the rectangular structure. shows the deformation plot when the maximum bending moment
Young’s modulus (MPa) Poisson’s Ratio happens, for different angles of incidence. More bending deformation is
Density (kg/m3 )
observed for small angles, and more rotation is noticed for larger an-
Soil 208 0.3 2000 gles. It should be noted that in all figures the resulting deformations are
Structure 32000 0.2 2500 magnified 30000 times.

Fig. 29. Time histories and the Fourier amplitudes of the horizontal and vertical accelerations.

18
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

Fig. 30. Time histories and the Fourier amplitudes of the racking displacements.

Fig. 31. Profiles of the maximum (a) axial force and (b) bending moment.

Fig. 32. Maximum deformation plot for different angles of incidence.

19
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

5. Conclusions Finally, we presented practical application examples on the computa-

tion of impedance functions, and investigations of the angle of in-
In this paper, we presented procedures for implementation of DRM cidence of seismic waves on response of buried structures. The pre-
and PML in ABAQUS for 2D and 3D problems, including a new version sented coupled DRM-PML technique is a key analysis tool for soil-
of DRM for inclined incident SV waves. We then carried out a com- structure interaction problems; and the presented ABAQUS im-
prehensive set of verification studies through problems involving ver- plementation, will be disseminated for broader use, should enable re-
tical and inclined incident SV waves, 2D and 3D domains, homogeneous searchers and practicing engineers to carry out state-of-the-art non-
and heterogeneous soil layers, and concentrated and distributed loads. linear seismic analyses of soil-structure systems in truncated domains.

Appendix A

For 2D domain, the semi-discrete form is:

MPML u
¨ + CPML u + KPML u = Fext (19)
where the element matrices MPML , CPML and KPML are defined as

Ma 0 Mb Ae Mc Ap
MPML = , CPML = , KPML =
0 Na Ae Nb Ap Nc (20a)

u = [ux uy T
xx yy xy ] , Fext = [Fx Fy 000]T (20b)
The submatrices in Eq. (20a) are defined as,

Mi = i diag( T, T)d , i = a , b, c
e (21a)

+ 2µ T T
4µ ( + µ) 4µ ( + µ)
T + 2µ T
Ni = e i 4µ ( + µ) 4µ ( + µ)
d , i = a , b, c
1 T
µ (21b)
i i
T T
,x y ,y x e p
Ai = e i i
d , i = e, p , j = j, j = j, j = x, y
T T
,y x ,x y (21c)
where a = x y, b = x y + y x , c = x y . and are the shape functions for the displacement and stress components. And here we use identical
4-node bilinear shape functions for and in 2D domain.
For 3D domain, the semi-discrete form is:
MPML u
¨ + CPML u + KPML u + GPML u¯ = Fext (22a)
t
u¯ = u( )d (22b)
0

where the element matrices MPML , CPML , KPML and GPML are defined as

Ma 0 Mb Ae
MPML = , CPML = ,
0 Na Ae Nb
Mc Ap Md Aw
KPML = , GPML =
Ap Nc Aw Nd (23a)

u = [ux uy u z xx yy zz xy xz yz ]T , Fext = [Fx Fy Fy 000000]T (23b)

The submatrices in Eq. (23a) are defined as,

Mi = i diag( T, T, T)d , i = a , b, c, d
e (24a)

+µ T T T
µ (3 + 2µ) 2µ (3 + 2µ) 2µ (3 + 2µ)
T +µ T T
2µ (3 + 2µ) µ (3 + 2µ) 2µ (3 + 2µ)
T T +µ T
2µ (3 + 2µ) 2µ (3 + 2µ) µ (3 + 2µ)
Ni = e i d ,
1 T
µ
1 T
µ
1 T
µ

i = a , b, c, d (24b)

20
W. Zhang, et al. Computers and Geotechnics 114 (2019) 103143

i i i
T T T
,x yz ,y xz ,z xy
i i i
Ai = e ,y
T
xz ,x
T
yz ,z
T
xy
d ,
i i i
T T T
,z xy ,x yz ,y xz
e p w
i = e , p, w, jk = j k, jk = j k + k j, jk = j k, j, k = x , y, z (24c)
where coefficients a, b, c, d are defined as
a= x y z (25a)

b= x y z + x y z + x y z (25b)

c= x y z + x y z + x y z (25c)

d= x y z (25d)

References [25] Basu U. Explicit finite element perfectly matched layer for transient three-dimen-
sional elastic waves. Int J Numer Meth Eng 2009;77(2):151–76.
[26] Livermore Software Technology Corporation. LS-DYNA Keyword User’s Manual;
[1] Wolf JP. Dynamic soil-structure interaction. Englewood Cliffs, N.J. 07632: Prentice- 2017.
Hall; 1985. [27] Poul MK, Zerva A. Time-domain pml formulation for modeling viscoelastic waves
[2] Mylonakis G, Gazetas G. Seismic soil-structure interaction: beneficial or detri- with rayleigh-type damping in an unbounded domain: Theory and application in
mental? J Earthquake Eng 2000;4(3):277–301. abaqus. Finite Elem Anal Des 2018;152:1–16.
[3] Bielak J, Christiano P. On the effective seismic input for non-linear soil-structure [28] Hibbit, Karlsson, Sorensen. ABAQUS/Standard Analysis User’s Manual. USA:
interaction systems. Earthquake Eng Struct Dynam 1984;12:107–19. Hibbitt, Karlsson, Sorensen Inc.; 2007.
[4] Lysmer J, Kuhlemeyer RL. Finite dynamic model for infinite media. J Eng Mech Div [29] Zheng Y, Huang X. Anisotropic perfectly matched layers for elastic waves in car-
1969;95(EM4):859–76. tesian and curvilinear coordinates. Tech. rep. Massachusetts Institute of
[5] Esmaeilzadeh Seylabi E. Reduced order modeling of soil structure interaction pro- Technology. Earth Resources Laboratory; 2002.
blems [Ph.D. thesis]. UCLA; 2016. [30] Hilber HM, Hughes TJ, Taylor RL. Improved numerical dissipation for time in-
[6] Marburg S, Nolte B. Computational acoustics of noise propagation in fluids: finite tegration algorithms in structural dynamics. Earthquake Eng Struct Dyn
and boundary element methods vol. 578. Springer; 2008. 1977;5(3):283–92.
[7] Higdon RL. Radiation boundary conditions for the scalar wave equation. SIAM J [31] Jeremić B, Kunnath S, Xiong F. Influence of soil–foundation–structure interaction
Numer Anal 1990;27(4):831–70. on seismic response of the i–880 viaduct. Eng Struct 2004;26(3):391–402.
[8] Givoli D. High-order local non-reflecting boundary conditions: a review. Wave [32] Zhang Y, Conte JP, Yang Z, Elgamal A, Bielak J, Acero G. Two-dimensional non-
Motion 2004;39:319–26. linear earthquake response analysis of a bridge-foundation-ground system.
[9] Joly P, Tsogka C. Higher order absorbing boundary conditions for elastodynamics. Earthquake Spectra 2008;24(2):343–86.
Proc. of the 4th European conference on computational mechanics (ECCM-2010), [33] Bielak J, Loukakis K, Hisada Y, Yoshimura C. Domain reduction method for three-
Paris. 2010. dimensional earthquake modeling in localized regions, part i: theory. Bull Seismol
[10] Rabinovich D, Givoli D, Bielak J, Hagstrom T. A finite element scheme with a high Soc Am 2003;93(2):817–24.
order absorbing boundary condition for elastodynamics. Comput Methods Appl [34] Fathi A. Full-waveform inversion in three-dimensional pml-truncated elastic media:
Mech Eng 2011;200(23–24):2048–66. theory, computations, and field experiments [Ph.D. thesis]. Austin: University of
[11] Baffet D, Bielak J, Givoli D, Hagstrom T, Rabinovich D. Long-time stable high-order Texas; 2015.
absorbing boundary conditions for elastodynamics. Comput Methods Appl Mech [35] Bécache E, Petropoulos PG, Gedney SD. On the long-time behavior of unsplit per-
Eng 2012;241(244):20–37. fectly matched layers. IEEE Trans Antennas Propag 2004;52(5):1335–42.
[12] Claerbout JF. Coarse grid calculations of waves in inhomogeneous media with [36] MATLAB, version R2018a. Natick, Massachusetts: The MathWorks Inc.; 2018.
application to delineation of complicated seismic structure. Geophysics [37] Poursartip B. Topographic amplification of seismic motion [Ph.D. thesis]. Austin:
1970;35(3):407–18. University of Texas; 2017.
[13] Claerbout JF, Johnson AG. Extrapolation of time-dependent waveforms along their [38] Wolf JP. Soil-structure-interaction analysis in time domain. Nucl Eng Des
path of propagation. Geophys J Int 1971;26(1–4):285–93. 1989;111(3):381–93.
[14] Clayton R, Engquist B. Absorbing boundary conditions for acoustic and elastic wave [39] Mylonakis G, Nikolaou A, Gazetas G. Soil–pile–bridge seismic interaction: kine-
equations. Bull Seismol Soc Am 1977;67(6):1529–40. matic and inertial effects. Part i: soft soil. Earthquake Eng Struct Dyn
[15] Akiyoshi T, Fuchida K, Fang H. Absorbing boundary conditions for dynamic ana- 1997;26(3):337–59.
lysis of fluid-saturated porous media. Soil Dyn Earthquake Eng 1994;13(6):387–97. [40] Seylabi EE, Jeong C, Taciroglu E. On numerical computation of impedance func-
[16] Kim HS. Finite element analysis with paraxial & viscous boundary conditions for tions for rigid soil-structure interfaces embedded in heterogeneous half-spaces.
elastic wave propagation. Engineering 2012;4(12):843. Comput Geotech 2016;72:15–27.
[17] Berenger JP. A perfectly matched layer for the absorption of electromagnetic waves. [41] Luco J, Westmann R. Dynamic response of a rigid footing bonded to an elastic half
J Comput Phys 1994;114:185–200. space. J Appl Mech 1972;39(2):527–34.
[18] Basu U, Chopra AK. Perfectly matched layers for time-harmonic elastodynamics of [42] Wang Y, Rajapakse R. Dynamics of rigid strip foundations embedded in orthotropic
unbounded domains: theory and finite-element implementation. Comput Methods elastic soils. Earthquake Eng Struct Dyn 1991;20(10):927–47.
Appl Mech Eng 2003;192(11–12):1337–75. [43] Luco JE, Westmann RA. Dynamic response of circular footings. J Eng Mech
[19] Basu U, Chopra AK. Perfectly matched layers for transient elastodynamics of un- 1971;97(5):1381–95.
bounded domains. Int J Numer Meth Eng 2004;59(8):1039–74. [44] Luco JE. Impedance functions for a rigid foundation on a layered medium. Nucl Eng
[20] Kucukcoban S, Kallivokas LF. Mixed pefectly-matched-layers for direct transient Des 1974;31(2):204–17.
analysis in 2d elastic heterogeneous media. Comput Methods Appl Mech Eng [45] Lin G, Zejun H, Jianbo L. An efficient approach for dynamic impedance of surface
2011;200:57–76. footing on layered half-space. Soil Dyn Earthquake Eng 2013;49:39–51.
[21] Kucukcoban S, Kallivokas L. A symmetric hybrid formulation for transient wave [46] Poursartip B, Fathi A, Kallivokas LF. Seismic wave amplification by topographic
simulations in pml-truncated heterogeneous media. Wave Motion features: a parametric study. Soil Dyn Earthquake Eng 2017;92:503–27.
2013;50(1):57–79. [47] Zhang W, Seylabi EE, Taciroglu E. Effects of soil stratigraphy on dynamic soil-
[22] Fathi A, Poursartip B, Kallivokas LF. Time-domain hybrid formulations for wave structure interaction behavior of large underground structures. Proc. of the 3rd int.
simulations in three-dimensional pml-truncated heterogeneous media. Int J Numer conf. on performance-based design in earthquake geotechnical engineering (PBD-
Meth Eng 2015;101(3):165–98. III), Vancouver, Canada. 2017.
[23] Brinkgreve RBJ, Swolfs WM, Engin E. Plaxis introductory: student pack and tutorial [48] Luo C, Lou M, Gui G, Wang H. A modified domain reduction method for numerical
manual 2010. Boca Raton, FL, USA: CRC Press, Inc; 2011. simulation of wave propagation in localized regions. Earthquake Eng Eng Vibr
[24] F. Itasca, Fast lagrangian analysis of continua, ver. 8.0, Itasca Consulting Group 2019;18(1):35–52.
Inc., Minneapolis, Minn.

21