Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering

journal homepage: http://www.elsevier.com/locate/soildyn

Method to evaluate the dynamic structure-soil-structure interaction of 3-D

buildings arrangement due to seismic excitation
Felipe Vicencio a, *, Nicholas A. Alexander b
a
Department of Civil Engineering, University of Bristol, Queen’s Building, Bristol, UK
b
Structural Dynamics, Department of Civil Engineering, University of Bristol, Queen’s Building, Bristol, UK

A R T I C L E I N F O A B S T R A C T

Keywords: This paper explores the influence of Structure-Soil-Structure Interaction (SSSI) between multiple buildings in a
Structure-soil-structure interaction three-dimensional arrangement under seismic excitation. A numerically simplified reduced-order model is pro­
Time history seismic analysis posed to significantly shorten computational run-times for the case of very large clusters of buildings. This
Reduced-order model
formulation has been developed to include the presence of any number of buildings with different heights, widths
and interbuilding distances. The auto-rotational spring stiffnesses and inter-rotational spring stiffnesses, of
foundations, used in this model are determined by an application of an empirical surficial displacement field
(loosely based on a form derived from a Boussinesq elastic half-space approximation that is calibrated against
FEA) and inverse system identification using either least-squares or an application of Kronecker products.
Different spatial building arrangements are presented, in order to compare with previous research and highlight
the versatility of the method. The results showed that the 3D multi-building SSSI can be significantly more
complex and difficult to predict than the previous 2D cases in the literature.

1. Introduction
Population growth and enhanced land costs have produced an in­
crease in city building surface density, i.e. more closely spaced build­
ings. While it is common practice to determine the seismic response of
buildings as singleton structures, the high surface density inevitably
results in the possibility of interaction and coupling effects via the un­
derlying soil. This phenomenon is referred to as Structure-Soil-Structure
Interaction (SSSI) and can either magnify or attenuate the seismic
response of a building. Some of the first work that highlighted the sig­
nificance of SSSI was Lee and Wesley [1], Luco and Contesse [2], Wong
and Trifunac [3], and Triantafyllidis and Prange [4] among others. The
results published in these papers showed that the coupled effects modify
the structural responses, and these coupled dynamic responses differ by
up to 30% when compared with the singleton case. Subsequently,
several authors have studied the SSSI problem through different meth­
odologies and techniques. Some made use of the Boundary element
method (BEM) [5–9], two or three-dimensional finite element modelling
(FEM) [10,11], or a combination of these two FEM/BEM procedures
[12–15]. These studies have identified the key factors that may control
the level of multi-structural interaction such as, (i) the interbuilding
distance, (ii) the relative dynamic characteristics of the adjacent build­
ings and (iii) the soil type. These numerical models have allowed rep­
resenting complex geometries and nonlinearities in the elements, and
have served as an essential validation data for simplified reduced-order
models. Nevertheless, these approaches have the disadvantages of
having a very large number of degrees of freedom involved, and a
considerable amount of system parameters, which leads to almost pro­
hibitive computational cost when modelling a city-site response.
While numerical SSSI models are vital, important validation/cali­
bration data have been sought from physical experimental tests. These
have been implemented in the last years for the study on SSSI effects.
Trombetta et al. [16] and Mason et al. [17], have investigated the in­
elastic structural response of two adjacent steel moment-resisting frames
in geotechnical centrifuge tests subjected to ground shaking. The results
show that SSSI can be beneficial or detrimental, depending on the
earthquake motion and the structural system. Knapett et al. [18] studied
similar and highly dissimilar buildings on shallow foundations using
centrifuge modelling, where permanent rotation (structural tilt) was
observed to increase significantly. Aldaikh et al. [19,20] have indicated
that the presence of one or two adjacent buildings could produce
favourable/detrimental changes in seismic response power when

* Corresponding author.
E-mail address: favicenc@gmail.com (F. Vicencio).

https://doi.org/10.1016/j.soildyn.2020.106494
Received 25 February 2020; Received in revised form 22 October 2020; Accepted 31 October 2020
0267-7261/© 2020 Elsevier Ltd. All rights reserved.

Please cite this article as: Felipe Vicencio, Nicholas A. Alexander, Soil Dynamics and Earthquake Engineering,
https://doi.org/10.1016/j.soildyn.2020.106494
F. Vicencio and N.A. Alexander Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

compared to a single building case. Jabary et al. [21] analysed the SSSI approximation that was calibrated against 3D FEA). Also, frequency-
effects on structures with a tuned mass damper. Li el al [22]. described independent lumped masses, springs and dampers are assumed, i.e.
the numerical analyses of a shaking table test on two identical 12-story frequency-dependant stiffness and frequency-dependant radiation
concrete buildings. These studies are an important validation point for damping is not modelled. All code used in this study was developed in
numerical models and provide interesting results in complex problems Matlab.
like SSSI. Nevertheless, they represent a small statistical sample and a Finally, it is worth stating that our reduced-order model has only 4n
limited parametric exploration of the problem. Note that these experi­ dofs (where n is the number of buildings) rather than perhaps hundreds
mental studies carried out in laminar boxes do not include all the ki­ of millions of dofs for a full 3D FE analysis of simple buildings and their
nematic effects due to these studies solely consider vertically underlying elastic half-space and hence does not required any full 3D FE
propagating shear waves that do not properly scatter on surface foun­ analyses.
dations. Studies of recorded responses of instrumented structures and
felt reports (like a macroseismic intensity survey) had been used for 2. A theoretical reduced-order model for SSSI between multiple
evidence of SSSI after an earthquake [23–26]. buildings in 3D
Lumped parameter models (or reduced-order models) with a limited
number of DOF have been well recognized and effectively applied in the 2.1. General preamble and limitations
evaluation of the SSSI effects on the dynamic response of buildings
[27–34]. In these mechanical models, it was considered that all masses, The reduced-order model derived in this paper consists of a set of n
springs and dashpots are lumped into a single mass, single spring and buildings belonging to a city block, distributed over the ground surface,
single damping constant for each mode of vibration. Also, as shown in Fig. 1. We do not attempt to model all interactions, only the
frequency-independent lumped masses, springs and dampers are significant rotational ones as in Ref. [30–34]. This system can be viewed
assumed. The coupled effects were incorporated into the solution by as an assemblage of individual buildings coupled through the common
means of empirical stiffness coefficients, being calibrated and validated homogeneous linear elastic soil medium. Thus, each building’s foun­
using FEM [30] and physical experimental tests [18–20]. Later, Aldaikh dation has two orthogonal auto-rotational springs kxi and kyi and a large
et al. [34] proposed an alternative closed-form analytical expression for number of inter-rotational springs κij (i.e. the coupling between build­
the interaction springs coefficient based on a Boussinesq approximation ings “i” and “j”). The planar mesh of rotational springs kxi , kyi and κij can
of the surficial displacement fields. The previous works highlight that be viewed as a complete replacement (master-slave condensation, sub­
dynamic loading, like seismic ground motion, can affect the response of structuring) of the ground [34] where only rotational dofs at the foun­
the structures, and this influence can either magnify or attenuate the dation θxi , θyj are retained. Note that due to the 3-dimensional nature of
structural response of a building. the problem κij should be viewed as a small matrix with inter-rotational
A related problem that has been addressed recently is site-city stiffness coefficients in the x -direction κxij , in the y -direction κyij , and the
interaction (SCI) or city effects, where the seismic free-field ground
cross-coupled term κxyij . The kinematic interaction [41–43], resulting in
motion is affected by building clusters in dense urban areas. Several
the differences between the stiff foundation and the ground motion, is
studies have been undertaken in the last decades, like Gueguen et al.
not considered in the present work, i.e. the transfer function between
[35], Boutin and Roussillon [36], Kham et al. [37], Ghergu and Ionescu
foundation input motion and free-field motion is one.
[38], Isbiliroglu et al. [39], Schwan et al. [40] just to cite a few. These
Each building’s superstructure has 2 dofs, xi and yi , and is assumed
works have shown that (i) the SCI effect is influenced by the separa­
regular (i.e. only plan-symmetric buildings are considered, unlike [33]).
tion/size of the structures and the relative stiffness to the soil, (ii) a part
The sway-flexural (modal) stiffness in both x and y directions of building
of the seismic energy transmitted to the buildings is redistributed in
“i” is defined by kbi . The lumped mass (the effective fundamental modal
their neighbourhood through multiple interaction between the soil
mass) of the building “i” in both x and y directions is mbi , and the
foundation and the buildings, and (iii) the ground motion is affected by
foundation/soil mass underneath the building “i” is msi . The soil/foun­
the presence of a large group of structures, producing a spatial variation
dation system for building “i” has two rotational DOFs at the foundation
of the ground acceleration.
levels θxi and θyi . The term “modal” used here refers to the first funda­
The objective of this work is to introduce a novel framework for
mental mode of this characteristic multiple degrees of freedom building.
producing an efficient reduced-order mathematical formulation. So that
Thus, the total system only contains 4n dofs where n is the number of
we can evaluate the multiple SSSI among a group of buildings of a city-
buildings.
block under seismic excitation in 3D. Additionally, this reduced-order
As an initial problem exploration, we further assume that all build
model is not reliant on the creation of any initial full 3D FEA analysis
of the soil. Thus, we are proposing a new methodology that avoids the
necessity of using any FEA of the soil. It relies solely on defining auto-
rotational and inter-rotational spring coefficients for the elastic half-
space. The only input information required to define our reduced-
order model is (i) the soil class, and (ii) the height, footprint di­
mensions and the planar coordinates of the buildings. Our approach
defines the problem in terms of a small number of system parameters
and degrees of freedom.
Therefore, it is computationally simple/efficient enough for an
extensive exploration of the problem of multiple buildings interacting.
The rocking spring stiffnesses of the soil-foundation (auto-rotational
spring stiffnesses) and the coupling spring stiffnesses between the
buildings (inter-rotational spring stiffnesses) are determined by using a
least-squares inverse system identification. A number of load cases of
static applied moments at foundations are employed, along with their
implied surficial displacement fields, to determine all the individual
stiffness coefficients of the system matrix. The implied surficial
displacement field is obtained by using an empirical formulation
(loosely based on a form derived from a Boussinesq elastic half-space Fig. 1. Idealization of the multi-buildings SSSI model in a 3D arrangement.

2
F. Vicencio and N.A. Alexander Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

ings are simultaneously excited only in the x direction. A known ground

∑ n ( )
displacement field xg is applied at all foundations, i.e. wave passage
n
( ) 1∑ 1∑n− 1
U= − μxi θxi + μyi θyi + kxi θ2xi + kyi θ2yi + ⋯ +
effects, coherence effects and spatially heterogeneous ground displace­ i=1
2 i=1 2 i=1
ments are neglected in the presented work. Nevertheless, this unidi­ ∑
n
( ( )2 ( )2 ( )2 ( )2 )
rectional spatially coherent ground motion produces vibration in the × κxij θxj − θxi + κyij θyj − θyi + κxyij θyj − θxi + κyxij θxj − θyi
two horizontal directions due to the unsymmetrical arrangement of
j=i+1

foundations. We assume that each building’s response could be affected (1)

by any building of the system, and the magnitude of these interactions
mainly depend on the inter-building distance [30–34]. This means that
we take into account multiple structure-soil-structure interactions
through the soil. Note that building pounding is not permitted as the
inter-building spacing is assumed large enough to avoid pounding.
2.2. How are the stiffness coefficients of all ground rotational springs
obtained?
where n is the number of buildings, kxi and kyi are the rotational spring
stiffnesses of each foundation and bi are the width of the building’s
foundation. In the general case for rectangle foundations, the last cross-
= κxyij , and must be
couped terms of the equation (1) are different κxyij ∕
derived. The Euler-Lagrange formulation provides a set of 2n linear
equations that relate the external moments and their implied rotations
(associated with the surface rotation field). These equations written in
matrix form are formulated in Equation (2).

In order to determine the inter-rotational springs κij and the auto-

⎡ ⎤ ⎡ ⎤⎡ ⎤
μx1 kx1 + κx1i + ⋯ + κxy1n ⋯ − κx1i ⋯ − κxy1n θx1
⎢ ⋮ ⎥ ⎢ ⋮ ⋱ ⋮ ⋱ ⋮ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥⎢ ⋮ ⎥
⎢ μxi ⎥ = ⎢
⎢ ⎥ ⎢ − κxi1 ⋯ kxi + κxi1 + ⋯ + κxyin ⋯ − κxyin ⎥ ⎢ θxi ⎥μ = KΘ
⎥⎢ ⎥ (2)
⎣ ⋮ ⎦ ⎣ ⋮ ⋱ ⋮ ⋱ ⋮ ⎦⎣ ⋮ ⎦
μyn − κxyn1 ⋯ − κxyni ⋯ kyn + κyn1 + ⋯ + κxyni θyn

rotational spring stiffnesses of the soil beneath each buildings kxi and
kyi , we extend the procedure described in Ref. [34] to the
three-dimensional case. This technique considers a reduced-order model
of the soil and the foundations, and is calibrated using FEM.
The geometry and nomenclature for the 3D foundation-soil-
foundation system studied in this paper are shown in Fig. 2. Each of
these foundations are coupled with the other foundations by the inter-
rotational springs κij between buildings “i” and “j”. The term κxij is the
component in the x-direction, κyij in the y-direction, and κxyij is the cross-
coupled part. All the foundations are assumed square, i.e. kxi = kyi , but
buildings can have different plan-size foundations i.e. kxi is not neces­
sarily equal to kxj .
The potential energy of the system is given by equation (1), and is
calculated by the sum of (i) the external work of applied static moments
μxi and μyi (summation term 1 in equation (1)), (ii) the internal work due
to auto-rotational springs (summation term 2 in equation (1)), and (iii)
the internal work due to inter-rotational springs κxij , κyij and the cross-
coupled term κxyij , where κxij = κxji , κyij = κyji , and κxyij = κyxij (double
summation term 3 in equation (1)). Hence,
Therefore, to determine the stiffness coefficients, i.e. the entire ma­
trix K of equation (2), we make use of the relationship between applied
surface moments and surficial rotations as in Ref. [34]. In this paper, two
approaches are used to develop an empirical applied moment-surficial
rotation relationship (i) a formulation based entirely on Boussinesq’s
approximation of elastic half-space behaviour and (ii) a 3D FEA cali­
brated empirical relationship loosely based on a Boussinesq form. Then,
we compare these results with previous 2D works [30,34].
For small deformations, i.e. a linear elastic soil half-space, the surface
displacement field Uz (x, y) is specified in Equation (3), where θy0 cor­
responds to the rotation of the rigid foundation about the y axis, and
Δ(x, y) is the decay function. The shape of the decay function is depicted
in Fig. 3, and is defined for any (x, y) arbitrary coordinate, located in the
free surface plane. This function is valid only for points outside the
foundation, i.e. (|x| ≥ b /2) ∧ (|y| ≥ b /2) where b is the foundation
width. The equation (3) can be nondimensionalized by the definition of

Fig. 2. Spatial distribution of multiple foundations and their interaction in 3D. Fig. 3. Decay function for the surface displacement field in 3D.

3
F. Vicencio and N.A. Alexander Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

the length x = ξx b and y = ξy b, where ξx and ξy are the nondimensional where the term Uz (x, y) = uz (ξx , ξy )b of the equation (9) becomes,
coordinates, and nondimensional surface vertical displacement uz (ξx ,ξy ) ⎧ ⎫
b = Uz (x, y). ⎪
⎪ ⎪
⎪
⎨ ⎬
b ( ) 1 ( ) ( ) 1 1 1 1 ( )
Uz (x, y) = Δ(x, y)θy0 ∴ uz ξx , ξy = Δ ξx , ξy θy0 (3) uz ξx , ξy = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
4π ⎪ ( )2 ( )2 ̅ θy0 = Δ ξx , ξy θy0
̅ − √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
2 2 ⎪ ⎪ 2
⎩ ξx − 12 + ξ2y ξx + 12 + ξ2y ⎪ ⎭
By differentiating the previous equation (3), the surface rotation
field in x and y directions can be calculated as,
(10)
( ) ( ) ( ) ( )
( ) ∂uz ξx , ξy
θ y ξx , ξy =
1 ∂ Δ ξx , ξy
= θy0 , |ξx | ≥
1 ⃒ ⃒ 1
∧ ⃒ξy ⃒ ≥ (4) Therefore, the decay function can be defined as,
∂ξx 2 ∂ξx 2 2 ⎧ ⎫
( ) ( ) ( ) ( ) ⎪
⎪ ⎪
⎪
( ) ∂uz ξx , ξy 1 ∂Δ ξx , ξy 1 ⃒ ⃒ 1 ⎨ ⎬ ( )
θx ξx , ξy = = θy0 , |ξx | ≥ ∧ ⃒ξy ⃒ ≥ (5) ( ) 1 1 1 1
∂ξy 2 ∂ξy 2 2 Δ ξx , ξy = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
̅ − √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
̅ , |ξx | ≫
2π ⎪ ( )2 ( )2
⎪ 2
⎪
⎩ ξx − 12 + ξ2y ξx + 12 + ξ2y ⎪ ⎭
2.3. Method 1: Applied moment-surficial rotation, using only (
⃒ ⃒ 1
)
Boussinesq’s approximation ∧ ⃒ξy ⃒ ≫
2
The Boussinesq surficial displacement field ρz due to presence of a (11)
vertical point load P applied at the origin of an elastic (linear and ho­ The surficial slopes are given by partial differentiation of equation
mogeneous) half-space is defined as, (11) as follows.
( ) ⎧ ⎫
P 1 − ν2s 1 ⎪ ⎪
ρz (x, y) = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅ (6) ⎪ ⎪
π Es x2 + y2 ( ) ⎨ ⎬
( ) ∂Δ ξx ,ξy 1 2ξx − 1 2ξx + 1
θy ξx ,ξy = = − )32 + ((
where νs is the Poisson’s ratio, Es is the elastic modulus of the soil, and ∂ξx 4π ⎪ (( )2 )2 )32
⎪
/ /

⎪
⎩ ξx − 12 + ξ2y ξx + 12 + ξ2y ⎪
⎭
(x, y) represent the Cartesian coordinates between the load and any
point in the half-space surface. Now, if we have a rigid square footing of
(12)
length b, we can use the equation (6) by applying a pair of equal and
opposite point loads P along the axis x at x = b/2 and x = − b/ 2. These ⎧ ⎫
loads produce a moment about the axis y equal to μy0 = Pb. An ⎪
⎪ ⎪
⎪
( ) ⎨ ⎬
approximation of the surface vertical displacement function Uz (x, y) in ( ) ∂Δ ξx ,ξy 1 ξy ξy
θx ξx ,ξy = = − )32 + ((
3D, due to the moment μy0 can be derived through the sum of the two ∂ξy 2π ⎪ (( )2 / )2 )32
⎪
/

⎪
⎩ + ξ2y ξx + 12 + ξ2y ⎪ ⎭
vertical loads as follows,
1
ξx − 2

(13)

⎧ ⎫
⎪
⎪ ⎪
⎪
( ) ( ) ( 2
)⎨ ⎬
b b μy0 1 − νs 1 1
Uz (x, y) = ρz x − , y − ρz x + , y = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2 ̅ − √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2 ̅ (7)
2 2 π Es b ⎪ ⎪
⎪
⎩ x− 2 +y b 2 x+2 +y ⎪b 2 ⎭

The relationship between the applied moment μy0 and the rotation of The limitations of this formulation, equation (11) are: (a) the
the foundation θy0 can be defined by using the empirical formulae displacement field for the position ξx = 12,ξy = 0 gives a singularity
deduced by Pais and Kausel [44]. Note that the frequency-dependent Δ(1 /2,0) = ∞, (b) for small inter-building distances, where the SSSI is
lumped springs are ignored, so the frequency-independent springs are important, the vertical deformation is less accurate, and (c) it does not
obtained by Pais and Kausel [44]. completely include the constraining effects of the rotation of the rigid
footing itself, as discussed in Refs. [34].
1 Gs b3 1 Es b3 Es
μy0 = θy0 = ( )θy0 , Gs = (8)
2 (1 − νs ) 4 1 − ν2s 2(1 + νs )
2.4. Method 2: Applied moment-surficial rotation, using a 3D FEA
where Gs is the shear modulus of the half-space. Then, the surface ver­ calibrated empirical fit
tical displacement can be written as,
⎧ ⎫ The weakness of Method 1 (Boussinesq approach) is that it predicts
⎪
⎪ ⎪
⎪ accurate vertical deformations only far away from the foundation.
2
⎨ ⎬ Therefore, a finite-element solution that provides a more accurate
b 1 1
Uz (x, y) = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ̅ θy0
̅ − √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ (9) response is required. The following high-order FE model, shown in
4π ⎪ ( )2 ( )2
⎪
⎪
⎩ x − b2 + y2 x + b2 + y2 ⎪ ⎭ Fig. 4, using the Matlab CALFEM toolbox [45] is performed. The soil was
modelled considering a fine mesh of 8-node isoparametric solid linear
elastic elements and a 3-D formulation. Note that Fig. 4 for a
We defining the nondimensional coordinates as x = ξx b and y = ξy b, single-footing case employed millions of dofs. The foundation is

4
F. Vicencio and N.A. Alexander Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

Fig. 4. Evaluation of surface deformation field in 3D due to the rotation of a rigid foundation using a finite element model (Calfem [45]).

modelled as a perfectly rigid plate element. ( )

( ) ∂Δ ξx , ξy
Note that in the formulation presented here we do not model uplift in θ y ξx , ξy = =− (
p1 (ξx − p2 )
( )2 )32
the foundation, i.e. the soil appears to be pulled up on one half of the ∂ ξx
/
(ξx − p2 )2 + p3 ξy
foundation. However, remember that as this system is assumed linear, p1 (ξx + p2 )
therefore this applied moment displacement field (load case) must be +( ( )2 )32
/ (15)
added to the vertical gravity displacement field (load case). It is in the (ξx + p2 )2 + p3 ξy
total load case scenario (lateral moment and gravity loads) that we as­ ( )
sume uplift is neglected, which is quite often the case. ( ) ∂ Δ ξx , ξy p1 ξy p23
θ x ξx , ξy = =− ( ( )2 )32
The empirical form of the decay function Δ(ξx , ξy ) that was loosely ∂ξy
/
(ξx − p2 )2 + p3 ξy
based on the form in method 1 was defined as follows, p1 ξy p23
( ) +( ( )2 )32 (16)
p1 p1 /

Δ ξx , ξy = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2 − √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2
(ξx − p2 )2 + p3 ξy (ξx + p2 )2 + p3 ξy
The optimal p1 , p2 , and p3 coefficients that resulted from least-
squares fitting of equation (14) to the full 3D FEA simulation shown in
Fig. 4, where as follows, p1 = 0.3555, p2 = 0.2453, and p3 = 0.8049.
The goodness of fit was very high with a coefficient of determination of
R2 = 0.956. It is worth noting that equation (11) is identical to equation
(14) when p1 = 0.159, p2 = 0.5, and p3 = 1.
This equation (14) represents an inverse power relationship between
the decay function Δ(ξx , ξy ) and the distances ξx and ξy . By differenti­
ating the equation (14) we can estimate the surface rotation function in
both directions as follows,
(14) (ξx + p2 )2 + p3 ξy
Fig. 5(a) displays the comparison for the surface decay function
Δ(ξx , ξy ) between Method 1, equation (11), and Method 2, equation (14).
The form of both surfaces are very similar. Method 2 is more accurate at
small interbuilding distances. Fig. 5(b) displays the comparison for the
surface rotations, derivative of the function Δx (ξx ,ξy ), between Method 1
(equation (12)) and Method 2 (equation (15)).
Method 2, the FEA calibrated empirical model loosely based on
Boussinesq’s approximation removes the errors at small inter-building
distances of Method 1 so is proposed as favourable and is used in all
subsequent analyses.

Fig. 5. Comparison between FEA empirical function and Boussinesq results for (a) surface decay function Δ(ξx , ξy ) and (b) surface rotation function θy (ξx , ξy ).

5
F. Vicencio and N.A. Alexander Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

2.5. Using static moment-surficial rotation field to determine global
stiffness coefficients of K matrix
Once the surface decay/rotation functions are obtained, it is possible
to estimate the stiffness matrix K of the foundation-soil-foundation
system by defining different load cases. The numbers of unknown co­
efficients in the symmetric stiffness matrix K, defined in equation (2),
are 2n2 + n (from the upper triangle of the symmetric matrix K ∈
R2n×2n ). In equation (2), for a particular load case, we know both applied
moments μ at foundations and their associated foundation rotation θ
(from equations (12) and (15)), this would give us 2n equations. Thus,
we need many independent load cases to produce a rank sufficient
system of algebraic equations in order to fully determine all the co­
efficients of K. Each load case i represents 2n algebraic equations in
terms of these 2n2 + nunknowns. Hence, we need at least q load cases,
where q is,
1
q≥n+ , q∈N
2
The presence of the (1/2) in eqn (17) suggests it is difficult to avoid
an overdetermined system. However, using more load cases then is
absolutely necessary can help via averaging to produce more stable re­
sults. Therefore, we use the least-squares method [46] to determine the
stiffness coefficients K, by minimizing the error ε in the following
use of eqns (15) and (16), thus θi contains all non-zero elements.
The weakness of equation (18) is that for large building arrange­
ments, the method becomes a little computationally time-consuming.
This is mainly because all the buildings are connected with each
other, so the number of unknowns grows quadratically. Therefore, for a
large cluster of buildings, we can follow the approach presented in
Appendix A based on Kronecker products. In the following section, the
least-squares method was used.
2.6. Comparison of 3D method 2 with the 2D solutions presented in Refs.
[34]
In order to verify our method, we compare the stiffness coefficients
ky1 and κy12 , with the values calculated in previous research [34], for
two foundations located with different centre to centre distances ξx and
ξy . The relationship, equation (2), between the applied moment μ and
(17) the rotation Θ in a two-building system simplifies to the following,
Aldaikh et al. [34] performed a finite element analysis in the soft­
ware PLAXIS 2D for two identical footings spaced at different distances.
This model is a plane strain model in 2D (per unit length in the
out-of-plane direction) with linear elastic materials. The soil was
modelled using an unstructured mesh of 15-node triangular elements. It

⎡ μ ⎤ ⎡k + κ + κ 0 − κx12 − κxy12
⎤⎡ ⎤
x1 x1 x12 xy12 θx1
⎢ μy1 ⎥ ⎢ 0 ky1 + κy12 + κxy12 − κxy12 − κy12 ⎥⎢ θy1 ⎥
⎣ μ ⎦=⎢ ⎣
⎥⎣
⎦ θx2 ⎦ (19)
x2 − κx12 − κxy12 kx2 + κx12 + κxy12 0
μy1 − κxy12 − κy12 0 ky2 + κy12 + κxy12 θy2

expression, is termed here as “Empirical 2D Aldaikh [34]”. In addition, Aldaikh et al.

[ ] [ ]
ε = minμ − KΘ2 , μ = μ1 , ⋯, μq , θ = θ1 , ⋯, θq
where moments μi and rotations θi are define for load case “i”. The sug­
gestion in this paper is that only one footing should have one applied
moment on it. Therefore μi is a single-entry vector. Each footing could
have 2 load cases (for x and y directions) and hence a total of q = 2n load
cases are proposed. The rotation vectors θi are determined by making
[34] performed a physical experiment for the case of two adjacent rigid
(18)
foundations, here termed “Experimental results Aldaikh [34]”.
Fig. 6(a) shows the variation of foundation stiffness auto-rotational
coefficients ky1 (with a neighbouring footing) relative to the stiffness
of a singleton, completely isolated, footing ks . Fig. 6(b) depicts the
variation of the inter-rotational stiffness κy12 (normalized by ky1 ) with
the nondimensional inter-foundation distances ξx and ξy . It is termed
“empirical 3D, equation (14)” in this paper. In general, the results of the

Fig. 6. Comparison of FEA empirical 3D analysis, FEA empirical 2D analysis [34] and experimental data [34] for: (a) foundation stiffness relative to a single footing;
(b) rotational interaction springs relative to individual footing stiffness.

6
F. Vicencio and N.A. Alexander
Fig. 7. Variation of cross-coupled interaction stiffness with the normalized
centre to centre spacing.
FEA empirical 3D agree very well with both FEA 2D and experimental
data, for ξy = 0 (2D case).
The increase of the auto-rotational stiffness ky1 reaches a maximum
value of 125% of ks , Fig. 6, while the inter-rotational stiffness ky12 is only
− 20% of ky1 at the same distance ξx = 1, ξy = 0. Also, the interaction
between the buildings is practically negligible at a spacing larger than
2.5 times the foundation’s width (i.e. ky1 ≈ ks and κy12 ≈ 0).
Using Method 2 (the FEA calibrated empirical 3D form, equation
(14)) it is possible to calculate the cross-coupled interaction stiffness
κyx12 , and this is plotted in Fig. 7. The value of κxy12 decreases as the
footing spacing increases and it is zero at a spacing equal to ξy = 0,
because the rotation θxi and θyi are decoupled in this position. The cross-
coupled interaction stiffness κyx12 reaches a value of only 8.6% of the
auto-rotational spring stiffness ky1 (at ξx = 1, ξy = ± 0.5). Thus, we
demonstrate that the cross-coupling inter-rotational springs κyxij are
likely to only produce second order effects.
2.7. Equation of motion of the complete system
Once the interaction springs between buildings κij and rotational
spring stiffnesses of the soil beneath each buildings kxi and kyi were
calculated, we can obtain the complete set of equations of motion, as
follows.
Mẍ + Cẋ + Kx = pẍg
where the mass matrix M, the damping matrix C, the stiffness matrix K,
the force vector p and the DOFs vector x correspond to the complete
system of n buildings. For a complete derivation of the matrices of
equation (20), please refer to Appendix B.
In this paper, the dynamic properties of the building “i” required for
Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx
equation (20) are calculated according to the following assumptions.
a. For the fundamental natural period of the structure on a rigid
foundation (i.e. with no foundation/soil rotation), we adopt the
approximate empirical relationship proposed by EC8 [47], Txi =
3/4
Tyi = Ti = c0 hi . Where the height of the building is taken in me­
tres, and the factor equal to c0 = 0.075.
b. Newmark and Rosenblueth [48] consider that the volume of soil
mass beneath a square base building is approximately equal to msi =
0.35ρs b3i .
c. The mass of the building can be approximated as mbi = ρb b2i hj .
d. The radius of gyration is calculated according to the Newmark’s
empirical expression ri = 0.33bi .
e. It has been demonstrated that SSSI on structures supported on loose
soil may exhibit significant interaction [30–33], so the soil properties
used correspond to loose sand, where the soil density is ρs =
1300[kg /m3 ], the shear wave velocity is Vs = 156[m /s] and the
Poisson’s ratio of the soil is νs = 0.3.
f. The average building density can be considered as ρb = 600[kg /m3 ].
2.8. Response performance measures of interest
We are interested in evaluating the change in the response between
the coupled (SSSI) and the uncoupled (soil-structure interaction, SSI)
system. First the coupled solution of equation (20) is determined
through time-history analysis. Then, the uncoupled response of each
buildings (i.e. SSI response) is computed similarly. The equations of
motion for the special case of uncoupled buildings response (SSI) are
obtained by completely removing all coupling springs and their stiff­
ening effects. As a measure of change in the response between SSSI and
SSI, we will employ total acceleration A1 (horizontal sway + ground +
rocking) for the top of the building i, denoted by,
Ai = ẍi + ẍg − hi θ̈yi (21)
In addition, it is valuable to characterize the change in total power
caused by the multiple SSSI among the buildings. So, the percentage
change in total power χ̈ i for the building i, when using the uncoupled SSI
analyses rather than coupled SSSI analyses is expressed in terms of the
total power spectral densities EPSD (Ai ),
{ }
χ̈ i = 100
[EPSD (Ai )]SSSI
− 1 (22)
[EPSD (Ai )]SSI
(20)
Fig. 8. Acceleration ground motion and power spectral density (Superstition
Hills-02 November 24, 1987).
7
F. Vicencio and N.A. Alexander Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

Fig. 9. Multi buildings SSSI model in 3D.

Fig. 11. Change in acceleration power χ̈ 1 for building 1 as a function of the

position of building 2 – Building 1 smaller than building 2, (T 1 = 0.6[s], T2 =
0.66[s], h1 /b = 1.6).

analyses are carried out for the Superstition Hill earthquake (11/24/
1987, Imperial Valley Wildlife SH-02 Station). Fig. 8 shows the ground
motion timeseries and the power spectral density of the record. The
record data was obtained from the Pacific Earthquake Engineering
(PEER) Research Center Database [49]. The ground motion was recor­
ded on weak soil, where the average shear wave velocity is less than
180[m /s] [47], i.e. loose sand.

3.1. Case 1: A comparison between 2D and 3D seismic analyses for two

adjacent buildings

In order to compare the differences between 2D and 3D analyses, the

system of two buildings shown in Fig. 9 is considered. Building 2 can be
in any centre to centre distance ξx , ξy respect to building 1. Both build­
ings have a similar square plan area of b2 .
Fig. 10. (a) Total acceleration responses, (b) Power spectra density of total
acceleration – Building 1 smaller than building 2, (T 1 = 0.6[s], T 2 = 0.66[s],
h1 /b = 1.6).

where EPSD (Ai ) is based on the Fourier Transform of all data points of the
response acceleration time-series Ai . The change of power χ̈ i would be
zero if there is no difference in overall response power between SSSI and
SSI analyses. Using the equation (22) as a comparative metric, delivers a
statistical estimate of magnitude that is more robust than employing a
single peak of the function (displacement or acceleration).

3. Numerical studies

In this section we apply the approach described previously, the

reduced order model equation (20), to explore some multi-building SSSI
3D cases. These anecdotal cases are designed to explore the performance
of the reduced order model equation (20) rather than explore a full
parametric, geometric and ground motion variation of the multi-
building SSSI 3D phenomenology as this is beyond the scope of this
current paper.
First, results (case 1) for a two-foundation system in 3D with
different interbuilding distances are presented. Subsequently, the solu­
tion for a set of seven buildings in an L shape arrangement (case 2) is Fig. 12. Change in acceleration power χ̈ 1 for building 1 as a function of the
analysed. Lastly, a city block of 12 buildings (case 3) is examined. These position of building 2 – Building 1 taller than building 2, (T 1 = 0.6[s], T2 =
0.54[s], h1 /b = 1.6).

8
F. Vicencio and N.A. Alexander
Fig. 10(a) shows the uncoupled SSI (blue line) and coupled SSSI (red
line) response for the top of building 1 (namely the total acceleration
A1 ), when the interbuilding distance is equal to ξx = 1.1 and ξy = 0. The
rigid base period of building 1 is T1 = 0.6[s] and T2 = 0.66[s] for building
2 (i.e. the second building is approximately 13.5% taller than the first
building). Building 1 has a height to width ratio equal to h1 / b = 1.6.
Fig. 10(b) shows the corresponding power spectra density for the total
acceleration for building 1 considering the coupled (SSSI) and uncou­
pled (SSI) case. Comparing the uncoupled and coupled responses,
building 1’s response appears to be significantly affected by the presence
of the taller building in all the time-history. The change in power,
defined in equation (22), is equal to χ̈ 1 = 37.4% for the acceleration.
If we compare the differences in the peak of the time-series, the
percentage of increase in the peak acceleration A1 for building 1 is of
12.5%.
Fig. 11 displays the variation of change in power for the acceleration
χ̈ 1 (A1 ) on the top of the building 1, where the centre-to-centre inter­
building spacing ξx , ξy varies between − 3.0 and 3.0. Note, building 1 is
smaller than building 2. The critical zones in this figure are red, i.e.
where the building’s total response power is amplified by the presence of
building 2. Blue indicates when the response power is reduced. The
inner black rectangle indicates the footprint of building 1, while the
outer black rectangle highlights the touching centre-to-centre distances
between buildings 1 and 2.
The maximum increase in the response occurred when building 2 is
close to building 1 and aligned to the direction of the earthquake (i.
⃒ ⃒
e. ⃒ξy ⃒ ≈ 0), where the change in power are χ̈ 1 = 50% for the accelera­
tion. As expected, the effects of SSSI diminish as the interbuilding dis­
tance increases, being practically negligible at ξx ,ξy ≥ 2.5. On the other
⃒ ⃒
hand, for the position of building 2 out of plane XZ (i.e. ⃒ξy ⃒ ≥ 1.1), the
building 1’s total power response attenuates, with a maximum of χ̈ 1 = −
22.5% for the acceleration when the buildings are close together. This
reduction on the response was not explored in the previous papers
[30–34] because these papers performed 2D analysis.
Therefore, while the 2D analysis seems to capture the worst case of
two buildings and unidirectional ground motions (the largest detri­
mental effect of SSSI), it does not allow a complete understanding of the
SSSI effects between two adjacent buildings, especially the cases of ad­
jacency out of plane that produce beneficial effects here.
Fig. 12 displays the variation of change in power for the acceleration
χ̈ 1 at the top of building 1 for the case where building 1 is taller than
building 2. These results correspond to the case where the rigid base
Fig. 13. Change in acceleration power due to 3D SSSI for L shape arrangement. Equispaced identical buildings, Ti = 0.5[s], hi /b = 1.5, ξx = 1.2 and ξy = 1.2.
9
Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx
period of building 1 is T1 = 0.6[s] and T2 = 0.54[s] for building 2 (i.e. the
second building is 13.1% smaller than the first building). Building 1 has
a height to width ratio equal to h1 /b = 1.6. Unlike the previous figure,
the position of the red peaks shifts between the graphs, where the
maximum increase in the response (χ̈ 1 = 14%) corresponds to the po­
⃒ ⃒
sition of building 2 equal to ⃒ξy ⃒ = 1.1 and ξx = 0. The maximum
decrease in the response occurred when the building 2 is closed to
⃒ ⃒
building 1 and aligned to the direction of the earthquake (i.e. ⃒ξy ⃒ ≈ 0),
where the change in power is χ̈ 1 = − 36% for the acceleration.
Thus, again the 2D analysis [30–34] highlights the worst case for the
case of two buildings and unidirectional ground motions (the largest
beneficial effect of SSSI) it does not capture the cases of adjacency out of
plane that produce detrimental effects here.
3.2. Case 2: L shape arrangement of identical equispaced buildings
In this section, a selected L shape arrangement is presented in order
to measure the influence of multi-building SSSI in 3D. All the buildings
are identical, having a rigid base period of Ti = 0.5[s], the same square
plan area and buildings have a height to width ratio equal to hi /b = 1.5.
The centre-to-centre interbuilding distances are equispaced at ξx = 1.2,
ξy = 1.2. Fig. 13 displays the variation of change in power for the ac­
celeration on top of the buildings. The maximum increase in total ac­
celeration power response occurred on the buildings aligned with the
earthquake, with a maximum of χ̈ 3 = 64% for the acceleration on
building 3. In addition, there is reduction in the response for some
buildings, with a maximum for the building 1 (χ̈ 1 = − 17.6%).
Fig. 14(a) and (b) shows the uncoupled SSI (blue line) and coupled
SSSI (red line) response for the top of the building 1 and 3 and Fig. 14(c)
and (d) shows the corresponding power spectra density for the total
acceleration for these buildings. Comparing the uncoupled and coupled
responses, building 3’s response appears to be significantly affected by
the presence of the adjacent buildings. The change in power, is equal to
χ̈ 3 = 64% for the acceleration. On the other hand, for building 1, there is
a reduction of the seismic response, with a change of power equal to
χ̈ 1 = − 17.6%.
If we compare the differences in the peak of the time-series, the
percentage of increase or reduction in the peak accelerations for build­
ings 1 and 3 are − 5.7% and 10.4% respectively.
F. Vicencio and N.A. Alexander Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

Fig. 14. L plan of equispaced identical buildings. Total acceleration responses for (a) building 1, (b) building 3, Power spectra density (c) building 1 and (d) building
3. T i = 0.5[s], hi /b = 1.5, ξx = 1.2 and ξy = 1.2.

Fig. 15. Change in acceleration power due to 3D SSSI for a city block of twelve equispaced identical buildings. T i = 0.5[s], hi /b = 1.5, ξx = 1.2 and ξy = 1.2.

Fig. 16. City block of twelve equispaced identical buildings. Total acceleration responses for (a) building 5, (b) building 10, Power spectra density (c) building 5 and
(d) building 10. T i = 0.5[s], hi /b = 1.5, ξx = 1.2 and ξy = 1.2.

10
F. Vicencio and N.A. Alexander Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

3.3. Case 3: A city block arrangement of identical equispaced buildings
Fig. 15 shows the variation of change in power for the acceleration at
the top of the buildings, for a city block of twelve identical equispaced
buildings. All the buildings have a rigid base period equal to Ti = 0.5[s]
and the buildings have a height to width ratio equal to h1 / b = 1.6. The
centre-to-centre interbuilding distance are the same between the adja­
cent buildings, ξx = 1.2, ξy = 1.2. The maximum increase in total ac­
celeration power response occurred in the central buildings, with a
maximum of χ̈ 10 = 65.1% for the acceleration on building 2, 3, 10 and
11. Due to the symmetry of the buildings’ arrangement, the change in
the seismic response is also symmetric.
Fig. 16(a) and (b) show the acceleration response for the top of the
building 5 and 10, and Fig. 16(c) and (d) show the corresponding power
spectral density for these buildings. Comparing the uncoupled and
coupled responses, building 10’s response appears to be significantly
affected by the presence of the adjacent buildings. The change in power,
is equal to χ̈ 10 = 65% for the acceleration. On the other hand, for
building 5, there is a reduction of the seismic response, with a change of
power equal to χ̈ 5 = − 10.4%. If we compare the differences in the peak
of the time-series, the percentage of increase or reduction in the peak
accelerations for buildings 5 and 10 are − 3.3% and 13% respectively.
4. Conclusion
In this paper, we present a theoretical linear elastic formulation (a
reduced-order model) for the evaluation of Structure-Soil-Structure
interaction between several buildings in a three-dimensional arrange­
ment, that are coupled through the soil. The buildings can have different
sizes, properties, and inter-building spacing, but it just allows square
plan buildings (although it can be readily extended to rectangular plan
shapes). This conceptual model reduces the number of dofs from perhaps
hundreds of millions in a conventional full 3D FEA to 4n (where n is the
number of buildings). Thus, city-site, and city block seismic time history
analyses are far more tractable.
The key computational stage is to assign values to all the auto-
rotational and inter-rotational soil/foundations spring stiffnesses. We
present a novel approach for estimating these stiffnesses that makes use
of least-square inverse system identification based on an FEA calibrated
empirical relationship between applied surficial moments and rotations.
We explore two different numerical methods for determining stiffness
coefficients using (i) the explicit symmetrical form of the stiffness matrix
(using a least squares toolbox) and (ii) based on the theory of Kronecker
products and Moore-Penrose pseudo inverse (which is far more
computationally efficient but doesn’t completely guarantee symmetry).
Results indicate that both methods, which are conceptually least-squares
approaches, produce very similar results.
The proposed FEA calibrated empirical fit for the surficial displace­
ment field (loosely based on a Boussinesq form) showed excellent
agreement with 3D FEA and experimental results. In addition, the auto-
rotational and inter-rotational soil spring stiffness coefficients showed
excellent agreement with previous FEA 2D and experimental data in the
literature. Results indicate that the hitherto unused (in 2D) cross-
coupled inter-rotational soil spring stiffnesses κyxij may only have sec­
ond order effects on system responses. We did, however, include them in
our analyses.
We presented three numerical cases to explore the performance of
the reduced-order model. These cases were anecdotal, and designed to
explore the versatility of the algorithm rather than an extensive explo­
ration of the phenomenology of 3D SSSI. Case 1 compared the differ­
ences between 2D and 3D SSSI for the case of two similar but not
identical buildings. Results suggest that the 2D analysis seems to capture
the worst case (the largest detrimental or largest beneficial effect of
SSSI). However, it does not tell the whole story. Two buildings that are
adjacent out of plane behave completely differently to those adjacent in
plane. Case 2 and Case 3 explore a set of equispaced identical buildings
in an L plan configuration and a square block configuration respectively.
Both cases indicate large detrimental effects are possible for buildings
within a group parallel to the main seismic direction of excitation and
suggest that corner buildings do not suffer the most detrimental effects.
Declaration of competing interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
Acknowledgements
The Ministry of Education, Chile and Commission for Scientific and
Technological Research (CONICYT) through grant BCH 72170305 has
granted financial support to the PhD student during this research. The
researchers are very grateful for the support of the Faculty of Engi­
neering, at the University of Bristol. Most of the computations necessary
for this work were carried out on the BlueCrystal supercomputers of
advanced computing research centre of the University of Bristol. This
work was supported by the EPSRC-funded research project EP/
R012806/, UKCRIC - Bristol Soil-Foundation-Structure Interaction Fa­
cility (SoFSI).

Appendix A. Stiffness matrix estimate using the theory of Kronecker products

A different approach for the same problem is using the theory of Kronecker products for inverse system identification. Given that there is, in
principle, a unique solution to this inverse problem we are left with the problem of rearranging equations,
[ ] [ ]
μ = KΘ, μ = μ1 , ⋯, μq , Θ = θ1 , ⋯, θq (23)

into a form that is easily tractable. This is a set of algebraic equations where, non-typically, the unknowns are in a matrix rather than a vector. The
theory of Kronecker products enables us to solve this kind of problem. Consider the following matrix equations,
AXB = C (24)
In which we want to determine all terms in matrix X. Consider a vectorization function vec( •) which is an isomorphism (a linear transformation)
that converts a matrix into a vector [50]. The vectorization function vec( •) is performed in Matlab using the reshape( ) function. When this vecto­
rization function is applied to equation (24) we obtain
( T )
B ⊗ A vec(X) = vec(C) (25)

where the symbol ⊗ is the Kronecker product (using the kron( ) function in Matlab). Hence, we can state the solution,

11
F. Vicencio and N.A. Alexander Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

(( )+ )
X̃ = vec− 1
BT ⊗ A vec(C) (26)

where vec− 1 ( •) is the inverse vectorization isomorphism and (•) is the Moore-Penrose pseudo-inverse (which can be determined using a singular-
+

value decomposition in Matlab using the backslash operator). Thus, for equation (23) we substitute A = I, X̃ = K̃, B = Θ, and C = μ into equation
(26) and therefore,
(( )+ )
K̃ = vec− 1 ΘT ⊗ I vec(μ) (27)

This is achieved by the following one line of Matlab script

Kest = reshape(kron(Theta’, eye(n)) \ reshape(mu, q*n, 1),n, n)

The weakness of this method is that it does not enforce symmetry in K̃. This must be imposed through averaging. Additionally, this method doesn’t
constrain the zero elements of equation (2) to be exactly zero. Thus, the methods presented in equations (18) and (27) are fundamentally both least
squares approaches. However, equation (18) adds additional constraints which are not explicitly present in equation (27). Nevertheless, the differ­
ences between both approaches should be considered second order effects, with equation (27) significantly quicker computationally.

Appendix B. Derivation of the equations of motion

Employing Euler-Lagrange energy mechanics, the equations of motion describing the dynamics of the discretised system of Fig. 1 is formulated.
The kinetic energy TE and the potential energy UE for this system are written respectively as follows,
n ( ( )2 ( )2 ( ))
1∑
TE = mbi ẋi + ẋg − hi θyi + mbi ẏi − hi θxi + msi ri2 θ2xi + θ2yi (28)
2 i=1

n ( ) 1∑
1∑ n− 1 ∑n
( ( )2 ( )2 ( )2 )
UE = kbi xi2 + kbi y2i + kxi θ2xi + kyi θ2yi + κxij θxj − θxi + κyij θyj − θyi + κxyij θyj − θxi (29)
2 i=1 2 i=1 j=i+1

where hi is the height, mbi is the total lumped modal mass of building i and msi is the foundation/soil mass underneath building i. kbi is the modal
building lateral stiffness and ri is the soil/foundation mass radius of gyration. The stiffness coefficient of the inter-rotational springs between buildings
i and j, namely κxij , κyij and the cross-coupled coefficient κxyij , and the auto-rotational spring stiffnesses of the soil beneath each building kxi and kyi .
Hence, the Euler-Lagrange equations of motion describing the dynamics of the discretised system can be derived in the standard way by calculus and is
written in matrix form, as follows,
Mẍ + Cẋ + Kx = pẍg (30)

where the mass matrix M, the damping matrix C, the stiffness matrix K, the force vector p and the DOFs vector x of the complete system of n buildings
are stated as follows,
⎡ ⎤ ⎡ ⎤
̂
M ⋯ 0 ⋯ 0 ̂
K ⋯ κ ̂ 1i ⋯ κ ̂ 1n
⎢ 1 ⎥ ⎢ 1 ⎥
⎢ ⋮ ⋱ ⋮ ⋱ ⋮ ⎥ ⎢ ⋮ ⋱ ⋮ ⋱ ⋮ ⎥
⎢ ⎥ ⎢ ⎥
̂
M = ⎢ 0 ⋯ M i ⋯ 0 ⎥, K = ⎢ κ ̂ ̂
⋯ Ki ⋯ κ ̂ in ⎥ (31)
⎢ ⎥ ⎢ i1 ⎥
⎣ ⋮ ⋱ ⋮ ⋱ ⋮ ⎦ ⎣ ⋮ ⋱ ⋮ ⋱ ⋮ ⎦
0 ⋯ 0 ⋯ ̂ Mn ̂ n1 ⋯ κ
κ ̂ ni ⋯ K ̂n

[ ]T [ ]T
p= p̂1 ⋯ ̂
pi ⋯ ̂
pn , x= ̂x̂
x1 ⋯ ̂
x̂xi ⋯ ̂
x̂xn , x i = [ xi
̂ yi θyi θxi ]T (32)
( )
∑4n
2ξj ωj
C=M φ φT
M (33)
j=1
φTj Mφj j j

The system’s linear viscous damping matrix C assumes that each natural mode is damped at ξj = 0.05 of the critical damping, φj is the modal
eigenvector of the mode j, ωj are the natural frequencies of the system. Thus, the Caughey orthogonal damping matrix C can be calculated as [51], by
equation (33). Note that the damping matrix refers to the coupled system. The fundamental frequencies ωj of the coupled system do not change very
much compared to the uncoupled system [32], with a maximum of 9% variation in the natural frequencies. Hence, the damping matrix does not vary
substantially between the SSSI and SSI systems.
The global mass matrix M corresponds to a diagonal block matrix, where the different blocks ̂ M i represent the mass matrix for each building i.
⎡ ⎤
mbi 0 − mbi hi 0
⎢ ⎥
⎢ 0 mbi 0 − mbi hi ⎥
⎢ ⎥
̂
Mi = ⎢ 2 2 ⎥ (34)
⎢ − mbi hi 0 mbi hi + msi ri 0 ⎥
⎣ ⎦
0 − mbi hi 0 mbi h2i + msi ri2

The global stiffness matrix K includes the interaction effects between the buildings. So, the diagonal block terms K ̂ i is the stiffness matrix for each
building i, including the additional stiffening effect of the adjacent footings. The off-diagonal block terms ̂
κ ij represent the interaction between the
buildings i and j.

12
F. Vicencio and N.A. Alexander Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

⎡ ⎤
kbi 0 0 0
⎢0 ⎥ ⎡ ⎤
⎢ kbi 0 0 ⎥ 0 0 0 0
⎢ ∑
n
( ) ⎥ ⎢0
̂i =⎢
⎥ 0 0 0 ⎥
K ⎢0 0 kxi + κxij + κxyij 0 ⎥,
⎥ ̂ 1i = ⎢
κ ⎢
⎥
⎥ (35)
⎢ j=1
⎥ ⎣0 0 − κx1i − κxy1i ⎦
⎢ ∑
n
( )⎥
⎣ 0 0 − κxy1i − κy1i
0 0 0 kyi + κyij + κxyij ⎦
j=1

Finally, the global excitation vector p is assembled by the block vector ̂

p i of each building i.
̂ i = [ − mbi
p 0 − mbi hi 0 ]T (36)

References [25] Laurenzano G, Priolo E, Gallipoli MR, Mucciarelli M, Ponzo FC. Effect of vibrating
buildings on free-field motion and on adjacent structures: the Bonefro (Italy) case
history. Bull Seismol Soc Am 2010;100:802–18.
[1] Lee TH, Wesley DA. Soil–structure interaction of nuclear reactor structures
[26] Gueguen P, Colombi A. Experimental and Numerical evidence of the clustering
considering through-soil coupling between adjacent structures. Nucl Eng Des 1973;
effect of structures on their response during an earthquake: a case study of three
24:374–87.
identical towers in the city of Grenoble, France. Bull Seismol Soc Am 2016;106(6):
[2] Luco JE, Contesse L. Dynamic structure–soil–structure interaction. Bull Seismol Soc
2855–64.
Am 1973;63:1289–303.
[27] Karabalis DL, Huang F. 3-D foundation-soil-foundation interaction. In: Kassab AJ,
[3] Wong HL, Trifunac MD. Two-dimensional, antiplane, building–soil–building
Brebbia CA, editors. Boundary element technology IX. Southampton:
interaction for two or more buildings and for incident planet SH waves. Bull
Computational Mechanics Publications; 1994.
Seismol Soc Am 1975;65:1863–85.
[28] Mulliken JS, Karabalis DL. Discrete model for foundation-soil-foundation
[4] Triantafyllidis T, Prange B. Dynamic subsoil coupling between rigid, rectangular
interaction. Soil dynamics and earthquake engineering VII. Southampton: CMP,
foundations. Soil Dynam Earthq Eng 1987;6(3):164–79.
Publications; 1995.
[5] Qian J, Beskos DE. Dynamic interaction between 3-D rigid surface foundations and
[29] Mulliken JS, Karabalis DL. Discrete model for dynamic through-the-soil coupling of
comparison with the ATC-3 provisions. Earthq Eng Struct Dynam 1995;24:419–37.
3D foundations and structures. Soil Dynam Earthq Eng 1998;27:678–710.
[6] Qian J, Beskos DE. Harmonic wave response of two 3-D rigid surface foundations.
[30] Alexander NA, Ibraim E, Aldaikh H. A simple discrete model for interaction of
Soil Dynam Earthq Eng 1996;15(2):95–110.
adjacent buildings during earthquakes. Comput Struct 2013;124:1–10.
[7] Karabalis DL, Mohammadi M. 3-D dynamic foundation-soil-foundation interaction
[31] Vicencio F, Alexander NA. Dynamic interaction between adjacent buildings
on layered soil. Soil Dynam Earthq Eng 1998;17:139–52.
through nonlinear soil during earthquakes. Soil Dynam Earthq Eng 2018;108:
[8] Lehmann L, Antes H. Dynamic structure–soil–structure interaction applying the
130–41.
symmetric galerkin boundary element method (SGBEM). Mech Res Commun 2001;
[32] Vicencio F, Alexander NA. Higher mode seismic structure-soil-structure interaction
28:297–304.
between adjacent building during earthquakes. Eng Struct 2018;174:322–37.
[9] Liang J, Han B, Todorovska MI, Trifunac M. 2D dynamic structure-soil-structure
[33] Vicencio F, Alexander NA. Dynamic Structure-Soil-Structure Interaction in
interaction for twin buildings in layered half-space II: incident SV-waves. Soil
unsymmetrical plan buildings due to seismic excitation. Soil Dynam Earthq Eng
Dynam Earthq Eng 2018;113:356–90.
2019;127:105817.
[10] Ghandil M, Behnamfar F, Vafaeian M. Dynamic responses of structure-soil-
[34] Aldaikh H, Alexander NA, Ibraim E, Knappett J. Evaluation of rocking and coupling
structure systems with an extension of the equivalent linear soil modelling. Soil
rotational stiffness coefficient of adjacent foundations. International Journal of
Dynam Earthq Eng 2016;80:149–62.
Geomechanics. ASCE 2018;18(1). 04017131.
[11] Wang HF, Lou ML, Zhang RL. Influence of presence of adjacent surface structure on
[35] Gueguen P, Bard PY, Chavez-Garcia F. Site-city seismic interaction in Mexico city-
seismic response of underground structure. Soil Dynam Earthq Eng 2017;100:
like environments: an analytical study. Bull Seismol Soc Am 2002;92(2):794–811.
131–43.
[36] Boutin C, Roussillon P. Assessment of the urbanization effect on seismic response.
[12] Wang S, Schmid G. Dynamic structure–soil–structure interaction by FEM and BEM.
Bull Seismol Soc Am 2004;94(1):251–68.
Comput Mech 1992;5:347–57.
[37] Marc Kham, Semblat JF, Bard PY, Dangla P. Seismic site-city interaction: main
[13] Qian J, Tham LG, Cheung YK. Dynamic cross-interaction between flexible surface
governing phenomena through simplified numerical models. Bull Seismol Soc Am
footings by combined BEM & FEM. Earthq Eng Struct Dynam 1996;25:509–26.
2006;96(5):1934–51.
[14] Padron LA, Aznarez JJ, Maeso O. Dynamic structure–soil–structure interaction
[38] Ghergu M, Ionescu IR. Structure-soil-structure coupling in seismic excitation and
between nearby piled buildings under seismic excitation by BEM–FEM model. Soil
“city effect”. Int J Eng Sci 2009;47:342–54.
Dynam Earthq Eng 2009;29:1084–96.
[39] Isbiliroglu Y, Taborda R, Bielak J. Coupled soil-structure interaction effects of
[15] Alamo GM, Padron LA, Aznares JJ, Maeso O. Structure-soil-structure interaction
buildings clusters during earthquakes. Erthq Spectra 2015;31(1):463–500.
effects on the dynamic response of piled structures under obliquely incident
[40] Schwan L, Boutin C, Padron LA, Dietz MS, Bard PY, Taylor C. Site-city interaction:
seismic shear waves. Soil Dynam Earthq Eng 2015;78:142–53.
theoretical, numerical and experimental crossed-analysis. Geophs J Int 2016;205:
[16] Trombetta NW, Mason HB, Chen Z, Hutchinson TC, Bray JD, Kutter BL. Nonlinear
1006–31.
dynamic foundation and frame structure response observed in geotechnical
[41] Newmark NM. Torsion of symmetrical buildings. In: Proceedings at 4th world
centrifuge experiments. Soil Dynam Earthq Eng 2013;50:117–33.
conference on earthquake engineering; 1969. Santiago, Chile.
[17] Mason HB, Trombett NW, Chen Z, Bray JD, Hutchinson TC, Kutter BL. Seismic
[42] Veletsos AS, Prasad AM, Wu WH. Transfer functions for rigid rectangular
soil–foundation–structure interaction observed in geotechnical centrifuge
foundations. Earthq Eng Struct Dynam 1997;26(1):5–17.
experiments. Soil Dynam Earthq Eng 2013;48:162–74.
[43] Nist Gcr 12-917-21. Soil-structure interaction for building structures. National
[18] Knappett J, Massen P, Caucis K. Seismic structure-soil-structure interaction
Institute of Standards and Technology. NEHRP Consultants Joint Venture; 2012.
between pairs of adjacent building structures. Geotechnique 2015;65(5):429–41.
[44] Pais A, Kausel E. Approximate formulas for dynamic stiffnesses of rigid
[19] Aldaikh H, Alexander N, Ibraim E, Knappett J. Shake table testing of the dynamic
foundations. Soil Dynam Earthq Eng 1988;7(4):213–27.
interaction between two and three adjacent buildings (SSSI). Soil Dynam Earthq
[45] Austrell PE, Dahlblom O, Lindemann J, Olsson A, Olsson KG, Persson K,
Eng 2016;89:219–32.
Petersson H, Ristinmaa M, Sandberg G, Wernberg PA. In: CALFEM A finite element
[20] Aldaikh H, Alexander NA, Ibraim E, Oddbjornsson O. Two dimensional numerical
toolbox to MATLAB Version 3.4. Lund University, Division of Structural Mechanics
and experimental models for the study of structure-soil-structure interaction
and Division of Solid Mechanic; 2004.
involving three buildings. Comput Struct 2015;150:79–91.
[46] Woodgate KG. Efficient stiffness matrix estimation for elastic structures. Comput
[21] Jabary RN, Madabhushi SPG. Structure-soil-structure interaction effects on
Struct 1998;69:79–84.
structures retrofitted with tuned mass dampers. Soil Dynam Earthq Eng 2017;100:
[47] European Standard En 1998-1. Eurocode 8: design of structures for earthquake
301–15.
resistance.
[22] Li P, Liu S, Lu Z, Yang J. Numerical analysis of a shaking table test on dynamic
[48] Newmark NM, Rosenblueth E. Fundamentals of earthquake engineering. Prentice-
structure-soil-structure interaction under earthquake excitation. Struct Des Tall
Hall; 1971.
Special Build 2017;26:e1382.
[49] PEER strong motion Database. http://ngawest2.berkeley.edu/.
[23] Çelebi M. Seismic response of two adjacent buildings. I: data and analysis. J Struct
[50] Petersen KB, Pedersen MS. The matrix cookbook. Technical university of Denmark.
Eng 1993;119(8):2461–76.
2007.
[24] Çelebi M. Seismic response of two adjacent buildings. II: interaction. J Struct Eng
[51] Clough RW, Penzien J. Dynamics of structures. second ed. McGrawHill Int; 1993.
1993;119(8):2477–92.

13