Computers and Structures 71 (1999) 397±412

Some modeling aspects in the nonlinear ®nite element

analysis of cable supported bridges
Raid Karoumi
Department of Structural Engineering, Royal Institute of Technology, S-100 44, Stockholm, Sweden
Received 2 December 1997; accepted 12 November 1998

Abstract

This paper presents a method for modeling cable supported bridges for nonlinear ®nite element analysis. A two-
node catenary cable element, derived using exact analytical expressions for the elastic catenary, is proposed for the
modeling of cables. The cable element tangent sti€ness matrix and internal force vector are evaluated accurately and
eciently using an iterative procedure. The reliability and eciency of the formulations used are demonstrated by
studying the behavior of the Great Belt suspension bridge during girder erection and the behavior of a cable-stayed
bridge. Eigenfrequency analyses are also conducted and the results show good agreement when compared with
previously published data. # 1999 Elsevier Science Ltd. All rights reserved.

Keywords: Cable element; Beam element; Finite element method; Nonlinear analysis; Cable-stayed bridges; Suspension bridges;
Great Belt suspension bridge

1. Introduction ¯exible and undergo large displacements before attain-

ing their equilibrium con®guration. Due to this inher-
Due to their esthetic appearance, ecient utilization ently nonlinear behavior, conventional linear analysis
of structural materials and other notable advantages, which assumes small elastic deformations and displace-
cable supported structures, such as suspended roofs, ments is often not applicable [3,4].
guyed towers, cable-stayed bridges or suspension A brief early history of the research into the beha-
bridges, have gained much popularity in recent dec- vior of cables has been published in [5], as well as a
ades. Among bridge engineers the popularity of cable- more recent history in [6]. Methods of static and
stayed bridges has increased tremendously. Bridges of dynamic analysis and the behavior of cable structures
this type are now entering a new era with main span are thoroughly presented in [1,3,5,7]. In [6,8,9], trial-
lengths reaching 1000 m [1,2,41]. This fact is due, on and-error search procedures have been proposed for
one hand to the relatively small size of the substruc- the nonlinear computer analysis of simple cable pro-
tures required and on the other hand to the develop- blems. For cable roof structures, analysis methods and
ment of ecient construction techniques and to the several very illustrative design details, e.g. details of
rapid progress in the analysis and design of this type connections, are presented in [10].
of bridge. For cable supported bridges, i.e. suspension and
The increasing attention on cable structures is not cable-stayed bridges, the trend today is to use more
only due to their inherent beauty but also to their shallow or slender sti€ening girders combined with
stubborn nature in not easily revealing the secret of increasing span lengths. For that reason, it is highly
their nonlinear behavior. Cable structures exhibit desirable in bridge engineering to develop accurate
highly geometrically nonlinear behavior, they are very procedures that can lead to a thorough understanding

0045-7949/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 4 5 - 7 9 4 9 ( 9 8 ) 0 0 2 4 4 - 2
398 R. Karoumi / Computers and Structures 71 (1999) 397±412

and a realistic prediction of the structural response. structures. The formulation of these elements is
Although several investigators [4,11±18] have studied included in this paper for the sake of clarity and also
the nonlinear behavior of cable supported bridges, for the purpose of having a paper as self-contained as
very few [16,17] have tackled the problem of using possible. Despite the fact that the cable modeling tech-
cable elements for modeling the cables. nique, based on the expressions given in [25], has been
Commercial ®nite element programs used in civil en- available for many years now, it has, to the author's
gineering today cannot be readily used for the model- knowledge, not yet been used for analysis of cable-
ing and analysis of modern cable structures as they stayed and suspension bridges. It is also believed that
lack suitable cable elements that can accurately model the beam element adopted here has not been used ear-
the actual cable curvature. As the cable represents a lier for analysis of this type of structures.
¯exible member with virtually no resistance to applied The expressions of the internal force vectors and
moments, the idea of replacing each cable by a bar el- tangent sti€ness matrices for the elements used were
ement with equivalent cable sti€ness or by several derived using the Maple software package for symbolic
beam elements with negligible moment of inertia has computations [27]. Samples of these Maple procedures
found wide acceptance and has been adopted by many are given in Appendix A.
investigators and designers using commercial codes.
As the popularity of cable structures has increased,
the search for more ecient methods has intensi®ed 2. Structure modeling
and today various other cable modeling techniques
than the crude modeling with a bar element mentioned The cable-supported structures considered in this
above can be found in the literature. In [19], a two- paper are cable-stayed and suspension bridges. Such
node curved ®nite element was developed, using cubic bridges consist of cables, pylons and girders (bridge
polynomial interpolation functions, and used for the decks) and are usually modeled using beam and bar el-
static and dynamic analysis of three-dimensional (3-D) ements for the analysis of the global structural re-
prestressed cable nets. In [20], another two-node sponse [14,37]. In the following, an alternative
curved ®nite element was developed using Lagrangian approach is presented where accurate and ecient
functions for the interpolation of element geometry. In cable and beam elements are used for the modeling.
[3,21], derivations of isoparametric cable elements, All sources of geometric nonlinearity, such as change
which includes the element curvature, are presented, of cable geometry under di€erent tension load levels
and in [16] a four-node isoparametric cable element (cable sag e€ect), change of the bridge geometry due
was used for modeling cables in cable-stayed bridges. to large displacements, and axial force±bending
An iterative analysis procedure for cables, based on moment interaction in the bridge deck and pylons (P±
using exact analytical expressions for the elastic caten- d e€ect), are considered in the present analysis.
ary, was suggested in [22,23]. This approach was later For simplicity the derivation and the modeling pre-
adopted by other investigators, developed and used for sented in this study are only in two dimensions.
the analysis of very simple cable structures [24,25] and Consequently, torsional e€ects and torsional modes of
of power transmission lines [26]. In [5], the same vibration are disregarded. As modern cable supported
approach was also suggested for the analysis of cable structures are ¯exible three-dimensional structures,
structures with appreciable sag and the applicability of two-dimensional models are of course not adequate
this method was later demonstrated in [17] on numeri- when studying the response of such structures under
cal examples of cable supported bridges. the action of environmental loads like wind, trac and
The cable element used in this study is derived using earthquakes [37,39]. However, simpli®ed two-dimen-
the exact analytical expressions for the elastic catenary sional bridge models are still very useful for bridge
given in [25]. The procedure presented later on in this designers in the preliminary design stage, e.g. for inves-
paper, to drive this cable element, is similar to that tigating the feasibility of alternative structural sol-
described in [5,17]. However, the analytical expressions utions. For the interested reader, accurate three-
for the elastic catenary adopted here are somewhat dimensional cable and beam elements can be found in
simpler and therefore easier to handle. This element [25,31].
can be used for modeling large sag cables such as sus- The formulation of cable and beam element matrices
pension bridge main cables, cables in long span cable- will be described later in this paper, where these will
stayed bridges, cables in large cable roofs, etc., where be given in the element local coordinate system. For
straight elements are not readily applicable. each individual element in the model, the evaluated el-
The study presented in this paper emphasizes the ement matrices in the local coordinate system are
modeling aspects of cable supported bridges. Accurate transformed to global coordinate system by the usual
and ecient cable and beam ®nite elements, found in coordinate transformation technique [21]. The struc-
the literature, are adopted for modeling the bridge ture matrices (i.e. the global tangent sti€ness matrix
 R. Karoumi / Computers and Structures 71 (1999) 397±412 399

Fig. 1. Catenary cable element.

Kt, global mass matrix M, and global internal force gravity, Lx is the horizontal projected length of the
vector p) are constructed from the transformed cable, and s the tensile stress in the cable. It can be
matrices of the individual elements of the structure by noticed that increasing the tensile stress in the cable
the general assembly procedure [21]. will, as the sag decreases, lead to an increase in the
apparent axial sti€ness of the cable. It has been shown
in [16] and will be shown in Section 2.1.2 of this paper
2.1. Modeling of cables
that the equivalent modulus approach results in softer
The problem of analyzing cables under di€erent con- cable response as it accounts for the sag e€ect but
®gurations and loading conditions is very complex. does not account for the sti€ening e€ect due to large
This is because stress/strain relationships for cables are displacements. Still, for some cases, e.g. for short span
highly nonlinear and also because large displacements cable-stayed bridges, linear analysis utilizing the equiv-
introduce nonlinearities in the geometric sense. alent modulus approach is often sucient [1,2,14], es-
There are mainly three approaches to the nonlinear pecially in the feasibility design stage, whereas, as
behavior of cable elements. In the ®rst approach each mentioned in the introduction, long span cable-stayed
cable is replaced by one bar element with equivalent bridge structures built today or proposed for future
cable sti€ness. This approach, often adopted when bridges are highly ¯exible, undergo large displace-
modeling cables in cable-stayed bridges, is referred to ments, and should therefore be analyzed taking into
as the equivalent modulus approach and has been used account all sources of geometric nonlinearity.
by several investigators [2,4,11,12,15,18,28,29]. The The second approach is to divide each cable into
equivalent tangent modulus of elasticity, used to take several straight elements, as done in [30], to adequately
account of the sag e€ect, is derived in [1] and can be model the curved geometry of the cable. This would
written as: introduce many added degrees of freedom with a con-
sequent increase in computer storage requirements and
E cable computational cost. In addition, numerical problems
E bar
eq ˆ 1†
rgLx †2 cable can occur and spurious results can be obtained if equi-
1‡ E librium conditions, at those nodes, are not ful®lled.
12s3
The third approach to model cables is to use iso-
where E cable is the cable modulus of elasticity, r is the parametric elements. In [3,16,21], derivations of iso-
density of the cable material, g is the acceleration of parametric cable elements, which include the element
400 R. Karoumi / Computers and Structures 71 (1999) 397±412

curvature, are presented. Using such elements one can because P1, P2, P3, P4, Ti and Tj are related by the fol-
model the curved geometry of a cable with fewer el- lowing equations:
ements compared to using straight elements and obtain
a better convergence [3]. However, those elements are P4 ˆ wLu ÿ P2 ; P3 ˆ ÿP1 5†
sti€er and require numerical integration to formulate
the element sti€ness matrix [3]. q q
The alternative approach presented in this paper is Ti ˆ P 21 ‡ P 22 ; Tj ˆ P 23 ‡ P 24 6†
based on exact analytical expressions for the elastic
catenary. In contrast to other modeling techniques Di€erentiating Eq. (4) and rewriting the results using
mentioned above, each cable may be represented by a matrix notation gives:
single two-node catenary element, which exactly con-
@ Lx @ Lx
siders the curved geometry of the cable, making this dLx ˆ dP1 ‡ dP2 ;
method very attractive for static response calculations. @ P1 @ P2
Even if each cable must be divided into several caten-
ary cable elements, to include cable modes of vibration @ Ly @ Ly
dLy ˆ dP1 ‡ dP2 7†
in the dynamic analysis or external loads acting @ P1 @ P2
between cable ends, the author still believes that this
approach is more ecient to adopt. This is mainly 2 3
because fewer internal nodes need to be de®ned for @ Lx @ Lx
  6 @P 7   
each cable in the model. Thus, the main advantages of dLx 6 1 @ P2 7 dP1 dP1
ˆ6 7 ˆF 8†
the proposed cable element are the reduction of dLy 4 @ Ly @ Ly 5 dP2 dP2
degrees of freedom, the simplicity of ®nding the dead @ P1 @ P2
load geometry of the cable system, the exact treatment
of cable sag, the exact treatment of cable weight as it where F is the incremental ¯exibility matrix. The sti€-
is included in the equations used for element formu- ness matrix K is given by the inverse of F as:
lation, and the simplicity of including the e€ect of pre-  
tension of the cable by simply giving the unstressed k 1 k2
K ˆ Fÿ1 ˆ 9†
cable length. k3 k4

2.1.1. Cable element formulation The tangent sti€ness matrix Kt and the corresponding
The procedure presented in this section determines internal force vector p for the cable element can now
the complete geometry of the cable, the cable element be obtained in terms of the four nodal degrees of free-
internal force vector, and its tangent sti€ness matrix dom as (noting that k2=k3):
from a given unstressed cable length and given pos- 2 3 8 9
ÿk1 ÿk2 k1 k2 >
> P1 >
itions of the ends of the cable. 6 < > =
6 ÿk4 k2 k4 77 P2
Consider an elastic cable element, stretched in the Kt ˆ 4 5 ; pˆ 10†
ÿk1 ÿk2 > P3 >
> >
vertical plane, with an unstressed length Lu, modulus : ;
sym: ÿk4 P4
of elasticity E, cross section area A, and weight per
unit length w, as shown in Fig. 1. The exact relations The element tangent sti€ness matrix Kt relates the
between the element projections and cable force com- incremental element nodal force vector {DP1, DP2,
ponents at the ends of the element are [25]: DP3, DP4}T to the incremental element nodal displace-
  ment vector {Du1, Du2, Du3, Du4}T. The Maple soft-
Lu 1 P4 ‡ Tj
Lx ˆ ÿP1 ‡ ln 2† ware package for symbolic computations [27] was used
EA w Ti ÿ P2 to perform the above mentioned operations and pro-
duce the necessary Fortran code. This Maple pro-
1 Tj ÿ Ti cedure is listed in Appendix A. However, if this
Ly ˆ T 2 ÿ T 2i † ‡ 3† package is not available the following expressions,
2EAw j w
obtained by derivation of Eq. (4), may be used to
where Ti and Tj are the cable tension at the two nodes evaluate the matrices K, Kt, and p:
of the element. For the above expressions it is assumed   
that the cable is perfectly ¯exible and Hooke's law is 1 Lu 1 P4 P2
k1 ˆ ÿ ‡ ‡ 11†
applicable to the cable material. The expressions above det F EA w Tj Ti
for Lx and Ly may be written, in terms of the end
forces P1 and P2 only, as:   
1 P1 1 1
k2 ˆ k3 ˆ ÿ ÿ 12†
Lx ˆ Lx P1 ,P2 †; Ly ˆ Ly P1 ,P2 † 4† det F w Tj Ti
 R. Karoumi / Computers and Structures 71 (1999) 397±412 401

Fig. 2. Cable under its own weight subjected to tensile force at both ends.
  
1 Lx 1 P4 P2
k4 ˆ ‡ ‡ 13† where the sti€ness matrix K is given in Eq. (9) and i is
det F P1 w Tj Ti the iteration number. For the present study, this iter-
ation process continued until DLx and DLy are less
det than 1  10ÿ5. As will be demonstrated later in this
paper, this iterative procedure converges very rapidly.
   
Lu 1 P4 P2 Lx 1 P4 To determine the unstressed cable length, Lu, for
Fˆ ÿ ÿ ‡ ‡
EA w Tj Ti P1 w Tj 14† cases where the initial cable tension is known instead,
   2 a similar iteration procedure can be adopted. A start-
P2 P1 1 1 ing value for the unstressed cable length is assumed,
‡ ÿ ÿ
Ti w Tj Ti e.g. equal to the cable chord length, and cable end
forces P1 and P2 are computed using the iterative pro-
To evaluate the tangent sti€ness matrix Kt, the end cedure described above. Using Eq. (6), cable tension
forces P1 and P2 must be determined ®rst. Those can now be computed. This is then compared with the
forces are adopted as the redundant forces and are given initial tension to obtain a better approximation
determined from given positions of cable end nodes, for Lu for the next iteration step.
using an iterative sti€ness procedure. This procedure If the complete geometry of the cable is to be deter-
requires starting values for the redundant forces. Based mined, coordinates for a number of points along the
on the well-known catenary relationships the following cable must be computed. This is very simple because
expressions will be used for the starting values [25]: P1 and P2 are now known, so Eqs. (2) and (3) can be
  used to compute the coordinates of any new point
wLx w cosh l along the cable by simply replacing Lu by any fraction
P1 ˆ ÿ ; P2 ˆ ÿ Ly ‡ Lu 15†
2l 2 sinh l of Lu.
For the frequency analysis, mass discretization is
where
simply done by static lumping of the element mass at
v
! both ends giving the following lumped mass matrix (r
u
u L2 ÿ L2
t
lˆ 3
u y
ÿ1 16† is the mass density of the cable element):
L2x
2 3
1 0 0 0
In cases where Eq. (16) cannot be used because the rALu 6
60 1 0 077
Mˆ 18†
unstressed cable length is less than the chord length, a 2 40 0 1 05
conservative value of 0.2 for l is assumed [25]. 0 0 0 1
Another diculty arises in Eq. (16) for vertical cables.
In that case an arbitrary large value of 106 for l is
used. Using Eqs. (5) and (6), new cable projections 2.1.2. Analytical veri®cation
corresponding to the assumed end forces P1 and P2 A cable hanging under its own weight and subjected
are now determined directly from Eqs. (2) and (3) and to a tensile force at both ends along its chord, as
the misclosure vector {DLx, DLy }T is evaluated as the shown in Fig. 2, was studied to verify the cable el-
positions of the end nodes are given. Corrections to ement and the analysis procedure described in Section
the assumed end forces can now be made using the 2.1.1. This problem was earlier studied in [16] using
computed misclosure vector as: isoparametric cable elements and the published results
    can now be compared to results obtained here.
DP1 DLx A cable with an unstressed length Lu=312.7 m,
ˆK ;
DP2 DLy modulus of elasticity E = 1.31  1011 N/m2, cross sec-
 i‡1  i   17† tion area A = 5.48  10ÿ4 m2, and weight per unit
P1 P1 DP1 length w = 46.11 N/m, was studied using three di€er-
ˆ ‡
P2 P2 DP2 ent models. For the three models: the cable was
402 R. Karoumi / Computers and Structures 71 (1999) 397±412

Fig. 3. Response of the cable de®ned in Fig. 2.

replaced by one catenary cable element, one bar el- tudinal displacement along the chord of the cable were
ement with an equivalent modulus of elasticity, and by determined for di€erent values of the tensile force T
20 beam elements with negligible moment of inertia. and the results are plotted in Fig. 3. Good agreement
The beam element used is described in the next sec- is observed when comparing the curves, in Fig. 3, for
tion. the proposed catenary cable element with those for the
To span the distance of 304.8 m, using a cable with isoparametric cable element presented in [16]. In Fig.
the above given properties, a horizontal force of 3(a), a signi®cant di€erence can be observed when
To=1.7794  104 N was needed at both ends. This comparing the one bar equivalent modulus curve with
force gave a mid point cable sag of 30.48 m and was the proposed catenary cable element curve. This di€er-
adopted as the initial force when calculating the curves ence is ascribed to the fact that the equivalent modulus
in Fig. 3. approach accounts for the sag e€ect but does not
Using the di€erent models, the sag and the longi- account for the sti€ening e€ect due to large displace-
 R. Karoumi / Computers and Structures 71 (1999) 397±412 403

Fig. 4. Deformed and undeformed con®guration of the beam element.

ments [16]. This leads to a softer cable model. Fig. 3 Referring to Fig. 4, the current deformed con®gur-
shows also that the replacement of the curved cable by ation of the beam axis is described by a regular curve
several beam elements with negligible ¯exural sti€ness de®ned by the position vector:
can give acceptable results. As the number of beam el-
ements increases and the moment of inertia decreases, ro x† ˆ ‰x ‡ u x†Ši ‡ w x†j 19†
the results should converge to those of the catenary
cable element model. One should only bear in mind where the abscissa x $ [0, L ] is measured on the
that replacing the cables by several straight elements straight reference con®guration of the beam, u(x ),
will give a sti€er structural model and consequently an w(x ) represent the axial and transversal displacement
underestimation of the displacements. components, and i and j are unit axis vectors. By intro-
The results from this simple numerical experiment ducing the angle y(x ) as the rotation of the cross sec-
provide con®dence in the application of the proposed tions (S ') in the deformed con®guration, the unit
catenary cable element for nonlinear modeling of cable vectors orthogonal and parallel to the cross sections
supported bridges. for each point on the deformed beam are obtained as:

a x† ˆ cos yi ‡ sin yj; b x† ˆ ÿsin yi ‡ cos yj 20†

2.2. Modeling of bridge deck and pylons
Further, de®ning the deformation measures E, g, k
The pylons and the bridge deckÐgirder or sti€ening according to:
girder as it is also calledÐare modeled using beam el-
ements able to resist bending, shear and axial forces. dro dy
ro,x ˆ ˆ 1 ‡ E†a ‡ gb; kˆ ˆ y,x 21†
For the present study, the simplest 2D beam element dx dx
introduced in [31] is adopted and the treatment given
there will be followed below when deriving the element and using Eqs. (19±20), the following expressions are
matrices. This ®nite element is developed following the obtained:
total Lagrangian approach and using a linear interp-
E ˆ 1 ‡ u,x †cos y ‡ w,x sin y ÿ 1 22†
olation scheme for the displacement components.
Previous studies, reported in [31], have shown that the
element is ecient and accurate. This element is also g ˆ w,x cos y ÿ 1 ‡ u,x †sin y 23†
chosen because it can handle large displacements and
shear deformations and because it is very simple to
formulate the element matrices. k ˆ y,x 24†
404 R. Karoumi / Computers and Structures 71 (1999) 397±412

If the constitutive relations are assumed as linear, the For more details concerning the formulation of the
strain energy can be written as: kinetic energy and mass matrix and the performance
of this element in dynamic problems, the reader is
L
1 referred to [40,41].
Pi u† ˆ EAE2 ‡ GAg2 ‡ EIk2 † d x 25†
2 0

where EA, GA and EI represent the axial, shear and 3. The nonlinear analysis procedure
¯exural rigidities. For u, w and y a linear interpolation
scheme is used according to: For a nonlinear structural system, the sti€ness
u ˆ hi x†qi ‡ hj x†qj 26† matrix Kt is a function of nodal displacements and el-
ement forces, and therefore the matrix Kt should be
where uT={u, w, y }, qTi ={u1, u2, u3} and qTj ={u4, u5, reformulated as the structure deforms. There are sev-
u6} contain the corresponding values of the displace- eral procedures for solving the nonlinear ®nite element
ments at the two nodes of the element, and hi(x ) = equations and the most frequently used are the
1 ÿ x/L; hj(x ) = x/L are the interpolation functions. Newton±Raphson iteration schemes [4,21,32]. In the
Finally, the expressions for the internal force vector present study, an incremental/iterative procedure using
p and the element tangent sti€ness matrix Kt are full Newton±Raphson iterations is adopted. This pro-
obtained through successive di€erentiation of the ex- cedure is generally expected to give quadratic conver-
pression for the strain energy according to: gence [32]. For the frequency analysis, the eigenvalue
problem of the system is solved based on the utiliz-
@ Pi @p @ 2 Pi ation of the tangent sti€ness matrix Kt of the bridge in
pˆ ; Kt ˆ ˆ 27†
@q @q @ q2 the dead load deformed state.
The nonlinear ®nite element code for the static and
where q is the nodal displacement vector {u1, u2, u3, u4, dynamic analysis of cable supported structures has
u5, u6}T. The Maple procedure which performs the been developed using Matlab [33].
above mentioned operations and produces the necess-
ary Fortran code for p and Kt, is listed in Appendix A.
The kinetic energy is expressed as the integral over 4. Numerical examples
the volume V:
Two numerical examples are presented to demon-
1
Pk ˆ rÇr x†T rÇ x†dV 28† strate the applicability of the catenary cable element
2 V for modeling cables in cable supported structures. In
where r is the mass density and Çr(x ) represents the vel- the ®rst example the erection procedure for the Great
ocity in a general point of the beam. The position of Belt suspension bridge girder sections was simulated
this point is de®ned by the vector r(x ), as shown in and in the second example the nonlinear behavior of a
Fig. 4. For this element, the ®nal expression for the medium-long symmetric cable-stayed bridge was stu-
kinetic energy becomes [40]: died. Eigenfrequency analyses were also conducted for
L L
both examples. To illustrate the eciency of the pre-
r r sented method, the results obtained are compared to
Pk ˆ Au_ x†2 dx ‡ Aw_ x†2
2 0 2 0 those reported by other researchers.
29†
L
r
dx ‡ Iy_ x†2 dx 4.1. Erection of the Great Belt suspension bridge
2 0

Using the interpolation functions, as in Eq. (26), the In this numerical example, the behavior of the Great
kinetic energy is written as a function of the velocity Belt suspension bridge during girder erection was stu-
components in the nodal degrees of freedom of the el- died and a frequency analysis for the completed bridge
ement. From the resulting expression for the kinetic was also conducted using the cable and beam elements
energy, the consistent element mass matrix is evaluated described earlier.
as: The Great Belt (Storebñlt) suspension bridge in
2 3 Denmark has, with its 1624 m main span, the second
2A 0 0 A 0 0 largest span in the world (Akashi Kaikyo suspension
6 0 2A 0 0 A 0 7 bridge in Japan holds currently the world record with
6 7
@ 2 Pk rL 6
6 0 0 2I 0 0 I 77 a main span length of 1990 m. Both suspension bridges
Mˆ ˆ 30†
@ qÇ 2 6 6
6 A 0 0 2A 0 0 77 were opened for trac in 1998). The bridge deck is a
4 0 A 0 0 2A 0 5 31 m wide and 4 m deep steel box. This girder is con-
0 0 I 0 0 2I tinuous over the full cable supported length of 2694 m,
 R. Karoumi / Computers and Structures 71 (1999) 397±412 405

Fig. 5. Geometry of the Great Belt suspension bridge.

i.e. without expansion joints at the pylons. Like many model consisting of 126 elements, 89 nodes and a total
other modern suspension bridges, the girder section is of 215 active degrees of freedom, was used. The bridge
streamlined (almost wing shaped) to resist strong wind geometry shown in Fig. 5 and the properties given in
action. The two main cables, each including a total of Table 1 were taken from [34,37]. The section properties
18,648 high strength galvanized steel wires 5.38 mm in for the pylons were assumed by the author to give the
diameter with a ®nal diameter of 0.85 m, were erected ®rst and second pylon frequencies 0.147 and 0.803 Hz,
in the autumn of 1996 by the air-spinning method [1]. which are acceptable when compared to the measured
Much the same method was used for suspension pylon frequencies reported in [38]. In the bridge model,
bridges more than 100 years ago, e.g. for the Brooklyn the girder was assumed to be pinned at the ends, i.e.
Bridge in New York dating from 1883. To minimize only rotations were allowed, and every third hanger
de¯ections under asymmetric trac load, the main from the original bridge was included and modeled
cables are ®xed to the sti€ening girder through rigid using catenary cable elements. The main cables were
clamps at mid span. The ratio of cable sag to main assumed to be ®xed at the pylon tops and the pylons
span length was chosen to be 1/9, as this was found to were rigidly ®xed to the piers. Further, the shear mod-
be economically favorable. This gives higher pylons ulus of all beam elements in the model was evaluated
than usual and therefore a more ¯exible structure. The as G = E/2(1 + n ); n = 0.3. To simplify the data input
total height of the concrete pylons, including cable and process, all internal cable nodes were vertically posi-
cable saddle, is approximately 258 m. tioned at the same level as the girder, and the con®gur-
Erection of a suspension bridge involves many chal- ation of the main cable under dead load was
lenging problems, especially aerodynamic stability pro- determined accurately, during analysis, after few iter-
blems, which relate to the fact that the bridge structure ations.
is incomplete, thus various structural components do To start the analysis, the initial side span and main
not receive or render the kind of support intended in span cable lengths were determined using a trial and
the complete structure. The erection of the girder may error procedure. Thus the initial cable lengths were
proceed in a number of di€erent ways [34±36]. For the estimated and the ®nal dead load pro®le, i.e. cable sag
analysis presented in this paper it was assumed that at mid span and pylon top displacements, were deter-
the erection proceeded simultaneously from mid span mined and compared with the desired dead load pro®le
and anchor blocks towards the pylons. To study the shown in Fig. 5. The estimated values were then
erection of the Great Belt suspension bridge and calcu- improved and the calculation was repeated until the
late the initial pro®le for the free hanging cables, i.e. required dead load pro®le was obtained. It was found
the initial cable lengths, the cable sag at mid span, and that if the initial lengths of the side span and main
the pylon tops horizontal displacement, a simple bridge span cables were chosen as 564.8 and 1672.7 m, the

Table 1
Parameters for the model of the Great Belt suspension bridge

E (N/m2) A (m2) I (m4) w (t/m)

Girder (31  4 m) 2.1  1011 1.00 3.32 14.78

Pylons, 0±75.5 m 0.4  1011 2  37.5 2  750 2  90
Pylons, 75.5±257.6 m 0.4  1011 2  (32.5, 30, 25) 2  (275, 200, 150) 2  (78, 72, 60)
Cable side spans 2.1  1011 2  0.41 ± 2  3.45
Cable main span 2.1  1011 2  0.40 ± 2  3.36
Hangers 2.1  1011 2  0.025 ± a

a
Mass of hangers and clamps is considered distributed uniformly along the main cable and included in the cable mass.
406 R. Karoumi / Computers and Structures 71 (1999) 397±412

Fig. 6. Cable sag variation (a) and horizontal displacement of pylon tops (b) during erection of the bridge girder.

calculated mid span cable sag and pylon tops horizon- pensate for the later erection of the girder, the saddles
tal displacements would be 180.09 and 0.04 m. Those at the pylon tops must therefore, prior to cable erec-
initial cable lengths were therefore considered to be tion, be horizontally displaced in relation to the verti-
good enough and were used for all the following cal pylon axis. This is done either by displacing the
results presented for this example. saddles in relation to the pylon tops or by pulling back
When the initial cable lengths were known, di€erent the pylons with so called tie-back cables or a combi-
erection stages were analyzed and the results are nation of the two methods [36]. It is therefore necess-
plotted in Fig. 6. Studying this ®gure one can notice ary to specify this displacement of the saddles to arrive
that, to arrive at the desired dead load pro®le, the at vertical pylons with zero bending in the ®nal dead
pylon tops must be displaced about 0.85 m outwards load condition. During the constructing of the Great
and the cable sag should be about 173 m, for 0% Belt suspension bridge, tie-back cables running from
erected girder. This pylon top displacement is needed each pylon top to the nearest anchor block were used,
to counteract the displacement caused by the pulling back each pylon 1.24 m [37] prior to main
elongation of the side span main cable during the sub- cable erection. When the main cables were erected and
sequent erection of the girder. Consequently, to com- the tie-back cables were dismantled, the pylon tops
 R. Karoumi / Computers and Structures 71 (1999) 397±412 407

Fig. 7. Natural frequencies and mode shapes for the lowest three vertical bending modes of vibration. Values inside brackets are
reported in [35].

moved back about 0.20 m [37]. Thus, the remaining model, made for wind tunnel testing, are presented.
displacement before erecting the girder was about 1 m. These results are also given in Fig. 7 within brackets
Acceptable agreement is found, according to the and, as can be seen, the agreement is very good when
author's opinion, when comparing this value with the comparing with the result obtained from the present
one obtained from the present analysis, i.e. 0.85 m. It analysis.
is worth noting that the aim of this study and the The CPU time used by the MATLAB process, to
essential goal of this example was to check the e- ®nd the tangent sti€ness matrix at the dead load
ciency and applicability of the presented elements for deformed state and to solve the system eigenvalue pro-
modeling cable supported bridges. For this reason blem determining all 215 modes of vibration, was
some simpli®cations and assumptions were made when about 200 s. The calculations were performed on a
modeling the bridge structure and not much emphasis Pentium Pro 200 MHz system.
was put on using the exact properties etc. for each
bridge member. This is believed to be the major expla- 4.2. Nonlinear behavior of a fan-shaped cable-stayed
nation for the di€erences in the results. bridge
Frequency analysis was also conducted for the com-
pleted bridge and the ®rst three natural frequencies A 2D model of the cable-stayed bridge described in
and mode shapes are given in Fig. 7. In [34,35], ana- [4] was adopted for this investigation. The bridge geo-
lytical frequency results from a 3D ®nite element metry is shown in Fig. 8 and the properties are given
model and results from a 1:200 scale aeroelastic bridge in Table 2. This bridge is similar in con®guration to an

Fig. 8. Geometry of the cable-stayed bridge studied in example 2.

408 R. Karoumi / Computers and Structures 71 (1999) 397±412

Table 2
Parameters for the cable-stayed bridge model de®ned in Fig. 8

E (N/m2) A (m2) I (m4) w (t/m)

Girder 2.0  1011 0.93 0.26 19.64a

Girder, central part 2.0  1011 1.11 1.29 19.64a
Pylons above deck level 2.8  1010 13.01 34.52 30.65
Pylons below deck level 2.8  1010 18.58 86.31 43.78
Links, deck to pylons 2.0  1011 0.56 0.10 4.38

Cable no. E (N/m2) A (m2) Lu (m) w (t/m)

1, 24 2.0  1011 0.0362 158.13 0.398

2, 11, 14, 23 2.0  1011 0.0232 134.66 0.255
3, 10, 15, 22 2.0  1011 0.0204 111.64 0.225
4, 9, 16, 21 2.0  1011 0.0176 89.43 0.194
5, 8, 17, 20 2.0  1011 0.0139 68.80 0.153
6, 7, 18, 19 2.0  1011 0.0113 51.69 0.125
12, 13 2.0  1011 0.0372 158.12 0.409

a
Including weight of cross beams.

existing bridge in Japan (The Meiko-Nishi Bridge in in the design process. However, as the span length
Nagoya) with a few modi®cations in dimension. The increases this nonlinearity will get more pronounced [4]
static and dynamic behavior of this bridge has been and linear analysis will no longer be adequate.
studied earlier by several other investigators [4,13,39]. For the frequency analysis, to include cable motions,
For the model, it was assumed that the girder was natural frequencies were also determined replacing
pinned at the ends, i.e. only rotations were allowed, each stay cable by 3, 5 and 7 catenary cable elements.
and it was connected to the pylons by vertical links. This was easily done as the preprocessor code devel-
The pylons were assumed to be rigidly ®xed to the oped can automatically re®ne the model if it is
piers and all cables were assumed ®xed to the pylons requested by the user. For the simplest model, i.e. the
and to the girder at their joints of attachment. model with one element per cable, the resulting modes
Further, the shear modulus of all beam elements in the of vibration only include the vibrating girder and
model was evaluated as G = E/2(1 + n ); n = 0.3. The pylons. Thus, cable modes and the dynamic interaction
simplest model analyzed in this example, i.e. the model between the vibrating cables and the bridge were disre-
with one element per cable, was composed of 66 el- garded. For the ®ner models, pure cable modes, i.e.
ements and 43 nodal points. additional new mode shapes characterized only by
Fig. 9 shows the nonlinear behavior of the model vibrating cables, were obtained between those basic
under static dead load, described in terms of the verti- bending modes. Moreover, it is evident that cable
cal displacement at the center of the bridge girder and motions are associated with every mode of vibration,
the tension in cable 12 and 13. Examining this ®gure, a as can be noticed in Fig. 10. For the four alternative
hardening characteristic with respect to the applied models, Table 3 presents a comparative frequency
load is apparent. It is also evident that at the start study of the ®rst 10 vertical bending modes of vi-
there is a signi®cant nonlinear behavior during the sta- bration. The order in which these modes appear is
tic application of the dead load. Thus, a nonlinear sta- given inside brackets.
tic analysis under dead load is essential to arrive at the Satisfactory agreement is found when comparing the
deformed dead load tangent sti€ness matrix. To the results from the static and frequency analysis presented
author's knowledge, this highly nonlinear ®rst part of here with those reported in [4,13,39]. The disagreement
the curve in Fig. 9a has not been presented in any of in the frequency results shown in Fig. 10 is believed to
the previous studies. For this cable-stayed bridge with be due to the fact that the catenary cable element used
modest main span length, as the nonlinearity is not so in the present study is sti€er, see Fig. 3a, than the one
strong above this dead load equilibrium point, one can bar element with an equivalent modulus used in [4,39].
assume that this bridge behaves as a linear system, Moreover, the girder and pylons were modeled in
when a€ected by live static and dynamic loads, starting [4,39] using conventional beam elements modi®ed by
from this dead load deformed state. This means that the stability functions and a diagonal lumped mass
in¯uence lines and superposition technique can be used matrix was adopted for all elements.
 R. Karoumi / Computers and Structures 71 (1999) 397±412 409

Fig. 9. Nonlinear behavior of the cable-stayed bridge de®ned by Fig. 8: (a) vertical displacement; (b) cable tension.

For the simplest model with a total of 119 active study has shown that the adopted elements are accu-
degrees of freedom, the CPU time used by the rate and ecient for nonlinear analysis of cable-stayed
MATLAB process, to ®nd the tangent sti€ness matrix and suspension bridges. It has been established that
at the dead load deformed state and to solve the sys- the main advantages of the cable element are the sim-
tem eigenvalue problem determining all 119 modes of plicity of including the e€ect of pretension of the cable
vibration, was about 15 s. The calculations were per- and the exact treatment of cable sag and cable weight.
formed on a Pentium Pro 200 MHz system. Moreover, the iterative process adopted, to ®nd the in-
ternal force vector and tangent sti€ness matrix for the
cable element, was found to converge very rapidly.
5. Conclusions According to the author's opinion, linear analysis
utilizing the traditional equivalent modulus approach
The paper has presented a method for modeling is not satisfactory for modern cable-stayed bridges.
cable supported bridges for nonlinear ®nite element Modern long span cable-stayed bridges built today or
analysis. A two-node catenary cable element was proposed for future bridges are, as they are highly ¯ex-
adopted for modeling the cables and a beam element ible, subjected to large displacements. The equivalent
for modeling the bridge deck and the pylons. This modulus approach however accounts only for the sag
410 R. Karoumi / Computers and Structures 71 (1999) 397±412

Fig. 10. Natural frequencies and mode shapes for the lowest three vertical bending modes of vibration. Values inside brackets are
reported in [39].

e€ect and not for the sti€ening e€ect due to large dis- commercial codes, e.g. ABAQUS, enable the users to
placements. It is concluded that the catenary cable el- de®ne their own elements. This disadvantage applies
ement is simple to formulate, accurate, and can also to the one bar element equivalent modulus
correctly model the geometric change of the stay cable approach.
at any tension level. This makes the element very It is apparent that cable supported bridges have a
attractive, especially for static response calculations, hardening characteristic with respect to the applied
and the author strongly recommends the use of this el- load. Furthermore, the highly nonlinear behavior
ement. However, one drawback is when using commer- during the static application of the dead load requires
cial ®nite element codes for analysis, as only few a nonlinear static analysis to arrive at the deformed

Table 3
Comparison of the ®rst 10 natural frequencies for vertical bending modes of vibration. Each cable was modeled using 1, 3, 5 and 7
catenary cable elements

Natural frequencies (Hz) and mode order

Vertical bending mode no. 1 element/cable 3 elements/cable 5 elements/cable 7 elements/cable

1 0.332 (1) 0.334 (1) 0.334 (1) 0.334 (1)

2 0.436 (2) 0.437 (2) 0.437 (2) 0.437 (2)
3 0.692 (3) 0.700 (7) 0.702 (7) 0.703 (7)
4 0.734 (4) 0.739 (8) 0.741 (8) 0.741 (8)
5 0.867 (5) 0.873 (13) 0.874 (13) 0.874 (13)
6 1.044 (6) 1.051 (18) 1.052 (18) 1.053 (18)
7 1.211 (7) 1.217 (27) 1.219 (23) 1.219 (23)
8 1.213 (8) 1.219 (28) 1.221 (24) 1.221 (24)
9 1.378 (9) 1.387 (33) 1.385 (29) 1.387 (29)
10 1.669 (10) 1.687 (42) 1.679 (38) 1.685 (38)
 R. Karoumi / Computers and Structures 71 (1999) 397±412 411

dead load tangent sti€ness matrix. Starting from this with(linalg);

dead load deformed state, a linear static and dynamic Ly: = 1/(2EAw)(Tj U 2 ÿ Ti U 2) + (Tj ÿ Ti)/w;
analysis can sometimes be sucient for short and med- Lx: = ÿ P1(Lu/E/A + 1/wln((P4 + Tj)/(Ti ÿ P2)));
ium span cable supported bridges [11,39]. P3: = ÿ P1;
Replacing each cable by several catenary cable el- P4: = wLu ÿ P2;
ements has demonstrated that, in addition to obtaining Ti: = sqrt(P1 U 2 + P2 U 2);
new pure cable modes of vibration, cable motions are Tj: = sqrt(P3 U 2 + P4 U 2);
also associated with every bending mode of vibration. f11: = di€(Lx,P1):
For dynamic analysis, this multi-element cable discreti- f12: = di€(Lx,P2):
zation should always be utilized, instead of the one-el- f21: = di€(Ly,P1):
ement cable discretization. This is essential in order to f22: = di€(Ly,P2):
include the dynamic interaction between the vibrating f: = matrix(2,2,[f11,f12,f21,f22]):
cables and the bridge and to realistically predict the re- k: = inverse(f):
sponse to dynamic loads such as earthquake, wind and k1: = k[1,1]:
trac. To simplify the data input process when utiliz- k2: = k[1,2]:
ing the multi-element cable discretization, one can start k4: = k[2,2]:
from a straight cable con®guration and during analysis Kt: = matrix(4,4,[ÿk1, ÿ k2,k1,k2, ÿ k2, ÿ k4,k2,k4,
the cable con®guration under its own weight is deter- k1,k2, ÿ k1, ÿ k2,k2,k4, ÿ k2, ÿ k4]):
mined accurately after few iterations. fortran(Kt,optimized):
Finally, this work has only focused on two-dimen-
sional modeling of cable supported bridges. However, A.0.2. Beam element
the catenary cable element used in this study is also # Internal force vector p and;
applicable for modeling cables in other types of cable # tanget sti€ness matrix Kt for;
structures [24±26] such as: suspended roofs, guyed # the beam element;
masts, electric transmission lines, moored drilling plat- readlib(fortran);
forms, etc. Moreover, with some minor modi®cations with(linalg);
of the cable element matrices this element can also be ux: = (u4 ÿ u1)/L; wx: = (u5 ÿ u2)/L;
used for modeling cables for three-dimensional analy- t: = (u3 + u6)/2; tx: = (u6 ÿ u3)/L;
sis. For such analysis, three-dimensional catenary cable e: = (1 + ux)cos(t) + wxsin(t) ÿ 1;
element and beam elements can be found in [25,31]. g: = wxcos(t) ÿ (1 + ux)sin(t);
k: = tx;
Fie: = 1/2LEAe U 2;
Acknowledgements Fig: = 1/2LGAg U 2;
Fik: = 1/2LEIk U 2;
The author wishes to acknowledge Professor HaÊkan Fi: = Fie + Fig + Fik;
Sundquist and Dr Costin C. Pacoste at the Royal p: = grad(Fi,[u1,u2,u3,u4,u5,u6]);
Institute of Technology in Stockholm for many helpful Kt: = hessian(Fi,[u1,u2,u3,u4,u5,u6]);
discussions. The author also wishes to thank Professor fortran(p,optimized);
Niels J. Gimsing at the Technical University of fortran(Kt,optimized)
Denmark, Professor Hiroshi Tanaka at the Danish
Maritime Institute, and Mr Sùren Esdahl at
COWIconsult A/S, Denmark for their help on provid- References
ing information on the Great Belt suspension bridge.
[1] Gimsing NJ. Cable supported bridges, second.
Chichester: John Wiley, 1997.
[2] Karoumi R. Dynamic response of cable-stayed bridges
Appendix A subjected to moving vehicles. Licentiate Thesis, TRITA-
BKN Bulletin 22, Department of Structural Engineering,
The Maple procedures used to generate the Fortran Royal Institute of Technology, Stockholm 1996.
code for the elements presented in Sections 2.1 and [3] Leonard JW. Tension structures. New York: McGraw-
2.2, are given below. Hill, 1988.
[4] Nazmy AS, Abdel-Gha€ar AM. Three-dimensional non-
A.0.1. Cable element linear static analysis of cable-stayed bridges. Comput
# Tanget sti€ness matrix Kt for; Struct 1990;34:257±71.
# the catenary cable element; [5] Irvine HM. Cable structures. Cambridge, MA: MIT
readlib(fortran); Press, 1992.
412 R. Karoumi / Computers and Structures 71 (1999) 397±412

[6] Judd BJ, Wheen RJ. Nonlinear cable behavior. J Struct [26] Peyrot AH, Goulois AM. Analysis of ¯exible trans-
Div ASCE 1978;104:567±75. mission lines. J Struct Div ASCE 1978;104:763±79.
[7] Dean DL. Static and dynamic analysis of guy cables. J [27] Anon.MAPLE V language reference manual. New York:
Struct Div ASCE 1961;87:1±21. Springer, 1991.
[8] Huddleston JV. Computer analysis of extensible cables. J [28] Karoumi R. Dynamic response of cable-stayed bridges
Engng Mech Div ASCE 1981;107:27±37. subjected to moving vehicles. In: IABSE 15th Congress,
[9] Huddleston JV, Ham HJ. Poisson e€ect in extensible Denmark, 1996. p. 87±92.
cables with both ends ®xed. J Engng Mech Div ASCE [29] Walther R, Houriet B, Isler W, MoõÈ a P. Cable stayed
1994;120:1590±5. bridges. London: Thomas Telford, 1988.
[10] Buchholdt HA. An introduction to cable roof structures. [30] Abdel-Gha€ar AM, Khalifa MA. Importance of cable
Cambridge: Cambridge University Press, 1985. vibration in dynamics of cable-stayed bridges. J Engng
[11] Fleming JF, Egeseli EA. Dynamic behaviour of a cable- Mech ASCE 1991;117:2571±89.
stayed bridge. Earthqu Engng Struct Dyn 1980;8:1±16. [31] Pacoste C, Eriksson A. Beam elements in instability pro-
[12] Adeli H, Zhang J. Fully nonlinear analysis of composite blems. Comput Meth Appl Mech Engng 1997;144:163±
girder cable-stayed bridges. Comput Struct 1995;54:267± 97.
77. [32] Cris®eld MA. Non-linear ®nite element analysis of solids
[13] Kanok-Nukulchai W, Hong G. Nonlinear modelling of and structures. Chichester: Wiley, 1991.
cable-stayed bridges. J Construct Steel Res 1993;26:249± [33] Anon.MATLAB Using Matlab version 5. Natick:
66. MathWorks, 1996.
[14] Kanok-Nukulchai W, Yiu PKA, Brotton DM. [34] Larsen A, Jacobsen AS. Aerodynamic design of the
Mathamatical modelling of cable-stayed bridges. Struct Great Belt East Bridge. In: Proceedings of 1st
Engng Int 1992;2:108±13. International Symposium on Aerodynamics of Large
[15] Bruno D, Grimaldi A. Nonlinear behaviour of long-span Bridges, Copenhagen, Denmark, 1992. p. 269±83.
cable-stayed bridges. Meccanica 1985;20:303±13. [35] Larsen A. Aerodynamic aspects of the ®nal design of the
[16] Ali HM, Abdel-Gha€ar AM. Modeling the nonlinear 1624 m suspension bridge across the Great Belt. J Wind
seismic behavior of cable-stayed bridges with passive Engng Industr Aerodyn 1993;48:261±85.
control bearings. Comput Struct 1995;54:461±92. [36] Jensen G, Petersen A. Erection of suspension bridges. In:
[17] Abbas S, Scordelis A. Nonlinear analysis of cable-stayed Proceedings of International Conference on Cable-Stayed
bridges. In: Proceedings of the International Conference and Suspension Bridges. Deauville, France, 1994. Vol. 2.
on Cable-Stayed and Suspension Bridges. Deauville, p. 351±362.
France, 1994. Vol. 2. p. 195±210. [37] Gimsing NJ (Technical University of Denmark), Tanaka
[18] Boonyapinyo V, Yamada H, Miyata T. Wind-induced H (Danish Maritime Institute), Esdahl S (COWI consult
nonlinear lateral-torsional buckling of cable-stayed A/S). Personal communication, 1997.
bridges. J Struct Engng ASCE 1994;120:486±506. [38] Larose GL, Zasso A, Melelli S, Casanova D. Field
[19] Gambhir ML, Batchelor B. A ®nite element for 3-D pre- measurements of the wind-induced response of a 254 m
stressed cablenets. Int J Numer Meth Engng high free-standing bridge pylon. In: Proceedings of the
1977;11:1699±718. 2nd European African Conference on Wind Engineering.
[20] Ozdemir H. A ®nite element approach for cable pro- Italy: Genova, 1997. p. 1±8.
blems. Int J Solids Structures 1979;15:427±37. [39] Abdel-Gha€ar AM, Nazmy AS. 3-D nonlinear seismic
[21] Bathe KJ. Finite element procedures. Englewood Cli€s, behavior of cable-stayed bridges. J Struct Engng ASCE
New Jersey: Prentice-Hall, 1996. 1991;117:3456±76.
[22] O'Brien T, Francis AJ. Cable movements under two- [40] Forsell K. Dynamic analyses of static instability phenom-
dimensional loads. J Struct Div ASCE 1964;90:89±123. ena. Licentiate Thesis, TRITA-BKN Bulletin 34,
[23] O'Brien T. General solution of suspended cable pro- Department of Structural Engineering, Royal Institute of
blems. J Struct Div ASCE 1967;93:1±26. Technology, Stockholm 1997.
[24] Peyrot AH, Goulois AM. Analysis of cable structures. [41] Karoumi R. Vibration of cable supported bridges under
Comput Struct 1979;10:805±13. the action of moving trac. Doctoral Thesis,
[25] Jayaraman HB, Knudson WC. A curved element for the Department of Structural Engineering, Royal Institute of
analysis of cable structures. Comput Struct 1981;14:325± Technology, Stockholm in press.
33.