Engineering Structures 31 (2009) 19031915

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Coupled lateraltorsional frequencies of asymmetric, three-dimensional

structures comprising shear-wall and core assemblies with stepwise
variable cross-section
B. Rafezy a , W.P. Howson b,
a

Sahand University of Technology, PO. Box 51335/1996, Tabriz, Iran

Cardiff School of Engineering, Cardiff University, The Parade, Cardiff CF24 3AA, UK

article

info

Article history:
Received 14 April 2008
Received in revised form
5 September 2008
Accepted 16 January 2009
Available online 11 April 2009
Keywords:
Thin-walled structures
Asymmetric structures
Torsional vibration
Coupling
Wall-core structures
Continuum mechanics
Three-dimensional models

abstract
A simple and accurate model for asymmetric, three-dimensional wall-core structures is developed that
enables any desired natural frequency to be determined by a method which guarantees that no natural
frequencies can be missed. The model assumes that the primary walls and cores run in two orthogonal
directions and that their properties may vary in a stepwise fashion at one or more storey levels. A vectorial
approach is used to generate the governing differential equations for coupled flexural-torsional motion
that are finally incorporated into an exact dynamic stiffness matrix (exact finite element) that can model
any uniform segment of the structure. A model of the original structure can then be assembled in the
usual way. Since the mass of each segment is assumed to be uniformly distributed, it is necessary to solve a
transcendental eigenvalue problem, which is accomplished using the WittrickWilliams algorithm. When
the structure can be represented realistically by a uniform cantilever, solutions can be found easily, by
hand. A parametric study comprising five, three-dimensional, asymmetric wall-core structures is given
to compare the accuracy of the current approach with that of a full finite element analysis.
2009 Elsevier Ltd. All rights reserved.

1. Introduction
The type of mathematical model developed herein can be
classified as a simplified global model. This implies that the original
structure is treated holistically, but simplified prior to analysis so
that only its dominant characteristics are retained. The resulting
model is solved exactly so that no additional accuracy is lost
in solution. The use of such models can be quite compelling in
appropriate circumstances [1], such as preliminary design, when
the concept may be evolving rapidly, or when it is necessary
to check solutions developed elsewhere. Moreover, the following
review of related work serves to highlight the growing popularity
of such techniques.
Approximate methods have recently been developed that
can deal with the vibration of asymmetric three-dimensional
structures, in which the translational and torsional modes of
vibration are coupled. Kuang and Ng [2,3] considered the problem
of doubly asymmetric, proportional structures in which the motion
is dominated by shear walls. For the analysis, the structure is

Corresponding author. Tel.: +44 0 2920 874263; fax: +44 0 2920 874597.
E-mail addresses: rafezyb@sut.ac.ir, rafezy@yahoo.co.uk (B. Rafezy),
howson@cf.ac.uk (W.P. Howson).
0141-0296/$ see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2009.01.024

replaced by an equivalent uniform cantilever whose deformation

is coupled in flexure and warping torsion. The same authors
extended this concept to the case of wall-frame structures by
allowing for bending and shear. In this case however, the wall
and frame systems are independently proportional, but result in a
non-proportional structural form [4]. In a recent publication, they
have extended their work to tall building structures comprising
frames, walls, structural cores and coupled walls [5]. As in their
previous work, they have replaced the structure with a uniform
cantilever, derived the governing differential equation for free
vibration and then solved the corresponding eigenvalue problem
using a generalised method based on the Galerkin technique. Wallframe structures have also been addressed by Wang et al. [6],
who used an equivalent eccentricity technique that is appropriate
for non-proportional structures. However, the analysis is limited
to finding the first two coupled natural frequencies of uniform
structures with singly asymmetric plan form.
Hand methods have also received considerable attention and
are particularly suitable for check calculations. In recent papers
by Zalka [7,8], such a method is presented which can deal with
the three-dimensional frequency analysis of buildings braced by
frameworks, coupled shear-walls and cores.
In a relatively recent publication, Potzta and Kollar [9] replaced
the original structure by an equivalent sandwich beam that can

1904

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

model both slender and wide structures consisting of frames,

trusses and coupled shear walls. In a subsequent paper, an
alternative approach is adopted in which the natural frequencies
of the replacement beam are solved approximately. This, together
with other simplifying assumptions, leads to simple formulae for
determining the required natural frequencies [10]. Reference [10]
also includes a useful tabulated summary of related work by the
following authors [7,1118].
The most recent contribution has been made by Rafezy
et al. [19] who presented a simple, accurate model for the calculation of natural frequencies of asymmetric, three dimensional
frame structures whose properties may vary through the height of
the structure in a stepwise fashion at one or more storey levels.
Their stiffness formulation enables the structure to be modelled as
a stepped shear-torsion cantilever which yields the lower natural
frequencies for medium to tall structures surprisingly accurately.
The methods developed in the references above offer solutions
of varying accuracy, depending on the assumptions employed.
Surprisingly, apart from the latter paper, none of them allows
for step changes of properties along the height of the structure,
despite the fact that this is almost inevitably the case in practical
building structures of reasonable height. This study therefore seeks
to extend the concept of the paper by Rafezy et al. [19] to wall-core
structures.
2. Problem statement
The class of building structure considered herein comprises two
sets of orthogonal shear walls that are additionally stiffened by
cores whose principal axes run parallel to the same orthogonal
directions. Since walls and cores deflect predominantly in a
flexural configuration, it is assumed that they obey BernoulliEuler
bending theory that allows for bending deformation but not
shearing deformation. In the case of an asymmetric arrangement
of walls and cores, torsional effects are produced and may become
significant or even critical in tall buildings. In addition, walls
and cores in buildings do not warp freely as they are restrained
against warping at foundation level, thus the effect of warping
should be taken into account in addition to St. Venant torsion. The
warping-restrained torsion is often referred to as Vlasovs torsion
and can lead to longitudinal stresses in the walls and cores that
are sometimes greater than longitudinal stresses due to overall
bending of the structure. In this study both the St. Venant and
warping rigidity of the cores are taken into account, but the St.
Venant torsional component of the walls, which is dependent on
circular shear flows within individual wall elements, is small in
comparison and can therefore be ignored.
The underlying approach adopted with the model is to dissect
the original building structure into segments, by cutting through
the structure horizontally at those storey levels corresponding to
changes in storey properties. Thus the storeys contained within
a segment between any two adjacent cut planes are identical. A
typical segment is then considered in isolation. Initially, one of the
two orthogonal plan directions is selected. If there is a primary
wall running in this direction it is replaced by an equivalent
flexural beam. If there is a core that has a component of flexure
in this direction, it is replaced by an equivalent flexuretorsion
beam located on the cores original shear centre. In each case, the
substitute members have uniformly distributed mass and stiffness.
The flexural beam only allows for bending deformation, while the
flexuretorsion beam allows for bending deformation, St. Venant
and warping torsion. In turn, each additional wall or core that
contributes to the structural action in the current direction is
replaced by its own substitute beam and the effect of all these
beams is summed to model the effect of the original structure. This
leads directly to the differential equation governing the motion

Fig. 1. Typical floor plan of an asymmetric three-dimensional wall-core structure.

O and C denote the locations of the rigidity and mass centres, respectively. The floor
system EFGH is considered to be rigid in its plane.

of the segment in the chosen direction. The same procedure is

then adopted for all the cores and those walls that contribute
to the structural action in the orthogonal direction. Once both
equations are available it requires little effort to write down the
substitute expressions for the coupled torsional motion. The three
equations thus formed are subsequently solved exactly and posed
in dynamic stiffness form. The resulting coupled flexuretorsion
beam element can then be used to reconstitute the original
structure by assembling the dynamic stiffness matrices for the
individual segments in the usual manner.
It is clear from the element formulation that the final model
has a transcendental dependence upon the frequency parameter.
The required natural frequencies are therefore determined by
solving the model using an exact technique, based on the
WittrickWilliams algorithm, that can be arrested after achieving
any desired accuracy and which also ensures that no natural
frequencies can be missed.
3. Theory
Consider the hypothetical layout of a typical floor plan of
the asymmetric, three-dimensional wall-core structure shown in
Fig. 1. The walls run in two orthogonal directions and the cores
principal axes are parallel to these directions. It is assumed that
the rigidity centre of the structure, O, at each floor level lies on a
vertical line that runs through the height of the structure.
It is now assumed that the origin of the co-ordinate system
is located at the rigidity centre, O, with the x and y co-ordinates
running parallel to the walls. The z-axis then runs vertically from
the base of the building and coincides with the rigidity axis. Point
C (xc , yc ) denotes the centre of mass at a typical floor level. It is
assumed that the floor system is rigid in its plane and that the
centre of mass at each level lies on a vertical line, the mass axis,
that runs through the height of the structure. When the rigidity and
mass axes of a structure do not coincide, the lateral and torsional
motion of the building will always be coupled in one or more
planes.
During vibration, the displacement of the mass centre at any
time t in the xy plane can be determined as the result of a pure
translation followed by a pure rotation about the rigidity centre,
see Fig. 2. During the translation phase the rigidity centre moves
to O0 and the mass centre C moves to C 0 , displacements in each
case of u(z , t ) and v(z , t ) in the x and y directions, respectively.
During rotation, the mass centre moves additionally from C 0 to C 00 ,
an angular rotation of (z , t ) about O0 . The resulting translations,
(uc , vc ) of the mass centre in the x and y directions, respectively,
are
uc (z , t ) = u(z , t ) yc (z , t ) and

(1a)

vc (z , t ) = v(z , t ) + xc (z , t ).

(1b)

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

Fig. 2. Coupled translational-torsional vibration of the structure. O and C move to

O0 and C 0 , respectively, during translation and C 0 moves additionally to C 00 during
rotation about O0 .

1905

Fig. 4. Co-ordinate system and positive sign convention for the substitute
two-dimensional flexural beam in the local yz plane. (a) Member convention.
(b) Element convention.

and time, respectively. Likewise, we consider a typical core, core i,

and replace it by a single flexuretorsion substitute beam, beam
i, shown in Fig. 5, which is then located at the shear centre of
the former core. This beam is a three-dimensional flexuretorsion
beam of length L and has uniformly distributed mass, flexural and
torsional stiffness. The elastic axis therefore coincides with the
local z-axis and is permitted bending deformation v i (z , t ) in the y
direction and torsional rotation (z , t ). The mass axis runs parallel
to the elastic axis and is not generally coincident with the elastic
axis. The distances of the local mass axis of core i from its own
elastic axis are denoted by exi and eyi in the x and y directions,
respectively.
The equations of motion for the substitute flexural beam can be
developed from Fig. 4 by considering a typical elemental length of
the beam, z, and employing DAlemberts principle. The dynamic
equilibrium equations for the motion of the element in the yz
plane may then be written as

Fig. 3. Typical segment formed by cutting the structure through planes Ek Fk Gk Hk

and Ek+1 Fk+1 Gk+1 Hk+1 that correspond to the kth and k + 1th changes in storey
properties. (Some lines have been omitted/included for clarity.).

More generally, it is clear that the displacements of a typical point

(xi , yi ) are given by Eqs. (1a) and (1b) when c = i.
The structure is now divided into segments along the z axis by
notionally cutting the structure along horizontal planes at those
storey levels corresponding to changes in storey properties. Fig. 3
shows a typical segment formed by cutting the structure through
planes Ek Fk Gk Hk and Ek+1 Fk+1 Gk+1 Hk+1 that correspond to the kth
and k + 1th changes in storey properties. The number of storeys in
any one segment can vary from one to the total number of storeys
in the structure if it is uniform throughout its height. However, in
any one segment each storey must have the same properties.
We now consider a typical segment in isolation and seek to
replace each primary wall by a substitute flexural beam that
replicates its in-plane motion. We also substitute each core with
a flexuretorsion beam which can undergo flexural deflection as
well as St. Venant and warping torsion. We start by considering a
typical wall, wall i, that runs parallel to the yz plane, see Fig. 1.
This wall is replaced by a single substitute beam, beam i, shown
in Fig. 4. This beam is a two-dimensional flexural beam of length L
and has uniformly distributed mass and flexural stiffness. The mass
and elastic axes therefore coincide with the local z-axis and the
elastic axis is only permitted bending deformation vi (z , t ) in the
y direction, where z and t denote distance from the local origin

Qwyi (z , t )
2 vi (z , t )
z = mwyi
z
z
t2

(2a)

or


EIwyi

3 vi (z , t )
z3


= mwyi

2 vi ( z , t )
t2

(2b)

where E is Youngs modulus, Qwyi (z , t ) is the shear force, mwyi is the

uniformly distributed mass per unit length and EIwyi is the flexural
rigidity of the element for motion in the y direction.
With a similar argument, the motion for the substitute
flexuretorsion beam can be studied by considering a typical
elemental length of the beam, z, from Fig. 5. The equations of
motion governing the element when translating in the yz plane
may be written as follows using Eq. (1b) and noting that exi is the
eccentricity between the elastic and mass axes of the original core i.

Qcyi (z , t )
2
z = mci 2 (vi (z , t ) + exi (z , t )) z
z
t

(3a)

or

3 v i (z , t )
EIcyi
z3


= mci

2
(vi (z , t ) + exi (z , t ))
t2

(3b)

where Qcyi (z , t ) is the shear force, mci is the uniformly distributed

mass per unit length and EIcyi is the flexural rigidity of the element
for motion in the y direction.
If the equivalent procedure is carried out for all of the i walls
that run parallel to the yz plane and all of the cores, the dynamic

1906

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

nwy
X

mwyi

i =1

nc
X

mci

i =1

2 (v(z , t ) + xi (z , t ))
t2

2 (v(z , t ) + (xi + exi )(z , t ))

=0
t2

(5)

where xi and x i are the distances of wall i and core i from the
rigidity centre, O, respectively. Since O is the centre of rigidity,
P nw y
Pnc
i = 0 and Eq. (5) can be simplified to
i=1 EIw yi xi +
i=1 EIcyi x
EIy

4 v(z , t )
2 v(z , t )
2 (z , t )
+ my
+ xc my
=0
4
2
z
t
t2

(6)

in which
EIy =

nw y
X

EIwyi +

i=1

nc
X

EIcyi

(7a)

i=1

and
m y xc =

nw y
X

mwyi xi +

i =1

nc
X

mci (xi + exi )

i=1

where my =

nw y
X
i =1

mwyi +

nc
X

mci .

(7b)

i =1

Since the total mass of the segment contributes to its vibration,

including the mass of the walls running in the x direction and
the rigid diaphragms, my should be replaced by m, where m is
the equivalent distributed mass over the height of the segment.
Therefore
EIy

2 v(z , t )
2 (z , t )
4 v(z , t )
+
m
+
mx
= 0.
c
z4
t2
t2

(8)

In an identical fashion, the dynamic equilibrium relationship for

motion in the xz plane can be written as
EIx

Fig. 5. Co-ordinate system and positive sign convention for the substitute threedimensional flexural-torsion beam (a) Member and element convention for the xz
plane. (b) Member and element convention for the yz plane.

equilibrium equation for motion in the yz plane may be written

as
nw y
nc
X
X
3 vi ( z , t )
3 v i (z , t )
EIwyi
+

EI
cyi
z i =1
z3
z i=1
z

nw y
X

mwyi

i=1

nc
2 vi (z , t ) X
2
+
m
(vi (z , t ) + exi (z , t ))
ci
t2
t2
i=1

EIwyi

i =1

4 (v(z , t ) + xi (z , t ))
z4

nc

X
i=1

EIcyi

2 (v(z , t ) + x i (z , t ))
z2

(9)

Finally, it should be noted that the cores and walls running

parallel to the xz and yz planes also provide the torsional
stiffness of the building. Thus the required equation for torsion
can be developed from a consideration of the torsional equilibrium
about O, which yields
nw y
X

EIwyi xi

i=1

nc
X

4 (v(z , t ) + xi (z , t ))
z4

EIcyi x i

i =1

nwy
X

mwyi xi

i =1

(4)

where nwy and nc are the number of walls and cores, respectively.
Noting that EIwyi and EIcyi are constant over the length of the
member and substituting for vi (z , t ) and v i (z , t ) from Eq. (1b) with
c replaced by i gives
nwy

4 u(z , t )
2 u(z , t )
2 (z , t )
+
m

my
= 0.
c
z4
t2
t2

nc
X

4 (v(z , t ) + x i (z , t ))
z4

2 (v(z , t ) + xi (z , t ))
t2

mci (xi + exi )

i =1

"
nw x
X

EIwxj yj

j =1
nc

EIcxj y j

j=1
nw x

X
j=1

mwxj yj

2 (v(z , t ) + (xi + exi )(z , t ))

t2

4 (u(z , t ) yj (z , t ))
z4

4 (u(z , t ) y j (z , t ))
z4

2 (u(z , t ) yj (z , t ))
t2

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

nc
X

mcj (yj + eyj )

j =1

nc
X

EIwci

i=1

2 (u(z , t ) (yj + eyj )(z , t ))

t2

nc
4 (z , t ) X
2 (z , t )
+
GJ
=0
ci
z4
z2
i =1

1907

(10)

deflected
deflected

2 v(z , t )
4 (z , t )
2 (z , t )

GJ
+
m
x
y
c
z4
z2
t2
2
2
(z , t )
u(z , t )
+ Ig
=0
mx yc
t2
t2

undeflected

EIw

undeflected

where EIwci and GJci are the warping and St. Venant torsional
rigidity of core i about its own shear centre, respectively. Eq. (10)
can be simplified to

(11)

where
EIw =

nw y
X

EIwyi x2i +

i=1

nc
X

nwx
X

j =1
nc

EIcxj y 2j +

j =1

GJ =

nc
X

EIwxj y2j +

nc
X

EIcyi x 2i

i=1

EIwci

(12a)

i=1

GJci

(12b)

i =1

Ig =

nwy
X

mwyi x2i +

i =1

nw x
X

mwxj y2j

j =1

nc
X

mci (xi + exi )2 +

i =1

nc
X

mcj (yj + eyj )2 .

(12c)

j =1

EI w is the warping torsional rigidity of the walls and cores about the
flexural rigidity centre O; GJ is the sum of the St. Venant torsional
rigidity of the cores about their own shear centres and Ig is the polar
second moment of mass about the rigidity centre O.
As before, the total mass of the cores and walls running in the x
and y directions, as well as that of the rigid diaphragms, should
be taken into account. Thus Eqs. (8), (9) and (11) can finally be
rearranged and written in the following form

4 u(z , t )
2 u(z , t )
2 (z , t )
+
m

my
=0
c
z4
t2
t2
4 v(z , t )
2 v(z , t )
2 (z , t )
EIy
+m
+ mxc
=0
4
2
z
t
t2
4 (z , t )
2 (z , t )
2 u(z , t )
GJ
myc
EIw
4
2
z
z
t2
2
2 v(z , t )

(
z
,
t
)
+ mxc
+ mrm2
=0
2
2
t
t
EIx

(13a)
(13b)

(13c)

where rm is the polar mass radius of gyration of the structure about

the rigidity centre O. Eqs. (13a)(13c) are the required differential
equations of motion.
The minimum requirements for Eqs. (13) to be valid are that
the structural system should contain either a single core or a single
wall running in each of the x and y directions.
4. Eigenvalue problem
Eqs. (13) are now solved and posed in dynamic stiffness
form. Although each equation was developed individually from
a consideration of the planar flexural and flexuretorsion beams
of Figs. 4 and 5, they now describe the motion of a threedimensional, flexuretorsion coupled beam whose co-ordinate
system and sign convention are shown in Fig. 6. This beam (exact
finite element) will replace a typical segment of the original,

Fig. 6. Co-ordinate system and positive sign convention for forces and
displacements of the three-dimensional flexuretorsion coupled beam. (a) Member
and element convention for the xz plane. (b) Member and element convention for
the yz plane.

asymmetric, three-dimensional wall-core structure. The whole of

the original structure can then be reconstituted by assembling the
exact finite elements corresponding to each segment in the usual
way.
Eqs. (13) are solved on the assumption of harmonic motion, so
that the instantaneous displacements can be written as
u(z , t ) = U (z ) sin t

(14a)

v(z , t ) = V (z ) sin t
(z , t ) = (z ) sin t

(14b)
(14c)

where U (z ), V (z ) and (z ) are the amplitudes of the sinusoidally

varying displacements and is the circular frequency.
Substituting Eqs. (14) into Eqs. (13) and re-writing in nondimensional form gives
U 0000 ( ) x2 2 U ( ) + yc 2 x2 ( ) = 0
V

0000

( ) V ( ) xc
2
y

2
y

( ) = 0

(15a)
(15b)

1908

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

0000 ( ) 2 00 ( ) 2 2 ( )

+ yc 2

It follows that the solution of Eq. (18) is of the form

x2
U ( ) xc 2 2 V ( ) = 0
2
x
y
2
y

(15c)

W ( ) = C1 cosh + C2 sinh + C3 cosh + C4 sinh

+ C5 cosh + C6 sinh + C7 cos + C8 sin

+ C9 cos + C10 sin + C11 cos + C12 sin . (21)

where

2 =
=
2
x

y2 =

GJ
EIw

L2 ,

mL4
EIx
mL4
EIy

2 = rm2

(16a)

U ( ) = C1u cosh + C2u sinh + C3u cosh + C4u sinh

(16b)

+ C5u cosh + C6u sinh

u
+ C7u cos + C8u sin + C9u cos + C10
sin

(16c)

u
u
+ C11
cos + C12
sin
(22a)
v
v
V ( ) = C1 cosh + C2 sinh + C3v cosh + C4v sinh
+ C5v cosh + C6v sinh
v
+ C7v cos + C8v sin + C9v cos + C10
sin

mL4

(16d)

EIw

x2 =

EIw

y2 =

EIw

EIx

(16e)
and

EIy

(16f)

= .

(16g)

D4 2 x2
0

yc

2
x
2
x

0
D
4

x c

2
y
2
y

y2

yc 2 x2
xc 2 y2

v
v
+ C11
cos + C12
sin
(22b)
( ) = C1 cosh + C2 sinh + C3 cosh + C4 sinh
+ C5 cosh + C6 sinh
+ C7 cos + C8 sin + C9 cos + C10 sin
+ C11 cos + C12 sin .
(22c)

The relationship between the constants Cju , Cjv and Cj (j = 1, 12)

also follows from Eq. (18) as

Eqs. (15) can be re-written in the following matrix form

Eq. (21) represents the solution for U ( ), V ( ) and ( ), since they

are related via Eq. (18). Hence they can be written individually as

"
#
U ( )
V ( ) = 0

( )
D4 2 D2 2 2
(17)

u
u
C2j
1 = tj C2j1
u
C2j

tju C2j

C2j1 = tj C2j1
v

in which D = d/d .
Eq. (17) can be combined into one equation by eliminating
either U , V or to give the twelfth-order differential equation

C2j = tj C2j

 4
D 2 x2


0

2

 yc 2 x

x2

tju =

0
D4 2 y2

x c 2

y2

y2



yc

2 2

x c y
 W ( ) = 0

D4 2 D2 2 2 
2

2
x

(18)

where W = U, V or .
The solution of Eq. (18) is found by substituting the trial solution
W ( ) = ea to yield the characteristic equation

 2
 2 x2


0

2

 yc 2 x

x2

2 2 y2
y2
x c 2 2
y

yc 2 x2
xc 2 y2





 W ( ) = 0

2
2
2 2


,
i, i

,
,
i, i

tjv =

= 2 ,
1 ,

= 5 and = 6

and i = 1.

3 ,

4 ,

(j = 1, 2, 3, 4, 5, 6).

j2 2 y2

(23c)
(23d)

(24b)

Following the sign convention of Fig. 6(a) and (b), expressions for
the bending rotations x ( ), y ( ) and the gradient of the twist
0 ( ) are easily established as
1 dU ( )
L

(25a)

1 dV ( )
L

0 ( ) =

1 d ( )
L

and

d
d

(25b)
(25c)

The corresponding bending moments Mx ( ), My ( ), the bimoment B( ), lateral shear forces Qx ( ), Qy ( ) and torsional
moment T ( ) are likewise easily determined from the appropriate
stress/strain relationships as
Mx ( ) =

EIx d2 U ( )
,
L2
d 2

(26a)

My ( ) =

EIy d2 V ( )
and
L2
d 2

(26b)

where

(23b)

(24a)

xc 2 y2

y ( ) =

(20a)

(j = 1, 6)

yc 2 x2
,
j2 2 x2

(19)

i, i

and

(23a)

where

x ( ) =

where = a2 .
Eq. (19) is a sixth order equation in and it can be proven
(Appendix) that it always has three positive and three negative real
roots. Let these six roots be 1 , 2 , 3 , 4 , 5 and 6 , where
j (j = 1, 6) are all real and positive. Therefore the twelve roots of
Eq. (19) can be obtained as

(j = 1, 6)

and

(20b)
B( ) =

EIw d2 ( )
L2

d 2

(26c)

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

EIx d3 U ( )
,
L3
d 3

Qx ( ) =

T ( ) =

(26d)

In similar fashion the vector of nodal forces can be determined

from Eqs. (26) and (27) as

(26e)

 

EIy d3 V ( )
Qy ( ) =
L3
d 3
EIw d ( ) GJ d ( )
+
.
L3
d 3
L d

(26f)

The nodal displacements and forces can now be defined in the

member co-ordinate system of Fig. 6(a) and (b), as follows

x = 1x ,

V = V1 ,

y = 1y ,
= 1 ,
0 = 10
At = 1 U = U2 ,
x = 2x ,
V = V2 ,

(27a)

y = 2y ,
= 2 ,
At = 0 Qx = Q1x ,

(27b)

My = M1y ,

0 = 20
Mx = M1x ,

T = T1 ,

At = 1 Qx = Q2x ,
My = M2y ,

Qy = Q1y ,

B = B1

(27c)

Mx = M2x ,

T = T2 ,

Qy = Q2y ,

B = B2 .

d1
d2

E1
0
=
E1 C h
E1 R1 Sh

E2
0
E2 C
E2 R2 S

0
E1 R1
E 1 Sh
E1 R1 Ch

0
 
E2 R2 Co

E2 S
Ce
E2 R2 C

(28)

where

V1
1

d1 =
1x ,

1y
10
C

V2
2

d2 =
2x ,

2y
20
C

C3
C5

Co =
C7 ,

C4
C6

Ce =
C8 ,

C9
C11

"
E1 =

R1 =

t1u
t1v
1

"
1

0
0

R2 =

0
sinh
0

cos
0
0

0
cos
0

t4u
t4v
1

t5u
t5v
1

0
0
cosh
0
0
sinh

0
0
,
cos

t6u
t6v
1

"

sinh
0
0

"

"

E2 =

0
cosh
0

Ch =

Sh =

0
0

0
0

0
0

Co
Ce

E1
0
=
E1 Ch
E1 R1 Sh

1x

(32a)

M1y
B1

Q
2x

Q2y
T2

p2 =
M2x ,

(32b)

M2y
B2

t1u 3 Bx

Q1 =
t1v 3 By
3
Eo F

t2u 3 Bx
t2v 3 By
3 Eo F

t3u 3 Bx

t3v 3 By
3
Eo F

(32c)

t4u 3 Bx
t5u 3 Bx
t6u 3 Bx

Q2 = t4v 3 By
t5v 3 By
t6v 3 By
3
3
3
Eo F Eo F Eo F
u 2

t1 ( Ax ) t2u ( 2 Ax ) t3u ( 2 Ax )
M1 = t1v ( 2 Ay ) t2v ( 2 Ay ) t3v ( 2 Ay ) ,
2 Do
2 Do
2 Do
u 2

t4 ( Ax ) t5u (2 Ax ) t6u (2 Ax )
M2 = t4v ( 2 Ay ) t5v (2 Ay ) t6v (2 Ay )
2 Do
2 Do
2 Do

EIx

L3
EIy

Do =
Eo =

#
,

(32d)

(32e)

(32f)

sin
0
0

0
sin
0

0
0
.
sin

0
E2 R2
E2 S
E2 R2 C

(33a)

(33b)

(33c)

L3
EIw
L2
EIw

L3
GJ

(33d)

(33e)

(33f)

.
(33g)
L
Thus the required stiffness matrix can be developed by substituting
Eqs. (30) into Eqs. (31) to give
F =

"

0
E1 R1
E1 Sh
E1 R1 Ch

L2
EIx

By =

S=

L2
EIy

Bx =

E2
0
E2 C
E2 R2 S

(31)

Q1y
T1

p1 =
M1x ,

Hence the vector of constants [Co Ce ]T can be determined from

Eq. (28) as

Q2
 
0
Co

Ce
Q 2 C
M2 S

Ay =

(29)

 

Q1
0
Q 1 C h
M1 Sh

where

Ax =

"

cosh
0
0

"

C=

t3u
t3v ,
1

0
M2
Q2 S
M2 C

where

C10
C12

t2u
t2v
1

0
M1
=
Q1 Sh
M 1 C h

(27d)

Then the nodal displacements can be determined from Eqs. (22)

and (25) as

 

p1
p2

At = 0 U = U1 ,

1909

 
p1
p2

0
M1
=
Q 1 S h
M1 Ch

 

d1
.
d2

(30)

E1
0

E1 Ch
E1 R1 Sh

0
M2
Q2 S
M2 C

Q1
0
Q 1 C h
M1 Sh

E2
0
E2 C
E2 R2 S

Q2
0

Q 2 C
M 2 S

0
E1 R1
E1 Sh
E1 R1 Ch

0
E2 R2
E2 S
E2 R2 C

 
d1
d2

(34)

1910

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

or

or

p = kd.

(35)

The stiffness relationship of Eq. (35) is general and can be

used in the normal way to assemble more complex forms.
The required natural frequencies of the resulting structure are
determined by evaluating its overall dynamic stiffness matrix at
a trial frequency and using the WittrickWilliams algorithm to
establish how many natural frequencies have been exceeded by .
This clearly provides the basis for a convergence procedure that
can yield the required natural frequencies to any desired accuracy.
The corresponding mode shapes can then be recovered by any
appropriate method [20].
5. WittrickWilliams algorithm
The WittrickWilliams algorithm [21,22] has been available
for over thirty years and has received considerable attention. The
algorithm states that
J = J 0 + s{ K }

(36)

where J is the number of natural frequencies of the structure

exceeded by some trial frequency, , Jo is the number of natural
frequencies that would still be exceeded if all members were
clamped at their ends so as to make D = 0 and s{K} is the sign
count of the dynamic structure stiffness matrix K. s{K} is defined
in reference [22] and is equal to the number of negative elements
on the leading diagonal of the upper triangular matrix obtained
from K, when = , by the standard form of Gauss elimination
without row interchanges.
From the definition of Jo , it can be seen that
Jo =

Jm

(37)

where Jm is the number of natural frequencies of a member,

with its ends clamped, which have been exceeded by , and
the summation extends over all members of the structure. In
some cases it is possible to determine the value of Jm for an
individual member symbolically using a direct approach [20] that
gives an analytical expression for Jm . However this is impractical
in the present case due to the algebraic complexity. Instead, Jm
is evaluated using an argument based on Eq. (37) that applies
the WittrickWilliams algorithm [22] in reverse. The procedure
corresponds to the one originally proposed by Howson and
Williams [23] and is described as follows.
Consider an element which has been isolated from its
neighbours by clamping its ends. Treating this member as a
complete structure, it is evident that the required value of Jm could
be evaluated if its natural frequencies were known. Unfortunately
this simple structure can rarely be solved easily. We therefore
seek to establish a different set of boundary conditions (other than
clampedclamped) which admit a simple solution from which the
solution for the clampedclamped case can be deduced. This is
most easily achieved in the present case by imposing what will be
defined as simply supported boundary conditions, i.e. at

= 0 and = 1,
Mx = My = B = 0.

U =V ==0

2y

11,4

11,5

11,6

11,10

11,11

11,12

K12,4

K12,5

K12,6

K12,10

K12,11

K12,12

(40)

where the Ki,j are the remaining elements of k with their original
row i and column j subscripts and kss is the required 6 6 matrix
for this simple one member structure.
Application of the WittrickWilliams algorithm [22] to this
simple structure gives
Jss = Jm + s{kss }

or

(41)

Jm = Jss s{kss }

(42)

where Jss is the number of natural frequencies of the simply

supported member that lie below the trial frequency , Jm is the
required number of clampedclamped natural frequencies of the
member lying below , s{kss } is the number of negative elements

on the leading diagonal of k

ss , where kss is the upper triangular
matrix obtained by applying the usual form of Gauss elimination
to kss .
The evaluation of s{kss } is clearly straightforward and the
problem thus lies in determining Jss .
Based on Eqs. (22) and (26a)(26c), the boundary conditions of
Eq. (38) are satisfied by assuming solutions for the displacements
U ( ), V ( ) and ( ) of the form
U ( ) = Ck sin(k ),
V ( ) = Dk sin(k )

(43a)
and

(43b)

( ) = Ek sin(k ) (k = 1, 2, 3, . . .)

(43c)

where Ck , Dk and Ek are constants.

Substituting Eqs. (43) into Eq. (17) gives

(k )4 2 x2
0

" #

yc 2

2
x
2
x

0
(k )4 2 y2

yc 2 x2
xc 2 y2

x c

(k )4 + 2 (k )2 2

2
2 y

y2

Ck

Dk = 0

(44)

Ek
in which represents the coupled natural frequencies of the
member with simply supported ends. The non-trivial solution of
Eq. (44) is obtained when



(k )4 2 2 0
yc 2 x2
x




4
2 2
2 2
0
(k ) y xc y



 = 0. (45)
2
2


2 x
2 y
4
2
2
2 2
 yc


(
k

)
+

(
k



x2
y2
Eq. (45) is a cubic equation in 2 and yields three positive
values of for each value of k. It is then possible to calculate Jss
by substituting progressively larger values of k until all of those
natural frequencies lying below have been accounted for. Once
Jss is known, Jm can be calculated from Eq. (42).
6. Special case: Uniform structures

and

(38)

The stiffness relationship for a single member subject to these

boundary conditions can then be obtained by deleting appropriate
rows and columns from Eq. (35) to leave

M1x
K4,4
K4,5
K4,6
K4,10
K4,11
K4,12
1x
M1y K5,4 K5,5
K5,6
K5,10
K5,11
K5,12 1y
0

B1 K6,4 K6,5

K6,6
K6,10
K6,11
K6,12

1
M2x = K10,4 K10,5 K10,6 K10,10 K10,11 K10,12 2x (39)

M K

K
K
K
K
K
B2

pss = kss dss

2y

20

When all the storeys of a wall-core building can be considered

to be identical, the whole building may be modelled as a single
flexuretorsion beam that is clamped at one end and free at the
other. The end conditions for such a beam are
d1 = 0;

(46a)

p2 = 0.

(46b)

The natural frequencies of such a beam can be found easily by

hand if the St. Venant torsional rigidity of the cores is ignored.
Such a course of action is considered in the Numerical Results
Section, where its effect has been assessed through a parametric

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

1911

study. The study shows that the St. Venant torsional rigidity makes
little contribution towards the overall behaviour of structures
comprising open cores and may safely be ignored in most practical
cases. When this is the case, Eq. (46b) can be written in the
following form using Eq. (26)
U 00 ( = 1)
V 00 ( = 1)
00

( = 1)
U 000 ( = 1) = 0.

V 000 ( = 1)
000 ( = 1)

(47)

If Eqs. (47) and (46a) are substituted into Eqs. (28), suitably
differentiated, it is clear that the condition for non-trivial solutions
is


 E1
 0

E R2 C
 1 1 h
 E R3 S
1

E2
0
E2 R21 C
E2 R31 S

1 h

0
E1 R1
E1 R22 Sh
E1 R32 Ch

0


E2 R2

2  = 0.
E2 R2 S 
E2 R32 C

then be calculated from

Noting that the St. Venant torsional rigidity of the cores has been
ignored, it is easy to show that the three positive and three negative
roots of Eq. (19) are symmetrical and therefore
R1 = R2

and

E1 = E2 .

0
E1
0
0

0
0
E1
0

0
0
0
E1


 I
 0

 0

 0

0
I
0
0

I
0

0
R1
0
0


0  I

0 0

0  Ch
3  S
R

0
0
R21
0

I
0
C

0
I
Sh
Ch





=0
S Sh 

C C

0
I

C Ch
S Sh


0

I
 = 0. (50)
S
C

(56)

Substituting Eq. (53) into Eq. (56) gives

j(2) =
j(3) =
i(4) =

3.5156
bj

22.0336
bj
61.7010
bj

(57a)

(57b)

(57c)

120.9120
bj

(j = 1, 2, 3)

(57d)

and


(51)

(j = 1, 2, 3).

bj

j(1) =

It is clear in this form that only the right-hand determinant can pass
through zero for non-trivial solutions. Noting that the elements
of the right hand determinant all are diagonal matrices, it can be
simplified to


I

0
0

0

j =

(49)

Hence Eq. (48) can be written in the following form


E1

0
0

0

Fig. 7. Floor plan of all structures considered in Section 8.

(48)

j(k) =

(k 0.5)2 2
bj


. . . k = 5, 6, 7, . . .

(j = 1, 2, 3). (57e)

or

7. Examples

|I + CCh | = 0.

(52)

The solution of Eq. (52) can be calculated to any desired

accuracy, although for illustrative purposes the solutions are given
approximately as

or or = 1.875, 4.694, 7.855, 10.996 and



1
k
for k = 5, 6, 7, . . . .

(53)

Substituting Eq. (53) into Eq. (19) and solving for yields the
required natural frequencies. In addition, it is interesting to note
that by combining and into a single parameter, an even simpler
solution procedure can be obtained as follows. Let
b = /

(54)

then Eq. (19) can be written in the following form

 2
b x2

 0

 yc 2

0
b2 y2
xc 2


yc x2

 W ( ) = 0.
xc y2

2
rm
(b2 2 )


(55)

Eq. (55) is a cubic equation in the frequency parameter b2 and it

can be proven [24] that it always has three positive roots. Let these
roots be b21 , b22 and b23 . The natural frequency of the building can

The work of this section consolidates the foregoing theory by

performing a parametric study on five wall-core structures of
varying height and comparing the lower natural frequencies with
those obtained from a full finite element analysis. The structures,
which have 5, 10, 20, 40 and 60 storeys, respectively, all have the
same doubly asymmetric floor plan and equal storey heights of 3 m.
Each structure comprises three walls in the y direction (W1-W3),
three walls in the x direction (W4-W6) and two cores (C1-C2), each
of which are connected to each other by typical rigid diaphragms at
each floor level with the arrangement shown in Fig. 7. In the 5, 10
and 20 storey buildings, the properties of the structural elements
do not change along the height of the structure, so each structure
can be modelled using a single substitute beam element and the
natural frequencies can be determined from the theory of Section 6.
In the 40 and 60 storey buildings, the properties of the structural
elements change in a step-wise fashion every 20 storeys. Table 1
shows the the thickness of the walls and cores of all the buildings
on different floor levels.
For simplicity in determining member masses, half the mass
of the walls and cores framing into and emanating from a floor
diaphragm, together with the mass of the diaphragm is stated as
an equivalent uniformly distributed floor mass at that storey level.
Thus the centre of mass is at the geometric centre of the floor plan.

1912

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

Table 1
Thickness of the walls and cores for each building.
Building height (storeys)

Floors

Thickness of the walls and cores

5
10
20
40

1st to 5th
1st to 10th
1st to 20th
1st to 20th
21st to 40th
1st to 20th
21st to 40th
41st to 60th

0.20 m
0.20 m
0.20 m
0.25 m
0.20 m
0.30 m
0.25 m
0.20 m

60

The shear modulus and Youngs modulus for all members are taken
to be G = 9 109 N/m2 and E = 2 1010 N/m2 , respectively.
The properties of Core C1 in Fig. 7, with regard to its own shear
centre, are given in Table 2. The properties of Core C2 can therefore
be obtained from the same table by exchanging the values of Ix
and Iy .
The location of the centre of rigidity, O, should satisfy the
following equations as discussed in Section 3.
3
X

EIwyi xi +

i=1
3
X

Table 2
Properties of Core C1 of Fig. 7.
Thickness of the core

Icx1 (m4 )

Icy1 (m4 )

Iwc1 (m6 )

Jc1 (m4 )

0.20 m
0.25 m
0.30 m

2.250
2.820
3.390

14.416
18.030
21.654

14.175
17.719
21.263

0.032
0.063
0.108

Table 3
Properties of the substitute flexuretorsion beam calculated from Eq. (7), the
equivalent equations for motion in the xz plane and Eq. (12).
Thickness of the walls and cores

Ix (m4 )

Iy (m4 )

Iw (m6 )

J (m4 )

0.20 m
0.25 m
0.30 m

36.013
45.040
54.074

27.466
34.350
41.244

4263.60
5331.01
6397.21

0.064
0.125
0.216

This corresponds precisely to the automatic idealisation process in

ETABS [25] and additionally only requires the total mass of the floor
to be converted into the equivalent uniformly distributed mass of
the member in the substitute beam approach. Arbitrarily the mass
is assumed to have a constant value of 360 kg/m2 at each floor
level, even where the stiffness properties of the member change.

2
X

EIcyi x i = 0

(58a)

EIcxj y j = 0.

(58b)

i=1

EIwxj yj +

j =1

2
X
j =1

Since all the structures have the same doubly asymmetric floor
plan and the thickness of the walls and cores in the 40 and 60
storey buildings change in the same ratio, it is clear that the rigidity
centre at each floor level lies in a vertical line through the building.
Utilising Eqs. (58a) and (58b), the eccentricities in the x and y
directions can be calculated as
xc = 6.79 m,

yc = 2.89 m.

The properties of the substitute beam for different thicknesses

of the walls and cores have been obtained using Eq. (7), the
equivalent equations for motion in the xz plane and Eq. (12) and
are given in Table 3.
The distributed mass of the substitute beam (smeared from
the diaphragm) and the polar mass radius of gyration of the
diaphragms about the shear centre can be calculated as follows
m = 18 30 360/3 = 64, 800 kg/m
2
rm
=

182 + 302
12

+ 6.792 + 2.892 = 156.46 m2 .

Table 4a
Coupled natural frequencies (Hz) of the 5-storey building from the continuum and FEM models. The frequency reduction factor of Eq. (59), rf = 0.8416.
Freq. no.

1
2
3

3D flexuretorsion beam
(Eq. (35))

3D flexuretorsion beam
(Eq. (57a)) (GJ = 0)

Modified 3D flexuretorsion beam

(Eq. (59))

ETABS (FEM)

Difference %

(1)

(2)

(3) = (1) rf

(4)

(1)(4)
(4)

(2)(4)
(4)

(3)(4)
(4)

5.7715
8.1179
11.4532

5.7694
8.1170
11.4499

4.8573
6.8320
9.6390

4.5111
6.0192
8.4687

27.94
34.87
35.24

27.89
34.85
35.20

7.67
13.50
13.82

32.68

32.65

11.67

Av.

Table 4b
Coupled natural frequencies (Hz) of the 10-storey building from the continuum and FEM models. The frequency reduction factor of Eq. (59), rf = 0.9106.
Freq. no.

1
2
3

3D flexuretorsion beam
(Eq. (35))

3D flexuretorsion beam
(Eq. (57a)) (GJ = 0)

Modified 3D flexuretorsion beam

(Eq. (59))

ETABS (FEM)

Difference %

(1)

(2)

(3) = (1) rf

(4)

(1)(4)
(4)

(2)(4)
(4)

(3)(4)
(4)

1.4434
2.0295
2.8645

1.4423
2.0293
2.8625

1.3144
1.8481
2.6084

1.2883
1.7813
2.5100

12.04
13.93
14.12

11.95
13.92
14.04

2.02
3.75
3.92

13.37

13.31

3.23

Av.

Table 4c
Coupled natural frequencies (Hz) of the 20-storey building from the continuum and FEM models. The frequency reduction factor of Eq. (59), rf = 0.9522.
Freq. no.

1
2
3
Av.

3D flexuretorsion beam
(Eq. (35))

3D flexuretorsion beam
(Eq. (57a)) (GJ = 0)

Modified 3D flexuretorsion beam

(Eq. (59))

ETABS (FEM)

Difference %

(1)

(2)

(3) = (1) rf

(4)

(1)(4)
(4)

(2)(4)
(4)

(3)(4)
(4)

0.3615
0.5074
0.7173

0.3606
0.5073
0.7156

0.3442
0.4831
0.6830

0.3421
0.4792
0.6756

5.67
5.88
6.17

5.41
5.86
5.92

0.62
0.82
1.10

5.91

5.73

0.85

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

1913

Table 4d
Coupled natural frequencies (Hz) of the 40-storey building from the continuum and FEM models. The frequency reduction factor of Eq. (59), rf = 0.9752.
Freq. no.

1
2
3

3D flexuretorsion beam
(Eq. (35))

3D flexuretorsion beam
(Eq. (35)) (GJ = 0)

Modified 3D flexuretorsion beam

(Eq. (59))

ETABS (FEM)

Difference %

(1)

(2)

(3) = (1) rf

(4)

(1)(4)
(4 )

(2)(4)
(4)

(3)(4)
(4)

0.1011
0.1410
0.2005

0.1002
0.1410
0.1988

0.0986
0.1375
0.1955

0.0977
0.1373
0.1937

3.48
2.69
3.51

2.56
2.69
2.63

0.91
0.15
0.94

3.23

2.63

0.67

Av.

Table 4e
Coupled natural frequencies (Hz) of the 60-storey building from the continuum and FEM models. The frequency reduction factor of Eq. (59), rf = 0.9833.
Freq. no.

1
2
3

3D flexuretorsion beam
(Eq. (35))

3D flexuretorsion beam
(Eq. (35)) (GJ = 0)

Modified 3D flexuretorsion beam

ETABS (FEM)

Difference %

(1)

(2)

(3) = (1) rf

(4)

(1)(4)
(4)

(2)(4)
(4)

(3)(4)
(4)

0.0492
0.0678
0.0975

0.0481
0.0677
0.0956

0.0483
0.0666
0.0958

0.0474
0.0666
0.0940

3.80
1.80
3.72

1.48
1.65
1.70

1.96
0.00
1.89

3.11

1.61

1.28

Av.

8. Numerical results
Column 2 of Tables 4a4e shows the first three coupled natural
frequencies (Hz) of the 5,10, 20, 40 and 60 storey buildings,
respectively, obtained from the proposed three-dimensional
flexuretorsion beam theory. The third column in each table shows
the natural frequencies when the St. Venant torsional rigidity of
the cores is ignored. (These figures can easily be obtained using
the theory of Section 6 and Eq. (57a) in the case of the 5, 10 and
20 storey buildings since they have uniform properties throughout
their height.) In the fourth column, the results of Column 2
have been modified by multiplying the natural frequencies by a
reduction factor rf , which allows for the fact that a proportion
of the mass of the building is concentrated at floor levels and
is therefore not uniformly distributed over the height of the
building, as assumed in the derivation of the proposed model. It
is clear that this assumption has considerable influence on short
buildings, although it can normally be ignored when dealing with
tall buildings. Zalka [8] suggests that the reduction factor rf can be
calculated as

r
rf =

n
n + 2.06

(59)

where n is the number of storeys in the building.

Finally, the fifth column in each table shows the results of a full
finite element analysis of the original structures obtained using the
vibration programme ETABS, in which the automatic idealisation
process was utilised that assumes uniformly distributed mass on
rigid floor diaphragms. Relevant comparisons are made in columns
six to eight.
9. Discussion
The results in Table 4a for the five storey building are the least
accurate of those presented due to the considerable difference
between the concentrated and distributed mass of the floors over
the height of the structure. However, the difference between
the model results and those of the finite element analysis still
lie below 12% when the reduction factor of Eq. (59) is applied.
As the number of storeys increases, the difference between the
results becomes significantly less, as shown in Tables 4b4e. The
results for the 20, 40 and 60 storey buildings would appear to be
perfectly satisfactory, even without applying the reduction factor.
Comparison of the results in columns 2 and 3 of each table indicates

Fig. 8. Graphs of the difference between the averaged results from the proposed
model for the three cases indicated in Tables 4a4e and those from the full finite
element analysis of the original structures.

that the St. Venant rigidity of open cores can safely be ignored. The
omission of this effect might have greater influence in structures
comprising closed cores and this requires further investigation.
Finally, Fig. 8 shows the difference between the results from the
full finite element analysis of the original structures and those from
the three forms of the proposed model i.e. with GJ, without GJ
and modified. This suggests that the proposed model is likely to
be satisfactory for buildings with 10 to 60 storeys, depending on
the accuracy required, although more investigation is required for
shorter and taller structures.
10. Conclusions
A simple and accurate model has been developed for calculating
the lower natural frequencies corresponding to overall modes
of vibration of medium and tall building structures. Within this
scope it can encompass many geometric configurations ranging
from uniform structures with doubly symmetric floor plans to
doubly asymmetric ones with step changes of member properties
at any number of storey levels. The model has been developed
on the assumption of uniformly distributed mass and stiffness
and thus necessitates the solution of a transcendental eigenvalue
problem. This can be solved to any desired accuracy by use of
the WittrickWilliams algorithm, which also guaranties that no
natural frequencies can be missed. When all the storeys within a
building can be considered to be identical, the required solutions

1914

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

can be found easily by hand. Results of a parametric study show

that the model is likely to yield results of sufficient accuracy for
engineering calculations when the number of storeys is greater
than about ten and less than about sixty. As is inevitably the case
when using simplified models, their accuracy should be thoroughly
checked prior to use against datum results for the class of structure
being considered.
Appendix. The roots of Eq. (19)

Fig. A.1. (a) Graph of A( ) versus ; (b) Graph of B( ) versus .

The nature of the roots of the characteristic Eq. (19) is

investigated in this appendix. For this purpose, Eq. (19) is rewritten again for convenience.

 2
 2 x2


0

2

 yc 2 x

x2



yc


x c
 W ( ) = 0.

2
2
2 2

2

2 2 y2
y2
x c 2 2
y

2
x
2 2
y

(A.1)

Since 2 , x2 , y2 , 2 , x2 , y2 , xc and yc are all real constants, the

coefficients in Eq. (A.1) are all real. It will be convenient to note
that the determinantal part of Eq. (A.1) is a 6th order polynomial
function f ( ) so that


 A( )
 0

f ( ) =  1
 D( )
2

0
B( )
1
E ( )
2


D( )
E ( ) 

C ( ) 

(A.2)

in which
A( ) = ,

(A.3)

B( ) = 2 2 y2 ,

(A.4)

C ( ) = 2 2 2 2

(A.5)

D( ) = yc 2 x2 ,

(A.6)

E ( ) = xc 2 y2 .

(A.7)

2
x

The quantity f ( ) is a smooth, continuous function and it is easy

to show that f ( ) as . Additionally f ( ) is negative
when = 0, as shown below.
Substituting = 0 in Eq. (A.2) gives


 2 x2

 0
f (0) = 
2
yc 2 x

x2

yc 2 x2 

xc 2 y2 

x c

2
y

2
2 y

2 2

y2

(A.8)

or
f (0) = 2 x2

4 y2 2

+ yc 2 x2

y2

!
x2c 4 y2

y 2 x2 y2
2 c

(A.9)

Eq. (A.9) can be simplified to

2
f (0) = (1/rm
)(rm2 x2c y2c )x2 y2 2 6

(A.12)

The right hand side of Eq. (A.10) is the product of six positive
parameters multiplied by a negative sign and therefore f (0) is
always negative.
Consider now the more general form of f ( ). From Eq. (A.3) it is
clear that A( ) is a second order equation in terms of and always
has one positive (x1 ) and one negative (x2 ) root, as follows
x1,2 = x .

(A.13)

In similar fashion, B( ) has one positive (y1 ) and one negative (y2 )
root given by
y1,2 = y .

(A.14)

Before discussing the roots of f ( ) = 0, it is useful to calculate

the quantity f ( ) when = x1 . Then
f (x1 ) =

x2

D2 (x1 )B(x1 ).

(A.15)

Since x2 and D2 (x1 ) always have positive values, the sign of

f (x1 ) only depends on the sign of B(x1 ). Similarly
f (y1 ) =

y2

E 2 (y1 )A(y1 )

(A.16)

and the sign of f (y1 ) only depends on the sign of A(y1 ).

Now Eq. (A.3) and Fig. A.1(a) show that
A( ) < 0

when x2 < < x1

and

(A.17a)

A( ) > 0

when > x1 or < x2 .

(A.17b)

B( ) < 0

when y2 < < y1

and

(A.18a)

B( ) > 0

when > y1 or < y2 .

(A.18b)

We now wish to consider the two cases in which x1 < y1 and

x1 > y1 .
Fig. A.1(a) and A.1(b) show that when x1 < y1 , B(x1 ) < 0 and
A(y1 ) > 0 so that f (x1 ) > 0 and f (y1 ) < 0.
Similarly, when x1 > y1 , B(x1 ) > 0 and A(y1 ) < 0 so that
f (x1 ) < 0 and f (y1 ) > 0.
If 1 represents the minimum value of x1 and y1 , i.e. 1 =
Min[x1 , y1 ], and 2 represents the maximum value of x1 and y1 ,
i.e. 2 = Max[x1 , y1 ], then
f (0) < 0,

(A.10)

in which rm is the polar mass radius of gyration about the flexural

rigidity centre O and can be related to the polar mass radius
of gyration about the centre of mass, rmc , through the following
equation
2
2
rm
= rmc
+ x2c + y2c .

(rm2 x2c y2c ) > 0.

Similarly, Eq. (A.4) and Fig. A.1(b) show that

Therefore

(A.11)

(A.19a)

f (1 ) > 0,
f (2 ) < 0
f () > 0.

(A.19b)
and

(A.19c)
(A.19d)

This implies that there are at least three positive real roots of
the function f ( ) in the intervals (0, 1 ), (1 , 2 ) and (2 , ). See
Fig. A.2.

B. Rafezy, W.P. Howson / Engineering Structures 31 (2009) 19031915

Fig. A.2. Graph of f ( ) versus .

Using an identical argument, the roots of f ( ) for negative

values of can be defined as
f (0) < 0,

(A.20a)

f (3 ) > 0,

(A.20b)

f (4 ) < 0 and

(A.20c)

f () > 0

(A.20d)

where 3 represents the maximum value of x2 and y2 , i.e. 3 =

Max[x2 , y2 ] and 4 represents the minimum value of x2 and y2 ,
i.e. 4 = Min[x2 , y2 ]. This implies that there are at least three
negative real roots of the function f ( ) in the intervals (, 4 ),
(4 , 3 ) and (3 , 0).
Since Eq. (A.1) is a sixth order equation in terms of , it has been
proven that it will always have three negative and three positive
real roots.
References
[1] Howson WP. Global analysis: Back to the future. Struct Eng 2006;84(3):1821.
[2] Kuang JS, Ng SC. Coupled lateraltorsion vibration of asymmetric shear-wall
structures. Thin-Walled Struct 2000;38(2):93104.
[3] Kuang JS, Ng SC. Dynamic coupling of asymmetric shear wall structures: An
analytical solution. Internat J Solids Struct 2001;38(4849):872333.
[4] Ng SC, Kuang JS. Triply coupled vibration of asymmetric wall-frame structures.
J Struct Eng ASCE 2000;126(9):9827.

1915

[5] Kuang JS, S.C. Ng. Coupled vibration of tall building structures. Struct Des Tall
Special Build 2004;13(4):291303.
[6] Wang Y, Arnaouti C, Guo S. A simple approximate formulation for the first
two frequencies of asymmetric wall-frame multi-storey building structures.
J Sound Vib 2000;236(1):14160.
[7] Zalka KA. A simplified method for calculation of the natural frequencies of
wall-frame buildings. Eng Struct 2001;23(12):154455.
[8] Zalka KA. Global structural analysis of buildings. London: E&FN Spon; 2001.
[9] Potzta G, Kollar LP. Analysis of building structures by replacement sandwich
beams. Internat J Solids Struct 2003;40:53553.
[10] Tarjan G, Kollar LP. Approximate analysis of building structures with identical
stories subjected to earthquakes. Internat J Solids Struct 2004;41:141133.
[11] Skattum SK. Dynamic analysis of coupled shear walls and sandwich beams.
California Institute of Technology; 1971.
[12] Basu AK. Seismic design charts for coupled shear walls. J Struct Eng ASCE 1983;
109(2):33552.
[13] Rosman R. Stability and dynamics of shear wall frame structures. Build Sci
1974;9:5563.
[14] Rutenberg A. Approximate natural frequencies for coupled shear walls. Earthq
Eng Struct Dyn 1975;4:95100.
[15] Smith BS, Crowe E. Estimating periods of vibration of tall buildings. J Struct Eng
ASCE 1986;112(5):100519.
[16] Smith BS, Yoon YS. Estimating seismic base shears of tall wall-frame buildings.
J Struct Eng ASCE 1991;117(10):302641.
[17] Kopecsiri A, Kollar LP. Approximate seismic analysis of building structures by
the continuum method. Acta Tech Acad Sci Hung 1999;108(34):41746.
[18] Kopecsiri A, Kollar LP. Simple formulas for the analysis of symmetric (plane)
bracing structures subjected to earthquakes. Acta Tech Acad Sci Hung 1999;
108(34):44773.
[19] Rafezy B, Zare A, Howson WP. Coupled lateraltorsional frequencies of
asymmetric, three dimensional frame structures. Internat J Solids Struct 2007;
44(1):12844.
[20] Howson WP. A compact method for computing the eigenvalues and
eigenvectors of plane frames. Adv Eng Softw Workstations 1979;1(4):18190.
[21] Williams FW, Wittrick WH. An automatic computational procedure for
calculating natural frequencies of skeletal structures. Int J Mech Sci 1970;12:
78191.
[22] Wittrick WH, Williams FW. A general algorithm for computing natural
frequencies of elastic structures. Quart J Mech Appl Math 1971;24(3):26384.
[23] Howson WP, Williams FW. Natural frequencies of frames with axially loaded
Timoshenko members. J Sound Vib 1973;26(5):50315.
[24] Rafezy B, Howson WP. Exact dynamic stiffness matrix of a three-dimensional
shear beam with doubly asymmetric cross-section. J Sound Vib 2006;289:
93851.
[25] Wilson EL, Hollings JP, Dovey HH. ETABS version 6, three-dimensional analysis
of building structures. 6 ed. Berkeley (CA); 1995.