Proceedings of the Institution of

Civil Engineers
Structures & Buildings 159
August 2006 Issue SB4
Pages 217–228

Paper 14270
Received 19/05/2005
Accepted 22/03/2006
Keywords: Bassam A. Izzuddin
buildings, structures & design/ Reader in Computational
mathematical modelling/structural Structural Mechanics, Imperial
frameworks College, London, UK

Simplified buckling analysis of skeletal structures

B. A. Izzuddin MSc, PhD, DIC, CEng, MIStructE

A simplified approach is proposed for buckling analysis of deal with a new class of problem, it sheds new light on the
skeletal structures, which employs a ‘rotational spring buckling analysis of skeletal structures, enabling better
analogy’ for the formulation of the geometric stiffness understanding of the buckling mechanisms, and it provides a
matrix. The benefit of this analogy is that it offers an simplified and practical framework for buckling predictions,
intuitive framework, which is based on the common importantly, using linear analysis principles.
notions of linear structural analysis. Assuming that the
structural deflections prior to buckling are negligible, a As mentioned above, buckling can be related to the singularity
linear eigenvalue problem, utilising the geometric and of the tangent stiffness matrix, which in turn consists of two
material stiffness matrices, can be easily formulated and parts. 6,7 The first part is the material stiffness matrix, which is
solved for the critical buckling loads. This can be further related to the deformational stiffness of the components, taking
simplified using an assumed mode, where the rotational into account the connectivity of components in the current
spring analogy is shown to provide considerable geometric configuration of the structure. For linear elastic
computational benefits and significant insight into the components, the material stiffness is identical with the linear
buckling of various forms of skeletal structure. In this elastic stiffness, but updating the structural geometry to
context, the use of different assumed modes can be include the effect of any displacements. The second part is the
conceived as a process of probing the structure to geometric stiffness matrix, which is related to the component
establish the most likely mode for buckling and the forces, and in some cases to the applied loading, taking into
corresponding critical load. It is also shown that the account the effect of a change in geometry from the current
approximation inherent in the assumed mode approach configuration. For typical structures, the material stiffness is
together with the discrete form of the rotational spring positive for all deformation modes, mathematically referred to
analogy can be significantly improved through modal as positive–definite, whereas the geometric stiffness can admit
combinations and increasing the number of elements, negative values for certain modes, depending on the
respectively, where convergence to the exact buckling component forces and applied loading. It is therefore the effect
solution is demonstrated. Several illustrative examples of a negative geometric stiffness that can lead to a singular
are provided in this paper, which highlight the simplicity overall tangent stiffness matrix, and hence buckling.
of the proposed approach, its application using a linear
structural analysis tool and its ability to shed significant The current paper presents a practical approach to the buckling
light on important issues in buckling analysis of skeletal analysis of skeletal structures, based on a ‘rotational spring
structures. analogy’ for formulating the geometric stiffness contribution.
The paper proceeds with describing the rotational spring
1. INTRODUCTION analogy, highlighting its benefits in relation to simplified
The buckling of various structural forms has attracted buckling analysis. After outlining the applicability of this
significant research interest over many years,1–5 since it analogy to detailed buckling predictions using iterative
normally provides a practical limit on the load-carrying techniques, its practical benefits in simplified buckling analysis
capacity of structures. The lowest buckling load is of most using an assumed mode are discussed in detail. In the latter
practical significance, and is normally achieved when the context, two approaches are proposed, based on individual
tangent stiffness associated with a mode of deformation probing modes and combined modes, respectively, both of
becomes zero, such a mode is then referred to as the buckling which maintain the practical benefits of the rotational spring
mode. Of course, numerous sophisticated procedures and analogy. Finally, the conceptual benefits of the new approach
computational tools have been developed over the past few are highlighted, especially in relation to the buckling
decades that deal with structural buckling, both in terms of interaction between various applied loads. Several examples
simplified linear eigenvalue analysis and through tracing the are provided throughout the paper, which demonstrate the
geometrically nonlinear response, in the latter case often practicality of the proposed approach and the conceptual
dealing with the influence of material nonlinearity as well. power it furnishes towards understanding the buckling of
While the approach proposed in the present paper does not skeletal structures.

Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin 217

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.
 2. ROTATIONAL SPRING ANALOGY Since, however, the local spring deformations diminish relative
The geometric stiffness of skeletal structures, subjected to to the global spring deformations as the number of elements is
conservative nodal loads that maintain a fixed direction, can increased (Fig. 3), the local distributed springs are ignored in
be determined from equivalent rotational springs using linear the proposed approach for simplified buckling analysis.
analysis principles, hence the term ‘rotational spring analogy’.
In the present context, the main source of geometric The above rotational spring analogy can be used for a variety
nonlinearity is considered to be the change to the path of axial of structural forms, including trusses, frames, cables, and link-
forces within the structure owing to a change in the deformed spring idealisations, for both two- and three-dimensional (2D
geometry. and 3D) structures. The most general application of this
analogy entails the formulation of the geometric stiffness
To illustrate this analogy, consider an axially loaded element matrix from the equivalent rotational springs using the
which remains straight, as shown in Fig. 1. The only first-order principle of virtual work, where only simple linear kinematics
change in forces required to maintain equilibrium under an are required, as discussed next for 2D and 3D analysis. A
infinitesimal rigid body rotation (
r) is represented by a couple significant benefit of the rotational spring analogy, however, is
proportional to the axial force (F) and the element length (L), that buckling analysis, in its most simplified form, may be
and hence this geometric stiffness is equivalent to a rotational undertaken without the formulation of a geometric stiffness
spring with a stiffness matrix, as elaborated in section 4.

1 kr ¼ FL 2.1. 2D analysis
In 2D analysis, the contribution of an equivalent rotational
spring to the geometric stiffness matrix is determined by the
Clearly, the equivalent rotational spring is stabilising (kr . 0: ) linear kinematic relationship between the spring rotation and
for a tensile axial force (F . 0) and is conversely destabilising the four translational freedoms
(kr , 0) for a compressive axial force (F , 0).

s c
When the element is prone to bending owing to buckling, as 2a r¼ ð U2
U1 Þ þ ð V 2
V 1 Þ ¼ Tr U
L L
shown in Fig. 2, the equivalent rotational springs become
distributed over the length. This effect, however, is equivalent
to a single discrete rotational spring connected to the element
with
chord, as given by equation (1), in addition to local distributed
rotational springs connecting the element chord to the
deformed element reference line (Fig. 2). These two sets of 2b U ¼ h U1 V1 U2 V2 iT
equivalent spring are responsible for global and local
geometric nonlinearity, or the P–˜ and P–
effects, 8
respectively, and both can be considered in the formulation of
1
the geometric stiffness matrix using linear analysis principles. 2c Tr ¼ h s
c
s ci
L

Fär F
F 2d c ¼ cos (Æ); s ¼ sin (Æ)
är

L ,
55. k r 5 FL where L ¼ corresponding element length and Æ ¼ element
angle with global X-axis, as shown in Fig. 4(a).
F
Fär F Using the principle of virtual work, the contribution of the
equivalent spring to the geometric stiffness matrix is simply
obtained as
Fig. 1. Equivalent geometric stiffness for straight element

F
F F
är 5 0
är
 äè Þ 0 , |är|
|äè| ,
äè
One element Two elements
L ,
55. k r 5 FL 1 kè 5 F

F F F

Fig. 2. Equivalent geometric stiffness for bending element Fig. 3. Reduction of local rotations with refinement

218 Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.
 Y, V
V2 U2

ry
Y, V kr
cy W2
V2

rz
W1
r U2
cz
X, U
kr U1
V1
U1 á
L

V1

X, U

Z, W

(a) 2D analysis (b) 3D analysis

Fig. 4. Linear kinematics of equivalent rotational springs

 
3 K G ¼ kr TrT Tr 1
c y,1
c y,2
c y,3 c y,1 c y,2 c y,3
5c Tr ¼
L
c z,1
c z,2
c z,3 c z,1 c z,2 c z,3

with kr given by equation (1). If the structural deflections up to where (c y , c z ) ¼ direction cosines of the transverse local
buckling are negligible, then L and Æ correspond to the initial element axes.
undeformed configuration, although equation (3) is equally
applicable for the case involving significant pre-buckling Again, using the principle of virtual work, the contribution of
deflections, in which case F, L and Æ must be obtained for the the equivalent spring to the geometric stiffness matrix is
deformed configuration. simply obtained as

Importantly, in the context of simplified buckling analysis, it is

often sufficient to evaluate the quadratic work associated with 6 K G ¼ kr TrT Tr
the geometric stiffness, which is simply represented by

4 U T K G U ¼ kr U T TrT Tr U ¼ kr r2 and its contribution to the quadratic work is represented by

 
7 U T K G U ¼ kr U T TrT Tr U ¼ kr r2y þ r2z
Clearly, in such a case, the rotational spring analogy offers
considerable simplification, where the contribution of the
equivalent springs to quadratic work is obtained from equation
(4) with the spring rotations r corresponding to U determined where r y , r z are readily obtained from equation (5).
readily from equation (2).

3. BUCKLING ANALYSIS
2.2. 3D analysis
Consider a structure subjected to a combination of constant
In 3D analysis, each axial component is associated with two
initial loads (P0 ) and proportional loads consisting of nominal
orthogonal equivalent rotational springs, as shown in Fig. 4(b),
loads (Pn ) varying according to a common load factor (º).
where the corresponding rotations are determined as
Neglecting pre-buckling deflections and assuming a linear

 elastic material response, the axial forces within the members
ry similarly consist of initial constant forces (F o ) and nominal
5a ¼ Tr U
rz forces (F n ), and the stiffnesses of the corresponding equivalent
rotational springs are obtained from equation (1) as (kro ) and
(krn ), respectively. This leads to contributions that can be
with assembled into two corresponding geometric stiffness matrices,
(K oG ) and (K Gn ), which combine with the material stiffness
5b U ¼ h U1 V1 W1 U2 V2 W2 iT matrix (K E ) to determine the overall tangent stiffness matrix as
follows

Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin 219

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.
 using equation (1), and the corresponding rotations owing to
8 K T ¼ K E þ K oG þ º K Gn
the considered mode (U) as determined from equations (2) or
(5).

where (K oG , K Gn ) can be obtained from (kro , krn ), respectively, as The prediction of the lowest critical load factor in equation (12)
discussed in sections 2.1 and 2.2. may be improved by revising the approximate mode using, for
example, the method of iteration. 9 With this method, the load
Buckling analysis is concerned with the solution of the pattern is iteratively revised after each estimate of U and º c as
following linear eigenvalue problem for the critical load factor the forces resulting from the geometric stiffness
(º ¼ ºc ) and associated buckling mode (U)
 
  13 P ¼
K oG þ º c K Gn U
9 K T U ¼ K E þ K oG þ º K G U ¼ 0 n
ðU 6¼ 0, º ¼ º c Þ

leading to new values for U and ºc using equations (11) and

If the buckling mode (U) is known, then the critical load factor (12).
can be obtained from pre-multiplying equation (9) by U T as
follows One disadvantage of the method of iteration is that it requires
in equation (13) the assembly of the geometric stiffness
U T K E U þ U T K oG U matrices, K oG and K Gn , or at least the corresponding geometric
10 ºc ¼
forces. An alternative, yet approximate, approach is proposed
U T K Gn U
in section 4, which allows an improvement of the initial
buckling prediction using the rotational spring analogy without
In practice, however, the buckling mode is unknown, although the need for assembly of the geometric stiffness matrix.
an approximate mode may be determined from linear analysis
using some load pattern (P)
3.1. Example: column with clamped–pinned ends
Consider an elastic column with one end clamped and the other
11 P ¼ K E U ) U ¼ K
1
E P pinned, as shown in Fig. 5(a), where buckling analysis is
undertaken using the rotational spring analogy with iteration.
With four elements, four equivalent rotational springs are
Of course, when the mode is approximate the prediction in
employed to represent the geometric stiffness, where the spring
equation (10) becomes also approximate, although always
properties are provided in Fig. 5(b). Clearly, and as easily
providing an upper bound on the lowest critical load factor for
verified from equation (2), the equivalent spring rotations are
typical structures that are stable in the unloaded state (i.e. KT
dependent only on the transverse deflections of the three
positive definite for º ¼ 0).
internal nodes. Similarly, the only forces arising from the
geometric stiffness of the equivalent springs in equation (13)
In view of equations (4) and (7), the rotational spring analogy
are transverse forces at the three internal nodes. The proposed
simplifies the expression in equation (10) such that the
buckling approach can therefore be applied with a linear
assembly of the geometric stiffness matrices becomes
unnecessary
EI
Pn 5 , L 5 3 m, EI 5 3 3 106 Nm2
X
m L2
o
UT P þ kr,i r2i
k E þ k oG i¼1
12a ºc ¼
¼
(for 2D analysis) U
k Gn X
m
n
r2i 1 2 3
kr,i ëP n
i¼1

X
m  
o
UT P þ kr,i r2y,i þ r2z,i (a) Problem characteristics
kE þ k oG i¼1
ºc ¼
¼
 
k Gn X
m
12b n
kr,i r2y,i þ r2z,i
i¼1
(for 3D analysis) L n
Fni 5 2P n, Li 5 , k 5 Fni Li, kro,i 5 0 (i 5 1, 4)
4 r ,i

where m ¼ number of equivalent rotational springs. 1 2 3 4

Clearly, the term (k E ¼ U T P) represents the work done by the

considered load pattern, which can be easily obtained using a (b) Equivalent rotational springs
linear analysis tool. The remaining terms simply represent the
energy stored or imparted by the equivalent rotational springs, Fig. 5. Elastic column with clamped-pinned ends
involving the spring stiffnesses (kro , krn ) as obtained for (F o , F n )

220 Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.
 analysis tool that models the response of the elastic column
under three arbitrary transverse forces at the internal nodes,
Iteration 1 Iteration 4
although the use of inverse iteration requires also the assembly
of the geometric stiffness matrix as discussed previously.

In applying inverse iteration, the starting mode is obtained

from a transverse load applied to the middle node, where in all
iterations the load pattern and the corresponding mode are
identically scaled to normalise the work done (i.e.
UT P ¼ 1 Nm). The rotations of the equivalent springs (rr) are
easily determined from U using equation (2), and the estimated
critical load factor (ºc ) is simply obtained from equation (12a).
Table 1 provides the iterative estimates, and Fig. 6 depicts the
starting/final modes, where it is clear that convergence to the
lowest critical load factor (ºc ¼ 22.36) is achieved after only
four iterations. This prediction overestimates the exact
theoretical solution (º exact
c ¼ 20:19) by around 11%, which is
caused by the neglect of the local geometric stiffness.
Considering Fig. 6, the significance of the local geometric
Fig. 6. Lowest buckling mode for elastic column using four
stiffness for the predicted mode is manifested in the elements
discrepancy between the element chords (dotted line) and the
column reference line (solid line), where it is clear that
improved accuracy is most effectively achieved with more
elements next to the clamped support and near midspan.
Iteration 1 2 3 4
Following on from the last point, the use of eight elements
with the same starting load pattern leads to the iterative ºc 22.14 20.86 20.74 20.73
estimates given in Table 2, where the converged value
(º c ¼ 20:73) overestimates the exact solution by only 2.7%. Table 2. Critical load factor for elastic column using eight
elements
This improved accuracy is reflected in Fig. 7 by a smaller
discrepancy between the element chords and the column
reference line in comparison with the previous case using four
elements. Clearly, the proposed simplified buckling approach
converges to the exact solution as the number of elements is In this context, the rotational spring analogy offers
increased, although the neglect of the local geometric stiffness, considerable simplification, where only a linear analysis model
which is equivalent to local distributed rotational springs that of the skeletal structure is required
are all negative, leads to convergence from above.
(a) to determine the axial forces (F o , F n ) corresponding to the
4. SIMPLIFIED APPROACH initial and nominal loads (Po , P n ), hence establishing the
Considering a specific mesh of elements, when the starting stiffnesses of the equivalent rotational springs (kro , krn )
mode is only an approximation of the actual buckling mode (b) to determine the mode (U) associated with the load pattern
that satisfies equation (9), inverse iteration may be used to (P), from which the corresponding spring rotations are
converge to the lowest buckling solution. It is often possible, obtained according to equation (2) or (5), leading to the
however. to select a load pattern (P) leading to a deflection estimation of the buckling load factor (ºc ) using equation
mode (U) such that the starting buckling prediction is already (12).
of sufficient accuracy, thus avoiding the need for inverse
iteration. For example, Stevens 10 suggested that a reasonable Clearly, since the load pattern is assumed, there is no need with
approximation for multi-storey frames may be achieved with a the rotational spring analogy for assembly of the geometric
load pattern consisting of horizontal forces that are directly stiffness matrix, though the result is an approximate ºc which
proportional to the applied vertical loading at various floor typically overestimates the lowest buckling solution of the
levels. eigenvalue problem in equation (9). The consideration of

Iteration P: kN U: mm r: mrad ºc

1 h0 3:492 0iT k0.1279 0.2864 0.2199lT k0.1705 0.2114
0.0887
0.2932lT 23.87
2 h
0:3181 2:333 1:591iT k0.1163 0.2799 0.2414lT k0.1551 0.2181
0.0513
0.3219lT 22.49
3 h
0:4718 2:016 2:024iT k0.1115 0.2756 0.2456lT k0.1486 0.2188
0.0400
0.3274lT 22.37
4 h
0:5237 1:930 2:143iT k0.1099 0.2741 0.2466lT k0.1466 0.2190
0.0368
0.3288lT 22.36

Table 1. Critical load factor for elastic column using four elements

Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin 221

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.
 ( ) 2 3
U TA U TA PA U TA PB
Iteration 1 Iteration 4 15 kE ¼ K E h UA UB i ¼ 4 5
U TB U TB PA U TB PB

On the other hand, the reduced 2 3 2 geometric stiffness

matrices are easily determined using the rotational spring
analogy, and are obtained in terms of the energy stored/
imparted by the equivalent rotational springs over the rotations
corresponding to U A and U B

( )
oj n U TA oj n
16a kG ¼ K G h UA UB i )
U TB

2X
m X
m 3
oj n oj n
6 kr,i r2A,i kr,i r A,i r B,i 7
6 i¼1 i¼1 7
k
16b G
oj n
¼ 6 7(2D analysis)
6X X oj n 7
4 m oj n
m
2 5
Fig. 7. Lowest buckling mode for elastic column using eight kr,i r B,i r A,i kr,i r B,i
elements i¼1 i¼1

2 ! 3
X
m
oj n
r2yA,i þ
6 kr,i 7
6 i¼1 r2zA,i 7
various reasonable load patterns and the selection of the 6 7
6 ! 7
6 X oj n r yA,i r yB,i þ 7
smallest ºc , however, often enables a good estimate of the 6 m
7
6 kr,i 7
lowest buckling load factor. This can be visualised as a probing 6 r zA,i r zB,i 7
16c k oj n 6 i¼1 7
process, where the structure is subjected to various G ¼6
6 m ! 7(3D analysis)
7
disturbances (U) with the objective of determining the most 6 X oj n r yB,i r yA,i þ 7
6 k 7
6 r,i 7
likely buckling mode associated with the smallest ºc . For each 6 i¼1 r zB,i r zA,i 7
6 7
probing mode (U), the value of ºc is most easily conceived as 6 2 ! 7
6 Xm r þ 7
that for which the energy stored in the linear structure, 4 oj n
kr,i
yB,i 5
1 T r2zB,i
2 (U K E U), is identical to the energy imparted by the equivalent
i¼1

rotational springs, 12 (
U T [K oG þ º c K Gn ]U).

Of course, the solution of the reduced eigenvalue problem in

4.1. Improvement of assumed mode approximation
equation (14a) is straightforward, where an improved estimate
Despite the practical benefits of the rotational spring analogy
of the lowest buckling load factor (ºc ) is easily obtained from
with the assumed mode approach, there are cases when it is
the condition (det(k T ) ¼ 0) representing a quadratic equation.
difficult to establish a sufficient number of probing modes to
converge on the lowest buckling load factor with reasonable
Importantly, the above simplified approach allows reasonably
accuracy. Such cases benefit from a practical modification,
accurate predictions of the lowest ºc using the rotational spring
where two previously considered modes, U A and U B , are
analogy with a linear analysis tool, without the need for
combined to formulate a rank 2 reduced eigenvalue problem9
assembly of the overall geometric stiffness matrices (K oG , K Gn ).
of the following form
Obviously, the accuracy of these predictions depends on the
extent to which modes U A and U B can, in combination,
  represent the lowest buckling mode. The use of a few
14a k T u ¼ k E þ k oG þ º k Gn u ¼ 0 ðu 6¼ 0, º ¼ º c Þ
representative probing modes for the selection of U A and U B ,
however, often produces very good estimates of the lowest ºc .
with
Finally, the most effective mode U B which is complementary to
a probing mode U A , leading to the closest approximation of the
14b U ¼ h UA UB i u lowest º c from the reduced problem, can be shown to
correspond to the following load pattern

where u ¼ hu1 u2 iT represents the two weighting parameters of h   i

the reduced problem. 17a P B ¼
P A þ K oG þ º cA K Gn U A

In performing problem reduction, the reduced 2 3 2 material

stiffness matrix is simply given in terms of the work done by 17b U B ¼ K
1
E P
B

the load patterns, P A and P B , over their respective modes

222 Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.
 Clearly, the determination of the above mode requires the iteration, where (º exact
c ¼ 0:9765) and the corresponding mode
assembly of the geometric stiffness matrices, or at least the is shown in Fig. 9(a).
forces resulting from the equivalent rotational springs, and
hence suffers from the same shortcoming of the method of Simplified buckling analysis is first undertaken by considering
inverse iteration. However, the use of the two modes, U A and three probing modes, (I, II, III), corresponding to horizontal
U B , within a reduced eigenvalue problem leads to much faster loads applied at three different nodes, as illustrated in Figs 9(b)
convergence than with standard inverse iteration as discussed to (d). Table 3 provides the load patterns, nodal displacements
in section 3. Typically with the reduced problem, consisting of and rotations of the equivalent springs for the three probing
a probing mode U A and a related mode U B using equation (17), modes, normalised such that (UT P ¼ 1 Nm). The buckling load
the value of ºc obtained as the solution of equation (14a) factor (ºc ) is estimated for each of the modes directly from
provides a very good approximation of the lowest buckling equation (12a), where as expected all predictions overestimate
load factor. It is, however, possible to refine the prediction º exact
c , although probing mode (I) offers an excellent
further with the reduced problem, where the mode U obtained approximation to within 1.1%.
from equation (14b) may be used as U A , with U B determined
again from equation (17), for a subsequent iteration. Next, improved buckling prediction is considered by
combining any two of the three previous probing modes in a
reduced problem, where the reduced material and geometric
stiffness matrices are obtained directly from equations (15)
4.2. Example: frame subject to vertical loading and (16), as given in Table 4. The solution of the reduced
Consider an elastic plane frame subject to proportional vertical problem, expressed by equation (14), leads to improved
loading applied directly to the columns, as shown in Fig. 8(a), estimates of the lowest ºc and the corresponding weighting
where two elements are used to represent each column. With factors of the two combined modes (u). Considering the
negligible axial forces in the beams, and neglecting pre- results in Table 4, it is clear that among the three possible
buckling deflections, the geometric stiffness is represented by combinations, combining modes (I=II) leads to the greatest
ten equivalent rotational springs in the columns, as illustrated improvement in simplified buckling prediction, now to within
in Fig. 8(b), where the nominal spring stiffnesses are provided 0.4%. On the other hand, combinations involving mode (III)
in the same figure. Clearly, the rotations of these springs with modes (I) or (II) lead only to a marginal improvement
depend only on the horizontal displacements of ten nodes, and in the single-mode predictions. If, however, related mode
therefore only these components, denoted by U, need to be (III B ) is obtained according to equation (17) from mode (III),
extracted from linear analysis under probing load patterns. considered hence as mode (III A ), a near-perfect prediction is
achieved to within 0.02%. Of course, the only shortcoming of
For comparison purposes, the exact lowest buckling mode for the last combination (III A =III B ) is that it requires the
the frame is determined using a sufficient number of inverse assembly of forces resulting from the equivalent rotational
springs owing to mode (III A ), whereas the other combinations
have no such requirement.

L 5 3 m, P n 5 EI0 /L2
EA0 5 4 3 109 N, EI0 5 3 3 106 Nm2  n
Finally, it is noted that the relatively coarse discretisation with
9 ëP ëP n 10
Columns: EA 5 EA0, EI 5 EI0 ten column elements leads to further inaccuracy in the above
Beams: EA 5 0·5 EA0, EI 5 2EI0 7 predictions owing to the neglect of local geometric
8 L
n
ëP n
nonlinearity, the extent of which is manifested in the small
4 ëP n
5 ë(2P ) 6
discrepancy between the element chords and reference lines for
Y 1 2 3 the lowest buckling mode, as shown in Fig. 9(a). This, however,
L
results in only a 3.4% discrepancy in the lowest buckling load
X, U
factors between the adopted coarse discretisation
2L 2L (º coarse
c ¼ 0:9765) and a very fine discretisation (º cfine ¼ 0:944),
thus highlighting the practicality of the proposed simplified
(a) Problem characteristics
buckling approach.

5. CONCEPTUAL BENEFITS
The buckling of conservative structural systems under static
2P n (i = 1; 4; 7, 10)
F ni 5 23P n (i 5 2; 5) loading is defined by the singularity of the tangent stiffness
22P n (i 5 3; 6) matrix (K T ), which consists of the material stiffness (K E ) and
L 10 the geometric stiffness (K G ), where the latter is directly
Li 5 , knr,i 5 FniLni, kor,i 5 0 (i 5 1, 10) 9
2 proportional to the applied loading and the equilibrating
7 8
internal forces. For typical skeletal structures, buckling occurs
4 5 6 at a load level that produces a sufficiently ‘negative’ K G that
1
overcomes the ‘positive’ K E in a specific deformation mode,
2 3
leading to a singular K T and the presence of infinitesimally
(b) Equivalent rotational springs adjacent equilibrium states.

Fig. 8. Elastic frame subject to vertical loading The relative influence of K E and K G for a specific mode can be
established in terms of the quadratic energy which they store

Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin 223

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.
 ëc 5 0·9765 ëIc 5 0·9874

(a) Exact buckling mode (b) Mode (I)

ëIIc 5 1·0656 ëIII

c 5 1·2650

(c) Mode (II) (d) Mode (III) and related mode

Fig. 9. Buckling and probing modes for elastic frame

Mode P: kN U: mm r : mrad ºc

(I) P4 ¼ 0.8451 k0.7280 0.7787 0.7413 k
0.4853
0.5191
0.4942 0.9874

Pi ¼ 0 (i 6¼ 4) 1.1832 1.1814 1.1805
0.3035
0.2685
0.2928
1.2737 1.3058
0.0615
0.0835
1.3392 1.3393lT
0.0436
0.0223lT
(II) P2 ¼ 1.1826 k0.6618 0.8456 0.6815 k
0.4412
0.5637
0.4544 1.0656
Pi ¼ 0 (i 6¼ 2) 1.0896 1.0902 1.0893
0.2852
0.1631
0.2718
1.1321 1.1906
0.0279
0.0675
1.1938 1.1940lT
0.0411
0.0023lT
(III) P9 ¼ 0.6369 k0.6274 0.6429 0.6026 k
0.4183
0.4286
0.4017 1.2650
Pi ¼ 0 (i 6¼ 9) 1.0092 1.0099 1.0101
0.2545
0.2446
0.2717
1.2872 1.3231
0.1849
0.2087
1.5701 1.5693lT
0.1886
0.1641lT

Table 3. Probing modes for elastic frame

Modes (A/B) kE : Nm knG : Nm u ºc

   

1:0 0:9209
1:0127
0:9637 0:7809
(I/II) 0.9804
0:9209 1:0
0:9637
0:9384 0:2334
   

1:0 0:8529
1:0127
0:8708 0:9742
(I/III) 0.9872
0:8529 1:0
0:8708
0:7905 0:0301
   

1:0 0:7603
0:9384
0:8209 0:7330
(II/III) 1.0269
0:7603 1:0
0:8209
0:7905 0:3221
   

1:0 0:0
0:7905
0:3836 0:8453
(III A /III B ) 0.9767
0:0 1:0
0:3836
0:3932 0:5197

Table 4. Combined modes for elastic frame

224 Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.
 and impart, respectively, where the sum indicates the nature of (a) If all axial forces owing to the incremental initial/nominal
the overall structural response to an external disturbance. Prior loads are compressive, the corresponding additional
to buckling, the negative influence of K G is not sufficient to equivalent springs will all be negative. Therefore, the
overcome the positive resistance of K E for any mode, hence K T increment in quadratic energy will either be zero resulting
is positive definite and structural equilibrium is thoroughly in (º cP ¼ º Oc ), if none of the additional equivalent springs
stable. At buckling, the negative influence of K G becomes are rotated owing to the original mode, or more typically
equal to the positive resistance of K E for a specific buckling negative leading to (º cP , º Oc ). Therefore, the exact and
mode, hence K T becomes positive semi-definite and predicted values can in general be related for this case
equilibrium becomes critical. Finally, when the negative according to (º cN < º cP < º Oc ).
influence of K G exceeds the positive resistance of K E for some (b) If all axial forces owing to the incremental initial/nominal
modes, K T admits negative values and equilibrium becomes loads are tensile, the corresponding additional equivalent
thoroughly unstable. springs will all be positive. Therefore, the increment in
quadratic energy will either be zero resulting in (º cP ¼ º Oc ),
For elastic structures with negligible pre-buckling deflections, if none of the additional equivalent springs are rotated
buckling along the fundamental equilibrium path can be owing to the original mode, or more typically positive
expressed as a linear eigenvalue problem, where K E remains leading to (º cP . º Oc ). It can be shown that the exact and
constant and K G is linearly dependent on the loading. Of predicted values are generally related for this case
course, buckling predictions based on an assumed mode, as according to (º Oc < º cN < º cP ).
discussed in section 4, cannot by definition correspond to a (c) If the incremental axial forces are of different signs, the
positive definite K T . Therefore, for such a structure, assumed same will apply to the additional equivalent springs.
mode predictions lead to K T which admits negative values, and Therefore, the increment in quadratic energy may be
hence the lowest buckling load is overestimated, as previously evaluated as negative, zero or positive, depending on the
indicated. The only exception is when the assumed mode is relative contributions of the equivalent springs over the
identical with the buckling mode, in which case the exact original mode, leading to (º cP , º Oc ), (º cP ¼ º Oc ) or (º cP . º Oc ),
buckling load is obtained, and the resulting K T is positive semi- respectively. Clearly, for the two former scenarios, the
definite. exact and predicted values are related according to
(º cN < º cP < º Oc ), while for the latter scenario no such
Now, the use of the rotational spring analogy brings significant ordering in possible, only that (º cN < º cP , º cP . º Oc ).
conceptual benefits, whether it is employed for formulating K G Typically, however, º cP offers a good estimate of º cN ,
in detailed buckling analysis or for establishing the especially for relatively small changes to the load
corresponding quadratic energy in simplified buckling analysis. distribution for which the original buckling mode remains
Since the stiffness of an equivalent rotational spring is reasonably accurate.
proportional to the axial force and length of the represented
element, the greater the compressive axial force, element
length and the spring rotation corresponding to a given mode, Finally, another conceptual benefit of the rotational spring
the greater its contribution to the negative influence of K G for analogy relates to the significance of neglecting local
that mode, as reflected by the quadratic energy contribution. geometric nonlinearity. Along similar lines of argument as
Therefore, buckling can be conceived as a phenomenon in above, if local geometric nonlinearity is ignored in elements
which the negative equivalent springs can just about apply which are all compressive, this corresponds to the neglect of
enough moment couples in a buckling mode to overcome the distributed equivalent rotational springs which are all negative.
material resistance, and possibly that of other positive This in turn leads to an underestimation of the negative
equivalent springs, such that equilibrium can be maintained to quadratic energy and, consequently, to an overestimation of
a second order. the lowest buckling load. On the other hand, if the neglected
local geometric nonlinearity is only for elements which are
Perhaps the greatest conceptual benefit of the rotational spring tensile, then the lowest buckling load is underestimated. In any
analogy, however, arises in considering the buckling case, it is emphasised again that the influence of local
interaction between different loads or, alternatively, in geometric nonlinearity reduces as the number of elements is
establishing the relationship between buckling predictions for increased, owing to the reduction in the local rotations relative
related loads. To elaborate, consider a structure for which the to the element chord for a given mode, thus guaranteeing
lowest buckling load factor (º Oc ) and corresponding mode have convergence to the exact solution using the proposed
already been established under a specific loading. The simplified approach.
influence of a small variation in the initial/nominal load
distribution can be assessed, using the original mode, in terms
of the effect of the corresponding load increment on the
quadratic energy. This in turn depends on the corresponding
increment in the geometric stiffness, and can be simply 5.1. Example: space truss subject to vertical loading
evaluated in terms of the additional equivalent rotational Consider the elastic space truss, shown in Fig. 10(a), which is
springs due to the change in element axial forces. Given that, symmetric about its midspan. The buckling of the truss is to be
with the neglect of pre-buckling deflections, any prediction º cP investigated under three related load cases, as depicted in Figs
assuming the original mode is an upper bound on the new 10(a) to (c), where the exact ‘lateral–torsional’ buckling mode
lowest buckling load factor (i.e. º cN < º cP ), the following and the corresponding load factor are determined for load case
outcomes can be ascertained. (I) as follows (Fig. 11)

Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin 225

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.
 L53m EA0 5 109 N

2 ëP n P n 5 0·01EA0 P o 5 0·005EA0
ëP n
1 EA 5 2EA0 EA 5 EA0
5
4 Po
Po
8
ëP n ëP n
3 7

Po
6 Po
3L/4
9
L/2
Z

3L/4 X 5L/4

4L

(a) Truss configuration and load case (I) (b) Load case (II)

Pn
3P n 3P n ëP n Pn ë
ë ë ëP n ë 4
4 4 4

3 2

5 1

Pn

ë
Pn
2 ë
2

(c) Load case (III)

Fig. 10. Elastic space truss subject to vertical loading

h
6:230, 53:70, 6:267,
11:05, 54:64,
10:52,
1:929,
16:27, 1:537,

18a U ¼ 0:0, 274:6, 94:98, 0:0, 277:1,
99:30, 0:0,
15:63,
1:706,
(
m)
6:230, 53:70, 6:267, 11:05, 54:64,
10:52, 1:929,
16:27, 1:537i T

kE X
m  
18b º cI ¼
¼ 1:3645 k Gn (Nm) ¼ n
kr,i r2y,i þ r2z,i ¼
0:732 87
k Gn 19b
i¼1

In the above, kE and k Gn represent the quadratic energy stored

and imparted by the material and nominal geometric stiffness,
Now with load case (II), consider the influence of initial loads
respectively. Alternatively, they correspond to the reduced
applied directly to the four corner truss members, as shown in
material and nominal geometric stiffness, and are obtained
Fig. 10(b). These lead to initial axial forces (F o ) in the various
with reference to equation (12b) as
truss members, the contribution of which to the geometric
stiffness is determined in terms of initial equivalent rotational
19a kE (Nm) ¼ U T P ¼ 1 springs (kro ). The corresponding quadratic energy associated
with the original mode in equation (18a) is represented by

226 Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.
 22b ˜krn (MNm) ¼ h
1:25 6:25 6:25 i T
ëc 5 1·3645

r y (mrad) ¼ h 0:129 52 0:130 10 0:128 98 i T ;

22c r z (mrad) ¼ 0

X
m  
22d ˜k Gn (Nm) ¼ ˜kr,i
n
r2y,i þ r2z,i ¼ 0:188 80
i¼1

Therefore, in view of equation (12b), the lowest buckling load

factor is estimated as

kE
23a º cIII 
¼ 1:838 0
k Gn þ ˜k Gn
Fig. 11. Buckling mode of truss under load case (I)
compared with the exact value

X
m  
20 k oG (Nm) ¼ o
kr,i r2y,i þ r2z,i ¼ 0:010 984 23b º cIII ¼ 1:766 4
i¼1

The above increase in the buckling load factor (i.e. º cIII . º cI ) is

and hence the lowest buckling load factor is estimated from clearly caused by the relatively large incremental tensile forces
equation (12b) as in elements 2 and 3 in comparison with the compressive force
of element 1, as obtained in equation (22a). Considering the
stiffness of the additional equivalent springs in equation (22b),
kE þ k oG
21a º cII 
¼ 1:379 5 and with the corresponding rotations due to the original mode
k Gn
being almost identical, the incremental quadratic energy in
equation (22d) is positive, leading to increased resistance to the
original buckling mode. Importantly, the prediction of º cIII in
compared with the exact value
equation (23a) using the original mode offers a good
assessment to within 4%, again reflecting that the new mode is
21b º cII ¼ 1:375 5 similar to the original mode.

6. CONCLUSION
Interestingly for load case (II), the application of initial vertical This paper presents a simplified approach for buckling analysis
loading increases the buckling load factor (i.e. º cII . º cI ), which of skeletal structures, which adopts a rotational spring analogy
results from the positive additional quadratic energy in to assess the influence of the geometric stiffness, enabling the
equation (20). Although the initial vertical loading causes use of common notions from linear structural analysis. The
compression in the vertical members, it also causes significant proposed approach combines simplicity with accuracy, thus
tension in the top horizontal chords, thus leading to positive providing a practical framework for buckling analysis. It also
equivalent rotational springs that restrain the lateral buckling offers significant conceptual benefits, shedding considerable
of these chords. Clearly, the prediction of º cII in equation (22a) light that can aid in the teaching and understanding of
using the original mode offers a very good assessment of the structural buckling.
effect of initial loading to within 0.3%, reflecting that the new
mode is similar to the original mode. The rotational spring analogy is first described, and the
formulation of the geometric stiffness matrix from equivalent
Finally, load case (III) considers the effect of applying part of discrete springs is discussed for 2D and 3D analysis. It is noted,
the nominal vertical load on the lower chord, which can be however, that the main practical benefit of the rotational
seen in terms of an additional self-equilibrating set of nominal spring analogy arises in the context of simplified buckling
loads, as demonstrated in Fig. 10(c). This additional set of analysis. In this context, only the quadratic energy stored/
nominal loads leads to axial forces in only three members, imparted by the equivalent springs owing to a deformation
highlighted in Fig. 10(c) with a thicker line, where the mode is required, while the assembly of the geometric stiffness
corresponding equivalent springs, spring rotations due to the matrix becomes unnecessary.
original mode and quadratic energy are obtained as follows
Detailed buckling analysis is then outlined, where the
22a ˜F (MN) ¼ h
0:833 33
n
2:635 2 2:635 2 i T applicability of the rotational spring analogy to buckling
predictions using a linear structural analysis tool is

Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin 227

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.
 highlighted. It is shown that iterative methods, such as inverse significant conceptual benefits, however, arise in considering
iteration, may be used to establish the exact lowest buckling the buckling interaction between different loads, where it is
mode, although this requires the assembly of the geometric shown that the rotational spring analogy offers a powerful
stiffness matrix, or at least the resulting forces owing to an means to predict and explain the change in the lowest buckling
arbitrary deformation mode. In this context, the rotational load factor owing to additional initial/nominal loading.
spring analogy does not offer major practical benefits, apart
from the ease of formulation of the geometric stiffness. Several illustrative examples are provided throughout the paper
to demonstrate the practical framework and conceptual power
Simplified buckling analysis using an assumed mode is offered by the rotational spring analogy for buckling analysis
subsequently discussed, where it is noted that the approximate of skeletal structures. It is believed that this new view of
predictions represent an upper bound on the exact lowest structural buckling can play an important role in providing
buckling load. For this type of analysis, the rotational spring simplified design-oriented tools and explaining the results of
analogy offers a very simple framework, where only the more detailed buckling analysis.
rotations of the equivalent springs corresponding to the
assumed mode are required. These are directly used to establish
the quadratic energy owing to the geometric stiffness, enabling REFERENCES
the approximation of the lowest buckling load factor with 1. TIMOSHENKO S. P. and GERE J. M. Theory of Elastic Stability,
relative ease. McGraw Hill, New York, 1961.
2. NETHERCOT D. A. and ROCKEY K. C. A Unified approach to
Two approaches are then proposed for simplified buckling the elastic lateral buckling of beams. The Structural
analysis. The first approach involves the consideration of Engineer, 1971, 49, No. 7, 321–330.
individual candidate modes, corresponding to individual 3. HORNE M. R. An approximate method for calculating the
probing load patterns, and the selection of the mode which elastic critical loads of multi-storey plane frames. The
achieves the lowest buckling load. The second approach is a Structural Engineer, 1975, 53, No. 6, 242–248.
refinement that combines two probing modes in a reduced 4. ALLEN H. G. and BULSON P. S. Background to Buckling,
eigenvalue problem, enabling a more accurate prediction of the McGraw Hill, New York, 1980.
lowest buckling load. Both approaches maintain the practical 5. BRADFORD M. A. Lateral-distortional buckling of steel
benefits of the rotational spring analogy in that the assembly I-section members. Journal of Constructional Steel
of the geometric stiffness matrix is not required. Research, 1992, 23, No. 1–3, 97–116.
Notwithstanding, an even more accurate version of the latter 6. CRISFIELD M. A. Nonlinear Analysis of Solids and
approach is proposed, where the second probing mode is Structures, Vol. 1, John Wiley & Sons, London, 1991.
directly related to the first, although this modification requires 7. BELYTSCHKO T., LIU W. K. and MORAN B. Nonlinear Finite
the assembly of the geometric stiffness contributions. Elements for Continua and Structures. John Wiley & Sons,
London, 2000.
Finally, the conceptual benefits of the rotational spring 8. CHAN S. L. and ZHOU Z. H. Non-linear integrated design
analogy are highlighted. It is suggested that buckling can be and analysis of skeletal structures by 1 element per
conceived as a phenomenon driven by negative equivalent member. Engineering Structures, 2000, 22, No. 3, pp.
rotational springs capable of overcoming the positive stiffness 246–257.
of the material and other equivalent springs. When seen in 9. HUGHES T. J. R. The Finite Element Method, Prentice-Hall,
these terms, the influence of parameters such as the magnitude New York, 1987.
of compressive/tensile forces and the member lengths on the 10. STEVENS L. K. Elastic stability of practical multi-storey
buckling load and associated mode can be naturally assessed frames. Proceedings of the Institution of Civil Engineers,
using common principles of linear analysis. The most 1967, 36, 99–117.

What do you think?

To comment on this paper, please email up to 500 words to the editor at journals@ice.org.uk
Proceedings journals rely entirely on contributions sent in by civil engineers and related professionals, academics and students. Papers
should be 2000–5000 words long, with adequate illustrations and references. Please visit www.thomastelford.com/journals for author
guidelines and further details.

228 Structures & Buildings 159 Issue SB4 Simplified buckling analysis of skeletal structures Izzuddin

Downloaded by [ UNIV OF ALABAMA LIB/SERIALS] on [16/09/16]. Copyright © ICE Publishing, all rights reserved.