Journal of Constructional Steel Research 62 (2006) 605–613

www.elsevier.com/locate/jcsr

Buckling lengths of irregular frame columns

Konuralp Girgin ∗ , Gunay Ozmen, Engin Orakdogen
Faculty of Civil Engineering, Istanbul Technical University, Maslak 34469, Istanbul, Turkey

Received 27 September 2004; accepted 28 September 2005

Abstract

In several design codes and specifications, simplified formulae and diagrams are given for determining the buckling lengths of frame columns.
It is shown that these formulae may yield rather erroneous results, especially for irregular frames. This is due to the fact that the code formulae
utilise only local stiffness distributions. In this paper, a simplified procedure for determining approximate values for the buckling loads of both
regular and irregular frames is developed. The procedure utilises lateral load analysis of frames and yields errors on the order of 5%, which may
be considered suitable for design purposes. The proposed procedure is applied to several numerical examples and it is shown that all the errors are
in the acceptable range and on the safe side.
c 2005 Elsevier Ltd. All rights reserved.

Keywords: Buckling load; Buckling length; Effective length; Sway mode; Unbraced frames; Irregular frames; Isolated subassembly; Design codes

1. Introduction subassembly approach include modifications made by Duan

Determination of the buckling (effective) lengths of frame
columns is one of the significant phases of frame design.
Theoretically, the buckling length of an individual column
is determined by calculating the system-buckling load of the
frame. Since a full system instability analysis may be quite
involved for frames encountered in practical applications,
simplified formulae and diagrams are given for determining the
buckling lengths of frame columns in most of the design codes
and specifications [1,2]. The so-called “isolated subassembly
approach” of specifications was originally developed by
Galambos [3]. Similar formulae and diagrams exist in other
widely applied specifications such as Eurocode 3 and DIN
18800 [4,5].
A major limitation of the methods based on the isolated
subassembly approach is that they do not properly recognise
the interaction effects of adjacent elements other than those
in the immediate neighbourhood of the joints. Hellesland
and Bjorhovde have shown that this approach may result in
significant errors even for “regular” frames [6]. The errors
encountered in the case of “irregular” frames are even greater,
as will be shown. Efforts to improve the applicability of the
∗ Corresponding author.
E-mail address: kgirgin@ins.itu.edu.tr (K. Girgin).
and Chen [7,8] and an iterative procedure developed by Bridge
and Fraser [9]. Another method of improvement for unbraced
regular frames is the so-called “storey buckling approach”,
which accounts for the horizontal interaction between columns
in a storey [10,11]. White and Hajjar have shown that this
approach may result in significant errors in asymmetrical
cases [12]. The storey buckling approach has been the subject
of several papers, among which the papers by Lui, Aristizabal-
Ochoa, and Cheong-Siat-Moy may be highlighted [13–15].
The works of Aristizabal-Ochoa and Cheong-Siat-Moy provide
solutions for both braced and unbraced frames as well
as “partially braced frames”. Aristizabal-Ochoa has further
extended his studies to cover three-dimensional structures
[16,17]. Another interesting improvement approach is proposed
by Hellesland and Bjorhovde, which involves a post-processing
procedure using weighted mean values of buckling lengths [18].
It has been stated that it is necessary to consider a wider
range of unbraced frames in order to confirm the practical
applicability of the proposed method. It is found that the
Hellesland–Bjorhovde improvement approach is applicable to
irregular frames as well.
Apart from the above-mentioned improvement studies,
certain independent methods for determining an approximate
value for the overall buckling load of plane frames are also
developed, whereby the lateral displacements due to a fictitious
loading is utilised. Among these, the methods developed by

0143-974X/$ - see front matter 

c 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jcsr.2005.09.006
606 K. Girgin et al. / Journal of Constructional Steel Research 62 (2006) 605–613
(a) Regular frame.
(b) Irregular frame.
Fig. 1. Regular and irregular frames.
Cakiroglu [19] and Stevens [20] are the earliest. The approach
of Stevens was later improved by Horne [21]. All of these
methods, which are developed for regular frames, are also
applicable to irregular frames by means of slight modifications.
It is interesting to note that all the fictitious load approaches
yield better results compared with the isolated subassembly
approach.
Recently, in AISC (1999), the isolated subassembly
approach has been abandoned and it has been stated that “. . . the
effective length factor K of compression members shall be
determined by structural analysis” [22]. However in several
widely used codes (such as Eurocode 3) the subassembly
approach and related charts and formulae are still being used.
In this paper, a practical method that is applicable to both
regular and irregular frames will be explained and applied to
numerical examples. The method, which is developed by using
the procedure given by Cakiroglu, is performed by applying a
simple quotient based on the results of a fictitious lateral load
analysis [19].
2. Irregular frames
A plane frame may be considered as being regular when all
the beams are continuous along the width of the frame at all
levels, as shown in Fig. 1(a).
The frame becomes “irregular” when the beams of at least
one level are curtailed, as shown in Fig. 1(b). In other words, it
is not possible to define a “storey” for certain levels of irregular
Fig. 2. Irregular frame examples.
(a) Frame and loading. (b)
Buckling
mode
displace-
ments.
Fig. 3. Multi-storey frame and buckling mode.
frames. In practice, several frames of this nature exist, as shown
in the examples in Fig. 2.
In the case of irregular frames, the error orders of
code procedures are far greater, mainly because the isolated
subassembly assumptions are hardly satisfied. Moreover,
almost all of the improvement studies mentioned above hardly
offer any remedy, since most of them use the storey buckling
approach and it is not possible to define a storey at certain
(or all) levels of an irregular frame. On the other hand, the
method presented in this paper offers an approximate but simple
solution for both regular and irregular frames.
3. System buckling load of unbraced multi-storey frames
A multi-storey frame that is composed of beams and
columns made of linear elastic material is under the effect of
vertical loads, as shown in Fig. 3(a).
 K. Girgin et al. / Journal of Constructional Steel Research 62 (2006) 605–613 607

Each axial load may be expressed as

Ni j = n i j P (1)
where n i j is a dimensionless coefficient and P is an arbitrarily
chosen load parameter. The frame is in a state of “Stable
Equilibrium” and, if the axial deformations are neglected, all
the displacements and deformations are zero. The internal
forces of the frame columns consist of only axial forces
Ni j , while all the internal forces of the beams are zero.
However, when the load parameter reaches a critical value Pcr ,
another state of “Unstable Equilibrium” may exist. The lateral
displacement diagram corresponding to this new state, which
is shown schematically in Fig. 3(b), is called the “Buckling
Mode” of the structure [23]. Once the buckling load parameter
Fig. 4. Stiffness coefficients and distribution factors.
Pcr is determined, the buckling length si j of an individual
column can be computed from

Stiffness coefficients for various members are shown in
E Ii j Fig. 4.
si j = π (2)
n i j Pcr Buckling (effective) length multiplier β is obtained first by
solving the following equation for γ
where E Ii j is the bending stiffness of the column.   
In certain simple cases, the buckling load parameter may γ 1 γ 1
− −
be determined by using the so-called stability functions [23]. 3(1/co − 1) tan γ 3(1/cu − 1) tan γ
For general cases however, it is necessary to utilise specially 1
prepared software. − 2 =0 (5)
sin γ
In this study, a practical method is developed for determining
the buckling lengths of columns in unbraced frames. The and then by computing
method is based on computing an approximate value for system π
β= , (6)
buckling load by using the results of a fictitious lateral loading. γ
4. Buckling lengths according to design codes [24].
Alternatively, instead of using Eqs. (5) and (6), buckling
In several design codes and specifications, simplified length multiplier β may also be read from the diagrams given
formulae and diagrams are given for calculating the buckling in the code. The buckling length s of an individual column is
lengths of individual columns. These simple formulae have computed from
the advantage of enabling the designer to obtain the buckling
lengths, without applying the tedious computations (or special s = βL (7)
software) which are necessary for the calculation of the overall where L is the length of the column.
buckling load. In the following, the formulae of Eurocode 3 are The application of code formulae on several numerical
presented as an example [4]. examples has shown that erroneous results may be encountered
In Annex B1 of Eurocode 3, calculation of the “buckling for both sway and non-sway modes. This is mainly because
of components of building structures” is supplied as follows. only local stiffness distributions are considered in these
First, the so-called distribution factors co and cu for columns in formulae, while the general behaviour of the frame is not
a sway mode are computed from taken into account. Discussion of buckling lengths of non-sway
1 frames is left out of the scope of this study for the sake of
co = 
a Ko
(3) brevity. The erroneous results encountered for sway mode will
1+ K s +K s,o presently be demonstrated on several numerical examples.
1
cu = 
a Ku
. (4)
1+ K s +K s,u
4.1. Example 1

Here The dimensions and loading of a regular three-storey frame

K s , K s,o , K s,u are the stiffness coefficients (I /L values) of the
columns,
K o and K u are the stiffness coefficients (I /L values) of the
beams,
a is a coefficient varying between 1 and 4, depending on
the end conditions of the beams.
are shown in the schematic elevation in Fig. 5.
Using special software prepared by Girgin [25] which uses
the system-buckling approach, the exact value of the buckling
load for the frame is found to be
EI
Pcr = 4.556 . (8)
h2
608 K. Girgin et al. / Journal of Constructional Steel Research 62 (2006) 605–613

(a) Frame and loading.

Fig. 5. Schematic elevation of Example 1.

Substituting this value in Eq. (2) and taking n i j = 1

s = 1.47h (9)
is obtained for all the columns. Buckling length multipliers
β calculated by using Eqs. (3)–(6) yield values of 1.18, 1.32
and 1.19 for 1st, 2nd and 3rd storeys, respectively. Relative
errors in β values vary between −19.8% and −10.5%, which
may be considered rather high, because the structure under
consideration has been chosen as being as regular as possible.
Hence it satisfies all the assumptions in deriving the isolated
subassembly equations. (b) Column designations.

4.2. Example 2 Fig. 6. Schematic elevation of Example 2.

The dimensions and loading of an irregular frame are shown Table 1

in the schematic elevation in Fig. 6. Buckling length multipliers for Example 2
The exact value of the buckling load for this frame is found Column n L β (Exact) β (Eurocode 3) Relative error (%)
to be
C1 1.00 h 1.72 1.24 −27.9
EI C2 2.00 h 1.21 1.16 −4.1
Pcr = 3.352 2 . (10) C3 1.00 h 1.72 1.08 −37.2
h
C4 1.00 2.3h 0.75 1.04 38.7
The exact values for buckling length multipliers β are found by
utilising Eqs. (2) and (7), which are shown in the 4th column
of Table 1. β values according to Eurocode 3 are also shown in Table 2
Table 1 together with the relative errors. Relative error ranges for several examples and different codes (%)
The relative errors of β values vary between −37.2% Type Eurocode 3 (2002) AISC (1988) ACI (1989)
and 38.7%, which may be considered as excessive, i.e. not (DIN 18800)
acceptable for design purposes. In other numerical examples, Min Max Min Max Min Max
even larger errors may be encountered, as explained below. a −52.8 17.6 −44.9 27.6 −42.6 31.8
b −48.9 31.2 −40.3 36.3 −39.8 39.2
4.3. Other examples and other design codes c −37.1 38.1 −32.6 43.1 −30.4 46.4
d −52.5 18.2 −41.5 32.8 −42.1 41.7
Several other irregular examples are tested using code e −57.4 58.5 −47.8 77.5 −48.6 88.5
formulae, and the error ranges are summarised in Table 2. f −45.6 43.1 −37.1 56.3 −36.3 63.0
g −46.8 2.4 −39.2 22.0 −37.5 19.6
Types a–g in the first column of the Table 2 refer to the h −60.1 13.9 −55.4 28.7 −54.3 31.3
designations shown in Fig. 2, while type h represents the
irregular frame shown in Fig. 1(b). The characteristics of type c
are as shown in Fig. 6 (Example 2). Detailed characteristics of calculations for the above-given examples have been carried
other examples are not included for the sake of brevity. out using AISC (1988) charts and ACI (1989) formulae, and
Similar formulae and diagrams for calculating buckling similar results are obtained. The ranges of errors for the codes
length multipliers are given in other design codes. The under consideration are shown together in Table 2.
 K. Girgin et al. / Journal of Constructional Steel Research 62 (2006) 605–613 609

Fig. 8. Displacement diagram of a column portion.

(a) Lateral loading. (b)
Displace-
ments. 5.1. Determination of W1
Fig. 7. Multi-storey frame and fictitious lateral loading.
According to the Principle of Virtual Works, W1 can be
It is clearly seen that all the codes considered yield errors, computed as the work done by the internal forces of the loading
which are almost of the same order. This is due to the fact shown in Fig. 3, in conjunction with the deformations induced
that all codes use similar formulae, which consider only the by the fictitious lateral loading. The displacement diagram of
local (isolated) stiffness distributions. However, investigations an infinitely small portion of one of the columns together with
carried out on a number of numerical examples have shown that the internal forces is shown in Fig. 8.
buckling length multipliers are dependent on If the axial deformations are neglected, the virtual work in
• Axial force distribution, this small portion can be computed from the product of the
dv
• Overall stiffness distribution, couple Ndv and the rotation dx . Hence, the virtual work on
• Position of the individual element any column can be obtained from
 h
together with local stiffness distributions. It is concluded that dv
w= N dv (12)
the buckling length multipliers should be determined by taking x=0 dx
into account all these factors, i.e. considering not only the local or
stiffness distributions but also the overall characteristics of the   2
h dv
structure.
w = nP dx (13)
x=0 dx
5. A simplified procedure for determining the buckling load
where h denotes the height of the individual column. The total
virtual work can be expressed as
In the following, a practical method will be explained and
applied to the numerical examples. The method, which is    h  dv 2
developed by using the procedure given by Cakiroglu [19], is W1 = w=P n dx. (14)
x=0 dx
applied by using a simple quotient based on the results obtained
from standard frame analysis software. Here the summation will be carried out for all the columns.
Consider the fictitious lateral loading shown in Fig. 7 applied It must be noted that the indices are omitted for the sake of
to the frame shown in Fig. 3. It is assumed that this loading simplicity. In the following, calculations related to the integral
provides displacements identical to (or proportional with) those expression on the right hand side of Eq. (14) will be performed.
corresponding to the buckling mode. Since the integrand contains the derivative of the lateral
The buckling load parameter can be determined by using displacements, it suffices to consider relative displacements.
Betti’s Reciprocal Theorem applied to the states shown in The bending moment and relative displacement diagrams of
Figs. 3 and 7. According to this theorem, it may be written that an individual column are shown in Fig. 9. α is a dimensionless
coefficient designating the location of the point of contraflexure
W1 = W2 (11) and δ denotes the relative storey displacement.
where W1 is the virtual work of the force system in Fig. 3(a) in The expression for the deformation of the column is
conjunction with the displacements in Fig. 7(b), and W2 is the
d2 v M(x)
virtual work of the force system in Fig. 7(a) in conjunction with
2
=− (15)
the displacements in Fig. 3(b) [26]. Since the displacements dx EI
of Figs. 3(b) and 7(b) are assumed as to be the same, the where the bending moment function M(x) may be expressed as
displacements and deformations corresponding to the lateral x

fictitious loading will be used in the following. M(x) = M A 1 − . (16)

αh
610 K. Girgin et al. / Journal of Constructional Steel Research 62 (2006) 605–613

Both summations in the above formula will be carried out on

columns. It must be noted that this formula is approximate,
since the lateral loading corresponding to the buckling load
displacements are not known initially. However, application to
several numerical examples has shown that the value of Pcr is
not strongly dependent on the initial choice of lateral loads. It
may be recommended that the lateral load at each joint should
be selected to be proportional to the vertical load Pi j existing at
the joint.
(a) Bending moments. (b) Relative
displacements.
5.4. The χ coefficients
Fig. 9. Column bending moment and relative displacement diagrams.
It is seen that, when applying Eq. (22), it is necessary to
Substituting M(x) into Eq. (15) and integrating twice with the compute χi j coefficients for each individual column. As can be
boundary conditions seen in Eq. (19), these coefficients are dependent on the bending
v=0 for x = 0 and moments; hence, a tedious amount of computation is required.
However, it can be shown that χ values vary in a rather narrow
v=δ for x = h range and can easily be simplified.
yields Let us consider the basic equation used in the approximate
  methods of lateral load analysis, which may be expressed as
δ M A h(3α − 1) MA
v(x) = + x+ x 2 (x − 3αh). (17) Q
h 6E I α 12E I αh δ= . (23)
k 12E I
After substituting the derivative of v(x) into Eq. (14) and h3
carrying out the integration Here Q denotes the shear force of the individual column and
 δ2 k is a dimensionless coefficient varying between 0 and 1,
W1 = P n χ (18) which depends on the stiffness of the beams at each end of the
h column [27]. Eq. (23) can alternatively be written as
is obtained. Here χ denotes a dimensionless coefficient given MA
by αh M A h2
δ= = (24)
 2   k 12E
h3
I 12kα E I
M A h2 1 1 1
χ =1+ − + . (19)
EIδ 12 12α 45α 2 from which
The rather interesting variation of χ will be discussed presently. M A h2
= 12kα (25)
EIδ
5.2. Determination of W2 is obtained. Substituting this into Eq. (19)
 
The virtual work of the force system in Fig. 7(a) in 1 1 1
χ = 1 + 144k 2α 2 − + (26)
conjunction with the displacements in Fig. 3(b) (Fig. 7(b)) can 12 12α 45α 2
be written simply as
 is found. It is seen that this new expression is dependent only
W2 = Hi d i (20) on two dimensionless variables, namely k and α. The variation
of χ is shown in the diagrams in Fig. 10.
where Hi and di represent the lateral storey loads and storey It is well known that, when k approaches unity, α assumes
displacements, respectively. The summation will be performed values near 0.50. Moreover, calculations carried out on the
for all stories. Eq. (20) may more conveniently be expressed columns of several regular and irregular numerical examples
in terms of column shear forces Q i and relative storey have yielded the results shown as dots and circles in Fig. 11.
displacements δi as
 It may be concluded that approximately the lower half of the
W2 = Q i δi . (21) figure is valid for practical purposes. Considering this narrow
range for values of χ, it is reasonable to assume a constant and
5.3. Simplified buckling load formula conservative value of
χ = 1.20
Substituting the expressions for W1 and W2 , given
respectively by Eqs. (18) and (21), into Eq. (11) and solving for practical purposes. Thus, Eq. (22) takes the rather practical
for P (Pcr ), the buckling load is obtained as form of
 
Q i δi Q i δi
Pcr =  . (22) Pcr =  δi2 . (27)
δ2
n i j hii χi j 1.20 n i j h i
 K. Girgin et al. / Journal of Constructional Steel Research 62 (2006) 605–613 611

Fig. 12. Fictitious lateral loading for Example 1.

Table 3
Buckling load calculations for Example 1

Fig. 10. Theoretical variation of χ values. EI δ E I Qδ (E I )2 δ 2

Storey Q n nh
h3 h3 h5
3 1.00 0.2063 0.2063 1.00 0.0426
2 1.00 0.2237 0.2237 1.00 0.0500
1 1.00 0.1057 0.1057 1.00 0.0112
Sum 0.5357 0.1038

6.1. Example 1

Fig. 11. Variation of χ values for numerical examples.
5.5. Analysis procedure
The buckling lengths of frame columns can be determined
as follows:
• Apply lateral forces proportional to the vertical loads at each
joint,
• Compute relative storey displacements using any existing
software,
• Compute the critical load Pcr by using Eq. (27),
• Determine the buckling lengths of columns by using Eq. (2).
This procedure is easily applicable to both regular and
irregular frames.
The dimensions and loading of the first example are the same
as shown in the schematic elevation in Fig. 5 of Section 4.1. The
fictitious lateral loading is shown in Fig. 12.
After carrying out a lateral load analysis for the fictitious
loading, relative storey displacements are obtained. The terms
used for the application of Eq. (27) are shown in Table 3.
One column for each storey is considered due to symmetry.
Applying Eq. (27) yields
0.5357 EI EI
Pcr = = 4.301 2
1.20 × 0.1038 h 2 h
with an error of −5.6%. Computing the buckling lengths of
columns by using Eq. (2) gives
s = 1.51h
for all the columns, with an error of 2.7%. It is interesting to
note that the buckling lengths (and errors) of all the columns
are the same due to the fact that they are computed by using the
same equation used for exact calculation, namely Eq. (2).
Buckling load calculations are repeated by using wind
and earthquake loadings for the same frame, and the errors
in buckling lengths are found to be −3.5% and −0.5%,
respectively. It can be deduced that any lateral loading can
be used in determining the approximate buckling load value,
without largely effecting the results.
6.2. Example 2

The dimensions and loading of the second example are

6. Numerical examples the same as shown in the schematic elevation in Fig. 6 of
Section 4.2. The fictitious lateral loading is shown in Fig. 13,
In the following, the procedure outlined above will be where loads are chosen to be proportional to vertical loads at
applied to examples and the results will be discussed. the joints.
612 K. Girgin et al. / Journal of Constructional Steel Research 62 (2006) 605–613

Table 6
Errors on buckling length multipliers

Type Relative error (%)

a 1.8
b 3.0
c 1.8
d 1.6
e 0.7
f 2.5
g 2.9
h 1.6
Minimum 0.7
Maximum 3.0
Average 2.0

Fig. 13. Fictitious lateral loading for Example 2.

Table 7
Table 4 Errors for wind and earthquake loadings (%)
Buckling load calculations for Example 2 Type Loading
EI δ E I Qδ (E I )2 δ 2 Wind Earthquake
Column Q hi n nh
h3 h3 h5 i
a 3.0 3.0
C1 1.501 h 0.3306 0.4963 1.00 0.1093 b 3.0 2.2
C2 1.403 h 0.2333 0.3271 2.00 0.1088 c 2.2 2.7
C3 2.099 h 0.2333 0.4895 1.00 0.0544 d 0.7 1.7
C4 0.499 2.3h 0.5639 0.2814 1.00 0.1382 e −1.0 0.1
Sum 1.5943 0.4107 f 0.3 −1.9
g 3.0 2.0
h 1.9 0.4
Minimum −1.0 −1.9
Table 5
Maximum 3.0 3.0
Buckling length multipliers for Example 2
Average 1.9 1.8
Column β (Exact) β (Prop. method) Relative error (%)
C1 1.716 1.747 1.8
It is seen that all the errors are on the safe side and in the
C2 1.213 1.235 1.8
C3 1.716 1.747 1.8 acceptable range. All the irregular examples are solved again by
C4 0.746 0.760 1.8 using wind and earthquake loadings too, and the corresponding
error orders are shown in Table 7.
It is seen that the error orders do not vary considerably due
Lateral load analysis yields relative displacements δ and to the kind of lateral loading, and they are on the safe side for
column shear forces Q. Using these values, the terms used for the great majority of cases.
the application of Eq. (27) are calculated as shown in Table 4.
Eq. (27) yields 6.4. Other approximate methods
1.5943 EI EI
Pcr = = 3.235 2
1.20 × 0.4107 h 2 h All the improvement methods mentioned in Section 1 are
developed for regular frames and most of them use the “storey
with an error of −3.5%. Buckling length multipliers β, which
buckling approach”. Since it is not possible to define a “storey”
are calculated by means of Eq. (2), are shown and compared
at least for certain levels of irregular frames, these methods
with the exact values in Table 5.
are not applicable to them. The only improvement method that
Here again, all the buckling length parameters have the same
considers all the frame columns is the Hellesland–Bjorhovde
error and are on the safe side.
approach. The methods using the “fictitious loading approach”
Buckling load calculations are repeated by using wind and
of Cakiroglu and Horne are also applicable to irregular frames
earthquake loadings for the same frame, and the errors on
by applying slight modifications. The irregular examples of
buckling lengths are found to be 2.2% and 2.7%, respectively.
Section 4.3 have once again been solved by using these three
methods, and the errors encountered are shown in Table 8.
6.3. Other examples The errors for all the columns of a certain frame are the same
for these methods too. It is seen that all the errors encountered
Using the proposed procedure, the irregular examples of are less than those found by using code formulae. However,
Section 4.3 are solved and the errors encountered in buckling they are generally greater than the errors found by applying the
length multipliers are shown in Table 6. method presented in this paper.
 K. Girgin et al. / Journal of Constructional Steel Research 62 (2006) 605–613 613

Table 8 Institute of Steel Construction; 1988.

Errors in buckling length multipliers for approximate methods [2] ACI 318-89. Building code requirements for reinforced concrete with
design applications. Skokie (IL): Portland Cement Association; 1989.
Type Relative errors (%) [3] Galambos TV. Structural members and frames. New York: Prentice-Hall,
Cakiroglu [19] Horne [21] Hellesland–Bjorhovde [18] Inc.; 1968.
a 13.1 −0.8 −3.0 [4] Eurocode 3. Design of steel structures. Final draft. Brussels (Belgium):
b 8.8 −7.5 3.5 CEN; 2002.
c 8.6 −4.2 3.8 [5] DIN 18800. Part 2: Analysis of safety against buckling of linear members
d 16.6 8.2 0.1 and frames. Berlin: Beuth Verlag GmbH; 1990.
e 12.8 −2.4 11.7 [6] Hellesland J, Bjorhovde R. Restraint demand factors and effective lengths
f 10.2 7.9 3.5 of braced columns. J Struct Engrg ASCE 1996;122(10):1216–24.
g 2.0 9.4 −6.9 [7] Duan L, Chen WF. Effective length factors for columns in braced frames.
h −1.9 5.5 −14.7 J Struct Engrg ASCE 1988;114(10):2357–70.
Minimum −1.9 −7.5 −14.7 [8] Duan L, Chen WF. Effective length factors for columns in unbraced
Maximum 16.6 9.4 11.7 frames. J Struct Engrg ASCE 1989;115(1):149–65.
Average 9.3 5.7 5.9 [9] Bridge RQ, Fraser DJ. Improved G-factor method for evaluating effective
lengths of columns. J Struct Engrg ASCE 1987;113(6):1341–56.
[10] Yura JA. Effective length of columns in unbraced frames. Engrg J AISC
7. Conclusions 1971;8(2):37–42.
[11] LeMessurier WJ. A practical method of second-order analysis, 2: Rigid
In this paper, the determination of buckling lengths of multi- frames. Engrg J AISC 1977;14(2):49–67.
[12] White DW, Hajjar JF. Buckling models and stability design of steel
storey frame columns is investigated. The main conclusions
frames: A unified approach. J Constr Steel Res 1997;42(3):171–207.
derived may be summarised as follows: [13] Lui EM. A novel approach for K factor determination. Engrg J AISC
1. It is shown that simplified formulae and diagrams, which are 1992;29(4):150–9.
[14] Aristizabal-Ochoa JD. Braced, partially braced and unbraced frames:
given in several design codes and specifications, may yield Classical approach. J Struct Engrg ASCE 1997;123(6):799–807.
rather erroneous results for buckling lengths of the columns, [15] Cheong-Siat-Moy F. An improved K-factor formula. J Struct Engrg ASCE
especially for irregular frames. This is due to the fact that 1999;125(2):169–74.
the code formulae refer only to local stiffness distributions, [16] Aristizabal-Ochoa JD. Classic buckling of three-dimensional multi-
instead of the overall behaviour of the structure. column systems under gravity loads. J Engrg Mech ASCE 2002;128(6):
613–24.
2. A simplified procedure for determining an approximate [17] Aristizabal-Ochoa JD. Elastic stability and second-order analysis of three-
value for system buckling load has been developed. The dimensional frames: Effects of column orientation. J Engrg Mech ASCE
procedure utilises a simple quotient based on the results of a 2003;129(11):1254–67.
fictitious lateral load analysis. Buckling lengths of columns [18] Hellesland J, Bjorhovde R. Improved frame stability analysis with
may then be calculated by means of a simple formula. effective lengths. J Struct Engrg ASCE 1997;122(11):1275–83.
[19] Cakiroglu A. Buckling analysis of multi-storey frames. In: Proc. of
3. The proposed procedure is applicable equally easily to both technical conference of Turkish civil engineers. 1962 [in Turkish].
regular and irregular cases. [20] Stevens LK. Elastic stability of practical multi-storey frames. Proc Inst
4. The proposed procedure yields errors that are less than 5% Civ Eng 1967;36.
for all the examples considered. This order may be regarded [21] Horne MR. An approximate method for calculating the elastic critical
as acceptable from the designer’s point of view. loads of multi-storey plane frames. Struct Eng 1975;53(6).
[22] AISC. Load and resistance factor design specification for structural steel
5. The buckling load value is not strongly dependent on the buildings. Chicago (IL): American Institute of Steel Construction; 1999.
choice of lateral loading. Hence, any existing lateral loading [23] Horne MR, Merchant W. The stability of frames. London: Pergamon
on the frame under consideration may be used without losing Press; 1965.
a significant amount of accuracy. [24] Schneider KJ. Bautabellen für ingenieure. Düsseldorf: 12. Auflage,
6. The proposed procedure is applied to several numerical Werner Verlag; 1996.
[25] Girgin K. A method of load increments for determining the second-order
examples and it is seen that all the errors are in the acceptable limit load and collapse safety in R/C structures. Ph.D. thesis. Istanbul:
range and on the safe side for most cases. Istanbul Technical University; 1996 [in Turkish].
[26] Neal BG. Structural theorems and their applications. London: Pergamon
References Press; 1964.
[27] Muto K. Aseismic design analysis of buildings. Tokyo: Maruzen Co. Ltd.;
[1] AISC. Specification for structural steel buildings. Chicago (IL): American 1974.