This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elseviers archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright

Author's personal copy

Journal of Constructional Steel Research 66 (2010) 670679

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

journal homepage: www.elsevier.com/locate/jcsr

On the theoretical background of the generalization of AyrtonPerry type

resistance formulas
Jzsef Szalai a, , Ferenc Papp b
a

KESZ Ltd., 1095-Hungary, Budapest, Mester utca 87, Hungary

Department of Structural Engineering, Budapest University of Technology and Economics, 1111-Hungary, Budapest, Bertalan L. utca 2, Hungary

article

info

Article history:
Received 18 May 2009
Accepted 18 December 2009
Keywords:
AyrtonPerry formula
Column buckling
Lateral-torsional buckling
Buckling curve
Imperfection factor
Amplification relationship

abstract
The AyrtonPerry type formulae are very popular models for the standard definition of the buckling
resistance of steel members. These standards benefit from the simplicity and flexibility, but most of all
from the clear mechanical background of this model, which apparently defines the appropriate meaning
of the model parameters. While this mechanical background has been properly clarified for column
buckling, however, for the case of the lateraltorsional buckling problem despite the various numerical
verifications of the standard models the exact derivation is still an unresolved question. This paper
introduces a possible way for the rigorous generalization of the AyrtonPerry formula, so it can be applied
to the description of lateraltorsional buckling problems. It is demonstrated that the shape of the initial
geometric imperfection has a key role in the solution; an appropriate choice can simplify the problems
considerably through a convenient form of the amplification relationship. The equations obtained have
various consequences regarding the form of the generalized and equivalent imperfection factors and the
multiple curves for lateraltorsional buckling resistance.
2009 Elsevier Ltd. All rights reserved.

1. Introduction
This paper deals with the theoretical background of design
curves for member buckling problems included in many modern standards as a basis for the stability design of steel structures. In the Eurocode 3 Chapter 1-1 (EC3) [1] member buckling
curves are used for two basic cases: flexural buckling of columns
and lateraltorsional buckling (LTB) of beams (pure bending) and
beamcolumns (compression and bending). These two buckling
curves are highly significant and are necessary for the standard
design of general member stability. The accurate theoretical and
experimental verification of these is therefore very important.
Column buckling has been thoroughly researched for many years
as the simplest and most basic case of stability problems. This
research has included a huge number of experimental tests, complete theoretical investigations using both analytical and numerical models, and advanced probabilistic examinations [24]. As a
result, the multiple column curves concept has been developed and
the AyrtonPerry formula (APF) has been adopted [57] as the basic design model. This formula has the following main advantages:
a clear mechanical background, simplicity and flexibility. It is important to note, however, that it is connected directly with flexural buckling phenomena. LTB is far more complicated to deal with;

Corresponding author. Tel.: +36 1 4766 574; fax: +36 1 4766 593.
E-mail address: szalaija@gmail.com (J. Szalai).

0143-974X/$ see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jcsr.2009.12.013

experimental research usually runs into modelling difficulties and

accordingly considerably less results have been obtained [8], the
analytical description is limited to the most basic cases [9]. Nowadays the most appropriate analysis is numerical simulation [10].
Exploiting the flexibility of the APF, this form was chosen as the design curve for LTB [8,11] adjusting the design parameters carefully
to correspond with the results of the numerical simulations [12].
Although the LTB curves can be regarded as adequately verified in
this way; the necessary conditions for the exact derivation of the
APF for LTB have not yet been clarified.
This paper introduces the suitable theoretical background for
the proper generalization of the APF, mainly for the purpose
of straightforward application to LTB problems of beams and
beamcolumns. The shape of the initial geometric imperfection is
found to be highly significant, and this should be defined appropriately to obtain the exact APF [13,14]. It is also found that in the analytical description of stability problems the proposed imperfection
shape is the most plausible choice and has a number of advantageous features. Accordingly, one of the most important results of
the paper is the clarification of the system of amplification effects
for the case of interactive loading and multiple imperfection components, which may be the starting point for any design formulae
for the LTB of beams and beamcolumns. It is important to note
that this paper deals only with the theoretical side of the problem and contains analytical solutions only; numerical results are
widely available in the literature [913,15].

Author's personal copy

J. Szalai, F. Papp / Journal of Constructional Steel Research 66 (2010) 670679

The structure of the paper is as follows: first of all the problem is described in detail through the analysis of the original
AyrtonPerry formula. The main points and the targeted generalization are introduced. The solutions for lateraltorsional buckling
of beams and beamcolumns are presented in the next two sections, then some particular features of the new formulae are analyzed in connection with the new form of imperfection factors, the
effect of torsional rigidity and the suitable section type division for
the multiple buckling curves.

671

V0

2. Definition of the problem

2.1. The original AyrtonPerry formula for column buckling
The APF was originally applied for geometrically imperfect,
elastic columns loaded by uniform compression, where the load
carrying capacity corresponds to the onset of yielding at the most
compressed fiber [5]. This problem can be described by the following differential equation:
EIz

2 v(x)
+ N (v(x) + v0 (x)) = 0
x2

(1)

where EIz is the appropriate lateral stiffness, v and v0 are the lateral displacement and imperfection, respectively, depending on
the axial coordinate x of the column, and N is the compressive force
(Fig. 1). Assuming sinusoidal imperfection, the maximum total second order lateral displacement of such a simply supported prismatic column writes:

vTot =

1
1 N /Ncr ,z

v0

(2)

where v0 is the midspan amplitude of the half-sine wave imperfection and Ncr .z is the elastic critical buckling load about the minor
axis. This equation has key significance in the generalization of the
APF so it is referred to as amplification relationship in this paper. At
the midspan cross-section, the most compressed fiber should reach
the yield stress (first yield criterion):
N
A

N vTot
Wz

= fy

(3)

where A is the cross-sectional area, Wz is the elastic sectional modulus and fy is the yield stress. If substituting Eq. (2) into Eq. (3), the
original form of the APF can be written:

(cr .z b )(fy b ) = b cr .z z

(4)

where cr .z is the elastic critical compressive stress, b is the actual

compressive stress, and z = v0 WA is the generalized imperfecz
tion factor. Eq. (4) can also be written by the standard notations
according to the EC3 [1]:

z2 + z 1

2z

2z

z +

2z

=0

where z is the well-known slenderness (z =

buckling reduction factor (z

N
Afy

(5)

Afy
Ncr .z

), z is the

). The solution of Eq. (5)

yields the standard form of the APF-based column buckling

curve [6], where the exact form and values of the generalized imperfection factor were defined by deep numerical and experimental calibrations, and probabilistic verification [7].
2.2. Generalization of the formula
Researchers working on unified design methods for the Structural Eurocodes quickly realized the advantages of the APF. The

Fig. 1. The basic model of the original APF.

clear mechanical background, the simplicity and flexibility of the

formula made it an optimal choice for application in stability design procedures. Accordingly, the reduction factor form of Eq. (5)
was chosen for application to other types of stability problems, especially to the lateraltorsional buckling of beams and
beamcolumns [8,11,12,15]. Although the design parameters have
also been calibrated in these cases, the mechanical background has
not been appropriately developed only approximate solutions
have been published [11] as the APF has not been extended rigorously to lateraltorsional buckling. The main difficulty was that
the greater complexity of the lateraltorsional buckling type problems resulting in much more intricate formulae proved to be inappropriate for standard procedures. The most important source
of this complexity is the form of the amplification relationship. At
this point it is beneficial to introduce two important definitions for
the amplification relationship:

the amplification relationship is called linear if the quotient

term in the denominator is to the first power (as in Eq. (2)).
Similarly, quadratic, cubic etc. forms are distinguished;
the amplification relationship is called consistent if in the case
of several governing displacement components it is identical
for all the components.
Using these definitions, it is shown later that the conventional
form of the APF can be obtained only if the amplification
relationship is linear and consistent. Accordingly, in this paper we
illustrate that the most important stage of the generalization is the
properly chosen shape for the initial geometric imperfection. It is
demonstrated that the following statements apply to the shape of
the first buckling mode of the perfect member:
1. this mode is always identical to the shape of the total second
order displacements of the imperfect member as the load
approaches the elastic critical load, when applying imperfection
of whatever shape;
2. applying the shape of the first buckling mode for imperfection,
the shape of the displacements of the member is always the
same throughout the whole loading history; i.e. the shape of
displacements is constant, only the amplitude changes;
3. it is necessary to apply the shape of the first buckling mode for
imperfection, in order to obtain a linear and consistent amplification relationship.
The last statement is found to be essential when constructing
the APF, and most of the difficulties arise from the fact that, for
other types of imperfection shapes, the amplification relationship
becomes nonlinear (usually taking on some kind of quadratic
form [14,16]) and inconsistent, as presented later for the case of

Author's personal copy

672

J. Szalai, F. Papp / Journal of Constructional Steel Research 66 (2010) 670679

governing displacement components (the effect of deflection is

neglected as stated earlier) [9]:

2 v(x)
+ My (x) = 0
x2
3 (x)
(x)
v(x)
EIw
GIt
+ My
=0
x3
x
x
EIz

(6)
(7)

with the boundary conditions associated with the assumptions

made in the previous section:

2 (x)
= v(x) = 0
(8)
x2
at both ends; i.e., at x = 0 and x = L. In Eqs. (6)(7) E and G are the
(x) =
Fig. 2. The descriptive geometry of the LTB problem.

the lateraltorsional buckling problem. Consequently, the proofs

of the above-mentioned statements for lateraltorsional buckling
of beams and beamcolumns are the most significant results in
this paper, and provide the basis for the theoretical background of
generalization of the AyrtonPerry type resistance formulae.
In order to remain simple and to obtain a clear and transparent
demonstration of the method, only the most basic cases are analyzed. Accordingly, the following general assumptions are made
for the subsequent examinations:

the member is prismatic, simply supported to prevent both

lateral deflection and twist at the ends, but free to warp;

the cross section is a doubly symmetric I-section;

the material is homogeneous, isotropic and linear elastic;
the loads are conservative, uniform (uniform compression (N )
and uniform major axis bending moment (My )) and applied
centrically;
the second order effects of in-plane curvature and the in-plane
bending moment are neglected;
the usual approximations for small displacements and rotations
and Bernoullis hypothesis apply;
the coordinate system and rotations follow the right hand rule
(Fig. 2).
According to the shapes of the displacements and imperfections,
it is further considered that the components have a half sine wave
shape along the member length. This is consistent with the statements made earlier, since the components of the first buckling
mode always have a half sine wave shape [9]. Moreover, it can
be shown that any kind of imperfection can be approximated by a
Fourier sine series, the components of which are the proper higher
order buckling modes with several waves. Using this Fourier approximation for the shape of imperfection and having arbitrary
amplitudes for the different modes, it was proved in [9] that, as
the load approaches the critical load, the total second order displacements are always dominated by the half sine wave mode (first
buckling mode). In the subsequent sections, therefore, for the displacements and imperfections, a half sine wave shape is assumed
which is completely defined by the midspan amplitude or for the
case with several governing displacement components in the system (e.g. LTB problem) the ratio between the midspan amplitudes
of the components.
3. Solution for lateraltorsional buckling of beams
3.1. Analysis of the perfect system
LTB of a perfectly straight beam subjected to pure, uniform
strong axis bending moment can be analyzed by the following
linear, homogeneous system of differential equations with two

elastic and shear modulus respectively, Iz is the second moment of

inertia about the minor axis, Iw and It are the warping moment of
inertia and the St. Venant torsional constant respectively.
The solution for the first buckling mode takes the following
form:

v(x) = vA sin
(x) = A sin

(9)

x
L

(10)

In Eqs. (9)(10) a special relationship exists between the two

amplitudes vA and A (and obviously between the two whole
shapes of the components as well):

vA
Mcr
=
= r0
A
Ncr ,z

Ncr ,x

(11)

Ncr ,z

where the right-hand side terms are the polar radius of gyration
(r0 ), the elastic critical axial force associated with the pure lateral
(Ncr ,z ) and pure torsional buckling (Ncr ,x ) and the elastic critical
bending moment (Mcr ):
Ncr ,z =
Ncr ,x =

EIz 2

(12)

L2

1
r02

Mcr = r0

EIw 2
L2


+ GIt

Ncr ,z Ncr ,x .

(13)
(14)

3.2. Analysis of the imperfect system

The same system with geometric imperfections can be described by the following linear and inhomogeneous system of differential equations [9]:

2 v(x)
+ My (x) = My 0 (x)
x2
3 (x)
(x)
v(x)
v0 (x)
EIw
GIt
+ My
= My
x3
x
x
x
EIz

(15)
(16)

where the imperfections are of the same form as Eqs. (9) and (10)
as discussed earlier:

v0 (x) = v0A sin

0 (x) = 0A sin

x
L

x
L

(17)
(18)

with arbitrary amplitudes at this stage to keep generality.

Substituting Eqs. (9), (10), (17) and (18) into Eqs. (15) and (16) and
omitting the sine and cosine terms, the following equation can be
obtained:

Author's personal copy

J. Szalai, F. Papp / Journal of Constructional Steel Research 66 (2010) 670679

Ncr ,z

My

  
My
0
vA
=
My
A
r02 Ncr ,x

My
0

v0A
.
0A




(19)

Hereinafter all the equations are expressed in terms of these amplitudes only, so the A subscript is omitted for the sake of simplicity.
From Eq. (19) it is now possible to write the general relationship
between the total second order displacements and the initial imperfections; i.e., the amplification relationship:

vTot
Tot

   
v
v
=
+ 0
0

1 My /Mcr

My

1

2
My

Ncr ,z
v0

 

0 .

r02 Ncr ,x

(20)

Eq. (20) is apparently nonlinear and inconsistent in general. This

feature makes the derivation of the general quadratic equation of
APF Eq. (5) impossible if the amplitudes of the components of imperfection are defined arbitrarily.
3.3. Analysis of the imperfection
In this section, the three main statements of Section 2.2 are
demonstrated for LTB of beams. For this purpose, it is enough to
analyze the ratio of the two components of the total displacements
in terms of the ratio of the two components of the imperfection.
First, the displacements are examined as the external bending
moment is approaching the elastic critical value:
M

lim

My Mcr

v0 + Ncry,z 0
vTot
= lim
=
My
My Mcr
Tot
v0 + 0
2
r0 Ncr ,x

v0
0

Mcr
Ncr ,z

Ncr ,z v0
Mcr 0

+1

Mcr
Ncr ,z

. (21)

The result is identical to the first buckling mode of Eq. (11), so the
first statement is proven. Now let us apply this first buckling mode
for the shape of imperfection and express the same ratio but for
external bending moment of arbitrary amplitude:

vTot
=
Tot

v0
0

My
Ncr ,z

v0
r02 Ncr ,x 0
My

+1

Mcr
Ncr ,z

My

1+

Mcr
r02 Ncr ,x Ncr ,z

My
Mcr

+1

Mcr
Ncr ,z

(22)

This result verifies the second statement, so the shape of displacement is constant and equal to the first buckling mode throughout
the whole loading history. It is clear now that the first buckling
mode has particular meaning among the possible shapes for imperfection. The proof of the third and most important statement
requires the solution for the following equation for the (v0 /0 )
ratio:

My

 

1

1
v0
vTot
Ncr ,z

=

2 My
0
Tot
1 My /Mcr
1
r02 Ncr ,x

 
v0
=
.

0
1 My /Mcr
1

(23)

The left-hand side of the equation expresses the general amplification relationship of Eq. (20), and the right-hand side is the aimed
linear and consistent amplification relationship. The equality requires the following shape for the imperfection components:

v0
Mcr
=
0
Ncr ,z

3.4. Solution for the AyrtonPerry formula

After having the appropriate linear and consistent amplification
relationship of Eq. (23), the next step is the construction of the
first yield criterion in terms of the second order internal forces at
midspan:
My
Wy

(24)

which is again the first buckling mode. Accordingly, with this special requirement for the imperfection, the system is applicable for
the construction of the APF for LTB of beams.

673

MzII
Wz

BII

= fy

Ww

(25)

where Wy , Wz and Ww are the elastic major axis, minor axis

and warping sectional moduli, respectively, fy is the yield stress,
and the second order internal forces can be expressed in terms
of the external bending moment and the total displacement
components [17]:
MzII = My Tot = My 0

(26)

1 My /Mcr

BII = My vTot GIt



1
1
= My v0
GIt 0
0 .
1 My /Mcr
1 My /Mcr
Introducing the usual slenderness LT =
factor LT

My
for
Wy f y

Wy f y
Mcr

(27)

and reduction

LTB, the standard quadratic APF can be

obtained:

+ LT 1
2
LT

2
LT

2
LT

LT


+

2LT

=0

(28)

in which the generalized imperfection factor has the following

form:

LT = v0

Wy
W

+ 0

Wy
Wz

GIt Wy
Mcr W

(29)

Eq. (28) has exactly the same form as Eq. (5) for the column
buckling case that verifies the correctness of the application of
APF for LTB, considering, however, the new meaning and effects
of the generalized imperfection factor of Eq. (29); we discuss this
problem in Section 5. Using these expressions the buckling curve
for LTB can be written as the solution of Eq. (28) in the well-known
form of the EC3 [1]:
1

LT =
LT

(30)

2
LT + LT
2LT


= 0.5 1 + LT + 2LT .

(31)

This is the fundamental solution for the APF based LTB curve
belonging to the first yield criterion of Eqs. (25)(27) and the
specially shaped initial geometric imperfection defined by Eq. (24).
4. Solution for lateraltorsional buckling of beamcolumns
4.1. Analysis of lateraltorsional buckling of columns
Before examining the beam-column problem for the interaction
of major axis bending moment and compression, it is useful to
review the LTB problem of columns; i.e., when a member subjected
only to axial compression has initial lateral deflection and twist
simultaneously as geometric imperfection. The most important
point is the form of the amplification relationship, which can be
expressed as follows [16]:

vTot

= 1 N /Ncr ,z
Tot

0
1
1 N /Ncr ,x

 
v0
.
0

(32)

Author's personal copy

674

J. Szalai, F. Papp / Journal of Constructional Steel Research 66 (2010) 670679

In this case there is no interaction between the two displacement

components; both depend only on the corresponding imperfection
component. The first yield criterion takes the following form:
N

MzII
Wz

BII
Ww

= fy

(33)

where the second order internal forces can be expressed in terms

of the axial force and the total displacement components [17]:
MzII

= N vTot = N v0

(34)

1 N /Ncr ,z

BII = r02 N Tot GIt



1
1
= r02 N 0
GIt 0
0 .
1 N /Ncr ,x
1 N /Ncr ,x

(35)

Substituting Eqs. (34)(35) into Eq. (33) yields the basic equation,
which should be solved in order to obtain the reduction factor:
N

N v0

Wz 1 N /Ncr ,z
GIt 0

Ww 1 N /Ncr ,x

1 = fy .

(36)

Ncr ,z N
My

r02

  
N
v
=

My

My

Ncr ,x N

(42)

My
r02 N

 
v0
.
0

(43)

From Eq. (43) it is now possible to write the general relationship

between the total second order displacements and the initial
imperfections; i.e., the amplification relationship:


1
vTot
=

2
Tot
1 My /McrN

This form indicates that although the displacement components

have linear amplification factors separately, since these factors
are different the whole system has an inconsistent amplification
relationship. The consequence is that the conventional quadratic
APF cannot be developed for this problem since the governing
equation is cubic and has no practical straightforward solution.
Accordingly, the main conclusion is that if the axial force N
causes an increased effect on the second order displacement
the axial force is an active part of the loading history then only
one governing imperfection component can be considered, or the
usual solution cannot be obtained. This phenomenon has a very
important meaning for the appropriate treatment of the beamcolumn problem as presented in the following sections.

(41)

where the imperfections are of the same form as Eqs. (17)(18) as

discussed earlier.
Substituting Eqs. (9)(10) and Eqs. (17)(18) into Eqs. (41)(42)
and omitting the sine and cosine terms, the following equation can
be obtained:

1 N /Ncr ,x

Ww

r02 N 0

2 v(x)
+ My (x) + N v(x) = My 0 (x) N v0 (x)
x2
3 (x)
(x)
v(x)
(x)
EIw
GIt
+ My
+ Nr02
x3
x
x
x
v0 (x)
0 (x)
= My
Nr02
x
x
EIz

My

1 N /Ncr ,z
My
1

Ncr ,z N 1 N /Ncr ,x

r02 Ncr ,x N 1 N /Ncr ,z

1 N /Ncr ,x

 
v
0 .
0

(44)

Eq. (44) is also apparently nonlinear and inconsistent in general so

the derivation of the general quadratic equation of APF is impossible again if the amplitudes of the components of imperfection are
defined arbitrarily.
4.4. Analysis of the imperfection

4.2. Analysis of the perfect system

At this point, it becomes necessary to examine the possible

options for the loading history:

The LTB of a perfectly straight beam-column subjected to uniform strong axis bending moment and axial compression can be
analyzed by the following linear, homogeneous system of differential equations with two governing displacement components [9]:

1. governing (changing) bending moment beside a constant (fix)

compression
2. governing (changing) compression beside a constant (fix)
bending moment
3. governing (changing) bending moment and compression beside
a constant (fix) eccentricity (constant bending moment compression ratio).

2 v(x)
+ My (x) + N v(x) = 0
x2
3 (x)
(x)
v(x)
(x)
EIw
GIt
+ My
+ Nr02
=0
x3
x
x
x
EIz

(37)
(38)

with the same boundary conditions, as expressed in Eq. (8). The

solution for the first buckling mode takes also the same form as
Eqs. (9)(10), but with the following relationship between the two
amplitudes:

v
= r0

Ncr ,x N
Ncr ,z N

McrN

Ncr ,z 1 N /Ncr ,z

(39)

where the one new term is the elastic critical bending moment
including the compression effect:
McrN = r0

Ncr ,z N

Ncr ,x N .

(40)

Since it has been shown in Section 4.1 that a system with governing
axial force cannot be solved appropriately with more displacement
and imperfection components, the second and third options are
inadequate for the system described in Section 4.3, and susceptible
for lateral deflection and twist. Accordingly, the derivation of APF
for the beam-column problem is only possible assuming the first
loading option; therefore all the following equations are expressed
for governing external bending moment considering a certain
constant compression level (N ).
First, the displacement ratio is examined again as the external
bending moment is approaching the elastic critical value:
lim

My McrN

vTot
=
Tot

v0
McrN
1
1
1N /Ncr ,z 0
Ncr ,z N 1N /Ncr ,x
McrN
v0
1
1
1N /Ncr ,x
r 2 Ncr ,x N 1N /Ncr ,z 0

4.3. Analysis of the imperfect system

The same system with geometric imperfections can be described by the following linear, inhomogeneous system of differential equations [9]:

v0
0

1 N /Ncr ,x
1 N /Ncr ,z
McrN

Ncr ,z
McrN

Ncr ,z 1 N /Ncr ,z

McrN
1
Ncr ,z 1N /Ncr ,x

1 N /Ncr ,x 0 + 1
0

(45)

Author's personal copy

J. Szalai, F. Papp / Journal of Constructional Steel Research 66 (2010) 670679

The result is identical to the first buckling mode of Eq. (39), so

the first statement is proven for the case as well. Now if we
examine the second statement and substitute the first buckling
mode of Eq. (39) into the expression of the general vTot /Tot ratio,
the same result as in Eq. (45) cannot be obtained; thus, applying
the first buckling mode for the imperfection the shape of the
displacements will not be constant. The reason for this unexpected
result lies in the considered loading history. If the axial force is
held constant throughout the loading history, it can be treated as
an initial condition of the system. The consequence of this concept
is an altered state in the initial geometry and stiffness; i.e., a LTB
problem of a beam subjected to pure uniform, strong axis bending
moment can be examined with the following initial imperfections
and stiffness:

v0 = v0
0 = 0

(46)

1 N /Ncr ,z
1

(47)

1 N /Ncr ,x

N cr ,z = Ncr ,z N

(48)

N cr ,x = Ncr ,x N

(49)

M cr = McrN = r0

Ncr ,z N

Ncr ,x N .

(50)

Accordingly, the components of the initial geometric imperfection

are amplified by the correspondent factor due to the effect of axial
force (see Eq. (32)), and the elastic critical buckling forces are
reduced by the axial force. If we substitute these expressions into
the Eqs. (19)(21) of the system with pure bending, we receive
the corresponding Eqs. (43)(45) of the system with bending and
compression that have already been obtained. In order to gain the
further results we should change the shape of the imperfection
consequently, as defined by Eq. (39):

v0
M cr
1
McrN
=
=

0
Ncr ,z 1 N /Ncr ,z
N cr ,z
v0
v0 1 N /Ncr ,z
McrN
1
=
=
.
0
0 1 N /Ncr ,x
Ncr ,z 1 N /Ncr ,x

(51)

Applying this new formula for the imperfection, the second statement can be proved again:

vTot
=
Tot

My
McrN
1
1
1
1N /Ncr ,z Ncr ,z 1N /Ncr ,x
Ncr ,z N 1N /Ncr ,x
My
McrN
1
1
1
1N /Ncr ,x
r 2 Ncr ,x N 1N /Ncr ,z Ncr ,z 1N /Ncr ,x
0

1
1 My /McrN

1
1 N /Ncr ,z

McrN
Ncr ,z

My
McrN

My
Ncr ,z

+1

1
vTot
=

2
Tot
1 My /McrN

1 N /Ncr ,z
My
1
r02 Ncr ,x N 1 N /Ncr ,z

1 N /Ncr ,z
.
1

(53)

1 N /Ncr ,x

4.5. Solution for the AyrtonPerry formula

After having the appropriate linear, consistent amplification
relationship of Eq. (53), the next step is the construction of the
first yield criterion in terms of the second order internal forces at
midspan.
My
Wy

N
A

MzII

Wz

BII
Ww

= fy

(54)

where the second order internal forces can be expressed in terms

of the external bending moment, the effect of the constant axial
force, and the total displacement components [17]:
MzII = My Tot + N v0
1
1
1
+ N v0
= My 0
1 N /Ncr ,x 1 My /McrN
1 N /Ncr ,z
BII = My vTot GIt + r02 N 0
1
1
= My v0
1 N /Ncr ,z 1 My /McrN

1
1
GIt 0
0
1 My /McrN 1 N /Ncr ,x
1
2
+ r0 N 0
.
1 N /Ncr ,x

(55)

(56)

These equations may also be obtained by substituting Eqs. (46)

(50) into Eqs. (26)(27) considering the additional effect from
the initial stresses caused by the axial force. Introducing the
slenderness and reduction factor for beamcolumns

s
BC =

Wy fy
McrN

(57)

My

(58)

W y fy

the standard quadratic APF can be obtained:

McrN

Ncr ,z 1 N /Ncr ,z

(52)



1
1
1
2
BC
+ BC N 2 2 BC + 2 N = 0
BC
BC
BC

My

Ncr ,z N 1 N /Ncr ,x

1 N /Ncr ,x

(59)

in which the new generalized imperfection factor (BC ) and the

compression effect factor (N ) are defined as follows:

BC = v0

Wy

+ 0

Wy

1 N /Ncr ,z W
1 N /Ncr ,x Wz
GIt Wy
1

1 N /Ncr ,x Mcr W

(60)

N v0
1
r 2 N 0
1
+
+ 0
Afy Wz fy 1 N /Ncr
W
f
1

N
/Ncr ,x
,z
w y
GIt 0
1

1
.
(61)
Ww fy 1 N /Ncr ,x
The compression effect term N apparently expresses the same

N = 1

v0

Both equalities can be satisfied only if Eq. (51) is valid for the imperfection components, so considering this special requirement for
the imperfection, the system is applicable for the construction of
the APF for LTB of beamcolumns.

BC =

Obviously, this relationship can also be obtained by substituting

Eqs. (46)(50) into Eq. (22), so the shape of displacement proved
to be constant and equal to the first buckling mode throughout
the whole loading history for the interaction problem as well.
The simplest way of the proof of the third statement is the solution of Eq. (23) by applying the expressions of Eqs. (46)(50), but
the solution of the following expounded equation also yields the
same:

675

 
v
0
0

as the main equation for LTB of columns Eq. (36); accordingly,

this term clearly indicates the utilization due to compression.

Author's personal copy

676

J. Szalai, F. Papp / Journal of Constructional Steel Research 66 (2010) 670679

Obviously Eq. (60) becomes identical to Eq. (29) (consequently

Eq. (59) becomes identical to Eq. (28)) if the axial force is equal
to zero, since it involves the pure bending case. Moreover, if the
compression effect term N is equal to zero (which means that
the buckling of the purely compressed member is the dominant
failure mode), the reduction factor BC also becomes zero, since the
member has no further capacity for bending moment. Using these
expressions, the buckling curve for LTB of beamcolumns can be
written as the solution of Eq. (59) in the well-known form of the
EC3 [1]:

N
q
2
BC + BC
N 2BC


= 0.5 1 + BC + N 2BC .

BC =

(62)

BC

(63)

3.0

2.5

2.0

1.5

1.0

0.5

This is the fundamental solution for the APF based LTB curve for
beam-column problem belonging to the first yield criterion of Eqs.
(54)(56) and the specially shaped initial geometric imperfection
defined by Eq. (51).

5000

10000

15000

Fig. 3. Slenderness of a HEA300 cross section at various compression levels.

1.0

5. Discussion of the results

This paper does not undertake to develop formulae appropriate
for design purposes in practice, as a comprehensive work on deterministic and probabilistic calibration would be needed. The main
objective is the mathematical derivation of a generalized APF and
the resulting clear description of LTB. For the sake of a better understanding, however, some particular features of the derived formulae are now illustrated. In these discussions the original domain
and assumptions of the previous chapters are retained, so an analysis of the additional effect of partial plasticity and residual stresses
which would be needed for complete application purposes is
still neglected.

0.5

0.0
0.5
0.0
0.0

1.0
0.5

1.5

n
1.0

2.0

Fig. 4. Interpretation of the LTB reduction factor of Eqs. (62)(63).

5.1. Appropriate interpretation of the derived equations

In examining and understanding the real meaning of the formulae developed, it is very important to note once more that we
have only analyzed the bending moment, considering constant
compression as an initial condition of the system, as stated in
Section 4.4. This means that a check should first be carried out
for pure compression (the pure flexural buckling problem), and
the equations derived are only valid and only yield meaningful
results if the system has strength in reserve for the bending moment. Accordingly, the special slenderness of Eq. (57) depends not
only on the usual geometrical and material parameters, but also
on the compression level. In consequence, the slenderness has no
mechanical meaning if the compression level exceeds the buckling
load (minimum of Ncr .z or Ncr .x ). This feature is illustrated in Fig. 3,
where the slenderness of an HEA300 section is plotted against the
member length at various compression levels, where the usual
compression utilization is defined as: n = N /Npl and the slenderness of pure compression (comp ) as defined in Section 2.1 is also
plotted as a thin solid line, for the sake of comparison. In addition,
a dashed line represents the limit where the whole capacity is utilized by the buckling resistance of the purely compressed member; i.e. the compression effect term N of Eq. (61) is equal to zero.
Above this limiting line, the curves obviously have no meaning. It
can be seen that the slenderness curves approach infinity asymptotically at a certain length associated with the elastic critical
(Euler) compression force (since the McrN value in Eq. (57) becomes
zero), but prior to that the slenderness is limited by the dashed line.
The major part of the curves lies between the slenderness of pure
compression and pure bending; at higher values, they exceed the
slenderness of pure compression slightly.

If the reduction factor of Eqs. (62)(63) is evaluated at various

member lengths and compression levels, a resistance surface can
be obtained (Fig. 4). For a certain problem i.e. with a certain
length and a certain compression level the reduction factor can
be represented graphically as an intersection line of this resistance
surface and a vertical plane defined by the compression utilization
n. It can easily be seen that this intersection line reaches zero at the
slenderness value associated with the buckling resistance of the
purely compressed member; and beyond this limit the reduction
factor has no mechanical meaning, as explained in the previous
paragraph. The intersection line at n = 0 becomes identical to the
LTB curve of Eqs. (30)(31).
5.2. The form of the imperfection factors
One of the most important results of the exact derivation
of the APF formula for LTB of beams and beamcolumns is the
appropriate form of the generalized imperfection factors of Eqs.
(29) and (60). First of all, these factors are the basis upon which the
multiple buckling curves for LTB can be determined and correctly
calibrated as discussed in the following section. A further
advantage is that the equivalent member imperfections (included
in EC3 for column buckling only) can also be defined for LTB. In this
section, we deal with the structure and specialties of the analytical
description of the introduced imperfection factors. It is important
to see that the results are only valid if the shape of imperfection
satisfies the criterion of Eq. (11), or more generally that of
Eq. (39). It is also relevant that the initial deflection component has
been studied much more comprehensively both experimentally
and analytically [24] due to its importance in column buckling

Author's personal copy

J. Szalai, F. Papp / Journal of Constructional Steel Research 66 (2010) 670679

0.035

677

0.035

0.030

0.030

0.025
0.025
0.020
0.020
0.015
0.015
0.010
0.010
0.005
0.005
5000

10000

15000

20000
0

Fig. 5. The initial twist of a member with different cross sections at the same
amplitude for initial out-of straightness.

and the simplicity of measuring it compared to the initial twist.

Consequently, only the initial deflection component has usually
been considered in numerical research of LTB [10,12]; the initial
twist is rarely accounted for (mainly in analytical solutions [11,
16]).
Accordingly, the effect of initial twist is examined first. In
Fig. 5 the initial twist is plotted against the member length for
four different profiles (calculated from Eq. (11) at v0 = L/1000
amplitude for the initial deflection). It can be seen that, although
the initial deflection increases linearly with the member length,
the initial twist has a maximum value around the medium
slenderness range. It is also worth noting that the heavier the cross
section, the smaller the initial twist. This follows from the fact that
heavier cross sections have greater torsional rigidity (especially the
St. Venant term) compared with lateral rigidity. The next question
which may arise is how the initial twist contributes to the total
imperfection effect. In Fig. 6 the LTB imperfection factor (LT
defined by Eq. (29)) normalized by the initial deflection amplitude
(v0 ) is plotted against the member length for the heaviest and
lightest profile of the previous cross sections. For comparison
purposes, the imperfection factors of lateral column buckling (z
defined in Section 2.1) are also plotted (dashed lines), and the
contributions of the initial deflection (first term in Eq. (29)) are
drawn separately. The following observations can be made from
this figure:

the LTB imperfection factors vary (decrease) by the member

length because of the contribution of initial twist; this effect
vanishes towards higher slenderness;
the imperfection factors of column buckling are generally
higher then those of LTB except for the light cross section in the
low slenderness range;
while the initial deflection part (first term) is the same for the
two profiles (the imperfection factors of column buckling are
also quite similar), the light cross section shows considerably
higher LTB imperfection factors, especially in the important
medium slenderness range.
5.3. The effect of cross-section geometry
In modern structural standards, the APF type buckling model
provides a suitable base for a sophisticated stability design method,
known as the multiple buckling curve approach. This method takes
into account the fact (detected mainly by experiments) that the
buckling resistance of a member in a non-dimensional form (i.e. in
the form of non-dimensional buckling curve) depends highly on
the shape and geometry of the cross section [68]. This dependence
comes from two primary sources:

5000

10000

15000

20000

Fig. 6. The shape and structure of the LTB imperfection factors for stocky and
slender cross sections.

1. different imperfection sensitivity of members with different

cross sections i.e. members with equal imperfections have
different buckling resistances;
2. different imperfection magnitude arising during production in
members with different cross sections i.e. members with
unequal imperfections have different buckling resistances.
It is important to note that the type 2 problem is a measurement
task, and is thus not relevant to the mechanical investigations.
The source of imperfections can be geometrical (member out of
straightness, cross sectional imperfections) and material (residual
stresses), but as stated earlier in the APF only the geometric imperfections are considered. An example for the type 1 dependence
is the distinction between column buckling about the major axis
(y) or the minor axis (z ) [1], where the relevant imperfection formulae can be expressed as follows:

y = w0
z = v0

(64)

(65)

Wy
A

Wz

If we consider the same amplitude for both directions (i.e. w0 =

v0 ). it is apparent that the major axis imperfection factor has a
considerably lower value (since Wy  Wz ) so the buckling about
the major axis is less sensitive to imperfections and consequently
yields higher values for the non-dimensional buckling curves.
In the recent EC3, a distinction is made between the welded and
hot-rolled profiles of the I-shaped section in the case of LTB, which
obviously reflects the different magnitude of residual stresses, but
the code also differentiates the sections by the depth over width
(h/b) ratio, and this attempts to represent the effect of the cross
section geometry on the imperfection sensitivity and thus on the
LTB resistance. This latter effect already appears in the APF based
resistance definition, so it can be examined here. Fig. 7 shows the
LTB reduction factor for pure bending Eqs. (30)(31) with 8 wide
flange profiles (solid lines from thick to thin: HEM300, HEB300,
HEA300, HEAA300; dashed lines from thick to thin: HEM900,
HEB900, HEA900, HEAA900) and Fig. 8 shows the LTB reduction
factor at n = 0.3 compression level Eqs. (62)(63) with the same
profiles; in both figures the elastic critical curve is also plotted as
a reference. The curves show considerable scatter mainly in the
domain of medium slenderness, which is obviously caused by the
different influences of the imperfection factor on different cross
section shapes. As demonstrated in Section 5.2, in this region the
heavier profiles produce a lower imperfection factor due to their
significantly higher St. Venant torsional rigidity. It is important to
note that there are much greater differences between the solid

Author's personal copy

678

J. Szalai, F. Papp / Journal of Constructional Steel Research 66 (2010) 670679

1.0

0.20
0.8

0.6
0.15

0.4

1.5

0.2

0.0

0.5

1.0

1.5

2.0

2.5

2.0

2.5

3.0

3.0

Fig. 9. The LTB imperfection factors of various cross sections against h/b ratio.
Fig. 7. LTB reduction factors for stocky and slender cross sections for pure bending.

1.0
0.20
0.8

0.6

0.15

0.4
15

20

25

30

0.2

0.0

0.5

1.0

1.5

2.0

2.5

3.0

Fig. 8. LTB reduction factors for stocky and slender cross sections for interactive
compression and bending (n = 0.3).

lines (300 profiles), although these sections have similar h/b ratios.
In order to study this peculiarity from the point of view of reduction
factors for multiple buckling curves, in Fig. 9 we have plotted the
LTB imperfection factors at v0 = L/1000 amplitude for all IPE,
HEA and HEB type profiles against the h/b ratio (small circle
IPE, medium circle HEA, large circle HEB). The EC3 separates
the cross sections by the h/b = 2 value, for profiles under this
limit, the use of a more favourable buckling curve is allowed.
Observing Fig. 9, an increasing tendency can actually be seen
towards higher h/b ratios, but there are also significant differences
between profiles on the same side of the limit, especially for HEA
and HEB sections where the h/b ratio is around 1, and IPE sections
where the h/b ratio is under 2. This is apparently the consequence
of the different torsional rigidity of cross sections with similar
h/b ratio, as described in the previous section. Although the h/b
ratio is an important geometrical measurement, the St. Venant
torsional rigidity of I profiles is highly influenced by the b/tf
ratio, so the product of these two measurements should be more
appropriate. Accordingly, in Fig. 10 the same imperfection factors
are plotted against the h/b b/tf = h/tf ratio. From this figure, it
can be concluded unambiguously that the h/tf ratio describes the
imperfection sensitivity of the profiles significantly better, and is
a more adequate basis for the multiple buckling curves in the case
of LTB.
6. Conclusion
In this paper, a theoretical model is presented for the possible generalization of the APF, which is widely used for the stability design of members in modern structural standards. The direct

Fig. 10. The LTB imperfection factors of various cross sections against h/tf ratio.

aim of the generalization is the rigorous derivation of the APF for

lateraltorsional buckling problems of beams (pure bending) and
beamcolumns (bending and compression). It has been demonstrated that the key component of the APF is the amplification
relationship, which should be linear and consistent in the cases examined, in order to obtain the basic quadratic equation of the conventional APF. It has been proved, that if we apply the first buckling
mode for the shape of initial geometric imperfection of the member, the amplification relationship becomes linear and consistent.
According to this statement, the APF was derived for the pure bending case and a special generalized imperfection factor was defined
for the buckling curve of LTB. This procedure was also extended to
the case of interactive loading (bending and compression), where
the appropriate way of handling the compression effect was first
introduced, and it was demonstrated that the actual amplification
relationship becomes linear and consistent only if the compression
is kept constant throughout the loading history. Accordingly, this
interaction problem can be treated as an equivalent LTB problem of
a beam subjected to a pure, uniform, strong axis bending moment
with special initial conditions, resulting from an altered state in
the initial geometry and stiffness caused by the constant compression effect. One of the most important results of the exact derivation of the APF formula for LTB of beams and beamcolumns is the
appropriate form of the generalized imperfection factors. Firstly,
these factors can be the basis upon which the multiple buckling
curves for LTB can be determined and calibrated correctly. A further advantage is that the equivalent member imperfections (included in EC3 for column buckling only) can also be defined for
LTB. An important result of the short, introductory discussions of
the formulae obtained is that since the initial twist of the member has strong significance in the issues covered by this paper the
torsional rigidity of the cross section is of great importance and has

Author's personal copy

J. Szalai, F. Papp / Journal of Constructional Steel Research 66 (2010) 670679

to be considered in the correct definition of the generalized imperfection factor and multiple buckling curves.
References
[1] European Standard, EuroCode 3. Design of Steel Structures Part1-1: General
rules and rules for buildings, EN 1993-1-1, 2005.
[2] Galambos TV. Guide to stability design of metal structures. Wiley; 1995.
[3] Strating J, Vos H. Computer simulation of the E.C.C.S. buckling curves using a
Monte-Carlo method. HERON 1973;19(2).
[4] Bjorhovde R. Deterministic and probabilistic approaches to the strength of
steel columns. Ph.D. dissertation. Bethlehem: Lehigh University; 1972.
[5] Ayrton WE, Perry J. On struts. The engineer; 1886.
[6] Beer H, Schultz G. Theoretical basis for the European column curves. Construction Mtalique 1970;3.
[7] Maquoi R, Rondal J. Mise en Equation des Nouvelles Courbes Europennes de
Flambement. Revue Construction Mtalique 1978;1.
[8] Janns J, Sedlacek G, Maquoi R, Ungermann D, Kuck J. Evaluation of test results
on columns, beams and beamcolumns with cross-sectional classes 1-3 in
order to obtain strength functions and suitable model factors. Background
report to Eurocode 3 Common unified rules for steel structures 1992.

679

[9] Chen WF, Atsuta T. Theory of beamcolumns. McGraw-Hill; 1977.

[10] Boissonade N, Jaspart JP, Muzeau JP, Villette M. Improvement of the interaction formulae for beam columns in Eurocode 3. Computers and Structures
2002;80:237585.
[11] Maquoi R, Janss J. EC3 design model for lateral torsional buckling resistance.
In: International conference on steel and aluminium structures. 1991.
[12] Greiner R, Salzgeber G, Ofner R. New lateraltorsional buckling curves
LT - numerical simulations and design formulae. ECCS TC8 report no.
TC8-2000-014, 2000.
[13] Szalai J. Analysis of the resistance of steel beamcolumns on probabilistic
basis. Ph.D. dissertation. Budapest University of Technology and Economics;
2005.
[14] Szalai J, Papp F. On the design curves for buckling problems. In: 5th European
conference on steel and composite structures. Eurosteel 2008.
[15] Greiner R. Background information on the beam-column interaction formulae
at level 1. ECCS TC 8, Ad-hoc working group on beamcolumns, Technical
University Graz; 2001.
[16] Boissonnade N, Villette M, Muzeau JP. About amplification factors for lateraltorsional buckling and torsional buckling. In: Festschrift Richard Greiner,
TU Graz; 2001.
[17] Papp F. Computer aided design of steel beam-column structures. Ph.D. dissertation. Edinburgh: Budapest University of Technology and Economics; 1994.