Journal of Constructional Steel Research 63 (2007) 894–908

www.elsevier.com/locate/jcsr

On the evaluation of elastic critical moments in doubly and singly symmetric

I-section cantilevers
A. Andrade a , D. Camotim b,∗ , P. Providência e Costa a
a Department of Civil Engineering, University of Coimbra, Pólo II, Pinhal de Marrocos, 3030-290 Coimbra, Portugal
b Department of Civil Engineering and Architecture, ICIST/IST, Technical University of Lisbon, Av. Rovisco Pais, 1049-001 Lisboa, Portugal

Received 15 June 2006; accepted 31 August 2006

Abstract

The so-called 3-factor formula is one of the most commonly employed general formulae to estimate the elastic critical moment of steel beams
prone to lateral-torsional buckling. This work extends its domain of application to I-section cantilevers (i) with equal or unequal flanges, (ii) fully
built-in or free to warp at the support and (iii) acted on by uniformly distributed or concentrated tip loads (applied either at the shear centre or at
one of the flanges). The paper includes (i) a discussion of the theoretical basis of elastic lateral-torsional buckling, (ii) the description of the main
steps involved in posing the buckling problem in a non-dimensional form over a fixed reference domain, features that are particularly convenient
for the purpose of this work, (iii) the numerical results of a parametric study, obtained by the Rayleigh–Ritz method, and (iv) their use for the
development of approximate analytical expressions for the C1 , C2 and C3 factors appearing in the aforementioned formula.
c 2006 Elsevier Ltd. All rights reserved.

Keywords: Lateral-torsional buckling; Elastic critical moments; I-section cantilevers; Dimensional analysis; Rayleigh–Ritz method; Eurocode 3

1. Introduction One of the most commonly employed general formulae

to estimate elastic critical moments (Mcr ) is the so-called
According to virtually all the design codes currently 3-factor formula, which was included in the ENV version
employed in the steel construction industry (e.g., [1–4]), the of Eurocode 3 [5]—the recently completed EN version of
load carrying capacity of laterally unsupported beams bent this design standard [1] provides no information concerning
in their stiffer principal plane is estimated on the basis of the determination of Mcr . In theory, this formula should be
their (i) cross-sectional (direct stresses) and (ii) elastic lateral- applicable to uniform beams (or beam segments) subjected
torsional buckling (LTB) resistances—therefore, designers to major axis bending, exhibiting doubly or singly symmetric
must have easy access to reliable methods for computing cross-sections (with the proviso that the cross-sections are
the elastic critical moment or load of a given beam. Of symmetric with respect to the minor central axis) and arbitrary
course, this calculation can be carried out by means of a support and loading conditions. But some situations are
number of numerical techniques, which basically involve the currently not covered, most notably the case of cantilevers,
approximation of either (i) the individual terms in the governing which are free to twist and deflect laterally at one end—
differential equations (e.g., finite difference methods) or (ii) information about the factors that must be introduced in the
the function spaces associated with an underlying variational formula is rather scarce and/or incomplete. In a 1960 review
principle (e.g., finite element methods). However, until user- article, Clark and Hill [6] presented lower and upper bounds for
friendly software packages implementing such numerical the C1 factor applicable to cantilevers restrained from warping
procedures become widely available to practitioners, simplified at the fixed end and acted on by a uniform load or by a
methods such as design charts and approximate formulae will point load applied at the tip—in the latter case, a C2 factor
retain their popularity and continue to be extensively used. was also proposed. Moreover, C1 and C2 factors for these
two types of loading were also published by Galéa [7] and
∗ Corresponding author. Tel.: +351 21 8418403; fax: +351 21 8497650. Baláz and Koleková [8], but the warping restraint condition
E-mail address: [email protected] (D. Camotim). at the support was not specified. However, the authors did

c 2006 Elsevier Ltd. All rights reserved.

0143-974X/$ - see front matter
doi:10.1016/j.jcsr.2006.08.015
 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908 895

not find in the literature any proposal for the C3 factor, a

fact that automatically rules out the possibility of dealing with
singly symmetric cantilevers. At this point, a word of caution is
required: there are slight differences in the formulae developed
by the above authors—for instance, none of the formulae
proposed by Clark and Hill or Galéa contain an effective
length factor kw , unlike the formula included in [5] and also
considered by Baláz and Koleková.
Over the years, a fair amount of research work has been
devoted to develop approximate formulae to assess the buckling
behaviour of I-beams. In the following paragraphs, a brief Fig. 1. (a) Undeformed singly symmetric I-section cantilever and external
state-of-the-art survey on this subject is presented, focusing loads. (b) Schematic representation of the warping conditions at the fixed end.
specifically on cantilevers.
The elastic LTB of equal-flanged I-section cantilevers with equal or unequal flanges, (ii2 ) fully built-in or free to
built-in at the support and under shear centre loading was warp at the support and (ii3 ) acted by uniform or tip loads
investigated by Timoshenko [9] (point load at the tip) and by (applied at one of the flanges or at the shear centre), thus making
Poley [10] (uniform load). In the 70s, Nethercot [11] extended available to practitioners a reliable and useful design tool. The
these early works by considering the effects of (i) top flange paper includes a discussion of the theoretical basis of elastic
loading, (ii) the type of restraint condition at the tip and (iii) LTB, as well as the description of the main steps involved in
the continuity between overhanging and internal segments—in posing the buckling problem in a non-dimensional form over
addition, this author proposed a set of effective length factors a fixed reference domain—a particularly adequate setting for
to estimate Mcr . On the other hand, Trahair [12] developed the systematic development of approximate buckling formulae.
approximate formulae to estimate the elastic buckling capacity This approach, which has its roots in dimensional analysis,
of built-in cantilevers and of overhanging segments free to differs significantly from the one proposed by Attard [19],
warp at the support, without any interaction with the internal who has non-dimensionalised a general expression for Mcr
segments—he also proposed a simplified method to account for written in terms of an assumed mode shape. Finally, note that
the interaction between adjacent segments and studied the effect preliminary and partial versions of this work were reported
of the presence of an elastic torsional restraint at the support. in [20–22].
However, the above studies were restricted to doubly symmetric
I-section members—note that the elastic critical moment of 2. Theoretical background
doubly symmetric cantilevers free to warp at the support can
also be estimated by a single factor formula included in a 2.1. Problem statement
recently published design guide [13]. By applying curve fitting
techniques to a collection of finite element results, Doswell [14] We consider a prismatic cantilever with a doubly or singly
proposed an improvement of the approximate Mcr formula symmetric I-section (i.e. equal or unequal flanges) and length
included in the current AISC specification [3]—this proposal l, as the one shown in Fig. 1(a). It is convenient to introduce
takes explicitly into account, through individual coefficients, a fixed rectangular right-handed Cartesian reference system
the effects of (i) the bending moment diagram (two loading such that, in the undeformed configuration, (i) the longitudinal
case are analysed: tip load and uniform load), (ii) the level of centroidal axis of the beam coincides with the x-axis, (ii) y and
load application (only the cases of shear centre and top flange z are the cross-section major and minor central axes and (iii)
loading are considered) and (iii) the effect of top flange bracing the top flange corresponds to negative z values (see Fig. 1(a))—
(either continuous bracing or a discrete bracing at the free end). thus, y = 0 is a symmetry plane for the undeformed member.
Once more, only equal-flanged I-sections were dealt with and The end cross-sections are initially contained in the planes x =
the cantilevers are always assumed to be fully built-in at the 0 and x = l and the locus of shear centres is the straight-line
support. segment defined by {(x, y, z) ∈ R3 : x ∈ [0, l], y = 0, z = z S }.
The elastic LTB of singly symmetric I-section cantilevers With no loss of generality, we assume that x = 0 corresponds
fully built-in at the support and acted on by a tip load or to the fixed end section, where the lateral deflections (along
a uniform load was studied by Anderson and Trahair [15], y), minor axis rotations (about z) and torsional rotations (about
Roberts and Burt [16] and Wang and Kitipornchai [17], who the shear centre axis) are prevented and warping is either fully
provided tables, charts and/or approximate expressions to restrained or completely free—these two warping conditions at
calculate elastic critical moments/loads. Moreover, the effect of the fixed end are schematically represented in Fig. 1(b) and will
intermediate restraints was investigated by Wang et al. [18]. be subsequently identified by the acronyms NW and FW, which
This paper attempts to fill-in two of the insufficiencies stand for No Warping and Free Warping, respectively.
identified in the above studies, by (i) examining the influence The cantilever is made of a Saint–Venant/Kirchhoff
of the warping restraint condition on the elastic LTB of material [23], with Young modulus E and shear modulus G.
singly symmetric cantilevers and (ii) extending the domain of It is deemed to be subjected to the conservative transverse
application of the 3-factor formula to I-section cantilevers (ii1 ) loads depicted in Fig. 1(a) – a uniform load q = qk and a
896 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908

tip load Q = Qk, where k is the unit vector directed along (recall that V and Φ are both identically zero in the fundamental
the z-axis −, both with the same direction and initially acting state), (ii) Iz , It and Iw are standard geometrical properties
on the plane y = 0. Their magnitudes q, Q are assumed to of the undeformed cross-section, (iii) β y is a cross-sectional
be proportional to a single factor λ (therefore, we may write asymmetry property, defined by
q = q0 λ and Q = Q 0 λ, where q0 , Q 0 are non-negative Z
1
reference magnitudes defining the loading profile) and their βy = z( y 2 + z 2 )dA − 2z S , (2)
conservative character is ensured by the fact that they follow Iy A
the beam deformation, always remaining parallel to the z-axis. f
(iv) M y denotes the bending moment distribution in the
For convenience, we also assume throughout this section that
fundamental state, given by
q0 and Q 0 are strictly positive.  
It is well-known that such a cantilever is prone to lateral- f 1
M y (x, λ) = − q0 (l − x) + Q 0 (l − x) λ
2
torsional buckling, a bifurcation-type of instability (from the 2
Latin word bifurcus—‘with two branches’) where (i) the f
fundamental path corresponds to equilibrium configurations = M y0 (x)λ, (3)
(parameterised by λ) that are symmetric with respect to the f
where M y0 is a function of x alone and denotes the bending
plane y = 0 (i.e. the cantilever is subjected solely to major
moment diagram caused by the reference loads q0 k and Q 0 k,
axis bending) and (ii) the buckled states are associated with
and (v) z q and z Q identify the point of load application—
non-symmetric configurations—the cantilever deflects laterally,
see Fig. 1(a). In the sequel, we exclude the limiting cases of
along y, and twists. This phenomenon is an obvious instance
T, inverted T and narrow rectangular cross-sections, so as to
of symmetry breaking. However, not all symmetry is lost:
ensure that Iw 6= 0.
given any possible buckled state, its reflection upon the plane
In order to completely define the functional δ 2 Π , it is
y = 0 is also a possible buckled state—i.e. instability breaks
necessary to specify the class of admissible functions v and
the symmetry of the equilibrium configuration, but not the
φ. There are two properties that must be characterised: (i) the
symmetry of the solution set. smoothness required of these functions and (ii) the boundary
In this section, we aim at formulating the problem conditions they must satisfy. The integrals in (1) only make
of identifying the bifurcation points along the cantilever’s sense if v and φ are square-integrable in (0, l) and possess
fundamental path in a way that is well suited for the square-integrable first and second derivatives. In addition, the
systematic development of approximate buckling formulae. In admissible functions v and φ must satisfy the homogeneous
order to linearise the problem, we neglect the pre-buckling form of the essential (Dirichlet) boundary conditions, i.e.
flexural deflections, thus assuming that the cantilever remains
straight up until the onset of buckling—in design, it is v(0) = 0 v,x (0) = 0 φ(0) = 0 φ, x (0) = 0, (4)
prudent to disregard the beneficial effect of pre-buckling
the last one applying only if warping is prevented at the fixed
deflections because (i) it may be significantly reduced, or
end section (NW). The real-valued functions fulfilling the above
even fully removed, by any pre-cambering of the member
requirements are termed kinematically admissible.
and (ii) it depends heavily on the actual conditions of lateral
According to Trefftz’s criterion [27,28] – for a critical
restraint [24]. Moreover, the analysis is carried out under the appraisal of this criterion, together with a brief historical
assumption of small strains. account and references to its application, the interested reader is
referred to [29] –, the variational form of the buckling problem
2.2. Mathematical modelling
can be stated as follows:
Using Vlassov’s classical thin-walled beam theory hypothe- Problem A (Variational Form). Find real scalars λ and
ses [25] – the cross-sections do not deform in their own plane kinematically admissible functions v, φ 6= 0 rendering δ 2 Π
and the shear strains on the mid-surface are negligible −, it is stationary, i.e. satisfying the variational condition
possible to show that the second variation, from a given funda-
mental state, of the cantilever total potential energy Π is defined δ (δ 2 Π ) = 0. (5)
by (e.g., [26]) Applying standard Calculus of Variations techniques [30],
E Iz l 2
Z
G It l 2
Z
E Iw l 2
Z we obtain the classical or strong form of Problem A, which
δ2Π = v,x x dx + φ,x dx + φ dx may be phrased as follows:
2 0 2 0 2 0 ,x x
1 l f Problem A (Strong Form). Find λ ∈ R and real-valued
Z
1
+ M y (2v,x x φ + β y φ,x
2
)dx + (z q − z S )q functions v, φ ∈ C 4 ([0, l]), with v, φ 6= 0, satisfying the
2 0 2
Z l differential equations
1
× φ 2 dx + (z Q − z S )Qφ(l)2 , (1) 
f

0 2 E Iz v,x x x x + M y0 φ λ=0 (6)
,x x
where (i) v(x) and φ(x) are variations of V (x, λ) (shear f f
(M y0 v,x x + q0 (z q − z S )φ − β y (M y0 φ,x ),x )λ − G It φ,x x
centre displacement along y) and Φ(x, λ) (rotation about the
undeformed shear centre axis), which are independent from λ + E Iw φ,x x x x = 0 (7)
 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908 897

in (0, l) (Euler–Lagrange equations of δ 2 Π ) and the boundary 2.3. Equivalent scaled problems
conditions
By an appropriate change of variable and subsequent scaling
v(0) = 0 Q 0 φ(l)λ + E Iz v,x x x (l) = 0 (8) of the functions v and φ, Problems A and B may be transformed
v,x (0) = 0 E Iz v,x x (l) = 0 (9) into equivalent ones, posed in a fixed reference domain
φ(0) = 0 (z Q − z S )Q 0 φ(l)λ + G It φ,x (l) (i.e. independent of the cantilever length l) and written in a non-
dimensional form—this is most convenient because it enables
− E Iw φ,x x x (l) = 0 (10)
a general and systematic representation of the cantilever LTB
φ,x (0) = 0 (NW) or E Iw φ,x x (0) = 0 (FW) behaviour.
E Iw φ,x x (l) = 0. (11) Since the interval [0, 1] is taken as the fixed reference
domain, the associated change of variable is defined by the
Due to the regularity of the data – constant mechanical and
f stretch ξ : [0, l] → [0, 1], ξ = x/l. Moreover, we introduce
geometrical properties and M y0 ∈ C ∞ ([0, l]) –, the strong and
variational forms of the problem are equivalent. p scaled non-dimensional functions v : [0, 1] → R, v(ξ ) =
the
E Iz /(G It )v(ξl)/l and φ : [0, 1] → R, φ(ξ ) = φ(ξl). Then,
From a mathematical viewpoint, Problem A is an the strong form of the scaled versions of Problems A and B may
eigenvalue problem—the elastic bifurcation load factors and be stated as follows:
the corresponding buckling modes are its eigenvalues and Problem C (Strong Form). Find λ ∈ R and v, φ ∈ C 4 ([0, 1]),
eigenfunctions. Given the self-adjoint character of this problem, with v, φ 6= 0, satisfying the differential equations
all eigenvalues are real and the eigenfunctions may be taken to
f
be real [31,32]—this is the mathematical justification for having v ,ξ ξ ξ ξ + (µ0 φ),ξ ξ λ = 0 (14)
formulated Problem A in terms of real scalars and real function f f
spaces, in line with its physical nature. The lowest positive (µ0 v ,ξ ξ + 2εq γq φ − δ y (µ0 φ ,ξ ),ξ )λ − φ ,ξ ξ
eigenvalue is termed elastic critical load factor and denoted by K2
λcr —accordingly, the corresponding buckling mode (vcr , φcr ) + φ =0 (15)
π 2 ,ξ ξ ξ ξ
is labelled critical as well. We also speak of the elastic critical
in (0, 1) and the boundary conditions
moment, defined as
f v(0) = 0 γ Q φ(1)λ + v ,ξ ξ ξ (1) = 0 (16)
Mcr = |M y (x, λcr )|max = (1/2q0l 2 + Q 0l)λcr
f v ,ξ (0) = 0 v ,ξ ξ (1) = 0 (17)
= −M y0 (0)λcr .
K2
φ(0) = 0 ε Q γ Q φ(1)λ + φ ,ξ (1) − φ (1) = 0 (18)
Integrating twice the field equation (6), together with the π 2 ,ξ ξ ξ
natural (Neumann) boundary conditions (82 )–(92 ), we are led
to the conclusion that v,x x and φ are related by the equation φ ,ξ (0) = 0 (NW) or φ ,ξ ξ (0) = 0 (FW) φ ,ξ ξ (1) = 0,(19)

f
where
M y0 l
v,x x = − φλ. (12) f
µ0 (ξ ) = √
f
M (ξl)
E Iz E Iz G It y0
This equation can be viewed as a holonomic kinematical = −(1 − ξ )((1 − ξ )γq + γ Q ) (20)
constraint [33], which can be used to eliminate v and write the
and
buckling problem in terms of a single unknown field φ. Indeed, s
the incorporation of (12) into (7) yields q0 l 3 Q 0l 2 π E Iw
γq = √ γQ = √ K = (21)
f2
M y0 2 E I z G It E I z G It l G It
f
− φλ2 + (q0 (z q − z S )φ − β y (M y0 φ,x ),x )λ s s
E Iz β y E Iz z q − z S E Iz
δy = εq =
− G It φ,x x + E Iw φ,x x x x = 0 (13) l G It l G It
s
and Problem A may then be replaced by the following one, z Q − z S E Iz
which takes on the form of a quadratic eigenvalue problem: εQ = (22)
l G It
Problem B (Strong Form). Find λ ∈ R and φ ∈ C 4 ([0, l]), are non-dimensional parameters, very much identical to the
with φ 6= 0, satisfying the field equation (13) and the boundary ones proposed by Anderson and Trahair [15].
conditions (10) and (11).
Problem D (Strong Form). Find λ ∈ R and φ ∈ C 4 ([0, 1]),
It is obvious that a variational characterisation can be given for with φ 6= 0, satisfying the differential equation
Problem B: it suffices to introduce (12) into (1) and to consider f2 f
again Eq. (5). If required, v can be obtained through the −µ0 φλ2 + (2εq γq φ − δ y (µ0 φ ,ξ ),ξ )λ − φ ,ξ ξ
mere integration of Eq. (12) subjected to the initial conditions K2
(81 )–(91 ). + φ =0 (23)
π 2 ,ξ ξ ξ ξ
898 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908

in (0, 1) and the boundary conditions in terms of a complete set of n − r = 6 non-dimensional

parameters:
K2
φ(0) = 0 ε Q γ Q φ(1)λ + φ ,ξ (1) − φ (1) = 0 (24) π1 = g(π2 , π3 , π4 , π5 , π6 ).
π 2 ,ξ ξ ξ (30)
φ ,ξ (0) = 0 (NW) or φ ,ξ ξ (0) = 0 (FW) φ ,ξ ξ (1) = 0. (25) There is an infinite number of different complete sets of non-
dimensional parameters. One possible choice is π1 = γq λcr
Again, it is readily seen that an equivalent variational
(or, alternatively, π1 = γ Q λcr ), π2 = K , π3 = δ y , π4 =
characterisation can be provided for Problems C and D. For
εq , π5 = ε Q and π6 = q0l/Q 0 , which arises naturally during
instance, (14) and (15), (162 ), (172 ), (182 ), (192 ) and (191 ) (the
the manipulation of the governing equations.
last one only if the fixed end section is free to warp—FW) are
It should be stressed that, up to this point, the shape of
the Euler–Lagrange equations and natural boundary conditions
the cross-section has not yet been invoked in any of the
related to the vanishing of the first variation of the functional
derivations. Thus, all the previous results are valid for thin-
1 1 2 1 1 2 K2 1 2 walled cantilevers with arbitrary singly symmetric open cross-
Z Z Z
δ Π =
2
v dξ + φ dξ + φ dξ
2 0 ,ξ ξ 2 0 ,ξ 2π 2 0 ,ξ ξ sections under major axis bending.
Since the parameters K , δ y , εq , ε Q do not convey a clear
1 1 f
Z  Z 1
2 2
+ µ0 2v ,ξ ξ φ + δ y φ ,ξ dξ + εq γq φ dξ perception of the cantilever slenderness, asymmetry and load
2 0 0 position, we adopt a new set of non-dimensional parameters,
! which (i) apply specifically to I-section beams and (ii) are
1
+ ε Q γ Q φ(1) λ,
2
(26) closely related to the geometrical and mechanical properties
2 governing their LTB behaviour. They are similar to the ones
proposed by Kitipornchai and Trahair [36] and given by
whose admissible arguments v, φ (i) lie in the space of square-
s
integrable functions in (0, 1) with equally square-integrable first
π E Iz h 2S Ib f − It f zq − z S
and second derivatives and (ii) satisfy (161 ), (171 ), (181 ) and, K = ψf = ζq = 2
when warping is prevented at the fixed end section (NW), also l 4G It Ib f + It f hS (31)
(191 )—note that zQ − zS
ζQ = 2 ,
hS
l 2
δ2Π = δ Π. (27) where h S is the distance between the flange mid-lines and Ibf
G It
and Itf are the moments of inertia, about the z-axis, of the
Once λcr has been obtained, the elastic critical moment is bottom and top flanges. It is worth noting that:
given simply by
(i) The parameter ψ f (i1 ) is purely geometrical (i.e. inde-
√ f
E I z G It pendent of the M y sign), unlike the ones defined in [5,
Mcr = (γq + γ Q )λcr . (28)
l 37], (i2 ) provides an easy-to-feel measure of the flange
From either of the above two problems, we conclude that asymmetry and (i3 ) varies from ψ f = −1 (T-section) to
“cantilever families” sharing the same (i) K , δ y , εq , ε Q and ψ f = 1 (inverted T-section), with ψ f = 0 for an equal
q0l/Q 0 values and (ii) warping restraint condition at the flanged I-section. It is related to the warping constant Iw
fixed end (NW or FW) exhibit identical values of γq λ or through [36,37]
γ Q λ at buckling. In other words, the cantilevers belonging to Iw = (1 − ψ 2f )Iz h 2S /4 (32)
such “families” are similar, as far as the LTB behaviour is
concerned—thus, γq λcr or γ Q λcr can be written as functions and may also be used to approximate β y by means of [37]
of K , δ y , εq , ε Q and q0l/Q 0 . βy ≈
It is worth pointing out that the same conclusion can be (
reached through the use of Buckingham’s π-theorem [34], −0.8ψ f h S if ψ f > 0 (i.e. favourable flange asymmetry)
(33)
which summarises the entire theory of dimensional analysis— −ψ f h S if ψ f < 0 (i.e. unfavourable flange asymmetry)
Buckingham himself did not rigorously prove the theorem (note that β y obviously vanishes for an equal flanged
that bears his name, although he presented evidence making I-section). An assessment of the errors involved in the
its validity seem plausible (a proof of this theorem can be approximation (33) and in the subsequent evaluation of
found, for instance, in [35]). On the basis of a purely physical γq λcr and γ Q λcr is presented in an appendix at the end
(i.e. phenomenological) reasoning, the following unspecified of this paper.
functional relationship, involving n = 9 quantities, may be (ii) In commonly used I-section cantilevers, the beam
stated: parameter K ranges from 0.1 to 2.5 [17,36]. Low (high)

q0 l
 K values mean long (short) cantilevers and/or compact
Mcr = f E Iz , G It , E Iw , β y , l, (z q −z S ), (z Q −z S ), .(29) (slender) cross-sections. Furthermore, one has K = K for
Q0
an equal flanged cantilever (ψ f = 0).
The rank of the associated dimensional matrix is r = 3 (iii) The load position parameters ζq and ζ Q enable an easy
and Buckingham’s π -theorem asserts that (29) can be written visualisation of the location of the load point of application
 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908 899

along the cross-section height (or, to be exact, its location factorisation of K (K = LLT , where L stands for a lower
in between the flange mid-lines). Typical values of these triangular matrix with positive diagonal entries) and reduce (39)
parameters are (iii1 ) ζq(Q) = 0 (load applied at the shear to the standard symmetric form
centre), (iii2 ) ζq(Q) = −(1 + ψ f ) (load applied at the top
1
flange centroid) and (iii3 ) ζq(Q) = 1 − ψ f (load applied at (L−1 GL−T )y = y, (40)
the bottom flange centroid) [17]. λ
(iv) Since the set {K , δ y , εq , ε Q } can be uniquely determined where y = LT x. The standard problem (40) can be solved by the
from {K , ψ f , ζq , ζ Q } by means of the expressions symmetric QR algorithm (e.g., [41]) or any other eigensolver
for real symmetric matrices.
q 1
K = K 1 − ψ 2f εq (Q) = K ζq (Q) (34) The discretisation scheme employed in this work is the
π Rayleigh–Ritz method, named after Strutt, 3rd Baron Rayleigh,

2 × 0.8 and Ritz [42,43]—for a comprehensive analysis of the method
−
 ψ f K if ψ f > 0
δy ≈ π (35) and a broad discussion of its application areas, see works by
2
− ψ f K
 if ψ f < 0, Mikhlin [32,44]. The functions v and φ were approximated by
π the sequences of linear combinations of coordinate functions
the latter can also be used to write the governing equations
n
(2k − 1)π
  
in non-dimensional form. X (n)
v n (ξ ) = ak 1 − cos ξ
k=1
2
3. Numerical implementation
n = 1, 2, . . . (41)
n
(2k − 1)π
  
The application of a structure-preserving discretisation X (n)
φ n (ξ ) = bk 1 − cos ξ
method (such as the methods of Rayleigh–Ritz and Bubnov– 2
k=1
Galerkin) to Problem D generates an algebraic quadratic
eigenvalue problem of the form n = 1, 2, . . . (NW) (42)
n
(n) (n)
X
(A + λB + λ2 C)x = 0, (36) φ n (ξ ) = b1 ξ + bk sin ((k − 1)π ξ )
k=2
where A, B and C are real symmetric matrices (of order n, n = 1, 2, . . . (FW), (43)
say)—A and C are positive and negative definite, respectively. (n) (n)
In the case of small-to-moderate n, the standard approach where the superscript in the coefficients ak , bk stresses their
to solve (36) numerically involves its linearisation, i.e. its dependence on n. Since the same number n of terms is used
conversion into an equivalent linear generalised eigenproblem to approximate v and φ, one is led to a discrete problem
by means of an augmentation procedure resembling the with 2n degrees of freedom—note also that (43) satisfies the
reduction of a second-order linear differential equation to a natural boundary condition (191 ). The numerical procedure was
system of two first-order ones [38,39]. In practice, the most continued until the relative differences between the buckling
commonly used linearisations involve one of the companion capacities γq λcr or γ Q λcr computed with three consecutive
forms values of n did not exceed 0.1%—it was found that this criterion
could always be satisfied by considering up to nine terms
(n ≤ 9) in (41)–(43).
     
0 N x N 0 x
=λ (37)
−A −B u 0 C u
4. Parametric study
     
−A 0 x B C x
=λ , (38) A parametric study was conducted to determine the non-
0 N u N 0 u dimensional elastic buckling capacities γq λcr or γ Q λcr of
where N is any non-singular n × n real matrix and u = λx. I-section cantilevers and the associated normalised buckling
This approach has two major drawbacks: (i) the dimension of mode shapes φ cr as a function of the beam parameter K (0.1 ≤
the linearised problem is twice that of the original quadratic one K ≤ 2.5), for (i) selected values of the flange asymmetry
and (ii) backward stability cannot always be guaranteed for the parameter ψ f (in the range −0.8 ≤ ψ f ≤ 0.8) and (ii) three
quadratic problem, even if a backward stable algorithm is used load positions (shear centre, top and bottom flange centroids).
to solve the linearised problem [40]. The cantilevers were acted on by a uniform load or a tip load
Therefore, we opted for the discretisation of Problem C, (i.e. Q 0 = 0 or q0 = 0)—the simultaneous application of the
leading directly to an algebraic linear generalised eigenproblem two loads was not considered in this work. As discussed earlier,
of the form warping at the fixed end section was either fully prevented (NW)
or completely free (FW).
1
Gx = Kx, (39) The numerical results were compared with values reported
λ by (i) Wang and Kitipornchai [17], for cantilevers with no
where G and K are real symmetric matrices and K is warping at the fixed end and both equal and unequal flanges,
positive definite. Then, it is possible to perform the Cholesky and by (ii) Trahair [12], for cantilevers free to warp at the fixed
900 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908

Fig. 2. Cantilevers acted on by a uniform load applied at the top flange centroid (ζq = −1 − Ψ f ): variation of γq λcr with K and Ψ f .

Fig. 3. Cantilevers acted on by a uniform load applied at the shear centre (ζq = 0): variation of γq λcr with K and Ψ f .

end and equal flanges—to the best of our knowledge, no Mcr extent, on the warping torsion generated during buckling—
values for cantilevers with unequal flanges and free to warp this effect is hardly mobilised in the absence of an external
at the fixed end are available in the literature. An excellent warping restraint. On the other hand, the warping torsion is
agreement was found in all cases. not relevant in cantilevers with very low K , which implies
Figs. 2–7 show plots of the elastic buckling capacities γq λcr that the warping restraint at the support has little impact on
and γ Q λcr against the beam parameter K , for selected values of γq λcr or γ Q λcr .
the flange asymmetry parameter ψ f and the three load positions (ii) Although the plots concerning cantilevers subjected to
considered. These plots show that: uniform and tip loads display similar qualitative features,
(i) As expected, the cantilevers restrained from warping at the γq λcr values are always higher: for equal end values (at
the fixed end (NW) exhibit higher buckling capacities than ξ = 0 and ξ = 1), the magnitude of the non-dimensional
f
the ones that are free to warp at that section (FW). While bending moment distribution |µ0 (ξ )λ| associated with the
the difference is only marginal for very low values of K , uniform load is smaller at any given interior point (0 <
it gradually increases with this parameter and becomes ξ < 1).
substantial for high K values. This indicates that the (iii) The so-called Wagner’s effect [15,45], which may be
buckling capacity of a cantilever restrained from warping viewed as an increase (reduction) in the actual torsional
at the support and having a high K depends, to a large rigidity of the cross-sections when the compression
 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908 901

Fig. 4. Cantilevers acted on by a uniform load applied at the bottom flange centroid (ζq = 1 − Ψ f ): variation of γq λcr with K and Ψ f .

Fig. 5. Cantilevers acted on by a tip load applied at the top flange centroid (ζ Q = −1 − Ψ f ): variation of γ Q λcr with K and Ψ f .

flange is larger (smaller) than the tension one, is clearly The critical buckling mode shapes φ cr of cantilevers with
noticeable for the cantilevers loaded at the shear centre: K = 1.0 and acted on by tip loads are shown in Figs. 8–10
both γq λcr and γ Q λcr increase monotonically with ψ f . (these shapes are normalised so that φ cr max = 1)—since the
(iv) In the cantilevers loaded at the top or bottom flange, the buckling mode shapes of the cantilevers subjected to uniform
variation of the buckling capacity with ψ f stems from loads are qualitatively similar, they are not presented here. The
the interplay between (iv1 ) Wagner’s effect and (iv2 ) most striking difference between the two graphs in each of
the location of the point of load application relative to these figures concerns the shape of the critical modes near the
the shear centre—recall that, for top or bottom flange supported end—while the cantilevers prevented from warping
loading, the value of the parameters ζq and ζ Q depends at the support (NW) display a considerable curvature, the ones
on ψ f . This explains why, in some cases, this variation that are free to warp (FW) exhibit a linear ascending branch,
is not monotonic (e.g., warping restrained cantilevers with which is an immediate consequence of the natural boundary
ψ f > 0 and bottom flange loading). It is also worth noting condition φ ,ξ ξ (0) = 0. It is also interesting to note that, in
that, for top flange loading and moderate-to-high K values, the cases of shear centre and top flange loading, φ cr increases
the maximum buckling capacity occurs for sections with monotonically with ξ , so that the maximum torsional rotation
equal (or nearly equal) flanges. occurs at the cantilever free end. Conversely, for bottom flange
902 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908

Fig. 6. Cantilevers acted on by a tip load applied at the shear centre (ζ Q = 0): variation of γ Q λcr with K and Ψ f .

Fig. 7. Cantilevers acted on by a tip load applied at the bottom flange centroid (ζ Q = 1 − Ψ f ): variation of γ Q λcr with K and Ψ f .

loading the critical mode shapes reach a maximum for ξ < 1 referred to [47,48]. The 3-factor formula, as it appeared in [5,
and then decrease towards the cantilever free end, a feature that 37], reads
becomes gradually more significant as the value of the flange π 2 E Iz
asymmetry parameter ψ f diminishes. Mcr = C1
(k z L)2
s 
 2
5. Approximate formulae to estimate elastic critical kz Iw (k z L)2 G It
2
+ C2 z g − C3 z j − C2 z g − C3 z j  .


×  +
moments kw Iz π 2 E Iz

5.1. The three-factor formula of Eurocode 3 (44)

Because of its long and widespread use (at least in Europe),
One of the most well-known and widely employed formulae
it was decided to adopt formula (44) and to extend its range of
to estimate elastic critical moments is the so-called 3-factor
application to cantilevers, by deriving appropriate Ci factors—
formula, which was adopted by the ENV version of Eurocode 3
bearing this in mind, it is convenient to make some preliminary
[5] and retained up until the “Stage 34” Draft of its EN
observations:
version [37]. It was subsequently set aside, without being
replaced (see [1,46])—for brief but illuminating accounts of the (i) In [5,37], L is defined as the beam length between
origin and eventful past of this formula, the interested reader is consecutive laterally restrained cross-sections. When
 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908 903

Fig. 8. Cantilevers (K = 1.0) acted on by a tip load applied at the top flange centroid (ζ Q = −1 − Ψ f ): normalised critical mode shapes φ cr .

Fig. 9. Cantilevers (K = 1.0) acted on by a tip load applied at the shear centre (ζ Q = 0): normalised critical mode shapes φ cr .

dealing with cantilevers, this definition has to be modified defined (there is no direct relation to the distance between
and, therefore, we take L as the cantilever length—recall inflection points of the critical buckling mode shape
that it was previously denoted in this paper by the lower {vcr , φcr }) and (iii2 ) do not fully account for the end
case letter l. support conditions—indeed, when k z changes so do the Ci
(ii) According to [5,37], the Ci factors depend only on factors. Thus, the values assigned to k z and kw should be
(ii1 ) the bending moment diagram and (ii2 ) the end regarded as merely conventional.
restraint conditions. However, it has been recognised by (iv) For gravity loads, we define z g = −(z q(Q) − z S ), which
several authors (e.g., [13,47,49]) that, except in a few means that one has z g > 0 for loads applied above the
particular cases, these factors also depend on (ii1 ) the shear centre and there is no need for any sign convention—
beam parameter K , (ii2 ) the location of the point of load incidentally, the convention adopted in [37] is incorrect.
application (C2 , C3 ) and (ii3 ) the degree of cross-section Moreover, it should be made clear that formula (44) is
asymmetry with respect to the y-axis (C3 ). not valid when the beam is subjected to transverse loads
(iii) The quantities k z and kw are effective length factors, applied at different levels (e.g., if one has z q 6= z Q ).
(v) The cross-sectional property z j is defined as
the former associated with the end rotations about the
βy
Z
z-axis and the latter with the end warping restraint. It 1
z j = zS − z(y 2 + z 2 )dA = − , (45)
should be noticed that these factors (iii1 ) are not properly 2I y A 2
904 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908

Fig. 10. Cantilevers (K = 1.0) acted on by a tip load applied at the bottom flange centroid (ζ Q = 1 − Ψ f ): normalised critical mode shapes φ cr .

Table 1
Cantilevers with warping prevented at the fixed end (NW)—C1 and C2 factors (L = l; k z = 2; kw = 1; 0.1 ≤ K ≤ 2.5)

C1 C2
Top flange loading Bottom flange loading
q q
2 2 2 2
Point load 2.462/ 1 + K + 2.383K / 1 + K 0.380 + 2.092K − 0.318K 0.512 + 0.370K − 0.033K
q q
2 2 2 2
Uniform load 3.962/ 1 + K + 5.531K / 1 + K 1.130 + 1.539K − 0.176K 1.049 + 0.234K − 0.020K

which means that, once again, no sign convention is γq(Q) λcr , this approach yields more accurate estimates
needed. than an alternative one that consists of fitting a polynomial
of the same degree to tabulated C1 vs. K data.
5.2. C1 , C2 and C3 factors for cantilevers (iii) Next, the cases corresponding to z j = 0 and z g 6= 0
(i.e. cantilevers with equal flanges and loaded at the top
The numerical results presented in Section 4 are now used to or bottom flange) are considered. The numerical results
develop approximate analytical expressions for C1 , C2 and C3 , given in Section 4 and the C1 approximations obtained in
by adopting the following step-by-step strategy: the preceding step are used to evaluate C2 by means of the
(i) Fixed values are assigned to the effective lengths factors: expression
k z = 2.0 (the most intuitive value, in view of the analogy 2
with column flexural buckling) and kw = 1.0.
2
4γq(Q) λ2cr − π 2 C12 (1 + K )
C2 = . (47)
(ii) Initially, only the cases corresponding to z g = z j = 0 (or, 2πC1 γq(Q) λcr K ζq(Q)
equivalently, ζq(Q) = ψ f = 0—i.e. cantilevers with equal
Polynomials in K are then fitted to the tabulated C2 vs. K
flanges and loaded at the shear centre) are dealt with. Then,
values—see Tables 1 and 3.
the non-dimensional version of formula (44) reads
(iv) Finally, the cases associated with z j 6= 0 (i.e. cantilevers
π
q
2 with unequal flanges) lead to the C3 values, which
γq(Q) λcr = C1 K + 1. (46) are obtained using the approximations for C1 and C2
2
For the NW case, Figs. 3 and 6 show that the variation of developed earlier and the expression
2
the buckling capacity with K is practically linear. Thus, 2
4γq(Q) λ2cr − π 2 C12 (1 + K (1 − ψ 2f )) − 2πC1 C2 γq(Q) λcr K ζq(Q)
straight lines are fitted to the γq(Q) λcr vs. K numerical C3 = ( ) .
0.8 if ψ f > 0
results—the approximate analytical expressions for C1 , 2π × × C1 γq(Q) λcr K ψ f
1.0 if ψ f < 0
shown in Table 1, stem from these fits.
(48)
For the FW case, on the other hand, a quadratic
polynomial is fitted to the γq(Q) λcr vs. K numerical Then, curve fitting techniques are again employed to
results, as shown in Table 3—it was found that, concerning construct polynomials in K and ψ f matching the tabulated
the statistical parameters qualifying the difference (or C3 , K and ψ f values as closely as possible—see Tables 2
distance) between the numerical and proposed values of and 4. It should be noted that the step function appearing
 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908 905

Table 2
Cantilevers with warping prevented at the fixed end (NW)—C3 factor (L = l; k z = 2; kw = 1; 0.1 ≤ K ≤ 2.5; −0.8 ≤ ψ f ≤ 0.8, ψ f 6= 0)

Top flange loading 1.520 − 1.342ψ f − 0.010ψ 2f − 0.424ψ 3f + (0.162 + 2.419ψ f + 0.875ψ 2f + 0.400ψ 3f )K
2
+ (0.148 − 0.623ψ f − 0.216ψ 2f + 0.141ψ 3f )K

Point load Shear centre loading 1.808 − 0.944ψ f + 0.299ψ 2f − 0.061ψ 3f + (0.060 + 1.235ψ f − 0.574ψ 2f − 0.337ψ 3f )K
2
+ (0.128 − 0.409ψ f + 0.047ψ 2f + 0.237ψ 3f )K

Bottom flange loading 1.966 − 0.792ψ f + 0.139ψ 2f − 0.341ψ 3f + (0.061 + 0.549ψ f + 0.077ψ 2f − 0.206ψ 3f )K
2
+ (0.064 − 0.135ψ f − 0.050ψ 2f + 0.058ψ 3f )K

Top flange loading 2.441 − 1.589ψ f + 0.176ψ 2f − 0.658ψ 3f + (−0.412 + 2.442ψ f + 0.635ψ 2f +
2
0.261ψ 3f )K + (0.273 − 0.601ψ f − 0.140ψ 2f + 0.205ψ 3f )K

Uniform load Shear centre loading 2.609 − 1.801ψ f + 0.522ψ 2f + 0.461ψ 3f + (−0.445 + 2.251ψ f − 0.620ψ 2f −
2
1.443ψ 3f )K + (0.244 − 0.710ψ f + 0.044ψ 2f + 0.611ψ 3f )K

Bottom flange loading 2.793 − 1.235ψ f + 0.428ψ 2f − 0.630ψ 3f + (−0.492 + 1.008ψ f − 0.134ψ 2f −
2
0.095ψ 3f )K + (0.194 − 0.263ψ f − 0.003ψ 2f + 0.060ψ 3f )K

Table 3
Cantilevers free to warp at the fixed end (FW)—C1 and C2 factors (L = l; k z = 2; kw = 1; 0.1 ≤ K ≤ 2.5)

C1 C2
Top flange loading Bottom flange loading
q q q
2 2 2 2 2 2
Point load 2.437/ 1 + K + 0.613K / 1 + K − 0.105K / 1 + K 0.409 + 1.444K + 0.070K 0.529 + 0.234K + 0.149K
q q q
2 2 2 2 2 2
Uniform load 3.840/ 1 + K + 1.496K / 1 + K − 0.247K / 1 + K 0.987 + 1.420K + 0.165K 1.028 + 0.388K + 0.150K

Table 4
Cantilevers free to warp at the fixed end (FW)—C3 factor (L = l; k z = 2; kw = 1; 0.1 ≤ K ≤ 2.5; −0.8 ≤ ψ f ≤ 0.8, ψ f 6= 0)

Top flange loading 1.732 − 0.648ψ f − 0.062ψ 2f + 0.059ψ 3f + (0.066 + 1.539ψ f + 0.520ψ 2f − 0.032ψ 3f )K
2
+ (0.446 + 0.221ψ f − 0.037ψ 2f + 0.066ψ 3f )K

Point load Shear centre loading 2.021 + 0.361ψ f + 0.176ψ 2f − 0.655ψ 3f + (0.242 + 0.120ψ f − 0.426ψ 2f + 0.891ψ 3f )K
2
+ (0.337 + 0.052ψ f − 0.198ψ 2f − 0.099ψ 3f )K

Bottom flange loading 2.156 − 0.055ψ f + 0.101ψ 2f − 0.079ψ 3f + (0.435 + 0.168ψ f − 0.083ψ 2f − 0.077ψ 3f )K
2
+ (0.238 − 0.022ψ f − 0.011ψ 2f − 0.030ψ 3f )K

Top flange loading 2.669 − 0.815ψ f + 0.071ψ 2f − 0.066ψ 3f + (0.113 + 1.812ψ f + 0.359ψ 2f + 0.007ψ 3f )K
2
+ (0.499 + 0.289ψ f + 0.043ψ 2f + 0.081ψ 3f )K

Uniform load Shear centre loading 3.036 + 0.310ψ f + 0.306ψ 2f − 0.888ψ 3f + (0.066 + 0.036ψ f − 0.585ψ 2f + 1.180ψ 3f )K
2
+ (0.462 + 0.098ψ f − 0.227ψ 2f − 0.123ψ 3f )K

Bottom flange loading 3.277 − 0.350ψ f + 0.348ψ 2f − 0.263ψ 3f + (0.190 + 0.348ψ f − 0.195ψ 2f − 0.137ψ 3f )K
2
+ (0.395 − 0.071ψ f − 0.009ψ 2f + 0.009ψ 3f )K

in the denominator of (48) is obviously incompatible with to-date treatment of this topic, see [50]. Table 5 provides
the smooth curves provided by continuous polynomials a synopsis of the statistical parameters measuring the
over the whole domain (i.e. for −0.8 ≤ ψ f ≤ 0.8). This quality (accuracy) of the estimates yielded by the proposed
problem is attenuated by the fact that one has C3 ≈ 0 in expressions. The degree of the polynomials adopted in the
the vicinity of the discontinuity (i.e. for ψ f ≈ 0) and its curve fitting process, which, for consistency, was kept the same
effect on the accuracy of the proposed expressions is only for similar cases, was selected in order to ensure a targeted
marginal.
accuracy. In view of the plethora of possible accuracy measures,
All the above curve fittings are performed using standard we opted to aim at guaranteeing that the maximum absolute
linear least squares procedures—for a comprehensive and up- value of the relative error never exceeds 5%, with the relative
906 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908

Table 5
Statistical parameters

Top flange loading Shear centre loading Bottom flange loading

σerr 0.01068 0.00897 0.00792
err (%) 0.082 0.105 0.111
r2 0.99664 0.99985 0.99987
Point load
|err| (%) 0.808 0.655 0.531
max|err| (%) 4.021 3.365 3.687
(|err| > 2.5%) (%) 3 3 4
NW
σerr 0.01144 0.01095 0.01018
err (%) 0.104 0.108 0.110
r2 0.99497 0.99984 0.99979
Uniform load
|err| (%) 0.845 0.802 0.720
max|err| (%) 5.226 3.809 3.901
(|err| > 2.5%) (%) 5 4 3

σerr 0.01050 0.00718 0.01176

err (%) 0.060 0.081 0.081
r2 0.99896 0.99984 0.99961
Point load
|err| (%) 0.735 0.530 0.951
max|err| (%) 4.225 2.640 3.499
(|err| > 2.5%) (%) 4 2 5
FW
σerr 0.01184 0.00905 0.01165
err (%) 0.084 0.098 0.108
r2 0.99832 0.99982 0.99975
Uniform load
|err| (%) 0.802 0.657 0.873
max|err| (%) 4.727 3.582 4.912
(|err| > 2.5%) (%) 7 4 5

error defined as careful examination of the influence of the warping restraint

condition on the elastic LTB of singly symmetric cantilevers.
γq(Q) λcr − γq(Q) λcr



err =
num app
(49) The estimates yielded by the proposed expressions are very
γq(Q) λcr num


accurate: the differences with respect to the numerical (“exact”)
results only occasionally exceed 5%.
(as shown in Table 5, this criterion is only violated once
In summary, our approach consists of the following steps:
and by a very minute margin). Table 5 includes (i) the
standard deviation σerr and mean value err of the relative error (i) write the governing equations in non-dimensional form
distribution, (ii) the square of the correlation coefficient r over a fixed reference domain and, in the process, identify
(also known as product-moment correlation coefficient or an adequate complete set of non-dimensional parameters;
Pearson’s correlation coefficient), (iii) the more significant, (ii) implement numerically (i.e. discretise) the continuum
from a practical viewpoint, mean of the absolute values of the model stemming from the preceding step;
relative error |err|, (iv) the maximum absolute value of the (iii) perform a parametric study, on the basis of the non-
relative error and (v) the number of cases, in percentage, with dimensional parameters defined in step (i);
absolute values of the relative error greater than 2.5%—this (iv) obtain approximate buckling formulae by fitting a
number is always rather small, with the majority of the cases mathematical model to the results of the parametric study
occurring for K ≈ 0.1). using least squares techniques and qualify quantitatively
the accuracy of the proposed expressions.
6. Conclusions The first step of this procedure is of fundamental importance,
since it enables a general and systematic representation of the
The so-called 3-factor formula (included, for instance, in cantilever LTB behaviour, a deeper insight into the nature of
the ENV version of Eurocode 3 [5]) is one of the most the phenomenon and a reduction in the number of parameters
commonly used general formulae to estimate elastic critical appearing in the governing equations—this last feature bears
moments in steel beams prone to LTB. In this paper, the important consequences, as far as the time and effort that must
domain of application of this formula is extended to cantilevers, be put into the parametric study are concerned.
by providing approximate analytical expressions to determine
the C1 , C2 and C3 factors. The paper deals specifically with Appendix
I-section cantilevers (i) having equal or unequal flanges, (ii)
fully built-in or free to warp at the support and (iii) acted on by The approximation of β y (in terms of ψ f and h S ) given
uniformly distributed or concentrated tip loads (applied either by Eq. (33) is inherent to the results of the parametric study
at the shear centre or at one of the flanges) and includes a concerning singly symmetric I-section cantilevers (Section 4)
 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908 907

and, thus, also to the analytical expressions proposed for the C3 [7] Galéa Y. Abaques de déversement pour profilés laminés. Construction
factor (Section 5.2). Métallique 1981;4:39–51.
To assess the accuracy of Eq. (33), a number of HEB and IPE [8] Baláz I, Koleková Y. Factors C1 , C2 , C3 for computing the elastic
critical moment Mcr . In: Zbornı́k VI. sympózia Drevo v stavebných
sections, having h S values varying between 166 mm and 484 konstrukciách so zahranicnou úcast’ou, Kocovce. 2004. p. 29–34.
mm, are made singly symmetric through the modification of [9] Timoshenko SP. Einige stabilitätsprobleme der elastizitätstheorie. Z Math
one of the flanges, either by reducing its width or by increasing Phys 1910;58:337–85. Reprinted in The collected papers of Stephen P.
its thickness. The fillets are neglected (i.e. the cross-sections Timoshenko. London: McGraw-Hill; 1953.
are idealised as a combination of three thin rectangles). The [10] Poley S. Lateral buckling of cantilevered I-beams under uniform load.
Trans ASCE 1956;121:786–90.
asymmetry parameter ψ f of the modified cross-sections falls [11] Nethercot DA. The effective lengths of cantilevers as governed by lateral
within the range 0.10 ≤ |ψ f | ≤ 0.95—small-to-moderate buckling. Struct Eng 1973;51(5):161–8.
|ψ f | values are associated with an increased flange thickness, [12] Trahair N. Lateral buckling of overhanging beams. In: Morris LJ, editor.
whereas moderate-to-large |ψ f | values correspond to a reduced Instability and plastic collapse of steel structures. London: Granada; 1983.
flange width. The approximate β y values yielded by Eq. (33) p. 503–18.
[13] European Commission for Steel and Coal (ECSC). Lateral-torsional
were then compared with the exact ones, as defined by Eq. (2)—
buckling in steel and composite beams. Research project 7210-PR-183
this comparison shows that: final technical report (Book 2–Design guide). 2003.
1. Eq. (33) always overestimates β y (if ψ f > 0, then β y,exact < [14] Doswell B. Lateral-torsional buckling of wide flange cantilever beams.
In: Proc. 2002 annual stability conference. Seattle: Structural Stability
β y,app < 0, while 0 < β y, exact < β y,app if ψ f < 0). This Research Council; 2002. p. 267–90.
means that an approximate β y value invariably leads to an [15] Anderson J, Trahair N. Stability of monosymmetric beams and
underestimation of Mcr . cantilevers. J Struct Div 1972;98(ST1):269–86.
2. The relative errors (β y,app − β y,exact )/|β y,exact | are generally [16] Roberts TM, Burt CA. Instability of monosymmetric I-beams and
below 15%, even if they occasionally reach almost 20%. cantilevers. Int J Mech Sci 1985;27(5):313–24.
[17] Wang CM, Kitipornchai S. On stability of monosymmetric cantilevers.
However, no obvious correlation appears to exist between
Eng. Struct. 1986;8(3):169–80.
the value of |ψ f | and the magnitude of these relative errors. [18] Wang CM, Kitipornchai S, Thevendran V. Buckling of braced
At this point, what remains to be assessed is how an error monosymmetric cantilevers. Int J Mech Sci 1987;29(5):321–37.
[19] Attard MM. General non-dimensional equation for lateral buckling. Thin-
in β y affects the evaluation of Mcr . In order to obtain such
Wall Struct 1990;9(1–4):417–35.
an assessment, we considered the worst-case scenario: a 20% [20] Camotim D, Andrade A. Lateral-torsional buckling of I-section
overestimation of β y , regardless of |ψ f |. It was found that: cantilevers: Evaluation of critical moments via the EC3 formula. In:
Report TC8-2002-023, European convention for constructional steelwork,
(i) As expected, the impact of the β y error on Mcr grows with
technical committee 8 (stability) meeting. 2002.
|ψ f | and is more pronounced in the FW case, due to an [21] Camotim D, Andrade A. Evaluating critical moments of I-section
increased importance of the Wagner effect in the overall cantilevers via the EC3 formula: Theoretical background and additional
buckling phenomenon. results. In: Report TC8-2003-007, European convention for constructional
(ii) Neither the loading nature (tip point load or uniformly steelwork, technical committee 8 (stability) meeting. 2003.
[22] Andrade A, Camotim D, Providência P. Lateral-torsional buckling of steel
distributed load) nor its location (shear centre, top or
cantilevers: Evaluation of critical moments using the EC3 Formula. In:
bottom flange) appear to influence the magnitude of the Proc. 20th czech and slovak national conference on steel structures and
Mcr errors significantly, at least in a consistent way. bridges. 2003. p. 635–40.
(iii) As the beam parameter K increases, so does the impact [23] Ciarlet PG. Mathematical elasticity, vol. 1: Three-dimensional elasticity.
of the β y error on Mcr . In FW cantilevers with highly Amsterdam: Elsevier; 1988.
asymmetric cross-sections – |ψ f | = 0.8 –, the β y error is [24] Vacharajittiphan P, Woolcock ST, Trahair NS. Effect of in-plane
deformation on lateral buckling. J Struct Mech 1974;3(1):29–60.
fully reflected on the evaluation of Mcr for K = 2.5. As K [25] Vlassov B. Thin-walled elastic bars. Jerusalem: Israel Program for
decreases, this impact is slowly (but steadily) attenuated— Scientific Translations; 1961.
it does not exceed 10% for K ≤ 0.5. [26] Trahair NS. Flexural-torsional buckling of structures. London: E&FN
Spon (Chapman & Hall); 1993.
References [27] Trefftz E. Über die ableitung der stabilitätskriterien des elastischen
gleichgewichts aus der elastizitätstheorie endlicher deformationen. In:
[1] Comité Européen de Normalisation (CEN). Eurocode 3: Design of steel Proc. third intern. congr. appl. mech. vol. 3. 1930. p. 44–50.
structures, Part 1-1: General rules and rules for buildings (EN 1993-1-1). [28] Trefftz E. Zur theorie der stabilität des elastischen gleichgewichts. Z
Brussels; 2005. Angew Math Mech 1933;13(2):160–5.
[2] British Standards Institution (BSI). BS 5950-1: Structural use of steelwork [29] Knops RJ, Wilkes EW. Theory of elastic stability. In: Truesdell C, editor.
in buildings. Code of practice for design. Rolled and welded sections. Mechanics of solids, vol. 3. Berlin: Springer-Verlag; 1984. p. 125–302.
London; 2000. Reissue of Encyclopaedia of physics, vol. VIa/3. Berlin: Springer-Verlag;
[3] American Institute of Steel Construction (AISC). Manual of steel 1973.
construction—Load and resistance factor design. 3rd ed. Chicago; 2001. [30] Mikhlin SG. Mathematical physics, an advanced course. Amsterdam:
[4] Standards Australia (SA). AS 4100: Steel structures (including North-Holland; 1970.
amendments 1, 2 and 3). Homebush; 1990. [31] Coddington EA, Levinson N. Theory of ordinary differential equations.
[5] Comité Européen de Normalisation (CEN). Eurocode 3: Design of steel New York: McGraw-Hill; 1955.
structures, Part 1-1: General rules and rules for buildings (ENV 1993-1-1). [32] Mikhlin SG. Variational methods in mathematical physics. Oxford:
Brussels; 1992. Pergamon Press; 1964.
[6] Clark JW, Hill HN. Lateral buckling of beams. J Struct Div 1960;86(7): [33] Lanczos C. The variational principles of mechanics. Toronto: University
175–96. of Toronto Press; 1970.
908 A. Andrade et al. / Journal of Constructional Steel Research 63 (2007) 894–908

[34] Buckingham E. On physically similar systems; illustrations of the use of Reprinted in Gesammelte werke—oeuvres. Paris: Gauthier-Villars,
dimensional equations. Phys Rev 1914;IV(4):345–76. Société Suisse de Physique; 1911.
[35] Langhaar HL. Dimensional analysis and theory of models. New York: [44] Mikhlin SG. The numerical performance of variational methods.
John Wiley & Sons; 1951. Groningen: Wolters-Noordhoff Publishing; 1971.
[36] Kitipornchai S, Trahair NS. Buckling properties of monosymmetric [45] Wagner H. Torsion and buckling of open sections. NACA technical
I-beams. J Struct Div 1980;106(ST5):941–57. memorandum 807. 1936 [Translation, by Reiss S, of Verdrehung und
[37] Comité Européen de Normalisation (CEN). Eurocode 3: Design of steel knickung von offenen profilen, from the 25th Anniversary Number of the
structures, Part 1-1: General rules and rules for buildings (prEN 1993-1- Technische Hochschule; Danzig, 1929. p. 329–43].
1). Stage 34 draft. Brussels; 2002. [46] Comité Européen de Normalisation (CEN). Eurocode 3: Design of steel
[38] Afolabi D. Linearization of the quadratic eigenvalue problem. Comput structures, Part 1-1: General rules and rules for buildings (prEN 1993-1-
Struct 1987;26(6):1039–40. 1). Stage 49 draft, Brussels. 2003.
[39] Tisseur F, Meerbergen K. The quadratic eigenvalue problem. SIAM Rev [47] Baláz I, Koleková Y. Critical moments of beams and girders—
2001;43(2):235–86. Clark–Mrázik formula. In: Proceedings of the 19th Czech and Slovak
[40] Tisseur F. Backward error and condition of polynomial eigenvalue national conference on steel structures and bridges. 2000. p. 87–94.
problems. Linear Algebra Appl 2000;309(1–3):339–61. [48] Braham M. Le déversement élastique des poutres en I à section
[41] Golub GH, Van Loan CF. Matrix computations. Baltimore: Johns Hopkins monosymétrique soumises à un gradient de moment de flexion. Constr
University Press; 1996. Métall 2001;(1):17–28.
[42] Strutt JW (3rd Baron Rayleigh). The theory of sound, vol. 2. New York: [49] Baláz I, Koleková Y. Buckling of monosymmetric beams—conjured
Dover; 1945. Unabridged republication of the second revised and enlarged problem (2 parts). In: Proceedings of the 2nd European conference on
edition of 1894. steel structures. 1999. p. 701–4. Full paper in CD - ROM.
[43] Ritz W. Über eine neue methode zur lösung gewisser variationsprobleme [50] Björck A. Numerical methods for least squares problems. Philadelphia:
der mathematischen physik. J Reine Angew Math 1908;135:1–61. SIAM; 1996.