This article was downloaded by: [York University Libraries]

On: 22 August 2013, At: 06:08

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

Mechanics Based Design of Structures and Machines: An

International Journal
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/lmbd20

Four New Methods for Finding Structural Critical

#
Points
a a a
M. Rezaiee-Pajand , H. R. Vejdani-Noghreiyan & A. R. Naghavi
a
Department of Civil Engineering , Ferdowsi University of Mashhad , Mashhad , Iran
Published online: 16 Jul 2013.

To cite this article: M. Rezaiee-Pajand , H. R. Vejdani-Noghreiyan & A. R. Naghavi (2013) Four New Methods for Finding
#
Structural Critical Points , Mechanics Based Design of Structures and Machines: An International Journal, 41:4, 399-420, DOI:
10.1080/15397734.2012.746917

To link to this article: http://dx.doi.org/10.1080/15397734.2012.746917

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
 Mechanics Based Design of Structures and Machines, 41: 399–420, 2013
Copyright © Taylor & Francis Group, LLC
ISSN: 1539-7734 print/1539-7742 online
DOI: 10.1080/15397734.2012.746917

FOUR NEW METHODS FOR FINDING STRUCTURAL

CRITICAL POINTS#

M. Rezaiee-Pajand, H. R. Vejdani-Noghreiyan, and A. R. Naghavi

Department of Civil Engineering, Ferdowsi University of Mashhad,
Mashhad, Iran
Downloaded by [York University Libraries] at 06:08 22 August 2013

Four systematic approaches are developed in the context of the indirect methods for
analyzing the nonlinear prebuckling behavior. In this study, an appropriate test function
to monitor bypassing of the critical points is utilized. The main idea of the article
is to find the critical loads by using the perturbation of the eigenvalue. Both linear
and quadratic variations of the eigenvalue and also parabolic changes of the structural
stiffness are considered. Moreover, other available techniques to find the critical points
are briefly reviewed. Finally, the accuracy and robustness of the presented procedures
are investigated by solving some well-known benchmark problems.

Keywords: Critical point; Eigenvalue; Equilibrium path; Perturbation; Structural stability; Tangent
stiffness matrix.

1. INTRODUCTION
Prior to the advent of the “finite element method,” only the analysis of simple
structures was possible. For the complex structures, approximate techniques were
mostly used. As a result, in order to obtain a rational reliability for structures, the high
safety factors were implemented for design provisions. In fact, developing the finite
element technique has made it possible for engineers to attain a more realistic behavior
of the structures. It is worthy to note that nowadays, structures are designed to sustain
complex behavior even under the service load. Therefore, any large deformation
analysis, and especially the stability of the structures with nonlinear prebuckling
behavior should be considered and examined despite their inherent complexities.
There are numerous algorithms for nonlinear analysis of structures
documented in literatures (Belytschko et al., 2000; Crisfield, 1991; Felippa, 1999).
Basically, these schemes are categorized in two groups: incremental procedures and
incremental-iterative methods. In incremental techniques, the linear assumption of
the structural behavior in each increment causes some error known as drifting error.
In incremental-iterative methods, this error is reduced by using the iterative (or
corrective) process. In other words, the incremental stage in incremental-iterative
algorithms is the same as the incremental strategies. Both techniques assume a linear

Received May 25, 2011; Accepted October 25, 2012

#
Communicated by I. Elishakoff.
Correspondence: M. Rezaiee-Pajand, Department of Civil Engineering, Ferdowsi University of
Mashhad, Mashhad, Iran; E-mail: [email protected]

399
 400 REZAIEE-PAJAND ET AL.

behavior for the structure in each step, but in an incremental-iterative algorithm, a

corrective phase also exists, which tries to eliminate the drifting error. Although the
incremental-iterative approaches could lead to the exact (or near exact) solution,
they may suffer from instability or time consumption.
Three common strategies used for the iterative phase are: the Full Newton–
Raphson procedure, the Modified Newton–Raphson method, and the Quasi-
Newton scheme. The Full Newton–Raphson technique uses an updated tangent
stiffness matrix in each iteration stage. Since updating of the tangent stiffness matrix
is time consuming, it is proposed to utilize the Modified Newton–Raphson scheme
in which the tangent stiffness matrix is updated after a number of iterations. In this
technique, despite decreasing the computational cost in each iteration process, the
numbers of the required iterations are increased, which is an indication of not being
a necessarily optimized algorithm.
Downloaded by [York University Libraries] at 06:08 22 August 2013

To solve both mentioned deficiencies, Quasi-Newton approaches were

proposed (Ohtmer, 1999). In these methods, the tangent stiffness matrix is estimated
based on the previous utilized tangent stiffness matrices. On the other hand, to
control the incremental stage of the analysis, various indicators may be used. Load
control is the first tactic that was utilized by analysts. By controlling the solution
process with load indicator, limit points may not be passed. Therefore, displacement
control techniques were proposed to pass the limit points. Despite passing limit
points, turning points on the equilibrium path may not be passed by this indicator.
To overcome these deficiencies, the arc-length method was proposed, which uses a
combination of load and displacement for tracing the equilibrium path (Riks, 1979;
Wempner, 1971). By using the arc-length technique, various kinds of equilibrium
paths with limit or turning points may be traced (Crisfield, 1991; Ritto-Correa
and Camotim, 2008). Since the arc-length technique is used in this article, a more
detailed investigation of this algorithm is presented in Section 2.
Structures have special performance at the critical points. These behaviors
are very attractive for researchers. As a result, various studies have so far been
concentrated on the stability analyses (Fabiena et al., 1991; Thacker et al., 1998).
In order to investigate the stability of structures, critical points on the equilibrium
path should be found. Generally, these points are determined by two main processes
called direct and indirect methods (Felippa, 1999).
In a direct procedure, a set of equations is solved to obtain the critical points
(Abbott, 1978; Moore and Spence, 1980; Planinc and Saje, 1999). In other words,
the critical points are the solutions of these set of equations. Since the set of the
equations is highly nonlinear, solving them requires an iterative procedure and
consequently has a high computational cost. One of the direct-approaches inherent
deficiencies is that only the critical points are calculated, and there would not be any
additional information available about the behavior of the structure. In addition,
when the system of equations becomes ill-conditioned, appropriate tactics should be
utilized to handle the problem (Shi, 1996; Wriggers and Simo, 1990).
In an indirect procedure, the behavior of the structure is monitored during
the analysis and bypassing the critical points is identified by a suitable test function
(Felippa, 1999). Using a powerful strategy to trace the highly nonlinear curves is
a crucial task in the success of the process. Among many techniques for tracing a
nonlinear equilibrium path, the arc-length method has outstanding characteristics and
also a good ability. This scheme was modified by Crisfield and called the cylindrical
 FINDING STRUCTURAL CRITICAL POINTS 401

arc-length method in the literature. In this study, the cylindrical arc-length procedure,
which is very appropriate for indirect techniques, is utilized to trace the nonlinear
equilibrium path. For finding the critical points by an indirect algorithm, researchers
introduced various tactics. Chan proposed an exact formulation for finding the limit
and bifurcation points by considering the nonlinear behavior of structures (Chan,
1993). On the other hand, Shi and Crisfield introduced a simple procedure to find
hidden instability paths and the critical points along the equilibrium path (Shi and
Crisfield, 1992). Their main idea was to find the concealed paths after imposing an
imperfection to the structure. Another simple scheme to find the buckling loads was
proposed by Fujikake, which considers the prebuckling nonlinear behavior of the
structures (Fujikake, 1985). He assumed a linear variation for the tangent stiffness
matrix between two increments. Using linear stability analysis is another technique to
overcome the nonlinear stability analysis complexities (Chang and Chen, 1986). To
Downloaded by [York University Libraries] at 06:08 22 August 2013

calculate the bifurcation points of large-scaled structures, other strategies have been
recently developed, which needs a large amount of computations (Fujii and Noguchi,
2002; Ikeda et al., 2007; Noguchi and Chen, 2003).
The problem of the stability analysis has been extensively examined so far. A
few of these studies were briefly discussed earlier. It is worth emphasizing that such
an analysis leads to the estimation of the load-carrying capacity of the structural
system. In this investigation, by using the perturbation theory of the eigenvalue, the
critical points will be identified (Vejdani-Noghreiyan, 2004). A monitoring function,
which has a better behavior than the other schemes, is utilized. The lowest entry
of the main diagonal of the matrix, in the parametric factorization of the tangent
stiffness matrix, is selected for the monitoring process. Four different formulations are
proposed to find the simple critical points in the presence of the nonlinear prebuckling
behavior of the structure. In the first procedure, the stiffness matrix is assumed to vary
linearly between the two available points on the equilibrium path, before and after
the critical point. Moreover, the lowest eigenvalue of the tangent stiffness matrix is
approximated by its first two terms in Taylor’s expansion. This strategy is simple and
leads to acceptable results in predicting the critical load of the structure. To improve
the accuracy of the results, more terms in Taylor’s expansion of the lowest eigenvalue
are utilized. As it can be seen in the numerical examples, the critical load predicted by
this technique does not necessarily approach to the exact value. The advantage of this
scheme is that the critical load is predicted lower than the actual value, and hence it is
suitable for design purposes. To reduce the error, an iterative counterpart of the first
procedure is also proposed. Although, the computational cost increases because of the
iteration processes, the error which comes from the linear assumption of the tangent
stiffness matrix diminishes. To check the influence of the variation of the tangent
stiffness matrix, in another presented procedure, a parabolic variation is proposed for
the tangent stiffness matrix between the available points on the equilibrium path. The
results show that this assumption improves the solutions significantly, but it needs
more memory storage.

2. NONLINEAR ANALYSIS OF STRUCTURES ALONG THE

EQUILIBRIUM PATH
Finding critical points by using the indirect method needs tracing the
structural equilibrium path. In order to obtain the bypassing of the singular points,
 402 REZAIEE-PAJAND ET AL.

the sign of the test function should be monitored after each increment. Therefore,
choosing a suitable and powerful strategy for tracing the equilibrium path plays a
crucial role in the success of the indirect procedure. The equilibrium condition must
be satisfied in any step of the applied force, and the unbalanced load has to be zero
for all states of the displacements. In other words, the states of
u p that satisfy
the following equation is the equilibrium state of the structure:

r = R
u − pe = 0 (1)

In this equation, r symbolizes the unbalanced load vector that should be

vanished and R
u represents the internal resistance of the structure due to the nodal
displacement. The reference load vector and the load parameter are shown by e
Downloaded by [York University Libraries] at 06:08 22 August 2013

and p, respectively. This system of equation is indeterminate, since it has
n + 1

unknowns and n equations. It is assumed that the structure has n nodal degrees of
freedom and one load parameters. The system can be determinate if an additional
equation is added to it. In the load control technique for example, it is the load
parameter that controls the progress of the analysis and the additional equation
is the increment of the load parameter that equals to an arbitrary value, which is
determined by the analyst. In the arc-length formulation, the augmented equation
is the increment of the arc, which is available by tracing the equilibrium path at
the end of the increment. The additional equation that should be added to the
equilibrium system is represented by the following arbitrary function in Eq. (3):

r = R
u − pe (2)
f
u p = 0 (3)

where, f is a function that relates the displacement parameter to the load parameter.
It is important to note that various selections of the arbitrary function constitute
different processes of the nonlinear analysis (Crisfield, 1991; Felippa, 1999). The
main goal of the structural nonlinear analysis is to solve Eq. (1). To achieve this
goal, two main techniques, called incremental schemes and incremental-iterative
procedures, may be used. In the incremental methods, the drift error is inevitable
even with a very small step length. On the other hand, in incremental-iterative
procedures, in addition to incremental steps, some iterative steps should be carried
out to vanish the drift from the equilibrium path caused by the linear assumption
of the stiffness matrix. These steps are for reducing the unbalanced load vector
and pushing back the structural behavior to the equilibrium path. It should be
mentioned that the probability of not converging increases in these tactics. Among
the available numerical process in the nonlinear analysis, the arc-length method is
one of the most powerful algorithms in iterative-incremental techniques (Crisfield,
1991). In the predictor part, the increment of the load and the displacements are
obtained by solving the following system of equations:

Ku − pe = 0 (4)

uT u + p2 2 eT e = l2 (5)
 FINDING STRUCTURAL CRITICAL POINTS 403

In these equations, K is the tangent stiffness matrix, which is the Hessian of

the total potential energy of the system, l is the length of the curve to be traced
in the increment, and  is a scalar parameter to show the role of the load in
this technique. By changing this parameter
, one can obtain various arc-length
strategies proposed by researchers. For example, setting  = 1 leads to the constant
arc-length suggested by Riks (1979) and Wempner (1971) or setting  = 0 leads to
the well-known cylindrical arc-length presented by Crisfield (1991).
It should be reminded that Eq. (5) is a constraint condition. The analysts have
formulated a variety of constraint conditions for nonlinear solver so far. Most of the
constraint equations have either linear or quadratic forms. Risks suggested a simple
linear type, whereas Crisfield proposed a quadratic one. It is worth emphasizing that
the quadratic option leads to a second-order control strategy and has the advantage
of shaping a closed constraint surface, which always intersects the equilibrium path.
Downloaded by [York University Libraries] at 06:08 22 August 2013

As a result, this kind of restriction leads to the nonlinear solution even in the
complex portion of the analysis domain.
In general, since the solution of the last system of equations is not on the
equilibrium path, the corrector phase should be started to vanish the created error.
The number of the iterations needed, for pushing back the state to the equilibrium
path, depends on the nonlinear behavior of the system. The load corrector and
displacement corrector can be calculated by the following equations:

Ku − pe = −r (6)

uT u + pp2 eT e = 0 (7)

In these equations, p and u are the load corrector and the displacement
corrector, respectively. At this stage, one can simply obtain the cylindrical
formulation by setting  = 0 in Eqs. (5) and (7). It should be noted that the
cylindrical arc-length method is a very robust technique for tracing the equilibrium
path and also has an acceptable convergence rate. Using this process leads to a
quadratic equation for the load corrector. To solve the above equations, an implicit
procedure is used. The corrector load parameter can be obtained by solving the
following equation:

a1
p2 + a2
p + a3 = 0 (8)

where, all the coefficients are given below:

a1 = uTe ue + 2 eT e
a2 = 2uTe
u0 + ur  + 2
p0  2 eT e
a3 =
u0 + ur T
u0 + ur  +
p0 2 2 eT e − l2 (9)
−1
ur = −K r
ue = K −1 e

In these equations, u0 and p0 are the difference between the current and
last converged state of displacements and the load parameter. For instance, based
 404 REZAIEE-PAJAND ET AL.
Downloaded by [York University Libraries] at 06:08 22 August 2013

Figure 1 The process of converging to the equilibrium path by using the arc-length method.

on Fig. 1, at the first iteration, point A is reached and a better solution point,
like B, is sought. Therefore, one can assume u0 = u1 and p0 = p1 , for the
second iteration (looking forward to C) u0 = u1 + u2 and p0 = p1 + p2 . This
algorithm is continued until r = 0, or it is equal to an acceptable tolerance.
The appropriate solution of the quadratic equation is given by Crisfield (1991).
It should be noted, if the other root of the quadratic equation is selected, the
solution will turn back to the traced branch. On the other hand, the sign of the
load predictor plays a crucial role in tracing the equilibrium path of the structure.
Inappropriate choosing of the sign might yield the process to trace the stable
branch when encountering bifurcation points. Consequently, the test function will
not identify bypassing the critical point, and as a result, this point will be useless.
In addition, encountering turning points might cause divergence of the algorithm by
inappropriate sign selection of the load predictor (Crisfield, 1991; De Souza Neto
and Feng, 1999; Feng et al., 1996).

3. MONITORING THE CRITICAL POINTS

Two main approaches, called direct and indirect tactics, may be used to find
the critical points. The direct scheme is used to find the critical points, by solving
an algebraic equation directly. On the other hand, an indirect strategy is used
to monitor the critical points as marching along the equilibrium path. In fact,
by utilizing this technique, the behavior of the structure is obtained before the
critical points. One of the test functions used for achieving the critical points is
the determinant of the tangent stiffness matrix. It is worth emphasizing that the
determinant of the tangent stiffness matrix is not an appropriate test function for
finding the critical points, especially for the large systems. Another way is using the
 FINDING STRUCTURAL CRITICAL POINTS 405

below determinant of the augmented system of the equation, which was suggested
by Abbott (1978):

r=0
(10)
Det
K = 0

On the other hand, Moore and Spence (1980) proposed a method using the
eigenvectors of the tangent stiffness matrix for the stability analysis of the structures.
According to their results, the solutions of the augmented system of equations
represent the instability of the structure. One of the most powerful techniques for
finding the critical points by a direct method was presented by Wriggers and Simo
(1990). In the modified version of this algorithm, the following system of equations
Downloaded by [York University Libraries] at 06:08 22 August 2013

has to be solved:

r=0
K = 0 (11)


= 1

The last term in this system of equations is used to rule out the trivial solution
of the equations
 = 0. In the indirect process, the existence of the critical points
is warned as the analysis traces the equilibrium path. Meanwhile, the sign of an
appropriate test function is monitored as the analysis progressed. Changing the sign
of the test function during tracing of the equilibrium path implies that a critical
point has just passed. It should be noted that the test function used in this method
is very important. The common test functions are as follows:
(1) Determinant of the tangent stiffness matrix
(2) The lowest eigenvalue of the tangent stiffness matrix
(3) The lowest entry on the main diagonal of the matrixD, in the parametric
factorization of the tangent stiffness matrix
K = LDLT 
Each of the mentioned functions has its inherent advantages and
disadvantages (Crisfield, 2000). Among the test functions used in indirect tactics
for finding the critical points, the lowest entry of the main diagonal of the matrix
D in the parametric factorization of the tangent stiffness matrix, may be superior
to the others. In fact, it needs less computational cost for solving the problem.
It is important to note that the number of negative, zero, and positive entries in
the main diagonal of D and the eigenvalue of K are the same, and because of the
symmetry of the tangent stiffness matrix, this factorization is always applicable.
Therefore, by monitoring the sign of the lowest entry of the main diagonal of D,
one may also monitor the sign of the lowest eigenvalue of the tangent stiffness
matrix, which needs much more computational effort to be calculated.
Selecting an appropriate analysis process is another important aspect in
stability analysis of structures. For instance, tracing the equilibrium path should be
such that, after reaching the critical points, the analysis is continued on the initial
path. If the analysis procedure continues on the stable branch derived from the
bifurcation point automatically, bypassing the critical points may not be alerted by
 406 REZAIEE-PAJAND ET AL.

the test function. One of the common and simple methods to find the critical load
is to interpolate between the stable and unstable points by a bisection technique or
fitting a polygonal between the available points. In the following sections, four new
procedures for finding the structural critical loads will be proposed.

4. FIRST-ORDER PERTURBATION OF THE LOWEST EIGENVALUE

As discussed earlier, bypassing the critical points can be monitored by using
an appropriate analysis procedure (like the arc-length algorithm) and a proper
test function to seek the existence of the critical point. In this study, it is assumed
that the tangent stiffness matrix varies linearly between the previous and the current
positions on the equilibrium path, hence it is needed to store the previous tangent
stiffness matrix. If the tangent stiffness matrix in the previous and current steps
Downloaded by [York University Libraries] at 06:08 22 August 2013

are called K0 and K1 , respectively, then K0 is a positive definite matrix, and K1

is indeterminate (for instance, it has some positive and some negative eigenvalue).
Based on the suggested scheme, the tangent stiffness matrix can be represented as
below:

K =
1 − K0 + K1 (12)

In this equation, K represents the tangent stiffness matrix between two

balanced points on the equilibrium path, and  is a positive number less than one.
From this point on, the goal is to find  such that the assumed tangent stiffness
matrix becomes singular. Another suitable form of K can be written as follows:

K = K0 + 
K1 − K0  (13)

It is very clear that the assumed tangent stiffness matrix is K0 at  = 0 and

when  approaches to 1, K approaches to K1 . According to Eq. (13), K consists of
two parts, the constant part
K0  and the variable part

K1 − K0 . The eigenvalue
of matrix K is a function of  and can be calculated from the following equation
(Stewart and Sun, 1990; Wilkinson, 1965):

 − xT
K1 − K0  x
lim = (14)
x→0  xT x
In this equation, and x represent the eigenvalue of the matrix K0 and its
corresponding eigenvector, respectively. For normalized eigenvectors, Eq. (14) can
be written as follows:


 = T

K1 − K0  (15)

where, stands for the Euclidian normalized vector of x. To determinate the

stability of the structure, the lowest eigenvalue of the matrix K should approach
zero to represent instability of the system. Therefore, Eq. (15) can be written in the
following form:


 = T

K1  − (16)
 FINDING STRUCTURAL CRITICAL POINTS 407

For convenience, an approximate value of the lowest eigenvalue of K is

calculated. By using the first two terms of the Taylor’s expansion of the lowest
eigenvalue

, the following simple equation can be obtained:



 = + (17)

To find the critical point,
 should be set to zero, as follows:



 = + =0⇒=− 
(18)

By substituting Eqs. (16) into (18), the following iterative equation can be
obtained:
Downloaded by [York University Libraries] at 06:08 22 August 2013

= (19)
− T
K1 

According to the presented formulations, the critical points of the structure

are calculated by assuming a linear variation of the tangent stiffness matrix between
the two adjacent points on the structural equilibrium path. Based on the earlier
discussion, the following algorithm is suggested for the proposed method:

Step 1 After tracing the equilibrium path and bypassing the critical point, matrices
K0 and K1 are calculated.
Step 2 Set p ← 0
Step 3 The lowest eigenvalue and corresponding eigenvector of K0 are found.
Step 4 Set  ← − T
K 
1
Step 5 p ← p + 
Step 6 If  ≤ Tol, then, p is printed as the critical load and end.
Step 7 The matrix K = K0 + 
K0 − K1  is calculated.
Step 8 If K is positive definite. Therefore, K0 ← K and go to step 3, Otherwise,
K1 ← K and go to step 4.

5. SECOND-ORDER PERTURBATION OF THE LOWEST EIGENVALUE

In the previous algorithm, the first two terms of the Taylor’s expansion of the
lowest eigenvalue were used to calculate the critical point. In this section, the third
term is also used to improve the accuracy of the results. Based on matrix algebra,
the eigenvalue of the matrix A can be calculated by the following equation:

Det
I − A = n
+ cn−1 n−1
+ cn−2 n−2
+ · · · + c0 = 0

On the other hand, if the A is perturbed by B, then, its eigenvalue can be
obtained from the following equation (Wilkinson, 1965):

Det
I − A − B = n
+ cn−1
 n−1
+ cn−2
 n−2
+ · · · + c0
 = 0 (20)
 408 REZAIEE-PAJAND ET AL.

In this equation, all the coefficients are functions of . For simple critical
points (bifurcation or limit point), the lowest eigenvalue
1  is a simple root of
Eq. (20). This root can be represented as follows (Wilkinson, 1965):

1
 = 1 + k 1  + k 2 2 + · · · (21)

In the previous section, the first two terms were only used to represent
1
 1
. Now, the third term will also be utilized. As a result, the following
relation is obtained. In this equation, x is the eigenvector that corresponds to 1 . It
is important to note that the new coefficient
k2  is very complicated. To calculate
this coefficient, all the eigenvalues and eigenvectors of the matrix A are needed. In
other words, the following equations should be utilized:
Downloaded by [York University Libraries] at 06:08 22 August 2013

ij = xiT Bxj
(22)
si = xiT xi
1
N
k2 = i1 1i
(23)
s1 i=2 si
1 − i

In these equations, xi and xj are the eigenvectors of i and j , respectively. It is

worth emphasizing that these equations are only valid for a simple critical point, in
which eigenvalues are distinct and hence prevent the vanishing of the denominator.
For multiple bifurcation points, other strategies should be used (Fujii et al., 2001;
Rezaiee-Pajand and Vejdani-Noghreiyan, 2006). For convenience, the presented
formulation will be more simplified, by normalizing the eigenvectors
si = 1. To
find the critical points, the lowest eigenvalue should approach zero, therefore, the
following equation can be obtained:

1
 = 1 + k 1  + k 2 2 = 0 (24)

By solving this quadratic equation,  can be calculated as below:


−k1 ± k12 − 4k2
= 1
(25)
2k2

While tracing the equilibrium path, the following substitutions are required to be
done:

A = K0
B = K1 − K 0

After Euclidian normalization of the eigenvectors
i , the following equations

will be obtained:

ij = T
i
K1 − K0  j
(26)
si = T
i i =1
 FINDING STRUCTURAL CRITICAL POINTS 409

If the difference between the lowest eigenvalue and others is large, which is the
case of the distinct separation of the eigenvalue, the change of the lowest eigenvalue
will be near linear. As it will be shown later, the first suggested procedure, described
in the previous section, would be satisfactory for some structures.

6. PARABOLIC STIFFNESS VARIATION

In order to develop the third formulation, it is assumed that the variation of
the tangent stiffness matrix is a parabolic function instead of a linear one. In this
case, the tangent stiffness matrix should be known in three states of the equilibrium
path. Based on this assumption, the tangent stiffness matrix, which is shown by K−1 ,
should be in hand before the state of K0 . Having three matrices, (K−1  K0 , and K1 ),
Downloaded by [York University Libraries] at 06:08 22 August 2013

which correspond to three successive points on the equilibrium path, the tangent
stiffness matrix
K will obey the following formula between points 0 and 1:


 − 1 + a
 + a
 − 1 + a 
 + a
K= K−1 + K0 + K1 (27)
a a −a
2 a−1

In this equation, a is an arbitrary number, and  varies between −a and

1 − a. To singularize K, its lowest eigenvalue should approach zero. Therefore, the
first and the second derivatives of the lowest eigenvalue can be calculated as follows:
 a−1 

 − xT K−1 + 2a−1
K
a2 −a 0
+ a
K
a−1 1
x

0 = lim = a
(28)
x→0  xT x
 



 −  x T 2
K−1 + 2
K
a2 −a 0
+ 2
K
a−1 1
x

0 = lim = a
T
(29)
x→0  x x

After normalizing the eigenvectors, the following equations will be available:

 
 a−1 2a − 1 a

0 = T
K−1 + 2 K0 + K (30)
a a −a a−1 1
 
 2 2 2

0 = T
K + K + K (31)
a −1 a2 − a 0 a − 1 1

Using the Taylor’s expansion of the lowest eigenvalue around zero, and then,
approaching it to zero will yield the singularized stiffness matrix. Consequently, the
following equation is held:
√
2   − ± 2 −2 

 = + + =0⇒= 
(32)
2

At this stage, the values of  between 0 and
1 − a will be acceptable. In the

following equation, a suitable choice for a is presented:

F0 − F−1
a= (33)
F1 − F−1
 410 REZAIEE-PAJAND ET AL.

Based on this equation, the variation of the load parameter is linear. It should
be noted that F−1  F0 , and F1 are the load parameters corresponding to K−1  K0 , and
K1 , respectively.

7. ACCURATE CRITICAL POINT

The arc-length method can be used instead of the load parameter technique
to calculate the critical load. Consequently, the inherent deficiency of utilizing the
load parameter to find the limit points will be obviated. In other words, as shown
in Fig. 2, although finding the bifurcation points are practical by using the load
parameter, but for evaluating the limit points, this parameter is unsuitable. In fact,
Downloaded by [York University Libraries] at 06:08 22 August 2013

the limit point can be passed without notice.

It is worth emphasizing that by using the arc-length, both the bifurcation and
limit points could be calculated efficiently. In this formulation, a new arc-length can
be obtained by the following relationship:

lN =  · l (34)

The arc-length between the two points on the equilibrium path is l, and
lN represents the length of the arc to reach the critical point or a new point on
the equilibrium path closer to the critical point. After reaching a new position on
the equilibrium path, the sign of the test function should be checked to determine
the state of the tangent stiffness matrix. If the tangent stiffness matrix in the new
position is positive definite, it can be substituted for K0 , and if it is an indeterminate
matrix, it can be substituted for K1 . Since after some steps, the speed of converging
to the critical point may decrease; authors’ numerical experiences show that using a
multiple of 0.8 of the required arc-length (0.8lN instead of lN ) after five iterations
will increase the ability of the suggested procedure and leads to more efficient
results.

Figure 2 Deficiency of load parameter in finding limit points.

 FINDING STRUCTURAL CRITICAL POINTS 411

Table 1 Name of the suggested methods

First order perturbation of eigenvalue FOP

Second order perturbation of eigenvalue SOP
Parabolic stiffness variation PSV
Arc-length updating ALU

8. NUMERICAL EXAMPLES
Some numerical examples are solved to show the accuracy and robustness
of the proposed procedures. To solve the problems a program was developed by
the authors in FORTRAN. Green–Lagrange strain and its compatible stress (PK2
Downloaded by [York University Libraries] at 06:08 22 August 2013

stress) were used for all the problems. The comparison of the numerical results is
made by utilizing the abbreviation of the suggested tactics, shown in Table 1.
For the sake of comparison, the exact value of the buckling load was obtained
by using a large number of iterations in each problem. Other strategies of finding
exact value of structural critical load are existed. One of them can obtain closed-
form solution for the equilibrium path of structure (Rezaiee-Pajand and Naghavi,
2011; Wang, 2000). For some structures, the buckling load without considering
prebuckling deformation was also presented.

Figure 3 Von–Mises truss.

 412 REZAIEE-PAJAND ET AL.

Table 2 Buckling load of Von–Mises Truss

Pcr = 105 × No. of iterations Analysis method

5 74978 1 FOP
5 74978 1 SOP
5.75092 1 PSV
5 75163 4 ALU
5 75163 – Exact

8.1. Von–Mises Truss

The first example is a 2D truss shown in Fig. 3, which is known as a Von–
Mises truss (Crisfield, 2000). For > 35 264 this plane truss possesses a simple
Downloaded by [York University Libraries] at 06:08 22 August 2013

bifurcation point. In this study, = 80 is assumed. The Von–Mises truss is a well-

known benchmark problem for stability analysis, because of its simplicity and also
having an available analytical solution (Crisfield, 2000). In this example, EA = 107
and L = 1 are utilized. All the members’ properties are dimensionless. Bypassing of
the critical load is sought using the parametric factorization of the tangent stiffness
matrix. Table 2 shows the results of the stability analysis.

8.2. Pyramidal Trusses

The solutions of the pyramidal trusses, shown in Fig. 4, were investigated
analytically by Ligaro and Valvo, in 2006. These are suitable structures to be
solved as benchmark problems. In these trusses, H = 80 B = 200, and EA = 100 are
assumed. By using these dimensions, the trusses have a simple limit point. It should
be added that all joints, except the top one, are fully restrained. Two cases of loading
are considered in this study.

Figure 4 Pyramid truss: (a) with two loads at the top and (b) with one vertical load at top.

Table 3 Buckling loads of pyramidal truss with two

loads at the top

pcr No. of iterations Analysis method

15.8117 1 FOP
15.7925 1 SOP
15.8117 1 PSV
15.8117 2 ALU
15.8117 – Exact
 FINDING STRUCTURAL CRITICAL POINTS 413

Table 4 Buckling loads of pyramidal truss with

one vertical load at the top

pcr No. of iterations Analysis method

18.7659 1 FOP
18.7501 1 SOP
18.7659 1 PSV
18.7663 3 ALU
18.7665 – Exact
Downloaded by [York University Libraries] at 06:08 22 August 2013

Figure 5 Plane arc-truss.

Tables 3 and 4 show the buckling load of the pyramidal trusses by utilizing
several suggested methods in this article.

8.3. Arc-Truss with 35 Members

The third example is a 35-member arc-truss shown in Fig. 5. This plane
structure has 35 members and 34 degrees of freedom (Onate and Matias, 1996). The
area of each member’s section is shown in Table 5, and the modulus of elasticity is
assumed to be E = 7 03 × 105 kgf/cm2 .
The results of the stability analysis are presented in Table 6.

Table 5 Section area of

truss members

Area
cm2  Member

51.61 1–10 and 35

64.52 11, 12
83.87 13–16
96.77 17, 18
102.23 19–22
161.29 23, 24
193.55 25, 26
256.06 27, 28
290.32 29–32
309.68 33, 34
 414 REZAIEE-PAJAND ET AL.

Table 6 Buckling loads of arc-truss

pcr No. of iterations Analysis method

2247.3771 1 FOP
2247.3451 1 SOP
2247.3500 1 PSV
2247.3604 6 ALU
2247.3602 – Exact
2415.2372 – Without considering
prebuckling
deformation
Downloaded by [York University Libraries] at 06:08 22 August 2013

Figure 6 Space truss.

8.4. Space Truss

Another example, which is shown in Fig. 6, is the well-known space truss
in stability analysis. This structure has 24 members and 21 degrees of freedom.
It should be reminded that because of its nonlinear and complex behavior, this
3D truss is widely used in the literature for investigating the robustness of the
algorithms (Crisfield, 2000; Onate and Matias, 1996; Shi, 1996).
As it is shown in Fig. 6, there are seven vertical loads exerted on the free nodes
of the structure. The structural properties are dimensionless, and are given below:

EA = 10000 W = 2 H = 6 22 L = 50 S = 25

In Table 7, the results of the stability analysis are presented.

Table 7 Critical loads of space truss

Pcr No. of iterations Analysis method

8.62344 1 FOP
8.6325 1 SOP
8.64367 1 PSV
8.64367 7 ALU
8.64360 – Exact
10.21 – Without considering
prebuckling
deformation
 FINDING STRUCTURAL CRITICAL POINTS 415

Figure 7 Shell structure.

Table 8 Critical loads of shell structure

pcr No. of iterations Analysis method

Downloaded by [York University Libraries] at 06:08 22 August 2013

11,235 1 FOP
11,225 1 SOP
11,250 1 PSV
11,252 11 ALU
11,252 – Exact

8.5. Shell Structure

Nonlinear and stability analysis of shell and plate structures are very attractive
for researchers. For example, Kere and Lyly investigate the buckling of cylindrical
shells (Kere and Lyly, 2011). Abed-Meraim et al. formulate a new six-node element
for nonlinear analysis of shells (Abed-Meraim et al., 2012). This example is a
single curved shell under a concentrated load as shown in Fig. 7 (Crisfield, 1991).
Mechanical properties of the material and also the geometry of the structure are
dimensionless, and are equal to: E = 31027 5 = 0 3. The thickness of the whole
meshes is t = 12 7. As it is shown in Fig. 7, 24 Lagrangian nine-node shell elements
are used for the analysis of this structure. Total number of degrees of freedom in
the curved shell is equal to 594. Table 8 shows the results of the stability analysis.

8.6. Dome Truss

To display the robustness of the new formulation in solving large structure,
a 3D truss will be analyzed. This structure has 216 nodes, 600 members, and 573
degrees of freedom. All the structural members are the same, and it is assumed that
A = 450, E = 2100, with the consistence units. The truss is shown in Fig. 8. Two
vertical loads act to the highest level of structure. Table 9 demonstrates the related
stability responses. To evaluate the efficiency of the new formulation, computing
time was measured. It was revealed that the authors’ technique was about 50 times
faster than the method of minimum eigenvalue interpolation (Crisfield, 2000), in the
stability part of the analysis.

8.7. Large Space Truss

Another 3D truss will be analyzed in this section. This structure has 391 nodes,
1470 members, and 1083 degrees of freedom. It is assumed that in all members
A = 450, E = 2100, with the consistence units. The space truss is shown in Fig. 9. One
 416 REZAIEE-PAJAND ET AL.
Downloaded by [York University Libraries] at 06:08 22 August 2013

Figure 8 Dome truss.

Table 9 Critical load of dome truss

Pcr No. of iterations Analysis method

94.708 1 FOP
94.601 1 SOP
94.715 1 PSV
94.716 8 ALU

vertical load is applied to the highest node of this structure. Table 10 demonstrates
the related stability responses. Based on the measured computing time, it was revealed
that the new formulation was about 250 times faster than the method of minimum
eigenvalue interpolation (Crisfield, 2000), in the stability part of the analysis.

Figure 9 Large space truss.

 FINDING STRUCTURAL CRITICAL POINTS 417

Table 10 Critical load of large space truss

Pcr No. of iterations Analysis method

17.224 1 FOP
17.352 1 SOP
17.015 1 PSV
17.017 8 ALU

9. CONCLUSIONS
The methodology presented in this article provides a comprehensive treatment
of structural critical points. All the four methods proposed here are categorized as
indirect procedures. The main idea is to utilize the perturbation of the eigenvalue
Downloaded by [York University Libraries] at 06:08 22 August 2013

when the stiffness matrix is changed. In the initial suggested scheme, a first-order
perturbation of the eigenvalue is used to obtain the critical point. The numerical
examples show that this tactic yields reasonable results. The critical points predicted
by this algorithm are near the exact values, sometimes lesser, and occasionally bigger
than the exact one.
In the second proposed algorithm, a more precise formulation is used for
calculating the eigenvalue. In other words, the first three terms of the Taylor’s
expansion were used. Once again, the tangent stiffness matrix is assumed to vary
linearly between the points. Since the actual variation of the tangent stiffness matrix
is not linear in this interval, increasing the accuracy of the eigenvalue calculation
does not necessarily increase the accuracy of the results. Based on the numerical
experiences, the results did not improve by this modification, and all the critical
points predicted by the presented algorithm are less than the exact values. Therefore,
the results are appropriate for design purposes.
To examine the effect of the pattern of the tangent stiffness matrix in the third
method, its variation is assumed to be quadratic between three available points on
the equilibrium path. Some benchmark problems are solved by this scheme. The
results obtained by this technique are better than before, but it needs more memory
storage. Considering the power of the computers nowadays, this problem is not a
serious drawback for the third suggested algorithm, as it was before.
In another approach, an iterative procedure was proposed to diminish the drift
error caused by the linear assumption of the tangent stiffness matrix. In this tactic,
the arc-length process is used as a perturbation parameter. This scheme has the
capability of reaching the exact values, but it needs more computational time.

REFERENCES
Abbott, J. P. (1978). An efficient algorithm for the determination of certain
bifurcation points. Journal of Computational and Applied Mathematics 4(1):19–27.
Abed-Meraim, F., Trinh, V. D., Combescure, A. (2012). Assumed-strain solid–
shell formulation for the six-node finite element SHB6: evaluation on non-linear
benchmark problems. European Journal of Computational Mechanics 21(2):52–71.
Belytschko, T., Liu, W. K., Moran, B. (2000). Nonlinear Finite Elements for Continua
and Structures. New York: John Wiley and Sons.
 418 REZAIEE-PAJAND ET AL.

Chan, S. L. (1993). Non-linear numerical method for accurate determination of limit

and bifurcation points. International Journal for Numerical Methods in Engineering
36(16):2779–2790.
Chang, S. C., Chen, J. J. (1986). Effectiveness of linear bifurication analysis for
predicting the nonlinear stability limits of structures. International Journal for
Numerical Methods in Engineering 23(5):831–846.
Crisfield, M. A. (1991). Non-Linear Finite Element Analysis of Solids and Structures.
Vol. 1: Basic Topics. New York: John Wiley & Sons.
Crisfield, M. A. (2000). Non-Linear Finite Element Analysis of Solid and Structures.
Vol. 2: Advanced Topics. New York: John Wiley and Sons.
De Souza Neto, E. A., Feng, Y. T. (1999). On the determination of the path
direction for arc-length methods in the presence of bifurcations and ‘snap-backs’.
Computer Methods in Applied Mechanics and Engineering 179(1–2):81–89.
Downloaded by [York University Libraries] at 06:08 22 August 2013

Felippa, C. A. (1999). Nonlinear Finite Element Methods. ASEN 5007, Lecture Notes.
University of Colorado at Boulder.
Feng, Y. T., Peric, D., Owen, D. R. J. (1996). A new criterion for determination
of initial loading parameter in arc-length methods. Computers and Structures
58(3):479–485.
Fujikake, M. (1985). Simple approach to bifurcation and limit point calculations.
International Journal for Numerical Methods in Engineering 21(1):183–191.
Fabiena, B. C., Longmana, R. W., Freudenstein, F. (1991). Stability of lifting rigs
with offset load. Mechanics Based Design of Structures and Machines 19(2):193–211.
Fujii, F., Ikeda, K., Noguchi, H., Okazawa, S. (2001). Modified stiffness iteration to
pinpoint multiple bifurcation points. Computer Methods in Applied Mechanics and
Engineering 190(18–19):2499–2522.
Fujii, F., Noguchi, H. (2002). The buckling mode extracted from the LDLT -
decomposed large-order stiffness matrix. Communications in Numerical Methods in
Engineering 18(7):459–467.
Ikeda, K., Murota, K., Yanagimoto, A., Noguchi, H. (2007). Improvement of
the scaled corrector method for bifurcation analysis using symmetry-exploiting
block-diagonalization. Computer Methods in Applied Mechanics and Engineering
196:1648–1661.
Kere, P., Lyly, M. (2011). Reissner-Mindlin-Von Kármán type shell facet model
for buckling simulation of imperfect cylindrical composite shells. Mechanics of
Advanced Materials and Structures 18(2):115–124.
Ligaro, S. S., Valvo, P. S. (2006). Large displacement analysis of elastic pyramidal
trusses. International Journal of Solids and Structures 43(16):4867–4887.
Moore, G., Spence, A. (1980). The calculation of turning points of nonlinear
equations. SIAM Journal on Numerical Analysis 17(4):567–576.
Noguchi, H., Chen, F. (2003). Eigenvector-free indicator, pinpointing and branch-
switching for bifurcation. Communications in Numerical Methods in Engineering
19(6):445–457.
Ohtmer, O. (1999). Nonlinear adaptive analysis via Quasi-Newton approach.
Advances in Engineering Software 30:595–605.
Onate, E., Matias, W. T. (1996). A critical displacement approach for predicting
structural instability. Computer Methods in Applied Mechanics and Engineering
134(1–2):135–161.
 FINDING STRUCTURAL CRITICAL POINTS 419

Planinc, I., Saje, M. (1999). A quadratically convergent algorithm for the computation
of stability points: the application of the determinant of the tangent stiffness matrix.
Computer Methods in Applied Mechanics and Engineering 169(1–2): 89–105.
Rezaiee-Pajand, M., Naghavi, A. R. (2011). Accurate solutions for geometric
nonlinear analysis of eight trusses. Mechanics Based Design of Structures and
Machines 39(1):46–82
Rezaiee-Pajand, M., Vejdani-Noghreiyan, H. R. (2006). Computation of multiple
bifurcation point. Engineering Computations 23(5):552–565.
Riks, E. (1979). An incremental approach to the solution of snapping and buckling
problems. International Journal of Solids and Structures 15:529–551.
Ritto-Correa, M., Camotim, D. (2008). On the arc-length and other quadratic
control methods: established, less known and new implementation procedures.
Computers and Structures 86:353–1368.
Downloaded by [York University Libraries] at 06:08 22 August 2013

Shi, J. (1996). Computing critical points and secondary paths in nonlinear structural
stability analysis by the finite element method. Computers and Structures 58(1):
203–220.
Shi, J., Crisfield, M. A. (1992). A simple indicator and branch switching technique
for hidden unstable equilibrium paths. Finite Elements in Analysis and Design
12(3–4):303–312.
Stewart, G. W., Sun, J. (1990). Matrix Perturbation Theory. Computer Science and
Scientific Computing. New York: Academic Press.
Thacker, W. I., Wang, C. Y., Watson, L. T. (1998). Stability and postbuckling of a
platform with flexible legs resting on a slippery surface. Mechanics Based Design
of Structures and Machines 26(3):277–286.
Vejdani-Noghreiyan, H. R. (2004). Geometrically Nonlinear Stability Analysis of
Structures. MS Thesis, Persian: Ferdowsi University of Mashhad.
Wang, C. Y. (2000). Analysis of nonlinear deformations of a triangular frame.
Mechanics Based Design of Structures and Machines 28(2):237–243
Wempner, G. A. (1971). Discrete approximations related to nonlinear theories of
solids. International Journal of Solids and Structures 7:1581–1599.
Wilkinson, J. H. (1965). The Algebraic Eigenvalue Problem. London: Oxford
University Press.
Wriggers, P., Simo, C. J. (1990). General procedure for direct computation of
turning and bifurcation points. International Journal for Numerical Methods in
Engineering 30:155–176.

APPENDIX A: SYMBOLS
u Nodal displacement vector
u Displacement increment
u Displacement corrector
p Load parameter
p Load corrector
p Load increment
r Unbalanced load vector
R Internal load resistance
e Reference load vector
K Tangent stiffness matrix
 420 REZAIEE-PAJAND ET AL.

Eigenvalue of tangent stiffness matrix

Normalized eigenvector of tangent stiffness matrix

APPENDIX B
In this section, the flowchart of the algorithm for finding the critical points is
presented.
Downloaded by [York University Libraries] at 06:08 22 August 2013