Journal of Structural Mechanics

ISSN: 0360-1218 (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/lmbd18

An Algorithm for Computing Critical Buckling

Loads of Elastic Structures

W. H. WITTRICK & F. W. WILLIAMS

To cite this article: W. H. WITTRICK & F. W. WILLIAMS (1973) An Algorithm for Computing
Critical Buckling Loads of Elastic Structures, Journal of Structural Mechanics, 1:4, 497-518, DOI:
10.1080/03601217308905354

To link to this article: https://doi.org/10.1080/03601217308905354

Published online: 03 Apr 2007.

Submit your article to this journal

Article views: 99

View related articles

Citing articles: 9 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=lmbd20
J. Strucl. Mech. 1 (4), pp. 497-518 (1973)

An Algorithm for Computing Critical

Buckling Loads o f Elastic Structures

W. H. WITTRICK
and
F. W. W I L L I A M S

ABSTRACT

The algorithm presented enables one to calculate the number of

critical buckling loads of an elastic structure which lie below any speci-
fied total load. and hence to converge on all required buckling loads
with ease and certainty. The algorithm is always valid when the total
load is applied proportionally. I t is also valid for nonproportional load-
ing, subject t o a single restriction.
The algorithm applies to continuous structures, as well as to struc-
tures with a finite number ofdegrees of freedom. I t is based on a stiffness
matrix analysis and substructures can be used.

497

Copyright 0 1973 by Marcel Dekker. Inc. All Rights Reserved. Neither this work nor any
part may be reproduced or transmitted in any form or by any means, electronic or mecha-
nical, including photocopying, microfilming, and wording. ar by any information storage
and retrieval syslem, without permission in writing from the publisher.
 W . H . Wittrick and F. W . Williams

I. INTRODUCTION

The authors recently presented a general algorithm for computing natural

frequencies o f elastic structures [I]. This algorithm enables one t o calculate
how many natural frequencies lie below any chosen frequency, without
determining them, and hence t o converge on any required natural frequency
to any specified accuracy. The problem of converging o n natural frequencies
is a n eigenvalue problem, as also is the problem of calculating the critical
buckling loads of structures. Indeed the two problems are s o similar that one
might expect an algorithm which can be applied in one c o n t c x t t o be ap-
plicable in the other also. However, there are two significant differences
between the vibration problem and the buckling problem, which are described
in the next two paragraphs. They both make the buckling problem more
complicated than the vibration problem. As a result the algorithm requires
modification t o make it applicable t o buckling problems. This is the purpose
of the present paper.

Fig. I Three simple structu;es. Lines represent pin-ended members and W is constant.
The values of Ffor which buckling accun are clearly all positive in case (a) and all
negative in case (b), whereas in case (c) some are negative and some are positive.
Algorithm for Computing Critical Buckling 499

The eigenvalues of a vibration problem are the squares of the natural

frequencies, and must therefore be positive. The eigenvalues of a buckling
problem are more difficult to define. In the simplest case, in which a single
load system is applied proportionally, so that each individual load is scaled
by a common load factor F, the eigenvalues are the various values of F for
which buckling can occur. They may be all positive, or all negative, or some
positive and some negative (see Fig. I). The derivation of the algorithm for
computing natural frequencies uses a theorem of Rayleigh's [2] which relates
to vibrations and was therefore proved for positive eigenvalues only. Thus
the foundation laid by Rayleigh is extended, in Section 11, to apply to
buckling as well as vibration problems, with the result that negative eigen-
values are now included.
Vibration problems include only one variable, namely, the square of the
frequency. However, buckling problems can involve the successive ap-
plication of two or more load systems, each of which has a load factor
associated with it (e.g., dead load and live load). This complication is con-
sidered in Section IV.
The algorithm which is presented in this paper enables one to converge
on all required buckling loads of any linearly elastic structure. Of course
usually only the lowest buckling load is required, and this can be obtained
without the use of the algorithm presented. However, the higher buckling
loads are sometimes required, as, for instance, in some postbuckling analyses.

I I . T H E O R E M OF CONSTRAINTS FOR A SYSTEM H A V I N G

BOTH POSITIVE A N D NEGATIVE EIGENVALUES

Consider a linearly elastic structure which, for the purposes of analysis,

is idealized as a system containing a finite number N of degrees of freedom.
The idealization may be effected, for example, by breaking it down into finite
elements. Let D, be a complete displacement vector, that is to say, a vector
containing N components corresponding to all the N degrees of freedom.
Also, let K, be the complete static stiffness matrix corresponding to D,.
Then, if the structure deforms, the strain energy stored is given by the quad-
ratic form

Suppose that the structure is subjected to a set of proportional loads.

whose magnitudes are defined by a single load factor F, and which are of
SO0 W . H . Wirtrick and F. W. Williams

such a nature that in the basic equilibrium state all the displacements D, are
zero. We consider first the problem of determining the critical values of F
(the eigenvalues) at which a bifurcation of equilibrium (buckling) can occur.
For sufficiently small deformations from the basic equilibrium state, the
loss of potential energy of the loads may be represented by the quadratic
form

It will be assumed that the structure is "anchored" so that it possesses no

rigid body freedoms. The stiffness matrix K, is therefore not only sym-
metrical but also positive definite. On the other hand the matrix C, though
symmetrical, is not necessarily positive definite.
The total energy E of the system may be written as

We now use the Weierstrass-Hilbert theorem which states that two

quadratic forms having symmetrical coefficients, one of which is positive
definite, can always be simultaneously diagonalized by a linear transforma-
tion of the form

where T is the ( N x N) transformation matrix and q is a new generalized

displacement vector. Moreover, in the new coordinate system the positive-
definite form has no negative coefficients. Applying this theorem to Eq. (3)
gives

where

A = T'K,T and B = T'CT

All the elements of the diagonal matrix A are positive, but those of the
diagonal matrix B may be either positive or negative.
Equation (5) may be simplified to
Algorifhm for Compufing Crifical Buckling 501

where Aj and Bj are the jth diagonal elements of A and B and q j is the jth
component of q.
Equilibrium o f t h e deformed state requires that

which implies that either qj = 0 or

Thus there are N real nonzero eigenvalues (buckling loads) given by

Eq. (a), which may be either positive or negative, depending upon the sign
of B,. The buckling mode corresponding to Fj is such that all components
o f q other than qj are zero. It follows from Eq. (4) that the displacement
vector ( D J j in the mode of buckling corresponding t o Fj is given by thejth
column of T.
.
Suppose that there are p positive eigenvalues F; ( k = 1, 2 , . . , p) and
n ( = N - p) negative ones - F ; (I = 1, 2 , . . . , n), both arranged in order
of numerical magnitude, i.c.,

We now consider the effect of applying a spring constraint, of stiffness y,

to one component, say the rth, of the displacement vector D,. From Eq. (4)
the rth component of D, is equal to z,"=, T,&, where T,, is the element in
row r, column s of T. Thus the additional strain energy stored in the spring is

a n d this must be added to the right-hand side of Eq. (6) to give the total
energy E of the new system. The equation of equilibrium (7) now becomes

Using Eq. (8) this becomes

502 W . H. Wittrick and F. W . Williams

Multiplying each of these equations by

and adding them a11 together we obtain the equation

where

Note that @ j 2 0, since all the A j are positive. Moreover 4j = 0 only if

T,j = 0, in which case it can be seen from Eq. (10) that an eigenvalue of the
constrained system is equal to the eigenvalue Fj of the unconstrained system.
The physical interpretation of this is that under these circumstances the rth
component of the vector D, is nodal in the jth buckling mode of the un-
constrained system and the introduction of the spring constraint obviously
has no erect on this particular eigenvalue. This is therefore a very special
case and in general T,;# 0 so that $ j > 0.
The criterion for buckling of the constrained system, from Eq. (I I), can
now be written as

If Eq. (13) is mul~ipliedthrough by F and then differentiated with respect

to F we obtain

and since 4, > 0 it follows that

Consider now the behavior of the function L(F). The following deductions
can be made from Eq. (13) and (14).
I. If - F ; iF < F: every term in the series in Eq. (13) is positive and
therefore L(F) > I/y.
Algorithm for Computing Criricol Buckling 503

2. If F -+ F,. L(F) -. m . Moreover, if F = Fj(l + E ) where E is infinites-

imally small, L(F) is given approximately by L(F) = -bj/&,
3. Equation (14) shows that L ( F ) l/y at any stationary value, i.e.,
whenever L'(F) = 0. Moreover, whenever L(F) = 0, FL'(F) > 0.
4. If IF1 is infinitely large, Eq. (13) shows that

where

Thus if a z 0, L(F) approaches l/y asymptotically from below a s F - t +a,

and from above as F -t - m. The converse occurs if a < 0.
These observations show that the function L(F) behaves as shown in
Fig. 2. ( F o r the purposes of illustration we have assumed that there are three
positive and two negative F,, i.e., p = 3 , 11 = 2, and N = 5.) The essential
feature is that there is one, and only one, zero of L(F) between each con-
secuLive pair of positive Fj, and between each consecutive pair of negative
F,. The following conclusions can be drawn.
a. If one (finite) spring constraint is imposed on a structure, the number
of positive eigenvalues p and the number of negative eigenvalues n remain
unchanged.
b. If the positive eigenvalues of the constrained structure are denoted by
F:,, FA,. . . , FL, and the negative ones by -F;, -F;, . . . , -F;, then
we have the following theorem.

and

If!he spring stiffness y is allowed t o become infinite the effect is to suppress

entirely one degree of freedom of the system. It follows that the new system
must then have ( N - I) eigenvalues, rather than the original N,and the ques-
tion arises as to whether the eigenvalue that is lost is a positive o r negative
one. The answer is easily seen from Fig. 2 by imagining that y -+ w , so that
the horizontal asymptote becomes coincident with the axis of F. It follows
W . H . Wi!!rick and F. W . Wi/lioms
Algorirhm for Comprdring Criricol Buckling 505

that F z - m if a > 0 and F,; -1 m if a < 0. Thus, as y - m the system

loses one positive eigenvalue if a > 0, or one negative one if a < 0.
I t may be noted that the equality signs have been inserted in Eq. (16) to
allow for two types of degenerate situation. First, if the displacement com-
ponent which is constrained is nodal in one of the buckling modes of the
unconstrained structure, we have previously seen that the corresponding
eigenvalue remains unchanged. Second, if the unconstrained structure has
m coincident eigenvalues, it can be seen from the inequalities (16). by letting
m of the Fjcoalesce, that the constrained structure must have at least (m - I)
equal coincident eigenvalues. As might be expected intuitively, it can be
proved that the corresponding (m - I) buckling modes of the constrained
structure are linear combinations of the m coincident buckling modes of
the unconstrained structure, such that the displacement component which
is being constrained is nodal.
Note that if the matrix C in Eq. (2) is positive definite, then so is B in
Eq. (5). All the Bj are then positive and hence the eigenvalues F, are all
positive. The first set of the inequalities (16) can then be interpreted in the
context of frequencies of vibration of the system, the result agreeing with the
theorem of Rayleigh [2] on which [I] was based.
Finally, we note that although the preceding argument has been based

is finite it remains valid as N-

upon the assumption that the number of degrees of freedom N of the system
w.

Ill. T H E E I G E N V A L U E PROBLEM FOR A N ASSEMBLY OF

SUBSTRUCTURES

For the purposes of deriving the theorem of constraints in Section I 1

the displacement vector D, was transformed into the principal coordinates
q so as to diagonalize the matrices Kc and C in Eq. (3). We now return to
this equation as it stands, so that the total energy E in the buckled state i:.

where

X, = K, - FC (18)

The equations of equilibrium are

506 W. H . Wirrrick and F. W. Williorns

where Dj is the jth component of D,. Hence

Equation (19) is the form in which the eigenvalue problem naturally arises
if. the structure is idealized into a n assembly of finite elements, with the
vector D, containing all the Ndegrees of freedom of the assembly. All buck-
ling displacements are zero if D, = 0 so that the criterion for buckling is

det X, = 0 (20)

This is a standard linear eigenvalue problem, for which numerous com-

putational procedures exist, and it is not the purpose o r this paper t o discuss
it in detail. However, as a preliminary step in the derivation of the later results
it is necessary to make certain observations, which we now proceed to do.
In general Eq. (19) will have both positive and negative eigenvalues, but
we shall concentrate entirely on those that are positive. In particular we shall
aim a t being able t o calculate how many eigenvalues lie in the range
0 < F < F * where F * is some chosen value of F. If this question can be
answered it will clearly be possible to answer the parallel question as t o how
many eigenvalues lie in the range -F* < F < 0, simply by replacing F
and C in Eq. (18) by - F and -C, respectively.
We note first the following two corollaries of Theorem I.
Corollary I: If one.constraint is released from a structure, the number
of eigenvalues in the range 0 < F < F* either remains unchanged o r in-
creases by one.
Corollary 2 : If t constraints are released from a structure, the number of
eigenvalues in the range 0 < F < F* increases by s, where 0 < s < r.
Let us denote by J(F*) the number of eigenvalues of the structure lying in
some chosen range 0 < F c F*. Now suppose that Nconstraints are imposed
so as to suppress all the N degrees of freedom of the structure. Clearly the
structure cannot buckle and J(F*) = 0. Suppose now that these constraints
are released one a t a time in the order of the vector D,. After (1 - I) con-
straints have been released it can be seen from Eq. (19) that the criterion for
,,
buckling is X,,_ = 0 where X i , _ , , is the leading principal minor of order
( t - I) of X,, i.e., the determinant of the matrix formed from the first
(r - I) rows and columns of X,. Similarly, after t constraints have been
released the buckling criterion is X , = 0. Figure 3 shows the variation of
,,
Xi,_ and X , with F. Theessential features of Figure 3 are that (i) both X ( , _ , ,
and X, are positive a t F = 0 since the structure is then stable and therefore
,,
X, is positive definite, and (ii) the roots of X ( , _ are intermediate between
those of X,, a fact that follows from Theorem I .
Algorithm for Computing Critical Buckling

,.,,
Fig. 3 Variation of X, and X,with F.

Now suppose that X,, X , , and X ( , - , , have the particular values X:, X:,
and X : - , , when F = F*. It is immediately apparent that if X,' and X & , ,
have the same sign (e.g., if F* = F: in Fig. 3 ) the release of the tth constraint
leaves J(F*) unaltered, whereas if X,' and X $ _ , , have opposite signs (e.g.,
if F* = F: in Fig. 3) J(F*) increases by one when the tth constraint is released.
Hence after all the constraints have been released J ( F t ) is equal to s { X : ] ,
the number of changes of sign between consecutive members of the Sturm
sequence

+
where X g is defined as 1 and XT is the leading principal minor of order
j of X r . s { X r } is called the sign count of the matrix X:. The sign count of a
real symmetrical matrix is more fully defined in [I], where it is shown that it
can be evaluated by the procedure outlined in the next paragraph.
Suppose XT is reduced to upper triangular form XZA by a simple Gaussian
elimination procedure in which rows are taken as pivotal in order, and suit-
able multiples of each pivotal row are added to all following rows, so that all
elements in the pivotal column which are below the pivot become zero. The
elements on the leading diagonal of X:A are XT, (X,*/X:), (X:/X:), . . . ,
(X,*/X& ,,). It follows that s { X : } is equal to the number of negative elements
on the leading diagonal of X:A. Also, for the unconstrained structure,

Furthermore, if J,,(F*) is the value of J(F*) when only the first t constraints
have been released, then
508 W. H . Wittrick and F. W. Williams

where Xz is the matrix formed from the first t rows and columns of X:.
Moreover s{X:] is the number of negative elements among the first t leading
diagonal elements of X:&.
Now consider Eq. (19) in the partitioned form

If constraints are applied to make D, = 0 then P, is the vector of required

clamping forces; on the other hand if the structure is completely uncon-
strained, then P, = 0. If the Gaussian elimination procedure in the form
described earlier is applied to Eq. (23), but is arrested after the first r rows
have been pivotal, we obtain

and completion of the procedure, for the unconstrained structure (i.e., for
P, = 0) gives

\
It is apparent from Eq. (24) that X, is the stiffness matrix corresponding
to the displacements D, only. Clearly it is not a linear function of F. Moreover
as F approaches values for which IX,J = 0, IX,I usually approaches infinity.
This is because, from Eq. (23), IX,J = 0 is the criterion for buckling when
constraints are applied to make D, = 0 whereas, from Eq. (25), IX,,I IX,I = 0
is' the criterion for buckling of the unconstrained structure, and these two
criteria are not usually both satisfied by the same value of F.
It now follows from Eq. (21) and (25) that

for the unconstrained structure, where X: is the value of X, when F = F*.

Substitution from Eq. (22) then gives the following result for the uncon-
strained structure.
Algorithm for Computing Critical Buckling 509

The significance of this result becomes clearer if it is stated a s follows.

The number of buckling loads, J(F*), of a structure in the range 0 < F < F*
is equal to the sum of the sign count, s{X:], of the stiffness matrix X : cor-
respondmg to any incomplete set o f buckling displacements D, plus the
number of buckling loads, J,(F*), which would still lie in the same range if
all the displacements D, were constrained with all other possible buckling
displacements unconstrained. This result has particular significance if D,
represents the buckling displacements a t the points of interconnection of a
set of substructures, a s J,(F*) is then the number o f buckling loads, in the
range 0 < F < F*, of the individual substructures when they are fully
constrained a t their points of interconnection.
Equation (26) remains valid if N -t m while D, remains of finite order.
I t can therefore be applied when structures are idealized a s a set of inter-
connected members which are treated as continuous, with an infinite number
of degrees of freedom. The vector D, can then be chosen to represent all
the buckling displacements at the nodes at which the members are connected
together.' X, will then depend upon F i n a transcendental way. However,
s{X:} can still be computed, and J,(F*) can be obtained from

where the summation includes all the individual members and J J F * ) is the
number of buckling loads, lying in the range 0 < F < F*, for an individual
member with its ends clamped. For some types of member (e.g., members
of plane frames, see Eq. (39)) a simple formula for JJF*) can be derived.
In some other cases the derivation of J,,,(F*) is a significant problem in its
own right.
Finally, since neither J,(F*) nor s(X:} can be negative, the criterion for
stability of a structure is

I V . THE CASE O F SEVERAL INDEPENDENT LOAD

SYSTEMS

In many practical situations it is not possible to express the loading in

terms of a single load system multiplied by a load factor F. In particular, the

'However substructures may be used, in which case D. will not contain all the displace-
ments at these nodes. Nevertheless Eq. (26) can still be used by applying i t to the sub-
structures before applying it to the whole structure, as described, for the similar problem
of finding natural frequencies, in [ I , 31.
510 W . H . Wirrrick and F. W . Williams

structure may be subjected t o a set of dead loads of fixed magnitude and a

set of live loads which are all proportional t o F. Consider, therefore, a
structure which is subjected t o two independent load systems, defined in
magnitude by load factors / and F. Then Eq. ( 2 ) for the loss of potential
energy of the loads during buckling must be replaced by

where C, and C are symmetrical, but not necessarily positive definite,

matrices. Equation (3) may now be replaced by

where

K, = K, - /Cd (31)

Equation (30) is identical t o Eq. (3) except that K, has been modified t o
K,. The derivation based on Eq. ( 3 ) makes use of the symmetry of Kcand C
and of the fact that K, is positive definite. Since K, and C in Eq. (30) are
symmetrical the conclusions drawn from Eq. ( 3 ) in Seciions I 1 and Ill remain
valid if Eq. (30) is used in place of Eq. ( 3 ) so long as K,,,, 1141ichreplaces K,,
is posiliw definire. Consequently Eq. (26) can be used t o compute J ( F 8 )
provided that the load system which depends o n / i s fully applied, ~ i r l ~ o u r
causing insrabilir?,. before F i s increased from zero. The criterion for stability
when / has its full w l u e / * and F = 0 is Eq. (28), with F* = /', as there is
only one load system applied a t this stage.
If the structure is still stable after F reaches its full value, it is clear that
the two load systems can be added together t o form a single load system
which could equally well have been reached by proportional loading using a
single load factor, instead of the two actually used. Such loading is governed
by Sections I I and Ill, and the argument of this section shows that Eq. (26)
can now be used a s a further load system is applied. S o long as the structure
is still stable Eq. (26) can still be used if yet another load system is applied,
and so on. However, a s soon a s one o r more buckling loads have been
exceeded during the application of a load system Eq. ( 2 6 ) cannot be used if
any further load systems are then applied.'
It is always possible t o think of the total load, W,, applied t o a structure
a s being the sum of any number ( j , say) o f other load systems, Wj. which are
chosen so that

21i Eq. (26) is used i t will sometimes give the correct answcr. However the derivation of the
equation given in this paper is not then valid, and the authon know of no way of predicting
whether the result obtained in such cases is correct or not.
Algorirhm for Comprrting Critical Buckling 511

Ifthese load systems are applied i n order, and the structure i s still stable
after Wi ( i = I, 2, . . . . ( j - I)) has been applied, the above argument shows
that Eq. (26) can be used as W , , , , , is applied. I n such cases the total load
will be said t o have been applied "via a permissible path." I n all other cases
the total load has been reached "via a nonpermissible path."
The physical meaning o f the terms i n Eq. (26) is, o f course. modified if
more than one load system has been applied, as follows. J(F*) is the total
number o f buckling loads passed during the entire loading process, i.e., as
the structure i s taken from the unloaded state t o the loaded state via a
permissible path. J,(F*) has the same meaning as J(F*), except that i t ap-
plies when constraints are applied t o make D, = 0. X : i s the value o f X,
when all the load systems have been applied. I t is clearly independent o f the
load path, as also is Xz. Thus, using Eq. (22), the value o f J(F') given by
Eq. (26) i s independent o f the load path. Hence the following theorem can
be deduced.
Theormt 2: I f the total load on a stmcture i n applied via any permissible
path, the number o f buckling loads passed i s independent of the particular
permissible path chosen and i s given by Eq. (26).
A n obvious corollary i s that i f the total load coincides with the rnth
buckling load when the total load i s applied via one permissible path, then
i t will also coincide with the 111thbuckling load when the total load is applied
via any other permissible path.
The interaction diagram of Fig. 4 illustrates the theory o f this section i n
the case o f a simple problem, for which W, is a function o f the two independ-
ent load systems W, and W,. The number i n a region between curves denotes
the value o f J(F*) for any total load, W,. (e.g., one o f the two denoted by
crosses) i n that region. I t is also the number of buckling curves crossed as
W, is applied via any permissible path. T w o buckling loads coincide at the
points of intersection o f curves.

V. BUCKLING O F A R I G I D - J O I N T E D T R I A N G U L A T E D
PLANE FRAME: AN EXAMPLE

F o r the purposes o f illustrating the use o f the algorithm represented by

Eq. (26) we consider the rigid-jointed plane frame shown i n Fig. 5 . This
consists o f I I identical members, each o f length I and flexural rigidily EI,
which will be assumed t o be inextensional. Loads w,,u3,, . . . , w, are applied
 W.H. Wittrick and F. W. WiNioms

Fig. 4 Interaction curves for the first five buckling loads of a structure. Permissible and
nonpermissible loading paths are denoted by solid and dotted straight lines,
repctively.

a s shown, where loj i s a nondimensional factor of the Euler load PC(= nZE1/l2)
of a member when its ends are pinned. The loads w, are assembled inro a
total load vector W, given by

For any given vector WT the axial compressive forces ( P ) in the I I members
designated a, b, . . . , k in Fig. 5 can be calculated from statics. We let
$2r. . . . ~,be the clockwise rotations of the seven joints during buckling
of the frame in its own plane. Because of the inextensional assumption, the
translational displacements of the joints are zero.
Algorithm for Computing Crirical Buckling

i ' i '
I
2 4 6
a c e 9 1 k
1
b

W2 w4 'tws

Fig. 5 Rigid-jointed plane frame with identical members.

.-
I
- I

Fig. 6 Forces and displacements at the ends of a member.

Figure 6 shows a member of the buckled frame, its end rotations being
4,and 4, and end moments M , and M 2 . The slope-deflection equations,
modified to take account of the destabilizing effect of the axial force P, are

where s and care the stability functions of Livesley and Chandler [4], defined
by the equations
514 W . H. Wirrrick and F. W. Williams

The equations of rotational equilibrium o f the seven joints during buckling

can now be written as
X,Y = o (36)
where

and the nonzero elements lying on or above the leading diagonal of the sym.
metrical matrix X, are

X I , = s, f sb XI 2 = (sc), XI 3 = (sc)~
X,, = s. + s, + s, X,, = (scL x,. = (sc)*
X,, = s, + s, + s, + s, X,, = (sc), X3, = ( S C ) ~
x,, = s, + s, + s, + s, x,, = (SC)~ X4, = (sc)~ (37)
" x55=sl+s,+s;+s, X,,=(SC)~ X,,=(SC)~
x6, = Sh + S; + Sk x67 = (sc)I
X,, = sj + s,

Buckling of a n individual member with ils ends clamped will occur when
s a n d l o r s c become infinite, see Eq. (34) and the following Eq. (25).
Using Eq. ( 3 9 , this implies that buckling occurs whenever

(sin a - a cos a)sin a =0 (a > 0) (38)

I t follows that, i f a = a* when F = F*, the number of buckling loads of the

member exceeded when its ends are clamped, J,(F*), is zero for P < 0, and
is otherwise given by

J,(F*) = j, + j, - f{l - (- I)jlsg(sin a* - a* cos a*)} (39)

where j, is the highest integer < a*/n, j, is the highest integer c {(a*/n) +
(1/2)} and sg(sin a* - a* cos a*) is + 1 o r - I depending upon the sign of
(sin a* - a*cos a*).
Algorilhm for Computing Critical B u c k l i ~ ~ g 515

J(F*) can now be calculated using Eqs. (26). (37), (27), and (39). Any
required buckling load can then be calculated as follows.
First, check that the structure is stable after each of the load systems
W i ( i = I , 2, . . . , ( j - I)) has been applied (see Eqs. (32) and (28)). Then
apply FWj, varying F i n a suitably systematic manner and computing J(F*)
for each value F * of Funtil upper and lower bounds on the required buckling
load have been found. Bisection between the current upper and lower bounds
will then give a new value of F* which will be n new upper or lower bound
depending upon the corresponding value of J(F*). This bisection procedure
can be applied successively until the required accuracy is obtained, i.e., until
the lower and upper bounds differ by less than a specified amount.
A simple computer program based on the theory outlined so far in this
section was used to obtain the results given for the three problems which
follow.
The first problem was to find the values of F a t the lst, 2nd, 3rd, and 10th
buckling loads for the case when

for the various combinations of w, and wp given in Table I, which also gives
the results obtained.

TABLE 1
Buckling Loads for the Problem of Fig. 5, w , = w , = w., w, = w , = w ,
w4 = ws =0

Value of F a t buckling load number

516 W . H . Wirrrick and F. W . Williams

The above problem clearly involves only one load system and therefore
does not use the theory of Section IV. However, the same total load, which
is the mth buckling load (m = I, 2, 3, or lo), can be reached in any number
of ways, see Eq. (32). Thus W . can be treated a s

where

and

If W, is applied first, and the structure is then stable, a further load f W , can
be applied and the computer program may be used to find the mth buckling
load. According to the corollary below Theorem 2, the value o f f t h u s found
should be unity. The second problem was therefore to check that f was unity
when this procedure was applied to 16 of the values of F given in Table 1,
W , being different in all I6 cases. For each of these cases the result obtained
for f was unity, to within the bounds of accuracy set when calculating f and F.
The third problem was to find the 9th, 18th, 27th, 36th, 45th, and 54th
buckling loads for

It is shown in the next paragraph that the six values thus obtained for F
should be the first six buckling loads of one of the members with its ends
clamped, expressed in units of P,,i.e., 4, 8.183, 16, 24.187, 36, and 48.188.
The computer program in fact gave these results, which supplied one check
on the correct functioning of the program.
The axial forces in the members, when W, is given by Eq. (42), can be
calculated by statics. They are zero in members d and h (see Fig. 5), and I +
+
(where denotes compression) in the other nine members. Thus as F is
increased through a value, F,,, for which the axial force in one of these
nine members is equal to the mth buckling load of the member with its ends
clamped, J,.(F*) increases from (m - I) to ni and, from Eq. (27), Jo(F*)
increases from 9(n1 - I ) to 9m. However, s{X,} must lie in the range
0 d s{X,} $ 7, since X, is (7 x 7), see Eq. (37). It follows from Eq. (26)
that J ( F a ) must increase from a t most (9m - 2) to a t least 9m as F passes
through F,,. Hence a t least two buckling loads, the (9m - I)-th and 9mth,
Algorithm for Comprrring C r i l i c o l Buckling 517

Fig. 7 Possible buckling modes for the frame of Fig. 5, with W, given by Eq. (42). when
F has the value for which the axial compre&ve force in all members except dand
h coincides with the mlh buckling load of a member with its ends clamped. (a)
Three possible independent modes form = I. (b) Two possible independent modes
for tn = 2.

of the structure occur a t F = F,,. Putting m = I , it follows that at least the

8th and 9th buckling loads for the structure occur when the axial forces in
all the members except d a n d h a r e equal to the lowest buckling load of these
members with their ends clamped. The existence of two independent buckling
modes in this case can easily be visualized (see Fig. 7(a))-indeed in this
case there is a third obvious possibility, as shown. In contrast, there only
appear to be two independent modes f o r m = 2, see Fig. 7(b). The computed
results confirmed these deductions to the extent that they showed three
coincident buckling loads for m = 1, 3, and 5 and two coincident buckling
loads for m = 2, 4, and 6. For computation purposes "coincidence" was
equated with "agreement to a t least one in lo9."

V I . CONCLUSIONS

Equation (26) enables one to calculate J(F*), the number of critical buck-
ling loads lying below any given total load. This information enables one to
converge on all required buckling loads with ease and certainty. In Eq. (26).
s{X:] denotes the sign count (as defined above Eq. (21)) of the overall
stiffness matrix of the loaded structure and J,,(F*) is the value J(F*) would
518 W . H . Wittrick and F. W . Williams

have if all the degrees of freedom corresponding t o X: were suppressed.

(J,(F*) is therefore zero if X: corresponds t o all the possible degrees of
freedom of the structure.) A particularly useful feature of Eq. (26) is that it
applies t o continuous structures, for which the number of degrees of freedom
is infinite. It can also be used when the final structure is assembled from
substructures.
Equation (26) is valid when the total load is applied proportionally. It is
also valid for nonproportional loading s o long a s such loading is applied via
a permissible path, a s defined below Eq. (32) and illustrated in Fig. 4.

REFERENCES

I. W. H. Wittrick and F. W. Williams, A general algorithm for computing natural

frequencies ofelasticstructures, Quart. J. Mech. Appl. Math. 24: 263-284(1971).
2. J. W. S. Rayleigh, The Theory of Sound, Vol. I , Dover, New York, 1945, p. 119.
3. F. W. Williams, Natural frequencies of repetitive structures, Quart. J. Mech.
Appl. Math. 24: 285-310 (1971).
4. R. K. Livesley and D. B. Chandler. Stabiliry Funcrions for Strucrural Frame-
works. Manchester University Press, Manchester. England. 1956.

Received May 8, 1972