Conversion From Imperfection-Sensitive Into Imperfection Insensitive Elastic Structures

Conversion from imperfection-sensitive into
imperfection-insensitive elastic structures I: Theory

Herbert A. Mang
a,
*
, Christian Schranz
a
, Peter Mackenzie-Helnwein
b
a
Institute for Mechanics of Materials and Structures, Vienna University of Technology, Karlsplatz 13, A-1040 Vienna, Austria
b
Department of Civil and Environmental Engineering, University of Washington, 12D More Hall, Seattle, WA 98195-2700, USA
Received 6 December 2004; received in revised form 4 February 2005; accepted 9 May 2005
Dedicated to Professor Thomas J.R. Hughes on the occasion of his 60th birthday
Abstract
A qualitative improvement of the initial postbuckling behavior of imperfection-sensitive elastic structures is their
conversion into imperfection-insensitive structures by means of modications of the original design. Such a conversion
is restricted to symmetric bifurcation. Koiters initial postbuckling analysis is applied in the framework of the FEM to
deduce mathematical relations for the transition from imperfection sensitivity to insensitivity, which may be achieved
by additional supports of the structure. This conclusion as well as several other conclusions from the theoretical inves-
tigation reported in Part I of this paper are corroborated by the results from a comprehensive numerical investigation
documented in Part II of this work.
2005 Elsevier B.V. All rights reserved.
Keywords: Symmetric bifurcation buckling; Imperfection sensitivity; Conversion into imperfection insensitivity; Koiters initial post-
buckling analysis; Finite element method
1. Introduction
This Euler column is imperfection insensitive and that cylindrical shell is imperfection sensitive . . .
students of structural engineering all over the world have heard such statements in the classroom, and
0045-7825/$ - see front matter 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.cma.2005.05.024
*
Corresponding author. Tel.: +43 1 58801 20210; fax: +43 1 58801 20299.
E-mail addresses: herbert.mang@tuwien.ac.at (H.A. Mang), christian.schranz@tuwien.ac.at (C. Schranz), pmackenz@u.washing-
ton.edu (P. Mackenzie-Helnwein).
Comput. Methods Appl. Mech. Engrg. 195 (2006) 14221457
www.elsevier.com/locate/cma
practicing structural engineers from all parts of the globe have read them in the technical literature. Some of
them may have come to the conclusion that the unfavorable mechanical diagnosis of imperfection sensitiv-
ity must be accepted as it is.
In the opinion of the writers, however, the acceptance of such a diagnosis would ignore the existence of
remedies. An obvious remedy is the conversion of imperfection-sensitive into imperfection-insensitive struc-
tures by means of modications of the original design. It leads to a qualitative improvement of the initial
postbuckling behavior of the structures, which often has a strong inuence on the entire postbuckling struc-
tural response. The title of the present paper refers to such a conversion which is the objective of this work.
One of several topics treated in this paper is design sensitivity analysis of the initial postbuckling behav-
ior of elastic structures. It plays a great role in the state-of-the-art. In a paper on design sensitivity analysis
of non-linear structures, Mro z and Haftka [9] also discussed the postbuckling behavior. The rst analyt-
ical work on design sensitivity analysis of the postbuckling behavior was presented by Godoy [4]. It is re-
stricted to consideration of the rst non-vanishing term in a series expansion for the load parameter. In an
extension of [9], Mro z and Piekarski [10] included imperfection insensitivity as a constraint condition for
optimization of the structural behavior. Bochenek and Kru_ zelecki [2] proposed an approach to optimize
the postbuckling behavior, which is based on determination of the maximum buckling load under the con-
straint that the structure is just no longer imperfection sensitive. In a paper on structural optimization of
the postbuckling behavior, Bochenek [1] mentioned the necessity to introduce constraints assuring symmet-
ric bifurcation, which is obviously needed for the conversion of an imperfection-sensitive into an imperfec-
tion-insensitive structure.
The present paper consists of two parts: (I) Theory and (II) Numerical investigation. Section 2 of Part I
deals with Koiters initial postbuckling analysis in the context of the nite element method (FEM). It is
emphasized that this mode of analysis primarily serves the purpose of deducing important theoretical re-
sults in Part I, which facilitate the verication of specic numerical results in Part II. The majority of
the structural analyses reported in Part II, is performed by means of the FEM, but without regard for Koi-
ters initial postbuckling analysis. Section 3 is devoted to symmetric bifurcation. In Section 4, new mathe-
matical conditions for the transition from imperfection sensitivity to imperfection insensitivity are
presented. Section 5 refers to the general case of non-linear prebuckling paths. In the framework of sensi-
tivity analyses also topics such as hilltop bifurcation and transition from bifurcation buckling to no loss of
stability are addressed. Section 6 covers the special case of linear prebuckling paths. Section 7 deals with the
completeness of solutions from Koiters initial postbuckling analysis, which are characterized by the van-
ishing of the rst one of those terms in the aforementioned series expansion for the load parameter, that
normally do not vanish for symmetric bifurcation. The conclusions of Part I (Section 8) are followed by
four Appendices. Appendix A is devoted to the computation of the coecient tensors for Koiters initial
postbuckling analysis in the context of the FEM. Appendices B and C contain mathematical details related
to Sections 2, 3 and 7, respectively. Appendix D contains a description of mathematical properties of the so-
called consistently linearized eigenvalue problem. The investigation of these properties is motivated by the
need to ensure the completeness of those specic solutions for the initial postbuckling paths, which were
mentioned in the context of the brief description of the contents of Section 7. Moreover, these properties
permit verication of theoretical results for limiting cases by inspection of corresponding eigenvalue
curves.
Part II of the paper consists of three sections: (1) Introduction, (2) Numerical investigation, and (3)
Conclusions.
Among the topics that are not treated in this paper are (a) multiple bifurcation, (b) material non-line-
arity, and (c) imperfections.
Re (a): The increase in mathematical complexity would outweigh the task-specic added value of informa-
tion resulting from consideration of multiple bifurcation.
H.A. Mang et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 14221457 1423
Re (b): The increase in programming work would outweigh the task-specic added value resulting from an
extension of the present work to material non-linearity, the more as such an extension would still
not include plasticity.
Re (c): Consideration of imperfections must be preceded by comprehension of the perfect situation, which
is the main goal of the present work.
This work is rmly embedded in the FEM. As mentioned previously, mathematical details that are re-
lated to the paper are given in Appendix A. Preliminary numerical studies on the inuence of mesh rene-
ment on the numerical results presented in Part II of this paper [12] were performed routinely. They are not
documented in the paper.
2. Koiters initial postbuckling analysis in the context of the FEM
The starting point of the theoretical investigation is Koiters initial postbuckling analysis [8]. Fig. 1 refers
to such an analysis. Point C denotes the bifurcation point. Point D is located on the secondary path. This
point is characterized by the load level k = k
D
and the corresponding displacement ~u(k
D
) v
D
. Primary and
secondary paths are represented by piecewise smooth curves u = ~u(k) and u = u
(g) = ~u(
~
k(g)) v
(g),
respectively, where g is a path parameter. For static, proportional loading,
G(u; k) := F
I
(u) P(k) = 0 (1)
is a necessary and sucient condition for equilibrium of mechanical systems discretized by the FEM [14].
F
I
(u) is the vector of the internal node forces, whereas P(k) = P
0
kP is the vector of the external node
forces. P
0
and P are given vectors of reference nodal loads.
One solution of (1) is the primary path u = ~u(k). Assuming that ~u(k) is known, a (non-linear) coordinate
transformation
(v; g) (u; k) = (~u(
~
k(g)) v;
~
k(g)) (2)
is performed such that (v = 0, g = 0) # (u
C
, k
C
). Substituting (2) into the expression for G(u, k) contained in
(1), yields the denition of the out-of-balance force in terms of v and g as
G
(v; g) := G(~u(
~
k(g)) v;
~
k(g)). (3)
Fig. 1. Initial postbuckling analysis at the bifurcation point C.
1424 H.A. Mang et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 14221457
The secondary path represents a non-trivial solution v = v
+
(g) of G
+
(v, g) = 0. Adopting Koiters concept of
an initial postbuckling analysis [8], G
+
(v, g) can be expressed as a Taylor series. Choosing the bifurcation
point C as the reference point characterized by v
0
= 0 and g
0
= 0, one obtains
G
(v; g) = G
(0; 0) G
;v
v G
;g
g
1
2
G
;vv
: v v G
;vg
vg
1
2
G
;gg
g
2
1
6
G
;vvv
.
.
.
v v v
1
2
G
;vvg
: v vg
1
2
G
;vgg
vg
2
1
6
G
;ggg
g
3
O(v v v v; v v vg; v vg
2
; vg
3
; g
4
). (4)
The computation of the rst-order tensors (vectors) G
;g
; G
;gg
; G
;ggg
; . . . ; second-order tensors (matrices)
G
;v
; G
;vg
; G
;vgg
; . . . ; third-order tensors G
;vv
; G
;vvg
; . . . ; fourth-order tensors G
;vvv
; . . . ; etc., is explained in
Appendix A. Each single term in (4) represents a vector-valued function of v and g.
Assuming a suciently smooth solution for both v
+
(g) and
~
k(g), these functions can be formulated as
series expansions [8,11]:
v
(g) = v
1
g v
2
g
2
v
3
g
3
; (5)
~
k(g) = k
C
k
1
g k
2
g
2
k
3
g
3
; (6)
where v
1
, v
2
, v
3
, . . . are the residual vectors and k
1
, k
2
, k
3
, . . . are load coecients associated with the secondary
(postbuckling) path. The residual vectors determine the deformation pattern of this path. The load coe-
cients govern the type of bifurcation, i.e. symmetric/unsymmetric and imperfection sensitive/insensitive, not-
ing that unsymmetric bifurcation is always imperfection sensitive.
Substituting (6) and (5) into (4) and re-ordering the resulting relation according to the order of g, yields
G
(v
(g); g) = g
1
(
K
T
v
1
) g
2

K
T
v
2
k
1
K
T;k
v
1

1
2
K
T;u
: v
1
v
1
_ _
g
3

K
T
v
3
k
2
K
T;k
v
1
k
1
K
T;k
v
2

1
2
k
2
1
K
T;kk
v
1
K
T;u
: v
1
v
2
_
1
2
k
1
K
T;uk
: v
1
v
1

1
6
K
T;uu
.
.
.
v
1
v
1
v
1
_
g
4
(see Appendix B)
g
5
(see Appendix B) g
6
(see Appendix B) O(g
7
) = 0. (7)
For this work the coecient vectors of g
1
, g
2
, . . . , and g
6
are needed. Because of the relatively great length of
the expressions for the coecients g
4
, g
5
, and g
6
, they have been transferred to Appendix B. Underlined
matrices in (7) and in other expressions in this section vanish in case of linear primary (prebuckling) paths.
Such matrices also occur in Sections 35 and 7, and in Appendices AD. However, in these sections and in
Appendix D they are not underlined because the special case of linear prebuckling paths is not explicitly
considered therein.
For (7) to be satised, each expression in parentheses must vanish separately. The resulting relations
allow to compute v
1
, k
1
and v
2
, k
2
and v
3
, . . . , k
5
and v
6
successively [11]. The following brief demonstration
is restricted to computation of v
1
, k
1
and v
2
, and k
2
. It starts with setting the coecient vector of g
1
equal to
zero, which gives
K
T
v
1
= 0; (8)
where v
1
is the eigenvector. Setting the coecient vector of g
2
equal to zero and premultiplying it by v
T
1
,
enables computation of k
1
as
k
1
=
1
2
v
T
1
K
T;u
: v
1
v
1
v
T
1
K
T;k
v
1
. (9)
Eq. (9) can be rewritten formally as
a
0
k
1
b
0
= 0; (10)
where
a
0
=
v
T
1
K
T;k
v
1
v
T
1
K
T;k
v
1
= 1 (11)
and
b
0
=
1
2
v
T
1
K
T;u
: v
1
v
1
v
T
1
K
T;k
v
1
. (12)
The motivation for this formulation is consistency with relations which will be presented in the following.
Setting the expression for the coecient vector of g
2
equal to zero, yields
K
T
v
2
=
K
T;k
v
1
k
1

1
2
K
T;u
: v
1
v
1
(13)
with k
1
according to (9). From (13) v
2
can be computed, noting that

K
T
has a rank one deciency at the
stability limit, with v
1
as the eigenvector (see (8)). Eq. (9) ensures that the right-hand side of (13) is orthog-
onal to the eigenvector v
1
. Setting the coecient vector of g
3
equal to zero and premultiplying it by v
T
1
, en-
ables computation of k
2
as
k
2
=
1
v
T
1
K
T;k
v
1
1
2
k
2
1
v
T
1
K
T;kk
v
1
k
1
v
T
1
K
T;k
v
2

1
2
v
T
1
K
T;uk
: v
1
v
1
_ _
v
T
1
K
T;u
: v
1
v
2
_
1
6
v
T
1
K
T;uu
.
.
.
v
1
v
1
v
1
_
. (14)
Eq. (14) can be rewritten formally as
a
1
k
2
1
b
1
k
1
c
1
= 0; (15)
where
a
1
=
1
2
v
T
1
K
T;kk
v
1
v
T
1
K
T;k
v
1
(16)
represents a so-called non-linearity coecient which vanishes trivially, i.e. because of

K
T;kk
= 0, in case of
linear prebuckling paths,
b
1
=
1
v
T
1
K
T;k
v
1
v
T
1
K
T;k
v
2

1
2
v
T
1
K
T;uk
: v
1
v
1
_ _
(17)
and
c
1
= d
1
k
2
(18)
with
d
1
=
1
v
T
1
K
T;k
v
1
v
T
1
K
T;u
: v
1
v
2

1
6
v
T
1
K
T;uu
.
.
.
v
1
v
1
v
1
_ _
. (19)
By analogy to (10) and (15), the following formal relations are obtained by setting the expressions for the
coecient vectors of g
4
, g
5
, and g
6
in (B.1) (see Eqs. (B.4)(B.6)) equal to zero and premultiplying the result-
ing relations by v
T
1
=v
T
1
K
T;k
v
1
:
a
+
1
k
3
1
b
+
1
k
2
1
c
+
1
k
1
d
+
1
= 0 with d
+
1
= e
+
1
k
3
; (20)
â
1
k
4
1

^
b
1
k
3
1
^c
1
k
2
1

^
d
1
k
1
ê
1
= 0 with ê
1
=
^
f
1
k
4
; (21)
~a
1
k
5
1

~
b
1
k
4
1
~c
1
k
3
1

~
d
1
k
2
1
~e
1
k
1

~
f
1
= 0 with
~
f
1
= ~ g
1
k
5
. (22)
The coecients a
+
1
, â
1
, and ~a
1
are given as
a
+
1
=
1
6
v
T
1
K
T;kkk
v
1
v
T
1
K
T;k
v
1
; â
1
=
1
24
v
T
1
K
T;kkkk
v
1
v
T
1
K
T;k
v
1
; ~a
1
=
1
120
v
T
1
K
T;kkkkk
v
1
v
T
1
K
T;k
v
1
. (23)
Just as a
1
, these coecients vanish trivially in case of linear prebuckling paths. Hence, they also represent
non-linearity coecients. The coecients b
+
1
,
^
b
1
, and
~
b
1
are given as
b
+
1
=
1
v
T
1
K
T;k
v
1
1
2
v
T
1
K
T;kk
v
2

1
4
v
T
1
K
T;ukk
v
1
v
1
_ _
; (24)
^
b
1
=
1
v
T
1
K
T;k
v
1
1
6
v
T
1
K
T;kkk
v
2

1
12
v
T
1
K
T;ukkk
v
1
v
1
_ _
; (25)
~
b
1
=
1
v
T
1
K
T;k
v
1
1
24
v
T
1
K
T;kkkk
v
2

1
48
v
T
1
K
T;ukkkk
v
1
v
1
_ _
. (26)
The expressions for the coecients c
+
1
,
^
f
1
, ~e
1
, which will be needed later, are given in Appendix C.
3. Symmetric bifurcation
Symmetric bifurcation is characterized by
k
1
= k
3
= = 0. (27)
If this condition is not satised by the original structure and for the given loading, it must be enforced in the
course of the conversion process [1]. This may require modications of the original design which, for dif-
ferent reasons, are unfeasible. Irrespective of the feasibility of such modications, this step is beyond the
scope of the present work.
For symmetric bifurcation, some of the coecients of (10), (15), and (20)(22) must vanish. In this con-
text, the term vanish means that the respective coecient is zero for arbitrary values of a design param-
eter j. Additionally, some of the remaining coecients may vanish. This may either be the case for arbitrary
or for specic values of j. The ones that must vanish are underlined in the following array:
a
0
b
0
; (28)
a
1
b
1
c
1
; (29)
a
+
1
b
+
1
c
+
1
d
+
1
; (30)
â
1
^
b
1
^c
1
^
d
1
ê
1
; (31)
~a
1
~
b
1
~c
1
~
d
1
~e
1
~
f
1
. (32)
Substitution of (18) into
c
1
= 0 (33)
(see (29)) yields
k
2
= d
1
; (34)
with d
1
according to (19). Substitution of (C.3) with k
3
= 0 into (21.2) and insertion of the result into
ê
1
= 0 (35)
(see (31)) gives
k
4
= a
1
k
2
2
b
2
k
2
d
3
(36)
with a
1
, b
2
, and d
3
according to (16), (C.2) and (C.4), and with k
2
according to (34).
The solution k
1
= 0 requires the vanishing of b
0
, c
1
, d
+
1
, ê
1
, and
~
f
1
(see (28)(32)). k
1
= 0 is a necessary but
not a sucient condition for symmetric bifurcation. To understand, why also b
1
, b
+
1
,
^
b
1
and
^
d
1
, and
~
b
1
and
~
d
1
must vanish for symmetric bifurcation (see (29)(32)), at rst, (15) is rewritten as
k
1

b
1
a
1
_ _
k
1

c
1
a
1
= 0 (a
1
,= 0). (37)
The relations
b
1
= b
1
(j) = 0 and c
1
= c
1
(j) = 0 (38)
(see (29)) indicate that k
1
= 0 is a double root of (15). Eq. (38.1) results in the disintegration of
d
+
1
=
(B.4)
b
1
k
2
d
+
= d
+
1
(j) =
(30)
0
with
d
+
=
1
v
T
1
K
T;k
v
1
v
T
1
K
T;u
v
1
v
3

1
2
v
T
1
K
T;u
v
2
v
2

1
2
v
T
1
K
T;uu
v
1
v
1
v
2

1
24
v
T
1
K
T;uuu
v
1
v
1
v
1
v
1
_ _
(39)
into b
1
k
2
= 0 and d* = 0, reecting the fact that symmetric bifurcation (see (27)) is a stronger requirement
than k
1
= 0.
Substitution of
b
+
1
= 0; d
+
1
= 0; (40)
(see (30)) into (20) yields
k
2
1

c
+
1
a
+
1
_ _
k
1
= 0 (a
+
1
,= 0) k
1
( )
1
= 0; k
1
( )
2;3
=
c
+
1
a
+
1
. (41)
Hence, k
1
= 0 is a single root of (20). Substitution of
^
b
1
= 0;
^
d
1
= 0 (42)
(see (31)) into (21) gives
k
2
1

^c
1
â
1
_ _
k
2
1
= 0 (â
1
,= 0) k
1
( )
1;2
= 0; k
1
( )
3;4
=
^c
1
â
1
. (43)
Hence, k
1
= 0 is a double root of (21). Substitution of
~
b
1
= 0;
~
d
1
= 0;
~
f
1
= 0 (44)
(see (32)) into (22) results in
k
4
1

~c
1
~a
1
k
2
1

~e
1
~a
1
_ _
k
1
= 0 (~a
1
,= 0) k
1
( )
1
= 0; (k
1
)
2;3;4;5
=
~c
1
2~a
1
~c
1
2~a
1
_ _
2
~e
1
~a
1
_
. (45)
Hence, k
1
= 0 is a single root of (22). Thus, k
1
= 0 is a single (double) root of polynomials of odd (even)
degree in k
1
.
In addition to the underlined coecients in (28)(32), which must vanish for arbitrary values of j, c
+
1
is a
coecient that may vanish for arbitrary values of j. Substitution of
c
+
1
= c
+
1
(j) = 0 (46)
into (41.1) yields
k
3
1
= 0 k ( )
1;2;3
= 0. (47)
Hence, k
1
= 0 is a triple root of (20). Substitution of (C.1) into (46) results in
2a
1
k
2
b
2
= 0 (48)
with b
2
according to (C.2). Solving (36) for k
2
, gives
k
2
=
b
2
2a
1
b
2
2a
1
_ _
2
d
3
k
4
a
1
(a
1
,= 0); (49)
where the positive sign in front of the square root correlates with (34).
Because of (48), the discriminant in (49) vanishes:
b
2
2
4a
1
(d
3
k
4
) = 0. (50)
Elimination of b
2
in (50) by means of (48) yields
k
4
= a
1
k
2
2
d
3
. (51)
If (46) holds, then also
^c
1
= ^c
1
(j) = 0; ~c
1
= ~c
1
(j) = 0; ~e
1
= ~e
1
(j) = 0 (52)
must hold. Substitution of (52.2) and (52.3) into (45.1) yields
(k
1
)
1;2;3;4;5
= 0. (53)
Hence, k
1
= 0 is a quintuple root of (22). Substitution of (C.5) with k
3
= 0 into (52.3) results in
2a
1
k
4
b
4
= 0 (54)
with b
4
according to (C.6). Thus, k
1
= 0 is a multiple root of polynomials in k
1
, with the obvious exception
of the linear polynomial (10).
If (46) holds true, a
1
and k
2
vanish for the same value of j, for which, following from (48), b
2
= 0. This is
not the case, if (46) is not valid. Rewriting (36) as
1
k
2
_ _
2
b
2
d
3
k
4
1
k
2
_ _
a
1
d
3
k
4
= 0 (d
3
k
4
,= 0) (55)
and specializing (55) for
a
1
d
3
k
4
= 0; (56)
gives
1
k
2
b
2
d
3
k
4
_ _
1
k
2
= 0. (57)
The two solutions of (57) are
1
k
2
_ _
1
= 0 and
1
k
2
_ _
2
=
b
2
d
3
k
4
. (58)
As follows from (58.2) for b
2
50 and d
3
k
4
50, (k
2
)
2
50.
Remarkably, symmetric bifurcation from non-linear prebuckling paths is associated either with
v
+T
j

K
T;kk
v
1
= 0; j = 2; 3; . . . ; n; (59)
where v
+
j
is the jth eigenvector of the so-called consistently linearized eigenproblem (see Appendix D), or
with
3 v
T
1
K
T;kk
v
1
_ _
2
2 v
T
1
K
T;k
v
1
_ _
v
T
1
K
T;kkk
v
1
_ _
= 0. (60)
Eq. (59) may but need not occur together with c
+
1
(j) = 0 Eqs. (48) and (51), whereas Eq. (60) only oc-
curs together with c
+
1
(j) = 0. Eqs. (59) and (60) result in two dierent modes of disintegration of an expres-
sion that holds for unsymmetric bifurcation from non-linear prebuckling paths (see (145)(147)). This
signals two dierent modes of transition from imperfection sensitivity to insensitivity.
Substitution of (16) and (23.1) into (60) yields
a
2
1
a
+
1
= 0. (61)
As a special case of (61), a
1
and a
+
1
may vanish for arbitrary values of j, i.e.
a
1
= a
1
(j) = 0; a
+
1
= a
+
1
(j) = 0; (62)
or for specic values of j. As a special case of (59) as well as of (62.1),
K
T;kk
v
1
= 0. (63)
4. Triples of values k
4
, a
1
, a
+
1
for k
2
= 0
This section refers to the general case of non-linear prebuckling paths. Hence,

K
T;kk
,= 0;

K
T;kkk
,= 0; . . .
Following from (34),
d
1
= 0 k
2
= 0. (64)
Substitution of (64.2) into (36) gives
k
4
= d
3
. (65)
The triples of values k
4
, a
1
, a
+
1
for k
2
= 0, which will be presented in the following, determine whether a
transition from imperfection sensitivity to imperfection insensitivity occurs and, if it does, how it occurs.
For (64.1) together with
v
+T
j

K
T;kk
v
1
= 0; j = 2; 3; . . . ; n; a
1
,= 0 and a
+
1
,= 0; (66)
or, in case of (46), together with
K
T;kk
v
1
= 0
(16)
a
1
= 0 and a
+
1
,= 0; (67)
the following result is obtained:
k
2k
= 0; k = 1; 2; . . . (68)
(see Part II of this work [12]). Eq. (68) requires that (59) holds for d
1
,= 0
(34)
k
2
,= 0. Substitution of (27)
and (68) into (6) yields
~
k(g) = k
C
= const. (69)
Such a transition was investigated by Tarnai [13] and by the authors of Part II of this work [12] for the case
of a bar-and-joint assembly composed of rigid members and elastic springs. It also occurs for a von Mises
truss with an elastic spring attached to the load point [12].
v
T
1
K
T;kk
v
1
= 0
(16)
a
1
= 0 and v
T
1
K
T;kkk
v
1
= 0
(23.1)
a
+
1
= 0; (70)
satisfying (61), the following result is obtained:
k
2
= 0 and k
4
< 0 (see Part II of this work [12]). (71)
The relations (71) require that (60) holds for d
1
,= 0
(34)
k
2
,= 0.
K
T;kk
v
1
= 0
(16)
a
1
= 0 and v
T
1
K
T;kkk
v
1
= 0
(23.1)
a
+
1
= 0; (72)
satisfying (61), the following result is obtained:
k
2
= 0 and k
4
= 0 (see Part II of this work [12]). (73)
Eqs. (73) are associated with
k
6
< 0 (74)
and, thus, with imperfection sensitivity (see Part II of this work [12]). The relations (73) and (74) require
that (60) holds for d
1
,= 0
(34)
k
2
,= 0.
Hence, for k
2
= 0, the following triples of values k
4
, a
1
, a
+
1
are obtained:
k
4
= 0; a
1
< 0; a
+
1
< 0; (75)
k
4
= 0; a
1
= 0 (with

K
T;kk
v
1
= 0); a
+
1
> 0; (76)
k
4
< 0; a
1
= 0 (with

K
T;kk
v
1
,= 0); a
+
1
= 0; (77)
k
4
= 0; a
1
= 0 (with

K
T;kk
v
1
= 0); a
+
1
= 0. (78)
The thick parts of the two lines in Fig. 2 show the geometric loci of all points in the k
4
a
1
plane of the k
2
k
4
a
1
space, which are solutions of (36) with k
2
= 0. It is seen that these geometric loci are restricted to the
two half-axes k
4
6 0 and a
1
6 0. This seems to be the consequence of the restriction to symmetric bifurca-
tion, excluding, e.g. the possibility of k
1
= 0, k
2
= 0, k
3
50, . . . Another consequence of this restriction is
the fact that only ve out of the eight octants into which the three-dimensional space can be divided by the
k
2
k
4
a
1
coordinate system are geometric loci of triples of values (k
2
, k
4
, a
1
) that are solutions of (36) (see
Fig. 3). The octants I, II, and IV are characterized by
v
T
1
K
T;k
v
1
= 1 (see (D.11)) and v
T
1
K
T;kk
v
1
P0; (79)
octant V by
v
T
1
K
T;k
v
1
= 1 (see (D.11)) and v
T
1
K
T;kk
v
1
6 0; (80)
and octant VII by
v
T
1
K
T;k
v
1
= 1 (see (D.11)) and v
T
1
K
T;kk
v
1
6 0 (81)
or by
v
T
1
K
T;k
v
1
= 1 (see (D.11)) and v
T
1
K
T;kk
v
1
P0; (82)
all of which correlate with k = k
C
> 0 (see Fig. 1). A positive value of k
C
can always be achieved by means
of a suitable denition of a positive reference load.
5. General case: Non-linear prebuckling paths
5.1. Sensitivity analysis
Fig. 4 shows qualitative plots of eight curves k
2
= k
2
(j), k
4
= k
4
(j), a
1
= a
1
(j). For each point on these
curves Eq. (8) holds. Each curve contains at least one point T, characterized by
k
2
= 0. (83)
The plane curve S = T F in Fig. 4(e), which is also located in this plane, is the limiting case of space curves
of the form shown in Fig. 4(d). The curve STF in Fig. 4(b) is located in the plane k
4
= 0. The vertical line
S = T F = T in Fig. 4(f), the horizontal line S = T F = T in Fig. 4(h), and point S = F = T in Fig. 4(g)
Fig. 2. Half-axes k
4
6 0 and a
1
6 0 as geometric loci of all points associated with k
2
= 0.
Fig. 3. Five octants as geometric loci of triples of values (k
2
, k
4
, a
1
) for k
C
> 0.
(a)
(c) (d)
(e)
(f)
(g) (h)
(b)
Fig. 4. Qualitative plots of curves k
2
= k
2
(j), k
4
= k
4
(j), a
1
= a
1
(j), with (at least) one point T(k
2
= 0, k
4
, a
1
). [j refers to the stiness of
a vertical elastic spring attached to the vertex of a pin-jointed bar (a) and of a von Mises truss (b) and to the center of a cylindrical panel
with two dierent thicknesses ((d) and (e)), respectively, at which a vertical load is applied; to the thickness of the panel (c); and to the
initial rise of the pin-jointed bar (f), of the truss (g), and of the panel without and with the spring ((g) and (h), respectively); details of
the three structures are given in Figs. 1, 4, and 9 in Part II of this work [12].]
represent degenerations of curves. The arrows on the curves in Fig. 4 correspond to the increase of the value
of the design parameter j. The starting point of such a curve is denoted as S and the nal point as F. The
starting points in Fig. 4(a)(d) and the nal points in Fig. 4(a)(e) are arbitrarily chosen points. The nal
points in Fig. 4(f)(h) refer to nal situations (transition from bifurcation buckling to no buckling). The
corresponding values of j are j
S
and j
F
> j
S
. In line with the purpose of this paper, reected by its title, S is
restricted to
k
2
6 0. (84)
If the sign of inequality holds in (84) (see Fig. 4(a)(d)),
j
S
< j
T
; (85)
where j
T
is the value of j corresponding to point T. The sign of equality in (84) holds for S = T (see
Fig. 4(e)(h)) with
j
S
= j
T
. (86)
In Fig. 4(e), k
2
(j) P0, k
4
(j) = 0, a
1
(j) P0, in Fig. 4(f), k
2
(j) = 0, k
4
(j) = 0, a
1
(j) < 0, in Fig. 4(g),
k
2
(j) = 0, k
4
(j) = 0, a
1
(j) = 0, and in Fig. 4(h), k
2
(j) = 0, k
4
(j) 6 0, a
1
(j) = 0. As shown in Fig. 4(b),
k
2
(j = j
F
) may be negative. Point H is a bifurcation point coinciding with a snap-through point. In
Fig. 4(f)(h), F = T represents the point of transition from bifurcation buckling to no buckling, denoted
as N. It is characterized by the degeneration of the secondary paths to one point each on the respective
loaddisplacement diagram, coinciding with a saddle point on this curve. The dashed curves in Fig. 4
are the projections of the respective curves onto the k
2
k
4
plane. S
0
and F
0
are the projections of S and
F onto this plane.
According to Fig. 4, at point T, with the exception of Fig. 4(a) and (f) for which (48) does not hold,
a
1
= 0. A distinctive feature at point T(k
2
= 0, k
4
, a
1
= 0) are the two dierent modes of vanishing of a
1
.
In contrast to the situation at point T in Fig. 4(d) and (h), which is characterized by k
4
50, at point T
in Fig. 4(b), (c), (e), and (g), with k
4
= 0,
K
T;kk
v
1
= 0
(16)
a
1
= 0; (87)
which is a stronger condition than the vanishing of the quadratic form v
T
1
K
T;kk
v
1
.
Another distinctive feature of the curves k
2
= k
2
(j), k
4
= k
4
(j), a
1
= a
1
(j) at point T(k
2
= 0, k
4
, a
1
= 0)
follows from the rst two partial derivatives of (48) with respect to j, which are obtained as
2 a
1;j
k
2
a
1
k
2;j
( ) b
2;j
= 0 and 2(a
1;jj
k
2
2a
1;j
k
2;j
a
1
k
2;jj
) b
2;jj
= 0. (88)
Specialization of Eq. (88) for k
2
= 0 and a
1
= 0 gives
b
2;j
= 0 and 2a
1;j
k
2;j
b
2;jj
= 0. (89)
At point T(k
2
= 0, k
4
, a
1
= 0) also k
2,j
may vanish, which is the case at this point in Fig. 4(c) and (g), and at
points T in Fig. 4(h). In contrast to point T in Fig. 4(c), at point T in Fig. 4(g) and at points T in Fig. 4(h),
a
1;j
= 0. (90)
Dierentiation of (16) with respect to j yields
a
1;j
=
1
2
(2v
T
1
K
T;kk
v
1;j
v
T
1
K
T;kkj
v
1
)v
T
1
K
T;k
v
1
(v
T
1
K
T;k
v
1
)
2

(2v
T
1
K
T;k
v
1;j
v
T
1
K
T;kj
v
1
)v
T
1
K
T;kk
v
1
(v
T
1
K
T;k
v
1
)
2
_ _
. (91)
Specialization of (91) for a
1
= 0, considering (16), and for a
1
,
j
= 0 results in
2v
T
1
K
T;kk
v
1;j
v
T
1
K
T;kkj
v
1
= 0; (92)
where
K
T;kkj
=

K
T;kkk
k
;j
K
T;kku
u
;j
with k
;j
= k
C;j
and u
;j
= u
C;j
. (93)
Substitution of (93) into (92) gives
2v
T
1
K
T;kk
v
1;j
v
T
1
K
T;kkk
v
1
k
C;j
v
T
1
K
T;kku
u
;j
v
1
= 0. (94)
At point T in Fig. 4(g) (von Mises truss), because of (87.1), (94) disintegrates into two parts. At points T in
Fig. 4(h), because of (62.2), (94) disintegrates into two other parts. At point T in Fig. 4(g) (cylindrical
panel), because of (62.2) and (87.1), (94) disintegrates into three parts.
Table 1 contains the values of k
2
,
j
, k
4
, and a
1
for points T(k
2
= 0, k
4
, a
1
) in Fig. 4(a)(h). The topic of
completeness of these solutions will be addressed in Section 7.
In the following, the individual plots of curves k
2
= k
2
(j), k
4
= k
4
(j), a
1
= a
1
(j) in Fig. 4 will be
discussed.
5.2. Discussion of Fig. 4(a)(h)
Each point of the eight curves in Fig. 4 is associated with Eq. (8). An arbitrary point of the curves in
Figs. 4(a) and (b) is additionally associated with Eq. (59) and an arbitrary point of the curves in
Fig. 4(c)(e) with Eq. (60).
Each point T of the eight curves in Fig. 4 is associated with Eq. (64). Point T in Fig. 4(a) and points T in
Fig. 4(f) are additionally associated with (66). Point T in Fig. 4(b), (c), (e), and (g) is additionally associated
with (67.1) and (72.1), respectively. Moreover, point T in Fig. 4(b) is also associated with (67.2), whereas
point T in Fig. 4(c) and (e) is also associated with (72.2); point T in Fig. 4(g) may also be associated with
this relation. Point T in Fig. 4(d) and point T in Fig. 4(h) are additionally associated with (70).
Fig. 4(a). At point T,
k
2
= 0; k
4
= 0; k
6
= 0; . . . ; a
1
< 0; a
+
1
< 0. (95)
At this point, the transition from imperfection sensitivity to imperfection insensitivity occurs. At point I,
k
2
> 0; k
4
> 0; k
6
> 0; . . . ;

K
T;kk
v
1
= 0
(16)
a
1
= 0; a
+
1
> 0. (96)
Fig. 4(b). At point T,
k
2
= 0; k
4
= 0; k
6
= 0; . . . ;

K
T;kk
v
1
= 0
(16)
a
1
= 0; a
+
1
> 0. (97)
At all other points, k
2k
= 0, k = 2, 3, . . .. The transition from imperfection sensitivity to imperfection
insensitivity occurs in the same way as at point T in Fig. 4(a). Dierences between postbuckling paths
referring to points in Fig. 4(a) and (b) other than points T are shown in Part II of this work [12].
Fig. 4(c). In contrast to Fig. 4(a) and (b), Fig. 4(c) does not indicate a conversion from an imperfection-
sensitive into an imperfection-insensitive structure at point T. However, a transition from k
4
> 0 to k
4
< 0
Table 1
Values of k
2,j
, k
4
, and a
1
for points T(k
2
= 0, k
4
, a
1
) in Fig. 4(a)(h)
Fig. 4
(a) (b) (c) (d) (e) (f) (g) (h)
k
2,j
>0 >0 0 >0 >0 0 0 0
k
4
0 0 0 <0 0 0 0 60
a
1
<0 0 0 0 0 <0 0 0
occurs at this point of the space curve. As mentioned previously, point H is a bifurcation point coinciding
with a snap-through point. This situation is referred to as hilltop bifurcation [3]. As will be shown in the
following, point H represents an improper cusp of the space curve in Fig. 4(b), characterized by
a
1
= . Hilltop bifurcation is restricted to octant VII in Fig. 3. Thus,
k
2
< 0; k
4
< 0. (98)
Obviously, hilltop bifurcation is associated with imperfection sensitivity.
In order to prove that a
1
= for point H,

K
T;k
and

K
T;kk
are expressed in terms of a path parameter n
and inserted in (16):
a
1
=
1
2
v
T
1
K
T;nn
k
;n

K
T;n
k
;nn
(k
;n
)
3
v
1
v
T
1
K
T;n
k
;n
v
1
=
1
2k
;n
v
T
1
K
T;nn
v
1
v
T
1
K
T;n
v
1
k
;nn
k
;n
_ _
. (99)
(In contrast to the path parameter g (see (2)), n refers to the primary path.)
At point H,
dk = 0 k
;n
= 0. (100)
Insertion of (100.2) into (99) and consideration of
k
;nn
< 0 (101)
and of the fact that the rst term in parentheses in (99) remains nite, results in
a
1
= . (102)
Furthermore, at point H,
a
1;j
= (; a
1;jj
= ; (103)
indicating the aforementioned improper cusp of the space curve k
2
= k
2
(j), k
4
= k
4
(j), a
1
= a
1
(j) in
Fig. 4(c).
As mentioned previously, the parameter of the space curve illustrated in Fig. 4(c) refers to the thickness
of a cylindrical panel (see Part II of this work [12]). According to expectation, the uniform increase of the
thickness of this structure results in an increase of the buckling pressure. However, it does not result in
the desired conversion of the initial postbuckling behavior from imperfection sensitive into imperfection
insensitive.
Fig. 4(d). Point S is assumed to be a point located on the space curve shown in Fig. 4(c). Hence, the
space curve in Fig. 4(d) can be thought of as being the second part of a sequence of two curves, the rst
of which is a portion of the space curve shown in Fig. 4(c). Because of the transition from k
2
< 0 to k
2
> 0
at point T, a conversion from an imperfection-sensitive into an imperfection-insensitive structure occurs.
In contrast to the situation at point T in Fig. 4(a)(c), characterized by k
4
= 0, however,
k
4
< 0. (104)
Fig. 4(e). This gure refers to the special case S = T of the general situation illustrated in Fig. 4(d).
Hence, point S = T agrees with point T in Fig. 4(c). Fig. 4(e) is characterized by
k
4
(j) = 0
(51)
a
1
k
2
2
d
3
= 0
(48) b
2
2
k
2
d
3
= 0. (105)
Fig. 4(f). Point S = T is assumed to agree with point T in Fig. 4(a). Hence, the thick vertical line in
Fig. 4(f) can be thought of as being the second part of a sequence of two curves, the rst of which is
a portion of the space curve shown in Fig. 4(a). Point F = N = T refers to the previously mentioned lim-
iting case of transition from bifurcation buckling to no buckling. It is characterized by
k
;n
= 0; k
;nn
= 0; v
1
= 0; (106)
indicating saddle points on the primary paths and the degeneration of the secondary paths to these
points, respectively (see point C in Fig. 5(a)). Hence, the second term in parentheses in (99) is an inde-
terminate expression. Application of de LHospitals rule to this expression gives
k
;nn
k
;n
=
0
0
=
k
;nnn
k
;nn
= . (107)
Since the rst term in parentheses in (99) remains nite,
a
1
= . (108)
The eigenvector of the singular matrix

K
T
follows from specialization of the innitesimally incremental
equilibrium equation
K
T
d~u = dkP (109)
for dk = 0. Thus, d~u is the eigenvector of

K
T
.
Fig. 4(g). Point S = T is assumed to agree with point T in Fig. 4(b) and (c), respectively. Hence, the origin
of the system of reference in Fig. 4(g) can be thought of as being the second part of a sequence of two
curves, the rst of which is a portion of the plane curve shown in Fig. 4(b) and the space curve in
Fig. 4(c) respectively. At point T, (67.1) holds true. Expressing

K
T;kk
in terms of the path parameter n, gives
K
T;nn
k
;n

K
T;n
k
;nn
(k
;n
)
3
. (110)
Substitution of (110) into (67.1) yields
K
T;nn

k
;nn
k
;n
K
T;n
_ _
v
1
= 0. (111)
Point F = N = T refers to the limiting case of transition from bifurcation buckling to no buckling, char-
acterized by (106) and Fig. 5(a). Application of de LHospitals rule to the indeterminate expression (k,
nn
/
k,
n
)v
1
gives
k
;nn
k
;n
v
1
=
0
0
0 =
k
;nnn
v
1
k
;nn
v
1;n
k
;nn
=
k
;nnnn
v
1
2k
;nnn
v
1;n
k
;nn
v
1;nn
k
;nnn
= 2v
1;n
. (112)
(a) (b) (c)
Fig. 5. Degeneration of secondary paths to a point on loaddisplacement curves [saddle point (Fig. 5(a)) and point of inection
(Fig. 5(b) and (c)), respectively].
Hence, for this limiting case,
K
T;n
v
1;n
= 0; (113)
which indicates that

K
T;n
is a singular matrix with v
1
,
n
as the eigenvector. As occurs for the limiting case
associated with point F = N = T in Fig. 4(f),
K
T
d~u = 0. (114)
Fig. 4(h). Point S = T is assumed to agree with point T in Fig. 4(d). Hence, the thick horizontal line in
Fig. 4(h) can be thought of as being the second part of a sequence of two curves, the rst of which is a
portion of the space curve shown in Fig. 4(d). At point T, (70.1) holds true. Consequently, the expression
in parentheses in (99) must vanish. Thus,
k
;nn
k
;n
=
v
T
1
K
T;nn
v
1
v
T
1
K
T;n
v
1
. (115)
Point F = N = T refers to the limiting case of transition from bifurcation buckling to no buckling, char-
acterized by (106) and Fig. 5(a). Application of de LHospitals rule to the indeterminate expressions in
(115) yields
k
;nnn
k
;nn
=
v
T
1;n
K
T;nn
v
1;n
v
T
1;n
K
T;n
v
1;n
. (116)
Because of k,
nn
= 0 and k,
nnn
50,
v
T
1;n
K
T;n
v
1;n
= 0. (117)
Just as point C in Fig. 5(a), also point C* in Fig. 5(b) and (c) refers to the transition to no loss of sta-
bility. (As far as Fig. 5(c) is concerned, this transition is irrelevant because it is preceded by snap-
through.) However, in contrast to the situation at point C in Fig. 5(a), where

K
T
is just still singular,
at point C* in Fig. 5(b) and (c),

K
T
has just become regular.
To investigate the situation at this point, Eq. (109) is rewritten as
K
T
~u
;k
= P. (118)
Dierentiation of (118) with respect to k gives
K
T
~u
;kk

K
T;k
~u
;k
~u
;k
= 0. (119)
At point C*, (119) disintegrates into
K
T
~u
;kk
= 0 and

K
T;k
~u
;k
~u
;k
= 0. (120)
Since

K
T
is regular,
~u
;kk
= 0. (121)
Expressing ~u
;kk
in terms of the path parameter n, yields
~u
;kk
=
~u
;nn
k
;n
~u
;n
k
;nn
(k
;n
)
3
. (122)
Substitution of (122) into (121) and consideration of k,
n
50 results in
~u
;nn
k
;n
~u
;n
k
;nn
= 0. (123)
Since (121) represents a global property in the sense that all active degrees of freedom are concerned,
(123) must disintegrate into
~u
;nn
= 0 and k
;nn
= 0. (124)
Eq. (121) must not be confused with the vanishing of the second derivative of specic degrees of freedom
with respect to k, such as
~u
i;kk
= 0
(123)
~u
i;nn
k
;n
~u
i;n
k
;nn
= 0; (125)
where
k
;n
,= 0; ~u
i;n
,= 0; k
;nn
,= 0; ~u
i;nn
,= 0. (126)
Numerical examples concerning the modes of transition from bifurcation buckling to no buckling illus-
trated in Fig. 5 will be presented in Part II of this work [12].
6. Special case: Linear prebuckling paths
In the rst paragraph of Section 5.2, referring to the general case of symmetric bifurcation from non-lin-
ear prebuckling paths, the assertion was made that an arbitrary point of the rst two curves in Fig. 4 is
associated with (see (59))
v
+T
j

K
T;kk
v
1
= 0; j = 2; 3; . . . ; n; (127)
and of the next three curves with (see (60))
3(v
T
1
K
T;kk
v
1
)
2
2(v
T
1
K
T;k
v
1
)(v
T
1
K
T;kkk
v
1
) = 0
(16),(23.1)
a
2
1
a
+
1
= 0 (61). (128)
These conditions are the reason for restrictions on the curves k
2
= k
2
(j), k
4
= k
4
(j), a
1
= a
1
(j) in Fig. 4.
For the special case of bifurcation from linear prebuckling paths,
K
T;kk
= 0
(16)
a
1
= 0;

K
T;kkk
= 0
(23.1)
a
+
1
= 0. (129)
Hence, the relations (127) and (128) are satised trivially. Because of the absence of non-trivial relations
replacing (127) and (128) for the special case considered, there are no restrictions on the plane curves
k
2
= k
2
(j), k
4
= k
4
(j), analogous to the ones for the general case.
Eq. (34) does not contain a term that vanishes for the special case of linear prebuckling paths. Hence,
irrespective of whether the prebuckling paths are non-linear or linear,
k
2
= d
1
. (130)
Because of (129.1), the rst term on the right-hand side of (36) vanishes trivially for the special case of linear
prebuckling paths. Hence, (36) is reduced to
k
4
= b
2
k
2
d
3
. (131)
The expressions for d
1
(see (19)), b
2
(see (C.2)), and d
3
(see (C.4)) do not contain terms that vanish for the
special case of linear prebuckling paths. Following from (130),
d
1
= 0 k
2
= 0. (132)
Hence, the condition for k
2
= 0 is the same as for the general case (see (64)). However, the additional con-
ditions associated with (64) for the general case, which where mentioned in the second paragraph of Section
5.2, are satised trivially for the special case of linear prebuckling paths.
Substitution of (132.2) into (131) gives
k
4
= d
3
; (133)
which agrees with (65). Following from (131),
b
2
k
2
d
3
= 0 k
4
= 0. (134)
Eq. (134.1) diers from (105.3). This reects the dierence between the situation at point Q in Fig. 6
and the one in Fig. 4(e), characterized by k
4,j
50 and k
4,j
= 0, respectively. Following from (130) and
(133), for
d
1
= 0 and d
3
= 0; (135)
k
2
= 0 and k
4
= 0. (136)
Fig. 6(a) and (b) shows plots of two curves k
2
= k
2
(j), k
4
= k
4
(j), which contain one point T and one point
Q each, for which
k
2
= 0 and k
4
= 0; (137)
respectively. In Fig. 6(c), these two points coincide.
7. Completeness of solutions from Koiters initial postbuckling analysis, containing k
2
= 0
For bifurcation from non-linear prebuckling paths,
v
+
1;k
= c
11
v
1
; (138)
where (see (D.14))
c
11
= a
1
. (139)
In general, a
1
50. Eq. (138) follows from specialization of (D.6) for
c
1j
= 0; j = 2; 3; . . . ; n; (140)
resulting from substitution of k
+
1
= k into (D.10). [k
+
1
k is the rst eigenvalue and v
+
1
is the corresponding
eigenvector of the so-called consistently linearized eigenproblem (see Appendix D). At the stability limit,
k
+
1
= k = k
C
and v
+
1
= v
1
.]
(a) (b) (c)
Fig. 6. Plots of curves k
2
= k
2
(j), k
4
= k
4
(j), with one point T(k
2
= 0, k
4
) and one point Q(k
2
, k
4
= 0) each. [j refers to the ratio of the
length of the rods of a pin-jointed bar. All rods are connected by hinges. Rotational and extensional springs are attached to the hinges
(the three illustrations refer to dierent values of the spring constant of the extensional spring); details of the structure are given in
Fig. 17 of Part II of this work [12].]
Distinctive features between unsymmetric and symmetric bifurcation from non-linear prebuckling paths
follow from
v
+
1;kk
=
n
j=1
c
1j;k
v
+
j
c
11
v
+
1;k
; (141)
where
c
11;k
=
1
2
2v
T
1
K
T;kk
v
+
1;k
v
T
1
K
T;kkk
v
1
v
T
1
K
T;k
v
1
; (142)
resulting from derivation of (D.13) with respect to k, consideration of (D.11), and specialization of the re-
sult for v
+
1
= v
1
, and
c
1j;k
=
1
k
+
1
k
+
j
v
+T
j

K
T;kk
v
1
v
+T
j

K
T;k
v
+
j
; j = 2; 3; . . . ; n; (143)
following from derivation of (D.10) with respect to k and specialization of the result for k
+
1
= k and v
+
1
= v
1
.
[k
+
j
k is the jth eigenvalue and v
+
j
is the corresponding eigenvector of the consistently linearized eigenprob-
lem (see Appendix D).]
Substitution of (138) into (142) yields
c
11;k
=
1
2
2(v
T
1
K
T;kk
v
1
)c
11
v
T
1
K
T;kkk
v
1
v
T
1
K
T;k
v
1
= 2a
2
1
3a
+
1
; (144)
where use of (16), (23.1), and (139) was made. Substitution of (138) with (139), and of (144) into (141) gives
v
+
1;kk
= 3(a
2
1
a
+
1
)v
1

n
j=2
c
1j;k
v
+
j
. (145)
Following from (59) and (60), respectively, for symmetric bifurcation from non-linear prebuckling paths,
(145) disintegrates either into
v
+
1;kk
= 3(a
2
1
a
+
1
)v
1
. c
1j;k
= 0; j = 2; 3; . . . ; n; (146)
or into
v
+
1;kk
=
n
j=2
c
1j;k
v
+
j
. a
2
1
a
+
1
= 0. (147)
For the special case of k
2
= 0 within the framework of symmetric bifurcation from non-linear prebuckling
paths (points T in Fig. 4(a)(h)), d
1
= 0 (see (64)). For point T in
Fig. 4(a): v
+
1;k
= a
1
v
1
; v
+
1;kk
= 3(a
2
1
a
+
1
)v
1
;
(146.2)
c
1j;k
= 0; j = 2; 3; . . . ; n; (148)
Fig. 4(b):

K
T;kk
v
1
= 0
(16)
a
1
= 0; (149)
(138),(139)
v
+
1;k
= 0;
(146.1)
v
+
1;kk
= 3a
+
1
v
1
;
(146.2)
c
1j;k
= 0; j = 2; 3; . . . ; n; (150)
Fig. 4(c):

K
T;kk
v
1
= 0
(16)
a
1
= 0
(147.2)
a
+
1
= 0; (151)
(138),(139)
v
+
1;k
= 0;
(147.1)
v
+
1;kk
= 0;
(147.1)
c
1j;k
= 0; j = 2; 3; . . . ; n; (152)
Fig. 4(d): v
T
1
K
T;kk
v
1
= 0
(16)
a
1
= 0
(147.2)
a
+
1
= 0; (153)
(138),(139)
v
+
1;k
= 0;
(147.1)
v
+
1;kk
=
n
j=2
c
1j;k
v
+
j
; (154)
Fig. 4(e): same as in Fig. 4(c);
Fig. 4(f): same as in Fig. 4(a);
Fig. 4(g): same as in Fig. 4(b) or (c);
Fig. 4(h): same as in Fig. 4(d).
Hence, for the special case of k
2
= 0 within the framework of symmetric bifurcation, d
1
= 0, and v
+
1;k
is
either parallel to v
1
(see (148.1)) or zero (see (150.1), (152.1) and (154.1)). If, for this special case, v
+
1;k
is par-
allel to v
1
, then also v
+
1;kk
is parallel to v
1
(see (148.2)). If v
+
1;k
is zero, then v
+
1;kk
is either parallel to v
1
(see
(150.2)) or orthogonal to v
1
with respect to

K
T;k
(see (154.2) and (D.7.2)), or zero (see (152.2)). Apart from
one exception, the following disjunction holds:
c
1j;k
= c
1j;k
(j) = 0; j = 2; 3; . . . ; n . a
2
1
a
+
1
= a
2
1
(j) a
+
1
(j) = 0. (155)
Eqs. (148) and (150) represent a complete subset of solutions for v
+
1;k
and v
+
1;kk
associated with k
2
= 0. Eqs.
(148.2) and (150.2) follow from Eq. (146) which meet the disjunction (155). Eqs. (152) and (154) represent
another complete subset of solutions for v
+
1;k
and v
+
1;kk
associated with k
2
= 0. Eq. (154.2) follows from Eqs.
(147) which also meet the disjunction (155). Because of (151.1),
c
1j;k
(k
2
= 0) = 0; j = 2; 3; . . . ; n. (156)
Hence, for the respective value of j the disjunction expressed by (155) exceptionally becomes a conjunction.
It is met by Eq. (152.2) which follows from Eq. (147). For the situation illustrated in Fig. 4(g), for the case
of the cylindrical panel with an elastic spring, this conjunction is not restricted to a specic value of j. The
sum of the two complete subsets of solutions for v
+
1;k
and v
+
1;kk
associated with k
2
= 0 represents the complete
set of such solutions. Fig. 2 illustrates the preceding remarks.
For symmetric bifurcation from linear prebuckling paths, the relations (127) and (128) are satised triv-
ially. Hence, the preceding considerations are irrelevant. There is no condition for k
2
= 0 in addition to
d
1
= 0 (see (132)).
8. Conclusions
Conversion from imperfection-sensitive into imperfection-insensitive structures requires symmetric
bifurcation. If this condition is not satised by the original structure and for the given loading, it must
be enforced in the course of the conversion process. This may require modications of the original design
which, for dierent reasons, are unfeasible.
Symmetric bifurcation from non-linear prebuckling paths is associated either with
v
+T
j

K
T;kk
v
1
= 0; j = 2; 3; . . . ; n (see (59))
or with
3 v
T
1
K
T;kk
v
1
_ _
2
2 v
T
1
K
T;k
v
1
_ _
v
T
1
K
T;kkk
v
1
_ _
= 0 (see (60)).
Eq. (59) may but need not occur together with c
+
1
(j) = 0. If it does,
k
2k
(j) = 0; k = 2; 3; . . .
Eq. (60) only occurs together with c
+
1
(j) = 0 Eqs. (48) and (51). Eqs. (59) and (60) result in two dif-
ferent disintegrations of an expression that holds for unsymmetric bifurcation from non-linear prebuck-
ling paths (see (145)(147)).
The geometric loci of all points in the k
4
a
1
plane of the k
2
k
4
a
1
space, which are solutions of
k
4
= a
1
k
2
2
b
2
k
2
d
3
(see (36))
with k
2
= 0, are restricted to the two half-axes k
4
6 0 and a
1
6 0 (see Fig. 2). This restriction is a con-
sequence of the condition for symmetric bifurcation,
k
1
= k
3
= = 0 (see (27));
which is stronger than the conditions k
1
= 0 and k
3
= 0 on which the above expression for k
4
is based.
The relations (66) and (67), respectively, refer to modes of conversion from imperfection-sensitive into
imperfection-insensitive structures, which are characterized by
k
2
= k
4
= k
6
= = 0 (see (68)).
The relations (70) refer to a mode of conversion from imperfection-sensitive into imperfection-insensitive
structures, which is characterized by
k
2
= 0; k
4
< 0 (see (71)).
k
2
= 0 is a necessary but not a sucient condition for the transition from imperfection sensitivity to
imperfection insensitivity. For
k
2
= 0; k
4
= 0; k
6
< 0;
such a transition does not occur (see point T in Fig. 4(c)). This situation is characterized by
K
T;kk
v
1
= 0 and v
T
1
K
T;kkk
v
1
= 0 (see (78.2) and (78.3)).
Hilltop bifurcation is characterized by
k
2
< 0; k
4
< 0; a
1
= (see (98) and (102); respectively).
The transition from bifurcation buckling to no loss of stability is characterized by
k
;n
= 0; k
;nn
= 0; v
1
= 0 (see (106));
indicating the existence of saddle points on the primary paths and the degeneration of the secondary
paths to these points, respectively (see point C in Fig. 5(a)).
Alternatively,
~u
;nn
= 0; k
;nn
= 0; (see (126));
indicating the existence of points of inection on the primary paths and the degeneration of the second-
ary paths to these points, respectively (see point C* in Fig. 5(b) and (c)).
For the special case of linear prebuckling paths,
K
T;kk
= 0;

K
T;kkk
= 0; . . . (see (129)).
Hence, (59) and (60) are satised trivially.
For this special case, in contrast to the general case of symmetric bifurcation from non-linear prebuck-
ling paths, k
2
= 0 may also occur jointly with k
4
> 0 (see Fig. 6(b)).
To each point on a space curve k
2
= k
2
(j), k
4
= k
4
(j), a
1
= a
1
(j) (see Fig. 4) a curve k
+
1
(k) can be related,
which is part of the solution of the consistently linearized eigenproblem (see Appendix D). It was shown
that the curves k
+
1
(k) related to point T(k
2
= 0, k
4
, a
1
) in Fig. 4 (with the exception of Fig. 4(a) and (f))
have specic geometric properties at the bifurcation point k
+
1
= k.
The investigation of these properties was motivated by the need to ensure the completeness of the solu-
tions for the initial postbuckling paths with k
2
= 0 for the general case of symmetric bifurcation from
non-linear prebuckling paths (see (148)(154)).
Reducing the initial rise of an imperfection-sensitive structure eventually results in the transition from
bifurcation buckling to no loss of stability. However, such a reduction is associated with a decrease of
the stability limit. Increasing the stiness of a structure by means of a uniform increase of its thickness does
not result in the conversion from imperfection sensitivity into insensitivity. Increasing the stiness of an
elastic spring, suitably attached to the structure, however, enables its conversion from an imperfection-sen-
sitive into an imperfection-insensitive structure. Based on these conclusions from Part II of this work [12],
it seems that additional supports of a structure are eective means to achieve the desired conversion.
A challenge for future scientic work is to investigate the eectiveness of dierent modes of additional
support of the original structure, which are feasible from the design standpoint, to accomplish such a
conversion.
Acknowledgments
The authors are indebted to B. Krenn for many helpful remarks and particularly for informing them
about the existence of a case of conversion from an imperfection-sensitive into an imperfection-insensitive
structure (see Fig. 4(a)), which they have missed originally. The second author thankfully acknowledges
partial nancial support by the Austrian Science Fund under the contract P14808.
Appendix A. Coecient tensors for Koiters postbuckling analysis in the context of the FEM
Because of successive application of the chain rule, the expressions for some of the coecient tensors in
(4) become relatively lengthy. Introduction of special tensor-valued functions and of a rule for dierentia-
tion, which combines partial and directional derivatives, allows to write these expressions in comparatively
compact form.
In the standard FEM, G,
u
is referred to as the tangent stiness matrix K
T
(u). For proportional loading,
G,
u
does not explicitly depend on k. Nevertheless, a matrix curve
K
T
(k) := K
T
(~u(k)) = G
;u
(~u(k)); k J R; (A:1)
may be dened along the equilibrium path u = ~u(k). This matrix function is identical with the tangent sti-
ness matrix K
T
(k) in papers by Helnwein [5], Helnwein and Mang [6], and Helnwein et al. [7]. In the present
paper,

K
T
(k) indicates equilibrium states on the primary path whereas K
T
(u) refers to congurations which,
in general, represent out-of-balance states. The main objective of introducing the above denition of

K
T
(k)
is to increase the compactness of the expressions for the coecient tensors in (4).
The tangent, curvature, and higher-order derivatives of the matrix curve (A.1) along the equilibrium
path are computed as follows:
K
T;k
(k) = G
;uu
~u
;k
; (A:2)
K
T;kk
(k) = G
;uuu
: ~u
;k
~u
;k
G
;uu
~u
;kk
; (A:3)
K
T;kkk
(k) = G
;uuuu
.
.
.
~u
;k
~u
;k
~u
;k
3G
;uuu
: ~u
;k
~u
;kk
G
;uu
~u
;kkk
; (A:4)
.
.
.
.
.
.
The chosen notation emphasizes that these relations only hold for points located on the primary path ~u(k).
To increase the compactness of the notation, a special rule for dierentiation of derivatives of a tensor-
valued function A(u, k) with respect to the load parameter k is introduced:
A(u; k) [ [
;k
:=
o
ok
A(u; k)
d
da
A(u a~u
;k
(k); k)[
a=0
. (A:5)
Applying (A.5) to the tangent stiness matrix K
T
(u), yields
K
T;k
(u; k) = K
T
(u) [ [
;k
= G
;u
(u) [ [
;k
=
o
ok
G
;u
(u)
d
da
a=0
G
;u
(u a~u
;k
(k)) = G
;uu
(u) ~u
;k
(k). (A:6)
Specialization of (A.6) for points on the primary path gives K
T;k
(~u(k); k) =

K
T;k
(k). The partial derivative
oK
T
(u)/ok = oG,
u
(u)/ok vanishes because, as mentioned previously, G,
u
does not explicitly depend on k.
Nevertheless, dierentiation according to (A.5) generates functions which are dened in the whole domain
of (u, k). In contrast to these functions, the ones according to the Eqs. (A.2)(A.4) are only dened along the
primary path ~u(k).
As an example, K
T
,
ukk
(u,k) = [K
T
,
uk
(u)],
k
will be computed in the following:
K
T;ukk
(u; k) = [K
T;uk
(u)[
;k
= G
;uuu
(u) ~u
;k
(k) [ [
;k
=
o
ok
(G
;uuu
~u
;k
)
d
da
a=0
G
;uuu
(u a~u
;k
(k)) ~u
;k
[ [
= G
;uuu
~u
;kk
G
;uuuu
: ~u
;k
~u
;k
. (A:7)
Comparison of this result with the one obtained from dierentiation of (A.3) with respect to u yields the
symmetry relation K
T
,
ukk
= K
T
,
kku
. Analogous symmetry relations hold for all other mixed derivatives.
Further simplications follow from the function
~
G(k): J R
n
as:
G(k) := G(~u(k); k) = 0; (A:8)
G
;k
= G
;u
~u
;k
G
;k
=

K
T
~u
;k
P = 0; (A:9)
G
;kk
= G
;uu
: ~u
;k
~u
;k
G
;u
~u
;kk
= 0; (A:10)
G
;kkk
= G
;uuu
.
.
.
~u
;k
~u
;k
~u
;k
3G
;uu
: ~u
;k
~u
;kk
G
;u
~u
;kkk
= 0; (A:11)
.
.
.
.
.
.
Eq. (A.8) expresses a trivial identity which directly results from the denition of the primary path ~u(k). Eqs.
(A.9), (A.10), (A.11), . . . allow successive computation of the vectors ~u
;k
, ~u
;kk
, ~u
;kkk
, . . . Because of
det G
;u
[
u
C
= det

K
T
[
k
C
= 0, the evaluation of these vectors at a bifurcation point C requires use of their lim-
its, as k k
C
. Moreover, the above relations cause the vanishing of the vectors G
;g
[
C
; G
gg
[
C
; G
ggg
[
C
; . . .
Computation of the rst-order coecient tensors (vectors) G
;g
; G
;gg
; G
;ggg
; . . . ; second-order tensors
(matrices) G
;v
; G
;vg
; G
;vgg
; . . . ; third-order tensors G
;vv
; G
;vvg
; . . . ; fourth-order tensors G
;vvv
; . . . ; etc., appear-
ing in (4), follows a simple pattern. All of them depend on v and g and must be expressed in terms of G(u, k).
Let A
(v; g) := A(~u(
~
k(g)) v;
~
k(g)) be one of these tensor-valued functions. The dierential of this function
can be written as
dA
(v; g) = A
;v
(v; g) dv A
;g
(v; g) dg (A:12)
= dA(~u(
~
k(g)) v;
~
k(g))
= A
;u
dv A
;u
~u
;k
A
;k
[ [
~
k
;g
(g) dg. (A:13)
Comparison of Eqs. (A.12) and (A.13) yields
A
;v
(v; g) = A
;u
[
(u;k)=(~u(
~
k(g))v;
~
k(g))
(A:14)
A
;g
(v; g) = [(A
;u
~u
;k
A
;k
)
~
k
;g
[[
(u;k)=(~u(
~
k(g))v;
~
k(g))
. (A:15)
Eventually, the coecient tensors in (4) are obtained by evaluating the expressions for these relations at the
bifurcation point C: (v, g) = (0, 0) (or (u; k) = (~u(k
C
); k
C
)).
Table 2 contains the coecient tensors as occurring in (4), evaluated at C. Note that G
;uk
=

K
T;k
=
G
;ku
= P
;u
= 0.
Appendix B. Coecient vectors of g
1
, g
2
, g
3
, g
4
, g
5
, and g
6
Eq. (7) can formally be expressed as
G
(v
(g); g) = g
1
a g
2
b g
3
c g
4
d g
5
e g
6
f O(g
7
) = 0; (B:1)
where the coecients a, b, . . . , f represent vector-valued expressions. The expressions for a, b, and c are con-
tained in (7). In order to write the comparatively lengthy expressions for d, e, and f, not contained in (7),
more concisely, the abbreviations
A : u v Auv; B
.
.
.
u v w Buvw; . . . ; u; v; w R
n
; (B:2)
will be used in what follows. It is noteworthy that all of these expressions result in vectors in R
n
. To obtain
the expressions for d, e, and f, the symmetry relations
K
T;u
: v w = G
;uu
: v w = G
;uu
: w v = K
T;u
: w v (B:3)
for arbitrary vectors v; w R
n
, must be used.
Table 2
Coecient tensors for Koiters postbuckling analysis in the context of the FEM, evaluated at the bifurcation point C
G
;v
[
C
= (G
;u
)[
C
=

K
T
(k
C
)
G
;g
[
C
= (G
;u
~u
;k
G
;k
)[
C
~
k
;g
=
(A.9)
k
1
G
;k
= 0
G
;vv
[
C
= (G
;uu
)[
C
= K
T;u
(u
C
)
G
;vg
[
C
= (G
;uu
~u
;k
~
k
;g
)[
C
=
(A.6)
k
1
K
T;k
(u
C
; k
C
) =
(A.2)
k
1
K
T;k
(k
C
)
G
;gg
[
C
= (G
;uu
: ~u
;k
~u
;k
G
;u
~u
;kk
)[
C
(
~
k
;g
)
2
(G
;u
~u
;k
G
;k
)[
C
~
k
;gg
= (k
1
)
2
G
;kk
2k
2
G
;k
=
(A.9); (A.10)
0
G
;vvv
[
C
= (G
;uuu
)[
C
= K
T;uu
(u
C
)
G
;vvg
[
C
= (G
;uuu
~u
;k
~
k
;g
)[
C
= k
1
K
T;uu
(u
C
) ~u
;k
(k
C
) = k
1
K
T;uk
u
C
; k
C
)
G
;vgg
[
C
= (
~
k
2
;g
K
T;kk

~
k
;gg
K
T;k
)[
C
= (k
1
)
2
K
T;kk
(k
C
) 2k
2
K
T;k
(k
C
)
G
;ggg
[
C
= 0
G
;vvvv
[
C
= (G
;uuuu
)[
C
= K
T;uuu
(u
C
)
G
;vvvg
[
C
= (G
;uuuu
~u
;k
~
k
;g
[
C
= k
1
K
T;uuk
(u
C
; k
C
)
G
;vvgg
[
C
= [(G
;uuuu
: ~u
;k
~u
;k
G
;uuu
~u
;kk
)(
~
k
;g
)
2
(G
;uuu
~u
;k
~
k
;gg
)[[
C
= (k
1
)
2
K
T;ukk
(u
C
; k
C
) 2k
2
K
T;uk
u
C
; k
C
)
G
;vggg
[
C
= (
~
k
;g
)
3
K
T;kkk
(k
C
) 3
~
k
;g
~
k
;gg
K
T;kk
(k
C
)
~
k
;ggg
K
T;k
(k
C
)
G
;gggg
[
C
= 0
Making use of Table 2 and of Eqs. (4) and (5), the coecients d, e, and f are obtained as
d = k
3
1
1
6
K
T;kkk
v
1
k
2
1
1
2
K
T;kk
v
2

1
4
K
T;ukk
v
1
v
1
_ _
k
1
k
2
K
T;kk
v
1

K
T;k
v
3
K
T;uk
v
1
v
2

1
6
K
T;uuk
v
1
v
1
v
1
_ _
k
2

K
T;k
v
2

1
2
K
T;uk
v
1
v
1
_ _
k
3
K
T;k
v
1

K
T
v
4
K
T;u
v
1
v
3
1
2
K
T;u
v
2
v
2

1
2
K
T;uu
v
1
v
1
v
2

1
24
K
T;uuu
v
1
v
1
v
1
v
1
; (B:4)
e = k
4
1
1
24
K
T;kkkk
v
1
k
3
1
1
6
K
T;kkk
v
2

1
12
K
T;ukkk
v
1
v
1
_ _
k
2
1
k
2
1
2
K
T;kkk
v
1

1
2
K
T;kk
v
3

1
2
K
T;ukk
v
1
v
2

1
12
K
T;uukk
v
1
v
1
v
1
_ _
k
1
k
3
K
T;kk
v
1
k
2

K
T;kk
v
2

1
2
K
T;ukk
v
1
v
1
_ _

K
T;k
v
4
K
T;uk
v
1
v
3
_
1
2
K
T;uk
v
2
v
2

1
2
K
T;uuk
v
1
v
1
v
2

1
24
K
T;uuuk
v
1
v
1
v
1
v
1
_
k
2
2
1
2
K
T;kk
v
1
k
2

K
T;k
v
3
K
T;uk
v
1
v
2

1
6
K
T;uuk
v
1
v
1
v
1
_ _
k
3

K
T;k
v
2

1
2
K
T;uk
v
1
v
1
_ _
k
4
K
T;k
v
1

K
T
v
5
K
T;u
v
1
v
4
K
T;u
v
2
v
3

1
2
K
T;uu
v
1
v
1
v
3

1
2
K
T;uu
v
1
v
2
v
2
1
6
K
T;uuu
v
1
v
1
v
1
v
2

1
120
K
T;uuuu
v
1
v
1
v
1
v
1
v
1
(B:5)
and
f = k
5
1
1
120
K
T;kkkkk
v
1
k
4
1
1
24
K
T;kkkk
v
2

1
48
K
T;ukkkk
v
1
v
1
_ _
k
3
1
k
2
1
6
K
T;kkkk
v
1

1
6
K
T;kkk
v
3

1
6
K
T;ukkk
v
1
v
2

1
36
K
T;uukkk
v
1
v
1
v
1
_ _
k
2
1
k
3
1
2
K
T;kkk
v
1
k
2
1
2
K
T;kkk
v
2

1
4
K
T;ukkk
v
1
v
1
_ _
1
2
K
T;kk
v
4

1
2
K
T;ukk
v
1
v
3

1
4
K
T;ukk
v
2
v
2
_
1
4
K
T;uukk
v
1
v
1
v
2

1
48
K
T;uuukk
v
1
v
1
v
1
v
1
_
k
1
k
4
K
T;kk
v
1
k
3

K
T;kk
v
2

1
2
K
T;ukk
v
1
v
1
_ _
k
2
2
1
2
K
T;kkk
v
1
_
k
2

K
T;kk
v
3
K
T;ukk
v
1
v
2

1
6
K
T;uukk
v
1
v
1
v
1
_ _

K
T;k
v
5
K
T;uk
v
1
v
4
K
T;uk
v
2
v
3

1
2
K
T;uuk
v
1
v
1
v
3
1
2
K
T;uuk
v
1
v
2
v
2

1
6
K
T;uuuk
v
1
v
1
v
1
v
2

1
120
K
T;uuuuk
v
1
v
1
v
1
v
1
v
1
_
k
2
2
1
2
K
T;kk
v
2

1
4
K
T;ukk
v
1
v
1
_ _
k
2

K
T;k
v
4
K
T;uk
v
1
v
3

1
2
K
T;uk
v
2
v
2

1
2
K
T;uuk
v
1
v
1
v
2

1
24
K
T;uuuk
v
1
v
1
v
1
v
1
_ _
k
3
k
2
K
T;kk
v
1

K
T;k
v
3
K
T;uk
v
1
v
2

1
6
K
T;uuk
v
1
v
1
v
1
_ _
k
4

K
T;k
v
2

1
2
K
T;uk
v
1
v
1
_ _
k
5
K
T;k
v
1

K
T
v
6
K
T;u
v
1
v
5
K
T;u
v
2
v
4

1
2
K
T;u
v
3
v
3

1
2
K
T;uu
v
1
v
1
v
4
K
T;uu
v
1
v
2
v
3

1
6
K
T;uu
v
2
v
2
v
2
1
6
K
T;uuu
v
1
v
1
v
1
v
3

1
4
K
T;uuu
v
1
v
1
v
2
v
2

1
24
K
T;uuuu
v
1
v
1
v
1
v
1
v
2

1
720
K
T;uuuuu
v
1
v
1
v
1
v
1
v
1
v
1
. (B:6)
Appendix C. Coecients c
+
1
,
^
f
1
, ~e
1
and b
2
, d
3
, b
4
For the general case of non-linear prebuckling paths, the coecients c
+
1
,
^
f
1
, and ~e
1
, occurring in Eqs.
(20)(22), respectively, are needed. Abbreviations according to (B.2) are used. Premultiplying the coecient
of k
1
in (B.4) by v
T
1
=v
T
1
K
T;k
v
1
, yields
c
+
1
= 2a
1
k
2
b
2
(C:1)
with a
1
according to (16) and
b
2
=
1
v
T
1
K
T;k
v
1
v
T
1
K
T;k
v
3
v
T
1
K
T;uk
v
1
v
2

1
6
v
T
1
K
T;uuk
v
1
v
1
v
1
_ _
. (C:2)
Premultiplying those terms in (B.5) by v
T
1
=v
T
1
K
T;k
v
1
, which do not contain k
1
, and considering (8), gives
^
f
1
k
4
, where
^
f
1
= b
1
k
3
a
1
k
2
2
b
2
k
2
d
3
(C:3)
with a
1
, b
1
, and b
2
according to Eqs. (16), (17), and (C.2), respectively, and
d
3
=
1
v
T
1
K
T;k
v
1
v
T
1
K
T;u
v
1
v
4
v
T
1
K
T;u
v
2
v
3

1
120
v
T
1
K
T;uuuu
v
1
v
1
v
1
v
1
v
1

1
2
v
T
1
K
T;uu
v
1
v
1
v
3
_
1
2
v
T
1
K
T;uu
v
1
v
2
v
2

1
6
v
T
1
K
T;uuu
v
1
v
1
v
1
v
2
_
. (C:4)
Premultiplying the coecient of k
1
in (B.6) by v
T
1
=v
T
1
K
T;k
v
1
, yields
~e
1
= 2a
1
k
4
2b
+
1
k
3
b
4
(C:5)
with a
1
and b
+
1
according to (16) and (24), respectively, and
b
4
= 3a
+
1
k
2
2
b
+
2
k
2
c
+
2
(C:6)
with a
+
1
according to (23.1),
b
+
2
=
1
v
T
1
K
T;k
v
1
v
T
1
K
T;kk
v
3
v
T
1
K
T;ukk
v
1
v
2

1
6
v
T
1
K
T;uukk
v
1
v
1
v
1
_ _
(C:7)
and
c
+
2
=
1
v
T
1
K
T;k
v
1
v
T
1
K
T;k
v
5
v
T
1
K
T;uk
v
1
v
4
v
T
1
K
T;uk
v
2
v
3

1
2
v
T
1
K
T;uuk
v
1
v
1
v
3

1
2
v
T
1
K
T;uuk
v
1
v
2
v
2
_
1
6
v
T
1
K
T;uuuk
v
1
v
1
v
1
v
2

1
120
v
T
1
K
T;uuuuk
v
1
v
1
v
1
v
1
v
1
_
. (C:8)
Appendix D. Mathematical properties of the consistently linearized eigenproblem
To each point on a space curve k
2
= k
2
(j), k
4
= k
4
(j), a
1
= a
1
(j) (see Fig. 4) a curve k
+
1
(k) can be related,
which is part of the solution of the so-called consistently linearized eigenproblem. The mathematical for-
mulation of this eigenproblem reads [5]
[
K
T
(k
+
k)
K
T;k
[v
+
= 0; (D:1)
where k* k is the eigenvalue and v* is the eigenvector. For v
+
= v
+
1
= v
1
, because of (8),
k
+
= k
+
1
= k; (D:2)
representing the load level at the stability limit.
It will be shown that, in general, the curves k
+
1
(k) related to point T(k
2
= 0, k
4
, a
1
) in Fig. 4 have specic
geometric properties at the bifurcation point k
+
1
= k. The investigation of these properties is motivated by
the need to ensure the completeness of the solutions for the initial postbuckling paths with k
2
= 0 for the
general case of symmetric bifurcation from non-linear primary paths (see Section 7). Moreover, these prop-
erties permit verication of theoretical results for limiting cases by inspection of the corresponding curves
k
+
1
(k). In view of the complexity of some of the relevant mathematical expressions, no practical alternative is
available.
Derivation of (D.1) with respect to k gives
[k
+
;k
K
T;k
(k
+
k)
K
T;kk
[v
+
[
K
T
(k
+
k)
K
T;k
[v
+
;k
= 0. (D:3)
Writing (D.3) for the rst eigenpair, which is a function of k, yields
[k
+
1;k
K
T;k
(k
+
1
k)
K
T;kk
[v
+
1
[
K
T
(k
+
1
k)
K
T;k
[v
+
1;k
= 0. (D:4)
Premultiplication of (D.4) by v
+T
1
and use of (D.1) gives
k
+
1;k
= (k
+
1
k)
v
+T
1
K
T;kk
v
+
1
v
+T
1
K
T;k
v
+
1
. (D:5)
Expressing v
+
1;k
in terms of the eigenvectors v
+
j
, j = 1, 2, . . . , n, results in
v
+
1;k
=
n
j=1
c
1j
v
+
j
. (D:6)
Inserting (D.6) into (D.4), premultiplying the obtained relation by v
+T
j
, j 51, and making use of the orthog-
onality conditions
v
+T
j

K
T
v
+
1
= 0; v
+T
j

K
T;k
v
+
1
= 0; (D:7)
following from (D.1), gives
(k
+
1
k)v
+T
j

K
T;kk
v
+
1
v
+T
j

K
T
(k
+
1
k)
K
T;k
_ _
c
1j
v
+
j
= 0. (D:8)
Writing (D.1) for the jth eigenpair, which is a function of k, and premultiplying the obtained relation by v
+T
j
,
yields
v
+T
j
[
K
T
(k
+
j
k)
K
T;k
[v
+
j
= 0. (D:9)
Insertion of (D.9) into (D.8) results in
c
1j
=
k
+
1
k
k
+
1
k
+
j
v
+T
j

K
T;kk
v
+
1
v
+T
j

K
T;k
v
+
j
. (D:10)
In order to determine c
11
, v
+
1
is normalized such that
v
+T
1
K
T;k
v
+
1
= 1 . 1; (D:11)
implying k
+
1
(k = 0) > 0, which can always be achieved by means of a suitable denition of a positive refer-
ence load. Derivation of (D.11) with respect to k gives
2v
+T
1
K
T;k
v
+
1;k
v
+T
1
K
T;kk
v
+
1
= 0. (D:12)
Substitution of (D.6) into (D.12) and consideration of (D.7.2) results in
c
11
=
1
2
v
+T
1
K
T;kk
v
+
1
v
+T
1
K
T;k
v
+
1
. (D:13)
Specializing (D.13) for the stability limit by setting v
+
1
= v
1
and comparing the relation for c
11
with (16), it is
seen that
c
11
= a
1
. (D:14)
Apart from exceptional cases, which will be treated later, specialization of (D.5) and (D.10) for the stability
limit k
+
1
= k yields
k
+
1;k
= 0 (D:15)
and
c
1j
= 0; j = 2; 3; . . . ; n; (D:16)
respectively. Substituting (D.16) into (D.6), gives
v
+
1;k
= c
11
v
+
1
with v
+
1
= v
1
. (D:17)
In order to show that the non-linearity coecient a
1
is proportional to the curvature of the eigenvalue curve
at the bifurcation point, (D.3) is dierentiated with respect to k:
[k
+
;kk
K
T;k
(2k
+
;k
1)
K
T;kk
(k
+
k)
K
T;kkk
[v
+
2[k
+
;k
K
T;k
(k
+
k)
K
T;kk
[v
+
;k
[
K
T
(k
+
k)
K
T;k
[v
+
;kk
= 0. (D:18)
Writing (D.18) for the rst eigenpair and specializing the obtained relation for the bifurcation point by
inserting Eqs. (D.2) and (D.15) and setting v
+
1
= v
1
, gives
k
+
1;kk
K
T;k

K
T;kk
_ _
v
1

K
T
v
+
1;kk
= 0. (D:19)
T
1
and consideration of (8) yields
k
+
1;kk
=
v
T
1
K
T;kk
v
1
v
T
1
K
T;k
v
1
. (D:20)
Comparison of (D.20) with (16) shows that
a
1
=
1
2
k
+
1;kk
k
+
1
=k
; (D:21)
which proves the correctness of the preceding assertion.
Specializing (D.21) for a
1
= 0 and (D.17) for c
11
= a
1
= 0, gives
k
+
1;kk
= 0 (D:22)
and
v
+
1;k
; = 0; (D:23)
respectively. Hence, point I in Fig. 4(a), point T in Fig. 4(b)(e) and (g), and points T in Fig. 4(h) correlate
with special points on the corresponding curves k
+
1
(k) and v
+
1
(k) at the bifurcation point k
+
1
= k.
Point I in Fig. 4(a). Substitution of (148.2) into (D.19) and consideration of (8) gives
k
+
1;kk
K
T;k

K
T;kk
_ _
v
1
= 0. (D:24)
At point I,
K
T;kk
v
1
= 0
(16)
a
1
= 0. (D:25)
Following from (D.24) and (D.25),
k
+
1;kk
= 0. (D:26)
In Part II of this work [12] it is shown numerically that the curve k
+
1
(k) related to point I in Fig. 4(a) has a
saddle point at the bifurcation point k
+
1
= k.
Point T in Fig. 4(b). At this point, (D.24) and (D.25) (which is equal to (149)) hold. Consequently,
k
+
1;kk
= 0. In Part II of this work [12] it is shown numerically that the curve k
+
1
(k) related to point T in
Fig. 4(b) has a saddle point at the bifurcation point k
+
1
= k. The reason for this correspondence is (46). This
relation does not hold for the situation illustrated in Fig. 4(a).
Point T in Fig. 4(c). At this point, (D.24) and (D.25) (which is equal to (151)) hold. Consequently,
k
+
1;kk
= 0. However, in contrast to the situation at point T in Fig. 4(b), the curve k
+
1
(k) related to point T
in Fig. 4(c) has a planar point at the bifurcation point k
+
1
= k, k
+
1;k
= 0. Hence, in addition to k
+
1;kk
= 0, also
k
+
1;kkk
= 0. This assertion is based on the following observation: the curvature of the curve k
+
1
(k) related to an
arbitrary point on the space curve in Fig. 4(c), has a maximum value at the bifurcation point k
+
1
= k,
k
+
1;k
= 0. Hence,
k
+
1;kk
[1 (k
+
1;k
)
2
[
3
2
_ _
;k
k
+
1;k
=0
= 0;
k
+
1;kk
[1 (k
+
1;k
)
2
[
3
2
_ _
;kk
k
+
1;k
=0
< 0. (D:27)
From (D.27.1),
k
+
1;kkk
[1 (k
+
1;k
)
2
[
3
2
3(k
+
1;kk
)
2
k
+
1;k
[1 (k
+
1;k
)
2
[
1
2
[1 (k
+
1;k
)
2
[
3
k
+
1;k
=0
= k
+
1;kkk
= 0. (D:28)
Derivation of (D.18) with respect to k results in
k
+
;kkk
K
T;k
3k
+
;kk
K
T;kk
(3k
+
;k
2)
K
T;kkk
(k
+
k)
K
T;kkkk
_ _
v
+
3 k
+
;kk
K
T;k
(2k
+
;k
1)
K
T;kk
(k
+
k)
K
T;kkk
_ _
v
+
;k
3 k
+
;k
K
T;k
(k
+
k)
K
T;kk
_ _
v
+
;kk

K
T
(k
+
k)
K
T;k
_ _
v
+
;kkk
= 0. (D:29)
Writing (D.29) for the rst eigenpair and specializing the result for the bifurcation point, i.e. for k
+
1
= k,
k
+
1;k
= 0, k
+
1;kkk
= 0, and v
+
1
= v
1
, gives
3k
+
1;kk
K
T;kk
2
K
T;kkk
_ _
v
1
3 k
+
1;kk
K
T;k

K
T;kk
_ _
v
+
1;k

K
T
v
+
1;kkk
= 0. (D:30)
T
1
and use of (8), (16), (23.1), (138), (139), and (D.21) results in a
2
1
a
+
1
= 0
(see (61)).
At point T in Fig. 4(c), (D.19) disintegrates into
k
+
1;kk
= 0;

K
T;kk
v
1
= 0
(16)
a
1
= 0 (see (151)); v
+
1;kk
= 0 (see (152.2)). (D:31)
Hence, the only dierence between the disintegration of (D.19) at point T in Fig. 4(c) and the one at point T
in Fig. 4(b) is the vanishing of v
+
1;kk
at the former. Eq. (D.31.1) holds in addition to k
+
1
= k, k
+
1;k
= 0, and
k
+
1;kkk
= 0. Therefore, as shown numerically in Part II of this work [12], the curve k
+
1
(k) related to point
T in Fig. 4(c) has a planar point at the bifurcation point k
+
1
= k.
Inserting (D.13) into (D.17) and specializing the result for (D.31.2), gives
v
+
1;k
= 0. (D:32)
Substitution of (D.31.1) and (D.32) into (D.30) results in
2
K
T;kkk
v
1

K
T
v
+
1;kkk
= 0. (D:33)
T
1
and consideration of (8) results in
v
T
1
K
T;kkk
v
1
= 0
(23.1)
a
+
1
= 0 (see (152.3)). (D:34)
For reasons of completeness, the situation at point H in Fig. 4(c) will be investigated in the following.
Specializing the ratio of the two quadratic forms in (D.5) for the stability limit by setting v
+
1
= v
1
and mak-
ing use of (D.20) and (D.21), gives
v
T
1
K
T;kk
v
1
v
T
1
K
T;k
v
1
= 2a
1
. (D:35)
Substituting (D.35) into (D.5), yields
k
+
1;k
[
k
+
1
=k
= 2 k
+
1
k
_ _
a
1
. (D:36)
Because of
k
+
1
k = 0 and a
1
= (see (102)); (D:37)
the expression for k
+
1;k
[
k
+
1
=k
is an indeterminate expression. (In the following, [
k
+
1
=k
will be omitted.)
Inserting (99) into (D.36), results in
k
+
1;k
=
k
+
1
k
_ _
k
;nn
(k
;n
)
2

v
T
1
K
T;nn
v
1
v
T
1
K
T;n
v
1
k
;n
k
;nn
1
_ _
. (D:38)
Application of de LHospitals rule to the indeterminate expression (k
+
1
k)=(k
;n
)
2
in (D.38) and consider-
ation of (100.2) gives
k
+
1;k
=
(k
+
1;n
k
;n
)k
;nn
2k
;n
k
;nn
=
1
2
(k
+
1;k
1); (D:39)
resulting in
k
+
1;k
= 1; (D:40)
which agrees with the numerical result for hilltop bifurcation reported in Part II of this work [12]. Substi-
tution of (102) into (D.21) yields
k
+
1;kk
= ; (D:41)
which indicates that the curve k
+
1
(k) has a singular point at k
+
1
= k.
Point T in Fig. 4(d). With exception of (D.31.2) and (D.31.3), the relations for point T in Fig. 4(c) also
hold for point T in Fig. 4(d). At this point, (D.19) disintegrates into
k
+
1;kk
= 0 and

K
T;kk
v
1

K
T
v
+
1;kk
= 0. (D:42)
Premultiplication of (D.42.2) by v
T
1
and consideration of (8) yields
v
T
1
K
T;kk
v
1
= 0
(16)
a
1
= 0. (D:43)
T
1
and consideration of (8) results in
v
T
1
K
T;kkk
v
1
= 0
(16)
a
+
1
= 0. (D:44)
Eqs. (D.43) and (D.44) are associated with k
+
1;kk
= 0 and k
+
1;kkk
= 0. Hence, as shown numerically in Part II
of this work [12], also the curve k
+
1
(k) related to point T in Fig. 4(d) has a planar point at the bifurcation
point k
+
1
= k.
Point T in Fig. 4(e). The situation at point S = T in Fig. 4(e) is the same as the one at point T in
Fig. 4(c). Therefore, the curve k
+
1
(k) related to point S = T in Fig. 4(e) has a planar point at the bifurcation
point k
+
1
= k.
Point F = N = T in Fig. 4(f). Substitution of (106) into (99) yields an indeterminate expression for a
1
.
With the help of de LHospitals rule, the result for this expression is obtained as
a
1
= . (D:45)
Because of
k
+
1
k = 0 and a
1
= ; (D:46)
k
+
1;k
[
k
+
1
=k
= 2 k
+
1
k
_ _
a
1
(see (D.36)) (D:47)
is an indeterminate expression. By means of de LHospitals rule, the result for this expression is obtained as
k
+
1;k
= 1; (D:48)
which agrees with (D.40). Moreover,
k
+
1;kk
= (D:49)
(see (D.41)).
Point T in Fig. 4(g). The situation at this point is the same as the one at point T in Fig. 4(b) (von Mises
truss) and in Fig. 4(c) (cylindrical panel), respectively. Hence, the curves k
+
1
(k) related to point T in Fig. 4(f)
have a saddle point (von Mises truss) and a planar point (cylindrical panel), respectively, at the bifurcation
point.
Fig. 7 (von Mises truss): The curve in Fig. 7(a) that contains the bifurcation point (point C) and the dash-
dotted curve in this Figure show the functions k
+
1
(k) and k
+
j
(k), both related to point T in Fig. 4(b) (von
Mises truss), which represents a limiting case (see (97)). At point C, k
+
1
= k, k
+
1;k
= 0, and k
+
1;kk
= 0. For
k > k
+
1
,

K
T
is an indenite matrix. Consequently, eigenvalue functions may become complex functions.
For k > k
R
, where k
R
refers to point R in Fig. 7(a), k
+
1
(k) and k
+
j
(k) are conjugate complex functions.
To understand the situation at point R = C in Fig. 7(b), which is associated with the nal situation
F = N = T of the aforementioned limiting case, the one at point R in Fig. 7(a) must be understood. To
understand the latter, the situation for k > k
R
in Fig. 7(a) must be investigated. For that purpose, k*
and v* in (D.1) are replaced by
k
+
1
= Re(k
+
1
) iIm(k
+
1
) and v
+
1
= Re(v
+
1
) iIm(v
+
1
); (D:50)
where Re( ) and Im( ) denote the real and the imaginary part, respectively, of the term in parentheses. This
gives
K
T
Re(k
+
1
) iIm(k
+
1
) k
_
K
T;k
Re(v
+
1
) iIm(v
+
1
) = 0. (D:51)
The real and the imaginary part of (D.51) are obtained as
K
T
[Re(k
+
1
) k[
K
T;k
Re(v
+
1
) Im(k
+
1
)
K
T;k
Im(v
+
1
) = 0 (D:52)
and
K
T
[Re(k
+
1
) k[
K
T;k
Im(v
+
1
) Im(k
+
1
)
K
T;k
Re(v
+
1
) = 0. (D:53)
Premultiplication of (D.52) by Re(v
+T
1
) and of (D.53) by Im(v
+T
1
) and addition of the so-obtained relations
yields
Re(v
+T
1
)
K
T
[Re(k
+
1
) k[
K
T;k
Re(v
+
1
) Im(v
+T
1
)

K
T
Re(k
+
1
) k
_
K
T;k
_ _
Im(v
+
1
) = 0. (D:54)
Premultiplication of (D.52) by Im(v
+T
1
) and of (D.53) by Re(v
+T
1
) and subtraction of the rst one of the so-
obtained relations from the second one results in
Re(v
+T
1
)
K
T;k
Re(v
+
1
) Im(v
+T
1
)
K
T;k
Im(v
+
1
) = 0. (D:55)
Substitution of (D.55) into (D.54) gives
Re(v
+T
1
)
K
T
Re(v
+
1
) Im(v
+T
1
)
K
T
Im(v
+
1
) = 0. (D:56)
At this point, the eigenvalue k
+
1
k is still real. Hence,
Im(k
+
1
) = 0
(D.50.1)
Re(k
+
1
) = k
+
1
. (D:57)
The eigenvalue represents a double root of the consistently linearized eigenproblem. The dash-dotted curve
in Fig. 7(a) is the eigenvalue curve that joins the eigenvalue curve, which contains point C, at point R. Since,
for k > k
R
, the two eigenvalue functions k
+
1
k and k
+
j
k are conjugate complex functions:
k
+
1
= Re(k
+
1
) iIm(k
+
1
); k
+
j
= k
+
1
= Re(k
+
1
) iIm(k
+
1
). (D:58)
(a) (b)
Fig. 7. Eigenvalue curves related to point T in Fig. 4(b) representing a limiting case, and point F = N = T (see Fig. 4(g)) representing
the nal situation of this limiting case.
Substitution of (D.57) into (D.52) and (D.53) yields
[
K
T
(k
+
1
k)
K
T;k
[Re(v
+
1
) = 0 and [
K
T
(k
+
1
k)
K
T;k
[Im(v
+
1
) = 0; (D:59)
respectively, resulting in
Im(v
+
1
) = aRe(v
+
1
); (D:60)
where a is an arbitrary scalar factor.
Substitution of (D.60) into (D.50.2) gives
v
+
1
= (1 ai)Re(v
+
1
). (D:61)
Consequently,
v
+
j
= v
+
1
= (1 ai)Re(v
+
1
). (D:62)
Inserting (D.60) into (D.56) and (D.55), yields
Re(v
+T
1
)
K
T
Re(v
+
1
) = 0 and Re(v
+T
1
)
K
T;k
Re(v
+
1
) = 0. (D:63)
Substitution of (D.61) into (D.5) results in
k
+
1;k
= (k
+
1
k)
Re(v
+T
1
)
K
T;kk
Re(v
+
1
)
Re(v
+T
1
)
K
T;k
Re(v
+
1
)
. (D:64)
Because of (D.63.2) and of
k
+
1
k ,= 0 and Re(v
+T
1
)
K
T;kk
Re(v
+
1
) ,= 0; (D:65)
k
+
1;k
[
k=k
R
= (D:66)
(see Fig. 7(a)). By analogy,
k
+
j;k
[
k=k
R
= k
+
1;k
[
k=k
R
= (D:67)
(see Fig. 7(b)).
The nal situation F = N = T of the limiting case T is characterized by the coincidence of points C and R
(see Fig. 7(b)). Hence, the eigenvalue represents a double root of value zero of the consistently linearized
eigenproblem, i.e.
k
+
1
k = k
+
j
k = 0. (D:68)
For this limiting case, Eqs. (106), (113), and (114) hold, i.e.
k
;n
= 0; k
;nn
= 0; v
1
= 0;

K
T;n
v
1;n
= 0;

K
T
d~u = 0. (D:69)
Following from (D.61), (D.62) and (D.69.3),
v
+
j
= v
j
= 0. (D:70)
As regards the eigenvalue function k
+
1
(k) k, the term limiting case means that at point C of the eigen-
value curve (see Fig. 7(a)),
k
+
1
k = 0; k
+
1;k
= 0; k
+
1;kk
= 0; (D:71)
indicating a saddle point. In contrast to the limiting case, the underlined relation in (D.71) does not hold for
the standard case. The term nal situation (of the limiting case) means that at point R = C of the eigen-
value curve (see Fig. 7(b)),
k
+
1
k = 0; k
+
1;k
= 0; k
+
1;kk
= 0; k
;k
+
1
k
+
1
k
+
1
= 0; k
;k
+
1
k
+
1
k
+
1
k
+
1
= 0. (D:72)
In contrast to this nal situation, which is associated with the transition to no buckling (see Fig. 5(a)), the
doubly underlined relations in (D.72) do not hold for the standard situation of the limiting case.
Fig. 8 (cylindrical panel): The curve in Fig. 8 that contains the bifurcation point (point C) and the dash-
dotted curve in this gure shows the functions k
+
1
(k) and k
+
j
(k), both related to point T in Fig. 4(c) (cylin-
drical panel). At point C, k
+
1
= k, k
+
1;k
= 0, k
+
1;kk
= 0, and k
+
1;kkk
= 0. For k > k
R
, where k
R
refers to point R in
Fig. 8, k
+
1
(k) and k
+
j
(k) are conjugate complex functions.
As regards the eigenvalue function k
+
1
(k) k, the term limiting case means that at point C of the eigen-
value curve (see Fig. 8),
k
+
1
k = 0; k
+
1;k
= 0; k
+
1;kk
= 0; k
+
1;kkk
= 0; (D:73)
indicating a planar point. In contrast to the limiting case, the underlined relation in (D.73) does not hold for
the standard case which diers from the one for the von Mises truss by the existence of the condition
(D.73.4). The term nal situation (of the limiting case) means that at point C,
k
+
1
k = 0; k
+
1;k
= 0; k
+
1;kk
= 0; k
+
1;kkk
= 0; k
+
1;kkkk
= 0; (D:74)
indicating saddle point of higher order. In contrast to the nal situation, which is associated with the
transition to no buckling (see Fig. 5(a)), the doubly underlined relation in (D.74) does not hold for the stan-
dard situation of the limiting case.
Points T in Fig. 4(h). The situation at points T in Fig. 4(h) is the same as the one at point T in Fig. 4(d)
(cylindrical panel). Hence, the curves k
+
1
(k) related to points T in Fig. 4(h) have a planar point at the bifur-
cation point. Consequently, Fig. 8 and (D.73) also apply to the points T in Fig. 4(h), whereas (D.74) also
applies to the nal situation F = N = T of this limiting case.
References
[1] B. Bochenek, Problems of structural optimization for post-buckling behaviour, Struct. Multidisciplinary Optim. 25 (56) (2003)
423435.
[2] B. Bochenek, J. Kru_ zelecki, A new concept of optimization for postbuckling behaviour, Engrg. Optimiz. 33 (2001) 503522.
[3] F. Fujii, Multiple hill-top branching, in: C.M. Wang et al. (Eds.), Proceedings of the 2nd International Conference on Structural
Stability and Dynamics, World Scientic, Singapore, 2002.
[4] L. Godoy, Sensitivity of post-critical states to changes in design parameters, Int. J. Solids Struct. 33 (15) (1996) 21772192.
Fig. 8. Eigenvalue curves related to point T (see Fig. 4(c)) representing a limiting case.
[5] P. Helnwein, Zur initialen Abschatzbarkeit von Stabilitatsgrenzen auf nichtlinearen Last-Verschiebungspfaden elastischer
Strukturen mittels der Methode der Finiten Elemente [On ab initio estimates of stability limits on nonlinear loaddisplacement
paths of elastic structures using the nite element method], Volume 79 of Dissertationen an der Technischen Universitat Wien,
O
sterreichischer Kunst- und Kulturverlag, Wien, 1997 (in German).

[6] P. Helnwein, H. Mang, An asymptotic approach for the evaluation of errors resulting from estimations of stability limits in
nonlinear elasticity, Acta Mech. 125 (1997) 235254.
[7] P. Helnwein, H. Mang, B. Pichler, Ab initio estimates of stability limits on nonlinear loaddisplacement paths: potential and
limitations, Comput. Assist. Mech. Engrg. Sci. 6 (3/4) (1999) 345360.
[8] W. Koiter, On the stability of elastic equilibrium, Translation of Over de Stabiliteit van het Elastisch Evenwicht (1945), in: NASA
TT F-10833, Polytechnic Institute Delft, H.J. Paris Publisher, Amsterdam, 1967.
[9] Z. Mro z, R. Haftka, Design sensitivity analysis of non-linear structures in regular and critical states, Int. J. Solids Struct. 31 (15)
(1994) 20712098.
[10] Z. Mro z, J. Piekarski, Sensitivity analysis and optimal design of non-linear structures, Int. J. Numer. Methods Engrg. 42 (1998)
12311262.
[11] R. Reitinger, Stabilitat und Optimierung imperfektionsempndlicher Tragwerke [Stability and optimization of imperfection-
sensitive structures], Dr.-Ing. dissertation, Universitat Stuttgart, Institut fu r Baustatik, 1994, Bericht Nr. 17 (in German).
[12] C. Schranz, B. Krenn, H. Mang, Conversion of imperfection sensitive to -insensitive elastic structures II: Numerical investigation,
Comput. Methods Appl. Mech. Engrg., in press, doi:10.1016/j.cma.2005.05.025.
[13] T. Tarnai, Zero stiness elastic structures, Int. J. Mech. Sci. 45 (2003) 425431.
[14] O. Zienkiewicz, R. Taylor, The Finite Element Method, fourth ed., vol. 2, McGraw-Hill, London, England, 1994.

Conversion From Imperfection-Sensitive Into Imperfection Insensitive Elastic Structures

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Conversion From Imperfection-Sensitive Into Imperfection Insensitive Elastic Structures

Caricato da

Copyright:

Formati disponibili

Conversion from imperfection-sensitive into

imperfection-insensitive elastic structures I: Theory

G(k) := G(~u(k); k) = 0; (A:8)

sterreichischer Kunst- und Kulturverlag, Wien, 1997 (in German).

Potrebbero piacerti anche