Colin B. Brown
Department of Civil Engineering,
Keywords:
static
stability
system,
stability
criterion
is~suefrom the lower and upper bound theorems of rigidplastic analysis. However, the authors have identified
the results of rigid-plastic analysis of masonry arches
with the minimization of the potential energy of a conservative system subject to kinematic constraints.
It is recognized that plastic limit analysis is connected
with constrained optimization. The theory has been
developed using duality relationships between static and
kinematic variables for objective functions which are
linear, such as the maximization of a load factor, or
quadratic, such as the minimization of complementary
energy. The constraints are linear equalities or inequalities representing compatibility in the kinematic formulation of the problem or equilibrium in the static
formulation. Where a quadratic objective function is
used, the problem is generally limited to a strictly convex objective function. A thorough review of the
literature to 1982 has been given by Maier and
Munro . More recently, Maier and Nappi I4 applied
this formulation to the prediction of the failure of a
general no-tension system, such as a masonry wall,
0141-0296/93/030189-08
@ 1993 Butterworth - Heinemann Ltd
Engng Struct.
189
General lower and upper bound theorem of static stability: T. E. Boothby and C. B. Brown
z =K(x)
(5)
Preliminaries
In this section particular primal and dual problems
which define the energy of the system and the constraints on the system will be identified. These are the
mathematical programs which are to be used subsequently to characterize the static stability of the elastic
and elastic-plastic systems and to establish bounds on the
stability of these systems.
(6)
where
Primal p r o b l e m
This paper is concerned with solutions of the following
mathematical program.
minimize %V(x)
(1)
subject to G(x) _ 0
(2)
H(x) = 0
(3)
The state variable x is a vector of dimension n representing some spatial coordinate x may be further
required to belong to an underlying space X C 6l". The
state variable is subject to kinematic constraints: the j
inequality constraints G and the k equality constraints H.
The solution of equations (1) - (3) requires the introduction of Lagrange multipliers p and r with components
pi, i = l , 2, . . . , j and ri, i = 1, 2 . . . . .
k. The
scalar-valued work function
(4)
190
-Xl
Xn ~ I
(7)
Xt=
Zl
ZI
Dual problem
Genera/lower and upper bound theorem of static stability: 7". E. Boothby and C. B. Brown
(9)
(10)
(11)
[A'qT:~x]
X = Ary
(12)
= ~aXrVx'q(8)
"Jr- I / / 2 ~ 2 ( ~ x T [ v 2 ~ ( , , ~ ) ] ( ~ X
"
"
"
(14)
(15)
(16)
Upper b o u n d theorem
The total change in %V due to an infinitesimal disturbance ~x will be written [A'W:/~x].
[x+]
_
(17)
by the n transformations
x~=[1
-1][ x[]
xi+ _> 0
(19)
x~- _< 0
(20)
x~+ x x7 = 0
(21)
0-
axi
Ox,
(13)
(22)
[AXV:e6x] < 0
(18)
Lx/- J
~(o)
= x+ ~
c3~ ( o ) + x7 ~ ~ ( o )
+ R(x + - x T )
(23)
191
General lower and upper bound theorem of static stability: T. E. Boothby and C. B. Brown
(24)
if x,.+ ___x,-
(25)
xi+ = 0
ifxi- >-xi +
(26)
Dissipative systems
(29)
(30)
where
V : + = O+/Ozk
(31)
V z = O-/Ozk
(32)
192
(33)
f(x) - ~ = 0
(34)
Equations (33) and (34) can be treated as additional constraint equations, with the dissipation term of the objective function transformed by equation (30). The upper
bound condition can be seen to correspond to the statement that if the virtual work in an infinitesimal
kinematically admissible disturbance about a state is
negative, the system is unstable in that state. This statement applies to general disturbances, including rigid
body displacements, elastic deformations, and disturbances involving slip or plastic flow.
(27)
z=f(x)
~_>0
Iv,w(0)1
p
H]
(35)
[I
H]
Iv,w(0)1
p
=0
(36)
is satisfied.
(2)/~ ___ 0
(3) The submatrix of [G H I formed of the columns
corresponding to nonzero/~, ? has full row rank.
If these conditions are satisfied, then the space of
feasible directions from O is spanned by the gradient
vectors of the active constraints with strictly positive
Lagrange multipliers p and nonzero r. Hence, the
second-order sufficient condition is satisfied ~7, implying that "q is at a local minimum at the origin, satisfying
the stability criterion.
The n x n matrix ~, such that r4' = I, transforms x
into the orthonormal basis.
x = [~+
~0 4,_1
(37)
General lower and upper bound theorem of static stability: 7". E. Boothby and C. B. Brown
where x+ are the coordinates associated with positive
roots of the Hessian matrix, x0 are associated with zero
roots, representing rigid body translations and x_ are
associated with negative roots, and represent rigid body
rotations. Transformed to this basis, equation (35) is
X+
L(x+, x0, x_, p, r) = V2
Dual program
L* J
(38)
where A+, A_ are diagonal matrices of the positive and
negative eigenvalues of X72'W(x). A sharper form of
the lower bound theorem can now be stated.
,~0r ]
&r/[I
G H]
is satisfied.
(2)
p___0
(42)
xEX
IVx%(0)
1
/i
=0
(39)
X0
=]cl'r/[l
x_
L, _j
(43)
/]TE J
x+
~r_ [G H ]
formed of the columns corresponding to nonzero
p, r has rank greater than or equal to the dimension of x_.
(4) The state
(40)
G H]
I_X_
X+
x
A+
A_
X+ FX+
+
Xo
Xo
x_
X_
(44)
(45)
subject to _ _> 0
(46)
I]TLA
x j
General lower and upper bound theorem of static stability: T. E. Boothby and C. B. Brown
x0
x_
[1 E:]TI ]
X+
X+
x0
E x a m p l e - b l o c k on a plane
X+
X0
" x_
(47)
ATX_
where B = A rAJA is negative semidefinite. The constraint equations (46) are introduced by redefining X, the
underlying space which contains the feasible set.
X=
x0
-Re+#Wl~l
(48)
(49)
ATX_ > 0
(50)
The first five terms are the potential energy, and the last
term is the energy dissipation. The absolute values and
the max[ } operator are eliminated by the procedure of
the section on the piecewise differentiable objective
function. The existence of some elastic deformation
degrees of freedom q is supposed. Retaining the first
two terms of the MacLaurin series approximation, the
following objective function results.
~*(X,
p , r) = - X r F X
(51)
v,~(0)]
subject to
X=
[1 G
HI
(52)
o~X= 0
(53)
Ar~I'r_X > 0
(54)
p ___ 0
(55)
(56)
[
-
Wh
Wb + Rh
tzW-R
IzW + R
q
Wh
Rb
Wh
0-
- Wh
where F = ,I[,+A
-~ ~ +T. This may be recognized as the
problem of maximizing the complementary energy of
the system subject to equilibrium constraints, with the
added condition that the positive forces of constraint
prevent each potential rigid body rotation of the system.
Because it is constructed about a local rather than a
global minimum in x, and the introduction of the
arbitrarily restricted subspace Y is required for the construction, %V* in equation (51) above may be called a
'pseudodual' function. The lower bound theorem has a
more straightforward form for this pseudodual problem.
Theorem 4 (lower bound theorem)
194
~-
"will/l/Ill~
Figure 1 Example block on a plane
0
K
(57)
General lower
of static
stability:
29.6mm
----A
e+ ro
(58)
o- ro
(59)
E+20
(60)
E-10
(61)
and C. 8. Brown
T. E. Boothby
LD
.z,
(62)
Wb+Rh>O
(63)
/.tW-R>O
(64)
/.LW+R>O
(65)
W*(Q) = - % Q&Q
(66)
subject to
8 +=Wb-Rh>O
(67)
8-=Wb+Rh>O
(68)
E+=/_LW-R-p;=0
(69)
E-=pW+R-p;=O
(70)
p:,p;~
(71)
Geometry
of block
system
8.ijyqyiiR4
7.20
0.91
1.9zi
87
(x10-2N
10.83
.l
50
where Q, 8 +, 8 _, E+, E- represent the force quantities conjugate to the displacements q, 8+, 8-, E+, Eand p+ and p; are the Lagrange multipliers for the constraint equations (60) and (61), and F is the elastic flexibility matrix. The constraint equations (67) -(71) can
be seen to express the same criteria as equations
(62)-(65).
Example-rigid
Figure
A system of blocks prestressed by a rubber band, bearing on roller supports is found to be stable when the
blocks are placed in an interlocking pattern as in Figure
2, and unstable when the blocks are placed as in Figure
4. The dimensions of the system are indicated in Figure
2. The mass of each of the blocks is 10.4 g. The elastic
constant of the rubber band is 135.2 N/m. The rubber
band in the initial position exerts a force of
228 X lo- N at the point indicated by the arrows in
Figure 2. A reasonable lower bound to the coefficient of
friction between the blocks is 0.45. Figure 3 shows a set
of forces for which rigid body equilibrium is satisfied
for each block and for which the equilibrium equations
at the joints are satisfied, as in equation (53). At each
joint, the normal force times the coefficient of friction
is greater than or equal to the tangential force, as
Figure3
Lower
r7
ri
r.7
bound
condition-stable
configuration
(72)
i=l
where W, is the weight of block i, xi is the horizontal
distance from the centre of rotation 0 or 0 to the centroid of block i and P is the force exerted by the rubber
band, found by measurement to be 2.23 N at 8 = 0, and
z is the moment arm of the rubber band force, 14 mm.
The first variation
of the potential energy
is
-6.32 x lo- Nm, so equation (15) is satisfied and the
system is not stable at 0 = 0. However, the displacement
shown in Figure 4 is not a rigid body displacement
because of the strain energy accumulated in the rubber
band. To ensure that there is no stable state in the
Engng Struct.
1993
Volume
15 Number
195
General lower and upper bound theorem of static stability: T. E. Boothby and C. B. Brown
the positive forces of constraint for inequality constraints and nonzero forces of constraint for equality
constraints span the space of rigid body rotations, then
the system is stable. A generalized upper bound on the
stability of the system is a set of loads for which the
system is known to be unstable. If, under a given loading
condition, the virtual work in a rigid body displacement
is negative, then the system is unstable.
xi
Acknowledgment
Figure 4
References
vicinity of 0 = 0, the second variation at 0 = 0 is
checked.
a-'n]
't0 ~ J0=0
= ]~
w,>,, + 2kz 2
(73)
i=1
Conclusions
A local minimum energy state associated with a
statically stable state of a mechanical system can be
found for a system with bounded state space, or a system
for which the energy function has a discontinuous first
derivative, by the methods of optimization under constraints. A meaningful local dual function may also be
constructed in the vicinity of such a local minimum
state, even though the energy function may not be convex in the state variable. Generalized lower and upper
bounds on the loads causing such a system to be unstable
can be found by methods which may be much simpler
than the determination of the exact critical loads. A
generalized lower bound on the stability of a system is
defined as a set of loads for which the system is known
to be stable. If a statically admissible set of internal
forces can be found for which the direction vectors of
196