Professional Documents
Culture Documents
Series Editor
Associate Editors
E. J. P. G. Schmidt
Dept. of Mathematics and Statistics
McGill University
Montreal. Canada H3A 2K6
Library of Congress Cataloging-in-Publication Data
Lagnese, J.
Modelling, analysis, and control of dynamic elastic multi-link
struetures I J. E. Lagnese, Giinter Leugering, E. J. P. G. Sehmidl.
p. em. -- (Systems & control)
Includes bibliographieal references and index.
ISBN 978-1-4612-6689-1
9 8 7 6 5 432 1
Contents
List of Figures xi
Preface xiii
Chapter I. Introduction 1
1. General Overview 1
2. On the Contents of the Book 6
Chapter II. Modeling of Networks of Elastic Strings 13
1. Modeling of Nonlinear Elastic Strings 13
2. Networks of Nonlinear Elastic Strings 16
3. Linearization 19
4. Well-posedness of the Network Equations 23
5. Controllability of Networks of Elastic Strings 33
5.1. Exact Controllability of Tree Networks 36
5.2. Lack of Controllability for Networks with Closed Circuits 40
6. Stabilizability of String Networks 42
7. String Networks with Masses at the Nodes 44
Chapter III. Networks of Thermoelastic Beams 47
1. Modeling of a Thin Thermoelastic Curved Beam 47
2. The Equations of Motion 60
2.1. Some Remarks on Warping and Torsion 66
3. Rotating Beams 67
3.1. Dynamic Stiffening 70
vi CONTENTS
8. Networks of Beams 83
8.1. Geometric Joint Conditions 83
8.1.1. Rigid Joints
8.1.2. Pinned Joints
8.2. Dynamic Joint Conditions 85
8.2.1. Rigid Joints
8.2.2. Pinned Joints
1. Preliminaries 135
1.1. Notation 136
1.2. Networks of Strings 138
1.3. Networks of Timoshenko Beams 139
1.4. Networks of Euler-Bernoulli Beams 140
CONTENTS vii
2. Linearization 266
2.1. Linearization of Equations of Motion 266
2.2. Linearization of Edge Conditions 269
2.3. Hamilton's Principle for the Reissner Model 270
2.4. Linearization of the Vector Rotation Angle 270
2.5. The Kirchhoff Plate Model 272
3. Systems of Linked Reissner Plates 273
3.1. Geometric Junction Conditions 274
3.2. Linearization of the Geometric Joint Conditions 275
3.3. Dynamic Joint Conditions 277
3.3.1. Dynamic conditions at a connected joint
3.3.2. Dynamic conditions at a hinged joint
3.3.3. Dynamic conditions at a semi-rigid joint
3.3.4. Dynamic conditions at a rigid joint
3.4. Junctions With Masses and Applied Forces 283
3.4.1. Dynamic conditions at a connected joint
3.4.2. Dynamic conditions at a hinged joint
3.4.3. Dynamic conditions at a semi-rigid joint
3.4.4. Dynamic conditions at a rigid joint
4. Well-posedness of Systems of Linked Reissner Plates 286
4.1. Function Spaces for Linked Plates 287
4.2. Existence and Uniqueness of Solutions 289
5. Controllability of Linked Reissner Plates 292
5.1. Controllability in Transmission Problems for Thin Plates 293
5.1.1. Observability Estimates and Consequences
6. Systems of Linked Kirchhoff Plates 304
6.1. Semi-rigid Joints 305
6.1.1. Interpretation of Theorem 6.1
6.2. Proof of Theorem 6.1 313
6.3. Connected, Hinged or Rigid Joints 315
Chapter VIII. Plate-Beam Systems 319
1. Introduction 319
2. Modeling of the Plate-Beam Junction: I 321
2.1. Geometric Conditions 321
2.2. Dynamic conditions 325
3. Function Spaces and Well-Posedness 330
4. The Reachable Set 345
4.1. Observabilityestimates 346
4.2. Proofs of Theorems 4.1 and 4.2 362
CONTENTS ix
large rotations in the joints. However, a global rigid rotation of the overall
structure may very well be applied in both cases. There are many different
possibilities which, because of space limitations, cannot all be considered
here. When dealing with the problem of stabilization and, in particular,
with robustness one has to specify classes of disturbances which need to
be counterbalanced within the structure. The classical viewpoint is that
regardless of the nature of the disturbance, the result of any disturbance is
a new set of initial conditions distinct from the equilibrium state. There-
fore, a primary goal is to guarantee uniform energy decay of all possible
trajectories starting near the equilibrium.
We shall also provide, in Chapter V, a number of numerical simulations
of structures composed of one dimensional elements. The purpose of the
simulations is to validate the models and to illustrate the control concepts
and their effectiveness. We believe that these simulations strongly support
the efficacy of the models developed in this monograph, although much
work along these lines, especially with regard to structures containing two
dimensional elements, remains to be done. In particular, a rigorous nu-
merical analysis of the transient behavior of elastic structures and their
associated control systems based on the models presented in this book is
beyond the scope of this work.
It is our hope that this book will prove to be useful not only to engineers
and mathematicians interested in design, modeling and control of flexible
structures, but also to those wishing to develop some understanding of
the sorts of complicated interactions between elements that can and do
occur in complex elastic systems. While a substantial knowledge of partial
differential equations and functional analysis is necessary for a complete
understanding of the details of the mathematical analyses, we believe that
this book will also be useful to those who simply wish to understand the
dynamics of elastic networks at a descriptive level. To this end we have
tried, in particular, to present a self-contained treatment of the mechanical
modeling aspects of the research described in this book.
The authors wish to convey their appreciation to those organizations
whose financial support made possible the research which led to this work.
The research of the first two authors was supported by the Air Force of
Scientific Research. The research of the second author was also supported
by the Deutsche Forschungsgemeinschaft (DFG), Heisenbergreferat. The
research of the third author was supported by grants from the Natural Sci-
ences and Engineering Research Council of Canada and from the Ministere
de l'Education du Quebec.
CHAPTER I
Introduction
1. General Overview
The need to control multi-body structures is evident and does not re-
quire further justification. What may be necessary to justify however, is
the relevance of distributed parameter models and their control properties
in such a context. Indeed, ultimately every practically implemented con-
trol law has to be computed in one sense or another from measurements
and data taken from the structure via a finite dimensional model of the
structure. It is, however, well known that controls which are computed
from a reduced order model may lead - and in many cases do lead - to
the spillover of energy into unmodeled higher modes of the system. This
is one reason for studying control problems for flexible structures in the
context of distributed systems: control results for highly flexible structures
obtained without taking into account the infinite dimensional nature of the
system are, in principle, questionable.
Another, even more important reason for distributed parameter system
(DPS) modeling and control of flexible structures stems from the fact that
such structures are often deployed in hostile environments in the sense that
the location may be difficult to access (space-structures), that some of the
states are hard to observe and to difficult to control, that the physical
parameters are subject to drastic changes (e.g., the temperature), and so
forth. Often the system cannot be adequately tested under its expected
working conditions. All of these considerations make precise mechanical
and mathematical modeling of such structures mandatory. There are nu-
merous examples of situations where poor modeling has led to catastrophic
results, even in the presence of superb control techniques. We only need
mention EXPLORER 1, launched in 1958, which began to tumble after a
few orbits because of unmodeled internal damping stemming from flexible
antennae attached to the main body, an effect which was not accounted for
in the control device. Another famous example is provided by MARINER
2 I. INTRODUCTION
10 (1973). In that case the control regime activated the 7th frequency
of oscillation, which was not considered in the model. See Bremer [9].
A third example is provided by the HUBBLE telescope (1989) which has
not operated as anticipated, partly because the abrupt and large changes
in temperature when the spacecraft passes from the earth's shadow into
sunlight cause tremendous thermal stresses in the beams carrying the so-
lar panels. Those thermal stresses are converted into long bending waves,
which ultimately cause some panels to flip. Examples of this kind are by
no means restricted to ambitious projects in outer space. Consider, for
instance, a simple system consisting of a beam with a mass attached to
it, rotating around its centerline. The combined objects can be viewed as
a rigid body as long as the angular velocity is small. However for large
angular velocities the system behaves like a deformable elastic body. This
is even more apparent if the axis of rotation is perpendicular to the cen-
terline, that is, if one considers robot arms, crane mechanisms, etc. In all
of these examples it is clear that a thorough mechanical and mathematical
modeling of the system and controls is indispensable. This monograph is
intended precisely to contribute to this aspect of flexible structure control.
We do not claim that the approach to modeling which we take here
is conceptually new from a mechanical viewpoint. (However, we do not
know of any similar work on the mechanical modeling of interconnected
membranes and plates). Nor do we suggest that our approach is supe-
rior to others as far as practical implementations of control stragegies are
concerned. Rather, in this work we wish to emphasis the tmnsient behav-
ior of solutions, the intemction phenomena at mechanical junctions, the
occurrence and propagation of singularities through junctions regions and
to study questions of well-posedness and stability, spectml properties and
exact, spectml or approximate controllability of the transients. In these
respects this monograph is almost disjoint from the vast literature on flex-
ible structure control. Indeed, the sparsity of mathematical litemture on
interaction problems of the kind considered here is very surprising. In
most papers on the subject, even after some mechanical justification of
balance laws the system is represented by a finite element approximation
or a modal reduction in terms of generalized coordinates, which necessar-
ily have finitely many degrees of freedom only. We may cite Hughes [39],
Kane et al. [41], Shabana [99] as exceptional examples, where the model-
ing is based on principles of continuum mechanics and where the relation
between the distributed model and its finite dimensional approximation is
kept transparent. In most engineering papers, however, multi-link struc-
tures are introduced through their finite dimensional representations rather
than through a system of partial differential equations or integrodifferential
equations with appropriate junction and boundary conditions, as it is done
in this monograph.
It is well known that many phenomena and distinctions which are present
1. GENERAL OVERVIEW 3
in (linear) infinite dimensional systems are lost in its finite dimensional ap-
proximations. Examples are the distinctions which exist in infinite dimen-
sional systems between various notions of stability (weak, strong, uniform)
and different concepts of controllability (spectral, approximate, exact). In
the context of an elastic frame containing closed circuits, the role of comea-
surability of the optical lengths of the structural elements in a circuit, which
is crucial to understanding the lack of (even weak) stabilizabilityand (even
approximate) controllability of such a frame, disappears in finite dimen-
sional approximations of the frame. In fact, it is the fine structure of the
frequency spectrum and its asymptotic behavior which can be viewed as be-
ing responsible for these differences. Because of the tradition and tendency
in the engineering literature to analyze finite dimensional approximations
only, the fine structure of the spectrum of multi-link structures is generally
not considered (see, however, Wittrick and Williams [110], [111]). In the
mathematical literature, exceptions are to found in the work of von Below
[106], [105] and Nicaise [75], [77], who consider the spectrum of second
order differential operators acting on scalar functions on 1-d networks.
We do not have sufficient space to give appropriate credit to the huge
engineering literature related to control of flexible structures. We simply
cite [39], [99] and [41] as general references and refer the reader to their
excellent bibliographies. Rather, we concentrate on the mathematical liter-
ature concerned with the problems to be addressed in this monograph. One
may identify two distinct groups of papers within that literature. The first
group considers multi-link flexible structures as being obtained from 3-d
elasticity. In this philosophy a frame (with rigid joints) is obtained from a
cube of dense material by "carving" out appropriate parts. The approach
is made mathematically rigorous by a singular perturbation (asymptotic)
analysis with respect to a certain ''thickness'' parameter using ideas due to
Ph. Ciarlet and Dstuynder [18]. It has subsequently been used by many
authors to obtain models of elastic junctions for a variety of multi-link
structures. (See in particular the monographs of Ciarlet [19] and LeDret
[57] and references therein.) This method is very powerful and universal. It
has, however, some shortcoming when it comes to dynamical problems. In
particular, as a consequence of the scaling of the Lame moduli with respect
to the asymptotic parameter, it occurs that the in-plane (or longitudinal)
stiffness of the material becomes infinite in the limit of the asymptotic pa-
rameter . This being the case, the interaction conditions at the junctions,
in particular, the continuity condition on the vector displacements of all of
the elements meeting at the junction, can be violated. Therefore, from our
perspective, the transmission of waves at the joints is not as transparent
as in the models developed here. In some important cases, however, the
junction conditions obtained here can be specialized to the ones obtained
within the framework of asymptotic analysis.
The second group of papers is concerned with constructive approaches
4 I. INTRODUCTION
since they balance forces and moments at the centerline of the beam with
avemges of forces and moments in the plate taken over the junction region.
We also investigate the exact controllability problem in the case of the
first configuration and show that the plate-beam system may be exactly
controlled through the application of controls in the forces and moments
along a portion of the edge of the plate which excludes the junction region.
Further, in the case of the first configuration, a simpler model is obtained
by the process of passing to the limit as the shear stiffnesses on the plate
and beam increase beyond bound, in analogy with the process carried out
in Chapter VII. The result is a combined Kirchhoff plate - Raleigh beam
system coupled to an ordinary differential equation which governs a time
dependent (rigid) rotation of the beam around its centerline.
CHAPTER II
This chapter deals with three dimensional networks of elastic strings. Such
networks, originally introduced in [94], provide a paradigm for the much
more complicated networks of beams, membranes and plates which are
the central subject matter of this monograph. Many of the main ideas of
this monograph can be illustrated in the relatively simple context of string
networks and certain questions, such as those concerning controllability
and stabilizability, can be answered more fully than is possible for the
more complex structures to be considered later.
(1.1) K(t) = -
2
111 pIR(x, tW dx.
0
and
1 t
(1.2) U(t) = 210 h[lR,x(x, t)1 - 1]2 dx,
respectively, where I . I denotes the Euclidean norm, the dot "." indicates
partial differentiation with respect to t and the subscript ",x" partial differ-
14 II. MODELING OF NETWORKS OF ELASTIC STRINGS
for all admissible 8R. Assuming that R(x, t) is twice differentiable (and
that the integrals are indeed convergent) one integrates by parts to obtain
(1.6) .. [ Rx]
pR = h R,x - IR:x x' 1,
If one now takes variations which do not vanish at lone is led to the free
(or naturaQ boundary condition
.. Rx] x +F,
h [R,x - IR:xl
.
(1.9) pR=
(1.10)
(1.11) [R(O, t) - c] x b = 0,
where x denotes the cross product. Then the perturbations 8R(x, t) must
be such that 8R(0, t) is a multiple of bj in that case 8R(0, t) = cf>(t)b
and, since cf>(t) can be an arbitrary smooth function, one concludes from
Hamilton's Principle that (1.11) must be complemented by the dynamic
boundary condition
REMARK 1.2. The assumption that Hooke's law holds is certainly only
reasonable for a limited range of extension (corresponding to IR,xl > 1)
or contraction (corresponding to IR,xl < 1). Hooke's law can be replaced
by replacing the quadratic potential energy function h/2[1R,xl - 1]2 by
more general functions of the form u(IR,xl- 1); this leads to a variety of
intriguing nonlinear systems.
REMARK 1.3. For related nonlinear string models derived in other ways
see Carrier [14] and Antman [2].
(2.2)
2. NETWORKS OF NONLINEAR ELASTIC STRINGS 17
(2.3)
where the control functions U i (t) belongs to a specified set of admissible
controls. Other nodes, indexed by j E .IN, may be subject to a Neumann
control which physically corresponds to the application of a force U i (t) at
the j-th node. The precise form of such a condition (namely (2.12) below)
will emerge from the variational argument. The remaining nodes will be
referred to as free nodes and correspond to indices j E .JF; at these nodes
the free (or natumQ node conditions (2.13) are a consequence of Hamilton's
principle. We refine the index notation by allowing both subscripts and
superscripts with the obvious interpretation: for example .Jfj = .IN n
.JM. In most of our applications, and certainly for the string networks we
consider in this chapter, these index sets will be disjoint. This point will be
remarked upon again in Section 2.6. To complete our terminology, we say
that the clamped and Dirichlet conditions are geometric conditions, while
the free and Neumann conditions are dynamic conditions. The former act
directly as constraints on the underlying geometric structure, while the
latter are embodied in the formulation of the Lagrangian functional and
hence a consequence of Hamilton's principle.
The kinetic and potential energies of the system are respectively given
by
(2.4)
and
(2.5)
(2.7) .c(R) = [T
10
[L [I. [~i IRi(x, t)12 + Fi(x, t) . Ri(x, t)
iEI10
One obtains
-h[R:x-lll:~xlOR:x](X,t)dX+ L Uj(t).ONj(t)]dt=O.
~ jE3 N
By using variations oR which vanish near the nodes one obtains the partial
differential equations
(2.10)
forjE:J N ,
(2.11)
for j E :JF.
The above equations together with the geometric conditions (2.1), (2.2)
and (2.3) have to be supplemented by initial conditions which prescribe
R(O) and R(O).
3. LINEARIZATION 19
If the Dirichlet data are time independent and the forces Fi and U j all
vanish it is easy to verify, at least formally, that the total energy
t
'EI10 ,x
where we have used the node conditions (2.1), (2.2), (2,3), and (2.11) to-
gether with the time independence of the Dirichlet data and the vanishing
of the Neumann data.
3. Linearization
In this section we linearize the network equations about equilibrium
configurations which we shall denote by R. The latter are obtained by
solving
(3.1) . Rix
hi [R~x - 'R~x'
1 (x) = F'(x),
-. for i E I,
,x
subject to (2.1), (2.2), (2,3) and (2.11) with all data Fi and iJj independent
of time. In fact there are generally many possibilities. In this regard see [95]
(and Section 2.7) for networks in which gravitational forces are taken into
account. Here we shall suppose that at equilibrium there are no distributed
forces acting along the strings so that Fi(x) = O. The individual equations
in (3.1) then have two evident classes of solutions. The first, very large,
class of solutions consists of all functions Ri(X) for which ,Rix(x), = 1 and
describe limp strings. The second class of stretched or comp~ssed solutions
have the linear form
(3.2)
where e i is the unit direction pointing along the i-th string segment and
Si is a constant measuring the extent to which the string is stretched (if
Si > 1) or compressed (if Si < 1). For networks one must also satisfy the
20 II. MODELING OF NETWORKS OF ELASTIC STRINGS
geometric node conditions (2.1), (2.2) and (2.3) as well as the dynamic
conditions (2.11) which take the form
for j E :r N ,
(3.3)
for j E :r F
If Vi -# 0 the strings adjoining the j-th node cannot all be limp. If all
Vi = 0 networks of limp strings automatically satisfy (3.3). It is easy
to give examples of networks and boundary conditions with equilibria in
which different elements may be limp, stretched or compressed. We shall
avoid such situations and suppose that the boundary conditions are such
that in the equilibrium configuration all strings are stretched. We then
linearize about such a stretched equilibrium configuration.
To carry out the linearization, we let
Obviously jii(X, t) = ri(x, t) so that the left hand side of equation (2.10)
can be immediately rewritten and remains linear. One possibility is now
to linearize the right hand side of (2.10) as well as the left hand sides of
the node conditions (2.11) directly, using the following quadratic Taylor
expansion in ri about 0:
1 1. 2 1 '2
=si[1+-e'r' +-Ir' I --(e'r') + ... ].
s, .
,x ,
2s~ ,x ,
2s~ ,x
3. LINEARIZATION 21
(3.6)
and
(3.8)
+ L U i . 8ni ] dt = O.
iE.:J N
Here the variations 8ri and the corresponding 8ni must respect the condi-
tions (3.4), (3.5) and (3.6) and can also be required to vanish for t = 0 and
22 II. MODELING OF NETWORKS OF ELASTIC STRINGS
[T
Jo
[L Jo[l, [-p.'r' + Q,r~xx]
tEI
. 8r' dx
for j E :IN,
L [t'Jh,(s, -
'EI,
l)e' + t'JQ,r:x] {~' for j E :IF.
for j E .IN,
(3.10)
for j E .JF.
/C(t)
t,
= L Jo p,Ir'1 2 dx,
W(t) = uJ nJ,
Ill.
'EI 0 JE.:JF
(4.4)
defined in a formal fashion for the moment. Two variants of this operator
will be defined below on domains which incorporate several of the side
conditions occurring in the system (4.1).
The precise formulation of the problem depends on the definition of
appropriate spaces. We shall usually require i-(-, t) and the initial data ro
to belong to
where L 2 (0, lii IR3) is the space of square integrable functions mapping
(0, li) into IR3. We provide H with the inner product
which corresponds to the kinetic energy. We shall require r(, t) and the
initial data ro to belong to the space
where Hk(O, lii IR 3 ) denotes the space of functions which have square inte-
grable distributional derivatives up to order k. On V we we define
(4.6)
For)' > 0, II r IIv,>.= (r, r)~~ is a norm for V equivalent to the standard
.:r
product norm. In case c =f 0 it follows from the connectedness of the
network that II r Ilv=1i r liv,o is also an equivalent norm. We shall need the
following subspace of V:
1 = V x H, 10 = Vo x H.
In view of (4.3) it is natural to try to obtain the solution of (4.1) with the
help of semigroups (which will turn out to be unitary groups) acting on 1
or 10 generated by operators of the form
(4.8)
Ar-Ar = f,
ri(xij) coincide for j E .JM, i E I j ,
(4.9)
1J1r = u j for j E .JD, 1J1r = 0 for j E .Jo,
J+/jr = u j for j E .IN, J+/jr = 0 for j E .JF.
for all 4> E Yo. It is easy to convince oneself that any sufficiently smooth
weak solution indeed satisfies the equation as well as the latter two dynamic
conditions in (4.9).
The uniqueness of solution to (4.9) is easily verified by subtracting the
weak identities for two given solutions, then setting 4> equal to the difference
of solutions to get (4),4>)v,>. = 0 and hence 4> = O. Existence of weak
solutions is obtained by a standard application of the Riesz representation
theorem in case A > 0 and also, if .Jo U.JD t=
0, when A = O. One
introduces rEV satisfying i+l j
= u for j E .JD, where i.r1
denote the
nodal positions corresponding to r. One then seeks a solution to (4.9) in
the form r = r + s. Then s must belong to V 0 satisfy
for all 4> E Yo. The right hand side of the latter identity defines a linear
functional on Yo, which is then representable by some s in terms of the
inner product (., )v,>..
In particular we can solve the weak stationary problem with homoge-
neous Dirichlet and Neumann data, in which case the solution lies in Yo.
One can for A > 0 (and, in case .Jo U.JD t=
0 also for A = 0) define the
solution operator S>. : H t-+ Vo c H by S>.f = r where r is the unique
solution of (4.10) with each u j = O. The operator is injective, since if
S>.f = 0 (4.10) implies that (f,4H = 0 for all 4> E Yo; thus f = 0 since
Vo is dense in H. Setting 4> = S>.g in (4.10) one gets the identity
(4.12) (S>.f, S>.g)v,>. = (f, S>.g)H.
From this it follows that S>. : H t-+ H is bounded, symmetric and, taking
into account the injectivity, positive definite. If we set f = g and then
26 II. MODELING OF NETWORKS OF ELASTIC STRINGS
apply Schwarz's inequality to the right hand side we also get the estimate
II SAf Ilv,A::;1I f IIH. It then follows from Rellich's theorem, which asserts
the compactness of the imbedding of H1 (0, Ii; IR3) into L2 (0, Ii; IR 3), that
SA : H I-t H is a compact operator. Consequently the spectrum consists
of a sequence of eigenvalues of finite multiplicity accumulating at and
corresponding to a complete orthonormal system of eigenfunctions. The
eigenvalues are positive.
The image of SA is dense in 1t. We define the self adjoint operator
AA = _8,;:1 on dom(AA) = im(SA)' The domain does not depend on A
because
(4.13)
implies
Ao[r x r] = r x Aor,
(4.15) {
with r x r E dom(Ao) = dom(Ao) x Yo.
In case :Tc U:T D =I- 0 (so that (', )v is an inner product on Yo) it is a
simple exercise in definitions to show that this is a skew adjoint operator
on 'H.o with respect to the energy inner product
(r x r,s x S)E = (r,s)v + (r,s)H
4. WELL-POSEDNESS OF THE NETWORK EQUATIONS 27
An[r x r] = r x AOlr,
(4.16) {
with r x r E dom(An) = dom(AOl) x VOl
is a skew adjoint operator on 101 = VOl X HOI with respect to the energy
inner product, and as before generates a unitary group U OI (t )tE.fl on 101.
We define 100 = Voo x Hoo and define UOO(t)tElR on 100 by
Uoo(t)[roo x roo] = (roo + troo) x roo.
This is a group of operators which is generated by the bounded operator
Alo[roo x roo] = roo x o. This group preserves energy but cannot be called
unitary since energy is not a norm on 1oo! Given ro x ro E 10 we can
decompose it into the sum of elements of 100 and 101:
weak formulation of Ar1 = 0 and ./Vjr = 0 for j E .:IN U .:JF. The initial
data can also be decomposed into
A.[r x r] = r x Ar,
{
with r x r E [dom(Ao) + V 1] xV.
In case Vo =V (or V 1 = {O}) we shall from now on simply write
U(t) = Uo(t).
Given initial data ro x ro E 1 and f(t) E L1 (0, T; H) one obtains a
solution to the system (4.1) with time independent Dirichlet data u j =
'lYro and vanishing Neumann data by setting
Solutions obtained in this way are called "mild solutions" and are exten-
sively discussed in Chapter 4 of pazy [83]. In particular the following
regularity result can be inferred from Theorem 2.4 and Corollary 2.5 of
that Chapter.
In particular one can conclude that if each fi E Coo ([0, Ii] x [0, T]; JR3)
and if ro x r E dom(Aoo) the solution r(t) x r(t) lies in Coo([O, T]; V x H)
and consequently that each ri(x, t) E COO ([0, Ii] x [0, T]; JR3).
We next consider time dependent Dirichlet data and inhomogeneous
Neumann data in (4.1). The Dirichlet data has to be compatible with the
initial data. We start with data satisfying
The next lemma states a multiplier identity together with a resulting esti-
mate; such identities go back at least to Rellich and have played an impor-
tant role in the control of hyperbolic equations (see Lop Fat Ho [37] and
Lions [60]).
As a consequence
PROOF. The idea is to multiply the equation by m(x)r,x and then in-
tegrate by parts to get the desired identity. We first treat the two terms
separately. Integration by parts with respect to t gives
1T 11 pf . mr,x dxdt = -1
T
11 mpr . r,x dxdt
- i [m
o
]
T -2 plrl2 x=1 dt +
x=o
11 [
0
]
mpr . r,x t=T dx,
t=o
[T [I [T [m
- Jo Jo [Qr,xl,x . mr,x dxdt = - Jo "2 Qr ,x' r,x x=o dt
]X-I
(4.25)
[T
+ Jo
t
Jo [ t
m
Qr,x' r,x -
m
"2 Q,x r,x . r,x] dxdt
If we add together the two identities (4.24) and (4.25) and regroup terms
we get (4.22).
Finally we obtain (4.23) by setting m(x) = -1 + 2xjl in (4.22) and by
estimating all the integrands on the right hand side of that identity by
suitable multiples of e(x, t). 0
=~ i
~ [Qr
L...J L...J 'J 'J' T)
.,x . ri](x
iE.7 iEIj
L[L
(4.27)
+ liti(TW + L IUi(TW].
jEJ D jEJ N
32 II. MODELING OF NETWORKS OF ELASTIC STRINGS
Note that
+ LJot [L
iE:JD
Itii (r)1 2 + L
iE:J N
lui (r)1 2] dr
where G is a positive constant. We can choose 10 so that lOG < 1 and then
bring the term (t) to the left and divide by 1- lOG to obtain the estimate
jE.::J N
(4.32)
for all 4> E Yo. Now we can take as test functions the eigenfunctions of
Ao. If such an eigenfunction corresponds to the eigenvalue -J.L we have
(s(t),4v = J.L(s(t), 4H. Consequently (4.31) becomes
where m D and m N are the respective cardinalities of .:rD and .:rN and
where HI (( 0, T1; IR 3) is the subspace of HI (0, T; IR3) consisting of functions
vanishing at t = O. On these spaces we can use the inner products
and
(UN, vN)UN = L iT ui(t). vi(t) dt.
iE.:J N 0
(5.2) lI,xo x ..\0 111t:S C I C;,[-Xo x ..\olllu, for all ,xo x ..\0 E 'H..
Here the spaces 'H. and U have been identified with their dual spaces. First
we need to identify C:}[,xo x ..\01. The form of this operator does of course
depend on the choice of inner products.
We introduce the following adjoint system:
.. i .
Pi,x = Qi,x:xx for i E I,
,xi(Xii,t) coincide for j .:rM, i E Ii'
E
PROPOSITION 5.1. Suppose that :fe -::f:. 0. Let ,x(t) be the solution of
(4.3) corresponding to terminal data ,xo x ~o E 1. Then
jE.:J D jE.:J N
Hence (5.4) follows by Riesz identification, taking into account the inner
products we have chosen for U D and UN and noting that
o
36 II. MODELING OF NETWORKS OF ELASTIC STRINGS
. 2 h(l - 8- 1 ) 2 h
x ILkt = c, for k = 1,2 wzth ILl = , IL2 = -.
P P
Then for t2 - t1 ~ T* = 28* = 2l/ ILl one has the estimate
THEOREM 5.1. Suppose that the network contains no closed circuit, that
all multiple nodes and no simple nodes are free, that exactly one simple node
is clamped and that the remaining simple nodes are controlled by Dirichlet
or Neumann controls. Then the network is exactly controllable for each
time T 2: T*, where
REMARK 5.1. The time T* is the best possible for the general statement
of the theorem as can easily be verified by considering the degenerate net-
work consisting of a single string clamped at one end and controlled from
the other, for which T* coincides with the minimum time for controllability
established, for example, with the help of the d'Alembert solution.
REMARK 5.2. the fact that one simple node is clamped ensures that
the energy in fact is a norm on 1(.. One can in fact allow a clamped
multiple (rather than simple) node with controls operative at all simple
nodes. In case no node is clamped the energy is no longer a norm, but our
methods would easily give controllability in the quotient space of 1(. with
the subspace consisting of states of zero energy. We are not sure whether
exact controllability also holds in the whole space!
{
ceij{Xj,t) ~ IN?(t)~ ~ Ceij{Xj,t) for j E .JR,
Ceij{Xj,t) ~ I1)J.,X{t) I ~ Ceij{Xj,t) for j E .J:.
THEOREM 5.2. Assume that the hypotheses on the network made in the
previous theorem are satisfied. Let r be a solution of the system (5.1) with
the homogeneous initial condition replaced by the requirement r{O) x r{O) E
?t. Then
(5.1O)
indices 1 :::; i < jo and the element leading to the jo-th node has index jo.
By the induction hypothesis we have for the reduced network
(5.11) ~ ,t
L.,;i(2):::; G
"
L.,;
i .-
ij(xj,O,T*),
i=jo jEjs\{m}
and that
(5.12) j8 = {jo} U [..78 \ {j : 1 :::; j < jo}].
By a shift of the time variable (5.11) implies
(5.13) ~ T*
L.,; i( 2) :::; G "i
L.,; ij (Xj; T*
2 T* T*
- 2' 2 + T*
2)
i=jo jEjs\{m}
The proof of Theorem 5.2 is completed by noting that for i = 1, ... ,jo -1,
Lemma 5.1 gives
T* . T* T* T*
i
i ( 2 ) + (Xijo, 2 - 2' 2 + T*
2)
T*
i(
:::; Gi Xi; 2 - T*
2 - i
S*,
T*
2 + T*
2 + s*i )
:::; Gl (Xi; 0, T*).
o
40 II. MODELING OF NETWORKS OF ELASTIC STRINGS
pw2q, = Qq"xx,
(5.19) {
q,(0) = q,(l) = O.
One can associate with Q an orthonormal basis {el' e2, e3} of IR3 consisting
of eigenvectors of Q. We can take el to be the longitudinal direction of the
string at equilibrium. Then
(5.20) 2
Qek = wkek, . WI2 = h, w22 = w32 =
with h( 1 - -1) .
s
One now easily verifies that if for some k
(5.21) is an integer,
#.,fiJliw
l
(5.23) ~ is rational for each i = 1, ... ,r and k = 1,2,3.
pIllwk
From this it follows readily that one can find W so that
..fiiili i .
(5.24) W - .- = nk for each z = 1, ... ,r and k = 1,2,3,
Wkll'
L c1+1{~;x1(0)e1+1
3 3
Q, L cl:{k,x(l,)el: = Q'+1 for i = 1, ... , r,
k=1 k=1
or, taking into account (5.20) and (5.24) and cancelling W from both sides,
L ckwkJp;el: = L C1+
3 3
1W kVP,+1 e~+1 for i = 1, ... ,r.
k=1 k=1
To see that this can always be solved we introduce the orthogonal matrix 8,
which has ai, a2 and a~ as its columns and we incorporate the coefficients
ckwk..;p;, with k = 1, ... , 3 into a column vector c'. The latter equations
then take the form
8 ,c' = 8 ,+ 1c ,+1 lor
z = 1, ... , r,
and it is a trivial matter to see that this has a non-trivial solution given by
c' = 8;8 1 c 1 where c 1 can be chosen arbitrarily.
If we construct such an eigenfunction on a circuit within a larger network
it can of course be extended to the other elements by simply setting ljJt = 0
for i = r + 1, ... ,n. One is then led to the following theorem:
THEOREM 5.3. Suppose that a stnng network contains a circuit along
which the vanous pammeters satzsfy conditwn (5.24). Then ker(C;') i= 0
so that the system is not approxzmately controllable for any positive T.
REMARK 5.3. The result of this section raises a variety of mathemati-
cally intrigueing questions which are, perhaps, of limited interest from the
point of view of applications where a distinction between rational and ir-
rational parameters does not seen "physically stable." Regardless, it may
be of interest to pose some of these questions:
Is the existence of an eigenfunction satisfying the overdetermined
system (5.16) and (5.17) equivalent to the conditions ker(C;') i= 0?
In general, what is the structure of the image of CT?
Is there any case of a network containing closed loops which is
exactly controllable?
(6.1) dE (t) =
dt
L Afi r .1)if(t).
iEJ"s\{m}
In the case of tree networks we shall see that for this linear feedback the
decay is in fact exponential. This depends on the same energy estimate,
namely that given by Theorem 5.2, which was previously used to prove
exact controllability. Before showing this we point out that the solution to
(6.2) can be given in terms of a contraction semigroup Us(t) on the space
11.. We do not prove this here but refer to the more general result Theorem
IV.6.1. Assuming the existence of this semi-group we now use an argument
due to Rauch and Taylor [87].
It follows from (6.1) and the feedback law that
Since 1)if(t) = fij (Xi, t) and Afir(t) = iQiJr~Hxi' t) the feedback law
implies that
Applying Theorem 5.2 to r and reversing the inequality given there one
44 II. MODELING OF NETWORKS OF ELASTIC STRINGS
~ -C IIUs(T*)[ro x ro]II~,
where in the last step we have used the contraction property of the semi-
group. From (6.3) it then follows that
REMARK 6.1. It follows from the construction in Section 6.2 of the aber-
rant eigenfunction cP corresponding to a closed loop under the condition
(5.23) that the feedback system will not be stabilizable. One simply con-
siders r = (coswt)cP which is a solution of (6.2) but does not decay.
Here we shall merely sketch some of the features of a model which in-
corporates point masses and the action of gravity.
Let mj be the mass located at the multiple node indexed by j. Let
v denote the vertical direction and 9 be the gravitational constant The
kinetic and potential energies of the system are respectively
and
(7.2) U(t) =L
iEIJo
ti ~i [[/R~x(x, t)/- 1]2
+ gpiRi(X, t) . v] dx + L gmjNj . v.
jE.JM
For simplicity, suppose that no additional forces are acting on the strings
or on the masses and that we apply Dirichlet controls (2.2) and (2.3) at all
the simple nodes, so that JS = JC U JD. Hamilton's principle then gives
(7.3) i EI,
(7.5)
along with the conditions
The triad ei(xt}, the initial body frame, can be written in terms of the
standard base e? in R 3 by the way of a rotation r 0 as
ei = roe?,
where ror~ = I. We have
8 8 0 8 ) t
-8 ei = -8 rei = (-8 ro rOei =: flOei
Xl Xl Xl
Obviously, flo is skew-symmetric, and hence flOii = 0, fl Oij = -flOji . With
this notation the initial curvatures and twist of the beam or, more pre-
cisely, of its centerline in the reference configuration, can be described
using Frenet-type formulae as
-K.3)
-T .
ro = r(xI,O,O),
is deformed to
The triads gi, G i represent the tangents along the coordinate lines in the
undeformed and deformed configuration, respectively. These are right-
handed systems as the determinant of the deformation matrix (Gi)i given
with respect to the base gi is positive definite. In particular, at the reference
line we have the triads
Note, however, that the second triad, Ei is not necessarily orthogonal, due
to shearing. We introduce the vector Ell = R,dIIR,lll in the direction of
the tangent R,l(Xl,O,O,t) reference curve. We postulate that the location
of the deformed point can be written as
In (1.1) the function > is related to the solution X (called the torsion func-
tion) of the St.Venant torsion problem in the following way: X satisfies the
equation LlX+2 = 0 (Ll signifies the Laplace operator) over the cross section
and vanishes at the boundary of the cross section. This function X can be
written as >l - !(x~ + x~) with a harmonic function >1, the Prandtl stress
function. With this notation, > above is the conjugate harmonic function
to >l' Hence, > is itself harmonic. By the assumed double symmetry and
the application of Schwartz' reflection principle, > has the following symme-
tries: >(X2' X3) = >( -X2, -X3), >( -X2, X3) = >(X2' -X3) = ->( -X2, -X3).
For a canonical example, we may think of the case of an elliptical cross
2 2
section ~ + ~ = 1 where the stress function >l is given by
8a 2 ~ (_l)n sinhkn x 3 .
<p = X2 X 3 - ~ ;:0 (2n + 1)3 cosh(knb/2) smknX 2
with kn = (2n + l)7r/a. The above mentioned properties of <p are easily
verified in these examples. The function 'Y(XI) measures the amount of
warping of the cross-section at Xl. Warping is considered small if the beam
is thin, and hence 'Y<P is small compared to the breadth of the beam. We
note that (1.1) restricts the geometry of the deformed beam, and that it
would be more accurate to account for higher order terms in the variables
X2,X3
Some remarks are in order.
REMARK 1.1. (1) First of all, with the assumption above we are looking
for a first order theory, that is a theory with Taylor series of the deformed
point in the variables X2, X3 truncated after the linear terms. There are
various papers in the literature in which theories of arbitrary order are
developed, but we shall not dwell on these here; see for instance Antman
and Warner [4].
(2) There are various ways to introduce moving triads in order to rep-
resent the deformed position R; see Green [29], Antman and Warner [4],
Antman [1], Reissner [89], Hodges [22], Kane et al [41], Simo [100], Mars-
den, Krishnaprasad, Simo [101] etc. We will essentially follow the presen-
tation of Wempner [108], which is based on the same philosophy as in [29],
[4]. This will be explained in due course and the differences between the
possible choices will be discussed. At this point it is sufficient to say that
we do not, at the onset, assume that the there is no distortion within the
plane of cross section. As a result, the vectors E 2 , E 3 , which span the
deformed cross section, are not necessarily perpendicular; neither are they
necessarily perpendicular to Eb due to shearing. In [22], [100], [101] and
in many other publications, where a director approach (as in the Cosserat
continuum) is taken, the moving triads representing the deformed body
are taken to be orthogonal. Orthogonal triads are more appealing from the
point of view of geometrically exact theories of rods, as the configuration
space can entirely be represented by translations and rotations. However,
whenever it comes down to explicit relations between strains and displace-
ments, additional restrictions on the admissible configurations have to be
introduced anyway and then there is no apparent advantage of one method
over the other. For this reason we want to choose a representation in which
the role of the rotation of infinitesimal elements and the shear angles are
most transparent and one which leads to a practical set of equations of
motions under a sufficiently general set of assumptions.
1. NONLINEAR ELASTIC BEAMS 51
(3) In order to define the strains at the reference curve, there are also
at least two possibilities. One can either take the triad Ei itself, as we
will do here, and measure the angles between the plane of cross section
spanned by E 2, E3 and the tangent to the deformed reference curve, E!,
or one can work with the triad E 2, E 3, E2 X E3 = N and measure the
angle between N and the tangent. We will see that in the case where
there is no distortion within the plane of cross sections, these two concepts
coincide. The choice made here seems to have the advantage of revealing a
more transparent relation between the Green-St.Venant- (Lagrange) strain
tensor to be introduced below and the reference strains just mentioned.
(4) As we are also interested in superimposed rigid body dynamics, as
in Ballieul and Levi [5], [6], Kane [41] and others, we also want to present
a frame indifferent setup of strain and stretch tensors. The equations of
motion will first be derived within the context of spatial variables, that is
with respect to a moving triad, and then, in order to obtain the differential
equations in terms of displacements and (local and global) rotations, those
equations are either to be transformed to material coordinates by means of
a pull back, see [101], or to convected coordinates using the Piola transform,
see [101], [18]. D
Let us now attempt to make the notion of strain explicit. We have,
Note that in the presence of an initial twist the triad gi is not orthogo-
nal. Introducing the length Si of an undeformed xl-coordinate line passing
through (X2' X3), we have the increments
dS l = Ir,lldx!, dS 2 = dX2, dS3 = dX3.
With this notation we are able to write down the strain components as
follows:
1 8R 8R 8r 8r
Cij := - ( - . - - - . -).
2 8s i 8s j8s i 8s j
Note that the reference configuration already includes curvatures and twist.
The matrices (:~ . :~) and (Z! .
Z~) give rise to Riemannian structures
on the reference body and the deformed body, respectively, and the strain
C just defined is then the Green-St. Venant strain tensor, see Ciarlet [18]
or Antman and Warner [4]. This definition of a strain has an obvious
physical interpretation, as it compares the length of an infinitesimal line in
the deformed configuration with the respective length in the undeformed
configuration. It is also important to note that this strain measure is
obviously invariant under rigid body motions.
Now
52 III. NETWORKS OF THERMOELASTIC BEAMS
where
dX2 __ dX3 __ 1, dS l 1
-d =lr11=(gigi)'2.
dS 2 dS 3 Xl'
Hence the components of the strain tensor can be written as
(1.2)
The last three strains are strains within the cross section. Those are usually
neglected in theories of thin beams.
The deformation of r into R(, t) will be considered as a succession of
two motions: (1) a rotation r 1 carrying the triad ei(xd to an intermediate
triad ei(xb t), followed by (2) a deformation into the non-orthogonal triad
Ei(Xb t). The two triads ei and Ei then differ on account of a strain to
be specified below. We choose to orient the intermediate (right-handed)
triad ei, which serves as a moving orthonormal reference frame, such that
holds. Note that the triad e is neither the canonical Frenet-system nor
the one employed in, e.g., [100]. Nevertheless, it is uniquely determined by
property (1.3).
A strain , related to the deformation carrying the triad ei to the triad
E i , is defined by
e1 . E1 =: 1 + 11,
e2 . E2 =: 1 + 22,
e3' E3 =: 1 + 33,
(1.4)
e1 . E2 =: 212,
e1 . E3 =: 213,
e2 . E3 =: 23
(1.6)
However, the latter is exactly the strain Cij(X1, 0, 0) at the deformed refer-
ence curve. 0
We proceed to express the motion carrying the triad ei into the triad
E i . We have, for a certain orthogonal matrix r 1 (x),
ei(xt} = r1(Xt}ei(Xt} = r1(Xt} r O(xt}e?
and
+ (r 1(xt)88 rO(X1)rO(X1)t)r1(xde(xt).
Xl
Therefore, upon defing the skew-symmetric matrices
{K2 A A A A
(1.8)
-K3)
-T ,
o
where we have omitted the arguments. We also define the axial vector
satisfying Ow = 0 and
(1.9)
el = (cos 8 2 cos 8 3)el + (cos e 2 sin 8 3)e2 - sin e 2e3,
e2 = (sin 8 1 sin 8 2 cos 8 3 - cos 8 1 sin ( 3)el
+ (cos 8 1 cos 8 3 +sine l sin 8 2 sin ( 3)e2 + (sinSl cOS( 2)e3,
e3 = (cose l sin 8 2 cos 8 3 + sine l sin ( 3)el
- (sine l cos 8 3 - cose l sin 8 2 sin 8 3)e2 + (cos 8 1 cos ( 2)e3.
This makes it possible to express the curvature and twist, with respect to
1. NONLINEAR ELASTIC BEAMS 55
In addition to that, we can write down the exact relations between the
reference frame ei and the tangent frame E i .
(1.11)
El = (1 + 11)[(cos8 2 cos 8 3)el + (sin 8 3 cos 9 2)e2 - (sin 9 2)e3],
E2 = [- sin 8 3 cos 8 1 + sin 9 1 sin 8 2 sin 8 3
+ 221 cos 8 2 cos 9 3
- 22(sin 9 3 cos 8 1 - sin 9 1 sin 8 2 cos 8 3)
+ 23(sin 8 1 sin 8 3 + sin 9 2 cos 9 1 cos 8 3)]el
+ [cos 8 1 cos 8 3 + sin 8 1 sin 8 2 sin 9 3
+ 221 sin 9 3 cos 9 2
+ 22( cos 9 1 cos 9 2 + sin 8 1 sin 8 2 sin 8 3)
- 23(sin 9 1 cos 9 2 - sin 8 2 sin 8 3 cos 8t}]e2
+ [sin 8 1 cos 8 2 - 2[12 sin 8 2
+ [22 sin 8 1 cos 8 2 + [23 cos 8 1 cos 8 2]e3,
E3 = [sin 9 1 sin 8 3 + sin 8 2 cos 8 1 cos 9 3
+ 231 cos 9 2 cos 9 3
+ [33 (sin 8 1 sin 8 3 + sin 8 2 cos 9 1 cos 8 3)
- [23 (sin 8 3 cos 8 1 - sin 8 1 sin 8 2 cos 83)]eb
+ [- sin 8 1 cos 8 3 + sin 8 2 sin 8 3 cos 8 1
+ 231 sin 8 3 cos 8 2
- 33(sin 8 1 cos 8 3 - sin 82 sin 9 3 cos 8 1)
+ 23(cos8l cos 9 3 + sin 9 1 sin 9 2 sin 8 3]e2
+ [cos 9 1 cos 8 2 - 231 sin 8 2
+ 33 cos 9 1 cos 8 2 + 23 sin 8 1 cos 8 2]e3.
These calculations amply demonstrate that in order to obtain any practi-
cal set of equations of motions we have to approximate the rotations. There
56 III. NETWORKS OF THERMOELASTIC BEAMS
are two stages of approximation, which have been termed finite rotation,
and moderate rotation, based on power series expansions, to be found in
the literature. A finite rotation approximation to any given rotation de-
scribed in terms of the three angles above amounts to keeping all quadratic
terms in the angles, while a moderate rotation, which has also come to be
known as infinitesimal rigid displacement (see Ciarlet[18, Section 6.3]), re-
flects the linear terms only. Another way of thinking of moderate and finite
rotations is to consider a one parameter group of rotations in R 3 generated
by a skew symmetric matrix O. The linear approximation is precisely the
moderate rotation, while the quadratic approximation coincides with the
finite rotation. The finite rotation approximation to (1.9) is as follows. We
have
+ r818 3 - 1 (2
K38 28 3 - 2K2 8 1 + 8 32) ,
1
E1=(1 + tll - 2(8~ + 8~))e1 + (1 + tll)83e2 - (1 + tll)82e3,
E 2 =( -83 + 8 1 8 2 + 2t21 + t2382 - t2283)e1
1
+ (1 + t22 - 2(8~ + 8~) + 2t2183 - t238 t}e2
+ (8 1 + t23 - 2t21 8 2 + t228 1)e3,
E3=(82 + 8 18 3 + 2t31 - t23 8 3 + t338 2)e1
+ (-81 + 8 28 3 + t23 + 2t31 8 3 - t338 t}e2
1
+ (1 + t33 - 2(8~ + 8~) - 2t3182 + t238 1)e3.
1. NONLINEAR ELASTIC BEAMS 57
'191=81,
{ '192=82 + 2t31 + 28 1 e21,
(1.13)
'193=83 - 2e21 + 281e3b
{} :='19 1e1 + 'I9 2e2 + 'I9 3e3.
These angles are now being interpreted as the global rotations. The shear
strains can then be expressed as
It is obvious from these relations that the shear strains vanish if and only if
the angles 8 i , 'I9i coincide, respectively, or, what is the same, if the normal
N to the cross section coincides with E 1 .
In order to complete the representation of the reference strains in terms
of the angles above and the displacements Wi := W . ei , we compute
58 III. NETWORKS OF THERMOELASTIC BEAMS
We rewrite G 1 . G 1 in (1.15) as
(1.18)
1
f31 =2(W3,1 + t?,-+ /\;3 W 1 + TW2
+ 'l?1('1?3 - W2,1 -/\;2W1 + TW3)),
and
K2 ='I?3,1 + /\;3'1?1 + T'I?2,
(1.19) { K3 = - 'l?2,1 - /\;2'1?1 + T'I?3,
T ='191,1 - "'2'192 - "'3'193,
60 III. NETWORKS OF THERMOELASTIC BEAMS
r:= Ii 2dA.
The second identity for J follows from the double symmetry of the cross
section. Note that these quantities are still functions of Xl since A(XI)
varies with Xl. From now on we assume in addition that
the cross section dimensions are negligible compared to the initial
radii of curvature (Le. l/K-i,l/T);
the material is linearly elastic and transversely isotropic.
The free energy per unit volume ~(Xb X2, X3) may then be taken to be
of the form
where '\, /-L are the Lame moduli and tr() indicated the trace of a matrix.
2. EQUATIONS OF MOTION 61
(2.5)
The resultant force, F, and couple, M, on a cross section are given by
Let fo, Co, to denote the external body force, body couple, and surface
traction, respectively, per unit undeformed volume and surface, respec-
tively. Also, denote by Po the density per unit reference volume and by
mO(x1) := POA(X1) the density per unit reference length (m := mOdx1 J:
is then the total mass). Then the body force P and the body couple C are
given by
Is
+ (R - Ito) x todS,
where S is the curve circumscribing the cross section. As we are mainly
interested in forces and couples applied to the end points of the beam we
set fo, Co, to equal to zero. We consider a small segment (slice) OQ of the
beam located at a point Q on the undeformed centerline with length 2Llx1
in the direction of e1 . The traction on the cross section at Q - LlX1 is
-t(Q - 6X1) and the corresponding traction at Q + 6X1 is t(Q + 6xt}.
Let us denote the end surfaces of the slice OQ by E Q , and the surface
tractions by g. The body force density on that volume element is provided
by the inertial force density -R,tt. Let us finally denote by A: B the inner
2. EQUATIONS OF MOTION 63
n
for all sufficiently smooth vector fields 8R: - t IR3 (see Ciarlet [18]), in
fact, for all variations 8W, MJ, 8, satisfying some specified geometrical
boundary conditions. It is also possible to express the principle of virtual
work in terms of the second Piola Kirchhoff tensor S and the St.Venant-
strain tensor , since T : V'8R = T : 8G = S : G T 8G = S : 8. Using the
notation introduced above we find that (2.9) is equivalent to
[Q+Ll
JQ-Ll
[!r [ {t. 8R,1 + (S12G l . 8R,2 + S13G 1 .8R,3)}dA]dx1
JA
+ Jr [ t . 8llilA - Jr [ t . 8llilA.
JA(Q+Ll) JA(Q-Ll)
It : = el . JL (-poR,tt)4>dA
= e1 . JL TV'4>dA,
D:= JL t4>dA,
~ := Eh . !L (S124>,2 + S134>,3)G1dA.
We obtain the following approximations for the variations of R:
8R = 8W + M) x R + 8'el,
8R,l = 8W,1 + 8-&,1 X R + 8-& X R,l + 8,,1el + 8,4>(K2e2 + K3e3),
8R,2 = 8-& x e2 + 8,,2el, 8R,3 = 8-& x e3 + 8,,3el,
it,l = -x2(K2e l - Te3) - x3(K3e l + Te2) + "lel + ,(K2e2 + K3e3).
64 III. NETWORKS OF THERMOELASTIC BEAMS
F,l +P = 0,
(2.10) { M,l +el x F + C = 0,
DI,1 - { + II = 0.
(2.11)
with given boundary forces F and couples 1\1: (see [108]). In this formu-
lation, equations (2.10) are the classical equations of motion. Boundary
forces are not, of course, to be prescribed where geometric boundary con-
ditions are imposed.
To equations (2.10) we have to add an equation governing the flow of
heat. We have assumed that the body has, at its reference configuration, a
homogeneous reference temperature To. Introducing the heat capacity Cv
and relying on Newton's law for the heat flux, we obtain
(2.12)
In order to explicate the system of balance and heat equations we could use
the moving reference frame ei, as it is done in Wempner's book. However,
as we have expressed the deformation of the initial triad ei into the triad Ei
in terms of the initial triad, it is most natural to express all the quantities
in terms of the triad ei, e.g., F = I:i Fiei, M = I:i Miei,
M 2 ,l + K,2Ml - r M3 - iJ2 F l - F3 + C2 = 0,
+ K,3Ml + r M2 -
M 3 ,l iJ3Fl + F2 + C3 = 0,
Dl,l - ~ + II = 0.
(2.17)
REMARK 2.1. The equations (2.14) together with the constitutive rela-
tions (2.16) and the expression of forces and moments (2.17), (2.17) consti-
tute a complete set of partial differential equations governing the motion of
a general beam in terms of displacements and rotations, if complemented
by the relations (1.18), (1.19), the heat equation (2.12) and appropriate
boundary conditions (2.11), (2.13). 0
(2.20) 1 ( 2
2G(I _ J) G(I - J)19 1,1 - (Er191,1l),1) .
3. ROTATING BEAMS 67
This is done in Kane et al [41J. It is obvious that in this case the first
approximation, based on the potential energy (2.3), results in a fourth order
differential equation governing the torsion angle f}l, whereas the second
approximation, based on (2.3) with (2.20) as the only term to represent
torsion, results in a sixth order differential equation in f}l. In this paper
we either take warping into account while keeping the body force h, to the
extent that the approximations just mentioned do not hold, or we neglect
the warping entirely.
3. Rotating Beams
In this section we consider the motion of a flexible beam, as described
in the previous section, which is supposed to be attached to a rigid body.
The rigid body together with its attachment is assumed to undergo a rigid
body transformation: a translation and a rotation. We do not take into
account warping of the cross section. There is a considerable literature
dealing with precisely this problem. See Antman and Nachman [3], Kane
et al [41], Ballieul and Levi [5], [6], Delfour et al [26] and many others.
In most papers, however, attention is devoted to either special kinematical
assumptions or special material laws. The equations which are derived here
are most easily compared with those obtained in [5].
Let mb denote the mass of the rigid body and h its moment of inertia.
For simplicity we assume that the center of mass of the rigid body is in the
origin of the body frame. We consider planar rotations about the e2 -axis,
only. That is, we consider a rotation
COS(O(t)) 0 - Sin(O(t)))
II(t) = ( 0 1 0 .
sin(O(t 0 cos(O(t
We also consider a translation in that plane
R = D(t)R + y(t).
The velocity R,t in the inertia frame is given by
where
J:= (00-1)
0 0
1 0
0
0
.
DR := R,t + OJR,
so that the velocity in the inertia frame is given by
n,t = TIDR+Y
We can now express the body forces and couples (2.8) as follows:
Ji
(3.2)
C = Po('R- - 'R-o) x (-'R-,tt)dA
(3.3)
We note that the deformation gradient in the rotated frame is TIG, and
therefore the corresponding stress vector is TIt. Hence the forces and mo-
ments in the rotated frame are TIF, TIM, respectively. We also note that
TIel x TIF = TI(el x F). Therefore, the balance equations (2.10) with
3. ROTATING BEAMS 69
(3.6) In
11 Po'R- X'R-,tt dX l dx 2dx 3
In
= 11 PoD(Ro + X3 E 3) x (D(D2Ro + X3 D2E3) + y)dx
As Ro = xel + W, we find
- 29W3 ,t - 9 (WI + xel
2 2
D Ito = (W1,tt - 9W3
.. . 2 ..
+ (W3,tt + OWl + 20WI,t - 0 W3 + xO)e3.
The first equation of (3.5) can be rewritten as
and
(3.9)
(mb + m)Yl
=- 10 {cos(O)(Wl,tt - 2
oW3 - 20W3,t - 0 (WI + x))
.. . 2 ..
- sin(O) (W3,tt + OWl + 20Wl ,t - 0 W3 + XO)}dXl,
10 2
(mb + m)jh = - {sin(O) (Wl,tt - OW3 - 20W3,t - 0 (WI + x))
.. . 2 ..
+ cos(O) (W3,tt + OWl + 20Wl,t - 0 W3 + XO)}dXl,
(h + i)O = - I~ mo {- W3 (Wl,tt - OW3 - 20W3,t)
+(Wl + x)(W3,tt + O(WI + x) + 20Wl,t) + 133 '19 2,tt
+W3(COS(0)Yl + sin(0)Y2)
-(x + W l )( - sin(O)Yl + cos(0)Y2)}dxb
1 2 1 ( 2 2
(3.10) T = "2hO +"2 Jo I('I92,t + 0 )dXl
where we take into account that Fl() = o. As the origin is assumed fixed,
we neglect the balance of total linear momentum. Then the set of equations
(3.8), (3.9) reduces to
F3,1 = m oW3,tt - 02W3 + xO,
{ M 2,1 2 2 2 ..
(3.11) - TO ( - x )112 - F3 = m oI 33 (11 2,tt - 0),
(Ib + i)o = -mo I~(xW3,tt + 133 112,tt)dx.
If we further differentiate the second equation with respect to x, we can
eliminate F3 by and obtain instead
mo2 2 2 )
(2 2 ..
M2,11 - 0 ( - x )112 ,1 = mOW3,tt + PO!a3 112,ltt - 0 W3 + xo.
4. NONSHEARABLE NONLINEAR 3-D BEAMS 71
In the case where there is no shear strain (e3l = 0), the Euler- Bernoulli
hypothesis, we have '!92 = -W3,1 and M2 reduces to M2 = -E133 W3,1l.
Then the equation above is equivalent to
2 ..
mo W3,tt - Po 133 W3,lltt + E133 W3,111 1 - 0 W3 + xO
( mo2 2
- TO ( - x 2)W3,1 ) 1 = o.
,
Pi = -moUri, i = 1,2,3,
.. f f
Cl = -Po1iJl, C2 = Po133W 3, C3 = -Po122W 2
We make use of the approximations (2.16) and insert the relations above
into the balance laws (2.14). Then we differentiate with respect to Xl the
second and third of the equations representing the balance of angular mo-
mentum, and insert the second and third equation of the balance of forces,
respectively. The result is a set of five equations governing the longitudinal
displacement WI, the two lateral displacements W 2, W3, the twist iJ l and
the warping 'Y. This set has to be complemented by the equations governing
the temperatures Oi, and by boundary and initial conditions.
Equations of Motion
72 III. NETWORKS OF THERMOELASTIC BEAMS
[Longitudinal motion]
(4.1 ) .. = [A('
mOWI E WI +"21 (W2')2 +"21 (')2)]'
W3 - 0'.0,1.
[Vertical motion]
[Warping]
Upon taking the average and the first moments with respect to the Xi axis,
we obtain a set of three 1-d heat equations for the variables Oi as follows:
[Heat conduction]
Wi(O) = 0, i = 1,2,3,
(4.7) { W~(O) = WHO) = 0,
1?1 (0) = 0, ,,(0) = 0.
[Insulated boundary conditions]
(4.8) O~(O) = 0, i = 1,2,3.
[Dynamical boundary conditions]
(4.14)
3 t
'"' f
U+ f:t 10 Pocv (}2d
2To i x.
The reason for this is the fact that the energy (2.3) is not purely quadratic
as it contains terms mixed in strains and temperatures. That energy, how-
ever, is easily seen to be bounded above by some constant times the poten-
tial energy just defined. Therefore, dissipativity with respect to the new
"energy" is transferred to the physical energy. If only planar isothermal
motion is considered, where we may look at the 1-2 or 1-3 plane (with
19 1 = 0), the system above reduces to the one we have discussed in Lagnese
and Leugering [47] which, in turn, reduces to a one-dimensional analog
to the von Karman plate if the inertia of the longitudinal motion is re-
placed with a constant load. If, finally, we neglect the rotational inertia
(I22 = 133 = 0), the model reduces to an Euler-Bernoulli type system.
Then the second and the third equations can be solved by variation of
parameters to give
Pocih = O~ - !
OTo {W{ + ~{(W~)2 + (W~)2)).
Equations (4.16)-{4.19) are now integro-partial differential equations, de-
scribing a thin thermoelastic beam in three dimensions. Upon replacing
the convolution kernels e-I';t by more general functions, for instance com-
pletely monotone kernels, we can refine the assumptions made in order to
arrive at (4.15). Also, this could be viewed as a particular kind of a vis-
coelastic, in fact thermo-viscoelastic beam model. Furthermore, instead
of the simplifying assumption that the original heat equations (4.6) are
replaced with (4.15), we could replace Fourier's law with a more general
76 III. NETWORKS OF THERMOELASTIC BEAMS
heat flux law accounting for a long memory of the material, and end up
with a even more general thermo-viscoelastic model. If then kernels are
chosen to approximate the Dirac distribution, we arrive at what has come
to be known as the slow flow approximation, or the Kelvin-Voigt viscoelas-
tic damping. There are other models for internal damping, in particular
the so called structural damping of Chen and Russell [16] or, as an almost
equivalent analog to that, the shear diffusion damping model described in
[90], which can be introduced here.
1 1
t11 = W{, t21 = 2(W~ -1?3), t31 = 2(W~ + 1?2),
F1 = EAW{, F2 = GA(W~ -1?3), F3 = GA(W~ + 1?2),
M1 = GI1?~, M2 = EI331?~, M3 = EI22 1?;,
Pi = -mo Wi, Ci = -moi9i.
This leads to the following system of 3-d Bresse-Timoshenko beam equa-
tions.
Equations of motion
REMARK 5.1. It is obvious from (5.1) that there is neither a linear cou-
pling between longitudinal and vertical/lateral motion nor between any
displacement and twist. D
13 = ~('I?2 + W~ + 1I:3W1).
F, M are given by (2.16), and C, P reduce to
Pi = -moWi ,
C2 = -PoI33 J2 , C1 = C3 = 0.
We obtain the following system of equations.
Equations of motion
[Longitudinal motion]
..
mOhW1 = [Eh (W 1, - 1 2) - o:(h ]'
11:3 W3 + 2'1?2
- 1I:3Gh('I?2 + W~ + 1I:3Wd + 0:[11:303]'
[Vertical motion]
.. "
mOW3 = [Gh('I?2 + W3 + 1I:3Wd] + 11:3 [Eh(W1, -1I:3W3 + 2'1?2)
1 2
- 0:01]
[Shear motion]
[Heat flux]
(6.1)
POCV~l = Or - aTo $t {W{ -
{ POCv03 ~3 W3 + ~t?n,
= 01- aTot?~.
Again, this set of equations has to be complemented by appropriate bound-
ary and initial conditions. We choose the situation analogous to the pre-
ceding section where, at x = 0, the beam is clamped and insulated, and at
x = f it is driven by forces, moments, and heat supplies.
Boundary conditions
[Geometric boundary conditions]
(6.4)
Initial conditions
and integrate the second of these equations. Then we obtain the system of
integro--partial differential equations
[Longitudinal motion]
[Vertical motion]
[Shear motion]
8{,
PoC" 01 = 0"1 - o:To at WI + 2"1((W2')2 + (')2)}
W3 ,
POC" 02 = e"2 + o:.Lo Wo'2'"
rJ"I
[Lateral motion)
[Shear motion]
[Heat flux]
. "
POC v Ol = 01 -
() { ,
aTo 8t Wi - "'3 W3 + 2"'1 I?22} ,
POCj)3 = O~ - aTOt?;.
[Vertical motion]
[Shear motion]
[Shear motion]
Rotating beams
See Section 3. For each particular beam model we can easily write
down the corresponding rotating beam models: the fully nonlinear and the
linearizations (including dynamic stiffening and damping).
8. Networks of Beams
Once we have achieved a sufficiently rich theory for a single beam, we
may proceed to develop models of networks of such beams. This is the
main purpose of this section. As with the reference arc in the case of a
single beam, we have now a connected graph, consisting of a set of smooth
open arcs which are connected through multiple nodes, labelled with N,
and which mayor may not end at a simple node. This network of arcs
serves then as the reference net. It would be possible to impose a static
prest retching on this net. Then additional equilibrium conditions would
have to be taken into account. This has been carried out for a network of
strings in Schmidt [93], and for Euler-Bernoulli beams in Leugering and
Schmidt [58]. It turns out that a prestretching leads to a coupling be-
tween axial strains and bending. For simplicity, we assume that no such
prestretching is present here. In the case of strings and Euler-Bernoulli
beams it is sufficient to consider the centerline only, which makes the the-
ory much simpler. In all the other cases, where we deal with shear, a joint
(multiple node) between two or more beams is considerably more compli-
cated to model. However, apart from the multiple nodes each beam satisfies
the requirements of the preceding sections, in particular, the balance laws.
This implies that each beam, individually, satisfies the equations of motion
which we have derived so far. This also is easily seen from the viewpoint
of energy principles; the potential and kinetic energy have to be replaced
with their sum over all members. Coupling between longitudinal, transver-
sal, lateral and torsional motion occurs at the multiple nodes. It is, for
instance, obvious that in the case of a carpenter's square, the longitudinal
displacement of one beam translates to a lateral displacement of the other
and vice versa. Also, a twist of one beam results in a bending moment at
the other beam, and vice versa.
8.1. Geometric Joint Conditions.
8.1.1. Rigid Joints. Here we consider two or more beams (of length f i )
of the nature described in the previous sections which are supposed to end
at a joint. Let I(N) denote the set of indices i such that the i-th beam
meets at the node N. When talking about a connected structure at all,
the first condition which has to be satisfied at any joint, regardless of its
particular nature, is the continuity relation
(8.1) Wi(N) = Wj(N), Vi,j E I(N).
In addition, we always assume that the temperatures flTi are all the same
at each joint.
(8.2) o{(N) = O~(N), Vi,j E I(N), k = 1,2,3.
In order to specify other features of the joints, we have to distinguish
between geometrical requirements and force/moment balances at the joints.
84 III. NETWORKS OF THERMOELASTIC BEAMS
Two different types of joints will be investigated: rigid joints and pinned
joints. Rigid joints are usually encountered in frames, whereas pinned joints
occur in trusses. In this section we discuss rigid joints. A joint will be called
rigid if the cross sections of the beams connected to each other at the node
N undergo the same "moderate rotation," characterized by the vector {J,
i.e.,
This condition allows the joint to rotate as a whole and to twist, but the
shape of the joint before and after deformation is forced to be same. If
there is no shear involved, rigidity of a joint means nothing more than
the tangents to the deformed reference curves span the same angle as the
tangents to undeformed reference curves.
It would be of great interest to derive a mathematically precise model of
joints that are essentially rigid, but where the angle between two consecu-
tive beam elements is allowed to switch if the deflection exceeds a certain
threshold. Joints of this type are of common occurrence in satellite struc-
tures, where arms or solar panels are to be unfolded in space.
It should be noted that flexible joint conditions, such as the ones de-
scribed here, are usually not reflected in the engineering literature on frames
and trusses. The reason for this is that the equations of motions are formu-
lated in a modal representation to begin with. In effect, this means that
a joint of two flexible elements is replaced with a joint described by the
two first finite elements of the connecting beams, which of course behave
like rigid bodies. As has been demonstrated by Howard in his recent thesis
[38], the transmission of elastic energy in a truss or a frame can only be
fully understood if the flexibility of the links is taken into account at the
joints. He shows that (and how) bending energy of a truss structure (or a
frame) is converted to axial energy, and vice versa, through the joints.
This implies that the first joint condition (8.1) can be written as
Although the first set of conditions obviously coincides with intuition, the
second set deserves further explanation. We focus on the Euler-Bernoulli
framework first. Here we find (elj = 0, j = 2,3)
tEI
One may, however, treat the joints N as mass points, so that IN is taken
to be zero. The inertia part in (8.4) is not to be confused with the rotary
inertia of the cross section which is present in the Rayleigh beam model.
One may suppose the joint to be massless if the beams are considered as
welded to each other.
REMARK 8.1. We note that the node conditions obtained in this case
are in accordance with the results of Le Dret [56] if the inertia term in
the equation governing the longitudinal motion in the vertical beam is
neglected, and if the resulting longitudinal stress is expressed in terms of
the weight of the vertical beam. Also, Le Dret allows the vertical beam
to move out of the plane causing a twist of the horizontal beam expressed
in terms of the weight and of surface tractions. The corresponding node
condition is obtained from our more general ones by neglecting the inertia
term in the equation governing the torsional motion and by expressing the
twist U~, which appears in the node conditions, in terms of appropriate
forces and couples.
iEI
88 III. NETWORKS OF THERMOELASTIC BEAMS
while (8.11) remains the same. These are the dynamic node conditions for
the Bresse-Timoshenko system.
(8.15)
in the 2-d case. As for the Bresse-Timoshenko system we obtain the dy-
namic node conditions (8.10) together with
(8.16)
9. ROTATING TWO-LINK NONLINEAR SHEARABLE BEAMS 89
and
= lli~ + Yi
'R.i
We write ~ instead of Roi. In particular ri = (x, 0, O)T and ~ = ri +Wi.
We require that the beams are connected at the joint, in the undeformed
and the deformed configuration:
rl(lt) = r2(0),
'R.I(lt) = 'R.2(0).
This amounts to
and
hence
We proceed to write down the kinetic and strain energies. The kinetic
energy is given by
90 III. NETWORKS OF THERMOELASTIC BEAMS
E:=T+U, E=O.
We remark, however, that using this principle one has to take special care
of Coriolis forces which do not contribute to work. We first calculate (h
We have
where a~i etc. are Frtkhet derivatives. As for the time derivative of the
kinetic energy, we find
We denote the last integral, which stems from the elastic deformation, by
9. ROTATING TWO-LINK NONLINEAR SHEARABLE BEAMS 91
We require that
and
The terms involving ih, ih are taken together with the corresponding terms
coming from the expression for the time derivative of the kinetic energy.
The equation E = 0 has to hold for all test functions satisfying the bound-
ary conditions and the continuity condition. We obtain the following set of
92 III. NETWORKS OF THERMOELASTIC BEAMS
. 1ll .
(Ir1 + h1)fJ1 +
o
h0:1 dx =< W 1(1),J III II2 a- (0) >
T T aU2
C2
2R lIT.. aU1
Po D 1 1 + 1 Y1 = aR1 '
2 T aU2
POD2R 2 + II2 II1r 1(1) = aR2
1 a-c1
II aU1 ( ) = II aU2 (0)
1 2 a-c2 ,
o:~ (1) = 0:~(0).
It is straightforward to show that the case where ()1 = ()2, m r 2 =
0, Ir2 = reduces to the rotating beam model described in Section 3.
One can also derive from this general set of equations an Euler-Bernoulli
type rotating two-link beam model, and one with dynamic stiffening. We
refer to Khorrami [42] where a rotating two-link Euler-Bernoulli beam has
been developed. His equations however, do not reduce to a single rotating
9. ROTATING TWO-LINK NONLINEAR SHEARABLE BEAMS 93
beam equation if the rotation angles are the same and the mass at the joint
is not considered. Rather, in that situation the second beam may oscillate
independent of the first in his model.
CHAPTER IV
(1.3)
where u j (t) is a preassigned control function. We shall later also allow
conditions of Neumann type. The form that these should take will emerge
from the variational argument which follows.
Let the system have associated with it kinetic and potential energy func-
tions of the form
i=l 0
.
t)12 dx,
and
1. THE GENERAL MODEL 97
where
Pi, Qi, Si and ~ are continuous Pi x Pi matrix valued functions
of X;
Pi and Qi are symmetric and positive definite;
Si is either positive definite symmetric or the zero matrix.
One can also introduce the work functional
where .:IN c .:J \ [.:Je U .:JD] is the set of nodal indices at which we wish
to impose a Neumann control uj(t). We apply Hamilton's principle to the
Lagrangian functional
(1.7)
(where "t" denotes the transpose of a matrix). One then considers arbitrary
variations satisfying the geometric conditions and uses the equations (4.7)
to get the variational identity
(1.8)
where "p(t) and "p'J(t) are arbitrary smooth functions taking their respec-
or
tive values in lRq, and in ker( C'J). Evidently 1)1 = "p or 0 corresponding
to the two cases. The last variational identity then yields the following dy-
namic conditions:
(1.9)
where
and
(1.13)
In fact there are many ways in which Dirichlet and Neumann conditions
could be combined, also at multiple nodes. To insist on full generality
would further complicate our notation and we choose not to pursue this
theme further.
where the Dji are rj x Pi matrices. One can then get dynamic node condi-
tions from the variational argument if one makes use of a basis Vjk = (V;k)
of solutions to the homogeneous linear system
L Djiv i = o.
iEIj
We do not work out the details here but refer to the third variant on
the example of Subsection 2.5 as well as Remark 5.3, which illustrate this
generalization of the multiple node conditions.
REMARK 2.1. In the above model it is natural for the matrices to have
no spatial dependence. When the network is subject to gravity the equilib-
rium configuration will "sag" and then the matrices entering the linearized
equations do depend on x (see Section 7, Chapter II and [95]).
r = (WI, W 3 , -0 2 ),
p 2 = diag(mo, mo, POi33), Q2 = diag(Eh, Gh, EIa3),
(2.1)
R = (~ ~ ~), S = O.
000
The network is planar if eil and ei3 span the same planar subspace of IR3 for
all i E I, so that all motions occur in that plane and ei2 = n can be chosen
independently of i as the normal to the plane of motion. The geometric
constraints require continuity of the displacements Wi = Wileil + Wi3ei3
of the center lines at multiple nodes, and for" rigid joints" one also demands
100 IV. A GENERAL HYPERBOLIC MODEL FOR NETWORKS
for j E .IN,
for j E .JF.
These can be decoupled into planar and normal components with the aid of
any pair of orthonormal vectors el and e2 which are chosen independently
of i and span the network plane. One gets
and
(2.4)
Physically, at multiple nodes, these two conditions assert that the resultants
of bending moments and forces both are zero or balanced by those of the
imposed forces.
If a multiple joint, indexed by j, is pinned rather than rigid there is
no constraint on the angles iJ ,2 (X'3 , t) and the appropriate form of the
continuity condition is given by (1.1) with C'3 = EtE, where E, is again
the 3 x 3 matrix having ed, e,3 and n as successive columns while E is a
2 x 3 matrix having as its rows the vectors el and e2 which were introduced
above. In this case the dynamic conditions (1.9) reduce to (1.3) alone, while
(1.10) is no longer vacuous and yields
(2.5)
In this case at the pinned joint all the iJ,2,x are free.
It is of course possible that at some multiple nodes the joint is rigid
while at others it is pinned.
2. SOME SPECIAL CASES 101
p2 = diag(mo,mo,mo,pI,pI33 ,pI22 ),
Q2 = diag(EA, GA, GA, GI, EI33 , EI22 ),
0 0 0 0 0 0
(2.6) 0 0 0 0 0 -1
0 0 0 0 1 0
R= 8=0.
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
At rigid joints the geometric multiple node condition (8.1) and (8.3) of
Chapter III can be written as
for j E .IN
for j E.JN '
and
For a pin-joint (or perhaps a ball-joint is a better term for a three dimen-
sional joint allowing all proper rotations) there is no geometric condition
102 IV. A GENERAL HYPERBOLIC MODEL FOR NETWORKS
One can use the same matrices C ij as before and obtain the following minor
variation of the conditions (2.3):
the beam the choice of spaces is made in accordance with Subsection 2.2.
In the absence of Neumann controls this leads to the Lagrangian function
(2.13) = iT ( L
o
2 [
i=l
[Qi r :x 6ri](x, t)
]X=lo
='
x 0
(2.15)
We then consider variations 6rl, 6WI and 6W3 which respect the first of
the above conditions and vanish at the other endpoints and set 6r2 = 0
104 IV. A GENERAL HYPERBOLIC MODEL FOR NETWORKS
{
Qlr;x(o, t) . el+ EhWl,x(O, t) = 0,
(2.16)
Qlr;AO, t) . e3 + Gh[W3,x + 1?2](0, t) = 0.
Similarly
Again 1?2 is not involved and one obtains the condition 1?2,x(0, t) = 0. Then
we take variations with 81?2 = and 8rl, 8r2, 8Wl and 8W3 having support
near the junction, to obtain from (2.13) that
This gives rise to the dynamic conditions (2.18) as before, together with
the additional condition
Note that in this last situation, unlike the previous one where the strings
are both joined to the center line of the beam, the variables rl(O, t) and
r2(0, t) together determine, and are determined by, the beam variables
WI (0, t), W3 (0, t) and 'i?2(0, t).
REMARK 2.2. The three different arrangements of beam and strings lead
to the same equations with different geometric and dynamic coupling con-
ditions where they are joined together. Supplementing these by boundary
conditions, controls and initial conditions one is led to systems for which
existence of solutions will follow from Section 3. The systems behave in
different ways (for example see Remark 5.3 for comments on exact control-
lability).
(3.2) ~ ( ~ ) = (~ ~) ( ~ ) +( ~/p )
106 IV. A GENERAL HYPERBOLIC MODEL FOR NETWORKS
(3.4)
which corresponds to the potential energy, as well as (r, s)v ,A = (r, s)v +
>'(r, S)H. It is not difficult to see that for>. > 0 the norm associated with
< r, r >V,A is equivalent to the standard norm. In some important cases
this is also true for >. = O. The following lemma gives some sufficient (but
not necessary) conditions for this to happen.
LEMMA 3.1. Suppose that either
(i) every Si is positive definite; or
(ii) there is at least one fully clamped node, and for each element labelled
i l , say, there exists a path denoted by [ib i 2, ... , irl leading along the net-
work elements labelled successively by ib i 2, ... ,ir and concluding at a fully
clamped node with the property that if jk is the index of the node between
the ik-l-th and the ik-th element the matrix C ikjk is invertible.
Then II r IIv is a norm on V and is equivalent to the product norm
(1.1) and the invertibility of each C ikik , we can conclude that indeed each
ri vanishes.
From the continuity of Qi(X), ~(x) and Si(X) it is obvious that I r Ilv~
C I r II, for some suitable constant C. The reverse inequality is obvious in
the case of the first hypothesis but more delicate in the alternative case. We
again use an idea from differential equations. Let Ti(X) be a (nonsingular)
Pi x Pi matrix-valued solution of
Then
Hence
Ti(x)ri(x) = Ti(O)ri(O) + fox [Ti(y)ri(y)l,xdy,
Since Ti(X) is continuous and continuously invertible one also has, after
redefinition of C,
The latter holds also with li and 0 interchanged. One again proceeds from
element to element along paths starting at a fully clamped node, getting
alternately estimates on Iri(O)12 or Ir i(li)12 and on J~i Iri l2dx in terms of
II r IIv. This gives the desired estimate II r I ~ C II r IIv. 0
REMARK 3.1. In most applications the elements will all be of the same
nature and the conclusion of the lemma will be valid whenever all the C ij
are invertible and there is at least one fully clamped node, since then the
second of the alternative hypotheses is trivially valid. For an exception see
Remark 5.3 concerning the mixed string/beam network.
Exactly as before one constructs a semigroup U(t) with a generator .A, and
uses this semigroup to solve (3.1) for initial data in dom(.A,OO) and control
functions which satisfy
One then has to make a priori estimates which allow the passage to
less regular data. This involves the same steps as before, but the explicit
computations need to be redone. For one thing the energy density is now
given by
and
1 (Xi tb t2) = [t2 e(x, t) dt.
1tl
We have to prove the analog of Lemma 4.2 of Chapter II which, in the
prosent context, takes the following form.
LEMMA 3.2. Let r be a smooth solution of
where
As a consequence
= - i 11
o
T
0
mP2r (r,x + Rr) dxdt + 11
0
[mP2r . (r,x + Rr)] ~ dx
t=O
1 {T {I
= 2" 10 10 [m,x p2 + m(p 2),x - mRtp2]r. rdxdt
- i [m
o
T
-2 IPrl 2
] x=1
x=o
dt +
11 [
0
mP 2r (r,x + Rr) ]
t=T
t=o
dx,
(3.11) -1 11 T
[[Q2(r,x + Rr)J,x - RtQ2(r,x + Rr)
By adding identities (3.10) and (3.11) and then rearranging we get (3.8).
The estimate (3.9) then follows by setting m(x) = -1 + 2xjl as before. 0
PROOF. The proof follows exactly the same lines as before. The follow-
ing calculation has to be made:
~~ (r) =:L
iEIJO
ti [P~ri . fi + Q~[r:x + ~riJ . [r:x + ~riJ] (x, r) dx
= :L [Q~ [r:x + ~riJ . ri] ::~i
iEI
Now
ri(Xij, r) = nb ri (Xij, r) + nijri(Xij, r) = C~Cijri(Xij, r) + nijri (Xij, r)
= C~1)ir(r) + nijri (Xij, r)
Substituting this back into the previous identity and taking into account
the form of the various operators involved we obtain
(3.13)
Since
l:vJr(TW ::; Cei(Xij, t) for any i E I j,
IN'jr(TW ::; C:L ei(xij, t),
iEIj
the remainder of the argument, using the estimate (3.9), goes through as
before. 0
One is led to the following general existence theorem:
THEOREM 3.1. Suppose that
(i) ro E V, ro E H, f E L 2 (0, T; H);
(ii) u j E Hl (0, T : IRqj) and satisfies u j (0) = 1)j ro for each j E J" D ;
(iii) u j (t) E L 2 (0, T : IRq,) for each j E .IN.
Then there exists a unique r E C([O, TJ; V) n C1([0, TJ; H) such that
1)jr(t) = uj(t) for j E J"D
and which satisfies the remaining requirements of (4.1) in the following weak
sense: for any </J E Vo the function (r( t), </J)H has an absolutely continuous
first derivative in t and almost everywhere satisfies
(3.14)
d2
dt 2 (r(t), </J)H + (r(t), </J)v (f(t)j p, </J)H + :L u j (t) . 1)j </J
jE.:J N
4. ENERGY ESTIMATES FOR HYPERBOLIC SYSTEMS 111
(4.3)
and
(4.4)
where
(4.5)
112 IV. A GENERAL HYPERBOLIC MODEL FOR NETWORKS
and
We let C(xo, to) and k(XO' to) denote the characteristics corresponding
to w and ~ respectively; these are described by functions x(t) and
4. ENERGY ESTIMATES FOR HYPERBOLIC SYSTEMS 113
~(t). Note that the characteristic curves can also be described in terms
of graphs t = t~(x) obtained by solving
(4.12)
PROOF. We apply the divergence theorem to the equation (4.3) over the
region ~(t) bounded above and below by the intervals {t} x [x+(t),x_(t)]
and {O} x [Xl, X2] and laterally by the characteristics C+(Xl, 0) and C-(X2' 0).
One has
(4.13) jjql(r,r,t,r,x)dXdr=
A(t)
f V.(-~Mw.w,e)dS,
M(t)
where v denotes the outward pointing unit normal to the boundary curve
8~(t) and ds denotes the element of (positive) arc length along the curve.
Along the top bounding interval one has v = (0,1), while along the base
interval v = (0, -1). Along the lateral characteristics one has, respectively,
v = [1 + W2]-1/2(=f1,W) and therefore, noting the third identity of (4.8),
v (-~Mw. w,e) is a positive multiple of
1 1
['2Mw. w + we] = '2 [Mw . w + wlwl 2 + wSw w].
By use of Schwarz's inequality, the fact that IMw wi ~ wlwl 2 and the
hypothesis that S vanishes or is positive definite, one concludes that along
the lateral characteristic curve segments v (-~Mw. w, e) is positive. If
we write out the line integral in (4.15) as the sum of four integrals corre-
sponding to the four different curve segments we can discard the positive
114 IV. A GENERAL HYPERBOLIC MODEL FOR NETWORKS
E(t;x+(t),:L(t)) :SE(0;Xl,X2)+ 11
.6(t)
ql(r,r,t,r,x)dxdr.
for some suitable constant C. The estimate (4.14) then follows from Gron-
wall's lemma. 0
II(x, t) =~
27rZ
1
'H
[zI - M(x, t)t 1 dz,
LEMMA 4.2. Let r(x, t) E W 2((0, l) x (0, T); IRP) be a solution of (4.1).
Suppose that ifS = 0, P is independent oft. Let a E [0, l] and C(a, T) be
associated with the functions x(t) respectively. If a is equal to 0 or l only
one of these characteristics is defined. If x+(t) meets the initial interval
{O} x [0, l] one has the estimates
and
and
8w 8w
(4.18) ~ =M-{} +l(r,rt,r x),
~L x "
with
Now
(4.19)
There are three contributions to the line integral. The contribution corre-
sponding to the characteristic curve C+ (a, T) is positive since along that
curve v = [1 +w2J-l/2( -l,w), so that V (-Mw+ w+, Iw+12) is a positive
multiple of the positive function
(T[Mw+ .w+J(a,t)dt-
Jo
r
Jx+(O)
Iw+(x,0)12dx::; jr{ q+(r,r,t, r,x) dxdr.
J
d+
Then (4.14) follows from Iw+12 ::; e, Iq1 ::; Ce and ~lw+12 ::; Mw+ . w+.
The estimate (4.15) is vacuous if S = 0; otherwise it is obtained by
applying the divergence theorem over ~+ to
{} {}
-{} [Sr.r]+ ~[Sr.r]=q(r,rt,rx),
x UL ' ,
where is chosen so that ::; W. One has q ::; Ceo Along the characteristic,
V(Srr, Srr) is a positive multiple of the positive function (-+w)Srr.
116 IV. A GENERAL HYPERBOLIC MODEL FOR NETWORKS
Discarding the line integral over the characteristic curve segment one is left
with
E fT [Sr. r](a, t) dt -
10
r
1x+(0)
[Sr. r](x, 0) dx ::; fr1f q(r, r,t, r,x) dx dr.
d+
Let : ;
THEOREM 4.1. Let r(x, t) E W 2((0, l) x (0, T); lRP ) be a solution of(4.1).
Xl ::; X2 ::; l. Let x+(t) and x-(t) correspond to the characteristics
C+(XI,O) andC_(x2,0), respectively, and suppose that these characteristics
do not cross for 0::; t < T. Let x+(T) ::; a ::; x+(T). Then
(4.21) (T;x+(T),x_(T)) +l(a;O,T)::; G(0;XI,X2).
PROOF. In view of lemma 4.1 it remains only to prove that
(4.22)
We note that e = Iw+12 + Iw_12 + Sr r. Note also that the characteristics
C(a, T) lie between C+(Xll 0) and C-(X2,0) so that it follows from the
estimates (4.14), (4.15) and (4.16) that
where we have used the easily verified fact that N is the inverse of M.
The latter fact also implies that the eigenvalue of N are the reciprocals of
those of M, so that, in particular, the largest eigenvalue of N is 1/~. The
characteristics are also the same. We rewrite Theorem 4.1 in the form that
is needed for proofs of exact controllability.
THEOREM 4.2. Letr(x,t) E W 2((0,l)x(0,T);lRP ) beasolutionof(4.1).
Let 0::; tl ::; t2 ::; T. Let ~+(t) and ~-(t) correspond to the characteristics
~+ (l, t2) and k- (l, t I) respectively and suppose that these characteristics do
not cross for 0::; X < 1. Let ~_(O) ::; a::; x+(O). Then
(4.24) 1 (0; x+(T), x_(T)) + (a; 0, l) ::; Gl (l; til t2).
A similar estimate holds if we interchange the roles of
and use the characteristics k+ (0, tl) and k_ (0, t2).
and l throughout
4. ENERGY ESTIMATES FOR HYPERBOLIC SYSTEMS 117
-1/~~y)
given by
fact we shall estimate the term &T (l; 0, 0); the estimate of the term at the
endpoint is similar. Now in Lemma 4.2 we set a = l and T = 0 chosen
118 IV. A GENERAL HYPERBOLIC MODEL FOR NETWORKS
sufficiently small to ensure that C+(l, 8) meets the initial interval. Then it
follows from the inequalities (4.14), (4.15) and (4.28) that
L [w(l, t) . Fk ]Fk
p
w+(l, t) = IT+w(l, t) =
k=l
k=l
p
(4.31)
Iw+(l, tW = L IO:kwk[P(r,x + Rr)](l, t) . fkl2
k=l
~ cl[Q(r,x + Rr)](l, t)12 = clw(l, tW
where
c = min IO:kWkl2 II QP- 1 11 2,
19~p
with the norm being the operator norm on matrices. From (4.30) and (4.31)
we get the desired estimate on 1 (l : 0,8), which completes the proof. 0
REMARK 4.1. When the coefficients are not time dependent this result,
as well as a corresponding result for networks, follows directly from Lemma
3.3. When needed, the above result is also readily generalized to networks
and allows for time dependence of the coefficients. Theorems 4.2 and 4.4
together provide the estimates which should allow the application of the
Hilbert Uniqueness method (as described in Lions [60]) to hyperbolic sys-
tems with time dependent coefficients.
5. EXACT CONTROLLABILITY OF THE NETWORK MODEL. 119
r(O) = 0, r(O) = o.
Here we make the same assumptions on the matrix coefficients as in Section
2.3; in particular they do not depend on t. We use the same notation for
the spaces of controls as in Section 5, Chapter II, namely,
and
(5.2) I ~o x .\0 II'H ~ C II CT[~o x .\01 IIu, for all ~o x .\0 E1{,.
As in Section 3, Chapter II, 1{, and U have been identified with their dual
spaces. We need to identify CT[~o x '\0], which depends on the of inner
products for U and 1{,.
120 IV. A GENERAL HYPERBOLIC MODEL FOR NETWORKS
P~t ~i = [Q~(>..i
",X
+ ~>..i)] ,x
- RtQ;(>"~x + ~>..i) + Si>"i for i E I,
>..i (Xij, t) coincide for j E 3M, i E I j,
(5.3) :z>i>..(t) = 0 for j E 3 D U3c ,
.Alj >..(t) = O(t) for j E 3 N U 3 F ,
.Al1>..(t) = 0 for j E 3 and i E I j ,
>"(T) = >"0, ~(T) = ~o.
along with the various nodal conditions in (5.1) and (5.3), we obtain
. iii
where
1
(5.7) 8! = -(-)dx.
wl x
o -
T
(5.8) Ei ("2) + Eii (OJ 8 + 8!,. T - . i
8 - 8!) ~ CiEi (lij 8, T - 8)
where
(5.11) T* = max{Til + Til + ... + Ti r : lib i 2, .. , i r ] leads from the
clamped node to a controlled node}.
REMARK 5.2. The result also holds for more general networks. For ex-
ample it also holds if for each path lib i 2, ... , i r ] which leads from the
clamped node to a controlled node the following condition is satisfied: let-
ting j be the index of the multiple node between the ik-th and the ik+1-th
elements, one has Qj = Pi j :::; PiHl .
6. STABILIZABILITY OF THE NETWORK MODEL. 123
THEOREM 5.2. Under the hypotheses of the previous theorem, the net-
work is exactly controllable for each time T 2': T*.
REMARK 5.3. The general model allows for elements of different nature
such as the networks of one beam and two strings introduced in Subsection
2.5. We can adapt the reasoning of this section to consider the controllabil-
ity of those networks. If one end of the beam is clamped and the strings are
joined to the extreme points of the other endface, the motion of the whole
system can indeed be exactly controlled from the remaining endpoints of
the strings. In the other cases the beam cannot be controlled by control
of the strings. The key element is the way in which energy is transmitted
through the multiple nodes; only in the one indicated case is the conclu-
sion of Lemma 5.2 valid; the main argument in deriving the energy estimate
goes through easily once that lemma is established.
[-N'ir(t) - 0] [1)1r(t) - 0] ~ 0,
PROOF. It helps ones intuition to note that (6.3) is just the weak form
of
For a given r x r there can exist at most one f in condition (6.3), so that
As is well defined as a single valued operator. To see this, let Vo denote
the subspace of V whose elements satisfy 1)1r = 0 for j E .:Is \ {m}.
Then Vo is dense in H. If f and f both satisfy the conditions of (6.3), the
difference of the two weak identities with tP E Vo gives (f - f, tP)H = 0,
from which it follows that f = f. We usually write -Ar in place of f so
that, in particular, As[r x r] = -r x (-Ar).
The monotonicity of A.s is immediate: for r x r and s x s in dom(As)
6. STABILIZABILITY OF THE NETWORK MODEL. 125
one has
(As[r x r]-A[s x s], r x r - s x S)l
= (-r + S, r - S)v + (-Ar + As, r - S)H
L [ui - vi] .1)i[r - s]
iE.1 S \{m}
where we have used the weak identity in (6.3) for r and s with q, = r-s, the
fact that -ui E gi('])ir) and -vi E gi('])is) as well as the monotonicity
of each gi.
To prove the maximal monotonicity it is enough to prove that the op-
erator 1 + A is surjective (see Proposition 2.2 of [11]). The equation
[I + A](r x r) = f x f needs to be solved for each f x f E V x H. The
equation translates into
r- r = f, r - Ar = f, -}fir E gi('])ir) for j E .Js \ {m},
which can be rewritten in the form
r-Ar = f+f,
(6.4) r=r - f, {
-ftfir E gi('])i[r - f]) for j E .Js \ {m}
It is enough to solve the latter equations (which no longer involve r); the
first equation then determines r. The weak form of the equations for r is
This functional does achieve a minimum. To see this note first that
where to estimate the convex terms we use the fact that each Gi is bounded
below by an affine function, as well as the continuity of 1)i : V 1--+ lRP. It is
126 IV. A GENERAL HYPERBOLIC MODEL FOR NETWORKS
that
-f[(f,r - f)v + (f,r - f)H - (f + f,r - f)H] - f2[11r - flli=. + Ilr - flit]
< ~ [Gj(1)1[f+f(r-f)-fD-Gj(1)1[f-f])]
jE.JS\{m}
In the last step we used the following consequence of the convexity of Gj:
1(4)) = ~ ui .1)i4>,
iE.JS\{m}
with -ui E gi (1)1 [r - fl). Since the righthand side of (6.8) vanishes
for~ E Vo (the subspace introduced above), it follows (by replacing 4>
with -4> in the inequality) that 1(4)) == 0 on Yo. Now let VI be the
orthogonal complement of Vo with respect to the inner product (., )v,
and let 4> = 4>0 + 4>1 be the corresponding decomposition of elements. It
6. STABILIZABILITY OF THE NETWORK MODEL. 127
is easy to see that the map CPl I--t (tJi CPl) iE.1 S \{m}
is a bijective map of
V 1 onto [JRPjl.1 s l-l and thus one can conclude that
That -ui belongs to gi (tJi[r - fD for each j also follows easily from (6.8).
This completes the proof of existence. We now turn to the proof of the
compactness property. It is enough to show that (I + As)-l : V x H I--t
V x H is compact. Since [1+Asj(r x r) = f x f is equivalent to (6.4) (with
the weak formulation (6.5)), it is enough by Rellich's compactness theorem
to show that solutions of the latter system satisfy the a priori estimate
so that
L II ri IIW
iEI
2(0,1;;RP)
o
We now prove a theorem which guarantees that for any initial state the
solution to the feedback system decays asymptotically to the zero energy
state.
THEOREM 6.2. Suppose that the network has no closed circuits and that
the hypotheses of the previous existence theorem are satisfied. Suppose also
that the minimum of each Gi(u) is achieved only at 0 and that
PROOF. The proof follows standard lines. From the compactness of the
resolvent of.As it follows that the trajectories {Us(t)[ro x fo]h~o are rela-
tively compact in 1 and hence that the w-limit set w(ro x fo) is non-empty.
We shall see that, under the stated hypotheses, it consists of the zero state
alone. Since the energy norm is nonincreasing along each trajectory a stan-
dard proof shows that the energy is constant on w(ro x fo). Moreover it
is easily seen that if So x So E w(ro x fo) so does the whole trajectory
[s x s](t) = Us(t)[so x so]. Consequently, energy is conserved along the
trajectory. It follows from (6.2) that for all positive t,
Therefore for each j E 3 \ {m} one has ei (xi , t) = 0 and it follows from
the estimate of Theorem 5.1 that the solution s(t) x s(t) has zero energy
for all t so that, in particular, So x So is the zero state. 0
THEOREM 6.3. Suppose that the network contains no closed circuits and
that the functions Gi are continuously differentiable, so that the gi are
continuous, single valued and maximal monotone. Suppose also that
(6.14)
/lUs(T*)[ro x rolll~
fT.
=llro x roll~ - L In gj(1Yr(t)) . 1Yr(t) dt
jE..7s\{m} 0
(6.15)
~/lro x ro/l~ -8 L
jE..7s\{m} 0
I T
[11) jr(tW + Igj(1Yr(tW] dt
fT.
=/1 ro x ro II~ -8 L In [11Yr(tW + INjr(t)1 2] dt.
jE..7s\{m} 0
Since
ei,(Xj,t) ~ C[I1Yr(t)1 2 + INjr(t)1 2],
it follows from (6.15) that
If we note that I Us (T*)[ro x rolll~ ~/I Us (T*/2)[ro x rolll~ and then use
the estimate of Theorem 5.1, we get from (6.17)
LEMMA 6.1. Under the hypotheses of Theorem 5.1 we have for all T >0
(6.19) (t)~C L
jE..7s\{m} t-S.
I t +S
e., (xj,r) dr.
6. STABILIZABILITY OF THE NETWORK MODEL. 131
Now
l o
T &(t) dt = ls*
0
&(t) dt + i T
T-S.
&(t) dt + iS.
T S
- &(t) dt.
T* sup &(t),
O:St:ST
while, by (6.5),
i s.
T
-
S
&(t) dt ::; L i [I
jE.7s\{m} s*
T
-
S
t s
+
t-S.
* ei j (Xj, r) dr] dt
::; T* L lT
jE.7s\{m} 0
eij (Xj, r) dr,
where we have changed the order of integration. The result follows from
the latter estimates. 0
The next lemma, due to Lasiecka and Tataru [54], will be stated without
proof.
(6.20)
THEOREM 6.4. Suppose that the network contains no closed circuits and
that the functions Gj are continuously differentiable, so that the gj are
continuous, single valued and maximal monotone. Suppose also that
L
jE.1s\{m} 0
iT [11)if I2 + Igj (1)if)1 2 ] dt
iT
jE.1s\{m} 0
where we have used HOlder's inequality. We substitute this, along with the
identity
i
jE.1s\{m} 0
+ T1/q [ L T
[gi(1)if).1)if]dtf/P].
jE.1s\{m} 0
1. Preliminaries
mat ions but also for the infinite dimensional full problem has been observed
by Destyunder [27] and, very recently, by Balakrishnan [7], who exploited
the ideas of [110], [111]. These papers are among of the second group
attacking the full infinite dimensional problem. A conceptually quite dif-
ferent subset of this second group contains the works of Nicaise, von Below,
Ali Mehmeti and others ([75], [77], [76], [106], [105], [67], [69], [71], [68]).
The latter papers are essentially devoted to the eigenvalue problem for the
Laplace- Beltrami operator on smooth, connected graphs . In particular,
for one-dimensional elastic links the method of von Below [106] is closest
to the ideas which we will pursue here. The basis of this method is the
possibility of rescaling the spatial variables associated with each member
of the structure to the uniform interval [0,1]. Once such a rescaled for-
mulation and a transformation to a first order system is obtained, some
classical results on the spectrum of general first order systems can be used.
See the work of Schiiike and Schneider and references in the survey arti-
cle by Schneider [98]. See also von Below's Habilitation thesis [107]. We
will follow [106], where a slightly more direct and constructive approach is
taken: the rescaled system is kept as a second order system and then ex-
plicitly solved using Hadamard's operational calculus and an appropriate
version of the variation of constants formula. The eigenvalues can then,
in principle, be computed from this representation. We will also provide
some numerical evidence based on a slicing technique similar to the one by
Wittrick and Williams [110].
We remark that the following sections are very much in the spirit of
von Below's works, even though the problems encountered here are more
difficult in that the models we study deal with in-plane or even fully 3-d
displacements rather than scalar functions which are exclusively considered
in von Below's work and, incidentally, also in the work of Nicaise [75], [77],
etc. However, for the sake of completeness, the arguments will be provided
in some detail.
incidence matrix
if 7I"j(lj) = Vi,
(1.1) if 7I"j(O) = Vi,
else,
I if ViVh E E(G),
(1.2) eih = { 0
else.
The individual length lS(i,h) of the edge joining Vi, Vh can then be expressed
by lih := lS(i,h)eih' where may set s(i, h) = 1 if eih = O. This notation
makes it possible to normalize the description of the running variable in
each individual edge as follows: for x E [0, 1],
._ (1 +
"Yih(X) .-lih
dis(i,h)
2 - xdis(i,h)
)
G =Uf=I kj,
aV(G) ={Vi E V(G)I d(Vi) = I},
o
V (G) ={Vi E V(G)I d(Vi) > I}.
Obviously, aV(G) signifies the set of simple nodes, where only one edge
o
starts or ends, while V (G) signifies the set of multiple nodes, where two
or more edges meet. A function r : G f-t lRT can then be viewed as a
collection of functions with values along the individual edges given by
(1.3)
138 V. SPECTRAL ANALYSIS AND NUMERICAL SIMULATIONS
and also
rhi(X) = rih(l - x).
This notation is adopted from [106].
Here 00 V(G) C oV(G) denotes the set of simple nodes with clamped
boundary conditions, 01 V (G) is the set of stress free (or externally loaded
as is the case when fv f- 0) simple nodes and ind f(Vh ) is the index set of
f(Vh)' The third equation expresses the continuity of the network, or its
connectedness also in the deformed configuration, while the fourth equation
represents the balance of forces (with a possible external loading modeled
1. PRELIMINARIES 139
.. Z(-2)K /I 0
rth - th s(t,h)rth = ,
(1.5) {
rht(x) = r t h(l - x), x E [0,1]'
(1.6)
(1.8)
(1.9)
rt(V) = 0, Vv E OoV(G),
(1.12)
The continuity condition is exactly of the form as in (1.4), with the only
difference that the vectors have now three components. Let us consider
the homogeneous situation. The balance of forces requirement can then be
expressed as
(1.15)
as
(1.18)
and in the case of pinned joints we do not have any additional condition of
the type (1.18).
Let us now look at the dynamical node conditions or, in the terminology
of Lurner, at the connection/interaction operators. These are
where the summation is taken over all h E ind r( vt ). In the case of a pinned
joint (1.20) is replaced by the stronger condition
(1.22)
We are now going to express this system in terms of the matrix notation
developed earlier. We have
(1.23)
rth + -L,h
o( -4)K ""
s(t,h)rth o( -2)H
- -Lth " - O'
s(t,h)rth - ,
3. E R 2N : r,h(O) = e,h(i),
(1.24) {
(z) = 0, 't/v, E 8oV(G);
(1.28)
The matrices e, V are the adjacency and incidence matrix of the underly-
ing graph as defined by (1.2) and (1.1), respectively. For the case of scalar
valued functions r and vanishing matrices S, M this system has been stud-
ied by von Below [106J and, based on a slightly different matrix setup, by
Nicaise [75J. It should be noted that the restriction to the second order
case is just a matter of convenience and that higher order problems can
be cast into a similar form. This is particularly important for networks
composed of Euler-Bernoulli beams.
We define B:= . K,(-1/2). Then using the functional calculus (which has
been attributed to Hadamard) introduced above, we may define
1
(2.3) R(x) = cos(xAB) . <P + AB(-l) . sin(xAB) .IJ!,
with the initial matrices <P = e* . e, IJ! = R' (0). It is easy to see that R( x)
given by (2.3) is the general solution to the system of equations (1.5) in
which the inertia term is replaced by -A2rih. The problem then consists
of finding matrices <P, IJ! such that the conditions (1.6) and (1.7), trans-
formed into matrix notation as in the second and third equation of (2.2),
are satisfied. The possibility of representing the general solution to the
matrix-differential equation in a comprehensible form is a crucial step in
this particular approach to the analysis of the eigenvalues. The analo-
gous representation in the case of a network of Timoshenko beams is more
involved and far less obvious. However, the representation (2.3) can be
utilized in essentially the same way it was done in [106J (see also [75J for
the case of homogeneous networks). Here we give the necessary conditions
for A to be an eigenvalue.
+ qS(i,h)(sin(Aq~!,h))-1 cos(Afs(i,h)q~!,h)P*i,h)'
The eigenelements, which are chamcterized through the nodal vectors , 'II,
then satisfy
(i) i=1, Vi=2, .... ,n; Wih=O, Vi,h=I, ... ,n;
PS(i,h)h = (-I)kP s(i,h)i, Ps(i,h)Wih = _(-I)kP s(i,h)Whi,
(ii) {
P*i,h)j = P*i,h) 'II jk = 0, Vj,k=I, ... ,n;
(iii) {
P*i'h)~ : (-1-1)kP*i'h~i' P*i,h) Wih = -( -1)-1p*i,h) Whi,
P s(i,h)J - P s(i,h) 'II Jk - 0, Vj,k=I, ... ,n;
(iv) E ker Z(C - diagHi ), and
Wih = Aeihfs(i,h){P~!,h)(sin(AfS(i,h)P~!,h)-1
(P s(i,h) (~ih - cos(Afs(i,h)P~!,h)~ih)
+ q~~,h)(sin(Aq~~,h)-1(P*i,h)(~:h - cos(Al8(i,h)q~:,h)~ih}'
respectively.
PROOF. The proof is given in the spirit of [106]. However, the situation
here is slightly more complicated in that we have vector-valued functions
and also allow for Dirichlet conditions in the simple nodes. Therefore, we
provide the arguments.
(i) We first deal with the case where A is zero. Suppose that 80 V(G) = 0.
Then (2.2) reduces to
(-2) .J(. R"(x) = 0,
3: R(O) = e* . ; i = 0, i: Vi E 8oV(G),
((-1) .J(. R'(O))z = 0,
R*(x) = R(I- x), Vx E [0,1].
The general solution to the first equation, under the requirement given by
the last equation, yields
In contrast to [106], we cannot invert the operator on the right side of this
equation. Nevertheless,
This equation is obviously satisfied if all nodal deformations >i are the
same, which means that the graph as a whole is shifted by a vector > =
>i, 'Vi.
Let now ooV(G) i= 0. It is sufficient to consider the case 8oV(G) =
{>t}. Then, however, the representation of the solution above gives rl,2(x)
== 0 which then implies >2 = 0 and so on. Hence, zero cannot be an
eigenvalue if one exterior node is clamped. (Note that we do not consider
a possibly rigid rotation around node 1 of the whole graph in this model.)
(ii) We are now going to demonstrate the occurrence of the eigenvalues of
each individual string under some periodic boundary conditions, dictated
by the" connectivity" conditions.
Let us consider the case where Aj,k = AS(i,h),k = f;(!,h)PS(i,h)1l"k, with j =
s(i, h) for some pair (i, h)j the case corresponding to (iii) in the statement
of the theorem is done in the same way. We set to zero all initial data
<Pm,n, Wm,n such that AS(i,h),k i= f~!n,n)Ps(m,n)1l"k. Then we are left with
Lf~!,m)etmP~(t,m) Wtm = O.
m
Note that if the coefficients of the s( i, m )-th string joining the i-th and
m-th vertices are different from the ones of the s(i, h)-th string, then by
assumption the corresponding Wtm = O. For the others the condition above
has to be taken into account. (We do not claim that the conditions (ii)
in the theorem are sufficient.) Note also that if one end of the j-th string
(j = s(i, h)) is clamped, the balance condition above is not effective for the
corresponding vertex, so that the vertex adjacent to the clamped end stays
fixed and the derivatives have to satisfy (ii).
(iii) is handled as (ii).
(iv) We now turn to the case where>. is neither zero nor of the form
discussed in (ii) or (iii):
rth(l) = PS(t,h)eth[cos(>.fS(t,h)P~!,h)<[>th
1
+ :\f~!,h)PS(t,h) sin(>'fs(t,h)P~!,h)Wthl
+ P~t,h)eth [cos (>'fS(t,h)q~!,h)<[>th
1
+ :\f~!,h)qS(t,h) sin(Aes(t,h)q~!,h)Wthl = rht(O) = <[>ht
The strategy is in principal taken from [106], but again we cannot invert
operators here. The idea is to eliminate Wth from the balance condition
and thereby obtain a system of linear equations in the unknowns t. We
recall that
I: {eihis(i,h)Ps(i,h) sin(Ais(i,h)P;!,h)-l
h
We are now going to give sufficient conditions in the special case where
ratios of the physical and geometrical quantities are uniform. In particular,
we will be looking at completely uniform networks.
(I) We shall first discuss the Neumann case, aoV(G) = 0. Here we can
fully adopt the results of von Below [106] and Nicaise [75]. In the case
where sin(d) i= 0 we find
and
sin( CAX)
(2.4) (rih)ih = COS(CAX)' >e* + sm
. (CA) (e>* - COs(cA)>e*) . ,
(II) Let us now discuss the case where 8oV(G) # 0. The only difference
from the previous case is that the balance condition now is
Z(C (e* - cos(c.\)e*) . )e = 0.
By a remark in [106], it is sufficient to consider the case C = . The system
then reduces to
Z = cos(.\)diag (e).
Now the application of Z just deletes the first k rows (which are to be
viewed as rows with 2-d columns) of . But also the first k entries (which
in fact are (2,1)-vectors themselves) in are zero. Therefore, the system is
n-k square. However, a complete characterisation as above does not seem
to be known even in the scalar case.
In the purely serial case (collinear or not) this means that the multiplic-
ities of the former eigenvalues are always equal to one.
2.3.1. Examples. We discuss the simplest situation in which all con-
stants are taken to be equal to one. Then the spectrum of (2.2) can be
determined as above from the eigenvalue problem
(2.6) = diag(d(vi))cos(.\).
Recall that d(Vi) is the edge degree of the vertex Vi. The simplest case, in
which the eigenvalues of type (iv) can be determined by the eigenvalues of
the adjacency matrix alone, is the case where d(vd == d, 'Vi = 1, ... , n. The
latter is true for a simple closed loop, where the edge degree is equal to 2
for all vertices. We have
(2.7) = dcos(.\).
EXAMPLE 2.1. The unit square.
In this case G is bipartite and 2-regular in the sense that the edge degree
is 2 for all vertices, i.e., d(Vi) = 2. Here is given by
(0101)
I I 0
= I I .
I I
The eigenvalues of are 0,2, - 2 and those of Z are 0, 1, -1 modulo multi-
plicities, hence.\ = krr including k = 0, and.\ = (k+ !)1T. This shows that
only the value 1T /2 gives rise to a new string.')f eigenvalues, i.e., eigenvalues
which are not among those of an individual string of length 1, with mul-
tiplicity 4. We have chosen this example because the eigenvalues can also
be obtained by elementary means from classical results on the spectrum
of the second order derivative operator with constant coefficients and with
periodic boundary conditions. Indeed, if all the material properties are
one and the same constant, so that the governing operator is the same on
150 V. SPECTRAL ANALYSIS AND NUMERICAL SIMULATIONS
all links, then the continuity and equilibrium conditions at the nodes are
equivalent to periodic boundary conditions. The spectrum of this opera-
tor is exactly as above. This analogy is not, however, apparent when the
material constants differ.
There are four different configurations of nodal values according to the
four eigenvalues 7r /2. These four nodal configurations are precisely the four
corresponding eigenvectors 4>i of the adjacency matrix:
-0.0940 -0.5659
-0.6897 0.1421
0.0157 -0.1256
0.1233 0.3792
4>1 = 4>2 =
0.0940 0.5659
0.6897 -0.1421
-0.0157 0.1256
-0.1233 -0.3792
-0.3713 -0.1818
-0.0110 -0.0631
0.4481 -0.5321
-0.4015 -0.4240
4>3 = 4>4 =
0.3713 0.1818
0.0110 0.0631
-0.4481 0.5321
0.4015 0.4240
1.5
':r
0.5
I-=-J 1I
0.5
[]
-oj
0
-0.5 0 0.5 1
I
1.5
-0.5
-0.5 0 0.5 1.5
.. 1.5
::r-(~_-]
----------,--.~-
0.5
[~-]
_o.~l
0
-0.5 0 0.5 1
1
1.5
-0.5
-0.5 0 0.5 1.5
10
\
\ ) \
4
o
o 4 10
o
o 4 10
matrix above and then invoke Sylvester's theorem on the invariance of the
number of positive, vanishing and negative eigenvalues. These numbers can
be recovered from the diagonal of the upper triangular matrix R. This is
essentially the argument of Wittrick and Williams [110]. However, in their
work it is applied to the finite element approximation, whereas here we ap-
ply it to the fully infinite dimensional system by using the exact solution.
Note that zero is a two-fold eigenvalue and that at individual eigenvalues
there is pole of the determinant of C - diag(Hi). At odd multiples of 7r we
recognize a jump by four indicating, as was proved above, the multiplicity
four. In passing the poles the multiplicity reduces by four, indicating that
also the individual eigenvalues have multiplicity four.
This analysis becomes particularly interesting when applied to the case
where the material properties are not equal to the same constant. This is
the case where we do not have explicit knowledge of the eigenvalue distri-
bution. We concentrate on the first eigenvalue coming from the graph (that
is, the first one related to the vanishing of the determinant of C - diag( Hi) ).
In order to be able to compare to the previous case, we take as a base the
same constant 1 for all the material constants. Then, however, we perturb
these by 0.1 x random numbers. The result is displayed in Figure V.3. We
clearly observe the splitting of the spectrum into simple eigenvalues. We
remark that, at this time, we do not have a proof of this natural result in
a general context.
\0
0
1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.1 2.2
2
1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.1 2.2
the graph is bipartite, but not regular, and does not contain a circuit. The
adjacency matrix is given by
1o 0I) .
o 0
We first consider the case where both exterior nodes satisfy the Neumann
condition. Since this time the edge degree is not uniform, we have to
consider the generalized eigenvalue problem
e = JLdiag (21, I, I)
and then take >. = arccos(JL). We obtain the eigenvalues >. = 7r /2 plus
integer multiples of 7r. We also have all integer multiples of 7r including
zero. The same is true for the serial beam, where the angle between the
strings is 180 degrees rather than 90 degrees. In fact, our analysis clearly
shows that the eigenvalues (eigenfrequencies) are independent of the angle
at the joint. In general, we can show that the serial string (collinear or non-
collinear) having m elements possesses all multiples of 7r / m plus multiples
of 7r as eigenvalues (if, as assumed, the exterior nodes satisfy Neumann
conditions) .
In the case of the carpenter's square as above but with both exterior
nodes satisfying Dirichlet conditions we find
L elhh = 2 + 3 = 2JL1
h
154 V. SPECTRAL ANALYSIS AND NUMERlCAL SIMULATIONS
20--~~-~--
15
I
Ii
JO
I1
i i
5 I j
or ~_J j
-L~
15 2 25
and 4>2 = 4>3 = 0, so that J.t = 0 and hence cos().) = 0 and, therefore,
7r /2 plus all integer multiples of 7r are eigenvalues. In addition, we have all
integer multiples of 7r with the exception of O.
If we take the same model, but instead of two Dirichlet nodes we take
one node satisfying the Dirichlet and the other the Neumann condition we
obtain the equations
=(~~~).
I I 0
Here we obtain). = (k + 1/3)7r for odd k and (k + 2/3)7r for even k. In
addition, we obtain 0 and all even multiples of 7r. We display the same
numerical evidence as in the first example: in the first graph of Figure VA
we see the plot of the determinant of C - diag( Hi) and in the second we
show the corresponding multiplicities, these being in agreement with the
theory.
We also perform the perturbation of constants analogous to the one in
Example 2.1. The result is displayed in Figure V.5. Again we observe the
splitting of multiple eigenvalues.
2. EIGENVALUE PROBLEMS 155
///\ -l
21
15
05
05
j
I
15f
21
1 05 05 15 25
[I rV
1~1~-1~-1--
'l
6f
2
.
(\
III,
II
I
I 1\ 5~
lO'
I
'C;~l
0
0 10 12 14 16 18 22 2' 26
L:h[~h1)Ks(i,h)r~h(0) + dis(i,h)Ms(i,h)rih(O)] = 0,
(2.8)
3: rih(O) = eih(i),
rhi(x) = rih(l- x).
As in the preceding case, we want to derive a general solution formula for
the Cauchy problem. In order to do this it is convenient to eliminate the
shear angle as usual. To this end we have to write down the equations of
motion in terms of Uih, Wih, 'ih as follows:
-2 2 " A2
s(i,h)Ps(i,h) Uih = - Uih,
We differentiate the second equation twice and the third equation once.
Also, we express '~h in terms of the vertical displacement and its second
derivative, using the second equation. The result is
-2 2 111'
s(i,h)qs(i,h)wih + -1 2
s(i,h)qs(i,h)'f'ih =
./,1"
-A
\2 "
Wih,
-2 2 ./,1" -1 2 " 2 ./,1 A2./,I
s(i,h)'Ys(i,h)'f'ih - s(i,h)qs(i,h)Wih - qS(i,h)'I'ih = - 'l'ih'
./,1 -1 "A2 -2
'l'ih = - s(i,h) Wih - s(i,h)qs(i,h) Wih
We can now use the second and the third equation to eliminate 'ih. We
obtain
(2.9)
""
Wih
~s(i,h) )
+ ( 1 + -2-- 2" 2 4 2
qS(i,h) 2 2
A Wih - A s(i,h)-2--(qs(i,h) - A )Wih = O.
'Ys(i,h) 'Ys(i,h)
2. EIGENVALUE PROBLEMS 157
Since we want to solve the initial value problem, we have to use the com-
patibility conditions
(2.10)
T =~
2
(1 + (S(i,h)
'Ys(i,h)
)2) >..2
[ ~4 (1+ (qS(i,h)2)
'Ys(i,h)
2
>..4+!ih (QS(i,h)2 >..2(q;ih _>..2)
(,) 'Ys(i,h) ( , )
]1/2
Note that, in order to obtain oscillatory solutions, >.. has to be larger than
qS(i,h) if it is not equal to zero. We define J-Lih1 := ,.;r+, J-Lih2 := r-,
J-Lih3 := -J-Lihb J-Lih4 := -J-Lih2 We can now write down the general solution
158 V. SPECTRAL ANALYSIS AND NUMERICAL SIMULATIONS
o/.h +
.1./1 (f S(',h)2 (>.2 _ Qs(.,h)
2 ).1. o/.h -_ fs(.,h) (QS(',h)2,
W.h
'Ys(.,h) 'Ys(.,h)
c .- fs( h)
..,.h .-
V'2 _ 2
1\ Q ( h)
'V
IS ( .,h ) S "
2. EIGENVALUE PROBLEMS 159
(2.12)
1)
ih fLihl ih fLih2
+ cos (
fLihl X ) (
0 qS(i,h) 0
(,s(i,h) - - .- ihfLihl c2 _ 2
l's( t,h)
e _1)]
"'ih fLihl
+ cos (
fLih2 X ) (
0 qS(i,h) 01
(,s(i,h)--. -OihfLih2 2
'Ys(t,h) ih fLih2
+ _/,1
'f'ihO
[~Sin(c.
c "'th
"'ih
x) (1+. s(t,h)
(qS(i,h)2 (
l's( i,h)
c2 2
"'ih - fLihl
O}~fLThl + c2
0}lfLTh2 ) )
2
"'ih - fLih2
- -
1
fLihl
.
SIll(fLihl X )
qs(i,h) 2
( is(i,h) ( - - )
'Ys(i,h)
e010 2
ihfLihl
ih - fLihl
)
2
- - 1 SIll
. (fLih2 X ) (0
(,s(i,h)
(qS(i,h) ) 2 OtlfLTh2 ) ]
-- 2 2
fLih2 'Ys(i,h) ~ih - fLih2
'
+WthO [~Sin(c.
c "'th
"'ih
x) (1-. s(t,h)
(qS(i,h)2 (
'Ys(i,h)
c2 2
"'ih - fLihl
P?~fLThl + c2P?lfLTh22
"'ih - fLih2
))
- - 1 SIll
. ( fLihl X ) (0
(,s(i,h)
(qS(i,h)2
-- 2 2
P?~fLThl
)
fLihl 'Ys(i,h) ~ih - fLihl
- - 1 SIll
. ( fLih2 X ) (0(,s(i,h)
(qS(i,h)2
--
P?lfLTh2 )]
2 2
fLih2 'Ys(i,h) ~ih - fLih2
'
+ WihO ",th X (
[cos(c.) 1 _ o. ( qs(.,h)
(,s(t,h)
.. )2 ( c2PihfL.hl
10.
_ 2
11.))
+ c2PihfLth2
_ 2
'Ys(t,h) "'ih fLihl "'ih JLih2
+ COS(fLihl X )
2 PihfLihl
10 )
- .- ) e _ 2
(is(i,h) ( -qs(i,h)
'Ys(t,h) ih fLihl
+ COS(JLih2 X ) (
is(i,h) (qS(i,h)
- - .-
'Ys(t,h)
)2e 11_ )]
PihfLih2
ih
2
JLih2
.
Given the deformations (2.11), (2.12) and (2.13) we can go back to the
matrix formulation and write
160 V. SPECTRAL ANALYSIS AND NUMERlCAL SIMULATIONS
We want to express this in terms of the initial values. To this end we define
0)
Wi~(A, x)
(UihO)
WihO
ZI~(A, x) s{i,h) "pihO s{i,h)
+( 0
U1h 0 0)
WBt(A,X) Wi~(A,X)
(U~hO)
W~hO ,
o ZPl(A,X) Zll(A,x) s{i,h) "p~hO s{i,h)
where the matrices and the vectors are given in the base es{i,h) , e*i,h)' n,
as indicated by the suffix s(i, h). This representation is particularly useful
if one wants to have a compact notation. However, one has to keep in
mind that for each different pair of indices i, h one is then talking about a
different meaning of the component form. We recall
l~~,h)P~(i,h)U~h = 0,
0-2 0-1 0
{.s(i,h)qs(i,h)wih + (.s(i,h)qs(i,h)'Pih = ,
2 II 2 .1,1
But then
and hence
and Wih to
Wih(X) = WihO + X(W;hO - WihO).
With this information we go back to the differential equation for 'l/Jih' which
now reads
W~h(X) = is(i,h)'l/Jiho = W;hO - WihO
But since 'I/J;hO = 'l/JihO, we conclude that w;hO = WihO which, in turn, reduces
Wih (x) to a constant
Wih(X) = WihO = W;hO
and, in addition, 'l/Jih(X) to zero. Therefore, the necessary conditions for
zero to be an eigenvalue are
0= e;h \II
or, in matrix form,
o
Wi~l(>.,x) 0)
Wll (>., x) (U~hO)
w~hO
Z?l(>', x) zg (>., x) s(i,h) 'I/J~ho s(i,h)
Since in this equation u~hO is uncoupled from w~hO' 'I/J~ho, we can set the
latter equal to zero and choose the eigenvalue >. to be a multiple of 11".
2. EIGENVALUE PROBLEMS 163
Likewise, we can take u~hO to be zero and look for values ). such that the
sub-matrix
has determinant equal to zero. Going back to the formula for the vertical
displacement, we see that this determinant condition describes precisely the
eigenvalues of a single Timoshenko beam. Unfortunately, there is no simple
explicit representation of those values. Note that we have dispersion in the
Timoshenko model. The determinant of the matrix Vi1
can be computed
more easily in the case of homogeneous networks. That case will be studied
later. Let us display the formula for the determinant in the case where all
coefficients are equal to one:
det(Vi1) = det(Vl) =
2.4.3. Eigenvalues for the Entire Groph. In this section we assume that
the eigenvalue). is neither zero nor of the form just discussed. We will use
the general solution formula (2.14), as before. Recall
o
0)
WB:-(>,x) Wli:-(>',x) (U~hO)
w~hO
Zrl(>',x) ztl (>., x) s(i,h) tP~hO s(i,h)
and
o
WB?().,x) Wii?o
()., x) ) (UihO)
WihO
(2.15)
164 V. SPECTRAL ANALYSIS AND NUMERICAL SIMULATIONS
(wU~hO)
ihO
I = (81ih )-1 (UihO)
*
wihO UihO )
- A ( WihO ,
'IfJ~hO s(i,h) 'lfJihO s(i,h) 'lfJihO s(i,h)
where A is given by
UPh 0
0
Ulh
A= 0 ~(W~OZ~l-
det t t W11Z~~)
t,
~(W.~Z~l-
det , t
W11Z~~)
"
0 ~(Z~OW~l1
det ,h ,h
_ Z~lW~O)
,h ,h
~(Z~OW~l
det ,h ,h
_ Z~lW.lO)
,h ,h s(i,h)
We have suppressed the argument (>', 1) and introduced the notation det =
det(8}h(>.,I)). We are going to use this expression for \l1 in terms of ~ in
the balance equations in order to obtain a system of equations involving
just ~. The idea is the same as in the string case; however, the system is
much more complicated. We have
o
for all i such that Vi EV (G) u (it V(G). Even though the equation (2.16)
has a very compact form, one has to keep in mind that the components
are taken with respect to the individual bases associated with each edge.
Therefore (2.16) is not suitable for computations.
2. EIGENVALUE PROBLEMS 165
UO(A,x) = cOS(Ax) ,
Ul(A,x) =~sin(Ax),
1 Asin(JLlx) - 21 Asin(JL2x) ,
ZOO(A,x) =-2
JLl JL2
W A-I A+l
Z (A,x) =--V:-COS(JL1X) + --V:-COS(JL2X),
"{ 1 1
(2.17) L...J Ul PS(i,h)<Ph + det (Z PS(i,h)<Ph - w MS(i,h)<Ph)
11.1 11
+~
det
(W lO Z11 _ Wll ZlO)M. + ~ (ZOOW 01 _ ZOI WOO)N .
s(t,h) det s(t,h)
The matrix elements W kl , Zkl are evaluated at (>.,1). It is clear that even
in this simple situation equation (2.17) is too complicated to analyze in
general. It is possible, however, to compute the eigenvalues from (2.17) for
very simple configurations such as serial collinear beams or certain non-
colinear networks. But the computations are enormous, even if only few
edges are involved. Things turn out to be much simpler if we concentrate
on the asymptotic behavior of the eigenvalues, that is, the behavior of large
>.. If all constants are equal to one, we have
We multiply equation (2.17) by det 1/ 2 = (sin >')/ >. to obtain the asymptotic
equation
(2.19) = cos(.x)diag (d i ),
which we recognize as the same equation as in the string case. This, how-
ever, comes as no surprise if one takes into account that, by assumption,
all wave speeds with respect to the longitudinal, vertical and shearing mo-
tions, respectively, are the same and that the effect of dispersion is only
present in the lower modes.
The controls just derived steer the initial conditions to rest in time T =
2. It is a classical result that these controls are also time optimal and, on the
optimal time interval, nonn-minimal. Similar arguments can be applied to
the Timoshenko system. We note, however, that because of the dispersion
of waves in that case, we do not have exact cancellation of outgoing signals
at the boundary with the consequence that exact controllability by velocity
feedback controls cannot be achieved. Nevertheless, exact controllability
can clearly be proved for open-loop controls. It is apparent that besides
the necessity of discretizing the state equation and the boundary conditions
in an appropriate way, the total energy as a function of time plays a key
role. In particular, it appears to be necessary to use a numerical scheme
which conserves the total energy of the discretized system as long as no
controls are applied. This can, for instance, be achieved by the classical
Lax-Friedrichs scheme discussed by Le Roux [91]. See also Halpern [59] for
various schemes simulating absorbing boundaries. This scheme is applied
to the system above in terms of the variables Ui, and it is also appropriate
to quasi-linear models. To avoid unnecessary indexing, we write U = w',
v = W, where w solves
w= (O"( w'))'
170 V. SPECTRAL ANALYSIS AND NUMERICAL SIMULATIONS
u = v', V = u'.
Upon introducing the grid X t = i . ~x, i = 0, ... , n, tJ = j . ~t, j = 0, ... m,
where ~x = lin, ~t = q~x, q being the appropriate mesh ratio, we have
(see [91])
(3.2)
(3.3) {
J +1
vo = u J +1
'n =
,
Now that the control strategy is essentially settled, we are going to attack
some exemplaric multi-string models. We will, however, only consider out-
of-plane displacement models in this section and leave the planar models to
3. NUMERICAL SIMULATIONS 171
20
15 -
10
5
I
l~-,--~ ---'----'....... ___._~~_
20 40 60 80 100
Total Energy
29.347~--~--~---~--~---~--~
14.6731-------------------
oL---~--~---~--~---~---
o 2 3 4 5 6
Total Energy
l6.---~----~----~----~----~-----,
14
12
\0
OL-----~--~----~----~----~----~
0 4 5 6
~_ _---'..:'ls"'-'tS~t,-"nng______ _
O L - - - - - L - -_ _~ O~-----~----~
o 5 10 o 5 10
15~---~3r~d~S~tr~In~--~
O~----~---~
5 10 o 5 10
1st Stnn
0 0
0 5 10 15 0 5 10 15
6 -- ~nllre System
15
4
I
05 2
0 0 I
0 5 10 15 0 5 10 15
(\
2.5
(\ r'\ f\
4 2 \ (\
3 1.5
0.5
0 \1 \/ \;
10 0 5 10
3 1.5
0.5
0
; J. \ 0 JJ J V
0 5 10 0 5 10
planar system, is only for the sake of a convenient display of the plots. We
will consider planar systems later.
together with initial conditions. This system describes the motion of three
strings with obvious orientation of the running variable x. The orientation
is chosen in order to avoid mixed boundary conditions. We perform a
succession of transformations as above, namely,
U i1.'-
- W i , ui2'.= (wi)' ,
ii = -(Zi)', i = 1,2,3,
zi = (Zi)', i = 4,5,6,
Zl(O) = Z4(0), zl(l) - z4(1) = z2(1) - z5(1),
z3(1) = z6(1), Z2(0) - Z5(0) = Z3(0) - Z6(0),
1 1 1 2
z4(1) + z2(1) = .j21 , Z5(0) + Z3(0) = .j21 ,
together with initial conditions. Our goal now is to steer the energy of the
entire system to the second string by applying absorbing controls at the
two multiple joints connecting the second string with the first and the third
strings. We adopt the strategy of P. Hagedorn and J. Schmidt, see [97]. To
this end, we define
(3.5)
By this choice the outgoing signal Z3(0) at x = 0 and z4(1) at the boundary
x = 1 are canceled. For the sake of convenience, we write down the bound-
ary conditions in a matrix form reflecting incoming and outgoing signals
3. NUMERICAL SIMULATIONS 177
as in Russell [92]:
~
at
- x=o
~ .
This shows that after two reflections at the boundaries the only residual mo-
tion left is in the second string. This is what is meant by directing or diode-
type controls. This concept is due to P. Hagedorn and J. Schmidt [97], who
suggested the use of this kind of control in flexible multi-structures. The
paper [97], however, considers only the one-link case for out-of-plane dis-
placements (which for two such controls can also be viewed as a three-link
situation). We will demonstrate that concept for planar frames shortly.
Let us first, however, provide some numerical evidence for the situation de-
scribed above. In Figure V.16 we take an initial condition distributed along
the three strings which are chosen to have individual lengths ~, so that the
length of the entire system is one. That the energy remains constant after
two reflections (that is at t = ~) is seen in Figure V.1 7.
In the context of the original system (3.4) the controls are chosen to be
as follows:
178 V. SPECTRAL ANALYSIS AND NUMERICAL SIMULATIONS
Total Energy
21
20
19
18
17[
1
16
15
::t
f I
0 2 3 4 5 6
It has been shown in [25] that absorbing boundary controls for the 1-
dimensional wave equation are sensitive to time delays in the velocity feed-
back. In fact, it has subsequently been shown by R. Datko [23], [24] that
this phenomenon is not peculiar to strings and that it is not even circum-
vented by the introduction of severe damping. This instability for our three
string model (3.4) is seen in Figures V.1S and V.19.
Even though this instability is rather discouraging, the situation in a real
structure is not as bad as this seems to suggest. First of all, some of the
energy absorbing feedback controls can be implemented as passive devices,
for which a time lag will not be present. But even if the feedback law is to
be realized by numerical computation (in real time), the time-lag due to the
action of the microprocessor will not be a constant, but, rather, distributed
by a certain stochastic process. There is, however, another remedy to this
problem: the growing instability of the solutions under a feedback with
time-lag is caused by feeding back an increasing amount of control energy
to the system, once it is out-of-phase. In a real structure, however, a
controller will have to respect certain a priori given bounds. There is an
extensive literature devoted to the problem of controllability/stabilizability
by bounded controls. We do not wish to dwell on this question in this
monograph. We refer the reader to a recent paper by Rao [86J and to the
180 V. SPECTRAL ANALYSIS AND NUMERICAL SIMULATIONS
Total Energy
140,--~---~--~--~--~------,
120
r
100
80
l
60
i
401
I
i
20'
Ol----
o 4 6
Total Energy
16"----~------,_----__----~----~----_,
14
12
10
Total Energy
14
12
10
:l
J
OL-----~----~----~------~----~-----
o 4 6
This model consists of a unit square (labels 1,2,3,4) with another string
(#5) attached to the node connecting string #2 and string #3. We have
an initial displacement in the first string only. In the first experiment we
apply an absorbing control only at the free end of the attached string.
Figures V.24 and V.25 reveal what one expects. Energy is taken out of
the circuit until only a residual motion remains. In particular, the plot of
the total energy of the 5th string shows that with each cycle less energy
can be absorbed from the circuit because of the gradual diminishing of
the nodal excitation. The situation completely changes once "directing"
controls are introduced around the vertex connecting strings #2, #3 and
#5. See Figures V.26 and V.27. D
EA -2 EI22 -2
WI = 10( UdXI = 10( {Tcn -2 -2
+ 2GA(CI2 + C13) + -2-"'2
EI33 -2 GI -2}
+ -2-"'3 + 27" dx.
The equations of motion are given by (2.10) which, in the present context,
reduce to
FI,1 + PI = 0,
F 2 ,1 + P2 = 0,
F 3 ,1 + P3 = 0,
MI,1+CI = 0,
M2,1 - F3 + C2 = 0,
M3,1 + F2 + C3 = 0.
Under our assumptions, (2.15) now reads
FI = EAn = EAWI,I,
F2 = 2GA12 = GA(W2,1 - 'l?3),
F3 = 2GA13 = GA(W3,1 + 'l?2),
MI = GIf = GN I ,
M2 = - EI33K-3 = EI33'1?2,1,
M3 = EI22K-2 = EI22'1?3,1.
Sometimes the area A appearing in the shear strains is replaced by
"effective shear areas AX". The continuity condition and the balance laws
184 V. SPECTRAL ANALYSIS AND NUMERICAL SIMULATIONS
4 2
3 1.5
0.5
0 0
0 5 10 15 20 0 5 10 15 20
2
3
1.5
2
0.5
0 0
0 5 10 15 20 0 5 10 15 20
5th Strin
1.5
0.5
0
0 2 4 6 8 10 12 14 16
2
0 2 4 6 8 10 12 14 16
:0 2
(\
3 I.S
0
\ o,SII
0 \'"
0 2 4 6 8 0 2 4 6 8
II
2.S 2.S
2 2
(\
I.S I.S
rl
\
O.S
0
) \
0.5
0
0 2 4 6 8 0 2 4 6 8
Sth Strin
3
0
0 2 3 4 S 6 7
0
0 2 3 4 S 6 7
of force and momentum at the nodes can be read into the global reference
frame by way of the rotations r~ as follows: in the global frame r~ := r~ W~
is the representation of W~. If there is no deflection and rotation prescribed
and if there are prescribed moments MN and forces F N at the node N, we
have
r~W~(N) = r J WJ(N), 'rii,j E (N),
I>~(N)r~F~ + FN = 0,
Lc~(N)r~M~ + MN = O.
If we introduce now the vector quantities
U= , f= , h=
Here
D= (G~2
S
G~3)' Db = (E~22 E~33)
are the shear and bending matrices and
i = (~12)
C13
are the bending and shearing strains, respectively. The quantity u is sought
such that (4.1) holds for all ii in a given space of test functions satisfying
homogeneous boundary and node conditions. In order to derive the finite
element approximation, we follow the presentation in Hughes [39]. We
are looking for an isoparametric description, that is we assume that the
4. FINITE ELEMENT APPROXIMATIONS 187
displacements and the rotations are represented by nodal values via the
same shape functions. We divide the elements of the frame in finite pieces,
i.e., the finite elements, labelled by the superscript e to be distinguished
from the frame element number j. For each finite element we have a number
en of element nodes with displacements wfa' and rotations -&fa' a = 1, ... , en;
the superscript h indicates the dependence on the "mesh", that is, on the
number of finite elements per structural element. We interpolate between
the nodal data using some shape function Na (linear, quadratic or cubic
functions are commonly used). Thus we write
en en
Wih = LNawfa, -&f = L Na-&fa
a=i a=i
At each node there are six degrees of freedom. We may express K. and
"y by the corresponding nodal values as follows:
e
wiha
h
w2a
h
0 0 0 w3a
en 0
=L 0 0 0 0 -N~ ~~) -&~a
a=i
-&~a
-&~a
en (0 N'a o o 0
=L 0
a=i
o -N~o Na
We define
0 0 0
Bba = (00
0
N~)
0 0 0 -N~ o '
N'a 0 0 0 -Na)
B 8a = (00
0 -N~ 0 Na o '
B~x = [N~, 0, 0, 0, 0, 0], B~o = [0, 0, 0, N~, 0, OJ,
Bb = [B~, ... , B: n ], B8 = [B~, ... ,B=nj,
188 V. SPECTRAL ANALYSIS AND NUMERICAL SIMULATIONS
B ax = [Bax
l ' ... , Bax]
en' B
to = [B to to ]
l ' ... , B en
The superscripts ax, to stand for axial and torsional quantities, respectively.
Then the element stiffness matrix k e can be written as
where
Using this technique, linear functions can be used to achieve the accuracy of
cubic shape functions in bending; see Hughes [39]. Once the local stiffness
and mass matrices are defined, they are read into the global stiffness and
global mass matrices using the local rotations ri. We do not go into detail,
because it is a classical procedure.
Let, finally, K, C, M denote the global stiffness, global damping and
global mass matrices, respectively. Then the system of equations governing
the evolution of the time dependent nodal variables can be written as
(4.2) MU + CD + KU = F,
where U is the vector containing the nodal variables of all nodes and F
represents all applied forces. We recall the family of Newmark schemes,
which are of the predictor-corrector type. Again we follow Hughes [39]. To
this end we put
(4.3)
4. FINITE ELEMENT APPROXIMATIONS 189
_ t2
{ ~n+l : dn + t:..tvn + t:.."2(l - 2.8)an,
(4.4)
vn+1 - Vn + (1- ')')t:..t~.
As for starting values, we have
This equation can be solved for ao. At step n + 1 the acceleration a n+1 is
calculated from previous quantities and a priori known data according to
the recursion
(4.5)
Once ~+1 is known from (4.5), we update the predictors via the correctors
The result of (4.6) is then inserted into (4.4) which, in turn, will be used
in (4.5) and so on. The two Newmark parameters can be chosen in various
EA 0 0 EA
T -T 0 0
AS AS AB AS
0 T 2 0 -T 2
0
AS
T
ABe + EI2 0 -2
AS Ase
4-2
EI
(4.7) ke = EA 0
4
0 EA 0
-2 T 0
AS AS AS AS
0 -T -2 0 T -2
0
AS
T
ABe
4-2
EI
0 -2
AS Ase
4
+ EI2
1 0 0 0 0 0
0 1 0 0 0 0
I
e moAt' 0 0 A 0 0 0
(4.8) m =--
2 0 0 0 1 0 0
0 0 0 0 1 0
I
0 0 0 0 0 A
see [39, p. 515], for the planar beam without longitudinal displacement.
With the bar wave velocity c = vE/mo and the beam shear wave velocity
Cs = (GAS)/(moA), the critical time step Ilt can be estimated as
Total Energy
3.5
2.5
1.5
0.5
0.8
0.6r
0.4
0.2
o
o 2 3
0.8
0.6
0.4
0.2
OL---~----~----~----~--~----~----~--~
o 2 4 5 6 7
ones of another element, only the minimal number of variables are retained.
It is actually a major part of the finite element implementation to accom-
modate such a reduced setup. Only after this reduction is a numerical
treatment as described in the last section meaningful. There is a recent
tendency in the literature to treat boundary and node conditions in numer-
ical simulations of multi-body structures as algebraic constraints. One of
the reasons for this methodology is to retain the simple band structure of
the stiffness and mass matrices. To make this more transparent, consider
a network with a circuit. Then, because of the fact that the first and the
last node of that circuit coincide, the part of the global stiffness matrix
containing that circuit will have nonzero entries far away from the typi-
cal band. This makes the procedure of solving equations apparently more
complicated. It appears, therefore, to be appealing to keep a simpler struc-
ture of the global mass and stiffness matrices and account for the couplings
by way of algebraic constraints. This approach is taken in Shabana [99]
for multilink flexible body dynamics, including rotating structures, and has
also be used in the context of multilink rigid body dynamics by Hairer and
Wanner [33] and Lubich [64], [65]. We will not dwell on this point further.
In this section we are mainly interested in applying the methods of
differential algebraic equations to a particular problem which naturally
arises in the context of damping of mechanical structures. The phenomenon
we want look at is dry friction at the joints. It has long been known
by mechanical and civil engineers that dry friction in the joints can be
a particularly helpful source of damping of residual modes in structures.
The idea is very simple. If one has a sufficient amount of dry (Coulomb)
friction, say at a joint, then a considerable amount of energy is necessary to
move that joint or to bring the element 'behind' that joint into oscillatory
motion. It is reasonable to expect that under certain circumstances it
will be possible to achieve "locking" of elements, that is, prevent certain
elements from oscillating, by applying dry friction.
We begin by considering a simple model for this phenomenon, which we
take from Cabannes [12], [13], and modify it later on. In [12], Cabannes
introduced the following model of a string with dry friction. Let w denote
the displacement. Again, we consider the spatial interval to be [0,1]. The
model under consideration is
is in the interval [8n, 8(n + 1)] the motion will stop after time t = n. He
used the method of characteristics to prove his startling results. The model
was later generalized to the 2-d wave equation by Haraux [35], [36], who
provided a treatment of the asymptotic behavior of solutions using the
method of multi-valued monotone operators. The locking phenomenon,
however, was not pursued by Haraux in those papers and, in fact, appears
to be an open problem, at least for the 2-d situation. In the 1-d case we
consider the problem of dry friction at a point of the string rather than
everywhere within the string. This model is also a modification of one
considered in Brasey and Hairer [8] of a violin string struck by a violin bow
locally within the string. The model in [8] is
Here v denotes the velocity at which the bow is moved across the string, Ps
is proportional to the shear force, Pa is proportional to the adhesive forces
(Coulomb frictional forces); these two constants are not necessarily the
same, which gives some control on the string. The function X is assumed
to have local support. In our application it is taken to be a point mass.
By this choice we intend to model dry friction at a joint. To this end we
model is as follows:
W"i= (i)"
W , i = 1,2,
Wl(O) = W2(0) = 0, w 1(1) = w 2(1),
(5.2)
(w l )'(l) + (w 2 )'(1) = -PsZ,
wl(x, 0) = wMx), w2 (x, 0) = 0, wl(x,O) = w2 (x, 0) = 0,
where
_ {sgn(WI(l ifwl(l)::I 0,
Z - E [Pa
- - ,Pal
- I
ese.
Ps Ps
We perform the familiar transformations
force ('kills' the outgoing signal v4(1, t)) at the boundary x = 1 which, in
our model, represents the multiple node. 0
We claim that in case (ii), the motion propagates into the second string.
The latter inclusion says that there is nonzero motion in the second string.
The other cases are handled in a similar way. 0
(5.3)
Y/ = f(y, z),
{
0= g(y),
198 V. SPECTRAL ANALYSIS AND NUMERICAL SIMULATIONS
where f : IRn x IRm 1-+ IRn , g : IRn 1-+ IRn are C 1 and C2 functions,
respectively. Denote the Jacobian off with respect to y, z by fy, fz . If we
differentiate (5.6h with respect to the variable x we obtain
0= gy(y)f(y,z).
Another differentiation with respect to x yields
(5.6) 0 = gyy(y)f(y, z) + gy(y)fy(y, z))f(y, z) + gy(y)fz(y, z)z'.
Obviously, if gy(y)fz(y,z) is invertible in a neighborhood of a solution,
say (y, z), then (5.6) is equivalent to an ordinary differential equation. In
this case for the problem
G(y, z) := gy(y)f(y, z) = 0, G(y, z) = 0,
there exists, by the implicit function theorem, a neighborhood U(y, z) and
a solution z = z(y) of G(y, z) = 0 such that (5.6) holds. See Hairer and
Wanner [33J for the theory of differential algebraic equations. With this
remark in mind we go back to our system, where the case just considered
corresponds to (5.3), (5.4h.
If, now, z is in mode two of (5.4) and reaches the value Pal Ps, it has to
switch to the first case in (5.4) which, together with (5.3), then reduces to
an ordinary differential equation. As a result, we have to switch between
the differential algebraic system (5.3), (5.4h and (5.5) and the ordinary dif-
ferential equation (initial value problem) (5.3), (5.4h and (5.5). Switching
problems of this kind (in particular (5.1)) have been treated by Brasey and
Hairer [8]. They used the half-explicit 5-stage Runge-Kutta method of or-
der 4. We used the Radau5 implementation of the Radau IIa fully implicit
Runge-Kutta method of order 5, also given by Hairer and Wanner in [33].
The corresponding driver was originally written by J. Verscht (Bayreuth)
for the problem (5.1). We made the appropriate changes to use it for the
problem under consideration and also for beam applications (which we do
not reproduce here). The problem of switching between the two models
is a major difficulty which is usually solved by the use of "dense output
formulae" and some kind of extrapolation to find the switching manifold
g(y) = O. See Figures V.33 and V.34.
5. DRY F
RICTION 199
0 .0 5
0\8~~
~
- 0 .0 5
20 o
FIGURE
V.33. D
ry frictio
n : locking
0 .1
0 .0 5o
~:t=:--I-'1-
--'I
_ 0 .0 5
- 0 .1
20 o
FIGURE
V.34. D
ry frictio
n: spillover
CHAPTER VI
It is assumed that the motion of the cylinder obeys the following hypothesis.
Basic kinematic hypothesis. The position vector at time t to the
displaced location of the particle situated at r(xb X2, X3) in the reference
configuration is
(1.2)
1. NONLINEAR ELASTIC MEMBRANES 203
1
ca:{3 = 2(wa:,{3 + w{3,a: + w,a: . w,(3),
(1.3)
C33 = o.
1.1. Equations of Motion. Let sij denote the components of the sec-
ond Piola-Kirchhoff stress tensor. This symmetric tensor relates the stress
vectors t i to the quantities G i through
(1.4)
(see [108, Chapter 3]). If F denotes the body force per unit of undeformed
volume, the three-dimensional conditions for the balance of forces is
where p is the mass of the body per unit of reference volume. Use of (1.1)
and (1.4) in (1.5) allows that equation to be written
(1.6) ij
S,i aj + S,ii{3W ,{3 + sa:{3W ,a:{3 + F = p
W
,tt
(1.7)
where the coefficients of elasticity C ijkl depend only on Xl and X2. The
symmetry of sij and Ckl requires that
(1.9)
Then, in particular, the stresses sij do not depend on X3 and (1.6) reduces
to
(1.10)
This system comprises three scalar equations for the scalar components Wi
ofW.
204 VI. INTERCONNECTED MEMBRANES
(1.13) W =0 on r a , t > O.
Let f denote a given distribution of forces along AD +X3a3, -00 < X3 < +00
which is independent of X3. The condition for the balance of forces along
this part of 8B is
vata = f
which, in view of our previous hypotheses, may be written
(1.14)
1.3. Hamilton's Principle. Let W satisfy (1.10) and (1.14), and let
W = Wjaj be a smooth function defined on n x [0, TJ such that Wlt=o =
Wlt=T = 0, WlrCX(O,T) = 0. We form the scalar produce in IR3 of (1.10)
with Wand integrate over n x (0, T). After some integrations by parts we
obtain
(1.15) is known as the principle of virtual work for the membrane problem
under consideration.
We now recast (1.15) into a form known as Hamilton's principle. Let 8
denote the first Frechet derivative with respect to W. We may write
(1.16)
(1.17)
(1.18)
(1.19) 81 T
[K:(t) - U(t) + W(t)] dt = 0,
where
(1.20)
(1.21)
The mapping (Xl. X2) 1-+ Pi (Xl. X2) is a homeomorphism of Oi onto Pi and
of ani onto BPi, with
REMARK 2.1. As in the last section, Greek indices take values 1,2, but,
unlike before, we do not assume summation convention and Roman indices
are not necessarily restricted to set {I, 2, 3}. The range of such an index
will be explicitly indicated whenever it is not clear fom context. We do,
however, make the convention that
3
L:=L
j j=1
0
Pi n Pk = 0, Vi:l kj
Ui=IPi is a connected set in IR3 j
Pi n Pk is either empty or is a finite union of components, Vi :I k.
Each of the bodies Bi is assumed to satisfy the basic kinematic assump-
tion (1.1): the position vector Ri(Xb X2, X3) to the displaced particle orig-
inallyat ri(xl,x2,X3) is given by
(2.2) Ri(Xb X2, X3, t) = ri(xb X2, X3) + Wi(Xb X2, t),
(Xl,X2) E ni , IX31 < 00,1 = 1, ... ,no
2. INTERCONNECTED ELASTIC MEMBRANES 207
P1 : < sb 82 < 1, 83 = 0,
0
P2 : 0 < 82,83 < 1, 81 = 0,
C1 := P 1 n P 2 : 0::; 82 ::; 1, 81 = 83 = 0,
C2 := P 1 n P3: 1/4::; 82 ::; 3/4, 81 = 83 = 0,
C3 := P 2 nP 3 = P 1 nP3,
so that C 2 = C3 C C 1 . We replace C 1 and C 2 by 61 and 62 , where
61 = C 1 n C 2 = C 2 ,
62 = C1 \ (C1 n C2 ): 0::; 82 < 1/4, 3/4 < 82 ::; 1, 81 = 83 = O.
PI and P2 intersect along J 1 and J3 , while each pair among PI. P2 and
P 3 intersects along J 2
For any joint J of the multi-membrane system define
I(J) = {i E [1, ... ,nll J c BPd.
Thus I(J) is the index set of those Pi which share J. The deformation
of the overall system is constrained by the following geometric junction
condition at each joint J:
(2.3) Wi(pil(p), t) = Wj(pjl(p), t), p E J, i,j E I(J).
Condition (2.3) simply requires that each membrane sharing J have the
same displacement there.
2.2. Dynamic Conditions. The geometric joint conditions reflect the
continuity of the overall structure. In addition, there are dynamic junction
condition8 which represent the balance of linear momentum at the joints.
The simplest way to obtain these is to minimize the Hamiltonian associated
with the structure. In carrying out the calculation, the variation must be
taken with respect to displacements which satisfy the geometric boundary
conditions and the geometric joint conditions. Of course, in addition to the
dynamic junction conditions Hamilton's principle also yields the equations
of motion and dynamic boundary conditions of the structure, as usual.
Set Ai = BPi. We assume that
(2.4)
where Af, AP, Ai are disjoint and relatively open in Ai. This partition of
Ai induces a similar partition of r i := ani into
(2.5)
where
Af = Pi(rf), AP = Pi(rp), A{ = Pi(r{).
The set Af represents the part of Ai that is clamped, i.e.,
(2.6) Wi =0 on rf,
AP is the part of Ai along which "dynamic" boundary conditions will hold,
and Ai is that part of Ai such that
At n Ak =I- 0 for at least one k =I- i.
2. INTERCONNECTED ELASTIC MEMBRANES 209
where Pi (Xl, X2) is the mass density per unit of reference area of Pi. Let
sik denote the stress tensor associated with the i th membrane. The total
strain energy is defined to be
where
(2.7) 81 T
[1C(t) - U(t) + W(t)] dt = 0,
in which the variation 8 is taken with respect to displacements (Wl, ... ,wn)
which respect the geometric boundary conditions (2.6) and geometric junc-
tion conditions (2.3) along each joint Jl, . .. ,Jm . It is fairly obvious that
210 VI. INTERCONNECTED MEMBRANES
we will obtain, in particular, the equations of motion (1.10) for each mem-
brane of the system and the dynamic boundary conditions (1.14) along rp.
Thus we have for each i = 1, ... ,n the equations of motion
for all sufficiently smooth displacements Wi, i = 1, ... ,n, which satisfy
the geometric boundary and junction conditions.
We have
1 r.(Jk)
(Lv~s~ja; + LV~S~{3W:(3)
o:,j 0:,{3
. Wi(x)df
(x = X?k + S(71,72))
= jhak (Lv~s~ja;
o:,j
+ LV~sf{3W:(3) . Wi(pil(p))ds
0:,{3
(p = q~ + STik)
1.1
= { ('"' i io:j a ji
~ Vo:S + '"' ,{3. w i( Pi-1(P )) dJk
i O:{3Wi)
~ Vo:S
A
i
Jk o:,j 0:,{3
Thus (2.10) may be expressed as
f L 1 (Lv~sfja; + LV~sf{3W:(3)
k=1 iEI(Jk) Jk o:,j 0:,{3
. Wi(pil(p))dJk
= f (
k=1 1Jk
fk. Wik dJk.
Since the Jk'S are mutually disjoint and the geometric junction constraints
are local to each Jk individually, the last variational equation can hold if
and only if
(2.12) L
iEI(Jk)
(Lv~sfja; + LV~Sf{3W:(3)
o:,j 0:,{3
(pil(p),t)
Equation (2.12) is a balance law for forces along the joint Jk and we refer
to it as the dynamic junction condition there.
(2.16)
where C;kare the linearized strains in Bi and Ai and J..Li are its Lame param-
eters. It follows from (2.16) that the linearized stress-displacement relation
is
(2.17)
(2.23)
i,k,l,m i,k
for some Co > O.
214 VI. INTERCONNECTED MEMBRANES
H = {W = L Wjajl Wj E L2(V)},
j
with
For W E V, set
it Iv
1/2
( )
e(W) = eJk(W)dV
(2.24)
Lemma 2.1 shows that the right hand side of (2.24) defines a norm on
V which is equivalent to the II . IIv norm on this space. The inequality
(2.24) is highly nontrivial. Many proofs of it have been given under various
assumptions on the region V; a relatively simple proof when V is a bounded
Lipschitz domain may be found in [82J.
where r ;::: -1/2 and (OD)o is any nonempty, relatively open subset of OD.
In particular, one has
H = {W = L Wjajl Wj E L2(On,
j
with
j
n n
V = II Vi, H = II Hi
i=1 i=1
216 VI. INTERCONNECTED MEMBRANES
An element W E V is given by
(2.27)
and
(2.28)
where
(2.29)
where
It follows immediately from Lemma 2.1 and Remark 2.3, applied in each
region ni , that (2.24) and (2.25) remain valid with these new definitions of
V, H and eO.
Let V denote the closed subspace of V consisting of functions W =
(Wi, ... ,wn) such that
(2.30)
(2.31)
Then V is dense in H with compact embedding. The following lemma is
the analog of (2.26) for systems of interconnected membranes.
2. INTERCONNECTED ELASTIC MEMBRANES 217
+ f
k=I
r r
T
Jo Jrik(Jk)
fk . Wik dr
,
vW = (WI, ... , wn) E L2(0, T; V), T > 0.
The form
(2.35)
a(W, W) t
~ eo 12: c!j(W )dO
i=1 n, oj
i
~ 2~21IWII~ - eollWllk, VW E V.
= t (ln,l Fi .
,=1
WidO + [D fi . Widr)
lr,
m
+ 2: [ fk . wik dr, vW E V.
k=1 lrik (Jk)
F i EL 2 (O,T;L 2 (Oi, f i
=0,2=1, ... ,n, -k
f =0, k=l, ... ,m,
WO:= (W5, ,WO) E V, W 1 := (WL ,Wf) E H.
2. INTERCONNECTED ELASTIC MEMBRANES 219
where
G = (PI1F1, ... ,p~IFn) E L2 (0,T;H),
and then applying standard theory. Note that it follows from (2.36) that
W,tt E 0([0, Tj; V') n L2(0, T; H).
Next, let us consider the case of inhomogeneous boundary conditions
and assume that Fi = 0. Introduce the space
One has the dense and continuous embeddings D A eVe H c V' c D~.
Since
t if
form
which satisfies (2.37) for almost all t E [0, T] and for which W(O) = W o,
W,t(O) = WI.
Wo E D A , WI E V.
u i = on rf, i = 1, ... ,n,
Ui(p;l(p)) = u j (pjl (p)), P E J k , k = 1, ... ,m.
"L...J 8 iDti3
,Dt = Pi Wii3,tt,
(3.1) {
JL:~W4 = PiW{ttl X E ni , t > 0,
where i = 1, ... ,n, ~W~ = EDt W~,DtDt and where
8~i3 = JLi(W!,i3 + W~,Dt) + Ai(L W~''Y)8Dti3.
l'
(3.2)
222 VI. INTERCONNECTED MEMBRANES
and
LEMMA 3.1. Let </> be a sufficiently regular solution of the scalar equation
and set
1
Ca(3(t/J) = "2 ('a,(3 + '(3,a),
sa(3( t/J) = 2JLca(3( t/J) + )"Da(3 L cn(t/J),
LEMMA 3.2. Let h = h1al + h2a2 be any W1,oo vector field in 0. and
t/J = 'lal + '2a2, where 'i E HB(n) for some s > 3/2. Then
A lengthy, but straightforward, calculation shows that the right side of the
last equality may be written
and set
Then
~ ~ [ (m . p) [~p'p~" + ~ Va 8
a
'(1/ (0:: )] dI'dt
For completeness, we shall carry out the calculations here. We write sa(3
and ca(3 in place of sa(3(1/J) and ca(3(1/J). One has
(3.11) LiT1
(3 0 0
(m . \7(3 + ~(3)p'IjJ(3,ttdOdt
= p in (m \7(3 + ~(3)(3,tdOI6
p
- 2" 10
r 10r~ ()a (ma~,t) dOdt + 2"p 10r 10r~,t dOdt
T
{)X
T
= p in {m \7(3 + ~(3)(3,tdOI6
+ ~ loT in It dOdt - ~ loT i (m . v)lt dJ'dt,
where mOl = Xa - XOa' Also, with the help of Lemma 3.2 with h =m we
calculate
+~
1/J. (S(1/J)v) dI' - ~ In SOl(3 ("p)ca(3 ("p) dr!.
On r one has
(3.13) L
01,(3
1(m \7(3 + ~(3)s~ dO = L01,(3 {~1r (m v)va a(3a:/: dJ'
0
S
-~ 7: In
T
The identity (3.9) now follows from (3.10), (3.11) and (3.13). 0
3. CONTROLLABILITY OF LINKED MEMBRANES 227
i a
= 0,
{
LV~S~{3(1/Ji)
(3.15)
v: .V'W~ = 0 X E rp, t > OJ
(3.16)
(3.19)
Let E(t) denote the total energy associated with a solution of (3.14):
E(t) = lC(t) + U(t)
where
lC(t) = ~ t1
i=i
Pi 1"\}I:t1 2 dn,
t1
(li
Z(t) = t
i=l
Pi [
10.;
L W~,t(mi . VW~ + W~) dO.
j
PROOF. We apply Lemmas 3.1 and 3.3 with 0 = Oi, m = m i , rjJ = W~,
.,p = .,pi, add the resulting identities (3.6) and (3.9) and then sum over i
from 1 to n. 0
Now let \}I be a sufficiently regular solution of (3.14)-(3.18). For exam-
ple, it will suffice to have
\}I E C([0,T]jyS)nC1([0,T]jys-1)
for some s > 3/2 (see Remark 2.4). The identity (3.20) will be used
to obtain the a priori estimates needed to describe the set of reachable
sr
states. We shall write f3 and c~f3 as abbreviated notations for f3 (.,pi) sr
and caf3(.,pi), respectively. In the right side of (3.21) one may write
3. CONTROLLABILITY OF LINKED MEMBRANES 229
We shall consider separately each of the sums on the right side of the last
equality.
(a) Integmls over rf. Since W = 0 on rf, the sum of integrals over rf
on the right side of (3.21) reduces to
(3.23)
1
2' ~
n
t-l
hv
[ i 2
ii
(m v) PilW,tl - LTcrSi
cr,/3
i cr/3 8.T,i/3
';l"
(h. i - J..ti
(8w.3) 2]
8T i dr.
= f L
k=l iEI(Jk)
1 Jk
f(pil(p))dJk
1
Consider
L (mi. Vi )PiI W:tI 2dr .
iEI(Jk) Jk
We assume that
Then
(3.25)
in which we have again utilized (3.17). It follows from the dynamic junction
conditions (3.18) that the last integral vanishes. Similarly, consider
LEMMA 3.5. Assume (3.22), (3.24) and (3.28). [fiJ! = (iJ!1, ... ,iJ!n) is
a sufficiently regular solution of (3.14)-(3.18), then
Let qo be any point of JR3. It follows from (3.31) that condition (3.28) may
be satisfied by choosing x& to be the projection of qo - p& onto the plane
containing Pi, i E I(Jk), for then
hence
mi(pi1(p. 7'ik = (p - qo) . 7'ik, p E Jk, i E I(Jk ),
is independent of i. With this choice of the points X&, condition (3.24)
takes the form
0l,{3
and therefore
where
Then
234 VI. INTERCONNECTED MEMBRANES
The right side of (3.36) may be written (suppressing the argument "pi in
ST i ("pi) and SlI i ("pi))
LetO<8<landset
(Note that mi. vi must be of constant sign on ri(Jk)') The integral on the
right side of (3.37) does not exceed
3. CONTROLLABILITY OF LINKED MEMBRANES 235
(3.39) L (1i(Jk)ST i =0
iEI(Jk)
and
(3.41) L ai=O,
iEI(Jk)
and
(3.42) L (biv i + cia~) = 0,
iEI(Jk)
When (3.41) and (3.42) are satisfied it follows that for Dsufficiently small,
and
236 VI. INTERCONNECTED MEMBRANES
Let us define
COROLLARY 3.1. Assume (3.22), (3.24), (3.28), (3.41) and (3.42). Sup-
pose further that either
(i) rf '"
0 for at least one index i; or
(ii) rp+ '" 0 for at least one index i.
If \II = (\111, ... ,\lin) is a sufficiently regular solution of (3.14)-(3.18),
then there is a To > 0 such that for T > To there is a constant CT such
that
3. CONTROLLABILITY OF LINKED MEMBRANES 237
Therefore, on rp we have
so that
o:::; (Si( 1/Ji )Ti) . 81/J~ = (2J.ti + Ai)2 - AT (Ti. 81/J~) 2 :::; 8lJ!~ 12
C1
8T~ 2J.ti + Ai 8T' 8T'
The right side of (3.44) is therefore an upper bound for the right side of
(3.43). If rf =/: 0 for at least one index i, it follows easily from Lemma 2.2
that
IZ(t)1 :::; CoE(t) = CoE(O) '" 11(lJ!(O), lJ!,t(O))II~XH' 0:::; t :::; T,
for an appropriate constant Co. We therefore obtain (3.44) with To = 2Co.
On the other hand, if rf = 0 for all i but rf+ =/: 0 for some i we have (cf.
Remark 2.2)
o
238 VI. INTERCONNECTED MEMBRANES
REMARK 3.2. The inequalities in (3.41) and (3.42) will, of course, hold
if mi. Vi ~ 0 for all i E I(Jk). A less trivial sufficient condition for their
validity is that, at each joint Jk, at most one of the mi. Vi, say m q v q , is
positive, mi. vi < 0 if i i- q, i E I(Jk ), and (mq . v q)/ JLq and (m q . v q)/ Aq
are sufficiently small relative to Imi . Vil/JLi and Imi . vii/Ai, respectively,
for i i- q and i E I( J k ). Suppose, for example, that I( Jk) = [1,2, ... ,q]
and set mi = (mi. Vi)/JLi. By utilizing the constraint in (3.41) we may
write
satisfy
{
a 3i :.: a i3 , \.I..
vZ,) = 1, ... ,n,
m-n-1.
3. CONTROLLABILITY OF LINKED MEMBRANES 239
lT t [
(3.51)
~ 2?:
1 n
D
i i
(my) Pi 1'IJI,t I
2
- L TaSia{3u~{3
i ::j.T,i
O-ri - J.ti
( u~3
::j.T,i )
O-ri
2] drdt.
t=1 0 r. a,{3
f m. yk
JJ"
{[8;" _
J.tk
8;k+ 1]
J.tk+1
+ [J.tk (OW~)
Oyk
2 _ J.tk+1 (oW;+1) 2]
Oyk+ 1
_ [J.tk (OW~)2
o-r k
-J.tk+1 (OW;+1) 2] + [ 8;"
o-rk+ 1
_ 8;"+1 ]
2J.tk + Ak 2J.tk+1 + Ak+1
- 2[ Ak S k (-rk . o1/l ) -
2J.tk + Ak " o-r k
Ak+1
2J.tk+1 + Ak+1 " O-rk+1
1
S k+l (-rk+1 . 01jJk+ ) 1
_ 4 [J.tk{J.tk + Ak) (-rk . 01jJk) 2
2J.tk + Ak o-r k
By invoking (3.46) and (3.47) , the first three brackets in (3.52) are seen
to be nonpositive exactly when J.tk ~ J.tk+I, and the fifth bracket vanishes.
The rest may be written
3. CONTROLLABILITY OF LINKED MEMBRANES 241
When IJ.k ~ lJ.k+b it is easily checked that the quadratic form inside the
braces is nonpositive if and only if
REMARK 3.4. The same proof shows that Theorem 3.2 is valid also in
the more general situation where the Ji's are any Lipschitz continuous
curves. Of course, in this situation m . vi need not be of constant sign on
Ji so that it is necessary to assume both m Vi ~ 0 on Ji and (3.49), or
m Vi ~ 0 on Ji and (3.50). 0
THEOREM 3.3. Assume that m = 1 and that qo may be chosen such that
PROOF. Condition (3.55) means that qo may be chosen along the straight
line containing J. When it is satisfied, so are (3.24) (3.41) and (3.42) ((3.28)
holds because of our choice of multipliers) and all integrals over ri(Jk) in
(3.43) vanish. 0
lim o;(p) = 0, P E J,
r-+oo
t a~
~I~
{p - qo - raD . vi = lip - qo - raAIl (8~ + 8;) a~ :::; 0,
~ ~
p E J,
(3.58)
(3.59)
We introduce a seminorm on DA X V by defining
(3.60) FCVxH
.
W:tIED+ E L2(0, Tj .c2 (rp+)),
In the above
.c
and similarly for 2 (r i ,,(Jk )).
Let W = (Wt, ... ,wn) be the solution of (3.1)-(3.5) on (0, T) with
zero initial data
where T > To, and let '11 be the solution of (3.14)-(3.18), (3.59) with
('110, '111) E DA x V. Form the expression
where the "derivatives" on the right side of (3.63) are defined as follows:
/ (8) 2.
~D-
r Jrp-
Jo
r 8\I!i
T 8.
.
A )
Thus
(3.64)
and
(3.65)
REMARK 3.6. Note that the quantity on the left side of (3.65) is not the
t derivative of \I!:t in the sense of distributions. Also, one has (1l 1(rp-, =
1l- 1 {rp-) ifrf f. 0 and, in this case, the left side of (3.64) is the tangential
.
derivative of {8"iJ1 i /8T i )II:;D- in the sense of distributions. 0
246 VI. INTERCONNECTED MEMBRANES
Similarly, we select
(3.66)
THEOREM 3.5. Let the hypotheses of Corollary 3.1 hold and let T > To.
Assume that W o E H, WI E V' and that (3.58) holds. Then there are
controls fi, fk satisfying (3.67), (3.68) such that the solution of (3.1)-(3.5),
(3.61) satisfies
+ t hf-I::I' dI"+
As a consequence of Corollary 3.2 the following controllability result
which, unlike Theorem 3.5, does not involve controls in the joints, is ob-
tained.
THEOREM 3.6. Let the hypotheses of Theorem 3.2 hold, assume that
(3.58) holds and let T > To. Then
(3.70) H x V' c {(W(T) , W,t(T))1 r satisfies (3.67) and fk = O},
where W is the solution of (3.1}-(3.5), (3.61) corresponding to rand
fk = O. Further, if
(3.71)
Then, by the process of ''weakening the norm" (see, e.g., [60, p. 200]), one
may obtain
(3.74)
(3.75) w 2 = 0, wi = 0, x E r C , t > OJ
Eo v~s~,8(w2) = 1,8,
(3.76) {
awl D
J.L2 ay2 = fa, x E r ,t > 0,
(x - xo) . 1/ 1 ~ on r 1 ,
(x - Xo) . 1/2 ~ on rD,
(x - xo) . 1/2 ~ on rC.
{W W2 1 -
3 - 3'
awl ow;
JL1 aI/I = -JL2 01/2 ' x E rb t> 0,
and the other for an analogous transmission problem involving the equa-
tions of plane elasticity. The former controllability result has been obtained
by Lions [60, Chapter VI] in the case where the control is in the Dirichlet,
rather than Neumann, boundary condition (i.e., wi = 13 on rD). In this
case, geometric restrictions on r are unnecessary.
CHAPTER VII
(1.2)
1. NONLINEAR ELASTIC PLATES 253
It follows from (1.2) and (1.2) that the components eij of the strain tensor
are given by
(1.4) W =p - p, U = A3 - a3.
The quantity W(Xb X2, t) is the displacement of the point P(Xb X2) in the
reference surface and U represents the "rotation" of the a3 line (this will
be made precise below). Set
1.1. Equations of Motion. Let sij denote the components of the sec-
ond Piola-Kirchhoff stress tensor and t i be the stress vectors, so that
(1.6)
If F denotes the body force per unit of undeformed volume, the three-
dimensional conditions for the balance of forces is
(1.7)
where p is the mass of the plate per unit of reference volume. If (1.7) is
integrated in X3 through the thickness of the plate the result is
Let
(1.9)
[t 3 - = J1 + IVh212 t+,
h2,ata]x3=h2
_[t 3 + h1,ata]x3=-hl = J1 + IVhl l2 t-,
where t = ntti denote the applied tractions on the upper and lower
surfaces, respectively. Set
where
= mp W,tt + IpU,tt.
where
where
(1.15)
where
(1.18)
Let (VI, V2) denote the unit exterior normal vector to r and (- V2, VI) be
the unit tangent vector to r which is positively oriented with respect to the
vector (0,0,1). Let f denote a given distribution offorces along AD +X3a3,
-hI:::; x3 :::; h2 .The three-dimensional condition for the balance of forces
along the edge of the plate corresponding to AD is
(1.19)
(1.20)
1. NONLINEAR ELASTIC PLATES 257
From (1.7)
where ni~ is the exterior unit normal to 8B (so that (nb n2, n3) is the unit
exterior normal to 8Q). If in this variational equation we consider only test
functions r = Vi + X3U of the form (1.5), it may be written
Conversely, if the variational equation (1.23) holds for all test functions
r = Vi + X3 U as described above, then W, U and the resultant stresses
and moments satisfy the two-dimensional plate equations and edge condi-
tions (1.11), (1.15) and (1.20). The variational formulation (1.23) of the
boundary value problem is the principle of virtual work for the problem
under consideration.
258 VII. SYSTEMS OF LINKED PLATES
(1.27) 81 T
[K(t) + W(t) -U(t)]dt = 0,
(1.28) 11 .
U{t) = -
2 Q
s'JcijdQ.
condition. Because of assumption (1.8) the resultants Nij, Mij ,10:(3 are
now expressed in terms of the strains Cij and various resultants in X3 of the
coefficients of elasticity Cijkl. Consequently, under assumption (1.8) equa-
tions (1.13) and (1.16) comprise six equations for the six scalar components
of the displacements W and U.
W = 0, U =0 on rC,
W = 0, vo:Mo: = c on rS. D
(1.29)
(1.30)
(1.31)
where
C = Ci8i, C = Ci8i.
We assume the geometric constraint (1.22). The necessary conditions that
be stationary are
ow = 0, ou = 0,
which comprise five equations for the five unknowns Wi, Uw
The variational equation 8w = 0 again leads to (1.13) and (1.21) in
which U is to be replaced in terms of u by means of (1.31). We obtain,
therefore
Let us next calculate oue. One has for the first variation of kinetic
energy
where
U'Ut
U3 ,t = - (1 - lul~)1/2'
A u,tU+uU,t (uu,t)(UU)
(Ou U3,t, u) = - (1 _ luI2)1/2 - (1 _ luI2)3/2 .
Therefore
Similarly,
One has
1
Ouca(3 = '2(Ga . OuG(3 + G(3 . ouGa),
1
Ouc3(3 = '2(G3 . OuG(3 + G(3 . OuG3),
where
Consequently,
(0uG 3, U
A) A U tl
= U - (1 _ luI2)1/2 a3,
hence
(1.39) jh2 sa/3G/3' (OuGa, tl) dX3 = [Ma/3(w/3 + a/3) + [O/3U/3]' tl,a
-hl
(1.40) jh2 s3/3G/3' (OuG 3, tl) dX3 = [N3/3(w/3 + a/3) + M 3/3u/3]' tl,a
-hl
~ ~ Utl
- (N W 3,/3 + M U3,/3) (1 -luI2)1/2'
and
where NI , ... ,N4 , Mo: are nonlinear partial differential operators defined
as follows:
U
N4 (u) = 1 -luI 2 '
(1.42)
o
oXo: [MO:/3(w,/3 + a/3) + IO:/3u,/3 + M 3 o:u - Mo:(W, u)]
- [N3/3(w,/3 + a/3) + M 3/3u,/3 - N 3(W, u)] + C,..a,.. - C3N4(U)
o
= at [mpw,t + Ipu,t + N 1 (W3, u)]- N 2 (W3, u),
264 VII. SYSTEMS OF LINKED PLATES
and
(1.43) vo[M,B(w,,B + a,B) + I,Bu,,B + M 30 u - Mo(u)]
= c!-,a!-, - C3N4(U) on rD.
The complete set of partial differential equations and edge conditions
for the functions W and u are equations (1.32) and (1.42), edge conditions
(1.33) and (1.43), and the geometric boundary conditions
(1.44) W = 0, u = 0 on rC.
REMARK 1.2. In addition to (1.29), a kinematic assumption that is stan-
dard in thin plate theory is the Kirchhoff hypothesis that A3 be normal the
the deformed reference surface. Then
A3 = Al X A2 = (W,I +aI) x (W,2 +a2)
IAI x A21 I(W,I + at} x (W,2 + a2)1'
so that U = A3 - a3 is expressed in terms of W,o. The kinetic energy
therefore involves the derivatives W,t and W,to while the strain energy con-
tains spatial derivatives of W up to second order. The geometric boundary
condition u = 0 in (1.44) becomes
W2,1 W3,2 - W2,2 W3 ,1 = W3,b WI ,2 W3,1 - WI,l W3,2 = W3,2 on rC.
When Hamilton's principle is applied in this context, the only variation to
be taken is with respect to W. The result is a nonlinear system of three
scalar equations for the components of W which involves spatial derivatives
of W up to fourth order and of W,t up to second order, together with ap-
propriate edge conditions on rD. When hl = h2 == const., the linearization
of this system in the case of an homogeneous, elastically isotropic mate-
rial is the Kirchhoff plate system; another specialization of it is the von
Karman system (see, e.g., [51, Chapter I] and Section 2.5 below.) 0
1.5. Rotations Associated with Plate Deformation. The kine-
matic assumption for this section is (1.1). Recall that the vectors Ao are
tangent to the deformed reference surface. Let us set
Ao ao + Wio~ Al X A2
(1.45)
to = IAol = lao + Wi:o~I' n = IAl x A21'
so that to is a unit tangent vector and n a unit normal vector to the
deformed reference surface. Introduce a right-handed orthonormal basis
A~ such that
where
Thus
(1.51)
In view of (1.45), (1.49) and (1.50), (1.51) expresses '3 in terms of the dis-
placements, but the relationship is obviously very complicated in general.
We next consider the rotation angles l around the aI-axis and 2
around the a2-axis needed to map a3 to a unit vector along A 3. Since
A3 = UOo f1o. + (U3 + 1)a3, one has
U2 UI
(1.52) tanl = -1 + U3 ' tan2 = 1 + U3 '
The differences Oo - 'Oo are the tmnsverse shear angles.
We now define the vector rotation angle associated with the deformation
of the plate as
(1.53) ~(XI' X2, t) = Oo{XI, X2, t)f1o. + '3{XI, X2, t)a3.
266 VII. SYSTEMS OF LINKED PLATES
The part >ofia measures rotations due to bending and transverse shearing
in the plate, while '3a3 is a measure of rotation due to in-plane twisting of
the plate. Equations (1.45), (1.47), (1.48), (1.51)-(1.53) give the (nonlin-
ear) relationship between the displacements W, U and the vector rotation
angle ~. Let us note that when (X1,X2) is a point on an
we may express
>oao in terms of the normal vector v = V1 a1 + V2a2 and the tangent vector
'T = -V2a1 + V1a2, in which case (1.53) takes the form
(1.54)
where c/> = >ofia.
2. Linearization
The kinematic conditions (1.1) and (1.29) are assumed. It is further
assumed that
and that
p(Xb X2, X3) = P(Xb X2, -X3), IX31 ~ h/2.
It follows, in particular, that mp = O.
2.1. Linearization of Equations of Motion. We are going to lin-
earize (1.32) and (1.42) around the equilibrium Wi = Uo = O. It is apparent
from the structure of these equations that the linearization of this system
about zero is the same as the linearization of the system
N,c;j + Pj = mWj,tt,
(2.1) {
M{3
,0:
- N3{3 + C{3 = I PU{3 "tt
since all of the omitted terms are at least of second order in the displace-
ments.
In order to calculate the generalized stresses Noj and generalized mo-
ments M{3 in terms of the strains, a specific stress-strain relation will be
assumed. We set, following Reissner [88],
1 -
Cl1 = E(sl1 - as 22 ) - ;s33,
C22
1
= _(s22 -as l1 ) _ ~s33,
a
(2.2) E E
1
C33 = ~(s33 - as OO ),
E
= --e s , c13 = 2G S C23 = 2G s
1+a 12 1 13 1 23
c12
~2
1- a
[1 + 5.i7(1E+ a)] 2-J.t + 5.,
=
E [ 5.i7(l+a)] 2 5.
1 - a2 a + E = TJ + .
Note that fi > TJ since a < 1. With these definitions it may be verified that
11 -
= 2fic11 + 2TJc22 + ACkk,
8
= 2TJc11 + 2fic22 + ACkk,
22 -
{ 8
(2.3)
where
J.t = fi - TJ, A = 5. + 2TJ
REMARK 2.1. When E = E and a = i7, then TJ = 0, it = fi = J.t, 5. = A
and A, J.t reduce to the Lame parameters
A_ Ea E
- (1 + a)(l- 2a)' J.t = 2(1 + a)" o
From (1.3) one has
(2.4) j h/2
ca{3 dX 3
h
= "2(Wa,{3 + W{3,a + Wi,a W i,{3) + 24Ui,aUi,{3,
h3
-h/2
268 VII. SYSTEMS OF LINKED PLATES
(2.6)
(2.7)
(2.8)
(2.9)
Similarly, the generalized moments Mo./3 are given by
3
(2.1O) M 11 = h12 [{2J.L +,X ) (Ul,l + Ui,l Wi,l)
3
2.12
( ) M 12 = h12J.L ( Ul,2 + U2,1 + Ui,2Wi,1 + Ui,l Wi,2).
The linearized versions of (2.6)-(2.12) are
The linearized equations of motion are therefore (2.1) with NOtj and MOt/3
given by (2.13) and (2.14), respectively. This system is sometimes referred
to as the Reissner model, or the Reissner-Mindlin model. Let us rewrite it
in a different form. Let S denote the set of 2 x 2 symmetric tensors. We
define a mapping C : S I---t S by
(2.15)
where I is the identity in S. Note that
(2.17)
(2.18)
where NOtj and MOt/3 are given by (2.13) and (2.14), respectively. In terms
of the matrix C these boundary conditions read
(2.20)
(2.22)
270 VII. SYSTEMS OF LINKED PLATES
(2.23) 81 T
[JC(t) + W(t) - U(t)J dt = 0,
where
1 [ h2
(2.26) U(t) = 2 in {Ca{j [c:(w)]C:a{j (w) + 12 Ca{j[c:(u)]c:a,8(u)
+ Ghlu + V'W312}dO,
where VW3 = W 3,aaa. In (2.23) the variation is taken with respect to suf-
ficiently smooth functions which satisfy the geometric boundary condition
(2.22).
(2.27)
We also have
and therefore
, t ai . (al + Wi ,I8i)
al' 1 = lal + Wi ,I8i1
It follows that
- ~{Wi,1 + Wf,2 + W;,1 + W;,2 + tP~ + tPn + tPl W3,2 - tP2W3,1 + ....
From (1.47) and (1.48) we have
tPl = W3 ,2 + ... , tP2 = -W3 ,1 + ... ,
so that
(2.28)
(2.29)
At a point of an,
W 2,1 - WI ,2 =T
&w
. - -
&w
v . -OT
OV
and (2.29) may then be written
=
a
fa - aT ('T . C) - v . C on rD,
where
B2(W3) = VIV2(W3,22 - W3,u) + (V~ - V~)W3,12'
Starting from the Reissner system, a rigorous derivation of the system
(2.32)-{2.35) may be given by passing to the limit as G ~ 00; see [51,
Chapter II] and Section 6 below.
3. LINKED REISSNER PLATES 273
The set 'Pi is considered as the reference surface of a thin plate whose
reference configuration is
where A~ =1= o.
As before, we set A~ = R:a lx3 =o,
.- Wi -- "Wia
P i _ p'z- ~
i
j j'
j
A; - a;:= Ui = LUla;,
j
Wi =L W!a~, u i = L U~a~.
a a
274 VII. SYSTEMS OF LINKED PLATES
might require that J neither bend nor stretch under deformation. The
latter constraint is formulated as
a . . 1
Ji-:-{-r' . w')(pi (p)) = 0, 'lip E J, Vi E I(J).
uT'
= ---(TJ. . uJ)v
..
J + (v
'"
J . UJ)TJ + -21 ( T J. -
{)wj . {)wj) .
. --- v3 . ----. a J
avJ aT) 3
if Ti = Tj on J
(3.9)
ifTi = ---Tj on J.
(3.10)
276 VII. SYSTEMS OF LINKED PLATES
(3.12) aW3
-- -v
aT t
t+ (tv aw
- ) a t =---v
aTt
t
3
awg J+ (J
aTJ
v .aw
aTJ
J J
- - ) a,
3
Tt . (u' + VWa) = O.
The conditions I{J) = [i,j], v t = -vJ, mean that only two plates are
connected along J and that their reference surfaces are coplanar. In this
case we say that the two plates are sertally connected along J.
( )
3.15
aWa
avt (P t-1{p)
) avJ (PJ-1{P ,
= a'J awg \fp E J.
where n t denotes the unit normal vector to the deformed reference surface
Pt. Equation (3.16) states the geometric constraint that the angle between
the normals along J to any two plates meeting there is the same after
deformation as before. This is why such a joint is called semi-rigid. D
3. LINKED REISSNER PLATES 277
Let us now verify (3.12). From the continuity condition (3.2) we obtain
(3.17)
so that
(3.18) , aw' _
V . - a - a'J
(, awJ awg ( ' J})
V - a + - a V a 3 .
T' TJ TJ
Since
aw
aTJ
J aw
J
aTJ
J aw
- - = ( v J - - ) v1+ ( T J - - ) T J ,
aTJ
it follows from (3.17) that
v ,aw'
. --
aT'
= atJ [(_.7 awJ )
V" - -
aTJ
, vJ + -
v - (v' . a J3 )]
awg
aTJ
.
hence
(3.19)
(3.20)
278 VII. SYSTEMS OF LINKED PLATES
where Af, AP, At are disjoint and relatively open in Ai. We then have
the associated partition
(3.21)
where
Af = Pi(rf), AP = Pi{rp), Af = Pi{rf)
The set Af represents the part of Ai that is clamped, i.e.,
(3.22) W
i i i
= 0, U = 0, W3 = on r Ci ,
AP is the part of Ai along which "dynamic" boundary conditions hold, and
Af consists of those joints of the multi-link system which belong to Ai. We
label the entire collection of joints by J}, J2 , .. ,Jm . Obviously, geometric
constraints other than, or in addition to, (3.22) could also be considered
(cf. Remark 1.1).
Let Ai, JLi, Gi , mi, IPi be the elastic parameters associated with the plate
whose reference surface is Pi and set
(3.23) 6ln
T
[~(t) - U(t) + W(t)] dt = 0,
where
where
a
In addition to equations (3.24), (3.25) and the clamped conditions (3.22),
the following variational junction condition must be satisfied:
for all displacements vi + W~a; and "rotations" u i which satisfy the geo-
metric junction conditions. By employing the notation of Section 2.2, (3.26)
may be expressed as
280 VII. SYSTEMS OF LINKED PLATES
(3.27) L
iEI(Jk)
1 h
({Ci[e(wi)]v i + Gihvi . (ui + VW~)an Wi(pil(p))
(3.28)
L {Ci[e(Wi)]Vi + Gihvi . (ui + VW~)a~}(pil(p)) = 0, Vp E Jk,
iEI(Jk)
1
(3.30)
L {Ci[e{Wi)]V i + Gihvi . (ui + VW~)an Wik{pi;;l{p)) dJk =
iEI(Jk) Jk
(3.31) L
iEI(h)
l ak
bk
{Ci[e{Wi)]V i + Gihvi . (ui + VW~)ands = 0,
3. LINKED REISSNER PLATES 281
where
J k = {q~ + s-rikl ak ~ s ~ bk }.
The first is a balance law for the average of forces along Jk and the second
is a balance law for the angular momentum with respect to the point q~.
for all sufficiently regular test functions W, iI which satisfy (3.2), (3.10) and
(3.14). We resolve Ci[(Ui)]v i into its normal and tangential components:
Ci[(Ui)]v i = Clli(Ui)v i + CTi(Ui)-ri,
Clli(Ui ) = vi. Ci[(ui)]V i , CTi(Ui ) = -ri. Ci[(Ui)]V i .
Equation (3.33) may be written in terms of these components as
(3.34) L
iEI(Jk) Jk
1 ({Ci[(wi)]V i + Gihv i . (ui + VW~)an Wi
, . [) .
+ W3[)-riCTi(U') ]) dJk = O.
We have integrated by parts along Jk to obtain the last integral, utilizing
test function W~ which vanish at the ends of Jk. By utilizing the geometric
constraint (3.2) we may conclude from (3.34) that
and
(3.36)
L 1
iEI(Jk) Jk
[(Vi. iIi)Clli(Ui ) + (-r i . iIi + ~!!) CTi(ui )] dJk = O.
The i th term on the right side of (3.35) represents the force exerted on the
joint due to twisting of the i th plate about vi In addition, by utilizing
282 VII. SYSTEMS OF LINKED PLATES
(3.37) L (J'i(Jk)C"'(U i ) = 0,
iEI(Jk)
Let us note that when (3.38) holds, ie., when two plates are serially con-
nected, the right side of (3.35) vanishes since Ti = (J'iiTi and a~ = -(J'iia~
in this situation, so (3.35) reduces to
C".(ui ) = (J'iiC,,;(Ui )
which, when combined with (3.38), yields
for sufficiently regular test functions satisfying (3.2), (3.3), (3.10) and
(3.14). Unlike the case of a semi-rigid joint, it is not permissible to utilize
test functions such that W~ = 0 at the ends of Jk, since Wi must be linear
along the joint. From (3.2) and (3.30),
Wi(pil(p)) = C~ + scl, (p = q~ + ST ik ),
3. LINKED REISSNER PLATES 283
aw~(
~ Pi-l())
P = Jk 1
(Ji ()C k . a3i
uT l
along Jk. We then deduce from (3.41) that (3.37), (3.38) must hold and
that
L
iEI(J,,)
l
a"
bk
(s{Cdc(wi)]v i + Gihv i . (ui + VWi)an
h2 . .)
- 12(Ji(Jk)Cr i(u l )a; ds = O.
3.4. Junctions With Masses and Applied Forces. In the deriva-
tion of the dynamic junction conditions it is tacitly assumed that the junc-
tion regions are massless and that no external forces act there. Suppose
now that there is a linear distribution of point masses along the joint Jk,
and let mk denote the linear mass density along Jk. Furthermore, let fk and
ck denote a distribution of forces and couples, respectively, applied along
J k . One may imagine that fk and ck are applied along J k to the edge of
a particular plate and that the masses are likewise attached to that plate.
These forces and couples, and the energy due to the mass distribution,
are then transmitted to other plates of the system through the connection
conditions along Jk. For definiteness, we suppose that fk, ck and the mass
distribution are associated with the plate of index ik = min{il i E I(Jk)}'
(In fact, as explained above fk and ck are resultants in z of a distribution
offorces and couples applied over the full edge Ek := Jk x [-h/2, h/2] a;" ,
and mk is the z-resultant of a planar density of point masses distributed
over Ek') The effect of the distribution of masses and applied forces and
couples is to modify the kinetic energy K of the system and the work W
done on the system. The new expressions are
284 VII. SYSTEMS OF LINKED PLATES
+f 1r{
k=1 'k(Jk)
(fk.Wik+ck.uik)dI'.
One may now proceed to derive the dynamic conditions via Hamilton's
principle. The details of the new calculation, which is similar to that carried
out in the last subsection, will not be provided. The only change is in the
variational junction condition (3.27) which, in the present situation, is
(3.42) L
iEI(Jk) Jk
1 ({Ci[c(Wi)]Vi + Gihvi . (ui + VW~)an Wi(pil(p))
(3.45) L
iEI(Jk)
lak
bk
{Ci[c(wi)]v i + Gihv i (ui + V'W~)ands
and
(3.46) L
iEI(Jk)
lak
bk
s{Cdc(wi)]vi + vi. (Ui + V'W~)ands
2
-_ - h 0 Cr' (U i) a3i
12 "L..J o-ri + r-k - -
mk
Wik
,tt
iEI(Jk)
3 2
H = EBHO(o; raj]) EBHO(O; [llo]),
j=l a=l
3 2
V = EBH1(O; raj]) EBH1(O; [llo]).
j=l a=1
where C a,8[c()] and ca,8(') are defined in (2.15) and (2.17) and w =
La Wallo It follows from Korn's lemma (see Section 2.4) that there is
a constant K > 0 depending only on 0, h and the elastic parameters such
that
One may replace IIRIIH on the right side of (4.1) by any continuous semi-
norm p(.) defined on V such that I1(R) = 0, p(R) = 0 implies R = 0 (cf.
Remark 2.2). In particular,
4.1. Function Spaces for Linked Plates. For i = 1, ... ,n, let ni ,
al, a~, a~ be as defined in Section 3. We form the spaces H and V above
in which n is replaced by ni and aj by a~, and denote these spaces by Hi
and Vi, respectively. We then set
n n
H= II Hi, V= II Vi.
i=l i=l
and
where
(4.9)
(4.10)
(4.12)
so that, in particular,
(4.13)
(4.14)
(4.16)
and
(4.20) L O'i(Jk)Clli(Ui ) = O.
iEI(Jk)
We identify the Hilbert space H with its dual space. Since V is dense in
H with continuous injection, we may write V c H c V', where V' denotes
the dual space of V. The duality between elements REV and R' E V' is
denoted by (R', R)v. If R' E H, one has (R', R)v = (R', R)H, VR E V.
290 VII. SYSTEMS OF LINKED PLATES
Let a(R, it) denote the symmetric bilinear form associated with the
seminorm e, that is
n
a(R, it) = L a,(R\ it'),
,=1
where
(4.23)
where
(R,tt, it) + c5(R,t, it)H + u{R, it) + c5 2 (R, it)H = (H, it)H, V'it E V,
where H(t) = e- 6t G(t) and where, according to (4.1), the form u(,) +
c52 ("
.) H is strictly coercive on V. In either case, standard variational theory
gives the following result.
Then
292 VII. SYSTEMS OF LINKED PLATES
t lD
i=1 ri
(i. Wi + ci . ui ) df = (g,R)v, 'VB. E V,
WI = W 2 , u l = u 2 ,
(5.4)
Let
and define
(5.9)
(5.10)
We shall write
Let V' denote the dual space of V with respect to H and A denote the
Riesz isomorphism of V onto V'. The weak formulation of (5.6)-(5.9) is
i=l {l;
. A A A1 A2
+ (Cd(I;")],(l;'i))}dn= 0, VcfJ=(cfJ ,cfJ )EV,
5. CONTROLLABILITY OF LINKED REISSNER PLATES 295
i
+ Z'k.
A A A
Let (C)o, C)1) E V x Hand C) be the solution of (5.6)-(5.10). Then there is
a To> such that for T > To,
REMARK 5.1. A similar result has been proved in [46J in the case where
the controls act through the Dirichlet, rather than Neumann, boundary
r
conditions on D , i.e., u 2 = C, W 2 = f. In that situation no restriction on
the geometry of r is needed and one obtains the sharper result H x V' =
RT. 0
= Oi x (O,T), ~i = r i x (O,T).
Qi
LEMMA 5.1. Let ~ = (~1, ~2) be a sufficiently regular solution of (5.6)
and define
1
Q.
0=1 Q.
(m V'Zi + t5Zi )[m,ii - Gihdiv{V'Zi + rp')J dQ
5. CONTROLLABILITY OF LINKED REISSNER PLATES 297
1 Q,
(m VZi + 6Zt)mi Zt dQ = Pi(t) 16'
+ (1 - 6) [ mil Zi l2dQ - ~i
lE, (m. v )IZ I dE.
i i2
lQ, [
Consider
1 Q,
(V/pim + 6/pi) . {Ip, <Pi - div Ci[c(/pi)]} dQ
=L
~
1~
(m VCP~ + 6cp~){Ip.<p~ - L C~~[c(c,oi)]} dQ,
Q
Keeping this last equality in mind, one may proceed in a manner identical
to the proof of Lemma 3.3, Chapter VI, to show that
L l (m.V'cp~+8cp~){Ip.<p:a- LC~~[c(/pi)]}dQ
{3 ~
1
Q
(5.16)
In.
r [(V'<p'm + (1- 8)<p') (V'Z' + <p')
- (m . V' Z' + 8Z') div(V' z' + <p')} dO
= 8 r (IV'Z'1 2 -1<p'12) ~2 Jr.r(m. v') {(OZ'
In.
dO _
ov,
+ v' . <p,)2
OZ'
- ( -+T'<p , ,)2 - 2 (OZ'
-+vcp , ') ('v'<p')} dr
aT' ov'
-1 r.
(OZ' + v' . <p')
OV'
[(m' T'/Z'
aT'
+ 8Z'] dr.
5. CONTROLLABILITY OF LINKED REISSNER PLATES 299
By using the boundary conditions (5.8) and (5.9), it is seen that the inte-
grals over 1: D and 1: c reduce to .1(1: D ) and .1(1:C ), respectively. On r 1
we have v 2 = _VI, -r2 = _-rl. Therefore
L 1(m vi)(miIZiI2 +
2
Ipi l<piI2) d1:
i=1 ~l
2 . { . . ocp'. . . ocp'.
'" { (m v') (Ci[e(cp')]V')' - . - (Ci[e(cp')]-r') . - .
,=1 J"'.
~ u.
oV' o-r'
5. CONTROLLABILITY OF LINKED REISSNER PLATES 301
By noting that
and that
o
COROLLARY 5.1. Let (cpl, cp2) be a sufficiently smooth solution of (5.6)-
(5.9). Suppose that the geometric conditions (5.12) hold and that Pl :s: P2,
G1 2: G 2 Then
a2 a2
(C2[c(~2)lv2). a~2 = L v~C~{3[c(~2)1 a:~
(5.20) a,{3
= L C~{3[c(~2)lca{3(~2) ~ 0,
a,{3
while on rD ,
a2 a2
(C2[c(~2)1'T2). a~2 = LT~C~{3[c(~2)J a:~
a,{3
ar.p2
= L C~{3[c(~2)]ca{3(~2) - L v~C~{3[c(~2)1 av~
a,{3 a,{3
=L
a~2
C~{3[c(~2)Jca{3(~2) - (C2[c(~2)lv2). av2
a,{3
= L C~{3[c(~2)Jca{3(~2) ~ o.
a,{3
LEMMA 5.3. Suppose that (~o, ~t) E D A X V and let ~ = (~1, ~2) be
the solution of (5.6)-{5.1O). Assume that the geometric conditions (5.12)
hold and that r C and r D meet in a strictly convex angk Suppose further
that
P1 ~ P2, G1 ~ G2, ILl ~ IL2, ).1 ~ ).2
PROOF. Our assumptions assure that the solution ~ has the regular-
ity required in the two previous lemmas. We first note the elementary
inequality
~
2
i=l
1 Q.
{(I - o)milZil2 + Mp.l~iI2
provided
(5.23)
(5.27)
and that the transmission conditions on r 1 are the same as (5.26) if Si(cpi)
is replaced by Ci[c(cpi)]. Thus, comparing (5.27) with (5.25), the same
argument used in the case of membranes applies in the present context and
we may conclude that (5.22) is valid provided (5.23) holds. D
Proposition 5.1 now follows from the last lemma by ''weakening the
norm" (cf. Lions [60]).
t 1n.
t=l
(m,W' vr - Ip. u' V'W3)dO = 0,
which is formally equivalent to the system
= w' (w').1 = 0
w'00 '00 ,
Wai E H 1(Oi'.
) W3i E H 2()
Oi,
aW3. = on r Ci ,
(6.2) {
i
W = 0, -a
v'
Wi(Pi1(P)) = Wi(pj1(p)),
(6.3) {
aw~ -1
avi (Pi (p)) =
aWl (Pi-1( P)) ,
(1ii avi
Vp E Jk, Vi, j E I( Jk), k = 1, ... ,m.
(6.4)
where it stands for R,t, R for R,tt, and where the superscript G reflects
the dependence of the solution on the shear moduli. The initial conditions
are
(6.5)
6. LINKED KIRCHHOFF PLATES 307
(6.6)
(In (6.6) and in what follows, we shall write L2(H), L2(VG), C(H), etc., to
mean L2(0, T; H), L2(0, T; VG), C([O, Tj; H), etc., where it is understood
that T > is fixed.)
REMARK 6.1. The hypotheses on the initial data mean that each plate
is initially free of transverse shear and has finite (Kirchhoff) kinetic en-
ergy. Thus they are assumptions of consistency with the Kirchhoff plate
model. 0
With assumption (6.6), the problem (6.4), (6.5) has a unique solution
RG E C(VG) with nP
E C(H) and ii E L2(V a).
We may decompose RG
and nP
into
G G G.L G.L .L
where Roo, Roo E C(Hoo) and (Roo) ,(Roo) E C(Hoo). Similarly, we
write
G = Goo + G~, Goo E L2(Hoo), G~ E L2(H~).
Let (ii~,)(t) E V:x, denote the restriction of iiG(t) to Voo. If R E W1(Hoo)
satisfies R(O) = R(T) = 0, one easily verifies that
so that
We write
G = (Gl, ... ,Gn ), G i = (m:;lFi,Ip, C i ),
and R = (R\ ... ,Rn) with
Then
We have
+ I: 1r, W~vhc~/3[C:(Wi)) dr
a,/3
We may deduce the dynamic boundary and junction conditions from (6.3)
and (6.16). We write V'Wa and Cdc(V'wDJv i in terms of normal and
tangential components:
310 VII. SYSTEMS OF LINKED PLATES
where
Ci[c(Wi)JVi = 0,
C". (VW~) = 0,
(6.18) . 2
aw~ a . h a .
I p'a-. -Di -a .(~W3)-12-a. CT' (VW3)
v' v' -r'
= _vi. Ci on rP .
It may be verified that these are the same boundary conditions which were
formally obtained in Section 2.5.
The dynamic junction conditions along a junction Jk also follow from
(6.17). They are, in variational form,
(6.19)
We have utilized (6.3) in order to obtain (6.19) from (6.17). The dynamic
6. LINKED KIRCHHOFF PLATES 311
(6.20) L
tEI(Jk)
{CI[C(WI)]V I + [Ip< ~:! - 8~1
Dl (LlW3)
and
In each of the last two equations, the i th term of the sum is evaluated at
p;l(p), p E Jk' Equation (6.20) is a balance law for the transmission of
linear momentum along Jk, while (6.21) describes how bending is transmit-
ted there. Therefore, the dynamics of a system of linked Kirchhoff plates
with semi-rigid joints are described by the equations of motion (6.15); the
geometric boundary conditions (6.2); the geometric junction conditions
(6.3); the dynamic boundary conditions (6.18); and the dynamic junctions
conditions (6.20) and (6.21).
There is a very simple geometric interpretation of the junction condition
(8W3I8v') = ulJ (8Wi/8vJ). Suppose that the reference surfaces 'P, and
'PJ are joined along J. The cosine of the angle between 'PI and 'PJ is a3 . a~.
On the other hand, the cosine of the angle between the deformed reference
surfaces is
where n" a unit vector normal to the deformed reference surface 'P" is
given by
Ai x A2
I
(6.22)
Condition (6.22) obviously means that the angle between 'PI and 'PJ is
the same after deformation as before.
312 VII. SYSTEMS OF LINKED PLATES
(6.33)
Statement (6.28) implies that t ~ RG(t) : [O,T] ~ H is equicontinuous in
G> O. Further (6.28) and the first assumption in (6.6) imply that
RG is bounded in LOO(H),
314 VII. SYSTEMS OF LINKED PLATES
so, in particular,
(6.34) wi,G is bounded in VXl{HO{fh)).
Taken together, (6.33) and (6.34) give
(6.35)
which, when combined with (6.32), yields
(6.36) Rl,G is bounded in VXl{Vl).
Suppose that P2 is connected to P l along a joint J. We wish to use
(6.36) together with the geometric junction conditions (6.1) to deduce that
(6.37)
Once (6.37) is established, we may conclude by iteration that
(6.38) RG is bounded in LOO{V).
From (6.36) and (6.1) with i = 1, j = 2, we obtain
W 2,Glr2 (J) is bounded in L OO {H l / 2{r 2{J)),
u 2 ,Glr2 (J) is bounded in L OO {H- l / 2 {r 2 {J)).
It then follows from (6.29), (6.30) and Korn's lemma (see Remark 2.2) that
W~,G, U~,G are bounded in LOO{Hl{fh))
and then, as above, that
W;,G is bounded in L OO {H l {(2)).
This proves (6.37).
We may now conclude from (6.28), (6.38) and (6.31) that upon passing
to the limit as G - 00 through an appropriate subnet of G > 0 we obtain,
for some R E LOO(V),
RG _ R weak* in LOO{V),
itG _ it weak* in LOO{H),
u~G + vw;;~ _ 0 strongly in LOO{HO{Oi)).
Let it E L2(Voo ) and take the scalar product in H of (6.4) with it. One
obtains
Therefore
It follows that
Thus
it~ ~ it weakly in L2 (V:,.J.
By the Arzela-Ascoli theorem,
R~ ~ R strongly in L 00 (V:.c') ,
which is just the variational form of {6.7}. On the other hand, the problem
(6.7), (6.8) has a unique solution with R E C(Voo) n CI(Hoo }. Therefore,
convergence of R G to this solution is through the entire net G > O.
where e2, el are fixed but arbitrary vectors in 1R3. We therefore obtain
from (6.45) the nonlocal balance laws
(6.46)
6. LINKED KIRCHHOFF PLATES 317
(6.47)
Finally, the dynamic junction conditions at a rigid joint are (6.46), (6.47)
and (6.21).
CHAPTER VIII
1. Introduction
angles associated with deformation of the beam. Thus the linearized model
involving the latter quantities has six degrees of freedom. This model con-
nects the Reissner-Mindlin plate equation to the Timoshenko beam equa-
tion and the linearized equation of torsion in a rod, with all variables cou-
pled through the geometric and dynamic junction conditions. In order to
obtain a simpler model, in Section 5 we pass to the limit in the original
model as the shear modulus of the plate and the shear modulus of the
beam both approach infinity. This process leads to a model in which trans-
verse shear is not present in either the plate or beam and in which the
torsional angle is spatially constant and amounts to a (time dependent)
rigid rotation of the beam about its center line. The limit model is a type
of hybrid system, involving coupled partial and ordinary differential equa-
tions. It has only three degrees of freedom and couples a Kirchhoff plate
to an Euler-Bernoulli beam with the transverse displacements of each and
the spatially constant torsional angle coupled through the geometric and
dynamic junction conditions.
Finally, in Section 6, a distributed parameter model is derived for the
plate-beam configuration in which the centerline of the beam is orthogonal
to one of the faces of the plate in the equilibrium configuration. As usual,
this model is obtained by imposing ad hoc kinematic and geometric con-
traints on the deformation and then applying Hamilton's principle. This
model turns out to be much more complicated than the one derived in
Section 2 because all variables (displacements and rotation angles of the
plate and beam) are coupled. It is clear that other physically plausible
constraints could be imposed instead of, or in addition to, the ones pro-
posed below and that each different choice would lead to a different set of
dynamic junction conditions. This fact emphasizes the necessity of further
research devoted to simulations and model validation; for example, it would
be extremely interesting to compare spectral values predicted by the model
derived below with experimentally obtained spectral data.
where summation convention for repeated indices has been assumed. Recall
that repeated Roman indices are summed from 1 to 3, while repeated Greek
indices are summed from 1 to 2.
322 VIII. PLATE-BEAM SYSTEMS
(A4) Set
Then Ja C r.
The region Ja x (-h/2, h/2) is the junction region. The above assump-
tions mean that the junction region is fiat, that the beam is prismatic and
its centerline is orthogonal to r at Xo and lies in the Xlx2-plane. In terms
of the ei basis, the junction region is given by
The assumption (A3) that the cross sections of the beam be rectangular
is unessential. At the cost of a slightly more complicated formalism, we
may just as easily consider beams with doubly symmetric cross sections w
such that {xo + 6e2 + 6e31 (6,6) E w} is contained within the lateral
boundary of the plate.
We denote by R(XI' X2, X3, t) the position vector at time t to the de-
formed position of the material particle in the plate which is located at
(Xl, X2, X3) when the plate is in equilibrium, and by W(Xl, X2, t) the dis-
placement vector of the material particle which occupies position (Xl, X2, 0)
in the reference surface in equilibrium. Similarly, r(6, 6, 6, t) will denote
the position vector to the deformed position of the particle originally at po-
sition Xo + {iei of the beam in equilibrium, and w(6, t) the displacement
vector of the point Xo + 6 el on the centerline of the beam. The kinematics
of the plate and beam are assumed to obey the following hypotheses (cf.
Chapters III and VII).
Basic kinematic hypotheses.
(2.1) R(XI' X2, X3, t) = x(3a(3 + W(Xl, X2, t) + X3A3(Xl, X2, t),
(Xl, X2) En, IX31::; h/2,
2. MODELING OF THE PLATE-BEAM JUNCTION: I 323
E3 = (sin {}1 sin {}3 + cos {}1 sin {}2 cos {}3 )e1
+ (cos {}1 sin {}2 sin {}3 - sin {}1 cos {}3)e2 + (cos {}1 cos {}2)e3.
REMARK 2.1. Assumption (2.2) is not entirely consistent with the in-
troduction of torsion since the latter may induce cross-sectional warping,
as discussed in Chapter III. However, as noted there, warping is considered
negligible if the beam is thin and the torsional rotation is small, which is
the situation we shall eventually consider.
We now define the geometric junction conditwns. Consider a point Xo +
6'T(xo) + 6a3 within the junction region. We require that
a h
(2.4) R(Xo + 6'T + 6 a 3, t) = r(O,6,6, t), 161 < 2' 161 < 2'
324 VIII. PLATE-BEAM SYSTEMS
W/3(Xo +6T, t)a{3 = W{3(O, t)e{3 +~2[(sin '19 1 sin '192 cos '193 - cos '191 sin'l93)e1
+ (cos '19 1 cos '193 + sin '19 1 sin '19 2 sin '193 - 1)e2](0, t),
U(Xo + 6T, t) = [(sin '19 1 sin '193 + cos '191 sin '192 cos'l93)e1
+ (cos '19 1 sin '192 sin '193 - sin '191 cos 'I9 3 )e2] (0, t).
If we now linearize these last relations around the equilibrium state we
obtain
W3(XO + 6T, t) = W3(O, t) + ~2'19:(O, t),
(2.6) {
u(xo + 6T, t) = -fl(O, t), 161 < 2'
and
where
-fl = 'I9 2 e1 - '19 1e2.
It should be noted that in (2.6) and(2.7) there is no coupling between the
variables WI', wI', '193 related to motion in the a1a2-plane and the variables
W 3, U, w3,-fl.
If we recall the expression (1.52) relating the rotation angles >/3 of the
plate to the U/3, then the second equation in (2.6) is (after linearization of
(1.52) )
>2a1 - >1a2 = 'I9 1e2 - 'I9 2e1
and so expresses the continuity of (vector) rotation angles across the inter-
face.
We shall take linearized expressions (2.6) and (2.7) as the geometric
junction conditions of our problem.
2. MODELING OF THE PLATE-BEAM JUNCTION: I 325
where K, and U represent the kinetic and strain energies, respectively, of the
plate-beam system, W is the work done on the system by external forces
and 8 denotes the first variation with respect to the class of admissible
displacements. The latter must obey, in particular, the geometric junction
conditions (2.6) and (2.7). They must, in addition, satisfy any other ge-
ometric restrictions that are imposed on the motion. Since W 3 , u, W3, iJ
are not coupled to the quantities W,B, w,B, {)3 in the geometric junction con-
ditions (nor will they be coupled in the geometric boundary conditions
imposed below), in considering the dynamic conditions related to these
quantities we need only take into account those portions of K" U and W
which involve W3 , U, W, iJ. We shall derive the dynamic conditions only in
this situation and leave the (similar) derivation of the dynamic conditions
related to motion in the ala2-plane to the reader.
In what follows we shall usually write W and W in place of W3 and W3,
-D =<J -
respectively, and ~ instead of 6. We assume that r = r u r u J, where
r D , r C and J are relatively open in r and mutually disjoint. For later
(2.9) W = 0, U = 0 on r C,
while r D corresponds to mechanical boundary conditions. We further as-
sume that J is connected and that J a C J. The part J \ Ja may be either
free, or it will be geometrically restricted in the following manner. Let ( be
a COO (J) function such that
((x) =6 if x = Xo + 6r(Xo) E Ja ,
and let ai, i = 0,1, be Co(J) functions such that ai = 1 on Ja . It is
required that
where ICp and ICB are the kinetic energies of the plate and beam, respec-
tively, related to the quantities W, u, W, iJ, and similarly for Up,UB and
Wp, WB' We assume that the plate is homogeneous and that the energy of
the plate is that associated with linear Reissner-Mindlin plate theory (see
(2.24), (2.26), Chapter VII). Thus we set
. 2
(2.11) ICp{t) = 21 in[ ( mllWI + IPlliIl 2) dO,
where ml is the mass of the plate per unit of reference area, IPl = m 1 h 2 /12,
G 1 is the shear modulus of the plate,
h3
G[cJ = 12 [>'1 caa I + 2PlcJ,
1
c(u) = 2[Vu + (Vu)*],
and (G[c] , c) = Gaf1[cJcaf1'
The total potential energy of the beam is given by
folL ~~~,
where cP = {1/2)Siicii is the free energy per unit ofreference volume. Here
the strains cii are given by
1
c
t]
= -2(r, t.. r ,]. - 8t ].)
Thus the linearized strains are found to be
2. MODELING OF THE PLATE-BEAM JUNCTION: I 327
where JL2,)..2 are the Lame parameters of the beam. It follows that q> is
given by (cf. Chapter III)
E 2 2 2
q> = "2eu + 2G2(e12 + e13)'
where G2 = JL2 is the shear modulus and E = (2JL2 + )..2) is the effective
Young's modulus of the beam. The strain energy density is therefore
~ 10
1
1
P2IrI 2dwd{,
where P2 is the mass of the beam per unit of reference volume. It is easily
found that
11 1p2lrl2dwd{
= 11 P2A [lwl2 + ~~IJ312 + ~~IJ212 + (~~ + ~;) IJlI2] d{,
hence
(2.14) KB(t) =
1((m21wl 2 + P2hlt?21
210 '2 '2
+ P2 I It?11 ) d{,
where m2 = P2A is the mass of the beam per unit of reference length.
To obtain the work done on the plate-beam system the applied forces
have to be specified. Let Fp (XI, X2, X3) be a distributed body force acting
328 VIII. PLATE-BEAM SYSTEMS
-h/21r D
Consider the resultant forces
h/2 jh/2
Fp(Xl,X2) = j Fpdx3, f p (Xl,X2)= fpdx3,
-h/2 -h/2
and moments
h/2 jh/2
Mp(Xt,X2) = j X3Fpdx3, mp(xt,x2) = Xip dx3 0
-h/2 -h/2
The portion of the work done on the plate related to its transverse motion
may then be written
where
Similarly, let F B (e, 6,6) be a distributed body force acting on the beam
and B(6,6) be a force distributed over the cross-section at = lo The e
work done on the beam by these forces is approximately
and set
The quantity M23 is a twisting moment around et, Ml3 is a bending mo-
ment about e2 and similarly for m23 and m13. Then the portion of the
work done on the beam related its transverse motion and torsion is
where
F2 = e3 F B , h = e3 fB,
M2 = Ml3e l - M 23 e 2, m2 = ml3el - m23e2
+ l V (C[c(u)]v) dr.
ml IV - G1hdiv(V'W + u) = F 1,
(2.18) {
Ipl ii - divC[c(u)] + G1h(V'W + u) = MI.
G1h(V'W + u) v = iI,
(2.20) {
C[c(u)]v = mI.
330 VIII. PLATE-BEAM SYSTEMS
Boundary conditions at ~ = f:
for all test functions W, 11, W, iJ which satisfy (2.10), where iJ = t?2el-t?le2.
The following junction conditions may be deduced from the last equation
and (2.10).
Dynamic junction conditions:
-ale2 (C[c(u)]v)}dr,
(2.22)
Eh19~(O) = i aIel (C[c(u)]v) dr,
+ Glh(VW + u) = 0,
I Pl ii.. - divC[(u)]
(3.1) {
mlW - Glhdiv(VW + u) = in 0;
m2W - G2A(w' + iJ 2)' = 0,
{
(3.2) P2I~~2 - EIhiJ~ + G2A(w' + iJ2) = 0,
P2IiJi - G2IiJT = 0, < ~ < l;
(3.4) W = 0, u =0 on rC ;
(C[(u)]v)}dr,
i
-0'Ie2'
(3.6)
EhO~(O) = O'lel' (C[(u)]v) dr,
When the controls are in the mechanical boundary conditions the remaining
boundary conditions are
G1h(VW + u) . v = /,
(3.7) {
C[(u)]v = m on rD ,
where / and m = m,aa,a are the controls. When controls are in the geo-
metrical boundary conditions, the boundary conditions along rD are
(3.8) w= /, u = m on rD.
332 VIII. PLATE-BEAM SYSTEMS
and
e = {} + we3 E 1-l(0,f), {} = 'l?2el - 'l?le2.
We define the norms
and we set
H = 1-l(O) x 1-l(0,f)
with its natural product topology. For s > the norms on 1-lB(O) and
1-l B(O, f) are those induced by the corresponding HB spaces with their stan-
dard norms.
Let "( be a nonempty, relatively open subset of r \ J, and consider the
set
1-l~(O) = {'l' E 1-ll(O)1 'l' = on "(}.
For 'l' E 1-l~(O) we set
Note that the condition "iJI = 0"8(0) just expresses the geometric condition
(3.5). We define a norm on V"( by setting
1I("iJI,8)llv., = {II"iJlII~~(n)
}1/2
+ 10 (G2II~~12 + Ehl~~12 + G 2 Alw' + ~212) d1.
The Hilbert space V"( is dense in H with compact injection, so that if H
is identified with its dual space we have the compact embeddings V"( c
H c V~, where V~ denotes the dual of V"(. We denote by (v', v) v., the
scalar product of elements v' E V~ and v E V"(. Let A denote the Riesz
isomorphism of V"( onto V~ and set
DA = {("iJI, 8) E V"(I A("iJI, 8) E H}, 11("iJI, 8)IIDA = IIA("iJI,8)IIH.
Then A is an isomorphism of D A onto H, of V"( onto V~ and may be ex-
tended by continuity to an isomorphism of H onto DA through the formula
(Au,v)v., = (U,AV)H, 'rIu E V, vEDA).
REMARK 3.1. It may be verified through integrations by parts that
"iJI E C" (0; tal]) EB C" (0; [a2]) EB C" (0; [a3])
334 VIII. PLATE-BEAM SYSTEMS
We also set
and
(3.12) ((~, 8), (~, 9))H + (('l1, 8), (~, S))v = f [m u + jW] dr,
Jr D
(~,e)+A('l1,e)=Bf in V'
(3.14) X=AX+Bf,
O:::D,
where
k
= {-G1hdiv(VW + u)W - [divC[(u)]- G1h(VW + u)]. u}dO
- u (C[c(u)]v) + u (C[(u)]v)) dr
+ { [m (C[e(u)]v) + Glh/aaW]dr.
JrD v
336 VIII. PLATE-BEAM SYSTEMS
Also
Since (\].i,8) and (+,8) each satisfy the geometric and dynamic junction
conditions, it follows that the sum of the integrals over J with the boundary
terms at zero in the last two equations vanishes; that is
- u (C[c:(u)lv) + u (C[c:(u)lv)) df
+ G2A(w' + 192)(O)w(O) + G2H~(O)Jl(O) + Eh19~(O)J2(O)
- G2A(w' + J 2)(O)w(O) - G2IJ~(O)191(O) - ElhJ~(O)192(O).
Therefore (3.11) may be written (see Remark 3.1)
irD
[m (C[C:(u)lv) + G1h/-
aw
av 1df ::; Cllfliuli\].iliw(n)
A
(3.17) [[m,(c[C:(U)lV)+G1h/aawldf
JrD v
= (Bf, (., 8))H, (~, 8) = A- 1 (., 8),
3. FUNCTION SPACES AND WELL-POSEDNESS 337
(((t, 8), A-1(~, S))H + (("IJ1, 8), (~, 8))H = (Bf, (~, 8))H,
'v'(~, S) E H.
We need to show that if the initial data satisfy the stronger regularity
assumptions of Theorem 3.1, then the solution has the regularity stated
in that theorem. This result cannot be obtained from abstract semigroup
theory but rather is a consequence of the following regularity estimate. Let
us consider the homogeneous system
(3.21)
~ = cp + Za3, S = 8 + ze3,
and ~o, ~l, So, Sl with obvious connotations.
LEMMA 3.1. Let (~,S) be the solution of (3.19)-(3.25) with initial data
and set ~l := r 1 x (0, T), where r 1 is any relatively open subset of r such
that r\ n] = 0. Then
is given by
One has
A' = (~ -oA), D(A') = V x DA.
(3.29) .,
Y=-AY, O~t~r, Y(r)=Y.
Hwe write
In addition,
(Bf,A(., 8H = 1 rD a 1dr,
[m (C[c(<p)]V) + Kf-
az
v
hence
(3.31)
340 VIII. PLATE-BEAM SYSTEMS
We insert this expression into (3.28) and utilize Lemma 3.1 with r l = rD
to obtain the estimate
REMARK 3.3. That X E C([O, T]j H x V') may also be obtained directly
from a "lifting theorem" of Lasiecka and Triggiani [55] once Lemma 3.1 is
established. 0
DA x V. We first use a trick from [85]. Let X E Coo(O) such that X = 1 is
a neighborhood of r 1 and X = in a neighborhood of J, and set ci = X().
(Such a X exists since 1\ n] = 0.) For any cp = CPlal + CP2a2 with
CPo E H2(n) we have
div C[c(Xcp)]
and
_ h 3 (X,l1 + JLIX,22
12 (.xl + JLI)X,12
the boundary conditions
(3.35) i = 0, ip = 0 on E;
and the initial conditions
i(O) = X~o, ~(O) = X~l.
For a solution of (3.34), (3.35) with initial data (i(O), i(O E 1i},(n) x
1iO(n) := 1i we have the standard energy estimate
h
loT In (V~h) divC[c(~)ldOdt = -~ (C[c(~)l,c(~))dE
+ h(V~h) (C[c(~)lll)
. dE -loT In Q(V~) dOdt,
where the functional Q is defined by
When the last four formulas are inserted into (3.37) the result is
Therefore, on r,
(3.41)
and
(3.42)
3
IC[e(rp)]vI2 = ( h12 )2 [ J.t~ ( T' a-)2
a~ + (2J.tl + At}2 ( V a-
a~ )2] .
344 VIII. PLATE-BEAM SYSTEMS
Hence, on r,
for some positive constant K. It follows from (3.38) and (3.43) that
:<; C [ [ (IIF (t) 1Il,.(o) + 11(4'>, ,f.) II;') dt + II (4'> , ,f.) lIi~ (O,T,,,) 1
:<; C [loT IIF(t) II;'. (0) dt + (T + 1)II( 4'>,,f.m~(O.T''')]' D
4. THE REACHABLE SET 345
(4.2) (x - xo) . v ~ 0 on rD
is needed.
THEOREM 4.1. (Geometrical control.) Assume that (4.1) holds and let
'Y = r - J. Then there is a time To > 0 such that RT = H x v~ for T > To.
THEOREM 4.2. (Mechanical control.) Assume that (4.1) and (4.2) hold,
that r C '" 0, that F and I'D either do not intersect or else intersect in a
strictly convex corner, and set 'Y = rC. Then there is a time To > 0 such
that V'"( x H C RT for T > To.
REMARK 4.1. One may eliminate the geometric condition (4.2) in The-
orem 4.2 but at the expense of enlarging the control space and working in
a weaker state space. For example, if one admits controls in the space
(4.3)
then one may prove that H x V~ C RT for T large enough, without as-
sumption (4.2). In (4.3),
(4.6)
+ h, ((C[o(<P)[V).: +G'hl:J').n:]
REMARK 4.2. In view of (3.43), (C[(rp)]v) . (8rp/8v) ~ 0 on ro and
this quantity be replaced by IC[(IP)]vI 2 in the right side of (4.12). Also,
since C[(IP)]v = 0 on rl, it follows from (3.39) that
Then there is a constant To such that ifT > To, the solution of (4.4)-(4.11)
satisfies
r r
and that 0 and 1 either do not intersect or else intersect in a strictly
convex corner. Then
(4.14)
Since () also satisfies the boundary conditions (4.7) and (4.8), the integrals
on the right side of (4.15) may be written as
where
(4.16)
(4.19)
(4.21)
and therefore
where C depends on the Hl(J) norms of 0"0,0"1. The estimate (4.17) follows
from (4.23) and (4.24) by choosing = Cl/2, for example.
By combining (4.16) with (4.17) we obtain the estimate
(4.25)
It follows from [49, Proposition 3.1] that the following identity holds for 8:
for some Co
> and the following estimate holds for the solution E:
~(o) (T
(4.27) - -2- 10 EB(O, t) dt
~(O) {T 2 2 2
=--2-10 (m2Ii(O,t)1 +P2hI02(O,t)1 +p2 I IOl(O,t)1 )dt
of the form
h 1:1') <II:]
I
+ t (m v) ((Cle('P)Jv) : +G,h
Since
we have
+ t (m v) ((CI'(<p)]v) . :~ + G,h
for some positive constants Co, co.
Since the solution of (4.4)-(4.11) is
given by a unitary group on Vro x H, we have
+ l-
1
(C[c(cp)]r). ~~ +G1 hlr (VZ+cp)1 2 )dE
+h t ((C[e(<p)]v). ~+Glhl:1') dE+ [[5(0,t)['dt].
where
Co
CT = coT_2CmFlm,vl.
Except for the last term on the right side, (4.31) is the estimate of Propo-
sition 4.1. Therefore, to complete the proof, it suffices to establish the
following result.
+ l-
1
(C[c(cp)]r) . ~~ + G1 hlr (VZ + CP)1 2 ) dE
+ ~_ (C[c(It')]-r). ~~ +G1hl-r.(\7Z+It')12)dI.:
h
1
(4.34)
Estimate (4.34) may be proved by contradiction, using (4.31) and the com-
pactness of the injection (see Simon [102, Section 8])
(+,8) E Loo(O, T; V")') n W1,00(0, T; H) 1-+ Loo(O, T; [V"),, H]o:), 0< 0: ::; 1,
where [-, .]0: denotes the interpolation space of order 0:. In fact, if (4.34)
is false, there is a sequence of initial data +~, 8~), (+~, 8~)) E D A X V
(where V:= Vro ) such that the corresponding solution (+n,8 n) of (4.4)-
(4.11) satisfies
(4.35) loT 18 n 2
(0, t)1 dt = 1,
(4.36) lI+nIILOO(O,T;1tO(O -+ 0,
It follows from (4.31), (4.35) and (4.37) that +~, 8~), (+~, 8~)) is bounded
in V x H and, therefore
4. THE REACHABLE SET 359
and
From (4.38),
(4)>,E) E Loo(0,T;V)nW1,00(0,T;H),
(4)>, E) satisfy (4.4)-(4.10),
cp = 0, Z = on Et,
arp az
T aT = 0, aT + T . rp = 0 on El
with norm
weakly in Hand
(4.42)
(4.44) {
'P = 0, Z = 0, C[c('P)]v = 0, Bv =
BZ
on r o+,
cp = 0, Z = 0, C[c:(<p)]v = 0, -Bv + v
BZ
. 'P = on rt
Since (4.43) is a second order elliptic system with constant coefficients
whose solution has vanishing Cauchy data on rt, if rt =f. 0 we may con-
clude that 'P =
and Z =
in n x (0, T). The geometric and dynamic
junction conditions then give 8(0,) = and 8'(0,) = 0, hence 8 = in
(0, ) x (0, T) since 8 = (J + ze3 satisfies the system of ordinary differential
equations
m2'T12Z - G2A(z' + ( 2 )' = 0,
{ P2h'T12(h - EhO~ + G A(z' + ( = 0,
2 2)
cp = 0, Z = 0, C[c:(cp)]v = 0, ~: = on rt.
4. THE REACHABLE SET 361
G2A(ZI + (2)' = 0,
{ EhO~ - G A(ZI + ( ) = 0,
2 2
= 0, 0 < ~ < j
G2 IOr
G2A(ZI + ( 2)() = EhO~() = G2IO~ () = OJ
Z = 0, <p = 0 on roj
C[(<P)]v = 0, G1h(VZ + <p). v =0 on r 1j
Z = O"o[z(O) + (01(0)], <p = 0"18(0) on Jj
-0"1e2 . (C[(<p)]V)}dr,
It follows from (4.31) that ((CI~, E~), (~~, E~)) is bounded in V x Hand,
therefore
(CIn' En) is bounded in LOO(O, Tj V) n W 1,00(0, Tj H).
362 VIII. PLATE-BEAM SYSTEMS
If (~, E) denotes the limit function, then (~, E) satisfies, in view of (4.45)-
(4.47),
C[ ] v = 0, BZ
c( <p) Bv = +
on Eo ,
cp = 0, .i = on Et,
T BT =
B<p
0,
BZ
BT +T . <p = _
on El ,
that is, (~, E) EX. Therefore (~, E) = 0, contradicting (4.48). 0
4.2. Proofs of Theorems 4.1 and 4.2. We begin with the proof of
Theorem 4.1. Consider the control-to-state map
We have already proved that ST E (L2(0, T; U), H x V~) for each T >0
(see (3.33)). Since
the dual map S~ : H x V')' I-t L2(0, T; U) is given by S~Yo = B'Y(), where
Y = ~l,El)' (~,E)) satisfies (3.30) with T = T. Therefore Range(ST) =
H x V~ is equivalent to
The proof of Theorem 4.2 is similar. In this case one again has (4.49),
where now B' is given by (see (3.13))
STS~ is the Riesz isomorphism of:F onto :F', the dual space of :F. Therefore
RT = :F' for T > O. On the other hand, according to Corollary 4.2 one has
:F c Vi x H for T > To and, therefore, V-y x H c :F' for T > To. If XO E RT
and Y = (STS~)-lXO, then f = S~Yo is the control of minimum norm
in L2(0, Tj U) for which STf = Xo.
(5.3)
(5.4) W = 0, u = 0 on foj
G1h(V'W + u) v = 0,
(5.5) {
C[(u)]v = 0 on f1j
G2It?~(0)= [ {G l h6v(V'W+U)
JJ a
-e2' (C[(U)]V)}dr,
(5.7)
Eht?~(O) = [ (C[(U)]V) dr,
JJ
el'
a
\Ii = U + Waa,
9 = f} + wea,
with corresponding initial functions \li o, \iiI, eO, e l . We further denote
the pair (\Ii, e) by R and the pairs (\lio,eo), (\lil,e l ) by Ro and Rl,
respectively. As discussed earlier, the appropriate function space setting
for the above problem is V x H, where V .- Vr o , and its variational
formulation is
(5.11)
where G = (Gl, G 2 ) with
Gl = I~lMl + mil Fla3, G2 = P2Ii:l M22el - 1-1 M2le2 + m2"l F2e3'
Our goal is to pass to the limit as G - 00 in (5.10) and (5.11). To this
end, we introduce a closed subspace Hoo of H as follows:
Hoo = {(w, 9) E HI WE Hl(O), wE Hl(O,I), 19 1 E Hl(O,I),
u = -VW, 192 = -w', 19~ = O}.
If R = (~, 9) E Hoo, then
~ = -VW + Wa3, 9 = -w'el - 19le2 + we3, 19 1 == const,
and
I fl 19 l d{ = Ipl
Jo
1J
a
6(v, u) dr, P2h19 2(0) = -Ipl f V udr.
JJ a
We further introduce
Voo = VG n Hoo = {(w, 8) E VGI WE H2(O), wE H2(0,1),
u = -VW, 192 = -w', 19~ = o}.
366 VIII. PLATE-BEAM SYSTEMS
(5.13) W= aw =0 onro
av
and the geometric junction conditions
W = w(O) + 6~1' - \7W = -w' (O)v(xo) - ~1 T(Xo) on Ja ,
where ~1 :=const.j equivalently
(5.14) W = w () cVol, aw
0 +.,,2 av = w'o
(0) n 1a'
If R E Voo , then
(5.15)
l }1/2
IIRllvoo = IIRllvG = { in (C[c(\7W)],c(\7W)) dO. + Eh llwll12~
The space (Voo, 1Illvoo ) is dense in (Hoo, 11IIH oo ) with continuous injection
so that we may write, as usual,
Voo C Hoo c V~.
Let AG denote the Riesz isomorphism of VG onto Va and Aoo the Riesz
isomorphism of V00 onto V~. If REV00 then AooR is the restriction of
AGR to Voo , that is
(AooR,R)voo = (AGR,R)vG' VR,R E Voo'
In order to pass to the limit in (5.10), (5.11), we assume that the data
of the problem satisfy the following conditions:
(5.16)
As in Section 6, Chapter VII, the restrictions on the initial data are con-
ditions of consistency of the data with the limit model.
The solution RG of (5.10), (5.11) may be decomposed into
RG = R~ + (R~Y-, liP = R~ + (R~)l.,
G Roo
where Roo, 'G E C(Hoo) and (Roo)
Gl. ,(Roo)
Gl. l. Similarly, we
E C(Hoo)'
write
G = Goo + G;;" Goo E L2(Hoo), G;;, E L2(H~,).
Let (R~)(t) E V~ denote the restriction of RG(t) to Voo' Then one has
R G
00
= .!!..RG .. .. G d G
dt 00 and, SImIlarly, Roo = dt Roo
RG -t R strongly in Loo(H)j
RG -t R weakly* in Loo(1'lfo(O) x 1-e(O,f))j
it~ -t it strongly in L 00 (V~) j
itG - t it weakly* in Loo(H)j
R~ -t R weakly in LZ(V~).
Therefore
In particular, the map t f-+ R(t) : [0, T) f-+ H is continuous and R(O) = Ro.
Let us write
(5.28)
wG is bounded in L oo (1ifo(O)),
/VW G + u G / _ strongly in Loo(L2(0
and therefore
hence
Let us calculate each of the three terms in (5.31). One has, at least formally,
The first term on the right was evaluated in Chapter VI (see (6.12)) with
the result
Thus
oW
(5.36) W= - =0 onro;
OV
Ipl OW
OV - Dl [OLlW 0
--a,;- + OT CT (VW)
]
= -v . M l ,
(5.37) {
CII(VW) =0 on r l ;
(5.39)
(5.40) W = W () ., oW
0, + 6'!?l () ov = W
I()
0, on Ja .
5. DERIVATION OF THE LIMIT MODEL 371
1JaIpl W oW
{ ( [oaw 0
OV dI' - Dl 1Ja W a;;- + OT CT(VW) dI'
A A ]
( oW .
+ Dl 1Ja OV C,,(VW) dI' - P2hw'(0)w(0) + !l1?11?
A
+ Eh[-w"(O)w'(O) + W"'(O)W(O)]
= - { (V, Ml)W dI' + t?l (i M2l df. + M22 (0)W(0), vit E Voo.
1Ja 10
The test functions it satisfy
where W(O),W'(O) and t?l are arbitrary real numbers. We may therefore
deduce from the last relation the following dynamic junction conditions.
1.
~ {I p, : -Dl [8~: + !CT(VWJJ}dr;
(5.42) Ehw"(O,) = Dl ( C,,(VW) dI';
1Ja
Since 'l9y and 'l9t are constants, it follows that '19 1 is independent of ~ and
amounts to a (time dependent) rigid rotation of the beam around its axis
(caused by the deflection of the plate). The relation (5.41) may be recog-
nized as a balance law for linear momentum in the a3 direction along Ja ,
(5.42) as a balance law for angular momentum about 'T(xo) and (5.43) as a
balance law for angular momentum about the axis of the beam. Of course,
'19 1 may be eliminated from the system by integrating (5.43) and substitut-
ing for '19 1 in (5.40), but this would make the physical interpretation of the
junction condition (5.40) less transparent.
(6.3) h
W2(XI. X2, t) + iU2(XI, X2, t) = W3(0, t) + xl'i?l (0, t),
W3(XI. X2, t) = WI(O, t) - x l 'i?3(0, t) + x 2'i?2(0, t).
We note that in (6.3) the quantities W3 , WI. 'i?2, 'i?3 are not coupled to
W a , Ua, W2, W3, 'i?l. However, all quantities will turn out to be coupled once
the dynamic conditions are taken into account. We shall take the linearized
expressions (6.3) as the geometric junction conditions of our problem.
6.2. Dynamic Conditions. As usual, the dynamic conditions will be
obtained from Hamilton's Principle:
(6.4) 61 T
[K(t) - U(t) + W(t)] dt = 0,
where K and U represent the kinetic and strain energies, respectively, of the
plate-beam system, W is the work done on the system by external forces
and 6 denotes the first variation with respect to the class of admissible
374 VIII. PLATE-BEAM SYSTEMS
where IC p and ICB are the kinetic energies of the plate and beam, respec-
tively, and similarly for Up,UB and Wp, WB. We assume that the beam is
homogeneous and elastically isotropic, that the plate is homogeneous and
that the energy of the plate is that associated with linear Reissner-Mindlin
plate theory. Thus we set
(6.7) Up(t) =
f
21 }n[(C[c(v)),c(v)) +
h2
12 (C[c(u)],c(u))
+ G 1 hlVW3 + u1 2 ] dO,
where we have set v = Wo:aa and where ml, Ipl = m 1 h 2 /12 and G 1 are
as in Section 2,
C[c] = h[>'lCo:o:I + 2JLIC),
1
c(u) = 2[Vu + (Vu)*),
and (C[c),c) = CO:fj[c]cO:fj. We further have
where P2 is the mass of the beam per unit of reference volume, m2 = P2A
is mass of the beam per unit of reference length, A denoting the area of w,
and
Also
(6.10) UB(t) = 1
~ e{EAlw~12 + EI21'1?~12 + EI31'1?~12 + G2II'I?~12
+ G2Alw~ - 'l?312 + G2Alw~ + 'l?212} ~b
where E and G 2 have the same meaning as in Section 2, and
(6.11) WB(t) 1
= e(FB . W - M23'1?1 + M13'1?2 - M12'1?3)(~, t) ~
+ fB . w(l, t) - m23'1?1(l, t) + m13'1?2(l, t) - m12'1?3(l, t),
where F B , fB, M23 , M13 , m23, m13 are various resultant (over w) forces and
moments as defined in Section 2 and
EAwi = fB eb
G2A(w~ - t?3) = fB . e2,
G2A(w~ + t?2) = fB . e3,
(6.15)
G2It?i = -m23,
EI3t?~ = m13,
EI2t?~ = -mI2.
i
w 12
+ {G 1hdiv(V'W3 + U)W3 - G 1h(V' W3 + u) u}dw
for all test functions Vi, U, w, Ji which satisfy (6.3). Because of the de-
coupling of W3, WI, J2 , J3 from the remaining variables in (6.3), the last
equation is equivalent to the two variational equations
and
These three equations may be recognized as a balance law for linear mo-
mentum in the a3 direction, a balance law for angular momentum around
the a2 axis and a balance law for angular momentum around the al axis,
respectively.
In the variational equation (6.17), we may set
o= G2A(w~ -193)(0)w2(0)
+ G2A(w; + 192)(0)w3(0) + G2I19~ (0)t9 1 (0)
+ !!.1
2
{m(v - div C[c( v)]} . U cU.v
-L
w
= ~ l {m1 v - divC[c{v)]}dw
The first two relations in (6.19) are balance laws of linear momentum in
the al and a2 directions, respectively, and the third is a balance law for
angular momentum around a3. The right side of (6.20) is a torque applied
to the reference surface of the plate (a force in the a1a2 plane at 6 = 0
acting through the distance h/2) caused by the rotation of the cross-section
of the beam at 6 = O. Equation (6.20) matches this torque with the one
applied at the base of the beam by the rotation of the plate filaments
{X3a311x31 :::; h/2, (XI,X2) E w}.
The complete model of the plate-beam configuration under consideration
therefore consists of (6.3), (6.5), (6.12)-{6.15) and (6.18)-{6.20).
REMARK 6.1. Although we shall not provide details, since the above
model is derived from Hamilton's principle it is fairly obvious that it has a
natural variational formulation (the Principle of Virtual Work) which, un-
der reasonable assumptions on the external loads, comprises a well-posed
problem in an appropriate "energy" space (a product of HI spaces such
that the geometric junction and boundary conditions are respected). We
have not investigated the reachability problem or the stabilizability prob-
lem for this system but it is to be expected that the system may be exactly
controlled, or uniformly stabilized, through the action of appropriate con-
trols acting on rD alone, provided the geometry of the triple (O, r C, rD)
is suitably restricted.
REMARK 6.2. The above model has eleven degrees of freedom. As in
section 5, a model with fewer degrees of freedom may be obtained by passing
to the limit as G 1 , G 2 go to infinity. The limit model will have seven degrees
of freedom: Wi, Wi and '!9 1, where '!9 1 is spatially constant.
Bibliography
[1) S. S. Antman, The theory of rods, Handbuch der Physik Vol. 6/2a (Berlin)
(Fliigge, ed.), Springer-Verlag, Berlin, 1972, pp. 643 -703.
[2) ___ , The equations for large vibrations of strings, Math Monthly 81 (1980),
359-370.
[3) S. S. Antman and Arje Nachman, Large buckled states of rotating rods, Nonlinear
Analysis: Theory, Methods and Appl. 4 (1980), 303-327.
[4) S. S. Antman and W. H. Warner, Dynamical theory of hyperelastic rods, Arch.
Rational Mech. Anal. 40 (1966), 135-162.
[5) J. Bailleul and M. Levi, Rotational elastic dynamics, Pfysica D 21 (1987), 43-62.
[6) ___ , Constrained relative motions in rotational mechanics, Arch. Rational
Mech. Anal. 115 (1991), 101-135.
[7) A. V. Balakrishnan, On the Wittrick-Williams algorithm for modes of flexible
strucutres, to appear.
[8) V. Brasey and E. Hairer, Half-explicit Runge-Kutta methods for differential-
algebraic systems of index 2, SIAM J. Numer. Anal. 30 (1993), no. 2, 538-552.
[9) H. Bremer, Dynamik und Regelung mechanischer Systeme, Teubner Studi-
enbiicher Mechanik, vol. 67, B.G. Teubner, Stuttgart, 1988.
[10) J. A. C. Bresse, Cours de mechanique applique, Mallet Bachellier, 1859.
(11) H. Brezis, Operateurs maximaux monotones, North-Holland/American Elsevier,
Amsterdam, London, New York, 1973.
[12) H. Cabannes, Mouvement d'une corde vibrante soumise d un fottement solide,
C. R. Acad. Sc. Paris 281 (1978), 671-673.
[13) ___ , Propagation des discontinuites dans les cordes vibrante soumises d un
frottement solide, C. R. Acad. Sc. Paris 289 (1979), 127-130.
[14] G. F. Carrier, On the non-linear vibration problem of the elastic string, Quart.
Appl. Math. 2 (1945), 157-165.
[15] G. Chen, M. Coleman, and H.H. West, Pointwise stabilization in the middle of
the span for second order systems, nonuniform and uniform exponential decay
of solutions, SIAM J. Control Optim. 41 (1987), no. 4, 751-779.
[16] G. Chen and D. L. Russell, A mathematical model for linear elastic systems with
structural damping, Quartely Appl. Math 39 (1982), 433-455.
[17] P. G. Ciarlet, H. Le Dret, and R. Nzengwa, Junctions between three-dimensional
and two-dimensional linearly elastic structures, J. Math. Pures Appl. 68 (1989),
261-295.
380 BIBLIOGRAPHY
[66] G. Lumer, Espaces mmijies et diffusions sur des resaux topologiques, Comptes
Rendus Acad. Science Paris 201 (1980), 627-630.
[67] F. Ali Mehmeti, Regular solutions of tmnsmission and intemction problems for
wave equations, Math. Meth. in the Appl. Sciences (1989), 656-685.
[68] ___ , Existence and regularity of solutions of cauchy problems for inhomoge-
nom wave equations with intemction, Oper. Theory Adv. and Appl. 50 (1991),
23-34.
[69] ___ , Nonlinear intemction problems, Nonlinear Analysis: Theory, Methods
and Appl. 20 (1993), 281-295.
[70] F. Ali Mehmeti and S. Nicaise, Nonlinear intemction problems, Nonlinear Anal-
ysis: Theory, Methods and Appl., to appear.
[71] ___ , Some realizations of intemction problems, Semigroup theory and evo-
lution equations (New York) (E. Mitidieri Ph. Clement and B. de Pagter, eds.),
Lect. Notes Pure and Appl. Math., vol. 135, Marcel Dekker, New York, 1991,
pp.15-27.
[72] Wei Ming, Controllability of networks of strings and masses, Ph.D. thesis, McGill
University, December 1993.
[73] S. Nicaise, Geneml interface problems I, Math. Methods Appl. Sci., to appear.
[74] ___ , Geneml interface problems II, Math. Methods Appl. Sci., to appear.
[75] ___ , Some results on spectral theory over networks applied to nerve impulse
tmnsmission, Lecture Notes in Mathematics (Brezinski, ed.), Springer-Verlag,
1985, pp. 533-541.
[76] ___ , Diffusion sur les espaces mmijies, Ph.D. thesis, These du Doctorat,
Mons, 1986.
[77] _ _ , Spectre des esaux topologiquejines, Bull. Sci. Math. III (1987),401-413.
[78] ___ , About the Lame system in a polygonal or a polyhedml domain and a
coupled problem between the Lame system and the plate equation. I: regularity of
the solutions, Ann. Scuola Normale Sup. Pisa XIX (1992), 327-361.
[79) ___ , Exact controllability of a pluridimensional coupled problem, Revista
Matematica 5 (1992),91-135.
[80) ___ , About the Lame system in a polygonal or a polyhedml domain and a
coupled problem between the Lame system and the plate equation. II: exact con-
trollability, Ann. Scuola Normale Sup. Pisa XX (1993), 163-191.
[81] ___ , Polygonal interface problems, Peter Lang, Frankfurt am Main, 1993.
[82] O. A. Oleinik, A. S. Shamaev, and G. A. Yosifian, Mathematical problems in
elasticity and homogenization, North Holland, Amsterdam, 1992.
[83] A. Pazy, Semigroups of linear opemtors and applications to partial differential
equations, Springer Verlag, New York, Berlin, Heidelberg, Tokyo, 1983.
[84] J. P. Puel and E. Zuazua, Controllability of a multi-dimensional system of
Schrodinger equations: application to a system of plate and beam equations,
To appear, 1992.
[85] ___ , Exact controllability for a model of multidimensional flexible structure,
Proc. Royal Soc. Edinburgh (1993), 323-344.
[86) Bopeng Rao, Stabilisation du modele SCOLE par un contr6le a priori borne,
C.R. Acad. Sci. Paris 316 (1993), 1061-1066.
[87) J. Rauch and M. Taylor, Exponential decay of solutions to hyperbolic equations
in bounded domains, Indiana Univ. Math. Jour. 24 (1974), 79-86.
[88} E. Reissner, On bending of elastic plates, Quart. Appl. Math. V (1947), 55- 68.
[89} ___ , On one dimensional large-displacement jinite-stmin beam theory, Stud-
ies in Applied Mathematics 52 (1973), 87-95.
[90) R. C. Rogers, Derivation of linear beam equations using nonlinear continuum
mechanics, VPI preprint 1989, 1989.
BIBLIOGRAPHY 383
Series Editor
Christopher I. Byrnes
School of Engineering and Applied Science
Washington University
Campus P.O. 1040
One Brookings Drive
St. Louis, MO 63130-4899
U.S.A.
Systems & Control: Foundations & Applications publishes research monographs and
advanced graduate texts dealing with areas of current research in all areas of systems and
control theory and its applications to a wide variety of scientific disciplines.
Proposals should be sent directly to the editor or to: Birkhiiuser Boston, 675 Massachusetts
Avenue, Cambridge, MA 02139, U.S.A.
Set-Valued Analysis
Jean-Pierre Aubin and Helene Frankowska
Disease Dynamics
Alexander Asachenkov, Guri Marchuk, Ronald Mohler, Serge Zuev