J. E. Lagnese, Günter Leugering, E. J. P. G. Schmidt Auth. Modeling, Analysis and Control of Dynamic Elastic Multi-Link Structures

Systems & Control: Foundations & Applications
Series Editor
Christopher I. Byrnes, Washington University
Associate Editors
S. - I. Amari, University of Tokyo

B.D.O. Anderson, Australian National University, Canberra
Karl Johan Astrom, Lund Institute of Technology, Sweden
Jean-Pierre Aubin, EDOMADE, Paris
H.T. Banks, North Carolina State University, Raleigh
John S. Baras, University of Maryland, College Park
A. Bensoussan, INRIA, Paris
John Bums, Virginia Polytechnic Institute, Blacksburg
Han-Fu Chen, Academia Sinica, Beijing
M.H.A. Davis, Imperial College of Science and Technology, London
Wendell Fleming, Brown University, Providence, Rhode Island
Michel Fliess, CNRS-ESE, Gif-sur-Yvette, France
Keith Glover, University of Cambridge, England
Diederich Hinrichsen, University of Bremen, Germany
Alberto Isidori, University of Rome
B. Jakubczyk, Polish Academy of Sciences, Warsaw
Hidenori Kimura, University of Osaka
Arthur J. Krener, University of California, Davis
H. Kunita, Kyushu University, Japan
Alexander Kurzhanski, Russian Academy of Sciences, Moscow
Harold J. Kushner, Brown University, Providence, Rhode Island
Anders Lindquist, Royal Institute of Technology, Stockholm
Andrzej Manitius, George Mason University, Fairfax, Virginia
Clyde F. Martin, Texas Tech University, Lubbock, Texas
Sanjoy K. Mitter, Massachusetts Institute of Technology, Cambridge
Giorgio Picci, University of Padova, Italy
Boris Pshenichnyj, Glushkov Institute of Cybernetics, Kiev
HJ. Sussman, Rutgers University, New Brunswick, New Jersey
TJ. Tam, Washington University, St. Louis, Missouri
V.M. Tikhomirov, Institute for Problems in Mechanics, Moscow
Pravin P. Varaiya, University of California, Berkeley
Jan C. Willems, University of Gronigen, The Netherlands
W.M. Wonham, University of Toronto
J. E. Lagnese
Giinter Leugering
E. J. P. G. Schmidt
Modeling, Analysis and Control of

Dynamic Elastic Multi-Link Structures
Springer Science+Business Media, LLC

J. E. Lagnese Giinter Leugering
Department of Mathematics FakuItt fiir Mathematik und Physik
Georgetown University University of Bayreuth
Washington, ne 20057, USA D-W 8580 Bayreuth, Germany
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 ISBN 978-1-4612-0273-8 (eBook)

DOI 10.1007/978-1-4612-0273-8
1. Aexible struetures--Mathematieal models. 2. Distributed
parameter systems--Mathematical models. 3. Elastic analysis
(Engineering) 1. Leugering, Giinter, 1953- . II. Schmidt, E. J.
P. G., 1940- . III. Title. IV. Series.
TA660.F53L34 1994 94-3096
624.1'7 --dc20 CIP
Printed on acid-free paper
Springer Science+Business Media New York 1994
Originally published by Birkhuser Boston in 1994
Softcover reprint of the hardcover 1st edition 1994
Copyright is not claimed for works ofU.S. Govemment employees.
AII rights reserved. No part ofthis publication may be reproduced, stored in a retrieval system,
or transmilled, in any form or by any means, electronic, mechanical, photocopying, recording,
or otherwise, without prior permission of the copyright owner.
Permission to photocopy for internal or personal use of specific clients is granted by

Springer Science+Business Media, LLC for libraries and other users registered with the
Copyright Clearance Center (CCC), provided that the base fee of$6.00 per copy, plus
$0.20 per page is paid direct1y to CCC, 222 Rosewood Drive, Danvers, MA 01923, V.S.A.
Special requests should be addressed direct1y to Springer Science+Business Media, LLC
V.S.A.
ISBN 978-1-4612-6689-1
Typeset by the authors.
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
4. Straight, Untwisted, Nonshearable Nonlinear 3-d Beams 71

4.1. Approximation-Generalizations 74
5. Straight, Untwisted Shearable Linear 3-d Beams 76
6. Shearable Nonlinear 2-d Beams with Curvature 77

6.1. Approximation-Generalizations 79
7. A List of Beam Models 80
Damping 82
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
9. Rotating Two-link Flexible Nonlinear Shearable Beams 89
Chapter IV. A General Hyperbolic Model for Networks 95
1. The General Model 95
2. Some Special Cases 99

2.1. String Networks 99
2.2. Networks of Planar Timoshenko Beams 99
2.4. Networks of Initially Curved Bresse Beams 102
2.5. Beams and Strings 102
3. Existence and Regularity of Solutions 105
4. Energy Estimates for Hyperbolic Systems 111
5. Exact Controllability of the Network Model. 119
6. Stabilizability of the Network Model. 123
Chapter V. Spectral Analysis and Numerical Simulations 135
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. Eigenvalue Problems for Networks of I-d Elements 142

2.1. Introduction 142
2.2. General String Networks 143
2.3. Homogeneous String Networks 147
2.3.1. Examples
2.4. Networks of Timoshenko Beams 156
2.4.1. The Case Where A = 0
2.4.2. The Case Where A Belongs to an Individual Beam
2.4.3. Eigenvalues for the Entire Graph
2.5. Homogeneous Timoshenko Networks 165
3. Numerical Simulations of Controlled I-d Networks 167
3.1. Introductory Remarks 167
3.2. Networks of Strings 168
3.2.1. Absorbing Controls
3.2.2. Directing Controls
4. Finite Element Approximations of Timoshenko Networks 183
5. Implicit Runge-Kutta Method: Dry Friction at Joints 191
Chapter VI. Interconnected Membranes 201
1. Modeling of Dynamic Nonlinear Elastic Membranes 202
1.1. Equations of Motion 203
1.2. Edge Conditions 204
1.3. Hamilton's Principle 204
2. Systems of Interconnected Elastic Membranes 206
2.1. Geometric Junction Conditions 207
2.2. Dynamic Conditions 208
2.3. Linearization 212
2.4. Well-Posedness of the Linear Model 213
3. Controllability of Linked Isotropic Membranes 221
3.1. Observability Estimates for the Homogeneous Problem 223
3.2. A Priori Estimates for Serially Connected Membranes 238
3.3. A Priori Estimates for Single Jointed Membrane Systems 241
3.4. The Reachable States 243
3.4.1. Serially Connected Membranes
3.4.2. Membrane Transmission Problems
Chapter VII. Systems of Linked Plates 251
1. Modeling of Dynamic Nonlinear Elastic Plates 252
1.1. Equations of Motion 253
1.2. Edge Conditions 256
1.3. Hamilton's Principle 257
1.4. Additional Kinematic and Material Assumptions 259
1.5. Rotations Associated with Plate Deformation 264
viii CONTENTS
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
5. Limit Model as the Shear Moduli Approach Infinity 363

5.1. Interpretation of the Limit Model 369
6. Modeling of a Plate-Beam Junction: II 372
6.1. Geometric Conditions 372
6.2. Dynamic Conditions 373
Bibliography 379
Index 385
List of Figures
V.1 Some eigenmodes of the unit square 151

V.2 Eigenvalues of unit square 152
V.3 Splitting of eigenvalues 153
V.4 Eigenvalues of a triangle 154
V.5 Splitting of the spectrum 154
V.6 A typical frame 155
V.7 Distribution of eigenvalues 156
V.8 Uncontrolled motion of a string 171
V.9 Contour plot 171
V.1O Conservation of energy 172
V.U Absorbing boundary control 172
V.12 Energy plot of disappearing solution 173
V.13 Case (a) 174
V.14 Case (b) 174
V.15 Transfer of energy through multiple joints 175
V.16 Directing controls 178
V.17 Energy of entire system 178
V.18 Effect of time delay 179
V.19 Instability due to time delay 180
V.20 Bang-bang control 181
V.21 Improved dissipation 181
V.22 Robustness of bang-bang control 182
V.23 Nonlinear dissipation with time delay 182
V.24 Energy in individual strings 184
V.25 Energy in 5th string and entire system 184
V.28 A comparison of controls 191
V.29 Unit square: uncontrolled motion 192
xii LIST OF FIGURES
V.30 Unit square: conservation of energy 192

V.31 Unit square: controlled motion 193
V.32 Unit square: saturation of energy 193
V.33 Dry friction: locking 199
V.34 Dry friction: spillover 199
Preface
The purpose of this monograph is threefold. First, mathematical models

of the transient behavior of some or all of the state variables describing the
motion of multiple-link flexible structures will be developed. The structures
which we have in mind consist of finitely many interconnected flexible ele-
ments such as strings, beams, plates and shells or combinations thereof and
are representative of trusses, frames, robot arms, solar panels, antennae,
deformable mirrors, etc., currently in use. For example, a typical subsys-
tem found in almost all aircraft and space vehicles consists of beam, plate
and/or shell elements attached to each other in a rigid or flexible manner.
Due to limitations on their weights, the elements themselves must be highly
flexible, and due to limitations on their initial configuration (i.e., before de-
ployment), those aggregates often have to contain several links so that the
substructure may be unfolded or telescoped once it is deployed. The point
of view we wish to adopt is that in order to understand completely the
dynamic response of a complex elastic structure it is not sufficient to con-
sider only its global motion but also necessary to take into account the
flexibility of individual elements and the interaction and transmission of
elastic effects such as bending, torsion and axial deformations at junctions
where members are connected to each other. The second object of this
book is to provide rigorous mathematical analyses of the resulting models.
Its third purpose is to develop control-theoretic properties of multiple-link
(multi-link) flexible structures based on the control-theoretic properties of
the models.
To be more specific, distributed pammeter models of multi-link flexible
structures will be derived. While there is already a vast literature concern-
ing various aspects of modeling, control and stabilization of multi-body
systems, it deals almost exclusively with rigid rather than flexible elements,
with finite-dimensional modal (hence global) modeling rather than adopt-
ing a spatially dependent (local) approach through distributed parameter
xiv PREFACE
modeling. As a result, the transmission and conversion of bending, torsion,

and axial deformation at the joints is usually not taken into account. Re-
cent engineering research strongly indicates, however, that it is important
to account for such phenomena in order to understand completely transmis-
sion effects in trusses and frames and in other aggregates of this kind. The
starting points in our derivations of models of the structures considered
in this work are the principles of continuum mechanics. From these are
obtained, under reasonable constraints on the geometry of the admissible
deformations, a set of coupled partial differential equations governing the
evolution of the displacements and other internal variables of the structure.
The models will account for full three dimensional transient behavior, var-
ious possible joint interactions, couplings between flexible motions and the
dynamics of joints and/or actuators, superimposed rigid rotations, viscous
and thermal effects, initial curvature and other effects.
Once the models are developed, the next objective is to establish that
they comprise well-posed mathematical problems. It at this point that the
various possible choices of controls will find their mathematical formula-
tions. There are normally three different locations within a multi-link body
where controls can reasonably be implemented: (a) base points, where the
substructure (as a whole) is attached to a massive main body, (b) joints
(nodes) within the structure, and (c) endpoints, or tips of the structure.
The types of control instruments that can be employed depend very much
on where the controllers are to be located. For example, at base points one
can usually only control the strains, whereas at endpoints one can influ-
ence the stresses, shear forces and moments. The controllers employed at
internal joints usually perform a rigid rotation, or amount to dry friction.
The object of the control action is to change the system response in some
specified way. Our emphasis is on the controllability of the full infinite
dimensional system. To this end we shall employ methods from control
theory of distributed parameter systems to formulate and solve some spe-
cific control problems which are motivated by typical control objectives for
elastic structures, such as stabilizabilizing or exactly controlling the tran-
sient motion of the structure. The first control objective is related to he
problem of maintaining the stability of a structure under various distur-
bances. The second control objective is concerned with the configuration of
the structure and amounts to describing its complete configuration space.
In any consideration of controllability or stabilizability it is necessary to
distinguish between controls at base points, at interior joints and at tips,
as indicated above. In particular, the control of structures through appli-
cation of control mechanisms at joints may be viewed as a particular issue
within the general question of appropriate joint conditions as such, and
represents a major research topic in itself. It is clear that a distinction be-
tween various applications has to be made: in contrast to a robot arm (or a
satellite arm unfolding a solar panel), in a truss one does not consider any
PREFACE xv
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
to modeling interconnected flexible bodies. This is the approach which

will be followed in this work. One begins by modeling single elements (see
Wempner [108], Danielson and Hodges [22], Kane et. al. [41] and many oth-
ers) and then considers various possible connections between such elements
such as rigid, revolutive, prismatic, spherical or pinned (in I-d networks).
Such considerations lead to various continuity conditions, sometimes called
compatibility conditions or geometric junction conditions, between elements
sharing a joint and to dynamic transmission conditions, also called dynamic
junction conditions. The latter conditions result from energy principles of
mechanics. The physical interpretations of the dynamic conditions are as
balance laws for forces and moments at the joints. The energy principles
are to be used within the context of appropriate energy spaces, also known
as ramification spaces (see Lumer [66]). This approach has been used by
Schmidt [93], Schmidt and Leugering [58], Lagnese, Leugering and Schmidt
[49], [50], [40], Lagnese and Leugering [48] and Lagnese [46], Schmidt and
Ming [95], Puel and Zuazua [85] for networks of strings, beams, membranes,
plates and combinations thereof. In the context of diffusion problems,
reaction-diffusion problems, nerve-pulse propagation problems, and scalar
wave propagation on networks (out-of-plane displacement), this approach
has been used by Nicaise [75], [77], Ali-Mehmeti [67], [69], Ali-Mehmeti
and Nicaise [70], von Below [106], [105] and others.
As for numerical simulations of structures based on the types of mod-
els discussed above, we may refer to [19] for simulations utilizing models
obtained from 3-d elasticity via asymptotic analysis and to [39] for a dis-
cussion of simulations of models obtained via energy principles. In [39],
however, only the general outline is given and no particular network is
simulated. Here we shall utilize finite difference, finite element and im-
plicit Runge-Kutta techniques in simulations of both uncontrolled and con-
trolled structures. Finite differences are used for simple one-dimensional
models involving absorbing boundary conditions; specifically, we use the
Lax-Friedrichs scheme, see Le Roux [91]. For more complex structures, in
particular for networks of Timoshenko beams, a finite element implemen-
tation along the lines of [39] is employed. Finally, for problems involving
nonsmooth damping (such as dry friction), a fully implicit Runge-Kutta
method such as Radau IIa in its radau5-implementation of Hairer and
Wanner [33] is utilized; see also [8]. Other current research in this area
is to be found in the work of Lubich [64], [65]. The latter is in the spirit
of [99], where the geometric joint conditions are treated as kinematic con-
straints. This approach leads to the occurrence of generalized forces and
Lagrange multipliers in the state equations and has the advantage of lead-
ing to simply structured stiffness matrices even in the case of structures
containing closed circuits of elements.
To resolve the question of well-posedness of the mathematical models,
which is implicit in the Ciarlet-Dystunder approach, we note that we con-
1. GENERAL OVERVIEW 5
sider here mainly models obtained by linearization around an equilibrium of

the structure. Then well-posedness is easily settled by using hyperbolic esti-
mates (for I-d networks of strings and beams) as in [40], or energy estimates
obtained from standard energy multipliers, followed by generic semi-group
arguments. While in the I-d networks no singularities develop at the junc-
tions, the situation is much more complicated in 2-d and higher dimensional
cases. Indeed, the regularity issues in those cases are for the most part un-
settled at this point. There is substantial work in this direction by Nicaise
[78], [81] for scalar functions on 2-d networks. We note that Ali-Mehmeti
[67], [69], von Below [106], [105] and others have obtained well-posedness
of some quasilinear scalar multi-domain models. Again, well-posedness of
the quasilinear problems describing the three dimensional, or even planar,
motion of (even one-dimensional) networks appear to be open problems.
Once appropriate models are developed and analyzed, control problems
for the models can be formulated in a mathematically meaningful way. As
mentioned above, we will concentrate on basic control concepts such as
exact, spectral or approximate controllability, or on stabilizability. In the
case of exact (nUll) controllability, one looks for a finite time T and a set of
admissible controls, which are to be applied at junctions or "boundaries" of
the structure, such that for each set of initial data (with certain regularity)
there is a control which steers the corresponding solution of the system
to rest. This turns out to be equivalent to the problem of describing the
entire space of reachable configurations of the structure, if the structure
is initially in an equilibrium state. The notion of spectral controllability
involves the possibility of exactly controlling the span of any set of finitely
many of the eigenmodes of the underlying system operator. Approximate
controllability, finally, is the most natural weakening of the concept of ex-
act controllability. In this case the goal is to reach an arbitrarily small
neighborhood of a desired target without requiring that the target be hit
exactly. Questions of uniform or nonuniform stabilizability are known to be
intimately related to the controllability concepts just mentioned. Control
theory of distributed parameter systems has become a rich field of research
in the last three decades. Readers who are not familiar with the concepts
of control of distributed parameter systems are referred to Curtain and
Pritchard [21] for an elementary introduction, to Russell [92] for and excel-
lent review of methods until 1978, and to Lions [61] for a description of the
Hilbert Uniqueness Method and its application to exact controllability prob-
lems for distributed parameter systems. As for the vast literature treating
controllability /stabilizability problems for single elastic elements such as
strings, beams, membranes and plates, we refer only to the monographs
by Lagnese and Lions [51], Lagnese [52], Lions [60] and the bibliographies
include therein.
Controllability problems for nonlinear dynamic networks are not investi-
gated in this monograph. Let us note, however, that through an appropri-
6 1. INTRODUCTION
ate application of the implicit function theorem, generically there is local

controllability in the nonlinear (semi-linear) situation as soon as the corre-
sponding global results for the linear approximation have been established.
We do, on the other hand, present some stabilizability results for linear
dynamic models of I-d networks to which are applied nonlinear feedback
controls at some of the nodes of the network.
Although optimal control problems, other than minimum norm control
reachability problems, are not considered in this monograph, it is our hope
that the availability of distributed parameter models of multi-link flexi-
ble structures will stimulate future research in this direction. There are
important (sometimes combinatorial) optimization problems which we do
consider numerically but for which a rigorous mathematical analyses is
lacking. One such example involves the optimal location of a given finite
number of (directing or absorbing) controllers at junctions and boundaries
of a large truss or frame. Another optimization problems of considerable
importance related to structures is that of optimal design. In addition,
there are obviously many problems of parameter identification which can
be addressed within the framework presented below but which we shall not
consider.
2. On the Contents of the Book

In this section the organization of the book will be described. Chapter
II is devoted to dynamic networks of strings and is intended to serve as a
paradigm for the philosophy and approach used throughout the entire book.
First, the nonlinear equations of motion for a single element are derived
from first principles of continuum mechanics. Next, our basic approach to
obtaining the network models by the use of energy principles is developed.
The resulting nonlinear model is then linearized about an equilibrium and
well-posedness of the linear model is investigated by means of energy esti-
mates. Some of the energy estimates developed in Chapters II and IV are
particular to I-d networks and are based on the possibility of recasting the
dynamic equations as a system of symmetric hyperbolic equations of first
order. The energy estimates are then used to establish a result on exact
controllability of tree-like networks of elastic strings. In particular, given a
tree structure as the underlying reference configuration and given that the
root node is clamped while the nodes at the tips of all the branches are
controlled, then generically one obtains exact controllability. The optimal
control time is related the longest path in the graph. It is shown that even
approximate controllability fails to hold for networks containing circuits. It
is also shown that exact controllability is sensitive to boundary conditions
at uncontrolled boundaries.
Modeling of thin, thermoelastic nonlinear 3-d beams and networks of
such elements is considered in Chapter III. Even though there are many
papers dealing with one or another class of beams models, we were unable
2. ON THE CONTENTS OF THE BOOK 7
to find in the literature a treatment in which all of the features of interest

are incorporated. Indeed, most papers do not go beyond the stress-strain
constitutive relations and, therefore, do not contain a practical set of partial
differential equations in terms of the displacements and internal variables.
We therefore took some pains to provide a concise development of beam
theory. The approach used has been outlined by Wempner [108] and also by
Danielson and Hodges [22] and Kane et. al. [41]. We provide a list of beam
models which includes all of the well known models but also some models
which appear not to have been described in the literature before, such as
a 3-d Timoshenko beam, a 2-d thermoelastic nonlinear shearable beam,
etc. The point is that all these models can be derived immediately and
without difficulty from our general set of equations. Once the single element
models are obtained we proceed to formulate the network models following
the general procedure outlined in Chapter II. As we have to account for
shearing and twisting at the joints, the continuity conditions at the joints
generally combine local displacements and shear/twist angles. In the plan~,
however, there is no twist and since in this case all shear angles have their
axes normal to the plane of motion, the displacements decouple from the
shear angles in the continuity conditions. As in Chapter II, balance laws
of linear and angular momenta at the junctions are obtained from energy
principles.
In Chapter IV we introduce a general model for networks of one-dimen-
sional elements whose dynamics are described by hyperbolic systems. This
model includes not only the string network of Chapter II but a variety of
planar and three dimensional beam models of the Bresse-Timoshenko type
which are derived in Chapter III. It also includes "mixed networks" which
we illustrate in a rudimentary way. The existence theory for the rather
complicated system of equations and joint conditions is obtained by the
same standard variational and semigroup methods which were developed
in detail in Chapter II. The state space and the spaces of controls are so
chosen that one can simultaneously use Dirichlet and Neumann controls.
The a priori energy estimates which are essential to results on controlla-
bility and stabilizability are then proved in a rather general form. The
results on controllability which were given in Chapter II generalize readily
as do the results on stabilizability. In fact the treatment of stabilizability is
considerably extended from what was done for string networks in Chapter
II, incorporating recent treatments of maximal monotone velocity feedback
systems.
Chapter V contains a discussion of the eigenvalue problems for networks
of strings and of Timoshenko beams and also contains numerical simula-
tions of a variety 1-d network problems. In the first part of Chapter V, an
appropriate notation for eigenvalue problems on networks, due to von Below
[106] for scalar problems, is introduced. The existence and cqmpleteness
of an orthonormal eigensystem for the string operator follows from results
8 1. INTRODUCTION
of Chapter II. Our principle goal is to give an explicit description of the

spectrum and of the eigenfunctions. This is accomplished along the lines of
[106]. The most explicit insight into the spectrum of a network of strings
is achieved for uniform networks with Neumann conditions at all nodes.
It turns out - exactly as in the scalar case discussed by von Below [106],
[105] and Nicaise [75] - that the spectrum is determined by the underlying
graph. Indeed, the eigenvalues are computed from the generalized eigen-
values of the adjacency matrix and the diagonal matrix carrying the edge
degrees at the diagonal entries. To some extent these results also carried
over to planar networks of Timoshenko beams. It is found that the asymp-
totic behavior of the spectrum of such a network is similar to that of a
string network on the same graph, whose asymptotic behavior is governed
by a Weyl's formula. These spectral properties are extremely important in
order to fully understand the degree of controllability of the network. For
example, the lack of approximate controllability can easily be checked by
looking at the eigenmodes. If the control is exerted at a vertex where some
eigenmode has a zero displacement, then approximate controllability fails
to hold if the control is in the Neumann condition at that vertex. On the
other hand, if the spectrum is simple and a Weyl formula holds, we may
deduce spectral controllability on a time interval which can be computed
by the Beuerling Malliavin density. At this point, however, the picture
is far from complete and it remains a challenging problem to achieve a
more detailed knowledge of the spectrum and the corresponding spectral
family. For instance, there does not seem to be a mathematical proof for
the fact that the eigenvalues, which have multiplicities proportional to the
number of vertices in the case of uniform networks, split into a sequence
of simple eigenvalues in the case of rationally independent optical lengths.
This, however, is strongly suggested by a numerical computation of the
spectrum, which is based on methods developed in this chapter. Another
important open problem is to completely characterize approximately or
spectrally controllable networks.
In the second part of Chapter V a variety of numerical simulations of 1-d
networks are presented. The controlled and uncontrolled motion of a single
string is treated first. Attention is restricted to closed-loop controls, in par-
ticular, to absorbing feedback controls. A unit square network of strings is
used to demonstrate how the energy is transmitted through the joints. We
consider also the problem of stabilizing networks containing circuits. It is
shown that dissipative feedbacks at the joints are not capable of extracting
all of the total energy out of the circuit. Rather, there remains a residual
motion, which has zero displacement at the vertices of the circuit, in the
network. Also presented is numerical support to techniques of "channeling
the energy" through the network. These techniques employ either the di-
rection of the flux of energy or dry friction at the joints. In this connection
there are many open problems of both numerical and theoretical natures.
In particular, any generalization to 2-d elements would be of considerable

interest. It is striking how sparse the literature is on the numerical work
related to such networks.
Our consideration of interconnected 2-d elements ;.3 initiated in Chapter
VI with a study of interconnected membranes. We first give a rigorous
derivation of a model of a single nonlinear elastic membrane in 3-d space
and then turn to the problem of modeling interconnected membranes. The
notion of a joint between two membranes, which is necessarily more in-
volved than in the I-d case, is defined and the appropriate compatibility
conditions at the joint are described. The corresponding dynamical joint
conditions are derived from Hamilton's principle. The overall system is
then linearized by replacing the nonlinear strain components by their lin-
ear approximation. Existence and uniqueness of weak solutions of the linear
system is settled along lines of Chapter II. However, as mentioned above,
in this situation singularities may be expected to develop at corners and
possibly along the joints and it is an open problem to describe those mem-
brane networks for which the weak solution is actually a classical solutions
in the case of regular data. As discussed in Chapter II, exact controllability
(exact reachability) is obtained by investigating the homogeneous adjoint
problem, in particular, appropriate observability estimates on solutions of
the adjoint problem are sought. This is a very difficult problem since in-
plane motions are typically coupled to the transverse displacement at the
joints as may be seen by considering, for example, a 3-d configurations of
interconnected membranes such as a thin-walled rectangular tube. Never-
theless, under various geometrical assumptions on the equilibrium shape of
the network and under the standing assumption of sufficient regularity of
solutions, the crucial observability inequality and corresponding exact con-
trollability result are obtained. A particular consequence of our analysis is
an exact controllability result for serially connected membranes and, as a
further consequence, the same controllability result for so-called transmis-
sion problems.
Chapter VII is devoted to the modeling, analysis and control of linked
thin, elastic plates. According to the general program outlined in the pre-
vious chapters, the equations of motions for a single nonlinear plate in
three dimensions are first derived. The model is again an infinitesimal
strain -moderate rotation model along the lines of Wempner [108] (as in
the beam case) and accounts for transverse shearing of cross-sections. Upon
introduction of the Kirchhoff hypothesis (the analog of the Euler-Bernoulli
hypothesis for beams) one may obtain the von Karman system, and, by
linearization, the Kirchhoff plate system. If the general nonlinear equa-
tions are properly linearized, the classical equations of Reissner-Mindlin
plate theory result. Modeling of interconnected plates is considered next.
It turns out to be a very subtle task to properly formulate the problem
of interconnected plates. We first study linked Reissner-Mindlin plates.
10 1. INTRODUCTION
As in the case of beams, there are a number of distinct types of connec-

tions between plates which can be imagined. For example, a joint between
plates can be considered as very stiff and rigid in the sense that it does
not bend and twist, or it might bend in one direction but not in another.
It might rotate if it is straight, and so on. Various possible types of joints
between membranes are defined in terms of appropriate restrictions on the
deformation of the joint. These involve compatibility relations among dis-
placements, tangential and transversal shearing angles as well as in-plane
twisting angles at the joint and possible restrictions on the geometry of the
deformation. The compatibility conditions are nonlinear, in general, but
are linearized in order to obtain a model whose the analysis is tractable.
Upon the introduction of appropriate ramification spaces, the correspond-
ing dynamic joint conditions are derived and the well-posedness of the linear
model is established. Only the existence of weak solutions is ascertained.
As in the case of linked membranes, the problem of when weak solutions
are classical solutions, if the data are smooth enough, is completely open.
Exact controllability results for linked Reissner-Mindlin plates may be ob-
tained in a way similar to that in the linked membrane case. The com-
putations, however, are much more involved and the results which could
be obtained would in the end be only formal because of the unresolved
regularity issues. For these reasons attention is restricted to transmission
problems in which a smooth domain is embedded into a larger (not nec-
essarily smooth) master domain. Observability estimates are derived and
the configuration space is described. The final section of this Chapter is
devoted to linked Kirchhoff plates. Rather than give a separate derivation
of the model, which is possible, we obtain the system of equations from the
Reissner-Mindlin model by letting the shear stiffness increase to infinity.
In this manner, models of interconnected Kirchhoff plates corresponding
to various types of joints are obtained and convergence of the solutions of
the parameterized Reissner-Mindlin model to those of the Kirchhoff model
is established.
The final Chapter studies the coupling of plates and beams. Two spe-
cific configurations are considered. In the first one a beam is attached
to the edge of a plate in such a manner that the centerline of the beam
is orthogonal to the tangent along the edge of the plate at its reference
surface. The second configuration is one in which a beam is orthogonally
attached to one of the faces of the plate. In each case the compatibility
condition at the interface is the requirement that the deformation of the
cross-section of the beam match that of the (3-d) plate throughout the junc-
tion region and thus this condition describes a rigid connection between the
plate and beam. The dynamic conditions are then obtained from Hamil-
ton's principle and lead (upon linearization) to a system which couples the
Reissner-Mindlin plate system to the Timoshenko beam system and the
equation for torsion in a rod. The dynamic coupling equations are nonlocal
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
Modeling of Networks of Elastic Strings
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. Modeling of Nonlinear Elastic Strings

We begin by considering vibrations in IR3 of a single elastic string seg-
ment. If the "natural" length of that string is l it is reasonable to consider
the string as parameterized by the rest arc length x with x E [0, l]. The
position at time t of the point corresponding to the parameter x is to be
denoted by the vector R(x, t). If p is the density of the string and if the
string is assumed to satisfy Hooke's law with constant h the kinetic and
potential energies are given by
(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
entiation with respect to x. The total energy is then
We apply Hamilton's principle to the Lagrangian functional
This requires C to be stationary at R. Of course the domain of C needs to

be specified. On the one hand this involves requirements on the regularity
of R about which we do not wish to be precise at this point. On the other
hand R may also be required to satisfy some geometric condition such as
a Dirichlet boundary condition. Suppose, for example, that we require
(1.3) R(O, t) = Uo(t)

and that no condition is placed on R(l, t), in which case we say that
the right end point is free. In that case one can consider a perturbation
R(x, t) + .MR(x, t) where 8R(O, t) = 0 and there is again no condition at
the other end. In fact we also assume that 8R and 8it vanish for t = 0 and
t = T. Now the stationary condition becomes
d
(1.4) 0 = d)" C(R+ )..8R)IA=o =
[T [I [ . . R x ]
Jo J o pR(x, t) . 8R(x, t) - h[IR,xl- 111R:xl (x, t) . 6R,x(x, t) dxdt.
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
(1.5) [1' [-pii.+h[R"-I~::ILl (x,t) 6R(x, t) dxdt
- h [R,x - 1:::1] (I, t) . 8R(l, t) = O.

Considering first arbitrary perturbations 8R(x, t) satisfying 8R(I, t) = 0,
one obtains the non-linear system of partial differential equations
(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
(1.7) h [R,x - 1:::1] (I, t) = o.

1. MODELING OF NONLINEAR ELASTIC STRINGS 15
The boundary conditions (1.3) and (1.7) need to be complemented by

initial conditions prescribing R(, 0) and R(., 0)
It is easy to verify by differentiation with respect to t that the total
energy (t) is conserved for solutions of the equation (1.6) with boundary
conditions (1.3) (with data which is independent of t) and (1.7).
We can also suppose that a distributed force F(x, t) is acting along the
string and that a force UI(t) acts on the endpoint at x = l. Then we
introduce the following work functional:
(1.8) W(t) = 11 F(x, t) . R(x, t) dx + UI(t) . R(l, t).
Applying Hamilton's principle to the modified Lagrangian functional
(R) = 1T [K(t) + W(t) - U(t)] dt

we get the inhomogeneous equation
.. Rx] x +F,
h [R,x - IR:xl
.
(1.9) pR=
as well as the inhomogeneous free boundary condition
(1.10)
More complicated boundary conditions are possible. Suppose for exam-

ple that the endpoint of the string corresponding to x = 0 is constrained
to slide along a frictionless rod located along the line given parametrically
by c + ab, a geometric boundary condition which can be expressed as
(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
(1.12) h [R.x - 1:::1] (l,t) b = O.

REMARK 1.1. We have no idea how existence theory for the highly non-
linear equation (1.6) can be approached, except in the stationary case.
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. Networks of Nonlinear Elastic Strings

A network consists of more than one string segment each of which is
modeled as above. We suppose that the strings are indexed by i E I =
{1, ... , n}. These strings may have different physical characteristics associ-
ated with constants Ii, Pi and hi. Let the spatial location of the i-th string
be described by Ri(X, t). We then let R denote the n-tuple of all these
location functions. We suppose that strings are joined together at certain
endpoints in such a way that the network is connected.
The model involves conditions at the nodes by which we mean the lo-
cations of the endpoints of the strings. We suppose that the positions of
these nodes are given by Ni (t) with j E J = {j = 1 ... m}. We distinguish
between simple nodes, indexed by j E JS which are the endpoints of only
one string, and multiple nodes, indexed by j E J M at which several strings
meet. Corresponding to j E .:TM we introduce the notation
Ii = {i E I: the i-th string has an end at Ni}.
For i E Ii we also introduce the parameter values Xii to be equal to Ii, if

Ni(t) = Ri(li' t) or 0 if Ni(t) = Ri(O, t). For the purposes of integration
by parts we introduce the sign function Eii to equal 1 if Xii = Ii and -1 if
Xii = O. Then EiiR:x (xii , t) denotes the "outward pointing" derivative at
Xii'
Implicit in the above notation is the following geometric multiple node
condition which requires the continuity of the location functions at each
multiple node:
(2.1) the Ri(Xii' t) are equal for all i E Ii, j E JM,
the common value of course being Ni(t).

At both simple and multiple nodes various conditions can be imposed.
We shall say that a node is clamped if Ni (t) is prescribed in a fixed and
time independent way. Clamped nodes will be indexed by j E JC so that
the explicit condition is
(2.2)
2. NETWORKS OF NONLINEAR ELASTIC STRINGS 17
Some nodes, indexed by j E .JD, may be subject to a Dirichlet condition

which takes the form
(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)
The work functional corresponding to distributed forces Fi(X, t) acting on

the i-th string as well as forces Ui(t) acting on the nodes indexed by j E :TN
is
(2.6) W(t) = L 1 Fi(x, t) Ri(x, t) dx + L

1
Ui(t) Ni(t).
iEI 0 iE3N
Then the Lagrangian functional takes the form
(2.7) .c(R) = [T
10
[L [I. [~i IRi(x, t)12 + Fi(x, t) . Ri(x, t)
iEI10
- ~i [lR:", (x, t)l- 1]2] dx + L Ui(t). Ni(t)] dt.

iE3 N
We apply Hamilton's principle to the Lagrangian (2.7) with variations

oR which respect the geometric node conditions (2.1), (2.2) and (2.3) so
that
ORi,(Xij,t) = oNj(t) for i E I j ,

(2.8) {
oNJ(t) = 0 for j E :Je U :JD.
One obtains
(2.9) d~ C(R + AoR)I,x=o = {T

10
[L (Ii [pi Ri .oR +
iEI10
i Fi oRi
-h[R:x-lll:~xlOR:x](X,t)dX+ L Uj(t).ONj(t)]dt=O.
~ jE3 N
Integrating by parts with respect to both t and x, taking into account

(2.8) and assuming that oR(, t) vanishes for t = 0 and t = T one gets
+ L [- ~ fijhi [R:x - 1:~xll (Xij, t) + uj(t)] oNj(t)

jE3N >EIj ,x
- L [~fijhi [R:x - 1:~xll (Xij, t)] . oNj(t)] dt = O.

jE3F .EI; ,x
By using variations oR which vanish near the nodes one obtains the partial
differential equations
(2.10)
If one now considers oR with support concentrated about a particular

Neumann controlled or free node one is led to the dynamic node conditions
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
of the network is conserved. If one differentiates with respect to t and

integrates by parts with respect to x, taking into account the equation
(2.10), one finds
: = ~ (i [iti(X' t) . PiRi(x, t) + it:x(x, t) . hi [R:x - ,::x,] (x, t)] dx
t
'EI10 ,x
~ [Ri(X,t). h;[R:. - 1:;:I](X,tJ[

= "~ NJ
. .. ~
" Eijh i [. Rix , ](Xij, t) = 0,
R~x - 'R~
j E.:J iEIj ,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
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
Ri(X, t) = Ili(x) + ri(x, t) = :R.b + xSiei + ri(x, t),

Ni(t) = Ni + ni(t) for j E :r,
Ui(t) = Vi + ui(t) for j E :r D,
Fi(X, t) = fi(X, t) for i E I,
where r i (x, t) and ni (t) are "small" displacements corresponding to "small"

perturbations u j (t) of the nodal data, "small" forces rex,
t) distributed
along the strings and "small" perturbations of the initial data. Conse-
quently, the geometric conditions (2.1), (2.2) and (2.3) require the dis-
placement functions to be subject to the node conditions
(3.4) ri(xii,t)=ni(t) foriEIj ,
(3.5) ni(t) =0 for j E :rC,ni(t) = uj(t) for j E :rD.
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:
= Si [1 + ~ei . ri + 2.. lri 12]1/2

Si ,x Si ,x
1 1. 2 1 '2
=si[1+-e'r' +-Ir' I --(e'r') + ... ].
s, .
,x ,
2s~ ,x ,
2s~ ,x
3. LINEARIZATION 21
Instead we return to the variational formulation expanding up to quadratic

terms in r i,x . The kinetic energy and the work functionals are simply
(3.6)
and
(3.7) W(t) = L iii r(x, t) . [Ri + ri] dx + L u j [Ni + nil.

iEI 0 iE.:JN
Both of these expressions contain no terms of order higher than quadratic.

As far as the potential energy is concerned, note first that
The quadratic terms in this expression can be written as Qirix

, . rix
, with
(3.8)
where I is the 3 x 3 identity matrix, ei is written as a column and ei T

denotes the transpose. We then obtain from (2.5)
If in the above expression for U we discard terms of order higher than

quadratic in r and if we then apply Hamilton's principle to the correspond-
ing Lagrangian functional of r = (ri)iEI we obtain
+ 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
t = T. Integrating by parts and grouping terms we then get
[T
Jo
[L Jo[l, [-p.'r' + Q,r~xx]
tEI
. 8r' dx
- L [L [t'Jht(s, - l)e' + t'JQ,r:x]- UJ] 8nJ

JE.:J F 'EI,
- L L [ttJh,(St - l)e' + t'JQ,r~x] . 8nJ] dt O.

JE.:J F 'EI,
Hence one is led to the linear system
(3.9) p,r.. , = Q'

,r,xx lor
t'
z'EI,
as well as the conditions
for j E :IN,
L [t'Jh,(s, -
'EI,
l)e' + t'JQ,r:x] {~' for j E :IF.
In view of (3.3) the latter simplify to
for j E .IN,
(3.10)
for j E .JF.
Now we can streamline the formulation of the linearized problem by

associating it with the following kinetic energy, potential energy and work
functionals:
/C(t)
t,
= L Jo p,Ir'1 2 dx,
L Jot, r' C'dx+ L

'EI 0
W(t) = uJ nJ,
Ill.
'EI 0 JE.:JF
U(t) = ~ - Q,r'x . r'x dx.

L.,., 2 "
,EI 0
One easily verifies that Hamilton's principle applied to the corresponding

Lagrangian directly yields (3.1O)and (3.11).
We note that Q, as defined by (3.9) is a symmetric matrix with eigen-
values h, and h,(l - S;l) corresponding to the eigenspace spanned bye'
and the orthogonal complement of e', respectively. Both eigenvalues are
positive because s, > 1 so that the equations (3.10) are hyperbolic. For
each string the vibration can be uncoupled into transverse and longitudinal
vibrations. The node conditions scramble these modes.
4. WELL-POSEDNESS OF THE NETWORK EQUATIONS 23
The total energy of the network in the linearized model is
(3.11) E{t) = ~ L [I, [p,le{x, t)12 + Q,r:x{x, t) . r:x{x, t)] dx.

'EI Jo
It is easy to check that if u J (t) = 0 for j E .JF and u J (t) is independent of
t for j E .Jj then energy is conserved, as before.
4. Well-posedness of the Network Equations

In this section we consider the well-posedness of the linear problem ob-
tained in the last section. In its entirety this system is as follows:
r'{x'J' t) coincide for j E .JM, i E I J,

(4.1) 1)3r{t) = uJ{t) for j E .JD, 1)3r{t) = 0 for j E .J0 ,
,NJr{t) = uJ{t) for j E .IN, ,NJr{t) = 0 for j E .JF,
r{O) = ro, r{O) = ro,
where r = {r,),EI, r = (r')'EI and the Dirichlet and Neumann boundary

operators 1)3 and,NJ are defined at the nodes by
1)3r{t) = nJ{t) = r'{x'J' t) for i E I J,

(4.2) {
,NJr{t) = I:'EI, f'JQ,r:x {x'J , t).
The governing equations can be rewritten in the form
(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
H = II 2{0, l,; JR3),

'EI
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
(4.5) (r, S)H = L '111;

iEI 0
..
2 Pi rt . s' dx,
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)
which corresponds to the potential energy, as well as
(4.7) (r, s)v,>. = (r, s)v + ).(r, S)H.
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:
Vo = {r E V : 'I)i r = for j E :rD }.

This space is densely imbedded in H. Only in the case of homogeneous
Dirichlet data can the solution r(, t) be expected to remain in Vo. Note
that 11llv is an inner product on Vo if and only if :r c u :r D :I 0. Finally
we introduce the energy spaces
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)
acting on appropriate domains.

In order to specify suitable domains for A we consider first the following

stationary problem which if A = 0 and f is replaced by f / p is just the
stationary version of (4.1):
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.
Given f E H and vectors Uj E 1R3 for j E .JD U.JN we say that r is a

weak solution of (4.9) if and only if r belongs to V and satisfies 1J1r =
uj for j E .JD as well as the weak identity
(4.10) (r,4v,>. = (f,4H+ L u j 1J14>

jE.J N
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
(4.11) (s,4v,>. = (r,4v,>. - (f,4H + L uJ .1J14>

jE.J N
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
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
(4.14) (r, cp )v,1' = (f + [JL - A]r, CP)H

and therefore if r = SAf one has r = SI'(f + [JL - A]r). One can in fact ver-
ify that dom(AA) is the subset of I1iEI H2(0, Ii; IR 3 ) consisting of elements
satisfying all the homogeneous geometric and dynamic side conditions asso-
ciated with (4.10). Finally, in case (', )v is not an inner product on Vo so
that the above arguments did not apply for A = 0, we define Ao = AA + AI
with dom(Ao) = dom(AA)' This definition is indeed independent of A be-
cause it follows from (4.13) and (4.14) that Al'r = AAr - !JL - A]r so that
[AA + AI]r = [AI' + JLI]r. In this way we obtain a self-adjoint operator Ao
in H for which it is easy to verify that
(Aor, ~)H = -(r, ~)v for r E dom(Ao), cP E Yo
Therefore Ao is negative semi-definite (and definite if (', )v is an inner
product on Yo) and the spectrum consists of an decreasing sequence of
non-positive real eigenvalues of finite multiplicity accumulating at 00 and
corresponding to a complete orthonormal system of eigenfunctions. The
first eigenvalue is zero if and only if there are neither clamped nor Dirich-
let controlled nodes and the corresponding eigenspace then consists of r's
which are identically constant across the network.
Returning to the non-stationary problem we introduce the operator Ao
on 'H.o by
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
and hence generates a group of unitary operators UO(t)tElR on 10. If, on

the other hand, .:Jc U.:J D = 0, this is not quite true. We then use the
inner product (., )V,l on Vo = V and set Voo = ker(Ao) and VOl equal
to the orthogonal complement of Voo in V. In the present situation Voo
consists of r's which are constant across the network. On VOl (., )v is
again an inner product. We also set Hoo = Voo and HOI equal to the
orthogonal complement of Hoo in Hj the latter is simply the direct sum
of all the eigenspaces corresponding to non-zero eigenvalues. One easily
verifies that V(n is dense in HOI and that the restriction AOI of Ao to
dom(AOl) = dom(Ao) n HOI is a self-adjoint operator in HOI. Then
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:
ro x ro = roo x roo + rOl x rOl.

One can then define
(4.17) Uo{t)[ro x rol = Uoo{t)[roo x rool + U 01 {t)[rOl x r01l.
Vo is a proper subspace of V if and only if .:JD #- 0. In that case we would
like to extend the semigroup from Vo to V. In order to do so let VI denote
the orthogonal complement of V 0 with respect to the inner product (., )v.
Then the condition that rl belongs to VIis equivalent to it being a weak
solution of
Ar=O,
ri(xij) coincide for j E .:JM, i E I j ,
1Yr = 1)jrl for j E .:JD, 1Yr = 0 for j E .:Jc ,
}/jr = 0 for j E .:IN U .:JF.
Then rl can be regarded as a stationary solution to the hyperbolic system

with Dirichlet data determined by itself. Hence given ro x ro E 1 we can
first decompose ro in V as ro = roo + rl with roo E Vo and rl E VI. One
has 1Yrl = 1Yro for each j E .:JD and the orthogonality condition is the
weak formulation of Ar1 = 0 and ./Vjr = 0 for j E .:IN U .:JF. The initial
data can also be decomposed into
ro x ro = [roo x ro] + [r1 x 0],
and one then defines
U(t)[ro x ro] = Uo(t)[ro x ro] + [r1 x 0].

This gives a unitary semigroup on 1 which solves the initial value problem
for initial data in 'H., with vanishing Neumann data and distributed forces
and time independent Dirichlet data u j = 'lY ro. The generator of this
group is the operator defined by
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
(4.18) r(t) x r(t) = U(t)[ro x r] + lot U(t - s)[O x f(s)/ p] ds.
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.
LEMMA 4.1. Let B be the generator of a strongly continuous semigroup

T(t) on a Banach space X. Suppose that f(t) E C~((O,T];X) and that
Xo E dom(Bk). Then the mild solution of the equation
d
dtx(t) = Bx(t) + f(t), x(O) = x o
belongs to Ck([O,T];X) n Ck-1([0, T]; dom(B)), where dom(B) is normed

by the graph norm II x II~=II Bx II.~ + I x II~. Moreover, for i = 1, ... , k,
the i-th order time derivatives x(t)(t) satisfy the equation
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
Uj(t) -'Djro E cgo((O, T]; JR3) for j E 3 D,

(4.19) {
Uj(t) E cgo((O, T]; JR3) for j E 3 N.
Since we already know how to obtain mild solutions corresponding to initial
data and distributed forces we can, for the moment, let these vanish. In
that case compatibility requires that Uj (t) E Cgo (( 0, T]; JR3) for j E 3 D.
Note that functions in Cgo((O, T]; JR3) vanish near t = 0. Let r(t) denote
the weak solution of (4.9) for arbitrary, but fixed, t in [0, T] with f = 0,
uj = uj(t) for j E 3 D U3N . Then r(t) E Coo((O,T];V) and it is also
easy to check that each ri(x, t) E COO ([0, Ii] x [0, T]; JR3). Now one seeks a
solution to (4.1) having the form r = r + s where s has to satisfy
PiSi = QiSi,xx + fi for i E I,

Si(Xij, t) coincide for j E 3M, i E I j,
(4.20) 'Dir(t) = 0 for j E 3 D U 3 C ,

Njr(t) = 0 for j E 3 F U 3 N ,
s(O) = 0, s(O) = 0,
with i = -pi + .xri. This system has the mild solution
s(t) x s(t) = lot U(t - 8)[0 x (8)/p] d8.

It again follows from Lemma (4.1) that each ri (x, t) E Coo ([0, Ii] x [0, T]; JR3)
and hence the same is true for each component ri(x; t) of the solution
r = r+ s.
We have therefore proved the existence of smooth solutions to (4.1) cor-
responding to given smooth data. This existence result can be extended to
less regular data by first proving an a priori estimate for solutions corre-
sponding to smooth data.
Here and later the following notation will be useful:
ei(x, t) = ~ [pilri(x, t)12 + Qir~x(x, t) . r~x(x, t)],

and
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]).
LEMMA 4.2. Let r be a smooth solution of
(4.21) pf{x,t) = [Qr,xl,x for (x,t) E [O,l] x [O,T],

where we allow Q and p to depend on x and m{x) is a given smooth mul-
tiplier function. Then
(4.22) [T[e{x,t)m{x)]X-I dt = [T e{x,t)m,x dxdtt

10 x=o 10 10
+ ~ loT 101[Pxlrl2 - Q,xr,x . r,x] (x, t)m{x) dxdt
+ [I [mpr . r,x] t=T dx.
10 t=O
As a consequence
(4.23) 1 (0; 0, T) + 1 (I; 0, T) :S C [1 T

(t) dt+ (0) + (T)].
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 [mpr . r,x] t=T dx

(4.24)
10 t=o
= ~ loT Io [m,xplrI 2+ mp,xlrI2] dxdt

l
- i [m
o
]
T -2 plrl2 x=1 dt +
x=o
11 [
0
]
mpr . r,x t=T dx,
t=o
where we have used the identity
mpr r., = ~mp[lrI2l., = ~[mplrI21,x - ~m,xplrI2 - ~mp,xlrI2,

and integrated [mplrl21,x with respect to x. To obtain the next identity we
note that
1 1
[Qr ,x ],x . r ,x = -[Qr
2 ,x . r ,x ],x + -Q
2 ,oX r ,x . r ,x
4. WELL-POSED NESS OF THE NETWORK EQUATIONS 31
and then integrate by parts with respect to x to get
[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
In the remainder of this section we shall be making many estimates

involving a constant C, which will simply be redefined from estimate to
estimate with no change of notation.
LEMMA 4.3. r be a smooth solution of (4.1). Then
PROOF. We differentiate (t), integrate by parts with respect to x and

regroup the terms taking into account the system (4.1). This gives
=~ i
~ [Qr
L...J L...J 'J 'J' T)
.,x . ri](x
iE.7 iEIj
By using the estimate ab ::; w 2 + b2 j 4, this gives
: (T) ::; [ L IN'ir(T) 12 + L l1)i r (TW]

iE.7 D iE.7 N
L[L
(4.27)
+ liti(TW + L IUi(TW].
jEJ D jEJ N
Note that
11)1r(r)1 2 ~ Gei(Xii,t) for any i E Ii'

for any J' E .:J {
.. I" 2
IIVJr(r)1 ~ GL:iEIj ei(xii,t).
Hence, integrating (4.27) from 0 to t and using (4.23) we obtain
(t) ~ (0) + 10 t [L INir l2 + L l'Dir(rW] dr

Jo iE:JD iE:J N
+ LJot [L
iE:JD
Itii (r)1 2 + L
iE:J N
lui (r)1 2] dr
~ (0) + lOG [lot (r) dr + (0) + (t)] dr

+ 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
(4.28) E(t) $ C [E(O) + l' E(T) dT
+ f.\'FTD lui(T)I' + ;'FTN IU;(T)I']d+
An easy application of Gronwall's inequality yields (4.26). 0
THEOREM 4.1. Suppose that

(i) ro E V, ro E H, f E L 2 (0, Tj H);
(ii) ui E H1(0, Tj IR3) and satisfies ui(O) = 1)1ro for each j E .:JD;
(iii) ui (t) E L 2 (0, Tj IR3) for each j E .:IN.
Then there exists a unique r E G([O, T] : V) n G1 ([0, T] : H) such that
1)1r(t) = ui(t) for j E .:JD
and which satisfies the remaining requirements of (4.1) in the following

weak sense: for any c/> E Vo the function (r(t), c/H has an absolutely
continuous first derivative in t and almost everywhere satisfies
(4.29) ~:2 (r(t),c/H + (r(t),c/v = (f(t)/p,c/H + L ui(t) .1)1c/>

iE:J N
5. CONTROLLABILITY OF NETWORKS OF ELASTIC STRINGS 33
as well as the initial conditions
PROOF. Note first that the requirements on the regularity of (r(t),4H

as well as (4.29) and (4.30) are all neatly contained in the identity
(4.31) (r(t),4H = (ro, 4H + t(ro, 4H +lot loT [-(r(a), 4v

+ (f(a)/p,4H + L u (a).1)i4>]dadr.
j
jE.::J N
Next we prove the uniqueness of solutions. Given two solutions corre-

sponding to the same data we let s denote the difference. Then s(t) E Vo
and
(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
(s(t),4H = lot loT [-J.L(s(a),4H] dadr.

This implies that (s(t),4H = y(t) satisfies
ii(t) = -J.Ly(t), y(O) = 0, y(O) = 0,
and hence vanishes identically. From the completeness of the system of
eigenfunctions one concludes that s(t) = 0 for all t, which completes the
proof of uniqueness.
For the proof of existence we note that this has already been established
for smooth data. It is easy to verify that the corresponding smooth so-
lutions satisfy the identity (4.31). Smooth data is dense in the class of
data specified by the hypotheses. We then obtain solutions corresponding
to arbitrary data by "continuity", making use of the a priori inequality
(4.26). 0
5. Controllability of Networks of Elastic Strings

We introduce the spaces of controls. First let uD(t) and uN(t) respec-
tively denote (Uj(t))jE.::JD and (uj(t))jE.::JN. These are required to belong
to the spaces
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
(u D , VD)UD = L iT Ui(t). vi(t) dt

iE.:J D 0
and
(UN, vN)UN = L iT ui(t). vi(t) dt.
iE.:J N 0
It is also convenient to let U = U D X UN with elements u = u D X uN.

By Theorem 4.1 we know that for u E U there exists a unique solution
to the following system, in which the distributed forces and the initial data
have been set equal to zero:
p/ii = Qir~xx for i E I,

ri (xii , t) coincide for j E .:rM, i E Ii,
(5.1) 1Yr(t) = ui(t) for j E .:rD, 'Dir(t) = 0 for j E .:rc ,
,N'ir(t) = ui(t) for j E .:rN, ,N'ir(t) = 0 for j E .:rF,
r(O) = 0, r(O) = o.
The solution defines a continuous trajectory r(t) X r(t) in 'H.. We define the
control to state map CT : U 1-+ 'It by CTu = r(T) xr(T). This is a bounded
linear operator. We would like to prove that this operator is surjective for
T sufficiently large, which is equivalent to the assertion that the network
is exactly controllable, and also closely related to the stabilizability of the
network which will be discussed later.
It is well known that the surjectivity of the bounded linear operator CT
is equivalent to the following estimate on the adjoint operator C:} : 'It 1-+ U:
(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
(5.3) 1Y,x(t) = 'Di,xo for j E .:rD, 1Y,x(t) = 0 for j E .:rc ,

,N'i ,x(t) = 0 for j E .:rNU.:J F,
,x(T) = ,xo, "\(T) = ..\0.
We then have the following characterization of C~.
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
(5.4) C~[,xo x ~ol = {t

Jo Jt/i ,x(T) dT} jE.:J D
x {'!)i ~(t)}
jE.:J N
,
where ,x is the solution of the adjoint system (5.3).
PROOF. The hypothesis allows us to use (', ')v as inner product on V-

components of elements of 1. We evaluate
~t [(r, ,x)v + (r, ~)H]

= -ddt '""
~
iEI 0
1- 1; 1[
2 Qr
t..,xt
t ,x
' + pe..,x
,x t
, dx i)
+ L I>ij [Qir~x . ~i + Qiri . ,x~x](Xij, t),

jE.:J iEIj
where we have integrated by parts with respect to x. Thus we obtain
~t (r x r,,x x ~)'H = L ['!)i ~. N'jr +N'j,x .'!)ir]

jE.:J
(5.5)
= L N'j,x. u + L 'D j ~ . u
j j .
jE.:J D jE.:J N
Integrating with respect to t from 0 to T one finds
(CTU,,xO x ~o)'H = [T [ L N'j,x. uj + L Dj~ u j ] dt.

Jo jE.:JD 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
Because of Proposition 5.1 the estimate (5.2), characterising exact con-

trollability in time T, becomes
(5.6) 11.\0 x ..\0 II~ ~ C iT [L INj

o jE.:rD
.\(tW + L
jE.:r N
11)i ..\(t) 12 ] dt.
This is a nontrivial estimate which will be established below under the

strong hypothesis that the network forms a "tree" with controls acting at
all simple nodes except for one, which is clamped. The estimate will in
fact hold when T is larger than a certain critical time T*, for which we give
an explicit expression. Examples will then be given of networks containing
"closed circuits" for which the kernel of the adjoint operator is non- C;,
trivial for all T > 0, so that the network is not exactly controllable over
any finite time interval.
5.1. Exact Controllability of Tree Networks. The proof of the

estimate (5.6) for tree networks depends on the energy estimate stated in
the following Lemma. This is a special case of the more general Theorem
4.3 of Chapter IV and no proof will be given for the moment. However
the estimate is intuitively plausible if one remembers that in hyperbolic
equations the energy flows along and between characteristics.
LEMMA 5.1. Let r E W 2([0, l] X [tl, t2];.IR3 be a solution of
pi = Qr,xx on [0, l] x [tl, t2],

where Q has eigenvalues hand h(l - 8- 1 ), and the corresponding families
of chamcteristics are given by
. 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
which is valid for all t E [t1

and l interchanged.
+ 8*, t2 - 8*]. The same estimate holds with
This lemma can be applied to each element of the network; in the process
we need to introduce the times T! = 2S! associated with the i-th element.
These times represent the maximal time it takes for a wave to move back
and forth along the i-th element.
A second lemma is an immediate consequence of the geometric and dy-
namic conditions at multiple nodes.
LEMMA 5.2. Suppose that r is a solution of the network equations (5.1)

with the homogeneous initial condition replaced by the requirement r(O) x
r(O) E 1(.. Let io E I j for j E JM' Then
(5.8) Ei~(Xioj,tl,t2)::; C L El(Xij,tb t2)

iEIj \{io}
We are now in a position to state and prove our principal result on

exact controllability. We let [iI, i 2 , , i r ] denote the path which leads
along the network elements labelled successively by ib i 2 , ,ir , with the
understanding that successive indices differ. A closed circuit in a network
is then a path for which ir = i l .
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
T* = max: { T;l + T.!1+ ... + T;r : lib h, ... , i r ] leads from

(5.9)
the clamped node to a controlled node}.
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!
We introduce some notational conventions involving simple nodes. For

j E JS the set I j consists of a single index ij. We then write Xj = Xijj and
f.j = f.ijj.
To prove Theorem 1.1 one needs to establish the estimate (5.6) under
the given hypotheses. This estimate is an immediate consequence of the
more general estimate contained in the following theorem as it applies to
the solution ~ of (5.3), since
The boundary conditions satisfied by .,X imply that,
{
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:.
The boundary conditions also imply energy conservation so that

(~) = (T).
By Remark 5.2, (T) is equivalent to II.,Xo x Ao II~.
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)
where m is the index of the clamped node.
PROOF. We note first that because of the estimate in Lemma 4.3 it is

enough to establish (5.1O) for smooth solutions corresponding to smooth
data, thus avoiding problems with regularity. The proof is by induction on
the number n of elements. For n = 1 the result follows from Lemma 5.l.
Suppose now that the estimate (5.10) holds for networks with fewer than
n elements. Given a network with n elements, we can locate a multiple
node, indexed by jo say, such that all "neighbouring" nodes except at most
one are simple. To find this node one chooses a "longest" path leading
from the single clamped node to another simple node. Here "length" is
to be understood in terms of the number of elements traversed, with the
understanding that no element is to be traversed more than once. Then we
choose the penultimate node along that path. That node cannot have more
than two neighbouring multiple nodes. If it did one would have to lie along
the longest path from the clamped node and the path could be continued
beyond the other to yield a longer path and hence a contradiction. Gen-
erally, unless the longest path traverses only two elements, the jo-th node
has exactly one multiple node as neighbour. Note that the step we have
just described depends on the assumptions that the network is connected
and contains no closed circuits.
We now consider the reduced network obtained by discarding the ele-
ments leading from our selected node to a neighbouring simple node other
than, possibly, the clamped node. The reduced network has fewer than
n elements and the I-th node is now simple. It is convenient to relable
the original network in such a way that the discarded nodes have indices
1 ~ j < jo, the discarded elements leading to the discarded nodes have
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}
where the tilde identifies quantities pertaining to the reduced network. We

note that
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}
:::;G l" _L.,;

E.1s\{m,jo}
i . T* T* T* T*
iji(xj,0,T*)+jo(Xjo'2-2'2+2)
. 1
Here there is a small notational inconsistency arising from the fact that the
node indexed by jo is multiple for the original network but simple for the
reduced network: we have written Xjo in place of Xjojo. Now by Lemma
5.2
Using (5.13) and (5.14) we now get
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
5.2. Lack of Controllability for Networks with Closed Circuits.

In this section we shall show that if certain "rationalitv" condition!'; on
the physical constants are satisfied a. system with closed loops will not be
exactly controllable. These conditions allow one to show that the adjoint
operator C;, has a non-trivial kernel for all positive T, in which case CT
can never be surjective. This raises a number of questions concerning the
image of CT which we have not been able to resolve.
We try to construct an element of the kernel of C;,
with the help of an
"aberrant" eigenfunction cjJ = (cjJi)iEI corresponding to an eigenvalue w2
which satisfies not only the eigenvalue equations
Pi W2 cjJi = QicjJ~xx for i E I,

cjJi{Xij) coincide for j E .:JM, i E I j ,
(5.16)
1J3cjJ= 0 for j E .:JD U.:J c ,
N'jcjJ{t) = o for j E .:IN U .:JF.
but also the supplementary conditions
1J3cjJ = 0 for j E .:IN,

(5.17) {
N'j cjJ = 0 for j E .:J D.
If we then define ~ = cosw{t - T)cjJ we obtain a solution of the adjoint

system (5.3) corresponding to terminal data ~o = cjJ and .xo = 0 with
CH~ x .xol = 0 by (5.4).
The system (5.16) determines the eigenvalues associated with the sta-
tionary network equations and hence, as discussed in Section 1.4, gives
rise to a sequence of eigenvalues of finite multiplicity. The supplemental
conditions (5.17) lead to an overdetermined system which can only be ex-
pected to have a solution under special circumstances. We shall illustrate
a situation where these arise.
We consider a single closed circuit of r strings indexed by i = 1, ... ,r,
where such a circuit might be thought of as standing alone or might in fact
be part of a larger network. We suppose that the indexing and parametri-
sat ion are such that the circuit is described by the path [1,2, ... , rl with
parameter x increasing from 0 to li as the path traverses the i-th string. It
is also convenient to identify the index r + 1 with 0 in accordance with the
cyclic nature of the circuit. We rewrite the systems (5.16) together with
(5.17) explicitly for (q,)i=1 as

PiW2q,i = Qiq,~xx for i = 1, ... ,r,
(5.18) { q,i(li; = q,i+1(0) = 0 for i = 1, ... r, I
Qiq"x(li) = Qi+1q,~:I(O) fori=l, ... ,r.

This corresponds to a situation where in the primal system controls are
active at all the nodes.
In an attempt to solve (5.18) we consider first a single equation, dropping
the index i,
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,
then (5.19) has the solution
(5.22) q,dx) = ~k(x)ek with ~k(X) = sin(v'P

Wk
wx).
For a single element such a choice of W is of course possible and in fact one
then obtains a sequence of suitable values for w.
Returning to the system (5.18) we again index all quantities by i and
make the very strong, yet easily satisfied, assumption that
#.,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'
where n~ is an integer which we can, without loss of generality, assume to

be even. If we then let
3
q,i(x) = Lc~~l{x)e~ with ~l(x) = sin(ffwx),
k=1 k
where the c~ are arbitrary coefficients we obtain a solution to the first

two conditions of (5.18). We now show that the coefficients can be chosen
in such a way that the third condition is also satisfied. That condition
translates into
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. 8tabilizability of 8tring Networks

In this section we shall prove only a rather rudimentary result, leaving
some of the details and much of the elaboration to Chapter N, where we
shall be dealing with a more general class of networks including beam net-
works. We proceed along the usual lines of introducing a velocity feedback
into the system in such a way that energy is obviously dissipated; one then
6. STABILIZABILITY OF STRING NETWORKS 43
verifies that the decay is actually exponential or polynomial, depending on

the nature of the feedback.
To motivate the selection of a suitable feedback we once again (as, for
example, in the proof of Lemma 4.3) differentiate the energy with respect
to t to get, in case all multiple nodes are free and one simple node, indexed
by m, is clamped,
(6.1) dE (t) =
dt
L Afi r .1)if(t).
iEJ"s\{m}
Consequently the "feedback law" Afir(t) = -ai1)if(t) for j E :/s \

{m}, where each ai is a positive scalar, ensures energy dissipation of the
"feedback system"
Piri Qir~xx for i E I,

ri(xii' t) coincide for j E :/M, i E Ii,
Afir(t) = 0 for j E :/M,
(6.2)
1)mr = 0,
Afir(t) = -ai1)if(t) for j E :/s \ {m},
r(O) = ro, f(O) = fo.
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
has after readjustment of the constant C
L OJ iT< l1)ir(tWdt ~ -C I\Us(~*)[ro x ro]ll~

jE.7s\{m} 0
~ -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
IIUs(T*)[ro x ro]lll~ (3llro x ro 1\1, with (3 = (1 + C)-! < l.

Therefore IIUs(T*) 11< 1 and the group Us(t) decays exponentially.
The argument used above also applies, with no significant changes, to
the non-linear feedback law
under the assumption of continuity and
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.
In Chapter 4 we treat these matters for a wide class of networks. There

we shall also introduce nonlinear feedbacks yielding only polynomial decay
rates, as well as setvalued feedbacks for which we can prove only that the
energy does decay to zero.
7. String Networks with Masses at the Nodes

In complex flexible structures it often happens that the various elements
are linked together by joints which are rigid bodies. Since these are usually
small in relation to the flexible elements it seems reasonable, at least as
a first approximation, to think of these as point masses. Gravitation will
of course act on these nodal masses as well as on the flexible elements. If
the latter are very flexible the nature of the equilibria, and hence of the
equations of motion linearized about an equilibrium, will be different from
what we have described in the earlier parts of this chapter. These modelling
issues, as they relate to networks of elastic strings, have been treated in the
technical report [95]. Many aspects of the controllability of such a system
are discussed in the doctoral thesis of Wei Ming [72]. Hanson and Zuazua
have treated the exact controllability of a "two string with one point mass"
system in [34]. A generalization of their results and of those of Wei Ming
are the subject of [96].
7. STRING NETWORKS WITH MASSES AT THE NODES 45
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
(7.1) JC(t) =L ~ [Ii pi/Ri(x, tW dx + L ~mj/Nj(t)/2

iEI Jo jE.JM
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,
as well as the multiple node condition
One can begin by studying the equilibria subject to time independen-

dent Dirichlet data. In [95] the existence of unique "stretched equilibria" ,
described by functions iti satisfying /it!x/ > 1, is proved by showing that
the potential energy function achieves its minimum. What is interesting
about this argument is that the potential energy function is not convex and
that there are sometimes also other equilibria which involve "compressed
strings" for which /itix/ < 1; this latter class of solutions is presumably
unstable. In a stretched equilibrium each string occupies a convex curve
in a vertical plane, which is of course consistent with ones intuition. If
one uses the procedure of Section 3 to linearize about the equilibrium one
obtains a linear system
(7.5)
along with the conditions
(7.6) mjiij(t) + L ijQi(xij)r~x(xij,t) o for j E JM.

iEIJ
Here Q{x) is the symmetric 3 x 3 matrix whose eigenvalues are hand

h (1-1R.,x{x)I- 1 ), with the corresponding eigenspaces consisting of the
span of R.,x{x) and the orthogonal complement of R.,x{x). Since 1R.,xl > 1,
Q is positive definite.
Well-posedness of the system can be established using variational and
semigroup arguments; we refer to [95] for details. The question of exact
controllability has an interesting feature which was first noticed by Hanson
and Zuazua. They consider two strings which are both joined to a point
mass, with the other end of one string clamped and the remaining end
of the second string controlled. There is then an increase of regularity as
one crosses the mass from the controlled to the clamped string. One way
of seeing this is to note that the multiple node condition (7.6) together
with energy estimates on the terms appearing on the left hand side of that
condition ensure that at the mass ii{t) is square integrable. Therefore the
equation for the string with a clamped end now has Dirichlet data with
an extra order of smoothness which raises the order of regularity of the
solution. We can therefore no longer expect exact controllability into the
energy space which was previously used for string networksl, although ap-
proximate controllability is easily proved within that space. In [34] and
in [72] (which also treats in detail serial networks with arbitrarily many
elements) exact controllability results are obtained using spaces which re-
flect the increase in regularity. In [96] these ideas are developed for general
networks.
CHAPTER III
Modeling of Networks of Thermoelastic Beams
1. Modeling of a Thin Thermoelastic Curved Beam

We consider the deformation of a thin beam of length i with a given
initial curvature and torsion, and with variable doubly symmetric cross
section. To be precise, the undeformed beam, in its initial reference con-
figuration, occupies the region
n := {r := ro(xd + X2e2(xd + X3(Xl)e31 Xl E [0, iJ,

(X2' X3) := X2 e2(xd + X3 e3(Xl) E A(xd},
where ro : [0, i] t--+ IR3 is a smooth function representing the centerline

of the beam at rest, and the orthonormal triads el(-), e2('), e3(-) are
chosen to be smooth functions of Xl such that el is the direction of the
tangent of the centerline with respect to the variable x!, i.e. d:
1 ro(xd =
el(xl), and such that e2(xl), e3(xl) span the orthogonal cross section at
Xl' The meaning of the variables Xi are as follows: Xl denotes the length
along the undeformed centerline, X2 and X3 denote the lengths along lines
orthogonal to the reference line. The set n can then be viewed as obtained
by translating the reference curve rO(xl) to the position X2e2 +X3e3 within
the cross section vertical to the tangent of roo The cross section at Xl is
defined as
A(Xl) is assumed to vary smoothly with respect to Xl in a sense that will be

made clear later on. In order that certain moments vanish, we will restrict
our attention to cross sections which are doubly symmetric in the sense that
(X2' X3) = X2e2 + X3e3 E A(xd implies that (-X2' X3), (X2' -X3) E A(xd,
and hence -(X2' X3) E A(Xl)'
48 III. NETWORKS OF THERMOELASTIC BEAMS
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.2 =, e2 . el,l = -el . e2,1 = -flOI2'

K.3 = e3 . el,l = - el . e3,1 = -flOI3 ,
T = e3 . e2,1 = -e2 . e3,1 = -fl023'
The matrix flo, in the basis ei, is given by
-K.3)
-T .
Let us also introduce the axial vector

which satisfies flowo = 0, hence
After deformation, the particle has the position

R(Xb X2, X3, t).
We introduce the displacement vector by
V:=R-r.
A position on the reference line (X2 = X3 = 0), denoted by
ro = r(xI,O,O),
is deformed to
Accordingly, the displacement vector of a particle on the reference line is

W := V(XI' 0, 0) = Ro - roo
1. NONLINEAR ELASTIC BEAMS 49
The tangent vector at any particle in the undeformed configuration is

denoted by
where the subscript separated by a comma means spatial derivation with

respect to variable Xi' Accordingly, the tangent at this particle in the
deformed configuration is
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
with the conjugate function

Another canonical example is a rectangular cross section with length a, b

along the X2, X3 directions, respectively. We only reproduce from Sokol-
nikoff [104] the formula for <p:
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.
(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,
r,l = (1 - X2K.2 - X3K.3)el + X2'Te3 - X3'Te2
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
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
The remaining strains are defined by requiring symmetry ij = ji. We

may now use the strains ij in order to express the triad Ei in tenns of the
moving orthonormal reference triad ei as follows:

E1 = e1 + t11e1,
{
(1.5) E2 = ~2 + ~22~2 + ~23~3 + 2~21 ~1'
E3 = e3 + C33e3 + C23e2 + 2c31 e1
REMARK 1.2. If the distortion of the plane of cross section is neglected,
that is if we neglect the strains t22, t33, t23, then we can express the normal
N to the cross section as
N = E2 X E3 = e1 - 2t21e2 - 2t31e3,
and then the strain vector "y introduced by
"y := E1 - N = t11 e1 + 2t21 e2 + 2t31 e3
is precisely the one introduced in [22], [89], [100, 101] etc., where the as-
sumption of rigidly moved cross sections is made at the onset. In addition,
in the case that the strains tij are small in the sense that the products tijtkl
may be ignored regardless of the magnitude of the rotation that carries the
triad ei into ei, we obtain
(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
88Xl ei(xt} = (88 r 1(X1))r1(X1)t ei (X1)

Xl
+ (r 1(xt)88 rO(X1)rO(X1)t)r1(xde(xt).
Xl
Therefore, upon defing the skew-symmetric matrices
01(X1) := (88 r 1(X1))r1(X1)t,

Xl
O(X1) := 01(X1) + r1(Xt}OO(X1)r1(X1)t,
we can express the derivatives of the triads with respect to Xl as
88Xl ei(x1) = O(x1)ei(xt},
88Xl ei(x1) = OO(x1)ei(x1).

We may now introduce "curvatures" Ki and "twist" T in the same way

as with the undeformed reference curve, but now based on the moving triad
ei, as follows
{K2 A A A A
:= ~2 . ~l,l = -~l . ~2,1,

(1.7) K3 := e3 . el,l = -el . e3,1,
T:= e3 . e2,1 = -e2 . e3,1.
With this notation O(Xl) can be represented as
(1.8)
-K3)
-T ,
o
where we have omitted the arguments. We also define the axial vector
satisfying Ow = 0 and
It is possible to express the rotations above by means of the Rodrigues

formula, see (22), [101) (in fact, one might want to use Euler or Rodrigues
parameters). However, it turns out that the use of dextral rotations provides
a more transparent interpretation of the angles within the reference frame
ei. The latter approach has been used by Wempner [108) and Kane et
al [41). It is our goal to express the strains Cij in terms of the reference
strains tij and the strains related to the rotation. That rotation is obtained
by successive dextral rotations about the el, e2 and e3 axes, respectively,
through respective angles 81, e 2 and 8 3. The result is
(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
the frame ei, explicitly as
K2 =9~ - 9 1 sin 9 2 + K2 cos 9 1 cos 9 2

+ K3(sin 9 1 cos 9 3 - sin 9 2 sin 9 3 cos 8t}
+ r(sin 8 1 sin 8 3 + sin 8 2 cos 9 1 cos 8 3,
K3 =9~ cos 8 3 - 8 1 sin 8 3 cos 8 2 - K2 sin 8 1 cos 8 2
(1.10)
+ K3 (cos 8 1 cos 8 3 sin 8 1 sin 9 2 sin 8 3)
+ r(sin 8 3 cos 9 1 - sin 83 sin 9 2 cos 9 3),
T =9~ cos 8 2 cos 8 3 - 8~ sin 8 3
- K2 sin 8 2 - K3 sin 8 3 cos 9 2 + r cos 82 cos 9 3.
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
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
e1=(1- ~(8~ + 8~))e1 + 83e2 - 82e3,
(1.12) e2=(818 2 - 8 3)e1 + (1 - ~(8~ + 8~))e2 + 81e3,

e3=(82 + 8 18 3)e1 - (81 - 8 28 3)e2 + (1- ~(8~ + 8~))e3.
The dot above the equality sign indicates an approximate identity. The
finite rotation approximation of (1.10) is
+ r818 3 - 1 (2
K38 28 3 - 2K2 8 1 + 8 32) ,
K3 - K3= - 8 2,1 - K281 + r83 - 8 38 1,1 - r818 2 - ~K3(8~ + 8~),

T - r=81,1 - K282 - K383 - 82,183 - ~r(8~ + 8~).
These expressions are analogous to the corresponding terms in Reissner
[89], where a slightly different triad is used. The quadratic approximation
to (1.11) is given by
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.
It is possible to derive the equations of motion including all of the strains

above, in particular including the strains which are responsible for the
distortion of the plane of cross section. In theories of thin beams, however,
the strains t22, t23, t33 are usually neglected, and theories based on this
assumption are still called geometrically exact within the literature. Under
this assumption, we can express the triad Ei as follows:
1
E1 =[1 + tn - 2(8~ + 8~)]e1 + (1 + tn)83e2 - (1 + tn)82e3,
1
E2=( -83 + 8 18 2 + 2t21)e1 + [1 - 2(8~ + 8~) + 2t2183]e2
+ (8 1 - 2t2182)e3,
E3=(82 + 8 18 3 + 2t31)e1 + (-81 + 8 28 3 + 2t3183)e2
+ [1- ~(8~ + 8~) - 2t3182]e3.
We calculate the normal N = E2 X E3 to the cross section with respect to

the reference triad ei and take the quadratic approximation
1
N=[1 + 2t2183 - 2t3182 - 2(8~ + 8~)]e1
+ (83 - 2t21 + 281t31)e2 - (8 2 + 2t31 + 281t21)e3.
Let us now introduce yet another rotation, namely, the one taking the
reference triad ei into the triad N, E 2 , E 3 Once again, we describe this
rotation by the way of three dextral rotations 'I9 1 ,{h, '193 about e1, e2, e3,
respectively. If we take the quadratic approximation (the finite rotation
approximation) of the latter, we can make the following identifications:
'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
2e21 = - '193 + 8 3 + 8 1('192 - 8 2 ),

(1.14) {
2e31 ='192 - 8 2 + 8 1('19 3 - 8 3 ).
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
tll = (W,I + el) . el - 1

1 2
= WI,I - ~2W2 - ~3W3 + 2"(8 2 + 8 3).
2
8 2 = - W,I . e3 = -(W3,1 + ~3 WI + TW2),

8 3 = W,I e3 = W2,1 + ~2WI - TW3
Therefore, we can express the reference strains as
tll =Wl,l - ~2W2 - ~3W3 + ~[(W3'1 + ~3Wl + TW2)2

+ (W2,1 + ~2WI - TW3)2],
1
t2I =2"(W2,1 - f)3 + ~2WI - TW3 + f)1(f)2 + W3,1 + ~3WI + TW2)),
1
t3l =2"(W3,1 + f)2 + ~3WI + TW2 + f)1(f)3 - W2,1 - ~2WI + TW3)).
The columns of the deformation gradient (Gi)i with respect to the moving
triad ei can be expressed as follows:
G I =[1 + tll+ X2( -K2 + 2t21,1) + X3( -K3 + 2t31) + 'Y,Ijel

+ [x22t2lKl + X3(2t3lKl - T) + 'YK2]e2
+ [X2(2t2I + T) + X32t3IK3 + 'YK3]e3,
G 2 =e2 + 2t2l el + 'Y,2eb
G 3 =e3 + 2t3l el + 'Y,3e l,
We are going to evaluate the scalar products G i . G j required for the
Green -St.Venant strain tensor introduced above. Again, we take the
quadmtic approximations in all variables, i.e., in the strains tll, t2b t3b
Kb K 2,T, t21,b t3l,1, 'Y,'Y,1 and the coordinates X2,X3. Thus we write
GI . G I =1 + 2tu + 2'Y,1 - 2X2(K2 - 2t21,t}

{ - 2X3(K3 - 2t3l,t} + t~l'
(1.15)
G I . G 2 = 2t2I + 'Y,2 + 2tllt2I - X3T ,
G I . G 3 = 2t3I + 'Y,3 + 2tllt3I + x2T.
As for the norm of the tangent to the reference line we take the approxi-
mation
We rewrite G 1 . G 1 in (1.15) as
G1 . G1 = 1- 2/\;2 - 2/\;3 + 2fll + 2')',1 c/> - 2X2(K2 - /\;2 - 2f21,t}
- 2X3(K3 - /\;3 - 2f31,t}.
Now let us introduce our final set of bending strains by setting

K2 :=K2 - /\;2 - 2f12,l,
(1.16) { K3 :=K3 - /\;3 - 2f13,l,

T:=T-T.
In order to evaluate Cij we recall the identities g1 . g2 = -X3T and g1 . g3 =
X2T. Then the quadratic approximation of Cij is given by
1
Cll =(1 + 2fll)fll - X2 K2 - X3 K3 + ,),,1 C/>,
1
(1.17) C12 =(1 + fll)f12 - 2(X3 T - ')'c/>,2 ),
1
C13 =(1 + fll)f13 + 2(X2 T + ')'c/>,3)'
It is important to note that the introduction of the new strains (1.16) is
not just a matter of notational convenience; in fact it has a physical inter-
pretation in that the quantities f21,1, f31,1 can be viewed as the entries of
the axial vector (0, f21,1, f31,t) describing the rotation of the cross section
into the plane spanned by e2, e3. Let us also note once again that the only
smallness assumption we have adopted up to this point is the consistent use
of quadratic approximations. It is customary within the engineering liter-
ature, however, to use two different degrees of approximation: a quadratic
approximation in all rotations, and a linear approximation in all strains fij'
This is what has come to be known as the theory of rods with infinitesimal
strains and moderate rotations. With our approximations, we obtain
(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,
2. The Equations of Motion

Let LlT(Xb X2, X3) denote the difference of the temperature T(XbX2' X3)
of the deformed state and the uniform temperature To of the reference state.
As the body is supposed to be very thin, we may take the temperature
difference LlT to satisfy
LlT = (h(xd + X282(XI) + X383(XI).
In addition, let E denote Young's modulus and G the shear modulus of the
beam at the point Xl on the reference curve. We have assumed from the
beginning that the cross section A(xd is doubly symmetric with respect
to the lines X2 = 0, X3 = o.
We introduce the moments of the cross section as follows:
122 := Ii x~dA, 133 := Ii x~dA, 1:= 122 + 133 ,
J: = Ii {(,2 - X3)2 + (,3 + X2)2}dA

= I + Ii {-,2 X3 + ,3 X2}dA,
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
(2.1) ~ (Xl, X2,

E2
X3 ) = 2"c + 2G (2 2 -
cl2 + C13) acuLlT.
u
Note that the trace of the strain tensor, which is the coefficient of the
temperature difference in (2.1), is just cu. The total strain energy then is
(2.2) WI:= IIIn {~ C~l + 2G(C~1 + C~l) - acuLlT}dxldx2dx3,
This is similar to the corresponding expression typical for 3-d St.Venant

-Kirchhoff materials, namely
WI = IIIn {~(tr(c2 + /-Ltr (c2 ) - (3'\ + 2/-L)iitr(c)LlT}dxldx2dx3,
where '\, /-L are the Lame moduli and tr() indicated the trace of a matrix.
2. EQUATIONS OF MOTION 61
We may express (2.2) in terms of our reference strains eij as follows. We

have
C~l =e~l - 2X2~2ell - 2X3~3ell + 2,,1ell + X~~~ + 2X3X2~3~2
- 2X2,,1~2 + '~122x3,,1~3 + X~~~,
2 -2 1 2 2 _
c21 =c21 + 4' ,2 + ,,2 c21 -
1 - -_
"2X3,,2T - X3 TC 21 + 41 X32-2
T ,
2 -2 1 3 2 _ 1 - __ 1 2-2
c31 =c31 + 4' ,2 - ,,3 c31 + "2 X2'Y,3 T - X2 Tc 31 + 4X2T .
In order to evaluate (2.2), we use the fact that due to the assumption of dou-
ble symmetry of the cross section and due to the properties of the harmonic
function related to the Prandtl-stress function as indicated above, the
following integrals vanish: IIA
dX2dx3, IIA
X2dx2dx3, X3dx2dx3, IIA
IIA ,2dx2dx3, IIA
,3dx2dx3. The vanishing of the first three integrals is
easily seen to follow from the symmetry properties of and the double
symmetry of the cross section, which implies that the domain A(X1) is a
succession of four reflections of a domain given in the first quadrant. As for
the other two integrals, we apply Green's theorem and use the symmetry
properties of the normal. The total strain (or potential) energy is then
calculated as
[t ({EA -2 -2 -2 EI22 _
(2.3) WI = 10 UdX1 = 10 TCu + 2GA(C12 + C13) + -2-"'22
EI33 -2 GI -2 Er 2 _ G 2
+ -2-"'3 + 2T + T"l -G(I - J)rT + "2(1 - J)r
- oAeu(h + oI22~2(h + oI33~303}dx.
Let us now focus on the stress-strain relations. Let Sij denote the entries
of the second Piola-Kirchhoff stress tensor S. Under the standing assump-
tion that the body is very thin, we have S22 = S33 = S23 = O. By the
symmetry of this stress tensor, the only three remaining entries are to be
given in terms of the strains and the temperature. As for the stress-strain
constitutive relation we require
Su = Ecu - o!:l.T = E(eu - X2~2 - X3~3 + ,,1 ) - o!:l.T
(2.4) { 812 = 2Gc12 = G(2e12 - X3 T + ,,2 )
813 = 2Gc13 = G(2e13 + X2 T + ,,3 ).
This is the assumption of a St. Venant-Kirchhoff material, see Ciarlet
[18]. Note that in taking the St.Venant strain tensor e we keep quadratic
terms in the reference strains. Dropping those reduces (2.4) to the classical
Hooke's law. By definition, the deformation matrix is G := (GI, G 2 , G 2)T,
where the Gi's are taken to be columns. Hence,
au 1 au 1 au
811 = a t 11' 812 = "2 a t 21' S13 = "2 at31 '
as they should be. If we apply the stress matrix 8 := (Sij) to G T we

obtain the transpose of the first Piola-Kirchhoff stress tensor T. Upon
introducing the Piola-Kirchhoff stress vector by
t = (Sl1 G 1 + S12G2 + S13G3),

we can express T as
T = (t, S12Gb S13Gt).
We can also introduce the second Piola-Kirchhoff stress vector s = 81,(.),
the first row of the second Piola-Kirchhoff matrix. As we have T = G8,
we can write
(2.5)
The resultant force, F, and couple, M, on a cross section are given by
(2.6) F:= IL tdA,
(2.7) M:= IL(R-Ito) xtdA.
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
P:= !L po(fo - R,tt)dA + Is todS,
(2.8) C:= IL [co + (R - Ito)] x Po(fo - R,tt)dA
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
product of matrices, i.e. A: B = tr(ATB), and by V'8R the Jacobian of

the virtual displacement
8R = 8W + x28E2 + x38E3 + 8,el := 8W + 8R,

where R := R - Ro and 8Ro = 8W. Then the principle of virtual work
can be expressed as
(2.9) !!lQ T : V'8llilx = !!lQ f 8llilx + flQ g. 8llila
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
= [Q+Ll {Jr [ (-PoR,tt) .8llilA}dx l

JQ-Ll JA
+ Jr [ t . 8llilA - Jr [ t . 8llilA.
JA(Q+Ll) JA(Q-Ll)
Following Wempner [108] (see also Hodges [22]), we introduce an aver-

age external force It, an avemge resultant force D and a surface action ~
reflecting the warping of the cross sections, as follows:
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).
After some standard calculations, using these approximations, we obtain

from the principle of virtual work the following balance laws, formulated
for the whole beam:
F,l +P = 0,
(2.10) { M,l +el x F + C = 0,
DI,1 - { + II = 0.
Associated boundary conditions are
(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)
together with the boundary conditions, e.g.

i - i
(2.13) ~Tlo =~Tlo'
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,
F,l = (FI,1 - /'i,2 F2 - /'i,3F3)el + (F2,1 + /'i,2FI - rF3)e2

+ (F3,1 + /'i,3FI + rF2)e2,
Ultimately we want to express rotations in terms of the angles {)i rather

than through the auxiliary angles 8 i . We can accomplish this by our as-
sumption that the shear is small compared to the rotation, thus expressions
like cabFa, 8aFa, where a, b = 2,3, are to be neglected. Therefore, we use
the approximation
This helps us to express the balance equations more explicitly as
Fl,l - K,2F2 - K,3F3 + PI = 0,
F 2,l + K,2 F l - rF3 + P2 = 0,

F 3,l + K,3 F l + rF2 + P3 = 0,
(2.14) Ml,l - K,2M2 - K,3M3 + Cl = 0,
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.
According to our assumptions, we obtain
Fi = ffA 8li dA,

Ml = ffA (S13 X2 - S12 X3)dA,
(2.15)
M2 = ffA 811 X3dA , M3 = - ffA Sl1 X2dA
Dl = ffA 8114>dA, ~= ffA (8124>,2 + 8l34>,3)dA.
We may then use the stress-strain constitutive equations (2.4) in (2.15) to
obtain
Fl = EAtl1 - 0.01,
F2 = 2GAt12,
F3 = 2GAe13,
Ml = GU - G(I - J)-y,
(2.16)
M2 = -EI33K3 - 0.03,
M3 = EI22K2 + 0.02,
Dl = Er"b
~ = G(I - J)(, - r).
Relations (2.16) constitute a set of eight functions of the eight strains

el1,t12,t13,K2, K3,r"",1. The components of P and C may be approxi-
mated as
(2.17)
whereas the average force II is

(2.18)
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
We remark that the first three equations of (2.14) are determined by

the axial and in-plane forces Fi , and the second three equations describe
twist (M1) about the e1-axis and bending M 2, M3 in the e1 e3 and e1 e2
planes, respectively. The last equation of (2.14) accounts for warping of
the cross section. It is apparent that the initial curvatures /'i,i and the
twist T introduce a linear coupling between stretching, bending, twisting
and the temperature. It is also apparent that bending and twisting are
nonlinearly coupled to shear. Furthermore, under the Euler-Bernoulli-
Kirchhoff hypothesis, namely, that the cross section remains perpendicular
to the centerline, the shear strains 12, 13 are then equal to zero, implying
that F2 , F3 are identically zero.
2.1. Some Remarks on Warping and Torsion. In the absence of

the average body force It and in view of (2.16), the last equation of (2.14)
reduces to the equation
(Er"l ),1 -G(I - J)(r - f) =0

or, equivalently, to
(2.19) G(I - J)f = G(I - J)r - (Er"l ),1 .
If the warping strain ,,1 is neglected and if the beam is initially straight
and untwisted, then (2.19) implies, = 19 1,1, and hence M1 = GJ19 1,1, or
M1 = GI19 1,1 if also <p == O. If the torsion is considered to be nonuniform
to the effect that the warping strain ,,1 is present, the right hand side of
(2.19) can be used as an approximation of the torsion if, is still taken to
be equal to 191,1, Note that in this case the torsion is given by a third order
differential expression in the variable 191 (see Kane et al [41]).
Before writing down the differential equations in some examples of par-
ticular interest, let us reflect on the potential energy (2.3). It is interesting
to observe that in the case where no warping is present (in particular if
r = 0, J = I, , = f), the only term representing torsion is ~GIf2, whereas
if warping is present and, is set equal to 19 1,1, torsion is represented by
~GJ19i,l + ~Er19i,ll (if the beam is initially straight and untwisted). As
has been indicated above, it is in this case also possible to take (2.19) as
an approximation for the torsion and to account for torsion through
(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
In the global inertia frame the position vector R is given by
R = D(t)R + y(t).
The velocity R,t in the inertia frame is given by
R,t = D(t) (D(t?i!(t)R + D(t)R,t) + y,

where the dot denotes the derivative of functions depending only on the
time parameter t. In the given case of planar rotations the matrix DTi!
has a simple representation,
where
J:= (00-1)
0 0
1 0
0
0
.
It is convenient to introduce the notation, adopted from [6],
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:
(3.1) P = -moTI(Ro,tt + 20JRo - o2iRo + OJRo) + moy,
where we have used n,tt = TID2R, R = Ro + X2E2 + X3E3 and the

assumption of double symmetry of the cross section. Also, we have used
the notation i = diag(l, 0,1). As for the couples we obtain
Ji
(3.2)
C = Po('R- - 'R-o) x (-'R-,tt)dA
= -Po Ji ('R- - 'R-o) x ('R-,tt - 'R-o,tt)dA
= -Po Ji X3TIE3 x (x3TI (E3,tt + 20JE3,t - o2iE3 + OJE3)dA
= -PoI33TI(E3 x E 3,tt + 20E3 X JE3,t + OE3 x JE3)

Again, we base our calculations on the quadratic approximation of rota-
tions. As we are working in the el e3 plane, we have
1 2
E3 = 'l?2 e l + (1 - i'l?2)e3,
E3,t = 'l?2,te l - 'l?2'1?2,te3,
E3,tt = 'l?2,tte l - ('I?~,t + 'l?2'1?2,tt)e3.
Taking the cross products in (3.2) and retaining only quadratic terms, we
obtain
(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
respect to the rotated frame can multiplied through by the transpose of D.

We obtain
F,l = moD2R - moDTy,

(3.4) {
M,l + el x F = m oI 33 {'i92,tt - O}e2.
This set of equations has to be complemented by the balance of linear and

total angular momentum
mbY + fffn PO'R-,ttdXldx2dx3 = 0,

(3.5) {
hO - fffn Po'R- x 'R-,ttdXldx2dx3 = o.
We can express the integral in the second equation of (3.5) as
(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
~D(11 In PoRo x D 2Ro + PoRo x DT ydXI dX2dx3
-1 mOI33dx IO+ 1 133'i92,ttdxI.
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
(3.7) {mb + moi)y + moD 1 D 2 Rodx = o.
Therefore, (3.4) and (3.6) can be written as

.. . 2
FI,1 - "'3F3 = mo(WI,tt - OW3 - 20W3,t - 0 (WI + x))
-mo(cos(O)jh + sin{O)Y3),
.. . 2 ..
(3.8) F3,1 + "'3FI = mo(W3,tt + OWl + 20WI,t - 0 W3 + xO)
-moe - sin{O)YI + COS(O)ii3),
M 2,1 - rhFI - F3 = Po 133 {-iJ2,tt + O},
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
where i = I~ 133modxl. These equations are precisely the ones obtained

in [5] for a uniform rod. We can also easily write down the kinetic energy
which is also found to be exactly as in Ballieul and Levi [6]:
1 2 1 ( 2 2
(3.10) T = "2hO +"2 Jo I('I92,t + 0 )dXl
+ ~mbllYll2 + ~ 1 PoliTIDRo + YII 2dx l
3.1. Dynamic Stiffening. In this subsection we consider an initially

straight beam (11:3 = 0) rotating about the origin in the ele3 plane (as
above). We neglect the inertia forces with respect to stretching and re-
tain as the only nonlinearity in the kinematics the terms with factor 02
(centrifugal forces). We can then integrate the first equation of (3.8) with
respect to x and obtain
Fl(X) = ~oiP(f? - X2),
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.
,
In this equation it becomes apparent that the term involving -(mo82 (2 -

X 2 )W3 ,t},1 introduces a dynamic stiffening. The last equation can be found
for instance in Ballieul and Levi [5] and other papers. In most of the engi-
neering papers, such a stiffening term is introduced by the way of additional
forces, if it is considered at all. The point here is that we obtain this sig-
nificant feature as a result of the infinitesimal stmin/large displacement or
the small stmin/finite rotation theory of rods.
4. Straight, Untwisted, Nonshearable Nonlinear 3-d Beams

As we have only one spatial variable Xl to take into account, we will
consequently replace the subscript ,1 with a prime in order to indicate
spatial derivation, and replace ,t by a superimposed dot. We set "'2 = "'3 =
T = O. By the Euler-Bernoulli-Kirchhoff hypothesis, e12 = e13 = O. In
addition, we neglect the nonlinearities in the shear stmins. We have
'!93 = W~, '!92 = -W~,

-
"'2 = W"2 , "'3
- = W"3' T- = 'VI
.af
The reference strains are

1
ell = W{ + 2W~? + (W~)2), e12 = e13 = 0
As for the body forces and couples, we find
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
[Longitudinal motion]
(4.1 ) .. = [A('
mOWI E WI +"21 (W2')2 +"21 (')2)]'
W3 - 0'.0,1.
[Vertical motion]
(4.2) mOW2 - [PoJ22W~1' + [EJ22W~']" + aO~ =

[(EA(W{ + ~(W~)2 + ~(W~)2) - aOl)W~l'.
[Lateral motion]
(4.3) moW3 - [PoJ33W~1' + [EJ33W~']" + aO~ =

[(EA(W{ + ~(W~)2 + ~(W~)2) - aOl)W~]'.
[Torsional motion]
(4.4) PoJJ I = [Glt?~ - G(J - Jh]'.
[Warping]
(4.5) Pori = [Er,,], - G(J - J)(-y - t?D.

The equation of heat conduction reduces to
PoCv{OI + 02X2 + 03X3} = O~ + X20~ + X30~

- Toa{ll - x2k2 - X3k3}.
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]
Or - aTo~t {Wi + ~((W~)2 + (W~)2)),

{ POCv02
PoCV~1 =
(4.6) = O~ + aTo W~',

PoCv03 = og + aTo W~'.
These equations have to be complemented by appropriate boundary con-
ditions. As a typical example, we choose a beam which is clamped and
thermally insulated at the boundary x = 0, and to which bending mo-
ments and shear forces as well as heat are applied at the end x = [.
Boundary conditions
[Geometric boundary conditions]
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]
EA(W{ + ~((W~)2 + (Wa)2))(f) - a01(f) = II,

(4.9) { -EI33Wa'(f) - a03(f) = m2,
EI22W~'(f) + a02(f) = m3,
(4.10) - [EI33W~'l'(f) - aO~(f)

+ [EA(W: + ~((W~)2 + (W~)2)) - a01]W~(f)
+ PoI33W~(f) = 12,
(4.11) [EI22W~'l'(f) + aO~(f)
- [EA(W: + ~((W~)2 + (W~)2)) - a01]W~(f)
- PoI22W~(f) = fa,
(4.12) GIt,}~(f) = ml, ,,'(f) = 0.
[Heat flux boundary conditions]
(4.13)
Initial conditions
Wi(t = 0,) = WiO (), Wi(t = 0,) = Wi1 (),

1?l(t = 0,) = 0, J1(t = 0,) = 0,
,,(t = 0, .) = 0, i'(t = 0, .) = 0,
Oi(t = 0,) = OiO(), i = 1,2,3.
As the body is supposed to be initially straight and untwisted, we have set
the torsion and warping distributed along the beam at the time t = to
be equal zero. A twist can be exerted, however, at the free end of beam

(i.e. through m1).
In the foregoing, we have assumed that the material is homogeneous

to the effect that the mass, the Young and shear moduli, as well as the
thermal coefficient Q and the reference temperature To can be treated as
constant. The cross section, however, is not constant in Xl, and so neither
are the moments h It is clear that the general inhomogeneous material
can be included without any extra effort, as long as we have isotropy. It
is important to note that the isothermal homogeneous process (4.1)-{4.5)
with (}i set equal to zero conserves the total energy U + T, where in this
case
(4.14)
The temperature dependent system is easily seen to be dissipative. In fact,

in the temperature dependent case, we may work with T as in (4.14), but
with U in (4.14) replaced with
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.
4.1. Approximation-Generalizations. When the dimensions in the

X2, X3 directions are very small, the heat conduction with respect to these
directions may be represented by
PocJIt = Or - aTO! {W:,+ ~((W;)' + (W~)')},

(4.15) { POC,,02 = -J.L202 + OTOW2 ,
POC,,03 = -J.L3 03 + oTOW3
II
Then the second and the third equations can be solved by variation of
parameters to give
Oi{t) =oTo jt e-I';(t-s)Wns)ds =: jt ai{t _ s)W:'{s)ds, i = 2,3.

POc" -00 -00
If we use this in (4.1)-{4.4), and neglect warping we obtain

(4.16) moWl = [EA{W{ + ~{W~)2 + ~(W~)2)]' - oO~.

[Vertical motion]
(4.17) moW2 - poIW~' + [EI22W~'l" + a J~ a2{t - s)W~"'{s)ds

= [{EA{W{ + ~{W~)2 + ~(W~)2) - OOl)W~]'.
[Lateral motion]
(4.18) moW3 - PoIW~' + [EI33W~']" + a J~ a3(t - s)W~"'{s)ds

= [{EA(W{ + ~(W~)2 + ~(W~)2) - oOl)W~l'.
[Torsional motion]
(4.19)
The equation of heat conduction which with we are left is
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
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.
5. Straight, Untwisted Shearable Linear 3-d Beams

We also do not consider warping and thermal coupling. We obtain
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
moW1 = [EAW1l' [Longitudinal motion]

mOW2 = [GA(W~ -1?3)]' [Lateral motion]
mOW3 = [GA(W3 + 1?2]' [Vertical motion]
(5.1)
POli91 = [GI1?il' [Torsional motion]
PoI33 i9 2 = [EIa31?~]' - GA(W~ + 1?2) [Shear around e2]
poI22i93 = [EI221?;]' + GA(W~ -1?3) [Shear around e3]
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
As for the boundary conditions, we consider a beam clamped at x = 0

and free at x = I and obtain:
Boundary conditions
Wi(O) = 0, 1?i(O) = 0, i = 1,2,3.

6. SHEARABLE NONLINEAR 2-D BEAMS 77
[Dynamic boundary conditions]
EAW{(i) = ft, GA(W~ - 'l?3)(i) = 12, GA(W~ + 'l?2)(i) = /3,

GI'I?~ (i) = m1, EI33'1?~(i) = m2, EI22'1?~(i) = m3.
We also have initial conditions as usual:

Initial conditions
Wi(t = 0,,) = WiO (-), Wi(t = 0,,) = Wi1 (),
'l?i(t=O,) =0, Ji(t=O,.)=O,
i = 1,2,3.
The approximations and generalizations are as in the previous case.
6. Shearable Nonlinear 2-d Beams with Curvature

Here we set W2 , T, 11:2, 'l?3, 'l?1 equal to zero. Since there is no torsion,
obviously neither is there warping. We have
K,2 = 0, K,3 = -'I?~,

T = 0,
1
u = W{ - 1I:3W3 + 2'1?~, 12 = 0,
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
..
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]
PoI33'1..?2 = Eh3'1?2" 1 2) - 0:01]

[Eh (W , - 11:3 W3 + 2'1?2
- 'l?2 1
- Gh(iJ 2 + W~ + 1I:3Wd - 0:0;
[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
(6.2) Wi(O) = 0, i = 1,3, t?2(0) = o.

[Insulated boundary conditions]
(6.3) OHo) = 0, i = 1,3.
[Dynamical boundary conditions]
Eh[W{ - ~3W3 + ~t?~](f) - aOl(f) = h,

Gh[t?2 + W~ + ~3Wl](f) = 12,
EI33t?~(f) - a03(f) -'I?2[Eh(W{ - ~3W3 + ~'I?~) - aOI](f) = mI.
[Heat flux boundary conditions]
(6.4)
Initial conditions
Wi(t=O,')=WiO ('), Wi(t=0,)=Wi1 (), i=1,3,

'l?2(t = 0,) = 'l?20, {)2(t = 0,) = 'l?21.
Oi(t = 0,) = OiO('), i = 1,3.

From this seemingly complicated system very interesting specializations
can be obtained. In particular, the isothermal linear system is exactly
the system obtained by Bresse [10] (1856), which is, in fact, more general
than the Timoshenko system (this has been pointed out by J. Schmidt in
[97]); for this reason we should in the future call the latter the Bresse-
Timoshenko system. We should also mention that the equations of balance
for precurved and pretwisted beams have been discussed in Love [63] (also
before Timoshenko), who traced the literature back to Kirchhoff in 1859.
This once again shows that it might be more appropriate not to attach
names to the various beam models to begin with, as many of these equations
seem to have been rediscovered repeatedly. Again, let us note that the
6. SHEARABLE NONLINEAR 2-D BEAMS 79
homogeneous isothermal model conserves the total energy U + T, where

now
T = 11{~o (W[ + Wi) + P~I Jndx.
The temperature dependent system is then easily seen to be dissipative.
6.1. Approximation-Generalizations. We apply the same consid-

eration as in the preceding section to the linear equations, and replace the
heat equations (6.1) with
[Heat flux]
(6.5) { POCV~I = ()~ - aTott {:v{ - K.3 W3},

POCv ()3 = -jt()3 - aTo'l?~,
and integrate the second of these equations. Then we obtain the system of
integro--partial differential equations
(6.6) mOWI = [Eh(W{ - K.3W3) - a()ll'
[Vertical motion]
(6.7) mOW3 = [Gh('I?2 + W~ + K.3WI)1'
[Shear motion]
(6.8) POh3J2 = [EI33'1?~1' - Gh('I?2 + W~ + K.3Wd

+ a 2 jt e-J.l(t-s)[K.3 To J~l'(s)ds.
-00 POCv
The same remarks as in the preceding section apply to this system. Systems
of this type do not seem to have appeared in the literature. They offer
the opportunity of introducing very subtle damping mechanisms into the
model.
7. A List of Beam Models

For easier reference, we represent the beam models obtained so far in
form of a compact list. We use the same labelling as in the text. We restrict
ourselves to display the equations of motion.
Initially straight and untwisted, nonshearable, nonlinear beams

.. = [EA (WI, + 2"1 (W2')2 + 2"1 (')2)]'

mOWI W3 - 0:0,1 ,
[Vertical motion]
mo W2 - [POI22 W~l' + [EI22 W~']" + 0:0~
= [(EA(W{ + ~(W~)2 + ~(W~)2) - o:OdW~l'.

[Lateral motion]
mo W3 - [POI33 W~l' + [EI33 W~']" + 0:0~
= [(EA(W{ + ~(W~)2 + ~(W~)2) - o:Ol)W~l'.

[Torsional motion]
POIi9 1 = [GI'!?~ - G(I - Jhl'

[Warping]
pori' = [Er'Y')' - G(I - J)('Y - '!?D
[Heat conduction]
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
POC,,03 = O~ + o:To W~'.
Planar isothermal approximation

[Lateral motion)
mOW3 - [pOI33W~)' + [EI33W~')" = [(EA(W{ + ~(W~)2))W~)'

A reduction of model to transverse motion
7. A LIST OF BEAM MODELS 81
We replace the inertial force with respect to the longitudinal motion by

a load h, integrate the first equation and insert the result into the second
equation of 3.2. We define F = - J:
h(s)ds. In the two dimensional
analog, this procedure leads to the Airy stress function and hence to the
von Karman system. In the one-dimensional case we just get an axial force:
moW3 - [poI33W~]' + [EI33W~'l" = [FW~]'.
The Rayleigh beam model

We neglect the longitudinal motion entirely:
mo W3 - [Po 133 W~]' + [EI33 W~']" = O.
The Euler-Bernoulli beam model

We neglect the rotatory inertia of the cross sections:
moW3 + [EI33W~']" = O.
Initially straight and untwisted linear shearable beam
moWI = [EAW1]' [Longitudinal motion]

moW2 = [GA(W~ - '193 )]' [Lateral motion]
moW3 = [GA(W3 + '192 )]' [Vertical motion]
POIt?1 = [GI'I9i]' [Torsional motion]
PoI33t?2 = [EI33'19~]' - GA(W~ + '192), [Shear around e2]
PoI22 i'J 3 = [EI22 '19al' + GA(W2 - '193) [Shear around e3]
Nonlinear planar shearable beams with initial curvature

.. , 1 2 ,
mOWI = [Eh(WI -1I:3W3 + 2''192 ) - a(h]
- 1I:3Gh('I92 + W~ + 1I:3WI) + a[1I:383]'
[Vertical motion]
moW3 = [Gh('I92 + W~ + 1I:3WI)]' + 11:3 [Eh(W; - 1I:3W3 + ~'I9~) - a81]
[Shear motion]
Po 133 '1..9 2 = EI33'192II - '19 2[

Eh(WI
'
- 11:3 W3 + 2'1'1922) - a8 1]
- Gh('I92 + W~ + 1I:3WI) - aO~
[Heat flux]
. "
POC v Ol = 01 -
() { ,
aTo 8t Wi - "'3 W3 + 2"'1 I?22} ,
POCj)3 = O~ - aTOt?;.
The Bresse system

[Vertical motion]
[Shear motion]
The Timoshenko system

[Vertical 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).
Damping. Damping can be introduced in all of the models as outlined

in the text. Possible sources of (internal) damping are:
thermal damping
viscous damping
viscoelastic damping
thermo-viscoelastic damping
structural or fractional power-type damping.
8. NETWORKS OF BEAMS 83
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.
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.,
(8.3) {Ji(N) = {)i(N), \fi,j E I(N).
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.
EXAMPLE 8.1. We confine ourselves to the illustration of a 3-dimensional

carpenter's square, a rectangular configuration of two beams. We neglect
the warping.
We have two initial triads ei, e~, e~, i = 1,2. However, in this special
situation we easily see that we may choose
This implies that the first joint condition (8.1) can be written as
wi = -wi, wi = Wi, wi = W[.

The temperature conditions (8.2) are independent of the triads. Accord-
ingly, (8.3) is then
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)
{h = -(W~ + "'3 WI + rW2),

{}3 = (W~ + "'2W2 - rW3).
Hence, (8.3) becomes
{}t = -Wi)' + ",~wi - r 2wi),

(wI)' + "'~ wi + rl wi = (Wi)' + ",~wl + r 2wi,
(Wi)' + ",~wi - riwi = {}~.
The interpretation of these conditions is particularly simple if the beams
are initially straight ("'~ = ri = 0). Then it is apparent that a twist {}i
of one beam causes a vertical strain -(Wi)' on the other, whereas the
orthogonality for the tangents is preserved in the 1-3 plane throughout the
deformation.
8.1.2. Pinned Joints. A joint is called a pinned joint if no bending waves

are transmitted, i.e., if the bending moment of each member at the joint
is equal to zero, or prescribed by a moment applied from the outside. In
addition, the sum of all shear forces has to be zero, or equal to a driving
force. This means that the members, individually, may perform a moderate
rotation about the joint. Joints of this type are typical for truss structures.
The geometrical or compatibility conditions are now (8.1), (8.2), but not
(8.3).
8.2. Dynamic Joint Conditions.
8.2.1. Rigid Joints. Dynamic joint conditions can obtained by balance

of force and momentum, just as in (2.14). Then the geometric node con-
ditions (8.1), (8.2), (8.3) serve as compatibility conditions. The dynamic
node conditions can also be obtained via the variational formulation of
the boundary value problem associated with the structure. Then the test
functions have to satisfy the geometric joint conditions (8.1), (8.2), (8.3),
in addition to the boundary conditions at the simple nodes (where only
one beams ends). We assume that all multiple nodes (joints) are rigid.
We give a discussion only in the case of the linearized dynamics. The
nonlinear node conditions are far from being completely understood in the
general case. They are, however, fully understood in the context of nonlin-
ear Euler-Bernoulli beam networks, but for the sake of brevity we do not
elaborate on this here.
In order to be able to assign to a typical element an outward pointing

normal at a multiple node N, we introduce the notation
In words, this means that if beam #i ends in the node at x = 0, then we

assign -1 to that end, and we assign +1 otherwise. This is, essentially, the
notation used in the paper by Wittenburg [109], where networks of rigid
bodies have been discussed.
We will give the dynamic node conditions in the following two exemplaric
cases:
The initially straight untwisted Euler-Bernoulli beam in JR3;
The initially curved Bresse beam.
In addition, we will then reduce the first class to planar motion, and the
second to the Timoshenko system.
Dynamic node condztions for a network of initzally straight and untwzsted
linear, thermoelastic Euler-Bernoulli beams. The geometrical node condi-
tions are (8.1), (8.3) with 19; = -(WaY, 193 = (W2)'. With the above
notation we obtain the following condition on the sum of the shear forces
at N:
(8.4) L Ct(N)[-((Etl t33(W3}")' + (iBa)ea - ((Etl t22 (W2)")' + atBi)ei
tEI
+ (EtAt(Wn' - atBDeil(N) = FN + mNW'(N),
where FN is a force applied to the joint N, and mN is the mass of this
joint. The bar on top of an index indicates that all values are the same for
all i according to a particular node condition, and that one of those values
is taken as a representative. The sum of moments at N satisfies
(8.5) I:>tEI(N)[GJt(l9i)'ei - (Etlt33(Wa)" + atBj)e;

-
+ (Etl t22 (W2)" + atB;)ea](N) = MN + INJt(N),
where MN are couples applied to the joint N, which has a moment of
inertia IN. In the case of planar motion in the 1-3 plane, conditions (8.4)
and (8.5) reduce to, respectively,
tEI
(8.7) - LCt(N)[Etlt33(W3)" + atBj](N) = MN - IN(WJ)'(N).

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.
EXAMPLE 8.2. In the example of the carpenter's square, with no forces

applied at the joint, which is also assumed to carry no extra mass, condition
(8.4) at N is
(EII133(Wj )")' + oJO! = E2A2(Wr)' - a20~,

(8.8) { (E21233 (Wj)")' + a20~ = EIAI(Wl)' - alO},
(EII122 (Wi )")' + a~ = -(E21222 (W?)")' + a20~.
The moment condition (8.5) at N may be written
Elh 22 (Wi)" + alO~ = -G212(U~)',

(8.9) { E21222 (W?)"+ a20~ = -Glh(UD',
E l h33(WI)" + alO! = -E21233 (Wj)" - a20~.
The reduction to the planar case is obvious, as is the reduction to the

isothermal situation.
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.
Dynamic node conditions for a planar network of linear initially curved

Bresse beams. Here we confine ourselves to the 1-3 plane, and proceed as
mentioned above. The balance law for the forces at N is
iEI
For the transmission of moments at N we obtain
(8.11) L ci(N)[Ei1i33 (1?;)' - (8;1~l(N) = MN + INJ~(N).

iEI
It is obvious that, for zero initial curvature, (8.10) reduces to
(8.12) L ci(N)((Eihi(Wf)' - o8;1i)el + GiAi(1?; + (W~)')e~)(N)

iEI
while (8.11) remains the same. These are the dynamic node conditions for
the Bresse-Timoshenko system.
EXAMPLE 8.3. Let us again consider the case of a carpenter's square as

above. The condition (8.10) on the sum of forces at N now reads
Elhl((Wl)' -l\;lWJ) - aIel = G2h2(1?~ + (Wl)' + I\;~Wf),

(8.13) {
E2h2((Wf)' -1\;~Wl) - a2e~ = -Glhl(1?~ + (WJ), + I\;lwl).
The condition (8.11) is already independent of the triads e l , e 2 and remains
unchanged. From this the corresponding conditions in the isothermal situ-
ation, and also in the uncurved case, are easily obtained.
8.2.2. Pinned Joints. As mentioned above, in the case of pinned joints

we retain (8.1), (8.2), but we do not impose (8.3) as geometrical or compat-
ibility conditions. Also, for the Euler-Bernoulli system, the sum of shear
forces at the node has to satisfy (8.4) or (8.6) in the 3-d or 2-d case, re-
spectively. The conditions on the sum of moments, however, is replaced by
the much simpler condition
in the 3-d case, or
(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
9. Rotating Two-link Flexible Nonlinear Shearable Beams

We consider two nonlinear shearable beams which are coupled through
a revolutive joint (pinned joint) with mass m r 2 and moment f r2 . This
aggregate is attached to a cylinder of mass mrl and moment f rl . For
simplicity we assume that the radius of the cylinder is zero (the more
general case is also easily dealt with). Both the cylinder and the attached
beams are then supposed to move rigidly in the el e3 plane, i.e. there is a
translation YI and a rotation III of the entire system. In addition we allow
the second beam to rotate about the revolutive joint, the rotation being
ll2. We assume, for simplicity, that the beams are uniform, and neglect
warping and thermal effects as well as damping effects. All these features
can be handled with some extra computational work.
In the inertial frame we have
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
The strain energy can be written as
We derive the equations of motion and the dynamic node-and boundary

conditions by the requirement that the total energy be conserved, i.e.
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
Tie. By straightforward calculations we find
We require that
R 1(0) = 0, ~~2 (2) = 0,

uC2
0:1(0) = 0, 0:;(2) = 0,
n 1W 1(1) = n2W2(0).
Hence,
R1(0) = 0, 0: 1(0) = 0
and
Using these expressions and the boundary conditions in E = L~=l (Ti+Ui),

we can express the nodal terms as
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
equations governing the motion of the system. Note that Y2 = II 1r 1(1).
. 1ll .
(Ir1 + h1)fJ1 +
o
h0:1 dx =< W 1(1),J III II2 a- (0) >
T T aU2
C2
-11 < R1,JT(D~R1 +IIfih > dx,
2R lIT.. aU1
Po D 1 1 + 1 Y1 = aR1 '
2 T aU2
POD2R 2 + II2 II1r 1(1) = aR2
The boundary and node conditions are given by
R 1(0) = 0, aa~2 (2) = 0,

c2
0:1(0) = 0, 0:~(2) = 0,
II1W1(t} = II 2W 2(0),
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
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
A General Hyperbolic Model for Networks

of One Dimensional Elements
In this chapter we study a general linear, hyperbolic model for vibrating

networks of one-dimensional elements. This model accommodates not only
the elastic string networks of Chapter II, but also applies to networks of
beams governed by one or other of the linear, isothermal, hyperbolic beam
models of Chapter III and to mixed networks which contain both beams
and strings. The existence and regularity of solutions to the general net-
work system is established, along lines identical to those used in Section
4 of Chapter II. Energy estimates involving characteristics are proved and
then used to obtain the estimates from which exact controllability and sta-
bilizability follow for networks containing no closed loops, again following
the argument used in Chapter II.
1. The General Model

We begin by describing the network configuration and establishing some
notation which generalizes or supplements that of Chapter II. We suppose
as before that the network consists of n elements located in JR3 (or JR2 in
the case of planar models) indexed by i E I and parameterized by x E [0, liJ.
We suppose also that these elements are joined at some of the endpoints
in such a way that the resulting network is connected. The elements are
associated with curves which generally describe the location of a physical
object such as a string or of the centerline of a three dimensional object
such as a beam. Associated with each curve we have a function ri(x, t) :
[0, liJ x [0, TJ ~ JRP', which describes the displacements of the variables
describing the configuration of the i-th element. For string networks one
has Pi = 3 and ri(x, t) describes the displacement of the i-th string at
time t and at the point parameterized by the natural arc length x. More
generally, some components of ri may describe the spatial displacement
of the centerline while others may describe changes in "internal variables"
96 IV. A GENERAL HYPERBOLIC MODEL FOR NETWORKS
introduced into the model by a kinematic hypothesis. As in Chapter II we

let r denote the n-tuple of functions (ri)iEI.
In order to formulate geometric constraints on the network we associate
with each node qj x Pi matrices C ij of rank qj indexed by j E 3 and i E I j ,
where necessarily qj :::; Pi. These matrices pick out that "component" of ri
which is to be involved in the geometric condition at the j-th node. At the
multiple nodes of the network we then impose the geometric node condition
(1.1) for j E 3M, Cijri(xij, t) is the same for all i E I j .
This is a continuity condition on the configuration functions at the j-
th node which reflects the way in which the corresponding elements are
joined together. For a given type of element there are many possible joint
conditions. Sometimes, as will be illustrated in the next section, it may
happen that the node conditions (1.1) indeed involve only some compo-
nents of the configuration displacements (see the reference to pinned joints
in Section 2.2 below). We remark also that the multiple node condition as
we have formulated it is convenient in the most important applications but
a more general condition is sometimes more appropriate (see Remark 1.2
and Subsection 2.5). In this section we want to walk a delicate line between
generality and intelligibility; there are many physically reasonable combi-
nations of geometric and dynamic conditions which can be accommodated
within the general model at the cost of a proliferation of notation.
For each j E 3 we define 1)i r to be the value (common value in the
case of multiple nodes) of Cijri(xij). We now suppose that certain nodes
are clamped:
(1.2)
In case C ij is injective for each i E I j this latter condition is equivalent to
ri (Xij) = 0 and we say that the node is fully clamped; otherwise it is only
partially clamped. Other nodes may be subject to Dirichlet controls
(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
(1.4) K(t) = L"21 10[Ii IPie(x,

n
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
(1.6) W(t) = L iii fi(x, t) . ri(x, t) dx + L uj(t). 1)jr(t),

iEI 0 jE.:JN
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
C(r) = loT [K(t) + W(t) - U(t)] dt,

allowing only variations in r which respect the geometric conditions (4.1),
(4.2) and (4.3). First one uses, for each i, variations in ri having compact
support in (0, li). In a standard way one obtains the system of equations
(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)
In order to write down the dynamic conditions resulting from (4.8) we

let IIij denote orthogonal projection onto the the kernel of the transforma-
tion defined by C ij and IIh
be orthogonal projection onto the orthogonal
complement of that kernel. Denote by C~ the generalized inverse of C ij .
This is the Pi x qj matrix which satisfies
CijC~ = Ij and C~Cij = IIh,

where I j denotes the qj x qj identity matrix. The variations 8ri can now
be chosen so that
J: i( .. ) _ {C~'I/J(t) + 'l/Jij(t), for j E .:IN,

ur Xt)' t - . F
'l/Jij(t), for J E.J ,
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)
(1.10) N:r = 0, for all j E J and i EI J,
where
(1.11) NJ r = L [C~ltQ~ (r:x + Rzr') (x'J' t) for j E JM,

'EI,
and
(1.12) N:r = 'Jn'JQ~ (r:x + Rzr') (x'J' t) for j E JM and i E IJ"

The condition (1.10) simply expresses the fact that the component of each
r'(x tJ , t) belonging to the kernel of C'J is free.
REMARK 1.1. One could of course place further geometric (Dirichlet)

or dynamic (Neumann) conditions on the components which belong to the
kernel of C'J. For example, if we add the following expression to the work
functional (1.6)
where IN! is any subset of J, the condition (1.10) becomes
(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.
REMARK 1.2. A reasonable generalization of the multiple node condi-

tion (4.1) takes the form
(1.14) 1)1r =L DJ,rt(x'J' t) = 0,

tEI,
2. SOME SPECIAL CASES 99
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.
2. Some Special Cases

After briefly describing how the string networks of Chapter II fit the
general model we draw on the some of the linear, hyperbolic and isothermal
beam models of Chapter III, Section 7 to specify a variety of beam networks
as well as some simple networks involving both strings and beams.
2.1. String Networks. Here Pi = 3 for each i E I and each C ij is

set equal to the 3 x 3 identity matrix I. Furthermore Qi is the positive
definite square root of the matrix called Qi in Chapter II, Pi = "/piI
and ~ = Si = o. Everything reduces exactly to the model considered in
Chapter II.
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]).
2.2. Networks of Planar Timoshenko Beams. Such networks were

considered in [49] and the treatment closely parallels that of string net-
works. The equations, given in Section 7, Chapter III are obtained from
(1.7) if we make the following assignment of matrices to each element, for
the moment suppressing the index i:
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
continuity of the rotation angles iJ,2' These conditions can be expressed

in the form (1.1) if we set C'3 equal to the orthogonal matrix having as
successive columns e,l, e,3 and n. One then has II'3 = 0 and C~ = C!3'
The condition (1.10) becomes vacuous and (1.9) gives
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.
Here we must set
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
{ E~=l Wik:1 c~in~ide for.i E I j ,
Ek=l 'l?ikek comclde for Z E I j .

These can be expressed in the form (1.1) with C ij = diag(EtE i , EtEi ),
where Ei and E respectively denote the 3 x 3 matrices having as columns the
triad el, e~, e~ associated with the i-th beam or orthonormal basis vectors
el, e2, e2 chosen independently of i. In this case each C ij is invertible so
condition (1.10) is degenerate. One is left with the dynamic conditions
(1.9) which can be written out explicitly. We note that [C~lt = C ij . One
can decouple the first three components of N'j r from the second three.
Denote the 6 components of the control vectors u j by ut
with k = 1, ... ,6.
If one carefully rewrites the expression (1.11) for ,N'j, the conditions (1.9)
become
(2.7) L fij [EiAiWil,xel + GiAi(Wi2,x - 'l?i3)e~
iEIj
for j E .IN
for j E.JN '
and
(2.8) L fij [Gil i'l?i1,x ei + Eil i33 '1?i2,xe; + Eili22'1?i3,x)e~1

iEIj
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
on the (infinitesimal) rotations and it is an easy exercise to work out what

the dynamic conditions are in that case.
2.4. Networks of Initially Curved Bresse Beams.. These planar

beam models are treated much like the Timoshenko beams. We have
r = (WI, W 3, t?2),
p2 = diag{mo, mo, poI 33 ), Q2 = diag{Eh, Gh, EI33 ),
(2.9)
R= (:3o -;3 ~),
0 0
S= o.
One can use the same matrices C ij as before and obtain the following minor
variation of the conditions (2.3):
(2.10) L ij [Eihi{Wil,X - lI:i3Wi3)eil + Gihi{Wi3 ,x +

iEIj
The condition (2.4) remains unchanged.

2.5. Beams and Strings. As an example of a structure involving el-
ements of different natures we consider some ways in which strings and
beams can be joined together. This will also illustrate the need for the
generalization of the multiple node condition (1.1), which was indicated in
Remark 1.2.
It is useful to recall from Chapter III that in the planar Timoshenko
model the beam is thought of as occupying a narrow rectangle {xlel +X3e3 :
Xl E (O, I), x3 E (-h, h)}. According to the kinematic hypotheses the
displacement at the point corresponding to (XI, X3) is then given by
(2.11)
In setting up the expressions for potential and kinetic energies the variable
X3 is integrated away and enters the constants occurring in the Timo-
shenko equations, and if we set X = Xl we end up with the apparently
one dimensional model of this section. However, in modeling the joints
between Timoshenko beams and other objects it is helpful to think of the
displacements in terms of both variables X := Xl and X3.
We shall now consider a planar network consisting of two strings indexed
by 1 and 2 attached in various ways to one beam indexed by 3: since there
is only one beam we do not bother to use the index for the quantities
associated with it. For i = 1,2 the spaces Vi and Hi as well as their inner
products are chosen as in Chapter II and Subsection 2.1 above, while for
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
+ fa l 3 {[moWf + moWl + poI33t?~1

- [EhW1,x 2+ Gh(W3,x + fh)2 + EI33 iJ2,x 2]} dX] dt.
Now one needs to specify the nature of the joints, which will involve the
endpoints of the string and endfaces of the beam. We note first that in
all cases variations of ri inside (0, li) will lead to the equations (3.10) of
Chapter II for the two strings and the Timoshenko system of Chapter III,
Section 7 for the beam. General variations, for the moment not subject to
any geometric constraints, then lead one to the variational identity
(2.13) = iT ( L
o
2 [
i=l
[Qi r :x 6ri](x, t)
]X=lo
='
x 0
+ [[EhW1 ,.6Wl + Gh(Wa,. + ~2)olWa + EI"D 2 ,.6D2 i(x, t) [:~ ) dt.
We do not write down the various controls or free conditions at the

endpoints which are not joined to another element, where one has all the
possibilities discussed in the previous sections.
Consider first the situation where the two strings are attached respec-
tively to the two endpoints of the centerline of the beam. Here, if one takes
into account the form (2.11) of the beam displacement, one has the geo-
metric constraints
rl(O, t) = WI (0, t)el + W3 (0, t)e3,
(2.14) {
r2(0, t) = WI (l3, t)el + W3(l3, t)e3.
These conditions at the two multiple nodes located at the endpoints of
the beam are certainly of the form (1.1). One can use the general scheme
to write down explicitly the corresponding dynamic conditions (1.9) and
(1.10) but it is also instructive to derive these directly. Since the angular
variable iJ 2 is not involved in the geometric constraint, one can choose
6iJ 2 arbitrarily while letting the other variations vanish. One obtains from
(2.13) that
(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
and Mh = 0. Then, because

8rl(0, t) = 8Wl (0, t)el + 8W3(0, t)e3,
(2.13) reduces to
[-Qlr~x.el - EhWl,x](O, t)8Wl (0, t)

+ [-Qlr~x.e3 - Gh(W3,x + 1?2)](0, t)8W3(0, t) = 0.
One obtains the following two dynamic node conditions associated with the
junction between the first string and the beam:
{
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
Q28r2(0' t) . el - EhWl ,x(l3, t) = 0,

(2.17) {
Q2r~AO, t) . e3 - Gh[W3,x + 1?2](l3, t) = 0.
Next we consider the situation where the two strings are both attached to
the centerline of the beam at one end, which corresponds to the geometric
constraint
(2.18) rl(O, t) = r2(0, t) = Wl (0, t)el + W3(0, t)e3.

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
[QlrlAO, t) + Q2r2x(0, t)] . el + EhWl x(O, t) = 0,

{
(2.19)" ,
[Qlr;x(O, t) + Q2r~x(0, t)] . e3 + Gh[W3,x + 1?2](0, t) = 0.
Finally we consider a third variant which illustrates that the continuity
condition (1.1) is somewhat restrictive. Suppose that the two strings are
attached to the two extreme points of one endface of the beam. This leads
to the constraints
rl(O' t) = [Wl(O, t) + M 2(0, t)]el + W3(0, t)e3,
(2.20) {
r2(0, t) = [Wl(O, t) - h1?2(0, t)]el + W3(0, t)e3.
These two multiple node conditions cannot be written in the form (1.1) in
a way which satisfies the conditions on C ij but can be expressed in the
form discussed in Remark 1.2. Again we can proceed in an ad hoc way.
For the variations we must require
8rl(0' t) = [8Wl (0, t) + h81?2(0, t)]el + 8W3(0, t)e3,

{
8r2(0, t) = [8Wl (0, t) - h81?2(0, t)]el + 8W3(0, t)e3.
3. EXISTENCE AND REGULARlTY OF SOLUTIONS 105
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. Existence and Regularity of Solutions

In this section we shall sketch how the standard variational and semi-
group methods used in Section 4 of Chapter II to obtain existence and
regularity results for solutions of the string network system apply with
only very minor changes to the general system we have derived in Section
1 of this Chapter. This general system is
P~ri = [Q~(r:x + ~ri)] ,x

- R~Q~(r~x + ~ri) + Siri + fi for i E I,
ri(xii' t) coincide for j E .lM, i E Ii,
(3.1) 1)ir(t) = u i (t) for j E .lD, 1)ir(t) = 0 for j E .lc ,
}/ir(t) = ui(t) for j E .lN, }/ir(t) = 0 for j E .IF,
}/ir(t) = 0 for j E .l and i E Ii,
r(O) = ro, r(O) = roo
The governing equations can again be written in the form
(3.2) ~ ( ~ ) = (~ ~) ( ~ ) +( ~/p )
As state space for our system we take the Hilbert space 1t = V x H

where
n
H = IT L (0, li; lRPi )
2
i=l
and
V = {r E IT Hl(O, li; lRPi) {Ci!ri(Xi j ) coincidecfor j E JM, i E I j , }.

iEI 1)J r = 0 for j E J
On H we define the inner product
(3.4)
corresponding to kinetic energy. On V we define

(3.5)
(r, s)v = L ~ 10[Ii [Q~[r:x + ~ri]. [s:x + ~si] + [Siri . si] (x, t)] dx,
iEI
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
PROOF. The triangle property and the homogeneity property of II r IIv

are evident, so to verify that this is a norm one need only show that II r Ilv=
o implies r = O. This is obvious if each Si is positive definite. Otherwise
one concludes that rix + Rri = 0 for each i. If ri vanishes at one of the
endpoints of [0, Iil it' follows from a uniqueness theorem for solutions of
differential equations that ri vanishes identically. If we now start at a fully
clamped node and work our way from element to element along paths such
as those described in the alternative hypothesis, using the node condition
3. EXISTENCE AND REGULARITY OF SOLUTIONS 107
(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,
and so one gets the estimate
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.
The argument leading to well-posedness of our model now proceeds word

for word as in Section 3 of Chapter II, with only minor notational adjust-
ments:
The continuity condition becomes: the vectors Cijri (Xij, t) coin-
cide for j E JM and i E Ij;
the condition N1 r = 0 for j E J and i E I j needs to be stated
along with the other conditions;
IR3 needs to be replaced by IRPi or IRq, as appropriate.
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
Uj(t) -1)jro E C8"((O,T)iIRQj) for j E .lD,

(3.6) {
Uj(t) E C8"((O,T)iIRQj) for j E .IN.
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
ei(x, t) = ~ [IPiril2 + IQi(r:x + ~riW + Si ri . ri].

It is in fact also useful to introduce
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 m(x) is a given smooth multiplier function. Then
(3.8) ioT[e(x, t)m(x) ]Xx=o=I dt iT 11 e(x, t)m,x dxdt

=
0 0
1 [T [I
+ '210 10 [M1r. r + M2(r,x + Rr) . (r,x + Rr)
([ t=T
- mSr (r,x + Rr) ] dxdt + 10 mP 2r (r,x + Rr)L=o dx,
where
As a consequence
(3.9) 1 (0; 0, T) + 1 (I; 0, T) S C [I T

&(t) dt + &(0) + &(T)].
3. EXISTENCE AND REGULARITY OF SOLUTIONS 109
PROOF. Here we multiply the equation by m(r,x + Rr) and integrate

by parts. First we get
(3.10) 1T 11 p 2f . m(r,x + Rr) dxdt
= - 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,
where we have used the identity
mP2r r", = ~[mp2r. rl,x - ~[m,xp2 + m(p2),xlr. r.

To obtain the next identity we first note that
1
m[Q2(r,x + Rr)J,x(r,x + Rr) = 2" [mQ2(r,x + Rr) . (r,x + Rr)J,x
- ~[m,xQ2 - m(Q2),x](r,x + Rr) (r,x + Rr).
Then we integrate by parts with respect to x to get
(3.11) -1 11 T
[[Q2(r,x + Rr)J,x - RtQ2(r,x + Rr)
- Sr] . m(r,x + Rr) dxdt = -loT [~ IQ(r,x + Rr)12] ::: dt
+~ 1T 11 [[m,xQ2 - m(Q2),x + mRtQ2] (r,x + Rr) . (r,x + Rr)

- mSr (r,x + Rr)] dxdt.
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
The next step is to prove the a priori estimate analogous to that of

Lemma 4.3 of Chapter II. We restate the result.
LEMMA 3.3. r be a smooth solution of (3.1). Then

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
= :L :L Eij[Q~[r:x + ~riJ . r i] (Xij, r).

jE.:J iEIj
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
as well as the initial conditions
The solution satisfies the estimate
4. Energy Estimates for Hyperbolic Systems

In this section we derive the energy estimates which are needed to prove
results on controllability and stabilizability.
We consider a single system
which is associated with the energy density

1
(4.2) e(x, t) = "2 [IPr,tI 2 + IQ(r,x + Rr)12 + Sr r].
Here we find it convenient to use ",t" as an alternative to the"" notation
for partial derivatives with respect to t. For the purposes of this section
we shall suppose
P and Q are smooth, positive definite and symmetric p x p matrix
valued functions of x and tj
S is either a positive definite and symmetric p x p matrix valued
function of x and t or, alternatively, it is the zero matrixj
R is a p x p matrix valued functions of x and possibly, in case S is
positive definite, of t as well.
The dependence on t is not needed for the applications we have in mind
here but may be useful in other contexts and does not introduce additional
difficulties.
We shall need the following two identities
(4.3)
and
(4.4)
where
(4.5)
and
(4.6) Q2(r,r,t,r,x) = ~[p21,xr,t. r,t - ~[Q21,x(r,x + Rr) (r,x + Rr)

1
+ "2S,Xr. r- P r,t R,tr + (r,x + Rr) R Q (r,x + Rr)
2 t 2
+ (r,x + Rr) . Sr + Sr r,x - p 2r,t Rr,t.

Under the given hypotheses on the matrices it is easily verified that
(4.7) Iqil ~ Ce for i = 1,2.

We note that
Q2(r,x + Rr) r,t = Q(r,x + Rr) Qp-1[Pr,t] = !Mw. w,
(4.8) { p 2r,t . (r,x + Rr) = Pr,t PQ-l[Q(r,x + Rr)] = !Nw. w,
e = Hlwl2 + Sr r),
where
Prt )
w = ( Q(r,x -+- Rr)
and
M=(Q:_1 P~Q), N=(Q~lP P~-l).
One easily verifies that
(4.9) W,t = Mw,x + l(r, r,t, r,x),
where l(r, r,t, r,x) is equal to
-P- 1[RtQ 2(r,x + Rr) + Sr + P,tPr,t - Q,xQ(r,~ + Rr)] + P,tr,t) .

(
QR,tr + QRr,t + Q,t(r,x + Rr) - QP- P,xr,t
Then Il(r, r,t, r,x)i2 ~ Ceo
The estimates we wish to obtain involve the characteristics, which are
defined in terms of the eigenvalues of M. It is an easy exercise to show
that these eigenvalues have the form Wk where {w~ }~=l are the eigenvalues
of the positive definite matrix P-1Q2p-l and the Wk are positive. The
eigenvalues will in general be functions of x and t. We denote the largest
and the smallest of the Wk by wand ~ respectively. We let Cl(xo, to) denote
the characteristic curve x = x~(t) obtained by solving the system
dx
(4.10) dt = Wk(X, t), x(to) = Xo
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
~(t). Note that the characteristic curves can also be described in terms
of graphs t = t~(x) obtained by solving
(4.11) dt = 1 ,t(xo) = to.

dx Wk(X, t)
We shall obtain estimates by using the divergence theorem over certain
regions in the x, t-plane bounded by characteristics and time-like or space-
like intervals. Let us recall the notation
We can formulate and prove various estimates in terms of these energy

integrals. The first is a very familiar hyperbolic energy estimate.
LEMMA 4.1. Let r(x,t) E W 2((0,1) x (O,T);lRP ) be a solution 0/(4.1).

Suppose that 0::; Xl ::; X2 ::; 1. Then i/x+(t) and x_(t) correspond to the
chamcteristics C+(Xl, 0) and C-(X2' 0), respectively, one has, as long as
x+(t) < x_(t),
(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
contributions along the characteristic curves and obtain
E(t;x+(t),:L(t)) :SE(0;Xl,X2)+ 11
.6(t)
ql(r,r,t,r,x)dxdr.
In view of (4.7) it follows that
for some suitable constant C. The estimate (4.14) then follows from Gron-
wall's lemma. 0
To formulate the next technical lemma we need to introduce the or-

thogonal projections II+ and II_ onto the direct sums of the eigenspaces
corresponding, respectively, to positive and negative eigenvalues of the sym-
metric matrix M. These are smooth functions of x and t. To see this one
notes that the eigenvalues of M never vanish so that one can, for given x
and t, use the representation
II(x, t) =~
27rZ
1
'H
[zI - M(x, t)t 1 dz,
where r are contours surrounding the positive and negative eigenvalues

of M respectively. The smoothness of the eigenvalues follows easily from
this. We also introduce w = II w, which have the same regularity as w.
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
(4.15) 1,TIS, . ,[(a, t) dt 5, C [(0; x+(O), a) + 1,T (T; x+(T), a) dT].

Similarly, ifx_(t) meets the initial interval {O} x [O,l] one has
and
(4.17) IT ISr rJ(a, t) dt S C [f(O;a, :'-{O)) + IT f(T; a, :L(T)) dT].
PROOF. It follows from (4.9) that
8w 8w
(4.18) ~ =M-{} +l(r,rt,r x),
~L x "
with
Now
(4.19)
where q = 2w l - M,x w .W are quadratic functions of the arguments

r, r,t and r,x satisfying Iq1 ::; Ceo
To obtain the estimate (4.14) we integrate q+ (r, r ,t, r ,x) over the region
~+ bounded by the intervals [x+(O), a] x 0 and a x [0, T] along with the
characteristic curve C+(a, T). By using (4.19) and the divergence theorem
we obtain
(4.20) jjq+(r,r,t,r,x)dxdr= f v.(-Mw+.w+,lw+12)ds.

d+ od+
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
(-l,w) (-Mw+ w+, Iw+12) = Mw+ w+ +wlw+12.

Discarding this term one is left with
(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.
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+
The estimate (4.15) is an immediate consequence.

The estimates (4.16) and (4.17) are proved in an analogous way. 0
We can combine our previous results to obtain the following theorem.
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
Et(a;O,T)'; C [E(O;Xl,X2) + 1,T E(T;X+(T),X_(T))dT].
Now (4.21) follows from (4.12). 0

Note that the roles of x and t can be reversed by rewriting (4.9) as
Ow Ow
(4.23) ! l = N!l - NI(r, r t, r x),
uX uX "
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
In case P and Q are independent of t, both N and its eigenvalues 1/Wk

are functions of x alone. The characteristics k+ (l, t2) and k- (l, t1) are then
-1/~~y)
given by
t = t2 dy and t = t1 + 11 ~~y) dy.

Taking this into account we obtain immediately a variant on the previous
theorem.
THEOREM 4.3. Letr(x,t) E W 2((0,l)x(0,T);lRP ) beasolutionof(4.1),
where P and Q are assumed to be independent oft. Suppose that t2 -t1 ~
2S*, where
[I 1
(4.25) S* = 10 ~(x) dx.
Then for any t E [t1 + S*, t2 - S*],
(4.26) &T (0; t1 + S*, t2 - S*) + &(t; 0, l) ~ G&T (l; tll t2)'
The same estimate holds with the roles of and l reversed.
We can use our methods to derive another estimate which plays a com-
plementary role to the estimates of the previous two theorems.
THEOREM 4.4. Let r(x, t) E W 2((0, l) x (0, T); lRP ) be a solution of the
equation (4.1) which satisfies the boundary conditions
{ r(o, t) = 0 or [r,x + Rrj(O, t) = 0,

r(l, t) = 0 or [r,x + Rr](l, t) = 0
for t E (0, T). Then
(4.27) &T (0; 0, T) + &T (I; 0, T) ~ G&(O).
PROOF. This theorem depends on Lemma 4.2 and is somewhat subtle.
First we note that an easy computation, taking into account the bound-
ary conditions, gives
~t &(t) ~ G(t),
from which it follows that
(4.28) &(t) ~ G&(O) for t E [0, T].
It is therefore enough to show that
(4.29) &T (0; t, t + 0) + &T (l; t, t + 0) ~ G&(t)

for each t and some corresponding positive 0. We do this for t = 0. In

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
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
(4.30) 1 [Iw+(l, tW + Sr r] dt :S C(O).
The final step is to see how Iw+(l, tW is related to IW(l, tW = IPr,t1 2 +

IQ(r,x +Rr)12. Note first that, because of the boundary conditions, one or
other of the terms in the last expression vanishes. We have to look more
closely at the projection defining w+ = IT+w. In order to do this pick an
orthonormal basis {fk}~=l of IR,P consisting of eigenvectors of p-1Q2p-l
corresponding to the eigenvalues {w~ }~=l' Then one easily verifies that one
has an orthonormal basis of eigenvectors {Fi }~=l U {F~ }~=l' where F~ =
O:k (fk , WkQ-l Pfk)}~=l with suitable normalizing constants O:k. The first
set of eigenfunctions corresponds to the positive eigenvalues Wk while the
second set corresponds to negative eigenvalues -Wk. Now, if we assume the
boundary condition w(l, t) = 0 and remember that w = (Pr,t, Q(r,x+Rr))
we find
L [w(l, t) . Fk ]Fk
p
w+(l, t) = IT+w(l, t) =
k=l
= L [O:kwk[Q(r,X + Rr))(l, t) . Q-1Pfk] Fk

p
k=l
p
= L [O:kwk[P(r,X + Rr)](l, t) . fk] Fk.

k=l
It follows that
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
5. Exact Controllability of the Network Model.

The primal problem we wish to consider is the following:
P;rt = [Q;(r:x + ~rt)] ,x

- R!Q;(r:x + ~rt) + St rt for i E I,
rt(XtJ' t) coincide for j E .JM, i EI J,
(5.1) "DJr(t) = uJ(t) for j E = 0 for j E .Jc ,
.JD, "DJr(t)
J\{Jr(t) = uJ(t) for j E .IN, J\{Jr(t) = 0 for j E .JF,
N:r(t) = 0 for j E .J and i EI J,
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,
We denote elements of these spaces by u D and UN respectively, and use

the inner products
and
As before we let U = U D X UN with elements u = u D X uN.

Because of Theorem 3.1 we can then define the control to state map
CT : U ~ 1{, by CTU = r(T) x r(T) where r is the solution of (5.1). The
surjectivity of the bounded linear operator CT (Le. the exact controllability
of the networks) is then equivalent to the following estimate on the adjoint
operator CT : 1{, ~ U:
(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{,.
The adjoint system will be
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.
We then have the following characterization of C;'.
PROPOSITION 5.1. Suppose that 11llv is a norm on V (which is guar-

anteed, for example, by the hypotheses of Lemma 3.1). Let >..(t) be the
solution of (4.3) corresponding to terminal data >"0 x ~o E 'It. Then
(5.4) C;,[>"o x ~l = {ft.Alj >..(r) dr}. x {:z>i ~(t)}. ,

Jo JE.J D JE.J N
where >.. is the solution of the adjoint system (5.3).
PROOF. The hypothesis allows us to use (', '}v as inner product on V-

components of elements of 'It. As in the proof of Lemma 3.3, we evaluate
~t [(r, >")v + (r, ~)H]

=L
iEI
1~
0
1
; [P;ri . ~i + Q;(r:x + ~ri) . (~~x + ~~i) + Siri. ~i
+ P;ri . ~i + Q;(r:x + ~ri) . (>"~x + ~>"i) + Siri . >"i] dx

= L [Q;(r:x + ~ri) . ~i + ri . Q;(>"~x + ~>..i)] x=l;
iEI x=o
where we have integrated by parts with respect to x. Thus, taking into

account
5. EXACT CONTROLLABILITY OF THE NETWORK MODEL. 121
along with the various nodal conditions in (5.1) and (5.3), we obtain
~t(r x r,~ x ~)"H = L[1)1~Njr+Nj~.1)1r]

(5.5)
jE.:T
= L Nj ~ . 11/ + L 1)1 ~ . u j .
Upon integrating with respect to t from 0 to T one finds
(CTU,.xo x ~O)"H = iT[ L Nj~uj + L Dj~.uj] dt.

o jE.:TD jE.:T N
Hence (5.4) follows by Riesz identification. 0
Because of this lemma the estimate (5.2), characterizing exact control-

lability in time T, becomes
(5.6) II ~o x ~ II~ ~ C iT [L INj ~(tW + L 11)1 ~(tW] dt.

o jE.:TD jE.:T N
This nontrivial estimate will again be established, as in Section 5, Chapter
II, for tree networks with one fully clamped node and for T larger than a
certain critical time T*, for which we will give an explicit expression. The
counterexamples of that section of course tell us that we cannot expect a
better result.
We shall need two lemmas. The first one is a reformulation of Theorem
4.3 with 8* replaced by 8!, t = T /2, tl = 8 and t2 = T - 8.
LEMMA 5.1. Letri E W 2 ((0,li) x (O,T)jlRPi ) be a solution of the equa-

tion of the form (4.1) for the i-th string and suppose that T ;::: T! = 28!,
. iii
where
1
(5.7) 8! = -(-)dx.
wl x
o -
Then there exists a constant Ci such that
T
(5.8) Ei ("2) + Eii (OJ 8 + 8!,. T - . i
8 - 8!) ~ CiEi (lij 8, T - 8)
for all nonnegative 8 satisfying 28 + T! ~ T. The same inequality holds

with 0 and li interchanged.
A second lemma is a consequence of the geometric and dynamic condi-

tions which hold at multiple nodes.
LEMMA 5.2. Suppose that qj = Pi = P for all i = 1, , nand j E

J M, in which case the matrices Cij are invertible. Let r be a solution
of the system (5.1) with the initial condition replaced by the requirement
r(O) x r(O) E 1. Then for j E JM and io E I j one has the estimate
(5.9) Elo(xioj;tt,t2):::; C L El(xij;tb t 2),

iEIj-{io}
where C is a suitable constant.
PROOF. Under the hypothesis of the lemma the node condition (8) is
redundant and, in view of (7) and the positive definiteness of each Qi, it
follows that
IQio[r~o +~orio](xiok,tW:::; C L IQ~[r~ +~rd(xij,tW
iEI;-{io}
From (1) it follows that
IPioriO(xioj, tW :::; CIPiri(xij, t)12 for any i E I j - rio}.
From these estimates (49) is immediate. 0
REMARK 5.1. The estimate (5.9) is valid under the assumption that
qj = Pio :::; Pi for i E I j .
The statement and proof of the next main theorem is very close to that
of Theorem 5.2, Chapter II and depends on the previous two lemmas.
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
(indexed by j = m) is fully clamped and that the remaining simple nodes are
controlled either geometrically or dynamically. In the interests of simplicity
suppose also that Pi = qj for all i and j, in which case the matrices C ij are
all invertible. Let r be a solution of the system (5.1) with the homogeneous
initial conditions replaced by the requirement ro x ro E 1. Then
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
Finally we state the result on exact controllability, which, as before,

follows immediately from the previous theorem.
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.
6. Stabilizability of the Network Model.

We return to the question of nonlinear feedbacks with a view to stabi-
lizing the vibrations of a general network as modeled in this chapter. We
make some simplifying, but unessential assumptions. In particular we as-
sume that qi = Pi = P for all i and j and that the m-th node is clamped.
We then consider the following nonlinear velocity feedback system:
P~ri = [Q~(r:x + ~ri)J,x

t
- RiQ 2( i
i r,x + ~ri) + Siri for i E I,
ri(xii' t) coincide for j E 3M, i E Ii,
(6.1) 'Dmr(t) = 0,
-ftfir(t) E gi('Dir(t)) for j E 3s \ {m},
,N'ir(t) = 0 for j E 3M,
r(O) = ro, r(O) = roo
Here each gi(u) will be a maximal, monotone function obtained as the

subdifferential 8Gi(u) of a convex function Gi(u) : lRP 1-+ lR taking its
minimum value at O. From (3.12) it then follows that for any sufficiently
regular solution of (6.1)
de (t) = ,N'ir(t) . 'Dir(t)

(6.2) dt
iE.7s\{m}
with -,N'ir(t) E gi('Dir(t)).
Since necessarily 0 E gi (0), monotonicity implies
[-N'ir(t) - 0] [1)1r(t) - 0] ~ 0,
so that it follows from (6.2) that the energy is decreasing monotonely.

Our two main tasks are to establish the existence of solutions to the feed-
back system (6.) and to find conditions on the functions gi which imply
that the energy actually decays to zero, preferably with a specified decay
rate. We follow closely Conrad and Pierre [20] who provide an abstract set-
ting for such problems (see also Lasiecka [53]). For general facts concerning
convex functions we refer to Ekeland and Temam [28], while extensive infor-
mation on maximal monotone operators and contraction semigroups may
be found in Brezis [11] (noting in particular [11, Theorem 3.1]).
THEOREM 6.1. Let the maximal monotone functions gi be as described

above. Let dom( As) c V x H consist of all r x r with rEV and r satisfying
there exists f E H such that

(6.3) { (r, tP)v = (f, tP)H + Ei~.1S\~m} ~i .1)i tP
for all tP E V, with -uJ E gJ(1J1r).
For r x r E dom(As) define As by setting As[r x r] = -r x f. Then

As is a single valued, maximal monotone operator on 1t which has dense
domain. Thus -As generates a contraction semigroup {Us(t)}t>o on 1t
so that Us(t)[ro x ro] provides a solution to the system (6.1). Moreover,
As has compact resolvent.
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)
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}
= L [( -ui ) - (-vi)] . [1)ir -'])is]

iE.1 S \{m}
~O,
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
(r, q,)H + (r, q,)v = (f -: t, q,?H -: E iE .1 S \{m} ui .'])i q,

(6.5) {
for all q, E V, with -uJ E gJ(1)3[r - f]),
The solution of this equation is obtained by showing that the following con-
vex functional on V achieves its minimum at an element r which necessarily
satisfies (6.5):
1 1
(6.6) Q(r) = 2(r, r)H + 2(r, r)v - (f + f, r)H
+ L Gi('])i[r - f]).
iE.1 S \{m}
This functional does achieve a minimum. To see this note first that
-(r, f)H + L Gi(1)i[r - fD ~ a + b IlrliH +cllrllv,

iE.1 S \{m}
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
then evident that Q(r) - 00 as II r Ilv- 00. The existence of a minimizing

element, which for the moment we denote by f, is then a consequence of a
standard result on convex functions (see [28, Proposition 1.2, Chapter II]).
It follows easily from
Q(f) ~ Q(f + f[r - f]) for all rEV
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}
< ~ f[Gj(1)1[r - f]) - Gj(1)1[f - fD].

jE.JS\{m}
In the last step we used the following consequence of the convexity of Gj:
Dividing by f and letting f go to zero we get
-[(f,r - f)v + (f,r - f)H - (f + f,r - f)H]

(6.7) < ~ [Gj(1)1[r-f])-Gj(1)1[f-f])].
iE.JS\{m}
Thus the linear functional 1(4)) = (f, 4v + (f, 4H - (f + f, 4H satisfies
(6.8) -1(4))~ ~ [Gi(1)1[f-f+4>])-Gi(1)1[f-f])].

iE.JS\{m}
so that -I belongs to the subdifferential at f of the convex function
Q(r) = ~ Gj(1)1[r - fD.

iE.JS\{m}
To complete the proof of existence one must show that
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
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
l(cp) = l(CPl) = L u i . 'Dicp.

iE.1 S \{m}
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
(6.9) L[llrillw2(O,li;.IlP) + IIrllw1(O,li;RP)] ~ C[lIfllv + IIfIlH]'

iEI
The first step is to set cP =r - f in (6.5). After simple rearrangement of

terms one gets
IIr - fll~ + Ilrllt-= (r,f)v + (f,r - f)H + L ui .tJi[r - fj.

iE.1 S \{m}
Since -ui E gi(tJi[r - fD and 0 E gi(O) it follows from the monotonicity

of gj that
From the previous identity we get the estimate
so that
(6.10) IIr - fll~ + IIrllt-~ C[lIfll~ + IIfllt-]

Now we are essentially finished. Since
IIrllv=llr - fllv~lIrllv + Ilfllv ,

if we take into account Lemma 3.1 the estimate (6.10) implies that part of
(6.9) which pertains to r. To get the compactness estimate on r we note
that the norm defined by
is equivalent to the norm
L II ri IIW
iEI
2(0,1;;RP)
Therefore (6.10) implies estimate (6.9) for r since

II ArlIH=lIr - f - fllH::;llr - fliH + IIfllH
and
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
(6.11) g1(0) = 8Gj(0) = {O} for all j E .J \ {m}.

Then
(6.12) lim IIUs(t)[ro x folll1t = 0 for all ro x fo E 1.

t-+oo
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,
L oAf1s(t) .1)1s(t) = 0, with -Njs(t) E gj(1)1s(t)).

jE..1s\{m}
It follows from the monotonicity of each gj that

Njs(t) .1)1s(t) = o.
Since
Gj(O) ~ Gj(1)1s(t)) - Njs(t) [0 -1)1s(t)] = Gj(1)1s(t)),
we conclude that Gj(O) = Gj(1)1s(t)). From the hypothesis it follows that
1)1s(t) = 0 and -Njs(t) E gj(O); consequently we also have Njs(t) = O.
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
REMARK 6.1. The assumption (1.11) is in fact a necessary condition for

the conclusion of the theorem to hold. In order to see this suppose that, in
contradiction to (6.11), some gk(O) contains a non-zero vector uk . Then
there exists a non-zero solution of
[Q~(r~x + Rri)] ,x - R~QHr~x + Rri) + Siri = 0 for i E I,
ri(xii) coincide for j E 3M, i E Ii,

(6.13) ,N'ir = 0 for j E 3M
'Dmr = 0,
,N'kr = uk, ,N'ir = 0 for j E 3s \ {m,k},
Consequently, since 0 E gi(O) and uk E gk(O), r x 0 is a non-trivial

stationary solution of (6.1); its energy does not decay to zero.
On the other hand the condition that the minimizing set min(Gi) of Gi
contain only 0 is unnecessarily stringent. Decay is, for example, guaranteed
ifmin(Gi)nmin(8Gi) = {O}. For details see [20J, where one can also find a
condition which together with (1.11) will be necessary and sufficient for the
conclusion (1.12) to hold. That condition again involves the non-existence
of an overdetermined eigenvector.
We now use Theorem 5.1 to prove a result on the exponential decay of

energy for the feedback system (6.1) in the case of a tree network.
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)
Then the feedback semi-group decays exponentially: there exist constants f3

and C such that
IIUs(t)[ro x rolll'H ~ C exp( -f3t) II ro x ro II'H .

PROOF. From (6.2) it follows that
/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
(6.16) IIUs(T*)[roxrolll~~/lroxro/l~ -8C- 1 L ~(xj,O,T*).

jE..7s\{m}
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)
where the constant C has been redefined. It follows that
the exponential decay follows from this. 0
Finally we discuss feedbacks yielding only polynomial decay rates. This

requires a much more subtle argument and depends on the following two
lemmas.
The first lemma is a consequence of Theorem 5.2.
LEMMA 6.1. Under the hypotheses of Theorem 5.1 we have for all T >0
(6.18) fT (t) dt ~ C [ sup (t) + L fT ei, (Xj, t) dtj.

Jo O::;t:ST jE..7s\{m} Jo
PROOF. It is obviously enough to prove this when T > T* = 28*. From
Theorem 5.2 we can conclude, after a time shift, that for any t
(6.19) (t)~C L
jE..7s\{m} t-S.
I t +S
e., (xj,r) dr.
Now
l o
T &(t) dt = ls*
0
&(t) dt + i T
T-S.
&(t) dt + iS.
T S
- &(t) dt.
The sum of the first two integrals on the right is bounded by
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.
LEMMA 6.2. Let hl be a positive, increasing function with hl (0) = 0

Suppose that a sequence of positive numbers Sk satisfies the estimate
(6.20)
Then Sk ::; S(k) where S(t) is the solution of

d
(6.21) dt 8{t) + h{S{t)) = 0, 8(0) = so,
with h a positive, increasing function satisfying h{O) =0 defined by h =
id - lid + hl]-l, "id" denoting the identity map.
We can now prove the following theorem.
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
(6.22) g. (u) u> { 8[1u12 + Ig(uW] for lui> 1,

J - 8[1ul 2+ Ig(uW]P for lui::; 1,
where p > 1 and 8> O. Then the feedback semi-group has decay-rate
(6.23)
PROOF. Since the energy is decreasing we get from (6.18) that
where here and below we repeatedly adjust the value of C. Let
Ij = {t> 01 l1)if(t) 1~ I}, Jj = {t> 01 l1)if(t) 1> I}.

Then, taking into account the hypothesis (6.22),
L
jE.1s\{m} 0
iT [11)if I2 + Igj (1)if)1 2 ] dt
~c L [lgi(1)if).1)ifdt+l [gi(1)if).1)if]l/ P dt]

jE.1s\{m} Ij Jj
~c[ L iT gi('Dj f).1)ifdt
iT
jE.1s\{m} 0
+ T1/q [ ~ gi(1)if). 'Djfdtf/P] ,

jE.1s\{m} 0
where we have used HOlder's inequality. We substitute this, along with the
identity
(6.25) (T) = (0) - L

jE.1s\{m} 0
iT gi(1)if) 1)ifdt
(which follows from (6.2)) into (6.24) to get
(6.26) T(T)~C[(T)+2 L iT gi(1)if).'Dj fdt
i
jE.1s\{m} 0
+ T1/q [ L T
[gi(1)if).1)if]dtf/P].
jE.1s\{m} 0
We can rearrange this, again using (6.25), to obtain

which for T sufficiently large gives
(T) ::; [((0) - (T)) + ((0) - (T))I/P]

(6.27)
= lid + ho]((O) - (T)),
where ho(s) = slip. If we then set hI = lid + hot l it follows that
(6.28) (T) + hl((T)) ::; (0).
Let Sk = (kT). Then it follows from the previous estimate, which is valid
over the interval [kT, kT + TJ, that the hypothesis (6.20) of Lemma 6.2
holds. This allows us to prove the decay rate. We define h( s) as in the
lemma. It is then an easy exercise to show that the two identities
hl(s + slip) = s and h(s + hl(s)) = hl(s)
together with the hypothesis p > 1 imply
(6.29) lim hl(S) =1 and lim h(s) = l.

8-+0 sP 8-+0 sP
Since Theorem 6.2 implies that limk-+oo Sk = 0 we can assume So to be
as small as we please. The solution S(t) of (6.21) satisfies S(t) ::; So for
t > O. Given E (0,1) So can be assumed so small that Ih(s) - sPI < Sp
for s ::; so. Then, in particular it follows from the equation (6.21) that
d
dt S(t) + (1 - )S(t)P ::; o.
It follows that S(t) = O(tl/(l-p) so that (kT) = O([kTj-l/(p-I)). Since
(t) is decreasing this implies the decay estimate (6.23). D
CHAPTER V
Spectral Analysis and Numerical

Simulation of I-d Networks
1. Preliminaries
In this and in the following section we shall consider the problem of

determining the frequency spectrum and the corresponding modes of a
given network of elastic strings or elastic beams. This problem is of great
importance for purposes of obtaining Fourier series expansions (Galerkin
approximations) of solutions to the dynamic equations of motion and also
for certain approaches to the theories of control and optimal design. There
is a huge literature devoted to this problem, within which we can identify
two groups of papers. The first group, comprising the majority of these
papers, is concerned with a finite element approximation of the underlying
system and with the generalized eigenvalue problem associated with that
approximation. The latter problem is then solved using a suitable numer-
ical procedure, such as the shifted Q-R algorithm or Jacobi's method, if
the dimension of the approximating space is comparatively small, or by
other methods such as the Lancos algorithm, the Raleigh-Ritz method,
etc., if one is interested in obtaining a particular subset of eigenvalues.
For repetitive structures, which are typical in network problems, yet an-
other method has been employed with great success, namely, the method
of slicing the spectrum, which is based on Sylvester's theorem of invari-
ance of the number of positive, vanishing and negative eigenvalues under
similarity transformations. This method has been developed for repetitive
structures by Wittrick and Williams [110], [111] and others [103]. The un-
derlying idea is that of substructuring, that is, of dividing the structure
into smaller identical pieces. The analysis is then performed on the level
of the substructure and the results are read into the master structure in an
appropriate way.
That this strategy is not only admissible for finite dimensional approxi-
136 V. SPECTRAL ANALYSIS AND NUMERICAL SIMULATIONS
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.
1.1. Notation. We consider a nonempty, finite, simple and connected

graph G in JR2 with n vertices V(G) := {Vi, i = 1, ... ,n}, and N edges
E(G) := {kj, j = 1, ... , N}. The edges k j are parameterized by 'Trj :
[O,i j ] ~ JR2, where the running variable Xj E [O,ij] represents the arc
length. The maps 'Trj are assumed to be C 2 -smooth. For i,j E V(G) we
set ks(i,h) := ViVj = VjVi, where ViVj signifies the edge joining the vertices
Vi, Vj, if ViVj E E(G). In fact, as we will be concerned here only with
networks consisting of straight edges, we assign to each edge ks(i,h) the unit
edge vector eS(i,h), directed along the edge, and its orthogonal complement
e~i,h)' We also introduce the unit vector n normal to the plane of the
graph. To each triad es(i,h), e~i,h)' n we associate the corresponding set
of orthogonal projections P s(i,h), P~i,h)' Pn' Further, we introduce the
1. PRELIMINARIES 137
incidence matrix
if 7I"j(lj) = Vi,
(1.1) if 7I"j(O) = Vi,
else,
and the adjacency matrix
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)
)
lS(i'h)(1 - x) ~f 7I"s(i,h) (ls(i,h) = Vi, eih = 1,

{
= ls(i,h)X If 7I"s(i,h) (0) = Vi, eih = 1,
o if eih = O.
For each vertex we define the set r(Vi) := {Vh E V(G)leih = I} of all
vertices adjacent to the vertex Vi, and d(Vi) := Ir(Vi)1 the edge degree of
the vertex Vi. We can now characterize the network as follows:
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
It is, in some circumstances, more convenient to express those functions r j

in the normalized form introduced above, i.e.,
(1.3)
With this definition we find
rS(i,h) (s(i,h)) if 7!'s(i,h) (s(i,h)) = Vi,

rih (0) =eih {
rs(i,h) (0) if 7!'s(i,h) (0) = Vi,
=eih r s(i,h) (7!';(;,h) (Vi)) = eihr( Vi)'
and also
rhi(X) = rih(l - x).
This notation is adopted from [106].
1.2. Networks of Strings. In this section we consider a network of

strings which, in its reference configuration, coincides with a graph G as
described in the last section. The equations governing the motion are
formally derived in Chapter II. The model we consider here, however, is
also easily developed along the lines of Chapter IlIon the modeling of beam
networks. Let r denote the deviation from the reference configuration. By
the definition above, r j denotes the displacement of the string (edge) with
label j. We define
so that represent the longitudinal and vertical displacements of the

Uj, Wj
j th string, respectively. Let PJ,
qJ denote the longitudinal and flexural
rigidities. FUrthermore, let K j := P;Pj + q;pl
be the stiffness operator
associated with j th string. Then the equations governing the motion of
the network of strings can be written as follows:
(1.4)
j = 1, ... ,N,
rj(v) = 0, Vv E ooV(G),
Kjrj(v) = iv, Vv E 01 V(G),
o
ri(vh) = rj(vh), Vi,j E indf(vh), Vh EV (G),
o
Lh:VhEr(V;) dis(i,h)Ks(i,h)r:(i,h) (7!';(;,h) (Vi)) = gv, Vi : Vi EV (G),
rjCO) = rjO, rj(',O) = rj1
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
through the functions gv). These equations may be expressed in terms of

the functions r th defined above as follows:
.. 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)
(1.7) Lf~h1)Ks(t,h)r~(t,h)(0) = gt, Vi: Vt E V(G),

h
where (i) E lR2 in (1.6) represents r(vt ). If V t E 8oV(G), then t = o. If

Vt in (1.7) is in 8 1 V(G), then there is exactly one vertex in f(v t ), and (1.7),
for that i, expresses the Neumann boundary condition for that node. The
balance condition (1.7) is the analog of the Kirchhoff law in the context of
electrical networks. Of course, f~h1) means f~!,h)eth.
The above setup is in complete analogy with the one to be found in
[106], where a static lR-valued Neumann eigenvalue problem is considered
in the context of diffusion equations on a network.
1.3. Networks of Timoshenko Beams. The situation is notationally

very similar to the last section but mathematically more involved in the
case where each edge is considered as a Timoshenko beam in its reference
configuration. In addition to the longitudinal and vertical displacements
(uJ , wJ ) of the centerline of the beam we also have a rotation 'lj;J :=<
PnrJ , n > of the cross section out of its position perpendicular to the
centerline, i.e., we have to account for shearing. We denote the shear
modulus of the j th beam by q;,
its flexural rigidity by m; and, as before, p;
denotes its longitudinal stiffness. We only consider here an initially straight
and untwisted linear Timoshenko-Bresse system as described above:
(1.8)
(1.9)
(1.10) L dts(t,h){P~(t,h)U~(t,h)es(t,h) + q~(t,h)(W:(t,h) + 'lj;s(t,h)e~t,h)

h
rt(V) = 0, Vv E OoV(G),
(1.11) { {p~u~eJ + q~(W~ + 'ljJJ)ef }(Vt ) = ft,
m~'ljJ;(Vt) = ht, VVt E 01 V(G).

In analogy to the previous section we define the operators
K J :=P;PJ + q;p/ + m;P n ,

MJ :=q; < Pn(-),n > ef,
NJ :=q; < P/(-),ei > n,
SJ :=NJ - M J ~ S; = -SJ'
Then the state equations (1.8) can be written as
(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
(1.13) L dt8(t,h) (Ks(t,h)r:(t,h) + Ms(t,h)rS(t,h(7l'~~,h(V,) = 0,

h
VV t E V(G) \ ooV(G).
We can as well express these equations in terms of matrix entries as follows:
3</J: rth(O) = eth</J(i)j </J(i) = 0, VV t E ooV(G),

(1.14) Eh[l~~I)Ks(t,h)r~h(O) + dts(t,h)Ms(t,h)rth(O)] = 0,
Vv, E V(G) \ 00 V(G),
rht(X) = r th(l- x), r,h(',O) = r,hO, r,h(',0) = r thl.
1.4. Networks of Euler-Bernoulli Beams. Here we consider the
case where all edges are taken to represent Euler-Bernoulli beams. By this
we mean that shear strains are not taken into account or, what amounts
to the same thing, that the cross sections of ~he deformed beam stay per-
pendicular to the deformed centerline. As we are going to discuss planar
networks only, we have again two displacements, the longitudinal, uJ ' and
the vertical, wJ We denote the longitudinal and flexural stiffnesses by
p~, q;, respectively. We will consider only homogeneous boundary and
nodal conditions. We use the notation developed in section 1.1. By setting
rJ = uJeJ + wJef, K J = q;PJ, H J = p;PJ,
we can rewrite the state equations
(1.15)
as
(1.16) rJ - H Jr"J + K Jr""

J = 0.
The continuity (connectedness) condition again reads

o
(1.17) rJ(vh) = r,(vh), 't/Vh EV (G).
This time, however, depending on the particular nature ofthejoint, we have
other geometric or compatibility conditions. In the case of rigid joints, the
case of most interest, we have
(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
(1.19) L dts("h) (Ks(t,h)r~(t,h) - Hs("h)r~(t,h)K7r~!,h) (vt )) = 0,

h
(1.20) L dts(t,h)Ks(t,h)r~("h) (1r~!,h) (vt )) = 0,

h
o
't/vt EV (G),
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
(1.21) K8(t,h)r~(t,h) (1r~!,h) (v,)) = 0, 't/h E ind r( v,).

Of course we also have initial conditions
(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);
(1.25) { 3~ E RN : i~hl) dts(t,h)P~"h)r~h (0) = eth~t,

~, = 0, 't/v, E 8oV(G);
(1.26) L {i~~3)Ks(i,h)r~% (0) - i~~l)Hs(i,h)r~h (On = 0;

h
(1.27) LdiS(i,h)i~~2)Ks(i,h)r~~(0) = 0, Vh E V(G) \ 8oV(G);

h
(1.28)
As remarked earlier, in the case of pinned joints we have to neglect (1.25)

and instead have to strengthen (1.27) to
(1.29) Ks(i,h)r~~(O) = 0, VVh E V(G) \ 8oV(G).
2. Eigenvalue Problems for Networks of 1-<1 Elements

2.1. Introduction. As in [106] we introduce a matrix notation which
provides a general framework for eigenvalue problems on multiconnected
domains. The notation is slightly different from the one introduced by
Nicaise [75] who, independently, discussed the eigenvalue problem for the
second derivative operator on homogeneous networks. Given two matrices
A, B we denote by A B the matrix with entries (aihbih) (coordinatewise
mUltiplication). It is understood that the same notation is used also in the
case where the entries of B are vectors (in which case we write B in place
of B) while those of A are scalars: i.e. A B = (aihbih)' One can also
introduce the following functional calculus with matrices: given a function
f (.) and a matrix A we define f (A) to be the matrix (f (aih) ). This notation
can be extended to the case of vector-valued matrix entries in the following
way. If B = (b)ih = (bihes(i,h) + bihe*i,h)) then f(B) = (f(bih)es(i,h) +
f(bih)e*i,h))' Let us also introduce the vector e := (1, ... ,1)* such that
Ae = Lh aih; an analogous notation is used for vector-valued matrices.
Without loss of generality we may choose the indices of the set 81 V (G) to
be i = k + 1, ... ,n. We introduce the diagonal matrix Z with zeros in the
first k entries and ones in the others. We will in general have to consider a
function r : G f-t IRT describing the deformation of the mechanical system
which, in its rest configuration, covers the graph G. The matrix R := (rih)
is then defined following the procedure outlined in Section 1.1. The general
structure of an eigenvalue problem associated with a network of the kind
described above will then be of the following type:
(-2) . K. R" + (-1) . S R' + Q. R = _).2 R,
3<p: R'(O) = <pe* =: <1>; <Pi = 0, Vi: Vi E 8oV(G),

(2.1)
Z((-l) . K . R'(O) + V MR(O))e = 0,
R*(x) = R(1 - x), "Ix E [0,1].
2. EIGENVALUE PROBLEMS 143
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.
2.2. General String Networks. We use the notation above with K,

given as in Section 2. We obtain the following problem
(-2). K,. R"(x) = -A 2 R(x),
3: R(O) = e* . ej i = 0, Vi: Vi E ooV(G),

(2.2)
Z((-l) . K,. R'(O))e = 0,
R*(x) = R(1 - x), Vx E [O,IJ.
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.
THEOREM 2.1. If A is an eigenvalue of the system (2.2), then A satisfies

one of the following conditions:
(i) A = 0 (iff ooV(G) = 0);
(ii) A = i~~,h)P8(i,h)7I'k for some i, h = 1, ... ,nj k E Z;
(iii) A = i~!,h)q8(i,h)7I'k for some i, h = 1, ... ,nj k E Z;
(iv) A is such that ker Z(C - diagHi ) =F 0, where
Cih := fS(i,h)Ps(i,h) (sin(Afs(i,h)P~!,h))-1eihP s(i,h)

+ f s( i,h) qs( i,h) (sin( Aq~!,h)) -1 eih P * i,h)'
Hi := L eihfs(i,h)(PS(i,h)(sin(Ap~!,h))-1 cos(Afs(i,h)P~!,h)P s(i,h)

h
+ 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
R(x) = ~ + x(~* - ~),

whereas the third equation leads to
(-1) ./C. (<p* - <p)e = o.
We recall the meaning of <p* = E . e>* and obtain
((-1) ./C)> = diag (((-1) ./C)e)>.
In contrast to [106], we cannot invert the operator on the right side of this
equation. Nevertheless,
diag ((( -1) /C)e) = diag (2":: .e~!,h)P~(i,h)eihP s(i,h)

h
hence the balance condition reduces to
L .e~!,h)P~(i,h)eihPs(i,h) (>h) + L .e~!,h)q~(i,h)eihP~i,h) (>h)

h h
= (L .e;(!,h)P~(i,h)eihP s(i,h) + L .e;(!,h)q;(i,h)eihP~i,h)>i.

h h
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
and the reflection condition

This implies (ii) as stated in the theorem. In addition, if V t is in V(G) \

8oV(G) we have to satisfy the balance condition, the second equation in
(2.2), which now reads
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):
We will extensively use the reflection formula:
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
and rewrite the reflection condition obtained above as
U~hO = eth>,p;(!,h) sin(>.fs(t,h)P;(!,h)-l(u:hO - cos(>'fs(t,h)P~!,h)UthO)'

W:hO = eth>.q~~,h) sin(>'fs(t,h)q~~,h)-l(w:hO - cos(>'fs(t,h)q~~,h)WthO).
This result is transferred back to the old notation
Wih = eihAis(i,h)P;!,h) sin(Ais(i,h)P;!,h)-l P s(i,h)

(CP;h - cos(Ais(i,h)P;(!,h)cpih)
- 1 (\0 -1 )-1
+ eihAf-s(i,h)qs(i,h)
\0
sm Af-s(i,h)qs(i,h)
P~i,h)(cp;h - cos(Ais(i,h)q;!,h)cpih).
We are now in the position to eliminate Wih from the balance condition
(2.2h. The result is
I: {eihis(i,h)Ps(i,h) sin(Ais(i,h)P;!,h)-l
h
P s(i,h) (cp;h - cos(Ais(i,h)P;!,h)cpih)

+ eihis(i,h)qs(i,h) sin(Ais(i,h)q;!,h)-l
P~i,h)(cp;h - cos(Ais(i,h)q;!,h)cpih) = 0
o
for all i such that Vi EV (G) U (h V (G). In order to make this equation a
bit more transparent we introduce the operators C, Hi as in statement (iv)
of the theorem and verify that, in fact, the balance equation reduces to the
equation given in (iv).
This concludes the proof of the theorem. 0
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.
2.3. Homogeneous String Networks. We consider the special case

where the quantities is(i,h)P;!,h) , is(i,h)q;!,h) are equal to a single number
c, implying that PS(i,h) = qS(i,h) and hence is(i,h)Ps(i,h) = is(i,h)qs(i,h) is a
songle number b. Hence, the operator .c( -2) . K reduces to cI and (2.2)
reduces to
R"(x) = -(CA)2 R(x),

3: R(O) = e* . E; i = 0, Vi: Vi E 8oV(G),
Z(.c R'(O)e) = 0,
R*(x) = R(I - x).
(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
r~h(O) = Wih = . C(A \) (e* - cos(d)e*) .

sm CA
and
sin( CAX)
(2.4) (rih)ih = COS(CAX)' >e* + sm
. (CA) (e>* - COs(cA)>e*) . ,
which are used in the balance conditions to obtain
(.c. (e>* - cos(d)>e*) . )e = O.
This, in turn, can be expressed as
.c> = cos( CA )diag (.ce )>.
Upon introducing the matrix
(2.5) Z := (diag (.ce))-l.c
we see that the equation to be satisfied by > is in fact an eigenvalue problem

for Z:
Z> = JL>
if we put A = arccos(JL)/c. It turns out that the case where sin(cA) = 0, i.e.,
where A = c- l k7r, corresponds to the eigenvalues 1, -1 of Z. Therefore,
all possible eigenvalues of (2.2) can be discussed entirely in terms of the
matrix Z. For easier reference, we display the results obtained in [106] or
in equivalent form in [75]. First it is shown in [106] that Z has eigenvalues
JLi : i = 1, ... ,n satisfying 1 = JLl > JL2 2: ... 2: JLn 2: -1, with corre-
sponding eigenvectors ~l = e, 6, ... '~n' respectively. In particular, -1 is
an eigenvalue if and only if the graph G contains no odd cycle, which is the
case if and only if it is bipartite (i.e., the vertex set V(G) can be separated
into two disjoint sets such that each edge in E(G) has one end point in
either set). Using these eigenvalues and eigenelements, von Below derives
the following characterization of the eigenvalues and eigenelements of (2.2).
Label the strings of eigenvalues by Ask, where s denotes the number of a
string of eigenvalues while kEN is the running index in that string. The
corresponding eigenelement is denoted by >sk. For simplicity, we change
CA to A. Let a be a generic real number.
(1) A10 = 0, >1o = aej
(2) Ask = - arccos(JLs) + (k + 1)7r if k == 1 mod 2,
2 ~ s ~ n and -1 < JLs, kEN, >sk = a~sj
(3) Ask = arccos(JLs) + k7r if k == 0 mod 2,
2 ~ s ~ n and -1 < JLs, kEN, >sk = aesj
(4) Alk = 27rk, k :f:. 0, >lk = ae, \11* = -\II, (.c. \II)e = OJ
(5) An+l,k = (2k + 1)7r, kEN, >n+l,k = 0,
\11* = \II, (.c. \II)e = 0 if G is not bipiU"titej
(6) Ank = (2k + 1)7r, kEN, >nk = aen,
\11* = \II, (.c. \II)e = 0 if G is bipartite.
(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
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
It is easy to interpret these node configurations geometrically; we omit the

details. Note that in the notation developed here the component 4>t of the
first eigenvector consists of the first two entries in 4>1 above and signifies
the deformed position of the first node, and so on. To be very explicit, let
us identify the deformation matrix R = r and then let us go back to the
original setup in which each individual string is labeled counterclockwise
along the square. We display only nonzero entries.
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
FIGURE V.I. Some eigenmodes of the unit square
rI2(x) =cos(x~)i +sin(x~)</>2 = r2(x),

rI4(x) =cos(x~ )</>1 + sin(x~ )</>4 = rl(x),
r21(x) =cos(x~ )</>2 + sin(x~ )</>1 = r2(1- x),
r23(x) = cos(x~ )</>2 + sin(x~ )</>3 = r3(1- x),
r32(x) =cos(x~)</>3+sin(x~)</>2 =r3(x),
r34(x) = cos(xi )</>3 + sin(xi )</>4 = r4(x),
r41(x) =cos(x~ )</>4 + sin(x~ )</>1 = rl(l- x),
r43(x) = cos(xi )</>4 + sin(xi )</>3 = r4(1 - x).
It is easy to check the continuity and balance conditions in either setting

directly using this table. We display some typical configurations in Figure
V.l.
We now want to use the matrix in (iv) of Theorem 2.1 directly in order
to compute eigenvalues and corresponding multiplicities even in the case
where the material properties are not assumed be equal to a single con-
stant. In order to obtain numerical evidence, we plot the determinant of
the matrix C - diag(Hi ), chopping off all values beyond 10. See the first
graph of Figure V.2. The zeros indicate eigenvalues of the entire graph.
In the second graph of Figure V.2 we show the multiplicities. This is done
in the following way. We perform a Doolittle L-R decomposition of the
10
\
\ ) \
4
o
o 4 10
o
o 4 10
FIGURE V.2. Eigenvalues of unit square
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.
EXAMPLE 2.2. The carpenter's square.

We have two links connected together in a right angle. We label the only
interior node with #1, and the two simple nodes by #2 and #3. Again
\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
FIGURE V.3. Splitting of eigenvalues
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
FIGURE V.4. Eigenvalues FIGURE V.5. Splitting

of a triangle of the spectrum
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
L elh4>h = 4>2 + 4>3 = 2J.t4>b L e3,h4>h = 4>1 = J.t4>3

h h
which, because of 4>2 = 0, reduce to 4>1 = 2J.t24>1 or to J.t = 1/..fi and

imply that ). = 7r / 4, 37r / 4 plus all integer multiples of 7r.
This example shows how the "softening" of boundary conditions (Le.,
going from Dirichlet to Neumann conditions) leads to a decrease of the
lowest eigenvalue, as it should be.
EXAMPLE 2.3. The equal sided triangle.
This case is also easily treated. This is the simplest case of a non-
bipartite graph. Also N = n, and G is 2-regular. In fact this graph is also
complete in the sense that = ee* - I :
=(~~~).
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.
///\ -l
21
15
05
05
j
I
15f
21
1 05 05 15 25
FIGURE V.6. A typical frame
EXAMPLE 2.4. The triangle with appendices.

We now take the triangle of the last example and attach a string at each
node. The exterior nodes are first assumed to satisfy Neumann conditions.
We label the multiple nodes as 1, 2, 3 and the simple nodes of the attach-
ments as 4, 5, 6, respectively. We still have N = n, but now the graph
is not regular (and, of course, not complete). Here the adjacency matrix
takes on the form
0 I I I 0 0
I 0 I 0 I 0
I I 0 0 0 I
= I 0 0 0 0 0
0 I 0 0 0 0
0 0 I 0 0 0
As a result we obtain from Z three different strings of eigenvalues with
arccos(J.l.s)/3 = O.35701T, 0.60821T, O.77851T.
The other eigenvalues are 0 and again the even multiples of 1T.
If we now consider the three attached strings as being clamped, then the
adjacency matrix of the sub-graph problem, obtained by deleting the rows
associated with clamped vertices, is the same as in the previous example.
However, the edge degree still is 3 rather than 2 as in the previous exam-
ple. In this case we do not have 0 but rather the multiples of 1T and the
eigenvalues associated with arccos(J.l.)/4 = .60821T, .26771T.
EXAMPLE 2.5. Outlook for geneml frames.
In the general case we have to deal with repetitive structures as in Wit-
trick and Williams [110J, [111J, Simpson and Tabarrok [103J. For homo-
geneous networks the eigenvalues can be given explicitly according to the
theory. This can also be done in the 3-dimensional case. We close the col-
lection of examples with a suggestive picture which appears to be typical
in this context; see Figure V.6. In this situation we obtain evidence on the
distribution of eigenvalues and their multiplicities; see Figure V.7.
[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) III /11 r~i. _~~~~

o 4 6 10 12
FIGURE V. 7. Distribution of eigenvalues

[4 16 18 22 24 26
2.4. Networks of Timoshenko Beams. We are now going to discuss

the eigenvalue problem for the Timoshenko system (1.8), (1.9), (1.10) and
(1.11), i.e., the system
P _
(-2)K
ih ,,(-1)8
s(i,h)rih ih + '
s(i,h)rih + qS(i,h)
2
nrih - -A
\2
rih,
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,
:;(~,h)q~(i,h)W~h + :;(!,h)q~(i,h)'~h = - A2wih,

:;(~,h)'Y~(i,h)'~h - :;(!,h)q~(i,h)W~h - q~(i,h)'ih = -A 2'ih.
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)
Since we want to solve the initial value problem, we have to use the com-
patibility conditions
Wih(O) = WihO, W~h(O) = W~hO'

II
Wih
(0) -- -
>..22s(i,h)qs(i,h)
-2 WihO - s(i,h) ./,1
'f'ihO'
(2.10)
The characteristic equation for the fourth order equation (2.9) is
and it has the roots
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
to the initial value problem (2.9), (2.10). To this end we define
Then W.h is given by
W.h (x) = (p?::. COS(/-L.hl x) + p?~ COS(/-L.h2X ))W.hO

+ (p~~ sin(/-L.hlx) + p~~ sin(/-L.h2x))w~ho
+ (()~~ COS(/-L.hl x) + ()~~ COS(/-L.h2 X))'I/J:hO
+ (()?~ sin(/-L.hl x) + ()?~ sin(/-L.h2x) )'I/J.hO.
We use this representation to solve the equation
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)
governing 'I/J.h by the method of variation of parameters. Let us introduce
c .- fs( h)
..,.h .-
V'2 _ 2
1\ Q ( h)
'V
IS ( .,h ) S "
Then 'l/Jih is given by
(2.12)
'l/Jih ='l/Jiho [ COS(~ihX) ( 1- is(i,h)

qs( t,h)
.)2 (
( - - .-
l's( t,h)
e_ ih fLihl
00
2 + e0 ih_fLih22
01 ))
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
.
We also have the longitudinal deformation
(2.13) Uih = cos

, is(i,h))
( 1 \ - - UihO
+~
Ps(i,h) . ( , S(i,h) ,
SIll UihO. 1\--
PS(i,h) s(i,h) Ps(i,h)
Given the deformations (2.11), (2.12) and (2.13) we can go back to the
matrix formulation and write
We want to express this in terms of the initial values. To this end we define
Uih(X) + Ulh(\ X)U~hO'

=: UPh(A, X)UihO
Wih(X) =: WB?(A, X)WihO + WiO~(A, X)W~hO

+ Wlf(A,X)"pihO + Wl;;(A,X)"p;hO'
"pih(X) =: ZP~(A, X)WihO + ZPl(\ X)W;hO
+ ZI~(A,x)"pihO + Zll(\x)"p~hO'
where the functions on the right hand side are defined using the formulas
(2.11), (2.12) and (2.13). Thus
Another way of writing this is
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
rih(O) = q,ih = UihOes{i,h) + WihOe*i,h) + "pihO n ,

r~h(O) = Wih = U~hOes{i,h) + w~hOe*i,h) + "p~hOn.
We may now define
e?h(A,X) = U~(A,X)Ps{i,h) + WP'?(A,X)PS{i,h)

+ Wi~(A,X)Ms{i,h) + Zf,?NS{i,h) + ZI~Pn,
e}h(,x,X) = Ulh(,x,x)Ps(i,h) + Wi<r(,x,X)PS(i,h)

+ Wll(,x,x)Ms(i,h) + Z?l(,x,x)Ns(i,h) + zflPn.
Then the solution to the Cauchy problem can be written as
(2.14)
2.4.1. The Case Where,x = O. In the case where ,x = 0 we have to solve
0(-2)K
{.ih
II
s(i,h)rih
0(-1)8
+ (.ih ' P 0
s(i,h)rih + qS(i,h) nrih = ,
2
or, what amounts to be same thing,
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
l-2 2 .1,1' 0-1 2 , 2 .1. 0

s(i,h(Ys(i,h) 'Pih - (.s(i,h)qs(i,h)Wih - qS(i,h) 'Pih = .
This implies
Uih(X) = UihO + x(uthO - UihO).
Furthermore, w~h(x) = -lS(i,h)1/J~h(X) implies 1/J~h(x) == O. Hence
1/Jih(X) = 1/Jiho + x1/Jih1 + x 21/Jih2

for some 1/Jih2. This implies
W~h (x) = -1/Jih1 - 2X1/Jih2'

The identity (wth)"(X) = w~h(l - x) implies
1/Jih2 = ~ (1/J:h1 - 1/Jiht).
However, we also have 1/J;h(x) = 1/Jih(l - x) and (1/J~h)*(x) = -1/J~h(l - x),

which implies that 1/J;h1 = 1/Jih1. Thus, 1/Jih(X) is reduced to
1/Jih (x) = 1/JihO + X1/Jihl,
which implies
But then
and hence
Wih(X) = _~X2(1/J:hO -1/JihO)ls(i,h) + Wih1 X + WihO.

Again, by reflection at 1 we deduce 1/J:hO = 1/Jiho and Wih1 = w;hO - WihO
This reduces 1/Jih(X) to a constant
1/Jih (x) = 1/JihO = 1/JthO
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
Uih(X) =UihO + x(u;hO - UihO),

Wih(X) =WihO = w;hO'
'l/Jih(X) =0.
As Uih(X) is the only variable with a non-vanishing derivative, the balance

equation reduces to
L dis(i,h) (P~(i,h) U~h (O)es(i,h)) = 0,

h
which turns out to be equivalent to
L dis(i,h) (P~(i,h)u;hOeS(i,h)) = L dis(i,h) (P~(i,h) UihOes(i,h)).

h h
By the same argument as in [106] we deduce u;hO = UihO, which then

reduces Uih(X) to a constant. This shows that zero is an eigenvalue if and
only if no vertex in av (G) is clamped.
2.4.2. The Case Where >. Belongs to an Individual Beam. Suppose we

keep all the nodes fixed, i.e., CPih = o. Then we can use the reflection
formula r;h(O) = rih(l) in order to obtain
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".
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 /\~2 (1 - COS(JL1) COS(JL2 + 2), ~

).2 - 1
sin(JLd sin(JL2).
The roots of the equation det(Vl) = 0 can be calculated to be asymptot-

ically equal to the integer multiples of 7r. Again, the fact that they are
not exactly (possibly non-integer) multiples of 7r reflects the phenomenon
of dispersion of bending and shear waves in this model.
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
Zr~(>" x) zt~()., x) s(i,h) tPihO s(i,h)
Then, by the reflection property, we have the identity
(2.15)
which, in matrix form, is
By assumption, the matrices 8}h are invertible. Therefore, we obtain for

\l1 ih the equation
(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.
We collect the results in a theorem. In contrast to the first theorem, we

dispense with the characterization of the corresponding eigenstates. Those
can be seen from the preceding discussion.
THEOREM 2.2. If A2 is an eigenvalue of (2.8), then A satisfies one of

the following conditions:
(i) A = 0 (iff 8oV(G) = 0).
Ul
(ii) A is a solution of h = 0 or of det(~~) = 0 for some pair i, h.
(iii) A is the solution of (2.16).
2.5. Homogeneous Timoshenko Networks. In the case of the com-

pletely homogeneous network all quantities det(~~), zf,!, Wi,:
are indepen-
dent of i, h. We have
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),
ZOl(A,x) = - 2~ COS(JL1X) + 2~ COS(J.L2X),

Zl1(A, x) =_1_ sin(JL1) + _1_ sin(JL2x),
21'1 21'2
In the situation under consideration equation (2.16) can be rewritten in

terms of the projections P s(i,h)' P~i,h)' P n as
"{ 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
+ d~t (-ZOlNs(i,h)<Ph + W01Pn<Ph)}

" { U OP 1 (WOO Zll Wll ZOO)p.l
= L...J Ul s(i,h) + det - s(i,h)
h
+~
det
(W lO Z11 _ Wll ZlO)M. + ~ (ZOOW 01 _ ZOI WOO)N .
s(t,h) det s(t,h)
+ d~t (ZOlW lO - ZlOW0 1)Pn}<Pi + (L diS(i,h)MS(i,h)) <Pi.

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
J-Ll = ..;'>.2 + )., J-L2 = ..;'>.2 - >..

Consequently J-Lk ~ >. for large >.. We find, therefore, the asymptotic for-
mulas
det ~ ;2 sin2 >.,

ZOO(>., 1) ~ 0, ZlO(>., 1) ~ cos(>.) ,
ZOl(>.,l) = o(~), Z11(>., 1) ~ ~ sin(>.),

W OO (>', 1) ~ cos(>.), W lO (>., 1) ~ 0,
W0 1 (>., 1) ~ ~ sin(>'), Wll(>., 1) = o( ~).
We multiply equation (2.17) by det 1/ 2 = (sin >')/ >. to obtain the asymptotic
equation
(2.18) L eih {Ps(i,h)<Ph + P~i,h)<Ph + P n<Ph}

h
= L eih{UO()., l)P s(i,h) + coS(>')P~i,h) + cos(>')Pn}<Pi.

h
3. NUMERICAL SIMULATIONS 167
Equation (2.18) obviously reduces to
(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.
3. Numerical Simulations of Controlled I--d Networks

3.1. Introductory Remarks. We do not attempt to give an overview
of numerical methods in the general context of multilink flexible structures.
There is a vast engineering literature on this topic. Instead, we want to con-
centrate on the very specific problems treated in this monograph. Indeed,
we shall only discuss one dimensional links and leave the higher dimensional
case for future study. Within the context of one dimensional links such as
strings and beams, we focus on the simulation of the transient behavior of
the solutions. From the point of view of open loop control concepts such
as approximate, spectral or exact controllability, it is exactly the transient
behavior which is subjected to controls. This is also true in the context
of stabilizability by means of feedback controls. In particular, the total
energy as a function of time is the object of interest. We therefore need
numerical schemes which conserve the total energy of the approximated
system as long as no controls are applied to the system. That is to say, we
have to rule out methods which exhibit algorithmic damping. In addition,
the frequencies of the approximating system should be close to those of the
original problem or, in other words, there should only be a small amount of
numerical dispersion. These requirements, taken together, are very restric-
tive if large structures are involved. In this respect, numerical simulations
of the kind considered here are quite different from those pursued in the
engineering sciences as far as large flexible structures are concerned. In the
engineering context one is not interested in energy conserving methods but
rather in schemes guamnteeing a certain degree of algorithmic damping.
This is due to the fact that the generalized eigenvalues corresponding to
the finite element approximation (which is used almost exclusively) usually
differ quite substantially from the corresponding eigenvalues of the original
system if the number of modes considered is large. In particular, the largest
eigenvalues of the finite element approximation sometimes have little to do
with the spectrum of the original problem. Therefore, damping of high
modes is a sought after property of a numerical scheme. Also, because the
condition numbers of the condensed stiffness and mass matrices literally
explode with the number of modes, iterative schemes are mostly used to
solve the corresponding algebraic problem after complete discretization.
As we are mainly interested in giving numerical support to theoretical

results which have been obtained in the preceding sections and because
of the above requirements and their natural consequences, we will discuss
examples of low dimension, that is networks with only a small number of
flexible elements (say ~ 200 DOF). As a result, we do not have to worry
too much about CPU time and we can resort to classical techniques which
have been described in the literature. The schemes which we will consider
are:
The (~, ~)- Newmark scheme applied to the semi-discrete equations
resulting from the finite element procedure.
The Lax-Friedrichs scheme for the fully discretized finite difference
approach.
High order implicit Runge-Kutta schemes for semi-discretized sys-
tems and differential-algebraic systems.
The calculations have been carried out on a 486MS-DOS PC and on a
HP 9000 Apollo work station using MATLAB and also FORTRAN codes.
We are not going to be any more specific about source codes or CPU times.
3.2. Networks of Strings.
3.2.1. Absorbing Controls. In order to give an exemplaric introduction,
we will consider the numerical realization of stabilizing feedback controls.
These controls are of more practical interest than open loop controls. In
particular, the feedback controls we discuss are of the absorbing type. They
can be viewed as being at the interface between controllability concepts
and stabilizability concepts. Loosely speaking, an absorbing control is one
which passes the entire energy flux in one direction. The history of these
controls goes back to the beginning of control theory of distributed pa-
rameter systems and is originally connected with D.L. Russell; see [92] for
an excellent review. From the mathematical viewpoint it is most easily
explained for the I-d wave equation in its characteristic description as in
[92].
EXAMPLE 3.1. The l-d wave equation.
Let us write down the scaled classicall-d-wave equation as follows:
W(X, t) = w"(x, t), (x, t) E (0,1) x (0, T),

{ w(O, t) = 0, w'(I, t) = ,
t E (0, T),
w(x,O) = wo(x), w(x,O) = Wl(X), X E [0,1].
Under the transformations
we have the first order system
v=rv', r= (~1 ~),

Vl(O) = V2(0), vl(l) + v2(1) = ../2.
In this simple case the total energy and its time derivative are given by
It is obvious that the optimal choice of a control is the one leading to

v2(1) = 0 for all t ~ O. In fact, it is readily seen that this choice amounts
to canceling the outgoing signals at the boundary point x = 1. This means
that the control mimics the response of the half-infinite string. This is Rus-
sell's philosophy in [92]: "let nature take its course". In tenns of feedback
controls the goal of canceling outgoing signals can be achieved by letting
(t) = vl(l, t)/V2, which is equivalent to the closed loop boundary condi-
tion w'(I, t) = -w(l, t) which, in turn, is a continous system analogue to
the classical dead beat control. It is also possible to interpret this in tenns
of an open loop control as follows:
(3.1) (t) = ~ {w 1 (t) - wb(t) t E (0,1),

2 -(wb(l-t)+Wl(t)), tE(I,2).
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'))'
for a suitable function (7 such that
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)
The boundary conditions are represented by
(3.3) {
J +1
vo = u J +1
'n =
,
u Jo+ 1 -- u J1 + q . v J1> vJ+1

n = vn-l
J+ 1 _ q . u Jn-l'
The Courant-Friedrichs-Lewy condition requires that q ::; 1; we always

use q = 1. This setup is particularly useful if the boundary feedback above
is to be implemented, because it prescribes the strain at the boundary
x = 1. We simply have to set u~+1 = -v~+1. For further information,
in particular on other realizations of absorbing boundary conditions , see
Halpern [59].
We first look at the uncontrolled string. We take 45 spatial mesh points
and 225 time steps, which amounts to time span of [0,5]. As initial con-
ditions we choose w(x,O) = sin(31l'x2 ), 1iJ(x,O) = O. The reason for this
choice is that we do not want to look at eigenmodes but rather at the
response to sufficiently rich data; see Figure V.B. Note that we display
the evolutwn of stmin, first because of the choice of variables (indeed we
could have integrated the strain in order to obtain the displacement), but
secondly because the strain is subjected to controls.
We also look at the contour plot for the uncontrolled motion; see figure
V.9. This plot clearly shows the propagation along the characteristics.
The total energy, which is also easily computed numerically using the
trapezoidal rule, is displayed in Figure V.lO.
We are now going to apply the absorbing boundary control at the bound-
ary x = 1. The result is shown in Figure v.n
The total energy is displayed in Figure V.12.
These plots serve as evidence for the remarks above. The numerical
procedure can be (and has been) applied also to planar strings and planar
Timoshenko beams. For similar simulations using, however, the classical
direct discretization of the wave equation, see J. Schmidt [97]. 0
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
FIGURE V.8. Uncontrolled motion of a string
20
15 -
10
5
I
l~-,--~ ---'----'....... ___._~~_
20 40 60 80 100
FIGURE V.9. Contour plot

Total Energy
29.347~--~--~---~--~---~--~
14.6731-------------------
oL---~--~---~--~---~---
o 2 3 4 5 6
FIGURE V.lO. Conservation of energy
FIGURE V.ll. Absorbing boundary control

Total Energy
l6.---~----~----~----~----~-----,
14
12
\0
OL-----~--~----~----~----~----~
0 4 5 6
FIGURE V.12. Energy plot of disappearing solution

the finite element section. The first multilink system we look at is a system
with three strings coupled together at a multiple node of vertex degree 3.
EXAMPLE 3.2. Three strings.
We do not write down the equations of motion, the boundary and node
conditions for this model again. Rather, we just want to show the numer-
ical simulation which is performed as in Example 3.1. Two situations are
considered:
(a) two exterior nodes are clamped, the remaining one is controlled;
(b) one exterior node is clamped, one is free and the remaining one is
controlled.
We label the strings such that the first and the second string are always
uncontrolled, while string #3 is subjected to an absorbing control at its
end as described above. We have put the initial displacement into the first
string. See Figure V.13 for the first case and Figure V.14 for the second
case.
Figure V.13 clearly reveals the fact that the system is not even approx-
imately controllable: energy which initially is located in the first string
enters the second and the third string through the multiple node. Then,
energy is absorbed at the boundary of the third string. Indeed, energy is
absorbed there until the motion in strings #1 and #2 settles to a residual
motion which vanishes at the multiple node.
Figure V.14 shows to what extent the previous situation is sensitive
to changes in the boundary conditions. This case parallels the two-string
model with in-span control discussed by Chen et al [151. There it was
shown that for two Dirichlet (Neumann) boundary conditions and control
at the mid-span, one does not have approximate controllability, while if
~_ _---'..:'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
FIGURE V.13. Case (a)
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
FIGURE V.14. Case (b)

5~____~I~st~S~tr~inh-g~__~ 2nd String
(\
2.5
(\ r'\ f\
4 2 \ (\
3 1.5
0.5
0 \1 \/ \;
10 0 5 10
3rd String 4th String

5 2.5
(\ (\ f\ (\ (\ (\
4 2
3 1.5
0.5
0
; J. \ 0 JJ J V
0 5 10 0 5 10
FIGURE V.15. Transfer of energy through multiple joints
one boundary condition changes from Dirichlet to Neumann (Neumann to

Dirichlet) exact controllability results. In our three string model such a
mathematical analysis does not seem to be known. The numerical simula-
tions, however, strongly support an affirmative conjecture. 0
EXAMPLE 3.3. The unit square.

We now turn to yet another example, which we have seen before: the unit
square. We know that this system is not approximately controllable with
controls at the joints. But in this example we do not dwell on numerical
evidence concerning controllability, which we defer until the finite element
section. Instead, we want to use this model to demonstrate the correctness
of our modeling. We consider the unit square with initial displacement in
the first (bottom) string. As in the previous cases, we take 20 spatial mesh
points for each string. Figure V.15 nicely represents the physically evident
flow pattern: motion penetrates from the bottom string to the two parallel
upright strings #2, #4 (with half of the energy in each) and crosses into
the top string #3 (which had been at rest for one time unit). At time 2 the
energy is concentrated entirely in the top string, and the reverse pattern
takes place. Rage See Figure V.15 0
3.2.2. Directing Controls. We pursue the concept of absorbing controls

a bit further and consider now three identical strings and out-of-plane dis-
placement. The fact that we concentrate on this model, rather than on the
planar system, is only for the sake of a convenient display of the plots. We
will consider planar systems later.
EXAMPLE 3.4. Three serial strings.

We have the following system:
iii = (wi)" on (0,1) x (0, T),

(3.4) { w1(0) = 0, w 1(1) = w2 (1), (w 1)'(I) + (w 2 )'(I) = II,
w3(1) = 0, W2(0) = w3(0), (w 2)'(0) + (w3)'(0) = j2,
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)' ,
With these transformations we are able to write the system as
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
as in Russell [92]:
(Zl) (0) = (10

Z2
Z3
0 0
1 0)
001 (Z4)
Z5
Z6
(0),
(3.6)
Z4)
( Z5
~
(1)= (0-1001
0100) (Zl) Z2
~
(1).
We consider a particular path of characteristics. For t = 0 we start

with the three outgoing signals Zl,
Z2, Z3 moving to the boundary x = 1
with characteristic speed equal to 1. These characteristics are exactly the
incoming signals at that boundary. According to the reflection condition
(3.6) at x = 1, Z4 is canceled at x = 1. Now, the triple Z4, Z5, Z6 constitutes
the set of outgoing signals at x = 1 and, accordingly, the set of incoming
signals at x = o. As Z4 is already equal to zero at x = 0 (because of
properties of characteristics), the set of nonvanishing outgoing signals at
x = 0 reduces to the variable Z2, according to the reflection condition (3.6).
Therefore, after reflection at x = 0 the only nonvanishing incoming signal
at x = 1 is Z2 which, in turn, is reflected to Z5 as the only nonvanishing
outgoing signal at x = 1. This, however, closes the cycle, because Z5
is converted to Z2 upon reflection at x = 0, and so on. We depict this
situation in the following diagram:
(0) (0) (0) (0)

(
=- : -
Zl) at x=l at x=o
~ -
at x=l
~
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:
FIGURE V.16. Directing controls
Total Energy
21
20
19
18
17[
1
16
15
::t
f I
0 2 3 4 5 6
FIGURE V.17. Energy of entire system

FIGURE V.1B. Effect of time delay
which results in the feedback controls
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
Total Energy
140,--~---~--~--~--~------,
120
r
100
80
l
60
i
401
I
i
20'
Ol----
o 4 6
FIGURE V.19. Instability due to time delay

classical work of Haraux [35], [36] and also to Cabannes [12], [13]. Without
going into the theoretical detail, we present the response of the system (3.4)
without time delay but with respect to the nonlinear boundary controls:
/1 = -sgn(w 2 (1)) + (w 2)'(1), /2 = sgn(w 2 (O)) + (w2)'(O)
See Figures V.20 and V.21. It is apparent that these controls lead to
considerably improved decay to a residual motion in the second string. It
is interesting to observe the pattern of the strain in that string. If we now
take these nonlinear boundary controls and if we account for a small time
delay in the control (say one time step), then no instability is observed.
See Figure V.22 and Figure V.23. Whether this observed robustness is
actually true for the infinite dimensional system appears to be an open
problem. The kind of nonlinearity of boundary controls just considered is
in fact derived from a physical principle, namely dry friction. We will come
back to this point later when we give an example of implicit Runge-Kutta
methods applied to differential algebraic equations. 0
In the next example we want to demonstrate how directing controls can
be used to improve the system response by "directing" the energy to a
"safe" part of the structure. This is of great practical importance because
large flexible structures usually contain many uncontrollable circuits. Even
if one puts the classical dissipative controls at each node, there will be
a residual vibration in the system which will not decay to zero. Use of
"directing" controls, however, can serve to "channel" the energy through
the network to a particular set of nodes where either the energy influx is
not important or where incoming energy is more easily absorbed.
EXAMPLE 3.5. Five strings containing a circuit.
FIGURE V.20. Bang-bang control
Total Energy
16"----~------,_----__----~----~----_,
14
12
10
FIGURE V.21. Improved dissipation

FIGURE V.22. Robustness of bang-bang control
Total Energy
14
12
10
:l
J
OL-----~----~----~------~----~-----
o 4 6
FIGURE V.23. Nonlinear dissipation with time delay

4. FINITE ELEMENT APPROXIMATIONS 183
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
4. Finite Element Approximations of Timoshenko Networks

We consider a 3-d network of isothermal Timoshenko beams without
warping. We recall that the total strain energy (2.3) reflecting the addi-
tional restrictions is given by
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
1st Strin 2nd Strin

5 2.5
4 2
3 1.5
0.5
0 0
0 5 10 15 20 0 5 10 15 20
3rd Strin 2.5 4th Strin

4
2
3
1.5
2
0.5
0 0
0 5 10 15 20 0 5 10 15 20
FIGURE V.24. Energy in individual strings
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
FIGURE V.25. Energy in 5th string and entire system

1st Strin!! 2nd String

2.5
:0 2
(\
3 I.S
0
\ o,SII
0 \'"
0 2 4 6 8 0 2 4 6 8
3rd Strin!! 4th String
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=
together with their barred counterparts, we arrive at the standard problem

(4.1) a(ii, u) = (ii, f) + (ii, h)r,
where the form a(,) and the other products are given by
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
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,
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
Similar arguments lead to the representation of the element mass matrix.

In most of the calculations we used a lumped mass matrix and linear shape
functions such that a one-point Gauss quadrature exactly integrates the
bending stiffness and appropriately underintegrates the shear stiffness to
avoid shear locking. Also we use a so-called residual bending flexibility to
update the effective shear area according to
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)
for the acceleration, velocity and displacement, respectively. We then define

the predictors by
_ 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
Mao = F - CVo - Kdo
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
dn+1 = ~n+1 + .8t:..t2~+1'

(4.6) {
v n+1 = v n+1 + ')' t:..t~+1.
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
the choice f3 = 1/4, ,=

ways and lead to different properties of the scheme. We will always make
1/2. This choice is known to lead to conservation
of the energy, as discussed above.
In the calculations which follow we confine attention to planar frames
of Timoshenko beams. In that case the local stiffness matrix k e is a 6 x
6 matrix since there are vertical and longitudinal displacements and one
rotation in the plane for each of the two nodes of a single element. For the
sake of completeness we write k e out explicitly:
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
As for the lumped mass matrix we take

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
EXAMPLE 4.1. A single planar Timoshenko beam.

In this example we want to consider very briefly a single planar Timo-
shenko beam which is clamped at x = 0 and subject to controls at the free
end at x = 1. We implement a dissipative feedback with impedance 1 in
the shear and axial forces as well as in the rotation. The system is known
to be exponentially stabilized by usual velocity type feedback controls. Be-
cause of dispersion of waves, however, exact cancellation as in the string
case does not take place. Questions which naturally arise are whether the
feedback gains can be chosen in an optimal way and whether an optimal
feedback may also contain the full state feedback rather than a pure veloc-
ity feedback. To give at least a little insight into these interesting issues we
compare, without going into detail, the effects of typical absorbing controls
versus those of an LQR-optimal control. In fact, we use as a cost functional
the total energy plus the L 2 (0, 00)- norm of the control. We computed the
LQR-feedback operator using the MATLAB control toolbox; see Figure
V.28. It is interesting to note that the effects of LQR-controls are almost
identical to those obtained using classical feedback controls. 0
While it is fairly simple to set up finite difference schemes, such as the
Lax-Friedrichs scheme, for planar systems of multi-link strings and Tim-
oshenko beams as long as the number of structural elements is small and
the graph is very simple, this method is unsuited for more complicated
networks, for which the finite element method seems more appropriate.
As remarked earlier, we consider our numerical simulations as tutorial
and, therefore, we restrict our attention to some simple models which are
nonetheless rich enough to reveal interesting properties.
EXAMPLE 4.2. The unit square of planar Timoshenko beams
We first consider the uncontrolled system. Again there is an initial
displacement only in the bottom beam (# 1). The difficulty in exhibiting
5. DRY FRICTION 191
Total Energy
3.5
2.5
1.5
0.5
-. = absorbmg control; - = LQR-control
FIGURE V.2B. A comparison of controls

graphically all of the displacements and rotations is obvious. At the nodes,
local vertical motion of beam # 1 converts to local longitudinal motion of
beams #2 and #4 which, in turn, is converted to local vertical motion in the
top beam (#3). Because in the finite element approach we automatically
have the local rotations r i available, we can easily go back to the global
reference frame. In this way the locally longitudinal motion of the 2nd
and 4th beam is, indeed, globally vertical; see Figure V.29. This plot
clearly shows the collision of waves in the 3rd beam after travelling safely
through beams #2 and #4. The initial bump next to the corner at the
bottom indicates the location of the first beam. Figure V.30 shows the
conservation of energy.
In the second experiment we apply dissipative controls at all four nodes;
see Figure V.31 and Figure V.32. Figure V.31 clearly shows how the motion
settles to a residual oscillation which vanishes at the vertices. Accordingly,
the energy settles to a constant value; see Figure V.32.
5. Implicit Runge-Kutta Method: Dry Friction at Joints

In deriving the semi-discrete approximation to the infinite dimensional
distributed parameter system by the finite element approach in the last
section, we have used a condensed description in the sense that only inde-
pendent nodal variables have been used. This means that, for instance at
a multiple node where, because of the continuity requirement, some of the
displacements of one structural element can be expressed by corresponding
FIGURE V.29. Unit square: uncontrolled motion
Total Energy of Entire System

1.2
0.8
0.6r
0.4
0.2
o
o 2 3
FIGURE V.30. Unit square: conservation of energy

5. DRY FRICTION 193
FIGURE V.31. Unit square: controlled motion
Total Energy of Entire System
0.8
0.6
0.4
0.2
OL---~----~----~----~--~----~----~--~
o 2 4 5 6 7
FIGURE V.32. Unit square: saturation of energy

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
{ ill - w" = -4sgn(tiI) if til:/: 0,

lill - w"l :::; 4 if til = 0.
This system is complemented by the boundary and initial conditions
w(O, t) = w(l, t) = 0, w(x,O) = wo(x), w(x, 0) = WI(X).

Cabannes showed that if WI == 0, wo(O) = wo(l) = and wg(x) :::; 4 for
all x E [0,1], the string does not move. He also studied cases in which the
initial curvature is beyond the value 4 and he showed that if the curvature
5. DRY FRICTION 195
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
iD - w" = -PsX(x)sgn(w - v) if w ::I v,

(5.1) {
liD - w"l ::; PaX(x) if W = v.
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
on both strings. We assume clamped boundaries at x = for both strings

and also assume dry friction at the common point x = 1. The resulting

consider two identical strings of length one joined at the ends where x = 1
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
UI := WI, U2:= W2, U3:= (WI)', U4:= (w 2 )',

1 1
VI := v'2( -UI + U3), v3:= J2(UI + U3),
1 1
V2 := yI2( -U2 + U4), V4:= y'2(U2 + U4).
We then have the system

v = rV', r = diag([-l, -1, 1, 1]),
with boundary and node conditions
(V3 - vd(O) = 0, (V3 - vd(l) = (V4 - v2)(1),

(V4 - V2)(0) = 0, (V3 + vd(l) + (V4 + v2)(1) = -Ps V2z
with
_ {~(V3 - vI)(l) if (V3 - vd(l) # 0,
z - E [Pa
- - , -Pal I
ese.
Ps Ps
The last set of equations can be rewritten as
(V2VI) (0) = (V3) (0),

V4
(V3) (1) = _ (0 1) (VI) (1) _
V4 1 0 V2
Ps~z
V2 (1)
1
.
As a consequence of our special choice of initial conditions (we could con-

sider more general data but wish to treat this problem only as an example),
we have V2("0) = V4("0) = O.
We will consider two cases:
(i) I(WI)'(.,O)I ~ Pa/V2;
(ii) I(WI)'(.,O)I > PaV2.
We claim that in case (i), no motion takes place in the second string for
t:2: O.
PROOF. We have
(V3 - vd(l) = 0 at t = O.
Take
z= - V2 vI(l, t) E [_Pa, Pal,
t E [0,1].
Ps Ps Ps
We follow the reflections at x = 1 and subsequently at x = 0, i.e.
v3(1, t) = -v2o(1 - t) + vlO(1- t) = vlO(1- t),

v4(1, t) = 0, t E [0,1],
VI (0,1 + t) = vlO(1- t),
v2(0,1 + t) = O.
Hence, upon reflection, Vi does not change magnitude and, with z as above,
V4 remains at zero, causing V2 to also equal zero.
The analogous case starting with V30, V40 and traveling first towards the
boundary at x = 0 is settled in a similar way. The main point is that the
variable z (representing the adhesive force) always compensates the shear
5. DRY FRlCTION 197
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.
PROOF. We have initially (V3 - vt}(I,O) = 0. Assume now that
(V3 - vt}(I, t) = 0, t E (0, e).

We have
z(t) E [_Pa, Pal =} z(t) =J - v'2 V1 (I,t) on (O,e)

Ps Ps
=} v4(1, t) =J on (0, e) =}
Ps
V2(0, 1 + t) =J 0.
The latter inclusion says that there is nonzero motion in the second string.
The other cases are handled in a similar way. 0
We are going to give numerical support to these remarks. We consider

the mesh points Xi = i/N, N = 40, and use a classical finite difference
approximation to the second order operator (which, incidentally, equals
the corresponding finite element stiffness matrix). Indeed, we can view
this process as occuring in one string with dry friction at the midpoint.
We have the following system:
(5.3)
where now z is described by
(5.4) { z=Sgn(W~(t)) ifw~(t)=JO,

psz E Pa[-I, 1] if WIt.(t)
2
= 0.
We also have the boundary and initial conditions
In the above system the variable z can be viewed as a control variable

which forces the solution of the system (5.3) to satisfy Wit.2 = 0. In order
to do that the variable z, which physically represents the adhesive force,
must vary in time. In order to make the role of z more transparent, we
consider a general differential algebraic equation of the following kind:
Y/ = f(y, z),
{
0= g(y),
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
Modeling and Controllability

of Interconnected Membranes
We shall consider in this Chapter the motion of an elastic body consisting

of several interconnected elastic membranes. Our goals are to obtain the
dynamic equations of motion of such a configuration, especially to eluci-
date the geometric and dynamic conditions which must hold in junction
regions where two or more such elastic elements are joined together, and
to study the controllability properites of the resulting (linearized) system
of equations. The modeling proceeds through two stages. In Section 1
the equations of motion and boundary conditions of a general nonlinear
membrane and the corresponding Hamilton's principle will be derived, as-
suming a linear stress-strain relation. The dynamic equations of a system
of interconnected membranes are derived in Section 2 by minimizing an ap-
propriate Lagrangian over the class of deformations which satisfies certain
geometric constraints. In particular, the geometric constraint imposed in
the junction regions is the very natural one that the initially connected sys-
tem of membranes remains connected throughout the deformation process.
When this constraint is imposed on the admissible deformations, the opti-
mal motion is forced to satisfy a second, dynamic constaint in the junction
regions that has the interpretation of a balance of forces law there. Both
the equations of motion and the dynamic junction conditions involve non-
linear couplings among the components of the displacement vector. When
these are linearized around the trivial equilibrium the familiar equations of
motion of a single elastic membrane, in which in-plane displacements are
not coupled to transverse motion, are obtained for the displacement of each
individual element. However, all displacements remain coupled through the
geometric and (linearized) dynamic junction conditions. The question of
exact controllability of this linearized coupled system is then considered in
Section 3.
Let us first fix the notation.
202 VI. INTERCONNECTED MEMBRANES
Roman indices i, j, ... take values 1,2,3.

Greek indices a, /3, ... take values 1,2.
Summation convention is assumed unless explicitly noted other-
wise.
A vector x E IR3 has coordinates (Xb X2, X3) with respect to the
natural basis of IR3.
1. Modeling of Dynamic Nonlinear Elastic Membranes

Let n be a bounded, open, connected set in IR2 with Lipschitz continuous
boundary r, and let al,a2,a3 be a fixed, right- handed orthonormal basis
for IR3. We consider an infinite elastic cylinder in IR3 whose reference
configuration is
where Po is a fixed vector in IR3. We set
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
Hypothesis (1.1) means that the deformation of a cross-section orthogonal

to the axis of the cylinder is independent of the variable X3 along the axis of
the cylinder. Thus the deformation of the cylinder is completely determined
by the deformation of any of its cross-sections. The latter is specified by
the vector function W which measures the displacement of a cross-section.
Let {Wi} denote the components of W in the {~} basis: W = Wi~. We
refer to a cross-section as an elastic membrane and to the planar region
as the reference cross-section.

Set G i = R,i, where a subscript after a comma means differentiation
with respect to the indicated spatial variable. The components ij of the
strain tensor are given by
(1.2)
1. NONLINEAR ELASTIC MEMBRANES 203
It follows from (1.1) and (1.2) that
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
(1.5) ti,... +F = pR,tt,
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
In order to be consistent with our basic kinematic hypothesis, it is assumed

that F and p are independent of X3. Let us further assume that the material
in question is linearly elastic (hookean), i.e.,
(1.7)
where the coefficients of elasticity C ijkl depend only on Xl and X2. The
symmetry of sij and Ckl requires that
(1.8) C ijkl = C jikl = C ij1k ,
and we further assume 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.
1.2. Edge Conditions. Let A denote the two-dimensional boundary

of the reference cross-section P and assume that
(1.11 )
where Aa, AD are mutually disjoint, relatively open subsets of A and AD i=
0. (The superscripts C and D are used to indicate "clamped" and "dy-
namic," respectively.) The partition (1.11) induces a similar partition of
r:= 8n into
(1.12)
through the mapping
r ~ A: (XbX2) ~ Po + Xaaa'
Let (Vb V2) denote a unit exterior normal vector and (-V2' vd a unit tan-
gent vector at a point (Xb X2) ofr. Since the mapping (Xl, X2, X3) ~ Xi~ is
orthogonal with determinant unity, v := vIal +V2a2 and T := -V2al +vla2
are vectors which are normal and tangent, respectively, to A at the point
Po + Xaaa with v pointing out of P and T positively oriented with respect
to a3, i.e., T x a3 = v.
The edge conditions along the part of 8B corresponding to Aa are de-
scribed by
(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) loT In (pW,t . W,t - SajWj,a - sa{3W,{3 . W,a) dn

+ fT f F. WdO + fT f f. Wdr = O.
lo In lo lrD
Conversely, ifW satisfies (1.15) for all such test functions W, then W sat-
isfies (in an appropriate sense) (1.10) and (1.14). The variational equation
1. NONLINEAR ELASTIC MEMBRANES 205
(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)
where (8Cij' W) denotes the value of 8c ij at W. For any material obeying

(1.7)-(1.9) one has
(1.17)
Under assumption (1.1),
(1.18)
It then follows from (1.16)-(1.18) that (1.15) may be written
(1.19) 81 T
[K:(t) - U(t) + W(t)] dt = 0,
where
(1.20)
is the kinetic energy of the deformation,
(1.21)
is the strain energy, and
(1.22) W(t) = { F W dO + { f W dI'

1n 1rD
is the work of external forces. The variational formulation (1.19) with K:, U
and W so defined, is Hamilton's principle for the membrane problem under
consideration.
2. Systems of Interconnected Elastic Membranes

For i = 1, ... ,n, let at, a;, a~ be a right-handed orthonormal system
in IR 3, let ph be a fixed vector in IR3, and ni be a bounded, open, con-
nected region in IR2 with Lipschitz boundary consisting of a finite number
of smooth simple curves. We set
(2.1) Pi(XI, X2) = p~ + LXo<a~,

0<
ri(XI, X2, X3) = Pi(Xl, X2) + X3 a ;,

Pi = {Pi(Xl,X2)1 (XI,X2) E nil.
The mapping (Xl. X2) 1-+ Pi (Xl. X2) is a homeomorphism of Oi onto Pi and
of ani onto BPi, with
Pi-1 (P) = ((P - i (P - Poi) . a2i) .

Poi) . ai'
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
The set Pi is considered as a reference cross-section of an infinite cylinder

whose reference configuration is
We make the following hypotheses regarding the membranes Pi.
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
2.1. Geometric Junction Conditions. We tentatively define a joint,

or junction, of the multi-membrane system as a component (a maximal line
segment; by definition, line segments are connected) which is a common
edge of two or more reference cross-sections, that is, a maximal line segment
where a Pi and Pk intersect for some k > i. However, this definition allows
the possibility that the set of all such components does not form a mutually
disjoint collection. If C 1 and C 2 are components such that C 1 n C 2 =I- 0, we
replace C 1 and C 2 by C1 nC2 , C 1 \ (C1 nC2 ) and C 2 \ (C1 nC2 ), whose union
is C 1 U C 2 Each set Co. \ (C1 n C 2 ) is either a single line segment or the
union of disjoint segments. In the latter case we consider Co. \ (C1 nC2 ) to
consist of several separate components. In this manner one may eliminate
the overlapping of components. Finally, we define the joints of the system
to be the collection of disjoint, open line segments which are the interiors
of the collection of non-overlapping components.
EXAMPLE 2.1. Let PI. P2 , P3 be the planar regions in lR3 defined by
P1 : < sb 82 < 1, 83 = 0,
0
P2 : 0 < 82,83 < 1, 81 = 0,
P3: 81 = 83, 0 < 81 < 1, 1/4 < 82 < 3/4.
To write these in the parametric form used above set
fh = O2 = {(XbX2)! 0< XbX2 < I},

03 = {(X1,X2)! 0 < Xl < 1, 1/4 < X2 < 3/4},
a~ = (1,0, 0), a~ = (0,1,0),

a~ = (0,1, 0), a~ = (0,0,1),
a~ = (1,0, 1)/h, a~ = (0,1,0).
The membranes intersect in components Pin P k given by
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.
Note that C2 \ (C1 n C2 ) = 0. Since C2 = C3 , the same process eliminates

C3 as a joint. Finally, since 61 contains two disjoint segments, the joints
of the multi-link system are taken to be
J1 0 < 82 < 1/4, 81 = 83 = 0,

:
J2 : 1/4 < 82 < 3/4, 81 = 83 = 0,

J3 : 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.
Thus At consists of those joints of the multi-membrane system which be-

long to Ai. We label the entire collection of joints by Jl, J2 , ,Jm .
The total kinetic energy of the system is defined to be
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
Let Fi denote an external body force acting on the i th membrane Pi,

fi be a force acting along AP and fk be an applied force along the junction
region Jk. The work of these external forces is defined to be
where
i = 1, ... ,nj k = 1, ... , m,

and where we may take
in accordance with the geometric junction conditions (2.3). In the integrals

over Pi and AP, the integrands are evaluated at pil(p) with p E Pi and
pE Af' respectively, while the integrand in the integral over Jk is evaluated
at P~ (p), p E Jk. The dynamic equations of the multi-membrane system
are postulated to be those associated with the Hamilton's principle
(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
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
(2.8) L>~~a~ + 2:(s~/3W:/3),a +Fi = PiW:w

a,j a,/3
and the boundary conditions
(2.9) 2: v~s~j a~ + 2: v~s~/3W:/3 = fi on rf, t > O.

a,j a,/3
There are, in addition, junction conditions given by the following varia-

tional equation:
for all sufficiently smooth displacements Wi, i = 1, ... ,n, which satisfy
the geometric boundary and junction conditions.
We have
The segment Jk may be parameterized as
where qZ E Jk is fixed but arbitrary. Since Jk = Pi(ri(Jk )), this parame-

terization of Jk induces a parameterization of the segment ri(Jk) given by
given by
both i and k. Therefore
1 r.(Jk)
(Lv~s~ja; + LV~S~{3W:(3)
o:,j 0:,{3
. Wi(x)df
= jbakk (Lv~s~ja; + LV~S~{3W:(3)

o:,j0:,{3
. Wi(x)ds
(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
From (2.3) we have

Wi(pil(p)) = Wik(pi:l(p)), '<Ii E I(Jk).
Since Wik(pi:l(p)) is arbitrary, it follows that
(2.12) L
iEI(Jk)
(Lv~sfja; + LV~Sf{3W:(3)
o:,j 0:,{3
(pil(p),t)
= fk(pi: 1 (p), t), P E Jk

Equation (2.12) is a balance law for forces along the joint Jk and we refer
to it as the dynamic junction condition there.
2.3. Linearization. The nonlinear equations of motion (2.8), dynamic

boundary conditions (2.9) and dynamic junction conditions (2.12) are geo-
metrically exact under the kinematic hypotheses imposed above (Le., (1.1)
and Hooke's Law). It is apparent that their linearizations about the equi-
librium Wi = 0, i = 1, ... ,n, are the same as the linearizations of the
equations
(2.13) Ls~~a;+Fi=PiW:W {Xl,X2)EOi,t>Oj

a,j
(2.14) Lv~sfja; = fi on rp, t > OJ

a,j
P E Jk, t > 0, k = 1, ... ,m.

The linearizations of (2.13)-{2.15) are obtained by replacing the nonlinear
strain components ca{3 in the stress-strain law (1.7) by their linear approx-
imations
(2.16)
Let us consider the case of an elastically isotropic body Bi . Its stress-

strain law is then
si'k = 2J..Li Cijk + Ai "L..J" cU,8
i j k,
l
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)
Substitution of (2.17) into (2.13) gives the linear equations of motion
~[(.\; + I'W!., + 1'>"1.[+ F~ = p;"1.",

(2.18) {
J..Li L W3
a
,ao: + F3 = Pi W 3,tt in 0, t > 0,
where Fj = Fi . a~. The boundary conditions (2.14) linearize to
(2.19) { ~[JLW~(~~'!3 + Wh,a) +D'\iW~,aV~]

I>~ W:j,a I~ on r t > 0,
JLi = i ,
= Ih'
where Ij = fi . a~. The dynamic junction condition (2.15) may similarly

be expressed as a linear equation in the displacements, using (2.17), but
we shall not write it explicitly. The description of the linear system is
completed by adjoining the geometric conditions
.
(2.20) W' = 0 on r ic , t > 0,
(2.21) Wi(pil(p), t) = Wi(pjl(p), t),

P E Jk; i,j E I(Jk); k = 1, ... ,m,
and the initial conditions
(2.22) Wilt=o = W~, W:tlt=o = wi in fl, ,i = 1, ... ,n.

Equations (2.18)-(2.21), and (2.15) with s{k given by (2.17), describe the
linearized motion of a system of interconnected isotropic elastic membranes.
Let us note that the above system is also directly derivable from the vari-
ational principle
CloT [K(t) - U(t) + W(t)] dt = 0,

where K and W are as above and
U(t) ~ ~ t, In, (~8~'w,i. + IIVWiI') dIl,

with V'W~ = Ea W~,aa~ and s~!3 given by (2.17).
2.4. Well-Posedness of the Linear Model. In this section it is
proved that the linear model consisting of (2.13)-(2.15), (2.20)-(2.22) forms
a well-posed problem in an appropriate function space. A linear strain ten-
sor is assumed for each body Bi , which is also supposed to be hookean
with elasticity tensor Crlm satisfying (1.8) and (1.9), but not necessarily
isotropic. We do assume, however, that its elasticity tensor is elliptic: for
every 3 by 3 symmetric tensor cik,
(2.23)
i,k,l,m i,k
for some Co > O.
We begin by recalling the general Korn's lemma of 3-dimensionallinear

elasticity. Let V be a bounded, open connected set in 1R3 with Lipschitz
boundary. We denote by Hk (V) the standard Sobolev space of order k
over V. When k = 0 we write L2(V), as is customary, instead of HO(V).
Let ab a2, a3 be an orthonormal basis for 1R3 and define
H = {W = L Wjajl Wj E L2(V)},
j
with
For W E V, set
it Iv
1/2
( )
e(W) = eJk(W)dV
With this notation, Korn's lemma may be stated as follows.
LEMMA 2.1. There is a constant K depending only on V such that for

allWEV,
(2.24)
Since it is trivial that
e(W) ::; KIIWllv,
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.
REMARK 2.2. A well-known (and easy to prove) consequence of Lemma

2.1 and the compactness of the embedding V ~ H is the following. Let p(.)
be any continuous seminorm defined on V such that e(W) = 0, p(W) = 0
implies W = O. Then
{2.25} IIWllv ::; K[e(W) + p(W)], VW E V.

Therefore, e() defines a norm on the closed subspace {W E VI p(W) = O}.

One may choose, for example,
1/2
( )
p(W) = ~ IIWj 1I~-r8V)o) ,
where r ;::: -1/2 and (OD)o is any nonempty, relatively open subset of OD.
In particular, one has
(2.26) IIWllv ~ Ke(W), VW E V n {W = 0 on (OD)o}. 0

REMARK 2.3. Apropos to the membrane problem, let 0 be a bounded,
open set in JR2 with Lipschitz boundary and define
H = {W = L Wjajl Wj E L2(On,
j
with
Then c33(W) = 0 and co:3(W) = W3 ,o:/2. Define
These definitions of V, H and e() coincide with the previous ones if we

take V to be the cylinder 0 x (0,1) and consider only functions W which
are independent of X3. It is then obvious that (2.24) remains valid with
V, H and e() so defined, as does Remark 2.2 with (OD)o replaced by r o,
where r 0 is any nonempty, relatively open subset of 80 and K depends
only on o. 0
Now for i = 1, ... ,n, let Oi and al,~,a~ be as defined at the beginning
of this section, and set
j
n n
V = II Vi, H = II Hi
i=1 i=1
An element W E V is given by
W = (Wi, ... ,Wn ), Wi = 2: Wja; E Vi,

j
and similarly for H. The norms on H and V are respectively
(2.27)
and
(2.28)
where
IIW;II" = ( ~ IIWill:'.(",) 1/'

With these definitions, V is dense in H with compact embedding.
A continuous seminorm e() is defined on V by setting
(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.
LEMMA 2.2. Assume that rf

=I 0 for at least one index i. Then there
is a constant K depending only on 0 1 , ... ,On such that
(2.32) IIWllv ~ Ke(W), VW E V.
PROOF. Since the map W 1--+ IIWiIlHl/2(rf) defines a continuous semi-

norm on V, it suffices to show that
WE V, e(W) = 0, Wilra, = ==> W = 0.
Without loss of generality, we may assume that rf

=I 0. Then since,
in particular, eI (WI) = 0, it follows from Remark 2.2 that WI = 0.
Suppose that P2 is joined to PI along a joint J. Since WI(PII(P)) = 0,
P E J, the geometric junction condition (2.31) implies W2(p2'I(p)) = 0,
P E J, and then Remark 2.2 gives W 2 = 0. Since the collection of plates
forms a connected set, we may conclude in the same manner that Wi = 0,
i = 1, ... ,no D
Having set up an appropriate function space framework, we may show

that the problem (2.13)-(2.15), (2.20)-(2.22) is well-posed if the data F i , f',
W~, wi are chosen in the proper spaces. Apart from the initial conditions
(2.22), we adopt as the definition of a solution to our problem a solution
of the variational equation
+ f
k=I
r r
T
Jo Jrik(Jk)
fk . Wik dr
,
vW = (WI, ... , wn) E L2(0, T; V), T > 0.
Equivalently, for almost all t > 0,

Equation (2.33) is obtained from (2.13)-(2.15), (2.20), (2.21) by multi-

plying (2.13) by Wi, integrating the product over Oi x (0, T) and using
integration by parts in the usual manner.
One may write
The form
is a symmetric, continuous bilinear form on V and, because of the ellipticity

assumption (2.23) and Korn's lemma,
(2.35)
a(W, W) t
~ eo 12: c!j(W )dO
i=1 n, oj
i
~ 2~21IWII~ - eollWllk, VW E V.
Therefore, [a(W, W) + Co(W, W)k]I/2 defines a norm on V (and even on

V) equivalent to that introduced above. We retopologize V with this new
norm and denote by V' the dual space of V. With H identified with its
dual space we then have V c H c V' with dense and compact embeddings.
Let A denote the Riesz isomorphism of V onto V', so that
(W, W)v = a(W, W) + eo(W, W)H = (AW, W)v,
where (W', W)v denotes the pairing between an element W' E V' and
WE V in the V'-V duality. If W' E H one has (W', W)v = (W', W)H,
vW E V. With this notation, (2.34) may be written
(2.36) (W,tt + AW - CoW, W)v
= t (ln,l Fi .
,=1
WidO + [D fi . Widr)
lr,
m
+ 2: [ fk . wik dr, vW E V.
k=1 lrik (Jk)
The variational equation (2.36) falls within standard variational theory

as found, for example, in [62, Chapter 3]. Let us first consider the case of
homogeneous boundary and junction conditions.
THEOREM 2.1. Assume that
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.
Then there is a unique function
WE 0([0, Tj; V) n 0 1 ([0, Tj; H) n 0 2([0, Tj; V')

which satisfies (2.36) for almost all t E [0, Tj and for which W(O) = W o,
W,t(O) = WI.
The proof is just a matter of observing that
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
the operator may be extended to a mapping H I-t D~, also a homeomor-

phism, through the formula
(AW,W)DA = (W,AW)H, 'VW E H, WE DA.

The variational equation (2.36) may therefore be written in the weaker
t if
form
(2.37) (W,tt + AW - CoW, W)DA = r Widr

2: r
m
+ fk . Wik dr, vW E DA.
k=1 Jrik(Jk)
THEOREM 2.2. Assume that Fi = 0,

fi E L2(0, T; L 2(rp)), i = 1, ... ,n,
-k 2 2
f E L (O,T;L (rik(Jk ))), k = 1, ... ,m,
Wo E H, WI E V'.
Then there is a unique function
WE C([O, TJ; H) n C 1 ([0, TJ; V') n C 2 ([0, TJ; D~)
which satisfies (2.37) for almost all t E [0, T] and for which W(O) = W o,
W,t(O) = WI.
In fact, one has
hence there exists an element g E L2(0, Tj V') such that
Standard variational theory then gives the conclusion of the Theorem.
REMARK 2.4. Consider the homogeneous problem Fi = 0, fi = 0, i =

1, ... ,n, fk = 0, k = 1, ... ,m, and assume that the initial data satisfy
Wo E D A , WI E V.
It is standard that the solution W then has the regularity
A fundamental question is whether or not W is then a classical solution.

The answer amounts to determining the regularity of an element U =
(U I , ... , un) E D A. Such an element is characterized by the variational
equation
a(U, U) = (F, U)H, vir E V,
for some F E H, together with the geometric conditions

u i = on rf, i = 1, ... ,n,
Ui(p;l(p)) = u j (pjl (p)), P E J k , k = 1, ... ,m.
If we introduce the notation

n
(2.38) V/ = {U = LUja;1 Uj E H 8
(Oi)}, V 8 = IIV/,
j i=1
the issue may be phrased as whether or not
(2.39) DA C V 8 n V for some s > 3/2

with continuous injection.
3. CONTROLLABILITY OF LINKED MEMBRANES 221
It is known that even for a single isotropic membrane the validity of

(2.39) depends on the geometry of n and the particular boundary con-
ditions. For example, (2.39) is true for any Lipschitz domain if F
and
I,v are disjoint. If I' n I'D =f:. 0, (2.39) will hold if r C and r D meet in
C
an angle smaller than 7r (measured in the interior of n). In this matter
the reader is referred to Grisvard [30J, [31], and Nicaise [78J. For systems
of interconnected isotropic membranes, from the regularity results for a
single membrane it is clear that there is HS regularity (8 > 3/2) in a neigh-
borhood of each point of rf, of rp and of rf n rp provided the angle
condition just mentioned is satisfied. (Interior regularity is not at issue.)
However, a precise description of the regularity possessed by U i near rf
or, more specifically, its regularity at mixed corners (i.e., points of rf nrf
and rf nrp), remains unavailable. Let us mention, however, an important
contribution of Nicaise [81J towards resolving this issue, wherin the precise
singular behavior of solutions of the Laplace and biharmonic equations on
networks at mixed corners is described. See also Nicaise [73J, [74J, where
regularity of solutions (among other things) of" transmission problems" for
general elliptic problems with general boundary and interface conditions is
investigated. D
3. Controllability of Linked Isotropic Membranes

In this section we shall consider the question of exact controllability of
solutions of the linearized equations of motion of the system of intercon-
nected, elastically isotropic membranes which was derived in Section 2.3.
We recall that such a system may be written (in the absence of distributed
forces) as follows. The equations of motion are
"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'
The boundary conditions are
(3.2)
and
(3.3) Wt = 0, x E r~, t > O.

The junction conditions are
(3.4) W t (p;l(p),t) = WJ(p,;-l(p),t),

pEJk, i,jEI(Jk), k=l, ... ,m,
and
(3.5) l: l:v~ (l:s~{3a~+J.tW3,aa3) (p;l(p),t)

tEI(A) a {3
= f-k (Ptk-1 (p),t), P E Jk , t > O.

Given a suitable T > 0 and "arbitrary" initial data (Wb, Wi) and final
data (Wb, Wi), the above system is exactly controllable in time T if there
are control functions ft = L J f;a~, fk such that the solution of (3.1)-(3.5)
satisfying the initial conditions
Wtlt=o = W~, W:tlt=o = Wi, x E Ot
achieves the state (Wb, Wi) at time T:

Wtlt=T = W~, W:tlt=T = Wi, i = 1, ... ,n.
Of course, the data and controls must be chosen from appropriate function
spaces for which the initial value problem is (at least) well- posed in some
sense. Since the problem (3.1)-(3.5) is time reversible, one may assume that
either the final state or the initial state vanishes. In the latter situation the
exact controllability problem is called the reachability problem. Thus the
reachability problem is that of showing that starting from the zero state,
an "arbitrary state" (Wb, Wi), i = 1, ... ,n, may be achieved in time T
through an appropriate choice of control functions.
It is not to be expected that every system of interconnected membranes,
no matter how configured, is exactly controllable. Indeed, our purpose is
to determine those geometric properties of such a system which will guar-
antee its exact controllability. It will be fairly apparent that the sufficient
conditions for exact controllability obtained below are (in all probability)
rather far from necessary. However, it seems that methods other than those
employed here will be required in order to further develop the theory.
Another point is that in order to obtain any exact controllability results
at all we usually (but not always) find it necessary to employ controls not
only along the exterior boundaries Af but also in the junction regions A[.
It is not clear to what extent the latter requirement is due to the methods
employed or whether there is something intrinsic to wave propagation in
general systems of interconnected membranes which necessitates the use of

junction based controls.
3.1. Observability Estimates for the Homogeneous Problem.

In this section the a priori estimates needed to describe the set of reachable
states will be derived. We begin by deriving certain basic identities. Let
o and at, a2, a3 be as defined in Section 1. We denote by m(xt, X2) the
radial vector field
m(Xl,X2) = Z)x a - xoa)ao, (Xl,X2) E IR?,

a
where (XOl,X02) is a fixed point in JR 2.
LEMMA 3.1. Let </> be a sufficiently regular solution of the scalar equation
p</>,tt - J.Ltl</> = 0 in 0 x (0, T),
and set
X(t) =p In </>,t(m V</> + ~</ dO,

where V</> = La </>,aao. Then
PROOF. The identity (3.6) follows from
o= [T [(m. V</>+ !</>)(p</>,tt - J.Ltl</ dOdt

io in 2
by appropriate integration by parts. It has been proved, either explicitly
or implicitly, by many authors in various contexts (cf. [60, p. 143]). It will
be valid for solutions having, for example, the regularity
for some s > 3/2. 0

Let t/J : JR2 ~ JR2 be a C 1 function. We introduce the notation
1
Ca(3(t/J) = "2 ('a,(3 + '(3,a),
sa(3( t/J) = 2JLca(3( t/J) + )"Da(3 L cn(t/J),
and we note the Green's formula
(3.7) In (S(t/J),c(cp))dO. = -In cp. divS(t/J)dO. h

+ cp. (S(t/J)v)dJ',
where (S( t/J), c( t/J)) = trace (S( t/J )c( cp)).
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
(3.8) In (V't/Jh) . div S(t/J) dO. = -~ h (h. v) (S(t/J), c(t/J)) dJ'
+ h(V't/Jh)'S(t/J)VdJ'-1n Q(V't/J) dO.,

where Q(V't/J) is a quadratic form in 'i,j given by
Q(V't/J) = ~(hl,l - h2 ,2)[(2JL + )..)('~,l - '~,2) + JL('~,l - '~,2)l

+ (2JL + )..)(h1,2'2,1'2,2 + h2,1'1,1'1,2)
+ JL(h1,2'1,1 + h2,1 '2,2)('1,2 + '2,1)
+ )..(h1,2'1,1 '2,1 + h2,1 '1,2'2,2).
PROOF. From Green's formula (3.7) we have
In (V't/Jh) . div S( t/J) dO. = -In (S( t/J), c(V't/Jh)) dJ'
+ h (V't/Jh) . (S(t/J)v) dJ',
so to prove (3.8) we need to calculate (S( t/J), c(V't/Jh)). We have
(S( t/J), c(V't/Jh)) = (2JL + )..)['l,l (V''1 . h),l + '2,2(V''2 . hbl

+ )..['1,1(V''2 h),2 + '2,2(V'1 . h),d
+ JL('1,2 + '2,1)[(V''1 . h),2 + (V'2' hbl
A lengthy, but straightforward, calculation shows that the right side of the
last equality may be written
~ div{h[(2JL + ,x)(1/1~,1 + 1/1~,2) + 2,x1/11,11/12,2 + JL(1/11,2 + 1/12,d 2]} + Q(V1/J)

= ~ div{h[(2JL + ,x)(C~l + C~2 + 2,xC11c22 + 4JLc~2]} + Q(V1/J)
= ~ div{h(S(1/J), c(1/J))} + Q(V1/J) ,
from which Lemma 3.2 follows. 0
LEMMA 3.3. Let 1/J be a sufficiently regular solution of the system
fYIP(3,tt - ~::>1\1/J) = 0, f3 = 1,2,

cc
and set
Then
(3.9) Y(T)-Y(O)+~ io[T in[ [LfYlPlt+ L (3 cc,(3

SCC (3(1/J)eCC(3(1/J)] dOdt
~ ~ [ (m . p) [~p'p~" + ~ Va 8
a
'(1/ (0:: )] dI'dt
-~ ~ lT i (m V)Tcc s cc(3(1/J) (a::) drdt
+ r;[ v s'(1/ [(mT):- + ~""l

a a dI'dt
PROOF. A proof of Lemma 3.3 (in a somewhat different form) may

be found in [45, Lemma 3.2]. It is obtained by carrying out appropriate
integrations by parts in the identity
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
(3.12) L r{m \7(3 + ~(3)s~ dO = ~r (\71/Jm + ~1/J) . div S(1/J) dO

a~~
= -~ i (m v)sa(3 (1/J)ca(3 (1/J) dJ' + i (\71/Jm) . (S(1/J)v) dJ'
+~
1/J. (S(1/J)v) dI' - ~ In SOl(3 ("p)ca(3 ("p) dr!.
On r one has
Insertion of the last two expressions into (3.12) yields
(3.13) L
01,(3
1(m \7(3 + ~(3)s~ dO = L01,(3 {~1r (m v)va a(3a:/: dJ'
0
S
+ rva a(3[(m. T/a(3 + !,al dJ'

1r
s
2
-~ 7: In
T
(m . V)Ta sa(3 dJ' - ~ sa(3 COl,a dO} .
The identity (3.9) now follows from (3.10), (3.11) and (3.13). 0
Let us now consider the homogeneous problem describing the motion of

a system of interconnected membranes. Let the displacement vector of the
i-th membrane be
i a
The system under consideration is then

' " ' a{3 (~/.i)
~ Si,a 'I' = Pi Wi(3,tt,
(3.14) {
JL:ilW~ = PiWttt, X E ni , t> OJ
= 0,
{
LV~S~{3(1/Ji)
(3.15)
v: .V'W~ = 0 X E rp, t > OJ
(3.16)
(3.17) "\}Ii(pii(p), t) = "\}Ii (pji(p), t),

P E Jk, i,j E I(Jk), k = 1, ... ,mj
(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
U(t) = ~ [L s~{3(1/Ji)a{3(1/Ji) + JLilV'w~121 dn.

i=i {li a,{3
If "\}I = ("\}Ii, ... , "\}In) satisfies (3.14)-(3.18), as a consequence of Hamilton's

principle E(t) is independent oft:
E(t) = E(O), t ~ O.
For i = 1, ... ,n, set

(3.20) mi(xl, X2) = :~:)Xa - xha)a~, (Xl, X2) E JR?,
a
where (Xh1,xb2) is a point in IR? which may depend on i.

LEMMA 3.4. Let \}I = (\}I 1 , ... , \}In) be a sufficiently regular solution of
(3.14) and set
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
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
The quantity in brackets is nonnegative since on rf

" i cr/3 8w~ _ " cr/3 i
~ vcrsi 8v i - ~ si ccr/3 ~ O.
cr,/3 cr,/3
Let us assume that the multipliers mi may be chosen so that
(3.22) mi. Vi sOon rf, i = 1, ... ,n.
Then L~=l fro 5 O.

(b) Integmk over rp.
It follows from the boundary conditions (3.15)
that the sum of integrals over rp
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.
(c) Integmls over rf, For any f ELl(rf) we may write
t.lt fix) dl' ~ t. ,)'(J') fix) dl'
= f L
k=l iEI(Jk)
1 Jk
f(pil(p))dJk
1
Consider
L (mi. Vi )PiI W:tI 2dr .
iEI(Jk) Jk
By utilizing the geometric junction conditions (3.17), this sum is seen to

be the same as
We assume that
(3.24) L Pi(mi . vi)(pil(p)) 5 0, 'vip E Jk, k = 1, ... ,m.

iEI(Jk)
Then
(3.25)
Next, let us consider
in which we have again utilized (3.17). It follows from the dynamic junction
conditions (3.18) that the last integral vanishes. Similarly, consider
Along Jk one has Ti = T ik which, along with (3.17), implies that
Therefore, (3.26) yields
Our next assumption on the multipliers is that
It then follows from (3.18) that the integral (3.27) vanishes.

We summarize what we have obtained so far in the following lemma.
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
(3.29) Z(T) - Z(O) + TE(O)
REMARK 3.1. Note that
(mi. v i )(pil(p)) = ~)(p - p~) . a~ - X~a)lI~

(3.30) a
where x& = La x&aa~. Similarly,
(3.31) (mi. 7'i)(pil(p = (p - p~ -~). 7'\ p E Ai.
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
(3.32) { (m~Z. 7'~)(pil(p = (p - qo) . 7'>

(m . VZ)(pil(p = (p - qo)' v', p E Ai, i E I(Jk),
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
(3.33) L Pi(P - qo) . vi :-::; 0, p E Jk.

iEI(Jk)
Note that (p - qo) . Vi is either strictly positive, strictly negative or identi-

cally zero on Jk since vi is constant there. It is clear that (3.32) will hold
if at most one of the mi. vi, say m q v q, is positive, if mi. vi :-::; 0 for i 1= q
and if pq(mq . v q) is sufficiently small relative to Pilmi . vii for those i 1= q
for which mi . Vi is not identically zero. 0
We are going to write the integral over ri(Jk) in (3.29) in terms of

normal and tangential derivatives. On r i one may write
from which follows
0l,{3
It may then be verified that

and therefore
The next thing to be proved is that, under some additional conditions on

the multipliers mi and parameters /-ti, Ai, one may, with the aid of the dy-
namic junction conditions, eliminate the normal derivatives from the right
side of (3.36). This is important in order to establish controllability results
which utilize only physically meaningful controls (forces and moments) in
the junction regions.
To this end, let us denote by Svi(1/i) and STi(t/Ji) the normal and tan-
gential components, respectively, of Si(t/Ji)v i in (3.34):
where
Then
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
where Ci is a constant depending on the set of elastic parameters J.ti, Ai

and on
max I(mi . ,})(pi 1 (p))l.
pEJk
The dynamic junction condition (3.18) may be written
(3.38) L (Sl/iV i + Sr i7"i + J.ti ~~: a~) = o.

iEI(Jk)
We define, for i E I(Jk),

I if 7"i = 7"ik
(1.,( Jk) ={ 1 1f 7" i -_ -7"ik .
-
Equation (3.38) may be decoupled into
(3.39) L (1i(Jk)ST i =0
iEI(Jk)
and
(3.40) ,,( . O\}l~.)

L..J Sl/iV' + ov aa = O.
J.ti i
iEI(Jk)
Our final conditions on the multipliers are as follows:
(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,
"L..J (1i(Jk)ST i = 0 ==9- " . S2. ~ 0,

. . v')~
L..J (1 + Di)(m'
iEI(Jk) iEI(h) J.t,
and
Therefore, we have the estimate
where Ck = L:iEI(Jk) Ci . As a consequence of this inequality and Lemma

3.5, we obtain the following result.
THEOREM 3.1. Assume (3.22), (3.24), (3.28), (3.41) and (3.42). If

\II = (\111, . .. ,\lin) is a sufficiently regular solution of (3.14)-(3.18), then
(3.43) Z(T) - Z(O) + TE(O)
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
(see (3.35), (3.35)). On rp, Si(1/Ji)vi = 0 hence, from (3.34),

i 81/Ji i 81/Ji
_=-v-
T
8vi 8Ti '
i 81/Ji Ai i 81/Ji
V-.=- T -..
8v~ 2J.ti + Ai 8T~
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)
IZ(')I ,; Co ( E(O) + t, if' 1.'(')1' dl') .
A calculation of Komornik [43] gives
sup [ llJ!i(t)12 dI' :::; (1 + liT) [T [ llJ!i(t)12 dI'dt

[O,T] 1rp+ 10 1rp+
+ [T [ 1lJ!~t(t)12 dI'dt.
10 1rp+
The estimate (3.44) then follows from the last two inequalities since
o
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
t.m;a1~ (~air + ~m;a1,

mq
which is negative if mq ~ 0, mi < 0 for i = 1, ... ,q-1 and mq is sufficiently

small with respect to Imil, ... ,Imq-il. A more precise condition is that
the determinants
satisfy
(3.45) (-l)i D i >0 for j = 1, ... ,q - 1.
Similarly, set mi = (mi.vi)/(2JLi+Ai) and write, utilizing the constraint

in (3.42),
~(""b1 + m;c;) ~ [~[bi("i .flq "q) + c,(a\ . v")lj'
+ mq [~[b;("i .an + c,(a; . ani]' + ~(""b1 + m;o1),

which is negative if mi, mi are negative for i = 1, ... ,q -1, and if mq, mq
are positive but sufficiently small with respect to the values of Imil, Imil,
respectively, for i = 1, ... ,q - 1. Again, one may write a more precise
condition analogous to (3.45), in which one should note that a~ . a~ =
_Vi. v q and a~ . v q = vi . a~. 0
3.2. A Priori Estimates for Serially Connected Membranes.

Consider n interconnected membranes with joints J1, ... ,Jm . The config-
uration is called serially connected if
{
a 3i :.: a i3 , \.I..
vZ,) = 1, ... ,n,
m-n-1.
By reindexing the membranes and joints, if necessary, we may assume that
{:: ~::+::, J;; ~;:' ~:. ,n -1,
ThusI(Jk) = [k, k+l]. Further, without loss of generality, we may suppose

= a~ = (0, 0, 1) (hence
that a~
,} = _v Hi , Ti = _T Hi on Ji )
and that
Pi = ni , Af = rf, Af = rf,
rk(Jk) = rk+l(Jk) = Jk, k = 1, ... ,n - 1,
{
ri(Jk) = 0, Ii - kl > 1.
If Pi is serially connected to PHi, the geometric and dynamic conditions
along Ji reduce to, respectively,
(3.46)
and
STi(t/Ji) = ST i+ l (t/JH1),
(3.47) S,A t/Ji) = S"Hl (t/JHi),

aiJI i aiJIH 1
J.ti a) = -J.tHi a)+l along Ji ,
so that the transverse motion is not coupled to the in-plane motion, as is
to be expected.
When membranes are serially connected, it is possible to eliminate the
integrals over Jk in the estimate (3.29), provided the parameters Pi, J.ti, Ai
satisfy certain monotonicity conditions. To prove this statement, we choose
as multipliers
mi(x)=x_XO' xEIR?, i=I, ... ,n,
where Xo E JR2 does not depend on i, and we write m = mi.
THEOREM 3.2. Assume that Xo E JR2 may be chosen so that
(3.48) (x - xo) . vi ~ 0 on rf, i = 1, ... ,n.
Suppose further that for the given choice of Xo and each i = 1, ... ,n - 1,
(3.49)
if m . vi > 0 on Ji ;
(3.50)
if m yi < 0 on Ji . If'IJI = ('IJI 1 , . ,'lJl n ) is a sufficiently regular solution

of (3.14)-{3.16), (3.46), (3.47), then
Z{T) - Z{O) + TE{O)
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
PROOF. Since mi is independent of i, clearly (3.28) will be satisfied.

Suppose that m yi > 0 on Ji . Condition (3.24) will then hold if and only
if Pi ~ Pi+I, since yi = _yi+1 on Ji . The sum of integrals over fi{Jk) in
(3.29) may, in the present context, be written {see (3.37))
(3.52)
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 [ 8r " ( yk. 01/l)

o-r k - 8rk+1 ( yk+1. 01/l+1)]
o-r k+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
_ J.tk+1 {J.tk+1 + Ak+d (-rk+1 . 01jJk+1) 2] }dr.

2J.tk+1 + Ak+1 O-rk+1
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
When IJ.k ~ lJ.k+b it is easily checked that the quadratic form inside the
braces is nonpositive if and only if
both of which will hold if .Ak ~ .Ak+ 1.

The proof is similar when (3.50) is satisfied. 0
REMARK 3.3. Recall that m . vi must be of constant sign on k If, for

a particular index i, m . Vi = 0 on Ji , no relation between the material
parameters of Pi and PH 1 need be assumed. Also, since the choice of an Xo
satisfying (3.48) is not unique, for certain configurations it may be possible
to choose distinct values of Xo such that for certain indices i, m . vi > 0
on Ji for the first choice of Xo while, for the second, m Vi < 0 on J i .
In such situations either (3.49) or (3.50) is sufficient for the conclusion of
the theorem. These observations suggest that the above restrictions on the
material parameters are due to the method of proof employed and are not
intrinsic to the validity of the theorem. 0
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
3.3. A Priori Estimates for Single Jointed Membrane Systems.

We consider a membrane system containing a single junction region J, i.e.,
m = 1. In this case it is possible to prove, at least for some specific
configurations, a stronger result than Theorem 3.1.
We first choose the multipliers mi as in Remark 3.1, that is,
(mi. v i )(p;1(p)) = (p - qo) . vi,

(3.53) {
(mi. ri)(p;1(p)) = (p - qo)' ri, p E Ai,
where qo is to be determined. Then we have the following result.
THEOREM 3.3. Assume that m = 1 and that qo may be chosen such that
(3.54) (p - qo)v i ~ 0, P E Af, i = 1, ... ,n,
(3.55) (p - qo)vi = 0, P E J, i = 1, ... ,n.

If '11 = ('11 1 , . ,'I1n) is a sufficiently regular solution of (3.14)-(3.18),

then
(3.56) Z(T) - Z(O) + TE(O)
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
Unfortunately, in many cases of interest, it will not be possible to choose

qo according to (3.55) while still maintaining the inequalities (3.54). We
shall consider one such situation and show that qo may be chosen in such
a way that (3.24), (3.41) and (3.42) are satisfied without any restrictions
on the parameters Pi, /-Li Ai.
To this end, we consider a folded membrane, i.e., n = 2, m = 1, we
assume that Af
= 0 and that qo is such that
(3.57)
Condition (3.57) will continue to hold if qo is replaced by qo + ra!, where

r is any real number. Let
i (p - qo - ra!) . vi .
0r(P) = II P- (qo +ra31)11 ,z = 1,2,
which is, of course, nothing but the cosine of the angle between P- (qo+ra!)
and the normal vector Vi at p. Then
lim o;(p) = 0, P E J,
r-+oo
and, since the membranes are nonserial,
with the convergence uniform on J. By properly choosing the sign of r, we

may assure that
It then follows that

2
LPi(p - qo - ra~) vi = [PI0;(P) + P2o;(p)lllp - qo - ra~1I :::; 0, P E J,
i=l
if r is sufficiently large and of proper sign. Similarly
t a~
~I~
{p - qo - raD . vi = lip - qo - raAIl (8~ + 8;) a~ :::; 0,
~ ~
p E J,
and likewise for (3.42). We therefore have the following result.
THEOREM 3.4. Assume that n = 2, m = 1, that (3.57) holds and that

rf = 0. IfiI! = (iI!I, \11 2 ) is a sufficiently regular solution of (3.14)-{3.18),
then the estimate (3.43) is valid for \II.
REMARK 3.5. A completely analogous result is valid in the situation
where n ~ 2 membranes are connected along a single joint, provided that
rf = 0 for i ~ 2 and the configuration is such that
lim8~{p)<0,
r
pEJ, i=2, ... ,n,
for a suitable choice of the sign of r. 0
3.4. The Reachable States. The a priori estimate of Corollary 3.1

will be utilized in this section to determine the reachable states of the
system (3.1)-{3.5). Since this estimate is proved only for "sufficiently reg-
ular" solutions of (3.14)-{3.18), the following regularity hypothesis will be
imposed (see Remark 2.4):
(3.58)
with continuous injection for some s > 3/2.

This regularity assumption imposes a further constraint on the configu-
ration of the system of membranes, but we do not know how to quantify
this constraint in terms of simple geometric properties of the configuration.
To describe the set of reachable states, we apply the Hilbert Uniqueness
Method introduced by J.-L. Lions in [61]. Let (iI!0, \III) E DA and let \II be
the solution of (3.14)-{3.18) with final data (\11 0, \lid at time T:
(3.59)
We introduce a seminorm on DA X V by defining
According to Corollary 3.1 (with E{O) replaced by E{T), which is permis-

sible since the problem for iI! is time reversible), II . IIF defines a norm if
T > To and if (3.24), (3.28), (3.41) and (3.42) are satisfied. Let F denote
the Hilbert space which is the completion of DA x V with respect to IIIIF.

Corollary 3.1 gives
(3.60) FCVxH
with dense and continuous injection. If (l}10, l}1d E F then, by definition

of the space F, the corresponding solution l}1 of (3.14)-(3.18) satisfies
.
W:tIED+ E L2(0, Tj .c2 (rp+)),
In the above
.c2 (rp) = {Lha~lh E L 2 (r i ), supph c rp},

j
.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
(3.61) Wlt=o = W,tlt=o = 0,
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 wi = La W~a~. We formally integrate by parts to transfer the

differentiations from W to l}1 and obtain
This last equality may be written, in view of (3.60), as
where (g',g)F denotes the duality pairing between an element 9 E F and

g'E F', the dual space of F.
In (3.62), let us choose
(3.63) ' ~- [(!)' "" -"''Lp C!,)' "" Ef~ +-
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 )
\ - 8Ti \I!t ,\I! = 8Ti . 8Ti df'dt,
'v'. E L 2 (0, Tj 1-{l(rp-)),

where
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
Similarly, we select
(3.66)
With these choices of f i , fk, (W,t(T), -W(T)) is a linear function of

(1I'0,1I'd:
(W,t(T), -W(T)) = L(1I'0, 1I'd
for some linear operator L and (3.62) becomes
Therefore, L is just the Riesz isomorphism of F onto F'. It follows that

the set of states (W(T), W,t(T which are reachable in time T > To,
starting from the zero state, is characterized by (W,t(T),-W(T)) E F' if
one utilizes controls
along EP, and

(3.68)
along Eik(Jk ). Since F c V x H one has V' x H C F'. We have therefore

obtained the following result.
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
REMARK 3.7. When f i , fk satisfy (3.67) and (3.68), respectively, it may

be proved that the problem (3.1)-(3.5), (3.61) has a unique solution, defined
by transposition. This solution exists only in a very weak sense and satisfies
(3.62) by virtue of the sense in which solution is defined. The reader is
referred to [44] for a discussion of this point in a general context. 0
3.4.1. Serially Connected Membmnes. It follows from Theorem 3.2 that

for n serially connected membranes the following estimate is valid.
COROLLARY 3.2. Suppose that the hypotheses of Theorem 3.2 are satis-
fied. If'iJ! = ('iJ!1, . .. , 'iJ!n) is a sufficiently regular solution of (3.14)-(3.16),
(3.46), (3.47) on (0, T) with ('iJ!(T), 'iJ!,t(T)) = ('iJ!o, 'iJ!t}, there is a To> 0
such that for T > To,
(3.69) II (iI'0, iI',) II~, H ., CT t. {t h f + (I iI':.I' + Iii" I') dI"dt
+ 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)
(that is, ifrp- =0) then
(3.72) V x H c ((W(T) , WAT)) 1 r E ,2(EP) and fk = o}.

PROOF. The inclusion (3.70) follows from Corollary 3.2 in the same way
that Theorem 3.5 was deduced from Corollary 3.1. If (3.71) is satisfied,
the estimate (3.69) reduces to
II ('iJ!o, 'iJ!t}II~xH ~ CT~ loT h(1'iJ!~tI2

p + l'iJ!iI2) dI'dt.
Then, by the process of ''weakening the norm" (see, e.g., [60, p. 200]), one
may obtain
(3.73) II ('iJ!o, 'iJ!t}II~xvl ~C Tt.1T hp I'iJ!i 12 dI'dt.
As above, for T > To one may introduce a norm IIIIG through
II ('iJ!o, 'iJ!1}11~ = ~ loT h p 1'iJ!i 12 dI'dt

and a space G which is the completion of DA X V with respect to this norm.

The set of reachable states (W(T), W,t(T, using controls fi E L 2 (EP),
fk = 0, is characterized by (W,t(T), - W(T E G'. The estimate (3.73)
shows that G is continuously and densely embedded in H x V', hence
H x V c G'. 0
3.4.2. Membrane Transmission Problems. Let 0 be any bounded, con-
nected open set in IR2 with a Lipschitz continuous boundary consisting of
a finite number of smooth simple arcs, and let 0 1 be a simply connected
open set in IR2 with smooth boundary such that '01 c O. Set
O2 = 0\ Ob r = a~, r 1 = aO b r 2 = a0 2 = r u r 2
We imagine that 0 represents the equilibrium position of a membrane com-
posed of two parts, 0 1 and O2 , each of which is homogeneous and elastically
isotropic but with possibly differing material properties. Let yi denote the
exterior unit normal to r i , so that yl = _y2 on rb and let i,j, k be the
natural basis for IR3. If
Wi = Wti + W~ + W~k := wi + W~k
is the displacement vector of Oi, for small displacements the motion is
described by
(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,
where r = F UI'D as above and f = hi + I~ + fak is an applied force on

rD.,
S.,.1(W1) = S.,.2(W2 ),
(3.77) SV1(W1) = Sv 2 (W 2),

aWJ aWl
J.Ll ayl = -J.L2 ay2' x E rl, t> 0,
where 7"i is the unit tangent vector to r i , positively oriented with respect
to k (hence 7"1 = _7"2 on ri). The initial conditions are
The problem (3.74)-(3.77) is known in the literature as a problem of

transmission. It obviously falls within the framework of serially connected
membranes discussed above in which r 1 plays the role of the junction region
(cf. Remark 3.4). However, the present problem is somewhat simpler in
that there is no difficulty concerning regularity of the solution since r 1 is
a smooth curve entirely contained within OJ one has in this case
WJ E C([0,T]jH2(01)), WJ E C([O, T]j H S (02)), j = 1,2,3,
for some s > 3/2 provided F
and I'D meet in a strictly convex angle.
(The assumption that r 1 is a smooth curve may be weakened to allow for
certain piecewise smooth curveSj see Nicaise [73], [74].) The techniques
above therefore yield, for example, the following result.
THEOREM 3.7. Suppose there is a point Xo E 0 1 such that
(x - xo) . 1/ 1 ~ on r 1 ,
(x - Xo) . 1/2 ~ on rD,
(x - xo) . 1/2 ~ on rC.
Assume further that
Then for T > To,

V x H c ((W(T), W,t(T)) If E .c2(~D)}.
REMARK 3.8. Since there is no coupling between the pair WI, w 2 and
the pair WJ, wi in the system (3.74)- (3.77), the last result is equivalent
to two separate reachability results, one for the problem
JLi~W~ = PiW~,tt, X E Oi, t> 0,

W; = 0, x E r C , t > 0,
owi = 13,
JL2 01/2 xE
D
r ,t > 0,
{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

of Systems of Linked Plates
The main purpose of this Chapter is to derive the dynamic equations of

motion of a system of interconnected thin, elastic plates. In particular, we
seek to determine the transmission conditions which describe how distur-
bances in one member of the system are propagated into the others. The
approach taken is similar to that of Chapter VI. We first derive the equa-
tions of motion and boundary conditions of a general nonlinear plate, and
corresponding Hamilton's principle, under a specific kinematic hypothesis
on the deformation process and assuming a linear stress-strain relation.
Linearizing this system around the trivial equilibrium leads to a system
known in the literature as the Reissner, or Reissner-Mindlin, plate model
for the transverse displacement and shear angles, and the usual equations
of plane elasticity for in-plane deformation.
The dynamical system describing the transient behavior of a collection of
interconnected plates is derived by minimizing an appropriate Lagrangian
over a class of deformations which satisfies certain geometric constraints,
in particular, in the junction regions where two or more plates are joined
together. The form of these constraints at a joint depends on the nature of
the joint (e.g., rigid, hinged, etc.) and each choice leads to particular bal-
ance laws for the transmission of linear and angular momentum through the
joint. In the resulting dynamical system there are five unknown functions
associated with each plate of the system: three components of the dis-
placement vector and two shear angles. All displacements and shear angles
are coupled through the geometric and dynamic junction conditions. The
shear angles and transverse displacement of each plate are further coupled
in the equations of motion through a coupling parameter called the shear
modulus of the plate. In Section 6, we consider the limit of the dynamical
system (in the case of semi-rigid joints) as all shear moduli tend to infinity
(this corresponds to the absence of transverse shearing) and obtain a limit
252 VII. SYSTEMS OF LINKED PLATES
system which models the transient motion of a collection of interconnected

Kirchhoff plates with semi-rigid joints.
Exact controllability of a system of linked Reissner plates is considered
in Section 7.6, but only in the case of serially connected plates.
1. Modeling of Dynamic Nonlinear Elastic Plates

In this section the dynamic equations and edge conditions which de-
scribe the motion of a class of nonlinear elastic thin plates will be derived.
Although the models which are presented are known in the literature, their
derivations are included here for the sake of completeness and because they
may not be familiar to the reader. Our presentation is adapted from Wemp-
ner [108, Chapter 7]. Summation convention will be assumed throughout
this section and the next.
Let 0 is a bounded, open, connected set in lR? with Lipschitz continuous
boundary r consisting of a finite number of smooth, simple curves, and let
al.a2,a3 be a fixed, right-handed orthonormal basis for IR3. We set
Q = {(XI,X2,X3)1 (Xl.X2) En, -hl(Xl.X2)::; X3::; h 2 (XI,X2)},
where hI, h2 are positive, differentiable functions on n. The reference con-

figuration of the plate is given by
B = {Po + Xa8a + X3 a 3 := P(XI, X2) + X3a 31 (Xl. X2, X3) E Q},

where Po is a fixed vector in IR3 and P(Xl. X2) = Po + Xa8a The planar
region
p = {P(XI,X2)1 (XI,X2) E O}
is called the reference surface. Note that the plate may have variable
thickness and is not necessarily symmetric with respect to its reference
surface.
Basic kinematic hypothesis. The position vector at time t to the
displaced location of the particle situated at P(Xl. X2)+X3a3 in the reference
configuration is
(1.1) R(Xl. X2, X3, t) = P(Xl. X2, t) + X3A3(XI, X2, t),

where A3 =1= o.
Assumption (1.1) means that the X3 line remains straight in the de-
formed plate. Set
where a subscript after a comma means differentiation with respect to the

indicated spatial variable. Then Al. A2 are vectors which are tangent to
the deformed reference surface. From (1.1) one has
(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
eO:f3 = ![Ao: . A(3 + x3(Ao: . A 3,(3 + A(3' A 3,0:)

{ +X~A3,0: . A3,(3 - 80:(3],
(1.3)
e0:3 = !(Ao: . A3 + X3 A 3 . A3,0:),
e33 = !(A3 . A3 - 1).
We introduce
(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
r(xl' X2, X3, t) = W(Xb X2, t) + X3 U(Xl' X2, t)

(1.5)
: = Wi(Xb X2, t)~ + X3Ui(Xb X2, t)~,
which is the displacement of the point P(Xl' X2) + X3a3. The strains may
then be written in terms of the displacement W and rotation U by noting
that A3 = U + a3 and Ao: = W,O: + ao;.
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
(1.8) fh 2 t:'o:dX3 + t31~:~~2hI + fh F dX3 =

2
fh 2 pR,ttdX3.
-hI -hI -hI
Let
(1.9)
Ni is the resultant stress per unit of reference area. One has

and therefore (1.8) may be written
Since the mapping (Xl,X2,X3) f-+ Xiai is orthogonal with determinant

unity, the vector
n+ = (-h2,aaa + a3)1 vII + IVh212

(resp., n- = -(h 1 ,aaa+a3)IJ1 + IVhlI2), where VhjJ. = hjJ.,aaa, is a unit
normal vector to the upper surface p(Xl, X2) + h 2(xl, x2)a3 (resp., to the
lower surface P(Xl,X2) - h1 (xl,X2)a3) of the plate and points towards the
exterior of the body. Therefore
[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
F = J1 + IVh212 t+ + J1 + IVh l l2 t- + [:21Fdx3,

the resultant body force per unit of reference area, and
With this notation, equation (1.10) becomes
(1.11) N~ +F = mW,tt + mpU,tt.

From (1.2) and (1.4) we have
Ga = (W,a +aa) +X3Ua, G3= U +a3

and therefore
where
(1.12) Nij = jh2 sij dX3, Mij = jh2 X3 sij dx 3

-hI -hI
are called the generalized stresses and generalized moments, respectively.

Equation (1.11) may therefore be written
(1.13) a~o [NnJ3(W,J3 + aJ3) + M oJ3U,J3 + No3(U -+ a3)) + F

= mW,tt + mpU,tt.
Next, multiply (1.7) by X3 and integrate through the plate thickness.
One obtains
(1.14) jh2 x3t~dx3 _jh2

-hI -hI
t3dx3 + x3t31~:~~2hI + jh2
-hI
X3 F dX3
= mp W,tt + IpU,tt.
where
The resultant stress couple per unit of reference area is
where
Equation (1.14) may be written
(1.15)
where
is the resultant applied couple per unit of reference area and

When written in terms of generalized stresses and generalized moments,

(1.15) becomes
(1.16) a~a [Ma,B(W,,B + a,B) + Ia,BU,,B + Ma3(u + a3)]

- [N 3,B(W,,B + a,B) + M 3,BU,,B + N33(U + a3)] + C
= mpW,tt + IpU,tt.
The equations of motion are (1.13) and (1.16). They comprise six scalar
equations for the 21 scalar functions Wi, Ui , Nij, Mij, Ia,B (taking into ac-
count the symmetry of sij).
1.2. Edge Conditions. Let OP denote the two-dimensional boundary

of the reference surface P and suppose that
-rlJ -D
(1.17) OP=A UA ,
where Ac, AD are mutually disjoint, relatively open subsets of OP and
AD i- 0. (As in Section 1.2, the superscripts C and D are used to indicate
"clamped" and "dynamic," respectively.) The partition (1.17) induces a
similar partition of r := an into
(1.18)
through the mapping
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)
Proceeding as above, we impose the conditions
One thereby obtains the edge conditions
(1.20)
or, in expanded form,
ZlQ[NQ,B(W,,B + a,B) + MQ,BU,,B + NQ3(U + a3)] = f,

(1.21) {
ZlQ[MQ,B(W,,B + a,B) + IQ,BU,,B + MQ3(U + a3)] = c.
1.3. Hamilton's Principle. Let r be a sufficiently smooth function

defined on Q x [0, T] such that rlt=o = rlt=T = 0 and assume that r satisfies
the same "geometric" conditions as the actual displacement r = W + X3 U
of the plate. For definiteness, let us suppose that the geometric condition
is
From (1.7)
This equation may be written
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
(1.23) [T [(F, Vi + C. U) dOdt + [T [ (f. Vi + c U) dl'dt

~ ~ ~ ~D
- loT In (NQ . Vi,Q + MQ . U,Q + N3 . U) dOdt
= -loT l [(m W,t + mpU,t) . Vi,t + (mp W,t + IpU,t) . U,t] dOdt.
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.
For a given admissible displacement r satisfying (1.22), the associated

kinetic energy is
K(t) = ~ 10 plr,tl 2dQ

(1.24)
=~ l (mIW,tI 2 + 2mp W,t . U,t + IpIU,tI 2 ) dO.,
and the associated work done by the external forces is
W(t)= { FrdQ+ { nitirdS

(1.25)
iQ iaQ
= { (F . W + C . U) dO. + { (f W + c . U) dr.
in irD
By utilizing the notation (1.24), (1.25) and recalling (1.2) and the symmetry
of si j , the principle of virtual work may be written
(1.26) 8 loT [K(t) + W(t)] dt -loT 10 sijOcijdndx3dt = 0,

where 8 represents the Frechet derivative with respect to an admissible
displacement r, i.e., a sufficiently smooth displacement of the form (1.5)
defined on Q x [0, T] which matches given displacements of the body at
t = 0 and t = T and satisfies the geometric boundary condition (1.22).
If it is assumed that the material in question is hookean, ie., satisfies
(1.8), with elasticity tensor Cijkl satisfying (1.8) and (1.9), it follows from
(1.17) that (1. 26) may be written
(1.27) 81 T
[K(t) + W(t) -U(t)]dt = 0,
where U is the strain energy functional given by
(1.28) 11 .
U{t) = -
2 Q
s'JcijdQ.
Thus, the actual evolution of deformations is a stationary state of the

Lagrangian
L = loT [K(t) + W(t) - U{t)]dt.

This is Hamilton's principle for our plate problem. Since W and U are
independent quantities, (1.27) is equivalent to the two equations
8WL=O and8uL=O,
where the subscript denotes differentiation with respect to the indicated
variable. The equation 8WL = 0 yields (1.11) and the first of the edge
conditions in (1.20), while the second gives (1.15) and the remaining edge
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.
REMARK 1.1. One may, of course, consider geometric constraints other

than, or in addition to, (1.22). for example, suppose that
where Ac, AD are as before and AS corresponds to a simply supported part

of OP, that is,
W=o onr s .
Here rS is the part of an corresponding to AS. Application of Hamilton's
principle leads to the dynamic condition
The complete set on edge conditions in this situation is therefore (1.20),
W = 0, U =0 on rC,
W = 0, vo:Mo: = c on rS. D
1.4. Additional Kinematic and Material Assumptions. In this

subsection the additional kinematic requirement
(1.29)
is imposed. The first condition in (1.29) is equivalent to the vanishing of

the extensional strain C33 and the second asserts that the x3-line rotates
through an angle of less than 90 degrees. The two conditions together
imply that
(1.30)
where u = Uo:ao:. Therefore
(1.31)
The kinetic energy is written

where w = Walla and U3 given by (1.30). The work functional is expressed

as
W(t) = In [F W + (CJSaJS) . u + C3U3]dO

+ f [f W + (cJSaJS) . + C3U3] dr
irD U
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
(1.32) 8~a {Na!3(W,!3 + a!3) + Ma!3[u,!3 + (vl-luI 2 ),!3a3]

+ N a3 [u + (VI -lul 2 )a3]) + F
= mW,tt + mp[u,tt + (vl-luI 2),tt a3],
(1.33) va{Na!3(W,!3 + a!3) + M a.B[u,!3 + h!1-luI 2 ),!3a3]

+ N a3 [u + (vl-luI 2 )a3]) =f on rD.
In terms of components, (1.32) is expressed as
(1.34)
8~a [Na!3(WJS ,!3 + 0JS!3) + M a!3UJS ,!3 + N a3 UJS ] + FJS
= mWJS,tt + mpUJS,tt,
8
8x a [N a!3W3,!3 + M a!3( VI -lul 2 ),13 + N a3 ( VI -luI 2 )] + F3
= mW3,tt + mp( VI - lul 2 ),tt,
where F = Fi8i, while the edge conditions (1.33) are
Va [Na!3(WJS ,!3 + Oa!3) + Ma!3UJS ,!3 + N a3UJS ] = IJS'

(1.35) { va [Na!3(W3,!3 + M a!3( VI -l u l2 ),13
+Na3(VI-luI2)] = h on rD ,
where f = li8i.
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,
(OuW,U) = 10 [(Cp.ap.) . U+ C3(OuU3,U)] dO

+ [ [(cp.ap.) U + C3(OuU3, u)] dr
(1.37)
Jr D
= 10 [(Cp.ap.) . U - C3(1-~~~)1/2 ] dO
+ hD [(Cp.ap.)' U - C3 (1-~~~)1/2 ] dr.
The variation ouU of the strain energy is given by
One has
1
Ouca(3 = '2(Ga . OuG(3 + G(3 . ouGa),
1
Ouc3(3 = '2(G3 . OuG(3 + G(3 . OuG3),
where
G(3 = R,(3 = W,(3 + a(3 + X3 U,(3,

G3 = A3 = U + a3 = U + (U3 + 1)a3'
Consequently,
(1.38) (ouU, tl) = k [sa/3G/3 . (OuGa, tl) + s3/3G/3 . (OuG3, tl)

+ s3/3G 3 . (OuGa, tl)] dQ.
A straightforward calculation gives
(0uG 3, U
A) A U tl
= U - (1 _ luI2)1/2 a3,
hence
G/3' (OuGa, tl) =X3(W,/3 + a/3 + X3 U ,/3)

A [u,a . tl + U tl,a (U U,a)(U' tl)] }
. U,a - (1 _ luI2)1/2 + (1 _ luI2)3/2 a3
{
=X3(W,/3 + a/3 + X3U,/3) . tl,a

u a . tl + U tl a
- X3(W3,/3 + X3 U3,/3) [ '(1-luI2)1/2'
(U' u,a)(U' tl)]

+ (1 -luI 2)3/2
Therefore
(1.39) jh2 sa/3G/3' (OuGa, tl) dX3 = [Ma/3(w/3 + a/3) + [O/3U/3]' tl,a
-hl
0/3 0/3 [u,a . tl + U . tl,a (u u,a)(U' tl)]

- (M W 3,/3 + I U3,/3) (1 -luI2)1/2 + (1 -luI2)3/2 .
Similarly,
(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
(1.41) jh2 s3/3G3 . (8uG.B, tl) dx3

-hl
= M 3.Bu tl,.B
3/3 [u,a . tl + U tl,a (U U,a)(U' tl)]

- M (U3 + 1) (1 -luI2)1/2 + (1 -luI2)3/2 .
It follows from (1.36)-(1.41) that the variational equation ouL = 0 may be

written
0= loT In {[(mpw,t + Ipu,t) . fl,t + NI (W3, u)) . U,t + N 2(W3, u) . fl]

- [MO:/3(w,/3 + a/3) + IO:/3U,/3 + M 3o:u - Mo:(W, u)] fl,o:
- [N3/3(w,/3 + a/3) + M 3/3u,/3 - N 3(W, u)] fl}dOdt
+ loT In (C,..a,.. - C3N4(U))' fldOdt
+ r r (c,..a,.. - c3N4(U)) . fldOdt,

T
Jo Jr D
where NI , ... ,N4 , Mo: are nonlinear partial differential operators defined
as follows:
U
N4 (u) = 1 -luI 2 '
Mo:(W, u) = [MO:/3W3,/3 + /,j;/3U3,/3 + M 3O:(U3 + 1)] (1 -1:1 2)1/2'
In all of these expressions, U3 is given by (1.30). The corresponding equa-

tions of motion and edge conditions are, respectively,
(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),
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
(1.46) A~ = n, A~. A2 = A~ . A l , A~ Al > o.

These conditions uniquely determine the A~ basis.
The A~ basis may be obtained from the ~ basis through an orthogonal
transformation. This transformation may be considered to consist of three
separate rotations: a rotation through an angle 'l around the aI-axis,

followed by a rotation through an angle '2 around the rotated a2-axis,
followed by a rotation through an angle '3 around the twice rotated a3-
axis. The angles 'l, '2 are associated with plate bending, while '3 is
associated with twisting of the reference surface. We choose the positive
direction of rotation to be counterclockwise. One then has
(1.47) tan 'l = _ n . a2 = W3 ,2 (WI,1 + 1) - WI ,2 W3 ,1 ,

n a3 (WI,1 + 1)(W2,2 + 1) - WI,2W2,1
(1.48) tan '2 = _ n . al = W3 ,1 (W2,2 + 1) - W2,1 W3 ,2 .
n a3 (WI,1 + 1)(W2,2 + 1) - WI,2W2,1
In order to express '3 in terms of displacements, we note that the initial
rotation through 'l around the aI-axis maps a2 to the vector a.;, where
(1.49)
while the rotation through '2 about a~ maps al to a~, where
(1.50)
The image of a3 after these two rotations is A a, and a~, a~ are in the plane
spanned by t l , t2' The final rotation through '3 maps the orthonormal
vectors a~, a~ onto A~, A~, respectively. A little geometry reveals that '3
is the average of two rotations (h and 92 about A a, the first mapping a~
to tl and the second mapping a~ to t2' Consequently,
sin2 '3 = ~(1 - cos91 cos 92 + sin91 sin ( 2),
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.
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
The quantities E, E, a, a, G are positive constants which satisfy

0< a < 1/2, E(1- a) - Ejj > 0, E(1- a) - 2Ejj2 > 0.
2. LINEARIZATION 267
When E = E and a = i7, (2.2) are the strain-stress relations of an isotropic

material. Otherwise, they indicate that the material is isotropic in direc-
tions parallel to the reference plane but have different material properties
in the transverse direction.
To calculate the generalized stresses and moments in terms of displace-
ments it is necessary to solve (2.2) for 8 ij in terms of Ckl. We introduce
E[E(l - a) - Ei7] - EEi7
J.t = 2[E(1 _ a) _ 2Ea 2 ]' A = E(l - a) - 2Ea 2 '
and further define TJ and fi through the equations
~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
8 33 = 2fJ,c33 + 5.c kk,
8 12 = 2(fi - TJ)c12, 8 13 = 2Gc13, 8 23 = 2Gc23
Under the further assumption C33 = 0 which is in effect, these reduce to
(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
where U3 is given by (1.30). It follows from (2.3)-{2.5) that the generalized

stresses are
(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
N11 = h[{2J.L + ,X)Wl,l + ,XW2,2],
(2.13) { N 22 = h[(2J.L + ,X)W2,2 + ,XW1,1],

N12 = hJ.L{Wl ,2 + W 2,1),
NO.3 = Gh{Uo. + W 3,o.),
M11 = ~~ [{2J.L + ,X)Ul,l + ,XU2,2],

(2.14) M22 = ~~ [(2J.L + ,X)U2,2 + ,XU1,1],
3
M 12 = h12J.L ( Ul ,2 + U2 ,1 ) .
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
(2.16) COt/3[e]iOt/3 = h[AeOtOt i/3/3 + 2JLeOt/3iOt /3j, Ve, i E S,

where C[e] := (COt/3[e]). For any differentiable function w = wOtaa : n I---t
JR2 define e(W) E S by
(2.17)
Then we see that
Thus the Reissner model may be expressed as
(2.18)
Gh 8~Ot (UOt + W3 ,Ot) + F3 = mW3 ,tt,

(2.19) {
h2 8
12 8xOt COt/3[e(U)]- Gh(U/3 + W3,/3) + C/3 = I p U/3,tt.
2.2. Linearization of Edge Conditions. The same argument as in
the last subsection shows that the linearization of the edge conditions
(1.33), (1.43) leads to
Vot N Otj -
- f j, Vot M
Ot /3 -- c/3 on rD ,
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)
GhVOt(UOt + W3 ,ot) = fa,

(2.21) {
h2
12VOtCOt/3[e(u)] = c/3 on rD.
Of course, one retains the geometric boundary condition
(2.22)
It is seen that w = Waaa is not coupled to W3, U = Uaaa in either the

equations of motion or the boundary conditions.
2.3. Hamilton's Principle for the Reissner Model. It is easily

verified that the system (2.18)-(2.21) is equivalent to the variational prin-
ciple
(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).
2.4. Linearization of the Vector Rotation Angle. We seek the

linearization of the vector rotation angle (1.53). It follows immediately
from (1.52) that a linearizes to
(2.27)
To linearize (1.51) we retain up to quadratic terms on each side of that

equation. The approximations of a~ and a~ are
a~ = [1 - (1/2)1/1~]a1 + 1/111/12a2 -1/12a3,

a; = [1 - (1/2)1/1~]a2 + 1/11a3'
We also have
la1 + W i ,18i1 -1 = 1- W1,1 - 1

21W,1 12 + 2W1,1
3 2
+ ... ,
Ia2 + Wi ,28i 1-1 = 1 - 11 W,2 12 + 2
W2,2 - 2 3 W 2,2 + ...
2 ,
and therefore
, t ai . (al + Wi ,I8i)
al' 1 = lal + Wi ,I8i1
= 1 - "211 W,1 12 +"21W1,1

2 - "2tP2
1 2- tP2 W3,1 + ... ,
a ,2 . t2 = 1 - "211 W,2 12 +"21 W22,2 - "2tPl

1 2 + tPl W3,2 + ... ,
a~ . t2 = W1,2 - W1,2 W2,2 + tPl tP2 - tP2 W3,2 + ... ,

a'2 . tl = WI ,2 - W2,
1WI 01'1'W3
,1'+f ,1 + ... .
It follows that
(a~ . tl){a~ t2) + (a~ . t2)(a~ t 1) = 1 + W1,2W2,1
- ~{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
(a~ . tl){a~ . t 2) + (a~ . t2){a~ . t 1) = 1- ~(Wl'2 - W 2,1? + ...
and therefore from (1.51)
tP32 ~ 4"1{W 1,2 - W 2,1) 2 = 4"1

1 2
curl wi ,
where w = Wa8a. Since curl w = {W2,1 - W 1 ,2)a3 corresponds to coun-

terclockwise rotation about a3 we take
(2.28)
It follows from (1.53)-{2.28) that the linearization of the vector rotation

angle is
(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
(2.30) ( = -(T' u)v + (v U)T + -1 (&w

T' - - V -&w) a3,
2 OV OT
where w = Walia and u = Ualia.

2.5. The Kirchhoff Plate Model. The Kirchhoff model corresponds
to the absence of transverse shearing. This may be expressed by saying
that A3 is a unit vector which is normal the the deformed reference plane
or, what is the same thing, that <Pi = 'l/Ji. In the linear situation, this means
that
(2.31)
To obtain the Kirchhoff model in a formal way, we eliminate Ua + W3 ,a
between the equations in (2.19). This leads to the equation
DUa,a{3{3 + Ca,a + F3 - mW3,tt = IpUa,att,
where D = {2J.L + >..)h3 /12. If (2.31) is used to substitute for Ua in terms
of W 3 ,a the result is the Kirchhoff plate equation
(2.32) mW3,tt - I pW3,aatt + DW3,aa{3{3 = F3 + Ca,a.
The geometric boundary condition for W3 on r C follows from (2.22) and
(2.31):
oW3
(2.33) W3 = 0, - - = 0 on
ov rC.
To obtain an edge condition for W3 on rD we substitute (2.31) into the
last two equations in (2.21), multiply the first of these equations by Vb the
second by V2 and add the products. The result may be written
(2.34) - D[W3,aa + (1 - ,8)B1(W3)] = v c on rD,
where ,8 = >../(2J.L + >..) satisfies 0 < ,8 < 1 and where
B 1(W3) = 2VIV2W3,12 - V~W3,22 - V~W3,11'
This boundary condition is a balance law for the bending moment about
the tangent vector 'T. There is a second edge condition for W3 along rD ,
representing the balance of shear forces, which cannot be deduced from
purely formal calculations on (2.20) and (2.21), using (2.31). It is
(2.35) - D [:v W3,aa + (1 - ,8) ! B2{W3)] + Ip :v W3,tt
=
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
3. Systems of Linked Reissner Plates

For i = 1, ... ,n let at, a~, a~ be a right- handed orthonormal system in
IR3, let pb be a fixed vector in IR 3, and ni be a bounded, open, connected
region in IR2 with Lipschitz boundary. Summation convention is no longer
assumed but, as above, Greek indices take values 1,2 and Lj := LJ=l'
We set
Pi{Xb X2) = pb + LXaa~,
a
'Pi = {Pi (Xl, X2)! (Xl, X2) E nil

The mapping (XbX2) 1--+ Pi(XbX2) is a homeomorphism ofOi onto 'Pi and
of 8n onto 8'P, with
Pi-1 (P) = (( P - Poi) . aI' i) . a i)

i ( P - Po
2
The set 'Pi is considered as the reference surface of a thin plate whose
reference configuration is
where h is a positive constant not depending on i. We make the following

assumptions.
'Pi n 'Pk = 0, 'Vi =1= k;

Ui=l 'Pi is a connected set in IR3;
'Pi n 'Pk is either empty or is a finite union of components, 'Vi =1= k
Each plate is assumed to satisfy the basic kinematic assumption (1.1):

the position vector Ri{Xb X2, X3) to the displaced particle originally at
Pi{Xb X2) + x3a~ is given by
(3.1) Ri{Xb X2, X3) = pi{Xb X2) + X3A;{Xb X2),

(XI, X2) E ni , IX31 < hj2, 1 = 1, ... ,n,
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
3.1. Geometric Junction Conditions. The joints, or junction re-

gions of the multi-plate system are defined exactly as in Section 2.1. Thus
each joint is an open line segment that is a common edge of two or more
reference surfaces. As in Section 2.1, it may be assumed without loss of
generality that the joints are mutually disjoint. The index set (J) of a
joint J is
I(J) = {i E [1, ... , nll J C Wi}.
Thus I(J) is the index set of those 'Pi which share J. We shall distinguish
several types of joints by considering various geometric constraints on joint
deformation.
(I) J is a connected joint if
Condition (3.2) simply requires continuity of displacements along the joints.

Let Ti denote the unit tangent vector to Wi that is positively oriented
with respect to a~.
(II) J is a hinged joint if it is a connected joint and if
(3.3) (a~i) 2 Wi(p;l(p)) = 0, Vp E J, Vi E I(J).
Condition (3.3) means that J remains straight under deformation.

Let ~i denote the vector rotation angle associated with the deformation
of 'Pi'
(III) J is a semi-rigid joint if J is a connected joint, if
(3.4) ~i(p;l(p)) = ~j(pjl(p)), Vp E J, Vi,j E I(J),
and if
Conditions (3.4) and (3.5) require continuity of vector rotation angles

and vanishing of the in-plane twisting strains, respectively, along the joint,
but allow for bending along the length of the joint and rotation of the entire
multi-link system around the joint.
(IV) J is a rigid joint if J is a semi-rigid joint and if
(3.6) (a~i )2 Wi(p;l(p)) = 0, Vp E J, Vi E I(J).
Thus a rigid joint is a semi-rigid joint which remains straight under

deformation.
It is obvious that other ad hoc, but physically reasonable, geometric
constraints on the deformation of J could be imposed; for example, one
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'
3.2. Linearization of the Geometric Joint Conditions. All of the

geometric joint conditions are linear in the displacements with the excep-
tion of those appearing in the definition of a semi-rigid joint. Let us examine
those conditions in the linear context of the last section. Let Vi = La v~a~
denote the unit exterior normal vector to ffPi . Then Ti = ---v~ai + v1~ is
the unit tangent vector to ffPi that is positively oriented with respect to
a~. It follows from (2.30) that (3.4) then takes the form
(3.7) --- (T i ui) Vi + (i

V ui) T i + -1 ( T'
i {)wi
- {)wi)
. ---V i . - . a3i
2 av' aT'
= ---(TJ. . uJ)v
..
J + (v
'"
J . UJ)TJ + -21 ( T J. -
{)wj . {)wj) .
. --- v3 . ----. a J
avJ aT) 3
while (3.5) becomes

a il . w,2
i + a2i . W,li = 0.
The latter equation is the same as

. {)wi . {)wi
T'-. +v'-. =0
av' aT'
(3.8) ___ (Ti . Ui)vi + (Vi. Ui)Ti ___ (Vi. {)w~) a~

aT'
= ---(T j . uj)vi + (v j Uj)Tj --- (v j ::) a~.

Along J one has Ti = Tj for every i, k E I(J). We define
if Ti = Tj on J
(3.9)
ifTi = ---Tj on J.
It follows that (3.8) is equivalent to
(3.10)
for all i, k E I{J). However, as will be proved in a moment, the continuity

condition (3.2) implies that
(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
and therefore (3.11) reduces to
where VW3 = La W3,aa~. If the cardinality i ofI{J) is greater than two,

or ifI{J) = [i,j] and v' i= -vJ, then (3.13) implies that
Tt . (u' + VWa) = O.
On the other hand, if I{J) = [i,j] and v t = -vJ , then a~ = -atJa~ so

the continuity condition Wt = WJ implies W3 = -atJ wg and so Tt .
VW3 = -TJ . vwg. Therefore (3.13) simplifies to T' . u t = TJ . u J which,
when combined with (3.1O), implies that u' = -atJ u J Thus with (3.2),
conditions (3.4) and (3.5) linearize to
Ut = -atJuJ if I{J) = [i,j] and v' = -vJ,

(v' . u t ) = atJ(vJ . uJ ), T t . (u' + VW3) = 0
{
(3.14)
otherwise, Vi,j E I(J).
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.
REMARK 3.1. In the absence of transverse shearing, the condition Tt .

(ut + VWa) = 0 is automatically satisfied since u' = -VWa in this case,
while (3.10) then becomes
( )
3.15
aWa
avt (P t-1{p)
) avJ (PJ-1{P ,
= a'J awg \fp E J.
It will be shown in section 6 that (3.15) is the linearization of the nonlinear

constraint
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
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
.
It also follows from (3.17) that
hence
(3.19)
Since a~ . a~ = a'J(v' . v J } and a~ . v J = -a'J(a~ . v'}, (3.12) follows from

(3.19).
3.3. Dynamic Joint Conditions. We wish to derive the balance laws

of linear momentum and angular momentum at a joint. As usual, this
is done through a variational formulation of the boundary value problem
associated with the structure in which the variation is taken with respect
to test functions which satisfy, in particular, the geometric joint conditions.
In what follows attention will be restricted to lznear dynamics and con-
straints. Set A, = {)P,. We assume that
(3.20)
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
Ci[C]= h[Ai(cll + c22)I + 2JLiC], 'ic E S.

For C, E S we write (c, ) = La,[3 Ca[3a[3. The dynamic equations of the
multi-link system of plates are postulated to be those associated with the
Hamilton's principle for a system of linked Reissner plates (cf. Section 2.3):
(3.23) 6ln
T
[~(t) - U(t) + W(t)] dt = 0,
where
W(t) = t, {lni (F~ . Wi + F~W~ + c i . u i ) dO
+ l/ti. w; + Jiw; +c; U;)dl'}.

Here Fb = La F~a~ denotes the resultant in-plane component of applied
force in the i th plate, C i = La C~a~ is the resultant in-plane component
of the applied couple, and similarly for f~ and c i . The variation in (3.23)
must be taken with respect to displacements which respect the geomet-

ric junction conditions imposed along J 1 , .. ,Jm , as well as the clamped
conditions (3.22).
As a result of the calculation (3.23) we obtain, in particular, the Reiss-
ner plate equations (2.18), (2.19) for each plate in the multi-link system,
and the boundary conditions (2.20), (2.21) along each rp' We write the
equations of motion in the following manner:
miw:tt - divCi[c(wi )] = Fh,

(3.24) miW~,tt - Gihdiv(ui + V'W~) = F~,
Ipi u:tt - ~; div Cdc (ui )] + Gih(ui + V'wJ) = C i,

where
The boundary conditions along rp may be written

Cdc{Wi)]Vi = fJ,
(3.25) Gihvi . {ui + V'wD = f~,

h2 ... D
12 Ci[c{u')]v' = c' on r i ,
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
which can hold if and only if
(3.27) L
iEI(Jk)
1 h
({Ci[e(wi)]v i + Gihvi . (ui + VW~)an Wi(pil(p))
+ ~; 6.i . (Cde(Ui)]Vi)(Pil(p))) dJk = 0, k = 1, ... ,m.
The dynamic junction conditions may be read off from (3.27).
3.3.1. Dynamic conditions at a connected joint. If Jk is a connected

joint, then
Wi(pil(p)) = Wik(pi;;l(p)) Vi E I(Jk)'
Since Wik(pi;;l(p)) is arbitrary, as is 6.i (i E I(Jk)) one obtains from (3.27)
(3.28)
L {Ci[e(Wi)]Vi + Gihvi . (ui + VW~)a~}(pil(p)) = 0, Vp E Jk,
iEI(Jk)
(3.29) (Ci [e(Ui )]Vi)(pil(p)) = 0, Vi E I(Jk), Vp E Jk.

Equation (3.28) is a balance law for forces along the joint while (3.29) states
that the bending moment of each plate sharing the joint Jk must vanish
there.
3.3.2. Dynamic conditions at a hinged joint. In this case, the variational
equation (3.27) must hold for arbitrary (sufficiently regular) 6.i and for Wi
satisfying (3.2) and (3.3). One therefore obtains once again the condition
(3.29) for the vanishing of bending moments. In addition,
1
(3.30)
L {Ci[e{Wi)]V i + Gihvi . (ui + VW~)an Wik{pi;;l{p)) dJk =
iEI(Jk) Jk
for all functions W ik of the form

Wi"{pi;;l{p)) = C~ + sC~, (p = q~ + sri,,),
where C~, C~ are fixed but arbitrary vectors in IR3. It is then a consequence
of (3.30) that the following non local balance laws must hold along Jk:
(3.31) L
iEI(h)
l ak
bk
{Ci[e{Wi)]V i + Gihvi . (ui + VW~)ands = 0,
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~.
3.3.3. Dynamic conditions at a semi-rigid joint. We have once again
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
+ ~; [(V', 0')0 (U') + (T" 0' + : : ) OT'(U')
, . [) .
+ 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
(3.35) L {Ci[(wi)]V i + Gihv i . (ui + VW~)an

iEI(Jk)
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
(3.10) and (3.14) in (3.36), we obtain the following conditions on balance

of moments, where (J'i(Jk) := (J'iik:
(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
(3.39) Gihv i . (ui + VW~) = (J'iiGihvi . (ui + vw1),

Ci[c(wi)]Vi = -Ci[c(wi)]vi if I(Jk) = [i,j] and vi = -vi.
Furthermore, at a serial joint, (3.37) reads
C".(ui ) = (J'iiC,,;(Ui )
which, when combined with (3.38), yields
Therefore, the dynamic junction conditions at a semi-rigid serial joint are

(3.39) and (3.40); at a semi-rigid non serial joint they are (3.35) and (3.37).
As is to be expected, at a serial joint there is no coupling in either the geo-
metric or dynamic junction conditions between the in-plane displacements
Wi, wi and the components w~, w1, u i , u i related to bending.
3.3.4. Dynamic conditions at a rigid joint. We write (3.33) as
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 ),
hence W~ = (C2 + sCl) a~ and
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
for arbitrary vectors C2, Cl. By setting Cl = 0 the balance of average

forces law (3.31) is obtained. In addition, we deduce the following balance
law for angular momentum along Jk:
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
where Iih is the moment of inertia of Jk, and
W(tl = t, {k/ Fi . Wi + C i . nil d{l + /r'. W' + ci . nil dI' }
+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))
+ ~~ ui . (Ci[C(Ui)]vi)(Pi 1 (P))) dJk

+ { [(mk W:tt - fk) . Wik + (IPk u:tt c k ) . Uik ] dJk = 0,
1Jk
-
k=l, ... ,m.
The dynamic junction conditions may be derived from (3.42)
3.4.1. Dynamic conditions at a connected joint. By use of the geometric

joint condition at a connected joint one immediately obtains from (3.42)
the following conditions along Jk.
(3.43) L {Ci[c(wi)]Vi + Gihv i . (ui + VW~)an = fk - mkW:tt ,

iEI(Jk)
Conditions (3.44) indicate that at a connected joint neither the rotational

energy nor the applied couple are transmitted to the connecting plates of
the system.
3.4.2. Dynamic conditions at a hinged joint. One obtains, once again,

(3.44) for the balance of bending moments, while (3.43) is replaced by the
non local balance laws
(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
3.4.3. Dynamic conditions at a semi-rigid joint. The balance law for

linear momentum now takes the form
(3.47) L {Ci[c(wi)]vi + Gihvi . (ui + V'W~)a~}

iEI(Jk)
2
-_ - h 0 Cr' (U i) a3i
12 "L..J o-ri + r-k - -
mk
Wik
,tt
iEI(Jk)
while the balance law for angular momentum is
(3.48) L O'i(Jk)Cv.(Ui ) = V ik . (c k - Ipku~~t)

iEI(Jk)
3.4.4. Dynamic conditions at a rigid joint. The balance law (3.48) is

retained, but (3.47) is replaced by (3.45) together with
4. Well-posedness of Systems of Linked Reissner Plates

In this section it will be demonstrated that the mathematical model of
a system of linked Reissner plates derived in the last section comprises a
well-posed mathematical problem in a function space appropriate to the
model.
Let 0, a1,a2,a3 be as defined at the beginning of this chapter. For
any unit vector a E 1R3 and s ~ 0 let H8(O; [a]) denote the Hilbert space
consisting offunctions lOa, where IE H8(O), the standard (real) Sobolev
space of order s. Set
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
A function R in H (resp., in V) is given by the pair R = (W, u), where

W() = Lj Wj()aj, u() = La UaOllo with Wj, Ua E HO(O) (resp.,
Wj, Ua E H1(O)). The space V is dense in H with compact injection.
Let us further introduce a continuous seminorm on V, denoted by 11(')
and defined by
I1(R) = (In ((C[c(w)],c(w)) + ~;(C[c(u)],c(u))

+ Ghlu+ VW312}dO) 1/2 ,
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
(4.1) IIRliv ::; K[I1(R) + IIRIIH], VR E V.
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,
(4.2) IIRliv ::; KI1(R), VR E V such that R= 0 on r o,

where r is any nonempty, relatively open subset of a~.
4. WELL-POSEDNESS 287
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
If R E H then R = (R1, R 2 , . ,Rn), where
(4.3) Ri = (Wi, u i ), Wi =L Wja~, Ui =L U!a~.

j a
The norms on H and V are defined to be, respectively,
and
IIRliv ~ (t.IIRilll" ) 'I'

A continuous seminorm a() is defined on V by setting
where
ai(Ri ) = (In; ((Ci[c(Wi)],c(Wi )) + ~;(Ci[C(Ui)l,c(Ui))

+ Gihlui + VW412}dO) 1/2 ,
It follows immediately from (4.1), applied in each region 0i using the spaces
Vi, Hi and the seminorm ai, that
IIRllv :::; K[a{R) + IIRIIHl, \fR E V.
Let V denote the closed subspace of V consisting of functions R =
(R1 , ... ,Rn) such that
Vi,
.
(4.4) Ri E
(4.5) Ri = 0 on r~ ,
(4.6) R satisfies the geometric junction conditions.
For instance, if Jk is a connected joint, (4.6) requires that
Wi(p;l(p = Wj{pj1(p, tIi,j E I(Jk),
while if Jk is hinged it is also required that
In each of these two cases there are no constraints on the u i components

along Jk. Such constraints would be present, however, if Jk is semi-rigid
or rigid, according to the definitions of the last section.
The following lemma is the analog of (4.2) for systems of linked plates.
LEMMA 4.1. Assume that rf

i= 0 for at least one index i and that each
joint is either semi-rigid or rigid. Then
(4.7) a(R) ~ kllRllv, VR E V,

for some k > o.
PROOF. Without loss of generality, we may assume that rf i= 0 and

that each joint is semi-rigid. Since the map V 1-+ 1R+ : R 1-+ IIRIII2(rf)
defines a continuous seminorm on V, it is sufficient to prove that
(4.8) a(R) = 0, RIlrc1 = 0 ===} R = o.

Since a(R) = 0 implies ai(Ri) = 0, i = 1, ... ,n, it follows from (4.8)
and (4.2) that R I = O. Suppose that 'P2 is connected to 'PI along a joint
J. According to the definition of a semi-rigid joint we have, for all P E J,
(4.9)
(4.10)
(4.11) [T I . (u l + V'wj)v l ](pll(p)) = [T2 . (u2 + V'W;)V 2](P21(p)).

Since RI = 0, it follows from (4.9) and (4.10) that
(4.12)
so that, in particular,
(4.13)
Then (4.11) gives
(4.14)
Therefore R2 = 0, and so forth. 0

4. WELL-POSEDNESS 289
4.2. Existence and Uniqueness of Solutions. For reference, let us

write down the complete system under consideration.
Equations of motion. For i = 1, ... ,n,
miw:tt - div Ci[c(Wi )] = Fb,

(4.15) miW~,tt - Gihdiv(ui + vwD = F~,
IPi u:tt - ~~ div Cdc(ui )] + Gih(ui + vw;) = C i.

Boundary conditions along rf, i = 1, ... ,n:
(4.16)
Boundary conditions along rp, i = 1, ... ,n:

Ci[c(Wi)]Vi = t6,
(4.17) Gihv i . (ui + VW~) = f~,

h2 ...
12 Ci [c(ut )]v t = ct.
m addition, the geometric and associated dynamic junction conditions must
hold along the joints J 1, ... ,Jm . If, for instance, Jk is a semi-rigid, non
serial joint, the geometric conditions along Jk are
W i (Pi 1(P)) = Wi (pjl (p)),

(4.18) { (vi. u i )(pil(p)) = (vi. Ui )(pjl(p)),
-r i . (ui + VW~)(pil(p)) = 0, Vi,j E I(Jk), p E Jk,
while the dynamic conditions along Jk are
(4.19) L {Cdc(wi)]vi + Gihv i . (ui + VW~)an

iEI(Jk)
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.
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
a,(R"it') = [ ((C,[c:(w')],c:(w')) + ~~(C,[c:(u')],c:(fI'))

In.
+ G,h(u' + VW3) . (fIJ + VWl)}dO, VR, it E K
Regardless of what the specific geometric and dynamic junction conditions
happen to be, a solution of the system of linked plates is, by definition, a
solution of the variational equation
(4.21) [T [(R,tt, it) + a(R, it)] dt =

Jo
t Jo[T In.[
,=1
(F' W' + C' fI') dOdt
or, equivalently, for almost all t > 0,
(4.22) (R,tt,R)v + a(R,R) = ~ In. (F' W' + C' fI')dO

+ t1
,=1 r~
(f'. W' + c' . fI') dr, vR E V.
Equation (4.21) is exactly the variational equation associated with the

Hamilton's principle which was used to derive the system in the first place.
As is usual, the geometric boundary condition (4.16) and the geometric
junction conditions are built into the space V while, at least formally, all of
the dynamic requirements are satisfied as a consequence of the variational
equation (4.22). We have not as yet specified the regularity a solution is
required to possess; of course, this will depend on the assumptions placed
on the data F', C' , f' ,c' and on the initial data
(4.23)
Let us first consider the case of homogeneous boundary data, f' = 0,

c' = 0, and assume of the rest of the data that
(4.24)
4. WELL-POSED NESS 291
The right side of (4.21) may be rewritten
where
G = (Gl, ... ,Gn ) E 2(0,TjH), G i = (m;lFi,I;/C i ),

(4.25) (R,tt, it)v + u(R, it) = (G, it)H, V'it E V.

The bilinear form u(, .) is nonnegative on V. If this form is strictly coercive
on V, the problem (4.25), (4.23) falls within the framework of standard
variational theory as may be found, for example, in Lions-Magenes [62,
Chapter 3J. Otherwise, the change of variable R = e6t R (c5 > 0) leads to
the equation
(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.
THEOREM 4.1. Under assumptions (4.24), the problem (4.25), (4.23)

has a unique solution R with regularity
(4.26) R E C{[O, TJj V) n C1([0, Tjj H).

It follows from (4.25) and (4.26) that
R,tt E C([O, T]j V') n 2(0, Tj H).

Now consider the case when the boundary conditions are non homoge-
neous. Without loss of generality we may assume that Fi = 0 and C i = O.
Write
and let us assume that
Then
for some constant C(fi , Ci ). It follows that there exists an element g E

L2(0, T; V') such that
t lD
i=1 ri
(i. Wi + ci . ui ) df = (g,R)v, 'VB. E V,
so that (4.22) may be written
(4.28) (R,tt, R) + a(R, R) = (g, R)v, 'VB. E V.

Once again, standard variational theory may be applied in order to deduce
the following result.
THEOREM 4.2. Under assumptions (4.27), the problem (4.28), (4.23)

has a unique solution R with regularity
(4.29) R E C([O, T]; H) n C1 ([0, T]; V').

In the context of the last theorem, equation (4.28) must be understood
in D A - D~ duality, as in Section 2.4.
REMARK 4.1. As in the case of systems of linked membranes, the issue of

regularity of the solution remains open, in particular, under what conditions
on the data will the system admit a classical solution? This question is open
even for the pure initial value problem (Le., Fi = fi = 0, C i = c i = 0). The
main question, of course, is regularity of the solution in a neighborhood of
a junction region (cf. Remark 2.4).
5. Controllability of Linked Reissner Plates

It is possible to prove exact controllability results for interconnected
Reissner plates similar to those obtained in Chapter VI for interconnected
membranes. However, derivation of the necessary observability estimates
is even more computationally intensive than before and in the end the
results could only be considered as formal, depending as they necessarily
will on the assumption that solutions of the homogeneous system have the
necessary regularity to justify the computations. The desired regularity
properties will depend on the geometry of the configuration of plates in
an unknown (to us) manner and it seems difficult to resolve this problem.
For this reason, we shall consider the exact controllability question only for
tmnsmission problems for thin plates, where regularity of solutions can be
made precise.
Let us consider a bounded, open, connected set 0 in 1R.2 with Lipschitz
boundary consisting of a finite number of smooth simple arcs, a bounded,
simply connected open set 0 1 with smooth boundary such that 0 1 C 0,
and set
5. CONTROLLABILITY OF LINKED REISSNER PLATES 293
We consider the following system of dynamic equations, transmission con-

ditions and boundary conditions (where' = 8/at):
(5.1) { Ipi~i.- divCi[c(ui )] ~ Gih.(Y'W + u i i) = 0,

miW' - Gihdiv(Y'W' + u') =0 in ni , t > 0, i = 1,2;
WI = W 2 , u l = u 2 ,
(5.2) { Glh(Y'Wl + u l ) . VI + G2h(Y'W2 + u 2 ) . v 2 = 0,
Cdc(u l )]v l + C2[c(U2)]V2 = 0 on r l , t > 0;
(5.3) W2 = 0, u 2 = 0 on r C " t > O'
(5.4)
Initial conditions are
(5.5) Wilt=o = Wilt=o = 0, uilt=o = liilt=o = 0 in ni , i = 1,2.

Equations (5.1)-(5.4) are those governing the transverse displacements
Wi and the rotations u i = (ut, u~) of two serial Reissner plates connected
by a semi-rigid joint r l (see (3.14), (3.39) and (3.40)) except that we have
written Ci instead of (h 2 /12)Ci . In (5.4), f and c = (Cl' C2) are the control
inputs. Of course, there are other equations and transmission conditions
for the in-plane displacements, but we have already studied exact control-
lability of this system in Section 3.4.2, Chapter VI.
5.1. Controllability in 'fransmission Problems for Thin Plates.
5.1.1. Observability Estimates and Consequences. Let i,j, k denote the
natural basis for IR3 and set
'}-l8(Oi) = H 8(Oi; [i]) EB H 8(Oi; u]) EB H 8(Oi; [k]),
Hi = '}-l(Oi)' Vi = '}-ll(ni)' H = HI X H 2, V = VI X V2.
If ~ belongs to H or V, then ~ = (~1, ~2) where
~i = <pi + Zik, <pi = L CP~3a.

As a norm on H we take
Let
and define
Then II . Ilv is a semi-norm on V which is a norm equivalent to the 11llv

norm on V if r C =f:. 0. To simplify the discussion, such shall be assumed to
be the case.
We also consider the homogeneous problem
{ Ip;~' - divCi[(l;'i)]. + G~h('VZi + I;'i) = 0,

(5.6)
miZ' - Gih div('V Z' + 1;") = in ni , t > 0, i = 1,2;
Zl = Z2, 1;'1 = 1;'2,

(5.7) { G h('V Z1 + 1;'1) . v 1 + G2h('V Z2 + 1;'2) . v 2 = 0,
1
Cd(1;'1 )]v 1 + C2[(1;'2)]V 2 = on rb t > 0;
(5.8) Z2 = 0, 1;'2 =0 on r C , t > OJ
(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
L 1{mi ZiZi + Ip;~icPi + Gih('VZi + I;'i) . ('Vz i + cPi )

2
i=l {l;
. A A A1 A2
+ (Cd(I;")],(l;'i))}dn= 0, VcfJ=(cfJ ,cfJ )EV,
i
+ Z'k.
A A A
where C) = cpi Equivalently,
(5.11) { ~ + Ac) = in V',

C)!t=o = C)o, ci>!t=o = C)1'
Standard theory gives: if (C)o, C)1) E V x H (resp., DA x V), (5.11) has a

unique solution with
C) E C([O,T);V) x C 1 ([0,T);H) x C2 ([0,T);V')
(resp.,
C) E C([O, T); DA) x C 1 ([0,T);V) x C2 ([0,T);H)).
Let ("\}II,"\}I2) E DA. Since r 1 is smooth and strictly contained in 0,
one may use standard techniques of elliptic regularity to show that C)1 E
1{2(01)' From some results of Nicaise [78) one also has C)2 E 1{8(02) for
some s E (1,2) which depends on the angle of intersection of r C and r D ,
where s > 3/2 if this angle is strictly smaller than 7r (measured in the
interior of 0). In this case, C) will satisfy the transmission conditions in
the sense of traces.
The observability estimate to be proved is as follows.
PROPOSITION 5.1. Assume that r C and r D meet in a strictly convex

angle and that there is a point Xo E 0 1 such that
(x-xo).V1:2:0 onr 1,
(5.12) { (x - xo) . v 2 :2:
on r D ,
(x - xo) . v 2 ~
on rC.
Suppose further that
PI ::; P2, G 1 :2: G2, J.L1 :2: J.L2, }.1 :2: }.2

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,
II (C)o, C)l)lI~xv' ::; CT {T ( !C)!2drdt.

Jo Jr D
From this proposition one may immediately deduce a reachability theo-

rem for the problem (5.1)-(5.5). Write
R=(R1,R2), Ri=Ui+Wik, f=c+fk.
If f E .c2 (ED), where ED = rD x (0, T), problem (5.1)-(5.5) has a unique

solution R with
R E C([O, Tj; H) x C 1 ([0, Tj; V') x C2 ([0, Tj; (D A )')
(see Theorem 4.2). Let
RT = ((R(T) , R(T)) If E 2(~D)}.
THEOREM 5.1. Under the assumptions of Proposition 5.1, for T > To

one has V x H C RT.
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
PROOF OF PROPOSITION 5.1. The proof is based on the following iden-

tities, in which we have written
= 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
Pi(t)=mi [ Zi(m. V'Zi + 6Z i ) dO,

In.
Qi(t) = Ip. [ cPi . (V'rpim + 6rpi) dO,
In.
where m(x) = x - Xo and {j E JR. Then
(5.13) Pi{T) - Pi (0) + (1 - 15) 1 Q.

mil Zi l2dQ
- [ Gih(m. V'Zi + t5Zi ) div(V'Zi + rpi) dQ = ~i [ (m. vi)IZiI2d~;

4 h
(5.14) Qi(T)-Qi(O)+ 1 {(1-t5)Ip.lcPi I2 +t5(Ci [c:(rpi)],c:(rpi))}dQ
1
Q.
+ Gih(V'rpim + t5rpi) . (V'Zi + rpi) dQ

Q.
1 [ {. arpi . . arpi }
2 lE. (mv')
2 ..
= Ip.lrpil + (Ci[c:(rp')Jv') avi - (Ci[c:(rp')JT') aT i d~
+ l. (Cdc:(rpi)Jv i ) . [(m. Ti) ~~: + t5rpi] dL

PROOF. The relation (5.13) follows from
0=1 Q.
(m V'Zi + t5Zi )[m,ii - Gihdiv{V'Zi + rp')J dQ
upon noting that
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, [
To prove (5.14) we form
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
where C~{3 are the components of the matrix Ci . According to (2.16) we

have
L C~{3[clcQ{3 = ~~ [Ai(L cuu)2 + 2/l-i L c~~l

Q~ u 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
= Qi(t) - Qi(O) + {(I - 8)Ip.lrPi I2 + 8 L C~~[c(/pi))cQ~(/pi)} dQ

Q. Q,~
-H. (m . ",+,.1";'1 2 + ~ Cf~I*"')1 [v~ :j -7~ ~ J} dE

-1 L v~C~~[c(/pt)l [(m'
E. Q,{3
Ti) ~j + 6CP~l dE.
When written in vector form, it is seen that the integrals over Ei in the
last relation are exactly those on the right side of (5.14), while the integral
over Qt on the right is the same as the first integral on the left side of
(5.14). 0
Let us replace 8 by (1 - 8) in (5.14) and then add the result to (5.13).

We obtain
(5.15) ~(T) - ~(O) + 1 Q.

{(I - 8)m,IZ'12 + 8Ip .lcP'1 2
+ (1- 8) (C, [c(<p')], c(<p'))} dQ

+ 1
Q.
G,h{(V'<p'm+(1-8)<p')(V'Z'+<p')
- (m V'Z' + 8Z') div(V'Z' + <p')} dQ

= ~ r (m. v') {m,IZ'1 2 + I .lcP'1
JE. p 2
+ (C,[c(<p')]v') . ~<p' - (C,[c(<p')]T') ~<p'} dE

uv' uT'
+ l. (C,[c(<p')]v') . [(m. T') ~~: + (1 - 8)<P'] dE,
where ~(t) = P,(t) + Q,(t). By an elementary calculation we obtain
(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.
Use of (5.16) in (5.15) gives the identity
LEMMA 5.2. Let ~ = (~1, ~2) be a sufficiently smooth solution of (5.6)-

(5.9). Then
L 1{(I - 8)m,IZ'12 + 8Ip.I~'12

2
(5.18) R(t) - R(O) +
,=1 Q.
+ (1- 8)(C,[c(cp')],c(cp')) + 8G,h(IVZ'1 2 -lcp'12)} dQ
= .1(ED) + .1(EC ) + .1(Ed,
where R(t) = R 1 (t) + R2(t) and
.1(ED) = ~ hD (m v 2) {P21Z212 + I p2 1cP212} dE
-~ L (m ",) {(c,[e(IC')]r')' Z: + (:!: +r'~') }~,

G,h
PROOF. W; sum (5.17) for i = 1,2 and note that
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
= { (m VI) [(PI - P2)IZ 1 12 + (IPI - I p2 )1<p 1 12] d1:,

J~l
since ZI = Z2 and ~1 = ~2 on 1: 1 Also, in view of the transmission

conditions (5.7),
2 . { . . ocp'. . . ocp'.
'" { (m v') (Ci[e(cp')]V')' - . - (Ci[e(cp')]-r') . - .
,=1 J"'.
~ u.
oV' o-r'
By noting that
one finds 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
PROOF. On r one has

Therefore, since Cf{3 = cfa , on r C we have
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
The estimate (5.19) therefore follows from (5.18). 0
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
Then there is a To > 0 such that if T > To,
PROOF. Our assumptions assure that the solution ~ has the regular-
ity required in the two previous lemmas. We first note the elementary
inequality
Also, since r C =1= 0,

Therefore, if 0 > 0 is small enough we have the estimate
~
2
i=l
1 Q.
{(I - o)milZil2 + Mp.l~iI2
+ (1 - o)(Cdc(cpi)], c(cpi)) + oGih(lV' zil2 _lcp i I2)} dQ

~ r . 2 2 2
? Co L.J J(. {mtlZil + Ip.lcpil + (Ci[C(CPi)],C(CPi)) + GihlV'Zi + CPil }dQ
i=l Q.
= Co loT (l14!(t)ll~ + 11()(t)II~) dt = CoTII(()O, )II~XH

()1
for some constant Co > O. Furthermore,
IR(t)1 ~ C1 (0)1I(()(t), 4!(t))II~XH = C1 (0)11(()o, ()l)II~XH'

It then follows from (5.19) that
where To = 2C1 (0). Therefore, it remains to prove that
provided
(5.23)
The proof of (5.22) is completely analogous to that given for serially

connected membranes (cf. Theorem 3.2 and Remark 3.4, Chapter VI). To
see the connection, let us recall that for two serially connected membranes
or, more generally, in a transmission problem for membranes, the analog
of (5.22) is
(5.24) r (m. y1) {(81(cp1)y1) . Ocp1

JEI Oy1
_ (82(cp2)y2) . Ocp2
Oy2
1) 1) Ocp1
- (8 1 (cP T . OT 1 + (82 (cp 2)T2). Ocp2 }
OT 2 dE ~ 0,
where Si(Vi) = (s~/3(cpi)) is the stress tensor and
(5.25) S~/3(cpi) = 2J..Lica/3(cpi) + Ai8a/3 LC1'1'(cpi).

l'
The transmission conditions for the in-plane motion of serial membranes

are
(5.26) cp1 = cp2, Sl(cp1)v 1 + S2(cp2)V2 = 0 on r 1,
and it was demonstrated in the proof of Theorem 3.2, Chapter VI (where
things are written in terms of normal and tangential components) that
(5.24) is valid provided J..L1 ~ J..L2 and A1 ~ A2'
Now recall the relation between the strains ca/3(cpi) and the generalized
moments C~/3[c(cpi)] is
(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]).
6. Systems of Linked Kirchhoff Plates

As was noted in Section 2.5, the Kirchhoff thin plate model formally
corresponds to absence of transverse shearing in the Reissner thin plate
model, that is, Ua = -W3 ,a. An equivalent way of viewing the Kirchhoff
model is that it corresponds to the Reissner model with an infinite shear
modulus G = 00. In fact, a rigorous derivation of the Kirchhoff model as
the limit of the Reissner model as G --+ 00 may be found in [51]. Here
a similar approach will be followed in order to obtain a model for a sys-
tem of linked Kirchhoff plates. This program will be carried out under
the assumption that all joints of the linked Reissner system are semi-rigid
and non serial. This is done mainly for convenience; there is no essential
difficulty in treating other types of joints by a similar method and we shall
comment on these other situations in Section 6.3 below. In addition, since
our main interest is in determining the form of the junction conditions for
a system of linked Kirchhoff plates, it shall be assumed that all boundary
forces and couples vanish and, additionally, that rf i= 0 for at least one
index i. These assumptions do not alter the form of the junction condi-
tions but do allow for a simpler convergence analysis. Thus, the problem
under consideration is the limiting behavior as Gi --+ 00, i = 1, ... ,n, of
6. LINKED KIRCHHOFF PLATES 305
solutions of the system (4.15)-(4.20) with f~ = c' = 0, 13 = O. Let us

note that the model which will be derived may also be obtained by ad hoc
considerations similar to those used to derive the model of linked Reissner
plates, as was done in [48], but that approach would not reveal the close
relationship between the two models.
6.1. Semi-rigid Joints. We write G for the n-tuple (G 1 , . ,Gn ) of

shear moduli, G > 0 to mean that every G, > 0 and G - 00 to mean that
G, - 00 for every i = 1, ... ,n. Since rf "# 0 for some i, according to
Lemma 4.1 the seminorm e() defines a norm equivalent to the 11llv norm
on the space V. The space V and the norm e() on V obviously depend
on G and we write Va in place of V and IIRllvG in place of ea(R) when
REVa
A typical element of H is described by (4.3). We introduce the following
closed subspace Hoo of H:
Hoo = {R E HI W3 E H 1 (O,), u' = -V'W3, W3 = 0 on rf}.
If R E H oo , then R' = (W\ -V'Ws ) and
An element R in the orthogonal complement H~ of Hoo in H is charac-

terized by the variational equation
t 1n.
t=l
(m,W' vr - Ip. u' V'W3)dO = 0,
which is formally equivalent to the system
w = 0, m,W3+ Ip. divu' = 0 in 0" v' u' = 0 on r, \ rf.

The decomposition of an element R E H into Roo +R~, where Roo E Hoo
and R~ E H~, is given by
= w' (w').1 = 0
w'00 '00 ,
m,(W3)oo - Ip. div(u~).1 = m,W3, -V'(W3)oo + (u~).1 = u"

(W3)oo = 0 on rf, v' (u~).1 = 0 on r, \ rf.
We similarly define a closed subspace Voo of Va by
Voo = Va n Hoo = {R E Val W3 E H 2 (O,), u' = -V'Wn

Since we are assuming that all joints are semi-rigid and non serial, functions
in Va satisfy the geometric conditions
R' E v., R' = 0 on rf,

W i (Pi 1(P)) = Wi (pj1(p)),

(6.1) { (~i. Ui)(pi1(~)) = (1ii(Vi . Ui)(pj1(p)),
T' . (u' + V'W3)(pi1(p)) = 0,
Vp E Jk, Vi,j E I(Jk), k = 1, ... ,m.
It follows that functions in V00 satisfy the geometric conditions
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.
Furthermore, for REV00 one has
Note that Voo is independent of G and is dense in Hoo with continuous

injection, so that if Hoo is identified with its dual space we may write
Voo C Hoo c V:x"

where V:x, denotes the dual space of Voo. Let Aa denote the Riesz isomor-
phism of Va onto Va and Aoo denote the Riesz isomorphism of Voo onto
V:x,. If R E Voo , then AooR is simply the restriction of AaR to Voo , i.e.,
(AooR, R)voo = (AaR, R)va, 'v'R, R E Voo

The variational formulation of the system under consideration is given
by (4.25). This equation may be written in the equivalent form
(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)
We assume that the data of the problem satisfy
(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
THEOREM 6.1. Assume (6.6). There is a function R E C(Voo) with

It E C(Hoo) and ii E L2(V:x,) such that as G -+ 00,
RG -+ R strongly in LOO(H) and weakly in Loo(V);
nP -+It weakly in L (H);
00
It~ -+ It strongly in L (V:X'>;

00
ii~ -+ ii weakly in L2(V:x,).
Moreover, R is the unique solution of
(6.7) ii + AooR = Goo,

(6.8) R(O) = Ito, R(O) = R 1.
6.1.1. Interpretation of Theorem 6.1. Before proving Theorem 6.1, let

us give its interpretation. Equation (6.7) signifies that R is a solution of
the variational equation
We write
G = (Gl, ... ,Gn ), G i = (m:;lFi,Ip, C i ),
and R = (R\ ... ,Rn) with
Wi = I:W~a~, VW~ = I:W~,aa~.

a a
Then
(6.10) (R, R)voo = ~ In, ((Ci[C:(Wi )], c:(wi))

h2 .
+ 12 (Ci[c:(VW3)l,c:(VW~))}dO.
A
We have
(6.11) ( (Ci[C:(W i )), c:(wi )) dO

10.,
= ~ I:
a,/3
( C~/3[C:(Wi))(W~,/3 + wj,a)
10.,
= - I: 10.,( W~~C~/3[C:(Wi))
a,/3 OX/3
dO
+ I: 1r, W~vhc~/3[C:(Wi)) dr
a,/3
= - ( wi. divCi[c:(w i )) dO + { w' (Ci[C:(Wi))V i ) dr.

10., lr,
Therefore
(6.12) ((Ci[C:(VW~)], c:(VW~)) dO

10.,
= I:
a,/3
( W~ oX O~x
10., a /3
C~/3[c(VW~)) dO -1 (VW~) (Ci[c(VW~))vi)
r,
. dr
=(2Jli + )..i)h { W~~2W3 dO - (2Jli + )..i)h { {W~ O~i (~W~)

10., lr,
- (VW~) . (Cdc:(VW~))vi)} dr,
where a = Eo 8 2 / 8x~ is the harmonic operator and a 2 = aa the bihar-

monic operator in JR2 In addition, at least formally we have
(6.13) (R, R)voo = t1

t=1 n,
(mi Wi. Wi + I p, V'W3 . V'W~) dO
Furthermore, for every R E Voo we have

(6.14) (Goo, R)Hoo = (G, R)H = t1
i=1 n,
(F i . Wi - Ci . V'W~) dO
where div C i = Eo C~,o and where we have written

- Fi0 + p'ia
F i -- '"' i
L.J Fia
i i '.-
i
3 3'
i
It follows from (6.9)-(6.14) that (WI, ... ,wn) and (WI, ... ,W;) satisfy
the system of partial differential equations
m(Wi - divCi[c(wi)J = Fb,

(6.15) {
miW~ - Ip,aW~ + Dia2w~ = F~ + divCi,
in agreement with our formal calculation in Section 2.5, where Di = (2ft. +
Ai)h3 /12. Since R takes is values in Voo , the geometric boundary conditions
(6.2) and geometric junction conditions (6.3) are also satisfied by R. In
addition, for all R E Voo we must have
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:
where
C".(VW~) = Vi. (Ci[c(VW~)lvi),

CT' (VW~) = -ri . (Ci[c(VW~}lvi).
Then (6.16) may be written
The dynamic boundary conditions on rp follow immediately from (6.17);

we obtain
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
junctions conditions along Jk are therefore
(6.20) L
tEI(Jk)
{CI[C(WI)]V I + [Ip< ~:! - 8~1
Dl (LlW3)
- ~~ 8~1 CTVW3)] aa} = - L (VI. Cl)aa, k = 1, ... ,m,

lEI(Jk)
and
(6.21) L U t (Jk)CII VW3) = O.

tEI(Jk)
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
n = IAi x A21' A~ = a~ + W:a
PROPOSITION 6.1. The )unctzon condztzon
zs the lznea1"tzatzon of the nonlmear constmmt
(6.22)
Condition (6.22) obviously means that the angle between 'PI and 'PJ is
the same after deformation as before.
PROOF. The proof is by direct calculation. We have
Ai1 X Ai2 = a i3 + a1i x Wi,2 - a2 Wi + Wi,1 X Wi,2'

i X,1
so that (where .. represent higher order terms in the displacement)

IAi x A~I = 1 + ai . W\, + a~ . W i,2 + '" ,
IAi x A~IIA{ x A~I = 1 + ai . W:1 + a~. W:2

+ a{ . W~1 + ~ . W~2 + ... ,
(Ai x A~) . (A{ x A~) = a~ . a~ + (a~ x a{) . W~2

- (a~ x ~) . W~1 + (a~ x aD . W:2 - (a~ x a~) . W:1 + ....
If only linear terms are retained in (6.22) we obtain
(6.23) 0 = (a~ x a{) . W~2 - (a~ . a~)(a~ . W~2) - (a~ x ~) . W~1

- (a~ . a~)(a{ . W~1) + (a~ x aD . W:2 - (a~ . a~)(a~ . W:2)
- (a~ x a~) . W:1 - (a~ . a~)(ai . W:1).
In the cross products, express a~ in terms of the a{, a~, a~ basis and a~ in
terms of the aL a;, a~ basis. Then (6.23) simplifies to
o= (a~ . a~)(a~ . W!2) + (a~ . a{)(a~ . W!l)

+ (a~ . a;)(a~ . W:2) + (a~ . aD(a~ . W:1),
that is,
(6.24) a~. V'w~ +a~ V'wg = 0,
where V'w~ = La w~,aa~. If we write
a~ = (a~. a~)a~ + (a~. Vi)V i ,

then
. . ' . aW3i
(6.25) a J3 "W' a 3 v') -a
v 3-- (J ..
v
Similarly, write
a~ = (a~ . a~)a~ + (a~ . vi)vi

= (a~ . a~)a~ - O'ij(a~ . vi)vi ,
so that
.' ' . awg
(6.26) a 3 V'wg = -O'ii(a~ v')-a..
vJ
It follows from (6.24)-(6.26) that
and we note that a~ . vi i= 0 when Pi and Pj are non coplanar. 0

6.2. Proof of Theorem 6.1. We easily deduce from (6.4) the energy
identity
IlitG(t)ll~ + IIRG(t)II~G = IIRlll~oo + IlRoII~oo + lot (G(s), itG(s))Hds,

from which it follows that
(6.27) IlitG(t)llioo(H) + IIRG(t)llioo(vG)
~ C(IIRlll~oo + IIRoII~oo + IIGlli2(H)
for some constant C independent of G. It follows from (6.27) that as
G-+oo,
(6.28) itG is bounded in LOO(H).
In addition, by using (2.16) and the definition of the VG norm, we may
conclude from (6.27) that for a = 1,2 and i = 1, ... ,n,
(6.29) w!',~ + w~',~ is bounded in LOO(HO(Oi)),

(6.30) u!',~ + u~,~ is bounded in LOO(HO(Od),
(6.31) v'a:lui,G + V'W;,GI is bounded in LOO(HO(Oi))'
(Our notation is as follows:
RG = (R1,G, ... ,Rn,G), Ri,G = (Wi,G,Ui,G),
ui,G = "'Ui,Gai Wi,G = "'W~,Gai.)
L...i Q Q' L...i J J'
Q j
Since, by assumption, rf i= 0 for at least one index i, we may assume

that rf i= 0. It then follows from (6.29)-(6.31) and Korn's lemma for
two-dimensional linear elasticity that
(6.32) w1,G U1,G are bounded in LOO(Hl(O 1 )) .
Q , Q
Then, from (6.31) and (6.32) we may conclude that
(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),
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)).
Then, writing R = (Rl , ... ,Rn), Ri = (Wi, u i ), we have

U i -- "Wi3.
-v
Therefore W~ E L OO {H 2{Oi)), R E LOO(Voo ), it E LOO(Hoo). Since RG is

uniformly bounded and equicontinuous from [0, T] into H we also have
RG _ R in C(H).
In particular, R( 0) = Ro.
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') ,
so that t t--+ R( t) is strongly continuous into V:x, and

R(O) = limR~(O) = R I .
Upon passing to the limit in (6.39) we obtain
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.
6.3. Connected, Hinged or Rigid Joints. Here we briefly consider

in a formal manner the forms of the dynamic junction conditions at a
connected joint, a hinged joint and at a rigid joint. In all cases, the ap-
propriate balance laws along a joint Jk are obtained from the variational
junction condition (cf. (6.19))
in which the test functions Wi must satisfy geometric constraints appro-

priate to the particular type of joint under consideration. At a connected
joint, these constraints are
at a hinged joint they are (6.41) and
(6.42) ( a'Tai )2 W'(pi

,. 1(p)) = 0, Vi E I(Jk), Vp E Jk;
and at a rigid joint they are (6.41), (6.42) and

, . , j
aw~( Pi-1( p )) =
(6.43) -8 (Jij - a . PJ
8W3( -1( P )) , Vi,j E I(Jk), Vp E Jk.
v' vJ
We immediately deduce from (6.40) and (6.41) the dynamic junction
conditions at a connected joint. They are (6.20) and
At a hinged joint, the dynamic conditions are (6.44) and
for all functions Wik of the form (cf. Section 3.3.2)
Wik (pi;;l(p)) = e2 + sel, (p = q2 + S'T ik , ak ~ S ~ bk ),
where e2, el are fixed but arbitrary vectors in 1R3. We therefore obtain
from (6.45) the nonlocal balance laws
(6.46)
(6.47)
Finally, the dynamic junction conditions at a rigid joint are (6.46), (6.47)
and (6.21).
CHAPTER VIII

of Coupled Plate-Beam Systems
1. Introduction
In this chapter we shall develop a distributed parameter models for two

dynamic elastic plate-beam configurations. The first consists of a thin plate
to which a thin beam is rigidly and orthogonally attached to the edge of
the plate in such a manner that the centerline of the beam is coplanar with
the middle plane of the plate (in the equilibrium state). The second model
is that of a thin plate to which a thin beam is attached orthogonally to a
face of the plate (Le., the centerline of the beam is orthogonal to a face of
the plate in the equilibrium configuration). In the case of the configuration
in which the beam is orthogonal to the edge of the plate, it will be proved
that the dynamical system obtained is exactly controllable by means of
controls applied along an appropriate portion of the edge of the plate that
excludes the junction region between the plate and beam. We shall also
consider a simplified model of that system obtained by passing to the limit
in the original model as the shear moduli go to infinity.
Modeling of junctions between elastic elements of differing dimensions
has been carried out by Gruais [32] for 2d-ld junctions and by Ciarlet,
Le Dret and Nzengwa [17] for 3d-2d junctions using the Ciarlet-Dystunder
method of asymptotic expansions discussed earlier. This approach requires
that the "lower dimensional" body be embedded in some manner into the
higher dimensional one. For the configuration in which a thin beam is at-
tached orthogonally to the edge of a thin plate, in the limit problem (as
the "thickness" parameters approach zero) the junction region appears as
a slit in the middle surface of the plate. This geometry presents partic-
ular difficulties in the study of controllability of the limit system because
the poor regularity properties of solutions of the uncontrolled problem in
the neighborhood of the slit create problems for the multiplier methods
320 VIII. PLATE-BEAM SYSTEMS
commonly used to derive the necessary observability estimates, since such

techniques can be justified only if it is known a priori that the solutions are
sufficiently regular. (In this context see, however, the controllability results
of Nicaise [79], [80] which pertain to a 3-d region with an embedded plate
and to a plate with and embedded beam.) In contrast, in the framework
presented below the junction region appears as a part of the boundary of
the reference surface of the plate. Following the approach employed above
for linked membranes and for linked plates, we obtain geometric junction
conditions from the very natural requirements that the interface between
the (three dimensional) plate and the (three dimensional) beam be flat and
that the displacements of the plate match those of the beam along that in-
terface, combined with certain ad hoc assumptions on the kinematics of the
deformation. Then Hamilton's principle is employed to derive the dynamic
conditions of the model. In particular, the dynamic junction conditions will
turn out to be nonlocal, coupling averaged forces and moments induced by
plate deformation along the plate-beam interface to forces and moments
induced by beam deformation at that point on the centerline of the beam
where the beam meets the plate.
The configuration in which the beam is attached orthogonally to the
edge of the plate will be studied in Sections 2 - 5. The model equations
are derived in Section 2 and, in Section 3, it is shown that this system
comprises a well-posed problem (in appropriate function spaces). Exact
controllability of the model equations is the subject of Section 4. There
it will be established that the system is exactly controllable by means of
controls acting in either geometric or mechanical boundary conditions along
a certain portion of the edge of the plate that excludes the junction region.
We remark that the methods employed below could also be used without
essential new difficulties to treat more general configurations such as a plate
with several beams attached orthogonally along its edge, or even a plate to
which is attached a network of rigidly joined beams of the sort discussed in
Chapter III. Let us also note also that exact controllability of a plate-beam
distributed parameter model, but one of a completely different character
than that considered below, has been established by Puel and Zuazua [84].
The reader is also referred to the paper [85], where exact controllability of
a model of an elastic string connected to an elastic membrane (or, more
generally, to an n-dimensional body modeled as an n-dimensional wave
equation) is proved.
The geometric junction conditions which we derive in the Section 2 in-
volve nonlinear couplings among the displacement and rotation vectors of
the plate and beam. Upon linearizing these relations one finds that quan-
tities related to motion in the reference plane of the plate are no longer
coupled to the remaining quantities, which are: the transverse displace-
ment and two transverse shear angles associated with deformation of the
plate; and the transverse displacement, torsional angle and one of the shear
2. MODELING OF THE PLATE-BEAM JUNCTION: I 321
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.
2. Modeling of the Plate-Beam Junction: I

2.1. Geometric Conditions. Let n be a bounded, open, connected
set in 1R? with Lipschitz continuous boundary r consisting of a finite num-
ber of smooth simple curves. We consider n to be the reference surface of
a thin plate of uniform thickness h. Points within the plate will be denoted
by coordinates (Xl,X2,X3) with respect to the natural basis a1.a2,a3 for
IR3. Thus the equilibrium position of the plate is given by
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.
Let Xo be a point of JR3 and el, e2, e3 be a right-handed orthonormal

basis for JR3. Let w be a bounded, simply connected open set in JR2 such
that 0 E w and which is doubly symmetric with respect to the origin. The
beam, in its reference configuration, is given by
The following assumptions are made:

(AI) Xo E r and v(Xo) exists.
(A2) el = v(xo), e2 = T(Xo), e3 = a3.
(A3)
(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,
where IA31 = 1 and a3' A3 > OJ and

(2.2) r(6,6,6, t) = Xo + 6e1 + w(6, t) + 6E2(6, t) + 6E3(6, t),
6~ ~ l, (~2' ~3) E w,
where
E 2(6, t) = 0(6, t)e2, E 3(6, t) = 0(6, t)e3
for some orthogonal matrix 0(6, t) with determinant one.
Assumption (2.2) means that cross-sections of the beam move rigidly,
while (2.1) is the hypothesis of Reissner-Mindlin plate theory that the X3
line remains straight and the extensional strain C33 vanishes.
The orthogonal matrix 0 may be considered to be the product of three
rotation matrices, the first representing rotation through an angle -0 1 around
the 6 line, the second rotation through an angle -0 2 around the 6 line and
the third rotation through an angle -03 around the 6 line. It may therefore
be represented with respect to the e, basis as the product
where, in this representation, positive values of -0. correspond to clockwise

rotation around the ~. line. The angle -0 1 is a torsional rotation about the
e1 axis and -02 , -03 correspond to flexure of the beam in the e1e3 and e1e2
planes, respectively. It follows from (2.3) that (cf. (1.9), Chapter III)
E2 = (sin -01sin {}2 cos {}3 - cos {}1 sin {}3 )e1
+ (cos {}1 cos {}3 + sin {}1 sin {}2 sin {}3 )e2 + (sin {}1 cos {}3 )e3,
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'
where T = T(Xo), that is
W(Xo + 6T) = w(o) + ~2(E2(0, t) - e2),

(2.5) {
U(xo + 6T, t) = (E3(0, t) - e3), 161 < ~,
where U = A3 - a3 is the "rotation" of the a3 line. Conditions (2.5) are of
rigid type and simply require that the displacements of the plate and beam
match throughout the junction region.
Write
W = Wi8.i, U = Ui8.i, U = U{3a{3, W = Wiei,
where U3 = -1 + (1-luI2)1/2. Then (2.5) may be written
W3(XO + 6T, t) = W3(0, t) + 6 (sin '19 1 cos '19 2)(0, t),
U3(xo + 6T, t) = (cos '19 1 cos '19 2)(0, t) - 1,
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.
2.2. Dynamic conditions. The dynamic conditions will be obtained

from Hamilton's Principle:
(2.8) 8foT[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 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
purposes we also assume that I'D n J = 0, which implies in particular that

r C i- 0 and I'C n J i- 0. The part r C corresponds to geometric boundary
conditions
(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
W(X) = ao(x)[w(O) + ((X){)l(O)],

(2.10) {
u(x) = al (x)iJ(O), x E J.
Conditions (2.10) require that the displacement Wand rotation U along J

change smoothly from their values on Ja to zero, and give prescribed profiles
to these quantities along J. The quantities ai and ( may be considered
design parameters. We refer to Puel-Zuazua [85], where a similar boundary
condition is introduced in a string-membrane problem in a neighborhood

of the point where the string meets the membrane.
To summarize, the geometric conditions which the admissible displace-
ments must obey are (2.6) and (2.9) when J\ Ja is free, or (2.9) and (2.1O)
otherwise.
We write
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,
(2.12) Up{t) =~ k [(G[c{u)J,c{u)) + G1hlVW + ul 2 JdO,
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
where I = 8/86. We assume that the beam is homogeneous and elastically

isotropic, so that
sij = 2JL2eij + )..2 Dij e kk,
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
where A = ah is the cross-sectional area of the beam, so that the part of

the potential energy containing the quantities W3, t?I, t?2 may be written
(2.13) UB(t) = ~ 101 (G211t?~12 + Ehlt?~12 + G2Alw' + t?212)) d{,

where W = W3 and
h2 a2
h := A 12' la:= A 12 1:= h + la
The kinetic energy of the beam is
~ 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
on the plate and fp(Xl,X2,X3) be an edge force distributed along rD x

(-h/2, h/2)o The work done on the plate by these forces is approximately
j h/2 {Fp (W + X3 U ) dndx3 + jh/2 { fp (W + X3 U ) dr dX 3

-h/2 in
0 0 0
-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
F1 (Xl,X2) = a3 Fp(Xt,X2), !t(Xl,X2) = a3 f p (Xl,X2),

0 0
Ml = (al Mp)al + (a2 M p )a2, ml = (al mp)al + (a2 mp)a2o

0 0 0 0
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
10 1 FB 0 (w(e) + 6(191(e)e3 - 193 (e)e1) + 6(192(e)e1 - 191(e)e2)) dwde
+1 fB 0 (w(l) + 6(191(l)e3 - 193 (l)e1) + 6(19 2(l)e1 - 19 1(l)e2))dwo

Consider the resultant forces
and set
M 23 (e) = 1 (6 e 2 - 6 e3) FB dw,

0 M 13 (e) = 16e1 FB dw,0
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
(2.16) WB(t) = lot [F 2w + M2 . ('I92el - 'I9 l e2)] d{

+ hw(f) + m2 . ('I92el - 'I9 l e2)(f),
where
F2 = e3 F B , h = e3 fB,
M2 = Ml3e l - M 23 e 2, m2 = ml3el - m23e2
We are now in the position to calculate (2.8), where the variation is

taken with respect to displacements satisfying (2.6) and (2.9) if J - Ja is
free, or (2.9) and (2.10) when J - Ja is geometrically restricted. We first
discuss the latter situation, since this is the one which will be considered
in subsequent sections. In the calculation we utilize the Green's formula
(3.7), Chapter VI, in the form
(2.17) k(C[c(u)],c(v))dn= - k v(divC[c(u)])dn
+ l V (C[c(u)]v) dr.
The following dynamic equations of motion and boundary conditions are

obtained.
Equations of motion of the plate:
ml IV - G1hdiv(V'W + u) = F 1,
(2.18) {
Ipl ii - divC[c(u)] + G1h(V'W + u) = MI.
Equations of motion of the beam:
m2W - G 2 A(w' + '192 )' = F2 ,

{
(2.19) P2I~~2 - Eh'l9~ + G 2 A(w' + '19 2 ) = M 13 ,
P2 I'!9 1 - G2I'I9~ = -M23 .
Boundary conditions along rD:
G1h(V'W + u) v = iI,
(2.20) {
C[c(u)]v = mI.
Boundary conditions at ~ = f:
G2 A (W' + 192 ) (f) = 12,

(2.21) { Eh19~(f) = m13,
G2119~ (f) = -m23.

In addition, we obtain the variational junction condition
0= i [11. (C[c(u)]v) + GlhW(V'W + u). v] dr
- G2N~ (O)JI(O) - Eh19~(0)t?2(0) - G2A(w' + 192)(0)w(0)
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:
G2N~ (0) = i {Glh(aov. (V'W + u)
-ale2 (C[c(u)]v)}dr,
(2.22)
Eh19~(O) = i aIel (C[c(u)]v) dr,
G2A(w' + 192 )(0) = Glh I aov (V'W + u) dr
in which it should be recalled that el = v(Xo), e2 = r(xo). The complete

description of the system when J \ J a is geometrically restricted therefore
consists of (2.18)-(2.22) together with the geometric constraints (2.9) and
(2.10).
In the case when J \ Ja is free rather than geometrically restricted, we
again obtain (2.18)-(2.22) except that the integrals are to be taken over Ja
rather than all of J, and we obtain on J \ Ja the free edge conditions
(2.23) Glh(V'W + u) . v = 0, C[c(u)]v = 0 on J \ Ja .

If desired, one may in the same manner obtain the equations of mo-
tion and boundary and junction conditions for the components w(3, W(3, 193
related to the motion of the plate-beam system in the ala2 plane.
3. Function Spaces and Well-Posedness

We shall consider below the question of exact controllability of a plate-
beam system by means of controls applied along rD only, which may act
through either mechanical boundary conditions or geometrical boundary
conditions. Only the subsystem for the components W, u, w, 191, 192 will
3. FUNCTION SPACES AND WELL-POSEDNESS 331
be considered; however, a similar analysis could be carried out on the

subsystem satisfied by W,a, w,a, iJ 3 . In preparation for this study, well-
posedness of the controlled system will be considered in this section.
We suppose that all distributed forces and moments vanish, as do those
at the free end of the beam. The system to be considered may then be
written
+ 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 ;
(3.5) W = O'o[w(O) + (iJI(O)], u = 0'1'0(0) on J;
G2IiJ~(0) = L {Glh(O'ov (VW + u)
(C[(u)]v)}dr,
i
-0'Ie2'
(3.6)
EhO~(O) = O'lel' (C[(u)]v) dr,
G2A(w' + O2 )(0) = Glh L O'oV' (VW + u) 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.
To complete the description of the system, initial conditions are specified:
Wlt=o = Wo, Wlt=o = WI, ult=o = uo, tilt=o = uI,

(3.9) {
wlt=o = wo, wlt=o = wI, 'l?ilt=o = 'I??, Jilt=o = 'l?t
For s 2: we set
1-lB(O) := HB(O; tal]) EBHB(O; [a2]) EBHB(O; [a3]),
1-l B(O, f) = HB(O, f; led) EB HB(O, f; [e2]) EB HB(O, f; [e3]).

Let
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
11'l' II1t w1) = {fo [(C[(u)], (u)) + GlhlVW + u1 2] dO} 1/2
According to Korn's Lemma, 11111t~(n) defines a norm on 1-l~(O) equivalent

to that induced by 1-ll(O). We further introduce
V"( = {('l', 8)1 'l' E 1-l~(O), 8 E 1-ll(O, f) and 'l' = 0"8(0) on J},
where 0" is a 3 by 3 matrix which with respect to the el, e2, e3 basis is given
by
o
0"1
0) .

-O"o( 0"0
3. FUNCTION SPACES AND WELL-POSED NESS 333
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
DA :::> D~ := {("iJI, 8)1 ("iJI, 8) E (rt2(O) x rt2(O,)) n V"('

("iJI, 8) satisfy the dynamic junction and boundary conditions}
and that, if ("iJI, 8) E D~ and (,j,,8) E H,
(A("iJI, 8), (,j" 8))H = -In {[div C[c(u)]- Glh(V'W + u)] u
+ Glhdiv(V'W + u)W}dn -1 {G2IiJ~JI

- [EhiJ~ - G 2A(w' + ~2)]J2 + G 2A(w' + ~2)'w}dn.
It follows that the injection of D~ (with the topology of 1{2 (n) x 1{2 (0, ))
into DAis continuous. On the other hand, we know from results of
Nicaise [78] that under the stated assumptions on the region n, an ele-
ment ("iJI, 8) E D A has the following spatial regularity:
(3.10) "iJI E 1C(O), 'rIs < So, s:::; 2, for some So > 3/2,
r
in the case of geometrical conditions on D , and also in the case of mechani-
cal conditions on rD provided rP
and I'D either have an empty intersection
or else intersect in a strictly convex corner. If 1'0 and I'D meet in a non-
convex corner, (3.10) holds for all S < So, S < 3/2, for some So > 5/4.
Since dim(O) = 2, it follows from a Sobolev imbedding theorem that
"iJI E C" (0; tal]) EB C" (0; [a2]) EB C" (0; [a3])
all a E (0, s - 1] [30, Theorem 1.4.5.2]. With regard to e, we have 9 E

1i2(O, f). The injection of DA into 1iS(O) x 1i2(O, f) is continuous. 0
We denote by (,i,[f0, 9), (\111, e 1) the initial data (3.9), i.e.,
= u O+ WOa3, \111 = u 1 + W1a3,

\110
eO = {J0 + wOe3, 9 1 = {J1 + w1e3'
We also set
and
THEOREM 3.1. (Well-posedness of geometrical control problem.) Let

'Y = r - J and suppose that
(\110, eO) E H, (\II 1,e 1) E V~, f E L2(O,T;U).
Then (3.1)--(3.6), (3.8), (3.9) has a unique solution
(\II, e) E C([O, T]; H) n C 1 ([0, T]; V~).

THEOREM 3.2. (Well-posedness of mechanical control problem.) As-
sume that r a =I- 0 and set 'Y = ra. Let
(\110, eO) E H, (\111,9 1 ) E V~, f E L2(O,T;U).
Then (3.1)--(3.7), (3.9) has a unique solution
(\II, e) E C([O, T]; H) n C 1 ([0, T]; V~).
REMARK 3.2. The assumption in Theorem 3.2 that r a =I- 0 is unessen-
tial in Theorem 3.2, but if r a = 0 we need to replace the norm on V by
In
(11(\11, e)ll~ + 1\II1 2dO)1/2. 0
PROOF OF THEOREM 3.2. The system (3.1)-(3.7), may be written as a
variational equation by forming
(3.11) l ([m1 W - G1hdiv(VW + u)]W

=
+ [IPI ii - divC[(u)] + G 1h(VW + u)] u}dO

+ 1 {(m2 W - G2A(w' + '!92)')W + (p2 I i9 1 - G2I'!9nJ 1
+ (p2hi92 - Eh'!9~ + G2A(w' + '!92))J2}~'

Set
and suppose that Cit,S) E V')' := V. Upon carrying out integrations by

parts in (3.11) one obtains with the aid of (2.17)
(3.12) ((~, 8), (~, 9))H + (('l1, 8), (~, S))v = f [m u + jW] dr,
Jr D
so there is an operator B E (U, V') such that
(3.13) f [m u+ jW] dr = (Bf, (~, 8))v.

Jr D
Therefore (3.12) may be written
(~,e)+A('l1,e)=Bf in V'
or, alternately, as the first order system
(3.14) X=AX+Bf,
O:::D,
where
X= A= (_OA ~), Bf= (~f)'

It is well known that A, as an unbounded operator in H x V' (resp., in
V x H, resp., in V' x D~) with domain V x H (resp, with domain DA x V,
resp., with domain H x V'), generates a Co-unitary group on H x V' (resp.,
on V x H, resp., on V' x D~). The conclusion of Theorem 3.2 follows
immediately, since Bf E L2(0, Tj H x V'). 0
PROOF OF THEOREM 3.1. To obtain a variational equation for the
geometrical control problem (3.1)-(3.6), (3.8) we again begin with (3.11)
where we now assume that (~, 8) E DA. Since (~, 9) = 0 on 'Y = r \]
we have
k {-G1hdiv(VW + u)W - [div C[(u)]- G1h(VW + u)] u} dO
k
= {-G1hdiv(VW + u)W - [divC[(u)]- G1h(VW + u)]. u}dO
+ I {-G1hv . (VW + u)W + G1hv . (VW + u)W
- u (C[c(u)]v) + u (C[(u)]v)) dr
+ { [m (C[e(u)]v) + Glh/aaW]dr.
JrD v
Also
1 {-G2 A(W' + 192)')w - G2119~Jl - [Eh19~ - G2A(w' + 192)lJ2}~
= 1 {-G 2A(w' + J 2)')W - G2IJ~191 - [EhJ~ - G2A(w' + J2)1192}~
+ 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) - EhJ~(O)192(O).
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
0= 1 {-G1hv . (VW + u)W + G1hv . (VW + u)W
- 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)
(3.15) (Cq" 8), (+, 8))H + \].i, 8), A(+, 8))H
= [ [m (C[c:(u)lv) + G1h/aaWl df.

JrD v
Write (~, 8) = A- 1 ( . , 9) where (.,9) E H. Then (3.15) takes the form
(3.16) q" 8), A- 1 ( . , e))H + \].i, 8), (., 8))H
= [ [m.(C[C:(U)lV)+G1h/aaWldf.
JrD v
One has, for some s > 3/2 (see Remark 3.1),
irD
[m (C[C:(u)lv) + G1h/-
aw
av 1df ::; Cllfliuli\].iliw(n)
A
::; Cllfliull(+, 8)IIDA

= Cllflluli(., 9)IiH'
Thus
(3.17) [[m,(c[C:(U)lV)+G1h/aawldf
JrD v
= (Bf, (., 8))H, (~, 8) = A- 1 (., 8),
for some B E C(U, H), so that (3.16) may be written
(((t, 8), A-1(~, S))H + (("IJ1, 8), (~, 8))H = (Bf, (~, 8))H,
'v'(~, S) E H.
This is the same as the equation
(3.18) ((t, 8) + A("IJ1, 8) = ABf in D~,
where A is an isomorphism of H onto D~. Equation (3.18) may be written

as (3.14) with
Bf = (A~f)' BE C(U, V~ x D~).

The operator A, as an unbounded operator in V~ x D~ with domain H x V~,
generates a Co-group of unitary operators. Therefore, for initial data
('11,8) E V~, ('111,8 1 ) E D~,

the initial value problem for (3.18) has a unique solution with
('11,8) E C([O,T], V~) x C1([0,T],D~).
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
IPl ~ - div C[( c,o)] + G 1 h(V Z + c,o) = 0,

(3.19) {
m1 Z - G 1 hdiv(VZ + c,o) = OJ
in
m2Z - G2A(ZI + ( 2 )' = 0,

(3.20) {
P2I~.ij2 - Eh9~ + G2A(ZI + ( 2 ) = 0,
P2I91 - G2I9? = 0, < { < ij
(3.21)
(3.22) Z = 0, c,o = on 'Y := r - Jj
(3.23) z = 0'0 [z(O) + (h(O)], c,o = 0'18(0) on J,

G2IO~(0) = [{Glh(CTOV. (\1Z + cp)

-CTle2 (C[c(cp)JV)}dr,
(3.24)
ElhO;(O) = [CTle l . (C[c(cp)Jv) dr,
G2A(z' + ( 2)(0) = Glh [ CTOV (\1Z + cp) dr on Jj
Zlt=o = ZO, Zlt=o = Zl, cplt=o = cpo, cplt=o = cpl,

(3.25) { o . 1
Zlt=o = zO, ilt=o = Zl, 81t=0 = 8 ,8It=0 = 8 .
As in the last section, we define
~ = 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
(~o,SO) E V,),, (~l,Sl) E H,
and set ~l := r 1 x (0, T), where r 1 is any relatively open subset of r such
that r\ n] = 0. Then
(3.26) L[IClo( ",)1 vi' + G , h ( ~) '] dE

::; Co(T + 1)II((~o, SO), (~l, Sl ))II~'YXH.
The proof of Lemma 3.1 is deferred until the end of this section.
To complete the proof of Theorem 3.1, we utilize the idea of transposi-
tion. Write V := V')'" The solution of (3.18) having initial data
is given by
X(t) = exp(tA)XO + lot expeCt - s)A)Bf(s) ds, 0::; t ::; T,

where exp(tA) is the unitary group on V' x DA generated by A. Suppose
that XO E H X V', let yO E V X DA and B' E.c(V x DA;U) be the dual
of B, defined by
(Bf,yO)VXDA = (f,B'YO)u, "If E U, yO E V X DA.
Let r E (0, T] be fixed. We have
(3.27) (X(r), yO}VXDA = (XO,exp(rA')yO}HXV
+ lT (f(s),8' exp((r - s)A')yO)uds.
Here A' is the dual of A, defined by
(AXO, yO}VXDA = (XO,A'yO}HxV, \/XO E H X V', yO E V X DA.
One has
A' = (~ -oA), D(A') = V x DA.
As is well-known, A' generates a unitary group exp(tA') on H x V and

exp(tA') is the dual of the restriction of exp(tA) to H x V'. Therefore
(3.27) is the same as
(3.28) (X(r), yO}VXDA = (XO, Y(r)}Hxv + lT (f(s), 8'Y(suds,
where Y(t) := exp((r - t)A')Yo, 0 ~ t ~ r, satisfies
(3.29) .,
Y=-AY, O~t~r, Y(r)=Y.
Hwe write
Y = ((.... 0;::;0) (....

"rl, ....1, 0;::;0
"r,.... , yO = (( .... 1,0;::;0
"r .... 1), (....0, . . . . . 0,
"r
(3.29) signifies that
(i,S) + A(.,8) = 0, (.1181) = -(ci;,8),

(3.30) {
(.(r),8(r = (.,8), (ci;(r),8(r = _(.1,8 1 ).
In addition,
From (3.17) we have
(Bf,A(., 8H = 1 rD a 1dr,
[m (C[c(<p)]V) + Kf-
az
v
hence
(3.31)
We insert this expression into (3.28) and utilize Lemma 3.1 with r l = rD
to obtain the estimate
(3.32) I(X(r),yO)vXDAI:::; IIXoIIHxV,lIyOIIHXV

az
+ Il f ll2(O,T;U)IIC[c(cp)]v + Glh av a31IL2(O,T;U)
:::; Co(T + l){IIXOIIHxv' + 11f112(O,T;U)}ll y OIIHXV, 0:::; r :::; T.
It follows that X E Loo(O, Tj H x V'). One may pass from LOO to C by a
standard approximation argument. 0
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
REMARK 3.4. When XO = 0, it follows from (3.32) that
(3.33) IIX(T)IIHxV' :::; CT ll f ll2(O,T;U)

so that the control-to-state map f I--t X(T) : L2(0, Tj U) I--t H x V'is
continuous.
PROOF OF LEMMA 3.1. It suffices to prove (3.26) for initial data in

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)]
= h3 (a~l ((Al + 2J.tdc ll(XCP) + Alc22(XCP)) + 2J.tl a~2 Cl2 (XCP))

12 a a ((Al + 2J.tdc22(XCP) + AlCll(XCP))
-a
aXl
2J.tl- Cl2(XCP) +
X2
= xdivC[c(cp)] + 2C[c(cp)]Vx + ~~ (J.tl - Al) ('P-'Pl,2

2 ,2 -'P2,l) Vx
'Pl,l
h (X,ll + J.tlX,22
3 (Al + J.tl)x,l2)
+- cp.
12 (Al + J.tl)x,l2 X,22 + J.tlX,ll
It follows that ci satisfies the equations
mlZ - Glhdiv(V Z + ip) = P,
(3.34) {
I p1 iP - divC[c(ip)] + Glh(VZ + ip) = M,
where
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
II(i(t), ~(tII~ ~ lI(i(O), ~(OII~ + c lot IIF(t)lI~o(n)dt,

where F = 1\1: + Fa3' We shall show that
(3.36) L[1C[e(~)IVI2 +G,h (~:)'l dE

s c [{ IIF(t)IIJ."(o)dt + (T + 1)11(+, +)II1.-(o.T'1<)]'
Since
II(i, ~)lIioo(O,T;'H) ~ II(i(O), ~(OII~ + c loT IIF(t)ll~o(n)dt

SC ("(.0,., )1IJ.\(0)x1<"(O) + /,T IIF(t)IIJ."(o)dt) ,
and
10[T IIF(t)ll~o(n)dt ~ c 10[T 1I~(t)lI~l(n)dt

~
~ C loT 11((~(t),8(t, (4;(t),8(t)))II~~XHdt

= CTII((~0,80), (~l, 8111~ xH'
~
and since i = ~ on E I , Lemma 5.1 follows from (3.36).

The idea leading to (3.36) is standard (cf. [60], for example). One
multiplies the two equations in (3.34) by h vi and (Vip)h respectively,
where h is a W 1 ,oo vector field in 0 such that h = II on r, adds the products

and integrates the sum over 0 x (0, T). One thereby obtains
(3.37) loT In {[mi Z- G1hdiv(VZ + ~)lh VZ

+ [IPl iP - divC[c(~)l + G 1h(VZ + ~)l' V~)h)} dOdt
= loT In[(h' VZ)F + V~)h)Ml dOdt.
One has
loT In (h VZ)div(VZ + ~)dOdt

= loT In {div[(h VZ)VZ1- (VhVZ). VZ
- ~ div(hIVZ/)2 + ~(divh)IVZI2 + (div~)(h. VZ)}dOdt

= fT f {!(divh)IVZI2 _ (VhVZ). VZ
10 in 2
+ (div <i (h . 'i1 Z)}dOdt + H(:)' dE,
iT In iP, (V~h)dOdt = In cP' (V~h)dOI5 + ~ iT In (divh)lcPI 2dOdt.
Also, from Lemma 3.2, Chapter VI, we have
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
Q(Vcp) = ~(hl,l - h2,2)[(2JLl + Ad(<ptl - <P~,2) + JLl(<p~,l - <p~,2)1

+ (2JLl + Ad(h1,2<P2,1<P2,2 + h2,1<Pl,1<Pl,2)
+ JLl(h1,2<Pl,1 + h2,1<P2,2)(<Pl,2 + <P2,d
+ Al (h1,2<Pl,1 <P2,1 + h2,1 <Pl,2<P2,2).
When the last four formulas are inserted into (3.37) the result is
in {mIZ(h. VZ) + Ip1tP (Vrph)}dO[
+ ~ loT in (divh){mIIZI2 + IplltPl2}dOdt
+ loT in {Q(Vrp) + GIh[-~(divh)IVZI2 + (VhVZ) . VZ
-1. {G~h (~: r

- (divrp)(h. VZ) + (VZ + rp). ((Vrp)h)]}dOdt
+ (V~h) (C[e(~)Jv) - ~(Ck(~)J, e(~))}.n:

= loT in [F(h . VZ) + M (Vrph)] dOdt.
On E on has V rph = arp / av, hence
(3.38) (Vrph) (C[e(rp)]v) - ~(C[e(rp)],e(rp)) = ~(C[e(rp)]v). ~~

(see (5.20), Chapter VII). In addition (cf. (3.34), (3.35), Chapter VI)
3
(3.39) C[e(rp)]v = Ca /3[e(rp)]va a/3 = h12 { J.t1 T' (a-a~ + v fr)
a~ T
+ [(2J.t1 + AI) (v. ~~) + Al (T' ~!) ]v},
(3.40) C[e(rp)]T = Ca .B[e(rp)]Ta a/3
= ~; { [(2J.t1 + At) (T . ~!) + Al (v .~~) ]T

+ J.tl a-av + v . -.!
( -.!
T' a-)}
aT
v .
Therefore, on r,
(3.41)
a~ 2+ (2J.t1 + AI) (V aa~- ) 2]

a- 3 [
(C[e(rp)]v) a~ = h12 (
J.t1 T'
a- )
,
and
(3.42)
3
IC[e(rp)]vI2 = ( h12 )2 [ J.t~ ( T' a-)2
a~ + (2J.tl + At}2 ( V a-
a~ )2] .
Hence, on r,
for some positive constant K. It follows from (3.38) and (3.43) that
(3.44) (Vi,Oh) (C[e(i,O)]v) - ~(C[e(i,O)],e(i,O)) ~ 2~IC[e(i,O)]vI2.

We also have
loT In {F(h. VZ) + iII (Vi,Oh)}dOdt

~ C JofT (IIFII~o(o) + 1I~1I~1r (0)) dt,
loT l {Q(Vi,O) + Glh[-~(divh)IVZI2 + (VhVZ). VZ

- (divi,O)(h VZ) + (VZ + i,O) . ((Vi,O)h)]}dOdt I
~ C JofT 1I~1I~1r (0)) dt.
It follows from the above estimates that
L[ICI I<')Ivl' + G1h ( : : r] dL
:<; 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. The Reachable Set

We consider the mechanical control problem (3.1)-(3.7), or the geo-
metrical control problem (3.1)-(3.6), (3.8), with vanishing initial data:
'11 0 = '111 = 0 and 8 0 = 8 1 = O. The reachable set at time T > 0
is
RT = ((('II(T), 8(T)) , (,j,(T), 8(T))) If E L2(0, Tj Un.
In order to say something useful about RT we shall need to impose geomet-
ric restrictions on r. We therefore assume that there is a point Xo E lR.2
such that
(4.1) (x - xo) . v ~ 0 for x E r C , (x - xo) . v < 0 for x E J.
Set
r~ = {x E rDI (x - xo) . v > O}, r~ = {x E rDI (x - xo)' v ~ O}.
When the controls act in the geometrical boundary conditions, one may
assume without loss of generality that r~ = rD or, to say the same thing,
that the controls f are supported in r~, which amounts to redefining r C to
be r C u r~. However, when the controls act in the mechanical boundary
conditions the explicit assumption
(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),
U+ = L2(r~j [a1]) E9 L2(r~j [a2]) E9L2(r~j [a2]),

1i-1(r~) = H-1(r~j [a1]) E9 H-1(r~j [a2]) E9H-1(r~j [a3])' 0
In the next subsection the observability estimates needed to prove The-
orems 4.1 and 4.2 are presented.
4.1. Observabilityestimates. Let us write r = 8n as r = four\ U"],

where r o,r l , J are relatively open in r, mutually disjoint, ro - 0 and
r l n"] = 0. We consider the problem
+ Glh(VZ + <p) =
(4.4) { Ipl~ - divC[(<p)] 0,
mlZ - Glhdiv(VZ + <p) = 0 in nj
m2Z - G 2A(z' + ( 2)' = 0,

{
(4.5) P2I~.jj2 - EhO~ + G 2A(z' + ( 2) = 0,
P2IOI - G 2IOr = 0, 0 < ~ < j
(4.6)
(4.7) Z = 0, <p = 0 on roj

(4.8) C[(<P)]v = 0, Glh(VZ + <p) . v = 0 on rlj
(4.9) Z = O'o[z(O) + (OI(O)], <p = 0'18(0) on J,
G2IO~(0) = [{Glh(O'ov' (VZ + <p)

-O'l~' (C[(<p)]v)}dI',
(4.10)
EhO~(O) = [O'lel ' (C[(<p)]v)dI',
G2A(z' + ( 2)(0) = Glh [ O'OV . (V Z + <p) dI'j
Zlt=o = ZO, Zlt=o = Zl, <plt=o = <po, <plt=o = <pI,

(4.11) {
zlt=o = zO, ilt=o = ZI, 81t=0 = 8 0, 0lt=o = 8 1 .
As in the last section, we write
~ = <p + Zk, 8 = 8 + ze3

with corresponding initial values ~o, ~1 ,80 ,8 1 at t = O.
Let Xc E JR2 and define
rt = {x Era I(X - Xc) . V > O},

r~ = {x E r al (x - Xc) . v:::; O}, a = 0, 1.
These sets of course depend on Xo. The relation between r 0, r 1

and r C,r D
is the following: when considering the control problem with controls in the
mechanical boundary conditions, we set 1 = r rD
and = r rC.
The
assumptions of Theorem 4.2 then require, in particular, that rt r
= 1 = 0.
When considering the control problem with controls in the geometrical
boundary conditions, we set r1 r r-
= 0, o = J, = and = r C ro rD rt.
As usual, we write m(x) = x - Xo and E~ = r~ x (0, T). The main
observability estimate is as follows.
PROPOSITION 4.1. Let Xo be such that m v < 0 on J. Assume that 1'0

and 1'1 either do not intersect or else intersect in a strictly convex corner
and that r o 1= 0. Let
(.,8) E DA, (.1,8 1 ) E Vro.
There is a constant To such that if T > To, the solution of (4.4)-(4.11)

satisfies
(4.12) 11((+,5), (+"5'))II~.xH ~ CT [ht 1+I'.n:

+ h- 1
(C[(IP)]T). ~~ + G1hIT (V'Z + IP)1 2 ) dE
+ 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
and therefore from (3.40)
The following corollary is an immediate consequence of Proposition 4.1

-
and Lemma 3.1 (note that r o n J = 0) .
~
COROLLARY 4.1. Let Xo be such that m v < 0 on 1. Assume that

r 1 = 0 and that
Then there is a constant To such that ifT > To, the solution of (4.4)-(4.11)
satisfies
(4.13) 1I((~o,EO), (~l,El))ll~roxH
$ CT ht (C[<P)]V). : + G,h 1:1') dE.
The corollary which follows is a consequence of Proposition 4.1 and the

process of "weakening the norm."
COROLLARY 4.2. Let xo be such that m . v < 0 on J. Assume that

rt = r 1 = 0,
that ra =I 0 and let
(~o,Eo) E Vro' (~l,El) E H.
r r
and that 0 and 1 either do not intersect or else intersect in a strictly
convex corner. Then
(4.14)
Corollary 4.1 pertains to the problem of control in the geometric bound-

ary conditions wherein r D = rt, r C = r a, while Corollary 4.2 pertains
to the problem of control in the mechanical boundary conditions, wherein
rD = rt = rl and r C= ro = roo
PROOF OF PROPOSITION 4.1. As our starting point we apply the iden-
tity (5.17), Chapter VII, to the solution ~ of (4.4). We thus have
(4.15) R(T) - R(O) +h {(I - 8)ml1Z12 + 8Ip1 1r,e12

+ (1- 8)(C[c(cp)],c(cP)) + 8G1h(IVZI2 -lcpI2)} dQ
= ~h (m. v){mllZl 2 + I p1 1r,e12} d'E
+! [(m. v) {(C[c(cp)]v). 8cp _ (C[c(cp)]T)' 8cp

2h ~ ~
where 8 > 0 and
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
As in the proof of Lemma 5.3, Chapter VII, we have
k {(1- 8)ml1Z12 + 8IP1 1cp12 + (1- 8)(C[c(cp)],c(cp))
+ 8G1h(IV Zl2 -lcpI2)} dQ 2:: Co loT (11ci>II~o(o) + 11()11~~o (0) dt

for some constant Co > O. Therefore, for some constant C,
(4.16)
R(t) lif + { (11+11;"'(0) + II.II~" (n) ) dt :5 G [1", (m . v)1+1' dE

- hJm'lI) ((C[C(rp)]T)' :~ +GlhIT.(vz+rp)12)d~
1
+ h,(m'v) (( G1e('P)lv): +G'hl~1') dE] +.1(E J)
since (C[c(rp)]T)' arp/aT ~ 0 on ~l (see Remark 4.2 above).
We proceed to estimate .J(~J) and are going to show that
(4.17) .J(~J):::; ~ f (m 'lI){mlIZI 2 + Ipllcpl2} ~

lEJ
+ G, 1", (m (I: I' + 1:: + 'PI') dE +
v) v. G, { 13(0)1' dt,
where C1. C2 are positive constants independent of T. To this end, from

(3.39) and (3.40) we calculate
Therefore, since m . v < 0 on J,
for some constants C1 ,C2

For any > 0 and for some constant Ce we have
(4.19)
Since m II < 0 on J, we also have
(4.20) IhJ (~~ + ~ [(m' r) ~! + 6Z] dEl

II' )
~'l, 1m- VII: + _1'1' dE+ c, l, [I~I' +IZI'] dE,

V
Further, from (3.39),
(4.21)
ICIeCI')lvl' ~ 2G:)' [C2~1 Ul)'IZI' +maxC~l,~l) I~J']

and therefore
(4.22) 1~}C[c(CP)lV). [(m.r)~~ + (1-8)CP] dEl

~ f h, Imv11:I' ,/laC, L (I~I' + I~I') dE.
It follows from (4.18)-(4.22) that for any > 0,
(4.23) 3(E J )::; ~

JEJ
r (m v){mllZl 2 + Ipl Icp12}dE
+(C, -f) L(mv) (1:1' + I: +v~1') dE
+c, h, (1~12 + 1~12 + I~!T + IZI2) dE
Since () = 0"8(0) on J, we have
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)
R(t)lif + [ (11+11;"(0) + 1I+lIltf,(0) dt ~ c [hi (m .v)I+I' dE

- ~Jm. v) (C[c(cp)]r). ~~ + G1 hlr (\i'Z + cp)1 2 ) dE
1
+h,(m.v) (Cle(~)lv): +G'hl~~1') dE]

+ C, h,<m. v) (I~:I' + I~~ + ~I') dE V
+ ~ r (m. v){ ml1Z1 2 + IPllcpl2} dE + C2Jor

JEJ
T
18(0, tW dt.
We next need to obtain an estimate analogous to (4.25) on the time

integral of the total beam energy
CB(t) = 11 (m21i1 2+ P2hl021 2+ P2 I IOlI 2) d1.
+ 11 (G2IIO~12 + Eh10;1 2 + G2Alz' + (2 12 ) d1..
Denote by EB(~' t) the energy density:
CB(t) = 11 EB(~' t) d1..
It follows from [49, Proposition 3.1] that the following identity holds for 8:
r(t)16 + ~ 1T 11 [~' EB - 2~m2i02 + 2~G2AO;(z' + (2)] d1.dt

~(l) [T ~(o) [T
= -2- 10 EB(l, t) dt - -2- 10 EB(O, t) dt,
where ~ is an arbitrary C 1 [0, l] function and
r(t) = 11 ~[P2I010~ + m2i(z' + ( 2) + P2h020;] d1..
(This identity is proved by multiplying the three equations in (4.5) by ~z',

~O~ and ~O~, respectively, adding the three products, integrating the sum
over [0, l] x [0, T] and using integration by parts.)
We choose ~ so that
~ ~ 0, ~'> 0, ~2m2 - (~')2p2h < 0,

~2G2A - (~')2Eh < 0, 0:::; ~:::; l.
For example, one may choose
Then the quadratic forms
~'m2q~ 2~m2qlq2 + ~'P2hq~

and
~'Ehq~ 2~G2Aqlq2 + ~'G2Aq~
are uniformly positive definite on : :; : :;
~ l, hence
~'EB - 2~m2i02 + 2~G2AO;(z' + (2)

= ~'[p2IIOlI2 + G2II0~12] + [~'m2IiI2 - 2~m2i02 +~' P2h1021 2]
+ [~I Eh18~12 + 2~G2A8~(Z' + (2 ) + ~/G2Alz' + 0212] > eoEB
for some Co
> and the following estimate holds for the solution E:
We estimate in the right side of (4.26), using the junction conditions

(4.10), as follows:
~(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
- ~~O) loT (G2I10~ (0, t)12 + EhIO~(O, tW + G2AI(z' + ( 2)(0, t)12) dt

oS C (LT IS(O, t) I'dt + 1q~ h,dI' Iv, (VZ + '1') I'dE
+ [O'~ df [J IC[c(cp)]V I2 dE) .
We then obtain from (4.26) the estimate
(4.28) r(t)li + Co (loT IISIIl,..(O,l)dt

+ loT l (G,Ilo; I' + EhIO;I' + G,AIz' + 0,1') d{dt )
oS C (loT IS(O, t)I'dt + h, (IC['I')]vl' + v (VZ + '1')1') dE)

oS C [loT (IS(O, t)I' + 15(0, t)I'dt
+ h, 1m vi (1:1\ I~~ +v '1'1') dE] ,
in view of (4.21) and the geometric junction condition, where C depends
the material parameters and on the HI (J) norms of 0'0 and 0'1.
Multiply (4.28) by E and add the product to (4.25) to obtain an estimate
of the form
[R(t) + fr(t)]~ + C{) loT U4.;1I~O(Cl)dt
+ C{)f [T "8"~O(O,l)dt + C{) [T 11.1I~1 (Cl)dt

10 10"Y
+ C{)f 1T 1'- (G2IIO~12 + EI 10;1 2 + G2Alz' + (
h 2 12 ) d1.dt
<: Co [hi (m . v)I+I' <II: + f.T 15(0, ')1' at

-lJm. v) (C[c(cp)]-r). ~~ + G hl-r (V'Z + cpW) dI::
1
h 1:1') <II:]
I
+ t (m v) ((Cle('P)Jv) : +G,h
+ (C, - ,C) h}m. v) 0:1' I:~ + +V 'PI') <II:

+~ [ (m. v){ m11Z1 2 + Ipll~12} dI:: + fC [T 18(0, t)12 dt.
lEJ 10
for some constants Co, C1 , C independent of f and T.
Choose f > 0 so that C1 - fC ~ O. Also, it is possible to choose f so
small that
In fact, by utilizing the geometric junction conditions one obtains
[ (m v) [m11Z12 + I pl ltPl 2 ] dI::

lEJ
= [ (m,v){m10'5Iz(0,t)+(01(O,t)1 2
lEJ
+ Ipi 0'~(l01(0, t)12 + 102(0, t)12)}dI::.
Since m v < 0 on J, the region of integration on the right may be replaced
by I:: Ja , where I::Ja = Ja x (0, T). On Ja we have O'i(X) = 1, (x) = ~2'
where x = Xo + 6-r(Xo), and
(m v)(x) = (Xo + 6-r(Xo) - fco) . v(Xo)

= (xo - fco) . v(Xo) = constant < O.
Since
we have
from which (4.29) immediately follows.

We have therefore proved the inequality
(4.30) [R(t) + r(t)]~ + Co [T II~, 8), (;, 8))II~r xHdt

Jo
[h,
0
,; Co (m v)I;j.I' dE + J,T 18(0,1)1' dl

-kJm. v) (C[c(cp)]-r)' ~~ + G1hl-r (\i'Z + cpW) dE
h I:: J') dEl

1
+ 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
Moreover, it is easy to see that
IR(t) + r(t)/ ::; C"~, 8), (;, E))"~ro xH

= CII~O,80),(~1,81))II~roxH'
Therefore, (4.30) may be rewritten
(coT - 2C)II((CI>,So), (CI>1, S1))11~I'oXH
~ Co [ht (m . v)[ .. [, dE + J,T [5(0, t)[' dt
-lJm.v) (C[c(cp)]r)' ~~ +G hlr(VZ+cpW) dE 1

1
+ ht'mV) ((C[e(<p)[V)' ~~ +Glhl~~I') dE],

hence for T > To := 2C/CO,
(4.31) II (( CI>0, So), (CI>1, S1)) II~I'o ~ CT [[JEt 1c)1

xH 2 dE
+ 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.
LEMMA 4.1. Under the assumptions of Proposition 4.1 we have
(4.32) [[5(0, t)['dt ~C [ht [..[, dE
+ l-
1
(C[c(cp)]r) . ~~ + G1 hlr (VZ + CP)1 2 ) dE
+ h t ((c[e(<p)]v), ~~ + G1h I~m dE].
PROOF. The proof uses a standard compactness-uniqueness argument.

One first proves that
(4.33) II( (.0,8"), (.',8' ))II;'" ,H " CT [hi 1+1' dE
+ ~_ (C[c(It')]-r). ~~ +G1hl-r.(\7Z+It')12)dI.:
h
1
+ t ( qe('P )]V) . : + G,h 1~~ I') dE + II. II l.-(O,T;1t'(O 1

for T > To by showing that
(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,
(4.37) hi 1+.1' dE + JEt (c]e('Pn)]V)' OJ.: + G,h I~ I') dE

+ ~_ (C[c(lt'n)]-r) . 8;n + G1hl-r (\7Zn + It'nW) dI.: -+ 0.
1
It follows from (4.31), (4.35) and (4.37) that +~, 8~), (+~, 8~)) is bounded
in V x H and, therefore
We may then conclude that, as n ---+ 00 through some subset of positive

integers,
(4)>n' En) converges strongly in VX>(O, T; [V, H]aJ, 0< 0: ~ l.
In view of (4.35) and (4.36) we may then conclude that
and
(4.39) loT IEn(O, tWdt loT IE(O, t)1 2dt = l.

---+
From (4.38),
Since the geometric junction condition 4n = aEn(O) holds along J,
En(O,) ---+ strongly in Loo(O, T),
which contradicts (4.39). This proves (4.34) and hence (4.33).

Let X be the linear space of all (4)>, E) which satisfy the conditions
(4)>,E) E Loo(0,T;V)nW1,00(0,T;H),
(4)>, E) satisfy (4.4)-(4.10),
(4.40) C[c:(<p)]v = 0, ~~ = on Et,
cp = 0, Z = on Et,
arp az
T aT = 0, aT + T . rp = 0 on El
with norm
11(4)>, E)lIx = 11((4, E), (ci;, 8))llv"'(o,T;vxH).

We wish to show that X = {O}. To this end, we set X = (4)>, E) and suppose
that X E X. Then the difference quotient (X( + h) - X())/h E X and
from (4.33) we have
(4.41) II (X(h) - X(O))/h, (X(h) - X(O))/h)llvXH

~ CTII(4)>( + h) - 4(-))/hllv"'(O,T;1tO(O).
Since X E X, the right side of (4.41) converges to 11ci;IILOO(O,T;1tO(O) as h---+

0, hence (X(h) - X(O))/h ---+ X(O) weakly in V, (X(h) - X(O))/h ---+ X(O)
weakly in Hand
II(X,X)IILoo(o,T;VXH) ~ CTII<)IILoo(o,T;'H O(fl)

~ CT II (X, X)IILoo(o,T;vxH) = CTIIXlix.
Hence XEX and the linear map
(4.42)
is continuous. It then follows by the Arzela-Ascoli Theorem that the injec-

tion
{(~,8) E XI (<),8) E X} f--+ X
is compact, hence X itself is compact and, therefore, finite dimensional.
Since the map (4.42) is continuous, it has an eigenvalue 'T1 (here we need
to work in the complexification of X). In particular, <) = 'T1~ so that for
each t E (0, T), ~ = <p + Zk satisfies
ml'T12Z - G 1 hdiv(\7Z + <p) = 0,

(4.43) {
Ipl'T12<p - div(C[(<p)] + G 1h(\7Z + <p) = n,in
(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)
P2I'T1281 - G2IOr = 0, < ~ < f.

Therefore X = {o} if rt =f. 0. On the other hand, if rt = 0, then
necessarily rt =f. 0. The pair cp, Z also satisfies (4.43) and the Cauchy
conditions
cp = 0, Z = 0, C[c:(cp)]v = 0, ~: = on rt.
Thus cp = 0 and Z = 0 in 0 x (0, T) and then, as above, we obtain 8 = 0

(0,) x (O,T). It then follows from (4.4)-(4.10) that (~,E) satisfies the
homogeneous, stationary problem
diVC[(<P)]- G1h(VZ + <p) = 0,

{
G1hdiv(VZ + <p) = 0 in OJ
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
G2IO~(0) = L {G1h(O"ov. (VZ + <p)
-0"1e2 . (C[(<p)]V)}dr,
EhO~(O) = L 0"1 e1 . (C[(<p)]v) dr,
G2A(ZI + (2)(0) = G1h L O"oV' (VZ + <p) dr on J.
If r 0 =I- 0, the only solution of this problem is ~ = 0, E = 0 and once again

we conclude that X = {o}.
To complete the proof of the lemma, let us suppose that (4.32) is false.
Then there is a sequence of initial data (( ~~, E~), (CI~, E~)) E D A X V
such that the corresponding solution (CIn' En) of (4.4)-(4.11) satisfies
(4.46) h/I>,I'dE + ht (IC[e( <Po)[vi' + I8:",1') dl:

+ ~_ (C[(<Pn)]T)' a:,.n +G1hIT'(VZn+<Pn)12)d~~0.
1
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).
We may then conclude that, as n ---+ 00 through some subset of positive

integers,
(4.47) (~n' En) converges weak* in Loo(O, T; V) n W1,00(0, T; H)

and strongly in Loo(O, T; [V, H]a), 0< a ~ 1.
If (~, E) denotes the limit function, then (~, E) satisfies, in view of (4.45)-
(4.47),
(4.48) IT IE(O, tWdt = 1,
(~, E) E Loo(O, T; V) n W1,00(0, T; H),

(~, E) satisfy (4.4)-(4.10),
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
STf = IT expT - s)A)Bf(s) ds.
We have already proved that ST E (L2(0, T; U), H x V~) for each T >0
(see (3.33)). Since
(4.49) (STf,yO)HXV., = IT(f(S),B'Y(S))udS,
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
(4.50) IIB'Y()IIL2(O,TU) ~ eo II yO IIHX V., , WO E H X V,),,

for some eo > o. From (3.31) it is seen that (4.50) is equivalent to showing
that
~D {IC[c(<p)]vI 2 + Kl 1~~ 12 }dE ~ eoll~o, EO), (~l, E1))II}.,XH

For T large enough the last estimate follows from Corollary 4.1.
5. DERIVATION OF THE LIMIT MODEL 363
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))
B'y = ~b~1 = cp + ZkI EI

However, (4.50) will not hold in general, so to prove Theorem 4.2 we use
the Hilbert Uniqueness Method. We define a semi-norm
lIyOIiF = II B'Y()II2(o,T;U), yO E H X V-y

which, according to Corollary 4.2, is a norm if T > To, and we define a
space:F which is the completion of H x V-y with respect to IIIIF. Then S~
is an isometry of:F onto L2(0, Tj U). Since
(STS~ZO, yO)F = (ZO, yo).?,.,
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. Limit Model as the Shear Moduli Approach Infinity

In this section we consider the limit as G1 -+ 00 and G2 -+ 00 in the
model
(5.1) { IPI~. - divC[(u)] + G1h(V'W + u) = Mt,

m1W - G1hdiv(V'W + u) = Fl in OJ
m 2iiJ - G 2 A(w' + th)' = F2 ,

{
(5.2) P2I~~2 - EhiJ~ + G 2A(w' + iJ 2) = M22 ,
P2Iiit - G2[iJ~ = M 2b 0 < ~ < Ij
(5.3)
(5.4) W = 0, u = 0 on foj
G1h(V'W + u) v = 0,
(5.5) {
C[(u)]v = 0 on f1j
(5.6) W = w(O) + 6iJ l (0), u = -6(0) on J a ,

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
G2A(w' + t?2)(0) = Glh [ (V'W + U) dr.

JJ
V
a
It is assumed that n is bounded, open and connected with a Lipschitz

boundary consisting of a finite number of smooth, simple curves, that := an
r = foUI\ uJa where r o, rl, Ja are relatively open in r, mutually disjoint
and Ja #- 0. For simplicity we assume that r o #- 0. This model corresponds
to the situation, discussed in Section 2, where J - J a is free, where there
r
are no external forces or moments acting on D and where we set r = r c ,
rl=r D U ( J-J- a ).
We write G for the pair (Gl, G2 ) of shear moduli, G > 0 and G -+ 00 to
mean that G i > 0 and G i -+ 00 respectively, for i = 1,2. As above, we set
\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
(5.8) (ii, R)v + (R, R)v
=lFl.~dn+ llF2.8d/;" vR=(~,8)EV,
(5.9) R(O) = Ro, R(O) = Rl,

where
Fl = Ml + Flaa, F2 = M22 e l - M2l e 2 + F2ea
It follows immediately that if
Ro E V, Rl E H, (Fl, F 2) E Ll(H),
(5.8), (5.9) has a unique solution with regularity
R E C(V) n Cl(H) n Ll(V').
We have written, in order to simplify the notation, Ll(H), C(V), etc., in
place of Ll(O, Tj H), C([O, T]j V), etc.
The space V and solution R both depend on G. We indicate this de-

pendence explicitly by writing VG and RG in their places. Thus we write
(5.8), (5.9) as
". G ' "
(5.10) (RG,R)va+(R ,R)va=(G,R)H, 'v'R=(W,8)EVG,
(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
(5.12) IIRllHoo = IIRIIH = [In (mllWl + 2 IplIVWI2) dO
+ [(m,lwl' + p,lh Iw' I') de + flIDll'r"
An element R in the orthogonal complement H ~ of H 00 in H is charac-

terized by the variational equation
In (mlWW - Ip1u, VW)dO
+ fol (m2ww - P2h19 2w' + Uh-Ol) d{ = 0, vB. E Hoo,

which is formally equivalent to the system
ml W + Ipl div u = 0 in 0, V Ulrl = 0;

m2W + I p1 19; = 0 in (0, I), 192(1) = 0;
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}.
Functions in V00 satisfy the geometric boundary conditions
(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
In analogy with Theorem 6.1 of Chapter VII, we have

THEOREM 5.1. Assume (5.16). There is a function R E C(Voo) with

it E C(Hoo) and R E LZ(V~) such that as G - t 00,
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~).
Moreover, R is the unique solution of
(5.17) R + AooR = Goo,

(5.18) R(O) = Ro, it(O) = R 1.
PROOF. From (5.10), (5.11) and assumption (5.16) we have
IlitG(t)II~ + IIRG(t)II~a = IIR1II~oo + llRoII~oo + I T

(G(S),itG (S))HdS
~ IIRtII~oo + llRolit + IIitGIILOO(H)IIGIIL1(H),

hence
Therefore
(5.19) itG is bounded in Loo{H),

(5.20) IIRGIILOO(Va ) is bounded,.
From (5.19) and assumption (5.16) we may conclude that
RG is bounded in W1,00(H).
Therefore, upon passing to the limit as G - t 00 through some subnet of

G > 0 we obtain, for some R E W1,00(H),
(5.21) { ~G -t ~ strongly in Loo(H),

RG -t R weakly* in Loo(H).
In particular, the map t f-+ R(t) : [0, T) f-+ H is continuous and R(O) = Ro.
Let us write
RG = Ci{rG, e G), q,G = u G + WGa3, e G = 1?~el -1?fez + wGe3,

R = ('11, e), '11 = u + Wa3, e = 1?Zel -1?lez + we3.
From (5.20) and the definition of the VG norm we have
(5.22) ea.B(UG ) is bounded in ,,0(L2(0)),

(5.23) #t/VW G + u G / is bounded in Loo(L2(0)),
(5.24) (19f)' is bounded in Loo (L 2 (0,)),
(5.25) JQ;/(w G ), + 19f/ is bounded in L oo (L 2 (0, f~,
(5.26) JQ;(19?)' is bounded in L oo (L 2 (0,)).
Since, by assumption, fo i- 0, (5.22) and Korn's Lemma imply that
(5.27) u~ is bounded in Loo(Hl(O))

and then, by (5.23),
(5.28)
Thus by (5.21), (5.23), (5.27) and (5.28),
wG is bounded in L oo (1ifo(O)),

/VW G + u G / _ strongly in Loo(L2(0
and therefore
\fiG _ \fI weakly* in Loo(1tf o (O)),

(5.29) {
U= -Vw.
Similarly, from (5.21), (5.24)-(5.26) we conclude that
19f is bounded in L oo (H 1 (0, f~,

w G is bounded in L oo (H 1 (0,,
(19?)' -
strongly in L oo (L 2 (0,)),
(w G )' + 19f -+ strongly in L oo (L 2 (0,,
hence
eG _ e weakly* in Loo(1t 1 (0,,

(5.30) {
19 2 = -w', 19i = 0.
It follows from (5.21), (5.29) and (5.30) that R E Loo(Voo) n C(Hoo) and
It E Loo(Hoo ).
The remainder of the proof is identical to the last part of the proof of
Theorem 6.1, Chapter VII, beginning with (6.39), and is omitted. 0
5.1. Interpretation of the Limit Model. It remains to interpret

(5.17), (5.18). Equation (5.17) may be written
Let us calculate each of the three terms in (5.31). One has, at least formally,
(5.32) (ii, R)voo = In (ml WW + IPi VW . VW) dO

+ 1'-(m2WW + P2hw'w') ~ + If:OlJl
= r (ml W - Ipi dW)W dO + 1f\r
1n
r Ipi w88Wv dr
o
+ 1'-(m2W- p2hw")w~ + P2h[w'wl& + u-OlJ.

Next, we calculate
The first term on the right was evaluated in Chapter VI (see (6.12)) with
the result
where Dl = (2J.Ll + Adh3 /12 and
CT(VW) = T (C[(VW)]v), Cv(VW) = v (C[(VW)]v).

Thus
(5.33) (R, it)voo = Dl { f W Ll2W dO + f

In Jr\ro
OW CII(VW)
v
dr
_ f
Jr\ro
W [OLlW
ov
+ ~CT(VW)]
OT
dr}
+ Eh {/, uI"'wd{ + lui',;,' ~ w"'wli } .
Finally, let us calculate, for it E H 00,
(5.34) (Goo, it)Hoo = (G, it)H = In (Fl W - Ml . VW) dO
+ It (F2w - M22W l + M 2l Jl ) df.
= In (Fl + divMt}W dO + It(F2 + M~2)wdf.

- f (v. Ml)W dr + J ft df. - [M22wl~.
l M2l
Jr\ro Jo
It follows from (5.31)-(5.34) that W, w, {)1 satisfy the following system
of equations, boundary and junction conditions.
(5.35)
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 ;
(5.38) m2 W.. - P2 I h W.. /1 + EIh W 1111 = r2

D
+ Mol22
ill
(0 ,~;
D)
(5.39)
(5.40) W = W () ., oW
0, + 6'!?l () ov = W
I()
0, on Ja .
There is; in addition, the variational junction condition
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
W = w(O) + 6t?b ~~ = w'(O) on Ja ,
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.
(5.41) P2hw'(O,') - Ehw"(O,) + M22 (0,')
1.
~ {I p, : -Dl [8~: + !CT(VWJJ}dr;
(5.42) Ehw"(O,) = Dl ( C,,(VW) dI';
1Ja
UD l (-) - 101 M2l(~,') dt;,

e+
(5.43)
~ t. p, : -Dl [8~ + !CT(VWJJ}dr.

The complete system of equations, boundary and junction conditions is
(5.35)-(5.43). It is a type of hybrid system in that it couples partial differ-
ential equations for w, W with an ordinary differential equation for 1?1.
Under assumption (5.16), the initial conditions (5.18) reduce to
0' 1 .
Wlt=o = W , Wlt=o = W III n,
Wlt=o = wO, wlt=o = WI in (0,1),
1?llt=o = 1?~ == const., Jllt=o = 1?~ == const.,
where WO E Hfo(n), WI E Hfo(n), wO E H2(O,f), WI E Hl(O,f) and
WO = wO(O) + 61?b 0;;0 = (wo)'(O) on J a.

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. Modeling of a Plate-Beam Junction: II

6.1. Geometric Conditions. Let n and w be bounded, open, con-
nected sets in IR2 with Lipschitz continuous boundaries consisting of a
finite number of smooth simple curves. We further assume that w is simply
connected, that 0 E w and that w is doubly symmetric with respect to O.
We consider n to be the reference surface of a thin plate of uniform thick-
ness h. Points within the plate will be denoted by coordinates (Xl, X2, X3)
with respect to the natural basis a!, a2, a3 for IR3. Thus the equilibrium
position of the plate is given by
{xi~1 (Xl,X2) EO, IX31:::; h/2}.
where summation convention for repeated indices has been assumed as in
Section 2.
Let Xo be a point of IR3 and ell e2, e3 be a right-handed orthonormal
basis for IR3. The beam, in its reference configuration, is given by
The following assumptions are made:

(AI) wen.
(A2) Xo = (h/2)a3, e1 = a3, e2 = a!, e3 = a2.
Thus the beam is attached to and orthogonal to the upper face of the plate
and w is the junction region between the plate and the beam.
As in Section 2, we denote by R(X1' X2, X3, t) the position vector at time
t to the deformed position of the material particle in the plate which is
located at (Xl, X2, X3) when the plate is in equilibrium, by W(Xl, X2, t)
the displacement vector of the material particle which occupies position
(Xl,X2,0) in the reference surface in equilibrium, by r(6'~2,6,t) the po-
sition vector to the deformed position of the particle originally at position
Xo + ~iei of the beam in equilibrium, and w(6, t) the displacement vector
of the point Xo + 6 e1 on the centerline of the beam. We further retain the
kinematic hypotheses (2.1) and (2.2) of Section 2.
As a geometric junction condition we impose the continuity condition
6. MODELING OF A PLATE-BEAM JUNCTION: II 373
Since al = e2 and a2 = e3, (6.1) is the same as

h
(6.2) W(XI. X2, t) + i U(XI. X2, t) = w(O, t)
+ Xl (E2(0, t) - e2) + x2(E3(0, t) - e3), (Xl. X2) E W, t > 0.
In terms of components, the last relation is
h
WI (Xl. X2, t) + iUI (Xl. X2, t) = W2(0, t)
+ Xl (cos 'i?l cos'i?3 + sin 'i?l sin'i?2sin'i?3)(0, t)
+ X2(COS'i?1 sin'i?2 sin'i?3 - sin 'i?l cos'i?3)(0, t),
h
W2(XI.X2,t) + iU2(XI.X2,t) = W3(0,t)
+ Xl (sin 'i?l cos 'i?3)(0, t) + X2( cos 'i?l cOS'i?2 - 1)(0, t),
and
h
W3(XI.X2,t) + iU3(XI.X2,t) = WI(O,t)
+ Xl (sin 'i?l sin 'i?2 cos 'i?3 - cos 'i?l sin 'i?3)(0, t)
+ X2 (sin 'i?l sin'i?3 + cos 'i?l sin 'i?2 cos 'i?3) (0, t),
where
W = Wi 8.i, U = Ui 8.i, ,W = Wiei.
When linearized, the above scalar relations reduce to
h
WI (Xl. X2, t) + 2"UI (Xl. X2, t) = W2(0, t) - x2'i?1 (0, t),
(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
displacements. The latter must obey, in particular, the geometric junction

conditions (6.3).
-D D
We assume that r = r u r where r and r are relatively open in
-0 0
r and mutually disjoint. The part r O corresponds to geometric boundary

conditions
(6.5) W=o, v=O, u=O onro ,
while rD corresponds to mechanical boundary conditions.

We write
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
(6.8) Wp(t) = f(Fp.W+Mp.U)dO+ f (fpW+mpU)dr,

}n }r D
where F p , M p , fp , mp are various resultant (in X3) forces and moments as
discussed in Section 2.
The total kinetic and potential energies of the beam are calculated as in
Section 2 (cf. Chapter III). One finds that
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
M12 = 1t 6FB . el~' m12 1

= e6fB . el~'
We may now calculate (6.4), where the variation is taken with respect
to displacements satisfying (6.3) and (6.5). In the calculation we utilize
the Green's formula (2.17). Standard computations lead to the following
dynamic equations.
Equations of motion of the plate: For (Xl, X2) w and t > 0, En \
mlV - divC[e(v)] = (F p . ap)ap,
(6.12) m l W3 - G l hdiv(VW3 + u) = Fp a3,
Ipl Ii - ~~ div C[e(U)] + Glh(VW3 + u) = (Mp . ap)ap.
Equations of motion of the beam: For 6 E (0, l) and t > 0,
m2Wl - EAw~ = fB el,

m2w2 - G2A(w; - 'l?3)' = FB . e2,
m2w3 - G2A(w; + 'l?2)' = FB . e3,
(6.13)
P2It9l - G2I'!?~ = -M23 ,
P213t92 - EI3'1?~ + G2A(w; + 'l?2) = M13,
P212t93 - EI2'1?~ - G2A(w; - t?3) = -M12'
Boundary conditions along rD , t > 0:
C[c(V)]v = (fp . a(3)a{3,
(6.14) G1h(V'W3 + u) . v = fp . a3,

h2
12 C[c(u)]v = (mp . a(3)a{3.
Boundary conditions at { = f, t > 0:
EAwi = fB eb
G2A(w~ - t?3) = fB . e2,
G2A(w~ + t?2) = fB . e3,
(6.15)
G2It?i = -m23,
EI3t?~ = m13,
EI2t?~ = -mI2.
There is, in addition, the variational junction condition
0= EAwi (O)WI(O) + G2A(w~ - t?3)(0)W2(0) + G2A(w~ + t?2)(0)W3(0)

+ G2 It?i(0)J1 (0) + EI3t?~(0)J2(0) + EI2t?~(0)J3(0)
+ l{v. divC[c(v)] + h2 u divC[c(u)]}dw
i
w 12
+ {G 1hdiv(V'W3 + U)W3 - G 1h(V' W3 + u) u}dw
-i (mIW. Vi +Ip1u 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
(6.16) 0= i {G1hdiv(V'W3 + u) - ml W3}W3 dw
+ EAwi (O)WI (0) + EI3t?~(0)J2(0) + EI2t?;(0)J3(0)

and
(6.17) 0 = G2A(w~ -193)(0)w2(0)

+ G2A( w; + 192 ) (0)W3(0) + G2I19~ (0)t9 1 (0)
L
+ {~~ divC[c(u)] - G 1 h( VW3 + u) - Ipiu} UcU.v
+ L{divC[c(v)] - mlU} VcU.v.
By utilizing the last of the junction conditions in (6.3) in (6.16) we obtain

the dynamic junction conditions
L {ml W3 - G1 hdiv(VW3 + u)} cU.v = EAw~ (0),

(6.18) L x2{ml W3 - G1 hdiv(VW3 + u)} cU.v = EI319~(0),
L xdmlW3 - G1 hdiv(VW3 + u)}cU.v = EI219;(0).
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
in view of (6.3). We thereby obtain
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
{Ipi ii - ~~ divC[c(u)] + G1h(VW3 + u)} ucU.v

-L {ml"v -divC[c(v)]} {w2(0)al +w3(0)a2 + (xla2 -x2a dt9 1(0)} cU.v,
where U, W2(0), W3(0), J 1(O) are arbitrary. Therefore
l (m1 WI - a1 . divC[c{v)]) dw = G2A{w~ - '!93){0),
(6.19) l (m1 W2 - a2 . divC[c{v)]) dw = G2A{w; + '!92){0),
l (X1 a2 - X2 a1) . (ml V -. divC[c{v)]) dw = G2I'!9~(0),

and
(6.20) l {Ip1ii- ~~ divC[c{u)J +G h{V'W 3 +u)}dw

1
= ~ l {m1 v - divC[c{v)]}dw
= ~G2A[{w~ - '!93)a1 + (w; + '!92)a2J{0).
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
[18J Ph. G. Ciarlet, Mathematical Elasticity, Vol. I: Three-Dimensional Elasticity,

North-Holland Elsevier Science Publ., 1988.
[19J ___ , Plates and junctions in elastic multi-structures: An asymptotic analysis,
Masson and Springer- Verlag, 1990.
[20J F. Conrad and M. Pierre, Stabilization of second order evolution equations by
unbounded nonlinear feedback, Tech. Report RR 1679, INRIA, 1992.
[21J R. F Curtain and A. J. Pritchard, Infinite dimensional system theory, Lect. Notes
in Control and Inf. Sc., vol. 8, Springer-Verlag, 1978.
[22J D. A. Danielson and D. H. Hodges, A beam theory for large global rotation,
moderate local rotation and small strain, J. Appl. Mech. 55 (1988), 179-185.
[23J R. Datko, Not all feedback stabilized hyperbolic systems aer robust with respect to
small time delays in their feedbacks, SIAM J. Control Optim. 26 (1988), 697-713.
[24J ___ , A rank-one perturbation result on the spectra of certain operators, Pro-
ceedings of the Royal Society of Edinburgh UIA (1989), 337-346.
[25J R. Datko, J. E. Lagnese, and M. P. Polis, An example of the effect of time delays
in boundary feedback stabilization of wave equations, SIAM J. Control Optim. 24
(1986), 152-156.
[26J M. C. Delfour, M. Kern, L. Passeron, and B. Sevenne, Modelling of a rotating
flexible beam, Control of distributed parameter systems, 4th IFAC Symposium,
Los Angeles 1986 (H. E. Rauch, ed.), Pergamon Press, 1986, pp. 383-387.
[27J Ph. Dystunder, Remarks on substructuring, Eur. J. Mech. 8 (1989), 201-218.
[28J 1. Ekeland and T. Temam, Convex analysis and variational problems, North-
Holland/American Elsevier, Amsterdam, Oxford, New York, 1976.
[29J A. E. Green, The equilibrium of rods, Arch. Rational Mech. Anal. 3 (1959), 417-
421.
[30J P. Grisvard, Elliptic problems in nonsmooth domains, Pitman, London, 1985.
[31J _ _ , Singularites en elasticite, J. Math. Pures Appl. 107 (1989),157-180.
[32J 1. Gruais, Modelisation de la jonction entre une poutre et une plaque, Ph.D.
thesis, Universite Pierre et Marie Curie, 1990.
[33J E. Hairer and G. Wanner, Solving ordinary differential equations ii. stiff and
differential algebraic systems, Springer-Verlag, Berlin, 1987.
[34J S. Hanson and E. Zuazua, Exact controllability and stabilization of a vibrating
string with an interior point mass, Tech. Report 1140, Institute for Mathematics
and its Applications, University of Minnesota, 1993.
[35J A. Haraux, Comportement a l'infini pour une equation d'ondes non lineaire dis-
sipative, C. R. Acad. Sc. Paris 287 (1978), 507-509.
[36J ___ , Nonlinear evolution equations - global behavior of solutions, Lect. Notes
in Mathematics, vol. 841, Springer-Verlag, Berlin, 1981.
[37J Lop Fat Ho, Exact controllability of second order hyperbolic systems with control
in the dirichlet boundary conditions, J. Math. Pures Appl. 66 (1987), 363-368.
[38J S. M. Howard, Transient stress waves in trusses and frames, Ph.D. thesis, Cornell
University, 1990.
[39J T. J. R. Hughes, The finite element method. Linear static and dynamic finite
element analysis, Prentice-Hall, New Yersey, 1987.
[40J G.Leugering J.E. Lagnese and E.J.P.G. Schmidt, On the analysis and control
of hyperbolic systems associated with vibrating networks, Proc. Royal Soc. Edin-
burgh, to appear.
[41J T. R. Kane, R.R. Ryan, and A. K. Barnerjee, Dynamics of a beam attached to a
moving base, AIAA J.Guidance, Control and Dynamics 10 (1987), 139-151.
[42J F. Khorrami, Analysis of multi-link flexible manipulators via asymptotic expan-
sions, Proceedings of the 28th Conference on Decision and Control, Tampa,Fl,
1989 (L. Shaw, ed.), IEEE, 1989, pp. 2089-2094.
[43J V. Komornik, A mathematics paper, J. Math. Pures Appl. U5 (1987), 1-18.
BIBLIOGRAPHY 381
[44] J. E. Lagnese, The Hilbert uniqueness method: A retrospective, Optimal Control

of Partial Differential Equations (Berlin) (K. H. Hoffmann and W.Krabs, eds.),
Springer-Verlag, Berlin, 1991, pp. 158-181.
[45] ___ , Uniform asymptotic energy estimates for solutions of the equations of
dynamic plane elasticity with nonlinear dissipation at the boundary, Nonlinear
Analysis: Theory, Methods and Appl. 16 (1991), 35-54.
[46] ___ , Boundary controllability in transmission problems for thin plates, Dif-
ferential Equations, Dynamical Systems and Control Science: A Festschrift in
Honor of Lawrence Markus (New York) (K. D. Elworthy, W. N. Everitt, and
E. B. Lee, eds.), Marcel Dekker, New York, 1993.
[47] J. E. Lagnese and G. Leugering, Uniform stabilization of a nonlinear beam by
nonlinear boundary feedback, J. Diff. Eqns. 91 (1991), 355-388.
[48] ___ , Modelling of dynamic networks of thin elastic plates, Math. Methods
Appl. Sci. 16 (1993), 379-407.
[49] J. E. Lagnese, G. Leugering, and E. J. P. G. Schmidt, Controllability of planar
networks of Timoshenko beams, SIAM J. Control Optim. 31 (1993),780- 811.
[50] ___ , Modelling of dynamic networks of thin thermoelastic beams, Math. Meth-
ods. Appl. Sci. 16 (1993), 327-358.
[51] J. E. Lagnese and J. L. Lions, Modelling, analysis and control of thin plates,
Collection RMA, vol. 6, Masson, Paris, 1988.
[52] J.E. Lagnese, Boundary stabilization of thin plates, Studies in Applied Mathe-
matics, vol. 10, SIAM, Philadelphia, 1989.
[53] I. Lasiecka, Stabilization of wave and plate-like equations with nonlinear dissipa-
tion on the boundary, J.Diff.. Equs. 79 (1989), 340-381.
[54] I. Lasiecka and D. Tataru, Uniform boundary stabilization of semilinear wave
equations with nonlinear boundary damping, J.Diff.lnt.Equ. (19), -.
[55] I. Lasiecka and R. Triggiani, A lifting theorem for the time regularity of solutions
to abstract equations with unbounded operators and applications to hyperbolic
equations, Proc. Amer. Math. Soc. 104 (1988), 745-755.
[56] H. LeDret, Modelling of the junction between two rods, J. Math. Pures et Appl.
68 (1989), 365-397.
[57] ___ , Problemes variationnels dans les multi-domains: Modelisation des jonc-
tions et applications, Masson, Paris, 1991.
[58] G. Leugering and E. J. P. G. Schmidt, On the control of networks of vibrating
strings and beams, Proceedings of the 28th IEEE Conference on Decision and
Control, vol. 3, IEEE, 1989, pp. 2287-2290.
[59] L.Halpern, Absorbing boundary conditions for the discretization of the one-
dimensional wave equation, Mathematics of Computation 38 (1982), no. 158,
415-429.
[60] J.-L. Lions, Controlabilite exacte, perturbations et stabilisation de systemes dis-
tribues, Tome I: Controlabilte exacte, Collection RMA, vol. 8, Masson, Paris,
1988.
[61] ___ , Exact controllability, stabilization and perturbations for distributed sys-
tems, SIAM Review 30 (1988), 1-68.
[62] J.-L. Lions and E. Magenes, Non-homogeneous boundary value problems and
applications, vol. I, Springer-Verlag, Berlin, 1972.
[63] A. E. H. Love, A treatise on the mathematical theory of elasticity, Cambridge
Univ. Press, 1927.
[64] Ch. Lubich, Extrapolation integrators for constrained multibody systems, Impact
of Computing in Science and Engineering (1991), 213-234.
[65] Ch. Lubich, U. Nowak, U. Pohle, and Ch. Engstler, MEXX-Numerical software
for the integration of constrained multibody systems, Preprint SC 92-12, 1992.
382 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
[91] A. Y. Le Roux, Stability of numerical schemes solving quasi-linear wave equa-

tions, Mathematics of Computation 36 (1981), no. 153,93-105.
[92] D. L. Russell, Controllability and stabilizability theory for linear partial differ-
ential equations: recent progress and open questions, SIAM Review 20 (1978),
639-739.
[93] E. J. P. G. Schmidt, On the modelling and exact controllability of networks of
vibrating strings, SIAM J. of Control and Optimization 30 (1992), 229-245.
[94] ___ , On the modelling and exact controllability of networks of vibrating
strings, SIAM J. Control Optim. 30 (1992), 229-245.
[95] E. J. P. G. Schmidt and Wei Ming, On the modelling and analysis of networks
of vibrating strings and masses, Tech. Report 91-13, Department of Mathematics
and Statistics, McGill University, 1991.
[96] E.J.P.G. Schmidt and Wei Ming, On the modelling, analysis and control of vi-
brating networks of strings and masses, submitted manuscript, 1993.
[97] J. Schmidt, Entwurj von Reglern zur aktiven Schwingungsdiimpfung an flexiblen
mechanischen Strukturen, Ph.D. thesis, Technische Hochschule Darmstadt, Fach-
bereich Mechanik, Darmstadt Germany, 1987.
[98] A. Schneider, On spectral theory for the linear sel fadjoint equation fy = Agy,
Ordinary and Partial Differential equations, Lecture Notes in Mathematics, vol.
846, Springer-Verlag, 1981, pp. 306-332.
[99] A. A. Shabana, Dynamics of multibody systems, John Wiley, New York, 1989.
[100] J. C. Simo, A finite strain beam formulation, Part 1: the three-dimensional
dynamic problem, Compo Meth. in Appl. Mech. and Engng. 28 (1986), 79-116.
[101] J. C. Simo, J.E. Marsden, and P. S. Krishnaprasad, The hamiltonian structure
of nonlinear elasticity: the material and convective representation of solids, rods
and plates, Arch. Rat. Mech. and Anal. 104 (1988), 125-185.
[102] J. Simon, Compact sets in LP(O, Tj B), Ann. Mat. Pura et Applicate CXLVI
(1987), 65-96.
[103] A. Simpson and B. Tabarrok, On Korn's eigenvalue procedure and related meth-
ods of frequency analysis, Quart. Journal Mech. and Applied Math. 21 (1968),
1-39.
[104] 1. S. Sokolnikoff, Mathematical theory of elasticity, R. E. Krieger, Maiabar,FL,
1983.
[105] J. von Below, A characteristic equation associated to an eigenvalue problem on
c2 -networks, Lin. Alg. and its Appl. 71 (1985), 309-325.
[106] ___ , Classical solvability of linear equations on networks, J. Diff. Eqns. 12
(1988), 316- 337.
[107] ___ , Parabolic network equations, Ph.D. thesis, Habilitationsschrift, Univer-
sitat Tiibingen, 1993.
[108] G. Wempner, Mechanics of solids with applications to thin bodies, Sijthoff and
Noordhoff, Rockville, Maryland, 1981.
[109] J. Wittenburg, Nonlinear equations of motion for arbitrary systems of intercon-
nected rigid bodies, Dynamics of Multibody Systems (K. Magnus, ed.), Springer-
Verlag, 1978, pp. 357-369.
[110] W. H. Wittrick and F. W. Williams, A general algortihm for computing natu-
ral frequencies of elastic structures, Quart. Journal Mech. and Applied Math.
XXIV /3 (1971), 263-284.
[111] ___ , Natural frequencies of repetitive structures, Quart. Journal Mech. and
Applied Math. XXIV /3 (1971), 285-308.
Index
Beam networks 99, 47, 83, 101, 102, strain energy 60

eigenvalue problems 142 structural damping 76
Euler-Bernoulli beams 140 shear-diffusion damping 76
frames 84, 189 thermoelastic 47, 75
joint conditions Timoshenko system 78, 183, 189
dynamic 85, 87 total energy 74
rigid joints 83, 141 viscoelastic 75
pinned joints 83, 85, 88, 141
rotating 89 Characteristics 168, 177
Timoshenko beams 139, 183, 189 Coefficients of elasticity 203, 213
trusses 84 Control functions 168,222
see also: Networks of general1-d absorbing controls 168, 170, 175,
elements 176
Beams (single elements) bang-bang controls 182
dead-beat-controls 169
boundary conditions
directing controls 175, 177, 180
geometric 72
L-Q-R optimal controls 190
dynamic 73
robust controls 180
Bresse system 78
with time delay 179
curved 47
Control to state map 34,35,119,120
elastic 47
Curvature 47,48, 54
equations of motion 60, 71
list of models 80 Deformation 48
free energy 60 Differential algebraic systems 191
Kelvin-Voigt model 76 Displacement 48
nonlinear 47 Dry friction 180, 194
rotating 67 Dynamic stiffening 71
shearable
linear straight 76 Effective shear area 183
nonlinear curved 76 Exact controllability
386 INDEX
for networks of elastic strings 33, equation 64

37 flux 64
failure of exact controllability 40 Hilbert Uniqueness Method 243, 363
for general networks of one- Hookean 13, 61, 16, 203, 212, 213
dimensional elements 119, 123 Hybrid system 321, 371
for interconnected membranes Hyperbolic estimates 111, 116, 117
221,222
for systems of linked plates 292 Interconnected membranes 206, 213,
for plate-beam systems 345 222, 227
Elastic strings edge conditions 210, 221
non-linear equation 14 equations of motion
linear equation 22 linear 212, 221,
Euler-Bernoulli nonlinear 210
hypothesis 66, 71 exact controllability 221, 222
isotropic 212, 213, 227
Finite Elements 168, 183 junction conditions 222, 239
matrices dynamic 208, 210
bending matrix 186 geometric 207, 208, 213, 235
condensed stiffness matrix 167 kinetic energy 209
element stiffness matrix 188 observability estimates 223
lumped mass matrix 188 serially connected 238, 247, 303
shear matrix 187 strain energy 209
Newmark scheme 188 total energy 227
predictor-corrector scheme 188 Interconnected beams and strings
shear locking 188 102
Isotropic body 212
Generalized moments 255, 268
Generalized stresses 255, 268 Joint
Graph 136 membrane 207
plate 274
Hamilton's principle connected 274, 280, 284, 315
for elastic strings 14, 15 hinged 274, 280, 284, 315
for interconnected membranes semi-rigid 274, 276, 281, 285
208, 209, 227 rigid 274, 282, 285, 315
for linked plates 278 plate-beam 322
for networks of elastic strings 17, rigid 324
18,22,21 Junction
for networks of general one di- membrane 207
mensional elements 97 plate 274
for plate-beam systems 325, 378 connected 274, 280, 284, 315
for single membranes 201, 204, hinged 274, 280, 284, 315
205 semi-rigid 274, 276, 281, 285
for single plates 257, 270 rigid 274, 282, 285, 315
Heat plate-beam 322
INDEX 387
rigid 324 potential energy 13, 15, 22

spectral properties 138
Kirchhoff hypothesis 264 stabilizability 42
Kirchhoff plate system 264, 272 well-posedness of linear equations
Korn's lemma 214, 218, 332 of motion 23, 32
Networks of general one dimensional
Lagrangian 14, 15, 18, 22, 21, 97, elements 95
201,258 derivation of equations of motion
Lame parameters 267, 327 97
Linearly elastic 203, 212 eigenvalue problems 142, 156
Linked Kirchhoff plates 305 exact controllability 119, 123
equations of motion 309 kinetic energy 96
edge conditions 310 nodes and node conditions
junction conditions potential energy 97
geometric 305, 315 spectral properties 138, 139
dynamic 310, 315 stabilizability 123, 128, 129
Linked Reissner plates 273, 278, 286 well-posedness of linear equations
edge conditions 279 of motion 105, 110
equations of motion 279, 289 Nodes 16
exact controllability 292 multiple and simple nodes 16
junction conditions geometric and dynamic node con-
geometric 274, 275, 289 ditions 16, 16, 17, 18, 20, 22,
dynamic 277, 280, 281, 284, 85, 87, 83, 141, 83, 85, 88,
285,289 141,96, 98, 100, 101, 101, 103,
observability estimates 293 104, 105
Nonlocal320
Membrane 201, 202 Numerical Schemes 167
edge conditions 204 algorithmic damping 167
equations of motion 203 dispersion 167
folded 242 finite elements 168, 183
kinetic energy 205 Lax-Friedrich scheme 168, 169
strain energy 205 Newmark scheme 168
Multiplier identities 30, 108 Runge-Kutta implicit scheme
168, 191
Networks of elastic strings 14, 99
derivation of non-linear equations Plates
of motion 15 edge conditions 256, 263, 269
derivation of linear equations of equations of motion 253, 263, 269
motion 22 kinetic energy 258
eigenvalue problems 142,143,147 nonlinear 252
exact controllability 33, 37, 40 rotations 264
kinetic energy 13, 17, 22 simply supported 259
nodes and node conditions 16, 17, strain energy 258
18, 22 Plate-beam systems 319, 372
388 INDEX
boundary conditions Shear modulus 304, 326

geometric 325, 330, 331, 348, Spectrum 135
374 frequencies 135
mechanical 325, 329, 331, 348, slicinl 135
374,376 splitting 152
equations of motion 329, 375 Stabilizability
junction conditions for networks of elastic strings 42
geometric 321, 323, 324, 330, for general networks of one-
372 dimensional elements 123, 128,
dynamic 325, 330, 377 129
limit model 363, 369 Strain tensor 51, 202, 253, 326
boundary conditions 370 Green - St. Venant 51
equations of motion 370 Stress-strain law 212, 266
junction conditions 370 Stress tensor
observability estimates 346 first Piola-Kirchhoff 62
reachable set 345 second Piola-Kirchhoff 60,203
Prandtl stress function 49 Stress vet:~Ol" 62, 203
St. Venant
Principle of virtual work 63, 205,
torsion 49
257,378
-Kirchhoff material 60
Reachability problem 222 Substructuring 135
Reachable states 228, 243, 247, 248
Temperature 60
Reference cross-section 202, 204, 206
Torsion 47,323
Reference surface 252, 273
St. Venant torsion 49,66
Repetitive Structures 135 Transmission problem
Reissner model 269, 279 for membranes 248, 249, 303
Reissner-Mindlin plate 269,323,326 for plates 292
Rotation Transverse shear angles 265
angles 265, 324 Twist 48,54
dextral 54
finite 56 Vector rotation angle 265, 270, 324
moderate 56 von Karman system 264
Serially connected Warping 50, 66,323

membranes 238,247,303 Work of external forces 15, 17, 22,
plates 276 97, 205, 209, 258
Systems & Control: Foundations & Applications
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.
We encourage the preparation of manuscripts in TeX, preferably in Plain or AMS TeX-

laTeX is also acceptable-for delivery as camera-ready hard copy which leads to rapid
publication, or on a diskette that can interface with laser printers or typesetters.
Proposals should be sent directly to the editor or to: Birkhiiuser Boston, 675 Massachusetts
Avenue, Cambridge, MA 02139, U.S.A.
Estimation Techniques for Distributed Parameter Systems

H. T. Banks and K. Kunisch
Set-Valued Analysis
Jean-Pierre Aubin and Helene Frankowska
Weak Convergence Methods and Singularly Perturbed

Stochastic Control and Filtering Problems
Harold J. Kushner
Methods of Algebraic Geometry in Control Theory: Part I

Scalar Linear Systems and Affine Algebraic Geometry
Peter Falb
H - Optimal Control and Related Minimax Design Problems

.
Tamer Basar and Pierre Bernhard
Identification and Stochastic Adaptive Control

Han-Fu Chen and Lei Guo
Viability Theory
Jean-Pierre Aubin
Representation and Control of Infinite Dimensional Systems, Vol. I

A. Bensoussan, G. Da Prato, M.e. Delfour and S.K. Mitter
Disease Dynamics
Alexander Asachenkov, Guri Marchuk, Ronald Mohler, Serge Zuev
Theory of Chattering Control with Applications to Astronautics,

Robotics, Economics, and Engineering
Michaill. Zelikin and Vladimir F. Borisov
Modeling, Analysis and Control of Dynamic Elastic

Multi-Link Structures
J. E. Lagnese, Gunter Leugering, E. J. P. G. Schmidt

J. E. Lagnese, Günter Leugering, E. J. P. G. Schmidt Auth. Modeling, Analysis and Control of Dynamic Elastic Multi-Link Structures

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

J. E. Lagnese, Günter Leugering, E. J. P. G. Schmidt Auth. Modeling, Analysis and Control of Dynamic Elastic Multi-Link Structures

Uploaded by

Copyright:

Available Formats

Systems & Control: Foundations & Applications

Christopher I. Byrnes, Washington University

S. - I. Amari, University of Tokyo

Modeling, Analysis and Control of

Springer Science+Business Media, LLC

ISBN 978-1-4612-6689-1 ISBN 978-1-4612-0273-8 (eBook)

Permission to photocopy for internal or personal use of specific clients is granted by

Typeset by the authors.

4. Straight, Untwisted, Nonshearable Nonlinear 3-d Beams 71

5. Straight, Untwisted Shearable Linear 3-d Beams 76

6. Shearable Nonlinear 2-d Beams with Curvature 77

9. Rotating Two-link Flexible Nonlinear Shearable Beams 89

Chapter IV. A General Hyperbolic Model for Networks 95

1. The General Model 95

2. Some Special Cases 99

3. Existence and Regularity of Solutions 105

4. Energy Estimates for Hyperbolic Systems 111

5. Exact Controllability of the Network Model. 119

6. Stabilizability of the Network Model. 123

Chapter V. Spectral Analysis and Numerical Simulations 135

2. Eigenvalue Problems for Networks of I-d Elements 142

5. Limit Model as the Shear Moduli Approach Infinity 363

V.1 Some eigenmodes of the unit square 151

V.30 Unit square: conservation of energy 192

The purpose of this monograph is threefold. First, mathematical models

modeling. As a result, the transmission and conversion of bending, torsion,

to modeling interconnected flexible bodies. This is the approach which

sider here mainly models obtained by linearization around an equilibrium of

ate application of the implicit function theorem, generically there is local

2. On the Contents of the Book

to find in the literature a treatment in which all of the features of interest

of Chapter II. Our principle goal is to give an explicit description of the

In particular, any generalization to 2-d elements would be of considerable

As in the case of beams, there are a number of distinct types of connec-

Modeling of Networks of Elastic Strings

1. Modeling of Nonlinear Elastic Strings

entiation with respect to x. The total energy is then

We apply Hamilton's principle to the Lagrangian functional

This requires C to be stationary at R. Of course the domain of C needs to

(1.3) R(O, t) = Uo(t)

(1.5) [1' [-pii.+h[R"-I~::ILl (x,t) 6R(x, t) dxdt

- h [R,x - 1:::1] (I, t) . 8R(l, t) = O.

(1.7) h [R,x - 1:::1] (I, t) = o.

The boundary conditions (1.3) and (1.7) need to be complemented by

(1.8) W(t) = 11 F(x, t) . R(x, t) dx + UI(t) . R(l, t).

Applying Hamilton's principle to the modified Lagrangian functional

(R) = 1T [K(t) + W(t) - U(t)] dt

as well as the inhomogeneous free boundary condition

More complicated boundary conditions are possible. Suppose for exam-

(1.12) h [R.x - 1:::1] (l,t) b = O.

2. Networks of Nonlinear Elastic Strings

For i E Ii we also introduce the parameter values Xii to be equal to Ii, if

(2.1) the Ri(Xii' t) are equal for all i E Ii, j E JM,

the common value of course being Ni(t).

Some nodes, indexed by j E .JD, may be subject to a Dirichlet condition

The work functional corresponding to distributed forces Fi(X, t) acting on

(2.6) W(t) = L 1 Fi(x, t) Ri(x, t) dx + L

Then the Lagrangian functional takes the form

- ~i [lR:", (x, t)l- 1]2] dx + L Ui(t). Ni(t)] dt.

We apply Hamilton's principle to the Lagrangian (2.7) with variations

ORi,(Xij,t) = oNj(t) for i E I j ,

(2.9) d~ C(R + AoR)I,x=o = {T

Integrating by parts with respect to both t and x, taking into account

+ L [- ~ fijhi [R:x - 1:~xll (Xij, t) + uj(t)] oNj(t)