Stability results for damped multilayer composite beams and plates

by
Aaron Andrew Allen
A dissertation submitted to the graduate faculty
in partial fulllment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
Major: Applied Mathematics
Program of Study Committee:
Scott W. Hansen, Major Professor
Paul Sacks
Elgin Johnston
Domenico DAlessandro
Steve Hou
Iowa State University
Ames, Iowa
2009
Copyright c _ Aaron Andrew Allen, 2009. All rights reserved.
ii
DEDICATION
First of all, I want to dedicate this thesis to my lovely wife Alison. Without her love and
support I would not have been able to complete this work. I would also like to dedicate this
thesis to my late father Roy W. Allen, who taught me that I could succeed regardless of my
circumstances. I would like to also thank my mother Karla J. Allen for all of her love and
encouragement through my long journey. Last, but not least, I give thanks, praise, and glory
to God for blessing me with the strength to persevere to this point of my life, and for sending
His son Jesus to pay the price for my sins.
iii
TABLE OF CONTENTS
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
CHAPTER 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Outline of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
CHAPTER 2. Basic Denitions and Theorems from Semigroup Theory . . 8
CHAPTER 3. Basic Beam and Plate Models . . . . . . . . . . . . . . . . . . 13
3.1 Basics of Beams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.1 Euler Bernoulli Beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.2 Rayleigh Beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.3 Timoshenko Beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Basic Plate Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.1 Kirchho Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.2 Mindlin-Timoshenko Plate . . . . . . . . . . . . . . . . . . . . . . . . . . 21
CHAPTER 4. Multi-layer Beams and Plates . . . . . . . . . . . . . . . . . . . 23
4.1 3-Layer Sandwich Beams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.1.1 3-Layer Rao-Nakra Beam . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.1.2 3-Layer Mead-Markus Beam . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2 3-Layer Damped beams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3 Multi-layer Beam Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3.1 Equations for the multilayer Rao-Nakra and Mead-Markus Beams . . . 29
4.4 Multi-layer Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
iv
4.4.1 Equations for the multilayer Rao-Nakra and Mead-Markus Plates . . . . 32
4.4.2 Korns Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.4.3 Continuity of

i
on (H
1
())
2
. . . . . . . . . . . . . . . . . . . . . . . . 35
CHAPTER 5. Analyticity, Exponential Stability, and Optimal Damping of
a Multilayer Mead-Markus Beam . . . . . . . . . . . . . . . . . . . . . . . . 37
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2 Semigroup Formulation: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.2.1 Semigroup Well-Posedness and Dissipativity . . . . . . . . . . . . . . . . 42
5.2.2 Semigroup Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.3 Analysis of the Semigroup Generator /
0
. . . . . . . . . . . . . . . . . . . . . . 46
5.3.1 Eigenvalues and Eigenfunctions of /
0
. . . . . . . . . . . . . . . . . . . 47
5.3.2 Riesz Basis Property for /
1
. . . . . . . . . . . . . . . . . . . . . . . . . 51
5.3.3 Analyticity of the semigroups generated by /
1
, /
0
, and / . . . . . . . . 53
5.3.4 Calculation of . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.4 Optimal Damping Angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.4.1 Preliminaries for proving Theorem 5.5 . . . . . . . . . . . . . . . . . . . 62
5.4.2 Properties for a critical point . . . . . . . . . . . . . . . . . . . . . . . . 62
5.4.3 Showing that the critical point is optimal . . . . . . . . . . . . . . . . . 66
5.4.4 Recovering the optimal damping coecients for analyticity . . . . . . . 69
5.5 Exponential Stability and Optimal Decay Rate . . . . . . . . . . . . . . . . . . 70
5.5.1 Exponential stability of the semigroup e
,t
. . . . . . . . . . . . . . . . . 70
5.5.2 Optimal damping for decay rate in a 3-Layer Mead-Markus Beam . . . 73
5.6 A 3-layer Mead-Markus Beam example . . . . . . . . . . . . . . . . . . . . . . . 77
5.6.1 Optimal damping coecient for angle of analyticity . . . . . . . . . . . 77
5.6.2 Optimal damping coecient for decay rate . . . . . . . . . . . . . . . . 78
5.7 Proof of Theorem 5.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
CHAPTER 6. Analyticity of a Multilayer Mead-Markus Plate: A Direct
Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
v
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6.2 Semigroup Formulation: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.2.1 Semigroup Well-Posedness . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.3 Analyticity of the Semigroup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.3.1 Eigenvalues of / . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.3.2 Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.3.3 A-priori estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.3.4 Main Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
CHAPTER 7. Exponential Stability of a Multilayer Rao-Nakra Beam . . . 111
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.2 Semigroup Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.2.1 Semigroup Well-Posedness . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3 Eigenvalues of / . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
7.3.1 Imaginary eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
7.3.2 Other eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
7.4 Exponential Stability of / . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
CHAPTER 8. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
8.1 Importance of the results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
8.2 Open problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
APPENDIX A. Analyticity of a Multilayer Mead-Markus Plate: An Indi-
rect Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
APPENDIX B. Proofs of various results in the thesis . . . . . . . . . . . . . 158
BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
ACKNOWLEDGEMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
vi
LIST OF FIGURES
Figure 1.1 A 3-layer sandwich beam . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Figure 1.2 Shear in viscoelastic material . . . . . . . . . . . . . . . . . . . . . . . 2
Figure 3.1 Normal cross-sections in an Euler-Bernoulli beam. . . . . . . . . . . . . . . 14
Figure 3.2 Transverse displacement w and bending angle . . . . . . . . . . . . . . . 15
Figure 3.3 Shear in a deformed Timoshenko beam. . . . . . . . . . . . . . . . . . . . 18
Figure 3.4 A Kirchho plate in equilibrium position. . . . . . . . . . . . . . . . . . . 19
Figure 3.5 A deformed Kirchho plate. . . . . . . . . . . . . . . . . . . . . . . . . . 20
Figure 3.6 A deformed Mindlin-Timoshenko plate. . . . . . . . . . . . . . . . . . . . 21
Figure 4.1 A 3-layer sandwich beam in equilibrium position. . . . . . . . . . . . . . . 23
Figure 4.2 Indexing of the layers in a sandwich beam. . . . . . . . . . . . . . . . . . 24
Figure 4.3 A multi-layer sandwich beam with m = 4. . . . . . . . . . . . . . . . . . . 28
Figure 4.4 A multi-layer sandwich plate with m = 2. . . . . . . . . . . . . . . . . . . 30
Figure 5.1 Eigenvalues of /
0
and of /. . . . . . . . . . . . . . . . . . . . . . . . . . 57
Figure 5.2 Negative eigenvalues of R when the
i
are distinct. . . . . . . . . . . . . . 60
Figure 5.3 Typical eigenvalues of / as varies. . . . . . . . . . . . . . . . . . . . . . 61
Figure 5.4 Eigenvalues of /
0
when

G
2
=

G

2
, and the sets

and

. . . . . . 78
Figure 5.5 The spectral bound for /

2
. . . . . . . . . . . . . . . . . . . . . . . . 79
Figure B.1 Directed graph of P. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
1
CHAPTER 1. Introduction
The vibrational properties of composite structures depend in a complex way upon the
elastic properties of the components making up the composite structure. Multilayer composite
beams and plates are commonly used in a wide variety of applications, from sporting goods to
aerospace engineering and in robotic arms and oor joists. When they are designed properly,
the favorable characteristics of the materials in each of the layers can be taken advantage
of. Standard beam and plate models are usually accurate enough for engineering applications
when the materials in each layer have similar elastic properties. On the other hand, when the
elastic properties vary greatly, the standard models (Euler-Bernoulli, Timoshenko, Kirchho,
Reissner-Mindlin, etc.) lead to poor predictions of vibrational properties.
In order to better understand the physics of a vibrating composite structure, much attention
has been focused on studying the simplest layered structure; the three-layer sandwich beam.
A sandwich beam typically consists of sti face plates and a compliant middle layer. Fig. 1.1
shows an example of such a beam.
Figure 1.1 A 3-layer sandwich beam
Some possible considerations in the design of such structures are to improve the strength-
to-mass ratios and improve vibration suppression through inclusion of damping. It has been
known for many years that the main source of damping of most sandwich structures is from the
2
shear in the compliant layer [43]. Shear is a phenomenon in which the normal cross-sections of
a viscoelastic material in the equilibrium state become oblique when the material is deformed
(see Fig. 1.2).
Figure 1.2 Shear in viscoelastic material
Typically, it is assumed that the layers are bonded together so that no slip occurs along the
interfaces of the beam layers. Thus, in-plane or bending motions in the outer layers force
the shearing motion in the core layer. In this dissertation, we study the eects that viscous
shear damping in the compliant layers has on the stability of composite beams and plates. We
also consider the problem of how to optimally damp exural vibrations in sandwich structures.
1.1 Background
Constrained-layer damping is a technique of designing a sandwich structure in order to
suppress vibrations. A typical constrained-layer damping application involves applying a layer
of compliant material onto a surface (of a base structure) and then adding a thin constraining
layer layer of material on top of the damping layer. Without the thin layer on top, the damping
material is unconstrained, and it deforms, mainly with stretching displacements, with the base
structure. Consequently, the damping that results comes mostly from the extension of the
viscoelastic material [43]. It has been found [45] that the promotion of shear greatly improves
the level of damping seen in the frequencies measured in experiments (this tends to be the
lower frequencies).
Perhaps the rst ones to analyze the eects of constrained-layer damping were Ross, Ungar,
3
and Kerwin in 1959. In [25] and [43], a three-layer system with a layer of damping tape bonded
to a face plate that was in turn constrained by a backing layer was considered. They measured
the shear damping by the energy-loss factor, which is based on a heuristic analogy to the
damped, forced spring, and is related to the energy loss per cycle of harmonic forcing. Their
analysis suggested that the primary source of damping comes from the shear motion in the
damped core layer. In addition, they reasoned, heuristically, that an optimal level of shear
damping exists in the core layer that leads to the best damping: If the damping in the core
layer is too small, little energy is lost, even though a large amount of shear may be induced in
the core layer; on the other hand, if the damping is too large, little energy is lost once again
because the core layer will resist much of the shearing motion. For years, the energy-loss factor
analysis done by Ross, Ungar, and Kerwin [43] (sometimes referred to as the RKU analysis; see
[38] and [45]) has been a widely used method for describing the behavior of damping treatments
because it has been veried in both experiment and in a number of mathematical models for
sandwich beams and plates. However, the loss factor approach is limited to harmonic motions
with harmonic forcing and hence is incompatible with a general PDE description of damping.
We do not pursue these methods in this thesis.
Ever since the late 1950s, a number of sandwich beam and plate models have been pro-
posed. In 1965, DiTaranto [8] derived a sixth-order dierential equation for a freely vibrating
sandwich beam that has no boundary conditions. The dierential equation he derived is in
terms of the longitudinal displacement of one of the face plates. In 1972, Yan and Dowell [47]
derived a set of ve partial dierential equations that govern the motions of vibrating damped
sandwich beams and plates. Two of the more widely accepted models (due to their simplicity
and good agreement with experiments) are those proposed by Mead and Markus [37], and Rao
and Nakra [41].
The three-layer sandwich beam of D. J. Mead and S. Markus [37] (derived in 1969) consists
of two sti outer layers and a compliant core layer that is elastic with respect to shear (similar
in structure to the one shown in Figure 1). The three layers are assumed to be bonded together
so that no slip occurs between the beam layers. They derived a sixth-order partial dierential
4
equation in terms of the transverse displacement (see Chapter 4 for an equivalent equation)
for such a beam.
The three-layer sandwich beam of Y.V.K.S. Rao and B.C. Nakra [41] was derived in 1974
under assumptions similar to what Mead and Markus used. The main dierence is that the ef-
fects of longitudinal and rotational momentum (which are ignored in the Mead-Markus model)
were considered. Rao and Nakra also did a spectral analysis in comparing their model to the
Mead-Markus model. Their analysis indicated certain conditions (in terms of thicknesses, sti-
nesses, frequencies, etc.) under which it is necessary to include the longitudinal and rotational
inertia terms in the model. Moreover, they deduced similar conclusions in the same paper for
a three-layer sandwich plate.
Many composite structures consist of more than three layers of varying stinesses. Hence,
it is natural to consider a multilayer sandwich structure based on alternating layers of sti
and compliant materials (more details will be discussed in Chapter 4). Over the past 30 years,
a great deal of material has been published with regard to models for multilayer beams and
plates. Some examples include Nashif, Jones, and Henderson [38]; and some articles published
by S.W. Hansen (see e.g., [10], [12], and [16]).
Much of the research on the topic of sandwich structures was conducted by engineers up
until the late 1990s. Consequently, existence, uniqueness, and well-posedness issues for the
existing sandwich plate and beam models had not been addressed, and modern PDE methods
had not been applied to any of these models. Then in the late 1990s and early 2000s,
S.W. Hansen established the well-posedness of composite beam and plate models using the
variational method (see e.g., [9], [10], [12], [13], and [16]) and semigroup theory (see e.g., [17],
[15], and [21]). In this thesis, we will look at multilayer composite beam and plate models in
the framework of semigroup theory.
In the semigroup formulation, the equations of motion for a sandwich beam or plate struc-
ture are written:
d
dt
x = /x, (1.1)
where / is a densely dened operator on Hilbert space H, and x H is the state of the
5
system. (see e.g., [14], [17], [18], and [20] for beams; and [1] and [15] for plates). It can often
be shown that the operator / is a generator of a C
0
-semigroup (see Chapter 2 for further
denitions). Then well-posedness of the problem follows from semigroup theory.
In some of the papers mentioned above, some promising results have been proven with
regard to the stability of three-layer damped sandwich structures. In Hansen and Lasiecka
[17], for example, it has been proven that the solutions corresponding to a three-layer sandwich
beam with a thin core (which is actually a special case of the three-layer Mead-Markus beam)
are analytic and exponentially stable. The same analysis works just as well for the general
three-layer Mead-Markus beam. Hansen and Fabiano [9] described explicitly how to nd the
best choice of damping in a three-layer Mead-Markus beam. On the other hand, the three-layer
Rao-Nakra beam has been proven to be exponentially stable in [20], but not analytic.
Up until now, much of the work done with regard to applying modern PDE methods to
multilayer composite beam and plate structures includes (i) establishing the well-posedness of
multilayer models and (ii) proving stability results for three-layer sandwich structures using
semigroup theory. The results we establish in this thesis are summarized as follows:
1. We prove analyticity and uniform exponential stability of the semigroup associated with
the multilayer Mead-Markus beam. More precisely, we formulate the equations of motions
in the form (1.1), and show that the operator / is the generator of an analytic semigroup.
To do this, we view / as a bounded perturbation of another semigroup generator /
0
.
We then show that the eigenfunctions and generalized eigenfunctions of /
0
form a Riesz
Basis. This can be used to prove that the semigroup generated by /
0
is analytic. A
perturbation theorem leads us to the conclusion that the semigroup generated by /
is analytic. We then prove the exponential stability of the semigroup associated with
the multilayer Mead-Markus beam. This follows from the fact that analytic semigroups
satisfy a spectrum-determined growth condition (see Chapter 2 for more information).
2. Next, we solve the problem of how to optimize the angle of analyticity of the semigroup
associated with the multilayer Mead-Markus beam with respect to the damping coe-
cients. We prove that either the system becomes over-damped, or there exists a unique
6
optimum for which we are able to compute the optimal damping coecients. We also
consider the problem of optimizing the decay rate of the semigroup. In the three-layer
case, we prove an explicit formula for computing the optimal damping coecients.
3. We prove the analyticity of the semigroup associated with a multilayer Mead-Markus
plate. Once again, we use a semigroup formulation. In comparison to the (one-dimensional)
beam problem, the (two-dimensional) plate theory is signicantly more complex since pla-
nar elasticity involves the Lame operator, which reduces to
d
2
dx
2
in the case of a beam.
Two dierent proofs of analyticity are presented in this thesis. In Chapter 6, we use a
direct proof of analyticity that involves the use of numerous estimates including Sobolev
inequalities and application of Korns inequality. An indirect proof (proof by contradic-
tion) of the same result is in Appendix A. It is a draft of the publication Analyticity
of a multilayer Mead-Markus plate, [1] which will appear in the Journal of Nonlinear
Analysis.
4. We also prove that the semigroup associated with a multilayer Rao-Nakra beam may
or may not be exponentially stable, and it is not analytic. Again we use a semigroup
formulation, and we establish conditions for the wave speeds so that the semigroup
generator / has no spectrum on the imaginary axis. We then apply an indirect argument
for proving that the semigroup generated by / is exponentially stable if and only if /
has no spectrum on the imaginary axis.
1.2 Outline of the Thesis
The remainder of the thesis is organized as follows. The next three chapters are devoted
to providing preliminary material. In Chapter 2 we give some useful denitions and theorems
from the basic theory of semigroups. We will use this theory to obtain many of the results
in this thesis. In Chapter 3, we discuss basic beam and plate models. This section will focus
on the physical assumptions of the individual layers. Chapter 4 builds on the discussion in
Chapter 3 by looking at examples of composite sandwich beams. In addition, we mention
7
and prove Korns inequality, which will be a valuable tool in analyzing the semigroups arising
from composite plate models. The next three chapters discuss the main results; Chapter 5 is
summarized by items 1 and 2, Chapter 6 is mentioned in item 3, and Chapter 7 is summarized
by item 4. In Chapter 8, we conclude with a discussion of open problems and directions for
future research. Furthermore, we include two appendices. Appendix A contains a draft of the
publication Analyticity of a multilayer Mead-Markus plate, [1] which was recently submitted
and accepted into the Journal of Nonlinear Analysis. It considers the same problem that
Chapter 6 does, but with additional boundary conditions. Finally, Appendix B contains proofs
of some formulas and facts used throughout thesis. It also contains detailed proofs of some
assertions that are not included in the main portion of the thesis for the sake of clarity and
readability.
8
CHAPTER 2. Basic Denitions and Theorems from Semigroup Theory
Throughout this thesis, we will formulate the dierential equations arising from beam and
plate models into a problem of the form (1.1). We then use ideas from the theory of semigroups
to analyze these dierential equations. In this chapter, we give some denitions and theorems
from basic semigroup theory. The books by A. Pazy [39]; Z. Liu and S. Zheng [35]; and Z.H.
Luo, B.Z. Guo, and O. Morg ul [36] are excellent resources on this subject.
Denition 2.1. Let X be a Banach Space. We say that a family of bounded operators on
X, T (t)
t0
is a semigroup of bounded operators, or a semigroup on X if the following two
conditions hold:
i.) T (0) = I, where I is the identity operator on X.
ii.) T (t +s) = T (t)T (s); s 0, t 0.
Denition 2.2. Suppose T (t)
t0
is a semigroup on a Banach Space X. If
lim
t0
|T (t)x x|
X
= 0, for all x X,
then T (t)
t0
is a C
0
-semigroup on X.
Denition 2.3. Let T (t)
t0
be a C
0
-semigroup on X. Dene the operator / as follows:
/x := lim
t0
T (t)x x
t
, x T(/), where
T(/) :=
_
x X : lim
t0
T (t)x x
t
X
_
.
Then / is called the innitesimal generator of the group T (t).
Remark 2.1. T (t) is sometimes denoted e
,t
.
9
As mentioned before, a common technique for analyzing beam and plate equations is to recast
them into a semigroup problem (1.1) and show that the associated operator generates a C
0
semigroup. The next theorem is important because it states that if we can formulate a problem
into the form (1.1) and show that / generates a C
0
semigroup, then the problem is well-posed
(see Theorem 2.64 in [36]).
Theorem 2.1. Let / be a densely dened linear operator in a Banach Space X with a non-
empty resolvent. The Cauchy problem (1.1) has a unique solution for all x T(/) which is
continuously dierentiable for t 0 if and only if / generates a C
0
-semigroup T (t) on X.
In many situations, it will be useful to consider a special class of C
0
semigroups.
Denition 2.4. Let T (t)
t0
be a C
0
-semigroup on X. T (t)
t0
is a C
0
semigroup of
contractions on X if
|T (t)|
/(X)
1.
Since we will be incorporating damping into the beam and plate models, we expect that energy
will dissipate over time. Hence, it will be useful to dene what it means for an operator to be
dissipative on a Hilbert space. The following denition is taken from Liu and Zheng [35].
Denition 2.5. Let H be a real or complex Hilbert space equipped with the inner product , )
and the induced norm | |. The operator / is dissipative on H if for all x T(/),
i.) / is densely dened on H, and
ii.) /x, x)
1
0.
A commonly used theorem to prove that an operator / is the generator of a C
0
-semigroup of
contractions is to use the well-known L umer-Phillips Theorem (see Theorem 2.7 in [36]).
Theorem 2.2. (L umer-Phillips) Let / be a linear operator on a Hilbert space H. Then /
generates a C
0
semigroup of contractions on H if and only if
i.) The domain of /, T(/), dense on H .
ii.) / is dissipative on H and R(
0
, /) = H for some
0
> 0.
10
In the above, R(
0
, /) denotes the resolvent operator (
0
I /)
1
.
The following corollary to the L umer-Phillips theorem will also be a useful way to determine
if / generates a C
0
semigroup of contractions (See Corollary 4.4 on p.15 of [39]).
Theorem 2.3. Let / be a densely dened, closed linear operator on a Hilbert Space H. Then
/ generates a C
0
semigroup of contractions on H if and only if both / and /

are dissipative
on H.
In the above theorem, /

denotes the adjoint operator of / on H.

Next, we consider analytic semigroups. So far, the semigroups we have mentioned T (t) have
a domain on the real nonnegative axis. With analytic semigroups, we will consider domains
that are sectors containing the nonnegative real axis.
Denition 2.6. Let X be a Banach space, and suppose /2
1
< 0 <
2
/2. Dene a
sector
(
1
,
2
)
by
(
1
,
2
)
:= z C :
1
< arg z <
2
, and let T (z) be a bounded operator
for all z
(
1
,
2
)
. The family of operators T (z)
z
(
1
,
2
)
is called an analytic semigroup
on
(
1
,
2
)
if the following holds:
i.) T (z) is analytic in
(
1
,
2
)
ii.) T (0) = I and lim
z0
= T (z)x = x, for all x X
iii.) T (z
1
+z
2
) = T (z
1
)T (z
2
), for all z
1
and z
2
in
(
1
,
2
)
.
Remark 2.2. If there exists some sector
(
1
,
2
)
in the complex plane on which T (z) is
analytic, then T (z) is said to be an analytic semigroup on X.
There are several useful theorems for proving that an operator / is the generator of an analytic
semigroup. In the next theorem, it will be useful to dene the following:

:=
(,)
= z C : [ arg z[ < ,

:= z C : [ arg z[ < +/2.

In the above denitions, 0 < < /2. The following theorem is from Pazy [39].
Theorem 2.4. Let T (t) be a uniformly bounded C
0
semigroup on a Banach space X generated
by /. Furthermore, assume 0 (/). Then the following are equivalent:
11
(i.) T (t) can be extended to an analytic semigroup in

, and |T (t)| is uniformly bounded

in every closed sub-sector

of

, where
t
< .
(ii.) There exists a constant C such that for every > 0, ,= 0 such that
|R( +i, /)| C/[[.
(iii.) There exist 0 < < /2 and M > 0 such that

0 (/) and
|R(, /)| M/[[, for

, ,= 0.
In Theorem 2.4, (/) denotes the resolvent set of /.
Not only does analyticity give us regularity results of the solution, but it also gives us valuable
information about the spectrum of the semigroup generator /.
Denition 2.7. Let / be the generator of a C
0
-semigroup T (t). The growth rate of a semi-
group T (t) is dened as follows:
(/) := lim
z0
ln|T (t)|
t
.
Denition 2.8. The spectral bound of an operator / is dened as follows:
S(/) := supRe : (/),
where (/) denotes the spectrum of /.
For any C
0
-semigroup we have S(/) (/). However, this inequality can be strict (see
Renardys counter-example [42], [36]). In many situations, it turns out that the spectral
bound and the growth rate are the same.
Denition 2.9. If (/) = S(/), then the semigroup T (z) = e
,z
satises the spectrum-
determined growth condition.
There is an important relationship between an analytic semigroup and the spectrum-determined
growth condition, which is summarized in the following theorem (see Corollary 3.14 in [36]).
12
Theorem 2.5. If / is the generator of an analytic semigroup, then the spectrum-determined
growth condition holds.
Therefore, knowing that / generates an analytic semigroup is valuable because one can deter-
mine the growth rate of the semigroup e
t,
just by looking at the spectrum of /. In addition,
the domain of / is often compactly dened. Since the spectrum of such operators consists
only of eigenvalues, we can deduce the growth rate simply by investigating the eigenvalues of
/ in this case.
Another important stability result we consider in this thesis is exponential stability.
Denition 2.10. The semigroup T (t) is exponentially stable if there exists constants M 1
and > 0 such that
|T(t)| Me
t
.
The following is a very useful theorem for proving exponential stability (see [40] and Corollary
3.36 in [36]).
Theorem 2.6. If e
,t
= T (t) is a uniformly bounded C
0
-semigroup on H, then T(t) is expo-
nentially stable if and only if (/) contains the imaginary axis and
sup
R
_
_
(iI /)
1
_
_
B(1)
< +.
Another useful theorem is due to F. Huang [23].
Theorem 2.7. Let e
,t
= T (t) be a C
0
-semigroup on H. T (t) is exponentially stable if and
only if
(i) supRe : (/) < 0, and
(ii) sup
Re 0
|(I /)
1
| < .
13
CHAPTER 3. Basic Beam and Plate Models
Before we derive the multi-layer beam and plate models, we look at the modeling assump-
tions of the individual layers.
3.1 Basics of Beams
We rst consider a beam of length L, width R, and thickness h in rest position on the
x
1
-x
2
-x
3
right-handed, three-dimensional coordinate system. Hence, if we dene Q = (0, L)
(0, R) (0, h), then the beam occupies the region Q at equilibrium. Also, the horizontal
midplane has equation x
3
= h/2 at equilibrium. When the beam is bent, or deformed, we must
consider the displacements in each direction and the rotation of the normal cross-sections of
the beam. In typical sandwich beam structures, the sti layers are either Euler-Bernoulli or
Rayleigh beams, while the compliant layers follow the Timoshenko beam theory. An excellent
explanation regarding the modeling of these three types of beams can be found in an article
done by D.L. Russell [44]. Some of the derivation of these beam models included in this section
is borrowed from his work.
3.1.1 Euler Bernoulli Beam
We discuss a way to obtain the Euler-Bernoulli equations. According to Euler-Bernoulli
beam theory, the following assumptions hold:
(i) Normal cross-sections remain straight.
(ii) Normal cross-sections stay the same length.
(iii) Normal cross-sections remain normal.
14
Figure 3.1 illustrates assumptions (i) - (iii).
Figure 3.1 Normal cross-sections in an Euler-Bernoulli beam.
In addition, we make the following further assumptions:
(iv) Longitudinal displacements (displacements in the x
1
-direction) vary linearly with respect
to x
3
, the transverse coordinate.
(v) The transverse displacement (displacement in the x
3
-direction) is constant with respect
to x
3
, the transverse coordinate.
(vi) All displacements and deformations are independent of the x
2
direction.
To dene some of the quantities, it will be convenient to refer to Figure 3.2. Let w denote
the transverse displacement. Assumptions (v) and (vi) imply that w depends only on x
1
and
t. It will be convenient to let the variable x denote x
1
in this context; hence, w = w(x, t) with
this notation. In the equilibrium state, the beam is parallel to the line x
3
= 0. In the deformed
state, the beam is parallel to the curve x
3
= w(x, t). Moreover, the normal cross-sections of
the beam at each x (0, L) are rotated by a bending angle of
(x, t) = tan
1
(w
x
(x, t)).
In a linear theory, displacements are assumed to be small (i.e w(x, t) is small). Thus
(x, t) w
x
(x, t). (3.1)
15
Figure 3.2 Transverse displacement w and bending angle .
The total energy of the beam is given by
E(t) = /(t) +T(t), (3.2)
where /(t) and T(t) denote the kinetic and potential energy respectively and are given by
/(t) =
1
2
_
L
0
m[ w[
2
dx and T(t) =
1
2
_
L
0
K[w
xx
[
2
dx.
In the above, the dots denote dierentiation with respect to t, m is the mass per unit length,
and K is the rigidity constant. Using variational methods and (3.2), one can derive the Euler-
Bernoulli beam equations. We dene the Lagrangian L on (0, T) by
L(w) =
_
T
0
[/(t) T(t)] dt. (3.3)
According to the principle of virtual work, the solution trajectory is the trajectory which
renders the Lagrangian stationary under all kinematically admissible displacements. Let w
denote a test function on (0, L) (0, T) that is compactly supported on (0, T). Using (3.3),
one can compute
lim
0
L(w + w) L(w)

=
_
T
0
m w,

w)
L
2
(0,L)
dt
_
T
0
Kw
xx
, w
xx
)
L
2
(0,L)
dt.
16
Setting the above expression equal to zero, we obtain the following variational equation:
m w, w)
L
2
(0,L)
+Kw
xx
, w
xx
)
L
2
(0,L)
= 0. (3.4)
Applying integration by parts in (3.4), one obtains
m w +Kw
xxxx
, w)
L
2
(0,L)
Kw
xxx
w

L
0
+Kw
xx
w
x

L
0
= 0.
Many boundary conditions may be considered in order to annihilate the boundary terms in
the above expression. If we use clamped boundary conditions; that is
w(0, t) = w(L, t) = w
x
(0, t) = w
x
(L, t) = 0, (3.5)
and impose the same boundary conditions on w, we obtain the following equations of motion:
_

_
m w +Kw
xxxx
= 0 x (0, L), t > 0
w(x, t) = w
x
(x, t) = 0 x = 0, L, t > 0
. (3.6)
Another boundary condition that can be applied is the simply-supported (or hinged) boundary
conditions:
w(0, t) = w(L, t) = w
xx
(0, t) = w
xx
(L, t) = 0, t > 0. (3.7)
We can also formulate (3.6) as a semigroup problem of the form (1.1). If we let y denote
w and let
_

_
w
y
_

_ be the state vector, then H = H

2
(0, L) L
2
(0, L) is an appropriate choice
for the state space. We then rewrite (3.6) as follows:
d
dt
_

_
w
y
_

_ = /
_

_
w
y
_

_, where / =
_

_
0 1

K
m
D
4
x
0
_

_, (3.8)
and D
k
x
denotes

k
x
k
. The domain of / is dened to be
T(/) = [w, y]
T
H
4
(0, L) H
2
(0, L) + BCs,
where the +BCs refers to the boundary conditions that are satised by the image of /. For
instance, the clamped boundary conditions would impose the following conditions on y:
y(0, t) = y(L, t) = y
x
(0, t) = y
x
(L, t) = 0, t > 0.
17
3.1.2 Rayleigh Beam
The Rayleigh beam is similar to the Euler-Bernoulli beam, with the exception that a mass
moment of inertia term is added. The total energy of the beam is given by
E(t) =
1
2
_
L
0
_
m[ w[
2
+[ w
x
[
2
+K[w
xx
[
2

dx,
where is the moment of inertia constant. Using a variational formulation similar to the one
done for the Euler-Bernoulli beam, one arrives at the following equations of motion for the
Rayleigh beam:
m w w
xx
+Kw
xxxx
= 0 x (0, L), t > 0 (3.9)
The hinged and clamped BCs ((3.7) and (3.5) respectively) can be applied to this model.
Using the same state variables as before and letting H = H
2
(0, L) H
1
(0, L) be the state
space, we can formulate the Rayleigh beam as a semigroup problem. Solving (3.9) for w,
however, requires that we dene an operator J : H
1
(0, L) H
1
(0, L). by
Jv = mv v
xx
.
One can show that J is coercive and invertible using the Lax-Milgram lemma. Therefore we
have
w = J
1
(Kw
xxxx
) .
We then formulate (3.9) as an abstract Cauchy problem of the form (1.1) as follows:
d
dt
_

_
w
w
_

_ = /
_

_
w
w
_

_ where / =
_

_
0 1
J
1
_
KD
4
x
_
0
_

_,
where the domain of / is dened to be
T(/) = [w, w]
T
H
3
(0, L) H
2
(0, L) + BCs,
and +BCs has the same meaning as before.
18
3.1.3 Timoshenko Beam
The Timoshenko beam assumptions are similar to the Euler-Bernoulli beam assumptions,
with the exception that assumption (iii) is dropped. This means that the cross-sections which
are normal in the rest state are allowed to become oblique in the deformed state, thus intro-
ducing shear.
Figure 3.3 Shear in a deformed Timoshenko beam.
In the above gure, represents the angle of the shear, and is the bending angle. Using small
displacement theory, the total rotation angle is approximately (x, t) w
x
(x, t) according to
(3.1). The equations of motion for such a beam is given in the following coupled PDE (see p.
263 in [36]):
_

_
m w +hG(
x
w
xx
) = 0 x (0, L), t > 0

K
xx
+hG( w
x
) = 0 x (0, L), t > 0
(3.10)
In the above, hG is the shear modulus of elasticity, and the constants m, , h, and K are
dened as before.
Remark 3.1. It is interesting to note that as the modulus of elasticity becomes large (i.e.
hG ), the Timoshenko beam equations resemble the Rayleigh beam equations. To see this,
we dierentiate the second line of (3.10) with respect to x then subtract it from the rst line
to obtain
m w

x
+K
xxx
= 0. (3.11)
19
Solving the rst line of (3.10) for
x
and substituting into (3.11) leads to
m w
_
w
xx

m
hG
....
w
_
+K
_
w
xxxx

m
hG
w
xx
_
= 0. (3.12)
As hG , (3.12) approaches (3.9).
3.2 Basic Plate Models
The theory of beams can easily be extended to the theory of plates. The main dierence
is that we will now need to consider bending and rotations in two directions when a plate is
in the deformed state.
3.2.1 Kirchho Plate
The sti layers in the plate models will follow Kirchho plate theory. The Kirchho assump-
tions are analogous to the Euler-Bernoulli assumptions for beams. Suppose is a bounded
domain in the x
2
-x
3
plane with a smooth boundary . We consider a plate of thickness h and
assume that it occupies the region Q = (0, h) in rest position.
Figure 3.4 A Kirchho plate in equilibrium position.
In addition, the midplane to the plate has equation x
3
= h/2 at equilibrium. The assumption
that the transverse displacement is constant with respect to the x
3
coordinate (analogous to
assumption (v) for the beam) is still valid. With a deformed plate, however, we must now
20
consider the longitudinal displacements and rotations that occur in the x
2
direction. Hence
the transverse displacement w depends on x
1
and x
2
; that is, w = w(x
1
, x
2
, t) . It will be
convenient to denote x = (x
1
, x
2
) in this context. Let
1
denote the bending rotation in the
x
1
direction and
2
denote the bending rotation in the x
2
direction. Using small displacement
theory (see (3.1)),

1

w
x
1
and
2

w
x
2
.
Figure 3.5 A deformed Kirchho plate.
Under the Kirchho assumptions, the total energy of a Kirchho plate is given by
E(t) =
_

_
m[ w[
2
+[ w[
2
+K[w[
2

dx. (3.13)
Using (3.13), one can derive the following equation of motion for the Kirchho plate:
m w w +K
2
w = 0, x , t > 0. (3.14)
In general, boundary conditions for a plate are much more complicated to describe. However,
the clamped and simply-supported BCs are fairly simple to describe (especially if we impose
the same conditions on all of ) Let n = (n
1
, n
2
) denote the outward unit normal to . The
clamped boundary conditions are of the form
w =
w
n
= 0 on . (3.15)
21
To describe the simply-supported boundary conditions, we dene a boundary operator B by
Bw = 2n
1
n
2

2
w
x
1
x
2
n
2
1

2
w
x
2
2
n
2
2

2
w
x
2
1
.
The simply-supported boundary conditions take the form
w = 0, w + (1 )Bw = 0 on , (3.16)
where (0, 1/2) denotes Poissons ratio for the plate [26].
3.2.2 Mindlin-Timoshenko Plate
The compliant layers in our multi-layer plate model will obey the Mindlin-Timoshenko
assumptions, which are analogous to the Timoshenko assumptions for a beam. They are
obtained by taking the same hypothesis as the Kirchho plate and removing the assumption
that the normal cross-sections remain normal under deformation. Since shear will occur in
two directions, we will assume is a function of x
1
and x
2
with 2 components; i.e. (x) =

1
(x
1
, x
2
),
2
(x
1
, x
2
).
Figure 3.6 A deformed Mindlin-Timoshenko plate.
22
The energy of the Mindlin-Timoshenko plate is
E(t) =
1
2
_

_
m[ w[
2
+[

1
+

2
[
2
_
dx +
1
2
_

hG
_

w
x
1
+
1

2
+

w
x
2
+
2

2
_
dx
+
1
2
_

K
_

1
x
1

2
+

2
x
2

2
+ 2

1
x
1

2
x
2
+
1
2

1
x
2
+

2
x
1

2
_
dx.
The equations of motion for a Mindlin-Timoshenko plate are of the following form (see p. 15
in Lagnese and Lions [26]):
_

1
K
_

1
x
2
1
+
1
2

1
x
2
2
+
1 +
2

2
x
1
x
2
_
+hG
_

1
+
w
x
1
_
= 0

2
K
_

2
x
2
2
+
1
2

2
x
2
1
+
1 +
2

1
x
1
x
2
_
+hG
_

2
+
w
x
2
_
= 0
m w hG
_

x
1
_

1
+
w
x
1
_
+

x
2
_

2
+
w
x
2
__
= 0
(3.17)
The clamped boundary conditions take the same form as before (see (3.15)). The simply-
supported boundary conditions are as follows:
w = 0 on ,
_

1
x
1
+

2
x
2
_
n
1
+
_
1
2
__

1
x
2
+

2
x
1
_
n
2
= 0 on ,
_

i
2
x
2
+

1
x
1
_
n
2
+
_
1
2
__

2
x
1
+

1
x
2
_
n
1
= 0 on .
Remark 3.2. Using a method similar to the one used in Remark 3.1, one can see that the
Mindlin-Timoshenko plate equations approach the Kirchho plate equation as the modulus of
elasticity approaches innity. See pp 16-17 in [26] for further details.
23
CHAPTER 4. Multi-layer Beams and Plates
4.1 3-Layer Sandwich Beams
The classical sandwich beam models of Rao-Nakra [41] and Mead-Markus [37] consist of
two outer beam layers of sti material that sandwich a exible core layer. We will make the
following physical assumptions:
(i) No slip occurs along the interfaces
(ii) The central beam layer allows shear (Timoshenko assumptions)
(iii) The outer beam layers do not shear (Euler-Bernoulli assumptions)
We will assume that the sandwich beam occupies the region Q = (0, h) at equilibrium,
where = (0, L) (0, R) is a rectangle of length L and width R.
Figure 4.1 A 3-layer sandwich beam in equilibrium position.
The layers will be indexed from bottom to top. Hence we dene the thickness of the i
th
layer
by
h
i
= z
i
z
i1
, i = 1, 2, 3,
24
where 0 = z
0
< z
1
< z
2
< z
3
= h, and the planes x
3
= z
i
denote the interfaces of each beam
for i = 0, 1, 2, 3.
Figure 4.2 Indexing of the layers in a sandwich beam.
In this context, the spatial variable x will denote the longitudinal direction x
1
.
4.1.1 3-Layer Rao-Nakra Beam
Let
i
=
i
(x, t) denote the shear of the i
th
layer. Since the outer layers have no shear,

i
= 0 for i odd. In addition, let v
i
= v
i
(x, t) denote the longitudinal displacement of the i
th
layer. We dene the following constants:

i
denotes the mass density per unit volume in the i
th
layer,
E
i
denotes the longitudinal Youngs modulus in the i
th
layer,
G
i
denotes the transverse shear modulus in the i
th
layer.
D
i
:=
E
i
12(1
2
i
)
The quantity D
i
h
3
i
is known as the modulus of exural rigidity, and
i
is the in-plane Poissons
ratio in the i
th
layer (0 <
i
< 1/2). Since the outer layers have no shear, G
1
= G
3
=
0. Otherwise, each of the constants dened above are positive. In addition, dene N =
h
1
+ 2h
2
+h
3
2h
2
. According to our assumptions, it is possible to relate the shear in the middle
layer to the longitudinal displacement in the odd layers; namely
v
3
v
1
= h
2

2
h
2
Nw
x
. (4.1)
25
We now derive the total energy and the equations of motion. If and are matrices in R
lm
,
we denote the scalar product in R
lm
by . We also dene the following inner products:
, )

=
_

dx, , )

=
_

d. (4.2)
The total energy (see [13] and [21]) is given by
E(t) =
R
2
_
m w, w)

+ w
x
, w
x
)

+h
1

1
v
1
, v
1
)

+h
3

3
v
3
, v
3
)

+
R
2
_
Kw
xx
, w
xx
)

+h
1

1
v
1
x
, v
1
x
)

+h
3

3
v
3
x
, v
3
x
)

+h
2
G
2

2
,
2
)

. (4.3)
In the above, m is the mass density per unit length, is the moment of inertia parameter, and
K is the bending stiness. They are dened as follows:
m = h
1

1
+h
2

2
+h
3

3
, =
1
12
_

1
h
3
1
+
3
h
3
3
_
, K =
1
12
_
E
1
h
3
1
+E
3
h
3
3
_
. (4.4)
This leads to the equations of motion for the 3-Layer Rao-Nakra beam: (as formulated in
Hansens Optimal Damping Paper [13])
_

_
m w D
2
x
w +KD
4
x
w D
x
Nh
2
G
2

2
= 0 on (0, L) (0, )
h
1

1
v
1
h
1
E
1
D
2
x
v
1
G
2

2
= 0 on (0, L) (0, )
h
3

3
v
3
h
3
E
3
D
2
x
v
3
+G
2

2
= 0 on (0, L) (0, )
(4.5)
4.1.2 3-Layer Mead-Markus Beam
The Mead-Markus model ignores longitudinal and rotational momentum. Hence, we can
obtain it by simply omitting terms involving D
2
x
w and v
O
from (4.5). Thus, the energy for the
three-layer Mead-Markus beam is
E(t) =
R
2
_
m w, w)

+Kw
xx
, w
xx
)

+h
1

1
v
1
x
, v
1
x
)

+h
3

3
v
3
x
, v
3
x
)

+h
2
G
2

2
,
2
)

,
(4.6)
and the equations of motion are
_

_
m w +KD
4
x
w D
x
Nh
2
G
2

2
= 0 on (0, L) (0, )
h
1
E
1
D
2
x
v
1
G
2

2
= 0 on (0, L) (0, )
h
3
E
3
D
2
x
v
3
+G
2

2
= 0 on (0, L) (0, )
(4.7)
26
Remark 4.1. The last two lines of (4.5) and (4.7) can be written in a more concise way.
Dene the following:
h
O
=
_

_
h
1
0
0 h
3
_

_, p
O
=
_

1
0
0
3
_

_, E
O
=
_

_
E
1
0
0 E
3
_

_,
v
O
=
_

_
v
1
v
3
_

_, B = [1, 1].
With this notation, the last two equations of (4.5) read
h
O
p
O
v
O
h
O
E
O
D
2
x
v
O
+B
T
G
2

2
= 0 on (0, L) (0, ), (4.8)
and the last two equations of (4.7) read
h
O
E
O
D
2
x
v
O
+B
T
G
2

2
= 0 on (0, L) (0, ). (4.9)
Remark 4.2. If we multiply (4.9) by B(h
O
E
O
)
1
, we can write (4.7) as follows:
_

_
m w +KD
4
x
w D
x
Nh
2
G
2

2
= 0 on (0, L) (0, )
Bv
O
+PG
2

2
= 0 on (0, L) (0, )
, (4.10)
where P = B(h
O
E
O
)
1
B
T
is a positive number. Notice that (4.1) reads
Bv
O
= h
2

2
h
2
Nw
x
.
Using this, along with the change in variables
C = h
2
P, H = h
2
N, = G
2
/h
2
, s = h
2

2
,
(4.10) can be written as follows (compare with equations (15) and (16) in [9]):
_

_
m w +Kw
xxxx
Hs
x
= 0 on (0, L) (0, )
Cs s
xx
+Hw
xxx
= 0 on (0, L) (0, )
This implies
_

_
s
x
=
m
H
w +
K
H
w
xxxx
Cs
x
s
xxx
+Hw
xxxx
= 0
(4.11)
27
Substituting the rst line of (4.11) into the second yields the equation
[m w +Kw
xxxx
]
xx
C
_
m w +
_
K +
H
2
C
_
w
xxxx
_
= 0.
This equation is equivalent to the sixth order equation obtained by Mead and Markus [37].
4.2 3-Layer Damped beams
In this thesis, we are interested in studying the eects of shear damping in the compliant
layer(s). If we use strain-rate proportional damping, we replace G
2

2
by G
2

2
+

G
2

2
according
to the viscoelastic constitutive law. Here,

G
2
> 0 denotes the coecient for damping in the
core layer. Using (4.5), (4.8), and the notation introduced in Remark 4.1, we can write the
following damped three-layer Rao-Nakra beam equations as follows:
_

_
m w D
2
x
w +KD
4
x
w D
x
Nh
2
_
G
2

2
+

G
2

2
_
= 0 on (0, L) (0, )
h
O
p
O
v
O
h
O
E
O
D
2
x
v
O
+B
T
_
G
2

2
+

G
2

2
_
= 0 on (0, L) (0, )
(4.12)
We modify (4.7) and (4.9) in a similar manner to obtain the damped three-layer Mead-Markus
equations:
_

_
m w +KD
4
x
w D
x
Nh
2
_
G
2

2
+

G
2

2
_
= 0 on (0, L) (0, )
h
O
E
O
D
2
x
v
O
+B
T
_
G
2

2
+

G
2

2
_
= 0 on (0, L) (0, )
(4.13)
4.3 Multi-layer Beam Models
The above discussion for three-layer beams can be generalized easily to multi-layer beams.
We will consider a beam consisting of n = 2m + 1 layers, in which m + 1 sti beam layers
sandwich m compliant layers (See Figure 4.3).
In addition, we use physical assumptions analogous to the ones for the three-layer model.
We apply shear damping in each of the compliant layers. Denote and v to be the n 1
vectors having the i
th
row of
i
and v
i
respectively. Let v
O
denote the (m + 1) 1 vector
[v
1
, v
3
, , v
n
]
T
consisting of the odd-indexed layers of v. Similarly, let
E
denote the m1
vector [
2
,
4
, ,
2m
]
T
consisting of the shears in the compliant layers. Dene the following
28
Figure 4.3 A multi-layer sandwich beam with m = 4.
matrices
h = diag(h
1
, h
2
, , h
n
), p = diag(
1
,
2
, ,
n
),
E = diag(E
1
, E
2
, , E
n
), G = diag(G
1
, G
2
, , G
n
),

G = diag(

G
1
,

G
2
, ,

G
n
), D = diag(D
1
, D
2
, , D
n
).
By our assumptions, we have
G
i
= 0, and

G
i
= 0 for i odd.
Throughout this thesis, we will use the following convention: If M is one of the above diagonal
matrices, then M
O
is the (m + 1) (m + 1) diagonal matrix of odd-indexed diagonal entries
of M, and M
E
is the mm diagonal matrix of even-indexed diagonal entries of M. We have
a relationship between
E
and v
O
similar to the one in (4.1) that holds; namely
Bv
O
= h
E

E
h
E

Nw
x
. (4.14)
In the above,

N = h
1
E
Ah
O

1
O
+

1
E
, where A and B are the (m+ 1) m matrices
a
ij
=
_

_
1/2 if j = i or j = i + 1
0 otherwise
b
ij
=
_

_
(1)
i+j+1
if j = i or j = i + 1
0 otherwise
,
and

1
O
and

1
E
are vectors with m + 1 and m ones respectively. For example, in the 5-layer
case m = 2 and
A =
_

_
1/2 1/2 0
0 1/2 1/2
_

_, B =
_

_
1 1 0
0 1 1
_

_,

N =
_

_
1
2h
2
(h
1
+ 2h
2
+h
3
)
1
2h
4
(h
3
+ 2h
4
+h
5
)
_

_.
29
4.3.1 Equations for the multilayer Rao-Nakra and Mead-Markus Beams
The total energy for the multilayer Rao-Nakra beam is similar to (4.3); namely,
E(t) =
R
2
[m w, w)

+ w
x
, w
x
)

+h
O
p
O
v
O
, v
O
)

]
+
R
2
[Kw
tt
, w
tt
)

+h
O
E
O
v
t
O
, v
t
O
)

+h
E
G
E

E
,
E
)

]. (4.15)
In the above, m, K, and are dened in a way analogous to (4.4); that is:
m =
n

i=1
h
i

i
, =
1
12

i=1,3, ,n

i
h
3
i
, K =
1
12

i=1,3, ,n
E
i
h
3
i
.
If we include the viscous damping in the even layers, the equations for the damped multilayer
Rao-Nakra beam are as follows (compare to (4.12)):
_

_
m w D
2
x
w +KD
4
x
w D
x

N
T
h
E
_
G
E

E
+

G
E

E
_
= 0 on (0, L) (0, )
h
O
p
O
v
O
h
O
E
O
D
2
x
v
O
+B
T
_
G
E

E
+

G
E

E
_
= 0 on (0, L) (0, )
. (4.16)
The total energy for the multilayer Mead-Markus beam is similar to (4.6); namely,
E(t) =
R
2
[m w, w)

+Kw
tt
, w
tt
)

+h
O
E
O
v
t
O
, v
t
O
)

+h
E
G
E

E
,
E
)

]. (4.17)
The damped multi-layer Mead-Markus sandwich beam equations are (compare to (4.13)):
_

_
m w +KD
4
x
w D
x

N
T
h
E
_
G
E

E
+

G
E

E
_
= 0 on (0, L) (0, )
h
O
E
O
D
2
x
v
O
+B
T
_
G
E

E
+

G
E

E
_
= 0 on (0, L) (0, )
. (4.18)
4.4 Multi-layer Plates
Now we generalize the multi-layer beam model to the multi-layer plate. We will use the
same type of structure as before; the plate will consist of m+1 sti plate layers that sandwich
m compliant plate layers. We will make the following physical assumptions:
(i) No slip occurs along the interfaces
(ii) The compliant plate layers allow shear (Mindlin-Timoshenko assumptions)
(iii) The sti plate layers do not shear (Kirchho assumptions)
30
Figure 4.4 A multi-layer sandwich plate with m = 2.
Let be a smooth, bounded set in R
2
, and set = , Also, let x = (x
1
, x
2
) denote the
points in . Since we must consider what happens in the two directions, the longitudinal
displacement and the shear have two components, and each component depends on x
1
and
x
2
. Let v
i
(x) = v
i
1
(x), v
i
2
(x), where v
i
1
denotes the in-plane displacement in the x
1
-direction
and v
i
2
denotes the in-plane displacement in the x
2
-direction in the i
th
layer (For brevity in
notation, the dependence on time is suppressed). Similarly, we denote the shear in the i
th
layer
by
i
(x) =
i
1
(x),
i
2
(x). Then denote v and to be the n 2 matrices whose i
th
entry is
v
i
and
i
respectively. Similar to before, we will use v
O
to denote the (m + 1) 2 matrix of
the odd-indexed rows of v and
E
to denote the m2 matrix of the even-indexed rows of .
From our assumptions, we have a relationship similar to (4.1) and (4.14); namely
Bv
O
= h
E

E
h
E

Nw. (4.19)
Before writing the equations of motion, it will be necessary to dene some forms and introduce
further notation. Let
i
denote the Poissons ratio of the i
th
layer, let

i
(; ) denote the
following bilinear form:

i
(
i
;

i
) :=
_

i
1
x
1
,

i
1
x
1
_

+
_

i
2
x
2
,

i
2
x
2
_

+
_

i
2
x
2
,

i
1
x
1
_

+
_

i
1
x
1
,

i
2
x
2
_

+
_
_
1
i
2
__

i
1
x
2
+

i
2
x
1
_
,
_

i
1
x
2
+

i
2
x
1
__

. (4.20)
31
We dene the Lame operator L

i
(
i
) =
_
L

i
1
(
i
1
,
i
2
), L

i
2
(
i
1
,
i
2
)
_
as follows:
L

i
1
(
i
1
,
i
2
) =

x
1
_

i
1
x
1
+
i

i
2
x
2
_
+

x
2
__
1
i
2
__

i
1
x
2
+

i
2
x
1
__
L

i
2
(
i
1
,
i
2
) =

x
2
_

i
2
x
2
+
i

i
1
x
1
_
+

x
1
__
1
i
2
__

i
2
x
1
+

i
1
x
2
__
.
Also dene the boundary operator B

i
() =
_
B

i
1
(
i
1
,
i
2
), B

i
2
(
i
1
,
i
2
)
_
as follows:
B

i
1
(
i
1
,
i
2
) =
_

i
1
x
1
+
i

i
2
x
2
_
n
1
+
_
1
i
2
__

i
1
x
2
+

i
2
x
1
_
n
2
B

i
2
(
i
1
,
i
2
) =
_

i
2
x
2
+
i

i
1
x
1
_
n
2
+
_
1
i
2
__

i
2
x
1
+

i
1
x
2
_
n
1
,
where n = (n
1
, n
2
) denotes the outward unit normal to . With these denitions, the following
Greens formula (which we prove in Appendix B) holds for suciently smooth scalar functions

i
and

i
:

i
(
i
,

i
) = B

i
,

i
)

i
,

i
)

. (4.21)
Next, we dene the bilinear form
O
(; ) as

O
(
O
;

O
) =

i=1,3, ,n

i
(
i
;

i
). (4.22)
For any suciently smooth n2 matrix = (
i
j
), i = 1, 2, , n, j = 1, 2, dene the matrices
L and B by
(L)
ij
= (L

i
j

i
), (B)
ij
= (B

i
j

i
), i = 1, 2, , n, j = 1, 2.
Furthermore, dene the operators L
O
and B
O
from the operators L and B acting on the odd
rows. Thus,
L
O
= diag (L

1
, L

3
, , L

2m+1
) , B
O
= diag (B

1
, B

3
, , B

2m+1
) .
Similar to (4.21), the following Greens formula holds:

O
(,

) = B
O
,

)

L
O
,

)

for suciently smooth (m+ 1) 2 matrices ,

. (4.23)
32
4.4.1 Equations for the multilayer Rao-Nakra and Mead-Markus Plates
The total energy for the Rao-Nakra plate is given as follows:
E(t) = m w, w)

+ w, w)

+h
O
p
O
v
O
, v
O
)

+
O
(h
3
O
D
O

1
O
w,

1
O
w) + 12
O
(h
O
D
O
v
O
, v
O
) +G
E
h
E

E
,
E
)

.
The equations of motion for the damped multilayer Rao-Nakra are (see [15]):
_

_
m w w +K
2
w div

N
T
h
E
_
G
E

E
+

G
E

E
_
= 0 in R
+
h
O
p
O
v
O
12h
O
D
O
L
O
v
O
+B
T
_
G
E

E
+

G
E

E
_
= 0 in R
+
. (4.24)
The total energy for the Mead-Markus plate is given as follows:
E(t) = m w, w)

+
O
(h
3
O
D
O

1
O
w,

1
O
w) +12
O
(h
O
D
O
v
O
, v
O
) +G
E
h
E

E
,
E
)

. (4.25)
The equations of motion for the damped multi-layer Mead-Markus are (see [15]):
_

_
m w +K
2
w div

N
T
h
E
_
G
E

E
+

G
E

E
_
= 0 in R
+
12h
O
D
O
L
O
v
O
+B
T
_
G
E

E
+

G
E

E
_
= 0 in R
+
. (4.26)
4.4.2 Korns Inequality
A very important inequality that arises in the study of multi-layer beams is Korns in-
equality. It proves that the bilinear form

i
dened in (4.20) is coercive on (H
1
0
())
2
. A
proof of Korns Inequality can be found in Lagnese and Lions [26], but we will include it for
completeness.
Theorem 4.1 (Korns Inequality:). If is a bounded open set with a suciently regular
boundary, and =
1
,
2
is such that
1
and
2
belong to L
2
() then for all > 0, there
exists a c

> 0 depending only on and such that

i
(; ) +||
2
(L
2
())
2
c

||
2
(H
1
())
2
. (4.27)
In order to prove this theorem, we will need a lemma from G. Duvaut and J.L. Lions [7]:
33
Lemma 4.1. Let be a bounded open set with regular boundary. Let f be a distribution on
such that f H
1
() and all partial derivatives of f belong to H
1
(). Then f L
2
().
The proof of Lemma 4.1 is very technical, and we will omit it. The details can be found in [7].
Proof of Korns Inequality: The following holds for all test functions on :
_

x
1
_

1
x
2
_
,
_
=
_

1
x
2
,

x
1
_
=
_

1
,

2

x
2
x
1
_
=
_

1
,

2

x
1
x
2
_
=
_

1
x
1
,

x
2
_
=
_

x
2
_

1
x
1
_
,
_
. (4.28)
Since

1
x
1
L
2
(), (4.28) implies that

x
1
_

1
x
2
_
=

x
2
_

1
x
1
_
H
1
(). (4.29)
Using a similar argument, along with the fact

2
x
2
L
2
(), we obtain

x
2
_

2
x
1
_
=

x
1
_

2
x
2
_
H
1
(). (4.30)
To show that

x
2
_

1
x
2
_
H
1
(), we consider the following:

x
2
_

1
x
2
_
=

x
2
_

1
x
2
+

2
x
1
_

x
2
_

2
x
1
_
. (4.31)
Since

1
x
2
+

2
x
1
L
2
(),

x
2
_

1
x
2
+

2
x
1
_
H
1
(). Furthermore,

x
2
_

2
x
1
_
H
1
(),
by (4.29). Therefore, (4.31) implies that

x
2
_

1
x
2
_
H
1
(). (4.32)
Using a similar argument,

x
1
_

2
x
1
_
H
1
(). (4.33)
34
Therefore, by Lemma 4.1,

1
x
2
L
2
() by (4.29) and (4.32), and

2
x
1
L
2
() by (4.30) and (4.33).
Hence, there exists a constant C > 0 depending only on such that
_
_
_
_

1
x
2
+

2
x
1
_
_
_
_
2
L
2
()
C
_
_
_
_
_

1
x
2
_
_
_
_
2
L
2
()
+
_
_
_
_

2
x
1
_
_
_
_
2
L
2
()
_
. (4.34)
Since 0 <
i
< 1/2, (4.20), and (4.34) imply that for all > 0, there exists a c

such that

i
(; ) +
_
|
1
|
2
L
2
()
+|
2
|
2
L
2
()
_
c

_
|
1
|
2
H
1
()
+|
2
|
2
H
1
()
_
. (4.35)
Hence, (4.27) follows.
Theorem 4.2.

i
is coercive on (H
1
0
())
2
.
Proof. If vanishes on the boundary, the Poincarre inequality implies that there exist positive
constants C
1
and C
2
such that
|
1
|
L
2
()
C
1
|
1
|
(L
2
())
2 and |
2
|
L
2
()
C
2
|
2
|
(L
2
())
2.
If C = maxC
1
, C
2
, then
|
j
|
(L
2
())
2 |
j
|
H
1
()
(1 +C)|
j
|
(L
2
())
2; j = 1, 2. (4.36)
Therefore, (4.35) and (4.36) imply

i
(; ) +C
_
|
1
|
(L
2
())
2 +|
2
|
(L
2
())
2

1 +C
_
|
1
|
(L
2
())
2 +|
2
|
(L
2
())
2

.
(4.37)
Choose small enough in (4.37) to obtain

i
(; ) c
_
|
1
|
(L
2
())
2 +|
2
|
(L
2
())
2

.
This result along with (4.36) implies

i
(; )
c
1 +C
_
|
1
|
H
1
()
+|
2
|
H
1
()

,
which is our desired result.
35
We have similar results for the form
O
. Dene the following spaces for s > 0:
H
s
O
() := v
O
= (v
i
j
), i = 1, 3, 5, . . . 2m+ 1, j = 1, 2 : v
i
j
H
s
()
H
s
E
() :=
E
= (
i
j
), i = 2, 4, . . . 2m, j = 1, 2 :
i
j
H
s
()
H
s
O,0
() := v
O
H
s
O
: v
O
= 0 on
H
s
E,0
() :=
E
H
s
E
:
E
= 0 on .
In addition, denote L
2
O
(), L
2
E
(), L
2
O,0
(), and L
2
E,0
() as the spaces corresponding to the
ones listed above with s = 0. Furthermore, dene the norms on these spaces as follows:
|
E
|
H
s
E
()
=

i=2,4,2m
_
_

j=1,2
|
i
j
|
H
s
()
_
_
, |v
O
|
H
s
O
()
=

i=1,3,n
_
_

j=1,2
|v
i
j
|
H
s
()
_
_
.
With this notation and the denition of
O
in (4.22), the next two corollaries easily follow from
Theorems 4.1 and 4.2.
Corollary 4.1. Suppose is a bounded open set with a suciently regular boundary. Also
suppose v
O
is such that v
i
1
and v
i
2
belong to L
2
() for all i = 1, 3, , n. then for all > 0,
there exists a constant c
O,
> 0 depending only on and such that

O
(v
O
; v
O
) +|v
O
|
2
L
2
O
()
c
O,
|v
O
|
2
H
1
O
()
. (4.38)
Corollary 4.2.
O
is coercive on H
1
O,0
().
4.4.3 Continuity of

i
on (H
1
())
2
It will also be useful to prove an inequality for

i
on (H
1
())
2
similar in nature to the
Schwartz Inequality for the inner product on L
2
().
Proposition 4.1. If =
1
,
2
(H
1
())
2
and =
1
,
2
(H
1
())
2
then

i
(; ) ||
(H
1
())
2||
(H
1
())
2. (4.39)
36
Proof. Since 0 <
i
< 1/2, (4.20) and the Schwartz inequality imply the following:

i
(, )
_
_
_
_

1
x
1
_
_
_
_
L
2
_
_
_
_

1
x
1
_
_
_
_
L
2
+
_
_
_
_

2
x
2
_
_
_
_
L
2
_
_
_
_

2
x
2
_
_
_
_
L
2
+
1
2
_
_
_
_

2
x
2
_
_
_
_
L
2
_
_
_
_

1
x
1
_
_
_
_
L
2
+
1
2
_
_
_
_

1
x
1
_
_
_
_
L
2
_
_
_
_

2
x
2
_
_
_
_
L
2
+
1
2
_
_
_
_

1
x
2
_
_
_
_
L
2
_
_
_
_

1
x
2
_
_
_
_
L
2
+
1
2
_
_
_
_

1
x
2
_
_
_
_
L
2
_
_
_
_

2
x
1
_
_
_
_
L
2
+
1
2
_
_
_
_

2
x
1
_
_
_
_
L
2
_
_
_
_

1
x
2
_
_
_
_
L
2
+
1
2
_
_
_
_

2
x
1
_
_
_
_
L
2
_
_
_
_

2
x
1
_
_
_
_
L
2

__
_
_
_

1
x
1
_
_
_
_
L
2
+
_
_
_
_

1
x
2
_
_
_
_
L
2
+
_
_
_
_

2
x
1
_
_
_
_
L
2
+
_
_
_
_

2
x
2
_
_
_
_
L
2
_

__
_
_
_

1
x
1
_
_
_
_
L
2
+
_
_
_
_

1
x
2
_
_
_
_
L
2
+
_
_
_
_

2
x
1
_
_
_
_
L
2
+
_
_
_
_

2
x
2
_
_
_
_
L
2
_
.
Thus, (4.39) follows.
It is easy to extend this result to the form
O
.
Corollary 4.3. If v
O
and v
O
belong to H
1
O
(), then

O
(v
O
, v
O
) |v
O
|
H
1
O
()
| v
O
|
H
1
O
()
. (4.40)
37
CHAPTER 5. Analyticity, Exponential Stability, and Optimal Damping of
a Multilayer Mead-Markus Beam
5.1 Introduction
In this chapter, we consider the multilayer Mead-Markus sandwich beam and prove that
the semigroup associated with this model is analytic and exponentially stable. Using hinged
boundary conditions, the equations of motion (4.18) are
_

_
m w +Kw
tttt

N
T
h
E
_
G
E

E
+

G
E

E
_
t
= 0 on (0, L) (0, )
h
O
E
O
v
tt
O
+B
T
_
G
E

E
+

G
E

E
_
= 0 on (0, L) (0, )
w = w
tt
=
t
E
= 0 for x = 0, L, t > 0
(5.1)
Throughout this chapter, we will use the convention that dots denote dierentiation with re-
spect to t and the primes denote dierentiation with respect to x. We rst formulate (5.1)
as a semigroup problem with the generator denoted by /. Then we view / as a bounded
perturbation of another operator (which we denote /
0
). In Section 5.3, we look at the eigen-
structure of /
0
, and show that the eigenfunctions and generalized eigenfunctions of /
0
form
a Riesz Basis. Then we use a semigroup perturbation result to establish the analyticity of the
semigroup generated by /. We also describe how to compute the angle of analyticity.
The remainder of the chapter discusses two optimal damping problems. As is mentioned
before, it is typical for composite beams to have an optimal level of damping, beyond which,
additional damping is counterproductive. The rst optimal problem we consider is choosing
damping parameters to achieve the optimal angle of analyticity. This problem was solved
by Hansen, but only a sketch of the proof exists in [13], and some details on proving that a
critical point leads to an optimum are lacking. We include these details in Section 5.4. In
the last section, we look at exponential stability and the growth (actually decay) rate of the
38
semigroup generated by /. Recall that the growth rate is the supremum of the real part of
the spectrum of / according to Theorem 2.5. Finally, we consider the other optimal problem:
choosing damping parameters to achieve the optimal decay rate. We solve this explicitly for
the three-layer case.
5.2 Semigroup Formulation:
We let w, v = w, and
E
be the state variables for the system (5.1). First of all, it will be
convenient to write (5.1) in terms of the state variables (eliminate the dependence on v
O
). To
this end, it will be useful to dene the mm matrix P by
P = B(h
O
E
O
)
1
B
T
.
It is known that P is an invertible, positive denite, symmetric M-matrix (see Theorem B.1
in Appendix B and [15]). If we multiply the second line of (5.1) on the right by B(h
O
E
O
)
1
,
we obtain
G
E

E
+

G
E

E
= P
1
Bv
tt
O
= P
1
h
E
(
tt
E


Nw
ttt
) by (4.14). (5.2)
Substituting this result into the rst line of (5.1), we obtain
_

_
w v = 0 on (0, L) (0, )
m v +
_
K +

N
T
h
E
P
1
h
E

N
_
w
tttt

N
T
h
E
P
1
h
E

ttt
E
= 0 on (0, L) (0, )
h
E

tt
E
+h
E

Nw
ttt
+P
_
G
E

E
+

G
E

E
_
= 0 on (0, L) (0, )
w = w
tt
=
t
E
= 0 for x = 0, L, t > 0
(5.3)
Remark 5.1. The formulation in (5.3) is valid even though the second line of (5.1) has m+1
rows while the third line of (5.3) has only m rows. A similar technicality arises in Hansens
paper on the semigroup well-posedness of a Mead-Markus plate [15], and more details on how
to deal with this issue are included there. Notice that since

1
O
is in the null space of matrix
B, multiplication of the second line of (5.1) on the left by

1
T
O
leads to

1
T
O
h
O
E
O
v
tt
O
= 0.
39
Then the boundary condition v
t
O
(0) = v
t
O
(L) = 0 implies

1
T
O
h
O
E
O
v
t
O
= 0. (5.4)
In addition, the following (m+ 1) (m+ 1) matrix is dened in [15]:
B
C
=
_

_
B

1
T
O
h
O
E
O
_

_
=
_

_
1 1
1 1
.
.
.
.
.
.
1 1
h
1
E
1
h
3
E
3
h
n2
E
n2
h
n
E
n
_

_
.
This matrix is invertible, so one can dene a pseudo-inverse of B, say S, as follows:
B
C

b =
_

_
a

0
_

_
y = Sa.
This matrix S satises the following conditions:
BSa =a a R
m
(5.5)
SB

b =

b R
m+1
with

1
T
O
h
O
E
O

b = 0. (5.6)
With our current setup, we have the following by (4.14) and (5.4):
B
C
v
t
O
=
_

_
h
E
(
t
E


Nw
tt
)

0
_

_
.
Therefore, (5.6) implies
v
t
O
= Sh
E
(
t
E


Nw
tt
). (5.7)
Dene the state space, H as follows:
H :=
_
[w, v,
E
]
T
: w H
2
(0, L) H
1
0
(0, L), v L
2
(0, L),
E
(H
1
(0, L))
m
_
,
40
We formulate (5.3) as a semigroup problem of the form (1.1) as follows:
d
dt
_

_
w
v

E
_

_
= /
_

_
w
v

E
_

_
,
where
/ :=
_

_
0 1

0
T

1
m
_
K +

N
T
h
E
P
1
h
E

N
_
D
4
x
0
1
m

N
T
h
E
P
1
h
E
D
3
x

G
1
E
P
1
h
E

ND
3
x

0

G
1
E
P
1
h
E
D
2
x

G
1
E
G
E
_

_
. (5.8)
In (5.8),

0 denotes the column vector of m zeros, and the domain of / is dened as follows:
T(/) = [w, v,
E
]
T
H : w H
4
(0, L), v H
1
0
(0, L) H
2
(0, L),

E
(H
3
(0, L))
m
, w
tt
(x) =
t
(x) = 0 if x = 0, L.
Next, we dene the following bilinear form on H:
z, z)
c
=
1
2
_
mv, v)
L
2
(0,L)
+Kw
tt
,

w
tt
)
L
2
(0,L)
+G
E
h
E

E
,

E
)
(L
2
(0,L))
m
_
+
1
2
_
P
1
h
E
(
t
E


Nw
tt
), h
E
(

E
t

N w
tt
))
(L
2
(0,L))
m.
_
(5.9)
In the above z = [w, v,
E
]
T
, and z =
_
w, v,

E
_
T
.
Remark 5.2. In (4.17), the term h
O
E
O
v
t
O
, v
t
O
)

can be rewritten as follows:

h
O
E
O
v
t
O
, v
t
O
)

= h
O
E
O
v
t
O
, Sh
E
(
t
E


Nw
tt
))

by (5.7)
= S
T
h
O
E
O
v
t
O
, Bv
t
O
)

= P
1
B(h
O
E
O
)
1
B
T
S
T
h
O
E
O
v
t
O
, Bv
t
O
)

= P
1
Bv
t
O
, Bv
t
O
)

by (5.5).
Hence, (5.9) corresponds to the total energy of the beam (4.17).
We show that (5.9) denes an inner product on H equivalent to the usual Sobolev topology
on H
2
(0, L) L
2
(0, L) (H
1
(0, L))
m
. By (5.9), we have the following upper bound on |z|
2
c
:
|z|
2
c
m|v|
2
L
2
()
+K|w
tt
|
2
L
2
()
+C
1
|
E
|
2
(L
2
())
m
+C
2
|
t
E
|
2
(L
2
())
m
+C
3
|w
tt
|
2
(L
2
())
m
C
_
|w|
2
H
2
()
+|v|
2
L
2
()
+|
E
|
2
(H
1
())
m
_
. (5.10)
41
To get a corresponding lower bound on |z|
2
c
, we look at each term on the right side of (5.9).
For w, we note that the Euler-Lagrange equation corresponding to the problem of minimizing
the Rayleigh quotient
w
tt
, w
tt
)
w, w)
on H
1
0
(0, L) H
2
(0, L) is given by
_

_
u
tttt
= u
u(0) = u(L) = u
tt
(0) = u
tt
(L) = 0
The eigenvalues of this problem are
k
=
k
4

4
L
4
, where k is a positive integer. Therefore, we
obtain
|w|
L
2
()

L
2

2
|w
tt
|
L
2
()
. (5.11)
Using integration by parts along with the Schwartz Inequality yields
|w
t
|
2
L
2
()
|w
tt
|
L
2
()
|w|
L
2
()

L
2

2
|w
tt
|
2
L
2
()
by (5.11). (5.12)
It follows from (5.11) and (5.12) that
|w|
H
2
()
|w
tt
|
L
2
()
=
|w|
L
2
()
+|w
t
|
L
2
()
+|w
tt
|
L
2
()
|w
tt
|
L
2
()
M, where M =
L
2

2
+
L

+ 1. (5.13)
Next we look for a bound involving |
E
|
(H
1
())
m. Since P
1
and h
E
are positive matrices,
there is a constant c
0
> 0 such that
c
0
|
t
E


Nw
tt
|
2
(L
2
())
m

_
P
1
h
E
(
t
E


Nw
tt
), h
E
(

E
t

N w
tt
)
_
.
Hence, the reverse triangle inequality implies that for all
1
(0, c
0
],

1
_
|
t
E
|
(L
2
())
m |

Nw
tt
|
(L
2
())
m
_

_
P
1
Bv
O
, Bv
O
)

1
|
t
E
|
(L
2
())
m
_

1
(

N
T
N)|w
tt
|
L
2
()

_
P
1
Bv
O
, Bv
O
). (5.14)
Add
_
K
2
|w
tt
|
L
2
()
to both sides of (5.14) to obtain

1
|
t
E
|
(L
2
())
m +
_
_
K
2

_

1
(

N
T
N)
_
|w
tt
|
L
2
()

_
P
1
Bv
O
, Bv
O
) +
_
K
2
w
tt
, w
tt
)
(5.15)
42
Fix
1
= min
_
c
0
,
K
2

N
T
N
_
. Then
_
K
2

_

1
(

N
T
N) 0, and (5.15) implies

1
|
t
E
|
(L
2
())
m
_
P
1
Bv
O
, Bv
O
) +
_
K
2
w
tt
, w
tt
).
Thus

1
|
t
E
|
2
(L
2
())
m
C
1
_
P
1
Bv
O
, Bv
O
) +
K
2
w
tt
, w
tt
)
_
, for some C
1
> 0. (5.16)
Let c
2
= min
_
K
2M
2
,

1
C
1
_
. Then (5.13) and (5.16) imply
c
2
_
|
t
E
|
2
(L
2
())
m
+|w|
2
H
2
()
_
P
1
Bv
O
, Bv
O
) +
K
2
w
tt
, w
tt
) +
K
2
w
tt
, w
tt
). (5.17)
Next we consider the term G
E
h
E

E
,
E
). Since h
E
and G
E
are positive matrices there exists
c
3
> 0 such that
c
3
|
E
|
2
(L
2
())
m
G
E
h
E

E
,
E
). (5.18)
Let c = minm, c
2
, c
3
. Then by (5.9), (5.13), (5.17), and (5.18), we have that
c
_
|w|
2
H
2
()
+|v|
2
L
2
()
+|
E
|
2
(H
1
())
m
_
|z|
2
c
. (5.19)
Therefore, by (5.10) and (5.19), we have that (5.9) determines a norm on H equivalent to the
usual norm on the Sobolev space H
2
(0, L) L
2
(0, L) (H
1
(0, L))
m
.
5.2.1 Semigroup Well-Posedness and Dissipativity
In this subsection, let /(

G
E
) denote the dependence of / on the parameter

G
E
.
Lemma 5.1. The adjoint of / is
[/(

G
E
)]

= /(

G
E
).
Consequently, T(/) = T(/

).
Proof. Let z = [w, v,
E
]
T
and u = [s, t,
E
]
T
be any two state vectors in T(/). This implies
the following boundary conditions:
w = w
tt
= s = s
tt
= v = t = 0,
t
E
=
t
E
= 0; x = 0, L. (5.20)
43
According to the denition of / in (5.8) and equation (4.14), we have
/(

G
E
)z =
_

_
v
1
m
_
Kw
tttt
+

N
T
h
E
P
1
h
E
(
ttt
E


Nw
tttt
)
_

G
1
E
(P
1
h
E
(
tt
E


Nw
ttt
) G
E

E
)
_

_
.
Dene := Kw
tttt
+

N
T
h
E
P
1
h
E
(
ttt
E

Nw
tttt
) and :=

G
1
E
(P
1
h
E
(
tt
E

Nw
ttt
) G
E

E
).
By (5.9), we have the following:
/(

G
E
)z, u)
c
= [v, m
1
, ]
T
, [s, t,
E
]
T
)
c
= , t) +Kv
tt
, s
tt
) +G
E
h
E
,
E
) +P
1
h
E
(
t

Nv
tt
), h
E
(
t
E


Ns
tt
))
Expanding this leads to
/(

G
E
)z, u)
c
= Kw
tttt
, t) +P
1
h
E
(
ttt
E


Nw
tttt
), h
E

Nt) +Kv
tt
, s
tt
)
+P
1
h
E
(
tt
E


Nw
ttt
),

G
1
E
h
E
G
E

E
) h
E

G
1
E
G
E

E
, G
E

E
)
+h
E

G
1
E
P
1
h
E
(
ttt
E


Nw
tttt
), P
1
h
E
(
t
E


Ns
tt
))
h
E

G
1
E
G
E

t
E
, P
1
h
E
(
t
E


Ns
tt
)) h
E

Nv
tt
, P
1
h
E
(
t
E


Ns
tt
))
Using integrations by parts along with the boundary conditions (5.20), we have the following:
/(

G
E
)z, u)
c
= Kw
tt
, t
tt
) P
1
h
E
(
tt
E


Nw
ttt
), h
E

Nt
t
) +Kv
tt
, s
tt
)
+P
1
h
E
(
tt
E


Nw
ttt
),

G
1
E
h
E
G
E

E
) h
E

G
1
E
G
E

E
, G
E

E
)
h
E

G
1
E
P
1
h
E
(
tt
E


Nw
ttt
), P
1
h
E
(
tt
E


Ns
ttt
))
+h
E

G
1
E
G
E

E
, P
1
h
E
(
tt
E


Ns
ttt
)) +h
E

Nv
t
, P
1
h
E
(
tt
E


Ns
ttt
))
Rearranging the terms in the last expression and using the fact that h
E
and

G
1
E
commute,
we have the following
/(

G
E
)z, u)
c
=
_
h
E

Nv
t
, P
1
h
E
(
tt
E


Ns
ttt
)) P
1
h
E
(
tt
E


Nw
ttt
), h
E

Nt
t
)
_
+
_
Kv
tt
, s
tt
) Kw
tt
, t
tt
)
_
h
E

G
1
E
G
E

E
, G
E

E
)
+h
E

G
1
E
G
E

E
, P
1
h
E
(
tt
E


Ns
ttt
)) +P
1
h
E
(
tt
E


Nw
ttt
), h
E

G
1
E
G
E

E
)
h
E

G
1
E
P
1
h
E
(
tt
E


Nw
ttt
), P
1
h
E
(
tt
E


Ns
ttt
)). (5.21)
44
Next, let := Ks
tttt

N
T
h
E
P
1
h
E
(
ttt
E

Ns
tttt
) and :=

G
1
E
(P
1
h
E
(
tt
E

Ns
ttt
) G
E

E
).
Then we have
z, /(

G
E
)u)
c
= [w, v,
E
]
T
, [t, m
1
, ]
T
)
c
= v, ) +Kw
tt
, t
tt
) +G
E
h
E

E
, ) +P
1
h
E
(
t
E


Nw
tt
), h
E
(
t

N(t
tt
)))
Expanding this leads to
z, /(

G
E
)u)
c
= Kv, s
tttt
) h
E

Nv, P
1
h
E
(
ttt
E


Ns
tttt
)) Kw
tt
, t
tt
)
+G
E
h
E

E
,

G
1
E
P
1
h
E
(
tt
E


Ns
ttt
))

G
1
E
G
E
h
E

E
, G
E

E
)
+P
1
h
E
(
t
E


Nw
tt
), h
E

G
1
E
P
1
h
E
(
ttt
E


Ns
tttt
))
P
1
h
E
(
t
E


Nw
tt
), h
E

G
1
E
G
E

t
E
) +P
1
h
E
(
t
E


Nw
tt
), h
E

Nt
tt
)
Using integrations by parts along with the boundary conditions (5.20), we have the following:
z, /(

G
E
)u)
c
= Kv
tt
, s
tt
) +h
E

Nv
t
, P
1
h
E
(
tt
E


Ns
ttt
)) Kw
tt
, t
tt
)
+G
E
h
E

E
,

G
1
E
P
1
h
E
(
tt
E


Ns
ttt
))

G
1
E
G
E
h
E

E
, G
E

E
)
P
1
h
E
(
tt
E


Nw
ttt
), h
E

G
1
E
P
1
h
E
(
tt
E


Ns
ttt
))
+P
1
h
E
(
tt
E


Nw
ttt
), h
E

G
1
E
G
E

E
) P
1
h
E
(
tt
E


Nw
ttt
), h
E

Nt
t
)
Rearranging the terms in the last expression and using the fact that h
E
and

G
1
E
commute,
we have the following
z, /(

G
E
)u)
c
=
_
h
E

Nv
t
, P
1
h
E
(
tt
E


Ns
ttt
)) P
1
h
E
(
tt
E


Nw
ttt
), h
E

Nt
t
)
_
+
_
Kv
tt
, s
tt
) Kw
tt
, t
tt
)
_

G
1
E
G
E
h
E

E
, G
E

E
)
+h
E

G
1
E
G
E

E
, P
1
h
E
(
tt
E


Ns
ttt
)) +P
1
h
E
(
tt
E


Nw
ttt
), h
E

G
1
E
G
E

E
)
h
E

G
1
E
P
1
h
E
(
tt
E


Nw
ttt
), P
1
h
E
(
tt
E


Ns
ttt
)). (5.22)
Since h
E
, G
E
, and

G
E
commute, the right hand side of (5.22) agrees with that of (5.21). Then
the conclusion of the theorem follows.
45
Theorem 5.1. / is the generator of a C
0
semigroup of contractions on H. Moreover, for all
z T(/), we have
Re /z, z)
c
=
_
_
_h
1/2
E

G
1/2
E
_
P
1
h
E
(
tt
E


Nw
ttt
) G
E

E
__
_
_
2
(L
2
(0,L))
m
. (5.23)
Proof. According to (5.21) and (5.22), we have
/z, z)
c
= 2i Im h
E

Nv
t
, P
1
h
E
(
tt
E


Nw
ttt
)) + 2i Im Kv
tt
, w
tt
)
h
E

G
1
E
G
E

E
, G
E

E
)
+ 2 Reh
E

G
1
E
G
E

E
, P
1
h
E
(
tt
E


Nw
ttt
))
h
E

G
1
E
P
1
h
E
(
tt
E


Nw
ttt
), P
1
h
E
(
tt
E


Nw
ttt
)). (5.24)
Taking the real part of (5.24) implies
Re /z, z)
c
=
_
h
E

G
1
E
_
G
E

E
P
1
h
E
(
tt
E


Nw
ttt
)
_
, G
E

E
P
1
h
E
(
tt
E


Nw
ttt
)
_
.
Thus (5.23) holds, and / is dissipative. Then Lemma 5.1 and Theorem 2.3 imply that / is
the generator of a C
0
semigroup of contractions on H.
5.2.2 Semigroup Decomposition
We can decompose / as follows:
/ = /
0
+G
E
B, where
/
0
:=
_

_
0 1

0
T

1
m
_
K +

N
T
h
E
P
1
h
E

N
_
D
4
x
0
1
m

N
T
h
E
P
1
h
E
D
3
x

G
1
E
P
1
h
E

ND
3
x

0

G
1
E
P
1
h
E
D
2
x
_

_
, and
B :=
_

_
0 0

0
T
0 0

0
T

0

0

G
E
_

_
. (5.25)
Moreover, T(/
0
) = T(/), and T(B) = H. Thus, we will view / as a bounded perturbation
of /
0
. In the next section, we will focus primarily on /
0
.
46
5.3 Analysis of the Semigroup Generator /
0
Since the dissipativity result in (5.23) remains valid when G
E
= 0, Theorem 5.1 remains
valid in this case. Hence, /
0
is the generator of a C
0
semigroup of contractions on H. The
partial dierential equation corresponding to /
0
is obtained by omitting terms involving G
E
in (5.3), yielding the system
_

_
w v = 0 on (0, L) (0, )
m v +
_
K +

N
T
h
E
P
1
h
E

N
_
w
tttt

N
T
h
E
P
1
h
E

ttt
E
= 0 on (0, L) (0, )
h
E

tt
E
+h
E

Nw
ttt
+P

G
E

E
= 0 on (0, L) (0, )
w = w
tt
=
t
E
= 0 for x = 0, L, t > 0
(5.26)
Separation of variables applied to (5.26) yields solutions of the form
w = 0, v = 0,
E
=

C
0
; k = 0 (5.27)
w =
a

2
k
e
s
k
t
sin
k
x, v = be
s
k
t
sin
k
x,
E
=

C
k

k
e
s
k
t
cos
k
x; k N. (5.28)
In (5.27),

C
0
is an m-vector of constants; and in (5.28),
k
= k/L, and

C
k
= (c
2
, c
4
, , c
2m
)
T
for each k N. Substituting the modal solutions (5.28) into (5.26) yields the following:
_

_
a
s
k

2
k
b = 0
bms
k
+a
_
K +

N
T
h
E
P
1
h
E

N
_

2
k

N
T
h
E
P
1
h
E

2
k

C
k
= 0
ah
E

N
k
+h
E

k

C
k
+
P

k
s
k

G
E

C
k
= 0
(5.29)
For convenience, well suppress the subscript k on s, , and

C. In addition, we dene
y :=
s

2
,

H = h
E

N, and = h
1
E

G
E
,
for brevity in notation. If we divide the second line of (5.29) by
2
and the third line by , we
can apply the above denitions and write (5.29) as follows:
_

_
ay b = 0
a(K +

H
T
P
1

H) +bmy

H
T
P
1
h
E

C = 0
a

H +h
E

C +Py

G
E

C = 0
(5.30)
47
The above can be rewritten as the following (m+ 2) (m+ 2) matrix system:
_

_
y 1

0
T
K +

H
T
P
1

H my

H
T
P
1
h
E

H

0 (I
m
+Py) h
E
_

_
a
b

C
_

_
=
_

_
0
0

0
_

_
. (5.31)
Remark 5.3. It is possible to rewrite (5.31) as a standard eigenvalue problem. In particular,
(5.31) is equivalent to
_
_
_
_
_
_
yI
m+2

_
0 1/m

0
T
(K +

H
T
P
1

H) 0

H
T
P
1

1/2

1/2
P
1

H

0
1/2
P
1

1/2
_

_
_
_
_
_
_
_
_

_
a
mb

1/2
h
E

C
_

_
=
_

_
0
0

0
_

_
.
Therefore, elementary matrix theory implies that (5.29) has a set of m+2 eigenvalues (including
multiplicities) with corresponding eigenvectors and generalized eigenvectors.
5.3.1 Eigenvalues and Eigenfunctions of /
0
Next, we determine the eigenstructure of /
0
. Let e
j

m+2
j=1
denote the sequence of standard
basis elements in R
m+2
. Dene the following for all k N:
E
1,k
= 1/
2
k
sin(
k
x)e
1
, E
2,k
= sin(
k
x)e
2
,
E
3,k
= 1/
k
cos(
k
x)e
3
, , E
(m+2),k
= 1/
k
cos(
k
x)e
(m+2)
, and

k
=
_
E
1,k
.
.
. E
2,k
.
.
.
.
.
. E
(m+2),k
_
.
In addition, it will be convenient to dene the following (m+ 2) (m+ 2) matrix:
R :=
_

_
0 1

0
T
1/m(K +

H
T
P
1

H) 0 1/m

H
T
P
1
h
E
h
1
E

1
P
1

H

0 h
1
E

1
P
1
h
E
_

_
. (5.32)
With this notation in place, we can prove the following result.
Lemma 5.2. For all k N,
/
0

k
=
2
k

k
R, k N. (5.33)
48
Proof. First of all, it is easy to show that
/
0
E
2,k
=
2
k
E
1,k
.
Therefore,
/
0
E
2,k
=
2
k

k
_

_
1
0

0
_

_
. (5.34)
For proving similar results with the remaining E
j,k
, it will be helpful to dene the matrices
[f
1
, f
2
, , f
m
] =
1
m

H
T
P
1
h
E
,
[g
1
, g
2
, , g
m
]
T
= h
1
E

1
P
1

H,
and
_

_
c
11
c
12
c
1m
c
21
c
22
c
2m
.
.
.
.
.
.
.
.
.
.
.
.
c
m1
c
m2
c
mm
_

_
=
1
h
1
E
P
1
h
E
Then
/
0
E
1,k
=
_

_
0 1 0 0

1
m
(

H
T
P
1

H +K)D
4
x
0 f
1
D
3
x
f
m
D
3
x
g
1
D
3
x
0 c
11
D
2
x
c
1m
D
2
x
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
g
m
D
3
x
0 c
m1
D
2
x
c
mm
D
2
x
_

_
_

_
1

2
k
sin(
k
x)
0
0
.
.
.
0
_

_
=
_

_
0

1
m
(

H
T
P
1

H +K)
2
k
sin(
k
x)
g
1

k
cos(
k
x)
.
.
.
g
1

k
cos(
k
x)
_

_
.
49
Thus,
/
0
E
1,k
=
2
k
_
E
1,k
.
.
.E
2,k
.
.
.
.
.
.E
(m+2),k
_
_

_
0

1
m
(

H
T
P
1

H +K)
g
1
.
.
.
g
m
_

_
.
Therefore,
/
0
E
1,k
=
2
k

k
_

_
0

1
m
(

H
T
P
1

H +K)
h
1
E

1
P
1

H
_

_
. (5.35)
Using a similar procedure, one can obtain
/
0
E
(j+2),k
=
2
k

k
_

_
0
f
j
c
1j
.
.
.
c
mj
_

_
for all j = 1, 2, , m.
This implies
/
0
_
E
3,k
.
.
.E
4,k
.
.
.
.
.
.E
(m+2),k
_
=
2
k

k
_

_
0 0 0
f
1
f
2
f
m
c
11
c
12
c
1m
.
.
.
.
.
.
.
.
.
.
.
.
c
m1
c
m2
c
mm
_

_
,
which can be rewritten as
/
0
_
E
3,k
.
.
.E
4,k
.
.
.
.
.
.E
(m+2),k
_
=
_

0
T
1
m

H
T
P
1
h
E

1
h
1
E
P
1
h
E
_

_
. (5.36)
Putting the results of (5.34), (5.35), and (5.36) together proves (5.33).
50
Next dene
E
j,0
= e
j+2
, j = 1, 2, , m.
Note that (5.27) implies
/
0
E
j,0
= 0; j = 1, 2, , m. (5.37)
Let Q = (j, k) : j 1, 2, , m if k = 0; j 1, 2, , m + 2 if k N. Note that
E
j,k

(j,k)Q
forms an orthogonal basis for H that satises < |E
j,k
| < 1/ for some > 0,
for all (j, k) Q. In addition, note that the null space of /
0
, ^(/
0
), is orthogonal to the
range of /
0
, 1(/
0
). Hence, it will be convenient to dene the following:
1 := [^(/
0
)]

in H, and /
1
:= /
0

1
.
Then E
j,k

(j,k)I
forms a basis for 1, where I := (j, k) : j 1, 2, , m + 2, k N. Let

m+2
j=1
be the sequence of eigenvalues of R (they need not be distinct), and let
j

m+2
j=1
the
corresponding sequence of eigenvectors and generalized eigenvectors. From (5.33) and (5.37),
we have the following result:
Lemma 5.3.
k

(j,k)I
is the sequence of eigenfunctions and generalized eigenfunctions
for /
1
with corresponding eigenvalues
2
k

(j,k)I
. Furthermore, E
0,j

m
j=1
is the sequence
of eigenvectors for /
0
with eigenvalue 0.
Proof. The second assertion is trivial according to (5.37). For the sake of simplicity, we will
consider the case when the
j
are distinct. Then (5.33) implies
/
0

j
=
2
k

j
, (j, k) I.
Remark 5.4. The analysis becomes more complicated when at least one of the eigenvalues are
repeated. The details involve the theory of Jordan canonical matrices. For completeness, we
include these details in Appendix B.
51
5.3.2 Riesz Basis Property for /
1
To show that the eigenfunctions and generalized eigenfunctions of /
1
form a Riesz Basis
on 1, we show that there exists a bounded operator that maps an orthogonal basis of 1 to the
basis of eigenvectors and generalized eigenvectors of /
1
(see Theorem 7 on pp. 25-26 in [48]).
We proceed using a method similar to the one used in [11]. First, we will need the following
lemma:
Lemma 5.4. Let X be a Hilbert space with an orthogonal basis E
j,k

(j,k)I
. Moreover, suppose
there is an > 0 such that < |E
j,k
| < 1/ for all (j, k) I. If T : X X has matrix
representation

T := diag(M, M, ), (5.38)
where M is an (m+2)(m+2) real, invertible matrix, then T is a bounded, invertible operator
on X.
Proof. We rst show that T is bounded. Without loss of generality, we can assume that
E
j,k

(j,k)I
is an orthonormal basis for X. Furthermore, since X is a separable Hilbert space,
we may also assume without loss of generality that X =
2
(the space of square-summable
sequences). Let x X be represented as follows:
x =

k=1
x
k
In the above, x
k
has the representation
x
k
=
_
0, , 0
.
.
.
.
.
. 0, , 0
.
.
. x
k
.
.
. 0, , 0
.
.
.
_
T
,
where
x
k
=
_
x
1,k
, , x
(m+2),k

R
(m+2)
.
With this notation, one can prove the following (see Appendix B for details):
|Tx|
2
X
=

k=1
|M x
T
k
|
2
R
(m+2)
(5.39)
52
Since M is a bounded, invertible matrix, there is a constant C > 0 (independent of k) such
that
|M x
T
k
|
R
(m+2) C| x
T
k
|
R
(m+2) (5.40)
For all x X, (5.39) and (5.40) imply the following:
|Tx|
2
X
= C
2

k=1
| x
T
k
|
2
R
(m+2)
= C
2

k=1
|x
k
|
2
X
by Parsevals Identity
= C
2
|x|
2
X
.
Hence T is bounded on X. To prove that T is invertible, we consider the operator V with the
matrix representation

V = diag(M
1
, M
1
, ).
This is well-dened, since M is invertible. Since M
1
is bounded, we repeat the same argument
as above to conclude that V is bounded on X. Clearly, V T = TV = I, and so V = T
1
.
Theorem 5.2. The eigenfunctions and generalized eigenfunctions of /
1
form a Riesz basis
on 1.
Proof. Dene the (m+ 2) (m+ 2) matrix
:=
_

1
.
.
.
2
.
.
.
.
.
.
(m+2)
_
.
Standard matrix theory implies that is a bounded, invertible matrix which transforms R
into Jordan canonical form. Dene M :=
T
. By (5.33), we have for every k N,
_

_
(
k

1
)
T
(
k

2
)
T
.
.
.
(
k

(m+2)
)
T
_

_
= M
_

_
E
T
1,k
E
T
2,k
.
.
.
E
T
(m+2),k
_

_
. (5.41)
Dene the operator T on 1 by
T(E
j,k
) :=
k

j
, for (j, k) I.
53
Then, using M in (5.41), T has the matrix representation (5.38) in Lemma 5.4. Since M is a
bounded, invertible matrix, and E
j,k

(j,k)I
satises < |E
j,k
| < 1/ for some > 0, Lemma
5.4 implies that T is a bounded, invertible transformation. Therefore, the eigenfunctions of
/
1
form a Riesz basis on 1.
5.3.3 Analyticity of the semigroups generated by /
1
, /
0
, and /
We rst show that /
1
is analytic and describe the sector of analyticity. Before doing so,
we will need the following lemma:
Lemma 5.5. The operator /
0
has no nonzero eigenvalues on the imaginary axis.
Proof. Suppose on the contrary that i ( ,= 0) is an eigenvalue for /
0
with eigenvector
z = [ w, v,

E
]
T
. Then (5.23) (with G
E
= 0) implies
0 = Re z, (iI /
0
) z)
c
= Re z, /
0
z)
c
=
_
_
_h
1/2
E

G
1/2
E
P
1
h
E
(

E
tt

N w
ttt
)
_
_
_
2
(L
2
(0,L))
m
. (5.42)
Since h
E
,

G
E
and P
1
are positive, invertible matrices, then (5.42) implies

E
tt

N w
ttt
= 0. (5.43)
Furthermore, solutions z = [w, v,
E
]
T
to
d
dt
z = /
0
z, z(0) = z
must satisfy the following (see the third line of (5.26)):
d
dt

t=0
=

G
1
E
P
1
h
E
(

E
tt

N w
ttt
) = 0 by (5.43). (5.44)
However, the eigenvector condition i z = /
0
z = z

t=0
, implies that i

E
=
d
dt

t=0
. Since
,= 0, it follows from (5.44) that

E
= 0. (5.45)
Going back to (5.43), the boundary conditions

E
t
= 0 and w
tt
= 0 for x = 0, L imply

E
t

N w
tt
= 0.
54
Then it follows immediately from (5.45) that
w
tt
= 0.
The boundary conditions on w imply | w|
L
2
(0,L)

L
2

2
| w
tt
|
L
2
(0,L)
(see (5.11)), and so
w = 0.
Finally it easily follows from the eigenvector condition i z = /
0
z = z

t=0
that y = 0, and
hence z = 0.
In addition to the previous lemma, we know that the nonzero eigenvalues of /
0
are scalar
multiples of the eigenvalues of R, according to (5.33). Hence, it follows from (5.23) with
G
E
= 0 that all eigenvalues of R and nonzero eigenvalues of /
0
are in the open left-half plane.
Let
j

m+2
j=1
be the sequence of eigenvalues for R, and dene
:= min[arg (
j
)[ : j 1, 2, , m+ 2.
Then > /2. Next, we dene the following sets:

:= z C : [arg z[ < /2 ,

:= z C : [arg z[ .
Theorem 5.3. /
1
generates a semigroup on 1 analytic in the sector

.
Proof. It is enough to show that /
1
is the generator of an analytic semigroup on any sector

1
, where 0 <
1
< [39]. Since T(/
0
) is compactly embedded in H, the spectrum of /
0
consists of eigenvalues. In addition, Lemma 5.5 implies /
1
has no zero eigenvalues. Thus,
0 (/
1
) and (/
1
) is contained in the interior of

. This implies
|R(, /
1
)|
/(1)
C, for all (C

1
) B(0, 1). (5.46)
For = z
0
, where > 0 and [z
0
[ = 1, it follows [39] that /
1
will be analytic on

1
if
|R(z
0
, /
1
)|
/(1)

C

for all 1. (5.47)

Let e
k,t
k

be a Riesz basis for 1 which are the eigenvectors and generalized eigenvectors of
/
1
corresponding to eigenvalues
k

k=1
. The index set is dened as follows:
55
:= (k, t
k
) : k N, t
k
is the length of the Jordan chain associated with
k
.
If is in the resolvent of /
1
, and if e
k,1
, , e
k,t
k
is a Jordan chain associated with the
eigenvalue
k
, then we have the following for each j = 1, 2, t
k
:
(I /
0
)
1
e
k,j
=
1
(
k
)
j
e
k,1
+
1
(
k
)
j1
e
k,2
+ +
1
(
k
)
e
k,j
. (5.48)
In addition, note that min[z
0

k
[ : k N exists, and it is proportional to [z
0
[ = .
Hence, there exists a constant

C > 0 (independent of k) such that
1
[z
0

k
[
<

for all k N. (5.49)

Thus, for each k N, and j = 1, 2, t
k
, (5.48) and (5.49) imply
_
_
(z
0
I /
1
)
1
e
k,j
_
_

|e
k,1
|
[z
0

k
[
j
+
|e
k,2
|
[z
0

k
[
(j1)
+ +
|e
k,j
|
[z
0

k
[

C
_
|e
k,1
|

j
+
|e
k,2
|

j1
+ +
|e
k,j
|

C
_
|e
k,1
|

t
k
+
|e
k,2
|

t
k
1
+ +
|e
k,t
k
|

_
, since t
k
j

[|e
k,1
| +|e
k,2
| + +|e
k,t
k
|] , since 1
Therefore, for each k N and 1,
_
_
_
_
_
_
t
k

j=1
(z
0
I /
1
)
1
e
k,j
_
_
_
_
_
_

t
k

j=1
[|e
k,1
| +|e
k,2
| + +|e
k,t
k
|]

m+2

j=1
[|e
k,1
| +|e
k,2
| + +|e
k,t
k
|]
=
(m+ 2)

C

[|e
k,1
| +|e
k,2
| + +|e
k,t
k
|] ,
since no Jordan chain can have length greater than m + 2. Thus, for each k N and 1,
there is a constant C
k
such that
_
_
_
_
_
_
t
k

j=1
(z
0
I /
1
)
1
e
k,j
_
_
_
_
_
_
2

(m+ 2)
2

C
2
C
k

2
t
k

j=1
|e
k,j
|
2
. (5.50)
56
Since t
k
m + 2 for all k N, there is a C
1
< such that C
k
< C
1
for all k N. Then for
all x 1 and 1, we have
_
_
(z
0
I /
1
)
1
x
_
_
2
=
_
_
_
_
_
_

k=1
t
k

j=1
(iI /
1
)
1
c
k,j
e
k,j
_
_
_
_
_
_
2

(m+ 2)
2

C
2
C
1

k=1
t
k

j=1
[c
k,j
[
2
|e
k,j
|
2
by (5.50)
=
(m+ 2)
2

C
2
C
1

2
|x|
2
.
Since the constant

C in (5.49) is uniform for all z
0
C

1
on the unit circle, (5.47) holds.
This completes the proof.
Corollary 5.1. / generates a semigroup on H analytic in the sector

.
Proof. We show that /
0
is analytic on every sub-sector

1
of

. With respect to the basis

E
j,k

(j,k)Q
dened earlier, /
0
has the representation diag (0
m
, /
1
), where 0
m
denotes the
mm matrix of zeros. Thus, for all such that [arg [
1
and ,= 0,
_
_
(I /
0
)
1
_
_
/(1)

_
_
_
(I
m
)
1
_
_
/(A(,
0
))
+
_
_
(I /
1
)
1
_
_
/(1)
_

1
[[
+
C
[[
by (5.46) and (5.47)

C
[[
.
Next, since / is a bounded perturbation of /
0
(see (5.25)), then Theorem 3.2.1 in Pazy [39]
implies / generates a semigroup on H that is analytic on

.
Remark 5.5. Even though / is a perturbation of /
0
, the sector of analyticity remains un-
changed. This is because for high frequencies (k ), the argument of the non-real eigenval-
ues of / are close to the argument of the eigenvalues for /
0
. This type of eigenvalue behavior is
known as frequency-proportional damping (see [13] for an explanation). Figure 5.1 illustrates
this phenomenon for a three-layer Mead-Markus beam. The typical location of the eigenvalues
of / with the matrix G
E
having all positive entries are marked by +, while the typical loca-
tion of the corresponding operator /
0
(/ with G
E
= 0) are marked by o. The rst 20 modes
of eigenvalues are plotted (k = 1, 2, , 20).
57
Figure 5.1 Eigenvalues of /
0
and of /.
Looking at this gure, it appears that the parameter G
E
being a positive matrix destabilizes
the complex eigenvalues and stabilizes the real eigenvalues.
5.3.4 Calculation of
Since the angle of analyticity is determined by the complex eigenvalues of R, the following
theorem will be very useful.
Theorem 5.4. R has at least m negative eigenvalues. Consequently, there are at most two
non-real eigenvalues which must occur as a complex conjugate pair.
Proof. In order to characterize the eigenvalues, it will be more convenient to rewrite (5.31) as
follows:
_

_
y 1

0
T
K +

H
T
P
1

H my

H
T
P
1/2
P
1/2

H

0 I +P
1/2
(y)P
1/2
_

_
_

_
a
b
P
1/2
h
E

C
_

_
=
_

_
0
0

0
_

_
. (5.51)
Since P is a symmetric, positive denite matrix, the matrix P
1/2
P
1/2
is also positive denite
and symmetric. Hence the Spectral Theorem for Hermitian Matrices (see Theorem 5.4 in [2])
58
implies there exists a real unitary matrix U such that
UP
1/2
P
1/2
U
T
=
_

1
.
.
.

m
_

_
, and
i
> 0 for each i = 1, , m. (5.52)
Next, we dene the following (m+ 2)-by-(m+ 2) matrix:
=
_

_
1 0

0
T
0 1

0
T

0

0 U
_

_
.
It is clear that is a real, unitary matrix, and that

1
=
_

_
1 0

0
T
0 1

0
T

0

0 U
T
_

_
.
Using , one can modify (5.51) as follows:

_
y 1

0
T
K +

H
T
P
1

H my

H
T
P
1/2
P
1/2

H

0 I +P
1/2
(y)P
1/2
_

_
a
mb
P
1/2
h
E

C
_

_
=
_

_
0
0

0
_

_
y 1

0
T
K +

H
T
P
1

H my

H
T
P
1/2
U
T
UP
1/2

H

0 I +UP
1/2
(y)P
1/2
U
T
_

_
_

_
a
b
UP
1/2
h
E

C
_

_
=
_

_
0
0

0
_

_
.
Dene

b = (b
1
, b
2
, , b
m
)
T
by

b := UP
1/2

H. With this denition, we have that

H
T
P
1/2
U
T
=

b
T
and

H
T
P
1

H = |

b|
2
. Then we use (5.52) to rewrite the above matrix system as follows:
_

_
y 1
0 0
K +|

b|
2
my
b
1
b
m
b
1

b
m
0
.
.
.
0
1 +
1
y 0
.
.
.
.
.
.
.
.
.
0 1 +
m
y
_

_
_

_
a
b
UP
1/2
h
E

C
_

_
=
_

_
0
0

0
_

_
. (5.53)
59
The determinant of the matrix on the left side of (5.53) is zero if
(my
2
+K +|

b|
2
)(1 +
1
y)(1 +
2
y) (1 +
m
y)
b
2
1
(1 +
2
y)(1 +
3
y) (1 +
m
y)
b
2
m
(1 +
1
y)(1 +
2
y) (1 +
m1
y) = 0. (5.54)
We rst consider the case of distinct
i
and assume without loss of generality that
m
< <

1
. Factoring (1 +
1
y) (1 +
m
y) out of (5.54) implies
my
2
+K +|

b|
2
= F(y), (5.55)
where
F(y) :=
b
2
1
1 +
1
y
+ +
b
2
m
1 +
m
y
.
We will solve (5.55) by considering the intersection points of the graph of z = F(y) with that
of z = my
2
+K +|

b|
2
. Notice that for each i = 1, 2, , m, we have
lim
y1/

i
F(y) = and lim
y1/
+
i
F(y) = +.
Also note that F(0) = |

b|
2
, so the z-intercept of F(y) is below that of the graph of z =
my
2
+K +|

b|. Moreover, the graph of z = my

2
+K +|

b| is a parabola opening upward, and

it will intersect F(y) in at least m distinct values of y near 1/
i
for each i. (see Figure 5.2
below). Since each
i
is positive, the real solutions to (5.55) are all negative.
Next, suppose for example that
1
=
2
= =
k
for k > 1 in (5.52), and the remaining
i
are distinct. Set b = b
2
1
+ b
2
k
. Then we have the following by (5.54):
(my
2
+K +|

b|
2
)(1 +
1
y)
k
(1 +
k+1
y) (1 +
m
y)
+
_
b
2
1 +
1
y
_
(1 +
1
y)
k
(1 +
k+1
y) (1 +
m
y)
+ +
_
b
2
m
1 +
m
y
_
(1 +
1
y)
k
(1 +
k+1
y) (1 +
m
y) = 0. (5.56)
If we factor (1
1
y)
k1
out of the left side of (5.56) then divide both sides by
(1 +
k
y) (1 +
m
y), we obtain
(1
1
y)
k1
_
my
2
+K +|

b|
2
+
b
2
1 +
k
y
+ +
b
2
m
1 +
m
y
_
= 0. (5.57)
60
Figure 5.2 Negative eigenvalues of R when the
i
are distinct.
Notice that setting the expression in the brackets set equal to 0 gives us precisely (5.55), except
that there are m k + 1 distinct values for the
i
instead of m of them. Therefore it follows
from the distinct root case that there are at least m k + 1 distinct real roots associated
with the term in the brackets on the left side of (5.57). Moreover, y =
1

1
is clearly a
zero of (5.55) of multiplicity k 1 according to (5.57). Therefore, the equation has at least
(mk + 1) + (k 1) = m real roots up to multiplicity. We can repeat this argument if there
are multiple repeated roots. This completes the proof.
5.4 Optimal Damping Angle
We now consider how to choose the damping parameters

G
E
in order to achieve an optimal
angle of analyticity. In particular, we will study the eigenvalues of / as varies (recall that

G
E
= h
E
, so has the damping information in it). Figure 5.3 shows a typical trajectory for
the eigenvalues in the second quadrant of the complex plane for / with k = 1 as varies. In
the gure, the optimal value of , say

, is indicated by o.
61
Figure 5.3 Typical eigenvalues of / as varies.
Thus, we pose the following optimization problem:
Maximize : diagonal matrices in R
mm
(5.58)
The following theorem explicitly describes how to nd the optimal damping parameters for
analyticity (see Theorem 4.1 in [13])
Theorem 5.5. Suppose

H
T
P
1

H
K
< 8. Then there exists a unique value of , say

, that
solves (5.58) and satises
P

H =

H, where =
m
1/2
K
1/4
(K +

H
T
P
1
H)
3/4
. (5.59)
Remark 5.6. A sketch of the proof of this theorem is given in Hansens Optimal Damping
paper [13]. However, it is incomplete because no proof is given that any critical point must be
an optimal point. For completeness, we include a detailed proof.
62
5.4.1 Preliminaries for proving Theorem 5.5
We can use (5.30) to calculate explicitly . First, we claim that if y is a non-real eigenvalue
of (5.30), then a ,= 0. For if a = 0, then b = 0 by (5.30). Inserting a = 0 and b = 0 into (5.31)
leads to
(I +Py)h
E

C = 0.
If we dene V := Ph
E

C, we have
P
1

1
V = yV. (5.60)
Using (5.60), we have
yV
T
PV = V
T
P(yV ) = V
T
P(P
1

1
V )
= V
T

1
V
= V
T

1
P
1
PV
= (P
1

1
V )
T
PV = yV
T
PV .
Therefore,
y = y.
Thus we have proven that the case a = 0 is consistent with real eigenvalues, and we can assume
without loss of generality that a = 1. Then (5.30) becomes
_

_
my
2
+
_
K +

H
T
P
1

H
_

H
T
P
1
h
E

C
k
= 0

H +h
E

C
k
+Pyh
E

C
k
= 0
(5.61)
Solving the second line of (5.61) for h
E

C
k
and substituting into the rst line leads to
my
2
+K +

H
T
P
1
_
I (I +Py)
1

H = 0. (5.62)
5.4.2 Properties for a critical point
We rewrite the characteristic equation (5.62) as follows:
my
2
+K +

H
T
Q
1

H = 0; Q =
1
y
1
+P. (5.63)
63
As a consequence of Theorem 5.4, complex non-real roots of (5.63) are simple and thus vary
analytically with respect to
i
(an element on the diagonal of , i = 1, 2, , m) in a neigh-
borhood of a non-real root. Let

denote a value of corresponding to

, the optimal value

of (5.58). Since

< , there exists a unique root of (5.63), say y

c
, in the upper half of the
complex plane. Then y
c
varies analytically with respect to in a neighborhood of

[24]. If
y
c
is a non-real root in the upper half of the complex plane, then = arg y
c
. Using (5.63), we
calculate implicitly the derivative y
t
c
of y
c
with respect to in the direction of M where M is
a real diagonal matrix. (Since is positive and diagonal, all valid variations should be in a
real, diagonal-matrix direction.) One nds
2my
c
y
t
c

H
T
Q
1
Q
t
Q
1

H
T
= 0
2my
c
y
t
c
+

H
T
(
1
y
1
c
+P)
1
(y
c
)
1
_
My
c
+ y
t
c

(y
c
)
1
(
1
y
1
c
+P)
1

H
T
= 0.
If we evaluate this at =

, we obtain
y
t
c
=

H
T
(I +

y
c
P)
1
My
c
(I +P

y
c
)
1

H
2my
c
+

H
T
(I +

y
c
P)
1

(I +P

y
c
)
1
H
. (5.64)
If y
c
= Re
i
, then
y
t
c
y
c
=
R
t
R
+i
t
. (5.65)
Therefore, in order for

to be a critical point (i.e.

t
= 0), (5.65) implies that y
t
c
/y
c
must be
real. Thus, (5.64) implies that we must have
y
t
c
y
c
=

Z
T
M

Z
2my
c
+

Z
T

Z
is real, (5.66)
where

Z = [z
1
, , z
m
]
T
:= (I +P

y
c
)
1

H.
By considering M with all zeros except for a 1 on the k
th
diagonal entry (k 1, 2, , m),
(5.66) implies
y
t
c
y
c
=
z
2
k
2my
c
+

Z
T

Z
, k = 1, 2, m.
Therefore, arg (z
2
k
) is independent of k. Thus, there exists a constant C and a

W R
m
such that

Z =

W. (5.67)
64
Then the denition of

Z and (5.67) imply

H =

W +y
c
P

W.
Taking the imaginary part of this yields
0 = Im ()

W + Im (y
c
)P

W. (5.68)
Note that the imaginary part of y
c
is nonzero because if were zero, then (5.68) would imply
that Im y
c
= 0, which is a contradiction to our assumption that y
c
CR. Thus, if we dene
:=
Im ()
Im (y
c
)
we have by (5.68),
P

W =

W.
But
P

W =

W P

Z =

Z by (5.67)
P

(I +P

y
c
)
1

H = (I +P

y
c
)
1

H

1
y
c
_
I (I +P

y
c
)
1

H = (I +P

y
c
)
1

H

H (I +P

y
c
)
1

H = y
c
(I +P

y
c
)
1

H
(I +P

y
c
)

H

H = y
c

H.
Therefore,
P

H =

H. (5.69)
Then (5.69) implies
(I +P

y
c
)
1

H =
1
1 +y
c

H. (5.70)
From (5.64) we obtain
y
t
c
y
c
=

H
T
M

H
2my
c
(1 +y
c
)
2
+

H
T

H
. (5.71)
To show that

is a critical point it is enough that (5.71) is real, independent of the direction

of variation M. However, since M and

are real, it is enough to show that:

y
c
(1 +y
c
)
2
is real (5.72)
65
Using the eigenvector condition in (5.69) we have that

H
T

H =

H
T
P
1

H. Thus (5.63)
becomes
(my
2
c
+K)(1 +y
c
) +y
c

H
T
P
1

H = 0.
This can be rewritten as
y
3
c
+
1

y
2
c
+
K +

H
T
P
1

H
m
y
c
+
K
m
= 0. (5.73)
We make the substitution
y
c
= x
_
K/m,
and rewrite equation (5.73) as follows:
x
3
+
1

_
m
K
x
2
+
K +

H
T
P
1

H
K
x +
1

_
m
K
= 0 (5.74)
The polynomial on the left side of (5.74) is of the form
p(x) := x
3
+x
2
+x + = 0. (5.75)
This cubic polynomial was analyzed by Hansen and Fabiano, and they proved the following
(See Theorem 2 in [9]):
Theorem 5.6. If (1, 9), then p(x) has exactly one negative root, and the other two roots
are non-real complex roots which occur as a complex conjugate pair, both with negative real
part. Moreover, the argument of the non-real eigenvalue in the upper half of the complex
plane is maximized precisely when

=
3/4
, (5.76)
and the optimal angle is

= tan
1
_
_
_
(3
1/2
)(1 +
1/2
)

1/2
1
_
_
. (5.77)
According to this theorem, equation (5.74) has exactly one negative real solution when
(K +

H
T
P
1

H)/K (1, 9),
66
(notice that this corresponds to the condition

H
T
P
1

H/K < 8) and the argument of the
solution in the second quadrant of the complex plane is optimized when
1

_
m
K
=
_
K +

H
T
P
1

H
K
_
3/4
.
This implies
=
m
1/2
K
1/4
(K +

H
T
P
1
H)
3/4
. (5.78)
Next, we use the substitution
t =
1

_
m
K
along with the value of in (5.78) to rewrite (5.74) as
x
3
+tx
2
+t
4/3
x +t = 0. (5.79)
For 1 t

27 the real root of (5.79) is t

1/3
, and the complex roots are
1
2
t
1/3
_
(1 t
2/3
) i
_
3 + 2t
2/3
t
4/3
_
. (5.80)
Furthermore, direct multiplication will verify that if x is a complex root of (5.79) then
x(t +x)
2
= t(t
2/3
+ 1).
Upon application of the reverse substitutions we nd that if y
c
is a complex root of (5.63) then
y
c
(1 +y
c
)
2
=

K
m

__
K +

H
T
P
1
H +

K
_
. (5.81)
Hence, the condition (5.72) has been veried. The condition that t <

27 translates to the
condition that

H
T
P
1

H/K < 8 given in the hypothesis.
5.4.3 Showing that the critical point is optimal
To show that this critical point is optimal (i.e. not a saddle point), we consider the second
derivative of (5.63) with respect to in the direction of a real diagonal matrix M. This leads
to the following:
2my
c
y
tt
c
+ 2m(y
t
c
)
2
+ 2

H
T
Q
1
Q
t
Q
1
Q
t
Q
1

H

H
T
Q
1
Q
tt
Q
1

H = 0.
67
Using the denition of Q, one arrives at
2m
y
tt
c
y
c
+ 2m
_
y
t
c
y
c
_
2
+
2
y
c

H
T
(I + Py
c
)
1

1
M +
y
t
c
y
c
I
_
(I +Py
c
)
1
_

1
M +
y
t
c
y
c
I
_
(I +Py
c
)
1

H
+
1
y
c

H
T
(I + Py
c
)
1

_
y
tt
c
y
c
I + 2
1
M
_
y
t
c
y
c
_
2
_

1
M +
y
t
c
y
c
I
_
2
_
(I +Py
c
)
1

H = 0.
Since (I +Py
c
)
1
I = (I +Py
c
)
1
Py
c
, the above can be written as
2m
y
tt
c
y
c
+ 2m
_
y
t
c
y
c
_
2

2
y
c

H
T
(I + Py
c
)
1
_
M +
y
t
c
y
c

_
(I +Py
c
)
1
Py
c
_
M +
y
t
c
y
c

_
(I +Py
c
)
1

H
+
y
tt
c
y
2
c

H
T
(I + Py
c
)
1
(I +Py
c
)
1

H +
2y
t
c
y
2
c

H
T
(I + Py
c
)
1
M(I +Py
c
)
1

H = 0.
Evaluating this equation at =

and applying (5.70) and (5.69), one obtains

_
2my
c
(1 +y
c
)
2
+

H
T
P
1

H
_
y
tt
c
y
c
+ 2my
c
(1 +y
c
)
2
_
y
t
c
y
c
_
2
+ 2

H
T
M

H
y
t
c
y
c
= 2y
c

H
T
_
M +
y
t
c
y
c

_
(I +Py
c
)
1
P
_
M +
y
t
c
y
c

_

H. (5.82)
Expanding the right side of (5.82) and applying (5.70) and (5.69) once again, we rewrite (5.82)
as follows:
_
2my
c
(1 +y
c
)
2
+

H
T
P
1

H
_
y
tt
c
y
c
+ 2my
c
(1 +y
c
)
2
_
y
t
c
y
c
_
2
+ 2

H
T
M

H
y
t
c
y
c
= 2y
c

H
T
M(I +P

y
c
)
1
PM

H +
4y
c

H
T
M

H
1 +y
c

y
t
c
y
c
+
2y
c

H
T

H
1 +y
c

_
y
t
c
y
c
_
2
. (5.83)
Next, dene a weighted inner product u, v)
P
on R
m
by
u, v)
P
= u
T
P
1
v, u, v R
m
.
Notice that this is a valid inner product because the matrix P
1
is positive denite. In addition,
P

is symmetric with respect to this inner product, and it is positive denite. Hence, the
spectral theorem implies that P

has a sequence of eigenvalues

k

m
k=1
(all of which are
positive) with a corresponding sequence of orthogonal eigenvectors
k

m
k=1
that span the
space R
m
, < , >
P
. Notice that for all i, j = 1, 2, , m, we have

T
i

j
= P

i
,
j
)
P
=
i

i
,
j
)
P
=
i

(i,j)
|
i
|
2
P
. (5.84)
68
We will denote
1
= and
1
=

H. Letting [H] = diag (

H) (note that [H] is invertible since

all entries of

H are positive) and [
k
] = diag (
k
), dene the matrix M
k
as follows:
M
k
:= [H]
1
[
k
]

; k = 1, 2, , m.
Notice that M
k
is diagonal, M
1
=

, and
M
k

H =

k
, for k = 1, 2, m. (5.85)
If we consider real diagonal matrix variations M = M
k
, and apply (5.84) and (5.85) to equation
(5.83), we obtain
_
2my
c
(1 +y
c
)
2
+

H
T
P
1

H
_
y
tt
c
y
c
+ 2my
c
(1 +y
c
)
2
_
y
t
c
y
c
_
2
+ 2
1
,
k
)
P
y
t
c
y
c
=
2y
c

2
k
|
k
|
2
P
1 +
k
y
c
+
4y
c

1
,
k
)
P
1 +y
c

y
t
c
y
c
+
2y
c

2
|
1
|
2
P
1 +y
c

_
y
t
c
y
c
_
2
(5.86)
If y = Re
i
, we can use (5.65) and compute
y
tt
c
y
c
=
_
y
t
c
y
c

R
t
R
__
y
t
c
y
c
+
R
t
R
_
+
R
tt
R
+i
tt
.
But when =

, (5.65) implies that y

t
c
/y
c
= R
t
/R, and so
y
tt
c
y
c
=
R
tt
R
+i
tt
.
Therefore, if

is a critical point, then

tt
= Im
_
y
tt
c
y
c
_
. Thus, we take the imaginary part of
(5.86) and obtain the following:
_
2my
c
(1 +y
c
)
2
+

H
T
P
1

H
_

tt
= 2
2
k
|
k
|
2
P
Im
_
y
c
1 +
k
y
c
_
+ 4
2

1
,
k
)
P
Im
_
y
c
1 +y
c

_
y
t
c
y
c
+ 2
2
|
1
|
2
P
Im
_
y
c
1 +y
c

__
y
t
c
y
c
_
2
. (5.87)
We next consider 2 cases:
Case 1: Suppose k = 1. Then (5.87) becomes
_
2my
c
(1 +y
c
)
2
+

H
T
P
1

H
_

tt
= 2
2
|
1
|
2
P
Im
_
y
c
1 +y
c
__
1 +
y
t
c
y
c
_
2
.
If y
c
= a +ib, then
Im
_
y
c
1 +y
c
_
=
b
(1 +a)
2
+ (b)
2
. (5.88)
69
Thus, we have
_
2my
c
(1 +y
c
)
2
+

H
T
P
1

H
_

tt
=
2b
2
|
1
|
2
P
_
1 +
y

c
yc
_
2
[(1 +a)
2
+ (b)
2
]
. (5.89)
Case 2: Suppose k = 2, 3, , m. Then
1
,
k
) = 0 and (5.87) becomes
_
2my
c
(1 +y
c
)
2
+

H
T
P
1

H
_

tt
= 2
2
k
|
k
|
2
P
Im
_
y
c
1 +
k
y
c
_
+ 2
2
|
1
|
2
P
Im
_
y
c
1 +y
c

__
y
t
c
y
c
_
2
.
Letting y
c
= a +ib again, we obtain the following by (5.88):
_
2my
c
(1 +y
c
)
2
+

H
T
P
1

H
_

tt
= 2b
_

2
k
|
k
|
2
P
(1 +a
k
)
2
+ (b
k
)
2
+

2
|
1
|
2
P
(1 +a)
2
+ (b)
2
_
y
t
c
y
c
_
2
_
.
(5.90)
Our nal step is to determine the sign of 2my
c
(1 +y
c
)
2
+

H
T
P
1

H. According to (5.81),
2my
c
(1 +y
c
)
2
+

H
T
P
1

H = 2

K
__
K +

H
T
P
1
H +

K
_
+

H
T
P
1

H
=
__
K +

H
T
P
1
H +

K
___
K +

H
T
P
1
H 3

K
_
But the condition

H
T
P
1

H
K
< 8 implies that
_
K +

H
T
P
1
H < 3

K. Therefore
2my
c
(1 +y
c
)
2
+

H
T
P
1

H is negative. (5.91)
Furthermore,
y

c
yc
,= 1 when M =

by (5.71), so the quantity

_
1 +
y

c
yc
_
2
in (5.89) is positive.
Finally, since y
c
is in the upper half of the complex plane, we have that b > 0. Hence it follows
from (5.89), (5.90), and (5.91) that
tt
is negative at

for all matrix variations M

k
in (5.85).
Moreover, any arbitrary real diagonal matrix M can be expressed as a nite (at most m terms)
linear combination of the M
k
in (5.85). Therefore,
tt
is negative for all real diagonal matrix
variations. Hence =

maximizes .
5.4.4 Recovering the optimal damping coecients for analyticity
Dene [H] to be the diagonal matrix consisting of the entries of

H and

be the vector of
the diagonal entries of

. From Theorem 5.5 we have

= [H]
1
P
1

H. (5.92)
70
Then, we can recover the optimal damping coecients,

G

E
, using the change of variables and
obtain

E
= diag(h
E

),
Note that since P
1
is a positive matrix and all entries of

H are positive constants, all entries
of

G

E
are positive. Moreover, one can explicitly compute the angle of analyticity using (5.77),
and obtain:

= tan
1
_
_
_
_
_
_
3

K
_
K +

H
T
P
1
H
__

K +
_
K +

H
T
P
1
H
_
_
K +

H
T
P
1
H

K
_
_
_
_
.
5.5 Exponential Stability and Optimal Decay Rate
In this section we show that the multi-layer Mead-Markus beam is exponentially stable.
Then we consider an optimal problem similar in principle to the one in the last section, only
this time we will try to nd damping parameters that achieve the optimal decay rate for the
semigroup e
,t
. For simplicity, we look at this optimal problem in the three-layer case.
5.5.1 Exponential stability of the semigroup e
,t
Since we have that / generates an analytic semigroup, Theorem 2.5 implies that the
spectrum-determined growth condition holds. Since T(/) is compactly embedded in H, the
spectrum of / consists only of eigenvalues. Hence, if we let denote the growth rate of the
semigroup e
,t
, then
= supRe : is an eigenvalue of /. (5.93)
Furthermore, one can repeat the analysis used in Lemma 5.5 to prove that all of the eigenvalues
of / are in the open, left half of the complex plane, except for possibly the zero eigenvalue.
Our next step is to show that 0 is not an eigenvalue.
Lemma 5.6. = 0 is not an eigenvalue of /.
71
Proof. Suppose = 0 is an eigenvalue with corresponding eigenvector z = [w, v,
E
]
T
. Then
w, v, and
E
must satisfy
v = 0 (5.94)
Kw
tttt
+

H
T
P
1
h
E
_

ttt
E


Nw
tttt
_
= 0 (5.95)
P
1
h
E
_

tt
E


Nw
ttt
_
G
E

E
= 0 (5.96)
Taking the L
2
-inner product of (5.95) with

H
T
P
1
h
E
_

t
E


Nw
tt
_
leads to
_
Kw
tttt
,

H
T
P
1
h
E
_

t
E


Nw
tt
__
+
_

H
T
P
1
h
E
_

ttt
E


Nw
tttt
_
,

H
T
P
1
h
E
_

t
E


Nw
tt
__
= 0
Integration by parts and the boundary conditions w
tt
= 0,
t
E
= 0 at x = 0, L imply
_
Kw
ttt
,

H
T
P
1
h
E
_

tt
E


Nw
ttt
__

_
_
_

H
T
P
1
h
E
_

tt
E


Nw
ttt
__
_
_
2
L
2
(0,L)
= 0. (5.97)
Next, take the (L
2
)
m
-inner product of (5.96) (multiply on the left) by K

Hw
ttt
and obtain
_
K

Hw
ttt
, P
1
h
E
_

tt
E


Nw
ttt
__
K

Hw
ttt
, G
E

E
) = 0.
Integration by parts on the second term on the left side leads to
_
Kw
ttt
,

H
T
P
1
h
E
_

tt
E


Nw
ttt
__
+K

Hw
tt
, G
E

t
E
) = 0. (5.98)
Subtract (5.97) from (5.98) to get
_
_
_

H
T
P
1
h
E
_

tt
E


Nw
ttt
__
_
_
2
L
2
(0,L)
+K

Hw
tt
, G
E

t
E
) = 0. (5.99)
Next take the (L
2
)
m
-inner product of (5.96) with KPG
E

E
and get
_
P
1
h
E
_

tt
E


Nw
ttt
_
, KPG
E

E
_
G
E

E
, KPG
E

E
) = 0.
Apply once again integrations by parts and the boundary conditions on
t
E
and w
tt
to get
Kh
E

t
E
, G
E

t
E
) +K

Hw
tt
, G
E

t
E
) KG
E

E
, PG
E

E
) = 0. (5.100)
72
Subtract (5.100) from (5.99) to get
_
_
_

H
T
P
1
h
E
_

tt
E


Nw
ttt
__
_
_
2
L
2
(0,L)
+K|h
1/2
E
G
1/2
E

t
E
|
2
(L
2
(0,L))
m
+K|P
1/2
G
E

E
|
2
(L
2
(0,L))
m
= 0.
Notice that this is valid since h
E
and G
E
are diagonal matrices whose entries on the diagonal
are positive, and P is a positive denite, symmetric matrix. Since K is a positive constant
and the matrices h
E
, G
E
, and P
1/2
are invertible, we have the following:

H
T
P
1
h
E
_

tt
E


Nw
ttt
_
= 0 (5.101)

t
E
= 0 (5.102)

E
= 0. (5.103)
Substitution of (5.102) into (5.101) implies

H
T
P
1

Hw
ttt
= 0.
But since

H
T
P
1

H is a positive constant, we have w
ttt
= 0, and so w
tt
is constant. But the
boundary conditions w
tt
(0) = w
tt
(L) = 0 imply that
w
tt
= 0.
Finally, the boundary conditions along with inequality (5.11) imply that
w = 0. (5.104)
Therefore, it follows from (5.94), (5.103), and (5.104) that z = 0. Hence = 0 cannot be an
eigenvalue of /.
Theorem 5.7. / generates an exponentially stable semigroup on H.
Proof. The results of the previous lemma along with Lemma 5.5 and the dissipation relation
(5.23) imply that all eigenvalues of / are in the open, left-half of the complex plane. Thus ,
as dened in (5.93) is negative. Moreover, sup
Re 0
|(I /)
1
| < , since the closed right half
plane is contained in the resolvent of /. Hence it follows from Theorem 2.7 that
|e
,t
| Me
[[t
, M 1.
73
5.5.2 Optimal damping for decay rate in a 3-Layer Mead-Markus Beam
We are now interested in the following optimization problem
Maximize : diagonal matrices in R
mm
(5.105)
No expression for the optimal growth (decay) rate is known for the general multilayer case. It
is possible, however, to solve (5.105) in closed form for the 3-layer case.
In nding the decay rate, it will be helpful once again to look at the behavior of the roots of
the polynomial
p(x) = x
3
+x
2
+x +,
which was already dened in (5.75). Dene a() to be the negative real root of p(x) and b()
to be the negative real part of the complex conjugate pair of roots of p(x). It is true that a
and b are smooth functions of . In the following theorem, we locate the value of , say

that solves the optimization problem:

nd sup
>0
Re(r) : p(r) = 0. (5.106)
Theorem 5.8. Let (1, 9) be xed. Then
i.) If (1, 5], then max
>0
[min(a(), b())] =
1
4
, and it is attained at

=
+ 1
2
.
ii.) If (5, 9), then max
>0
[min(a(), b())] =

3
2
, and it is attained at

= 3
_
3
2
.
We will prove this theorem at the end of this chapter. In the meantime, we will use this result
to nd the optimal decay rate.
The 3-layer Mead Markus beam equations take the form
_

_
m w +Kw
tttt
Nh
2
_
G
2

t
+

G
2

t
_
= 0 on (0, L) (0, )
h
2
Nw
ttt
h
2

tt
+P
_
G
2
+

G
2

_
= 0 on (0, L) (0, )
w = w
tt
=
t
= 0 for x = 0, L, t > 0
(5.107)
If we insert the modal solutions
w = e
s
k
t
sin
k
x, = ce
s
k
t
cos
k
x;
k
= k/L,
74
into (5.107), we obtain the characteristic equation
_

_
ms
2
k
+K
4
k
+Nh
2
_

k
G
2
+s
k

k

G
2
_
c = 0
c
2
k
h
2
+P
_
G
2
+s
k

G
2
_
c =
3
k
h
2
N
(5.108)
Letting y be dened as before, (5.108) can be written as
_

_
my
2
+K +
Nh
2

k
_
G
2

2
k
+

G
2
y
_
c = 0
c
_
h
2
+
PG
2

2
k
+P

G
2
y
_
= h
2
N
(5.109)
Solving the second line of (5.109) for c and substituting into the rst line yields
my
2
+K +
Nh
2
_
G
2
/
2
k
+

G
2
y
_
P
1
h
2
N
P
1
h
2
+G
2
/
2
k
+

G
2
y
= 0.
Thus we obtain the characteristic equation
m

G
2
y
3
+m
_
P
1
h
2
+G
2
/
2
k
_
y
2
+
_
K

G
2
+Nh
2

G
2
P
1
h
2
N
_
y
+
_
K
_
P
1
h
2
+G
2
/
2
k
_
+
Nh
2
G
2
P
1
h
2
N

2
k
_
= 0. (5.110)
Introduce the change of variables
y =
_
K
m
x, =
_
m
K

G
1
2
P
1
h
2
, = 1 +
Nh
2
P
1
h
2
N
K
. (5.111)
Then (5.110) can be written as
x
3
+
_
1 +
G
2
P

2
k
h
2
_
x
2
+x +
_
1 +
G
2
P

2
k
h
2
+
Nh
2
G
2
N

2
k
K
_
= 0. (5.112)
Introduce the substitutions

k
=
_
1 +
G
2
P

2
k
h
2
_
,
k
=
Nh
2
G
2
h
2
N
K(
2
k
h
2
+G
2
P)
, (5.113)
and rewrite (5.112) as
x
3
+
k
x
2
+x +
k
(1 +
k
) = 0. (5.114)
A nal change in variables
x = t
_
1 +
k
(5.115)
75
leads to
q
k
(t) := t
3
+

k

1 +
k
t
2
+

1 +
k
t +

k

1 +
k
= 0. (5.116)
Note that this cubic polynomial is of the form p(t) in (5.75). Since the eigenvalue s
1
of / is
closest to the imaginary axis, we will only be concerned with the rst mode (k = 1) in order
to determine the optimal decay rate. To solve this problem, we will utilize Theorem 5.8 to
optimize the real part of the roots of the cubic polynomial q
1
(t).
If

1 +
1
(1, 5], then the optimal real part for a root of q
1
is

1
4
_

1 +
1
1
_
.
Using (5.111), this implies
sup

G
2
>0
Re t : q
1
(t) = 0 =
(h
2
2
N
2
PK
1
)
4PK(1 +
1
)
(5.117)
By applying the reverse substitutions in (5.111) and (5.115), this leads to
sup

G
2
>0
Re s
1
=

2
1
(h
2
2
N
2
PK
1
)
4P
_
mK(1 +
1
)
(5.118)
In this case, (5.118) is achieved when

1 +
1
=
1
2
_

1 +
1
+ 1
_
.
Upon applying the reverse substitutions in (5.111) and (5.113), one nds

G
2
=
2(
2
1
h
2
+G
2
P)
_
mK(1 +
1
)

2
(h
2
2
N
2
+ 2PK +PK
1
)
. (5.119)
If

1 +
1
(5, 9), then the optimal real part for a root of q
1
is

1
2
_

1 +
1
3
_
.
Using (5.111), this implies
sup

G
2
>0
Re t : q
1
(t) = 0 =

h
2
2
N
2
2PK 3PK
1
2PK(1 +
1
)
(5.120)
76
By applying the reverse substitutions in (5.111) and (5.115), this leads to
sup

G
2
>0
Re s
1
=

2
1
_
h
2
2
N
2
2PK 3PK
1

2mP
(5.121)
In this case, (5.121) is achieved when

1 +
1
= 3

h
2
2
N
2
2PK 3PK
1
2PK(1 +
1
)
,
using (5.120). Upon applying the reverse substitutions in (5.111) and (5.113), one nds

G
2
=
(
2
1
h
2
+G
2
P)

2m
3
2
1
_
P(h
2
2
N
2
2PK 3PK
1
)
. (5.122)
We have thus proven the following result.
Theorem 5.9. Let /

G
2
denote the dependence of / on

G
2
, and let

denote the decay rate

of the semigroup generated by /

2
. Then we have the following:
(i) If
h
2
2
N
2
PK
4 + 5
1
, then the optimal decay rate of / is

2
1
(h
2
2
N
2
PK
1
)
4P
_
mK(1 +
1
)
,
where

2
=
2(
2
1
h
2
+G
2
P)
_
mK(1 +
1
)

2
1
(h
2
2
N
2
+ 2PK +PK
1
)
.
(ii) If
h
2
2
N
2
PK
(4 + 5
1
, 8 + 9
1
), then the optimal decay rate of / is

2
1
_
h
2
2
N
2
2PK 3PK
1

2mP
,
where

2
=
(
2
1
h
2
+G
2
P)

2m
3
2
1
_
P(h
2
2
N
2
2PK 3PK
1
)
.
Furthermore, the semigroup T (t) generated by /

2
is exponentially stable with
|T (t)| Me
[

[t
.
77
5.6 A 3-layer Mead-Markus Beam example
Suppose we have a sandwich beam with length L = 10 and the following quantities dened:
h = diag (1.1, 0.5, 0.9), E = diag (1.4 10
7
, 10000, 1.2 10
7
), m = 1, G
2
= 100
Then, we compute P and N:
P = [1 1](h
O
E
O
)
1
_

_
1
1
_

_ = 1.57528 10
7
, N = 2[0.5 0.5]h
O
_

_
1
1
_

_ + 1 = 3.
Therefore,
h
2
2
N
2
P
= 1.4283 10
7
Using the denition of K,
K =
1
12
_
1.4 10
7
(1.1)
3
+ 1.2 10
7
(0.9)
3
_
= 2.2818 10
6
,
and hence
h
2
2
N
2
PK
= 6.2595. (5.123)
5.6.1 Optimal damping coecient for angle of analyticity
Theorem 5.6 and (5.111) imply

= (1 + 6.2595)
3/4
= 4.42264, (5.124)
Using the reverse substitution in (5.111), we recover the following optimal value of

G
2
from
(5.124)

2
= 475.105 (optimal damping for angle of analyticity). (5.125)
In addition, we have by (5.77) that the optimal damping angle is

= tan
1
_
_
(3 (7.2595)
1/2
)(1 + (7.2595)
1/2
)
(7.2595)
1/2
1
_
= 2.58144 rad 147.9

. (5.126)
Figure 5.4, shows a plot of the eigenvalues of /
0
when

G
2
=

G

2
= 475.105 (marked with
o). Notice that the non-real eigenvalues with positive real part coincide with the polar line
=

= 2.58144. Therefore, the spectrum of /

0
when

G
2
= 475.105 is contained in the set

. The sector of analyticity

is also shown.
78
Figure 5.4 Eigenvalues of /
0
when

G
2
=

G

2
, and the sets

and

.
5.6.2 Optimal damping coecient for decay rate
According to (5.113)

1
=
225
2.2818 10
6
(
2
/100 0.5 + 0.000016)
0.001998.
Since 4+5
1
4.01, we have that h
2
2
N
2
/PK (4+5
1
, 8+9
1
) by (5.123). Hence, Theorem
5.9 implies

2
1
_
h
2
2
N
2
2PK 3PK
1

2mP
217.421,
where

2
=
(
2
1
h
2
+G
2
P)

2m
3
2
1
_
P(h
2
2
N
2
2PK 3PK
1
)
480.428 (optimal damping for decay rate).
(5.127)
Furthermore, the semigroup T (t) = exp(/

2
t) satises
|T (t)| Me
217.421t
; M > 1.
In Figure 5.5, we graphically verify that Theorem 5.9 holds for our three-layer beam example.
The eigenvalues of the semigroup /

G
2
are plotted in the rst mode (k=1) as varies. The o
marks indicate exactly where s
1
is for /

2
. According to (5.127), the optimal decay rate will
79
occur when = 4.3736 (

G

2
= 480.428

= 4.3736 by (5.111)). The + marks indicate the

location of s
1
for /

G
2
for = 3.0, 3.5, 4.0. Notice that the optimal real part for each of these
choices of will coincide with the negative real eigenvalues. Each of these are to the right of
the vertical line Re x = 217.421 : x C. The x marks indicate the location of s
1
for
= 4.5, 5.0, 5.5. The optimal real part for these choices of will coincide with the real part
of the complex conjugate pair of eigenvalues, which are also to the right of the vertical line
Re x = 217.421 : x C. Hence, if we choose

G

2
= 480.428, the spectrum of /

2
is on or
left of the vertical line. Moreover, (as is clear from Figure 5.5) this choice for

G

2
forces the
supremum of the real part of the eigenvalues of / to be as far into the left half of the complex
plane as possible.
Figure 5.5 The spectral bound for /

2
.
Remark 5.7. Looking at Figure 5.5, it appears that the optimal decay rate occurs precisely
when the real part of the complex conjugate pair coincides with the negative real eigenvalue. In
fact, this is true because for our example,
h
2
2
N
2
PK
(4 + 5
1
, 8 + 9
1
). The fact that the real
part of the complex conjugate pair coincide with the negative real eigenvalue in this case will
become more apparent in the next section as we prove Theorem 5.8 (see Lemma 5.9).
80
Remark 5.8. It is interesting to note that the values of

G

2
in (5.125) and (5.127) are nearly
the same. Although we currently have no optimal damping result for a general multilayer beam,
the results we obtained for the three layer example suggest that the optimal choice of damping
for the analyticity of a multilayer beam must be close to the choice of optimal damping needed
for the decay rate of a multilayer beam.
5.7 Proof of Theorem 5.8
If a() is the negative real root of p(x) and b() is the negative real part of the complex
conjugate pair of roots of p(x), we can write p(x) as follows:
p(x) = (x+a)(x+b +ic)(x+b ic) = x
3
+(a +2b)x
2
+(b
2
+2ab +c
2
)x+a(a
2
+c
2
). (5.128)
Throughout this proof, the prime symbol will denote dierentiation with respect to .
In order to prove Theorem 5.8, it will be helpful to rst prove 3 lemmas.
Lemma 5.7. Suppose > 1. Then
max
>0
b() =
1
4
,
and it is attained when =
+ 1
2
. Moreover,
a() = a
t
() = 1 when =
+ 1
2
. (5.129)
Proof. From the second coecient of (5.75) and (5.128), we know that
= a + 2b. (5.130)
By the third coecient, we have 2ab = b
2
+c
2
. Using this, along with the fourth coecient,
we get that
= a( 2ab). (5.131)
81
Next, we seek a value of such that b
t
() = 0. Dierentiating both sides of (5.130) with
respect to gives 1 = a
t
+ 2b
t
. So if b
t
= 0, we have
a
t
= 1. (5.132)
Dierentiating both sides of (5.131) with respect to and using the result in (5.132) leads to
1 = 4ab. This implies
ab =
1
4
. (5.133)
Substitution of (5.133) into (5.131) implies
a =
2
+ 1
. (5.134)
By substituting this result into (5.133) and solving for b, we obtain
b =
( 1)( + 1)
8
. (5.135)
Next, substitute equations (5.134) and (5.135) into (5.130) to obtain the following equation:
4
2
(1 ) = (1 )( + 1)
2
.
Since > 1, then 4
2
= ( + 1)
2
. Therefore
=
+ 1
2
. (5.136)
Hence, we have determined that b
t
() = 0 when =
+ 1
2
.
Next, we substitute the result of (5.136) into (5.134) and obtain
a = 1. (5.137)
Finally, we substitute this result into (5.133) and obtain
b =
1
4
. (5.138)
Now as 0
+
, p(x) x(x
2
+ ), and so the roots approach 0 and i

. Thus, b 0 as
0
+
. As , p(x) (x
2
+ 1), and so the complex roots approach i. Thus, b 0
as . Therefore, b attains a maximum value of
1
4
at =
+ 1
2
according to (5.136)
and (5.138). Furthermore, equations (5.132) and (5.137) give (5.129).
82
Lemma 5.8. a
t
() > 0 for all > 0 and lim

a() = .
Proof. We rst show that a
t
() ,= 0 for all > 0. Suppose on the contrary a
t
(c) = 0 for some
c > 0. If we dierentiate (5.130) with respect to and evaluate at c, we get
b
t
(c) =
1
2
. (5.139)
Then dierentiating (5.131) with respect to and evaluating at c implies:
1 = a(c)[2a(c)b
t
(c)]
By (5.139), this implies [a(c)]
2
= 1, which is absurd. Thus,
a
t
() ,= 0 for all > 0. (5.140)
We next show that a
t
() > 0 for all > 0. Suppose there is a c > 0 so that a
t
() < 0. But
Lemma 5.7 tells us that a
t
= 1 > 0 at =
+ 1
2
. So there exists a t between c and
+ 1
2
such
that a
t
(t) = 0. But this contradicts (5.140). Thus a
t
() > 0 for all > 0. Finally, 2b = a
by (5.130), and b 0 as . Therefore, a as .
Lemma 5.9. If (1, 3], then a() > b() for all > 0. If > 3, then
a() = b() =
_
3
2
when = 3
_
3
2
.
Proof. Suppose a = b for some > 0. Then (5.130) implies
= 3a. (5.141)
Similarly by (5.132), we have
= a[ 2a
2
]. (5.142)
Equations (5.141) and (5.142) together imply that 3 = 2a
2
since a > 0. Hence,
a
2
=
3
2
. (5.143)
If (1, 3), then (5.143) has no real solution. Moreover, if = 3, (5.143) implies that a = 0,
but this contradicts that a > 0 (Lemma 5.8). Thus
a() ,= b() for all > 0 and for any (1, 3]. (5.144)
83
To show a() > b() for all > 0 for any xed (1, 3], suppose by way of contradiction
there exists a c > 0 such that a(c) < b(c). Since b 0 and a as , then there is
an M > 0 such that a(M) > b(M). So there is a t between c and M such that a(t) = b(t),
contradicting (5.144). Thus a() > b() for all > 0 when (1, 3].
If > 3, then we can solve (5.143) and obtain
a = b =

3
2
. (5.145)
Equations (5.141) and (5.145) then imply that a and b are equal whenever
= 3

3
2
. (5.146)
Proof of Theorem 5.8: First of all, since b 0 when 0
+
and , then min(a, b) 0
when 0
+
and . Furthermore, Lemma 5.8 implies that a monotonically as
. So the only possible critical points of min(a, b) are (i) where a crosses b and (ii) where
b attains its maximum. We consider 3 cases.
Case 1: If (1, 3], Lemma 5.9 implies min(a, b) = b. Thus by Lemma 5.7,
max
>0
[min(a(), b())] =
1
4
is attained at =
+ 1
2
.
Case 2: If (3, 5], then Lemmas 5.7 and 5.9 give us the critical points
1
=
+ 1
2
and

2
= 3
_
3
2
respectively. Since
1
4
1 for all (3, 5], then Lemma 5.7 implies
min(a(
1
), b(
1
)) =
1
4
. In addition, min(a(
2
), b(
2
)) =
_
3
2
by Lemma 5.9. Since
_
3
2

1
4
for all (3, 5], we have
max
>0
[min(a(), b())] =
1
4
is attained at =
+ 1
2
.
Case3: If (5, 9), then Lemmas 5.7 and 5.9 imply the critical points are
1
=
+ 1
2
and
2
= 3
_
3
2
respectively. Since
1
4
> 1 for all (5, 9), then Lemma 5.7 im-
plies min(a(
1
), b(
1
)) = 1. In addition, min(a(
2
), b(
2
)) =
_
3
2
by Lemma 5.9. Since
84
_
3
2
> 1 for (5, 9), we have
max
>0
[min(a(), b())] =

3
2
is attained at = 3

3
2
.
85
CHAPTER 6. Analyticity of a Multilayer Mead-Markus Plate: A Direct
Proof
6.1 Introduction
One of the main results in the previous chapter is the analyticity of the multilayer Mead-
Markus beam. We now consider an analogous model for a plate, and show that the semigroup
associated with a multilayer Mead-Markus plate is analytic. Recall from Chapter 4 that the
two-dimensional set describes the shape of the face of each plate layer, and is the boundary
of . In addition, we will make use of the forms and
O
, and the operators L, L
O
, B, and
B
O
(see Chapter 4 for appropriate denitions). We are interested in the equations of motion
for the multilayer Mead-Markus plate (4.26). Many boundary conditions could be considered.
One form of hinged boundary conditions is obtained under the assumption that the lateral
stress and bending moment vanish with zero transverse displacement imposed along :
B

w n = w = 0, B
O
v
O
= 0 on (6.1)
We can also consider clamped boundary conditions:
w = 0,
w
n
= 0,
E
= 0 on . (6.2)
For the sake of simplicity, we consider only the clamped boundary conditions (6.2) in this
chapter (we consider both hinged and clamped BCs in Appendix A) So we consider the
system (4.26), (6.2):
_

_
m w +K
2
w div

N
T
h
E
_
G
E

E
+

G
E

E
_
= 0 in R
+
12h
O
D
O
L
O
v
O
+B
T
_
G
E

E
+

G
E

E
_
= 0 in R
+
w = 0, w = 0,
E
= 0 on
. (6.3)
86
Notice that since w = 0 and
w
n
= 0 on , we have w = 0 on .
In Section 6.2, we formulate (6.3) as a semigroup problem with generator /, and we show
that / generates a C
0
semigroup of contractions. The main result, the analyticity of the
Mead-Markus plate, is shown in Section 6.3. Our proof relies on establishing the estimate
|R(i, /)| C/[[ for all real : [[ 1. (6.4)
In Appendix A, we prove this result by a contradiction argument similar to the one by Hansen
and Liu [18]. In the present chapter, we prove this result by a direct argument similar to the
one used by Hansen and Lasiecka [17].
6.2 Semigroup Formulation:
Dene y := w. We formulate the semigroup to this problem using the state variables w, y,
and
E
. As is seen in [15], there is diculty in solving (6.3) for

E
. In order to do this, we
assume that the Poisson Ratio in each of the sti layers is the same, i.e.
:=
1
=
3
= =
2m+1
.
In this case, we have the following (see Appendix B for proof):
BL
O
v
O
= L
E
Bv
O
, (6.5)
where L
E
is dened as
L
E
= I
m
L

,
and I
m
is the identity operator on R
m
. Analogously, we dene
B
E
= I
m
B

.
Furthermore, dene the quadratic form
E
as follows:

E
(
E
,

E
) = B
E

E
,

E
) L
E

E
,

E
)

. (6.6)
We also have the following analogues to Corollaries 4.1, 4.2, and 4.3:
87
Corollary 6.1. Suppose is a bounded open set with a suciently regular boundary. Also
suppose
E
is such that
i
1
and
i
2
belong to L
2
() for all i = 2, 4, , 2m. then for all > 0,
there exists a constant c
E,
> 0 depending only on and such that

O
(
E
;
E
) +|
E
|
2
L
2
E
()
c
E,
|
E
|
2
H
1
E
()
. (6.7)
Corollary 6.2.
E
is coercive on H
1
E,0
().
Corollary 6.3. If
E
and

E
belong to H
1
E
(), then

E
(
E
,

E
) |
E
|
H
1
E
()
|

E
|
H
1
E
()
. (6.8)
Dene the matrix P by
P =
1
12
BD
1
O
h
1
O
B
T
.
It is shown in Appendix B that P is a positive denite, symmetric, and invertible M-matrix.
If we multiply the last line of (6.3) by
1
12
BD
1
O
h
1
O
, we obtain the following:
_

_
m w +K
2
w div

N
T
_
h
E
P
1
L
E
Bv
O

= 0 in R
+
G
E

E
+

G
E

E
P
1
L
E
Bv
O
= 0 in R
+
w = 0, w = 0,
E
= 0 on
(6.9)
Note that the above formulation is completely independent of v
O
since the term Bv
O
is easily
eliminated using (4.19).
Remark 6.1. Similar to Chapter 5, the multiplication of the last line of (6.3) by
1
12
BD
1
O
h
1
O
reduces the number of rows in the equation by one. If we multiply by

1
T
O
, we obtain
12

1
O
h
O
D
O
L
O
v
O
= 0, which implies
12L

i
1
O
h
O
D
O
v
O
= 0.
Since v
O
= 0 on , we have
L

i
1
O
h
O
D
O
v
O
,

1
O
h
O
D
O
v
O
) =

i
(

1
O
h
O
D
O
v
O
,

1
O
h
O
D
O
v
O
) c
0
|

1
O
h
O
D
O
v
O
|
2
(H
1
())
2
,
88
by Korns Inequality (see Theorem 4.2). Therefore,

1
O
h
O
D
O
v
O
= 0. (6.10)
Hence the denition of B
C
in Remark 5.1 implies
B
C
v
O
=
_

_
h
E
(
E


Nw)

0
_

_
.
Therefore, (5.6) implies
v
O
= Sh
E
(
E


Nw). (6.11)
Recall from Chapter 4 that we dened the spaces H
s
E
(), H
s
O
(), H
s
E,0
(), and H
s
O,0
() for
s 0 (s = 0 corresponds to L
2
). Throughout this chapter, it will be convenient to use the
following notation:
[w[
s
:= |w|
H
s
()
, [w[
0
:= |w|
L
2
()
, [
E
[
s,E
:= |
E
|
H
s
E
()
, [v
O
[
s,O
:= |v
O
|
H
s
O
()
In terms of the state variables w, y = w, and
E
, we can rewrite the previous system as:
d
dt
_

_
w
y

E
_

_
= /
_

_
w
y

E
_

_
:=
_

_
y
1
m
_
K
2
w + div

N
T
_
h
E
P
1
L
E
Bv
O

G
1
E
_
G
E

E
+P
1
L
E
Bv
O
_
_

_
. (6.12)
The energy inner product associated with z = [w, y,
E
], z = [ w, y,

E
] is
z, z)
c
=
1
2
_
my, y)

+K
2
w, w)

+G
E
h
E

E
,

E
)

P
1
L
E
Bv
O
, B v
O
_

_
. (6.13)
Let c be [w, y,
E
]
T
H
2
() L
2
() H
1
E
(). We let H denote the closure in c of all
elements in c that satisfy the boundary conditions.
Remark 6.2. In (4.25), the term
O
(h
3
O
D
O

1
O
w,

1
O
w) is the same as K
2
w, w)

since
the Poisson ratio of the odd layers are the same (see (B.13) in Appendix B). Since v
O
= 0 on
89
, the term 12
O
(h
O
D
O
v
O
, v
O
) can be rewritten as follows:
12
O
(h
O
D
O
v
O
, v
O
) = 12L
E
h
O
D
O
v
O
, v
O
)
= 12h
O
D
O
L
E
D
O
v
O
, Sh
E
(
E


Nw)) by (6.11)
=
_
12P
1
1
12
B(h
O
D
O
)
1
B
T
S
T
h
O
D
O
L
E
D
O
v
O
, Bv
O
_
=

P
1
BL
O
v
O
, Bv
O
_

P
1
L
E
Bv
O
, Bv
O
_

by (6.5).
Hence (6.13) corresponds to the total energy of the plate (4.25).
Using an argument similar to the one done in Section 5.2, one nds that the topology induced
(6.13) is equivalent to the usual Sobolev topology on H. It is easy to bound (6.13) above
by C
_
[w[
2
2
+[y[
2
0
+[
E
[
2
1,E
_
. To show coercivity, the terms my, y)

and G
E
h
E

E
,
E
)

are
bounded below by c
1
_
[y[
2
0
+[
E
[
2
0,E
_
because G
E
and h
E
are positive matrices. To bound the
term K
2
w, w)

below by c
3
[w[
2
2
, we note that
2
w, w)

(w, w) (see Appendix B).

Since w (H
1
0
())
2
, Theorem 4.2 implies there is a c
2
> 0 such that
c
2
|w|
2
(H
1
())
2

(w, w).
But since w = 0 on , the Poincare Inequality implies there is a C
1
> 0 such that [w[
0

C
1
|w|
(L
2
())
2, and so there is a c
3
> 0 such that
c
3
[w[
2
2
K
2
w, w)

. (6.14)
Next we look for a bound involving [
E
[
1,E
. Since P
1
is a positive matrix, there is a c
4
> 0
such that
c
4
[L
E
Bv
O
, B v
O
)

P
1
L
E
Bv
O
, B v
O
_

.
But since Bv
O
= 0 on , the Greens formula (6.6) implies
L
E
Bv
O
, B v
O
)

=
E
(Bv
O
, Bv
O
).
But Bv
O
H
1
E,0
(), and so Corollary 6.2 implies that there is a c
5
> 0 such that
c
5
[Bv
O
[
2
0,E

E
(Bv
O
, Bv
O
).
90
Thus, (4.19) implies
c
4
c
5
[h
E
(
E


Nw)[
2
1,E

P
1
L
E
Bv
O
, B v
O
_

.
But since h
E
is a positive matrix, there is a c
6
> 0 such that
c
6
[
E


Nw[
2
1,E

P
1
L
E
Bv
O
, Bv
O
_

.
Hence, the reverse triangle inequality implies that for all
1
(0, c
6
],

1
_
[
E
[
1,E
[

Nw[
1,E
_

_
P
1
L
E
Bv
O
, Bv
O
)

1
[
E
[
1,E

_

1
(

N
T
N)|w|
(H
1
())
2
_
P
1
L
E
Bv
O
, Bv
O
). (6.15)
Add
_
K
2
|w|
(H
1
())
2 to both sides of (6.15) to obtain

1
[
E
[
1,E
+
_
_
K
2

_

1
(

N
T
N)
_
|w|
(H
1
())
2

_
P
1
L
E
Bv
O
, Bv
O
) +
_
K
2
|w|
(H
1
())
2

1
[
E
[
1,E
+
_
_
K
2

_

1
(

N
T
N)
_
|w|
(H
1
())
2

_
P
1
L
E
Bv
O
, Bv
O
) +
_
K
2
w, w), (6.16)
where the last line follows from the boundary conditions on w and the Poincare Inequality.
Fix
1
= min
_
c
6
,
K
2

N
T
N
_
. Then
_
K
2

_

1
(

N
T
N) 0, and (6.16) implies

1
[
E
[
1,E

_
P
1
L
E
Bv
O
, Bv
O
) +
_
K
2

2
w, w)
Thus

1
[
E
[
1,E
C
1
_
P
1
L
E
Bv
O
, Bv
O
) +
K
2

2
w, w)
_
, for some C
1
> 0. (6.17)
Let c
7
= min
_
c
3
2
,

1
C
1
_
. Then (6.14) and (6.17) imply
c
7
_
[
E
[
2
1,E
+[w[
2
2

P
1
L
E
Bv
O
, Bv
O
) +
K
2

2
w, w) +
K
2

2
w, w). (6.18)
Finally, if c = minc
1
, c
7
, (6.14) and (6.18) imply
c
_
[w[
2
2
+[y[
2
0
+[
E
[
2
1,E

[z[
2
c
.
91
Thus, (6.13) is an inner product that induces a norm on H equivalent to the usual Sobolev
topology on H
2
() L
2
() H
1
E
(). Henceforth, we denote the norm induced by (6.13) as
[z[
c
:=
_
z, z)
c
, where z = [w, y,
E
]
T
.
The domain of / is as follows:
T(/) = [w, y,
E
]
T
H : w H
4
(), y H
2
(),
E
H
3
E
(), + BCs,
where +BCs refers to the clamped boundary conditions (6.2) along with boundary condi-
tions induced by the image of /[w, y,
E
]
T
satisfying boundary conditions in H; that is,
y = 0, y = 0, G
E

E
+P
1
L
E
Bv
O
= 0 on . (6.19)
6.2.1 Semigroup Well-Posedness
In Hansen [15] (see Theorem 2), the semigroup associated with a similar plate model is
shown to be the generator of a strongly continuous semigroup. Using a procedure similar to
the one Theorem 2 of [15], we prove the following result:
Theorem 6.1. The operator / in (6.12) is the generator of a strongly continuous semigroup of
contractions on H. Furthermore the following dissipativity condition holds: For all z T(/),
Re z, /z)
c
=

G
1/2
E
h
1/2
E
(P
1
L
E
Bv
O
G
E

E
)

2
0,E
. (6.20)
Proof. We rst show that / is dissipative on H. By (5.9), we have the for all z T(/):
z, /z)
c
=y, K
2
w + div

N
T
h
E
P
1
L
E
h
E
(
E


Nw)) +K
2
w, y)
+G
E
h
E

E
,

G
1
E
[G
E

E
+P
1
L
E
h
E
(
E


Nw)])
P
1
L
E
h
E
(
E


Nw), h
E
[

G
1
E
(G
E

E
+P
1
L
E
h
E
(
E


Nw))

Ny]).
92
Since y = 0 on , the Divergence Theorem implies
z, /z)
c
=Ky,
2
w) y,

N
T
h
E
P
1
L
E
h
E
(
E


Nw)) +K
2
w, y)
G
E
h
E

E
,

G
1
E
G
E

E
) +G
E
h
E

E
,

G
1
E
P
1
L
E
h
E
(
E


Nw))
+P
1
L
E
h
E
(
E


Nw), h
E

G
1
E
G
E

E
)
P
1
L
E
h
E
(
E


Nw), h
E

G
1
E
P
1
L
E
h
E
(
E


Nw))
+P
1
L
E
h
E
(
E


Nw), h
E

Ny)).
Gathering terms leads to
z, /z)
c
=
_
K
2
w, y) Ky,
2
w)
_
+
_
P
1
L
E
h
E
(
E


Nw), h
E

Ny)) h
E

Ny, P
1
L
E
h
E
(
E


Nw))
_
h
E

G
1
E
G
E

E
, G
E

E
) +h
E

G
1
E
G
E

E
, P
1
L
E
h
E
(
E


Nw))
+P
1
L
E
h
E
(
E


Nw), h
E

G
1
E
G
E

E
)
h
E

G
1
E
P
1
L
E
h
E
(
E


Nw), P
1
L
E
h
E
(
E


Nw)). (6.21)
The rst two pairs of terms on the right side of (6.21) (grouped by ) are imaginary.
Therefore, taking the real part of (6.21) gives us
Re z, /z)
c
=
_

G
1/2
E
h
1/2
E
(P
1
L
E
h
E
(
E


Nw) G
E

E
), P
1
L
E
h
E
(
E


Nw) G
E

E
_
,
which is precisely (6.20).
Next, we prove that the mapping (I /) : T(/) H is onto. Suppose
/z +z = v, (6.22)
where v = [s, t,
E
]
T
is an arbitrary element in H. Our goal is to show that z = [w, y,
E
]
T(/). In particular, we prove that the following estimate holds:
[w[
4
+[y[
2
+[
E
[
3,E
C [[s[
2
+[t[
0
+[
E
[
1,E
] . (6.23)
(Throughout this chapter, C will denote some generic constant that may vary from line to line,
but it is independent of w, y, , s, t, and .)
93
(6.22) can be explicitly written as follows:
y +w = s (6.24)
K
2
w div

N
T
h
E
P
1
L
E
h
E
(
E


Nw) +my = mt (6.25)
G
E

E
P
1
L
E
h
E
(
E


Nw) +

G
E

E
=

G
E

E
(6.26)
Multiply (6.26) by

N
T
and take the divergence of both sides to obtain
div

N
T
G
E

E
+ div

N
T
P
1
L
E
h
E
(
E


Nw) div

N
T

G
E

E
= div

N
T

G
E

E
.
Add this to equation (6.25) and obtain
K
2
w div

N
T
(

G
E
+G
E
)
E
+my = mt div

N
T

G
E

E
.
Thus (6.24) and (6.26) imply
K
2
w +mw = div

N
T
(

G
E
+G
E
)
E
+m(s +t) div

N
T

G
E

E
(6.27)
(

G
E
+G
E
)
E
P
1
L
E
h
E

E
= P
1
L
E
h
E

Nw +

G
E

E
. (6.28)
Notice that for all w H
4
() H
2
0
() we have
K
2
w +mw, w) = K[w[
2
0
+m[w[
2
0
.
Thus the operator K
2
+ mI is associated with a coercive bilinear form, and then it follows
from Lax-Milgram lemma that this operator is an isomorphism from H
4
() H
2
0
() to L
2
().
Therefore, (6.27) implies
[w[
4
C
_
[div

N
T
(

G
E
+G
E
)
E
[
0
+[m(s +t)[
0
+[div

N
T

G
E

E
[
0
_
[w[
4
C [[
E
[
1,E
+[s[
0
+[t[
0
+[
E
[
1,E
] (6.29)
Next, note that for all
E
H
2
E
() H
1
E,0
(), we have the following:
(

G
E
+G
E
)
E
P
1
L
E
h
E

E
,
E
) = [(

G
E
+G
E
)
1/2

E
[
2
0,E
P
1
h
E
L
E

E
,
E
),
since

G
E
and G
E
are positive matrices and h
E
is diagonal. Since all matrix quantities in the
above equation are positive, there exist positive constants
1
and
2
such that
[(

G
E
+G
E
)
1/2

E
[
2
0,E
P
1
h
E
L
E

E
,
E
)
1
[
E
[
2
0,E

2
L
E

E
,
E
).
94
Therefore,
(

G
E
+G
E
)
E
P
1
L
E
h
E

E
,
E
)
1
[
E
[
2
0,E
+
2

E
(
E
,
E
)

1
[
E
[
2
0,E
+
2
c
0
[
E
[
2
1,E
by Corollary 6.2.
Hence the operator (

G
E
+G
E
)I P
1
L
E
h
E
is associated with a coercive bilinear form, and
then it follows from Lax-Milgram lemma that this operator is an isomorphism from H
2
E
()
H
1
E,0
() to L
2
E
(). Therefore, (6.28) implies
[
E
[
2,E
C
_
[P
1
L
E
h
E

Nw[
0,E
+[

G
E

E
[
0,E
_
[
E
[
2,E
C [[w[
3
+[
E
[
1,E
] . (6.30)
Moreover, equation (6.24) implies
[y[
2
C [[w[
2
+[s[
2
] . (6.31)
Notice that the dissipativity result (6.20) implies that the resolvent set of / is contained in
the open right half of the complex plane. Thus the operator (/ + I) is a one-to-one map,
and the z that solves (6.22) is unique. Applying the usual compactness/uniqueness argument
(see [17] and [15]) to the estimates (6.29), (6.30), and (6.31) implies
[w[
4
+[y[
2
+[
E
[
2,E
C [[s[
2
+[t[
0
+[
E
[
1,E
] . (6.32)
Next consider each row of equation (6.28). For i = 2, 4, , 2m, we have
(

G
i
+G
i
)
i
c
i
L

i
= d
i
L

i
w +

G
E

i
,
where c
i
and d
i
are positive since P
1
, h
E
, and

N are positive matrix quantities. Solving the
above equation for c
i
L

i
and taking the divergence of both sides leads to the following:
c
i
div L

i
= (

G
i
+G
i
)div
i
+d
i
div L

i
w

G
i
div
i
c
i
div
i
= (

G
i
+G
i
)div
i
+d
i

2
w

G
i
div
i
since div L

i
= div.
Hence,
[div
i
[
0
C
_
[div
i
[
0
+[
2
w[
0
+[div
i
[
0

.
95
This leads to

i=2,4, ,m
[div
i
[
0
C [[
E
[
1,E
+[w[
4
+[
E
[
1,E
]
C [[s[
2
+[t[
0
+[
E
[
1,E
] by (6.32). (6.33)
Inequalities (6.32) and (6.33) together give our desired inequality (6.23). Thus z T(/).
Finally, it follows from the L umer-Phillips Theorem (Theorem 2.2) that / is the generator of
a strongly continuous semigroup of contractions on H.
6.3 Analyticity of the Semigroup
The main result of this chapter is to show that the semigroup generated by / is in fact
analytic. We begin with a lemma.
6.3.1 Eigenvalues of /
Lemma 6.1. The operator / has no nonzero eigenvalues on the imaginary axis.
Proof. The proof is similar to Lemma 5.5. Suppose i is an eigenvalue of / with eigenvector
z = [ w, y,

E
]
T
, where R 0. Then (6.20) implies that
0 = Re z, (iI /) z)
c
=

G
1/2
E
h
1/2
E
(P
1
L
E
h
E
(

E


N w) G
E

E
)

2
0,E
,
and thus
P
1
L
E
h
E
(

E


N w) = 0 (6.34)
In addition, the solution z = [w, y,
E
]
T
to z = /z, z(0) = z satises

E
[
t=0
= 0.
Since ,= 0, it immediately follows from the eigenvector condition that

E
= 0. (6.35)
Equations (6.34) and (6.35) imply
L
E
h
E

N w = 0
96
Using the Greens formula (6.6) and the boundary conditions, it follows that

E
(h
E

N w, h
E

N w) = 0.
Then Korns inequality applied to
E
(see Corollary 6.2) implies that

N w = 0. (6.36)
Thus w is linear since

N is a vector of positive constants. However the boundary condition
w = 0 on implies w = 0 in . Finally, it easily follows from the eigenvector condition that
y = 0, and hence z = 0.
6.3.2 Main Result
Theorem 6.2. The semigroup generated by / is analytic.
Proof. First note, since T(/) is compactly embedded in H, the spectrum of / consists only of
eigenvalues. Since by Lemma 6.1 there are no eigenvalues on the nonzero imaginary axis, and
by Theorem 6.1 the semigroup is dissipative, the closed right half-plane is in the resolvent set
of /, with the only possible exception being an eigenvalue at the origin.
By Theorem 2.4, a sucient condition for analyticity is that there exist M,
0
for which
the resolvent operator R(, /) satises
|R( +i, /)|
M
[[
,= 0, >
0
. (6.37)
We rst prove that (6.4) implies (6.37). Indeed, with = +i the resolvent identity
R(, /) = R(i, /) + (i )R(, /)R(i, /)
together with (6.4) and the bound |R(, /)| 1/[[ (see Pazy [39], p. 11) gives
|R( +is, /)|
C
[[
+[[
1
[[

C
[[
=
2C
[[
, [[ 1.
For [[ < 1 once
0
> 0 is picked, |R(
0
+ i, /)| is uniformly bounded for [[ < 1, hence
(6.37) follows for any
0
> 0. Thus analyticity follows if we prove the resolvent estimate (6.4).
97
For the sake of convenience, we will denote
E
by and v
O
by v throughout the rest of
this chapter. For all R with [[ 1, and for an arbitrary [w
0
, y
0
,
0
]
T
c, dene
[w(), y(), ()]
T
:= R(i, /)[w
0
, y
0
,
0
]
T
.
According to (6.12), we can write this explicitly as follows:
iw()
1

y() =
1

w
0
(6.38)
imy() +
1

K
2
w()
1

div

N
T
h
E
P
1
L
E
h
E
_
()

Nw()
_
=
1

my
0
(6.39)
i

G
E
() +
1

G
E
()
1

P
1
L
E
h
E
_
()

Nw()
_
=
1

G
E

0
. (6.40)
Since i : R, [[ 1 belongs to the resolvent set of /, R(i, /) maps into T(/) for
[[ 1, and thus [w(), y(), ()]
T
T(/). Therefore, we have the following boundary
conditions:
w() = w() = v() = 0 on . (6.41)
Since () = h
1
E
Bv() +Nw(), we have that
() = 0 on . (6.42)
Furthermore, (6.38) and (6.41) imply that
y() = y() = 0, G
E

E
() P
1
L
E
h
E
(
E
()

Nw()) = 0 on .
In order to prove (6.4), it is enough to show that there exists a constant C > 0 independent of
such that for all [w
0
, y
0
,
0
]
T
c,
[w()[
2
+[y()[
0
+[()[
1,E

C
[[
([w
0
[
2
+[y
0
[
0
+[
0
[
1,E
) for all R, [[ 1. (6.43)
It will be convenient to dene the following:
z() := [w(), y(), ()]
T
, z
0
:= [w
0
, y
0
,
0
]
T
.
In order to prove (6.43), we need to prove the following lemma:
98
Lemma 6.2. Let z() = [w(), y(), ()]
T
be a solution to (6.38)-(6.40) with clamped bound-
ary conditions. Then for any > 0 there exists a constant C

> 0 independent of such

that
[()[
2
1
[z()[
2
c
+
C

[[
2
[z
0
[
2
c
(6.44)
[w()[
2
2
[z()[
2
c
+
C

[[
2
[z
0
[
2
c
(6.45)
[y()[
2
0
[z()[
2
c
+
C

[[
2
[z
0
[
2
c
(6.46)
Pending the proof of this lemma, we have the following for all z() T(/) and
0
> 0:
[z()[
2
c
< C
_
[w()[
2
2
+[v()[
2
0
+[s()[
2
1

<
0
[z()[
2
c
+
C

0
[[
2
[z
0
[
2
c
.
Taking
0
to be suciently small, we get
[z()[
2
c

C
[[
2
[z
0
[
2
c
.
This completes the proof of the main theorem.
6.3.3 A-priori estimates
In order to prove Lemma 6.2, we need to establish two more lemmas.
Lemma 6.3. Let z() = [w(), y(), ()]
T
be a solution to (6.38)-(6.40) with clamped bound-
ary conditions. Then, there exists a C > 0 independent of such that
[t()[
2
0,E
C[z
0
[
c
[z()[
c
, (6.47)
where t() := i

G
E
()

G
E

0
.
Proof. Equations (6.38)-(6.40) can be written in the following form:
iz() /z() = z
0
. (6.48)
If we take the energy inner product of (6.48) with z(), we obtain
i z(), z())
c
/z(), z())
c
= z
0
, z())
c
. (6.49)
99
Take the real part of (6.49) and obtain
Re /z(), z())
c
= Re z
0
, z())
c
. (6.50)
Therefore, by (6.20), (6.50) implies

G
1/2
E
h
1/2
E
_
P
1
L
E
h
E
(()

Nw()) G
E
()
_

2
0,E
= Re z
0
, z())
c
. (6.51)
Since

G
1/2
E
h
1/2
E
is a positive, invertible matrix, (6.51) implies that there is a constant C > 0
such that

G
E
() P
1
L
E
h
E
(()

Nw())

2
0,E
C[z
0
[
c
[z()[
c
. (6.52)
But according to (6.40)
i

G
E
()

G
E

0
= G
E
() P
1
L
E
h
E
(()

Nw()).
This result, along with (6.52) implies that

i

G
E
()

G
E

2
0,E
C[z
0
[
c
[z()[
c
.
This is precisely (6.47), according to the denition of t().
Lemma 6.4. Let z() = [w(), y(), ()]
T
be a solution to (6.38) -(6.40) with clamped bound-
ary conditions. Then there exists a constant C > 0 independent of such that
1
[[
[y()[
2
+
1
_
[[
[y()[
1
C
_
[z()[
c
+
1
[[
[z
0
[
c
_
(6.53)
1
[[
[w()[
4
+
1
_
[[
[w()[
3
C
_
[z()[
c
+
1
[[
[z
0
[
c
_
(6.54)
1
_
[[
[L
E
()[
0,E
C
_
[z()[
c
+
1
[[
[z
0
[
c
_
(6.55)
6.3.3.1 A-priori estimates for y():
According to (6.38), we have that
1

y() = iw()
1

w
0
.
100
Thus,
1
[[
[y()[
2
[w()[
2
+
1
[[
[w
0
[
2
C
_
[z()[
c
+
1
[[
[z
0
[
c
_
. (6.56)
But we certainly have [y()[
0
C[z()[
c
. If we interpolate this with respect to (6.56), we
obtain the estimate
1
_
[[
[y()[
1
C
_
[z()[
c
+
1
[[
[z
0
[
c
_
. (6.57)
(6.56) and (6.57) together imply (6.53).
6.3.3.2 A-priori estimates for w():
According to (6.40) we have
i

G
E
() +
1

G
E
()
1

P
1
L
E
h
E
_
()

Nw()
_
=
1

G
E

0
.
Multiplying both sides of this equation by

N
T
h
E
and taking the divergence of both sides
implies
i div

N
T
h
E

G
E
() +
1

div

N
T
h
E
G
E
()

div

N
T
h
E
P
1
L
E
h
E
_
()

Nw()
_
=
1

div

N
T
h
E

G
E

0
. (6.58)
If we subtract (6.58) from (6.39), we obtain
imy() +
1

K
2
w() i div

N
T
h
E

G
E
()

div

N
T
h
E
G
E
() =
1

my
0

div

N
T
h
E

G
E

0
,
which leads to
1

K
2
w() = i div

N
T
h
E

G
E
() imy() +
1

div

N
T
h
E
G
E
()
+
1

my
0

div

N
T
h
E

G
E

0
.
Therefore,
1
[[
[
2
w()[
0
C
_
[div

N
T
()[
0
+[y()[
0
+
1
[[
[div

N
T
()[
0
+
1
[[
[y
0
[
0
+
1
[[
[div

N
T

0
[
0
_
C
_
2[div

N
T
()[
0
+[y()[
0
+
1
[[
[y
0
[
0
+
1
[[
[div

N
T

0
[
0
_
, since
1
[[
1
C
_
[()[
1,E
+[y()[
0
+
1
[[
([y
0
[
0
+[
0
[
1,E
)
_
101
Hence, we have
1
[[
[
2
w()[
0
C
_
[z()[
c
+
1
[[
[z
0
[
c
_
. (6.59)
Since
2
is a positive operator associated with a coercive bilinear form, the Lax-Milgram
lemma implies that it is an isomorphism from H
4
() H
2
0
() to L
2
(). Therefore, (6.59)
implies
1
[[
[w()[
4
C
_
[z()[
c
+
1
[[
[z
0
[
c
_
. (6.60)
Then interpolation of (6.60) with respect to [w()[
2
[z()[
c
implies
1
_
[[
[w()[
3
C
_
[z()[
c
+
1
[[
[z
0
[
c
_
. (6.61)
This result along with (6.60) gives us (6.54).
6.3.3.3 A-priori estimates for ():
According to (6.40),
i

G
E
() +
1

_
G
E
() P
1
L
E
h
E
_
()

Nw()
__
=
1

G
E

0
.
Hence

E
((), i

G
E
()) +
1

E
_
(), G
E
() P
1
L
E
h
E
(()

Nw())
_
=
1

E
((),

G
E

0
).
Since G
E

E
() P
1
L
E
h
E
(
E
()

Nw()) = 0 on , we have the following:

E
((), i

G
E

E
())
1

_
L
E
(), G
E
() P
1
L
E
h
E
_
()

Nw()
__
=
1

E
((),

G
E

0
)

E
((), i

G
E
())
1

L
E
(), G
E
()) +
1

L
E
(), P
1
L
E
h
E
()
_

_
L
E
(), P
1
L
E
h
E

Nw()
_
=
1

E
((),

G
E

0
)

E
((), i

G
E
()) +
1

E
((), G
E
()) +
1

L
E
(), P
1
L
E
h
E
()
_

_
L
E
(), P
1
L
E
h
E

Nw()
_
=
1

E
((),

G
E

0
)
102
Therefore,
1
[[

L
E
(), P
1
L
E
h
E
()
_

E
((),

G
E

E
()) +
1
[[

E
((), G
E
())
+
1
[[

_
L
E
(), P
1
L
E
h
E

Nw()
_

+
1
[[

E
((),

G
E

0
). (6.62)
Left side of (6.62): Since P
1
and h
E
are positive matrices, there is an
1
> 0 such that

1
[L
E
(), L
E
())[

L
E
(), P
1
L
E
h
E
()
_

. (6.63)
First term on right side of (6.62): Since

G
E
is a positive matrix, there exists a
1
> 0 such
that

E
((),

G
E
())
1

E
((), ()).
Then the continuity of
E
on H
1
E
() (see Corollary 6.3) implies

E
((),

G
E
())
1
[()[
2
1,E
. (6.64)
Second term on right side of (6.62): Repeating the same argument as above, there exists a

2
> 0 (since G
E
is a positive matrix) such that

E
((), G
E
())
2
[()[
2
1,E
. (6.65)
Third term on right side of (6.62): By the Schwartz Inequality

_
L
E
(), P
1
L
E
h
E

Nw()
_

[L
E
()[
0,E
[P
1
L
E
h
E

Nw()[
0,E
.
Since P
1
, h
E
, and

N are positive matrix quantities, there exists a
3
> 0 such that
[P
1
L
E
h
E

Nw()[
0,E

3
|L

w()|
(L
2
())
2.
But L

w() = (w()) (see Appendix B). Thus,

_
L
E
(), P
1
L
E
h
E

Nw()
_

3
[L
E
()[
0,E
|(w())|
(L
2
())
2. (6.66)
103
Fourth term on right side of (6.62): We use an argument similar to the ones used for the rst
and second terms on the right side of (6.62). Since

G
E
is a positive matrix, there exists a

4
> 0 such that

E
((),

G
E

0
)
4
[()[
1,E
[
0
[
1,E
. (6.67)
Combining the results of (6.62 - 6.67) together leads to the following:
1
[[
[L
E
()[
2
0,E
C
_
[()[
2
1,E
+
1
[[
[()[
2
1,E
+
1
[[
[L
E
()[
0,E
|(w())|
(L
2
())
2
_
+
C
[[
[()[
1,E
[
0
[
1,E
C
_
2[()[
2
1,E
+
_
1
_
[[
[L
E
()[
0,E
__
1
_
[[
[w()[
3
__
+C[()[
1,E
_
1
[[
[
0
[
1,E
_
, since [[ 1.
Then (6.61) implies
1
[[
[L
E
()[
2
0,E
C
_
[z()[
2
c
+
1
_
[[
[L
E
()[
0,E
_
[z()[
c
+
1
[[
[z
0
[
c
_
+[z()[
c
_
1
[[
[z
0
[
c
_
_
.
(6.68)
Next, we apply Youngs Inequality on the second and third terms on the right side of (6.68).
For all > 0, we have
1
[[
[L
E
()[
2
0,E
C[z()[
2
c
+

[[
[L
E
()[
2
0,E
+C

_
[z()[
2
c
+
1
[[
2
[z
0
[
2
c
_
+[z()[
c
+
C

[[
2
[z
0
[
2
c
,
which implies
(1 )
1
[[
[L
E
()[
2
0,E


C

_
[z()[
2
c
+
1
[[
2
[z
0
[
2
c
_
. (6.69)
Letting be suciently small (in particular, < 1) in (6.69) leads to
1
[[
[L
E
()[
2
0,E
C
_
[z()[
2
c
+
1
[[
2
[z
0
[
2
c
_
.
Taking the square root of both sides gives us our desired a-priori estimate for () in (6.55).
6.3.4 Main Estimates
The remainder of this section is devoted to proving the main estimates in Lemma 6.2. Once
again, we will prove this in three parts.
104
6.3.4.1 Main estimates for ():
From the denition of t(), we have
i

G
E
()
1

t() =
1

G
E

0
.
Thus,

E
((), i

G
E
())
1

E
((), t()) =
1

E
((),

G
E

0
)

E
((), i

G
E
()) +
1

L
E
(), t())

=
1

E
((),

G
E

0
),
since t() = 0 on . The continuity of
E
on H
1
E
() (see Corollary 6.3) along with the positivity
of the matrix

G
E
implies

E
((),

G
E

0
) C[()[
1,E
[
0
[
1,E
.
Therefore,

E
(

G
1/2
E
(),

G
1/2
E
())
C
[[
[L
E
()[
0,E
[t()[
0,E
+
C
[[
[()[
1,E
[
0
[
1,E
. (6.70)
Left side of (6.70): Since

G
1/2
E
() H
1
E,0
(), Corollary 6.2 implies that there is an
1
> 0
such that

1
[

G
1/2
E
()[
2
1,E

E
(

G
1/2
E
(),

G
1/2
E
()).
Since

G
1/2
E
is a positive matrix, there exists an
2
> 0 such that

2
[()[
2
1,E
[

G
1/2
E
()[
2
1,E
.
Therefore,

2
[()[
2
1,E

E
(

G
1/2
E
(),

G
1/2
E
()). (6.71)
First term on the right side of (6.70): According to Lemma 6.3 and the a-priori estimate for
() (see (6.55)), we have the following:
1
[[
[L
E
()[
0,E
[t()[
0,E
=
_
1
_
[[
[L
E
()[
0,E
__
1
_
[[
[t()[
0,E
_
C
_
[z()[
c
+
1
[[
[z
0
[
c
_
[z()[
1/2
c
_
1
_
[[
[z
0
[
1/2
c
_
105
By Youngs Inequality, we have for all
1
> 0
1
[[
[L
E
()[
0,E
[t()[
0,E
C
_
[z()[
c
+
1
[[
[z
0
[
c
__

1
[z()[
c
+
C

1
[[
[z
0
[
c
_
.
Applying Youngs Inequality once again for > 0, we have
1
[[
[L
E
()[
0,E
[t()[
0,E

_
[z()[
2
c
+
1
[[
2
[z
0
[
2
c
_
+C

2
1
[z()[
2
c
+
C
2

1
[[
2
[z
0
[
c
_
.
If we let
1
be arbitrarily small, we obtain the following inequality for > 0:
1
[[
[L
E
()[
0,E
[t()[
0,E
[z()[
2
c
+
C

[[
2
[z
0
[
2
c
(6.72)
Second term on the right side of (6.70): By Youngs Inequality, we have for all > 0:
1
[[
[()[
1,E
[
0
[
1,E
[z()[
c
_
1
[[
[z
0
[
c
_
[z()[
2
c
+
C

[[
2
[z
0
[
2
c
. (6.73)
Finally (6.70 - 6.73) together imply the main estimate for () in (6.44).
6.3.4.2 Main estimates for w():
Going back to (6.40), we multiply both sides by

N
T
and obtain
i

N
T

G
E
() +
1

N
T
G
E
()
1

N
T
P
1
L
E
h
E
_
()

Nw()
_
=
1

N
T

G
E

0
. (6.74)
Taking the L
2
-inner product of (6.74) with y() yields
i

N
T

G
E
(), y())

+
1

N
T
G
E
(), y())

N
T
P
1
L
E
h
E
_
()

Nw()
_
, y()
_

=
1

N
T

G
E

0
, y())

N
T
P
1
L
E
h
E

Nw(), y()
_

=
1

N
T
P
1
L
E
h
E
(), y()
_

N
T

G
E
(), y())

N
T
G
E
(), y())

+
1

N
T

G
E

0
, y())

. (6.75)
But according to (6.38),
1

y() = iw()
1

w
0
. Thus, (6.75) becomes
i
_

N
T
P
1
L
E
h
E

Nw(), w()
_

=
1

N
T
P
1
L
E
h
E

Nw(), w
0
_

+
1

N
T
P
1
L
E
h
E
(), y()
_

N
T

G
E
(), y())

N
T
G
E
(), y())

+
1

N
T

G
E

0
, y())

. (6.76)
106
Since P
1
is a positive matrix, and G
E
and

G
E
are invertible matrices, (6.76) implies

_
L
E
h
E

Nw(),

Nw()
_

C
[[

_
L
E
h
E

Nw(),

Nw
0
_

+
C
[[

_
L
E
h
E
(),

Ny()
_

+C

(),

Ny())

+
C
[[

(),

Ny())

+
C
[[

0
,

Ny())

. (6.77)
Since

N is a vector of positive constants and h
E
is a diagonal matrix whose diagonal entries
are positive, there exists an > 0 such that
[L

w(), w())

_
L
E
h
E

Nw(),

Nw()
_

. (6.78)
But,
[L

w(), w())

[ = [div L

w(), w())

[ , since w() = 0 on
=

2
w(), w()
_

, since divL

w() =
2
w()
= [w(), w())

[ , since w() = w() = 0 on .

Therefore,
[w()[
2
0

_
L
E
h
E

Nw(),

Nw()
_

. (6.79)
Using (6.78) and (6.79), we can rewrite (6.77) as follows:
[w()[
2
0

C
[[

_
L
E
h
E

Nw(),

Nw
0
_

+
C
[[

_
L
E
h
E
(),

Ny()
_

+C

(),

Ny())

+
C
[[

(),

Ny())

+
C
[[

0
,

Ny())

. (6.80)
First term on the right side of (6.80): Using an argument analogous to the one used to obtain
the inequality in (6.79) (except with the inequality reversed), there exists a constant > 0
such that

_
L
E
h
E

Nw(),

Nw
0
_

[w(), w
0
)

[ . (6.81)
Note that (6.81) is valid since we have the appropriate boundary conditions on w
0
(w
0
=
107
w
0
= 0 on ). Therefore, (6.81) implies
1
[[

_
L
E
h
E

Nw(),

Nw
0
_

[[
[w()[
0
[w
0
[
0

C
[[
[z()[
c
[z
0
[
c
[z()[
2
c
+
C

[[
2
[z
0
[
2
c
by Youngs inequality. (6.82)
Second term on the right side of (6.80): Since

Ny() = 0 on , (6.6) implies
_
L
E
h
E
(),

Ny()
_

=
E
(h
E
(),

Ny()).
Therefore, by Corollary 6.3 (see (6.8)), we have
1
[[

_
L
E
h
E
(),

Ny()
_

1
[[
[h
E
()[
1,E
[

Ny()[
1,E
C ([()[
1,E
)
_
1
[[
[y()[
2
_
. (6.83)
Using the a-priori estimate for y() in Lemma 6.4 (see (6.53)) and the main estimate for ()
(see (6.44)), (6.83) implies the following for all
1
> 0:
1
[[

_
L
E
h
E
(),

Ny()
_

C
_

1
[z()[
c
+
C

1
[[
[z
0
[
c
__
[z()[
c
+
1
[[
[z
0
[
c
_
. (6.84)
Applying Youngs inequality for > 0 in (6.84), we obtain
1
[[

_
L
E
h
E
(),

Ny()
_

_
[z()[
2
c
+
1
[[
2
[z
0
[
2
c
_
+C

2
1
[z()[
2
c
+
C
2

1
[[
2
[z
0
[
2
c
_
.
Then by letting
1
be small enough, we get
1
[[

_
L
E
h
E
(),

Ny()
_

[z()[
2
c
+
C

[[
2
[z
0
[
2
c
. (6.85)
Third term on the right side of (6.80): Since y() = 0 on , the Divergence Theorem implies

(),

Ny())

N
T
(), y())

div

N
T
(), y())

Since

N has only positive constants,

(),

Ny())

C[()[
0,E
[y()[
0
C[()[
1,E
[z()[
c
(6.86)
108
Using the main estimate for () in (6.44), (6.86) implies that for all
1
> 0 we have

(),

Ny())

C
_

1
[z()[
c
+
C

1
[[
[z
0
[
c
_
[z()[
c
. (6.87)
Applying Youngs inequality for > 0 in (6.87), we obtain

(),

Ny())

[z()[
2
c
+C

2
1
[z()[
2
c
+
C
2

1
[[
2
[z
0
[
2
c
_
If
1
is suciently small in the above inequality, we have the estimate

(),

Ny())

[z()[
2
c
+
C

[[
2
[z
0
[
2
c
. (6.88)
Fourth term on the right side of (6.80): Since

N consists only of positive constants,
1
[[

(),

Ny())

C[()[
0,E
_
1
[[
[y()[
0
_
C[()[
1,E
_
1
_
[[
[y()[
1
_
, since [[ 1
C
_

1
[z()[
c
+
C

1
[[
[z
0
[
c
_
_
1
_
[[
[y()[
1
_
by (6.44) for all
1
> 0
C
_

1
[z()[
c
+
C

1
[[
[z
0
[
c
__
[z()[
c
+
1
[[
[z
0
[
c
_
by (6.53). (6.89)
For all > 0, we apply Youngs inequality in (6.89) to get
1
[[

(),

Ny())

_
[z()[
2
c
+
1
[[
2
[z
0
[
2
c
_
+C

2
1
[z()[
2
c
+
C
2

1
[[
2
[z
0
[
2
c
_
.
By choosing
1
small enough in the previous inequality, we obtain
1
[[

(),

Ny())

[z()[
2
c
+
C

[[
2
[z
0
[
2
c
. (6.90)
Fifth term on the right side of (6.80): Since y() = 0 on , the Divergence Theorem implies

0
,

Ny())

N
T

0
, y())

div

N
T

0
, y())

.
Since

N is a vector of positive constants, we have
1
[[

0
,

Ny())

C
[[
[
0
[
0,E
[y()[
0

C
[[
[z
0
[
c
[z()[
c
[z()[
2
c
+
C

[[
2
[z
0
[
2
c
by Youngs inequality for > 0. (6.91)
109
Now we combine (6.82), (6.85), (6.88), (6.90), and (6.91) with (6.80) to get the following
estimate:
[w()[
2
0
[z()[
2
c
+
C

[[
2
[z
0
[
2
c
. (6.92)
Since w() = 0 on , elliptic estimates applied to (6.92) implies the desired estimate (6.45) for
w().
6.3.4.3 Main estimates for y():
According to (6.39), we have
imy() =
1

div

N
T
h
E
P
1
L
E
h
E
()
1

div

N
T
h
E
P
1
L
E
h
E

Nw()
1

K
2
w() +
1

my
0
.
(6.93)
Taking the L
2
-inner product of (6.93) with y() implies
im[y()[
2
0
=
1

div

N
T
h
E
P
1
L
E
h
E
(), y())

div

N
T
h
E
P
1
L
E
h
E

Nw(), y())

K
2
w(), y())

+
1

my
0
, y())

.
(6.94)
Since h
E
and P
1
are invertible matrices, (6.94) implies
[y()[
2
0

C
[[

div

N
T
L
E
h
E
(), y())

+
C
[[

div

N
T
L
E
h
E

Nw(), y())

+
C
[[

2
w(), y())

+
C
[[
[y
0
, y())

[ . (6.95)
Since

N is a vector of positive constants and div L

w() =
2
w(), we have

div

N
T
L
E
h
E

Nw(), y())

C [div L

w(), y())

[ = C

2
w(), y())

Therefore, (6.95) can be written as follows:

[y()[
2
0

C
[[

div

N
T
L
E
h
E
(), y())

+
2C
[[

2
w(), y())

+
C
[[
[y
0
, y())

[ . (6.96)
First term on the right side of (6.96): Since y() = 0 on , the Divergence Theorem implies

div

N
T
L
E
h
E
(), y())

N
T
L
E
h
E
(), y())

L
E
h
E
(),

Ny())

.
110
Then, by (6.85), we have for all > 0,
1
[[

div

N
T
L
E
h
E
(), y())

[z()[
2
c
+
C

[[
2
[z
0
[
2
c
. (6.97)
Second term on the right side of (6.96): Since y() = y() = 0 on , integration by parts
twice implies

2
w(), y())

= [w(), y())

[ .
Therefore we have,
1
[[

2
w(), y())

C[w()[
0
_
1
[[
[y()[
0
_
C[w()[
0
_
1
_
[[
[y()[
1
_
, since [[ 1
C
_

1
[z()[
c
+
C

1
[[
[z
0
[
c
_
_
1
_
[[
[y()[
1
_
by (6.92) for every
1
> 0
C
_

1
[z()[
c
+
C

1
[[
[z
0
[
c
__
[z()[
c
+
1
[[
[z
0
[
c
_
by (6.53). (6.98)
Applying once more Youngs inequality for > 0, we have from (6.98)
1
[[

2
w(), y())

_
[z()[
2
c
+
1
[[
2
[z
0
[
2
c
_
+C

2
1
[z()[
2
c
+
C
2

1
[[
2
[z
0
[
2
c
_
.
Taking
1
suciently small, we obtain the estimate
1
[[

2
w(), y())

[z()[
2
c
+
C

[[
2
[z
0
[
2
c
. (6.99)
Third term on the right side of (6.96):
1
[[
[y
0
, y())

[
C
[[
[y
0
[
0
[y()[
0

C
[[
[z
0
[
c
[z()[
c
[z()[
2
c
+
C

[[
2
[z
0
[
2
c
by Youngs inequality for > 0. (6.100)
Finally, if we combine (6.97), (6.99), and (6.100) with (6.96), we get our desired estimate (6.46)
for y(). This completes the proof of Lemma 6.2, as well as the proof of the main theorem.
111
CHAPTER 7. Exponential Stability of a Multilayer Rao-Nakra Beam
7.1 Introduction
The previous two chapters investigate stability results for a beam and a plate, both of which
are Mead-Markus models. We next look at a multilayer Rao-Nakra beam in this chapter, and
prove that the associated semigroup is exponentially stable under certain circumstances. Recall
that the equations of motion for a Mead-Markus beam (4.18) are obtained by simply removing
the moment of inertia term in the equations for a Rao-Nakra beam (4.16). We consider (4.16)
with hinged boundary conditions
_

_
m w w
tt
+Kw
tttt

N
T
h
E
_
G
E

E
+

G
E

E
_
t
= 0 on (0, L) (0, )
h
O
p
O
v
O
h
O
E
O
v
tt
O
+B
T
_
G
E

E
+

G
E

E
_
= 0 on (0, L) (0, )
w(x, t) = w
tt
(x, t) = 0, v
t
O
(x, t) = 0, x 0, L, t > 0
. (7.1)
In Section 7.2, we formulate this as a semigroup problem with semigroup generator /. We
then show that / generates a semigroup of contractions on the appropriate space. Our main
goal is to establish the exponential stability of the multilayer Rao-Nakra beam. This will
require some extra conditions because, as we will see in Section 7.3, the semigroup generator
/ has eigenvalues on the imaginary axis. In Section 7.3, we determine necessary and sucient
conditions for / to have no eigenvalues on the imaginary axis. Then in Section 7.4 we show
that the conditions we establish in Section 7.3 are necessary and sucient for / to be the
generator of an exponentially stable semigroup. Our proof relies on showing that the estimate
|R(i, /)| C for all real (7.2)
is valid if and only if the conditions we nd in Section 7.3 hold. We prove this result using a
contradiction argument similar to the one in Hansen and Liu [18].
112
7.2 Semigroup Formulation
Let w, w, v
O
, and v
O
be the state variables. Dene
U = [u, u]
T
:= [w, v
O
]
T
, V = [v, v]
T
:= [ w, v
O
]
T
, and Y := [U, V ]
T
.
With the notation introduced above, we have the following (see (4.14)):

E
= h
1
E
Bu +

Nu
t
, and

E
= h
1
E
Bv +

Nv
t
.
Also, dene the operator J : H
2
(0, L)H
1
0
(0, L) L
2
(0, L) by J = m
tt
. It follows from
the Lax-Milgram Lemma that J is an isomorphism from H
1
0
(0, L) to H
1
(0, L). Moreover, it
is coercive and invertible as an operator that maps H
2
(0, L) H
1
0
(0, L) to L
2
(0, L). Thus, we
rewrite (7.1) as follows:
dY
dt
= /Y :=
_

_
0 I
A
1
A
2
_

_
_

_
U
V
_

_, (7.3)
where
A
1
U :=
_

_
J
1
_
Ku
tttt
+

N
T
h
E
G
E
(h
1
E
Bu
t
+

Nu
tt
)
_
h
1
O
p
1
O
_
h
O
E
O
u
tt
B
T
G
E
(h
1
E
Bu +

Nu
t
)
_
_

_, (7.4)
and
A
2
V :=
_

_
J
1
_

N
T
h
E

G
E
(h
1
E
Bv
t
+

Nv
tt
)
_
h
1
O
p
1
O
_
B
T

G
E
(h
1
E
Bv +

Nv
t
)
_
_

_, (7.5)
Remark 7.1. From the denition of / and the boundary conditions
u = u
tt
= 0, u
t
= 0; x = 0, L
it can be shown that = 0 is a double eigenvalue of /, with eigenvector
u = 0, u =

1
O
, v = 0, v =

0
O
,
and generalized eigenvector
u = 0, u =

0
O
, v = 0, v =

1
O
113
(see Hansen [14] for a proof of this). The solutions [u, u, v, v]
T
= [0,

1
T
O
, 0,

0
T
O
]
T
correspond to
inertial sliding solutions.
Because we have a nontrivial null space for /, some care must be taken when dening the
state space and the domain of /. Dene the space
/ = span
_

_
0

1
O
f(x)
_

_,
where f(x) is a scalar function satisfying f
t
(0) = f
t
(L) = 0. We dene the state space H to be
H := (X
1
+/) X
0
,
where
X
1
:=
_
[u, u]
T
: u H
2
(0, L) H
1
0
(0, L), u (H
1
(0, L))
(m+1)
_
X
0
:=
_
[v, v]
T
: v H
1
0
(0, L), v (L
2
(0, L))
(m+1)
_
.
In addition, X
1
+/ denotes the quotient space X
1
/.
Using the total energy of the multilayer Rao-Nakra beam (see (4.15)), we dene the fol-
lowing bilinear form:
Y,

Y )
c
=
1
2
_
mv, v)

+v
t
, v
t
)

+h
O
p
O
v, v)

+Ku
tt
, u
tt
)

+
1
2
_
h
O
E
O
u
t
, u
t
)

+G
E
h
1
E
(Bu +h
E

Nu
t
), (B u +h
E

N u
t
))

_
. (7.6)
In (7.6),

Y denotes [ u, u, v, v]
T
. We claim that this denes an inner product on H. Then main
problem is showing that this bilinear form is coercive on H. In particular, we need to nd
a bound from below by |u|
2
(H
1
(0,L))
m+1
. If u =

1
O
f(x), then Bu = 0. However, for every
nonzero vector
_

_
u
u
_

_ X
1
+/, we have that there exists a constant c
0
> 0 such that
c
0
|u|
2
(L
2
(0,L))
m+1
|Bu|
2
(L
2
(0,L))
m
. (7.7)
Since G
E
and h
E
are invertible matrices and

N is a vector of positive constants, there exists
a constant c
1
> 0 such that
c
1
_
|Bu +h
E

Nu
t
|
2
(L
2
(0,L))
m
_
G
E
h
1
E
(Bu +h
E

Nu
t
), (Bu +h
E

Nu
t
))

.
114
The reverse triangle inequality then implies that for all
1
(0, c
1
],

1
|Bu|
(L
2
(0,L))
m

1
|h
E

Nu
t
|
(L
2
(0,L))
m |G
1/2
E
h
1/2
E
(Bu +h
E

Nu
t
)|
(L
2
(0,L))
m

1
|Bu|
(L
2
(0,L))
m
_

1
(

N
T
h
2
E

N)|u
t
|
(L
2
(0,L))
m |G
1/2
E
h
1/2
E
(Bu +h
E

Nu
t
)|
(L
2
(0,L))
m.
(7.8)
Add

L
_
K
2
|u
t
|
(L
2
(0,L))
m to both sides of (7.8) to obtain

1
|Bu|
(L
2
(0,L))
m +
_

L
_
K
2

_

1
(

N
T
h
2
E

N)
_
|u
t
|
(L
2
(0,L))
m
|G
1/2
E
h
1/2
E
(Bu +h
E

Nu
t
)|
(L
2
(0,L))
m +

L
_
K
2
|u
t
|
(L
2
(0,L))
m.
Fix
1
= min
_
c
1
,

2
K
2L
2
(

N
T
h
2
E

N)
_
. Then

L
_
K
2

_

1
(

N
T
h
2
E

N) 0, and

1
|Bu|
(L
2
(0,L))
m |G
1/2
E
h
1/2
E
(Bu +h
E

Nu
t
)|
(L
2
(0,L))
m +

L
_
K
2
|u
t
|
(L
2
(0,L))
m
|G
1/2
E
h
1/2
E
(Bu +h
E

Nu
t
)|
(L
2
(0,L))
m +
_
K
2
|u
tt
|
(L
2
(0,L))
m by (5.12).
Therefore, for some C
1
> 0, we have

1
|Bu|
2
(L
2
(0,L))
m
C
1
_
G
E
h
1
E
(Bu +h
E

Nu
t
), (Bu +h
E

Nu
t
))

+
K
2
u
tt
, u
tt
)

1
c
0
|u|
2
(L
2
(0,L))
m+1
C
1
_
G
E
h
1
E
(Bu +h
E

Nu
t
), (Bu +h
E

Nu
t
))

+
K
2
u
tt
, u
tt
)

_
, (7.9)
where we used (7.7) in the last line. Recall from (5.13) that
|u|
2
H
2
()
M
2
|u
tt
|
2
L
2
()
; M =
L
2

2
+
L

+ 1.
Thus, by setting c
2
= min
_

1
c
0
C
1
,
2
KM
2
_
, we have by (7.9) that
c
2
_
|u|
H
2
()
+|u|
2
(L
2
(0,L))
m+1
_

K
2
u
tt
, u
tt
)

+G
E
h
1
E
(Bu +h
E

Nu
t
), (Bu +h
E

Nu
t
))

+
K
2
u
tt
, u
tt
)

. (7.10)
Finally, since h
O
E
O
is a positive matrix, there is a c
3
> 0 such that
c
3
|u
t
|
2
(L
2
(0,L))
m+1
h
O
E
O
u
t
, u
t
)

.
115
Then if c
4
= minc
2
, c
3
, we have by (7.10)
c
4
_
|u|
H
2
()
+|u|
2
(H
1
(0,L))
m+1
_
Ku
tt
, u
tt
)

+G
E
h
1
E
(Bu +h
E

Nu
t
), (Bu +h
E

Nu
t
))

+h
O
E
O
u
t
, u
t
)

. (7.11)
The remaining terms in (7.6) are easily bounded from below by |v|
H
1
()
and |v|
(L
2
(0,L))
m+1,
and so it follows from (7.11) that (7.6) is coercive on H. Furthermore, it is easy to bound each
of the terms from above. Therefore, (7.6) is a valid energy inner product on H, and the norm
on H can be determined by (7.6). The domain of / is
T(/) = (X
2
+/) X
1
,
where
X
2
:= [u, u]
T
X
1
: u H
3
(0, L), u (H
2
(0, L))
(m+1)
, u
tt
(0) = u
tt
(L) = 0, u
t
(0) = u
t
(L) = 0.
Remark 7.2. In order for / to map into H, the I that appears in the denition of / (see
(7.3)) must be the identity operator on the quotient space X
1
+/.
Remark 7.3. Since
/
_
[0,

0
O
, 0,

1
O
f(x)]
T
_
= [0,

1
O
f(x), 0,

0
O
]
T
and [0,

1
O
f(x), 0,

0
O
]
T
can be viewed as a zero element in H, we now have that the vector
u = 0, u =

0
O
, v = 0, v =

1
O
f(x)
is the eigenvector for / (not a generalized eigenvector) corresponding to the simple eigenvalue
= 0.
7.2.1 Semigroup Well-Posedness
In order to establish the well-posedness of the semigroup generated by /, we rst show
that the adjoint /

exists on H, and it is given by /(

G
E
)
116
Lemma 7.1. Let /(

G
E
) denote the dependence of / on the parameter

G
E
. Then
_
/(

G
E
)
_

= /(

G
E
) on H.
Proof. Let Y = [u, u, v, v]
T
and W = [s, s, t, t]
T
be any state vectors in T(/). This implies
the following boundary conditions:
u = u
tt
= s = s
tt
= v = t = 0, u
t
= s
t
= 0; x = 0, L. (7.12)
Dene := J
1
_
Ku
tttt
+D
x

N
T
h
E
(G
E
(Bu +h
E

Nu
t
) +

G
E
(Bv +h
E

Nv
t
))
_
and
:= h
1
O
p
1
O
_
h
O
E
O
u
tt
B
T
h
1
E
(G
E
(Bu +h
E

Nu
t
) +

G
E
(Bv +h
E

Nv
t
))
_
. Then we have
the following by (7.6):
/(

G
E
)Y, W)
c
=
_
v, v, ,

T
, [s, s, t, t]
T
)
c
= m
tt
, t)

+h
O
p
O
, t)

+Kv
tt
, s
tt
)

+h
O
E
O
v
t
, s
t
)

+G
E
h
1
E
(Bv +h
E

Nv
t
), (Bs +h
E

Ns
t
))

,
where we used the boundary condition t = 0 at x = 0, L and integration by parts on the term
on involving . Therefore,
/(

G
E
)Y, W)
c
=
_
Ku
tttt
+D
x

N
T
_
G
E
(Bu +h
E

Nu
t
) +

G
E
(Bv +h
E

Nv
t
)
_
, t
_

+
_
h
O
E
O
u
tt
B
T
h
1
E
_
G
E
(Bu +h
E

Nu
t
) +

G
E
(Bv +h
E

Nv
t
)
_
, t
_

+Kv
tt
, s
tt
)

+h
O
E
O
v
t
, s
t
)

+G
E
h
1
E
(Bv +h
E

Nv
t
), (Bs +h
E

Ns
t
))

.
Using integrations by parts along with the boundary conditions (7.12), we obtain
/(

G
E
)Y, W)
c
= Ku
tt
, t
tt
)

G
E
(Bu +h
E

Nu
t
),

Nt
t
)

G
E
(Bv +h
E

Nv
t
),

Nt
t
)

h
O
E
O
u
t
, t
t
)

G
E
(Bu +h
E

Nu
t
), h
1
E
Bt)

G
E
(Bv +h
E

Nv
t
), h
1
E
Bt)

+Kv
tt
, s
tt
)

+h
O
E
O
v
t
, s
t
)

+G
E
h
1
E
(Bv +h
E

Nv
t
), (Bs +h
E

Ns
t
))

.
117
Gathering terms, we get
/(

G
E
)Y, W)
c
=
_
Kv
tt
, s
tt
)

Ku
tt
, t
tt
)

_
+
_
h
O
E
O
v
t
, s
t
)

h
O
E
O
u
t
, t
t
)

_
+
_
G
E
h
1
E
(Bv +h
E

Nv
t
), (Bs +h
E

Ns
t
))

G
E
(Bu +h
E

Nu
t
), h
1
E
(Bt +h
E

Nt
t
))

G
E
(Bv +h
E

Nv
t
), h
1
E
(Bt +h
E

Nt
t
))

. (7.13)
Next dene := J
1
_
Ku
tttt
D
x

N
T
h
E
G
E
(Bu +h
E

Nu
t
) +D
x

N
T
h
E

G
E
(Bv +h
E

Nv
t
)
_
and
:= h
1
O
p
1
O
_
h
O
E
O
u
tt
+B
T
G
E
(Bu +h
E

Nu
t
) B
T

G
E
(Bv +h
E

Nv
t
)
_
. Then we have
the following by (7.6):
Y, /(

G
E
)W)
c
= [u, u, v, v]
T
,
_
t, t, ,

T
)
c
= v, m
tt
)

+h
O
p
O
v, )

+Ku
tt
, t
tt
)

+h
O
E
O
u
t
, t
t
)

+G
E
h
1
E
(Bu +h
E

Nu
t
), (Bt +h
E

Nt
t
))

,
where we used the boundary condition v = 0 at x = 0, L and integration by parts on the term
on involving . Therefore,
Y, /(

G
E
)W)
c
=
_
v, Ks
tttt
D
x

N
T
G
E
(Bs +h
E

Ns
t
) +D
x

N
T

G
E
(Bt +h
E

Nt
t
)
_

+
_
v, h
O
E
O
s
tt
+B
T
h
1
E
G
E
(Bs +h
E

Ns
t
) B
T
h
1
E

G
E
(Bt +h
E

Nt
t
)
_

Ku
tt
, t
tt
)

h
O
E
O
u
t
, t
t
)

G
E
h
1
E
(Bu +h
E

Nu
t
), (Bt +h
E

Nt
t
))

.
Using integrations by parts along with the boundary conditions (7.12), we obtain
Y, /(

G
E
)W)
c
= Kv
tt
, s
tt
)

Nv
t
, G
E
(Bs +h
E

Ns
t
))

Nv
t
,

G
E
(Bt +h
E

Nt
t
))

+v
t
, h
O
E
O
s
t
)

+h
1
E
Bv, G
E
(Bs +h
E

Ns
t
))

h
1
E
Bv,

G
E
(Bt +h
E

Nt
t
))

Ku
tt
, t
tt
)

h
O
E
O
u
t
, t
t
)

G
E
h
1
E
(Bu +h
E

Nu
t
), (Bt +h
E

Nt
t
))

.
118
Gathering terms, we get
Y, /(

G
E
)W)
c
=
_
Kv
tt
, s
tt
)

Ku
tt
, t
tt
)

_
+
_
v
t
, h
O
E
O
s
t
)

h
O
E
O
u
t
, t
t
)

_
+
_
h
1
E
(Bv +h
E

Nv
t
), G
E
(Bs +h
E

Ns
t
))

h
1
E
(Bu +h
E

Nu
t
), G
E
(Bt +h
E

Nt
t
))

G
E
(Bv +h
E

Nv
t
), h
1
E
(Bt +h
E

Nt
t
))

. (7.14)
The right side of (7.13) and (7.14) agree, and thus the result of this lemma follows.
Remark 7.4. As a consequence of the previous lemma,
T(/) = T(/

).
Next, we show that / is dissipative on H. According to (7.13),
/Y, Y )
c
=
_
Kv
tt
, u
tt
)

Ku
tt
, v
tt
)

_
+
_
v
t
, h
O
E
O
u
t
)

h
O
E
O
u
t
, v
t
)

_
+
_
(Bv +h
E

Nv
t
), G
E
h
1
E
(Bu +h
E

Nu
t
))

G
E
h
1
E
(Bu +h
E

Nu
t
), (Bv +h
E

Nv
t
))

G
E
h
1
E
(Bv +h
E

Nv
t
), (Bv +h
E

Nv
t
))

.
Notice that the rst three pairs of terms on the right side of the above expression (grouped
together by ) are purely imaginary since
d
dt
[u, u]
T
= [v, v]
T
. Therefore,
Re /Y, Y )
c
= h
1
E

G
E
(Bv +h
E

Nv
t
), (Bv +h
E

Nv
t
))

. (7.15)
From Remark 7.3 we know that
^(/) = ^(/

) = span([0,

0
O
, 0,

1
O
f(x)]
T
).
Hence, we dene the following subspace of H:
1 := [^(/)]

.
Therefore, we will view H as 1 ^(/). Using the results of (7.15), Lemma 7.1, and Theorem
2.3, we have the following theorem.
Theorem 7.1. / is dissipative on H, and it is strictly dissipative on 1. Consequently, / is
the generator of a C
0
semigroup of contractions on 1.
119
7.3 Eigenvalues of /
Suppose is an eigenvalue for / and Y is the corresponding eigenvalue. Then /Y = Y
implies
V = U (7.16)
A
1
U +A
2
U =
2
U. (7.17)
7.3.1 Imaginary eigenvalues
We are particularly interested in establishing necessary and sucient conditions needed
for / to have no eigenvalues on the imaginary axis. Suppose i is an eigenvalue of /, and
Z = [ u, u, v, v]
T
is the corresponding eigenvector. Since / has no zero eigenvalues in 1, we
must have R 0. (7.15) implies the following:
0 = Re iZ /Z, Z)
c
= h
1
E

G
E
(B v h
E

N v
t
), (B v h
E

N v
t
))

. (7.18)
Therefore, if Y = [u, u, v, v]
T
is a solution to

Y = /Y, Y (0) = Z,
then Y must satisfy
Bv h
E

Nv
t
= 0 B v h
E

N v
t
= 0. (7.19)
But
d
dt
[u, u]
T
= [v, v]
T
= i[u, u]
T
, and since ,= 0, we have
Bu h
E

Nu
t
= 0 B u h
E

N u
t
. (7.20)
Substitution of (7.19) and (7.20) into (7.17) leads to
_

_
J
1
[K u
tttt
]
h
1
O
p
1
O
h
O
E
O
u
tt
_

_ +i
_

_
0
0
_

_ =
2
_

_
u
u
_

_
This can be rewritten as follows:
_

_
K u
tttt
+
2
u
tt
m
2
u = 0
u
tt
=
2
E
1
O
p
O
u
B u = h
E
N u
t
(7.21)
120
For convenience, dene

k
=
k
L
,
k
=

K
+
m

2
k
; k N,
and let

i
=

E
i

i
, i = 1, 3, 5, , n
denote the wave speeds in the odd-indexed layers.
Lemma 7.2. Any solution [ u, u]
T
of (7.21) is one of the following:
u(x) = 0 and u(x) = c

1
O
cos(
k
x); k N, c R, (7.22)
or
u(x) = a sin(
k
x) and u =
_

_
C
1
cos
_

1
x
_
C
3
cos
_

3
x
_
.
.
.
C
n
cos
_

k
n
x
_
_

_
; k N, C
i
R, a R 0. (7.23)
Proof. If u = 0, then the third line of (7.21) reads B u = 0. Therefore,
u = c

1
O
f(x).
But the second line of (7.21) along with the boundary conditions u
t
(0) = u
t
(L) = 0 imply
f(x) = c cos(
k
x) for any k N, and so
u = c

1
O
cos(
k
x); k N.
Next, we assume that u ,= 0. Solutions to the second equation in (7.21) are of the following
form due to the boundary condition u
t
(0) = 0:
u =
_

_
C
1
cos
_

1
x
_
C
3
cos
_

3
x
_
.
.
.
C
n
cos
_

n
x
_
_

_
(7.24)
121
Let the vector [H
2
, H
4
, , H
n1
]
T
denote h
E

N. Then (7.24) and the third line of (7.21) imply
C
1
cos
_

1
x
_
C
3
cos
_

3
x
_
= H
2
u
t
(x)
.
.
.
C
n2
cos
_

n2
x
_
C
n
cos
_

n
x
_
= H
n1
u
t
(x) (7.25)
Next, we claim that u must be of the form
u(x) = a sin(bx) +c sin(dx).
To see this, let i, j be any pair of indices in 1, 3, , n with i < j, and consider the following
subsystem of (7.25):
C
i
cos
_

i
x
_
C
i+2
cos
_

i+2
x
_
= H
i+1
u
t
(x)
.
.
.
C
j2
cos
_

j2
x
_
C
j
cos
_

j
x
_
= H
j1
u
t
(x)
Adding all of the lines of this system together leads to
C
i
cos
_

i
x
_
C
j
cos
_

j
x
_
= (H
i+1
+H
i+3
+ +H
j1
) u
t
(x).
Then the boundary condition u(0) = 0 implies
u(x) =
C
i

i
H(i, j)
sin
_

i
x
_
+
C
j

j
H(i, j)
sin
_

j
x
_
, (7.26)
where
H(i, j) = H
i+1
+H
i+3
+ +H
j1
.
Therefore, in order for u(x) in (7.26) to be of the desired form a sin(
k
x), one of the following
must hold for any pair of indices i, j 1, 3, , n:

i
=
j
, C
i
= 0, or C
j
= 0.
By way of contradiction, suppose there is a pair of indices i, j 1, 3, , n with
i
,=
j
,
C
i
,= 0, and C
j
,= 0. Then the boundary condition u(L) = 0 implies the following:

i
=
k
,

j
=

; k, N.
122
Since we are assuming
i
,=
k
, we have k ,= in the above. Therefore, (7.26) implies
u(x) =
C
i

k
H(i, j)
sin(
k
x) +
C
j

H(i, j)
sin(

x).
Substitution of this result into the rst line of (7.21) leads to
C
i

k
H(i, j)
_
K
4
k

2
k
m
2

sin(
k
x) =
C
j

H(i, j)
_
K
4

m
2

sin(

x).
Since
k
,=

, we have the equations

K
4
k

2
k
m
2
= 0, K
4

m
2
= 0.
Hence

2
k
=
2

=

2
+

4
+ 4Km
2
2K
,
since
2

4
+ 4Km
2
< 0, and
2
k
and
2

are positive. By our assumptions,

k
,=

,
so we must have
k
=

. But this is impossible since

k
and

are both positive. Thus we

have reached a contradiction. Hence,
u
k
(x) = a
k
sin(
k
x); k N.
Substitution of these solutions into the rst line of (7.21) leads to:
K
4
k

2
m
2
= 0.
Solving this equation for
2
gives

2
=
2
k
_
K
+
m

2
k
_
=
2
k

2
k
,
but since
k
and
k
are positive, we have
=
k

k
.
Insert this result into (7.24) and obtain
u =
_

_
C
1
cos
_

1
x
_
C
3
cos
_

3
x
_
.
.
.
C
n
cos
_

k
n
x
_
_

_
. (7.27)
123
Remark 7.5. The c R in (7.22) is a constant in terms of x, but it may be a function of t.
In fact, the next lemma will consider a case when this happens.
Lemma 7.3. If there are 2, 3, , m wave speeds that do not intersect
k

kN
= , then /
has no imaginary eigenvalues. If none of the wave speeds intersect
k

kN
= , then / has
an innite family of imaginary eigenvalues provided that all of the wave speeds are equal.
Proof. Suppose i, j 1, 3, , n with i < j. In addition, suppose
i
and
j
are the only two
wave speeds that do not intersect
k

kN
. If we insert (7.27) into the last line of (7.21), one
arrives at the following system of m equations:
C
1
cos
_

1
x
_
C
3
cos
_

3
x
_
= a
k
H
2
cos(
k
x) (7.28)
.
.
.
C
i
cos
_

i
x
_
C
i+2
cos
_

i+2
x
_
= a
k
H
i+1
cos(
k
x) (7.29)
.
.
.
C
j2
cos
_

j2
x
_
C
j
cos
_

j
x
_
= a
k
H
j1
cos(
k
x) (7.30)
.
.
.
C
n2
cos
_

n2
x
_
C
n
cos
_

n
x
_
= a
k
H
n1
cos(
k
x) (7.31)
Add lines (7.29) through (7.30) together and obtain
C
i
cos
_

i
x
_
C
j
cos
_

j
x
_
= a
k
H(i, j) cos(
k
x),
where H(i, j) is dened as before. Since
i
and
j
are not equal to
k
for any k and H(i, j) > 0,
then
a = 0.
According to Lemma 7.2, this leads to u = 0 and u = c

1
0
cos(
k
x). If c ,= 0, then all of the
wave speeds must be the same. But this is not the case, and so we must have c = 0. Hence,
u = 0 and u = 0, which leads to v = 0, v = 0, and Z = 0.
124
If there are m wave speeds that do not intersect
k

kN
, then there is a j 1, 3, , n and
a k
0
N such that
j
=
k
0
. The equations
C
j2
cos
_

k
0

k
0

j2
x
_
C
j
cos
_

k
0

k
0

j
x
_
= a
k
0
H
j1
cos(
k
0
x)
C
j
cos
_

k
0

k
0

j
x
_
C
j+2
cos
_

k
0

k
0

j+2
x
_
= a
k
0
H
j+1
cos(
k
0
x)
imply that
C
j2
= C
j+2
= 0 and C
j
= a
k
0
H
j1
= a
k
0
H
j+1
.
This implies that a
k
0
(H
j1
+H
j+1
) = 0, and since h
E

N has only positive constants, we must
have a = 0. Then we again have u = c

1
O
cos(
k
x) by Lemma 7.2. But c = 0 because not all
waves speeds are identical and thus we arrive at Z = 0 once again.
If there are 3, 4, , or m1 wave speeds that do not intersect
k

kN
, an argument similar
to the above ones can be used to arrive at Z = 0 once again.
Finally, suppose none of the wave speeds intersect the set
k

kN
. Then for every i =
1, 3, , n, the equation
C
i
cos
_

i
x
_
C
i+2
cos
_

i+2
x
_
= a
k
H
i+1
cos(
k
x)
implies that a = 0. This leads to u = 0, and then Lemma 7.2 implies
u = c(t)

1
O
cos(
k
x); k N.
If c(t) ,= 0, then all of the wave speeds are the same; that is
=
i
=
k
, k N, i = 1, 3, , n.
Thus p
1
O
E
O
=
2
I
m+1
. Using the denition of (7.4) and (7.5), we have
A
1
_

_
0
c(t)

1
O
cos(
k
x)
_

_ =
_

_
0
c(t)
2
k
p
1
O
E
O

1
O
cos(
k
x)
_

_ =
2

2
k
_

_
0
c(t)

1
O
cos(
k
x)
_

_,
A
2
_

_
0
c(t)

1
O
cos(
k
x)
_

_ =
_

_
0
0
_

_
125
Thus, if is an eigenvalue of /, (7.17) implies
A
1
_

_
0
c(t)

1
O
cos(
k
x)
_

_ =
2
_

_
0
c(t)

1
O
cos(
k
x)
_

_.
Therefore,
= i
k
.
Substitute this into (7.16) and obtain
_

_
0
c(t)

1
O
cos(
k
x)
_

_ = i
k
_

_
0
c(t)

1
O
cos(
k
x)
_

_
Therefore,
c(t) = c
0
e
i(
k
t)
; c
0
R.
Since c
0
= 0 leads to Z = 0 we may assume without loss of generality that c
0
= 1. Therefore,
for all k N, i
k
are the eigenvalues for / with corresponding eigenvectors
u = 0, u
k
= e
i(
k
t)

1
O
cos(
k
x), v = 0, v
k
= i
k
e
i(
k
t)

1
O
cos(
k
x).
Remark 7.6. The above lemma holds even if
i
and
j
are the only two wave speeds that do
not intersect
k

kN
and
i
=
j
.
Dene the set of wave speeds
S :=
1
,
3
, ,
n
.
Therefore, in order for / to have nonzero imaginary eigenvalues, Remark 7.6 implies that we
can rule out possibilities for the set S such as

k
1
, ,
k
i1
, ,
k
i+1
, ,
k
j1
, ,
k
j+1
, ,
km
,

k
1
, ,
k
i1
, , ,
k
i+2
, ,
km
, etc.
Therefore, the only way / can have any imaginary eigenvalues is if either S = , , , ,
where ,=
k
for all k N (as in Lemma 7.3), or there is at most one wave speed that does
not intersect the set
k

kN
.
126
Lemma 7.4. If there exists one wave speed
j
that is distinct from the others, then / has a
pair of imaginary eigenvalues. Furthermore, all of the rest of the wave speeds must equal
k
0
for some k
0
N in this case.
Proof. Consider once again the system of m equations in (7.28 - 7.31), namely:
C
1
cos
_

1
x
_
C
3
cos
_

3
x
_
= a
k
H
2
cos(
k
x) (7.32)
C
3
cos
_

3
x
_
C
5
cos
_

5
x
_
= a
k
H
4
cos(
k
x) (7.33)
C
5
cos
_

5
x
_
C
7
cos
_

7
x
_
= a
k
H
6
cos(
k
x) (7.34)
.
.
.
C
n6
cos
_

n6
x
_
C
n4
cos
_

n4
x
_
= a
k
H
n5
cos(
k
x) (7.35)
C
n4
cos
_

n4
x
_
C
n2
cos
_

n2
x
_
= a
k
H
n3
cos(
k
x) (7.36)
C
n2
cos
_

n2
x
_
C
n
cos
_

n
x
_
= a
k
H
n1
cos(
k
x) (7.37)
Suppose
1
,=
3
and
3
=
k
0
for some k
0
N. Then (7.32) (at k = k
0
) implies
C
1
cos
_

k
0

k
0

1
x
_
= (C
3
+a
k
0
H
2
) cos(
k
0
x).
Since cos
_

k
0

k
0

1
x
_
and cos(
k
0
x) are linearly independent, we have
C
1
= 0, and C
3
= a
k
0
H
2
. (7.38)
Substitute this result into (7.33) (with k = k
0
) and obtain
C
5
cos
_

k
0

k
0

5
x
_
= a
k
0
(H
2
+H
4
) cos(
k
0
x)
Note that we must have
5
=
k
0
. If this were not the case, we would get a = 0, which leads
to solutions of the form (7.22). Furthermore, since we have a distinct wave speed, we would
get u = 0, and then Z = 0. So since
5
=
k
0
, we have
C
5
= a
k
0
(H
2
+H
4
).
Substitute this result into (7.34) (with k = k
0
) and obtain
C
7
cos
_

k
0

k
0

7
x
_
= a
k
0
(H
2
+H
4
+H
6
) cos(
k
0
x)
127
This forces
7
=
k
0
(otherwise we get Z = 0 again by the same reasoning as before). If we
continue in this fashion we obtain the following if
1
,=
k
0
:

i
=
k
0
and C
i
= a
k
0
(H
2
+H
4
+ +H
i1
); i = 3, 5, 7, , n. (7.39)
Now suppose
n
,=
n2
and
n2
=
k
0
for some k
0
N. Then (7.37) (at k = k
0
) implies
C
n
cos
_

k
0

k
0

n
x
_
= (a
k
0
H
n1
+C
n2
) cos(
k
0
x)
Thus,
C
n
= 0, and C
n2
= a
k
0
H
n1
.
Substitution of this result into (7.36) (with k = k
0
) leads to
C
n4
cos
_

k
0

k
0

n4
x
_
= a
k
0
(H
n3
+H
n1
) cos(
k
0
x).
This forces

n4
=
k
0
,
and then we obtain
C
n4
= a
k
0
(H
n3
+H
n1
).
Similar to before, substituting the last result into (7.35) (with k = k
0
) forces

n6
=
k
0
, and C
n6
= a
k
0
(H
n5
+H
n3
+H
n1
).
Continuing upward in this fashion, we have

i
=
k
0
and C
i
= a
k
0
(H
i+1
+H
i+3
+ +H
n1
); i = 1, 3, 5, , n 2. (7.40)
Next, suppose there is some j 3, 5, , n2 such that
j
,=
j+2
and
j+2
=
k
0
for some
k
0
N. Using an argument similar to before, the equation
C
j2
cos
_

k
0

k
0

j2
x
_
C
j
cos
_

k
0

k
0

j
x
_
= a
k
0
H
j1
cos(
k
0
x)
forces

j2
=
k
0
128
and leads to
C
j
= 0, and C
j2
= a
k
0
H
j1
.
Then substitution of this result into
C
j4
cos
_

k
0

k
0

j4
x
_
C
j2
cos
_

k
0

k
0

j2
x
_
= a
k
0
H
j3
cos(
k
0
x)
(if j ,= 3) forces

j4
=
k
0
and implies
C
j4
= a
k
0
(H
j3
+H
j1
).
This continues to propagate upward, and we obtain

i
=
k
0
and C
i
= a
k
0
(H
i+1
+H
i+3
+ +H
j1
); i = 1, 3, , j 2.
Since C
j
= 0, the equation
C
j
cos
_

k
0

k
0

j
x
_
C
j+2
cos
_

k
0

k
0

j+2
x
_
= a
k
0
H
j+1
cos(
k
0
x)
leads to

j+2
=
k
0
and C
j+2
= a
k
0
H
j+1
Similar to before, we can propagate downward for i = j + 2, , n and obtain

i
=
k
0
and C
i
= a
k
0
(H
i+1
+H
i+3
+ +H
n1
); i = j + 2, j + 4, , n 2.
Therefore for any j 1, 3, , n, if
j
is the distinct wave speed, we have that u must be of
the form
u = c

A
j
cos(
k
0
x); k
0
N,
129
where

A
j
=
_

_
H
2
+H
4
+ +H
j1
H
4
+ +H
j1
.
.
.
H
j1
0
H
j+1
.
.
.
(H
j+1
+ +H
n1
)
_

_
.
In the case of Lemma 7.4, we have that S is of the form
S =
k
0
, ,
k
0
,
j
,
k
0
, ,
k
0
(
j
,=
k
0
). (7.41)
Remark 7.7. In the previous lemma, the distinct wave speed
j
could either be dierent than

k
for all k N, or it could equal
k
1
for some k
1
N other than k
0
.
Remark 7.8. If S is of the form (7.41), one can verify directly (by substituting into (7.3 - 7.5)
and using (7.16) and (7.17)) that i
k
0

k
0
are the eigenvalues of / with the corresponding
eigenvectors Z

= [ u

, u

, v

, v

]
T
, where
u

(x, t) = e
i(
k
0

k
0
)t
sin(
k
0
x), u

(x, t) =
k
0
e
i(
k
0

k
0
)t

A
j
cos(
k
0
x)
v

(x, t) = i
k
0

k
0
e
i(
k
0

k
0
)t
sin(
k
0
x), v

(x, t) = i
2
k
0

k
0
e
i(
k
0

k
0
)t

A
j
cos(
k
0
x).
Finally, if the set S has the form
S =
k
0
,
k
0
, ,
k
0
. (7.42)
for some k
0
N, then a solution u of (7.21) must be of the form
u(x) =

C cos(
k
0
x);

C = [C
1
, C
3
, , C
n
].
130
One can show that if S is of the form (7.42), the eigenvalues i
k
0

k
0
correspond to the
eigenvectors Z

= [ u

, u

, v

, v

]
T
, which are given by the following:
u

(x, t) = e
i(
k
0

k
0
)t
sin(
k
0
x),
u

(x, t) =
k
0
e
i(
k
0

k
0
)t
_

A
n
+

1
O
_
cos(
k
0
x)
v

(x, t) = i
k
0

k
0
e
i(
k
0

k
0
)t
sin(
k
0
x),
v

(x, t) = i
2
k
0

k
0
e
i(
k
0

k
0
)t
_

A
n
+

1
O
_
cos(
k
0
x).
Putting all of the results of this section together, we have the following theorem:
Theorem 7.2. / has eigenvalues on the imaginary axis if and only if one of the following
occurs:
S = , , , ,=
k
for all k N,
S =
k
0
, ,
k
0
,
j
,
k
0
, ,
k
0

j
,=
k
0
for some k
0
N, or
S =
k
0
,
k
0
, ,
k
0
for some k
0
N.
Equivalently, / has no spectrum on the imaginary axis if and only if both of the following
conditions are satised:
(i.) There exist at least two distinct wave speeds.
(ii.) The following does not occur:
There is an i 1, 3, , n such that
i
is distinct, and all of the other
j
are
equal to
k
0
for some k
0
N.
7.3.2 Other eigenvalues
Suppose that the wave speeds
_
K

,
1
,
3
, ,
n
are distinct. Hansen [14] has shown that
under these conditions, the eigenvalues of / occur asymptotically as conjugate pairs as follows:

k,0
=

N
T

G
E
h
E

N
2
i
k
_
K

+O
_
1
k
_

k,j
=

j1
+
j+1
2h
j

j
i
k

j
+O
_
1
k
_
; j = 1, 3, n,
131
where
0
=
n+1
= 0 and
i
=

G
i
h
i
for i = 2, 4, , 2m. Therefore, for each j = 0, 1, 3, , n
we have
lim
k
arg(

k,j
) = /2.
As a consequence, we have the following:
Theorem 7.3. If the wave speeds
_
K

,
1
,
3
, ,
n
are distinct, then the semigroup gener-
ated by / is not analytic.
7.4 Exponential Stability of /
We are now prepared to prove the following result.
Theorem 7.4. / is the generator of an exponentially stable semigroup on 1 if and only if
both of the following conditions are satised:
(i.) There exist at least two distinct wave speeds.
(ii.) The following does not occur:
There is an i 1, 3, , n such that
i
is distinct, and all of the other
j
are
equal to
k
0
for some k
0
N.
Proof. We rst prove the reverse implication by contradiction. Conditions (i) and (ii) being
satised implies iR (/). According to Theorem 2.6, / is the generator of an exponentially
stable semigroup if and only if iR (/) and sup
R
_
_
(iI /)
1
_
_
< . Equivalently, we show
that there exists > 0 such that inf
|Y |
E
=1
|(isI /)Y |
c
. Suppose on the contrary that this
condition does not hold. Then there exists a sequence of real numbers s
k
and a sequence of
state vectors Y
k
in T(/) with |Y
k
|
c
= 1 such that
lim
k
|is
k
Y
k
/Y
k
|
c
= 0. (7.43)
Note that Y
k
:= [u
k
, u
k
, v
k
, v
k
]
T
. Since the norm of the resolvent is symmetric about the
imaginary axis, we can assume without loss of generality that s
k
> 0. Moreover, iR (/),
so if the sequence s
k

k=1
were to remain in a compact set, (7.43) would be violated. So we
132
can also assume without loss of generality that s
k
+. We can write (7.43) explicitly as
follows:
is
k
u
k
v
k
:= g
k
0 in H
2
(0, L) H
1
0
(0, L) (7.44)
is
k
u
k
v
k
:= g
k
0 in (H
1
(0, L))
m+1
(7.45)
is
k
v
k
+J
1
_
Ku
tttt
k


N
T
h
E
G
E

t
k

N
T
h
E

G
E

t
k
_
:= h
k
0 in H
1
0
(0, L) (7.46)
is
k
h
O
p
O
v
k
h
O
E
O
u
tt
k
+B
T
_
G
E

k
+

G
E

k
_
:= h
k
0 in (L
2
(0, L))
m+1
(7.47)
In addition, dene
z
k
:= p
1
O
E
O
u
k
. (7.48)
We proceed by showing the following steps in the indicated order:
1. s
k

k
0 in (L
2
(0, L))
m
.
2. is
k
v
k
z
tt
k
0 in (L
2
(0, L))
m+1
.
3. v
k
0 in L
2
(0, L).
4. is
k
Jv
k
, u
k
)

+K|u
tt
k
|
2
L
2
0.
5. is
k
Jv
k
, u
k
)

+|v
t
k
|
2
L
2
0.
6. u
n
0 in H
2
().
7. v
n
0 in H
1
().
8. Bz
t
k
0 in (L
2
(0, L))
m
.
9. u
k
0 in (H
1
(0, L))
m+1
.
10. v
k
0 in (L
2
(0, L))
m+1
.
Proof of (1) and (2): If we assume (7.43) holds, then
Re is
k
Y
k
/Y
k
, Y
k
)
c
0.
133
But (7.15) implies

k
0 in (L
2
(0, L))
m
. (7.49)
According to (7.44),
is
k

Nu
t
k

Nv
t
k
0 in (L
2
(0, L))
m
, (7.50)
and according to (7.45),
is
k
h
1
E
Bu
k
h
1
E
Bv
k
0 in (L
2
(0, L))
m
(7.51)
Adding (7.50) and (7.51) together, yields
is
k

k
0 in (L
2
(0, L))
m
. (7.52)
Then (1) follows from (7.49) and (7.52). Since s
k
, it immediately follows from (7.52)
that

k
0 in (L
2
(0, L))
m
. (7.53)
Insert the results of (7.49) and (7.53) into (7.47) and obtain
is
k
h
O
p
O
v
k
h
O
E
O
u
tt
k
0 in (L
2
(0, L))
m+1
.
Multiplying the last result on the left by (h
O
p
O
)
1
yields (2).
Proof of (3): Next, we look at (7.46). Note that
Ku
tttt
k
=
K

_
u
tttt
k
mu
tt
k

+
Km

u
tt
k
=
Km

u
tt
k

J(u
tt
k
).
Therefore, (7.46) implies
is
k
v
k

u
tt
k
+J
1
_
Km

u
tt
k

N
T
h
E
G
E

t
k

N
T
h
E

G
E

t
k
_
0 in L
2
(0, L). (7.54)
Since Y
k
T(/), we have
Km

u
tt
k

Nh
E
G
E

t
k

Nh
E

G
E

t
k
L
2
(0, L).
Therefore, it follows from the denition of J that
J
1
_
Km

u
tt
k

N
T
h
E
G
E

t
k

N
T
h
E

G
E

t
k
_
H
1
0
(0, L). (7.55)
134
Moreover, the sequences u
tt
k
,

N
T
h
E

t
k
, and

N
T
h
E

t
k
are bounded in H
1
(0, L), and
J
1
is an isomorphism from H
1
(0, L) to H
1
(0, L). Therefore,
_
J
1
_
Km

u
tt
k

N
T
h
E
G
E

t
k

N
T
h
E

G
E

t
k
__

k=1
is a bounded sequence in H
1
(0, L),
and so it certainly follows that
_
J
1
_
Km

u
tt
k

N
T
h
E
G
E

t
k

N
T
h
E

G
E

t
k
__

k=1
is a bounded sequence in L
2
(0, L).
Combine this result with (7.54) to deduce that
the sequence s
k
v
k
u
tt
k

k=1
is bounded in L
2
(0, L).
But since |u
tt
k
|
L
2 |Y
k
|
c
1, we then have that
the sequence s
k
v
k

k=1
is bounded in L
2
(0, L). (7.56)
Since s
k
, (3) follows from (7.56).
Proof of (4): Next we take the L
2
inner product of (7.54) with u
tt
k
(multiply on the left)
and get the following:
is
k
v
k
, u
tt
k
)

+K|u
tt
k
|
2
L
2

_
J
1
_
Km

u
tt
k

N
T
h
E
G
E

t
k

N
T
h
E

G
E

t
k
__
, u
tt
k
_

0, (7.57)
We use (7.55) to integrate the last term on the left side of (7.57) by parts, and we use the
boundary condition v
k
= 0 for x = 0, L for integrating the rst term by parts. This leads to
is
k
v
t
k
, u
t
k
)

+K|u
tt
k
|
2
L
2
+
_
D
x
_
J
1
_
Km

u
tt
k

N
T
h
E
G
E

t
k

N
T
h
E

G
E

t
k
__
, u
t
k
_

0.
Using the boundary condition u
k
= 0 for x = 0, L, we can apply further integrations by parts
to arrive at
is
k
v
tt
k
, u
k
)

+K|u
tt
k
|
2
L
2

_
D
2
x
_
J
1
_
Km

u
tt
k

N
T
h
E
G
E

t
k

N
T
h
E

G
E

t
k
__
, u
k
_

0. (7.58)
135
Next, take the L
2
inner product of (7.54) with mu
k
(multiply on the left) and obtain the
following:
is
k
mv
k
, u
k
)

Km

u
tt
k
, u
k
)

+
_
mJ
1
_
Km

u
tt
k

N
T
h
E
G
E

t
k

N
T
h
E

G
E

t
k
_
, u
k
_

0. (7.59)
Adding (7.58) and (7.59) together and using the denition of the operator J leads to the
following:
is
k
Jv
k
, u
k
)

+K|u
tt
k
|
2
L
2

Km

u
tt
k
, u
k
)

+
_
Km

u
tt
k

N
T
h
E
G
E

t
k

N
T
h
E

G
E

t
k
, u
k
_

0.
Using integration by parts along with the boundary condition u
k
(0) = u
k
(L) = 0, the above
can be written
is
k
Jv
k
, u
k
)

+K|u
tt
k
|
2
L
2
+

N
T
h
E
G
E

k
, u
t
k
)

N
T
h
E

G
E

k
, u
t
k
)

0. (7.60)
Since |u
t
k
|
L
2 |Y
k
|
c
= 1, (7.53) implies that the third term on the right side of (7.60) goes
to zero. Similarly, (7.49) implies that the fourth term on the right side of (7.60) goes to zero.
Therefore, (4) follows from (7.60).
Proof of (5): Going back to (7.44), we have that
is
k
u
k
v
k
0 in L
2
(0, L).
Therefore, (3) implies that
s
k
u
k
0 in L
2
(0, L). (7.61)
So by (3) and (7.61) we have
is
k
u
k
+v
k
0 in L
2
(0, L). (7.62)
Take the L
2
inner product of (7.62) with Jv
k
and obtain
is
k
Jv
k
, u
k
)

+m|v
k
|
2
L
2
+|v
t
k
|
2
L
2
0, (7.63)
where we use integration by parts in the last term on the right side of (7.63) along with the
boundary condition v
k
(0) = v
k
(L) = 0. The second term on the right side of (7.63) goes to 0
by (3). Thus (5) follows.
136
Proof of (6) and (7): Add (4) and (5) together and obtain
K|u
tt
k
|
2
L
2
+|v
t
k
|
2
L
2
0, (7.64)
Since all terms on the left side of (7.64) are positive real numbers, we have the following:
u
tt
k
0 in L
2
(0, L), and (7.65)
v
t
k
0 in L
2
(0, L). (7.66)
Since u
k
(0) = u
k
(L) = 0, standard elliptic estimates applied to (7.65) implies (6) In addition,
the results of (3) and (7.66) imply (7).
Proof of (8): Since Bv
k
= h
E
(

Nv
t
k
), (7.49) and (7.66) imply
Bv
k
0 in (L
2
(0, L))
m
. (7.67)
Since the sequence v
k
is bounded in (L
2
(0, L))
m+1
, (7.45) implies that
s
k
u
k

k=1
is a bounded sequence in (L
2
(0, L))
m+1
. (7.68)
According to (2), we have
is
k
Bv
k
Bz
tt
k
0 in (L
2
(0, L))
m
. (7.69)
Take the L
2
inner product of this with respect to Bz
k
and obtain
Bv
k
, is
k
Bz
k
)

+|Bz
t
k
|
2
L
2
0, (7.70)
where in the last term on the left side we used z
t
k
= 0 for x = 0, L with hinged boundary
conditions. Note that the rst term on the right side of (7.70) goes to zero by (7.67) and
(7.68). This implies (8)
Proof of (9): The analysis for this step is a bit more dicult because Bz
t
k
0 does not
necessarily mean z
t
k
0 because

1
O
is in the null space of B. We will need to consider a
system of limits, and show that each component of z
t
k
goes to 0. Using (7.48), (7.45) can be
written as
is
k
E
1
O
p
O
z
k
v
k
= g
k
0 in (H
1
(0, L))
m+1
. (7.71)
137
Therefore, we have by (2) and (7.71) that
f
k
0 in (L
2
(0, L))
m+1
, where f
k
:= s
2
k
E
1
O
p
O
z
k
+z
tt
k
+is
k
g
k
. (7.72)
Dene
i
:=
_

i
E
i
_
=
1

i
_
for i = 1, 3, , n. If we write out (7.72) explicitly, we have the
following:
s
2
k

2
1
z
k,1
+z
tt
k,1
= f
k,1
is
k
g
k,1
0 in L
2
(0, L)
s
2
k

2
3
z
k,3
+z
tt
k,3
= f
k,3
is
k
g
k,3
0 in L
2
(0, L)
.
.
.
s
2
k

2
n
z
k,n
+z
tt
k,n
= f
k,n
is
k
g
k,n
0 in L
2
(0, L). (7.73)
Using Laplace transforms, one nds that the solution of (7.73) is the following:
z
k
=
_

_
C
k,1
cos(s
k

1
x)
C
k,3
cos(s
k

3
x)
.
.
.
C
k,n
cos(s
k

n
x)
_

_
+
_

k,1
(x)

k,3
(x)
.
.
.

k,n
(x)
_

_
, (7.74)
where for each i = 1, 3, , n,

k,i
(x) :=
1

i
_
L
0
sin(s
k

i
(x y))
_
s
1
k
f
k,1
(y) ig
k,1
(y)
_
dy,
and C
k,i
is a constant.
We next show that s
k

k,i
0 in L
2
(0, L). For all i = 1, 3, , n, we have the following:
[s
k

k,i
(x)[
1

i
_
L
0
[f
k,i
(y)[ dy +
1

_
L
0
s
k
sin(s
k

i
(x y))g
k,i
(y) dy

i
|f
k,i
|
(L
2
)
m+1 +
1

2
i

cos(s
k

i
(x y))g
k,i
(y)

L
0

2
i
|g
t
k,i
|
(L
2
)
m+1

i
|f
k,i
|
(L
2
)
m+1 +
C

2
i
|g
k,i
|
(H
1
)
m+1 +

2
i
|g
t
k,i
|
(L
2
)
m+1 (7.75)
Note that in the second line of (7.75) we used integration by parts and the Holder inequality,
and in the third line we used the fact that there is a 0 < C < such that |g
k,i
|
(L

)
m+1
C|g
k,i
|
(H
1
)
m+1. Hence, (7.71), (7.72), and (7.75) imply
s
k

k,i
0 in L
2
(0, L), for i = 1, 3, , n. (7.76)
138
Going back to (7.68), since s
k
u
k

k=1
is bounded in (L
2
(0, L))
m+1
, we must have that s
k
z
k

k=1
is bounded in (L
2
(0, L))
m+1
. Therefore, by (7.74), (7.76), and the fact that s
k
z
k
is bounded
in (L
2
(0, L))
m+1
we have that
s
k
C
k,i

k=1
is bounded in R for i = 1, 3, , n (7.77)
Next, if we dierentiate (7.74), we have
z
t
k
=
_

_
s
k

1
C
k,1
sin(s
k

1
x)
s
k

3
C
k,3
sin(s
k

3
x)
.
.
.
s
k

n
C
k,n
sin(s
k

n
x)
_

_
+
_

t
k,1
(x)

t
k,3
(x)
.
.
.

t
k,n
(x)
_

_
. (7.78)
For i = 1, 3, , n, notice that

t
k,i
(x) =
_
L
0
cos(s
k

i
(x y))(f
k,i
(y) is
k
g
k,i
(y)) dy.
By following an argument similar to (7.75), one can show
[
t
k,i
(x)[

L|f
k,i
|
(L
2
)
m+1 +

i
|g
t
k,i
|
(L
2
)
m+1 +
C

i
|g
k,i
|
(H
1
)
m+1.
Therefore, (7.71) and (7.72) imply that

t
k,i
0 in L
2
(0, L) for i = 1, 3, , n. (7.79)
Thus, (8), (7.78), and (7.79) imply the following:
s
k

1
C
k,1
sin(s
k

1
x) s
k

3
C
k,3
sin(s
k

3
x) 0 in L
2
(0, L)
s
k

3
C
k,3
sin(s
k

3
x) s
k

5
C
k,5
sin(s
k

5
x) 0 in L
2
(0, L)
.
.
.
s
k

n
C
k,n2
sin(s
k

n2
x) s
k

n
C
k,n
sin(s
k

n
x) 0 in L
2
(0, L). (7.80)
According condition (i), there is an i 1, 3, , n 2 such that
i
,=
i+2
We can explicitly
139
write the limit expression corresponding to i and i + 2 in (7.80) as follows:
|s
k

i
C
k,i
sin(s
k

i
x)|
2
L
2
+|s
k

i+2
C
k,i+2
sin(s
k

i+2
x)|
2
L
2
+
i

i+2
(s
k
C
k,i
) (s
k
C
k,i+2
)
_
L
0
cos(s
k
(
i
+
i+2
)x) dx
+
i

i+2
(s
k
C
k,i
) (s
k
C
k,i+2
)
_
L
0
cos(s
k
(
i

i+2
)x) dx 0. (7.81)
But according to (7.77), there is a constant M > 0 such that

i+2
(s
k
C
k,i
) (s
k
C
k,i+2
) M for all k.
Therefore,

i+2
(s
k
C
k,i
) (s
k
C
k,i+2
)
_
L
0
cos(s
k
(
i
+
i+2
)x) dx

M
_
L
0
[ cos(s
k
(
i
+
i+2
)x)[ dx
However, the Riemann-Lebesgue Lemma implies that
_
L
0
[ cos(s
k
(
i
+
i+2
)x)[ dx 0.
Therefore,

i+2
(s
k
C
k,i
) (s
k
C
k,i+2
)
_
L
0
cos(s
k
(
i
+
i+2
)x) dx 0. (7.82)
Similarly,

i+2
(s
k
C
k,i
) (s
k
C
k,i+2
)
_
L
0
cos(s
k
(
i

i+2
)x) dx

M
_
L
0
[ cos(s
k
(
i

i+2
)x)[ dx
Since we have
i
,=
i+2
, the Riemann-Lebesgue Lemma implies:
_
L
0
[ cos(s
k
(
i

i+2
)x)[ dx 0.
Therefore,

i+2
(s
k
C
k,i
) (s
k
C
k,i+2
)
_
L
0
cos(s
k
(
i

i+2
)x) dx 0. (7.83)
Thus (7.81), (7.82), and (7.83) imply that
s
k

i
C
k,i
sin(s
k

i
x) 0 in L
2
(0, L), and
s
k

i+2
C
k,i+2
sin(s
k

i+2
x) 0 in L
2
(0, L).
(7.84)
140
Using (7.84), we complete a chain of backwards substitutions in (7.80) (similar in fashion to
the procedure in Lemma 7.4) to deduce
s
k

j
C
k,j
sin(s
k

j
x) 0 in L
2
(0, L) for all j 1, 3, , n.
Then it follows from (7.78) and (7.79) that
z
t
k
0 in (L
2
(0, L))
m+1
. (7.85)
Since p
1
O
E
O
is an invertible matrix, (7.85) implies
u
t
k
0 in (L
2
(0, L))
m+1
. (7.86)
Since (7.68) implies that u
k
0 in (L
2
(0, L))
m+1
, we nally conclude that (9) holds.
Proof of (10): Take the (L
2
)
m+1
-inner product of (2) with u
k
(multiply on the left) and
obtain
is
k
v
k
, u
k
)

+p
1
O
E
O
u
t
k
, u
t
k
)

0, (7.87)
where in the last term on the left side we used u
t
k
= 0 for x = 0, L. By (7.86), the second term
in (7.87) goes to 0. Thus,
is
k
v
k
, u
k
)

0. (7.88)
Next, go back to (7.45), and take the (L
2
)
2
inner product of this with v
k
to obtain the following:
is
n
v
k
, u
k
)

|v
k
|
2
(L
2
)
m+1
0. (7.89)
But the rst term in (7.89) goes to 0 according to (7.88), hence (10) follows.
Finally, if we combine the results of (6), (7), (9), and (10), we conclude that
Y
k
0 in c. (7.90)
But this contradicts our assumption that |Y
k
|
c
= 1 for all k. Therefore, (7.43) is false, and /
must be the generator of an exponentially stable semigroup provided that conditions (i) and
(ii) hold.
141
The forward implication can be proven by contrapositive. If S is one of the sets in Theorem
7.2 (which is the negation of conditions (i) and (ii) both holding true), the spectral bound of
/ is 0. According to semigroup theory (see [39] and [36]), the growth rate of e
,t
is greater
than or equal to the spectral bound of /; hence
|e
,t
| Me
t
for all M > 1,
but 0. Thus, /is not exponentially stable if either condition (i) or (ii) fails. This completes
the proof.
142
CHAPTER 8. Conclusion
In this thesis, we focused primarily on the following 3 dierent structures:
1. The multilayer Mead-Markus beam
2. The multilayer Mead-Markus plate
3. The multilayer Rao-Nakra beam
In Chapter 5, we proved that the semigroup associated with the multilayer Mead-Markus beam
is analytic and exponentially stable. The method of proof was direct and involved the use of
Riesz bases. Moreover, two optimal damping problems were considered:
1. Choosing the damping parameters (viscosity) for the materials in the compliant layers
to achieve the optimal angle of analyticity for the associated semigroup.
2. Choosing the damping parameters in the compliant layers to achieve the optimal growth
rate for exponential stability of the associated semigroup.
The rst optimal damping problem is completely solved for the multilayer beam, but the sec-
ond problem is solved only in the three-layer case. When comparing the optimal damping
parameters for angle of analyticity to those for the optimal growth rate in the three-layer case,
one nds that they are in close agreement. In Chapter 6, the analyticity of the multilayer
Mead-Markus plate is proved via a direct argument (very similar to the method used in [17]).
However, it is assumed that the Poisson ratio in the odd layers are all the same. In addition,
the only boundary conditions that are considered are the clamped ones. As will be shown in
Appendix A, the result holds true for the hinged boundary conditions as well via the indirect
method. In Chapter 7, the exponential stability of the semigroup associated with a Rao-Nakra
143
beam is proved via the indirect method, provided that the wave speeds satised certain condi-
tions. This time, stress free boundary conditions with respect to the longitudinal displacements
are considered. One technical diculty is that stress free boundary conditions admit inertial
sliding solutions. These solutions are eigenvectors of the semigroup generator corresponding
to the zero eigenvalue, which violates exponential stability. This diculty is overcome by using
a quotient space formulation that equates solutions up to these initial sliding states.
8.1 Importance of the results
The results in Chapter 5 can be very useful for industrial purposes. As mentioned in the
introduction, many multilayer beams and plates can be accurately modeled eectively using
Mead-Markus theory. In particular, a manufacturer can design a beam with the optimal
damping characteristics desired by simply making the materials in the compliant layers the
appropriate viscosity. One can compute these viscosities by way of Theorems 5.5 or 5.9.
According to the theory of semigroups, there are some benets of showing analyticity. One of
which is the regularizing property of analytic semigroups: for t > 0, solutions to x = /x are in
T(/
k
) for any k. Often this implies that solutions are C

in the space variables. Additional

regularity can be exploited in application to coupled problems (see e.g., [28], [34], and [4])
and nonlinear problems (see e.g., [29] and [30]). Another benet is that analytic semigroups
satisfy the spectrum-determined growth condition according to Theorem 2.5. This allows us
to determine the growth rate of semigroups associated with beam and plate models by looking
at the spectrum. Furthermore, the spectrum of these semigroup generators typically consist
only of eigenvalues; therefore a great deal of information can be gleaned about the exponential
stability of a semigroup just by looking at the eigenvalues. Finally, analyticity and exponential
stability are useful for proving controllability results for the beam and plate models (see e.g.,
[20] and [19]).
8.2 Open problems
There are a number of open questions regarding the stability of plates and beams.
144
1. A closed formula for recovering the optimal damping coecients needed to achieve the
optimal decay rate in a multilayer (m > 1) Mead-Markus beam would be very useful.
The author believes that an eigenvalue condition somewhat similar to the one in (5.59)
must hold, but the calculations have not been promising so far.
2. A direct proof for the analyticity of a Mead-Markus plate using hinged boundary condi-
tions has not been found yet. It is likely that it can be proved using some clever identities
and/or inequalities involving the form (, ) dened in Chapter 4.
3. A proof of the exponential stability (either direct or indirect) of the Rao-Nakra beam
with clamped boundary conditions has not been done yet. However, the indirect proof
of this should be easier than the one in Chapter 7 because we avoid the possibility of
having inertial sliding solutions with the clamped boundary conditions.
4. The exponential stability of a multilayer Rao-Nakra plate has not been proven, but the
author suspects that it is true with both hinged and clamped boundary conditions.
5. A direct proof of the exponential stability of a multilayer Mead-Markus beam with either
clamped or hinged boundary conditions. The author has tried to mimic the proof of
exponential stability for a laminated beam done in Hansen and Lasiecka [17] and apply
it to the multilayer Mead-Markus beam case, but has done so with little success so far.
145
APPENDIX A. Analyticity of a Multilayer Mead-Markus Plate: An
Indirect Proof
The multilayer Mead-Markus plate consists of m+1 Kirchho plate layers bound together
by m shear-deformable layers. We consider the case in which the Poisson ratios are the same in
each Kirchho plate layer and linear viscous shear damping is included in the shear deformable
layers. We show, via a contradiction argument, that the associated semigroup is analytic.
Introduction
The classical (three layer) sandwich beam of Mead and Markus [37] models a composite
material consisting of two relatively sti outer layers and a much more compliant interior layer.
The outer layers are modeled under Euler-Bernoulli assumptions and hence do not allow shear.
The inner layer is elastic with respect to shear, but is assumed to be much less rigid than the
face layers, hence the bending stiness is considered to be negligible. Some form of linear shear
damping is usually assumed in the interior layer. In the case of linear viscous shear damping,
the Mead-Markus model can be written:
_

_
mw
tt
+ (A+
B
2
C
)w
xxxx

B
C
s
xxx
= 0, 0 < x < L; t > 0
s
t
+s
1
C
s
xx
+
B
C
w
xxx
= 0, 0 < x < L; t > 0
, (A.1)
where A, B, C, m, , are positive physical constants, w represents the transverse displace-
ment and s is proportional to the shear of the middle layer. (See [9] for a detailed explanation.)
The semigroup exp(t/) associated with the (three layer) Mead-Markus beam (A.1) was
shown to be analytic in Hansen and Lasiecka, [17] in the cases of either clamped or hinged
boundary conditions. (Actually [17] considers a special case of (A.1), but the proof applies
equally well to (A.1).) The proof relies on showing the dissipativity of the generator / and
146
establishing the following estimate for the resolvent operator R(, /) = (I /)
1
:
|R(is, /)| C/[s[ for all real s : [s[ 1. (A.2)
A similar analyticity result, but for a symmetric multilayer beam system was proved in
Hansen and Liu [18], where (A.2) was proved by the same idea, but with an indirect argument.
In Hansen, [16], a multilayer plate analog of the Mead-Markus model is derived in which
m + 1 sti plate layers modeled under Kirchho plate assumptions are bonded together
by m compliant plate layers. The compliant layers are elastic in shear (i.e., include linear
elastic constants for shear modulus and shear damping modulus) but the bending stiness is
considered negligible in comparison to those of the surrounding sti layers. The resulting
model (A.5), (A.6) is described by a scalar transverse displacement w, the shear of the even
layers,
E
= (
i
j
), i = 2, 4, . . . 2m, j = 1, 2 and the in-plane displacements of the odd layers
v
O
= (v
i
j
), i = 1, 3, . . . 2m + 1, j = 1, 2. In the special case in which the Poisson ratios of the
odd layers are all the same, it is possible to eliminate v
O
and obtain a system (A.10), (A.6)
that resembles Mead-Markus beam system (A.1) in structure. It was shown in [15] that system
(A.10), (A.6) is governed by a C
0
contraction semigroup.
In this paper we prove the analyticity of the semigroup associated with the multilayer
Mead-Markus plate model (A.10), (A.6) of [15]. We consider the cases of hinged or clamped
boundary conditions. Our proof relies on establishing the estimate (A.2), which we accomplish
by a contradiction argument similar to the one in [18].
Analyticity results for beam and plate systems are often of critical importance in appli-
cation to control and stability problems. For application of the Riccati theory to analytic
semigroups, see [32]; for coupled systems involving analytic semigroups, see e.g., [28], [34], [4].
For stability of nonlinear systems having an analytic linearization, see e.g., [29], [30]. For local
null controllability problems associated with analytic semigroups, see e.g., [3].
We remark that analyticity has been found in a number of other beam and plate systems
based on the Euler-Bernoulli model w
tt
+
2
w = 0. Chen and Russell [5] proposed the
square-root damping model w
tt
+ Bw
t
+ Aw = 0, where B is proportional to A
1/2
, and
showed (under appropriate assumptions for A) that the spectrum is frequency-proportional,
147
i.e, asymptotically, the eigenvalues have proportional real and imaginary parts. Chen and
Triggiani [6] showed, essentially, that when B is proportional to A

, for 1/2 1, the

associated semigroup is analytic, but not for < 1/2. Similar to the Euler-Bernoulli beam
with square-root damping, the thermoelastic Euler-Bernoulli beam: w
t
+ w
xxxx
+ w = 0,

xx
w
xxt
= 0, > 0 was shown [22] to have frequency-proportional spectrum and to be
associated with an analytic semigroup. Analogous analyticity results for thermoelastic plate
model were later shown to hold; [33], [31].
There is a close similarity between the Mead-Markus model described here (A.10), (A.6)
and thermoelastic systems. In fact, Triggiani [46] showed that (A.1) is equivalent (modulo
lower-order terms) to the thermoelastic Euler-Bernoulli beam through a change of variables.
This does not remain true for the plate model in this paper since the dissipation results from
a damped Lame system (not a heat equation.)
Equations of Motion:
Let be a smooth bounded domain in the plane with boundary and let x = x
1
, x
2

denote the points in . If and are matrices in R

mn
, then we denote : as the scalar
product in R
mn
. Then we denote
, )

=
_

: dx, , )

=
_

: d
Dene the form for functions (x) =
1
(x),
2
(x) by

(;

) =
_

1
x
1
,

1
x
1
_

+
_

2
x
2
,

2
x
2
_

+
_

2
x
2
,

1
x
1
_

+
_

1
x
1
,

2
x
2
_

+
_
_
1
2
__

1
x
2
+

2
x
1
_
,
_

1
x
2
+

2
x
1
__

,
where is the Poisson ratio of each layer (0 < < 1/2). We dene the Lame operator
L

= L

1
(
1
,
2
), L

2
(
1
,
2
) as follows:
L

1
(
1
,
2
) =

x
1
_

1
x
1
+

2
x
2
_
+

x
2
__
1
2
__

1
x
2
+

2
x
1
__
L

2
(
1
,
2
) =

x
2
_

2
x
2
+

1
x
1
_
+

x
1
__
1
2
__

2
x
1
+

1
x
2
__
.
148
Also dene the boundary operator B

= B

1
(
1
,
2
), B

2
(
1
,
2
) as follows:
B

1
(
1
,
2
) =
_

1
x
1
+

2
x
2
_
n
1
+
_
1
2
__

1
x
2
+

2
x
1
_
n
2
B

2
(
1
,
2
) =
_

2
x
2
+

1
x
1
_
n
2
+
_
1
2
__

2
x
1
+

1
x
2
_
n
1
,
where n = (n
1
, n
2
) denotes the outward unit normal to . With these denitions, the following
Greens formula holds for suciently smooth scalar functions and

:

(,

) = B

,

)

,

)

. (A.3)
Let
i
, i = 1, 3, . . . 2m+ 1 denote the Poisson ratios of the odd layers. Dene the operator
L
O
by (L
O

O
)
ij
= (L

i
)
j
, where
i
= (
i
1
,
i
2
) is the i
th
row of
O
= (
i
j
), i = 1, 3, . . . , 2m+1,
j = 1, 2. Associated with L
O
is the boundary operator B
O
for which the following Greens
formula is valid:

i=1,3,...2m+1

i
(
i
,

i
) = B
O

O
,

O
)

L
O

O
,

O
)

(A.4)
for all suciently smooth
O
,

O
. Alternatively, L
O
= diag (L

1
, L

3
, , L

2m+1
), and B
O
=
diag (B

1
, B

3
, , B

2m+1
).
The multilayer plate version of the Mead-Markus model derived in Hansen [16] takes the
form: _

_
mw
tt
+K
2
w div

N
T
h
E
_
G
E

E
+

G
E

E
_
= 0 in R
+
12h
O
D
O
L
O
v
O
+B
T
_
G
E

E
+

G
E

t
E
_
= 0 in R
+
, (A.5)
where
h
E

E
= Bv
O
+h
E

Nw. (A.6)
In the above, m, K are positive constants, matrices h
E
, h
O
, G
E
,

G
E
, and D
O
are diagonal with
positive diagonal elements. The E subscripts indicate quantities with m rows (corresponding
to even-indexed rows) and the O subscripts indicate quantities with m+1 rows (corresponding
to odd-indexed rows). In addition,

N is a column vector of constants greater that 1; B = (b
ij
)
is the m(m+ 1) matrix dened by
b
ij
=
_

_
(1)
i+j+1
if j = i or j = i + 1
0 otherwise.
149
As described in the introduction, w,
E
and v
O
represent respectively, the transverse displace-
ment, the shear of the even layers, and in-plane displacements of the odd layers.
Many boundary conditions could be considered. One form of hinged boundary conditions
is obtained under the assumption that the lateral stress and bending moment vanish with zero
transverse displacement imposed along :
Bw n = w = 0, B
O
v
O
= 0 on (A.7)
We can also consider clamped boundary conditions:
w = 0,
w
n
= 0,
E
= 0 on . (A.8)
In the remainder of this paper we will assume the Poisson ratios of each layer are the same.
Hence we dene
:=
1
=
3
= =
2m+1
.
In this case it is easy to see that
BL
O
v
O
= L
E
Bv
O
,
where L
E
= diag (L

, L

, . . . L

) = I
m
L

, where I
m
is the identity on R
m
. Also analogously
dene boundary operator B
E
= I
m
B

and quadratic form

E
(
E
,

E
) = B
E

E
,

E
)

L
E

E
,

E
)

. (A.9)
Dene the matrix P by
P =
1
12
BD
1
O
h
1
O
B
T
.
It is shown in [15] that P is a positive denite, symmetric, and invertible M-matrix. If we
multiply the last line of (A.5) by
1
12
BD
1
O
h
1
O
, we obtain the following:
_

_
mw
tt
+K
2
w div

N
T
_
h
E
P
1
L
E
Bv
O

= 0 in R
+
G
E

E
+

G
E

t
E
P
1
L
E
Bv
O
= 0 in R
+
.
(A.10)
Hinged boundary conditions (A.7) take the form:
Bw n = w = 0, B
E
Bv
O
= 0 in (A.11)
150
while clamped boundary conditions (A.8) can be written
w = 0, w = 0,
E
= 0 in , (A.12)
since w = 0 and
w
n
= 0 on .
Note that the above formulation is completely independent of v
O
since the term Bv
O
is easily
eliminated using (A.6).
Semigroup Formulation and Well-Posedness:
Dene the following spaces for s > 0:
H
s
O
() := v
O
= (v
i
j
), i = 1, 3, 5, . . . 2m+ 1, j = 1, 2 : v
i
j
H
s
()
H
s
E
() :=
E
= (
i
j
), i = 2, 4, . . . 2m, j = 1, 2 :
i
j
H
s
().
In terms of the state variables w, y = w, and
E
, we can rewrite the previous system as:
d
dt
_

_
w
y

E
_

_
= /
_

_
w
y

E
_

_
:=
_

_
y
1
m
_
K
2
w + div

N
T
_
h
E
P
1
L
E
Bv
O

G
1
E
_
G
E

E
+P
1
L
E
Bv
O
_
_

_
. (A.13)
The energy inner product associated with z = [w, y,
E
], z = [ w, y,

E
] is
z, z)
c
=
1
2
_
my, y)

+K
2
w, w)

+G
E
h
E

E
,

E
)

P
1
L
E
Bv
O
, B v
O
_

_
.
Let c be [w, y,
E
]
T
H
2
() L
2
() H
1
E
(). We let H denote the closure in c of all
elements in c that satisfy the boundary conditions. The domain of / is as follows:
T(/) = [w, y,
E
]
T
H : w H
4
(), y H
2
(),
E
H
3
E
(), + BCs,
where +BCs refers to either the hinged boundary conditions (A.11) or the clamped bound-
ary conditions (A.12) along with boundary conditions induced by the image of /[w, y,
E
]
T
satisfying boundary conditions in H; that is,
y = 0 on , (A.14)
for hinged boundary conditions, and
y = 0, y = 0, G
E

E
+P
1
L
E
Bv
O
= 0 on , (A.15)
for clamped boundary conditions.
151
Semigroup Well-Posedness
Throughout the rest of the discussion, it will be convenient to use the following notation:
[w[
s
:= |w|
H
s
()
, [w[
0
:= |w|
L
2
()
, [
E
[
s,E
:= |
E
|
H
s
E
()
[v
O
[
s,O
:= |v
O
|
H
s
O
()
, [z[
c
:=
_
z, z)
c
, where z = [w, y,
E
]
T
.
The following theorem is proved in Hansen [15]:
Theorem A.1. The operator / in (A.13) is the generator of a strongly continuous semigroup
of contractions on H. Furthermore the following dissipativity condition holds: For all z
T(/),
Re z, /z)
c
=

G
1/2
E
h
1/2
E
(P
1
L
E
Bv
O
G
E

E
)

2
0,E
. (A.16)
Analyticity of Semigroup
The main result of the this paper is that the semigroup generated by / is in fact analytic.
We begin with a lemma.
Lemma A.1. The operator / has no nonzero eigenvalues on the imaginary axis.
Proof: Suppose i is an eigenvalue of / with eigenvector z = [w
0
, y
0
,
0
E
]
T
, where R0.
Then (A.16) implies that
0 = Re z, (iI /)z)
c
=

G
1/2
E
h
1/2
E
(P
1
L
E
Bv
O
G
E

E
)

2
0,E
. (A.17)
Therefore the solution v = [w, y,
E
]
T
to v
t
= /v, v(0) = z satises

t
E
= 0.
Since ,= 0, it immediately follows that

E
= 0,
0
E
= 0. (A.18)
152
Equations (A.17) and (A.18) imply
L
E
Bv
O
= 0
Using the Greens formula (A.4) and the boundary conditions, it follows that

((Bv
O
)
i
, (Bv
O
)
i
) = 0,
where (Bv
O
)
i
is the ith row of Bv
O
. Then Korns inequality applied to

(see e.g., [26])

implies for the case of hinged boundary conditions that
Bv
O
= constant (A.19)
For clamped boundary conditions, we obtain Bv
O
= 0. In either case, (A.6), (A.18), and
(A.19) imply
w = constant.
Thus w is linear. However the boundary condition w = 0 on implies w = 0 in . Finally, it
easily follows from the eigenvector equation that y = 0, and hence z = 0.
Theorem A.2. The semigroup generated by / is analytic.
Proof: First note, since T(/) is compactly embedded in H, the spectrum of / consists of
eigenvalues. Since by Lemma A.1 there are no eigenvalues on the nonzero imaginary axis, and
by Theorem A.1 the semigroup is dissipative, the closed right half-plane is in the resolvent set
of /, with the only possible exception being an eigenvalue at the origin.
By Theorem 2.4, a sucient condition for analyticity is that there exist M,
0
for which
the resolvent operator R(, /) satises
|R( +is, /)|
M
[s[
s ,= 0, >
0
. (A.20)
We rst prove that (A.2) implies (A.20). Indeed, with = +is the resolvent identity
R(, /) = R(is, /) + (is )R(, /)R(is, /)
together with (A.2) and the bound |R(, /)| 1/[[ (see Pazy [39], p. 11) gives
|R( +is, /)|
C
[s[
+[[
1
[[

C
[s[
=
2C
[s[
, [s[ 1.
153
For [s[ < 1 once
0
> 0 is picked, |R(
0
+ is, /)| is uniformly bounded for [s[ < 1, hence
(A.20) follows for any
0
> 0. Thus analyticity follows if we prove the resolvent estimate (A.2).
Equivalently, we need to show that there is a > 0 such that
inf
[z[
E
=1,zT(,)
[/z isz[
c
> [is[.
Suppose on the contrary that there exists s
n

n=1
in R and z
n

n=1
in T(/) with [z
n
[
c
= 1
such that
lim
n
[iz
n
s
1
n
/z
n
[
c
= 0. (A.21)
We can take s
n
> 0 because the norm of the resolvent is symmetric with respect to the real
axis. Furthermore, since the imaginary axis is in the resolvent set, (A.21) is violated if s
n

remains in any compact set. Hence without loss of generality we assume s

n
.
For convenience, we will drop the subscript O from v
O
and the subscript E from
E
for
the rest of this proof. With this notation, we have that
Bv
n
= h
E
(
n

Nw
n
) (see (A.6)). (A.22)
Our goal is to show that z
n
0 in H, in contradiction to the assumption that [z
n
[
c
= 1. First,
we write out (A.21) explicitly as follows:
W
n
:= iw
n
s
1
n
y
n
0 in H
2
() (A.23)
Y
n
:= iy
n
s
1
n
m
1
_
K
2
w + div

N
T
_
h
E
P
1
L
E
Bv
n

_
0 in L
2
() (A.24)

n
:= i
n
s
1
n

G
1
E
_
G
E

n
+P
1
L
E
Bv
n
_
0 in H
1
E
() (A.25)
We proceed by showing the following steps in the indicated order:
1.
n
0 in L
2
E
().
2. s
1/2
n
L
E
Bv
n
0 in L
2
E
().
3.
_
s
1/2
n
y
n
_

n=1
is a bounded sequence in L
2
().
4. Bv
n

n=1
is a bounded sequence in H
1
E
().
154
5. Bv
n
0 in H
1
E
().
6. im[y
n
[
2
0
+Ks
1
n
(w
n
, y
n
) 0.
7. y
n
0 in L
2
().
8. w
n
0 in H
2
().
9.
n
0 in H
1
E
().
Proof of (1) and (2): Assuming that (A.21) holds, we have that
Re z
n
, iz
n
s
1
n
/z
n
) 0.
But the dissipativity relation (A.16) implies
s
1
n

G
1/2
E
h
1/2
E
(P
1
L
E
Bv
n
G
E

n
)

2
0,E
0. (A.26)
Thus, (A.26) reads
s
1/2
n

G
1/2
E
h
1/2
E
(P
1
L
E
Bv
n
G
E

n
) 0 in L
2
E
(),
which certainly implies
s
1
n

G
1
E
(P
1
L
E
Bv
n
G
E

n
) 0 in L
2
E
(). (A.27)
By (A.25),
i
n
s
1
n

G
1
E
_
G
E

n
+P
1
L
E
Bv
n
_
0 in L
2
E
(). (A.28)
By combining (A.27), and (A.28), we obtain (1). Then substitution of (1) into (A.26) leads to
s
1
n

G
1/2
E
h
1/2
E
(P
1
L
E
Bv
n
)

2
0,E
0.
Since

G
E
, h
E
, and P are all invertible matrices, (2) follows.
Proof of (3) and (4): Since [w
n
[
2
[z
n
[
c
= 1, by (A.23) the sequence
_
s
1
n
y
n
_

n=1
is
bounded in H
2
(). But also, [y
n
[
0
[z
n
[
c
= 1. Using interpolation, we conclude that
_
s
1/2
n
y
n
_

n=1
is bounded in H
1
().
155
From this, (3) immediately follows. To show (4), we note that
[
n

Nw
n
[
1,E
C ([
n
[
1,E
+[w
n
[
2
) for some constant C. (A.29)
But, [
n
[
1,E
[z
n
[
c
= 1 and [w
n
[
2
[z
n
[
c
= 1. Hence, (A.22) and (A.29) imply (4).
Proof of (5): Equation (A.23) implies
w
n
s
1
n
y
n
0 in L
2
() (A.30)
By (3), s
1
n
y
n
0 in L
2
(). This, along with (A.30), implies that
w
n
0 in L
2
(). (A.31)
Therefore, (A.22), (1), and (A.31) imply that
Bv
n
0 in L
2
E
(). (A.32)
Next note that according to (A.23) and (A.25), we have that
h
E
_

NW
n
_
0 in H
1
E
(). (A.33)
Since the form
E
is continuous on H
1
E
(), (4) and (A.33) imply

E
_
Bv
n
, h
E
(
n

NW
n
)
_
0. (A.34)
By (A.23) and (A.25), we can rewrite (A.34) as follows:
i
E
(Bv
n
, Bv
n
) s
1
n

E
_
Bv
n
, h
E

G
1
E
_
G
E

n
+P
1
L
E
Bv
n
_
h
E

Ny
n
_
0. (A.35)
Next, we will look at the second term in (A.35) and show that it goes to 0 as n . For the
hinged boundary conditions, we have B
E
Bv
O
= 0 on by (A.11). For the clamped boundary
conditions, we have G
E

n
+ P
1
L
E
Bv
n
= 0 on , and y
n
= 0 on by (A.15). In either
case, the Greens formula for
E
in (A.9) implies
s
1
n

E
_
Bv
n
, h
E

G
1
E
_
G
E

n
+P
1
L
E
Bv
n
_
h
E

Ny
n
_
= s
1
n
_
L
E
Bv
n
, h
E

G
1
E
G
E

n
_

s
1
n

L
E
Bv
n
, P
1
L
E
Bv
n
_

+s
1
n
_
L
E
Bv
n
, h
E

Ny
n
_

. (A.36)
156
The rst term on the right side of (A.36) goes to 0 by (1) and (2), the second term goes to 0
by (2), and the third term goes to 0 by (2) and (3). Therefore,

E
(Bv
n
, Bv
n
) 0. (A.37)
By Korns inequality (Corollary 4.1) for any > 0 there exists C > 0 such that

E
(Bv
n
, Bv
n
) +[Bv
n
[
2
0,E
C[Bv
n
[
2
1,E
.
Therefore, (5) follows from (A.32) and (A.37).
Proof of (6): Our next step is to take the inner product of (A.24) with my
n
. Since y
n
is
bounded in L
2
(), we have that
im[y
n
[
2
0
+Ks
1
n

2
w
n
, y
n
_

s
1
n
_
div

N
T
h
E
P
1
L
E
Bv
n
, y
n
_

0. (A.38)
Lets examine the second term in (A.38). Using the Divergence Theorem and the fact that

2
w
n
= div Lw
n
, we have
Ks
1
n

2
w
n
, y
n
_

= Ks
1
n
(Lw
n
) n, y
n
)

Ks
1
n
Lw
n
, y
n
)

. (A.39)
Since y
n
= 0 on , (A.39) implies that
Ks
1
n

2
w
n
, y
n
_

= Ks
1
n
Lw
n
, y
n
)

. (A.40)
But according to the Greens formula in (A.3),
Ks
1
n
Lw
n
, y
n
)

= Ks
1
n
Bw
n
, y
n
)

Ks
1
n
(w
n
, y
n
) (A.41)
For the hinged boundary conditions, y
n
= 0 on implies that y
n
=
n
(y
n
)n, where n is
the outward unit normal to and
n
(y
n
) denotes the normal derivative. Then we can apply
the condition Bw n = 0 on to conclude Bw
n
, y
n
)

= 0. For the clamped boundary

conditions, we have y
n
= 0 on by (A.14) and (A.15). Thus using either of the boundary
conditions, (A.41) becomes
Ks
1
n
Lw
n
, y
n
)

= Ks
1
n
(w
n
, y
n
) . (A.42)
157
Putting the results of (A.38), (A.40) and (A.42) together, we have
im[y
n
[
2
0
+Ks
1
n
(w
n
, y
n
) s
1
n
_
div

N
T
h
E
P
1
L
E
Bv
n
, y
n
_

0. (A.43)
Next, consider the third term in (A.43). The Divergence Theorem and the fact that y
n
= 0 on
implies
s
1
n
_
div

N
T
h
E
P
1
L
E
Bv
n
, y
n
_

= s
1
n
_

N
T
h
E
P
1
L
E
Bv
n
, y
n
_

. (A.44)
The right side of (A.44) goes to 0 by (2) and (3), and thus (A.44) implies
s
1
n
_
div

N
T
h
E
P
1
L
E
Bv
n
, y
n
_

0. (A.45)
Thus (6) follows from (A.43) and (A.45).
Proof of (7), (8), and (9): Since w
n

n=1
is a bounded sequence in H
2
(), w
n

n=1
is
a bounded sequence in H
1
(). In addition, W
n
0 in H
1
() by (A.23). Thus,
K(w
n
, W
n
) 0 iK(w
n
, w
n
) Ks
1
n
(w
n
, y
n
) 0. (A.46)
Now add the results of (6) and (A.46) together and obtain the following:
im[y
n
[
2
0
+iK(w
n
, w
n
) 0. (A.47)
Since is nonnegative, it follows from (A.47) that (7) is true, and also
(w
n
, w
n
) 0 (A.48)
Since w
n
0 in L
2
() by (A.31), Korns inequality applied to (A.48) implies that
w
n
0 in H
1
(). (A.49)
But w
n
= 0 on , and so Poincares Inequality with (A.49) implies (8). Then (A.22), (5), and
(A.49) imply (9). Finally, (7), (8), and (9) imply that z
n
0 in c, which contradicts that
|z
n
|
c
= 1. Therefore, (A.21) cannot be true, and / must be the generator of an analytic
semigroup.
158
APPENDIX B. Proofs of various results in the thesis
In this appendix, we will give the proofs of a variety of useful results and formulas in this
thesis.
Properties of the matrix P
Recall that the matrix P is dened in Chapter 5 as
P := B(h
O
E
O
)
1
B
T
,
and in Chapter 6 it is dened
P :=
1
12
B(h
O
D
O
)
1
B
T
.
With either denition, we have the following.
Theorem B.1. P has the following properties:
1. P is an M-matrix.
2. P is symmetric.
3. P is positive denite.
4. P
1
consists entirely of positive entries.
Denition B.1. A nonsingular matrix P R
mm
is said to be an M-matrix if all entries of
P
1
are nonnegative, and all o-diagonal entries of P are non-positive.
Proof. Recall that the matrices h
O
, E
O
, and D
O
are diagonal matrices with positive entries
along the diagonal. Thus the matrices (h
O
E
O
)
1
and
1
12
(h
O
D
O
)
1
are of the form
159
diag (a
1
, a
2
, , a
m+1
), where a
i
> 0 for i = 1, 2, , m+ 1.
Thus, we have from the denition of B that
P =
_

_
1 1
1 1
.
.
.
.
.
.
1 1
_

_
_

_
a
1
a
2
.
.
.
a
m+1
_

_
_

_
1
1 1
1
.
.
.
.
.
. 1
1
_

_
=
_

_
a
1
+a
2
a
2
a
2
a
2
+a
3
a
3
a
3
.
.
.
.
.
.
.
.
.
.
.
. a
m
a
m
a
m
+a
m+1
_

_
. (B.1)
The directed graph of this matrix has the structure shown in Figure C.1.
Figure B.1 Directed graph of P.
It is clear from the directed graph that for any two distinct points, there exists a directed
path connecting them. Thus the directed graph is strongly connected, and P is an irreducible
matrix (see Theorem 15.1.1 in Lancaster and Tismenetsky [27]). Moreover, the following is
clear from (B.1):
p
ii
=
m

j=1,j,=i
[p
i,j
[ for i = 2, 3, , m1, and p
ii
>
m

j=1,j,=i
[p
i,j
[ for i = 1 and m.
160
Therefore, by Theorem 15.2.3 in [27], P is an M-matrix. Furthermore, it is clear that P is
symmetric from (B.1). This proves the rst two assertions. Finally, the last two conclusions
are proved in [15].
Greens formulas
In Chapter 6 and Appendix A, we analyzed the multilayer plate models of Mead-Markus.
Let n = (n
1
, n
2
) denote the outward unit normal to , and
i
denote the Poisson Ratio in
the i
th
layer. In Chapter 4, we dened the quadratic forms

i
,
E
, and
O
(see (4.20) and
(4.22)), and corresponding Lame and boundary operators L

i
, L
E
, L
O
, B

i
, B
E
, and B
O
. The
following theorem proves the Greens formula (4.21).
Theorem B.2. For suciently smooth functions = (x
1
, x
2
) =
1
(x
1
, x
2
),
2
(x
1
, x
2
) and
= (x
1
, x
2
) =
1
(x
1
, x
2
),
2
(x
1
, x
2
), we have

i
(, ) = B

i
, )

i
, )

.
Proof. Using the denition of in (4.20), we have

i
(; ) =
_

1
x
1

1
x
1
dx +
_

2
x
2

2
x
2
dx +
_

2
x
2

1
x
1
dx +
_

1
x
1

2
x
2
dx
+
_

_
1
i
2
__

1
x
2
+

2
x
1
__

1
x
2
+

2
x
1
_
dx
=
_

1
x
1

1
x
1
dx +
_

2
x
2

2
x
2
dx +
i
_

2
x
2

1
x
1
dx +
i
_

1
x
1

2
x
2
dx
+
1
i
2
_

1
x
2

1
x
2
dx +
1
i
2
_

1
x
2

2
x
1
dx
+
1
i
2
_

2
x
1

1
x
2
dx +
1
i
2
_

2
x
1

2
x
1
dx
161
Applying integration by parts, we obtain the following:

i
(; ) =
_

1
x
1

1
n
1
dS
_

1
x
2
1

1
dx +
_

2
x
2

2
n
2
dS
_

2
x
2
2

2
dx
+
i
_

2
x
2

1
n
1
dS
i
_

2
x
1
x
2

1
dx +
i
_

1
x
1

2
n
2
dS

i
_

1
x
2
x
1

2
dx +
1
i
2
_

1
x
2

1
n
2
dS
1
i
2
_

2
x
2

1
dx
+
1
i
2
_

1
x
2

2
n
1
dS
1
i
2
_

1
x
1
x
2

2
dx +
1
i
2
_

2
x
1

1
n
2
dS

1
i
2
_

2
x
2
x
1

1
dx +
1
i
2
_

2
x
1

2
n
1
dS
1
i
2
_

2
x
2
1

2
dx.
Gathering like terms, we obtain the following:

i
(; ) =
_

1
x
1
+
i

2
x
2
_

1
n
1
dS
_

1
x
2
1
+
i

2
x
1
x
2
_

1
dx
+
_

2
x
2
+
i

1
x
1
_

2
n
2
dS
_

2
x
2
2
+
i

1
x
2
x
1
_

2
dx
+
1
i
2
_

1
x
2
+

2
x
1
_

1
n
2
dS
1
i
2
_

2
x
2
+

2

2
x
2
x
1
_

1
dx
+
1
i
2
_

1
x
2
+

2
x
1
_

2
n
1
dS
1
i
2
_

_

2

1
x
1
x
2
+

2

2
x
2
1
_

2
dx
=
_

__

1
x
1
+
i

2
x
2
_
n
1
+
_
1
i
2
__

1
x
2
+

2
x
1
_
n
2
_

1
dS

__

1
x
2
1
+
i

2
x
1
x
2
_
+
_
1
i
2
__

2
x
2
+

2

2
x
2
x
1
__

1
dx
+
_

__

2
x
2
+
i

1
x
1
_
n
2
+
_
1
i
2
__

1
x
2
+

2
x
1
_
n
1
_

2
dS

__

2
x
2
2
+
i

1
x
2
x
1
_
+
_
1
i
2
__

2

1
x
1
x
2
+

2

2
x
2
1
__

2
dx.
Therefore,

i
(; ) =
_

i
1
(
1
,
2
)
1
dS
_

i
1
(
1
,
2
)
1
dx
+
_

i
2
(
1
,
2
)
2
dS
_

i
2
(
1
,
2
)
2
dx
=
_

i
1
(
1
,
2
), B

i
2
(
1
,
2
)
1
,
2
dS

i
1
(
1
,
2
), L

i
2
(
1
,
2
)
1
,
2
dS
= B

i
, )

i
, )

.
162
Corollary B.1. Suppose and are suciently smooth n2 matrices such that the i
th
row of
and are
i
=
i
1
,
i
2
and
i
=
i
1
,
i
2
respectively. Let
O
and
O
denote the (m+1) 2
matrices of the odd rows of and respectively, and
E
and
E
denote the m2 matrices of
the even rows of and respectively. Then the following Greens formulas hold:

O
(
O
,
O
) = B
O

O
,
O
)

L
O

O
,
O
)

E
(
E
,
E
) = B
E

E
,
E
)

L
E

E
,
E
)

.
Proof. The scalar product of B
O

O
and
O
on R
(m+1),2
is
B
O

O

O
=
_

_
B

1
1
(
1
1
,
1
2
) B

1
2
(
1
1
,
1
2
)
B

3
1
(
3
1
,
3
2
) B

3
2
(
3
1
,
3
2
)
.
.
.
.
.
.
B
n
1
(
n
1
,
n
2
) B
n
2
(
n
1
,
n
2
)
_

1
1

1
2

3
1

3
2
.
.
.
.
.
.

n
1

n
2
_

_
=

i=1,3, ,n
_
B

i
1
(
i
1
,
i
2
)
i
1
+B

i
2
(
i
1
,
i
2
)
i
2
_
.
Therefore
B
O

O
,
O
)

=
_

i=1,3, ,n
_
B

i
1
(
i
1
,
i
2
)
i
1
+B

i
2
(
i
1
,
i
2
)
i
2
_
d
=

i=1,3, ,n
_

i
1
(
i
1
,
i
2
), B

i
2
(
i
1
,
i
2
)
i
1
,
i
2
d
=

i=1,3, ,n
B

i
,
i
)

. (B.2)
Using a similar argument
L
O

O
,
O
)

i=1,3, ,n
L

i
,
i
)

. (B.3)
163
Therefore,

O
(
O
,
O
) =

i=1,3, ,n

i
(
i
,
i
)
=

i=1,3, ,n
_
B

i
,
i
)

i
,
i
)

_
by Theorem B.2
=

i=1,3, ,n
B

i
,
i
)

i=1,3, ,n
L

i
,
i
)

= B
O

O
,
O
)

L
O

O
,
O
)

by (B.2) and (B.3).

Hence (4.23) is proved. The proof of (6.6) is similar.
Identities involving the divergence and operators
Several lines in the proofs of the analyticity of the Mead-Markus plate in Chapter 6 and
Appendix A rely on some key identities.
Theorem B.3. For all suciently smooth functions = (x
1
, x
2
) =
1
(x
1
, x
2
),
2
(x
1
, x
2
),
div L

i
= div for all i = 1, 2, , n (B.4)
Proof. Using the denition of L

i
1
, we have the following:

x
1
L

i
1
(
1
,
2
) =

x
1
_

x
1
_

1
x
1
+
i

2
x
2
_
+

x
2
__
1
i
2
__

1
x
2
+

2
x
1
___
=

3

1
x
3
1
+
i

2
x
2
1
x
2
+
1
i
2

1
x
1
x
2
2
+
1
i
2

2
x
2
1
x
2
. (B.5)
Similarly

x
2
L

i
2
(
1
,
2
) =

3

2
x
3
2
+
i

1
x
1
x
2
2
+
1
i
2

2
x
2
1
x
2
+
1
i
2

1
x
1
x
2
2
. (B.6)
Add the results of (B.5) and (B.6) together and combine like terms to obtain the following for
all i = 1, 2, , n:
div L

i
=

3

1
x
3
1
+

3

2
x
3
2
+

3

2
x
2
1
x
2
+

3

1
x
1
x
2
2
. (B.7)
On the other hand, we have
div
1
,
2
=
_

1
x
1
+

2
x
2
_
=

3

1
x
3
1
+

3

1
x
1
x
2
2
+

3

2
x
3
2
+

3

2
x
2
1
x
2
(B.8)
164
Since the right hand sides of (B.7) and (B.8) agree, the proof is complete.
Theorem B.4. For all suciently smooth scalar functions w = w(x
1
, x
2
),
L

i
w = (w) for all i = 1, 2, , n (B.9)
Proof. For all i = 1, 2, , n,
L

i
1
w =

x
1
_

x
1
_
w
x
1
_
+
i

x
2
_
w
x
2
__
+

x
2
__
1
i
2
__

x
2
_
w
x
1
_
+

x
1
_
w
x
2
___
=

x
1
_

2
w
x
2
1
+
i

2
w
x
2
2
_
+

x
2
_
(1
i
)

2
w
x
1
x
2
_
=

3
w
x
3
1
+
i

3
w
x
1
x
2
2
+ (1
i
)

3
w
x
1
x
2
2
=

3
w
x
3
1
+

3
w
x
1
x
2
2
=

x
1
_

2
w
x
2
1
+

2
w
x
2
2
_
=

x
1
(w).
A similar calculation shows that
L

i
2
w =

x
2
(w) for all i = 1, 2, , n.
Thus (B.9) follows.
Corollary B.2. For all suciently smooth scalar functions w = w(x
1
, x
2
),
div L

i
w =
2
w for all i = 1, 2, , n.
Proof. Using (B.9), we have the following for all i = 1, 2, , n:
div L

i
w = div (w) = (w) =
2
w.
Remark B.1. Another way to prove the last result is to use (B.4). For all i = 1, 2, , n:
div L

i
w = (div w) = (w) =
2
w.
165
Useful identities when odd plate layers have identical Poissons ratio
In Chapter 6 and Appendix A, we examined the multilayer Mead-Markus plate and worked
under the assumption that the Poissons ratio in the odd layers is the same. So let
=
1
=
3
= =
n
.
In addition, recall that v
i
= v
i
(x
1
, x
2
) = v
i
1
(x
1
, x
2
), v
i
2
(x
1
, x
2
) is the longitudinal displace-
ment in the i
th
layer, and v
O
is the (m+1) 2 matrix in which the rows are the longitudinal
displacements in the odd layers.
Theorem B.5. (6.5) holds. That is,
BL
O
v
O
= L
E
Bv
O
.
Proof. First we note that the operators L

1
and L

2
are linear. For example:
L

1
(
1
+
1
,
2
+
2
)
=

x
1
_

x
1
(
1
+
1
) +

x
2
(
2
+
2
)
_
+

x
2
__
1
2
__

x
2
(
1
+
1
) +

x
1
(
2
+
2
)
__
=

x
1
_

1
x
1
+

2
x
2
_
+
_

1
x
1
+

2
x
2
__
+

x
2
_

__
1
2
__

1
x
2
+

2
x
1
__
+
__
1
2
__

1
x
2
+

2
x
1
___
=
_

x
1
_

1
x
1
+

2
x
2
_
+

x
2
__
1
2
__

1
x
2
+

2
x
1
___
+
_

x
2
_

2
x
2
+

1
x
1
_
+

x
1
__
1
2
__

2
x
1
+

1
x
2
___
Hence
L

1
(
1
+
1
,
2
+
2
) = L

1
(
1
,
2
) +L

1
(
1
,
2
). (B.10)
Similarly,
L

2
(
1
+
1
,
2
+
2
) = L

2
(
1
,
2
) +L

2
(
1
,
2
). (B.11)
166
From the denition of L
O
, L
O
v
O
is the (m+ 1) 2 matrix
L
O
v
O
=
_

_
L

1
(v
1
1
, v
1
2
) L

2
(v
1
1
, v
1
2
)
L

1
(v
3
1
, v
3
2
) L

2
(v
3
1
, v
3
2
)
.
.
.
.
.
.
L

1
(v
n
1
, v
n
2
) L

2
(v
n
1
, v
n
2
)
_

_
.
Hence,
BL
O
v
O
=
_

_
L

1
(v
1
1
, v
1
2
) L

1
(v
3
1
, v
3
2
) L

2
(v
1
1
, v
1
2
) L

2
(v
3
1
, v
3
2
)
L

1
(v
3
1
, v
3
2
) L

1
(v
5
1
, v
5
2
) L

2
(v
3
1
, v
3
2
) L

2
(v
5
1
, v
5
2
)
.
.
.
.
.
.
L

1
(v
(n2)
1
, v
(n2)
2
) L

1
(v
n
1
, v
n
2
) L

2
(v
(n2)
1
, v
(n2)
2
) L

2
(v
n
1
, v
1
2
)
_

_
=
_

_
L

1
((v
1
1
v
3
1
), (v
1
2
v
3
2
)) L

2
((v
1
1
v
3
1
), (v
1
2
v
3
2
))
L

1
((v
3
1
v
5
1
), (v
3
2
v
5
2
)) L

2
((v
3
1
v
5
1
), (v
3
2
v
5
2
))
.
.
.
.
.
.
L

1
((v
(n2)
1
v
n
1
), (v
(n2)
2
v
n
2
)) L

2
((v
(n2)
1
v
n
1
), (v
(n2)
2
v
n
2
))
_

_
, (B.12)
where in the last line we used (B.10) and (B.11). Note that the matrices on the right side of
(B.12) have only m rows. Thus (B.12) reads
BL
O
v
O
= L
E
_

_
v
1
1
v
3
1
v
1
2
v
3
2
v
3
1
v
5
1
v
3
2
v
5
2
.
.
.
.
.
.
v
(n2)
1
v
n
1
v
(n2)
2
v
n
2
_

_
= L
E
Bv
O
.
Remark B.2. If the Poisson ratios were dierent, then the matrix on the right side in the
rst line of (B.12) would not be equal to the one on the second line.
Next, we look at the energy of the multilayer Mead-Markus plate, namely
E(t) = m w, w)

+
O
(h
3
O
D
O

1
O
w,

1
O
w) + 12
O
(h
O
D
O
v
O
, v
O
) +G
E
h
E

E
,
E
)

.
167
In particular, we consider what happens to the term
O
(h
3
O
D
O

1
O
w,

1
O
w) when we assume
the Poisson ratios in the odd layers are the same. We shall prove the following result is valid
for both clamped and hinged boundary conditions.
Theorem B.6. For suciently smooth w, we have

O
(h
3
O
D
O

1
O
w,

1
O
w) = K
2
w, w)

. (B.13)
Proof. First note that

O
(h
3
O
D
O

1
O
w,

1
O
w) =

i=1,3, ,n

i
(h
3
i
D
i
w, w).
But since
i
= for all i = 1, 3, , n, we have

O
(h
3
O
D
O

1
O
w,

1
O
w) =

i=1,3, ,n

(h
3
i
D
i
w, w)
=

_
_

i=1,3, ,n
h
3
i
D
i
w, w
_
_
=

(Kw, w). (B.14)

Using the Greens formula (4.21) for

(Kw, w) = KB

w, w)

KL

w, w)

.
The clamped boundary conditions imply that w = 0 on , and the hinged boundary condi-
tions imply B

w n = 0 on . In either case, we have

(Kw, w) = KL

w, w)

. (B.15)
The divergence theorem implies
L

w, w)

= (L

w) n, w)

div L

w, w)

.
Since w = 0 on and div L

w =
2
w, we have
L

w, w)

=
2
w, w)

. (B.16)
Finally, the results of (B.14), (B.15), and (B.16) put together imply (B.13).
168
Some detailed proofs of two lemmas in Chapter 5
Recall from Chapter 5 that we wanted to prove a Riesz Basis property for the semigroup
generator /
1
. In doing so, we needed to prove that there exists a bounded, invertible operator
which maps an orthogonal basis of 1 to the basis of eigenvectors and generalized eigenvectors
for /
1
.
Proof of Lemma 5.3 in general case
We rst recall Lemma 5.3.
Lemma 5.3.
k

(j,k)I
is the sequence of eigenfunctions and generalized eigenfunctions
for /
1
with corresponding eigenvalues
2
k

(j,k)I
. Furthermore, E
0,j

m
j=1
is the sequence
of eigenvectors for /
0
with eigenvalue 0.
The proof of this lemma is easy when the eigenvalues
j
are all distinct. We now want to
verify it for the case of repeated eigenvalues. Since the notation is cumbersome, we will follow
this proof with an example that illustrates the meaning of the notation.
Proof. Suppose that R has has (1 m + 2) distinct eigenvalues
1
,
2
, ,

. For
each i = 1, 2, , , let r
i
be the algebraic multiplicity and p
i
be the geometric multiplic-
ity of
i
. Hence, there are p
i
Jordan chains corresponding to
i
. Let t
1
i
, t
2
i
, , t
p
i
i
be the
lengths of the Jordan chains corresponding to
i
. Dene for each xed i = 1, 2, , ,

1
i,1
,
1
i,2
, ,
1
i,p
i
to be the linearly independent eigenvectors for
i
. Then for each j =
1, 2, , p
i
, let
2
i,j
,
3
i,j
, ,
t
j
i
i,j
be the generalized eigenvectors corresponding to
i
. Hence,
we have the following:
_

_
R
1
i,j
=
i

1
i,j
,
R
s
i,j
=
i

s
i,j
+
s1
i,j
, for s = 2, 3, , t
j
i
(i, j, k xed).
169
Therefore, (5.33) implies
_

_
/
1
(
k

1
i,j
) =
2
k

i
(
k

1
i,j
),
/
1
(
k

s
i,j
) =
2
k

i
(
k

s
i,j
) +
2
k
(
k

s1
i,j
), for s = 2, 3, , t
j
i
(i, j, k xed). (B.17)
Thus, the linearly independent eigenfunctions of /
1
corresponding to
2
k

i
are

1
i,1
,
k

1
i,2
, ,
k

1
i,p
i
,
and the generalized eigenfunctions are

2
i,j
,
k

3
i,j
, ,
k

t
j
i
i,j
, for j = 1, 2, , p
i
Notice that the sequence
j

m+2
j=1
of eigenvalues and generalized eigenvalues of R is given by

m+2
j=1
=

_
i=1
p
i
_
j=1
t
j
i
_
s=1

s
i,j
.
Furthermore, the sequence of eigenvalues
j

m+2
j=1
includes
i

i=1
with
i
appearing r
i
times,
and it corresponds to the sequence of eigenvectors and generalized eigenvectors
j

m+2
j=1
. Then
the conclusion of the theorem follows.
Example B.1. To clarify the meaning of the notation in the above proof, suppose m = 6.
Then the matrix R is 8 8. Suppose R has distinct eigenvalues
1
= 5,
2
= 3, and

3
= 1 with algebraic multiplicities of r
1
= 3, r
2
= 1, and r
3
= 4 respectively. Suppose also
that the geometric multiplicities are p
1
= 1, p
2
= 1, and p
3
= 3 respectively. Then we have the
following:
_

1
1,1
is an eigenvector for
1
= 5,

2
1,1
and
3
1,1
are generalized eigenvectors for
1
= 5,

1
2,1
is an eigenvector for
2
= 3,
and
_

1
3,1
,
1
3,2
, and
1
3,3
are eigenvectors for
3
= 1,

2
3,3
is a generalized eigenvector for
3
= 1.
Thus,
t
1
1
= 3, t
1
2
= 1, t
1
3
= 1, t
2
3
= 1, and t
3
3
= 2.
170
In addition, the sequence
j

8
j=1
can be written as

1
1,1
,
2
1,1
,
3
1,1
,
1
2,1
,
1
3,1
,
1
3,2
,
1
3,3
,
2
3,3
,
with the corresponding sequence
j

8
j=1
given by
5, 5, 5, 3, 1, 1, 1, 1.
The Jordan canonical form for R is
_

_
5 1 0
0 5 1
0 0 5
3
1
1
1 1
0 1
_

_
.
A detailed proof of Lemma 5.4
The following is the statement of Lemma 5.4:
Lemma 5.4. Let X be a Hilbert space with an orthogonal basis E
j,k

(j,k)I
. Moreover,
suppose there is an > 0 such that < |E
j,k
| < 1/ for all (j, k) I. If T : X X has
matrix representation

T := diag(M, M, ), (B.18)
where M is an (m+2)(m+2) real, invertible matrix, then T is a bounded, invertible operator
on X.
One crucial step in the proof is to establish the validity of (5.39), namely
|Tx|
2
X
=

k=1
|M x
T
k
|
2
R
(m+2)
.
171
Considering the structure of the operator T and the representation of x A, this equality is
easy to see, even though the proof is involved. We include the entire proof for completeness
and clarity.
Proof. We rst show that T is bounded. Without loss of generality, we can assume that
E
j,k

(j,k)I
is an orthonormal basis for X. Furthermore, since X is a separable Hilbert space,
we may also assume without loss of generality that X =
2
(the space of square-summable
sequences). Let x X be represented as follows:
x =

k=1
x
k
In the above, x
k
has the representation
x
k
=
_
0, , 0
.
.
.
.
.
. 0, , 0
.
.
. x
k
.
.
. 0, , 0
.
.
.
_
T
,
where
x
k
=
_
x
1,k
, , x
(m+2),k

R
(m+2)
,
and
x
j,k
= x
k
, E
j,k
)

2
(j, k) I.
Thus for each k N,
x
k
=

=1
m+2

j=1
x
k
, E
j,
)E
j,
=
m+2

j=1
x
k
, E
j,k
)E
j,k
, since x
k
, E
j,
) =
,k
.
Therefore,
x
k
=
m+2

j=1
x
j,k
E
j,k
, for each k N. (B.19)
Hence we have the following for each k N:
|x
k
|
2
X
=
_
m+2

j=1
x
j,k
E
j,k
,
m+2

i=1
x
i,k
E
i,k
_
=
m+2

j=1
m+2

i=1
x
j,k
x
i,k
E
j,k
, E
i,k
)
=
m+2

j=1
x
j,k
x
j,k
, since E
j,k
, E
i,k
) =
i,j
.
172
Therefore,
|x
k
|
2
X
= | x
T
k
|
2
R
m+2
, for each k N. (B.20)
This implies for all x X the following:
|x|
2
X
= x, x)
=
_

k=1
m+2

j=1
x
j,k
E
j,k
,

=1
m+2

i=1
x
i,
E
i,
_
=

k=1
m+2

j=1

=1
m+2

i=1
x
j,k
x
i,
E
j,k
, E
i,
)
=

k=1
m+2

j=1
x
j,k
x
j,k
, since E
j,k
, E
i,
) =
(j,k),(i,)
.
For all x X we have,
|x|
2
X
=

k=1
| x
T
k
|
2
R
m+2
=

k=1
|x
k
|
2
X
, by (B.20). (B.21)
According to (B.18), we have for all k N, M = m
(i,j)
, where
m
(i,j)
= m
ijk
, m
ijk
= T(E
j,k
), E
i,k
), for i, j = 1, 2, , m+ 2.
In addition, (B.18) implies that T is an invariant operator in each block; that is
T(E
j,k
), E
i,
) =
_

_
0 if ,= k,
m
ijk
if = k
(B.22)
Hence, for each xed (j, k) I we have
T(E
j,k
) =

=1
m+2

i=1
T(E
j,k
), E
i,
)E
i,
=
m+2

i=1
m
ijk
E
i,k
by (B.22) (B.23)
Then for each k N, we have
Tx
k
= T
_
_
_
m+2

j=1
x
j,k
E
j,k
_
_
_
by (B.19)
=
m+2

j=1
x
j,k
T(E
j,k
) since T is linear by denition
=
m+2

j=1
x
j,k
_
m+2

i=1
m
ijk
E
i,k
_
by (B.23).
173
Therefore,
Tx
k
=
m+2

j=1
m+2

i=1
x
j,k
m
ijk
E
i,k
for each k N. (B.24)
Hence, for all x X,
Tx = T
_

k=1
x
k
_
=

k=1
Tx
k
by linearity of T
=

k=1
m+2

j=1
m+2

i=1
x
j,k
m
ijk
E
i,k
by (B.24). (B.25)
Thus, for all x X,:
|Tx|
2
X
= Tx, Tx)
=
_

k=1
m+2

j=1
m+2

i=1
x
j,k
m
ijk
E
i,k
,

=1
m+2

b=1
m+2

a=1
x
b,
m
ab
E
a,
_
by (B.25)
=

k=1
m+2

j=1
m+2

i=1

=1
m+2

b=1
m+2

a=1
x
j,k
m
ijk
x
b,
m
ab
E
i,k
, E
a,
).
Since E
i,k
, E
a,
) =
(i,k),(a,)
, the previous equation simplies as follows:
|T
x
|
2
X
=

k=1
_
_
m+2

i=1
m+2

j=1
m+2

b=1
x
j,k
m
ijk
x
b,k
m
ibk
_
_
for all x X. (B.26)
On the other hand, we have for each k N
M x
T
k
=
_

m+2
j=1
m
1jk
x
j,k
.
.
.

m+2
j=1
m
(m+2)jk
x
j,k
_

_
.
Hence, for each k N,
|M x
T
k
|
2
R
m+2
=
m+2

j=1
m
1jk
x
jk
_
m+2

b=1
m
1bk
x
bk
_
+ +
m+2

j=1
m
(m+2)jk
x
jk
_
m+2

b=1
m
(m+2)bk
x
bk
_
.
Thus,
|M x
T
k
|
2
R
m+2
=
m+2

i=1
m+2

j=1
m+2

b=1
m
ijk
x
jk
m
ibk
x
bk
, for all k N (B.27)
Combining the results of (B.26) and (B.27) yields
|Tx|
2
X
=

k=1
|M x
T
k
|
2
R
(m+2)
(B.28)
174
Since M is a bounded, invertible matrix, there is a constant C > 0 (independent of k) such
that
|M x
T
k
|
R
(m+2) C| x
T
k
|
R
(m+2) (B.29)
For all x X, (B.28) and (B.29) imply the following:
|Tx|
2
X
= C
2

k=1
| x
T
k
|
2
R
(m+2)
= C
2
|x|
2
X
by (B.21).
Hence T is bounded on X. To prove that T is invertible, we consider the operator V with the
matrix representation

V = diag(M
1
, M
1
, ).
This is well-dened, since M is invertible. Since M
1
is bounded, we repeat the same argument
as above to conclude that V is bounded on X. Clearly, V T = TV = I, and so V = T
1
.
175
BIBLIOGRAPHY
[1] A. A. Allen and S. W. Hansen, Analyticity of a multilayer Mead-Markus plate, J. Nonlinear
Analysis, 2009 (to appear).
[2] M. Artin, Algebra, Prentice Hall, New Jersey, 1991.
[3] G. Avalos, Concerning the exact null controllability of a nonlinear thermoelastic plate, Nonlinear
Analysis 63 (2005), e1455e1465.
[4] G. Avalos, Dierential Riccati equations for the active control of a problem in structural acoustic,
J. Optim. Theory, Appl. 91 (1996), 695728.
[5] G. Chen and D. L. Russell, A mathematical model for linear elastic systems with structural
damping Quarterly of Applied Mathematics 39 no. 4 (1982), 433454.
[6] S. Chen and R. Triggiani, Proof of extensions of two conjectures on structural damping for
elastic systems, Pacic J. Math. 136 no. 1 (1989), 1555.
[7] G. Duvaut and J.L Lions, Inequalities in Mathematics and Physics, Collection: Grundlehren
der mathematischen Wissenschaften 219, Springer-Verlag, Heidelburg, Germany, 1976.
[8] R. A. DiTaranto, Theory of vibratory bending for elastic and viscoelastic layered nite-length
beams, J. Appl. Mech. 32 (1965), 881-886.
[9] R. H. Fabiano and S. W. Hansen, Modelling and analysis of a three-layer damped sandwich
beam, J. Discrete Contin. Dynam. Sys. Added Volume (2001), 143155.
[10] S. W. Hansen, A dynamical model for multilayered plates with independent shear defor-
mations, Quarterly of Applied Mathematics 55 no. 4 (1997), 601621.
176
[11] S. W. Hansen, Exponential energy decay in a linear thermoelastic rod, J. Math. Anal. Appl
167 (1992), 429442.
[12] S. W. Hansen, Modeling and analysis of multilayer laminated plates, ESAIM: Proc. 4
(1998), 117135.
[13] S. W. Hansen, Optimal damping in multilayer sandwich beams, Smart Structures and Materials
in Proc. of SPIE 5049 (2003), 201208.
[14] S. W. Hansen, Exact controllability of a multilayer Rao-Nakra beam with minimal number of
boundary controls, Modeling, Signal Processing, and Control for Smart Structures in Proc.
of SPIE 6523 (2007), 65230P-165230P-8.
[15] S. W. Hansen, Semigroup well-posedness of a multilayer Mead-Markus plate with shear damp-
ing, Chapter XIX: Control and Boundary Analysis (eds. J. Cagnol, J.P. Zolesio), Series:
Lecture Notes in Applied Mathematics 240 (2005), 243256.
[16] S. W. Hansen, Several related models for multilayer sandwich plates, Math. Models & Methods
in Appl. Sci. 14 no. 8 (2004), 11031132.
[17] S. W. Hansen and I. Lasiecka, Analyticity, hyperbolicity and uniform stability of semigroups
arising in models of composite beams Math. Models Methods Appl. Sci. 10 no. 4 (2000),
555580.
[18] S. W. Hansen and Z. Liu, Analyticity of semigroup associated with a laminated composite beam,
Control of Distributed Parameter and Stochastic Systems (eds. S. Chen, X. Li, J. Yong,
X.Y. Zhou) (Hangzhou, 1998) 4754, Kluwer Acad. Publ., Boston, MA, 1999.
[19] S.W. Hansen and R. Rajaram, Null controllability of a damped Mead-Markus sandwich beam,
Proc. of the AIMS Fifth Int. Conference on Dynamical Systems and Dierential Equations,
2004.
[20] S.W. Hansen and R. Rajaram, Riesz basis property and related results for a Rao-Nakra sandwich
beam, Proc. Fifth Int. Conference on Dynamical Systems and Dierential Equations, 2005.
177
[21] S.W. Hansen and R. Rajaram, Simultaneous boundary control of a Rao-Nakra sandwich beam,
Proc. 44th IEEE Conference on Decision and Control and the European Control Confer-
ence, 2005.
[22] S. W. Hansen and B-Y. Zhang Boundary control of a linear thermoelastic beam, J. Math.
Anal., Appl. 210, (1997), 182205.
[23] F.L Huang, Characteristic condition for the exponential stability of linear dynamical sys-
tems in Hilbert spaces, Ann. of Di. Eqs. 1 (1985), 4356.
[24] T. Kato, Perturbation Theory for Linear Operators, Springer-Verlag, Heidelburg, Ger-
many, 1976.
[25] E.M Kerwin, Jr, Damping of exural waves by a constrained visco-elastic layer, J. Acoustical
Soc. Amer. 31 no.7 (1959) 952962.
[26] J.E. Lagnese and J.-L Lions, Modelling, Analysis and Control of Thin Plates, Collection:
Recherches en Mathematiques Appliquees RMA 6, Masson, Paris, 1988.
[27] P. Lancaster and M. Tismenetsky, The Theory of Matrices, Series: Computer Science and
Applied Mathematics, Academic Press, San Diego, 1985.
[28] I. Lasiecka, Mathematical Control Theory of Coupled PDEs, CMBS-NSF, Lecture Notes,
SIAM, Philadelphia, 2002.
[29] I. Lasiecka, Uniform decay rates for full von Karman system of dynamic thermoelasticity with free
boundary conditions and partial boundary dissipation, Comm. Partial Dierential Equations 24
no. 9-10 (1999), 18011847.
[30] I. Lasiecka and C. Lebiedzik, Asymptotic behaviour of nonlinear structural acoustic interactions
with thermal eects on the interface, Nonlinear Analysis 49 (2002), 703735.
[31] I. Lasiecka and R. Triggiani, Analyticity of thermo-elastic semigroups with free B.C., Ann.
Scuola Normale Superior 27 (1998), 457482.
178
[32] I. Lasiecka and R. Triggiani, Dierential and algebraic Riccati equations with application to
boundary/point control problems: continuous theory and approximation theory, Lecture Notes in
Control and Information Sciences 164, Springer-Verlag, Berlin, 1991.
[33] I. Lasiecka and R. Triggiani, Two direct proofs on the analyticity of the s.c. semigroup arising
in abstract thermoelastic equations. Adv. Dierential Equations 3 (1998), 387416.
[34] I. Lasiecka and A. Tuaha, Riccati equations for the Bolza problem arising in boundary/point
control problems governed by C
0
semigroups satisfying a singular estimate, J. Optim. Theory
Appl. 136 no. 2 (2008), 229246.
[35] Z. Liu and S. Zheng, Semigroups Associated with Dissipative Systems, Series: Research Notes
in Mathematics, Chapman and Hall, London, 1999.
[36] Z. H. Luo, B. Z. Guo, O. Morgul, Stability and Stabilization of Innite Dimensional Systems
with Applications, Springer, London, 1999.
[37] D. J. Mead and S. Markus, The forced vibration of a three-layer, damped sandwich beam with
arbitrary boundary conditions, J. Sound Vibr. 10 (1969), 163175.
[38] A. D. Nashif, D. I. G. Jones, J. P. Henderson, Vibration Damping, John Wiley & Sons, Inc.,
New York, 1985.
[39] A. Pazy, Semigroups of Linear Operators and Applications to Partial Dierential Equations,
Springer, New York, 1983.
[40] J. Pr uss, On the spectrum of C
0
-semigroups, Trans. of Amer. Math. Soc. 284 (1984),
847857.
[41] Y.V.K.S. Rao and B.C. Nakra, Vibrations of unsymmetrical sandwich beams and plates with
viscoelastic cores, J. Sound Vibr. 34 no.3 (1974), 309326.
[42] M. Renardy, On the linear stability of hyperbolic PDEs and viscoelastic ows, Z. Agnew. Math.
Phys. 45 (1994), 854865.
179
[43] D. Ross, E. Ungar, E. M. Kerwin, Damping of plate exural vibrations by means of viscoelastic
laminae, Structural Damping, Amer. Soc. Mech. Engis. (1959).
[44] D. L. Russell, Mathematical models for the elastic beam and their control-theoretic implications,
Semigroups, Theory and Applications II, Pittman Research Notes 152 (1986), 177217.
[45] C.T. Sun and Y.P. Liu, Vibration Damping of Structural Elements, 1995, Prentice Hall.
[46] R. Triggiani, The coupled PDE system of a composite (sandwich) beam revisited, Discrete Con-
tin. Dyn. Syst. Ser. B 3 no. 2 (2003), 285298.
[47] M.-J. Yan and E. H. Dowell, Governing equations for vibrating constrained-layer damping sand-
wich plates and beams, J. Appl. Mech 39 (1972), 10411046.
[48] R. M. Young, An Introduction to Non-Harmonic Fourier Series, 1980, New York.
180
ACKNOWLEDGEMENTS
I would like to take this opportunity to express my thanks to those who helped me with
various aspects of conducting research and the writing of this thesis. First of all, I would like
to thank professors Johnston, Sacks, DAlessandro, and Hou for serving on my committee.
Finally, I would like to express my gratitude for Dr. Scott W. Hansen. His guidance, patience,
and support led to many of the results in this thesis.